perf sched: Sanity check context switch events

Use 'perf sched latency' to track the current task based on
context-switch events, and flag the cases where there's some
impossible transition: such as a PID being switched out that
was not switched in.

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

1 files changed, 26 insertions, 1 deletions

diff --git a/tools/perf/builtin-sched.c b/tools/perf/builtin-sched.c
index 1f0f9be34faa..2d542368de3c 100644
--- a/tools/perf/builtin-sched.c
+++ b/tools/perf/builtin-sched.c
@@ -119,6 +119,7 @@ static unsigned long		replay_repeat = 10;
 static unsigned long            nr_timestamps;
 static unsigned long            nr_unordered_timestamps;
 static unsigned long            nr_state_machine_bugs;
+static unsigned long            nr_context_switch_bugs;
 static unsigned long            nr_events;
 static unsigned long            nr_lost_chunks;
 static unsigned long            nr_lost_events;
@@ -1399,6 +1400,14 @@ static void __cmd_lat(void)
                         printf(" (due to lost events?)");
                 printf("\n");
         }
+        if (nr_context_switch_bugs && nr_timestamps) {
+                printf("  INFO: %.3f%% context switch bugs (%ld out of %ld)",
+                        (double)nr_context_switch_bugs/(double)nr_timestamps*100.0,
+                        nr_context_switch_bugs, nr_timestamps);
+                if (nr_lost_events)
+                        printf(" (due to lost events?)");
+                printf("\n");
+        }
         printf("\n");
 
 }
@@ -1425,10 +1434,16 @@ process_sched_wakeup_event(struct raw_event_sample *raw,
         trace_handler->wakeup_event(&wakeup_event, event, cpu, timestamp, thread);
 }
 
+/*
+ * Track the current task - that way we can know whether there's any
+ * weird events, such as a task being switched away that is not current.
+ */
+static u32 curr_pid[MAX_CPUS] = { [0 ... MAX_CPUS-1] = -1 };
+
 static void
 process_sched_switch_event(struct raw_event_sample *raw,
                            struct event *event,
-                           int cpu __used,
+                           int cpu,
                            u64 timestamp __used,
                            struct thread *thread __used)
 {
@@ -1444,6 +1459,16 @@ process_sched_switch_event(struct raw_event_sample *raw,
         FILL_FIELD(switch_event, next_pid, event, raw->data);
         FILL_FIELD(switch_event, next_prio, event, raw->data);
 
+        if (curr_pid[cpu] != (u32)-1) {
+                /*
+                 * Are we trying to switch away a PID that is
+                 * not current?
+                 */
+                if (curr_pid[cpu] != switch_event.prev_pid)
+                        nr_context_switch_bugs++;
+        }
+        curr_pid[cpu] = switch_event.next_pid;
+
         trace_handler->switch_event(&switch_event, event, cpu, timestamp, thread);
 }