Re: NMI hardlock stacktrace deadlock [was Re: Linux 5.2-rc5]

[Chris Wilson]

Quoting Thomas Gleixner (2019-06-21 20:33:36)
> On Fri, 21 Jun 2019, Chris Wilson wrote:
> 
> > Quoting Thomas Gleixner (2019-06-21 16:30:52)
> > > Chris, do you have the actual NMI lockup detector splats somewhere?
> > 
> > Sorry, I'm having a hard time reproducing this at will now. The test
> > case depends on the right timing of the wrong event to cause the GPU to
> > hang.
> > 
> > From memory, I got the
> >       "Watchdog detected hard LOCKUP on cpu foo"
> > followed by the register dump and then nothing. At which point I had to
> > power cycle the machine.
> 
> Hmm. Do you have a serial log of that incident?

I use netconsole. I think Tomi has a serial console for most things
available, but not permanently hooked up. And I didn't have it in a tee
as it was late, with the lockup an annoyance to the bug I was trying
to solve. I'll keep trying to recreate that bug as once I do have that
recipe, it should be possible to bisect. I can check with Tomi on Monday
if he can pull a machine out of the farm and see how it locked up.
-Chris