Break out of an SDDM login loop after an NVIDIA DKMS update
Symptom, diagnosis and the diagnostic commands are sound. limine-mkinitcpio is real (shipped by limine-mkinitcpio-hook, listed in install/omarchy-other.packages) and SDDM is indeed the DM. Two substantive errors. (1) Driver package selection is wrong: install/hardware/nvidia.sh picks nvidia-open-dkms nvidia-utils lib32-nvidia-utils libva-nvidia-driver for GSP-capable (Turing+) GPUs and nvidia-580xx-dkms nvidia-580xx-utils lib32-nvidia-580xx-utils for pre-Turing. The record says 'nvidia-dkms for pre-Turing', which is wrong: pre-Turing needs the 580xx legacy branch. (2) sudo mkinitcpio -P followed by sudo limine-mkinitcpio is redundant: per the comment in bin/omarchy-hibernation-setup, limine-mkinitcpio 'rebuilds initramfs/UKI for all kernels and updates the' boot entries itself. Also pacman -S nvidia-open-dkms nvidia-utils without -u is a partial-upgrade pattern, and linux-lts-headers is pointless unless linux-lts is installed.
Regenerating initramfs/UKI incorrectly can leave the machine unbootable. Keep the LTS kernel entry and at least one snapshot available in the Limine menu before running mkinitcpio.
Symptom
After omarchy update and a reboot, SDDM accepts the password and then immediately bounces straight back to the login screen, forever. No error is shown. Switching to a TTY works. Some users see atomic drm request: failed to commit: invalid argument or Aquamarine errors when launching Hyprland by hand from the TTY.
Cause
The NVIDIA DKMS module failed to compile against the newly installed kernel (often a GCC version bump). mkinitcpio still "succeeded" and regenerated the image/UKI without the NVIDIA modules, while userspace nvidia-utils was upgraded. The kernel-module/userspace version mismatch makes Hyprland fail to get a DRM device, so the session dies instantly and SDDM re-prompts.
Fix
Drop to a TTY with Ctrl + Alt + F2 and confirm the mismatch:
cat /proc/driver/nvidia/version # kernel module version
pacman -Q nvidia-utils nvidia-580xx-utils 2>/dev/null # userspace version
dkms status
Rebuild the module and the boot image. On Omarchy, limine-mkinitcpio rebuilds the initramfs/UKI for every kernel AND regenerates the Limine entries, so it replaces mkinitcpio -P rather than following it:
sudo dkms autoinstall -k "$(uname -r)"
sudo limine-mkinitcpio
sudo reboot
If dkms autoinstall fails to build, read the build log, then reinstall the driver that matches your GPU generation. Which branch you need is not a guess. Omarchy's own detector tells you:
sudo cat /var/lib/dkms/nvidia*/*/build/make.log | tail -40
omarchy-hw-nvidia-gsp && echo "Turing or newer -> nvidia-open-dkms"
omarchy-hw-nvidia-without-gsp && echo "pre-Turing -> nvidia-580xx-dkms"
Install headers for the kernels you actually have, then the matching driver set (use -Syu, not -S, so you do not end up in a partial upgrade):
sudo env OMARCHY_ALLOW_DIRECT_PACMAN=1 pacman -Syu --needed linux-headers
# Turing (GTX 16xx / RTX 20xx) and newer:
sudo env OMARCHY_ALLOW_DIRECT_PACMAN=1 pacman -Syu --needed nvidia-open-dkms nvidia-utils lib32-nvidia-utils
# Pre-Turing (Maxwell / Pascal / Volta) - the 580xx legacy branch, NOT nvidia-dkms:
sudo env OMARCHY_ALLOW_DIRECT_PACMAN=1 pacman -Syu --needed nvidia-580xx-dkms nvidia-580xx-utils lib32-nvidia-580xx-utils
sudo limine-mkinitcpio
To get a working session right now without rebooting:
sudo systemctl stop sddm
sudo modprobe -r nvidia_drm nvidia_uvm nvidia_modeset nvidia
sudo modprobe nvidia_drm
sudo systemctl start sddm
If none of this works, reboot and select the pre-update snapshot in the Limine menu.
Verify
modinfo -F version nvidia matches pacman -Q nvidia-utils, and hyprctl version returns JSON after logging in.