]> err.no Git - linux-2.6/commitdiff
sched: remove prio preference from balance decisions
authorPeter Zijlstra <a.p.zijlstra@chello.nl>
Fri, 27 Jun 2008 11:41:31 +0000 (13:41 +0200)
committerIngo Molnar <mingo@elte.hu>
Fri, 27 Jun 2008 12:31:42 +0000 (14:31 +0200)
Priority looses much of its meaning in a hierarchical context. So don't
use it in balance decisions.

Signed-off-by: Peter Zijlstra <peterz@infradead.org>
Cc: Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com>
Cc: Mike Galbraith <efault@gmx.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
kernel/sched.c

index 5e2aa394a81259bcb678f2cf4a7c3ca276aa86ca..10d43f5bf0fcee2d57ee2bb3676942480e4eeb3d 100644 (file)
@@ -2896,7 +2896,7 @@ balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest,
              enum cpu_idle_type idle, int *all_pinned,
              int *this_best_prio, struct rq_iterator *iterator)
 {
-       int loops = 0, pulled = 0, pinned = 0, skip_for_load;
+       int loops = 0, pulled = 0, pinned = 0;
        struct task_struct *p;
        long rem_load_move = max_load_move;
 
@@ -2912,14 +2912,8 @@ balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest,
 next:
        if (!p || loops++ > sysctl_sched_nr_migrate)
                goto out;
-       /*
-        * To help distribute high priority tasks across CPUs we don't
-        * skip a task if it will be the highest priority task (i.e. smallest
-        * prio value) on its new queue regardless of its load weight
-        */
-       skip_for_load = (p->se.load.weight >> 1) > rem_load_move +
-                                                        SCHED_LOAD_SCALE_FUZZ;
-       if ((skip_for_load && p->prio >= *this_best_prio) ||
+
+       if ((p->se.load.weight >> 1) > rem_load_move ||
            !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) {
                p = iterator->next(iterator->arg);
                goto next;