Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Mon, 7 Jan 2019 18:56:45 +0100
From:   Oleg Nesterov <oleg@redhat.com>
To:     Qian Cai <cai@lca.pw>
Cc:     Peter Zijlstra <peterz@infradead.org>,
        Ingo Molnar <mingo@redhat.com>,
        linux kernel <linux-kernel@vger.kernel.org>,
        gkohli@codeaurora.org
Subject: Re: kernel BUG at kernel/sched/core.c:3490!
Message-ID: <20190107175644.GB7636@redhat.com>
References: <ba9c5ad8-d78c-904d-0430-a67d8e3ea729@lca.pw>
 <20190107135215.GG14122@hirez.programming.kicks-ass.net>
 <46089f1c-ad72-c96c-2f35-c2f60e726462@lca.pw>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <46089f1c-ad72-c96c-2f35-c2f60e726462@lca.pw>
User-Agent: Mutt/1.5.24 (2015-08-30)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On 01/07, Qian Cai wrote:
>
>
> On 1/7/19 8:52 AM, Peter Zijlstra wrote:
> > On Tue, Jan 01, 2019 at 12:44:35AM -0500, Qian Cai wrote:
> >> Running some mmap() workloads to put the system on low memory situation with
> >> swapping and OOM, and then it trigger this BUG(),
> >>
> >> void __noreturn do_task_dead(void)
> >> {
> >>         /* Causes final put_task_struct in finish_task_switch(): */
> >>         set_special_state(TASK_DEAD);
> >>
> >>         /* Tell freezer to ignore us: */
> >>         current->flags |= PF_NOFREEZE;
> >>
> >>         __schedule(false);
> >>         BUG();
> >>
> >>         /* Avoid "noreturn function does return" - but don't continue if BUG()
> >> is a NOP: */
> >>         for (;;)
> >>                 cpu_relax();
> >> }
> >
> > This would mean that we somehow loose the TASK_DEAD state before hitting
> > schedule(), but that is something that should be avoided by
> > set_special_state(), which is supposed to serialize against concurrent
> > wake-ups.

or may be pick_next_task() somehow returns the deactivated TASK_DEAD task?

> > How readily does this reproduce?
>
> Running LTP oom01 [1] triggered it at least once in five attempts every time so
> far on v4.20+. Have not tried much on v5.0-rc1 yet.

Can you add

	pr_crit("XXX: %ld %d\n", current->state, current->on_rq);

before that BUG() and reproduce?

Oleg.