3D apps freeze the machine temporarily

[pierrec]

Hello everybody,

I have an odd problem related to the nvidia gl : when I run a 3D app, like oolite, tuxracer or 3D screen savers, after a very short time doing 3D things, the entire screen and keyboard freeze hard for a moment, then the mouse pointer seems to come back to life, then the X session comes back to life for a second or two, then it freezes again etc... Sometimes it freezes for 10 seconds, sometimes for 15 minutes, but the machine stays alive (I can ssh to it) and if I'm patient enough and quick enough with the mouse, I can usually kill the offending app during the "recovery" period and everything's back to normal in 2D-land. There are no error messages from the kernel or from the X log.

My configuration is as follow:

- Athlon XP2000+ on a KT400-based VIA motherboard (yes, I know...)
- GeForce FX5200
- Linux 2.6.17.6
- nVidia driver 8762
- Xorg 1.0.2

I've tried everything I could try to fix the problem, including reinstalling older nVidia drivers (the 7667 and the 7182 with legacy GPU support) disabling RenderAccel, disabling AGP altogether, using NvAGP instead of agpgart, removing APIC support from the kernel and BIOS, increasing/decreasing the AGP aperture size in the BIOS from 4M to 1G, enabling/disabling AGP fast-writes and SBA, backing down to AGP 4X instead of 8X (the BIOS won't let me set 2X or 1X), manually setting values of 0xEA and above in the "AGP driving value" setting in the BIOS, etc etc... but nothing makes a bit of difference and the problem is still there.

Does anybody know what might be going on here, and how I might work around the problem? is this an issue known to nVidia?

EDIT: Actually there is a kernel message each time after a freeze occur. It's always the same:

NVRM: Xid (0001:00): 8, Channel 00000002

I don't know what that might mean though...

[pe1chl]

I don't have this problem. But when I see your description I would consider an issue with interrupt sharing. What other devices is your video card sharing interrupts with (do cat /proc/interrupts) and is there a way to move some of those out of the way? (e.g. putting a PCI card in another slot)

[pierrec]

Originally Posted by pe1chl

I don't have this problem. But when I see your description I would consider an issue with interrupt sharing. What other devices is your video card sharing interrupts with (do cat /proc/interrupts) and is there a way to move some of those out of the way? (e.g. putting a PCI card in another slot)

That was my first thought, but no, the video card sits all alone on IRQ11:

ppc@akula:~$ cat /proc/interrupts
          CPU0
  0:      63476          XT-PIC  timer
  1:       1182          XT-PIC  i8042
  2:          0          XT-PIC  cascade
  8:          4          XT-PIC  rtc
 10:       2994          XT-PIC  EMU10K1, EMU10K1
 11:      13875          XT-PIC  nvidia
 12:       4512          XT-PIC  i8042
 14:       9642          XT-PIC  ide0
 15:         53          XT-PIC  ide1
NMI:          0
ERR:          1