[PATCH 3/4] drm/radeon: consolidate uvd/vce initialization, resume and suspend.

[Daniel Vetter]

On Sat, Mar 19, 2016 at 10:41:59AM +0100, Daniel Vetter wrote:
> On Fri, Mar 18, 2016 at 09:16:08AM +0100, Daniel Vetter wrote:
> > On Fri, Mar 18, 2016 at 01:00:26PM +0900, Michel Dänzer wrote:
> > > On 17.03.2016 16:36, Daniel Vetter wrote:
> > > > On Thu, Mar 17, 2016 at 06:41:14AM +1000, Dave Airlie wrote:
> > > >> Just an aside,
> > > >>
> > > >> So is there no way to do hibernate with these blocks?
> > > >>
> > > >> Like can you not cleanly shut them down without doing a power cycle.
> > > >>
> > > >> I have to say UVD is a real pain in the ass from a stability pov, I'd
> > > >> kinda wished I'd enforced AMD creating something like intel-gpu-tools
> > > >> and having tests to make sure GPU reset etc stayed working before
> > > >> merging it.
> > > > 
> > > > igt already supports running on any kind of drm device, and it has a bunch
> > > > of vc4 specific testcases on top. If anyone finds offence in the "intel"
> > > > part, we can rename it to igt gpu tools/tests ;-)
> > > 
> > > Any tips for running the tests on non-Intel GPUs? I tried piglit igt.py,
> > > but it was generating tens of thousands of failures from tests which
> > > look Intel specific.
> > 
> > Yeah Chris again broke the SKIP logic in gem_concurrent_blt/all testcases.
> > Just explicitly exclude those with -x gem_concurrent. The problem is that
> > hw/kernel feature checks aren't properlty encapsulated in the right
> > igt_fixture or igt_subtest blocks, so it falls over. Specifically the
> > access_mode->require() test is only protetected by
> > igt_only_list_subtests(), which is the wrong way to do it.
> > 
> > Adding Chris.
> 
> Ok, fixed pushed now using the just added igt_subtest_group blocks. Please
> scream when anything else falls apart.

Ok, just realized that Chris' combinatorial stress-test crusade reached a
few 100k tests now ;-) So just run with -x gem_ to get rid of all the i915
gem stuff. If we get more along the lines of vc4, we probably need to give
them all a i915_ prefix. Same for the i915 ioctl wrappers probably.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch