[PATCH]sched/rt: Reduce race between inc_rt_group and do_sched_rt_period_timer

[Kirill Tkhai]

Function do_sched_rt_period_timer() iterates through runqueues sequentially.
This makes it racy with inc_rt_group(). Sometimes when timer callback is
running we may skip hrtimer restart.

On the one hand it's not good to skip bandwidth timer. On the other hand,
there is no simple fix to complitelly get rid of the race.

So, let's use a fix, which is fast and simple and almost complitelly
reduces the race. It makes race probability very small.

Signed-off-by: Kirill Tkhai <tkhai@xxxxxxxxx>
CC: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
CC: Ingo Molnar <mingo@xxxxxxxxxx>
---
 kernel/sched/rt.c | 16 ++++++++++++++--
 1 file changed, 14 insertions(+), 2 deletions(-)
diff --git a/kernel/sched/rt.c b/kernel/sched/rt.c
index facc824..db5303f 100644
--- a/kernel/sched/rt.c
+++ b/kernel/sched/rt.c
@@ -923,11 +923,23 @@ static void update_curr_rt(struct rq *rq)
 		rt_rq = rt_rq_of_se(rt_se);
 
 		if (sched_rt_runtime(rt_rq) != RUNTIME_INF) {
+			bool exceed;
 			raw_spin_lock(&rt_rq->rt_runtime_lock);
 			rt_rq->rt_time += delta_exec;
-			if (sched_rt_runtime_exceeded(rt_rq))
-				resched_task(curr);
+			exceed = sched_rt_runtime_exceeded(rt_rq);
 			raw_spin_unlock(&rt_rq->rt_runtime_lock);
+
+			if (exceed) {
+				resched_task(curr);
+				/*
+				 * The pair do_sched_rt_period_timer() and
+				 * inc_rt_group() is racy. Sometimes we may
+				 * have enqueued task and stopped timer
+				 * at this point. We reduce the race and
+				 * make its probability extremely small.
+				 */
+				start_rt_bandwidth(sched_rt_bandwidth(rt_rq));
+			}
 		}
 	}
 }
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/