Re: [PATCH] perf: fix assertion failure in x86_pmu_start()

[Ingo Molnar <mingo@elte.hu>]

* Eric Dumazet <eric.dumazet@gmail.com> wrote:

> Le lundi 06 février 2012 à 21:53 +0100, Stephane Eranian a écrit :
> > The following patch fixes an issue introduced by the following
> > commit:
> >         e050e3f0a71b ("perf: Fix broken interrupt rate throttling")
> > 
> > The patch caused the following warning to pop up depending on
> > the sampling frequency adjustments:
> > 
> > [89214.962603] ------------[ cut here ]------------
> > [89214.967441] WARNING: at arch/x86/kernel/cpu/perf_event.c:995 x86_pmu_start+0x79/0xd4()
> > [89214.975825] Hardware name: X8DTN
> > [89214.979268] Modules linked in:
> > [89214.982560] Pid: 0, comm: swapper/6 Not tainted 3.3.0-rc2-tip+ #1
> > [89214.988865] Call Trace:
> > [89214.991533]  <IRQ>  [<ffffffff81065cc7>] warn_slowpath_common+0x7e/0x97
> > [89214.998379]  [<ffffffff81065cf5>] warn_slowpath_null+0x15/0x17
> > [89215.004428]  [<ffffffff8103f626>] x86_pmu_start+0x79/0xd4
> > [89215.010042]  [<ffffffff810e30d1>] perf_adjust_freq_unthr_context.part.63+0xef/0x123
> > [89215.018123]  [<ffffffff810e318c>] perf_event_task_tick+0x87/0x1c1
> > [89215.024463]  [<ffffffff810a2370>] ? tick_nohz_handler+0xda/0xda
> > [89215.030595]  [<ffffffff8108b819>] scheduler_tick+0xd1/0xf3
> > [89215.036296]  [<ffffffff810720b0>] update_process_times+0x5e/0x6f
> > [89215.042512]  [<ffffffff810a23e0>] tick_sched_timer+0x70/0x99
> > [89215.048387]  [<ffffffff810823f9>] __run_hrtimer+0x8c/0x148
> > [89215.054087]  [<ffffffff81082add>] hrtimer_interrupt+0xc1/0x18c
> > 
> > It was caused by the following call sequence:
> > 
> > perf_adjust_freq_unthr_context.part() {
> >      stop()
> >      if (delta > 0) {
> >           perf_adjust_period() {
> >               if (period > 8*...) {
> >                   stop()
> >                   ...
> >                   start()
> >               }
> >           }
> >       }
> >       start()
> > }
> > 
> > Which caused a double start and a double stop, thus triggering the assert
> > in x86_pmu_start().
> > 
> > The patch fixes the problem by avoiding the double calls. We pass a new
> > argument to perf_adjust_period() to indicate whether or not the event
> > is already stopped. We can't just remove the start/stop from that function
> > because it's called from __perf_event_overflow where the event needs to
> > be reloaded via a stop/start back-toback call.
> > 
> > The patch reintroduces the assertion in x86_pmu_start() which was removed
> > by commit:
> > 	84f2b9b perf: Remove deprecated WARN_ON_ONCE()
> > 
> > Signed-off-by: Stephane Eranian <eranian@google.com>
> > ---
> 
> This indeed fix the WARNING for me
> 
> Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
> 
> But I still have these messages when doing a perf session.
> 
> Machine seems to recover properly.
> 
> Previous kernels were working without notice.
> 
> [  300.553017] Uhhuh. NMI received for unknown reason 31 on CPU 2.
> [  300.553071] Do you have a strange power saving mode enabled?

Were these messages introduced by:

 e050e3f0a71b: perf: Fix broken interrupt rate throttling

as well?

In any case I'm holding off on applying the patch before this is 
resolved.

Thanks,

	Ingo