sched: Make sure timers have migrated before killing the migration_thread
Amit K. Arora [Tue, 25 May 2010 13:23:46 +0000 (18:23 +0530)]
Problem: In a stress test where some heavy tests were running along with
regular CPU offlining and onlining, a hang was observed. The system seems
to be hung at a point where migration_call() tries to kill the
migration_thread of the dying CPU, which just got moved to the current
CPU. This migration thread does not get a chance to run (and die) since
rt_throttled is set to 1 on current, and it doesn't get cleared as the
hrtimer which is supposed to reset the rt bandwidth
(sched_rt_period_timer) is tied to the CPU which we just marked dead!

Solution: This patch pushes the killing of migration thread to
"CPU_POST_DEAD" event. By then all the timers (including
sched_rt_period_timer) should have got migrated (along with other

Signed-off-by: Amit Arora <>
Signed-off-by: Gautham R Shenoy <>
Acked-by: Tejun Heo <>
Signed-off-by: Peter Zijlstra <>
Cc: Thomas Gleixner <>
LKML-Reference: <>
Signed-off-by: Ingo Molnar <>


index b4e7431..70f8d90 100644 (file)
@@ -321,7 +321,7 @@ static int __cpuinit cpu_stop_cpu_callback(struct notifier_block *nfb,
        case CPU_UP_CANCELED:
-       case CPU_DEAD:
+       case CPU_POST_DEAD:
                struct cpu_stop_work *work;