sched: Ensure a task has a non-normalized vruntime when returning back to CFS
authorByungchul Park <byungchul.park@lge.com>
Mon, 10 Aug 2015 09:02:55 +0000 (18:02 +0900)
committerIngo Molnar <mingo@kernel.org>
Wed, 12 Aug 2015 10:06:09 +0000 (12:06 +0200)
Current code ensures that a task has a normalized vruntime when switching away
from the fair class, but it does not ensure the task has a non-normalized
vruntime when switching back to the fair class.

This is an example breaking this consistency:

  1. a task is in fair class and !queued
  2. changes its class to RT class (still !queued)
  3. changes its class to fair class again (still !queued)

Signed-off-by: Byungchul Park <byungchul.park@lge.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: http://lkml.kernel.org/r/1439197375-27927-1-git-send-email-byungchul.park@lge.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
kernel/sched/fair.c

index 858b94ab1bd2a46ba585b1eeffcaf43e7e64790e..f0950fde1f5ba3b1004fd25b2ba7bba59eba86ff 100644 (file)
@@ -7930,16 +7930,31 @@ static void switched_from_fair(struct rq *rq, struct task_struct *p)
  */
 static void switched_to_fair(struct rq *rq, struct task_struct *p)
 {
-#ifdef CONFIG_FAIR_GROUP_SCHED
        struct sched_entity *se = &p->se;
+
+#ifdef CONFIG_FAIR_GROUP_SCHED
        /*
         * Since the real-depth could have been changed (only FAIR
         * class maintain depth value), reset depth properly.
         */
        se->depth = se->parent ? se->parent->depth + 1 : 0;
 #endif
-       if (!task_on_rq_queued(p))
+
+       if (!task_on_rq_queued(p)) {
+
+               /*
+                * Ensure the task has a non-normalized vruntime when it is switched
+                * back to the fair class with !queued, so that enqueue_entity() at
+                * wake-up time will do the right thing.
+                *
+                * If it's queued, then the enqueue_entity(.flags=0) makes the task
+                * has non-normalized vruntime, if it's !queued, then it still has
+                * normalized vruntime.
+                */
+               if (p->state != TASK_RUNNING)
+                       se->vruntime += cfs_rq_of(se)->min_vruntime;
                return;
+       }
 
        /*
         * We were most likely switched from sched_rt, so