Re: [PATCH v4 1/2] sched/fair: Check a task has a fitting cpu when updating misfit

[Dietmar Eggemann]

On 06/02/2024 16:06, Qais Yousef wrote:
> On 02/05/24 20:49, Dietmar Eggemann wrote:
>> On 26/01/2024 02:46, Qais Yousef wrote:
>>> On 01/25/24 18:40, Vincent Guittot wrote:
>>>> On Wed, 24 Jan 2024 at 23:30, Qais Yousef <qyousef@xxxxxxxxxxx> wrote:
>>>>>
>>>>> On 01/23/24 09:26, Vincent Guittot wrote:
>>>>>> On Fri, 5 Jan 2024 at 23:20, Qais Yousef <qyousef@xxxxxxxxxxx> wrote:
>>>>>>>
>>>>>>> From: Qais Yousef <qais.yousef@xxxxxxx>

[...]

>>> It seems we flatten topologies but not sched domains. I see all cpus shown as
>>> core_siblings. The DT for apple silicon sets clusters in the cpu-map - which
>>> seems the flatten topology stuff detect LLC correctly but still keeps the
>>> sched-domains not flattened. Is this a bug? I thought we will end up with one
>>> sched domain still.
>>
>> IMHO, if you have a cpu_map entry with > 1 cluster in your dtb, you end
>> up with MC and PKG (former DIE) Sched Domain (SD) level. And misfit load
> 
> Hmm, okay. I thought the detection of topology where we know the LLC is shared
> will cause the sched domains to collapse too.
> 
>> balance takes potentially longer on PKG than to MC.
> 
> Why potentially longer? We iterate through the domains the CPU belong to. If
> the first iteration (at MC) pulled something, then once we go to PKG then we're
> less likely to pull again?

There are a couple of mechanisms in place to let load-balance on higher
sd levels happen less frequently, eg:

  load_balance() -> should_we_balance() + continue_balancing

  interval = get_sd_balance_interval(sd, busy) in rebalance_domains()

  rq->avg_idle versus sd->max_newidle_lb_cost

> Anyway. I think I am hitting a bug here. The behavior doesn't look right to me
> given the delays I'm seeing and the fact we do the ilb but for some reason fail
> to pull

[...]