sched/rt: Fix rq's cpupri leak while enqueue/dequeue child RT entities
authorKirill Tkhai <tkhai@yandex.ru>
Wed, 27 Nov 2013 15:59:13 +0000 (19:59 +0400)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Thu, 9 Jan 2014 20:24:21 +0000 (12:24 -0800)
commit 757dfcaa41844595964f1220f1d33182dae49976 upstream.

This patch touches the RT group scheduling case.

Functions inc_rt_prio_smp() and dec_rt_prio_smp() change (global) rq's
priority, while rt_rq passed to them may be not the top-level rt_rq.
This is wrong, because changing of priority on a child level does not
guarantee that the priority is the highest all over the rq. So, this
leak makes RT balancing unusable.

The short example: the task having the highest priority among all rq's
RT tasks (no one other task has the same priority) are waking on a
throttle rt_rq.  The rq's cpupri is set to the task's priority
equivalent, but real rq->rt.highest_prio.curr is less.

The patch below fixes the problem.

Signed-off-by: Kirill Tkhai <tkhai@yandex.ru>
Signed-off-by: Peter Zijlstra <peterz@infradead.org>
CC: Steven Rostedt <rostedt@goodmis.org>
Link: http://lkml.kernel.org/r/49231385567953@web4m.yandex.ru
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
kernel/sched/rt.c

index 127a2c4..15334e6 100644 (file)
@@ -964,6 +964,13 @@ inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio)
 {
        struct rq *rq = rq_of_rt_rq(rt_rq);
 
+#ifdef CONFIG_RT_GROUP_SCHED
+       /*
+        * Change rq's cpupri only if rt_rq is the top queue.
+        */
+       if (&rq->rt != rt_rq)
+               return;
+#endif
        if (rq->online && prio < prev_prio)
                cpupri_set(&rq->rd->cpupri, rq->cpu, prio);
 }
@@ -973,6 +980,13 @@ dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio)
 {
        struct rq *rq = rq_of_rt_rq(rt_rq);
 
+#ifdef CONFIG_RT_GROUP_SCHED
+       /*
+        * Change rq's cpupri only if rt_rq is the top queue.
+        */
+       if (&rq->rt != rt_rq)
+               return;
+#endif
        if (rq->online && rt_rq->highest_prio.curr != prev_prio)
                cpupri_set(&rq->rd->cpupri, rq->cpu, rt_rq->highest_prio.curr);
 }