sched/rt: Do not throttle when PI boosting
authorPeter Zijlstra <peterz@infradead.org>
Tue, 18 Oct 2011 20:03:48 +0000 (22:03 +0200)
committerIngo Molnar <mingo@elte.hu>
Thu, 1 Mar 2012 09:28:02 +0000 (10:28 +0100)
When a runqueue has rt_runtime_us = 0 then the only way it can
accumulate rt_time is via PI boosting. That causes the runqueue
to be throttled and replenishing does not change anything due to
rt_runtime_us = 0. So avoid that situation by clearing rt_time and
skip the throttling alltogether.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
[ Changelog ]
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/n/tip-7x70cypsotjb4jvcor3edctk@git.kernel.org
Signed-off-by: Ingo Molnar <mingo@elte.hu>
kernel/sched/rt.c

index 6d1eb0b..7f7e7cd 100644 (file)
@@ -857,8 +857,24 @@ static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq)
                return 0;
 
        if (rt_rq->rt_time > runtime) {
-               rt_rq->rt_throttled = 1;
-               printk_once(KERN_WARNING "sched: RT throttling activated\n");
+               struct rt_bandwidth *rt_b = sched_rt_bandwidth(rt_rq);
+
+               /*
+                * Don't actually throttle groups that have no runtime assigned
+                * but accrue some time due to boosting.
+                */
+               if (likely(rt_b->rt_runtime)) {
+                       rt_rq->rt_throttled = 1;
+                       printk_once(KERN_WARNING "sched: RT throttling activated\n");
+               } else {
+                       /*
+                        * In case we did anyway, make it go away,
+                        * replenishment is a joke, since it will replenish us
+                        * with exactly 0 ns.
+                        */
+                       rt_rq->rt_time = 0;
+               }
+
                if (rt_rq_throttled(rt_rq)) {
                        sched_rt_rq_dequeue(rt_rq);
                        return 1;