Re: Re: Re: [PATCH] sched/fair: remove next_buddy_marked

[Abel Wu <wuyun.abel@bytedance.com>]

On 12/14/23 9:02 PM, Wang Jinchao Wrote:
> On Thu, Dec 14, 2023 at 08:21:53PM +0800, Abel Wu wrote:
>> On 12/14/23 4:18 PM, Vincent Guittot Wrote:
>>> On Thu, 14 Dec 2023 at 06:20, Wang Jinchao <wangjinchao@xfusion.com> wrote:
>>>>
>>>> Remove unused `next_buddy_marked` in `check_preempt_wakeup_fair`
>>>>
>>>
>>> Fixes: 5e963f2bd465 ("sched/fair: Commit to EEVDF")
>>
>> After this commit @pse preempts curr without being the NEXT_BUDDY, but
>> IMHO it should be, so how about this?
>>
>> @@ -8259,8 +8259,11 @@ static void check_preempt_wakeup_fair(struct rq *rq, struct task_struct *p, int
>>          /*
>>           * XXX pick_eevdf(cfs_rq) != se ?
>>           */
>> -       if (pick_eevdf(cfs_rq) == pse)
>> +       if (pick_eevdf(cfs_rq) == pse) {
>> +               if (!next_buddy_marked)
>> +                       set_next_buddy(pse);
>>                  goto preempt;
>> +       }
>>
>>          return;
>>
>> which will align with before.
> Seizing this opportunity to inquire about a question:
> What does "buddy" mean in the context of the scheduler?

struct sched_entity

> 
> Is the effect the same between
>      preempting after pick_evfd(cfs_rq) == pse
> and
>      preempting after set_next_buddy(pse) followed by pick_evfd(cfs_rq) == pse?
> Would both scenarios result in pse becoming the next scheduled se?"

Probably, since pse is the one preempts curr, pick_next_entity() could
return pse directly without walking the rbtree. So the difference is in
performance.