sched: fix buddies for group scheduling
authorPeter Zijlstra <a.p.zijlstra@chello.nl>
Tue, 4 Nov 2008 20:25:10 +0000 (21:25 +0100)
committerIngo Molnar <mingo@elte.hu>
Wed, 5 Nov 2008 09:30:15 +0000 (10:30 +0100)
Impact: scheduling order fix for group scheduling

For each level in the hierarchy, set the buddy to point to the right entity.
Therefore, when we do the hierarchical schedule, we have a fair chance of
ending up where we meant to.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Acked-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
kernel/sched_fair.c

index a6b1db8..51aa3e1 100644 (file)
@@ -1319,6 +1319,18 @@ wakeup_preempt_entity(struct sched_entity *curr, struct sched_entity *se)
        return 0;
 }
 
+static void set_last_buddy(struct sched_entity *se)
+{
+       for_each_sched_entity(se)
+               cfs_rq_of(se)->last = se;
+}
+
+static void set_next_buddy(struct sched_entity *se)
+{
+       for_each_sched_entity(se)
+               cfs_rq_of(se)->next = se;
+}
+
 /*
  * Preempt the current task with a newly woken task if needed:
  */
@@ -1352,8 +1364,8 @@ static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int sync)
         * obvious reasons its a bad idea to schedule back to the idle thread.
         */
        if (sched_feat(LAST_BUDDY) && likely(se->on_rq && curr != rq->idle))
-               cfs_rq_of(se)->last = se;
-       cfs_rq_of(pse)->next = pse;
+               set_last_buddy(se);
+       set_next_buddy(pse);
 
        /*
         * We can come here with TIF_NEED_RESCHED already set from new task