Created attachment 120448 [details]
OpenGL version string: 3.0 Mesa 11.2.0-devel (git-83e8e07)
OpenGL renderer string: Mesa DRI Intel(R) HD Graphics 530 (Skylake GT2)
00:02.0 VGA compatible controller : Intel Corporation Sky Lake Integrated Graphics [8086:1912] (rev 06)
Intel(R) Core(TM) i5-6600K CPU @ 3.50GHz
Mesa built with: --enable-debug
When running all the layout_binding tests of the OpenGL ES 3.1 CTS, execution stops after 3 or 4 test with:
"intel_do_flush_locked failed: Input/output error"
Created attachment 120449 [details]
dmesg after hang
If ARB_compute_shader are not override enabled, the hang is not reproducible.
To clarify above comment. The hang is related to compute shader funktionality. It has nothing to do with how compute shaders are enabled.
I wonder if this is related to the GPU hang that Jason found a while back, where out of bounds values in an array (of say, samplers or images) can cause us to access an invalid surface...or a surface with a message that's invalid for that surface (i.e. try to use the sampler on a RAW surface).
Cc'ing Jason in case.
I am not sure, but maybe BUG 93407 is related to this. It is a SKL only failing test: ES31-CTS.compute_shader.resources-texture
(In reply to Kenneth Graunke from comment #4)
> I wonder if this is related to the GPU hang that Jason found a while back,
> where out of bounds values in an array (of say, samplers or images) can
> cause us to access an invalid surface...or a surface with a message that's
> invalid for that surface (i.e. try to use the sampler on a RAW surface).
Going off nothing but the test name and bug report, I think that's likely.
(In reply to Marta Löfstedt from comment #5)
> I am not sure, but maybe BUG 93407 is related to this. It is a SKL only
> failing test: ES31-CTS.compute_shader.resources-texture
Again, I haven't actually looked at the test, but based on names, that seems like a reasonable guess.
If it is the bug Ken is referring to above, then it comes from doing an out-of-bounds array access on an array of textures or buffers. We need to add some code to the glsl compiler that clamps array indices for arrays of external resources to range [0, size).
I ran through all layout binding cases one by one.
It is only:
that cause GPU HANG. All other layout binding tests actually pass.
I can not reproduce this issue with Mesa: git@ad20be1f30ef5 and Curros patch from BUG 92589.
*** This bug has been marked as a duplicate of bug 92589 ***