The issue can be reproduced with the media-driver on ICL LP by following the steps described in https://github.com/intel/media-driver/issues/267.
It can be fixed with https://github.com/intel/media-driver/pull/271, but this media-driver patch depends on the i915 "Per context dynamic (sub)slice power-gating" series -- https://patchwork.freedesktop.org/series/48194/.
Assigned to Tvrtko.
Priority changed to High as the issue hangs GPU.
Tony, could you help me creating a simplest possible batch buffer which hangs on ICL if dynamic SSEU is not working or not present? I need to add a more proper functional test to the existing IGT (https://patchwork.freedesktop.org/patch/248544/) since currently all we have is the register read-back.
Tvrtko, the VME sampler cannot be accessed in a simple batch. You need to prepare some states, surfaces, and then launch a kernel to do something with VME. It's similar to https://cgit.freedesktop.org/xorg/app/intel-gpu-tools/tree/lib/media_fill.c, but more efforts are required. Unfortunately, I do not have something which is already working to share with you.
The simplest way to reproduce the issue to use the media driver to encode a stream.
Would you or someone from your team be able to estimate the amount of work needed, and then maybe add this test case as follow-up?
We could also look at the feasibility of integrating any test suites you have into the i915 continuous integration testing. That would ensure i915 changes do not break the media driver ever.
Tvrtko, I think we now have a testcase for this, is that right?
Correct, as far as I understand it is all mostly there apart from some process bits and a final rebase pending.