Re: [PATCH v4] sched/freq: move call to cpufreq_update_util

[Rafael J. Wysocki]

On Fri, Nov 15, 2019 at 11:46 AM Vincent Guittot
<vincent.guittot@xxxxxxxxxx> wrote:
>
> On Fri, 15 Nov 2019 at 11:37, Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
> >
> > On Fri, Nov 15, 2019 at 11:18:00AM +0100, Vincent Guittot wrote:
> > > On Fri, 15 Nov 2019 at 10:55, Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
> > > >
> > > > On Thu, Nov 14, 2019 at 06:07:31PM +0100, Vincent Guittot wrote:
> > > > > update_cfs_rq_load_avg() calls cfs_rq_util_change() everytime pelt decays,
> > > > > which might be inefficient when cpufreq driver has rate limitation.
> > > > >
> > > > > When a task is attached on a CPU, we have call path:
> > > > >
> > > > > update_load_avg()
> > > > >   update_cfs_rq_load_avg()
> > > > >     cfs_rq_util_change -- > trig frequency update
> > > > >   attach_entity_load_avg()
> > > > >     cfs_rq_util_change -- > trig frequency update
> > > > >
> > > > > The 1st frequency update will not take into account the utilization of the
> > > > > newly attached task and the 2nd one might be discard because of rate
> > > > > limitation of the cpufreq driver.
> > > >
> > > > Doesn't this just show that a dumb rate limit in the driver is broken?
> > >
> > > But the rate limit may come from HW constraints that forces to wait
> > > let say 4ms or even 10ms between each frequency update.
> >
> > Sure, but then it can still remember the value passed in last and use
> > that state later.
> >
> > It doesn't _have_ to completely discard values.
>
> yes but it means that we run at the "wrong" frequency during this
> period and also that the cpufreq must in this case set a kind of timer
> to resubmit a new frequency change out of scheduler event

The driver would need to do that, because from the cpufreq core
perspective it is in-band.

Which would kind of defeat the purpose of driving it from the
scheduler, wouldn't it?