Re: [PATCH] sched/proxy_exec: Optimize proxy_tag_curr() pushable removal

[John Stultz]

On Tue, Mar 3, 2026 at 9:33 PM <soolaugust@xxxxxxxxx> wrote:
>
> From: zhidao su <suzhidao@xxxxxxxxxx>
>
> On Wed, Mar 4, 2026 at 10:28 AM John Stultz <jstultz@xxxxxxxxxx> wrote:
> > Yeah, I agree we should use a sched_class-> method rather then doing
> > a big conditional tree.
> >
> > Oh hmm, good catch! So I think you're right that right now we're
> > leaning on the owner to be run without a donor before we return it
> > to the pushable list, which isn't great.
> >
> > I'll take a swing at addressing this (along with your other feedback
> > - this year has been hectic and I've only just now finally managed
> > to get back to proxy-exec upstreaming efforts!) for v25.
>
> Understood, and thanks to both of you for the thorough review.
>
> The sched_class->proxy_tag_curr() direction makes sense -- it avoids
> the conditional tree and keeps each class's pushable logic self-
> contained. And the missing proxy_untag_current() is a real gap I
> had not considered; good to know it will be addressed in v25.
>
> I'll drop this RFC and wait for v25. Looking forward to it.

Thanks! I'm going to try to send just two patches to address this
issue now as an RFC, so its not stuck behind the rest of my v25
rework.

Again, I really appreciate your submission here, the analysis of the
overhead cost you shared here is very clear and helps motivate fixing
this properly.

thanks
-john