perf/core: Fix group scheduling with mixed hw and sw events
authorSong Liu <songliubraving@fb.com>
Thu, 3 May 2018 19:47:16 +0000 (12:47 -0700)
committerIngo Molnar <mingo@kernel.org>
Fri, 25 May 2018 06:11:10 +0000 (08:11 +0200)
commita1150c202207cc8501bebc45b63c264f91959260
tree17b8820b35d29a2e82875b8339ca036fe6452db2
parentbd9c67ad9693bacef086d65c1c6744645d4777e7
perf/core: Fix group scheduling with mixed hw and sw events

When hw and sw events are mixed in the same group, they are all attached
to the hw perf_event_context. This sometimes requires moving group of
perf_event to a different context.

We found a bug in how the kernel handles this, for example if we do:

   perf stat -e '{faults,ref-cycles,faults}'  -I 1000

     1.005591180              1,297      faults
     1.005591180        457,476,576      ref-cycles
     1.005591180    <not supported>      faults

First, sw event "faults" is attached to the sw context, and becomes the
group leader. Then, hw event "ref-cycles" is attached, so both events
are moved to the hw context. Last, another sw "faults" tries to attach,
but it fails because of mismatch between the new target ctx (from sw
pmu) and the group_leader's ctx (hw context, same as ref-cycles).

The broken condition is:
   group_leader is sw event;
   group_leader is on hw context;
   add a sw event to the group.

Fix this scenario by checking group_leader's context (instead of just
event type). If group_leader is on hw context, use the ->pmu of this
context to look up context for the new event.

Signed-off-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: <kernel-team@fb.com>
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: Stephane Eranian <eranian@google.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vince Weaver <vincent.weaver@maine.edu>
Fixes: b04243ef7006 ("perf: Complete software pmu grouping")
Link: http://lkml.kernel.org/r/20180503194716.162815-1-songliubraving@fb.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
include/linux/perf_event.h
kernel/events/core.c