[PATCH 0/6 v5] sched: reflect sched_entity move into task_group's load

[PATCH 0/6 v5] sched: reflect sched_entity move into task_group's load

[Vincent Guittot]

Ensure that the move of a sched_entity will be reflected in load and
utilization of the task_group hierarchy.

When a sched_entity moves between groups or CPUs, load and utilization
of cfs_rq don't reflect the changes immediately but converge to new values.
As a result, the metrics are no more aligned with the new balance of the
load in the system and next decisions will have a biased view.

This patchset synchronizes load/utilization of sched_entity with its child
cfs_rq (se->my-q) only when tasks move to/from child cfs_rq:
-move between task group
-migration between CPUs
Otherwise, PELT is updated as usual.

This version doesn't include any changes related to discussion that have
started during the review of the previous version about:
- encapsulate the sequence for changing the propoerty of a task
- remove a cfs_rq from list during update_blocked_averages  
These topics don't gain anything from being added in this patchset as they
are fairly independent and deserve a separate patch.

Changes since v4:
- minor typo and commit message changes
- move call to cfs_rq_clock_task(cfs_rq) in post_init_entity_util_avg

Changes since v3:
- Replaced the 2 arguments of update_load_avg by 1 flags argument
- Propagated move in runnable_load_avg when sched_entity is already on_rq
- Ensure that intermediate value will not reach memory when updating load and
  utilization
- Optimize the the calculation of load_avg of the sched_entity
- Fixed some typo

Changes since v2:
- Propagate both utilization and load
- Synced sched_entity and se->my_q instead of adding the delta

Changes since v1:
- This patch needs the patch that fixes issue with rq->leaf_cfs_rq_list
  "sched: fix hierarchical order in rq->leaf_cfs_rq_list" in order to work
  correctly. I haven't sent them as a single patchset because the fix is
  independent of this one
- Merge some functions that are always used together
- During update of blocked load, ensure that the sched_entity is synced
  with the cfs_rq applying changes
- Fix an issue when task changes its cpu affinity

Vincent Guittot (6):
  sched: factorize attach entity
  sched: fix hierarchical order in rq->leaf_cfs_rq_list
  sched: factorize PELT update
  sched: propagate load during synchronous attach/detach
  sched: propagate asynchrous detach
  sched: fix task group initialization

 kernel/sched/core.c  |   1 +
 kernel/sched/fair.c  | 359 +++++++++++++++++++++++++++++++++++++++++----------
 kernel/sched/sched.h |   2 +
 3 files changed, 292 insertions(+), 70 deletions(-)

-- 
2.7.4

[PATCH 1/6 v5] sched: factorize attach entity

[Vincent Guittot]

Factorize post_init_entity_util_avg and part of attach_task_cfs_rq
in one function attach_entity_cfs_rq

Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
---
 kernel/sched/fair.c | 24 +++++++++++++-----------
 1 file changed, 13 insertions(+), 11 deletions(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 543b2f2..584b86f 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -701,9 +701,7 @@ void init_entity_runnable_average(struct sched_entity *se)
 }
 
 static inline u64 cfs_rq_clock_task(struct cfs_rq *cfs_rq);
-static int update_cfs_rq_load_avg(u64 now, struct cfs_rq *cfs_rq, bool update_freq);
-static void update_tg_load_avg(struct cfs_rq *cfs_rq, int force);
-static void attach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se);
+static void attach_entity_cfs_rq(struct sched_entity *se);
 
 /*
  * With new tasks being created, their initial util_avgs are extrapolated
@@ -735,7 +733,6 @@ void post_init_entity_util_avg(struct sched_entity *se)
 	struct cfs_rq *cfs_rq = cfs_rq_of(se);
 	struct sched_avg *sa = &se->avg;
 	long cap = (long)(SCHED_CAPACITY_SCALE - cfs_rq->avg.util_avg) / 2;
-	u64 now = cfs_rq_clock_task(cfs_rq);
 
 	if (cap > 0) {
 		if (cfs_rq->avg.util_avg != 0) {
@@ -763,14 +760,12 @@ void post_init_entity_util_avg(struct sched_entity *se)
 			 * such that the next switched_to_fair() has the
 			 * expected state.
 			 */
-			se->avg.last_update_time = now;
+			se->avg.last_update_time = cfs_rq_clock_task(cfs_rq);
 			return;
 		}
 	}
 
-	update_cfs_rq_load_avg(now, cfs_rq, false);
-	attach_entity_load_avg(cfs_rq, se);
-	update_tg_load_avg(cfs_rq, false);
+	attach_entity_cfs_rq(se);
 }
 
 #else /* !CONFIG_SMP */
@@ -8697,9 +8692,8 @@ static void detach_task_cfs_rq(struct task_struct *p)
 	update_tg_load_avg(cfs_rq, false);
 }
 
-static void attach_task_cfs_rq(struct task_struct *p)
+static void attach_entity_cfs_rq(struct sched_entity *se)
 {
-	struct sched_entity *se = &p->se;
 	struct cfs_rq *cfs_rq = cfs_rq_of(se);
 	u64 now = cfs_rq_clock_task(cfs_rq);
 
@@ -8711,10 +8705,18 @@ static void attach_task_cfs_rq(struct task_struct *p)
 	se->depth = se->parent ? se->parent->depth + 1 : 0;
 #endif
 
-	/* Synchronize task with its cfs_rq */
+	/* Synchronize entity with its cfs_rq */
 	update_cfs_rq_load_avg(now, cfs_rq, false);
 	attach_entity_load_avg(cfs_rq, se);
 	update_tg_load_avg(cfs_rq, false);
+}
+
+static void attach_task_cfs_rq(struct task_struct *p)
+{
+	struct sched_entity *se = &p->se;
+	struct cfs_rq *cfs_rq = cfs_rq_of(se);
+
+	attach_entity_cfs_rq(se);
 
 	if (!vruntime_normalized(p))
 		se->vruntime += cfs_rq->min_vruntime;
-- 
2.7.4

[PATCH 2/6 v5] sched: fix hierarchical order in rq->leaf_cfs_rq_list

[Vincent Guittot]

Fix the insertion of cfs_rq in rq->leaf_cfs_rq_list to ensure that
a child will always be called before its parent.

The hierarchical order in shares update list has been introduced by
commit 67e86250f8ea ("sched: Introduce hierarchal order on shares update list")

With the current implementation a child can be still put after its parent.

Lets take the example of
       root
        \
         b
         /\
         c d*
           |
           e*

with root -> b -> c already enqueued but not d -> e so the leaf_cfs_rq_list
looks like: head -> c -> b -> root -> tail

The branch d -> e will be added the first time that they are enqueued,
starting with e then d.

When e is added, its parents is not already on the list so e is put at the
tail : head -> c -> b -> root -> e -> tail

Then, d is added at the head because its parent is already on the list:
head -> d -> c -> b -> root -> e -> tail

e is not placed at the right position and will be called the last whereas
it should be called at the beginning.

Because it follows the bottom-up enqueue sequence, we are sure that we
will finished to add either a cfs_rq without parent or a cfs_rq with a parent
that is already on the list. We can use this event to detect when we have
finished to add a new branch. For the others, whose parents are not already
added, we have to ensure that they will be added after their children that
have just been inserted the steps before, and after any potential parents that
are already in the list. The easiest way is to put the cfs_rq just after the
last inserted one and to keep track of it untl the branch is fully added.

Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
---
 kernel/sched/core.c  |  1 +
 kernel/sched/fair.c  | 54 +++++++++++++++++++++++++++++++++++++++++++++-------
 kernel/sched/sched.h |  1 +
 3 files changed, 49 insertions(+), 7 deletions(-)

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index aae08ce..59bce4f 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -7582,6 +7582,7 @@ void __init sched_init(void)
 #ifdef CONFIG_FAIR_GROUP_SCHED
 		root_task_group.shares = ROOT_TASK_GROUP_LOAD;
 		INIT_LIST_HEAD(&rq->leaf_cfs_rq_list);
+		rq->tmp_alone_branch = &rq->leaf_cfs_rq_list;
 		/*
 		 * How much cpu bandwidth does root_task_group get?
 		 *
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 584b86f..ecc66d6 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -290,19 +290,59 @@ static inline struct cfs_rq *group_cfs_rq(struct sched_entity *grp)
 static inline void list_add_leaf_cfs_rq(struct cfs_rq *cfs_rq)
 {
 	if (!cfs_rq->on_list) {
+		struct rq *rq = rq_of(cfs_rq);
+		int cpu = cpu_of(rq);
 		/*
 		 * Ensure we either appear before our parent (if already
 		 * enqueued) or force our parent to appear after us when it is
-		 * enqueued.  The fact that we always enqueue bottom-up
-		 * reduces this to two cases.
+		 * enqueued. The fact that we always enqueue bottom-up
+		 * reduces this to two cases and a special case for the root
+		 * cfs_rq. Furthermore, it also means that we will always reset
+		 * tmp_alone_branch either when the branch is connected
+		 * to a tree or when we reach the beg of the tree
 		 */
 		if (cfs_rq->tg->parent &&
-		    cfs_rq->tg->parent->cfs_rq[cpu_of(rq_of(cfs_rq))]->on_list) {
-			list_add_rcu(&cfs_rq->leaf_cfs_rq_list,
-				&rq_of(cfs_rq)->leaf_cfs_rq_list);
-		} else {
+		    cfs_rq->tg->parent->cfs_rq[cpu]->on_list) {
+			/*
+			 * If parent is already on the list, we add the child
+			 * just before. Thanks to circular linked property of
+			 * the list, this means to put the child at the tail
+			 * of the list that starts by parent.
+			 */
+			list_add_tail_rcu(&cfs_rq->leaf_cfs_rq_list,
+				&(cfs_rq->tg->parent->cfs_rq[cpu]->leaf_cfs_rq_list));
+			/*
+			 * The branch is now connected to its tree so we can
+			 * reset tmp_alone_branch to the beginning of the
+			 * list.
+			 */
+			rq->tmp_alone_branch = &rq->leaf_cfs_rq_list;
+		} else if (!cfs_rq->tg->parent) {
+			/*
+			 * cfs rq without parent should be put
+			 * at the tail of the list.
+			 */
 			list_add_tail_rcu(&cfs_rq->leaf_cfs_rq_list,
-				&rq_of(cfs_rq)->leaf_cfs_rq_list);
+				&rq->leaf_cfs_rq_list);
+			/*
+			 * We have reach the beg of a tree so we can reset
+			 * tmp_alone_branch to the beginning of the list.
+			 */
+			rq->tmp_alone_branch = &rq->leaf_cfs_rq_list;
+		} else {
+			/*
+			 * The parent has not already been added so we want to
+			 * make sure that it will be put after us.
+			 * tmp_alone_branch points to the beg of the branch
+			 * where we will add parent.
+			 */
+			list_add_rcu(&cfs_rq->leaf_cfs_rq_list,
+				rq->tmp_alone_branch);
+			/*
+			 * update tmp_alone_branch to points to the new beg
+			 * of the branch
+			 */
+			rq->tmp_alone_branch = &cfs_rq->leaf_cfs_rq_list;
 		}
 
 		cfs_rq->on_list = 1;
diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
index 19b9986..f1185c3 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -623,6 +623,7 @@ struct rq {
 #ifdef CONFIG_FAIR_GROUP_SCHED
 	/* list of leaf cfs_rq on this cpu: */
 	struct list_head leaf_cfs_rq_list;
+	struct list_head *tmp_alone_branch;
 #endif /* CONFIG_FAIR_GROUP_SCHED */
 
 	/*
-- 
2.7.4

[PATCH 3/6 v5] sched: factorize PELT update

[Vincent Guittot]

Every time, we modify load/utilization of sched_entity, we start to sync
it with its cfs_rq. This update is done is different ways:
-when attaching/detaching a sched_entity, we update cfs_rq and then we
sync the entity with the cfs_rq.
-when enqueueing/dequeuing the sched_entity, we update both sched_entity
and cfs_rq metrics to now.

Use update_load_avg everytime we have to update and sync cfs_rq and
sched_entity before changing the state of a sched_enity

Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
---
 kernel/sched/fair.c | 75 +++++++++++++++++------------------------------------
 1 file changed, 24 insertions(+), 51 deletions(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index ecc66d6..52cbc92 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -3098,8 +3098,14 @@ update_cfs_rq_load_avg(u64 now, struct cfs_rq *cfs_rq, bool update_freq)
 	return decayed || removed_load;
 }
 
+/*
+ * Optional action to be done while updating the load average
+ */
+#define UPDATE_TG	0x1
+#define SKIP_AGE_LOAD	0x2
+
 /* Update task and its cfs_rq load average */
-static inline void update_load_avg(struct sched_entity *se, int update_tg)
+static inline void update_load_avg(struct sched_entity *se, int flags)
 {
 	struct cfs_rq *cfs_rq = cfs_rq_of(se);
 	u64 now = cfs_rq_clock_task(cfs_rq);
@@ -3110,11 +3116,12 @@ static inline void update_load_avg(struct sched_entity *se, int update_tg)
 	 * Track task load average for carrying it to new CPU after migrated, and
 	 * track group sched_entity load average for task_h_load calc in migration
 	 */
-	__update_load_avg(now, cpu, &se->avg,
+	if (se->avg.last_update_time && !(flags & SKIP_AGE_LOAD))
+		__update_load_avg(now, cpu, &se->avg,
 			  se->on_rq * scale_load_down(se->load.weight),
 			  cfs_rq->curr == se, NULL);
 
-	if (update_cfs_rq_load_avg(now, cfs_rq, true) && update_tg)
+	if (update_cfs_rq_load_avg(now, cfs_rq, true) && (flags & UPDATE_TG))
 		update_tg_load_avg(cfs_rq, 0);
 }
 
@@ -3128,26 +3135,6 @@ static inline void update_load_avg(struct sched_entity *se, int update_tg)
  */
 static void attach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
 {
-	if (!sched_feat(ATTACH_AGE_LOAD))
-		goto skip_aging;
-
-	/*
-	 * If we got migrated (either between CPUs or between cgroups) we'll
-	 * have aged the average right before clearing @last_update_time.
-	 *
-	 * Or we're fresh through post_init_entity_util_avg().
-	 */
-	if (se->avg.last_update_time) {
-		__update_load_avg(cfs_rq->avg.last_update_time, cpu_of(rq_of(cfs_rq)),
-				  &se->avg, 0, 0, NULL);
-
-		/*
-		 * XXX: we could have just aged the entire load away if we've been
-		 * absent from the fair class for too long.
-		 */
-	}
-
-skip_aging:
 	se->avg.last_update_time = cfs_rq->avg.last_update_time;
 	cfs_rq->avg.load_avg += se->avg.load_avg;
 	cfs_rq->avg.load_sum += se->avg.load_sum;
@@ -3167,9 +3154,6 @@ static void attach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *s
  */
 static void detach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
 {
-	__update_load_avg(cfs_rq->avg.last_update_time, cpu_of(rq_of(cfs_rq)),
-			  &se->avg, se->on_rq * scale_load_down(se->load.weight),
-			  cfs_rq->curr == se, NULL);
 
 	sub_positive(&cfs_rq->avg.load_avg, se->avg.load_avg);
 	sub_positive(&cfs_rq->avg.load_sum, se->avg.load_sum);
@@ -3184,34 +3168,20 @@ static inline void
 enqueue_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
 {
 	struct sched_avg *sa = &se->avg;
-	u64 now = cfs_rq_clock_task(cfs_rq);
-	int migrated, decayed;
-
-	migrated = !sa->last_update_time;
-	if (!migrated) {
-		__update_load_avg(now, cpu_of(rq_of(cfs_rq)), sa,
-			se->on_rq * scale_load_down(se->load.weight),
-			cfs_rq->curr == se, NULL);
-	}
-
-	decayed = update_cfs_rq_load_avg(now, cfs_rq, !migrated);
 
 	cfs_rq->runnable_load_avg += sa->load_avg;
 	cfs_rq->runnable_load_sum += sa->load_sum;
 
-	if (migrated)
+	if (!sa->last_update_time) {
 		attach_entity_load_avg(cfs_rq, se);
-
-	if (decayed || migrated)
 		update_tg_load_avg(cfs_rq, 0);
+	}
 }
 
 /* Remove the runnable load generated by se from cfs_rq's runnable load average */
 static inline void
 dequeue_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *se)
 {
-	update_load_avg(se, 1);
-
 	cfs_rq->runnable_load_avg =
 		max_t(long, cfs_rq->runnable_load_avg - se->avg.load_avg, 0);
 	cfs_rq->runnable_load_sum =
@@ -3285,7 +3255,10 @@ update_cfs_rq_load_avg(u64 now, struct cfs_rq *cfs_rq, bool update_freq)
 	return 0;
 }
 
-static inline void update_load_avg(struct sched_entity *se, int not_used)
+#define UPDATE_TG	0x0
+#define SKIP_AGE_LOAD	0x0
+
+static inline void update_load_avg(struct sched_entity *se, int not_used1)
 {
 	struct cfs_rq *cfs_rq = cfs_rq_of(se);
 	struct rq *rq = rq_of(cfs_rq);
@@ -3433,6 +3406,7 @@ enqueue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
 	if (renorm && !curr)
 		se->vruntime += cfs_rq->min_vruntime;
 
+	update_load_avg(se, UPDATE_TG);
 	enqueue_entity_load_avg(cfs_rq, se);
 	account_entity_enqueue(cfs_rq, se);
 	update_cfs_shares(cfs_rq);
@@ -3507,6 +3481,7 @@ dequeue_entity(struct cfs_rq *cfs_rq, struct sched_entity *se, int flags)
 	 * Update run-time statistics of the 'current'.
 	 */
 	update_curr(cfs_rq);
+	update_load_avg(se, UPDATE_TG);
 	dequeue_entity_load_avg(cfs_rq, se);
 
 	update_stats_dequeue(cfs_rq, se, flags);
@@ -3594,7 +3569,7 @@ set_next_entity(struct cfs_rq *cfs_rq, struct sched_entity *se)
 		 */
 		update_stats_wait_end(cfs_rq, se);
 		__dequeue_entity(cfs_rq, se);
-		update_load_avg(se, 1);
+		update_load_avg(se, UPDATE_TG);
 	}
 
 	update_stats_curr_start(cfs_rq, se);
@@ -3712,7 +3687,7 @@ entity_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr, int queued)
 	/*
 	 * Ensure that runnable average is periodically updated.
 	 */
-	update_load_avg(curr, 1);
+	update_load_avg(curr, UPDATE_TG);
 	update_cfs_shares(cfs_rq);
 
 #ifdef CONFIG_SCHED_HRTICK
@@ -4601,7 +4576,7 @@ enqueue_task_fair(struct rq *rq, struct task_struct *p, int flags)
 		if (cfs_rq_throttled(cfs_rq))
 			break;
 
-		update_load_avg(se, 1);
+		update_load_avg(se, UPDATE_TG);
 		update_cfs_shares(cfs_rq);
 	}
 
@@ -4660,7 +4635,7 @@ static void dequeue_task_fair(struct rq *rq, struct task_struct *p, int flags)
 		if (cfs_rq_throttled(cfs_rq))
 			break;
 
-		update_load_avg(se, 1);
+		update_load_avg(se, UPDATE_TG);
 		update_cfs_shares(cfs_rq);
 	}
 
@@ -8715,7 +8690,6 @@ static void detach_task_cfs_rq(struct task_struct *p)
 {
 	struct sched_entity *se = &p->se;
 	struct cfs_rq *cfs_rq = cfs_rq_of(se);
-	u64 now = cfs_rq_clock_task(cfs_rq);
 
 	if (!vruntime_normalized(p)) {
 		/*
@@ -8727,7 +8701,7 @@ static void detach_task_cfs_rq(struct task_struct *p)
 	}
 
 	/* Catch up with the cfs_rq and remove our load when we leave */
-	update_cfs_rq_load_avg(now, cfs_rq, false);
+	update_load_avg(se, 0);
 	detach_entity_load_avg(cfs_rq, se);
 	update_tg_load_avg(cfs_rq, false);
 }
@@ -8735,7 +8709,6 @@ static void detach_task_cfs_rq(struct task_struct *p)
 static void attach_entity_cfs_rq(struct sched_entity *se)
 {
 	struct cfs_rq *cfs_rq = cfs_rq_of(se);
-	u64 now = cfs_rq_clock_task(cfs_rq);
 
 #ifdef CONFIG_FAIR_GROUP_SCHED
 	/*
@@ -8746,7 +8719,7 @@ static void attach_entity_cfs_rq(struct sched_entity *se)
 #endif
 
 	/* Synchronize entity with its cfs_rq */
-	update_cfs_rq_load_avg(now, cfs_rq, false);
+	update_load_avg(se, sched_feat(ATTACH_AGE_LOAD) ? 0 : SKIP_AGE_LOAD);
 	attach_entity_load_avg(cfs_rq, se);
 	update_tg_load_avg(cfs_rq, false);
 }
-- 
2.7.4

Re: [PATCH 3/6 v5] sched: factorize PELT update

[Peter Zijlstra]

On Mon, Oct 17, 2016 at 11:14:10AM +0200, Vincent Guittot wrote:
> @@ -3110,11 +3116,12 @@ static inline void update_load_avg(struct sched_entity *se, int update_tg)
>  	 * Track task load average for carrying it to new CPU after migrated, and
>  	 * track group sched_entity load average for task_h_load calc in migration
>  	 */
> -	__update_load_avg(now, cpu, &se->avg,
> +	if (se->avg.last_update_time && !(flags & SKIP_AGE_LOAD))
> +		__update_load_avg(now, cpu, &se->avg,
>  			  se->on_rq * scale_load_down(se->load.weight),
>  			  cfs_rq->curr == se, NULL);
>  

{ } for any multi line stmt pls.

Re: [PATCH 3/6 v5] sched: factorize PELT update

[Vincent Guittot]

On 26 October 2016 at 13:41, Peter Zijlstra <peterz@infradead.org> wrote:
> On Mon, Oct 17, 2016 at 11:14:10AM +0200, Vincent Guittot wrote:
>> @@ -3110,11 +3116,12 @@ static inline void update_load_avg(struct sched_entity *se, int update_tg)
>>        * Track task load average for carrying it to new CPU after migrated, and
>>        * track group sched_entity load average for task_h_load calc in migration
>>        */
>> -     __update_load_avg(now, cpu, &se->avg,
>> +     if (se->avg.last_update_time && !(flags & SKIP_AGE_LOAD))
>> +             __update_load_avg(now, cpu, &se->avg,
>>                         se->on_rq * scale_load_down(se->load.weight),
>>                         cfs_rq->curr == se, NULL);
>>
>
> { } for any multi line stmt pls.

OK

[PATCH 4/6 v5] sched: propagate load during synchronous attach/detach

[Vincent Guittot]

When a task moves from/to a cfs_rq, we set a flag which is then used to
propagate the change at parent level (sched_entity and cfs_rq) during
next update. If the cfs_rq is throttled, the flag will stay pending until
the cfs_rw is unthrottled.

For propagating the utilization, we copy the utilization of group cfs_rq to
the sched_entity.

For propagating the load, we have to take into account the load of the
whole task group in order to evaluate the load of the sched_entity.
Similarly to what was done before the rewrite of PELT, we add a correction
factor in case the task group's load is greater than its share so it will
contribute the same load of a task of equal weight.

Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
---
 kernel/sched/fair.c  | 200 ++++++++++++++++++++++++++++++++++++++++++++++++++-
 kernel/sched/sched.h |   1 +
 2 files changed, 200 insertions(+), 1 deletion(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 52cbc92..91fc949 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -3031,6 +3031,162 @@ static inline void cfs_rq_util_change(struct cfs_rq *cfs_rq)
 }
 
 /*
+ * Signed add and clamp on underflow.
+ *
+ * Explicitly do a load-store to ensure the intermediate value never hits
+ * memory. This allows lockless observations without ever seeing the negative
+ * values.
+ */
+#define add_positive(_ptr, _val) do {				\
+	typeof(_ptr) ptr = (_ptr);				\
+	typeof(_val) res, val = (_val);				\
+	typeof(*ptr) var = READ_ONCE(*ptr);			\
+	res = var + val;					\
+	if (res < 0)						\
+		res = 0;					\
+	WRITE_ONCE(*ptr, res);					\
+} while (0)
+
+#ifdef CONFIG_FAIR_GROUP_SCHED
+/* Take into account change of utilization of a child task group */
+static inline void
+update_tg_cfs_util(struct cfs_rq *cfs_rq, struct sched_entity *se)
+{
+	struct cfs_rq *gcfs_rq =  group_cfs_rq(se);
+	long delta = gcfs_rq->avg.util_avg - se->avg.util_avg;
+
+	/* Nothing to update */
+	if (!delta)
+		return;
+
+	/* Set new sched_entity's utilization */
+	se->avg.util_avg = gcfs_rq->avg.util_avg;
+	se->avg.util_sum = se->avg.util_avg * LOAD_AVG_MAX;
+
+	/* Update parent cfs_rq utilization */
+	add_positive(&cfs_rq->avg.util_avg, delta);
+	cfs_rq->avg.util_sum = cfs_rq->avg.util_avg * LOAD_AVG_MAX;
+}
+
+/* Take into account change of load of a child task group */
+static inline void
+update_tg_cfs_load(struct cfs_rq *cfs_rq, struct sched_entity *se)
+{
+	struct cfs_rq *gcfs_rq = group_cfs_rq(se);
+	long delta, load = gcfs_rq->avg.load_avg;
+
+	/*
+	 * If the load of group cfs_rq is null, the load of the
+	 * sched_entity will also be null so we can skip the formula
+	 */
+	if (load) {
+		long tg_load;
+
+		/* Get tg's load and ensure tg_load > 0 */
+		tg_load = atomic_long_read(&gcfs_rq->tg->load_avg) + 1;
+
+		/* Ensure tg_load >= load and updated with current load*/
+		tg_load -= gcfs_rq->tg_load_avg_contrib;
+		tg_load += load;
+
+		/*
+		 * We need to compute a correction term in the case that the
+		 * task group is consuming more cpu than a task of equal
+		 * weight. A task with a weight equals to tg->shares will have
+		 * a load less or equal to scale_load_down(tg->shares).
+		 * Similarly, the sched_entities that represent the task group
+		 * at parent level, can't have a load higher than
+		 * scale_load_down(tg->shares). And the Sum of sched_entities'
+		 * load must be <= scale_load_down(tg->shares).
+		 */
+		if (tg_load > scale_load_down(gcfs_rq->tg->shares)) {
+			/* scale gcfs_rq's load into tg's shares*/
+			load *= scale_load_down(gcfs_rq->tg->shares);
+			load /= tg_load;
+		}
+	}
+
+	delta = load - se->avg.load_avg;
+
+	/* Nothing to update */
+	if (!delta)
+		return;
+
+	/* Set new sched_entity's load */
+	se->avg.load_avg = load;
+	se->avg.load_sum = se->avg.load_avg * LOAD_AVG_MAX;
+
+	/* Update parent cfs_rq load */
+	add_positive(&cfs_rq->avg.load_avg, delta);
+	cfs_rq->avg.load_sum = cfs_rq->avg.load_avg * LOAD_AVG_MAX;
+
+	/*
+	 * If the sched_entity is already enqueued, we also have to update the
+	 * runnable load avg.
+	 */
+	if (se->on_rq) {
+		/* Update parent cfs_rq runnable_load_avg */
+		add_positive(&cfs_rq->runnable_load_avg, delta);
+		cfs_rq->runnable_load_sum = cfs_rq->runnable_load_avg * LOAD_AVG_MAX;
+	}
+}
+
+static inline void set_tg_cfs_propagate(struct cfs_rq *cfs_rq)
+{
+	/* set cfs_rq's flag */
+	cfs_rq->propagate_avg = 1;
+}
+
+static inline int test_and_clear_tg_cfs_propagate(struct sched_entity *se)
+{
+	/* Get my cfs_rq */
+	struct cfs_rq *cfs_rq = group_cfs_rq(se);
+
+	/* Nothing to propagate */
+	if (!cfs_rq->propagate_avg)
+		return 0;
+
+	/* Clear my cfs_rq's flag */
+	cfs_rq->propagate_avg = 0;
+
+	return 1;
+}
+
+/* Update task and its cfs_rq load average */
+static inline int propagate_entity_load_avg(struct sched_entity *se)
+{
+	struct cfs_rq *cfs_rq;
+
+	if (entity_is_task(se))
+		return 0;
+
+	if (!test_and_clear_tg_cfs_propagate(se))
+		return 0;
+
+	/* Get parent cfs_rq */
+	cfs_rq = cfs_rq_of(se);
+
+	/* Propagate to parent */
+	set_tg_cfs_propagate(cfs_rq);
+
+	/* Update utilization */
+	update_tg_cfs_util(cfs_rq, se);
+
+	/* Update load */
+	update_tg_cfs_load(cfs_rq, se);
+
+	return 1;
+}
+#else
+static inline int propagate_entity_load_avg(struct sched_entity *se)
+{
+	return 0;
+}
+
+static inline void set_tg_cfs_propagate(struct cfs_rq *cfs_rq) {}
+#endif
+
+/*
  * Unsigned subtract and clamp on underflow.
  *
  * Explicitly do a load-store to ensure the intermediate value never hits
@@ -3111,6 +3267,7 @@ static inline void update_load_avg(struct sched_entity *se, int flags)
 	u64 now = cfs_rq_clock_task(cfs_rq);
 	struct rq *rq = rq_of(cfs_rq);
 	int cpu = cpu_of(rq);
+	int decayed;
 
 	/*
 	 * Track task load average for carrying it to new CPU after migrated, and
@@ -3121,7 +3278,11 @@ static inline void update_load_avg(struct sched_entity *se, int flags)
 			  se->on_rq * scale_load_down(se->load.weight),
 			  cfs_rq->curr == se, NULL);
 
-	if (update_cfs_rq_load_avg(now, cfs_rq, true) && (flags & UPDATE_TG))
+	decayed = update_cfs_rq_load_avg(now, cfs_rq, true);
+
+	decayed |= propagate_entity_load_avg(se);
+
+	if (decayed && (flags & UPDATE_TG))
 		update_tg_load_avg(cfs_rq, 0);
 }
 
@@ -3140,6 +3301,7 @@ static void attach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *s
 	cfs_rq->avg.load_sum += se->avg.load_sum;
 	cfs_rq->avg.util_avg += se->avg.util_avg;
 	cfs_rq->avg.util_sum += se->avg.util_sum;
+	set_tg_cfs_propagate(cfs_rq);
 
 	cfs_rq_util_change(cfs_rq);
 }
@@ -3159,6 +3321,7 @@ static void detach_entity_load_avg(struct cfs_rq *cfs_rq, struct sched_entity *s
 	sub_positive(&cfs_rq->avg.load_sum, se->avg.load_sum);
 	sub_positive(&cfs_rq->avg.util_avg, se->avg.util_avg);
 	sub_positive(&cfs_rq->avg.util_sum, se->avg.util_sum);
+	set_tg_cfs_propagate(cfs_rq);
 
 	cfs_rq_util_change(cfs_rq);
 }
@@ -8704,6 +8867,22 @@ static void detach_task_cfs_rq(struct task_struct *p)
 	update_load_avg(se, 0);
 	detach_entity_load_avg(cfs_rq, se);
 	update_tg_load_avg(cfs_rq, false);
+
+#ifdef CONFIG_FAIR_GROUP_SCHED
+	/*
+	 * Propagate the detach across the tg tree to make it visible to the
+	 * root
+	 */
+	se = se->parent;
+	for_each_sched_entity(se) {
+		cfs_rq = cfs_rq_of(se);
+
+		if (cfs_rq_throttled(cfs_rq))
+			break;
+
+		update_load_avg(se, UPDATE_TG);
+	}
+#endif
 }
 
 static void attach_entity_cfs_rq(struct sched_entity *se)
@@ -8722,6 +8901,22 @@ static void attach_entity_cfs_rq(struct sched_entity *se)
 	update_load_avg(se, sched_feat(ATTACH_AGE_LOAD) ? 0 : SKIP_AGE_LOAD);
 	attach_entity_load_avg(cfs_rq, se);
 	update_tg_load_avg(cfs_rq, false);
+
+#ifdef CONFIG_FAIR_GROUP_SCHED
+	/*
+	 * Propagate the attach across the tg tree to make it visible to the
+	 * root
+	 */
+	se = se->parent;
+	for_each_sched_entity(se) {
+		cfs_rq = cfs_rq_of(se);
+
+		if (cfs_rq_throttled(cfs_rq))
+			break;
+
+		update_load_avg(se, UPDATE_TG);
+	}
+#endif
 }
 
 static void attach_task_cfs_rq(struct task_struct *p)
@@ -8783,6 +8978,9 @@ void init_cfs_rq(struct cfs_rq *cfs_rq)
 	cfs_rq->min_vruntime_copy = cfs_rq->min_vruntime;
 #endif
 #ifdef CONFIG_SMP
+#ifdef CONFIG_FAIR_GROUP_SCHED
+	cfs_rq->propagate_avg = 0;
+#endif
 	atomic_long_set(&cfs_rq->removed_load_avg, 0);
 	atomic_long_set(&cfs_rq->removed_util_avg, 0);
 #endif
diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
index f1185c3..47385e2 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -404,6 +404,7 @@ struct cfs_rq {
 	unsigned long runnable_load_avg;
 #ifdef CONFIG_FAIR_GROUP_SCHED
 	unsigned long tg_load_avg_contrib;
+	unsigned long propagate_avg;
 #endif
 	atomic_long_t removed_load_avg, removed_util_avg;
 #ifndef CONFIG_64BIT
-- 
2.7.4

Re: [PATCH 4/6 v5] sched: propagate load during synchronous attach/detach

[Dietmar Eggemann]

On 10/17/2016 10:14 AM, Vincent Guittot wrote:
> When a task moves from/to a cfs_rq, we set a flag which is then used to
> propagate the change at parent level (sched_entity and cfs_rq) during
> next update. If the cfs_rq is throttled, the flag will stay pending until
> the cfs_rw is unthrottled.

minor nit:

s/cfs_rw/cfs_rq

[...]

> @@ -8704,6 +8867,22 @@ static void detach_task_cfs_rq(struct task_struct *p)
>  	update_load_avg(se, 0);
>  	detach_entity_load_avg(cfs_rq, se);
>  	update_tg_load_avg(cfs_rq, false);
> +
> +#ifdef CONFIG_FAIR_GROUP_SCHED
> +	/*
> +	 * Propagate the detach across the tg tree to make it visible to the
> +	 * root
> +	 */
> +	se = se->parent;
> +	for_each_sched_entity(se) {
> +		cfs_rq = cfs_rq_of(se);
> +
> +		if (cfs_rq_throttled(cfs_rq))
> +			break;
> +
> +		update_load_avg(se, UPDATE_TG);
> +	}
> +#endif
>  }
>
>  static void attach_entity_cfs_rq(struct sched_entity *se)
> @@ -8722,6 +8901,22 @@ static void attach_entity_cfs_rq(struct sched_entity *se)
>  	update_load_avg(se, sched_feat(ATTACH_AGE_LOAD) ? 0 : SKIP_AGE_LOAD);
>  	attach_entity_load_avg(cfs_rq, se);
>  	update_tg_load_avg(cfs_rq, false);
> +
> +#ifdef CONFIG_FAIR_GROUP_SCHED
> +	/*
> +	 * Propagate the attach across the tg tree to make it visible to the
> +	 * root
> +	 */
> +	se = se->parent;
> +	for_each_sched_entity(se) {
> +		cfs_rq = cfs_rq_of(se);
> +
> +		if (cfs_rq_throttled(cfs_rq))
> +			break;
> +
> +		update_load_avg(se, UPDATE_TG);
> +	}
> +#endif
>  }

The 'detach across' and 'attach across' in detach_task_cfs_rq() and 
attach_entity_cfs_rq() do the same so couldn't you not create a function 
propagate_foo() for it? This would avoid this ifdef as well.

You could further create in your '[PATCH 1/6 v5] sched: factorize attach 
entity':

detach_entity_cfs_rq() {
   update_load_avg()
   detach_entity_load_avg()
   update_tg_load_avg()
   propagate_load_avg()
}

and then we would have:

attach_task_cfs_rq() -> attach_entity_cfs_rq() -> propagate_foo()
detach_task_cfs_rq() -> detach_entity_cfs_rq() -> propagate_foo()

I guess you didn't because it would be only called one time but this 
symmetric approaches are easier to remember (at least for me).

[...]

Re: [PATCH 4/6 v5] sched: propagate load during synchronous attach/detach

[Vincent Guittot]

On 21 October 2016 at 14:19, Dietmar Eggemann <dietmar.eggemann@arm.com> wrote:
>
> On 10/17/2016 10:14 AM, Vincent Guittot wrote:
>>
>> When a task moves from/to a cfs_rq, we set a flag which is then used to
>> propagate the change at parent level (sched_entity and cfs_rq) during
>> next update. If the cfs_rq is throttled, the flag will stay pending until
>> the cfs_rw is unthrottled.
>
>
> minor nit:
>
> s/cfs_rw/cfs_rq

yes

>
> [...]
>
>
>> @@ -8704,6 +8867,22 @@ static void detach_task_cfs_rq(struct task_struct *p)
>>         update_load_avg(se, 0);
>>         detach_entity_load_avg(cfs_rq, se);
>>         update_tg_load_avg(cfs_rq, false);
>> +
>> +#ifdef CONFIG_FAIR_GROUP_SCHED
>> +       /*
>> +        * Propagate the detach across the tg tree to make it visible to the
>> +        * root
>> +        */
>> +       se = se->parent;
>> +       for_each_sched_entity(se) {
>> +               cfs_rq = cfs_rq_of(se);
>> +
>> +               if (cfs_rq_throttled(cfs_rq))
>> +                       break;
>> +
>> +               update_load_avg(se, UPDATE_TG);
>> +       }
>> +#endif
>>  }
>>
>>  static void attach_entity_cfs_rq(struct sched_entity *se)
>> @@ -8722,6 +8901,22 @@ static void attach_entity_cfs_rq(struct sched_entity *se)
>>         update_load_avg(se, sched_feat(ATTACH_AGE_LOAD) ? 0 : SKIP_AGE_LOAD);
>>         attach_entity_load_avg(cfs_rq, se);
>>         update_tg_load_avg(cfs_rq, false);
>> +
>> +#ifdef CONFIG_FAIR_GROUP_SCHED
>> +       /*
>> +        * Propagate the attach across the tg tree to make it visible to the
>> +        * root
>> +        */
>> +       se = se->parent;
>> +       for_each_sched_entity(se) {
>> +               cfs_rq = cfs_rq_of(se);
>> +
>> +               if (cfs_rq_throttled(cfs_rq))
>> +                       break;
>> +
>> +               update_load_avg(se, UPDATE_TG);
>> +       }
>> +#endif
>>  }
>
>
> The 'detach across' and 'attach across' in detach_task_cfs_rq() and attach_entity_cfs_rq() do the same so couldn't you not create a function propagate_foo() for it? This would avoid this ifdef as well.
>
> You could further create in your '[PATCH 1/6 v5] sched: factorize attach entity':
>
> detach_entity_cfs_rq() {
>   update_load_avg()
>   detach_entity_load_avg()
>   update_tg_load_avg()
>   propagate_load_avg()
> }
>
> and then we would have:
>
> attach_task_cfs_rq() -> attach_entity_cfs_rq() -> propagate_foo()
> detach_task_cfs_rq() -> detach_entity_cfs_rq() -> propagate_foo()
>
> I guess you didn't because it would be only called one time but this symmetric approaches are easier to remember (at least for me).

Yes i haven't created attach_entity_cfs_rq because it would be used only once.
Regarding the creation of a propagate_foo function, i have just
followed a similar skeleton as what is done in
enqueue/dequeue_task_fair

I don't have strong opinion about creating this indirection for code
readability. Others, have you got a preference ?

>
> [...]

Re: [PATCH 4/6 v5] sched: propagate load during synchronous attach/detach

[Peter Zijlstra]

On Wed, Oct 26, 2016 at 09:05:49AM +0200, Vincent Guittot wrote:
> >
> > The 'detach across' and 'attach across' in detach_task_cfs_rq() and
> > attach_entity_cfs_rq() do the same so couldn't you not create a
> > function propagate_foo() for it? This would avoid this ifdef as
> > well.
> >
> > You could further create in your '[PATCH 1/6 v5] sched: factorize attach entity':
> >
> > detach_entity_cfs_rq() {
> >   update_load_avg()
> >   detach_entity_load_avg()
> >   update_tg_load_avg()
> >   propagate_load_avg()
> > }
> >
> > and then we would have:
> >
> > attach_task_cfs_rq() -> attach_entity_cfs_rq() -> propagate_foo()
> > detach_task_cfs_rq() -> detach_entity_cfs_rq() -> propagate_foo()
> >
> > I guess you didn't because it would be only called one time but this
> > symmetric approaches are easier to remember (at least for me).
> 
> Yes i haven't created attach_entity_cfs_rq because it would be used only once.
> Regarding the creation of a propagate_foo function, i have just
> followed a similar skeleton as what is done in
> enqueue/dequeue_task_fair
> 
> I don't have strong opinion about creating this indirection for code
> readability. Others, have you got a preference ?

I think I agree with Dietmar. Duplicate code needs a helper function and
it would be nice to keep symmetry, even if there's only a single
call site.

Re: [PATCH 4/6 v5] sched: propagate load during synchronous attach/detach

[Vincent Guittot]

On 26 October 2016 at 13:16, Peter Zijlstra <peterz@infradead.org> wrote:
> On Wed, Oct 26, 2016 at 09:05:49AM +0200, Vincent Guittot wrote:
>> >
>> > The 'detach across' and 'attach across' in detach_task_cfs_rq() and
>> > attach_entity_cfs_rq() do the same so couldn't you not create a
>> > function propagate_foo() for it? This would avoid this ifdef as
>> > well.
>> >
>> > You could further create in your '[PATCH 1/6 v5] sched: factorize attach entity':
>> >
>> > detach_entity_cfs_rq() {
>> >   update_load_avg()
>> >   detach_entity_load_avg()
>> >   update_tg_load_avg()
>> >   propagate_load_avg()
>> > }
>> >
>> > and then we would have:
>> >
>> > attach_task_cfs_rq() -> attach_entity_cfs_rq() -> propagate_foo()
>> > detach_task_cfs_rq() -> detach_entity_cfs_rq() -> propagate_foo()
>> >
>> > I guess you didn't because it would be only called one time but this
>> > symmetric approaches are easier to remember (at least for me).
>>
>> Yes i haven't created attach_entity_cfs_rq because it would be used only once.
>> Regarding the creation of a propagate_foo function, i have just
>> followed a similar skeleton as what is done in
>> enqueue/dequeue_task_fair
>>
>> I don't have strong opinion about creating this indirection for code
>> readability. Others, have you got a preference ?
>
> I think I agree with Dietmar. Duplicate code needs a helper function and
> it would be nice to keep symmetry, even if there's only a single
> call site.

OK

Re: [PATCH 4/6 v5] sched: propagate load during synchronous attach/detach

[Peter Zijlstra]

On Mon, Oct 17, 2016 at 11:14:11AM +0200, Vincent Guittot wrote:
>  /*
> + * Signed add and clamp on underflow.
> + *
> + * Explicitly do a load-store to ensure the intermediate value never hits
> + * memory. This allows lockless observations without ever seeing the negative
> + * values.
> + */
> +#define add_positive(_ptr, _val) do {				\
> +	typeof(_ptr) ptr = (_ptr);				\
> +	typeof(_val) res, val = (_val);				\
> +	typeof(*ptr) var = READ_ONCE(*ptr);			\
> +	res = var + val;					\
> +	if (res < 0)						\
> +		res = 0;					\

I think this is broken, and inconsistent with sub_positive().

The thing is, util_avg, on which you use this, is an unsigned type.
Checking for unsigned underflow can be done by comparing against either
one of the terms.

> +	WRITE_ONCE(*ptr, res);					\
> +} while (0)

> +	add_positive(&cfs_rq->avg.util_avg, delta);

Re: [PATCH 4/6 v5] sched: propagate load during synchronous attach/detach

[Vincent Guittot]

On 26 October 2016 at 12:54, Peter Zijlstra <peterz@infradead.org> wrote:
> On Mon, Oct 17, 2016 at 11:14:11AM +0200, Vincent Guittot wrote:
>>  /*
>> + * Signed add and clamp on underflow.
>> + *
>> + * Explicitly do a load-store to ensure the intermediate value never hits
>> + * memory. This allows lockless observations without ever seeing the negative
>> + * values.
>> + */
>> +#define add_positive(_ptr, _val) do {                                \
>> +     typeof(_ptr) ptr = (_ptr);                              \
>> +     typeof(_val) res, val = (_val);                         \
>> +     typeof(*ptr) var = READ_ONCE(*ptr);                     \
>> +     res = var + val;                                        \
>> +     if (res < 0)                                            \
>> +             res = 0;                                        \
>
> I think this is broken, and inconsistent with sub_positive().

I agree that the behavior is different from sub_positive which deals
with unsigned value, but i was not able to come with a short name that
highlight this signed/unsigned difference

>
> The thing is, util_avg, on which you use this, is an unsigned type.

The delta that is added to util_avg, is a signed value

> Checking for unsigned underflow can be done by comparing against either
> one of the terms.
>
>> +     WRITE_ONCE(*ptr, res);                                  \
>> +} while (0)
>
>> +     add_positive(&cfs_rq->avg.util_avg, delta);

Re: [PATCH 4/6 v5] sched: propagate load during synchronous attach/detach

[Peter Zijlstra]

On Wed, Oct 26, 2016 at 02:31:01PM +0200, Vincent Guittot wrote:
> On 26 October 2016 at 12:54, Peter Zijlstra <peterz@infradead.org> wrote:
> > On Mon, Oct 17, 2016 at 11:14:11AM +0200, Vincent Guittot wrote:
> >>  /*
> >> + * Signed add and clamp on underflow.
> >> + *
> >> + * Explicitly do a load-store to ensure the intermediate value never hits
> >> + * memory. This allows lockless observations without ever seeing the negative
> >> + * values.
> >> + */
> >> +#define add_positive(_ptr, _val) do {                                \
> >> +     typeof(_ptr) ptr = (_ptr);                              \
> >> +     typeof(_val) res, val = (_val);                         \
> >> +     typeof(*ptr) var = READ_ONCE(*ptr);                     \
> >> +     res = var + val;                                        \
> >> +     if (res < 0)                                            \
> >> +             res = 0;                                        \
> >
> > I think this is broken, and inconsistent with sub_positive().
> 
> I agree that the behavior is different from sub_positive which deals
> with unsigned value, but i was not able to come with a short name that
> highlight this signed/unsigned difference
> 
> >
> > The thing is, util_avg, on which you use this, is an unsigned type.
> 
> The delta that is added to util_avg, is a signed value

Doesn't matter, util_avg is unsigned, this means MSB set is a valid and
non-negative number, while the above will truncate it to 0.

So you really do need an alternative method of underflow. And yes, delta
being signed makes it slightly more complicated.

How about something like the below, that will, if val is negative and we
thus end up doing a subtraction (assumes 2s complement, which is fine,
we do all over anyway), check the result isn't larger than we started
out with.


#define add_positive(_ptr, _val) do {                           \
        typeof(_ptr) ptr = (_ptr);                              \
        typeof(_val) val = (_val);                              \
        typeof(*ptr) res, var = READ_ONCE(*ptr);                \
								\
	res = var + val;					\
								\
	if (val < 0 && res > var)				\
		res = 0;					\
								\
        WRITE_ONCE(*ptr, res);                                  \
} while (0)

Re: [PATCH 4/6 v5] sched: propagate load during synchronous attach/detach

[Vincent Guittot]

On 26 October 2016 at 16:28, Peter Zijlstra <peterz@infradead.org> wrote:
> On Wed, Oct 26, 2016 at 02:31:01PM +0200, Vincent Guittot wrote:
>> On 26 October 2016 at 12:54, Peter Zijlstra <peterz@infradead.org> wrote:
>> > On Mon, Oct 17, 2016 at 11:14:11AM +0200, Vincent Guittot wrote:
>> >>  /*
>> >> + * Signed add and clamp on underflow.
>> >> + *
>> >> + * Explicitly do a load-store to ensure the intermediate value never hits
>> >> + * memory. This allows lockless observations without ever seeing the negative
>> >> + * values.
>> >> + */
>> >> +#define add_positive(_ptr, _val) do {                                \
>> >> +     typeof(_ptr) ptr = (_ptr);                              \
>> >> +     typeof(_val) res, val = (_val);                         \
>> >> +     typeof(*ptr) var = READ_ONCE(*ptr);                     \
>> >> +     res = var + val;                                        \
>> >> +     if (res < 0)                                            \
>> >> +             res = 0;                                        \
>> >
>> > I think this is broken, and inconsistent with sub_positive().
>>
>> I agree that the behavior is different from sub_positive which deals
>> with unsigned value, but i was not able to come with a short name that
>> highlight this signed/unsigned difference
>>
>> >
>> > The thing is, util_avg, on which you use this, is an unsigned type.
>>
>> The delta that is added to util_avg, is a signed value
>
> Doesn't matter, util_avg is unsigned, this means MSB set is a valid and
> non-negative number, while the above will truncate it to 0.
>
> So you really do need an alternative method of underflow. And yes, delta
> being signed makes it slightly more complicated.
>
> How about something like the below, that will, if val is negative and we
> thus end up doing a subtraction (assumes 2s complement, which is fine,
> we do all over anyway), check the result isn't larger than we started
> out with.
>
>
> #define add_positive(_ptr, _val) do {                           \
>         typeof(_ptr) ptr = (_ptr);                              \
>         typeof(_val) val = (_val);                              \
>         typeof(*ptr) res, var = READ_ONCE(*ptr);                \
>                                                                 \
>         res = var + val;                                        \
>                                                                 \
>         if (val < 0 && res > var)                               \
>                 res = 0;                                        \
>                                                                 \
>         WRITE_ONCE(*ptr, res);                                  \
> } while (0)

Indeed, looks better like that

>
>

Re: [PATCH 4/6 v5] sched: propagate load during synchronous attach/detach

[Peter Zijlstra]

On Mon, Oct 17, 2016 at 11:14:11AM +0200, Vincent Guittot wrote:
> +/* Take into account change of load of a child task group */
> +static inline void
> +update_tg_cfs_load(struct cfs_rq *cfs_rq, struct sched_entity *se)
> +{
> +	struct cfs_rq *gcfs_rq = group_cfs_rq(se);
> +	long delta, load = gcfs_rq->avg.load_avg;
> +
> +	/*
> +	 * If the load of group cfs_rq is null, the load of the
> +	 * sched_entity will also be null so we can skip the formula
> +	 */

Does it make sense to do:

	if (!load)
		goto no_load;

and avoid the indent?

> +	if (load) {
> +		long tg_load;
> +
> +		/* Get tg's load and ensure tg_load > 0 */
> +		tg_load = atomic_long_read(&gcfs_rq->tg->load_avg) + 1;
> +
> +		/* Ensure tg_load >= load and updated with current load*/
> +		tg_load -= gcfs_rq->tg_load_avg_contrib;
> +		tg_load += load;
> +
> +		/*
> +		 * We need to compute a correction term in the case that the
> +		 * task group is consuming more cpu than a task of equal
> +		 * weight. A task with a weight equals to tg->shares will have
> +		 * a load less or equal to scale_load_down(tg->shares).
  +		 *
> +		 * Similarly, the sched_entities that represent the task group
> +		 * at parent level, can't have a load higher than
> +		 * scale_load_down(tg->shares). And the Sum of sched_entities'
> +		 * load must be <= scale_load_down(tg->shares).
> +		 */
> +		if (tg_load > scale_load_down(gcfs_rq->tg->shares)) {
> +			/* scale gcfs_rq's load into tg's shares*/
> +			load *= scale_load_down(gcfs_rq->tg->shares);
> +			load /= tg_load;
> +		}
> +	}
> +
no_load:
> +	delta = load - se->avg.load_avg;
> +
> +	/* Nothing to update */
> +	if (!delta)
> +		return;
> +
> +	/* Set new sched_entity's load */
> +	se->avg.load_avg = load;
> +	se->avg.load_sum = se->avg.load_avg * LOAD_AVG_MAX;
> +
> +	/* Update parent cfs_rq load */
> +	add_positive(&cfs_rq->avg.load_avg, delta);
> +	cfs_rq->avg.load_sum = cfs_rq->avg.load_avg * LOAD_AVG_MAX;
> +
> +	/*
> +	 * If the sched_entity is already enqueued, we also have to update the
> +	 * runnable load avg.
> +	 */
> +	if (se->on_rq) {
> +		/* Update parent cfs_rq runnable_load_avg */
> +		add_positive(&cfs_rq->runnable_load_avg, delta);
> +		cfs_rq->runnable_load_sum = cfs_rq->runnable_load_avg * LOAD_AVG_MAX;
> +	}
> +}

[PATCH 5/6 v5] sched: propagate asynchrous detach

[Vincent Guittot]

A task can be asynchronously detached from cfs_rq when migrating
between CPUs. The load of the migrated task is then removed from
source cfs_rq during its next update. We use this event to set propagation
flag.

During the load balance, we take advanatge of the update of blocked load
to propagate any pending changes.The propagation relies on patch
"sched: fix hierarchical order in rq->leaf_cfs_rq_list", which orders
children and parents, to ensure that it's done in one pass.

Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
---
 kernel/sched/fair.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 91fc949..4c0a404 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -3231,6 +3231,7 @@ update_cfs_rq_load_avg(u64 now, struct cfs_rq *cfs_rq, bool update_freq)
 		sub_positive(&sa->load_avg, r);
 		sub_positive(&sa->load_sum, r * LOAD_AVG_MAX);
 		removed_load = 1;
+		set_tg_cfs_propagate(cfs_rq);
 	}
 
 	if (atomic_long_read(&cfs_rq->removed_util_avg)) {
@@ -3238,6 +3239,7 @@ update_cfs_rq_load_avg(u64 now, struct cfs_rq *cfs_rq, bool update_freq)
 		sub_positive(&sa->util_avg, r);
 		sub_positive(&sa->util_sum, r * LOAD_AVG_MAX);
 		removed_util = 1;
+		set_tg_cfs_propagate(cfs_rq);
 	}
 
 	decayed = __update_load_avg(now, cpu_of(rq_of(cfs_rq)), sa,
@@ -6803,6 +6805,10 @@ static void update_blocked_averages(int cpu)
 
 		if (update_cfs_rq_load_avg(cfs_rq_clock_task(cfs_rq), cfs_rq, true))
 			update_tg_load_avg(cfs_rq, 0);
+
+		/* Propagate pending load changes to the parent */
+		if (cfs_rq->tg->se[cpu])
+			update_load_avg(cfs_rq->tg->se[cpu], 0);
 	}
 	raw_spin_unlock_irqrestore(&rq->lock, flags);
 }
-- 
2.7.4

[PATCH 6/6 v5] sched: fix task group initialization

[Vincent Guittot]

The moves of tasks are now propagated down to root and the utilization
of cfs_rq reflects reality so it doesn't need to be estimated at init.

Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
---
 kernel/sched/fair.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 4c0a404..d0cf597 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -9098,7 +9098,7 @@ void online_fair_sched_group(struct task_group *tg)
 		se = tg->se[i];
 
 		raw_spin_lock_irq(&rq->lock);
-		post_init_entity_util_avg(se);
+		attach_entity_cfs_rq(se);
 		sync_throttle(tg, i);
 		raw_spin_unlock_irq(&rq->lock);
 	}
-- 
2.7.4