Pull xen updates from Juergen Gross:
"Xen features and fixes:
- a series from Boris Ostrovsky adding support for booting Linux as
Xen PVH guest
- a series from Juergen Gross streamlining the xenbus driver
- a series from Paul Durrant adding support for the new device model
hypercall
- several small corrections"
* tag 'for-linus-4.11-rc0-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/xen/tip:
xen/privcmd: add IOCTL_PRIVCMD_RESTRICT
xen/privcmd: Add IOCTL_PRIVCMD_DM_OP
xen/privcmd: return -ENOTTY for unimplemented IOCTLs
xen: optimize xenbus driver for multiple concurrent xenstore accesses
xen: modify xenstore watch event interface
xen: clean up xenbus internal headers
xenbus: Neaten xenbus_va_dev_error
xen/pvh: Use Xen's emergency_restart op for PVH guests
xen/pvh: Enable CPU hotplug
xen/pvh: PVH guests always have PV devices
xen/pvh: Initialize grant table for PVH guests
xen/pvh: Make sure we don't use ACPI_IRQ_MODEL_PIC for SCI
xen/pvh: Bootstrap PVH guest
xen/pvh: Import PVH-related Xen public interfaces
xen/x86: Remove PVH support
x86/boot/32: Convert the 32-bit pgtable setup code from assembly to C
xen/manage: correct return value check on xenbus_scanf()
x86/xen: Fix APIC id mismatch warning on Intel
xen/netback: set default upper limit of tx/rx queues to 8
xen/netfront: set default upper limit of tx/rx queues to 8
Pull x86 platform updates from Ingo Molnar:
"Misc platform updates: SGI UV4 support additions, intel-mid Merrifield
enhancements and purge of old code"
* 'x86-platform-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (21 commits)
x86/platform/UV/NMI: Fix uneccessary kABI breakage
x86/platform/UV: Clean up the NMI code to match current coding style
x86/platform/UV: Ensure uv_system_init is called when necessary
x86/platform/UV: Initialize PCH GPP_D_0 NMI Pin to be NMI source
x86/platform/UV: Verify NMI action is valid, default is standard
x86/platform/UV: Add basic CPU NMI health check
x86/platform/UV: Add Support for UV4 Hubless NMIs
x86/platform/UV: Add Support for UV4 Hubless systems
x86/platform/UV: Clean up the UV APIC code
x86/platform/intel-mid: Move watchdog registration to arch_initcall()
x86/platform/intel-mid: Don't shadow error code of mp_map_gsi_to_irq()
x86/platform/intel-mid: Allocate RTC interrupt for Merrifield
x86/ioapic: Return suitable error code in mp_map_gsi_to_irq()
x86/platform/UV: Fix 2 socket config problem
x86/platform/UV: Fix panic with missing UVsystab support
x86/platform/intel-mid: Enable RTC on Intel Merrifield
x86/platform/intel: Remove PMIC GPIO block support
x86/platform/intel-mid: Make intel_scu_device_register() static
x86/platform/intel-mid: Enable GPIO keys on Merrifield
x86/platform/intel-mid: Get rid of duplication of IPC handler
...
Pull x86 mm updates from Ingo Molnar:
"A laundry list of changes: KASAN improvements/fixes for ptdump, a
self-test fix, PAT cleanup and wbinvd() avoidance, removal of stale
code and documentation updates"
* 'x86-mm-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86/mm/ptdump: Add address marker for KASAN shadow region
x86/mm/ptdump: Optimize check for W+X mappings for CONFIG_KASAN=y
x86/mm/pat: Use rb_entry()
x86/mpx: Re-add MPX to selftests Makefile
x86/mm: Remove CONFIG_DEBUG_NX_TEST
x86/mm/cpa: Avoid wbinvd() for PREEMPT
x86/mm: Improve documentation for low-level device I/O functions
Pull x86 microcode updates from Ingo Molnar:
"The main changes are further simplification and unification of the
code between the AMD and Intel microcode loaders, plus other
simplifications - by Borislav Petkov"
* 'x86-microcode-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86/microcode/AMD: Remove struct cont_desc.eq_id
x86/microcode/AMD: Remove AP scanning optimization
x86/microcode/AMD: Simplify saving from initrd
x86/microcode/AMD: Unify load_ucode_amd_ap()
x86/microcode/AMD: Check patch level only on the BSP
x86/microcode: Remove local vendor variable
x86/microcode/AMD: Use find_microcode_in_initrd()
x86/microcode/AMD: Get rid of global this_equiv_id
x86/microcode: Decrease CPUID use
x86/microcode/AMD: Rework container parsing
x86/microcode/AMD: Extend the container struct
x86/microcode/AMD: Shorten function parameter's name
x86/microcode/AMD: Clean up find_equiv_id()
x86/microcode: Convert to bare minimum MSR accessors
x86/MSR: Carve out bare minimum accessors
Pull x86 fpu updates from Ingo Molnar:
"The main changes relate to fixes between (lack of) CPUID and FPU
detection that should only affect old or weird CPUs, by Andy
Lutomirski"
* 'x86-fpu-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86/fpu: Fix the "Giving up, no FPU found" test
x86/fpu: Fix CPUID-less FPU detection
x86/fpu: Fix "x86/fpu: Legacy x87 FPU detected" message
x86/cpu: Re-apply forced caps every time CPU caps are re-read
x86/cpu: Factor out application of forced CPU caps
x86/cpu: Add X86_FEATURE_CPUID
x86/fpu/xstate: Move XSAVES state init to a function
Pull x86 cpufeature updates from Ingo Molnar:
"The main changes in this cycle were related to enable ring-3
MONITOR/MWAIT instructions support on supported CPUs, by Grzegorz
Andrejczuk and Piotr Luc"
* 'x86-cpufeature-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86/cpufeature: Move RING3MWAIT feature to avoid conflicts
x86/cpufeature: Enable RING3MWAIT for Knights Mill
x86/cpufeature: Enable RING3MWAIT for Knights Landing
x86/cpufeature: Add RING3MWAIT to CPU features
x86/elf: Add HWCAP2 to expose ring 3 MONITOR/MWAIT
x86/msr: Add MSR_MISC_FEATURE_ENABLES and RING3MWAIT bit
x86/cpufeature: Add AVX512_VPOPCNTDQ feature
Pull x86 boot updates from Ingo Molnar:
"Misc updates:
- fix e820 error handling
- convert page table setup code from assembly to C
- fix kexec environment bug
- ... plus small cleanups"
* 'x86-boot-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86/kconfig: Remove misleading note regarding hibernation and KASLR
x86/boot: Fix KASLR and memmap= collision
x86/e820/32: Fix e820_search_gap() error handling on x86-32
x86/boot/32: Convert the 32-bit pgtable setup code from assembly to C
x86/e820: Make e820_search_gap() static and remove unused variables
Pull locking updates from Ingo Molnar:
"The main changes in this cycle were:
- Implement wraparound-safe refcount_t and kref_t types based on
generic atomic primitives (Peter Zijlstra)
- Improve and fix the ww_mutex code (Nicolai Hähnle)
- Add self-tests to the ww_mutex code (Chris Wilson)
- Optimize percpu-rwsems with the 'rcuwait' mechanism (Davidlohr
Bueso)
- Micro-optimize the current-task logic all around the core kernel
(Davidlohr Bueso)
- Tidy up after recent optimizations: remove stale code and APIs,
clean up the code (Waiman Long)
- ... plus misc fixes, updates and cleanups"
* 'locking-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (50 commits)
fork: Fix task_struct alignment
locking/spinlock/debug: Remove spinlock lockup detection code
lockdep: Fix incorrect condition to print bug msgs for MAX_LOCKDEP_CHAIN_HLOCKS
lkdtm: Convert to refcount_t testing
kref: Implement 'struct kref' using refcount_t
refcount_t: Introduce a special purpose refcount type
sched/wake_q: Clarify queue reinit comment
sched/wait, rcuwait: Fix typo in comment
locking/mutex: Fix lockdep_assert_held() fail
locking/rtmutex: Flip unlikely() branch to likely() in __rt_mutex_slowlock()
locking/rwsem: Reinit wake_q after use
locking/rwsem: Remove unnecessary atomic_long_t casts
jump_labels: Move header guard #endif down where it belongs
locking/atomic, kref: Implement kref_put_lock()
locking/ww_mutex: Turn off __must_check for now
locking/atomic, kref: Avoid more abuse
locking/atomic, kref: Use kref_get_unless_zero() more
locking/atomic, kref: Kill kref_sub()
locking/atomic, kref: Add kref_read()
locking/atomic, kref: Add KREF_INIT()
...
Pull scheduler updates from Ingo Molnar:
"The main changes in this (fairly busy) cycle were:
- There was a class of scheduler bugs related to forgetting to update
the rq-clock timestamp which can cause weird and hard to debug
problems, so there's a new debug facility for this: which uncovered
a whole lot of bugs which convinced us that we want to keep the
debug facility.
(Peter Zijlstra, Matt Fleming)
- Various cputime related updates: eliminate cputime and use u64
nanoseconds directly, simplify and improve the arch interfaces,
implement delayed accounting more widely, etc. - (Frederic
Weisbecker)
- Move code around for better structure plus cleanups (Ingo Molnar)
- Move IO schedule accounting deeper into the scheduler plus related
changes to improve the situation (Tejun Heo)
- ... plus a round of sched/rt and sched/deadline fixes, plus other
fixes, updats and cleanups"
* 'sched-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (85 commits)
sched/core: Remove unlikely() annotation from sched_move_task()
sched/autogroup: Rename auto_group.[ch] to autogroup.[ch]
sched/topology: Split out scheduler topology code from core.c into topology.c
sched/core: Remove unnecessary #include headers
sched/rq_clock: Consolidate the ordering of the rq_clock methods
delayacct: Include <uapi/linux/taskstats.h>
sched/core: Clean up comments
sched/rt: Show the 'sched_rr_timeslice' SCHED_RR timeslice tuning knob in milliseconds
sched/clock: Add dummy clear_sched_clock_stable() stub function
sched/cputime: Remove generic asm headers
sched/cputime: Remove unused nsec_to_cputime()
s390, sched/cputime: Remove unused cputime definitions
powerpc, sched/cputime: Remove unused cputime definitions
s390, sched/cputime: Make arch_cpu_idle_time() to return nsecs
ia64, sched/cputime: Remove unused cputime definitions
ia64: Convert vtime to use nsec units directly
ia64, sched/cputime: Move the nsecs based cputime headers to the last arch using it
sched/cputime: Remove jiffies based cputime
sched/cputime, vtime: Return nsecs instead of cputime_t to account
sched/cputime: Complete nsec conversion of tick based accounting
...
Pull RAS updates from Ingo Molnar:
"The main changes in this cycle were:
- Assign notifier chain priorities for all RAS related handlers to
make the ordering explicit (Borislav Petkov)
- Improve the AMD MCA banks sysfs output (Yazen Ghannam)
- Various cleanups and restructuring of the x86 RAS code (Borislav
Petkov)"
* 'ras-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
x86/ras, EDAC, acpi: Assign MCE notifier handlers a priority
x86/ras: Get rid of mce_process_work()
EDAC/mce/amd: Dump TSC value
EDAC/mce/amd: Unexport amd_decode_mce()
x86/ras/amd/inj: Change dependency
x86/ras: Flip the TSC-adding logic
x86/ras/amd: Make sysfs names of banks more user-friendly
x86/ras/therm_throt: Do not log a fake MCE for thermal events
x86/ras/inject: Make it depend on X86_LOCAL_APIC=y
Pull EFI updates from Ingo Molnar:
"The main changes in this cycle were:
- Changes to the EFI init code to establish whether secure boot
authentication was performed at boot time. (Josh Boyer, David
Howells)
- Wire up the UEFI memory attributes table for x86. This eliminates
any runtime memory regions that are both writable and executable,
on recent firmware versions. (Sai Praneeth)
- Move the BGRT init code to an earlier stage so that we can still
use efi_mem_reserve(). (Dave Young)
- Preserve debug symbols in the ARM/arm64 UEFI stub (Ard Biesheuvel)
- Code deduplication work and various other cleanups (Lukas Wunner)
- ... plus various other fixes and cleanups"
* 'efi-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
efi/libstub: Make file I/O chunking x86-specific
efi: Print the secure boot status in x86 setup_arch()
efi: Disable secure boot if shim is in insecure mode
efi: Get and store the secure boot status
efi: Add SHIM and image security database GUID definitions
arm/efi: Allow invocation of arbitrary runtime services
x86/efi: Allow invocation of arbitrary runtime services
efi/libstub: Preserve .debug sections after absolute relocation check
efi/x86: Add debug code to print cooked memmap
efi/x86: Move the EFI BGRT init code to early init code
efi: Use typed function pointers for the runtime services table
efi/esrt: Fix typo in pr_err() message
x86/efi: Add support for EFI_MEMORY_ATTRIBUTES_TABLE
efi: Introduce the EFI_MEM_ATTR bit and set it from the memory attributes table
efi: Make EFI_MEMORY_ATTRIBUTES_TABLE initialization common across all architectures
x86/efi: Deduplicate efi_char16_printk()
efi: Deduplicate efi_file_size() / _read() / _close()
Pull timer updates from Thomas Gleixner:
"Nothing exciting, just the usual pile of fixes, updates and cleanups:
- A bunch of clocksource driver updates
- Removal of CONFIG_TIMER_STATS and the related /proc file
- More posix timer slim down work
- A scalability enhancement in the tick broadcast code
- Math cleanups"
* 'timers-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (23 commits)
hrtimer: Catch invalid clockids again
math64, tile: Fix build failure
clocksource/drivers/arm_arch_timer:: Mark cyclecounter __ro_after_init
timerfd: Protect the might cancel mechanism proper
timer_list: Remove useless cast when printing
time: Remove CONFIG_TIMER_STATS
clocksource/drivers/arm_arch_timer: Work around Hisilicon erratum 161010101
clocksource/drivers/arm_arch_timer: Introduce generic errata handling infrastructure
clocksource/drivers/arm_arch_timer: Remove fsl-a008585 parameter
clocksource/drivers/arm_arch_timer: Add dt binding for hisilicon-161010101 erratum
clocksource/drivers/ostm: Add renesas-ostm timer driver
clocksource/drivers/ostm: Document renesas-ostm timer DT bindings
clocksource/drivers/tcb_clksrc: Use 32 bit tcb as sched_clock
clocksource/drivers/gemini: Add driver for the Cortina Gemini
clocksource: add DT bindings for Cortina Gemini
clockevents: Add a clkevt-of mechanism like clksrc-of
tick/broadcast: Reduce lock cacheline contention
timers: Omit POSIX timer stuff from task_struct when disabled
x86/timer: Make delay() work during early bootup
delay: Add explanation of udelay() inaccuracy
...
Recently a new dm_op[1] hypercall was added to Xen to provide a mechanism
for restricting device emulators (such as QEMU) to a limited set of
hypervisor operations, and being able to audit those operations in the
kernel of the domain in which they run.
This patch adds IOCTL_PRIVCMD_DM_OP as gateway for __HYPERVISOR_dm_op.
NOTE: There is no requirement for user-space code to bounce data through
locked memory buffers (as with IOCTL_PRIVCMD_HYPERCALL) since
privcmd has enough information to lock the original buffers
directly.
[1] http://xenbits.xen.org/gitweb/?p=xen.git;a=commit;h=524a98c2
Signed-off-by: Paul Durrant <paul.durrant@citrix.com>
Acked-by: Stefano Stabellini <sstabellini@kernel.org>
Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
Commit:
a33d331761 ("x86/CPU/AMD: Fix Bulldozer topology")
restored the initial approach we had with the Fam15h topology of
enumerating CU (Compute Unit) threads as cores. And this is still
correct - they're beefier than HT threads but still have some
shared functionality.
Our current approach has a problem with the Mad Max Steam game, for
example. Yves Dionne reported a certain "choppiness" while playing on
v4.9.5.
That problem stems most likely from the fact that the CU threads share
resources within one CU and when we schedule to a thread of a different
compute unit, this incurs latency due to migrating the working set to a
different CU through the caches.
When the thread siblings mask mirrors that aspect of the CUs and
threads, the scheduler pays attention to it and tries to schedule within
one CU first. Which takes care of the latency, of course.
Reported-by: Yves Dionne <yves.dionne@gmail.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Cc: <stable@vger.kernel.org> # 4.9
Cc: Brice Goglin <Brice.Goglin@inria.fr>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Yazen Ghannam <yazen.ghannam@amd.com>
Link: http://lkml.kernel.org/r/20170205105022.8705-1-bp@alien8.de
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Introduce ELF_HWCAP2 variable for x86 and reserve its bit 0 to expose the
ring 3 MONITOR/MWAIT.
HWCAP variables contain bitmasks which can be used by userspace
applications to detect which instruction sets are supported by CPU. On x86
architecture information about CPU capabilities can be checked via CPUID
instructions, unfortunately presence of ring 3 MONITOR/MWAIT feature cannot
be checked this way. ELF_HWCAP cannot be used as well, because on x86 it is
set to CPUID[1].EDX which means that all bits are reserved there.
HWCAP2 approach was chosen because it reuses existing solution present
in other architectures, so only minor modifications are required to the
kernel and userspace applications. When ELF_HWCAP2 is defined
kernel maps it to AT_HWCAP2 during the start of the application.
This way the ring 3 MONITOR/MWAIT feature can be detected using getauxval()
API in a simple and fast manner. ELF_HWCAP2 type is u32 to be consistent
with x86 ELF_HWCAP type.
Signed-off-by: Grzegorz Andrejczuk <grzegorz.andrejczuk@intel.com>
Cc: Piotr.Luc@intel.com
Cc: dave.hansen@linux.intel.com
Link: http://lkml.kernel.org/r/1484918557-15481-3-git-send-email-grzegorz.andrejczuk@intel.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Define new MSR MISC_FEATURE_ENABLES (0x140).
On supported CPUs if bit 1 of this MSR is set, then calling MONITOR and
MWAIT instructions outside of ring 0 will not cause invalid-opcode
exception.
The MSR MISC_FEATURE_ENABLES is not yet documented in the SDM. Here is the
relevant documentation:
Hex Dec Name Scope
140H 320 MISC_FEATURE_ENABLES Thread
0 Reserved
1 If set to 1, the MONITOR and MWAIT instructions do not
cause invalid-opcode exceptions when executed with CPL > 0
or in virtual-8086 mode. If MWAIT is executed when CPL > 0
or in virtual-8086 mode, and if EAX indicates a C-state
other than C0 or C1, the instruction operates as if EAX
indicated the C-state C1.
63:2 Reserved
Signed-off-by: Grzegorz Andrejczuk <grzegorz.andrejczuk@intel.com>
Cc: Piotr.Luc@intel.com
Cc: dave.hansen@linux.intel.com
Link: http://lkml.kernel.org/r/1484918557-15481-2-git-send-email-grzegorz.andrejczuk@intel.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Having tracepoints to the MSR accessors makes them unsuitable for early
microcode loading: think 32-bit before paging is enabled and us chasing
pointers to test whether a tracepoint is enabled or not. Results in a
reliable triple fault.
Convert to the bare ones.
Signed-off-by: Borislav Petkov <bp@suse.de>
Link: http://lkml.kernel.org/r/20170120202955.4091-4-bp@alien8.de
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Add __rdmsr() and __wrmsr() which *only* read and write an MSR with
exception handling. Those are going to be used in early code, like the
microcode loader, which cannot stomach tracing code piggybacking on the
MSR operation.
While at it, get rid of __native_write_msr_notrace().
Signed-off-by: Borislav Petkov <bp@suse.de>
Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/20170120202955.4091-3-bp@alien8.de
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Vector population count instructions for dwords and qwords are going to be
available in future Intel Xeon & Xeon Phi processors. Bit 14 of
CPUID[level:0x07, ECX] indicates that the instructions are supported by a
processor.
The specification can be found in the Intel Software Developer Manual (SDM)
and in the Instruction Set Extensions Programming Reference (ISE).
Populate the feature bit and clear it when xsave is disabled.
Signed-off-by: Piotr Luc <piotr.luc@intel.com>
Reviewed-by: Borislav Petkov <bp@suse.de>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: kvm@vger.kernel.org
Cc: Radim Krčmář <rkrcmar@redhat.com>
Link: http://lkml.kernel.org/r/20170110173403.6010-2-piotr.luc@intel.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
There are a handful of callers to save_stack_trace_tsk() and
show_stack() which try to unwind the stack of a task other than current.
In such cases, it's remotely possible that the task is running on one
CPU while the unwinder is reading its stack from another CPU, causing
the unwinder to see stack corruption.
These cases seem to be mostly harmless. The unwinder has checks which
prevent it from following bad pointers beyond the bounds of the stack.
So it's not really a bug as long as the caller understands that
unwinding another task will not always succeed.
In such cases, it's possible that the unwinder may read a KASAN-poisoned
region of the stack. Account for that by using READ_ONCE_NOCHECK() when
reading the stack of another task.
Use READ_ONCE() when reading the stack of the current task, since KASAN
warnings can still be useful for finding bugs in that case.
Reported-by: Dmitry Vyukov <dvyukov@google.com>
Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Brian Gerst <brgerst@gmail.com>
Cc: Dave Jones <davej@codemonkey.org.uk>
Cc: Denys Vlasenko <dvlasenk@redhat.com>
Cc: H. Peter Anvin <hpa@zytor.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Miroslav Benes <mbenes@suse.cz>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/4c575eb288ba9f73d498dfe0acde2f58674598f1.1483978430.git.jpoimboe@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Since on Intel we're required to do CPUID(1) first, before reading
the microcode revision MSR, let's add a special helper which does the
required steps so that we don't forget to do them next time, when we
want to read the microcode revision.
Signed-off-by: Borislav Petkov <bp@suse.de>
Link: http://lkml.kernel.org/r/20170109114147.5082-4-bp@alien8.de
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
In commit 6290602709 ("mm: add PageWaiters indicating tasks are
waiting for a page bit") Nick Piggin made our page locking no longer
unconditionally touch the hashed page waitqueue, which not only helps
performance in general, but is particularly helpful on NUMA machines
where the hashed wait queues can bounce around a lot.
However, the "clear lock bit atomically and then test the waiters bit"
sequence turns out to be much more expensive than it needs to be,
because you get a nasty stall when trying to access the same word that
just got updated atomically.
On architectures where locking is done with LL/SC, this would be trivial
to fix with a new primitive that clears one bit and tests another
atomically, but that ends up not working on x86, where the only atomic
operations that return the result end up being cmpxchg and xadd. The
atomic bit operations return the old value of the same bit we changed,
not the value of an unrelated bit.
On x86, we could put the lock bit in the high bit of the byte, and use
"xadd" with that bit (where the overflow ends up not touching other
bits), and look at the other bits of the result. However, an even
simpler model is to just use a regular atomic "and" to clear the lock
bit, and then the sign bit in eflags will indicate the resulting state
of the unrelated bit #7.
So by moving the PageWaiters bit up to bit #7, we can atomically clear
the lock bit and test the waiters bit on x86 too. And architectures
with LL/SC (which is all the usual RISC suspects), the particular bit
doesn't matter, so they are fine with this approach too.
This avoids the extra access to the same atomic word, and thus avoids
the costly stall at page unlock time.
The only downside is that the interface ends up being a bit odd and
specialized: clear a bit in a byte, and test the sign bit. Nick doesn't
love the resulting name of the new primitive, but I'd rather make the
name be descriptive and very clear about the limitation imposed by
trying to work across all relevant architectures than make it be some
generic thing that doesn't make the odd semantics explicit.
So this introduces the new architecture primitive
clear_bit_unlock_is_negative_byte();
and adds the trivial implementation for x86. We have a generic
non-optimized fallback (that just does a "clear_bit()"+"test_bit(7)"
combination) which can be overridden by any architecture that can do
better. According to Nick, Power has the same hickup x86 has, for
example, but some other architectures may not even care.
All these optimizations mean that my page locking stress-test (which is
just executing a lot of small short-lived shell scripts: "make test" in
the git source tree) no longer makes our page locking look horribly bad.
Before all these optimizations, just the unlock_page() costs were just
over 3% of all CPU overhead on "make test". After this, it's down to
0.66%, so just a quarter of the cost it used to be.
(The difference on NUMA is bigger, but there this micro-optimization is
likely less noticeable, since the big issue on NUMA was not the accesses
to 'struct page', but the waitqueue accesses that were already removed
by Nick's earlier commit).
Acked-by: Nick Piggin <npiggin@gmail.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Bob Peterson <rpeterso@redhat.com>
Cc: Steven Whitehouse <swhiteho@redhat.com>
Cc: Andrew Lutomirski <luto@kernel.org>
Cc: Andreas Gruenbacher <agruenba@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
