sched: Fix task affinity for select_task_rq_fair

While merging select_task_rq_fair() and sched_balance_self() I made
a mistake that leads to testing the wrong task affinty.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This commit is contained in:
Peter Zijlstra 2009-09-11 12:45:38 +02:00 committed by Ingo Molnar
parent 83f54960c1
commit d7c33c4930

View file

@ -1318,7 +1318,6 @@ find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu)
*/
static int select_task_rq_fair(struct task_struct *p, int flag, int sync)
{
struct task_struct *t = current;
struct sched_domain *tmp, *sd = NULL;
int cpu = smp_processor_id();
int prev_cpu = task_cpu(p);
@ -1393,13 +1392,13 @@ static int select_task_rq_fair(struct task_struct *p, int flag, int sync)
continue;
}
group = find_idlest_group(sd, t, cpu);
group = find_idlest_group(sd, p, cpu);
if (!group) {
sd = sd->child;
continue;
}
new_cpu = find_idlest_cpu(group, t, cpu);
new_cpu = find_idlest_cpu(group, p, cpu);
if (new_cpu == -1 || new_cpu == cpu) {
/* Now try balancing at a lower domain level of cpu */
sd = sd->child;