Re: Abysmal scheduler performance in Linus' tree?

[Peter Zijlstra]

On Wed, Sep 06, 2017 at 11:18:46AM +0100, Chris Wilson wrote:

> > +static void get_llc_stats(struct llc_stats *stats, int cpu)
> > +{
> > +       struct sched_domain_shared *sds = rcu_dereference(per_cpu(sd_llc_shared, cpu));
> > +
> > +       if (!sds) {
> > +               memset(&stats, 0, sizeof(*stats));
> 
> Yes, I even sent you a mail about it ;)

Bah, too much email, sorry :-(

> > +       /*
> > +        * The has_capacity stuff is not SMT aware, but by trying to balance
> > +        * the nr_running on both ends we try and fill the domain at equal
> > +        * rates, thereby first consuming cores before siblings.
> > +        */
> > +
> > +       /* if the old cache has capacity, stay there */
> > +       if (prev_stats.has_capacity && prev_stats.nr_running < this_stats.nr_running+1)
> > +               return false;
> > +
> > +       /* if this cache has capacity, come here */
> > +       if (this_stats.has_capacity && this_stats.nr_running < prev_stats.nr_running+1)
> > +               return true;
> 
> This is still not working as intended, it should be 
> 
> 	if (this_stats.has_capacity && this_stats.nr_running+1 < prev_stats.nr_running)
> 		return true;
> 
> to fix the regression.

Argh, you're quite right. Let me do a patch for that.