Re: [PATCH] sched: update the rq->avg_idle when a task is moved to an idle CPU

[Madadi Vineeth Reddy <vineethr@linux.ibm.com>]

On 25/11/25 14:29, Shijie Huang wrote:
> 
> On 25/11/2025 15:08, Madadi Vineeth Reddy wrote:
>> I traced the activate_task() call paths and found that load balancing migrations
>> through attach_task() in kernel/sched/fair.c may not be covered.
> thanks for pointing this, I did not notice them.
>>
>> activate_task <- attach_task <- attach_tasks <- sched_balance_rq
>> activate_task <- attach_task <- attach_one_task <- active_load_balance_cpu_stop
>>
>> These paths are called in periodic load balancing and when tasks are pulled
>> towards an idle CPU via attach_task(), it doesn't update rq->avg_idle or clear
>> idle_stamp.
> 
> Yes, we should update the rq->avg_idle for them.
> 
> I will add it in version 2.
> 
>>
>> Should attach_task() in kernel/sched/fair.c also call update_rq_avg_idle()
>> after activation?
> 
> IMHO,  we should not call the update_rq_avg_idle() directly in attach_task().
> 
> In the current attach_task(), there is no information for the context(newidle, idle, busy).
> 
> The attach_task() is also called in the newidle code path.
> 
> But we can call the update_rq_avg_idle() in attach_tasks() with a condition check.
> 

IIUC, update_rq_avg_idle() already checks if (rq->idle_stamp) internally and
in attach_task() we have rq available and the guard in update_rq_avg_idle()
ensures we only update when the CPU was actually idle. Whether it's called
during newidle, idle, or busy balancing shouldn't matter.

Let me know if I am missing something.

Thanks,
Vineeth

>>
>> Also, can update_rq_avg_idle() be placed inside activate_task() to avoid
>> all these?
> 
> The same reason as above.
> 
> 
> Thanks
> 
> Huang Shijie
>