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: <4781d250-9a29-cef3-268d-7d83c98bf16a@gmail.com> <87wo4lekm5.fsf@kurt>
In-Reply-To: <87wo4lekm5.fsf@kurt>
From:   Vladimir Oltean <olteanv@gmail.com>
Date:   Fri, 5 Jun 2020 16:19:59 +0300
Message-ID: <CA+h21hqbKasMAuHL+B-2Gb-YQ3QGF+_pWGCxr8LTcusjvuqFeg@mail.gmail.com>
Subject: Re: stress-ng --hrtimers hangs system
To:     Kurt Kanzenbach <kurt.kanzenbach@linutronix.de>
Cc:     linux-rt-users@vger.kernel.org,
        lkml <linux-kernel@vger.kernel.org>, rcu@vger.kernel.org,
        Colin King <colin.king@canonical.com>
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Hi Kurt,

On Fri, 5 Jun 2020 at 15:57, Kurt Kanzenbach
<kurt.kanzenbach@linutronix.de> wrote:
>
> Hi Vladimir,
>
> On Fri Jun 05 2020, Vladimir Oltean wrote:
> > Hi,
> >
> > I was testing stress-ng on an ARM64 box and I found that it can be killed instantaneously with a --hrtimers 1 test:
> > https://github.com/ColinIanKing/stress-ng/blob/master/stress-hrtimers.c
> > The console shell locks up immediately after starting the process, and I get this rcu_preempt splat after 21 seconds,
> > letting me know that the grace-periods kernel thread could not run:
>
> interesting. Just tested this on an ARM64 box with v5.6-rt and the
> stress-ng hrtimer test works fine. No lockups, cyclictest results are
> looking good. So maybe this is v5.7 related.
>
> Thanks,
> Kurt

This is not actually v5.7 related. I could also reproduce it on a
downstream 5.4 kernel which is how I originally saw it.

Thanks,
-Vladimir