Need help debugging possible graphics-related crash

[Dard]

Okay, so I tried again.
I deleted my whole /usr/local directory. Finally all traces of the mesa drivers are gone and no longer loaded. My system is now rock solid, as far as I can see. But only 2D.
Apart from the crashes I fixed another bug this way that caused graphics glitches when I moved areas, for example when a tooltip is opened inside Firefox and I scrolled with the wheelmouse. I decided to investigate further, maybe it was my window manager.

My previous mesa installation was ages old, so I tried to reinstall the current mesa, this time in /usr/local/xorg, so that I can remove it quickly without destroying other applications I have installed in /usr/local.

My problem are probably the following messages in Xorg.0.log:

(EE) AIGLX error: dlopen of /usr/lib/dri/r600_dri.so failed (/usr/lib/dri/r600_dri.so: cannot open shared object file: No such file or directory)
(EE) AIGLX: reverting to software rendering
(II) AIGLX: Screen 0 is not DRI capable
(EE) AIGLX error: dlopen of /usr/lib/dri/swrast_dri.so failed (/usr/lib/dri/swrast_dri.so: cannot open shared object file: No such file or directory)
(EE) GLX: could not load software renderer

I tried to follow the instructions at http://www.x.org/wiki/radeonBuildHowTo.

These are basically my commands when I installed:

git clone git://anongit.freedesktop.org/xorg/driver/xf86-video-ati
git clone git://anongit.freedesktop.org/mesa/drm
git clone git://anongit.freedesktop.org/mesa/mesa

cd drm/
./autogen.sh --prefix=/usr/local/xorg
make -j 3 ; make install ; cd ..

export PKG_CONFIG_PATH="/usr/local/xorg/lib/pkgconfig"
cd xf86-video-ati/
./autogen.sh --prefix=/usr/local/xorg
make -j 3 ; make install ; cd ..

cd mesa/
./autogen.sh --prefix=/usr/local/xorg --enable-gallium-r600 --with-dri-driverdir=/usr/local/xorg/lib/dri/
make -j 3 ; make install ; cd ..

Then I added the line "/usr/local/xorg/lib" to /etc/ld.so.conf and executed "ldconfig".
Then I added "/usr/local/xorg/lib/modules" to the ModulePath entry in xorg.conf, although this seems useless since there isn't a modules folder.
I set this variable:
LIBGL_DRIVERS_PATH=/usr/local/xorg/lib/dri/
In xorg.conf I still have this section:

Section "DRI"
    Group      "dard"
    Mode       0666
EndSection

I thought this is everything I need, but glxinfo gives me:

name of display: :0.0
Error: couldn't find RGB GLX visual or fbconfig

What am I missing?

[Dard]

Okay, thank you so far, but my system seems to be seriously screwed up.
First I removed the git drivers as far as I could by manually removing them from /usr/local. (there wasn't a "make uninstall" option)
Xorg seems to successfully fall back on the Debian drivers. Games like Chromium still work but with small glitches. And Xorg still crashes.

Then I removed Section "DRI" in xorg.conf.
To my surprise, the games still run as fast as if they had direct rendering, although glxinfo says no.
And the crashes still happen.

I have no way out, I guess I have to live with it for the time being.
At least it has become unlikely that the current Mesa driver is responsible.

[Dard]

I am using Debian testing.
I hoped this is a new enough system.
xserver-xorg 1:7.5+8
xserver-xorg-core 2:1.7.7-11

[Dard]

I have made some progress, but not enough
As it turned out, the computer wasn't completely dead once the crash appens.
When connected from another computer by ssh, I got an NMI-error. So I enabled the NMI-Watchdog. Maybe it is the watchdog that keeps things running, I don't know.
But now can still access the computer by ssh after crash. However, it is so slow that I can only get one command executed every 2 minutes. top shows varying tasks running at several hundred percent CPU time, among them the watchdogs.

By now the crashes are quite frequent, if I can't debug this further, I have to try go get back to a stable old version.
I have also another reliable crasher: Mincraft crashes within usually five minutes after start.

Now that I know I can still access the computer after the crash, I could debug, but I don't know how.
I can compile Mesa with debug options, but I don't know how to start it with gdb. After all, I only know how to start X by startx->xinit->X. And that isn't even Mesa.
Besides that, I have no idea how to debug a live-lock like this.

Any ideas?

BTW, after seeing the NMI-message, I thought about a hardware problem, which seems to be the most likely explanation. But I don't think I have experienced a Windows crash so far.

[Dard]

Most games run fine here, too.
Sauerbraten once caused a crash, but not repeatable.
Neverball, Chromium, Kobo Deluxe and several others ran too.
Gothic 1 in Wine is the only one that crashes the system every time.

[MoRZiLLo]

I upgraded to 2.6.38-rc6.
It still crashes reliably, still without any messages.

I'm on 2.6.38-rc6 and runs fine.
I send some test to phoronix Global today.
My rig is Core2DUO [email protected] 4gb DDR 1066 , and RV770(Ati HD4870 512mb old model) with Gallium 0.4 OGL 2.1 Mesa 7.11 devel

No crash , no black screens.
Only few games doe's not run(etqw and ut2004 and some Unigine tests)

[Dard]

I will not install a new system because it would take over a week to do properly and apart from some crashes now and then my current Debian system works just fine.

Also, if the crashes would stop that way, I could no longer provide a bug report. That way the bug (if there is one) would be corrected and not ignored.
I was glad to be able to reproduce it reliably but sadly have no idea where that bug could be located or how to reproduce it under more clearly defined conditions.

[Dard]

I upgraded to 2.6.38-rc6.
It still crashes reliably, still without any messages.

[Dard]

Recently my system becomes more and more unstable.
I more or less frequently update mesa from git and apart from my soundcard, my video driver is the only possible offender I can think of.

Symptoms:
Occasionally, the system just locks up hard. I get a black screen, the monitor turns off, no more hard disc or sound activity.
Reboot is the only option.

When does this happen:
Randomly, but it seems to happen more often when I do more extended graphic commands, like xrandr or 3D.
Just once it may have happened in Windows 7 as well, but I am not sure about that.

Yesterday, I found a sure way to reproduce this error: I use a fairly recent wine and try to start Gothic 1.

I tried to get a message by monitoring system messages via RS232 on my laptop, but once the hangup occurs, I do not get any messages.

I would like to debug this further to make a proper bug report, but I need suggestions.
The mesa git is just two days old, I compiled it with gallium enabled. I use kernel 2.6.36 on a debian system. My graphics card is a Radeon 4350.

Thanks!