Bug 16615

When you say "fairly recent", can you point out a version of the driver that works without artifacts? Even better, can you use git bisect to pinpoint the commit where it turns bad?

(In reply to comment #1)
> When you say "fairly recent", can you point out a version of the driver that
> works without artifacts? Even better, can you use git bisect to pinpoint the
> commit where it turns bad?

When I say "recent", I mean that it used to work in the not-too-distant past. However, that was still many kernels, builds of wine and one prior release of Fedora ago. WoW also went through an extended period of not running *at all* under Mesa, and crashing immediately on start-up:

$ warcraft
fixme:mixer:ALSA_MixerInit No master control found on USB Device 0x46d:0x8b4, disabling mixer
fixme:mixer:ALSA_MixerInit No master control found on USB Device 0x46d:0x992, disabling mixer
fixme:advapi:SetSecurityInfo stub
archive Data\patch.MPQ opened
archive Data\enGB\patch-enGB.MPQ opened
archive Data\enGB\patch-enGB-2.MPQ opened
archive Data\patch-2.MPQ opened
archive Data\expansion.MPQ opened
archive Data\common.MPQ opened
archive Data\enGB\locale-enGB.MPQ opened
archive Data\enGB\speech-enGB.MPQ opened
archive Data\enGB\expansion-locale-enGB.MPQ opened
archive Data\enGB\expansion-speech-enGB.MPQ opened
fixme:win:EnumDisplayDevicesW ((null),0,0x32eda4,0x00000000), stub!
fixme:win:EnumDisplayDevicesW ((null),0,0x32ec94,0x00000000), stub!
fixme:win:EnumDisplayDevicesW ((null),0,0x32f2c8,0x00000000), stub!
fixme:win:EnumDisplayDevicesW ((null),0,0x32f42c,0x00000000), stub!
fixme:win:EnumDisplayDevicesW ((null),0,0x32f5a8,0x00000000), stub!
fixme:win:EnumDisplayDevicesW ((null),0,0x32f5a0,0x00000000), stub!
fixme:win:EnumDisplayDevicesW ((null),0,0x32f528,0x00000000), stub!
fixme:win:EnumDisplayDevicesW ((null),0,0x32f518,0x00000000), stub!
X Error of failed request:  GLXBadDrawable
  Major opcode of failed request:  156 (GLX)
  Minor opcode of failed request:  5 (X_GLXMakeCurrent)
  Serial number of failed request:  585
  Current serial number in output stream:  585

The only reason that I have been able to play again on this machine has been because I have suddenly discovered that running:

# prelink /usr/lib/libGL.so.1.2
# prelink /usr/lib/libGLU.so.1.3.070100

on new Mesa installations seems to do some "magic" of its own that avoids the X_GLXMakeCurrent error.

Having said all that, I do think that Mesa git was working for me when Marcus Amsler was checking WoW/Radeon patches in.

> X Error of failed request:  GLXBadDrawable
>  Major opcode of failed request:  156 (GLX)
>  Minor opcode of failed request:  5 (X_GLXMakeCurrent)
>  Serial number of failed request:  585
>  Current serial number in output stream:  585
I remember this error too, but I' haven't seen it for a long time. I think (not sure) it was a compile/installation issue.

The artifacts look quite similar to the ones I had and fixed with

commit acb47dee69a165f242d88f9eac60fc5646e33410
Author: Markus Amsler <markus.amsler@oribi.org>
Date:   Fri May 2 01:58:18 2008 +0000

    r300: Set correct VAP_CNTL per vertex program.

Does disabling pixelshaders help?

Just checked it with a RV370, its the multitexture problem with s3tc. you should see a

*********************************WARN_ONCE*********************************
File r300_texstate.c function compute_tex_image_offset line 231
DXT 3/5 suffers from multitexturing problems!
***************************************************************************

warning in the console. Disabling s3tc in driconf should do the trick.
Strange is that disabling pixelshaders removes all terrain related artifacts, I'm not sure why. Also on r500 I see no artifacts at all, although it should suffer from the same multitexture problems.

(In reply to comment #4)
> Just checked it with a RV370, its the multitexture problem with s3tc. you
> should see a
> 
> *********************************WARN_ONCE*********************************
> File r300_texstate.c function compute_tex_image_offset line 231
> DXT 3/5 suffers from multitexturing problems!
> ***************************************************************************

Aaaggh! I can't currently test anything with WoW/Mesa because the

X Error of failed request:  GLXBadDrawable
  Major opcode of failed request:  156 (GLX)
  Minor opcode of failed request:  5 (X_GLXMakeCurrent)
  Serial number of failed request:  585
  Current serial number in output stream:  585

crash is back, and this time I can't get rid of it by running prelink. And this is not a build error, because I've rebuilt mesa from scratch by running "make realclean; make linux-dri-x86" first.

OK, I've managed to run WoW again by "applying patience". The X_GLXMakeCurrent error seems to occur randomly, which I suppose suggests a race condition somewhere. (This is a SMP machine with 2 physical hyper-threaded CPUs, providing 4 logical CPUs.) Retry WoW often enough and it runs eventually.

(In reply to comment #4)
> Just checked it with a RV370, its the multitexture problem with s3tc. you
> should see a
> 
> *********************************WARN_ONCE*********************************
> File r300_texstate.c function compute_tex_image_offset line 231
> DXT 3/5 suffers from multitexturing problems!
> ***************************************************************************
> 
> warning in the console. Disabling s3tc in driconf should do the trick.

No, I am no longer seeing this message at all. I used to see it all the time, but it no longer matters if I enable S3TC via driconf or not.

> Strange is that disabling pixelshaders removes all terrain related artifacts,

Yes, adding 'SET pixelShaders "0"' to Config.wtf does indeed remove all the terrain-related artifacts.

Can you reproduce the bug with current mesa master?

Mass version move, cvs -> git

Closing due to lack of user input.