The commit 67692435c411 ("sched: Rework pick_next_task() slow-path") and
commit 6e2df0581f56 ("sched: Fix pick_next_task() vs 'change' pattern race")
changed the original pick_next_task() loop logic a little.

Previously we always do class->pick_next_task(rq, prev, rf) from highest
to lowest, and balance happened in that class->pick_next_task(), so we
don't need to pull RT task if we can pick a DL task.

The current code do the balancing pass from prev->class, say prev task
is a RT task to sleep, we will try balance for rt sched_class even though
we have a DL task on rq.

This patch changes to always do the balancing pass from dl_sched_class,
since we will check if balance is really needed in each class->balance(),
we can terminate the balance pass as soon as we know there is a runnable
task of @class priority or higher.

Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com>
---
 kernel/sched/core.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index ee28253c9ac0..ff5e7ac236bd 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -5774,7 +5774,7 @@ static void put_prev_task_balance(struct rq *rq, struct task_struct *prev,
 	 * We can terminate the balance pass as soon as we know there is
 	 * a runnable task of @class priority or higher.
 	 */
-	for_class_range(class, prev->sched_class, &idle_sched_class) {
+	for_class_range(class, &dl_sched_class, &idle_sched_class) {
 		if (class->balance(rq, prev, rf))
 			break;
 	}
-- 
2.37.2