[Mesa-dev] [PATCH] st/dri: wait for thread to finish before unbinding context

[Michel Dänzer]

On 20/03/17 06:40 AM, Timothy Arceri wrote:
> On 17/03/17 18:38, Michel Dänzer wrote:
>> On 17/03/17 09:02 AM, Timothy Arceri wrote:
>>> Fixes a bunch of piglit crashes that hit an assert() when trying
>>> to delete the framebuffer. The assert() was triggered because
>>> WinSysDrawBuffer was set to NULL before glDeleteFramebuffers()
>>> was called.
>>
>> Tested-by: Michel Dänzer <michel.daenzer at amd.com>
>>
>>
>> With this patch, the remaining regressions with mesa_glthread=true vs
>> without for piglit gpu.py are 3 crashes and 14 failures for me. Are you
>> seeing the same?
>>
> 
> I'm seeing 3 crashes 9 failures but some of those are the timeing tests
> which are always intermittent.

As discussed on IRC: The

 glx at glx_arb_sync_control@timing *

tests produce unstable results, apparently due to a test bug:

 <ickle> it doesn't align the first wait to a vblank

However, with mesa_glthread=true these tests consistently fail for me,
which is probably a glthread bug.

Possibly related, glx at glx-swap-event_async also fails.


I'm also still seeing crashes with

spec@!opengl 1.1 at fdo23670-drawpix_stencil
spec at arb_clip_control@arb_clip_control-viewport

and failures with

shaders at glsl-uniform-out-of-bounds
spec@!opengl 1.4 at gl-1.4-dlist-multidrawarrays


-- 
Earthling Michel Dänzer               |               http://www.amd.com
Libre software enthusiast             |             Mesa and X developer