[PATCH v5 3/5] sched: Remove double_rq_lock() from __migrate_task()

[Kirill Tkhai <ktkhai@parallels.com>]

Avoid double_rq_lock() and use TASK_ON_RQ_MIGRATING for __migrate_task().
The advantage is (obviously) not holding two 'rq->lock's at the same time
and thereby increasing parallelism.

The important point to note is that because we acquire dst->lock
immediately after releasing src->lock the potential wait time of
task_rq_lock() callers on TASK_ON_RQ_MIGRATING is not longer than it would
have been in the double rq lock scenario.

Signed-off-by: Kirill Tkhai <ktkhai@parallels.com>
---
 kernel/sched/core.c |   23 +++++++++++++++--------
 1 file changed, 15 insertions(+), 8 deletions(-)

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index cef1a13..a50393b 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -4663,20 +4663,20 @@ EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr);
  */
 static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu)
 {
-	struct rq *rq_dest, *rq_src;
+	struct rq *rq;
 	int ret = 0;
 
 	if (unlikely(!cpu_active(dest_cpu)))
 		return ret;
 
-	rq_src = cpu_rq(src_cpu);
-	rq_dest = cpu_rq(dest_cpu);
+	rq = cpu_rq(src_cpu);
 
 	raw_spin_lock(&p->pi_lock);
-	double_rq_lock(rq_src, rq_dest);
+	raw_spin_lock(&rq->lock);
 	/* Already moved. */
 	if (task_cpu(p) != src_cpu)
 		goto done;
+
 	/* Affinity changed (again). */
 	if (!cpumask_test_cpu(dest_cpu, tsk_cpus_allowed(p)))
 		goto fail;
@@ -4686,15 +4686,22 @@ static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu)
 	 * placed properly.
 	 */
 	if (task_on_rq_queued(p)) {
-		dequeue_task(rq_src, p, 0);
+		dequeue_task(rq, p, 0);
+		p->on_rq = TASK_ON_RQ_MIGRATING;
 		set_task_cpu(p, dest_cpu);
-		enqueue_task(rq_dest, p, 0);
-		check_preempt_curr(rq_dest, p, 0);
+		raw_spin_unlock(&rq->lock);
+
+		rq = cpu_rq(dest_cpu);
+		raw_spin_lock(&rq->lock);
+		BUG_ON(task_rq(p) != rq);
+		p->on_rq = TASK_ON_RQ_QUEUED;
+		enqueue_task(rq, p, 0);
+		check_preempt_curr(rq, p, 0);
 	}
 done:
 	ret = 1;
 fail:
-	double_rq_unlock(rq_src, rq_dest);
+	raw_spin_unlock(&rq->lock);
 	raw_spin_unlock(&p->pi_lock);
 	return ret;
 }