powerpc: Merge enough to start building in arch/powerpc.
authorPaul Mackerras <paulus@samba.org>
Mon, 26 Sep 2005 06:04:21 +0000 (16:04 +1000)
committerPaul Mackerras <paulus@samba.org>
Mon, 26 Sep 2005 06:04:21 +0000 (16:04 +1000)
This creates the directory structure under arch/powerpc and a bunch
of Kconfig files.  It does a first-cut merge of arch/powerpc/mm,
arch/powerpc/lib and arch/powerpc/platforms/powermac.  This is enough
to build a 32-bit powermac kernel with ARCH=powerpc.

For now we are getting some unmerged files from arch/ppc/kernel and
arch/ppc/syslib, or arch/ppc64/kernel.  This makes some minor changes
to files in those directories and files outside arch/powerpc.

The boot directory is still not merged.  That's going to be interesting.

Signed-off-by: Paul Mackerras <paulus@samba.org>
89 files changed:
arch/powerpc/Kconfig [new file with mode: 0644]
arch/powerpc/Kconfig.debug [new file with mode: 0644]
arch/powerpc/Makefile [new file with mode: 0644]
arch/powerpc/kernel/Makefile [new file with mode: 0644]
arch/powerpc/kernel/asm-offsets.c [new file with mode: 0644]
arch/powerpc/kernel/fpu.S [new file with mode: 0644]
arch/powerpc/kernel/head.S [new file with mode: 0644]
arch/powerpc/kernel/head_44x.S [new file with mode: 0644]
arch/powerpc/kernel/head_4xx.S [new file with mode: 0644]
arch/powerpc/kernel/head_64.S [new file with mode: 0644]
arch/powerpc/kernel/head_8xx.S [new file with mode: 0644]
arch/powerpc/kernel/head_fsl_booke.S [new file with mode: 0644]
arch/powerpc/kernel/idle_6xx.S [new file with mode: 0644]
arch/powerpc/kernel/process.c [new file with mode: 0644]
arch/powerpc/kernel/semaphore.c [new file with mode: 0644]
arch/powerpc/kernel/traps.c [new file with mode: 0644]
arch/powerpc/kernel/vector.S [new file with mode: 0644]
arch/powerpc/kernel/vmlinux.lds [new file with mode: 0644]
arch/powerpc/kernel/vmlinux.lds.S [new file with mode: 0644]
arch/powerpc/lib/Makefile [new file with mode: 0644]
arch/powerpc/lib/checksum.S [new file with mode: 0644]
arch/powerpc/lib/checksum64.S [new file with mode: 0644]
arch/powerpc/lib/copy32.S [new file with mode: 0644]
arch/powerpc/lib/copypage.S [new file with mode: 0644]
arch/powerpc/lib/copyuser.S [new file with mode: 0644]
arch/powerpc/lib/div64.S [new file with mode: 0644]
arch/powerpc/lib/e2a.c [new file with mode: 0644]
arch/powerpc/lib/memcpy.S [new file with mode: 0644]
arch/powerpc/lib/rheap.c [new file with mode: 0644]
arch/powerpc/lib/sstep.c [new file with mode: 0644]
arch/powerpc/lib/strcase.c [new file with mode: 0644]
arch/powerpc/lib/string.S [new file with mode: 0644]
arch/powerpc/lib/usercopy.c [new file with mode: 0644]
arch/powerpc/mm/44x_mmu.c [new file with mode: 0644]
arch/powerpc/mm/4xx_mmu.c [new file with mode: 0644]
arch/powerpc/mm/Makefile [new file with mode: 0644]
arch/powerpc/mm/fault.c [new file with mode: 0644]
arch/powerpc/mm/fsl_booke_mmu.c [new file with mode: 0644]
arch/powerpc/mm/hash_32.S [new file with mode: 0644]
arch/powerpc/mm/init.c [new file with mode: 0644]
arch/powerpc/mm/init64.c [new file with mode: 0644]
arch/powerpc/mm/mem.c [new file with mode: 0644]
arch/powerpc/mm/mem64.c [new file with mode: 0644]
arch/powerpc/mm/mem_pieces.c [new file with mode: 0644]
arch/powerpc/mm/mem_pieces.h [new file with mode: 0644]
arch/powerpc/mm/mmu_context.c [new file with mode: 0644]
arch/powerpc/mm/mmu_context64.c [new file with mode: 0644]
arch/powerpc/mm/mmu_decl.h [new file with mode: 0644]
arch/powerpc/mm/pgtable.c [new file with mode: 0644]
arch/powerpc/mm/pgtable64.c [new file with mode: 0644]
arch/powerpc/mm/ppc_mmu.c [new file with mode: 0644]
arch/powerpc/mm/tlb.c [new file with mode: 0644]
arch/powerpc/platforms/4xx/Kconfig [new file with mode: 0644]
arch/powerpc/platforms/85xx/Kconfig [new file with mode: 0644]
arch/powerpc/platforms/8xx/Kconfig [new file with mode: 0644]
arch/powerpc/platforms/apus/Kconfig [new file with mode: 0644]
arch/powerpc/platforms/embedded6xx/Kconfig [new file with mode: 0644]
arch/powerpc/platforms/iseries/Kconfig [new file with mode: 0644]
arch/powerpc/platforms/powermac/Makefile [new file with mode: 0644]
arch/powerpc/platforms/powermac/pmac.h [new file with mode: 0644]
arch/powerpc/platforms/powermac/pmac_backlight.c [new file with mode: 0644]
arch/powerpc/platforms/powermac/pmac_cache.S [new file with mode: 0644]
arch/powerpc/platforms/powermac/pmac_cpufreq.c [new file with mode: 0644]
arch/powerpc/platforms/powermac/pmac_feature.c [new file with mode: 0644]
arch/powerpc/platforms/powermac/pmac_low_i2c.c [new file with mode: 0644]
arch/powerpc/platforms/powermac/pmac_nvram.c [new file with mode: 0644]
arch/powerpc/platforms/powermac/pmac_pci.c [new file with mode: 0644]
arch/powerpc/platforms/powermac/pmac_pic.c [new file with mode: 0644]
arch/powerpc/platforms/powermac/pmac_pic.h [new file with mode: 0644]
arch/powerpc/platforms/powermac/pmac_setup.c [new file with mode: 0644]
arch/powerpc/platforms/powermac/pmac_sleep.S [new file with mode: 0644]
arch/powerpc/platforms/powermac/pmac_smp.c [new file with mode: 0644]
arch/powerpc/platforms/powermac/pmac_time.c [new file with mode: 0644]
arch/powerpc/platforms/prep/Kconfig [new file with mode: 0644]
arch/powerpc/platforms/pseries/Kconfig [new file with mode: 0644]
arch/powerpc/sysdev/Makefile [new file with mode: 0644]
arch/powerpc/sysdev/mpic.c [new file with mode: 0644]
arch/ppc/kernel/Makefile
arch/ppc/kernel/setup.c
arch/ppc/platforms/prep_setup.c
arch/ppc/syslib/Makefile
drivers/macintosh/via-pmu.c
fs/proc/proc_misc.c
include/asm-powerpc/kdebug.h [new file with mode: 0644]
include/asm-powerpc/kprobes.h [new file with mode: 0644]
include/asm-powerpc/mpic.h [new file with mode: 0644]
include/asm-powerpc/reg.h [new file with mode: 0644]
include/asm-powerpc/system.h [new file with mode: 0644]
include/asm-ppc/smp.h

diff --git a/arch/powerpc/Kconfig b/arch/powerpc/Kconfig
new file mode 100644 (file)
index 0000000..edfac46
--- /dev/null
@@ -0,0 +1,861 @@
+# For a description of the syntax of this configuration file,
+# see Documentation/kbuild/kconfig-language.txt.
+#
+
+mainmenu "Linux/PowerPC Kernel Configuration"
+
+config PPC64
+       bool "64-bit kernel"
+       default n
+       help
+         This option selects whether a 32-bit or a 64-bit kernel
+         will be built.
+
+config PPC32
+       bool
+       default y if !PPC64
+
+config 64BIT
+       bool
+       default y if PPC64
+
+config PPC_MERGE
+       def_bool y
+
+config MMU
+       bool
+       default y
+
+config UID16
+       bool
+
+config GENERIC_HARDIRQS
+       bool
+       default y
+
+config RWSEM_GENERIC_SPINLOCK
+       bool
+
+config RWSEM_XCHGADD_ALGORITHM
+       bool
+       default y
+
+config GENERIC_CALIBRATE_DELAY
+       bool
+       default y
+
+config PPC
+       bool
+       default y
+
+config EARLY_PRINTK
+       bool
+       default y if PPC64
+
+config COMPAT
+       bool
+       default y if PPC64
+
+config SYSVIPC_COMPAT
+       bool
+       depends on COMPAT && SYSVIPC
+       default y
+
+# All PPC32s use generic nvram driver through ppc_md
+config GENERIC_NVRAM
+       bool
+       default y if PPC32
+
+config SCHED_NO_NO_OMIT_FRAME_POINTER
+       bool
+       default y
+
+config ARCH_MAY_HAVE_PC_FDC
+       bool
+       default y
+
+menu "Processor support"
+choice
+       prompt "Processor Type"
+       depends on PPC32
+       default 6xx
+
+config 6xx
+       bool "6xx/7xx/74xx"
+       select PPC_FPU
+       help
+         There are four families of PowerPC chips supported.  The more common
+         types (601, 603, 604, 740, 750, 7400), the Motorola embedded
+         versions (821, 823, 850, 855, 860, 52xx, 82xx, 83xx), the AMCC
+         embedded versions (403 and 405) and the high end 64 bit Power
+         processors (POWER 3, POWER4, and IBM PPC970 also known as G5).
+         
+         Unless you are building a kernel for one of the embedded processor
+         systems, 64 bit IBM RS/6000 or an Apple G5, choose 6xx.
+         Note that the kernel runs in 32-bit mode even on 64-bit chips.
+
+config PPC_52xx
+       bool "Freescale 52xx"
+       
+config PPC_82xx
+       bool "Freescale 82xx"
+
+config PPC_83xx
+       bool "Freescale 83xx"
+
+config 40x
+       bool "AMCC 40x"
+
+config 44x
+       bool "AMCC 44x"
+
+config PPC64BRIDGE
+       select PPC_FPU
+       bool "POWER3, POWER4 and PPC970 (G5)"
+
+config 8xx
+       bool "Freescale 8xx"
+
+config E200
+       bool "Freescale e200"
+
+config E500
+       bool "Freescale e500"
+endchoice
+
+config POWER4_ONLY
+       bool "Optimize for POWER4"
+       depends on PPC64 || PPC64BRIDGE
+       default n
+       ---help---
+         Cause the compiler to optimize for POWER4/POWER5/PPC970 processors.
+         The resulting binary will not work on POWER3 or RS64 processors
+         when compiled with binutils 2.15 or later.
+
+config POWER3
+       bool
+       depends on PPC64 || PPC64BRIDGE
+       default y if !POWER4_ONLY
+
+config POWER4
+       depends on PPC64 || PPC64BRIDGE
+       def_bool y
+
+config PPC_FPU
+       bool
+       default y if PPC64
+
+config BOOKE
+       bool
+       depends on E200 || E500
+       default y
+
+config FSL_BOOKE
+       bool
+       depends on E200 || E500
+       default y
+
+config PTE_64BIT
+       bool
+       depends on 44x || E500
+       default y if 44x
+       default y if E500 && PHYS_64BIT
+
+config PHYS_64BIT
+       bool 'Large physical address support' if E500
+       depends on 44x || E500
+       default y if 44x
+       ---help---
+         This option enables kernel support for larger than 32-bit physical
+         addresses.  This features is not be available on all e500 cores.
+
+         If in doubt, say N here.
+
+config ALTIVEC
+       bool "AltiVec Support"
+       depends on 6xx || POWER4
+       ---help---
+         This option enables kernel support for the Altivec extensions to the
+         PowerPC processor. The kernel currently supports saving and restoring
+         altivec registers, and turning on the 'altivec enable' bit so user
+         processes can execute altivec instructions.
+
+         This option is only usefully if you have a processor that supports
+         altivec (G4, otherwise known as 74xx series), but does not have
+         any affect on a non-altivec cpu (it does, however add code to the
+         kernel).
+
+         If in doubt, say Y here.
+
+config SPE
+       bool "SPE Support"
+       depends on E200 || E500
+       ---help---
+         This option enables kernel support for the Signal Processing
+         Extensions (SPE) to the PowerPC processor. The kernel currently
+         supports saving and restoring SPE registers, and turning on the
+         'spe enable' bit so user processes can execute SPE instructions.
+
+         This option is only useful if you have a processor that supports
+         SPE (e500, otherwise known as 85xx series), but does not have any
+         effect on a non-spe cpu (it does, however add code to the kernel).
+
+         If in doubt, say Y here.
+
+config PPC_STD_MMU
+       bool
+       depends on 6xx || POWER3 || POWER4 || PPC64
+       default y
+
+config PPC_STD_MMU_32
+       def_bool y
+       depends on PPC_STD_MMU && PPC32
+
+config SMP
+       depends on PPC_STD_MMU
+       bool "Symmetric multi-processing support"
+       ---help---
+         This enables support for systems with more than one CPU. If you have
+         a system with only one CPU, say N. If you have a system with more
+         than one CPU, say Y.  Note that the kernel does not currently
+         support SMP machines with 603/603e/603ev or PPC750 ("G3") processors
+         since they have inadequate hardware support for multiprocessor
+         operation.
+
+         If you say N here, the kernel will run on single and multiprocessor
+         machines, but will use only one CPU of a multiprocessor machine. If
+         you say Y here, the kernel will run on single-processor machines.
+         On a single-processor machine, the kernel will run faster if you say
+         N here.
+
+         If you don't know what to do here, say N.
+
+config NR_CPUS
+       int "Maximum number of CPUs (2-32)"
+       range 2 128
+       depends on SMP
+       default "32" if PPC64
+       default "4"
+
+config NOT_COHERENT_CACHE
+       bool
+       depends on 4xx || 8xx || E200
+       default y
+endmenu
+
+source "init/Kconfig"
+
+menu "Platform support"
+       depends on PPC64 || 6xx
+
+choice
+       prompt "Machine type"
+       default PPC_MULTIPLATFORM
+
+config PPC_MULTIPLATFORM
+       bool "Generic desktop/server/laptop"
+       help
+         Select this option if configuring for an IBM pSeries or
+         RS/6000 machine, an Apple machine, or a PReP, CHRP,
+         Maple or Cell-based machine.
+
+config PPC_ISERIES
+       bool "IBM Legacy iSeries"
+       depends on PPC64
+
+config EMBEDDED6xx
+       bool "Embedded 6xx/7xx/7xxx-based board"
+       depends on PPC32
+
+config APUS
+       bool "Amiga-APUS"
+       depends on PPC32 && BROKEN
+       help
+         Select APUS if configuring for a PowerUP Amiga.
+         More information is available at:
+         <http://linux-apus.sourceforge.net/>.
+endchoice
+
+config PPC_PSERIES
+       depends on PPC_MULTIPLATFORM && PPC64
+       bool "  IBM pSeries & new (POWER5-based) iSeries"
+       default y
+
+config PPC_CHRP
+       bool "  Common Hardware Reference Platform (CHRP) based machines"
+       depends on PPC_MULTIPLATFORM && PPC32
+       default y
+
+config PPC_PMAC
+       bool "  Apple PowerMac based machines"
+       depends on PPC_MULTIPLATFORM
+       default y
+
+config PPC_PMAC64
+       bool
+       depends on PPC_PMAC && POWER4
+       default y
+
+config PPC_PREP
+       bool "  PowerPC Reference Platform (PReP) based machines"
+       depends on PPC_MULTIPLATFORM && PPC32
+       default y
+
+config PPC_MAPLE
+       depends on PPC_MULTIPLATFORM && PPC64
+       bool "  Maple 970FX Evaluation Board"
+       select U3_DART
+       select MPIC_BROKEN_U3
+       default n
+       help
+          This option enables support for the Maple 970FX Evaluation Board.
+         For more informations, refer to <http://www.970eval.com>
+
+config PPC_BPA
+       bool "  Broadband Processor Architecture"
+       depends on PPC_MULTIPLATFORM && PPC64
+
+config PPC_OF
+       bool
+       depends on PPC_MULTIPLATFORM    # for now
+       default y
+
+config XICS
+       depends on PPC_PSERIES
+       bool
+       default y
+
+config U3_DART
+       bool 
+       depends on PPC_MULTIPLATFORM && PPC64
+       default n
+
+config MPIC
+       depends on PPC_PSERIES || PPC_PMAC || PPC_MAPLE
+       bool
+       default y
+
+config MPIC_BROKEN_U3
+       bool
+       depends on PPC_MAPLE
+       default y
+
+config BPA_IIC
+       depends on PPC_BPA
+       bool
+       default y
+
+config IBMVIO
+       depends on PPC_PSERIES || PPC_ISERIES
+       bool
+       default y
+
+source "drivers/cpufreq/Kconfig"
+
+config CPU_FREQ_PMAC
+       bool "Support for Apple PowerBooks"
+       depends on CPU_FREQ && ADB_PMU && PPC32
+       select CPU_FREQ_TABLE
+       help
+         This adds support for frequency switching on Apple PowerBooks,
+         this currently includes some models of iBook & Titanium
+         PowerBook.
+
+config PPC601_SYNC_FIX
+       bool "Workarounds for PPC601 bugs"
+       depends on 6xx && (PPC_PREP || PPC_PMAC)
+       help
+         Some versions of the PPC601 (the first PowerPC chip) have bugs which
+         mean that extra synchronization instructions are required near
+         certain instructions, typically those that make major changes to the
+         CPU state.  These extra instructions reduce performance slightly.
+         If you say N here, these extra instructions will not be included,
+         resulting in a kernel which will run faster but may not run at all
+         on some systems with the PPC601 chip.
+
+         If in doubt, say Y here.
+
+config TAU
+       bool "Thermal Management Support"
+       depends on 6xx
+       help
+         G3 and G4 processors have an on-chip temperature sensor called the
+         'Thermal Assist Unit (TAU)', which, in theory, can measure the on-die
+         temperature within 2-4 degrees Celsius. This option shows the current
+         on-die temperature in /proc/cpuinfo if the cpu supports it.
+
+         Unfortunately, on some chip revisions, this sensor is very inaccurate
+         and in some cases, does not work at all, so don't assume the cpu
+         temp is actually what /proc/cpuinfo says it is.
+
+config TAU_INT
+       bool "Interrupt driven TAU driver (DANGEROUS)"
+       depends on TAU
+       ---help---
+         The TAU supports an interrupt driven mode which causes an interrupt
+         whenever the temperature goes out of range. This is the fastest way
+         to get notified the temp has exceeded a range. With this option off,
+         a timer is used to re-check the temperature periodically.
+
+         However, on some cpus it appears that the TAU interrupt hardware
+         is buggy and can cause a situation which would lead unexplained hard
+         lockups.
+
+         Unless you are extending the TAU driver, or enjoy kernel/hardware
+         debugging, leave this option off.
+
+config TAU_AVERAGE
+       bool "Average high and low temp"
+       depends on TAU
+       ---help---
+         The TAU hardware can compare the temperature to an upper and lower
+         bound.  The default behavior is to show both the upper and lower
+         bound in /proc/cpuinfo. If the range is large, the temperature is
+         either changing a lot, or the TAU hardware is broken (likely on some
+         G4's). If the range is small (around 4 degrees), the temperature is
+         relatively stable.  If you say Y here, a single temperature value,
+         halfway between the upper and lower bounds, will be reported in
+         /proc/cpuinfo.
+
+         If in doubt, say N here.
+endmenu
+
+source arch/powerpc/platforms/embedded6xx/Kconfig
+source arch/powerpc/platforms/4xx/Kconfig
+source arch/powerpc/platforms/85xx/Kconfig
+source arch/powerpc/platforms/8xx/Kconfig
+
+menu "Kernel options"
+
+config HIGHMEM
+       bool "High memory support"
+       depends on PPC32
+
+source kernel/Kconfig.hz
+source kernel/Kconfig.preempt
+source "fs/Kconfig.binfmt"
+
+# We optimistically allocate largepages from the VM, so make the limit
+# large enough (16MB). This badly named config option is actually
+# max order + 1
+config FORCE_MAX_ZONEORDER
+       int
+       depends on PPC64
+       default "13"
+
+config MATH_EMULATION
+       bool "Math emulation"
+       depends on 4xx || 8xx || E200 || E500
+       ---help---
+         Some PowerPC chips designed for embedded applications do not have
+         a floating-point unit and therefore do not implement the
+         floating-point instructions in the PowerPC instruction set.  If you
+         say Y here, the kernel will include code to emulate a floating-point
+         unit, which will allow programs that use floating-point
+         instructions to run.
+
+config IOMMU_VMERGE
+       bool "Enable IOMMU virtual merging (EXPERIMENTAL)"
+       depends on EXPERIMENTAL && PPC64
+       default n
+       help
+         Cause IO segments sent to a device for DMA to be merged virtually
+         by the IOMMU when they happen to have been allocated contiguously.
+         This doesn't add pressure to the IOMMU allocator. However, some
+         drivers don't support getting large merged segments coming back
+         from *_map_sg(). Say Y if you know the drivers you are using are
+         properly handling this case.
+
+config HOTPLUG_CPU
+       bool "Support for enabling/disabling CPUs"
+       depends on SMP && HOTPLUG && EXPERIMENTAL && (PPC_PSERIES || PPC_PMAC)
+       ---help---
+         Say Y here to be able to disable and re-enable individual
+         CPUs at runtime on SMP machines.
+
+         Say N if you are unsure.
+
+config KEXEC
+       bool "kexec system call (EXPERIMENTAL)"
+       depends on PPC_MULTIPLATFORM && EXPERIMENTAL
+       help
+         kexec is a system call that implements the ability to shutdown your
+         current kernel, and to start another kernel.  It is like a reboot
+         but it is indepedent of the system firmware.   And like a reboot
+         you can start any kernel with it, not just Linux.
+
+         The name comes from the similiarity to the exec system call.
+
+         It is an ongoing process to be certain the hardware in a machine
+         is properly shutdown, so do not be surprised if this code does not
+         initially work for you.  It may help to enable device hotplugging
+         support.  As of this writing the exact hardware interface is
+         strongly in flux, so no good recommendation can be made.
+
+config EMBEDDEDBOOT
+       bool
+       depends on 8xx || 8260
+       default y
+
+config PC_KEYBOARD
+       bool "PC PS/2 style Keyboard"
+       depends on 4xx || CPM2
+
+config PPCBUG_NVRAM
+       bool "Enable reading PPCBUG NVRAM during boot" if PPLUS || LOPEC
+       default y if PPC_PREP
+
+config IRQ_ALL_CPUS
+       bool "Distribute interrupts on all CPUs by default"
+       depends on SMP && !MV64360
+       help
+         This option gives the kernel permission to distribute IRQs across
+         multiple CPUs.  Saying N here will route all IRQs to the first
+         CPU.  Generally saying Y is safe, although some problems have been
+         reported with SMP Power Macintoshes with this option enabled.
+
+source "arch/powerpc/platforms/pseries/Kconfig"
+
+config ARCH_SELECT_MEMORY_MODEL
+       def_bool y
+       depends on PPC64
+
+config ARCH_FLATMEM_ENABLE
+       def_bool y
+       depends on PPC64 && !NUMA
+
+config ARCH_DISCONTIGMEM_ENABLE
+       def_bool y
+       depends on SMP && PPC_PSERIES
+
+config ARCH_DISCONTIGMEM_DEFAULT
+       def_bool y
+       depends on ARCH_DISCONTIGMEM_ENABLE
+
+config ARCH_FLATMEM_ENABLE
+       def_bool y
+       depends on PPC64
+
+config ARCH_SPARSEMEM_ENABLE
+       def_bool y
+       depends on ARCH_DISCONTIGMEM_ENABLE
+
+source "mm/Kconfig"
+
+config HAVE_ARCH_EARLY_PFN_TO_NID
+       def_bool y
+       depends on NEED_MULTIPLE_NODES
+
+# Some NUMA nodes have memory ranges that span
+# other nodes.  Even though a pfn is valid and
+# between a node's start and end pfns, it may not
+# reside on that node.
+#
+# This is a relatively temporary hack that should
+# be able to go away when sparsemem is fully in
+# place
+
+config NODES_SPAN_OTHER_NODES
+       def_bool y
+       depends on NEED_MULTIPLE_NODES
+
+config NUMA
+       bool "NUMA support"
+       default y if DISCONTIGMEM || SPARSEMEM
+
+config SCHED_SMT
+       bool "SMT (Hyperthreading) scheduler support"
+       depends on PPC64 && SMP
+       default off
+       help
+         SMT scheduler support improves the CPU scheduler's decision making
+         when dealing with POWER5 cpus at a cost of slightly increased
+         overhead in some places. If unsure say N here.
+
+config PROC_DEVICETREE
+       bool "Support for Open Firmware device tree in /proc"
+       depends on PPC_OF && PROC_FS
+       help
+         This option adds a device-tree directory under /proc which contains
+         an image of the device tree that the kernel copies from Open
+         Firmware. If unsure, say Y here.
+
+source "arch/powerpc/platforms/prep/Kconfig"
+
+config CMDLINE_BOOL
+       bool "Default bootloader kernel arguments"
+       depends on !PPC_ISERIES
+
+config CMDLINE
+       string "Initial kernel command string"
+       depends on CMDLINE_BOOL
+       default "console=ttyS0,9600 console=tty0 root=/dev/sda2"
+       help
+         On some platforms, there is currently no way for the boot loader to
+         pass arguments to the kernel. For these platforms, you can supply
+         some command-line options at build time by entering them here.  In
+         most cases you will need to specify the root device here.
+
+if !44x || BROKEN
+source kernel/power/Kconfig
+endif
+
+config SECCOMP
+       bool "Enable seccomp to safely compute untrusted bytecode"
+       depends on PROC_FS
+       default y
+       help
+         This kernel feature is useful for number crunching applications
+         that may need to compute untrusted bytecode during their
+         execution. By using pipes or other transports made available to
+         the process as file descriptors supporting the read/write
+         syscalls, it's possible to isolate those applications in
+         their own address space using seccomp. Once seccomp is
+         enabled via /proc/<pid>/seccomp, it cannot be disabled
+         and the task is only allowed to execute a few safe syscalls
+         defined by each seccomp mode.
+
+         If unsure, say Y. Only embedded should say N here.
+
+endmenu
+
+config ISA_DMA_API
+       bool
+       default y
+
+menu "Bus options"
+
+config ISA
+       bool "Support for ISA-bus hardware"
+       depends on PPC_PREP || PPC_CHRP
+       help
+         Find out whether you have ISA slots on your motherboard.  ISA is the
+         name of a bus system, i.e. the way the CPU talks to the other stuff
+         inside your box.  If you have an Apple machine, say N here; if you
+         have an IBM RS/6000 or pSeries machine or a PReP machine, say Y.  If
+         you have an embedded board, consult your board documentation.
+
+config GENERIC_ISA_DMA
+       bool
+       depends on PPC64 || POWER4 || 6xx && !CPM2
+       default y
+
+config EISA
+       bool
+
+config SBUS
+       bool
+
+# Yes MCA RS/6000s exist but Linux-PPC does not currently support any
+config MCA
+       bool
+
+config PCI
+       bool "PCI support" if 40x || CPM2 || 83xx || 85xx || PPC_MPC52xx || (EMBEDDED && PPC_ISERIES)
+       default y if !40x && !CPM2 && !8xx && !APUS && !83xx && !85xx
+       default PCI_PERMEDIA if !4xx && !CPM2 && !8xx && APUS
+       default PCI_QSPAN if !4xx && !CPM2 && 8xx
+       help
+         Find out whether your system includes a PCI bus. PCI is the name of
+         a bus system, i.e. the way the CPU talks to the other stuff inside
+         your box.  If you say Y here, the kernel will include drivers and
+         infrastructure code to support PCI bus devices.
+
+config PCI_DOMAINS
+       bool
+       default PCI
+
+config MPC83xx_PCI2
+       bool "  Supprt for 2nd PCI host controller"
+       depends on PCI && MPC834x
+       default y if MPC834x_SYS
+
+config PCI_QSPAN
+       bool "QSpan PCI"
+       depends on !4xx && !CPM2 && 8xx
+       help
+         Say Y here if you have a system based on a Motorola 8xx-series
+         embedded processor with a QSPAN PCI interface, otherwise say N.
+
+config PCI_8260
+       bool
+       depends on PCI && 8260
+       default y
+
+config 8260_PCI9
+       bool "  Enable workaround for MPC826x erratum PCI 9"
+       depends on PCI_8260 && !ADS8272
+       default y
+
+choice
+       prompt "  IDMA channel for PCI 9 workaround"
+       depends on 8260_PCI9
+
+config 8260_PCI9_IDMA1
+       bool "IDMA1"
+
+config 8260_PCI9_IDMA2
+       bool "IDMA2"
+
+config 8260_PCI9_IDMA3
+       bool "IDMA3"
+
+config 8260_PCI9_IDMA4
+       bool "IDMA4"
+
+endchoice
+
+source "drivers/pci/Kconfig"
+
+source "drivers/pcmcia/Kconfig"
+
+source "drivers/pci/hotplug/Kconfig"
+
+endmenu
+
+menu "Advanced setup"
+       depends on PPC32
+
+config ADVANCED_OPTIONS
+       bool "Prompt for advanced kernel configuration options"
+       help
+         This option will enable prompting for a variety of advanced kernel
+         configuration options.  These options can cause the kernel to not
+         work if they are set incorrectly, but can be used to optimize certain
+         aspects of kernel memory management.
+
+         Unless you know what you are doing, say N here.
+
+comment "Default settings for advanced configuration options are used"
+       depends on !ADVANCED_OPTIONS
+
+config HIGHMEM_START_BOOL
+       bool "Set high memory pool address"
+       depends on ADVANCED_OPTIONS && HIGHMEM
+       help
+         This option allows you to set the base address of the kernel virtual
+         area used to map high memory pages.  This can be useful in
+         optimizing the layout of kernel virtual memory.
+
+         Say N here unless you know what you are doing.
+
+config HIGHMEM_START
+       hex "Virtual start address of high memory pool" if HIGHMEM_START_BOOL
+       default "0xfe000000"
+
+config LOWMEM_SIZE_BOOL
+       bool "Set maximum low memory"
+       depends on ADVANCED_OPTIONS
+       help
+         This option allows you to set the maximum amount of memory which
+         will be used as "low memory", that is, memory which the kernel can
+         access directly, without having to set up a kernel virtual mapping.
+         This can be useful in optimizing the layout of kernel virtual
+         memory.
+
+         Say N here unless you know what you are doing.
+
+config LOWMEM_SIZE
+       hex "Maximum low memory size (in bytes)" if LOWMEM_SIZE_BOOL
+       default "0x30000000"
+
+config KERNEL_START_BOOL
+       bool "Set custom kernel base address"
+       depends on ADVANCED_OPTIONS
+       help
+         This option allows you to set the kernel virtual address at which
+         the kernel will map low memory (the kernel image will be linked at
+         this address).  This can be useful in optimizing the virtual memory
+         layout of the system.
+
+         Say N here unless you know what you are doing.
+
+config KERNEL_START
+       hex "Virtual address of kernel base" if KERNEL_START_BOOL
+       default "0xc0000000"
+
+config TASK_SIZE_BOOL
+       bool "Set custom user task size"
+       depends on ADVANCED_OPTIONS
+       help
+         This option allows you to set the amount of virtual address space
+         allocated to user tasks.  This can be useful in optimizing the
+         virtual memory layout of the system.
+
+         Say N here unless you know what you are doing.
+
+config TASK_SIZE
+       hex "Size of user task space" if TASK_SIZE_BOOL
+       default "0x80000000"
+
+config CONSISTENT_START_BOOL
+       bool "Set custom consistent memory pool address"
+       depends on ADVANCED_OPTIONS && NOT_COHERENT_CACHE
+       help
+         This option allows you to set the base virtual address
+         of the the consistent memory pool.  This pool of virtual
+         memory is used to make consistent memory allocations.
+
+config CONSISTENT_START
+       hex "Base virtual address of consistent memory pool" if CONSISTENT_START_BOOL
+       default "0xff100000" if NOT_COHERENT_CACHE
+
+config CONSISTENT_SIZE_BOOL
+       bool "Set custom consistent memory pool size"
+       depends on ADVANCED_OPTIONS && NOT_COHERENT_CACHE
+       help
+         This option allows you to set the size of the the
+         consistent memory pool.  This pool of virtual memory
+         is used to make consistent memory allocations.
+
+config CONSISTENT_SIZE
+       hex "Size of consistent memory pool" if CONSISTENT_SIZE_BOOL
+       default "0x00200000" if NOT_COHERENT_CACHE
+
+config BOOT_LOAD_BOOL
+       bool "Set the boot link/load address"
+       depends on ADVANCED_OPTIONS && !PPC_MULTIPLATFORM
+       help
+         This option allows you to set the initial load address of the zImage
+         or zImage.initrd file.  This can be useful if you are on a board
+         which has a small amount of memory.
+
+         Say N here unless you know what you are doing.
+
+config BOOT_LOAD
+       hex "Link/load address for booting" if BOOT_LOAD_BOOL
+       default "0x00400000" if 40x || 8xx || 8260
+       default "0x01000000" if 44x
+       default "0x00800000"
+
+config PIN_TLB
+       bool "Pinned Kernel TLBs (860 ONLY)"
+       depends on ADVANCED_OPTIONS && 8xx
+endmenu
+
+source "net/Kconfig"
+
+source "drivers/Kconfig"
+
+source "fs/Kconfig"
+
+# XXX source "arch/ppc/8xx_io/Kconfig"
+
+# XXX source "arch/ppc/8260_io/Kconfig"
+
+source "arch/powerpc/platforms/iseries/Kconfig"
+
+source "lib/Kconfig"
+
+source "arch/powerpc/oprofile/Kconfig"
+
+source "arch/powerpc/Kconfig.debug"
+
+source "security/Kconfig"
+
+config KEYS_COMPAT
+       bool
+       depends on COMPAT && KEYS
+       default y
+
+source "crypto/Kconfig"
diff --git a/arch/powerpc/Kconfig.debug b/arch/powerpc/Kconfig.debug
new file mode 100644 (file)
index 0000000..61653cb
--- /dev/null
@@ -0,0 +1,73 @@
+menu "Kernel hacking"
+
+source "lib/Kconfig.debug"
+
+config KGDB
+       bool "Include kgdb kernel debugger"
+       depends on DEBUG_KERNEL && (BROKEN || PPC_GEN550 || 4xx)
+       select DEBUG_INFO
+       help
+         Include in-kernel hooks for kgdb, the Linux kernel source level
+         debugger.  See <http://kgdb.sourceforge.net/> for more information.
+         Unless you are intending to debug the kernel, say N here.
+
+choice
+       prompt "Serial Port"
+       depends on KGDB
+       default KGDB_TTYS1
+
+config KGDB_TTYS0
+       bool "ttyS0"
+
+config KGDB_TTYS1
+       bool "ttyS1"
+
+config KGDB_TTYS2
+       bool "ttyS2"
+
+config KGDB_TTYS3
+       bool "ttyS3"
+
+endchoice
+
+config KGDB_CONSOLE
+       bool "Enable serial console thru kgdb port"
+       depends on KGDB && 8xx || CPM2
+       help
+         If you enable this, all serial console messages will be sent
+         over the gdb stub.
+         If unsure, say N.
+
+config XMON
+       bool "Include xmon kernel debugger"
+       depends on DEBUG_KERNEL
+       help
+         Include in-kernel hooks for the xmon kernel monitor/debugger.
+         Unless you are intending to debug the kernel, say N here.
+
+config BDI_SWITCH
+       bool "Include BDI-2000 user context switcher"
+       depends on DEBUG_KERNEL
+       help
+         Include in-kernel support for the Abatron BDI2000 debugger.
+         Unless you are intending to debug the kernel with one of these
+         machines, say N here.
+
+config BOOTX_TEXT
+       bool "Support for early boot text console (BootX or OpenFirmware only)"
+       depends PPC_OF
+       help
+         Say Y here to see progress messages from the boot firmware in text
+         mode. Requires either BootX or Open Firmware.
+
+config SERIAL_TEXT_DEBUG
+       bool "Support for early boot texts over serial port"
+       depends on 4xx || LOPEC || MV64X60 || PPLUS || PRPMC800 || \
+               PPC_GEN550 || PPC_MPC52xx
+
+config PPC_OCP
+       bool
+       depends on IBM_OCP || XILINX_OCP
+       default y
+
+endmenu
diff --git a/arch/powerpc/Makefile b/arch/powerpc/Makefile
new file mode 100644 (file)
index 0000000..8a65e11
--- /dev/null
@@ -0,0 +1,222 @@
+# This file is included by the global makefile so that you can add your own
+# architecture-specific flags and dependencies. Remember to do have actions
+# for "archclean" and "archdep" for cleaning up and making dependencies for
+# this architecture.
+#
+# This file is subject to the terms and conditions of the GNU General Public
+# License.  See the file "COPYING" in the main directory of this archive
+# for more details.
+#
+# Copyright (C) 1994 by Linus Torvalds
+# Changes for PPC by Gary Thomas
+# Rewritten by Cort Dougan and Paul Mackerras
+#
+
+# This must match PAGE_OFFSET in include/asm-powerpc/page.h.
+KERNELLOAD     := $(CONFIG_KERNEL_START)
+
+HAS_BIARCH     := $(call cc-option-yn, -m32)
+
+ifeq ($(CONFIG_PPC64),y)
+SZ     := 64
+
+# Set default 32 bits cross compilers for vdso and boot wrapper
+CROSS32_COMPILE ?=
+
+CROSS32CC              := $(CROSS32_COMPILE)gcc
+CROSS32AS              := $(CROSS32_COMPILE)as
+CROSS32LD              := $(CROSS32_COMPILE)ld
+CROSS32OBJCOPY         := $(CROSS32_COMPILE)objcopy
+
+ifeq ($(HAS_BIARCH),y)
+ifeq ($(CROSS32_COMPILE),)
+CROSS32CC      := $(CC) -m32
+CROSS32AS      := $(AS) -a32
+CROSS32LD      := $(LD) -m elf32ppc
+CROSS32OBJCOPY := $(OBJCOPY)
+endif
+endif
+
+export CROSS32CC CROSS32AS CROSS32LD CROSS32OBJCOPY
+
+new_nm := $(shell if $(NM) --help 2>&1 | grep -- '--synthetic' > /dev/null; then echo y; else echo n; fi)
+
+ifeq ($(new_nm),y)
+NM             := $(NM) --synthetic
+endif
+
+else
+SZ     := 32
+endif
+
+ifeq ($(HAS_BIARCH),y)
+override AS    += -a$(SZ)
+override LD    += -m elf$(SZ)ppc
+override CC    += -m$(SZ)
+endif
+
+LDFLAGS_vmlinux        := -Ttext $(KERNELLOAD) -Bstatic -e $(KERNELLOAD)
+
+# The -Iarch/$(ARCH)/include is temporary while we are merging
+CPPFLAGS       += -Iarch/$(ARCH) -Iarch/$(ARCH)/include
+AFLAGS         += -Iarch/$(ARCH)
+CFLAGS         += -Iarch/$(ARCH) -msoft-float -pipe
+ifeq ($(CONFIG_PPC64),y)
+CFLAGS         += -mminimal-toc -mtraceback=none  -mcall-aixdesc
+else
+CFLAGS         += -ffixed-r2 -mmultiple
+endif
+CPP            = $(CC) -E $(CFLAGS)
+# Temporary hack until we have migrated to asm-powerpc
+LINUXINCLUDE    += -Iarch/$(ARCH)/include
+
+CHECKFLAGS     += -m$(SZ) -D__powerpc__ -D__powerpc$(SZ)__
+
+ifeq ($(CONFIG_PPC64),y)
+GCC_VERSION     := $(call cc-version)
+GCC_BROKEN_VEC := $(shell if [ $(GCC_VERSION) -lt 0400 ] ; then echo "y"; fi)
+
+ifeq ($(CONFIG_POWER4_ONLY),y)
+ifeq ($(CONFIG_ALTIVEC),y)
+ifeq ($(GCC_BROKEN_VEC),y)
+       CFLAGS += $(call cc-option,-mcpu=970)
+else
+       CFLAGS += $(call cc-option,-mcpu=power4)
+endif
+else
+       CFLAGS += $(call cc-option,-mcpu=power4)
+endif
+else
+       CFLAGS += $(call cc-option,-mtune=power4)
+endif
+endif
+
+# Enable unit-at-a-time mode when possible. It shrinks the
+# kernel considerably.
+CFLAGS += $(call cc-option,-funit-at-a-time)
+
+ifndef CONFIG_FSL_BOOKE
+CFLAGS         += -mstring
+endif
+
+cpu-as-$(CONFIG_PPC64BRIDGE)   += -Wa,-mppc64bridge
+cpu-as-$(CONFIG_4xx)           += -Wa,-m405
+cpu-as-$(CONFIG_6xx)           += -Wa,-maltivec
+cpu-as-$(CONFIG_POWER4)                += -Wa,-maltivec
+cpu-as-$(CONFIG_E500)          += -Wa,-me500
+cpu-as-$(CONFIG_E200)          += -Wa,-me200
+
+AFLAGS += $(cpu-as-y)
+CFLAGS += $(cpu-as-y)
+
+# Default to the common case.
+KBUILD_DEFCONFIG := common_defconfig
+
+head-y                         := arch/powerpc/kernel/head.o
+head-$(CONFIG_PPC64)           := arch/powerpc/kernel/head_64.o
+head-$(CONFIG_8xx)             := arch/powerpc/kernel/head_8xx.o
+head-$(CONFIG_4xx)             := arch/powerpc/kernel/head_4xx.o
+head-$(CONFIG_44x)             := arch/powerpc/kernel/head_44x.o
+head-$(CONFIG_FSL_BOOKE)       := arch/powerpc/kernel/head_fsl_booke.o
+
+ifeq ($(CONFIG_PPC32),y)
+head-$(CONFIG_6xx)             += arch/powerpc/kernel/idle_6xx.o
+head-$(CONFIG_POWER4)          += arch/powerpc/kernel/idle_power4.o
+head-$(CONFIG_PPC_FPU)         += arch/powerpc/kernel/fpu.o
+endif
+
+core-y                         += arch/powerpc/kernel/ \
+                                  arch/powerpc/mm/ \
+                                  arch/powerpc/lib/ \
+                                  arch/powerpc/sysdev/
+core-$(CONFIG_PPC32)           += arch/ppc/kernel/ \
+                                  arch/ppc/syslib/
+core-$(CONFIG_PPC64)           += arch/ppc64/kernel/
+core-$(CONFIG_PPC_PMAC)                += arch/powerpc/platforms/powermac/
+core-$(CONFIG_4xx)             += arch/ppc/platforms/4xx/
+core-$(CONFIG_83xx)            += arch/ppc/platforms/83xx/
+core-$(CONFIG_85xx)            += arch/ppc/platforms/85xx/
+core-$(CONFIG_MATH_EMULATION)  += arch/ppc/math-emu/
+core-$(CONFIG_XMON)            += arch/powerpc/xmon/
+core-$(CONFIG_APUS)            += arch/ppc/amiga/
+drivers-$(CONFIG_8xx)          += arch/ppc/8xx_io/
+drivers-$(CONFIG_4xx)          += arch/ppc/4xx_io/
+drivers-$(CONFIG_CPM2)         += arch/ppc/8260_io/
+
+drivers-$(CONFIG_OPROFILE)     += arch/powerpc/oprofile/
+
+BOOT_TARGETS = zImage zImage.initrd znetboot znetboot.initrd vmlinux.sm
+
+.PHONY: $(BOOT_TARGETS)
+
+all: uImage zImage
+
+CPPFLAGS_vmlinux.lds   := -Upowerpc
+
+# All the instructions talk about "make bzImage".
+bzImage: zImage
+
+boot := arch/$(ARCH)/boot
+
+$(BOOT_TARGETS): vmlinux
+       $(Q)$(MAKE) $(build)=$(boot) $@
+
+uImage: vmlinux
+       $(Q)$(MAKE) $(build)=$(boot)/images $(boot)/images/$@
+
+define archhelp
+  @echo '* zImage          - Compressed kernel image (arch/$(ARCH)/boot/images/zImage.*)'
+  @echo '  uImage          - Create a bootable image for U-Boot / PPCBoot'
+  @echo '  install         - Install kernel using'
+  @echo '                    (your) ~/bin/installkernel or'
+  @echo '                    (distribution) /sbin/installkernel or'
+  @echo '                    install to $$(INSTALL_PATH) and run lilo'
+  @echo '  *_defconfig     - Select default config from arch/$(ARCH)/ppc/configs'
+endef
+
+archclean:
+       $(Q)$(MAKE) $(clean)=arch/ppc/boot
+       # Temporary hack until we have migrated to asm-powerpc
+       $(Q)rm -rf arch/$(ARCH)/include
+
+archprepare: checkbin
+
+# Temporary hack until we have migrated to asm-powerpc
+ifeq ($(CONFIG_PPC64),y)
+include/asm: arch/$(ARCH)/include/asm
+arch/$(ARCH)/include/asm:
+       $(Q)if [ ! -d arch/$(ARCH)/include ]; then mkdir -p arch/$(ARCH)/include; fi
+       $(Q)ln -fsn $(srctree)/include/asm-ppc64 arch/$(ARCH)/include/asm
+else
+include/asm: arch/$(ARCH)/include/asm
+arch/$(ARCH)/include/asm:
+       $(Q)if [ ! -d arch/$(ARCH)/include ]; then mkdir -p arch/$(ARCH)/include; fi
+       $(Q)ln -fsn $(srctree)/include/asm-ppc arch/$(ARCH)/include/asm
+endif
+
+# Use the file '.tmp_gas_check' for binutils tests, as gas won't output
+# to stdout and these checks are run even on install targets.
+TOUT   := .tmp_gas_check
+# Ensure this is binutils 2.12.1 (or 2.12.90.0.7) or later for altivec
+# instructions.
+# gcc-3.4 and binutils-2.14 are a fatal combination.
+GCC_VERSION    := $(call cc-version)
+
+checkbin:
+       @if test "$(GCC_VERSION)" = "0304" ; then \
+               if ! /bin/echo mftb 5 | $(AS) -v -mppc -many -o $(TOUT) >/dev/null 2>&1 ; then \
+                       echo -n '*** ${VERSION}.${PATCHLEVEL} kernels no longer build '; \
+                       echo 'correctly with gcc-3.4 and your version of binutils.'; \
+                       echo '*** Please upgrade your binutils or downgrade your gcc'; \
+                       false; \
+               fi ; \
+       fi
+       @if ! /bin/echo dssall | $(AS) -many -o $(TOUT) >/dev/null 2>&1 ; then \
+               echo -n '*** ${VERSION}.${PATCHLEVEL} kernels no longer build ' ; \
+               echo 'correctly with old versions of binutils.' ; \
+               echo '*** Please upgrade your binutils to 2.12.1 or newer' ; \
+               false ; \
+       fi
+
+CLEAN_FILES += $(TOUT)
+
diff --git a/arch/powerpc/kernel/Makefile b/arch/powerpc/kernel/Makefile
new file mode 100644 (file)
index 0000000..62c4a51
--- /dev/null
@@ -0,0 +1,18 @@
+#
+# Makefile for the linux kernel.
+#
+
+extra-$(CONFIG_PPC_STD_MMU)    := head.o
+extra_$(CONFIG_PPC64)          := head_64.o
+extra-$(CONFIG_40x)            := head_4xx.o
+extra-$(CONFIG_44x)            := head_44x.o
+extra-$(CONFIG_FSL_BOOKE)      := head_fsl_booke.o
+extra-$(CONFIG_8xx)            := head_8xx.o
+extra-$(CONFIG_6xx)            += idle_6xx.o
+extra-$(CONFIG_POWER4)         += idle_power4.o
+extra-$(CONFIG_PPC_FPU)                += fpu.o
+extra-y                                += vmlinux.lds
+
+obj-y                          := semaphore.o traps.o process.o
+
+obj-$(CONFIG_ALTIVEC)          += vecemu.o vector.o
diff --git a/arch/powerpc/kernel/asm-offsets.c b/arch/powerpc/kernel/asm-offsets.c
new file mode 100644 (file)
index 0000000..16cf0b7
--- /dev/null
@@ -0,0 +1,262 @@
+/*
+ * This program is used to generate definitions needed by
+ * assembly language modules.
+ *
+ * We use the technique used in the OSF Mach kernel code:
+ * generate asm statements containing #defines,
+ * compile this file to assembler, and then extract the
+ * #defines from the assembly-language output.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version
+ * 2 of the License, or (at your option) any later version.
+ */
+
+#include <linux/config.h>
+#include <linux/signal.h>
+#include <linux/sched.h>
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/string.h>
+#include <linux/types.h>
+#include <linux/ptrace.h>
+#include <linux/suspend.h>
+#include <linux/mman.h>
+#include <linux/mm.h>
+#include <linux/time.h>
+#include <linux/hardirq.h>
+#include <asm/io.h>
+#include <asm/page.h>
+#include <asm/pgtable.h>
+#include <asm/processor.h>
+
+#include <asm/cputable.h>
+#include <asm/thread_info.h>
+#ifdef CONFIG_PPC64
+#include <asm/paca.h>
+#include <asm/lppaca.h>
+#include <asm/iSeries/HvLpEvent.h>
+#include <asm/rtas.h>
+#include <asm/cache.h>
+#include <asm/systemcfg.h>
+#include <asm/compat.h>
+#endif
+
+#define DEFINE(sym, val) \
+       asm volatile("\n->" #sym " %0 " #val : : "i" (val))
+
+#define BLANK() asm volatile("\n->" : : )
+
+int main(void)
+{
+       /* thread struct on stack */
+       DEFINE(TI_FLAGS, offsetof(struct thread_info, flags));
+       DEFINE(TI_PREEMPT, offsetof(struct thread_info, preempt_count));
+       DEFINE(TI_CPU, offsetof(struct thread_info, cpu));
+#ifdef CONFIG_PPC32
+       DEFINE(TI_LOCAL_FLAGS, offsetof(struct thread_info, local_flags));
+#endif
+#ifdef CONFIG_PPC64
+       DEFINE(TI_SC_NOERR, offsetof(struct thread_info, syscall_noerror));
+       DEFINE(THREAD_SHIFT, THREAD_SHIFT);
+#endif
+       DEFINE(THREAD_SIZE, THREAD_SIZE);
+
+       /* task_struct->thread */
+       DEFINE(THREAD, offsetof(struct task_struct, thread));
+       DEFINE(THREAD_INFO, offsetof(struct task_struct, thread_info));
+       DEFINE(MM, offsetof(struct task_struct, mm));
+       DEFINE(PTRACE, offsetof(struct task_struct, ptrace));
+       DEFINE(KSP, offsetof(struct thread_struct, ksp));
+       DEFINE(PGDIR, offsetof(struct thread_struct, pgdir));
+       DEFINE(LAST_SYSCALL, offsetof(struct thread_struct, last_syscall));
+       DEFINE(PT_REGS, offsetof(struct thread_struct, regs));
+       DEFINE(THREAD_FPEXC_MODE, offsetof(struct thread_struct, fpexc_mode));
+       DEFINE(THREAD_FPR0, offsetof(struct thread_struct, fpr[0]));
+       DEFINE(THREAD_FPSCR, offsetof(struct thread_struct, fpscr));
+#if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
+       DEFINE(THREAD_DBCR0, offsetof(struct thread_struct, dbcr0));
+       DEFINE(PT_PTRACED, PT_PTRACED);
+#endif
+#ifdef CONFIG_PPC64
+       DEFINE(KSP_VSID, offsetof(struct thread_struct, ksp_vsid));
+#endif
+
+#ifdef CONFIG_ALTIVEC
+       DEFINE(THREAD_VR0, offsetof(struct thread_struct, vr[0]));
+       DEFINE(THREAD_VRSAVE, offsetof(struct thread_struct, vrsave));
+       DEFINE(THREAD_VSCR, offsetof(struct thread_struct, vscr));
+       DEFINE(THREAD_USED_VR, offsetof(struct thread_struct, used_vr));
+#endif /* CONFIG_ALTIVEC */
+#ifdef CONFIG_SPE
+       DEFINE(THREAD_EVR0, offsetof(struct thread_struct, evr[0]));
+       DEFINE(THREAD_ACC, offsetof(struct thread_struct, acc));
+       DEFINE(THREAD_SPEFSCR, offsetof(struct thread_struct, spefscr));
+       DEFINE(THREAD_USED_SPE, offsetof(struct thread_struct, used_spe));
+#endif /* CONFIG_SPE */
+       /* Interrupt register frame */
+       DEFINE(STACK_FRAME_OVERHEAD, STACK_FRAME_OVERHEAD);
+#ifndef CONFIG_PPC64
+       DEFINE(INT_FRAME_SIZE, STACK_FRAME_OVERHEAD + sizeof(struct pt_regs));
+#else
+       DEFINE(SWITCH_FRAME_SIZE, STACK_FRAME_OVERHEAD + sizeof(struct pt_regs));
+
+       /* 288 = # of volatile regs, int & fp, for leaf routines */
+       /* which do not stack a frame.  See the PPC64 ABI.       */
+       DEFINE(INT_FRAME_SIZE, STACK_FRAME_OVERHEAD + sizeof(struct pt_regs) + 288);
+#endif
+       /* in fact we only use gpr0 - gpr9 and gpr20 - gpr23 */
+       DEFINE(GPR0, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[0]));
+       DEFINE(GPR1, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[1]));
+       DEFINE(GPR2, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[2]));
+       DEFINE(GPR3, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[3]));
+       DEFINE(GPR4, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[4]));
+       DEFINE(GPR5, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[5]));
+       DEFINE(GPR6, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[6]));
+       DEFINE(GPR7, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[7]));
+       DEFINE(GPR8, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[8]));
+       DEFINE(GPR9, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[9]));
+       DEFINE(GPR10, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[10]));
+       DEFINE(GPR11, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[11]));
+       DEFINE(GPR12, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[12]));
+       DEFINE(GPR13, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[13]));
+       DEFINE(GPR14, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[14]));
+       DEFINE(GPR15, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[15]));
+       DEFINE(GPR16, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[16]));
+       DEFINE(GPR17, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[17]));
+       DEFINE(GPR18, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[18]));
+       DEFINE(GPR19, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[19]));
+       DEFINE(GPR20, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[20]));
+       DEFINE(GPR21, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[21]));
+       DEFINE(GPR22, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[22]));
+       DEFINE(GPR23, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[23]));
+       DEFINE(GPR24, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[24]));
+       DEFINE(GPR25, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[25]));
+       DEFINE(GPR26, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[26]));
+       DEFINE(GPR27, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[27]));
+       DEFINE(GPR28, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[28]));
+       DEFINE(GPR29, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[29]));
+       DEFINE(GPR30, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[30]));
+       DEFINE(GPR31, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, gpr[31]));
+       /*
+        * Note: these symbols include _ because they overlap with special
+        * register names
+        */
+       DEFINE(_NIP, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, nip));
+       DEFINE(_MSR, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, msr));
+       DEFINE(_CTR, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, ctr));
+       DEFINE(_LINK, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, link));
+       DEFINE(_CCR, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, ccr));
+       DEFINE(_MQ, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, mq));
+       DEFINE(_XER, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, xer));
+       DEFINE(_DAR, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, dar));
+       DEFINE(_DSISR, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, dsisr));
+       /* The PowerPC 400-class & Book-E processors have neither the DAR nor the DSISR
+        * SPRs. Hence, we overload them to hold the similar DEAR and ESR SPRs
+        * for such processors.  For critical interrupts we use them to
+        * hold SRR0 and SRR1.
+        */
+       DEFINE(_DEAR, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, dar));
+       DEFINE(_ESR, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, dsisr));
+       DEFINE(ORIG_GPR3, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, orig_gpr3));
+       DEFINE(RESULT, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, result));
+       DEFINE(TRAP, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, trap));
+       DEFINE(CLONE_VM, CLONE_VM);
+       DEFINE(CLONE_UNTRACED, CLONE_UNTRACED);
+       DEFINE(MM_PGD, offsetof(struct mm_struct, pgd));
+
+       /* About the CPU features table */
+       DEFINE(CPU_SPEC_ENTRY_SIZE, sizeof(struct cpu_spec));
+       DEFINE(CPU_SPEC_PVR_MASK, offsetof(struct cpu_spec, pvr_mask));
+       DEFINE(CPU_SPEC_PVR_VALUE, offsetof(struct cpu_spec, pvr_value));
+       DEFINE(CPU_SPEC_FEATURES, offsetof(struct cpu_spec, cpu_features));
+       DEFINE(CPU_SPEC_SETUP, offsetof(struct cpu_spec, cpu_setup));
+
+#ifdef CONFIG_PPC64
+       DEFINE(MM, offsetof(struct task_struct, mm));
+       DEFINE(AUDITCONTEXT, offsetof(struct task_struct, audit_context));
+
+       DEFINE(DCACHEL1LINESIZE, offsetof(struct ppc64_caches, dline_size));
+       DEFINE(DCACHEL1LOGLINESIZE, offsetof(struct ppc64_caches, log_dline_size));
+       DEFINE(DCACHEL1LINESPERPAGE, offsetof(struct ppc64_caches, dlines_per_page));
+       DEFINE(ICACHEL1LINESIZE, offsetof(struct ppc64_caches, iline_size));
+       DEFINE(ICACHEL1LOGLINESIZE, offsetof(struct ppc64_caches, log_iline_size));
+       DEFINE(ICACHEL1LINESPERPAGE, offsetof(struct ppc64_caches, ilines_per_page));
+       DEFINE(PLATFORM, offsetof(struct systemcfg, platform));
+
+       /* paca */
+        DEFINE(PACA_SIZE, sizeof(struct paca_struct));
+        DEFINE(PACAPACAINDEX, offsetof(struct paca_struct, paca_index));
+        DEFINE(PACAPROCSTART, offsetof(struct paca_struct, cpu_start));
+        DEFINE(PACAKSAVE, offsetof(struct paca_struct, kstack));
+       DEFINE(PACACURRENT, offsetof(struct paca_struct, __current));
+        DEFINE(PACASAVEDMSR, offsetof(struct paca_struct, saved_msr));
+        DEFINE(PACASTABREAL, offsetof(struct paca_struct, stab_real));
+        DEFINE(PACASTABVIRT, offsetof(struct paca_struct, stab_addr));
+       DEFINE(PACASTABRR, offsetof(struct paca_struct, stab_rr));
+        DEFINE(PACAR1, offsetof(struct paca_struct, saved_r1));
+       DEFINE(PACATOC, offsetof(struct paca_struct, kernel_toc));
+       DEFINE(PACAPROCENABLED, offsetof(struct paca_struct, proc_enabled));
+       DEFINE(PACASLBCACHE, offsetof(struct paca_struct, slb_cache));
+       DEFINE(PACASLBCACHEPTR, offsetof(struct paca_struct, slb_cache_ptr));
+       DEFINE(PACACONTEXTID, offsetof(struct paca_struct, context.id));
+#ifdef CONFIG_HUGETLB_PAGE
+       DEFINE(PACALOWHTLBAREAS, offsetof(struct paca_struct, context.low_htlb_areas));
+       DEFINE(PACAHIGHHTLBAREAS, offsetof(struct paca_struct, context.high_htlb_areas));
+#endif /* CONFIG_HUGETLB_PAGE */
+       DEFINE(PACADEFAULTDECR, offsetof(struct paca_struct, default_decr));
+        DEFINE(PACA_EXGEN, offsetof(struct paca_struct, exgen));
+        DEFINE(PACA_EXMC, offsetof(struct paca_struct, exmc));
+        DEFINE(PACA_EXSLB, offsetof(struct paca_struct, exslb));
+        DEFINE(PACA_EXDSI, offsetof(struct paca_struct, exdsi));
+        DEFINE(PACAEMERGSP, offsetof(struct paca_struct, emergency_sp));
+       DEFINE(PACALPPACA, offsetof(struct paca_struct, lppaca));
+       DEFINE(PACAHWCPUID, offsetof(struct paca_struct, hw_cpu_id));
+       DEFINE(LPPACASRR0, offsetof(struct lppaca, saved_srr0));
+       DEFINE(LPPACASRR1, offsetof(struct lppaca, saved_srr1));
+       DEFINE(LPPACAANYINT, offsetof(struct lppaca, int_dword.any_int));
+       DEFINE(LPPACADECRINT, offsetof(struct lppaca, int_dword.fields.decr_int));
+
+       /* RTAS */
+       DEFINE(RTASBASE, offsetof(struct rtas_t, base));
+       DEFINE(RTASENTRY, offsetof(struct rtas_t, entry));
+
+       DEFINE(_TRAP, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, trap));
+       DEFINE(SOFTE, STACK_FRAME_OVERHEAD+offsetof(struct pt_regs, softe));
+
+       /* Create extra stack space for SRR0 and SRR1 when calling prom/rtas. */
+       DEFINE(PROM_FRAME_SIZE, STACK_FRAME_OVERHEAD + sizeof(struct pt_regs) + 16);
+       DEFINE(RTAS_FRAME_SIZE, STACK_FRAME_OVERHEAD + sizeof(struct pt_regs) + 16);
+
+       /* These _only_ to be used with {PROM,RTAS}_FRAME_SIZE!!! */
+       DEFINE(_SRR0, STACK_FRAME_OVERHEAD+sizeof(struct pt_regs));
+       DEFINE(_SRR1, STACK_FRAME_OVERHEAD+sizeof(struct pt_regs)+8);
+
+       /* systemcfg offsets for use by vdso */
+       DEFINE(CFG_TB_ORIG_STAMP, offsetof(struct systemcfg, tb_orig_stamp));
+       DEFINE(CFG_TB_TICKS_PER_SEC, offsetof(struct systemcfg, tb_ticks_per_sec));
+       DEFINE(CFG_TB_TO_XS, offsetof(struct systemcfg, tb_to_xs));
+       DEFINE(CFG_STAMP_XSEC, offsetof(struct systemcfg, stamp_xsec));
+       DEFINE(CFG_TB_UPDATE_COUNT, offsetof(struct systemcfg, tb_update_count));
+       DEFINE(CFG_TZ_MINUTEWEST, offsetof(struct systemcfg, tz_minuteswest));
+       DEFINE(CFG_TZ_DSTTIME, offsetof(struct systemcfg, tz_dsttime));
+       DEFINE(CFG_SYSCALL_MAP32, offsetof(struct systemcfg, syscall_map_32));
+       DEFINE(CFG_SYSCALL_MAP64, offsetof(struct systemcfg, syscall_map_64));
+
+       /* timeval/timezone offsets for use by vdso */
+       DEFINE(TVAL64_TV_SEC, offsetof(struct timeval, tv_sec));
+       DEFINE(TVAL64_TV_USEC, offsetof(struct timeval, tv_usec));
+       DEFINE(TVAL32_TV_SEC, offsetof(struct compat_timeval, tv_sec));
+       DEFINE(TVAL32_TV_USEC, offsetof(struct compat_timeval, tv_usec));
+       DEFINE(TZONE_TZ_MINWEST, offsetof(struct timezone, tz_minuteswest));
+       DEFINE(TZONE_TZ_DSTTIME, offsetof(struct timezone, tz_dsttime));
+#endif
+
+       DEFINE(pbe_address, offsetof(struct pbe, address));
+       DEFINE(pbe_orig_address, offsetof(struct pbe, orig_address));
+       DEFINE(pbe_next, offsetof(struct pbe, next));
+
+       DEFINE(NUM_USER_SEGMENTS, TASK_SIZE>>28);
+       return 0;
+}
diff --git a/arch/powerpc/kernel/fpu.S b/arch/powerpc/kernel/fpu.S
new file mode 100644 (file)
index 0000000..665d7d3
--- /dev/null
@@ -0,0 +1,133 @@
+/*
+ *  FPU support code, moved here from head.S so that it can be used
+ *  by chips which use other head-whatever.S files.
+ *
+ *  This program is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU General Public License
+ *  as published by the Free Software Foundation; either version
+ *  2 of the License, or (at your option) any later version.
+ *
+ */
+
+#include <linux/config.h>
+#include <asm/processor.h>
+#include <asm/page.h>
+#include <asm/mmu.h>
+#include <asm/pgtable.h>
+#include <asm/cputable.h>
+#include <asm/cache.h>
+#include <asm/thread_info.h>
+#include <asm/ppc_asm.h>
+#include <asm/asm-offsets.h>
+
+/*
+ * This task wants to use the FPU now.
+ * On UP, disable FP for the task which had the FPU previously,
+ * and save its floating-point registers in its thread_struct.
+ * Load up this task's FP registers from its thread_struct,
+ * enable the FPU for the current task and return to the task.
+ */
+       .globl  load_up_fpu
+load_up_fpu:
+       mfmsr   r5
+       ori     r5,r5,MSR_FP
+#ifdef CONFIG_PPC64BRIDGE
+       clrldi  r5,r5,1                 /* turn off 64-bit mode */
+#endif /* CONFIG_PPC64BRIDGE */
+       SYNC
+       MTMSRD(r5)                      /* enable use of fpu now */
+       isync
+/*
+ * For SMP, we don't do lazy FPU switching because it just gets too
+ * horrendously complex, especially when a task switches from one CPU
+ * to another.  Instead we call giveup_fpu in switch_to.
+ */
+#ifndef CONFIG_SMP
+       tophys(r6,0)                    /* get __pa constant */
+       addis   r3,r6,last_task_used_math@ha
+       lwz     r4,last_task_used_math@l(r3)
+       cmpwi   0,r4,0
+       beq     1f
+       add     r4,r4,r6
+       addi    r4,r4,THREAD            /* want last_task_used_math->thread */
+       SAVE_32FPRS(0, r4)
+       mffs    fr0
+       stfd    fr0,THREAD_FPSCR-4(r4)
+       lwz     r5,PT_REGS(r4)
+       add     r5,r5,r6
+       lwz     r4,_MSR-STACK_FRAME_OVERHEAD(r5)
+       li      r10,MSR_FP|MSR_FE0|MSR_FE1
+       andc    r4,r4,r10               /* disable FP for previous task */
+       stw     r4,_MSR-STACK_FRAME_OVERHEAD(r5)
+1:
+#endif /* CONFIG_SMP */
+       /* enable use of FP after return */
+       mfspr   r5,SPRN_SPRG3           /* current task's THREAD (phys) */
+       lwz     r4,THREAD_FPEXC_MODE(r5)
+       ori     r9,r9,MSR_FP            /* enable FP for current */
+       or      r9,r9,r4
+       lfd     fr0,THREAD_FPSCR-4(r5)
+       mtfsf   0xff,fr0
+       REST_32FPRS(0, r5)
+#ifndef CONFIG_SMP
+       subi    r4,r5,THREAD
+       sub     r4,r4,r6
+       stw     r4,last_task_used_math@l(r3)
+#endif /* CONFIG_SMP */
+       /* restore registers and return */
+       /* we haven't used ctr or xer or lr */
+       b       fast_exception_return
+
+/*
+ * FP unavailable trap from kernel - print a message, but let
+ * the task use FP in the kernel until it returns to user mode.
+ */
+       .globl  KernelFP
+KernelFP:
+       lwz     r3,_MSR(r1)
+       ori     r3,r3,MSR_FP
+       stw     r3,_MSR(r1)             /* enable use of FP after return */
+       lis     r3,86f@h
+       ori     r3,r3,86f@l
+       mr      r4,r2                   /* current */
+       lwz     r5,_NIP(r1)
+       bl      printk
+       b       ret_from_except
+86:    .string "floating point used in kernel (task=%p, pc=%x)\n"
+       .align  4,0
+
+/*
+ * giveup_fpu(tsk)
+ * Disable FP for the task given as the argument,
+ * and save the floating-point registers in its thread_struct.
+ * Enables the FPU for use in the kernel on return.
+ */
+       .globl  giveup_fpu
+giveup_fpu:
+       mfmsr   r5
+       ori     r5,r5,MSR_FP
+       SYNC_601
+       ISYNC_601
+       MTMSRD(r5)                      /* enable use of fpu now */
+       SYNC_601
+       isync
+       cmpwi   0,r3,0
+       beqlr-                          /* if no previous owner, done */
+       addi    r3,r3,THREAD            /* want THREAD of task */
+       lwz     r5,PT_REGS(r3)
+       cmpwi   0,r5,0
+       SAVE_32FPRS(0, r3)
+       mffs    fr0
+       stfd    fr0,THREAD_FPSCR-4(r3)
+       beq     1f
+       lwz     r4,_MSR-STACK_FRAME_OVERHEAD(r5)
+       li      r3,MSR_FP|MSR_FE0|MSR_FE1
+       andc    r4,r4,r3                /* disable FP for previous task */
+       stw     r4,_MSR-STACK_FRAME_OVERHEAD(r5)
+1:
+#ifndef CONFIG_SMP
+       li      r5,0
+       lis     r4,last_task_used_math@ha
+       stw     r5,last_task_used_math@l(r4)
+#endif /* CONFIG_SMP */
+       blr
diff --git a/arch/powerpc/kernel/head.S b/arch/powerpc/kernel/head.S
new file mode 100644 (file)
index 0000000..d05509f
--- /dev/null
@@ -0,0 +1,1545 @@
+/*
+ *  PowerPC version
+ *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
+ *
+ *  Rewritten by Cort Dougan (cort@cs.nmt.edu) for PReP
+ *    Copyright (C) 1996 Cort Dougan <cort@cs.nmt.edu>
+ *  Adapted for Power Macintosh by Paul Mackerras.
+ *  Low-level exception handlers and MMU support
+ *  rewritten by Paul Mackerras.
+ *    Copyright (C) 1996 Paul Mackerras.
+ *  MPC8xx modifications Copyright (C) 1997 Dan Malek (dmalek@jlc.net).
+ *  Amiga/APUS changes by Jesper Skov (jskov@cygnus.co.uk).
+ *
+ *  This file contains the low-level support and setup for the
+ *  PowerPC platform, including trap and interrupt dispatch.
+ *  (The PPC 8xx embedded CPUs use head_8xx.S instead.)
+ *
+ *  This program is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU General Public License
+ *  as published by the Free Software Foundation; either version
+ *  2 of the License, or (at your option) any later version.
+ *
+ */
+
+#include <linux/config.h>
+#include <asm/processor.h>
+#include <asm/page.h>
+#include <asm/mmu.h>
+#include <asm/pgtable.h>
+#include <asm/cputable.h>
+#include <asm/cache.h>
+#include <asm/thread_info.h>
+#include <asm/ppc_asm.h>
+#include <asm/asm-offsets.h>
+
+#ifdef CONFIG_APUS
+#include <asm/amigappc.h>
+#endif
+
+#ifdef CONFIG_PPC64BRIDGE
+#define LOAD_BAT(n, reg, RA, RB)       \
+       ld      RA,(n*32)+0(reg);       \
+       ld      RB,(n*32)+8(reg);       \
+       mtspr   SPRN_IBAT##n##U,RA;     \
+       mtspr   SPRN_IBAT##n##L,RB;     \
+       ld      RA,(n*32)+16(reg);      \
+       ld      RB,(n*32)+24(reg);      \
+       mtspr   SPRN_DBAT##n##U,RA;     \
+       mtspr   SPRN_DBAT##n##L,RB;     \
+
+#else /* CONFIG_PPC64BRIDGE */
+
+/* 601 only have IBAT; cr0.eq is set on 601 when using this macro */
+#define LOAD_BAT(n, reg, RA, RB)       \
+       /* see the comment for clear_bats() -- Cort */ \
+       li      RA,0;                   \
+       mtspr   SPRN_IBAT##n##U,RA;     \
+       mtspr   SPRN_DBAT##n##U,RA;     \
+       lwz     RA,(n*16)+0(reg);       \
+       lwz     RB,(n*16)+4(reg);       \
+       mtspr   SPRN_IBAT##n##U,RA;     \
+       mtspr   SPRN_IBAT##n##L,RB;     \
+       beq     1f;                     \
+       lwz     RA,(n*16)+8(reg);       \
+       lwz     RB,(n*16)+12(reg);      \
+       mtspr   SPRN_DBAT##n##U,RA;     \
+       mtspr   SPRN_DBAT##n##L,RB;     \
+1:
+#endif /* CONFIG_PPC64BRIDGE */
+
+       .text
+       .stabs  "arch/ppc/kernel/",N_SO,0,0,0f
+       .stabs  "head.S",N_SO,0,0,0f
+0:
+       .globl  _stext
+_stext:
+
+/*
+ * _start is defined this way because the XCOFF loader in the OpenFirmware
+ * on the powermac expects the entry point to be a procedure descriptor.
+ */
+       .text
+       .globl  _start
+_start:
+       /*
+        * These are here for legacy reasons, the kernel used to
+        * need to look like a coff function entry for the pmac
+        * but we're always started by some kind of bootloader now.
+        *  -- Cort
+        */
+       nop     /* used by __secondary_hold on prep (mtx) and chrp smp */
+       nop     /* used by __secondary_hold on prep (mtx) and chrp smp */
+       nop
+
+/* PMAC
+ * Enter here with the kernel text, data and bss loaded starting at
+ * 0, running with virtual == physical mapping.
+ * r5 points to the prom entry point (the client interface handler
+ * address).  Address translation is turned on, with the prom
+ * managing the hash table.  Interrupts are disabled.  The stack
+ * pointer (r1) points to just below the end of the half-meg region
+ * from 0x380000 - 0x400000, which is mapped in already.
+ *
+ * If we are booted from MacOS via BootX, we enter with the kernel
+ * image loaded somewhere, and the following values in registers:
+ *  r3: 'BooX' (0x426f6f58)
+ *  r4: virtual address of boot_infos_t
+ *  r5: 0
+ *
+ * APUS
+ *   r3: 'APUS'
+ *   r4: physical address of memory base
+ *   Linux/m68k style BootInfo structure at &_end.
+ *
+ * PREP
+ * This is jumped to on prep systems right after the kernel is relocated
+ * to its proper place in memory by the boot loader.  The expected layout
+ * of the regs is:
+ *   r3: ptr to residual data
+ *   r4: initrd_start or if no initrd then 0
+ *   r5: initrd_end - unused if r4 is 0
+ *   r6: Start of command line string
+ *   r7: End of command line string
+ *
+ * This just gets a minimal mmu environment setup so we can call
+ * start_here() to do the real work.
+ * -- Cort
+ */
+
+       .globl  __start
+__start:
+/*
+ * We have to do any OF calls before we map ourselves to KERNELBASE,
+ * because OF may have I/O devices mapped into that area
+ * (particularly on CHRP).
+ */
+       mr      r31,r3                  /* save parameters */
+       mr      r30,r4
+       mr      r29,r5
+       mr      r28,r6
+       mr      r27,r7
+       li      r24,0                   /* cpu # */
+
+/*
+ * early_init() does the early machine identification and does
+ * the necessary low-level setup and clears the BSS
+ *  -- Cort <cort@fsmlabs.com>
+ */
+       bl      early_init
+
+/*
+ * On POWER4, we first need to tweak some CPU configuration registers
+ * like real mode cache inhibit or exception base
+ */
+#ifdef CONFIG_POWER4
+       bl      __970_cpu_preinit
+#endif /* CONFIG_POWER4 */
+
+#ifdef CONFIG_APUS
+/* On APUS the __va/__pa constants need to be set to the correct
+ * values before continuing.
+ */
+       mr      r4,r30
+       bl      fix_mem_constants
+#endif /* CONFIG_APUS */
+
+/* Switch MMU off, clear BATs and flush TLB. At this point, r3 contains
+ * the physical address we are running at, returned by early_init()
+ */
+       bl      mmu_off
+__after_mmu_off:
+#ifndef CONFIG_POWER4
+       bl      clear_bats
+       bl      flush_tlbs
+
+       bl      initial_bats
+#if !defined(CONFIG_APUS) && defined(CONFIG_BOOTX_TEXT)
+       bl      setup_disp_bat
+#endif
+#else /* CONFIG_POWER4 */
+       bl      reloc_offset
+       bl      initial_mm_power4
+#endif /* CONFIG_POWER4 */
+
+/*
+ * Call setup_cpu for CPU 0 and initialize 6xx Idle
+ */
+       bl      reloc_offset
+       li      r24,0                   /* cpu# */
+       bl      call_setup_cpu          /* Call setup_cpu for this CPU */
+#ifdef CONFIG_6xx
+       bl      reloc_offset
+       bl      init_idle_6xx
+#endif /* CONFIG_6xx */
+#ifdef CONFIG_POWER4
+       bl      reloc_offset
+       bl      init_idle_power4
+#endif /* CONFIG_POWER4 */
+
+
+#ifndef CONFIG_APUS
+/*
+ * We need to run with _start at physical address 0.
+ * On CHRP, we are loaded at 0x10000 since OF on CHRP uses
+ * the exception vectors at 0 (and therefore this copy
+ * overwrites OF's exception vectors with our own).
+ * If the MMU is already turned on, we copy stuff to KERNELBASE,
+ * otherwise we copy it to 0.
+ */
+       bl      reloc_offset
+       mr      r26,r3
+       addis   r4,r3,KERNELBASE@h      /* current address of _start */
+       cmpwi   0,r4,0                  /* are we already running at 0? */
+       bne     relocate_kernel
+#endif /* CONFIG_APUS */
+/*
+ * we now have the 1st 16M of ram mapped with the bats.
+ * prep needs the mmu to be turned on here, but pmac already has it on.
+ * this shouldn't bother the pmac since it just gets turned on again
+ * as we jump to our code at KERNELBASE. -- Cort
+ * Actually no, pmac doesn't have it on any more. BootX enters with MMU
+ * off, and in other cases, we now turn it off before changing BATs above.
+ */
+turn_on_mmu:
+       mfmsr   r0
+       ori     r0,r0,MSR_DR|MSR_IR
+       mtspr   SPRN_SRR1,r0
+       lis     r0,start_here@h
+       ori     r0,r0,start_here@l
+       mtspr   SPRN_SRR0,r0
+       SYNC
+       RFI                             /* enables MMU */
+
+/*
+ * We need __secondary_hold as a place to hold the other cpus on
+ * an SMP machine, even when we are running a UP kernel.
+ */
+       . = 0xc0                        /* for prep bootloader */
+       li      r3,1                    /* MTX only has 1 cpu */
+       .globl  __secondary_hold
+__secondary_hold:
+       /* tell the master we're here */
+       stw     r3,4(0)
+#ifdef CONFIG_SMP
+100:   lwz     r4,0(0)
+       /* wait until we're told to start */
+       cmpw    0,r4,r3
+       bne     100b
+       /* our cpu # was at addr 0 - go */
+       mr      r24,r3                  /* cpu # */
+       b       __secondary_start
+#else
+       b       .
+#endif /* CONFIG_SMP */
+
+/*
+ * Exception entry code.  This code runs with address translation
+ * turned off, i.e. using physical addresses.
+ * We assume sprg3 has the physical address of the current
+ * task's thread_struct.
+ */
+#define EXCEPTION_PROLOG       \
+       mtspr   SPRN_SPRG0,r10; \
+       mtspr   SPRN_SPRG1,r11; \
+       mfcr    r10;            \
+       EXCEPTION_PROLOG_1;     \
+       EXCEPTION_PROLOG_2
+
+#define EXCEPTION_PROLOG_1     \
+       mfspr   r11,SPRN_SRR1;          /* check whether user or kernel */ \
+       andi.   r11,r11,MSR_PR; \
+       tophys(r11,r1);                 /* use tophys(r1) if kernel */ \
+       beq     1f;             \
+       mfspr   r11,SPRN_SPRG3; \
+       lwz     r11,THREAD_INFO-THREAD(r11);    \
+       addi    r11,r11,THREAD_SIZE;    \
+       tophys(r11,r11);        \
+1:     subi    r11,r11,INT_FRAME_SIZE  /* alloc exc. frame */
+
+
+#define EXCEPTION_PROLOG_2     \
+       CLR_TOP32(r11);         \
+       stw     r10,_CCR(r11);          /* save registers */ \
+       stw     r12,GPR12(r11); \
+       stw     r9,GPR9(r11);   \
+       mfspr   r10,SPRN_SPRG0; \
+       stw     r10,GPR10(r11); \
+       mfspr   r12,SPRN_SPRG1; \
+       stw     r12,GPR11(r11); \
+       mflr    r10;            \
+       stw     r10,_LINK(r11); \
+       mfspr   r12,SPRN_SRR0;  \
+       mfspr   r9,SPRN_SRR1;   \
+       stw     r1,GPR1(r11);   \
+       stw     r1,0(r11);      \
+       tovirt(r1,r11);                 /* set new kernel sp */ \
+       li      r10,MSR_KERNEL & ~(MSR_IR|MSR_DR); /* can take exceptions */ \
+       MTMSRD(r10);                    /* (except for mach check in rtas) */ \
+       stw     r0,GPR0(r11);   \
+       SAVE_4GPRS(3, r11);     \
+       SAVE_2GPRS(7, r11)
+
+/*
+ * Note: code which follows this uses cr0.eq (set if from kernel),
+ * r11, r12 (SRR0), and r9 (SRR1).
+ *
+ * Note2: once we have set r1 we are in a position to take exceptions
+ * again, and we could thus set MSR:RI at that point.
+ */
+
+/*
+ * Exception vectors.
+ */
+#define EXCEPTION(n, label, hdlr, xfer)                \
+       . = n;                                  \
+label:                                         \
+       EXCEPTION_PROLOG;                       \
+       addi    r3,r1,STACK_FRAME_OVERHEAD;     \
+       xfer(n, hdlr)
+
+#define EXC_XFER_TEMPLATE(n, hdlr, trap, copyee, tfer, ret)    \
+       li      r10,trap;                                       \
+       stw     r10,TRAP(r11);                                  \
+       li      r10,MSR_KERNEL;                                 \
+       copyee(r10, r9);                                        \
+       bl      tfer;                                           \
+i##n:                                                          \
+       .long   hdlr;                                           \
+       .long   ret
+
+#define COPY_EE(d, s)          rlwimi d,s,0,16,16
+#define NOCOPY(d, s)
+
+#define EXC_XFER_STD(n, hdlr)          \
+       EXC_XFER_TEMPLATE(n, hdlr, n, NOCOPY, transfer_to_handler_full, \
+                         ret_from_except_full)
+
+#define EXC_XFER_LITE(n, hdlr)         \
+       EXC_XFER_TEMPLATE(n, hdlr, n+1, NOCOPY, transfer_to_handler, \
+                         ret_from_except)
+
+#define EXC_XFER_EE(n, hdlr)           \
+       EXC_XFER_TEMPLATE(n, hdlr, n, COPY_EE, transfer_to_handler_full, \
+                         ret_from_except_full)
+
+#define EXC_XFER_EE_LITE(n, hdlr)      \
+       EXC_XFER_TEMPLATE(n, hdlr, n+1, COPY_EE, transfer_to_handler, \
+                         ret_from_except)
+
+/* System reset */
+/* core99 pmac starts the seconary here by changing the vector, and
+   putting it back to what it was (UnknownException) when done.  */
+#if defined(CONFIG_GEMINI) && defined(CONFIG_SMP)
+       . = 0x100
+       b       __secondary_start_gemini
+#else
+       EXCEPTION(0x100, Reset, UnknownException, EXC_XFER_STD)
+#endif
+
+/* Machine check */
+/*
+ * On CHRP, this is complicated by the fact that we could get a
+ * machine check inside RTAS, and we have no guarantee that certain
+ * critical registers will have the values we expect.  The set of
+ * registers that might have bad values includes all the GPRs
+ * and all the BATs.  We indicate that we are in RTAS by putting
+ * a non-zero value, the address of the exception frame to use,
+ * in SPRG2.  The machine check handler checks SPRG2 and uses its
+ * value if it is non-zero.  If we ever needed to free up SPRG2,
+ * we could use a field in the thread_info or thread_struct instead.
+ * (Other exception handlers assume that r1 is a valid kernel stack
+ * pointer when we take an exception from supervisor mode.)
+ *     -- paulus.
+ */
+       . = 0x200
+       mtspr   SPRN_SPRG0,r10
+       mtspr   SPRN_SPRG1,r11
+       mfcr    r10
+#ifdef CONFIG_PPC_CHRP
+       mfspr   r11,SPRN_SPRG2
+       cmpwi   0,r11,0
+       bne     7f
+#endif /* CONFIG_PPC_CHRP */
+       EXCEPTION_PROLOG_1
+7:     EXCEPTION_PROLOG_2
+       addi    r3,r1,STACK_FRAME_OVERHEAD
+#ifdef CONFIG_PPC_CHRP
+       mfspr   r4,SPRN_SPRG2
+       cmpwi   cr1,r4,0
+       bne     cr1,1f
+#endif
+       EXC_XFER_STD(0x200, MachineCheckException)
+#ifdef CONFIG_PPC_CHRP
+1:     b       machine_check_in_rtas
+#endif
+
+/* Data access exception. */
+       . = 0x300
+#ifdef CONFIG_PPC64BRIDGE
+       b       DataAccess
+DataAccessCont:
+#else
+DataAccess:
+       EXCEPTION_PROLOG
+#endif /* CONFIG_PPC64BRIDGE */
+       mfspr   r10,SPRN_DSISR
+       andis.  r0,r10,0xa470           /* weird error? */
+       bne     1f                      /* if not, try to put a PTE */
+       mfspr   r4,SPRN_DAR             /* into the hash table */
+       rlwinm  r3,r10,32-15,21,21      /* DSISR_STORE -> _PAGE_RW */
+       bl      hash_page
+1:     stw     r10,_DSISR(r11)
+       mr      r5,r10
+       mfspr   r4,SPRN_DAR
+       EXC_XFER_EE_LITE(0x300, handle_page_fault)
+
+#ifdef CONFIG_PPC64BRIDGE
+/* SLB fault on data access. */
+       . = 0x380
+       b       DataSegment
+#endif /* CONFIG_PPC64BRIDGE */
+
+/* Instruction access exception. */
+       . = 0x400
+#ifdef CONFIG_PPC64BRIDGE
+       b       InstructionAccess
+InstructionAccessCont:
+#else
+InstructionAccess:
+       EXCEPTION_PROLOG
+#endif /* CONFIG_PPC64BRIDGE */
+       andis.  r0,r9,0x4000            /* no pte found? */
+       beq     1f                      /* if so, try to put a PTE */
+       li      r3,0                    /* into the hash table */
+       mr      r4,r12                  /* SRR0 is fault address */
+       bl      hash_page
+1:     mr      r4,r12
+       mr      r5,r9
+       EXC_XFER_EE_LITE(0x400, handle_page_fault)
+
+#ifdef CONFIG_PPC64BRIDGE
+/* SLB fault on instruction access. */
+       . = 0x480
+       b       InstructionSegment
+#endif /* CONFIG_PPC64BRIDGE */
+
+/* External interrupt */
+       EXCEPTION(0x500, HardwareInterrupt, do_IRQ, EXC_XFER_LITE)
+
+/* Alignment exception */
+       . = 0x600
+Alignment:
+       EXCEPTION_PROLOG
+       mfspr   r4,SPRN_DAR
+       stw     r4,_DAR(r11)
+       mfspr   r5,SPRN_DSISR
+       stw     r5,_DSISR(r11)
+       addi    r3,r1,STACK_FRAME_OVERHEAD
+       EXC_XFER_EE(0x600, AlignmentException)
+
+/* Program check exception */
+       EXCEPTION(0x700, ProgramCheck, ProgramCheckException, EXC_XFER_STD)
+
+/* Floating-point unavailable */
+       . = 0x800
+FPUnavailable:
+       EXCEPTION_PROLOG
+       bne     load_up_fpu             /* if from user, just load it up */
+       addi    r3,r1,STACK_FRAME_OVERHEAD
+       EXC_XFER_EE_LITE(0x800, KernelFP)
+
+/* Decrementer */
+       EXCEPTION(0x900, Decrementer, timer_interrupt, EXC_XFER_LITE)
+
+       EXCEPTION(0xa00, Trap_0a, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0xb00, Trap_0b, UnknownException, EXC_XFER_EE)
+
+/* System call */
+       . = 0xc00
+SystemCall:
+       EXCEPTION_PROLOG
+       EXC_XFER_EE_LITE(0xc00, DoSyscall)
+
+/* Single step - not used on 601 */
+       EXCEPTION(0xd00, SingleStep, SingleStepException, EXC_XFER_STD)
+       EXCEPTION(0xe00, Trap_0e, UnknownException, EXC_XFER_EE)
+
+/*
+ * The Altivec unavailable trap is at 0x0f20.  Foo.
+ * We effectively remap it to 0x3000.
+ * We include an altivec unavailable exception vector even if
+ * not configured for Altivec, so that you can't panic a
+ * non-altivec kernel running on a machine with altivec just
+ * by executing an altivec instruction.
+ */
+       . = 0xf00
+       b       Trap_0f
+
+       . = 0xf20
+       b       AltiVecUnavailable
+
+Trap_0f:
+       EXCEPTION_PROLOG
+       addi    r3,r1,STACK_FRAME_OVERHEAD
+       EXC_XFER_EE(0xf00, UnknownException)
+
+/*
+ * Handle TLB miss for instruction on 603/603e.
+ * Note: we get an alternate set of r0 - r3 to use automatically.
+ */
+       . = 0x1000
+InstructionTLBMiss:
+/*
+ * r0: stored ctr
+ * r1: linux style pte ( later becomes ppc hardware pte )
+ * r2: ptr to linux-style pte
+ * r3: scratch
+ */
+       mfctr   r0
+       /* Get PTE (linux-style) and check access */
+       mfspr   r3,SPRN_IMISS
+       lis     r1,KERNELBASE@h         /* check if kernel address */
+       cmplw   0,r3,r1
+       mfspr   r2,SPRN_SPRG3
+       li      r1,_PAGE_USER|_PAGE_PRESENT /* low addresses tested as user */
+       lwz     r2,PGDIR(r2)
+       blt+    112f
+       lis     r2,swapper_pg_dir@ha    /* if kernel address, use */
+       addi    r2,r2,swapper_pg_dir@l  /* kernel page table */
+       mfspr   r1,SPRN_SRR1            /* and MSR_PR bit from SRR1 */
+       rlwinm  r1,r1,32-12,29,29       /* shift MSR_PR to _PAGE_USER posn */
+112:   tophys(r2,r2)
+       rlwimi  r2,r3,12,20,29          /* insert top 10 bits of address */
+       lwz     r2,0(r2)                /* get pmd entry */
+       rlwinm. r2,r2,0,0,19            /* extract address of pte page */
+       beq-    InstructionAddressInvalid       /* return if no mapping */
+       rlwimi  r2,r3,22,20,29          /* insert next 10 bits of address */
+       lwz     r3,0(r2)                /* get linux-style pte */
+       andc.   r1,r1,r3                /* check access & ~permission */
+       bne-    InstructionAddressInvalid /* return if access not permitted */
+       ori     r3,r3,_PAGE_ACCESSED    /* set _PAGE_ACCESSED in pte */
+       /*
+        * NOTE! We are assuming this is not an SMP system, otherwise
+        * we would need to update the pte atomically with lwarx/stwcx.
+        */
+       stw     r3,0(r2)                /* update PTE (accessed bit) */
+       /* Convert linux-style PTE to low word of PPC-style PTE */
+       rlwinm  r1,r3,32-10,31,31       /* _PAGE_RW -> PP lsb */
+       rlwinm  r2,r3,32-7,31,31        /* _PAGE_DIRTY -> PP lsb */
+       and     r1,r1,r2                /* writable if _RW and _DIRTY */
+       rlwimi  r3,r3,32-1,30,30        /* _PAGE_USER -> PP msb */
+       rlwimi  r3,r3,32-1,31,31        /* _PAGE_USER -> PP lsb */
+       ori     r1,r1,0xe14             /* clear out reserved bits and M */
+       andc    r1,r3,r1                /* PP = user? (rw&dirty? 2: 3): 0 */
+       mtspr   SPRN_RPA,r1
+       mfspr   r3,SPRN_IMISS
+       tlbli   r3
+       mfspr   r3,SPRN_SRR1            /* Need to restore CR0 */
+       mtcrf   0x80,r3
+       rfi
+InstructionAddressInvalid:
+       mfspr   r3,SPRN_SRR1
+       rlwinm  r1,r3,9,6,6     /* Get load/store bit */
+
+       addis   r1,r1,0x2000
+       mtspr   SPRN_DSISR,r1   /* (shouldn't be needed) */
+       mtctr   r0              /* Restore CTR */
+       andi.   r2,r3,0xFFFF    /* Clear upper bits of SRR1 */
+       or      r2,r2,r1
+       mtspr   SPRN_SRR1,r2
+       mfspr   r1,SPRN_IMISS   /* Get failing address */
+       rlwinm. r2,r2,0,31,31   /* Check for little endian access */
+       rlwimi  r2,r2,1,30,30   /* change 1 -> 3 */
+       xor     r1,r1,r2
+       mtspr   SPRN_DAR,r1     /* Set fault address */
+       mfmsr   r0              /* Restore "normal" registers */
+       xoris   r0,r0,MSR_TGPR>>16
+       mtcrf   0x80,r3         /* Restore CR0 */
+       mtmsr   r0
+       b       InstructionAccess
+
+/*
+ * Handle TLB miss for DATA Load operation on 603/603e
+ */
+       . = 0x1100
+DataLoadTLBMiss:
+/*
+ * r0: stored ctr
+ * r1: linux style pte ( later becomes ppc hardware pte )
+ * r2: ptr to linux-style pte
+ * r3: scratch
+ */
+       mfctr   r0
+       /* Get PTE (linux-style) and check access */
+       mfspr   r3,SPRN_DMISS
+       lis     r1,KERNELBASE@h         /* check if kernel address */
+       cmplw   0,r3,r1
+       mfspr   r2,SPRN_SPRG3
+       li      r1,_PAGE_USER|_PAGE_PRESENT /* low addresses tested as user */
+       lwz     r2,PGDIR(r2)
+       blt+    112f
+       lis     r2,swapper_pg_dir@ha    /* if kernel address, use */
+       addi    r2,r2,swapper_pg_dir@l  /* kernel page table */
+       mfspr   r1,SPRN_SRR1            /* and MSR_PR bit from SRR1 */
+       rlwinm  r1,r1,32-12,29,29       /* shift MSR_PR to _PAGE_USER posn */
+112:   tophys(r2,r2)
+       rlwimi  r2,r3,12,20,29          /* insert top 10 bits of address */
+       lwz     r2,0(r2)                /* get pmd entry */
+       rlwinm. r2,r2,0,0,19            /* extract address of pte page */
+       beq-    DataAddressInvalid      /* return if no mapping */
+       rlwimi  r2,r3,22,20,29          /* insert next 10 bits of address */
+       lwz     r3,0(r2)                /* get linux-style pte */
+       andc.   r1,r1,r3                /* check access & ~permission */
+       bne-    DataAddressInvalid      /* return if access not permitted */
+       ori     r3,r3,_PAGE_ACCESSED    /* set _PAGE_ACCESSED in pte */
+       /*
+        * NOTE! We are assuming this is not an SMP system, otherwise
+        * we would need to update the pte atomically with lwarx/stwcx.
+        */
+       stw     r3,0(r2)                /* update PTE (accessed bit) */
+       /* Convert linux-style PTE to low word of PPC-style PTE */
+       rlwinm  r1,r3,32-10,31,31       /* _PAGE_RW -> PP lsb */
+       rlwinm  r2,r3,32-7,31,31        /* _PAGE_DIRTY -> PP lsb */
+       and     r1,r1,r2                /* writable if _RW and _DIRTY */
+       rlwimi  r3,r3,32-1,30,30        /* _PAGE_USER -> PP msb */
+       rlwimi  r3,r3,32-1,31,31        /* _PAGE_USER -> PP lsb */
+       ori     r1,r1,0xe14             /* clear out reserved bits and M */
+       andc    r1,r3,r1                /* PP = user? (rw&dirty? 2: 3): 0 */
+       mtspr   SPRN_RPA,r1
+       mfspr   r3,SPRN_DMISS
+       tlbld   r3
+       mfspr   r3,SPRN_SRR1            /* Need to restore CR0 */
+       mtcrf   0x80,r3
+       rfi
+DataAddressInvalid:
+       mfspr   r3,SPRN_SRR1
+       rlwinm  r1,r3,9,6,6     /* Get load/store bit */
+       addis   r1,r1,0x2000
+       mtspr   SPRN_DSISR,r1
+       mtctr   r0              /* Restore CTR */
+       andi.   r2,r3,0xFFFF    /* Clear upper bits of SRR1 */
+       mtspr   SPRN_SRR1,r2
+       mfspr   r1,SPRN_DMISS   /* Get failing address */
+       rlwinm. r2,r2,0,31,31   /* Check for little endian access */
+       beq     20f             /* Jump if big endian */
+       xori    r1,r1,3
+20:    mtspr   SPRN_DAR,r1     /* Set fault address */
+       mfmsr   r0              /* Restore "normal" registers */
+       xoris   r0,r0,MSR_TGPR>>16
+       mtcrf   0x80,r3         /* Restore CR0 */
+       mtmsr   r0
+       b       DataAccess
+
+/*
+ * Handle TLB miss for DATA Store on 603/603e
+ */
+       . = 0x1200
+DataStoreTLBMiss:
+/*
+ * r0: stored ctr
+ * r1: linux style pte ( later becomes ppc hardware pte )
+ * r2: ptr to linux-style pte
+ * r3: scratch
+ */
+       mfctr   r0
+       /* Get PTE (linux-style) and check access */
+       mfspr   r3,SPRN_DMISS
+       lis     r1,KERNELBASE@h         /* check if kernel address */
+       cmplw   0,r3,r1
+       mfspr   r2,SPRN_SPRG3
+       li      r1,_PAGE_RW|_PAGE_USER|_PAGE_PRESENT /* access flags */
+       lwz     r2,PGDIR(r2)
+       blt+    112f
+       lis     r2,swapper_pg_dir@ha    /* if kernel address, use */
+       addi    r2,r2,swapper_pg_dir@l  /* kernel page table */
+       mfspr   r1,SPRN_SRR1            /* and MSR_PR bit from SRR1 */
+       rlwinm  r1,r1,32-12,29,29       /* shift MSR_PR to _PAGE_USER posn */
+112:   tophys(r2,r2)
+       rlwimi  r2,r3,12,20,29          /* insert top 10 bits of address */
+       lwz     r2,0(r2)                /* get pmd entry */
+       rlwinm. r2,r2,0,0,19            /* extract address of pte page */
+       beq-    DataAddressInvalid      /* return if no mapping */
+       rlwimi  r2,r3,22,20,29          /* insert next 10 bits of address */
+       lwz     r3,0(r2)                /* get linux-style pte */
+       andc.   r1,r1,r3                /* check access & ~permission */
+       bne-    DataAddressInvalid      /* return if access not permitted */
+       ori     r3,r3,_PAGE_ACCESSED|_PAGE_DIRTY
+       /*
+        * NOTE! We are assuming this is not an SMP system, otherwise
+        * we would need to update the pte atomically with lwarx/stwcx.
+        */
+       stw     r3,0(r2)                /* update PTE (accessed/dirty bits) */
+       /* Convert linux-style PTE to low word of PPC-style PTE */
+       rlwimi  r3,r3,32-1,30,30        /* _PAGE_USER -> PP msb */
+       li      r1,0xe15                /* clear out reserved bits and M */
+       andc    r1,r3,r1                /* PP = user? 2: 0 */
+       mtspr   SPRN_RPA,r1
+       mfspr   r3,SPRN_DMISS
+       tlbld   r3
+       mfspr   r3,SPRN_SRR1            /* Need to restore CR0 */
+       mtcrf   0x80,r3
+       rfi
+
+#ifndef CONFIG_ALTIVEC
+#define AltivecAssistException UnknownException
+#endif
+
+       EXCEPTION(0x1300, Trap_13, InstructionBreakpoint, EXC_XFER_EE)
+       EXCEPTION(0x1400, SMI, SMIException, EXC_XFER_EE)
+       EXCEPTION(0x1500, Trap_15, UnknownException, EXC_XFER_EE)
+#ifdef CONFIG_POWER4
+       EXCEPTION(0x1600, Trap_16, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x1700, Trap_17, AltivecAssistException, EXC_XFER_EE)
+       EXCEPTION(0x1800, Trap_18, TAUException, EXC_XFER_STD)
+#else /* !CONFIG_POWER4 */
+       EXCEPTION(0x1600, Trap_16, AltivecAssistException, EXC_XFER_EE)
+       EXCEPTION(0x1700, Trap_17, TAUException, EXC_XFER_STD)
+       EXCEPTION(0x1800, Trap_18, UnknownException, EXC_XFER_EE)
+#endif /* CONFIG_POWER4 */
+       EXCEPTION(0x1900, Trap_19, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x1a00, Trap_1a, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x1b00, Trap_1b, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x1c00, Trap_1c, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x1d00, Trap_1d, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x1e00, Trap_1e, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x1f00, Trap_1f, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x2000, RunMode, RunModeException, EXC_XFER_EE)
+       EXCEPTION(0x2100, Trap_21, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x2200, Trap_22, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x2300, Trap_23, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x2400, Trap_24, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x2500, Trap_25, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x2600, Trap_26, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x2700, Trap_27, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x2800, Trap_28, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x2900, Trap_29, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x2a00, Trap_2a, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x2b00, Trap_2b, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x2c00, Trap_2c, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x2d00, Trap_2d, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x2e00, Trap_2e, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x2f00, MOLTrampoline, UnknownException, EXC_XFER_EE_LITE)
+
+       .globl mol_trampoline
+       .set mol_trampoline, i0x2f00
+
+       . = 0x3000
+
+AltiVecUnavailable:
+       EXCEPTION_PROLOG
+#ifdef CONFIG_ALTIVEC
+       bne     load_up_altivec         /* if from user, just load it up */
+#endif /* CONFIG_ALTIVEC */
+       EXC_XFER_EE_LITE(0xf20, AltivecUnavailException)
+
+#ifdef CONFIG_PPC64BRIDGE
+DataAccess:
+       EXCEPTION_PROLOG
+       b       DataAccessCont
+
+InstructionAccess:
+       EXCEPTION_PROLOG
+       b       InstructionAccessCont
+
+DataSegment:
+       EXCEPTION_PROLOG
+       addi    r3,r1,STACK_FRAME_OVERHEAD
+       mfspr   r4,SPRN_DAR
+       stw     r4,_DAR(r11)
+       EXC_XFER_STD(0x380, UnknownException)
+
+InstructionSegment:
+       EXCEPTION_PROLOG
+       addi    r3,r1,STACK_FRAME_OVERHEAD
+       EXC_XFER_STD(0x480, UnknownException)
+#endif /* CONFIG_PPC64BRIDGE */
+
+#ifdef CONFIG_ALTIVEC
+/* Note that the AltiVec support is closely modeled after the FP
+ * support.  Changes to one are likely to be applicable to the
+ * other!  */
+load_up_altivec:
+/*
+ * Disable AltiVec for the task which had AltiVec previously,
+ * and save its AltiVec registers in its thread_struct.
+ * Enables AltiVec for use in the kernel on return.
+ * On SMP we know the AltiVec units are free, since we give it up every
+ * switch.  -- Kumar
+ */
+       mfmsr   r5
+       oris    r5,r5,MSR_VEC@h
+       MTMSRD(r5)                      /* enable use of AltiVec now */
+       isync
+/*
+ * For SMP, we don't do lazy AltiVec switching because it just gets too
+ * horrendously complex, especially when a task switches from one CPU
+ * to another.  Instead we call giveup_altivec in switch_to.
+ */
+#ifndef CONFIG_SMP
+       tophys(r6,0)
+       addis   r3,r6,last_task_used_altivec@ha
+       lwz     r4,last_task_used_altivec@l(r3)
+       cmpwi   0,r4,0
+       beq     1f
+       add     r4,r4,r6
+       addi    r4,r4,THREAD    /* want THREAD of last_task_used_altivec */
+       SAVE_32VRS(0,r10,r4)
+       mfvscr  vr0
+       li      r10,THREAD_VSCR
+       stvx    vr0,r10,r4
+       lwz     r5,PT_REGS(r4)
+       add     r5,r5,r6
+       lwz     r4,_MSR-STACK_FRAME_OVERHEAD(r5)
+       lis     r10,MSR_VEC@h
+       andc    r4,r4,r10       /* disable altivec for previous task */
+       stw     r4,_MSR-STACK_FRAME_OVERHEAD(r5)
+1:
+#endif /* CONFIG_SMP */
+       /* enable use of AltiVec after return */
+       oris    r9,r9,MSR_VEC@h
+       mfspr   r5,SPRN_SPRG3           /* current task's THREAD (phys) */
+       li      r4,1
+       li      r10,THREAD_VSCR
+       stw     r4,THREAD_USED_VR(r5)
+       lvx     vr0,r10,r5
+       mtvscr  vr0
+       REST_32VRS(0,r10,r5)
+#ifndef CONFIG_SMP
+       subi    r4,r5,THREAD
+       sub     r4,r4,r6
+       stw     r4,last_task_used_altivec@l(r3)
+#endif /* CONFIG_SMP */
+       /* restore registers and return */
+       /* we haven't used ctr or xer or lr */
+       b       fast_exception_return
+
+/*
+ * AltiVec unavailable trap from kernel - print a message, but let
+ * the task use AltiVec in the kernel until it returns to user mode.
+ */
+KernelAltiVec:
+       lwz     r3,_MSR(r1)
+       oris    r3,r3,MSR_VEC@h
+       stw     r3,_MSR(r1)     /* enable use of AltiVec after return */
+       lis     r3,87f@h
+       ori     r3,r3,87f@l
+       mr      r4,r2           /* current */
+       lwz     r5,_NIP(r1)
+       bl      printk
+       b       ret_from_except
+87:    .string "AltiVec used in kernel  (task=%p, pc=%x)  \n"
+       .align  4,0
+
+/*
+ * giveup_altivec(tsk)
+ * Disable AltiVec for the task given as the argument,
+ * and save the AltiVec registers in its thread_struct.
+ * Enables AltiVec for use in the kernel on return.
+ */
+
+       .globl  giveup_altivec
+giveup_altivec:
+       mfmsr   r5
+       oris    r5,r5,MSR_VEC@h
+       SYNC
+       MTMSRD(r5)                      /* enable use of AltiVec now */
+       isync
+       cmpwi   0,r3,0
+       beqlr-                          /* if no previous owner, done */
+       addi    r3,r3,THREAD            /* want THREAD of task */
+       lwz     r5,PT_REGS(r3)
+       cmpwi   0,r5,0
+       SAVE_32VRS(0, r4, r3)
+       mfvscr  vr0
+       li      r4,THREAD_VSCR
+       stvx    vr0,r4,r3
+       beq     1f
+       lwz     r4,_MSR-STACK_FRAME_OVERHEAD(r5)
+       lis     r3,MSR_VEC@h
+       andc    r4,r4,r3                /* disable AltiVec for previous task */
+       stw     r4,_MSR-STACK_FRAME_OVERHEAD(r5)
+1:
+#ifndef CONFIG_SMP
+       li      r5,0
+       lis     r4,last_task_used_altivec@ha
+       stw     r5,last_task_used_altivec@l(r4)
+#endif /* CONFIG_SMP */
+       blr
+#endif /* CONFIG_ALTIVEC */
+
+/*
+ * This code is jumped to from the startup code to copy
+ * the kernel image to physical address 0.
+ */
+relocate_kernel:
+       addis   r9,r26,klimit@ha        /* fetch klimit */
+       lwz     r25,klimit@l(r9)
+       addis   r25,r25,-KERNELBASE@h
+       li      r3,0                    /* Destination base address */
+       li      r6,0                    /* Destination offset */
+       li      r5,0x4000               /* # bytes of memory to copy */
+       bl      copy_and_flush          /* copy the first 0x4000 bytes */
+       addi    r0,r3,4f@l              /* jump to the address of 4f */
+       mtctr   r0                      /* in copy and do the rest. */
+       bctr                            /* jump to the copy */
+4:     mr      r5,r25
+       bl      copy_and_flush          /* copy the rest */
+       b       turn_on_mmu
+
+/*
+ * Copy routine used to copy the kernel to start at physical address 0
+ * and flush and invalidate the caches as needed.
+ * r3 = dest addr, r4 = source addr, r5 = copy limit, r6 = start offset
+ * on exit, r3, r4, r5 are unchanged, r6 is updated to be >= r5.
+ */
+copy_and_flush:
+       addi    r5,r5,-4
+       addi    r6,r6,-4
+4:     li      r0,L1_CACHE_LINE_SIZE/4
+       mtctr   r0
+3:     addi    r6,r6,4                 /* copy a cache line */
+       lwzx    r0,r6,r4
+       stwx    r0,r6,r3
+       bdnz    3b
+       dcbst   r6,r3                   /* write it to memory */
+       sync
+       icbi    r6,r3                   /* flush the icache line */
+       cmplw   0,r6,r5
+       blt     4b
+       sync                            /* additional sync needed on g4 */
+       isync
+       addi    r5,r5,4
+       addi    r6,r6,4
+       blr
+
+#ifdef CONFIG_APUS
+/*
+ * On APUS the physical base address of the kernel is not known at compile
+ * time, which means the __pa/__va constants used are incorrect. In the
+ * __init section is recorded the virtual addresses of instructions using
+ * these constants, so all that has to be done is fix these before
+ * continuing the kernel boot.
+ *
+ * r4 = The physical address of the kernel base.
+ */
+fix_mem_constants:
+       mr      r10,r4
+       addis   r10,r10,-KERNELBASE@h    /* virt_to_phys constant */
+       neg     r11,r10                  /* phys_to_virt constant */
+
+       lis     r12,__vtop_table_begin@h
+       ori     r12,r12,__vtop_table_begin@l
+       add     r12,r12,r10              /* table begin phys address */
+       lis     r13,__vtop_table_end@h
+       ori     r13,r13,__vtop_table_end@l
+       add     r13,r13,r10              /* table end phys address */
+       subi    r12,r12,4
+       subi    r13,r13,4
+1:     lwzu    r14,4(r12)               /* virt address of instruction */
+       add     r14,r14,r10              /* phys address of instruction */
+       lwz     r15,0(r14)               /* instruction, now insert top */
+       rlwimi  r15,r10,16,16,31         /* half of vp const in low half */
+       stw     r15,0(r14)               /* of instruction and restore. */
+       dcbst   r0,r14                   /* write it to memory */
+       sync
+       icbi    r0,r14                   /* flush the icache line */
+       cmpw    r12,r13
+       bne     1b
+       sync                            /* additional sync needed on g4 */
+       isync
+
+/*
+ * Map the memory where the exception handlers will
+ * be copied to when hash constants have been patched.
+ */
+#ifdef CONFIG_APUS_FAST_EXCEPT
+       lis     r8,0xfff0
+#else
+       lis     r8,0
+#endif
+       ori     r8,r8,0x2               /* 128KB, supervisor */
+       mtspr   SPRN_DBAT3U,r8
+       mtspr   SPRN_DBAT3L,r8
+
+       lis     r12,__ptov_table_begin@h
+       ori     r12,r12,__ptov_table_begin@l
+       add     r12,r12,r10              /* table begin phys address */
+       lis     r13,__ptov_table_end@h
+       ori     r13,r13,__ptov_table_end@l
+       add     r13,r13,r10              /* table end phys address */
+       subi    r12,r12,4
+       subi    r13,r13,4
+1:     lwzu    r14,4(r12)               /* virt address of instruction */
+       add     r14,r14,r10              /* phys address of instruction */
+       lwz     r15,0(r14)               /* instruction, now insert top */
+       rlwimi  r15,r11,16,16,31         /* half of pv const in low half*/
+       stw     r15,0(r14)               /* of instruction and restore. */
+       dcbst   r0,r14                   /* write it to memory */
+       sync
+       icbi    r0,r14                   /* flush the icache line */
+       cmpw    r12,r13
+       bne     1b
+
+       sync                            /* additional sync needed on g4 */
+       isync                           /* No speculative loading until now */
+       blr
+
+/***********************************************************************
+ *  Please note that on APUS the exception handlers are located at the
+ *  physical address 0xfff0000. For this reason, the exception handlers
+ *  cannot use relative branches to access the code below.
+ ***********************************************************************/
+#endif /* CONFIG_APUS */
+
+#ifdef CONFIG_SMP
+#ifdef CONFIG_GEMINI
+       .globl  __secondary_start_gemini
+__secondary_start_gemini:
+        mfspr   r4,SPRN_HID0
+        ori     r4,r4,HID0_ICFI
+        li      r3,0
+        ori     r3,r3,HID0_ICE
+        andc    r4,r4,r3
+        mtspr   SPRN_HID0,r4
+        sync
+        b       __secondary_start
+#endif /* CONFIG_GEMINI */
+
+       .globl  __secondary_start_pmac_0
+__secondary_start_pmac_0:
+       /* NB the entries for cpus 0, 1, 2 must each occupy 8 bytes. */
+       li      r24,0
+       b       1f
+       li      r24,1
+       b       1f
+       li      r24,2
+       b       1f
+       li      r24,3
+1:
+       /* on powersurge, we come in here with IR=0 and DR=1, and DBAT 0
+          set to map the 0xf0000000 - 0xffffffff region */
+       mfmsr   r0
+       rlwinm  r0,r0,0,28,26           /* clear DR (0x10) */
+       SYNC
+       mtmsr   r0
+       isync
+
+       .globl  __secondary_start
+__secondary_start:
+#ifdef CONFIG_PPC64BRIDGE
+       mfmsr   r0
+       clrldi  r0,r0,1                 /* make sure it's in 32-bit mode */
+       SYNC
+       MTMSRD(r0)
+       isync
+#endif
+       /* Copy some CPU settings from CPU 0 */
+       bl      __restore_cpu_setup
+
+       lis     r3,-KERNELBASE@h
+       mr      r4,r24
+       bl      identify_cpu
+       bl      call_setup_cpu          /* Call setup_cpu for this CPU */
+#ifdef CONFIG_6xx
+       lis     r3,-KERNELBASE@h
+       bl      init_idle_6xx
+#endif /* CONFIG_6xx */
+#ifdef CONFIG_POWER4
+       lis     r3,-KERNELBASE@h
+       bl      init_idle_power4
+#endif /* CONFIG_POWER4 */
+
+       /* get current_thread_info and current */
+       lis     r1,secondary_ti@ha
+       tophys(r1,r1)
+       lwz     r1,secondary_ti@l(r1)
+       tophys(r2,r1)
+       lwz     r2,TI_TASK(r2)
+
+       /* stack */
+       addi    r1,r1,THREAD_SIZE-STACK_FRAME_OVERHEAD
+       li      r0,0
+       tophys(r3,r1)
+       stw     r0,0(r3)
+
+       /* load up the MMU */
+       bl      load_up_mmu
+
+       /* ptr to phys current thread */
+       tophys(r4,r2)
+       addi    r4,r4,THREAD    /* phys address of our thread_struct */
+       CLR_TOP32(r4)
+       mtspr   SPRN_SPRG3,r4
+       li      r3,0
+       mtspr   SPRN_SPRG2,r3   /* 0 => not in RTAS */
+
+       /* enable MMU and jump to start_secondary */
+       li      r4,MSR_KERNEL
+       FIX_SRR1(r4,r5)
+       lis     r3,start_secondary@h
+       ori     r3,r3,start_secondary@l
+       mtspr   SPRN_SRR0,r3
+       mtspr   SPRN_SRR1,r4
+       SYNC
+       RFI
+#endif /* CONFIG_SMP */
+
+/*
+ * Those generic dummy functions are kept for CPUs not
+ * included in CONFIG_6xx
+ */
+_GLOBAL(__setup_cpu_power3)
+       blr
+_GLOBAL(__setup_cpu_generic)
+       blr
+
+#if !defined(CONFIG_6xx) && !defined(CONFIG_POWER4)
+_GLOBAL(__save_cpu_setup)
+       blr
+_GLOBAL(__restore_cpu_setup)
+       blr
+#endif /* !defined(CONFIG_6xx) && !defined(CONFIG_POWER4) */
+
+
+/*
+ * Load stuff into the MMU.  Intended to be called with
+ * IR=0 and DR=0.
+ */
+load_up_mmu:
+       sync                    /* Force all PTE updates to finish */
+       isync
+       tlbia                   /* Clear all TLB entries */
+       sync                    /* wait for tlbia/tlbie to finish */
+       TLBSYNC                 /* ... on all CPUs */
+       /* Load the SDR1 register (hash table base & size) */
+       lis     r6,_SDR1@ha
+       tophys(r6,r6)
+       lwz     r6,_SDR1@l(r6)
+       mtspr   SPRN_SDR1,r6
+#ifdef CONFIG_PPC64BRIDGE
+       /* clear the ASR so we only use the pseudo-segment registers. */
+       li      r6,0
+       mtasr   r6
+#endif /* CONFIG_PPC64BRIDGE */
+       li      r0,16           /* load up segment register values */
+       mtctr   r0              /* for context 0 */
+       lis     r3,0x2000       /* Ku = 1, VSID = 0 */
+       li      r4,0
+3:     mtsrin  r3,r4
+       addi    r3,r3,0x111     /* increment VSID */
+       addis   r4,r4,0x1000    /* address of next segment */
+       bdnz    3b
+#ifndef CONFIG_POWER4
+/* Load the BAT registers with the values set up by MMU_init.
+   MMU_init takes care of whether we're on a 601 or not. */
+       mfpvr   r3
+       srwi    r3,r3,16
+       cmpwi   r3,1
+       lis     r3,BATS@ha
+       addi    r3,r3,BATS@l
+       tophys(r3,r3)
+       LOAD_BAT(0,r3,r4,r5)
+       LOAD_BAT(1,r3,r4,r5)
+       LOAD_BAT(2,r3,r4,r5)
+       LOAD_BAT(3,r3,r4,r5)
+#endif /* CONFIG_POWER4 */
+       blr
+
+/*
+ * This is where the main kernel code starts.
+ */
+start_here:
+       /* ptr to current */
+       lis     r2,init_task@h
+       ori     r2,r2,init_task@l
+       /* Set up for using our exception vectors */
+       /* ptr to phys current thread */
+       tophys(r4,r2)
+       addi    r4,r4,THREAD    /* init task's THREAD */
+       CLR_TOP32(r4)
+       mtspr   SPRN_SPRG3,r4
+       li      r3,0
+       mtspr   SPRN_SPRG2,r3   /* 0 => not in RTAS */
+
+       /* stack */
+       lis     r1,init_thread_union@ha
+       addi    r1,r1,init_thread_union@l
+       li      r0,0
+       stwu    r0,THREAD_SIZE-STACK_FRAME_OVERHEAD(r1)
+/*
+ * Do early bootinfo parsing, platform-specific initialization,
+ * and set up the MMU.
+ */
+       mr      r3,r31
+       mr      r4,r30
+       mr      r5,r29
+       mr      r6,r28
+       mr      r7,r27
+       bl      machine_init
+       bl      MMU_init
+
+#ifdef CONFIG_APUS
+       /* Copy exception code to exception vector base on APUS. */
+       lis     r4,KERNELBASE@h
+#ifdef CONFIG_APUS_FAST_EXCEPT
+       lis     r3,0xfff0               /* Copy to 0xfff00000 */
+#else
+       lis     r3,0                    /* Copy to 0x00000000 */
+#endif
+       li      r5,0x4000               /* # bytes of memory to copy */
+       li      r6,0
+       bl      copy_and_flush          /* copy the first 0x4000 bytes */
+#endif  /* CONFIG_APUS */
+
+/*
+ * Go back to running unmapped so we can load up new values
+ * for SDR1 (hash table pointer) and the segment registers
+ * and change to using our exception vectors.
+ */
+       lis     r4,2f@h
+       ori     r4,r4,2f@l
+       tophys(r4,r4)
+       li      r3,MSR_KERNEL & ~(MSR_IR|MSR_DR)
+       FIX_SRR1(r3,r5)
+       mtspr   SPRN_SRR0,r4
+       mtspr   SPRN_SRR1,r3
+       SYNC
+       RFI
+/* Load up the kernel context */
+2:     bl      load_up_mmu
+
+#ifdef CONFIG_BDI_SWITCH
+       /* Add helper information for the Abatron bdiGDB debugger.
+        * We do this here because we know the mmu is disabled, and
+        * will be enabled for real in just a few instructions.
+        */
+       lis     r5, abatron_pteptrs@h
+       ori     r5, r5, abatron_pteptrs@l
+       stw     r5, 0xf0(r0)    /* This much match your Abatron config */
+       lis     r6, swapper_pg_dir@h
+       ori     r6, r6, swapper_pg_dir@l
+       tophys(r5, r5)
+       stw     r6, 0(r5)
+#endif /* CONFIG_BDI_SWITCH */
+
+/* Now turn on the MMU for real! */
+       li      r4,MSR_KERNEL
+       FIX_SRR1(r4,r5)
+       lis     r3,start_kernel@h
+       ori     r3,r3,start_kernel@l
+       mtspr   SPRN_SRR0,r3
+       mtspr   SPRN_SRR1,r4
+       SYNC
+       RFI
+
+/*
+ * Set up the segment registers for a new context.
+ */
+_GLOBAL(set_context)
+       mulli   r3,r3,897       /* multiply context by skew factor */
+       rlwinm  r3,r3,4,8,27    /* VSID = (context & 0xfffff) << 4 */
+       addis   r3,r3,0x6000    /* Set Ks, Ku bits */
+       li      r0,NUM_USER_SEGMENTS
+       mtctr   r0
+
+#ifdef CONFIG_BDI_SWITCH
+       /* Context switch the PTE pointer for the Abatron BDI2000.
+        * The PGDIR is passed as second argument.
+        */
+       lis     r5, KERNELBASE@h
+       lwz     r5, 0xf0(r5)
+       stw     r4, 0x4(r5)
+#endif
+       li      r4,0
+       isync
+3:
+#ifdef CONFIG_PPC64BRIDGE
+       slbie   r4
+#endif /* CONFIG_PPC64BRIDGE */
+       mtsrin  r3,r4
+       addi    r3,r3,0x111     /* next VSID */
+       rlwinm  r3,r3,0,8,3     /* clear out any overflow from VSID field */
+       addis   r4,r4,0x1000    /* address of next segment */
+       bdnz    3b
+       sync
+       isync
+       blr
+
+/*
+ * An undocumented "feature" of 604e requires that the v bit
+ * be cleared before changing BAT values.
+ *
+ * Also, newer IBM firmware does not clear bat3 and 4 so
+ * this makes sure it's done.
+ *  -- Cort
+ */
+clear_bats:
+       li      r10,0
+       mfspr   r9,SPRN_PVR
+       rlwinm  r9,r9,16,16,31          /* r9 = 1 for 601, 4 for 604 */
+       cmpwi   r9, 1
+       beq     1f
+
+       mtspr   SPRN_DBAT0U,r10
+       mtspr   SPRN_DBAT0L,r10
+       mtspr   SPRN_DBAT1U,r10
+       mtspr   SPRN_DBAT1L,r10
+       mtspr   SPRN_DBAT2U,r10
+       mtspr   SPRN_DBAT2L,r10
+       mtspr   SPRN_DBAT3U,r10
+       mtspr   SPRN_DBAT3L,r10
+1:
+       mtspr   SPRN_IBAT0U,r10
+       mtspr   SPRN_IBAT0L,r10
+       mtspr   SPRN_IBAT1U,r10
+       mtspr   SPRN_IBAT1L,r10
+       mtspr   SPRN_IBAT2U,r10
+       mtspr   SPRN_IBAT2L,r10
+       mtspr   SPRN_IBAT3U,r10
+       mtspr   SPRN_IBAT3L,r10
+BEGIN_FTR_SECTION
+       /* Here's a tweak: at this point, CPU setup have
+        * not been called yet, so HIGH_BAT_EN may not be
+        * set in HID0 for the 745x processors. However, it
+        * seems that doesn't affect our ability to actually
+        * write to these SPRs.
+        */
+       mtspr   SPRN_DBAT4U,r10
+       mtspr   SPRN_DBAT4L,r10
+       mtspr   SPRN_DBAT5U,r10
+       mtspr   SPRN_DBAT5L,r10
+       mtspr   SPRN_DBAT6U,r10
+       mtspr   SPRN_DBAT6L,r10
+       mtspr   SPRN_DBAT7U,r10
+       mtspr   SPRN_DBAT7L,r10
+       mtspr   SPRN_IBAT4U,r10
+       mtspr   SPRN_IBAT4L,r10
+       mtspr   SPRN_IBAT5U,r10
+       mtspr   SPRN_IBAT5L,r10
+       mtspr   SPRN_IBAT6U,r10
+       mtspr   SPRN_IBAT6L,r10
+       mtspr   SPRN_IBAT7U,r10
+       mtspr   SPRN_IBAT7L,r10
+END_FTR_SECTION_IFSET(CPU_FTR_HAS_HIGH_BATS)
+       blr
+
+flush_tlbs:
+       lis     r10, 0x40
+1:     addic.  r10, r10, -0x1000
+       tlbie   r10
+       blt     1b
+       sync
+       blr
+
+mmu_off:
+       addi    r4, r3, __after_mmu_off - _start
+       mfmsr   r3
+       andi.   r0,r3,MSR_DR|MSR_IR             /* MMU enabled? */
+       beqlr
+       andc    r3,r3,r0
+       mtspr   SPRN_SRR0,r4
+       mtspr   SPRN_SRR1,r3
+       sync
+       RFI
+
+#ifndef CONFIG_POWER4
+/*
+ * Use the first pair of BAT registers to map the 1st 16MB
+ * of RAM to KERNELBASE.  From this point on we can't safely
+ * call OF any more.
+ */
+initial_bats:
+       lis     r11,KERNELBASE@h
+#ifndef CONFIG_PPC64BRIDGE
+       mfspr   r9,SPRN_PVR
+       rlwinm  r9,r9,16,16,31          /* r9 = 1 for 601, 4 for 604 */
+       cmpwi   0,r9,1
+       bne     4f
+       ori     r11,r11,4               /* set up BAT registers for 601 */
+       li      r8,0x7f                 /* valid, block length = 8MB */
+       oris    r9,r11,0x800000@h       /* set up BAT reg for 2nd 8M */
+       oris    r10,r8,0x800000@h       /* set up BAT reg for 2nd 8M */
+       mtspr   SPRN_IBAT0U,r11         /* N.B. 601 has valid bit in */
+       mtspr   SPRN_IBAT0L,r8          /* lower BAT register */
+       mtspr   SPRN_IBAT1U,r9
+       mtspr   SPRN_IBAT1L,r10
+       isync
+       blr
+#endif /* CONFIG_PPC64BRIDGE */
+
+4:     tophys(r8,r11)
+#ifdef CONFIG_SMP
+       ori     r8,r8,0x12              /* R/W access, M=1 */
+#else
+       ori     r8,r8,2                 /* R/W access */
+#endif /* CONFIG_SMP */
+#ifdef CONFIG_APUS
+       ori     r11,r11,BL_8M<<2|0x2    /* set up 8MB BAT registers for 604 */
+#else
+       ori     r11,r11,BL_256M<<2|0x2  /* set up BAT registers for 604 */
+#endif /* CONFIG_APUS */
+
+#ifdef CONFIG_PPC64BRIDGE
+       /* clear out the high 32 bits in the BAT */
+       clrldi  r11,r11,32
+       clrldi  r8,r8,32
+#endif /* CONFIG_PPC64BRIDGE */
+       mtspr   SPRN_DBAT0L,r8          /* N.B. 6xx (not 601) have valid */
+       mtspr   SPRN_DBAT0U,r11         /* bit in upper BAT register */
+       mtspr   SPRN_IBAT0L,r8
+       mtspr   SPRN_IBAT0U,r11
+       isync
+       blr
+
+#if !defined(CONFIG_APUS) && defined(CONFIG_BOOTX_TEXT)
+setup_disp_bat:
+       /*
+        * setup the display bat prepared for us in prom.c
+        */
+       mflr    r8
+       bl      reloc_offset
+       mtlr    r8
+       addis   r8,r3,disp_BAT@ha
+       addi    r8,r8,disp_BAT@l
+       lwz     r11,0(r8)
+       lwz     r8,4(r8)
+       mfspr   r9,SPRN_PVR
+       rlwinm  r9,r9,16,16,31          /* r9 = 1 for 601, 4 for 604 */
+       cmpwi   0,r9,1
+       beq     1f
+       mtspr   SPRN_DBAT3L,r8
+       mtspr   SPRN_DBAT3U,r11
+       blr
+1:     mtspr   SPRN_IBAT3L,r8
+       mtspr   SPRN_IBAT3U,r11
+       blr
+
+#endif /* !defined(CONFIG_APUS) && defined(CONFIG_BOOTX_TEXT) */
+
+#else /* CONFIG_POWER4 */
+/*
+ * Load up the SDR1 and segment register values now
+ * since we don't have the BATs.
+ * Also make sure we are running in 32-bit mode.
+ */
+
+initial_mm_power4:
+       addis   r14,r3,_SDR1@ha         /* get the value from _SDR1 */
+       lwz     r14,_SDR1@l(r14)        /* assume hash table below 4GB */
+       mtspr   SPRN_SDR1,r14
+       slbia
+       lis     r4,0x2000               /* set pseudo-segment reg 12 */
+       ori     r5,r4,0x0ccc
+       mtsr    12,r5
+#if 0
+       ori     r5,r4,0x0888            /* set pseudo-segment reg 8 */
+       mtsr    8,r5                    /* (for access to serial port) */
+#endif
+#ifdef CONFIG_BOOTX_TEXT
+       ori     r5,r4,0x0999            /* set pseudo-segment reg 9 */
+       mtsr    9,r5                    /* (for access to screen) */
+#endif
+       mfmsr   r0
+       clrldi  r0,r0,1
+       sync
+       mtmsr   r0
+       isync
+       blr
+
+#endif /* CONFIG_POWER4 */
+
+#ifdef CONFIG_8260
+/* Jump into the system reset for the rom.
+ * We first disable the MMU, and then jump to the ROM reset address.
+ *
+ * r3 is the board info structure, r4 is the location for starting.
+ * I use this for building a small kernel that can load other kernels,
+ * rather than trying to write or rely on a rom monitor that can tftp load.
+ */
+       .globl  m8260_gorom
+m8260_gorom:
+       mfmsr   r0
+       rlwinm  r0,r0,0,17,15   /* clear MSR_EE in r0 */
+       sync
+       mtmsr   r0
+       sync
+       mfspr   r11, SPRN_HID0
+       lis     r10, 0
+       ori     r10,r10,HID0_ICE|HID0_DCE
+       andc    r11, r11, r10
+       mtspr   SPRN_HID0, r11
+       isync
+       li      r5, MSR_ME|MSR_RI
+       lis     r6,2f@h
+       addis   r6,r6,-KERNELBASE@h
+       ori     r6,r6,2f@l
+       mtspr   SPRN_SRR0,r6
+       mtspr   SPRN_SRR1,r5
+       isync
+       sync
+       rfi
+2:
+       mtlr    r4
+       blr
+#endif
+
+
+/*
+ * We put a few things here that have to be page-aligned.
+ * This stuff goes at the beginning of the data segment,
+ * which is page-aligned.
+ */
+       .data
+       .globl  sdata
+sdata:
+       .globl  empty_zero_page
+empty_zero_page:
+       .space  4096
+
+       .globl  swapper_pg_dir
+swapper_pg_dir:
+       .space  4096
+
+/*
+ * This space gets a copy of optional info passed to us by the bootstrap
+ * Used to pass parameters into the kernel like root=/dev/sda1, etc.
+ */
+       .globl  cmd_line
+cmd_line:
+       .space  512
+
+       .globl intercept_table
+intercept_table:
+       .long 0, 0, i0x200, i0x300, i0x400, 0, i0x600, i0x700
+       .long i0x800, 0, 0, 0, 0, i0xd00, 0, 0
+       .long 0, 0, 0, i0x1300, 0, 0, 0, 0
+       .long 0, 0, 0, 0, 0, 0, 0, 0
+       .long 0, 0, 0, 0, 0, 0, 0, 0
+       .long 0, 0, 0, 0, 0, 0, 0, 0
+
+/* Room for two PTE pointers, usually the kernel and current user pointers
+ * to their respective root page table.
+ */
+abatron_pteptrs:
+       .space  8
diff --git a/arch/powerpc/kernel/head_44x.S b/arch/powerpc/kernel/head_44x.S
new file mode 100644 (file)
index 0000000..599245b
--- /dev/null
@@ -0,0 +1,778 @@
+/*
+ * arch/ppc/kernel/head_44x.S
+ *
+ * Kernel execution entry point code.
+ *
+ *    Copyright (c) 1995-1996 Gary Thomas <gdt@linuxppc.org>
+ *      Initial PowerPC version.
+ *    Copyright (c) 1996 Cort Dougan <cort@cs.nmt.edu>
+ *      Rewritten for PReP
+ *    Copyright (c) 1996 Paul Mackerras <paulus@cs.anu.edu.au>
+ *      Low-level exception handers, MMU support, and rewrite.
+ *    Copyright (c) 1997 Dan Malek <dmalek@jlc.net>
+ *      PowerPC 8xx modifications.
+ *    Copyright (c) 1998-1999 TiVo, Inc.
+ *      PowerPC 403GCX modifications.
+ *    Copyright (c) 1999 Grant Erickson <grant@lcse.umn.edu>
+ *      PowerPC 403GCX/405GP modifications.
+ *    Copyright 2000 MontaVista Software Inc.
+ *     PPC405 modifications
+ *      PowerPC 403GCX/405GP modifications.
+ *     Author: MontaVista Software, Inc.
+ *             frank_rowand@mvista.com or source@mvista.com
+ *             debbie_chu@mvista.com
+ *    Copyright 2002-2005 MontaVista Software, Inc.
+ *      PowerPC 44x support, Matt Porter <mporter@kernel.crashing.org>
+ *
+ * This program is free software; you can redistribute  it and/or modify it
+ * under  the terms of  the GNU General  Public License as published by the
+ * Free Software Foundation;  either version 2 of the  License, or (at your
+ * option) any later version.
+ */
+
+#include <linux/config.h>
+#include <asm/processor.h>
+#include <asm/page.h>
+#include <asm/mmu.h>
+#include <asm/pgtable.h>
+#include <asm/ibm4xx.h>
+#include <asm/ibm44x.h>
+#include <asm/cputable.h>
+#include <asm/thread_info.h>
+#include <asm/ppc_asm.h>
+#include <asm/asm-offsets.h>
+#include "head_booke.h"
+
+
+/* As with the other PowerPC ports, it is expected that when code
+ * execution begins here, the following registers contain valid, yet
+ * optional, information:
+ *
+ *   r3 - Board info structure pointer (DRAM, frequency, MAC address, etc.)
+ *   r4 - Starting address of the init RAM disk
+ *   r5 - Ending address of the init RAM disk
+ *   r6 - Start of kernel command line string (e.g. "mem=128")
+ *   r7 - End of kernel command line string
+ *
+ */
+       .text
+_GLOBAL(_stext)
+_GLOBAL(_start)
+       /*
+        * Reserve a word at a fixed location to store the address
+        * of abatron_pteptrs
+        */
+       nop
+/*
+ * Save parameters we are passed
+ */
+       mr      r31,r3
+       mr      r30,r4
+       mr      r29,r5
+       mr      r28,r6
+       mr      r27,r7
+       li      r24,0           /* CPU number */
+
+/*
+ * Set up the initial MMU state
+ *
+ * We are still executing code at the virtual address
+ * mappings set by the firmware for the base of RAM.
+ *
+ * We first invalidate all TLB entries but the one
+ * we are running from.  We then load the KERNELBASE
+ * mappings so we can begin to use kernel addresses
+ * natively and so the interrupt vector locations are
+ * permanently pinned (necessary since Book E
+ * implementations always have translation enabled).
+ *
+ * TODO: Use the known TLB entry we are running from to
+ *      determine which physical region we are located
+ *      in.  This can be used to determine where in RAM
+ *      (on a shared CPU system) or PCI memory space
+ *      (on a DRAMless system) we are located.
+ *       For now, we assume a perfect world which means
+ *      we are located at the base of DRAM (physical 0).
+ */
+
+/*
+ * Search TLB for entry that we are currently using.
+ * Invalidate all entries but the one we are using.
+ */
+       /* Load our current PID->MMUCR TID and MSR IS->MMUCR STS */
+       mfspr   r3,SPRN_PID                     /* Get PID */
+       mfmsr   r4                              /* Get MSR */
+       andi.   r4,r4,MSR_IS@l                  /* TS=1? */
+       beq     wmmucr                          /* If not, leave STS=0 */
+       oris    r3,r3,PPC44x_MMUCR_STS@h        /* Set STS=1 */
+wmmucr:        mtspr   SPRN_MMUCR,r3                   /* Put MMUCR */
+       sync
+
+       bl      invstr                          /* Find our address */
+invstr:        mflr    r5                              /* Make it accessible */
+       tlbsx   r23,0,r5                        /* Find entry we are in */
+       li      r4,0                            /* Start at TLB entry 0 */
+       li      r3,0                            /* Set PAGEID inval value */
+1:     cmpw    r23,r4                          /* Is this our entry? */
+       beq     skpinv                          /* If so, skip the inval */
+       tlbwe   r3,r4,PPC44x_TLB_PAGEID         /* If not, inval the entry */
+skpinv:        addi    r4,r4,1                         /* Increment */
+       cmpwi   r4,64                           /* Are we done? */
+       bne     1b                              /* If not, repeat */
+       isync                                   /* If so, context change */
+
+/*
+ * Configure and load pinned entry into TLB slot 63.
+ */
+
+       lis     r3,KERNELBASE@h         /* Load the kernel virtual address */
+       ori     r3,r3,KERNELBASE@l
+
+       /* Kernel is at the base of RAM */
+       li r4, 0                        /* Load the kernel physical address */
+
+       /* Load the kernel PID = 0 */
+       li      r0,0
+       mtspr   SPRN_PID,r0
+       sync
+
+       /* Initialize MMUCR */
+       li      r5,0
+       mtspr   SPRN_MMUCR,r5
+       sync
+
+       /* pageid fields */
+       clrrwi  r3,r3,10                /* Mask off the effective page number */
+       ori     r3,r3,PPC44x_TLB_VALID | PPC44x_TLB_256M
+
+       /* xlat fields */
+       clrrwi  r4,r4,10                /* Mask off the real page number */
+                                       /* ERPN is 0 for first 4GB page */
+
+       /* attrib fields */
+       /* Added guarded bit to protect against speculative loads/stores */
+       li      r5,0
+       ori     r5,r5,(PPC44x_TLB_SW | PPC44x_TLB_SR | PPC44x_TLB_SX | PPC44x_TLB_G)
+
+        li      r0,63                    /* TLB slot 63 */
+
+       tlbwe   r3,r0,PPC44x_TLB_PAGEID /* Load the pageid fields */
+       tlbwe   r4,r0,PPC44x_TLB_XLAT   /* Load the translation fields */
+       tlbwe   r5,r0,PPC44x_TLB_ATTRIB /* Load the attrib/access fields */
+
+       /* Force context change */
+       mfmsr   r0
+       mtspr   SPRN_SRR1, r0
+       lis     r0,3f@h
+       ori     r0,r0,3f@l
+       mtspr   SPRN_SRR0,r0
+       sync
+       rfi
+
+       /* If necessary, invalidate original entry we used */
+3:     cmpwi   r23,63
+       beq     4f
+       li      r6,0
+       tlbwe   r6,r23,PPC44x_TLB_PAGEID
+       isync
+
+4:
+#ifdef CONFIG_SERIAL_TEXT_DEBUG
+       /*
+        * Add temporary UART mapping for early debug.
+        * We can map UART registers wherever we want as long as they don't
+        * interfere with other system mappings (e.g. with pinned entries).
+        * For an example of how we handle this - see ocotea.h.       --ebs
+        */
+       /* pageid fields */
+       lis     r3,UART0_IO_BASE@h
+       ori     r3,r3,PPC44x_TLB_VALID | PPC44x_TLB_4K
+
+       /* xlat fields */
+       lis     r4,UART0_PHYS_IO_BASE@h         /* RPN depends on SoC */
+#ifndef CONFIG_440EP
+       ori     r4,r4,0x0001            /* ERPN is 1 for second 4GB page */
+#endif
+
+       /* attrib fields */
+       li      r5,0
+       ori     r5,r5,(PPC44x_TLB_SW | PPC44x_TLB_SR | PPC44x_TLB_I | PPC44x_TLB_G)
+
+        li      r0,0                    /* TLB slot 0 */
+
+       tlbwe   r3,r0,PPC44x_TLB_PAGEID /* Load the pageid fields */
+       tlbwe   r4,r0,PPC44x_TLB_XLAT   /* Load the translation fields */
+       tlbwe   r5,r0,PPC44x_TLB_ATTRIB /* Load the attrib/access fields */
+
+       /* Force context change */
+       isync
+#endif /* CONFIG_SERIAL_TEXT_DEBUG */
+
+       /* Establish the interrupt vector offsets */
+       SET_IVOR(0,  CriticalInput);
+       SET_IVOR(1,  MachineCheck);
+       SET_IVOR(2,  DataStorage);
+       SET_IVOR(3,  InstructionStorage);
+       SET_IVOR(4,  ExternalInput);
+       SET_IVOR(5,  Alignment);
+       SET_IVOR(6,  Program);
+       SET_IVOR(7,  FloatingPointUnavailable);
+       SET_IVOR(8,  SystemCall);
+       SET_IVOR(9,  AuxillaryProcessorUnavailable);
+       SET_IVOR(10, Decrementer);
+       SET_IVOR(11, FixedIntervalTimer);
+       SET_IVOR(12, WatchdogTimer);
+       SET_IVOR(13, DataTLBError);
+       SET_IVOR(14, InstructionTLBError);
+       SET_IVOR(15, Debug);
+
+       /* Establish the interrupt vector base */
+       lis     r4,interrupt_base@h     /* IVPR only uses the high 16-bits */
+       mtspr   SPRN_IVPR,r4
+
+#ifdef CONFIG_440EP
+       /* Clear DAPUIB flag in CCR0 (enable APU between CPU and FPU) */
+       mfspr   r2,SPRN_CCR0
+       lis     r3,0xffef
+       ori     r3,r3,0xffff
+       and     r2,r2,r3
+       mtspr   SPRN_CCR0,r2
+       isync
+#endif
+
+       /*
+        * This is where the main kernel code starts.
+        */
+
+       /* ptr to current */
+       lis     r2,init_task@h
+       ori     r2,r2,init_task@l
+
+       /* ptr to current thread */
+       addi    r4,r2,THREAD    /* init task's THREAD */
+       mtspr   SPRN_SPRG3,r4
+
+       /* stack */
+       lis     r1,init_thread_union@h
+       ori     r1,r1,init_thread_union@l
+       li      r0,0
+       stwu    r0,THREAD_SIZE-STACK_FRAME_OVERHEAD(r1)
+
+       bl      early_init
+
+/*
+ * Decide what sort of machine this is and initialize the MMU.
+ */
+       mr      r3,r31
+       mr      r4,r30
+       mr      r5,r29
+       mr      r6,r28
+       mr      r7,r27
+       bl      machine_init
+       bl      MMU_init
+
+       /* Setup PTE pointers for the Abatron bdiGDB */
+       lis     r6, swapper_pg_dir@h
+       ori     r6, r6, swapper_pg_dir@l
+       lis     r5, abatron_pteptrs@h
+       ori     r5, r5, abatron_pteptrs@l
+       lis     r4, KERNELBASE@h
+       ori     r4, r4, KERNELBASE@l
+       stw     r5, 0(r4)       /* Save abatron_pteptrs at a fixed location */
+       stw     r6, 0(r5)
+
+       /* Let's move on */
+       lis     r4,start_kernel@h
+       ori     r4,r4,start_kernel@l
+       lis     r3,MSR_KERNEL@h
+       ori     r3,r3,MSR_KERNEL@l
+       mtspr   SPRN_SRR0,r4
+       mtspr   SPRN_SRR1,r3
+       rfi                     /* change context and jump to start_kernel */
+
+/*
+ * Interrupt vector entry code
+ *
+ * The Book E MMUs are always on so we don't need to handle
+ * interrupts in real mode as with previous PPC processors. In
+ * this case we handle interrupts in the kernel virtual address
+ * space.
+ *
+ * Interrupt vectors are dynamically placed relative to the
+ * interrupt prefix as determined by the address of interrupt_base.
+ * The interrupt vectors offsets are programmed using the labels
+ * for each interrupt vector entry.
+ *
+ * Interrupt vectors must be aligned on a 16 byte boundary.
+ * We align on a 32 byte cache line boundary for good measure.
+ */
+
+interrupt_base:
+       /* Critical Input Interrupt */
+       CRITICAL_EXCEPTION(0x0100, CriticalInput, UnknownException)
+
+       /* Machine Check Interrupt */
+#ifdef CONFIG_440A
+       MCHECK_EXCEPTION(0x0200, MachineCheck, MachineCheckException)
+#else
+       CRITICAL_EXCEPTION(0x0200, MachineCheck, MachineCheckException)
+#endif
+
+       /* Data Storage Interrupt */
+       START_EXCEPTION(DataStorage)
+       mtspr   SPRN_SPRG0, r10         /* Save some working registers */
+       mtspr   SPRN_SPRG1, r11
+       mtspr   SPRN_SPRG4W, r12
+       mtspr   SPRN_SPRG5W, r13
+       mfcr    r11
+       mtspr   SPRN_SPRG7W, r11
+
+       /*
+        * Check if it was a store fault, if not then bail
+        * because a user tried to access a kernel or
+        * read-protected page.  Otherwise, get the
+        * offending address and handle it.
+        */
+       mfspr   r10, SPRN_ESR
+       andis.  r10, r10, ESR_ST@h
+       beq     2f
+
+       mfspr   r10, SPRN_DEAR          /* Get faulting address */
+
+       /* If we are faulting a kernel address, we have to use the
+        * kernel page tables.
+        */
+       lis     r11, TASK_SIZE@h
+       cmplw   r10, r11
+       blt+    3f
+       lis     r11, swapper_pg_dir@h
+       ori     r11, r11, swapper_pg_dir@l
+
+       mfspr   r12,SPRN_MMUCR
+       rlwinm  r12,r12,0,0,23          /* Clear TID */
+
+       b       4f
+
+       /* Get the PGD for the current thread */
+3:
+       mfspr   r11,SPRN_SPRG3
+       lwz     r11,PGDIR(r11)
+
+       /* Load PID into MMUCR TID */
+       mfspr   r12,SPRN_MMUCR          /* Get MMUCR */
+       mfspr   r13,SPRN_PID            /* Get PID */
+       rlwimi  r12,r13,0,24,31         /* Set TID */
+
+4:
+       mtspr   SPRN_MMUCR,r12
+
+       rlwinm  r12, r10, 13, 19, 29    /* Compute pgdir/pmd offset */
+       lwzx    r11, r12, r11           /* Get pgd/pmd entry */
+       rlwinm. r12, r11, 0, 0, 20      /* Extract pt base address */
+       beq     2f                      /* Bail if no table */
+
+       rlwimi  r12, r10, 23, 20, 28    /* Compute pte address */
+       lwz     r11, 4(r12)             /* Get pte entry */
+
+       andi.   r13, r11, _PAGE_RW      /* Is it writeable? */
+       beq     2f                      /* Bail if not */
+
+       /* Update 'changed'.
+       */
+       ori     r11, r11, _PAGE_DIRTY|_PAGE_ACCESSED|_PAGE_HWWRITE
+       stw     r11, 4(r12)             /* Update Linux page table */
+
+       li      r13, PPC44x_TLB_SR@l    /* Set SR */
+       rlwimi  r13, r11, 29, 29, 29    /* SX = _PAGE_HWEXEC */
+       rlwimi  r13, r11, 0, 30, 30     /* SW = _PAGE_RW */
+       rlwimi  r13, r11, 29, 28, 28    /* UR = _PAGE_USER */
+       rlwimi  r12, r11, 31, 26, 26    /* (_PAGE_USER>>1)->r12 */
+       rlwimi  r12, r11, 29, 30, 30    /* (_PAGE_USER>>3)->r12 */
+       and     r12, r12, r11           /* HWEXEC/RW & USER */
+       rlwimi  r13, r12, 0, 26, 26     /* UX = HWEXEC & USER */
+       rlwimi  r13, r12, 3, 27, 27     /* UW = RW & USER */
+
+       rlwimi  r11,r13,0,26,31         /* Insert static perms */
+
+       rlwinm  r11,r11,0,20,15         /* Clear U0-U3 */
+
+       /* find the TLB index that caused the fault.  It has to be here. */
+       tlbsx   r10, 0, r10
+
+       tlbwe   r11, r10, PPC44x_TLB_ATTRIB     /* Write ATTRIB */
+
+       /* Done...restore registers and get out of here.
+       */
+       mfspr   r11, SPRN_SPRG7R
+       mtcr    r11
+       mfspr   r13, SPRN_SPRG5R
+       mfspr   r12, SPRN_SPRG4R
+
+       mfspr   r11, SPRN_SPRG1
+       mfspr   r10, SPRN_SPRG0
+       rfi                     /* Force context change */
+
+2:
+       /*
+        * The bailout.  Restore registers to pre-exception conditions
+        * and call the heavyweights to help us out.
+        */
+       mfspr   r11, SPRN_SPRG7R
+       mtcr    r11
+       mfspr   r13, SPRN_SPRG5R
+       mfspr   r12, SPRN_SPRG4R
+
+       mfspr   r11, SPRN_SPRG1
+       mfspr   r10, SPRN_SPRG0
+       b       data_access
+
+       /* Instruction Storage Interrupt */
+       INSTRUCTION_STORAGE_EXCEPTION
+
+       /* External Input Interrupt */
+       EXCEPTION(0x0500, ExternalInput, do_IRQ, EXC_XFER_LITE)
+
+       /* Alignment Interrupt */
+       ALIGNMENT_EXCEPTION
+
+       /* Program Interrupt */
+       PROGRAM_EXCEPTION
+
+       /* Floating Point Unavailable Interrupt */
+#ifdef CONFIG_PPC_FPU
+       FP_UNAVAILABLE_EXCEPTION
+#else
+       EXCEPTION(0x2010, FloatingPointUnavailable, UnknownException, EXC_XFER_EE)
+#endif
+
+       /* System Call Interrupt */
+       START_EXCEPTION(SystemCall)
+       NORMAL_EXCEPTION_PROLOG
+       EXC_XFER_EE_LITE(0x0c00, DoSyscall)
+
+       /* Auxillary Processor Unavailable Interrupt */
+       EXCEPTION(0x2020, AuxillaryProcessorUnavailable, UnknownException, EXC_XFER_EE)
+
+       /* Decrementer Interrupt */
+       DECREMENTER_EXCEPTION
+
+       /* Fixed Internal Timer Interrupt */
+       /* TODO: Add FIT support */
+       EXCEPTION(0x1010, FixedIntervalTimer, UnknownException, EXC_XFER_EE)
+
+       /* Watchdog Timer Interrupt */
+       /* TODO: Add watchdog support */
+#ifdef CONFIG_BOOKE_WDT
+       CRITICAL_EXCEPTION(0x1020, WatchdogTimer, WatchdogException)
+#else
+       CRITICAL_EXCEPTION(0x1020, WatchdogTimer, UnknownException)
+#endif
+
+       /* Data TLB Error Interrupt */
+       START_EXCEPTION(DataTLBError)
+       mtspr   SPRN_SPRG0, r10         /* Save some working registers */
+       mtspr   SPRN_SPRG1, r11
+       mtspr   SPRN_SPRG4W, r12
+       mtspr   SPRN_SPRG5W, r13
+       mfcr    r11
+       mtspr   SPRN_SPRG7W, r11
+       mfspr   r10, SPRN_DEAR          /* Get faulting address */
+
+       /* If we are faulting a kernel address, we have to use the
+        * kernel page tables.
+        */
+       lis     r11, TASK_SIZE@h
+       cmplw   r10, r11
+       blt+    3f
+       lis     r11, swapper_pg_dir@h
+       ori     r11, r11, swapper_pg_dir@l
+
+       mfspr   r12,SPRN_MMUCR
+       rlwinm  r12,r12,0,0,23          /* Clear TID */
+
+       b       4f
+
+       /* Get the PGD for the current thread */
+3:
+       mfspr   r11,SPRN_SPRG3
+       lwz     r11,PGDIR(r11)
+
+       /* Load PID into MMUCR TID */
+       mfspr   r12,SPRN_MMUCR
+       mfspr   r13,SPRN_PID            /* Get PID */
+       rlwimi  r12,r13,0,24,31         /* Set TID */
+
+4:
+       mtspr   SPRN_MMUCR,r12
+
+       rlwinm  r12, r10, 13, 19, 29    /* Compute pgdir/pmd offset */
+       lwzx    r11, r12, r11           /* Get pgd/pmd entry */
+       rlwinm. r12, r11, 0, 0, 20      /* Extract pt base address */
+       beq     2f                      /* Bail if no table */
+
+       rlwimi  r12, r10, 23, 20, 28    /* Compute pte address */
+       lwz     r11, 4(r12)             /* Get pte entry */
+       andi.   r13, r11, _PAGE_PRESENT /* Is the page present? */
+       beq     2f                      /* Bail if not present */
+
+       ori     r11, r11, _PAGE_ACCESSED
+       stw     r11, 4(r12)
+
+        /* Jump to common tlb load */
+       b       finish_tlb_load
+
+2:
+       /* The bailout.  Restore registers to pre-exception conditions
+        * and call the heavyweights to help us out.
+        */
+       mfspr   r11, SPRN_SPRG7R
+       mtcr    r11
+       mfspr   r13, SPRN_SPRG5R
+       mfspr   r12, SPRN_SPRG4R
+       mfspr   r11, SPRN_SPRG1
+       mfspr   r10, SPRN_SPRG0
+       b       data_access
+
+       /* Instruction TLB Error Interrupt */
+       /*
+        * Nearly the same as above, except we get our
+        * information from different registers and bailout
+        * to a different point.
+        */
+       START_EXCEPTION(InstructionTLBError)
+       mtspr   SPRN_SPRG0, r10         /* Save some working registers */
+       mtspr   SPRN_SPRG1, r11
+       mtspr   SPRN_SPRG4W, r12
+       mtspr   SPRN_SPRG5W, r13
+       mfcr    r11
+       mtspr   SPRN_SPRG7W, r11
+       mfspr   r10, SPRN_SRR0          /* Get faulting address */
+
+       /* If we are faulting a kernel address, we have to use the
+        * kernel page tables.
+        */
+       lis     r11, TASK_SIZE@h
+       cmplw   r10, r11
+       blt+    3f
+       lis     r11, swapper_pg_dir@h
+       ori     r11, r11, swapper_pg_dir@l
+
+       mfspr   r12,SPRN_MMUCR
+       rlwinm  r12,r12,0,0,23          /* Clear TID */
+
+       b       4f
+
+       /* Get the PGD for the current thread */
+3:
+       mfspr   r11,SPRN_SPRG3
+       lwz     r11,PGDIR(r11)
+
+       /* Load PID into MMUCR TID */
+       mfspr   r12,SPRN_MMUCR
+       mfspr   r13,SPRN_PID            /* Get PID */
+       rlwimi  r12,r13,0,24,31         /* Set TID */
+
+4:
+       mtspr   SPRN_MMUCR,r12
+
+       rlwinm  r12, r10, 13, 19, 29    /* Compute pgdir/pmd offset */
+       lwzx    r11, r12, r11           /* Get pgd/pmd entry */
+       rlwinm. r12, r11, 0, 0, 20      /* Extract pt base address */
+       beq     2f                      /* Bail if no table */
+
+       rlwimi  r12, r10, 23, 20, 28    /* Compute pte address */
+       lwz     r11, 4(r12)             /* Get pte entry */
+       andi.   r13, r11, _PAGE_PRESENT /* Is the page present? */
+       beq     2f                      /* Bail if not present */
+
+       ori     r11, r11, _PAGE_ACCESSED
+       stw     r11, 4(r12)
+
+       /* Jump to common TLB load point */
+       b       finish_tlb_load
+
+2:
+       /* The bailout.  Restore registers to pre-exception conditions
+        * and call the heavyweights to help us out.
+        */
+       mfspr   r11, SPRN_SPRG7R
+       mtcr    r11
+       mfspr   r13, SPRN_SPRG5R
+       mfspr   r12, SPRN_SPRG4R
+       mfspr   r11, SPRN_SPRG1
+       mfspr   r10, SPRN_SPRG0
+       b       InstructionStorage
+
+       /* Debug Interrupt */
+       DEBUG_EXCEPTION
+
+/*
+ * Local functions
+ */
+       /*
+        * Data TLB exceptions will bail out to this point
+        * if they can't resolve the lightweight TLB fault.
+        */
+data_access:
+       NORMAL_EXCEPTION_PROLOG
+       mfspr   r5,SPRN_ESR             /* Grab the ESR, save it, pass arg3 */
+       stw     r5,_ESR(r11)
+       mfspr   r4,SPRN_DEAR            /* Grab the DEAR, save it, pass arg2 */
+       EXC_XFER_EE_LITE(0x0300, handle_page_fault)
+
+/*
+
+ * Both the instruction and data TLB miss get to this
+ * point to load the TLB.
+ *     r10 - EA of fault
+ *     r11 - available to use
+ *     r12 - Pointer to the 64-bit PTE
+ *     r13 - available to use
+ *     MMUCR - loaded with proper value when we get here
+ *     Upon exit, we reload everything and RFI.
+ */
+finish_tlb_load:
+       /*
+        * We set execute, because we don't have the granularity to
+        * properly set this at the page level (Linux problem).
+        * If shared is set, we cause a zero PID->TID load.
+        * Many of these bits are software only.  Bits we don't set
+        * here we (properly should) assume have the appropriate value.
+        */
+
+       /* Load the next available TLB index */
+       lis     r13, tlb_44x_index@ha
+       lwz     r13, tlb_44x_index@l(r13)
+       /* Load the TLB high watermark */
+       lis     r11, tlb_44x_hwater@ha
+       lwz     r11, tlb_44x_hwater@l(r11)
+
+       /* Increment, rollover, and store TLB index */
+       addi    r13, r13, 1
+       cmpw    0, r13, r11                     /* reserve entries */
+       ble     7f
+       li      r13, 0
+7:
+       /* Store the next available TLB index */
+       lis     r11, tlb_44x_index@ha
+       stw     r13, tlb_44x_index@l(r11)
+
+       lwz     r11, 0(r12)                     /* Get MS word of PTE */
+       lwz     r12, 4(r12)                     /* Get LS word of PTE */
+       rlwimi  r11, r12, 0, 0 , 19             /* Insert RPN */
+       tlbwe   r11, r13, PPC44x_TLB_XLAT       /* Write XLAT */
+
+       /*
+        * Create PAGEID. This is the faulting address,
+        * page size, and valid flag.
+        */
+       li      r11, PPC44x_TLB_VALID | PPC44x_TLB_4K
+       rlwimi  r10, r11, 0, 20, 31             /* Insert valid and page size */
+       tlbwe   r10, r13, PPC44x_TLB_PAGEID     /* Write PAGEID */
+
+       li      r10, PPC44x_TLB_SR@l            /* Set SR */
+       rlwimi  r10, r12, 0, 30, 30             /* Set SW = _PAGE_RW */
+       rlwimi  r10, r12, 29, 29, 29            /* SX = _PAGE_HWEXEC */
+       rlwimi  r10, r12, 29, 28, 28            /* UR = _PAGE_USER */
+       rlwimi  r11, r12, 31, 26, 26            /* (_PAGE_USER>>1)->r12 */
+       and     r11, r12, r11                   /* HWEXEC & USER */
+       rlwimi  r10, r11, 0, 26, 26             /* UX = HWEXEC & USER */
+
+       rlwimi  r12, r10, 0, 26, 31             /* Insert static perms */
+       rlwinm  r12, r12, 0, 20, 15             /* Clear U0-U3 */
+       tlbwe   r12, r13, PPC44x_TLB_ATTRIB     /* Write ATTRIB */
+
+       /* Done...restore registers and get out of here.
+       */
+       mfspr   r11, SPRN_SPRG7R
+       mtcr    r11
+       mfspr   r13, SPRN_SPRG5R
+       mfspr   r12, SPRN_SPRG4R
+       mfspr   r11, SPRN_SPRG1
+       mfspr   r10, SPRN_SPRG0
+       rfi                                     /* Force context change */
+
+/*
+ * Global functions
+ */
+
+/*
+ * extern void giveup_altivec(struct task_struct *prev)
+ *
+ * The 44x core does not have an AltiVec unit.
+ */
+_GLOBAL(giveup_altivec)
+       blr
+
+/*
+ * extern void giveup_fpu(struct task_struct *prev)
+ *
+ * The 44x core does not have an FPU.
+ */
+#ifndef CONFIG_PPC_FPU
+_GLOBAL(giveup_fpu)
+       blr
+#endif
+
+/*
+ * extern void abort(void)
+ *
+ * At present, this routine just applies a system reset.
+ */
+_GLOBAL(abort)
+        mfspr   r13,SPRN_DBCR0
+        oris    r13,r13,DBCR0_RST_SYSTEM@h
+        mtspr   SPRN_DBCR0,r13
+
+_GLOBAL(set_context)
+
+#ifdef CONFIG_BDI_SWITCH
+       /* Context switch the PTE pointer for the Abatron BDI2000.
+        * The PGDIR is the second parameter.
+        */
+       lis     r5, abatron_pteptrs@h
+       ori     r5, r5, abatron_pteptrs@l
+       stw     r4, 0x4(r5)
+#endif
+       mtspr   SPRN_PID,r3
+       isync                   /* Force context change */
+       blr
+
+/*
+ * We put a few things here that have to be page-aligned. This stuff
+ * goes at the beginning of the data segment, which is page-aligned.
+ */
+       .data
+_GLOBAL(sdata)
+_GLOBAL(empty_zero_page)
+       .space  4096
+
+/*
+ * To support >32-bit physical addresses, we use an 8KB pgdir.
+ */
+_GLOBAL(swapper_pg_dir)
+       .space  8192
+
+/* Reserved 4k for the critical exception stack & 4k for the machine
+ * check stack per CPU for kernel mode exceptions */
+       .section .bss
+        .align 12
+exception_stack_bottom:
+       .space  BOOKE_EXCEPTION_STACK_SIZE
+_GLOBAL(exception_stack_top)
+
+/*
+ * This space gets a copy of optional info passed to us by the bootstrap
+ * which is used to pass parameters into the kernel like root=/dev/sda1, etc.
+ */
+_GLOBAL(cmd_line)
+       .space  512
+
+/*
+ * Room for two PTE pointers, usually the kernel and current user pointers
+ * to their respective root page table.
+ */
+abatron_pteptrs:
+       .space  8
+
+
diff --git a/arch/powerpc/kernel/head_4xx.S b/arch/powerpc/kernel/head_4xx.S
new file mode 100644 (file)
index 0000000..8562b80
--- /dev/null
@@ -0,0 +1,1016 @@
+/*
+ *    Copyright (c) 1995-1996 Gary Thomas <gdt@linuxppc.org>
+ *      Initial PowerPC version.
+ *    Copyright (c) 1996 Cort Dougan <cort@cs.nmt.edu>
+ *      Rewritten for PReP
+ *    Copyright (c) 1996 Paul Mackerras <paulus@cs.anu.edu.au>
+ *      Low-level exception handers, MMU support, and rewrite.
+ *    Copyright (c) 1997 Dan Malek <dmalek@jlc.net>
+ *      PowerPC 8xx modifications.
+ *    Copyright (c) 1998-1999 TiVo, Inc.
+ *      PowerPC 403GCX modifications.
+ *    Copyright (c) 1999 Grant Erickson <grant@lcse.umn.edu>
+ *      PowerPC 403GCX/405GP modifications.
+ *    Copyright 2000 MontaVista Software Inc.
+ *     PPC405 modifications
+ *      PowerPC 403GCX/405GP modifications.
+ *     Author: MontaVista Software, Inc.
+ *             frank_rowand@mvista.com or source@mvista.com
+ *             debbie_chu@mvista.com
+ *
+ *
+ *    Module name: head_4xx.S
+ *
+ *    Description:
+ *      Kernel execution entry point code.
+ *
+ *    This program is free software; you can redistribute it and/or
+ *    modify it under the terms of the GNU General Public License
+ *    as published by the Free Software Foundation; either version
+ *    2 of the License, or (at your option) any later version.
+ *
+ */
+
+#include <linux/config.h>
+#include <asm/processor.h>
+#include <asm/page.h>
+#include <asm/mmu.h>
+#include <asm/pgtable.h>
+#include <asm/ibm4xx.h>
+#include <asm/cputable.h>
+#include <asm/thread_info.h>
+#include <asm/ppc_asm.h>
+#include <asm/asm-offsets.h>
+
+/* As with the other PowerPC ports, it is expected that when code
+ * execution begins here, the following registers contain valid, yet
+ * optional, information:
+ *
+ *   r3 - Board info structure pointer (DRAM, frequency, MAC address, etc.)
+ *   r4 - Starting address of the init RAM disk
+ *   r5 - Ending address of the init RAM disk
+ *   r6 - Start of kernel command line string (e.g. "mem=96m")
+ *   r7 - End of kernel command line string
+ *
+ * This is all going to change RSN when we add bi_recs.......  -- Dan
+ */
+       .text
+_GLOBAL(_stext)
+_GLOBAL(_start)
+
+       /* Save parameters we are passed.
+       */
+       mr      r31,r3
+       mr      r30,r4
+       mr      r29,r5
+       mr      r28,r6
+       mr      r27,r7
+
+       /* We have to turn on the MMU right away so we get cache modes
+        * set correctly.
+        */
+       bl      initial_mmu
+
+/* We now have the lower 16 Meg mapped into TLB entries, and the caches
+ * ready to work.
+ */
+turn_on_mmu:
+       lis     r0,MSR_KERNEL@h
+       ori     r0,r0,MSR_KERNEL@l
+       mtspr   SPRN_SRR1,r0
+       lis     r0,start_here@h
+       ori     r0,r0,start_here@l
+       mtspr   SPRN_SRR0,r0
+       SYNC
+       rfi                             /* enables MMU */
+       b       .                       /* prevent prefetch past rfi */
+
+/*
+ * This area is used for temporarily saving registers during the
+ * critical exception prolog.
+ */
+       . = 0xc0
+crit_save:
+_GLOBAL(crit_r10)
+       .space  4
+_GLOBAL(crit_r11)
+       .space  4
+
+/*
+ * Exception vector entry code. This code runs with address translation
+ * turned off (i.e. using physical addresses). We assume SPRG3 has the
+ * physical address of the current task thread_struct.
+ * Note that we have to have decremented r1 before we write to any fields
+ * of the exception frame, since a critical interrupt could occur at any
+ * time, and it will write to the area immediately below the current r1.
+ */
+#define NORMAL_EXCEPTION_PROLOG                                                     \
+       mtspr   SPRN_SPRG0,r10;         /* save two registers to work with */\
+       mtspr   SPRN_SPRG1,r11;                                              \
+       mtspr   SPRN_SPRG2,r1;                                               \
+       mfcr    r10;                    /* save CR in r10 for now          */\
+       mfspr   r11,SPRN_SRR1;          /* check whether user or kernel    */\
+       andi.   r11,r11,MSR_PR;                                              \
+       beq     1f;                                                          \
+       mfspr   r1,SPRN_SPRG3;          /* if from user, start at top of   */\
+       lwz     r1,THREAD_INFO-THREAD(r1); /* this thread's kernel stack   */\
+       addi    r1,r1,THREAD_SIZE;                                           \
+1:     subi    r1,r1,INT_FRAME_SIZE;   /* Allocate an exception frame     */\
+       tophys(r11,r1);                                                      \
+       stw     r10,_CCR(r11);          /* save various registers          */\
+       stw     r12,GPR12(r11);                                              \
+       stw     r9,GPR9(r11);                                                \
+       mfspr   r10,SPRN_SPRG0;                                              \
+       stw     r10,GPR10(r11);                                              \
+       mfspr   r12,SPRN_SPRG1;                                              \
+       stw     r12,GPR11(r11);                                              \
+       mflr    r10;                                                         \
+       stw     r10,_LINK(r11);                                              \
+       mfspr   r10,SPRN_SPRG2;                                              \
+       mfspr   r12,SPRN_SRR0;                                               \
+       stw     r10,GPR1(r11);                                               \
+       mfspr   r9,SPRN_SRR1;                                                \
+       stw     r10,0(r11);                                                  \
+       rlwinm  r9,r9,0,14,12;          /* clear MSR_WE (necessary?)       */\
+       stw     r0,GPR0(r11);                                                \
+       SAVE_4GPRS(3, r11);                                                  \
+       SAVE_2GPRS(7, r11)
+
+/*
+ * Exception prolog for critical exceptions.  This is a little different
+ * from the normal exception prolog above since a critical exception
+ * can potentially occur at any point during normal exception processing.
+ * Thus we cannot use the same SPRG registers as the normal prolog above.
+ * Instead we use a couple of words of memory at low physical addresses.
+ * This is OK since we don't support SMP on these processors.
+ */
+#define CRITICAL_EXCEPTION_PROLOG                                           \
+       stw     r10,crit_r10@l(0);      /* save two registers to work with */\
+       stw     r11,crit_r11@l(0);                                           \
+       mfcr    r10;                    /* save CR in r10 for now          */\
+       mfspr   r11,SPRN_SRR3;          /* check whether user or kernel    */\
+       andi.   r11,r11,MSR_PR;                                              \
+       lis     r11,critical_stack_top@h;                                    \
+       ori     r11,r11,critical_stack_top@l;                                \
+       beq     1f;                                                          \
+       /* COMING FROM USER MODE */                                          \
+       mfspr   r11,SPRN_SPRG3;         /* if from user, start at top of   */\
+       lwz     r11,THREAD_INFO-THREAD(r11); /* this thread's kernel stack */\
+       addi    r11,r11,THREAD_SIZE;                                         \
+1:     subi    r11,r11,INT_FRAME_SIZE; /* Allocate an exception frame     */\
+       tophys(r11,r11);                                                     \
+       stw     r10,_CCR(r11);          /* save various registers          */\
+       stw     r12,GPR12(r11);                                              \
+       stw     r9,GPR9(r11);                                                \
+       mflr    r10;                                                         \
+       stw     r10,_LINK(r11);                                              \
+       mfspr   r12,SPRN_DEAR;          /* save DEAR and ESR in the frame  */\
+       stw     r12,_DEAR(r11);         /* since they may have had stuff   */\
+       mfspr   r9,SPRN_ESR;            /* in them at the point where the  */\
+       stw     r9,_ESR(r11);           /* exception was taken             */\
+       mfspr   r12,SPRN_SRR2;                                               \
+       stw     r1,GPR1(r11);                                                \
+       mfspr   r9,SPRN_SRR3;                                                \
+       stw     r1,0(r11);                                                   \
+       tovirt(r1,r11);                                                      \
+       rlwinm  r9,r9,0,14,12;          /* clear MSR_WE (necessary?)       */\
+       stw     r0,GPR0(r11);                                                \
+       SAVE_4GPRS(3, r11);                                                  \
+       SAVE_2GPRS(7, r11)
+
+       /*
+        * State at this point:
+        * r9 saved in stack frame, now saved SRR3 & ~MSR_WE
+        * r10 saved in crit_r10 and in stack frame, trashed
+        * r11 saved in crit_r11 and in stack frame,
+        *      now phys stack/exception frame pointer
+        * r12 saved in stack frame, now saved SRR2
+        * CR saved in stack frame, CR0.EQ = !SRR3.PR
+        * LR, DEAR, ESR in stack frame
+        * r1 saved in stack frame, now virt stack/excframe pointer
+        * r0, r3-r8 saved in stack frame
+        */
+
+/*
+ * Exception vectors.
+ */
+#define        START_EXCEPTION(n, label)                                            \
+       . = n;                                                               \
+label:
+
+#define EXCEPTION(n, label, hdlr, xfer)                                \
+       START_EXCEPTION(n, label);                              \
+       NORMAL_EXCEPTION_PROLOG;                                \
+       addi    r3,r1,STACK_FRAME_OVERHEAD;                     \
+       xfer(n, hdlr)
+
+#define CRITICAL_EXCEPTION(n, label, hdlr)                     \
+       START_EXCEPTION(n, label);                              \
+       CRITICAL_EXCEPTION_PROLOG;                              \
+       addi    r3,r1,STACK_FRAME_OVERHEAD;                     \
+       EXC_XFER_TEMPLATE(hdlr, n+2, (MSR_KERNEL & ~(MSR_ME|MSR_DE|MSR_CE)), \
+                         NOCOPY, crit_transfer_to_handler,     \
+                         ret_from_crit_exc)
+
+#define EXC_XFER_TEMPLATE(hdlr, trap, msr, copyee, tfer, ret)  \
+       li      r10,trap;                                       \
+       stw     r10,TRAP(r11);                                  \
+       lis     r10,msr@h;                                      \
+       ori     r10,r10,msr@l;                                  \
+       copyee(r10, r9);                                        \
+       bl      tfer;                                           \
+       .long   hdlr;                                           \
+       .long   ret
+
+#define COPY_EE(d, s)          rlwimi d,s,0,16,16
+#define NOCOPY(d, s)
+
+#define EXC_XFER_STD(n, hdlr)          \
+       EXC_XFER_TEMPLATE(hdlr, n, MSR_KERNEL, NOCOPY, transfer_to_handler_full, \
+                         ret_from_except_full)
+
+#define EXC_XFER_LITE(n, hdlr)         \
+       EXC_XFER_TEMPLATE(hdlr, n+1, MSR_KERNEL, NOCOPY, transfer_to_handler, \
+                         ret_from_except)
+
+#define EXC_XFER_EE(n, hdlr)           \
+       EXC_XFER_TEMPLATE(hdlr, n, MSR_KERNEL, COPY_EE, transfer_to_handler_full, \
+                         ret_from_except_full)
+
+#define EXC_XFER_EE_LITE(n, hdlr)      \
+       EXC_XFER_TEMPLATE(hdlr, n+1, MSR_KERNEL, COPY_EE, transfer_to_handler, \
+                         ret_from_except)
+
+
+/*
+ * 0x0100 - Critical Interrupt Exception
+ */
+       CRITICAL_EXCEPTION(0x0100, CriticalInterrupt, UnknownException)
+
+/*
+ * 0x0200 - Machine Check Exception
+ */
+       CRITICAL_EXCEPTION(0x0200, MachineCheck, MachineCheckException)
+
+/*
+ * 0x0300 - Data Storage Exception
+ * This happens for just a few reasons.  U0 set (but we don't do that),
+ * or zone protection fault (user violation, write to protected page).
+ * If this is just an update of modified status, we do that quickly
+ * and exit.  Otherwise, we call heavywight functions to do the work.
+ */
+       START_EXCEPTION(0x0300, DataStorage)
+       mtspr   SPRN_SPRG0, r10         /* Save some working registers */
+       mtspr   SPRN_SPRG1, r11
+#ifdef CONFIG_403GCX
+       stw     r12, 0(r0)
+       stw     r9, 4(r0)
+       mfcr    r11
+       mfspr   r12, SPRN_PID
+       stw     r11, 8(r0)
+       stw     r12, 12(r0)
+#else
+       mtspr   SPRN_SPRG4, r12
+       mtspr   SPRN_SPRG5, r9
+       mfcr    r11
+       mfspr   r12, SPRN_PID
+       mtspr   SPRN_SPRG7, r11
+       mtspr   SPRN_SPRG6, r12
+#endif
+
+       /* First, check if it was a zone fault (which means a user
+       * tried to access a kernel or read-protected page - always
+       * a SEGV).  All other faults here must be stores, so no
+       * need to check ESR_DST as well. */
+       mfspr   r10, SPRN_ESR
+       andis.  r10, r10, ESR_DIZ@h
+       bne     2f
+
+       mfspr   r10, SPRN_DEAR          /* Get faulting address */
+
+       /* If we are faulting a kernel address, we have to use the
+        * kernel page tables.
+        */
+       lis     r11, TASK_SIZE@h
+       cmplw   r10, r11
+       blt+    3f
+       lis     r11, swapper_pg_dir@h
+       ori     r11, r11, swapper_pg_dir@l
+       li      r9, 0
+       mtspr   SPRN_PID, r9            /* TLB will have 0 TID */
+       b       4f
+
+       /* Get the PGD for the current thread.
+        */
+3:
+       mfspr   r11,SPRN_SPRG3
+       lwz     r11,PGDIR(r11)
+4:
+       tophys(r11, r11)
+       rlwimi  r11, r10, 12, 20, 29    /* Create L1 (pgdir/pmd) address */
+       lwz     r11, 0(r11)             /* Get L1 entry */
+       rlwinm. r12, r11, 0, 0, 19      /* Extract L2 (pte) base address */
+       beq     2f                      /* Bail if no table */
+
+       rlwimi  r12, r10, 22, 20, 29    /* Compute PTE address */
+       lwz     r11, 0(r12)             /* Get Linux PTE */
+
+       andi.   r9, r11, _PAGE_RW       /* Is it writeable? */
+       beq     2f                      /* Bail if not */
+
+       /* Update 'changed'.
+       */
+       ori     r11, r11, _PAGE_DIRTY|_PAGE_ACCESSED|_PAGE_HWWRITE
+       stw     r11, 0(r12)             /* Update Linux page table */
+
+       /* Most of the Linux PTE is ready to load into the TLB LO.
+        * We set ZSEL, where only the LS-bit determines user access.
+        * We set execute, because we don't have the granularity to
+        * properly set this at the page level (Linux problem).
+        * If shared is set, we cause a zero PID->TID load.
+        * Many of these bits are software only.  Bits we don't set
+        * here we (properly should) assume have the appropriate value.
+        */
+       li      r12, 0x0ce2
+       andc    r11, r11, r12           /* Make sure 20, 21 are zero */
+
+       /* find the TLB index that caused the fault.  It has to be here.
+       */
+       tlbsx   r9, 0, r10
+
+       tlbwe   r11, r9, TLB_DATA               /* Load TLB LO */
+
+       /* Done...restore registers and get out of here.
+       */
+#ifdef CONFIG_403GCX
+       lwz     r12, 12(r0)
+       lwz     r11, 8(r0)
+       mtspr   SPRN_PID, r12
+       mtcr    r11
+       lwz     r9, 4(r0)
+       lwz     r12, 0(r0)
+#else
+       mfspr   r12, SPRN_SPRG6
+       mfspr   r11, SPRN_SPRG7
+       mtspr   SPRN_PID, r12
+       mtcr    r11
+       mfspr   r9, SPRN_SPRG5
+       mfspr   r12, SPRN_SPRG4
+#endif
+       mfspr   r11, SPRN_SPRG1
+       mfspr   r10, SPRN_SPRG0
+       PPC405_ERR77_SYNC
+       rfi                     /* Should sync shadow TLBs */
+       b       .               /* prevent prefetch past rfi */
+
+2:
+       /* The bailout.  Restore registers to pre-exception conditions
+        * and call the heavyweights to help us out.
+        */
+#ifdef CONFIG_403GCX
+       lwz     r12, 12(r0)
+       lwz     r11, 8(r0)
+       mtspr   SPRN_PID, r12
+       mtcr    r11
+       lwz     r9, 4(r0)
+       lwz     r12, 0(r0)
+#else
+       mfspr   r12, SPRN_SPRG6
+       mfspr   r11, SPRN_SPRG7
+       mtspr   SPRN_PID, r12
+       mtcr    r11
+       mfspr   r9, SPRN_SPRG5
+       mfspr   r12, SPRN_SPRG4
+#endif
+       mfspr   r11, SPRN_SPRG1
+       mfspr   r10, SPRN_SPRG0
+       b       DataAccess
+
+/*
+ * 0x0400 - Instruction Storage Exception
+ * This is caused by a fetch from non-execute or guarded pages.
+ */
+       START_EXCEPTION(0x0400, InstructionAccess)
+       NORMAL_EXCEPTION_PROLOG
+       mr      r4,r12                  /* Pass SRR0 as arg2 */
+       li      r5,0                    /* Pass zero as arg3 */
+       EXC_XFER_EE_LITE(0x400, handle_page_fault)
+
+/* 0x0500 - External Interrupt Exception */
+       EXCEPTION(0x0500, HardwareInterrupt, do_IRQ, EXC_XFER_LITE)
+
+/* 0x0600 - Alignment Exception */
+       START_EXCEPTION(0x0600, Alignment)
+       NORMAL_EXCEPTION_PROLOG
+       mfspr   r4,SPRN_DEAR            /* Grab the DEAR and save it */
+       stw     r4,_DEAR(r11)
+       addi    r3,r1,STACK_FRAME_OVERHEAD
+       EXC_XFER_EE(0x600, AlignmentException)
+
+/* 0x0700 - Program Exception */
+       START_EXCEPTION(0x0700, ProgramCheck)
+       NORMAL_EXCEPTION_PROLOG
+       mfspr   r4,SPRN_ESR             /* Grab the ESR and save it */
+       stw     r4,_ESR(r11)
+       addi    r3,r1,STACK_FRAME_OVERHEAD
+       EXC_XFER_STD(0x700, ProgramCheckException)
+
+       EXCEPTION(0x0800, Trap_08, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x0900, Trap_09, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x0A00, Trap_0A, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x0B00, Trap_0B, UnknownException, EXC_XFER_EE)
+
+/* 0x0C00 - System Call Exception */
+       START_EXCEPTION(0x0C00, SystemCall)
+       NORMAL_EXCEPTION_PROLOG
+       EXC_XFER_EE_LITE(0xc00, DoSyscall)
+
+       EXCEPTION(0x0D00, Trap_0D, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x0E00, Trap_0E, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x0F00, Trap_0F, UnknownException, EXC_XFER_EE)
+
+/* 0x1000 - Programmable Interval Timer (PIT) Exception */
+       START_EXCEPTION(0x1000, Decrementer)
+       NORMAL_EXCEPTION_PROLOG
+       lis     r0,TSR_PIS@h
+       mtspr   SPRN_TSR,r0             /* Clear the PIT exception */
+       addi    r3,r1,STACK_FRAME_OVERHEAD
+       EXC_XFER_LITE(0x1000, timer_interrupt)
+
+#if 0
+/* NOTE:
+ * FIT and WDT handlers are not implemented yet.
+ */
+
+/* 0x1010 - Fixed Interval Timer (FIT) Exception
+*/
+       STND_EXCEPTION(0x1010,  FITException,           UnknownException)
+
+/* 0x1020 - Watchdog Timer (WDT) Exception
+*/
+#ifdef CONFIG_BOOKE_WDT
+       CRITICAL_EXCEPTION(0x1020, WDTException, WatchdogException)
+#else
+       CRITICAL_EXCEPTION(0x1020, WDTException, UnknownException)
+#endif
+#endif
+
+/* 0x1100 - Data TLB Miss Exception
+ * As the name implies, translation is not in the MMU, so search the
+ * page tables and fix it.  The only purpose of this function is to
+ * load TLB entries from the page table if they exist.
+ */
+       START_EXCEPTION(0x1100, DTLBMiss)
+       mtspr   SPRN_SPRG0, r10         /* Save some working registers */
+       mtspr   SPRN_SPRG1, r11
+#ifdef CONFIG_403GCX
+       stw     r12, 0(r0)
+       stw     r9, 4(r0)
+       mfcr    r11
+       mfspr   r12, SPRN_PID
+       stw     r11, 8(r0)
+       stw     r12, 12(r0)
+#else
+       mtspr   SPRN_SPRG4, r12
+       mtspr   SPRN_SPRG5, r9
+       mfcr    r11
+       mfspr   r12, SPRN_PID
+       mtspr   SPRN_SPRG7, r11
+       mtspr   SPRN_SPRG6, r12
+#endif
+       mfspr   r10, SPRN_DEAR          /* Get faulting address */
+
+       /* If we are faulting a kernel address, we have to use the
+        * kernel page tables.
+        */
+       lis     r11, TASK_SIZE@h
+       cmplw   r10, r11
+       blt+    3f
+       lis     r11, swapper_pg_dir@h
+       ori     r11, r11, swapper_pg_dir@l
+       li      r9, 0
+       mtspr   SPRN_PID, r9            /* TLB will have 0 TID */
+       b       4f
+
+       /* Get the PGD for the current thread.
+        */
+3:
+       mfspr   r11,SPRN_SPRG3
+       lwz     r11,PGDIR(r11)
+4:
+       tophys(r11, r11)
+       rlwimi  r11, r10, 12, 20, 29    /* Create L1 (pgdir/pmd) address */
+       lwz     r12, 0(r11)             /* Get L1 entry */
+       andi.   r9, r12, _PMD_PRESENT   /* Check if it points to a PTE page */
+       beq     2f                      /* Bail if no table */
+
+       rlwimi  r12, r10, 22, 20, 29    /* Compute PTE address */
+       lwz     r11, 0(r12)             /* Get Linux PTE */
+       andi.   r9, r11, _PAGE_PRESENT
+       beq     5f
+
+       ori     r11, r11, _PAGE_ACCESSED
+       stw     r11, 0(r12)
+
+       /* Create TLB tag.  This is the faulting address plus a static
+        * set of bits.  These are size, valid, E, U0.
+       */
+       li      r12, 0x00c0
+       rlwimi  r10, r12, 0, 20, 31
+
+       b       finish_tlb_load
+
+2:     /* Check for possible large-page pmd entry */
+       rlwinm. r9, r12, 2, 22, 24
+       beq     5f
+
+       /* Create TLB tag.  This is the faulting address, plus a static
+        * set of bits (valid, E, U0) plus the size from the PMD.
+        */
+       ori     r9, r9, 0x40
+       rlwimi  r10, r9, 0, 20, 31
+       mr      r11, r12
+
+       b       finish_tlb_load
+
+5:
+       /* The bailout.  Restore registers to pre-exception conditions
+        * and call the heavyweights to help us out.
+        */
+#ifdef CONFIG_403GCX
+       lwz     r12, 12(r0)
+       lwz     r11, 8(r0)
+       mtspr   SPRN_PID, r12
+       mtcr    r11
+       lwz     r9, 4(r0)
+       lwz     r12, 0(r0)
+#else
+       mfspr   r12, SPRN_SPRG6
+       mfspr   r11, SPRN_SPRG7
+       mtspr   SPRN_PID, r12
+       mtcr    r11
+       mfspr   r9, SPRN_SPRG5
+       mfspr   r12, SPRN_SPRG4
+#endif
+       mfspr   r11, SPRN_SPRG1
+       mfspr   r10, SPRN_SPRG0
+       b       DataAccess
+
+/* 0x1200 - Instruction TLB Miss Exception
+ * Nearly the same as above, except we get our information from different
+ * registers and bailout to a different point.
+ */
+       START_EXCEPTION(0x1200, ITLBMiss)
+       mtspr   SPRN_SPRG0, r10         /* Save some working registers */
+       mtspr   SPRN_SPRG1, r11
+#ifdef CONFIG_403GCX
+       stw     r12, 0(r0)
+       stw     r9, 4(r0)
+       mfcr    r11
+       mfspr   r12, SPRN_PID
+       stw     r11, 8(r0)
+       stw     r12, 12(r0)
+#else
+       mtspr   SPRN_SPRG4, r12
+       mtspr   SPRN_SPRG5, r9
+       mfcr    r11
+       mfspr   r12, SPRN_PID
+       mtspr   SPRN_SPRG7, r11
+       mtspr   SPRN_SPRG6, r12
+#endif
+       mfspr   r10, SPRN_SRR0          /* Get faulting address */
+
+       /* If we are faulting a kernel address, we have to use the
+        * kernel page tables.
+        */
+       lis     r11, TASK_SIZE@h
+       cmplw   r10, r11
+       blt+    3f
+       lis     r11, swapper_pg_dir@h
+       ori     r11, r11, swapper_pg_dir@l
+       li      r9, 0
+       mtspr   SPRN_PID, r9            /* TLB will have 0 TID */
+       b       4f
+
+       /* Get the PGD for the current thread.
+        */
+3:
+       mfspr   r11,SPRN_SPRG3
+       lwz     r11,PGDIR(r11)
+4:
+       tophys(r11, r11)
+       rlwimi  r11, r10, 12, 20, 29    /* Create L1 (pgdir/pmd) address */
+       lwz     r12, 0(r11)             /* Get L1 entry */
+       andi.   r9, r12, _PMD_PRESENT   /* Check if it points to a PTE page */
+       beq     2f                      /* Bail if no table */
+
+       rlwimi  r12, r10, 22, 20, 29    /* Compute PTE address */
+       lwz     r11, 0(r12)             /* Get Linux PTE */
+       andi.   r9, r11, _PAGE_PRESENT
+       beq     5f
+
+       ori     r11, r11, _PAGE_ACCESSED
+       stw     r11, 0(r12)
+
+       /* Create TLB tag.  This is the faulting address plus a static
+        * set of bits.  These are size, valid, E, U0.
+       */
+       li      r12, 0x00c0
+       rlwimi  r10, r12, 0, 20, 31
+
+       b       finish_tlb_load
+
+2:     /* Check for possible large-page pmd entry */
+       rlwinm. r9, r12, 2, 22, 24
+       beq     5f
+
+       /* Create TLB tag.  This is the faulting address, plus a static
+        * set of bits (valid, E, U0) plus the size from the PMD.
+        */
+       ori     r9, r9, 0x40
+       rlwimi  r10, r9, 0, 20, 31
+       mr      r11, r12
+
+       b       finish_tlb_load
+
+5:
+       /* The bailout.  Restore registers to pre-exception conditions
+        * and call the heavyweights to help us out.
+        */
+#ifdef CONFIG_403GCX
+       lwz     r12, 12(r0)
+       lwz     r11, 8(r0)
+       mtspr   SPRN_PID, r12
+       mtcr    r11
+       lwz     r9, 4(r0)
+       lwz     r12, 0(r0)
+#else
+       mfspr   r12, SPRN_SPRG6
+       mfspr   r11, SPRN_SPRG7
+       mtspr   SPRN_PID, r12
+       mtcr    r11
+       mfspr   r9, SPRN_SPRG5
+       mfspr   r12, SPRN_SPRG4
+#endif
+       mfspr   r11, SPRN_SPRG1
+       mfspr   r10, SPRN_SPRG0
+       b       InstructionAccess
+
+       EXCEPTION(0x1300, Trap_13, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x1400, Trap_14, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x1500, Trap_15, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x1600, Trap_16, UnknownException, EXC_XFER_EE)
+#ifdef CONFIG_IBM405_ERR51
+       /* 405GP errata 51 */
+       START_EXCEPTION(0x1700, Trap_17)
+       b DTLBMiss
+#else
+       EXCEPTION(0x1700, Trap_17, UnknownException, EXC_XFER_EE)
+#endif
+       EXCEPTION(0x1800, Trap_18, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x1900, Trap_19, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x1A00, Trap_1A, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x1B00, Trap_1B, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x1C00, Trap_1C, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x1D00, Trap_1D, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x1E00, Trap_1E, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x1F00, Trap_1F, UnknownException, EXC_XFER_EE)
+
+/* Check for a single step debug exception while in an exception
+ * handler before state has been saved.  This is to catch the case
+ * where an instruction that we are trying to single step causes
+ * an exception (eg ITLB/DTLB miss) and thus the first instruction of
+ * the exception handler generates a single step debug exception.
+ *
+ * If we get a debug trap on the first instruction of an exception handler,
+ * we reset the MSR_DE in the _exception handler's_ MSR (the debug trap is
+ * a critical exception, so we are using SPRN_CSRR1 to manipulate the MSR).
+ * The exception handler was handling a non-critical interrupt, so it will
+ * save (and later restore) the MSR via SPRN_SRR1, which will still have
+ * the MSR_DE bit set.
+ */
+       /* 0x2000 - Debug Exception */
+       START_EXCEPTION(0x2000, DebugTrap)
+       CRITICAL_EXCEPTION_PROLOG
+
+       /*
+        * If this is a single step or branch-taken exception in an
+        * exception entry sequence, it was probably meant to apply to
+        * the code where the exception occurred (since exception entry
+        * doesn't turn off DE automatically).  We simulate the effect
+        * of turning off DE on entry to an exception handler by turning
+        * off DE in the SRR3 value and clearing the debug status.
+        */
+       mfspr   r10,SPRN_DBSR           /* check single-step/branch taken */
+       andis.  r10,r10,DBSR_IC@h
+       beq+    2f
+
+       andi.   r10,r9,MSR_IR|MSR_PR    /* check supervisor + MMU off */
+       beq     1f                      /* branch and fix it up */
+
+       mfspr   r10,SPRN_SRR2           /* Faulting instruction address */
+       cmplwi  r10,0x2100
+       bgt+    2f                      /* address above exception vectors */
+
+       /* here it looks like we got an inappropriate debug exception. */
+1:     rlwinm  r9,r9,0,~MSR_DE         /* clear DE in the SRR3 value */
+       lis     r10,DBSR_IC@h           /* clear the IC event */
+       mtspr   SPRN_DBSR,r10
+       /* restore state and get out */
+       lwz     r10,_CCR(r11)
+       lwz     r0,GPR0(r11)
+       lwz     r1,GPR1(r11)
+       mtcrf   0x80,r10
+       mtspr   SPRN_SRR2,r12
+       mtspr   SPRN_SRR3,r9
+       lwz     r9,GPR9(r11)
+       lwz     r12,GPR12(r11)
+       lwz     r10,crit_r10@l(0)
+       lwz     r11,crit_r11@l(0)
+       PPC405_ERR77_SYNC
+       rfci
+       b       .
+
+       /* continue normal handling for a critical exception... */
+2:     mfspr   r4,SPRN_DBSR
+       addi    r3,r1,STACK_FRAME_OVERHEAD
+       EXC_XFER_TEMPLATE(DebugException, 0x2002, \
+               (MSR_KERNEL & ~(MSR_ME|MSR_DE|MSR_CE)), \
+               NOCOPY, crit_transfer_to_handler, ret_from_crit_exc)
+
+/*
+ * The other Data TLB exceptions bail out to this point
+ * if they can't resolve the lightweight TLB fault.
+ */
+DataAccess:
+       NORMAL_EXCEPTION_PROLOG
+       mfspr   r5,SPRN_ESR             /* Grab the ESR, save it, pass arg3 */
+       stw     r5,_ESR(r11)
+       mfspr   r4,SPRN_DEAR            /* Grab the DEAR, save it, pass arg2 */
+       EXC_XFER_EE_LITE(0x300, handle_page_fault)
+
+/* Other PowerPC processors, namely those derived from the 6xx-series
+ * have vectors from 0x2100 through 0x2F00 defined, but marked as reserved.
+ * However, for the 4xx-series processors these are neither defined nor
+ * reserved.
+ */
+
+       /* Damn, I came up one instruction too many to fit into the
+        * exception space :-).  Both the instruction and data TLB
+        * miss get to this point to load the TLB.
+        *      r10 - TLB_TAG value
+        *      r11 - Linux PTE
+        *      r12, r9 - avilable to use
+        *      PID - loaded with proper value when we get here
+        *      Upon exit, we reload everything and RFI.
+        * Actually, it will fit now, but oh well.....a common place
+        * to load the TLB.
+        */
+tlb_4xx_index:
+       .long   0
+finish_tlb_load:
+       /* load the next available TLB index.
+       */
+       lwz     r9, tlb_4xx_index@l(0)
+       addi    r9, r9, 1
+       andi.   r9, r9, (PPC4XX_TLB_SIZE-1)
+       stw     r9, tlb_4xx_index@l(0)
+
+6:
+       /*
+        * Clear out the software-only bits in the PTE to generate the
+        * TLB_DATA value.  These are the bottom 2 bits of the RPM, the
+        * top 3 bits of the zone field, and M.
+        */
+       li      r12, 0x0ce2
+       andc    r11, r11, r12
+
+       tlbwe   r11, r9, TLB_DATA               /* Load TLB LO */
+       tlbwe   r10, r9, TLB_TAG                /* Load TLB HI */
+
+       /* Done...restore registers and get out of here.
+       */
+#ifdef CONFIG_403GCX
+       lwz     r12, 12(r0)
+       lwz     r11, 8(r0)
+       mtspr   SPRN_PID, r12
+       mtcr    r11
+       lwz     r9, 4(r0)
+       lwz     r12, 0(r0)
+#else
+       mfspr   r12, SPRN_SPRG6
+       mfspr   r11, SPRN_SPRG7
+       mtspr   SPRN_PID, r12
+       mtcr    r11
+       mfspr   r9, SPRN_SPRG5
+       mfspr   r12, SPRN_SPRG4
+#endif
+       mfspr   r11, SPRN_SPRG1
+       mfspr   r10, SPRN_SPRG0
+       PPC405_ERR77_SYNC
+       rfi                     /* Should sync shadow TLBs */
+       b       .               /* prevent prefetch past rfi */
+
+/* extern void giveup_fpu(struct task_struct *prev)
+ *
+ * The PowerPC 4xx family of processors do not have an FPU, so this just
+ * returns.
+ */
+_GLOBAL(giveup_fpu)
+       blr
+
+/* This is where the main kernel code starts.
+ */
+start_here:
+
+       /* ptr to current */
+       lis     r2,init_task@h
+       ori     r2,r2,init_task@l
+
+       /* ptr to phys current thread */
+       tophys(r4,r2)
+       addi    r4,r4,THREAD    /* init task's THREAD */
+       mtspr   SPRN_SPRG3,r4
+
+       /* stack */
+       lis     r1,init_thread_union@ha
+       addi    r1,r1,init_thread_union@l
+       li      r0,0
+       stwu    r0,THREAD_SIZE-STACK_FRAME_OVERHEAD(r1)
+
+       bl      early_init      /* We have to do this with MMU on */
+
+/*
+ * Decide what sort of machine this is and initialize the MMU.
+ */
+       mr      r3,r31
+       mr      r4,r30
+       mr      r5,r29
+       mr      r6,r28
+       mr      r7,r27
+       bl      machine_init
+       bl      MMU_init
+
+/* Go back to running unmapped so we can load up new values
+ * and change to using our exception vectors.
+ * On the 4xx, all we have to do is invalidate the TLB to clear
+ * the old 16M byte TLB mappings.
+ */
+       lis     r4,2f@h
+       ori     r4,r4,2f@l
+       tophys(r4,r4)
+       lis     r3,(MSR_KERNEL & ~(MSR_IR|MSR_DR))@h
+       ori     r3,r3,(MSR_KERNEL & ~(MSR_IR|MSR_DR))@l
+       mtspr   SPRN_SRR0,r4
+       mtspr   SPRN_SRR1,r3
+       rfi
+       b       .               /* prevent prefetch past rfi */
+
+/* Load up the kernel context */
+2:
+       sync                    /* Flush to memory before changing TLB */
+       tlbia
+       isync                   /* Flush shadow TLBs */
+
+       /* set up the PTE pointers for the Abatron bdiGDB.
+       */
+       lis     r6, swapper_pg_dir@h
+       ori     r6, r6, swapper_pg_dir@l
+       lis     r5, abatron_pteptrs@h
+       ori     r5, r5, abatron_pteptrs@l
+       stw     r5, 0xf0(r0)    /* Must match your Abatron config file */
+       tophys(r5,r5)
+       stw     r6, 0(r5)
+
+/* Now turn on the MMU for real! */
+       lis     r4,MSR_KERNEL@h
+       ori     r4,r4,MSR_KERNEL@l
+       lis     r3,start_kernel@h
+       ori     r3,r3,start_kernel@l
+       mtspr   SPRN_SRR0,r3
+       mtspr   SPRN_SRR1,r4
+       rfi                     /* enable MMU and jump to start_kernel */
+       b       .               /* prevent prefetch past rfi */
+
+/* Set up the initial MMU state so we can do the first level of
+ * kernel initialization.  This maps the first 16 MBytes of memory 1:1
+ * virtual to physical and more importantly sets the cache mode.
+ */
+initial_mmu:
+       tlbia                   /* Invalidate all TLB entries */
+       isync
+
+       /* We should still be executing code at physical address 0x0000xxxx
+        * at this point. However, start_here is at virtual address
+        * 0xC000xxxx. So, set up a TLB mapping to cover this once
+        * translation is enabled.
+        */
+
+       lis     r3,KERNELBASE@h         /* Load the kernel virtual address */
+       ori     r3,r3,KERNELBASE@l
+       tophys(r4,r3)                   /* Load the kernel physical address */
+
+       iccci   r0,r3                   /* Invalidate the i-cache before use */
+
+       /* Load the kernel PID.
+       */
+       li      r0,0
+       mtspr   SPRN_PID,r0
+       sync
+
+       /* Configure and load two entries into TLB slots 62 and 63.
+        * In case we are pinning TLBs, these are reserved in by the
+        * other TLB functions.  If not reserving, then it doesn't
+        * matter where they are loaded.
+        */
+       clrrwi  r4,r4,10                /* Mask off the real page number */
+       ori     r4,r4,(TLB_WR | TLB_EX) /* Set the write and execute bits */
+
+       clrrwi  r3,r3,10                /* Mask off the effective page number */
+       ori     r3,r3,(TLB_VALID | TLB_PAGESZ(PAGESZ_16M))
+
+        li      r0,63                    /* TLB slot 63 */
+
+       tlbwe   r4,r0,TLB_DATA          /* Load the data portion of the entry */
+       tlbwe   r3,r0,TLB_TAG           /* Load the tag portion of the entry */
+
+#if defined(CONFIG_SERIAL_TEXT_DEBUG) && defined(SERIAL_DEBUG_IO_BASE)
+
+       /* Load a TLB entry for the UART, so that ppc4xx_progress() can use
+        * the UARTs nice and early.  We use a 4k real==virtual mapping. */
+
+       lis     r3,SERIAL_DEBUG_IO_BASE@h
+       ori     r3,r3,SERIAL_DEBUG_IO_BASE@l
+       mr      r4,r3
+       clrrwi  r4,r4,12
+       ori     r4,r4,(TLB_WR|TLB_I|TLB_M|TLB_G)
+
+       clrrwi  r3,r3,12
+       ori     r3,r3,(TLB_VALID | TLB_PAGESZ(PAGESZ_4K))
+
+       li      r0,0                    /* TLB slot 0 */
+       tlbwe   r4,r0,TLB_DATA
+       tlbwe   r3,r0,TLB_TAG
+#endif /* CONFIG_SERIAL_DEBUG_TEXT && SERIAL_DEBUG_IO_BASE */
+
+       isync
+
+       /* Establish the exception vector base
+       */
+       lis     r4,KERNELBASE@h         /* EVPR only uses the high 16-bits */
+       tophys(r0,r4)                   /* Use the physical address */
+       mtspr   SPRN_EVPR,r0
+
+       blr
+
+_GLOBAL(abort)
+        mfspr   r13,SPRN_DBCR0
+        oris    r13,r13,DBCR0_RST_SYSTEM@h
+        mtspr   SPRN_DBCR0,r13
+
+_GLOBAL(set_context)
+
+#ifdef CONFIG_BDI_SWITCH
+       /* Context switch the PTE pointer for the Abatron BDI2000.
+        * The PGDIR is the second parameter.
+        */
+       lis     r5, KERNELBASE@h
+       lwz     r5, 0xf0(r5)
+       stw     r4, 0x4(r5)
+#endif
+       sync
+       mtspr   SPRN_PID,r3
+       isync                           /* Need an isync to flush shadow */
+                                       /* TLBs after changing PID */
+       blr
+
+/* We put a few things here that have to be page-aligned. This stuff
+ * goes at the beginning of the data segment, which is page-aligned.
+ */
+       .data
+_GLOBAL(sdata)
+_GLOBAL(empty_zero_page)
+       .space  4096
+_GLOBAL(swapper_pg_dir)
+       .space  4096
+
+
+/* Stack for handling critical exceptions from kernel mode */
+       .section .bss
+        .align 12
+exception_stack_bottom:
+       .space  4096
+critical_stack_top:
+_GLOBAL(exception_stack_top)
+
+/* This space gets a copy of optional info passed to us by the bootstrap
+ * which is used to pass parameters into the kernel like root=/dev/sda1, etc.
+ */
+_GLOBAL(cmd_line)
+       .space  512
+
+/* Room for two PTE pointers, usually the kernel and current user pointers
+ * to their respective root page table.
+ */
+abatron_pteptrs:
+       .space  8
diff --git a/arch/powerpc/kernel/head_64.S b/arch/powerpc/kernel/head_64.S
new file mode 100644 (file)
index 0000000..22a5ee0
--- /dev/null
@@ -0,0 +1,2011 @@
+/*
+ *  arch/ppc64/kernel/head.S
+ *
+ *  PowerPC version
+ *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
+ *
+ *  Rewritten by Cort Dougan (cort@cs.nmt.edu) for PReP
+ *    Copyright (C) 1996 Cort Dougan <cort@cs.nmt.edu>
+ *  Adapted for Power Macintosh by Paul Mackerras.
+ *  Low-level exception handlers and MMU support
+ *  rewritten by Paul Mackerras.
+ *    Copyright (C) 1996 Paul Mackerras.
+ *
+ *  Adapted for 64bit PowerPC by Dave Engebretsen, Peter Bergner, and
+ *    Mike Corrigan {engebret|bergner|mikejc}@us.ibm.com
+ *
+ *  This file contains the low-level support and setup for the
+ *  PowerPC-64 platform, including trap and interrupt dispatch.
+ *
+ *  This program is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU General Public License
+ *  as published by the Free Software Foundation; either version
+ *  2 of the License, or (at your option) any later version.
+ */
+
+#include <linux/config.h>
+#include <linux/threads.h>
+#include <asm/processor.h>
+#include <asm/page.h>
+#include <asm/mmu.h>
+#include <asm/systemcfg.h>
+#include <asm/ppc_asm.h>
+#include <asm/asm-offsets.h>
+#include <asm/bug.h>
+#include <asm/cputable.h>
+#include <asm/setup.h>
+#include <asm/hvcall.h>
+#include <asm/iSeries/LparMap.h>
+
+#ifdef CONFIG_PPC_ISERIES
+#define DO_SOFT_DISABLE
+#endif
+
+/*
+ * We layout physical memory as follows:
+ * 0x0000 - 0x00ff : Secondary processor spin code
+ * 0x0100 - 0x2fff : pSeries Interrupt prologs
+ * 0x3000 - 0x5fff : interrupt support, iSeries and common interrupt prologs
+ * 0x6000 - 0x6fff : Initial (CPU0) segment table
+ * 0x7000 - 0x7fff : FWNMI data area
+ * 0x8000 -        : Early init and support code
+ */
+
+/*
+ *   SPRG Usage
+ *
+ *   Register  Definition
+ *
+ *   SPRG0     reserved for hypervisor
+ *   SPRG1     temp - used to save gpr
+ *   SPRG2     temp - used to save gpr
+ *   SPRG3     virt addr of paca
+ */
+
+/*
+ * Entering into this code we make the following assumptions:
+ *  For pSeries:
+ *   1. The MMU is off & open firmware is running in real mode.
+ *   2. The kernel is entered at __start
+ *
+ *  For iSeries:
+ *   1. The MMU is on (as it always is for iSeries)
+ *   2. The kernel is entered at system_reset_iSeries
+ */
+
+       .text
+       .globl  _stext
+_stext:
+#ifdef CONFIG_PPC_MULTIPLATFORM
+_GLOBAL(__start)
+       /* NOP this out unconditionally */
+BEGIN_FTR_SECTION
+       b .__start_initialization_multiplatform
+END_FTR_SECTION(0, 1)
+#endif /* CONFIG_PPC_MULTIPLATFORM */
+
+       /* Catch branch to 0 in real mode */
+       trap
+
+#ifdef CONFIG_PPC_ISERIES
+       /*
+        * At offset 0x20, there is a pointer to iSeries LPAR data.
+        * This is required by the hypervisor
+        */
+       . = 0x20
+       .llong hvReleaseData-KERNELBASE
+
+       /*
+        * At offset 0x28 and 0x30 are offsets to the mschunks_map
+        * array (used by the iSeries LPAR debugger to do translation
+        * between physical addresses and absolute addresses) and
+        * to the pidhash table (also used by the debugger)
+        */
+       .llong mschunks_map-KERNELBASE
+       .llong 0        /* pidhash-KERNELBASE SFRXXX */
+
+       /* Offset 0x38 - Pointer to start of embedded System.map */
+       .globl  embedded_sysmap_start
+embedded_sysmap_start:
+       .llong  0
+       /* Offset 0x40 - Pointer to end of embedded System.map */
+       .globl  embedded_sysmap_end
+embedded_sysmap_end:
+       .llong  0
+
+#endif /* CONFIG_PPC_ISERIES */
+
+       /* Secondary processors spin on this value until it goes to 1. */
+       .globl  __secondary_hold_spinloop
+__secondary_hold_spinloop:
+       .llong  0x0
+
+       /* Secondary processors write this value with their cpu # */
+       /* after they enter the spin loop immediately below.      */
+       .globl  __secondary_hold_acknowledge
+__secondary_hold_acknowledge:
+       .llong  0x0
+
+       . = 0x60
+/*
+ * The following code is used on pSeries to hold secondary processors
+ * in a spin loop after they have been freed from OpenFirmware, but
+ * before the bulk of the kernel has been relocated.  This code
+ * is relocated to physical address 0x60 before prom_init is run.
+ * All of it must fit below the first exception vector at 0x100.
+ */
+_GLOBAL(__secondary_hold)
+       mfmsr   r24
+       ori     r24,r24,MSR_RI
+       mtmsrd  r24                     /* RI on */
+
+       /* Grab our linux cpu number */
+       mr      r24,r3
+
+       /* Tell the master cpu we're here */
+       /* Relocation is off & we are located at an address less */
+       /* than 0x100, so only need to grab low order offset.    */
+       std     r24,__secondary_hold_acknowledge@l(0)
+       sync
+
+       /* All secondary cpus wait here until told to start. */
+100:   ld      r4,__secondary_hold_spinloop@l(0)
+       cmpdi   0,r4,1
+       bne     100b
+
+#ifdef CONFIG_HMT
+       b       .hmt_init
+#else
+#ifdef CONFIG_SMP
+       mr      r3,r24
+       b       .pSeries_secondary_smp_init
+#else
+       BUG_OPCODE
+#endif
+#endif
+
+/* This value is used to mark exception frames on the stack. */
+       .section ".toc","aw"
+exception_marker:
+       .tc     ID_72656773_68657265[TC],0x7265677368657265
+       .text
+
+/*
+ * The following macros define the code that appears as
+ * the prologue to each of the exception handlers.  They
+ * are split into two parts to allow a single kernel binary
+ * to be used for pSeries and iSeries.
+ * LOL.  One day... - paulus
+ */
+
+/*
+ * We make as much of the exception code common between native
+ * exception handlers (including pSeries LPAR) and iSeries LPAR
+ * implementations as possible.
+ */
+
+/*
+ * This is the start of the interrupt handlers for pSeries
+ * This code runs with relocation off.
+ */
+#define EX_R9          0
+#define EX_R10         8
+#define EX_R11         16
+#define EX_R12         24
+#define EX_R13         32
+#define EX_SRR0                40
+#define EX_R3          40      /* SLB miss saves R3, but not SRR0 */
+#define EX_DAR         48
+#define EX_LR          48      /* SLB miss saves LR, but not DAR */
+#define EX_DSISR       56
+#define EX_CCR         60
+
+#define EXCEPTION_PROLOG_PSERIES(area, label)                          \
+       mfspr   r13,SPRG3;              /* get paca address into r13 */ \
+       std     r9,area+EX_R9(r13);     /* save r9 - r12 */             \
+       std     r10,area+EX_R10(r13);                                   \
+       std     r11,area+EX_R11(r13);                                   \
+       std     r12,area+EX_R12(r13);                                   \
+       mfspr   r9,SPRG1;                                               \
+       std     r9,area+EX_R13(r13);                                    \
+       mfcr    r9;                                                     \
+       clrrdi  r12,r13,32;             /* get high part of &label */   \
+       mfmsr   r10;                                                    \
+       mfspr   r11,SRR0;               /* save SRR0 */                 \
+       ori     r12,r12,(label)@l;      /* virt addr of handler */      \
+       ori     r10,r10,MSR_IR|MSR_DR|MSR_RI;                           \
+       mtspr   SRR0,r12;                                               \
+       mfspr   r12,SRR1;               /* and SRR1 */                  \
+       mtspr   SRR1,r10;                                               \
+       rfid;                                                           \
+       b       .       /* prevent speculative execution */
+
+/*
+ * This is the start of the interrupt handlers for iSeries
+ * This code runs with relocation on.
+ */
+#define EXCEPTION_PROLOG_ISERIES_1(area)                               \
+       mfspr   r13,SPRG3;              /* get paca address into r13 */ \
+       std     r9,area+EX_R9(r13);     /* save r9 - r12 */             \
+       std     r10,area+EX_R10(r13);                                   \
+       std     r11,area+EX_R11(r13);                                   \
+       std     r12,area+EX_R12(r13);                                   \
+       mfspr   r9,SPRG1;                                               \
+       std     r9,area+EX_R13(r13);                                    \
+       mfcr    r9
+
+#define EXCEPTION_PROLOG_ISERIES_2                                     \
+       mfmsr   r10;                                                    \
+       ld      r11,PACALPPACA+LPPACASRR0(r13);                         \
+       ld      r12,PACALPPACA+LPPACASRR1(r13);                         \
+       ori     r10,r10,MSR_RI;                                         \
+       mtmsrd  r10,1
+
+/*
+ * The common exception prolog is used for all except a few exceptions
+ * such as a segment miss on a kernel address.  We have to be prepared
+ * to take another exception from the point where we first touch the
+ * kernel stack onwards.
+ *
+ * On entry r13 points to the paca, r9-r13 are saved in the paca,
+ * r9 contains the saved CR, r11 and r12 contain the saved SRR0 and
+ * SRR1, and relocation is on.
+ */
+#define EXCEPTION_PROLOG_COMMON(n, area)                                  \
+       andi.   r10,r12,MSR_PR;         /* See if coming from user      */ \
+       mr      r10,r1;                 /* Save r1                      */ \
+       subi    r1,r1,INT_FRAME_SIZE;   /* alloc frame on kernel stack  */ \
+       beq-    1f;                                                        \
+       ld      r1,PACAKSAVE(r13);      /* kernel stack to use          */ \
+1:     cmpdi   cr1,r1,0;               /* check if r1 is in userspace  */ \
+       bge-    cr1,bad_stack;          /* abort if it is               */ \
+       std     r9,_CCR(r1);            /* save CR in stackframe        */ \
+       std     r11,_NIP(r1);           /* save SRR0 in stackframe      */ \
+       std     r12,_MSR(r1);           /* save SRR1 in stackframe      */ \
+       std     r10,0(r1);              /* make stack chain pointer     */ \
+       std     r0,GPR0(r1);            /* save r0 in stackframe        */ \
+       std     r10,GPR1(r1);           /* save r1 in stackframe        */ \
+       std     r2,GPR2(r1);            /* save r2 in stackframe        */ \
+       SAVE_4GPRS(3, r1);              /* save r3 - r6 in stackframe   */ \
+       SAVE_2GPRS(7, r1);              /* save r7, r8 in stackframe    */ \
+       ld      r9,area+EX_R9(r13);     /* move r9, r10 to stackframe   */ \
+       ld      r10,area+EX_R10(r13);                                      \
+       std     r9,GPR9(r1);                                               \
+       std     r10,GPR10(r1);                                             \
+       ld      r9,area+EX_R11(r13);    /* move r11 - r13 to stackframe */ \
+       ld      r10,area+EX_R12(r13);                                      \
+       ld      r11,area+EX_R13(r13);                                      \
+       std     r9,GPR11(r1);                                              \
+       std     r10,GPR12(r1);                                             \
+       std     r11,GPR13(r1);                                             \
+       ld      r2,PACATOC(r13);        /* get kernel TOC into r2       */ \
+       mflr    r9;                     /* save LR in stackframe        */ \
+       std     r9,_LINK(r1);                                              \
+       mfctr   r10;                    /* save CTR in stackframe       */ \
+       std     r10,_CTR(r1);                                              \
+       mfspr   r11,XER;                /* save XER in stackframe       */ \
+       std     r11,_XER(r1);                                              \
+       li      r9,(n)+1;                                                  \
+       std     r9,_TRAP(r1);           /* set trap number              */ \
+       li      r10,0;                                                     \
+       ld      r11,exception_marker@toc(r2);                              \
+       std     r10,RESULT(r1);         /* clear regs->result           */ \
+       std     r11,STACK_FRAME_OVERHEAD-16(r1); /* mark the frame      */
+
+/*
+ * Exception vectors.
+ */
+#define STD_EXCEPTION_PSERIES(n, label)                        \
+       . = n;                                          \
+       .globl label##_pSeries;                         \
+label##_pSeries:                                       \
+       HMT_MEDIUM;                                     \
+       mtspr   SPRG1,r13;              /* save r13 */  \
+       RUNLATCH_ON(r13);                               \
+       EXCEPTION_PROLOG_PSERIES(PACA_EXGEN, label##_common)
+
+#define STD_EXCEPTION_ISERIES(n, label, area)          \
+       .globl label##_iSeries;                         \
+label##_iSeries:                                       \
+       HMT_MEDIUM;                                     \
+       mtspr   SPRG1,r13;              /* save r13 */  \
+       RUNLATCH_ON(r13);                               \
+       EXCEPTION_PROLOG_ISERIES_1(area);               \
+       EXCEPTION_PROLOG_ISERIES_2;                     \
+       b       label##_common
+
+#define MASKABLE_EXCEPTION_ISERIES(n, label)                           \
+       .globl label##_iSeries;                                         \
+label##_iSeries:                                                       \
+       HMT_MEDIUM;                                                     \
+       mtspr   SPRG1,r13;              /* save r13 */                  \
+       RUNLATCH_ON(r13);                                               \
+       EXCEPTION_PROLOG_ISERIES_1(PACA_EXGEN);                         \
+       lbz     r10,PACAPROCENABLED(r13);                               \
+       cmpwi   0,r10,0;                                                \
+       beq-    label##_iSeries_masked;                                 \
+       EXCEPTION_PROLOG_ISERIES_2;                                     \
+       b       label##_common;                                         \
+
+#ifdef DO_SOFT_DISABLE
+#define DISABLE_INTS                           \
+       lbz     r10,PACAPROCENABLED(r13);       \
+       li      r11,0;                          \
+       std     r10,SOFTE(r1);                  \
+       mfmsr   r10;                            \
+       stb     r11,PACAPROCENABLED(r13);       \
+       ori     r10,r10,MSR_EE;                 \
+       mtmsrd  r10,1
+
+#define ENABLE_INTS                            \
+       lbz     r10,PACAPROCENABLED(r13);       \
+       mfmsr   r11;                            \
+       std     r10,SOFTE(r1);                  \
+       ori     r11,r11,MSR_EE;                 \
+       mtmsrd  r11,1
+
+#else  /* hard enable/disable interrupts */
+#define DISABLE_INTS
+
+#define ENABLE_INTS                            \
+       ld      r12,_MSR(r1);                   \
+       mfmsr   r11;                            \
+       rlwimi  r11,r12,0,MSR_EE;               \
+       mtmsrd  r11,1
+
+#endif
+
+#define STD_EXCEPTION_COMMON(trap, label, hdlr)                \
+       .align  7;                                      \
+       .globl label##_common;                          \
+label##_common:                                                \
+       EXCEPTION_PROLOG_COMMON(trap, PACA_EXGEN);      \
+       DISABLE_INTS;                                   \
+       bl      .save_nvgprs;                           \
+       addi    r3,r1,STACK_FRAME_OVERHEAD;             \
+       bl      hdlr;                                   \
+       b       .ret_from_except
+
+#define STD_EXCEPTION_COMMON_LITE(trap, label, hdlr)   \
+       .align  7;                                      \
+       .globl label##_common;                          \
+label##_common:                                                \
+       EXCEPTION_PROLOG_COMMON(trap, PACA_EXGEN);      \
+       DISABLE_INTS;                                   \
+       addi    r3,r1,STACK_FRAME_OVERHEAD;             \
+       bl      hdlr;                                   \
+       b       .ret_from_except_lite
+
+/*
+ * Start of pSeries system interrupt routines
+ */
+       . = 0x100
+       .globl __start_interrupts
+__start_interrupts:
+
+       STD_EXCEPTION_PSERIES(0x100, system_reset)
+
+       . = 0x200
+_machine_check_pSeries:
+       HMT_MEDIUM
+       mtspr   SPRG1,r13               /* save r13 */
+       RUNLATCH_ON(r13)
+       EXCEPTION_PROLOG_PSERIES(PACA_EXMC, machine_check_common)
+
+       . = 0x300
+       .globl data_access_pSeries
+data_access_pSeries:
+       HMT_MEDIUM
+       mtspr   SPRG1,r13
+BEGIN_FTR_SECTION
+       mtspr   SPRG2,r12
+       mfspr   r13,DAR
+       mfspr   r12,DSISR
+       srdi    r13,r13,60
+       rlwimi  r13,r12,16,0x20
+       mfcr    r12
+       cmpwi   r13,0x2c
+       beq     .do_stab_bolted_pSeries
+       mtcrf   0x80,r12
+       mfspr   r12,SPRG2
+END_FTR_SECTION_IFCLR(CPU_FTR_SLB)
+       EXCEPTION_PROLOG_PSERIES(PACA_EXGEN, data_access_common)
+
+       . = 0x380
+       .globl data_access_slb_pSeries
+data_access_slb_pSeries:
+       HMT_MEDIUM
+       mtspr   SPRG1,r13
+       RUNLATCH_ON(r13)
+       mfspr   r13,SPRG3               /* get paca address into r13 */
+       std     r9,PACA_EXSLB+EX_R9(r13)        /* save r9 - r12 */
+       std     r10,PACA_EXSLB+EX_R10(r13)
+       std     r11,PACA_EXSLB+EX_R11(r13)
+       std     r12,PACA_EXSLB+EX_R12(r13)
+       std     r3,PACA_EXSLB+EX_R3(r13)
+       mfspr   r9,SPRG1
+       std     r9,PACA_EXSLB+EX_R13(r13)
+       mfcr    r9
+       mfspr   r12,SRR1                /* and SRR1 */
+       mfspr   r3,DAR
+       b       .do_slb_miss            /* Rel. branch works in real mode */
+
+       STD_EXCEPTION_PSERIES(0x400, instruction_access)
+
+       . = 0x480
+       .globl instruction_access_slb_pSeries
+instruction_access_slb_pSeries:
+       HMT_MEDIUM
+       mtspr   SPRG1,r13
+       RUNLATCH_ON(r13)
+       mfspr   r13,SPRG3               /* get paca address into r13 */
+       std     r9,PACA_EXSLB+EX_R9(r13)        /* save r9 - r12 */
+       std     r10,PACA_EXSLB+EX_R10(r13)
+       std     r11,PACA_EXSLB+EX_R11(r13)
+       std     r12,PACA_EXSLB+EX_R12(r13)
+       std     r3,PACA_EXSLB+EX_R3(r13)
+       mfspr   r9,SPRG1
+       std     r9,PACA_EXSLB+EX_R13(r13)
+       mfcr    r9
+       mfspr   r12,SRR1                /* and SRR1 */
+       mfspr   r3,SRR0                 /* SRR0 is faulting address */
+       b       .do_slb_miss            /* Rel. branch works in real mode */
+
+       STD_EXCEPTION_PSERIES(0x500, hardware_interrupt)
+       STD_EXCEPTION_PSERIES(0x600, alignment)
+       STD_EXCEPTION_PSERIES(0x700, program_check)
+       STD_EXCEPTION_PSERIES(0x800, fp_unavailable)
+       STD_EXCEPTION_PSERIES(0x900, decrementer)
+       STD_EXCEPTION_PSERIES(0xa00, trap_0a)
+       STD_EXCEPTION_PSERIES(0xb00, trap_0b)
+
+       . = 0xc00
+       .globl  system_call_pSeries
+system_call_pSeries:
+       HMT_MEDIUM
+       RUNLATCH_ON(r9)
+       mr      r9,r13
+       mfmsr   r10
+       mfspr   r13,SPRG3
+       mfspr   r11,SRR0
+       clrrdi  r12,r13,32
+       oris    r12,r12,system_call_common@h
+       ori     r12,r12,system_call_common@l
+       mtspr   SRR0,r12
+       ori     r10,r10,MSR_IR|MSR_DR|MSR_RI
+       mfspr   r12,SRR1
+       mtspr   SRR1,r10
+       rfid
+       b       .       /* prevent speculative execution */
+
+       STD_EXCEPTION_PSERIES(0xd00, single_step)
+       STD_EXCEPTION_PSERIES(0xe00, trap_0e)
+
+       /* We need to deal with the Altivec unavailable exception
+        * here which is at 0xf20, thus in the middle of the
+        * prolog code of the PerformanceMonitor one. A little
+        * trickery is thus necessary
+        */
+       . = 0xf00
+       b       performance_monitor_pSeries
+
+       STD_EXCEPTION_PSERIES(0xf20, altivec_unavailable)
+
+       STD_EXCEPTION_PSERIES(0x1300, instruction_breakpoint)
+       STD_EXCEPTION_PSERIES(0x1700, altivec_assist)
+
+       . = 0x3000
+
+/*** pSeries interrupt support ***/
+
+       /* moved from 0xf00 */
+       STD_EXCEPTION_PSERIES(., performance_monitor)
+
+       .align  7
+_GLOBAL(do_stab_bolted_pSeries)
+       mtcrf   0x80,r12
+       mfspr   r12,SPRG2
+       EXCEPTION_PROLOG_PSERIES(PACA_EXSLB, .do_stab_bolted)
+
+/*
+ * Vectors for the FWNMI option.  Share common code.
+ */
+      .globl system_reset_fwnmi
+system_reset_fwnmi:
+      HMT_MEDIUM
+      mtspr   SPRG1,r13               /* save r13 */
+      RUNLATCH_ON(r13)
+      EXCEPTION_PROLOG_PSERIES(PACA_EXGEN, system_reset_common)
+
+      .globl machine_check_fwnmi
+machine_check_fwnmi:
+      HMT_MEDIUM
+      mtspr   SPRG1,r13               /* save r13 */
+      RUNLATCH_ON(r13)
+      EXCEPTION_PROLOG_PSERIES(PACA_EXMC, machine_check_common)
+
+#ifdef CONFIG_PPC_ISERIES
+/***  ISeries-LPAR interrupt handlers ***/
+
+       STD_EXCEPTION_ISERIES(0x200, machine_check, PACA_EXMC)
+
+       .globl data_access_iSeries
+data_access_iSeries:
+       mtspr   SPRG1,r13
+BEGIN_FTR_SECTION
+       mtspr   SPRG2,r12
+       mfspr   r13,DAR
+       mfspr   r12,DSISR
+       srdi    r13,r13,60
+       rlwimi  r13,r12,16,0x20
+       mfcr    r12
+       cmpwi   r13,0x2c
+       beq     .do_stab_bolted_iSeries
+       mtcrf   0x80,r12
+       mfspr   r12,SPRG2
+END_FTR_SECTION_IFCLR(CPU_FTR_SLB)
+       EXCEPTION_PROLOG_ISERIES_1(PACA_EXGEN)
+       EXCEPTION_PROLOG_ISERIES_2
+       b       data_access_common
+
+.do_stab_bolted_iSeries:
+       mtcrf   0x80,r12
+       mfspr   r12,SPRG2
+       EXCEPTION_PROLOG_ISERIES_1(PACA_EXSLB)
+       EXCEPTION_PROLOG_ISERIES_2
+       b       .do_stab_bolted
+
+       .globl  data_access_slb_iSeries
+data_access_slb_iSeries:
+       mtspr   SPRG1,r13               /* save r13 */
+       EXCEPTION_PROLOG_ISERIES_1(PACA_EXSLB)
+       std     r3,PACA_EXSLB+EX_R3(r13)
+       ld      r12,PACALPPACA+LPPACASRR1(r13)
+       mfspr   r3,DAR
+       b       .do_slb_miss
+
+       STD_EXCEPTION_ISERIES(0x400, instruction_access, PACA_EXGEN)
+
+       .globl  instruction_access_slb_iSeries
+instruction_access_slb_iSeries:
+       mtspr   SPRG1,r13               /* save r13 */
+       EXCEPTION_PROLOG_ISERIES_1(PACA_EXSLB)
+       std     r3,PACA_EXSLB+EX_R3(r13)
+       ld      r12,PACALPPACA+LPPACASRR1(r13)
+       ld      r3,PACALPPACA+LPPACASRR0(r13)
+       b       .do_slb_miss
+
+       MASKABLE_EXCEPTION_ISERIES(0x500, hardware_interrupt)
+       STD_EXCEPTION_ISERIES(0x600, alignment, PACA_EXGEN)
+       STD_EXCEPTION_ISERIES(0x700, program_check, PACA_EXGEN)
+       STD_EXCEPTION_ISERIES(0x800, fp_unavailable, PACA_EXGEN)
+       MASKABLE_EXCEPTION_ISERIES(0x900, decrementer)
+       STD_EXCEPTION_ISERIES(0xa00, trap_0a, PACA_EXGEN)
+       STD_EXCEPTION_ISERIES(0xb00, trap_0b, PACA_EXGEN)
+
+       .globl  system_call_iSeries
+system_call_iSeries:
+       mr      r9,r13
+       mfspr   r13,SPRG3
+       EXCEPTION_PROLOG_ISERIES_2
+       b       system_call_common
+
+       STD_EXCEPTION_ISERIES( 0xd00, single_step, PACA_EXGEN)
+       STD_EXCEPTION_ISERIES( 0xe00, trap_0e, PACA_EXGEN)
+       STD_EXCEPTION_ISERIES( 0xf00, performance_monitor, PACA_EXGEN)
+
+       .globl system_reset_iSeries
+system_reset_iSeries:
+       mfspr   r13,SPRG3               /* Get paca address */
+       mfmsr   r24
+       ori     r24,r24,MSR_RI
+       mtmsrd  r24                     /* RI on */
+       lhz     r24,PACAPACAINDEX(r13)  /* Get processor # */
+       cmpwi   0,r24,0                 /* Are we processor 0? */
+       beq     .__start_initialization_iSeries /* Start up the first processor */
+       mfspr   r4,SPRN_CTRLF
+       li      r5,CTRL_RUNLATCH        /* Turn off the run light */
+       andc    r4,r4,r5
+       mtspr   SPRN_CTRLT,r4
+
+1:
+       HMT_LOW
+#ifdef CONFIG_SMP
+       lbz     r23,PACAPROCSTART(r13)  /* Test if this processor
+                                        * should start */
+       sync
+       LOADADDR(r3,current_set)
+       sldi    r28,r24,3               /* get current_set[cpu#] */
+       ldx     r3,r3,r28
+       addi    r1,r3,THREAD_SIZE
+       subi    r1,r1,STACK_FRAME_OVERHEAD
+
+       cmpwi   0,r23,0
+       beq     iSeries_secondary_smp_loop      /* Loop until told to go */
+       bne     .__secondary_start              /* Loop until told to go */
+iSeries_secondary_smp_loop:
+       /* Let the Hypervisor know we are alive */
+       /* 8002 is a call to HvCallCfg::getLps, a harmless Hypervisor function */
+       lis     r3,0x8002
+       rldicr  r3,r3,32,15             /* r0 = (r3 << 32) & 0xffff000000000000 */
+#else /* CONFIG_SMP */
+       /* Yield the processor.  This is required for non-SMP kernels
+               which are running on multi-threaded machines. */
+       lis     r3,0x8000
+       rldicr  r3,r3,32,15             /* r3 = (r3 << 32) & 0xffff000000000000 */
+       addi    r3,r3,18                /* r3 = 0x8000000000000012 which is "yield" */
+       li      r4,0                    /* "yield timed" */
+       li      r5,-1                   /* "yield forever" */
+#endif /* CONFIG_SMP */
+       li      r0,-1                   /* r0=-1 indicates a Hypervisor call */
+       sc                              /* Invoke the hypervisor via a system call */
+       mfspr   r13,SPRG3               /* Put r13 back ???? */
+       b       1b                      /* If SMP not configured, secondaries
+                                        * loop forever */
+
+       .globl decrementer_iSeries_masked
+decrementer_iSeries_masked:
+       li      r11,1
+       stb     r11,PACALPPACA+LPPACADECRINT(r13)
+       lwz     r12,PACADEFAULTDECR(r13)
+       mtspr   SPRN_DEC,r12
+       /* fall through */
+
+       .globl hardware_interrupt_iSeries_masked
+hardware_interrupt_iSeries_masked:
+       mtcrf   0x80,r9         /* Restore regs */
+       ld      r11,PACALPPACA+LPPACASRR0(r13)
+       ld      r12,PACALPPACA+LPPACASRR1(r13)
+       mtspr   SRR0,r11
+       mtspr   SRR1,r12
+       ld      r9,PACA_EXGEN+EX_R9(r13)
+       ld      r10,PACA_EXGEN+EX_R10(r13)
+       ld      r11,PACA_EXGEN+EX_R11(r13)
+       ld      r12,PACA_EXGEN+EX_R12(r13)
+       ld      r13,PACA_EXGEN+EX_R13(r13)
+       rfid
+       b       .       /* prevent speculative execution */
+#endif /* CONFIG_PPC_ISERIES */
+
+/*** Common interrupt handlers ***/
+
+       STD_EXCEPTION_COMMON(0x100, system_reset, .system_reset_exception)
+
+       /*
+        * Machine check is different because we use a different
+        * save area: PACA_EXMC instead of PACA_EXGEN.
+        */
+       .align  7
+       .globl machine_check_common
+machine_check_common:
+       EXCEPTION_PROLOG_COMMON(0x200, PACA_EXMC)
+       DISABLE_INTS
+       bl      .save_nvgprs
+       addi    r3,r1,STACK_FRAME_OVERHEAD
+       bl      .machine_check_exception
+       b       .ret_from_except
+
+       STD_EXCEPTION_COMMON_LITE(0x900, decrementer, .timer_interrupt)
+       STD_EXCEPTION_COMMON(0xa00, trap_0a, .unknown_exception)
+       STD_EXCEPTION_COMMON(0xb00, trap_0b, .unknown_exception)
+       STD_EXCEPTION_COMMON(0xd00, single_step, .single_step_exception)
+       STD_EXCEPTION_COMMON(0xe00, trap_0e, .unknown_exception)
+       STD_EXCEPTION_COMMON(0xf00, performance_monitor, .performance_monitor_exception)
+       STD_EXCEPTION_COMMON(0x1300, instruction_breakpoint, .instruction_breakpoint_exception)
+#ifdef CONFIG_ALTIVEC
+       STD_EXCEPTION_COMMON(0x1700, altivec_assist, .altivec_assist_exception)
+#else
+       STD_EXCEPTION_COMMON(0x1700, altivec_assist, .unknown_exception)
+#endif
+
+/*
+ * Here we have detected that the kernel stack pointer is bad.
+ * R9 contains the saved CR, r13 points to the paca,
+ * r10 contains the (bad) kernel stack pointer,
+ * r11 and r12 contain the saved SRR0 and SRR1.
+ * We switch to using an emergency stack, save the registers there,
+ * and call kernel_bad_stack(), which panics.
+ */
+bad_stack:
+       ld      r1,PACAEMERGSP(r13)
+       subi    r1,r1,64+INT_FRAME_SIZE
+       std     r9,_CCR(r1)
+       std     r10,GPR1(r1)
+       std     r11,_NIP(r1)
+       std     r12,_MSR(r1)
+       mfspr   r11,DAR
+       mfspr   r12,DSISR
+       std     r11,_DAR(r1)
+       std     r12,_DSISR(r1)
+       mflr    r10
+       mfctr   r11
+       mfxer   r12
+       std     r10,_LINK(r1)
+       std     r11,_CTR(r1)
+       std     r12,_XER(r1)
+       SAVE_GPR(0,r1)
+       SAVE_GPR(2,r1)
+       SAVE_4GPRS(3,r1)
+       SAVE_2GPRS(7,r1)
+       SAVE_10GPRS(12,r1)
+       SAVE_10GPRS(22,r1)
+       addi    r11,r1,INT_FRAME_SIZE
+       std     r11,0(r1)
+       li      r12,0
+       std     r12,0(r11)
+       ld      r2,PACATOC(r13)
+1:     addi    r3,r1,STACK_FRAME_OVERHEAD
+       bl      .kernel_bad_stack
+       b       1b
+
+/*
+ * Return from an exception with minimal checks.
+ * The caller is assumed to have done EXCEPTION_PROLOG_COMMON.
+ * If interrupts have been enabled, or anything has been
+ * done that might have changed the scheduling status of
+ * any task or sent any task a signal, you should use
+ * ret_from_except or ret_from_except_lite instead of this.
+ */
+fast_exception_return:
+       ld      r12,_MSR(r1)
+       ld      r11,_NIP(r1)
+       andi.   r3,r12,MSR_RI           /* check if RI is set */
+       beq-    unrecov_fer
+       ld      r3,_CCR(r1)
+       ld      r4,_LINK(r1)
+       ld      r5,_CTR(r1)
+       ld      r6,_XER(r1)
+       mtcr    r3
+       mtlr    r4
+       mtctr   r5
+       mtxer   r6
+       REST_GPR(0, r1)
+       REST_8GPRS(2, r1)
+
+       mfmsr   r10
+       clrrdi  r10,r10,2               /* clear RI (LE is 0 already) */
+       mtmsrd  r10,1
+
+       mtspr   SRR1,r12
+       mtspr   SRR0,r11
+       REST_4GPRS(10, r1)
+       ld      r1,GPR1(r1)
+       rfid
+       b       .       /* prevent speculative execution */
+
+unrecov_fer:
+       bl      .save_nvgprs
+1:     addi    r3,r1,STACK_FRAME_OVERHEAD
+       bl      .unrecoverable_exception
+       b       1b
+
+/*
+ * Here r13 points to the paca, r9 contains the saved CR,
+ * SRR0 and SRR1 are saved in r11 and r12,
+ * r9 - r13 are saved in paca->exgen.
+ */
+       .align  7
+       .globl data_access_common
+data_access_common:
+       RUNLATCH_ON(r10)                /* It wont fit in the 0x300 handler */
+       mfspr   r10,DAR
+       std     r10,PACA_EXGEN+EX_DAR(r13)
+       mfspr   r10,DSISR
+       stw     r10,PACA_EXGEN+EX_DSISR(r13)
+       EXCEPTION_PROLOG_COMMON(0x300, PACA_EXGEN)
+       ld      r3,PACA_EXGEN+EX_DAR(r13)
+       lwz     r4,PACA_EXGEN+EX_DSISR(r13)
+       li      r5,0x300
+       b       .do_hash_page           /* Try to handle as hpte fault */
+
+       .align  7
+       .globl instruction_access_common
+instruction_access_common:
+       EXCEPTION_PROLOG_COMMON(0x400, PACA_EXGEN)
+       ld      r3,_NIP(r1)
+       andis.  r4,r12,0x5820
+       li      r5,0x400
+       b       .do_hash_page           /* Try to handle as hpte fault */
+
+       .align  7
+       .globl hardware_interrupt_common
+       .globl hardware_interrupt_entry
+hardware_interrupt_common:
+       EXCEPTION_PROLOG_COMMON(0x500, PACA_EXGEN)
+hardware_interrupt_entry:
+       DISABLE_INTS
+       addi    r3,r1,STACK_FRAME_OVERHEAD
+       bl      .do_IRQ
+       b       .ret_from_except_lite
+
+       .align  7
+       .globl alignment_common
+alignment_common:
+       mfspr   r10,DAR
+       std     r10,PACA_EXGEN+EX_DAR(r13)
+       mfspr   r10,DSISR
+       stw     r10,PACA_EXGEN+EX_DSISR(r13)
+       EXCEPTION_PROLOG_COMMON(0x600, PACA_EXGEN)
+       ld      r3,PACA_EXGEN+EX_DAR(r13)
+       lwz     r4,PACA_EXGEN+EX_DSISR(r13)
+       std     r3,_DAR(r1)
+       std     r4,_DSISR(r1)
+       bl      .save_nvgprs
+       addi    r3,r1,STACK_FRAME_OVERHEAD
+       ENABLE_INTS
+       bl      .alignment_exception
+       b       .ret_from_except
+
+       .align  7
+       .globl program_check_common
+program_check_common:
+       EXCEPTION_PROLOG_COMMON(0x700, PACA_EXGEN)
+       bl      .save_nvgprs
+       addi    r3,r1,STACK_FRAME_OVERHEAD
+       ENABLE_INTS
+       bl      .program_check_exception
+       b       .ret_from_except
+
+       .align  7
+       .globl fp_unavailable_common
+fp_unavailable_common:
+       EXCEPTION_PROLOG_COMMON(0x800, PACA_EXGEN)
+       bne     .load_up_fpu            /* if from user, just load it up */
+       bl      .save_nvgprs
+       addi    r3,r1,STACK_FRAME_OVERHEAD
+       ENABLE_INTS
+       bl      .kernel_fp_unavailable_exception
+       BUG_OPCODE
+
+/*
+ * load_up_fpu(unused, unused, tsk)
+ * Disable FP for the task which had the FPU previously,
+ * and save its floating-point registers in its thread_struct.
+ * Enables the FPU for use in the kernel on return.
+ * On SMP we know the fpu is free, since we give it up every
+ * switch (ie, no lazy save of the FP registers).
+ * On entry: r13 == 'current' && last_task_used_math != 'current'
+ */
+_STATIC(load_up_fpu)
+       mfmsr   r5                      /* grab the current MSR */
+       ori     r5,r5,MSR_FP
+       mtmsrd  r5                      /* enable use of fpu now */
+       isync
+/*
+ * For SMP, we don't do lazy FPU switching because it just gets too
+ * horrendously complex, especially when a task switches from one CPU
+ * to another.  Instead we call giveup_fpu in switch_to.
+ *
+ */
+#ifndef CONFIG_SMP
+       ld      r3,last_task_used_math@got(r2)
+       ld      r4,0(r3)
+       cmpdi   0,r4,0
+       beq     1f
+       /* Save FP state to last_task_used_math's THREAD struct */
+       addi    r4,r4,THREAD
+       SAVE_32FPRS(0, r4)
+       mffs    fr0
+       stfd    fr0,THREAD_FPSCR(r4)
+       /* Disable FP for last_task_used_math */
+       ld      r5,PT_REGS(r4)
+       ld      r4,_MSR-STACK_FRAME_OVERHEAD(r5)
+       li      r6,MSR_FP|MSR_FE0|MSR_FE1
+       andc    r4,r4,r6
+       std     r4,_MSR-STACK_FRAME_OVERHEAD(r5)
+1:
+#endif /* CONFIG_SMP */
+       /* enable use of FP after return */
+       ld      r4,PACACURRENT(r13)
+       addi    r5,r4,THREAD            /* Get THREAD */
+       ld      r4,THREAD_FPEXC_MODE(r5)
+       ori     r12,r12,MSR_FP
+       or      r12,r12,r4
+       std     r12,_MSR(r1)
+       lfd     fr0,THREAD_FPSCR(r5)
+       mtfsf   0xff,fr0
+       REST_32FPRS(0, r5)
+#ifndef CONFIG_SMP
+       /* Update last_task_used_math to 'current' */
+       subi    r4,r5,THREAD            /* Back to 'current' */
+       std     r4,0(r3)
+#endif /* CONFIG_SMP */
+       /* restore registers and return */
+       b       fast_exception_return
+
+       .align  7
+       .globl altivec_unavailable_common
+altivec_unavailable_common:
+       EXCEPTION_PROLOG_COMMON(0xf20, PACA_EXGEN)
+#ifdef CONFIG_ALTIVEC
+BEGIN_FTR_SECTION
+       bne     .load_up_altivec        /* if from user, just load it up */
+END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
+#endif
+       bl      .save_nvgprs
+       addi    r3,r1,STACK_FRAME_OVERHEAD
+       ENABLE_INTS
+       bl      .altivec_unavailable_exception
+       b       .ret_from_except
+
+#ifdef CONFIG_ALTIVEC
+/*
+ * load_up_altivec(unused, unused, tsk)
+ * Disable VMX for the task which had it previously,
+ * and save its vector registers in its thread_struct.
+ * Enables the VMX for use in the kernel on return.
+ * On SMP we know the VMX is free, since we give it up every
+ * switch (ie, no lazy save of the vector registers).
+ * On entry: r13 == 'current' && last_task_used_altivec != 'current'
+ */
+_STATIC(load_up_altivec)
+       mfmsr   r5                      /* grab the current MSR */
+       oris    r5,r5,MSR_VEC@h
+       mtmsrd  r5                      /* enable use of VMX now */
+       isync
+
+/*
+ * For SMP, we don't do lazy VMX switching because it just gets too
+ * horrendously complex, especially when a task switches from one CPU
+ * to another.  Instead we call giveup_altvec in switch_to.
+ * VRSAVE isn't dealt with here, that is done in the normal context
+ * switch code. Note that we could rely on vrsave value to eventually
+ * avoid saving all of the VREGs here...
+ */
+#ifndef CONFIG_SMP
+       ld      r3,last_task_used_altivec@got(r2)
+       ld      r4,0(r3)
+       cmpdi   0,r4,0
+       beq     1f
+       /* Save VMX state to last_task_used_altivec's THREAD struct */
+       addi    r4,r4,THREAD
+       SAVE_32VRS(0,r5,r4)
+       mfvscr  vr0
+       li      r10,THREAD_VSCR
+       stvx    vr0,r10,r4
+       /* Disable VMX for last_task_used_altivec */
+       ld      r5,PT_REGS(r4)
+       ld      r4,_MSR-STACK_FRAME_OVERHEAD(r5)
+       lis     r6,MSR_VEC@h
+       andc    r4,r4,r6
+       std     r4,_MSR-STACK_FRAME_OVERHEAD(r5)
+1:
+#endif /* CONFIG_SMP */
+       /* Hack: if we get an altivec unavailable trap with VRSAVE
+        * set to all zeros, we assume this is a broken application
+        * that fails to set it properly, and thus we switch it to
+        * all 1's
+        */
+       mfspr   r4,SPRN_VRSAVE
+       cmpdi   0,r4,0
+       bne+    1f
+       li      r4,-1
+       mtspr   SPRN_VRSAVE,r4
+1:
+       /* enable use of VMX after return */
+       ld      r4,PACACURRENT(r13)
+       addi    r5,r4,THREAD            /* Get THREAD */
+       oris    r12,r12,MSR_VEC@h
+       std     r12,_MSR(r1)
+       li      r4,1
+       li      r10,THREAD_VSCR
+       stw     r4,THREAD_USED_VR(r5)
+       lvx     vr0,r10,r5
+       mtvscr  vr0
+       REST_32VRS(0,r4,r5)
+#ifndef CONFIG_SMP
+       /* Update last_task_used_math to 'current' */
+       subi    r4,r5,THREAD            /* Back to 'current' */
+       std     r4,0(r3)
+#endif /* CONFIG_SMP */
+       /* restore registers and return */
+       b       fast_exception_return
+#endif /* CONFIG_ALTIVEC */
+
+/*
+ * Hash table stuff
+ */
+       .align  7
+_GLOBAL(do_hash_page)
+       std     r3,_DAR(r1)
+       std     r4,_DSISR(r1)
+
+       andis.  r0,r4,0xa450            /* weird error? */
+       bne-    .handle_page_fault      /* if not, try to insert a HPTE */
+BEGIN_FTR_SECTION
+       andis.  r0,r4,0x0020            /* Is it a segment table fault? */
+       bne-    .do_ste_alloc           /* If so handle it */
+END_FTR_SECTION_IFCLR(CPU_FTR_SLB)
+
+       /*
+        * We need to set the _PAGE_USER bit if MSR_PR is set or if we are
+        * accessing a userspace segment (even from the kernel). We assume
+        * kernel addresses always have the high bit set.
+        */
+       rlwinm  r4,r4,32-25+9,31-9,31-9 /* DSISR_STORE -> _PAGE_RW */
+       rotldi  r0,r3,15                /* Move high bit into MSR_PR posn */
+       orc     r0,r12,r0               /* MSR_PR | ~high_bit */
+       rlwimi  r4,r0,32-13,30,30       /* becomes _PAGE_USER access bit */
+       ori     r4,r4,1                 /* add _PAGE_PRESENT */
+       rlwimi  r4,r5,22+2,31-2,31-2    /* Set _PAGE_EXEC if trap is 0x400 */
+
+       /*
+        * On iSeries, we soft-disable interrupts here, then
+        * hard-enable interrupts so that the hash_page code can spin on
+        * the hash_table_lock without problems on a shared processor.
+        */
+       DISABLE_INTS
+
+       /*
+        * r3 contains the faulting address
+        * r4 contains the required access permissions
+        * r5 contains the trap number
+        *
+        * at return r3 = 0 for success
+        */
+       bl      .hash_page              /* build HPTE if possible */
+       cmpdi   r3,0                    /* see if hash_page succeeded */
+
+#ifdef DO_SOFT_DISABLE
+       /*
+        * If we had interrupts soft-enabled at the point where the
+        * DSI/ISI occurred, and an interrupt came in during hash_page,
+        * handle it now.
+        * We jump to ret_from_except_lite rather than fast_exception_return
+        * because ret_from_except_lite will check for and handle pending
+        * interrupts if necessary.
+        */
+       beq     .ret_from_except_lite
+       /* For a hash failure, we don't bother re-enabling interrupts */
+       ble-    12f
+
+       /*
+        * hash_page couldn't handle it, set soft interrupt enable back
+        * to what it was before the trap.  Note that .local_irq_restore
+        * handles any interrupts pending at this point.
+        */
+       ld      r3,SOFTE(r1)
+       bl      .local_irq_restore
+       b       11f
+#else
+       beq     fast_exception_return   /* Return from exception on success */
+       ble-    12f                     /* Failure return from hash_page */
+
+       /* fall through */
+#endif
+
+/* Here we have a page fault that hash_page can't handle. */
+_GLOBAL(handle_page_fault)
+       ENABLE_INTS
+11:    ld      r4,_DAR(r1)
+       ld      r5,_DSISR(r1)
+       addi    r3,r1,STACK_FRAME_OVERHEAD
+       bl      .do_page_fault
+       cmpdi   r3,0
+       beq+    .ret_from_except_lite
+       bl      .save_nvgprs
+       mr      r5,r3
+       addi    r3,r1,STACK_FRAME_OVERHEAD
+       lwz     r4,_DAR(r1)
+       bl      .bad_page_fault
+       b       .ret_from_except
+
+/* We have a page fault that hash_page could handle but HV refused
+ * the PTE insertion
+ */
+12:    bl      .save_nvgprs
+       addi    r3,r1,STACK_FRAME_OVERHEAD
+       lwz     r4,_DAR(r1)
+       bl      .low_hash_fault
+       b       .ret_from_except
+
+       /* here we have a segment miss */
+_GLOBAL(do_ste_alloc)
+       bl      .ste_allocate           /* try to insert stab entry */
+       cmpdi   r3,0
+       beq+    fast_exception_return
+       b       .handle_page_fault
+
+/*
+ * r13 points to the PACA, r9 contains the saved CR,
+ * r11 and r12 contain the saved SRR0 and SRR1.
+ * r9 - r13 are saved in paca->exslb.
+ * We assume we aren't going to take any exceptions during this procedure.
+ * We assume (DAR >> 60) == 0xc.
+ */
+       .align  7
+_GLOBAL(do_stab_bolted)
+       stw     r9,PACA_EXSLB+EX_CCR(r13)       /* save CR in exc. frame */
+       std     r11,PACA_EXSLB+EX_SRR0(r13)     /* save SRR0 in exc. frame */
+
+       /* Hash to the primary group */
+       ld      r10,PACASTABVIRT(r13)
+       mfspr   r11,DAR
+       srdi    r11,r11,28
+       rldimi  r10,r11,7,52    /* r10 = first ste of the group */
+
+       /* Calculate VSID */
+       /* This is a kernel address, so protovsid = ESID */
+       ASM_VSID_SCRAMBLE(r11, r9)
+       rldic   r9,r11,12,16    /* r9 = vsid << 12 */
+
+       /* Search the primary group for a free entry */
+1:     ld      r11,0(r10)      /* Test valid bit of the current ste    */
+       andi.   r11,r11,0x80
+       beq     2f
+       addi    r10,r10,16
+       andi.   r11,r10,0x70
+       bne     1b
+
+       /* Stick for only searching the primary group for now.          */
+       /* At least for now, we use a very simple random castout scheme */
+       /* Use the TB as a random number ;  OR in 1 to avoid entry 0    */
+       mftb    r11
+       rldic   r11,r11,4,57    /* r11 = (r11 << 4) & 0x70 */
+       ori     r11,r11,0x10
+
+       /* r10 currently points to an ste one past the group of interest */
+       /* make it point to the randomly selected entry                 */
+       subi    r10,r10,128
+       or      r10,r10,r11     /* r10 is the entry to invalidate       */
+
+       isync                   /* mark the entry invalid               */
+       ld      r11,0(r10)
+       rldicl  r11,r11,56,1    /* clear the valid bit */
+       rotldi  r11,r11,8
+       std     r11,0(r10)
+       sync
+
+       clrrdi  r11,r11,28      /* Get the esid part of the ste         */
+       slbie   r11
+
+2:     std     r9,8(r10)       /* Store the vsid part of the ste       */
+       eieio
+
+       mfspr   r11,DAR         /* Get the new esid                     */
+       clrrdi  r11,r11,28      /* Permits a full 32b of ESID           */
+       ori     r11,r11,0x90    /* Turn on valid and kp                 */
+       std     r11,0(r10)      /* Put new entry back into the stab     */
+
+       sync
+
+       /* All done -- return from exception. */
+       lwz     r9,PACA_EXSLB+EX_CCR(r13)       /* get saved CR */
+       ld      r11,PACA_EXSLB+EX_SRR0(r13)     /* get saved SRR0 */
+
+       andi.   r10,r12,MSR_RI
+       beq-    unrecov_slb
+
+       mtcrf   0x80,r9                 /* restore CR */
+
+       mfmsr   r10
+       clrrdi  r10,r10,2
+       mtmsrd  r10,1
+
+       mtspr   SRR0,r11
+       mtspr   SRR1,r12
+       ld      r9,PACA_EXSLB+EX_R9(r13)
+       ld      r10,PACA_EXSLB+EX_R10(r13)
+       ld      r11,PACA_EXSLB+EX_R11(r13)
+       ld      r12,PACA_EXSLB+EX_R12(r13)
+       ld      r13,PACA_EXSLB+EX_R13(r13)
+       rfid
+       b       .       /* prevent speculative execution */
+
+/*
+ * r13 points to the PACA, r9 contains the saved CR,
+ * r11 and r12 contain the saved SRR0 and SRR1.
+ * r3 has the faulting address
+ * r9 - r13 are saved in paca->exslb.
+ * r3 is saved in paca->slb_r3
+ * We assume we aren't going to take any exceptions during this procedure.
+ */
+_GLOBAL(do_slb_miss)
+       mflr    r10
+
+       stw     r9,PACA_EXSLB+EX_CCR(r13)       /* save CR in exc. frame */
+       std     r10,PACA_EXSLB+EX_LR(r13)       /* save LR */
+
+       bl      .slb_allocate                   /* handle it */
+
+       /* All done -- return from exception. */
+
+       ld      r10,PACA_EXSLB+EX_LR(r13)
+       ld      r3,PACA_EXSLB+EX_R3(r13)
+       lwz     r9,PACA_EXSLB+EX_CCR(r13)       /* get saved CR */
+#ifdef CONFIG_PPC_ISERIES
+       ld      r11,PACALPPACA+LPPACASRR0(r13)  /* get SRR0 value */
+#endif /* CONFIG_PPC_ISERIES */
+
+       mtlr    r10
+
+       andi.   r10,r12,MSR_RI  /* check for unrecoverable exception */
+       beq-    unrecov_slb
+
+.machine       push
+.machine       "power4"
+       mtcrf   0x80,r9
+       mtcrf   0x01,r9         /* slb_allocate uses cr0 and cr7 */
+.machine       pop
+
+#ifdef CONFIG_PPC_ISERIES
+       mtspr   SRR0,r11
+       mtspr   SRR1,r12
+#endif /* CONFIG_PPC_ISERIES */
+       ld      r9,PACA_EXSLB+EX_R9(r13)
+       ld      r10,PACA_EXSLB+EX_R10(r13)
+       ld      r11,PACA_EXSLB+EX_R11(r13)
+       ld      r12,PACA_EXSLB+EX_R12(r13)
+       ld      r13,PACA_EXSLB+EX_R13(r13)
+       rfid
+       b       .       /* prevent speculative execution */
+
+unrecov_slb:
+       EXCEPTION_PROLOG_COMMON(0x4100, PACA_EXSLB)
+       DISABLE_INTS
+       bl      .save_nvgprs
+1:     addi    r3,r1,STACK_FRAME_OVERHEAD
+       bl      .unrecoverable_exception
+       b       1b
+
+/*
+ * Space for CPU0's segment table.
+ *
+ * On iSeries, the hypervisor must fill in at least one entry before
+ * we get control (with relocate on).  The address is give to the hv
+ * as a page number (see xLparMap in LparData.c), so this must be at a
+ * fixed address (the linker can't compute (u64)&initial_stab >>
+ * PAGE_SHIFT).
+ */
+       . = STAB0_PHYS_ADDR     /* 0x6000 */
+       .globl initial_stab
+initial_stab:
+       .space  4096
+
+/*
+ * Data area reserved for FWNMI option.
+ * This address (0x7000) is fixed by the RPA.
+ */
+       .= 0x7000
+       .globl fwnmi_data_area
+fwnmi_data_area:
+
+       /* iSeries does not use the FWNMI stuff, so it is safe to put
+        * this here, even if we later allow kernels that will boot on
+        * both pSeries and iSeries */
+#ifdef CONFIG_PPC_ISERIES
+        . = LPARMAP_PHYS
+#include "lparmap.s"
+/*
+ * This ".text" is here for old compilers that generate a trailing
+ * .note section when compiling .c files to .s
+ */
+       .text
+#endif /* CONFIG_PPC_ISERIES */
+
+        . = 0x8000
+
+/*
+ * On pSeries, secondary processors spin in the following code.
+ * At entry, r3 = this processor's number (physical cpu id)
+ */
+_GLOBAL(pSeries_secondary_smp_init)
+       mr      r24,r3
+       
+       /* turn on 64-bit mode */
+       bl      .enable_64b_mode
+       isync
+
+       /* Copy some CPU settings from CPU 0 */
+       bl      .__restore_cpu_setup
+
+       /* Set up a paca value for this processor. Since we have the
+        * physical cpu id in r24, we need to search the pacas to find
+        * which logical id maps to our physical one.
+        */
+       LOADADDR(r13, paca)             /* Get base vaddr of paca array  */
+       li      r5,0                    /* logical cpu id                */
+1:     lhz     r6,PACAHWCPUID(r13)     /* Load HW procid from paca      */
+       cmpw    r6,r24                  /* Compare to our id             */
+       beq     2f
+       addi    r13,r13,PACA_SIZE       /* Loop to next PACA on miss     */
+       addi    r5,r5,1
+       cmpwi   r5,NR_CPUS
+       blt     1b
+
+       mr      r3,r24                  /* not found, copy phys to r3    */
+       b       .kexec_wait             /* next kernel might do better   */
+
+2:     mtspr   SPRG3,r13               /* Save vaddr of paca in SPRG3   */
+       /* From now on, r24 is expected to be logical cpuid */
+       mr      r24,r5
+3:     HMT_LOW
+       lbz     r23,PACAPROCSTART(r13)  /* Test if this processor should */
+                                       /* start.                        */
+       sync
+
+       /* Create a temp kernel stack for use before relocation is on.  */
+       ld      r1,PACAEMERGSP(r13)
+       subi    r1,r1,STACK_FRAME_OVERHEAD
+
+       cmpwi   0,r23,0
+#ifdef CONFIG_SMP
+       bne     .__secondary_start
+#endif
+       b       3b                      /* Loop until told to go         */
+
+#ifdef CONFIG_PPC_ISERIES
+_STATIC(__start_initialization_iSeries)
+       /* Clear out the BSS */
+       LOADADDR(r11,__bss_stop)
+       LOADADDR(r8,__bss_start)
+       sub     r11,r11,r8              /* bss size                     */
+       addi    r11,r11,7               /* round up to an even double word */
+       rldicl. r11,r11,61,3            /* shift right by 3             */
+       beq     4f
+       addi    r8,r8,-8
+       li      r0,0
+       mtctr   r11                     /* zero this many doublewords   */
+3:     stdu    r0,8(r8)
+       bdnz    3b
+4:
+       LOADADDR(r1,init_thread_union)
+       addi    r1,r1,THREAD_SIZE
+       li      r0,0
+       stdu    r0,-STACK_FRAME_OVERHEAD(r1)
+
+       LOADADDR(r3,cpu_specs)
+       LOADADDR(r4,cur_cpu_spec)
+       li      r5,0
+       bl      .identify_cpu
+
+       LOADADDR(r2,__toc_start)
+       addi    r2,r2,0x4000
+       addi    r2,r2,0x4000
+
+       bl      .iSeries_early_setup
+
+       /* relocation is on at this point */
+
+       b       .start_here_common
+#endif /* CONFIG_PPC_ISERIES */
+
+#ifdef CONFIG_PPC_MULTIPLATFORM
+
+_STATIC(__mmu_off)
+       mfmsr   r3
+       andi.   r0,r3,MSR_IR|MSR_DR
+       beqlr
+       andc    r3,r3,r0
+       mtspr   SPRN_SRR0,r4
+       mtspr   SPRN_SRR1,r3
+       sync
+       rfid
+       b       .       /* prevent speculative execution */
+
+
+/*
+ * Here is our main kernel entry point. We support currently 2 kind of entries
+ * depending on the value of r5.
+ *
+ *   r5 != NULL -> OF entry, we go to prom_init, "legacy" parameter content
+ *                 in r3...r7
+ *   
+ *   r5 == NULL -> kexec style entry. r3 is a physical pointer to the
+ *                 DT block, r4 is a physical pointer to the kernel itself
+ *
+ */
+_GLOBAL(__start_initialization_multiplatform)
+       /*
+        * Are we booted from a PROM Of-type client-interface ?
+        */
+       cmpldi  cr0,r5,0
+       bne     .__boot_from_prom               /* yes -> prom */
+
+       /* Save parameters */
+       mr      r31,r3
+       mr      r30,r4
+
+       /* Make sure we are running in 64 bits mode */
+       bl      .enable_64b_mode
+
+       /* Setup some critical 970 SPRs before switching MMU off */
+       bl      .__970_cpu_preinit
+
+       /* cpu # */
+       li      r24,0
+
+       /* Switch off MMU if not already */
+       LOADADDR(r4, .__after_prom_start - KERNELBASE)
+       add     r4,r4,r30
+       bl      .__mmu_off
+       b       .__after_prom_start
+
+_STATIC(__boot_from_prom)
+       /* Save parameters */
+       mr      r31,r3
+       mr      r30,r4
+       mr      r29,r5
+       mr      r28,r6
+       mr      r27,r7
+
+       /* Make sure we are running in 64 bits mode */
+       bl      .enable_64b_mode
+
+       /* put a relocation offset into r3 */
+       bl      .reloc_offset
+
+       LOADADDR(r2,__toc_start)
+       addi    r2,r2,0x4000
+       addi    r2,r2,0x4000
+
+       /* Relocate the TOC from a virt addr to a real addr */
+       sub     r2,r2,r3
+
+       /* Restore parameters */
+       mr      r3,r31
+       mr      r4,r30
+       mr      r5,r29
+       mr      r6,r28
+       mr      r7,r27
+
+       /* Do all of the interaction with OF client interface */
+       bl      .prom_init
+       /* We never return */
+       trap
+
+/*
+ * At this point, r3 contains the physical address we are running at,
+ * returned by prom_init()
+ */
+_STATIC(__after_prom_start)
+
+/*
+ * We need to run with __start at physical address 0.
+ * This will leave some code in the first 256B of
+ * real memory, which are reserved for software use.
+ * The remainder of the first page is loaded with the fixed
+ * interrupt vectors.  The next two pages are filled with
+ * unknown exception placeholders.
+ *
+ * Note: This process overwrites the OF exception vectors.
+ *     r26 == relocation offset
+ *     r27 == KERNELBASE
+ */
+       bl      .reloc_offset
+       mr      r26,r3
+       SET_REG_TO_CONST(r27,KERNELBASE)
+
+       li      r3,0                    /* target addr */
+
+       // XXX FIXME: Use phys returned by OF (r30)
+       sub     r4,r27,r26              /* source addr                   */
+                                       /* current address of _start     */
+                                       /*   i.e. where we are running   */
+                                       /*      the source addr          */
+
+       LOADADDR(r5,copy_to_here)       /* # bytes of memory to copy     */
+       sub     r5,r5,r27
+
+       li      r6,0x100                /* Start offset, the first 0x100 */
+                                       /* bytes were copied earlier.    */
+
+       bl      .copy_and_flush         /* copy the first n bytes        */
+                                       /* this includes the code being  */
+                                       /* executed here.                */
+
+       LOADADDR(r0, 4f)                /* Jump to the copy of this code */
+       mtctr   r0                      /* that we just made/relocated   */
+       bctr
+
+4:     LOADADDR(r5,klimit)
+       sub     r5,r5,r26
+       ld      r5,0(r5)                /* get the value of klimit */
+       sub     r5,r5,r27
+       bl      .copy_and_flush         /* copy the rest */
+       b       .start_here_multiplatform
+
+#endif /* CONFIG_PPC_MULTIPLATFORM */
+
+/*
+ * Copy routine used to copy the kernel to start at physical address 0
+ * and flush and invalidate the caches as needed.
+ * r3 = dest addr, r4 = source addr, r5 = copy limit, r6 = start offset
+ * on exit, r3, r4, r5 are unchanged, r6 is updated to be >= r5.
+ *
+ * Note: this routine *only* clobbers r0, r6 and lr
+ */
+_GLOBAL(copy_and_flush)
+       addi    r5,r5,-8
+       addi    r6,r6,-8
+4:     li      r0,16                   /* Use the least common         */
+                                       /* denominator cache line       */
+                                       /* size.  This results in       */
+                                       /* extra cache line flushes     */
+                                       /* but operation is correct.    */
+                                       /* Can't get cache line size    */
+                                       /* from NACA as it is being     */
+                                       /* moved too.                   */
+
+       mtctr   r0                      /* put # words/line in ctr      */
+3:     addi    r6,r6,8                 /* copy a cache line            */
+       ldx     r0,r6,r4
+       stdx    r0,r6,r3
+       bdnz    3b
+       dcbst   r6,r3                   /* write it to memory           */
+       sync
+       icbi    r6,r3                   /* flush the icache line        */
+       cmpld   0,r6,r5
+       blt     4b
+       sync
+       addi    r5,r5,8
+       addi    r6,r6,8
+       blr
+
+.align 8
+copy_to_here:
+
+#ifdef CONFIG_SMP
+#ifdef CONFIG_PPC_PMAC
+/*
+ * On PowerMac, secondary processors starts from the reset vector, which
+ * is temporarily turned into a call to one of the functions below.
+ */
+       .section ".text";
+       .align 2 ;
+
+       .globl  pmac_secondary_start_1  
+pmac_secondary_start_1:        
+       li      r24, 1
+       b       .pmac_secondary_start
+       
+       .globl pmac_secondary_start_2
+pmac_secondary_start_2:        
+       li      r24, 2
+       b       .pmac_secondary_start
+       
+       .globl pmac_secondary_start_3
+pmac_secondary_start_3:
+       li      r24, 3
+       b       .pmac_secondary_start
+       
+_GLOBAL(pmac_secondary_start)
+       /* turn on 64-bit mode */
+       bl      .enable_64b_mode
+       isync
+
+       /* Copy some CPU settings from CPU 0 */
+       bl      .__restore_cpu_setup
+
+       /* pSeries do that early though I don't think we really need it */
+       mfmsr   r3
+       ori     r3,r3,MSR_RI
+       mtmsrd  r3                      /* RI on */
+
+       /* Set up a paca value for this processor. */
+       LOADADDR(r4, paca)               /* Get base vaddr of paca array        */
+       mulli   r13,r24,PACA_SIZE        /* Calculate vaddr of right paca */
+       add     r13,r13,r4              /* for this processor.          */
+       mtspr   SPRG3,r13                /* Save vaddr of paca in SPRG3 */
+
+       /* Create a temp kernel stack for use before relocation is on.  */
+       ld      r1,PACAEMERGSP(r13)
+       subi    r1,r1,STACK_FRAME_OVERHEAD
+
+       b       .__secondary_start
+
+#endif /* CONFIG_PPC_PMAC */
+
+/*
+ * This function is called after the master CPU has released the
+ * secondary processors.  The execution environment is relocation off.
+ * The paca for this processor has the following fields initialized at
+ * this point:
+ *   1. Processor number
+ *   2. Segment table pointer (virtual address)
+ * On entry the following are set:
+ *   r1        = stack pointer.  vaddr for iSeries, raddr (temp stack) for pSeries
+ *   r24   = cpu# (in Linux terms)
+ *   r13   = paca virtual address
+ *   SPRG3 = paca virtual address
+ */
+_GLOBAL(__secondary_start)
+
+       HMT_MEDIUM                      /* Set thread priority to MEDIUM */
+
+       ld      r2,PACATOC(r13)
+       li      r6,0
+       stb     r6,PACAPROCENABLED(r13)
+
+#ifndef CONFIG_PPC_ISERIES
+       /* Initialize the page table pointer register. */
+       LOADADDR(r6,_SDR1)
+       ld      r6,0(r6)                /* get the value of _SDR1        */
+       mtspr   SDR1,r6                 /* set the htab location         */
+#endif
+       /* Initialize the first segment table (or SLB) entry             */
+       ld      r3,PACASTABVIRT(r13)    /* get addr of segment table     */
+       bl      .stab_initialize
+
+       /* Initialize the kernel stack.  Just a repeat for iSeries.      */
+       LOADADDR(r3,current_set)
+       sldi    r28,r24,3               /* get current_set[cpu#]         */
+       ldx     r1,r3,r28
+       addi    r1,r1,THREAD_SIZE-STACK_FRAME_OVERHEAD
+       std     r1,PACAKSAVE(r13)
+
+       ld      r3,PACASTABREAL(r13)    /* get raddr of segment table    */
+       ori     r4,r3,1                 /* turn on valid bit             */
+
+#ifdef CONFIG_PPC_ISERIES
+       li      r0,-1                   /* hypervisor call */
+       li      r3,1
+       sldi    r3,r3,63                /* 0x8000000000000000 */
+       ori     r3,r3,4                 /* 0x8000000000000004 */
+       sc                              /* HvCall_setASR */
+#else
+       /* set the ASR */
+       ld      r3,systemcfg@got(r2)    /* r3 = ptr to systemcfg         */
+       ld      r3,0(r3)
+       lwz     r3,PLATFORM(r3)         /* r3 = platform flags           */
+       andi.   r3,r3,PLATFORM_LPAR     /* Test if bit 0 is set (LPAR bit) */
+       beq     98f                     /* branch if result is 0  */
+       mfspr   r3,PVR
+       srwi    r3,r3,16
+       cmpwi   r3,0x37                 /* SStar  */
+       beq     97f
+       cmpwi   r3,0x36                 /* IStar  */
+       beq     97f
+       cmpwi   r3,0x34                 /* Pulsar */
+       bne     98f
+97:    li      r3,H_SET_ASR            /* hcall = H_SET_ASR */
+       HVSC                            /* Invoking hcall */
+       b       99f
+98:                                    /* !(rpa hypervisor) || !(star)  */
+       mtasr   r4                      /* set the stab location         */
+99:
+#endif
+       li      r7,0
+       mtlr    r7
+
+       /* enable MMU and jump to start_secondary */
+       LOADADDR(r3,.start_secondary_prolog)
+       SET_REG_TO_CONST(r4, MSR_KERNEL)
+#ifdef DO_SOFT_DISABLE
+       ori     r4,r4,MSR_EE
+#endif
+       mtspr   SRR0,r3
+       mtspr   SRR1,r4
+       rfid
+       b       .       /* prevent speculative execution */
+
+/* 
+ * Running with relocation on at this point.  All we want to do is
+ * zero the stack back-chain pointer before going into C code.
+ */
+_GLOBAL(start_secondary_prolog)
+       li      r3,0
+       std     r3,0(r1)                /* Zero the stack frame pointer */
+       bl      .start_secondary
+#endif
+
+/*
+ * This subroutine clobbers r11 and r12
+ */
+_GLOBAL(enable_64b_mode)
+       mfmsr   r11                     /* grab the current MSR */
+       li      r12,1
+       rldicr  r12,r12,MSR_SF_LG,(63-MSR_SF_LG)
+       or      r11,r11,r12
+       li      r12,1
+       rldicr  r12,r12,MSR_ISF_LG,(63-MSR_ISF_LG)
+       or      r11,r11,r12
+       mtmsrd  r11
+       isync
+       blr
+
+#ifdef CONFIG_PPC_MULTIPLATFORM
+/*
+ * This is where the main kernel code starts.
+ */
+_STATIC(start_here_multiplatform)
+       /* get a new offset, now that the kernel has moved. */
+       bl      .reloc_offset
+       mr      r26,r3
+
+       /* Clear out the BSS. It may have been done in prom_init,
+        * already but that's irrelevant since prom_init will soon
+        * be detached from the kernel completely. Besides, we need
+        * to clear it now for kexec-style entry.
+        */
+       LOADADDR(r11,__bss_stop)
+       LOADADDR(r8,__bss_start)
+       sub     r11,r11,r8              /* bss size                     */
+       addi    r11,r11,7               /* round up to an even double word */
+       rldicl. r11,r11,61,3            /* shift right by 3             */
+       beq     4f
+       addi    r8,r8,-8
+       li      r0,0
+       mtctr   r11                     /* zero this many doublewords   */
+3:     stdu    r0,8(r8)
+       bdnz    3b
+4:
+
+       mfmsr   r6
+       ori     r6,r6,MSR_RI
+       mtmsrd  r6                      /* RI on */
+
+#ifdef CONFIG_HMT
+       /* Start up the second thread on cpu 0 */
+       mfspr   r3,PVR
+       srwi    r3,r3,16
+       cmpwi   r3,0x34                 /* Pulsar  */
+       beq     90f
+       cmpwi   r3,0x36                 /* Icestar */
+       beq     90f
+       cmpwi   r3,0x37                 /* SStar   */
+       beq     90f
+       b       91f                     /* HMT not supported */
+90:    li      r3,0
+       bl      .hmt_start_secondary
+91:
+#endif
+
+       /* The following gets the stack and TOC set up with the regs */
+       /* pointing to the real addr of the kernel stack.  This is   */
+       /* all done to support the C function call below which sets  */
+       /* up the htab.  This is done because we have relocated the  */
+       /* kernel but are still running in real mode. */
+
+       LOADADDR(r3,init_thread_union)
+       sub     r3,r3,r26
+
+       /* set up a stack pointer (physical address) */
+       addi    r1,r3,THREAD_SIZE
+       li      r0,0
+       stdu    r0,-STACK_FRAME_OVERHEAD(r1)
+
+       /* set up the TOC (physical address) */
+       LOADADDR(r2,__toc_start)
+       addi    r2,r2,0x4000
+       addi    r2,r2,0x4000
+       sub     r2,r2,r26
+
+       LOADADDR(r3,cpu_specs)
+       sub     r3,r3,r26
+       LOADADDR(r4,cur_cpu_spec)
+       sub     r4,r4,r26
+       mr      r5,r26
+       bl      .identify_cpu
+
+       /* Save some low level config HIDs of CPU0 to be copied to
+        * other CPUs later on, or used for suspend/resume
+        */
+       bl      .__save_cpu_setup
+       sync
+
+       /* Setup a valid physical PACA pointer in SPRG3 for early_setup
+        * note that boot_cpuid can always be 0 nowadays since there is
+        * nowhere it can be initialized differently before we reach this
+        * code
+        */
+       LOADADDR(r27, boot_cpuid)
+       sub     r27,r27,r26
+       lwz     r27,0(r27)
+
+       LOADADDR(r24, paca)             /* Get base vaddr of paca array  */
+       mulli   r13,r27,PACA_SIZE       /* Calculate vaddr of right paca */
+       add     r13,r13,r24             /* for this processor.           */
+       sub     r13,r13,r26             /* convert to physical addr      */
+       mtspr   SPRG3,r13               /* PPPBBB: Temp... -Peter */
+       
+       /* Do very early kernel initializations, including initial hash table,
+        * stab and slb setup before we turn on relocation.     */
+
+       /* Restore parameters passed from prom_init/kexec */
+       mr      r3,r31
+       bl      .early_setup
+
+       /* set the ASR */
+       ld      r3,PACASTABREAL(r13)
+       ori     r4,r3,1                 /* turn on valid bit             */
+       ld      r3,systemcfg@got(r2)    /* r3 = ptr to systemcfg */
+       ld      r3,0(r3)
+       lwz     r3,PLATFORM(r3)         /* r3 = platform flags */
+       andi.   r3,r3,PLATFORM_LPAR     /* Test if bit 0 is set (LPAR bit) */
+       beq     98f                     /* branch if result is 0  */
+       mfspr   r3,PVR
+       srwi    r3,r3,16
+       cmpwi   r3,0x37                 /* SStar */
+       beq     97f
+       cmpwi   r3,0x36                 /* IStar  */
+       beq     97f
+       cmpwi   r3,0x34                 /* Pulsar */
+       bne     98f
+97:    li      r3,H_SET_ASR            /* hcall = H_SET_ASR */
+       HVSC                            /* Invoking hcall */
+       b       99f
+98:                                    /* !(rpa hypervisor) || !(star) */
+       mtasr   r4                      /* set the stab location        */
+99:
+       /* Set SDR1 (hash table pointer) */
+       ld      r3,systemcfg@got(r2)    /* r3 = ptr to systemcfg */
+       ld      r3,0(r3)
+       lwz     r3,PLATFORM(r3)         /* r3 = platform flags */
+       /* Test if bit 0 is set (LPAR bit) */
+       andi.   r3,r3,PLATFORM_LPAR
+       bne     98f                     /* branch if result is !0  */
+       LOADADDR(r6,_SDR1)              /* Only if NOT LPAR */
+       sub     r6,r6,r26
+       ld      r6,0(r6)                /* get the value of _SDR1 */
+       mtspr   SDR1,r6                 /* set the htab location  */
+98: 
+       LOADADDR(r3,.start_here_common)
+       SET_REG_TO_CONST(r4, MSR_KERNEL)
+       mtspr   SRR0,r3
+       mtspr   SRR1,r4
+       rfid
+       b       .       /* prevent speculative execution */
+#endif /* CONFIG_PPC_MULTIPLATFORM */
+       
+       /* This is where all platforms converge execution */
+_STATIC(start_here_common)
+       /* relocation is on at this point */
+
+       /* The following code sets up the SP and TOC now that we are */
+       /* running with translation enabled. */
+
+       LOADADDR(r3,init_thread_union)
+
+       /* set up the stack */
+       addi    r1,r3,THREAD_SIZE
+       li      r0,0
+       stdu    r0,-STACK_FRAME_OVERHEAD(r1)
+
+       /* Apply the CPUs-specific fixups (nop out sections not relevant
+        * to this CPU
+        */
+       li      r3,0
+       bl      .do_cpu_ftr_fixups
+
+       LOADADDR(r26, boot_cpuid)
+       lwz     r26,0(r26)
+
+       LOADADDR(r24, paca)             /* Get base vaddr of paca array  */
+       mulli   r13,r26,PACA_SIZE       /* Calculate vaddr of right paca */
+       add     r13,r13,r24             /* for this processor.           */
+       mtspr   SPRG3,r13
+
+       /* ptr to current */
+       LOADADDR(r4,init_task)
+       std     r4,PACACURRENT(r13)
+
+       /* Load the TOC */
+       ld      r2,PACATOC(r13)
+       std     r1,PACAKSAVE(r13)
+
+       bl      .setup_system
+
+       /* Load up the kernel context */
+5:
+#ifdef DO_SOFT_DISABLE
+       li      r5,0
+       stb     r5,PACAPROCENABLED(r13) /* Soft Disabled */
+       mfmsr   r5
+       ori     r5,r5,MSR_EE            /* Hard Enabled */
+       mtmsrd  r5
+#endif
+
+       bl .start_kernel
+
+_GLOBAL(hmt_init)
+#ifdef CONFIG_HMT
+       LOADADDR(r5, hmt_thread_data)
+       mfspr   r7,PVR
+       srwi    r7,r7,16
+       cmpwi   r7,0x34                 /* Pulsar  */
+       beq     90f
+       cmpwi   r7,0x36                 /* Icestar */
+       beq     91f
+       cmpwi   r7,0x37                 /* SStar   */
+       beq     91f
+       b       101f
+90:    mfspr   r6,PIR
+       andi.   r6,r6,0x1f
+       b       92f
+91:    mfspr   r6,PIR
+       andi.   r6,r6,0x3ff
+92:    sldi    r4,r24,3
+       stwx    r6,r5,r4
+       bl      .hmt_start_secondary
+       b       101f
+
+__hmt_secondary_hold:
+       LOADADDR(r5, hmt_thread_data)
+       clrldi  r5,r5,4
+       li      r7,0
+       mfspr   r6,PIR
+       mfspr   r8,PVR
+       srwi    r8,r8,16
+       cmpwi   r8,0x34
+       bne     93f
+       andi.   r6,r6,0x1f
+       b       103f
+93:    andi.   r6,r6,0x3f
+
+103:   lwzx    r8,r5,r7
+       cmpw    r8,r6
+       beq     104f
+       addi    r7,r7,8
+       b       103b
+
+104:   addi    r7,r7,4
+       lwzx    r9,r5,r7
+       mr      r24,r9
+101:
+#endif
+       mr      r3,r24
+       b       .pSeries_secondary_smp_init
+
+#ifdef CONFIG_HMT
+_GLOBAL(hmt_start_secondary)
+       LOADADDR(r4,__hmt_secondary_hold)
+       clrldi  r4,r4,4
+       mtspr   NIADORM, r4
+       mfspr   r4, MSRDORM
+       li      r5, -65
+       and     r4, r4, r5
+       mtspr   MSRDORM, r4
+       lis     r4,0xffef
+       ori     r4,r4,0x7403
+       mtspr   TSC, r4
+       li      r4,0x1f4
+       mtspr   TST, r4
+       mfspr   r4, HID0
+       ori     r4, r4, 0x1
+       mtspr   HID0, r4
+       mfspr   r4, SPRN_CTRLF
+       oris    r4, r4, 0x40
+       mtspr   SPRN_CTRLT, r4
+       blr
+#endif
+
+#if defined(CONFIG_KEXEC) || (defined(CONFIG_SMP) && !defined(CONFIG_PPC_ISERIES))
+_GLOBAL(smp_release_cpus)
+       /* All secondary cpus are spinning on a common
+        * spinloop, release them all now so they can start
+        * to spin on their individual paca spinloops.
+        * For non SMP kernels, the secondary cpus never
+        * get out of the common spinloop.
+        */
+       li      r3,1
+       LOADADDR(r5,__secondary_hold_spinloop)
+       std     r3,0(r5)
+       sync
+       blr
+#endif /* CONFIG_SMP && !CONFIG_PPC_ISERIES */
+
+
+/*
+ * We put a few things here that have to be page-aligned.
+ * This stuff goes at the beginning of the bss, which is page-aligned.
+ */
+       .section ".bss"
+
+       .align  PAGE_SHIFT
+
+       .globl  empty_zero_page
+empty_zero_page:
+       .space  PAGE_SIZE
+
+       .globl  swapper_pg_dir
+swapper_pg_dir:
+       .space  PAGE_SIZE
+
+/*
+ * This space gets a copy of optional info passed to us by the bootstrap
+ * Used to pass parameters into the kernel like root=/dev/sda1, etc.
+ */
+       .globl  cmd_line
+cmd_line:
+       .space  COMMAND_LINE_SIZE
diff --git a/arch/powerpc/kernel/head_8xx.S b/arch/powerpc/kernel/head_8xx.S
new file mode 100644 (file)
index 0000000..cb1a3a5
--- /dev/null
@@ -0,0 +1,860 @@
+/*
+ *  arch/ppc/kernel/except_8xx.S
+ *
+ *  PowerPC version
+ *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
+ *  Rewritten by Cort Dougan (cort@cs.nmt.edu) for PReP
+ *    Copyright (C) 1996 Cort Dougan <cort@cs.nmt.edu>
+ *  Low-level exception handlers and MMU support
+ *  rewritten by Paul Mackerras.
+ *    Copyright (C) 1996 Paul Mackerras.
+ *  MPC8xx modifications by Dan Malek
+ *    Copyright (C) 1997 Dan Malek (dmalek@jlc.net).
+ *
+ *  This file contains low-level support and setup for PowerPC 8xx
+ *  embedded processors, including trap and interrupt dispatch.
+ *
+ *  This program is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU General Public License
+ *  as published by the Free Software Foundation; either version
+ *  2 of the License, or (at your option) any later version.
+ *
+ */
+
+#include <linux/config.h>
+#include <asm/processor.h>
+#include <asm/page.h>
+#include <asm/mmu.h>
+#include <asm/cache.h>
+#include <asm/pgtable.h>
+#include <asm/cputable.h>
+#include <asm/thread_info.h>
+#include <asm/ppc_asm.h>
+#include <asm/asm-offsets.h>
+
+/* Macro to make the code more readable. */
+#ifdef CONFIG_8xx_CPU6
+#define DO_8xx_CPU6(val, reg)  \
+       li      reg, val;       \
+       stw     reg, 12(r0);    \
+       lwz     reg, 12(r0);
+#else
+#define DO_8xx_CPU6(val, reg)
+#endif
+       .text
+       .globl  _stext
+_stext:
+       .text
+       .globl  _start
+_start:
+
+/* MPC8xx
+ * This port was done on an MBX board with an 860.  Right now I only
+ * support an ELF compressed (zImage) boot from EPPC-Bug because the
+ * code there loads up some registers before calling us:
+ *   r3: ptr to board info data
+ *   r4: initrd_start or if no initrd then 0
+ *   r5: initrd_end - unused if r4 is 0
+ *   r6: Start of command line string
+ *   r7: End of command line string
+ *
+ * I decided to use conditional compilation instead of checking PVR and
+ * adding more processor specific branches around code I don't need.
+ * Since this is an embedded processor, I also appreciate any memory
+ * savings I can get.
+ *
+ * The MPC8xx does not have any BATs, but it supports large page sizes.
+ * We first initialize the MMU to support 8M byte pages, then load one
+ * entry into each of the instruction and data TLBs to map the first
+ * 8M 1:1.  I also mapped an additional I/O space 1:1 so we can get to
+ * the "internal" processor registers before MMU_init is called.
+ *
+ * The TLB code currently contains a major hack.  Since I use the condition
+ * code register, I have to save and restore it.  I am out of registers, so
+ * I just store it in memory location 0 (the TLB handlers are not reentrant).
+ * To avoid making any decisions, I need to use the "segment" valid bit
+ * in the first level table, but that would require many changes to the
+ * Linux page directory/table functions that I don't want to do right now.
+ *
+ * I used to use SPRG2 for a temporary register in the TLB handler, but it
+ * has since been put to other uses.  I now use a hack to save a register
+ * and the CCR at memory location 0.....Someday I'll fix this.....
+ *     -- Dan
+ */
+       .globl  __start
+__start:
+       mr      r31,r3                  /* save parameters */
+       mr      r30,r4
+       mr      r29,r5
+       mr      r28,r6
+       mr      r27,r7
+
+       /* We have to turn on the MMU right away so we get cache modes
+        * set correctly.
+        */
+       bl      initial_mmu
+
+/* We now have the lower 8 Meg mapped into TLB entries, and the caches
+ * ready to work.
+ */
+
+turn_on_mmu:
+       mfmsr   r0
+       ori     r0,r0,MSR_DR|MSR_IR
+       mtspr   SPRN_SRR1,r0
+       lis     r0,start_here@h
+       ori     r0,r0,start_here@l
+       mtspr   SPRN_SRR0,r0
+       SYNC
+       rfi                             /* enables MMU */
+
+/*
+ * Exception entry code.  This code runs with address translation
+ * turned off, i.e. using physical addresses.
+ * We assume sprg3 has the physical address of the current
+ * task's thread_struct.
+ */
+#define EXCEPTION_PROLOG       \
+       mtspr   SPRN_SPRG0,r10; \
+       mtspr   SPRN_SPRG1,r11; \
+       mfcr    r10;            \
+       EXCEPTION_PROLOG_1;     \
+       EXCEPTION_PROLOG_2
+
+#define EXCEPTION_PROLOG_1     \
+       mfspr   r11,SPRN_SRR1;          /* check whether user or kernel */ \
+       andi.   r11,r11,MSR_PR; \
+       tophys(r11,r1);                 /* use tophys(r1) if kernel */ \
+       beq     1f;             \
+       mfspr   r11,SPRN_SPRG3; \
+       lwz     r11,THREAD_INFO-THREAD(r11);    \
+       addi    r11,r11,THREAD_SIZE;    \
+       tophys(r11,r11);        \
+1:     subi    r11,r11,INT_FRAME_SIZE  /* alloc exc. frame */
+
+
+#define EXCEPTION_PROLOG_2     \
+       CLR_TOP32(r11);         \
+       stw     r10,_CCR(r11);          /* save registers */ \
+       stw     r12,GPR12(r11); \
+       stw     r9,GPR9(r11);   \
+       mfspr   r10,SPRN_SPRG0; \
+       stw     r10,GPR10(r11); \
+       mfspr   r12,SPRN_SPRG1; \
+       stw     r12,GPR11(r11); \
+       mflr    r10;            \
+       stw     r10,_LINK(r11); \
+       mfspr   r12,SPRN_SRR0;  \
+       mfspr   r9,SPRN_SRR1;   \
+       stw     r1,GPR1(r11);   \
+       stw     r1,0(r11);      \
+       tovirt(r1,r11);                 /* set new kernel sp */ \
+       li      r10,MSR_KERNEL & ~(MSR_IR|MSR_DR); /* can take exceptions */ \
+       MTMSRD(r10);                    /* (except for mach check in rtas) */ \
+       stw     r0,GPR0(r11);   \
+       SAVE_4GPRS(3, r11);     \
+       SAVE_2GPRS(7, r11)
+
+/*
+ * Note: code which follows this uses cr0.eq (set if from kernel),
+ * r11, r12 (SRR0), and r9 (SRR1).
+ *
+ * Note2: once we have set r1 we are in a position to take exceptions
+ * again, and we could thus set MSR:RI at that point.
+ */
+
+/*
+ * Exception vectors.
+ */
+#define EXCEPTION(n, label, hdlr, xfer)                \
+       . = n;                                  \
+label:                                         \
+       EXCEPTION_PROLOG;                       \
+       addi    r3,r1,STACK_FRAME_OVERHEAD;     \
+       xfer(n, hdlr)
+
+#define EXC_XFER_TEMPLATE(n, hdlr, trap, copyee, tfer, ret)    \
+       li      r10,trap;                                       \
+       stw     r10,TRAP(r11);                                  \
+       li      r10,MSR_KERNEL;                                 \
+       copyee(r10, r9);                                        \
+       bl      tfer;                                           \
+i##n:                                                          \
+       .long   hdlr;                                           \
+       .long   ret
+
+#define COPY_EE(d, s)          rlwimi d,s,0,16,16
+#define NOCOPY(d, s)
+
+#define EXC_XFER_STD(n, hdlr)          \
+       EXC_XFER_TEMPLATE(n, hdlr, n, NOCOPY, transfer_to_handler_full, \
+                         ret_from_except_full)
+
+#define EXC_XFER_LITE(n, hdlr)         \
+       EXC_XFER_TEMPLATE(n, hdlr, n+1, NOCOPY, transfer_to_handler, \
+                         ret_from_except)
+
+#define EXC_XFER_EE(n, hdlr)           \
+       EXC_XFER_TEMPLATE(n, hdlr, n, COPY_EE, transfer_to_handler_full, \
+                         ret_from_except_full)
+
+#define EXC_XFER_EE_LITE(n, hdlr)      \
+       EXC_XFER_TEMPLATE(n, hdlr, n+1, COPY_EE, transfer_to_handler, \
+                         ret_from_except)
+
+/* System reset */
+       EXCEPTION(0x100, Reset, UnknownException, EXC_XFER_STD)
+
+/* Machine check */
+       . = 0x200
+MachineCheck:
+       EXCEPTION_PROLOG
+       mfspr r4,SPRN_DAR
+       stw r4,_DAR(r11)
+       mfspr r5,SPRN_DSISR
+       stw r5,_DSISR(r11)
+       addi r3,r1,STACK_FRAME_OVERHEAD
+       EXC_XFER_STD(0x200, MachineCheckException)
+
+/* Data access exception.
+ * This is "never generated" by the MPC8xx.  We jump to it for other
+ * translation errors.
+ */
+       . = 0x300
+DataAccess:
+       EXCEPTION_PROLOG
+       mfspr   r10,SPRN_DSISR
+       stw     r10,_DSISR(r11)
+       mr      r5,r10
+       mfspr   r4,SPRN_DAR
+       EXC_XFER_EE_LITE(0x300, handle_page_fault)
+
+/* Instruction access exception.
+ * This is "never generated" by the MPC8xx.  We jump to it for other
+ * translation errors.
+ */
+       . = 0x400
+InstructionAccess:
+       EXCEPTION_PROLOG
+       mr      r4,r12
+       mr      r5,r9
+       EXC_XFER_EE_LITE(0x400, handle_page_fault)
+
+/* External interrupt */
+       EXCEPTION(0x500, HardwareInterrupt, do_IRQ, EXC_XFER_LITE)
+
+/* Alignment exception */
+       . = 0x600
+Alignment:
+       EXCEPTION_PROLOG
+       mfspr   r4,SPRN_DAR
+       stw     r4,_DAR(r11)
+       mfspr   r5,SPRN_DSISR
+       stw     r5,_DSISR(r11)
+       addi    r3,r1,STACK_FRAME_OVERHEAD
+       EXC_XFER_EE(0x600, AlignmentException)
+
+/* Program check exception */
+       EXCEPTION(0x700, ProgramCheck, ProgramCheckException, EXC_XFER_STD)
+
+/* No FPU on MPC8xx.  This exception is not supposed to happen.
+*/
+       EXCEPTION(0x800, FPUnavailable, UnknownException, EXC_XFER_STD)
+
+/* Decrementer */
+       EXCEPTION(0x900, Decrementer, timer_interrupt, EXC_XFER_LITE)
+
+       EXCEPTION(0xa00, Trap_0a, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0xb00, Trap_0b, UnknownException, EXC_XFER_EE)
+
+/* System call */
+       . = 0xc00
+SystemCall:
+       EXCEPTION_PROLOG
+       EXC_XFER_EE_LITE(0xc00, DoSyscall)
+
+/* Single step - not used on 601 */
+       EXCEPTION(0xd00, SingleStep, SingleStepException, EXC_XFER_STD)
+       EXCEPTION(0xe00, Trap_0e, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0xf00, Trap_0f, UnknownException, EXC_XFER_EE)
+
+/* On the MPC8xx, this is a software emulation interrupt.  It occurs
+ * for all unimplemented and illegal instructions.
+ */
+       EXCEPTION(0x1000, SoftEmu, SoftwareEmulation, EXC_XFER_STD)
+
+       . = 0x1100
+/*
+ * For the MPC8xx, this is a software tablewalk to load the instruction
+ * TLB.  It is modelled after the example in the Motorola manual.  The task
+ * switch loads the M_TWB register with the pointer to the first level table.
+ * If we discover there is no second level table (value is zero) or if there
+ * is an invalid pte, we load that into the TLB, which causes another fault
+ * into the TLB Error interrupt where we can handle such problems.
+ * We have to use the MD_xxx registers for the tablewalk because the
+ * equivalent MI_xxx registers only perform the attribute functions.
+ */
+InstructionTLBMiss:
+#ifdef CONFIG_8xx_CPU6
+       stw     r3, 8(r0)
+#endif
+       DO_8xx_CPU6(0x3f80, r3)
+       mtspr   SPRN_M_TW, r10  /* Save a couple of working registers */
+       mfcr    r10
+       stw     r10, 0(r0)
+       stw     r11, 4(r0)
+       mfspr   r10, SPRN_SRR0  /* Get effective address of fault */
+       DO_8xx_CPU6(0x3780, r3)
+       mtspr   SPRN_MD_EPN, r10        /* Have to use MD_EPN for walk, MI_EPN can't */
+       mfspr   r10, SPRN_M_TWB /* Get level 1 table entry address */
+
+       /* If we are faulting a kernel address, we have to use the
+        * kernel page tables.
+        */
+       andi.   r11, r10, 0x0800        /* Address >= 0x80000000 */
+       beq     3f
+       lis     r11, swapper_pg_dir@h
+       ori     r11, r11, swapper_pg_dir@l
+       rlwimi  r10, r11, 0, 2, 19
+3:
+       lwz     r11, 0(r10)     /* Get the level 1 entry */
+       rlwinm. r10, r11,0,0,19 /* Extract page descriptor page address */
+       beq     2f              /* If zero, don't try to find a pte */
+
+       /* We have a pte table, so load the MI_TWC with the attributes
+        * for this "segment."
+        */
+       ori     r11,r11,1               /* Set valid bit */
+       DO_8xx_CPU6(0x2b80, r3)
+       mtspr   SPRN_MI_TWC, r11        /* Set segment attributes */
+       DO_8xx_CPU6(0x3b80, r3)
+       mtspr   SPRN_MD_TWC, r11        /* Load pte table base address */
+       mfspr   r11, SPRN_MD_TWC        /* ....and get the pte address */
+       lwz     r10, 0(r11)     /* Get the pte */
+
+       ori     r10, r10, _PAGE_ACCESSED
+       stw     r10, 0(r11)
+
+       /* The Linux PTE won't go exactly into the MMU TLB.
+        * Software indicator bits 21, 22 and 28 must be clear.
+        * Software indicator bits 24, 25, 26, and 27 must be
+        * set.  All other Linux PTE bits control the behavior
+        * of the MMU.
+        */
+2:     li      r11, 0x00f0
+       rlwimi  r10, r11, 0, 24, 28     /* Set 24-27, clear 28 */
+       DO_8xx_CPU6(0x2d80, r3)
+       mtspr   SPRN_MI_RPN, r10        /* Update TLB entry */
+
+       mfspr   r10, SPRN_M_TW  /* Restore registers */
+       lwz     r11, 0(r0)
+       mtcr    r11
+       lwz     r11, 4(r0)
+#ifdef CONFIG_8xx_CPU6
+       lwz     r3, 8(r0)
+#endif
+       rfi
+
+       . = 0x1200
+DataStoreTLBMiss:
+#ifdef CONFIG_8xx_CPU6
+       stw     r3, 8(r0)
+#endif
+       DO_8xx_CPU6(0x3f80, r3)
+       mtspr   SPRN_M_TW, r10  /* Save a couple of working registers */
+       mfcr    r10
+       stw     r10, 0(r0)
+       stw     r11, 4(r0)
+       mfspr   r10, SPRN_M_TWB /* Get level 1 table entry address */
+
+       /* If we are faulting a kernel address, we have to use the
+        * kernel page tables.
+        */
+       andi.   r11, r10, 0x0800
+       beq     3f
+       lis     r11, swapper_pg_dir@h
+       ori     r11, r11, swapper_pg_dir@l
+       rlwimi  r10, r11, 0, 2, 19
+3:
+       lwz     r11, 0(r10)     /* Get the level 1 entry */
+       rlwinm. r10, r11,0,0,19 /* Extract page descriptor page address */
+       beq     2f              /* If zero, don't try to find a pte */
+
+       /* We have a pte table, so load fetch the pte from the table.
+        */
+       ori     r11, r11, 1     /* Set valid bit in physical L2 page */
+       DO_8xx_CPU6(0x3b80, r3)
+       mtspr   SPRN_MD_TWC, r11        /* Load pte table base address */
+       mfspr   r10, SPRN_MD_TWC        /* ....and get the pte address */
+       lwz     r10, 0(r10)     /* Get the pte */
+
+       /* Insert the Guarded flag into the TWC from the Linux PTE.
+        * It is bit 27 of both the Linux PTE and the TWC (at least
+        * I got that right :-).  It will be better when we can put
+        * this into the Linux pgd/pmd and load it in the operation
+        * above.
+        */
+       rlwimi  r11, r10, 0, 27, 27
+       DO_8xx_CPU6(0x3b80, r3)
+       mtspr   SPRN_MD_TWC, r11
+
+       mfspr   r11, SPRN_MD_TWC        /* get the pte address again */
+       ori     r10, r10, _PAGE_ACCESSED
+       stw     r10, 0(r11)
+
+       /* The Linux PTE won't go exactly into the MMU TLB.
+        * Software indicator bits 21, 22 and 28 must be clear.
+        * Software indicator bits 24, 25, 26, and 27 must be
+        * set.  All other Linux PTE bits control the behavior
+        * of the MMU.
+        */
+2:     li      r11, 0x00f0
+       rlwimi  r10, r11, 0, 24, 28     /* Set 24-27, clear 28 */
+       DO_8xx_CPU6(0x3d80, r3)
+       mtspr   SPRN_MD_RPN, r10        /* Update TLB entry */
+
+       mfspr   r10, SPRN_M_TW  /* Restore registers */
+       lwz     r11, 0(r0)
+       mtcr    r11
+       lwz     r11, 4(r0)
+#ifdef CONFIG_8xx_CPU6
+       lwz     r3, 8(r0)
+#endif
+       rfi
+
+/* This is an instruction TLB error on the MPC8xx.  This could be due
+ * to many reasons, such as executing guarded memory or illegal instruction
+ * addresses.  There is nothing to do but handle a big time error fault.
+ */
+       . = 0x1300
+InstructionTLBError:
+       b       InstructionAccess
+
+/* This is the data TLB error on the MPC8xx.  This could be due to
+ * many reasons, including a dirty update to a pte.  We can catch that
+ * one here, but anything else is an error.  First, we track down the
+ * Linux pte.  If it is valid, write access is allowed, but the
+ * page dirty bit is not set, we will set it and reload the TLB.  For
+ * any other case, we bail out to a higher level function that can
+ * handle it.
+ */
+       . = 0x1400
+DataTLBError:
+#ifdef CONFIG_8xx_CPU6
+       stw     r3, 8(r0)
+#endif
+       DO_8xx_CPU6(0x3f80, r3)
+       mtspr   SPRN_M_TW, r10  /* Save a couple of working registers */
+       mfcr    r10
+       stw     r10, 0(r0)
+       stw     r11, 4(r0)
+
+       /* First, make sure this was a store operation.
+       */
+       mfspr   r10, SPRN_DSISR
+       andis.  r11, r10, 0x0200        /* If set, indicates store op */
+       beq     2f
+
+       /* The EA of a data TLB miss is automatically stored in the MD_EPN
+        * register.  The EA of a data TLB error is automatically stored in
+        * the DAR, but not the MD_EPN register.  We must copy the 20 most
+        * significant bits of the EA from the DAR to MD_EPN before we
+        * start walking the page tables.  We also need to copy the CASID
+        * value from the M_CASID register.
+        * Addendum:  The EA of a data TLB error is _supposed_ to be stored
+        * in DAR, but it seems that this doesn't happen in some cases, such
+        * as when the error is due to a dcbi instruction to a page with a
+        * TLB that doesn't have the changed bit set.  In such cases, there
+        * does not appear to be any way  to recover the EA of the error
+        * since it is neither in DAR nor MD_EPN.  As a workaround, the
+        * _PAGE_HWWRITE bit is set for all kernel data pages when the PTEs
+        * are initialized in mapin_ram().  This will avoid the problem,
+        * assuming we only use the dcbi instruction on kernel addresses.
+        */
+       mfspr   r10, SPRN_DAR
+       rlwinm  r11, r10, 0, 0, 19
+       ori     r11, r11, MD_EVALID
+       mfspr   r10, SPRN_M_CASID
+       rlwimi  r11, r10, 0, 28, 31
+       DO_8xx_CPU6(0x3780, r3)
+       mtspr   SPRN_MD_EPN, r11
+
+       mfspr   r10, SPRN_M_TWB /* Get level 1 table entry address */
+
+       /* If we are faulting a kernel address, we have to use the
+        * kernel page tables.
+        */
+       andi.   r11, r10, 0x0800
+       beq     3f
+       lis     r11, swapper_pg_dir@h
+       ori     r11, r11, swapper_pg_dir@l
+       rlwimi  r10, r11, 0, 2, 19
+3:
+       lwz     r11, 0(r10)     /* Get the level 1 entry */
+       rlwinm. r10, r11,0,0,19 /* Extract page descriptor page address */
+       beq     2f              /* If zero, bail */
+
+       /* We have a pte table, so fetch the pte from the table.
+        */
+       ori     r11, r11, 1             /* Set valid bit in physical L2 page */
+       DO_8xx_CPU6(0x3b80, r3)
+       mtspr   SPRN_MD_TWC, r11                /* Load pte table base address */
+       mfspr   r11, SPRN_MD_TWC                /* ....and get the pte address */
+       lwz     r10, 0(r11)             /* Get the pte */
+
+       andi.   r11, r10, _PAGE_RW      /* Is it writeable? */
+       beq     2f                      /* Bail out if not */
+
+       /* Update 'changed', among others.
+       */
+       ori     r10, r10, _PAGE_DIRTY|_PAGE_ACCESSED|_PAGE_HWWRITE
+       mfspr   r11, SPRN_MD_TWC                /* Get pte address again */
+       stw     r10, 0(r11)             /* and update pte in table */
+
+       /* The Linux PTE won't go exactly into the MMU TLB.
+        * Software indicator bits 21, 22 and 28 must be clear.
+        * Software indicator bits 24, 25, 26, and 27 must be
+        * set.  All other Linux PTE bits control the behavior
+        * of the MMU.
+        */
+       li      r11, 0x00f0
+       rlwimi  r10, r11, 0, 24, 28     /* Set 24-27, clear 28 */
+       DO_8xx_CPU6(0x3d80, r3)
+       mtspr   SPRN_MD_RPN, r10        /* Update TLB entry */
+
+       mfspr   r10, SPRN_M_TW  /* Restore registers */
+       lwz     r11, 0(r0)
+       mtcr    r11
+       lwz     r11, 4(r0)
+#ifdef CONFIG_8xx_CPU6
+       lwz     r3, 8(r0)
+#endif
+       rfi
+2:
+       mfspr   r10, SPRN_M_TW  /* Restore registers */
+       lwz     r11, 0(r0)
+       mtcr    r11
+       lwz     r11, 4(r0)
+#ifdef CONFIG_8xx_CPU6
+       lwz     r3, 8(r0)
+#endif
+       b       DataAccess
+
+       EXCEPTION(0x1500, Trap_15, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x1600, Trap_16, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x1700, Trap_17, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x1800, Trap_18, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x1900, Trap_19, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x1a00, Trap_1a, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x1b00, Trap_1b, UnknownException, EXC_XFER_EE)
+
+/* On the MPC8xx, these next four traps are used for development
+ * support of breakpoints and such.  Someday I will get around to
+ * using them.
+ */
+       EXCEPTION(0x1c00, Trap_1c, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x1d00, Trap_1d, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x1e00, Trap_1e, UnknownException, EXC_XFER_EE)
+       EXCEPTION(0x1f00, Trap_1f, UnknownException, EXC_XFER_EE)
+
+       . = 0x2000
+
+       .globl  giveup_fpu
+giveup_fpu:
+       blr
+
+/*
+ * This is where the main kernel code starts.
+ */
+start_here:
+       /* ptr to current */
+       lis     r2,init_task@h
+       ori     r2,r2,init_task@l
+
+       /* ptr to phys current thread */
+       tophys(r4,r2)
+       addi    r4,r4,THREAD    /* init task's THREAD */
+       mtspr   SPRN_SPRG3,r4
+       li      r3,0
+       mtspr   SPRN_SPRG2,r3   /* 0 => r1 has kernel sp */
+
+       /* stack */
+       lis     r1,init_thread_union@ha
+       addi    r1,r1,init_thread_union@l
+       li      r0,0
+       stwu    r0,THREAD_SIZE-STACK_FRAME_OVERHEAD(r1)
+
+       bl      early_init      /* We have to do this with MMU on */
+
+/*
+ * Decide what sort of machine this is and initialize the MMU.
+ */
+       mr      r3,r31
+       mr      r4,r30
+       mr      r5,r29
+       mr      r6,r28
+       mr      r7,r27
+       bl      machine_init
+       bl      MMU_init
+
+/*
+ * Go back to running unmapped so we can load up new values
+ * and change to using our exception vectors.
+ * On the 8xx, all we have to do is invalidate the TLB to clear
+ * the old 8M byte TLB mappings and load the page table base register.
+ */
+       /* The right way to do this would be to track it down through
+        * init's THREAD like the context switch code does, but this is
+        * easier......until someone changes init's static structures.
+        */
+       lis     r6, swapper_pg_dir@h
+       ori     r6, r6, swapper_pg_dir@l
+       tophys(r6,r6)
+#ifdef CONFIG_8xx_CPU6
+       lis     r4, cpu6_errata_word@h
+       ori     r4, r4, cpu6_errata_word@l
+       li      r3, 0x3980
+       stw     r3, 12(r4)
+       lwz     r3, 12(r4)
+#endif
+       mtspr   SPRN_M_TWB, r6
+       lis     r4,2f@h
+       ori     r4,r4,2f@l
+       tophys(r4,r4)
+       li      r3,MSR_KERNEL & ~(MSR_IR|MSR_DR)
+       mtspr   SPRN_SRR0,r4
+       mtspr   SPRN_SRR1,r3
+       rfi
+/* Load up the kernel context */
+2:
+       SYNC                    /* Force all PTE updates to finish */
+       tlbia                   /* Clear all TLB entries */
+       sync                    /* wait for tlbia/tlbie to finish */
+       TLBSYNC                 /* ... on all CPUs */
+
+       /* set up the PTE pointers for the Abatron bdiGDB.
+       */
+       tovirt(r6,r6)
+       lis     r5, abatron_pteptrs@h
+       ori     r5, r5, abatron_pteptrs@l
+       stw     r5, 0xf0(r0)    /* Must match your Abatron config file */
+       tophys(r5,r5)
+       stw     r6, 0(r5)
+
+/* Now turn on the MMU for real! */
+       li      r4,MSR_KERNEL
+       lis     r3,start_kernel@h
+       ori     r3,r3,start_kernel@l
+       mtspr   SPRN_SRR0,r3
+       mtspr   SPRN_SRR1,r4
+       rfi                     /* enable MMU and jump to start_kernel */
+
+/* Set up the initial MMU state so we can do the first level of
+ * kernel initialization.  This maps the first 8 MBytes of memory 1:1
+ * virtual to physical.  Also, set the cache mode since that is defined
+ * by TLB entries and perform any additional mapping (like of the IMMR).
+ * If configured to pin some TLBs, we pin the first 8 Mbytes of kernel,
+ * 24 Mbytes of data, and the 8M IMMR space.  Anything not covered by
+ * these mappings is mapped by page tables.
+ */
+initial_mmu:
+       tlbia                   /* Invalidate all TLB entries */
+#ifdef CONFIG_PIN_TLB
+       lis     r8, MI_RSV4I@h
+       ori     r8, r8, 0x1c00
+#else
+       li      r8, 0
+#endif
+       mtspr   SPRN_MI_CTR, r8 /* Set instruction MMU control */
+
+#ifdef CONFIG_PIN_TLB
+       lis     r10, (MD_RSV4I | MD_RESETVAL)@h
+       ori     r10, r10, 0x1c00
+       mr      r8, r10
+#else
+       lis     r10, MD_RESETVAL@h
+#endif
+#ifndef CONFIG_8xx_COPYBACK
+       oris    r10, r10, MD_WTDEF@h
+#endif
+       mtspr   SPRN_MD_CTR, r10        /* Set data TLB control */
+
+       /* Now map the lower 8 Meg into the TLBs.  For this quick hack,
+        * we can load the instruction and data TLB registers with the
+        * same values.
+        */
+       lis     r8, KERNELBASE@h        /* Create vaddr for TLB */
+       ori     r8, r8, MI_EVALID       /* Mark it valid */
+       mtspr   SPRN_MI_EPN, r8
+       mtspr   SPRN_MD_EPN, r8
+       li      r8, MI_PS8MEG           /* Set 8M byte page */
+       ori     r8, r8, MI_SVALID       /* Make it valid */
+       mtspr   SPRN_MI_TWC, r8
+       mtspr   SPRN_MD_TWC, r8
+       li      r8, MI_BOOTINIT         /* Create RPN for address 0 */
+       mtspr   SPRN_MI_RPN, r8         /* Store TLB entry */
+       mtspr   SPRN_MD_RPN, r8
+       lis     r8, MI_Kp@h             /* Set the protection mode */
+       mtspr   SPRN_MI_AP, r8
+       mtspr   SPRN_MD_AP, r8
+
+       /* Map another 8 MByte at the IMMR to get the processor
+        * internal registers (among other things).
+        */
+#ifdef CONFIG_PIN_TLB
+       addi    r10, r10, 0x0100
+       mtspr   SPRN_MD_CTR, r10
+#endif
+       mfspr   r9, 638                 /* Get current IMMR */
+       andis.  r9, r9, 0xff80          /* Get 8Mbyte boundary */
+
+       mr      r8, r9                  /* Create vaddr for TLB */
+       ori     r8, r8, MD_EVALID       /* Mark it valid */
+       mtspr   SPRN_MD_EPN, r8
+       li      r8, MD_PS8MEG           /* Set 8M byte page */
+       ori     r8, r8, MD_SVALID       /* Make it valid */
+       mtspr   SPRN_MD_TWC, r8
+       mr      r8, r9                  /* Create paddr for TLB */
+       ori     r8, r8, MI_BOOTINIT|0x2 /* Inhibit cache -- Cort */
+       mtspr   SPRN_MD_RPN, r8
+
+#ifdef CONFIG_PIN_TLB
+       /* Map two more 8M kernel data pages.
+       */
+       addi    r10, r10, 0x0100
+       mtspr   SPRN_MD_CTR, r10
+
+       lis     r8, KERNELBASE@h        /* Create vaddr for TLB */
+       addis   r8, r8, 0x0080          /* Add 8M */
+       ori     r8, r8, MI_EVALID       /* Mark it valid */
+       mtspr   SPRN_MD_EPN, r8
+       li      r9, MI_PS8MEG           /* Set 8M byte page */
+       ori     r9, r9, MI_SVALID       /* Make it valid */
+       mtspr   SPRN_MD_TWC, r9
+       li      r11, MI_BOOTINIT        /* Create RPN for address 0 */
+       addis   r11, r11, 0x0080        /* Add 8M */
+       mtspr   SPRN_MD_RPN, r8
+
+       addis   r8, r8, 0x0080          /* Add 8M */
+       mtspr   SPRN_MD_EPN, r8
+       mtspr   SPRN_MD_TWC, r9
+       addis   r11, r11, 0x0080        /* Add 8M */
+       mtspr   SPRN_MD_RPN, r8
+#endif
+
+       /* Since the cache is enabled according to the information we
+        * just loaded into the TLB, invalidate and enable the caches here.
+        * We should probably check/set other modes....later.
+        */
+       lis     r8, IDC_INVALL@h
+       mtspr   SPRN_IC_CST, r8
+       mtspr   SPRN_DC_CST, r8
+       lis     r8, IDC_ENABLE@h
+       mtspr   SPRN_IC_CST, r8
+#ifdef CONFIG_8xx_COPYBACK
+       mtspr   SPRN_DC_CST, r8
+#else
+       /* For a debug option, I left this here to easily enable
+        * the write through cache mode
+        */
+       lis     r8, DC_SFWT@h
+       mtspr   SPRN_DC_CST, r8
+       lis     r8, IDC_ENABLE@h
+       mtspr   SPRN_DC_CST, r8
+#endif
+       blr
+
+
+/*
+ * Set up to use a given MMU context.
+ * r3 is context number, r4 is PGD pointer.
+ *
+ * We place the physical address of the new task page directory loaded
+ * into the MMU base register, and set the ASID compare register with
+ * the new "context."
+ */
+_GLOBAL(set_context)
+
+#ifdef CONFIG_BDI_SWITCH
+       /* Context switch the PTE pointer for the Abatron BDI2000.
+        * The PGDIR is passed as second argument.
+        */
+       lis     r5, KERNELBASE@h
+       lwz     r5, 0xf0(r5)
+       stw     r4, 0x4(r5)
+#endif
+
+#ifdef CONFIG_8xx_CPU6
+       lis     r6, cpu6_errata_word@h
+       ori     r6, r6, cpu6_errata_word@l
+       tophys  (r4, r4)
+       li      r7, 0x3980
+       stw     r7, 12(r6)
+       lwz     r7, 12(r6)
+        mtspr   SPRN_M_TWB, r4               /* Update MMU base address */
+       li      r7, 0x3380
+       stw     r7, 12(r6)
+       lwz     r7, 12(r6)
+        mtspr   SPRN_M_CASID, r3             /* Update context */
+#else
+        mtspr   SPRN_M_CASID,r3                /* Update context */
+       tophys  (r4, r4)
+       mtspr   SPRN_M_TWB, r4          /* and pgd */
+#endif
+       SYNC
+       blr
+
+#ifdef CONFIG_8xx_CPU6
+/* It's here because it is unique to the 8xx.
+ * It is important we get called with interrupts disabled.  I used to
+ * do that, but it appears that all code that calls this already had
+ * interrupt disabled.
+ */
+       .globl  set_dec_cpu6
+set_dec_cpu6:
+       lis     r7, cpu6_errata_word@h
+       ori     r7, r7, cpu6_errata_word@l
+       li      r4, 0x2c00
+       stw     r4, 8(r7)
+       lwz     r4, 8(r7)
+        mtspr   22, r3         /* Update Decrementer */
+       SYNC
+       blr
+#endif
+
+/*
+ * We put a few things here that have to be page-aligned.
+ * This stuff goes at the beginning of the data segment,
+ * which is page-aligned.
+ */
+       .data
+       .globl  sdata
+sdata:
+       .globl  empty_zero_page
+empty_zero_page:
+       .space  4096
+
+       .globl  swapper_pg_dir
+swapper_pg_dir:
+       .space  4096
+
+/*
+ * This space gets a copy of optional info passed to us by the bootstrap
+ * Used to pass parameters into the kernel like root=/dev/sda1, etc.
+ */
+       .globl  cmd_line
+cmd_line:
+       .space  512
+
+/* Room for two PTE table poiners, usually the kernel and current user
+ * pointer to their respective root page table (pgdir).
+ */
+abatron_pteptrs:
+       .space  8
+
+#ifdef CONFIG_8xx_CPU6
+       .globl  cpu6_errata_word
+cpu6_errata_word:
+       .space  16
+#endif
+
diff --git a/arch/powerpc/kernel/head_fsl_booke.S b/arch/powerpc/kernel/head_fsl_booke.S
new file mode 100644 (file)
index 0000000..eba5a5f
--- /dev/null
@@ -0,0 +1,1058 @@
+/*
+ * arch/ppc/kernel/head_fsl_booke.S
+ *
+ * Kernel execution entry point code.
+ *
+ *    Copyright (c) 1995-1996 Gary Thomas <gdt@linuxppc.org>
+ *      Initial PowerPC version.
+ *    Copyright (c) 1996 Cort Dougan <cort@cs.nmt.edu>
+ *      Rewritten for PReP
+ *    Copyright (c) 1996 Paul Mackerras <paulus@cs.anu.edu.au>
+ *      Low-level exception handers, MMU support, and rewrite.
+ *    Copyright (c) 1997 Dan Malek <dmalek@jlc.net>
+ *      PowerPC 8xx modifications.
+ *    Copyright (c) 1998-1999 TiVo, Inc.
+ *      PowerPC 403GCX modifications.
+ *    Copyright (c) 1999 Grant Erickson <grant@lcse.umn.edu>
+ *      PowerPC 403GCX/405GP modifications.
+ *    Copyright 2000 MontaVista Software Inc.
+ *     PPC405 modifications
+ *      PowerPC 403GCX/405GP modifications.
+ *     Author: MontaVista Software, Inc.
+ *             frank_rowand@mvista.com or source@mvista.com
+ *             debbie_chu@mvista.com
+ *    Copyright 2002-2004 MontaVista Software, Inc.
+ *      PowerPC 44x support, Matt Porter <mporter@kernel.crashing.org>
+ *    Copyright 2004 Freescale Semiconductor, Inc
+ *      PowerPC e500 modifications, Kumar Gala <kumar.gala@freescale.com>
+ *
+ * This program is free software; you can redistribute  it and/or modify it
+ * under  the terms of  the GNU General  Public License as published by the
+ * Free Software Foundation;  either version 2 of the  License, or (at your
+ * option) any later version.
+ */
+
+#include <linux/config.h>
+#include <linux/threads.h>
+#include <asm/processor.h>
+#include <asm/page.h>
+#include <asm/mmu.h>
+#include <asm/pgtable.h>
+#include <asm/cputable.h>
+#include <asm/thread_info.h>
+#include <asm/ppc_asm.h>
+#include <asm/asm-offsets.h>
+#include "head_booke.h"
+
+/* As with the other PowerPC ports, it is expected that when code
+ * execution begins here, the following registers contain valid, yet
+ * optional, information:
+ *
+ *   r3 - Board info structure pointer (DRAM, frequency, MAC address, etc.)
+ *   r4 - Starting address of the init RAM disk
+ *   r5 - Ending address of the init RAM disk
+ *   r6 - Start of kernel command line string (e.g. "mem=128")
+ *   r7 - End of kernel command line string
+ *
+ */
+       .text
+_GLOBAL(_stext)
+_GLOBAL(_start)
+       /*
+        * Reserve a word at a fixed location to store the address
+        * of abatron_pteptrs
+        */
+       nop
+/*
+ * Save parameters we are passed
+ */
+       mr      r31,r3
+       mr      r30,r4
+       mr      r29,r5
+       mr      r28,r6
+       mr      r27,r7
+       li      r24,0           /* CPU number */
+
+/* We try to not make any assumptions about how the boot loader
+ * setup or used the TLBs.  We invalidate all mappings from the
+ * boot loader and load a single entry in TLB1[0] to map the
+ * first 16M of kernel memory.  Any boot info passed from the
+ * bootloader needs to live in this first 16M.
+ *
+ * Requirement on bootloader:
+ *  - The page we're executing in needs to reside in TLB1 and
+ *    have IPROT=1.  If not an invalidate broadcast could
+ *    evict the entry we're currently executing in.
+ *
+ *  r3 = Index of TLB1 were executing in
+ *  r4 = Current MSR[IS]
+ *  r5 = Index of TLB1 temp mapping
+ *
+ * Later in mapin_ram we will correctly map lowmem, and resize TLB1[0]
+ * if needed
+ */
+
+/* 1. Find the index of the entry we're executing in */
+       bl      invstr                          /* Find our address */
+invstr:        mflr    r6                              /* Make it accessible */
+       mfmsr   r7
+       rlwinm  r4,r7,27,31,31                  /* extract MSR[IS] */
+       mfspr   r7, SPRN_PID0
+       slwi    r7,r7,16
+       or      r7,r7,r4
+       mtspr   SPRN_MAS6,r7
+       tlbsx   0,r6                            /* search MSR[IS], SPID=PID0 */
+#ifndef CONFIG_E200
+       mfspr   r7,SPRN_MAS1
+       andis.  r7,r7,MAS1_VALID@h
+       bne     match_TLB
+       mfspr   r7,SPRN_PID1
+       slwi    r7,r7,16
+       or      r7,r7,r4
+       mtspr   SPRN_MAS6,r7
+       tlbsx   0,r6                            /* search MSR[IS], SPID=PID1 */
+       mfspr   r7,SPRN_MAS1
+       andis.  r7,r7,MAS1_VALID@h
+       bne     match_TLB
+       mfspr   r7, SPRN_PID2
+       slwi    r7,r7,16
+       or      r7,r7,r4
+       mtspr   SPRN_MAS6,r7
+       tlbsx   0,r6                            /* Fall through, we had to match */
+#endif
+match_TLB:
+       mfspr   r7,SPRN_MAS0
+       rlwinm  r3,r7,16,20,31                  /* Extract MAS0(Entry) */
+
+       mfspr   r7,SPRN_MAS1                    /* Insure IPROT set */
+       oris    r7,r7,MAS1_IPROT@h
+       mtspr   SPRN_MAS1,r7
+       tlbwe
+
+/* 2. Invalidate all entries except the entry we're executing in */
+       mfspr   r9,SPRN_TLB1CFG
+       andi.   r9,r9,0xfff
+       li      r6,0                            /* Set Entry counter to 0 */
+1:     lis     r7,0x1000                       /* Set MAS0(TLBSEL) = 1 */
+       rlwimi  r7,r6,16,4,15                   /* Setup MAS0 = TLBSEL | ESEL(r6) */
+       mtspr   SPRN_MAS0,r7
+       tlbre
+       mfspr   r7,SPRN_MAS1
+       rlwinm  r7,r7,0,2,31                    /* Clear MAS1 Valid and IPROT */
+       cmpw    r3,r6
+       beq     skpinv                          /* Dont update the current execution TLB */
+       mtspr   SPRN_MAS1,r7
+       tlbwe
+       isync
+skpinv:        addi    r6,r6,1                         /* Increment */
+       cmpw    r6,r9                           /* Are we done? */
+       bne     1b                              /* If not, repeat */
+
+       /* Invalidate TLB0 */
+       li      r6,0x04
+       tlbivax 0,r6
+#ifdef CONFIG_SMP
+       tlbsync
+#endif
+       /* Invalidate TLB1 */
+       li      r6,0x0c
+       tlbivax 0,r6
+#ifdef CONFIG_SMP
+       tlbsync
+#endif
+       msync
+
+/* 3. Setup a temp mapping and jump to it */
+       andi.   r5, r3, 0x1     /* Find an entry not used and is non-zero */
+       addi    r5, r5, 0x1
+       lis     r7,0x1000       /* Set MAS0(TLBSEL) = 1 */
+       rlwimi  r7,r3,16,4,15   /* Setup MAS0 = TLBSEL | ESEL(r3) */
+       mtspr   SPRN_MAS0,r7
+       tlbre
+
+       /* Just modify the entry ID and EPN for the temp mapping */
+       lis     r7,0x1000       /* Set MAS0(TLBSEL) = 1 */
+       rlwimi  r7,r5,16,4,15   /* Setup MAS0 = TLBSEL | ESEL(r5) */
+       mtspr   SPRN_MAS0,r7
+       xori    r6,r4,1         /* Setup TMP mapping in the other Address space */
+       slwi    r6,r6,12
+       oris    r6,r6,(MAS1_VALID|MAS1_IPROT)@h
+       ori     r6,r6,(MAS1_TSIZE(BOOKE_PAGESZ_4K))@l
+       mtspr   SPRN_MAS1,r6
+       mfspr   r6,SPRN_MAS2
+       li      r7,0            /* temp EPN = 0 */
+       rlwimi  r7,r6,0,20,31
+       mtspr   SPRN_MAS2,r7
+       tlbwe
+
+       xori    r6,r4,1
+       slwi    r6,r6,5         /* setup new context with other address space */
+       bl      1f              /* Find our address */
+1:     mflr    r9
+       rlwimi  r7,r9,0,20,31
+       addi    r7,r7,24
+       mtspr   SPRN_SRR0,r7
+       mtspr   SPRN_SRR1,r6
+       rfi
+
+/* 4. Clear out PIDs & Search info */
+       li      r6,0
+       mtspr   SPRN_PID0,r6
+#ifndef CONFIG_E200
+       mtspr   SPRN_PID1,r6
+       mtspr   SPRN_PID2,r6
+#endif
+       mtspr   SPRN_MAS6,r6
+
+/* 5. Invalidate mapping we started in */
+       lis     r7,0x1000       /* Set MAS0(TLBSEL) = 1 */
+       rlwimi  r7,r3,16,4,15   /* Setup MAS0 = TLBSEL | ESEL(r3) */
+       mtspr   SPRN_MAS0,r7
+       tlbre
+       li      r6,0
+       mtspr   SPRN_MAS1,r6
+       tlbwe
+       /* Invalidate TLB1 */
+       li      r9,0x0c
+       tlbivax 0,r9
+#ifdef CONFIG_SMP
+       tlbsync
+#endif
+       msync
+
+/* 6. Setup KERNELBASE mapping in TLB1[0] */
+       lis     r6,0x1000               /* Set MAS0(TLBSEL) = TLB1(1), ESEL = 0 */
+       mtspr   SPRN_MAS0,r6
+       lis     r6,(MAS1_VALID|MAS1_IPROT)@h
+       ori     r6,r6,(MAS1_TSIZE(BOOKE_PAGESZ_16M))@l
+       mtspr   SPRN_MAS1,r6
+       li      r7,0
+       lis     r6,KERNELBASE@h
+       ori     r6,r6,KERNELBASE@l
+       rlwimi  r6,r7,0,20,31
+       mtspr   SPRN_MAS2,r6
+       li      r7,(MAS3_SX|MAS3_SW|MAS3_SR)
+       mtspr   SPRN_MAS3,r7
+       tlbwe
+
+/* 7. Jump to KERNELBASE mapping */
+       lis     r7,MSR_KERNEL@h
+       ori     r7,r7,MSR_KERNEL@l
+       bl      1f                      /* Find our address */
+1:     mflr    r9
+       rlwimi  r6,r9,0,20,31
+       addi    r6,r6,24
+       mtspr   SPRN_SRR0,r6
+       mtspr   SPRN_SRR1,r7
+       rfi                             /* start execution out of TLB1[0] entry */
+
+/* 8. Clear out the temp mapping */
+       lis     r7,0x1000       /* Set MAS0(TLBSEL) = 1 */
+       rlwimi  r7,r5,16,4,15   /* Setup MAS0 = TLBSEL | ESEL(r5) */
+       mtspr   SPRN_MAS0,r7
+       tlbre
+       mtspr   SPRN_MAS1,r8
+       tlbwe
+       /* Invalidate TLB1 */
+       li      r9,0x0c
+       tlbivax 0,r9
+#ifdef CONFIG_SMP
+       tlbsync
+#endif
+       msync
+
+       /* Establish the interrupt vector offsets */
+       SET_IVOR(0,  CriticalInput);
+       SET_IVOR(1,  MachineCheck);
+       SET_IVOR(2,  DataStorage);
+       SET_IVOR(3,  InstructionStorage);
+       SET_IVOR(4,  ExternalInput);
+       SET_IVOR(5,  Alignment);
+       SET_IVOR(6,  Program);
+       SET_IVOR(7,  FloatingPointUnavailable);
+       SET_IVOR(8,  SystemCall);
+       SET_IVOR(9,  AuxillaryProcessorUnavailable);
+       SET_IVOR(10, Decrementer);
+       SET_IVOR(11, FixedIntervalTimer);
+       SET_IVOR(12, WatchdogTimer);
+       SET_IVOR(13, DataTLBError);
+       SET_IVOR(14, InstructionTLBError);
+       SET_IVOR(15, Debug);
+       SET_IVOR(32, SPEUnavailable);
+       SET_IVOR(33, SPEFloatingPointData);
+       SET_IVOR(34, SPEFloatingPointRound);
+#ifndef CONFIG_E200
+       SET_IVOR(35, PerformanceMonitor);
+#endif
+
+       /* Establish the interrupt vector base */
+       lis     r4,interrupt_base@h     /* IVPR only uses the high 16-bits */
+       mtspr   SPRN_IVPR,r4
+
+       /* Setup the defaults for TLB entries */
+       li      r2,(MAS4_TSIZED(BOOKE_PAGESZ_4K))@l
+#ifdef CONFIG_E200
+       oris    r2,r2,MAS4_TLBSELD(1)@h
+#endif
+       mtspr   SPRN_MAS4, r2
+
+#if 0
+       /* Enable DOZE */
+       mfspr   r2,SPRN_HID0
+       oris    r2,r2,HID0_DOZE@h
+       mtspr   SPRN_HID0, r2
+#endif
+#ifdef CONFIG_E200
+       /* enable dedicated debug exception handling resources (Debug APU) */
+       mfspr   r2,SPRN_HID0
+       ori     r2,r2,HID0_DAPUEN@l
+       mtspr   SPRN_HID0,r2
+#endif
+
+#if !defined(CONFIG_BDI_SWITCH)
+       /*
+        * The Abatron BDI JTAG debugger does not tolerate others
+        * mucking with the debug registers.
+        */
+       lis     r2,DBCR0_IDM@h
+       mtspr   SPRN_DBCR0,r2
+       /* clear any residual debug events */
+       li      r2,-1
+       mtspr   SPRN_DBSR,r2
+#endif
+
+       /*
+        * This is where the main kernel code starts.
+        */
+
+       /* ptr to current */
+       lis     r2,init_task@h
+       ori     r2,r2,init_task@l
+
+       /* ptr to current thread */
+       addi    r4,r2,THREAD    /* init task's THREAD */
+       mtspr   SPRN_SPRG3,r4
+
+       /* stack */
+       lis     r1,init_thread_union@h
+       ori     r1,r1,init_thread_union@l
+       li      r0,0
+       stwu    r0,THREAD_SIZE-STACK_FRAME_OVERHEAD(r1)
+
+       bl      early_init
+
+       mfspr   r3,SPRN_TLB1CFG
+       andi.   r3,r3,0xfff
+       lis     r4,num_tlbcam_entries@ha
+       stw     r3,num_tlbcam_entries@l(r4)
+/*
+ * Decide what sort of machine this is and initialize the MMU.
+ */
+       mr      r3,r31
+       mr      r4,r30
+       mr      r5,r29
+       mr      r6,r28
+       mr      r7,r27
+       bl      machine_init
+       bl      MMU_init
+
+       /* Setup PTE pointers for the Abatron bdiGDB */
+       lis     r6, swapper_pg_dir@h
+       ori     r6, r6, swapper_pg_dir@l
+       lis     r5, abatron_pteptrs@h
+       ori     r5, r5, abatron_pteptrs@l
+       lis     r4, KERNELBASE@h
+       ori     r4, r4, KERNELBASE@l
+       stw     r5, 0(r4)       /* Save abatron_pteptrs at a fixed location */
+       stw     r6, 0(r5)
+
+       /* Let's move on */
+       lis     r4,start_kernel@h
+       ori     r4,r4,start_kernel@l
+       lis     r3,MSR_KERNEL@h
+       ori     r3,r3,MSR_KERNEL@l
+       mtspr   SPRN_SRR0,r4
+       mtspr   SPRN_SRR1,r3
+       rfi                     /* change context and jump to start_kernel */
+
+/* Macros to hide the PTE size differences
+ *
+ * FIND_PTE -- walks the page tables given EA & pgdir pointer
+ *   r10 -- EA of fault
+ *   r11 -- PGDIR pointer
+ *   r12 -- free
+ *   label 2: is the bailout case
+ *
+ * if we find the pte (fall through):
+ *   r11 is low pte word
+ *   r12 is pointer to the pte
+ */
+#ifdef CONFIG_PTE_64BIT
+#define PTE_FLAGS_OFFSET       4
+#define FIND_PTE       \
+       rlwinm  r12, r10, 13, 19, 29;   /* Compute pgdir/pmd offset */  \
+       lwzx    r11, r12, r11;          /* Get pgd/pmd entry */         \
+       rlwinm. r12, r11, 0, 0, 20;     /* Extract pt base address */   \
+       beq     2f;                     /* Bail if no table */          \
+       rlwimi  r12, r10, 23, 20, 28;   /* Compute pte address */       \
+       lwz     r11, 4(r12);            /* Get pte entry */
+#else
+#define PTE_FLAGS_OFFSET       0
+#define FIND_PTE       \
+       rlwimi  r11, r10, 12, 20, 29;   /* Create L1 (pgdir/pmd) address */     \
+       lwz     r11, 0(r11);            /* Get L1 entry */                      \
+       rlwinm. r12, r11, 0, 0, 19;     /* Extract L2 (pte) base address */     \
+       beq     2f;                     /* Bail if no table */                  \
+       rlwimi  r12, r10, 22, 20, 29;   /* Compute PTE address */               \
+       lwz     r11, 0(r12);            /* Get Linux PTE */
+#endif
+
+/*
+ * Interrupt vector entry code
+ *
+ * The Book E MMUs are always on so we don't need to handle
+ * interrupts in real mode as with previous PPC processors. In
+ * this case we handle interrupts in the kernel virtual address
+ * space.
+ *
+ * Interrupt vectors are dynamically placed relative to the
+ * interrupt prefix as determined by the address of interrupt_base.
+ * The interrupt vectors offsets are programmed using the labels
+ * for each interrupt vector entry.
+ *
+ * Interrupt vectors must be aligned on a 16 byte boundary.
+ * We align on a 32 byte cache line boundary for good measure.
+ */
+
+interrupt_base:
+       /* Critical Input Interrupt */
+       CRITICAL_EXCEPTION(0x0100, CriticalInput, UnknownException)
+
+       /* Machine Check Interrupt */
+#ifdef CONFIG_E200
+       /* no RFMCI, MCSRRs on E200 */
+       CRITICAL_EXCEPTION(0x0200, MachineCheck, MachineCheckException)
+#else
+       MCHECK_EXCEPTION(0x0200, MachineCheck, MachineCheckException)
+#endif
+
+       /* Data Storage Interrupt */
+       START_EXCEPTION(DataStorage)
+       mtspr   SPRN_SPRG0, r10         /* Save some working registers */
+       mtspr   SPRN_SPRG1, r11
+       mtspr   SPRN_SPRG4W, r12
+       mtspr   SPRN_SPRG5W, r13
+       mfcr    r11
+       mtspr   SPRN_SPRG7W, r11
+
+       /*
+        * Check if it was a store fault, if not then bail
+        * because a user tried to access a kernel or
+        * read-protected page.  Otherwise, get the
+        * offending address and handle it.
+        */
+       mfspr   r10, SPRN_ESR
+       andis.  r10, r10, ESR_ST@h
+       beq     2f
+
+       mfspr   r10, SPRN_DEAR          /* Get faulting address */
+
+       /* If we are faulting a kernel address, we have to use the
+        * kernel page tables.
+        */
+       lis     r11, TASK_SIZE@h
+       ori     r11, r11, TASK_SIZE@l
+       cmplw   0, r10, r11
+       bge     2f
+
+       /* Get the PGD for the current thread */
+3:
+       mfspr   r11,SPRN_SPRG3
+       lwz     r11,PGDIR(r11)
+4:
+       FIND_PTE
+
+       /* Are _PAGE_USER & _PAGE_RW set & _PAGE_HWWRITE not? */
+       andi.   r13, r11, _PAGE_RW|_PAGE_USER|_PAGE_HWWRITE
+       cmpwi   0, r13, _PAGE_RW|_PAGE_USER
+       bne     2f                      /* Bail if not */
+
+       /* Update 'changed'. */
+       ori     r11, r11, _PAGE_DIRTY|_PAGE_ACCESSED|_PAGE_HWWRITE
+       stw     r11, PTE_FLAGS_OFFSET(r12) /* Update Linux page table */
+
+       /* MAS2 not updated as the entry does exist in the tlb, this
+          fault taken to detect state transition (eg: COW -> DIRTY)
+        */
+       andi.   r11, r11, _PAGE_HWEXEC
+       rlwimi  r11, r11, 31, 27, 27    /* SX <- _PAGE_HWEXEC */
+       ori     r11, r11, (MAS3_UW|MAS3_SW|MAS3_UR|MAS3_SR)@l /* set static perms */
+
+       /* update search PID in MAS6, AS = 0 */
+       mfspr   r12, SPRN_PID0
+       slwi    r12, r12, 16
+       mtspr   SPRN_MAS6, r12
+
+       /* find the TLB index that caused the fault.  It has to be here. */
+       tlbsx   0, r10
+
+       /* only update the perm bits, assume the RPN is fine */
+       mfspr   r12, SPRN_MAS3
+       rlwimi  r12, r11, 0, 20, 31
+       mtspr   SPRN_MAS3,r12
+       tlbwe
+
+       /* Done...restore registers and get out of here.  */
+       mfspr   r11, SPRN_SPRG7R
+       mtcr    r11
+       mfspr   r13, SPRN_SPRG5R
+       mfspr   r12, SPRN_SPRG4R
+       mfspr   r11, SPRN_SPRG1
+       mfspr   r10, SPRN_SPRG0
+       rfi                     /* Force context change */
+
+2:
+       /*
+        * The bailout.  Restore registers to pre-exception conditions
+        * and call the heavyweights to help us out.
+        */
+       mfspr   r11, SPRN_SPRG7R
+       mtcr    r11
+       mfspr   r13, SPRN_SPRG5R
+       mfspr   r12, SPRN_SPRG4R
+       mfspr   r11, SPRN_SPRG1
+       mfspr   r10, SPRN_SPRG0
+       b       data_access
+
+       /* Instruction Storage Interrupt */
+       INSTRUCTION_STORAGE_EXCEPTION
+
+       /* External Input Interrupt */
+       EXCEPTION(0x0500, ExternalInput, do_IRQ, EXC_XFER_LITE)
+
+       /* Alignment Interrupt */
+       ALIGNMENT_EXCEPTION
+
+       /* Program Interrupt */
+       PROGRAM_EXCEPTION
+
+       /* Floating Point Unavailable Interrupt */
+#ifdef CONFIG_PPC_FPU
+       FP_UNAVAILABLE_EXCEPTION
+#else
+#ifdef CONFIG_E200
+       /* E200 treats 'normal' floating point instructions as FP Unavail exception */
+       EXCEPTION(0x0800, FloatingPointUnavailable, ProgramCheckException, EXC_XFER_EE)
+#else
+       EXCEPTION(0x0800, FloatingPointUnavailable, UnknownException, EXC_XFER_EE)
+#endif
+#endif
+
+       /* System Call Interrupt */
+       START_EXCEPTION(SystemCall)
+       NORMAL_EXCEPTION_PROLOG
+       EXC_XFER_EE_LITE(0x0c00, DoSyscall)
+
+       /* Auxillary Processor Unavailable Interrupt */
+       EXCEPTION(0x2900, AuxillaryProcessorUnavailable, UnknownException, EXC_XFER_EE)
+
+       /* Decrementer Interrupt */
+       DECREMENTER_EXCEPTION
+
+       /* Fixed Internal Timer Interrupt */
+       /* TODO: Add FIT support */
+       EXCEPTION(0x3100, FixedIntervalTimer, UnknownException, EXC_XFER_EE)
+
+       /* Watchdog Timer Interrupt */
+#ifdef CONFIG_BOOKE_WDT
+       CRITICAL_EXCEPTION(0x3200, WatchdogTimer, WatchdogException)
+#else
+       CRITICAL_EXCEPTION(0x3200, WatchdogTimer, UnknownException)
+#endif
+
+       /* Data TLB Error Interrupt */
+       START_EXCEPTION(DataTLBError)
+       mtspr   SPRN_SPRG0, r10         /* Save some working registers */
+       mtspr   SPRN_SPRG1, r11
+       mtspr   SPRN_SPRG4W, r12
+       mtspr   SPRN_SPRG5W, r13
+       mfcr    r11
+       mtspr   SPRN_SPRG7W, r11
+       mfspr   r10, SPRN_DEAR          /* Get faulting address */
+
+       /* If we are faulting a kernel address, we have to use the
+        * kernel page tables.
+        */
+       lis     r11, TASK_SIZE@h
+       ori     r11, r11, TASK_SIZE@l
+       cmplw   5, r10, r11
+       blt     5, 3f
+       lis     r11, swapper_pg_dir@h
+       ori     r11, r11, swapper_pg_dir@l
+
+       mfspr   r12,SPRN_MAS1           /* Set TID to 0 */
+       rlwinm  r12,r12,0,16,1
+       mtspr   SPRN_MAS1,r12
+
+       b       4f
+
+       /* Get the PGD for the current thread */
+3:
+       mfspr   r11,SPRN_SPRG3
+       lwz     r11,PGDIR(r11)
+
+4:
+       FIND_PTE
+       andi.   r13, r11, _PAGE_PRESENT /* Is the page present? */
+       beq     2f                      /* Bail if not present */
+
+#ifdef CONFIG_PTE_64BIT
+       lwz     r13, 0(r12)
+#endif
+       ori     r11, r11, _PAGE_ACCESSED
+       stw     r11, PTE_FLAGS_OFFSET(r12)
+
+        /* Jump to common tlb load */
+       b       finish_tlb_load
+2:
+       /* The bailout.  Restore registers to pre-exception conditions
+        * and call the heavyweights to help us out.
+        */
+       mfspr   r11, SPRN_SPRG7R
+       mtcr    r11
+       mfspr   r13, SPRN_SPRG5R
+       mfspr   r12, SPRN_SPRG4R
+       mfspr   r11, SPRN_SPRG1
+       mfspr   r10, SPRN_SPRG0
+       b       data_access
+
+       /* Instruction TLB Error Interrupt */
+       /*
+        * Nearly the same as above, except we get our
+        * information from different registers and bailout
+        * to a different point.
+        */
+       START_EXCEPTION(InstructionTLBError)
+       mtspr   SPRN_SPRG0, r10         /* Save some working registers */
+       mtspr   SPRN_SPRG1, r11
+       mtspr   SPRN_SPRG4W, r12
+       mtspr   SPRN_SPRG5W, r13
+       mfcr    r11
+       mtspr   SPRN_SPRG7W, r11
+       mfspr   r10, SPRN_SRR0          /* Get faulting address */
+
+       /* If we are faulting a kernel address, we have to use the
+        * kernel page tables.
+        */
+       lis     r11, TASK_SIZE@h
+       ori     r11, r11, TASK_SIZE@l
+       cmplw   5, r10, r11
+       blt     5, 3f
+       lis     r11, swapper_pg_dir@h
+       ori     r11, r11, swapper_pg_dir@l
+
+       mfspr   r12,SPRN_MAS1           /* Set TID to 0 */
+       rlwinm  r12,r12,0,16,1
+       mtspr   SPRN_MAS1,r12
+
+       b       4f
+
+       /* Get the PGD for the current thread */
+3:
+       mfspr   r11,SPRN_SPRG3
+       lwz     r11,PGDIR(r11)
+
+4:
+       FIND_PTE
+       andi.   r13, r11, _PAGE_PRESENT /* Is the page present? */
+       beq     2f                      /* Bail if not present */
+
+#ifdef CONFIG_PTE_64BIT
+       lwz     r13, 0(r12)
+#endif
+       ori     r11, r11, _PAGE_ACCESSED
+       stw     r11, PTE_FLAGS_OFFSET(r12)
+
+       /* Jump to common TLB load point */
+       b       finish_tlb_load
+
+2:
+       /* The bailout.  Restore registers to pre-exception conditions
+        * and call the heavyweights to help us out.
+        */
+       mfspr   r11, SPRN_SPRG7R
+       mtcr    r11
+       mfspr   r13, SPRN_SPRG5R
+       mfspr   r12, SPRN_SPRG4R
+       mfspr   r11, SPRN_SPRG1
+       mfspr   r10, SPRN_SPRG0
+       b       InstructionStorage
+
+#ifdef CONFIG_SPE
+       /* SPE Unavailable */
+       START_EXCEPTION(SPEUnavailable)
+       NORMAL_EXCEPTION_PROLOG
+       bne     load_up_spe
+       addi    r3,r1,STACK_FRAME_OVERHEAD
+       EXC_XFER_EE_LITE(0x2010, KernelSPE)
+#else
+       EXCEPTION(0x2020, SPEUnavailable, UnknownException, EXC_XFER_EE)
+#endif /* CONFIG_SPE */
+
+       /* SPE Floating Point Data */
+#ifdef CONFIG_SPE
+       EXCEPTION(0x2030, SPEFloatingPointData, SPEFloatingPointException, EXC_XFER_EE);
+#else
+       EXCEPTION(0x2040, SPEFloatingPointData, UnknownException, EXC_XFER_EE)
+#endif /* CONFIG_SPE */
+
+       /* SPE Floating Point Round */
+       EXCEPTION(0x2050, SPEFloatingPointRound, UnknownException, EXC_XFER_EE)
+
+       /* Performance Monitor */
+       EXCEPTION(0x2060, PerformanceMonitor, PerformanceMonitorException, EXC_XFER_STD)
+
+
+       /* Debug Interrupt */
+       DEBUG_EXCEPTION
+
+/*
+ * Local functions
+ */
+
+       /*
+        * Data TLB exceptions will bail out to this point
+        * if they can't resolve the lightweight TLB fault.
+        */
+data_access:
+       NORMAL_EXCEPTION_PROLOG
+       mfspr   r5,SPRN_ESR             /* Grab the ESR, save it, pass arg3 */
+       stw     r5,_ESR(r11)
+       mfspr   r4,SPRN_DEAR            /* Grab the DEAR, save it, pass arg2 */
+       andis.  r10,r5,(ESR_ILK|ESR_DLK)@h
+       bne     1f
+       EXC_XFER_EE_LITE(0x0300, handle_page_fault)
+1:
+       addi    r3,r1,STACK_FRAME_OVERHEAD
+       EXC_XFER_EE_LITE(0x0300, CacheLockingException)
+
+/*
+
+ * Both the instruction and data TLB miss get to this
+ * point to load the TLB.
+ *     r10 - EA of fault
+ *     r11 - TLB (info from Linux PTE)
+ *     r12, r13 - available to use
+ *     CR5 - results of addr < TASK_SIZE
+ *     MAS0, MAS1 - loaded with proper value when we get here
+ *     MAS2, MAS3 - will need additional info from Linux PTE
+ *     Upon exit, we reload everything and RFI.
+ */
+finish_tlb_load:
+       /*
+        * We set execute, because we don't have the granularity to
+        * properly set this at the page level (Linux problem).
+        * Many of these bits are software only.  Bits we don't set
+        * here we (properly should) assume have the appropriate value.
+        */
+
+       mfspr   r12, SPRN_MAS2
+#ifdef CONFIG_PTE_64BIT
+       rlwimi  r12, r11, 26, 24, 31    /* extract ...WIMGE from pte */
+#else
+       rlwimi  r12, r11, 26, 27, 31    /* extract WIMGE from pte */
+#endif
+       mtspr   SPRN_MAS2, r12
+
+       bge     5, 1f
+
+       /* is user addr */
+       andi.   r12, r11, (_PAGE_USER | _PAGE_HWWRITE | _PAGE_HWEXEC)
+       andi.   r10, r11, _PAGE_USER    /* Test for _PAGE_USER */
+       srwi    r10, r12, 1
+       or      r12, r12, r10   /* Copy user perms into supervisor */
+       iseleq  r12, 0, r12
+       b       2f
+
+       /* is kernel addr */
+1:     rlwinm  r12, r11, 31, 29, 29    /* Extract _PAGE_HWWRITE into SW */
+       ori     r12, r12, (MAS3_SX | MAS3_SR)
+
+#ifdef CONFIG_PTE_64BIT
+2:     rlwimi  r12, r13, 24, 0, 7      /* grab RPN[32:39] */
+       rlwimi  r12, r11, 24, 8, 19     /* grab RPN[40:51] */
+       mtspr   SPRN_MAS3, r12
+BEGIN_FTR_SECTION
+       srwi    r10, r13, 8             /* grab RPN[8:31] */
+       mtspr   SPRN_MAS7, r10
+END_FTR_SECTION_IFSET(CPU_FTR_BIG_PHYS)
+#else
+2:     rlwimi  r11, r12, 0, 20, 31     /* Extract RPN from PTE and merge with perms */
+       mtspr   SPRN_MAS3, r11
+#endif
+#ifdef CONFIG_E200
+       /* Round robin TLB1 entries assignment */
+       mfspr   r12, SPRN_MAS0
+
+       /* Extract TLB1CFG(NENTRY) */
+       mfspr   r11, SPRN_TLB1CFG
+       andi.   r11, r11, 0xfff
+
+       /* Extract MAS0(NV) */
+       andi.   r13, r12, 0xfff
+       addi    r13, r13, 1
+       cmpw    0, r13, r11
+       addi    r12, r12, 1
+
+       /* check if we need to wrap */
+       blt     7f
+
+       /* wrap back to first free tlbcam entry */
+       lis     r13, tlbcam_index@ha
+       lwz     r13, tlbcam_index@l(r13)
+       rlwimi  r12, r13, 0, 20, 31
+7:
+       mtspr   SPRN_MAS0,r12
+#endif /* CONFIG_E200 */
+
+       tlbwe
+
+       /* Done...restore registers and get out of here.  */
+       mfspr   r11, SPRN_SPRG7R
+       mtcr    r11
+       mfspr   r13, SPRN_SPRG5R
+       mfspr   r12, SPRN_SPRG4R
+       mfspr   r11, SPRN_SPRG1
+       mfspr   r10, SPRN_SPRG0
+       rfi                                     /* Force context change */
+
+#ifdef CONFIG_SPE
+/* Note that the SPE support is closely modeled after the AltiVec
+ * support.  Changes to one are likely to be applicable to the
+ * other!  */
+load_up_spe:
+/*
+ * Disable SPE for the task which had SPE previously,
+ * and save its SPE registers in its thread_struct.
+ * Enables SPE for use in the kernel on return.
+ * On SMP we know the SPE units are free, since we give it up every
+ * switch.  -- Kumar
+ */
+       mfmsr   r5
+       oris    r5,r5,MSR_SPE@h
+       mtmsr   r5                      /* enable use of SPE now */
+       isync
+/*
+ * For SMP, we don't do lazy SPE switching because it just gets too
+ * horrendously complex, especially when a task switches from one CPU
+ * to another.  Instead we call giveup_spe in switch_to.
+ */
+#ifndef CONFIG_SMP
+       lis     r3,last_task_used_spe@ha
+       lwz     r4,last_task_used_spe@l(r3)
+       cmpi    0,r4,0
+       beq     1f
+       addi    r4,r4,THREAD    /* want THREAD of last_task_used_spe */
+       SAVE_32EVRS(0,r10,r4)
+       evxor   evr10, evr10, evr10     /* clear out evr10 */
+       evmwumiaa evr10, evr10, evr10   /* evr10 <- ACC = 0 * 0 + ACC */
+       li      r5,THREAD_ACC
+       evstddx evr10, r4, r5           /* save off accumulator */
+       lwz     r5,PT_REGS(r4)
+       lwz     r4,_MSR-STACK_FRAME_OVERHEAD(r5)
+       lis     r10,MSR_SPE@h
+       andc    r4,r4,r10       /* disable SPE for previous task */
+       stw     r4,_MSR-STACK_FRAME_OVERHEAD(r5)
+1:
+#endif /* CONFIG_SMP */
+       /* enable use of SPE after return */
+       oris    r9,r9,MSR_SPE@h
+       mfspr   r5,SPRN_SPRG3           /* current task's THREAD (phys) */
+       li      r4,1
+       li      r10,THREAD_ACC
+       stw     r4,THREAD_USED_SPE(r5)
+       evlddx  evr4,r10,r5
+       evmra   evr4,evr4
+       REST_32EVRS(0,r10,r5)
+#ifndef CONFIG_SMP
+       subi    r4,r5,THREAD
+       stw     r4,last_task_used_spe@l(r3)
+#endif /* CONFIG_SMP */
+       /* restore registers and return */
+2:     REST_4GPRS(3, r11)
+       lwz     r10,_CCR(r11)
+       REST_GPR(1, r11)
+       mtcr    r10
+       lwz     r10,_LINK(r11)
+       mtlr    r10
+       REST_GPR(10, r11)
+       mtspr   SPRN_SRR1,r9
+       mtspr   SPRN_SRR0,r12
+       REST_GPR(9, r11)
+       REST_GPR(12, r11)
+       lwz     r11,GPR11(r11)
+       SYNC
+       rfi
+
+/*
+ * SPE unavailable trap from kernel - print a message, but let
+ * the task use SPE in the kernel until it returns to user mode.
+ */
+KernelSPE:
+       lwz     r3,_MSR(r1)
+       oris    r3,r3,MSR_SPE@h
+       stw     r3,_MSR(r1)     /* enable use of SPE after return */
+       lis     r3,87f@h
+       ori     r3,r3,87f@l
+       mr      r4,r2           /* current */
+       lwz     r5,_NIP(r1)
+       bl      printk
+       b       ret_from_except
+87:    .string "SPE used in kernel  (task=%p, pc=%x)  \n"
+       .align  4,0
+
+#endif /* CONFIG_SPE */
+
+/*
+ * Global functions
+ */
+
+/*
+ * extern void loadcam_entry(unsigned int index)
+ *
+ * Load TLBCAM[index] entry in to the L2 CAM MMU
+ */
+_GLOBAL(loadcam_entry)
+       lis     r4,TLBCAM@ha
+       addi    r4,r4,TLBCAM@l
+       mulli   r5,r3,20
+       add     r3,r5,r4
+       lwz     r4,0(r3)
+       mtspr   SPRN_MAS0,r4
+       lwz     r4,4(r3)
+       mtspr   SPRN_MAS1,r4
+       lwz     r4,8(r3)
+       mtspr   SPRN_MAS2,r4
+       lwz     r4,12(r3)
+       mtspr   SPRN_MAS3,r4
+       tlbwe
+       isync
+       blr
+
+/*
+ * extern void giveup_altivec(struct task_struct *prev)
+ *
+ * The e500 core does not have an AltiVec unit.
+ */
+_GLOBAL(giveup_altivec)
+       blr
+
+#ifdef CONFIG_SPE
+/*
+ * extern void giveup_spe(struct task_struct *prev)
+ *
+ */
+_GLOBAL(giveup_spe)
+       mfmsr   r5
+       oris    r5,r5,MSR_SPE@h
+       SYNC
+       mtmsr   r5                      /* enable use of SPE now */
+       isync
+       cmpi    0,r3,0
+       beqlr-                          /* if no previous owner, done */
+       addi    r3,r3,THREAD            /* want THREAD of task */
+       lwz     r5,PT_REGS(r3)
+       cmpi    0,r5,0
+       SAVE_32EVRS(0, r4, r3)
+       evxor   evr6, evr6, evr6        /* clear out evr6 */
+       evmwumiaa evr6, evr6, evr6      /* evr6 <- ACC = 0 * 0 + ACC */
+       li      r4,THREAD_ACC
+       evstddx evr6, r4, r3            /* save off accumulator */
+       mfspr   r6,SPRN_SPEFSCR
+       stw     r6,THREAD_SPEFSCR(r3)   /* save spefscr register value */
+       beq     1f
+       lwz     r4,_MSR-STACK_FRAME_OVERHEAD(r5)
+       lis     r3,MSR_SPE@h
+       andc    r4,r4,r3                /* disable SPE for previous task */
+       stw     r4,_MSR-STACK_FRAME_OVERHEAD(r5)
+1:
+#ifndef CONFIG_SMP
+       li      r5,0
+       lis     r4,last_task_used_spe@ha
+       stw     r5,last_task_used_spe@l(r4)
+#endif /* CONFIG_SMP */
+       blr
+#endif /* CONFIG_SPE */
+
+/*
+ * extern void giveup_fpu(struct task_struct *prev)
+ *
+ * Not all FSL Book-E cores have an FPU
+ */
+#ifndef CONFIG_PPC_FPU
+_GLOBAL(giveup_fpu)
+       blr
+#endif
+
+/*
+ * extern void abort(void)
+ *
+ * At present, this routine just applies a system reset.
+ */
+_GLOBAL(abort)
+       li      r13,0
+        mtspr   SPRN_DBCR0,r13         /* disable all debug events */
+       mfmsr   r13
+       ori     r13,r13,MSR_DE@l        /* Enable Debug Events */
+       mtmsr   r13
+        mfspr   r13,SPRN_DBCR0
+        lis    r13,(DBCR0_IDM|DBCR0_RST_CHIP)@h
+        mtspr   SPRN_DBCR0,r13
+
+_GLOBAL(set_context)
+
+#ifdef CONFIG_BDI_SWITCH
+       /* Context switch the PTE pointer for the Abatron BDI2000.
+        * The PGDIR is the second parameter.
+        */
+       lis     r5, abatron_pteptrs@h
+       ori     r5, r5, abatron_pteptrs@l
+       stw     r4, 0x4(r5)
+#endif
+       mtspr   SPRN_PID,r3
+       isync                   /* Force context change */
+       blr
+
+/*
+ * We put a few things here that have to be page-aligned. This stuff
+ * goes at the beginning of the data segment, which is page-aligned.
+ */
+       .data
+_GLOBAL(sdata)
+_GLOBAL(empty_zero_page)
+       .space  4096
+_GLOBAL(swapper_pg_dir)
+       .space  4096
+
+/* Reserved 4k for the critical exception stack & 4k for the machine
+ * check stack per CPU for kernel mode exceptions */
+       .section .bss
+        .align 12
+exception_stack_bottom:
+       .space  BOOKE_EXCEPTION_STACK_SIZE * NR_CPUS
+_GLOBAL(exception_stack_top)
+
+/*
+ * This space gets a copy of optional info passed to us by the bootstrap
+ * which is used to pass parameters into the kernel like root=/dev/sda1, etc.
+ */
+_GLOBAL(cmd_line)
+       .space  512
+
+/*
+ * Room for two PTE pointers, usually the kernel and current user pointers
+ * to their respective root page table.
+ */
+abatron_pteptrs:
+       .space  8
+
diff --git a/arch/powerpc/kernel/idle_6xx.S b/arch/powerpc/kernel/idle_6xx.S
new file mode 100644 (file)
index 0000000..1a2194c
--- /dev/null
@@ -0,0 +1,233 @@
+/*
+ *  This file contains the power_save function for 6xx & 7xxx CPUs
+ *  rewritten in assembler
+ *
+ *  Warning ! This code assumes that if your machine has a 750fx
+ *  it will have PLL 1 set to low speed mode (used during NAP/DOZE).
+ *  if this is not the case some additional changes will have to
+ *  be done to check a runtime var (a bit like powersave-nap)
+ *
+ *  This program is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU General Public License
+ *  as published by the Free Software Foundation; either version
+ *  2 of the License, or (at your option) any later version.
+ */
+
+#include <linux/config.h>
+#include <linux/threads.h>
+#include <asm/processor.h>
+#include <asm/page.h>
+#include <asm/cputable.h>
+#include <asm/thread_info.h>
+#include <asm/ppc_asm.h>
+#include <asm/asm-offsets.h>
+
+#undef DEBUG
+
+       .text
+
+/*
+ * Init idle, called at early CPU setup time from head.S for each CPU
+ * Make sure no rest of NAP mode remains in HID0, save default
+ * values for some CPU specific registers. Called with r24
+ * containing CPU number and r3 reloc offset
+ */
+_GLOBAL(init_idle_6xx)
+BEGIN_FTR_SECTION
+       mfspr   r4,SPRN_HID0
+       rlwinm  r4,r4,0,10,8    /* Clear NAP */
+       mtspr   SPRN_HID0, r4
+       b       1f
+END_FTR_SECTION_IFSET(CPU_FTR_CAN_NAP)
+       blr
+1:
+       slwi    r5,r24,2
+       add     r5,r5,r3
+BEGIN_FTR_SECTION
+       mfspr   r4,SPRN_MSSCR0
+       addis   r6,r5, nap_save_msscr0@ha
+       stw     r4,nap_save_msscr0@l(r6)
+END_FTR_SECTION_IFSET(CPU_FTR_NAP_DISABLE_L2_PR)
+BEGIN_FTR_SECTION
+       mfspr   r4,SPRN_HID1
+       addis   r6,r5,nap_save_hid1@ha
+       stw     r4,nap_save_hid1@l(r6)
+END_FTR_SECTION_IFSET(CPU_FTR_DUAL_PLL_750FX)
+       blr
+
+/*
+ * Here is the power_save_6xx function. This could eventually be
+ * split into several functions & changing the function pointer
+ * depending on the various features.
+ */
+_GLOBAL(ppc6xx_idle)
+       /* Check if we can nap or doze, put HID0 mask in r3
+        */
+       lis     r3, 0
+BEGIN_FTR_SECTION
+       lis     r3,HID0_DOZE@h
+END_FTR_SECTION_IFSET(CPU_FTR_CAN_DOZE)
+BEGIN_FTR_SECTION
+       /* We must dynamically check for the NAP feature as it
+        * can be cleared by CPU init after the fixups are done
+        */
+       lis     r4,cur_cpu_spec@ha
+       lwz     r4,cur_cpu_spec@l(r4)
+       lwz     r4,CPU_SPEC_FEATURES(r4)
+       andi.   r0,r4,CPU_FTR_CAN_NAP
+       beq     1f
+       /* Now check if user or arch enabled NAP mode */
+       lis     r4,powersave_nap@ha
+       lwz     r4,powersave_nap@l(r4)
+       cmpwi   0,r4,0
+       beq     1f
+       lis     r3,HID0_NAP@h
+1:     
+END_FTR_SECTION_IFSET(CPU_FTR_CAN_NAP)
+       cmpwi   0,r3,0
+       beqlr
+
+       /* Clear MSR:EE */
+       mfmsr   r7
+       rlwinm  r0,r7,0,17,15
+       mtmsr   r0
+
+       /* Check current_thread_info()->flags */
+       rlwinm  r4,r1,0,0,18
+       lwz     r4,TI_FLAGS(r4)
+       andi.   r0,r4,_TIF_NEED_RESCHED
+       beq     1f
+       mtmsr   r7      /* out of line this ? */
+       blr
+1:     
+       /* Some pre-nap cleanups needed on some CPUs */
+       andis.  r0,r3,HID0_NAP@h
+       beq     2f
+BEGIN_FTR_SECTION
+       /* Disable L2 prefetch on some 745x and try to ensure
+        * L2 prefetch engines are idle. As explained by errata
+        * text, we can't be sure they are, we just hope very hard
+        * that well be enough (sic !). At least I noticed Apple
+        * doesn't even bother doing the dcbf's here...
+        */
+       mfspr   r4,SPRN_MSSCR0
+       rlwinm  r4,r4,0,0,29
+       sync
+       mtspr   SPRN_MSSCR0,r4
+       sync
+       isync
+       lis     r4,KERNELBASE@h
+       dcbf    0,r4
+       dcbf    0,r4
+       dcbf    0,r4
+       dcbf    0,r4
+END_FTR_SECTION_IFSET(CPU_FTR_NAP_DISABLE_L2_PR)
+#ifdef DEBUG
+       lis     r6,nap_enter_count@ha
+       lwz     r4,nap_enter_count@l(r6)
+       addi    r4,r4,1
+       stw     r4,nap_enter_count@l(r6)
+#endif 
+2:
+BEGIN_FTR_SECTION
+       /* Go to low speed mode on some 750FX */
+       lis     r4,powersave_lowspeed@ha
+       lwz     r4,powersave_lowspeed@l(r4)
+       cmpwi   0,r4,0
+       beq     1f
+       mfspr   r4,SPRN_HID1
+       oris    r4,r4,0x0001
+       mtspr   SPRN_HID1,r4
+1:     
+END_FTR_SECTION_IFSET(CPU_FTR_DUAL_PLL_750FX)
+
+       /* Go to NAP or DOZE now */     
+       mfspr   r4,SPRN_HID0
+       lis     r5,(HID0_NAP|HID0_SLEEP)@h
+BEGIN_FTR_SECTION
+       oris    r5,r5,HID0_DOZE@h
+END_FTR_SECTION_IFSET(CPU_FTR_CAN_DOZE)
+       andc    r4,r4,r5
+       or      r4,r4,r3
+BEGIN_FTR_SECTION
+       oris    r4,r4,HID0_DPM@h        /* that should be done once for all  */
+END_FTR_SECTION_IFCLR(CPU_FTR_NO_DPM)
+       mtspr   SPRN_HID0,r4
+BEGIN_FTR_SECTION
+       DSSALL
+       sync
+END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
+       ori     r7,r7,MSR_EE /* Could be ommited (already set) */
+       oris    r7,r7,MSR_POW@h
+       sync
+       isync
+       mtmsr   r7
+       isync
+       sync
+       blr
+       
+/*
+ * Return from NAP/DOZE mode, restore some CPU specific registers,
+ * we are called with DR/IR still off and r2 containing physical
+ * address of current.
+ */
+_GLOBAL(power_save_6xx_restore)
+       mfspr   r11,SPRN_HID0
+       rlwinm. r11,r11,0,10,8  /* Clear NAP & copy NAP bit !state to cr1 EQ */
+       cror    4*cr1+eq,4*cr0+eq,4*cr0+eq
+BEGIN_FTR_SECTION
+       rlwinm  r11,r11,0,9,7   /* Clear DOZE */
+END_FTR_SECTION_IFSET(CPU_FTR_CAN_DOZE)
+       mtspr   SPRN_HID0, r11
+
+#ifdef DEBUG
+       beq     cr1,1f
+       lis     r11,(nap_return_count-KERNELBASE)@ha
+       lwz     r9,nap_return_count@l(r11)
+       addi    r9,r9,1
+       stw     r9,nap_return_count@l(r11)
+1:
+#endif
+       
+       rlwinm  r9,r1,0,0,18
+       tophys(r9,r9)
+       lwz     r11,TI_CPU(r9)
+       slwi    r11,r11,2
+       /* Todo make sure all these are in the same page
+        * and load r22 (@ha part + CPU offset) only once
+        */
+BEGIN_FTR_SECTION
+       beq     cr1,1f
+       addis   r9,r11,(nap_save_msscr0-KERNELBASE)@ha
+       lwz     r9,nap_save_msscr0@l(r9)
+       mtspr   SPRN_MSSCR0, r9
+       sync
+       isync
+1:
+END_FTR_SECTION_IFSET(CPU_FTR_NAP_DISABLE_L2_PR)
+BEGIN_FTR_SECTION
+       addis   r9,r11,(nap_save_hid1-KERNELBASE)@ha
+       lwz     r9,nap_save_hid1@l(r9)
+       mtspr   SPRN_HID1, r9
+END_FTR_SECTION_IFSET(CPU_FTR_DUAL_PLL_750FX)
+       b       transfer_to_handler_cont
+
+       .data
+
+_GLOBAL(nap_save_msscr0)
+       .space  4*NR_CPUS
+
+_GLOBAL(nap_save_hid1)
+       .space  4*NR_CPUS
+
+_GLOBAL(powersave_nap)
+       .long   0
+_GLOBAL(powersave_lowspeed)
+       .long   0
+
+#ifdef DEBUG
+_GLOBAL(nap_enter_count)
+       .space  4
+_GLOBAL(nap_return_count)
+       .space  4
+#endif
diff --git a/arch/powerpc/kernel/process.c b/arch/powerpc/kernel/process.c
new file mode 100644 (file)
index 0000000..f5a9d2a
--- /dev/null
@@ -0,0 +1,724 @@
+/*
+ *  arch/ppc/kernel/process.c
+ *
+ *  Derived from "arch/i386/kernel/process.c"
+ *    Copyright (C) 1995  Linus Torvalds
+ *
+ *  Updated and modified by Cort Dougan (cort@cs.nmt.edu) and
+ *  Paul Mackerras (paulus@cs.anu.edu.au)
+ *
+ *  PowerPC version
+ *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
+ *
+ *  This program is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU General Public License
+ *  as published by the Free Software Foundation; either version
+ *  2 of the License, or (at your option) any later version.
+ */
+
+#include <linux/config.h>
+#include <linux/errno.h>
+#include <linux/sched.h>
+#include <linux/kernel.h>
+#include <linux/mm.h>
+#include <linux/smp.h>
+#include <linux/smp_lock.h>
+#include <linux/stddef.h>
+#include <linux/unistd.h>
+#include <linux/ptrace.h>
+#include <linux/slab.h>
+#include <linux/user.h>
+#include <linux/elf.h>
+#include <linux/init.h>
+#include <linux/prctl.h>
+#include <linux/init_task.h>
+#include <linux/module.h>
+#include <linux/kallsyms.h>
+#include <linux/mqueue.h>
+#include <linux/hardirq.h>
+
+#include <asm/pgtable.h>
+#include <asm/uaccess.h>
+#include <asm/system.h>
+#include <asm/io.h>
+#include <asm/processor.h>
+#include <asm/mmu.h>
+#include <asm/prom.h>
+
+extern unsigned long _get_SP(void);
+
+#ifndef CONFIG_SMP
+struct task_struct *last_task_used_math = NULL;
+struct task_struct *last_task_used_altivec = NULL;
+struct task_struct *last_task_used_spe = NULL;
+#endif
+
+static struct fs_struct init_fs = INIT_FS;
+static struct files_struct init_files = INIT_FILES;
+static struct signal_struct init_signals = INIT_SIGNALS(init_signals);
+static struct sighand_struct init_sighand = INIT_SIGHAND(init_sighand);
+struct mm_struct init_mm = INIT_MM(init_mm);
+EXPORT_SYMBOL(init_mm);
+
+/* this is 8kB-aligned so we can get to the thread_info struct
+   at the base of it from the stack pointer with 1 integer instruction. */
+union thread_union init_thread_union
+       __attribute__((__section__(".data.init_task"))) =
+{ INIT_THREAD_INFO(init_task) };
+
+/* initial task structure */
+struct task_struct init_task = INIT_TASK(init_task);
+EXPORT_SYMBOL(init_task);
+
+/* only used to get secondary processor up */
+struct task_struct *current_set[NR_CPUS] = {&init_task, };
+
+/*
+ * Make sure the floating-point register state in the
+ * the thread_struct is up to date for task tsk.
+ */
+void flush_fp_to_thread(struct task_struct *tsk)
+{
+       if (tsk->thread.regs) {
+               /*
+                * We need to disable preemption here because if we didn't,
+                * another process could get scheduled after the regs->msr
+                * test but before we have finished saving the FP registers
+                * to the thread_struct.  That process could take over the
+                * FPU, and then when we get scheduled again we would store
+                * bogus values for the remaining FP registers.
+                */
+               preempt_disable();
+               if (tsk->thread.regs->msr & MSR_FP) {
+#ifdef CONFIG_SMP
+                       /*
+                        * This should only ever be called for current or
+                        * for a stopped child process.  Since we save away
+                        * the FP register state on context switch on SMP,
+                        * there is something wrong if a stopped child appears
+                        * to still have its FP state in the CPU registers.
+                        */
+                       BUG_ON(tsk != current);
+#endif
+                       giveup_fpu(current);
+               }
+               preempt_enable();
+       }
+}
+
+void enable_kernel_fp(void)
+{
+       WARN_ON(preemptible());
+
+#ifdef CONFIG_SMP
+       if (current->thread.regs && (current->thread.regs->msr & MSR_FP))
+               giveup_fpu(current);
+       else
+               giveup_fpu(NULL);       /* just enables FP for kernel */
+#else
+       giveup_fpu(last_task_used_math);
+#endif /* CONFIG_SMP */
+}
+EXPORT_SYMBOL(enable_kernel_fp);
+
+int dump_task_fpu(struct task_struct *tsk, elf_fpregset_t *fpregs)
+{
+       if (!tsk->thread.regs)
+               return 0;
+       flush_fp_to_thread(current);
+
+       memcpy(fpregs, &tsk->thread.fpr[0], sizeof(*fpregs));
+
+       return 1;
+}
+
+#ifdef CONFIG_ALTIVEC
+void enable_kernel_altivec(void)
+{
+       WARN_ON(preemptible());
+
+#ifdef CONFIG_SMP
+       if (current->thread.regs && (current->thread.regs->msr & MSR_VEC))
+               giveup_altivec(current);
+       else
+               giveup_altivec(NULL);   /* just enable AltiVec for kernel - force */
+#else
+       giveup_altivec(last_task_used_altivec);
+#endif /* CONFIG_SMP */
+}
+EXPORT_SYMBOL(enable_kernel_altivec);
+
+/*
+ * Make sure the VMX/Altivec register state in the
+ * the thread_struct is up to date for task tsk.
+ */
+void flush_altivec_to_thread(struct task_struct *tsk)
+{
+       if (tsk->thread.regs) {
+               preempt_disable();
+               if (tsk->thread.regs->msr & MSR_VEC) {
+#ifdef CONFIG_SMP
+                       BUG_ON(tsk != current);
+#endif
+                       giveup_altivec(current);
+               }
+               preempt_enable();
+       }
+}
+
+int dump_task_altivec(struct pt_regs *regs, elf_vrregset_t *vrregs)
+{
+       flush_altivec_to_thread(current);
+       memcpy(vrregs, &current->thread.vr[0], sizeof(*vrregs));
+       return 1;
+}
+#endif /* CONFIG_ALTIVEC */
+
+#ifdef CONFIG_SPE
+
+void enable_kernel_spe(void)
+{
+       WARN_ON(preemptible());
+
+#ifdef CONFIG_SMP
+       if (current->thread.regs && (current->thread.regs->msr & MSR_SPE))
+               giveup_spe(current);
+       else
+               giveup_spe(NULL);       /* just enable SPE for kernel - force */
+#else
+       giveup_spe(last_task_used_spe);
+#endif /* __SMP __ */
+}
+EXPORT_SYMBOL(enable_kernel_spe);
+
+void flush_spe_to_thread(struct task_struct *tsk)
+{
+       if (tsk->thread.regs) {
+               preempt_disable();
+               if (tsk->thread.regs->msr & MSR_SPE) {
+#ifdef CONFIG_SMP
+                       BUG_ON(tsk != current);
+#endif
+                       giveup_spe(current);
+               }
+               preempt_enable();
+       }
+}
+
+int dump_spe(struct pt_regs *regs, elf_vrregset_t *evrregs)
+{
+       flush_spe_to_thread(current);
+       /* We copy u32 evr[32] + u64 acc + u32 spefscr -> 35 */
+       memcpy(evrregs, &current->thread.evr[0], sizeof(u32) * 35);
+       return 1;
+}
+#endif /* CONFIG_SPE */
+
+static void set_dabr_spr(unsigned long val)
+{
+       mtspr(SPRN_DABR, val);
+}
+
+int set_dabr(unsigned long dabr)
+{
+       int ret = 0;
+
+#ifdef CONFIG_PPC64
+       if (firmware_has_feature(FW_FEATURE_XDABR)) {
+               /* We want to catch accesses from kernel and userspace */
+               unsigned long flags = H_DABRX_KERNEL|H_DABRX_USER;
+               ret = plpar_set_xdabr(dabr, flags);
+       } else if (firmware_has_feature(FW_FEATURE_DABR)) {
+               ret = plpar_set_dabr(dabr);
+       } else
+#endif
+               set_dabr_spr(dabr);
+
+       return ret;
+}
+
+static DEFINE_PER_CPU(unsigned long, current_dabr);
+
+struct task_struct *__switch_to(struct task_struct *prev,
+       struct task_struct *new)
+{
+       struct thread_struct *new_thread, *old_thread;
+       unsigned long flags;
+       struct task_struct *last;
+
+#ifdef CONFIG_SMP
+       /* avoid complexity of lazy save/restore of fpu
+        * by just saving it every time we switch out if
+        * this task used the fpu during the last quantum.
+        *
+        * If it tries to use the fpu again, it'll trap and
+        * reload its fp regs.  So we don't have to do a restore
+        * every switch, just a save.
+        *  -- Cort
+        */
+       if (prev->thread.regs && (prev->thread.regs->msr & MSR_FP))
+               giveup_fpu(prev);
+#ifdef CONFIG_ALTIVEC
+       /*
+        * If the previous thread used altivec in the last quantum
+        * (thus changing altivec regs) then save them.
+        * We used to check the VRSAVE register but not all apps
+        * set it, so we don't rely on it now (and in fact we need
+        * to save & restore VSCR even if VRSAVE == 0).  -- paulus
+        *
+        * On SMP we always save/restore altivec regs just to avoid the
+        * complexity of changing processors.
+        *  -- Cort
+        */
+       if (prev->thread.regs && (prev->thread.regs->msr & MSR_VEC))
+               giveup_altivec(prev);
+       /* Avoid the trap.  On smp this this never happens since
+        * we don't set last_task_used_altivec -- Cort
+        */
+       if (new->thread.regs && last_task_used_altivec == new)
+               new->thread.regs->msr |= MSR_VEC;
+#endif /* CONFIG_ALTIVEC */
+#ifdef CONFIG_SPE
+       /*
+        * If the previous thread used spe in the last quantum
+        * (thus changing spe regs) then save them.
+        *
+        * On SMP we always save/restore spe regs just to avoid the
+        * complexity of changing processors.
+        */
+       if ((prev->thread.regs && (prev->thread.regs->msr & MSR_SPE)))
+               giveup_spe(prev);
+       /* Avoid the trap.  On smp this this never happens since
+        * we don't set last_task_used_spe
+        */
+       if (new->thread.regs && last_task_used_spe == new)
+               new->thread.regs->msr |= MSR_SPE;
+#endif /* CONFIG_SPE */
+#endif /* CONFIG_SMP */
+
+#ifdef CONFIG_PPC64    /* for now */
+       if (unlikely(__get_cpu_var(current_dabr) != new->thread.dabr)) {
+               set_dabr(new->thread.dabr);
+               __get_cpu_var(current_dabr) = new->thread.dabr;
+       }
+#endif
+
+       new_thread = &new->thread;
+       old_thread = &current->thread;
+       local_irq_save(flags);
+       last = _switch(old_thread, new_thread);
+
+       local_irq_restore(flags);
+
+       return last;
+}
+
+void show_regs(struct pt_regs * regs)
+{
+       int i, trap;
+
+       printk("NIP: %08lX LR: %08lX SP: %08lX REGS: %p TRAP: %04lx    %s\n",
+              regs->nip, regs->link, regs->gpr[1], regs, regs->trap,
+              print_tainted());
+       printk("MSR: %08lx EE: %01x PR: %01x FP: %01x ME: %01x IR/DR: %01x%01x\n",
+              regs->msr, regs->msr&MSR_EE ? 1 : 0, regs->msr&MSR_PR ? 1 : 0,
+              regs->msr & MSR_FP ? 1 : 0,regs->msr&MSR_ME ? 1 : 0,
+              regs->msr&MSR_IR ? 1 : 0,
+              regs->msr&MSR_DR ? 1 : 0);
+       trap = TRAP(regs);
+       if (trap == 0x300 || trap == 0x600)
+               printk("DAR: %08lX, DSISR: %08lX\n", regs->dar, regs->dsisr);
+       printk("TASK = %p[%d] '%s' THREAD: %p\n",
+              current, current->pid, current->comm, current->thread_info);
+       printk("Last syscall: %ld ", current->thread.last_syscall);
+
+#ifdef CONFIG_SMP
+       printk(" CPU: %d", smp_processor_id());
+#endif /* CONFIG_SMP */
+
+       for (i = 0;  i < 32;  i++) {
+               long r;
+               if ((i % 8) == 0)
+                       printk("\n" KERN_INFO "GPR%02d: ", i);
+               if (__get_user(r, &regs->gpr[i]))
+                       break;
+               printk("%08lX ", r);
+               if (i == 12 && !FULL_REGS(regs))
+                       break;
+       }
+       printk("\n");
+#ifdef CONFIG_KALLSYMS
+       /*
+        * Lookup NIP late so we have the best change of getting the
+        * above info out without failing
+        */
+       printk("NIP [%08lx] ", regs->nip);
+       print_symbol("%s\n", regs->nip);
+       printk("LR [%08lx] ", regs->link);
+       print_symbol("%s\n", regs->link);
+#endif
+       show_stack(current, (unsigned long *) regs->gpr[1]);
+}
+
+void exit_thread(void)
+{
+#ifndef CONFIG_SMP
+       if (last_task_used_math == current)
+               last_task_used_math = NULL;
+#ifdef CONFIG_ALTIVEC
+       if (last_task_used_altivec == current)
+               last_task_used_altivec = NULL;
+#endif /* CONFIG_ALTIVEC */
+#ifdef CONFIG_SPE
+       if (last_task_used_spe == current)
+               last_task_used_spe = NULL;
+#endif
+#endif /* CONFIG_SMP */
+}
+
+void flush_thread(void)
+{
+#ifndef CONFIG_SMP
+       if (last_task_used_math == current)
+               last_task_used_math = NULL;
+#ifdef CONFIG_ALTIVEC
+       if (last_task_used_altivec == current)
+               last_task_used_altivec = NULL;
+#endif /* CONFIG_ALTIVEC */
+#ifdef CONFIG_SPE
+       if (last_task_used_spe == current)
+               last_task_used_spe = NULL;
+#endif
+#endif /* CONFIG_SMP */
+
+#ifdef CONFIG_PPC64    /* for now */
+       if (current->thread.dabr) {
+               current->thread.dabr = 0;
+               set_dabr(0);
+       }
+#endif
+}
+
+void
+release_thread(struct task_struct *t)
+{
+}
+
+/*
+ * This gets called before we allocate a new thread and copy
+ * the current task into it.
+ */
+void prepare_to_copy(struct task_struct *tsk)
+{
+       flush_fp_to_thread(current);
+       flush_altivec_to_thread(current);
+       flush_spe_to_thread(current);
+}
+
+/*
+ * Copy a thread..
+ */
+int
+copy_thread(int nr, unsigned long clone_flags, unsigned long usp,
+           unsigned long unused,
+           struct task_struct *p, struct pt_regs *regs)
+{
+       struct pt_regs *childregs, *kregs;
+       extern void ret_from_fork(void);
+       unsigned long sp = (unsigned long)p->thread_info + THREAD_SIZE;
+       unsigned long childframe;
+
+       CHECK_FULL_REGS(regs);
+       /* Copy registers */
+       sp -= sizeof(struct pt_regs);
+       childregs = (struct pt_regs *) sp;
+       *childregs = *regs;
+       if ((childregs->msr & MSR_PR) == 0) {
+               /* for kernel thread, set `current' and stackptr in new task */
+               childregs->gpr[1] = sp + sizeof(struct pt_regs);
+               childregs->gpr[2] = (unsigned long) p;
+               p->thread.regs = NULL;  /* no user register state */
+       } else {
+               childregs->gpr[1] = usp;
+               p->thread.regs = childregs;
+               if (clone_flags & CLONE_SETTLS)
+                       childregs->gpr[2] = childregs->gpr[6];
+       }
+       childregs->gpr[3] = 0;  /* Result from fork() */
+       sp -= STACK_FRAME_OVERHEAD;
+       childframe = sp;
+
+       /*
+        * The way this works is that at some point in the future
+        * some task will call _switch to switch to the new task.
+        * That will pop off the stack frame created below and start
+        * the new task running at ret_from_fork.  The new task will
+        * do some house keeping and then return from the fork or clone
+        * system call, using the stack frame created above.
+        */
+       sp -= sizeof(struct pt_regs);
+       kregs = (struct pt_regs *) sp;
+       sp -= STACK_FRAME_OVERHEAD;
+       p->thread.ksp = sp;
+       kregs->nip = (unsigned long)ret_from_fork;
+
+       p->thread.last_syscall = -1;
+
+       return 0;
+}
+
+/*
+ * Set up a thread for executing a new program
+ */
+void start_thread(struct pt_regs *regs, unsigned long nip, unsigned long sp)
+{
+       set_fs(USER_DS);
+       memset(regs->gpr, 0, sizeof(regs->gpr));
+       regs->ctr = 0;
+       regs->link = 0;
+       regs->xer = 0;
+       regs->ccr = 0;
+       regs->mq = 0;
+       regs->nip = nip;
+       regs->gpr[1] = sp;
+       regs->msr = MSR_USER;
+#ifndef CONFIG_SMP
+       if (last_task_used_math == current)
+               last_task_used_math = NULL;
+#ifdef CONFIG_ALTIVEC
+       if (last_task_used_altivec == current)
+               last_task_used_altivec = NULL;
+#endif
+#ifdef CONFIG_SPE
+       if (last_task_used_spe == current)
+               last_task_used_spe = NULL;
+#endif
+#endif /* CONFIG_SMP */
+       memset(current->thread.fpr, 0, sizeof(current->thread.fpr));
+       current->thread.fpscr = 0;
+#ifdef CONFIG_ALTIVEC
+       memset(current->thread.vr, 0, sizeof(current->thread.vr));
+       memset(&current->thread.vscr, 0, sizeof(current->thread.vscr));
+       current->thread.vrsave = 0;
+       current->thread.used_vr = 0;
+#endif /* CONFIG_ALTIVEC */
+#ifdef CONFIG_SPE
+       memset(current->thread.evr, 0, sizeof(current->thread.evr));
+       current->thread.acc = 0;
+       current->thread.spefscr = 0;
+       current->thread.used_spe = 0;
+#endif /* CONFIG_SPE */
+}
+
+#define PR_FP_ALL_EXCEPT (PR_FP_EXC_DIV | PR_FP_EXC_OVF | PR_FP_EXC_UND \
+               | PR_FP_EXC_RES | PR_FP_EXC_INV)
+
+int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
+{
+       struct pt_regs *regs = tsk->thread.regs;
+
+       /* This is a bit hairy.  If we are an SPE enabled  processor
+        * (have embedded fp) we store the IEEE exception enable flags in
+        * fpexc_mode.  fpexc_mode is also used for setting FP exception
+        * mode (asyn, precise, disabled) for 'Classic' FP. */
+       if (val & PR_FP_EXC_SW_ENABLE) {
+#ifdef CONFIG_SPE
+               tsk->thread.fpexc_mode = val &
+                       (PR_FP_EXC_SW_ENABLE | PR_FP_ALL_EXCEPT);
+#else
+               return -EINVAL;
+#endif
+       } else {
+               /* on a CONFIG_SPE this does not hurt us.  The bits that
+                * __pack_fe01 use do not overlap with bits used for
+                * PR_FP_EXC_SW_ENABLE.  Additionally, the MSR[FE0,FE1] bits
+                * on CONFIG_SPE implementations are reserved so writing to
+                * them does not change anything */
+               if (val > PR_FP_EXC_PRECISE)
+                       return -EINVAL;
+               tsk->thread.fpexc_mode = __pack_fe01(val);
+               if (regs != NULL && (regs->msr & MSR_FP) != 0)
+                       regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
+                               | tsk->thread.fpexc_mode;
+       }
+       return 0;
+}
+
+int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
+{
+       unsigned int val;
+
+       if (tsk->thread.fpexc_mode & PR_FP_EXC_SW_ENABLE)
+#ifdef CONFIG_SPE
+               val = tsk->thread.fpexc_mode;
+#else
+               return -EINVAL;
+#endif
+       else
+               val = __unpack_fe01(tsk->thread.fpexc_mode);
+       return put_user(val, (unsigned int __user *) adr);
+}
+
+int sys_clone(unsigned long clone_flags, unsigned long usp,
+             int __user *parent_tidp, void __user *child_threadptr,
+             int __user *child_tidp, int p6,
+             struct pt_regs *regs)
+{
+       CHECK_FULL_REGS(regs);
+       if (usp == 0)
+               usp = regs->gpr[1];     /* stack pointer for child */
+       return do_fork(clone_flags, usp, regs, 0, parent_tidp, child_tidp);
+}
+
+int sys_fork(unsigned long p1, unsigned long p2, unsigned long p3,
+            unsigned long p4, unsigned long p5, unsigned long p6,
+            struct pt_regs *regs)
+{
+       CHECK_FULL_REGS(regs);
+       return do_fork(SIGCHLD, regs->gpr[1], regs, 0, NULL, NULL);
+}
+
+int sys_vfork(unsigned long p1, unsigned long p2, unsigned long p3,
+             unsigned long p4, unsigned long p5, unsigned long p6,
+             struct pt_regs *regs)
+{
+       CHECK_FULL_REGS(regs);
+       return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->gpr[1],
+                       regs, 0, NULL, NULL);
+}
+
+int sys_execve(unsigned long a0, unsigned long a1, unsigned long a2,
+              unsigned long a3, unsigned long a4, unsigned long a5,
+              struct pt_regs *regs)
+{
+       int error;
+       char * filename;
+
+       filename = getname((char __user *) a0);
+       error = PTR_ERR(filename);
+       if (IS_ERR(filename))
+               goto out;
+       flush_fp_to_thread(current);
+       flush_altivec_to_thread(current);
+       flush_spe_to_thread(current);
+       if (error == 0) {
+               task_lock(current);
+               current->ptrace &= ~PT_DTRACE;
+               task_unlock(current);
+       }
+       putname(filename);
+out:
+       return error;
+}
+
+static int validate_sp(unsigned long sp, struct task_struct *p,
+                      unsigned long nbytes)
+{
+       unsigned long stack_page = (unsigned long)p->thread_info;
+
+       if (sp >= stack_page + sizeof(struct thread_struct)
+           && sp <= stack_page + THREAD_SIZE - nbytes)
+               return 1;
+
+#ifdef CONFIG_IRQSTACKS
+       stack_page = (unsigned long) hardirq_ctx[task_cpu(p)];
+       if (sp >= stack_page + sizeof(struct thread_struct)
+           && sp <= stack_page + THREAD_SIZE - nbytes)
+               return 1;
+
+       stack_page = (unsigned long) softirq_ctx[task_cpu(p)];
+       if (sp >= stack_page + sizeof(struct thread_struct)
+           && sp <= stack_page + THREAD_SIZE - nbytes)
+               return 1;
+#endif
+
+       return 0;
+}
+
+void dump_stack(void)
+{
+       show_stack(current, NULL);
+}
+
+EXPORT_SYMBOL(dump_stack);
+
+void show_stack(struct task_struct *tsk, unsigned long *stack)
+{
+       unsigned long sp, stack_top, prev_sp, ret;
+       int count = 0;
+       unsigned long next_exc = 0;
+       struct pt_regs *regs;
+       extern char ret_from_except, ret_from_except_full, ret_from_syscall;
+
+       sp = (unsigned long) stack;
+       if (tsk == NULL)
+               tsk = current;
+       if (sp == 0) {
+               if (tsk == current)
+                       asm("mr %0,1" : "=r" (sp));
+               else
+                       sp = tsk->thread.ksp;
+       }
+
+       prev_sp = (unsigned long) (tsk->thread_info + 1);
+       stack_top = (unsigned long) tsk->thread_info + THREAD_SIZE;
+       while (count < 16 && sp > prev_sp && sp < stack_top && (sp & 3) == 0) {
+               if (count == 0) {
+                       printk("Call trace:");
+#ifdef CONFIG_KALLSYMS
+                       printk("\n");
+#endif
+               } else {
+                       if (next_exc) {
+                               ret = next_exc;
+                               next_exc = 0;
+                       } else
+                               ret = *(unsigned long *)(sp + 4);
+                       printk(" [%08lx] ", ret);
+#ifdef CONFIG_KALLSYMS
+                       print_symbol("%s", ret);
+                       printk("\n");
+#endif
+                       if (ret == (unsigned long) &ret_from_except
+                           || ret == (unsigned long) &ret_from_except_full
+                           || ret == (unsigned long) &ret_from_syscall) {
+                               /* sp + 16 points to an exception frame */
+                               regs = (struct pt_regs *) (sp + 16);
+                               if (sp + 16 + sizeof(*regs) <= stack_top)
+                                       next_exc = regs->nip;
+                       }
+               }
+               ++count;
+               sp = *(unsigned long *)sp;
+       }
+#ifndef CONFIG_KALLSYMS
+       if (count > 0)
+               printk("\n");
+#endif
+}
+
+unsigned long get_wchan(struct task_struct *p)
+{
+       unsigned long ip, sp;
+       int count = 0;
+
+       if (!p || p == current || p->state == TASK_RUNNING)
+               return 0;
+
+       sp = p->thread.ksp;
+       if (!validate_sp(sp, p, 16))
+               return 0;
+
+       do {
+               sp = *(unsigned long *)sp;
+               if (!validate_sp(sp, p, 16))
+                       return 0;
+               if (count > 0) {
+                       ip = *(unsigned long *)(sp + 4);
+                       if (!in_sched_functions(ip))
+                               return ip;
+               }
+       } while (count++ < 16);
+       return 0;
+}
+EXPORT_SYMBOL(get_wchan);
diff --git a/arch/powerpc/kernel/semaphore.c b/arch/powerpc/kernel/semaphore.c
new file mode 100644 (file)
index 0000000..2f8c3c9
--- /dev/null
@@ -0,0 +1,135 @@
+/*
+ * PowerPC-specific semaphore code.
+ *
+ * Copyright (C) 1999 Cort Dougan <cort@cs.nmt.edu>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version
+ * 2 of the License, or (at your option) any later version.
+ *
+ * April 2001 - Reworked by Paul Mackerras <paulus@samba.org>
+ * to eliminate the SMP races in the old version between the updates
+ * of `count' and `waking'.  Now we use negative `count' values to
+ * indicate that some process(es) are waiting for the semaphore.
+ */
+
+#include <linux/sched.h>
+#include <linux/init.h>
+#include <linux/module.h>
+
+#include <asm/atomic.h>
+#include <asm/semaphore.h>
+#include <asm/errno.h>
+
+/*
+ * Atomically update sem->count.
+ * This does the equivalent of the following:
+ *
+ *     old_count = sem->count;
+ *     tmp = MAX(old_count, 0) + incr;
+ *     sem->count = tmp;
+ *     return old_count;
+ */
+static inline int __sem_update_count(struct semaphore *sem, int incr)
+{
+       int old_count, tmp;
+
+       __asm__ __volatile__("\n"
+"1:    lwarx   %0,0,%3\n"
+"      srawi   %1,%0,31\n"
+"      andc    %1,%0,%1\n"
+"      add     %1,%1,%4\n"
+       PPC405_ERR77(0,%3)
+"      stwcx.  %1,0,%3\n"
+"      bne     1b"
+       : "=&r" (old_count), "=&r" (tmp), "=m" (sem->count)
+       : "r" (&sem->count), "r" (incr), "m" (sem->count)
+       : "cc");
+
+       return old_count;
+}
+
+void __up(struct semaphore *sem)
+{
+       /*
+        * Note that we incremented count in up() before we came here,
+        * but that was ineffective since the result was <= 0, and
+        * any negative value of count is equivalent to 0.
+        * This ends up setting count to 1, unless count is now > 0
+        * (i.e. because some other cpu has called up() in the meantime),
+        * in which case we just increment count.
+        */
+       __sem_update_count(sem, 1);
+       wake_up(&sem->wait);
+}
+EXPORT_SYMBOL(__up);
+
+/*
+ * Note that when we come in to __down or __down_interruptible,
+ * we have already decremented count, but that decrement was
+ * ineffective since the result was < 0, and any negative value
+ * of count is equivalent to 0.
+ * Thus it is only when we decrement count from some value > 0
+ * that we have actually got the semaphore.
+ */
+void __sched __down(struct semaphore *sem)
+{
+       struct task_struct *tsk = current;
+       DECLARE_WAITQUEUE(wait, tsk);
+
+       __set_task_state(tsk, TASK_UNINTERRUPTIBLE);
+       add_wait_queue_exclusive(&sem->wait, &wait);
+
+       /*
+        * Try to get the semaphore.  If the count is > 0, then we've
+        * got the semaphore; we decrement count and exit the loop.
+        * If the count is 0 or negative, we set it to -1, indicating
+        * that we are asleep, and then sleep.
+        */
+       while (__sem_update_count(sem, -1) <= 0) {
+               schedule();
+               set_task_state(tsk, TASK_UNINTERRUPTIBLE);
+       }
+       remove_wait_queue(&sem->wait, &wait);
+       __set_task_state(tsk, TASK_RUNNING);
+
+       /*
+        * If there are any more sleepers, wake one of them up so
+        * that it can either get the semaphore, or set count to -1
+        * indicating that there are still processes sleeping.
+        */
+       wake_up(&sem->wait);
+}
+EXPORT_SYMBOL(__down);
+
+int __sched __down_interruptible(struct semaphore * sem)
+{
+       int retval = 0;
+       struct task_struct *tsk = current;
+       DECLARE_WAITQUEUE(wait, tsk);
+
+       __set_task_state(tsk, TASK_INTERRUPTIBLE);
+       add_wait_queue_exclusive(&sem->wait, &wait);
+
+       while (__sem_update_count(sem, -1) <= 0) {
+               if (signal_pending(current)) {
+                       /*
+                        * A signal is pending - give up trying.
+                        * Set sem->count to 0 if it is negative,
+                        * since we are no longer sleeping.
+                        */
+                       __sem_update_count(sem, 0);
+                       retval = -EINTR;
+                       break;
+               }
+               schedule();
+               set_task_state(tsk, TASK_INTERRUPTIBLE);
+       }
+       remove_wait_queue(&sem->wait, &wait);
+       __set_task_state(tsk, TASK_RUNNING);
+
+       wake_up(&sem->wait);
+       return retval;
+}
+EXPORT_SYMBOL(__down_interruptible);
diff --git a/arch/powerpc/kernel/traps.c b/arch/powerpc/kernel/traps.c
new file mode 100644 (file)
index 0000000..c7afbbb
--- /dev/null
@@ -0,0 +1,1047 @@
+/*
+ *  arch/powerpc/kernel/traps.c
+ *
+ *  Copyright (C) 1995-1996  Gary Thomas (gdt@linuxppc.org)
+ *
+ *  This program is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU General Public License
+ *  as published by the Free Software Foundation; either version
+ *  2 of the License, or (at your option) any later version.
+ *
+ *  Modified by Cort Dougan (cort@cs.nmt.edu)
+ *  and Paul Mackerras (paulus@samba.org)
+ */
+
+/*
+ * This file handles the architecture-dependent parts of hardware exceptions
+ */
+
+#include <linux/config.h>
+#include <linux/errno.h>
+#include <linux/sched.h>
+#include <linux/kernel.h>
+#include <linux/mm.h>
+#include <linux/stddef.h>
+#include <linux/unistd.h>
+#include <linux/ptrace.h>
+#include <linux/slab.h>
+#include <linux/user.h>
+#include <linux/a.out.h>
+#include <linux/interrupt.h>
+#include <linux/config.h>
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/prctl.h>
+#include <linux/delay.h>
+#include <linux/kprobes.h>
+#include <asm/kdebug.h>
+
+#include <asm/pgtable.h>
+#include <asm/uaccess.h>
+#include <asm/system.h>
+#include <asm/io.h>
+#include <asm/reg.h>
+#include <asm/xmon.h>
+#ifdef CONFIG_PMAC_BACKLIGHT
+#include <asm/backlight.h>
+#endif
+#include <asm/perfmon.h>
+
+#ifdef CONFIG_DEBUGGER
+int (*__debugger)(struct pt_regs *regs);
+int (*__debugger_ipi)(struct pt_regs *regs);
+int (*__debugger_bpt)(struct pt_regs *regs);
+int (*__debugger_sstep)(struct pt_regs *regs);
+int (*__debugger_iabr_match)(struct pt_regs *regs);
+int (*__debugger_dabr_match)(struct pt_regs *regs);
+int (*__debugger_fault_handler)(struct pt_regs *regs);
+
+EXPORT_SYMBOL(__debugger);
+EXPORT_SYMBOL(__debugger_ipi);
+EXPORT_SYMBOL(__debugger_bpt);
+EXPORT_SYMBOL(__debugger_sstep);
+EXPORT_SYMBOL(__debugger_iabr_match);
+EXPORT_SYMBOL(__debugger_dabr_match);
+EXPORT_SYMBOL(__debugger_fault_handler);
+#endif
+
+struct notifier_block *powerpc_die_chain;
+static DEFINE_SPINLOCK(die_notifier_lock);
+
+int register_die_notifier(struct notifier_block *nb)
+{
+       int err = 0;
+       unsigned long flags;
+
+       spin_lock_irqsave(&die_notifier_lock, flags);
+       err = notifier_chain_register(&powerpc_die_chain, nb);
+       spin_unlock_irqrestore(&die_notifier_lock, flags);
+       return err;
+}
+
+/*
+ * Trap & Exception support
+ */
+
+static DEFINE_SPINLOCK(die_lock);
+
+int die(const char *str, struct pt_regs *regs, long err)
+{
+       static int die_counter;
+       int nl = 0;
+
+       if (debugger(regs))
+               return 1;
+
+       console_verbose();
+       spin_lock_irq(&die_lock);
+       bust_spinlocks(1);
+#ifdef CONFIG_PMAC_BACKLIGHT
+       if (_machine == _MACH_Pmac) {
+               set_backlight_enable(1);
+               set_backlight_level(BACKLIGHT_MAX);
+       }
+#endif
+       printk("Oops: %s, sig: %ld [#%d]\n", str, err, ++die_counter);
+#ifdef CONFIG_PREEMPT
+       printk("PREEMPT ");
+       nl = 1;
+#endif
+#ifdef CONFIG_SMP
+       printk("SMP NR_CPUS=%d ", NR_CPUS);
+       nl = 1;
+#endif
+#ifdef CONFIG_DEBUG_PAGEALLOC
+       printk("DEBUG_PAGEALLOC ");
+       nl = 1;
+#endif
+#ifdef CONFIG_NUMA
+       printk("NUMA ");
+       nl = 1;
+#endif
+#ifdef CONFIG_PPC64
+       switch (systemcfg->platform) {
+       case PLATFORM_PSERIES:
+               printk("PSERIES ");
+               nl = 1;
+               break;
+       case PLATFORM_PSERIES_LPAR:
+               printk("PSERIES LPAR ");
+               nl = 1;
+               break;
+       case PLATFORM_ISERIES_LPAR:
+               printk("ISERIES LPAR ");
+               nl = 1;
+               break;
+       case PLATFORM_POWERMAC:
+               printk("POWERMAC ");
+               nl = 1;
+               break;
+       case PLATFORM_BPA:
+               printk("BPA ");
+               nl = 1;
+               break;
+       }
+#endif
+       if (nl)
+               printk("\n");
+       print_modules();
+       show_regs(regs);
+       bust_spinlocks(0);
+       spin_unlock_irq(&die_lock);
+
+       if (in_interrupt())
+               panic("Fatal exception in interrupt");
+
+       if (panic_on_oops) {
+               panic("Fatal exception");
+       }
+       do_exit(err);
+
+       return 0;
+}
+
+void _exception(int signr, struct pt_regs *regs, int code, unsigned long addr)
+{
+       siginfo_t info;
+
+       if (!user_mode(regs)) {
+               if (die("Exception in kernel mode", regs, signr))
+                       return;
+       }
+
+       memset(&info, 0, sizeof(info));
+       info.si_signo = signr;
+       info.si_code = code;
+       info.si_addr = (void __user *) addr;
+       force_sig_info(signr, &info, current);
+
+       /*
+        * Init gets no signals that it doesn't have a handler for.
+        * That's all very well, but if it has caused a synchronous
+        * exception and we ignore the resulting signal, it will just
+        * generate the same exception over and over again and we get
+        * nowhere.  Better to kill it and let the kernel panic.
+        */
+       if (current->pid == 1) {
+               __sighandler_t handler;
+
+               spin_lock_irq(&current->sighand->siglock);
+               handler = current->sighand->action[signr-1].sa.sa_handler;
+               spin_unlock_irq(&current->sighand->siglock);
+               if (handler == SIG_DFL) {
+                       /* init has generated a synchronous exception
+                          and it doesn't have a handler for the signal */
+                       printk(KERN_CRIT "init has generated signal %d "
+                              "but has no handler for it\n", signr);
+                       do_exit(signr);
+               }
+       }
+}
+
+#ifdef CONFIG_PPC64
+void system_reset_exception(struct pt_regs *regs)
+{
+       /* See if any machine dependent calls */
+       if (ppc_md.system_reset_exception)
+               ppc_md.system_reset_exception(regs);
+
+       die("System Reset", regs, SIGABRT);
+
+       /* Must die if the interrupt is not recoverable */
+       if (!(regs->msr & MSR_RI))
+               panic("Unrecoverable System Reset");
+
+       /* What should we do here? We could issue a shutdown or hard reset. */
+}
+#endif
+
+/*
+ * I/O accesses can cause machine checks on powermacs.
+ * Check if the NIP corresponds to the address of a sync
+ * instruction for which there is an entry in the exception
+ * table.
+ * Note that the 601 only takes a machine check on TEA
+ * (transfer error ack) signal assertion, and does not
+ * set any of the top 16 bits of SRR1.
+ *  -- paulus.
+ */
+static inline int check_io_access(struct pt_regs *regs)
+{
+#ifdef CONFIG_PPC_PMAC
+       unsigned long msr = regs->msr;
+       const struct exception_table_entry *entry;
+       unsigned int *nip = (unsigned int *)regs->nip;
+
+       if (((msr & 0xffff0000) == 0 || (msr & (0x80000 | 0x40000)))
+           && (entry = search_exception_tables(regs->nip)) != NULL) {
+               /*
+                * Check that it's a sync instruction, or somewhere
+                * in the twi; isync; nop sequence that inb/inw/inl uses.
+                * As the address is in the exception table
+                * we should be able to read the instr there.
+                * For the debug message, we look at the preceding
+                * load or store.
+                */
+               if (*nip == 0x60000000)         /* nop */
+                       nip -= 2;
+               else if (*nip == 0x4c00012c)    /* isync */
+                       --nip;
+               if (*nip == 0x7c0004ac || (*nip >> 26) == 3) {
+                       /* sync or twi */
+                       unsigned int rb;
+
+                       --nip;
+                       rb = (*nip >> 11) & 0x1f;
+                       printk(KERN_DEBUG "%s bad port %lx at %p\n",
+                              (*nip & 0x100)? "OUT to": "IN from",
+                              regs->gpr[rb] - _IO_BASE, nip);
+                       regs->msr |= MSR_RI;
+                       regs->nip = entry->fixup;
+                       return 1;
+               }
+       }
+#endif /* CONFIG_PPC_PMAC */
+       return 0;
+}
+
+#if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
+/* On 4xx, the reason for the machine check or program exception
+   is in the ESR. */
+#define get_reason(regs)       ((regs)->dsisr)
+#ifndef CONFIG_FSL_BOOKE
+#define get_mc_reason(regs)    ((regs)->dsisr)
+#else
+#define get_mc_reason(regs)    (mfspr(SPRN_MCSR))
+#endif
+#define REASON_FP              ESR_FP
+#define REASON_ILLEGAL         (ESR_PIL | ESR_PUO)
+#define REASON_PRIVILEGED      ESR_PPR
+#define REASON_TRAP            ESR_PTR
+
+/* single-step stuff */
+#define single_stepping(regs)  (current->thread.dbcr0 & DBCR0_IC)
+#define clear_single_step(regs)        (current->thread.dbcr0 &= ~DBCR0_IC)
+
+#else
+/* On non-4xx, the reason for the machine check or program
+   exception is in the MSR. */
+#define get_reason(regs)       ((regs)->msr)
+#define get_mc_reason(regs)    ((regs)->msr)
+#define REASON_FP              0x100000
+#define REASON_ILLEGAL         0x80000
+#define REASON_PRIVILEGED      0x40000
+#define REASON_TRAP            0x20000
+
+#define single_stepping(regs)  ((regs)->msr & MSR_SE)
+#define clear_single_step(regs)        ((regs)->msr &= ~MSR_SE)
+#endif
+
+/*
+ * This is "fall-back" implementation for configurations
+ * which don't provide platform-specific machine check info
+ */
+void __attribute__ ((weak))
+platform_machine_check(struct pt_regs *regs)
+{
+}
+
+void MachineCheckException(struct pt_regs *regs)
+{
+#ifdef CONFIG_PPC64
+       int recover = 0;
+
+       /* See if any machine dependent calls */
+       if (ppc_md.machine_check_exception)
+               recover = ppc_md.machine_check_exception(regs);
+
+       if (recover)
+               return;
+#else
+       unsigned long reason = get_mc_reason(regs);
+
+       if (user_mode(regs)) {
+               regs->msr |= MSR_RI;
+               _exception(SIGBUS, regs, BUS_ADRERR, regs->nip);
+               return;
+       }
+
+#if defined(CONFIG_8xx) && defined(CONFIG_PCI)
+       /* the qspan pci read routines can cause machine checks -- Cort */
+       bad_page_fault(regs, regs->dar, SIGBUS);
+       return;
+#endif
+
+       if (debugger_fault_handler(regs)) {
+               regs->msr |= MSR_RI;
+               return;
+       }
+
+       if (check_io_access(regs))
+               return;
+
+#if defined(CONFIG_4xx) && !defined(CONFIG_440A)
+       if (reason & ESR_IMCP) {
+               printk("Instruction");
+               mtspr(SPRN_ESR, reason & ~ESR_IMCP);
+       } else
+               printk("Data");
+       printk(" machine check in kernel mode.\n");
+#elif defined(CONFIG_440A)
+       printk("Machine check in kernel mode.\n");
+       if (reason & ESR_IMCP){
+               printk("Instruction Synchronous Machine Check exception\n");
+               mtspr(SPRN_ESR, reason & ~ESR_IMCP);
+       }
+       else {
+               u32 mcsr = mfspr(SPRN_MCSR);
+               if (mcsr & MCSR_IB)
+                       printk("Instruction Read PLB Error\n");
+               if (mcsr & MCSR_DRB)
+                       printk("Data Read PLB Error\n");
+               if (mcsr & MCSR_DWB)
+                       printk("Data Write PLB Error\n");
+               if (mcsr & MCSR_TLBP)
+                       printk("TLB Parity Error\n");
+               if (mcsr & MCSR_ICP){
+                       flush_instruction_cache();
+                       printk("I-Cache Parity Error\n");
+               }
+               if (mcsr & MCSR_DCSP)
+                       printk("D-Cache Search Parity Error\n");
+               if (mcsr & MCSR_DCFP)
+                       printk("D-Cache Flush Parity Error\n");
+               if (mcsr & MCSR_IMPE)
+                       printk("Machine Check exception is imprecise\n");
+
+               /* Clear MCSR */
+               mtspr(SPRN_MCSR, mcsr);
+       }
+#elif defined (CONFIG_E500)
+       printk("Machine check in kernel mode.\n");
+       printk("Caused by (from MCSR=%lx): ", reason);
+
+       if (reason & MCSR_MCP)
+               printk("Machine Check Signal\n");
+       if (reason & MCSR_ICPERR)
+               printk("Instruction Cache Parity Error\n");
+       if (reason & MCSR_DCP_PERR)
+               printk("Data Cache Push Parity Error\n");
+       if (reason & MCSR_DCPERR)
+               printk("Data Cache Parity Error\n");
+       if (reason & MCSR_GL_CI)
+               printk("Guarded Load or Cache-Inhibited stwcx.\n");
+       if (reason & MCSR_BUS_IAERR)
+               printk("Bus - Instruction Address Error\n");
+       if (reason & MCSR_BUS_RAERR)
+               printk("Bus - Read Address Error\n");
+       if (reason & MCSR_BUS_WAERR)
+               printk("Bus - Write Address Error\n");
+       if (reason & MCSR_BUS_IBERR)
+               printk("Bus - Instruction Data Error\n");
+       if (reason & MCSR_BUS_RBERR)
+               printk("Bus - Read Data Bus Error\n");
+       if (reason & MCSR_BUS_WBERR)
+               printk("Bus - Read Data Bus Error\n");
+       if (reason & MCSR_BUS_IPERR)
+               printk("Bus - Instruction Parity Error\n");
+       if (reason & MCSR_BUS_RPERR)
+               printk("Bus - Read Parity Error\n");
+#elif defined (CONFIG_E200)
+       printk("Machine check in kernel mode.\n");
+       printk("Caused by (from MCSR=%lx): ", reason);
+
+       if (reason & MCSR_MCP)
+               printk("Machine Check Signal\n");
+       if (reason & MCSR_CP_PERR)
+               printk("Cache Push Parity Error\n");
+       if (reason & MCSR_CPERR)
+               printk("Cache Parity Error\n");
+       if (reason & MCSR_EXCP_ERR)
+               printk("ISI, ITLB, or Bus Error on first instruction fetch for an exception handler\n");
+       if (reason & MCSR_BUS_IRERR)
+               printk("Bus - Read Bus Error on instruction fetch\n");
+       if (reason & MCSR_BUS_DRERR)
+               printk("Bus - Read Bus Error on data load\n");
+       if (reason & MCSR_BUS_WRERR)
+               printk("Bus - Write Bus Error on buffered store or cache line push\n");
+#else /* !CONFIG_4xx && !CONFIG_E500 && !CONFIG_E200 */
+       printk("Machine check in kernel mode.\n");
+       printk("Caused by (from SRR1=%lx): ", reason);
+       switch (reason & 0x601F0000) {
+       case 0x80000:
+               printk("Machine check signal\n");
+               break;
+       case 0:         /* for 601 */
+       case 0x40000:
+       case 0x140000:  /* 7450 MSS error and TEA */
+               printk("Transfer error ack signal\n");
+               break;
+       case 0x20000:
+               printk("Data parity error signal\n");
+               break;
+       case 0x10000:
+               printk("Address parity error signal\n");
+               break;
+       case 0x20000000:
+               printk("L1 Data Cache error\n");
+               break;
+       case 0x40000000:
+               printk("L1 Instruction Cache error\n");
+               break;
+       case 0x00100000:
+               printk("L2 data cache parity error\n");
+               break;
+       default:
+               printk("Unknown values in msr\n");
+       }
+#endif /* CONFIG_4xx */
+
+       /*
+        * Optional platform-provided routine to print out
+        * additional info, e.g. bus error registers.
+        */
+       platform_machine_check(regs);
+#endif /* CONFIG_PPC64 */
+
+       if (debugger_fault_handler(regs))
+               return;
+       die("Machine check", regs, SIGBUS);
+
+       /* Must die if the interrupt is not recoverable */
+       if (!(regs->msr & MSR_RI))
+               panic("Unrecoverable Machine check");
+}
+
+void SMIException(struct pt_regs *regs)
+{
+       die("System Management Interrupt", regs, SIGABRT);
+}
+
+void UnknownException(struct pt_regs *regs)
+{
+       printk("Bad trap at PC: %lx, SR: %lx, vector=%lx\n",
+              regs->nip, regs->msr, regs->trap);
+
+       _exception(SIGTRAP, regs, 0, 0);
+}
+
+void InstructionBreakpoint(struct pt_regs *regs)
+{
+       if (notify_die(DIE_IABR_MATCH, "iabr_match", regs, 5,
+                                       5, SIGTRAP) == NOTIFY_STOP)
+               return;
+       if (debugger_iabr_match(regs))
+               return;
+       _exception(SIGTRAP, regs, TRAP_BRKPT, regs->nip);
+}
+
+void RunModeException(struct pt_regs *regs)
+{
+       _exception(SIGTRAP, regs, 0, 0);
+}
+
+void SingleStepException(struct pt_regs *regs)
+{
+       regs->msr &= ~(MSR_SE | MSR_BE);  /* Turn off 'trace' bits */
+
+       if (notify_die(DIE_SSTEP, "single_step", regs, 5,
+                                       5, SIGTRAP) == NOTIFY_STOP)
+               return;
+       if (debugger_sstep(regs))
+               return;
+
+       _exception(SIGTRAP, regs, TRAP_TRACE, regs->nip);
+}
+
+/*
+ * After we have successfully emulated an instruction, we have to
+ * check if the instruction was being single-stepped, and if so,
+ * pretend we got a single-step exception.  This was pointed out
+ * by Kumar Gala.  -- paulus
+ */
+static void emulate_single_step(struct pt_regs *regs)
+{
+       if (single_stepping(regs)) {
+               clear_single_step(regs);
+               _exception(SIGTRAP, regs, TRAP_TRACE, 0);
+       }
+}
+
+/* Illegal instruction emulation support.  Originally written to
+ * provide the PVR to user applications using the mfspr rd, PVR.
+ * Return non-zero if we can't emulate, or -EFAULT if the associated
+ * memory access caused an access fault.  Return zero on success.
+ *
+ * There are a couple of ways to do this, either "decode" the instruction
+ * or directly match lots of bits.  In this case, matching lots of
+ * bits is faster and easier.
+ *
+ */
+#define INST_MFSPR_PVR         0x7c1f42a6
+#define INST_MFSPR_PVR_MASK    0xfc1fffff
+
+#define INST_DCBA              0x7c0005ec
+#define INST_DCBA_MASK         0x7c0007fe
+
+#define INST_MCRXR             0x7c000400
+#define INST_MCRXR_MASK                0x7c0007fe
+
+#define INST_STRING            0x7c00042a
+#define INST_STRING_MASK       0x7c0007fe
+#define INST_STRING_GEN_MASK   0x7c00067e
+#define INST_LSWI              0x7c0004aa
+#define INST_LSWX              0x7c00042a
+#define INST_STSWI             0x7c0005aa
+#define INST_STSWX             0x7c00052a
+
+static int emulate_string_inst(struct pt_regs *regs, u32 instword)
+{
+       u8 rT = (instword >> 21) & 0x1f;
+       u8 rA = (instword >> 16) & 0x1f;
+       u8 NB_RB = (instword >> 11) & 0x1f;
+       u32 num_bytes;
+       unsigned long EA;
+       int pos = 0;
+
+       /* Early out if we are an invalid form of lswx */
+       if ((instword & INST_STRING_MASK) == INST_LSWX)
+               if ((rT == rA) || (rT == NB_RB))
+                       return -EINVAL;
+
+       EA = (rA == 0) ? 0 : regs->gpr[rA];
+
+       switch (instword & INST_STRING_MASK) {
+               case INST_LSWX:
+               case INST_STSWX:
+                       EA += NB_RB;
+                       num_bytes = regs->xer & 0x7f;
+                       break;
+               case INST_LSWI:
+               case INST_STSWI:
+                       num_bytes = (NB_RB == 0) ? 32 : NB_RB;
+                       break;
+               default:
+                       return -EINVAL;
+       }
+
+       while (num_bytes != 0)
+       {
+               u8 val;
+               u32 shift = 8 * (3 - (pos & 0x3));
+
+               switch ((instword & INST_STRING_MASK)) {
+                       case INST_LSWX:
+                       case INST_LSWI:
+                               if (get_user(val, (u8 __user *)EA))
+                                       return -EFAULT;
+                               /* first time updating this reg,
+                                * zero it out */
+                               if (pos == 0)
+                                       regs->gpr[rT] = 0;
+                               regs->gpr[rT] |= val << shift;
+                               break;
+                       case INST_STSWI:
+                       case INST_STSWX:
+                               val = regs->gpr[rT] >> shift;
+                               if (put_user(val, (u8 __user *)EA))
+                                       return -EFAULT;
+                               break;
+               }
+               /* move EA to next address */
+               EA += 1;
+               num_bytes--;
+
+               /* manage our position within the register */
+               if (++pos == 4) {
+                       pos = 0;
+                       if (++rT == 32)
+                               rT = 0;
+               }
+       }
+
+       return 0;
+}
+
+static int emulate_instruction(struct pt_regs *regs)
+{
+       u32 instword;
+       u32 rd;
+
+       if (!user_mode(regs))
+               return -EINVAL;
+       CHECK_FULL_REGS(regs);
+
+       if (get_user(instword, (u32 __user *)(regs->nip)))
+               return -EFAULT;
+
+       /* Emulate the mfspr rD, PVR. */
+       if ((instword & INST_MFSPR_PVR_MASK) == INST_MFSPR_PVR) {
+               rd = (instword >> 21) & 0x1f;
+               regs->gpr[rd] = mfspr(SPRN_PVR);
+               return 0;
+       }
+
+       /* Emulating the dcba insn is just a no-op.  */
+       if ((instword & INST_DCBA_MASK) == INST_DCBA)
+               return 0;
+
+       /* Emulate the mcrxr insn.  */
+       if ((instword & INST_MCRXR_MASK) == INST_MCRXR) {
+               int shift = (instword >> 21) & 0x1c;
+               unsigned long msk = 0xf0000000UL >> shift;
+
+               regs->ccr = (regs->ccr & ~msk) | ((regs->xer >> shift) & msk);
+               regs->xer &= ~0xf0000000UL;
+               return 0;
+       }
+
+       /* Emulate load/store string insn. */
+       if ((instword & INST_STRING_GEN_MASK) == INST_STRING)
+               return emulate_string_inst(regs, instword);
+
+       return -EINVAL;
+}
+
+/*
+ * Look through the list of trap instructions that are used for BUG(),
+ * BUG_ON() and WARN_ON() and see if we hit one.  At this point we know
+ * that the exception was caused by a trap instruction of some kind.
+ * Returns 1 if we should continue (i.e. it was a WARN_ON) or 0
+ * otherwise.
+ */
+extern struct bug_entry __start___bug_table[], __stop___bug_table[];
+
+#ifndef CONFIG_MODULES
+#define module_find_bug(x)     NULL
+#endif
+
+struct bug_entry *find_bug(unsigned long bugaddr)
+{
+       struct bug_entry *bug;
+
+       for (bug = __start___bug_table; bug < __stop___bug_table; ++bug)
+               if (bugaddr == bug->bug_addr)
+                       return bug;
+       return module_find_bug(bugaddr);
+}
+
+int check_bug_trap(struct pt_regs *regs)
+{
+       struct bug_entry *bug;
+       unsigned long addr;
+
+       if (regs->msr & MSR_PR)
+               return 0;       /* not in kernel */
+       addr = regs->nip;       /* address of trap instruction */
+       if (addr < PAGE_OFFSET)
+               return 0;
+       bug = find_bug(regs->nip);
+       if (bug == NULL)
+               return 0;
+       if (bug->line & BUG_WARNING_TRAP) {
+               /* this is a WARN_ON rather than BUG/BUG_ON */
+#ifdef CONFIG_XMON
+               xmon_printf(KERN_ERR "Badness in %s at %s:%d\n",
+                      bug->function, bug->file,
+                      bug->line & ~BUG_WARNING_TRAP);
+#endif /* CONFIG_XMON */               
+               printk(KERN_ERR "Badness in %s at %s:%d\n",
+                      bug->function, bug->file,
+                      bug->line & ~BUG_WARNING_TRAP);
+               dump_stack();
+               return 1;
+       }
+#ifdef CONFIG_XMON
+       xmon_printf(KERN_CRIT "kernel BUG in %s at %s:%d!\n",
+              bug->function, bug->file, bug->line);
+       xmon(regs);
+#endif /* CONFIG_XMON */
+       printk(KERN_CRIT "kernel BUG in %s at %s:%d!\n",
+              bug->function, bug->file, bug->line);
+
+       return 0;
+}
+
+void ProgramCheckException(struct pt_regs *regs)
+{
+       unsigned int reason = get_reason(regs);
+       extern int do_mathemu(struct pt_regs *regs);
+
+#ifdef CONFIG_MATH_EMULATION
+       /* (reason & REASON_ILLEGAL) would be the obvious thing here,
+        * but there seems to be a hardware bug on the 405GP (RevD)
+        * that means ESR is sometimes set incorrectly - either to
+        * ESR_DST (!?) or 0.  In the process of chasing this with the
+        * hardware people - not sure if it can happen on any illegal
+        * instruction or only on FP instructions, whether there is a
+        * pattern to occurences etc. -dgibson 31/Mar/2003 */
+       if (!(reason & REASON_TRAP) && do_mathemu(regs) == 0) {
+               emulate_single_step(regs);
+               return;
+       }
+#endif /* CONFIG_MATH_EMULATION */
+
+       if (reason & REASON_FP) {
+               /* IEEE FP exception */
+               int code = 0;
+               u32 fpscr;
+
+               /* We must make sure the FP state is consistent with
+                * our MSR_FP in regs
+                */
+               preempt_disable();
+               if (regs->msr & MSR_FP)
+                       giveup_fpu(current);
+               preempt_enable();
+
+               fpscr = current->thread.fpscr;
+               fpscr &= fpscr << 22;   /* mask summary bits with enables */
+               if (fpscr & FPSCR_VX)
+                       code = FPE_FLTINV;
+               else if (fpscr & FPSCR_OX)
+                       code = FPE_FLTOVF;
+               else if (fpscr & FPSCR_UX)
+                       code = FPE_FLTUND;
+               else if (fpscr & FPSCR_ZX)
+                       code = FPE_FLTDIV;
+               else if (fpscr & FPSCR_XX)
+                       code = FPE_FLTRES;
+               _exception(SIGFPE, regs, code, regs->nip);
+               return;
+       }
+
+       if (reason & REASON_TRAP) {
+               /* trap exception */
+               if (debugger_bpt(regs))
+                       return;
+               if (check_bug_trap(regs)) {
+                       regs->nip += 4;
+                       return;
+               }
+               _exception(SIGTRAP, regs, TRAP_BRKPT, 0);
+               return;
+       }
+
+       /* Try to emulate it if we should. */
+       if (reason & (REASON_ILLEGAL | REASON_PRIVILEGED)) {
+               switch (emulate_instruction(regs)) {
+               case 0:
+                       regs->nip += 4;
+                       emulate_single_step(regs);
+                       return;
+               case -EFAULT:
+                       _exception(SIGSEGV, regs, SEGV_MAPERR, regs->nip);
+                       return;
+               }
+       }
+
+       if (reason & REASON_PRIVILEGED)
+               _exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
+       else
+               _exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
+}
+
+void AlignmentException(struct pt_regs *regs)
+{
+       int fixed;
+
+       fixed = fix_alignment(regs);
+
+       if (fixed == 1) {
+               regs->nip += 4; /* skip over emulated instruction */
+               emulate_single_step(regs);
+               return;
+       }
+
+       /* Operand address was bad */   
+       if (fixed == -EFAULT) {
+               if (user_mode(regs))
+                       _exception(SIGSEGV, regs, SEGV_ACCERR, regs->dar);
+               else
+                       /* Search exception table */
+                       bad_page_fault(regs, regs->dar, SIGSEGV);
+               return;
+       }
+       _exception(SIGBUS, regs, BUS_ADRALN, regs->dar);
+}
+
+void StackOverflow(struct pt_regs *regs)
+{
+       printk(KERN_CRIT "Kernel stack overflow in process %p, r1=%lx\n",
+              current, regs->gpr[1]);
+       debugger(regs);
+       show_regs(regs);
+       panic("kernel stack overflow");
+}
+
+void nonrecoverable_exception(struct pt_regs *regs)
+{
+       printk(KERN_ERR "Non-recoverable exception at PC=%lx MSR=%lx\n",
+              regs->nip, regs->msr);
+       debugger(regs);
+       die("nonrecoverable exception", regs, SIGKILL);
+}
+
+void trace_syscall(struct pt_regs *regs)
+{
+       printk("Task: %p(%d), PC: %08lX/%08lX, Syscall: %3ld, Result: %s%ld    %s\n",
+              current, current->pid, regs->nip, regs->link, regs->gpr[0],
+              regs->ccr&0x10000000?"Error=":"", regs->gpr[3], print_tainted());
+}
+
+#ifdef CONFIG_8xx
+void SoftwareEmulation(struct pt_regs *regs)
+{
+       extern int do_mathemu(struct pt_regs *);
+       extern int Soft_emulate_8xx(struct pt_regs *);
+       int errcode;
+
+       CHECK_FULL_REGS(regs);
+
+       if (!user_mode(regs)) {
+               debugger(regs);
+               die("Kernel Mode Software FPU Emulation", regs, SIGFPE);
+       }
+
+#ifdef CONFIG_MATH_EMULATION
+       errcode = do_mathemu(regs);
+#else
+       errcode = Soft_emulate_8xx(regs);
+#endif
+       if (errcode) {
+               if (errcode > 0)
+                       _exception(SIGFPE, regs, 0, 0);
+               else if (errcode == -EFAULT)
+                       _exception(SIGSEGV, regs, 0, 0);
+               else
+                       _exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
+       } else
+               emulate_single_step(regs);
+}
+#endif /* CONFIG_8xx */
+
+#if defined(CONFIG_40x) || defined(CONFIG_BOOKE)
+
+void DebugException(struct pt_regs *regs, unsigned long debug_status)
+{
+       if (debug_status & DBSR_IC) {   /* instruction completion */
+               regs->msr &= ~MSR_DE;
+               if (user_mode(regs)) {
+                       current->thread.dbcr0 &= ~DBCR0_IC;
+               } else {
+                       /* Disable instruction completion */
+                       mtspr(SPRN_DBCR0, mfspr(SPRN_DBCR0) & ~DBCR0_IC);
+                       /* Clear the instruction completion event */
+                       mtspr(SPRN_DBSR, DBSR_IC);
+                       if (debugger_sstep(regs))
+                               return;
+               }
+               _exception(SIGTRAP, regs, TRAP_TRACE, 0);
+       }
+}
+#endif /* CONFIG_4xx || CONFIG_BOOKE */
+
+#if !defined(CONFIG_TAU_INT)
+void TAUException(struct pt_regs *regs)
+{
+       printk("TAU trap at PC: %lx, MSR: %lx, vector=%lx    %s\n",
+              regs->nip, regs->msr, regs->trap, print_tainted());
+}
+#endif /* CONFIG_INT_TAU */
+
+void AltivecUnavailException(struct pt_regs *regs)
+{
+       static int kernel_altivec_count;
+
+#ifndef CONFIG_ALTIVEC
+       if (user_mode(regs)) {
+               /* A user program has executed an altivec instruction,
+                  but this kernel doesn't support altivec. */
+               _exception(SIGILL, regs, ILL_ILLOPC, regs->nip);
+               return;
+       }
+#endif
+       /* The kernel has executed an altivec instruction without
+          first enabling altivec.  Whinge but let it do it. */
+       if (++kernel_altivec_count < 10)
+               printk(KERN_ERR "AltiVec used in kernel (task=%p, pc=%lx)\n",
+                      current, regs->nip);
+       regs->msr |= MSR_VEC;
+}
+
+#ifdef CONFIG_ALTIVEC
+void AltivecAssistException(struct pt_regs *regs)
+{
+       int err;
+
+       preempt_disable();
+       if (regs->msr & MSR_VEC)
+               giveup_altivec(current);
+       preempt_enable();
+       if (!user_mode(regs)) {
+               printk(KERN_EMERG "VMX/Altivec assist exception in kernel mode"
+                      " at %lx\n", regs->nip);
+               die("Kernel Altivec assist exception", regs, SIGILL);
+       }
+
+       err = emulate_altivec(regs);
+       if (err == 0) {
+               regs->nip += 4;         /* skip emulated instruction */
+               emulate_single_step(regs);
+               return;
+       }
+
+       if (err == -EFAULT) {
+               /* got an error reading the instruction */
+               _exception(SIGSEGV, regs, SEGV_ACCERR, regs->nip);
+       } else {
+               /* didn't recognize the instruction */
+               /* XXX quick hack for now: set the non-Java bit in the VSCR */
+               if (printk_ratelimit())
+                       printk(KERN_ERR "Unrecognized altivec instruction "
+                              "in %s at %lx\n", current->comm, regs->nip);
+               current->thread.vscr.u[3] |= 0x10000;
+       }
+}
+#endif /* CONFIG_ALTIVEC */
+
+#ifdef CONFIG_E500
+void PerformanceMonitorException(struct pt_regs *regs)
+{
+       perf_irq(regs);
+}
+#endif
+
+#ifdef CONFIG_FSL_BOOKE
+void CacheLockingException(struct pt_regs *regs, unsigned long address,
+                          unsigned long error_code)
+{
+       /* We treat cache locking instructions from the user
+        * as priv ops, in the future we could try to do
+        * something smarter
+        */
+       if (error_code & (ESR_DLK|ESR_ILK))
+               _exception(SIGILL, regs, ILL_PRVOPC, regs->nip);
+       return;
+}
+#endif /* CONFIG_FSL_BOOKE */
+
+#ifdef CONFIG_SPE
+void SPEFloatingPointException(struct pt_regs *regs)
+{
+       unsigned long spefscr;
+       int fpexc_mode;
+       int code = 0;
+
+       spefscr = current->thread.spefscr;
+       fpexc_mode = current->thread.fpexc_mode;
+
+       /* Hardware does not neccessarily set sticky
+        * underflow/overflow/invalid flags */
+       if ((spefscr & SPEFSCR_FOVF) && (fpexc_mode & PR_FP_EXC_OVF)) {
+               code = FPE_FLTOVF;
+               spefscr |= SPEFSCR_FOVFS;
+       }
+       else if ((spefscr & SPEFSCR_FUNF) && (fpexc_mode & PR_FP_EXC_UND)) {
+               code = FPE_FLTUND;
+               spefscr |= SPEFSCR_FUNFS;
+       }
+       else if ((spefscr & SPEFSCR_FDBZ) && (fpexc_mode & PR_FP_EXC_DIV))
+               code = FPE_FLTDIV;
+       else if ((spefscr & SPEFSCR_FINV) && (fpexc_mode & PR_FP_EXC_INV)) {
+               code = FPE_FLTINV;
+               spefscr |= SPEFSCR_FINVS;
+       }
+       else if ((spefscr & (SPEFSCR_FG | SPEFSCR_FX)) && (fpexc_mode & PR_FP_EXC_RES))
+               code = FPE_FLTRES;
+
+       current->thread.spefscr = spefscr;
+
+       _exception(SIGFPE, regs, code, regs->nip);
+       return;
+}
+#endif
+
+#ifdef CONFIG_BOOKE_WDT
+/*
+ * Default handler for a Watchdog exception,
+ * spins until a reboot occurs
+ */
+void __attribute__ ((weak)) WatchdogHandler(struct pt_regs *regs)
+{
+       /* Generic WatchdogHandler, implement your own */
+       mtspr(SPRN_TCR, mfspr(SPRN_TCR)&(~TCR_WIE));
+       return;
+}
+
+void WatchdogException(struct pt_regs *regs)
+{
+       printk (KERN_EMERG "PowerPC Book-E Watchdog Exception\n");
+       WatchdogHandler(regs);
+}
+#endif
+
+void __init trap_init(void)
+{
+}
diff --git a/arch/powerpc/kernel/vector.S b/arch/powerpc/kernel/vector.S
new file mode 100644 (file)
index 0000000..12cb90b
--- /dev/null
@@ -0,0 +1,197 @@
+#include <linux/config.h>
+#include <asm/ppc_asm.h>
+#include <asm/processor.h>
+
+/*
+ * The routines below are in assembler so we can closely control the
+ * usage of floating-point registers.  These routines must be called
+ * with preempt disabled.
+ */
+#ifdef CONFIG_PPC32
+       .data
+fpzero:
+       .long   0
+fpone:
+       .long   0x3f800000      /* 1.0 in single-precision FP */
+fphalf:
+       .long   0x3f000000      /* 0.5 in single-precision FP */
+
+#define LDCONST(fr, name)      \
+       lis     r11,name@ha;    \
+       lfs     fr,name@l(r11)
+#else
+
+       .section ".toc","aw"
+fpzero:
+       .tc     FD_0_0[TC],0
+fpone:
+       .tc     FD_3ff00000_0[TC],0x3ff0000000000000    /* 1.0 */
+fphalf:
+       .tc     FD_3fe00000_0[TC],0x3fe0000000000000    /* 0.5 */
+
+#define LDCONST(fr, name)      \
+       lfd     fr,name@toc(r2)
+#endif
+
+       .text
+/*
+ * Internal routine to enable floating point and set FPSCR to 0.
+ * Don't call it from C; it doesn't use the normal calling convention.
+ */
+fpenable:
+#ifdef CONFIG_PPC32
+       stwu    r1,-64(r1)
+#else
+       stdu    r1,-64(r1)
+#endif
+       mfmsr   r10
+       ori     r11,r10,MSR_FP
+       mtmsr   r11
+       isync
+       stfd    fr0,24(r1)
+       stfd    fr1,16(r1)
+       stfd    fr31,8(r1)
+       LDCONST(fr1, fpzero)
+       mffs    fr31
+       mtfsf   0xff,fr1
+       blr
+
+fpdisable:
+       mtlr    r12
+       mtfsf   0xff,fr31
+       lfd     fr31,8(r1)
+       lfd     fr1,16(r1)
+       lfd     fr0,24(r1)
+       mtmsr   r10
+       isync
+       addi    r1,r1,64
+       blr
+
+/*
+ * Vector add, floating point.
+ */
+_GLOBAL(vaddfp)
+       mflr    r12
+       bl      fpenable
+       li      r0,4
+       mtctr   r0
+       li      r6,0
+1:     lfsx    fr0,r4,r6
+       lfsx    fr1,r5,r6
+       fadds   fr0,fr0,fr1
+       stfsx   fr0,r3,r6
+       addi    r6,r6,4
+       bdnz    1b
+       b       fpdisable
+
+/*
+ * Vector subtract, floating point.
+ */
+_GLOBAL(vsubfp)
+       mflr    r12
+       bl      fpenable
+       li      r0,4
+       mtctr   r0
+       li      r6,0
+1:     lfsx    fr0,r4,r6
+       lfsx    fr1,r5,r6
+       fsubs   fr0,fr0,fr1
+       stfsx   fr0,r3,r6
+       addi    r6,r6,4
+       bdnz    1b
+       b       fpdisable
+
+/*
+ * Vector multiply and add, floating point.
+ */
+_GLOBAL(vmaddfp)
+       mflr    r12
+       bl      fpenable
+       stfd    fr2,32(r1)
+       li      r0,4
+       mtctr   r0
+       li      r7,0
+1:     lfsx    fr0,r4,r7
+       lfsx    fr1,r5,r7
+       lfsx    fr2,r6,r7
+       fmadds  fr0,fr0,fr2,fr1
+       stfsx   fr0,r3,r7
+       addi    r7,r7,4
+       bdnz    1b
+       lfd     fr2,32(r1)
+       b       fpdisable
+
+/*
+ * Vector negative multiply and subtract, floating point.
+ */
+_GLOBAL(vnmsubfp)
+       mflr    r12
+       bl      fpenable
+       stfd    fr2,32(r1)
+       li      r0,4
+       mtctr   r0
+       li      r7,0
+1:     lfsx    fr0,r4,r7
+       lfsx    fr1,r5,r7
+       lfsx    fr2,r6,r7
+       fnmsubs fr0,fr0,fr2,fr1
+       stfsx   fr0,r3,r7
+       addi    r7,r7,4
+       bdnz    1b
+       lfd     fr2,32(r1)
+       b       fpdisable
+
+/*
+ * Vector reciprocal estimate.  We just compute 1.0/x.
+ * r3 -> destination, r4 -> source.
+ */
+_GLOBAL(vrefp)
+       mflr    r12
+       bl      fpenable
+       li      r0,4
+       LDCONST(fr1, fpone)
+       mtctr   r0
+       li      r6,0
+1:     lfsx    fr0,r4,r6
+       fdivs   fr0,fr1,fr0
+       stfsx   fr0,r3,r6
+       addi    r6,r6,4
+       bdnz    1b
+       b       fpdisable
+
+/*
+ * Vector reciprocal square-root estimate, floating point.
+ * We use the frsqrte instruction for the initial estimate followed
+ * by 2 iterations of Newton-Raphson to get sufficient accuracy.
+ * r3 -> destination, r4 -> source.
+ */
+_GLOBAL(vrsqrtefp)
+       mflr    r12
+       bl      fpenable
+       stfd    fr2,32(r1)
+       stfd    fr3,40(r1)
+       stfd    fr4,48(r1)
+       stfd    fr5,56(r1)
+       li      r0,4
+       LDCONST(fr4, fpone)
+       LDCONST(fr5, fphalf)
+       mtctr   r0
+       li      r6,0
+1:     lfsx    fr0,r4,r6
+       frsqrte fr1,fr0         /* r = frsqrte(s) */
+       fmuls   fr3,fr1,fr0     /* r * s */
+       fmuls   fr2,fr1,fr5     /* r * 0.5 */
+       fnmsubs fr3,fr1,fr3,fr4 /* 1 - s * r * r */
+       fmadds  fr1,fr2,fr3,fr1 /* r = r + 0.5 * r * (1 - s * r * r) */
+       fmuls   fr3,fr1,fr0     /* r * s */
+       fmuls   fr2,fr1,fr5     /* r * 0.5 */
+       fnmsubs fr3,fr1,fr3,fr4 /* 1 - s * r * r */
+       fmadds  fr1,fr2,fr3,fr1 /* r = r + 0.5 * r * (1 - s * r * r) */
+       stfsx   fr1,r3,r6
+       addi    r6,r6,4
+       bdnz    1b
+       lfd     fr5,56(r1)
+       lfd     fr4,48(r1)
+       lfd     fr3,40(r1)
+       lfd     fr2,32(r1)
+       b       fpdisable
diff --git a/arch/powerpc/kernel/vmlinux.lds b/arch/powerpc/kernel/vmlinux.lds
new file mode 100644 (file)
index 0000000..d62c288
--- /dev/null
@@ -0,0 +1,174 @@
+/* Align . to a 8 byte boundary equals to maximum function alignment. */
+/* sched.text is aling to function alignment to secure we have same
+ * address even at second ld pass when generating System.map */
+/* spinlock.text is aling to function alignment to secure we have same
+ * address even at second ld pass when generating System.map */
+  /* DWARF debug sections.
+               Symbols in the DWARF debugging sections are relative to
+               the beginning of the section so we begin them at 0.  */
+  /* Stabs debugging sections.  */
+OUTPUT_ARCH(powerpc:common)
+jiffies = jiffies_64 + 4;
+SECTIONS
+{
+  /* Read-only sections, merged into text segment: */
+  . = + SIZEOF_HEADERS;
+  .interp : { *(.interp) }
+  .hash : { *(.hash) }
+  .dynsym : { *(.dynsym) }
+  .dynstr : { *(.dynstr) }
+  .rel.text : { *(.rel.text) }
+  .rela.text : { *(.rela.text) }
+  .rel.data : { *(.rel.data) }
+  .rela.data : { *(.rela.data) }
+  .rel.rodata : { *(.rel.rodata) }
+  .rela.rodata : { *(.rela.rodata) }
+  .rel.got : { *(.rel.got) }
+  .rela.got : { *(.rela.got) }
+  .rel.ctors : { *(.rel.ctors) }
+  .rela.ctors : { *(.rela.ctors) }
+  .rel.dtors : { *(.rel.dtors) }
+  .rela.dtors : { *(.rela.dtors) }
+  .rel.bss : { *(.rel.bss) }
+  .rela.bss : { *(.rela.bss) }
+  .rel.plt : { *(.rel.plt) }
+  .rela.plt : { *(.rela.plt) }
+/*  .init          : { *(.init)        } =0*/
+  .plt : { *(.plt) }
+  .text :
+  {
+    *(.text)
+    . = ALIGN(8); __sched_text_start = .; *(.sched.text) __sched_text_end = .;
+    . = ALIGN(8); __lock_text_start = .; *(.spinlock.text) __lock_text_end = .;
+    *(.fixup)
+    *(.got1)
+    __got2_start = .;
+    *(.got2)
+    __got2_end = .;
+  }
+  _etext = .;
+  PROVIDE (etext = .);
+  .rodata : AT(ADDR(.rodata) - 0) { *(.rodata) *(.rodata.*) *(__vermagic) } .rodata1 : AT(ADDR(.rodata1) - 0) { *(.rodata1) } .pci_fixup : AT(ADDR(.pci_fixup) - 0) { __start_pci_fixups_early = .; *(.pci_fixup_early) __end_pci_fixups_early = .; __start_pci_fixups_header = .; *(.pci_fixup_header) __end_pci_fixups_header = .; __start_pci_fixups_final = .; *(.pci_fixup_final) __end_pci_fixups_final = .; __start_pci_fixups_enable = .; *(.pci_fixup_enable) __end_pci_fixups_enable = .; } __ksymtab : AT(ADDR(__ksymtab) - 0) { __start___ksymtab = .; *(__ksymtab) __stop___ksymtab = .; } __ksymtab_gpl : AT(ADDR(__ksymtab_gpl) - 0) { __start___ksymtab_gpl = .; *(__ksymtab_gpl) __stop___ksymtab_gpl = .; } __kcrctab : AT(ADDR(__kcrctab) - 0) { __start___kcrctab = .; *(__kcrctab) __stop___kcrctab = .; } __kcrctab_gpl : AT(ADDR(__kcrctab_gpl) - 0) { __start___kcrctab_gpl = .; *(__kcrctab_gpl) __stop___kcrctab_gpl = .; } __ksymtab_strings : AT(ADDR(__ksymtab_strings) - 0) { *(__ksymtab_strings) } __param : AT(ADDR(__param) - 0) { __start___param = .; *(__param) __stop___param = .; }
+  .fini : { *(.fini) } =0
+  .ctors : { *(.ctors) }
+  .dtors : { *(.dtors) }
+  .fixup : { *(.fixup) }
+ __ex_table : {
+  __start___ex_table = .;
+  *(__ex_table)
+  __stop___ex_table = .;
+ }
+ __bug_table : {
+  __start___bug_table = .;
+  *(__bug_table)
+  __stop___bug_table = .;
+ }
+  /* Read-write section, merged into data segment: */
+  . = ALIGN(4096);
+  .data :
+  {
+    *(.data)
+    *(.data1)
+    *(.sdata)
+    *(.sdata2)
+    *(.got.plt) *(.got)
+    *(.dynamic)
+    CONSTRUCTORS
+  }
+
+  . = ALIGN(4096);
+  __nosave_begin = .;
+  .data_nosave : { *(.data.nosave) }
+  . = ALIGN(4096);
+  __nosave_end = .;
+
+  . = ALIGN(32);
+  .data.cacheline_aligned : { *(.data.cacheline_aligned) }
+
+  _edata = .;
+  PROVIDE (edata = .);
+
+  . = ALIGN(8192);
+  .data.init_task : { *(.data.init_task) }
+
+  . = ALIGN(4096);
+  __init_begin = .;
+  .init.text : {
+ _sinittext = .;
+ *(.init.text)
+ _einittext = .;
+  }
+  /* .exit.text is discarded at runtime, not link time,
+     to deal with references from __bug_table */
+  .exit.text : { *(.exit.text) }
+  .init.data : {
+    *(.init.data);
+    __vtop_table_begin = .;
+    *(.vtop_fixup);
+    __vtop_table_end = .;
+    __ptov_table_begin = .;
+    *(.ptov_fixup);
+    __ptov_table_end = .;
+  }
+  . = ALIGN(16);
+  __setup_start = .;
+  .init.setup : { *(.init.setup) }
+  __setup_end = .;
+  __initcall_start = .;
+  .initcall.init : {
+ *(.initcall1.init)
+ *(.initcall2.init)
+ *(.initcall3.init)
+ *(.initcall4.init)
+ *(.initcall5.init)
+ *(.initcall6.init)
+ *(.initcall7.init)
+  }
+  __initcall_end = .;
+
+  __con_initcall_start = .;
+  .con_initcall.init : { *(.con_initcall.init) }
+  __con_initcall_end = .;
+
+  .security_initcall.init : AT(ADDR(.security_initcall.init) - 0) { __security_initcall_start = .; *(.security_initcall.init) __security_initcall_end = .; }
+
+  __start___ftr_fixup = .;
+  __ftr_fixup : { *(__ftr_fixup) }
+  __stop___ftr_fixup = .;
+
+  . = ALIGN(32);
+  __per_cpu_start = .;
+  .data.percpu : { *(.data.percpu) }
+  __per_cpu_end = .;
+
+  . = ALIGN(4096);
+  __initramfs_start = .;
+  .init.ramfs : { *(.init.ramfs) }
+  __initramfs_end = .;
+
+  . = ALIGN(4096);
+  __init_end = .;
+
+  . = ALIGN(4096);
+  _sextratext = .;
+  _eextratext = .;
+
+  __bss_start = .;
+  .bss :
+  {
+   *(.sbss) *(.scommon)
+   *(.dynbss)
+   *(.bss)
+   *(COMMON)
+  }
+  __bss_stop = .;
+
+  _end = . ;
+  PROVIDE (end = .);
+
+  /* Sections to be discarded. */
+  /DISCARD/ : {
+    *(.exitcall.exit)
+    *(.exit.data)
+  }
+}
diff --git a/arch/powerpc/kernel/vmlinux.lds.S b/arch/powerpc/kernel/vmlinux.lds.S
new file mode 100644 (file)
index 0000000..09c6525
--- /dev/null
@@ -0,0 +1,172 @@
+#include <asm-generic/vmlinux.lds.h>
+
+OUTPUT_ARCH(powerpc:common)
+jiffies = jiffies_64 + 4;
+SECTIONS
+{
+  /* Read-only sections, merged into text segment: */
+  . = + SIZEOF_HEADERS;
+  .interp : { *(.interp) }
+  .hash          : { *(.hash)          }
+  .dynsym        : { *(.dynsym)                }
+  .dynstr        : { *(.dynstr)                }
+  .rel.text      : { *(.rel.text)              }
+  .rela.text     : { *(.rela.text)     }
+  .rel.data      : { *(.rel.data)              }
+  .rela.data     : { *(.rela.data)     }
+  .rel.rodata    : { *(.rel.rodata)    }
+  .rela.rodata   : { *(.rela.rodata)   }
+  .rel.got       : { *(.rel.got)               }
+  .rela.got      : { *(.rela.got)              }
+  .rel.ctors     : { *(.rel.ctors)     }
+  .rela.ctors    : { *(.rela.ctors)    }
+  .rel.dtors     : { *(.rel.dtors)     }
+  .rela.dtors    : { *(.rela.dtors)    }
+  .rel.bss       : { *(.rel.bss)               }
+  .rela.bss      : { *(.rela.bss)              }
+  .rel.plt       : { *(.rel.plt)               }
+  .rela.plt      : { *(.rela.plt)              }
+/*  .init          : { *(.init)        } =0*/
+  .plt : { *(.plt) }
+  .text      :
+  {
+    *(.text)
+    SCHED_TEXT
+    LOCK_TEXT
+    *(.fixup)
+    *(.got1)
+    __got2_start = .;
+    *(.got2)
+    __got2_end = .;
+  }
+  _etext = .;
+  PROVIDE (etext = .);
+
+  RODATA
+  .fini      : { *(.fini)    } =0
+  .ctors     : { *(.ctors)   }
+  .dtors     : { *(.dtors)   }
+
+  .fixup   : { *(.fixup) }
+
+       __ex_table : {
+               __start___ex_table = .;
+               *(__ex_table)
+               __stop___ex_table = .;
+       }
+
+       __bug_table : {
+               __start___bug_table = .;
+               *(__bug_table)
+               __stop___bug_table = .;
+       }
+
+  /* Read-write section, merged into data segment: */
+  . = ALIGN(4096);
+  .data    :
+  {
+    *(.data)
+    *(.data1)
+    *(.sdata)
+    *(.sdata2)
+    *(.got.plt) *(.got)
+    *(.dynamic)
+    CONSTRUCTORS
+  }
+
+  . = ALIGN(4096);
+  __nosave_begin = .;
+  .data_nosave : { *(.data.nosave) }
+  . = ALIGN(4096);
+  __nosave_end = .;
+
+  . = ALIGN(32);
+  .data.cacheline_aligned : { *(.data.cacheline_aligned) }
+
+  _edata  =  .;
+  PROVIDE (edata = .);
+
+  . = ALIGN(8192);
+  .data.init_task : { *(.data.init_task) }
+
+  . = ALIGN(4096);
+  __init_begin = .;
+  .init.text : {
+       _sinittext = .;
+       *(.init.text)
+       _einittext = .;
+  }
+  /* .exit.text is discarded at runtime, not link time,
+     to deal with references from __bug_table */
+  .exit.text : { *(.exit.text) }
+  .init.data : {
+    *(.init.data);
+    __vtop_table_begin = .;
+    *(.vtop_fixup);
+    __vtop_table_end = .;
+    __ptov_table_begin = .;
+    *(.ptov_fixup);
+    __ptov_table_end = .;
+  }
+  . = ALIGN(16);
+  __setup_start = .;
+  .init.setup : { *(.init.setup) }
+  __setup_end = .;
+  __initcall_start = .;
+  .initcall.init : {
+       *(.initcall1.init)
+       *(.initcall2.init)
+       *(.initcall3.init)
+       *(.initcall4.init)
+       *(.initcall5.init)
+       *(.initcall6.init)
+       *(.initcall7.init)
+  }
+  __initcall_end = .;
+
+  __con_initcall_start = .;
+  .con_initcall.init : { *(.con_initcall.init) }
+  __con_initcall_end = .;
+
+  SECURITY_INIT
+
+  __start___ftr_fixup = .;
+  __ftr_fixup : { *(__ftr_fixup) }
+  __stop___ftr_fixup = .;
+
+  . = ALIGN(32);
+  __per_cpu_start = .;
+  .data.percpu  : { *(.data.percpu) }
+  __per_cpu_end = .;
+
+  . = ALIGN(4096);
+  __initramfs_start = .;
+  .init.ramfs : { *(.init.ramfs) }
+  __initramfs_end = .;
+
+  . = ALIGN(4096);
+  __init_end = .;
+
+  . = ALIGN(4096);
+  _sextratext = .;
+  _eextratext = .;
+
+  __bss_start = .;
+  .bss       :
+  {
+   *(.sbss) *(.scommon)
+   *(.dynbss)
+   *(.bss)
+   *(COMMON)
+  }
+  __bss_stop = .;
+
+  _end = . ;
+  PROVIDE (end = .);
+
+  /* Sections to be discarded. */
+  /DISCARD/ : {
+    *(.exitcall.exit)
+    *(.exit.data)
+  }
+}
diff --git a/arch/powerpc/lib/Makefile b/arch/powerpc/lib/Makefile
new file mode 100644 (file)
index 0000000..347f979
--- /dev/null
@@ -0,0 +1,9 @@
+#
+# Makefile for ppc-specific library files..
+#
+
+obj-y                  := strcase.o string.o
+obj-$(CONFIG_PPC32)    += div64.o copy32.o checksum.o
+obj-$(CONFIG_PPC64)    += copypage.o copyuser.o memcpy.o usercopy.o \
+                          sstep.o checksum64.o
+obj-$(CONFIG_PPC_ISERIES) += e2a.o
diff --git a/arch/powerpc/lib/checksum.S b/arch/powerpc/lib/checksum.S
new file mode 100644 (file)
index 0000000..7874e8a
--- /dev/null
@@ -0,0 +1,225 @@
+/*
+ * This file contains assembly-language implementations
+ * of IP-style 1's complement checksum routines.
+ *     
+ *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
+ *
+ *  This program is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU General Public License
+ *  as published by the Free Software Foundation; either version
+ *  2 of the License, or (at your option) any later version.
+ *
+ * Severely hacked about by Paul Mackerras (paulus@cs.anu.edu.au).
+ */
+
+#include <linux/sys.h>
+#include <asm/processor.h>
+#include <asm/errno.h>
+#include <asm/ppc_asm.h>
+
+       .text
+
+/*
+ * ip_fast_csum(buf, len) -- Optimized for IP header
+ * len is in words and is always >= 5.
+ */
+_GLOBAL(ip_fast_csum)
+       lwz     r0,0(r3)
+       lwzu    r5,4(r3)
+       addic.  r4,r4,-2
+       addc    r0,r0,r5
+       mtctr   r4
+       blelr-
+1:     lwzu    r4,4(r3)
+       adde    r0,r0,r4
+       bdnz    1b
+       addze   r0,r0           /* add in final carry */
+       rlwinm  r3,r0,16,0,31   /* fold two halves together */
+       add     r3,r0,r3
+       not     r3,r3
+       srwi    r3,r3,16
+       blr
+
+/*
+ * Compute checksum of TCP or UDP pseudo-header:
+ *   csum_tcpudp_magic(saddr, daddr, len, proto, sum)
+ */    
+_GLOBAL(csum_tcpudp_magic)
+       rlwimi  r5,r6,16,0,15   /* put proto in upper half of len */
+       addc    r0,r3,r4        /* add 4 32-bit words together */
+       adde    r0,r0,r5
+       adde    r0,r0,r7
+       addze   r0,r0           /* add in final carry */
+       rlwinm  r3,r0,16,0,31   /* fold two halves together */
+       add     r3,r0,r3
+       not     r3,r3
+       srwi    r3,r3,16
+       blr
+
+/*
+ * computes the checksum of a memory block at buff, length len,
+ * and adds in "sum" (32-bit)
+ *
+ * csum_partial(buff, len, sum)
+ */
+_GLOBAL(csum_partial)
+       addic   r0,r5,0
+       subi    r3,r3,4
+       srwi.   r6,r4,2
+       beq     3f              /* if we're doing < 4 bytes */
+       andi.   r5,r3,2         /* Align buffer to longword boundary */
+       beq+    1f
+       lhz     r5,4(r3)        /* do 2 bytes to get aligned */
+       addi    r3,r3,2
+       subi    r4,r4,2
+       addc    r0,r0,r5
+       srwi.   r6,r4,2         /* # words to do */
+       beq     3f
+1:     mtctr   r6
+2:     lwzu    r5,4(r3)        /* the bdnz has zero overhead, so it should */
+       adde    r0,r0,r5        /* be unnecessary to unroll this loop */
+       bdnz    2b
+       andi.   r4,r4,3
+3:     cmpwi   0,r4,2
+       blt+    4f
+       lhz     r5,4(r3)
+       addi    r3,r3,2
+       subi    r4,r4,2
+       adde    r0,r0,r5
+4:     cmpwi   0,r4,1
+       bne+    5f
+       lbz     r5,4(r3)
+       slwi    r5,r5,8         /* Upper byte of word */
+       adde    r0,r0,r5
+5:     addze   r3,r0           /* add in final carry */
+       blr
+
+/*
+ * Computes the checksum of a memory block at src, length len,
+ * and adds in "sum" (32-bit), while copying the block to dst.
+ * If an access exception occurs on src or dst, it stores -EFAULT
+ * to *src_err or *dst_err respectively, and (for an error on
+ * src) zeroes the rest of dst.
+ *
+ * csum_partial_copy_generic(src, dst, len, sum, src_err, dst_err)
+ */
+_GLOBAL(csum_partial_copy_generic)
+       addic   r0,r6,0
+       subi    r3,r3,4
+       subi    r4,r4,4
+       srwi.   r6,r5,2
+       beq     3f              /* if we're doing < 4 bytes */
+       andi.   r9,r4,2         /* Align dst to longword boundary */
+       beq+    1f
+81:    lhz     r6,4(r3)        /* do 2 bytes to get aligned */
+       addi    r3,r3,2
+       subi    r5,r5,2
+91:    sth     r6,4(r4)
+       addi    r4,r4,2
+       addc    r0,r0,r6
+       srwi.   r6,r5,2         /* # words to do */
+       beq     3f
+1:     srwi.   r6,r5,4         /* # groups of 4 words to do */
+       beq     10f
+       mtctr   r6
+71:    lwz     r6,4(r3)
+72:    lwz     r9,8(r3)
+73:    lwz     r10,12(r3)
+74:    lwzu    r11,16(r3)
+       adde    r0,r0,r6
+75:    stw     r6,4(r4)
+       adde    r0,r0,r9
+76:    stw     r9,8(r4)
+       adde    r0,r0,r10
+77:    stw     r10,12(r4)
+       adde    r0,r0,r11
+78:    stwu    r11,16(r4)
+       bdnz    71b
+10:    rlwinm. r6,r5,30,30,31  /* # words left to do */
+       beq     13f
+       mtctr   r6
+82:    lwzu    r9,4(r3)
+92:    stwu    r9,4(r4)
+       adde    r0,r0,r9
+       bdnz    82b
+13:    andi.   r5,r5,3
+3:     cmpwi   0,r5,2
+       blt+    4f
+83:    lhz     r6,4(r3)
+       addi    r3,r3,2
+       subi    r5,r5,2
+93:    sth     r6,4(r4)
+       addi    r4,r4,2
+       adde    r0,r0,r6
+4:     cmpwi   0,r5,1
+       bne+    5f
+84:    lbz     r6,4(r3)
+94:    stb     r6,4(r4)
+       slwi    r6,r6,8         /* Upper byte of word */
+       adde    r0,r0,r6
+5:     addze   r3,r0           /* add in final carry */
+       blr
+
+/* These shouldn't go in the fixup section, since that would
+   cause the ex_table addresses to get out of order. */
+
+src_error_4:
+       mfctr   r6              /* update # bytes remaining from ctr */
+       rlwimi  r5,r6,4,0,27
+       b       79f
+src_error_1:
+       li      r6,0
+       subi    r5,r5,2
+95:    sth     r6,4(r4)
+       addi    r4,r4,2
+79:    srwi.   r6,r5,2
+       beq     3f
+       mtctr   r6
+src_error_2:
+       li      r6,0
+96:    stwu    r6,4(r4)
+       bdnz    96b
+3:     andi.   r5,r5,3
+       beq     src_error
+src_error_3:
+       li      r6,0
+       mtctr   r5
+       addi    r4,r4,3
+97:    stbu    r6,1(r4)
+       bdnz    97b
+src_error:
+       cmpwi   0,r7,0
+       beq     1f
+       li      r6,-EFAULT
+       stw     r6,0(r7)
+1:     addze   r3,r0
+       blr
+
+dst_error:
+       cmpwi   0,r8,0
+       beq     1f
+       li      r6,-EFAULT
+       stw     r6,0(r8)
+1:     addze   r3,r0
+       blr
+
+.section __ex_table,"a"
+       .long   81b,src_error_1
+       .long   91b,dst_error
+       .long   71b,src_error_4
+       .long   72b,src_error_4
+       .long   73b,src_error_4
+       .long   74b,src_error_4
+       .long   75b,dst_error
+       .long   76b,dst_error
+       .long   77b,dst_error
+       .long   78b,dst_error
+       .long   82b,src_error_2
+       .long   92b,dst_error
+       .long   83b,src_error_3
+       .long   93b,dst_error
+       .long   84b,src_error_3
+       .long   94b,dst_error
+       .long   95b,dst_error
+       .long   96b,dst_error
+       .long   97b,dst_error
diff --git a/arch/powerpc/lib/checksum64.S b/arch/powerpc/lib/checksum64.S
new file mode 100644 (file)
index 0000000..ef96c6c
--- /dev/null
@@ -0,0 +1,229 @@
+/*
+ * This file contains assembly-language implementations
+ * of IP-style 1's complement checksum routines.
+ *     
+ *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
+ *
+ *  This program is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU General Public License
+ *  as published by the Free Software Foundation; either version
+ *  2 of the License, or (at your option) any later version.
+ *
+ * Severely hacked about by Paul Mackerras (paulus@cs.anu.edu.au).
+ */
+
+#include <linux/sys.h>
+#include <asm/processor.h>
+#include <asm/errno.h>
+#include <asm/ppc_asm.h>
+
+/*
+ * ip_fast_csum(r3=buf, r4=len) -- Optimized for IP header
+ * len is in words and is always >= 5.
+ *
+ * In practice len == 5, but this is not guaranteed.  So this code does not
+ * attempt to use doubleword instructions.
+ */
+_GLOBAL(ip_fast_csum)
+       lwz     r0,0(r3)
+       lwzu    r5,4(r3)
+       addic.  r4,r4,-2
+       addc    r0,r0,r5
+       mtctr   r4
+       blelr-
+1:     lwzu    r4,4(r3)
+       adde    r0,r0,r4
+       bdnz    1b
+       addze   r0,r0           /* add in final carry */
+        rldicl  r4,r0,32,0      /* fold two 32-bit halves together */
+        add     r0,r0,r4
+        srdi    r0,r0,32
+       rlwinm  r3,r0,16,0,31   /* fold two halves together */
+       add     r3,r0,r3
+       not     r3,r3
+       srwi    r3,r3,16
+       blr
+
+/*
+ * Compute checksum of TCP or UDP pseudo-header:
+ *   csum_tcpudp_magic(r3=saddr, r4=daddr, r5=len, r6=proto, r7=sum)
+ * No real gain trying to do this specially for 64 bit, but
+ * the 32 bit addition may spill into the upper bits of
+ * the doubleword so we still must fold it down from 64.
+ */    
+_GLOBAL(csum_tcpudp_magic)
+       rlwimi  r5,r6,16,0,15   /* put proto in upper half of len */
+       addc    r0,r3,r4        /* add 4 32-bit words together */
+       adde    r0,r0,r5
+       adde    r0,r0,r7
+        rldicl  r4,r0,32,0      /* fold 64 bit value */
+        add     r0,r4,r0
+        srdi    r0,r0,32
+       rlwinm  r3,r0,16,0,31   /* fold two halves together */
+       add     r3,r0,r3
+       not     r3,r3
+       srwi    r3,r3,16
+       blr
+
+/*
+ * Computes the checksum of a memory block at buff, length len,
+ * and adds in "sum" (32-bit).
+ *
+ * This code assumes at least halfword alignment, though the length
+ * can be any number of bytes.  The sum is accumulated in r5.
+ *
+ * csum_partial(r3=buff, r4=len, r5=sum)
+ */
+_GLOBAL(csum_partial)
+        subi   r3,r3,8         /* we'll offset by 8 for the loads */
+        srdi.  r6,r4,3         /* divide by 8 for doubleword count */
+        addic   r5,r5,0         /* clear carry */
+        beq    3f              /* if we're doing < 8 bytes */
+        andi.  r0,r3,2         /* aligned on a word boundary already? */
+        beq+   1f
+        lhz     r6,8(r3)        /* do 2 bytes to get aligned */
+        addi    r3,r3,2
+        subi    r4,r4,2
+        addc    r5,r5,r6
+        srdi.   r6,r4,3         /* recompute number of doublewords */
+        beq     3f              /* any left? */
+1:      mtctr   r6
+2:      ldu     r6,8(r3)        /* main sum loop */
+        adde    r5,r5,r6
+        bdnz    2b
+        andi.  r4,r4,7         /* compute bytes left to sum after doublewords */
+3:     cmpwi   0,r4,4          /* is at least a full word left? */
+       blt     4f
+       lwz     r6,8(r3)        /* sum this word */
+       addi    r3,r3,4
+       subi    r4,r4,4
+       adde    r5,r5,r6
+4:     cmpwi   0,r4,2          /* is at least a halfword left? */
+        blt+   5f
+        lhz     r6,8(r3)        /* sum this halfword */
+        addi    r3,r3,2
+        subi    r4,r4,2
+        adde    r5,r5,r6
+5:     cmpwi   0,r4,1          /* is at least a byte left? */
+        bne+    6f
+        lbz     r6,8(r3)        /* sum this byte */
+        slwi    r6,r6,8         /* this byte is assumed to be the upper byte of a halfword */
+        adde    r5,r5,r6
+6:      addze  r5,r5           /* add in final carry */
+       rldicl  r4,r5,32,0      /* fold two 32-bit halves together */
+        add     r3,r4,r5
+        srdi    r3,r3,32
+        blr
+
+/*
+ * Computes the checksum of a memory block at src, length len,
+ * and adds in "sum" (32-bit), while copying the block to dst.
+ * If an access exception occurs on src or dst, it stores -EFAULT
+ * to *src_err or *dst_err respectively, and (for an error on
+ * src) zeroes the rest of dst.
+ *
+ * This code needs to be reworked to take advantage of 64 bit sum+copy.
+ * However, due to tokenring halfword alignment problems this will be very
+ * tricky.  For now we'll leave it until we instrument it somehow.
+ *
+ * csum_partial_copy_generic(r3=src, r4=dst, r5=len, r6=sum, r7=src_err, r8=dst_err)
+ */
+_GLOBAL(csum_partial_copy_generic)
+       addic   r0,r6,0
+       subi    r3,r3,4
+       subi    r4,r4,4
+       srwi.   r6,r5,2
+       beq     3f              /* if we're doing < 4 bytes */
+       andi.   r9,r4,2         /* Align dst to longword boundary */
+       beq+    1f
+81:    lhz     r6,4(r3)        /* do 2 bytes to get aligned */
+       addi    r3,r3,2
+       subi    r5,r5,2
+91:    sth     r6,4(r4)
+       addi    r4,r4,2
+       addc    r0,r0,r6
+       srwi.   r6,r5,2         /* # words to do */
+       beq     3f
+1:     mtctr   r6
+82:    lwzu    r6,4(r3)        /* the bdnz has zero overhead, so it should */
+92:    stwu    r6,4(r4)        /* be unnecessary to unroll this loop */
+       adde    r0,r0,r6
+       bdnz    82b
+       andi.   r5,r5,3
+3:     cmpwi   0,r5,2
+       blt+    4f
+83:    lhz     r6,4(r3)
+       addi    r3,r3,2
+       subi    r5,r5,2
+93:    sth     r6,4(r4)
+       addi    r4,r4,2
+       adde    r0,r0,r6
+4:     cmpwi   0,r5,1
+       bne+    5f
+84:    lbz     r6,4(r3)
+94:    stb     r6,4(r4)
+       slwi    r6,r6,8         /* Upper byte of word */
+       adde    r0,r0,r6
+5:     addze   r3,r0           /* add in final carry (unlikely with 64-bit regs) */
+        rldicl  r4,r3,32,0      /* fold 64 bit value */
+        add     r3,r4,r3
+        srdi    r3,r3,32
+       blr
+
+/* These shouldn't go in the fixup section, since that would
+   cause the ex_table addresses to get out of order. */
+
+       .globl src_error_1
+src_error_1:
+       li      r6,0
+       subi    r5,r5,2
+95:    sth     r6,4(r4)
+       addi    r4,r4,2
+       srwi.   r6,r5,2
+       beq     3f
+       mtctr   r6
+       .globl src_error_2
+src_error_2:
+       li      r6,0
+96:    stwu    r6,4(r4)
+       bdnz    96b
+3:     andi.   r5,r5,3
+       beq     src_error
+       .globl src_error_3
+src_error_3:
+       li      r6,0
+       mtctr   r5
+       addi    r4,r4,3
+97:    stbu    r6,1(r4)
+       bdnz    97b
+       .globl src_error
+src_error:
+       cmpdi   0,r7,0
+       beq     1f
+       li      r6,-EFAULT
+       stw     r6,0(r7)
+1:     addze   r3,r0
+       blr
+
+       .globl dst_error
+dst_error:
+       cmpdi   0,r8,0
+       beq     1f
+       li      r6,-EFAULT
+       stw     r6,0(r8)
+1:     addze   r3,r0
+       blr
+
+.section __ex_table,"a"
+       .align  3
+       .llong  81b,src_error_1
+       .llong  91b,dst_error
+       .llong  82b,src_error_2
+       .llong  92b,dst_error
+       .llong  83b,src_error_3
+       .llong  93b,dst_error
+       .llong  84b,src_error_3
+       .llong  94b,dst_error
+       .llong  95b,dst_error
+       .llong  96b,dst_error
+       .llong  97b,dst_error
diff --git a/arch/powerpc/lib/copy32.S b/arch/powerpc/lib/copy32.S
new file mode 100644 (file)
index 0000000..420a912
--- /dev/null
@@ -0,0 +1,543 @@
+/*
+ * Memory copy functions for 32-bit PowerPC.
+ *
+ * Copyright (C) 1996-2005 Paul Mackerras.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version
+ * 2 of the License, or (at your option) any later version.
+ */
+#include <linux/config.h>
+#include <asm/processor.h>
+#include <asm/cache.h>
+#include <asm/errno.h>
+#include <asm/ppc_asm.h>
+
+#define COPY_16_BYTES          \
+       lwz     r7,4(r4);       \
+       lwz     r8,8(r4);       \
+       lwz     r9,12(r4);      \
+       lwzu    r10,16(r4);     \
+       stw     r7,4(r6);       \
+       stw     r8,8(r6);       \
+       stw     r9,12(r6);      \
+       stwu    r10,16(r6)
+
+#define COPY_16_BYTES_WITHEX(n)        \
+8 ## n ## 0:                   \
+       lwz     r7,4(r4);       \
+8 ## n ## 1:                   \
+       lwz     r8,8(r4);       \
+8 ## n ## 2:                   \
+       lwz     r9,12(r4);      \
+8 ## n ## 3:                   \
+       lwzu    r10,16(r4);     \
+8 ## n ## 4:                   \
+       stw     r7,4(r6);       \
+8 ## n ## 5:                   \
+       stw     r8,8(r6);       \
+8 ## n ## 6:                   \
+       stw     r9,12(r6);      \
+8 ## n ## 7:                   \
+       stwu    r10,16(r6)
+
+#define COPY_16_BYTES_EXCODE(n)                        \
+9 ## n ## 0:                                   \
+       addi    r5,r5,-(16 * n);                \
+       b       104f;                           \
+9 ## n ## 1:                                   \
+       addi    r5,r5,-(16 * n);                \
+       b       105f;                           \
+.section __ex_table,"a";                       \
+       .align  2;                              \
+       .long   8 ## n ## 0b,9 ## n ## 0b;      \
+       .long   8 ## n ## 1b,9 ## n ## 0b;      \
+       .long   8 ## n ## 2b,9 ## n ## 0b;      \
+       .long   8 ## n ## 3b,9 ## n ## 0b;      \
+       .long   8 ## n ## 4b,9 ## n ## 1b;      \
+       .long   8 ## n ## 5b,9 ## n ## 1b;      \
+       .long   8 ## n ## 6b,9 ## n ## 1b;      \
+       .long   8 ## n ## 7b,9 ## n ## 1b;      \
+       .text
+
+       .text
+       .stabs  "arch/powerpc/lib/",N_SO,0,0,0f
+       .stabs  "copy32.S",N_SO,0,0,0f
+0:
+
+CACHELINE_BYTES = L1_CACHE_LINE_SIZE
+LG_CACHELINE_BYTES = LG_L1_CACHE_LINE_SIZE
+CACHELINE_MASK = (L1_CACHE_LINE_SIZE-1)
+
+/*
+ * Use dcbz on the complete cache lines in the destination
+ * to set them to zero.  This requires that the destination
+ * area is cacheable.  -- paulus
+ */
+_GLOBAL(cacheable_memzero)
+       mr      r5,r4
+       li      r4,0
+       addi    r6,r3,-4
+       cmplwi  0,r5,4
+       blt     7f
+       stwu    r4,4(r6)
+       beqlr
+       andi.   r0,r6,3
+       add     r5,r0,r5
+       subf    r6,r0,r6
+       clrlwi  r7,r6,32-LG_CACHELINE_BYTES
+       add     r8,r7,r5
+       srwi    r9,r8,LG_CACHELINE_BYTES
+       addic.  r9,r9,-1        /* total number of complete cachelines */
+       ble     2f
+       xori    r0,r7,CACHELINE_MASK & ~3
+       srwi.   r0,r0,2
+       beq     3f
+       mtctr   r0
+4:     stwu    r4,4(r6)
+       bdnz    4b
+3:     mtctr   r9
+       li      r7,4
+#if !defined(CONFIG_8xx)
+10:    dcbz    r7,r6
+#else
+10:    stw     r4, 4(r6)
+       stw     r4, 8(r6)
+       stw     r4, 12(r6)
+       stw     r4, 16(r6)
+#if CACHE_LINE_SIZE >= 32
+       stw     r4, 20(r6)
+       stw     r4, 24(r6)
+       stw     r4, 28(r6)
+       stw     r4, 32(r6)
+#endif /* CACHE_LINE_SIZE */
+#endif
+       addi    r6,r6,CACHELINE_BYTES
+       bdnz    10b
+       clrlwi  r5,r8,32-LG_CACHELINE_BYTES
+       addi    r5,r5,4
+2:     srwi    r0,r5,2
+       mtctr   r0
+       bdz     6f
+1:     stwu    r4,4(r6)
+       bdnz    1b
+6:     andi.   r5,r5,3
+7:     cmpwi   0,r5,0
+       beqlr
+       mtctr   r5
+       addi    r6,r6,3
+8:     stbu    r4,1(r6)
+       bdnz    8b
+       blr
+
+_GLOBAL(memset)
+       rlwimi  r4,r4,8,16,23
+       rlwimi  r4,r4,16,0,15
+       addi    r6,r3,-4
+       cmplwi  0,r5,4
+       blt     7f
+       stwu    r4,4(r6)
+       beqlr
+       andi.   r0,r6,3
+       add     r5,r0,r5
+       subf    r6,r0,r6
+       srwi    r0,r5,2
+       mtctr   r0
+       bdz     6f
+1:     stwu    r4,4(r6)
+       bdnz    1b
+6:     andi.   r5,r5,3
+7:     cmpwi   0,r5,0
+       beqlr
+       mtctr   r5
+       addi    r6,r6,3
+8:     stbu    r4,1(r6)
+       bdnz    8b
+       blr
+
+/*
+ * This version uses dcbz on the complete cache lines in the
+ * destination area to reduce memory traffic.  This requires that
+ * the destination area is cacheable.
+ * We only use this version if the source and dest don't overlap.
+ * -- paulus.
+ */
+_GLOBAL(cacheable_memcpy)
+       add     r7,r3,r5                /* test if the src & dst overlap */
+       add     r8,r4,r5
+       cmplw   0,r4,r7
+       cmplw   1,r3,r8
+       crand   0,0,4                   /* cr0.lt &= cr1.lt */
+       blt     memcpy                  /* if regions overlap */
+
+       addi    r4,r4,-4
+       addi    r6,r3,-4
+       neg     r0,r3
+       andi.   r0,r0,CACHELINE_MASK    /* # bytes to start of cache line */
+       beq     58f
+
+       cmplw   0,r5,r0                 /* is this more than total to do? */
+       blt     63f                     /* if not much to do */
+       andi.   r8,r0,3                 /* get it word-aligned first */
+       subf    r5,r0,r5
+       mtctr   r8
+       beq+    61f
+70:    lbz     r9,4(r4)                /* do some bytes */
+       stb     r9,4(r6)
+       addi    r4,r4,1
+       addi    r6,r6,1
+       bdnz    70b
+61:    srwi.   r0,r0,2
+       mtctr   r0
+       beq     58f
+72:    lwzu    r9,4(r4)                /* do some words */
+       stwu    r9,4(r6)
+       bdnz    72b
+
+58:    srwi.   r0,r5,LG_CACHELINE_BYTES /* # complete cachelines */
+       clrlwi  r5,r5,32-LG_CACHELINE_BYTES
+       li      r11,4
+       mtctr   r0
+       beq     63f
+53:
+#if !defined(CONFIG_8xx)
+       dcbz    r11,r6
+#endif
+       COPY_16_BYTES
+#if L1_CACHE_LINE_SIZE >= 32
+       COPY_16_BYTES
+#if L1_CACHE_LINE_SIZE >= 64
+       COPY_16_BYTES
+       COPY_16_BYTES
+#if L1_CACHE_LINE_SIZE >= 128
+       COPY_16_BYTES
+       COPY_16_BYTES
+       COPY_16_BYTES
+       COPY_16_BYTES
+#endif
+#endif
+#endif
+       bdnz    53b
+
+63:    srwi.   r0,r5,2
+       mtctr   r0
+       beq     64f
+30:    lwzu    r0,4(r4)
+       stwu    r0,4(r6)
+       bdnz    30b
+
+64:    andi.   r0,r5,3
+       mtctr   r0
+       beq+    65f
+40:    lbz     r0,4(r4)
+       stb     r0,4(r6)
+       addi    r4,r4,1
+       addi    r6,r6,1
+       bdnz    40b
+65:    blr
+
+_GLOBAL(memmove)
+       cmplw   0,r3,r4
+       bgt     backwards_memcpy
+       /* fall through */
+
+_GLOBAL(memcpy)
+       srwi.   r7,r5,3
+       addi    r6,r3,-4
+       addi    r4,r4,-4
+       beq     2f                      /* if less than 8 bytes to do */
+       andi.   r0,r6,3                 /* get dest word aligned */
+       mtctr   r7
+       bne     5f
+1:     lwz     r7,4(r4)
+       lwzu    r8,8(r4)
+       stw     r7,4(r6)
+       stwu    r8,8(r6)
+       bdnz    1b
+       andi.   r5,r5,7
+2:     cmplwi  0,r5,4
+       blt     3f
+       lwzu    r0,4(r4)
+       addi    r5,r5,-4
+       stwu    r0,4(r6)
+3:     cmpwi   0,r5,0
+       beqlr
+       mtctr   r5
+       addi    r4,r4,3
+       addi    r6,r6,3
+4:     lbzu    r0,1(r4)
+       stbu    r0,1(r6)
+       bdnz    4b
+       blr
+5:     subfic  r0,r0,4
+       mtctr   r0
+6:     lbz     r7,4(r4)
+       addi    r4,r4,1
+       stb     r7,4(r6)
+       addi    r6,r6,1
+       bdnz    6b
+       subf    r5,r0,r5
+       rlwinm. r7,r5,32-3,3,31
+       beq     2b
+       mtctr   r7
+       b       1b
+
+_GLOBAL(backwards_memcpy)
+       rlwinm. r7,r5,32-3,3,31         /* r0 = r5 >> 3 */
+       add     r6,r3,r5
+       add     r4,r4,r5
+       beq     2f
+       andi.   r0,r6,3
+       mtctr   r7
+       bne     5f
+1:     lwz     r7,-4(r4)
+       lwzu    r8,-8(r4)
+       stw     r7,-4(r6)
+       stwu    r8,-8(r6)
+       bdnz    1b
+       andi.   r5,r5,7
+2:     cmplwi  0,r5,4
+       blt     3f
+       lwzu    r0,-4(r4)
+       subi    r5,r5,4
+       stwu    r0,-4(r6)
+3:     cmpwi   0,r5,0
+       beqlr
+       mtctr   r5
+4:     lbzu    r0,-1(r4)
+       stbu    r0,-1(r6)
+       bdnz    4b
+       blr
+5:     mtctr   r0
+6:     lbzu    r7,-1(r4)
+       stbu    r7,-1(r6)
+       bdnz    6b
+       subf    r5,r0,r5
+       rlwinm. r7,r5,32-3,3,31
+       beq     2b
+       mtctr   r7
+       b       1b
+
+_GLOBAL(__copy_tofrom_user)
+       addi    r4,r4,-4
+       addi    r6,r3,-4
+       neg     r0,r3
+       andi.   r0,r0,CACHELINE_MASK    /* # bytes to start of cache line */
+       beq     58f
+
+       cmplw   0,r5,r0                 /* is this more than total to do? */
+       blt     63f                     /* if not much to do */
+       andi.   r8,r0,3                 /* get it word-aligned first */
+       mtctr   r8
+       beq+    61f
+70:    lbz     r9,4(r4)                /* do some bytes */
+71:    stb     r9,4(r6)
+       addi    r4,r4,1
+       addi    r6,r6,1
+       bdnz    70b
+61:    subf    r5,r0,r5
+       srwi.   r0,r0,2
+       mtctr   r0
+       beq     58f
+72:    lwzu    r9,4(r4)                /* do some words */
+73:    stwu    r9,4(r6)
+       bdnz    72b
+
+       .section __ex_table,"a"
+       .align  2
+       .long   70b,100f
+       .long   71b,101f
+       .long   72b,102f
+       .long   73b,103f
+       .text
+
+58:    srwi.   r0,r5,LG_CACHELINE_BYTES /* # complete cachelines */
+       clrlwi  r5,r5,32-LG_CACHELINE_BYTES
+       li      r11,4
+       beq     63f
+
+#ifdef CONFIG_8xx
+       /* Don't use prefetch on 8xx */
+       mtctr   r0
+       li      r0,0
+53:    COPY_16_BYTES_WITHEX(0)
+       bdnz    53b
+
+#else /* not CONFIG_8xx */
+       /* Here we decide how far ahead to prefetch the source */
+       li      r3,4
+       cmpwi   r0,1
+       li      r7,0
+       ble     114f
+       li      r7,1
+#if MAX_COPY_PREFETCH > 1
+       /* Heuristically, for large transfers we prefetch
+          MAX_COPY_PREFETCH cachelines ahead.  For small transfers
+          we prefetch 1 cacheline ahead. */
+       cmpwi   r0,MAX_COPY_PREFETCH
+       ble     112f
+       li      r7,MAX_COPY_PREFETCH
+112:   mtctr   r7
+111:   dcbt    r3,r4
+       addi    r3,r3,CACHELINE_BYTES
+       bdnz    111b
+#else
+       dcbt    r3,r4
+       addi    r3,r3,CACHELINE_BYTES
+#endif /* MAX_COPY_PREFETCH > 1 */
+
+114:   subf    r8,r7,r0
+       mr      r0,r7
+       mtctr   r8
+
+53:    dcbt    r3,r4
+54:    dcbz    r11,r6
+       .section __ex_table,"a"
+       .align  2
+       .long   54b,105f
+       .text
+/* the main body of the cacheline loop */
+       COPY_16_BYTES_WITHEX(0)
+#if L1_CACHE_LINE_SIZE >= 32
+       COPY_16_BYTES_WITHEX(1)
+#if L1_CACHE_LINE_SIZE >= 64
+       COPY_16_BYTES_WITHEX(2)
+       COPY_16_BYTES_WITHEX(3)
+#if L1_CACHE_LINE_SIZE >= 128
+       COPY_16_BYTES_WITHEX(4)
+       COPY_16_BYTES_WITHEX(5)
+       COPY_16_BYTES_WITHEX(6)
+       COPY_16_BYTES_WITHEX(7)
+#endif
+#endif
+#endif
+       bdnz    53b
+       cmpwi   r0,0
+       li      r3,4
+       li      r7,0
+       bne     114b
+#endif /* CONFIG_8xx */
+
+63:    srwi.   r0,r5,2
+       mtctr   r0
+       beq     64f
+30:    lwzu    r0,4(r4)
+31:    stwu    r0,4(r6)
+       bdnz    30b
+
+64:    andi.   r0,r5,3
+       mtctr   r0
+       beq+    65f
+40:    lbz     r0,4(r4)
+41:    stb     r0,4(r6)
+       addi    r4,r4,1
+       addi    r6,r6,1
+       bdnz    40b
+65:    li      r3,0
+       blr
+
+/* read fault, initial single-byte copy */
+100:   li      r9,0
+       b       90f
+/* write fault, initial single-byte copy */
+101:   li      r9,1
+90:    subf    r5,r8,r5
+       li      r3,0
+       b       99f
+/* read fault, initial word copy */
+102:   li      r9,0
+       b       91f
+/* write fault, initial word copy */
+103:   li      r9,1
+91:    li      r3,2
+       b       99f
+
+/*
+ * this stuff handles faults in the cacheline loop and branches to either
+ * 104f (if in read part) or 105f (if in write part), after updating r5
+ */
+       COPY_16_BYTES_EXCODE(0)
+#if L1_CACHE_LINE_SIZE >= 32
+       COPY_16_BYTES_EXCODE(1)
+#if L1_CACHE_LINE_SIZE >= 64
+       COPY_16_BYTES_EXCODE(2)
+       COPY_16_BYTES_EXCODE(3)
+#if L1_CACHE_LINE_SIZE >= 128
+       COPY_16_BYTES_EXCODE(4)
+       COPY_16_BYTES_EXCODE(5)
+       COPY_16_BYTES_EXCODE(6)
+       COPY_16_BYTES_EXCODE(7)
+#endif
+#endif
+#endif
+
+/* read fault in cacheline loop */
+104:   li      r9,0
+       b       92f
+/* fault on dcbz (effectively a write fault) */
+/* or write fault in cacheline loop */
+105:   li      r9,1
+92:    li      r3,LG_CACHELINE_BYTES
+       mfctr   r8
+       add     r0,r0,r8
+       b       106f
+/* read fault in final word loop */
+108:   li      r9,0
+       b       93f
+/* write fault in final word loop */
+109:   li      r9,1
+93:    andi.   r5,r5,3
+       li      r3,2
+       b       99f
+/* read fault in final byte loop */
+110:   li      r9,0
+       b       94f
+/* write fault in final byte loop */
+111:   li      r9,1
+94:    li      r5,0
+       li      r3,0
+/*
+ * At this stage the number of bytes not copied is
+ * r5 + (ctr << r3), and r9 is 0 for read or 1 for write.
+ */
+99:    mfctr   r0
+106:   slw     r3,r0,r3
+       add.    r3,r3,r5
+       beq     120f                    /* shouldn't happen */
+       cmpwi   0,r9,0
+       bne     120f
+/* for a read fault, first try to continue the copy one byte at a time */
+       mtctr   r3
+130:   lbz     r0,4(r4)
+131:   stb     r0,4(r6)
+       addi    r4,r4,1
+       addi    r6,r6,1
+       bdnz    130b
+/* then clear out the destination: r3 bytes starting at 4(r6) */
+132:   mfctr   r3
+       srwi.   r0,r3,2
+       li      r9,0
+       mtctr   r0
+       beq     113f
+112:   stwu    r9,4(r6)
+       bdnz    112b
+113:   andi.   r0,r3,3
+       mtctr   r0
+       beq     120f
+114:   stb     r9,4(r6)
+       addi    r6,r6,1
+       bdnz    114b
+120:   blr
+
+       .section __ex_table,"a"
+       .align  2
+       .long   30b,108b
+       .long   31b,109b
+       .long   40b,110b
+       .long   41b,111b
+       .long   130b,132b
+       .long   131b,120b
+       .long   112b,120b
+       .long   114b,120b
+       .text
diff --git a/arch/powerpc/lib/copypage.S b/arch/powerpc/lib/copypage.S
new file mode 100644 (file)
index 0000000..733d616
--- /dev/null
@@ -0,0 +1,121 @@
+/*
+ * arch/ppc64/lib/copypage.S
+ *
+ * Copyright (C) 2002 Paul Mackerras, IBM Corp.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version
+ * 2 of the License, or (at your option) any later version.
+ */
+#include <asm/processor.h>
+#include <asm/ppc_asm.h>
+
+_GLOBAL(copy_page)
+       std     r31,-8(1)
+       std     r30,-16(1)
+       std     r29,-24(1)
+       std     r28,-32(1)
+       std     r27,-40(1)
+       std     r26,-48(1)
+       std     r25,-56(1)
+       std     r24,-64(1)
+       std     r23,-72(1)
+       std     r22,-80(1)
+       std     r21,-88(1)
+       std     r20,-96(1)
+       li      r5,4096/32 - 1
+       addi    r3,r3,-8
+       li      r12,5
+0:     addi    r5,r5,-24
+       mtctr   r12
+       ld      r22,640(4)
+       ld      r21,512(4)
+       ld      r20,384(4)
+       ld      r11,256(4)
+       ld      r9,128(4)
+       ld      r7,0(4)
+       ld      r25,648(4)
+       ld      r24,520(4)
+       ld      r23,392(4)
+       ld      r10,264(4)
+       ld      r8,136(4)
+       ldu     r6,8(4)
+       cmpwi   r5,24
+1:     std     r22,648(3)
+       std     r21,520(3)
+       std     r20,392(3)
+       std     r11,264(3)
+       std     r9,136(3)
+       std     r7,8(3)
+       ld      r28,648(4)
+       ld      r27,520(4)
+       ld      r26,392(4)
+       ld      r31,264(4)
+       ld      r30,136(4)
+       ld      r29,8(4)
+       std     r25,656(3)
+       std     r24,528(3)
+       std     r23,400(3)
+       std     r10,272(3)
+       std     r8,144(3)
+       std     r6,16(3)
+       ld      r22,656(4)
+       ld      r21,528(4)
+       ld      r20,400(4)
+       ld      r11,272(4)
+       ld      r9,144(4)
+       ld      r7,16(4)
+       std     r28,664(3)
+       std     r27,536(3)
+       std     r26,408(3)
+       std     r31,280(3)
+       std     r30,152(3)
+       stdu    r29,24(3)
+       ld      r25,664(4)
+       ld      r24,536(4)
+       ld      r23,408(4)
+       ld      r10,280(4)
+       ld      r8,152(4)
+       ldu     r6,24(4)
+       bdnz    1b
+       std     r22,648(3)
+       std     r21,520(3)
+       std     r20,392(3)
+       std     r11,264(3)
+       std     r9,136(3)
+       std     r7,8(3)
+       addi    r4,r4,640
+       addi    r3,r3,648
+       bge     0b
+       mtctr   r5
+       ld      r7,0(4)
+       ld      r8,8(4)
+       ldu     r9,16(4)
+3:     ld      r10,8(4)
+       std     r7,8(3)
+       ld      r7,16(4)
+       std     r8,16(3)
+       ld      r8,24(4)
+       std     r9,24(3)
+       ldu     r9,32(4)
+       stdu    r10,32(3)
+       bdnz    3b
+4:     ld      r10,8(4)
+       std     r7,8(3)
+       std     r8,16(3)
+       std     r9,24(3)
+       std     r10,32(3)
+9:     ld      r20,-96(1)
+       ld      r21,-88(1)
+       ld      r22,-80(1)
+       ld      r23,-72(1)
+       ld      r24,-64(1)
+       ld      r25,-56(1)
+       ld      r26,-48(1)
+       ld      r27,-40(1)
+       ld      r28,-32(1)
+       ld      r29,-24(1)
+       ld      r30,-16(1)
+       ld      r31,-8(1)
+       blr
diff --git a/arch/powerpc/lib/copyuser.S b/arch/powerpc/lib/copyuser.S
new file mode 100644 (file)
index 0000000..a0b3fbb
--- /dev/null
@@ -0,0 +1,576 @@
+/*
+ * arch/ppc64/lib/copyuser.S
+ *
+ * Copyright (C) 2002 Paul Mackerras, IBM Corp.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version
+ * 2 of the License, or (at your option) any later version.
+ */
+#include <asm/processor.h>
+#include <asm/ppc_asm.h>
+
+       .align  7
+_GLOBAL(__copy_tofrom_user)
+       /* first check for a whole page copy on a page boundary */
+       cmpldi  cr1,r5,16
+       cmpdi   cr6,r5,4096
+       or      r0,r3,r4
+       neg     r6,r3           /* LS 3 bits = # bytes to 8-byte dest bdry */
+       andi.   r0,r0,4095
+       std     r3,-24(r1)
+       crand   cr0*4+2,cr0*4+2,cr6*4+2
+       std     r4,-16(r1)
+       std     r5,-8(r1)
+       dcbt    0,r4
+       beq     .Lcopy_page
+       andi.   r6,r6,7
+       mtcrf   0x01,r5
+       blt     cr1,.Lshort_copy
+       bne     .Ldst_unaligned
+.Ldst_aligned:
+       andi.   r0,r4,7
+       addi    r3,r3,-16
+       bne     .Lsrc_unaligned
+       srdi    r7,r5,4
+20:    ld      r9,0(r4)
+       addi    r4,r4,-8
+       mtctr   r7
+       andi.   r5,r5,7
+       bf      cr7*4+0,22f
+       addi    r3,r3,8
+       addi    r4,r4,8
+       mr      r8,r9
+       blt     cr1,72f
+21:    ld      r9,8(r4)
+70:    std     r8,8(r3)
+22:    ldu     r8,16(r4)
+71:    stdu    r9,16(r3)
+       bdnz    21b
+72:    std     r8,8(r3)
+       beq+    3f
+       addi    r3,r3,16
+23:    ld      r9,8(r4)
+.Ldo_tail:
+       bf      cr7*4+1,1f
+       rotldi  r9,r9,32
+73:    stw     r9,0(r3)
+       addi    r3,r3,4
+1:     bf      cr7*4+2,2f
+       rotldi  r9,r9,16
+74:    sth     r9,0(r3)
+       addi    r3,r3,2
+2:     bf      cr7*4+3,3f
+       rotldi  r9,r9,8
+75:    stb     r9,0(r3)
+3:     li      r3,0
+       blr
+
+.Lsrc_unaligned:
+       srdi    r6,r5,3
+       addi    r5,r5,-16
+       subf    r4,r0,r4
+       srdi    r7,r5,4
+       sldi    r10,r0,3
+       cmpldi  cr6,r6,3
+       andi.   r5,r5,7
+       mtctr   r7
+       subfic  r11,r10,64
+       add     r5,r5,r0
+       bt      cr7*4+0,28f
+
+24:    ld      r9,0(r4)        /* 3+2n loads, 2+2n stores */
+25:    ld      r0,8(r4)
+       sld     r6,r9,r10
+26:    ldu     r9,16(r4)
+       srd     r7,r0,r11
+       sld     r8,r0,r10
+       or      r7,r7,r6
+       blt     cr6,79f
+27:    ld      r0,8(r4)
+       b       2f
+
+28:    ld      r0,0(r4)        /* 4+2n loads, 3+2n stores */
+29:    ldu     r9,8(r4)
+       sld     r8,r0,r10
+       addi    r3,r3,-8
+       blt     cr6,5f
+30:    ld      r0,8(r4)
+       srd     r12,r9,r11
+       sld     r6,r9,r10
+31:    ldu     r9,16(r4)
+       or      r12,r8,r12
+       srd     r7,r0,r11
+       sld     r8,r0,r10
+       addi    r3,r3,16
+       beq     cr6,78f
+
+1:     or      r7,r7,r6
+32:    ld      r0,8(r4)
+76:    std     r12,8(r3)
+2:     srd     r12,r9,r11
+       sld     r6,r9,r10
+33:    ldu     r9,16(r4)
+       or      r12,r8,r12
+77:    stdu    r7,16(r3)
+       srd     r7,r0,r11
+       sld     r8,r0,r10
+       bdnz    1b
+
+78:    std     r12,8(r3)
+       or      r7,r7,r6
+79:    std     r7,16(r3)
+5:     srd     r12,r9,r11
+       or      r12,r8,r12
+80:    std     r12,24(r3)
+       bne     6f
+       li      r3,0
+       blr
+6:     cmpwi   cr1,r5,8
+       addi    r3,r3,32
+       sld     r9,r9,r10
+       ble     cr1,.Ldo_tail
+34:    ld      r0,8(r4)
+       srd     r7,r0,r11
+       or      r9,r7,r9
+       b       .Ldo_tail
+
+.Ldst_unaligned:
+       mtcrf   0x01,r6         /* put #bytes to 8B bdry into cr7 */
+       subf    r5,r6,r5
+       li      r7,0
+       cmpldi  r1,r5,16
+       bf      cr7*4+3,1f
+35:    lbz     r0,0(r4)
+81:    stb     r0,0(r3)
+       addi    r7,r7,1
+1:     bf      cr7*4+2,2f
+36:    lhzx    r0,r7,r4
+82:    sthx    r0,r7,r3
+       addi    r7,r7,2
+2:     bf      cr7*4+1,3f
+37:    lwzx    r0,r7,r4
+83:    stwx    r0,r7,r3
+3:     mtcrf   0x01,r5
+       add     r4,r6,r4
+       add     r3,r6,r3
+       b       .Ldst_aligned
+
+.Lshort_copy:
+       bf      cr7*4+0,1f
+38:    lwz     r0,0(r4)
+39:    lwz     r9,4(r4)
+       addi    r4,r4,8
+84:    stw     r0,0(r3)
+85:    stw     r9,4(r3)
+       addi    r3,r3,8
+1:     bf      cr7*4+1,2f
+40:    lwz     r0,0(r4)
+       addi    r4,r4,4
+86:    stw     r0,0(r3)
+       addi    r3,r3,4
+2:     bf      cr7*4+2,3f
+41:    lhz     r0,0(r4)
+       addi    r4,r4,2
+87:    sth     r0,0(r3)
+       addi    r3,r3,2
+3:     bf      cr7*4+3,4f
+42:    lbz     r0,0(r4)
+88:    stb     r0,0(r3)
+4:     li      r3,0
+       blr
+
+/*
+ * exception handlers follow
+ * we have to return the number of bytes not copied
+ * for an exception on a load, we set the rest of the destination to 0
+ */
+
+136:
+137:
+       add     r3,r3,r7
+       b       1f
+130:
+131:
+       addi    r3,r3,8
+120:
+122:
+124:
+125:
+126:
+127:
+128:
+129:
+133:
+       addi    r3,r3,8
+121:
+132:
+       addi    r3,r3,8
+123:
+134:
+135:
+138:
+139:
+140:
+141:
+142:
+
+/*
+ * here we have had a fault on a load and r3 points to the first
+ * unmodified byte of the destination
+ */
+1:     ld      r6,-24(r1)
+       ld      r4,-16(r1)
+       ld      r5,-8(r1)
+       subf    r6,r6,r3
+       add     r4,r4,r6
+       subf    r5,r6,r5        /* #bytes left to go */
+
+/*
+ * first see if we can copy any more bytes before hitting another exception
+ */
+       mtctr   r5
+43:    lbz     r0,0(r4)
+       addi    r4,r4,1
+89:    stb     r0,0(r3)
+       addi    r3,r3,1
+       bdnz    43b
+       li      r3,0            /* huh? all copied successfully this time? */
+       blr
+
+/*
+ * here we have trapped again, need to clear ctr bytes starting at r3
+ */
+143:   mfctr   r5
+       li      r0,0
+       mr      r4,r3
+       mr      r3,r5           /* return the number of bytes not copied */
+1:     andi.   r9,r4,7
+       beq     3f
+90:    stb     r0,0(r4)
+       addic.  r5,r5,-1
+       addi    r4,r4,1
+       bne     1b
+       blr
+3:     cmpldi  cr1,r5,8
+       srdi    r9,r5,3
+       andi.   r5,r5,7
+       blt     cr1,93f
+       mtctr   r9
+91:    std     r0,0(r4)
+       addi    r4,r4,8
+       bdnz    91b
+93:    beqlr
+       mtctr   r5      
+92:    stb     r0,0(r4)
+       addi    r4,r4,1
+       bdnz    92b
+       blr
+
+/*
+ * exception handlers for stores: we just need to work
+ * out how many bytes weren't copied
+ */
+182:
+183:
+       add     r3,r3,r7
+       b       1f
+180:
+       addi    r3,r3,8
+171:
+177:
+       addi    r3,r3,8
+170:
+172:
+176:
+178:
+       addi    r3,r3,4
+185:
+       addi    r3,r3,4
+173:
+174:
+175:
+179:
+181:
+184:
+186:
+187:
+188:
+189:   
+1:
+       ld      r6,-24(r1)
+       ld      r5,-8(r1)
+       add     r6,r6,r5
+       subf    r3,r3,r6        /* #bytes not copied */
+190:
+191:
+192:
+       blr                     /* #bytes not copied in r3 */
+
+       .section __ex_table,"a"
+       .align  3
+       .llong  20b,120b
+       .llong  21b,121b
+       .llong  70b,170b
+       .llong  22b,122b
+       .llong  71b,171b
+       .llong  72b,172b
+       .llong  23b,123b
+       .llong  73b,173b
+       .llong  74b,174b
+       .llong  75b,175b
+       .llong  24b,124b
+       .llong  25b,125b
+       .llong  26b,126b
+       .llong  27b,127b
+       .llong  28b,128b
+       .llong  29b,129b
+       .llong  30b,130b
+       .llong  31b,131b
+       .llong  32b,132b
+       .llong  76b,176b
+       .llong  33b,133b
+       .llong  77b,177b
+       .llong  78b,178b
+       .llong  79b,179b
+       .llong  80b,180b
+       .llong  34b,134b
+       .llong  35b,135b
+       .llong  81b,181b
+       .llong  36b,136b
+       .llong  82b,182b
+       .llong  37b,137b
+       .llong  83b,183b
+       .llong  38b,138b
+       .llong  39b,139b
+       .llong  84b,184b
+       .llong  85b,185b
+       .llong  40b,140b
+       .llong  86b,186b
+       .llong  41b,141b
+       .llong  87b,187b
+       .llong  42b,142b
+       .llong  88b,188b
+       .llong  43b,143b
+       .llong  89b,189b
+       .llong  90b,190b
+       .llong  91b,191b
+       .llong  92b,192b
+       
+       .text
+
+/*
+ * Routine to copy a whole page of data, optimized for POWER4.
+ * On POWER4 it is more than 50% faster than the simple loop
+ * above (following the .Ldst_aligned label) but it runs slightly
+ * slower on POWER3.
+ */
+.Lcopy_page:
+       std     r31,-32(1)
+       std     r30,-40(1)
+       std     r29,-48(1)
+       std     r28,-56(1)
+       std     r27,-64(1)
+       std     r26,-72(1)
+       std     r25,-80(1)
+       std     r24,-88(1)
+       std     r23,-96(1)
+       std     r22,-104(1)
+       std     r21,-112(1)
+       std     r20,-120(1)
+       li      r5,4096/32 - 1
+       addi    r3,r3,-8
+       li      r0,5
+0:     addi    r5,r5,-24
+       mtctr   r0
+20:    ld      r22,640(4)
+21:    ld      r21,512(4)
+22:    ld      r20,384(4)
+23:    ld      r11,256(4)
+24:    ld      r9,128(4)
+25:    ld      r7,0(4)
+26:    ld      r25,648(4)
+27:    ld      r24,520(4)
+28:    ld      r23,392(4)
+29:    ld      r10,264(4)
+30:    ld      r8,136(4)
+31:    ldu     r6,8(4)
+       cmpwi   r5,24
+1:
+32:    std     r22,648(3)
+33:    std     r21,520(3)
+34:    std     r20,392(3)
+35:    std     r11,264(3)
+36:    std     r9,136(3)
+37:    std     r7,8(3)
+38:    ld      r28,648(4)
+39:    ld      r27,520(4)
+40:    ld      r26,392(4)
+41:    ld      r31,264(4)
+42:    ld      r30,136(4)
+43:    ld      r29,8(4)
+44:    std     r25,656(3)
+45:    std     r24,528(3)
+46:    std     r23,400(3)
+47:    std     r10,272(3)
+48:    std     r8,144(3)
+49:    std     r6,16(3)
+50:    ld      r22,656(4)
+51:    ld      r21,528(4)
+52:    ld      r20,400(4)
+53:    ld      r11,272(4)
+54:    ld      r9,144(4)
+55:    ld      r7,16(4)
+56:    std     r28,664(3)
+57:    std     r27,536(3)
+58:    std     r26,408(3)
+59:    std     r31,280(3)
+60:    std     r30,152(3)
+61:    stdu    r29,24(3)
+62:    ld      r25,664(4)
+63:    ld      r24,536(4)
+64:    ld      r23,408(4)
+65:    ld      r10,280(4)
+66:    ld      r8,152(4)
+67:    ldu     r6,24(4)
+       bdnz    1b
+68:    std     r22,648(3)
+69:    std     r21,520(3)
+70:    std     r20,392(3)
+71:    std     r11,264(3)
+72:    std     r9,136(3)
+73:    std     r7,8(3)
+74:    addi    r4,r4,640
+75:    addi    r3,r3,648
+       bge     0b
+       mtctr   r5
+76:    ld      r7,0(4)
+77:    ld      r8,8(4)
+78:    ldu     r9,16(4)
+3:
+79:    ld      r10,8(4)
+80:    std     r7,8(3)
+81:    ld      r7,16(4)
+82:    std     r8,16(3)
+83:    ld      r8,24(4)
+84:    std     r9,24(3)
+85:    ldu     r9,32(4)
+86:    stdu    r10,32(3)
+       bdnz    3b
+4:
+87:    ld      r10,8(4)
+88:    std     r7,8(3)
+89:    std     r8,16(3)
+90:    std     r9,24(3)
+91:    std     r10,32(3)
+9:     ld      r20,-120(1)
+       ld      r21,-112(1)
+       ld      r22,-104(1)
+       ld      r23,-96(1)
+       ld      r24,-88(1)
+       ld      r25,-80(1)
+       ld      r26,-72(1)
+       ld      r27,-64(1)
+       ld      r28,-56(1)
+       ld      r29,-48(1)
+       ld      r30,-40(1)
+       ld      r31,-32(1)
+       li      r3,0
+       blr
+
+/*
+ * on an exception, reset to the beginning and jump back into the
+ * standard __copy_tofrom_user
+ */
+100:   ld      r20,-120(1)
+       ld      r21,-112(1)
+       ld      r22,-104(1)
+       ld      r23,-96(1)
+       ld      r24,-88(1)
+       ld      r25,-80(1)
+       ld      r26,-72(1)
+       ld      r27,-64(1)
+       ld      r28,-56(1)
+       ld      r29,-48(1)
+       ld      r30,-40(1)
+       ld      r31,-32(1)
+       ld      r3,-24(r1)
+       ld      r4,-16(r1)
+       li      r5,4096
+       b       .Ldst_aligned
+
+       .section __ex_table,"a"
+       .align  3
+       .llong  20b,100b
+       .llong  21b,100b
+       .llong  22b,100b
+       .llong  23b,100b
+       .llong  24b,100b
+       .llong  25b,100b
+       .llong  26b,100b
+       .llong  27b,100b
+       .llong  28b,100b
+       .llong  29b,100b
+       .llong  30b,100b
+       .llong  31b,100b
+       .llong  32b,100b
+       .llong  33b,100b
+       .llong  34b,100b
+       .llong  35b,100b
+       .llong  36b,100b
+       .llong  37b,100b
+       .llong  38b,100b
+       .llong  39b,100b
+       .llong  40b,100b
+       .llong  41b,100b
+       .llong  42b,100b
+       .llong  43b,100b
+       .llong  44b,100b
+       .llong  45b,100b
+       .llong  46b,100b
+       .llong  47b,100b
+       .llong  48b,100b
+       .llong  49b,100b
+       .llong  50b,100b
+       .llong  51b,100b
+       .llong  52b,100b
+       .llong  53b,100b
+       .llong  54b,100b
+       .llong  55b,100b
+       .llong  56b,100b
+       .llong  57b,100b
+       .llong  58b,100b
+       .llong  59b,100b
+       .llong  60b,100b
+       .llong  61b,100b
+       .llong  62b,100b
+       .llong  63b,100b
+       .llong  64b,100b
+       .llong  65b,100b
+       .llong  66b,100b
+       .llong  67b,100b
+       .llong  68b,100b
+       .llong  69b,100b
+       .llong  70b,100b
+       .llong  71b,100b
+       .llong  72b,100b
+       .llong  73b,100b
+       .llong  74b,100b
+       .llong  75b,100b
+       .llong  76b,100b
+       .llong  77b,100b
+       .llong  78b,100b
+       .llong  79b,100b
+       .llong  80b,100b
+       .llong  81b,100b
+       .llong  82b,100b
+       .llong  83b,100b
+       .llong  84b,100b
+       .llong  85b,100b
+       .llong  86b,100b
+       .llong  87b,100b
+       .llong  88b,100b
+       .llong  89b,100b
+       .llong  90b,100b
+       .llong  91b,100b
diff --git a/arch/powerpc/lib/div64.S b/arch/powerpc/lib/div64.S
new file mode 100644 (file)
index 0000000..3527569
--- /dev/null
@@ -0,0 +1,58 @@
+/*
+ * Divide a 64-bit unsigned number by a 32-bit unsigned number.
+ * This routine assumes that the top 32 bits of the dividend are
+ * non-zero to start with.
+ * On entry, r3 points to the dividend, which get overwritten with
+ * the 64-bit quotient, and r4 contains the divisor.
+ * On exit, r3 contains the remainder.
+ *
+ * Copyright (C) 2002 Paul Mackerras, IBM Corp.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version
+ * 2 of the License, or (at your option) any later version.
+ */
+#include <asm/ppc_asm.h>
+#include <asm/processor.h>
+
+_GLOBAL(__div64_32)
+       lwz     r5,0(r3)        # get the dividend into r5/r6
+       lwz     r6,4(r3)
+       cmplw   r5,r4
+       li      r7,0
+       li      r8,0
+       blt     1f
+       divwu   r7,r5,r4        # if dividend.hi >= divisor,
+       mullw   r0,r7,r4        # quotient.hi = dividend.hi / divisor
+       subf.   r5,r0,r5        # dividend.hi %= divisor
+       beq     3f
+1:     mr      r11,r5          # here dividend.hi != 0
+       andis.  r0,r5,0xc000
+       bne     2f
+       cntlzw  r0,r5           # we are shifting the dividend right
+       li      r10,-1          # to make it < 2^32, and shifting
+       srw     r10,r10,r0      # the divisor right the same amount,
+       add     r9,r4,r10       # rounding up (so the estimate cannot
+       andc    r11,r6,r10      # ever be too large, only too small)
+       andc    r9,r9,r10
+       or      r11,r5,r11
+       rotlw   r9,r9,r0
+       rotlw   r11,r11,r0
+       divwu   r11,r11,r9      # then we divide the shifted quantities
+2:     mullw   r10,r11,r4      # to get an estimate of the quotient,
+       mulhwu  r9,r11,r4       # multiply the estimate by the divisor,
+       subfc   r6,r10,r6       # take the product from the divisor,
+       add     r8,r8,r11       # and add the estimate to the accumulated
+       subfe.  r5,r9,r5        # quotient
+       bne     1b
+3:     cmplw   r6,r4
+       blt     4f
+       divwu   r0,r6,r4        # perform the remaining 32-bit division
+       mullw   r10,r0,r4       # and get the remainder
+       add     r8,r8,r0
+       subf    r6,r10,r6
+4:     stw     r7,0(r3)        # return the quotient in *r3
+       stw     r8,4(r3)
+       mr      r3,r6           # return the remainder in r3
+       blr
diff --git a/arch/powerpc/lib/e2a.c b/arch/powerpc/lib/e2a.c
new file mode 100644 (file)
index 0000000..d2b8348
--- /dev/null
@@ -0,0 +1,108 @@
+/*
+ *  arch/ppc64/lib/e2a.c
+ *
+ *  EBCDIC to ASCII conversion
+ *
+ * This function moved here from arch/ppc64/kernel/viopath.c
+ *
+ * (C) Copyright 2000-2004 IBM Corporation
+ *
+ * This program is free software;  you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License as
+ * published by the Free Software Foundation; either version 2 of the
+ * License, or (at your option) anyu later version.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software Foundation,
+ * Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ *
+ */
+
+#include <linux/module.h>
+
+unsigned char e2a(unsigned char x)
+{
+       switch (x) {
+       case 0xF0:
+               return '0';
+       case 0xF1:
+               return '1';
+       case 0xF2:
+               return '2';
+       case 0xF3:
+               return '3';
+       case 0xF4:
+               return '4';
+       case 0xF5:
+               return '5';
+       case 0xF6:
+               return '6';
+       case 0xF7:
+               return '7';
+       case 0xF8:
+               return '8';
+       case 0xF9:
+               return '9';
+       case 0xC1:
+               return 'A';
+       case 0xC2:
+               return 'B';
+       case 0xC3:
+               return 'C';
+       case 0xC4:
+               return 'D';
+       case 0xC5:
+               return 'E';
+       case 0xC6:
+               return 'F';
+       case 0xC7:
+               return 'G';
+       case 0xC8:
+               return 'H';
+       case 0xC9:
+               return 'I';
+       case 0xD1:
+               return 'J';
+       case 0xD2:
+               return 'K';
+       case 0xD3:
+               return 'L';
+       case 0xD4:
+               return 'M';
+       case 0xD5:
+               return 'N';
+       case 0xD6:
+               return 'O';
+       case 0xD7:
+               return 'P';
+       case 0xD8:
+               return 'Q';
+       case 0xD9:
+               return 'R';
+       case 0xE2:
+               return 'S';
+       case 0xE3:
+               return 'T';
+       case 0xE4:
+               return 'U';
+       case 0xE5:
+               return 'V';
+       case 0xE6:
+               return 'W';
+       case 0xE7:
+               return 'X';
+       case 0xE8:
+               return 'Y';
+       case 0xE9:
+               return 'Z';
+       }
+       return ' ';
+}
+EXPORT_SYMBOL(e2a);
+
+
diff --git a/arch/powerpc/lib/memcpy.S b/arch/powerpc/lib/memcpy.S
new file mode 100644 (file)
index 0000000..9ccacdf
--- /dev/null
@@ -0,0 +1,172 @@
+/*
+ * arch/ppc64/lib/memcpy.S
+ *
+ * Copyright (C) 2002 Paul Mackerras, IBM Corp.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version
+ * 2 of the License, or (at your option) any later version.
+ */
+#include <asm/processor.h>
+#include <asm/ppc_asm.h>
+
+       .align  7
+_GLOBAL(memcpy)
+       mtcrf   0x01,r5
+       cmpldi  cr1,r5,16
+       neg     r6,r3           # LS 3 bits = # bytes to 8-byte dest bdry
+       andi.   r6,r6,7
+       dcbt    0,r4
+       blt     cr1,.Lshort_copy
+       bne     .Ldst_unaligned
+.Ldst_aligned:
+       andi.   r0,r4,7
+       addi    r3,r3,-16
+       bne     .Lsrc_unaligned
+       srdi    r7,r5,4
+       ld      r9,0(r4)
+       addi    r4,r4,-8
+       mtctr   r7
+       andi.   r5,r5,7
+       bf      cr7*4+0,2f
+       addi    r3,r3,8
+       addi    r4,r4,8
+       mr      r8,r9
+       blt     cr1,3f
+1:     ld      r9,8(r4)
+       std     r8,8(r3)
+2:     ldu     r8,16(r4)
+       stdu    r9,16(r3)
+       bdnz    1b
+3:     std     r8,8(r3)
+       beqlr
+       addi    r3,r3,16
+       ld      r9,8(r4)
+.Ldo_tail:
+       bf      cr7*4+1,1f
+       rotldi  r9,r9,32
+       stw     r9,0(r3)
+       addi    r3,r3,4
+1:     bf      cr7*4+2,2f
+       rotldi  r9,r9,16
+       sth     r9,0(r3)
+       addi    r3,r3,2
+2:     bf      cr7*4+3,3f
+       rotldi  r9,r9,8
+       stb     r9,0(r3)
+3:     blr
+
+.Lsrc_unaligned:
+       srdi    r6,r5,3
+       addi    r5,r5,-16
+       subf    r4,r0,r4
+       srdi    r7,r5,4
+       sldi    r10,r0,3
+       cmpdi   cr6,r6,3
+       andi.   r5,r5,7
+       mtctr   r7
+       subfic  r11,r10,64
+       add     r5,r5,r0
+
+       bt      cr7*4+0,0f
+
+       ld      r9,0(r4)        # 3+2n loads, 2+2n stores
+       ld      r0,8(r4)
+       sld     r6,r9,r10
+       ldu     r9,16(r4)
+       srd     r7,r0,r11
+       sld     r8,r0,r10
+       or      r7,r7,r6
+       blt     cr6,4f
+       ld      r0,8(r4)
+       # s1<< in r8, d0=(s0<<|s1>>) in r7, s3 in r0, s2 in r9, nix in r6 & r12
+       b       2f
+
+0:     ld      r0,0(r4)        # 4+2n loads, 3+2n stores
+       ldu     r9,8(r4)
+       sld     r8,r0,r10
+       addi    r3,r3,-8
+       blt     cr6,5f
+       ld      r0,8(r4)
+       srd     r12,r9,r11
+       sld     r6,r9,r10
+       ldu     r9,16(r4)
+       or      r12,r8,r12
+       srd     r7,r0,r11
+       sld     r8,r0,r10
+       addi    r3,r3,16
+       beq     cr6,3f
+
+       # d0=(s0<<|s1>>) in r12, s1<< in r6, s2>> in r7, s2<< in r8, s3 in r9
+1:     or      r7,r7,r6
+       ld      r0,8(r4)
+       std     r12,8(r3)
+2:     srd     r12,r9,r11
+       sld     r6,r9,r10
+       ldu     r9,16(r4)
+       or      r12,r8,r12
+       stdu    r7,16(r3)
+       srd     r7,r0,r11
+       sld     r8,r0,r10
+       bdnz    1b
+
+3:     std     r12,8(r3)
+       or      r7,r7,r6
+4:     std     r7,16(r3)
+5:     srd     r12,r9,r11
+       or      r12,r8,r12
+       std     r12,24(r3)
+       beqlr
+       cmpwi   cr1,r5,8
+       addi    r3,r3,32
+       sld     r9,r9,r10
+       ble     cr1,.Ldo_tail
+       ld      r0,8(r4)
+       srd     r7,r0,r11
+       or      r9,r7,r9
+       b       .Ldo_tail
+
+.Ldst_unaligned:
+       mtcrf   0x01,r6         # put #bytes to 8B bdry into cr7
+       subf    r5,r6,r5
+       li      r7,0
+       cmpldi  r1,r5,16
+       bf      cr7*4+3,1f
+       lbz     r0,0(r4)
+       stb     r0,0(r3)
+       addi    r7,r7,1
+1:     bf      cr7*4+2,2f
+       lhzx    r0,r7,r4
+       sthx    r0,r7,r3
+       addi    r7,r7,2
+2:     bf      cr7*4+1,3f
+       lwzx    r0,r7,r4
+       stwx    r0,r7,r3
+3:     mtcrf   0x01,r5
+       add     r4,r6,r4
+       add     r3,r6,r3
+       b       .Ldst_aligned
+
+.Lshort_copy:
+       bf      cr7*4+0,1f
+       lwz     r0,0(r4)
+       lwz     r9,4(r4)
+       addi    r4,r4,8
+       stw     r0,0(r3)
+       stw     r9,4(r3)
+       addi    r3,r3,8
+1:     bf      cr7*4+1,2f
+       lwz     r0,0(r4)
+       addi    r4,r4,4
+       stw     r0,0(r3)
+       addi    r3,r3,4
+2:     bf      cr7*4+2,3f
+       lhz     r0,0(r4)
+       addi    r4,r4,2
+       sth     r0,0(r3)
+       addi    r3,r3,2
+3:     bf      cr7*4+3,4f
+       lbz     r0,0(r4)
+       stb     r0,0(r3)
+4:     blr
diff --git a/arch/powerpc/lib/rheap.c b/arch/powerpc/lib/rheap.c
new file mode 100644 (file)
index 0000000..42c5de2
--- /dev/null
@@ -0,0 +1,693 @@
+/*
+ * arch/ppc/syslib/rheap.c
+ *
+ * A Remote Heap.  Remote means that we don't touch the memory that the
+ * heap points to. Normal heap implementations use the memory they manage
+ * to place their list. We cannot do that because the memory we manage may
+ * have special properties, for example it is uncachable or of different
+ * endianess.
+ *
+ * Author: Pantelis Antoniou <panto@intracom.gr>
+ *
+ * 2004 (c) INTRACOM S.A. Greece. This file is licensed under
+ * the terms of the GNU General Public License version 2. This program
+ * is licensed "as is" without any warranty of any kind, whether express
+ * or implied.
+ */
+#include <linux/types.h>
+#include <linux/errno.h>
+#include <linux/mm.h>
+#include <linux/slab.h>
+
+#include <asm/rheap.h>
+
+/*
+ * Fixup a list_head, needed when copying lists.  If the pointers fall
+ * between s and e, apply the delta.  This assumes that
+ * sizeof(struct list_head *) == sizeof(unsigned long *).
+ */
+static inline void fixup(unsigned long s, unsigned long e, int d,
+                        struct list_head *l)
+{
+       unsigned long *pp;
+
+       pp = (unsigned long *)&l->next;
+       if (*pp >= s && *pp < e)
+               *pp += d;
+
+       pp = (unsigned long *)&l->prev;
+       if (*pp >= s && *pp < e)
+               *pp += d;
+}
+
+/* Grow the allocated blocks */
+static int grow(rh_info_t * info, int max_blocks)
+{
+       rh_block_t *block, *blk;
+       int i, new_blocks;
+       int delta;
+       unsigned long blks, blke;
+
+       if (max_blocks <= info->max_blocks)
+               return -EINVAL;
+
+       new_blocks = max_blocks - info->max_blocks;
+
+       block = kmalloc(sizeof(rh_block_t) * max_blocks, GFP_KERNEL);
+       if (block == NULL)
+               return -ENOMEM;
+
+       if (info->max_blocks > 0) {
+
+               /* copy old block area */
+               memcpy(block, info->block,
+                      sizeof(rh_block_t) * info->max_blocks);
+
+               delta = (char *)block - (char *)info->block;
+
+               /* and fixup list pointers */
+               blks = (unsigned long)info->block;
+               blke = (unsigned long)(info->block + info->max_blocks);
+
+               for (i = 0, blk = block; i < info->max_blocks; i++, blk++)
+                       fixup(blks, blke, delta, &blk->list);
+
+               fixup(blks, blke, delta, &info->empty_list);
+               fixup(blks, blke, delta, &info->free_list);
+               fixup(blks, blke, delta, &info->taken_list);
+
+               /* free the old allocated memory */
+               if ((info->flags & RHIF_STATIC_BLOCK) == 0)
+                       kfree(info->block);
+       }
+
+       info->block = block;
+       info->empty_slots += new_blocks;
+       info->max_blocks = max_blocks;
+       info->flags &= ~RHIF_STATIC_BLOCK;
+
+       /* add all new blocks to the free list */
+       for (i = 0, blk = block + info->max_blocks; i < new_blocks; i++, blk++)
+               list_add(&blk->list, &info->empty_list);
+
+       return 0;
+}
+
+/*
+ * Assure at least the required amount of empty slots.  If this function
+ * causes a grow in the block area then all pointers kept to the block
+ * area are invalid!
+ */
+static int assure_empty(rh_info_t * info, int slots)
+{
+       int max_blocks;
+
+       /* This function is not meant to be used to grow uncontrollably */
+       if (slots >= 4)
+               return -EINVAL;
+
+       /* Enough space */
+       if (info->empty_slots >= slots)
+               return 0;
+
+       /* Next 16 sized block */
+       max_blocks = ((info->max_blocks + slots) + 15) & ~15;
+
+       return grow(info, max_blocks);
+}
+
+static rh_block_t *get_slot(rh_info_t * info)
+{
+       rh_block_t *blk;
+
+       /* If no more free slots, and failure to extend. */
+       /* XXX: You should have called assure_empty before */
+       if (info->empty_slots == 0) {
+               printk(KERN_ERR "rh: out of slots; crash is imminent.\n");
+               return NULL;
+       }
+
+       /* Get empty slot to use */
+       blk = list_entry(info->empty_list.next, rh_block_t, list);
+       list_del_init(&blk->list);
+       info->empty_slots--;
+
+       /* Initialize */
+       blk->start = NULL;
+       blk->size = 0;
+       blk->owner = NULL;
+
+       return blk;
+}
+
+static inline void release_slot(rh_info_t * info, rh_block_t * blk)
+{
+       list_add(&blk->list, &info->empty_list);
+       info->empty_slots++;
+}
+
+static void attach_free_block(rh_info_t * info, rh_block_t * blkn)
+{
+       rh_block_t *blk;
+       rh_block_t *before;
+       rh_block_t *after;
+       rh_block_t *next;
+       int size;
+       unsigned long s, e, bs, be;
+       struct list_head *l;
+
+       /* We assume that they are aligned properly */
+       size = blkn->size;
+       s = (unsigned long)blkn->start;
+       e = s + size;
+
+       /* Find the blocks immediately before and after the given one
+        * (if any) */
+       before = NULL;
+       after = NULL;
+       next = NULL;
+
+       list_for_each(l, &info->free_list) {
+               blk = list_entry(l, rh_block_t, list);
+
+               bs = (unsigned long)blk->start;
+               be = bs + blk->size;
+
+               if (next == NULL && s >= bs)
+                       next = blk;
+
+               if (be == s)
+                       before = blk;
+
+               if (e == bs)
+                       after = blk;
+
+               /* If both are not null, break now */
+               if (before != NULL && after != NULL)
+                       break;
+       }
+
+       /* Now check if they are really adjacent */
+       if (before != NULL && s != (unsigned long)before->start + before->size)
+               before = NULL;
+
+       if (after != NULL && e != (unsigned long)after->start)
+               after = NULL;
+
+       /* No coalescing; list insert and return */
+       if (before == NULL && after == NULL) {
+
+               if (next != NULL)
+                       list_add(&blkn->list, &next->list);
+               else
+                       list_add(&blkn->list, &info->free_list);
+
+               return;
+       }
+
+       /* We don't need it anymore */
+       release_slot(info, blkn);
+
+       /* Grow the before block */
+       if (before != NULL && after == NULL) {
+               before->size += size;
+               return;
+       }
+
+       /* Grow the after block backwards */
+       if (before == NULL && after != NULL) {
+               after->start = (int8_t *)after->start - size;
+               after->size += size;
+               return;
+       }
+
+       /* Grow the before block, and release the after block */
+       before->size += size + after->size;
+       list_del(&after->list);
+       release_slot(info, after);
+}
+
+static void attach_taken_block(rh_info_t * info, rh_block_t * blkn)
+{
+       rh_block_t *blk;
+       struct list_head *l;
+
+       /* Find the block immediately before the given one (if any) */
+       list_for_each(l, &info->taken_list) {
+               blk = list_entry(l, rh_block_t, list);
+               if (blk->start > blkn->start) {
+                       list_add_tail(&blkn->list, &blk->list);
+                       return;
+               }
+       }
+
+       list_add_tail(&blkn->list, &info->taken_list);
+}
+
+/*
+ * Create a remote heap dynamically.  Note that no memory for the blocks
+ * are allocated.  It will upon the first allocation
+ */
+rh_info_t *rh_create(unsigned int alignment)
+{
+       rh_info_t *info;
+
+       /* Alignment must be a power of two */
+       if ((alignment & (alignment - 1)) != 0)
+               return ERR_PTR(-EINVAL);
+
+       info = kmalloc(sizeof(*info), GFP_KERNEL);
+       if (info == NULL)
+               return ERR_PTR(-ENOMEM);
+
+       info->alignment = alignment;
+
+       /* Initially everything as empty */
+       info->block = NULL;
+       info->max_blocks = 0;
+       info->empty_slots = 0;
+       info->flags = 0;
+
+       INIT_LIST_HEAD(&info->empty_list);
+       INIT_LIST_HEAD(&info->free_list);
+       INIT_LIST_HEAD(&info->taken_list);
+
+       return info;
+}
+
+/*
+ * Destroy a dynamically created remote heap.  Deallocate only if the areas
+ * are not static
+ */
+void rh_destroy(rh_info_t * info)
+{
+       if ((info->flags & RHIF_STATIC_BLOCK) == 0 && info->block != NULL)
+               kfree(info->block);
+
+       if ((info->flags & RHIF_STATIC_INFO) == 0)
+               kfree(info);
+}
+
+/*
+ * Initialize in place a remote heap info block.  This is needed to support
+ * operation very early in the startup of the kernel, when it is not yet safe
+ * to call kmalloc.
+ */
+void rh_init(rh_info_t * info, unsigned int alignment, int max_blocks,
+            rh_block_t * block)
+{
+       int i;
+       rh_block_t *blk;
+
+       /* Alignment must be a power of two */
+       if ((alignment & (alignment - 1)) != 0)
+               return;
+
+       info->alignment = alignment;
+
+       /* Initially everything as empty */
+       info->block = block;
+       info->max_blocks = max_blocks;
+       info->empty_slots = max_blocks;
+       info->flags = RHIF_STATIC_INFO | RHIF_STATIC_BLOCK;
+
+       INIT_LIST_HEAD(&info->empty_list);
+       INIT_LIST_HEAD(&info->free_list);
+       INIT_LIST_HEAD(&info->taken_list);
+
+       /* Add all new blocks to the free list */
+       for (i = 0, blk = block; i < max_blocks; i++, blk++)
+               list_add(&blk->list, &info->empty_list);
+}
+
+/* Attach a free memory region, coalesces regions if adjuscent */
+int rh_attach_region(rh_info_t * info, void *start, int size)
+{
+       rh_block_t *blk;
+       unsigned long s, e, m;
+       int r;
+
+       /* The region must be aligned */
+       s = (unsigned long)start;
+       e = s + size;
+       m = info->alignment - 1;
+
+       /* Round start up */
+       s = (s + m) & ~m;
+
+       /* Round end down */
+       e = e & ~m;
+
+       /* Take final values */
+       start = (void *)s;
+       size = (int)(e - s);
+
+       /* Grow the blocks, if needed */
+       r = assure_empty(info, 1);
+       if (r < 0)
+               return r;
+
+       blk = get_slot(info);
+       blk->start = start;
+       blk->size = size;
+       blk->owner = NULL;
+
+       attach_free_block(info, blk);
+
+       return 0;
+}
+
+/* Detatch given address range, splits free block if needed. */
+void *rh_detach_region(rh_info_t * info, void *start, int size)
+{
+       struct list_head *l;
+       rh_block_t *blk, *newblk;
+       unsigned long s, e, m, bs, be;
+
+       /* Validate size */
+       if (size <= 0)
+               return ERR_PTR(-EINVAL);
+
+       /* The region must be aligned */
+       s = (unsigned long)start;
+       e = s + size;
+       m = info->alignment - 1;
+
+       /* Round start up */
+       s = (s + m) & ~m;
+
+       /* Round end down */
+       e = e & ~m;
+
+       if (assure_empty(info, 1) < 0)
+               return ERR_PTR(-ENOMEM);
+
+       blk = NULL;
+       list_for_each(l, &info->free_list) {
+               blk = list_entry(l, rh_block_t, list);
+               /* The range must lie entirely inside one free block */
+               bs = (unsigned long)blk->start;
+               be = (unsigned long)blk->start + blk->size;
+               if (s >= bs && e <= be)
+                       break;
+               blk = NULL;
+       }
+
+       if (blk == NULL)
+               return ERR_PTR(-ENOMEM);
+
+       /* Perfect fit */
+       if (bs == s && be == e) {
+               /* Delete from free list, release slot */
+               list_del(&blk->list);
+               release_slot(info, blk);
+               return (void *)s;
+       }
+
+       /* blk still in free list, with updated start and/or size */
+       if (bs == s || be == e) {
+               if (bs == s)
+                       blk->start = (int8_t *)blk->start + size;
+               blk->size -= size;
+
+       } else {
+               /* The front free fragment */
+               blk->size = s - bs;
+
+               /* the back free fragment */
+               newblk = get_slot(info);
+               newblk->start = (void *)e;
+               newblk->size = be - e;
+
+               list_add(&newblk->list, &blk->list);
+       }
+
+       return (void *)s;
+}
+
+void *rh_alloc(rh_info_t * info, int size, const char *owner)
+{
+       struct list_head *l;
+       rh_block_t *blk;
+       rh_block_t *newblk;
+       void *start;
+
+       /* Validate size */
+       if (size <= 0)
+               return ERR_PTR(-EINVAL);
+
+       /* Align to configured alignment */
+       size = (size + (info->alignment - 1)) & ~(info->alignment - 1);
+
+       if (assure_empty(info, 1) < 0)
+               return ERR_PTR(-ENOMEM);
+
+       blk = NULL;
+       list_for_each(l, &info->free_list) {
+               blk = list_entry(l, rh_block_t, list);
+               if (size <= blk->size)
+                       break;
+               blk = NULL;
+       }
+
+       if (blk == NULL)
+               return ERR_PTR(-ENOMEM);
+
+       /* Just fits */
+       if (blk->size == size) {
+               /* Move from free list to taken list */
+               list_del(&blk->list);
+               blk->owner = owner;
+               start = blk->start;
+
+               attach_taken_block(info, blk);
+
+               return start;
+       }
+
+       newblk = get_slot(info);
+       newblk->start = blk->start;
+       newblk->size = size;
+       newblk->owner = owner;
+
+       /* blk still in free list, with updated start, size */
+       blk->start = (int8_t *)blk->start + size;
+       blk->size -= size;
+
+       start = newblk->start;
+
+       attach_taken_block(info, newblk);
+
+       return start;
+}
+
+/* allocate at precisely the given address */
+void *rh_alloc_fixed(rh_info_t * info, void *start, int size, const char *owner)
+{
+       struct list_head *l;
+       rh_block_t *blk, *newblk1, *newblk2;
+       unsigned long s, e, m, bs, be;
+
+       /* Validate size */
+       if (size <= 0)
+               return ERR_PTR(-EINVAL);
+
+       /* The region must be aligned */
+       s = (unsigned long)start;
+       e = s + size;
+       m = info->alignment - 1;
+
+       /* Round start up */
+       s = (s + m) & ~m;
+
+       /* Round end down */
+       e = e & ~m;
+
+       if (assure_empty(info, 2) < 0)
+               return ERR_PTR(-ENOMEM);
+
+       blk = NULL;
+       list_for_each(l, &info->free_list) {
+               blk = list_entry(l, rh_block_t, list);
+               /* The range must lie entirely inside one free block */
+               bs = (unsigned long)blk->start;
+               be = (unsigned long)blk->start + blk->size;
+               if (s >= bs && e <= be)
+                       break;
+       }
+
+       if (blk == NULL)
+               return ERR_PTR(-ENOMEM);
+
+       /* Perfect fit */
+       if (bs == s && be == e) {
+               /* Move from free list to taken list */
+               list_del(&blk->list);
+               blk->owner = owner;
+
+               start = blk->start;
+               attach_taken_block(info, blk);
+
+               return start;
+
+       }
+
+       /* blk still in free list, with updated start and/or size */
+       if (bs == s || be == e) {
+               if (bs == s)
+                       blk->start = (int8_t *)blk->start + size;
+               blk->size -= size;
+
+       } else {
+               /* The front free fragment */
+               blk->size = s - bs;
+
+               /* The back free fragment */
+               newblk2 = get_slot(info);
+               newblk2->start = (void *)e;
+               newblk2->size = be - e;
+
+               list_add(&newblk2->list, &blk->list);
+       }
+
+       newblk1 = get_slot(info);
+       newblk1->start = (void *)s;
+       newblk1->size = e - s;
+       newblk1->owner = owner;
+
+       start = newblk1->start;
+       attach_taken_block(info, newblk1);
+
+       return start;
+}
+
+int rh_free(rh_info_t * info, void *start)
+{
+       rh_block_t *blk, *blk2;
+       struct list_head *l;
+       int size;
+
+       /* Linear search for block */
+       blk = NULL;
+       list_for_each(l, &info->taken_list) {
+               blk2 = list_entry(l, rh_block_t, list);
+               if (start < blk2->start)
+                       break;
+               blk = blk2;
+       }
+
+       if (blk == NULL || start > (blk->start + blk->size))
+               return -EINVAL;
+
+       /* Remove from taken list */
+       list_del(&blk->list);
+
+       /* Get size of freed block */
+       size = blk->size;
+       attach_free_block(info, blk);
+
+       return size;
+}
+
+int rh_get_stats(rh_info_t * info, int what, int max_stats, rh_stats_t * stats)
+{
+       rh_block_t *blk;
+       struct list_head *l;
+       struct list_head *h;
+       int nr;
+
+       switch (what) {
+
+       case RHGS_FREE:
+               h = &info->free_list;
+               break;
+
+       case RHGS_TAKEN:
+               h = &info->taken_list;
+               break;
+
+       default:
+               return -EINVAL;
+       }
+
+       /* Linear search for block */
+       nr = 0;
+       list_for_each(l, h) {
+               blk = list_entry(l, rh_block_t, list);
+               if (stats != NULL && nr < max_stats) {
+                       stats->start = blk->start;
+                       stats->size = blk->size;
+                       stats->owner = blk->owner;
+                       stats++;
+               }
+               nr++;
+       }
+
+       return nr;
+}
+
+int rh_set_owner(rh_info_t * info, void *start, const char *owner)
+{
+       rh_block_t *blk, *blk2;
+       struct list_head *l;
+       int size;
+
+       /* Linear search for block */
+       blk = NULL;
+       list_for_each(l, &info->taken_list) {
+               blk2 = list_entry(l, rh_block_t, list);
+               if (start < blk2->start)
+                       break;
+               blk = blk2;
+       }
+
+       if (blk == NULL || start > (blk->start + blk->size))
+               return -EINVAL;
+
+       blk->owner = owner;
+       size = blk->size;
+
+       return size;
+}
+
+void rh_dump(rh_info_t * info)
+{
+       static rh_stats_t st[32];       /* XXX maximum 32 blocks */
+       int maxnr;
+       int i, nr;
+
+       maxnr = sizeof(st) / sizeof(st[0]);
+
+       printk(KERN_INFO
+              "info @0x%p (%d slots empty / %d max)\n",
+              info, info->empty_slots, info->max_blocks);
+
+       printk(KERN_INFO "  Free:\n");
+       nr = rh_get_stats(info, RHGS_FREE, maxnr, st);
+       if (nr > maxnr)
+               nr = maxnr;
+       for (i = 0; i < nr; i++)
+               printk(KERN_INFO
+                      "    0x%p-0x%p (%u)\n",
+                      st[i].start, (int8_t *) st[i].start + st[i].size,
+                      st[i].size);
+       printk(KERN_INFO "\n");
+
+       printk(KERN_INFO "  Taken:\n");
+       nr = rh_get_stats(info, RHGS_TAKEN, maxnr, st);
+       if (nr > maxnr)
+               nr = maxnr;
+       for (i = 0; i < nr; i++)
+               printk(KERN_INFO
+                      "    0x%p-0x%p (%u) %s\n",
+                      st[i].start, (int8_t *) st[i].start + st[i].size,
+                      st[i].size, st[i].owner != NULL ? st[i].owner : "");
+       printk(KERN_INFO "\n");
+}
+
+void rh_dump_blk(rh_info_t * info, rh_block_t * blk)
+{
+       printk(KERN_INFO
+              "blk @0x%p: 0x%p-0x%p (%u)\n",
+              blk, blk->start, (int8_t *) blk->start + blk->size, blk->size);
+}
diff --git a/arch/powerpc/lib/sstep.c b/arch/powerpc/lib/sstep.c
new file mode 100644 (file)
index 0000000..e79123d
--- /dev/null
@@ -0,0 +1,141 @@
+/*
+ * Single-step support.
+ *
+ * Copyright (C) 2004 Paul Mackerras <paulus@au.ibm.com>, IBM
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version
+ * 2 of the License, or (at your option) any later version.
+ */
+#include <linux/kernel.h>
+#include <linux/ptrace.h>
+#include <asm/sstep.h>
+#include <asm/processor.h>
+
+extern char system_call_common[];
+
+/* Bits in SRR1 that are copied from MSR */
+#define MSR_MASK       0xffffffff87c0ffff
+
+/*
+ * Determine whether a conditional branch instruction would branch.
+ */
+static int branch_taken(unsigned int instr, struct pt_regs *regs)
+{
+       unsigned int bo = (instr >> 21) & 0x1f;
+       unsigned int bi;
+
+       if ((bo & 4) == 0) {
+               /* decrement counter */
+               --regs->ctr;
+               if (((bo >> 1) & 1) ^ (regs->ctr == 0))
+                       return 0;
+       }
+       if ((bo & 0x10) == 0) {
+               /* check bit from CR */
+               bi = (instr >> 16) & 0x1f;
+               if (((regs->ccr >> (31 - bi)) & 1) != ((bo >> 3) & 1))
+                       return 0;
+       }
+       return 1;
+}
+
+/*
+ * Emulate instructions that cause a transfer of control.
+ * Returns 1 if the step was emulated, 0 if not,
+ * or -1 if the instruction is one that should not be stepped,
+ * such as an rfid, or a mtmsrd that would clear MSR_RI.
+ */
+int emulate_step(struct pt_regs *regs, unsigned int instr)
+{
+       unsigned int opcode, rd;
+       unsigned long int imm;
+
+       opcode = instr >> 26;
+       switch (opcode) {
+       case 16:        /* bc */
+               imm = (signed short)(instr & 0xfffc);
+               if ((instr & 2) == 0)
+                       imm += regs->nip;
+               regs->nip += 4;
+               if ((regs->msr & MSR_SF) == 0)
+                       regs->nip &= 0xffffffffUL;
+               if (instr & 1)
+                       regs->link = regs->nip;
+               if (branch_taken(instr, regs))
+                       regs->nip = imm;
+               return 1;
+       case 17:        /* sc */
+               /*
+                * N.B. this uses knowledge about how the syscall
+                * entry code works.  If that is changed, this will
+                * need to be changed also.
+                */
+               regs->gpr[9] = regs->gpr[13];
+               regs->gpr[11] = regs->nip + 4;
+               regs->gpr[12] = regs->msr & MSR_MASK;
+               regs->gpr[13] = (unsigned long) get_paca();
+               regs->nip = (unsigned long) &system_call_common;
+               regs->msr = MSR_KERNEL;
+               return 1;
+       case 18:        /* b */
+               imm = instr & 0x03fffffc;
+               if (imm & 0x02000000)
+                       imm -= 0x04000000;
+               if ((instr & 2) == 0)
+                       imm += regs->nip;
+               if (instr & 1) {
+                       regs->link = regs->nip + 4;
+                       if ((regs->msr & MSR_SF) == 0)
+                               regs->link &= 0xffffffffUL;
+               }
+               if ((regs->msr & MSR_SF) == 0)
+                       imm &= 0xffffffffUL;
+               regs->nip = imm;
+               return 1;
+       case 19:
+               switch (instr & 0x7fe) {
+               case 0x20:      /* bclr */
+               case 0x420:     /* bcctr */
+                       imm = (instr & 0x400)? regs->ctr: regs->link;
+                       regs->nip += 4;
+                       if ((regs->msr & MSR_SF) == 0) {
+                               regs->nip &= 0xffffffffUL;
+                               imm &= 0xffffffffUL;
+                       }
+                       if (instr & 1)
+                               regs->link = regs->nip;
+                       if (branch_taken(instr, regs))
+                               regs->nip = imm;
+                       return 1;
+               case 0x24:      /* rfid, scary */
+                       return -1;
+               }
+       case 31:
+               rd = (instr >> 21) & 0x1f;
+               switch (instr & 0x7fe) {
+               case 0xa6:      /* mfmsr */
+                       regs->gpr[rd] = regs->msr & MSR_MASK;
+                       regs->nip += 4;
+                       if ((regs->msr & MSR_SF) == 0)
+                               regs->nip &= 0xffffffffUL;
+                       return 1;
+               case 0x164:     /* mtmsrd */
+                       /* only MSR_EE and MSR_RI get changed if bit 15 set */
+                       /* mtmsrd doesn't change MSR_HV and MSR_ME */
+                       imm = (instr & 0x10000)? 0x8002: 0xefffffffffffefffUL;
+                       imm = (regs->msr & MSR_MASK & ~imm)
+                               | (regs->gpr[rd] & imm);
+                       if ((imm & MSR_RI) == 0)
+                               /* can't step mtmsrd that would clear MSR_RI */
+                               return -1;
+                       regs->msr = imm;
+                       regs->nip += 4;
+                       if ((imm & MSR_SF) == 0)
+                               regs->nip &= 0xffffffffUL;
+                       return 1;
+               }
+       }
+       return 0;
+}
diff --git a/arch/powerpc/lib/strcase.c b/arch/powerpc/lib/strcase.c
new file mode 100644 (file)
index 0000000..36b5210
--- /dev/null
@@ -0,0 +1,23 @@
+#include <linux/ctype.h>
+
+int strcasecmp(const char *s1, const char *s2)
+{
+       int c1, c2;
+
+       do {
+               c1 = tolower(*s1++);
+               c2 = tolower(*s2++);
+       } while (c1 == c2 && c1 != 0);
+       return c1 - c2;
+}
+
+int strncasecmp(const char *s1, const char *s2, int n)
+{
+       int c1, c2;
+
+       do {
+               c1 = tolower(*s1++);
+               c2 = tolower(*s2++);
+       } while ((--n > 0) && c1 == c2 && c1 != 0);
+       return c1 - c2;
+}
diff --git a/arch/powerpc/lib/string.S b/arch/powerpc/lib/string.S
new file mode 100644 (file)
index 0000000..15d40e9
--- /dev/null
@@ -0,0 +1,203 @@
+/*
+ * String handling functions for PowerPC.
+ *
+ * Copyright (C) 1996 Paul Mackerras.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version
+ * 2 of the License, or (at your option) any later version.
+ */
+#include <linux/config.h>
+#include <asm/processor.h>
+#include <asm/errno.h>
+#include <asm/ppc_asm.h>
+
+       .text
+       .stabs  "arch/powerpc/lib/",N_SO,0,0,0f
+       .stabs  "string.S",N_SO,0,0,0f
+0:
+
+       .section __ex_table,"a"
+#ifdef CONFIG_PPC64
+       .align  3
+#define EXTBL  .llong
+#else
+       .align  2
+#define EXTBL  .long
+#endif
+       .text
+       
+_GLOBAL(strcpy)
+       addi    r5,r3,-1
+       addi    r4,r4,-1
+1:     lbzu    r0,1(r4)
+       cmpwi   0,r0,0
+       stbu    r0,1(r5)
+       bne     1b
+       blr
+
+/* This clears out any unused part of the destination buffer,
+   just as the libc version does.  -- paulus */
+_GLOBAL(strncpy)
+       cmpwi   0,r5,0
+       beqlr
+       mtctr   r5
+       addi    r6,r3,-1
+       addi    r4,r4,-1
+1:     lbzu    r0,1(r4)
+       cmpwi   0,r0,0
+       stbu    r0,1(r6)
+       bdnzf   2,1b            /* dec ctr, branch if ctr != 0 && !cr0.eq */
+       bnelr                   /* if we didn't hit a null char, we're done */
+       mfctr   r5
+       cmpwi   0,r5,0          /* any space left in destination buffer? */
+       beqlr                   /* we know r0 == 0 here */
+2:     stbu    r0,1(r6)        /* clear it out if so */
+       bdnz    2b
+       blr
+
+_GLOBAL(strcat)
+       addi    r5,r3,-1
+       addi    r4,r4,-1
+1:     lbzu    r0,1(r5)
+       cmpwi   0,r0,0
+       bne     1b
+       addi    r5,r5,-1
+1:     lbzu    r0,1(r4)
+       cmpwi   0,r0,0
+       stbu    r0,1(r5)
+       bne     1b
+       blr
+
+_GLOBAL(strcmp)
+       addi    r5,r3,-1
+       addi    r4,r4,-1
+1:     lbzu    r3,1(r5)
+       cmpwi   1,r3,0
+       lbzu    r0,1(r4)
+       subf.   r3,r0,r3
+       beqlr   1
+       beq     1b
+       blr
+
+_GLOBAL(strlen)
+       addi    r4,r3,-1
+1:     lbzu    r0,1(r4)
+       cmpwi   0,r0,0
+       bne     1b
+       subf    r3,r3,r4
+       blr
+
+_GLOBAL(memcmp)
+       cmpwi   0,r5,0
+       ble-    2f
+       mtctr   r5
+       addi    r6,r3,-1
+       addi    r4,r4,-1
+1:     lbzu    r3,1(r6)
+       lbzu    r0,1(r4)
+       subf.   r3,r0,r3
+       bdnzt   2,1b
+       blr
+2:     li      r3,0
+       blr
+
+_GLOBAL(memchr)
+       cmpwi   0,r5,0
+       ble-    2f
+       mtctr   r5
+       addi    r3,r3,-1
+1:     lbzu    r0,1(r3)
+       cmpw    0,r0,r4
+       bdnzf   2,1b
+       beqlr
+2:     li      r3,0
+       blr
+
+_GLOBAL(__clear_user)
+       addi    r6,r3,-4
+       li      r3,0
+       li      r5,0
+       cmplwi  0,r4,4
+       blt     7f
+       /* clear a single word */
+11:    stwu    r5,4(r6)
+       beqlr
+       /* clear word sized chunks */
+       andi.   r0,r6,3
+       add     r4,r0,r4
+       subf    r6,r0,r6
+       srwi    r0,r4,2
+       andi.   r4,r4,3
+       mtctr   r0
+       bdz     7f
+1:     stwu    r5,4(r6)
+       bdnz    1b
+       /* clear byte sized chunks */
+7:     cmpwi   0,r4,0
+       beqlr
+       mtctr   r4
+       addi    r6,r6,3
+8:     stbu    r5,1(r6)
+       bdnz    8b
+       blr
+90:    mr      r3,r4
+       blr
+91:    mfctr   r3
+       slwi    r3,r3,2
+       add     r3,r3,r4
+       blr
+92:    mfctr   r3
+       blr
+
+       .section __ex_table,"a"
+       EXTBL   11b,90b
+       EXTBL   1b,91b
+       EXTBL   8b,92b
+       .text
+
+_GLOBAL(__strncpy_from_user)
+       addi    r6,r3,-1
+       addi    r4,r4,-1
+       cmpwi   0,r5,0
+       beq     2f
+       mtctr   r5
+1:     lbzu    r0,1(r4)
+       cmpwi   0,r0,0
+       stbu    r0,1(r6)
+       bdnzf   2,1b            /* dec ctr, branch if ctr != 0 && !cr0.eq */
+       beq     3f
+2:     addi    r6,r6,1
+3:     subf    r3,r3,r6
+       blr
+99:    li      r3,-EFAULT
+       blr
+
+       .section __ex_table,"a"
+       EXTBL   1b,99b
+       .text
+
+/* r3 = str, r4 = len (> 0), r5 = top (highest addr) */
+_GLOBAL(__strnlen_user)
+       addi    r7,r3,-1
+       subf    r6,r7,r5        /* top+1 - str */
+       cmplw   0,r4,r6
+       bge     0f
+       mr      r6,r4
+0:     mtctr   r6              /* ctr = min(len, top - str) */
+1:     lbzu    r0,1(r7)        /* get next byte */
+       cmpwi   0,r0,0
+       bdnzf   2,1b            /* loop if --ctr != 0 && byte != 0 */
+       addi    r7,r7,1
+       subf    r3,r3,r7        /* number of bytes we have looked at */
+       beqlr                   /* return if we found a 0 byte */
+       cmpw    0,r3,r4         /* did we look at all len bytes? */
+       blt     99f             /* if not, must have hit top */
+       addi    r3,r4,1         /* return len + 1 to indicate no null found */
+       blr
+99:    li      r3,0            /* bad address, return 0 */
+       blr
+
+       .section __ex_table,"a"
+       EXTBL   1b,99b
diff --git a/arch/powerpc/lib/usercopy.c b/arch/powerpc/lib/usercopy.c
new file mode 100644 (file)
index 0000000..5eea6f3
--- /dev/null
@@ -0,0 +1,41 @@
+/*
+ * Functions which are too large to be inlined.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version
+ * 2 of the License, or (at your option) any later version.
+ */
+#include <linux/module.h>
+#include <asm/uaccess.h>
+
+unsigned long copy_from_user(void *to, const void __user *from, unsigned long n)
+{
+       if (likely(access_ok(VERIFY_READ, from, n)))
+               n = __copy_from_user(to, from, n);
+       else
+               memset(to, 0, n);
+       return n;
+}
+
+unsigned long copy_to_user(void __user *to, const void *from, unsigned long n)
+{
+       if (likely(access_ok(VERIFY_WRITE, to, n)))
+               n = __copy_to_user(to, from, n);
+       return n;
+}
+
+unsigned long copy_in_user(void __user *to, const void __user *from,
+                          unsigned long n)
+{
+       might_sleep();
+       if (likely(access_ok(VERIFY_READ, from, n) &&
+           access_ok(VERIFY_WRITE, to, n)))
+               n =__copy_tofrom_user(to, from, n);
+       return n;
+}
+
+EXPORT_SYMBOL(copy_from_user);
+EXPORT_SYMBOL(copy_to_user);
+EXPORT_SYMBOL(copy_in_user);
+
diff --git a/arch/powerpc/mm/44x_mmu.c b/arch/powerpc/mm/44x_mmu.c
new file mode 100644 (file)
index 0000000..3d79ce2
--- /dev/null
@@ -0,0 +1,120 @@
+/*
+ * Modifications by Matt Porter (mporter@mvista.com) to support
+ * PPC44x Book E processors.
+ *
+ * This file contains the routines for initializing the MMU
+ * on the 4xx series of chips.
+ *  -- paulus
+ *
+ *  Derived from arch/ppc/mm/init.c:
+ *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
+ *
+ *  Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
+ *  and Cort Dougan (PReP) (cort@cs.nmt.edu)
+ *    Copyright (C) 1996 Paul Mackerras
+ *  Amiga/APUS changes by Jesper Skov (jskov@cygnus.co.uk).
+ *
+ *  Derived from "arch/i386/mm/init.c"
+ *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
+ *
+ *  This program is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU General Public License
+ *  as published by the Free Software Foundation; either version
+ *  2 of the License, or (at your option) any later version.
+ *
+ */
+
+#include <linux/config.h>
+#include <linux/signal.h>
+#include <linux/sched.h>
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/string.h>
+#include <linux/types.h>
+#include <linux/ptrace.h>
+#include <linux/mman.h>
+#include <linux/mm.h>
+#include <linux/swap.h>
+#include <linux/stddef.h>
+#include <linux/vmalloc.h>
+#include <linux/init.h>
+#include <linux/delay.h>
+#include <linux/highmem.h>
+
+#include <asm/pgalloc.h>
+#include <asm/prom.h>
+#include <asm/io.h>
+#include <asm/mmu_context.h>
+#include <asm/pgtable.h>
+#include <asm/mmu.h>
+#include <asm/uaccess.h>
+#include <asm/smp.h>
+#include <asm/bootx.h>
+#include <asm/machdep.h>
+#include <asm/setup.h>
+
+#include "mmu_decl.h"
+
+extern char etext[], _stext[];
+
+/* Used by the 44x TLB replacement exception handler.
+ * Just needed it declared someplace.
+ */
+unsigned int tlb_44x_index = 0;
+unsigned int tlb_44x_hwater = 62;
+
+/*
+ * "Pins" a 256MB TLB entry in AS0 for kernel lowmem
+ */
+static void __init
+ppc44x_pin_tlb(int slot, unsigned int virt, unsigned int phys)
+{
+       unsigned long attrib = 0;
+
+       __asm__ __volatile__("\
+       clrrwi  %2,%2,10\n\
+       ori     %2,%2,%4\n\
+       clrrwi  %1,%1,10\n\
+       li      %0,0\n\
+       ori     %0,%0,%5\n\
+       tlbwe   %2,%3,%6\n\
+       tlbwe   %1,%3,%7\n\
+       tlbwe   %0,%3,%8"
+       :
+       : "r" (attrib), "r" (phys), "r" (virt), "r" (slot),
+         "i" (PPC44x_TLB_VALID | PPC44x_TLB_256M),
+         "i" (PPC44x_TLB_SW | PPC44x_TLB_SR | PPC44x_TLB_SX | PPC44x_TLB_G),
+         "i" (PPC44x_TLB_PAGEID),
+         "i" (PPC44x_TLB_XLAT),
+         "i" (PPC44x_TLB_ATTRIB));
+}
+
+/*
+ * MMU_init_hw does the chip-specific initialization of the MMU hardware.
+ */
+void __init MMU_init_hw(void)
+{
+       flush_instruction_cache();
+}
+
+unsigned long __init mmu_mapin_ram(void)
+{
+       unsigned int pinned_tlbs = 1;
+       int i;
+
+       /* Determine number of entries necessary to cover lowmem */
+       pinned_tlbs = (unsigned int)
+               (_ALIGN(total_lowmem, PPC44x_PIN_SIZE) >> PPC44x_PIN_SHIFT);
+
+       /* Write upper watermark to save location */
+       tlb_44x_hwater = PPC44x_LOW_SLOT - pinned_tlbs;
+
+       /* If necessary, set additional pinned TLBs */
+       if (pinned_tlbs > 1)
+               for (i = (PPC44x_LOW_SLOT-(pinned_tlbs-1)); i < PPC44x_LOW_SLOT; i++) {
+                       unsigned int phys_addr = (PPC44x_LOW_SLOT-i) * PPC44x_PIN_SIZE;
+                       ppc44x_pin_tlb(i, phys_addr+PAGE_OFFSET, phys_addr);
+               }
+
+       return total_lowmem;
+}
diff --git a/arch/powerpc/mm/4xx_mmu.c b/arch/powerpc/mm/4xx_mmu.c
new file mode 100644 (file)
index 0000000..b7bcbc2
--- /dev/null
@@ -0,0 +1,141 @@
+/*
+ * This file contains the routines for initializing the MMU
+ * on the 4xx series of chips.
+ *  -- paulus
+ *
+ *  Derived from arch/ppc/mm/init.c:
+ *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
+ *
+ *  Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
+ *  and Cort Dougan (PReP) (cort@cs.nmt.edu)
+ *    Copyright (C) 1996 Paul Mackerras
+ *  Amiga/APUS changes by Jesper Skov (jskov@cygnus.co.uk).
+ *
+ *  Derived from "arch/i386/mm/init.c"
+ *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
+ *
+ *  This program is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU General Public License
+ *  as published by the Free Software Foundation; either version
+ *  2 of the License, or (at your option) any later version.
+ *
+ */
+
+#include <linux/config.h>
+#include <linux/signal.h>
+#include <linux/sched.h>
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/string.h>
+#include <linux/types.h>
+#include <linux/ptrace.h>
+#include <linux/mman.h>
+#include <linux/mm.h>
+#include <linux/swap.h>
+#include <linux/stddef.h>
+#include <linux/vmalloc.h>
+#include <linux/init.h>
+#include <linux/delay.h>
+#include <linux/highmem.h>
+
+#include <asm/pgalloc.h>
+#include <asm/prom.h>
+#include <asm/io.h>
+#include <asm/mmu_context.h>
+#include <asm/pgtable.h>
+#include <asm/mmu.h>
+#include <asm/uaccess.h>
+#include <asm/smp.h>
+#include <asm/bootx.h>
+#include <asm/machdep.h>
+#include <asm/setup.h>
+#include "mmu_decl.h"
+
+extern int __map_without_ltlbs;
+/*
+ * MMU_init_hw does the chip-specific initialization of the MMU hardware.
+ */
+void __init MMU_init_hw(void)
+{
+       /*
+        * The Zone Protection Register (ZPR) defines how protection will
+        * be applied to every page which is a member of a given zone. At
+        * present, we utilize only two of the 4xx's zones.
+        * The zone index bits (of ZSEL) in the PTE are used for software
+        * indicators, except the LSB.  For user access, zone 1 is used,
+        * for kernel access, zone 0 is used.  We set all but zone 1
+        * to zero, allowing only kernel access as indicated in the PTE.
+        * For zone 1, we set a 01 binary (a value of 10 will not work)
+        * to allow user access as indicated in the PTE.  This also allows
+        * kernel access as indicated in the PTE.
+        */
+
+        mtspr(SPRN_ZPR, 0x10000000);
+
+       flush_instruction_cache();
+
+       /*
+        * Set up the real-mode cache parameters for the exception vector
+        * handlers (which are run in real-mode).
+        */
+
+        mtspr(SPRN_DCWR, 0x00000000);  /* All caching is write-back */
+
+        /*
+        * Cache instruction and data space where the exception
+        * vectors and the kernel live in real-mode.
+        */
+
+        mtspr(SPRN_DCCR, 0xF0000000);  /* 512 MB of data space at 0x0. */
+        mtspr(SPRN_ICCR, 0xF0000000);  /* 512 MB of instr. space at 0x0. */
+}
+
+#define LARGE_PAGE_SIZE_16M    (1<<24)
+#define LARGE_PAGE_SIZE_4M     (1<<22)
+
+unsigned long __init mmu_mapin_ram(void)
+{
+       unsigned long v, s;
+       phys_addr_t p;
+
+       v = KERNELBASE;
+       p = PPC_MEMSTART;
+       s = 0;
+
+       if (__map_without_ltlbs) {
+               return s;
+       }
+
+       while (s <= (total_lowmem - LARGE_PAGE_SIZE_16M)) {
+               pmd_t *pmdp;
+               unsigned long val = p | _PMD_SIZE_16M | _PAGE_HWEXEC | _PAGE_HWWRITE;
+
+               spin_lock(&init_mm.page_table_lock);
+               pmdp = pmd_offset(pgd_offset_k(v), v);
+               pmd_val(*pmdp++) = val;
+               pmd_val(*pmdp++) = val;
+               pmd_val(*pmdp++) = val;
+               pmd_val(*pmdp++) = val;
+               spin_unlock(&init_mm.page_table_lock);
+
+               v += LARGE_PAGE_SIZE_16M;
+               p += LARGE_PAGE_SIZE_16M;
+               s += LARGE_PAGE_SIZE_16M;
+       }
+
+       while (s <= (total_lowmem - LARGE_PAGE_SIZE_4M)) {
+               pmd_t *pmdp;
+               unsigned long val = p | _PMD_SIZE_4M | _PAGE_HWEXEC | _PAGE_HWWRITE;
+
+               spin_lock(&init_mm.page_table_lock);
+               pmdp = pmd_offset(pgd_offset_k(v), v);
+               pmd_val(*pmdp) = val;
+               spin_unlock(&init_mm.page_table_lock);
+
+               v += LARGE_PAGE_SIZE_4M;
+               p += LARGE_PAGE_SIZE_4M;
+               s += LARGE_PAGE_SIZE_4M;
+       }
+
+       return s;
+}
diff --git a/arch/powerpc/mm/Makefile b/arch/powerpc/mm/Makefile
new file mode 100644 (file)
index 0000000..9f52c26
--- /dev/null
@@ -0,0 +1,12 @@
+#
+# Makefile for the linux ppc-specific parts of the memory manager.
+#
+
+obj-y                          := fault.o mem.o
+obj-$(CONFIG_PPC32)            += init.o pgtable.o mmu_context.o \
+                                  mem_pieces.o tlb.o
+obj-$(CONFIG_PPC64)            += init64.o pgtable64.o mmu_context64.o
+obj-$(CONFIG_PPC_STD_MMU_32)   += ppc_mmu.o hash_32.o
+obj-$(CONFIG_40x)              += 4xx_mmu.o
+obj-$(CONFIG_44x)              += 44x_mmu.o
+obj-$(CONFIG_FSL_BOOKE)                += fsl_booke_mmu.o
diff --git a/arch/powerpc/mm/fault.c b/arch/powerpc/mm/fault.c
new file mode 100644 (file)
index 0000000..3df641f
--- /dev/null
@@ -0,0 +1,391 @@
+/*
+ *  arch/ppc/mm/fault.c
+ *
+ *  PowerPC version
+ *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
+ *
+ *  Derived from "arch/i386/mm/fault.c"
+ *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
+ *
+ *  Modified by Cort Dougan and Paul Mackerras.
+ *
+ *  Modified for PPC64 by Dave Engebretsen (engebret@ibm.com)
+ *
+ *  This program is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU General Public License
+ *  as published by the Free Software Foundation; either version
+ *  2 of the License, or (at your option) any later version.
+ */
+
+#include <linux/config.h>
+#include <linux/signal.h>
+#include <linux/sched.h>
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/string.h>
+#include <linux/types.h>
+#include <linux/ptrace.h>
+#include <linux/mman.h>
+#include <linux/mm.h>
+#include <linux/interrupt.h>
+#include <linux/highmem.h>
+#include <linux/module.h>
+#include <linux/kprobes.h>
+
+#include <asm/page.h>
+#include <asm/pgtable.h>
+#include <asm/mmu.h>
+#include <asm/mmu_context.h>
+#include <asm/system.h>
+#include <asm/uaccess.h>
+#include <asm/tlbflush.h>
+#include <asm/kdebug.h>
+#include <asm/siginfo.h>
+
+/*
+ * Check whether the instruction at regs->nip is a store using
+ * an update addressing form which will update r1.
+ */
+static int store_updates_sp(struct pt_regs *regs)
+{
+       unsigned int inst;
+
+       if (get_user(inst, (unsigned int __user *)regs->nip))
+               return 0;
+       /* check for 1 in the rA field */
+       if (((inst >> 16) & 0x1f) != 1)
+               return 0;
+       /* check major opcode */
+       switch (inst >> 26) {
+       case 37:        /* stwu */
+       case 39:        /* stbu */
+       case 45:        /* sthu */
+       case 53:        /* stfsu */
+       case 55:        /* stfdu */
+               return 1;
+       case 62:        /* std or stdu */
+               return (inst & 3) == 1;
+       case 31:
+               /* check minor opcode */
+               switch ((inst >> 1) & 0x3ff) {
+               case 181:       /* stdux */
+               case 183:       /* stwux */
+               case 247:       /* stbux */
+               case 439:       /* sthux */
+               case 695:       /* stfsux */
+               case 759:       /* stfdux */
+                       return 1;
+               }
+       }
+       return 0;
+}
+
+static void do_dabr(struct pt_regs *regs, unsigned long error_code)
+{
+       siginfo_t info;
+
+       if (notify_die(DIE_DABR_MATCH, "dabr_match", regs, error_code,
+                       11, SIGSEGV) == NOTIFY_STOP)
+               return;
+
+       if (debugger_dabr_match(regs))
+               return;
+
+       /* Clear the DABR */
+       set_dabr(0);
+
+       /* Deliver the signal to userspace */
+       info.si_signo = SIGTRAP;
+       info.si_errno = 0;
+       info.si_code = TRAP_HWBKPT;
+       info.si_addr = (void __user *)regs->nip;
+       force_sig_info(SIGTRAP, &info, current);
+}
+
+/*
+ * For 600- and 800-family processors, the error_code parameter is DSISR
+ * for a data fault, SRR1 for an instruction fault. For 400-family processors
+ * the error_code parameter is ESR for a data fault, 0 for an instruction
+ * fault.
+ * For 64-bit processors, the error_code parameter is
+ *  - DSISR for a non-SLB data access fault,
+ *  - SRR1 & 0x08000000 for a non-SLB instruction access fault
+ *  - 0 any SLB fault.
+ *
+ * The return value is 0 if the fault was handled, or the signal
+ * number if this is a kernel fault that can't be handled here.
+ */
+int __kprobes do_page_fault(struct pt_regs *regs, unsigned long address,
+                           unsigned long error_code)
+{
+       struct vm_area_struct * vma;
+       struct mm_struct *mm = current->mm;
+       siginfo_t info;
+       int code = SEGV_MAPERR;
+       int is_write = 0;
+       int trap = TRAP(regs);
+       int is_exec = trap == 0x400;
+
+#if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
+       /*
+        * Fortunately the bit assignments in SRR1 for an instruction
+        * fault and DSISR for a data fault are mostly the same for the
+        * bits we are interested in.  But there are some bits which
+        * indicate errors in DSISR but can validly be set in SRR1.
+        */
+       if (trap == 0x400)
+               error_code &= 0x48200000;
+       else
+               is_write = error_code & DSISR_ISSTORE;
+#else
+       is_write = error_code & ESR_DST;
+#endif /* CONFIG_4xx || CONFIG_BOOKE */
+
+       if (notify_die(DIE_PAGE_FAULT, "page_fault", regs, error_code,
+                               11, SIGSEGV) == NOTIFY_STOP)
+               return 0;
+
+       if (trap == 0x300) {
+               if (debugger_fault_handler(regs))
+                       return 0;
+       }
+
+       /* On a kernel SLB miss we can only check for a valid exception entry */
+       if (!user_mode(regs) && (address >= TASK_SIZE))
+               return SIGSEGV;
+
+#if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
+       if (error_code & DSISR_DABRMATCH) {
+               /* DABR match */
+               do_dabr(regs, error_code);
+               return 0;
+       }
+#endif /* !(CONFIG_4xx || CONFIG_BOOKE)*/
+
+       if (in_atomic() || mm == NULL) {
+               if (!user_mode(regs))
+                       return SIGSEGV;
+               /* in_atomic() in user mode is really bad,
+                  as is current->mm == NULL. */
+               printk(KERN_EMERG "Page fault in user mode with"
+                      "in_atomic() = %d mm = %p\n", in_atomic(), mm);
+               printk(KERN_EMERG "NIP = %lx  MSR = %lx\n",
+                      regs->nip, regs->msr);
+               die("Weird page fault", regs, SIGSEGV);
+       }
+
+       /* When running in the kernel we expect faults to occur only to
+        * addresses in user space.  All other faults represent errors in the
+        * kernel and should generate an OOPS.  Unfortunatly, in the case of an
+        * erroneous fault occuring in a code path which already holds mmap_sem
+        * we will deadlock attempting to validate the fault against the
+        * address space.  Luckily the kernel only validly references user
+        * space from well defined areas of code, which are listed in the
+        * exceptions table.
+        *
+        * As the vast majority of faults will be valid we will only perform
+        * the source reference check when there is a possibilty of a deadlock.
+        * Attempt to lock the address space, if we cannot we then validate the
+        * source.  If this is invalid we can skip the address space check,
+        * thus avoiding the deadlock.
+        */
+       if (!down_read_trylock(&mm->mmap_sem)) {
+               if (!user_mode(regs) && !search_exception_tables(regs->nip))
+                       goto bad_area_nosemaphore;
+
+               down_read(&mm->mmap_sem);
+       }
+
+       vma = find_vma(mm, address);
+       if (!vma)
+               goto bad_area;
+       if (vma->vm_start <= address)
+               goto good_area;
+       if (!(vma->vm_flags & VM_GROWSDOWN))
+               goto bad_area;
+
+       /*
+        * N.B. The POWER/Open ABI allows programs to access up to
+        * 288 bytes below the stack pointer.
+        * The kernel signal delivery code writes up to about 1.5kB
+        * below the stack pointer (r1) before decrementing it.
+        * The exec code can write slightly over 640kB to the stack
+        * before setting the user r1.  Thus we allow the stack to
+        * expand to 1MB without further checks.
+        */
+       if (address + 0x100000 < vma->vm_end) {
+               /* get user regs even if this fault is in kernel mode */
+               struct pt_regs *uregs = current->thread.regs;
+               if (uregs == NULL)
+                       goto bad_area;
+
+               /*
+                * A user-mode access to an address a long way below
+                * the stack pointer is only valid if the instruction
+                * is one which would update the stack pointer to the
+                * address accessed if the instruction completed,
+                * i.e. either stwu rs,n(r1) or stwux rs,r1,rb
+                * (or the byte, halfword, float or double forms).
+                *
+                * If we don't check this then any write to the area
+                * between the last mapped region and the stack will
+                * expand the stack rather than segfaulting.
+                */
+               if (address + 2048 < uregs->gpr[1]
+                   && (!user_mode(regs) || !store_updates_sp(regs)))
+                       goto bad_area;
+       }
+       if (expand_stack(vma, address))
+               goto bad_area;
+
+good_area:
+       code = SEGV_ACCERR;
+#if defined(CONFIG_6xx)
+       if (error_code & 0x95700000)
+               /* an error such as lwarx to I/O controller space,
+                  address matching DABR, eciwx, etc. */
+               goto bad_area;
+#endif /* CONFIG_6xx */
+#if defined(CONFIG_8xx)
+        /* The MPC8xx seems to always set 0x80000000, which is
+         * "undefined".  Of those that can be set, this is the only
+         * one which seems bad.
+         */
+       if (error_code & 0x10000000)
+                /* Guarded storage error. */
+               goto bad_area;
+#endif /* CONFIG_8xx */
+
+       if (is_exec) {
+#ifdef CONFIG_PPC64
+               /* protection fault */
+               if (error_code & DSISR_PROTFAULT)
+                       goto bad_area;
+               if (!(vma->vm_flags & VM_EXEC))
+                       goto bad_area;
+#endif
+#if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
+               pte_t *ptep;
+
+               /* Since 4xx/Book-E supports per-page execute permission,
+                * we lazily flush dcache to icache. */
+               ptep = NULL;
+               if (get_pteptr(mm, address, &ptep) && pte_present(*ptep)) {
+                       struct page *page = pte_page(*ptep);
+
+                       if (! test_bit(PG_arch_1, &page->flags)) {
+                               flush_dcache_icache_page(page);
+                               set_bit(PG_arch_1, &page->flags);
+                       }
+                       pte_update(ptep, 0, _PAGE_HWEXEC);
+                       _tlbie(address);
+                       pte_unmap(ptep);
+                       up_read(&mm->mmap_sem);
+                       return 0;
+               }
+               if (ptep != NULL)
+                       pte_unmap(ptep);
+#endif
+       /* a write */
+       } else if (is_write) {
+               if (!(vma->vm_flags & VM_WRITE))
+                       goto bad_area;
+       /* a read */
+       } else {
+               /* protection fault */
+               if (error_code & 0x08000000)
+                       goto bad_area;
+               if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
+                       goto bad_area;
+       }
+
+       /*
+        * If for any reason at all we couldn't handle the fault,
+        * make sure we exit gracefully rather than endlessly redo
+        * the fault.
+        */
+ survive:
+       switch (handle_mm_fault(mm, vma, address, is_write)) {
+
+       case VM_FAULT_MINOR:
+               current->min_flt++;
+               break;
+       case VM_FAULT_MAJOR:
+               current->maj_flt++;
+               break;
+       case VM_FAULT_SIGBUS:
+               goto do_sigbus;
+       case VM_FAULT_OOM:
+               goto out_of_memory;
+       default:
+               BUG();
+       }
+
+       up_read(&mm->mmap_sem);
+       return 0;
+
+bad_area:
+       up_read(&mm->mmap_sem);
+
+bad_area_nosemaphore:
+       /* User mode accesses cause a SIGSEGV */
+       if (user_mode(regs)) {
+               _exception(SIGSEGV, regs, code, address);
+               return 0;
+       }
+
+       if (is_exec && (error_code & DSISR_PROTFAULT)
+           && printk_ratelimit())
+               printk(KERN_CRIT "kernel tried to execute NX-protected"
+                      " page (%lx) - exploit attempt? (uid: %d)\n",
+                      address, current->uid);
+
+       return SIGSEGV;
+
+/*
+ * We ran out of memory, or some other thing happened to us that made
+ * us unable to handle the page fault gracefully.
+ */
+out_of_memory:
+       up_read(&mm->mmap_sem);
+       if (current->pid == 1) {
+               yield();
+               down_read(&mm->mmap_sem);
+               goto survive;
+       }
+       printk("VM: killing process %s\n", current->comm);
+       if (user_mode(regs))
+               do_exit(SIGKILL);
+       return SIGKILL;
+
+do_sigbus:
+       up_read(&mm->mmap_sem);
+       if (user_mode(regs)) {
+               info.si_signo = SIGBUS;
+               info.si_errno = 0;
+               info.si_code = BUS_ADRERR;
+               info.si_addr = (void __user *)address;
+               force_sig_info(SIGBUS, &info, current);
+               return 0;
+       }
+       return SIGBUS;
+}
+
+/*
+ * bad_page_fault is called when we have a bad access from the kernel.
+ * It is called from the DSI and ISI handlers in head.S and from some
+ * of the procedures in traps.c.
+ */
+void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
+{
+       const struct exception_table_entry *entry;
+
+       /* Are we prepared to handle this fault?  */
+       if ((entry = search_exception_tables(regs->nip)) != NULL) {
+               regs->nip = entry->fixup;
+               return;
+       }
+
+       /* kernel has accessed a bad area */
+       die("Kernel access of bad area", regs, sig);
+}
diff --git a/arch/powerpc/mm/fsl_booke_mmu.c b/arch/powerpc/mm/fsl_booke_mmu.c
new file mode 100644 (file)
index 0000000..af9ca0e
--- /dev/null
@@ -0,0 +1,237 @@
+/*
+ * Modifications by Kumar Gala (kumar.gala@freescale.com) to support
+ * E500 Book E processors.
+ *
+ * Copyright 2004 Freescale Semiconductor, Inc
+ *
+ * This file contains the routines for initializing the MMU
+ * on the 4xx series of chips.
+ *  -- paulus
+ *
+ *  Derived from arch/ppc/mm/init.c:
+ *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
+ *
+ *  Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
+ *  and Cort Dougan (PReP) (cort@cs.nmt.edu)
+ *    Copyright (C) 1996 Paul Mackerras
+ *  Amiga/APUS changes by Jesper Skov (jskov@cygnus.co.uk).
+ *
+ *  Derived from "arch/i386/mm/init.c"
+ *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
+ *
+ *  This program is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU General Public License
+ *  as published by the Free Software Foundation; either version
+ *  2 of the License, or (at your option) any later version.
+ *
+ */
+
+#include <linux/config.h>
+#include <linux/signal.h>
+#include <linux/sched.h>
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/string.h>
+#include <linux/types.h>
+#include <linux/ptrace.h>
+#include <linux/mman.h>
+#include <linux/mm.h>
+#include <linux/swap.h>
+#include <linux/stddef.h>
+#include <linux/vmalloc.h>
+#include <linux/init.h>
+#include <linux/delay.h>
+#include <linux/highmem.h>
+
+#include <asm/pgalloc.h>
+#include <asm/prom.h>
+#include <asm/io.h>
+#include <asm/mmu_context.h>
+#include <asm/pgtable.h>
+#include <asm/mmu.h>
+#include <asm/uaccess.h>
+#include <asm/smp.h>
+#include <asm/bootx.h>
+#include <asm/machdep.h>
+#include <asm/setup.h>
+
+extern void loadcam_entry(unsigned int index);
+unsigned int tlbcam_index;
+unsigned int num_tlbcam_entries;
+static unsigned long __cam0, __cam1, __cam2;
+extern unsigned long total_lowmem;
+extern unsigned long __max_low_memory;
+#define MAX_LOW_MEM    CONFIG_LOWMEM_SIZE
+
+#define NUM_TLBCAMS    (16)
+
+struct tlbcam {
+       u32     MAS0;
+       u32     MAS1;
+       u32     MAS2;
+       u32     MAS3;
+       u32     MAS7;
+} TLBCAM[NUM_TLBCAMS];
+
+struct tlbcamrange {
+       unsigned long start;
+       unsigned long limit;
+       phys_addr_t phys;
+} tlbcam_addrs[NUM_TLBCAMS];
+
+extern unsigned int tlbcam_index;
+
+/*
+ * Return PA for this VA if it is mapped by a CAM, or 0
+ */
+unsigned long v_mapped_by_tlbcam(unsigned long va)
+{
+       int b;
+       for (b = 0; b < tlbcam_index; ++b)
+               if (va >= tlbcam_addrs[b].start && va < tlbcam_addrs[b].limit)
+                       return tlbcam_addrs[b].phys + (va - tlbcam_addrs[b].start);
+       return 0;
+}
+
+/*
+ * Return VA for a given PA or 0 if not mapped
+ */
+unsigned long p_mapped_by_tlbcam(unsigned long pa)
+{
+       int b;
+       for (b = 0; b < tlbcam_index; ++b)
+               if (pa >= tlbcam_addrs[b].phys
+                   && pa < (tlbcam_addrs[b].limit-tlbcam_addrs[b].start)
+                             +tlbcam_addrs[b].phys)
+                       return tlbcam_addrs[b].start+(pa-tlbcam_addrs[b].phys);
+       return 0;
+}
+
+/*
+ * Set up one of the I/D BAT (block address translation) register pairs.
+ * The parameters are not checked; in particular size must be a power
+ * of 4 between 4k and 256M.
+ */
+void settlbcam(int index, unsigned long virt, phys_addr_t phys,
+               unsigned int size, int flags, unsigned int pid)
+{
+       unsigned int tsize, lz;
+
+       asm ("cntlzw %0,%1" : "=r" (lz) : "r" (size));
+       tsize = (21 - lz) / 2;
+
+#ifdef CONFIG_SMP
+       if ((flags & _PAGE_NO_CACHE) == 0)
+               flags |= _PAGE_COHERENT;
+#endif
+
+       TLBCAM[index].MAS0 = MAS0_TLBSEL(1) | MAS0_ESEL(index) | MAS0_NV(index+1);
+       TLBCAM[index].MAS1 = MAS1_VALID | MAS1_IPROT | MAS1_TSIZE(tsize) | MAS1_TID(pid);
+       TLBCAM[index].MAS2 = virt & PAGE_MASK;
+
+       TLBCAM[index].MAS2 |= (flags & _PAGE_WRITETHRU) ? MAS2_W : 0;
+       TLBCAM[index].MAS2 |= (flags & _PAGE_NO_CACHE) ? MAS2_I : 0;
+       TLBCAM[index].MAS2 |= (flags & _PAGE_COHERENT) ? MAS2_M : 0;
+       TLBCAM[index].MAS2 |= (flags & _PAGE_GUARDED) ? MAS2_G : 0;
+       TLBCAM[index].MAS2 |= (flags & _PAGE_ENDIAN) ? MAS2_E : 0;
+
+       TLBCAM[index].MAS3 = (phys & PAGE_MASK) | MAS3_SX | MAS3_SR;
+       TLBCAM[index].MAS3 |= ((flags & _PAGE_RW) ? MAS3_SW : 0);
+
+#ifndef CONFIG_KGDB /* want user access for breakpoints */
+       if (flags & _PAGE_USER) {
+          TLBCAM[index].MAS3 |= MAS3_UX | MAS3_UR;
+          TLBCAM[index].MAS3 |= ((flags & _PAGE_RW) ? MAS3_UW : 0);
+       }
+#else
+       TLBCAM[index].MAS3 |= MAS3_UX | MAS3_UR;
+       TLBCAM[index].MAS3 |= ((flags & _PAGE_RW) ? MAS3_UW : 0);
+#endif
+
+       tlbcam_addrs[index].start = virt;
+       tlbcam_addrs[index].limit = virt + size - 1;
+       tlbcam_addrs[index].phys = phys;
+
+       loadcam_entry(index);
+}
+
+void invalidate_tlbcam_entry(int index)
+{
+       TLBCAM[index].MAS0 = MAS0_TLBSEL(1) | MAS0_ESEL(index);
+       TLBCAM[index].MAS1 = ~MAS1_VALID;
+
+       loadcam_entry(index);
+}
+
+void __init cam_mapin_ram(unsigned long cam0, unsigned long cam1,
+               unsigned long cam2)
+{
+       settlbcam(0, KERNELBASE, PPC_MEMSTART, cam0, _PAGE_KERNEL, 0);
+       tlbcam_index++;
+       if (cam1) {
+               tlbcam_index++;
+               settlbcam(1, KERNELBASE+cam0, PPC_MEMSTART+cam0, cam1, _PAGE_KERNEL, 0);
+       }
+       if (cam2) {
+               tlbcam_index++;
+               settlbcam(2, KERNELBASE+cam0+cam1, PPC_MEMSTART+cam0+cam1, cam2, _PAGE_KERNEL, 0);
+       }
+}
+
+/*
+ * MMU_init_hw does the chip-specific initialization of the MMU hardware.
+ */
+void __init MMU_init_hw(void)
+{
+       flush_instruction_cache();
+}
+
+unsigned long __init mmu_mapin_ram(void)
+{
+       cam_mapin_ram(__cam0, __cam1, __cam2);
+
+       return __cam0 + __cam1 + __cam2;
+}
+
+
+void __init
+adjust_total_lowmem(void)
+{
+       unsigned long max_low_mem = MAX_LOW_MEM;
+       unsigned long cam_max = 0x10000000;
+       unsigned long ram;
+
+       /* adjust CAM size to max_low_mem */
+       if (max_low_mem < cam_max)
+               cam_max = max_low_mem;
+
+       /* adjust lowmem size to max_low_mem */
+       if (max_low_mem < total_lowmem)
+               ram = max_low_mem;
+       else
+               ram = total_lowmem;
+
+       /* Calculate CAM values */
+       __cam0 = 1UL << 2 * (__ilog2(ram) / 2);
+       if (__cam0 > cam_max)
+               __cam0 = cam_max;
+       ram -= __cam0;
+       if (ram) {
+               __cam1 = 1UL << 2 * (__ilog2(ram) / 2);
+               if (__cam1 > cam_max)
+                       __cam1 = cam_max;
+               ram -= __cam1;
+       }
+       if (ram) {
+               __cam2 = 1UL << 2 * (__ilog2(ram) / 2);
+               if (__cam2 > cam_max)
+                       __cam2 = cam_max;
+               ram -= __cam2;
+       }
+
+       printk(KERN_INFO "Memory CAM mapping: CAM0=%ldMb, CAM1=%ldMb,"
+                       " CAM2=%ldMb residual: %ldMb\n",
+                       __cam0 >> 20, __cam1 >> 20, __cam2 >> 20,
+                       (total_lowmem - __cam0 - __cam1 - __cam2) >> 20);
+       __max_low_memory = max_low_mem = __cam0 + __cam1 + __cam2;
+}
diff --git a/arch/powerpc/mm/hash_32.S b/arch/powerpc/mm/hash_32.S
new file mode 100644 (file)
index 0000000..57278a8
--- /dev/null
@@ -0,0 +1,618 @@
+/*
+ *  arch/ppc/kernel/hashtable.S
+ *
+ *  $Id: hashtable.S,v 1.6 1999/10/08 01:56:15 paulus Exp $
+ *
+ *  PowerPC version
+ *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
+ *  Rewritten by Cort Dougan (cort@cs.nmt.edu) for PReP
+ *    Copyright (C) 1996 Cort Dougan <cort@cs.nmt.edu>
+ *  Adapted for Power Macintosh by Paul Mackerras.
+ *  Low-level exception handlers and MMU support
+ *  rewritten by Paul Mackerras.
+ *    Copyright (C) 1996 Paul Mackerras.
+ *
+ *  This file contains low-level assembler routines for managing
+ *  the PowerPC MMU hash table.  (PPC 8xx processors don't use a
+ *  hash table, so this file is not used on them.)
+ *
+ *  This program is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU General Public License
+ *  as published by the Free Software Foundation; either version
+ *  2 of the License, or (at your option) any later version.
+ *
+ */
+
+#include <linux/config.h>
+#include <asm/processor.h>
+#include <asm/page.h>
+#include <asm/pgtable.h>
+#include <asm/cputable.h>
+#include <asm/ppc_asm.h>
+#include <asm/thread_info.h>
+#include <asm/asm-offsets.h>
+
+#ifdef CONFIG_SMP
+       .comm   mmu_hash_lock,4
+#endif /* CONFIG_SMP */
+
+/*
+ * Sync CPUs with hash_page taking & releasing the hash
+ * table lock
+ */
+#ifdef CONFIG_SMP
+       .text
+_GLOBAL(hash_page_sync)
+       lis     r8,mmu_hash_lock@h
+       ori     r8,r8,mmu_hash_lock@l
+       lis     r0,0x0fff
+       b       10f
+11:    lwz     r6,0(r8)
+       cmpwi   0,r6,0
+       bne     11b
+10:    lwarx   r6,0,r8
+       cmpwi   0,r6,0
+       bne-    11b
+       stwcx.  r0,0,r8
+       bne-    10b
+       isync
+       eieio
+       li      r0,0
+       stw     r0,0(r8)
+       blr     
+#endif
+
+/*
+ * Load a PTE into the hash table, if possible.
+ * The address is in r4, and r3 contains an access flag:
+ * _PAGE_RW (0x400) if a write.
+ * r9 contains the SRR1 value, from which we use the MSR_PR bit.
+ * SPRG3 contains the physical address of the current task's thread.
+ *
+ * Returns to the caller if the access is illegal or there is no
+ * mapping for the address.  Otherwise it places an appropriate PTE
+ * in the hash table and returns from the exception.
+ * Uses r0, r3 - r8, ctr, lr.
+ */
+       .text
+_GLOBAL(hash_page)
+#ifdef CONFIG_PPC64BRIDGE
+       mfmsr   r0
+       clrldi  r0,r0,1         /* make sure it's in 32-bit mode */
+       MTMSRD(r0)
+       isync
+#endif
+       tophys(r7,0)                    /* gets -KERNELBASE into r7 */
+#ifdef CONFIG_SMP
+       addis   r8,r7,mmu_hash_lock@h
+       ori     r8,r8,mmu_hash_lock@l
+       lis     r0,0x0fff
+       b       10f
+11:    lwz     r6,0(r8)
+       cmpwi   0,r6,0
+       bne     11b
+10:    lwarx   r6,0,r8
+       cmpwi   0,r6,0
+       bne-    11b
+       stwcx.  r0,0,r8
+       bne-    10b
+       isync
+#endif
+       /* Get PTE (linux-style) and check access */
+       lis     r0,KERNELBASE@h         /* check if kernel address */
+       cmplw   0,r4,r0
+       mfspr   r8,SPRN_SPRG3           /* current task's THREAD (phys) */
+       ori     r3,r3,_PAGE_USER|_PAGE_PRESENT /* test low addresses as user */
+       lwz     r5,PGDIR(r8)            /* virt page-table root */
+       blt+    112f                    /* assume user more likely */
+       lis     r5,swapper_pg_dir@ha    /* if kernel address, use */
+       addi    r5,r5,swapper_pg_dir@l  /* kernel page table */
+       rlwimi  r3,r9,32-12,29,29       /* MSR_PR -> _PAGE_USER */
+112:   add     r5,r5,r7                /* convert to phys addr */
+       rlwimi  r5,r4,12,20,29          /* insert top 10 bits of address */
+       lwz     r8,0(r5)                /* get pmd entry */
+       rlwinm. r8,r8,0,0,19            /* extract address of pte page */
+#ifdef CONFIG_SMP
+       beq-    hash_page_out           /* return if no mapping */
+#else
+       /* XXX it seems like the 601 will give a machine fault on the
+          rfi if its alignment is wrong (bottom 4 bits of address are
+          8 or 0xc) and we have had a not-taken conditional branch
+          to the address following the rfi. */
+       beqlr-
+#endif
+       rlwimi  r8,r4,22,20,29          /* insert next 10 bits of address */
+       rlwinm  r0,r3,32-3,24,24        /* _PAGE_RW access -> _PAGE_DIRTY */
+       ori     r0,r0,_PAGE_ACCESSED|_PAGE_HASHPTE
+
+       /*
+        * Update the linux PTE atomically.  We do the lwarx up-front
+        * because almost always, there won't be a permission violation
+        * and there won't already be an HPTE, and thus we will have
+        * to update the PTE to set _PAGE_HASHPTE.  -- paulus.
+        */
+retry:
+       lwarx   r6,0,r8                 /* get linux-style pte */
+       andc.   r5,r3,r6                /* check access & ~permission */
+#ifdef CONFIG_SMP
+       bne-    hash_page_out           /* return if access not permitted */
+#else
+       bnelr-
+#endif
+       or      r5,r0,r6                /* set accessed/dirty bits */
+       stwcx.  r5,0,r8                 /* attempt to update PTE */
+       bne-    retry                   /* retry if someone got there first */
+
+       mfsrin  r3,r4                   /* get segment reg for segment */
+       mfctr   r0
+       stw     r0,_CTR(r11)
+       bl      create_hpte             /* add the hash table entry */
+
+#ifdef CONFIG_SMP
+       eieio
+       addis   r8,r7,mmu_hash_lock@ha
+       li      r0,0
+       stw     r0,mmu_hash_lock@l(r8)
+#endif
+
+       /* Return from the exception */
+       lwz     r5,_CTR(r11)
+       mtctr   r5
+       lwz     r0,GPR0(r11)
+       lwz     r7,GPR7(r11)
+       lwz     r8,GPR8(r11)
+       b       fast_exception_return
+
+#ifdef CONFIG_SMP
+hash_page_out:
+       eieio
+       addis   r8,r7,mmu_hash_lock@ha
+       li      r0,0
+       stw     r0,mmu_hash_lock@l(r8)
+       blr
+#endif /* CONFIG_SMP */
+
+/*
+ * Add an entry for a particular page to the hash table.
+ *
+ * add_hash_page(unsigned context, unsigned long va, unsigned long pmdval)
+ *
+ * We assume any necessary modifications to the pte (e.g. setting
+ * the accessed bit) have already been done and that there is actually
+ * a hash table in use (i.e. we're not on a 603).
+ */
+_GLOBAL(add_hash_page)
+       mflr    r0
+       stw     r0,4(r1)
+
+       /* Convert context and va to VSID */
+       mulli   r3,r3,897*16            /* multiply context by context skew */
+       rlwinm  r0,r4,4,28,31           /* get ESID (top 4 bits of va) */
+       mulli   r0,r0,0x111             /* multiply by ESID skew */
+       add     r3,r3,r0                /* note create_hpte trims to 24 bits */
+
+#ifdef CONFIG_SMP
+       rlwinm  r8,r1,0,0,18            /* use cpu number to make tag */
+       lwz     r8,TI_CPU(r8)           /* to go in mmu_hash_lock */
+       oris    r8,r8,12
+#endif /* CONFIG_SMP */
+
+       /*
+        * We disable interrupts here, even on UP, because we don't
+        * want to race with hash_page, and because we want the
+        * _PAGE_HASHPTE bit to be a reliable indication of whether
+        * the HPTE exists (or at least whether one did once).
+        * We also turn off the MMU for data accesses so that we
+        * we can't take a hash table miss (assuming the code is
+        * covered by a BAT).  -- paulus
+        */
+       mfmsr   r10
+       SYNC
+       rlwinm  r0,r10,0,17,15          /* clear bit 16 (MSR_EE) */
+       rlwinm  r0,r0,0,28,26           /* clear MSR_DR */
+       mtmsr   r0
+       SYNC_601
+       isync
+
+       tophys(r7,0)
+
+#ifdef CONFIG_SMP
+       addis   r9,r7,mmu_hash_lock@ha
+       addi    r9,r9,mmu_hash_lock@l
+10:    lwarx   r0,0,r9                 /* take the mmu_hash_lock */
+       cmpi    0,r0,0
+       bne-    11f
+       stwcx.  r8,0,r9
+       beq+    12f
+11:    lwz     r0,0(r9)
+       cmpi    0,r0,0
+       beq     10b
+       b       11b
+12:    isync
+#endif
+
+       /*
+        * Fetch the linux pte and test and set _PAGE_HASHPTE atomically.
+        * If _PAGE_HASHPTE was already set, we don't replace the existing
+        * HPTE, so we just unlock and return.
+        */
+       mr      r8,r5
+       rlwimi  r8,r4,22,20,29
+1:     lwarx   r6,0,r8
+       andi.   r0,r6,_PAGE_HASHPTE
+       bne     9f                      /* if HASHPTE already set, done */
+       ori     r5,r6,_PAGE_HASHPTE
+       stwcx.  r5,0,r8
+       bne-    1b
+
+       bl      create_hpte
+
+9:
+#ifdef CONFIG_SMP
+       eieio
+       li      r0,0
+       stw     r0,0(r9)                /* clear mmu_hash_lock */
+#endif
+
+       /* reenable interrupts and DR */
+       mtmsr   r10
+       SYNC_601
+       isync
+
+       lwz     r0,4(r1)
+       mtlr    r0
+       blr
+
+/*
+ * This routine adds a hardware PTE to the hash table.
+ * It is designed to be called with the MMU either on or off.
+ * r3 contains the VSID, r4 contains the virtual address,
+ * r5 contains the linux PTE, r6 contains the old value of the
+ * linux PTE (before setting _PAGE_HASHPTE) and r7 contains the
+ * offset to be added to addresses (0 if the MMU is on,
+ * -KERNELBASE if it is off).
+ * On SMP, the caller should have the mmu_hash_lock held.
+ * We assume that the caller has (or will) set the _PAGE_HASHPTE
+ * bit in the linux PTE in memory.  The value passed in r6 should
+ * be the old linux PTE value; if it doesn't have _PAGE_HASHPTE set
+ * this routine will skip the search for an existing HPTE.
+ * This procedure modifies r0, r3 - r6, r8, cr0.
+ *  -- paulus.
+ *
+ * For speed, 4 of the instructions get patched once the size and
+ * physical address of the hash table are known.  These definitions
+ * of Hash_base and Hash_bits below are just an example.
+ */
+Hash_base = 0xc0180000
+Hash_bits = 12                         /* e.g. 256kB hash table */
+Hash_msk = (((1 << Hash_bits) - 1) * 64)
+
+#ifndef CONFIG_PPC64BRIDGE
+/* defines for the PTE format for 32-bit PPCs */
+#define PTE_SIZE       8
+#define PTEG_SIZE      64
+#define LG_PTEG_SIZE   6
+#define LDPTEu         lwzu
+#define STPTE          stw
+#define CMPPTE         cmpw
+#define PTE_H          0x40
+#define PTE_V          0x80000000
+#define TST_V(r)       rlwinm. r,r,0,0,0
+#define SET_V(r)       oris r,r,PTE_V@h
+#define CLR_V(r,t)     rlwinm r,r,0,1,31
+
+#else
+/* defines for the PTE format for 64-bit PPCs */
+#define PTE_SIZE       16
+#define PTEG_SIZE      128
+#define LG_PTEG_SIZE   7
+#define LDPTEu         ldu
+#define STPTE          std
+#define CMPPTE         cmpd
+#define PTE_H          2
+#define PTE_V          1
+#define TST_V(r)       andi. r,r,PTE_V
+#define SET_V(r)       ori r,r,PTE_V
+#define CLR_V(r,t)     li t,PTE_V; andc r,r,t
+#endif /* CONFIG_PPC64BRIDGE */
+
+#define HASH_LEFT      31-(LG_PTEG_SIZE+Hash_bits-1)
+#define HASH_RIGHT     31-LG_PTEG_SIZE
+
+_GLOBAL(create_hpte)
+       /* Convert linux-style PTE (r5) to low word of PPC-style PTE (r8) */
+       rlwinm  r8,r5,32-10,31,31       /* _PAGE_RW -> PP lsb */
+       rlwinm  r0,r5,32-7,31,31        /* _PAGE_DIRTY -> PP lsb */
+       and     r8,r8,r0                /* writable if _RW & _DIRTY */
+       rlwimi  r5,r5,32-1,30,30        /* _PAGE_USER -> PP msb */
+       rlwimi  r5,r5,32-2,31,31        /* _PAGE_USER -> PP lsb */
+       ori     r8,r8,0xe14             /* clear out reserved bits and M */
+       andc    r8,r5,r8                /* PP = user? (rw&dirty? 2: 3): 0 */
+BEGIN_FTR_SECTION
+       ori     r8,r8,_PAGE_COHERENT    /* set M (coherence required) */
+END_FTR_SECTION_IFSET(CPU_FTR_NEED_COHERENT)
+
+       /* Construct the high word of the PPC-style PTE (r5) */
+#ifndef CONFIG_PPC64BRIDGE
+       rlwinm  r5,r3,7,1,24            /* put VSID in 0x7fffff80 bits */
+       rlwimi  r5,r4,10,26,31          /* put in API (abbrev page index) */
+#else /* CONFIG_PPC64BRIDGE */
+       clrlwi  r3,r3,8                 /* reduce vsid to 24 bits */
+       sldi    r5,r3,12                /* shift vsid into position */
+       rlwimi  r5,r4,16,20,24          /* put in API (abbrev page index) */
+#endif /* CONFIG_PPC64BRIDGE */
+       SET_V(r5)                       /* set V (valid) bit */
+
+       /* Get the address of the primary PTE group in the hash table (r3) */
+_GLOBAL(hash_page_patch_A)
+       addis   r0,r7,Hash_base@h       /* base address of hash table */
+       rlwimi  r0,r3,LG_PTEG_SIZE,HASH_LEFT,HASH_RIGHT    /* VSID -> hash */
+       rlwinm  r3,r4,20+LG_PTEG_SIZE,HASH_LEFT,HASH_RIGHT /* PI -> hash */
+       xor     r3,r3,r0                /* make primary hash */
+       li      r0,8                    /* PTEs/group */
+
+       /*
+        * Test the _PAGE_HASHPTE bit in the old linux PTE, and skip the search
+        * if it is clear, meaning that the HPTE isn't there already...
+        */
+       andi.   r6,r6,_PAGE_HASHPTE
+       beq+    10f                     /* no PTE: go look for an empty slot */
+       tlbie   r4
+
+       addis   r4,r7,htab_hash_searches@ha
+       lwz     r6,htab_hash_searches@l(r4)
+       addi    r6,r6,1                 /* count how many searches we do */
+       stw     r6,htab_hash_searches@l(r4)
+
+       /* Search the primary PTEG for a PTE whose 1st (d)word matches r5 */
+       mtctr   r0
+       addi    r4,r3,-PTE_SIZE
+1:     LDPTEu  r6,PTE_SIZE(r4)         /* get next PTE */
+       CMPPTE  0,r6,r5
+       bdnzf   2,1b                    /* loop while ctr != 0 && !cr0.eq */
+       beq+    found_slot
+
+       /* Search the secondary PTEG for a matching PTE */
+       ori     r5,r5,PTE_H             /* set H (secondary hash) bit */
+_GLOBAL(hash_page_patch_B)
+       xoris   r4,r3,Hash_msk>>16      /* compute secondary hash */
+       xori    r4,r4,(-PTEG_SIZE & 0xffff)
+       addi    r4,r4,-PTE_SIZE
+       mtctr   r0
+2:     LDPTEu  r6,PTE_SIZE(r4)
+       CMPPTE  0,r6,r5
+       bdnzf   2,2b
+       beq+    found_slot
+       xori    r5,r5,PTE_H             /* clear H bit again */
+
+       /* Search the primary PTEG for an empty slot */
+10:    mtctr   r0
+       addi    r4,r3,-PTE_SIZE         /* search primary PTEG */
+1:     LDPTEu  r6,PTE_SIZE(r4)         /* get next PTE */
+       TST_V(r6)                       /* test valid bit */
+       bdnzf   2,1b                    /* loop while ctr != 0 && !cr0.eq */
+       beq+    found_empty
+
+       /* update counter of times that the primary PTEG is full */
+       addis   r4,r7,primary_pteg_full@ha
+       lwz     r6,primary_pteg_full@l(r4)
+       addi    r6,r6,1
+       stw     r6,primary_pteg_full@l(r4)
+
+       /* Search the secondary PTEG for an empty slot */
+       ori     r5,r5,PTE_H             /* set H (secondary hash) bit */
+_GLOBAL(hash_page_patch_C)
+       xoris   r4,r3,Hash_msk>>16      /* compute secondary hash */
+       xori    r4,r4,(-PTEG_SIZE & 0xffff)
+       addi    r4,r4,-PTE_SIZE
+       mtctr   r0
+2:     LDPTEu  r6,PTE_SIZE(r4)
+       TST_V(r6)
+       bdnzf   2,2b
+       beq+    found_empty
+       xori    r5,r5,PTE_H             /* clear H bit again */
+
+       /*
+        * Choose an arbitrary slot in the primary PTEG to overwrite.
+        * Since both the primary and secondary PTEGs are full, and we
+        * have no information that the PTEs in the primary PTEG are
+        * more important or useful than those in the secondary PTEG,
+        * and we know there is a definite (although small) speed
+        * advantage to putting the PTE in the primary PTEG, we always
+        * put the PTE in the primary PTEG.
+        */
+       addis   r4,r7,next_slot@ha
+       lwz     r6,next_slot@l(r4)
+       addi    r6,r6,PTE_SIZE
+       andi.   r6,r6,7*PTE_SIZE
+       stw     r6,next_slot@l(r4)
+       add     r4,r3,r6
+
+#ifndef CONFIG_SMP
+       /* Store PTE in PTEG */
+found_empty:
+       STPTE   r5,0(r4)
+found_slot:
+       STPTE   r8,PTE_SIZE/2(r4)
+
+#else /* CONFIG_SMP */
+/*
+ * Between the tlbie above and updating the hash table entry below,
+ * another CPU could read the hash table entry and put it in its TLB.
+ * There are 3 cases:
+ * 1. using an empty slot
+ * 2. updating an earlier entry to change permissions (i.e. enable write)
+ * 3. taking over the PTE for an unrelated address
+ *
+ * In each case it doesn't really matter if the other CPUs have the old
+ * PTE in their TLB.  So we don't need to bother with another tlbie here,
+ * which is convenient as we've overwritten the register that had the
+ * address. :-)  The tlbie above is mainly to make sure that this CPU comes
+ * and gets the new PTE from the hash table.
+ *
+ * We do however have to make sure that the PTE is never in an invalid
+ * state with the V bit set.
+ */
+found_empty:
+found_slot:
+       CLR_V(r5,r0)            /* clear V (valid) bit in PTE */
+       STPTE   r5,0(r4)
+       sync
+       TLBSYNC
+       STPTE   r8,PTE_SIZE/2(r4) /* put in correct RPN, WIMG, PP bits */
+       sync
+       SET_V(r5)
+       STPTE   r5,0(r4)        /* finally set V bit in PTE */
+#endif /* CONFIG_SMP */
+
+       sync            /* make sure pte updates get to memory */
+       blr
+
+       .comm   next_slot,4
+       .comm   primary_pteg_full,4
+       .comm   htab_hash_searches,4
+
+/*
+ * Flush the entry for a particular page from the hash table.
+ *
+ * flush_hash_pages(unsigned context, unsigned long va, unsigned long pmdval,
+ *                 int count)
+ *
+ * We assume that there is a hash table in use (Hash != 0).
+ */
+_GLOBAL(flush_hash_pages)
+       tophys(r7,0)
+
+       /*
+        * We disable interrupts here, even on UP, because we want
+        * the _PAGE_HASHPTE bit to be a reliable indication of
+        * whether the HPTE exists (or at least whether one did once).
+        * We also turn off the MMU for data accesses so that we
+        * we can't take a hash table miss (assuming the code is
+        * covered by a BAT).  -- paulus
+        */
+       mfmsr   r10
+       SYNC
+       rlwinm  r0,r10,0,17,15          /* clear bit 16 (MSR_EE) */
+       rlwinm  r0,r0,0,28,26           /* clear MSR_DR */
+       mtmsr   r0
+       SYNC_601
+       isync
+
+       /* First find a PTE in the range that has _PAGE_HASHPTE set */
+       rlwimi  r5,r4,22,20,29
+1:     lwz     r0,0(r5)
+       cmpwi   cr1,r6,1
+       andi.   r0,r0,_PAGE_HASHPTE
+       bne     2f
+       ble     cr1,19f
+       addi    r4,r4,0x1000
+       addi    r5,r5,4
+       addi    r6,r6,-1
+       b       1b
+
+       /* Convert context and va to VSID */
+2:     mulli   r3,r3,897*16            /* multiply context by context skew */
+       rlwinm  r0,r4,4,28,31           /* get ESID (top 4 bits of va) */
+       mulli   r0,r0,0x111             /* multiply by ESID skew */
+       add     r3,r3,r0                /* note code below trims to 24 bits */
+
+       /* Construct the high word of the PPC-style PTE (r11) */
+#ifndef CONFIG_PPC64BRIDGE
+       rlwinm  r11,r3,7,1,24           /* put VSID in 0x7fffff80 bits */
+       rlwimi  r11,r4,10,26,31         /* put in API (abbrev page index) */
+#else /* CONFIG_PPC64BRIDGE */
+       clrlwi  r3,r3,8                 /* reduce vsid to 24 bits */
+       sldi    r11,r3,12               /* shift vsid into position */
+       rlwimi  r11,r4,16,20,24         /* put in API (abbrev page index) */
+#endif /* CONFIG_PPC64BRIDGE */
+       SET_V(r11)                      /* set V (valid) bit */
+
+#ifdef CONFIG_SMP
+       addis   r9,r7,mmu_hash_lock@ha
+       addi    r9,r9,mmu_hash_lock@l
+       rlwinm  r8,r1,0,0,18
+       add     r8,r8,r7
+       lwz     r8,TI_CPU(r8)
+       oris    r8,r8,9
+10:    lwarx   r0,0,r9
+       cmpi    0,r0,0
+       bne-    11f
+       stwcx.  r8,0,r9
+       beq+    12f
+11:    lwz     r0,0(r9)
+       cmpi    0,r0,0
+       beq     10b
+       b       11b
+12:    isync
+#endif
+
+       /*
+        * Check the _PAGE_HASHPTE bit in the linux PTE.  If it is
+        * already clear, we're done (for this pte).  If not,
+        * clear it (atomically) and proceed.  -- paulus.
+        */
+33:    lwarx   r8,0,r5                 /* fetch the pte */
+       andi.   r0,r8,_PAGE_HASHPTE
+       beq     8f                      /* done if HASHPTE is already clear */
+       rlwinm  r8,r8,0,31,29           /* clear HASHPTE bit */
+       stwcx.  r8,0,r5                 /* update the pte */
+       bne-    33b
+
+       /* Get the address of the primary PTE group in the hash table (r3) */
+_GLOBAL(flush_hash_patch_A)
+       addis   r8,r7,Hash_base@h       /* base address of hash table */
+       rlwimi  r8,r3,LG_PTEG_SIZE,HASH_LEFT,HASH_RIGHT    /* VSID -> hash */
+       rlwinm  r0,r4,20+LG_PTEG_SIZE,HASH_LEFT,HASH_RIGHT /* PI -> hash */
+       xor     r8,r0,r8                /* make primary hash */
+
+       /* Search the primary PTEG for a PTE whose 1st (d)word matches r5 */
+       li      r0,8                    /* PTEs/group */
+       mtctr   r0
+       addi    r12,r8,-PTE_SIZE
+1:     LDPTEu  r0,PTE_SIZE(r12)        /* get next PTE */
+       CMPPTE  0,r0,r11
+       bdnzf   2,1b                    /* loop while ctr != 0 && !cr0.eq */
+       beq+    3f
+
+       /* Search the secondary PTEG for a matching PTE */
+       ori     r11,r11,PTE_H           /* set H (secondary hash) bit */
+       li      r0,8                    /* PTEs/group */
+_GLOBAL(flush_hash_patch_B)
+       xoris   r12,r8,Hash_msk>>16     /* compute secondary hash */
+       xori    r12,r12,(-PTEG_SIZE & 0xffff)
+       addi    r12,r12,-PTE_SIZE
+       mtctr   r0
+2:     LDPTEu  r0,PTE_SIZE(r12)
+       CMPPTE  0,r0,r11
+       bdnzf   2,2b
+       xori    r11,r11,PTE_H           /* clear H again */
+       bne-    4f                      /* should rarely fail to find it */
+
+3:     li      r0,0
+       STPTE   r0,0(r12)               /* invalidate entry */
+4:     sync
+       tlbie   r4                      /* in hw tlb too */
+       sync
+
+8:     ble     cr1,9f                  /* if all ptes checked */
+81:    addi    r6,r6,-1
+       addi    r5,r5,4                 /* advance to next pte */
+       addi    r4,r4,0x1000
+       lwz     r0,0(r5)                /* check next pte */
+       cmpwi   cr1,r6,1
+       andi.   r0,r0,_PAGE_HASHPTE
+       bne     33b
+       bgt     cr1,81b
+
+9:
+#ifdef CONFIG_SMP
+       TLBSYNC
+       li      r0,0
+       stw     r0,0(r9)                /* clear mmu_hash_lock */
+#endif
+
+19:    mtmsr   r10
+       SYNC_601
+       isync
+       blr
diff --git a/arch/powerpc/mm/init.c b/arch/powerpc/mm/init.c
new file mode 100644 (file)
index 0000000..f4d983a
--- /dev/null
@@ -0,0 +1,581 @@
+/*
+ *  PowerPC version
+ *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
+ *
+ *  Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
+ *  and Cort Dougan (PReP) (cort@cs.nmt.edu)
+ *    Copyright (C) 1996 Paul Mackerras
+ *  Amiga/APUS changes by Jesper Skov (jskov@cygnus.co.uk).
+ *  PPC44x/36-bit changes by Matt Porter (mporter@mvista.com)
+ *
+ *  Derived from "arch/i386/mm/init.c"
+ *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
+ *
+ *  This program is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU General Public License
+ *  as published by the Free Software Foundation; either version
+ *  2 of the License, or (at your option) any later version.
+ *
+ */
+
+#include <linux/config.h>
+#include <linux/module.h>
+#include <linux/sched.h>
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/string.h>
+#include <linux/types.h>
+#include <linux/mm.h>
+#include <linux/stddef.h>
+#include <linux/init.h>
+#include <linux/bootmem.h>
+#include <linux/highmem.h>
+#include <linux/initrd.h>
+#include <linux/pagemap.h>
+
+#include <asm/pgalloc.h>
+#include <asm/prom.h>
+#include <asm/io.h>
+#include <asm/mmu_context.h>
+#include <asm/pgtable.h>
+#include <asm/mmu.h>
+#include <asm/smp.h>
+#include <asm/machdep.h>
+#include <asm/btext.h>
+#include <asm/tlb.h>
+#include <asm/bootinfo.h>
+#include <asm/prom.h>
+
+#include "mem_pieces.h"
+#include "mmu_decl.h"
+
+#if defined(CONFIG_KERNEL_START_BOOL) || defined(CONFIG_LOWMEM_SIZE_BOOL)
+/* The ammount of lowmem must be within 0xF0000000 - KERNELBASE. */
+#if (CONFIG_LOWMEM_SIZE > (0xF0000000 - KERNELBASE))
+#error "You must adjust CONFIG_LOWMEM_SIZE or CONFIG_START_KERNEL"
+#endif
+#endif
+#define MAX_LOW_MEM    CONFIG_LOWMEM_SIZE
+
+DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
+
+unsigned long total_memory;
+unsigned long total_lowmem;
+
+unsigned long ppc_memstart;
+unsigned long ppc_memoffset = PAGE_OFFSET;
+
+int mem_init_done;
+int init_bootmem_done;
+int boot_mapsize;
+#ifdef CONFIG_PPC_PMAC
+unsigned long agp_special_page;
+#endif
+
+extern char _end[];
+extern char etext[], _stext[];
+extern char __init_begin, __init_end;
+
+#ifdef CONFIG_HIGHMEM
+pte_t *kmap_pte;
+pgprot_t kmap_prot;
+
+EXPORT_SYMBOL(kmap_prot);
+EXPORT_SYMBOL(kmap_pte);
+#endif
+
+void MMU_init(void);
+void set_phys_avail(unsigned long total_ram);
+
+/* XXX should be in current.h  -- paulus */
+extern struct task_struct *current_set[NR_CPUS];
+
+char *klimit = _end;
+struct mem_pieces phys_avail;
+struct device_node *memory_node;
+
+/*
+ * this tells the system to map all of ram with the segregs
+ * (i.e. page tables) instead of the bats.
+ * -- Cort
+ */
+int __map_without_bats;
+int __map_without_ltlbs;
+
+/* max amount of RAM to use */
+unsigned long __max_memory;
+/* max amount of low RAM to map in */
+unsigned long __max_low_memory = MAX_LOW_MEM;
+
+/*
+ * Read in a property describing some pieces of memory.
+ */
+static int __init get_mem_prop(char *name, struct mem_pieces *mp)
+{
+       struct reg_property *rp;
+       int i, s;
+       unsigned int *ip;
+       int nac = prom_n_addr_cells(memory_node);
+       int nsc = prom_n_size_cells(memory_node);
+
+       ip = (unsigned int *) get_property(memory_node, name, &s);
+       if (ip == NULL) {
+               printk(KERN_ERR "error: couldn't get %s property on /memory\n",
+                      name);
+               return 0;
+       }
+       s /= (nsc + nac) * 4;
+       rp = mp->regions;
+       for (i = 0; i < s; ++i, ip += nac+nsc) {
+               if (nac >= 2 && ip[nac-2] != 0)
+                       continue;
+               rp->address = ip[nac-1];
+               if (nsc >= 2 && ip[nac+nsc-2] != 0)
+                       rp->size = ~0U;
+               else
+                       rp->size = ip[nac+nsc-1];
+               ++rp;
+       }
+       mp->n_regions = rp - mp->regions;
+
+       /* Make sure the pieces are sorted. */
+       mem_pieces_sort(mp);
+       mem_pieces_coalesce(mp);
+       return 1;
+}
+
+/*
+ * Collect information about physical RAM and which pieces are
+ * already in use from the device tree.
+ */
+unsigned long __init find_end_of_memory(void)
+{
+       unsigned long a, total;
+       struct mem_pieces phys_mem;
+
+       /*
+        * Find out where physical memory is, and check that it
+        * starts at 0 and is contiguous.  It seems that RAM is
+        * always physically contiguous on Power Macintoshes.
+        *
+        * Supporting discontiguous physical memory isn't hard,
+        * it just makes the virtual <-> physical mapping functions
+        * more complicated (or else you end up wasting space
+        * in mem_map).
+        */
+       memory_node = find_devices("memory");
+       if (memory_node == NULL || !get_mem_prop("reg", &phys_mem)
+           || phys_mem.n_regions == 0)
+               panic("No RAM??");
+       a = phys_mem.regions[0].address;
+       if (a != 0)
+               panic("RAM doesn't start at physical address 0");
+       total = phys_mem.regions[0].size;
+
+       if (phys_mem.n_regions > 1) {
+               printk("RAM starting at 0x%x is not contiguous\n",
+                      phys_mem.regions[1].address);
+               printk("Using RAM from 0 to 0x%lx\n", total-1);
+       }
+
+       return total;
+}
+
+/*
+ * Check for command-line options that affect what MMU_init will do.
+ */
+void MMU_setup(void)
+{
+       /* Check for nobats option (used in mapin_ram). */
+       if (strstr(cmd_line, "nobats")) {
+               __map_without_bats = 1;
+       }
+
+       if (strstr(cmd_line, "noltlbs")) {
+               __map_without_ltlbs = 1;
+       }
+
+       /* Look for mem= option on command line */
+       if (strstr(cmd_line, "mem=")) {
+               char *p, *q;
+               unsigned long maxmem = 0;
+
+               for (q = cmd_line; (p = strstr(q, "mem=")) != 0; ) {
+                       q = p + 4;
+                       if (p > cmd_line && p[-1] != ' ')
+                               continue;
+                       maxmem = simple_strtoul(q, &q, 0);
+                       if (*q == 'k' || *q == 'K') {
+                               maxmem <<= 10;
+                               ++q;
+                       } else if (*q == 'm' || *q == 'M') {
+                               maxmem <<= 20;
+                               ++q;
+                       }
+               }
+               __max_memory = maxmem;
+       }
+}
+
+/*
+ * MMU_init sets up the basic memory mappings for the kernel,
+ * including both RAM and possibly some I/O regions,
+ * and sets up the page tables and the MMU hardware ready to go.
+ */
+void __init MMU_init(void)
+{
+       if (ppc_md.progress)
+               ppc_md.progress("MMU:enter", 0x111);
+
+       /* parse args from command line */
+       MMU_setup();
+
+       /*
+        * Figure out how much memory we have, how much
+        * is lowmem, and how much is highmem.  If we were
+        * passed the total memory size from the bootloader,
+        * just use it.
+        */
+       if (boot_mem_size)
+               total_memory = boot_mem_size;
+       else
+               total_memory = ppc_md.find_end_of_memory();
+
+       if (__max_memory && total_memory > __max_memory)
+               total_memory = __max_memory;
+       total_lowmem = total_memory;
+#ifdef CONFIG_FSL_BOOKE
+       /* Freescale Book-E parts expect lowmem to be mapped by fixed TLB
+        * entries, so we need to adjust lowmem to match the amount we can map
+        * in the fixed entries */
+       adjust_total_lowmem();
+#endif /* CONFIG_FSL_BOOKE */
+       if (total_lowmem > __max_low_memory) {
+               total_lowmem = __max_low_memory;
+#ifndef CONFIG_HIGHMEM
+               total_memory = total_lowmem;
+#endif /* CONFIG_HIGHMEM */
+       }
+       set_phys_avail(total_lowmem);
+
+       /* Initialize the MMU hardware */
+       if (ppc_md.progress)
+               ppc_md.progress("MMU:hw init", 0x300);
+       MMU_init_hw();
+
+       /* Map in all of RAM starting at KERNELBASE */
+       if (ppc_md.progress)
+               ppc_md.progress("MMU:mapin", 0x301);
+       mapin_ram();
+
+#ifdef CONFIG_HIGHMEM
+       ioremap_base = PKMAP_BASE;
+#else
+       ioremap_base = 0xfe000000UL;    /* for now, could be 0xfffff000 */
+#endif /* CONFIG_HIGHMEM */
+       ioremap_bot = ioremap_base;
+
+       /* Map in I/O resources */
+       if (ppc_md.progress)
+               ppc_md.progress("MMU:setio", 0x302);
+       if (ppc_md.setup_io_mappings)
+               ppc_md.setup_io_mappings();
+
+       /* Initialize the context management stuff */
+       mmu_context_init();
+
+       if (ppc_md.progress)
+               ppc_md.progress("MMU:exit", 0x211);
+
+#ifdef CONFIG_BOOTX_TEXT
+       /* By default, we are no longer mapped */
+               boot_text_mapped = 0;
+       /* Must be done last, or ppc_md.progress will die. */
+       map_boot_text();
+#endif
+}
+
+/* This is only called until mem_init is done. */
+void __init *early_get_page(void)
+{
+       void *p;
+
+       if (init_bootmem_done) {
+               p = alloc_bootmem_pages(PAGE_SIZE);
+       } else {
+               p = mem_pieces_find(PAGE_SIZE, PAGE_SIZE);
+       }
+       return p;
+}
+
+/* Free up now-unused memory */
+static void free_sec(unsigned long start, unsigned long end, const char *name)
+{
+       unsigned long cnt = 0;
+
+       while (start < end) {
+               ClearPageReserved(virt_to_page(start));
+               set_page_count(virt_to_page(start), 1);
+               free_page(start);
+               cnt++;
+               start += PAGE_SIZE;
+       }
+       if (cnt) {
+               printk(" %ldk %s", cnt << (PAGE_SHIFT - 10), name);
+               totalram_pages += cnt;
+       }
+}
+
+void free_initmem(void)
+{
+#define FREESEC(TYPE) \
+       free_sec((unsigned long)(&__ ## TYPE ## _begin), \
+                (unsigned long)(&__ ## TYPE ## _end), \
+                #TYPE);
+
+       printk ("Freeing unused kernel memory:");
+       FREESEC(init);
+       printk("\n");
+       ppc_md.progress = NULL;
+#undef FREESEC
+}
+
+#ifdef CONFIG_BLK_DEV_INITRD
+void free_initrd_mem(unsigned long start, unsigned long end)
+{
+       if (start < end)
+               printk ("Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
+       for (; start < end; start += PAGE_SIZE) {
+               ClearPageReserved(virt_to_page(start));
+               set_page_count(virt_to_page(start), 1);
+               free_page(start);
+               totalram_pages++;
+       }
+}
+#endif
+
+/*
+ * Initialize the bootmem system and give it all the memory we
+ * have available.
+ */
+void __init do_init_bootmem(void)
+{
+       unsigned long start, size;
+       int i;
+
+       /*
+        * Find an area to use for the bootmem bitmap.
+        * We look for the first area which is at least
+        * 128kB in length (128kB is enough for a bitmap
+        * for 4GB of memory, using 4kB pages), plus 1 page
+        * (in case the address isn't page-aligned).
+        */
+       start = 0;
+       size = 0;
+       for (i = 0; i < phys_avail.n_regions; ++i) {
+               unsigned long a = phys_avail.regions[i].address;
+               unsigned long s = phys_avail.regions[i].size;
+               if (s <= size)
+                       continue;
+               start = a;
+               size = s;
+               if (s >= 33 * PAGE_SIZE)
+                       break;
+       }
+       start = PAGE_ALIGN(start);
+
+       min_low_pfn = start >> PAGE_SHIFT;
+       max_low_pfn = (PPC_MEMSTART + total_lowmem) >> PAGE_SHIFT;
+       max_pfn = (PPC_MEMSTART + total_memory) >> PAGE_SHIFT;
+       boot_mapsize = init_bootmem_node(&contig_page_data, min_low_pfn,
+                                        PPC_MEMSTART >> PAGE_SHIFT,
+                                        max_low_pfn);
+
+       /* remove the bootmem bitmap from the available memory */
+       mem_pieces_remove(&phys_avail, start, boot_mapsize, 1);
+
+       /* add everything in phys_avail into the bootmem map */
+       for (i = 0; i < phys_avail.n_regions; ++i)
+               free_bootmem(phys_avail.regions[i].address,
+                            phys_avail.regions[i].size);
+
+       init_bootmem_done = 1;
+}
+
+/*
+ * paging_init() sets up the page tables - in fact we've already done this.
+ */
+void __init paging_init(void)
+{
+       unsigned long zones_size[MAX_NR_ZONES], i;
+
+#ifdef CONFIG_HIGHMEM
+       map_page(PKMAP_BASE, 0, 0);     /* XXX gross */
+       pkmap_page_table = pte_offset_kernel(pmd_offset(pgd_offset_k
+                       (PKMAP_BASE), PKMAP_BASE), PKMAP_BASE);
+       map_page(KMAP_FIX_BEGIN, 0, 0); /* XXX gross */
+       kmap_pte = pte_offset_kernel(pmd_offset(pgd_offset_k
+                       (KMAP_FIX_BEGIN), KMAP_FIX_BEGIN), KMAP_FIX_BEGIN);
+       kmap_prot = PAGE_KERNEL;
+#endif /* CONFIG_HIGHMEM */
+
+       /*
+        * All pages are DMA-able so we put them all in the DMA zone.
+        */
+       zones_size[ZONE_DMA] = total_lowmem >> PAGE_SHIFT;
+       for (i = 1; i < MAX_NR_ZONES; i++)
+               zones_size[i] = 0;
+
+#ifdef CONFIG_HIGHMEM
+       zones_size[ZONE_HIGHMEM] = (total_memory - total_lowmem) >> PAGE_SHIFT;
+#endif /* CONFIG_HIGHMEM */
+
+       free_area_init(zones_size);
+}
+
+void __init mem_init(void)
+{
+       unsigned long addr;
+       int codepages = 0;
+       int datapages = 0;
+       int initpages = 0;
+#ifdef CONFIG_HIGHMEM
+       unsigned long highmem_mapnr;
+
+       highmem_mapnr = total_lowmem >> PAGE_SHIFT;
+#endif /* CONFIG_HIGHMEM */
+       max_mapnr = total_memory >> PAGE_SHIFT;
+
+       high_memory = (void *) __va(PPC_MEMSTART + total_lowmem);
+       num_physpages = max_mapnr;      /* RAM is assumed contiguous */
+
+       totalram_pages += free_all_bootmem();
+
+#ifdef CONFIG_BLK_DEV_INITRD
+       /* if we are booted from BootX with an initial ramdisk,
+          make sure the ramdisk pages aren't reserved. */
+       if (initrd_start) {
+               for (addr = initrd_start; addr < initrd_end; addr += PAGE_SIZE)
+                       ClearPageReserved(virt_to_page(addr));
+       }
+#endif /* CONFIG_BLK_DEV_INITRD */
+
+#ifdef CONFIG_PPC_OF
+       /* mark the RTAS pages as reserved */
+       if ( rtas_data )
+               for (addr = (ulong)__va(rtas_data);
+                    addr < PAGE_ALIGN((ulong)__va(rtas_data)+rtas_size) ;
+                    addr += PAGE_SIZE)
+                       SetPageReserved(virt_to_page(addr));
+#endif
+#ifdef CONFIG_PPC_PMAC
+       if (agp_special_page)
+               SetPageReserved(virt_to_page(agp_special_page));
+#endif
+       for (addr = PAGE_OFFSET; addr < (unsigned long)high_memory;
+            addr += PAGE_SIZE) {
+               if (!PageReserved(virt_to_page(addr)))
+                       continue;
+               if (addr < (ulong) etext)
+                       codepages++;
+               else if (addr >= (unsigned long)&__init_begin
+                        && addr < (unsigned long)&__init_end)
+                       initpages++;
+               else if (addr < (ulong) klimit)
+                       datapages++;
+       }
+
+#ifdef CONFIG_HIGHMEM
+       {
+               unsigned long pfn;
+
+               for (pfn = highmem_mapnr; pfn < max_mapnr; ++pfn) {
+                       struct page *page = mem_map + pfn;
+
+                       ClearPageReserved(page);
+                       set_page_count(page, 1);
+                       __free_page(page);
+                       totalhigh_pages++;
+               }
+               totalram_pages += totalhigh_pages;
+       }
+#endif /* CONFIG_HIGHMEM */
+
+        printk("Memory: %luk available (%dk kernel code, %dk data, %dk init, %ldk highmem)\n",
+              (unsigned long)nr_free_pages()<< (PAGE_SHIFT-10),
+              codepages<< (PAGE_SHIFT-10), datapages<< (PAGE_SHIFT-10),
+              initpages<< (PAGE_SHIFT-10),
+              (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10)));
+
+#ifdef CONFIG_PPC_PMAC
+       if (agp_special_page)
+               printk(KERN_INFO "AGP special page: 0x%08lx\n", agp_special_page);
+#endif
+
+       mem_init_done = 1;
+}
+
+/*
+ * Set phys_avail to the amount of physical memory,
+ * less the kernel text/data/bss.
+ */
+void __init
+set_phys_avail(unsigned long total_memory)
+{
+       unsigned long kstart, ksize;
+
+       /*
+        * Initially, available physical memory is equivalent to all
+        * physical memory.
+        */
+
+       phys_avail.regions[0].address = PPC_MEMSTART;
+       phys_avail.regions[0].size = total_memory;
+       phys_avail.n_regions = 1;
+
+       /*
+        * Map out the kernel text/data/bss from the available physical
+        * memory.
+        */
+
+       kstart = __pa(_stext);  /* should be 0 */
+       ksize = PAGE_ALIGN(klimit - _stext);
+
+       mem_pieces_remove(&phys_avail, kstart, ksize, 0);
+       mem_pieces_remove(&phys_avail, 0, 0x4000, 0);
+
+#if defined(CONFIG_BLK_DEV_INITRD)
+       /* Remove the init RAM disk from the available memory. */
+       if (initrd_start) {
+               mem_pieces_remove(&phys_avail, __pa(initrd_start),
+                                 initrd_end - initrd_start, 1);
+       }
+#endif /* CONFIG_BLK_DEV_INITRD */
+#ifdef CONFIG_PPC_OF
+       /* remove the RTAS pages from the available memory */
+       if (rtas_data)
+               mem_pieces_remove(&phys_avail, rtas_data, rtas_size, 1);
+#endif
+#ifdef CONFIG_PPC_PMAC
+       /* Because of some uninorth weirdness, we need a page of
+        * memory as high as possible (it must be outside of the
+        * bus address seen as the AGP aperture). It will be used
+        * by the r128 DRM driver
+        *
+        * FIXME: We need to make sure that page doesn't overlap any of the\
+        * above. This could be done by improving mem_pieces_find to be able
+        * to do a backward search from the end of the list.
+        */
+       if (_machine == _MACH_Pmac && find_devices("uni-north-agp")) {
+               agp_special_page = (total_memory - PAGE_SIZE);
+               mem_pieces_remove(&phys_avail, agp_special_page, PAGE_SIZE, 0);
+               agp_special_page = (unsigned long)__va(agp_special_page);
+       }
+#endif /* CONFIG_PPC_PMAC */
+}
+
+/* Mark some memory as reserved by removing it from phys_avail. */
+void __init reserve_phys_mem(unsigned long start, unsigned long size)
+{
+       mem_pieces_remove(&phys_avail, start, size, 1);
+}
diff --git a/arch/powerpc/mm/init64.c b/arch/powerpc/mm/init64.c
new file mode 100644 (file)
index 0000000..81f6745
--- /dev/null
@@ -0,0 +1,385 @@
+/*
+ *  PowerPC version
+ *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
+ *
+ *  Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
+ *  and Cort Dougan (PReP) (cort@cs.nmt.edu)
+ *    Copyright (C) 1996 Paul Mackerras
+ *  Amiga/APUS changes by Jesper Skov (jskov@cygnus.co.uk).
+ *
+ *  Derived from "arch/i386/mm/init.c"
+ *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
+ *
+ *  Dave Engebretsen <engebret@us.ibm.com>
+ *      Rework for PPC64 port.
+ *
+ *  This program is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU General Public License
+ *  as published by the Free Software Foundation; either version
+ *  2 of the License, or (at your option) any later version.
+ *
+ */
+
+#include <linux/config.h>
+#include <linux/signal.h>
+#include <linux/sched.h>
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/string.h>
+#include <linux/types.h>
+#include <linux/mman.h>
+#include <linux/mm.h>
+#include <linux/swap.h>
+#include <linux/stddef.h>
+#include <linux/vmalloc.h>
+#include <linux/init.h>
+#include <linux/delay.h>
+#include <linux/bootmem.h>
+#include <linux/highmem.h>
+#include <linux/idr.h>
+#include <linux/nodemask.h>
+#include <linux/module.h>
+
+#include <asm/pgalloc.h>
+#include <asm/page.h>
+#include <asm/prom.h>
+#include <asm/lmb.h>
+#include <asm/rtas.h>
+#include <asm/io.h>
+#include <asm/mmu_context.h>
+#include <asm/pgtable.h>
+#include <asm/mmu.h>
+#include <asm/uaccess.h>
+#include <asm/smp.h>
+#include <asm/machdep.h>
+#include <asm/tlb.h>
+#include <asm/eeh.h>
+#include <asm/processor.h>
+#include <asm/mmzone.h>
+#include <asm/cputable.h>
+#include <asm/ppcdebug.h>
+#include <asm/sections.h>
+#include <asm/system.h>
+#include <asm/iommu.h>
+#include <asm/abs_addr.h>
+#include <asm/vdso.h>
+#include <asm/imalloc.h>
+
+#if PGTABLE_RANGE > USER_VSID_RANGE
+#warning Limited user VSID range means pagetable space is wasted
+#endif
+
+#if (TASK_SIZE_USER64 < PGTABLE_RANGE) && (TASK_SIZE_USER64 < USER_VSID_RANGE)
+#warning TASK_SIZE is smaller than it needs to be.
+#endif
+
+int mem_init_done;
+unsigned long ioremap_bot = IMALLOC_BASE;
+static unsigned long phbs_io_bot = PHBS_IO_BASE;
+
+extern pgd_t swapper_pg_dir[];
+extern struct task_struct *current_set[NR_CPUS];
+
+unsigned long klimit = (unsigned long)_end;
+
+unsigned long _SDR1=0;
+unsigned long _ASR=0;
+
+/* max amount of RAM to use */
+unsigned long __max_memory;
+
+/* info on what we think the IO hole is */
+unsigned long  io_hole_start;
+unsigned long  io_hole_size;
+
+/*
+ * Do very early mm setup.
+ */
+void __init mm_init_ppc64(void)
+{
+#ifndef CONFIG_PPC_ISERIES
+       unsigned long i;
+#endif
+
+       ppc64_boot_msg(0x100, "MM Init");
+
+       /* This is the story of the IO hole... please, keep seated,
+        * unfortunately, we are out of oxygen masks at the moment.
+        * So we need some rough way to tell where your big IO hole
+        * is. On pmac, it's between 2G and 4G, on POWER3, it's around
+        * that area as well, on POWER4 we don't have one, etc...
+        * We need that as a "hint" when sizing the TCE table on POWER3
+        * So far, the simplest way that seem work well enough for us it
+        * to just assume that the first discontinuity in our physical
+        * RAM layout is the IO hole. That may not be correct in the future
+        * (and isn't on iSeries but then we don't care ;)
+        */
+
+#ifndef CONFIG_PPC_ISERIES
+       for (i = 1; i < lmb.memory.cnt; i++) {
+               unsigned long base, prevbase, prevsize;
+
+               prevbase = lmb.memory.region[i-1].base;
+               prevsize = lmb.memory.region[i-1].size;
+               base = lmb.memory.region[i].base;
+               if (base > (prevbase + prevsize)) {
+                       io_hole_start = prevbase + prevsize;
+                       io_hole_size = base  - (prevbase + prevsize);
+                       break;
+               }
+       }
+#endif /* CONFIG_PPC_ISERIES */
+       if (io_hole_start)
+               printk("IO Hole assumed to be %lx -> %lx\n",
+                      io_hole_start, io_hole_start + io_hole_size - 1);
+
+       ppc64_boot_msg(0x100, "MM Init Done");
+}
+
+void free_initmem(void)
+{
+       unsigned long addr;
+
+       addr = (unsigned long)__init_begin;
+       for (; addr < (unsigned long)__init_end; addr += PAGE_SIZE) {
+               memset((void *)addr, 0xcc, PAGE_SIZE);
+               ClearPageReserved(virt_to_page(addr));
+               set_page_count(virt_to_page(addr), 1);
+               free_page(addr);
+               totalram_pages++;
+       }
+       printk ("Freeing unused kernel memory: %luk freed\n",
+               ((unsigned long)__init_end - (unsigned long)__init_begin) >> 10);
+}
+
+#ifdef CONFIG_BLK_DEV_INITRD
+void free_initrd_mem(unsigned long start, unsigned long end)
+{
+       if (start < end)
+               printk ("Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
+       for (; start < end; start += PAGE_SIZE) {
+               ClearPageReserved(virt_to_page(start));
+               set_page_count(virt_to_page(start), 1);
+               free_page(start);
+               totalram_pages++;
+       }
+}
+#endif
+
+/*
+ * Initialize the bootmem system and give it all the memory we
+ * have available.
+ */
+#ifndef CONFIG_NEED_MULTIPLE_NODES
+void __init do_init_bootmem(void)
+{
+       unsigned long i;
+       unsigned long start, bootmap_pages;
+       unsigned long total_pages = lmb_end_of_DRAM() >> PAGE_SHIFT;
+       int boot_mapsize;
+
+       /*
+        * Find an area to use for the bootmem bitmap.  Calculate the size of
+        * bitmap required as (Total Memory) / PAGE_SIZE / BITS_PER_BYTE.
+        * Add 1 additional page in case the address isn't page-aligned.
+        */
+       bootmap_pages = bootmem_bootmap_pages(total_pages);
+
+       start = lmb_alloc(bootmap_pages<<PAGE_SHIFT, PAGE_SIZE);
+       BUG_ON(!start);
+
+       boot_mapsize = init_bootmem(start >> PAGE_SHIFT, total_pages);
+
+       max_pfn = max_low_pfn;
+
+       /* Add all physical memory to the bootmem map, mark each area
+        * present.
+        */
+       for (i=0; i < lmb.memory.cnt; i++)
+               free_bootmem(lmb.memory.region[i].base,
+                            lmb_size_bytes(&lmb.memory, i));
+
+       /* reserve the sections we're already using */
+       for (i=0; i < lmb.reserved.cnt; i++)
+               reserve_bootmem(lmb.reserved.region[i].base,
+                               lmb_size_bytes(&lmb.reserved, i));
+
+       for (i=0; i < lmb.memory.cnt; i++)
+               memory_present(0, lmb_start_pfn(&lmb.memory, i),
+                              lmb_end_pfn(&lmb.memory, i));
+}
+
+/*
+ * paging_init() sets up the page tables - in fact we've already done this.
+ */
+void __init paging_init(void)
+{
+       unsigned long zones_size[MAX_NR_ZONES];
+       unsigned long zholes_size[MAX_NR_ZONES];
+       unsigned long total_ram = lmb_phys_mem_size();
+       unsigned long top_of_ram = lmb_end_of_DRAM();
+
+       printk(KERN_INFO "Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
+              top_of_ram, total_ram);
+       printk(KERN_INFO "Memory hole size: %ldMB\n",
+              (top_of_ram - total_ram) >> 20);
+       /*
+        * All pages are DMA-able so we put them all in the DMA zone.
+        */
+       memset(zones_size, 0, sizeof(zones_size));
+       memset(zholes_size, 0, sizeof(zholes_size));
+
+       zones_size[ZONE_DMA] = top_of_ram >> PAGE_SHIFT;
+       zholes_size[ZONE_DMA] = (top_of_ram - total_ram) >> PAGE_SHIFT;
+
+       free_area_init_node(0, NODE_DATA(0), zones_size,
+                           __pa(PAGE_OFFSET) >> PAGE_SHIFT, zholes_size);
+}
+#endif /* ! CONFIG_NEED_MULTIPLE_NODES */
+
+static struct kcore_list kcore_vmem;
+
+static int __init setup_kcore(void)
+{
+       int i;
+
+       for (i=0; i < lmb.memory.cnt; i++) {
+               unsigned long base, size;
+               struct kcore_list *kcore_mem;
+
+               base = lmb.memory.region[i].base;
+               size = lmb.memory.region[i].size;
+
+               /* GFP_ATOMIC to avoid might_sleep warnings during boot */
+               kcore_mem = kmalloc(sizeof(struct kcore_list), GFP_ATOMIC);
+               if (!kcore_mem)
+                       panic("mem_init: kmalloc failed\n");
+
+               kclist_add(kcore_mem, __va(base), size);
+       }
+
+       kclist_add(&kcore_vmem, (void *)VMALLOC_START, VMALLOC_END-VMALLOC_START);
+
+       return 0;
+}
+module_init(setup_kcore);
+
+void __init mem_init(void)
+{
+#ifdef CONFIG_NEED_MULTIPLE_NODES
+       int nid;
+#endif
+       pg_data_t *pgdat;
+       unsigned long i;
+       struct page *page;
+       unsigned long reservedpages = 0, codesize, initsize, datasize, bsssize;
+
+       num_physpages = max_low_pfn;    /* RAM is assumed contiguous */
+       high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
+
+#ifdef CONFIG_NEED_MULTIPLE_NODES
+        for_each_online_node(nid) {
+               if (NODE_DATA(nid)->node_spanned_pages != 0) {
+                       printk("freeing bootmem node %x\n", nid);
+                       totalram_pages +=
+                               free_all_bootmem_node(NODE_DATA(nid));
+               }
+       }
+#else
+       max_mapnr = num_physpages;
+       totalram_pages += free_all_bootmem();
+#endif
+
+       for_each_pgdat(pgdat) {
+               for (i = 0; i < pgdat->node_spanned_pages; i++) {
+                       page = pgdat_page_nr(pgdat, i);
+                       if (PageReserved(page))
+                               reservedpages++;
+               }
+       }
+
+       codesize = (unsigned long)&_etext - (unsigned long)&_stext;
+       initsize = (unsigned long)&__init_end - (unsigned long)&__init_begin;
+       datasize = (unsigned long)&_edata - (unsigned long)&__init_end;
+       bsssize = (unsigned long)&__bss_stop - (unsigned long)&__bss_start;
+
+       printk(KERN_INFO "Memory: %luk/%luk available (%luk kernel code, "
+              "%luk reserved, %luk data, %luk bss, %luk init)\n",
+               (unsigned long)nr_free_pages() << (PAGE_SHIFT-10),
+               num_physpages << (PAGE_SHIFT-10),
+               codesize >> 10,
+               reservedpages << (PAGE_SHIFT-10),
+               datasize >> 10,
+               bsssize >> 10,
+               initsize >> 10);
+
+       mem_init_done = 1;
+
+       /* Initialize the vDSO */
+       vdso_init();
+}
+
+void __iomem * reserve_phb_iospace(unsigned long size)
+{
+       void __iomem *virt_addr;
+               
+       if (phbs_io_bot >= IMALLOC_BASE) 
+               panic("reserve_phb_iospace(): phb io space overflow\n");
+                       
+       virt_addr = (void __iomem *) phbs_io_bot;
+       phbs_io_bot += size;
+
+       return virt_addr;
+}
+
+static void zero_ctor(void *addr, kmem_cache_t *cache, unsigned long flags)
+{
+       memset(addr, 0, kmem_cache_size(cache));
+}
+
+static const int pgtable_cache_size[2] = {
+       PTE_TABLE_SIZE, PMD_TABLE_SIZE
+};
+static const char *pgtable_cache_name[ARRAY_SIZE(pgtable_cache_size)] = {
+       "pgd_pte_cache", "pud_pmd_cache",
+};
+
+kmem_cache_t *pgtable_cache[ARRAY_SIZE(pgtable_cache_size)];
+
+void pgtable_cache_init(void)
+{
+       int i;
+
+       BUILD_BUG_ON(PTE_TABLE_SIZE != pgtable_cache_size[PTE_CACHE_NUM]);
+       BUILD_BUG_ON(PMD_TABLE_SIZE != pgtable_cache_size[PMD_CACHE_NUM]);
+       BUILD_BUG_ON(PUD_TABLE_SIZE != pgtable_cache_size[PUD_CACHE_NUM]);
+       BUILD_BUG_ON(PGD_TABLE_SIZE != pgtable_cache_size[PGD_CACHE_NUM]);
+
+       for (i = 0; i < ARRAY_SIZE(pgtable_cache_size); i++) {
+               int size = pgtable_cache_size[i];
+               const char *name = pgtable_cache_name[i];
+
+               pgtable_cache[i] = kmem_cache_create(name,
+                                                    size, size,
+                                                    SLAB_HWCACHE_ALIGN
+                                                    | SLAB_MUST_HWCACHE_ALIGN,
+                                                    zero_ctor,
+                                                    NULL);
+               if (! pgtable_cache[i])
+                       panic("pgtable_cache_init(): could not create %s!\n",
+                             name);
+       }
+}
+
+pgprot_t phys_mem_access_prot(struct file *file, unsigned long addr,
+                             unsigned long size, pgprot_t vma_prot)
+{
+       if (ppc_md.phys_mem_access_prot)
+               return ppc_md.phys_mem_access_prot(file, addr, size, vma_prot);
+
+       if (!page_is_ram(addr >> PAGE_SHIFT))
+               vma_prot = __pgprot(pgprot_val(vma_prot)
+                                   | _PAGE_GUARDED | _PAGE_NO_CACHE);
+       return vma_prot;
+}
+EXPORT_SYMBOL(phys_mem_access_prot);
diff --git a/arch/powerpc/mm/mem.c b/arch/powerpc/mm/mem.c
new file mode 100644 (file)
index 0000000..345db08
--- /dev/null
@@ -0,0 +1,299 @@
+/*
+ *  PowerPC version
+ *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
+ *
+ *  Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
+ *  and Cort Dougan (PReP) (cort@cs.nmt.edu)
+ *    Copyright (C) 1996 Paul Mackerras
+ *  Amiga/APUS changes by Jesper Skov (jskov@cygnus.co.uk).
+ *  PPC44x/36-bit changes by Matt Porter (mporter@mvista.com)
+ *
+ *  Derived from "arch/i386/mm/init.c"
+ *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
+ *
+ *  This program is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU General Public License
+ *  as published by the Free Software Foundation; either version
+ *  2 of the License, or (at your option) any later version.
+ *
+ */
+
+#include <linux/config.h>
+#include <linux/module.h>
+#include <linux/sched.h>
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/string.h>
+#include <linux/types.h>
+#include <linux/mm.h>
+#include <linux/stddef.h>
+#include <linux/init.h>
+#include <linux/bootmem.h>
+#include <linux/highmem.h>
+#include <linux/initrd.h>
+#include <linux/pagemap.h>
+
+#include <asm/pgalloc.h>
+#include <asm/prom.h>
+#include <asm/io.h>
+#include <asm/mmu_context.h>
+#include <asm/pgtable.h>
+#include <asm/mmu.h>
+#include <asm/smp.h>
+#include <asm/machdep.h>
+#include <asm/btext.h>
+#include <asm/tlb.h>
+#include <asm/bootinfo.h>
+#include <asm/prom.h>
+
+#include "mem_pieces.h"
+#include "mmu_decl.h"
+
+#ifndef CPU_FTR_COHERENT_ICACHE
+#define CPU_FTR_COHERENT_ICACHE        0       /* XXX for now */
+#define CPU_FTR_NOEXECUTE      0
+#endif
+
+/*
+ * This is called by /dev/mem to know if a given address has to
+ * be mapped non-cacheable or not
+ */
+int page_is_ram(unsigned long pfn)
+{
+       unsigned long paddr = (pfn << PAGE_SHIFT);
+
+#ifndef CONFIG_PPC64   /* XXX for now */
+       return paddr < __pa(high_memory);
+#else
+       int i;
+       for (i=0; i < lmb.memory.cnt; i++) {
+               unsigned long base;
+
+               base = lmb.memory.region[i].base;
+
+               if ((paddr >= base) &&
+                       (paddr < (base + lmb.memory.region[i].size))) {
+                       return 1;
+               }
+       }
+
+       return 0;
+#endif
+}
+EXPORT_SYMBOL(page_is_ram);
+
+pgprot_t phys_mem_access_prot(struct file *file, unsigned long addr,
+                             unsigned long size, pgprot_t vma_prot)
+{
+       if (ppc_md.phys_mem_access_prot)
+               return ppc_md.phys_mem_access_prot(file, addr, size, vma_prot);
+
+       if (!page_is_ram(addr >> PAGE_SHIFT))
+               vma_prot = __pgprot(pgprot_val(vma_prot)
+                                   | _PAGE_GUARDED | _PAGE_NO_CACHE);
+       return vma_prot;
+}
+EXPORT_SYMBOL(phys_mem_access_prot);
+
+void show_mem(void)
+{
+       unsigned long total = 0, reserved = 0;
+       unsigned long shared = 0, cached = 0;
+       unsigned long highmem = 0;
+       struct page *page;
+       pg_data_t *pgdat;
+       unsigned long i;
+
+       printk("Mem-info:\n");
+       show_free_areas();
+       printk("Free swap:       %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10));
+       for_each_pgdat(pgdat) {
+               for (i = 0; i < pgdat->node_spanned_pages; i++) {
+                       page = pgdat_page_nr(pgdat, i);
+                       total++;
+                       if (PageHighMem(page))
+                               highmem++;
+                       if (PageReserved(page))
+                               reserved++;
+                       else if (PageSwapCache(page))
+                               cached++;
+                       else if (page_count(page))
+                               shared += page_count(page) - 1;
+               }
+       }
+       printk("%ld pages of RAM\n", total);
+#ifdef CONFIG_HIGHMEM
+       printk("%ld pages of HIGHMEM\n", highmem);
+#endif
+       printk("%ld reserved pages\n", reserved);
+       printk("%ld pages shared\n", shared);
+       printk("%ld pages swap cached\n", cached);
+}
+
+/*
+ * This is called when a page has been modified by the kernel.
+ * It just marks the page as not i-cache clean.  We do the i-cache
+ * flush later when the page is given to a user process, if necessary.
+ */
+void flush_dcache_page(struct page *page)
+{
+       if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
+               return;
+       /* avoid an atomic op if possible */
+       if (test_bit(PG_arch_1, &page->flags))
+               clear_bit(PG_arch_1, &page->flags);
+}
+EXPORT_SYMBOL(flush_dcache_page);
+
+void flush_dcache_icache_page(struct page *page)
+{
+#ifdef CONFIG_BOOKE
+       void *start = kmap_atomic(page, KM_PPC_SYNC_ICACHE);
+       __flush_dcache_icache(start);
+       kunmap_atomic(start, KM_PPC_SYNC_ICACHE);
+#elif defined(CONFIG_8xx)
+       /* On 8xx there is no need to kmap since highmem is not supported */
+       __flush_dcache_icache(page_address(page)); 
+#else
+       __flush_dcache_icache_phys(page_to_pfn(page) << PAGE_SHIFT);
+#endif
+
+}
+void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
+{
+       clear_page(page);
+
+       if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
+               return;
+       /*
+        * We shouldnt have to do this, but some versions of glibc
+        * require it (ld.so assumes zero filled pages are icache clean)
+        * - Anton
+        */
+
+       /* avoid an atomic op if possible */
+       if (test_bit(PG_arch_1, &pg->flags))
+               clear_bit(PG_arch_1, &pg->flags);
+}
+EXPORT_SYMBOL(clear_user_page);
+
+void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
+                   struct page *pg)
+{
+       copy_page(vto, vfrom);
+
+       /*
+        * We should be able to use the following optimisation, however
+        * there are two problems.
+        * Firstly a bug in some versions of binutils meant PLT sections
+        * were not marked executable.
+        * Secondly the first word in the GOT section is blrl, used
+        * to establish the GOT address. Until recently the GOT was
+        * not marked executable.
+        * - Anton
+        */
+#if 0
+       if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
+               return;
+#endif
+
+       if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
+               return;
+
+       /* avoid an atomic op if possible */
+       if (test_bit(PG_arch_1, &pg->flags))
+               clear_bit(PG_arch_1, &pg->flags);
+}
+
+void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
+                            unsigned long addr, int len)
+{
+       unsigned long maddr;
+
+       maddr = (unsigned long) kmap(page) + (addr & ~PAGE_MASK);
+       flush_icache_range(maddr, maddr + len);
+       kunmap(page);
+}
+EXPORT_SYMBOL(flush_icache_user_range);
+
+/*
+ * This is called at the end of handling a user page fault, when the
+ * fault has been handled by updating a PTE in the linux page tables.
+ * We use it to preload an HPTE into the hash table corresponding to
+ * the updated linux PTE.
+ * 
+ * This must always be called with the mm->page_table_lock held
+ */
+void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
+                     pte_t pte)
+{
+       /* handle i-cache coherency */
+       unsigned long pfn = pte_pfn(pte);
+#ifdef CONFIG_PPC32
+       pmd_t *pmd;
+#else
+       unsigned long vsid;
+       void *pgdir;
+       pte_t *ptep;
+       int local = 0;
+       cpumask_t tmp;
+       unsigned long flags;
+#endif
+
+       /* handle i-cache coherency */
+       if (!cpu_has_feature(CPU_FTR_COHERENT_ICACHE) &&
+           !cpu_has_feature(CPU_FTR_NOEXECUTE) &&
+           pfn_valid(pfn)) {
+               struct page *page = pfn_to_page(pfn);
+               if (!PageReserved(page)
+                   && !test_bit(PG_arch_1, &page->flags)) {
+                       if (vma->vm_mm == current->active_mm) {
+#ifdef CONFIG_8xx
+                       /* On 8xx, cache control instructions (particularly 
+                        * "dcbst" from flush_dcache_icache) fault as write 
+                        * operation if there is an unpopulated TLB entry 
+                        * for the address in question. To workaround that, 
+                        * we invalidate the TLB here, thus avoiding dcbst 
+                        * misbehaviour.
+                        */
+                               _tlbie(address);
+#endif
+                               __flush_dcache_icache((void *) address);
+                       } else
+                               flush_dcache_icache_page(page);
+                       set_bit(PG_arch_1, &page->flags);
+               }
+       }
+
+#ifdef CONFIG_PPC_STD_MMU
+       /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
+       if (!pte_young(pte) || address >= TASK_SIZE)
+               return;
+#ifdef CONFIG_PPC32
+       if (Hash == 0)
+               return;
+       pmd = pmd_offset(pgd_offset(vma->vm_mm, address), address);
+       if (!pmd_none(*pmd))
+               add_hash_page(vma->vm_mm->context, address, pmd_val(*pmd));
+#else
+       pgdir = vma->vm_mm->pgd;
+       if (pgdir == NULL)
+               return;
+
+       ptep = find_linux_pte(pgdir, ea);
+       if (!ptep)
+               return;
+
+       vsid = get_vsid(vma->vm_mm->context.id, ea);
+
+       local_irq_save(flags);
+       tmp = cpumask_of_cpu(smp_processor_id());
+       if (cpus_equal(vma->vm_mm->cpu_vm_mask, tmp))
+               local = 1;
+
+       __hash_page(ea, pte_val(pte) & (_PAGE_USER|_PAGE_RW), vsid, ptep,
+                   0x300, local);
+       local_irq_restore(flags);
+#endif
+#endif
+}
diff --git a/arch/powerpc/mm/mem64.c b/arch/powerpc/mm/mem64.c
new file mode 100644 (file)
index 0000000..ef765a8
--- /dev/null
@@ -0,0 +1,259 @@
+/*
+ *  PowerPC version 
+ *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
+ *
+ *  Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
+ *  and Cort Dougan (PReP) (cort@cs.nmt.edu)
+ *    Copyright (C) 1996 Paul Mackerras
+ *  Amiga/APUS changes by Jesper Skov (jskov@cygnus.co.uk).
+ *
+ *  Derived from "arch/i386/mm/init.c"
+ *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
+ *
+ *  Dave Engebretsen <engebret@us.ibm.com>
+ *      Rework for PPC64 port.
+ *
+ *  This program is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU General Public License
+ *  as published by the Free Software Foundation; either version
+ *  2 of the License, or (at your option) any later version.
+ *
+ */
+
+#include <linux/config.h>
+#include <linux/signal.h>
+#include <linux/sched.h>
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/string.h>
+#include <linux/types.h>
+#include <linux/mman.h>
+#include <linux/mm.h>
+#include <linux/swap.h>
+#include <linux/stddef.h>
+#include <linux/vmalloc.h>
+#include <linux/init.h>
+#include <linux/delay.h>
+#include <linux/bootmem.h>
+#include <linux/highmem.h>
+#include <linux/idr.h>
+#include <linux/nodemask.h>
+#include <linux/module.h>
+
+#include <asm/pgalloc.h>
+#include <asm/page.h>
+#include <asm/prom.h>
+#include <asm/lmb.h>
+#include <asm/rtas.h>
+#include <asm/io.h>
+#include <asm/mmu_context.h>
+#include <asm/pgtable.h>
+#include <asm/mmu.h>
+#include <asm/uaccess.h>
+#include <asm/smp.h>
+#include <asm/machdep.h>
+#include <asm/tlb.h>
+#include <asm/eeh.h>
+#include <asm/processor.h>
+#include <asm/mmzone.h>
+#include <asm/cputable.h>
+#include <asm/ppcdebug.h>
+#include <asm/sections.h>
+#include <asm/system.h>
+#include <asm/iommu.h>
+#include <asm/abs_addr.h>
+#include <asm/vdso.h>
+#include <asm/imalloc.h>
+
+/*
+ * This is called by /dev/mem to know if a given address has to
+ * be mapped non-cacheable or not
+ */
+int page_is_ram(unsigned long pfn)
+{
+       int i;
+       unsigned long paddr = (pfn << PAGE_SHIFT);
+
+       for (i=0; i < lmb.memory.cnt; i++) {
+               unsigned long base;
+
+               base = lmb.memory.region[i].base;
+
+               if ((paddr >= base) &&
+                       (paddr < (base + lmb.memory.region[i].size))) {
+                       return 1;
+               }
+       }
+
+       return 0;
+}
+EXPORT_SYMBOL(page_is_ram);
+
+pgprot_t phys_mem_access_prot(struct file *file, unsigned long addr,
+                             unsigned long size, pgprot_t vma_prot)
+{
+       if (ppc_md.phys_mem_access_prot)
+               return ppc_md.phys_mem_access_prot(file, addr, size, vma_prot);
+
+       if (!page_is_ram(addr >> PAGE_SHIFT))
+               vma_prot = __pgprot(pgprot_val(vma_prot)
+                                   | _PAGE_GUARDED | _PAGE_NO_CACHE);
+       return vma_prot;
+}
+EXPORT_SYMBOL(phys_mem_access_prot);
+
+void show_mem(void)
+{
+       unsigned long total = 0, reserved = 0;
+       unsigned long shared = 0, cached = 0;
+       struct page *page;
+       pg_data_t *pgdat;
+       unsigned long i;
+
+       printk("Mem-info:\n");
+       show_free_areas();
+       printk("Free swap:       %6ldkB\n", nr_swap_pages<<(PAGE_SHIFT-10));
+       for_each_pgdat(pgdat) {
+               for (i = 0; i < pgdat->node_spanned_pages; i++) {
+                       page = pgdat_page_nr(pgdat, i);
+                       total++;
+                       if (PageReserved(page))
+                               reserved++;
+                       else if (PageSwapCache(page))
+                               cached++;
+                       else if (page_count(page))
+                               shared += page_count(page) - 1;
+               }
+       }
+       printk("%ld pages of RAM\n", total);
+       printk("%ld reserved pages\n", reserved);
+       printk("%ld pages shared\n", shared);
+       printk("%ld pages swap cached\n", cached);
+}
+
+/*
+ * This is called when a page has been modified by the kernel.
+ * It just marks the page as not i-cache clean.  We do the i-cache
+ * flush later when the page is given to a user process, if necessary.
+ */
+void flush_dcache_page(struct page *page)
+{
+       if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
+               return;
+       /* avoid an atomic op if possible */
+       if (test_bit(PG_arch_1, &page->flags))
+               clear_bit(PG_arch_1, &page->flags);
+}
+EXPORT_SYMBOL(flush_dcache_page);
+
+void clear_user_page(void *page, unsigned long vaddr, struct page *pg)
+{
+       clear_page(page);
+
+       if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
+               return;
+       /*
+        * We shouldnt have to do this, but some versions of glibc
+        * require it (ld.so assumes zero filled pages are icache clean)
+        * - Anton
+        */
+
+       /* avoid an atomic op if possible */
+       if (test_bit(PG_arch_1, &pg->flags))
+               clear_bit(PG_arch_1, &pg->flags);
+}
+EXPORT_SYMBOL(clear_user_page);
+
+void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
+                   struct page *pg)
+{
+       copy_page(vto, vfrom);
+
+       /*
+        * We should be able to use the following optimisation, however
+        * there are two problems.
+        * Firstly a bug in some versions of binutils meant PLT sections
+        * were not marked executable.
+        * Secondly the first word in the GOT section is blrl, used
+        * to establish the GOT address. Until recently the GOT was
+        * not marked executable.
+        * - Anton
+        */
+#if 0
+       if (!vma->vm_file && ((vma->vm_flags & VM_EXEC) == 0))
+               return;
+#endif
+
+       if (cpu_has_feature(CPU_FTR_COHERENT_ICACHE))
+               return;
+
+       /* avoid an atomic op if possible */
+       if (test_bit(PG_arch_1, &pg->flags))
+               clear_bit(PG_arch_1, &pg->flags);
+}
+
+void flush_icache_user_range(struct vm_area_struct *vma, struct page *page,
+                            unsigned long addr, int len)
+{
+       unsigned long maddr;
+
+       maddr = (unsigned long)page_address(page) + (addr & ~PAGE_MASK);
+       flush_icache_range(maddr, maddr + len);
+}
+EXPORT_SYMBOL(flush_icache_user_range);
+
+/*
+ * This is called at the end of handling a user page fault, when the
+ * fault has been handled by updating a PTE in the linux page tables.
+ * We use it to preload an HPTE into the hash table corresponding to
+ * the updated linux PTE.
+ * 
+ * This must always be called with the mm->page_table_lock held
+ */
+void update_mmu_cache(struct vm_area_struct *vma, unsigned long ea,
+                     pte_t pte)
+{
+       unsigned long vsid;
+       void *pgdir;
+       pte_t *ptep;
+       int local = 0;
+       cpumask_t tmp;
+       unsigned long flags;
+
+       /* handle i-cache coherency */
+       if (!cpu_has_feature(CPU_FTR_COHERENT_ICACHE) &&
+           !cpu_has_feature(CPU_FTR_NOEXECUTE)) {
+               unsigned long pfn = pte_pfn(pte);
+               if (pfn_valid(pfn)) {
+                       struct page *page = pfn_to_page(pfn);
+                       if (!PageReserved(page)
+                           && !test_bit(PG_arch_1, &page->flags)) {
+                               __flush_dcache_icache(page_address(page));
+                               set_bit(PG_arch_1, &page->flags);
+                       }
+               }
+       }
+
+       /* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
+       if (!pte_young(pte))
+               return;
+
+       pgdir = vma->vm_mm->pgd;
+       if (pgdir == NULL)
+               return;
+
+       ptep = find_linux_pte(pgdir, ea);
+       if (!ptep)
+               return;
+
+       vsid = get_vsid(vma->vm_mm->context.id, ea);
+
+       local_irq_save(flags);
+       tmp = cpumask_of_cpu(smp_processor_id());
+       if (cpus_equal(vma->vm_mm->cpu_vm_mask, tmp))
+               local = 1;
+
+       __hash_page(ea, pte_val(pte) & (_PAGE_USER|_PAGE_RW), vsid, ptep,
+                   0x300, local);
+       local_irq_restore(flags);
+}
diff --git a/arch/powerpc/mm/mem_pieces.c b/arch/powerpc/mm/mem_pieces.c
new file mode 100644 (file)
index 0000000..3d63905
--- /dev/null
@@ -0,0 +1,163 @@
+/*
+ *    Copyright (c) 1996 Paul Mackerras <paulus@cs.anu.edu.au>
+ *      Changes to accommodate Power Macintoshes.
+ *    Cort Dougan <cort@cs.nmt.edu>
+ *      Rewrites.
+ *    Grant Erickson <grant@lcse.umn.edu>
+ *      General rework and split from mm/init.c.
+ *
+ *    Module name: mem_pieces.c
+ *
+ *    Description:
+ *      Routines and data structures for manipulating and representing
+ *      phyiscal memory extents (i.e. address/length pairs).
+ *
+ */
+
+#include <linux/config.h>
+#include <linux/kernel.h>
+#include <linux/stddef.h>
+#include <linux/init.h>
+#include <asm/page.h>
+
+#include "mem_pieces.h"
+
+extern struct mem_pieces phys_avail;
+
+static void mem_pieces_print(struct mem_pieces *);
+
+/*
+ * Scan a region for a piece of a given size with the required alignment.
+ */
+void __init *
+mem_pieces_find(unsigned int size, unsigned int align)
+{
+       int i;
+       unsigned a, e;
+       struct mem_pieces *mp = &phys_avail;
+
+       for (i = 0; i < mp->n_regions; ++i) {
+               a = mp->regions[i].address;
+               e = a + mp->regions[i].size;
+               a = (a + align - 1) & -align;
+               if (a + size <= e) {
+                       mem_pieces_remove(mp, a, size, 1);
+                       return (void *) __va(a);
+               }
+       }
+       panic("Couldn't find %u bytes at %u alignment\n", size, align);
+
+       return NULL;
+}
+
+/*
+ * Remove some memory from an array of pieces
+ */
+void __init
+mem_pieces_remove(struct mem_pieces *mp, unsigned int start, unsigned int size,
+                 int must_exist)
+{
+       int i, j;
+       unsigned int end, rs, re;
+       struct reg_property *rp;
+
+       end = start + size;
+       for (i = 0, rp = mp->regions; i < mp->n_regions; ++i, ++rp) {
+               if (end > rp->address && start < rp->address + rp->size)
+                       break;
+       }
+       if (i >= mp->n_regions) {
+               if (must_exist)
+                       printk("mem_pieces_remove: [%x,%x) not in any region\n",
+                              start, end);
+               return;
+       }
+       for (; i < mp->n_regions && end > rp->address; ++i, ++rp) {
+               rs = rp->address;
+               re = rs + rp->size;
+               if (must_exist && (start < rs || end > re)) {
+                       printk("mem_pieces_remove: bad overlap [%x,%x) with",
+                              start, end);
+                       mem_pieces_print(mp);
+                       must_exist = 0;
+               }
+               if (start > rs) {
+                       rp->size = start - rs;
+                       if (end < re) {
+                               /* need to split this entry */
+                               if (mp->n_regions >= MEM_PIECES_MAX)
+                                       panic("eek... mem_pieces overflow");
+                               for (j = mp->n_regions; j > i + 1; --j)
+                                       mp->regions[j] = mp->regions[j-1];
+                               ++mp->n_regions;
+                               rp[1].address = end;
+                               rp[1].size = re - end;
+                       }
+               } else {
+                       if (end < re) {
+                               rp->address = end;
+                               rp->size = re - end;
+                       } else {
+                               /* need to delete this entry */
+                               for (j = i; j < mp->n_regions - 1; ++j)
+                                       mp->regions[j] = mp->regions[j+1];
+                               --mp->n_regions;
+                               --i;
+                               --rp;
+                       }
+               }
+       }
+}
+
+static void __init
+mem_pieces_print(struct mem_pieces *mp)
+{
+       int i;
+
+       for (i = 0; i < mp->n_regions; ++i)
+               printk(" [%x, %x)", mp->regions[i].address,
+                      mp->regions[i].address + mp->regions[i].size);
+       printk("\n");
+}
+
+void __init
+mem_pieces_sort(struct mem_pieces *mp)
+{
+       unsigned long a, s;
+       int i, j;
+
+       for (i = 1; i < mp->n_regions; ++i) {
+               a = mp->regions[i].address;
+               s = mp->regions[i].size;
+               for (j = i - 1; j >= 0; --j) {
+                       if (a >= mp->regions[j].address)
+                               break;
+                       mp->regions[j+1] = mp->regions[j];
+               }
+               mp->regions[j+1].address = a;
+               mp->regions[j+1].size = s;
+       }
+}
+
+void __init
+mem_pieces_coalesce(struct mem_pieces *mp)
+{
+       unsigned long a, s, ns;
+       int i, j, d;
+
+       d = 0;
+       for (i = 0; i < mp->n_regions; i = j) {
+               a = mp->regions[i].address;
+               s = mp->regions[i].size;
+               for (j = i + 1; j < mp->n_regions
+                            && mp->regions[j].address - a <= s; ++j) {
+                       ns = mp->regions[j].address + mp->regions[j].size - a;
+                       if (ns > s)
+                               s = ns;
+               }
+               mp->regions[d].address = a;
+               mp->regions[d].size = s;
+               ++d;
+       }
+       mp->n_regions = d;
+}
diff --git a/arch/powerpc/mm/mem_pieces.h b/arch/powerpc/mm/mem_pieces.h
new file mode 100644 (file)
index 0000000..e2b700d
--- /dev/null
@@ -0,0 +1,48 @@
+/*
+ *    Copyright (c) 1996 Paul Mackerras <paulus@cs.anu.edu.au>
+ *      Changes to accommodate Power Macintoshes.
+ *    Cort Dougan <cort@cs.nmt.edu>
+ *      Rewrites.
+ *    Grant Erickson <grant@lcse.umn.edu>
+ *      General rework and split from mm/init.c.
+ *
+ *    Module name: mem_pieces.h
+ *
+ *    Description:
+ *      Routines and data structures for manipulating and representing
+ *      phyiscal memory extents (i.e. address/length pairs).
+ *
+ */
+
+#ifndef __MEM_PIECES_H__
+#define        __MEM_PIECES_H__
+
+#include <asm/prom.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+
+/* Type Definitions */
+
+#define        MEM_PIECES_MAX  32
+
+struct mem_pieces {
+    int n_regions;
+    struct reg_property regions[MEM_PIECES_MAX];
+};
+
+/* Function Prototypes */
+
+extern void    *mem_pieces_find(unsigned int size, unsigned int align);
+extern void     mem_pieces_remove(struct mem_pieces *mp, unsigned int start,
+                                  unsigned int size, int must_exist);
+extern void     mem_pieces_coalesce(struct mem_pieces *mp);
+extern void     mem_pieces_sort(struct mem_pieces *mp);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* __MEM_PIECES_H__ */
diff --git a/arch/powerpc/mm/mmu_context.c b/arch/powerpc/mm/mmu_context.c
new file mode 100644 (file)
index 0000000..a8816e0
--- /dev/null
@@ -0,0 +1,86 @@
+/*
+ * This file contains the routines for handling the MMU on those
+ * PowerPC implementations where the MMU substantially follows the
+ * architecture specification.  This includes the 6xx, 7xx, 7xxx,
+ * 8260, and POWER3 implementations but excludes the 8xx and 4xx.
+ *  -- paulus
+ *
+ *  Derived from arch/ppc/mm/init.c:
+ *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
+ *
+ *  Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
+ *  and Cort Dougan (PReP) (cort@cs.nmt.edu)
+ *    Copyright (C) 1996 Paul Mackerras
+ *  Amiga/APUS changes by Jesper Skov (jskov@cygnus.co.uk).
+ *
+ *  Derived from "arch/i386/mm/init.c"
+ *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
+ *
+ *  This program is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU General Public License
+ *  as published by the Free Software Foundation; either version
+ *  2 of the License, or (at your option) any later version.
+ *
+ */
+
+#include <linux/config.h>
+#include <linux/mm.h>
+#include <linux/init.h>
+
+#include <asm/mmu_context.h>
+#include <asm/tlbflush.h>
+
+mm_context_t next_mmu_context;
+unsigned long context_map[LAST_CONTEXT / BITS_PER_LONG + 1];
+#ifdef FEW_CONTEXTS
+atomic_t nr_free_contexts;
+struct mm_struct *context_mm[LAST_CONTEXT+1];
+void steal_context(void);
+#endif /* FEW_CONTEXTS */
+
+/*
+ * Initialize the context management stuff.
+ */
+void __init
+mmu_context_init(void)
+{
+       /*
+        * Some processors have too few contexts to reserve one for
+        * init_mm, and require using context 0 for a normal task.
+        * Other processors reserve the use of context zero for the kernel.
+        * This code assumes FIRST_CONTEXT < 32.
+        */
+       context_map[0] = (1 << FIRST_CONTEXT) - 1;
+       next_mmu_context = FIRST_CONTEXT;
+#ifdef FEW_CONTEXTS
+       atomic_set(&nr_free_contexts, LAST_CONTEXT - FIRST_CONTEXT + 1);
+#endif /* FEW_CONTEXTS */
+}
+
+#ifdef FEW_CONTEXTS
+/*
+ * Steal a context from a task that has one at the moment.
+ * This is only used on 8xx and 4xx and we presently assume that
+ * they don't do SMP.  If they do then this will have to check
+ * whether the MM we steal is in use.
+ * We also assume that this is only used on systems that don't
+ * use an MMU hash table - this is true for 8xx and 4xx.
+ * This isn't an LRU system, it just frees up each context in
+ * turn (sort-of pseudo-random replacement :).  This would be the
+ * place to implement an LRU scheme if anyone was motivated to do it.
+ *  -- paulus
+ */
+void
+steal_context(void)
+{
+       struct mm_struct *mm;
+
+       /* free up context `next_mmu_context' */
+       /* if we shouldn't free context 0, don't... */
+       if (next_mmu_context < FIRST_CONTEXT)
+               next_mmu_context = FIRST_CONTEXT;
+       mm = context_mm[next_mmu_context];
+       flush_tlb_mm(mm);
+       destroy_context(mm);
+}
+#endif /* FEW_CONTEXTS */
diff --git a/arch/powerpc/mm/mmu_context64.c b/arch/powerpc/mm/mmu_context64.c
new file mode 100644 (file)
index 0000000..714a84d
--- /dev/null
@@ -0,0 +1,63 @@
+/*
+ *  MMU context allocation for 64-bit kernels.
+ *
+ *  Copyright (C) 2004 Anton Blanchard, IBM Corp. <anton@samba.org>
+ *
+ *  This program is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU General Public License
+ *  as published by the Free Software Foundation; either version
+ *  2 of the License, or (at your option) any later version.
+ *
+ */
+
+#include <linux/config.h>
+#include <linux/sched.h>
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/string.h>
+#include <linux/types.h>
+#include <linux/mm.h>
+#include <linux/spinlock.h>
+#include <linux/idr.h>
+
+#include <asm/mmu_context.h>
+
+static DEFINE_SPINLOCK(mmu_context_lock);
+static DEFINE_IDR(mmu_context_idr);
+
+int init_new_context(struct task_struct *tsk, struct mm_struct *mm)
+{
+       int index;
+       int err;
+
+again:
+       if (!idr_pre_get(&mmu_context_idr, GFP_KERNEL))
+               return -ENOMEM;
+
+       spin_lock(&mmu_context_lock);
+       err = idr_get_new_above(&mmu_context_idr, NULL, 1, &index);
+       spin_unlock(&mmu_context_lock);
+
+       if (err == -EAGAIN)
+               goto again;
+       else if (err)
+               return err;
+
+       if (index > MAX_CONTEXT) {
+               idr_remove(&mmu_context_idr, index);
+               return -ENOMEM;
+       }
+
+       mm->context.id = index;
+
+       return 0;
+}
+
+void destroy_context(struct mm_struct *mm)
+{
+       spin_lock(&mmu_context_lock);
+       idr_remove(&mmu_context_idr, mm->context.id);
+       spin_unlock(&mmu_context_lock);
+
+       mm->context.id = NO_CONTEXT;
+}
diff --git a/arch/powerpc/mm/mmu_decl.h b/arch/powerpc/mm/mmu_decl.h
new file mode 100644 (file)
index 0000000..540f329
--- /dev/null
@@ -0,0 +1,85 @@
+/*
+ * Declarations of procedures and variables shared between files
+ * in arch/ppc/mm/.
+ *
+ *  Derived from arch/ppc/mm/init.c:
+ *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
+ *
+ *  Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
+ *  and Cort Dougan (PReP) (cort@cs.nmt.edu)
+ *    Copyright (C) 1996 Paul Mackerras
+ *  Amiga/APUS changes by Jesper Skov (jskov@cygnus.co.uk).
+ *
+ *  Derived from "arch/i386/mm/init.c"
+ *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
+ *
+ *  This program is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU General Public License
+ *  as published by the Free Software Foundation; either version
+ *  2 of the License, or (at your option) any later version.
+ *
+ */
+#include <asm/tlbflush.h>
+#include <asm/mmu.h>
+
+extern void mapin_ram(void);
+extern int map_page(unsigned long va, phys_addr_t pa, int flags);
+extern void setbat(int index, unsigned long virt, unsigned long phys,
+                  unsigned int size, int flags);
+extern void reserve_phys_mem(unsigned long start, unsigned long size);
+extern void settlbcam(int index, unsigned long virt, phys_addr_t phys,
+                     unsigned int size, int flags, unsigned int pid);
+extern void invalidate_tlbcam_entry(int index);
+
+extern int __map_without_bats;
+extern unsigned long ioremap_base;
+extern unsigned long ioremap_bot;
+extern unsigned int rtas_data, rtas_size;
+
+extern unsigned long total_memory;
+extern unsigned long total_lowmem;
+extern int mem_init_done;
+
+extern PTE *Hash, *Hash_end;
+extern unsigned long Hash_size, Hash_mask;
+
+extern unsigned int num_tlbcam_entries;
+
+/* ...and now those things that may be slightly different between processor
+ * architectures.  -- Dan
+ */
+#if defined(CONFIG_8xx)
+#define flush_HPTE(X, va, pg)  _tlbie(va)
+#define MMU_init_hw()          do { } while(0)
+#define mmu_mapin_ram()                (0UL)
+
+#elif defined(CONFIG_4xx)
+#define flush_HPTE(X, va, pg)  _tlbie(va)
+extern void MMU_init_hw(void);
+extern unsigned long mmu_mapin_ram(void);
+
+#elif defined(CONFIG_FSL_BOOKE)
+#define flush_HPTE(X, va, pg)  _tlbie(va)
+extern void MMU_init_hw(void);
+extern unsigned long mmu_mapin_ram(void);
+extern void adjust_total_lowmem(void);
+
+#else
+/* anything except 4xx or 8xx */
+extern void MMU_init_hw(void);
+extern unsigned long mmu_mapin_ram(void);
+
+/* Be careful....this needs to be updated if we ever encounter 603 SMPs,
+ * which includes all new 82xx processors.  We need tlbie/tlbsync here
+ * in that case (I think). -- Dan.
+ */
+static inline void flush_HPTE(unsigned context, unsigned long va,
+                             unsigned long pdval)
+{
+       if ((Hash != 0) &&
+           cpu_has_feature(CPU_FTR_HPTE_TABLE))
+               flush_hash_pages(0, va, pdval, 1);
+       else
+               _tlbie(va);
+}
+#endif
diff --git a/arch/powerpc/mm/pgtable.c b/arch/powerpc/mm/pgtable.c
new file mode 100644 (file)
index 0000000..81a3d74
--- /dev/null
@@ -0,0 +1,470 @@
+/*
+ * This file contains the routines setting up the linux page tables.
+ *  -- paulus
+ *
+ *  Derived from arch/ppc/mm/init.c:
+ *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
+ *
+ *  Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
+ *  and Cort Dougan (PReP) (cort@cs.nmt.edu)
+ *    Copyright (C) 1996 Paul Mackerras
+ *  Amiga/APUS changes by Jesper Skov (jskov@cygnus.co.uk).
+ *
+ *  Derived from "arch/i386/mm/init.c"
+ *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
+ *
+ *  This program is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU General Public License
+ *  as published by the Free Software Foundation; either version
+ *  2 of the License, or (at your option) any later version.
+ *
+ */
+
+#include <linux/config.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/types.h>
+#include <linux/mm.h>
+#include <linux/vmalloc.h>
+#include <linux/init.h>
+#include <linux/highmem.h>
+
+#include <asm/pgtable.h>
+#include <asm/pgalloc.h>
+#include <asm/io.h>
+
+#include "mmu_decl.h"
+
+unsigned long ioremap_base;
+unsigned long ioremap_bot;
+int io_bat_index;
+
+#if defined(CONFIG_6xx) || defined(CONFIG_POWER3)
+#define HAVE_BATS      1
+#endif
+
+#if defined(CONFIG_FSL_BOOKE)
+#define HAVE_TLBCAM    1
+#endif
+
+extern char etext[], _stext[];
+
+#ifdef CONFIG_SMP
+extern void hash_page_sync(void);
+#endif
+
+#ifdef HAVE_BATS
+extern unsigned long v_mapped_by_bats(unsigned long va);
+extern unsigned long p_mapped_by_bats(unsigned long pa);
+void setbat(int index, unsigned long virt, unsigned long phys,
+           unsigned int size, int flags);
+
+#else /* !HAVE_BATS */
+#define v_mapped_by_bats(x)    (0UL)
+#define p_mapped_by_bats(x)    (0UL)
+#endif /* HAVE_BATS */
+
+#ifdef HAVE_TLBCAM
+extern unsigned int tlbcam_index;
+extern unsigned long v_mapped_by_tlbcam(unsigned long va);
+extern unsigned long p_mapped_by_tlbcam(unsigned long pa);
+#else /* !HAVE_TLBCAM */
+#define v_mapped_by_tlbcam(x)  (0UL)
+#define p_mapped_by_tlbcam(x)  (0UL)
+#endif /* HAVE_TLBCAM */
+
+#ifdef CONFIG_PTE_64BIT
+/* 44x uses an 8kB pgdir because it has 8-byte Linux PTEs. */
+#define PGDIR_ORDER    1
+#else
+#define PGDIR_ORDER    0
+#endif
+
+pgd_t *pgd_alloc(struct mm_struct *mm)
+{
+       pgd_t *ret;
+
+       ret = (pgd_t *)__get_free_pages(GFP_KERNEL|__GFP_ZERO, PGDIR_ORDER);
+       return ret;
+}
+
+void pgd_free(pgd_t *pgd)
+{
+       free_pages((unsigned long)pgd, PGDIR_ORDER);
+}
+
+pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
+{
+       pte_t *pte;
+       extern int mem_init_done;
+       extern void *early_get_page(void);
+
+       if (mem_init_done) {
+               pte = (pte_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO);
+       } else {
+               pte = (pte_t *)early_get_page();
+               if (pte)
+                       clear_page(pte);
+       }
+       return pte;
+}
+
+struct page *pte_alloc_one(struct mm_struct *mm, unsigned long address)
+{
+       struct page *ptepage;
+
+#ifdef CONFIG_HIGHPTE
+       int flags = GFP_KERNEL | __GFP_HIGHMEM | __GFP_REPEAT;
+#else
+       int flags = GFP_KERNEL | __GFP_REPEAT;
+#endif
+
+       ptepage = alloc_pages(flags, 0);
+       if (ptepage)
+               clear_highpage(ptepage);
+       return ptepage;
+}
+
+void pte_free_kernel(pte_t *pte)
+{
+#ifdef CONFIG_SMP
+       hash_page_sync();
+#endif
+       free_page((unsigned long)pte);
+}
+
+void pte_free(struct page *ptepage)
+{
+#ifdef CONFIG_SMP
+       hash_page_sync();
+#endif
+       __free_page(ptepage);
+}
+
+#ifndef CONFIG_PHYS_64BIT
+void __iomem *
+ioremap(phys_addr_t addr, unsigned long size)
+{
+       return __ioremap(addr, size, _PAGE_NO_CACHE);
+}
+#else /* CONFIG_PHYS_64BIT */
+void __iomem *
+ioremap64(unsigned long long addr, unsigned long size)
+{
+       return __ioremap(addr, size, _PAGE_NO_CACHE);
+}
+
+void __iomem *
+ioremap(phys_addr_t addr, unsigned long size)
+{
+       phys_addr_t addr64 = fixup_bigphys_addr(addr, size);
+
+       return ioremap64(addr64, size);
+}
+#endif /* CONFIG_PHYS_64BIT */
+
+void __iomem *
+__ioremap(phys_addr_t addr, unsigned long size, unsigned long flags)
+{
+       unsigned long v, i;
+       phys_addr_t p;
+       int err;
+
+       /*
+        * Choose an address to map it to.
+        * Once the vmalloc system is running, we use it.
+        * Before then, we use space going down from ioremap_base
+        * (ioremap_bot records where we're up to).
+        */
+       p = addr & PAGE_MASK;
+       size = PAGE_ALIGN(addr + size) - p;
+
+       /*
+        * If the address lies within the first 16 MB, assume it's in ISA
+        * memory space
+        */
+       if (p < 16*1024*1024)
+               p += _ISA_MEM_BASE;
+
+       /*
+        * Don't allow anybody to remap normal RAM that we're using.
+        * mem_init() sets high_memory so only do the check after that.
+        */
+       if ( mem_init_done && (p < virt_to_phys(high_memory)) )
+       {
+               printk("__ioremap(): phys addr "PHYS_FMT" is RAM lr %p\n", p,
+                      __builtin_return_address(0));
+               return NULL;
+       }
+
+       if (size == 0)
+               return NULL;
+
+       /*
+        * Is it already mapped?  Perhaps overlapped by a previous
+        * BAT mapping.  If the whole area is mapped then we're done,
+        * otherwise remap it since we want to keep the virt addrs for
+        * each request contiguous.
+        *
+        * We make the assumption here that if the bottom and top
+        * of the range we want are mapped then it's mapped to the
+        * same virt address (and this is contiguous).
+        *  -- Cort
+        */
+       if ((v = p_mapped_by_bats(p)) /*&& p_mapped_by_bats(p+size-1)*/ )
+               goto out;
+
+       if ((v = p_mapped_by_tlbcam(p)))
+               goto out;
+
+       if (mem_init_done) {
+               struct vm_struct *area;
+               area = get_vm_area(size, VM_IOREMAP);
+               if (area == 0)
+                       return NULL;
+               v = (unsigned long) area->addr;
+       } else {
+               v = (ioremap_bot -= size);
+       }
+
+       if ((flags & _PAGE_PRESENT) == 0)
+               flags |= _PAGE_KERNEL;
+       if (flags & _PAGE_NO_CACHE)
+               flags |= _PAGE_GUARDED;
+
+       /*
+        * Should check if it is a candidate for a BAT mapping
+        */
+
+       err = 0;
+       for (i = 0; i < size && err == 0; i += PAGE_SIZE)
+               err = map_page(v+i, p+i, flags);
+       if (err) {
+               if (mem_init_done)
+                       vunmap((void *)v);
+               return NULL;
+       }
+
+out:
+       return (void __iomem *) (v + ((unsigned long)addr & ~PAGE_MASK));
+}
+
+void iounmap(volatile void __iomem *addr)
+{
+       /*
+        * If mapped by BATs then there is nothing to do.
+        * Calling vfree() generates a benign warning.
+        */
+       if (v_mapped_by_bats((unsigned long)addr)) return;
+
+       if (addr > high_memory && (unsigned long) addr < ioremap_bot)
+               vunmap((void *) (PAGE_MASK & (unsigned long)addr));
+}
+
+void __iomem *ioport_map(unsigned long port, unsigned int len)
+{
+       return (void __iomem *) (port + _IO_BASE);
+}
+
+void ioport_unmap(void __iomem *addr)
+{
+       /* Nothing to do */
+}
+EXPORT_SYMBOL(ioport_map);
+EXPORT_SYMBOL(ioport_unmap);
+
+int
+map_page(unsigned long va, phys_addr_t pa, int flags)
+{
+       pmd_t *pd;
+       pte_t *pg;
+       int err = -ENOMEM;
+
+       spin_lock(&init_mm.page_table_lock);
+       /* Use upper 10 bits of VA to index the first level map */
+       pd = pmd_offset(pgd_offset_k(va), va);
+       /* Use middle 10 bits of VA to index the second-level map */
+       pg = pte_alloc_kernel(&init_mm, pd, va);
+       if (pg != 0) {
+               err = 0;
+               set_pte_at(&init_mm, va, pg, pfn_pte(pa >> PAGE_SHIFT, __pgprot(flags)));
+               if (mem_init_done)
+                       flush_HPTE(0, va, pmd_val(*pd));
+       }
+       spin_unlock(&init_mm.page_table_lock);
+       return err;
+}
+
+/*
+ * Map in all of physical memory starting at KERNELBASE.
+ */
+void __init mapin_ram(void)
+{
+       unsigned long v, p, s, f;
+
+       s = mmu_mapin_ram();
+       v = KERNELBASE + s;
+       p = PPC_MEMSTART + s;
+       for (; s < total_lowmem; s += PAGE_SIZE) {
+               if ((char *) v >= _stext && (char *) v < etext)
+                       f = _PAGE_RAM_TEXT;
+               else
+                       f = _PAGE_RAM;
+               map_page(v, p, f);
+               v += PAGE_SIZE;
+               p += PAGE_SIZE;
+       }
+}
+
+/* is x a power of 2? */
+#define is_power_of_2(x)       ((x) != 0 && (((x) & ((x) - 1)) == 0))
+
+/* is x a power of 4? */
+#define is_power_of_4(x)       ((x) != 0 && (((x) & (x-1)) == 0) && (ffs(x) & 1))
+
+/*
+ * Set up a mapping for a block of I/O.
+ * virt, phys, size must all be page-aligned.
+ * This should only be called before ioremap is called.
+ */
+void __init io_block_mapping(unsigned long virt, phys_addr_t phys,
+                            unsigned int size, int flags)
+{
+       int i;
+
+       if (virt > KERNELBASE && virt < ioremap_bot)
+               ioremap_bot = ioremap_base = virt;
+
+#ifdef HAVE_BATS
+       /*
+        * Use a BAT for this if possible...
+        */
+       if (io_bat_index < 2 && is_power_of_2(size)
+           && (virt & (size - 1)) == 0 && (phys & (size - 1)) == 0) {
+               setbat(io_bat_index, virt, phys, size, flags);
+               ++io_bat_index;
+               return;
+       }
+#endif /* HAVE_BATS */
+
+#ifdef HAVE_TLBCAM
+       /*
+        * Use a CAM for this if possible...
+        */
+       if (tlbcam_index < num_tlbcam_entries && is_power_of_4(size)
+           && (virt & (size - 1)) == 0 && (phys & (size - 1)) == 0) {
+               settlbcam(tlbcam_index, virt, phys, size, flags, 0);
+               ++tlbcam_index;
+               return;
+       }
+#endif /* HAVE_TLBCAM */
+
+       /* No BATs available, put it in the page tables. */
+       for (i = 0; i < size; i += PAGE_SIZE)
+               map_page(virt + i, phys + i, flags);
+}
+
+/* Scan the real Linux page tables and return a PTE pointer for
+ * a virtual address in a context.
+ * Returns true (1) if PTE was found, zero otherwise.  The pointer to
+ * the PTE pointer is unmodified if PTE is not found.
+ */
+int
+get_pteptr(struct mm_struct *mm, unsigned long addr, pte_t **ptep)
+{
+        pgd_t  *pgd;
+        pmd_t  *pmd;
+        pte_t  *pte;
+        int     retval = 0;
+
+        pgd = pgd_offset(mm, addr & PAGE_MASK);
+        if (pgd) {
+                pmd = pmd_offset(pgd, addr & PAGE_MASK);
+                if (pmd_present(*pmd)) {
+                        pte = pte_offset_map(pmd, addr & PAGE_MASK);
+                        if (pte) {
+                               retval = 1;
+                               *ptep = pte;
+                               /* XXX caller needs to do pte_unmap, yuck */
+                        }
+                }
+        }
+        return(retval);
+}
+
+/* Find physical address for this virtual address.  Normally used by
+ * I/O functions, but anyone can call it.
+ */
+unsigned long iopa(unsigned long addr)
+{
+       unsigned long pa;
+
+       /* I don't know why this won't work on PMacs or CHRP.  It
+        * appears there is some bug, or there is some implicit
+        * mapping done not properly represented by BATs or in page
+        * tables.......I am actively working on resolving this, but
+        * can't hold up other stuff.  -- Dan
+        */
+       pte_t *pte;
+       struct mm_struct *mm;
+
+       /* Check the BATs */
+       pa = v_mapped_by_bats(addr);
+       if (pa)
+               return pa;
+
+       /* Allow mapping of user addresses (within the thread)
+        * for DMA if necessary.
+        */
+       if (addr < TASK_SIZE)
+               mm = current->mm;
+       else
+               mm = &init_mm;
+
+       pa = 0;
+       if (get_pteptr(mm, addr, &pte)) {
+               pa = (pte_val(*pte) & PAGE_MASK) | (addr & ~PAGE_MASK);
+               pte_unmap(pte);
+       }
+
+       return(pa);
+}
+
+/* This is will find the virtual address for a physical one....
+ * Swiped from APUS, could be dangerous :-).
+ * This is only a placeholder until I really find a way to make this
+ * work.  -- Dan
+ */
+unsigned long
+mm_ptov (unsigned long paddr)
+{
+       unsigned long ret;
+#if 0
+       if (paddr < 16*1024*1024)
+               ret = ZTWO_VADDR(paddr);
+       else {
+               int i;
+
+               for (i = 0; i < kmap_chunk_count;){
+                       unsigned long phys = kmap_chunks[i++];
+                       unsigned long size = kmap_chunks[i++];
+                       unsigned long virt = kmap_chunks[i++];
+                       if (paddr >= phys
+                           && paddr < (phys + size)){
+                               ret = virt + paddr - phys;
+                               goto exit;
+                       }
+               }
+       
+               ret = (unsigned long) __va(paddr);
+       }
+exit:
+#ifdef DEBUGPV
+       printk ("PTOV(%lx)=%lx\n", paddr, ret);
+#endif
+#else
+       ret = (unsigned long)paddr + KERNELBASE;
+#endif
+       return ret;
+}
+
diff --git a/arch/powerpc/mm/pgtable64.c b/arch/powerpc/mm/pgtable64.c
new file mode 100644 (file)
index 0000000..724f97e
--- /dev/null
@@ -0,0 +1,357 @@
+/*
+ *  This file contains ioremap and related functions for 64-bit machines.
+ *
+ *  Derived from arch/ppc64/mm/init.c
+ *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
+ *
+ *  Modifications by Paul Mackerras (PowerMac) (paulus@samba.org)
+ *  and Cort Dougan (PReP) (cort@cs.nmt.edu)
+ *    Copyright (C) 1996 Paul Mackerras
+ *  Amiga/APUS changes by Jesper Skov (jskov@cygnus.co.uk).
+ *
+ *  Derived from "arch/i386/mm/init.c"
+ *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
+ *
+ *  Dave Engebretsen <engebret@us.ibm.com>
+ *      Rework for PPC64 port.
+ *
+ *  This program is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU General Public License
+ *  as published by the Free Software Foundation; either version
+ *  2 of the License, or (at your option) any later version.
+ *
+ */
+
+#include <linux/config.h>
+#include <linux/signal.h>
+#include <linux/sched.h>
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/string.h>
+#include <linux/types.h>
+#include <linux/mman.h>
+#include <linux/mm.h>
+#include <linux/swap.h>
+#include <linux/stddef.h>
+#include <linux/vmalloc.h>
+#include <linux/init.h>
+#include <linux/delay.h>
+#include <linux/bootmem.h>
+#include <linux/highmem.h>
+#include <linux/idr.h>
+#include <linux/nodemask.h>
+#include <linux/module.h>
+
+#include <asm/pgalloc.h>
+#include <asm/page.h>
+#include <asm/prom.h>
+#include <asm/lmb.h>
+#include <asm/rtas.h>
+#include <asm/io.h>
+#include <asm/mmu_context.h>
+#include <asm/pgtable.h>
+#include <asm/mmu.h>
+#include <asm/uaccess.h>
+#include <asm/smp.h>
+#include <asm/machdep.h>
+#include <asm/tlb.h>
+#include <asm/eeh.h>
+#include <asm/processor.h>
+#include <asm/mmzone.h>
+#include <asm/cputable.h>
+#include <asm/ppcdebug.h>
+#include <asm/sections.h>
+#include <asm/system.h>
+#include <asm/iommu.h>
+#include <asm/abs_addr.h>
+#include <asm/vdso.h>
+#include <asm/imalloc.h>
+
+#if PGTABLE_RANGE > USER_VSID_RANGE
+#warning Limited user VSID range means pagetable space is wasted
+#endif
+
+#if (TASK_SIZE_USER64 < PGTABLE_RANGE) && (TASK_SIZE_USER64 < USER_VSID_RANGE)
+#warning TASK_SIZE is smaller than it needs to be.
+#endif
+
+int mem_init_done;
+unsigned long ioremap_bot = IMALLOC_BASE;
+static unsigned long phbs_io_bot = PHBS_IO_BASE;
+
+extern pgd_t swapper_pg_dir[];
+extern struct task_struct *current_set[NR_CPUS];
+
+unsigned long klimit = (unsigned long)_end;
+
+/* max amount of RAM to use */
+unsigned long __max_memory;
+
+/* info on what we think the IO hole is */
+unsigned long  io_hole_start;
+unsigned long  io_hole_size;
+
+#ifdef CONFIG_PPC_ISERIES
+
+void __iomem *ioremap(unsigned long addr, unsigned long size)
+{
+       return (void __iomem *)addr;
+}
+
+extern void __iomem *__ioremap(unsigned long addr, unsigned long size,
+                      unsigned long flags)
+{
+       return (void __iomem *)addr;
+}
+
+void iounmap(volatile void __iomem *addr)
+{
+       return;
+}
+
+#else
+
+/*
+ * map_io_page currently only called by __ioremap
+ * map_io_page adds an entry to the ioremap page table
+ * and adds an entry to the HPT, possibly bolting it
+ */
+static int map_io_page(unsigned long ea, unsigned long pa, int flags)
+{
+       pgd_t *pgdp;
+       pud_t *pudp;
+       pmd_t *pmdp;
+       pte_t *ptep;
+       unsigned long vsid;
+
+       if (mem_init_done) {
+               spin_lock(&init_mm.page_table_lock);
+               pgdp = pgd_offset_k(ea);
+               pudp = pud_alloc(&init_mm, pgdp, ea);
+               if (!pudp)
+                       return -ENOMEM;
+               pmdp = pmd_alloc(&init_mm, pudp, ea);
+               if (!pmdp)
+                       return -ENOMEM;
+               ptep = pte_alloc_kernel(&init_mm, pmdp, ea);
+               if (!ptep)
+                       return -ENOMEM;
+               set_pte_at(&init_mm, ea, ptep, pfn_pte(pa >> PAGE_SHIFT,
+                                                         __pgprot(flags)));
+               spin_unlock(&init_mm.page_table_lock);
+       } else {
+               unsigned long va, vpn, hash, hpteg;
+
+               /*
+                * If the mm subsystem is not fully up, we cannot create a
+                * linux page table entry for this mapping.  Simply bolt an
+                * entry in the hardware page table.
+                */
+               vsid = get_kernel_vsid(ea);
+               va = (vsid << 28) | (ea & 0xFFFFFFF);
+               vpn = va >> PAGE_SHIFT;
+
+               hash = hpt_hash(vpn, 0);
+
+               hpteg = ((hash & htab_hash_mask) * HPTES_PER_GROUP);
+
+               /* Panic if a pte grpup is full */
+               if (ppc_md.hpte_insert(hpteg, va, pa >> PAGE_SHIFT,
+                                      HPTE_V_BOLTED,
+                                      _PAGE_NO_CACHE|_PAGE_GUARDED|PP_RWXX)
+                   == -1) {
+                       panic("map_io_page: could not insert mapping");
+               }
+       }
+       return 0;
+}
+
+
+static void __iomem * __ioremap_com(unsigned long addr, unsigned long pa,
+                           unsigned long ea, unsigned long size,
+                           unsigned long flags)
+{
+       unsigned long i;
+
+       if ((flags & _PAGE_PRESENT) == 0)
+               flags |= pgprot_val(PAGE_KERNEL);
+
+       for (i = 0; i < size; i += PAGE_SIZE)
+               if (map_io_page(ea+i, pa+i, flags))
+                       return NULL;
+
+       return (void __iomem *) (ea + (addr & ~PAGE_MASK));
+}
+
+
+void __iomem *
+ioremap(unsigned long addr, unsigned long size)
+{
+       return __ioremap(addr, size, _PAGE_NO_CACHE | _PAGE_GUARDED);
+}
+
+void __iomem * __ioremap(unsigned long addr, unsigned long size,
+                        unsigned long flags)
+{
+       unsigned long pa, ea;
+       void __iomem *ret;
+
+       /*
+        * Choose an address to map it to.
+        * Once the imalloc system is running, we use it.
+        * Before that, we map using addresses going
+        * up from ioremap_bot.  imalloc will use
+        * the addresses from ioremap_bot through
+        * IMALLOC_END
+        * 
+        */
+       pa = addr & PAGE_MASK;
+       size = PAGE_ALIGN(addr + size) - pa;
+
+       if (size == 0)
+               return NULL;
+
+       if (mem_init_done) {
+               struct vm_struct *area;
+               area = im_get_free_area(size);
+               if (area == NULL)
+                       return NULL;
+               ea = (unsigned long)(area->addr);
+               ret = __ioremap_com(addr, pa, ea, size, flags);
+               if (!ret)
+                       im_free(area->addr);
+       } else {
+               ea = ioremap_bot;
+               ret = __ioremap_com(addr, pa, ea, size, flags);
+               if (ret)
+                       ioremap_bot += size;
+       }
+       return ret;
+}
+
+#define IS_PAGE_ALIGNED(_val) ((_val) == ((_val) & PAGE_MASK))
+
+int __ioremap_explicit(unsigned long pa, unsigned long ea,
+                      unsigned long size, unsigned long flags)
+{
+       struct vm_struct *area;
+       void __iomem *ret;
+       
+       /* For now, require page-aligned values for pa, ea, and size */
+       if (!IS_PAGE_ALIGNED(pa) || !IS_PAGE_ALIGNED(ea) ||
+           !IS_PAGE_ALIGNED(size)) {
+               printk(KERN_ERR "unaligned value in %s\n", __FUNCTION__);
+               return 1;
+       }
+       
+       if (!mem_init_done) {
+               /* Two things to consider in this case:
+                * 1) No records will be kept (imalloc, etc) that the region
+                *    has been remapped
+                * 2) It won't be easy to iounmap() the region later (because
+                *    of 1)
+                */
+               ;
+       } else {
+               area = im_get_area(ea, size,
+                       IM_REGION_UNUSED|IM_REGION_SUBSET|IM_REGION_EXISTS);
+               if (area == NULL) {
+                       /* Expected when PHB-dlpar is in play */
+                       return 1;
+               }
+               if (ea != (unsigned long) area->addr) {
+                       printk(KERN_ERR "unexpected addr return from "
+                              "im_get_area\n");
+                       return 1;
+               }
+       }
+       
+       ret = __ioremap_com(pa, pa, ea, size, flags);
+       if (ret == NULL) {
+               printk(KERN_ERR "ioremap_explicit() allocation failure !\n");
+               return 1;
+       }
+       if (ret != (void *) ea) {
+               printk(KERN_ERR "__ioremap_com() returned unexpected addr\n");
+               return 1;
+       }
+
+       return 0;
+}
+
+/*  
+ * Unmap an IO region and remove it from imalloc'd list.
+ * Access to IO memory should be serialized by driver.
+ * This code is modeled after vmalloc code - unmap_vm_area()
+ *
+ * XXX what about calls before mem_init_done (ie python_countermeasures())
+ */
+void iounmap(volatile void __iomem *token)
+{
+       void *addr;
+
+       if (!mem_init_done)
+               return;
+       
+       addr = (void *) ((unsigned long __force) token & PAGE_MASK);
+
+       im_free(addr);
+}
+
+static int iounmap_subset_regions(unsigned long addr, unsigned long size)
+{
+       struct vm_struct *area;
+
+       /* Check whether subsets of this region exist */
+       area = im_get_area(addr, size, IM_REGION_SUPERSET);
+       if (area == NULL)
+               return 1;
+
+       while (area) {
+               iounmap((void __iomem *) area->addr);
+               area = im_get_area(addr, size,
+                               IM_REGION_SUPERSET);
+       }
+
+       return 0;
+}
+
+int iounmap_explicit(volatile void __iomem *start, unsigned long size)
+{
+       struct vm_struct *area;
+       unsigned long addr;
+       int rc;
+       
+       addr = (unsigned long __force) start & PAGE_MASK;
+
+       /* Verify that the region either exists or is a subset of an existing
+        * region.  In the latter case, split the parent region to create 
+        * the exact region 
+        */
+       area = im_get_area(addr, size, 
+                           IM_REGION_EXISTS | IM_REGION_SUBSET);
+       if (area == NULL) {
+               /* Determine whether subset regions exist.  If so, unmap */
+               rc = iounmap_subset_regions(addr, size);
+               if (rc) {
+                       printk(KERN_ERR
+                              "%s() cannot unmap nonexistent range 0x%lx\n",
+                               __FUNCTION__, addr);
+                       return 1;
+               }
+       } else {
+               iounmap((void __iomem *) area->addr);
+       }
+       /*
+        * FIXME! This can't be right:
+       iounmap(area->addr);
+        * Maybe it should be "iounmap(area);"
+        */
+       return 0;
+}
+
+#endif
+
+EXPORT_SYMBOL(ioremap);
+EXPORT_SYMBOL(__ioremap);
+EXPORT_SYMBOL(iounmap);
diff --git a/arch/powerpc/mm/ppc_mmu.c b/arch/powerpc/mm/ppc_mmu.c
new file mode 100644 (file)
index 0000000..9a381ed
--- /dev/null
@@ -0,0 +1,296 @@
+/*
+ * This file contains the routines for handling the MMU on those
+ * PowerPC implementations where the MMU substantially follows the
+ * architecture specification.  This includes the 6xx, 7xx, 7xxx,
+ * 8260, and POWER3 implementations but excludes the 8xx and 4xx.
+ *  -- paulus
+ *
+ *  Derived from arch/ppc/mm/init.c:
+ *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
+ *
+ *  Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
+ *  and Cort Dougan (PReP) (cort@cs.nmt.edu)
+ *    Copyright (C) 1996 Paul Mackerras
+ *  Amiga/APUS changes by Jesper Skov (jskov@cygnus.co.uk).
+ *
+ *  Derived from "arch/i386/mm/init.c"
+ *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
+ *
+ *  This program is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU General Public License
+ *  as published by the Free Software Foundation; either version
+ *  2 of the License, or (at your option) any later version.
+ *
+ */
+
+#include <linux/config.h>
+#include <linux/kernel.h>
+#include <linux/mm.h>
+#include <linux/init.h>
+#include <linux/highmem.h>
+
+#include <asm/prom.h>
+#include <asm/mmu.h>
+#include <asm/machdep.h>
+
+#include "mmu_decl.h"
+#include "mem_pieces.h"
+
+PTE *Hash, *Hash_end;
+unsigned long Hash_size, Hash_mask;
+unsigned long _SDR1;
+
+union ubat {                   /* BAT register values to be loaded */
+       BAT     bat;
+#ifdef CONFIG_PPC64BRIDGE
+       u64     word[2];
+#else
+       u32     word[2];
+#endif
+} BATS[4][2];                  /* 4 pairs of IBAT, DBAT */
+
+struct batrange {              /* stores address ranges mapped by BATs */
+       unsigned long start;
+       unsigned long limit;
+       unsigned long phys;
+} bat_addrs[4];
+
+/*
+ * Return PA for this VA if it is mapped by a BAT, or 0
+ */
+unsigned long v_mapped_by_bats(unsigned long va)
+{
+       int b;
+       for (b = 0; b < 4; ++b)
+               if (va >= bat_addrs[b].start && va < bat_addrs[b].limit)
+                       return bat_addrs[b].phys + (va - bat_addrs[b].start);
+       return 0;
+}
+
+/*
+ * Return VA for a given PA or 0 if not mapped
+ */
+unsigned long p_mapped_by_bats(unsigned long pa)
+{
+       int b;
+       for (b = 0; b < 4; ++b)
+               if (pa >= bat_addrs[b].phys
+                   && pa < (bat_addrs[b].limit-bat_addrs[b].start)
+                             +bat_addrs[b].phys)
+                       return bat_addrs[b].start+(pa-bat_addrs[b].phys);
+       return 0;
+}
+
+unsigned long __init mmu_mapin_ram(void)
+{
+#ifdef CONFIG_POWER4
+       return 0;
+#else
+       unsigned long tot, bl, done;
+       unsigned long max_size = (256<<20);
+       unsigned long align;
+
+       if (__map_without_bats)
+               return 0;
+
+       /* Set up BAT2 and if necessary BAT3 to cover RAM. */
+
+       /* Make sure we don't map a block larger than the
+          smallest alignment of the physical address. */
+       /* alignment of PPC_MEMSTART */
+       align = ~(PPC_MEMSTART-1) & PPC_MEMSTART;
+       /* set BAT block size to MIN(max_size, align) */
+       if (align && align < max_size)
+               max_size = align;
+
+       tot = total_lowmem;
+       for (bl = 128<<10; bl < max_size; bl <<= 1) {
+               if (bl * 2 > tot)
+                       break;
+       }
+
+       setbat(2, KERNELBASE, PPC_MEMSTART, bl, _PAGE_RAM);
+       done = (unsigned long)bat_addrs[2].limit - KERNELBASE + 1;
+       if ((done < tot) && !bat_addrs[3].limit) {
+               /* use BAT3 to cover a bit more */
+               tot -= done;
+               for (bl = 128<<10; bl < max_size; bl <<= 1)
+                       if (bl * 2 > tot)
+                               break;
+               setbat(3, KERNELBASE+done, PPC_MEMSTART+done, bl, _PAGE_RAM);
+               done = (unsigned long)bat_addrs[3].limit - KERNELBASE + 1;
+       }
+
+       return done;
+#endif
+}
+
+/*
+ * Set up one of the I/D BAT (block address translation) register pairs.
+ * The parameters are not checked; in particular size must be a power
+ * of 2 between 128k and 256M.
+ */
+void __init setbat(int index, unsigned long virt, unsigned long phys,
+                  unsigned int size, int flags)
+{
+       unsigned int bl;
+       int wimgxpp;
+       union ubat *bat = BATS[index];
+
+       if (((flags & _PAGE_NO_CACHE) == 0) &&
+           cpu_has_feature(CPU_FTR_NEED_COHERENT))
+               flags |= _PAGE_COHERENT;
+
+       bl = (size >> 17) - 1;
+       if (PVR_VER(mfspr(SPRN_PVR)) != 1) {
+               /* 603, 604, etc. */
+               /* Do DBAT first */
+               wimgxpp = flags & (_PAGE_WRITETHRU | _PAGE_NO_CACHE
+                                  | _PAGE_COHERENT | _PAGE_GUARDED);
+               wimgxpp |= (flags & _PAGE_RW)? BPP_RW: BPP_RX;
+               bat[1].word[0] = virt | (bl << 2) | 2; /* Vs=1, Vp=0 */
+               bat[1].word[1] = phys | wimgxpp;
+#ifndef CONFIG_KGDB /* want user access for breakpoints */
+               if (flags & _PAGE_USER)
+#endif
+                       bat[1].bat.batu.vp = 1;
+               if (flags & _PAGE_GUARDED) {
+                       /* G bit must be zero in IBATs */
+                       bat[0].word[0] = bat[0].word[1] = 0;
+               } else {
+                       /* make IBAT same as DBAT */
+                       bat[0] = bat[1];
+               }
+       } else {
+               /* 601 cpu */
+               if (bl > BL_8M)
+                       bl = BL_8M;
+               wimgxpp = flags & (_PAGE_WRITETHRU | _PAGE_NO_CACHE
+                                  | _PAGE_COHERENT);
+               wimgxpp |= (flags & _PAGE_RW)?
+                       ((flags & _PAGE_USER)? PP_RWRW: PP_RWXX): PP_RXRX;
+               bat->word[0] = virt | wimgxpp | 4;      /* Ks=0, Ku=1 */
+               bat->word[1] = phys | bl | 0x40;        /* V=1 */
+       }
+
+       bat_addrs[index].start = virt;
+       bat_addrs[index].limit = virt + ((bl + 1) << 17) - 1;
+       bat_addrs[index].phys = phys;
+}
+
+/*
+ * Initialize the hash table and patch the instructions in hashtable.S.
+ */
+void __init MMU_init_hw(void)
+{
+       unsigned int hmask, mb, mb2;
+       unsigned int n_hpteg, lg_n_hpteg;
+
+       extern unsigned int hash_page_patch_A[];
+       extern unsigned int hash_page_patch_B[], hash_page_patch_C[];
+       extern unsigned int hash_page[];
+       extern unsigned int flush_hash_patch_A[], flush_hash_patch_B[];
+
+       if (!cpu_has_feature(CPU_FTR_HPTE_TABLE)) {
+               /*
+                * Put a blr (procedure return) instruction at the
+                * start of hash_page, since we can still get DSI
+                * exceptions on a 603.
+                */
+               hash_page[0] = 0x4e800020;
+               flush_icache_range((unsigned long) &hash_page[0],
+                                  (unsigned long) &hash_page[1]);
+               return;
+       }
+
+       if ( ppc_md.progress ) ppc_md.progress("hash:enter", 0x105);
+
+#ifdef CONFIG_PPC64BRIDGE
+#define LG_HPTEG_SIZE  7               /* 128 bytes per HPTEG */
+#define SDR1_LOW_BITS  (lg_n_hpteg - 11)
+#define MIN_N_HPTEG    2048            /* min 256kB hash table */
+#else
+#define LG_HPTEG_SIZE  6               /* 64 bytes per HPTEG */
+#define SDR1_LOW_BITS  ((n_hpteg - 1) >> 10)
+#define MIN_N_HPTEG    1024            /* min 64kB hash table */
+#endif
+
+#ifdef CONFIG_POWER4
+       /* The hash table has already been allocated and initialized
+          in prom.c */
+       n_hpteg = Hash_size >> LG_HPTEG_SIZE;
+       lg_n_hpteg = __ilog2(n_hpteg);
+
+       /* Remove the hash table from the available memory */
+       if (Hash)
+               reserve_phys_mem(__pa(Hash), Hash_size);
+
+#else /* CONFIG_POWER4 */
+       /*
+        * Allow 1 HPTE (1/8 HPTEG) for each page of memory.
+        * This is less than the recommended amount, but then
+        * Linux ain't AIX.
+        */
+       n_hpteg = total_memory / (PAGE_SIZE * 8);
+       if (n_hpteg < MIN_N_HPTEG)
+               n_hpteg = MIN_N_HPTEG;
+       lg_n_hpteg = __ilog2(n_hpteg);
+       if (n_hpteg & (n_hpteg - 1)) {
+               ++lg_n_hpteg;           /* round up if not power of 2 */
+               n_hpteg = 1 << lg_n_hpteg;
+       }
+       Hash_size = n_hpteg << LG_HPTEG_SIZE;
+
+       /*
+        * Find some memory for the hash table.
+        */
+       if ( ppc_md.progress ) ppc_md.progress("hash:find piece", 0x322);
+       Hash = mem_pieces_find(Hash_size, Hash_size);
+       cacheable_memzero(Hash, Hash_size);
+       _SDR1 = __pa(Hash) | SDR1_LOW_BITS;
+#endif /* CONFIG_POWER4 */
+
+       Hash_end = (PTE *) ((unsigned long)Hash + Hash_size);
+
+       printk("Total memory = %ldMB; using %ldkB for hash table (at %p)\n",
+              total_memory >> 20, Hash_size >> 10, Hash);
+
+
+       /*
+        * Patch up the instructions in hashtable.S:create_hpte
+        */
+       if ( ppc_md.progress ) ppc_md.progress("hash:patch", 0x345);
+       Hash_mask = n_hpteg - 1;
+       hmask = Hash_mask >> (16 - LG_HPTEG_SIZE);
+       mb2 = mb = 32 - LG_HPTEG_SIZE - lg_n_hpteg;
+       if (lg_n_hpteg > 16)
+               mb2 = 16 - LG_HPTEG_SIZE;
+
+       hash_page_patch_A[0] = (hash_page_patch_A[0] & ~0xffff)
+               | ((unsigned int)(Hash) >> 16);
+       hash_page_patch_A[1] = (hash_page_patch_A[1] & ~0x7c0) | (mb << 6);
+       hash_page_patch_A[2] = (hash_page_patch_A[2] & ~0x7c0) | (mb2 << 6);
+       hash_page_patch_B[0] = (hash_page_patch_B[0] & ~0xffff) | hmask;
+       hash_page_patch_C[0] = (hash_page_patch_C[0] & ~0xffff) | hmask;
+
+       /*
+        * Ensure that the locations we've patched have been written
+        * out from the data cache and invalidated in the instruction
+        * cache, on those machines with split caches.
+        */
+       flush_icache_range((unsigned long) &hash_page_patch_A[0],
+                          (unsigned long) &hash_page_patch_C[1]);
+
+       /*
+        * Patch up the instructions in hashtable.S:flush_hash_page
+        */
+       flush_hash_patch_A[0] = (flush_hash_patch_A[0] & ~0xffff)
+               | ((unsigned int)(Hash) >> 16);
+       flush_hash_patch_A[1] = (flush_hash_patch_A[1] & ~0x7c0) | (mb << 6);
+       flush_hash_patch_A[2] = (flush_hash_patch_A[2] & ~0x7c0) | (mb2 << 6);
+       flush_hash_patch_B[0] = (flush_hash_patch_B[0] & ~0xffff) | hmask;
+       flush_icache_range((unsigned long) &flush_hash_patch_A[0],
+                          (unsigned long) &flush_hash_patch_B[1]);
+
+       if ( ppc_md.progress ) ppc_md.progress("hash:done", 0x205);
+}
diff --git a/arch/powerpc/mm/tlb.c b/arch/powerpc/mm/tlb.c
new file mode 100644 (file)
index 0000000..6c3dc3c
--- /dev/null
@@ -0,0 +1,183 @@
+/*
+ * This file contains the routines for TLB flushing.
+ * On machines where the MMU uses a hash table to store virtual to
+ * physical translations, these routines flush entries from the
+ * hash table also.
+ *  -- paulus
+ *
+ *  Derived from arch/ppc/mm/init.c:
+ *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
+ *
+ *  Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
+ *  and Cort Dougan (PReP) (cort@cs.nmt.edu)
+ *    Copyright (C) 1996 Paul Mackerras
+ *  Amiga/APUS changes by Jesper Skov (jskov@cygnus.co.uk).
+ *
+ *  Derived from "arch/i386/mm/init.c"
+ *    Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
+ *
+ *  This program is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU General Public License
+ *  as published by the Free Software Foundation; either version
+ *  2 of the License, or (at your option) any later version.
+ *
+ */
+
+#include <linux/config.h>
+#include <linux/kernel.h>
+#include <linux/mm.h>
+#include <linux/init.h>
+#include <linux/highmem.h>
+#include <asm/tlbflush.h>
+#include <asm/tlb.h>
+
+#include "mmu_decl.h"
+
+/*
+ * Called when unmapping pages to flush entries from the TLB/hash table.
+ */
+void flush_hash_entry(struct mm_struct *mm, pte_t *ptep, unsigned long addr)
+{
+       unsigned long ptephys;
+
+       if (Hash != 0) {
+               ptephys = __pa(ptep) & PAGE_MASK;
+               flush_hash_pages(mm->context, addr, ptephys, 1);
+       }
+}
+
+/*
+ * Called by ptep_set_access_flags, must flush on CPUs for which the
+ * DSI handler can't just "fixup" the TLB on a write fault
+ */
+void flush_tlb_page_nohash(struct vm_area_struct *vma, unsigned long addr)
+{
+       if (Hash != 0)
+               return;
+       _tlbie(addr);
+}
+
+/*
+ * Called at the end of a mmu_gather operation to make sure the
+ * TLB flush is completely done.
+ */
+void tlb_flush(struct mmu_gather *tlb)
+{
+       if (Hash == 0) {
+               /*
+                * 603 needs to flush the whole TLB here since
+                * it doesn't use a hash table.
+                */
+               _tlbia();
+       }
+}
+
+/*
+ * TLB flushing:
+ *
+ *  - flush_tlb_mm(mm) flushes the specified mm context TLB's
+ *  - flush_tlb_page(vma, vmaddr) flushes one page
+ *  - flush_tlb_range(vma, start, end) flushes a range of pages
+ *  - flush_tlb_kernel_range(start, end) flushes kernel pages
+ *
+ * since the hardware hash table functions as an extension of the
+ * tlb as far as the linux tables are concerned, flush it too.
+ *    -- Cort
+ */
+
+/*
+ * 750 SMP is a Bad Idea because the 750 doesn't broadcast all
+ * the cache operations on the bus.  Hence we need to use an IPI
+ * to get the other CPU(s) to invalidate their TLBs.
+ */
+#ifdef CONFIG_SMP_750
+#define FINISH_FLUSH   smp_send_tlb_invalidate(0)
+#else
+#define FINISH_FLUSH   do { } while (0)
+#endif
+
+static void flush_range(struct mm_struct *mm, unsigned long start,
+                       unsigned long end)
+{
+       pmd_t *pmd;
+       unsigned long pmd_end;
+       int count;
+       unsigned int ctx = mm->context;
+
+       if (Hash == 0) {
+               _tlbia();
+               return;
+       }
+       start &= PAGE_MASK;
+       if (start >= end)
+               return;
+       end = (end - 1) | ~PAGE_MASK;
+       pmd = pmd_offset(pgd_offset(mm, start), start);
+       for (;;) {
+               pmd_end = ((start + PGDIR_SIZE) & PGDIR_MASK) - 1;
+               if (pmd_end > end)
+                       pmd_end = end;
+               if (!pmd_none(*pmd)) {
+                       count = ((pmd_end - start) >> PAGE_SHIFT) + 1;
+                       flush_hash_pages(ctx, start, pmd_val(*pmd), count);
+               }
+               if (pmd_end == end)
+                       break;
+               start = pmd_end + 1;
+               ++pmd;
+       }
+}
+
+/*
+ * Flush kernel TLB entries in the given range
+ */
+void flush_tlb_kernel_range(unsigned long start, unsigned long end)
+{
+       flush_range(&init_mm, start, end);
+       FINISH_FLUSH;
+}
+
+/*
+ * Flush all the (user) entries for the address space described by mm.
+ */
+void flush_tlb_mm(struct mm_struct *mm)
+{
+       struct vm_area_struct *mp;
+
+       if (Hash == 0) {
+               _tlbia();
+               return;
+       }
+
+       for (mp = mm->mmap; mp != NULL; mp = mp->vm_next)
+               flush_range(mp->vm_mm, mp->vm_start, mp->vm_end);
+       FINISH_FLUSH;
+}
+
+void flush_tlb_page(struct vm_area_struct *vma, unsigned long vmaddr)
+{
+       struct mm_struct *mm;
+       pmd_t *pmd;
+
+       if (Hash == 0) {
+               _tlbie(vmaddr);
+               return;
+       }
+       mm = (vmaddr < TASK_SIZE)? vma->vm_mm: &init_mm;
+       pmd = pmd_offset(pgd_offset(mm, vmaddr), vmaddr);
+       if (!pmd_none(*pmd))
+               flush_hash_pages(mm->context, vmaddr, pmd_val(*pmd), 1);
+       FINISH_FLUSH;
+}
+
+/*
+ * For each address in the range, find the pte for the address
+ * and check _PAGE_HASHPTE bit; if it is set, find and destroy
+ * the corresponding HPTE.
+ */
+void flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
+                    unsigned long end)
+{
+       flush_range(vma->vm_mm, start, end);
+       FINISH_FLUSH;
+}
diff --git a/arch/powerpc/platforms/4xx/Kconfig b/arch/powerpc/platforms/4xx/Kconfig
new file mode 100644 (file)
index 0000000..ed39d6a
--- /dev/null
@@ -0,0 +1,280 @@
+config 4xx
+       bool
+       depends on 40x || 44x
+       default y
+
+config WANT_EARLY_SERIAL
+       bool
+       select SERIAL_8250
+       default n
+
+menu "AMCC 4xx options"
+       depends on 4xx
+
+choice
+       prompt "Machine Type"
+       depends on 40x
+       default WALNUT
+
+config BUBINGA
+       bool "Bubinga"
+       select WANT_EARLY_SERIAL
+       help
+         This option enables support for the IBM 405EP evaluation board.
+
+config CPCI405
+       bool "CPCI405"
+       help
+         This option enables support for the CPCI405 board.
+
+config EP405
+       bool "EP405/EP405PC"
+       help
+         This option enables support for the EP405/EP405PC boards.
+
+config REDWOOD_5
+       bool "Redwood-5"
+       help
+         This option enables support for the IBM STB04 evaluation board.
+
+config REDWOOD_6
+       bool "Redwood-6"
+       help
+         This option enables support for the IBM STBx25xx evaluation board.
+
+config SYCAMORE
+       bool "Sycamore"
+       help
+         This option enables support for the IBM PPC405GPr evaluation board.
+
+config WALNUT
+       bool "Walnut"
+       help
+         This option enables support for the IBM PPC405GP evaluation board.
+
+config XILINX_ML300
+       bool "Xilinx-ML300"
+       help
+         This option enables support for the Xilinx ML300 evaluation board.
+
+endchoice
+
+choice
+       prompt "Machine Type"
+       depends on 44x
+       default EBONY
+
+config BAMBOO
+       bool "Bamboo"
+       select WANT_EARLY_SERIAL
+       help
+         This option enables support for the IBM PPC440EP evaluation board.
+
+config EBONY
+       bool "Ebony"
+       select WANT_EARLY_SERIAL
+       help
+         This option enables support for the IBM PPC440GP evaluation board.
+
+config LUAN
+       bool "Luan"
+       select WANT_EARLY_SERIAL
+       help
+         This option enables support for the IBM PPC440SP evaluation board.
+
+config OCOTEA
+       bool "Ocotea"
+       select WANT_EARLY_SERIAL
+       help
+         This option enables support for the IBM PPC440GX evaluation board.
+
+endchoice
+
+config EP405PC
+       bool "EP405PC Support"
+       depends on EP405
+
+
+# It's often necessary to know the specific 4xx processor type.
+# Fortunately, it is impled (so far) from the board type, so we
+# don't need to ask more redundant questions.
+config NP405H
+       bool
+       depends on ASH
+       default y
+
+config 440EP
+       bool
+       depends on BAMBOO
+       select PPC_FPU
+       default y
+
+config 440GP
+       bool
+       depends on EBONY
+       default y
+
+config 440GX
+       bool
+       depends on OCOTEA
+       default y
+
+config 440SP
+       bool
+       depends on LUAN
+       default y
+
+config 440
+       bool
+       depends on 440GP || 440SP || 440EP
+       default y
+
+config 440A
+       bool
+       depends on 440GX
+       default y
+
+config IBM440EP_ERR42
+       bool
+       depends on 440EP
+       default y
+
+# All 405-based cores up until the 405GPR and 405EP have this errata.
+config IBM405_ERR77
+       bool
+       depends on 40x && !403GCX && !405GPR && !405EP
+       default y
+
+# All 40x-based cores, up until the 405GPR and 405EP have this errata.
+config IBM405_ERR51
+       bool
+       depends on 40x && !405GPR && !405EP
+       default y
+
+config BOOKE
+       bool
+       depends on 44x
+       default y
+
+config IBM_OCP
+       bool
+       depends on ASH || BAMBOO || BUBINGA || CPCI405 || EBONY || EP405 || LUAN || OCOTEA || REDWOOD_5 || REDWOOD_6 || SYCAMORE || WALNUT
+       default y
+
+config XILINX_OCP
+       bool
+       depends on XILINX_ML300
+       default y
+
+config IBM_EMAC4
+       bool
+       depends on 440GX || 440SP
+       default y
+
+config BIOS_FIXUP
+       bool
+       depends on BUBINGA || EP405 || SYCAMORE || WALNUT
+       default y
+
+# OAK doesn't exist but wanted to keep this around for any future 403GCX boards
+config 403GCX
+       bool
+       depends OAK
+       default y
+
+config 405EP
+       bool
+       depends on BUBINGA
+       default y
+
+config 405GP
+       bool
+       depends on CPCI405 || EP405 || WALNUT
+       default y
+
+config 405GPR
+       bool
+       depends on SYCAMORE
+       default y
+
+config VIRTEX_II_PRO
+       bool
+       depends on XILINX_ML300
+       default y
+
+config STB03xxx
+       bool
+       depends on REDWOOD_5 || REDWOOD_6
+       default y
+
+config EMBEDDEDBOOT
+       bool
+       depends on EP405 || XILINX_ML300
+       default y
+
+config IBM_OPENBIOS
+       bool
+       depends on ASH || BUBINGA || REDWOOD_5 || REDWOOD_6 || SYCAMORE || WALNUT
+       default y
+
+config PPC4xx_DMA
+       bool "PPC4xx DMA controller support"
+       depends on 4xx
+
+config PPC4xx_EDMA
+       bool
+       depends on !STB03xxx && PPC4xx_DMA
+       default y
+
+config PPC_GEN550
+       bool
+       depends on 4xx
+       default y
+
+choice
+       prompt "TTYS0 device and default console"
+       depends on 40x
+       default UART0_TTYS0
+
+config UART0_TTYS0
+       bool "UART0"
+
+config UART0_TTYS1
+       bool "UART1"
+
+endchoice
+
+config SERIAL_SICC
+       bool "SICC Serial port support"
+       depends on STB03xxx
+
+config UART1_DFLT_CONSOLE
+       bool
+       depends on SERIAL_SICC && UART0_TTYS1
+       default y
+
+config SERIAL_SICC_CONSOLE
+       bool
+       depends on SERIAL_SICC && UART0_TTYS1
+       default y
+endmenu
+
+
+menu "IBM 40x options"
+       depends on 40x
+
+config SERIAL_SICC
+       bool "SICC Serial port"
+       depends on STB03xxx
+
+config UART1_DFLT_CONSOLE
+       bool
+       depends on SERIAL_SICC && UART0_TTYS1
+       default y
+
+config SERIAL_SICC_CONSOLE
+       bool
+       depends on SERIAL_SICC && UART0_TTYS1
+       default y
+
+endmenu
diff --git a/arch/powerpc/platforms/85xx/Kconfig b/arch/powerpc/platforms/85xx/Kconfig
new file mode 100644 (file)
index 0000000..c5bc282
--- /dev/null
@@ -0,0 +1,86 @@
+config 85xx
+       bool
+       depends on E500
+       default y
+
+config PPC_INDIRECT_PCI_BE
+       bool
+       depends on 85xx
+       default y
+
+menu "Freescale 85xx options"
+       depends on E500
+
+choice
+       prompt "Machine Type"
+       depends on 85xx
+       default MPC8540_ADS
+
+config MPC8540_ADS
+       bool "Freescale MPC8540 ADS"
+       help
+         This option enables support for the MPC 8540 ADS evaluation board.
+
+config MPC8548_CDS
+       bool "Freescale MPC8548 CDS"
+       help
+         This option enablese support for the MPC8548 CDS evaluation board.
+
+config MPC8555_CDS
+       bool "Freescale MPC8555 CDS"
+       help
+         This option enablese support for the MPC8555 CDS evaluation board.
+
+config MPC8560_ADS
+       bool "Freescale MPC8560 ADS"
+       help
+         This option enables support for the MPC 8560 ADS evaluation board.
+
+config SBC8560
+       bool "WindRiver PowerQUICC III SBC8560"
+       help
+         This option enables support for the WindRiver PowerQUICC III 
+         SBC8560 board.
+
+config STX_GP3
+       bool "Silicon Turnkey Express GP3"
+       help
+         This option enables support for the Silicon Turnkey Express GP3
+         board.
+
+endchoice
+
+# It's often necessary to know the specific 85xx processor type.
+# Fortunately, it is implied (so far) from the board type, so we
+# don't need to ask more redundant questions.
+config MPC8540
+       bool
+       depends on MPC8540_ADS
+       default y
+
+config MPC8548
+       bool
+       depends on MPC8548_CDS
+       default y
+
+config MPC8555
+       bool
+       depends on MPC8555_CDS
+       default y
+
+config MPC8560
+       bool
+       depends on SBC8560 || MPC8560_ADS || STX_GP3
+       default y
+
+config 85xx_PCI2
+       bool "Supprt for 2nd PCI host controller"
+       depends on MPC8555_CDS
+       default y
+
+config PPC_GEN550
+       bool
+       depends on MPC8540 || SBC8560 || MPC8555
+       default y
+
+endmenu
diff --git a/arch/powerpc/platforms/8xx/Kconfig b/arch/powerpc/platforms/8xx/Kconfig
new file mode 100644 (file)
index 0000000..c8c0ba3
--- /dev/null
@@ -0,0 +1,352 @@
+config FADS
+       bool
+
+choice
+       prompt "8xx Machine Type"
+       depends on 8xx
+       default RPXLITE
+
+config RPXLITE
+       bool "RPX-Lite"
+       ---help---
+         Single-board computers based around the PowerPC MPC8xx chips and
+         intended for embedded applications.  The following types are
+         supported:
+
+         RPX-Lite:
+         Embedded Planet RPX Lite. PC104 form-factor SBC based on the MPC823.
+
+         RPX-Classic:
+         Embedded Planet RPX Classic Low-fat. Credit-card-size SBC based on
+         the MPC 860
+
+         BSE-IP:
+         Bright Star Engineering ip-Engine.
+
+         TQM823L:
+         TQM850L:
+         TQM855L:
+         TQM860L:
+         MPC8xx based family of mini modules, half credit card size,
+         up to 64 MB of RAM, 8 MB Flash, (Fast) Ethernet, 2 x serial ports,
+         2 x CAN bus interface, ...
+         Manufacturer: TQ Components, www.tq-group.de
+         Date of Release: October (?) 1999
+         End of Life: not yet :-)
+         URL:
+         - module: <http://www.denx.de/PDF/TQM8xxLHWM201.pdf>
+         - starter kit: <http://www.denx.de/PDF/STK8xxLHWM201.pdf>
+         - images: <http://www.denx.de/embedded-ppc-en.html>
+
+         FPS850L:
+         FingerPrint Sensor System (based on TQM850L)
+         Manufacturer: IKENDI AG, <http://www.ikendi.com/>
+         Date of Release: November 1999
+         End of life: end 2000 ?
+         URL: see TQM850L
+
+         IVMS8:
+         MPC860 based board used in the "Integrated Voice Mail System",
+         Small Version (8 voice channels)
+         Manufacturer: Speech Design, <http://www.speech-design.de/>
+         Date of Release: December 2000 (?)
+         End of life: -
+         URL: <http://www.speech-design.de/>
+
+         IVML24:
+         MPC860 based board used in the "Integrated Voice Mail System",
+         Large Version (24 voice channels)
+         Manufacturer: Speech Design, <http://www.speech-design.de/>
+         Date of Release: March 2001  (?)
+         End of life: -
+         URL: <http://www.speech-design.de/>
+
+         HERMES:
+         Hermes-Pro ISDN/LAN router with integrated 8 x hub
+         Manufacturer: Multidata Gesellschaft fur Datentechnik und Informatik
+         <http://www.multidata.de/>
+         Date of Release: 2000 (?)
+         End of life: -
+         URL: <http://www.multidata.de/english/products/hpro.htm>
+
+         IP860:
+         VMEBus IP (Industry Pack) carrier board with MPC860
+         Manufacturer: MicroSys GmbH, <http://www.microsys.de/>
+         Date of Release: ?
+         End of life: -
+         URL: <http://www.microsys.de/html/ip860.html>
+
+         PCU_E:
+         PCU = Peripheral Controller Unit, Extended
+         Manufacturer: Siemens AG, ICN (Information and Communication Networks)
+               <http://www.siemens.de/page/1,3771,224315-1-999_2_226207-0,00.html>
+         Date of Release: April 2001
+         End of life: August 2001
+         URL: n. a.
+
+config RPXCLASSIC
+       bool "RPX-Classic"
+       help
+         The RPX-Classic is a single-board computer based on the Motorola
+         MPC860.  It features 16MB of DRAM and a variable amount of flash,
+         I2C EEPROM, thermal monitoring, a PCMCIA slot, a DIP switch and two
+         LEDs.  Variants with Ethernet ports exist.  Say Y here to support it
+         directly.
+
+config BSEIP
+       bool "BSE-IP"
+       help
+         Say Y here to support the Bright Star Engineering ipEngine SBC.
+         This is a credit-card-sized device featuring a MPC823 processor,
+         26MB DRAM, 4MB flash, Ethernet, a 16K-gate FPGA, USB, an LCD/video
+         controller, and two RS232 ports.
+
+config MPC8XXFADS
+       bool "FADS"
+       select FADS
+
+config MPC86XADS
+       bool "MPC86XADS"
+       help
+         MPC86x Application Development System by Freescale Semiconductor.
+         The MPC86xADS is meant to serve as a platform for s/w and h/w
+         development around the MPC86X processor families.
+       select FADS
+
+config MPC885ADS
+       bool "MPC885ADS"
+       help
+         Freescale Semiconductor MPC885 Application Development System (ADS).
+         Also known as DUET.
+         The MPC885ADS is meant to serve as a platform for s/w and h/w
+         development around the MPC885 processor family.
+
+config TQM823L
+       bool "TQM823L"
+       help
+         Say Y here to support the TQM823L, one of an MPC8xx-based family of
+         mini SBCs (half credit-card size) from TQ Components first released
+         in late 1999.  Technical references are at
+         <http://www.denx.de/PDF/TQM8xxLHWM201.pdf>, and
+         <http://www.denx.de/PDF/STK8xxLHWM201.pdf>, and an image at
+         <http://www.denx.de/embedded-ppc-en.html>.
+
+config TQM850L
+       bool "TQM850L"
+       help
+         Say Y here to support the TQM850L, one of an MPC8xx-based family of
+         mini SBCs (half credit-card size) from TQ Components first released
+         in late 1999.  Technical references are at
+         <http://www.denx.de/PDF/TQM8xxLHWM201.pdf>, and
+         <http://www.denx.de/PDF/STK8xxLHWM201.pdf>, and an image at
+         <http://www.denx.de/embedded-ppc-en.html>.
+
+config TQM855L
+       bool "TQM855L"
+       help
+         Say Y here to support the TQM855L, one of an MPC8xx-based family of
+         mini SBCs (half credit-card size) from TQ Components first released
+         in late 1999.  Technical references are at
+         <http://www.denx.de/PDF/TQM8xxLHWM201.pdf>, and
+         <http://www.denx.de/PDF/STK8xxLHWM201.pdf>, and an image at
+         <http://www.denx.de/embedded-ppc-en.html>.
+
+config TQM860L
+       bool "TQM860L"
+       help
+         Say Y here to support the TQM860L, one of an MPC8xx-based family of
+         mini SBCs (half credit-card size) from TQ Components first released
+         in late 1999.  Technical references are at
+         <http://www.denx.de/PDF/TQM8xxLHWM201.pdf>, and
+         <http://www.denx.de/PDF/STK8xxLHWM201.pdf>, and an image at
+         <http://www.denx.de/embedded-ppc-en.html>.
+
+config FPS850L
+       bool "FPS850L"
+
+config IVMS8
+       bool "IVMS8"
+       help
+         Say Y here to support the Integrated Voice-Mail Small 8-channel SBC
+         from Speech Design, released March 2001.  The manufacturer's website
+         is at <http://www.speech-design.de/>.
+
+config IVML24
+       bool "IVML24"
+       help
+         Say Y here to support the Integrated Voice-Mail Large 24-channel SBC
+         from Speech Design, released March 2001.  The manufacturer's website
+         is at <http://www.speech-design.de/>.
+
+config HERMES_PRO
+       bool "HERMES"
+
+config IP860
+       bool "IP860"
+
+config LWMON
+       bool "LWMON"
+
+config PCU_E
+       bool "PCU_E"
+
+config CCM
+       bool "CCM"
+
+config LANTEC
+       bool "LANTEC"
+
+config MBX
+       bool "MBX"
+       help
+         MBX is a line of Motorola single-board computer based around the
+         MPC821 and MPC860 processors, and intended for embedded-controller
+         applications.  Say Y here to support these boards directly.
+
+config WINCEPT
+       bool "WinCept"
+       help
+         The Wincept 100/110 is a Motorola single-board computer based on the
+         MPC821 PowerPC, introduced in 1998 and designed to be used in
+         thin-client machines.  Say Y to support it directly.
+
+endchoice
+
+#
+# MPC8xx Communication options
+#
+
+menu "MPC8xx CPM Options"
+       depends on 8xx
+
+config SCC_ENET
+       bool "CPM SCC Ethernet"
+       depends on NET_ETHERNET
+       help
+         Enable Ethernet support via the Motorola MPC8xx serial
+         communications controller.
+
+choice
+       prompt "SCC used for Ethernet"
+       depends on SCC_ENET
+       default SCC1_ENET
+
+config SCC1_ENET
+       bool "SCC1"
+       help
+         Use MPC8xx serial communications controller 1 to drive Ethernet
+         (default).
+
+config SCC2_ENET
+       bool "SCC2"
+       help
+         Use MPC8xx serial communications controller 2 to drive Ethernet.
+
+config SCC3_ENET
+       bool "SCC3"
+       help
+         Use MPC8xx serial communications controller 3 to drive Ethernet.
+
+endchoice
+
+config FEC_ENET
+       bool "860T FEC Ethernet"
+       depends on NET_ETHERNET
+       help
+         Enable Ethernet support via the Fast Ethernet Controller (FCC) on
+         the Motorola MPC8260.
+
+config USE_MDIO
+       bool "Use MDIO for PHY configuration"
+       depends on FEC_ENET
+       help
+         On some boards the hardware configuration of the ethernet PHY can be
+         used without any software interaction over the MDIO interface, so
+         all MII code can be omitted. Say N here if unsure or if you don't
+         need link status reports.
+
+config  FEC_AM79C874
+       bool "Support AMD79C874 PHY"
+       depends on USE_MDIO
+
+config FEC_LXT970
+       bool "Support LXT970 PHY"
+       depends on USE_MDIO
+
+config FEC_LXT971
+       bool "Support LXT971 PHY"
+       depends on USE_MDIO
+       
+config FEC_QS6612
+       bool "Support QS6612 PHY"
+       depends on USE_MDIO
+       
+config ENET_BIG_BUFFERS
+       bool "Use Big CPM Ethernet Buffers"
+       depends on SCC_ENET || FEC_ENET
+       help
+         Allocate large buffers for MPC8xx Ethernet. Increases throughput
+         and decreases the likelihood of dropped packets, but costs memory.
+
+config HTDMSOUND
+       bool "Embedded Planet HIOX Audio"
+       depends on SOUND=y
+
+# This doesn't really belong here, but it is convenient to ask
+# 8xx specific questions.
+comment "Generic MPC8xx Options"
+
+config 8xx_COPYBACK
+       bool "Copy-Back Data Cache (else Writethrough)"
+       help
+         Saying Y here will cause the cache on an MPC8xx processor to be used
+         in Copy-Back mode.  If you say N here, it is used in Writethrough
+         mode.
+
+         If in doubt, say Y here.
+
+config 8xx_CPU6
+       bool "CPU6 Silicon Errata (860 Pre Rev. C)"
+       help
+         MPC860 CPUs, prior to Rev C have some bugs in the silicon, which
+         require workarounds for Linux (and most other OSes to work).  If you
+         get a BUG() very early in boot, this might fix the problem.  For
+         more details read the document entitled "MPC860 Family Device Errata
+         Reference" on Motorola's website.  This option also incurs a
+         performance hit.
+
+         If in doubt, say N here.
+
+choice
+       prompt "Microcode patch selection"
+       default NO_UCODE_PATCH
+       help
+         Help not implemented yet, coming soon.
+
+config NO_UCODE_PATCH
+       bool "None"
+
+config USB_SOF_UCODE_PATCH
+       bool "USB SOF patch"
+       help
+         Help not implemented yet, coming soon.
+
+config I2C_SPI_UCODE_PATCH
+       bool "I2C/SPI relocation patch"
+       help
+         Help not implemented yet, coming soon.
+
+config I2C_SPI_SMC1_UCODE_PATCH
+       bool "I2C/SPI/SMC1 relocation patch"
+       help
+         Help not implemented yet, coming soon.
+
+endchoice
+
+config UCODE_PATCH
+       bool
+       default y
+       depends on !NO_UCODE_PATCH
+
+endmenu
+
diff --git a/arch/powerpc/platforms/apus/Kconfig b/arch/powerpc/platforms/apus/Kconfig
new file mode 100644 (file)
index 0000000..6bde3bf
--- /dev/null
@@ -0,0 +1,130 @@
+
+config AMIGA
+       bool
+       depends on APUS
+       default y
+       help
+         This option enables support for the Amiga series of computers.
+
+config ZORRO
+       bool
+       depends on APUS
+       default y
+       help
+         This enables support for the Zorro bus in the Amiga. If you have
+         expansion cards in your Amiga that conform to the Amiga
+         AutoConfig(tm) specification, say Y, otherwise N. Note that even
+         expansion cards that do not fit in the Zorro slots but fit in e.g.
+         the CPU slot may fall in this category, so you have to say Y to let
+         Linux use these.
+
+config ABSTRACT_CONSOLE
+       bool
+       depends on APUS
+       default y
+
+config APUS_FAST_EXCEPT
+       bool
+       depends on APUS
+       default y
+
+config AMIGA_PCMCIA
+       bool "Amiga 1200/600 PCMCIA support"
+       depends on APUS && EXPERIMENTAL
+       help
+         Include support in the kernel for pcmcia on Amiga 1200 and Amiga
+         600. If you intend to use pcmcia cards say Y; otherwise say N.
+
+config AMIGA_BUILTIN_SERIAL
+       tristate "Amiga builtin serial support"
+       depends on APUS
+       help
+         If you want to use your Amiga's built-in serial port in Linux,
+         answer Y.
+
+         To compile this driver as a module, choose M here.
+
+config GVPIOEXT
+       tristate "GVP IO-Extender support"
+       depends on APUS
+       help
+         If you want to use a GVP IO-Extender serial card in Linux, say Y.
+         Otherwise, say N.
+
+config GVPIOEXT_LP
+       tristate "GVP IO-Extender parallel printer support"
+       depends on GVPIOEXT
+       help
+         Say Y to enable driving a printer from the parallel port on your
+         GVP IO-Extender card, N otherwise.
+
+config GVPIOEXT_PLIP
+       tristate "GVP IO-Extender PLIP support"
+       depends on GVPIOEXT
+       help
+         Say Y to enable doing IP over the parallel port on your GVP
+         IO-Extender card, N otherwise.
+
+config MULTIFACE_III_TTY
+       tristate "Multiface Card III serial support"
+       depends on APUS
+       help
+         If you want to use a Multiface III card's serial port in Linux,
+         answer Y.
+
+         To compile this driver as a module, choose M here.
+
+config A2232
+       tristate "Commodore A2232 serial support (EXPERIMENTAL)"
+       depends on EXPERIMENTAL && APUS
+       ---help---
+         This option supports the 2232 7-port serial card shipped with the
+         Amiga 2000 and other Zorro-bus machines, dating from 1989.  At
+         a max of 19,200 bps, the ports are served by a 6551 ACIA UART chip
+         each, plus a 8520 CIA, and a master 6502 CPU and buffer as well. The
+         ports were connected with 8 pin DIN connectors on the card bracket,
+         for which 8 pin to DB25 adapters were supplied. The card also had
+         jumpers internally to toggle various pinning configurations.
+
+         This driver can be built as a module; but then "generic_serial"
+         will also be built as a module. This has to be loaded before
+         "ser_a2232". If you want to do this, answer M here.
+
+config WHIPPET_SERIAL
+       tristate "Hisoft Whippet PCMCIA serial support"
+       depends on AMIGA_PCMCIA
+       help
+         HiSoft has a web page at <http://www.hisoft.co.uk/>, but there
+         is no listing for the Whippet in their Amiga section.
+
+config APNE
+       tristate "PCMCIA NE2000 support"
+       depends on AMIGA_PCMCIA
+       help
+         If you have a PCMCIA NE2000 compatible adapter, say Y.  Otherwise,
+         say N.
+
+         To compile this driver as a module, choose M here: the
+         module will be called apne.
+
+config SERIAL_CONSOLE
+       bool "Support for serial port console"
+       depends on APUS && (AMIGA_BUILTIN_SERIAL=y || GVPIOEXT=y || MULTIFACE_III_TTY=y)
+
+config HEARTBEAT
+       bool "Use power LED as a heartbeat"
+       depends on APUS
+       help
+         Use the power-on LED on your machine as a load meter.  The exact
+         behavior is platform-dependent, but normally the flash frequency is
+         a hyperbolic function of the 5-minute load average.
+
+config PROC_HARDWARE
+       bool "/proc/hardware support"
+       depends on APUS
+
+source "drivers/zorro/Kconfig"
+
+config PCI_PERMEDIA
+       bool "PCI for Permedia2"
+       depends on !4xx && !8xx && APUS
diff --git a/arch/powerpc/platforms/embedded6xx/Kconfig b/arch/powerpc/platforms/embedded6xx/Kconfig
new file mode 100644 (file)
index 0000000..4f35514
--- /dev/null
@@ -0,0 +1,313 @@
+choice
+       prompt "Machine Type"
+       depends on EMBEDDED6xx
+
+config APUS
+       bool "Amiga-APUS"
+       depends on BROKEN
+       help
+         Select APUS if configuring for a PowerUP Amiga.
+         More information is available at:
+         <http://linux-apus.sourceforge.net/>.
+
+config KATANA
+       bool "Artesyn-Katana"
+       help
+         Select KATANA if configuring an Artesyn KATANA 750i or 3750
+         cPCI board.
+
+config WILLOW
+       bool "Cogent-Willow"
+
+config CPCI690
+       bool "Force-CPCI690"
+       help
+         Select CPCI690 if configuring a Force CPCI690 cPCI board.
+
+config POWERPMC250
+       bool "Force-PowerPMC250"
+
+config CHESTNUT
+       bool "IBM 750FX Eval board or 750GX Eval board"
+       help
+         Select CHESTNUT if configuring an IBM 750FX Eval Board or a
+         IBM 750GX Eval board.
+
+config SPRUCE
+       bool "IBM-Spruce"
+
+config HDPU
+       bool "Sky-HDPU"
+       help
+         Select HDPU if configuring a Sky Computers Compute Blade.
+
+config HDPU_FEATURES
+       depends HDPU
+       tristate "HDPU-Features"
+       help
+         Select to enable HDPU enhanced features.
+
+config EV64260
+       bool "Marvell-EV64260BP"
+       help
+         Select EV64260 if configuring a Marvell (formerly Galileo)
+         EV64260BP Evaluation platform.
+
+config LOPEC
+       bool "Motorola-LoPEC"
+
+config MVME5100
+       bool "Motorola-MVME5100"
+
+config PPLUS
+       bool "Motorola-PowerPlus"
+
+config PRPMC750
+       bool "Motorola-PrPMC750"
+
+config PRPMC800
+       bool "Motorola-PrPMC800"
+
+config SANDPOINT
+       bool "Motorola-Sandpoint"
+       help
+         Select SANDPOINT if configuring for a Motorola Sandpoint X3
+         (any flavor).
+
+config RADSTONE_PPC7D
+       bool "Radstone Technology PPC7D board"
+
+config PAL4
+       bool "SBS-Palomar4"
+
+config GEMINI
+       bool "Synergy-Gemini"
+       depends on BROKEN
+       help
+         Select Gemini if configuring for a Synergy Microsystems' Gemini
+         series Single Board Computer.  More information is available at:
+         <http://www.synergymicro.com/PressRel/97_10_15.html>.
+
+config EST8260
+       bool "EST8260"
+       ---help---
+         The EST8260 is a single-board computer manufactured by Wind River
+         Systems, Inc. (formerly Embedded Support Tools Corp.) and based on
+         the MPC8260.  Wind River Systems has a website at
+         <http://www.windriver.com/>, but the EST8260 cannot be found on it
+         and has probably been discontinued or rebadged.
+
+config SBC82xx
+       bool "SBC82xx"
+       ---help---
+         SBC PowerQUICC II, single-board computer with MPC82xx CPU
+         Manufacturer: Wind River Systems, Inc.
+         Date of Release: May 2003
+         End of Life: -
+         URL: <http://www.windriver.com/>
+
+config SBS8260
+       bool "SBS8260"
+
+config RPX8260
+       bool "RPXSUPER"
+
+config TQM8260
+       bool "TQM8260"
+       ---help---
+         MPC8260 based module, little larger than credit card,
+         up to 128 MB global + 64 MB local RAM, 32 MB Flash,
+         32 kB EEPROM, 256 kB L@ Cache, 10baseT + 100baseT Ethernet,
+         2 x serial ports, ...
+         Manufacturer: TQ Components, www.tq-group.de
+         Date of Release: June 2001
+         End of Life: not yet :-)
+         URL: <http://www.denx.de/PDF/TQM82xx_SPEC_Rev005.pdf>
+
+config ADS8272
+       bool "ADS8272"
+
+config PQ2FADS
+       bool "Freescale-PQ2FADS"
+       help
+         Select PQ2FADS if you wish to configure for a Freescale
+         PQ2FADS board (-VR or -ZU).
+
+config LITE5200
+       bool "Freescale LITE5200 / (IceCube)"
+       select PPC_MPC52xx
+       help
+         Support for the LITE5200 dev board for the MPC5200 from Freescale.
+         This is for the LITE5200 version 2.0 board. Don't know if it changes
+         much but it's only been tested on this board version. I think this
+         board is also known as IceCube.
+
+config MPC834x_SYS
+       bool "Freescale MPC834x SYS"
+       help
+         This option enables support for the MPC 834x SYS evaluation board.
+
+         Be aware that PCI buses can only function when SYS board is plugged
+         into the PIB (Platform IO Board) board from Freescale which provide
+         3 PCI slots.  The PIBs PCI initialization is the bootloader's
+         responsiblilty.
+
+config EV64360
+       bool "Marvell-EV64360BP"
+       help
+         Select EV64360 if configuring a Marvell EV64360BP Evaluation
+         platform.
+endchoice
+
+config PQ2ADS
+       bool
+       depends on ADS8272
+       default y
+
+config TQM8xxL
+       bool
+       depends on 8xx && (TQM823L || TQM850L || FPS850L || TQM855L || TQM860L)
+       default y
+
+config PPC_MPC52xx
+       bool
+
+config 8260
+       bool "CPM2 Support" if WILLOW
+       depends on 6xx
+       default y if TQM8260 || RPX8260 || EST8260 || SBS8260 || SBC82xx || PQ2FADS
+       help
+         The MPC8260 is a typical embedded CPU made by Motorola.  Selecting
+         this option means that you wish to build a kernel for a machine with
+         an 8260 class CPU.
+
+config 8272
+       bool
+       depends on 6xx
+       default y if ADS8272
+       select 8260
+       help
+         The MPC8272 CPM has a different internal dpram setup than other CPM2
+         devices
+
+config 83xx
+       bool
+       default y if MPC834x_SYS
+
+config MPC834x
+       bool
+       default y if MPC834x_SYS
+
+config CPM2
+       bool
+       depends on 8260 || MPC8560 || MPC8555
+       default y
+       help
+         The CPM2 (Communications Processor Module) is a coprocessor on
+         embedded CPUs made by Motorola.  Selecting this option means that
+         you wish to build a kernel for a machine with a CPM2 coprocessor
+         on it (826x, 827x, 8560).
+
+config PPC_GEN550
+       bool
+       depends on SANDPOINT || SPRUCE || PPLUS || \
+               PRPMC750 || PRPMC800 || LOPEC || \
+               (EV64260 && !SERIAL_MPSC) || CHESTNUT || RADSTONE_PPC7D || \
+               83xx
+       default y
+
+config FORCE
+       bool
+       depends on 6xx && POWERPMC250
+       default y
+
+config GT64260
+       bool
+       depends on EV64260 || CPCI690
+       default y
+
+config MV64360         # Really MV64360 & MV64460
+       bool
+       depends on CHESTNUT || KATANA || RADSTONE_PPC7D || HDPU || EV64360
+       default y
+
+config MV64X60
+       bool
+       depends on (GT64260 || MV64360)
+       default y
+
+menu "Set bridge options"
+       depends on MV64X60
+
+config NOT_COHERENT_CACHE
+       bool "Turn off Cache Coherency"
+       default n
+       help
+         Some 64x60 bridges lock up when trying to enforce cache coherency.
+         When this option is selected, cache coherency will be turned off.
+         Note that this can cause other problems (e.g., stale data being
+         speculatively loaded via a cached mapping).  Use at your own risk.
+
+config MV64X60_BASE
+       hex "Set bridge base used by firmware"
+       default "0xf1000000"
+       help
+         A firmware can leave the base address of the bridge's registers at
+         a non-standard location.  If so, set this value to reflect the
+         address of that non-standard location.
+
+config MV64X60_NEW_BASE
+       hex "Set bridge base used by kernel"
+       default "0xf1000000"
+       help
+         If the current base address of the bridge's registers is not where
+         you want it, set this value to the address that you want it moved to.
+
+endmenu
+
+config NONMONARCH_SUPPORT
+       bool "Enable Non-Monarch Support"
+       depends on PRPMC800
+
+config HARRIER
+       bool
+       depends on PRPMC800
+       default y
+
+config EPIC_SERIAL_MODE
+       bool
+       depends on 6xx && (LOPEC || SANDPOINT)
+       default y
+
+config MPC10X_BRIDGE
+       bool
+       depends on POWERPMC250 || LOPEC || SANDPOINT
+       default y
+
+config MPC10X_OPENPIC
+       bool
+       depends on POWERPMC250 || LOPEC || SANDPOINT
+       default y
+
+config MPC10X_STORE_GATHERING
+       bool "Enable MPC10x store gathering"
+       depends on MPC10X_BRIDGE
+
+config SANDPOINT_ENABLE_UART1
+       bool "Enable DUART mode on Sandpoint"
+       depends on SANDPOINT
+       help
+         If this option is enabled then the MPC824x processor will run
+         in DUART mode instead of UART mode.
+
+config HARRIER_STORE_GATHERING
+       bool "Enable Harrier store gathering"
+       depends on HARRIER
+
+config MVME5100_IPMC761_PRESENT
+       bool "MVME5100 configured with an IPMC761"
+       depends on MVME5100
+
+config SPRUCE_BAUD_33M
+       bool "Spruce baud clock support"
+       depends on SPRUCE
diff --git a/arch/powerpc/platforms/iseries/Kconfig b/arch/powerpc/platforms/iseries/Kconfig
new file mode 100644 (file)
index 0000000..3d957a3
--- /dev/null
@@ -0,0 +1,31 @@
+
+menu "iSeries device drivers"
+       depends on PPC_ISERIES
+
+config VIOCONS
+       tristate "iSeries Virtual Console Support"
+
+config VIODASD
+       tristate "iSeries Virtual I/O disk support"
+       help
+         If you are running on an iSeries system and you want to use
+         virtual disks created and managed by OS/400, say Y.
+
+config VIOCD
+       tristate "iSeries Virtual I/O CD support"
+       help
+         If you are running Linux on an IBM iSeries system and you want to
+         read a CD drive owned by OS/400, say Y here.
+
+config VIOTAPE
+       tristate "iSeries Virtual Tape Support"
+       help
+         If you are running Linux on an iSeries system and you want Linux
+         to read and/or write a tape drive owned by OS/400, say Y here.
+
+endmenu
+
+config VIOPATH
+       bool
+       depends on VIOCONS || VIODASD || VIOCD || VIOTAPE || VETH
+       default y
diff --git a/arch/powerpc/platforms/powermac/Makefile b/arch/powerpc/platforms/powermac/Makefile
new file mode 100644 (file)
index 0000000..37b7341
--- /dev/null
@@ -0,0 +1,9 @@
+obj-$(CONFIG_PPC_PMAC)         += pmac_pic.o pmac_setup.o pmac_time.o \
+                                  pmac_feature.o pmac_pci.o pmac_sleep.o \
+                                  pmac_low_i2c.o pmac_cache.o
+obj-$(CONFIG_PMAC_BACKLIGHT)   += pmac_backlight.o
+obj-$(CONFIG_CPU_FREQ_PMAC)    += pmac_cpufreq.o
+ifeq ($(CONFIG_PPC_PMAC),y)
+obj-$(CONFIG_NVRAM)            += pmac_nvram.o
+obj-$(CONFIG_SMP)              += pmac_smp.o
+endif
diff --git a/arch/powerpc/platforms/powermac/pmac.h b/arch/powerpc/platforms/powermac/pmac.h
new file mode 100644 (file)
index 0000000..40e1c50
--- /dev/null
@@ -0,0 +1,31 @@
+#ifndef __PMAC_H__
+#define __PMAC_H__
+
+#include <linux/pci.h>
+#include <linux/ide.h>
+
+/*
+ * Declaration for the various functions exported by the
+ * pmac_* files. Mostly for use by pmac_setup
+ */
+
+extern void pmac_get_boot_time(struct rtc_time *tm);
+extern void pmac_get_rtc_time(struct rtc_time *tm);
+extern int  pmac_set_rtc_time(struct rtc_time *tm);
+extern void pmac_read_rtc_time(void);
+extern void pmac_calibrate_decr(void);
+
+extern void pmac_pcibios_fixup(void);
+extern void pmac_pci_init(void);
+extern void pmac_setup_pci_dma(void);
+extern void pmac_check_ht_link(void);
+
+extern void pmac_setup_smp(void);
+
+extern unsigned long pmac_ide_get_base(int index);
+extern void pmac_ide_init_hwif_ports(hw_regs_t *hw,
+       unsigned long data_port, unsigned long ctrl_port, int *irq);
+
+extern void pmac_nvram_init(void);
+
+#endif /* __PMAC_H__ */
diff --git a/arch/powerpc/platforms/powermac/pmac_backlight.c b/arch/powerpc/platforms/powermac/pmac_backlight.c
new file mode 100644 (file)
index 0000000..8be2f7d
--- /dev/null
@@ -0,0 +1,202 @@
+/*
+ * Miscellaneous procedures for dealing with the PowerMac hardware.
+ * Contains support for the backlight.
+ *
+ *   Copyright (C) 2000 Benjamin Herrenschmidt
+ *
+ */
+
+#include <linux/config.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/stddef.h>
+#include <linux/reboot.h>
+#include <linux/nvram.h>
+#include <linux/console.h>
+#include <asm/sections.h>
+#include <asm/ptrace.h>
+#include <asm/io.h>
+#include <asm/pgtable.h>
+#include <asm/system.h>
+#include <asm/prom.h>
+#include <asm/machdep.h>
+#include <asm/nvram.h>
+#include <asm/backlight.h>
+
+#include <linux/adb.h>
+#include <linux/pmu.h>
+
+static struct backlight_controller *backlighter;
+static void* backlighter_data;
+static int backlight_autosave;
+static int backlight_level = BACKLIGHT_MAX;
+static int backlight_enabled = 1;
+static int backlight_req_level = -1;
+static int backlight_req_enable = -1;
+
+static void backlight_callback(void *);
+static DECLARE_WORK(backlight_work, backlight_callback, NULL);
+
+void register_backlight_controller(struct backlight_controller *ctrler,
+                                         void *data, char *type)
+{
+       struct device_node* bk_node;
+       char *prop;
+       int valid = 0;
+
+       /* There's already a matching controller, bail out */
+       if (backlighter != NULL)
+               return;
+
+       bk_node = find_devices("backlight");
+
+#ifdef CONFIG_ADB_PMU
+       /* Special case for the old PowerBook since I can't test on it */
+       backlight_autosave = machine_is_compatible("AAPL,3400/2400")
+               || machine_is_compatible("AAPL,3500");
+       if ((backlight_autosave
+            || machine_is_compatible("AAPL,PowerBook1998")
+            || machine_is_compatible("PowerBook1,1"))
+           && !strcmp(type, "pmu"))
+               valid = 1;
+#endif
+       if (bk_node) {
+               prop = get_property(bk_node, "backlight-control", NULL);
+               if (prop && !strncmp(prop, type, strlen(type)))
+                       valid = 1;
+       }
+       if (!valid)
+               return;
+       backlighter = ctrler;
+       backlighter_data = data;
+
+       if (bk_node && !backlight_autosave)
+               prop = get_property(bk_node, "bklt", NULL);
+       else
+               prop = NULL;
+       if (prop) {
+               backlight_level = ((*prop)+1) >> 1;
+               if (backlight_level > BACKLIGHT_MAX)
+                       backlight_level = BACKLIGHT_MAX;
+       }
+
+#ifdef CONFIG_ADB_PMU
+       if (backlight_autosave) {
+               struct adb_request req;
+               pmu_request(&req, NULL, 2, 0xd9, 0);
+               while (!req.complete)
+                       pmu_poll();
+               backlight_level = req.reply[0] >> 4;
+       }
+#endif
+       acquire_console_sem();
+       if (!backlighter->set_enable(1, backlight_level, data))
+               backlight_enabled = 1;
+       release_console_sem();
+
+       printk(KERN_INFO "Registered \"%s\" backlight controller,"
+              "level: %d/15\n", type, backlight_level);
+}
+EXPORT_SYMBOL(register_backlight_controller);
+
+void unregister_backlight_controller(struct backlight_controller
+                                           *ctrler, void *data)
+{
+       /* We keep the current backlight level (for now) */
+       if (ctrler == backlighter && data == backlighter_data)
+               backlighter = NULL;
+}
+EXPORT_SYMBOL(unregister_backlight_controller);
+
+static int __set_backlight_enable(int enable)
+{
+       int rc;
+
+       if (!backlighter)
+               return -ENODEV;
+       acquire_console_sem();
+       rc = backlighter->set_enable(enable, backlight_level,
+                                    backlighter_data);
+       if (!rc)
+               backlight_enabled = enable;
+       release_console_sem();
+       return rc;
+}
+int set_backlight_enable(int enable)
+{
+       if (!backlighter)
+               return -ENODEV;
+       backlight_req_enable = enable;
+       schedule_work(&backlight_work);
+       return 0;
+}
+
+EXPORT_SYMBOL(set_backlight_enable);
+
+int get_backlight_enable(void)
+{
+       if (!backlighter)
+               return -ENODEV;
+       return backlight_enabled;
+}
+EXPORT_SYMBOL(get_backlight_enable);
+
+static int __set_backlight_level(int level)
+{
+       int rc = 0;
+
+       if (!backlighter)
+               return -ENODEV;
+       if (level < BACKLIGHT_MIN)
+               level = BACKLIGHT_OFF;
+       if (level > BACKLIGHT_MAX)
+               level = BACKLIGHT_MAX;
+       acquire_console_sem();
+       if (backlight_enabled)
+               rc = backlighter->set_level(level, backlighter_data);
+       if (!rc)
+               backlight_level = level;
+       release_console_sem();
+       if (!rc && !backlight_autosave) {
+               level <<=1;
+               if (level & 0x10)
+                       level |= 0x01;
+               // -- todo: save to property "bklt"
+       }
+       return rc;
+}
+int set_backlight_level(int level)
+{
+       if (!backlighter)
+               return -ENODEV;
+       backlight_req_level = level;
+       schedule_work(&backlight_work);
+       return 0;
+}
+
+EXPORT_SYMBOL(set_backlight_level);
+
+int get_backlight_level(void)
+{
+       if (!backlighter)
+               return -ENODEV;
+       return backlight_level;
+}
+EXPORT_SYMBOL(get_backlight_level);
+
+static void backlight_callback(void *dummy)
+{
+       int level, enable;
+
+       do {
+               level = backlight_req_level;
+               enable = backlight_req_enable;
+               mb();
+
+               if (level >= 0)
+                       __set_backlight_level(level);
+               if (enable >= 0)
+                       __set_backlight_enable(enable);
+       } while(cmpxchg(&backlight_req_level, level, -1) != level ||
+               cmpxchg(&backlight_req_enable, enable, -1) != enable);
+}
diff --git a/arch/powerpc/platforms/powermac/pmac_cache.S b/arch/powerpc/platforms/powermac/pmac_cache.S
new file mode 100644 (file)
index 0000000..fb977de
--- /dev/null
@@ -0,0 +1,359 @@
+/*
+ * This file contains low-level cache management functions
+ * used for sleep and CPU speed changes on Apple machines.
+ * (In fact the only thing that is Apple-specific is that we assume
+ * that we can read from ROM at physical address 0xfff00000.)
+ *
+ *    Copyright (C) 2004 Paul Mackerras (paulus@samba.org) and
+ *                       Benjamin Herrenschmidt (benh@kernel.crashing.org)
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version
+ * 2 of the License, or (at your option) any later version.
+ *
+ */
+
+#include <linux/config.h>
+#include <asm/processor.h>
+#include <asm/ppc_asm.h>
+#include <asm/cputable.h>
+
+/*
+ * Flush and disable all data caches (dL1, L2, L3). This is used
+ * when going to sleep, when doing a PMU based cpufreq transition,
+ * or when "offlining" a CPU on SMP machines. This code is over
+ * paranoid, but I've had enough issues with various CPU revs and
+ * bugs that I decided it was worth beeing over cautious
+ */
+
+_GLOBAL(flush_disable_caches)
+#ifndef CONFIG_6xx
+       blr
+#else
+BEGIN_FTR_SECTION
+       b       flush_disable_745x
+END_FTR_SECTION_IFSET(CPU_FTR_SPEC7450)
+BEGIN_FTR_SECTION
+       b       flush_disable_75x
+END_FTR_SECTION_IFSET(CPU_FTR_L2CR)
+       b       __flush_disable_L1
+
+/* This is the code for G3 and 74[01]0 */
+flush_disable_75x:
+       mflr    r10
+
+       /* Turn off EE and DR in MSR */
+       mfmsr   r11
+       rlwinm  r0,r11,0,~MSR_EE
+       rlwinm  r0,r0,0,~MSR_DR
+       sync
+       mtmsr   r0
+       isync
+
+       /* Stop DST streams */
+BEGIN_FTR_SECTION
+       DSSALL
+       sync
+END_FTR_SECTION_IFSET(CPU_FTR_ALTIVEC)
+
+       /* Stop DPM */
+       mfspr   r8,SPRN_HID0            /* Save SPRN_HID0 in r8 */
+       rlwinm  r4,r8,0,12,10           /* Turn off HID0[DPM] */
+       sync
+       mtspr   SPRN_HID0,r4            /* Disable DPM */
+       sync
+
+       /* Disp-flush L1. We have a weird problem here that I never
+        * totally figured out. On 750FX, using the ROM for the flush
+        * results in a non-working flush. We use that workaround for
+        * now until I finally understand what's going on. --BenH
+        */
+
+       /* ROM base by default */
+       lis     r4,0xfff0
+       mfpvr   r3
+       srwi    r3,r3,16
+       cmplwi  cr0,r3,0x7000
+       bne+    1f
+       /* RAM base on 750FX */
+       li      r4,0
+1:     li      r4,0x4000
+       mtctr   r4
+1:     lwz     r0,0(r4)
+       addi    r4,r4,32
+       bdnz    1b
+       sync
+       isync
+
+       /* Disable / invalidate / enable L1 data */
+       mfspr   r3,SPRN_HID0
+       rlwinm  r3,r3,0,~(HID0_DCE | HID0_ICE)
+       mtspr   SPRN_HID0,r3
+       sync
+       isync
+       ori     r3,r3,(HID0_DCE|HID0_DCI|HID0_ICE|HID0_ICFI)
+       sync
+       isync
+       mtspr   SPRN_HID0,r3
+       xori    r3,r3,(HID0_DCI|HID0_ICFI)
+       mtspr   SPRN_HID0,r3
+       sync
+
+       /* Get the current enable bit of the L2CR into r4 */
+       mfspr   r5,SPRN_L2CR
+       /* Set to data-only (pre-745x bit) */
+       oris    r3,r5,L2CR_L2DO@h
+       b       2f
+       /* When disabling L2, code must be in L1 */
+       .balign 32
+1:     mtspr   SPRN_L2CR,r3
+3:     sync
+       isync
+       b       1f
+2:     b       3f
+3:     sync
+       isync
+       b       1b
+1:     /* disp-flush L2. The interesting thing here is that the L2 can be
+        * up to 2Mb ... so using the ROM, we'll end up wrapping back to memory
+        * but that is probbaly fine. We disp-flush over 4Mb to be safe
+        */
+       lis     r4,2
+       mtctr   r4
+       lis     r4,0xfff0
+1:     lwz     r0,0(r4)
+       addi    r4,r4,32
+       bdnz    1b
+       sync
+       isync
+       lis     r4,2
+       mtctr   r4
+       lis     r4,0xfff0
+1:     dcbf    0,r4
+       addi    r4,r4,32
+       bdnz    1b
+       sync
+       isync
+
+       /* now disable L2 */
+       rlwinm  r5,r5,0,~L2CR_L2E
+       b       2f
+       /* When disabling L2, code must be in L1 */
+       .balign 32
+1:     mtspr   SPRN_L2CR,r5
+3:     sync
+       isync
+       b       1f
+2:     b       3f
+3:     sync
+       isync
+       b       1b
+1:     sync
+       isync
+       /* Invalidate L2. This is pre-745x, we clear the L2I bit ourselves */
+       oris    r4,r5,L2CR_L2I@h
+       mtspr   SPRN_L2CR,r4
+       sync
+       isync
+
+       /* Wait for the invalidation to complete */
+1:     mfspr   r3,SPRN_L2CR
+       rlwinm. r0,r3,0,31,31
+       bne     1b
+
+       /* Clear L2I */
+       xoris   r4,r4,L2CR_L2I@h
+       sync
+       mtspr   SPRN_L2CR,r4
+       sync
+
+       /* now disable the L1 data cache */
+       mfspr   r0,SPRN_HID0
+       rlwinm  r0,r0,0,~(HID0_DCE|HID0_ICE)
+       mtspr   SPRN_HID0,r0
+       sync
+       isync
+
+       /* Restore HID0[DPM] to whatever it was before */
+       sync
+       mfspr   r0,SPRN_HID0
+       rlwimi  r0,r8,0,11,11           /* Turn back HID0[DPM] */
+       mtspr   SPRN_HID0,r0
+       sync
+
+       /* restore DR and EE */
+       sync
+       mtmsr   r11
+       isync
+
+       mtlr    r10
+       blr
+
+/* This code is for 745x processors */
+flush_disable_745x:
+       /* Turn off EE and DR in MSR */
+       mfmsr   r11
+       rlwinm  r0,r11,0,~MSR_EE
+       rlwinm  r0,r0,0,~MSR_DR
+       sync
+       mtmsr   r0
+       isync
+
+       /* Stop prefetch streams */
+       DSSALL
+       sync
+
+       /* Disable L2 prefetching */
+       mfspr   r0,SPRN_MSSCR0
+       rlwinm  r0,r0,0,0,29
+       mtspr   SPRN_MSSCR0,r0
+       sync
+       isync
+       lis     r4,0
+       dcbf    0,r4
+       dcbf    0,r4
+       dcbf    0,r4
+       dcbf    0,r4
+       dcbf    0,r4
+       dcbf    0,r4
+       dcbf    0,r4
+       dcbf    0,r4
+
+       /* Due to a bug with the HW flush on some CPU revs, we occasionally
+        * experience data corruption. I'm adding a displacement flush along
+        * with a dcbf loop over a few Mb to "help". The problem isn't totally
+        * fixed by this in theory, but at least, in practice, I couldn't reproduce
+        * it even with a big hammer...
+        */
+
+        lis     r4,0x0002
+        mtctr   r4
+       li      r4,0
+1:
+        lwz     r0,0(r4)
+        addi    r4,r4,32                /* Go to start of next cache line */
+        bdnz    1b
+        isync
+
+        /* Now, flush the first 4MB of memory */
+        lis     r4,0x0002
+        mtctr   r4
+       li      r4,0
+        sync
+1:
+        dcbf    0,r4
+        addi    r4,r4,32                /* Go to start of next cache line */
+        bdnz    1b
+
+       /* Flush and disable the L1 data cache */
+       mfspr   r6,SPRN_LDSTCR
+       lis     r3,0xfff0       /* read from ROM for displacement flush */
+       li      r4,0xfe         /* start with only way 0 unlocked */
+       li      r5,128          /* 128 lines in each way */
+1:     mtctr   r5
+       rlwimi  r6,r4,0,24,31
+       mtspr   SPRN_LDSTCR,r6
+       sync
+       isync
+2:     lwz     r0,0(r3)        /* touch each cache line */
+       addi    r3,r3,32
+       bdnz    2b
+       rlwinm  r4,r4,1,24,30   /* move on to the next way */
+       ori     r4,r4,1
+       cmpwi   r4,0xff         /* all done? */
+       bne     1b
+       /* now unlock the L1 data cache */
+       li      r4,0
+       rlwimi  r6,r4,0,24,31
+       sync
+       mtspr   SPRN_LDSTCR,r6
+       sync
+       isync
+
+       /* Flush the L2 cache using the hardware assist */
+       mfspr   r3,SPRN_L2CR
+       cmpwi   r3,0            /* check if it is enabled first */
+       bge     4f
+       oris    r0,r3,(L2CR_L2IO_745x|L2CR_L2DO_745x)@h
+       b       2f
+       /* When disabling/locking L2, code must be in L1 */
+       .balign 32
+1:     mtspr   SPRN_L2CR,r0    /* lock the L2 cache */
+3:     sync
+       isync
+       b       1f
+2:     b       3f
+3:     sync
+       isync
+       b       1b
+1:     sync
+       isync
+       ori     r0,r3,L2CR_L2HWF_745x
+       sync
+       mtspr   SPRN_L2CR,r0    /* set the hardware flush bit */
+3:     mfspr   r0,SPRN_L2CR    /* wait for it to go to 0 */
+       andi.   r0,r0,L2CR_L2HWF_745x
+       bne     3b
+       sync
+       rlwinm  r3,r3,0,~L2CR_L2E
+       b       2f
+       /* When disabling L2, code must be in L1 */
+       .balign 32
+1:     mtspr   SPRN_L2CR,r3    /* disable the L2 cache */
+3:     sync
+       isync
+       b       1f
+2:     b       3f
+3:     sync
+       isync
+       b       1b
+1:     sync
+       isync
+       oris    r4,r3,L2CR_L2I@h
+       mtspr   SPRN_L2CR,r4
+       sync
+       isync
+1:     mfspr   r4,SPRN_L2CR
+       andis.  r0,r4,L2CR_L2I@h
+       bne     1b
+       sync
+
+BEGIN_FTR_SECTION
+       /* Flush the L3 cache using the hardware assist */
+4:     mfspr   r3,SPRN_L3CR
+       cmpwi   r3,0            /* check if it is enabled */
+       bge     6f
+       oris    r0,r3,L3CR_L3IO@h
+       ori     r0,r0,L3CR_L3DO
+       sync
+       mtspr   SPRN_L3CR,r0    /* lock the L3 cache */
+       sync
+       isync
+       ori     r0,r0,L3CR_L3HWF
+       sync
+       mtspr   SPRN_L3CR,r0    /* set the hardware flush bit */
+5:     mfspr   r0,SPRN_L3CR    /* wait for it to go to zero */
+       andi.   r0,r0,L3CR_L3HWF
+       bne     5b
+       rlwinm  r3,r3,0,~L3CR_L3E
+       sync
+       mtspr   SPRN_L3CR,r3    /* disable the L3 cache */
+       sync
+       ori     r4,r3,L3CR_L3I
+       mtspr   SPRN_L3CR,r4
+1:     mfspr   r4,SPRN_L3CR
+       andi.   r0,r4,L3CR_L3I
+       bne     1b
+       sync
+END_FTR_SECTION_IFSET(CPU_FTR_L3CR)
+
+6:     mfspr   r0,SPRN_HID0    /* now disable the L1 data cache */
+       rlwinm  r0,r0,0,~HID0_DCE
+       mtspr   SPRN_HID0,r0
+       sync
+       isync
+       mtmsr   r11             /* restore DR and EE */
+       isync
+       blr
+#endif /* CONFIG_6xx */
diff --git a/arch/powerpc/platforms/powermac/pmac_cpufreq.c b/arch/powerpc/platforms/powermac/pmac_cpufreq.c
new file mode 100644 (file)
index 0000000..6d32d99
--- /dev/null
@@ -0,0 +1,728 @@
+/*
+ *  arch/ppc/platforms/pmac_cpufreq.c
+ *
+ *  Copyright (C) 2002 - 2005 Benjamin Herrenschmidt <benh@kernel.crashing.org>
+ *  Copyright (C) 2004        John Steele Scott <toojays@toojays.net>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * TODO: Need a big cleanup here. Basically, we need to have different
+ * cpufreq_driver structures for the different type of HW instead of the
+ * current mess. We also need to better deal with the detection of the
+ * type of machine.
+ *
+ */
+
+#include <linux/config.h>
+#include <linux/module.h>
+#include <linux/types.h>
+#include <linux/errno.h>
+#include <linux/kernel.h>
+#include <linux/delay.h>
+#include <linux/sched.h>
+#include <linux/adb.h>
+#include <linux/pmu.h>
+#include <linux/slab.h>
+#include <linux/cpufreq.h>
+#include <linux/init.h>
+#include <linux/sysdev.h>
+#include <linux/i2c.h>
+#include <linux/hardirq.h>
+#include <asm/prom.h>
+#include <asm/machdep.h>
+#include <asm/irq.h>
+#include <asm/pmac_feature.h>
+#include <asm/mmu_context.h>
+#include <asm/sections.h>
+#include <asm/cputable.h>
+#include <asm/time.h>
+#include <asm/system.h>
+#include <asm/mpic.h>
+#include <asm/keylargo.h>
+
+/* WARNING !!! This will cause calibrate_delay() to be called,
+ * but this is an __init function ! So you MUST go edit
+ * init/main.c to make it non-init before enabling DEBUG_FREQ
+ */
+#undef DEBUG_FREQ
+
+/*
+ * There is a problem with the core cpufreq code on SMP kernels,
+ * it won't recalculate the Bogomips properly
+ */
+#ifdef CONFIG_SMP
+#warning "WARNING, CPUFREQ not recommended on SMP kernels"
+#endif
+
+extern void low_choose_7447a_dfs(int dfs);
+extern void low_choose_750fx_pll(int pll);
+extern void low_sleep_handler(void);
+
+/*
+ * Currently, PowerMac cpufreq supports only high & low frequencies
+ * that are set by the firmware
+ */
+static unsigned int low_freq;
+static unsigned int hi_freq;
+static unsigned int cur_freq;
+static unsigned int sleep_freq;
+
+/*
+ * Different models uses different mecanisms to switch the frequency
+ */
+static int (*set_speed_proc)(int low_speed);
+static unsigned int (*get_speed_proc)(void);
+
+/*
+ * Some definitions used by the various speedprocs
+ */
+static u32 voltage_gpio;
+static u32 frequency_gpio;
+static u32 slew_done_gpio;
+static int no_schedule;
+static int has_cpu_l2lve;
+static int is_pmu_based;
+
+/* There are only two frequency states for each processor. Values
+ * are in kHz for the time being.
+ */
+#define CPUFREQ_HIGH                  0
+#define CPUFREQ_LOW                   1
+
+static struct cpufreq_frequency_table pmac_cpu_freqs[] = {
+       {CPUFREQ_HIGH,          0},
+       {CPUFREQ_LOW,           0},
+       {0,                     CPUFREQ_TABLE_END},
+};
+
+static struct freq_attr* pmac_cpu_freqs_attr[] = {
+       &cpufreq_freq_attr_scaling_available_freqs,
+       NULL,
+};
+
+static inline void local_delay(unsigned long ms)
+{
+       if (no_schedule)
+               mdelay(ms);
+       else
+               msleep(ms);
+}
+
+static inline void wakeup_decrementer(void)
+{
+       set_dec(tb_ticks_per_jiffy);
+       /* No currently-supported powerbook has a 601,
+        * so use get_tbl, not native
+        */
+       last_jiffy_stamp(0) = tb_last_stamp = get_tbl();
+}
+
+#ifdef DEBUG_FREQ
+static inline void debug_calc_bogomips(void)
+{
+       /* This will cause a recalc of bogomips and display the
+        * result. We backup/restore the value to avoid affecting the
+        * core cpufreq framework's own calculation.
+        */
+       extern void calibrate_delay(void);
+
+       unsigned long save_lpj = loops_per_jiffy;
+       calibrate_delay();
+       loops_per_jiffy = save_lpj;
+}
+#endif /* DEBUG_FREQ */
+
+/* Switch CPU speed under 750FX CPU control
+ */
+static int cpu_750fx_cpu_speed(int low_speed)
+{
+       u32 hid2;
+
+       if (low_speed == 0) {
+               /* ramping up, set voltage first */
+               pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, voltage_gpio, 0x05);
+               /* Make sure we sleep for at least 1ms */
+               local_delay(10);
+
+               /* tweak L2 for high voltage */
+               if (has_cpu_l2lve) {
+                       hid2 = mfspr(SPRN_HID2);
+                       hid2 &= ~0x2000;
+                       mtspr(SPRN_HID2, hid2);
+               }
+       }
+#ifdef CONFIG_6xx
+       low_choose_750fx_pll(low_speed);
+#endif
+       if (low_speed == 1) {
+               /* tweak L2 for low voltage */
+               if (has_cpu_l2lve) {
+                       hid2 = mfspr(SPRN_HID2);
+                       hid2 |= 0x2000;
+                       mtspr(SPRN_HID2, hid2);
+               }
+
+               /* ramping down, set voltage last */
+               pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, voltage_gpio, 0x04);
+               local_delay(10);
+       }
+
+       return 0;
+}
+
+static unsigned int cpu_750fx_get_cpu_speed(void)
+{
+       if (mfspr(SPRN_HID1) & HID1_PS)
+               return low_freq;
+       else
+               return hi_freq;
+}
+
+/* Switch CPU speed using DFS */
+static int dfs_set_cpu_speed(int low_speed)
+{
+       if (low_speed == 0) {
+               /* ramping up, set voltage first */
+               pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, voltage_gpio, 0x05);
+               /* Make sure we sleep for at least 1ms */
+               local_delay(1);
+       }
+
+       /* set frequency */
+#ifdef CONFIG_6xx
+       low_choose_7447a_dfs(low_speed);
+#endif
+       udelay(100);
+
+       if (low_speed == 1) {
+               /* ramping down, set voltage last */
+               pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, voltage_gpio, 0x04);
+               local_delay(1);
+       }
+
+       return 0;
+}
+
+static unsigned int dfs_get_cpu_speed(void)
+{
+       if (mfspr(SPRN_HID1) & HID1_DFS)
+               return low_freq;
+       else
+               return hi_freq;
+}
+
+
+/* Switch CPU speed using slewing GPIOs
+ */
+static int gpios_set_cpu_speed(int low_speed)
+{
+       int gpio, timeout = 0;
+
+       /* If ramping up, set voltage first */
+       if (low_speed == 0) {
+               pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, voltage_gpio, 0x05);
+               /* Delay is way too big but it's ok, we schedule */
+               local_delay(10);
+       }
+
+       /* Set frequency */
+       gpio =  pmac_call_feature(PMAC_FTR_READ_GPIO, NULL, frequency_gpio, 0);
+       if (low_speed == ((gpio & 0x01) == 0))
+               goto skip;
+
+       pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, frequency_gpio,
+                         low_speed ? 0x04 : 0x05);
+       udelay(200);
+       do {
+               if (++timeout > 100)
+                       break;
+               local_delay(1);
+               gpio = pmac_call_feature(PMAC_FTR_READ_GPIO, NULL, slew_done_gpio, 0);
+       } while((gpio & 0x02) == 0);
+ skip:
+       /* If ramping down, set voltage last */
+       if (low_speed == 1) {
+               pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, voltage_gpio, 0x04);
+               /* Delay is way too big but it's ok, we schedule */
+               local_delay(10);
+       }
+
+#ifdef DEBUG_FREQ
+       debug_calc_bogomips();
+#endif
+
+       return 0;
+}
+
+/* Switch CPU speed under PMU control
+ */
+static int pmu_set_cpu_speed(int low_speed)
+{
+       struct adb_request req;
+       unsigned long save_l2cr;
+       unsigned long save_l3cr;
+       unsigned int pic_prio;
+       unsigned long flags;
+
+       preempt_disable();
+
+#ifdef DEBUG_FREQ
+       printk(KERN_DEBUG "HID1, before: %x\n", mfspr(SPRN_HID1));
+#endif
+       pmu_suspend();
+
+       /* Disable all interrupt sources on openpic */
+       pic_prio = mpic_cpu_get_priority();
+       mpic_cpu_set_priority(0xf);
+
+       /* Make sure the decrementer won't interrupt us */
+       asm volatile("mtdec %0" : : "r" (0x7fffffff));
+       /* Make sure any pending DEC interrupt occuring while we did
+        * the above didn't re-enable the DEC */
+       mb();
+       asm volatile("mtdec %0" : : "r" (0x7fffffff));
+
+       /* We can now disable MSR_EE */
+       local_irq_save(flags);
+
+       /* Giveup the FPU & vec */
+       enable_kernel_fp();
+
+#ifdef CONFIG_ALTIVEC
+       if (cpu_has_feature(CPU_FTR_ALTIVEC))
+               enable_kernel_altivec();
+#endif /* CONFIG_ALTIVEC */
+
+       /* Save & disable L2 and L3 caches */
+       save_l3cr = _get_L3CR();        /* (returns -1 if not available) */
+       save_l2cr = _get_L2CR();        /* (returns -1 if not available) */
+
+       /* Send the new speed command. My assumption is that this command
+        * will cause PLL_CFG[0..3] to be changed next time CPU goes to sleep
+        */
+       pmu_request(&req, NULL, 6, PMU_CPU_SPEED, 'W', 'O', 'O', 'F', low_speed);
+       while (!req.complete)
+               pmu_poll();
+
+       /* Prepare the northbridge for the speed transition */
+       pmac_call_feature(PMAC_FTR_SLEEP_STATE,NULL,1,1);
+
+       /* Call low level code to backup CPU state and recover from
+        * hardware reset
+        */
+       low_sleep_handler();
+
+       /* Restore the northbridge */
+       pmac_call_feature(PMAC_FTR_SLEEP_STATE,NULL,1,0);
+
+       /* Restore L2 cache */
+       if (save_l2cr != 0xffffffff && (save_l2cr & L2CR_L2E) != 0)
+               _set_L2CR(save_l2cr);
+       /* Restore L3 cache */
+       if (save_l3cr != 0xffffffff && (save_l3cr & L3CR_L3E) != 0)
+               _set_L3CR(save_l3cr);
+
+       /* Restore userland MMU context */
+       set_context(current->active_mm->context, current->active_mm->pgd);
+
+#ifdef DEBUG_FREQ
+       printk(KERN_DEBUG "HID1, after: %x\n", mfspr(SPRN_HID1));
+#endif
+
+       /* Restore low level PMU operations */
+       pmu_unlock();
+
+       /* Restore decrementer */
+       wakeup_decrementer();
+
+       /* Restore interrupts */
+       mpic_cpu_set_priority(pic_prio);
+
+       /* Let interrupts flow again ... */
+       local_irq_restore(flags);
+
+#ifdef DEBUG_FREQ
+       debug_calc_bogomips();
+#endif
+
+       pmu_resume();
+
+       preempt_enable();
+
+       return 0;
+}
+
+static int do_set_cpu_speed(int speed_mode, int notify)
+{
+       struct cpufreq_freqs freqs;
+       unsigned long l3cr;
+       static unsigned long prev_l3cr;
+
+       freqs.old = cur_freq;
+       freqs.new = (speed_mode == CPUFREQ_HIGH) ? hi_freq : low_freq;
+       freqs.cpu = smp_processor_id();
+
+       if (freqs.old == freqs.new)
+               return 0;
+
+       if (notify)
+               cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
+       if (speed_mode == CPUFREQ_LOW &&
+           cpu_has_feature(CPU_FTR_L3CR)) {
+               l3cr = _get_L3CR();
+               if (l3cr & L3CR_L3E) {
+                       prev_l3cr = l3cr;
+                       _set_L3CR(0);
+               }
+       }
+       set_speed_proc(speed_mode == CPUFREQ_LOW);
+       if (speed_mode == CPUFREQ_HIGH &&
+           cpu_has_feature(CPU_FTR_L3CR)) {
+               l3cr = _get_L3CR();
+               if ((prev_l3cr & L3CR_L3E) && l3cr != prev_l3cr)
+                       _set_L3CR(prev_l3cr);
+       }
+       if (notify)
+               cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
+       cur_freq = (speed_mode == CPUFREQ_HIGH) ? hi_freq : low_freq;
+
+       return 0;
+}
+
+static unsigned int pmac_cpufreq_get_speed(unsigned int cpu)
+{
+       return cur_freq;
+}
+
+static int pmac_cpufreq_verify(struct cpufreq_policy *policy)
+{
+       return cpufreq_frequency_table_verify(policy, pmac_cpu_freqs);
+}
+
+static int pmac_cpufreq_target(        struct cpufreq_policy *policy,
+                                       unsigned int target_freq,
+                                       unsigned int relation)
+{
+       unsigned int    newstate = 0;
+
+       if (cpufreq_frequency_table_target(policy, pmac_cpu_freqs,
+                       target_freq, relation, &newstate))
+               return -EINVAL;
+
+       return do_set_cpu_speed(newstate, 1);
+}
+
+unsigned int pmac_get_one_cpufreq(int i)
+{
+       /* Supports only one CPU for now */
+       return (i == 0) ? cur_freq : 0;
+}
+
+static int pmac_cpufreq_cpu_init(struct cpufreq_policy *policy)
+{
+       if (policy->cpu != 0)
+               return -ENODEV;
+
+       policy->governor = CPUFREQ_DEFAULT_GOVERNOR;
+       policy->cpuinfo.transition_latency      = CPUFREQ_ETERNAL;
+       policy->cur = cur_freq;
+
+       cpufreq_frequency_table_get_attr(pmac_cpu_freqs, policy->cpu);
+       return cpufreq_frequency_table_cpuinfo(policy, pmac_cpu_freqs);
+}
+
+static u32 read_gpio(struct device_node *np)
+{
+       u32 *reg = (u32 *)get_property(np, "reg", NULL);
+       u32 offset;
+
+       if (reg == NULL)
+               return 0;
+       /* That works for all keylargos but shall be fixed properly
+        * some day... The problem is that it seems we can't rely
+        * on the "reg" property of the GPIO nodes, they are either
+        * relative to the base of KeyLargo or to the base of the
+        * GPIO space, and the device-tree doesn't help.
+        */
+       offset = *reg;
+       if (offset < KEYLARGO_GPIO_LEVELS0)
+               offset += KEYLARGO_GPIO_LEVELS0;
+       return offset;
+}
+
+static int pmac_cpufreq_suspend(struct cpufreq_policy *policy, pm_message_t pmsg)
+{
+       /* Ok, this could be made a bit smarter, but let's be robust for now. We
+        * always force a speed change to high speed before sleep, to make sure
+        * we have appropriate voltage and/or bus speed for the wakeup process,
+        * and to make sure our loops_per_jiffies are "good enough", that is will
+        * not cause too short delays if we sleep in low speed and wake in high
+        * speed..
+        */
+       no_schedule = 1;
+       sleep_freq = cur_freq;
+       if (cur_freq == low_freq && !is_pmu_based)
+               do_set_cpu_speed(CPUFREQ_HIGH, 0);
+       return 0;
+}
+
+static int pmac_cpufreq_resume(struct cpufreq_policy *policy)
+{
+       /* If we resume, first check if we have a get() function */
+       if (get_speed_proc)
+               cur_freq = get_speed_proc();
+       else
+               cur_freq = 0;
+
+       /* We don't, hrm... we don't really know our speed here, best
+        * is that we force a switch to whatever it was, which is
+        * probably high speed due to our suspend() routine
+        */
+       do_set_cpu_speed(sleep_freq == low_freq ?
+                        CPUFREQ_LOW : CPUFREQ_HIGH, 0);
+
+       no_schedule = 0;
+       return 0;
+}
+
+static struct cpufreq_driver pmac_cpufreq_driver = {
+       .verify         = pmac_cpufreq_verify,
+       .target         = pmac_cpufreq_target,
+       .get            = pmac_cpufreq_get_speed,
+       .init           = pmac_cpufreq_cpu_init,
+       .suspend        = pmac_cpufreq_suspend,
+       .resume         = pmac_cpufreq_resume,
+       .flags          = CPUFREQ_PM_NO_WARN,
+       .attr           = pmac_cpu_freqs_attr,
+       .name           = "powermac",
+       .owner          = THIS_MODULE,
+};
+
+
+static int pmac_cpufreq_init_MacRISC3(struct device_node *cpunode)
+{
+       struct device_node *volt_gpio_np = of_find_node_by_name(NULL,
+                                                               "voltage-gpio");
+       struct device_node *freq_gpio_np = of_find_node_by_name(NULL,
+                                                               "frequency-gpio");
+       struct device_node *slew_done_gpio_np = of_find_node_by_name(NULL,
+                                                                    "slewing-done");
+       u32 *value;
+
+       /*
+        * Check to see if it's GPIO driven or PMU only
+        *
+        * The way we extract the GPIO address is slightly hackish, but it
+        * works well enough for now. We need to abstract the whole GPIO
+        * stuff sooner or later anyway
+        */
+
+       if (volt_gpio_np)
+               voltage_gpio = read_gpio(volt_gpio_np);
+       if (freq_gpio_np)
+               frequency_gpio = read_gpio(freq_gpio_np);
+       if (slew_done_gpio_np)
+               slew_done_gpio = read_gpio(slew_done_gpio_np);
+
+       /* If we use the frequency GPIOs, calculate the min/max speeds based
+        * on the bus frequencies
+        */
+       if (frequency_gpio && slew_done_gpio) {
+               int lenp, rc;
+               u32 *freqs, *ratio;
+
+               freqs = (u32 *)get_property(cpunode, "bus-frequencies", &lenp);
+               lenp /= sizeof(u32);
+               if (freqs == NULL || lenp != 2) {
+                       printk(KERN_ERR "cpufreq: bus-frequencies incorrect or missing\n");
+                       return 1;
+               }
+               ratio = (u32 *)get_property(cpunode, "processor-to-bus-ratio*2", NULL);
+               if (ratio == NULL) {
+                       printk(KERN_ERR "cpufreq: processor-to-bus-ratio*2 missing\n");
+                       return 1;
+               }
+
+               /* Get the min/max bus frequencies */
+               low_freq = min(freqs[0], freqs[1]);
+               hi_freq = max(freqs[0], freqs[1]);
+
+               /* Grrrr.. It _seems_ that the device-tree is lying on the low bus
+                * frequency, it claims it to be around 84Mhz on some models while
+                * it appears to be approx. 101Mhz on all. Let's hack around here...
+                * fortunately, we don't need to be too precise
+                */
+               if (low_freq < 98000000)
+                       low_freq = 101000000;
+                       
+               /* Convert those to CPU core clocks */
+               low_freq = (low_freq * (*ratio)) / 2000;
+               hi_freq = (hi_freq * (*ratio)) / 2000;
+
+               /* Now we get the frequencies, we read the GPIO to see what is out current
+                * speed
+                */
+               rc = pmac_call_feature(PMAC_FTR_READ_GPIO, NULL, frequency_gpio, 0);
+               cur_freq = (rc & 0x01) ? hi_freq : low_freq;
+
+               set_speed_proc = gpios_set_cpu_speed;
+               return 1;
+       }
+
+       /* If we use the PMU, look for the min & max frequencies in the
+        * device-tree
+        */
+       value = (u32 *)get_property(cpunode, "min-clock-frequency", NULL);
+       if (!value)
+               return 1;
+       low_freq = (*value) / 1000;
+       /* The PowerBook G4 12" (PowerBook6,1) has an error in the device-tree
+        * here */
+       if (low_freq < 100000)
+               low_freq *= 10;
+
+       value = (u32 *)get_property(cpunode, "max-clock-frequency", NULL);
+       if (!value)
+               return 1;
+       hi_freq = (*value) / 1000;
+       set_speed_proc = pmu_set_cpu_speed;
+       is_pmu_based = 1;
+
+       return 0;
+}
+
+static int pmac_cpufreq_init_7447A(struct device_node *cpunode)
+{
+       struct device_node *volt_gpio_np;
+
+       if (get_property(cpunode, "dynamic-power-step", NULL) == NULL)
+               return 1;
+
+       volt_gpio_np = of_find_node_by_name(NULL, "cpu-vcore-select");
+       if (volt_gpio_np)
+               voltage_gpio = read_gpio(volt_gpio_np);
+       if (!voltage_gpio){
+               printk(KERN_ERR "cpufreq: missing cpu-vcore-select gpio\n");
+               return 1;
+       }
+
+       /* OF only reports the high frequency */
+       hi_freq = cur_freq;
+       low_freq = cur_freq/2;
+
+       /* Read actual frequency from CPU */
+       cur_freq = dfs_get_cpu_speed();
+       set_speed_proc = dfs_set_cpu_speed;
+       get_speed_proc = dfs_get_cpu_speed;
+
+       return 0;
+}
+
+static int pmac_cpufreq_init_750FX(struct device_node *cpunode)
+{
+       struct device_node *volt_gpio_np;
+       u32 pvr, *value;
+
+       if (get_property(cpunode, "dynamic-power-step", NULL) == NULL)
+               return 1;
+
+       hi_freq = cur_freq;
+       value = (u32 *)get_property(cpunode, "reduced-clock-frequency", NULL);
+       if (!value)
+               return 1;
+       low_freq = (*value) / 1000;
+
+       volt_gpio_np = of_find_node_by_name(NULL, "cpu-vcore-select");
+       if (volt_gpio_np)
+               voltage_gpio = read_gpio(volt_gpio_np);
+
+       pvr = mfspr(SPRN_PVR);
+       has_cpu_l2lve = !((pvr & 0xf00) == 0x100);
+
+       set_speed_proc = cpu_750fx_cpu_speed;
+       get_speed_proc = cpu_750fx_get_cpu_speed;
+       cur_freq = cpu_750fx_get_cpu_speed();
+
+       return 0;
+}
+
+/* Currently, we support the following machines:
+ *
+ *  - Titanium PowerBook 1Ghz (PMU based, 667Mhz & 1Ghz)
+ *  - Titanium PowerBook 800 (PMU based, 667Mhz & 800Mhz)
+ *  - Titanium PowerBook 400 (PMU based, 300Mhz & 400Mhz)
+ *  - Titanium PowerBook 500 (PMU based, 300Mhz & 500Mhz)
+ *  - iBook2 500/600 (PMU based, 400Mhz & 500/600Mhz)
+ *  - iBook2 700 (CPU based, 400Mhz & 700Mhz, support low voltage)
+ *  - Recent MacRISC3 laptops
+ *  - All new machines with 7447A CPUs
+ */
+static int __init pmac_cpufreq_setup(void)
+{
+       struct device_node      *cpunode;
+       u32                     *value;
+
+       if (strstr(cmd_line, "nocpufreq"))
+               return 0;
+
+       /* Assume only one CPU */
+       cpunode = find_type_devices("cpu");
+       if (!cpunode)
+               goto out;
+
+       /* Get current cpu clock freq */
+       value = (u32 *)get_property(cpunode, "clock-frequency", NULL);
+       if (!value)
+               goto out;
+       cur_freq = (*value) / 1000;
+
+       /*  Check for 7447A based MacRISC3 */
+       if (machine_is_compatible("MacRISC3") &&
+           get_property(cpunode, "dynamic-power-step", NULL) &&
+           PVR_VER(mfspr(SPRN_PVR)) == 0x8003) {
+               pmac_cpufreq_init_7447A(cpunode);
+       /* Check for other MacRISC3 machines */
+       } else if (machine_is_compatible("PowerBook3,4") ||
+                  machine_is_compatible("PowerBook3,5") ||
+                  machine_is_compatible("MacRISC3")) {
+               pmac_cpufreq_init_MacRISC3(cpunode);
+       /* Else check for iBook2 500/600 */
+       } else if (machine_is_compatible("PowerBook4,1")) {
+               hi_freq = cur_freq;
+               low_freq = 400000;
+               set_speed_proc = pmu_set_cpu_speed;
+               is_pmu_based = 1;
+       }
+       /* Else check for TiPb 400 & 500 */
+       else if (machine_is_compatible("PowerBook3,2")) {
+               /* We only know about the 400 MHz and the 500Mhz model
+                * they both have 300 MHz as low frequency
+                */
+               if (cur_freq < 350000 || cur_freq > 550000)
+                       goto out;
+               hi_freq = cur_freq;
+               low_freq = 300000;
+               set_speed_proc = pmu_set_cpu_speed;
+               is_pmu_based = 1;
+       }
+       /* Else check for 750FX */
+       else if (PVR_VER(mfspr(SPRN_PVR)) == 0x7000)
+               pmac_cpufreq_init_750FX(cpunode);
+out:
+       if (set_speed_proc == NULL)
+               return -ENODEV;
+
+       pmac_cpu_freqs[CPUFREQ_LOW].frequency = low_freq;
+       pmac_cpu_freqs[CPUFREQ_HIGH].frequency = hi_freq;
+
+       printk(KERN_INFO "Registering PowerMac CPU frequency driver\n");
+       printk(KERN_INFO "Low: %d Mhz, High: %d Mhz, Boot: %d Mhz\n",
+              low_freq/1000, hi_freq/1000, cur_freq/1000);
+
+       return cpufreq_register_driver(&pmac_cpufreq_driver);
+}
+
+module_init(pmac_cpufreq_setup);
+
diff --git a/arch/powerpc/platforms/powermac/pmac_feature.c b/arch/powerpc/platforms/powermac/pmac_feature.c
new file mode 100644 (file)
index 0000000..2cba670
--- /dev/null
@@ -0,0 +1,3062 @@
+/*
+ *  arch/ppc/platforms/pmac_feature.c
+ *
+ *  Copyright (C) 1996-2001 Paul Mackerras (paulus@cs.anu.edu.au)
+ *                          Ben. Herrenschmidt (benh@kernel.crashing.org)
+ *
+ *  This program is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU General Public License
+ *  as published by the Free Software Foundation; either version
+ *  2 of the License, or (at your option) any later version.
+ *
+ *  TODO:
+ *
+ *   - Replace mdelay with some schedule loop if possible
+ *   - Shorten some obfuscated delays on some routines (like modem
+ *     power)
+ *   - Refcount some clocks (see darwin)
+ *   - Split split split...
+ *
+ */
+#include <linux/config.h>
+#include <linux/types.h>
+#include <linux/init.h>
+#include <linux/delay.h>
+#include <linux/kernel.h>
+#include <linux/sched.h>
+#include <linux/spinlock.h>
+#include <linux/adb.h>
+#include <linux/pmu.h>
+#include <linux/ioport.h>
+#include <linux/pci.h>
+#include <asm/sections.h>
+#include <asm/errno.h>
+#include <asm/ohare.h>
+#include <asm/heathrow.h>
+#include <asm/keylargo.h>
+#include <asm/uninorth.h>
+#include <asm/io.h>
+#include <asm/prom.h>
+#include <asm/machdep.h>
+#include <asm/pmac_feature.h>
+#include <asm/dbdma.h>
+#include <asm/pci-bridge.h>
+#include <asm/pmac_low_i2c.h>
+
+#undef DEBUG_FEATURE
+
+#ifdef DEBUG_FEATURE
+#define DBG(fmt...) printk(KERN_DEBUG fmt)
+#else
+#define DBG(fmt...)
+#endif
+
+#ifdef CONFIG_6xx
+extern int powersave_lowspeed;
+#endif
+
+extern int powersave_nap;
+extern struct device_node *k2_skiplist[2];
+
+
+/*
+ * We use a single global lock to protect accesses. Each driver has
+ * to take care of its own locking
+ */
+static DEFINE_SPINLOCK(feature_lock);
+
+#define LOCK(flags)    spin_lock_irqsave(&feature_lock, flags);
+#define UNLOCK(flags)  spin_unlock_irqrestore(&feature_lock, flags);
+
+
+/*
+ * Instance of some macio stuffs
+ */
+struct macio_chip macio_chips[MAX_MACIO_CHIPS];
+
+struct macio_chip *macio_find(struct device_node *child, int type)
+{
+       while(child) {
+               int     i;
+
+               for (i=0; i < MAX_MACIO_CHIPS && macio_chips[i].of_node; i++)
+                       if (child == macio_chips[i].of_node &&
+                           (!type || macio_chips[i].type == type))
+                               return &macio_chips[i];
+               child = child->parent;
+       }
+       return NULL;
+}
+EXPORT_SYMBOL_GPL(macio_find);
+
+static const char *macio_names[] =
+{
+       "Unknown",
+       "Grand Central",
+       "OHare",
+       "OHareII",
+       "Heathrow",
+       "Gatwick",
+       "Paddington",
+       "Keylargo",
+       "Pangea",
+       "Intrepid",
+       "K2"
+};
+
+
+
+/*
+ * Uninorth reg. access. Note that Uni-N regs are big endian
+ */
+
+#define UN_REG(r)      (uninorth_base + ((r) >> 2))
+#define UN_IN(r)       (in_be32(UN_REG(r)))
+#define UN_OUT(r,v)    (out_be32(UN_REG(r), (v)))
+#define UN_BIS(r,v)    (UN_OUT((r), UN_IN(r) | (v)))
+#define UN_BIC(r,v)    (UN_OUT((r), UN_IN(r) & ~(v)))
+
+static struct device_node *uninorth_node;
+static u32 __iomem *uninorth_base;
+static u32 uninorth_rev;
+static int uninorth_u3;
+static void __iomem *u3_ht;
+
+/*
+ * For each motherboard family, we have a table of functions pointers
+ * that handle the various features.
+ */
+
+typedef long (*feature_call)(struct device_node *node, long param, long value);
+
+struct feature_table_entry {
+       unsigned int    selector;
+       feature_call    function;
+};
+
+struct pmac_mb_def
+{
+       const char*                     model_string;
+       const char*                     model_name;
+       int                             model_id;
+       struct feature_table_entry*     features;
+       unsigned long                   board_flags;
+};
+static struct pmac_mb_def pmac_mb;
+
+/*
+ * Here are the chip specific feature functions
+ */
+
+static inline int simple_feature_tweak(struct device_node *node, int type,
+                                      int reg, u32 mask, int value)
+{
+       struct macio_chip*      macio;
+       unsigned long           flags;
+
+       macio = macio_find(node, type);
+       if (!macio)
+               return -ENODEV;
+       LOCK(flags);
+       if (value)
+               MACIO_BIS(reg, mask);
+       else
+               MACIO_BIC(reg, mask);
+       (void)MACIO_IN32(reg);
+       UNLOCK(flags);
+
+       return 0;
+}
+
+#ifndef CONFIG_POWER4
+
+static long ohare_htw_scc_enable(struct device_node *node, long param,
+                                long value)
+{
+       struct macio_chip*      macio;
+       unsigned long           chan_mask;
+       unsigned long           fcr;
+       unsigned long           flags;
+       int                     htw, trans;
+       unsigned long           rmask;
+
+       macio = macio_find(node, 0);
+       if (!macio)
+               return -ENODEV;
+       if (!strcmp(node->name, "ch-a"))
+               chan_mask = MACIO_FLAG_SCCA_ON;
+       else if (!strcmp(node->name, "ch-b"))
+               chan_mask = MACIO_FLAG_SCCB_ON;
+       else
+               return -ENODEV;
+
+       htw = (macio->type == macio_heathrow || macio->type == macio_paddington
+               || macio->type == macio_gatwick);
+       /* On these machines, the HRW_SCC_TRANS_EN_N bit mustn't be touched */
+       trans = (pmac_mb.model_id != PMAC_TYPE_YOSEMITE &&
+                pmac_mb.model_id != PMAC_TYPE_YIKES);
+       if (value) {
+#ifdef CONFIG_ADB_PMU
+               if ((param & 0xfff) == PMAC_SCC_IRDA)
+                       pmu_enable_irled(1);
+#endif /* CONFIG_ADB_PMU */
+               LOCK(flags);
+               fcr = MACIO_IN32(OHARE_FCR);
+               /* Check if scc cell need enabling */
+               if (!(fcr & OH_SCC_ENABLE)) {
+                       fcr |= OH_SCC_ENABLE;
+                       if (htw) {
+                               /* Side effect: this will also power up the
+                                * modem, but it's too messy to figure out on which
+                                * ports this controls the tranceiver and on which
+                                * it controls the modem
+                                */
+                               if (trans)
+                                       fcr &= ~HRW_SCC_TRANS_EN_N;
+                               MACIO_OUT32(OHARE_FCR, fcr);
+                               fcr |= (rmask = HRW_RESET_SCC);
+                               MACIO_OUT32(OHARE_FCR, fcr);
+                       } else {
+                               fcr |= (rmask = OH_SCC_RESET);
+                               MACIO_OUT32(OHARE_FCR, fcr);
+                       }
+                       UNLOCK(flags);
+                       (void)MACIO_IN32(OHARE_FCR);
+                       mdelay(15);
+                       LOCK(flags);
+                       fcr &= ~rmask;
+                       MACIO_OUT32(OHARE_FCR, fcr);
+               }
+               if (chan_mask & MACIO_FLAG_SCCA_ON)
+                       fcr |= OH_SCCA_IO;
+               if (chan_mask & MACIO_FLAG_SCCB_ON)
+                       fcr |= OH_SCCB_IO;
+               MACIO_OUT32(OHARE_FCR, fcr);
+               macio->flags |= chan_mask;
+               UNLOCK(flags);
+               if (param & PMAC_SCC_FLAG_XMON)
+                       macio->flags |= MACIO_FLAG_SCC_LOCKED;
+       } else {
+               if (macio->flags & MACIO_FLAG_SCC_LOCKED)
+                       return -EPERM;
+               LOCK(flags);
+               fcr = MACIO_IN32(OHARE_FCR);
+               if (chan_mask & MACIO_FLAG_SCCA_ON)
+                       fcr &= ~OH_SCCA_IO;
+               if (chan_mask & MACIO_FLAG_SCCB_ON)
+                       fcr &= ~OH_SCCB_IO;
+               MACIO_OUT32(OHARE_FCR, fcr);
+               if ((fcr & (OH_SCCA_IO | OH_SCCB_IO)) == 0) {
+                       fcr &= ~OH_SCC_ENABLE;
+                       if (htw && trans)
+                               fcr |= HRW_SCC_TRANS_EN_N;
+                       MACIO_OUT32(OHARE_FCR, fcr);
+               }
+               macio->flags &= ~(chan_mask);
+               UNLOCK(flags);
+               mdelay(10);
+#ifdef CONFIG_ADB_PMU
+               if ((param & 0xfff) == PMAC_SCC_IRDA)
+                       pmu_enable_irled(0);
+#endif /* CONFIG_ADB_PMU */
+       }
+       return 0;
+}
+
+static long ohare_floppy_enable(struct device_node *node, long param,
+                               long value)
+{
+       return simple_feature_tweak(node, macio_ohare,
+               OHARE_FCR, OH_FLOPPY_ENABLE, value);
+}
+
+static long ohare_mesh_enable(struct device_node *node, long param, long value)
+{
+       return simple_feature_tweak(node, macio_ohare,
+               OHARE_FCR, OH_MESH_ENABLE, value);
+}
+
+static long ohare_ide_enable(struct device_node *node, long param, long value)
+{
+       switch(param) {
+       case 0:
+               /* For some reason, setting the bit in set_initial_features()
+                * doesn't stick. I'm still investigating... --BenH.
+                */
+               if (value)
+                       simple_feature_tweak(node, macio_ohare,
+                               OHARE_FCR, OH_IOBUS_ENABLE, 1);
+               return simple_feature_tweak(node, macio_ohare,
+                       OHARE_FCR, OH_IDE0_ENABLE, value);
+       case 1:
+               return simple_feature_tweak(node, macio_ohare,
+                       OHARE_FCR, OH_BAY_IDE_ENABLE, value);
+       default:
+               return -ENODEV;
+       }
+}
+
+static long ohare_ide_reset(struct device_node *node, long param, long value)
+{
+       switch(param) {
+       case 0:
+               return simple_feature_tweak(node, macio_ohare,
+                       OHARE_FCR, OH_IDE0_RESET_N, !value);
+       case 1:
+               return simple_feature_tweak(node, macio_ohare,
+                       OHARE_FCR, OH_IDE1_RESET_N, !value);
+       default:
+               return -ENODEV;
+       }
+}
+
+static long ohare_sleep_state(struct device_node *node, long param, long value)
+{
+       struct macio_chip*      macio = &macio_chips[0];
+
+       if ((pmac_mb.board_flags & PMAC_MB_CAN_SLEEP) == 0)
+               return -EPERM;
+       if (value == 1) {
+               MACIO_BIC(OHARE_FCR, OH_IOBUS_ENABLE);
+       } else if (value == 0) {
+               MACIO_BIS(OHARE_FCR, OH_IOBUS_ENABLE);
+       }
+
+       return 0;
+}
+
+static long heathrow_modem_enable(struct device_node *node, long param,
+                                 long value)
+{
+       struct macio_chip*      macio;
+       u8                      gpio;
+       unsigned long           flags;
+
+       macio = macio_find(node, macio_unknown);
+       if (!macio)
+               return -ENODEV;
+       gpio = MACIO_IN8(HRW_GPIO_MODEM_RESET) & ~1;
+       if (!value) {
+               LOCK(flags);
+               MACIO_OUT8(HRW_GPIO_MODEM_RESET, gpio);
+               UNLOCK(flags);
+               (void)MACIO_IN8(HRW_GPIO_MODEM_RESET);
+               mdelay(250);
+       }
+       if (pmac_mb.model_id != PMAC_TYPE_YOSEMITE &&
+           pmac_mb.model_id != PMAC_TYPE_YIKES) {
+               LOCK(flags);
+               if (value)
+                       MACIO_BIC(HEATHROW_FCR, HRW_SCC_TRANS_EN_N);
+               else
+                       MACIO_BIS(HEATHROW_FCR, HRW_SCC_TRANS_EN_N);
+               UNLOCK(flags);
+               (void)MACIO_IN32(HEATHROW_FCR);
+               mdelay(250);
+       }
+       if (value) {
+               LOCK(flags);
+               MACIO_OUT8(HRW_GPIO_MODEM_RESET, gpio | 1);
+               (void)MACIO_IN8(HRW_GPIO_MODEM_RESET);
+               UNLOCK(flags); mdelay(250); LOCK(flags);
+               MACIO_OUT8(HRW_GPIO_MODEM_RESET, gpio);
+               (void)MACIO_IN8(HRW_GPIO_MODEM_RESET);
+               UNLOCK(flags); mdelay(250); LOCK(flags);
+               MACIO_OUT8(HRW_GPIO_MODEM_RESET, gpio | 1);
+               (void)MACIO_IN8(HRW_GPIO_MODEM_RESET);
+               UNLOCK(flags); mdelay(250);
+       }
+       return 0;
+}
+
+static long heathrow_floppy_enable(struct device_node *node, long param,
+                                  long value)
+{
+       return simple_feature_tweak(node, macio_unknown,
+               HEATHROW_FCR,
+               HRW_SWIM_ENABLE|HRW_BAY_FLOPPY_ENABLE,
+               value);
+}
+
+static long heathrow_mesh_enable(struct device_node *node, long param,
+                                long value)
+{
+       struct macio_chip*      macio;
+       unsigned long           flags;
+
+       macio = macio_find(node, macio_unknown);
+       if (!macio)
+               return -ENODEV;
+       LOCK(flags);
+       /* Set clear mesh cell enable */
+       if (value)
+               MACIO_BIS(HEATHROW_FCR, HRW_MESH_ENABLE);
+       else
+               MACIO_BIC(HEATHROW_FCR, HRW_MESH_ENABLE);
+       (void)MACIO_IN32(HEATHROW_FCR);
+       udelay(10);
+       /* Set/Clear termination power */
+       if (value)
+               MACIO_BIC(HEATHROW_MBCR, 0x04000000);
+       else
+               MACIO_BIS(HEATHROW_MBCR, 0x04000000);
+       (void)MACIO_IN32(HEATHROW_MBCR);
+       udelay(10);
+       UNLOCK(flags);
+
+       return 0;
+}
+
+static long heathrow_ide_enable(struct device_node *node, long param,
+                               long value)
+{
+       switch(param) {
+       case 0:
+               return simple_feature_tweak(node, macio_unknown,
+                       HEATHROW_FCR, HRW_IDE0_ENABLE, value);
+       case 1:
+               return simple_feature_tweak(node, macio_unknown,
+                       HEATHROW_FCR, HRW_BAY_IDE_ENABLE, value);
+       default:
+               return -ENODEV;
+       }
+}
+
+static long heathrow_ide_reset(struct device_node *node, long param,
+                              long value)
+{
+       switch(param) {
+       case 0:
+               return simple_feature_tweak(node, macio_unknown,
+                       HEATHROW_FCR, HRW_IDE0_RESET_N, !value);
+       case 1:
+               return simple_feature_tweak(node, macio_unknown,
+                       HEATHROW_FCR, HRW_IDE1_RESET_N, !value);
+       default:
+               return -ENODEV;
+       }
+}
+
+static long heathrow_bmac_enable(struct device_node *node, long param,
+                                long value)
+{
+       struct macio_chip*      macio;
+       unsigned long           flags;
+
+       macio = macio_find(node, 0);
+       if (!macio)
+               return -ENODEV;
+       if (value) {
+               LOCK(flags);
+               MACIO_BIS(HEATHROW_FCR, HRW_BMAC_IO_ENABLE);
+               MACIO_BIS(HEATHROW_FCR, HRW_BMAC_RESET);
+               UNLOCK(flags);
+               (void)MACIO_IN32(HEATHROW_FCR);
+               mdelay(10);
+               LOCK(flags);
+               MACIO_BIC(HEATHROW_FCR, HRW_BMAC_RESET);
+               UNLOCK(flags);
+               (void)MACIO_IN32(HEATHROW_FCR);
+               mdelay(10);
+       } else {
+               LOCK(flags);
+               MACIO_BIC(HEATHROW_FCR, HRW_BMAC_IO_ENABLE);
+               UNLOCK(flags);
+       }
+       return 0;
+}
+
+static long heathrow_sound_enable(struct device_node *node, long param,
+                                 long value)
+{
+       struct macio_chip*      macio;
+       unsigned long           flags;
+
+       /* B&W G3 and Yikes don't support that properly (the
+        * sound appear to never come back after beeing shut down).
+        */
+       if (pmac_mb.model_id == PMAC_TYPE_YOSEMITE ||
+           pmac_mb.model_id == PMAC_TYPE_YIKES)
+               return 0;
+
+       macio = macio_find(node, 0);
+       if (!macio)
+               return -ENODEV;
+       if (value) {
+               LOCK(flags);
+               MACIO_BIS(HEATHROW_FCR, HRW_SOUND_CLK_ENABLE);
+               MACIO_BIC(HEATHROW_FCR, HRW_SOUND_POWER_N);
+               UNLOCK(flags);
+               (void)MACIO_IN32(HEATHROW_FCR);
+       } else {
+               LOCK(flags);
+               MACIO_BIS(HEATHROW_FCR, HRW_SOUND_POWER_N);
+               MACIO_BIC(HEATHROW_FCR, HRW_SOUND_CLK_ENABLE);
+               UNLOCK(flags);
+       }
+       return 0;
+}
+
+static u32 save_fcr[6];
+static u32 save_mbcr;
+static u32 save_gpio_levels[2];
+static u8 save_gpio_extint[KEYLARGO_GPIO_EXTINT_CNT];
+static u8 save_gpio_normal[KEYLARGO_GPIO_CNT];
+static u32 save_unin_clock_ctl;
+static struct dbdma_regs save_dbdma[13];
+static struct dbdma_regs save_alt_dbdma[13];
+
+static void dbdma_save(struct macio_chip *macio, struct dbdma_regs *save)
+{
+       int i;
+
+       /* Save state & config of DBDMA channels */
+       for (i = 0; i < 13; i++) {
+               volatile struct dbdma_regs __iomem * chan = (void __iomem *)
+                       (macio->base + ((0x8000+i*0x100)>>2));
+               save[i].cmdptr_hi = in_le32(&chan->cmdptr_hi);
+               save[i].cmdptr = in_le32(&chan->cmdptr);
+               save[i].intr_sel = in_le32(&chan->intr_sel);
+               save[i].br_sel = in_le32(&chan->br_sel);
+               save[i].wait_sel = in_le32(&chan->wait_sel);
+       }
+}
+
+static void dbdma_restore(struct macio_chip *macio, struct dbdma_regs *save)
+{
+       int i;
+
+       /* Save state & config of DBDMA channels */
+       for (i = 0; i < 13; i++) {
+               volatile struct dbdma_regs __iomem * chan = (void __iomem *)
+                       (macio->base + ((0x8000+i*0x100)>>2));
+               out_le32(&chan->control, (ACTIVE|DEAD|WAKE|FLUSH|PAUSE|RUN)<<16);
+               while (in_le32(&chan->status) & ACTIVE)
+                       mb();
+               out_le32(&chan->cmdptr_hi, save[i].cmdptr_hi);
+               out_le32(&chan->cmdptr, save[i].cmdptr);
+               out_le32(&chan->intr_sel, save[i].intr_sel);
+               out_le32(&chan->br_sel, save[i].br_sel);
+               out_le32(&chan->wait_sel, save[i].wait_sel);
+       }
+}
+
+static void heathrow_sleep(struct macio_chip *macio, int secondary)
+{
+       if (secondary) {
+               dbdma_save(macio, save_alt_dbdma);
+               save_fcr[2] = MACIO_IN32(0x38);
+               save_fcr[3] = MACIO_IN32(0x3c);
+       } else {
+               dbdma_save(macio, save_dbdma);
+               save_fcr[0] = MACIO_IN32(0x38);
+               save_fcr[1] = MACIO_IN32(0x3c);
+               save_mbcr = MACIO_IN32(0x34);
+               /* Make sure sound is shut down */
+               MACIO_BIS(HEATHROW_FCR, HRW_SOUND_POWER_N);
+               MACIO_BIC(HEATHROW_FCR, HRW_SOUND_CLK_ENABLE);
+               /* This seems to be necessary as well or the fan
+                * keeps coming up and battery drains fast */
+               MACIO_BIC(HEATHROW_FCR, HRW_IOBUS_ENABLE);
+               MACIO_BIC(HEATHROW_FCR, HRW_IDE0_RESET_N);
+               /* Make sure eth is down even if module or sleep
+                * won't work properly */
+               MACIO_BIC(HEATHROW_FCR, HRW_BMAC_IO_ENABLE | HRW_BMAC_RESET);
+       }
+       /* Make sure modem is shut down */
+       MACIO_OUT8(HRW_GPIO_MODEM_RESET,
+               MACIO_IN8(HRW_GPIO_MODEM_RESET) & ~1);
+       MACIO_BIS(HEATHROW_FCR, HRW_SCC_TRANS_EN_N);
+       MACIO_BIC(HEATHROW_FCR, OH_SCCA_IO|OH_SCCB_IO|HRW_SCC_ENABLE);
+
+       /* Let things settle */
+       (void)MACIO_IN32(HEATHROW_FCR);
+}
+
+static void heathrow_wakeup(struct macio_chip *macio, int secondary)
+{
+       if (secondary) {
+               MACIO_OUT32(0x38, save_fcr[2]);
+               (void)MACIO_IN32(0x38);
+               mdelay(1);
+               MACIO_OUT32(0x3c, save_fcr[3]);
+               (void)MACIO_IN32(0x38);
+               mdelay(10);
+               dbdma_restore(macio, save_alt_dbdma);
+       } else {
+               MACIO_OUT32(0x38, save_fcr[0] | HRW_IOBUS_ENABLE);
+               (void)MACIO_IN32(0x38);
+               mdelay(1);
+               MACIO_OUT32(0x3c, save_fcr[1]);
+               (void)MACIO_IN32(0x38);
+               mdelay(1);
+               MACIO_OUT32(0x34, save_mbcr);
+               (void)MACIO_IN32(0x38);
+               mdelay(10);
+               dbdma_restore(macio, save_dbdma);
+       }
+}
+
+static long heathrow_sleep_state(struct device_node *node, long param,
+                                long value)
+{
+       if ((pmac_mb.board_flags & PMAC_MB_CAN_SLEEP) == 0)
+               return -EPERM;
+       if (value == 1) {
+               if (macio_chips[1].type == macio_gatwick)
+                       heathrow_sleep(&macio_chips[0], 1);
+               heathrow_sleep(&macio_chips[0], 0);
+       } else if (value == 0) {
+               heathrow_wakeup(&macio_chips[0], 0);
+               if (macio_chips[1].type == macio_gatwick)
+                       heathrow_wakeup(&macio_chips[0], 1);
+       }
+       return 0;
+}
+
+static long core99_scc_enable(struct device_node *node, long param, long value)
+{
+       struct macio_chip*      macio;
+       unsigned long           flags;
+       unsigned long           chan_mask;
+       u32                     fcr;
+
+       macio = macio_find(node, 0);
+       if (!macio)
+               return -ENODEV;
+       if (!strcmp(node->name, "ch-a"))
+               chan_mask = MACIO_FLAG_SCCA_ON;
+       else if (!strcmp(node->name, "ch-b"))
+               chan_mask = MACIO_FLAG_SCCB_ON;
+       else
+               return -ENODEV;
+
+       if (value) {
+               int need_reset_scc = 0;
+               int need_reset_irda = 0;
+
+               LOCK(flags);
+               fcr = MACIO_IN32(KEYLARGO_FCR0);
+               /* Check if scc cell need enabling */
+               if (!(fcr & KL0_SCC_CELL_ENABLE)) {
+                       fcr |= KL0_SCC_CELL_ENABLE;
+                       need_reset_scc = 1;
+               }
+               if (chan_mask & MACIO_FLAG_SCCA_ON) {
+                       fcr |= KL0_SCCA_ENABLE;
+                       /* Don't enable line drivers for I2S modem */
+                       if ((param & 0xfff) == PMAC_SCC_I2S1)
+                               fcr &= ~KL0_SCC_A_INTF_ENABLE;
+                       else
+                               fcr |= KL0_SCC_A_INTF_ENABLE;
+               }
+               if (chan_mask & MACIO_FLAG_SCCB_ON) {
+                       fcr |= KL0_SCCB_ENABLE;
+                       /* Perform irda specific inits */
+                       if ((param & 0xfff) == PMAC_SCC_IRDA) {
+                               fcr &= ~KL0_SCC_B_INTF_ENABLE;
+                               fcr |= KL0_IRDA_ENABLE;
+                               fcr |= KL0_IRDA_CLK32_ENABLE | KL0_IRDA_CLK19_ENABLE;
+                               fcr |= KL0_IRDA_SOURCE1_SEL;
+                               fcr &= ~(KL0_IRDA_FAST_CONNECT|KL0_IRDA_DEFAULT1|KL0_IRDA_DEFAULT0);
+                               fcr &= ~(KL0_IRDA_SOURCE2_SEL|KL0_IRDA_HIGH_BAND);
+                               need_reset_irda = 1;
+                       } else
+                               fcr |= KL0_SCC_B_INTF_ENABLE;
+               }
+               MACIO_OUT32(KEYLARGO_FCR0, fcr);
+               macio->flags |= chan_mask;
+               if (need_reset_scc)  {
+                       MACIO_BIS(KEYLARGO_FCR0, KL0_SCC_RESET);
+                       (void)MACIO_IN32(KEYLARGO_FCR0);
+                       UNLOCK(flags);
+                       mdelay(15);
+                       LOCK(flags);
+                       MACIO_BIC(KEYLARGO_FCR0, KL0_SCC_RESET);
+               }
+               if (need_reset_irda)  {
+                       MACIO_BIS(KEYLARGO_FCR0, KL0_IRDA_RESET);
+                       (void)MACIO_IN32(KEYLARGO_FCR0);
+                       UNLOCK(flags);
+                       mdelay(15);
+                       LOCK(flags);
+                       MACIO_BIC(KEYLARGO_FCR0, KL0_IRDA_RESET);
+               }
+               UNLOCK(flags);
+               if (param & PMAC_SCC_FLAG_XMON)
+                       macio->flags |= MACIO_FLAG_SCC_LOCKED;
+       } else {
+               if (macio->flags & MACIO_FLAG_SCC_LOCKED)
+                       return -EPERM;
+               LOCK(flags);
+               fcr = MACIO_IN32(KEYLARGO_FCR0);
+               if (chan_mask & MACIO_FLAG_SCCA_ON)
+                       fcr &= ~KL0_SCCA_ENABLE;
+               if (chan_mask & MACIO_FLAG_SCCB_ON) {
+                       fcr &= ~KL0_SCCB_ENABLE;
+                       /* Perform irda specific clears */
+                       if ((param & 0xfff) == PMAC_SCC_IRDA) {
+                               fcr &= ~KL0_IRDA_ENABLE;
+                               fcr &= ~(KL0_IRDA_CLK32_ENABLE | KL0_IRDA_CLK19_ENABLE);
+                               fcr &= ~(KL0_IRDA_FAST_CONNECT|KL0_IRDA_DEFAULT1|KL0_IRDA_DEFAULT0);
+                               fcr &= ~(KL0_IRDA_SOURCE1_SEL|KL0_IRDA_SOURCE2_SEL|KL0_IRDA_HIGH_BAND);
+                       }
+               }
+               MACIO_OUT32(KEYLARGO_FCR0, fcr);
+               if ((fcr & (KL0_SCCA_ENABLE | KL0_SCCB_ENABLE)) == 0) {
+                       fcr &= ~KL0_SCC_CELL_ENABLE;
+                       MACIO_OUT32(KEYLARGO_FCR0, fcr);
+               }
+               macio->flags &= ~(chan_mask);
+               UNLOCK(flags);
+               mdelay(10);
+       }
+       return 0;
+}
+
+static long
+core99_modem_enable(struct device_node *node, long param, long value)
+{
+       struct macio_chip*      macio;
+       u8                      gpio;
+       unsigned long           flags;
+
+       /* Hack for internal USB modem */
+       if (node == NULL) {
+               if (macio_chips[0].type != macio_keylargo)
+                       return -ENODEV;
+               node = macio_chips[0].of_node;
+       }
+       macio = macio_find(node, 0);
+       if (!macio)
+               return -ENODEV;
+       gpio = MACIO_IN8(KL_GPIO_MODEM_RESET);
+       gpio |= KEYLARGO_GPIO_OUTPUT_ENABLE;
+       gpio &= ~KEYLARGO_GPIO_OUTOUT_DATA;
+
+       if (!value) {
+               LOCK(flags);
+               MACIO_OUT8(KL_GPIO_MODEM_RESET, gpio);
+               UNLOCK(flags);
+               (void)MACIO_IN8(KL_GPIO_MODEM_RESET);
+               mdelay(250);
+       }
+       LOCK(flags);
+       if (value) {
+               MACIO_BIC(KEYLARGO_FCR2, KL2_ALT_DATA_OUT);
+               UNLOCK(flags);
+               (void)MACIO_IN32(KEYLARGO_FCR2);
+               mdelay(250);
+       } else {
+               MACIO_BIS(KEYLARGO_FCR2, KL2_ALT_DATA_OUT);
+               UNLOCK(flags);
+       }
+       if (value) {
+               LOCK(flags);
+               MACIO_OUT8(KL_GPIO_MODEM_RESET, gpio | KEYLARGO_GPIO_OUTOUT_DATA);
+               (void)MACIO_IN8(KL_GPIO_MODEM_RESET);
+               UNLOCK(flags); mdelay(250); LOCK(flags);
+               MACIO_OUT8(KL_GPIO_MODEM_RESET, gpio);
+               (void)MACIO_IN8(KL_GPIO_MODEM_RESET);
+               UNLOCK(flags); mdelay(250); LOCK(flags);
+               MACIO_OUT8(KL_GPIO_MODEM_RESET, gpio | KEYLARGO_GPIO_OUTOUT_DATA);
+               (void)MACIO_IN8(KL_GPIO_MODEM_RESET);
+               UNLOCK(flags); mdelay(250);
+       }
+       return 0;
+}
+
+static long
+pangea_modem_enable(struct device_node *node, long param, long value)
+{
+       struct macio_chip*      macio;
+       u8                      gpio;
+       unsigned long           flags;
+
+       /* Hack for internal USB modem */
+       if (node == NULL) {
+               if (macio_chips[0].type != macio_pangea &&
+                   macio_chips[0].type != macio_intrepid)
+                       return -ENODEV;
+               node = macio_chips[0].of_node;
+       }
+       macio = macio_find(node, 0);
+       if (!macio)
+               return -ENODEV;
+       gpio = MACIO_IN8(KL_GPIO_MODEM_RESET);
+       gpio |= KEYLARGO_GPIO_OUTPUT_ENABLE;
+       gpio &= ~KEYLARGO_GPIO_OUTOUT_DATA;
+
+       if (!value) {
+               LOCK(flags);
+               MACIO_OUT8(KL_GPIO_MODEM_RESET, gpio);
+               UNLOCK(flags);
+               (void)MACIO_IN8(KL_GPIO_MODEM_RESET);
+               mdelay(250);
+       }
+       LOCK(flags);
+       if (value) {
+               MACIO_OUT8(KL_GPIO_MODEM_POWER,
+                       KEYLARGO_GPIO_OUTPUT_ENABLE);
+               UNLOCK(flags);
+               (void)MACIO_IN32(KEYLARGO_FCR2);
+               mdelay(250);
+       } else {
+               MACIO_OUT8(KL_GPIO_MODEM_POWER,
+                       KEYLARGO_GPIO_OUTPUT_ENABLE | KEYLARGO_GPIO_OUTOUT_DATA);
+               UNLOCK(flags);
+       }
+       if (value) {
+               LOCK(flags);
+               MACIO_OUT8(KL_GPIO_MODEM_RESET, gpio | KEYLARGO_GPIO_OUTOUT_DATA);
+               (void)MACIO_IN8(KL_GPIO_MODEM_RESET);
+               UNLOCK(flags); mdelay(250); LOCK(flags);
+               MACIO_OUT8(KL_GPIO_MODEM_RESET, gpio);
+               (void)MACIO_IN8(KL_GPIO_MODEM_RESET);
+               UNLOCK(flags); mdelay(250); LOCK(flags);
+               MACIO_OUT8(KL_GPIO_MODEM_RESET, gpio | KEYLARGO_GPIO_OUTOUT_DATA);
+               (void)MACIO_IN8(KL_GPIO_MODEM_RESET);
+               UNLOCK(flags); mdelay(250);
+       }
+       return 0;
+}
+
+static long
+core99_ata100_enable(struct device_node *node, long value)
+{
+       unsigned long flags;
+       struct pci_dev *pdev = NULL;
+       u8 pbus, pid;
+
+       if (uninorth_rev < 0x24)
+               return -ENODEV;
+
+       LOCK(flags);
+       if (value)
+               UN_BIS(UNI_N_CLOCK_CNTL, UNI_N_CLOCK_CNTL_ATA100);
+       else
+               UN_BIC(UNI_N_CLOCK_CNTL, UNI_N_CLOCK_CNTL_ATA100);
+       (void)UN_IN(UNI_N_CLOCK_CNTL);
+       UNLOCK(flags);
+       udelay(20);
+
+       if (value) {
+               if (pci_device_from_OF_node(node, &pbus, &pid) == 0)
+                       pdev = pci_find_slot(pbus, pid);
+               if (pdev == NULL)
+                       return 0;
+               pci_enable_device(pdev);
+               pci_set_master(pdev);
+       }
+       return 0;
+}
+
+static long
+core99_ide_enable(struct device_node *node, long param, long value)
+{
+       /* Bus ID 0 to 2 are KeyLargo based IDE, busID 3 is U2
+        * based ata-100
+        */
+       switch(param) {
+           case 0:
+               return simple_feature_tweak(node, macio_unknown,
+                       KEYLARGO_FCR1, KL1_EIDE0_ENABLE, value);
+           case 1:
+               return simple_feature_tweak(node, macio_unknown,
+                       KEYLARGO_FCR1, KL1_EIDE1_ENABLE, value);
+           case 2:
+               return simple_feature_tweak(node, macio_unknown,
+                       KEYLARGO_FCR1, KL1_UIDE_ENABLE, value);
+           case 3:
+               return core99_ata100_enable(node, value);
+           default:
+               return -ENODEV;
+       }
+}
+
+static long
+core99_ide_reset(struct device_node *node, long param, long value)
+{
+       switch(param) {
+           case 0:
+               return simple_feature_tweak(node, macio_unknown,
+                       KEYLARGO_FCR1, KL1_EIDE0_RESET_N, !value);
+           case 1:
+               return simple_feature_tweak(node, macio_unknown,
+                       KEYLARGO_FCR1, KL1_EIDE1_RESET_N, !value);
+           case 2:
+               return simple_feature_tweak(node, macio_unknown,
+                       KEYLARGO_FCR1, KL1_UIDE_RESET_N, !value);
+           default:
+               return -ENODEV;
+       }
+}
+
+static long
+core99_gmac_enable(struct device_node *node, long param, long value)
+{
+       unsigned long flags;
+
+       LOCK(flags);
+       if (value)
+               UN_BIS(UNI_N_CLOCK_CNTL, UNI_N_CLOCK_CNTL_GMAC);
+       else
+               UN_BIC(UNI_N_CLOCK_CNTL, UNI_N_CLOCK_CNTL_GMAC);
+       (void)UN_IN(UNI_N_CLOCK_CNTL);
+       UNLOCK(flags);
+       udelay(20);
+
+       return 0;
+}
+
+static long
+core99_gmac_phy_reset(struct device_node *node, long param, long value)
+{
+       unsigned long flags;
+       struct macio_chip *macio;
+
+       macio = &macio_chips[0];
+       if (macio->type != macio_keylargo && macio->type != macio_pangea &&
+           macio->type != macio_intrepid)
+               return -ENODEV;
+
+       LOCK(flags);
+       MACIO_OUT8(KL_GPIO_ETH_PHY_RESET, KEYLARGO_GPIO_OUTPUT_ENABLE);
+       (void)MACIO_IN8(KL_GPIO_ETH_PHY_RESET);
+       UNLOCK(flags);
+       mdelay(10);
+       LOCK(flags);
+       MACIO_OUT8(KL_GPIO_ETH_PHY_RESET, /*KEYLARGO_GPIO_OUTPUT_ENABLE | */
+               KEYLARGO_GPIO_OUTOUT_DATA);
+       UNLOCK(flags);
+       mdelay(10);
+
+       return 0;
+}
+
+static long
+core99_sound_chip_enable(struct device_node *node, long param, long value)
+{
+       struct macio_chip*      macio;
+       unsigned long           flags;
+
+       macio = macio_find(node, 0);
+       if (!macio)
+               return -ENODEV;
+
+       /* Do a better probe code, screamer G4 desktops &
+        * iMacs can do that too, add a recalibrate  in
+        * the driver as well
+        */
+       if (pmac_mb.model_id == PMAC_TYPE_PISMO ||
+           pmac_mb.model_id == PMAC_TYPE_TITANIUM) {
+               LOCK(flags);
+               if (value)
+                       MACIO_OUT8(KL_GPIO_SOUND_POWER,
+                               KEYLARGO_GPIO_OUTPUT_ENABLE |
+                               KEYLARGO_GPIO_OUTOUT_DATA);
+               else
+                       MACIO_OUT8(KL_GPIO_SOUND_POWER,
+                               KEYLARGO_GPIO_OUTPUT_ENABLE);
+               (void)MACIO_IN8(KL_GPIO_SOUND_POWER);
+               UNLOCK(flags);
+       }
+       return 0;
+}
+
+static long
+core99_airport_enable(struct device_node *node, long param, long value)
+{
+       struct macio_chip*      macio;
+       unsigned long           flags;
+       int                     state;
+
+       macio = macio_find(node, 0);
+       if (!macio)
+               return -ENODEV;
+
+       /* Hint: we allow passing of macio itself for the sake of the
+        * sleep code
+        */
+       if (node != macio->of_node &&
+           (!node->parent || node->parent != macio->of_node))
+               return -ENODEV;
+       state = (macio->flags & MACIO_FLAG_AIRPORT_ON) != 0;
+       if (value == state)
+               return 0;
+       if (value) {
+               /* This code is a reproduction of OF enable-cardslot
+                * and init-wireless methods, slightly hacked until
+                * I got it working.
+                */
+               LOCK(flags);
+               MACIO_OUT8(KEYLARGO_GPIO_0+0xf, 5);
+               (void)MACIO_IN8(KEYLARGO_GPIO_0+0xf);
+               UNLOCK(flags);
+               mdelay(10);
+               LOCK(flags);
+               MACIO_OUT8(KEYLARGO_GPIO_0+0xf, 4);
+               (void)MACIO_IN8(KEYLARGO_GPIO_0+0xf);
+               UNLOCK(flags);
+
+               mdelay(10);
+
+               LOCK(flags);
+               MACIO_BIC(KEYLARGO_FCR2, KL2_CARDSEL_16);
+               (void)MACIO_IN32(KEYLARGO_FCR2);
+               udelay(10);
+               MACIO_OUT8(KEYLARGO_GPIO_EXTINT_0+0xb, 0);
+               (void)MACIO_IN8(KEYLARGO_GPIO_EXTINT_0+0xb);
+               udelay(10);
+               MACIO_OUT8(KEYLARGO_GPIO_EXTINT_0+0xa, 0x28);
+               (void)MACIO_IN8(KEYLARGO_GPIO_EXTINT_0+0xa);
+               udelay(10);
+               MACIO_OUT8(KEYLARGO_GPIO_EXTINT_0+0xd, 0x28);
+               (void)MACIO_IN8(KEYLARGO_GPIO_EXTINT_0+0xd);
+               udelay(10);
+               MACIO_OUT8(KEYLARGO_GPIO_0+0xd, 0x28);
+               (void)MACIO_IN8(KEYLARGO_GPIO_0+0xd);
+               udelay(10);
+               MACIO_OUT8(KEYLARGO_GPIO_0+0xe, 0x28);
+               (void)MACIO_IN8(KEYLARGO_GPIO_0+0xe);
+               UNLOCK(flags);
+               udelay(10);
+               MACIO_OUT32(0x1c000, 0);
+               mdelay(1);
+               MACIO_OUT8(0x1a3e0, 0x41);
+               (void)MACIO_IN8(0x1a3e0);
+               udelay(10);
+               LOCK(flags);
+               MACIO_BIS(KEYLARGO_FCR2, KL2_CARDSEL_16);
+               (void)MACIO_IN32(KEYLARGO_FCR2);
+               UNLOCK(flags);
+               mdelay(100);
+
+               macio->flags |= MACIO_FLAG_AIRPORT_ON;
+       } else {
+               LOCK(flags);
+               MACIO_BIC(KEYLARGO_FCR2, KL2_CARDSEL_16);
+               (void)MACIO_IN32(KEYLARGO_FCR2);
+               MACIO_OUT8(KL_GPIO_AIRPORT_0, 0);
+               MACIO_OUT8(KL_GPIO_AIRPORT_1, 0);
+               MACIO_OUT8(KL_GPIO_AIRPORT_2, 0);
+               MACIO_OUT8(KL_GPIO_AIRPORT_3, 0);
+               MACIO_OUT8(KL_GPIO_AIRPORT_4, 0);
+               (void)MACIO_IN8(KL_GPIO_AIRPORT_4);
+               UNLOCK(flags);
+
+               macio->flags &= ~MACIO_FLAG_AIRPORT_ON;
+       }
+       return 0;
+}
+
+#ifdef CONFIG_SMP
+static long
+core99_reset_cpu(struct device_node *node, long param, long value)
+{
+       unsigned int reset_io = 0;
+       unsigned long flags;
+       struct macio_chip *macio;
+       struct device_node *np;
+       const int dflt_reset_lines[] = {        KL_GPIO_RESET_CPU0,
+                                               KL_GPIO_RESET_CPU1,
+                                               KL_GPIO_RESET_CPU2,
+                                               KL_GPIO_RESET_CPU3 };
+
+       macio = &macio_chips[0];
+       if (macio->type != macio_keylargo)
+               return -ENODEV;
+
+       np = find_path_device("/cpus");
+       if (np == NULL)
+               return -ENODEV;
+       for (np = np->child; np != NULL; np = np->sibling) {
+               u32 *num = (u32 *)get_property(np, "reg", NULL);
+               u32 *rst = (u32 *)get_property(np, "soft-reset", NULL);
+               if (num == NULL || rst == NULL)
+                       continue;
+               if (param == *num) {
+                       reset_io = *rst;
+                       break;
+               }
+       }
+       if (np == NULL || reset_io == 0)
+               reset_io = dflt_reset_lines[param];
+
+       LOCK(flags);
+       MACIO_OUT8(reset_io, KEYLARGO_GPIO_OUTPUT_ENABLE);
+       (void)MACIO_IN8(reset_io);
+       udelay(1);
+       MACIO_OUT8(reset_io, 0);
+       (void)MACIO_IN8(reset_io);
+       UNLOCK(flags);
+
+       return 0;
+}
+#endif /* CONFIG_SMP */
+
+static long
+core99_usb_enable(struct device_node *node, long param, long value)
+{
+       struct macio_chip *macio;
+       unsigned long flags;
+       char *prop;
+       int number;
+       u32 reg;
+
+       macio = &macio_chips[0];
+       if (macio->type != macio_keylargo && macio->type != macio_pangea &&
+           macio->type != macio_intrepid)
+               return -ENODEV;
+
+       prop = (char *)get_property(node, "AAPL,clock-id", NULL);
+       if (!prop)
+               return -ENODEV;
+       if (strncmp(prop, "usb0u048", 8) == 0)
+               number = 0;
+       else if (strncmp(prop, "usb1u148", 8) == 0)
+               number = 2;
+       else if (strncmp(prop, "usb2u248", 8) == 0)
+               number = 4;
+       else
+               return -ENODEV;
+
+       /* Sorry for the brute-force locking, but this is only used during
+        * sleep and the timing seem to be critical
+        */
+       LOCK(flags);
+       if (value) {
+               /* Turn ON */
+               if (number == 0) {
+                       MACIO_BIC(KEYLARGO_FCR0, (KL0_USB0_PAD_SUSPEND0 | KL0_USB0_PAD_SUSPEND1));
+                       (void)MACIO_IN32(KEYLARGO_FCR0);
+                       UNLOCK(flags);
+                       mdelay(1);
+                       LOCK(flags);
+                       MACIO_BIS(KEYLARGO_FCR0, KL0_USB0_CELL_ENABLE);
+               } else if (number == 2) {
+                       MACIO_BIC(KEYLARGO_FCR0, (KL0_USB1_PAD_SUSPEND0 | KL0_USB1_PAD_SUSPEND1));
+                       UNLOCK(flags);
+                       (void)MACIO_IN32(KEYLARGO_FCR0);
+                       mdelay(1);
+                       LOCK(flags);
+                       MACIO_BIS(KEYLARGO_FCR0, KL0_USB1_CELL_ENABLE);
+               } else if (number == 4) {
+                       MACIO_BIC(KEYLARGO_FCR1, (KL1_USB2_PAD_SUSPEND0 | KL1_USB2_PAD_SUSPEND1));
+                       UNLOCK(flags);
+                       (void)MACIO_IN32(KEYLARGO_FCR1);
+                       mdelay(1);
+                       LOCK(flags);
+                       MACIO_BIS(KEYLARGO_FCR1, KL1_USB2_CELL_ENABLE);
+               }
+               if (number < 4) {
+                       reg = MACIO_IN32(KEYLARGO_FCR4);
+                       reg &=  ~(KL4_PORT_WAKEUP_ENABLE(number) | KL4_PORT_RESUME_WAKE_EN(number) |
+                               KL4_PORT_CONNECT_WAKE_EN(number) | KL4_PORT_DISCONNECT_WAKE_EN(number));
+                       reg &=  ~(KL4_PORT_WAKEUP_ENABLE(number+1) | KL4_PORT_RESUME_WAKE_EN(number+1) |
+                               KL4_PORT_CONNECT_WAKE_EN(number+1) | KL4_PORT_DISCONNECT_WAKE_EN(number+1));
+                       MACIO_OUT32(KEYLARGO_FCR4, reg);
+                       (void)MACIO_IN32(KEYLARGO_FCR4);
+                       udelay(10);
+               } else {
+                       reg = MACIO_IN32(KEYLARGO_FCR3);
+                       reg &=  ~(KL3_IT_PORT_WAKEUP_ENABLE(0) | KL3_IT_PORT_RESUME_WAKE_EN(0) |
+                               KL3_IT_PORT_CONNECT_WAKE_EN(0) | KL3_IT_PORT_DISCONNECT_WAKE_EN(0));
+                       reg &=  ~(KL3_IT_PORT_WAKEUP_ENABLE(1) | KL3_IT_PORT_RESUME_WAKE_EN(1) |
+                               KL3_IT_PORT_CONNECT_WAKE_EN(1) | KL3_IT_PORT_DISCONNECT_WAKE_EN(1));
+                       MACIO_OUT32(KEYLARGO_FCR3, reg);
+                       (void)MACIO_IN32(KEYLARGO_FCR3);
+                       udelay(10);
+               }
+               if (macio->type == macio_intrepid) {
+                       /* wait for clock stopped bits to clear */
+                       u32 test0 = 0, test1 = 0;
+                       u32 status0, status1;
+                       int timeout = 1000;
+
+                       UNLOCK(flags);
+                       switch (number) {
+                       case 0:
+                               test0 = UNI_N_CLOCK_STOPPED_USB0;
+                               test1 = UNI_N_CLOCK_STOPPED_USB0PCI;
+                               break;
+                       case 2:
+                               test0 = UNI_N_CLOCK_STOPPED_USB1;
+                               test1 = UNI_N_CLOCK_STOPPED_USB1PCI;
+                               break;
+                       case 4:
+                               test0 = UNI_N_CLOCK_STOPPED_USB2;
+                               test1 = UNI_N_CLOCK_STOPPED_USB2PCI;
+                               break;
+                       }
+                       do {
+                               if (--timeout <= 0) {
+                                       printk(KERN_ERR "core99_usb_enable: "
+                                              "Timeout waiting for clocks\n");
+                                       break;
+                               }
+                               mdelay(1);
+                               status0 = UN_IN(UNI_N_CLOCK_STOP_STATUS0);
+                               status1 = UN_IN(UNI_N_CLOCK_STOP_STATUS1);
+                       } while ((status0 & test0) | (status1 & test1));
+                       LOCK(flags);
+               }
+       } else {
+               /* Turn OFF */
+               if (number < 4) {
+                       reg = MACIO_IN32(KEYLARGO_FCR4);
+                       reg |=  KL4_PORT_WAKEUP_ENABLE(number) | KL4_PORT_RESUME_WAKE_EN(number) |
+                               KL4_PORT_CONNECT_WAKE_EN(number) | KL4_PORT_DISCONNECT_WAKE_EN(number);
+                       reg |=  KL4_PORT_WAKEUP_ENABLE(number+1) | KL4_PORT_RESUME_WAKE_EN(number+1) |
+                               KL4_PORT_CONNECT_WAKE_EN(number+1) | KL4_PORT_DISCONNECT_WAKE_EN(number+1);
+                       MACIO_OUT32(KEYLARGO_FCR4, reg);
+                       (void)MACIO_IN32(KEYLARGO_FCR4);
+                       udelay(1);
+               } else {
+                       reg = MACIO_IN32(KEYLARGO_FCR3);
+                       reg |=  KL3_IT_PORT_WAKEUP_ENABLE(0) | KL3_IT_PORT_RESUME_WAKE_EN(0) |
+                               KL3_IT_PORT_CONNECT_WAKE_EN(0) | KL3_IT_PORT_DISCONNECT_WAKE_EN(0);
+                       reg |=  KL3_IT_PORT_WAKEUP_ENABLE(1) | KL3_IT_PORT_RESUME_WAKE_EN(1) |
+                               KL3_IT_PORT_CONNECT_WAKE_EN(1) | KL3_IT_PORT_DISCONNECT_WAKE_EN(1);
+                       MACIO_OUT32(KEYLARGO_FCR3, reg);
+                       (void)MACIO_IN32(KEYLARGO_FCR3);
+                       udelay(1);
+               }
+               if (number == 0) {
+                       if (macio->type != macio_intrepid)
+                               MACIO_BIC(KEYLARGO_FCR0, KL0_USB0_CELL_ENABLE);
+                       (void)MACIO_IN32(KEYLARGO_FCR0);
+                       udelay(1);
+                       MACIO_BIS(KEYLARGO_FCR0, (KL0_USB0_PAD_SUSPEND0 | KL0_USB0_PAD_SUSPEND1));
+                       (void)MACIO_IN32(KEYLARGO_FCR0);
+               } else if (number == 2) {
+                       if (macio->type != macio_intrepid)
+                               MACIO_BIC(KEYLARGO_FCR0, KL0_USB1_CELL_ENABLE);
+                       (void)MACIO_IN32(KEYLARGO_FCR0);
+                       udelay(1);
+                       MACIO_BIS(KEYLARGO_FCR0, (KL0_USB1_PAD_SUSPEND0 | KL0_USB1_PAD_SUSPEND1));
+                       (void)MACIO_IN32(KEYLARGO_FCR0);
+               } else if (number == 4) {
+                       udelay(1);
+                       MACIO_BIS(KEYLARGO_FCR1, (KL1_USB2_PAD_SUSPEND0 | KL1_USB2_PAD_SUSPEND1));
+                       (void)MACIO_IN32(KEYLARGO_FCR1);
+               }
+               udelay(1);
+       }
+       UNLOCK(flags);
+
+       return 0;
+}
+
+static long
+core99_firewire_enable(struct device_node *node, long param, long value)
+{
+       unsigned long flags;
+       struct macio_chip *macio;
+
+       macio = &macio_chips[0];
+       if (macio->type != macio_keylargo && macio->type != macio_pangea &&
+           macio->type != macio_intrepid)
+               return -ENODEV;
+       if (!(macio->flags & MACIO_FLAG_FW_SUPPORTED))
+               return -ENODEV;
+
+       LOCK(flags);
+       if (value) {
+               UN_BIS(UNI_N_CLOCK_CNTL, UNI_N_CLOCK_CNTL_FW);
+               (void)UN_IN(UNI_N_CLOCK_CNTL);
+       } else {
+               UN_BIC(UNI_N_CLOCK_CNTL, UNI_N_CLOCK_CNTL_FW);
+               (void)UN_IN(UNI_N_CLOCK_CNTL);
+       }
+       UNLOCK(flags);
+       mdelay(1);
+
+       return 0;
+}
+
+static long
+core99_firewire_cable_power(struct device_node *node, long param, long value)
+{
+       unsigned long flags;
+       struct macio_chip *macio;
+
+       /* Trick: we allow NULL node */
+       if ((pmac_mb.board_flags & PMAC_MB_HAS_FW_POWER) == 0)
+               return -ENODEV;
+       macio = &macio_chips[0];
+       if (macio->type != macio_keylargo && macio->type != macio_pangea &&
+           macio->type != macio_intrepid)
+               return -ENODEV;
+       if (!(macio->flags & MACIO_FLAG_FW_SUPPORTED))
+               return -ENODEV;
+
+       LOCK(flags);
+       if (value) {
+               MACIO_OUT8(KL_GPIO_FW_CABLE_POWER , 0);
+               MACIO_IN8(KL_GPIO_FW_CABLE_POWER);
+               udelay(10);
+       } else {
+               MACIO_OUT8(KL_GPIO_FW_CABLE_POWER , 4);
+               MACIO_IN8(KL_GPIO_FW_CABLE_POWER); udelay(10);
+       }
+       UNLOCK(flags);
+       mdelay(1);
+
+       return 0;
+}
+
+static long
+intrepid_aack_delay_enable(struct device_node *node, long param, long value)
+{
+       unsigned long flags;
+
+       if (uninorth_rev < 0xd2)
+               return -ENODEV;
+
+       LOCK(flags);
+       if (param)
+               UN_BIS(UNI_N_AACK_DELAY, UNI_N_AACK_DELAY_ENABLE);
+       else
+               UN_BIC(UNI_N_AACK_DELAY, UNI_N_AACK_DELAY_ENABLE);
+       UNLOCK(flags);
+
+       return 0;
+}
+
+
+#endif /* CONFIG_POWER4 */
+
+static long
+core99_read_gpio(struct device_node *node, long param, long value)
+{
+       struct macio_chip *macio = &macio_chips[0];
+
+       return MACIO_IN8(param);
+}
+
+
+static long
+core99_write_gpio(struct device_node *node, long param, long value)
+{
+       struct macio_chip *macio = &macio_chips[0];
+
+       MACIO_OUT8(param, (u8)(value & 0xff));
+       return 0;
+}
+
+#ifdef CONFIG_POWER4
+static long g5_gmac_enable(struct device_node *node, long param, long value)
+{
+       struct macio_chip *macio = &macio_chips[0];
+       unsigned long flags;
+
+       if (node == NULL)
+               return -ENODEV;
+
+       LOCK(flags);
+       if (value) {
+               MACIO_BIS(KEYLARGO_FCR1, K2_FCR1_GMAC_CLK_ENABLE);
+               mb();
+               k2_skiplist[0] = NULL;
+       } else {
+               k2_skiplist[0] = node;
+               mb();
+               MACIO_BIC(KEYLARGO_FCR1, K2_FCR1_GMAC_CLK_ENABLE);
+       }
+       
+       UNLOCK(flags);
+       mdelay(1);
+
+       return 0;
+}
+
+static long g5_fw_enable(struct device_node *node, long param, long value)
+{
+       struct macio_chip *macio = &macio_chips[0];
+       unsigned long flags;
+
+       if (node == NULL)
+               return -ENODEV;
+
+       LOCK(flags);
+       if (value) {
+               MACIO_BIS(KEYLARGO_FCR1, K2_FCR1_FW_CLK_ENABLE);
+               mb();
+               k2_skiplist[1] = NULL;
+       } else {
+               k2_skiplist[1] = node;
+               mb();
+               MACIO_BIC(KEYLARGO_FCR1, K2_FCR1_FW_CLK_ENABLE);
+       }
+       
+       UNLOCK(flags);
+       mdelay(1);
+
+       return 0;
+}
+
+static long g5_mpic_enable(struct device_node *node, long param, long value)
+{
+       unsigned long flags;
+
+       if (node->parent == NULL || strcmp(node->parent->name, "u3"))
+               return 0;
+
+       LOCK(flags);
+       UN_BIS(U3_TOGGLE_REG, U3_MPIC_RESET | U3_MPIC_OUTPUT_ENABLE);
+       UNLOCK(flags);
+
+       return 0;
+}
+
+static long g5_eth_phy_reset(struct device_node *node, long param, long value)
+{
+       struct macio_chip *macio = &macio_chips[0];
+       struct device_node *phy;
+       int need_reset;
+
+       /*
+        * We must not reset the combo PHYs, only the BCM5221 found in
+        * the iMac G5.
+        */
+       phy = of_get_next_child(node, NULL);
+       if (!phy)
+               return -ENODEV;
+       need_reset = device_is_compatible(phy, "B5221");
+       of_node_put(phy);
+       if (!need_reset)
+               return 0;
+
+       /* PHY reset is GPIO 29, not in device-tree unfortunately */
+       MACIO_OUT8(K2_GPIO_EXTINT_0 + 29,
+                  KEYLARGO_GPIO_OUTPUT_ENABLE | KEYLARGO_GPIO_OUTOUT_DATA);
+       /* Thankfully, this is now always called at a time when we can
+        * schedule by sungem.
+        */
+       msleep(10);
+       MACIO_OUT8(K2_GPIO_EXTINT_0 + 29, 0);
+
+       return 0;
+}
+
+static long g5_i2s_enable(struct device_node *node, long param, long value)
+{
+       /* Very crude implementation for now */
+       struct macio_chip *macio = &macio_chips[0];
+       unsigned long flags;
+
+       if (value == 0)
+               return 0; /* don't disable yet */
+
+       LOCK(flags);
+       MACIO_BIS(KEYLARGO_FCR3, KL3_CLK45_ENABLE | KL3_CLK49_ENABLE |
+                 KL3_I2S0_CLK18_ENABLE);
+       udelay(10);
+       MACIO_BIS(KEYLARGO_FCR1, K2_FCR1_I2S0_CELL_ENABLE |
+                 K2_FCR1_I2S0_CLK_ENABLE_BIT | K2_FCR1_I2S0_ENABLE);
+       udelay(10);
+       MACIO_BIC(KEYLARGO_FCR1, K2_FCR1_I2S0_RESET);
+       UNLOCK(flags);
+       udelay(10);
+
+       return 0;
+}
+
+
+#ifdef CONFIG_SMP
+static long g5_reset_cpu(struct device_node *node, long param, long value)
+{
+       unsigned int reset_io = 0;
+       unsigned long flags;
+       struct macio_chip *macio;
+       struct device_node *np;
+
+       macio = &macio_chips[0];
+       if (macio->type != macio_keylargo2)
+               return -ENODEV;
+
+       np = find_path_device("/cpus");
+       if (np == NULL)
+               return -ENODEV;
+       for (np = np->child; np != NULL; np = np->sibling) {
+               u32 *num = (u32 *)get_property(np, "reg", NULL);
+               u32 *rst = (u32 *)get_property(np, "soft-reset", NULL);
+               if (num == NULL || rst == NULL)
+                       continue;
+               if (param == *num) {
+                       reset_io = *rst;
+                       break;
+               }
+       }
+       if (np == NULL || reset_io == 0)
+               return -ENODEV;
+
+       LOCK(flags);
+       MACIO_OUT8(reset_io, KEYLARGO_GPIO_OUTPUT_ENABLE);
+       (void)MACIO_IN8(reset_io);
+       udelay(1);
+       MACIO_OUT8(reset_io, 0);
+       (void)MACIO_IN8(reset_io);
+       UNLOCK(flags);
+
+       return 0;
+}
+#endif /* CONFIG_SMP */
+
+/*
+ * This can be called from pmac_smp so isn't static
+ *
+ * This takes the second CPU off the bus on dual CPU machines
+ * running UP
+ */
+void g5_phy_disable_cpu1(void)
+{
+       UN_OUT(U3_API_PHY_CONFIG_1, 0);
+}
+#endif /* CONFIG_POWER4 */
+
+#ifndef CONFIG_POWER4
+
+static void
+keylargo_shutdown(struct macio_chip *macio, int sleep_mode)
+{
+       u32 temp;
+
+       if (sleep_mode) {
+               mdelay(1);
+               MACIO_BIS(KEYLARGO_FCR0, KL0_USB_REF_SUSPEND);
+               (void)MACIO_IN32(KEYLARGO_FCR0);
+               mdelay(1);
+       }
+
+       MACIO_BIC(KEYLARGO_FCR0,KL0_SCCA_ENABLE | KL0_SCCB_ENABLE |
+                               KL0_SCC_CELL_ENABLE |
+                               KL0_IRDA_ENABLE | KL0_IRDA_CLK32_ENABLE |
+                               KL0_IRDA_CLK19_ENABLE);
+
+       MACIO_BIC(KEYLARGO_MBCR, KL_MBCR_MB0_DEV_MASK);
+       MACIO_BIS(KEYLARGO_MBCR, KL_MBCR_MB0_IDE_ENABLE);
+
+       MACIO_BIC(KEYLARGO_FCR1,
+               KL1_AUDIO_SEL_22MCLK | KL1_AUDIO_CLK_ENABLE_BIT |
+               KL1_AUDIO_CLK_OUT_ENABLE | KL1_AUDIO_CELL_ENABLE |
+               KL1_I2S0_CELL_ENABLE | KL1_I2S0_CLK_ENABLE_BIT |
+               KL1_I2S0_ENABLE | KL1_I2S1_CELL_ENABLE |
+               KL1_I2S1_CLK_ENABLE_BIT | KL1_I2S1_ENABLE |
+               KL1_EIDE0_ENABLE | KL1_EIDE0_RESET_N |
+               KL1_EIDE1_ENABLE | KL1_EIDE1_RESET_N |
+               KL1_UIDE_ENABLE);
+
+       MACIO_BIS(KEYLARGO_FCR2, KL2_ALT_DATA_OUT);
+       MACIO_BIC(KEYLARGO_FCR2, KL2_IOBUS_ENABLE);
+
+       temp = MACIO_IN32(KEYLARGO_FCR3);
+       if (macio->rev >= 2) {
+               temp |= KL3_SHUTDOWN_PLL2X;
+               if (sleep_mode)
+                       temp |= KL3_SHUTDOWN_PLL_TOTAL;
+       }
+
+       temp |= KL3_SHUTDOWN_PLLKW6 | KL3_SHUTDOWN_PLLKW4 |
+               KL3_SHUTDOWN_PLLKW35;
+       if (sleep_mode)
+               temp |= KL3_SHUTDOWN_PLLKW12;
+       temp &= ~(KL3_CLK66_ENABLE | KL3_CLK49_ENABLE | KL3_CLK45_ENABLE
+               | KL3_CLK31_ENABLE | KL3_I2S1_CLK18_ENABLE | KL3_I2S0_CLK18_ENABLE);
+       if (sleep_mode)
+               temp &= ~(KL3_TIMER_CLK18_ENABLE | KL3_VIA_CLK16_ENABLE);
+       MACIO_OUT32(KEYLARGO_FCR3, temp);
+
+       /* Flush posted writes & wait a bit */
+       (void)MACIO_IN32(KEYLARGO_FCR0); mdelay(1);
+}
+
+static void
+pangea_shutdown(struct macio_chip *macio, int sleep_mode)
+{
+       u32 temp;
+
+       MACIO_BIC(KEYLARGO_FCR0,KL0_SCCA_ENABLE | KL0_SCCB_ENABLE |
+                               KL0_SCC_CELL_ENABLE |
+                               KL0_USB0_CELL_ENABLE | KL0_USB1_CELL_ENABLE);
+
+       MACIO_BIC(KEYLARGO_FCR1,
+               KL1_AUDIO_SEL_22MCLK | KL1_AUDIO_CLK_ENABLE_BIT |
+               KL1_AUDIO_CLK_OUT_ENABLE | KL1_AUDIO_CELL_ENABLE |
+               KL1_I2S0_CELL_ENABLE | KL1_I2S0_CLK_ENABLE_BIT |
+               KL1_I2S0_ENABLE | KL1_I2S1_CELL_ENABLE |
+               KL1_I2S1_CLK_ENABLE_BIT | KL1_I2S1_ENABLE |
+               KL1_UIDE_ENABLE);
+       if (pmac_mb.board_flags & PMAC_MB_MOBILE)
+               MACIO_BIC(KEYLARGO_FCR1, KL1_UIDE_RESET_N);
+
+       MACIO_BIS(KEYLARGO_FCR2, KL2_ALT_DATA_OUT);
+
+       temp = MACIO_IN32(KEYLARGO_FCR3);
+       temp |= KL3_SHUTDOWN_PLLKW6 | KL3_SHUTDOWN_PLLKW4 |
+               KL3_SHUTDOWN_PLLKW35;
+       temp &= ~(KL3_CLK49_ENABLE | KL3_CLK45_ENABLE | KL3_CLK31_ENABLE
+               | KL3_I2S0_CLK18_ENABLE | KL3_I2S1_CLK18_ENABLE);
+       if (sleep_mode)
+               temp &= ~(KL3_VIA_CLK16_ENABLE | KL3_TIMER_CLK18_ENABLE);
+       MACIO_OUT32(KEYLARGO_FCR3, temp);
+
+       /* Flush posted writes & wait a bit */
+       (void)MACIO_IN32(KEYLARGO_FCR0); mdelay(1);
+}
+
+static void
+intrepid_shutdown(struct macio_chip *macio, int sleep_mode)
+{
+       u32 temp;
+
+       MACIO_BIC(KEYLARGO_FCR0,KL0_SCCA_ENABLE | KL0_SCCB_ENABLE |
+                 KL0_SCC_CELL_ENABLE);
+
+       MACIO_BIC(KEYLARGO_FCR1,
+                 /*KL1_USB2_CELL_ENABLE |*/
+               KL1_I2S0_CELL_ENABLE | KL1_I2S0_CLK_ENABLE_BIT |
+               KL1_I2S0_ENABLE | KL1_I2S1_CELL_ENABLE |
+               KL1_I2S1_CLK_ENABLE_BIT | KL1_I2S1_ENABLE);
+       if (pmac_mb.board_flags & PMAC_MB_MOBILE)
+               MACIO_BIC(KEYLARGO_FCR1, KL1_UIDE_RESET_N);
+
+       temp = MACIO_IN32(KEYLARGO_FCR3);
+       temp &= ~(KL3_CLK49_ENABLE | KL3_CLK45_ENABLE |
+                 KL3_I2S1_CLK18_ENABLE | KL3_I2S0_CLK18_ENABLE);
+       if (sleep_mode)
+               temp &= ~(KL3_TIMER_CLK18_ENABLE | KL3_IT_VIA_CLK32_ENABLE);
+       MACIO_OUT32(KEYLARGO_FCR3, temp);
+
+       /* Flush posted writes & wait a bit */
+       (void)MACIO_IN32(KEYLARGO_FCR0);
+       mdelay(10);
+}
+
+
+void pmac_tweak_clock_spreading(int enable)
+{
+       struct macio_chip *macio = &macio_chips[0];
+
+       /* Hack for doing clock spreading on some machines PowerBooks and
+        * iBooks. This implements the "platform-do-clockspreading" OF
+        * property as decoded manually on various models. For safety, we also
+        * check the product ID in the device-tree in cases we'll whack the i2c
+        * chip to make reasonably sure we won't set wrong values in there
+        *
+        * Of course, ultimately, we have to implement a real parser for
+        * the platform-do-* stuff...
+        */
+
+       if (macio->type == macio_intrepid) {
+               if (enable)
+                       UN_OUT(UNI_N_CLOCK_SPREADING, 2);
+               else
+                       UN_OUT(UNI_N_CLOCK_SPREADING, 0);
+               mdelay(40);
+       }
+
+       while (machine_is_compatible("PowerBook5,2") ||
+              machine_is_compatible("PowerBook5,3") ||
+              machine_is_compatible("PowerBook6,2") ||
+              machine_is_compatible("PowerBook6,3")) {
+               struct device_node *ui2c = of_find_node_by_type(NULL, "i2c");
+               struct device_node *dt = of_find_node_by_name(NULL, "device-tree");
+               u8 buffer[9];
+               u32 *productID;
+               int i, rc, changed = 0;
+
+               if (dt == NULL)
+                       break;
+               productID = (u32 *)get_property(dt, "pid#", NULL);
+               if (productID == NULL)
+                       break;
+               while(ui2c) {
+                       struct device_node *p = of_get_parent(ui2c);
+                       if (p && !strcmp(p->name, "uni-n"))
+                               break;
+                       ui2c = of_find_node_by_type(ui2c, "i2c");
+               }
+               if (ui2c == NULL)
+                       break;
+               DBG("Trying to bump clock speed for PID: %08x...\n", *productID);
+               rc = pmac_low_i2c_open(ui2c, 1);
+               if (rc != 0)
+                       break;
+               pmac_low_i2c_setmode(ui2c, pmac_low_i2c_mode_combined);
+               rc = pmac_low_i2c_xfer(ui2c, 0xd2 | pmac_low_i2c_read, 0x80, buffer, 9);
+               DBG("read result: %d,", rc);
+               if (rc != 0) {
+                       pmac_low_i2c_close(ui2c);
+                       break;
+               }
+               for (i=0; i<9; i++)
+                       DBG(" %02x", buffer[i]);
+               DBG("\n");
+
+               switch(*productID) {
+               case 0x1182:    /* AlBook 12" rev 2 */
+               case 0x1183:    /* iBook G4 12" */
+                       buffer[0] = (buffer[0] & 0x8f) | 0x70;
+                       buffer[2] = (buffer[2] & 0x7f) | 0x00;
+                       buffer[5] = (buffer[5] & 0x80) | 0x31;
+                       buffer[6] = (buffer[6] & 0x40) | 0xb0;
+                       buffer[7] = (buffer[7] & 0x00) | (enable ? 0xc0 : 0xba);
+                       buffer[8] = (buffer[8] & 0x00) | 0x30;
+                       changed = 1;
+                       break;
+               case 0x3142:    /* AlBook 15" (ATI M10) */
+               case 0x3143:    /* AlBook 17" (ATI M10) */
+                       buffer[0] = (buffer[0] & 0xaf) | 0x50;
+                       buffer[2] = (buffer[2] & 0x7f) | 0x00;
+                       buffer[5] = (buffer[5] & 0x80) | 0x31;
+                       buffer[6] = (buffer[6] & 0x40) | 0xb0;
+                       buffer[7] = (buffer[7] & 0x00) | (enable ? 0xd0 : 0xc0);
+                       buffer[8] = (buffer[8] & 0x00) | 0x30;
+                       changed = 1;
+                       break;
+               default:
+                       DBG("i2c-hwclock: Machine model not handled\n");
+                       break;
+               }
+               if (!changed) {
+                       pmac_low_i2c_close(ui2c);
+                       break;
+               }
+               pmac_low_i2c_setmode(ui2c, pmac_low_i2c_mode_stdsub);
+               rc = pmac_low_i2c_xfer(ui2c, 0xd2 | pmac_low_i2c_write, 0x80, buffer, 9);
+               DBG("write result: %d,", rc);
+               pmac_low_i2c_setmode(ui2c, pmac_low_i2c_mode_combined);
+               rc = pmac_low_i2c_xfer(ui2c, 0xd2 | pmac_low_i2c_read, 0x80, buffer, 9);
+               DBG("read result: %d,", rc);
+               if (rc != 0) {
+                       pmac_low_i2c_close(ui2c);
+                       break;
+               }
+               for (i=0; i<9; i++)
+                       DBG(" %02x", buffer[i]);
+               pmac_low_i2c_close(ui2c);
+               break;
+       }
+}
+
+
+static int
+core99_sleep(void)
+{
+       struct macio_chip *macio;
+       int i;
+
+       macio = &macio_chips[0];
+       if (macio->type != macio_keylargo && macio->type != macio_pangea &&
+           macio->type != macio_intrepid)
+               return -ENODEV;
+
+       /* We power off the wireless slot in case it was not done
+        * by the driver. We don't power it on automatically however
+        */
+       if (macio->flags & MACIO_FLAG_AIRPORT_ON)
+               core99_airport_enable(macio->of_node, 0, 0);
+
+       /* We power off the FW cable. Should be done by the driver... */
+       if (macio->flags & MACIO_FLAG_FW_SUPPORTED) {
+               core99_firewire_enable(NULL, 0, 0);
+               core99_firewire_cable_power(NULL, 0, 0);
+       }
+
+       /* We make sure int. modem is off (in case driver lost it) */
+       if (macio->type == macio_keylargo)
+               core99_modem_enable(macio->of_node, 0, 0);
+       else
+               pangea_modem_enable(macio->of_node, 0, 0);
+
+       /* We make sure the sound is off as well */
+       core99_sound_chip_enable(macio->of_node, 0, 0);
+
+       /*
+        * Save various bits of KeyLargo
+        */
+
+       /* Save the state of the various GPIOs */
+       save_gpio_levels[0] = MACIO_IN32(KEYLARGO_GPIO_LEVELS0);
+       save_gpio_levels[1] = MACIO_IN32(KEYLARGO_GPIO_LEVELS1);
+       for (i=0; i<KEYLARGO_GPIO_EXTINT_CNT; i++)
+               save_gpio_extint[i] = MACIO_IN8(KEYLARGO_GPIO_EXTINT_0+i);
+       for (i=0; i<KEYLARGO_GPIO_CNT; i++)
+               save_gpio_normal[i] = MACIO_IN8(KEYLARGO_GPIO_0+i);
+
+       /* Save the FCRs */
+       if (macio->type == macio_keylargo)
+               save_mbcr = MACIO_IN32(KEYLARGO_MBCR);
+       save_fcr[0] = MACIO_IN32(KEYLARGO_FCR0);
+       save_fcr[1] = MACIO_IN32(KEYLARGO_FCR1);
+       save_fcr[2] = MACIO_IN32(KEYLARGO_FCR2);
+       save_fcr[3] = MACIO_IN32(KEYLARGO_FCR3);
+       save_fcr[4] = MACIO_IN32(KEYLARGO_FCR4);
+       if (macio->type == macio_pangea || macio->type == macio_intrepid)
+               save_fcr[5] = MACIO_IN32(KEYLARGO_FCR5);
+
+       /* Save state & config of DBDMA channels */
+       dbdma_save(macio, save_dbdma);
+
+       /*
+        * Turn off as much as we can
+        */
+       if (macio->type == macio_pangea)
+               pangea_shutdown(macio, 1);
+       else if (macio->type == macio_intrepid)
+               intrepid_shutdown(macio, 1);
+       else if (macio->type == macio_keylargo)
+               keylargo_shutdown(macio, 1);
+
+       /*
+        * Put the host bridge to sleep
+        */
+
+       save_unin_clock_ctl = UN_IN(UNI_N_CLOCK_CNTL);
+       /* Note: do not switch GMAC off, driver does it when necessary, WOL must keep it
+        * enabled !
+        */
+       UN_OUT(UNI_N_CLOCK_CNTL, save_unin_clock_ctl &
+              ~(/*UNI_N_CLOCK_CNTL_GMAC|*/UNI_N_CLOCK_CNTL_FW/*|UNI_N_CLOCK_CNTL_PCI*/));
+       udelay(100);
+       UN_OUT(UNI_N_HWINIT_STATE, UNI_N_HWINIT_STATE_SLEEPING);
+       UN_OUT(UNI_N_POWER_MGT, UNI_N_POWER_MGT_SLEEP);
+       mdelay(10);
+
+       /*
+        * FIXME: A bit of black magic with OpenPIC (don't ask me why)
+        */
+       if (pmac_mb.model_id == PMAC_TYPE_SAWTOOTH) {
+               MACIO_BIS(0x506e0, 0x00400000);
+               MACIO_BIS(0x506e0, 0x80000000);
+       }
+       return 0;
+}
+
+static int
+core99_wake_up(void)
+{
+       struct macio_chip *macio;
+       int i;
+
+       macio = &macio_chips[0];
+       if (macio->type != macio_keylargo && macio->type != macio_pangea &&
+           macio->type != macio_intrepid)
+               return -ENODEV;
+
+       /*
+        * Wakeup the host bridge
+        */
+       UN_OUT(UNI_N_POWER_MGT, UNI_N_POWER_MGT_NORMAL);
+       udelay(10);
+       UN_OUT(UNI_N_HWINIT_STATE, UNI_N_HWINIT_STATE_RUNNING);
+       udelay(10);
+
+       /*
+        * Restore KeyLargo
+        */
+
+       if (macio->type == macio_keylargo) {
+               MACIO_OUT32(KEYLARGO_MBCR, save_mbcr);
+               (void)MACIO_IN32(KEYLARGO_MBCR); udelay(10);
+       }
+       MACIO_OUT32(KEYLARGO_FCR0, save_fcr[0]);
+       (void)MACIO_IN32(KEYLARGO_FCR0); udelay(10);
+       MACIO_OUT32(KEYLARGO_FCR1, save_fcr[1]);
+       (void)MACIO_IN32(KEYLARGO_FCR1); udelay(10);
+       MACIO_OUT32(KEYLARGO_FCR2, save_fcr[2]);
+       (void)MACIO_IN32(KEYLARGO_FCR2); udelay(10);
+       MACIO_OUT32(KEYLARGO_FCR3, save_fcr[3]);
+       (void)MACIO_IN32(KEYLARGO_FCR3); udelay(10);
+       MACIO_OUT32(KEYLARGO_FCR4, save_fcr[4]);
+       (void)MACIO_IN32(KEYLARGO_FCR4); udelay(10);
+       if (macio->type == macio_pangea || macio->type == macio_intrepid) {
+               MACIO_OUT32(KEYLARGO_FCR5, save_fcr[5]);
+               (void)MACIO_IN32(KEYLARGO_FCR5); udelay(10);
+       }
+
+       dbdma_restore(macio, save_dbdma);
+
+       MACIO_OUT32(KEYLARGO_GPIO_LEVELS0, save_gpio_levels[0]);
+       MACIO_OUT32(KEYLARGO_GPIO_LEVELS1, save_gpio_levels[1]);
+       for (i=0; i<KEYLARGO_GPIO_EXTINT_CNT; i++)
+               MACIO_OUT8(KEYLARGO_GPIO_EXTINT_0+i, save_gpio_extint[i]);
+       for (i=0; i<KEYLARGO_GPIO_CNT; i++)
+               MACIO_OUT8(KEYLARGO_GPIO_0+i, save_gpio_normal[i]);
+
+       /* FIXME more black magic with OpenPIC ... */
+       if (pmac_mb.model_id == PMAC_TYPE_SAWTOOTH) {
+               MACIO_BIC(0x506e0, 0x00400000);
+               MACIO_BIC(0x506e0, 0x80000000);
+       }
+
+       UN_OUT(UNI_N_CLOCK_CNTL, save_unin_clock_ctl);
+       udelay(100);
+
+       return 0;
+}
+
+static long
+core99_sleep_state(struct device_node *node, long param, long value)
+{
+       /* Param == 1 means to enter the "fake sleep" mode that is
+        * used for CPU speed switch
+        */
+       if (param == 1) {
+               if (value == 1) {
+                       UN_OUT(UNI_N_HWINIT_STATE, UNI_N_HWINIT_STATE_SLEEPING);
+                       UN_OUT(UNI_N_POWER_MGT, UNI_N_POWER_MGT_IDLE2);
+               } else {
+                       UN_OUT(UNI_N_POWER_MGT, UNI_N_POWER_MGT_NORMAL);
+                       udelay(10);
+                       UN_OUT(UNI_N_HWINIT_STATE, UNI_N_HWINIT_STATE_RUNNING);
+                       udelay(10);
+               }
+               return 0;
+       }
+       if ((pmac_mb.board_flags & PMAC_MB_CAN_SLEEP) == 0)
+               return -EPERM;
+
+       if (value == 1)
+               return core99_sleep();
+       else if (value == 0)
+               return core99_wake_up();
+       return 0;
+}
+
+#endif /* CONFIG_POWER4 */
+
+static long
+generic_dev_can_wake(struct device_node *node, long param, long value)
+{
+       /* Todo: eventually check we are really dealing with on-board
+        * video device ...
+        */
+
+       if (pmac_mb.board_flags & PMAC_MB_MAY_SLEEP)
+               pmac_mb.board_flags |= PMAC_MB_CAN_SLEEP;
+       return 0;
+}
+
+static long generic_get_mb_info(struct device_node *node, long param, long value)
+{
+       switch(param) {
+               case PMAC_MB_INFO_MODEL:
+                       return pmac_mb.model_id;
+               case PMAC_MB_INFO_FLAGS:
+                       return pmac_mb.board_flags;
+               case PMAC_MB_INFO_NAME:
+                       /* hack hack hack... but should work */
+                       *((const char **)value) = pmac_mb.model_name;
+                       return 0;
+       }
+       return -EINVAL;
+}
+
+
+/*
+ * Table definitions
+ */
+
+/* Used on any machine
+ */
+static struct feature_table_entry any_features[] = {
+       { PMAC_FTR_GET_MB_INFO,         generic_get_mb_info },
+       { PMAC_FTR_DEVICE_CAN_WAKE,     generic_dev_can_wake },
+       { 0, NULL }
+};
+
+#ifndef CONFIG_POWER4
+
+/* OHare based motherboards. Currently, we only use these on the
+ * 2400,3400 and 3500 series powerbooks. Some older desktops seem
+ * to have issues with turning on/off those asic cells
+ */
+static struct feature_table_entry ohare_features[] = {
+       { PMAC_FTR_SCC_ENABLE,          ohare_htw_scc_enable },
+       { PMAC_FTR_SWIM3_ENABLE,        ohare_floppy_enable },
+       { PMAC_FTR_MESH_ENABLE,         ohare_mesh_enable },
+       { PMAC_FTR_IDE_ENABLE,          ohare_ide_enable},
+       { PMAC_FTR_IDE_RESET,           ohare_ide_reset},
+       { PMAC_FTR_SLEEP_STATE,         ohare_sleep_state },
+       { 0, NULL }
+};
+
+/* Heathrow desktop machines (Beige G3).
+ * Separated as some features couldn't be properly tested
+ * and the serial port control bits appear to confuse it.
+ */
+static struct feature_table_entry heathrow_desktop_features[] = {
+       { PMAC_FTR_SWIM3_ENABLE,        heathrow_floppy_enable },
+       { PMAC_FTR_MESH_ENABLE,         heathrow_mesh_enable },
+       { PMAC_FTR_IDE_ENABLE,          heathrow_ide_enable },
+       { PMAC_FTR_IDE_RESET,           heathrow_ide_reset },
+       { PMAC_FTR_BMAC_ENABLE,         heathrow_bmac_enable },
+       { 0, NULL }
+};
+
+/* Heathrow based laptop, that is the Wallstreet and mainstreet
+ * powerbooks.
+ */
+static struct feature_table_entry heathrow_laptop_features[] = {
+       { PMAC_FTR_SCC_ENABLE,          ohare_htw_scc_enable },
+       { PMAC_FTR_MODEM_ENABLE,        heathrow_modem_enable },
+       { PMAC_FTR_SWIM3_ENABLE,        heathrow_floppy_enable },
+       { PMAC_FTR_MESH_ENABLE,         heathrow_mesh_enable },
+       { PMAC_FTR_IDE_ENABLE,          heathrow_ide_enable },
+       { PMAC_FTR_IDE_RESET,           heathrow_ide_reset },
+       { PMAC_FTR_BMAC_ENABLE,         heathrow_bmac_enable },
+       { PMAC_FTR_SOUND_CHIP_ENABLE,   heathrow_sound_enable },
+       { PMAC_FTR_SLEEP_STATE,         heathrow_sleep_state },
+       { 0, NULL }
+};
+
+/* Paddington based machines
+ * The lombard (101) powerbook, first iMac models, B&W G3 and Yikes G4.
+ */
+static struct feature_table_entry paddington_features[] = {
+       { PMAC_FTR_SCC_ENABLE,          ohare_htw_scc_enable },
+       { PMAC_FTR_MODEM_ENABLE,        heathrow_modem_enable },
+       { PMAC_FTR_SWIM3_ENABLE,        heathrow_floppy_enable },
+       { PMAC_FTR_MESH_ENABLE,         heathrow_mesh_enable },
+       { PMAC_FTR_IDE_ENABLE,          heathrow_ide_enable },
+       { PMAC_FTR_IDE_RESET,           heathrow_ide_reset },
+       { PMAC_FTR_BMAC_ENABLE,         heathrow_bmac_enable },
+       { PMAC_FTR_SOUND_CHIP_ENABLE,   heathrow_sound_enable },
+       { PMAC_FTR_SLEEP_STATE,         heathrow_sleep_state },
+       { 0, NULL }
+};
+
+/* Core99 & MacRISC 2 machines (all machines released since the
+ * iBook (included), that is all AGP machines, except pangea
+ * chipset. The pangea chipset is the "combo" UniNorth/KeyLargo
+ * used on iBook2 & iMac "flow power".
+ */
+static struct feature_table_entry core99_features[] = {
+       { PMAC_FTR_SCC_ENABLE,          core99_scc_enable },
+       { PMAC_FTR_MODEM_ENABLE,        core99_modem_enable },
+       { PMAC_FTR_IDE_ENABLE,          core99_ide_enable },
+       { PMAC_FTR_IDE_RESET,           core99_ide_reset },
+       { PMAC_FTR_GMAC_ENABLE,         core99_gmac_enable },
+       { PMAC_FTR_GMAC_PHY_RESET,      core99_gmac_phy_reset },
+       { PMAC_FTR_SOUND_CHIP_ENABLE,   core99_sound_chip_enable },
+       { PMAC_FTR_AIRPORT_ENABLE,      core99_airport_enable },
+       { PMAC_FTR_USB_ENABLE,          core99_usb_enable },
+       { PMAC_FTR_1394_ENABLE,         core99_firewire_enable },
+       { PMAC_FTR_1394_CABLE_POWER,    core99_firewire_cable_power },
+       { PMAC_FTR_SLEEP_STATE,         core99_sleep_state },
+#ifdef CONFIG_SMP
+       { PMAC_FTR_RESET_CPU,           core99_reset_cpu },
+#endif /* CONFIG_SMP */
+       { PMAC_FTR_READ_GPIO,           core99_read_gpio },
+       { PMAC_FTR_WRITE_GPIO,          core99_write_gpio },
+       { 0, NULL }
+};
+
+/* RackMac
+ */
+static struct feature_table_entry rackmac_features[] = {
+       { PMAC_FTR_SCC_ENABLE,          core99_scc_enable },
+       { PMAC_FTR_IDE_ENABLE,          core99_ide_enable },
+       { PMAC_FTR_IDE_RESET,           core99_ide_reset },
+       { PMAC_FTR_GMAC_ENABLE,         core99_gmac_enable },
+       { PMAC_FTR_GMAC_PHY_RESET,      core99_gmac_phy_reset },
+       { PMAC_FTR_USB_ENABLE,          core99_usb_enable },
+       { PMAC_FTR_1394_ENABLE,         core99_firewire_enable },
+       { PMAC_FTR_1394_CABLE_POWER,    core99_firewire_cable_power },
+       { PMAC_FTR_SLEEP_STATE,         core99_sleep_state },
+#ifdef CONFIG_SMP
+       { PMAC_FTR_RESET_CPU,           core99_reset_cpu },
+#endif /* CONFIG_SMP */
+       { PMAC_FTR_READ_GPIO,           core99_read_gpio },
+       { PMAC_FTR_WRITE_GPIO,          core99_write_gpio },
+       { 0, NULL }
+};
+
+/* Pangea features
+ */
+static struct feature_table_entry pangea_features[] = {
+       { PMAC_FTR_SCC_ENABLE,          core99_scc_enable },
+       { PMAC_FTR_MODEM_ENABLE,        pangea_modem_enable },
+       { PMAC_FTR_IDE_ENABLE,          core99_ide_enable },
+       { PMAC_FTR_IDE_RESET,           core99_ide_reset },
+       { PMAC_FTR_GMAC_ENABLE,         core99_gmac_enable },
+       { PMAC_FTR_GMAC_PHY_RESET,      core99_gmac_phy_reset },
+       { PMAC_FTR_SOUND_CHIP_ENABLE,   core99_sound_chip_enable },
+       { PMAC_FTR_AIRPORT_ENABLE,      core99_airport_enable },
+       { PMAC_FTR_USB_ENABLE,          core99_usb_enable },
+       { PMAC_FTR_1394_ENABLE,         core99_firewire_enable },
+       { PMAC_FTR_1394_CABLE_POWER,    core99_firewire_cable_power },
+       { PMAC_FTR_SLEEP_STATE,         core99_sleep_state },
+       { PMAC_FTR_READ_GPIO,           core99_read_gpio },
+       { PMAC_FTR_WRITE_GPIO,          core99_write_gpio },
+       { 0, NULL }
+};
+
+/* Intrepid features
+ */
+static struct feature_table_entry intrepid_features[] = {
+       { PMAC_FTR_SCC_ENABLE,          core99_scc_enable },
+       { PMAC_FTR_MODEM_ENABLE,        pangea_modem_enable },
+       { PMAC_FTR_IDE_ENABLE,          core99_ide_enable },
+       { PMAC_FTR_IDE_RESET,           core99_ide_reset },
+       { PMAC_FTR_GMAC_ENABLE,         core99_gmac_enable },
+       { PMAC_FTR_GMAC_PHY_RESET,      core99_gmac_phy_reset },
+       { PMAC_FTR_SOUND_CHIP_ENABLE,   core99_sound_chip_enable },
+       { PMAC_FTR_AIRPORT_ENABLE,      core99_airport_enable },
+       { PMAC_FTR_USB_ENABLE,          core99_usb_enable },
+       { PMAC_FTR_1394_ENABLE,         core99_firewire_enable },
+       { PMAC_FTR_1394_CABLE_POWER,    core99_firewire_cable_power },
+       { PMAC_FTR_SLEEP_STATE,         core99_sleep_state },
+       { PMAC_FTR_READ_GPIO,           core99_read_gpio },
+       { PMAC_FTR_WRITE_GPIO,          core99_write_gpio },
+       { PMAC_FTR_AACK_DELAY_ENABLE,   intrepid_aack_delay_enable },
+       { 0, NULL }
+};
+
+#else /* CONFIG_POWER4 */
+
+/* G5 features
+ */
+static struct feature_table_entry g5_features[] = {
+       { PMAC_FTR_GMAC_ENABLE,         g5_gmac_enable },
+       { PMAC_FTR_1394_ENABLE,         g5_fw_enable },
+       { PMAC_FTR_ENABLE_MPIC,         g5_mpic_enable },
+       { PMAC_FTR_GMAC_PHY_RESET,      g5_eth_phy_reset },
+       { PMAC_FTR_SOUND_CHIP_ENABLE,   g5_i2s_enable },
+#ifdef CONFIG_SMP
+       { PMAC_FTR_RESET_CPU,           g5_reset_cpu },
+#endif /* CONFIG_SMP */
+       { PMAC_FTR_READ_GPIO,           core99_read_gpio },
+       { PMAC_FTR_WRITE_GPIO,          core99_write_gpio },
+       { 0, NULL }
+};
+
+#endif /* CONFIG_POWER4 */
+
+static struct pmac_mb_def pmac_mb_defs[] = {
+#ifndef CONFIG_POWER4
+       /*
+        * Desktops
+        */
+
+       {       "AAPL,8500",                    "PowerMac 8500/8600",
+               PMAC_TYPE_PSURGE,               NULL,
+               0
+       },
+       {       "AAPL,9500",                    "PowerMac 9500/9600",
+               PMAC_TYPE_PSURGE,               NULL,
+               0
+       },
+       {       "AAPL,7200",                    "PowerMac 7200",
+               PMAC_TYPE_PSURGE,               NULL,
+               0
+       },
+       {       "AAPL,7300",                    "PowerMac 7200/7300",
+               PMAC_TYPE_PSURGE,               NULL,
+               0
+       },
+       {       "AAPL,7500",                    "PowerMac 7500",
+               PMAC_TYPE_PSURGE,               NULL,
+               0
+       },
+       {       "AAPL,ShinerESB",               "Apple Network Server",
+               PMAC_TYPE_ANS,                  NULL,
+               0
+       },
+       {       "AAPL,e407",                    "Alchemy",
+               PMAC_TYPE_ALCHEMY,              NULL,
+               0
+       },
+       {       "AAPL,e411",                    "Gazelle",
+               PMAC_TYPE_GAZELLE,              NULL,
+               0
+       },
+       {       "AAPL,Gossamer",                "PowerMac G3 (Gossamer)",
+               PMAC_TYPE_GOSSAMER,             heathrow_desktop_features,
+               0
+       },
+       {       "AAPL,PowerMac G3",             "PowerMac G3 (Silk)",
+               PMAC_TYPE_SILK,                 heathrow_desktop_features,
+               0
+       },
+       {       "PowerMac1,1",                  "Blue&White G3",
+               PMAC_TYPE_YOSEMITE,             paddington_features,
+               0
+       },
+       {       "PowerMac1,2",                  "PowerMac G4 PCI Graphics",
+               PMAC_TYPE_YIKES,                paddington_features,
+               0
+       },
+       {       "PowerMac2,1",                  "iMac FireWire",
+               PMAC_TYPE_FW_IMAC,              core99_features,
+               PMAC_MB_MAY_SLEEP | PMAC_MB_OLD_CORE99
+       },
+       {       "PowerMac2,2",                  "iMac FireWire",
+               PMAC_TYPE_FW_IMAC,              core99_features,
+               PMAC_MB_MAY_SLEEP | PMAC_MB_OLD_CORE99
+       },
+       {       "PowerMac3,1",                  "PowerMac G4 AGP Graphics",
+               PMAC_TYPE_SAWTOOTH,             core99_features,
+               PMAC_MB_OLD_CORE99
+       },
+       {       "PowerMac3,2",                  "PowerMac G4 AGP Graphics",
+               PMAC_TYPE_SAWTOOTH,             core99_features,
+               PMAC_MB_MAY_SLEEP | PMAC_MB_OLD_CORE99
+       },
+       {       "PowerMac3,3",                  "PowerMac G4 AGP Graphics",
+               PMAC_TYPE_SAWTOOTH,             core99_features,
+               PMAC_MB_MAY_SLEEP | PMAC_MB_OLD_CORE99
+       },
+       {       "PowerMac3,4",                  "PowerMac G4 Silver",
+               PMAC_TYPE_QUICKSILVER,          core99_features,
+               PMAC_MB_MAY_SLEEP
+       },
+       {       "PowerMac3,5",                  "PowerMac G4 Silver",
+               PMAC_TYPE_QUICKSILVER,          core99_features,
+               PMAC_MB_MAY_SLEEP
+       },
+       {       "PowerMac3,6",                  "PowerMac G4 Windtunnel",
+               PMAC_TYPE_WINDTUNNEL,           core99_features,
+               PMAC_MB_MAY_SLEEP,
+       },
+       {       "PowerMac4,1",                  "iMac \"Flower Power\"",
+               PMAC_TYPE_PANGEA_IMAC,          pangea_features,
+               PMAC_MB_MAY_SLEEP
+       },
+       {       "PowerMac4,2",                  "Flat panel iMac",
+               PMAC_TYPE_FLAT_PANEL_IMAC,      pangea_features,
+               PMAC_MB_CAN_SLEEP
+       },
+       {       "PowerMac4,4",                  "eMac",
+               PMAC_TYPE_EMAC,                 core99_features,
+               PMAC_MB_MAY_SLEEP
+       },
+       {       "PowerMac5,1",                  "PowerMac G4 Cube",
+               PMAC_TYPE_CUBE,                 core99_features,
+               PMAC_MB_MAY_SLEEP | PMAC_MB_OLD_CORE99
+       },
+       {       "PowerMac6,1",                  "Flat panel iMac",
+               PMAC_TYPE_UNKNOWN_INTREPID,     intrepid_features,
+               PMAC_MB_MAY_SLEEP,
+       },
+       {       "PowerMac6,3",                  "Flat panel iMac",
+               PMAC_TYPE_UNKNOWN_INTREPID,     intrepid_features,
+               PMAC_MB_MAY_SLEEP,
+       },
+       {       "PowerMac6,4",                  "eMac",
+               PMAC_TYPE_UNKNOWN_INTREPID,     intrepid_features,
+               PMAC_MB_MAY_SLEEP,
+       },
+       {       "PowerMac10,1",                 "Mac mini",
+               PMAC_TYPE_UNKNOWN_INTREPID,     intrepid_features,
+               PMAC_MB_MAY_SLEEP | PMAC_MB_HAS_FW_POWER,
+       },
+       {       "iMac,1",                       "iMac (first generation)",
+               PMAC_TYPE_ORIG_IMAC,            paddington_features,
+               0
+       },
+
+       /*
+        * Xserve's
+        */
+
+       {       "RackMac1,1",                   "XServe",
+               PMAC_TYPE_RACKMAC,              rackmac_features,
+               0,
+       },
+       {       "RackMac1,2",                   "XServe rev. 2",
+               PMAC_TYPE_RACKMAC,              rackmac_features,
+               0,
+       },
+
+       /*
+        * Laptops
+        */
+
+       {       "AAPL,3400/2400",               "PowerBook 3400",
+               PMAC_TYPE_HOOPER,               ohare_features,
+               PMAC_MB_CAN_SLEEP | PMAC_MB_MOBILE
+       },
+       {       "AAPL,3500",                    "PowerBook 3500",
+               PMAC_TYPE_KANGA,                ohare_features,
+               PMAC_MB_CAN_SLEEP | PMAC_MB_MOBILE
+       },
+       {       "AAPL,PowerBook1998",           "PowerBook Wallstreet",
+               PMAC_TYPE_WALLSTREET,           heathrow_laptop_features,
+               PMAC_MB_CAN_SLEEP | PMAC_MB_MOBILE
+       },
+       {       "PowerBook1,1",                 "PowerBook 101 (Lombard)",
+               PMAC_TYPE_101_PBOOK,            paddington_features,
+               PMAC_MB_CAN_SLEEP | PMAC_MB_MOBILE
+       },
+       {       "PowerBook2,1",                 "iBook (first generation)",
+               PMAC_TYPE_ORIG_IBOOK,           core99_features,
+               PMAC_MB_CAN_SLEEP | PMAC_MB_OLD_CORE99 | PMAC_MB_MOBILE
+       },
+       {       "PowerBook2,2",                 "iBook FireWire",
+               PMAC_TYPE_FW_IBOOK,             core99_features,
+               PMAC_MB_MAY_SLEEP | PMAC_MB_HAS_FW_POWER |
+               PMAC_MB_OLD_CORE99 | PMAC_MB_MOBILE
+       },
+       {       "PowerBook3,1",                 "PowerBook Pismo",
+               PMAC_TYPE_PISMO,                core99_features,
+               PMAC_MB_MAY_SLEEP | PMAC_MB_HAS_FW_POWER |
+               PMAC_MB_OLD_CORE99 | PMAC_MB_MOBILE
+       },
+       {       "PowerBook3,2",                 "PowerBook Titanium",
+               PMAC_TYPE_TITANIUM,             core99_features,
+               PMAC_MB_MAY_SLEEP | PMAC_MB_HAS_FW_POWER | PMAC_MB_MOBILE
+       },
+       {       "PowerBook3,3",                 "PowerBook Titanium II",
+               PMAC_TYPE_TITANIUM2,            core99_features,
+               PMAC_MB_MAY_SLEEP | PMAC_MB_HAS_FW_POWER | PMAC_MB_MOBILE
+       },
+       {       "PowerBook3,4",                 "PowerBook Titanium III",
+               PMAC_TYPE_TITANIUM3,            core99_features,
+               PMAC_MB_MAY_SLEEP | PMAC_MB_HAS_FW_POWER | PMAC_MB_MOBILE
+       },
+       {       "PowerBook3,5",                 "PowerBook Titanium IV",
+               PMAC_TYPE_TITANIUM4,            core99_features,
+               PMAC_MB_MAY_SLEEP | PMAC_MB_HAS_FW_POWER | PMAC_MB_MOBILE
+       },
+       {       "PowerBook4,1",                 "iBook 2",
+               PMAC_TYPE_IBOOK2,               pangea_features,
+               PMAC_MB_MAY_SLEEP | PMAC_MB_HAS_FW_POWER | PMAC_MB_MOBILE
+       },
+       {       "PowerBook4,2",                 "iBook 2",
+               PMAC_TYPE_IBOOK2,               pangea_features,
+               PMAC_MB_MAY_SLEEP | PMAC_MB_HAS_FW_POWER | PMAC_MB_MOBILE
+       },
+       {       "PowerBook4,3",                 "iBook 2 rev. 2",
+               PMAC_TYPE_IBOOK2,               pangea_features,
+               PMAC_MB_MAY_SLEEP | PMAC_MB_HAS_FW_POWER | PMAC_MB_MOBILE
+       },
+       {       "PowerBook5,1",                 "PowerBook G4 17\"",
+               PMAC_TYPE_UNKNOWN_INTREPID,     intrepid_features,
+               PMAC_MB_HAS_FW_POWER | PMAC_MB_MOBILE,
+       },
+       {       "PowerBook5,2",                 "PowerBook G4 15\"",
+               PMAC_TYPE_UNKNOWN_INTREPID,     intrepid_features,
+               PMAC_MB_MAY_SLEEP | PMAC_MB_HAS_FW_POWER | PMAC_MB_MOBILE,
+       },
+       {       "PowerBook5,3",                 "PowerBook G4 17\"",
+               PMAC_TYPE_UNKNOWN_INTREPID,     intrepid_features,
+               PMAC_MB_MAY_SLEEP | PMAC_MB_HAS_FW_POWER | PMAC_MB_MOBILE,
+       },
+       {       "PowerBook5,4",                 "PowerBook G4 15\"",
+               PMAC_TYPE_UNKNOWN_INTREPID,     intrepid_features,
+               PMAC_MB_MAY_SLEEP | PMAC_MB_HAS_FW_POWER | PMAC_MB_MOBILE,
+       },
+       {       "PowerBook5,5",                 "PowerBook G4 17\"",
+               PMAC_TYPE_UNKNOWN_INTREPID,     intrepid_features,
+               PMAC_MB_MAY_SLEEP | PMAC_MB_HAS_FW_POWER | PMAC_MB_MOBILE,
+       },
+       {       "PowerBook5,6",                 "PowerBook G4 15\"",
+               PMAC_TYPE_UNKNOWN_INTREPID,     intrepid_features,
+               PMAC_MB_MAY_SLEEP | PMAC_MB_HAS_FW_POWER | PMAC_MB_MOBILE,
+       },
+       {       "PowerBook5,7",                 "PowerBook G4 17\"",
+               PMAC_TYPE_UNKNOWN_INTREPID,     intrepid_features,
+               PMAC_MB_MAY_SLEEP | PMAC_MB_HAS_FW_POWER | PMAC_MB_MOBILE,
+       },
+       {       "PowerBook6,1",                 "PowerBook G4 12\"",
+               PMAC_TYPE_UNKNOWN_INTREPID,     intrepid_features,
+               PMAC_MB_MAY_SLEEP | PMAC_MB_HAS_FW_POWER | PMAC_MB_MOBILE,
+       },
+       {       "PowerBook6,2",                 "PowerBook G4",
+               PMAC_TYPE_UNKNOWN_INTREPID,     intrepid_features,
+               PMAC_MB_MAY_SLEEP | PMAC_MB_HAS_FW_POWER | PMAC_MB_MOBILE,
+       },
+       {       "PowerBook6,3",                 "iBook G4",
+               PMAC_TYPE_UNKNOWN_INTREPID,     intrepid_features,
+               PMAC_MB_MAY_SLEEP | PMAC_MB_HAS_FW_POWER | PMAC_MB_MOBILE,
+       },
+       {       "PowerBook6,4",                 "PowerBook G4 12\"",
+               PMAC_TYPE_UNKNOWN_INTREPID,     intrepid_features,
+               PMAC_MB_MAY_SLEEP | PMAC_MB_HAS_FW_POWER | PMAC_MB_MOBILE,
+       },
+       {       "PowerBook6,5",                 "iBook G4",
+               PMAC_TYPE_UNKNOWN_INTREPID,     intrepid_features,
+               PMAC_MB_MAY_SLEEP | PMAC_MB_HAS_FW_POWER | PMAC_MB_MOBILE,
+       },
+       {       "PowerBook6,8",                 "PowerBook G4 12\"",
+               PMAC_TYPE_UNKNOWN_INTREPID,     intrepid_features,
+               PMAC_MB_MAY_SLEEP | PMAC_MB_HAS_FW_POWER | PMAC_MB_MOBILE,
+       },
+#else /* CONFIG_POWER4 */
+       {       "PowerMac7,2",                  "PowerMac G5",
+               PMAC_TYPE_POWERMAC_G5,          g5_features,
+               0,
+       },
+#ifdef CONFIG_PPC64
+       {       "PowerMac7,3",                  "PowerMac G5",
+               PMAC_TYPE_POWERMAC_G5,          g5_features,
+               0,
+       },
+       {       "PowerMac8,1",                  "iMac G5",
+               PMAC_TYPE_IMAC_G5,              g5_features,
+               0,
+       },
+       {       "PowerMac9,1",                  "PowerMac G5",
+               PMAC_TYPE_POWERMAC_G5_U3L,      g5_features,
+               0,
+       },
+       {       "RackMac3,1",                   "XServe G5",
+               PMAC_TYPE_XSERVE_G5,            g5_features,
+               0,
+       },
+#endif /* CONFIG_PPC64 */
+#endif /* CONFIG_POWER4 */
+};
+
+/*
+ * The toplevel feature_call callback
+ */
+long pmac_do_feature_call(unsigned int selector, ...)
+{
+       struct device_node *node;
+       long param, value;
+       int i;
+       feature_call func = NULL;
+       va_list args;
+
+       if (pmac_mb.features)
+               for (i=0; pmac_mb.features[i].function; i++)
+                       if (pmac_mb.features[i].selector == selector) {
+                               func = pmac_mb.features[i].function;
+                               break;
+                       }
+       if (!func)
+               for (i=0; any_features[i].function; i++)
+                       if (any_features[i].selector == selector) {
+                               func = any_features[i].function;
+                               break;
+                       }
+       if (!func)
+               return -ENODEV;
+
+       va_start(args, selector);
+       node = (struct device_node*)va_arg(args, void*);
+       param = va_arg(args, long);
+       value = va_arg(args, long);
+       va_end(args);
+
+       return func(node, param, value);
+}
+
+static int __init probe_motherboard(void)
+{
+       int i;
+       struct macio_chip *macio = &macio_chips[0];
+       const char *model = NULL;
+       struct device_node *dt;
+
+       /* Lookup known motherboard type in device-tree. First try an
+        * exact match on the "model" property, then try a "compatible"
+        * match is none is found.
+        */
+       dt = find_devices("device-tree");
+       if (dt != NULL)
+               model = (const char *) get_property(dt, "model", NULL);
+       for(i=0; model && i<(sizeof(pmac_mb_defs)/sizeof(struct pmac_mb_def)); i++) {
+           if (strcmp(model, pmac_mb_defs[i].model_string) == 0) {
+               pmac_mb = pmac_mb_defs[i];
+               goto found;
+           }
+       }
+       for(i=0; i<(sizeof(pmac_mb_defs)/sizeof(struct pmac_mb_def)); i++) {
+           if (machine_is_compatible(pmac_mb_defs[i].model_string)) {
+               pmac_mb = pmac_mb_defs[i];
+               goto found;
+           }
+       }
+
+       /* Fallback to selection depending on mac-io chip type */
+       switch(macio->type) {
+#ifndef CONFIG_POWER4
+           case macio_grand_central:
+               pmac_mb.model_id = PMAC_TYPE_PSURGE;
+               pmac_mb.model_name = "Unknown PowerSurge";
+               break;
+           case macio_ohare:
+               pmac_mb.model_id = PMAC_TYPE_UNKNOWN_OHARE;
+               pmac_mb.model_name = "Unknown OHare-based";
+               break;
+           case macio_heathrow:
+               pmac_mb.model_id = PMAC_TYPE_UNKNOWN_HEATHROW;
+               pmac_mb.model_name = "Unknown Heathrow-based";
+               pmac_mb.features = heathrow_desktop_features;
+               break;
+           case macio_paddington:
+               pmac_mb.model_id = PMAC_TYPE_UNKNOWN_PADDINGTON;
+               pmac_mb.model_name = "Unknown Paddington-based";
+               pmac_mb.features = paddington_features;
+               break;
+           case macio_keylargo:
+               pmac_mb.model_id = PMAC_TYPE_UNKNOWN_CORE99;
+               pmac_mb.model_name = "Unknown Keylargo-based";
+               pmac_mb.features = core99_features;
+               break;
+           case macio_pangea:
+               pmac_mb.model_id = PMAC_TYPE_UNKNOWN_PANGEA;
+               pmac_mb.model_name = "Unknown Pangea-based";
+               pmac_mb.features = pangea_features;
+               break;
+           case macio_intrepid:
+               pmac_mb.model_id = PMAC_TYPE_UNKNOWN_INTREPID;
+               pmac_mb.model_name = "Unknown Intrepid-based";
+               pmac_mb.features = intrepid_features;
+               break;
+#else /* CONFIG_POWER4 */
+       case macio_keylargo2:
+               pmac_mb.model_id = PMAC_TYPE_UNKNOWN_K2;
+               pmac_mb.model_name = "Unknown K2-based";
+               pmac_mb.features = g5_features;
+               break;
+#endif /* CONFIG_POWER4 */
+       default:
+               return -ENODEV;
+       }
+found:
+#ifndef CONFIG_POWER4
+       /* Fixup Hooper vs. Comet */
+       if (pmac_mb.model_id == PMAC_TYPE_HOOPER) {
+               u32 __iomem * mach_id_ptr = ioremap(0xf3000034, 4);
+               if (!mach_id_ptr)
+                       return -ENODEV;
+               /* Here, I used to disable the media-bay on comet. It
+                * appears this is wrong, the floppy connector is actually
+                * a kind of media-bay and works with the current driver.
+                */
+               if (__raw_readl(mach_id_ptr) & 0x20000000UL)
+                       pmac_mb.model_id = PMAC_TYPE_COMET;
+               iounmap(mach_id_ptr);
+       }
+#endif /* CONFIG_POWER4 */
+
+#ifdef CONFIG_6xx
+       /* Set default value of powersave_nap on machines that support it.
+        * It appears that uninorth rev 3 has a problem with it, we don't
+        * enable it on those. In theory, the flush-on-lock property is
+        * supposed to be set when not supported, but I'm not very confident
+        * that all Apple OF revs did it properly, I do it the paranoid way.
+        */
+       while (uninorth_base && uninorth_rev > 3) {
+               struct device_node *np = find_path_device("/cpus");
+               if (!np || !np->child) {
+                       printk(KERN_WARNING "Can't find CPU(s) in device tree !\n");
+                       break;
+               }
+               np = np->child;
+               /* Nap mode not supported on SMP */
+               if (np->sibling)
+                       break;
+               /* Nap mode not supported if flush-on-lock property is present */
+               if (get_property(np, "flush-on-lock", NULL))
+                       break;
+               powersave_nap = 1;
+               printk(KERN_INFO "Processor NAP mode on idle enabled.\n");
+               break;
+       }
+
+       /* On CPUs that support it (750FX), lowspeed by default during
+        * NAP mode
+        */
+       powersave_lowspeed = 1;
+#endif /* CONFIG_6xx */
+#ifdef CONFIG_POWER4
+       powersave_nap = 1;
+#endif
+       /* Check for "mobile" machine */
+       if (model && (strncmp(model, "PowerBook", 9) == 0
+                  || strncmp(model, "iBook", 5) == 0))
+               pmac_mb.board_flags |= PMAC_MB_MOBILE;
+
+
+       printk(KERN_INFO "PowerMac motherboard: %s\n", pmac_mb.model_name);
+       return 0;
+}
+
+/* Initialize the Core99 UniNorth host bridge and memory controller
+ */
+static void __init probe_uninorth(void)
+{
+       unsigned long actrl;
+
+       /* Locate core99 Uni-N */
+       uninorth_node = of_find_node_by_name(NULL, "uni-n");
+       /* Locate G5 u3 */
+       if (uninorth_node == NULL) {
+               uninorth_node = of_find_node_by_name(NULL, "u3");
+               uninorth_u3 = 1;
+       }
+       if (uninorth_node && uninorth_node->n_addrs > 0) {
+               unsigned long address = uninorth_node->addrs[0].address;
+               uninorth_base = ioremap(address, 0x40000);
+               uninorth_rev = in_be32(UN_REG(UNI_N_VERSION));
+               if (uninorth_u3)
+                       u3_ht = ioremap(address + U3_HT_CONFIG_BASE, 0x1000);
+       } else
+               uninorth_node = NULL;
+
+       if (!uninorth_node)
+               return;
+
+       printk(KERN_INFO "Found %s memory controller & host bridge, revision: %d\n",
+              uninorth_u3 ? "U3" : "UniNorth", uninorth_rev);
+       printk(KERN_INFO "Mapped at 0x%08lx\n", (unsigned long)uninorth_base);
+
+       /* Set the arbitrer QAck delay according to what Apple does
+        */
+       if (uninorth_rev < 0x11) {
+               actrl = UN_IN(UNI_N_ARB_CTRL) & ~UNI_N_ARB_CTRL_QACK_DELAY_MASK;
+               actrl |= ((uninorth_rev < 3) ? UNI_N_ARB_CTRL_QACK_DELAY105 :
+                       UNI_N_ARB_CTRL_QACK_DELAY) << UNI_N_ARB_CTRL_QACK_DELAY_SHIFT;
+               UN_OUT(UNI_N_ARB_CTRL, actrl);
+       }
+
+       /* Some more magic as done by them in recent MacOS X on UniNorth
+        * revs 1.5 to 2.O and Pangea. Seem to toggle the UniN Maxbus/PCI
+        * memory timeout
+        */
+       if ((uninorth_rev >= 0x11 && uninorth_rev <= 0x24) || uninorth_rev == 0xc0)
+               UN_OUT(0x2160, UN_IN(0x2160) & 0x00ffffff);
+}
+
+static void __init probe_one_macio(const char *name, const char *compat, int type)
+{
+       struct device_node*     node;
+       int                     i;
+       volatile u32 __iomem *  base;
+       u32*                    revp;
+
+       node = find_devices(name);
+       if (!node || !node->n_addrs)
+               return;
+       if (compat)
+               do {
+                       if (device_is_compatible(node, compat))
+                               break;
+                       node = node->next;
+               } while (node);
+       if (!node)
+               return;
+       for(i=0; i<MAX_MACIO_CHIPS; i++) {
+               if (!macio_chips[i].of_node)
+                       break;
+               if (macio_chips[i].of_node == node)
+                       return;
+       }
+       if (i >= MAX_MACIO_CHIPS) {
+               printk(KERN_ERR "pmac_feature: Please increase MAX_MACIO_CHIPS !\n");
+               printk(KERN_ERR "pmac_feature: %s skipped\n", node->full_name);
+               return;
+       }
+       base = ioremap(node->addrs[0].address, node->addrs[0].size);
+       if (!base) {
+               printk(KERN_ERR "pmac_feature: Can't map mac-io chip !\n");
+               return;
+       }
+       if (type == macio_keylargo) {
+               u32 *did = (u32 *)get_property(node, "device-id", NULL);
+               if (*did == 0x00000025)
+                       type = macio_pangea;
+               if (*did == 0x0000003e)
+                       type = macio_intrepid;
+       }
+       macio_chips[i].of_node  = node;
+       macio_chips[i].type     = type;
+       macio_chips[i].base     = base;
+       macio_chips[i].flags    = MACIO_FLAG_SCCB_ON | MACIO_FLAG_SCCB_ON;
+       macio_chips[i].name     = macio_names[type];
+       revp = (u32 *)get_property(node, "revision-id", NULL);
+       if (revp)
+               macio_chips[i].rev = *revp;
+       printk(KERN_INFO "Found a %s mac-io controller, rev: %d, mapped at 0x%p\n",
+               macio_names[type], macio_chips[i].rev, macio_chips[i].base);
+}
+
+static int __init
+probe_macios(void)
+{
+       /* Warning, ordering is important */
+       probe_one_macio("gc", NULL, macio_grand_central);
+       probe_one_macio("ohare", NULL, macio_ohare);
+       probe_one_macio("pci106b,7", NULL, macio_ohareII);
+       probe_one_macio("mac-io", "keylargo", macio_keylargo);
+       probe_one_macio("mac-io", "paddington", macio_paddington);
+       probe_one_macio("mac-io", "gatwick", macio_gatwick);
+       probe_one_macio("mac-io", "heathrow", macio_heathrow);
+       probe_one_macio("mac-io", "K2-Keylargo", macio_keylargo2);
+
+       /* Make sure the "main" macio chip appear first */
+       if (macio_chips[0].type == macio_gatwick
+           && macio_chips[1].type == macio_heathrow) {
+               struct macio_chip temp = macio_chips[0];
+               macio_chips[0] = macio_chips[1];
+               macio_chips[1] = temp;
+       }
+       if (macio_chips[0].type == macio_ohareII
+           && macio_chips[1].type == macio_ohare) {
+               struct macio_chip temp = macio_chips[0];
+               macio_chips[0] = macio_chips[1];
+               macio_chips[1] = temp;
+       }
+       macio_chips[0].lbus.index = 0;
+       macio_chips[1].lbus.index = 1;
+
+       return (macio_chips[0].of_node == NULL) ? -ENODEV : 0;
+}
+
+static void __init
+initial_serial_shutdown(struct device_node *np)
+{
+       int len;
+       struct slot_names_prop {
+               int     count;
+               char    name[1];
+       } *slots;
+       char *conn;
+       int port_type = PMAC_SCC_ASYNC;
+       int modem = 0;
+
+       slots = (struct slot_names_prop *)get_property(np, "slot-names", &len);
+       conn = get_property(np, "AAPL,connector", &len);
+       if (conn && (strcmp(conn, "infrared") == 0))
+               port_type = PMAC_SCC_IRDA;
+       else if (device_is_compatible(np, "cobalt"))
+               modem = 1;
+       else if (slots && slots->count > 0) {
+               if (strcmp(slots->name, "IrDA") == 0)
+                       port_type = PMAC_SCC_IRDA;
+               else if (strcmp(slots->name, "Modem") == 0)
+                       modem = 1;
+       }
+       if (modem)
+               pmac_call_feature(PMAC_FTR_MODEM_ENABLE, np, 0, 0);
+       pmac_call_feature(PMAC_FTR_SCC_ENABLE, np, port_type, 0);
+}
+
+static void __init
+set_initial_features(void)
+{
+       struct device_node *np;
+
+       /* That hack appears to be necessary for some StarMax motherboards
+        * but I'm not too sure it was audited for side-effects on other
+        * ohare based machines...
+        * Since I still have difficulties figuring the right way to
+        * differenciate them all and since that hack was there for a long
+        * time, I'll keep it around
+        */
+       if (macio_chips[0].type == macio_ohare && !find_devices("via-pmu")) {
+               struct macio_chip *macio = &macio_chips[0];
+               MACIO_OUT32(OHARE_FCR, STARMAX_FEATURES);
+       } else if (macio_chips[0].type == macio_ohare) {
+               struct macio_chip *macio = &macio_chips[0];
+               MACIO_BIS(OHARE_FCR, OH_IOBUS_ENABLE);
+       } else if (macio_chips[1].type == macio_ohare) {
+               struct macio_chip *macio = &macio_chips[1];
+               MACIO_BIS(OHARE_FCR, OH_IOBUS_ENABLE);
+       }
+
+#ifdef CONFIG_POWER4
+       if (macio_chips[0].type == macio_keylargo2) {
+#ifndef CONFIG_SMP
+               /* On SMP machines running UP, we have the second CPU eating
+                * bus cycles. We need to take it off the bus. This is done
+                * from pmac_smp for SMP kernels running on one CPU
+                */
+               np = of_find_node_by_type(NULL, "cpu");
+               if (np != NULL)
+                       np = of_find_node_by_type(np, "cpu");
+               if (np != NULL) {
+                       g5_phy_disable_cpu1();
+                       of_node_put(np);
+               }
+#endif /* CONFIG_SMP */
+               /* Enable GMAC for now for PCI probing. It will be disabled
+                * later on after PCI probe
+                */
+               np = of_find_node_by_name(NULL, "ethernet");
+               while(np) {
+                       if (device_is_compatible(np, "K2-GMAC"))
+                               g5_gmac_enable(np, 0, 1);
+                       np = of_find_node_by_name(np, "ethernet");
+               }
+
+               /* Enable FW before PCI probe. Will be disabled later on
+                * Note: We should have a batter way to check that we are
+                * dealing with uninorth internal cell and not a PCI cell
+                * on the external PCI. The code below works though.
+                */
+               np = of_find_node_by_name(NULL, "firewire");
+               while(np) {
+                       if (device_is_compatible(np, "pci106b,5811")) {
+                               macio_chips[0].flags |= MACIO_FLAG_FW_SUPPORTED;
+                               g5_fw_enable(np, 0, 1);
+                       }
+                       np = of_find_node_by_name(np, "firewire");
+               }
+       }
+#else /* CONFIG_POWER4 */
+
+       if (macio_chips[0].type == macio_keylargo ||
+           macio_chips[0].type == macio_pangea ||
+           macio_chips[0].type == macio_intrepid) {
+               /* Enable GMAC for now for PCI probing. It will be disabled
+                * later on after PCI probe
+                */
+               np = of_find_node_by_name(NULL, "ethernet");
+               while(np) {
+                       if (np->parent
+                           && device_is_compatible(np->parent, "uni-north")
+                           && device_is_compatible(np, "gmac"))
+                               core99_gmac_enable(np, 0, 1);
+                       np = of_find_node_by_name(np, "ethernet");
+               }
+
+               /* Enable FW before PCI probe. Will be disabled later on
+                * Note: We should have a batter way to check that we are
+                * dealing with uninorth internal cell and not a PCI cell
+                * on the external PCI. The code below works though.
+                */
+               np = of_find_node_by_name(NULL, "firewire");
+               while(np) {
+                       if (np->parent
+                           && device_is_compatible(np->parent, "uni-north")
+                           && (device_is_compatible(np, "pci106b,18") ||
+                               device_is_compatible(np, "pci106b,30") ||
+                               device_is_compatible(np, "pci11c1,5811"))) {
+                               macio_chips[0].flags |= MACIO_FLAG_FW_SUPPORTED;
+                               core99_firewire_enable(np, 0, 1);
+                       }
+                       np = of_find_node_by_name(np, "firewire");
+               }
+
+               /* Enable ATA-100 before PCI probe. */
+               np = of_find_node_by_name(NULL, "ata-6");
+               while(np) {
+                       if (np->parent
+                           && device_is_compatible(np->parent, "uni-north")
+                           && device_is_compatible(np, "kauai-ata")) {
+                               core99_ata100_enable(np, 1);
+                       }
+                       np = of_find_node_by_name(np, "ata-6");
+               }
+
+               /* Switch airport off */
+               np = find_devices("radio");
+               while(np) {
+                       if (np && np->parent == macio_chips[0].of_node) {
+                               macio_chips[0].flags |= MACIO_FLAG_AIRPORT_ON;
+                               core99_airport_enable(np, 0, 0);
+                       }
+                       np = np->next;
+               }
+       }
+
+       /* On all machines that support sound PM, switch sound off */
+       if (macio_chips[0].of_node)
+               pmac_do_feature_call(PMAC_FTR_SOUND_CHIP_ENABLE,
+                       macio_chips[0].of_node, 0, 0);
+
+       /* While on some desktop G3s, we turn it back on */
+       if (macio_chips[0].of_node && macio_chips[0].type == macio_heathrow
+               && (pmac_mb.model_id == PMAC_TYPE_GOSSAMER ||
+                   pmac_mb.model_id == PMAC_TYPE_SILK)) {
+               struct macio_chip *macio = &macio_chips[0];
+               MACIO_BIS(HEATHROW_FCR, HRW_SOUND_CLK_ENABLE);
+               MACIO_BIC(HEATHROW_FCR, HRW_SOUND_POWER_N);
+       }
+
+       /* Some machine models need the clock chip to be properly setup for
+        * clock spreading now. This should be a platform function but we
+        * don't do these at the moment
+        */
+       pmac_tweak_clock_spreading(1);
+
+#endif /* CONFIG_POWER4 */
+
+       /* On all machines, switch modem & serial ports off */
+       np = find_devices("ch-a");
+       while(np) {
+               initial_serial_shutdown(np);
+               np = np->next;
+       }
+       np = find_devices("ch-b");
+       while(np) {
+               initial_serial_shutdown(np);
+               np = np->next;
+       }
+}
+
+void __init
+pmac_feature_init(void)
+{
+       /* Detect the UniNorth memory controller */
+       probe_uninorth();
+
+       /* Probe mac-io controllers */
+       if (probe_macios()) {
+               printk(KERN_WARNING "No mac-io chip found\n");
+               return;
+       }
+
+       /* Setup low-level i2c stuffs */
+       pmac_init_low_i2c();
+
+       /* Probe machine type */
+       if (probe_motherboard())
+               printk(KERN_WARNING "Unknown PowerMac !\n");
+
+       /* Set some initial features (turn off some chips that will
+        * be later turned on)
+        */
+       set_initial_features();
+}
+
+int __init pmac_feature_late_init(void)
+{
+#if 0
+       struct device_node *np;
+
+       /* Request some resources late */
+       if (uninorth_node)
+               request_OF_resource(uninorth_node, 0, NULL);
+       np = find_devices("hammerhead");
+       if (np)
+               request_OF_resource(np, 0, NULL);
+       np = find_devices("interrupt-controller");
+       if (np)
+               request_OF_resource(np, 0, NULL);
+#endif
+       return 0;
+}
+
+device_initcall(pmac_feature_late_init);
+
+#if 0
+static void dump_HT_speeds(char *name, u32 cfg, u32 frq)
+{
+       int     freqs[16] = { 200,300,400,500,600,800,1000,0,0,0,0,0,0,0,0,0 };
+       int     bits[8] = { 8,16,0,32,2,4,0,0 };
+       int     freq = (frq >> 8) & 0xf;
+
+       if (freqs[freq] == 0)
+               printk("%s: Unknown HT link frequency %x\n", name, freq);
+       else
+               printk("%s: %d MHz on main link, (%d in / %d out) bits width\n",
+                      name, freqs[freq],
+                      bits[(cfg >> 28) & 0x7], bits[(cfg >> 24) & 0x7]);
+}
+
+void __init pmac_check_ht_link(void)
+{
+#if 0 /* Disabled for now */
+       u32     ufreq, freq, ucfg, cfg;
+       struct device_node *pcix_node;
+       u8      px_bus, px_devfn;
+       struct pci_controller *px_hose;
+
+       (void)in_be32(u3_ht + U3_HT_LINK_COMMAND);
+       ucfg = cfg = in_be32(u3_ht + U3_HT_LINK_CONFIG);
+       ufreq = freq = in_be32(u3_ht + U3_HT_LINK_FREQ);
+       dump_HT_speeds("U3 HyperTransport", cfg, freq);
+
+       pcix_node = of_find_compatible_node(NULL, "pci", "pci-x");
+       if (pcix_node == NULL) {
+               printk("No PCI-X bridge found\n");
+               return;
+       }
+       if (pci_device_from_OF_node(pcix_node, &px_bus, &px_devfn) != 0) {
+               printk("PCI-X bridge found but not matched to pci\n");
+               return;
+       }
+       px_hose = pci_find_hose_for_OF_device(pcix_node);
+       if (px_hose == NULL) {
+               printk("PCI-X bridge found but not matched to host\n");
+               return;
+       }       
+       early_read_config_dword(px_hose, px_bus, px_devfn, 0xc4, &cfg);
+       early_read_config_dword(px_hose, px_bus, px_devfn, 0xcc, &freq);
+       dump_HT_speeds("PCI-X HT Uplink", cfg, freq);
+       early_read_config_dword(px_hose, px_bus, px_devfn, 0xc8, &cfg);
+       early_read_config_dword(px_hose, px_bus, px_devfn, 0xd0, &freq);
+       dump_HT_speeds("PCI-X HT Downlink", cfg, freq);
+#endif
+}
+
+#endif /* CONFIG_POWER4 */
+
+/*
+ * Early video resume hook
+ */
+
+static void (*pmac_early_vresume_proc)(void *data);
+static void *pmac_early_vresume_data;
+
+void pmac_set_early_video_resume(void (*proc)(void *data), void *data)
+{
+       if (_machine != _MACH_Pmac)
+               return;
+       preempt_disable();
+       pmac_early_vresume_proc = proc;
+       pmac_early_vresume_data = data;
+       preempt_enable();
+}
+EXPORT_SYMBOL(pmac_set_early_video_resume);
+
+void pmac_call_early_video_resume(void)
+{
+       if (pmac_early_vresume_proc)
+               pmac_early_vresume_proc(pmac_early_vresume_data);
+}
+
+/*
+ * AGP related suspend/resume code
+ */
+
+static struct pci_dev *pmac_agp_bridge;
+static int (*pmac_agp_suspend)(struct pci_dev *bridge);
+static int (*pmac_agp_resume)(struct pci_dev *bridge);
+
+void pmac_register_agp_pm(struct pci_dev *bridge,
+                                int (*suspend)(struct pci_dev *bridge),
+                                int (*resume)(struct pci_dev *bridge))
+{
+       if (suspend || resume) {
+               pmac_agp_bridge = bridge;
+               pmac_agp_suspend = suspend;
+               pmac_agp_resume = resume;
+               return;
+       }
+       if (bridge != pmac_agp_bridge)
+               return;
+       pmac_agp_suspend = pmac_agp_resume = NULL;
+       return;
+}
+EXPORT_SYMBOL(pmac_register_agp_pm);
+
+void pmac_suspend_agp_for_card(struct pci_dev *dev)
+{
+       if (pmac_agp_bridge == NULL || pmac_agp_suspend == NULL)
+               return;
+       if (pmac_agp_bridge->bus != dev->bus)
+               return;
+       pmac_agp_suspend(pmac_agp_bridge);
+}
+EXPORT_SYMBOL(pmac_suspend_agp_for_card);
+
+void pmac_resume_agp_for_card(struct pci_dev *dev)
+{
+       if (pmac_agp_bridge == NULL || pmac_agp_resume == NULL)
+               return;
+       if (pmac_agp_bridge->bus != dev->bus)
+               return;
+       pmac_agp_resume(pmac_agp_bridge);
+}
+EXPORT_SYMBOL(pmac_resume_agp_for_card);
diff --git a/arch/powerpc/platforms/powermac/pmac_low_i2c.c b/arch/powerpc/platforms/powermac/pmac_low_i2c.c
new file mode 100644 (file)
index 0000000..f3f39e8
--- /dev/null
@@ -0,0 +1,523 @@
+/*
+ *  arch/ppc/platforms/pmac_low_i2c.c
+ *
+ *  Copyright (C) 2003 Ben. Herrenschmidt (benh@kernel.crashing.org)
+ *
+ *  This program is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU General Public License
+ *  as published by the Free Software Foundation; either version
+ *  2 of the License, or (at your option) any later version.
+ *
+ *  This file contains some low-level i2c access routines that
+ *  need to be used by various bits of the PowerMac platform code
+ *  at times where the real asynchronous & interrupt driven driver
+ *  cannot be used. The API borrows some semantics from the darwin
+ *  driver in order to ease the implementation of the platform
+ *  properties parser
+ */
+
+#undef DEBUG
+
+#include <linux/config.h>
+#include <linux/types.h>
+#include <linux/sched.h>
+#include <linux/init.h>
+#include <linux/module.h>
+#include <linux/adb.h>
+#include <linux/pmu.h>
+#include <asm/keylargo.h>
+#include <asm/uninorth.h>
+#include <asm/io.h>
+#include <asm/prom.h>
+#include <asm/machdep.h>
+#include <asm/pmac_low_i2c.h>
+
+#define MAX_LOW_I2C_HOST       4
+
+#ifdef DEBUG
+#define DBG(x...) do {\
+               printk(KERN_DEBUG "KW:" x);     \
+       } while(0)
+#else
+#define DBG(x...)
+#endif
+
+struct low_i2c_host;
+
+typedef int (*low_i2c_func_t)(struct low_i2c_host *host, u8 addr, u8 sub, u8 *data, int len);
+
+struct low_i2c_host
+{
+       struct device_node      *np;            /* OF device node */
+       struct semaphore        mutex;          /* Access mutex for use by i2c-keywest */
+       low_i2c_func_t          func;           /* Access function */
+       unsigned int            is_open : 1;    /* Poor man's access control */
+       int                     mode;           /* Current mode */
+       int                     channel;        /* Current channel */
+       int                     num_channels;   /* Number of channels */
+       void __iomem            *base;          /* For keywest-i2c, base address */
+       int                     bsteps;         /* And register stepping */
+       int                     speed;          /* And speed */
+};
+
+static struct low_i2c_host     low_i2c_hosts[MAX_LOW_I2C_HOST];
+
+/* No locking is necessary on allocation, we are running way before
+ * anything can race with us
+ */
+static struct low_i2c_host *find_low_i2c_host(struct device_node *np)
+{
+       int i;
+
+       for (i = 0; i < MAX_LOW_I2C_HOST; i++)
+               if (low_i2c_hosts[i].np == np)
+                       return &low_i2c_hosts[i];
+       return NULL;
+}
+
+/*
+ *
+ * i2c-keywest implementation (UniNorth, U2, U3, Keylargo's)
+ *
+ */
+
+/*
+ * Keywest i2c definitions borrowed from drivers/i2c/i2c-keywest.h,
+ * should be moved somewhere in include/asm-ppc/
+ */
+/* Register indices */
+typedef enum {
+       reg_mode = 0,
+       reg_control,
+       reg_status,
+       reg_isr,
+       reg_ier,
+       reg_addr,
+       reg_subaddr,
+       reg_data
+} reg_t;
+
+
+/* Mode register */
+#define KW_I2C_MODE_100KHZ     0x00
+#define KW_I2C_MODE_50KHZ      0x01
+#define KW_I2C_MODE_25KHZ      0x02
+#define KW_I2C_MODE_DUMB       0x00
+#define KW_I2C_MODE_STANDARD   0x04
+#define KW_I2C_MODE_STANDARDSUB        0x08
+#define KW_I2C_MODE_COMBINED   0x0C
+#define KW_I2C_MODE_MODE_MASK  0x0C
+#define KW_I2C_MODE_CHAN_MASK  0xF0
+
+/* Control register */
+#define KW_I2C_CTL_AAK         0x01
+#define KW_I2C_CTL_XADDR       0x02
+#define KW_I2C_CTL_STOP                0x04
+#define KW_I2C_CTL_START       0x08
+
+/* Status register */
+#define KW_I2C_STAT_BUSY       0x01
+#define KW_I2C_STAT_LAST_AAK   0x02
+#define KW_I2C_STAT_LAST_RW    0x04
+#define KW_I2C_STAT_SDA                0x08
+#define KW_I2C_STAT_SCL                0x10
+
+/* IER & ISR registers */
+#define KW_I2C_IRQ_DATA                0x01
+#define KW_I2C_IRQ_ADDR                0x02
+#define KW_I2C_IRQ_STOP                0x04
+#define KW_I2C_IRQ_START       0x08
+#define KW_I2C_IRQ_MASK                0x0F
+
+/* State machine states */
+enum {
+       state_idle,
+       state_addr,
+       state_read,
+       state_write,
+       state_stop,
+       state_dead
+};
+
+#define WRONG_STATE(name) do {\
+               printk(KERN_DEBUG "KW: wrong state. Got %s, state: %s (isr: %02x)\n", \
+                      name, __kw_state_names[state], isr); \
+       } while(0)
+
+static const char *__kw_state_names[] = {
+       "state_idle",
+       "state_addr",
+       "state_read",
+       "state_write",
+       "state_stop",
+       "state_dead"
+};
+
+static inline u8 __kw_read_reg(struct low_i2c_host *host, reg_t reg)
+{
+       return readb(host->base + (((unsigned int)reg) << host->bsteps));
+}
+
+static inline void __kw_write_reg(struct low_i2c_host *host, reg_t reg, u8 val)
+{
+       writeb(val, host->base + (((unsigned)reg) << host->bsteps));
+       (void)__kw_read_reg(host, reg_subaddr);
+}
+
+#define kw_write_reg(reg, val) __kw_write_reg(host, reg, val) 
+#define kw_read_reg(reg)       __kw_read_reg(host, reg) 
+
+
+/* Don't schedule, the g5 fan controller is too
+ * timing sensitive
+ */
+static u8 kw_wait_interrupt(struct low_i2c_host* host)
+{
+       int i, j;
+       u8 isr;
+       
+       for (i = 0; i < 100000; i++) {
+               isr = kw_read_reg(reg_isr) & KW_I2C_IRQ_MASK;
+               if (isr != 0)
+                       return isr;
+
+               /* This code is used with the timebase frozen, we cannot rely
+                * on udelay ! For now, just use a bogus loop
+                */
+               for (j = 1; j < 10000; j++)
+                       mb();
+       }
+       return isr;
+}
+
+static int kw_handle_interrupt(struct low_i2c_host *host, int state, int rw, int *rc, u8 **data, int *len, u8 isr)
+{
+       u8 ack;
+
+       DBG("kw_handle_interrupt(%s, isr: %x)\n", __kw_state_names[state], isr);
+
+       if (isr == 0) {
+               if (state != state_stop) {
+                       DBG("KW: Timeout !\n");
+                       *rc = -EIO;
+                       goto stop;
+               }
+               if (state == state_stop) {
+                       ack = kw_read_reg(reg_status);
+                       if (!(ack & KW_I2C_STAT_BUSY)) {
+                               state = state_idle;
+                               kw_write_reg(reg_ier, 0x00);
+                       }
+               }
+               return state;
+       }
+
+       if (isr & KW_I2C_IRQ_ADDR) {
+               ack = kw_read_reg(reg_status);
+               if (state != state_addr) {
+                       kw_write_reg(reg_isr, KW_I2C_IRQ_ADDR);
+                       WRONG_STATE("KW_I2C_IRQ_ADDR"); 
+                       *rc = -EIO;
+                       goto stop;
+               }
+               if ((ack & KW_I2C_STAT_LAST_AAK) == 0) {                        
+                       *rc = -ENODEV;
+                       DBG("KW: NAK on address\n");
+                       return state_stop;                   
+               } else {
+                       if (rw) {
+                               state = state_read;
+                               if (*len > 1)
+                                       kw_write_reg(reg_control, KW_I2C_CTL_AAK);
+                       } else {
+                               state = state_write;
+                               kw_write_reg(reg_data, **data);
+                               (*data)++; (*len)--;
+                       }
+               }
+               kw_write_reg(reg_isr, KW_I2C_IRQ_ADDR);
+       }
+
+       if (isr & KW_I2C_IRQ_DATA) {
+               if (state == state_read) {
+                       **data = kw_read_reg(reg_data);
+                       (*data)++; (*len)--;
+                       kw_write_reg(reg_isr, KW_I2C_IRQ_DATA);
+                       if ((*len) == 0)
+                               state = state_stop;
+                       else if ((*len) == 1)
+                               kw_write_reg(reg_control, 0);
+               } else if (state == state_write) {
+                       ack = kw_read_reg(reg_status);
+                       if ((ack & KW_I2C_STAT_LAST_AAK) == 0) {
+                               DBG("KW: nack on data write\n");
+                               *rc = -EIO;
+                               goto stop;
+                       } else if (*len) {
+                               kw_write_reg(reg_data, **data);
+                               (*data)++; (*len)--;
+                       } else {
+                               kw_write_reg(reg_control, KW_I2C_CTL_STOP);
+                               state = state_stop;
+                               *rc = 0;
+                       }
+                       kw_write_reg(reg_isr, KW_I2C_IRQ_DATA);
+               } else {
+                       kw_write_reg(reg_isr, KW_I2C_IRQ_DATA);
+                       WRONG_STATE("KW_I2C_IRQ_DATA"); 
+                       if (state != state_stop) {
+                               *rc = -EIO;
+                               goto stop;
+                       }
+               }
+       }
+
+       if (isr & KW_I2C_IRQ_STOP) {
+               kw_write_reg(reg_isr, KW_I2C_IRQ_STOP);
+               if (state != state_stop) {
+                       WRONG_STATE("KW_I2C_IRQ_STOP");
+                       *rc = -EIO;
+               }
+               return state_idle;
+       }
+
+       if (isr & KW_I2C_IRQ_START)
+               kw_write_reg(reg_isr, KW_I2C_IRQ_START);
+
+       return state;
+
+ stop:
+       kw_write_reg(reg_control, KW_I2C_CTL_STOP);     
+       return state_stop;
+}
+
+static int keywest_low_i2c_func(struct low_i2c_host *host, u8 addr, u8 subaddr, u8 *data, int len)
+{
+       u8 mode_reg = host->speed;
+       int state = state_addr;
+       int rc = 0;
+
+       /* Setup mode & subaddress if any */
+       switch(host->mode) {
+       case pmac_low_i2c_mode_dumb:
+               printk(KERN_ERR "low_i2c: Dumb mode not supported !\n");
+               return -EINVAL;
+       case pmac_low_i2c_mode_std:
+               mode_reg |= KW_I2C_MODE_STANDARD;
+               break;
+       case pmac_low_i2c_mode_stdsub:
+               mode_reg |= KW_I2C_MODE_STANDARDSUB;
+               break;
+       case pmac_low_i2c_mode_combined:
+               mode_reg |= KW_I2C_MODE_COMBINED;
+               break;
+       }
+
+       /* Setup channel & clear pending irqs */
+       kw_write_reg(reg_isr, kw_read_reg(reg_isr));
+       kw_write_reg(reg_mode, mode_reg | (host->channel << 4));
+       kw_write_reg(reg_status, 0);
+
+       /* Set up address and r/w bit */
+       kw_write_reg(reg_addr, addr);
+
+       /* Set up the sub address */
+       if ((mode_reg & KW_I2C_MODE_MODE_MASK) == KW_I2C_MODE_STANDARDSUB
+           || (mode_reg & KW_I2C_MODE_MODE_MASK) == KW_I2C_MODE_COMBINED)
+               kw_write_reg(reg_subaddr, subaddr);
+
+       /* Start sending address & disable interrupt*/
+       kw_write_reg(reg_ier, 0 /*KW_I2C_IRQ_MASK*/);
+       kw_write_reg(reg_control, KW_I2C_CTL_XADDR);
+
+       /* State machine, to turn into an interrupt handler */
+       while(state != state_idle) {
+               u8 isr = kw_wait_interrupt(host);
+               state = kw_handle_interrupt(host, state, addr & 1, &rc, &data, &len, isr);
+       }
+
+       return rc;
+}
+
+static void keywest_low_i2c_add(struct device_node *np)
+{
+       struct low_i2c_host     *host = find_low_i2c_host(NULL);
+       u32                     *psteps, *prate, steps, aoffset = 0;
+       struct device_node      *parent;
+
+       if (host == NULL) {
+               printk(KERN_ERR "low_i2c: Can't allocate host for %s\n",
+                      np->full_name);
+               return;
+       }
+       memset(host, 0, sizeof(*host));
+
+       init_MUTEX(&host->mutex);
+       host->np = of_node_get(np);     
+       psteps = (u32 *)get_property(np, "AAPL,address-step", NULL);
+       steps = psteps ? (*psteps) : 0x10;
+       for (host->bsteps = 0; (steps & 0x01) == 0; host->bsteps++)
+               steps >>= 1;
+       parent = of_get_parent(np);
+       host->num_channels = 1;
+       if (parent && parent->name[0] == 'u') {
+               host->num_channels = 2;
+               aoffset = 3;
+       }
+       /* Select interface rate */
+       host->speed = KW_I2C_MODE_100KHZ;
+       prate = (u32 *)get_property(np, "AAPL,i2c-rate", NULL);
+       if (prate) switch(*prate) {
+       case 100:
+               host->speed = KW_I2C_MODE_100KHZ;
+               break;
+       case 50:
+               host->speed = KW_I2C_MODE_50KHZ;
+               break;
+       case 25:
+               host->speed = KW_I2C_MODE_25KHZ;
+               break;
+       }       
+
+       host->mode = pmac_low_i2c_mode_std;
+       host->base = ioremap(np->addrs[0].address + aoffset,
+                                               np->addrs[0].size);
+       host->func = keywest_low_i2c_func;
+}
+
+/*
+ *
+ * PMU implementation
+ *
+ */
+
+
+#ifdef CONFIG_ADB_PMU
+
+static int pmu_low_i2c_func(struct low_i2c_host *host, u8 addr, u8 sub, u8 *data, int len)
+{
+       // TODO
+       return -ENODEV;
+}
+
+static void pmu_low_i2c_add(struct device_node *np)
+{
+       struct low_i2c_host     *host = find_low_i2c_host(NULL);
+
+       if (host == NULL) {
+               printk(KERN_ERR "low_i2c: Can't allocate host for %s\n",
+                      np->full_name);
+               return;
+       }
+       memset(host, 0, sizeof(*host));
+
+       init_MUTEX(&host->mutex);
+       host->np = of_node_get(np);     
+       host->num_channels = 3;
+       host->mode = pmac_low_i2c_mode_std;
+       host->func = pmu_low_i2c_func;
+}
+
+#endif /* CONFIG_ADB_PMU */
+
+void __init pmac_init_low_i2c(void)
+{
+       struct device_node *np;
+
+       /* Probe keywest-i2c busses */
+       np = of_find_compatible_node(NULL, "i2c", "keywest-i2c");
+       while(np) {
+               keywest_low_i2c_add(np);
+               np = of_find_compatible_node(np, "i2c", "keywest-i2c");
+       }
+
+#ifdef CONFIG_ADB_PMU
+       /* Probe PMU busses */
+       np = of_find_node_by_name(NULL, "via-pmu");
+       if (np)
+               pmu_low_i2c_add(np);
+#endif /* CONFIG_ADB_PMU */
+
+       /* TODO: Add CUDA support as well */
+}
+
+int pmac_low_i2c_lock(struct device_node *np)
+{
+       struct low_i2c_host *host = find_low_i2c_host(np);
+
+       if (!host)
+               return -ENODEV;
+       down(&host->mutex);
+       return 0;
+}
+EXPORT_SYMBOL(pmac_low_i2c_lock);
+
+int pmac_low_i2c_unlock(struct device_node *np)
+{
+       struct low_i2c_host *host = find_low_i2c_host(np);
+
+       if (!host)
+               return -ENODEV;
+       up(&host->mutex);
+       return 0;
+}
+EXPORT_SYMBOL(pmac_low_i2c_unlock);
+
+
+int pmac_low_i2c_open(struct device_node *np, int channel)
+{
+       struct low_i2c_host *host = find_low_i2c_host(np);
+
+       if (!host)
+               return -ENODEV;
+
+       if (channel >= host->num_channels)
+               return -EINVAL;
+
+       down(&host->mutex);
+       host->is_open = 1;
+       host->channel = channel;
+
+       return 0;
+}
+EXPORT_SYMBOL(pmac_low_i2c_open);
+
+int pmac_low_i2c_close(struct device_node *np)
+{
+       struct low_i2c_host *host = find_low_i2c_host(np);
+
+       if (!host)
+               return -ENODEV;
+
+       host->is_open = 0;
+       up(&host->mutex);
+
+       return 0;
+}
+EXPORT_SYMBOL(pmac_low_i2c_close);
+
+int pmac_low_i2c_setmode(struct device_node *np, int mode)
+{
+       struct low_i2c_host *host = find_low_i2c_host(np);
+
+       if (!host)
+               return -ENODEV;
+       WARN_ON(!host->is_open);
+       host->mode = mode;
+
+       return 0;
+}
+EXPORT_SYMBOL(pmac_low_i2c_setmode);
+
+int pmac_low_i2c_xfer(struct device_node *np, u8 addrdir, u8 subaddr, u8 *data, int len)
+{
+       struct low_i2c_host *host = find_low_i2c_host(np);
+
+       if (!host)
+               return -ENODEV;
+       WARN_ON(!host->is_open);
+
+       return host->func(host, addrdir, subaddr, data, len);
+}
+EXPORT_SYMBOL(pmac_low_i2c_xfer);
+
diff --git a/arch/powerpc/platforms/powermac/pmac_nvram.c b/arch/powerpc/platforms/powermac/pmac_nvram.c
new file mode 100644 (file)
index 0000000..8c9b008
--- /dev/null
@@ -0,0 +1,584 @@
+/*
+ *  arch/ppc/platforms/pmac_nvram.c
+ *
+ *  Copyright (C) 2002 Benjamin Herrenschmidt (benh@kernel.crashing.org)
+ *
+ *  This program is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU General Public License
+ *  as published by the Free Software Foundation; either version
+ *  2 of the License, or (at your option) any later version.
+ *
+ *  Todo: - add support for the OF persistent properties
+ */
+#include <linux/config.h>
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/stddef.h>
+#include <linux/string.h>
+#include <linux/nvram.h>
+#include <linux/init.h>
+#include <linux/slab.h>
+#include <linux/delay.h>
+#include <linux/errno.h>
+#include <linux/adb.h>
+#include <linux/pmu.h>
+#include <linux/bootmem.h>
+#include <linux/completion.h>
+#include <linux/spinlock.h>
+#include <asm/sections.h>
+#include <asm/io.h>
+#include <asm/system.h>
+#include <asm/prom.h>
+#include <asm/machdep.h>
+#include <asm/nvram.h>
+
+#define DEBUG
+
+#ifdef DEBUG
+#define DBG(x...) printk(x)
+#else
+#define DBG(x...)
+#endif
+
+#define NVRAM_SIZE             0x2000  /* 8kB of non-volatile RAM */
+
+#define CORE99_SIGNATURE       0x5a
+#define CORE99_ADLER_START     0x14
+
+/* On Core99, nvram is either a sharp, a micron or an AMD flash */
+#define SM_FLASH_STATUS_DONE   0x80
+#define SM_FLASH_STATUS_ERR            0x38
+#define SM_FLASH_CMD_ERASE_CONFIRM     0xd0
+#define SM_FLASH_CMD_ERASE_SETUP       0x20
+#define SM_FLASH_CMD_RESET             0xff
+#define SM_FLASH_CMD_WRITE_SETUP       0x40
+#define SM_FLASH_CMD_CLEAR_STATUS      0x50
+#define SM_FLASH_CMD_READ_STATUS       0x70
+
+/* CHRP NVRAM header */
+struct chrp_header {
+  u8           signature;
+  u8           cksum;
+  u16          len;
+  char          name[12];
+  u8           data[0];
+};
+
+struct core99_header {
+  struct chrp_header   hdr;
+  u32                  adler;
+  u32                  generation;
+  u32                  reserved[2];
+};
+
+/*
+ * Read and write the non-volatile RAM on PowerMacs and CHRP machines.
+ */
+static int nvram_naddrs;
+static volatile unsigned char *nvram_addr;
+static volatile unsigned char *nvram_data;
+static int nvram_mult, is_core_99;
+static int core99_bank = 0;
+static int nvram_partitions[3];
+static DEFINE_SPINLOCK(nv_lock);
+
+extern int pmac_newworld;
+extern int system_running;
+
+static int (*core99_write_bank)(int bank, u8* datas);
+static int (*core99_erase_bank)(int bank);
+
+static char *nvram_image;
+
+
+static unsigned char core99_nvram_read_byte(int addr)
+{
+       if (nvram_image == NULL)
+               return 0xff;
+       return nvram_image[addr];
+}
+
+static void core99_nvram_write_byte(int addr, unsigned char val)
+{
+       if (nvram_image == NULL)
+               return;
+       nvram_image[addr] = val;
+}
+
+
+static unsigned char direct_nvram_read_byte(int addr)
+{
+       return in_8(&nvram_data[(addr & (NVRAM_SIZE - 1)) * nvram_mult]);
+}
+
+static void direct_nvram_write_byte(int addr, unsigned char val)
+{
+       out_8(&nvram_data[(addr & (NVRAM_SIZE - 1)) * nvram_mult], val);
+}
+
+
+static unsigned char indirect_nvram_read_byte(int addr)
+{
+       unsigned char val;
+       unsigned long flags;
+
+       spin_lock_irqsave(&nv_lock, flags);
+       out_8(nvram_addr, addr >> 5);
+       val = in_8(&nvram_data[(addr & 0x1f) << 4]);
+       spin_unlock_irqrestore(&nv_lock, flags);
+
+       return val;
+}
+
+static void indirect_nvram_write_byte(int addr, unsigned char val)
+{
+       unsigned long flags;
+
+       spin_lock_irqsave(&nv_lock, flags);
+       out_8(nvram_addr, addr >> 5);
+       out_8(&nvram_data[(addr & 0x1f) << 4], val);
+       spin_unlock_irqrestore(&nv_lock, flags);
+}
+
+
+#ifdef CONFIG_ADB_PMU
+
+static void pmu_nvram_complete(struct adb_request *req)
+{
+       if (req->arg)
+               complete((struct completion *)req->arg);
+}
+
+static unsigned char pmu_nvram_read_byte(int addr)
+{
+       struct adb_request req;
+       DECLARE_COMPLETION(req_complete); 
+       
+       req.arg = system_state == SYSTEM_RUNNING ? &req_complete : NULL;
+       if (pmu_request(&req, pmu_nvram_complete, 3, PMU_READ_NVRAM,
+                       (addr >> 8) & 0xff, addr & 0xff))
+               return 0xff;
+       if (system_state == SYSTEM_RUNNING)
+               wait_for_completion(&req_complete);
+       while (!req.complete)
+               pmu_poll();
+       return req.reply[0];
+}
+
+static void pmu_nvram_write_byte(int addr, unsigned char val)
+{
+       struct adb_request req;
+       DECLARE_COMPLETION(req_complete); 
+       
+       req.arg = system_state == SYSTEM_RUNNING ? &req_complete : NULL;
+       if (pmu_request(&req, pmu_nvram_complete, 4, PMU_WRITE_NVRAM,
+                       (addr >> 8) & 0xff, addr & 0xff, val))
+               return;
+       if (system_state == SYSTEM_RUNNING)
+               wait_for_completion(&req_complete);
+       while (!req.complete)
+               pmu_poll();
+}
+
+#endif /* CONFIG_ADB_PMU */
+
+
+static u8 chrp_checksum(struct chrp_header* hdr)
+{
+       u8 *ptr;
+       u16 sum = hdr->signature;
+       for (ptr = (u8 *)&hdr->len; ptr < hdr->data; ptr++)
+               sum += *ptr;
+       while (sum > 0xFF)
+               sum = (sum & 0xFF) + (sum>>8);
+       return sum;
+}
+
+static u32 core99_calc_adler(u8 *buffer)
+{
+       int cnt;
+       u32 low, high;
+
+       buffer += CORE99_ADLER_START;
+       low = 1;
+       high = 0;
+       for (cnt=0; cnt<(NVRAM_SIZE-CORE99_ADLER_START); cnt++) {
+               if ((cnt % 5000) == 0) {
+                       high  %= 65521UL;
+                       high %= 65521UL;
+               }
+               low += buffer[cnt];
+               high += low;
+       }
+       low  %= 65521UL;
+       high %= 65521UL;
+
+       return (high << 16) | low;
+}
+
+static u32 core99_check(u8* datas)
+{
+       struct core99_header* hdr99 = (struct core99_header*)datas;
+
+       if (hdr99->hdr.signature != CORE99_SIGNATURE) {
+               DBG("Invalid signature\n");
+               return 0;
+       }
+       if (hdr99->hdr.cksum != chrp_checksum(&hdr99->hdr)) {
+               DBG("Invalid checksum\n");
+               return 0;
+       }
+       if (hdr99->adler != core99_calc_adler(datas)) {
+               DBG("Invalid adler\n");
+               return 0;
+       }
+       return hdr99->generation;
+}
+
+static int sm_erase_bank(int bank)
+{
+       int stat, i;
+       unsigned long timeout;
+
+       u8* base = (u8 *)nvram_data + core99_bank*NVRAM_SIZE;
+
+               DBG("nvram: Sharp/Micron Erasing bank %d...\n", bank);
+
+       out_8(base, SM_FLASH_CMD_ERASE_SETUP);
+       out_8(base, SM_FLASH_CMD_ERASE_CONFIRM);
+       timeout = 0;
+       do {
+               if (++timeout > 1000000) {
+                       printk(KERN_ERR "nvram: Sharp/Miron flash erase timeout !\n");
+                       break;
+               }
+               out_8(base, SM_FLASH_CMD_READ_STATUS);
+               stat = in_8(base);
+       } while (!(stat & SM_FLASH_STATUS_DONE));
+
+       out_8(base, SM_FLASH_CMD_CLEAR_STATUS);
+       out_8(base, SM_FLASH_CMD_RESET);
+
+       for (i=0; i<NVRAM_SIZE; i++)
+               if (base[i] != 0xff) {
+                       printk(KERN_ERR "nvram: Sharp/Micron flash erase failed !\n");
+                       return -ENXIO;
+               }
+       return 0;
+}
+
+static int sm_write_bank(int bank, u8* datas)
+{
+       int i, stat = 0;
+       unsigned long timeout;
+
+       u8* base = (u8 *)nvram_data + core99_bank*NVRAM_SIZE;
+
+               DBG("nvram: Sharp/Micron Writing bank %d...\n", bank);
+
+       for (i=0; i<NVRAM_SIZE; i++) {
+               out_8(base+i, SM_FLASH_CMD_WRITE_SETUP);
+               udelay(1);
+               out_8(base+i, datas[i]);
+               timeout = 0;
+               do {
+                       if (++timeout > 1000000) {
+                               printk(KERN_ERR "nvram: Sharp/Micron flash write timeout !\n");
+                               break;
+                       }
+                       out_8(base, SM_FLASH_CMD_READ_STATUS);
+                       stat = in_8(base);
+               } while (!(stat & SM_FLASH_STATUS_DONE));
+               if (!(stat & SM_FLASH_STATUS_DONE))
+                       break;
+       }
+       out_8(base, SM_FLASH_CMD_CLEAR_STATUS);
+       out_8(base, SM_FLASH_CMD_RESET);
+       for (i=0; i<NVRAM_SIZE; i++)
+               if (base[i] != datas[i]) {
+                       printk(KERN_ERR "nvram: Sharp/Micron flash write failed !\n");
+                       return -ENXIO;
+               }
+       return 0;
+}
+
+static int amd_erase_bank(int bank)
+{
+       int i, stat = 0;
+       unsigned long timeout;
+
+       u8* base = (u8 *)nvram_data + core99_bank*NVRAM_SIZE;
+
+               DBG("nvram: AMD Erasing bank %d...\n", bank);
+
+       /* Unlock 1 */
+       out_8(base+0x555, 0xaa);
+       udelay(1);
+       /* Unlock 2 */
+       out_8(base+0x2aa, 0x55);
+       udelay(1);
+
+       /* Sector-Erase */
+       out_8(base+0x555, 0x80);
+       udelay(1);
+       out_8(base+0x555, 0xaa);
+       udelay(1);
+       out_8(base+0x2aa, 0x55);
+       udelay(1);
+       out_8(base, 0x30);
+       udelay(1);
+
+       timeout = 0;
+       do {
+               if (++timeout > 1000000) {
+                       printk(KERN_ERR "nvram: AMD flash erase timeout !\n");
+                       break;
+               }
+               stat = in_8(base) ^ in_8(base);
+       } while (stat != 0);
+       
+       /* Reset */
+       out_8(base, 0xf0);
+       udelay(1);
+       
+       for (i=0; i<NVRAM_SIZE; i++)
+               if (base[i] != 0xff) {
+                       printk(KERN_ERR "nvram: AMD flash erase failed !\n");
+                       return -ENXIO;
+               }
+       return 0;
+}
+
+static int amd_write_bank(int bank, u8* datas)
+{
+       int i, stat = 0;
+       unsigned long timeout;
+
+       u8* base = (u8 *)nvram_data + core99_bank*NVRAM_SIZE;
+
+               DBG("nvram: AMD Writing bank %d...\n", bank);
+
+       for (i=0; i<NVRAM_SIZE; i++) {
+               /* Unlock 1 */
+               out_8(base+0x555, 0xaa);
+               udelay(1);
+               /* Unlock 2 */
+               out_8(base+0x2aa, 0x55);
+               udelay(1);
+
+               /* Write single word */
+               out_8(base+0x555, 0xa0);
+               udelay(1);
+               out_8(base+i, datas[i]);
+               
+               timeout = 0;
+               do {
+                       if (++timeout > 1000000) {
+                               printk(KERN_ERR "nvram: AMD flash write timeout !\n");
+                               break;
+                       }
+                       stat = in_8(base) ^ in_8(base);
+               } while (stat != 0);
+               if (stat != 0)
+                       break;
+       }
+
+       /* Reset */
+       out_8(base, 0xf0);
+       udelay(1);
+
+       for (i=0; i<NVRAM_SIZE; i++)
+               if (base[i] != datas[i]) {
+                       printk(KERN_ERR "nvram: AMD flash write failed !\n");
+                       return -ENXIO;
+               }
+       return 0;
+}
+
+static void __init lookup_partitions(void)
+{
+       u8 buffer[17];
+       int i, offset;
+       struct chrp_header* hdr;
+
+       if (pmac_newworld) {
+               nvram_partitions[pmac_nvram_OF] = -1;
+               nvram_partitions[pmac_nvram_XPRAM] = -1;
+               nvram_partitions[pmac_nvram_NR] = -1;
+               hdr = (struct chrp_header *)buffer;
+
+               offset = 0;
+               buffer[16] = 0;
+               do {
+                       for (i=0;i<16;i++)
+                               buffer[i] = nvram_read_byte(offset+i);
+                       if (!strcmp(hdr->name, "common"))
+                               nvram_partitions[pmac_nvram_OF] = offset + 0x10;
+                       if (!strcmp(hdr->name, "APL,MacOS75")) {
+                               nvram_partitions[pmac_nvram_XPRAM] = offset + 0x10;
+                               nvram_partitions[pmac_nvram_NR] = offset + 0x110;
+                       }
+                       offset += (hdr->len * 0x10);
+               } while(offset < NVRAM_SIZE);
+       } else {
+               nvram_partitions[pmac_nvram_OF] = 0x1800;
+               nvram_partitions[pmac_nvram_XPRAM] = 0x1300;
+               nvram_partitions[pmac_nvram_NR] = 0x1400;
+       }
+       DBG("nvram: OF partition at 0x%x\n", nvram_partitions[pmac_nvram_OF]);
+       DBG("nvram: XP partition at 0x%x\n", nvram_partitions[pmac_nvram_XPRAM]);
+       DBG("nvram: NR partition at 0x%x\n", nvram_partitions[pmac_nvram_NR]);
+}
+
+static void core99_nvram_sync(void)
+{
+       struct core99_header* hdr99;
+       unsigned long flags;
+
+       if (!is_core_99 || !nvram_data || !nvram_image)
+               return;
+
+       spin_lock_irqsave(&nv_lock, flags);
+       if (!memcmp(nvram_image, (u8*)nvram_data + core99_bank*NVRAM_SIZE,
+               NVRAM_SIZE))
+               goto bail;
+
+       DBG("Updating nvram...\n");
+
+       hdr99 = (struct core99_header*)nvram_image;
+       hdr99->generation++;
+       hdr99->hdr.signature = CORE99_SIGNATURE;
+       hdr99->hdr.cksum = chrp_checksum(&hdr99->hdr);
+       hdr99->adler = core99_calc_adler(nvram_image);
+       core99_bank = core99_bank ? 0 : 1;
+       if (core99_erase_bank)
+               if (core99_erase_bank(core99_bank)) {
+                       printk("nvram: Error erasing bank %d\n", core99_bank);
+                       goto bail;
+               }
+       if (core99_write_bank)
+               if (core99_write_bank(core99_bank, nvram_image))
+                       printk("nvram: Error writing bank %d\n", core99_bank);
+ bail:
+       spin_unlock_irqrestore(&nv_lock, flags);
+
+#ifdef DEBUG
+               mdelay(2000);
+#endif
+}
+
+void __init pmac_nvram_init(void)
+{
+       struct device_node *dp;
+
+       nvram_naddrs = 0;
+
+       dp = find_devices("nvram");
+       if (dp == NULL) {
+               printk(KERN_ERR "Can't find NVRAM device\n");
+               return;
+       }
+       nvram_naddrs = dp->n_addrs;
+       is_core_99 = device_is_compatible(dp, "nvram,flash");
+       if (is_core_99) {
+               int i;
+               u32 gen_bank0, gen_bank1;
+
+               if (nvram_naddrs < 1) {
+                       printk(KERN_ERR "nvram: no address\n");
+                       return;
+               }
+               nvram_image = alloc_bootmem(NVRAM_SIZE);
+               if (nvram_image == NULL) {
+                       printk(KERN_ERR "nvram: can't allocate ram image\n");
+                       return;
+               }
+               nvram_data = ioremap(dp->addrs[0].address, NVRAM_SIZE*2);
+               nvram_naddrs = 1; /* Make sure we get the correct case */
+
+               DBG("nvram: Checking bank 0...\n");
+
+               gen_bank0 = core99_check((u8 *)nvram_data);
+               gen_bank1 = core99_check((u8 *)nvram_data + NVRAM_SIZE);
+               core99_bank = (gen_bank0 < gen_bank1) ? 1 : 0;
+
+               DBG("nvram: gen0=%d, gen1=%d\n", gen_bank0, gen_bank1);
+               DBG("nvram: Active bank is: %d\n", core99_bank);
+
+               for (i=0; i<NVRAM_SIZE; i++)
+                       nvram_image[i] = nvram_data[i + core99_bank*NVRAM_SIZE];
+
+               ppc_md.nvram_read_val   = core99_nvram_read_byte;
+               ppc_md.nvram_write_val  = core99_nvram_write_byte;
+               ppc_md.nvram_sync       = core99_nvram_sync;
+               /* 
+                * Maybe we could be smarter here though making an exclusive list
+                * of known flash chips is a bit nasty as older OF didn't provide us
+                * with a useful "compatible" entry. A solution would be to really
+                * identify the chip using flash id commands and base ourselves on
+                * a list of known chips IDs
+                */
+               if (device_is_compatible(dp, "amd-0137")) {
+                       core99_erase_bank = amd_erase_bank;
+                       core99_write_bank = amd_write_bank;
+               } else {
+                       core99_erase_bank = sm_erase_bank;
+                       core99_write_bank = sm_write_bank;
+               }
+       } else if (_machine == _MACH_chrp && nvram_naddrs == 1) {
+               nvram_data = ioremap(dp->addrs[0].address + isa_mem_base,
+                                    dp->addrs[0].size);
+               nvram_mult = 1;
+               ppc_md.nvram_read_val   = direct_nvram_read_byte;
+               ppc_md.nvram_write_val  = direct_nvram_write_byte;
+       } else if (nvram_naddrs == 1) {
+               nvram_data = ioremap(dp->addrs[0].address, dp->addrs[0].size);
+               nvram_mult = (dp->addrs[0].size + NVRAM_SIZE - 1) / NVRAM_SIZE;
+               ppc_md.nvram_read_val   = direct_nvram_read_byte;
+               ppc_md.nvram_write_val  = direct_nvram_write_byte;
+       } else if (nvram_naddrs == 2) {
+               nvram_addr = ioremap(dp->addrs[0].address, dp->addrs[0].size);
+               nvram_data = ioremap(dp->addrs[1].address, dp->addrs[1].size);
+               ppc_md.nvram_read_val   = indirect_nvram_read_byte;
+               ppc_md.nvram_write_val  = indirect_nvram_write_byte;
+       } else if (nvram_naddrs == 0 && sys_ctrler == SYS_CTRLER_PMU) {
+#ifdef CONFIG_ADB_PMU
+               nvram_naddrs = -1;
+               ppc_md.nvram_read_val   = pmu_nvram_read_byte;
+               ppc_md.nvram_write_val  = pmu_nvram_write_byte;
+#endif /* CONFIG_ADB_PMU */
+       } else {
+               printk(KERN_ERR "Don't know how to access NVRAM with %d addresses\n",
+                      nvram_naddrs);
+       }
+       lookup_partitions();
+}
+
+int pmac_get_partition(int partition)
+{
+       return nvram_partitions[partition];
+}
+
+u8 pmac_xpram_read(int xpaddr)
+{
+       int offset = nvram_partitions[pmac_nvram_XPRAM];
+
+       if (offset < 0)
+               return 0xff;
+
+       return ppc_md.nvram_read_val(xpaddr + offset);
+}
+
+void pmac_xpram_write(int xpaddr, u8 data)
+{
+       int offset = nvram_partitions[pmac_nvram_XPRAM];
+
+       if (offset < 0)
+               return;
+
+       ppc_md.nvram_write_val(xpaddr + offset, data);
+}
+
+EXPORT_SYMBOL(pmac_get_partition);
+EXPORT_SYMBOL(pmac_xpram_read);
+EXPORT_SYMBOL(pmac_xpram_write);
diff --git a/arch/powerpc/platforms/powermac/pmac_pci.c b/arch/powerpc/platforms/powermac/pmac_pci.c
new file mode 100644 (file)
index 0000000..40bcd3e
--- /dev/null
@@ -0,0 +1,1341 @@
+/*
+ * Support for PCI bridges found on Power Macintoshes.
+ * At present the "bandit" and "chaos" bridges are supported.
+ * Fortunately you access configuration space in the same
+ * way with either bridge.
+ *
+ * Copyright (C) 2003 Benjamin Herrenschmuidt (benh@kernel.crashing.org)
+ * Copyright (C) 1997 Paul Mackerras (paulus@samba.org)
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version
+ * 2 of the License, or (at your option) any later version.
+ */
+
+#include <linux/kernel.h>
+#include <linux/pci.h>
+#include <linux/delay.h>
+#include <linux/string.h>
+#include <linux/init.h>
+#include <linux/bootmem.h>
+
+#include <asm/sections.h>
+#include <asm/io.h>
+#include <asm/prom.h>
+#include <asm/pci-bridge.h>
+#include <asm/machdep.h>
+#include <asm/pmac_feature.h>
+
+#undef DEBUG
+
+#ifdef DEBUG
+#define DBG(x...) printk(x)
+#else
+#define DBG(x...)
+#endif
+
+static int add_bridge(struct device_node *dev);
+extern void pmac_check_ht_link(void);
+
+/* XXX Could be per-controller, but I don't think we risk anything by
+ * assuming we won't have both UniNorth and Bandit */
+static int has_uninorth;
+#ifdef CONFIG_POWER4
+static struct pci_controller *u3_agp;
+#endif /* CONFIG_POWER4 */
+
+extern u8 pci_cache_line_size;
+extern int pcibios_assign_bus_offset;
+
+struct device_node *k2_skiplist[2];
+
+/*
+ * Magic constants for enabling cache coherency in the bandit/PSX bridge.
+ */
+#define BANDIT_DEVID_2 8
+#define BANDIT_REVID   3
+
+#define BANDIT_DEVNUM  11
+#define BANDIT_MAGIC   0x50
+#define BANDIT_COHERENT        0x40
+
+static int __init fixup_one_level_bus_range(struct device_node *node, int higher)
+{
+       for (; node != 0;node = node->sibling) {
+               int * bus_range;
+               unsigned int *class_code;
+               int len;
+
+               /* For PCI<->PCI bridges or CardBus bridges, we go down */
+               class_code = (unsigned int *) get_property(node, "class-code", NULL);
+               if (!class_code || ((*class_code >> 8) != PCI_CLASS_BRIDGE_PCI &&
+                       (*class_code >> 8) != PCI_CLASS_BRIDGE_CARDBUS))
+                       continue;
+               bus_range = (int *) get_property(node, "bus-range", &len);
+               if (bus_range != NULL && len > 2 * sizeof(int)) {
+                       if (bus_range[1] > higher)
+                               higher = bus_range[1];
+               }
+               higher = fixup_one_level_bus_range(node->child, higher);
+       }
+       return higher;
+}
+
+/* This routine fixes the "bus-range" property of all bridges in the
+ * system since they tend to have their "last" member wrong on macs
+ *
+ * Note that the bus numbers manipulated here are OF bus numbers, they
+ * are not Linux bus numbers.
+ */
+static void __init fixup_bus_range(struct device_node *bridge)
+{
+       int * bus_range;
+       int len;
+
+       /* Lookup the "bus-range" property for the hose */
+       bus_range = (int *) get_property(bridge, "bus-range", &len);
+       if (bus_range == NULL || len < 2 * sizeof(int)) {
+               printk(KERN_WARNING "Can't get bus-range for %s\n",
+                              bridge->full_name);
+               return;
+       }
+       bus_range[1] = fixup_one_level_bus_range(bridge->child, bus_range[1]);
+}
+
+/*
+ * Apple MacRISC (U3, UniNorth, Bandit, Chaos) PCI controllers.
+ *
+ * The "Bandit" version is present in all early PCI PowerMacs,
+ * and up to the first ones using Grackle. Some machines may
+ * have 2 bandit controllers (2 PCI busses).
+ *
+ * "Chaos" is used in some "Bandit"-type machines as a bridge
+ * for the separate display bus. It is accessed the same
+ * way as bandit, but cannot be probed for devices. It therefore
+ * has its own config access functions.
+ *
+ * The "UniNorth" version is present in all Core99 machines
+ * (iBook, G4, new IMacs, and all the recent Apple machines).
+ * It contains 3 controllers in one ASIC.
+ *
+ * The U3 is the bridge used on G5 machines. It contains an
+ * AGP bus which is dealt with the old UniNorth access routines
+ * and a HyperTransport bus which uses its own set of access
+ * functions.
+ */
+
+#define MACRISC_CFA0(devfn, off)       \
+       ((1 << (unsigned long)PCI_SLOT(dev_fn)) \
+       | (((unsigned long)PCI_FUNC(dev_fn)) << 8) \
+       | (((unsigned long)(off)) & 0xFCUL))
+
+#define MACRISC_CFA1(bus, devfn, off)  \
+       ((((unsigned long)(bus)) << 16) \
+       |(((unsigned long)(devfn)) << 8) \
+       |(((unsigned long)(off)) & 0xFCUL) \
+       |1UL)
+
+static unsigned long macrisc_cfg_access(struct pci_controller* hose,
+                                              u8 bus, u8 dev_fn, u8 offset)
+{
+       unsigned int caddr;
+
+       if (bus == hose->first_busno) {
+               if (dev_fn < (11 << 3))
+                       return 0;
+               caddr = MACRISC_CFA0(dev_fn, offset);
+       } else
+               caddr = MACRISC_CFA1(bus, dev_fn, offset);
+
+       /* Uninorth will return garbage if we don't read back the value ! */
+       do {
+               out_le32(hose->cfg_addr, caddr);
+       } while (in_le32(hose->cfg_addr) != caddr);
+
+       offset &= has_uninorth ? 0x07 : 0x03;
+       return ((unsigned long)hose->cfg_data) + offset;
+}
+
+static int macrisc_read_config(struct pci_bus *bus, unsigned int devfn,
+                                     int offset, int len, u32 *val)
+{
+       struct pci_controller *hose = bus->sysdata;
+       unsigned long addr;
+
+       addr = macrisc_cfg_access(hose, bus->number, devfn, offset);
+       if (!addr)
+               return PCIBIOS_DEVICE_NOT_FOUND;
+       /*
+        * Note: the caller has already checked that offset is
+        * suitably aligned and that len is 1, 2 or 4.
+        */
+       switch (len) {
+       case 1:
+               *val = in_8((u8 *)addr);
+               break;
+       case 2:
+               *val = in_le16((u16 *)addr);
+               break;
+       default:
+               *val = in_le32((u32 *)addr);
+               break;
+       }
+       return PCIBIOS_SUCCESSFUL;
+}
+
+static int macrisc_write_config(struct pci_bus *bus, unsigned int devfn,
+                                      int offset, int len, u32 val)
+{
+       struct pci_controller *hose = bus->sysdata;
+       unsigned long addr;
+
+       addr = macrisc_cfg_access(hose, bus->number, devfn, offset);
+       if (!addr)
+               return PCIBIOS_DEVICE_NOT_FOUND;
+       /*
+        * Note: the caller has already checked that offset is
+        * suitably aligned and that len is 1, 2 or 4.
+        */
+       switch (len) {
+       case 1:
+               out_8((u8 *)addr, val);
+               (void) in_8((u8 *)addr);
+               break;
+       case 2:
+               out_le16((u16 *)addr, val);
+               (void) in_le16((u16 *)addr);
+               break;
+       default:
+               out_le32((u32 *)addr, val);
+               (void) in_le32((u32 *)addr);
+               break;
+       }
+       return PCIBIOS_SUCCESSFUL;
+}
+
+static struct pci_ops macrisc_pci_ops =
+{
+       macrisc_read_config,
+       macrisc_write_config
+};
+
+/*
+ * Verifiy that a specific (bus, dev_fn) exists on chaos
+ */
+static int
+chaos_validate_dev(struct pci_bus *bus, int devfn, int offset)
+{
+       struct device_node *np;
+       u32 *vendor, *device;
+
+       np = pci_busdev_to_OF_node(bus, devfn);
+       if (np == NULL)
+               return PCIBIOS_DEVICE_NOT_FOUND;
+
+       vendor = (u32 *)get_property(np, "vendor-id", NULL);
+       device = (u32 *)get_property(np, "device-id", NULL);
+       if (vendor == NULL || device == NULL)
+               return PCIBIOS_DEVICE_NOT_FOUND;
+
+       if ((*vendor == 0x106b) && (*device == 3) && (offset >= 0x10)
+           && (offset != 0x14) && (offset != 0x18) && (offset <= 0x24))
+               return PCIBIOS_BAD_REGISTER_NUMBER;
+
+       return PCIBIOS_SUCCESSFUL;
+}
+
+static int
+chaos_read_config(struct pci_bus *bus, unsigned int devfn, int offset,
+                 int len, u32 *val)
+{
+       int result = chaos_validate_dev(bus, devfn, offset);
+       if (result == PCIBIOS_BAD_REGISTER_NUMBER)
+               *val = ~0U;
+       if (result != PCIBIOS_SUCCESSFUL)
+               return result;
+       return macrisc_read_config(bus, devfn, offset, len, val);
+}
+
+static int
+chaos_write_config(struct pci_bus *bus, unsigned int devfn, int offset,
+                  int len, u32 val)
+{
+       int result = chaos_validate_dev(bus, devfn, offset);
+       if (result != PCIBIOS_SUCCESSFUL)
+               return result;
+       return macrisc_write_config(bus, devfn, offset, len, val);
+}
+
+static struct pci_ops chaos_pci_ops =
+{
+       chaos_read_config,
+       chaos_write_config
+};
+
+#ifdef CONFIG_POWER4
+
+/*
+ * These versions of U3 HyperTransport config space access ops do not
+ * implement self-view of the HT host yet
+ */
+
+/*
+ * This function deals with some "special cases" devices.
+ *
+ *  0 -> No special case
+ *  1 -> Skip the device but act as if the access was successfull
+ *       (return 0xff's on reads, eventually, cache config space
+ *       accesses in a later version)
+ * -1 -> Hide the device (unsuccessful acess)
+ */
+static int u3_ht_skip_device(struct pci_controller *hose,
+                            struct pci_bus *bus, unsigned int devfn)
+{
+       struct device_node *busdn, *dn;
+       int i;
+
+       /* We only allow config cycles to devices that are in OF device-tree
+        * as we are apparently having some weird things going on with some
+        * revs of K2 on recent G5s
+        */
+       if (bus->self)
+               busdn = pci_device_to_OF_node(bus->self);
+       else
+               busdn = hose->arch_data;
+       for (dn = busdn->child; dn; dn = dn->sibling)
+               if (dn->data && PCI_DN(dn)->devfn == devfn)
+                       break;
+       if (dn == NULL)
+               return -1;
+
+       /*
+        * When a device in K2 is powered down, we die on config
+        * cycle accesses. Fix that here.
+        */
+       for (i=0; i<2; i++)
+               if (k2_skiplist[i] == dn)
+                       return 1;
+
+       return 0;
+}
+
+#define U3_HT_CFA0(devfn, off)         \
+               ((((unsigned long)devfn) << 8) | offset)
+#define U3_HT_CFA1(bus, devfn, off)    \
+               (U3_HT_CFA0(devfn, off) \
+               + (((unsigned long)bus) << 16) \
+               + 0x01000000UL)
+
+static unsigned long u3_ht_cfg_access(struct pci_controller* hose,
+                                            u8 bus, u8 devfn, u8 offset)
+{
+       if (bus == hose->first_busno) {
+               /* For now, we don't self probe U3 HT bridge */
+               if (PCI_SLOT(devfn) == 0)
+                       return 0;
+               return ((unsigned long)hose->cfg_data) + U3_HT_CFA0(devfn, offset);
+       } else
+               return ((unsigned long)hose->cfg_data) + U3_HT_CFA1(bus, devfn, offset);
+}
+
+static int u3_ht_read_config(struct pci_bus *bus, unsigned int devfn,
+                                   int offset, int len, u32 *val)
+{
+       struct pci_controller *hose = bus->sysdata;
+       unsigned long addr;
+
+       struct device_node *np = pci_busdev_to_OF_node(bus, devfn);
+       if (np == NULL)
+               return PCIBIOS_DEVICE_NOT_FOUND;
+
+       addr = u3_ht_cfg_access(hose, bus->number, devfn, offset);
+       if (!addr)
+               return PCIBIOS_DEVICE_NOT_FOUND;
+
+       switch (u3_ht_skip_device(hose, bus, devfn)) {
+       case 0:
+               break;
+       case 1:
+                       switch (len) {
+                       case 1:
+                               *val = 0xff; break;
+                       case 2:
+                               *val = 0xffff; break;
+                       default:
+                               *val = 0xfffffffful; break;
+                       }
+                       return PCIBIOS_SUCCESSFUL;
+       default:
+               return PCIBIOS_DEVICE_NOT_FOUND;
+               }
+           
+       /*
+        * Note: the caller has already checked that offset is
+        * suitably aligned and that len is 1, 2 or 4.
+        */
+       switch (len) {
+       case 1:
+               *val = in_8((u8 *)addr);
+               break;
+       case 2:
+               *val = in_le16((u16 *)addr);
+               break;
+       default:
+               *val = in_le32((u32 *)addr);
+               break;
+       }
+       return PCIBIOS_SUCCESSFUL;
+}
+
+static int u3_ht_write_config(struct pci_bus *bus, unsigned int devfn,
+                                    int offset, int len, u32 val)
+{
+       struct pci_controller *hose = bus->sysdata;
+       unsigned long addr;
+
+       struct device_node *np = pci_busdev_to_OF_node(bus, devfn);
+       if (np == NULL)
+               return PCIBIOS_DEVICE_NOT_FOUND;
+
+       addr = u3_ht_cfg_access(hose, bus->number, devfn, offset);
+       if (!addr)
+               return PCIBIOS_DEVICE_NOT_FOUND;
+
+       switch (u3_ht_skip_device(hose, bus, devfn)) {
+       case 0:
+               break;
+       case 1:
+               return PCIBIOS_SUCCESSFUL;
+       default:
+               return PCIBIOS_DEVICE_NOT_FOUND;
+       }
+
+       /*
+        * Note: the caller has already checked that offset is
+        * suitably aligned and that len is 1, 2 or 4.
+        */
+       switch (len) {
+       case 1:
+               out_8((u8 *)addr, val);
+               (void) in_8((u8 *)addr);
+               break;
+       case 2:
+               out_le16((u16 *)addr, val);
+               (void) in_le16((u16 *)addr);
+               break;
+       default:
+               out_le32((u32 *)addr, val);
+               (void) in_le32((u32 *)addr);
+               break;
+       }
+       return PCIBIOS_SUCCESSFUL;
+}
+
+static struct pci_ops u3_ht_pci_ops =
+{
+       u3_ht_read_config,
+       u3_ht_write_config
+};
+
+#endif /* CONFIG_POWER4 */
+
+/*
+ * For a bandit bridge, turn on cache coherency if necessary.
+ * N.B. we could clean this up using the hose ops directly.
+ */
+static void __init
+init_bandit(struct pci_controller *bp)
+{
+       unsigned int vendev, magic;
+       int rev;
+
+       /* read the word at offset 0 in config space for device 11 */
+       out_le32(bp->cfg_addr, (1UL << BANDIT_DEVNUM) + PCI_VENDOR_ID);
+       udelay(2);
+       vendev = in_le32(bp->cfg_data);
+       if (vendev == (PCI_DEVICE_ID_APPLE_BANDIT << 16) +
+                       PCI_VENDOR_ID_APPLE) {
+               /* read the revision id */
+               out_le32(bp->cfg_addr,
+                        (1UL << BANDIT_DEVNUM) + PCI_REVISION_ID);
+               udelay(2);
+               rev = in_8(bp->cfg_data);
+               if (rev != BANDIT_REVID)
+                       printk(KERN_WARNING
+                              "Unknown revision %d for bandit\n", rev);
+       } else if (vendev != (BANDIT_DEVID_2 << 16) + PCI_VENDOR_ID_APPLE) {
+               printk(KERN_WARNING "bandit isn't? (%x)\n", vendev);
+               return;
+       }
+
+       /* read the word at offset 0x50 */
+       out_le32(bp->cfg_addr, (1UL << BANDIT_DEVNUM) + BANDIT_MAGIC);
+       udelay(2);
+       magic = in_le32(bp->cfg_data);
+       if ((magic & BANDIT_COHERENT) != 0)
+               return;
+       magic |= BANDIT_COHERENT;
+       udelay(2);
+       out_le32(bp->cfg_data, magic);
+       printk(KERN_INFO "Cache coherency enabled for bandit/PSX\n");
+}
+
+
+/*
+ * Tweak the PCI-PCI bridge chip on the blue & white G3s.
+ */
+static void __init
+init_p2pbridge(void)
+{
+       struct device_node *p2pbridge;
+       struct pci_controller* hose;
+       u8 bus, devfn;
+       u16 val;
+
+       /* XXX it would be better here to identify the specific
+          PCI-PCI bridge chip we have. */
+       if ((p2pbridge = find_devices("pci-bridge")) == 0
+           || p2pbridge->parent == NULL
+           || strcmp(p2pbridge->parent->name, "pci") != 0)
+               return;
+       if (pci_device_from_OF_node(p2pbridge, &bus, &devfn) < 0) {
+               DBG("Can't find PCI infos for PCI<->PCI bridge\n");
+               return;
+       }
+       /* Warning: At this point, we have not yet renumbered all busses.
+        * So we must use OF walking to find out hose
+        */
+       hose = pci_find_hose_for_OF_device(p2pbridge);
+       if (!hose) {
+               DBG("Can't find hose for PCI<->PCI bridge\n");
+               return;
+       }
+       if (early_read_config_word(hose, bus, devfn,
+                                  PCI_BRIDGE_CONTROL, &val) < 0) {
+               printk(KERN_ERR "init_p2pbridge: couldn't read bridge control\n");
+               return;
+       }
+       val &= ~PCI_BRIDGE_CTL_MASTER_ABORT;
+       early_write_config_word(hose, bus, devfn, PCI_BRIDGE_CONTROL, val);
+}
+
+/*
+ * Some Apple desktop machines have a NEC PD720100A USB2 controller
+ * on the motherboard. Open Firmware, on these, will disable the
+ * EHCI part of it so it behaves like a pair of OHCI's. This fixup
+ * code re-enables it ;)
+ */
+static void __init
+fixup_nec_usb2(void)
+{
+       struct device_node *nec;
+
+       for (nec = NULL; (nec = of_find_node_by_name(nec, "usb")) != NULL;) {
+               struct pci_controller *hose;
+               u32 data, *prop;
+               u8 bus, devfn;
+               
+               prop = (u32 *)get_property(nec, "vendor-id", NULL);
+               if (prop == NULL)
+                       continue;
+               if (0x1033 != *prop)
+                       continue;
+               prop = (u32 *)get_property(nec, "device-id", NULL);
+               if (prop == NULL)
+                       continue;
+               if (0x0035 != *prop)
+                       continue;
+               prop = (u32 *)get_property(nec, "reg", NULL);
+               if (prop == NULL)
+                       continue;
+               devfn = (prop[0] >> 8) & 0xff;
+               bus = (prop[0] >> 16) & 0xff;
+               if (PCI_FUNC(devfn) != 0)
+                       continue;
+               hose = pci_find_hose_for_OF_device(nec);
+               if (!hose)
+                       continue;
+               early_read_config_dword(hose, bus, devfn, 0xe4, &data);
+               if (data & 1UL) {
+                       printk("Found NEC PD720100A USB2 chip with disabled EHCI, fixing up...\n");
+                       data &= ~1UL;
+                       early_write_config_dword(hose, bus, devfn, 0xe4, data);
+                       early_write_config_byte(hose, bus, devfn | 2, PCI_INTERRUPT_LINE,
+                               nec->intrs[0].line);
+               }
+       }
+}
+
+void __init
+pmac_find_bridges(void)
+{
+       struct device_node *np, *root;
+       struct device_node *ht = NULL;
+
+       root = of_find_node_by_path("/");
+       if (root == NULL) {
+               printk(KERN_CRIT "pmac_find_bridges: can't find root of device tree\n");
+               return;
+       }
+       for (np = NULL; (np = of_get_next_child(root, np)) != NULL;) {
+               if (np->name == NULL)
+                       continue;
+               if (strcmp(np->name, "bandit") == 0
+                   || strcmp(np->name, "chaos") == 0
+                   || strcmp(np->name, "pci") == 0) {
+                       if (add_bridge(np) == 0)
+                               of_node_get(np);
+               }
+               if (strcmp(np->name, "ht") == 0) {
+                       of_node_get(np);
+                       ht = np;
+               }
+       }
+       of_node_put(root);
+
+       /* Probe HT last as it relies on the agp resources to be already
+        * setup
+        */
+       if (ht && add_bridge(ht) != 0)
+               of_node_put(ht);
+
+       init_p2pbridge();
+       fixup_nec_usb2();
+       
+       /* We are still having some issues with the Xserve G4, enabling
+        * some offset between bus number and domains for now when we
+        * assign all busses should help for now
+        */
+       if (pci_assign_all_busses)
+               pcibios_assign_bus_offset = 0x10;
+
+#ifdef CONFIG_POWER4 
+       /* There is something wrong with DMA on U3/HT. I haven't figured out
+        * the details yet, but if I set the cache line size to 128 bytes like
+        * it should, I'm getting memory corruption caused by devices like
+        * sungem (even without the MWI bit set, but maybe sungem doesn't
+        * care). Right now, it appears that setting up a 64 bytes line size
+        * works properly, 64 bytes beeing the max transfer size of HT, I
+        * suppose this is related the way HT/PCI are hooked together. I still
+        * need to dive into more specs though to be really sure of what's
+        * going on. --BenH.
+        *
+        * Ok, apparently, it's just that HT can't do more than 64 bytes
+        * transactions. MWI seem to be meaningless there as well, it may
+        * be worth nop'ing out pci_set_mwi too though I haven't done that
+        * yet.
+        *
+        * Note that it's a bit different for whatever is in the AGP slot.
+        * For now, I don't care, but this can become a real issue, we
+        * should probably hook pci_set_mwi anyway to make sure it sets
+        * the real cache line size in there.
+        */
+       if (machine_is_compatible("MacRISC4"))
+               pci_cache_line_size = 16; /* 64 bytes */
+
+       pmac_check_ht_link();
+#endif /* CONFIG_POWER4 */
+}
+
+#define GRACKLE_CFA(b, d, o)   (0x80 | ((b) << 8) | ((d) << 16) \
+                                | (((o) & ~3) << 24))
+
+#define GRACKLE_PICR1_STG              0x00000040
+#define GRACKLE_PICR1_LOOPSNOOP                0x00000010
+
+/* N.B. this is called before bridges is initialized, so we can't
+   use grackle_pcibios_{read,write}_config_dword. */
+static inline void grackle_set_stg(struct pci_controller* bp, int enable)
+{
+       unsigned int val;
+
+       out_be32(bp->cfg_addr, GRACKLE_CFA(0, 0, 0xa8));
+       val = in_le32(bp->cfg_data);
+       val = enable? (val | GRACKLE_PICR1_STG) :
+               (val & ~GRACKLE_PICR1_STG);
+       out_be32(bp->cfg_addr, GRACKLE_CFA(0, 0, 0xa8));
+       out_le32(bp->cfg_data, val);
+       (void)in_le32(bp->cfg_data);
+}
+
+static inline void grackle_set_loop_snoop(struct pci_controller *bp, int enable)
+{
+       unsigned int val;
+
+       out_be32(bp->cfg_addr, GRACKLE_CFA(0, 0, 0xa8));
+       val = in_le32(bp->cfg_data);
+       val = enable? (val | GRACKLE_PICR1_LOOPSNOOP) :
+               (val & ~GRACKLE_PICR1_LOOPSNOOP);
+       out_be32(bp->cfg_addr, GRACKLE_CFA(0, 0, 0xa8));
+       out_le32(bp->cfg_data, val);
+       (void)in_le32(bp->cfg_data);
+}
+
+static int __init
+setup_uninorth(struct pci_controller* hose, struct reg_property* addr)
+{
+       pci_assign_all_busses = 1;
+       has_uninorth = 1;
+       hose->ops = &macrisc_pci_ops;
+       hose->cfg_addr = ioremap(addr->address + 0x800000, 0x1000);
+       hose->cfg_data = ioremap(addr->address + 0xc00000, 0x1000);
+       /* We "know" that the bridge at f2000000 has the PCI slots. */
+       return addr->address == 0xf2000000;
+}
+
+static void __init
+setup_bandit(struct pci_controller* hose, struct reg_property* addr)
+{
+       hose->ops = &macrisc_pci_ops;
+       hose->cfg_addr = ioremap(addr->address + 0x800000, 0x1000);
+       hose->cfg_data = ioremap(addr->address + 0xc00000, 0x1000);
+       init_bandit(hose);
+}
+
+static void __init
+setup_chaos(struct pci_controller* hose, struct reg_property* addr)
+{
+       /* assume a `chaos' bridge */
+       hose->ops = &chaos_pci_ops;
+       hose->cfg_addr = ioremap(addr->address + 0x800000, 0x1000);
+       hose->cfg_data = ioremap(addr->address + 0xc00000, 0x1000);
+}
+
+#ifdef CONFIG_POWER4
+
+static void __init setup_u3_agp(struct pci_controller* hose)
+{
+       /* On G5, we move AGP up to high bus number so we don't need
+        * to reassign bus numbers for HT. If we ever have P2P bridges
+        * on AGP, we'll have to move pci_assign_all_busses to the
+        * pci_controller structure so we enable it for AGP and not for
+        * HT childs.
+        * We hard code the address because of the different size of
+        * the reg address cell, we shall fix that by killing struct
+        * reg_property and using some accessor functions instead
+        */
+               hose->first_busno = 0xf0;
+       hose->last_busno = 0xff;
+       has_uninorth = 1;
+       hose->ops = &macrisc_pci_ops;
+       hose->cfg_addr = ioremap(0xf0000000 + 0x800000, 0x1000);
+       hose->cfg_data = ioremap(0xf0000000 + 0xc00000, 0x1000);
+
+       u3_agp = hose;
+}
+
+static void __init setup_u3_ht(struct pci_controller* hose)
+{
+       struct device_node *np = (struct device_node *)hose->arch_data;
+       int i, cur;
+
+       hose->ops = &u3_ht_pci_ops;
+
+       /* We hard code the address because of the different size of
+        * the reg address cell, we shall fix that by killing struct
+        * reg_property and using some accessor functions instead
+        */
+       hose->cfg_data = (volatile unsigned char *)ioremap(0xf2000000, 0x02000000);
+
+       /*
+        * /ht node doesn't expose a "ranges" property, so we "remove" regions that
+        * have been allocated to AGP. So far, this version of the code doesn't assign
+        * any of the 0xfxxxxxxx "fine" memory regions to /ht.
+        * We need to fix that sooner or later by either parsing all child "ranges"
+        * properties or figuring out the U3 address space decoding logic and
+        * then read its configuration register (if any).
+        */
+       hose->io_base_phys = 0xf4000000;
+       hose->io_base_virt = ioremap(hose->io_base_phys, 0x00400000);
+       isa_io_base = (unsigned long) hose->io_base_virt;
+       hose->io_resource.name = np->full_name;
+       hose->io_resource.start = 0;
+       hose->io_resource.end = 0x003fffff;
+       hose->io_resource.flags = IORESOURCE_IO;
+       hose->pci_mem_offset = 0;
+       hose->first_busno = 0;
+       hose->last_busno = 0xef;
+       hose->mem_resources[0].name = np->full_name;
+       hose->mem_resources[0].start = 0x80000000;
+       hose->mem_resources[0].end = 0xefffffff;
+       hose->mem_resources[0].flags = IORESOURCE_MEM;
+
+       if (u3_agp == NULL) {
+               DBG("U3 has no AGP, using full resource range\n");
+               return;
+       }
+
+       /* We "remove" the AGP resources from the resources allocated to HT, that
+        * is we create "holes". However, that code does assumptions that so far
+        * happen to be true (cross fingers...), typically that resources in the
+        * AGP node are properly ordered
+        */
+       cur = 0;
+       for (i=0; i<3; i++) {
+               struct resource *res = &u3_agp->mem_resources[i];
+               if (res->flags != IORESOURCE_MEM)
+                       continue;
+               /* We don't care about "fine" resources */
+               if (res->start >= 0xf0000000)
+                       continue;
+               /* Check if it's just a matter of "shrinking" us in one direction */
+               if (hose->mem_resources[cur].start == res->start) {
+                       DBG("U3/HT: shrink start of %d, %08lx -> %08lx\n",
+                           cur, hose->mem_resources[cur].start, res->end + 1);
+                       hose->mem_resources[cur].start = res->end + 1;
+                       continue;
+               }
+               if (hose->mem_resources[cur].end == res->end) {
+                       DBG("U3/HT: shrink end of %d, %08lx -> %08lx\n",
+                           cur, hose->mem_resources[cur].end, res->start - 1);
+                       hose->mem_resources[cur].end = res->start - 1;
+                       continue;
+               }
+               /* No, it's not the case, we need a hole */
+               if (cur == 2) {
+                       /* not enough resources to make a hole, we drop part of the range */
+                       printk(KERN_WARNING "Running out of resources for /ht host !\n");
+                       hose->mem_resources[cur].end = res->start - 1;
+                       continue;
+               }               
+               cur++;
+                       DBG("U3/HT: hole, %d end at %08lx, %d start at %08lx\n",
+                   cur-1, res->start - 1, cur, res->end + 1);
+               hose->mem_resources[cur].name = np->full_name;
+               hose->mem_resources[cur].flags = IORESOURCE_MEM;
+               hose->mem_resources[cur].start = res->end + 1;
+               hose->mem_resources[cur].end = hose->mem_resources[cur-1].end;
+               hose->mem_resources[cur-1].end = res->start - 1;
+       }
+}
+
+#endif /* CONFIG_POWER4 */
+
+void __init
+setup_grackle(struct pci_controller *hose)
+{
+       setup_indirect_pci(hose, 0xfec00000, 0xfee00000);
+       if (machine_is_compatible("AAPL,PowerBook1998"))
+               grackle_set_loop_snoop(hose, 1);
+#if 0  /* Disabled for now, HW problems ??? */
+       grackle_set_stg(hose, 1);
+#endif
+}
+
+static void __init pmac_process_bridge_OF_ranges(struct pci_controller *hose,
+                          struct device_node *dev, int primary)
+{
+       static unsigned int static_lc_ranges[2024];
+       unsigned int *dt_ranges, *lc_ranges, *ranges, *prev;
+       unsigned int size;
+       int rlen = 0, orig_rlen;
+       int memno = 0;
+       struct resource *res;
+       int np, na = prom_n_addr_cells(dev);
+
+       np = na + 5;
+
+       /* First we try to merge ranges to fix a problem with some pmacs
+        * that can have more than 3 ranges, fortunately using contiguous
+        * addresses -- BenH
+        */
+       dt_ranges = (unsigned int *) get_property(dev, "ranges", &rlen);
+       if (!dt_ranges)
+               return;
+       /*      lc_ranges = alloc_bootmem(rlen);*/
+       lc_ranges = static_lc_ranges;
+       if (!lc_ranges)
+               return; /* what can we do here ? */
+       memcpy(lc_ranges, dt_ranges, rlen);
+       orig_rlen = rlen;
+
+       /* Let's work on a copy of the "ranges" property instead of damaging
+        * the device-tree image in memory
+        */
+       ranges = lc_ranges;
+       prev = NULL;
+       while ((rlen -= np * sizeof(unsigned int)) >= 0) {
+               if (prev) {
+                       if (prev[0] == ranges[0] && prev[1] == ranges[1] &&
+                               (prev[2] + prev[na+4]) == ranges[2] &&
+                               (prev[na+2] + prev[na+4]) == ranges[na+2]) {
+                               prev[na+4] += ranges[na+4];
+                               ranges[0] = 0;
+                               ranges += np;
+                               continue;
+                       }
+               }
+               prev = ranges;
+               ranges += np;
+       }
+
+       /*
+        * The ranges property is laid out as an array of elements,
+        * each of which comprises:
+        *   cells 0 - 2:       a PCI address
+        *   cells 3 or 3+4:    a CPU physical address
+        *                      (size depending on dev->n_addr_cells)
+        *   cells 4+5 or 5+6:  the size of the range
+        */
+       ranges = lc_ranges;
+       rlen = orig_rlen;
+       while (ranges && (rlen -= np * sizeof(unsigned int)) >= 0) {
+               res = NULL;
+               size = ranges[na+4];
+               switch (ranges[0] >> 24) {
+               case 1:         /* I/O space */
+                       if (ranges[2] != 0)
+                               break;
+                       hose->io_base_phys = ranges[na+2];
+                       /* limit I/O space to 16MB */
+                       if (size > 0x01000000)
+                               size = 0x01000000;
+                       hose->io_base_virt = ioremap(ranges[na+2], size);
+                       if (primary)
+                               isa_io_base = (unsigned long) hose->io_base_virt;
+                       res = &hose->io_resource;
+                       res->flags = IORESOURCE_IO;
+                       res->start = ranges[2];
+                       break;
+               case 2:         /* memory space */
+                       memno = 0;
+                       if (ranges[1] == 0 && ranges[2] == 0
+                           && ranges[na+4] <= (16 << 20)) {
+                               /* 1st 16MB, i.e. ISA memory area */
+#if 0
+                               if (primary)
+                                       isa_mem_base = ranges[na+2];
+#endif
+                               memno = 1;
+                       }
+                       while (memno < 3 && hose->mem_resources[memno].flags)
+                               ++memno;
+                       if (memno == 0)
+                               hose->pci_mem_offset = ranges[na+2] - ranges[2];
+                       if (memno < 3) {
+                               res = &hose->mem_resources[memno];
+                               res->flags = IORESOURCE_MEM;
+                               res->start = ranges[na+2];
+                       }
+                       break;
+               }
+               if (res != NULL) {
+                       res->name = dev->full_name;
+                       res->end = res->start + size - 1;
+                       res->parent = NULL;
+                       res->sibling = NULL;
+                       res->child = NULL;
+               }
+               ranges += np;
+       }
+}
+
+/*
+ * We assume that if we have a G3 powermac, we have one bridge called
+ * "pci" (a MPC106) and no bandit or chaos bridges, and contrariwise,
+ * if we have one or more bandit or chaos bridges, we don't have a MPC106.
+ */
+static int __init add_bridge(struct device_node *dev)
+{
+       int len;
+       struct pci_controller *hose;
+       struct reg_property *addr;
+       char* disp_name;
+       int *bus_range;
+       int primary = 1;
+
+       DBG("Adding PCI host bridge %s\n", dev->full_name);
+
+               addr = (struct reg_property *) get_property(dev, "reg", &len);
+               if (addr == NULL || len < sizeof(*addr)) {
+                       printk(KERN_WARNING "Can't use %s: no address\n",
+                              dev->full_name);
+                       return -ENODEV;
+               }
+               bus_range = (int *) get_property(dev, "bus-range", &len);
+               if (bus_range == NULL || len < 2 * sizeof(int)) {
+                       printk(KERN_WARNING "Can't get bus-range for %s, assume bus 0\n",
+                                      dev->full_name);
+               }
+
+               hose = pcibios_alloc_controller();
+               if (!hose)
+                       return -ENOMEM;
+               hose->arch_data = dev;
+               hose->first_busno = bus_range ? bus_range[0] : 0;
+               hose->last_busno = bus_range ? bus_range[1] : 0xff;
+
+       disp_name = NULL;
+#ifdef CONFIG_POWER4
+               if (device_is_compatible(dev, "u3-agp")) {
+                       setup_u3_agp(hose, addr);
+                       disp_name = "U3-AGP";
+                       primary = 0;
+               } else if (device_is_compatible(dev, "u3-ht")) {
+                       setup_u3_ht(hose, addr);
+                       disp_name = "U3-HT";
+                       primary = 1;
+               } else
+#endif /* CONFIG_POWER4 */
+       if (device_is_compatible(dev, "uni-north")) {
+                       primary = setup_uninorth(hose, addr);
+                       disp_name = "UniNorth";
+               } else if (strcmp(dev->name, "pci") == 0) {
+                       /* XXX assume this is a mpc106 (grackle) */
+                       setup_grackle(hose);
+                       disp_name = "Grackle (MPC106)";
+               } else if (strcmp(dev->name, "bandit") == 0) {
+                       setup_bandit(hose, addr);
+                       disp_name = "Bandit";
+               } else if (strcmp(dev->name, "chaos") == 0) {
+                       setup_chaos(hose, addr);
+                       disp_name = "Chaos";
+                       primary = 0;
+               }
+               printk(KERN_INFO "Found %s PCI host bridge at 0x%08x. Firmware bus number: %d->%d\n",
+                       disp_name, addr->address, hose->first_busno, hose->last_busno);
+               DBG(" ->Hose at 0x%p, cfg_addr=0x%p,cfg_data=0x%p\n",
+                       hose, hose->cfg_addr, hose->cfg_data);
+
+               /* Interpret the "ranges" property */
+               /* This also maps the I/O region and sets isa_io/mem_base */
+               pci_process_bridge_OF_ranges(hose, dev, primary);
+
+               /* Fixup "bus-range" OF property */
+               fixup_bus_range(dev);
+
+       return 0;
+}
+
+static void __init
+pcibios_fixup_OF_interrupts(void)
+{
+       struct pci_dev* dev = NULL;
+
+       /*
+        * Open Firmware often doesn't initialize the
+        * PCI_INTERRUPT_LINE config register properly, so we
+        * should find the device node and apply the interrupt
+        * obtained from the OF device-tree
+        */
+       for_each_pci_dev(dev) {
+               struct device_node *node;
+               node = pci_device_to_OF_node(dev);
+               /* this is the node, see if it has interrupts */
+               if (node && node->n_intrs > 0)
+                       dev->irq = node->intrs[0].line;
+               pci_write_config_byte(dev, PCI_INTERRUPT_LINE, dev->irq);
+       }
+}
+
+void __init
+pmac_pcibios_fixup(void)
+{
+       /* Fixup interrupts according to OF tree */
+       pcibios_fixup_OF_interrupts();
+}
+
+int
+pmac_pci_enable_device_hook(struct pci_dev *dev, int initial)
+{
+       struct device_node* node;
+       int updatecfg = 0;
+       int uninorth_child;
+
+       node = pci_device_to_OF_node(dev);
+
+       /* We don't want to enable USB controllers absent from the OF tree
+        * (iBook second controller)
+        */
+       if (dev->vendor == PCI_VENDOR_ID_APPLE
+           && (dev->class == ((PCI_CLASS_SERIAL_USB << 8) | 0x10))
+           && !node) {
+               printk(KERN_INFO "Apple USB OHCI %s disabled by firmware\n",
+                      pci_name(dev));
+               return -EINVAL;
+       }
+
+       if (!node)
+               return 0;
+
+       uninorth_child = node->parent &&
+               device_is_compatible(node->parent, "uni-north");
+       
+       /* Firewire & GMAC were disabled after PCI probe, the driver is
+        * claiming them, we must re-enable them now.
+        */
+       if (uninorth_child && !strcmp(node->name, "firewire") &&
+           (device_is_compatible(node, "pci106b,18") ||
+            device_is_compatible(node, "pci106b,30") ||
+            device_is_compatible(node, "pci11c1,5811"))) {
+               pmac_call_feature(PMAC_FTR_1394_CABLE_POWER, node, 0, 1);
+               pmac_call_feature(PMAC_FTR_1394_ENABLE, node, 0, 1);
+               updatecfg = 1;
+       }
+       if (uninorth_child && !strcmp(node->name, "ethernet") &&
+           device_is_compatible(node, "gmac")) {
+               pmac_call_feature(PMAC_FTR_GMAC_ENABLE, node, 0, 1);
+               updatecfg = 1;
+       }
+
+       if (updatecfg) {
+               u16 cmd;
+       
+               /*
+                * Make sure PCI is correctly configured
+                *
+                * We use old pci_bios versions of the function since, by
+                * default, gmac is not powered up, and so will be absent
+                * from the kernel initial PCI lookup.
+                *
+                * Should be replaced by 2.4 new PCI mechanisms and really
+                * register the device.
+                */
+               pci_read_config_word(dev, PCI_COMMAND, &cmd);
+               cmd |= PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER | PCI_COMMAND_INVALIDATE;
+               pci_write_config_word(dev, PCI_COMMAND, cmd);
+               pci_write_config_byte(dev, PCI_LATENCY_TIMER, 16);
+               pci_write_config_byte(dev, PCI_CACHE_LINE_SIZE, pci_cache_line_size);
+       }
+
+       return 0;
+}
+
+/* We power down some devices after they have been probed. They'll
+ * be powered back on later on
+ */
+void __init
+pmac_pcibios_after_init(void)
+{
+       struct device_node* nd;
+
+#ifdef CONFIG_BLK_DEV_IDE
+       struct pci_dev *dev = NULL;
+
+       /* OF fails to initialize IDE controllers on macs
+        * (and maybe other machines)
+        *
+        * Ideally, this should be moved to the IDE layer, but we need
+        * to check specifically with Andre Hedrick how to do it cleanly
+        * since the common IDE code seem to care about the fact that the
+        * BIOS may have disabled a controller.
+        *
+        * -- BenH
+        */
+       for_each_pci_dev(dev) {
+               if ((dev->class >> 16) == PCI_BASE_CLASS_STORAGE)
+                       pci_enable_device(dev);
+       }
+#endif /* CONFIG_BLK_DEV_IDE */
+
+       nd = find_devices("firewire");
+       while (nd) {
+               if (nd->parent && (device_is_compatible(nd, "pci106b,18") ||
+                                  device_is_compatible(nd, "pci106b,30") ||
+                                  device_is_compatible(nd, "pci11c1,5811"))
+                   && device_is_compatible(nd->parent, "uni-north")) {
+                       pmac_call_feature(PMAC_FTR_1394_ENABLE, nd, 0, 0);
+                       pmac_call_feature(PMAC_FTR_1394_CABLE_POWER, nd, 0, 0);
+               }
+               nd = nd->next;
+       }
+       nd = find_devices("ethernet");
+       while (nd) {
+               if (nd->parent && device_is_compatible(nd, "gmac")
+                   && device_is_compatible(nd->parent, "uni-north"))
+                       pmac_call_feature(PMAC_FTR_GMAC_ENABLE, nd, 0, 0);
+               nd = nd->next;
+       }
+}
+
+#ifdef CONFIG_PPC64
+static void __init pmac_fixup_phb_resources(void)
+{
+       struct pci_controller *hose, *tmp;
+       
+       list_for_each_entry_safe(hose, tmp, &hose_list, list_node) {
+               unsigned long offset = (unsigned long)hose->io_base_virt - pci_io_base;
+               hose->io_resource.start += offset;
+               hose->io_resource.end += offset;
+               printk(KERN_INFO "PCI Host %d, io start: %lx; io end: %lx\n",
+                      hose->global_number,
+                      hose->io_resource.start, hose->io_resource.end);
+       }
+}
+
+void __init pmac_pci_init(void)
+{
+       struct device_node *np, *root;
+       struct device_node *ht = NULL;
+
+       /* Probe root PCI hosts, that is on U3 the AGP host and the
+        * HyperTransport host. That one is actually "kept" around
+        * and actually added last as it's resource management relies
+        * on the AGP resources to have been setup first
+        */
+       root = of_find_node_by_path("/");
+       if (root == NULL) {
+               printk(KERN_CRIT "pmac_find_bridges: can't find root of device tree\n");
+               return;
+       }
+       for (np = NULL; (np = of_get_next_child(root, np)) != NULL;) {
+               if (np->name == NULL)
+                       continue;
+               if (strcmp(np->name, "pci") == 0) {
+                       if (add_bridge(np) == 0)
+                               of_node_get(np);
+               }
+               if (strcmp(np->name, "ht") == 0) {
+                       of_node_get(np);
+                       ht = np;
+               }
+       }
+       of_node_put(root);
+
+       /* Now setup the HyperTransport host if we found any
+        */
+       if (ht && add_bridge(ht) != 0)
+               of_node_put(ht);
+
+       /* Fixup the IO resources on our host bridges as the common code
+        * does it only for childs of the host bridges
+        */
+       pmac_fixup_phb_resources();
+
+       /* Setup the linkage between OF nodes and PHBs */ 
+       pci_devs_phb_init();
+
+       /* Fixup the PCI<->OF mapping for U3 AGP due to bus renumbering. We
+        * assume there is no P2P bridge on the AGP bus, which should be a
+        * safe assumptions hopefully.
+        */
+       if (u3_agp) {
+               struct device_node *np = u3_agp->arch_data;
+               PCI_DN(np)->busno = 0xf0;
+               for (np = np->child; np; np = np->sibling)
+                       PCI_DN(np)->busno = 0xf0;
+       }
+
+       pmac_check_ht_link();
+
+       /* Tell pci.c to not use the common resource allocation mecanism */
+       pci_probe_only = 1;
+       
+       /* Allow all IO */
+       io_page_mask = -1;
+}
+#endif
+
+#ifdef CONFIG_PPC32
+void pmac_pci_fixup_cardbus(struct pci_dev* dev)
+{
+       if (_machine != _MACH_Pmac)
+               return;
+       /*
+        * Fix the interrupt routing on the various cardbus bridges
+        * used on powerbooks
+        */
+       if (dev->vendor != PCI_VENDOR_ID_TI)
+               return;
+       if (dev->device == PCI_DEVICE_ID_TI_1130 ||
+           dev->device == PCI_DEVICE_ID_TI_1131) {
+               u8 val;
+               /* Enable PCI interrupt */
+               if (pci_read_config_byte(dev, 0x91, &val) == 0)
+                       pci_write_config_byte(dev, 0x91, val | 0x30);
+               /* Disable ISA interrupt mode */
+               if (pci_read_config_byte(dev, 0x92, &val) == 0)
+                       pci_write_config_byte(dev, 0x92, val & ~0x06);
+       }
+       if (dev->device == PCI_DEVICE_ID_TI_1210 ||
+           dev->device == PCI_DEVICE_ID_TI_1211 ||
+           dev->device == PCI_DEVICE_ID_TI_1410 ||
+           dev->device == PCI_DEVICE_ID_TI_1510) {
+               u8 val;
+               /* 0x8c == TI122X_IRQMUX, 2 says to route the INTA
+                  signal out the MFUNC0 pin */
+               if (pci_read_config_byte(dev, 0x8c, &val) == 0)
+                       pci_write_config_byte(dev, 0x8c, (val & ~0x0f) | 2);
+               /* Disable ISA interrupt mode */
+               if (pci_read_config_byte(dev, 0x92, &val) == 0)
+                       pci_write_config_byte(dev, 0x92, val & ~0x06);
+       }
+}
+
+DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_TI, PCI_ANY_ID, pmac_pci_fixup_cardbus);
+
+void pmac_pci_fixup_pciata(struct pci_dev* dev)
+{
+       u8 progif = 0;
+
+       /*
+        * On PowerMacs, we try to switch any PCI ATA controller to
+       * fully native mode
+        */
+       if (_machine != _MACH_Pmac)
+               return;
+       /* Some controllers don't have the class IDE */
+       if (dev->vendor == PCI_VENDOR_ID_PROMISE)
+               switch(dev->device) {
+               case PCI_DEVICE_ID_PROMISE_20246:
+               case PCI_DEVICE_ID_PROMISE_20262:
+               case PCI_DEVICE_ID_PROMISE_20263:
+               case PCI_DEVICE_ID_PROMISE_20265:
+               case PCI_DEVICE_ID_PROMISE_20267:
+               case PCI_DEVICE_ID_PROMISE_20268:
+               case PCI_DEVICE_ID_PROMISE_20269:
+               case PCI_DEVICE_ID_PROMISE_20270:
+               case PCI_DEVICE_ID_PROMISE_20271:
+               case PCI_DEVICE_ID_PROMISE_20275:
+               case PCI_DEVICE_ID_PROMISE_20276:
+               case PCI_DEVICE_ID_PROMISE_20277:
+                       goto good;
+               }
+       /* Others, check PCI class */
+       if ((dev->class >> 8) != PCI_CLASS_STORAGE_IDE)
+               return;
+ good:
+       pci_read_config_byte(dev, PCI_CLASS_PROG, &progif);
+       if ((progif & 5) != 5) {
+               printk(KERN_INFO "Forcing PCI IDE into native mode: %s\n", pci_name(dev));
+               (void) pci_write_config_byte(dev, PCI_CLASS_PROG, progif|5);
+               if (pci_read_config_byte(dev, PCI_CLASS_PROG, &progif) ||
+                   (progif & 5) != 5)
+                       printk(KERN_ERR "Rewrite of PROGIF failed !\n");
+       }
+}
+DECLARE_PCI_FIXUP_FINAL(PCI_ANY_ID, PCI_ANY_ID, pmac_pci_fixup_pciata);
+#endif
+
+/*
+ * Disable second function on K2-SATA, it's broken
+ * and disable IO BARs on first one
+ */
+static void fixup_k2_sata(struct pci_dev* dev)
+{
+       int i;
+       u16 cmd;
+
+       if (PCI_FUNC(dev->devfn) > 0) {
+               pci_read_config_word(dev, PCI_COMMAND, &cmd);
+               cmd &= ~(PCI_COMMAND_IO | PCI_COMMAND_MEMORY);
+               pci_write_config_word(dev, PCI_COMMAND, cmd);
+               for (i = 0; i < 6; i++) {
+                       dev->resource[i].start = dev->resource[i].end = 0;
+                       dev->resource[i].flags = 0;
+                       pci_write_config_dword(dev, PCI_BASE_ADDRESS_0 + 4 * i, 0);
+               }
+       } else {
+               pci_read_config_word(dev, PCI_COMMAND, &cmd);
+               cmd &= ~PCI_COMMAND_IO;
+               pci_write_config_word(dev, PCI_COMMAND, cmd);
+               for (i = 0; i < 5; i++) {
+                       dev->resource[i].start = dev->resource[i].end = 0;
+                       dev->resource[i].flags = 0;
+                       pci_write_config_dword(dev, PCI_BASE_ADDRESS_0 + 4 * i, 0);
+               }
+       }
+}
+DECLARE_PCI_FIXUP_HEADER(PCI_VENDOR_ID_SERVERWORKS, 0x0240, fixup_k2_sata);
+
diff --git a/arch/powerpc/platforms/powermac/pmac_pic.c b/arch/powerpc/platforms/powermac/pmac_pic.c
new file mode 100644 (file)
index 0000000..bf3e189
--- /dev/null
@@ -0,0 +1,655 @@
+/*
+ *  Support for the interrupt controllers found on Power Macintosh,
+ *  currently Apple's "Grand Central" interrupt controller in all
+ *  it's incarnations. OpenPIC support used on newer machines is
+ *  in a separate file
+ *
+ *  Copyright (C) 1997 Paul Mackerras (paulus@samba.org)
+ *
+ *  Maintained by Benjamin Herrenschmidt (benh@kernel.crashing.org)
+ *
+ *  This program is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU General Public License
+ *  as published by the Free Software Foundation; either version
+ *  2 of the License, or (at your option) any later version.
+ *
+ */
+
+#include <linux/config.h>
+#include <linux/stddef.h>
+#include <linux/init.h>
+#include <linux/sched.h>
+#include <linux/signal.h>
+#include <linux/pci.h>
+#include <linux/interrupt.h>
+#include <linux/sysdev.h>
+#include <linux/adb.h>
+#include <linux/pmu.h>
+
+#include <asm/sections.h>
+#include <asm/io.h>
+#include <asm/smp.h>
+#include <asm/prom.h>
+#include <asm/pci-bridge.h>
+#include <asm/time.h>
+#include <asm/open_pic.h>
+#include <asm/xmon.h>
+#include <asm/pmac_feature.h>
+#include <asm/mpic.h>
+
+#include "pmac_pic.h"
+
+/*
+ * XXX this should be in xmon.h, but putting it there means xmon.h
+ * has to include <linux/interrupt.h> (to get irqreturn_t), which
+ * causes all sorts of problems.  -- paulus
+ */
+extern irqreturn_t xmon_irq(int, void *, struct pt_regs *);
+
+struct pmac_irq_hw {
+        unsigned int    event;
+        unsigned int    enable;
+        unsigned int    ack;
+        unsigned int    level;
+};
+
+/* Default addresses */
+static volatile struct pmac_irq_hw *pmac_irq_hw[4] = {
+        (struct pmac_irq_hw *) 0xf3000020,
+        (struct pmac_irq_hw *) 0xf3000010,
+        (struct pmac_irq_hw *) 0xf4000020,
+        (struct pmac_irq_hw *) 0xf4000010,
+};
+
+#define GC_LEVEL_MASK          0x3ff00000
+#define OHARE_LEVEL_MASK       0x1ff00000
+#define HEATHROW_LEVEL_MASK    0x1ff00000
+
+static int max_irqs;
+static int max_real_irqs;
+static u32 level_mask[4];
+
+static DEFINE_SPINLOCK(pmac_pic_lock);
+
+
+#define GATWICK_IRQ_POOL_SIZE        10
+static struct interrupt_info gatwick_int_pool[GATWICK_IRQ_POOL_SIZE];
+
+/*
+ * Mark an irq as "lost".  This is only used on the pmac
+ * since it can lose interrupts (see pmac_set_irq_mask).
+ * -- Cort
+ */
+void
+__set_lost(unsigned long irq_nr, int nokick)
+{
+       if (!test_and_set_bit(irq_nr, ppc_lost_interrupts)) {
+               atomic_inc(&ppc_n_lost_interrupts);
+               if (!nokick)
+                       set_dec(1);
+       }
+}
+
+static void
+pmac_mask_and_ack_irq(unsigned int irq_nr)
+{
+        unsigned long bit = 1UL << (irq_nr & 0x1f);
+        int i = irq_nr >> 5;
+        unsigned long flags;
+
+        if ((unsigned)irq_nr >= max_irqs)
+                return;
+
+        clear_bit(irq_nr, ppc_cached_irq_mask);
+        if (test_and_clear_bit(irq_nr, ppc_lost_interrupts))
+                atomic_dec(&ppc_n_lost_interrupts);
+       spin_lock_irqsave(&pmac_pic_lock, flags);
+        out_le32(&pmac_irq_hw[i]->enable, ppc_cached_irq_mask[i]);
+        out_le32(&pmac_irq_hw[i]->ack, bit);
+        do {
+                /* make sure ack gets to controller before we enable
+                   interrupts */
+                mb();
+        } while((in_le32(&pmac_irq_hw[i]->enable) & bit)
+                != (ppc_cached_irq_mask[i] & bit));
+       spin_unlock_irqrestore(&pmac_pic_lock, flags);
+}
+
+static void pmac_set_irq_mask(unsigned int irq_nr, int nokicklost)
+{
+        unsigned long bit = 1UL << (irq_nr & 0x1f);
+        int i = irq_nr >> 5;
+        unsigned long flags;
+
+        if ((unsigned)irq_nr >= max_irqs)
+                return;
+
+       spin_lock_irqsave(&pmac_pic_lock, flags);
+        /* enable unmasked interrupts */
+        out_le32(&pmac_irq_hw[i]->enable, ppc_cached_irq_mask[i]);
+
+        do {
+                /* make sure mask gets to controller before we
+                   return to user */
+                mb();
+        } while((in_le32(&pmac_irq_hw[i]->enable) & bit)
+                != (ppc_cached_irq_mask[i] & bit));
+
+        /*
+         * Unfortunately, setting the bit in the enable register
+         * when the device interrupt is already on *doesn't* set
+         * the bit in the flag register or request another interrupt.
+         */
+        if (bit & ppc_cached_irq_mask[i] & in_le32(&pmac_irq_hw[i]->level))
+               __set_lost((ulong)irq_nr, nokicklost);
+       spin_unlock_irqrestore(&pmac_pic_lock, flags);
+}
+
+/* When an irq gets requested for the first client, if it's an
+ * edge interrupt, we clear any previous one on the controller
+ */
+static unsigned int pmac_startup_irq(unsigned int irq_nr)
+{
+        unsigned long bit = 1UL << (irq_nr & 0x1f);
+        int i = irq_nr >> 5;
+
+       if ((irq_desc[irq_nr].status & IRQ_LEVEL) == 0)
+               out_le32(&pmac_irq_hw[i]->ack, bit);
+        set_bit(irq_nr, ppc_cached_irq_mask);
+        pmac_set_irq_mask(irq_nr, 0);
+
+       return 0;
+}
+
+static void pmac_mask_irq(unsigned int irq_nr)
+{
+        clear_bit(irq_nr, ppc_cached_irq_mask);
+        pmac_set_irq_mask(irq_nr, 0);
+        mb();
+}
+
+static void pmac_unmask_irq(unsigned int irq_nr)
+{
+        set_bit(irq_nr, ppc_cached_irq_mask);
+        pmac_set_irq_mask(irq_nr, 0);
+}
+
+static void pmac_end_irq(unsigned int irq_nr)
+{
+       if (!(irq_desc[irq_nr].status & (IRQ_DISABLED|IRQ_INPROGRESS))
+           && irq_desc[irq_nr].action) {
+               set_bit(irq_nr, ppc_cached_irq_mask);
+               pmac_set_irq_mask(irq_nr, 1);
+       }
+}
+
+
+struct hw_interrupt_type pmac_pic = {
+       .typename       = " PMAC-PIC ",
+       .startup        = pmac_startup_irq,
+       .enable         = pmac_unmask_irq,
+       .disable        = pmac_mask_irq,
+       .ack            = pmac_mask_and_ack_irq,
+       .end            = pmac_end_irq,
+};
+
+struct hw_interrupt_type gatwick_pic = {
+       .typename       = " GATWICK  ",
+       .startup        = pmac_startup_irq,
+       .enable         = pmac_unmask_irq,
+       .disable        = pmac_mask_irq,
+       .ack            = pmac_mask_and_ack_irq,
+       .end            = pmac_end_irq,
+};
+
+static irqreturn_t gatwick_action(int cpl, void *dev_id, struct pt_regs *regs)
+{
+       int irq, bits;
+
+       for (irq = max_irqs; (irq -= 32) >= max_real_irqs; ) {
+               int i = irq >> 5;
+               bits = in_le32(&pmac_irq_hw[i]->event) | ppc_lost_interrupts[i];
+               /* We must read level interrupts from the level register */
+               bits |= (in_le32(&pmac_irq_hw[i]->level) & level_mask[i]);
+               bits &= ppc_cached_irq_mask[i];
+               if (bits == 0)
+                       continue;
+               irq += __ilog2(bits);
+               __do_IRQ(irq, regs);
+               return IRQ_HANDLED;
+       }
+       printk("gatwick irq not from gatwick pic\n");
+       return IRQ_NONE;
+}
+
+int
+pmac_get_irq(struct pt_regs *regs)
+{
+       int irq;
+       unsigned long bits = 0;
+
+#ifdef CONFIG_SMP
+       void psurge_smp_message_recv(struct pt_regs *);
+
+               /* IPI's are a hack on the powersurge -- Cort */
+               if ( smp_processor_id() != 0 ) {
+               psurge_smp_message_recv(regs);
+               return -2;      /* ignore, already handled */
+        }
+#endif /* CONFIG_SMP */
+       for (irq = max_real_irqs; (irq -= 32) >= 0; ) {
+               int i = irq >> 5;
+               bits = in_le32(&pmac_irq_hw[i]->event) | ppc_lost_interrupts[i];
+               /* We must read level interrupts from the level register */
+               bits |= (in_le32(&pmac_irq_hw[i]->level) & level_mask[i]);
+               bits &= ppc_cached_irq_mask[i];
+               if (bits == 0)
+                       continue;
+               irq += __ilog2(bits);
+               break;
+       }
+
+       return irq;
+}
+
+/* This routine will fix some missing interrupt values in the device tree
+ * on the gatwick mac-io controller used by some PowerBooks
+ */
+static void __init
+pmac_fix_gatwick_interrupts(struct device_node *gw, int irq_base)
+{
+       struct device_node *node;
+       int count;
+
+       memset(gatwick_int_pool, 0, sizeof(gatwick_int_pool));
+       node = gw->child;
+       count = 0;
+       while(node)
+       {
+               /* Fix SCC */
+               if (strcasecmp(node->name, "escc") == 0)
+                       if (node->child) {
+                               if (node->child->n_intrs < 3) {
+                                       node->child->intrs = &gatwick_int_pool[count];
+                                       count += 3;
+                               }
+                               node->child->n_intrs = 3;
+                               node->child->intrs[0].line = 15+irq_base;
+                               node->child->intrs[1].line =  4+irq_base;
+                               node->child->intrs[2].line =  5+irq_base;
+                               printk(KERN_INFO "irq: fixed SCC on second controller (%d,%d,%d)\n",
+                                       node->child->intrs[0].line,
+                                       node->child->intrs[1].line,
+                                       node->child->intrs[2].line);
+                       }
+               /* Fix media-bay & left SWIM */
+               if (strcasecmp(node->name, "media-bay") == 0) {
+                       struct device_node* ya_node;
+
+                       if (node->n_intrs == 0)
+                               node->intrs = &gatwick_int_pool[count++];
+                       node->n_intrs = 1;
+                       node->intrs[0].line = 29+irq_base;
+                       printk(KERN_INFO "irq: fixed media-bay on second controller (%d)\n",
+                                       node->intrs[0].line);
+
+                       ya_node = node->child;
+                       while(ya_node)
+                       {
+                               if (strcasecmp(ya_node->name, "floppy") == 0) {
+                                       if (ya_node->n_intrs < 2) {
+                                               ya_node->intrs = &gatwick_int_pool[count];
+                                               count += 2;
+                                       }
+                                       ya_node->n_intrs = 2;
+                                       ya_node->intrs[0].line = 19+irq_base;
+                                       ya_node->intrs[1].line =  1+irq_base;
+                                       printk(KERN_INFO "irq: fixed floppy on second controller (%d,%d)\n",
+                                               ya_node->intrs[0].line, ya_node->intrs[1].line);
+                               }
+                               if (strcasecmp(ya_node->name, "ata4") == 0) {
+                                       if (ya_node->n_intrs < 2) {
+                                               ya_node->intrs = &gatwick_int_pool[count];
+                                               count += 2;
+                                       }
+                                       ya_node->n_intrs = 2;
+                                       ya_node->intrs[0].line = 14+irq_base;
+                                       ya_node->intrs[1].line =  3+irq_base;
+                                       printk(KERN_INFO "irq: fixed ide on second controller (%d,%d)\n",
+                                               ya_node->intrs[0].line, ya_node->intrs[1].line);
+                               }
+                               ya_node = ya_node->sibling;
+                       }
+               }
+               node = node->sibling;
+       }
+       if (count > 10) {
+               printk("WARNING !! Gatwick interrupt pool overflow\n");
+               printk("  GATWICK_IRQ_POOL_SIZE = %d\n", GATWICK_IRQ_POOL_SIZE);
+               printk("              requested = %d\n", count);
+       }
+}
+
+/*
+ * The PowerBook 3400/2400/3500 can have a combo ethernet/modem
+ * card which includes an ohare chip that acts as a second interrupt
+ * controller.  If we find this second ohare, set it up and fix the
+ * interrupt value in the device tree for the ethernet chip.
+ */
+static int __init enable_second_ohare(void)
+{
+       unsigned char bus, devfn;
+       unsigned short cmd;
+        unsigned long addr;
+       struct device_node *irqctrler = find_devices("pci106b,7");
+       struct device_node *ether;
+
+       if (irqctrler == NULL || irqctrler->n_addrs <= 0)
+               return -1;
+       addr = (unsigned long) ioremap(irqctrler->addrs[0].address, 0x40);
+       pmac_irq_hw[1] = (volatile struct pmac_irq_hw *)(addr + 0x20);
+       max_irqs = 64;
+       if (pci_device_from_OF_node(irqctrler, &bus, &devfn) == 0) {
+               struct pci_controller* hose = pci_find_hose_for_OF_device(irqctrler);
+               if (!hose)
+                   printk(KERN_ERR "Can't find PCI hose for OHare2 !\n");
+               else {
+                   early_read_config_word(hose, bus, devfn, PCI_COMMAND, &cmd);
+                   cmd |= PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER;
+                   cmd &= ~PCI_COMMAND_IO;
+                   early_write_config_word(hose, bus, devfn, PCI_COMMAND, cmd);
+               }
+       }
+
+       /* Fix interrupt for the modem/ethernet combo controller. The number
+          in the device tree (27) is bogus (correct for the ethernet-only
+          board but not the combo ethernet/modem board).
+          The real interrupt is 28 on the second controller -> 28+32 = 60.
+       */
+       ether = find_devices("pci1011,14");
+       if (ether && ether->n_intrs > 0) {
+               ether->intrs[0].line = 60;
+               printk(KERN_INFO "irq: Fixed ethernet IRQ to %d\n",
+                      ether->intrs[0].line);
+       }
+
+       /* Return the interrupt number of the cascade */
+       return irqctrler->intrs[0].line;
+}
+
+static int pmac_u3_cascade(struct pt_regs *regs, void *data)
+{
+       return mpic_get_one_irq((struct mpic *)data, regs);
+}
+
+#ifdef CONFIG_XMON
+static struct irqaction xmon_action = {
+       .handler        = xmon_irq,
+       .flags          = 0,
+       .mask           = CPU_MASK_NONE,
+       .name           = "NMI - XMON"
+};
+#endif
+
+static struct irqaction gatwick_cascade_action = {
+       .handler        = gatwick_action,
+       .flags          = SA_INTERRUPT,
+       .mask           = CPU_MASK_NONE,
+       .name           = "cascade",
+};
+
+void __init pmac_pic_init(void)
+{
+        int i;
+        struct device_node *irqctrler  = NULL;
+        struct device_node *irqctrler2 = NULL;
+       struct device_node *np;
+        unsigned long addr;
+       int irq_cascade = -1;
+       struct mpic *mpic1, *mpic2;
+
+       /* We first try to detect Apple's new Core99 chipset, since mac-io
+        * is quite different on those machines and contains an IBM MPIC2.
+        */
+       np = find_type_devices("open-pic");
+       while (np) {
+               if (np->parent && !strcmp(np->parent->name, "u3"))
+                       irqctrler2 = np;
+               else
+                       irqctrler = np;
+               np = np->next;
+       }
+       if (irqctrler != NULL && irqctrler->n_addrs > 0) {
+               unsigned char senses[128];
+
+               printk(KERN_INFO "PowerMac using OpenPIC irq controller at 0x%08x\n",
+                      (unsigned int)irqctrler->addrs[0].address);
+
+               prom_get_irq_senses(senses, 0, 128);
+               mpic1 = mpic_alloc(irqctrler->addrs[0].address,
+                                  MPIC_PRIMARY | MPIC_WANTS_RESET,
+                                  0, 0, 128, 256, senses, 128, " K2-MPIC  ");
+               BUG_ON(mpic1 == NULL);
+               mpic_init(mpic1);               
+
+               if (irqctrler2 != NULL && irqctrler2->n_intrs > 0 &&
+                   irqctrler2->n_addrs > 0) {
+                       printk(KERN_INFO "Slave OpenPIC at 0x%08x hooked on IRQ %d\n",
+                              (u32)irqctrler2->addrs[0].address,
+                              irqctrler2->intrs[0].line);
+
+                       pmac_call_feature(PMAC_FTR_ENABLE_MPIC, irqctrler2, 0, 0);
+                       prom_get_irq_senses(senses, 128, 128 + 128);
+
+                       /* We don't need to set MPIC_BROKEN_U3 here since we don't have
+                        * hypertransport interrupts routed to it
+                        */
+                       mpic2 = mpic_alloc(irqctrler2->addrs[0].address,
+                                          MPIC_BIG_ENDIAN | MPIC_WANTS_RESET,
+                                          0, 128, 128, 0, senses, 128, " U3-MPIC  ");
+                       BUG_ON(mpic2 == NULL);
+                       mpic_init(mpic2);
+                       mpic_setup_cascade(irqctrler2->intrs[0].line,
+                                          pmac_u3_cascade, mpic2);
+               }
+       }
+
+       /* Get the level/edge settings, assume if it's not
+        * a Grand Central nor an OHare, then it's an Heathrow
+        * (or Paddington).
+        */
+       if (find_devices("gc"))
+               level_mask[0] = GC_LEVEL_MASK;
+       else if (find_devices("ohare")) {
+               level_mask[0] = OHARE_LEVEL_MASK;
+               /* We might have a second cascaded ohare */
+               level_mask[1] = OHARE_LEVEL_MASK;
+       } else {
+               level_mask[0] = HEATHROW_LEVEL_MASK;
+               level_mask[1] = 0;
+               /* We might have a second cascaded heathrow */
+               level_mask[2] = HEATHROW_LEVEL_MASK;
+               level_mask[3] = 0;
+       }
+
+       /*
+        * G3 powermacs and 1999 G3 PowerBooks have 64 interrupts,
+        * 1998 G3 Series PowerBooks have 128,
+        * other powermacs have 32.
+        * The combo ethernet/modem card for the Powerstar powerbooks
+        * (2400/3400/3500, ohare based) has a second ohare chip
+        * effectively making a total of 64.
+        */
+       max_irqs = max_real_irqs = 32;
+       irqctrler = find_devices("mac-io");
+       if (irqctrler)
+       {
+               max_real_irqs = 64;
+               if (irqctrler->next)
+                       max_irqs = 128;
+               else
+                       max_irqs = 64;
+       }
+       for ( i = 0; i < max_real_irqs ; i++ )
+               irq_desc[i].handler = &pmac_pic;
+
+       /* get addresses of first controller */
+       if (irqctrler) {
+               if  (irqctrler->n_addrs > 0) {
+                       addr = (unsigned long)
+                               ioremap(irqctrler->addrs[0].address, 0x40);
+                       for (i = 0; i < 2; ++i)
+                               pmac_irq_hw[i] = (volatile struct pmac_irq_hw*)
+                                       (addr + (2 - i) * 0x10);
+               }
+
+               /* get addresses of second controller */
+               irqctrler = irqctrler->next;
+               if (irqctrler && irqctrler->n_addrs > 0) {
+                       addr = (unsigned long)
+                               ioremap(irqctrler->addrs[0].address, 0x40);
+                       for (i = 2; i < 4; ++i)
+                               pmac_irq_hw[i] = (volatile struct pmac_irq_hw*)
+                                       (addr + (4 - i) * 0x10);
+                       irq_cascade = irqctrler->intrs[0].line;
+                       if (device_is_compatible(irqctrler, "gatwick"))
+                               pmac_fix_gatwick_interrupts(irqctrler, max_real_irqs);
+               }
+       } else {
+               /* older powermacs have a GC (grand central) or ohare at
+                  f3000000, with interrupt control registers at f3000020. */
+               addr = (unsigned long) ioremap(0xf3000000, 0x40);
+               pmac_irq_hw[0] = (volatile struct pmac_irq_hw *) (addr + 0x20);
+       }
+
+       /* PowerBooks 3400 and 3500 can have a second controller in a second
+          ohare chip, on the combo ethernet/modem card */
+       if (machine_is_compatible("AAPL,3400/2400")
+            || machine_is_compatible("AAPL,3500"))
+               irq_cascade = enable_second_ohare();
+
+       /* disable all interrupts in all controllers */
+       for (i = 0; i * 32 < max_irqs; ++i)
+               out_le32(&pmac_irq_hw[i]->enable, 0);
+       /* mark level interrupts */
+       for (i = 0; i < max_irqs; i++)
+               if (level_mask[i >> 5] & (1UL << (i & 0x1f)))
+                       irq_desc[i].status = IRQ_LEVEL;
+
+       /* get interrupt line of secondary interrupt controller */
+       if (irq_cascade >= 0) {
+               printk(KERN_INFO "irq: secondary controller on irq %d\n",
+                       (int)irq_cascade);
+               for ( i = max_real_irqs ; i < max_irqs ; i++ )
+                       irq_desc[i].handler = &gatwick_pic;
+               setup_irq(irq_cascade, &gatwick_cascade_action);
+       }
+       printk("System has %d possible interrupts\n", max_irqs);
+       if (max_irqs != max_real_irqs)
+               printk(KERN_DEBUG "%d interrupts on main controller\n",
+                       max_real_irqs);
+
+#ifdef CONFIG_XMON
+       setup_irq(20, &xmon_action);
+#endif /* CONFIG_XMON */
+}
+
+#ifdef CONFIG_PM
+/*
+ * These procedures are used in implementing sleep on the powerbooks.
+ * sleep_save_intrs() saves the states of all interrupt enables
+ * and disables all interrupts except for the nominated one.
+ * sleep_restore_intrs() restores the states of all interrupt enables.
+ */
+unsigned long sleep_save_mask[2];
+
+/* This used to be passed by the PMU driver but that link got
+ * broken with the new driver model. We use this tweak for now...
+ */
+static int pmacpic_find_viaint(void)
+{
+       int viaint = -1;
+
+#ifdef CONFIG_ADB_PMU
+       struct device_node *np;
+
+       if (pmu_get_model() != PMU_OHARE_BASED)
+               goto not_found;
+       np = of_find_node_by_name(NULL, "via-pmu");
+       if (np == NULL)
+               goto not_found;
+       viaint = np->intrs[0].line;
+#endif /* CONFIG_ADB_PMU */
+
+not_found:
+       return viaint;
+}
+
+static int pmacpic_suspend(struct sys_device *sysdev, pm_message_t state)
+{
+       int viaint = pmacpic_find_viaint();
+
+       sleep_save_mask[0] = ppc_cached_irq_mask[0];
+       sleep_save_mask[1] = ppc_cached_irq_mask[1];
+       ppc_cached_irq_mask[0] = 0;
+       ppc_cached_irq_mask[1] = 0;
+       if (viaint > 0)
+               set_bit(viaint, ppc_cached_irq_mask);
+       out_le32(&pmac_irq_hw[0]->enable, ppc_cached_irq_mask[0]);
+       if (max_real_irqs > 32)
+               out_le32(&pmac_irq_hw[1]->enable, ppc_cached_irq_mask[1]);
+       (void)in_le32(&pmac_irq_hw[0]->event);
+       /* make sure mask gets to controller before we return to caller */
+       mb();
+        (void)in_le32(&pmac_irq_hw[0]->enable);
+
+        return 0;
+}
+
+static int pmacpic_resume(struct sys_device *sysdev)
+{
+       int i;
+
+       out_le32(&pmac_irq_hw[0]->enable, 0);
+       if (max_real_irqs > 32)
+               out_le32(&pmac_irq_hw[1]->enable, 0);
+       mb();
+       for (i = 0; i < max_real_irqs; ++i)
+               if (test_bit(i, sleep_save_mask))
+                       pmac_unmask_irq(i);
+
+       return 0;
+}
+
+#endif /* CONFIG_PM */
+
+static struct sysdev_class pmacpic_sysclass = {
+       set_kset_name("pmac_pic"),
+};
+
+static struct sys_device device_pmacpic = {
+       .id             = 0,
+       .cls            = &pmacpic_sysclass,
+};
+
+static struct sysdev_driver driver_pmacpic = {
+#ifdef CONFIG_PM
+       .suspend        = &pmacpic_suspend,
+       .resume         = &pmacpic_resume,
+#endif /* CONFIG_PM */
+};
+
+static int __init init_pmacpic_sysfs(void)
+{
+       if (max_irqs == 0)
+               return -ENODEV;
+
+       printk(KERN_DEBUG "Registering pmac pic with sysfs...\n");
+       sysdev_class_register(&pmacpic_sysclass);
+       sysdev_register(&device_pmacpic);
+       sysdev_driver_register(&pmacpic_sysclass, &driver_pmacpic);
+       return 0;
+}
+
+subsys_initcall(init_pmacpic_sysfs);
+
diff --git a/arch/powerpc/platforms/powermac/pmac_pic.h b/arch/powerpc/platforms/powermac/pmac_pic.h
new file mode 100644 (file)
index 0000000..664103d
--- /dev/null
@@ -0,0 +1,11 @@
+#ifndef __PPC_PLATFORMS_PMAC_PIC_H
+#define __PPC_PLATFORMS_PMAC_PIC_H
+
+#include <linux/irq.h>
+
+extern struct hw_interrupt_type pmac_pic;
+
+void pmac_pic_init(void);
+int pmac_get_irq(struct pt_regs *regs);
+
+#endif /* __PPC_PLATFORMS_PMAC_PIC_H */
diff --git a/arch/powerpc/platforms/powermac/pmac_setup.c b/arch/powerpc/platforms/powermac/pmac_setup.c
new file mode 100644 (file)
index 0000000..dbc921a
--- /dev/null
@@ -0,0 +1,662 @@
+/*
+ *  arch/ppc/platforms/setup.c
+ *
+ *  PowerPC version
+ *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
+ *
+ *  Adapted for Power Macintosh by Paul Mackerras
+ *    Copyright (C) 1996 Paul Mackerras (paulus@cs.anu.edu.au)
+ *
+ *  Derived from "arch/alpha/kernel/setup.c"
+ *    Copyright (C) 1995 Linus Torvalds
+ *
+ *  Maintained by Benjamin Herrenschmidt (benh@kernel.crashing.org)
+ *
+ *  This program is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU General Public License
+ *  as published by the Free Software Foundation; either version
+ *  2 of the License, or (at your option) any later version.
+ *
+ */
+
+/*
+ * bootup setup stuff..
+ */
+
+#include <linux/config.h>
+#include <linux/init.h>
+#include <linux/errno.h>
+#include <linux/sched.h>
+#include <linux/kernel.h>
+#include <linux/mm.h>
+#include <linux/stddef.h>
+#include <linux/unistd.h>
+#include <linux/ptrace.h>
+#include <linux/slab.h>
+#include <linux/user.h>
+#include <linux/a.out.h>
+#include <linux/tty.h>
+#include <linux/string.h>
+#include <linux/delay.h>
+#include <linux/ioport.h>
+#include <linux/major.h>
+#include <linux/initrd.h>
+#include <linux/vt_kern.h>
+#include <linux/console.h>
+#include <linux/ide.h>
+#include <linux/pci.h>
+#include <linux/adb.h>
+#include <linux/cuda.h>
+#include <linux/pmu.h>
+#include <linux/irq.h>
+#include <linux/seq_file.h>
+#include <linux/root_dev.h>
+#include <linux/bitops.h>
+#include <linux/suspend.h>
+
+#include <asm/reg.h>
+#include <asm/sections.h>
+#include <asm/prom.h>
+#include <asm/system.h>
+#include <asm/pgtable.h>
+#include <asm/io.h>
+#include <asm/pci-bridge.h>
+#include <asm/ohare.h>
+#include <asm/mediabay.h>
+#include <asm/machdep.h>
+#include <asm/dma.h>
+#include <asm/bootx.h>
+#include <asm/cputable.h>
+#include <asm/btext.h>
+#include <asm/pmac_feature.h>
+#include <asm/time.h>
+#include <asm/of_device.h>
+#include <asm/mmu_context.h>
+
+#include "pmac_pic.h"
+
+#undef SHOW_GATWICK_IRQS
+
+extern long pmac_time_init(void);
+extern unsigned long pmac_get_rtc_time(void);
+extern int pmac_set_rtc_time(unsigned long nowtime);
+extern void pmac_read_rtc_time(void);
+extern void pmac_calibrate_decr(void);
+extern void pmac_pcibios_fixup(void);
+extern void pmac_find_bridges(void);
+extern unsigned long pmac_ide_get_base(int index);
+extern void pmac_ide_init_hwif_ports(hw_regs_t *hw,
+       unsigned long data_port, unsigned long ctrl_port, int *irq);
+
+extern void pmac_nvram_update(void);
+extern unsigned char pmac_nvram_read_byte(int addr);
+extern void pmac_nvram_write_byte(int addr, unsigned char val);
+extern int pmac_pci_enable_device_hook(struct pci_dev *dev, int initial);
+extern void pmac_pcibios_after_init(void);
+extern int of_show_percpuinfo(struct seq_file *m, int i);
+
+unsigned char drive_info;
+
+int ppc_override_l2cr = 0;
+int ppc_override_l2cr_value;
+int has_l2cache = 0;
+
+static int current_root_goodness = -1;
+
+extern int pmac_newworld;
+
+#define DEFAULT_ROOT_DEVICE Root_SDA1  /* sda1 - slightly silly choice */
+
+extern void zs_kgdb_hook(int tty_num);
+static void ohare_init(void);
+#ifdef CONFIG_BOOTX_TEXT
+static void pmac_progress(char *s, unsigned short hex);
+#endif
+
+sys_ctrler_t sys_ctrler = SYS_CTRLER_UNKNOWN;
+
+#ifdef CONFIG_SMP
+extern struct smp_ops_t psurge_smp_ops;
+extern struct smp_ops_t core99_smp_ops;
+#endif /* CONFIG_SMP */
+
+static int
+pmac_show_cpuinfo(struct seq_file *m)
+{
+       struct device_node *np;
+       char *pp;
+       int plen;
+       int mbmodel = pmac_call_feature(PMAC_FTR_GET_MB_INFO,
+               NULL, PMAC_MB_INFO_MODEL, 0);
+       unsigned int mbflags = (unsigned int)pmac_call_feature(PMAC_FTR_GET_MB_INFO,
+               NULL, PMAC_MB_INFO_FLAGS, 0);
+       char* mbname;
+
+       if (pmac_call_feature(PMAC_FTR_GET_MB_INFO, NULL, PMAC_MB_INFO_NAME, (int)&mbname) != 0)
+               mbname = "Unknown";
+
+       /* find motherboard type */
+       seq_printf(m, "machine\t\t: ");
+       np = find_devices("device-tree");
+       if (np != NULL) {
+               pp = (char *) get_property(np, "model", NULL);
+               if (pp != NULL)
+                       seq_printf(m, "%s\n", pp);
+               else
+                       seq_printf(m, "PowerMac\n");
+               pp = (char *) get_property(np, "compatible", &plen);
+               if (pp != NULL) {
+                       seq_printf(m, "motherboard\t:");
+                       while (plen > 0) {
+                               int l = strlen(pp) + 1;
+                               seq_printf(m, " %s", pp);
+                               plen -= l;
+                               pp += l;
+                       }
+                       seq_printf(m, "\n");
+               }
+       } else
+               seq_printf(m, "PowerMac\n");
+
+       /* print parsed model */
+       seq_printf(m, "detected as\t: %d (%s)\n", mbmodel, mbname);
+       seq_printf(m, "pmac flags\t: %08x\n", mbflags);
+
+       /* find l2 cache info */
+       np = find_devices("l2-cache");
+       if (np == 0)
+               np = find_type_devices("cache");
+       if (np != 0) {
+               unsigned int *ic = (unsigned int *)
+                       get_property(np, "i-cache-size", NULL);
+               unsigned int *dc = (unsigned int *)
+                       get_property(np, "d-cache-size", NULL);
+               seq_printf(m, "L2 cache\t:");
+               has_l2cache = 1;
+               if (get_property(np, "cache-unified", NULL) != 0 && dc) {
+                       seq_printf(m, " %dK unified", *dc / 1024);
+               } else {
+                       if (ic)
+                               seq_printf(m, " %dK instruction", *ic / 1024);
+                       if (dc)
+                               seq_printf(m, "%s %dK data",
+                                          (ic? " +": ""), *dc / 1024);
+               }
+               pp = get_property(np, "ram-type", NULL);
+               if (pp)
+                       seq_printf(m, " %s", pp);
+               seq_printf(m, "\n");
+       }
+
+       /* find ram info */
+       np = find_devices("memory");
+       if (np != 0) {
+               int n;
+               struct reg_property *reg = (struct reg_property *)
+                       get_property(np, "reg", &n);
+
+               if (reg != 0) {
+                       unsigned long total = 0;
+
+                       for (n /= sizeof(struct reg_property); n > 0; --n)
+                               total += (reg++)->size;
+                       seq_printf(m, "memory\t\t: %luMB\n", total >> 20);
+               }
+       }
+
+       /* Checks "l2cr-value" property in the registry */
+       np = find_devices("cpus");
+       if (np == 0)
+               np = find_type_devices("cpu");
+       if (np != 0) {
+               unsigned int *l2cr = (unsigned int *)
+                       get_property(np, "l2cr-value", NULL);
+               if (l2cr != 0) {
+                       seq_printf(m, "l2cr override\t: 0x%x\n", *l2cr);
+               }
+       }
+
+       /* Indicate newworld/oldworld */
+       seq_printf(m, "pmac-generation\t: %s\n",
+                  pmac_newworld ? "NewWorld" : "OldWorld");
+
+
+       return 0;
+}
+
+static int
+pmac_show_percpuinfo(struct seq_file *m, int i)
+{
+#ifdef CONFIG_CPU_FREQ_PMAC
+       extern unsigned int pmac_get_one_cpufreq(int i);
+       unsigned int freq = pmac_get_one_cpufreq(i);
+       if (freq != 0) {
+               seq_printf(m, "clock\t\t: %dMHz\n", freq/1000);
+               return 0;
+       }
+#endif /* CONFIG_CPU_FREQ_PMAC */
+       return of_show_percpuinfo(m, i);
+}
+
+static volatile u32 *sysctrl_regs;
+
+void __init
+pmac_setup_arch(void)
+{
+       struct device_node *cpu;
+       int *fp;
+       unsigned long pvr;
+
+       pvr = PVR_VER(mfspr(SPRN_PVR));
+
+       /* Set loops_per_jiffy to a half-way reasonable value,
+          for use until calibrate_delay gets called. */
+       cpu = find_type_devices("cpu");
+       if (cpu != 0) {
+               fp = (int *) get_property(cpu, "clock-frequency", NULL);
+               if (fp != 0) {
+                       if (pvr == 4 || pvr >= 8)
+                               /* 604, G3, G4 etc. */
+                               loops_per_jiffy = *fp / HZ;
+                       else
+                               /* 601, 603, etc. */
+                               loops_per_jiffy = *fp / (2*HZ);
+               } else
+                       loops_per_jiffy = 50000000 / HZ;
+       }
+
+       /* this area has the CPU identification register
+          and some registers used by smp boards */
+       sysctrl_regs = (volatile u32 *) ioremap(0xf8000000, 0x1000);
+       ohare_init();
+
+       /* Lookup PCI hosts */
+       pmac_find_bridges();
+
+       /* Checks "l2cr-value" property in the registry */
+       if (cpu_has_feature(CPU_FTR_L2CR)) {
+               struct device_node *np = find_devices("cpus");
+               if (np == 0)
+                       np = find_type_devices("cpu");
+               if (np != 0) {
+                       unsigned int *l2cr = (unsigned int *)
+                               get_property(np, "l2cr-value", NULL);
+                       if (l2cr != 0) {
+                               ppc_override_l2cr = 1;
+                               ppc_override_l2cr_value = *l2cr;
+                               _set_L2CR(0);
+                               _set_L2CR(ppc_override_l2cr_value);
+                       }
+               }
+       }
+
+       if (ppc_override_l2cr)
+               printk(KERN_INFO "L2CR overriden (0x%x), backside cache is %s\n",
+                       ppc_override_l2cr_value, (ppc_override_l2cr_value & 0x80000000)
+                               ? "enabled" : "disabled");
+
+#ifdef CONFIG_KGDB
+       zs_kgdb_hook(0);
+#endif
+
+#ifdef CONFIG_ADB_CUDA
+       find_via_cuda();
+#else
+       if (find_devices("via-cuda")) {
+               printk("WARNING ! Your machine is Cuda based but your kernel\n");
+               printk("          wasn't compiled with CONFIG_ADB_CUDA option !\n");
+       }
+#endif
+#ifdef CONFIG_ADB_PMU
+       find_via_pmu();
+#else
+       if (find_devices("via-pmu")) {
+               printk("WARNING ! Your machine is PMU based but your kernel\n");
+               printk("          wasn't compiled with CONFIG_ADB_PMU option !\n");
+       }
+#endif
+#ifdef CONFIG_NVRAM
+       pmac_nvram_init();
+#endif
+#ifdef CONFIG_BLK_DEV_INITRD
+       if (initrd_start)
+               ROOT_DEV = Root_RAM0;
+       else
+#endif
+               ROOT_DEV = DEFAULT_ROOT_DEVICE;
+
+#ifdef CONFIG_SMP
+       /* Check for Core99 */
+       if (find_devices("uni-n") || find_devices("u3"))
+               ppc_md.smp_ops = &core99_smp_ops;
+       else
+               ppc_md.smp_ops = &psurge_smp_ops;
+#endif /* CONFIG_SMP */
+
+       pci_create_OF_bus_map();
+}
+
+static void __init ohare_init(void)
+{
+       /*
+        * Turn on the L2 cache.
+        * We assume that we have a PSX memory controller iff
+        * we have an ohare I/O controller.
+        */
+       if (find_devices("ohare") != NULL) {
+               if (((sysctrl_regs[2] >> 24) & 0xf) >= 3) {
+                       if (sysctrl_regs[4] & 0x10)
+                               sysctrl_regs[4] |= 0x04000020;
+                       else
+                               sysctrl_regs[4] |= 0x04000000;
+                       if(has_l2cache)
+                               printk(KERN_INFO "Level 2 cache enabled\n");
+               }
+       }
+}
+
+extern char *bootpath;
+extern char *bootdevice;
+void *boot_host;
+int boot_target;
+int boot_part;
+extern dev_t boot_dev;
+
+#ifdef CONFIG_SCSI
+void __init
+note_scsi_host(struct device_node *node, void *host)
+{
+       int l;
+       char *p;
+
+       l = strlen(node->full_name);
+       if (bootpath != NULL && bootdevice != NULL
+           && strncmp(node->full_name, bootdevice, l) == 0
+           && (bootdevice[l] == '/' || bootdevice[l] == 0)) {
+               boot_host = host;
+               /*
+                * There's a bug in OF 1.0.5.  (Why am I not surprised.)
+                * If you pass a path like scsi/sd@1:0 to canon, it returns
+                * something like /bandit@F2000000/gc@10/53c94@10000/sd@0,0
+                * That is, the scsi target number doesn't get preserved.
+                * So we pick the target number out of bootpath and use that.
+                */
+               p = strstr(bootpath, "/sd@");
+               if (p != NULL) {
+                       p += 4;
+                       boot_target = simple_strtoul(p, NULL, 10);
+                       p = strchr(p, ':');
+                       if (p != NULL)
+                               boot_part = simple_strtoul(p + 1, NULL, 10);
+               }
+       }
+}
+#endif
+
+#if defined(CONFIG_BLK_DEV_IDE) && defined(CONFIG_BLK_DEV_IDE_PMAC)
+static dev_t __init
+find_ide_boot(void)
+{
+       char *p;
+       int n;
+       dev_t __init pmac_find_ide_boot(char *bootdevice, int n);
+
+       if (bootdevice == NULL)
+               return 0;
+       p = strrchr(bootdevice, '/');
+       if (p == NULL)
+               return 0;
+       n = p - bootdevice;
+
+       return pmac_find_ide_boot(bootdevice, n);
+}
+#endif /* CONFIG_BLK_DEV_IDE && CONFIG_BLK_DEV_IDE_PMAC */
+
+static void __init
+find_boot_device(void)
+{
+#if defined(CONFIG_BLK_DEV_IDE) && defined(CONFIG_BLK_DEV_IDE_PMAC)
+       boot_dev = find_ide_boot();
+#endif
+}
+
+static int initializing = 1;
+/* TODO: Merge the suspend-to-ram with the common code !!!
+ * currently, this is a stub implementation for suspend-to-disk
+ * only
+ */
+
+#ifdef CONFIG_SOFTWARE_SUSPEND
+
+static int pmac_pm_prepare(suspend_state_t state)
+{
+       printk(KERN_DEBUG "%s(%d)\n", __FUNCTION__, state);
+
+       return 0;
+}
+
+static int pmac_pm_enter(suspend_state_t state)
+{
+       printk(KERN_DEBUG "%s(%d)\n", __FUNCTION__, state);
+
+       /* Giveup the lazy FPU & vec so we don't have to back them
+        * up from the low level code
+        */
+       enable_kernel_fp();
+
+#ifdef CONFIG_ALTIVEC
+       if (cur_cpu_spec[0]->cpu_features & CPU_FTR_ALTIVEC)
+               enable_kernel_altivec();
+#endif /* CONFIG_ALTIVEC */
+
+       return 0;
+}
+
+static int pmac_pm_finish(suspend_state_t state)
+{
+       printk(KERN_DEBUG "%s(%d)\n", __FUNCTION__, state);
+
+       /* Restore userland MMU context */
+       set_context(current->active_mm->context, current->active_mm->pgd);
+
+       return 0;
+}
+
+static struct pm_ops pmac_pm_ops = {
+       .pm_disk_mode   = PM_DISK_SHUTDOWN,
+       .prepare        = pmac_pm_prepare,
+       .enter          = pmac_pm_enter,
+       .finish         = pmac_pm_finish,
+};
+
+#endif /* CONFIG_SOFTWARE_SUSPEND */
+
+static int pmac_late_init(void)
+{
+       initializing = 0;
+#ifdef CONFIG_SOFTWARE_SUSPEND
+       pm_set_ops(&pmac_pm_ops);
+#endif /* CONFIG_SOFTWARE_SUSPEND */
+       return 0;
+}
+
+late_initcall(pmac_late_init);
+
+/* can't be __init - can be called whenever a disk is first accessed */
+void
+note_bootable_part(dev_t dev, int part, int goodness)
+{
+       static int found_boot = 0;
+       char *p;
+
+       if (!initializing)
+               return;
+       if ((goodness <= current_root_goodness) &&
+           ROOT_DEV != DEFAULT_ROOT_DEVICE)
+               return;
+       p = strstr(saved_command_line, "root=");
+       if (p != NULL && (p == saved_command_line || p[-1] == ' '))
+               return;
+
+       if (!found_boot) {
+               find_boot_device();
+               found_boot = 1;
+       }
+       if (!boot_dev || dev == boot_dev) {
+               ROOT_DEV = dev + part;
+               boot_dev = 0;
+               current_root_goodness = goodness;
+       }
+}
+
+static void
+pmac_restart(char *cmd)
+{
+#ifdef CONFIG_ADB_CUDA
+       struct adb_request req;
+#endif /* CONFIG_ADB_CUDA */
+
+       switch (sys_ctrler) {
+#ifdef CONFIG_ADB_CUDA
+       case SYS_CTRLER_CUDA:
+               cuda_request(&req, NULL, 2, CUDA_PACKET,
+                            CUDA_RESET_SYSTEM);
+               for (;;)
+                       cuda_poll();
+               break;
+#endif /* CONFIG_ADB_CUDA */
+#ifdef CONFIG_ADB_PMU
+       case SYS_CTRLER_PMU:
+               pmu_restart();
+               break;
+#endif /* CONFIG_ADB_PMU */
+       default: ;
+       }
+}
+
+static void
+pmac_power_off(void)
+{
+#ifdef CONFIG_ADB_CUDA
+       struct adb_request req;
+#endif /* CONFIG_ADB_CUDA */
+
+       switch (sys_ctrler) {
+#ifdef CONFIG_ADB_CUDA
+       case SYS_CTRLER_CUDA:
+               cuda_request(&req, NULL, 2, CUDA_PACKET,
+                            CUDA_POWERDOWN);
+               for (;;)
+                       cuda_poll();
+               break;
+#endif /* CONFIG_ADB_CUDA */
+#ifdef CONFIG_ADB_PMU
+       case SYS_CTRLER_PMU:
+               pmu_shutdown();
+               break;
+#endif /* CONFIG_ADB_PMU */
+       default: ;
+       }
+}
+
+static void
+pmac_halt(void)
+{
+       pmac_power_off();
+}
+
+void __init
+pmac_init(unsigned long r3, unsigned long r4, unsigned long r5,
+         unsigned long r6, unsigned long r7)
+{
+       /* isa_io_base gets set in pmac_find_bridges */
+       isa_mem_base = PMAC_ISA_MEM_BASE;
+       pci_dram_offset = PMAC_PCI_DRAM_OFFSET;
+       ISA_DMA_THRESHOLD = ~0L;
+       DMA_MODE_READ = 1;
+       DMA_MODE_WRITE = 2;
+
+       ppc_md.setup_arch     = pmac_setup_arch;
+       ppc_md.show_cpuinfo   = pmac_show_cpuinfo;
+       ppc_md.show_percpuinfo = pmac_show_percpuinfo;
+       ppc_md.irq_canonicalize = NULL;
+       ppc_md.init_IRQ       = pmac_pic_init;
+       ppc_md.get_irq        = pmac_get_irq; /* Changed later on ... */
+
+       ppc_md.pcibios_fixup  = pmac_pcibios_fixup;
+       ppc_md.pcibios_enable_device_hook = pmac_pci_enable_device_hook;
+       ppc_md.pcibios_after_init = pmac_pcibios_after_init;
+       ppc_md.phys_mem_access_prot = pci_phys_mem_access_prot;
+
+       ppc_md.restart        = pmac_restart;
+       ppc_md.power_off      = pmac_power_off;
+       ppc_md.halt           = pmac_halt;
+
+       ppc_md.time_init      = pmac_time_init;
+       ppc_md.set_rtc_time   = pmac_set_rtc_time;
+       ppc_md.get_rtc_time   = pmac_get_rtc_time;
+       ppc_md.calibrate_decr = pmac_calibrate_decr;
+
+       ppc_md.feature_call   = pmac_do_feature_call;
+
+#if defined(CONFIG_BLK_DEV_IDE) || defined(CONFIG_BLK_DEV_IDE_MODULE)
+#ifdef CONFIG_BLK_DEV_IDE_PMAC
+        ppc_ide_md.ide_init_hwif       = pmac_ide_init_hwif_ports;
+        ppc_ide_md.default_io_base     = pmac_ide_get_base;
+#endif /* CONFIG_BLK_DEV_IDE_PMAC */
+#endif /* defined(CONFIG_BLK_DEV_IDE) || defined(CONFIG_BLK_DEV_IDE_MODULE) */
+
+#ifdef CONFIG_BOOTX_TEXT
+       ppc_md.progress = pmac_progress;
+#endif /* CONFIG_BOOTX_TEXT */
+
+       if (ppc_md.progress) ppc_md.progress("pmac_init(): exit", 0);
+
+}
+
+#ifdef CONFIG_BOOTX_TEXT
+static void __init
+pmac_progress(char *s, unsigned short hex)
+{
+       if (boot_text_mapped) {
+               btext_drawstring(s);
+               btext_drawchar('\n');
+       }
+}
+#endif /* CONFIG_BOOTX_TEXT */
+
+static int __init
+pmac_declare_of_platform_devices(void)
+{
+       struct device_node *np;
+
+       np = find_devices("uni-n");
+       if (np) {
+               for (np = np->child; np != NULL; np = np->sibling)
+                       if (strncmp(np->name, "i2c", 3) == 0) {
+                               of_platform_device_create(np, "uni-n-i2c",
+                                                         NULL);
+                               break;
+                       }
+       }
+       np = find_devices("u3");
+       if (np) {
+               for (np = np->child; np != NULL; np = np->sibling)
+                       if (strncmp(np->name, "i2c", 3) == 0) {
+                               of_platform_device_create(np, "u3-i2c",
+                                                         NULL);
+                               break;
+                       }
+       }
+
+       np = find_devices("valkyrie");
+       if (np)
+               of_platform_device_create(np, "valkyrie", NULL);
+       np = find_devices("platinum");
+       if (np)
+               of_platform_device_create(np, "platinum", NULL);
+
+       return 0;
+}
+
+device_initcall(pmac_declare_of_platform_devices);
diff --git a/arch/powerpc/platforms/powermac/pmac_sleep.S b/arch/powerpc/platforms/powermac/pmac_sleep.S
new file mode 100644 (file)
index 0000000..88419c7
--- /dev/null
@@ -0,0 +1,396 @@
+/*
+ * This file contains sleep low-level functions for PowerBook G3.
+ *    Copyright (C) 1999 Benjamin Herrenschmidt (benh@kernel.crashing.org)
+ *    and Paul Mackerras (paulus@samba.org).
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version
+ * 2 of the License, or (at your option) any later version.
+ *
+ */
+
+#include <linux/config.h>
+#include <asm/processor.h>
+#include <asm/page.h>
+#include <asm/ppc_asm.h>
+#include <asm/cputable.h>
+#include <asm/cache.h>
+#include <asm/thread_info.h>
+#include <asm/asm-offsets.h>
+
+#define MAGIC  0x4c617273      /* 'Lars' */
+
+/*
+ * Structure for storing CPU registers on the stack.
+ */
+#define SL_SP          0
+#define SL_PC          4
+#define SL_MSR         8
+#define SL_SDR1                0xc
+#define SL_SPRG0       0x10    /* 4 sprg's */
+#define SL_DBAT0       0x20
+#define SL_IBAT0       0x28
+#define SL_DBAT1       0x30
+#define SL_IBAT1       0x38
+#define SL_DBAT2       0x40
+#define SL_IBAT2       0x48
+#define SL_DBAT3       0x50
+#define SL_IBAT3       0x58
+#define SL_TB          0x60
+#define SL_R2          0x68
+#define SL_CR          0x6c
+#define SL_R12         0x70    /* r12 to r31 */
+#define SL_SIZE                (SL_R12 + 80)
+
+       .section .text
+       .align  5
+
+#if defined(CONFIG_PM) || defined(CONFIG_CPU_FREQ_PMAC)
+
+/* This gets called by via-pmu.c late during the sleep process.
+ * The PMU was already send the sleep command and will shut us down
+ * soon. We need to save all that is needed and setup the wakeup
+ * vector that will be called by the ROM on wakeup
+ */
+_GLOBAL(low_sleep_handler)
+#ifndef CONFIG_6xx
+       blr
+#else
+       mflr    r0
+       stw     r0,4(r1)
+       stwu    r1,-SL_SIZE(r1)
+       mfcr    r0
+       stw     r0,SL_CR(r1)
+       stw     r2,SL_R2(r1)
+       stmw    r12,SL_R12(r1)
+
+       /* Save MSR & SDR1 */
+       mfmsr   r4
+       stw     r4,SL_MSR(r1)
+       mfsdr1  r4
+       stw     r4,SL_SDR1(r1)
+
+       /* Get a stable timebase and save it */
+1:     mftbu   r4
+       stw     r4,SL_TB(r1)
+       mftb    r5
+       stw     r5,SL_TB+4(r1)
+       mftbu   r3
+       cmpw    r3,r4
+       bne     1b
+
+       /* Save SPRGs */
+       mfsprg  r4,0
+       stw     r4,SL_SPRG0(r1)
+       mfsprg  r4,1
+       stw     r4,SL_SPRG0+4(r1)
+       mfsprg  r4,2
+       stw     r4,SL_SPRG0+8(r1)
+       mfsprg  r4,3
+       stw     r4,SL_SPRG0+12(r1)
+
+       /* Save BATs */
+       mfdbatu r4,0
+       stw     r4,SL_DBAT0(r1)
+       mfdbatl r4,0
+       stw     r4,SL_DBAT0+4(r1)
+       mfdbatu r4,1
+       stw     r4,SL_DBAT1(r1)
+       mfdbatl r4,1
+       stw     r4,SL_DBAT1+4(r1)
+       mfdbatu r4,2
+       stw     r4,SL_DBAT2(r1)
+       mfdbatl r4,2
+       stw     r4,SL_DBAT2+4(r1)
+       mfdbatu r4,3
+       stw     r4,SL_DBAT3(r1)
+       mfdbatl r4,3
+       stw     r4,SL_DBAT3+4(r1)
+       mfibatu r4,0
+       stw     r4,SL_IBAT0(r1)
+       mfibatl r4,0
+       stw     r4,SL_IBAT0+4(r1)
+       mfibatu r4,1
+       stw     r4,SL_IBAT1(r1)
+       mfibatl r4,1
+       stw     r4,SL_IBAT1+4(r1)
+       mfibatu r4,2
+       stw     r4,SL_IBAT2(r1)
+       mfibatl r4,2
+       stw     r4,SL_IBAT2+4(r1)
+       mfibatu r4,3
+       stw     r4,SL_IBAT3(r1)
+       mfibatl r4,3
+       stw     r4,SL_IBAT3+4(r1)
+
+       /* Backup various CPU config stuffs */
+       bl      __save_cpu_setup
+
+       /* The ROM can wake us up via 2 different vectors:
+        *  - On wallstreet & lombard, we must write a magic
+        *    value 'Lars' at address 4 and a pointer to a
+        *    memory location containing the PC to resume from
+        *    at address 0.
+        *  - On Core99, we must store the wakeup vector at
+        *    address 0x80 and eventually it's parameters
+        *    at address 0x84. I've have some trouble with those
+        *    parameters however and I no longer use them.
+        */
+       lis     r5,grackle_wake_up@ha
+       addi    r5,r5,grackle_wake_up@l
+       tophys(r5,r5)
+       stw     r5,SL_PC(r1)
+       lis     r4,KERNELBASE@h
+       tophys(r5,r1)
+       addi    r5,r5,SL_PC
+       lis     r6,MAGIC@ha
+       addi    r6,r6,MAGIC@l
+       stw     r5,0(r4)
+       stw     r6,4(r4)
+       /* Setup stuffs at 0x80-0x84 for Core99 */
+       lis     r3,core99_wake_up@ha
+       addi    r3,r3,core99_wake_up@l
+       tophys(r3,r3)
+       stw     r3,0x80(r4)
+       stw     r5,0x84(r4)
+       /* Store a pointer to our backup storage into
+        * a kernel global
+        */
+       lis r3,sleep_storage@ha
+       addi r3,r3,sleep_storage@l
+       stw r5,0(r3)
+
+       .globl  low_cpu_die
+low_cpu_die:
+       /* Flush & disable all caches */
+       bl      flush_disable_caches
+
+       /* Turn off data relocation. */
+       mfmsr   r3              /* Save MSR in r7 */
+       rlwinm  r3,r3,0,28,26   /* Turn off DR bit */
+       sync
+       mtmsr   r3
+       isync
+
+BEGIN_FTR_SECTION
+       /* Flush any pending L2 data prefetches to work around HW bug */
+       sync
+       lis     r3,0xfff0
+       lwz     r0,0(r3)        /* perform cache-inhibited load to ROM */
+       sync                    /* (caches are disabled at this point) */
+END_FTR_SECTION_IFSET(CPU_FTR_SPEC7450)
+
+/*
+ * Set the HID0 and MSR for sleep.
+ */
+       mfspr   r2,SPRN_HID0
+       rlwinm  r2,r2,0,10,7    /* clear doze, nap */
+       oris    r2,r2,HID0_SLEEP@h
+       sync
+       isync
+       mtspr   SPRN_HID0,r2
+       sync
+
+/* This loop puts us back to sleep in case we have a spurrious
+ * wakeup so that the host bridge properly stays asleep. The
+ * CPU will be turned off, either after a known time (about 1
+ * second) on wallstreet & lombard, or as soon as the CPU enters
+ * SLEEP mode on core99
+ */
+       mfmsr   r2
+       oris    r2,r2,MSR_POW@h
+1:     sync
+       mtmsr   r2
+       isync
+       b       1b
+
+/*
+ * Here is the resume code.
+ */
+
+
+/*
+ * Core99 machines resume here
+ * r4 has the physical address of SL_PC(sp) (unused)
+ */
+_GLOBAL(core99_wake_up)
+       /* Make sure HID0 no longer contains any sleep bit and that data cache
+        * is disabled
+        */
+       mfspr   r3,SPRN_HID0
+       rlwinm  r3,r3,0,11,7            /* clear SLEEP, NAP, DOZE bits */
+       rlwinm  3,r3,0,18,15            /* clear DCE, ICE */
+       mtspr   SPRN_HID0,r3
+       sync
+       isync
+
+       /* sanitize MSR */
+       mfmsr   r3
+       ori     r3,r3,MSR_EE|MSR_IP
+       xori    r3,r3,MSR_EE|MSR_IP
+       sync
+       isync
+       mtmsr   r3
+       sync
+       isync
+
+       /* Recover sleep storage */
+       lis     r3,sleep_storage@ha
+       addi    r3,r3,sleep_storage@l
+       tophys(r3,r3)
+       lwz     r1,0(r3)
+
+       /* Pass thru to older resume code ... */
+/*
+ * Here is the resume code for older machines.
+ * r1 has the physical address of SL_PC(sp).
+ */
+
+grackle_wake_up:
+
+       /* Restore the kernel's segment registers before
+        * we do any r1 memory access as we are not sure they
+        * are in a sane state above the first 256Mb region
+        */
+       li      r0,16           /* load up segment register values */
+       mtctr   r0              /* for context 0 */
+       lis     r3,0x2000       /* Ku = 1, VSID = 0 */
+       li      r4,0
+3:     mtsrin  r3,r4
+       addi    r3,r3,0x111     /* increment VSID */
+       addis   r4,r4,0x1000    /* address of next segment */
+       bdnz    3b
+       sync
+       isync
+
+       subi    r1,r1,SL_PC
+
+       /* Restore various CPU config stuffs */
+       bl      __restore_cpu_setup
+
+       /* Make sure all FPRs have been initialized */
+       bl      reloc_offset
+       bl      __init_fpu_registers
+
+       /* Invalidate & enable L1 cache, we don't care about
+        * whatever the ROM may have tried to write to memory
+        */
+       bl      __inval_enable_L1
+
+       /* Restore the BATs, and SDR1.  Then we can turn on the MMU. */
+       lwz     r4,SL_SDR1(r1)
+       mtsdr1  r4
+       lwz     r4,SL_SPRG0(r1)
+       mtsprg  0,r4
+       lwz     r4,SL_SPRG0+4(r1)
+       mtsprg  1,r4
+       lwz     r4,SL_SPRG0+8(r1)
+       mtsprg  2,r4
+       lwz     r4,SL_SPRG0+12(r1)
+       mtsprg  3,r4
+
+       lwz     r4,SL_DBAT0(r1)
+       mtdbatu 0,r4
+       lwz     r4,SL_DBAT0+4(r1)
+       mtdbatl 0,r4
+       lwz     r4,SL_DBAT1(r1)
+       mtdbatu 1,r4
+       lwz     r4,SL_DBAT1+4(r1)
+       mtdbatl 1,r4
+       lwz     r4,SL_DBAT2(r1)
+       mtdbatu 2,r4
+       lwz     r4,SL_DBAT2+4(r1)
+       mtdbatl 2,r4
+       lwz     r4,SL_DBAT3(r1)
+       mtdbatu 3,r4
+       lwz     r4,SL_DBAT3+4(r1)
+       mtdbatl 3,r4
+       lwz     r4,SL_IBAT0(r1)
+       mtibatu 0,r4
+       lwz     r4,SL_IBAT0+4(r1)
+       mtibatl 0,r4
+       lwz     r4,SL_IBAT1(r1)
+       mtibatu 1,r4
+       lwz     r4,SL_IBAT1+4(r1)
+       mtibatl 1,r4
+       lwz     r4,SL_IBAT2(r1)
+       mtibatu 2,r4
+       lwz     r4,SL_IBAT2+4(r1)
+       mtibatl 2,r4
+       lwz     r4,SL_IBAT3(r1)
+       mtibatu 3,r4
+       lwz     r4,SL_IBAT3+4(r1)
+       mtibatl 3,r4
+
+BEGIN_FTR_SECTION
+       li      r4,0
+       mtspr   SPRN_DBAT4U,r4
+       mtspr   SPRN_DBAT4L,r4
+       mtspr   SPRN_DBAT5U,r4
+       mtspr   SPRN_DBAT5L,r4
+       mtspr   SPRN_DBAT6U,r4
+       mtspr   SPRN_DBAT6L,r4
+       mtspr   SPRN_DBAT7U,r4
+       mtspr   SPRN_DBAT7L,r4
+       mtspr   SPRN_IBAT4U,r4
+       mtspr   SPRN_IBAT4L,r4
+       mtspr   SPRN_IBAT5U,r4
+       mtspr   SPRN_IBAT5L,r4
+       mtspr   SPRN_IBAT6U,r4
+       mtspr   SPRN_IBAT6L,r4
+       mtspr   SPRN_IBAT7U,r4
+       mtspr   SPRN_IBAT7L,r4
+END_FTR_SECTION_IFSET(CPU_FTR_HAS_HIGH_BATS)
+
+       /* Flush all TLBs */
+       lis     r4,0x1000
+1:     addic.  r4,r4,-0x1000
+       tlbie   r4
+       blt     1b
+       sync
+
+       /* restore the MSR and turn on the MMU */
+       lwz     r3,SL_MSR(r1)
+       bl      turn_on_mmu
+
+       /* get back the stack pointer */
+       tovirt(r1,r1)
+
+       /* Restore TB */
+       li      r3,0
+       mttbl   r3
+       lwz     r3,SL_TB(r1)
+       lwz     r4,SL_TB+4(r1)
+       mttbu   r3
+       mttbl   r4
+
+       /* Restore the callee-saved registers and return */
+       lwz     r0,SL_CR(r1)
+       mtcr    r0
+       lwz     r2,SL_R2(r1)
+       lmw     r12,SL_R12(r1)
+       addi    r1,r1,SL_SIZE
+       lwz     r0,4(r1)
+       mtlr    r0
+       blr
+
+turn_on_mmu:
+       mflr    r4
+       tovirt(r4,r4)
+       mtsrr0  r4
+       mtsrr1  r3
+       sync
+       isync
+       rfi
+
+#endif /* defined(CONFIG_PM) || defined(CONFIG_CPU_FREQ) */
+
+       .section .data
+       .balign L1_CACHE_LINE_SIZE
+sleep_storage:
+       .long 0
+       .balign L1_CACHE_LINE_SIZE, 0
+
+#endif /* CONFIG_6xx */
+       .section .text
diff --git a/arch/powerpc/platforms/powermac/pmac_smp.c b/arch/powerpc/platforms/powermac/pmac_smp.c
new file mode 100644 (file)
index 0000000..995e909
--- /dev/null
@@ -0,0 +1,716 @@
+/*
+ * SMP support for power macintosh.
+ *
+ * We support both the old "powersurge" SMP architecture
+ * and the current Core99 (G4 PowerMac) machines.
+ *
+ * Note that we don't support the very first rev. of
+ * Apple/DayStar 2 CPUs board, the one with the funky
+ * watchdog. Hopefully, none of these should be there except
+ * maybe internally to Apple. I should probably still add some
+ * code to detect this card though and disable SMP. --BenH.
+ *
+ * Support Macintosh G4 SMP by Troy Benjegerdes (hozer@drgw.net)
+ * and Ben Herrenschmidt <benh@kernel.crashing.org>.
+ *
+ * Support for DayStar quad CPU cards
+ * Copyright (C) XLR8, Inc. 1994-2000
+ *
+ *  This program is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU General Public License
+ *  as published by the Free Software Foundation; either version
+ *  2 of the License, or (at your option) any later version.
+ */
+#include <linux/config.h>
+#include <linux/kernel.h>
+#include <linux/sched.h>
+#include <linux/smp.h>
+#include <linux/smp_lock.h>
+#include <linux/interrupt.h>
+#include <linux/kernel_stat.h>
+#include <linux/delay.h>
+#include <linux/init.h>
+#include <linux/spinlock.h>
+#include <linux/errno.h>
+#include <linux/hardirq.h>
+#include <linux/cpu.h>
+
+#include <asm/ptrace.h>
+#include <asm/atomic.h>
+#include <asm/irq.h>
+#include <asm/page.h>
+#include <asm/pgtable.h>
+#include <asm/sections.h>
+#include <asm/io.h>
+#include <asm/prom.h>
+#include <asm/smp.h>
+#include <asm/residual.h>
+#include <asm/machdep.h>
+#include <asm/pmac_feature.h>
+#include <asm/time.h>
+#include <asm/open_pic.h>
+#include <asm/cacheflush.h>
+#include <asm/keylargo.h>
+
+/*
+ * Powersurge (old powermac SMP) support.
+ */
+
+extern void __secondary_start_pmac_0(void);
+
+/* Addresses for powersurge registers */
+#define HAMMERHEAD_BASE                0xf8000000
+#define HHEAD_CONFIG           0x90
+#define HHEAD_SEC_INTR         0xc0
+
+/* register for interrupting the primary processor on the powersurge */
+/* N.B. this is actually the ethernet ROM! */
+#define PSURGE_PRI_INTR                0xf3019000
+
+/* register for storing the start address for the secondary processor */
+/* N.B. this is the PCI config space address register for the 1st bridge */
+#define PSURGE_START           0xf2800000
+
+/* Daystar/XLR8 4-CPU card */
+#define PSURGE_QUAD_REG_ADDR   0xf8800000
+
+#define PSURGE_QUAD_IRQ_SET    0
+#define PSURGE_QUAD_IRQ_CLR    1
+#define PSURGE_QUAD_IRQ_PRIMARY        2
+#define PSURGE_QUAD_CKSTOP_CTL 3
+#define PSURGE_QUAD_PRIMARY_ARB        4
+#define PSURGE_QUAD_BOARD_ID   6
+#define PSURGE_QUAD_WHICH_CPU  7
+#define PSURGE_QUAD_CKSTOP_RDBK        8
+#define PSURGE_QUAD_RESET_CTL  11
+
+#define PSURGE_QUAD_OUT(r, v)  (out_8(quad_base + ((r) << 4) + 4, (v)))
+#define PSURGE_QUAD_IN(r)      (in_8(quad_base + ((r) << 4) + 4) & 0x0f)
+#define PSURGE_QUAD_BIS(r, v)  (PSURGE_QUAD_OUT((r), PSURGE_QUAD_IN(r) | (v)))
+#define PSURGE_QUAD_BIC(r, v)  (PSURGE_QUAD_OUT((r), PSURGE_QUAD_IN(r) & ~(v)))
+
+/* virtual addresses for the above */
+static volatile u8 __iomem *hhead_base;
+static volatile u8 __iomem *quad_base;
+static volatile u32 __iomem *psurge_pri_intr;
+static volatile u8 __iomem *psurge_sec_intr;
+static volatile u32 __iomem *psurge_start;
+
+/* values for psurge_type */
+#define PSURGE_NONE            -1
+#define PSURGE_DUAL            0
+#define PSURGE_QUAD_OKEE       1
+#define PSURGE_QUAD_COTTON     2
+#define PSURGE_QUAD_ICEGRASS   3
+
+/* what sort of powersurge board we have */
+static int psurge_type = PSURGE_NONE;
+
+/* L2 and L3 cache settings to pass from CPU0 to CPU1 */
+volatile static long int core99_l2_cache;
+volatile static long int core99_l3_cache;
+
+/* Timebase freeze GPIO */
+static unsigned int core99_tb_gpio;
+
+/* Sync flag for HW tb sync */
+static volatile int sec_tb_reset = 0;
+static unsigned int pri_tb_hi, pri_tb_lo;
+static unsigned int pri_tb_stamp;
+
+static void __devinit core99_init_caches(int cpu)
+{
+       if (!cpu_has_feature(CPU_FTR_L2CR))
+               return;
+
+       if (cpu == 0) {
+               core99_l2_cache = _get_L2CR();
+               printk("CPU0: L2CR is %lx\n", core99_l2_cache);
+       } else {
+               printk("CPU%d: L2CR was %lx\n", cpu, _get_L2CR());
+               _set_L2CR(0);
+               _set_L2CR(core99_l2_cache);
+               printk("CPU%d: L2CR set to %lx\n", cpu, core99_l2_cache);
+       }
+
+       if (!cpu_has_feature(CPU_FTR_L3CR))
+               return;
+
+       if (cpu == 0){
+               core99_l3_cache = _get_L3CR();
+               printk("CPU0: L3CR is %lx\n", core99_l3_cache);
+       } else {
+               printk("CPU%d: L3CR was %lx\n", cpu, _get_L3CR());
+               _set_L3CR(0);
+               _set_L3CR(core99_l3_cache);
+               printk("CPU%d: L3CR set to %lx\n", cpu, core99_l3_cache);
+       }
+}
+
+/*
+ * Set and clear IPIs for powersurge.
+ */
+static inline void psurge_set_ipi(int cpu)
+{
+       if (psurge_type == PSURGE_NONE)
+               return;
+       if (cpu == 0)
+               in_be32(psurge_pri_intr);
+       else if (psurge_type == PSURGE_DUAL)
+               out_8(psurge_sec_intr, 0);
+       else
+               PSURGE_QUAD_OUT(PSURGE_QUAD_IRQ_SET, 1 << cpu);
+}
+
+static inline void psurge_clr_ipi(int cpu)
+{
+       if (cpu > 0) {
+               switch(psurge_type) {
+               case PSURGE_DUAL:
+                       out_8(psurge_sec_intr, ~0);
+               case PSURGE_NONE:
+                       break;
+               default:
+                       PSURGE_QUAD_OUT(PSURGE_QUAD_IRQ_CLR, 1 << cpu);
+               }
+       }
+}
+
+/*
+ * On powersurge (old SMP powermac architecture) we don't have
+ * separate IPIs for separate messages like openpic does.  Instead
+ * we have a bitmap for each processor, where a 1 bit means that
+ * the corresponding message is pending for that processor.
+ * Ideally each cpu's entry would be in a different cache line.
+ *  -- paulus.
+ */
+static unsigned long psurge_smp_message[NR_CPUS];
+
+void psurge_smp_message_recv(struct pt_regs *regs)
+{
+       int cpu = smp_processor_id();
+       int msg;
+
+       /* clear interrupt */
+       psurge_clr_ipi(cpu);
+
+       if (num_online_cpus() < 2)
+               return;
+
+       /* make sure there is a message there */
+       for (msg = 0; msg < 4; msg++)
+               if (test_and_clear_bit(msg, &psurge_smp_message[cpu]))
+                       smp_message_recv(msg, regs);
+}
+
+irqreturn_t psurge_primary_intr(int irq, void *d, struct pt_regs *regs)
+{
+       psurge_smp_message_recv(regs);
+       return IRQ_HANDLED;
+}
+
+static void smp_psurge_message_pass(int target, int msg, unsigned long data,
+                                          int wait)
+{
+       int i;
+
+       if (num_online_cpus() < 2)
+               return;
+
+       for (i = 0; i < NR_CPUS; i++) {
+               if (!cpu_online(i))
+                       continue;
+               if (target == MSG_ALL
+                   || (target == MSG_ALL_BUT_SELF && i != smp_processor_id())
+                   || target == i) {
+                       set_bit(msg, &psurge_smp_message[i]);
+                       psurge_set_ipi(i);
+               }
+       }
+}
+
+/*
+ * Determine a quad card presence. We read the board ID register, we
+ * force the data bus to change to something else, and we read it again.
+ * It it's stable, then the register probably exist (ugh !)
+ */
+static int __init psurge_quad_probe(void)
+{
+       int type;
+       unsigned int i;
+
+       type = PSURGE_QUAD_IN(PSURGE_QUAD_BOARD_ID);
+       if (type < PSURGE_QUAD_OKEE || type > PSURGE_QUAD_ICEGRASS
+           || type != PSURGE_QUAD_IN(PSURGE_QUAD_BOARD_ID))
+               return PSURGE_DUAL;
+
+       /* looks OK, try a slightly more rigorous test */
+       /* bogus is not necessarily cacheline-aligned,
+          though I don't suppose that really matters.  -- paulus */
+       for (i = 0; i < 100; i++) {
+               volatile u32 bogus[8];
+               bogus[(0+i)%8] = 0x00000000;
+               bogus[(1+i)%8] = 0x55555555;
+               bogus[(2+i)%8] = 0xFFFFFFFF;
+               bogus[(3+i)%8] = 0xAAAAAAAA;
+               bogus[(4+i)%8] = 0x33333333;
+               bogus[(5+i)%8] = 0xCCCCCCCC;
+               bogus[(6+i)%8] = 0xCCCCCCCC;
+               bogus[(7+i)%8] = 0x33333333;
+               wmb();
+               asm volatile("dcbf 0,%0" : : "r" (bogus) : "memory");
+               mb();
+               if (type != PSURGE_QUAD_IN(PSURGE_QUAD_BOARD_ID))
+                       return PSURGE_DUAL;
+       }
+       return type;
+}
+
+static void __init psurge_quad_init(void)
+{
+       int procbits;
+
+       if (ppc_md.progress) ppc_md.progress("psurge_quad_init", 0x351);
+       procbits = ~PSURGE_QUAD_IN(PSURGE_QUAD_WHICH_CPU);
+       if (psurge_type == PSURGE_QUAD_ICEGRASS)
+               PSURGE_QUAD_BIS(PSURGE_QUAD_RESET_CTL, procbits);
+       else
+               PSURGE_QUAD_BIC(PSURGE_QUAD_CKSTOP_CTL, procbits);
+       mdelay(33);
+       out_8(psurge_sec_intr, ~0);
+       PSURGE_QUAD_OUT(PSURGE_QUAD_IRQ_CLR, procbits);
+       PSURGE_QUAD_BIS(PSURGE_QUAD_RESET_CTL, procbits);
+       if (psurge_type != PSURGE_QUAD_ICEGRASS)
+               PSURGE_QUAD_BIS(PSURGE_QUAD_CKSTOP_CTL, procbits);
+       PSURGE_QUAD_BIC(PSURGE_QUAD_PRIMARY_ARB, procbits);
+       mdelay(33);
+       PSURGE_QUAD_BIC(PSURGE_QUAD_RESET_CTL, procbits);
+       mdelay(33);
+       PSURGE_QUAD_BIS(PSURGE_QUAD_PRIMARY_ARB, procbits);
+       mdelay(33);
+}
+
+static int __init smp_psurge_probe(void)
+{
+       int i, ncpus;
+
+       /* We don't do SMP on the PPC601 -- paulus */
+       if (PVR_VER(mfspr(SPRN_PVR)) == 1)
+               return 1;
+
+       /*
+        * The powersurge cpu board can be used in the generation
+        * of powermacs that have a socket for an upgradeable cpu card,
+        * including the 7500, 8500, 9500, 9600.
+        * The device tree doesn't tell you if you have 2 cpus because
+        * OF doesn't know anything about the 2nd processor.
+        * Instead we look for magic bits in magic registers,
+        * in the hammerhead memory controller in the case of the
+        * dual-cpu powersurge board.  -- paulus.
+        */
+       if (find_devices("hammerhead") == NULL)
+               return 1;
+
+       hhead_base = ioremap(HAMMERHEAD_BASE, 0x800);
+       quad_base = ioremap(PSURGE_QUAD_REG_ADDR, 1024);
+       psurge_sec_intr = hhead_base + HHEAD_SEC_INTR;
+
+       psurge_type = psurge_quad_probe();
+       if (psurge_type != PSURGE_DUAL) {
+               psurge_quad_init();
+               /* All released cards using this HW design have 4 CPUs */
+               ncpus = 4;
+       } else {
+               iounmap(quad_base);
+               if ((in_8(hhead_base + HHEAD_CONFIG) & 0x02) == 0) {
+                       /* not a dual-cpu card */
+                       iounmap(hhead_base);
+                       psurge_type = PSURGE_NONE;
+                       return 1;
+               }
+               ncpus = 2;
+       }
+
+       psurge_start = ioremap(PSURGE_START, 4);
+       psurge_pri_intr = ioremap(PSURGE_PRI_INTR, 4);
+
+       /* this is not actually strictly necessary -- paulus. */
+       for (i = 1; i < ncpus; ++i)
+               smp_hw_index[i] = i;
+
+       if (ppc_md.progress) ppc_md.progress("smp_psurge_probe - done", 0x352);
+
+       return ncpus;
+}
+
+static void __init smp_psurge_kick_cpu(int nr)
+{
+       unsigned long start = __pa(__secondary_start_pmac_0) + nr * 8;
+       unsigned long a;
+
+       /* may need to flush here if secondary bats aren't setup */
+       for (a = KERNELBASE; a < KERNELBASE + 0x800000; a += 32)
+               asm volatile("dcbf 0,%0" : : "r" (a) : "memory");
+       asm volatile("sync");
+
+       if (ppc_md.progress) ppc_md.progress("smp_psurge_kick_cpu", 0x353);
+
+       out_be32(psurge_start, start);
+       mb();
+
+       psurge_set_ipi(nr);
+       udelay(10);
+       psurge_clr_ipi(nr);
+
+       if (ppc_md.progress) ppc_md.progress("smp_psurge_kick_cpu - done", 0x354);
+}
+
+/*
+ * With the dual-cpu powersurge board, the decrementers and timebases
+ * of both cpus are frozen after the secondary cpu is started up,
+ * until we give the secondary cpu another interrupt.  This routine
+ * uses this to get the timebases synchronized.
+ *  -- paulus.
+ */
+static void __init psurge_dual_sync_tb(int cpu_nr)
+{
+       int t;
+
+       set_dec(tb_ticks_per_jiffy);
+       set_tb(0, 0);
+       last_jiffy_stamp(cpu_nr) = 0;
+
+       if (cpu_nr > 0) {
+               mb();
+               sec_tb_reset = 1;
+               return;
+       }
+
+       /* wait for the secondary to have reset its TB before proceeding */
+       for (t = 10000000; t > 0 && !sec_tb_reset; --t)
+               ;
+
+       /* now interrupt the secondary, starting both TBs */
+       psurge_set_ipi(1);
+
+       smp_tb_synchronized = 1;
+}
+
+static struct irqaction psurge_irqaction = {
+       .handler = psurge_primary_intr,
+       .flags = SA_INTERRUPT,
+       .mask = CPU_MASK_NONE,
+       .name = "primary IPI",
+};
+
+static void __init smp_psurge_setup_cpu(int cpu_nr)
+{
+
+       if (cpu_nr == 0) {
+               /* If we failed to start the second CPU, we should still
+                * send it an IPI to start the timebase & DEC or we might
+                * have them stuck.
+                */
+               if (num_online_cpus() < 2) {
+                       if (psurge_type == PSURGE_DUAL)
+                               psurge_set_ipi(1);
+                       return;
+               }
+               /* reset the entry point so if we get another intr we won't
+                * try to startup again */
+               out_be32(psurge_start, 0x100);
+               if (setup_irq(30, &psurge_irqaction))
+                       printk(KERN_ERR "Couldn't get primary IPI interrupt");
+       }
+
+       if (psurge_type == PSURGE_DUAL)
+               psurge_dual_sync_tb(cpu_nr);
+}
+
+void __init smp_psurge_take_timebase(void)
+{
+       /* Dummy implementation */
+}
+
+void __init smp_psurge_give_timebase(void)
+{
+       /* Dummy implementation */
+}
+
+static int __init smp_core99_probe(void)
+{
+#ifdef CONFIG_6xx
+       extern int powersave_nap;
+#endif
+       struct device_node *cpus, *firstcpu;
+       int i, ncpus = 0, boot_cpu = -1;
+       u32 *tbprop = NULL;
+
+       if (ppc_md.progress) ppc_md.progress("smp_core99_probe", 0x345);
+       cpus = firstcpu = find_type_devices("cpu");
+       while(cpus != NULL) {
+               u32 *regprop = (u32 *)get_property(cpus, "reg", NULL);
+               char *stateprop = (char *)get_property(cpus, "state", NULL);
+               if (regprop != NULL && stateprop != NULL &&
+                   !strncmp(stateprop, "running", 7))
+                       boot_cpu = *regprop;
+               ++ncpus;
+               cpus = cpus->next;
+       }
+       if (boot_cpu == -1)
+               printk(KERN_WARNING "Couldn't detect boot CPU !\n");
+       if (boot_cpu != 0)
+               printk(KERN_WARNING "Boot CPU is %d, unsupported setup !\n", boot_cpu);
+
+       if (machine_is_compatible("MacRISC4")) {
+               extern struct smp_ops_t core99_smp_ops;
+
+               core99_smp_ops.take_timebase = smp_generic_take_timebase;
+               core99_smp_ops.give_timebase = smp_generic_give_timebase;
+       } else {
+               if (firstcpu != NULL)
+                       tbprop = (u32 *)get_property(firstcpu, "timebase-enable", NULL);
+               if (tbprop)
+                       core99_tb_gpio = *tbprop;
+               else
+                       core99_tb_gpio = KL_GPIO_TB_ENABLE;
+       }
+
+       if (ncpus > 1) {
+               mpic_request_ipis();
+               for (i = 1; i < ncpus; ++i)
+                       smp_hw_index[i] = i;
+#ifdef CONFIG_6xx
+               powersave_nap = 0;
+#endif
+               core99_init_caches(0);
+       }
+
+       return ncpus;
+}
+
+static void __devinit smp_core99_kick_cpu(int nr)
+{
+       unsigned long save_vector, new_vector;
+       unsigned long flags;
+
+       volatile unsigned long *vector
+                = ((volatile unsigned long *)(KERNELBASE+0x100));
+       if (nr < 0 || nr > 3)
+               return;
+       if (ppc_md.progress) ppc_md.progress("smp_core99_kick_cpu", 0x346);
+
+       local_irq_save(flags);
+       local_irq_disable();
+
+       /* Save reset vector */
+       save_vector = *vector;
+
+       /* Setup fake reset vector that does    
+        *   b __secondary_start_pmac_0 + nr*8 - KERNELBASE
+        */
+       new_vector = (unsigned long) __secondary_start_pmac_0 + nr * 8;
+       *vector = 0x48000002 + new_vector - KERNELBASE;
+
+       /* flush data cache and inval instruction cache */
+       flush_icache_range((unsigned long) vector, (unsigned long) vector + 4);
+
+       /* Put some life in our friend */
+       pmac_call_feature(PMAC_FTR_RESET_CPU, NULL, nr, 0);
+
+       /* FIXME: We wait a bit for the CPU to take the exception, I should
+        * instead wait for the entry code to set something for me. Well,
+        * ideally, all that crap will be done in prom.c and the CPU left
+        * in a RAM-based wait loop like CHRP.
+        */
+       mdelay(1);
+
+       /* Restore our exception vector */
+       *vector = save_vector;
+       flush_icache_range((unsigned long) vector, (unsigned long) vector + 4);
+
+       local_irq_restore(flags);
+       if (ppc_md.progress) ppc_md.progress("smp_core99_kick_cpu done", 0x347);
+}
+
+static void __devinit smp_core99_setup_cpu(int cpu_nr)
+{
+       /* Setup L2/L3 */
+       if (cpu_nr != 0)
+               core99_init_caches(cpu_nr);
+
+       /* Setup openpic */
+       mpic_setup_this_cpu();
+
+       if (cpu_nr == 0) {
+#ifdef CONFIG_POWER4
+               extern void g5_phy_disable_cpu1(void);
+
+               /* If we didn't start the second CPU, we must take
+                * it off the bus
+                */
+               if (machine_is_compatible("MacRISC4") &&
+                   num_online_cpus() < 2)              
+                       g5_phy_disable_cpu1();
+#endif /* CONFIG_POWER4 */
+               if (ppc_md.progress) ppc_md.progress("core99_setup_cpu 0 done", 0x349);
+       }
+}
+
+/* not __init, called in sleep/wakeup code */
+void smp_core99_take_timebase(void)
+{
+       unsigned long flags;
+
+       /* tell the primary we're here */
+       sec_tb_reset = 1;
+       mb();
+
+       /* wait for the primary to set pri_tb_hi/lo */
+       while (sec_tb_reset < 2)
+               mb();
+
+       /* set our stuff the same as the primary */
+       local_irq_save(flags);
+       set_dec(1);
+       set_tb(pri_tb_hi, pri_tb_lo);
+       last_jiffy_stamp(smp_processor_id()) = pri_tb_stamp;
+       mb();
+
+       /* tell the primary we're done */
+               sec_tb_reset = 0;
+       mb();
+       local_irq_restore(flags);
+}
+
+/* not __init, called in sleep/wakeup code */
+void smp_core99_give_timebase(void)
+{
+       unsigned long flags;
+       unsigned int t;
+
+       /* wait for the secondary to be in take_timebase */
+       for (t = 100000; t > 0 && !sec_tb_reset; --t)
+               udelay(10);
+       if (!sec_tb_reset) {
+               printk(KERN_WARNING "Timeout waiting sync on second CPU\n");
+               return;
+       }
+
+       /* freeze the timebase and read it */
+       /* disable interrupts so the timebase is disabled for the
+          shortest possible time */
+       local_irq_save(flags);
+       pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, core99_tb_gpio, 4);
+       pmac_call_feature(PMAC_FTR_READ_GPIO, NULL, core99_tb_gpio, 0);
+       mb();
+       pri_tb_hi = get_tbu();
+       pri_tb_lo = get_tbl();
+       pri_tb_stamp = last_jiffy_stamp(smp_processor_id());
+       mb();
+
+       /* tell the secondary we're ready */
+       sec_tb_reset = 2;
+       mb();
+
+       /* wait for the secondary to have taken it */
+       for (t = 100000; t > 0 && sec_tb_reset; --t)
+               udelay(10);
+       if (sec_tb_reset)
+               printk(KERN_WARNING "Timeout waiting sync(2) on second CPU\n");
+       else
+               smp_tb_synchronized = 1;
+
+       /* Now, restart the timebase by leaving the GPIO to an open collector */
+               pmac_call_feature(PMAC_FTR_WRITE_GPIO, NULL, core99_tb_gpio, 0);
+        pmac_call_feature(PMAC_FTR_READ_GPIO, NULL, core99_tb_gpio, 0);
+       local_irq_restore(flags);
+}
+
+void smp_core99_message_pass(int target, int msg, unsigned long data, int wait)
+{
+       cpumask_t mask = CPU_MASK_ALL;
+       /* make sure we're sending something that translates to an IPI */
+       if (msg > 0x3) {
+               printk("SMP %d: smp_message_pass: unknown msg %d\n",
+                      smp_processor_id(), msg);
+               return;
+       }
+       switch (target) {
+       case MSG_ALL:
+               mpic_send_ipi(msg, mask);
+               break;
+       case MSG_ALL_BUT_SELF:
+               cpu_clear(smp_processor_id(), mask);
+               mpic_send_ipi(msg, mask);
+               break;
+       default:
+               mpic_send_ipi(msg, cpumask_of_cpu(target));
+               break;
+       }
+}
+
+
+/* PowerSurge-style Macs */
+struct smp_ops_t psurge_smp_ops = {
+       .message_pass   = smp_psurge_message_pass,
+       .probe          = smp_psurge_probe,
+       .kick_cpu       = smp_psurge_kick_cpu,
+       .setup_cpu      = smp_psurge_setup_cpu,
+       .give_timebase  = smp_psurge_give_timebase,
+       .take_timebase  = smp_psurge_take_timebase,
+};
+
+/* Core99 Macs (dual G4s) */
+struct smp_ops_t core99_smp_ops = {
+       .message_pass   = smp_core99_message_pass,
+       .probe          = smp_core99_probe,
+       .kick_cpu       = smp_core99_kick_cpu,
+       .setup_cpu      = smp_core99_setup_cpu,
+       .give_timebase  = smp_core99_give_timebase,
+       .take_timebase  = smp_core99_take_timebase,
+};
+
+#ifdef CONFIG_HOTPLUG_CPU
+
+int __cpu_disable(void)
+{
+       cpu_clear(smp_processor_id(), cpu_online_map);
+
+       /* XXX reset cpu affinity here */
+       openpic_set_priority(0xf);
+       asm volatile("mtdec %0" : : "r" (0x7fffffff));
+       mb();
+       udelay(20);
+       asm volatile("mtdec %0" : : "r" (0x7fffffff));
+       return 0;
+}
+
+extern void low_cpu_die(void) __attribute__((noreturn)); /* in pmac_sleep.S */
+static int cpu_dead[NR_CPUS];
+
+void cpu_die(void)
+{
+       local_irq_disable();
+       cpu_dead[smp_processor_id()] = 1;
+       mb();
+       low_cpu_die();
+}
+
+void __cpu_die(unsigned int cpu)
+{
+       int timeout;
+
+       timeout = 1000;
+       while (!cpu_dead[cpu]) {
+               if (--timeout == 0) {
+                       printk("CPU %u refused to die!\n", cpu);
+                       break;
+               }
+               msleep(1);
+       }
+       cpu_callin_map[cpu] = 0;
+       cpu_dead[cpu] = 0;
+}
+
+#endif
diff --git a/arch/powerpc/platforms/powermac/pmac_time.c b/arch/powerpc/platforms/powermac/pmac_time.c
new file mode 100644 (file)
index 0000000..ff6adff
--- /dev/null
@@ -0,0 +1,291 @@
+/*
+ * Support for periodic interrupts (100 per second) and for getting
+ * the current time from the RTC on Power Macintoshes.
+ *
+ * We use the decrementer register for our periodic interrupts.
+ *
+ * Paul Mackerras      August 1996.
+ * Copyright (C) 1996 Paul Mackerras.
+ */
+#include <linux/config.h>
+#include <linux/errno.h>
+#include <linux/sched.h>
+#include <linux/kernel.h>
+#include <linux/param.h>
+#include <linux/string.h>
+#include <linux/mm.h>
+#include <linux/init.h>
+#include <linux/time.h>
+#include <linux/adb.h>
+#include <linux/cuda.h>
+#include <linux/pmu.h>
+#include <linux/hardirq.h>
+
+#include <asm/sections.h>
+#include <asm/prom.h>
+#include <asm/system.h>
+#include <asm/io.h>
+#include <asm/pgtable.h>
+#include <asm/machdep.h>
+#include <asm/time.h>
+#include <asm/nvram.h>
+
+/* Apparently the RTC stores seconds since 1 Jan 1904 */
+#define RTC_OFFSET     2082844800
+
+/*
+ * Calibrate the decrementer frequency with the VIA timer 1.
+ */
+#define VIA_TIMER_FREQ_6       4700000 /* time 1 frequency * 6 */
+
+/* VIA registers */
+#define RS             0x200           /* skip between registers */
+#define T1CL           (4*RS)          /* Timer 1 ctr/latch (low 8 bits) */
+#define T1CH           (5*RS)          /* Timer 1 counter (high 8 bits) */
+#define T1LL           (6*RS)          /* Timer 1 latch (low 8 bits) */
+#define T1LH           (7*RS)          /* Timer 1 latch (high 8 bits) */
+#define ACR            (11*RS)         /* Auxiliary control register */
+#define IFR            (13*RS)         /* Interrupt flag register */
+
+/* Bits in ACR */
+#define T1MODE         0xc0            /* Timer 1 mode */
+#define T1MODE_CONT    0x40            /*  continuous interrupts */
+
+/* Bits in IFR and IER */
+#define T1_INT         0x40            /* Timer 1 interrupt */
+
+extern struct timezone sys_tz;
+
+long __init
+pmac_time_init(void)
+{
+#ifdef CONFIG_NVRAM
+       s32 delta = 0;
+       int dst;
+       
+       delta = ((s32)pmac_xpram_read(PMAC_XPRAM_MACHINE_LOC + 0x9)) << 16;
+       delta |= ((s32)pmac_xpram_read(PMAC_XPRAM_MACHINE_LOC + 0xa)) << 8;
+       delta |= pmac_xpram_read(PMAC_XPRAM_MACHINE_LOC + 0xb);
+       if (delta & 0x00800000UL)
+               delta |= 0xFF000000UL;
+       dst = ((pmac_xpram_read(PMAC_XPRAM_MACHINE_LOC + 0x8) & 0x80) != 0);
+       printk("GMT Delta read from XPRAM: %d minutes, DST: %s\n", delta/60,
+               dst ? "on" : "off");
+       return delta;
+#else
+       return 0;
+#endif
+}
+
+unsigned long
+pmac_get_rtc_time(void)
+{
+#if defined(CONFIG_ADB_CUDA) || defined(CONFIG_ADB_PMU)
+       struct adb_request req;
+       unsigned long now;
+#endif
+
+       /* Get the time from the RTC */
+       switch (sys_ctrler) {
+#ifdef CONFIG_ADB_CUDA
+       case SYS_CTRLER_CUDA:
+               if (cuda_request(&req, NULL, 2, CUDA_PACKET, CUDA_GET_TIME) < 0)
+                       return 0;
+               while (!req.complete)
+                       cuda_poll();
+               if (req.reply_len != 7)
+                       printk(KERN_ERR "pmac_get_rtc_time: got %d byte reply\n",
+                              req.reply_len);
+               now = (req.reply[3] << 24) + (req.reply[4] << 16)
+                       + (req.reply[5] << 8) + req.reply[6];
+               return now - RTC_OFFSET;
+#endif /* CONFIG_ADB_CUDA */
+#ifdef CONFIG_ADB_PMU
+       case SYS_CTRLER_PMU:
+               if (pmu_request(&req, NULL, 1, PMU_READ_RTC) < 0)
+                       return 0;
+               while (!req.complete)
+                       pmu_poll();
+               if (req.reply_len != 4)
+                       printk(KERN_ERR "pmac_get_rtc_time: got %d byte reply\n",
+                              req.reply_len);
+               now = (req.reply[0] << 24) + (req.reply[1] << 16)
+                       + (req.reply[2] << 8) + req.reply[3];
+               return now - RTC_OFFSET;
+#endif /* CONFIG_ADB_PMU */
+       default: ;
+       }
+       return 0;
+}
+
+int
+pmac_set_rtc_time(unsigned long nowtime)
+{
+#if defined(CONFIG_ADB_CUDA) || defined(CONFIG_ADB_PMU)
+       struct adb_request req;
+#endif
+
+       nowtime += RTC_OFFSET;
+
+       switch (sys_ctrler) {
+#ifdef CONFIG_ADB_CUDA
+       case SYS_CTRLER_CUDA:
+               if (cuda_request(&req, NULL, 6, CUDA_PACKET, CUDA_SET_TIME,
+                                nowtime >> 24, nowtime >> 16, nowtime >> 8, nowtime) < 0)
+                       return 0;
+               while (!req.complete)
+                       cuda_poll();
+               if ((req.reply_len != 3) && (req.reply_len != 7))
+                       printk(KERN_ERR "pmac_set_rtc_time: got %d byte reply\n",
+                              req.reply_len);
+               return 1;
+#endif /* CONFIG_ADB_CUDA */
+#ifdef CONFIG_ADB_PMU
+       case SYS_CTRLER_PMU:
+               if (pmu_request(&req, NULL, 5, PMU_SET_RTC,
+                               nowtime >> 24, nowtime >> 16, nowtime >> 8, nowtime) < 0)
+                       return 0;
+               while (!req.complete)
+                       pmu_poll();
+               if (req.reply_len != 0)
+                       printk(KERN_ERR "pmac_set_rtc_time: got %d byte reply\n",
+                              req.reply_len);
+               return 1;
+#endif /* CONFIG_ADB_PMU */
+       default:
+               return 0;
+       }
+}
+
+/*
+ * Calibrate the decrementer register using VIA timer 1.
+ * This is used both on powermacs and CHRP machines.
+ */
+int __init
+via_calibrate_decr(void)
+{
+       struct device_node *vias;
+       volatile unsigned char __iomem *via;
+       int count = VIA_TIMER_FREQ_6 / 100;
+       unsigned int dstart, dend;
+
+       vias = find_devices("via-cuda");
+       if (vias == 0)
+               vias = find_devices("via-pmu");
+       if (vias == 0)
+               vias = find_devices("via");
+       if (vias == 0 || vias->n_addrs == 0)
+               return 0;
+       via = ioremap(vias->addrs[0].address, vias->addrs[0].size);
+
+       /* set timer 1 for continuous interrupts */
+       out_8(&via[ACR], (via[ACR] & ~T1MODE) | T1MODE_CONT);
+       /* set the counter to a small value */
+       out_8(&via[T1CH], 2);
+       /* set the latch to `count' */
+       out_8(&via[T1LL], count);
+       out_8(&via[T1LH], count >> 8);
+       /* wait until it hits 0 */
+       while ((in_8(&via[IFR]) & T1_INT) == 0)
+               ;
+       dstart = get_dec();
+       /* clear the interrupt & wait until it hits 0 again */
+       in_8(&via[T1CL]);
+       while ((in_8(&via[IFR]) & T1_INT) == 0)
+               ;
+       dend = get_dec();
+
+       tb_ticks_per_jiffy = (dstart - dend) / (6 * (HZ/100));
+       tb_to_us = mulhwu_scale_factor(dstart - dend, 60000);
+
+       printk(KERN_INFO "via_calibrate_decr: ticks per jiffy = %u (%u ticks)\n",
+              tb_ticks_per_jiffy, dstart - dend);
+
+       iounmap(via);
+       
+       return 1;
+}
+
+#ifdef CONFIG_PM
+/*
+ * Reset the time after a sleep.
+ */
+static int
+time_sleep_notify(struct pmu_sleep_notifier *self, int when)
+{
+       static unsigned long time_diff;
+       unsigned long flags;
+       unsigned long seq;
+
+       switch (when) {
+       case PBOOK_SLEEP_NOW:
+               do {
+                       seq = read_seqbegin_irqsave(&xtime_lock, flags);
+                       time_diff = xtime.tv_sec - pmac_get_rtc_time();
+               } while (read_seqretry_irqrestore(&xtime_lock, seq, flags));
+               break;
+       case PBOOK_WAKE:
+               write_seqlock_irqsave(&xtime_lock, flags);
+               xtime.tv_sec = pmac_get_rtc_time() + time_diff;
+               xtime.tv_nsec = 0;
+               last_rtc_update = xtime.tv_sec;
+               write_sequnlock_irqrestore(&xtime_lock, flags);
+               break;
+       }
+       return PBOOK_SLEEP_OK;
+}
+
+static struct pmu_sleep_notifier time_sleep_notifier = {
+       time_sleep_notify, SLEEP_LEVEL_MISC,
+};
+#endif /* CONFIG_PM */
+
+/*
+ * Query the OF and get the decr frequency.
+ * This was taken from the pmac time_init() when merging the prep/pmac
+ * time functions.
+ */
+void __init
+pmac_calibrate_decr(void)
+{
+       struct device_node *cpu;
+       unsigned int freq, *fp;
+
+#ifdef CONFIG_PM
+       pmu_register_sleep_notifier(&time_sleep_notifier);
+#endif /* CONFIG_PM */
+
+       /* We assume MacRISC2 machines have correct device-tree
+        * calibration. That's better since the VIA itself seems
+        * to be slightly off. --BenH
+        */
+       if (!machine_is_compatible("MacRISC2") &&
+           !machine_is_compatible("MacRISC3") &&
+           !machine_is_compatible("MacRISC4"))
+               if (via_calibrate_decr())
+                       return;
+
+       /* Special case: QuickSilver G4s seem to have a badly calibrated
+        * timebase-frequency in OF, VIA is much better on these. We should
+        * probably implement calibration based on the KL timer on these
+        * machines anyway... -BenH
+        */
+       if (machine_is_compatible("PowerMac3,5"))
+               if (via_calibrate_decr())
+                       return;
+       /*
+        * The cpu node should have a timebase-frequency property
+        * to tell us the rate at which the decrementer counts.
+        */
+       cpu = find_type_devices("cpu");
+       if (cpu == 0)
+               panic("can't find cpu node in time_init");
+       fp = (unsigned int *) get_property(cpu, "timebase-frequency", NULL);
+       if (fp == 0)
+               panic("can't get cpu timebase frequency");
+       freq = *fp;
+       printk("time_init: decrementer frequency = %u.%.6u MHz\n",
+              freq/1000000, freq%1000000);
+       tb_ticks_per_jiffy = freq / HZ;
+       tb_to_us = mulhwu_scale_factor(freq, 1000000);
+}
diff --git a/arch/powerpc/platforms/prep/Kconfig b/arch/powerpc/platforms/prep/Kconfig
new file mode 100644 (file)
index 0000000..673ac47
--- /dev/null
@@ -0,0 +1,22 @@
+
+config PREP_RESIDUAL
+       bool "Support for PReP Residual Data"
+       depends on PPC_PREP
+       help
+         Some PReP systems have residual data passed to the kernel by the
+         firmware.  This allows detection of memory size, devices present and
+         other useful pieces of information.  Sometimes this information is
+         not present or incorrect, in which case it could lead to the machine 
+         behaving incorrectly.  If this happens, either disable PREP_RESIDUAL
+         or pass the 'noresidual' option to the kernel.
+
+         If you are running a PReP system, say Y here, otherwise say N.
+
+config PROC_PREPRESIDUAL
+       bool "Support for reading of PReP Residual Data in /proc"
+       depends on PREP_RESIDUAL && PROC_FS
+       help
+         Enabling this option will create a /proc/residual file which allows
+         you to get at the residual data on PReP systems.  You will need a tool
+         (lsresidual) to parse it.  If you aren't on a PReP system, you don't
+         want this.
diff --git a/arch/powerpc/platforms/pseries/Kconfig b/arch/powerpc/platforms/pseries/Kconfig
new file mode 100644 (file)
index 0000000..7a3b6fc
--- /dev/null
@@ -0,0 +1,47 @@
+
+config PPC_SPLPAR
+       depends on PPC_PSERIES
+       bool "Support for shared-processor logical partitions"
+       default n
+       help
+         Enabling this option will make the kernel run more efficiently
+         on logically-partitioned pSeries systems which use shared
+         processors, that is, which share physical processors between
+         two or more partitions.
+
+config HMT
+       bool "Hardware multithreading"
+       depends on SMP && PPC_PSERIES && BROKEN
+       help
+         This option enables hardware multithreading on RS64 cpus.
+         pSeries systems p620 and p660 have such a cpu type.
+
+config EEH
+       bool "PCI Extended Error Handling (EEH)" if EMBEDDED
+       depends on PPC_PSERIES
+       default y if !EMBEDDED
+
+config PPC_RTAS
+       bool
+       depends on PPC_PSERIES || PPC_BPA
+       default y
+
+config RTAS_PROC
+       bool "Proc interface to RTAS"
+       depends on PPC_RTAS
+       default y
+
+config RTAS_FLASH
+       tristate "Firmware flash interface"
+       depends on PPC64 && RTAS_PROC
+
+config SCANLOG
+       tristate "Scanlog dump interface"
+       depends on RTAS_PROC && PPC_PSERIES
+
+config LPARCFG
+       tristate "LPAR Configuration Data"
+       depends on PPC_PSERIES || PPC_ISERIES
+       help
+       Provide system capacity information via human readable
+       <key word>=<value> pairs through a /proc/ppc64/lparcfg interface.
diff --git a/arch/powerpc/sysdev/Makefile b/arch/powerpc/sysdev/Makefile
new file mode 100644 (file)
index 0000000..26bdcd9
--- /dev/null
@@ -0,0 +1 @@
+obj-$(CONFIG_MPIC)     += mpic.o
diff --git a/arch/powerpc/sysdev/mpic.c b/arch/powerpc/sysdev/mpic.c
new file mode 100644 (file)
index 0000000..c660e7d
--- /dev/null
@@ -0,0 +1,904 @@
+/*
+ *  arch/powerpc/kernel/mpic.c
+ *
+ *  Driver for interrupt controllers following the OpenPIC standard, the
+ *  common implementation beeing IBM's MPIC. This driver also can deal
+ *  with various broken implementations of this HW.
+ *
+ *  Copyright (C) 2004 Benjamin Herrenschmidt, IBM Corp.
+ *
+ *  This file is subject to the terms and conditions of the GNU General Public
+ *  License.  See the file COPYING in the main directory of this archive
+ *  for more details.
+ */
+
+#undef DEBUG
+
+#include <linux/config.h>
+#include <linux/types.h>
+#include <linux/kernel.h>
+#include <linux/init.h>
+#include <linux/irq.h>
+#include <linux/smp.h>
+#include <linux/interrupt.h>
+#include <linux/bootmem.h>
+#include <linux/spinlock.h>
+#include <linux/pci.h>
+
+#include <asm/ptrace.h>
+#include <asm/signal.h>
+#include <asm/io.h>
+#include <asm/pgtable.h>
+#include <asm/irq.h>
+#include <asm/machdep.h>
+#include <asm/mpic.h>
+#include <asm/smp.h>
+
+#ifdef DEBUG
+#define DBG(fmt...) printk(fmt)
+#else
+#define DBG(fmt...)
+#endif
+
+static struct mpic *mpics;
+static struct mpic *mpic_primary;
+static DEFINE_SPINLOCK(mpic_lock);
+
+
+/*
+ * Register accessor functions
+ */
+
+
+static inline u32 _mpic_read(unsigned int be, volatile u32 __iomem *base,
+                           unsigned int reg)
+{
+       if (be)
+               return in_be32(base + (reg >> 2));
+       else
+               return in_le32(base + (reg >> 2));
+}
+
+static inline void _mpic_write(unsigned int be, volatile u32 __iomem *base,
+                             unsigned int reg, u32 value)
+{
+       if (be)
+               out_be32(base + (reg >> 2), value);
+       else
+               out_le32(base + (reg >> 2), value);
+}
+
+static inline u32 _mpic_ipi_read(struct mpic *mpic, unsigned int ipi)
+{
+       unsigned int be = (mpic->flags & MPIC_BIG_ENDIAN) != 0;
+       unsigned int offset = MPIC_GREG_IPI_VECTOR_PRI_0 + (ipi * 0x10);
+
+       if (mpic->flags & MPIC_BROKEN_IPI)
+               be = !be;
+       return _mpic_read(be, mpic->gregs, offset);
+}
+
+static inline void _mpic_ipi_write(struct mpic *mpic, unsigned int ipi, u32 value)
+{
+       unsigned int offset = MPIC_GREG_IPI_VECTOR_PRI_0 + (ipi * 0x10);
+
+       _mpic_write(mpic->flags & MPIC_BIG_ENDIAN, mpic->gregs, offset, value);
+}
+
+static inline u32 _mpic_cpu_read(struct mpic *mpic, unsigned int reg)
+{
+       unsigned int cpu = 0;
+
+       if (mpic->flags & MPIC_PRIMARY)
+               cpu = hard_smp_processor_id();
+
+       return _mpic_read(mpic->flags & MPIC_BIG_ENDIAN, mpic->cpuregs[cpu], reg);
+}
+
+static inline void _mpic_cpu_write(struct mpic *mpic, unsigned int reg, u32 value)
+{
+       unsigned int cpu = 0;
+
+       if (mpic->flags & MPIC_PRIMARY)
+               cpu = hard_smp_processor_id();
+
+       _mpic_write(mpic->flags & MPIC_BIG_ENDIAN, mpic->cpuregs[cpu], reg, value);
+}
+
+static inline u32 _mpic_irq_read(struct mpic *mpic, unsigned int src_no, unsigned int reg)
+{
+       unsigned int    isu = src_no >> mpic->isu_shift;
+       unsigned int    idx = src_no & mpic->isu_mask;
+
+       return _mpic_read(mpic->flags & MPIC_BIG_ENDIAN, mpic->isus[isu],
+                         reg + (idx * MPIC_IRQ_STRIDE));
+}
+
+static inline void _mpic_irq_write(struct mpic *mpic, unsigned int src_no,
+                                  unsigned int reg, u32 value)
+{
+       unsigned int    isu = src_no >> mpic->isu_shift;
+       unsigned int    idx = src_no & mpic->isu_mask;
+
+       _mpic_write(mpic->flags & MPIC_BIG_ENDIAN, mpic->isus[isu],
+                   reg + (idx * MPIC_IRQ_STRIDE), value);
+}
+
+#define mpic_read(b,r)         _mpic_read(mpic->flags & MPIC_BIG_ENDIAN,(b),(r))
+#define mpic_write(b,r,v)      _mpic_write(mpic->flags & MPIC_BIG_ENDIAN,(b),(r),(v))
+#define mpic_ipi_read(i)       _mpic_ipi_read(mpic,(i))
+#define mpic_ipi_write(i,v)    _mpic_ipi_write(mpic,(i),(v))
+#define mpic_cpu_read(i)       _mpic_cpu_read(mpic,(i))
+#define mpic_cpu_write(i,v)    _mpic_cpu_write(mpic,(i),(v))
+#define mpic_irq_read(s,r)     _mpic_irq_read(mpic,(s),(r))
+#define mpic_irq_write(s,r,v)  _mpic_irq_write(mpic,(s),(r),(v))
+
+
+/*
+ * Low level utility functions
+ */
+
+
+
+/* Check if we have one of those nice broken MPICs with a flipped endian on
+ * reads from IPI registers
+ */
+static void __init mpic_test_broken_ipi(struct mpic *mpic)
+{
+       u32 r;
+
+       mpic_write(mpic->gregs, MPIC_GREG_IPI_VECTOR_PRI_0, MPIC_VECPRI_MASK);
+       r = mpic_read(mpic->gregs, MPIC_GREG_IPI_VECTOR_PRI_0);
+
+       if (r == le32_to_cpu(MPIC_VECPRI_MASK)) {
+               printk(KERN_INFO "mpic: Detected reversed IPI registers\n");
+               mpic->flags |= MPIC_BROKEN_IPI;
+       }
+}
+
+#ifdef CONFIG_MPIC_BROKEN_U3
+
+/* Test if an interrupt is sourced from HyperTransport (used on broken U3s)
+ * to force the edge setting on the MPIC and do the ack workaround.
+ */
+static inline int mpic_is_ht_interrupt(struct mpic *mpic, unsigned int source_no)
+{
+       if (source_no >= 128 || !mpic->fixups)
+               return 0;
+       return mpic->fixups[source_no].base != NULL;
+}
+
+static inline void mpic_apic_end_irq(struct mpic *mpic, unsigned int source_no)
+{
+       struct mpic_irq_fixup *fixup = &mpic->fixups[source_no];
+       u32 tmp;
+
+       spin_lock(&mpic->fixup_lock);
+       writeb(0x11 + 2 * fixup->irq, fixup->base);
+       tmp = readl(fixup->base + 2);
+       writel(tmp | 0x80000000ul, fixup->base + 2);
+       /* config writes shouldn't be posted but let's be safe ... */
+       (void)readl(fixup->base + 2);
+       spin_unlock(&mpic->fixup_lock);
+}
+
+
+static void __init mpic_amd8111_read_irq(struct mpic *mpic, u8 __iomem *devbase)
+{
+       int i, irq;
+       u32 tmp;
+
+       printk(KERN_INFO "mpic:    - Workarounds on AMD 8111 @ %p\n", devbase);
+
+       for (i=0; i < 24; i++) {
+               writeb(0x10 + 2*i, devbase + 0xf2);
+               tmp = readl(devbase + 0xf4);
+               if ((tmp & 0x1) || !(tmp & 0x20))
+                       continue;
+               irq = (tmp >> 16) & 0xff;
+               mpic->fixups[irq].irq = i;
+               mpic->fixups[irq].base = devbase + 0xf2;
+       }
+}
+static void __init mpic_amd8131_read_irq(struct mpic *mpic, u8 __iomem *devbase)
+{
+       int i, irq;
+       u32 tmp;
+
+       printk(KERN_INFO "mpic:    - Workarounds on AMD 8131 @ %p\n", devbase);
+
+       for (i=0; i < 4; i++) {
+               writeb(0x10 + 2*i, devbase + 0xba);
+               tmp = readl(devbase + 0xbc);
+               if ((tmp & 0x1) || !(tmp & 0x20))
+                       continue;
+               irq = (tmp >> 16) & 0xff;
+               mpic->fixups[irq].irq = i;
+               mpic->fixups[irq].base = devbase + 0xba;
+       }
+}
+static void __init mpic_scan_ioapics(struct mpic *mpic)
+{
+       unsigned int devfn;
+       u8 __iomem *cfgspace;
+
+       printk(KERN_INFO "mpic: Setting up IO-APICs workarounds for U3\n");
+
+       /* Allocate fixups array */
+       mpic->fixups = alloc_bootmem(128 * sizeof(struct mpic_irq_fixup));
+       BUG_ON(mpic->fixups == NULL);
+       memset(mpic->fixups, 0, 128 * sizeof(struct mpic_irq_fixup));
+
+       /* Init spinlock */
+       spin_lock_init(&mpic->fixup_lock);
+
+       /* Map u3 config space. We assume all IO-APICs are on the primary bus
+        * and slot will never be above "0xf" so we only need to map 32k
+        */
+       cfgspace = (unsigned char __iomem *)ioremap(0xf2000000, 0x8000);
+       BUG_ON(cfgspace == NULL);
+
+       /* Now we scan all slots. We do a very quick scan, we read the header type,
+        * vendor ID and device ID only, that's plenty enough
+        */
+       for (devfn = 0; devfn < PCI_DEVFN(0x10,0); devfn ++) {
+               u8 __iomem *devbase = cfgspace + (devfn << 8);
+               u8 hdr_type = readb(devbase + PCI_HEADER_TYPE);
+               u32 l = readl(devbase + PCI_VENDOR_ID);
+               u16 vendor_id, device_id;
+               int multifunc = 0;
+
+               DBG("devfn %x, l: %x\n", devfn, l);
+
+               /* If no device, skip */
+               if (l == 0xffffffff || l == 0x00000000 ||
+                   l == 0x0000ffff || l == 0xffff0000)
+                       goto next;
+
+               /* Check if it's a multifunction device (only really used
+                * to function 0 though
+                */
+               multifunc = !!(hdr_type & 0x80);
+               vendor_id = l & 0xffff;
+               device_id = (l >> 16) & 0xffff;
+
+               /* If a known device, go to fixup setup code */
+               if (vendor_id == PCI_VENDOR_ID_AMD && device_id == 0x7460)
+                       mpic_amd8111_read_irq(mpic, devbase);
+               if (vendor_id == PCI_VENDOR_ID_AMD && device_id == 0x7450)
+                       mpic_amd8131_read_irq(mpic, devbase);
+       next:
+               /* next device, if function 0 */
+               if ((PCI_FUNC(devfn) == 0) && !multifunc)
+                       devfn += 7;
+       }
+}
+
+#endif /* CONFIG_MPIC_BROKEN_U3 */
+
+
+/* Find an mpic associated with a given linux interrupt */
+static struct mpic *mpic_find(unsigned int irq, unsigned int *is_ipi)
+{
+       struct mpic *mpic = mpics;
+
+       while(mpic) {
+               /* search IPIs first since they may override the main interrupts */
+               if (irq >= mpic->ipi_offset && irq < (mpic->ipi_offset + 4)) {
+                       if (is_ipi)
+                               *is_ipi = 1;
+                       return mpic;
+               }
+               if (irq >= mpic->irq_offset &&
+                   irq < (mpic->irq_offset + mpic->irq_count)) {
+                       if (is_ipi)
+                               *is_ipi = 0;
+                       return mpic;
+               }
+               mpic = mpic -> next;
+       }
+       return NULL;
+}
+
+/* Convert a cpu mask from logical to physical cpu numbers. */
+static inline u32 mpic_physmask(u32 cpumask)
+{
+       int i;
+       u32 mask = 0;
+
+       for (i = 0; i < NR_CPUS; ++i, cpumask >>= 1)
+               mask |= (cpumask & 1) << get_hard_smp_processor_id(i);
+       return mask;
+}
+
+#ifdef CONFIG_SMP
+/* Get the mpic structure from the IPI number */
+static inline struct mpic * mpic_from_ipi(unsigned int ipi)
+{
+       return container_of(irq_desc[ipi].handler, struct mpic, hc_ipi);
+}
+#endif
+
+/* Get the mpic structure from the irq number */
+static inline struct mpic * mpic_from_irq(unsigned int irq)
+{
+       return container_of(irq_desc[irq].handler, struct mpic, hc_irq);
+}
+
+/* Send an EOI */
+static inline void mpic_eoi(struct mpic *mpic)
+{
+       mpic_cpu_write(MPIC_CPU_EOI, 0);
+       (void)mpic_cpu_read(MPIC_CPU_WHOAMI);
+}
+
+#ifdef CONFIG_SMP
+static irqreturn_t mpic_ipi_action(int irq, void *dev_id, struct pt_regs *regs)
+{
+       struct mpic *mpic = dev_id;
+
+       smp_message_recv(irq - mpic->ipi_offset, regs);
+       return IRQ_HANDLED;
+}
+#endif /* CONFIG_SMP */
+
+/*
+ * Linux descriptor level callbacks
+ */
+
+
+static void mpic_enable_irq(unsigned int irq)
+{
+       unsigned int loops = 100000;
+       struct mpic *mpic = mpic_from_irq(irq);
+       unsigned int src = irq - mpic->irq_offset;
+
+       DBG("%s: enable_irq: %d (src %d)\n", mpic->name, irq, src);
+
+       mpic_irq_write(src, MPIC_IRQ_VECTOR_PRI,
+                      mpic_irq_read(src, MPIC_IRQ_VECTOR_PRI) & ~MPIC_VECPRI_MASK);
+
+       /* make sure mask gets to controller before we return to user */
+       do {
+               if (!loops--) {
+                       printk(KERN_ERR "mpic_enable_irq timeout\n");
+                       break;
+               }
+       } while(mpic_irq_read(src, MPIC_IRQ_VECTOR_PRI) & MPIC_VECPRI_MASK);    
+}
+
+static void mpic_disable_irq(unsigned int irq)
+{
+       unsigned int loops = 100000;
+       struct mpic *mpic = mpic_from_irq(irq);
+       unsigned int src = irq - mpic->irq_offset;
+
+       DBG("%s: disable_irq: %d (src %d)\n", mpic->name, irq, src);
+
+       mpic_irq_write(src, MPIC_IRQ_VECTOR_PRI,
+                      mpic_irq_read(src, MPIC_IRQ_VECTOR_PRI) | MPIC_VECPRI_MASK);
+
+       /* make sure mask gets to controller before we return to user */
+       do {
+               if (!loops--) {
+                       printk(KERN_ERR "mpic_enable_irq timeout\n");
+                       break;
+               }
+       } while(!(mpic_irq_read(src, MPIC_IRQ_VECTOR_PRI) & MPIC_VECPRI_MASK));
+}
+
+static void mpic_end_irq(unsigned int irq)
+{
+       struct mpic *mpic = mpic_from_irq(irq);
+
+       DBG("%s: end_irq: %d\n", mpic->name, irq);
+
+       /* We always EOI on end_irq() even for edge interrupts since that
+        * should only lower the priority, the MPIC should have properly
+        * latched another edge interrupt coming in anyway
+        */
+
+#ifdef CONFIG_MPIC_BROKEN_U3
+       if (mpic->flags & MPIC_BROKEN_U3) {
+               unsigned int src = irq - mpic->irq_offset;
+               if (mpic_is_ht_interrupt(mpic, src))
+                       mpic_apic_end_irq(mpic, src);
+       }
+#endif /* CONFIG_MPIC_BROKEN_U3 */
+
+       mpic_eoi(mpic);
+}
+
+#ifdef CONFIG_SMP
+
+static void mpic_enable_ipi(unsigned int irq)
+{
+       struct mpic *mpic = mpic_from_ipi(irq);
+       unsigned int src = irq - mpic->ipi_offset;
+
+       DBG("%s: enable_ipi: %d (ipi %d)\n", mpic->name, irq, src);
+       mpic_ipi_write(src, mpic_ipi_read(src) & ~MPIC_VECPRI_MASK);
+}
+
+static void mpic_disable_ipi(unsigned int irq)
+{
+       /* NEVER disable an IPI... that's just plain wrong! */
+}
+
+static void mpic_end_ipi(unsigned int irq)
+{
+       struct mpic *mpic = mpic_from_ipi(irq);
+
+       /*
+        * IPIs are marked IRQ_PER_CPU. This has the side effect of
+        * preventing the IRQ_PENDING/IRQ_INPROGRESS logic from
+        * applying to them. We EOI them late to avoid re-entering.
+        * We mark IPI's with SA_INTERRUPT as they must run with
+        * irqs disabled.
+        */
+       mpic_eoi(mpic);
+}
+
+#endif /* CONFIG_SMP */
+
+static void mpic_set_affinity(unsigned int irq, cpumask_t cpumask)
+{
+       struct mpic *mpic = mpic_from_irq(irq);
+
+       cpumask_t tmp;
+
+       cpus_and(tmp, cpumask, cpu_online_map);
+
+       mpic_irq_write(irq - mpic->irq_offset, MPIC_IRQ_DESTINATION,
+                      mpic_physmask(cpus_addr(tmp)[0]));       
+}
+
+
+/*
+ * Exported functions
+ */
+
+
+struct mpic * __init mpic_alloc(unsigned long phys_addr,
+                               unsigned int flags,
+                               unsigned int isu_size,
+                               unsigned int irq_offset,
+                               unsigned int irq_count,
+                               unsigned int ipi_offset,
+                               unsigned char *senses,
+                               unsigned int senses_count,
+                               const char *name)
+{
+       struct mpic     *mpic;
+       u32             reg;
+       const char      *vers;
+       int             i;
+
+       mpic = alloc_bootmem(sizeof(struct mpic));
+       if (mpic == NULL)
+               return NULL;
+       
+
+       memset(mpic, 0, sizeof(struct mpic));
+       mpic->name = name;
+
+       mpic->hc_irq.typename = name;
+       mpic->hc_irq.enable = mpic_enable_irq;
+       mpic->hc_irq.disable = mpic_disable_irq;
+       mpic->hc_irq.end = mpic_end_irq;
+       if (flags & MPIC_PRIMARY)
+               mpic->hc_irq.set_affinity = mpic_set_affinity;
+#ifdef CONFIG_SMP
+       mpic->hc_ipi.typename = name;
+       mpic->hc_ipi.enable = mpic_enable_ipi;
+       mpic->hc_ipi.disable = mpic_disable_ipi;
+       mpic->hc_ipi.end = mpic_end_ipi;
+#endif /* CONFIG_SMP */
+
+       mpic->flags = flags;
+       mpic->isu_size = isu_size;
+       mpic->irq_offset = irq_offset;
+       mpic->irq_count = irq_count;
+       mpic->ipi_offset = ipi_offset;
+       mpic->num_sources = 0; /* so far */
+       mpic->senses = senses;
+       mpic->senses_count = senses_count;
+
+       /* Map the global registers */
+       mpic->gregs = ioremap(phys_addr + MPIC_GREG_BASE, 0x1000);
+       mpic->tmregs = mpic->gregs + (MPIC_TIMER_BASE >> 2);
+       BUG_ON(mpic->gregs == NULL);
+
+       /* Reset */
+       if (flags & MPIC_WANTS_RESET) {
+               mpic_write(mpic->gregs, MPIC_GREG_GLOBAL_CONF_0,
+                          mpic_read(mpic->gregs, MPIC_GREG_GLOBAL_CONF_0)
+                          | MPIC_GREG_GCONF_RESET);
+               while( mpic_read(mpic->gregs, MPIC_GREG_GLOBAL_CONF_0)
+                      & MPIC_GREG_GCONF_RESET)
+                       mb();
+       }
+
+       /* Read feature register, calculate num CPUs and, for non-ISU
+        * MPICs, num sources as well. On ISU MPICs, sources are counted
+        * as ISUs are added
+        */
+       reg = mpic_read(mpic->gregs, MPIC_GREG_FEATURE_0);
+       mpic->num_cpus = ((reg & MPIC_GREG_FEATURE_LAST_CPU_MASK)
+                         >> MPIC_GREG_FEATURE_LAST_CPU_SHIFT) + 1;
+       if (isu_size == 0)
+               mpic->num_sources = ((reg & MPIC_GREG_FEATURE_LAST_SRC_MASK)
+                                    >> MPIC_GREG_FEATURE_LAST_SRC_SHIFT) + 1;
+
+       /* Map the per-CPU registers */
+       for (i = 0; i < mpic->num_cpus; i++) {
+               mpic->cpuregs[i] = ioremap(phys_addr + MPIC_CPU_BASE +
+                                          i * MPIC_CPU_STRIDE, 0x1000);
+               BUG_ON(mpic->cpuregs[i] == NULL);
+       }
+
+       /* Initialize main ISU if none provided */
+       if (mpic->isu_size == 0) {
+               mpic->isu_size = mpic->num_sources;
+               mpic->isus[0] = ioremap(phys_addr + MPIC_IRQ_BASE,
+                                       MPIC_IRQ_STRIDE * mpic->isu_size);
+               BUG_ON(mpic->isus[0] == NULL);
+       }
+       mpic->isu_shift = 1 + __ilog2(mpic->isu_size - 1);
+       mpic->isu_mask = (1 << mpic->isu_shift) - 1;
+
+       /* Display version */
+       switch (reg & MPIC_GREG_FEATURE_VERSION_MASK) {
+       case 1:
+               vers = "1.0";
+               break;
+       case 2:
+               vers = "1.2";
+               break;
+       case 3:
+               vers = "1.3";
+               break;
+       default:
+               vers = "<unknown>";
+               break;
+       }
+       printk(KERN_INFO "mpic: Setting up MPIC \"%s\" version %s at %lx, max %d CPUs\n",
+              name, vers, phys_addr, mpic->num_cpus);
+       printk(KERN_INFO "mpic: ISU size: %d, shift: %d, mask: %x\n", mpic->isu_size,
+              mpic->isu_shift, mpic->isu_mask);
+
+       mpic->next = mpics;
+       mpics = mpic;
+
+       if (flags & MPIC_PRIMARY)
+               mpic_primary = mpic;
+
+       return mpic;
+}
+
+void __init mpic_assign_isu(struct mpic *mpic, unsigned int isu_num,
+                           unsigned long phys_addr)
+{
+       unsigned int isu_first = isu_num * mpic->isu_size;
+
+       BUG_ON(isu_num >= MPIC_MAX_ISU);
+
+       mpic->isus[isu_num] = ioremap(phys_addr, MPIC_IRQ_STRIDE * mpic->isu_size);
+       if ((isu_first + mpic->isu_size) > mpic->num_sources)
+               mpic->num_sources = isu_first + mpic->isu_size;
+}
+
+void __init mpic_setup_cascade(unsigned int irq, mpic_cascade_t handler,
+                              void *data)
+{
+       struct mpic *mpic = mpic_find(irq, NULL);
+       unsigned long flags;
+
+       /* Synchronization here is a bit dodgy, so don't try to replace cascade
+        * interrupts on the fly too often ... but normally it's set up at boot.
+        */
+       spin_lock_irqsave(&mpic_lock, flags);
+       if (mpic->cascade)             
+               mpic_disable_irq(mpic->cascade_vec + mpic->irq_offset);
+       mpic->cascade = NULL;
+       wmb();
+       mpic->cascade_vec = irq - mpic->irq_offset;
+       mpic->cascade_data = data;
+       wmb();
+       mpic->cascade = handler;
+       mpic_enable_irq(irq);
+       spin_unlock_irqrestore(&mpic_lock, flags);
+}
+
+void __init mpic_init(struct mpic *mpic)
+{
+       int i;
+
+       BUG_ON(mpic->num_sources == 0);
+
+       printk(KERN_INFO "mpic: Initializing for %d sources\n", mpic->num_sources);
+
+       /* Set current processor priority to max */
+       mpic_cpu_write(MPIC_CPU_CURRENT_TASK_PRI, 0xf);
+
+       /* Initialize timers: just disable them all */
+       for (i = 0; i < 4; i++) {
+               mpic_write(mpic->tmregs,
+                          i * MPIC_TIMER_STRIDE + MPIC_TIMER_DESTINATION, 0);
+               mpic_write(mpic->tmregs,
+                          i * MPIC_TIMER_STRIDE + MPIC_TIMER_VECTOR_PRI,
+                          MPIC_VECPRI_MASK |
+                          (MPIC_VEC_TIMER_0 + i));
+       }
+
+       /* Initialize IPIs to our reserved vectors and mark them disabled for now */
+       mpic_test_broken_ipi(mpic);
+       for (i = 0; i < 4; i++) {
+               mpic_ipi_write(i,
+                              MPIC_VECPRI_MASK |
+                              (10 << MPIC_VECPRI_PRIORITY_SHIFT) |
+                              (MPIC_VEC_IPI_0 + i));
+#ifdef CONFIG_SMP
+               if (!(mpic->flags & MPIC_PRIMARY))
+                       continue;
+               irq_desc[mpic->ipi_offset+i].status |= IRQ_PER_CPU;
+               irq_desc[mpic->ipi_offset+i].handler = &mpic->hc_ipi;
+               
+#endif /* CONFIG_SMP */
+       }
+
+       /* Initialize interrupt sources */
+       if (mpic->irq_count == 0)
+               mpic->irq_count = mpic->num_sources;
+
+#ifdef CONFIG_MPIC_BROKEN_U3
+       /* Do the ioapic fixups on U3 broken mpic */
+       DBG("MPIC flags: %x\n", mpic->flags);
+       if ((mpic->flags & MPIC_BROKEN_U3) && (mpic->flags & MPIC_PRIMARY))
+               mpic_scan_ioapics(mpic);
+#endif /* CONFIG_MPIC_BROKEN_U3 */
+
+       for (i = 0; i < mpic->num_sources; i++) {
+               /* start with vector = source number, and masked */
+               u32 vecpri = MPIC_VECPRI_MASK | i | (8 << MPIC_VECPRI_PRIORITY_SHIFT);
+               int level = 0;
+               
+               /* if it's an IPI, we skip it */
+               if ((mpic->irq_offset + i) >= (mpic->ipi_offset + i) &&
+                   (mpic->irq_offset + i) <  (mpic->ipi_offset + i + 4))
+                       continue;
+
+               /* do senses munging */
+               if (mpic->senses && i < mpic->senses_count) {
+                       if (mpic->senses[i] & IRQ_SENSE_LEVEL)
+                               vecpri |= MPIC_VECPRI_SENSE_LEVEL;
+                       if (mpic->senses[i] & IRQ_POLARITY_POSITIVE)
+                               vecpri |= MPIC_VECPRI_POLARITY_POSITIVE;
+               } else
+                       vecpri |= MPIC_VECPRI_SENSE_LEVEL;
+
+               /* remember if it was a level interrupts */
+               level = (vecpri & MPIC_VECPRI_SENSE_LEVEL);
+
+               /* deal with broken U3 */
+               if (mpic->flags & MPIC_BROKEN_U3) {
+#ifdef CONFIG_MPIC_BROKEN_U3
+                       if (mpic_is_ht_interrupt(mpic, i)) {
+                               vecpri &= ~(MPIC_VECPRI_SENSE_MASK |
+                                           MPIC_VECPRI_POLARITY_MASK);
+                               vecpri |= MPIC_VECPRI_POLARITY_POSITIVE;
+                       }
+#else
+                       printk(KERN_ERR "mpic: BROKEN_U3 set, but CONFIG doesn't match\n");
+#endif
+               }
+
+               DBG("setup source %d, vecpri: %08x, level: %d\n", i, vecpri,
+                   (level != 0));
+
+               /* init hw */
+               mpic_irq_write(i, MPIC_IRQ_VECTOR_PRI, vecpri);
+               mpic_irq_write(i, MPIC_IRQ_DESTINATION,
+                              1 << hard_smp_processor_id());
+
+               /* init linux descriptors */
+               if (i < mpic->irq_count) {
+                       irq_desc[mpic->irq_offset+i].status = level ? IRQ_LEVEL : 0;
+                       irq_desc[mpic->irq_offset+i].handler = &mpic->hc_irq;
+               }
+       }
+       
+       /* Init spurrious vector */
+       mpic_write(mpic->gregs, MPIC_GREG_SPURIOUS, MPIC_VEC_SPURRIOUS);
+
+       /* Disable 8259 passthrough */
+       mpic_write(mpic->gregs, MPIC_GREG_GLOBAL_CONF_0,
+                  mpic_read(mpic->gregs, MPIC_GREG_GLOBAL_CONF_0)
+                  | MPIC_GREG_GCONF_8259_PTHROU_DIS);
+
+       /* Set current processor priority to 0 */
+       mpic_cpu_write(MPIC_CPU_CURRENT_TASK_PRI, 0);
+}
+
+
+
+void mpic_irq_set_priority(unsigned int irq, unsigned int pri)
+{
+       int is_ipi;
+       struct mpic *mpic = mpic_find(irq, &is_ipi);
+       unsigned long flags;
+       u32 reg;
+
+       spin_lock_irqsave(&mpic_lock, flags);
+       if (is_ipi) {
+               reg = mpic_ipi_read(irq - mpic->ipi_offset) & MPIC_VECPRI_PRIORITY_MASK;
+               mpic_ipi_write(irq - mpic->ipi_offset,
+                              reg | (pri << MPIC_VECPRI_PRIORITY_SHIFT));
+       } else {
+               reg = mpic_irq_read(irq - mpic->irq_offset, MPIC_IRQ_VECTOR_PRI)
+                       & MPIC_VECPRI_PRIORITY_MASK;
+               mpic_irq_write(irq - mpic->irq_offset, MPIC_IRQ_VECTOR_PRI,
+                              reg | (pri << MPIC_VECPRI_PRIORITY_SHIFT));
+       }
+       spin_unlock_irqrestore(&mpic_lock, flags);
+}
+
+unsigned int mpic_irq_get_priority(unsigned int irq)
+{
+       int is_ipi;
+       struct mpic *mpic = mpic_find(irq, &is_ipi);
+       unsigned long flags;
+       u32 reg;
+
+       spin_lock_irqsave(&mpic_lock, flags);
+       if (is_ipi)
+               reg = mpic_ipi_read(irq - mpic->ipi_offset);
+       else
+               reg = mpic_irq_read(irq - mpic->irq_offset, MPIC_IRQ_VECTOR_PRI);
+       spin_unlock_irqrestore(&mpic_lock, flags);
+       return (reg & MPIC_VECPRI_PRIORITY_MASK) >> MPIC_VECPRI_PRIORITY_SHIFT;
+}
+
+void mpic_setup_this_cpu(void)
+{
+#ifdef CONFIG_SMP
+       struct mpic *mpic = mpic_primary;
+       unsigned long flags;
+       u32 msk = 1 << hard_smp_processor_id();
+       unsigned int i;
+
+       BUG_ON(mpic == NULL);
+
+       DBG("%s: setup_this_cpu(%d)\n", mpic->name, hard_smp_processor_id());
+
+       spin_lock_irqsave(&mpic_lock, flags);
+
+       /* let the mpic know we want intrs. default affinity is 0xffffffff
+        * until changed via /proc. That's how it's done on x86. If we want
+        * it differently, then we should make sure we also change the default
+        * values of irq_affinity in irq.c.
+        */
+       if (distribute_irqs) {
+               for (i = 0; i < mpic->num_sources ; i++)
+                       mpic_irq_write(i, MPIC_IRQ_DESTINATION,
+                               mpic_irq_read(i, MPIC_IRQ_DESTINATION) | msk);
+       }
+
+       /* Set current processor priority to 0 */
+       mpic_cpu_write(MPIC_CPU_CURRENT_TASK_PRI, 0);
+
+       spin_unlock_irqrestore(&mpic_lock, flags);
+#endif /* CONFIG_SMP */
+}
+
+int mpic_cpu_get_priority(void)
+{
+       struct mpic *mpic = mpic_primary;
+
+       return mpic_cpu_read(MPIC_CPU_CURRENT_TASK_PRI);
+}
+
+void mpic_cpu_set_priority(int prio)
+{
+       struct mpic *mpic = mpic_primary;
+
+       prio &= MPIC_CPU_TASKPRI_MASK;
+       mpic_cpu_write(MPIC_CPU_CURRENT_TASK_PRI, prio);
+}
+
+/*
+ * XXX: someone who knows mpic should check this.
+ * do we need to eoi the ipi including for kexec cpu here (see xics comments)?
+ * or can we reset the mpic in the new kernel?
+ */
+void mpic_teardown_this_cpu(int secondary)
+{
+       struct mpic *mpic = mpic_primary;
+       unsigned long flags;
+       u32 msk = 1 << hard_smp_processor_id();
+       unsigned int i;
+
+       BUG_ON(mpic == NULL);
+
+       DBG("%s: teardown_this_cpu(%d)\n", mpic->name, hard_smp_processor_id());
+       spin_lock_irqsave(&mpic_lock, flags);
+
+       /* let the mpic know we don't want intrs.  */
+       for (i = 0; i < mpic->num_sources ; i++)
+               mpic_irq_write(i, MPIC_IRQ_DESTINATION,
+                       mpic_irq_read(i, MPIC_IRQ_DESTINATION) & ~msk);
+
+       /* Set current processor priority to max */
+       mpic_cpu_write(MPIC_CPU_CURRENT_TASK_PRI, 0xf);
+
+       spin_unlock_irqrestore(&mpic_lock, flags);
+}
+
+
+void mpic_send_ipi(unsigned int ipi_no, unsigned int cpu_mask)
+{
+       struct mpic *mpic = mpic_primary;
+
+       BUG_ON(mpic == NULL);
+
+       DBG("%s: send_ipi(ipi_no: %d)\n", mpic->name, ipi_no);
+
+       mpic_cpu_write(MPIC_CPU_IPI_DISPATCH_0 + ipi_no * 0x10,
+                      mpic_physmask(cpu_mask & cpus_addr(cpu_online_map)[0]));
+}
+
+int mpic_get_one_irq(struct mpic *mpic, struct pt_regs *regs)
+{
+       u32 irq;
+
+       irq = mpic_cpu_read(MPIC_CPU_INTACK) & MPIC_VECPRI_VECTOR_MASK;
+       DBG("%s: get_one_irq(): %d\n", mpic->name, irq);
+
+       if (mpic->cascade && irq == mpic->cascade_vec) {
+               DBG("%s: cascading ...\n", mpic->name);
+               irq = mpic->cascade(regs, mpic->cascade_data);
+               mpic_eoi(mpic);
+               return irq;
+       }
+       if (unlikely(irq == MPIC_VEC_SPURRIOUS))
+               return -1;
+       if (irq < MPIC_VEC_IPI_0) 
+               return irq + mpic->irq_offset;
+               DBG("%s: ipi %d !\n", mpic->name, irq - MPIC_VEC_IPI_0);
+       return irq - MPIC_VEC_IPI_0 + mpic->ipi_offset;
+}
+
+int mpic_get_irq(struct pt_regs *regs)
+{
+       struct mpic *mpic = mpic_primary;
+
+       BUG_ON(mpic == NULL);
+
+       return mpic_get_one_irq(mpic, regs);
+}
+
+
+#ifdef CONFIG_SMP
+void mpic_request_ipis(void)
+{
+       struct mpic *mpic = mpic_primary;
+
+       BUG_ON(mpic == NULL);
+       
+       printk("requesting IPIs ... \n");
+
+       /* IPIs are marked SA_INTERRUPT as they must run with irqs disabled */
+       request_irq(mpic->ipi_offset+0, mpic_ipi_action, SA_INTERRUPT,
+                   "IPI0 (call function)", mpic);
+       request_irq(mpic->ipi_offset+1, mpic_ipi_action, SA_INTERRUPT,
+                  "IPI1 (reschedule)", mpic);
+       request_irq(mpic->ipi_offset+2, mpic_ipi_action, SA_INTERRUPT,
+                  "IPI2 (unused)", mpic);
+       request_irq(mpic->ipi_offset+3, mpic_ipi_action, SA_INTERRUPT,
+                  "IPI3 (debugger break)", mpic);
+
+       printk("IPIs requested... \n");
+}
+#endif /* CONFIG_SMP */
index ce166e3de53b04b8cbf0ca6a07eff8aebf10604a..0649540bc7d967b4967a00426b5fe50415f178ed 100644 (file)
@@ -1,6 +1,7 @@
 #
 # Makefile for the linux kernel.
 #
+ifneq ($(CONFIG_PPC_MERGE),y)
 
 extra-$(CONFIG_PPC_STD_MMU)    := head.o
 extra-$(CONFIG_40x)            := head_4xx.o
@@ -37,3 +38,23 @@ endif
 
 # These are here while we do the architecture merge
 vecemu-y                       += ../../powerpc/kernel/vecemu.o
+
+else
+obj-y                          := entry.o irq.o idle.o time.o misc.o \
+                                       signal.o ptrace.o align.o \
+                                       syscalls.o setup.o \
+                                       cputable.o perfmon.o
+obj-$(CONFIG_6xx)              += l2cr.o cpu_setup_6xx.o
+obj-$(CONFIG_SOFTWARE_SUSPEND) += swsusp.o
+obj-$(CONFIG_POWER4)           += cpu_setup_power4.o
+obj-$(CONFIG_MODULES)          += module.o ppc_ksyms.o
+obj-$(CONFIG_NOT_COHERENT_CACHE)       += dma-mapping.o
+obj-$(CONFIG_PCI)              += pci.o
+obj-$(CONFIG_KGDB)             += ppc-stub.o
+obj-$(CONFIG_SMP)              += smp.o smp-tbsync.o
+obj-$(CONFIG_TAU)              += temp.o
+ifndef CONFIG_E200
+obj-$(CONFIG_FSL_BOOKE)                += perfmon_fsl_booke.o
+endif
+obj-$(CONFIG_KEXEC)            += machine_kexec.o relocate_kernel.o
+endif
index 77fecfbabe887c0f58baaa5b2c43e1a61771c548..1b891b806f3dfa9fd4a7781e6d745f8f9e82b7b7 100644 (file)
@@ -83,6 +83,8 @@ extern void pmac_init(unsigned long r3, unsigned long r4,
                unsigned long r5, unsigned long r6, unsigned long r7);
 extern void chrp_init(unsigned long r3, unsigned long r4,
                unsigned long r5, unsigned long r6, unsigned long r7);
+
+dev_t boot_dev;
 #endif /* CONFIG_PPC_MULTIPLATFORM */
 
 #ifdef CONFIG_MAGIC_SYSRQ
@@ -405,11 +407,13 @@ platform_init(unsigned long r3, unsigned long r4, unsigned long r5,
                        _machine = _MACH_prep;
        }
 
+#ifdef CONFIG_PPC_PREP
        /* not much more to do here, if prep */
        if (_machine == _MACH_prep) {
                prep_init(r3, r4, r5, r6, r7);
                return;
        }
+#endif
 
        /* prom_init has already been called from __start */
        if (boot_infos)
@@ -480,12 +484,16 @@ platform_init(unsigned long r3, unsigned long r4, unsigned long r5,
 #endif /* CONFIG_ADB */
 
        switch (_machine) {
+#ifdef CONFIG_PPC_PMAC
        case _MACH_Pmac:
                pmac_init(r3, r4, r5, r6, r7);
                break;
+#endif
+#ifdef CONFIG_PPC_CHRP
        case _MACH_chrp:
                chrp_init(r3, r4, r5, r6, r7);
                break;
+#endif
        }
 }
 
index fccafbcd4b586f7108394765288c2497e3ce80a3..8bc734fe6682e8c03078562b3b3782c0134fe3ff 100644 (file)
@@ -89,9 +89,6 @@ extern void prep_tiger1_setup_pci(char *irq_edge_mask_lo, char *irq_edge_mask_hi
 #define cached_21      (((char *)(ppc_cached_irq_mask))[3])
 #define cached_A1      (((char *)(ppc_cached_irq_mask))[2])
 
-/* for the mac fs */
-dev_t boot_dev;
-
 #ifdef CONFIG_SOUND_CS4232
 long ppc_cs4232_dma, ppc_cs4232_dma2;
 #endif
index b8d08f33f7ee6c1d2309c8aa92f96f37a434c9a3..1b0a84931afcdef8fd719ae2b4efd8e2d509e9b8 100644 (file)
@@ -5,6 +5,7 @@
 CFLAGS_prom_init.o      += -fPIC
 CFLAGS_btext.o          += -fPIC
 
+ifneq ($(CONFIG_PPC_MERGE),y)
 wdt-mpc8xx-$(CONFIG_8xx_WDT)   += m8xx_wdt.o
 
 obj-$(CONFIG_PPCBUG_NVRAM)     += prep_nvram.o
@@ -109,3 +110,16 @@ obj-$(CONFIG_PPC_MPC52xx)  += mpc52xx_setup.o mpc52xx_pic.o \
 ifeq ($(CONFIG_PPC_MPC52xx),y)
 obj-$(CONFIG_PCI)              += mpc52xx_pci.o
 endif
+
+else
+# Stuff still needed by the merged powerpc sources
+
+obj-$(CONFIG_PPCBUG_NVRAM)     += prep_nvram.o
+obj-$(CONFIG_PPC_OF)           += prom_init.o prom.o of_device.o
+obj-$(CONFIG_PPC_PMAC)         += indirect_pci.o
+obj-$(CONFIG_PPC_CHRP)         += indirect_pci.o i8259.o
+obj-$(CONFIG_PPC_PREP)         += indirect_pci.o i8259.o todc_time.o
+obj-$(CONFIG_BOOTX_TEXT)       += btext.o
+obj-$(CONFIG_MPC10X_BRIDGE)    += mpc10x_common.o indirect_pci.o ppc_sys.o
+
+endif
index 76719451e38402674b09b5d58ac1d823de610609..503461884528c76fd98ada1e770e0dbadcfc92a3 100644 (file)
@@ -405,7 +405,7 @@ static int __init via_pmu_start(void)
        bright_req_2.complete = 1;
        batt_req.complete = 1;
 
-#ifdef CONFIG_PPC32
+#if defined(CONFIG_PPC32) && !defined(CONFIG_PPC_MERGE)
        if (pmu_kind == PMU_KEYLARGO_BASED)
                openpic_set_irq_priority(vias->intrs[0].line,
                                         OPENPIC_PRIORITY_DEFAULT + 1);
index a3453555a94e76ca4e84952a993049faa257f2b4..5b6b0b6038a7bcaf0e1c2f65ff6faf422278db83 100644 (file)
@@ -629,12 +629,4 @@ void __init proc_misc_init(void)
        if (entry)
                entry->proc_fops = &proc_sysrq_trigger_operations;
 #endif
-#ifdef CONFIG_PPC32
-       {
-               extern struct file_operations ppc_htab_operations;
-               entry = create_proc_entry("ppc_htab", S_IRUGO|S_IWUSR, NULL);
-               if (entry)
-                       entry->proc_fops = &ppc_htab_operations;
-       }
-#endif
 }
diff --git a/include/asm-powerpc/kdebug.h b/include/asm-powerpc/kdebug.h
new file mode 100644 (file)
index 0000000..7c55abf
--- /dev/null
@@ -0,0 +1,42 @@
+#ifndef _POWERPC_KDEBUG_H
+#define _POWERPC_KDEBUG_H 1
+
+/* nearly identical to x86_64/i386 code */
+
+#include <linux/notifier.h>
+
+struct pt_regs;
+
+struct die_args {
+       struct pt_regs *regs;
+       const char *str;
+       long err;
+       int trapnr;
+       int signr;
+};
+
+/*
+   Note - you should never unregister because that can race with NMIs.
+   If you really want to do it first unregister - then synchronize_sched -
+   then free.
+ */
+int register_die_notifier(struct notifier_block *nb);
+extern struct notifier_block *powerpc_die_chain;
+
+/* Grossly misnamed. */
+enum die_val {
+       DIE_OOPS = 1,
+       DIE_IABR_MATCH,
+       DIE_DABR_MATCH,
+       DIE_BPT,
+       DIE_SSTEP,
+       DIE_PAGE_FAULT,
+};
+
+static inline int notify_die(enum die_val val,char *str,struct pt_regs *regs,long err,int trap, int sig)
+{
+       struct die_args args = { .regs=regs, .str=str, .err=err, .trapnr=trap,.signr=sig };
+       return notifier_call_chain(&powerpc_die_chain, val, &args);
+}
+
+#endif
diff --git a/include/asm-powerpc/kprobes.h b/include/asm-powerpc/kprobes.h
new file mode 100644 (file)
index 0000000..d9129d2
--- /dev/null
@@ -0,0 +1,67 @@
+#ifndef _ASM_KPROBES_H
+#define _ASM_KPROBES_H
+/*
+ *  Kernel Probes (KProbes)
+ *  include/asm-ppc64/kprobes.h
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
+ *
+ * Copyright (C) IBM Corporation, 2002, 2004
+ *
+ * 2002-Oct    Created by Vamsi Krishna S <vamsi_krishna@in.ibm.com> Kernel
+ *             Probes initial implementation ( includes suggestions from
+ *             Rusty Russell).
+ * 2004-Nov    Modified for PPC64 by Ananth N Mavinakayanahalli
+ *             <ananth@in.ibm.com>
+ */
+#include <linux/types.h>
+#include <linux/ptrace.h>
+
+struct pt_regs;
+
+typedef unsigned int kprobe_opcode_t;
+#define BREAKPOINT_INSTRUCTION 0x7fe00008      /* trap */
+#define MAX_INSN_SIZE 1
+
+#define IS_TW(instr)           (((instr) & 0xfc0007fe) == 0x7c000008)
+#define IS_TD(instr)           (((instr) & 0xfc0007fe) == 0x7c000088)
+#define IS_TDI(instr)          (((instr) & 0xfc000000) == 0x08000000)
+#define IS_TWI(instr)          (((instr) & 0xfc000000) == 0x0c000000)
+
+#define JPROBE_ENTRY(pentry)   (kprobe_opcode_t *)((func_descr_t *)pentry)
+
+#define is_trap(instr) (IS_TW(instr) || IS_TD(instr) || \
+                       IS_TWI(instr) || IS_TDI(instr))
+
+#define ARCH_SUPPORTS_KRETPROBES
+void kretprobe_trampoline(void);
+
+/* Architecture specific copy of original instruction */
+struct arch_specific_insn {
+       /* copy of original instruction */
+       kprobe_opcode_t *insn;
+};
+
+#ifdef CONFIG_KPROBES
+extern int kprobe_exceptions_notify(struct notifier_block *self,
+                                   unsigned long val, void *data);
+#else                          /* !CONFIG_KPROBES */
+static inline int kprobe_exceptions_notify(struct notifier_block *self,
+                                          unsigned long val, void *data)
+{
+       return 0;
+}
+#endif
+#endif                         /* _ASM_KPROBES_H */
diff --git a/include/asm-powerpc/mpic.h b/include/asm-powerpc/mpic.h
new file mode 100644 (file)
index 0000000..f1e24f4
--- /dev/null
@@ -0,0 +1,279 @@
+#include <linux/irq.h>
+
+/*
+ * Global registers
+ */
+
+#define MPIC_GREG_BASE                 0x01000
+
+#define MPIC_GREG_FEATURE_0            0x00000
+#define                MPIC_GREG_FEATURE_LAST_SRC_MASK         0x07ff0000
+#define                MPIC_GREG_FEATURE_LAST_SRC_SHIFT        16
+#define                MPIC_GREG_FEATURE_LAST_CPU_MASK         0x00001f00
+#define                MPIC_GREG_FEATURE_LAST_CPU_SHIFT        8
+#define                MPIC_GREG_FEATURE_VERSION_MASK          0xff
+#define MPIC_GREG_FEATURE_1            0x00010
+#define MPIC_GREG_GLOBAL_CONF_0                0x00020
+#define                MPIC_GREG_GCONF_RESET                   0x80000000
+#define                MPIC_GREG_GCONF_8259_PTHROU_DIS         0x20000000
+#define                MPIC_GREG_GCONF_BASE_MASK               0x000fffff
+#define MPIC_GREG_GLOBAL_CONF_1                0x00030
+#define MPIC_GREG_VENDOR_0             0x00040
+#define MPIC_GREG_VENDOR_1             0x00050
+#define MPIC_GREG_VENDOR_2             0x00060
+#define MPIC_GREG_VENDOR_3             0x00070
+#define MPIC_GREG_VENDOR_ID            0x00080
+#define        MPIC_GREG_VENDOR_ID_STEPPING_MASK       0x00ff0000
+#define        MPIC_GREG_VENDOR_ID_STEPPING_SHIFT      16
+#define        MPIC_GREG_VENDOR_ID_DEVICE_ID_MASK      0x0000ff00
+#define        MPIC_GREG_VENDOR_ID_DEVICE_ID_SHIFT     8
+#define        MPIC_GREG_VENDOR_ID_VENDOR_ID_MASK      0x000000ff
+#define MPIC_GREG_PROCESSOR_INIT       0x00090
+#define MPIC_GREG_IPI_VECTOR_PRI_0     0x000a0
+#define MPIC_GREG_IPI_VECTOR_PRI_1     0x000b0
+#define MPIC_GREG_IPI_VECTOR_PRI_2     0x000c0
+#define MPIC_GREG_IPI_VECTOR_PRI_3     0x000d0
+#define MPIC_GREG_SPURIOUS             0x000e0
+#define MPIC_GREG_TIMER_FREQ           0x000f0
+
+/*
+ *
+ * Timer registers
+ */
+#define MPIC_TIMER_BASE                        0x01100
+#define MPIC_TIMER_STRIDE              0x40
+
+#define MPIC_TIMER_CURRENT_CNT         0x00000
+#define MPIC_TIMER_BASE_CNT            0x00010
+#define MPIC_TIMER_VECTOR_PRI          0x00020
+#define MPIC_TIMER_DESTINATION         0x00030
+
+/*
+ * Per-Processor registers
+ */
+
+#define MPIC_CPU_THISBASE              0x00000
+#define MPIC_CPU_BASE                  0x20000
+#define MPIC_CPU_STRIDE                        0x01000
+
+#define MPIC_CPU_IPI_DISPATCH_0                0x00040
+#define MPIC_CPU_IPI_DISPATCH_1                0x00050
+#define MPIC_CPU_IPI_DISPATCH_2                0x00060
+#define MPIC_CPU_IPI_DISPATCH_3                0x00070
+#define MPIC_CPU_CURRENT_TASK_PRI      0x00080
+#define        MPIC_CPU_TASKPRI_MASK                   0x0000000f
+#define MPIC_CPU_WHOAMI                        0x00090
+#define        MPIC_CPU_WHOAMI_MASK                    0x0000001f
+#define MPIC_CPU_INTACK                        0x000a0
+#define MPIC_CPU_EOI                   0x000b0
+
+/*
+ * Per-source registers
+ */
+
+#define MPIC_IRQ_BASE                  0x10000
+#define MPIC_IRQ_STRIDE                        0x00020
+#define MPIC_IRQ_VECTOR_PRI            0x00000
+#define        MPIC_VECPRI_MASK                        0x80000000
+#define        MPIC_VECPRI_ACTIVITY                    0x40000000      /* Read Only */
+#define        MPIC_VECPRI_PRIORITY_MASK               0x000f0000
+#define        MPIC_VECPRI_PRIORITY_SHIFT              16
+#define        MPIC_VECPRI_VECTOR_MASK                 0x000007ff
+#define        MPIC_VECPRI_POLARITY_POSITIVE           0x00800000
+#define        MPIC_VECPRI_POLARITY_NEGATIVE           0x00000000
+#define        MPIC_VECPRI_POLARITY_MASK               0x00800000
+#define        MPIC_VECPRI_SENSE_LEVEL                 0x00400000
+#define        MPIC_VECPRI_SENSE_EDGE                  0x00000000
+#define        MPIC_VECPRI_SENSE_MASK                  0x00400000
+#define MPIC_IRQ_DESTINATION           0x00010
+
+#define MPIC_MAX_IRQ_SOURCES   2048
+#define MPIC_MAX_CPUS          32
+#define MPIC_MAX_ISU           32
+
+/*
+ * Special vector numbers (internal use only)
+ */
+#define MPIC_VEC_SPURRIOUS     255
+#define MPIC_VEC_IPI_3         254
+#define MPIC_VEC_IPI_2         253
+#define MPIC_VEC_IPI_1         252
+#define MPIC_VEC_IPI_0         251
+
+/* unused */
+#define MPIC_VEC_TIMER_3       250
+#define MPIC_VEC_TIMER_2       249
+#define MPIC_VEC_TIMER_1       248
+#define MPIC_VEC_TIMER_0       247
+
+/* Type definition of the cascade handler */
+typedef int (*mpic_cascade_t)(struct pt_regs *regs, void *data);
+
+#ifdef CONFIG_MPIC_BROKEN_U3
+/* Fixup table entry */
+struct mpic_irq_fixup
+{
+       u8 __iomem      *base;
+       unsigned int   irq;
+};
+#endif /* CONFIG_MPIC_BROKEN_U3 */
+
+
+/* The instance data of a given MPIC */
+struct mpic
+{
+       /* The "linux" controller struct */
+       hw_irq_controller       hc_irq;
+#ifdef CONFIG_SMP
+       hw_irq_controller       hc_ipi;
+#endif
+       const char              *name;
+       /* Flags */
+       unsigned int            flags;
+       /* How many irq sources in a given ISU */
+       unsigned int            isu_size;
+       unsigned int            isu_shift;
+       unsigned int            isu_mask;
+       /* Offset of irq vector numbers */
+       unsigned int            irq_offset;     
+       unsigned int            irq_count;
+       /* Offset of ipi vector numbers */
+       unsigned int            ipi_offset;
+       /* Number of sources */
+       unsigned int            num_sources;
+       /* Number of CPUs */
+       unsigned int            num_cpus;
+       /* cascade handler */
+       mpic_cascade_t          cascade;
+       void                    *cascade_data;
+       unsigned int            cascade_vec;
+       /* senses array */
+       unsigned char           *senses;
+       unsigned int            senses_count;
+
+#ifdef CONFIG_MPIC_BROKEN_U3
+       /* The fixup table */
+       struct mpic_irq_fixup   *fixups;
+       spinlock_t              fixup_lock;
+#endif
+
+       /* The various ioremap'ed bases */
+       volatile u32 __iomem    *gregs;
+       volatile u32 __iomem    *tmregs;
+       volatile u32 __iomem    *cpuregs[MPIC_MAX_CPUS];
+       volatile u32 __iomem    *isus[MPIC_MAX_ISU];
+
+       /* link */
+       struct mpic             *next;
+};
+
+/* This is the primary controller, only that one has IPIs and
+ * has afinity control. A non-primary MPIC always uses CPU0
+ * registers only
+ */
+#define MPIC_PRIMARY                   0x00000001
+/* Set this for a big-endian MPIC */
+#define MPIC_BIG_ENDIAN                        0x00000002
+/* Broken U3 MPIC */
+#define MPIC_BROKEN_U3                 0x00000004
+/* Broken IPI registers (autodetected) */
+#define MPIC_BROKEN_IPI                        0x00000008
+/* MPIC wants a reset */
+#define MPIC_WANTS_RESET               0x00000010
+
+/* Allocate the controller structure and setup the linux irq descs
+ * for the range if interrupts passed in. No HW initialization is
+ * actually performed.
+ * 
+ * @phys_addr: physial base address of the MPIC
+ * @flags:     flags, see constants above
+ * @isu_size:  number of interrupts in an ISU. Use 0 to use a
+ *              standard ISU-less setup (aka powermac)
+ * @irq_offset: first irq number to assign to this mpic
+ * @irq_count:  number of irqs to use with this mpic IRQ sources. Pass 0
+ *             to match the number of sources
+ * @ipi_offset: first irq number to assign to this mpic IPI sources,
+ *             used only on primary mpic
+ * @senses:    array of sense values
+ * @senses_num: number of entries in the array
+ *
+ * Note about the sense array. If none is passed, all interrupts are
+ * setup to be level negative unless MPIC_BROKEN_U3 is set in which
+ * case they are edge positive (and the array is ignored anyway).
+ * The values in the array start at the first source of the MPIC,
+ * that is senses[0] correspond to linux irq "irq_offset".
+ */
+extern struct mpic *mpic_alloc(unsigned long phys_addr,
+                              unsigned int flags,
+                              unsigned int isu_size,
+                              unsigned int irq_offset,
+                              unsigned int irq_count,
+                              unsigned int ipi_offset,
+                              unsigned char *senses,
+                              unsigned int senses_num,
+                              const char *name);
+
+/* Assign ISUs, to call before mpic_init()
+ *
+ * @mpic:      controller structure as returned by mpic_alloc()
+ * @isu_num:   ISU number
+ * @phys_addr: physical address of the ISU
+ */
+extern void mpic_assign_isu(struct mpic *mpic, unsigned int isu_num,
+                           unsigned long phys_addr);
+
+/* Initialize the controller. After this has been called, none of the above
+ * should be called again for this mpic
+ */
+extern void mpic_init(struct mpic *mpic);
+
+/* Setup a cascade. Currently, only one cascade is supported this
+ * way, though you can always do a normal request_irq() and add
+ * other cascades this way. You should call this _after_ having
+ * added all the ISUs
+ *
+ * @irq_no:    "linux" irq number of the cascade (that is offset'ed vector)
+ * @handler:   cascade handler function
+ */
+extern void mpic_setup_cascade(unsigned int irq_no, mpic_cascade_t hanlder,
+                              void *data);
+
+/*
+ * All of the following functions must only be used after the
+ * ISUs have been assigned and the controller fully initialized
+ * with mpic_init()
+ */
+
+
+/* Change/Read the priority of an interrupt. Default is 8 for irqs and
+ * 10 for IPIs. You can call this on both IPIs and IRQ numbers, but the
+ * IPI number is then the offset'ed (linux irq number mapped to the IPI)
+ */
+extern void mpic_irq_set_priority(unsigned int irq, unsigned int pri);
+extern unsigned int mpic_irq_get_priority(unsigned int irq);
+
+/* Setup a non-boot CPU */
+extern void mpic_setup_this_cpu(void);
+
+/* Clean up for kexec (or cpu offline or ...) */
+extern void mpic_teardown_this_cpu(int secondary);
+
+/* Get the current cpu priority for this cpu (0..15) */
+extern int mpic_cpu_get_priority(void);
+
+/* Set the current cpu priority for this cpu */
+extern void mpic_cpu_set_priority(int prio);
+
+/* Request IPIs on primary mpic */
+extern void mpic_request_ipis(void);
+
+/* Send an IPI (non offseted number 0..3) */
+extern void mpic_send_ipi(unsigned int ipi_no, unsigned int cpu_mask);
+
+/* Fetch interrupt from a given mpic */
+extern int mpic_get_one_irq(struct mpic *mpic, struct pt_regs *regs);
+/* This one gets to the primary mpic */
+extern int mpic_get_irq(struct pt_regs *regs);
+
+/* global mpic for pSeries */
+extern struct mpic *pSeries_mpic;
diff --git a/include/asm-powerpc/reg.h b/include/asm-powerpc/reg.h
new file mode 100644 (file)
index 0000000..f97a5f1
--- /dev/null
@@ -0,0 +1,446 @@
+/*
+ * Contains the definition of registers common to all PowerPC variants.
+ * If a register definition has been changed in a different PowerPC
+ * variant, we will case it in #ifndef XXX ... #endif, and have the
+ * number used in the Programming Environments Manual For 32-Bit
+ * Implementations of the PowerPC Architecture (a.k.a. Green Book) here.
+ */
+
+#ifdef __KERNEL__
+#ifndef __ASM_PPC_REGS_H__
+#define __ASM_PPC_REGS_H__
+
+#include <linux/stringify.h>
+
+/* Pickup Book E specific registers. */
+#if defined(CONFIG_BOOKE) || defined(CONFIG_40x)
+#include <asm/reg_booke.h>
+#endif
+
+/* Machine State Register (MSR) Fields */
+#define MSR_SF         (1<<63)
+#define MSR_ISF                (1<<61)
+#define MSR_VEC                (1<<25)         /* Enable AltiVec */
+#define MSR_POW                (1<<18)         /* Enable Power Management */
+#define MSR_WE         (1<<18)         /* Wait State Enable */
+#define MSR_TGPR       (1<<17)         /* TLB Update registers in use */
+#define MSR_CE         (1<<17)         /* Critical Interrupt Enable */
+#define MSR_ILE                (1<<16)         /* Interrupt Little Endian */
+#define MSR_EE         (1<<15)         /* External Interrupt Enable */
+#define MSR_PR         (1<<14)         /* Problem State / Privilege Level */
+#define MSR_FP         (1<<13)         /* Floating Point enable */
+#define MSR_ME         (1<<12)         /* Machine Check Enable */
+#define MSR_FE0                (1<<11)         /* Floating Exception mode 0 */
+#define MSR_SE         (1<<10)         /* Single Step */
+#define MSR_BE         (1<<9)          /* Branch Trace */
+#define MSR_DE         (1<<9)          /* Debug Exception Enable */
+#define MSR_FE1                (1<<8)          /* Floating Exception mode 1 */
+#define MSR_IP         (1<<6)          /* Exception prefix 0x000/0xFFF */
+#define MSR_IR         (1<<5)          /* Instruction Relocate */
+#define MSR_DR         (1<<4)          /* Data Relocate */
+#define MSR_PE         (1<<3)          /* Protection Enable */
+#define MSR_PX         (1<<2)          /* Protection Exclusive Mode */
+#define MSR_RI         (1<<1)          /* Recoverable Exception */
+#define MSR_LE         (1<<0)          /* Little Endian */
+
+/* Default MSR for kernel mode. */
+#ifdef CONFIG_APUS_FAST_EXCEPT
+#define MSR_KERNEL     (MSR_ME|MSR_IP|MSR_RI|MSR_IR|MSR_DR)
+#endif
+
+#ifndef MSR_KERNEL
+#define MSR_KERNEL     (MSR_ME|MSR_RI|MSR_IR|MSR_DR)
+#endif
+
+#define MSR_USER       (MSR_KERNEL|MSR_PR|MSR_EE)
+
+/* Floating Point Status and Control Register (FPSCR) Fields */
+#define FPSCR_FX       0x80000000      /* FPU exception summary */
+#define FPSCR_FEX      0x40000000      /* FPU enabled exception summary */
+#define FPSCR_VX       0x20000000      /* Invalid operation summary */
+#define FPSCR_OX       0x10000000      /* Overflow exception summary */
+#define FPSCR_UX       0x08000000      /* Underflow exception summary */
+#define FPSCR_ZX       0x04000000      /* Zero-divide exception summary */
+#define FPSCR_XX       0x02000000      /* Inexact exception summary */
+#define FPSCR_VXSNAN   0x01000000      /* Invalid op for SNaN */
+#define FPSCR_VXISI    0x00800000      /* Invalid op for Inv - Inv */
+#define FPSCR_VXIDI    0x00400000      /* Invalid op for Inv / Inv */
+#define FPSCR_VXZDZ    0x00200000      /* Invalid op for Zero / Zero */
+#define FPSCR_VXIMZ    0x00100000      /* Invalid op for Inv * Zero */
+#define FPSCR_VXVC     0x00080000      /* Invalid op for Compare */
+#define FPSCR_FR       0x00040000      /* Fraction rounded */
+#define FPSCR_FI       0x00020000      /* Fraction inexact */
+#define FPSCR_FPRF     0x0001f000      /* FPU Result Flags */
+#define FPSCR_FPCC     0x0000f000      /* FPU Condition Codes */
+#define FPSCR_VXSOFT   0x00000400      /* Invalid op for software request */
+#define FPSCR_VXSQRT   0x00000200      /* Invalid op for square root */
+#define FPSCR_VXCVI    0x00000100      /* Invalid op for integer convert */
+#define FPSCR_VE       0x00000080      /* Invalid op exception enable */
+#define FPSCR_OE       0x00000040      /* IEEE overflow exception enable */
+#define FPSCR_UE       0x00000020      /* IEEE underflow exception enable */
+#define FPSCR_ZE       0x00000010      /* IEEE zero divide exception enable */
+#define FPSCR_XE       0x00000008      /* FP inexact exception enable */
+#define FPSCR_NI       0x00000004      /* FPU non IEEE-Mode */
+#define FPSCR_RN       0x00000003      /* FPU rounding control */
+
+/* Special Purpose Registers (SPRNs)*/
+#define SPRN_CTR       0x009   /* Count Register */
+#define SPRN_DABR      0x3F5   /* Data Address Breakpoint Register */
+#define   DABR_TRANSLATION     (1UL << 2)
+#define SPRN_DAR       0x013   /* Data Address Register */
+#define        SPRN_DSISR      0x012   /* Data Storage Interrupt Status Register */
+#define   DSISR_NOHPTE         0x40000000      /* no translation found */
+#define   DSISR_PROTFAULT      0x08000000      /* protection fault */
+#define   DSISR_ISSTORE                0x02000000      /* access was a store */
+#define   DSISR_DABRMATCH      0x00400000      /* hit data breakpoint */
+#define   DSISR_NOSEGMENT      0x00200000      /* STAB/SLB miss */
+#define SPRN_TBRL      0x10C   /* Time Base Read Lower Register (user, R/O) */
+#define SPRN_TBRU      0x10D   /* Time Base Read Upper Register (user, R/O) */
+#define SPRN_TBWL      0x11C   /* Time Base Lower Register (super, R/W) */
+#define SPRN_TBWU      0x11D   /* Time Base Upper Register (super, R/W) */
+#define SPRN_HIOR      0x137   /* 970 Hypervisor interrupt offset */
+#define SPRN_DBAT0L    0x219   /* Data BAT 0 Lower Register */
+#define SPRN_DBAT0U    0x218   /* Data BAT 0 Upper Register */
+#define SPRN_DBAT1L    0x21B   /* Data BAT 1 Lower Register */
+#define SPRN_DBAT1U    0x21A   /* Data BAT 1 Upper Register */
+#define SPRN_DBAT2L    0x21D   /* Data BAT 2 Lower Register */
+#define SPRN_DBAT2U    0x21C   /* Data BAT 2 Upper Register */
+#define SPRN_DBAT3L    0x21F   /* Data BAT 3 Lower Register */
+#define SPRN_DBAT3U    0x21E   /* Data BAT 3 Upper Register */
+#define SPRN_DBAT4L    0x239   /* Data BAT 4 Lower Register */
+#define SPRN_DBAT4U    0x238   /* Data BAT 4 Upper Register */
+#define SPRN_DBAT5L    0x23B   /* Data BAT 5 Lower Register */
+#define SPRN_DBAT5U    0x23A   /* Data BAT 5 Upper Register */
+#define SPRN_DBAT6L    0x23D   /* Data BAT 6 Lower Register */
+#define SPRN_DBAT6U    0x23C   /* Data BAT 6 Upper Register */
+#define SPRN_DBAT7L    0x23F   /* Data BAT 7 Lower Register */
+#define SPRN_DBAT7U    0x23E   /* Data BAT 7 Upper Register */
+
+#define SPRN_DEC       0x016           /* Decrement Register */
+#define SPRN_DER       0x095           /* Debug Enable Regsiter */
+#define DER_RSTE       0x40000000      /* Reset Interrupt */
+#define DER_CHSTPE     0x20000000      /* Check Stop */
+#define DER_MCIE       0x10000000      /* Machine Check Interrupt */
+#define DER_EXTIE      0x02000000      /* External Interrupt */
+#define DER_ALIE       0x01000000      /* Alignment Interrupt */
+#define DER_PRIE       0x00800000      /* Program Interrupt */
+#define DER_FPUVIE     0x00400000      /* FP Unavailable Interrupt */
+#define DER_DECIE      0x00200000      /* Decrementer Interrupt */
+#define DER_SYSIE      0x00040000      /* System Call Interrupt */
+#define DER_TRE                0x00020000      /* Trace Interrupt */
+#define DER_SEIE       0x00004000      /* FP SW Emulation Interrupt */
+#define DER_ITLBMSE    0x00002000      /* Imp. Spec. Instruction TLB Miss */
+#define DER_ITLBERE    0x00001000      /* Imp. Spec. Instruction TLB Error */
+#define DER_DTLBMSE    0x00000800      /* Imp. Spec. Data TLB Miss */
+#define DER_DTLBERE    0x00000400      /* Imp. Spec. Data TLB Error */
+#define DER_LBRKE      0x00000008      /* Load/Store Breakpoint Interrupt */
+#define DER_IBRKE      0x00000004      /* Instruction Breakpoint Interrupt */
+#define DER_EBRKE      0x00000002      /* External Breakpoint Interrupt */
+#define DER_DPIE       0x00000001      /* Dev. Port Nonmaskable Request */
+#define SPRN_DMISS     0x3D0           /* Data TLB Miss Register */
+#define SPRN_EAR       0x11A           /* External Address Register */
+#define SPRN_HASH1     0x3D2           /* Primary Hash Address Register */
+#define SPRN_HASH2     0x3D3           /* Secondary Hash Address Resgister */
+#define SPRN_HID0      0x3F0           /* Hardware Implementation Register 0 */
+#define HID0_EMCP      (1<<31)         /* Enable Machine Check pin */
+#define HID0_EBA       (1<<29)         /* Enable Bus Address Parity */
+#define HID0_EBD       (1<<28)         /* Enable Bus Data Parity */
+#define HID0_SBCLK     (1<<27)
+#define HID0_EICE      (1<<26)
+#define HID0_TBEN      (1<<26)         /* Timebase enable - 745x */
+#define HID0_ECLK      (1<<25)
+#define HID0_PAR       (1<<24)
+#define HID0_STEN      (1<<24)         /* Software table search enable - 745x */
+#define HID0_HIGH_BAT  (1<<23)         /* Enable high BATs - 7455 */
+#define HID0_DOZE      (1<<23)
+#define HID0_NAP       (1<<22)
+#define HID0_SLEEP     (1<<21)
+#define HID0_DPM       (1<<20)
+#define HID0_BHTCLR    (1<<18)         /* Clear branch history table - 7450 */
+#define HID0_XAEN      (1<<17)         /* Extended addressing enable - 7450 */
+#define HID0_NHR       (1<<16)         /* Not hard reset (software bit-7450)*/
+#define HID0_ICE       (1<<15)         /* Instruction Cache Enable */
+#define HID0_DCE       (1<<14)         /* Data Cache Enable */
+#define HID0_ILOCK     (1<<13)         /* Instruction Cache Lock */
+#define HID0_DLOCK     (1<<12)         /* Data Cache Lock */
+#define HID0_ICFI      (1<<11)         /* Instr. Cache Flash Invalidate */
+#define HID0_DCI       (1<<10)         /* Data Cache Invalidate */
+#define HID0_SPD       (1<<9)          /* Speculative disable */
+#define HID0_DAPUEN    (1<<8)          /* Debug APU enable */
+#define HID0_SGE       (1<<7)          /* Store Gathering Enable */
+#define HID0_SIED      (1<<7)          /* Serial Instr. Execution [Disable] */
+#define HID0_DFCA      (1<<6)          /* Data Cache Flush Assist */
+#define HID0_LRSTK     (1<<4)          /* Link register stack - 745x */
+#define HID0_BTIC      (1<<5)          /* Branch Target Instr Cache Enable */
+#define HID0_ABE       (1<<3)          /* Address Broadcast Enable */
+#define HID0_FOLD      (1<<3)          /* Branch Folding enable - 745x */
+#define HID0_BHTE      (1<<2)          /* Branch History Table Enable */
+#define HID0_BTCD      (1<<1)          /* Branch target cache disable */
+#define HID0_NOPDST    (1<<1)          /* No-op dst, dstt, etc. instr. */
+#define HID0_NOPTI     (1<<0)          /* No-op dcbt and dcbst instr. */
+
+#define SPRN_HID1      0x3F1           /* Hardware Implementation Register 1 */
+#define HID1_EMCP      (1<<31)         /* 7450 Machine Check Pin Enable */
+#define HID1_DFS       (1<<22)         /* 7447A Dynamic Frequency Scaling */
+#define HID1_PC0       (1<<16)         /* 7450 PLL_CFG[0] */
+#define HID1_PC1       (1<<15)         /* 7450 PLL_CFG[1] */
+#define HID1_PC2       (1<<14)         /* 7450 PLL_CFG[2] */
+#define HID1_PC3       (1<<13)         /* 7450 PLL_CFG[3] */
+#define HID1_SYNCBE    (1<<11)         /* 7450 ABE for sync, eieio */
+#define HID1_ABE       (1<<10)         /* 7450 Address Broadcast Enable */
+#define HID1_PS                (1<<16)         /* 750FX PLL selection */
+#define SPRN_HID2      0x3F8           /* Hardware Implementation Register 2 */
+#define SPRN_IABR      0x3F2   /* Instruction Address Breakpoint Register */
+#define SPRN_HID4      0x3F4           /* 970 HID4 */
+#define SPRN_HID5      0x3F6           /* 970 HID5 */
+#if !defined(SPRN_IAC1) && !defined(SPRN_IAC2)
+#define SPRN_IAC1      0x3F4           /* Instruction Address Compare 1 */
+#define SPRN_IAC2      0x3F5           /* Instruction Address Compare 2 */
+#endif
+#define SPRN_IBAT0L    0x211           /* Instruction BAT 0 Lower Register */
+#define SPRN_IBAT0U    0x210           /* Instruction BAT 0 Upper Register */
+#define SPRN_IBAT1L    0x213           /* Instruction BAT 1 Lower Register */
+#define SPRN_IBAT1U    0x212           /* Instruction BAT 1 Upper Register */
+#define SPRN_IBAT2L    0x215           /* Instruction BAT 2 Lower Register */
+#define SPRN_IBAT2U    0x214           /* Instruction BAT 2 Upper Register */
+#define SPRN_IBAT3L    0x217           /* Instruction BAT 3 Lower Register */
+#define SPRN_IBAT3U    0x216           /* Instruction BAT 3 Upper Register */
+#define SPRN_IBAT4L    0x231           /* Instruction BAT 4 Lower Register */
+#define SPRN_IBAT4U    0x230           /* Instruction BAT 4 Upper Register */
+#define SPRN_IBAT5L    0x233           /* Instruction BAT 5 Lower Register */
+#define SPRN_IBAT5U    0x232           /* Instruction BAT 5 Upper Register */
+#define SPRN_IBAT6L    0x235           /* Instruction BAT 6 Lower Register */
+#define SPRN_IBAT6U    0x234           /* Instruction BAT 6 Upper Register */
+#define SPRN_IBAT7L    0x237           /* Instruction BAT 7 Lower Register */
+#define SPRN_IBAT7U    0x236           /* Instruction BAT 7 Upper Register */
+#define SPRN_ICMP      0x3D5           /* Instruction TLB Compare Register */
+#define SPRN_ICTC      0x3FB   /* Instruction Cache Throttling Control Reg */
+#define SPRN_ICTRL     0x3F3   /* 1011 7450 icache and interrupt ctrl */
+#define ICTRL_EICE     0x08000000      /* enable icache parity errs */
+#define ICTRL_EDC      0x04000000      /* enable dcache parity errs */
+#define ICTRL_EICP     0x00000100      /* enable icache par. check */
+#define SPRN_IMISS     0x3D4           /* Instruction TLB Miss Register */
+#define SPRN_IMMR      0x27E           /* Internal Memory Map Register */
+#define SPRN_L2CR      0x3F9           /* Level 2 Cache Control Regsiter */
+#define SPRN_L2CR2     0x3f8
+#define L2CR_L2E               0x80000000      /* L2 enable */
+#define L2CR_L2PE              0x40000000      /* L2 parity enable */
+#define L2CR_L2SIZ_MASK                0x30000000      /* L2 size mask */
+#define L2CR_L2SIZ_256KB       0x10000000      /* L2 size 256KB */
+#define L2CR_L2SIZ_512KB       0x20000000      /* L2 size 512KB */
+#define L2CR_L2SIZ_1MB         0x30000000      /* L2 size 1MB */
+#define L2CR_L2CLK_MASK                0x0e000000      /* L2 clock mask */
+#define L2CR_L2CLK_DISABLED    0x00000000      /* L2 clock disabled */
+#define L2CR_L2CLK_DIV1                0x02000000      /* L2 clock / 1 */
+#define L2CR_L2CLK_DIV1_5      0x04000000      /* L2 clock / 1.5 */
+#define L2CR_L2CLK_DIV2                0x08000000      /* L2 clock / 2 */
+#define L2CR_L2CLK_DIV2_5      0x0a000000      /* L2 clock / 2.5 */
+#define L2CR_L2CLK_DIV3                0x0c000000      /* L2 clock / 3 */
+#define L2CR_L2RAM_MASK                0x01800000      /* L2 RAM type mask */
+#define L2CR_L2RAM_FLOW                0x00000000      /* L2 RAM flow through */
+#define L2CR_L2RAM_PIPE                0x01000000      /* L2 RAM pipelined */
+#define L2CR_L2RAM_PIPE_LW     0x01800000      /* L2 RAM pipelined latewr */
+#define L2CR_L2DO              0x00400000      /* L2 data only */
+#define L2CR_L2I               0x00200000      /* L2 global invalidate */
+#define L2CR_L2CTL             0x00100000      /* L2 RAM control */
+#define L2CR_L2WT              0x00080000      /* L2 write-through */
+#define L2CR_L2TS              0x00040000      /* L2 test support */
+#define L2CR_L2OH_MASK         0x00030000      /* L2 output hold mask */
+#define L2CR_L2OH_0_5          0x00000000      /* L2 output hold 0.5 ns */
+#define L2CR_L2OH_1_0          0x00010000      /* L2 output hold 1.0 ns */
+#define L2CR_L2SL              0x00008000      /* L2 DLL slow */
+#define L2CR_L2DF              0x00004000      /* L2 differential clock */
+#define L2CR_L2BYP             0x00002000      /* L2 DLL bypass */
+#define L2CR_L2IP              0x00000001      /* L2 GI in progress */
+#define L2CR_L2IO_745x         0x00100000      /* L2 instr. only (745x) */
+#define L2CR_L2DO_745x         0x00010000      /* L2 data only (745x) */
+#define L2CR_L2REP_745x                0x00001000      /* L2 repl. algorithm (745x) */
+#define L2CR_L2HWF_745x                0x00000800      /* L2 hardware flush (745x) */
+#define SPRN_L3CR              0x3FA   /* Level 3 Cache Control Regsiter */
+#define L3CR_L3E               0x80000000      /* L3 enable */
+#define L3CR_L3PE              0x40000000      /* L3 data parity enable */
+#define L3CR_L3APE             0x20000000      /* L3 addr parity enable */
+#define L3CR_L3SIZ             0x10000000      /* L3 size */
+#define L3CR_L3CLKEN           0x08000000      /* L3 clock enable */
+#define L3CR_L3RES             0x04000000      /* L3 special reserved bit */
+#define L3CR_L3CLKDIV          0x03800000      /* L3 clock divisor */
+#define L3CR_L3IO              0x00400000      /* L3 instruction only */
+#define L3CR_L3SPO             0x00040000      /* L3 sample point override */
+#define L3CR_L3CKSP            0x00030000      /* L3 clock sample point */
+#define L3CR_L3PSP             0x0000e000      /* L3 P-clock sample point */
+#define L3CR_L3REP             0x00001000      /* L3 replacement algorithm */
+#define L3CR_L3HWF             0x00000800      /* L3 hardware flush */
+#define L3CR_L3I               0x00000400      /* L3 global invalidate */
+#define L3CR_L3RT              0x00000300      /* L3 SRAM type */
+#define L3CR_L3NIRCA           0x00000080      /* L3 non-integer ratio clock adj. */
+#define L3CR_L3DO              0x00000040      /* L3 data only mode */
+#define L3CR_PMEN              0x00000004      /* L3 private memory enable */
+#define L3CR_PMSIZ             0x00000001      /* L3 private memory size */
+#define SPRN_MSSCR0    0x3f6   /* Memory Subsystem Control Register 0 */
+#define SPRN_MSSSR0    0x3f7   /* Memory Subsystem Status Register 1 */
+#define SPRN_LDSTCR    0x3f8   /* Load/Store control register */
+#define SPRN_LDSTDB    0x3f4   /* */
+#define SPRN_LR                0x008   /* Link Register */
+#define SPRN_MMCR0     0x3B8   /* Monitor Mode Control Register 0 */
+#define SPRN_MMCR1     0x3BC   /* Monitor Mode Control Register 1 */
+#ifndef SPRN_PIR
+#define SPRN_PIR       0x3FF   /* Processor Identification Register */
+#endif
+#define SPRN_PMC1      0x3B9   /* Performance Counter Register 1 */
+#define SPRN_PMC2      0x3BA   /* Performance Counter Register 2 */
+#define SPRN_PMC3      0x3BD   /* Performance Counter Register 3 */
+#define SPRN_PMC4      0x3BE   /* Performance Counter Register 4 */
+#define SPRN_PTEHI     0x3D5   /* 981 7450 PTE HI word (S/W TLB load) */
+#define SPRN_PTELO     0x3D6   /* 982 7450 PTE LO word (S/W TLB load) */
+#define SPRN_PVR       0x11F   /* Processor Version Register */
+#define SPRN_RPA       0x3D6   /* Required Physical Address Register */
+#define SPRN_SDA       0x3BF   /* Sampled Data Address Register */
+#define SPRN_SDR1      0x019   /* MMU Hash Base Register */
+#define SPRN_SIA       0x3BB   /* Sampled Instruction Address Register */
+#define SPRN_SPRG0     0x110   /* Special Purpose Register General 0 */
+#define SPRN_SPRG1     0x111   /* Special Purpose Register General 1 */
+#define SPRN_SPRG2     0x112   /* Special Purpose Register General 2 */
+#define SPRN_SPRG3     0x113   /* Special Purpose Register General 3 */
+#define SPRN_SPRG4     0x114   /* Special Purpose Register General 4 */
+#define SPRN_SPRG5     0x115   /* Special Purpose Register General 5 */
+#define SPRN_SPRG6     0x116   /* Special Purpose Register General 6 */
+#define SPRN_SPRG7     0x117   /* Special Purpose Register General 7 */
+#define SPRN_SRR0      0x01A   /* Save/Restore Register 0 */
+#define SPRN_SRR1      0x01B   /* Save/Restore Register 1 */
+#ifndef SPRN_SVR
+#define SPRN_SVR       0x11E   /* System Version Register */
+#endif
+#define SPRN_THRM1     0x3FC           /* Thermal Management Register 1 */
+/* these bits were defined in inverted endian sense originally, ugh, confusing */
+#define THRM1_TIN      (1 << 31)
+#define THRM1_TIV      (1 << 30)
+#define THRM1_THRES(x) ((x&0x7f)<<23)
+#define THRM3_SITV(x)  ((x&0x3fff)<<1)
+#define THRM1_TID      (1<<2)
+#define THRM1_TIE      (1<<1)
+#define THRM1_V                (1<<0)
+#define SPRN_THRM2     0x3FD           /* Thermal Management Register 2 */
+#define SPRN_THRM3     0x3FE           /* Thermal Management Register 3 */
+#define THRM3_E                (1<<0)
+#define SPRN_TLBMISS   0x3D4           /* 980 7450 TLB Miss Register */
+#define SPRN_UMMCR0    0x3A8   /* User Monitor Mode Control Register 0 */
+#define SPRN_UMMCR1    0x3AC   /* User Monitor Mode Control Register 0 */
+#define SPRN_UPMC1     0x3A9   /* User Performance Counter Register 1 */
+#define SPRN_UPMC2     0x3AA   /* User Performance Counter Register 2 */
+#define SPRN_UPMC3     0x3AD   /* User Performance Counter Register 3 */
+#define SPRN_UPMC4     0x3AE   /* User Performance Counter Register 4 */
+#define SPRN_USIA      0x3AB   /* User Sampled Instruction Address Register */
+#define SPRN_VRSAVE    0x100   /* Vector Register Save Register */
+#define SPRN_XER       0x001   /* Fixed Point Exception Register */
+
+/* Bit definitions for MMCR0 and PMC1 / PMC2. */
+#define MMCR0_PMC1_CYCLES      (1 << 7)
+#define MMCR0_PMC1_ICACHEMISS  (5 << 7)
+#define MMCR0_PMC1_DTLB                (6 << 7)
+#define MMCR0_PMC2_DCACHEMISS  0x6
+#define MMCR0_PMC2_CYCLES      0x1
+#define MMCR0_PMC2_ITLB                0x7
+#define MMCR0_PMC2_LOADMISSTIME        0x5
+#define MMCR0_PMXE     (1 << 26)
+
+/* Processor Version Register */
+
+/* Processor Version Register (PVR) field extraction */
+
+#define PVR_VER(pvr)   (((pvr) >>  16) & 0xFFFF)       /* Version field */
+#define PVR_REV(pvr)   (((pvr) >>   0) & 0xFFFF)       /* Revison field */
+
+/*
+ * IBM has further subdivided the standard PowerPC 16-bit version and
+ * revision subfields of the PVR for the PowerPC 403s into the following:
+ */
+
+#define PVR_FAM(pvr)   (((pvr) >> 20) & 0xFFF) /* Family field */
+#define PVR_MEM(pvr)   (((pvr) >> 16) & 0xF)   /* Member field */
+#define PVR_CORE(pvr)  (((pvr) >> 12) & 0xF)   /* Core field */
+#define PVR_CFG(pvr)   (((pvr) >>  8) & 0xF)   /* Configuration field */
+#define PVR_MAJ(pvr)   (((pvr) >>  4) & 0xF)   /* Major revision field */
+#define PVR_MIN(pvr)   (((pvr) >>  0) & 0xF)   /* Minor revision field */
+
+/* Processor Version Numbers */
+
+#define PVR_403GA      0x00200000
+#define PVR_403GB      0x00200100
+#define PVR_403GC      0x00200200
+#define PVR_403GCX     0x00201400
+#define PVR_405GP      0x40110000
+#define PVR_STB03XXX   0x40310000
+#define PVR_NP405H     0x41410000
+#define PVR_NP405L     0x41610000
+#define PVR_601                0x00010000
+#define PVR_602                0x00050000
+#define PVR_603                0x00030000
+#define PVR_603e       0x00060000
+#define PVR_603ev      0x00070000
+#define PVR_603r       0x00071000
+#define PVR_604                0x00040000
+#define PVR_604e       0x00090000
+#define PVR_604r       0x000A0000
+#define PVR_620                0x00140000
+#define PVR_740                0x00080000
+#define PVR_750                PVR_740
+#define PVR_740P       0x10080000
+#define PVR_750P       PVR_740P
+#define PVR_7400       0x000C0000
+#define PVR_7410       0x800C0000
+#define PVR_7450       0x80000000
+#define PVR_8540       0x80200000
+#define PVR_8560       0x80200000
+/*
+ * For the 8xx processors, all of them report the same PVR family for
+ * the PowerPC core. The various versions of these processors must be
+ * differentiated by the version number in the Communication Processor
+ * Module (CPM).
+ */
+#define PVR_821                0x00500000
+#define PVR_823                PVR_821
+#define PVR_850                PVR_821
+#define PVR_860                PVR_821
+#define PVR_8240       0x00810100
+#define PVR_8245       0x80811014
+#define PVR_8260       PVR_8240
+
+#if 0
+/* Segment Registers */
+#define SR0    0
+#define SR1    1
+#define SR2    2
+#define SR3    3
+#define SR4    4
+#define SR5    5
+#define SR6    6
+#define SR7    7
+#define SR8    8
+#define SR9    9
+#define SR10   10
+#define SR11   11
+#define SR12   12
+#define SR13   13
+#define SR14   14
+#define SR15   15
+#endif
+
+/* Macros for setting and retrieving special purpose registers */
+#ifndef __ASSEMBLY__
+#define mfmsr()                ({unsigned int rval; \
+                       asm volatile("mfmsr %0" : "=r" (rval)); rval;})
+#define mtmsr(v)       asm volatile("mtmsr %0" : : "r" (v))
+
+#define mfspr(rn)      ({unsigned int rval; \
+                       asm volatile("mfspr %0," __stringify(rn) \
+                               : "=r" (rval)); rval;})
+#define mtspr(rn, v)   asm volatile("mtspr " __stringify(rn) ",%0" : : "r" (v))
+
+#define mfsrin(v)      ({unsigned int rval; \
+                       asm volatile("mfsrin %0,%1" : "=r" (rval) : "r" (v)); \
+                                       rval;})
+
+#define proc_trap()    asm volatile("trap")
+#endif /* __ASSEMBLY__ */
+#endif /* __ASM_PPC_REGS_H__ */
+#endif /* __KERNEL__ */
diff --git a/include/asm-powerpc/system.h b/include/asm-powerpc/system.h
new file mode 100644 (file)
index 0000000..be542ef
--- /dev/null
@@ -0,0 +1,350 @@
+/*
+ * Copyright (C) 1999 Cort Dougan <cort@cs.nmt.edu>
+ */
+#ifndef __PPC_SYSTEM_H
+#define __PPC_SYSTEM_H
+
+#include <linux/config.h>
+#include <linux/kernel.h>
+
+#include <asm/hw_irq.h>
+#include <asm/ppc_asm.h>
+
+/*
+ * Memory barrier.
+ * The sync instruction guarantees that all memory accesses initiated
+ * by this processor have been performed (with respect to all other
+ * mechanisms that access memory).  The eieio instruction is a barrier
+ * providing an ordering (separately) for (a) cacheable stores and (b)
+ * loads and stores to non-cacheable memory (e.g. I/O devices).
+ *
+ * mb() prevents loads and stores being reordered across this point.
+ * rmb() prevents loads being reordered across this point.
+ * wmb() prevents stores being reordered across this point.
+ * read_barrier_depends() prevents data-dependent loads being reordered
+ *     across this point (nop on PPC).
+ *
+ * We have to use the sync instructions for mb(), since lwsync doesn't
+ * order loads with respect to previous stores.  Lwsync is fine for
+ * rmb(), though.  Note that lwsync is interpreted as sync by
+ * 32-bit and older 64-bit CPUs.
+ *
+ * For wmb(), we use sync since wmb is used in drivers to order
+ * stores to system memory with respect to writes to the device.
+ * However, smp_wmb() can be a lighter-weight eieio barrier on
+ * SMP since it is only used to order updates to system memory.
+ */
+#define mb()   __asm__ __volatile__ ("sync" : : : "memory")
+#define rmb()  __asm__ __volatile__ ("lwsync" : : : "memory")
+#define wmb()  __asm__ __volatile__ ("sync" : : : "memory")
+#define read_barrier_depends()  do { } while(0)
+
+#define set_mb(var, value)     do { var = value; mb(); } while (0)
+#define set_wmb(var, value)    do { var = value; wmb(); } while (0)
+
+#ifdef CONFIG_SMP
+#define smp_mb()       mb()
+#define smp_rmb()      rmb()
+#define smp_wmb()      __asm__ __volatile__ ("eieio" : : : "memory")
+#define smp_read_barrier_depends()     read_barrier_depends()
+#else
+#define smp_mb()       barrier()
+#define smp_rmb()      barrier()
+#define smp_wmb()      barrier()
+#define smp_read_barrier_depends()     do { } while(0)
+#endif /* CONFIG_SMP */
+
+#ifdef __KERNEL__
+struct task_struct;
+struct pt_regs;
+
+#ifdef CONFIG_DEBUGGER
+
+extern int (*__debugger)(struct pt_regs *regs);
+extern int (*__debugger_ipi)(struct pt_regs *regs);
+extern int (*__debugger_bpt)(struct pt_regs *regs);
+extern int (*__debugger_sstep)(struct pt_regs *regs);
+extern int (*__debugger_iabr_match)(struct pt_regs *regs);
+extern int (*__debugger_dabr_match)(struct pt_regs *regs);
+extern int (*__debugger_fault_handler)(struct pt_regs *regs);
+
+#define DEBUGGER_BOILERPLATE(__NAME) \
+static inline int __NAME(struct pt_regs *regs) \
+{ \
+       if (unlikely(__ ## __NAME)) \
+               return __ ## __NAME(regs); \
+       return 0; \
+}
+
+DEBUGGER_BOILERPLATE(debugger)
+DEBUGGER_BOILERPLATE(debugger_ipi)
+DEBUGGER_BOILERPLATE(debugger_bpt)
+DEBUGGER_BOILERPLATE(debugger_sstep)
+DEBUGGER_BOILERPLATE(debugger_iabr_match)
+DEBUGGER_BOILERPLATE(debugger_dabr_match)
+DEBUGGER_BOILERPLATE(debugger_fault_handler)
+
+#ifdef CONFIG_XMON
+extern void xmon_init(int enable);
+#endif
+
+#else
+static inline int debugger(struct pt_regs *regs) { return 0; }
+static inline int debugger_ipi(struct pt_regs *regs) { return 0; }
+static inline int debugger_bpt(struct pt_regs *regs) { return 0; }
+static inline int debugger_sstep(struct pt_regs *regs) { return 0; }
+static inline int debugger_iabr_match(struct pt_regs *regs) { return 0; }
+static inline int debugger_dabr_match(struct pt_regs *regs) { return 0; }
+static inline int debugger_fault_handler(struct pt_regs *regs) { return 0; }
+#endif
+
+extern int set_dabr(unsigned long dabr);
+extern void print_backtrace(unsigned long *);
+extern void show_regs(struct pt_regs * regs);
+extern void flush_instruction_cache(void);
+extern void hard_reset_now(void);
+extern void poweroff_now(void);
+
+#ifdef CONFIG_6xx
+extern long _get_L2CR(void);
+extern long _get_L3CR(void);
+extern void _set_L2CR(unsigned long);
+extern void _set_L3CR(unsigned long);
+#else
+#define _get_L2CR()    0L
+#define _get_L3CR()    0L
+#define _set_L2CR(val) do { } while(0)
+#define _set_L3CR(val) do { } while(0)
+#endif
+
+extern void via_cuda_init(void);
+extern void pmac_nvram_init(void);
+extern void read_rtc_time(void);
+extern void pmac_find_display(void);
+extern void giveup_fpu(struct task_struct *);
+extern void enable_kernel_fp(void);
+extern void flush_fp_to_thread(struct task_struct *);
+extern void enable_kernel_altivec(void);
+extern void giveup_altivec(struct task_struct *);
+extern void load_up_altivec(struct task_struct *);
+extern void giveup_spe(struct task_struct *);
+extern void load_up_spe(struct task_struct *);
+extern int fix_alignment(struct pt_regs *);
+extern void cvt_fd(float *from, double *to, unsigned long *fpscr);
+extern void cvt_df(double *from, float *to, unsigned long *fpscr);
+
+#ifdef CONFIG_ALTIVEC
+extern void flush_altivec_to_thread(struct task_struct *);
+#else
+static inline void flush_altivec_to_thread(struct task_struct *t)
+{
+}
+#endif
+
+#ifdef CONFIG_SPE
+extern void flush_spe_to_thread(struct task_struct *);
+#else
+static inline void flush_spe_to_thread(struct task_struct *t)
+{
+}
+#endif
+
+extern int call_rtas(const char *, int, int, unsigned long *, ...);
+extern void cacheable_memzero(void *p, unsigned int nb);
+extern void *cacheable_memcpy(void *, const void *, unsigned int);
+extern int do_page_fault(struct pt_regs *, unsigned long, unsigned long);
+extern void bad_page_fault(struct pt_regs *, unsigned long, int);
+extern int die(const char *, struct pt_regs *, long);
+extern void _exception(int, struct pt_regs *, int, unsigned long);
+#ifdef CONFIG_BOOKE_WDT
+extern u32 booke_wdt_enabled;
+extern u32 booke_wdt_period;
+#endif /* CONFIG_BOOKE_WDT */
+
+/* EBCDIC -> ASCII conversion for [0-9A-Z] on iSeries */
+extern unsigned char e2a(unsigned char);
+
+struct device_node;
+extern void note_scsi_host(struct device_node *, void *);
+
+extern struct task_struct *__switch_to(struct task_struct *,
+       struct task_struct *);
+#define switch_to(prev, next, last)    ((last) = __switch_to((prev), (next)))
+
+struct thread_struct;
+extern struct task_struct *_switch(struct thread_struct *prev,
+                                  struct thread_struct *next);
+
+extern unsigned int rtas_data;
+
+/*
+ * Atomic exchange
+ *
+ * Changes the memory location '*ptr' to be val and returns
+ * the previous value stored there.
+ */
+static __inline__ unsigned long
+__xchg_u32(volatile void *p, unsigned long val)
+{
+       unsigned long prev;
+
+       __asm__ __volatile__(
+       EIEIO_ON_SMP
+"1:    lwarx   %0,0,%2 \n"
+       PPC405_ERR77(0,%2)
+"      stwcx.  %3,0,%2 \n\
+       bne-    1b"
+       ISYNC_ON_SMP
+       : "=&r" (prev), "=m" (*(volatile unsigned int *)p)
+       : "r" (p), "r" (val), "m" (*(volatile unsigned int *)p)
+       : "cc", "memory");
+
+       return prev;
+}
+
+#ifdef CONFIG_PPC64
+static __inline__ unsigned long
+__xchg_u64(volatile void *p, unsigned long val)
+{
+       unsigned long prev;
+
+       __asm__ __volatile__(
+       EIEIO_ON_SMP
+"1:    ldarx   %0,0,%2 \n"
+       PPC405_ERR77(0,%2)
+"      stdcx.  %3,0,%2 \n\
+       bne-    1b"
+       ISYNC_ON_SMP
+       : "=&r" (prev), "=m" (*(volatile unsigned long *)p)
+       : "r" (p), "r" (val), "m" (*(volatile unsigned long *)p)
+       : "cc", "memory");
+
+       return prev;
+}
+#endif
+
+/*
+ * This function doesn't exist, so you'll get a linker error
+ * if something tries to do an invalid xchg().
+ */
+extern void __xchg_called_with_bad_pointer(void);
+
+static __inline__ unsigned long
+__xchg(volatile void *ptr, unsigned long x, unsigned int size)
+{
+       switch (size) {
+       case 4:
+               return __xchg_u32(ptr, x);
+#ifdef CONFIG_PPC64
+       case 8:
+               return __xchg_u64(ptr, x);
+#endif
+       }
+       __xchg_called_with_bad_pointer();
+       return x;
+}
+
+#define xchg(ptr,x)                                                         \
+  ({                                                                        \
+     __typeof__(*(ptr)) _x_ = (x);                                          \
+     (__typeof__(*(ptr))) __xchg((ptr), (unsigned long)_x_, sizeof(*(ptr))); \
+  })
+
+#define tas(ptr) (xchg((ptr),1))
+
+/*
+ * Compare and exchange - if *p == old, set it to new,
+ * and return the old value of *p.
+ */
+#define __HAVE_ARCH_CMPXCHG    1
+
+static __inline__ unsigned long
+__cmpxchg_u32(volatile unsigned int *p, unsigned long old, unsigned long new)
+{
+       unsigned int prev;
+
+       __asm__ __volatile__ (
+       EIEIO_ON_SMP
+"1:    lwarx   %0,0,%2         # __cmpxchg_u32\n\
+       cmpw    0,%0,%3\n\
+       bne-    2f\n"
+       PPC405_ERR77(0,%2)
+"      stwcx.  %4,0,%2\n\
+       bne-    1b"
+       ISYNC_ON_SMP
+       "\n\
+2:"
+       : "=&r" (prev), "=m" (*p)
+       : "r" (p), "r" (old), "r" (new), "m" (*p)
+       : "cc", "memory");
+
+       return prev;
+}
+
+#ifdef CONFIG_PPC64
+static __inline__ unsigned long
+__cmpxchg_u64(volatile long *p, unsigned long old, unsigned long new)
+{
+       unsigned long prev;
+
+       __asm__ __volatile__ (
+       EIEIO_ON_SMP
+"1:    ldarx   %0,0,%2         # __cmpxchg_u64\n\
+       cmpd    0,%0,%3\n\
+       bne-    2f\n\
+       stdcx.  %4,0,%2\n\
+       bne-    1b"
+       ISYNC_ON_SMP
+       "\n\
+2:"
+       : "=&r" (prev), "=m" (*p)
+       : "r" (p), "r" (old), "r" (new), "m" (*p)
+       : "cc", "memory");
+
+       return prev;
+}
+#endif
+
+/* This function doesn't exist, so you'll get a linker error
+   if something tries to do an invalid cmpxchg().  */
+extern void __cmpxchg_called_with_bad_pointer(void);
+
+static __inline__ unsigned long
+__cmpxchg(volatile void *ptr, unsigned long old, unsigned long new,
+         unsigned int size)
+{
+       switch (size) {
+       case 4:
+               return __cmpxchg_u32(ptr, old, new);
+#ifdef CONFIG_PPC64
+       case 8:
+               return __cmpxchg_u64(ptr, old, new);
+#endif
+       }
+       __cmpxchg_called_with_bad_pointer();
+       return old;
+}
+
+#define cmpxchg(ptr,o,n)                                                \
+  ({                                                                    \
+     __typeof__(*(ptr)) _o_ = (o);                                      \
+     __typeof__(*(ptr)) _n_ = (n);                                      \
+     (__typeof__(*(ptr))) __cmpxchg((ptr), (unsigned long)_o_,          \
+                                   (unsigned long)_n_, sizeof(*(ptr))); \
+  })
+
+#ifdef CONFIG_PPC64
+/*
+ * We handle most unaligned accesses in hardware. On the other hand 
+ * unaligned DMA can be very expensive on some ppc64 IO chips (it does
+ * powers of 2 writes until it reaches sufficient alignment).
+ *
+ * Based on this we disable the IP header alignment in network drivers.
+ */
+#define NET_IP_ALIGN   0
+#endif
+
+#define arch_align_stack(x) (x)
+
+#endif /* __KERNEL__ */
+#endif /* __PPC_SYSTEM_H */
index 829481c0a9dc2fd6321a71a629045d3a5d94f4b1..79c1be3dfe619550e5a13b838a5d3bbe877551fe 100644 (file)
@@ -45,30 +45,21 @@ extern int __cpu_disable(void);
 extern void __cpu_die(unsigned int cpu);
 extern void cpu_die(void) __attribute__((noreturn));
 
-#define NO_PROC_ID             0xFF            /* No processor magic marker */
-#define PROC_CHANGE_PENALTY    20
-
 #define raw_smp_processor_id() (current_thread_info()->cpu)
 
 extern int __cpu_up(unsigned int cpu);
 
 extern int smp_hw_index[];
-#define hard_smp_processor_id() (smp_hw_index[smp_processor_id()])
-
-struct klock_info_struct {
-       unsigned long kernel_flag;
-       unsigned char akp;
-};
-
-extern struct klock_info_struct klock_info;
-#define KLOCK_HELD       0xffffffff
-#define KLOCK_CLEAR      0x0
+#define hard_smp_processor_id()        (smp_hw_index[smp_processor_id()])
+#define get_hard_smp_processor_id(cpu) (smp_hw_index[(cpu)])
 
 #endif /* __ASSEMBLY__ */
 
 #else /* !(CONFIG_SMP) */
 
 static inline void cpu_die(void) { }
+#define get_hard_smp_processor_id(cpu) 0
+#define hard_smp_processor_id() 0
 
 #endif /* !(CONFIG_SMP) */