linux-kernel - Re: [PATCH 4/4] cpu/hotplug: Fix CPU down rollback

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <20210121154314.GA439562@e120877-lin.cambridge.arm.com>
Date:   Thu, 21 Jan 2021 15:43:15 +0000
From:   Vincent Donnefort <vincent.donnefort@....com>
To:     Peter Zijlstra <peterz@...radead.org>
Cc:     tglx@...utronix.de, linux-kernel@...r.kernel.org,
        valentin.schneider@....com
Subject: Re: [PATCH 4/4] cpu/hotplug: Fix CPU down rollback

On Thu, Jan 21, 2021 at 03:57:03PM +0100, Peter Zijlstra wrote:
> On Mon, Jan 11, 2021 at 05:10:47PM +0000, vincent.donnefort@....com wrote:
> > From: Vincent Donnefort <vincent.donnefort@....com>
> > 
> > After the AP brought itself down to CPUHP_TEARDOWN_CPU, the BP will finish
> > the job. The steps left are as followed:
> > 
> >    +--------------------+
> >    | CPUHP_TEARDOWN_CPU |  -> If fails state is CPUHP_TEARDOWN_CPU
> >    +--------------------+
> >    |   ATOMIC STATES    |  -> Cannot Fail
> >    +--------------------+
> >    |  CPUHP_BRINGUP_CPU |  -> Cannot fail
> >    +--------------------+
> >    |        ...         |
> >    |        ...         |  -> Can fail and rollback
> 
> These are the PREPARE/DEAD states, right? It would be _really_ daft for
> a DEAD notifier to fail. But yeah, I suppose that if it does, it will
> indeed end up in CPUHP_AP_ONLINE_IDLE.
> 
> Do we want to WARN when a DEAD notifier fails?
> 
> 

Indeed, I couldn't find a dead callback which can return an error. So I
suppose we could go for another strategy here, that would be to not allow
failure for the dead states (i.e states < BRINGUP_CPU when hotunplug). In
that case, the fail interface should probably disallow selecting those
states and a WARN here would be good.

-- 
Vincent