Skip to content

L
linux
Switch branch/tag
  1. 09 Feb, 2009 2 commits
    • Mike Galbraith's avatar
      perf_counters: account NMI interrupts · d278c484
      Mike Galbraith authored 
      I noticed that kerneltop interrupts were accounted as NMI, but not their
perf counter origin.

Account NMI performance counter interrupts.
Signed-off-by: default avatarMike Galbraith  <efault@gmx.de>
Signed-off-by: default avatarIngo Molnar <mingo@elte.hu>

 arch/x86/kernel/cpu/perf_counter.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
      d278c484
    • Paul Mackerras's avatar
      perf_counters: make software counters work as per-cpu counters · 23a185ca
      Paul Mackerras authored 
      Impact: kernel crash fix

Yanmin Zhang reported that using a PERF_COUNT_TASK_CLOCK software
counter as a per-cpu counter would reliably crash the system, because
it calls __task_delta_exec with a null pointer.  The page fault,
context switch and cpu migration counters also won't function
correctly as per-cpu counters since they reference the current task.

This fixes the problem by redirecting the task_clock counter to the
cpu_clock counter when used as a per-cpu counter, and by implementing
per-cpu page fault, context switch and cpu migration counters.

Along the way, this:

- Initializes counter->ctx earlier, in perf_counter_alloc, so that
  sw_perf_counter_init can use it
- Adds code to kernel/sched.c to count task migrations into each
  cpu, in rq->nr_migrations_in
- Exports the per-cpu context switch and task migration counts
  via new functions added to kernel/sched.c
- Makes sure that if sw_perf_counter_init fails, we don't try to
  initialize the counter as a hardware counter.  Since the user has
  passed a negative, non-raw event type, they clearly don't intend
  for it to be interpreted as a hardware event.
Reported-by: default avatar"Zhang Yanmin" <yanmin_zhang@linux.intel.com>
Signed-off-by: default avatarPaul Mackerras <paulus@samba.org>
Signed-off-by: default avatarIngo Molnar <mingo@elte.hu>
      23a185ca
  3. 05 Feb, 2009 1 commit
    • Ingo Molnar's avatar
      perfcounters: fix "perf counters kills oprofile" bug, v2 · 82aa9a18
      Ingo Molnar authored 
      Impact: fix kernel crash

Both oprofile and perfcounters register an NMI die handler, but only one
can handle the NMI.  Conveniently, oprofile unregisters it's notifier
when not actively in use, so setting it's notifier priority higher than
perfcounter's allows oprofile to borrow the NMI for the duration of it's
run.  Tested/works both as module and built-in.

While testing, I found that if kerneltop was generating NMIs at very
high frequency, the kernel may panic when oprofile registered it's
handler.  This turned out to be because oprofile registers it's handler
before reset_value has been allocated, so if an NMI comes in while it's
still setting up, kabOom.  Rather than try more invasive changes, I
followed the lead of other places in op_model_ppro.c, and simply
returned in that highly unlikely event.  (debug warnings attached)
Signed-off-by: default avatarMike Galbraith <efault@gmx.de>
Signed-off-by: default avatarIngo Molnar <mingo@elte.hu>
      82aa9a18
  5. 04 Feb, 2009 1 commit
  7. 01 Feb, 2009 1 commit
  9. 29 Jan, 2009 1 commit
  11. 26 Jan, 2009 1 commit
  13. 23 Jan, 2009 12 commits
  15. 21 Jan, 2009 19 commits
  17. 20 Jan, 2009 2 commits
    • Nick Piggin's avatar
      x86: optimise x86's do_page_fault (C entry point for the page fault path) · 92181f19
      Nick Piggin authored 
      Impact: cleanup, restructure code to improve assembly

gcc isn't _all_ that smart about spilling registers to stack or reusing
stack slots, even with branch annotations. do_page_fault contained a lot
of functionality, so split unlikely paths into their own functions, and
mark them as noinline just to be sure. I consider this actually to be
somewhat of a cleanup too: the main function now contains about half
the number of lines so the normal path is easier to read, while the error
cases are also nicely split away.

Also, ensure the order of arguments to functions is always the same: regs,
addr, error_code. This can reduce code size a tiny bit, and just looks neater
too.

And add a couple of branch annotations.

Before:
  do_page_fault:
          subq    $360, %rsp      #,

After:
  do_page_fault:
          subq    $56, %rsp       #,

bloat-o-meter:
  add/remove: 8/0 grow/shrink: 0/1 up/down: 2222/-1680 (542)
  function                                     old     new   delta
  __bad_area_nosemaphore                         -     506    +506
  no_context                                     -     474    +474
  vmalloc_fault                                  -     424    +424
  spurious_fault                                 -     358    +358
  mm_fault_error                                 -     272    +272
  bad_area_access_error                          -      89     +89
  bad_area                                       -      89     +89
  bad_area_nosemaphore                           -      10     +10
  do_page_fault                               2464     784   -1680

Yes, the total size increases by 542 bytes, due to the extra function calls.
But these will very rarely be called (except for vmalloc_fault) in a normal
workload. Importantly, do_page_fault is less than 1/3rd it's original size,
and touches far less stack.

Existing gotos and branch hints did move a lot of the infrequently used text
out of the fastpath, but that's even further improved after this patch.
Signed-off-by: default avatarNick Piggin <npiggin@suse.de>
Acked-by: default avatarLinus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: default avatarIngo Molnar <mingo@elte.hu>
      92181f19
    • Ingo Molnar's avatar
      Merge commit 'v2.6.29-rc2' into x86/mm · 0ce1c383
      Ingo Molnar authored
      0ce1c383
GitLab Nexedi Edition | About GitLab | About Nexedi | 沪ICP备2021021310号-2 | 沪ICP备2021021310号-7