[Bug 108456] userptr deadlock in Vulkan CTS: mmu_notifier vs workers vs struct_mutex

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=108456

--- Comment #16 from Mark Janes <mark.a.janes at intel.com> ---
(In reply to Emil Velikov from comment #14)
> Mark, is this a regression? How often does the failure occur?

I've been treating this as a regression, because it began in the past few
months.  Because it is intermittent, I have been unable to figure out what
triggered the hangs.

The failure occurs on about 30% of our CI runs.  Each run invokes vulkancts on
more than a dozen machines.  The per-machine failure rate is about 1-2%. 
Unfortunately, the impact is high: the machines are unusable until they are
rebooted.

It's possible that a kernel update or some other change (eg CTS version)
triggered these failures.  I'll do more testing on our 18.2 branches to rule
that out.  If the hang reproduces on 18.2, then there is no reason to block the
release.

> I'm wondering if we could drop this from the Mesa 18.3.0 blocker list.
> We're at RC5, as we normally aim to have around 4-6 release candidates.

Rafael has been working to eliminate userptr in Mesa, which should fix this
bug.  Ideally, we would backport his work to 18.3 for the release.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are the QA Contact for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/intel-gfx-bugs/attachments/20181129/aa3baffa/attachment-0001.html>