Xorg freeze with GeForce 7900GT, 1.0-9746, Ubuntu 6.10

[MacSlow]

I constantly run into Xorg freezing when stressing my machine with some OpenGL (happens with compiz, blender, own OpenGL-programs, games, gl-screensavers etc.).

Once this freeze happens I can usually still work on the machine via a remote ssh session I logged into _before_ the freeze (afterwards I cannot log in via ssh, it has to be before the freeze). Although the connecting stalls from time to time but then recovers again. The load is at 1.0 or 2.0, and Xorg is listed in top with 99% on CPU-usage.

I cannot successfully recover form that with a "shutdown -r now" or a "kill -9 Xorg". I have to do a reset (and even after that the machine does not always properly come back up). Best thing to do is to power-off, wait a few seconds and pray the journaling filesystem does its job well on the next boot.

I tried disabling XrenderAccel and all 4 different values for NvAGP sofar. That did not fix the situation. I suffer form this bug for a long time now and no driver release up to now fixed that. It makes the machine nearly worthless to me as I need OpenGL constantly.

Setup looks like:
stock Ubuntu 6.10
Asus A8N-SLI deluxe (bios revision 1017-004)
Athlon64 3700+ (not overclocked)
evga GeForce 7900GT (not overclocked)
460W power-supply from Emermax
2 GBytes of RAM
and so on

The attached files are created/copied right after the freeze occured via a remote ssh connection.

Best regards...

Mirco "MacSlow" Müller

[am2020]

Hi. I have a machine with almost your same setup ! It's an ASUS SLI32 Deluxe (Is yours the 32?) and I have an AMD 4800+ and an XFX 7900GT card. I am going to install Kubuntu 6.10 (your is ubuntu - it's the same) tonight. My power supply is 600W though.

Give me an app list or things to try and I wil be happy to test them.

[netllama]

In your bug report I see:
(II) NVIDIA(0): The NVIDIA X driver has encountered an error; attempting to
(II) NVIDIA(0): recover...
(EE) NVIDIA(0): Error recovery failed.
(EE) NVIDIA(0): *** Aborting ***

Do you have a reliable means of triggering this problem, or is it seemingly random? Does it reproduce if you're not using the Composite extension?

Does this problem persist with a more recent kernel.org kernel (2.6.19.x)?

Also, it looks like you're using SBIOS version 1017, however I don't see that version listed on Asus's website. Where did you obtain 1017?

Thanks,
Lonni

[MacSlow]

Originally Posted by netllama

Do you have a reliable means of triggering this problem,

Yes, grab glitz-test from http://people.freedesktop.org/~macslow (either tarball or via git). Compile it (you'll need gtk+ 2.10.x, glitz 0.5.6, librsvg 2.14.x, cairo 1.2.x or better versions of those libraries). Compile and run it. Either with or without command-line option -n (disables use of glitz, thus OpenGL) or with or without compiz running. It will trigger the freeze after a few moments.

If this info helps... I've tried several other systems with a comparable nvidia hw/driver combo and never encountered this freeze. But all those systems are AGP and not PCIe.

Originally Posted by netllama

or is it seemingly random?

No, heavy OpenGL-usage will trigger it within a few minutes.

Originally Posted by netllama

Does it reproduce if you're not using the Composite extension?

I'll have to test that. I will report back once I've done that.

Originally Posted by netllama

Does this problem persist with a more recent kernel.org kernel (2.6.19.x)?

I have not tried it with newer kernel yet. Not sure I've the time to test that atm.

Originally Posted by netllama

Also, it looks like you're using SBIOS version 1017, however I don't see that version listed on Asus's website. Where did you obtain 1017?

I cannot remember anymore, it is several months ago. But it was an Asus-site for sure. I would not have flashed my motherboard's BIOS with a non-Asus BIOS-image.

Best regards...

Mirco "MacSlow" Müller

[MacSlow]

Originally Posted by netllama

Does it reproduce if you're not using the Composite extension?

Yes. The freeze still happens when I use my /etc/X11/xorg.conf looking like this...

...
Section "Screen"
    Identifier     "Default Screen"
    Device         "NVIDIA Corporation NV71 [GeForce 7900GT]"
    Monitor        "Samsung SyncMaster 215TW"
    DefaultDepth    24
    Option         "NoLogo" "true"
    Option         "NvAGP" "1"
    Option         "RenderAccel" "false"
    Option         "DisableGLXRootClipping" "true"
    Option         "AllowGLXWithComposite" "false"
    Option         "AddARGBGLXVisuals" "false"
    Option         "BackingStore" "true"
    SubSection     "Display"
        Depth       24
        Modes      "1680x1050"
    EndSubSection
EndSection

Section "Extensions"
    Option         "Composite" "false"
EndSection

[netllama]

I'm attempting to build glitz-test, and its failing:
gcc -Wall -c -O3 -std=c99 `pkg-config --cflags cairo gtk+-2.0 librsvg-2.0 glitz glitz-glx` -c cairo-rendering.c -o cairo-rendering.o
In file included from cairo-rendering.h:27,
from cairo-rendering.c:21:
glitz-helper.h:24:25: error: cairo-glitz.h: No such file or directory


I've got gtk+ 2.10.x, glitz 0.5.6, librsvg 2.14.x, & cairo 1.2.x installed, but there is no cairo-glitz.h included with any of them. Any ideas?

[MacSlow]

Originally Posted by netllama

I'm attempting to build glitz-test, and its failing:
gcc -Wall -c -O3 -std=c99 `pkg-config --cflags cairo gtk+-2.0 librsvg-2.0 glitz glitz-glx` -c cairo-rendering.c -o cairo-rendering.o
In file included from cairo-rendering.h:27,
from cairo-rendering.c:21:
glitz-helper.h:24:25: error: cairo-glitz.h: No such file or directory


I've got gtk+ 2.10.x, glitz 0.5.6, librsvg 2.14.x, & cairo 1.2.x installed, but there is no cairo-glitz.h included with any of them. Any ideas?

Oh sure... sorry forgot to mention that you need glitz-enabled cairo.

The best approach is to grab cairo from upstream...

http://cairographics.org/snapshots/cairo-1.3.10.tar.gz

unpack it and build with...

./configure --prefix=/tmp/testing --enable-glitz && make && make install

You might want to also reget glitz-test from here...

http://people.freedesktop.org/~macsl...z-test.tar.bz2

I fixed a few bugs in it, but they sadly are unable to address the xorg-freeze. Then cd to glitz-test's directory and do...

export PKG_CONFIG_PATH=/tmp/testing/lib/pkgconfig
export LD_LIBRARY_PATH=/tmp/testing/lib
make clean ; make

That should successfully compile the glitz-test binary. glitz-test offers some command-line options...

Usage:
  glitz-test [OPTION...] - cairo-based stress-test

Help Options:
  -?, --help                Show help options

Application Options:
  -n, --no-glitz            don't use glitz (default: use glitz)
  -r, --refresh-rate=Hz     render at frames/sec (default: 30 Hz)

Start it like...

./glitz-test

and wait a few moments. It should freeze Xorg. At least this happens on my box constantly.

You can quit the glitz-test with the q- or esc-key, change the demos with the F1..F4 keys, or mouse-wheel or cursor keys, or PageUp/PageDown keys. Space toggles between pause and animation. It's fun if it works...



Best regards...

MacSlow

[netllama]

Thanks, that helped and I'm able to run glitz-test now. In fact, its been running in a system with Asus A8N-SLI-Deluxe & 7900GT for the past 5 minutes without any problems. How long is typically needed to reproduce the hang?

Is this hang specific to a certain window manager or compositing manager?

Also, you should check on your SBIOS, as that version doesn't seem to be listed on Asus's website, so I'm wondering if it was a beta BIOS which they removed due to some bug. My system has 1016, which is the latest stable BIOS listed on their website.

Thanks,
Lonni

[MacSlow]

Originally Posted by netllama

Thanks, that helped and I'm able to run glitz-test now. In fact, its been running in a system with Asus A8N-SLI-Deluxe & 7900GT for the past 5 minutes without any problems. How long is typically needed to reproduce the hang?

Usually less than a minute.

Originally Posted by netllama

Is this hang specific to a certain window manager or compositing manager?

No it happens under metacity (2.16.3) and compiz (0.3.7).

Originally Posted by netllama

Also, you should check on your SBIOS, as that version doesn't seem to be listed on Asus's website, so I'm wondering if it was a beta BIOS which they removed due to some bug. My system has 1016, which is the latest stable BIOS listed on their website.

I downgraded the BIOS to 1016 now. No change. The freeze still happens with Composite enabled. Without Composite I'm running about 5 minutes now with a fullscreen OpenGL-app.

Another info that might be of help. The moment before the freeze the screen sort of "flashes" briefly once... as if something was rendered all over the desktop for a moment.

Best regards...

MacSlow

[netllama]

Does this only fail with compiz (not beryl, or any other compositing manager)?

[MacSlow]

Originally Posted by netllama

Does this only fail with compiz (not beryl, or any other compositing manager)?

I don't use beryl.

I just tried xcompmgr with enabled Composite, AllowGLXWithComposite and RenderAccel but disabled AddARGBGLXVisuals. I ran that with celestia and another self-written OpenGL app. I waited about 10 minutes before I quit that.

Now I'm repeating this with compiz and it's running for about 5 minutes now with no freeze. So for the moment I'm suspecting AddARGBGLXVisuals to be the trouble-maker.

Did you have AddARGBGLXVisuals enabled when you tried it?

Best regards...

MacSlow

[netllama]

Yes, I had AddARGBGLXVisuals enabled, however I wasn't using compiz. It would be useful to know whether the trigger here is compiz or AddARGBGLXVisuals or both.