Re: [PATCH 6/6 v3] sched/eevdf: Speedup short slice task scheduling

[Vincent Guittot]

On Fri, 26 Jun 2026 at 15:55, Vincent Guittot
<vincent.guittot@xxxxxxxxxx> wrote:
>
> On Fri, 26 Jun 2026 at 10:55, Vincent Guittot
> <vincent.guittot@xxxxxxxxxx> wrote:
> >
> > On Fri, 26 Jun 2026 at 09:55, Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
> > >
> > > On Fri, Jun 26, 2026 at 09:18:32AM +0200, Peter Zijlstra wrote:
> > > > On Fri, Jun 26, 2026 at 09:05:29AM +0200, Vincent Guittot wrote:
> > > > > On Fri, 26 Jun 2026 at 05:57, K Prateek Nayak <kprateek.nayak@xxxxxxx> wrote:
> > > > > >
> > > > > > Hello Peter,
> > > > > >
> > > > > > On 6/26/2026 3:58 AM, Peter Zijlstra wrote:
> > > > > > > +static u64 ineligible_vruntime(struct cfs_rq *cfs_rq)
> > > > > > > +{
> > > > > > > +     struct sched_entity *curr = cfs_rq->curr;
> > > > > > > +     long weight = cfs_rq->sum_weight;
> > > > > > > +     s64 delta = 0;
> > > > > > > +
> > > > > > > +     if (curr && !curr->on_rq)
> > > > > > > +             curr = NULL;
> > > > > > > +
> > > > > > > +     /*
> > > > > > > +      * This is called from set_next_task_fair(.first=true) /
> > > > > > > +      * set_protect_slice() so curr had better be set and on_rq.
> > > > > > > +      */
> > > > > > > +     WARN_ON_ONCE(!curr);
> > > > > >
> > > > > > set_protect_slice() is indeed called from set_next_entity(.first=true)
> > > > > > but it is done after __dequeue_entity() and before "cfs_rq->curr" is
> > > > > > set (both sched/flat and sched/core have the same pattern).
> > > > >
> > > > > Yes, I confirm that set_protect_slice() is called before setting curr
> > > > > in tip/sched/core but I think that Peter added flat hierarchy patches
> > > > > in his tree
> > > >
> > > > Indeed I did; I was planning to stick them in this cycle, so I figure I
> > > > should be doing these patches on top.
> > >
> > > I pushed out the now booting pile of gunk into queue:sched/core.
> > >
> > > I've made some rando edits to your (Vincent's) patches, if you disagree
> > > or want changes, let me know. This really is just my quilt stack of the
> > > moment for sched/core and it can easily be changed.
> >
> > Okay, I'm going to have a look
>
> The patches look good to me:
> - The addition of set_short_buddy() makes sense
> - The fix of ineligible_vruntime too
>

I run a bunch of perf and latency tests at roor level with your
sched/core on top of  7.2-rc2 and results looks good

> >
> >
> > >
> > > /me wonders how the fuck it can be 28C before 10am and goes in search of
> > > more water.