Problem with nvidia driver on x86_64 2.6.5 kernel

[fdm]

Hello,

I've got a problem with the NVIDIA_kernel-1.0-5332 driver on my x86_64 platform using kernel 2.6.5. After applying the patch from www.minion.de (NVIDIA_kernel-1.0-5332-2.6.diff) the kernel module gets build and installed, but modprobe complains the module doesn't exist.
I noticed that the kernel modules in 2.6 now have the extension .ko instead of .o, so I tried renaming the module. This however results in the following error message from modprobe:
FATAL: Error inserting nvidia (/lib/modules/2.6.5/kernel/drivers/video/nvidia.ko): Invalid module format.

Does anyone have an idea how to fix this?

Frans

[molecular]

I'm using 5336 on ia32 2.6.5.
it's a gentoo-system and I installed the driver via the gentoo package system "portage". I looked at the ebuild-file (used to describe how to install sth) to see what it would do to get the .ko-file. Indeed it just copies the .o-file as you suspected.
Maybe this helps a little. Probably not... posted anyway.

[Corona688]

I have heard the problem is related to the 2.6.5 kernel using 8K stacks as opposed to the 4K ones the nvidia driver expects..

[edit] OK I had it backwards. See here. Not sure if this is related though.

[tamran]

I've got a problem with the NVIDIA_kernel-1.0-5332 driver on my x86_64 platform using kernel 2.6.5. After applying the patch from www.minion.de (NVIDIA_kernel-1.0-5332-2.6.diff) the kernel module gets build and installed, but modprobe complains the module doesn't exist.

is there a symbolic link from your current kernel (/usr/src/kernel-source) to /usr/src/linux ?

Not having that link would cause what you described. To fix, use:

cd /usr/src/ && rm linux && ln -s linux-2.6.5 linux

Hope this helps,

Tamran

[fdm]

Hello,

First of all, thanks for all your replies.
Unfortunately, the problem is still there.

I've tried recompiling with the /usr/src/linux link pointing to linux-2.6.5, as tamran suggested, but this doesn't change the situation.

I don't know if my 2.6.5 kernel is using 4k stacks. From the link given by Corona668, I make up that it will be an experimental feature in the 2.6.6 kernels with a default of 'no'. I've checked if there is such an option in menuconfig for 2.6.5, but found none.

I think the problem is related to the way the module gets compiled and linked. Below you'll find:
-the compiler options used by the makefile for the kernel module;
-gcc version info
-ld version info

Compiler options:
-Wall -Wimplicit -Wreturn-type -Wswitch -Wformat -Wchar-subscripts -Wparentheses -Wpointer-arith -Wno-multichar -Werror -O -mno-red-zone -MD -D__KERNEL__ -DMODULE -D_LOOSE_KERNEL_NAMES -mcmodel=kernel -DNTRM -D_GNU_SOURCE -D_LOOSE_KERNEL_NAMES -D__KERNEL__ -DMODULE -DNV_MAJOR_VERSION=1 -DNV_MINOR_VERSION=0 -DNV_PATCHLEVEL=5332 -DNV_UNIX -DNV_LINUX -DNV_INT64_OK -DNV_64_BITS -DNVCPU_X86_64 -DREMAP_PAGE_RANGE_5 -I. -I/lib/modules/2.6.5/build/include -Wno-cast-qual -Wno-error

gcc version info:
Reading specs from /usr/lib64/gcc-lib/x86_64-suse-linux/3.3.1/specs
Configured with: ../configure --enable-threads=posix --prefix=/usr --with-local-prefix=/usr/local --infodir=/usr/share/info --mandir=/usr/share/man --libdir=/usr/lib64 --enable-languages=c,c++,f77,objc,java,ada --disable-checking --enable-libgcj --with-gxx-include-dir=/usr/include/g++ --with-slibdir=/lib64 --with-system-zlib --enable-shared --enable-__cxa_atexit x86_64-suse-linux
Thread model: posix
gcc version 3.3.1 (SuSE Linux)

ld version info:
GNU ld version 2.14.90.0.5 20030722 (SuSE Linux)

Perhaps this info holds the key?

Thanks again,

Frans

[Phorem]

Try this prebuilt one instead.......

http://www.sh.nu/download/nvidia/lin...-5332-pkg0.run

It seems to work fine when the Nvidia official ones fail.

[zander]

The "Invalid module format" insmod error message is typically accompanied by a bitter complaint from the kernel; check the output of 'dmesg', chances are it's a module version magic mismatch of some sort (e.g. mismatched compilers). FYI, there's no 4KSTACKS patch for x86_64, though it puts similar constraints on kernel stack usage (8K stacks, 64bit kernel).

[mikygee]

I know it's not a very good answer but the nvidia drivers work with 2.6.4, so why not sticking with this kernel ?

bye

[tamran]

Interesting indeed, I had an issue with 2.6.5_rc2 and rc3 kernels. Nothing I did would make the nvidia drivers work. This was not the case with 2.6.5. Go figure?

Tamran

[fdm]

Thanks Phorem, the modified installer you suggested works. I had a quick look and it seems that that the module makefile has been changed in some ways.

I noticed that the installer only provides 64 bit libGL* files in the /usr/lib64 and /usr/lib64/tls directories. The mesa files under /usr/lib have not been replaced. Will this be a problem?
Could it be the reason why glxgears is running with approx. 6000 FPS, but the atlantis screen save only has 2.5?

Frans

Edit: Hmmm, the UT2004 demo for AMD64 crashes my machine. Could be related to the wrong GL lib files, l'll keep you informed.

[fdm]

Originally Posted by zander

The "Invalid module format" insmod error message is typically accompanied by a bitter complaint from the kernel;

The message in dmesg was: No module found in object

Thanks for you help,

Frans

[fdm]

Originally Posted by mikygee

I know it's not a very good answer but the nvidia drivers work with 2.6.4, so why not sticking with this kernel ?

Well, I started with the 2.6.4 kernel, but unfortunately it had a problem with missing symbols in the kernel. After some searching I learned that this is related to some mysterious linker problem that hasn't been solved yet, but for which a work-around exists. The next day 2.6.5 got released, already including the work-around I needed.
With the proper patch for 2.6.4, I could go back, however with Phorem's solution available, I'd rather stick with 2.6.5.

Thanks for your suggestion,

Frans