[PATCH 9/9] drm/vc4: Add an interface for capturing the GPU state after a hang.

[Daniel Vetter]

On Wed, Dec 02, 2015 at 11:35:16AM -0800, Eric Anholt wrote:
> Emil Velikov <emil.l.velikov at gmail.com> writes:
> 
> > On 1 December 2015 at 20:35, Eric Anholt <eric at anholt.net> wrote:
> >> This can be parsed with vc4-gpu-tools tools for trying to figure out
> >> what was going on.
> >>
> > I might be pushing my luck here ... have you thought about basing
> > (forking) vc4-gpu-tools of intel-gpu-tools ? I'd imagine that the
> > macros and helpers will come in handy, despite that some are quite
> > intel specific.
> >
> > On a related note - with the above project in place I'd imagine you
> > have (re)considered about having libdrm-vc4 ? Copying hunks around
> > might lead to interesting moments (as you have already noticed :-P)
> >
> > On that note I'll stop now with beating the libdrm drum :-)
> 
> The headers and code that I copy between the two userspace locations
> will go in libdrm when I have a kernel ABI, but vc4_drm.h can't go in
> until merging to the kernel, and there's not a whole lot of point
> without that.
> 
> Yes, I have thought about basing vc4-gpu-tools off of intel-gpu-tools.
> I've actually tried to build and use the kms testing stuff on vc4, and
> it was a total bust.  Someone needs to do a lot of work to make igt
> useful for non-intel.  If you'd like me to build my vc4 testing inside
> of igt, I'd someone to demo one of my tests building inside of igt, with
> the test runner working and none of the intel-specific tests reporting
> failure, and get me permission to just push code to that repository
> (It's hard enough getting piglit tests reviewed, vc4-specific tests and
> tools would never get review).

Daniel Stone claimed that this Just Works but evidently it doesn't.
There's some autoconfig fail where igt wants too much intel crap that just
doesn't build on arm. Iirc Daniel had some patches floating around for
that.

Wrt building tests itself I think we should require libdrm_$vendor
everywhere, since without that you can embedded a vendor specific testcase
into a shared one (e.g. to test tiled pageflipping). And if that's not
possible then either we'll end up in #ifdef hell or code sharing is really
hard, rendering a shared testsuite pretty useless.

Wrt commit access I'm ok with that, atm rule for igt is to send to
intel-gfx and push after 24h of no complaints. For purely vc4-specific
stuff we should probably create a vc4 subdir.

Wrt tests failing: I sometimes run the entire suite here on a radeon, and
at least in the past that resulted in a perfect 100% skip score. So should
work, too.

Cheers, Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch