Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <20210117123104.27589-1-benbjiang@tencent.com> <CAKfTPtCPUnhiNF0SxK-=RTaq+h1D0tK-OfRsubb38V23KFEB_w@mail.gmail.com>
In-Reply-To: <CAKfTPtCPUnhiNF0SxK-=RTaq+h1D0tK-OfRsubb38V23KFEB_w@mail.gmail.com>
From:   Jiang Biao <benbjiang@gmail.com>
Date:   Mon, 18 Jan 2021 22:11:01 +0800
Message-ID: <CAPJCdBnHHk40uBtrOisp=hY=5K3OLeUwpkPmgUURVf2DvDmY_Q@mail.gmail.com>
Subject: Re: [PATCH] sched/fair: add protection for delta of wait time
To:     Vincent Guittot <vincent.guittot@linaro.org>
Cc:     Ingo Molnar <mingo@redhat.com>,
        Peter Zijlstra <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>,
        Daniel Bristot de Oliveira <bristot@redhat.com>,
        linux-kernel <linux-kernel@vger.kernel.org>,
        Jiang Biao <benbjiang@tencent.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

Hi, Vincent

On Mon, 18 Jan 2021 at 15:56, Vincent Guittot
<vincent.guittot@linaro.org> wrote:
>
> On Sun, 17 Jan 2021 at 13:31, Jiang Biao <benbjiang@gmail.com> wrote:
> >
> > From: Jiang Biao <benbjiang@tencent.com>
> >
> > delta in update_stats_wait_end() might be negative, which would
> > make following statistics go wrong.
>
> Could you describe the use case that generates a negative delta ?
>
> rq_clock is always increasing so this should not lead to a negative
> value even if update_stats_wait_end/start are not called in the right
> order,
Yes, indeed.

> This situation could happen after a migration if we forgot to call
> update_stats_wait_start
The migration case was what I worried about, but no regular use case
comes into my mind. :)
As an extreme case, would it be a problem if we disable/re-enable
sched_schedstats during migration?

static inline void
update_stats_wait_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
        u64 wait_start, prev_wait_start;

        if (!schedstat_enabled()) // disable during migration
                return; // return here, and skip updating wait_start
...
}

static inline void
update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se)
{
        struct task_struct *p;
        u64 delta;

        if (!schedstat_enabled())  // re-enable again
                return;

        /*
         * When the sched_schedstat changes from 0 to 1, some sched se
         * maybe already in the runqueue, the se->statistics.wait_start
         * will be 0.So it will let the delta wrong. We need to avoid this
         * scenario.
         */
        if (unlikely(!schedstat_val(se->statistics.wait_start)))
                return;
         //stale wait_start which might be bigger than rq_clock would
be used. -)
        delta = rq_clock(rq_of(cfs_rq)) -
schedstat_val(se->statistics.wait_start);
...

Thanks a lot.
Regards,
Jiang

}
>
> >
> > Add protection for delta of wait time, like what have been done in
> > update_stats_enqueue_sleeper() for deltas of sleep/block time.
> >
> > Signed-off-by: Jiang Biao <benbjiang@tencent.com>
> > ---
> >  kernel/sched/fair.c | 3 +++
> >  1 file changed, 3 insertions(+)
> >
> > diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> > index c0374c1152e0..ac950ac950bc 100644
> > --- a/kernel/sched/fair.c
> > +++ b/kernel/sched/fair.c
> > @@ -917,6 +917,9 @@ update_stats_wait_end(struct cfs_rq *cfs_rq, struct sched_entity *se)
> >
> >         delta = rq_clock(rq_of(cfs_rq)) - schedstat_val(se->statistics.wait_start);
> >
> > +       if ((s64)delta < 0)
> > +               delta = 0;
> > +
> >         if (entity_is_task(se)) {
> >                 p = task_of(se);
> >                 if (task_on_rq_migrating(p)) {
> > --
> > 2.21.0
> >