Re: [PATCH 7/9] sched/fair: Enforce proportional scan limits when scanning for an idle core

[Mel Gorman]

On Mon, Aug 02, 2021 at 10:52:01AM +0000, Song Bao Hua (Barry Song) wrote:
> > @@ -6265,30 +6265,35 @@ static int select_idle_cpu(struct task_struct *p, struct
> > sched_domain *sd, bool
> >  		if (has_idle_core) {
> >  			i = select_idle_core(p, cpu, cpus, &idle_cpu);
> >  			if ((unsigned int)i < nr_cpumask_bits)
> > -				return i;
> > +				break;
> > 
> > +			nr -= sched_smt_weight;
> >  		} else {
> > -			if (!--nr)
> > -				return -1;
> >  			idle_cpu = __select_idle_cpu(cpu, p);
> >  			if ((unsigned int)idle_cpu < nr_cpumask_bits)
> >  				break;
> > +			nr--;
> >  		}
> > +
> > +		if (nr < 0)
> > +			break;
> >  	}
> > 
> > -	if (has_idle_core)
> > -		set_idle_cores(target, false);
> > +	if ((unsigned int)idle_cpu < nr_cpumask_bits) {
> > +		if (has_idle_core)
> > +			set_idle_cores(target, false);
> > 
> 
> For example, if we have 16 cpus(8 SMT2 cores). In case core7 is idle,
> we only have scanned core0+core1(cpu0-cpu3) and if these two cores
> are not idle, but here we set has_idle_cores to false while core7 is
> idle. It seems incorrect.
> 

Yep, that block needs to be revisited.

-- 
Mel Gorman
SUSE Labs