perf: Do not send exit event twice

4e93ad60 · Jiri Olsa <[email protected]> · 2015-11-04 16:00:05 +01:00

In case we monitor events system wide, we get EXIT event
(when configured) twice for each task that exited.

Note doubled lines with same pid/tid in following example:

  $ sudo ./perf record -a
  ^C[ perf record: Woken up 1 times to write data ]
  [ perf record: Captured and wrote 0.480 MB perf.data (2518 samples) ]
  $ sudo ./perf report -D | grep EXIT

  0 60290687567581 0x59910 [0x38]: PERF_RECORD_EXIT(1250:1250):(1250:1250)
  0 60290687568354 0x59948 [0x38]: PERF_RECORD_EXIT(1250:1250):(1250:1250)
  0 60290687988744 0x59ad8 [0x38]: PERF_RECORD_EXIT(1250:1250):(1250:1250)
  0 60290687989198 0x59b10 [0x38]: PERF_RECORD_EXIT(1250:1250):(1250:1250)
  1 60290692567895 0x62af0 [0x38]: PERF_RECORD_EXIT(1253:1253):(1253:1253)
  1 60290692568322 0x62b28 [0x38]: PERF_RECORD_EXIT(1253:1253):(1253:1253)
  2 60290692739276 0x69a18 [0x38]: PERF_RECORD_EXIT(1252:1252):(1252:1252)
  2 60290692739910 0x69a50 [0x38]: PERF_RECORD_EXIT(1252:1252):(1252:1252)

The reason is that the cpu contexts are processes each time
we call perf_event_task. I'm changing the perf_event_aux logic
to serve task_ctx and cpu contexts separately, which ensure we
don't get EXIT event generated twice on same cpu context.

This does not affect other auxiliary events, as they don't
use task_ctx at all.

Signed-off-by: Jiri Olsa <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Cc: Arnaldo Carvalho de Melo <[email protected]>
Cc: Arnaldo Carvalho de Melo <[email protected]>
Cc: David Ahern <[email protected]>
Cc: Jiri Olsa <[email protected]>
Cc: Linus Torvalds <[email protected]>
Cc: Namhyung Kim <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Stephane Eranian <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Vince Weaver <[email protected]>
Link: http://lkml.kernel.org/r/[email protected]
Signed-off-by: Ingo Molnar <[email protected]>

Commit: 4e93ad601a4308d4a67673c81556580817d56940
Parents: 169b932a

Changes

1 files changed · +31 −11

File	Status	Add	Del	Content
kernel/events/core.c	modified	+31	−11	View file

kernel/events/core.c

View file

diff --git a/kernel/events/core.c b/kernel/events/core.c
index 49a5118..39cf4a4 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -5683,6 +5683,17 @@ perf_event_aux_ctx(struct perf_event_context *ctx,
 }
 
 static void
+perf_event_aux_task_ctx(perf_event_aux_output_cb output, void *data,
+			struct perf_event_context *task_ctx)
+{
+	rcu_read_lock();
+	preempt_disable();
+	perf_event_aux_ctx(task_ctx, output, data);
+	preempt_enable();
+	rcu_read_unlock();
+}
+
+static void
 perf_event_aux(perf_event_aux_output_cb output, void *data,
 	       struct perf_event_context *task_ctx)
 {
@@ -5691,14 +5702,23 @@ perf_event_aux(perf_event_aux_output_cb output, void *data,
 	struct pmu *pmu;
 	int ctxn;
 
+	/*
+	 * If we have task_ctx != NULL we only notify
+	 * the task context itself. The task_ctx is set
+	 * only for EXIT events before releasing task
+	 * context.
+	 */
+	if (task_ctx) {
+		perf_event_aux_task_ctx(output, data, task_ctx);
+		return;
+	}
+
 	rcu_read_lock();
 	list_for_each_entry_rcu(pmu, &pmus, entry) {
 		cpuctx = get_cpu_ptr(pmu->pmu_cpu_context);
 		if (cpuctx->unique_pmu != pmu)
 			goto next;
 		perf_event_aux_ctx(&cpuctx->ctx, output, data);
-		if (task_ctx)
-			goto next;
 		ctxn = pmu->task_ctx_nr;
 		if (ctxn < 0)
 			goto next;
@@ -5708,12 +5728,6 @@ perf_event_aux(perf_event_aux_output_cb output, void *data,
 next:
 		put_cpu_ptr(pmu->pmu_cpu_context);
 	}
-
-	if (task_ctx) {
-		preempt_disable();
-		perf_event_aux_ctx(task_ctx, output, data);
-		preempt_enable();
-	}
 	rcu_read_unlock();
 }
 
@@ -8803,10 +8817,8 @@ static void perf_event_exit_task_context(struct task_struct *child, int ctxn)
 	struct perf_event_context *child_ctx, *clone_ctx = NULL;
 	unsigned long flags;
 
-	if (likely(!child->perf_event_ctxp[ctxn])) {
-		perf_event_task(child, NULL, 0);
+	if (likely(!child->perf_event_ctxp[ctxn]))
 		return;
-	}
 
 	local_irq_save(flags);
 	/*
@@ -8890,6 +8902,14 @@ void perf_event_exit_task(struct task_struct *child)
 
 	for_each_task_context_nr(ctxn)
 		perf_event_exit_task_context(child, ctxn);
+
+	/*
+	 * The perf_event_exit_task_context calls perf_event_task
+	 * with child's task_ctx, which generates EXIT events for
+	 * child contexts and sets child->perf_event_ctxp[] to NULL.
+	 * At this point we need to send EXIT events to cpu contexts.
+	 */
+	perf_event_task(child, NULL, 0);
 }
 
 static void perf_free_event(struct perf_event *event,