Re: [RFC][PATCH 1/7] cpufreq / sched: Make schedutil access utilization data directly

From: Juri Lelli
Date: Tue Aug 02 2016 - 06:38:42 EST


Hi,

On 02/08/16 01:46, Rafael J. Wysocki wrote:
> On Monday, August 01, 2016 12:28:50 PM Steve Muckle wrote:
> > On Mon, Aug 01, 2016 at 01:34:36AM +0200, Rafael J. Wysocki wrote:
> > ...
> > > Index: linux-pm/kernel/sched/cpufreq_schedutil.c
> > > ===================================================================
> > > --- linux-pm.orig/kernel/sched/cpufreq_schedutil.c
> > > +++ linux-pm/kernel/sched/cpufreq_schedutil.c
> > > @@ -144,17 +144,47 @@ static unsigned int get_next_freq(struct
> > > return cpufreq_driver_resolve_freq(policy, freq);
> > > }
> > >
> > > -static void sugov_update_single(struct update_util_data *hook, u64 time,
> > > - unsigned long util, unsigned long max)
> > > +static void sugov_get_util(unsigned long *util, unsigned long *max)
> > > +{
> > > + unsigned long dl_util, dl_max;
> > > + unsigned long cfs_util, cfs_max;
> > > + int cpu = smp_processor_id();
> > > + struct dl_bw *dl_bw = dl_bw_of(cpu);
> > > + struct rq *rq = this_rq();
> > > +
> > > + if (rt_prio(current->prio)) {
> > > + *util = ULONG_MAX;
> > > + return;
> > > + }
> > > +
> > > + dl_max = dl_bw_cpus(cpu) << 20;
> > > + dl_util = dl_bw->total_bw;
> > > +
> > > + cfs_max = rq->cpu_capacity_orig;
> > > + cfs_util = min(rq->cfs.avg.util_avg, cfs_max);
> > > +
> > > + if (cfs_util * dl_max > dl_util * cfs_max) {
> > > + *util = cfs_util;
> > > + *max = cfs_max;
> > > + } else {
> > > + *util = dl_util;
> > > + *max = dl_max;
> > > + }
> > > +}
> >
> > Last Friday I had put together a similar patch based on Peter's. I need
> > the flags field for the remote wakeup support. My previous plan,
> > installing a late callback in check_preempt_curr that gets requested
> > from the earlier existing CFS callback, was not working out since those
> > two events don't always match up 1:1.
> >
> > Anyway one way that my patch differed was that I had used the flags
> > field to keep the behavior the same for both RT and DL.

Do you mean "go to max" policy for both, until proper policies will be
implemented in the future?

> That happens
> > later on in this series for RT but the DL policy is modified as above.
> > Can the DL policy be left as-is and discussed/modified in a separate
> > series?

Not that we want to start discussing this point now, if we postpone the
change for later, but I just wanted to point out a difference w.r.t.
what the schedfreq thing was doing: it used to sum contributions from
the different classes, instead of taking the max. We probably never
really discussed on the list what is the right thing to do, though.

Best,

- Juri