Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <20220311161406.23497-1-vincent.guittot@linaro.org>
 <20220311161406.23497-6-vincent.guittot@linaro.org> <CADjb_WTYPqenF_BhuiDyLduxpaHWCigg-jxAE3FYKTNkWvVz=Q@mail.gmail.com>
In-Reply-To: <CADjb_WTYPqenF_BhuiDyLduxpaHWCigg-jxAE3FYKTNkWvVz=Q@mail.gmail.com>
From:   Vincent Guittot <vincent.guittot@linaro.org>
Date:   Wed, 4 May 2022 14:39:10 +0200
Message-ID: <CAKfTPtCUxZVHOrj+1YTZ+z__HC53by=8X6Bxt1NeGJc2prqdzw@mail.gmail.com>
Subject: Re: [RFC 5/6] sched/fair: Take into account latency nice at wakeup
To:     Chen Yu <yu.chen.surf@gmail.com>
Cc:     Ingo Molnar <mingo@redhat.com>,
        "Peter Zijlstra (Intel)" <peterz@infradead.org>,
        Juri Lelli <juri.lelli@redhat.com>,
        Dietmar Eggemann <dietmar.eggemann@arm.com>,
        Steven Rostedt <rostedt@goodmis.org>,
        Ben Segall <bsegall@google.com>, Mel Gorman <mgorman@suse.de>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        parth@linux.ibm.com, qais.yousef@arm.com, chris.hyser@oracle.com,
        pkondeti@codeaurora.org,
        Valentin Schneider <valentin.schneider@arm.com>,
        patrick.bellasi@matbug.net, David.Laight@aculab.com,
        Paul Turner <pjt@google.com>, Pavel Machek <pavel@ucw.cz>,
        tj@kernel.org, dhaval.giani@oracle.com, qperret@google.com,
        Tim Chen <tim.c.chen@linux.intel.com>,
        Chen Yu <yu.c.chen@intel.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Wed, 4 May 2022 at 13:15, Chen Yu <yu.chen.surf@gmail.com> wrote:
>
> On Sat, Mar 12, 2022 at 7:11 AM Vincent Guittot
> <vincent.guittot@linaro.org> wrote:
> >
> > Take into account the nice latency priority of a thread when deciding to
> > preempt the current running thread. We don't want to provide more CPU
> > bandwidth to a thread but reorder the scheduling to run latency sensitive
> > task first whenever possible.
> >
> ---------->8-------------------
> >  #endif /* CONFIG_SMP */
> >
> > +static long wakeup_latency_gran(int latency_weight)
> > +{
> > +       long thresh = sysctl_sched_latency;
> If I understood correctly, this is to consider the latency weight and
> 'shrink/expand'
> current task's time slice thus to facilitate preemption. And may I
> know why don't we use
> __sched_period() but to use sysctl_sched_latency directly? Is it
> possible the rq has
> more than 8(sched_nr_latency) tasks thus the period is longer than
> sysctl_sched_latency?

Main reason is to be aligned with place_entity which also uses
sysctl_sched_latency to cap entity's vruntime to be higher than
min_vruntime-sysctl_sched_latency

>
> Thanks,
> Chenyu
> > +
> > +       if (!latency_weight)
> > +               return 0;
> > +
> > +       if (sched_feat(GENTLE_FAIR_SLEEPERS))
> > +               thresh >>= 1;
> > +
> > +       /*
> > +        * Clamp the delta to stay in the scheduler period range
> > +        * [-sysctl_sched_latency:sysctl_sched_latency]
> > +        */
> > +       latency_weight = clamp_t(long, latency_weight,
> > +                               -1 * NICE_LATENCY_WEIGHT_MAX,
> > +                               NICE_LATENCY_WEIGHT_MAX);
> > +
> > +       return (thresh * latency_weight) >> NICE_LATENCY_SHIFT;
> > +}
> > +