linux-kernel - Re: [Patch v4 4/6] sched/fair: update cpu

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:   Thu, 31 Oct 2019 16:48:34 +0100
From:   Vincent Guittot <vincent.guittot@...aro.org>
To:     Dietmar Eggemann <dietmar.eggemann@....com>
Cc:     Qais Yousef <qais.yousef@....com>,
        Peter Zijlstra <peterz@...radead.org>,
        Thara Gopinath <thara.gopinath@...aro.org>,
        Ingo Molnar <mingo@...hat.com>,
        Ionela Voinescu <ionela.voinescu@....com>,
        Zhang Rui <rui.zhang@...el.com>,
        Eduardo Valentin <edubezval@...il.com>,
        Quentin Perret <qperret@...gle.com>,
        linux-kernel <linux-kernel@...r.kernel.org>,
        Amit Kachhap <amit.kachhap@...il.com>,
        Javi Merino <javi.merino@...nel.org>,
        Daniel Lezcano <daniel.lezcano@...aro.org>
Subject: Re: [Patch v4 4/6] sched/fair: update cpu_capcity to reflect thermal pressure

On Thu, 31 Oct 2019 at 16:38, Dietmar Eggemann <dietmar.eggemann@....com> wrote:
>
> On 31.10.19 11:53, Qais Yousef wrote:
> > On 10/28/19 16:30, Peter Zijlstra wrote:
> >> On Wed, Oct 23, 2019 at 01:28:40PM +0100, Qais Yousef wrote:
> >>> On 10/22/19 16:34, Thara Gopinath wrote:
> >>>> cpu_capacity relflects the maximum available capacity of a cpu. Thermal
> >>>> pressure on a cpu means this maximum available capacity is reduced. This
> >>>> patch reduces the average thermal pressure for a cpu from its maximum
> >>>> available capacity so that cpu_capacity reflects the actual
> >>>> available capacity.
> >>>>
> >>>> Signed-off-by: Thara Gopinath <thara.gopinath@...aro.org>
> >>>> ---
> >>>>  kernel/sched/fair.c | 1 +
> >>>>  1 file changed, 1 insertion(+)
> >>>>
> >>>> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> >>>> index 4f9c2cb..be3e802 100644
> >>>> --- a/kernel/sched/fair.c
> >>>> +++ b/kernel/sched/fair.c
> >>>> @@ -7727,6 +7727,7 @@ static unsigned long scale_rt_capacity(struct sched_domain *sd, int cpu)
> >>>>
> >>>>    used = READ_ONCE(rq->avg_rt.util_avg);
> >>>>    used += READ_ONCE(rq->avg_dl.util_avg);
> >>>> +  used += READ_ONCE(rq->avg_thermal.load_avg);
> >>>
> >>> Maybe a naive question - but can we add util_avg with load_avg without
> >>> a conversion? I thought the 2 signals have different properties.
> >>
> >> Changelog of patch #1 explains, it's in that dense blob of text.
> >>
> >> But yes, you're quite right that that wants a comment here.
> >
> > Thanks for the pointer! A comment would be nice indeed.
> >
> > To make sure I got this correctly - it's because avg_thermal.load_avg
> > represents delta_capacity which is already a 'converted' form of load. So this
> > makes avg_thermal.load_avg a util_avg really. Correct?
> >
> > If I managed to get it right somehow. It'd be nice if we can do inverse
> > conversion on delta_capacity so that avg_thermal.{load_avg, util_avg} meaning
> > is consistent across the board. But I don't feel strongly about it if this gets
> > documented properly.
>
> So why can't we use rq->avg_thermal.util_avg here? Since capacity is
> closer to util than to load?
>
> Is it because you want to use the influence of ___update_load_sum(...,
> unsigned long load eq. per-cpu delta_capacity in your signal?
>
> Why not call it this way then?

util_avg tracks a binary state with 2 fixed weights: running(1024)  vs
not running (0)
In the case of thermal pressure, we want to track how much pressure is
put on the CPU: capping to half the max frequency is not the same as
capping only 10%
load_avg is not boolean but you set the weight  you want to apply and
this weight reflects the amount of pressure.

>
> diff --git a/kernel/sched/pelt.c b/kernel/sched/pelt.c
> index 38210691c615..d3035457483f 100644
> --- a/kernel/sched/pelt.c
> +++ b/kernel/sched/pelt.c
> @@ -357,9 +357,9 @@ int update_thermal_load_avg(u64 now, struct rq *rq,
> u64 capacity)
>  {
>         if (___update_load_sum(now, &rq->avg_thermal,
>                                capacity,
> -                              capacity,
> -                              capacity)) {
> -               ___update_load_avg(&rq->avg_thermal, 1, 1);
> +                              0,
> +                              0)) {
> +               ___update_load_avg(&rq->avg_thermal, 1, 0);
>                 return 1;
>         }