[elrepo-devel] kmod-nvidia.el6 question

[Phil Perry]

On 09/02/11 20:27, Steve Cleveland wrote:
> When I initially tested the nvidia-x11-drv and kmod-nvidia package with
> rhel6, it was working great.  But now I've run into a bit of a strange
> issue.
>
> My system build environment is based on the latest version of every
> package (using mrepo to download RHN updates and adding a repo option to
> ks.cfg to look at that repository during install).
>
> So for kernel, it is installing the latest available kernel only
> (currently 2.6.32-71.14.1.el6.x86_64).
>
> With that package in place, if I install nvidia-x11-drv and kmod-nvidia,
> I'm not able to load the nvidia module:
>


Hi Steve,

This is a known issue, although I'm not exactly sure of the cause.

The "issue" arises when installing kmod-nvidia from a kickstart. If you 
set up your machine with kickstart and then install kmod-nvidia manually 
afterwards, I'm guessing all works fine. The other solution should 
simply be to uninstall and reinstall kmod-nvidia.

Here is another instance of a user having troubles installing from a 
kickstart:

http://elrepo.org/bugs/view.php?id=98

I'd be particularly interested to see your /etc/xorg.conf and 
/boot/grub/grub.conf files immediately after a failed kickstart install, 
please.


> # modprobe nvidia
> FATAL: Error inserting nvidia
> (/lib/modules/2.6.32-71.14.1.el6.x86_64/weak-updates/nvidia/nvidia.ko):
> No such device
>
> But the file does exist:
>
> # ll /lib/modules/2.6.32-71.14.1.el6.x86_64/weak-updates/nvidia/nvidia.ko
> lrwxrwxrwx. 1 root root 56 Feb  9 11:09
> /lib/modules/2.6.32-71.14.1.el6.x86_64/weak-updates/nvidia/nvidia.ko ->
> /lib/modules/2.6.32-71.el6.x86_64/extra/nvidia/nvidia.ko
>
> # ll /lib/modules/2.6.32-71.el6.x86_64/extra/nvidia/nvidia.ko
> -rw-r--r--. 1 root root 14845520 Jan 21 11:35
> /lib/modules/2.6.32-71.el6.x86_64/extra/nvidia/nvidia.ko
>

Yes, that all looks as expected

>   From here, I can install the old kernel (2.6.32-71.el6.x86_64) and boot
> off of that.  The module loads fine.  But even with the old kernel in
> place, the newer kernel cannot load the module.

Yes, that's not right. Clearly the first install with 
kernel-2.6.32-71.14.1.el6.x86_64 has failed and only 
uninstall/reinstalling is going to fix it. As I said above, I suspect 
reinstalling either the kernel (as you've demonstrated) or kmod-nvidia 
would have the same effect.

> If I then remove the
> newer kernel, reinstall it, and then boot off of it, the module loads fine.
>
> Is there a piece in the stock kernel that is needed by kmod-nvidia to
> work correctly?  I'm not terribly familiar with how weak kernel modules
> work.  I get the impression that during the update process, modules get
> linked or copied over to the newer kernels.  But by having the newer
> kernel be the first one loaded, does that break what the kmod module
> does?  Is there a solution to get this to work on whatever the most
> current kernel version is?
>

No, there's no magic involved here - in theory the module should just 
work with any kernel and you certainly don't need any particular kernel 
to be installed on the system.

Regards,

Phil