Re: [RFC PATCH 5/6] sched/fair: Select an energy-efficient CPU on task wake-up

[Joel Fernandes]

Hi Morten,

On Fri, Mar 23, 2018 at 8:47 AM, Morten Rasmussen
<morten.rasmussen@xxxxxxx> wrote:
> On Thu, Mar 22, 2018 at 01:10:22PM -0700, Joel Fernandes wrote:
>> On Wed, Mar 21, 2018 at 8:35 AM, Patrick Bellasi
>> <patrick.bellasi@xxxxxxx> wrote:
>> > [...]
>> >
>> >> @@ -6555,6 +6613,14 @@ select_task_rq_fair(struct task_struct *p, int prev_cpu, int sd_flag, int wake_f
>> >>                       break;
>> >>               }
>> >>
>> >> +             /*
>> >> +              * Energy-aware task placement is performed on the highest
>> >> +              * non-overutilized domain spanning over cpu and prev_cpu.
>> >> +              */
>> >> +             if (want_energy && !sd_overutilized(tmp) &&
>> >> +                 cpumask_test_cpu(prev_cpu, sched_domain_span(tmp)))
>> >> +                     energy_sd = tmp;
>> >> +
>> >
>> > Not entirely sure, but I was trying to understand if we can avoid to
>> > modify the definition of want_affine (in the previous chunk) and move
>> > this block before the previous "if (want_affine..." (in mainline but
>> > not in this chunk), which will became an else, e.g.
>> >
>> >         if (want_energy && !sd_overutilized(tmp) &&
>> >                 // ...
>> >         else if (want_energy && !sd_overutilized(tmp) &&
>> >                 // ...
>> >
>> > Isn't that the same?
>> >
>> > Maybe there is a code path I'm missing... but otherwise it seems a
>> > more self contained modification of select_task_rq_fair...
>>
>> Just replying to this here Patrick instead of the other thread.
>>
>> I think this is the right place for the block from Quentin quoted
>> above because we want to search for the highest domain that is
>> !overutilized and look among those for the candidates. So from that
>> perspective, we can't move the block to the beginning and it seems to
>> be in the right place. My main concern on the other thread was
>> different, I was talking about the cases where sd_flag & tmp->flags
>> don't match. In that case, sd = NULL would trump EAS and I was
>> wondering if that's the right thing to do...
>
> You mean if SD_BALANCE_WAKE isn't set on sched_domains?

Yes.

> The current code seems to rely on that flag to be set to work correctly.
> Otherwise, the loop might bail out on !want_affine and we end up doing
> the find_energy_efficient_cpu() on the lowest level sched_domain even if
> there is higher level one which isn't over-utilized.
>
> However, SD_BALANCE_WAKE should be set if SD_ASYM_CPUCAPACITY is set so
> sd == NULL shouldn't be possible? This only holds as long as we only
> want EAS for asymmetric systems.

Yes, I see you had topology code that set SD_BALANCE_WAKE for ASYM. It
makes sense to me then, thanks for the clarification.

Still I feel it is a bit tedious/confusing when reading code to draw
the conclusion about why sd is checked first before doing
find_energy_efficient_cpu (and that sd will != NULL for ASYM systems).
If energy_sd is set, then we can just proceed with EAS without
checking that sd != NULL. This function in mainline is already pretty
confusing as it is :-(

Regards,

- Joel