[PATCH v1 6/6] perf: Allow set-output for task contexts of different types
- Date: Tue, 12 Jun 2018 10:51:17 +0300
- From: Alexander Shishkin <alexander.shishkin@xxxxxxxxxxxxxxx>
- Subject: [PATCH v1 6/6] perf: Allow set-output for task contexts of different types
From: Adrian Hunter <adrian.hunter@xxxxxxxxx>
Set-output must be limited to events that cannot be active on different
cpus at the same time. Thus either the event cpu must be the same, or
the event task must be the same. Current logic does not check the task
directly but checks whether the perf_event_context is the same, however
there are separate contexts for hardware and software events so in that
case the perf_event_context is different even though the task is the same.
This patch changes the logic to check the task directly.
Signed-off-by: Adrian Hunter <adrian.hunter@xxxxxxxxx>
Signed-off-by: Alexander Shishkin <alexander.shishkin@xxxxxxxxxxxxxxx>
kernel/events/core.c | 16 ++++++++++++++--
1 file changed, 14 insertions(+), 2 deletions(-)
diff --git a/kernel/events/core.c b/kernel/events/core.c
index 9bf5c7abb621..9ab9f21f58da 100644
@@ -9848,9 +9848,21 @@ perf_event_set_output(struct perf_event *event, struct perf_event *output_event)
- * If its not a per-cpu rb, it must be the same task.
+ * If it's not a per-cpu rb, it must be the same task.
+ * Since output_event is a per-task event, ->ctx is stable
+ * and should be around for as long as the file is around.
+ * The context switch optimization doesn't apply to output_event
+ * as well, so we can look at its ctx->task, which will be either
+ * a valid task or TASK_TOMBSTONE.
+ * The source event's task can also be TASK_TOMBSTONE, so look out
+ * for that also.
- if (output_event->cpu == -1 && output_event->ctx != event->ctx)
+ if (output_event->cpu == -1 &&
+ (output_event->ctx->task != event->ctx->task ||
+ output_event->ctx->task == TASK_TOMBSTONE))