perf core: Add PERF_COUNT_SW_CGROUP_SWITCHES event
authorNamhyung Kim <namhyung@kernel.org>
Wed, 10 Feb 2021 08:33:26 +0000 (17:33 +0900)
committerPeter Zijlstra <peterz@infradead.org>
Fri, 16 Apr 2021 16:58:52 +0000 (18:58 +0200)
commitd0d1dd628527c77db2391ce0293c1ed344b2365f
treed401d43e406558d2a54b772ef75f9f27c6185e69
parent7c8056bb366b1b2dc8e4a3cc0b876e15a8ebca2c
perf core: Add PERF_COUNT_SW_CGROUP_SWITCHES event

This patch adds a new software event to count context switches
involving cgroup switches.  So it's counted only if cgroups of
previous and next tasks are different.  Note that it only checks the
cgroups in the perf_event subsystem.  For cgroup v2, it shouldn't
matter anyway.

One can argue that we can do this by using existing sched_switch event
with eBPF.  But some systems might not have eBPF for some reason so
I'd like to add this as a simple way.

Signed-off-by: Namhyung Kim <namhyung@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Link: https://lkml.kernel.org/r/20210210083327.22726-2-namhyung@kernel.org
include/linux/perf_event.h
include/uapi/linux/perf_event.h