Re: [PATCH 2/3] sched,fair: Fix local starvation

[Mike Galbraith <umgwanakikbuti@gmail.com>]

On Sat, 2016-05-21 at 16:04 +0200, Mike Galbraith wrote:

> Wakees that were not migrated/normalized eat an unwanted min_vruntime,
> and likely take a size XXL latency hit.  Big box running master bled
> profusely under heavy load until I turned TTWU_QUEUE off.

The below made big box a happy camper again.

sched/fair: Move se->vruntime normalization state into struct sched_entity

Make ->vruntime normalization state explicit.

Signed-off-by: Mike Galbraith <umgwanakikbuti@gmail.com>
---
 include/linux/sched.h |    1 +
 kernel/sched/core.c   |    1 +
 kernel/sched/fair.c   |   42 ++++++++++++++----------------------------
 3 files changed, 16 insertions(+), 28 deletions(-)

--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -1348,6 +1348,7 @@ struct sched_entity {
 	struct rb_node		run_node;
 	struct list_head	group_node;
 	unsigned int		on_rq;
+	bool			normalized;
 
 	u64			exec_start;
 	u64			sum_exec_runtime;
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -2259,6 +2259,7 @@ static void __sched_fork(unsigned long c
 	p->se.prev_sum_exec_runtime	= 0;
 	p->se.nr_migrations		= 0;
 	p->se.vruntime			= 0;
+	p->se.normalized		= true;
 	INIT_LIST_HEAD(&p->se.group_node);
 
 #ifdef CONFIG_FAIR_GROUP_SCHED
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -3305,14 +3305,13 @@ static inline void check_schedstat_requi
 static void
 enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
 {
-	bool renorm = !(flags & ENQUEUE_WAKEUP) || (flags & ENQUEUE_MIGRATED);
 	bool curr = cfs_rq->curr == se;
 
 	/*
 	 * If we're the current task, we must renormalise before calling
 	 * update_curr().
 	 */
-	if (renorm && curr)
+	if (se->normalized && curr)
 		se->vruntime += cfs_rq->min_vruntime;
 
 	update_curr(cfs_rq);
@@ -3323,9 +3322,11 @@ enqueue_entity(struct cfs_rq *cfs_rq, st
 	 * placed in the past could significantly boost this task to the
 	 * fairness detriment of existing tasks.
 	 */
-	if (renorm && !curr)
+	if (se->normalized && !curr)
 		se->vruntime += cfs_rq->min_vruntime;
 
+	se->normalized = false;
+
 	enqueue_entity_load_avg(cfs_rq, se);
 	account_entity_enqueue(cfs_rq, se);
 	update_cfs_shares(cfs_rq);
@@ -3422,8 +3423,10 @@ dequeue_entity(struct cfs_rq *cfs_rq, st
 	 * update can refer to the ->curr item and we need to reflect this
 	 * movement in our normalized position.
 	 */
-	if (!(flags & DEQUEUE_SLEEP))
+	if (!(flags & DEQUEUE_SLEEP)) {
 		se->vruntime -= cfs_rq->min_vruntime;
+		se->normalized = true;
+	}
 
 	/* return excess runtime on last dequeue */
 	return_cfs_rq_runtime(cfs_rq);
@@ -5681,6 +5684,7 @@ static void migrate_task_rq_fair(struct
 #endif
 
 		se->vruntime -= min_vruntime;
+		se->normalized = true;
 	}
 
 	/*
@@ -8591,6 +8595,7 @@ static void task_fork_fair(struct task_s
 	}
 
 	se->vruntime -= cfs_rq->min_vruntime;
+	se->normalized = true;
 
 	raw_spin_unlock_irqrestore(&rq->lock, flags);
 }
@@ -8619,29 +8624,7 @@ prio_changed_fair(struct rq *rq, struct
 
 static inline bool vruntime_normalized(struct task_struct *p)
 {
-	struct sched_entity *se = &p->se;
-
-	/*
-	 * In both the TASK_ON_RQ_QUEUED and TASK_ON_RQ_MIGRATING cases,
-	 * the dequeue_entity(.flags=0) will already have normalized the
-	 * vruntime.
-	 */
-	if (p->on_rq)
-		return true;
-
-	/*
-	 * When !on_rq, vruntime of the task has usually NOT been normalized.
-	 * But there are some cases where it has already been normalized:
-	 *
-	 * - A forked child which is waiting for being woken up by
-	 *   wake_up_new_task().
-	 * - A task which has been woken up by try_to_wake_up() and
-	 *   waiting for actually being woken up by sched_ttwu_pending().
-	 */
-	if (!se->sum_exec_runtime || p->state == TASK_WAKING)
-		return true;
-
-	return false;
+	return p->se.normalized;
 }
 
 static void detach_task_cfs_rq(struct task_struct *p)
@@ -8656,6 +8639,7 @@ static void detach_task_cfs_rq(struct ta
 		 */
 		place_entity(cfs_rq, se, 0);
 		se->vruntime -= cfs_rq->min_vruntime;
+		se->normalized = true;
 	}
 
 	/* Catch up with the cfs_rq and remove our load when we leave */
@@ -8678,8 +8662,10 @@ static void attach_task_cfs_rq(struct ta
 	/* Synchronize task with its cfs_rq */
 	attach_entity_load_avg(cfs_rq, se);
 
-	if (!vruntime_normalized(p))
+	if (vruntime_normalized(p)) {
 		se->vruntime += cfs_rq->min_vruntime;
+		se->normalized = false;
+	}
 }
 
 static void switched_from_fair(struct rq *rq, struct task_struct *p)