This is the bug forcing us to blacklist Gallium in Firefox 6+.
Steps to reproduce:
0. This has been confirmed with Mesa 7.10.2, Gallium 0.4, and both R300 and R600 but seems device-independent. I didn't find the proper category for general Gallium bugs.
1. Grab a Nightly of Firefox:
2. Run it; If you're already using Firefox, you may want to do:
./path/to/firefox -no-remote -P
3. Make sure that WebGL is enabled (a bug in parsing version strings makes it unintentionally blacklist some systems). Go to about:config, set webgl.force-enabled=true if needed. You can go to about:support to check WebGL status.
4. Go to http://people.mozilla.com/~sicking/webgl/ray.html
Actual results: segfault
A stack trace is given there:
I discussed this with Corbin Simpson (https://bugzilla.mozilla.org/show_bug.cgi?id=624935#c6) who told me it was a known Gallium bug, been discussed on the mailing list already:
This is the same as bug 33188 isn't it?
(In reply to comment #1)
> This is the same as bug 33188 isn't it?
Looks like it (didn't know about the other bug), but this bug has more specific info, and also the testcase link is more specific.
Close the other one as duplicate of this one (despite anachronism) ?
(In reply to comment #2)
> Close the other one as duplicate of this one (despite anachronism) ?
That's probably a good idea. You could also reassign this bug to component "Mesa core" as it isn't a problem specific to r600g.
*** Bug 33188 has been marked as a duplicate of this bug. ***
This may well be a dupe of bug 33203.
The same glXMakeCurrent with drawable set to None and ctx set to NULL avoided this as well, at least.
It's probably fixed by http://cgit.freedesktop.org/mesa/mesa/commit/?id=bf69ce37f0dcbb479078ee676d5100ac63e20750
(In reply to comment #6)
> It's probably fixed by
What Mesa version number/string should I check for? 7.11 ?
Also, there is something that I would like to understand before un-blacklisting Gallium. In https://bugzilla.mozilla.org/show_bug.cgi?id=624935#c6 a Gallium bug about shaders with too many texture indirections was mentioned, see
The present bug/patch doesn't seem to have anything to do with that. Does that mean that the other bug (mentioned in these emails) still exists?
Karl: with the present fix, and the fix from https://bugzilla.mozilla.org/show_bug.cgi?id=659842 , is there still any serious Gallium WebGL crasher? Does the conformance test suite pass without a crash?
I have tried the conformance test with Firefox 7.0a1 and Mesa c0c0bb6cb140825f5bab3c40c0c9c0ec575fbc76
r600g on a HD5670:
Results: (5665 of 5785 passed, 2 timed out)
r300g on a RV570
Results: (5772 of 5907 passed, 1 timed out)
llvmpipe still crashes firefox.
Is this still an issue?
Firefox, WebGL and Mesa have worked together well for a long time, this was likely fixed years ago. Closing.