linux-kernel - Re: Deadlock due to "cpuset: Make cpuset hotplug synchronous"

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

Message-ID: <20200406105522.c66p4vzzzylety5d@e107158-lin.cambridge.arm.com>
Date:   Mon, 6 Apr 2020 11:55:23 +0100
From:   Qais Yousef <qais.yousef@....com>
To:     Tejun Heo <tj@...nel.org>
Cc:     Qian Cai <cai@....pw>, Prateek Sood <prsood@...eaurora.org>,
        Li Zefan <lizefan@...wei.com>, cgroups@...r.kernel.org,
        LKML <linux-kernel@...r.kernel.org>,
        Peter Zijlstra <peterz@...radead.org>
Subject: Re: Deadlock due to "cpuset: Make cpuset hotplug synchronous"

On 04/03/20 10:55, Tejun Heo wrote:
> On Thu, Mar 26, 2020 at 10:15:32AM +0000, Qais Yousef wrote:
> > On 03/25/20 15:19, Tejun Heo wrote:
> > > On Wed, Mar 25, 2020 at 03:16:56PM -0400, Qian Cai wrote:
> > > > The linux-next commit a49e4629b5ed (“cpuset: Make cpuset hotplug synchronous”)
> > > > introduced real deadlocks with CPU hotplug as showed in the lockdep splat, since it is
> > > > now making a relation from cpu_hotplug_lock —> cgroup_mutex.
> > > 
> > > Prateek, can you please take a look? Given that the merge window is just around
> > > the corner, we might have to revert and retry later if it can't be resolved
> > > quickly.
> > 
> > I've ran cpuset_hotplug and cpuhotplug LTP tests using next-20200325 but
> > couldn't reproduce it.
> > 
> > Hopefully that can be fixed, but if you had to revert it, do you mind picking
> > this instead to fix the LTP issue I encountered before?
> > 
> > 	https://lore.kernel.org/lkml/20200211141554.24181-1-qais.yousef@arm.com/
> 
> So, I'd rather not, for now anyway. It isn't a real problem and I don't wanna
> add a wait vector there.

What would be the right approach to get a fix in then? We have been skipping
this test for a while and we'd like to enable it but this failure is a
blocking issue.

Android relies on cpuset and some devices use hotplug to manage thermal/power.
So it's an interesting combination to be able to test for us.

Thanks

--
Qais Yousef