]> git.proxmox.com Git - mirror_ubuntu-eoan-kernel.git/commitdiff
perf_event: Fix broken calc_timer_values()
authorEric B Munson <emunson@mgebm.net>
Thu, 23 Jun 2011 20:34:38 +0000 (16:34 -0400)
committerIngo Molnar <mingo@elte.hu>
Wed, 31 Aug 2011 13:56:29 +0000 (15:56 +0200)
We detected a serious issue with PERF_SAMPLE_READ and
timing information when events were being multiplexing.

Samples would have time_running > time_enabled. That
was easy to reproduce with a libpfm4 example (ran 3
times to cause multiplexing on Core 2):

 $ syst_smpl -e uops_retired:freq=1 &
 $ syst_smpl -e uops_retired:freq=1 &
 $ syst_smpl -e uops_retired:freq=1 &
 IIP:0x0000000040062d ... PERIOD:2355332948 ENA=40144625315 RUN=60014875184
 syst_smpl: WARNING: time_running > time_enabled
63277537998 uops_retired:freq=1 , scaled

The bug was not present in kernel up to (and including) 3.0. It turns
out the bug was introduced by the following commit:

commit c4794295917ebeda8013b6cb9c8d71ab4f74a1fa

    events: Move lockless timer calculation into helper function

The parameters of the function got reversed yet the call sites
were not updated to reflect the change. That lead to time_running
and time_enabled being swapped. That had no effect when there was
no multiplexing because in that case time_running = time_enabled
but it would show up in any other scenario.

Signed-off-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/20110829124112.GA4828@quad
Signed-off-by: Ingo Molnar <mingo@elte.hu>
kernel/events/core.c

index 45847fbb599aba45e20f25af785ba7b6aee80f2d..0f857782d06f45eace0bc2bedb7bb1ccf3bc3512 100644 (file)
@@ -3396,8 +3396,8 @@ static int perf_event_index(struct perf_event *event)
 }
 
 static void calc_timer_values(struct perf_event *event,
-                               u64 *running,
-                               u64 *enabled)
+                               u64 *enabled,
+                               u64 *running)
 {
        u64 now, ctx_time;