linux-trace-kernel.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Qais Yousef <qyousef@layalina.io>
To: Vincent Guittot <vincent.guittot@linaro.org>
Cc: linux@armlinux.org.uk, catalin.marinas@arm.com, will@kernel.org,
	sudeep.holla@arm.com, rafael@kernel.org, viresh.kumar@linaro.org,
	agross@kernel.org, andersson@kernel.org,
	konrad.dybcio@linaro.org, mingo@redhat.com, peterz@infradead.org,
	juri.lelli@redhat.com, dietmar.eggemann@arm.com,
	rostedt@goodmis.org, bsegall@google.com, mgorman@suse.de,
	bristot@redhat.com, vschneid@redhat.com, lukasz.luba@arm.com,
	rui.zhang@intel.com, mhiramat@kernel.org,
	daniel.lezcano@linaro.org, amit.kachhap@gmail.com,
	corbet@lwn.net, gregkh@linuxfoundation.org,
	linux-arm-kernel@lists.infradead.org,
	linux-kernel@vger.kernel.org, linux-pm@vger.kernel.org,
	linux-arm-msm@vger.kernel.org,
	linux-trace-kernel@vger.kernel.org, linux-doc@vger.kernel.org
Subject: Re: [PATCH v4 2/5] sched: Take cpufreq feedback into account
Date: Tue, 30 Jan 2024 22:09:32 +0000	[thread overview]
Message-ID: <20240130220932.2wvnscmzqadmsrd2@airbuntu> (raw)
In-Reply-To: <CAKfTPtCJ9TXRRrCPsTam=uReDtay7dRqBpZCWQkOsbieWTuMbw@mail.gmail.com>

On 01/30/24 10:35, Vincent Guittot wrote:
> On Tue, 30 Jan 2024 at 01:50, Qais Yousef <qyousef@layalina.io> wrote:
> >
> > On 01/30/24 00:26, Qais Yousef wrote:
> > > On 01/09/24 17:46, Vincent Guittot wrote:
> > > > Aggregate the different pressures applied on the capacity of CPUs and
> > > > create a new function that returns the actual capacity of the CPU:
> > > >   get_actual_cpu_capacity()
> > > >
> > > > Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
> > > > Reviewed-by: Lukasz Luba <lukasz.luba@arm.com>
> > > > ---
> > > >  kernel/sched/fair.c | 45 +++++++++++++++++++++++++--------------------
> > > >  1 file changed, 25 insertions(+), 20 deletions(-)
> > > >
> > > > diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> > > > index 9cc20855dc2b..e54bbf8b4936 100644
> > > > --- a/kernel/sched/fair.c
> > > > +++ b/kernel/sched/fair.c
> > > > @@ -4910,13 +4910,22 @@ static inline void util_est_update(struct cfs_rq *cfs_rq,
> > > >     trace_sched_util_est_se_tp(&p->se);
> > > >  }
> > > >
> > > > +static inline unsigned long get_actual_cpu_capacity(int cpu)
> > > > +{
> > > > +   unsigned long capacity = arch_scale_cpu_capacity(cpu);
> > > > +
> > > > +   capacity -= max(thermal_load_avg(cpu_rq(cpu)), cpufreq_get_pressure(cpu));
> > >
> > > Does cpufreq_get_pressure() reflect thermally throttled frequency, or just the
> > > policy->max being capped by user etc? I didn't see an update to cpufreq when we
> > > topology_update_hw_pressure(). Not sure if it'll go through another path.
> >
> > It is done via the cooling device. And assume any limitations on freq due to
> > power etc are assumed to always to cause the policy->max to change.
> >
> > (sorry if I missed earlier discussions about this)
> 
> I assume that you have answered all your questions.
> 
> We have now 2 distinct signals:
> - hw high freq update which is averaged with PELT and go through
> topology_update_hw_pressure
> - cpufreq pressure which is not averaged (including cpufreq cooling
> device with patch 3)

Yes. I think a comment like suggested below is useful to help keeping the code
understandable to new comers. But FWIW

Reviewed-by: Qais Yousef <qyousef@layalina.io>

> 
> >
> > >
> > > maxing with thermal_load_avg() will change the behavior below where we used to
> > > compare against instantaneous pressure. The concern was that it not just can
> > > appear quickly, but disappear quickly too. thermal_load_avg() will decay
> > > slowly, no?  This means we'll lose a lot of opportunities for better task
> > > placement until this decays which can take relatively long time.
> > >
> > > So maxing handles the direction where a pressure suddenly appears. But it
> > > doesn't handle where it disappears.
> > >
> > > I suspect your thoughts are that if it was transient then thermal_load_avg()
> > > should be small anyway - which I think makes sense.
> > >
> > > I think we need a comment to explain these nuance differences.
> > >
> > > > +
> > > > +   return capacity;
> > > > +}
> > > > +
> > > >  static inline int util_fits_cpu(unsigned long util,
> > > >                             unsigned long uclamp_min,
> > > >                             unsigned long uclamp_max,
> > > >                             int cpu)
> > > >  {
> > > > -   unsigned long capacity_orig, capacity_orig_thermal;
> > > >     unsigned long capacity = capacity_of(cpu);
> > > > +   unsigned long capacity_orig;
> > > >     bool fits, uclamp_max_fits;
> > > >
> > > >     /*
> > > > @@ -4948,7 +4957,6 @@ static inline int util_fits_cpu(unsigned long util,
> > > >      * goal is to cap the task. So it's okay if it's getting less.
> > > >      */
> > > >     capacity_orig = arch_scale_cpu_capacity(cpu);
> > > > -   capacity_orig_thermal = capacity_orig - arch_scale_thermal_pressure(cpu);
> > > >
> > > >     /*
> > > >      * We want to force a task to fit a cpu as implied by uclamp_max.
> > > > @@ -5023,7 +5031,8 @@ static inline int util_fits_cpu(unsigned long util,
> > > >      * handle the case uclamp_min > uclamp_max.
> > > >      */
> > > >     uclamp_min = min(uclamp_min, uclamp_max);
> > > > -   if (fits && (util < uclamp_min) && (uclamp_min > capacity_orig_thermal))
> > > > +   if (fits && (util < uclamp_min) &&
> > > > +       (uclamp_min > get_actual_cpu_capacity(cpu)))
> > > >             return -1;
> > > >
> > > >     return fits;
> > > > @@ -7404,7 +7413,7 @@ select_idle_capacity(struct task_struct *p, struct sched_domain *sd, int target)
> > > >              * Look for the CPU with best capacity.
> > > >              */
> > > >             else if (fits < 0)
> > > > -                   cpu_cap = arch_scale_cpu_capacity(cpu) - thermal_load_avg(cpu_rq(cpu));
> > > > +                   cpu_cap = get_actual_cpu_capacity(cpu);
> > > >
> > > >             /*
> > > >              * First, select CPU which fits better (-1 being better than 0).
> > > > @@ -7897,8 +7906,8 @@ static int find_energy_efficient_cpu(struct task_struct *p, int prev_cpu)
> > > >     struct root_domain *rd = this_rq()->rd;
> > > >     int cpu, best_energy_cpu, target = -1;
> > > >     int prev_fits = -1, best_fits = -1;
> > > > -   unsigned long best_thermal_cap = 0;
> > > > -   unsigned long prev_thermal_cap = 0;
> > > > +   unsigned long best_actual_cap = 0;
> > > > +   unsigned long prev_actual_cap = 0;
> > > >     struct sched_domain *sd;
> > > >     struct perf_domain *pd;
> > > >     struct energy_env eenv;
> > > > @@ -7928,7 +7937,7 @@ static int find_energy_efficient_cpu(struct task_struct *p, int prev_cpu)
> > > >
> > > >     for (; pd; pd = pd->next) {
> > > >             unsigned long util_min = p_util_min, util_max = p_util_max;
> > > > -           unsigned long cpu_cap, cpu_thermal_cap, util;
> > > > +           unsigned long cpu_cap, cpu_actual_cap, util;
> > > >             long prev_spare_cap = -1, max_spare_cap = -1;
> > > >             unsigned long rq_util_min, rq_util_max;
> > > >             unsigned long cur_delta, base_energy;
> > > > @@ -7940,18 +7949,17 @@ static int find_energy_efficient_cpu(struct task_struct *p, int prev_cpu)
> > > >             if (cpumask_empty(cpus))
> > > >                     continue;
> > > >
> > > > -           /* Account thermal pressure for the energy estimation */
> > > > +           /* Account external pressure for the energy estimation */
> > > >             cpu = cpumask_first(cpus);
> > > > -           cpu_thermal_cap = arch_scale_cpu_capacity(cpu);
> > > > -           cpu_thermal_cap -= arch_scale_thermal_pressure(cpu);
> > > > +           cpu_actual_cap = get_actual_cpu_capacity(cpu);
> > > >
> > > > -           eenv.cpu_cap = cpu_thermal_cap;
> > > > +           eenv.cpu_cap = cpu_actual_cap;
> > > >             eenv.pd_cap = 0;
> > > >
> > > >             for_each_cpu(cpu, cpus) {
> > > >                     struct rq *rq = cpu_rq(cpu);
> > > >
> > > > -                   eenv.pd_cap += cpu_thermal_cap;
> > > > +                   eenv.pd_cap += cpu_actual_cap;
> > > >
> > > >                     if (!cpumask_test_cpu(cpu, sched_domain_span(sd)))
> > > >                             continue;
> > > > @@ -8022,7 +8030,7 @@ static int find_energy_efficient_cpu(struct task_struct *p, int prev_cpu)
> > > >                     if (prev_delta < base_energy)
> > > >                             goto unlock;
> > > >                     prev_delta -= base_energy;
> > > > -                   prev_thermal_cap = cpu_thermal_cap;
> > > > +                   prev_actual_cap = cpu_actual_cap;
> > > >                     best_delta = min(best_delta, prev_delta);
> > > >             }
> > > >
> > > > @@ -8037,7 +8045,7 @@ static int find_energy_efficient_cpu(struct task_struct *p, int prev_cpu)
> > > >                      * but best energy cpu has better capacity.
> > > >                      */
> > > >                     if ((max_fits < 0) &&
> > > > -                       (cpu_thermal_cap <= best_thermal_cap))
> > > > +                       (cpu_actual_cap <= best_actual_cap))
> > > >                             continue;
> > > >
> > > >                     cur_delta = compute_energy(&eenv, pd, cpus, p,
> > > > @@ -8058,14 +8066,14 @@ static int find_energy_efficient_cpu(struct task_struct *p, int prev_cpu)
> > > >                     best_delta = cur_delta;
> > > >                     best_energy_cpu = max_spare_cap_cpu;
> > > >                     best_fits = max_fits;
> > > > -                   best_thermal_cap = cpu_thermal_cap;
> > > > +                   best_actual_cap = cpu_actual_cap;
> > > >             }
> > > >     }
> > > >     rcu_read_unlock();
> > > >
> > > >     if ((best_fits > prev_fits) ||
> > > >         ((best_fits > 0) && (best_delta < prev_delta)) ||
> > > > -       ((best_fits < 0) && (best_thermal_cap > prev_thermal_cap)))
> > > > +       ((best_fits < 0) && (best_actual_cap > prev_actual_cap)))
> > > >             target = best_energy_cpu;
> > > >
> > > >     return target;
> > > > @@ -9441,8 +9449,8 @@ static inline void init_sd_lb_stats(struct sd_lb_stats *sds)
> > > >
> > > >  static unsigned long scale_rt_capacity(int cpu)
> > > >  {
> > > > +   unsigned long max = get_actual_cpu_capacity(cpu);
> > > >     struct rq *rq = cpu_rq(cpu);
> > > > -   unsigned long max = arch_scale_cpu_capacity(cpu);
> > > >     unsigned long used, free;
> > > >     unsigned long irq;
> > > >
> > > > @@ -9454,12 +9462,9 @@ static unsigned long scale_rt_capacity(int cpu)
> > > >     /*
> > > >      * avg_rt.util_avg and avg_dl.util_avg track binary signals
> > > >      * (running and not running) with weights 0 and 1024 respectively.
> > > > -    * avg_thermal.load_avg tracks thermal pressure and the weighted
> > > > -    * average uses the actual delta max capacity(load).
> > > >      */
> > > >     used = READ_ONCE(rq->avg_rt.util_avg);
> > > >     used += READ_ONCE(rq->avg_dl.util_avg);
> > > > -   used += thermal_load_avg(rq);
> > > >
> > > >     if (unlikely(used >= max))
> > > >             return 1;
> > > > --
> > > > 2.34.1
> > > >

  reply	other threads:[~2024-01-30 22:09 UTC|newest]

Thread overview: 17+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2024-01-09 16:46 [PATCH v4 0/5] Rework system pressure interface to the scheduler Vincent Guittot
2024-01-09 16:46 ` [PATCH v4 1/5] cpufreq: Add a cpufreq pressure feedback for " Vincent Guittot
2024-01-09 16:49   ` Rafael J. Wysocki
2024-01-09 17:00     ` Vincent Guittot
2024-01-10  6:43   ` Viresh Kumar
2024-01-30  0:05   ` Qais Yousef
2024-01-09 16:46 ` [PATCH v4 2/5] sched: Take cpufreq feedback into account Vincent Guittot
2024-01-30  0:26   ` Qais Yousef
2024-01-30  0:50     ` Qais Yousef
2024-01-30  9:35       ` Vincent Guittot
2024-01-30 22:09         ` Qais Yousef [this message]
2024-01-09 16:46 ` [PATCH v4 3/5] thermal/cpufreq: Remove arch_update_thermal_pressure() Vincent Guittot
2024-01-09 16:46 ` [PATCH v4 4/5] sched: Rename arch_update_thermal_pressure into arch_update_hw_pressure Vincent Guittot
2024-01-30  0:33   ` Qais Yousef
2024-01-30  0:47   ` Qais Yousef
2024-01-09 16:46 ` [PATCH v4 5/5] sched/pelt: Remove shift of thermal clock Vincent Guittot
2024-01-30  0:36   ` Qais Yousef

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20240130220932.2wvnscmzqadmsrd2@airbuntu \
    --to=qyousef@layalina.io \
    --cc=agross@kernel.org \
    --cc=amit.kachhap@gmail.com \
    --cc=andersson@kernel.org \
    --cc=bristot@redhat.com \
    --cc=bsegall@google.com \
    --cc=catalin.marinas@arm.com \
    --cc=corbet@lwn.net \
    --cc=daniel.lezcano@linaro.org \
    --cc=dietmar.eggemann@arm.com \
    --cc=gregkh@linuxfoundation.org \
    --cc=juri.lelli@redhat.com \
    --cc=konrad.dybcio@linaro.org \
    --cc=linux-arm-kernel@lists.infradead.org \
    --cc=linux-arm-msm@vger.kernel.org \
    --cc=linux-doc@vger.kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-pm@vger.kernel.org \
    --cc=linux-trace-kernel@vger.kernel.org \
    --cc=linux@armlinux.org.uk \
    --cc=lukasz.luba@arm.com \
    --cc=mgorman@suse.de \
    --cc=mhiramat@kernel.org \
    --cc=mingo@redhat.com \
    --cc=peterz@infradead.org \
    --cc=rafael@kernel.org \
    --cc=rostedt@goodmis.org \
    --cc=rui.zhang@intel.com \
    --cc=sudeep.holla@arm.com \
    --cc=vincent.guittot@linaro.org \
    --cc=viresh.kumar@linaro.org \
    --cc=vschneid@redhat.com \
    --cc=will@kernel.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).