Proprietary NVIDIA preventing upgrade to 9.4

Hi,
I’m upgrading from CentOS 7.9 to AlmaLinux 9.4. So far I have completed phase 1, and I’m currently running AlmaLinux 8.10. However, when I run leapp preupgrade, my only inhibitor is Proprietary NVIDIA driver detected, and it suggest uninstalling NVIDIA driver and switching back to nouveau. I uninstall NVIDIA with the run file used to install, but after rebooting the system, it continues to find and load the NVIDIA driver preventing me from completing phase 2. I have manually tried to uninstall the driver, removing any remnant files and modules, but it continues to find this particular driver version. Even if I upgrade to a higher NVIDIA version, nvidia-smi shows the new driver, but when I reboot, it finds the older version. If I can somehow get it to not find an NVIDIA and revert back to nouveau maybe I can finish the upgrade. Any help would be appreciated.

Thanks!

For years, the use of RPM packages has been recommended over the run file. This is one of the reasons why.

The driver is placed into two places: into directories, where kernel and X11 do look for it,
and into the initramfs images (in /boot/) from which kernel actually loads it from during boot.

The uninstall should recreate the initramfs image.


That is interesting. When you have booted, one version of the driver has been loaded into memory. If you update NVidia’s tools and driver, you do replace the files on disk, not the driver in memory. On that situation I usually get “version mismatch” from the nvidia-smi. Well, perhaps the run file does unload-load or something.


Anyway, while running CentOS 7, I would check all the command-line parameters of the “run file” installer to repeat the uninstallation. Perhaps rmmod nvidia to unload the driver from memory too, and then tell the ‘dracut’ to recreate the initramfs. (Although I have never learned how to do that.)

Does CentOS 7 have lsinitrd? That lists files in the image. If there is no “nvidia” anything, then things ought to be better.


Personally, I have never done the in-place conversion of one major version to another.
I really prefer fresh installs. Configs can be reproduced and user data transferred (or restored from backups).

Thanks @jlehtone for the reply. I do get NVML type mismatch on nvidia-smi when installing a newer driver. However once I unload the modules (nvidia_modeset, nvidia_drm, nvidia) nvidia-smi works with the newer driver installed until I reboot and it loads the previous nvidia drivers again. I can unload the nvidia drivers (init 3, rmmod nvidia*) and successfully run leap preupgrade and upgrade without errors but the upgrade will not start. The upgrade initramfs entry is created for the upgrade but the upgrade never appears in the menu and kernel 4.18.0-553.27.1.el8_10.x86_64 loads. The upgrade never starts and I’m back to square 1 with the old nvidia driver being loaded.

The common theme there seems to be that
the NVidia installer does not update initramfs
and the leapp does not update the initramfs.

Is there some issue in writing to /boot?

check if /boot has enough space.
then recreate the initramfs

e.g.
setenforce 0
dracut -f

then reboot