The code in changeset 83230590 can occasionally get tripped up
by the "src_rq->nr_running == 1 && dst_rq->nr_running == 1"
check, and lead to undesirable and/or useless task moves, as
well as idle CPUs.
Luckily this bug can be avoided in a way that also simplifies
the code.
Signed-off-by: Rik van Riel <r...@redhat.com>
---
kernel/sched/fair.c | 33 +++++++++++++--------------------
1 file changed, 13 insertions(+), 20 deletions(-)
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 376bc07c..32edd2c56 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -1159,7 +1159,7 @@ static void task_numa_compare(struct task_numa_env *env,
long src_load, dst_load;
long load;
long imp = env->p->numa_group ? groupimp : taskimp;
- long moveimp = imp;
+ long moveimp = imp - 1;
rcu_read_lock();
cur = ACCESS_ONCE(dst_rq->curr);
@@ -1206,8 +1206,18 @@ static void task_numa_compare(struct task_numa_env *env,
}
}
- if (imp <= env->best_imp && moveimp <= env->best_imp)
- goto unlock;
+ if (imp <= env->best_imp) {
+ if (moveimp <= env->best_imp)
+ goto unlock;
+
+ /*
+ * A task swap is not going to work; a task move may be
+ * required to consolidate this workload, especially if
+ * both nodes are overloaded and there are no idle CPUs.
+ */
+ imp = moveimp;
+ cur = NULL;
+ }
if (!cur) {
/* Is there capacity at our destination? */
@@ -1231,23 +1241,6 @@ static void task_numa_compare(struct task_numa_env *env,
dst_load = env->dst_stats.load + load;
src_load = env->src_stats.load - load;
- if (moveimp > imp && moveimp > env->best_imp) {
- /*
- * If the improvement from just moving env->p direction is
- * better than swapping tasks around, check if a move is
- * possible. Store a slightly smaller score than moveimp,
- * so an actually idle CPU will win.
- */
- if (!load_too_imbalanced(src_load, dst_load, env)) {
- imp = moveimp - 1;
- cur = NULL;
- goto assign;
- }
- }
-
- if (imp <= env->best_imp)
- goto unlock;
-
if (cur) {
load = task_h_load(cur);
dst_load -= load;
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
