Re: [PATCH] sched/fair: Simplify some logic in update_sd_pick_busiest()

[David Vernet <void@manifault.com>]

[-- Attachment #1: Type: text/plain, Size: 2678 bytes --]

On Fri, Feb 02, 2024 at 06:40:21PM +0100, Valentin Schneider wrote:
> On 02/02/24 11:07, David Vernet wrote:
> > On Fri, Feb 02, 2024 at 06:01:22PM +0100, Valentin Schneider wrote:
> >> On 02/02/24 01:02, David Vernet wrote:
> >> > When comparing the current struct sched_group with the yet-busiest
> >> > domain in update_sd_pick_busiest(), if the two groups have the same
> >> > group type, we're currently doing a bit of unnecessary work for any
> >> > group >= group_misfit_task. We're comparing the two groups, and then
> >> > returning only if false (the group in question is not the busiest).
> >> > Othewise, we break, do an extra unnecessary conditional check that's
> >> > vacuously false for any group type > group_fully_busy, and then always
> >> > return true.
> >> >
> >> > Let's just return directly in the switch statement instead. This doesn't
> >> > change the size of vmlinux with llvm 17 (not surprising given that all
> >> > of this is inlined in load_balance()), but it does shrink load_balance()
> >> > by 88 bytes on x86. Given that it also improves readability, this seems
> >> > worth doing.
> >> >
> >> > As a bonus, remove an unnecessary goto in update_sd_lb_stats().
> >> >
> >>
> >> Given that's a different scope than what the rest of the patch touches, I'd
> >> rather see this as a separate patch.
> >>
> >> Other than that:
> >> Reviewed-by: Valentin Schneider <vschneid@redhat.com>
> >
> > Thanks, would you like me to send a follow-on series split into two with
> > your tag on both? Or were you just letting me know for next time?
> >
> 
> Well, I'm not picking up any patches, just reviewing them :) So yes I'd say
> re-send with the split and feel free to apply the tag on both.

Sounds good, I'll send a follow up at some point tomorrow or early next
week.

> > We could also update this check to only do a strict greater than to
> > avoid unnecessary writes, but I figured it was preferable to have no
> > logical changes for this iteration:
> >
> > return sgs->group_misfit_task_load >= busiest->group_misfit_task_load;
> 
> That's a good point, I don't think there was a specific reason for going
> with a lower-than rather than a lower-or-equal back then:
>   cad68e552e77 ("sched/fair: Consider misfit tasks when load-balancing")

Yeah, the goal is to choose the group with the highest misfit task load,
so it seems pretty straightforward that we don't gain anything from
choosing a new group that has the same load as the previous busiest.

When I send out the follow-on patch set, I'll do it as 3 separate
patches (unless I hear otherwise from someone else).

Thanks,
David

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]