[Bug 111606] [CI][RESUME] igt at gem_exec_schedule@preempt-* - incomplete

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=111606

--- Comment #4 from Matthew Brost <matthew.brost at intel.com> ---
Took a quick look. This test is inserting a spin batch at NOISE level and write
batch at a LOW level on the same engine. On a different engine a write batch is
inserted at a HIGH level. This write should boost the LOW write batch past the
NOISE spin batch and preempt it. Once both writes complete the spin batch is
released and the test should complete. Since the test is killed by a SIGQUIT,
I'm guessing the spin batch is never preempted and the test just hangs. I'd
suggest adding some tracing to the driver to see if LOW write batch priority is
bumped by the HIGH write batch and if the preemption code path is triggered.

I'm specifically referring to preempt-other-blt.

It looks like other preemption sections are also failing, specifically
preempt-queue-* sections. These IMO are more complicated as they contain more
exec buf IOCTLs than preempt-other-blt so I'd focus on debugging
preempt-other-blt. All the failures are likely the same issue so if
preempt-other-blt gets fixed, I'd imagine all the other failing sections will
also get fixed.

I need to start running some of the new GuC code I've been developing on TGL
later this week so I plan on grabbing a RIL TGL system. Once I have a TGL
system, I'll see if I can recreate this failure and dig into it a bit.

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
You are the assignee for the bug.
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/intel-gfx-bugs/attachments/20190917/29ee0380/attachment.html>