[PATCH] perf/core: fix corner case in perf_rotate_context()

* [PATCH] perf/core: fix corner case in perf_rotate_context()
@ 2019-10-03  6:43 Song Liu
  2019-10-06  6:09 ` Song Liu
                   ` (2 more replies)
  0 siblings, 3 replies; 6+ messages in thread
From: Song Liu @ 2019-10-03  6:43 UTC (permalink / raw)
  To: linux-kernel
  Cc: kernel-team, Song Liu, stable, Peter Zijlstra,
	Arnaldo Carvalho de Melo, Thomas Gleixner

This is a rare corner case, but it does happen:

In perf_rotate_context(), when the first cpu flexible event fail to
schedule, cpu_rotate is 1, while cpu_event is NULL. Since cpu_event is
NULL, perf_rotate_context will _NOT_ call cpu_ctx_sched_out(), thus
cpuctx->ctx.is_active will have EVENT_FLEXIBLE set. Then, the next
perf_event_sched_in() will skip all cpu flexible events because of the
EVENT_FLEXIBLE bit.

In the next call of perf_rotate_context(), cpu_rotate stays 1, and
cpu_event stays NULL, so this process repeats. The end result is, flexible
events on this cpu will not be scheduled (until another event being added
to the cpuctx).

Similar issue may happen with the task_ctx. But it is usually not a
problem because the task_ctx moves around different CPU.

Fix this corner case by using cpu_rotate and task_rotate to gate calls for
(cpu_)ctx_sched_out and rotate_ctx. Also enable rotate_ctx() to handle
event == NULL case.

Fixes: 8d5bce0c37fa ("perf/core: Optimize perf_rotate_context() event scheduling")
Cc: stable@vger.kernel.org # v4.17+
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Arnaldo Carvalho de Melo <acme@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Song Liu <songliubraving@fb.com>
---
 kernel/events/core.c | 15 +++++++++++----
 1 file changed, 11 insertions(+), 4 deletions(-)

diff --git a/kernel/events/core.c b/kernel/events/core.c
index 4655adbbae10..50021735f367 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -3775,6 +3775,13 @@ static void rotate_ctx(struct perf_event_context *ctx, struct perf_event *event)
 	if (ctx->rotate_disable)
 		return;
 
+	/* if no event specified, try to rotate the first event */
+	if (!event)
+		event = rb_entry_safe(rb_first(&ctx->flexible_groups.tree),
+				      typeof(*event), group_node);
+	if (!event)
+		return;
+
 	perf_event_groups_delete(&ctx->flexible_groups, event);
 	perf_event_groups_insert(&ctx->flexible_groups, event);
 }
@@ -3816,14 +3823,14 @@ static bool perf_rotate_context(struct perf_cpu_context *cpuctx)
 	 * As per the order given at ctx_resched() first 'pop' task flexible
 	 * and then, if needed CPU flexible.
 	 */
-	if (task_event || (task_ctx && cpu_event))
+	if (task_rotate || (task_ctx && cpu_rotate))
 		ctx_sched_out(task_ctx, cpuctx, EVENT_FLEXIBLE);
-	if (cpu_event)
+	if (cpu_rotate)
 		cpu_ctx_sched_out(cpuctx, EVENT_FLEXIBLE);
 
-	if (task_event)
+	if (task_rotate)
 		rotate_ctx(task_ctx, task_event);
-	if (cpu_event)
+	if (cpu_rotate)
 		rotate_ctx(&cpuctx->ctx, cpu_event);
 
 	perf_event_sched_in(cpuctx, task_ctx, current);
-- 
2.17.1


^ permalink raw reply related	[flat|nested] 6+ messages in thread