GitHub/moto-9609/android_kernel_motorola_exynos9610.git
9 years agoarm64: mm: ensure patched kernel text is fetched from PoU
Will Deacon [Tue, 4 Aug 2015 16:49:36 +0000 (17:49 +0100)]
arm64: mm: ensure patched kernel text is fetched from PoU

The arm64 booting document requires that the bootloader has cleaned the
kernel image to the PoC. However, when a CPU re-enters the kernel due to
either a CPU hotplug "on" event or resuming from a low-power state (e.g.
cpuidle), the kernel text may in-fact be dirty at the PoU due to things
like alternative patching or even module loading.

Thanks to I-cache speculation with the MMU off, stale instructions could
be fetched prior to enabling the MMU, potentially leading to crashes
when executing regions of code that have been modified at runtime.

This patch addresses the issue by ensuring that the local I-cache is
invalidated immediately after a CPU has enabled its MMU but before
jumping out of the identity mapping. Any stale instructions fetched from
the PoC will then be discarded and refetched correctly from the PoU.
Patching kernel text executed prior to the MMU being enabled is
prohibited, so the early entry code will always be clean.

Reviewed-by: Mark Rutland <mark.rutland@arm.com>
Tested-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: alternatives: ensure secondary CPUs execute ISB after patching
Will Deacon [Tue, 4 Aug 2015 17:52:09 +0000 (18:52 +0100)]
arm64: alternatives: ensure secondary CPUs execute ISB after patching

In order to guarantee that the patched instruction stream is visible to
a CPU, that CPU must execute an isb instruction after any related cache
maintenance has completed.

The instruction patching routines in kernel/insn.c get this right for
things like jump labels and ftrace, but the alternatives patching omits
it entirely leaving secondary cores in a potential limbo between the old
and the new code.

This patch adds an isb following the secondary polling loop in the
altenatives patching.

Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: make ll/sc __cmpxchg_case_##name asm consistent
Mark Rutland [Tue, 4 Aug 2015 16:27:34 +0000 (17:27 +0100)]
arm64: make ll/sc __cmpxchg_case_##name asm consistent

The ll/sc __cmpxchg_case_##name assembly mostly uses symbolic names for
operands, but in a single case uses %2 to refer to what is otherwise
known as %[v]. This makes the code more painful to read than is
necessary.

Use %[v] instead.

Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: dma-mapping: Simplify pgprot handling
Robin Murphy [Fri, 31 Jul 2015 17:28:34 +0000 (18:28 +0100)]
arm64: dma-mapping: Simplify pgprot handling

Since __get_dma_pgprot() does The Right Thing(TM) in the non-coherent
case, and the non-cacheable alias for DMA buffers is private to the
kernel anyway, we can simplify things slightly and make the code more
readable by just using PAGE_KERNEL as the base pgprot.

Suggested-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Robin Murphy <robin.murphy@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: restore cpu suspend/resume functionality
Sudeep Holla [Fri, 31 Jul 2015 15:41:36 +0000 (16:41 +0100)]
arm64: restore cpu suspend/resume functionality

Commit 4b3dc9679cf7 ("arm64: force CONFIG_SMP=y and remove redundant #ifdefs")
accidentally retained code for !CONFIG_SMP in cpu_resume function. This
resulted in the hash index being zeroed in x7 after proper computation,
which is then used to get the cpu context pointer while resuming.

This patch removes the remanant code and restores back the cpu suspend/
resume functionality.

Fixes: 4b3dc9679cf7 ("arm64: force CONFIG_SMP=y and remove redundant #ifdefs")
Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>
Cc: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoARM64: PCI: do not enable resources on PROBE_ONLY systems
Lorenzo Pieralisi [Thu, 30 Jul 2015 13:13:59 +0000 (14:13 +0100)]
ARM64: PCI: do not enable resources on PROBE_ONLY systems

On ARM64 PROBE_ONLY PCI systems resources are not currently claimed,
therefore they can't be enabled since they do not have a valid
parent pointer; this in turn prevents enabling PCI devices on
ARM64 PROBE_ONLY systems, causing PCI devices initialization to
fail.

To solve this issue, resources must be claimed when devices are
added on PROBE_ONLY systems, which ensures that the resource hierarchy
is validated and the resource tree is sane, but this requires changes
in the ARM64 resource management that can affect adversely existing
PCI set-ups (claiming resources on !PROBE_ONLY systems might break
existing ARM64 PCI platform implementations).

As a temporary solution in preparation for a proper resources claiming
implementation in ARM64 core, to enable PCI PROBE_ONLY systems on ARM64,
this patch adds a pcibios_enable_device() arch implementation that
simply prevents enabling resources on PROBE_ONLY systems (mirroring ARM
behaviour).

This is always a safe thing to do because on PROBE_ONLY systems the
configuration space set-up can be considered immutable, and it is in
preparation of proper resource claiming that would finally validate
the PCI resources tree in the ARM64 arch implementation on PROBE_ONLY
systems.

For !PROBE_ONLY systems resources enablement in pcibios_enable_device()
on ARM64 is implemented as in current PCI core, leaving the behaviour
unchanged.

Signed-off-by: Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Bjorn Helgaas <bhelgaas@google.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: cmpxchg: truncate sub-word signed types before comparison
Will Deacon [Thu, 30 Jul 2015 18:19:43 +0000 (19:19 +0100)]
arm64: cmpxchg: truncate sub-word signed types before comparison

When performing a cmpxchg operation on a signed sub-word type (e.g. s8),
we need to ensure that the upper register bits of the "old" value used
for comparison are zeroed, otherwise we may erroneously fail the cmpxchg
which may even be interpreted as success by the caller (if the compiler
performs the truncation as part of its check). This has been observed
in mod_state, where negative values where causing problems with
this_cpu_cmpxchg.

This patch fixes the issue by explicitly casting 8-bit and 16-bit "old"
values using unsigned types in our cmpxchg wrappers. 32-bit types can be
left alone, since the underlying asm makes use of W registers in this
case.

Reported-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: alternative: put secondary CPUs into polling loop during patch
Will Deacon [Tue, 28 Jul 2015 18:07:28 +0000 (19:07 +0100)]
arm64: alternative: put secondary CPUs into polling loop during patch

When patching the kernel text with alternatives, we may end up patching
parts of the stop_machine state machine (e.g. atomic_dec_and_test in
ack_state) and consequently corrupt the instruction stream of any
secondary CPUs.

This patch passes the cpu_online_mask to stop_machine, forcing all of
the CPUs into our own callback which can place the secondary cores into
a dumb (but safe!) polling loop whilst the patching is carried out.

Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64/Documentation: clarify wording regarding memory below the Image
Ard Biesheuvel [Wed, 29 Jul 2015 11:30:39 +0000 (12:30 +0100)]
arm64/Documentation: clarify wording regarding memory below the Image

Clarify that the memory below the start of the image but inside the
region covered by the linear mapping has no special significance to
the kernel, and may be used by the firmware provided that it is marked
as reserved.

Also, fix up some whitespace errors.

Acked-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: lse: fix lse cmpxchg code indentation
Will Deacon [Wed, 29 Jul 2015 14:16:22 +0000 (15:16 +0100)]
arm64: lse: fix lse cmpxchg code indentation

For some reason, the ll/sc cmpxchg asm is all off to the left and
awkward to read in conjunction with the following (correctly indented)
LSE version.

This patch shifts the ll/sc code back to where it should be.

Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: remove redundant object file list
Jonas Rabenstein [Wed, 29 Jul 2015 11:13:20 +0000 (12:13 +0100)]
arm64: remove redundant object file list

Commit 4b3dc9679cf7 ("arm64: force CONFIG_SMP=y and remove redundant
#ifdefs") forces SMP on arm64. To build the necessary objects for SMP,
they were added to the arm64-obj-y rule in arch/arm64/kernel/Makefile,
without removing the arm64-obj-$(CONFIG_SMP) rule.

Remove redundant object file list depending on always-yes CONFIG_SMP in
arch/arm64/kernel/Makefile.

Signed-off-by: Jonas Rabenstein <jonas.rabenstein@studium.uni-erlangen.de>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: remove dead-code depending on CONFIG_UP_LATE_INIT
Jonas Rabenstein [Wed, 29 Jul 2015 11:07:57 +0000 (12:07 +0100)]
arm64: remove dead-code depending on CONFIG_UP_LATE_INIT

Commit 4b3dc9679cf7 ("arm64: force CONFIG_SMP=y and remove redundant
and therfore can not be selected anymore.

Remove dead #ifdef-block depending on UP_LATE_INIT in
arch/arm64/kernel/setup.c

Signed-off-by: Jonas Rabenstein <jonas.rabenstein@studium.uni-erlangen.de>
[will: kill do_post_cpus_up_work altogether]
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: pgtable: fix definition of pte_valid
Will Deacon [Tue, 28 Jul 2015 15:14:03 +0000 (16:14 +0100)]
arm64: pgtable: fix definition of pte_valid

pte_valid should check if the PTE_VALID bit (1 << 0) is set in the pte,
so fix the macro definition to use bitwise & instead of logical &&.

Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: spinlock: fix ll/sc unlock on big-endian systems
Will Deacon [Tue, 28 Jul 2015 13:48:00 +0000 (14:48 +0100)]
arm64: spinlock: fix ll/sc unlock on big-endian systems

When unlocking a spinlock, we perform a read-modify-write on the owner
ticket in order to increment it and store it back with release
semantics.

In the LL/SC case, we load the 16-bit ticket using a 32-bit load and
therefore store back the wrong halfword on a big-endian system,
corrupting the lock after the first unlock and killing the system dead.

This patch fixes the unlock code to use 16-bit accessors consistently.

Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: Use last level TLBI for user pte changes
Catalin Marinas [Wed, 18 Mar 2015 11:28:06 +0000 (11:28 +0000)]
arm64: Use last level TLBI for user pte changes

The flush_tlb_page() function is used on user address ranges when PTEs
(or PMDs/PUDs for huge pages) were changed (attributes or clearing). For
such cases, it is more efficient to invalidate only the last level of
the TLB with the "tlbi vale1is" instruction.

In the TLB shoot-down case, the TLB caching of the intermediate page
table levels (pmd, pud, pgd) is handled by __flush_tlb_pgtable() via the
__(pte|pmd|pud)_free_tlb() functions and it is not deferred to
tlb_finish_mmu() (as of commit 285994a62c80 - "arm64: Invalidate the TLB
corresponding to intermediate page table levels"). The tlb_flush()
function only needs to invalidate the TLB for the last level of page
tables; the __flush_tlb_range() function gains a fourth argument for
last level TLBI.

Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: Clean up __flush_tlb(_kernel)_range functions
Catalin Marinas [Fri, 24 Jul 2015 08:59:55 +0000 (09:59 +0100)]
arm64: Clean up __flush_tlb(_kernel)_range functions

This patch moves the MAX_TLB_RANGE check into the
flush_tlb(_kernel)_range functions directly to avoid the
undescore-prefixed definitions (and for consistency with a subsequent
patch).

Signed-off-by: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: mm: mark create_mapping as __init
Mark Rutland [Tue, 28 Jul 2015 09:31:06 +0000 (10:31 +0100)]
arm64: mm: mark create_mapping as __init

Currently create_mapping is marked with __ref, apparently because it
refers to early_alloc. However, create_mapping has no logic to prevent
erroneous use of early_alloc after it has been freed, and is only ever
called by __init functions anyway. Thus the __ref marker is misleading
and unnecessary.

Instead, this patch marks create_mapping as __init, resulting in
warnings if it is used from a a non __init functions, and allowing its
memory to be reclaimed.

Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: debug: rename enum debug_el to avoid symbol collision
Will Deacon [Mon, 27 Jul 2015 17:36:54 +0000 (18:36 +0100)]
arm64: debug: rename enum debug_el to avoid symbol collision

lib/list_sort.c defines a 'struct debug_el', where "el" is assumedly a
a contraction of "element". This conflicts with 'enum debug_el' in our
asm/debug-monitors.h header file, where "el" stands for Exception Level.

The result is build failure when targetting allmodconfig, so rename our
enum to 'dbg_active_el' to be slightly more explicit about what it is.

Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: mm: add __init section marker to free_initrd_mem
Wang Long [Mon, 27 Jul 2015 02:32:53 +0000 (03:32 +0100)]
arm64: mm: add __init section marker to free_initrd_mem

It is not needed after booting, this patch moves the
free_initrd_mem() function to the __init section.

This patch also make keep_initrd __initdata, to reduce kernel
size.

Signed-off-by: Wang Long <long.wanglong@huawei.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: elf: use cpuid_feature_extract_field for hwcap detection
Will Deacon [Mon, 27 Jul 2015 15:55:32 +0000 (16:55 +0100)]
arm64: elf: use cpuid_feature_extract_field for hwcap detection

cpuid_feature_extract_field takes care of the fiddly ID register
field sign-extension, so use that instead of rolling our own version.

Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: lse: use generic cpufeature detection for LSE atomics
Will Deacon [Mon, 27 Jul 2015 15:23:58 +0000 (16:23 +0100)]
arm64: lse: use generic cpufeature detection for LSE atomics

Rework the cpufeature detection to support ISAR0 and use that for
detecting the presence of LSE atomics.

Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: kconfig: group the v8.1 features together
Will Deacon [Mon, 27 Jul 2015 14:54:13 +0000 (15:54 +0100)]
arm64: kconfig: group the v8.1 features together

ARMv8 CPUs do not support any of the v8.1 features, so group them
together in Kconfig to make it clear that they're part of 8.1 and not
relevant to older cores.

Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: lse: rename ARM64_CPU_FEAT_LSE_ATOMICS for consistency
Will Deacon [Mon, 27 Jul 2015 13:11:55 +0000 (14:11 +0100)]
arm64: lse: rename ARM64_CPU_FEAT_LSE_ATOMICS for consistency

Other CPU features follow an 'ARM64_HAS_*' naming scheme, so do the same
for the LSE atomics.

Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: kconfig: select HAVE_CMPXCHG_LOCAL
Will Deacon [Fri, 29 May 2015 13:57:47 +0000 (14:57 +0100)]
arm64: kconfig: select HAVE_CMPXCHG_LOCAL

We implement an optimised cmpxchg_local macro, so let the kernel know.

Reviewed-by: Steve Capper <steve.capper@arm.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: atomic64_dec_if_positive: fix incorrect branch condition
Will Deacon [Fri, 29 May 2015 13:44:06 +0000 (14:44 +0100)]
arm64: atomic64_dec_if_positive: fix incorrect branch condition

If we attempt to atomic64_dec_if_positive on INT_MIN, we will underflow
and incorrectly decide that the original parameter was positive.

This patches fixes the broken condition code so that we handle this
corner case correctly.

Reviewed-by: Steve Capper <steve.capper@arm.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: atomics: implement atomic{,64}_cmpxchg using cmpxchg
Will Deacon [Thu, 4 Jun 2015 16:46:37 +0000 (17:46 +0100)]
arm64: atomics: implement atomic{,64}_cmpxchg using cmpxchg

We don't need duplicate cmpxchg implementations, so use cmpxchg to
implement atomic{,64}_cmpxchg, like we do for xchg already.

Reviewed-by: Steve Capper <steve.capper@arm.com>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: atomics: prefetch the destination word for write prior to stxr
Will Deacon [Fri, 29 May 2015 12:31:10 +0000 (13:31 +0100)]
arm64: atomics: prefetch the destination word for write prior to stxr

The cost of changing a cacheline from shared to exclusive state can be
significant, especially when this is triggered by an exclusive store,
since it may result in having to retry the transaction.

This patch makes use of prfm to prefetch cachelines for write prior to
ldxr/stxr loops when using the ll/sc atomic routines.

Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: atomics: tidy up common atomic{,64}_* macros
Will Deacon [Thu, 4 Jun 2015 15:41:36 +0000 (16:41 +0100)]
arm64: atomics: tidy up common atomic{,64}_* macros

The common (i.e. identical for ll/sc and lse) atomic macros in atomic.h
are needlessley different for atomic_t and atomic64_t.

This patch tidies up the definitions to make them consistent across the
two atomic types and factors out common code such as the add_unless
implementation based on cmpxchg.

Reviewed-by: Steve Capper <steve.capper@arm.com>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: cmpxchg: avoid memory barrier on comparison failure
Will Deacon [Fri, 29 May 2015 13:47:59 +0000 (14:47 +0100)]
arm64: cmpxchg: avoid memory barrier on comparison failure

cmpxchg doesn't require memory barrier semantics when the value
comparison fails, so make the barrier conditional on success.

Reviewed-by: Steve Capper <steve.capper@arm.com>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: cmpxchg: avoid "cc" clobber in ll/sc routines
Will Deacon [Fri, 29 May 2015 13:26:59 +0000 (14:26 +0100)]
arm64: cmpxchg: avoid "cc" clobber in ll/sc routines

We can perform the cmpxchg comparison using eor and cbnz which avoids
the "cc" clobber for the ll/sc case and consequently for the LSE case
where we may have to fall-back on the ll/sc code at runtime.

Reviewed-by: Steve Capper <steve.capper@arm.com>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: cmpxchg_dbl: patch in lse instructions when supported by the CPU
Will Deacon [Thu, 14 May 2015 17:05:50 +0000 (18:05 +0100)]
arm64: cmpxchg_dbl: patch in lse instructions when supported by the CPU

On CPUs which support the LSE atomic instructions introduced in ARMv8.1,
it makes sense to use them in preference to ll/sc sequences.

This patch introduces runtime patching of our cmpxchg_double primitives
so that the LSE casp instruction is used instead.

Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: cmpxchg: patch in lse instructions when supported by the CPU
Will Deacon [Thu, 23 Apr 2015 19:08:49 +0000 (20:08 +0100)]
arm64: cmpxchg: patch in lse instructions when supported by the CPU

On CPUs which support the LSE atomic instructions introduced in ARMv8.1,
it makes sense to use them in preference to ll/sc sequences.

This patch introduces runtime patching of our cmpxchg primitives so that
the LSE cas instruction is used instead.

Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: xchg: patch in lse instructions when supported by the CPU
Will Deacon [Tue, 31 Mar 2015 13:11:24 +0000 (14:11 +0100)]
arm64: xchg: patch in lse instructions when supported by the CPU

On CPUs which support the LSE atomic instructions introduced in ARMv8.1,
it makes sense to use them in preference to ll/sc sequences.

This patch introduces runtime patching of our xchg primitives so that
the LSE swp instruction (yes, you read right!) is used instead.

Reviewed-by: Steve Capper <steve.capper@arm.com>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: bitops: patch in lse instructions when supported by the CPU
Will Deacon [Thu, 12 Feb 2015 04:17:37 +0000 (04:17 +0000)]
arm64: bitops: patch in lse instructions when supported by the CPU

On CPUs which support the LSE atomic instructions introduced in ARMv8.1,
it makes sense to use them in preference to ll/sc sequences.

This patch introduces runtime patching of our bitops functions so that
LSE atomic instructions are used instead.

Reviewed-by: Steve Capper <steve.capper@arm.com>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: locks: patch in lse instructions when supported by the CPU
Will Deacon [Tue, 10 Feb 2015 03:03:15 +0000 (03:03 +0000)]
arm64: locks: patch in lse instructions when supported by the CPU

On CPUs which support the LSE atomic instructions introduced in ARMv8.1,
it makes sense to use them in preference to ll/sc sequences.

This patch introduces runtime patching of our locking functions so that
LSE atomic instructions are used for spinlocks and rwlocks.

Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: atomics: patch in lse instructions when supported by the CPU
Will Deacon [Tue, 3 Feb 2015 16:14:13 +0000 (16:14 +0000)]
arm64: atomics: patch in lse instructions when supported by the CPU

On CPUs which support the LSE atomic instructions introduced in ARMv8.1,
it makes sense to use them in preference to ll/sc sequences.

This patch introduces runtime patching of atomic_t and atomic64_t
routines so that the call-site for the out-of-line ll/sc sequences is
patched with an LSE atomic instruction when we detect that
the CPU supports it.

If binutils is not recent enough to assemble the LSE instructions, then
the ll/sc sequences are inlined as though CONFIG_ARM64_LSE_ATOMICS=n.

Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: introduce CONFIG_ARM64_LSE_ATOMICS as fallback to ll/sc atomics
Will Deacon [Tue, 3 Feb 2015 12:39:03 +0000 (12:39 +0000)]
arm64: introduce CONFIG_ARM64_LSE_ATOMICS as fallback to ll/sc atomics

In order to patch in the new atomic instructions at runtime, we need to
generate wrappers around the out-of-line exclusive load/store atomics.

This patch adds a new Kconfig option, CONFIG_ARM64_LSE_ATOMICS. which
causes our atomic functions to branch to the out-of-line ll/sc
implementations. To avoid the register spill overhead of the PCS, the
out-of-line functions are compiled with specific compiler flags to
force out-of-line save/restore of any registers that are usually
caller-saved.

Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: alternatives: add cpu feature for lse atomics
Will Deacon [Wed, 4 Feb 2015 12:17:55 +0000 (12:17 +0000)]
arm64: alternatives: add cpu feature for lse atomics

Add a CPU feature for the LSE atomic instructions, so that they can be
patched in at runtime when we detect that they are supported.

Reviewed-by: Steve Capper <steve.capper@arm.com>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: elf: advertise 8.1 atomic instructions as new hwcap
Will Deacon [Mon, 26 Jan 2015 18:46:19 +0000 (18:46 +0000)]
arm64: elf: advertise 8.1 atomic instructions as new hwcap

The ARM v8.1 architecture introduces new atomic instructions to the A64
instruction set for things like cmpxchg, so advertise their availability
to userspace using a hwcap.

Reviewed-by: Steve Capper <steve.capper@arm.com>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: atomics: move ll/sc atomics into separate header file
Will Deacon [Tue, 3 Feb 2015 11:26:53 +0000 (11:26 +0000)]
arm64: atomics: move ll/sc atomics into separate header file

In preparation for the Large System Extension (LSE) atomic instructions
introduced by ARM v8.1, move the current exclusive load/store (LL/SC)
atomics into their own header file.

Reviewed-by: Steve Capper <steve.capper@arm.com>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: cpufeature.h: add missing #include of kernel.h
Will Deacon [Thu, 30 Apr 2015 17:55:50 +0000 (18:55 +0100)]
arm64: cpufeature.h: add missing #include of kernel.h

cpufeature.h makes use of DECLARE_BITMAP, which in turn relies on the
BITS_TO_LONGS and DIV_ROUND_UP macros.

This patch includes kernel.h in cpufeature.h to prevent all users having
to do the same thing.

Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: rwlocks: don't fail trylock purely due to contention
Will Deacon [Wed, 22 Jul 2015 17:25:52 +0000 (18:25 +0100)]
arm64: rwlocks: don't fail trylock purely due to contention

STXR can fail for a number of reasons, so don't fail an rwlock trylock
operation simply because the STXR reported failure.

I'm not aware of any issues with the current code, but this makes it
consistent with spin_trylock and also other architectures (e.g. arch/arm).

Reported-by: Catalin Marinas <catalin.marinas@arm.com>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoMerge branch 'locking/arch-atomic' of git://git.kernel.org/pub/scm/linux/kernel/git...
Will Deacon [Mon, 27 Jul 2015 13:21:15 +0000 (14:21 +0100)]
Merge branch 'locking/arch-atomic' of git://git./linux/kernel/git/tip/tip into aarch64/for-next/core

Merge in PeterZ's logical atomic ops so that we can implement them in
our subsequent LSE atomics.

9 years agoatomic: Add simple atomic_t tests
Peter Zijlstra [Mon, 13 Jul 2015 10:55:58 +0000 (12:55 +0200)]
atomic: Add simple atomic_t tests

Add a few atomic_t tests, gets some compile coverage for the new
operations.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
9 years agoatomic: Replace atomic_{set,clear}_mask() usage
Peter Zijlstra [Thu, 23 Apr 2015 23:12:32 +0000 (01:12 +0200)]
atomic: Replace atomic_{set,clear}_mask() usage

Replace the deprecated atomic_{set,clear}_mask() usage with the now
ubiquous atomic_{or,andnot}() functions.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
9 years agoatomic: Collapse all atomic_{set,clear}_mask definitions
Peter Zijlstra [Thu, 23 Apr 2015 23:12:32 +0000 (01:12 +0200)]
atomic: Collapse all atomic_{set,clear}_mask definitions

Move the now generic definitions of atomic_{set,clear}_mask() into
linux/atomic.h to avoid endless and pointless repetition.

Also, provide an atomic_andnot() wrapper for those few archs that can
implement that.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
9 years agoatomic: Provide atomic_{or,xor,and}
Peter Zijlstra [Wed, 23 Apr 2014 17:32:50 +0000 (19:32 +0200)]
atomic: Provide atomic_{or,xor,and}

Implement atomic logic ops -- atomic_{or,xor,and}.

These will replace the atomic_{set,clear}_mask functions that are
available on some archs.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
9 years agotile: Provide atomic_{or,xor,and}
Chris Metcalf [Thu, 9 Jul 2015 20:38:17 +0000 (16:38 -0400)]
tile: Provide atomic_{or,xor,and}

Implement atomic logic ops -- atomic_{or,xor,and}.

For tilegx, these are relatively straightforward; the architecture
provides atomic "or" and "and", both 32-bit and 64-bit.  To support
xor we provide a loop using "cmpexch".

For the older 32-bit tilepro architecture, we have to extend
the set of low-level assembly routines to include 32-bit "and",
as well as all three 64-bit routines.  Somewhat confusingly,
some 32-bit versions are already used by the bitops inlines, with
parameter types appropriate for bitops, so we have to do a bit of
casting to match "int" to "unsigned long".

Signed-off-by: Chris Metcalf <cmetcalf@ezchip.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: http://lkml.kernel.org/r/1436474297-32187-1-git-send-email-cmetcalf@ezchip.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
9 years agoh8300: Provide atomic_{or,xor,and}
Peter Zijlstra [Fri, 10 Jul 2015 10:55:45 +0000 (12:55 +0200)]
h8300: Provide atomic_{or,xor,and}

Implement atomic logic ops -- atomic_{or,xor,and}

These will replace the atomic_{set,clear}_mask functions that are
available on some archs.

Also rework the atomic implementation in terms of CPP macros to avoid
the typical repetition -- I seem to have missed this arch the last
time around when I did that.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
9 years agofrv: Rewrite atomic implementation
Peter Zijlstra [Thu, 23 Apr 2015 22:49:20 +0000 (00:49 +0200)]
frv: Rewrite atomic implementation

Mostly complete rewrite of the FRV atomic implementation, instead of
using assembly files, use inline assembler.

The out-of-line CONFIG option makes a bit of a mess of things, but a
little CPP trickery gets that done too.

FRV already had the atomic logic ops but under a non standard name,
the reimplementation provides the generic names and provides the
intermediate form required for the bitops implementation.

The slightly inconsistent __atomic32_fetch_##op naming is because
__atomic_fetch_##op conlicts with GCC builtin functions.

The 64bit atomic ops use the inline assembly %Ln construct to access
the low word register (r+1), afaik this construct was not previously
used in the kernel and is completely undocumented, but I found it in
the FRV GCC code and it seems to work.

FRV had a non-standard definition of atomic_{clear,set}_mask() which
would work types other than atomic_t, the one user relying on that
(arch/frv/kernel/dma.c) got converted to use the new intermediate
form.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
9 years agox86: Provide atomic_{or,xor,and}
Peter Zijlstra [Wed, 23 Apr 2014 18:28:37 +0000 (20:28 +0200)]
x86: Provide atomic_{or,xor,and}

Implement atomic logic ops -- atomic_{or,xor,and}.

These will replace the atomic_{set,clear}_mask functions that are
available on some archs.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
9 years agos390: Provide atomic_{or,xor,and}
Peter Zijlstra [Wed, 23 Apr 2014 18:15:48 +0000 (20:15 +0200)]
s390: Provide atomic_{or,xor,and}

Implement atomic logic ops -- atomic_{or,xor,and}.

These will replace the atomic_{set,clear}_mask functions that are
available on some archs.

Acked-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
9 years agoxtensa: Provide atomic_{or,xor,and}
Peter Zijlstra [Wed, 23 Apr 2014 17:35:00 +0000 (19:35 +0200)]
xtensa: Provide atomic_{or,xor,and}

Implement atomic logic ops -- atomic_{or,xor,and}.

These will replace the atomic_{set,clear}_mask functions that are
available on some archs.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
9 years agosparc: Provide atomic_{or,xor,and}
Peter Zijlstra [Wed, 23 Apr 2014 17:40:25 +0000 (19:40 +0200)]
sparc: Provide atomic_{or,xor,and}

Implement atomic logic ops -- atomic_{or,xor,and}.

These will replace the atomic_{set,clear}_mask functions that are
available on some archs.

Acked-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
9 years agosh: Provide atomic_{or,xor,and}
Peter Zijlstra [Wed, 23 Apr 2014 17:44:20 +0000 (19:44 +0200)]
sh: Provide atomic_{or,xor,and}

Implement atomic logic ops -- atomic_{or,xor,and}.

These will replace the atomic_{set,clear}_mask functions that are
available on some archs.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
9 years agopowerpc: Provide atomic_{or,xor,and}
Peter Zijlstra [Wed, 23 Apr 2014 17:46:23 +0000 (19:46 +0200)]
powerpc: Provide atomic_{or,xor,and}

Implement atomic logic ops -- atomic_{or,xor,and}.

These will replace the atomic_{set,clear}_mask functions that are
available on some archs.

Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
9 years agoparisc: Provide atomic_{or,xor,and}
Peter Zijlstra [Wed, 23 Apr 2014 17:47:25 +0000 (19:47 +0200)]
parisc: Provide atomic_{or,xor,and}

Implement atomic logic ops -- atomic_{or,xor,and}.

These will replace the atomic_{set,clear}_mask functions that are
available on some archs.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
9 years agomn10300: Provide atomic_{or,xor,and}
Peter Zijlstra [Wed, 23 Apr 2014 17:50:20 +0000 (19:50 +0200)]
mn10300: Provide atomic_{or,xor,and}

Implement atomic logic ops -- atomic_{or,xor,and}.

These will replace the atomic_{set,clear}_mask functions that are
available on some archs.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
9 years agomips: Provide atomic_{or,xor,and}
Peter Zijlstra [Wed, 23 Apr 2014 17:51:36 +0000 (19:51 +0200)]
mips: Provide atomic_{or,xor,and}

Implement atomic logic ops -- atomic_{or,xor,and}.

These will replace the atomic_{set,clear}_mask functions that are
available on some archs.

Acked-by: Ralf Baechle <ralf@linux-mips.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
9 years agometag: Provide atomic_{or,xor,and}
Peter Zijlstra [Wed, 23 Apr 2014 17:53:39 +0000 (19:53 +0200)]
metag: Provide atomic_{or,xor,and}

Implement atomic logic ops -- atomic_{or,xor,and}.

These will replace the atomic_{set,clear}_mask functions that are
available on some archs.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
9 years agom68k: Provide atomic_{or,xor,and}
Peter Zijlstra [Wed, 23 Apr 2014 17:56:20 +0000 (19:56 +0200)]
m68k: Provide atomic_{or,xor,and}

Implement atomic logic ops -- atomic_{or,xor,and}.

These will replace the atomic_{set,clear}_mask functions that are
available on some archs.

Acked-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
9 years agom32r: Provide atomic_{or,xor,and}
Peter Zijlstra [Wed, 23 Apr 2014 17:57:49 +0000 (19:57 +0200)]
m32r: Provide atomic_{or,xor,and}

Implement atomic logic ops -- atomic_{or,xor,and}.

These will replace the atomic_{set,clear}_mask functions that are
available on some archs.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
9 years agoia64: Provide atomic_{or,xor,and}
Peter Zijlstra [Wed, 23 Apr 2014 18:00:01 +0000 (20:00 +0200)]
ia64: Provide atomic_{or,xor,and}

Implement atomic logic ops -- atomic_{or,xor,and}.

These will replace the atomic_{set,clear}_mask functions that are
available on some archs.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
9 years agohexagon: Provide atomic_{or,xor,and}
Peter Zijlstra [Wed, 23 Apr 2014 18:00:54 +0000 (20:00 +0200)]
hexagon: Provide atomic_{or,xor,and}

Implement atomic logic ops -- atomic_{or,xor,and}.

These will replace the atomic_{set,clear}_mask functions that are
available on some archs.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
9 years agoblackfin: Provide atomic_{or,xor,and}
Peter Zijlstra [Thu, 23 Apr 2015 19:44:42 +0000 (21:44 +0200)]
blackfin: Provide atomic_{or,xor,and}

Implement atomic logic ops -- atomic_{or,xor,and}.

These will replace the atomic_{set,clear}_mask functions that are
available on some archs.

TODO: use inline asm or at least asm macros to collapse the lot.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
9 years agoavr32: Provide atomic_{or,xor,and}
Peter Zijlstra [Wed, 23 Apr 2014 18:02:20 +0000 (20:02 +0200)]
avr32: Provide atomic_{or,xor,and}

Implement atomic logic ops -- atomic_{or,xor,and}.

These will replace the atomic_{set,clear}_mask functions that are
available on some archs.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
9 years agoarm64: Provide atomic_{or,xor,and}
Peter Zijlstra [Wed, 23 Apr 2014 18:03:13 +0000 (20:03 +0200)]
arm64: Provide atomic_{or,xor,and}

Implement atomic logic ops -- atomic_{or,xor,and}.

These will replace the atomic_{set,clear}_mask functions that are
available on some archs.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
9 years agoarm: Provide atomic_{or,xor,and}
Peter Zijlstra [Wed, 23 Apr 2014 18:04:39 +0000 (20:04 +0200)]
arm: Provide atomic_{or,xor,and}

Implement atomic logic ops -- atomic_{or,xor,and}.

These will replace the atomic_{set,clear}_mask functions that are
available on some archs.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
9 years agoarc: Provide atomic_{or,xor,and}
Peter Zijlstra [Wed, 23 Apr 2014 18:06:20 +0000 (20:06 +0200)]
arc: Provide atomic_{or,xor,and}

Implement atomic logic ops -- atomic_{or,xor,and}.

These will replace the atomic_{set,clear}_mask functions that are
available on some archs.

Acked-by: Vineet Gupta <vgupta@synopsys.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
9 years agoalpha: Provide atomic_{or,xor,and}
Peter Zijlstra [Wed, 23 Apr 2014 18:07:47 +0000 (20:07 +0200)]
alpha: Provide atomic_{or,xor,and}

Implement atomic logic ops -- atomic_{or,xor,and}.

These will replace the atomic_{set,clear}_mask functions that are
available on some archs.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
9 years agoatomic: Prepare generic atomic implementation for logic ops
Peter Zijlstra [Wed, 15 Jul 2015 13:47:25 +0000 (15:47 +0200)]
atomic: Prepare generic atomic implementation for logic ops

Clean up the #ifdef guards a bit to prepare for architectures to
supply their own logic ops.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
9 years agoarm64: include linux/types.h in asm/spinlock_types.h
Will Deacon [Mon, 27 Jul 2015 10:07:03 +0000 (11:07 +0100)]
arm64: include linux/types.h in asm/spinlock_types.h

Our ticket-based spinlock structures rely on a definition of u16, so
include linux/types.h explicitly to ensure the thing compiles.

Found by a module build failure in -next:

  arch/arm64/include/asm/spinlock_types.h:27:2: error: unknown type name 'u16'
  arch/arm64/include/asm/spinlock_types.h:28:2: error: unknown type name 'u16'
  arch/arm64/include/asm/spinlock_types.h:33:13: error: expected declaration specifiers or '...' before numeric constant
  include/linux/spinlock_types.h:21:2: error: unknown type name 'arch_spinlock_t'
  arch/arm64/include/asm/spinlock.h:34:35: error: unknown type name 'arch_spinlock_t'
  arch/arm64/include/asm/spinlock.h:65:37: error: unknown type name 'arch_spinlock_t'

Reported-by: Russell King <linux@arm.linux.org.uk>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64/BUG: Show explicit backtrace for WARNs
Dave P Martin [Fri, 24 Jul 2015 15:37:49 +0000 (16:37 +0100)]
arm64/BUG: Show explicit backtrace for WARNs

The generic slowpath WARN implementation prints a backtrace, but
the report_bug() based implementation does not, opting to print the
registers instead which is generally not as useful.

Ideally, report_bug() should be fixed to make the behaviour more
consistent, but in the meantime this patch generates a backtrace
directly from the arm64 backend instead so that this functionality
is not lost with the migration to report_bug().

As a side-effect, the backtrace will be outside the oops end
marker, but that's hard to avoid without modifying generic code.

This patch can go away if report_bug() grows the ability in the
future to generate a backtrace directly or call an arch hook at the
appropriate time.

Signed-off-by: Dave Martin <Dave.Martin@arm.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64/BUG: Use BRK instruction for generic BUG traps
Dave P Martin [Fri, 24 Jul 2015 15:37:48 +0000 (16:37 +0100)]
arm64/BUG: Use BRK instruction for generic BUG traps

Currently, the minimal default BUG() implementation from asm-
generic is used for arm64.

This patch uses the BRK software breakpoint instruction to generate
a trap instead, similarly to most other arches, with the generic
BUG code generating the dmesg boilerplate.

This allows bug metadata to be moved to a separate table and
reduces the amount of inline code at BUG and WARN sites.  This also
avoids clobbering any registers before they can be dumped.

To mitigate the size of the bug table further, this patch makes
use of the existing infrastructure for encoding addresses within
the bug table as 32-bit offsets instead of absolute pointers.
(Note that this limits the kernel size to 2GB.)

Traps are registered at arch_initcall time for aarch64, but BUG
has minimal real dependencies and it is desirable to be able to
generate bug splats as early as possible.  This patch redirects
all debug exceptions caused by BRK directly to bug_handler() until
the full debug exception support has been initialised.

Signed-off-by: Dave Martin <Dave.Martin@arm.com>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64/debug: Add missing #includes
Dave P Martin [Fri, 24 Jul 2015 15:37:47 +0000 (16:37 +0100)]
arm64/debug: Add missing #includes

<asm/debug-monitors.h> relies on <asm/ptrace.h>, but doesn't
declare this dependency.  This becomes a problem once
debug-monitors.h starts getting included all over the place to get
the BRK immedates.

The missing include of <asm/memory.h> (for UL()) in <asm/esr.h> is
also added.  The series no longer relies on this, but I spotted it
during development and it may as well get fixed.

No functional change.

Signed-off-by: Dave Martin <Dave.Martin@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64/debug: Simplify BRK insn opcode declarations
Dave P Martin [Fri, 24 Jul 2015 15:37:46 +0000 (16:37 +0100)]
arm64/debug: Simplify BRK insn opcode declarations

The way the KGDB_DYN_BRK_INS_BYTEx macros are declared is more
complex than it needs to be.  Also, the macros are only used in one
place, which is arch-specific anyway.

This patch refactors the macros to simplify them, and exposes an
argument so that we can have a single macro instead of 4.

As a side effect, this patch also fixes some anomalous spellings of
"KGDB".

These changes alter the compile types of some integer constants
that are harmless but trigger truncation warnings in gcc when
assigning to 32-bit variables.  This patch adds an explicit cast
for the affected cases.

Signed-off-by: Dave Martin <Dave.Martin@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64/debug: Move BRK ESR template macro into <asm/esr.h>
Dave P Martin [Fri, 24 Jul 2015 15:37:45 +0000 (16:37 +0100)]
arm64/debug: Move BRK ESR template macro into <asm/esr.h>

It makes sense to keep all the architectural exception syndrome
definitions in the same place.

Signed-off-by: Dave Martin <Dave.Martin@arm.com>
Acked-by: Mark Rutland <mark.rutland@arm.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64/debug: More consistent naming for the BRK ESR template macro
Dave P Martin [Fri, 24 Jul 2015 15:37:44 +0000 (16:37 +0100)]
arm64/debug: More consistent naming for the BRK ESR template macro

The naming of DBG_ESR_VAL_BRK is inconsistent with the way other
similar macros are named.

This patch makes the naming more consistent, and appends "64"
as a reminder that this ESR pattern only matches from AArch64
state.

Signed-off-by: Dave Martin <Dave.Martin@arm.com>
Acked-by: Mark Rutland <mark.rutland@arm.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64/debug: Eliminate magic number from ESR template definition
Dave P Martin [Fri, 24 Jul 2015 15:37:43 +0000 (16:37 +0100)]
arm64/debug: Eliminate magic number from ESR template definition

<asm/esr.h> has perfectly good constants for defining ESR values
already.  Let's use them.

Signed-off-by: Dave Martin <Dave.Martin@arm.com>
Acked-by: Mark Rutland <mark.rutland@arm.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64/debug: Mask off all reserved bits from generated ESR values
Dave P Martin [Fri, 24 Jul 2015 15:37:42 +0000 (16:37 +0100)]
arm64/debug: Mask off all reserved bits from generated ESR values

There are only 16 comment bits in a BRK instruction, which
correspond to ESR bits 15:0.  Bits 24:16 of the ESR are RES0,
and might have weird meanings in the future.

This code inserts 16 bits of comment in the ESR value instead of
20 (almost certainly a typo in the original code).

Signed-off-by: Dave Martin <Dave.Martin@arm.com>
Acked-by: Mark Rutland <mark.rutland@arm.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64/debug: Eliminate magic number for size of BRK instruction
Dave P Martin [Fri, 24 Jul 2015 15:37:41 +0000 (16:37 +0100)]
arm64/debug: Eliminate magic number for size of BRK instruction

The size of an A64 BRK instruction is the same as the size of all other
A64 instructions, because all A64 instructions are the same size.

BREAK_INSTR_SIZE is retained for readibility, but it should not be
an independent constant from AARCH64_INSN_SIZE.

Signed-off-by: Dave Martin <Dave.Martin@arm.com>
Acked-by: Mark Rutland <mark.rutland@arm.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: insn: use set_fixmap_offset to make it more clear
yalin wang [Fri, 24 Jul 2015 11:52:28 +0000 (12:52 +0100)]
arm64: insn: use set_fixmap_offset to make it more clear

A little change to patch_map() function,
use set_fixmap_offset() to make code more clear.

Signed-off-by: yalin wang <yalin.wang2010@gmail.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: efi: prefer AllocatePages() over efi_low_alloc() for vmlinux
Ard Biesheuvel [Fri, 24 Jul 2015 11:38:27 +0000 (12:38 +0100)]
arm64: efi: prefer AllocatePages() over efi_low_alloc() for vmlinux

When allocating memory for the kernel image, try the AllocatePages()
boot service to obtain memory at the preferred offset of
'dram_base + TEXT_OFFSET', and only revert to efi_low_alloc() if that
fails. This is the only way to allocate at the base of DRAM if DRAM
starts at 0x0, since efi_low_alloc() refuses to allocate at 0x0.

Tested-by: Haojian Zhuang <haojian.zhuang@linaro.org>
Reviewed-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: kernel: remove non-legit DT warnings when booting using ACPI
Sudeep Holla [Thu, 23 Jul 2015 17:28:26 +0000 (18:28 +0100)]
arm64: kernel: remove non-legit DT warnings when booting using ACPI

Since both CONFIG_ACPI and CONFIG_OF are enabled when booting using ACPI
tables on ARM64 platforms, we get few device tree warnings which are not
valid for ACPI boot. We can use of_have_populated_dt to check if the
device tree is populated or not before throwing out those errors.

This patch uses of_have_populated_dt to remove non legitimate device
tree warning when booting using ACPI tables.

Cc: Lorenzo Pieralisi <Lorenzo.Pieralisi@arm.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Acked-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: alternatives: add enable parameter to conditional asm macros
Will Deacon [Thu, 23 Jul 2015 12:51:46 +0000 (13:51 +0100)]
arm64: alternatives: add enable parameter to conditional asm macros

There are cases where we want to compile out both versions of an
alternative code block, so add an enable parameter to the new conditional
alternative assembly macros in the same way as alternative_insn.

Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: kernel: Add support for Privileged Access Never
James Morse [Wed, 22 Jul 2015 18:05:54 +0000 (19:05 +0100)]
arm64: kernel: Add support for Privileged Access Never

'Privileged Access Never' is a new arm8.1 feature which prevents
privileged code from accessing any virtual address where read or write
access is also permitted at EL0.

This patch enables the PAN feature on all CPUs, and modifies {get,put}_user
helpers temporarily to permit access.

This will catch kernel bugs where user memory is accessed directly.
'Unprivileged loads and stores' using ldtrb et al are unaffected by PAN.

Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: James Morse <james.morse@arm.com>
[will: use ALTERNATIVE in asm and tidy up pan_enable check]
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: Generalise msr_s/mrs_s operations
Suzuki K. Poulose [Wed, 22 Jul 2015 10:38:14 +0000 (11:38 +0100)]
arm64: Generalise msr_s/mrs_s operations

The system register encoding generated by sys_reg() works only
for MRS/MSR(Register) operations, as we hardcode Bit20 to 1 in
mrs_s/msr_s mask. This makes it unusable for generating instructions
accessing registers with Op0 < 2(e.g, PSTATE.x with Op0=0).

As per ARMv8 ARM, (Ref: ARMv8 ARM, Section: "System instruction class
encoding overview", C5.2, version:ARM DDI 0487A.f), the instruction
encoding reserves bits [20-19] for Op0.

This patch generalises the sys_reg, mrs_s and msr_s macros, so that
we could use them to access any of the supported system register.

Cc: James Morse <james.morse@arm.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Suzuki K. Poulose <suzuki.poulose@arm.com>
Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: kernel: Add optional CONFIG_ parameter to ALTERNATIVE()
James Morse [Tue, 21 Jul 2015 12:23:30 +0000 (13:23 +0100)]
arm64: kernel: Add optional CONFIG_ parameter to ALTERNATIVE()

Some uses of ALTERNATIVE() may depend on a feature that is disabled at
compile time by a Kconfig option. In this case the unused alternative
instructions waste space, and if the original instruction is a nop, it
wastes time and space.

This patch adds an optional 'config' option to ALTERNATIVE() and
alternative_insn that allows the compiler to remove both the original
and alternative instructions if the config option is not defined.

Suggested-by: Catalin Marinas <catalin.marinas@arm.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: James Morse <james.morse@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: kernel: Add min_field_value and use '>=' for feature detection
James Morse [Tue, 21 Jul 2015 12:23:29 +0000 (13:23 +0100)]
arm64: kernel: Add min_field_value and use '>=' for feature detection

When a new cpu feature is available, the cpu feature bits will have some
initial value, which is incremented when the feature is updated.
This patch changes 'register_value' to be 'min_field_value', and checks
the feature bits value (interpreted as a signed int) is greater than this
minimum.

Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: James Morse <james.morse@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: kernel: Add cpufeature 'enable' callback
James Morse [Tue, 21 Jul 2015 12:23:28 +0000 (13:23 +0100)]
arm64: kernel: Add cpufeature 'enable' callback

This patch adds an 'enable()' callback to cpu capability/feature
detection, allowing features that require some setup or configuration
to get this opportunity once the feature has been detected.

Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: James Morse <james.morse@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: kernel: Move config_sctlr_el1
James Morse [Tue, 21 Jul 2015 12:23:27 +0000 (13:23 +0100)]
arm64: kernel: Move config_sctlr_el1

Later patches need config_sctlr_el1 to set/clear bits in the sctlr_el1
register.

This patch moves this function into header a file.

Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: James Morse <james.morse@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: kvm: Adopt new alternative assembler macros
Daniel Thompson [Wed, 22 Jul 2015 11:21:04 +0000 (12:21 +0100)]
arm64: kvm: Adopt new alternative assembler macros

Convert the dynamic patching for ARM64_HAS_SYSREG_GIC_CPUIF over to
the newly added alternative assembler macros.

Acked-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Daniel Thompson <daniel.thompson@linaro.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: kernel: Adopt new alternative assembler macros
Daniel Thompson [Wed, 22 Jul 2015 11:21:03 +0000 (12:21 +0100)]
arm64: kernel: Adopt new alternative assembler macros

Convert the dynamic patching for ARM64_WORKAROUND_845719 over to
the newly added alternative assembler macros.

Signed-off-by: Daniel Thompson <daniel.thompson@linaro.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: mm: Adopt new alternative assembler macros
Daniel Thompson [Wed, 22 Jul 2015 11:21:02 +0000 (12:21 +0100)]
arm64: mm: Adopt new alternative assembler macros

Convert the dynamic patching for ARM64_WORKAROUND_CLEAN_CACHE over to
the newly added alternative assembler macros.

Signed-off-by: Daniel Thompson <daniel.thompson@linaro.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: alternative: Provide if/else/endif assembler macros
Daniel Thompson [Wed, 22 Jul 2015 11:21:01 +0000 (12:21 +0100)]
arm64: alternative: Provide if/else/endif assembler macros

The existing alternative_insn macro has some limitations that make it
hard to work with. In particular the fact it takes instructions from it
own macro arguments means it doesn't play very nicely with C pre-processor
macros because the macro arguments look like a string to the C
pre-processor. Workarounds are (probably) possible but things start to
look ugly.

Introduce an alternative set of macros that allows instructions to be
presented to the assembler as normal and switch everything over to the
new macros.

Signed-off-by: Daniel Thompson <daniel.thompson@linaro.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: kernel: Add cpuid_feature_extract_field() for 4bit sign extension
James Morse [Tue, 21 Jul 2015 12:23:26 +0000 (13:23 +0100)]
arm64: kernel: Add cpuid_feature_extract_field() for 4bit sign extension

Based on arch/arm/include/asm/cputype.h, this function does the
shifting and sign extension necessary when accessing cpu feature fields.

Signed-off-by: James Morse <james.morse@arm.com>
Suggested-by: Russell King <linux@arm.linux.org.uk>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: perf: condense event number maps
Mark Rutland [Tue, 21 Jul 2015 10:36:39 +0000 (11:36 +0100)]
arm64: perf: condense event number maps

Most of the cache events an architecture might support do not map well
to those provided by the ARM architecture, and as such most entries in
the event number maps are *_UNSUPPORTED. Unfortuantely as 0 is a valid
physical event identifier, the *_UNSUPPORTED macros expand to a non-zero
value and thus each unsupported event must be explicitly initialised as
such. This leads to large diffs when adding support for a new CPU, and
makes it difficult to spot the important information.

This patch follows arch/arm/ in making use of PERF_*_ALL_UNSUPPORTED
macros to initialise all entries to *_UNSUPPORTED before overriding this
for the specific events we actually support, resulting in a significant
source code reduction.

Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: hugetlb: remove paragraph about writing to FSF
Jisheng Zhang [Tue, 21 Jul 2015 14:43:58 +0000 (15:43 +0100)]
arm64: hugetlb: remove paragraph about writing to FSF

Remove paragraph about writing to the Free Software Foundation's
mailing address from GPL notice.

Signed-off-by: Jisheng Zhang <jszhang@marvell.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: dma-mapping: implement dma_get_sgtable()
Robin Murphy [Fri, 17 Jul 2015 15:58:21 +0000 (16:58 +0100)]
arm64: dma-mapping: implement dma_get_sgtable()

The default dma_common_get_sgtable() implementation relies on the CPU
address of the buffer being a regular lowmem address. This is not always
the case on arm64, since allocations from the various DMA pools may have
remapped vmalloc addresses, rendering the use of virt_to_page() invalid.

Fix this by providing our own implementation based on the fact that we
can safely derive a physical address from the DMA address in both cases.

CC: Jon Medhurst <tixy@linaro.org>
Signed-off-by: Robin Murphy <robin.murphy@arm.com>
[will: made static]
Signed-off-by: Will Deacon <will.deacon@arm.com>
9 years agoarm64: force CONFIG_SMP=y and remove redundant #ifdefs
Will Deacon [Fri, 29 May 2015 17:28:44 +0000 (18:28 +0100)]
arm64: force CONFIG_SMP=y and remove redundant #ifdefs

Nobody seems to be producing !SMP systems anymore, so this is just
becoming a source of kernel bugs, particularly if people want to use
coherent DMA with non-shared pages.

This patch forces CONFIG_SMP=y for arm64, removing a modest amount of
code in the process.

Signed-off-by: Will Deacon <will.deacon@arm.com>