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;
Subject: Re: [PATCH v2] watchdog: Fix possible soft lockup warning at bootup
To:     Thomas Gleixner <tglx@linutronix.de>,
        Robert Richter <rrichter@marvell.com>
Cc:     Andrew Morton <akpm@linux-foundation.org>,
        Ingo Molnar <mingo@kernel.org>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
        Mike Rapoport <rppt@linux.ibm.com>,
        Kees Cook <keescook@chromium.org>,
        Catalin Marinas <catalin.marinas@arm.com>,
        Will Deacon <will@kernel.org>,
        Peter Zijlstra <peterz@infradead.org>
References: <87ftgffc9z.fsf@nanos.tec.linutronix.de>
From:   Waiman Long <longman@redhat.com>
Organization: Red Hat
Message-ID: <7168b0aa-4904-4246-0de5-3906df13b5c8@redhat.com>
Date:   Thu, 16 Jan 2020 14:13:49 -0500
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:60.0) Gecko/20100101
 Thunderbird/60.7.2
MIME-Version: 1.0
In-Reply-To: <87ftgffc9z.fsf@nanos.tec.linutronix.de>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: quoted-printable
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On 1/16/20 2:10 PM, Thomas Gleixner wrote:
> Waiman Long <longman@redhat.com> writes:
>
>> On 1/16/20 11:57 AM, Thomas Gleixner wrote:
>>>> So your theory the MONOTONIC clock runs differently/wrongly could
>>>> explain that (assuming this drives the sched clock). Though, I am
>>> No. sched_clock() is separate. It uses a raw timestamp (in your case
>>> from the ARM arch timer) and converts it to something which is close =
to
>>> proper time. So my assumption was based on the printout Waiman had:
>>>
>>>  [ 1... ] CPU.... watchdog_fn now  170000000
>>>  [ 25.. ] CPU.... watchdog_fn now 4170000000
>>>
>>> I assumed that now comes from ktime_get() or something like
>>> that. Waiman?
>> I printed out the now parameter of the=C2=A0 __hrtimer_run_queues() ca=
ll.
> Yes. That's clock MONOTONIC.
>
>> So from the timer perspective, it is losing time. For watchdog, the so=
ft
>> expiry time is 4s. The watchdog function won't be called until the
>> timer's time advances 4s or more. That corresponds to about 24s in
>> timestamp time for that particular class of systems.
> Right. And assumed that the firmware call is the culprit this has an
> explanation.
>
> Could you please take sched_clock() timestamps before and after the
> firmware call which kicks the secondary CPUs into life to verify that?
>
> They should sum up to the amount of time which gets lost accross
> smp_init().

Sure, I will do that after I get hold of the arm64 system that can
reproduce the issue. That system is currently used by another engineer.

Cheers,
Longman