perf/core: Fix time tracking bug with multiplexing
authorPeter Zijlstra <peterz@infradead.org>
Tue, 29 Mar 2016 07:26:44 +0000 (09:26 +0200)
committerIngo Molnar <mingo@kernel.org>
Thu, 31 Mar 2016 07:54:06 +0000 (09:54 +0200)
commit8fdc65391c6ad16461526a685f03262b3b01bfde
tree3c96715b1cd1f2ca6fcf95e2b64a284f02f084dc
parentc932cf07ddc02f79e093596924e41cf6d7fda509
perf/core: Fix time tracking bug with multiplexing

Stephane reported that commit:

  3cbaa5906967 ("perf: Fix ctx time tracking by introducing EVENT_TIME")

introduced a regression wrt. time tracking, as easily observed by:

> This patch introduce a bug in the time tracking of events when
> multiplexing is used.
>
> The issue is easily reproducible with the following perf run:
>
>  $ perf stat -a -C 0 -e branches,branches,branches,branches,branches,branches -I 1000
>      1.000730239            652,394      branches   (66.41%)
>      1.000730239            597,809      branches   (66.41%)
>      1.000730239            593,870      branches   (66.63%)
>      1.000730239            651,440      branches   (67.03%)
>      1.000730239            656,725      branches   (66.96%)
>      1.000730239      <not counted>      branches
>
> One branches event is shown as not having run. Yet, with
> multiplexing, all events should run especially with a 1s (-I 1000)
> interval. The delta for time_running comes out to 0. Yet, the event
> has run because the kernel is actually multiplexing the events. The
> problem is that the time tracking is the kernel and especially in
> ctx_sched_out() is wrong now.
>
> The problem is that in case that the kernel enters ctx_sched_out() with the
> following state:
>    ctx->is_active=0x7 event_type=0x1
>    Call Trace:
>     [<ffffffff813ddd41>] dump_stack+0x63/0x82
>     [<ffffffff81182bdc>] ctx_sched_out+0x2bc/0x2d0
>     [<ffffffff81183896>] perf_mux_hrtimer_handler+0xf6/0x2c0
>     [<ffffffff811837a0>] ? __perf_install_in_context+0x130/0x130
>     [<ffffffff810f5818>] __hrtimer_run_queues+0xf8/0x2f0
>     [<ffffffff810f6097>] hrtimer_interrupt+0xb7/0x1d0
>     [<ffffffff810509a8>] local_apic_timer_interrupt+0x38/0x60
>     [<ffffffff8175ca9d>] smp_apic_timer_interrupt+0x3d/0x50
>     [<ffffffff8175ac7c>] apic_timer_interrupt+0x8c/0xa0
>
> In that case, the test:
>       if (is_active & EVENT_TIME)
>
> will be false and the time will not be updated. Time must always be updated on
> sched out.

Fix this by always updating time if EVENT_TIME was set, as opposed to
only updating time when EVENT_TIME changed.

Reported-by: Stephane Eranian <eranian@google.com>
Tested-by: Stephane Eranian <eranian@google.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Cc: kan.liang@intel.com
Cc: namhyung@kernel.org
Fixes: 3cbaa5906967 ("perf: Fix ctx time tracking by introducing EVENT_TIME")
Link: http://lkml.kernel.org/r/20160329072644.GB3408@twins.programming.kicks-ass.net
Signed-off-by: Ingo Molnar <mingo@kernel.org>
kernel/events/core.c
This page took 0.025777 seconds and 5 git commands to generate.