[PATCH 03/20 v3] sched/eevdf: Reset lag at wakeup when migrating and prev has been idle

[Vincent Guittot]

When prev CPU entered idle while a task was sleeping, we assume that the
latter received enough running time and the lag can be cleared similarly to
what happens for a negative lag.
While migrating the waking task to a new cpu, Check if prev CPU entered
idle in the meantime.

Signed-off-by: Vincent Guittot <vincent.guittot@xxxxxxxxxx>
---
 kernel/sched/fair.c | 27 +++++++++++++++++++++++++++
 1 file changed, 27 insertions(+)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 79f53811274e..945de0215abb 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -932,6 +932,27 @@ bool update_entity_lag(struct cfs_rq *cfs_rq, struct sched_entity *se)
 static inline unsigned long cfs_rq_load_avg(struct cfs_rq *cfs_rq);
 static unsigned long task_h_load(struct task_struct *p);
 
+static __always_inline
+void entity_clear_lag(struct task_struct *p)
+{
+	struct sched_entity *se;
+	struct cfs_rq *cfs_rq;
+
+	if (READ_ONCE(p->__state) != TASK_WAKING)
+		return;
+
+	se = &p->se;
+	/* Negative lag implies delayed dequeue */
+	if (se->vlag <= 0)
+		return;
+
+	cfs_rq = &task_rq(p)->cfs;
+
+	/* CPU has been idle in between so the lag has been removed */
+	if (se->vlag_seq != READ_ONCE(cfs_rq->idle_seq))
+		se->vlag = 0;
+}
+
 static __always_inline
 void decay_entity_lag(struct cfs_rq *cfs_rq, struct task_struct *p, int flags)
 {
@@ -10049,6 +10070,12 @@ static void migrate_task_rq_fair(struct task_struct *p, int new_cpu)
 	struct sched_entity *se = &p->se;
 
 	if (!task_on_rq_migrating(p)) {
+		/*
+		 * Clear the positive lag of a waking task before migrating to
+		 * the new cpu.
+		 */
+		entity_clear_lag(p);
+
 		remove_entity_load_avg(se);
 
 		/*
-- 
2.53.0