sched: sync wakeups vs avg_overlap
authorPeter Zijlstra <peterz@infradead.org>
Wed, 8 Oct 2008 07:16:04 +0000 (09:16 +0200)
committerIngo Molnar <mingo@elte.hu>
Wed, 8 Oct 2008 10:20:26 +0000 (12:20 +0200)
While looking at the code I wondered why we always do:

  sync && avg_overlap < migration_cost

Which is a bit odd, since the overlap test was meant to detect sync wakeups
so using it to specialize sync wakeups doesn't make much sense.

Hence change the code to do:

  sync || avg_overlap < migration_cost

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

index fcbe850..18fd171 100644 (file)
@@ -1103,6 +1103,11 @@ wake_affine(struct sched_domain *this_sd, struct rq *this_rq,
        if (!(this_sd->flags & SD_WAKE_AFFINE) || !sched_feat(AFFINE_WAKEUPS))
                return 0;
 
+       if (!sync && sched_feat(SYNC_WAKEUPS) &&
+           curr->se.avg_overlap < sysctl_sched_migration_cost &&
+           p->se.avg_overlap < sysctl_sched_migration_cost)
+               sync = 1;
+
        /*
         * If sync wakeup then subtract the (maximum possible)
         * effect of the currently running task from the load
@@ -1127,11 +1132,8 @@ wake_affine(struct sched_domain *this_sd, struct rq *this_rq,
         * a reasonable amount of time then attract this newly
         * woken task:
         */
-       if (sync && balanced) {
-               if (curr->se.avg_overlap < sysctl_sched_migration_cost &&
-                   p->se.avg_overlap < sysctl_sched_migration_cost)
-                       return 1;
-       }
+       if (sync && balanced)
+               return 1;
 
        schedstat_inc(p, se.nr_wakeups_affine_attempts);
        tl_per_task = cpu_avg_load_per_task(this_cpu);
@@ -1268,9 +1270,9 @@ static void check_preempt_wakeup(struct rq *rq, struct task_struct *p, int sync)
        if (!sched_feat(WAKEUP_PREEMPT))
                return;
 
-       if (sched_feat(WAKEUP_OVERLAP) && sync &&
-                       se->avg_overlap < sysctl_sched_migration_cost &&
-                       pse->avg_overlap < sysctl_sched_migration_cost) {
+       if (sched_feat(WAKEUP_OVERLAP) && (sync ||
+                       (se->avg_overlap < sysctl_sched_migration_cost &&
+                        pse->avg_overlap < sysctl_sched_migration_cost))) {
                resched_task(curr);
                return;
        }