Re: [PATCH v3 10/10] xen/arm: Enable errata for secondary CPU on hotplug after the boot

From: Mirela Simonovic <mirela.simonovic@aggios.com>
To: Dario Faggioli <dfaggioli@suse.com>
Cc: "Edgar E. Iglesias" <edgar.iglesias@xilinx.com>,
	Julien Grall <julien.grall@arm.com>,
	Stefano Stabellini <sstabellini@kernel.org>,
	Davorin Mista <dm@aggios.com>,
	Xen Devel <xen-devel@lists.xen.org>
Subject: Re: [PATCH v3 10/10] xen/arm: Enable errata for secondary CPU on hotplug after the boot
Date: Fri, 11 May 2018 12:41:19 +0200	[thread overview]
Message-ID: <CAKPH-NiZtBwioY1hfzx2PSgzKB6eQxjw0hXUzLg94+u+OVpL+Q@mail.gmail.com> (raw)
In-Reply-To: <6c0ba4fadcdb650e3a8651f13781da9caaae251e.camel@suse.com>

Hi Dario,

On Thu, May 10, 2018 at 6:24 PM, Dario Faggioli <dfaggioli@suse.com> wrote:
> On Thu, 2018-05-10 at 17:49 +0200, Mirela Simonovic wrote:
>> Regardless of the fact that the notifier returns an error or not, I
>> believe it would be good and safe to set priority and document that
>> priority zero would cause racing issue in the scenario I debugged
>> today. I'm pretty sure that this discussion would be forgotten soon
>> and it really should be documented in code/comment.
>>
> I may very well be missing or misunderstanding something, but I
> continue to think that the problem here is that CPU_STARTING can't,
> right now, fail, while you need it to be able to.
>
> If that is the case, giving different priorities to the notifier, is
> not a solution.
>

Let me try to clarify. The assumption is that the starting CPU can
fail. Additional requirement set by Julien is that panic or BUG_ON is
not acceptable. There are 2 ways to deal with this scenario:

1) Ignore and report the error, and let the erroneous CPU become online.
This cannot be done without changing the logic in either scheduler or
notify_cpu_starting(), or I least I don't see how would that be done.
In previous email when I said "escalating this to who knows where" I
did not refer to error escalation but the escalation of the scope of
this series.

2) Stop the erroneous CPU.
Taking this approach requires setting the priority for the notifier.
The key point is that notify_cpu_starting() and scheduler do not have
to be changed. If errata notifier has higher priority than scheduler's
notifier in the case of an error the CPU will not return into
notify_cpu_starting() and it will never execute BUG_ON because it will
be stopped. The rest of the system will continue to function without
that CPU.

> Regards,
> Dario
> --
> <<This happens because I choose it to happen!>> (Raistlin Majere)
> -----------------------------------------------------------------
> Dario Faggioli, Ph.D, http://about.me/dario.faggioli
> Software Engineer @ SUSE https://www.suse.com/

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel