Issues with loading and unloading nvidia driver modules

[Borodin Nikolay]

These problems also affect older versions of NVidia drivers.

Issue #1. The NVidia kernel driver doesn't have time to load and Xorg starts before the driver is loaded. A further restart of the X server loads the driver correctly. This problem may or may not appear depending on whether the driver has enough time to boot.

Solution: The package post-install script must add the following lines to /etc/modules and to /etc/modules.load.d/modules.conf respectively.

nvidia
nvidia-drm
nvidia-modeset

Issue #2. When the system shuts down or reboots. Warnings are thrown into the console.

This is a driver issue, which has already been reported in the open-source version of the NVidia drver.

https://github.com/NVIDIA/open-gpu-kernel-modules/issues/289

[ 281.772822] sd-umoun[3293]: Failed to unmount /oldroot: Device or resource busy
[ 281.774041] sd-umoun[3294]: Failed to unmount /oldroot/sys: Device or resource busy
[ 281.775539] shutdown[1]: Failed to finalize file systems, ignoreing.

Until this is fixed, the following workaround should be added for convenience.

1. Create file /usr/lib/systemd/system-shutdown/nvidia.shutdown and make it executable.
2. The file should have the following content.

#!/bin/sh
# Remove all Nvidia modules on shutdown
# to avoid errors like
# [35509.312925] sd-umoun[56020]: Failed to unmount /oldroot: Device or resource busy
# [35509.317505] sd-umoun[56021]: Failed to unmount /oldroot/sys: Device or resource busy
# [35509.322349] shutdown[1]: Failed to finalize file systems, ignoring.
#
for MODULE in nvidia_drm nvidia_modeset nvidia_uvm nvidia
   do
# we will have to check if the modules are still loaded because they aren't always
       if lsmod | grep "$MODULE" &> /dev/null ; then
          rmmod $MODULE
       fi
   done

OS: Ubuntu 20.04 (Focal Fossa).