sched: unthrottle rt runqueues in __disable_runtime()
Peter Boonstoppel [Thu, 17 May 2012 22:15:43 +0000 (15:15 -0700)]
migrate_tasks() uses _pick_next_task_rt() to get tasks from the
real-time runqueues to be migrated. When rt_rq is throttled
_pick_next_task_rt() won't return anything, in which case
migrate_tasks() can't move all threads over and gets stuck in an
infinite loop.

Instead unthrottle rt runqueues before migrating tasks.

Bug 976709

Change-Id: Ie3696702abc560fe8ffa7d2fb5dc5d54d532cc0d
Signed-off-by: Peter Boonstoppel <pboonstoppel@nvidia.com>
(cherry picked from commit 4d18ba5765c206bf9f37634f532d97dabd507a58)
Reviewed-on: http://git-master/r/103417
Reviewed-by: Automatic_Commit_Validation_User
Reviewed-by: Aleksandr Frid <afrid@nvidia.com>
Reviewed-by: Yu-Huan Hsu <yhsu@nvidia.com>

kernel/sched_rt.c

index af11778..046f429 100644 (file)
@@ -497,6 +497,7 @@ balanced:
                 * runtime - in which case borrowing doesn't make sense.
                 */
                rt_rq->rt_runtime = RUNTIME_INF;
+               rt_rq->rt_throttled = 0;
                raw_spin_unlock(&rt_rq->rt_runtime_lock);
                raw_spin_unlock(&rt_b->rt_runtime_lock);
        }