Skip to content

cloud-kernel
cloud-kernel

openanolis / cloud-kernel
1 年多 前同步成功

通知 160
Star 36
Fork 7
cloud-kernel
cloud-kernel
切换分支/标签
  1. 29 5月, 2009 5 次提交
  3. 16 5月, 2009 1 次提交
    • J
      x86: Fix performance regression caused by paravirt_ops on native kernels · b4ecc126
      由 Jeremy Fitzhardinge 提交于 
      Xiaohui Xin and some other folks at Intel have been looking into what's
behind the performance hit of paravirt_ops when running native.

It appears that the hit is entirely due to the paravirtualized
spinlocks introduced by:

 | commit 8efcbab6
 | Date:   Mon Jul 7 12:07:51 2008 -0700
 |
 |     paravirt: introduce a "lock-byte" spinlock implementation

The extra call/return in the spinlock path is somehow
causing an increase in the cycles/instruction of somewhere around 2-7%
(seems to vary quite a lot from test to test).  The working theory is
that the CPU's pipeline is getting upset about the
call->call->locked-op->return->return, and seems to be failing to
speculate (though I haven't seen anything definitive about the precise
reasons).  This doesn't entirely make sense, because the performance
hit is also visible on unlock and other operations which don't involve
locked instructions.  But spinlock operations clearly swamp all the
other pvops operations, even though I can't imagine that they're
nearly as common (there's only a .05% increase in instructions
executed).

If I disable just the pv-spinlock calls, my tests show that pvops is
identical to non-pvops performance on native (my measurements show that
it is actually about .1% faster, but Xiaohui shows a .05% slowdown).

Summary of results, averaging 10 runs of the "mmperf" test, using a
no-pvops build as baseline:

		nopv		Pv-nospin	Pv-spin
CPU cycles	100.00%		99.89%		102.18%
instructions	100.00%		100.10%		100.15%
CPI		100.00%		99.79%		102.03%
cache ref	100.00%		100.84%		100.28%
cache miss	100.00%		90.47%		88.56%
cache miss rate	100.00%		89.72%		88.31%
branches	100.00%		99.93%		100.04%
branch miss	100.00%		103.66%		107.72%
branch miss rt	100.00%		103.73%		107.67%
wallclock	100.00%		99.90%		102.20%

The clear effect here is that the 2% increase in CPI is
directly reflected in the final wallclock time.

(The other interesting effect is that the more ops are
out of line calls via pvops, the lower the cache access
and miss rates.  Not too surprising, but it suggests that
the non-pvops kernel is over-inlined.  On the flipside,
the branch misses go up correspondingly...)

So, what's the fix?

Paravirt patching turns all the pvops calls into direct calls, so
_spin_lock etc do end up having direct calls.  For example, the compiler
generated code for paravirtualized _spin_lock is:

<_spin_lock+0>:		mov    %gs:0xb4c8,%rax
<_spin_lock+9>:		incl   0xffffffffffffe044(%rax)
<_spin_lock+15>:	callq  *0xffffffff805a5b30
<_spin_lock+22>:	retq

The indirect call will get patched to:
<_spin_lock+0>:		mov    %gs:0xb4c8,%rax
<_spin_lock+9>:		incl   0xffffffffffffe044(%rax)
<_spin_lock+15>:	callq <__ticket_spin_lock>
<_spin_lock+20>:	nop; nop		/* or whatever 2-byte nop */
<_spin_lock+22>:	retq

One possibility is to inline _spin_lock, etc, when building an
optimised kernel (ie, when there's no spinlock/preempt
instrumentation/debugging enabled).  That will remove the outer
call/return pair, returning the instruction stream to a single
call/return, which will presumably execute the same as the non-pvops
case.  The downsides arel 1) it will replicate the
preempt_disable/enable code at eack lock/unlock callsite; this code is
fairly small, but not nothing; and 2) the spinlock definitions are
already a very heavily tangled mass of #ifdefs and other preprocessor
magic, and making any changes will be non-trivial.

The other obvious answer is to disable pv-spinlocks.  Making them a
separate config option is fairly easy, and it would be trivial to
enable them only when Xen is enabled (as the only non-default user).
But it doesn't really address the common case of a distro build which
is going to have Xen support enabled, and leaves the open question of
whether the native performance cost of pv-spinlocks is worth the
performance improvement on a loaded Xen system (10% saving of overall
system CPU when guests block rather than spin).  Still it is a
reasonable short-term workaround.

[ Impact: fix pvops performance regression when running native ]
Analysed-by: N"Xin Xiaohui" <xiaohui.xin@intel.com>
Analysed-by: N"Li Xin" <xin.li@intel.com>
Analysed-by: N"Nakajima Jun" <jun.nakajima@intel.com>
Signed-off-by: NJeremy Fitzhardinge <jeremy.fitzhardinge@citrix.com>
Acked-by: NH. Peter Anvin <hpa@zytor.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Xen-devel <xen-devel@lists.xensource.com>
LKML-Reference: <4A0B62F7.5030802@goop.org>
[ fixed the help text ]
Signed-off-by: NIngo Molnar <mingo@elte.hu>
      b4ecc126
  5. 22 4月, 2009 1 次提交
  7. 17 4月, 2009 1 次提交
    • Y
      x86/irq: mark NUMA_MIGRATE_IRQ_DESC broken · ca713c2a
      由 Yinghai Lu 提交于 
      It causes crash on system with lots of cards with MSI-X
when irq_balancer enabled...

The patches fixing it were both complex and fragile, according
to Eric they were also doing quite dangerous things to the
hardware.

Instead we now have patches that solve this problem via static
NUMA node mappings - not dynamic allocation and balancing.

The patches are much simpler than this method but are still too
large outside of the merge window, so we mark the dynamic balancer
as broken for now, and queue up the new approach for v2.6.31.

[ Impact: deactivate broken kernel feature ]
Reported-by: NSuresh Siddha <suresh.b.siddha@intel.com>
Signed-off-by: NYinghai Lu <yinghai@kernel.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
LKML-Reference: <49E68C41.4020801@kernel.org>
Signed-off-by: NIngo Molnar <mingo@elte.hu>
      ca713c2a
  9. 08 4月, 2009 1 次提交
  11. 07 4月, 2009 1 次提交
    • D
      x86, intel-iommu: fix X2APIC && !ACPI build failure · f7d7f866
      由 David Woodhouse 提交于 
      This build failure:

| drivers/pci/dmar.c:47: error: expected ‘=’, ‘,’, ‘;’, ‘asm’ or ‘__attribute__’ before ‘dmar_tbl_size’
| drivers/pci/dmar.c:62: warning: ‘struct acpi_dmar_device_scope’ declared inside parameter list
| drivers/pci/dmar.c:62: warning: its scope is only this definition or declaration, which is probably not what you want

Triggers due to this commit:

  d0b03bd1: x2apic/intr-remap: decouple interrupt remapping from x2apic

Which exposed a pre-existing but dormant fragility of the 'select X86_X2APIC'
it moved around and turned that fragility into a build failure.

Replace it with a proper 'depends on' construct.
Signed-off-by: NDavid Woodhouse <David.Woodhouse@intel.com>
LKML-Reference: <1239084280.22733.404.camel@macbook.infradead.org>
Signed-off-by: NIngo Molnar <mingo@elte.hu>
      f7d7f866
  13. 04 4月, 2009 1 次提交
  15. 01 4月, 2009 1 次提交
  17. 30 3月, 2009 1 次提交
  19. 26 3月, 2009 2 次提交
  21. 17 3月, 2009 1 次提交
  23. 13 3月, 2009 3 次提交
  25. 11 3月, 2009 2 次提交
  27. 27 2月, 2009 1 次提交
  29. 25 2月, 2009 1 次提交
    • A
      x86, mce, cmci: factor out threshold interrupt handler · b2762686
      由 Andi Kleen 提交于 
      Impact: cleanup; preparation for feature

The mce_amd_64 code has an own private MC threshold vector with an own
interrupt handler. Since Intel needs a similar handler
it makes sense to share the vector because both can not
be active at the same time.

I factored the common APIC handler code into a separate file which can
be used by both the Intel or AMD MC code.

This is needed for the next patch which adds an Intel specific
CMCI handler.

This patch should be a nop for AMD, it just moves some code
around.
Signed-off-by: NAndi Kleen <ak@linux.intel.com>
Signed-off-by: NH. Peter Anvin <hpa@zytor.com>
      b2762686
  31. 24 2月, 2009 1 次提交
    • T
      bootmem: clean up arch-specific bootmem wrapping · c1329375
      由 Tejun Heo 提交于 
      Impact: cleaner and consistent bootmem wrapping

By setting CONFIG_HAVE_ARCH_BOOTMEM_NODE, archs can define
arch-specific wrappers for bootmem allocation.  However, this is done
a bit strangely in that only the high level convenience macros can be
changed while lower level, but still exported, interface functions
can't be wrapped.  This not only is messy but also leads to strange
situation where alloc_bootmem() does what the arch wants it to do but
the equivalent __alloc_bootmem() call doesn't although they should be
able to be used interchangeably.

This patch updates bootmem such that archs can override / wrap the
backend function - alloc_bootmem_core() instead of the highlevel
interface functions to allow simpler and consistent wrapping.  Also,
HAVE_ARCH_BOOTMEM_NODE is renamed to HAVE_ARCH_BOOTMEM.
Signed-off-by: NTejun Heo <tj@kernel.org>
Cc: Johannes Weiner <hannes@saeurebad.de>
      c1329375
  33. 23 2月, 2009 2 次提交
    • I
      x86: remove the Voyager 32-bit subarch · 965c7eca
      由 Ingo Molnar 提交于 
      Impact: remove unused/broken code

The Voyager subarch last built successfully on the v2.6.26 kernel
and has been stale since then and does not build on the v2.6.27,
v2.6.28 and v2.6.29-rc5 kernels.

No actual users beyond the maintainer reported this breakage.
Patches were sent and most of the fixes were accepted but the
discussion around how to do a few remaining issues cleanly
fizzled out with no resolution and the code remained broken.

In the v2.6.30 x86 tree development cycle 32-bit subarch support
has been reworked and removed - and the Voyager code, beyond the
build problems already known, needs serious and significant
changes and probably a rewrite to support it.

CONFIG_X86_VOYAGER has been marked BROKEN then. The maintainer has
been notified but no patches have been sent so far to fix it.

While all other subarchs have been converted to the new scheme,
voyager is still broken. We'd prefer to receive patches which
clean up the current situation in a constructive way, but even in
case of removal there is no obstacle to add that support back
after the issues have been sorted out in a mutually acceptable
fashion.

So remove this inactive code for now.
Signed-off-by: NIngo Molnar <mingo@elte.hu>
      965c7eca
    • R
      x86: improve the help text of X86_EXTENDED_PLATFORM · 8425091f
      由 Ravikiran G Thirumalai 提交于 
      Change the CONFIG_X86_EXTENDED_PLATFORM help text to display the
32bit/64bit extended platform list. This is as suggested by Ingo.
Signed-off-by: NRavikiran Thirumalai <kiran@scalex86.org>
Cc: shai@scalex86.org
Cc: "Benzi Galili (Benzi@ScaleMP.com)" <benzi@scalemp.com>
Signed-off-by: NIngo Molnar <mingo@elte.hu>
      8425091f
  35. 20 2月, 2009 1 次提交
    • T
      x86: convert to the new dynamic percpu allocator · 11124411
      由 Tejun Heo 提交于 
      Impact: use new dynamic allocator, unified access to static/dynamic
        percpu memory

Convert to the new dynamic percpu allocator.

* implement populate_extra_pte() for both 32 and 64
* update setup_per_cpu_areas() to use pcpu_setup_static()
* define __addr_to_pcpu_ptr() and __pcpu_ptr_to_addr()
* define config HAVE_DYNAMIC_PER_CPU_AREA
Signed-off-by: NTejun Heo <tj@kernel.org>
      11124411
  37. 17 2月, 2009 2 次提交
  39. 12 2月, 2009 1 次提交
  41. 11 2月, 2009 1 次提交
  43. 10 2月, 2009 2 次提交
    • T
      x86: implement x86_32 stack protector · 60a5317f
      由 Tejun Heo 提交于 
      Impact: stack protector for x86_32

Implement stack protector for x86_32.  GDT entry 28 is used for it.
It's set to point to stack_canary-20 and have the length of 24 bytes.
CONFIG_CC_STACKPROTECTOR turns off CONFIG_X86_32_LAZY_GS and sets %gs
to the stack canary segment on entry.  As %gs is otherwise unused by
the kernel, the canary can be anywhere.  It's defined as a percpu
variable.

x86_32 exception handlers take register frame on stack directly as
struct pt_regs.  With -fstack-protector turned on, gcc copies the
whole structure after the stack canary and (of course) doesn't copy
back on return thus losing all changed.  For now, -fno-stack-protector
is added to all files which contain those functions.  We definitely
need something better.
Signed-off-by: NTejun Heo <tj@kernel.org>
Signed-off-by: NIngo Molnar <mingo@elte.hu>
      60a5317f
    • T
      x86: make lazy %gs optional on x86_32 · ccbeed3a
      由 Tejun Heo 提交于 
      Impact: pt_regs changed, lazy gs handling made optional, add slight
        overhead to SAVE_ALL, simplifies error_code path a bit

On x86_32, %gs hasn't been used by kernel and handled lazily.  pt_regs
doesn't have place for it and gs is saved/loaded only when necessary.
In preparation for stack protector support, this patch makes lazy %gs
handling optional by doing the followings.

* Add CONFIG_X86_32_LAZY_GS and place for gs in pt_regs.

* Save and restore %gs along with other registers in entry_32.S unless
  LAZY_GS.  Note that this unfortunately adds "pushl $0" on SAVE_ALL
  even when LAZY_GS.  However, it adds no overhead to common exit path
  and simplifies entry path with error code.

* Define different user_gs accessors depending on LAZY_GS and add
  lazy_save_gs() and lazy_load_gs() which are noop if !LAZY_GS.  The
  lazy_*_gs() ops are used to save, load and clear %gs lazily.

* Define ELF_CORE_COPY_KERNEL_REGS() which always read %gs directly.

xen and lguest changes need to be verified.
Signed-off-by: NTejun Heo <tj@kernel.org>
Cc: Jeremy Fitzhardinge <jeremy@xensource.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: NIngo Molnar <mingo@elte.hu>
      ccbeed3a
  45. 08 2月, 2009 2 次提交
    • S
      ftrace: change function graph tracer to use new in_nmi · 9a5fd902
      由 Steven Rostedt 提交于 
      The function graph tracer piggy backed onto the dynamic ftracer
to use the in_nmi custom code for dynamic tracing. The problem
was (as Andrew Morton pointed out) it really only wanted to bail
out if the context of the current CPU was in NMI context. But the
dynamic ftrace in_nmi custom code was true if _any_ CPU happened
to be in NMI context.

Now that we have a generic in_nmi interface, this patch changes
the function graph code to use it instead of the dynamic ftarce
custom code.
Reported-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NSteven Rostedt <srostedt@redhat.com>
      9a5fd902
    • S
      ring-buffer: add NMI protection for spinlocks · 78d904b4
      由 Steven Rostedt 提交于 
      Impact: prevent deadlock in NMI

The ring buffers are not yet totally lockless with writing to
the buffer. When a writer crosses a page, it grabs a per cpu spinlock
to protect against a reader. The spinlocks taken by a writer are not
to protect against other writers, since a writer can only write to
its own per cpu buffer. The spinlocks protect against readers that
can touch any cpu buffer. The writers are made to be reentrant
with the spinlocks disabling interrupts.

The problem arises when an NMI writes to the buffer, and that write
crosses a page boundary. If it grabs a spinlock, it can be racing
with another writer (since disabling interrupts does not protect
against NMIs) or with a reader on the same CPU. Luckily, most of the
users are not reentrant and protects against this issue. But if a
user of the ring buffer becomes reentrant (which is what the ring
buffers do allow), if the NMI also writes to the ring buffer then
we risk the chance of a deadlock.

This patch moves the ftrace_nmi_enter called by nmi_enter() to the
ring buffer code. It replaces the current ftrace_nmi_enter that is
used by arch specific code to arch_ftrace_nmi_enter and updates
the Kconfig to handle it.

When an NMI is called, it will set a per cpu variable in the ring buffer
code and will clear it when the NMI exits. If a write to the ring buffer
crosses page boundaries inside an NMI, a trylock is used on the spin
lock instead. If the spinlock fails to be acquired, then the entry
is discarded.

This bug appeared in the ftrace work in the RT tree, where event tracing
is reentrant. This workaround solved the deadlocks that appeared there.
Signed-off-by: NSteven Rostedt <srostedt@redhat.com>
      78d904b4
  47. 06 2月, 2009 1 次提交
  49. 05 2月, 2009 1 次提交
  51. 30 1月, 2009 2 次提交
  53. 29 1月, 2009 1 次提交