perf_counter: Full task tracing
In order to be able to distinguish between no samples due to
inactivity and no samples due to task ended, Arjan asked for
PERF_EVENT_EXIT events. This is useful to the boot delay
instrumentation (bootchart) app.
This patch changes the PERF_EVENT_FORK to be emitted on every
clone, and adds PERF_EVENT_EXIT to be emitted on task exit,
after the task's counters have been closed.
This task tracing is controlled through: attr.comm || attr.mmap
and through the new attr.task field.
Suggested-by: Arjan van de Ven <arjan@linux.intel.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Anton Blanchard <anton@samba.org>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
[ cleaned up perf_counter.h a bit ]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
diff --git a/include/linux/perf_counter.h b/include/linux/perf_counter.h
index bd15d7a..e604e6e 100644
--- a/include/linux/perf_counter.h
+++ b/include/linux/perf_counter.h
@@ -181,8 +181,9 @@
freq : 1, /* use freq, not period */
inherit_stat : 1, /* per task counts */
enable_on_exec : 1, /* next exec enables */
+ task : 1, /* trace fork/exit */
- __reserved_1 : 51;
+ __reserved_1 : 50;
__u32 wakeup_events; /* wakeup every n events */
__u32 __reserved_2;
@@ -311,6 +312,15 @@
/*
* struct {
* struct perf_event_header header;
+ * u32 pid, ppid;
+ * u32 tid, ptid;
+ * };
+ */
+ PERF_EVENT_EXIT = 4,
+
+ /*
+ * struct {
+ * struct perf_event_header header;
* u64 time;
* u64 id;
* u64 stream_id;
@@ -323,6 +333,7 @@
* struct {
* struct perf_event_header header;
* u32 pid, ppid;
+ * u32 tid, ptid;
* };
*/
PERF_EVENT_FORK = 7,
