Problems installing rpmfusion Nvidia drivers

[tilt-x]

Hi all,

I’ve been trying to install the Nvidia drivers from rpmfusion and just can’t get them to work. I’m running Rocky 9.3 and have followed this post:

dnf update -y
reboot
dnf install akmod-nvidia

I have waited a while until the installation process seemed completed (looking at “top”). After a reboot, the nvidia driver didn’t work though, there were messages about falling back to nouveau.

# rpm -q akmod-nvidia kernel
akmod-nvidia-545.29.06-1.el9.x86_64
kernel-5.14.0-362.18.1.el9_3.x86_64
# uname -r
5.14.0-362.18.1.el9_3.x86_64

I have read in another post that one could try running “akmods --force” which I did but that didn’t solve the problem.

I’ve tried to load the module manually and this is where it gets interesting:

# modprobe nvidia
modprobe: FATAL: Module nvidia not found in directory /lib/modules/5.14.0-362.18.1.el9_3.x86_64

The module isn’t in that directory but it can be found in another location: /lib/modules/5.14.0-362.8.1.el9_3.x86_64/extra/nvidia/nvidia.ko.xz

This is from a slightly older kernel version. The directory for the kernel I’m running doesn’t have the extra/nvidia subdirectories at all. I’ve moved that directory somewhere else, uninstalled that older kernel package and ran “akmods --force” again in case it just ended up in the wrong version, but that didn’t result in any changes I could see. Is there more logging output available than just the result of akmods which is:

Checking kmods exist for 5.14.0-362.18.1.el9_3.x86_64 [  OK  ]
Building and installing nvidia-kmod [  OK  ]

Where did the module go? Should I try renaming the modules folder from the old kernel version to the new one in case these are compatible? Could it be a problem with the nouveau driver being around? It’s blacklisted on the boot command line (rd.driver.blacklist=nouveau modprobe.blacklist=nouveau) as far as I can see, but it seems to be loaded according to lsmod.

[dali]

Check the logs in /var/log/akmods, they might contain a clue to what’s happening.

[tilt-x]

Thank you! Your hint eventually led me to the solution. Long story short: dnf thought kmod-nvidia was already installed so akmods --force didn’t finish (but also didn’t display an error).

Step by step solution in case somebody else stumbles upon this problem:

The akmods log file just contained

2024/02/09 10:20:37 akmods: Building RPM using the command '/sbin/akmodsbuild --kernels 5.14.0-362.18.1.el9_3.x86_64 /usr/src/akmods/nvidia-kmod.latest'
2024/02/09 10:21:26 akmods: Installing newly built rpms
2024/02/09 10:21:26 akmods: DNF detected
2024/02/09 10:22:46 akmods: Successful.

But by looking at the akmods script itself I found out that the generated rpm package ended up in /var/cache/akmods/nvidia/ alongside a log file. And that one contained:

Package kmod-nvidia-5.14.0-362.el9_3-3:545.29.06-1.el9.x86_64 is already installed.
Dependencies resolved.
Nothing to do.

RPM also showed me that kmod-nvidia was installed, but DNF wasn’t able to remove it using “remove kmod-nvidia”. Only when I told it to remove kmod-nvidia-5.14.0… (the full version string) it removed the package which it claimed was installed from rpmfusion.

I know I had installed both kmod-nvidia and akmod-nvidia, removed them, reinstalled them, during my efforts to get the Nvidia driver going (and I didn’t know the difference between kmod and akmod).

Eventually, once the kmod-nvidia package was gone, akmods --force rebuilt and installed the right package and now I have the nvidia kernel module in /lib/modules/$(uname -r) and the driver was loaded after a reboot.