Bug 42683

Summary: [r300g] GPU lockup while playing WoW
Product: Mesa Reporter: Chris Rankin <rankincj>
Component: Drivers/Gallium/r300Assignee: Default DRI bug account <dri-devel>
Status: RESOLVED FIXED QA Contact:
Severity: normal    
Priority: medium    
Version: git   
Hardware: x86 (IA32)   
OS: Linux (All)   
Whiteboard:
i915 platform: i915 features:
Attachments: dmesg log from GPU lockup
Two GPU resets while playing WoW

Description Chris Rankin 2011-11-07 13:31:56 UTC 
Created attachment 53271 [details]
dmesg log from GPU lockup

I'm not sure whether this is a new problem or just a particularly bad side effect of bug #41698, but WoW has now begun locking up after just a few moments of gameplay. I'm sure it wasn't as bad as this yesterday though.

The latest commit in git is:

commit dceb202297b39220fbbcb41267077fb3ff8d137a
Author: Kenneth Graunke <kenneth@whitecape.org>
Date:   Mon Nov 7 12:07:44 2011 -0800

    i965: Fix Sandybridge regression introduced by workaround-free math.

And I'm fairly sure it was working yesterday with:

commit 2318443ebd6bf9727676b530a3b057dcf13a3ca6
Author: Dave Airlie <airlied@redhat.com>
Date:   Sun Nov 6 19:38:50 2011 +0000

    gl3.txt: update for ARB_texture_storage.

There is something seriously wrong with r300g.
Comment 1 Marek Olšák 2011-11-07 14:50:34 UTC 
Could you please bisect?
Comment 2 Chris Rankin 2011-11-07 15:13:25 UTC 
(In reply to comment #1)
> Could you please bisect?

Haven't we already played the bisection game, though? I proved that commit:

363ff844753c46ac9c13866627e096b091ea81f8 is the first bad commit
commit 363ff844753c46ac9c13866627e096b091ea81f8
Author: Marek Olšák <maraeo@gmail.com>
Date:   Tue Sep 27 01:10:20 2011 +0200

    winsys/radeon: move GEM domains out of the drivers into winsys

was bad ages ago, and have already mentioned getting a GPU lockup with it applied. Any new bisection must therefore include this "known bad" commit.
Comment 3 Chris Rankin 2011-11-07 16:44:29 UTC 
(In reply to comment #1)
> Could you please bisect?

The problem does not occur between dceb202297b39220fbbcb41267077fb3ff8d137a and 2318443ebd6bf9727676b530a3b057dcf13a3ca6, so I can only assume that these two GPU lockups were due to an intermittent bug that was already present.

FWIW, there seems to be a small drop in FPS between Mesa 7.11 and Mesa git. 7.11 (strictly speaking, it's mesa-dri-drivers-7.11-1.fc15.i686) is giving me 10-11 fps in situations where git is currently giving me 7-8 fps. Possibly the lockup requires a certain level of graphical "business" which just isn't present at 12:30 AM.
Comment 4 Chris Rankin 2011-11-22 12:57:56 UTC 
Created attachment 53788 [details]
Two GPU resets while playing WoW

These two GPU resets happened in quick succession, although the kernel managed to recover each time. (Small comfort, I suppose).

The kernel is 3.1.2, with Mesa HEAD at:

commit 034e63b9f8ee23cfac769b1a2c431bdd2307a6c4
Author: Marek Olšák <maraeo@gmail.com>
Date:   Tue Nov 22 20:48:23 2011 +0100

    r600g: handle all remaining CAPs

This does not bode well for RV350 support with Mesa 7.12.
Comment 5 Chris Rankin 2012-06-08 06:44:53 UTC 
This hasn't happened for a long time.

Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.