Nvidia driver - GeForce 2 card locks up pc

[neeko]

I've downloaded and installed the Nvidia driver for Linux (Mdk 10 ce) and went through the installation process all ok.

But, after about half an hour of playing TuxRacer/GL-117/Chromium the pc locks up completely. No mouse, no keyboard - so unable to ctrl-alt-backspace or ctrl-alt-F1 etc.

Its nice to be able to play some of these OpenGL games, but not when I have to cold reboot every time.

Can anyone give me some ideas of what to look for to fix this?

Thanks
N

[LordMorgul]

First check if you can determine anything useful from the logs, if you can login remotely to the system before rebooting it that may get you the best info. There should be messages in /var/log/messages or /var/log/dmesg that pertain to the kernel panic (if that is indeed what is occuring).

Monitor the system in any way possible while forcing this condition, check temperatures, fan speeds, cpu load,e tc... anything you can find out.
Force the condition with other games and/or screensavers... time it to see if its frequently near the same time.

And most importantly... post your X log for the simple startup before the crash and all applicable information about the system (such as which AGP module is being used and what kernel version is running).

[neeko]

Thanks for your comments. I am not able to remote login to the pc as this is my only linux box - rest are Macs.

There are no errors in the XFree86.0.log file, but there are these warnings:


(WW) NVIDIA(0): The user specified HorizSync "30.000-97.000" has been adjusted
(WW) NVIDIA(0): to "30.000-92.000" (the intersection with EDID-specified
(WW) NVIDIA(0): HorizSync "30.000-92.000")
...
(WW) NVIDIA(0): Not using mode "1152x768":
(WW) NVIDIA(0): horizontal sync start (1178) not a multiple of 8
(WW) NVIDIA(0): Not using mode "768x576":
(WW) NVIDIA(0): horizontal sync width (846 - 832 = 14) not a multiple of 8
(WW) NVIDIA(0): Not using mode "576x384":
(WW) NVIDIA(0): horizontal sync start (589) not a multiple of 8
(WW) NVIDIA(0): Not using mode "360x200":
(WW) NVIDIA(0): horizontal sync start (378) not a multiple of 8
...

(WW) NVIDIA(0): Failed to verify AGP usage

The only thing that looks odd to me in the messages log file is this:
...
Apr 9 14:48:38 tarantula XFdrake[3154]: ### Program is starting ###
Apr 9 14:48:51 tarantula drakconf.real[3120]: modified file /etc/mcc.conf
Apr 9 14:48:51 tarantula su(pam_unix)[3114]: session closed for user root
Apr 9 14:49:14 tarantula mdkkdm[2318]: X server for display :0 terminated unexpectedly
Apr 9 14:49:14 tarantula gconfd (john-2901): Received signal 15, shutting down cleanly
Apr 9 14:49:14 tarantula gconfd (john-2901): Exiting
Apr 9 14:49:14 tarantula kde3(pam_unix)[2447]: session closed for user john
Apr 9 14:49:16 tarantula kernel: atkbd.c: Unknown key released (translated set 2, code 0x7a on isa0060/serio0).
Apr 9 14:49:16 tarantula kernel: atkbd.c: This is an XFree86 bug. It shouldn't access hardware directly.
Apr 9 14:49:16 tarantula kernel: atkbd.c: Unknown key released (translated set 2, code 0x7a on isa0060/serio0).
Apr 9 14:49:16 tarantula kernel: atkbd.c: This is an XFree86 bug. It shouldn't access hardware directly.

I changed the xserver and screen res to 1280x1024 at 16bpp from 24bpp.

And this from dmesg:
Linux agpgart interface v0.100 (c) Dave Jones
agpgart: Detected VIA Twister-K/KT133x/KM133 chipset
agpgart: Maximum main memory to use for agp memory: 556M
agpgart: AGP aperture is 4M @ 0xfe000000
nvidia: no version for "struct_module" found: kernel tainted.
nvidia: module license 'NVIDIA' taints kernel.
0: nvidia: loading NVIDIA Linux x86 NVIDIA Kernel Module 1.0-5336 Wed Jan 14 18:29:26 PST 2004

uname -r gives: 2.6.3-4mdk

The time thing was just my estimate based on how long I can play an OpenGL game before the pc locks up.

glxgears gives:
2314 frames in 5.0 seconds = 462.800 FPS
2686 frames in 5.0 seconds = 537.200 FPS
2681 frames in 5.0 seconds = 536.200 FPS
2626 frames in 5.0 seconds = 525.200 FPS
2670 frames in 5.0 seconds = 534.000 FPS

The device part of XF86Config-4 is:
Section "Device"
Identifier "device1"
VendorName "nVidia Corporation"
BoardName "NVIDIA GeForce2 DDR (generic)"
Driver "nvidia"
BusID "PCI:0:15:0"

Option "DPMS"

EndSection

Yes I'm using a PCI graphics card as my mobo has no agp slot. I've tried commenting out the DPMS option and restarting X but that made no difference.

I should also mention that my mobo has onboard video (not in use) which may account for the agpgart messages perhaps?

Thanks
N

[SourPearPirate]

If the Macs run OS X then you can indeed do a remote login. I quite frequently have to as the bug which makes the X server freeze up and only the mouse move affects my system.

[LordMorgul]

If the onboard video is not yet disabled in the BIOS shut it off. It should not create problems, but it is one less thing to consider. The performance speeds you're getting look reasonable for that setup, so the problem is just a kernel related issue.

If you cannot work out a way to stay remotely logged in while the machine crashes, then you should try using several different kernel versions, both newer and older.

[neeko]

Thanks for your comments.

The two Macs I have are not running OS X. But I did install NetAtalk on my linux box and then mount the share on one of the Macs. Sure enough when the linux box died (just running glxgears for about 10 minutes does the trick) the share failed too.

I've tried the obvious stuff like disabling AGP in the XF86Config-4 file and making sure that the onboard lan is disabled in bios, but still no go.

I'v just read (in an older thread) about ACPI causing trouble with some kernels - not really being disabled. So I'll try a different kernel next.

N