Re: [PATCH 1/2] x86,sched: Add support for frequency invariance

[Peter Zijlstra]

On Tue, Sep 17, 2019 at 04:27:46PM +0200, Giovanni Gherdovich wrote:
> Hello Quentin,
> 
> On Sat, 2019-09-14 at 12:57 +0200, Quentin Perret wrote:
> > Hi Giovanni
> > 
> > On Monday 09 Sep 2019 at 04:42:15 (+0200), Giovanni Gherdovich wrote:
> > > +static inline long arch_scale_freq_capacity(int cpu)
> > > +{
> > > +	if (static_cpu_has(X86_FEATURE_APERFMPERF))
> > > +		return per_cpu(arch_cpu_freq, cpu);
> > 
> > So, if this is conditional, perhaps you could also add this check in an
> > x86-specific implementation of arch_scale_freq_invariant() ? That would
> > guide sugov in the right path (see get_next_freq()) if APERF/MPERF are
> > unavailable.
> > 
> > > +	return 1024 /* SCHED_CAPACITY_SCALE */;
> > > +}
> >
> 
> Good remark. If the cpu doesn't have APERF/MPERF, the choice here is that
> freq_curr is constantly equal to freq_max, and the scaling factor is 1 all the
> time.
> 
> But I'm checking this static_cpu_has() every time I do a frequency update;
> arguably schedutil should be smarter and settle such a case once and for all
> at boot time.
> 
> I'll check what's the cost of static_cpu_has() and if it's non-negligible I'll
> do what you suggest (x86-specific version of arch_scale_freq_invariant().

static_cpu_has() is an alternative and ends up being a static branch
(similar to static_key) once the alternative patching runs.