perf sched: Remove nr_state_machine_bugs in perf latency
authorDongsheng Yang <yangds.fnst@cn.fujitsu.com>
Tue, 13 May 2014 01:38:21 +0000 (10:38 +0900)
committerJiri Olsa <jolsa@kernel.org>
Fri, 16 May 2014 07:17:36 +0000 (09:17 +0200)
As we do not use .success in sched_wakeup event any more, then
we can not guarantee that the task when wakeup event happen is
out of run queue. So the message of nr_state_machine_bugs is
not correct.

Signed-off-by: Dongsheng Yang <yangds.fnst@cn.fujitsu.com>
Link: http://lkml.kernel.org/r/1399945101-21736-1-git-send-email-yangds.fnst@cn.fujitsu.com
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
tools/perf/builtin-sched.c

index a3320f1cda694b94784df75d7402840aa3898189..0b4fe533e1a135bd7c4bc4e48c7dc83d4ec4fd11 100644 (file)
@@ -149,7 +149,6 @@ struct perf_sched {
        unsigned long    nr_runs;
        unsigned long    nr_timestamps;
        unsigned long    nr_unordered_timestamps;
-       unsigned long    nr_state_machine_bugs;
        unsigned long    nr_context_switch_bugs;
        unsigned long    nr_events;
        unsigned long    nr_lost_chunks;
@@ -1032,12 +1031,18 @@ static int latency_wakeup_event(struct perf_sched *sched,
        atom = list_entry(atoms->work_list.prev, struct work_atom, list);
 
        /*
+        * As we do not guarantee the wakeup event happens when
+        * task is out of run queue, also may happen when task is
+        * on run queue and wakeup only change ->state to TASK_RUNNING,
+        * then we should not set the ->wake_up_time when wake up a
+        * task which is on run queue.
+        *
         * You WILL be missing events if you've recorded only
         * one CPU, or are only looking at only one, so don't
-        * make useless noise.
+        * skip in this case.
         */
        if (sched->profile_cpu == -1 && atom->state != THREAD_SLEEPING)
-               sched->nr_state_machine_bugs++;
+               return 0;
 
        sched->nr_timestamps++;
        if (atom->sched_out_time > timestamp) {
@@ -1496,14 +1501,6 @@ static void print_bad_events(struct perf_sched *sched)
                        (double)sched->nr_lost_events/(double)sched->nr_events * 100.0,
                        sched->nr_lost_events, sched->nr_events, sched->nr_lost_chunks);
        }
-       if (sched->nr_state_machine_bugs && sched->nr_timestamps) {
-               printf("  INFO: %.3f%% state machine bugs (%ld out of %ld)",
-                       (double)sched->nr_state_machine_bugs/(double)sched->nr_timestamps*100.0,
-                       sched->nr_state_machine_bugs, sched->nr_timestamps);
-               if (sched->nr_lost_events)
-                       printf(" (due to lost events?)");
-               printf("\n");
-       }
        if (sched->nr_context_switch_bugs && sched->nr_timestamps) {
                printf("  INFO: %.3f%% context switch bugs (%ld out of %ld)",
                        (double)sched->nr_context_switch_bugs/(double)sched->nr_timestamps*100.0,