[POWERPC] Add general support for mpc7448hpc2 (Taiga) platform

Add support for Freescale mpc7448 (Taiga) board support

Signed-off-by: Roy Zang <tie-fei.zang@freescale.com>
Signed-off-by: Paul Mackerras <paulus@samba.org>

[POWERPC] Correct the MAX_CONTEXT definition

When we increased the address space per process to 2^44 bytes, the
number of contexts that we could actually use reduced, but we forgot
to decrease the MAX_CONTEXT definition.  (Fortunately this would only
cause problems if we actually had more than 512k user processes
running.)  This patch corrects the definition.

Signed-off-by: Paul Mackerras <paulus@samba.org>

Merge branch 'for_paulus' of /linux/kernel/git/galak/powerpc

powerpc: minor cleanups for mpc86xx

* Remove duplicated cputable entry for 8641 (matches w/7448)
* Removed __init from function prototypes in mpc86xx.h
* Moved pci fixups into board specific code
* Moved mpc86xx_exclude_device to generic mpc86xx pci code
* Fixed sparse warnings in mpc86xx_smp.c
* Removed board specific header include from asm-powerpc/mpc86xx.h

Signed-off-by: Kumar Gala <galak@kernel.crashing.org>

[POWERPC] Make sure we select CONFIG_NEW_LEDS if ADB_PMU_LED is set

Signed-off-by: Paul Mackerras <paulus@samba.org>

[POWERPC] Simplify the code defining the 64-bit CPU features

Signed-off-by: Paul Mackerras <paulus@samba.org>

[POWERPC] powerpc: kconfig warning fix

arch/powerpc/Kconfig:420:warning: leading whitespace ignored

Stop doing that.

Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>

[POWERPC] Consolidate some of kernel/misc*.S

There were some common functions (mainly i/o).

Also some small white space cleanups and remove a couple of small unused
functions.

Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Paul Mackerras <paulus@samba.org>

[POWERPC] Remove unused function call_with_mmu_off

Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Paul Mackerras <paulus@samba.org>

[POWERPC] update asm-powerpc/time.h

If we ever build a combined kernel including iSeries, then this will
be needed.

Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Paul Mackerras <paulus@samba.org>

[POWERPC] Clean up it_lp_queue.h

No more StudlyCaps.
Remove from a couple of places it is no longer needed.
Use C style comments.

Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Paul Mackerras <paulus@samba.org>

[POWERPC] Skip the "copy down" of the kernel if it is already at zero.

This patch allows the kernel to recognized that it was loaded at zero
and the copy down of the image is unnecessary.  This is useful for
Simulators and kexec models.
On a typical 3.8 MiB vmlinux.strip this saves about 2.3 million instructions.

Signed-off-by: Paul Mackerras <paulus@samba.org>

[POWERPC] Add the use of the firmware soft-reset-nmi to kdump.

With this patch, kdump uses the firmware soft-reset NMI for two purposes:
1) Initiate the kdump (take a crash dump) by issuing a soft-reset.
2) Break a CPU out of a deadlock condition that is detected during kdump
processing.

When a soft-reset is initiated each CPU will enter
system_reset_exception() and set its corresponding bit in the global
bit-array cpus_in_sr then call die(). When die() finds the CPU's bit set
in cpu_in_sr crash_kexec() is called to initiate a crash dump. The first
CPU to enter crash_kexec() is called the "crashing CPU". All other CPUs
are "secondary CPUs". The secondary CPU's pass through to
crash_kexec_secondary() and sleep. The crashing CPU waits for all CPUs
to enter via soft-reset then boots the kdump kernel (see
crash_soft_reset_check())

When the system crashes due to a panic or exception, crash_kexec() is
called by panic() or die(). The crashing CPU sends an IPI to all other
CPUs to notify them of the pending shutdown. If a CPU is in a deadlock
or hung state with interrupts disabled, the IPI will not be delivered.
The result being, that the kdump kernel is not booted. This problem is
solved with the use of a firmware generated soft-reset. After the
crashing_cpu has issued the IPI, it waits for 10 sec for all CPUs to
enter crash_ipi_callback(). A CPU signifies its entry to
crash_ipi_callback() by setting its corresponding bit in the
cpus_in_crash bit array. After 10 sec, if one or more CPUs have not set
their bit in cpus_in_crash we assume that the CPU(s) is deadlocked. The
operator is then prompted to generate a soft-reset to break the
deadlock. Each CPU enters the soft reset handler as described above.

Two conditions must be handled at this point:
1) The system crashed because the operator generated a soft-reset. See
2) The system had crashed before the soft-reset was generated ( in the
case of a Panic or oops).

The first CPU to enter crash_kexec() uses the state of the kexec_lock to
determine this state. If kexec_lock is already held then condition 2 is
true and crash_kexec_secondary() is called, else; this CPU is flagged as
the crashing CPU, the kexec_lock is acquired and crash_kexec() proceeds
as described above.

Each additional CPUs responding to the soft-reset will pass through
crash_kexec() to kexec_secondary(). All secondary CPUs call
crash_ipi_callback() readying them self's for the shutdown. When ready
they clear their bit in cpus_in_sr. The crashing CPU waits in
kexec_secondary() until all other CPUs have cleared their bits in
cpus_in_sr. The kexec kernel boot is then started.

Signed-off-by: Haren Myneni <haren@us.ibm.com>
Signed-off-by: David Wilder <dwilder@us.ibm.com>
Signed-off-by: Paul Mackerras <paulus@samba.org>

[POWERPC] spufs: fix class0 interrupt assignment

The class zero interrupt handling for spus was confusing alignment and
error interrupts, so swap them.

Signed-off-by: Arnd Bergmann <arnd.bergmann@de.ibm.com>
Signed-off-by: Paul Mackerras <paulus@samba.org>

[POWERPC] spufs: fix memory hotplug dependency

spufs_base.c calls __add_pages, which depends on CONFIG_MEMORY_HOTPLUG.

Moved the selection of CONFIG_MEMORY_HOTPLUG from CONFIG_SPUFS_MMAP
to CONFIG_SPU_FS.

Signed-off-by: Geoff Levand <geoffrey.levand@am.sony.com>
Signed-off-by: Arnd Bergmann <arnd.bergmann@de.ibm.com>
Signed-off-by: Paul Mackerras <paulus@samba.org>

[POWERPC] spufs: fix MFC command queue purge

In the context save/restore code, the SPU MFC command queue purge
code has a bug:

static inline void wait_purge_complete(struct spu_state *csa, struct
       spu *spu)
{
    struct spu_priv2 __iomem *priv2 = spu->priv2;

    /* Save, Step 28:
     *     Poll MFC_CNTL[Ps] until value '11' is
     *     read
     *      (purge complete).
     */
    POLL_WHILE_FALSE(in_be64(&priv2->mfc_control_RW)
     & MFC_CNTL_PURGE_DMA_COMPLETE);
}

This will exit as soon as _one_ of the 2 bits that compose
MFC_CNTL_PURGE_DMA_COMPLETE is set, and one of them happens to be
"purge in progress"...  which means that we'll happily continue
restoring the MFC while it's being purged at the same time.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Arnd Bergmann <arnd.bergmann@de.ibm.com>
Signed-off-by: Paul Mackerras <paulus@samba.org>

[POWERPC] spufs: map mmio space as guarded into user space

This fixes a bug where we don't properly map SPE MMIO space as guarded,
causing various test cases to fail, probably due to write combining and other
niceties caused by the lack of the G bit.

Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>

[POWERPC] Enable XMON in cell_defconfig

Now that we have the udbg callbacks we can enable XMON by default.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Paul Mackerras <paulus@samba.org>

[POWERPC] Enable the RTAS udbg console on IBM Cell Blade

Enable the RTAS udbg console on IBM Cell Blade, this allows xmon
to work.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Paul Mackerras <paulus@samba.org>

[POWERPC] Add udbg support for RTAS console

Add udbg hooks for the RTAS console, based on the RTAS put-term-char
and get-term-char calls. Along with my previous patches, this should
enable debugging as soon as early_init_dt_scan_rtas() is called.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Paul Mackerras <paulus@samba.org>

[POWERPC] Setup RTAS values earlier, to enable rtas_call() earlier

Althought RTAS is instantiated when we enter the kernel, we can't actually
call into it until we know its entry point address. Currently we grab that
in rtas_initialize(), however that's quite late in the boot sequence.

To enable rtas_call() earlier, we can grab the RTAS entry etc. values while
we're scanning the flattened device tree. There's existing code to retrieve
the values from /chosen, however we don't store them there anymore, so remove
that code.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Paul Mackerras <paulus@samba.org>

[POWERPC] Move RTAS exports next to their declarations

Move RTAS exports next to their declarations.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Paul Mackerras <paulus@samba.org>

[POWERPC] Make rtas_call() safe if RTAS hasn't been initialised

Currently it's unsafe to call rtas_call() prior to rtas_initialize(). This
is because the rtas.entry value hasn't been setup and so we don't know
where to enter, but we just try anyway.

We can't do anything intelligent without rtas.entry, so if it's not set, just
return. Code that calls rtas_call() early needs to be aware that the call
might fail.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Paul Mackerras <paulus@samba.org>

[POWERPC] Setup the boot cpu's paca pointer in C rather than asm

There's no need to set the boot cpu paca in asm, so do it in C so us
mere mortals can understand it.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Paul Mackerras <paulus@samba.org>

[POWERPC] Make kexec_setup() a regular initcall

There's no reason kexec_setup() needs to be called explicitly from
setup_system(), it can just be a regular initcall.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Paul Mackerras <paulus@samba.org>

[POWERPC] Use ppc_md.hpte_insert() in htab_bolt_mapping()

With the ppc_md htab pointers setup earlier, we can use ppc_md.hpte_insert
in htab_bolt_mapping(), rather than deciding which version to call by hand.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Paul Mackerras <paulus@samba.org>

[POWERPC] powerpc: Initialise ppc_md htab pointers earlier

Initialise the ppc_md htab callbacks earlier, in the probe routines. This
allows us to call htab_finish_init() from htab_initialize(), and makes it
private to hash_utils_64.c. Move htab_finish_init() and make_bl() above
htab_initialize() to avoid forward declarations.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Paul Mackerras <paulus@samba.org>

[POWERPC] Export flat device tree via debugfs for debugging

If DEBUG is turned on in prom.c, export the flat device tree via debugfs.
This has been handy on several occasions.

To look at it:
 # mount -t debugfs none /sys/kernel/debug
 # od -a /sys/kernel/debug/powerpc/flat-device-tree
 and/or
 # dtc -fI dtb /sys/kernel/debug/powerpc/flat-device-tree -O dts

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Paul Mackerras <paulus@samba.org>

[POWERPC] Remove remaining iSeries debugger cruft

None of this seems to be necessary, so let's see if can remove it and not
break anything. Booted on iSeries & pSeries here.

NB. we don't remove the hvReleaseData, we just move it down so that the file
reads more clearly.

Signed-off-by: Michael Ellerman <michael@ellerman.id.au>
Signed-off-by: Paul Mackerras <paulus@samba.org>

[POWERPC] kdump: Reserve the existing TCE mappings left by the first kernel

During kdump boot, noticed some machines checkstop on dma protection
fault for ongoing DMA left in the first kernel. Instead of initializing
TCE entries in iommu_init() for the kdump boot, this patch fixes this
issue by walking through the each TCE table and checks whether the
entries are in use by the first kernel. If so, reserve those entries by
setting the corresponding bit in tbl->it_map such that these entries
will not be available for the kdump boot.

However it could be possible that all TCE entries might be used up due
to the driver bug that does continuous mapping. My observation is around
1700 TCE  entries are used on some systems (Ex: P4) at some point of
time during kdump boot and saving dump (either write into the disk or
sending to remote machine). Hence, this patch will make sure that
minimum of 2048 entries will be available such that kdump boot could be
successful in some cases.

Signed-off-by: Haren Myneni <haren@us.ibm.com>
Acked-by: Olof Johansson <olof@lixom.net>
Signed-off-by: Paul Mackerras <paulus@samba.org>

[POWERPC] Remove obsolete #include <linux/config.h>.

Signed-off-by: Jon Loeliger <jdl@freescale.com>
Signed-off-by: Paul Mackerras <paulus@samba.org>

[POWERPC] Remove redundant STD_MMU selection.

Signed-off-by: Jon Loeliger <jdl@freescale.com>
Signed-off-by: Paul Mackerras <paulus@samba.org>

[POWERPC] Move I8259 selection under MPC8641HPCN board

Signed-off-by: Jon Loeliger <jdl@freescale.com>
Signed-off-by: Paul Mackerras <paulus@samba.org>

[POWERPC] Remove redundant PPC_86XX check.

Signed-off-by: Jon Loeliger <jdl@freescale.com>
Signed-off-by: Paul Mackerras <paulus@samba.org>

[POWERPC] Don't access HID registers if running on a Hypervisor.

The following patch avoids accessing Hypervisor privilege HID
registers when running on a Hypervisor (MSR[HV]=0).

Signed-off-by: Jimi Xenidis <jimix@watson.ibm.com>
Signed-off-by: Paul Mackerras <paulus@samba.org>

[POWERPC] Convert powermac ide blink to new led infrastructure

This patch removes the old pmac ide led blink code and
adds generic LED subsystem support for the LED.

It maintains backward compatibility with the old
BLK_DEV_IDE_PMAC_BLINK Kconfig option which now
simply selects the new code and influences the
default trigger.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Paul Mackerras <paulus@samba.org>

[PATCH] fix static linking of NFS

Builds on ARM report link problems with common configurations like
statically linked NFS (for nfsroot).  The symptom is that __init
section code references __exit section code; that won't work since
the exit sections are discarded (since they can never be called).

The best fix for these particular cases would be an "__init_or_exit"
section annotation.

Signed-off-by: David Brownell <dbrownell@users.sourceforge.net>
Acked-by: Trond Myklebust <trond.myklebust@fys.uio.no>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Input: fix resetting name, phys and uniq when unregistering device

It should be done before calling class_device_unregister() because
it will destroy the device and free memory if there are no other
references to the device.

Signed-off-by: Dmitry Torokhov <dtor@mail.ru>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Revert "kbuild: fix make -rR breakage"

This reverts commit e5c44fd88c146755da6941d047de4d97651404a9.

Thanks to Daniel Ritz and Michal Piotrowski for noticing the problem.

Daniel says:

  "[The] reason is a recent change that made modules always shows as
   module.mod.  it breaks modprobe and probably many scripts..besides
   lsmod looking horrible

   stuff like this in modprobe.conf:
        install pcmcia_core /sbin/modprobe --ignore-install pcmcia_core; /sbin/modprobe pcmcia
   makes modprobe fork/exec endlessly calling itself...until oom
   interrupts it"

Signed-off-by: Linus Torvalds <torvalds@osdl.org>

Merge branch 'upstream-linus' of git://git./linux/kernel/git/mfasheh/ocfs2

* 'upstream-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mfasheh/ocfs2: (56 commits)
  [PATCH] fs/ocfs2/dlm/: cleanups
  ocfs2: fix compiler warnings in dlm_convert_lock_handler()
  ocfs2: dlm_print_one_mle() needs to be defined
  ocfs2: remove whitespace in dlmunlock.c
  ocfs2: move dlm work to a private work queue
  ocfs2: fix incorrect error returns
  ocfs2: tune down some noisy messages during dlm recovery
  ocfs2: display message before waiting for recovery to complete
  ocfs2: mlog in dlm_convert_lock_handler() should be ML_ERROR
  ocfs2: retry operations when a lock is marked in recovery
  ocfs2: use cond_resched() in dlm_thread()
  ocfs2: use GFP_NOFS in some dlm operations
  ocfs2: wait for recovery when starting lock mastery
  ocfs2: continue recovery when a dead node is encountered
  ocfs2: remove unneccesary spin_unlock() in dlm_remaster_locks()
  ocfs2: dlm_remaster_locks() should never exit without completing
  ocfs2: special case recovery lock in dlmlock_remote()
  ocfs2: pending mastery asserts and migrations should block each other
  ocfs2: temporarily disable automatic lock migration
  ocfs2: do not unconditionally purge the lockres in dlmlock_remote()
  ...

Merge master.kernel.org:/home/rmk/linux-2.6-arm

* master.kernel.org:/home/rmk/linux-2.6-arm:
  [ARM] 3657/1: S3C24XX: Documentation update of Overview.txt
  [ARM] Update mach-types
  [ARM] 3656/1: S3C2412: Add S3C2412 and S3C2413 documenation
  [ARM] 3654/1: add ajeco 1arm sbc support
  [ARM] fix drivers/mfd/ucb1x00-core.c IRQ probing bug
  [ARM] 3651/1: S3C24XX: Make arch list more detailed
  [ARM] 3650/1: S3C2412: Update s3c2410_defconfig
  [ARM] 3649/1: S3C24XX: Fix capitalisation of CPU on SMDK2440
  [ARM] 3612/1: make pci bus optional for ixp4xx platform
  [ARM] Remove MODE_(SVC|IRQ|FIQ|USR) and DEFAULT_FIQ
  [ARM] Remove save_lr/restore_pc macros
  [ARM] Remove partial non-v6 binutils compatibility
  [ARM] Remove LOADREGS macro
  [ARM] Remove RETINSTR macro

Merge master.kernel.org:/home/rmk/linux-2.6-serial

* master.kernel.org:/home/rmk/linux-2.6-serial:
  [SERIAL] 8250_pnp: add support for other Wacom tablets

[ARM] 3657/1: S3C24XX: Documentation update of Overview.txt

Patch from Ben Dooks

Update the list of supported devices, and remove the
changelog. Add SMDK2413 information.--

Signed-off-by: Ben Dooks <ben-linux@fluff.org>Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>

[ARM] Update mach-types

Usual mach-types update.

Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>

[PATCH] fs/ocfs2/dlm/: cleanups

This patch #if 0's the no longer used dlm_dump_lock_resources().

Since this makes dlmdebug.h empty, this patch also removes this header.

Additionally, the needlessly global dlm_is_node_recovered() is made
static.

Signed-off-by: Adrian Bunk <bunk@stusta.de>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: fix compiler warnings in dlm_convert_lock_handler()

We need to cast to unsigned long long.

Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: dlm_print_one_mle() needs to be defined

Fixes compile breakage.

Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: remove whitespace in dlmunlock.c

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: move dlm work to a private work queue

The work that is done can block for long periods of time and so is not
appropriate for keventd.

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: fix incorrect error returns

Use DLM_REJECTED instead of DLM_RECOVERING.

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: tune down some noisy messages during dlm recovery

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: display message before waiting for recovery to complete

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: mlog in dlm_convert_lock_handler() should be ML_ERROR

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: retry operations when a lock is marked in recovery

Before checking for a nonexistent lock, make sure the lockres is not marked
RECOVERING. The caller will just retry and the state should be fixed up when
recovery completes.

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: use cond_resched() in dlm_thread()

yield() does not yield.  cond_resched() does.

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: use GFP_NOFS in some dlm operations

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: wait for recovery when starting lock mastery

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: continue recovery when a dead node is encountered

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: remove unneccesary spin_unlock() in dlm_remaster_locks()

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: dlm_remaster_locks() should never exit without completing

We cannot restart recovery. Once we begin to recover a node, keep the state
of the recovery intact and follow through, regardless of any other node
deaths that may occur.

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: special case recovery lock in dlmlock_remote()

If the previous master of the recovery lock dies, let calc_usage take it
down completely and let the caller completely redo the dlmlock() call.
Otherwise, there will never be an opportunity to re-master the lockres and
recovery wont be able to progress.

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: pending mastery asserts and migrations should block each other

Use the existing structure for blocking migrations when ASTs are pending to
achieve the same result. If we can catch the assert before it goes on the
wire, just cancel it and let the migration continue.

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: temporarily disable automatic lock migration

Now we never change the owner of a lock resource until unmount or node
death. This will be re-enabled once some issues in the algorithm used have
been resolved.

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: do not unconditionally purge the lockres in dlmlock_remote()

In dlmlock_remote(), do not call purge_lockres until the lock resource
actually changes. otherwise, the mastery info on the lockres will go away
underneath the caller.

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: increase backoff before waiting for recovery

When mastering non-recovery lock resources, additional time was frequently
needed to allow the disk heartbeat to catch up with the network timeout. the
recovery lock resource is time critical and avoids this path.

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: have dlm_pre_master_reco_lockres() ignore dead nodes

Recovery will spin in dlm_pre_master_reco_lockres if we do not ignore
timed-out network responses from dead nodes.

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: give the dlm dirty list a reference on the lockres

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: teach dlm_restart_lock_mastery() to wait on recovery

Change behavior of dlm_restart_lock_mastery() when a node goes down.  Dump
all responses that have been collected and start over.

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: gracefully handle stale create_lock messages.

This is an error on the sending side, so gracefully error out on the
receiving end.

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: update lvb immediately during recovery

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: do not send master requests to localhost

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: purge lockres' sooner

Immediately purge a lockress that the local node is not the master of.

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: dump mismatching migrated lvbs before BUG()

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: make dlm recovery finalization 2 stage

Makes it easier for the recovery process to deal with node death.

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: dlm recovery / lockres reference count fix

Take a reference on lockres structures while they are on the recovery list.

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: better error handling during assert master message

handle errors during lock assert master by either killing self or other node

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: dump lockres info before we BUG() on a bad reference

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: do LVB puts in place

Don't wait until the AST will be fired to do the LVB copy into the lock
resource.

Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: mle ref count debugging

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: allow for an assert message during lock mastery

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: take mle reference during migration

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: properly initialize the mle structure

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: detach mle from heartbeat events

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: mle ref counting fixes

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: better mle debugging

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: clean up recovery related messages

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: handle network errors during recovery

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: only recover one dead node at a time

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: Better tracking for recovery state changes

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: Fix empty lvb check

The check for an empty lvb should check the entire buffer not just the first
byte.

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: fix inverted logic in dlm_is_node_dead

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: recheck lockres master before sending an unlock request.

Recovery may have happened and it may now be mastered locally.

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: add a small delay after a failed migration

Otherwise we risk starving other threads.

Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: silence a compile warning in dlm_alloc_pagevec()

Reported by Andrew Morton.

Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

[PATCH] ocfs2: Alloc at least a page for the DLM hash

The OCFS2 DLM allocates a number of pages for a hash to lookup locks.
There was a bug where a PAGE_SIZE bigger than the hash size (eg, 64K
pages) would result in zero pages allocated.

Signed-off-by: Joel Becker <joel.becker@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: allocate lockres hash pages in an array

This allows us to have a hash table greater than a single page which greatly
improves dlm performance on some tests.

Signed-off-by: Daniel Phillips <phillips@google.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: inline dlm_lockres_get()

It's called on every lookup so this might help performance a bit.

Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

[PATCH] Clean up ocfs2 hash probe and make it faster

Signed-Off-By: Daniel Phillips <phillips@google.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: calculate lockid hash values outside of the spinlock

Fixes a performance bug - pointed out by Andrew.

Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>

ocfs2: move lockres qstr next to hlist_node structure

Gains us a bit of performance on loads which heavily hit the lockres hash.
Patch suggested by Daniel Phillips <phillips@google.com>.

Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>