Skip to contentSkip to conversation

sched: fix SCHED_OTHER balance iterator to include all tasks

PATCH In review Gregory Haskins 3 participants 3 messages May 5, 2008 – May 8, 2008

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 sched: fix SCHED_OTHER balance iterator to include all tasks

Hi Ingo,
The follow patch applies to sched-devel. I know the original function was somehow trying to protect against dequeueing by pre-iterating. I am not sure if I captured the intent properly in that regard. What I can say is that the old algorithm would miss tasks, and that is now fixed. It may need further adjustment if I missed the pre-iterating operation.

Hi Peter,
I was wrong earlier when I mentioned there might be a problem with entity_is_task(). I think this was the root cause of what I was seeing. Sorry for the red-herring.

Regards,
-Greg

-----------------------
sched: fix SCHED_OTHER balance iterator to include all tasks

The currently logic inadvertently skips the last task on the run-queue, resulting in missed balance opportunities.

---

 kernel/sched_fair.c |   19 +++++++------------
 1 files changed, 7 insertions(+), 12 deletions(-)
kernel/sched_fair.c +7 -12
index 87b7300..e8c289a 100644
@@ -1700,23 +1700,18 @@ __load_balance_iterator(struct cfs_rq *cfs_rq, struct list_head *next)
 	struct task_struct *p = NULL;
 	struct sched_entity *se;
 
-	if (next == &cfs_rq->tasks)
-		return NULL;
-
-	/* Skip over entities that are not tasks */
-	do {
+	while (next != &cfs_rq->tasks) {
 		se = list_entry(next, struct sched_entity, group_node);
 		next = next->next;
-	} while (next != &cfs_rq->tasks && !entity_is_task(se));
 
-	if (next == &cfs_rq->tasks)
-		return NULL;
+		/* Skip over entities that are not tasks */
+		if (entity_is_task(se)) {
+			p = task_of(se);
+			break;
+		}
+	}
 
 	cfs_rq->balance_iterator = next;
-
-	if (entity_is_task(se))
-		p = task_of(se);
-
 	return p;
 }
 
#1 lore raw
Quoted 8 lines · Gregory Haskins

Hi Ingo,
The follow patch applies to sched-devel. I know the original function was somehow trying to protect against dequeueing by pre-iterating. I am not sure if I captured the intent properly in that regard. What I can say is that the old algorithm would miss tasks, and that is now fixed. It may need further adjustment if I missed the pre-iterating operation.

Quoted 4 lines

Hi Peter,
I was wrong earlier when I mentioned there might be a problem with entity_is_task(). I think this was the root cause of what I was seeing. Sorry for the red-herring.

applied Gregory, thanks. Nice fix! Peter, do you concur with Gregory's analysis?

Ingo

#2 lore raw
Quoted 10 lines · Ingo Molnar
Quoted 8 lines · Gregory Haskins

Hi Ingo,
The follow patch applies to sched-devel. I know the original function was somehow trying to protect against dequeueing by pre-iterating. I am not sure if I captured the intent properly in that regard. What I can say is that the old algorithm would miss tasks, and that is now fixed. It may need further adjustment if I missed the pre-iterating operation.

Quoted 7 lines
Quoted 4 lines

Hi Peter,
I was wrong earlier when I mentioned there might be a problem with entity_is_task(). I think this was the root cause of what I was seeing. Sorry for the red-herring.

applied Gregory, thanks. Nice fix! Peter, do you concur with Gregory's analysis?

Yeah,