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;
From:   Valentin Schneider <vschneid@redhat.com>
To:     paulmck@kernel.org
Cc:     linux-kernel@vger.kernel.org, mingo@redhat.com,
        peterz@infradead.org, juri.lelli@redhat.com,
        vincent.guittot@linaro.org, dietmar.eggemann@arm.com,
        rostedt@goodmis.org, bsegall@google.com, mgorman@suse.de,
        bristot@redhat.com
Subject: Re: "Dying CPU not properly vacated" splat
In-Reply-To: <20220705172342.GB1790663@paulmck-ThinkPad-P17-Gen-1>
References: <20220421193821.GA173010@paulmck-ThinkPad-P17-Gen-1>
 <xhsmh4k2h9m26.mognet@vschneid.remote.csb>
 <20220425173320.GX4285@paulmck-ThinkPad-P17-Gen-1>
 <xhsmh1qxkakof.mognet@vschneid.remote.csb>
 <20220426000328.GY4285@paulmck-ThinkPad-P17-Gen-1>
 <xhsmhy1zr99zt.mognet@vschneid.remote.csb>
 <20220426162445.GG4285@paulmck-ThinkPad-P17-Gen-1>
 <20220622195856.GA1694215@paulmck-ThinkPad-P17-Gen-1>
 <xhsmhzghof1eo.mognet@vschneid.remote.csb>
 <20220705172342.GB1790663@paulmck-ThinkPad-P17-Gen-1>
Date:   Tue, 02 Aug 2022 10:30:02 +0100
Message-ID: <xhsmhsfmfc7s5.mognet@vschneid.remote.csb>
MIME-Version: 1.0
Content-Type: text/plain
Precedence: bulk

On 05/07/22 10:23, Paul E. McKenney wrote:
> The second of these occurred near shutdown, but the first was quite some
> time before shutdown.  In case that makes a difference.
>
> I have not seen this since.
>
> Any other diagnostics I should add?
>

Sorry, I let this get buried to the bottom of my inbox :(

I've had another look at rcutorture.c but just like for
rcu_torture_reader(), I don't see any obvious culprit (no
kthread_set_per_cpu() usage)).

One thing I think would help is a scheduling trace (say sched_switch,
sched_wakeup and cpuhp*, combined with ftrace_dump_on_oops + panic_on_warn
?) - that should at least tell us if the issue is in the wakeup placement
(if the task gets placed on a dying CPU *after* CPUHP_AP_ACTIVE), or in the
balance_push() mechanism (the task was *already* on the CPU when it started
dying and never moved away).

Neither make sense to me, but it has to be somewhere in there...

>                                                       Thanx, Paul