Re: [PATCH v3] sched/fair: bring back select_idle_smt, but differently

[Rik van Riel <riel@surriel.com>]

[-- Attachment #1: Type: text/plain, Size: 2217 bytes --]

On Tue, 2021-04-06 at 17:31 +0200, Vincent Guittot wrote:
> On Tue, 6 Apr 2021 at 17:26, Rik van Riel <riel@surriel.com> wrote:
> > On Tue, 2021-04-06 at 17:10 +0200, Vincent Guittot wrote:
> > > On Fri, 26 Mar 2021 at 20:19, Rik van Riel <riel@surriel.com>
> > > wrote:
> > > 
> > > > -static int select_idle_cpu(struct task_struct *p, struct
> > > > sched_domain *sd, int target)
> > > > +static int select_idle_cpu(struct task_struct *p, struct
> > > > sched_domain *sd, int prev, int target)
> > > >  {
> > > >         struct cpumask *cpus =
> > > > this_cpu_cpumask_var_ptr(select_idle_mask);
> > > >         int i, cpu, idle_cpu = -1, nr = INT_MAX;
> > > > @@ -6136,23 +6163,32 @@ static int select_idle_cpu(struct
> > > > task_struct *p, struct sched_domain *sd, int t
> > > > 
> > > >         cpumask_and(cpus, sched_domain_span(sd), p->cpus_ptr);
> > > > 
> > > > -       if (sched_feat(SIS_PROP) && !smt) {
> > > > -               u64 avg_cost, avg_idle, span_avg;
> > > > +       if (!smt) {
> > > > +               if (cpus_share_cache(prev, target)) {
> > > 
> > > Have you checked the impact on no smt system ? would worth a
> > > static
> > > branch.
> > > 
> > > Also, this doesn't need to be in select_idle_cpu() which aims to
> > > loop
> > > the sched_domain becaus you only compare  target and prev. So you
> > > can
> > > move this call to select_idle_smt() in select_idle_sibling()
> > 
> > After Mel's rewrite, there no longer are calls to
> > select_idle_core() or select_idle_smt() in select_idle_sibling().
> 
> select_idle_smt() had even disappeared that why it was not in
> select_idle_sibling
> 
> > Everything got folded into one single loop in select_idle_cpu()
> 
> but this is done completely out of the loop so we don't need to
> complify the function with unrelated stuff

Not entirely. The call to select_idle_smt() is still
conditional on test_idle_cores() returning false.

We only look for the
other sibling if there is no idle
core in the LLC. If there is an idle core, we prefer
that.

Pulling the select_idle_smt() call out of select_idle_cpu()
would mean having to test_idle_cores() twice.

-- 
All Rights Reversed.

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 488 bytes --]