[PATCH] cgroup, perf_event: Fix perf cgroup hierarchy support

[PATCH] cgroup, perf_event: Fix perf cgroup hierarchy support

[Tejun Heo]

864c2357ca89 ("perf/core: Do not set cpuctx->cgrp for unscheduled
cgroups") made list_update_cgroup_event() skip setting cpuctx->cgrp if
no cgroup event targets %current's cgroup.  This breaks perf_event's
hierarchical support because events which target one of the ancestors
get ignored.

Fix it by using cgroup_is_descendant() test instead of equality.

Signed-off-by: Tejun Heo <tj@kernel.org>
Fixes: 864c2357ca89 ("perf/core: Do not set cpuctx->cgrp for unscheduled cgroups")
Cc: stable@vger.kernel.org # v4.9+
Cc: David Carrillo-Cisneros <davidcc@google.com>
---
 kernel/events/core.c |    7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

diff --git a/kernel/events/core.c b/kernel/events/core.c
index 9d93db8..db2e630 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -901,9 +901,12 @@ list_update_cgroup_event(struct perf_event *event,
 	cpuctx_entry = &cpuctx->cgrp_cpuctx_entry;
 	/* cpuctx->cgrp is NULL unless a cgroup event is active in this CPU .*/
 	if (add) {
+		struct perf_cgroup *cgrp = perf_cgroup_from_task(current, ctx);
+
 		list_add(cpuctx_entry, this_cpu_ptr(&cgrp_cpuctx_list));
-		if (perf_cgroup_from_task(current, ctx) == event->cgrp)
-			cpuctx->cgrp = event->cgrp;
+		if (cgroup_is_descendant(cgrp->css.cgroup,
+					 event->cgrp->css.cgroup))
+			cpuctx->cgrp = cgrp;
 	} else {
 		list_del(cpuctx_entry);
 		cpuctx->cgrp = NULL;

Re: [PATCH] cgroup, perf_event: Fix perf cgroup hierarchy support

[Peter Zijlstra]

On Sat, Oct 28, 2017 at 09:49:37AM -0700, Tejun Heo wrote:
> 864c2357ca89 ("perf/core: Do not set cpuctx->cgrp for unscheduled
> cgroups") made list_update_cgroup_event() skip setting cpuctx->cgrp if
> no cgroup event targets %current's cgroup.  This breaks perf_event's
> hierarchical support because events which target one of the ancestors
> get ignored.
> 
> Fix it by using cgroup_is_descendant() test instead of equality.

Right similar to:

  e6a5203399d1 ("perf/core: Fix cgroup time when scheduling descendants")


> Signed-off-by: Tejun Heo <tj@kernel.org>
> Fixes: 864c2357ca89 ("perf/core: Do not set cpuctx->cgrp for unscheduled cgroups")
> Cc: stable@vger.kernel.org # v4.9+
> Cc: David Carrillo-Cisneros <davidcc@google.com>

I think David left Google (I've had this email bounce), lets add
Stephane.

Cc: Stephane Eranian <eranian@gmail.com>


Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>

> ---
>  kernel/events/core.c |    7 +++++--
>  1 file changed, 5 insertions(+), 2 deletions(-)
> 
> diff --git a/kernel/events/core.c b/kernel/events/core.c
> index 9d93db8..db2e630 100644
> --- a/kernel/events/core.c
> +++ b/kernel/events/core.c
> @@ -901,9 +901,12 @@ list_update_cgroup_event(struct perf_event *event,
>  	cpuctx_entry = &cpuctx->cgrp_cpuctx_entry;
>  	/* cpuctx->cgrp is NULL unless a cgroup event is active in this CPU .*/
>  	if (add) {
> +		struct perf_cgroup *cgrp = perf_cgroup_from_task(current, ctx);
> +
>  		list_add(cpuctx_entry, this_cpu_ptr(&cgrp_cpuctx_list));
> -		if (perf_cgroup_from_task(current, ctx) == event->cgrp)
> -			cpuctx->cgrp = event->cgrp;
> +		if (cgroup_is_descendant(cgrp->css.cgroup,
> +					 event->cgrp->css.cgroup))
> +			cpuctx->cgrp = cgrp;
>  	} else {
>  		list_del(cpuctx_entry);
>  		cpuctx->cgrp = NULL;

[tip:perf/urgent] perf/cgroup: Fix perf cgroup hierarchy support

[tip-bot for Tejun Heo]

Commit-ID:  be96b316deff35e119760982c43af74e606fa143
Gitweb:     https://git.kernel.org/tip/be96b316deff35e119760982c43af74e606fa143
Author:     Tejun Heo <tj@kernel.org>
AuthorDate: Sat, 28 Oct 2017 09:49:37 -0700
Committer:  Ingo Molnar <mingo@kernel.org>
CommitDate: Mon, 30 Oct 2017 11:58:51 +0100

perf/cgroup: Fix perf cgroup hierarchy support

The following commit:

  864c2357ca89 ("perf/core: Do not set cpuctx->cgrp for unscheduled cgroups")

made list_update_cgroup_event() skip setting cpuctx->cgrp if no cgroup event
targets %current's cgroup.

This breaks perf_event's hierarchical support because events which target one
of the ancestors get ignored.

Fix it by using cgroup_is_descendant() test instead of equality.

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: David Carrillo-Cisneros <davidcc@google.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: kernel-team@fb.com
Cc: stable@vger.kernel.org # v4.9+
Fixes: 864c2357ca89 ("perf/core: Do not set cpuctx->cgrp for unscheduled cgroups")
Link: http://lkml.kernel.org/r/20171028164237.GA972780@devbig577.frc2.facebook.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
---
 kernel/events/core.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/kernel/events/core.c b/kernel/events/core.c
index 9d93db8..10cdb9c 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -901,9 +901,11 @@ list_update_cgroup_event(struct perf_event *event,
 	cpuctx_entry = &cpuctx->cgrp_cpuctx_entry;
 	/* cpuctx->cgrp is NULL unless a cgroup event is active in this CPU .*/
 	if (add) {
+		struct perf_cgroup *cgrp = perf_cgroup_from_task(current, ctx);
+
 		list_add(cpuctx_entry, this_cpu_ptr(&cgrp_cpuctx_list));
-		if (perf_cgroup_from_task(current, ctx) == event->cgrp)
-			cpuctx->cgrp = event->cgrp;
+		if (cgroup_is_descendant(cgrp->css.cgroup, event->cgrp->css.cgroup))
+			cpuctx->cgrp = cgrp;
 	} else {
 		list_del(cpuctx_entry);
 		cpuctx->cgrp = NULL;