[PATCH] sched: don't check rq after newidle_balance return positive

[PATCH] sched: don't check rq after newidle_balance return positive

[chenxg1x]

From: Chen Xiaoguang <xiaoggchen@tencent.com>

In pick_next_task_fair, if CPU is going to idle newidle_balance
is called first trying to pull some tasks.
When newidle_balance returns positive which means it does
pulls tasks or some tasks enqueued then there is no need to check
sched_fair_runnable again.

Signed-off-by: He Chen <heddchen@tencent.com>
Signed-off-by: Xiaoguang Chen <xiaoggchen@tencent.com>
---
 kernel/sched/fair.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index ae7ceba..c2f7eac 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -7004,10 +7004,9 @@ struct task_struct *
 	struct task_struct *p;
 	int new_tasks;
 
-again:
 	if (!sched_fair_runnable(rq))
 		goto idle;
-
+again:
 #ifdef CONFIG_FAIR_GROUP_SCHED
 	if (!prev || prev->sched_class != &fair_sched_class)
 		goto simple;
-- 
1.8.3.1

Re: [PATCH] sched: don't check rq after newidle_balance return positive

[Peter Zijlstra]

On Tue, Dec 15, 2020 at 02:48:50PM +0800, chenxg1x@gmail.com wrote:
> From: Chen Xiaoguang <xiaoggchen@tencent.com>
> 
> In pick_next_task_fair, if CPU is going to idle newidle_balance
> is called first trying to pull some tasks.
> When newidle_balance returns positive which means it does
> pulls tasks or some tasks enqueued then there is no need to check
> sched_fair_runnable again.

No, I think it actually does need to, because while it counts the number
of tasks it pulled, it doesn't verify it still has them after it
re-acquires rq->lock. That is, someone could've come along and stolen
them right from under our noses.

> 
> Signed-off-by: He Chen <heddchen@tencent.com>
> Signed-off-by: Xiaoguang Chen <xiaoggchen@tencent.com>

This SoB chain is broken. The first SoB should be the author, but From
does not match.

答复: [PATCH] sched: don't check rq after newidle_balance return positive(Internet mail)

[xiaoggchen(陈小光)]

-----邮件原件-----
发件人: Peter Zijlstra <peterz@infradead.org> 
发送时间: 2020年12月15日 16:33
收件人: chenxg1x@gmail.com
抄送: linux-kernel@vger.kernel.org; mingo@redhat.com; juri.lelli@redhat.com; vincent.guittot@linaro.org; dietmar.eggemann@arm.com; rostedt@goodmis.org; bsegall@google.com; mgorman@suse.de; bristot@redhat.com; heddchen(陈贺) <heddchen@tencent.com>; xiaoggchen(陈小光) <xiaoggchen@tencent.com>
主题: Re: [PATCH] sched: don't check rq after newidle_balance return positive(Internet mail)

On Tue, Dec 15, 2020 at 02:48:50PM +0800, chenxg1x@gmail.com wrote:
>> From: Chen Xiaoguang <xiaoggchen@tencent.com>
>> 
>> In pick_next_task_fair, if CPU is going to idle newidle_balance is 
>> called first trying to pull some tasks.
>> When newidle_balance returns positive which means it does pulls tasks 
>> or some tasks enqueued then there is no need to check 
>> sched_fair_runnable again.

> No, I think it actually does need to, because while it counts the number of tasks it pulled, it doesn't verify it still has them after it re-acquires rq->lock. That is, someone could've come along and stolen them 
> right from under our noses.
Ah, yes, our change only make sense when pulling nothing in load_balance but some tasks enqueued this rq during the lock of this rq is released.
 
Thanks.

>> 
>> Signed-off-by: He Chen <heddchen@tencent.com>
>> Signed-off-by: Xiaoguang Chen <xiaoggchen@tencent.com>

> This SoB chain is broken. The first SoB should be the author, but From does not match.
We will fix this next time.