2.4.20+NVIDIA driver

[amgarching]

Anybody running 2.4.20 + with NVIDIA???

I am constantly having VM faults as
- kernel BUG at vmscan.c (many Oops in SYSLOG)
- often followed by "Segmentation violation"
of perfectly stable programs (vi, man)
- often followed by X crashes
- grave system crashes

Is reproduced with agpgart and internal
Nvidia AGP driver.
Reproduced with 1.0-3123 and 1.0-4191
With 2.4.19 and 1.0-2960 I had acceptable
stability. The hell started after the kernel
upgrade to 2.4.20

Because of these problem I had to revert
to GPL'd driver on the "mission critical"
machines. But I have ~15 "graphic critical"
terminals to run :-(

[amgarching]

All cards having AGP GeForce2 MX chips
CPU is Athlon (one of those with
advanced speculative caching bug)

[cjewell]

Hi,

I'm also getting problems with 2.4.20. Here's what I did. I have a RedHat 8.0 system that comes with the stock 2.4.18-14 kernel. I needed to upgrade to customise a bit for some non-RedHat hardware I have. I installed the new kernel ok. Then I quit X server, booted the new kernel and went to the NVIDIA_kernel directory and ran "make SYSINCLUDE=/root/linux-2.4.20/include". Make went fine and loaded the nvidia.o modules ok. I then tried to startx. I was met with a screen of odd-looking text characters and no X server. Any ideas what's going on? Do I need to recompile the NVIDIA_GLX package?

Regards,

Chris

[Noth]

I've had the nVidia drivers running 2.4.20 (with XFS patches) for some time, had ~50 days of uptime with heavy use recently.

I have dual Athlon 1.2s and a GF3.

[Delleon]

I too am using the 2.4.20 kernel with the NVidia kernel module. I initially had some problems with the 4191 drivers, however I managed to get it working after some messing around with my BIOS settings and recompiling the kernel a few times. Because I'm using the Nforce2 chipset, I had no choice but to use NVidia's AGP support.

If you're having so many problems, why don't you just revert back to 2.4.19? Do you absolutely need to use 2.4.20?


Gentoo Linux w/ 2.4.20 kernel + XFS patches + preemptive kernel patch
Athlon 2100+
Abit NV7 Nforce2 chipset
Asus Geforce4 Ti 4200

[bwkaz]

Originally posted by amgarching
Anybody running 2.4.20 + with NVIDIA???

I am constantly having VM faults as
- kernel BUG at vmscan.c (many Oops in SYSLOG) Are you running the STANDARD 2.4.20, or RedHat 8.1 beta's 2.4.20+massive-patches kernel?

If you're trying to use RH 8.1 beta, then these VM errors are very likely caused by the new 2.5 TLS code that RedHat backported into their kernels. It is completely incompatible with the nVidia drivers (specifically, the NVIDIA_GLX package uses the IA32 fs and gs registers, and so does the TLS code, and they step on each other) as-is, and must be disabled at runtime (and you also need to patch the NVIDIA_kernel package to get it to compile properly). If you really want to know how, do a forum search for "phoebe howto" and you'll find the thread where someone explains it.

If you're seeing kernel OOPSes and BUGs in the standard 2.4.20 kernel (the one off www.kernel.org), then you very likely have dodgy hardware; my first test would be to run memtest86 at least overnight to stress the memory and CPU->memory bus.

I run 2.4.20 plus -preempt with no problems whatsoever; my current uptime is 12 days and some change, and I run UT, Unreal Gold, UT2003, Rune, Descent 3, zsnes, lbreakout2, and other SDL/OpenGL games all the time. Plus mplayer, you can't forget mplayer. ;)

Using driver 3123, with fast writes and SBA both on. I've got a Via KT333 chipset, and am using agpgart's AGP. The card is a GF4 Ti 4200 made by MSI.

My Athlon (XP 1800) also has the advanced speculative caching feature (it's not a bug, it makes stuff run faster almost all the time, it just runs into problems with Linux and AGP), but my kernel (every one since 2.4.19 has done this) disables it on bootup due to the problem with AGP.

[cjewell]

Ok, so do you guys think there may be something wrong with the way I compilied the driver (4191)? Again, I pointed the SYSINCLUDE to the include directory in the linux-2.4.20 source. I compiled in AGPART for SIS generic in the kernel - could this be causing conflict?

Chris

[bwkaz]

Are you trying to use "NvAgp" "1"? Compiling agpgart into the kernel will cause "NvAgp" "1" setups to fail (but not like that).

I'd still make sure it's a vanilla kernel, then check for buggy hardware, personally.

[Andy Mecham]

I agree with bwkaz.

Compiling in AGPGART is only recommended if you have a chipset known to work well with AGPGART. Otherwise, you should compile it as a module.

If your system passes stress testing with memtest86, please send details about your system and software to linux-bugs@nvidia.com.

--andy