non-deterministic system freezes with 1.0-9625/9626 running compiz

[MacSlow]

Greetings everybody!

As nice as the native support for GLX_EXT_texture_from_pixmap is, but I get random system-freezes with the combination of the 9625/9626 driver and compiz. Sometimes I can run the system under full load (4xFSAA, 8xANIO, compiz, some xscreensaver's gl-hacks, videos or mp3s playing in totem) for hours before it freezes hard. Then there are times when the machine freezes right away after I successfully logged in or even as early as the gdm-login screen. Then I have to press the reset button (ssh-ing to the machine from remote nearly never works in order to shut it down cleanly). On the next reboot I then often see the kernel complaining about IRQs not working or not being received by my network-card or the nvidia-card. Sometimes it is so bad, that not even the BIOS can get the machine booted.

The freeze happens sometimes during resizing or moving a windows. But there are also times when it freezes just as is. I sadly cannot provide a procedure that triggers the freeze in a deterministic fashion.

Because I was assuming the system-RAM to be failing (memtest+ V1.65 reported errors... but only sometimes) I replaced the DIMM with a new one. Sadly that didn't fix anything.

Also unplugging all add-in cards except for the graphics-card didn't improve the situation.

I've even updated the mainboards BIOS to the newest beta-version from Asus in the hope it would fix these issues. That didn't help either.

The only way to use my machine without failure is to use the "nv" driver, which is not really an option for me as I need proper OpenGL-acceleration for development and the composited desktop-environment.

I'm using the nvidia-driver with zander's SMBUS-patch applied, because otherwise Xorg would not load up.

The machine looks like this:
Asus A8N SLI Deluxe
AMD Athlon64 3700+
eVGA 7900GT
(no component is overclocked)
running Ubuntu EdgyEft (always updated)

I hope I've supplied all information helpful for you engineering folks at nvidia to resolve this issue.

Best regards...

MacSlow

[prefect42]

Running the same motherboard with two cards I am force to run with the kernel argument of pci=conf1 or I suffer intermittent lockups.

[JaXXoN]

I have experienced that once an nvidia card crashes the system,
then you should power down your system for at least twenty seconds.
Otherwise it may easily happen that the system again crashes pretty
quickly: I'm sometimes playing around with kernel drivers, xen, realtime,
etc., and since a graphics driver is a pretty complex piece of software
it is typically one of the first parts that indicate that it is not happy with
a certain "improovement". Then it very oftenly happens that the system
crashes pretty quickly after pressing the reset button and re-booting into
an actually "known to be pretty stable" setup.

You may like to redirect the kernel messages to a serial console
so that you can capture any potential error messages from another
PC when the system crashes.

regards

Bernhard

[MacSlow]

Originally Posted by prefect42

Running the same motherboard with two cards I am force to run with the kernel argument of pci=conf1 or I suffer intermittent lockups.

Ah... but what's the meaning of "pci=conf1"?

Best regards...

MacSlow

[prefect42]

Originally Posted by MacSlow

Ah... but what's the meaning of "pci=conf1"?

It's been mentioned properly elsewhere on this board. Redhat has a very short comment on it: http://kbase.redhat.com/faq/FAQ_85_9249.shtm

[MacSlow]

I've ran into a way to reproduce the freeze in deterministic a fashion.

1.) Install and use any of the 1.0-9x.yz drivers
2.) Run compiz (or beryl).
3.) Start OpenOffice.

As soon as e.g. OpenOffice Writer has finished loading and is ready for input the afore mentioned freeze occurs.

Best regards...

MacSlow

[prefect42]

Originally Posted by MacSlow

I've ran into a way to reproduce the freeze in deterministic a fashion.

1.) Install and use any of the 1.0-9x.yz drivers
2.) Run compiz (or beryl).
3.) Start OpenOffice.

As soon as e.g. OpenOffice Writer has finished loading and is ready for input the afore mentioned freeze occurs.

9629 running on FC6 2239 on a Dell Precision 470 with nVidia Quadro FX 540. Compiz running with all the bells and whistles. Start openoffice. Open new document. Play with it. All is fine for me.

[albright]

running an amd64x2 with nvidia 6600. Mandriva 2007. I
find that with beryl I get random crashes of X; compiz
is rock solid (knock on wood). This with the 9629 drivers
(anyone know if the 9642 beta's are better).

have no idea what is the problem (I've tried pci=nommconf
(same as conf1 I guess), idle=poll with no success)

[MacSlow]

By chance I found out that my box doesn't really freeze. Rather Xorg/compiz stops repainting. The cpu-load is at about 2.0, music is still playing (although interrupted every few seconds) and I can ssh to my machine form another system. The question is now who's responsible... the driver, compiz or Xorg (or some other component)?

Best regards...

MacSlow

[prefect42]

Originally Posted by MacSlow

The question is now who's responsible... the driver, compiz or Xorg (or some other component)?

Easy to test if it's compiz. ssh in and killall -9 compiz.