Re: [RFC PATCH 3/4] sched/fair: delay update_tg_load_avg() for cfs_rq's removed load

[Dietmar Eggemann <dietmar.eggemann@arm.com>]

On 20/07/2023 17:02, Vincent Guittot wrote:
> On Thu, 20 Jul 2023 at 16:42, Aaron Lu <aaron.lu@intel.com> wrote:
>>
>> On Thu, Jul 20, 2023 at 03:10:30PM +0200, Vincent Guittot wrote:
>>> On Wed, 19 Jul 2023 at 15:29, Aaron Lu <aaron.lu@intel.com> wrote:
>>>>
>>>> On Wed, Jul 19, 2023 at 11:47:06AM +0200, Vincent Guittot wrote:
>>>>> On Wed, 19 Jul 2023 at 10:01, Aaron Lu <aaron.lu@intel.com> wrote:
>>>>>>
>>>>>> On Wed, Jul 19, 2023 at 01:18:26PM +0800, Aaron Lu wrote:

[...]

> What was wrong with your proposal to limit the update inside
> update_tg_load_avg()  ? except maybe s/1000000/NSEC_PER_MSEC/ and
> computing delta after testing the time since last update
> 
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index a80a73909dc2..e48fd0e6982d 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -3665,6 +3665,7 @@ static inline bool cfs_rq_is_decayed(struct
> cfs_rq *cfs_rq)
>  static inline void update_tg_load_avg(struct cfs_rq *cfs_rq)
>  {
>         long delta = cfs_rq->avg.load_avg - cfs_rq->tg_load_avg_contrib;
> +       u64 now = cfs_rq_clock_pelt(cfs_rq);

Could this be `u64 now = sched_clock_cpu()` like in
migrate_se_pelt_lag() or newidle_balance() to avoid the time morphing
due to PELT's frequency and uArch invariance?
> 
>         /*
>          * No need to update load_avg for root_task_group as it is not used.

[...]