Re: [PATCH 5.15 00/92] 5.15.126-rc1 review

[Joel Fernandes]

On Wed, Aug 9, 2023 at 2:35 PM Joel Fernandes <joel@xxxxxxxxxxxxxxxxx> wrote:
>
> On Wed, Aug 9, 2023 at 12:18 PM Guenter Roeck <linux@xxxxxxxxxxxx> wrote:
> >
> > On 8/9/23 06:53, Joel Fernandes wrote:
> > > On Wed, Aug 09, 2023 at 12:40:36PM +0200, Greg Kroah-Hartman wrote:
> > >> This is the start of the stable review cycle for the 5.15.126 release.
> > >> There are 92 patches in this series, all will be posted as a response
> > >> to this one.  If anyone has any issues with these being applied, please
> > >> let me know.
> > >>
> > >> Responses should be made by Fri, 11 Aug 2023 10:36:10 +0000.
> > >> Anything received after that time might be too late.
> > >>
> > >> The whole patch series can be found in one patch at:
> > >>      https://www.kernel.org/pub/linux/kernel/v5.x/stable-review/patch-5.15.126-rc1.gz
> > >> or in the git tree and branch at:
> > >>      git://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable-rc.git linux-5.15.y
> > >> and the diffstat can be found below.
> > >
> > > Not necesscarily new with 5.15 stable but 3 of the 19 rcutorture scenarios
> > > hang with this -rc: TREE04, TREE07, TASKS03.
> > >
> > > 5.15 has a known stop machine issue where it hangs after 1.5 hours with cpu
> > > hotplug rcutorture testing. Me and tglx are continuing to debug this. The
> > > issue does not show up on anything but 5.15 stable kernels and neither on
> > > mainline.
> > >
> >
> > Do you by any have a crash pattern that we could possibly use to find the crash
> > in ChromeOS crash logs ? No idea if that would help, but it could provide some
> > additional data points.
>
> The pattern shows as a hard hang, the system is unresponsive and all CPUs
> are stuck in stop_machine. Sometimes it recovers on its own from the
> hang and then RCU immediately gives stall warnings. It takes 1.5 hour
> to reproduce and sometimes never happens for several hours.
>
> It appears related to CPU hotplug since gdb showed me most of the CPUs
> are spinning in multi_cpu_stop() / stop machine after the hang.
>

Adding to this, it appears one of the CPUs is constantly firing and
reprogramming hrtimer events for some reason every few 100
microseconds (I see this in gdb). My debug angle right now is to
figure out why it does that but collecting a trace is hard as it
appears even trace collection may not be happening once hung and the
only traces I am getting are the ones after the hang recovers, not
during the hang.  I am also trying to see if multi_cpu_stop() can
panic the kernel if it sits there too long.

 - Joel