Hi,

This patchset mainly fix a race problem between cgroup_migrate_execute()
and perf_cgroup_sched_out/in().

v4:
 - delete patch 3 in v3 that remove event_filter_match() in
   merge_sched_in(), which is wrong since the hybrid system
   implemented PMU specific filter_match callback. Thanks Kan.

v3:
 - put two fix patches together in patch 4
 - delete percpu cpu_perf_cgroup, just use cpuctx->cgrp to compare
 - reorganize these patches' order

v2:
 - split into two patches to fix the race problem for easier review
 - use cpuctx->cgrp when start ctx time and delete unused task argument
 - use cpuctx->cgrp when update ctx time from cgroup perf_event
 - always set cpuctx->cgrp when the first cgroup event enabled
 - remove obselete comments

v1:
 - https://lore.kernel.org/lkml/20220308135948.55336-1-zhouchengming@bytedance.com/


Chengming Zhou (4):
  perf/core: Don't pass task around when ctx sched in
  perf/core: Use perf_cgroup_info->active to check if cgroup is active
  perf/core: Fix perf_cgroup_switch()
  perf/core: Always set cpuctx cgrp when enable cgroup event

 kernel/events/core.c | 209 +++++++++++--------------------------------
 1 file changed, 52 insertions(+), 157 deletions(-)

-- 
2.35.1

On Tue, Mar 29, 2022 at 11:45:19PM +0800, Chengming Zhou wrote:
> Chengming Zhou (4):
>   perf/core: Don't pass task around when ctx sched in
>   perf/core: Use perf_cgroup_info->active to check if cgroup is active
>   perf/core: Fix perf_cgroup_switch()
>   perf/core: Always set cpuctx cgrp when enable cgroup event

Thanks, queued for post -rc1