glamor versus AIGLX GL context series

[Eric Anholt]

Michel Dänzer <michel at daenzer.net> writes:

> On 19.04.2014 03:39, Eric Anholt wrote:
>> Here it is, at long last.  I was stumped for several days with a weird
>> fbo incomplete that was happening after my first round of fixes, and
>> it took a day's detour of cooking up a Xephyr with DRI3 to get a fast
>> debug environment and figure out what was going on (xephyr-glamor-egl
>> is that branch.  I warn you: that branch is almost entirely made up of
>> layering violations and bad assumptions, and should not be viewed by
>> people with a weak stomach).
>> 
>> With this series, I've run piglit on my glamor-using desktop, and also
>> played with some LIBGL_ALWAYS_INDIRECT=1 glxgears.  Those both used to
>> take down my server in different ways, so I think this fixes what we
>> need for 1.16.  However, swrast AIGLX (as used in xephyr-glamor-dri3)
>> has a problem with recursive flushes during the loader's GetImage, and
>> I don't have a solution to it.
>> 
>> This branch can be found on glamor-gl-context-2 of my tree.
>
> This didn't reliably survive LIBGL_ALWAYS_INDIRECT=1 glxgears for me, let
> alone piglit. valgrind found this:
>
> By changing the context, it may also invalidate the DRI2 buffer information,
> so we need to get that again.
>
> Fixes crashes due to use-after-free with LIBGL_ALWAYS_INDIRECT=1 glxgears
> and piglit.

Glad to see things are working for you.  I've added it to the series.
If we can get this and the render change reviewed, I think the must-do
items for glamor in 1.16 will be done.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 818 bytes
Desc: not available
URL: <http://lists.x.org/archives/xorg-devel/attachments/20140421/60965a61/attachment.sig>