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: <20210422120459.447350175@infradead.org> <20210422123308.196692074@infradead.org>
 <CABk29Ntop2nX+z1bV7giG8ToR_w3f_+GYGAw+hFQ6g9rCZunmw@mail.gmail.com>
 <YIZ6ZpkrMGQ9A9x2@hirez.programming.kicks-ass.net> <CABk29NvicqM_c2ssYnDrEy_FPsfD5GH38rB_XHooErALOabe5g@mail.gmail.com>
 <CAOY2WoyTq0AYMz+z=USxSpdcJqWMjkq-FDkuLN++brwhAkwTaQ@mail.gmail.com>
 <CABk29Nuz-FDCk23ajcr9gS4KD-wMpwyn=ASu+yuTTT445rwTvw@mail.gmail.com>
 <CAERHkrvU2Xzi5p9Dph3zZ7zkoYjSV1krK-UhqE7+Qb80FodEGg@mail.gmail.com> <5c289c5a-a120-a1d0-ca89-2724a1445fe8@linux.intel.com>
In-Reply-To: <5c289c5a-a120-a1d0-ca89-2724a1445fe8@linux.intel.com>
From:   Aubrey Li <aubrey.intel@gmail.com>
Date:   Wed, 28 Apr 2021 18:57:21 +0800
Message-ID: <CAERHkrsoCR7d3N2rhwKCeFDDBv4-S4HzD567mOaV_pngXn_Hkg@mail.gmail.com>
Subject: Re: [PATCH 04/19] sched: Prepare for Core-wide rq->lock
To:     Aubrey Li <aubrey.li@linux.intel.com>
Cc:     Josh Don <joshdon@google.com>, Don Hiatt <dhiatt@digitalocean.com>,
        Peter Zijlstra <peterz@infradead.org>,
        Joel Fernandes <joel@joelfernandes.org>,
        "Hyser,Chris" <chris.hyser@oracle.com>,
        Ingo Molnar <mingo@kernel.org>,
        Vincent Guittot <vincent.guittot@linaro.org>,
        Valentin Schneider <valentin.schneider@arm.com>,
        Mel Gorman <mgorman@suse.de>,
        linux-kernel <linux-kernel@vger.kernel.org>,
        Thomas Gleixner <tglx@linutronix.de>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Wed, Apr 28, 2021 at 2:05 PM Aubrey Li <aubrey.li@linux.intel.com> wrote:
>
> On 4/28/21 9:03 AM, Aubrey Li wrote:
> > On Wed, Apr 28, 2021 at 7:36 AM Josh Don <joshdon@google.com> wrote:
> >>
> >> On Tue, Apr 27, 2021 at 10:10 AM Don Hiatt <dhiatt@digitalocean.com> wrote:
> >>> Hi Josh and Peter,
> >>>
> >>> I've been running into soft lookups and hard lockups when running a script
> >>> that just cycles setting the cookie of a group of processes over and over again.
> >>>
> >>> Unfortunately the only way I can reproduce this is by setting the cookies
> >>> on qemu. I've tried sysbench, stress-ng but those seem to work just fine.
> >>>
> >>> I'm running Peter's branch and even tried the suggested changes here but
> >>> still see the same behavior. I enabled panic on hard lockup and here below
> >>> is a snippet of the log.
> >>>
> >>> Is there anything you'd like me to try or have any debugging you'd like me to
> >>> do? I'd certainly like to get to the bottom of this.
> >>
> >> Hi Don,
> >>
> >> I tried to repro using qemu, but did not generate a lockup. Could you
> >> provide more details on what your script is doing (or better yet,
> >> share the script directly)? I would have expected you to potentially
> >> hit a lockup if you were cycling sched_core being enabled and
> >> disabled, but it sounds like you are just recreating the cookie for a
> >> process group over and over?
> >>
> >
> > I saw something similar on a bare metal hardware. Also tried the suggested
> > patch here and no luck. Panic stack attached with
> > softlockup_all_cpu_backtrace=1.
> > (sorry, my system has 192 cpus and somehow putting 184 cpus offline causes
> > system hang without any message...)
>
> Can you please try the following change to see if the problem is gone on your side?
>

Please ignore this patch, as the change of double_rq_unlock() in
Peter's last patch
fixed the problem.

Thanks,
-Aubrey

>
> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> index f732642e3e09..1ef13b50dfcd 100644
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -493,14 +493,17 @@ void double_rq_lock(struct rq *rq1, struct rq *rq2)
>  {
>         lockdep_assert_irqs_disabled();
>
> -       if (rq1->cpu > rq2->cpu)
> -               swap(rq1, rq2);
> -
> -       raw_spin_rq_lock(rq1);
> -       if (__rq_lockp(rq1) == __rq_lockp(rq2))
> -               return;
> -
> -       raw_spin_rq_lock_nested(rq2, SINGLE_DEPTH_NESTING);
> +       if (__rq_lockp(rq1) == __rq_lockp(rq2)) {
> +               raw_spin_rq_lock(rq1);
> +       } else {
> +               if (__rq_lockp(rq1) < __rq_lockp(rq2)) {
> +                       raw_spin_rq_lock(rq1);
> +                       raw_spin_rq_lock_nested(rq2, SINGLE_DEPTH_NESTING);
> +               } else {
> +                       raw_spin_rq_lock(rq2);
> +                       raw_spin_rq_lock_nested(rq1, SINGLE_DEPTH_NESTING);
> +               }
> +       }
>  }
>  #endif
>