Fedora 25 - NVIDIA Quadro M2000M - Display Driver 375.20 - Kernel updates not forcing recompile via DKMS?

Hello NVIDIA community,

my employer just provided me a new notebook, a Dell Precision. I configured the notebook with a NVIDIA Quadro M2000M graphics.

I am running Fedora 25 OS. As I digged around, I found out, that there is the possibility of using some RPM Fusion repo driver for the Xorg server (https://rpmfusion.org/Howto/nVidia). As there was a recent switch in Fedora from X11 to Wayland, I decided to stick with the solution to install the official NVIDIA display driver and disabling the nouveau driver.

I downloaded the recent matching NVIDIA driver package (375.20) here: http://www.nvidia.de/download/driverResults.aspx/111627/. Then, I followed a procedure described here: https://www.if-not-true-then-false.com/2015/fedora-nvidia-guide/. This included booting Fedora in runlevel 3, disabling the nouveau driver, etc.

This was the best guide I found so far. Anyway, I had to add two seperate steps:

  1. Disabling secure boot while running the NVIDIA installer (due to a kernel error)
  2. After rebooting, I had to downgrade the “mutter” package from 3.22.2 to 3.22.1 (the most recent package caused a black screen after reboot, informations found here: https://www.reddit.com/r/archlinux/comments/5ce6a6/black_screen_on_gnome_login_and_xorgwayland_switch/)

So, after performing all those steps, I was able to startup Fedora running with the Quadro M2000M graphics. Calling the NVIDIA settings dialog (from shell: nvidia-settings), all informations there seem to prove a stable running NVIDIA graphics.

So, today, there was a Fedora update, including an upgrade of the kernel from 4.8.10 to 4.8.11. I downloaded the updates and rebooted my system and I ran into the same black screen problem. I was able to solve this problem by downloading the NVIDIA driver package again via wget from the shell in runlevel 3 and then running the installer again.

I suppose, this is not the way it is intended to work, as I registered the driver with DKMS and therefore I assume, the driver should have been recompiled for the new kernel. And I fairly doubt, I have to run this procedure over and over again, everytime there is a kernel update in my distribution.

Hence, this points toward my question to the community: What am I missing here? Why wasn’t the driver kernel module recompiled via DKMS? Is there anything in the nvidia-* command list, that provides some functionality regarding this problem?

After booting the new kernel, but before recompiling the driver, there were (obviously) error messages in the kernel ring buffer which I was able to grep throug dmesg output. After recompiling, there seems everything to be ok.

I appreciate any help.

Best regards
ne0h