Systemd service nvidia-cdi-refresh.service fails to identify kernel module

[ohadlb]

Similar to #1395 I think. I'll point out in advance that I don't recall how I've installed everything and I am using the Debian Testing version (currently Forky), which results in a less stable system.

I'm using the latest release of the container toolkit:

$ nvidia-container-toolkit --version
NVIDIA Container Runtime Hook version 1.18.1
commit: efe99418ef87500dbe059cadc9ab418b2815b9d5

where the regex condition is /(nvidia|nvidia-current)\.ko[:]. In my case I have

$ grep -E '/(nvidia|nvidia-current)' /lib/modules/$(uname -r)/modules.dep
kernel/drivers/net/ethernet/nvidia/forcedeth.ko.xz:
kernel/drivers/platform/x86/nvidia-wmi-ec-backlight.ko.xz: kernel/drivers/acpi/video.ko.xz kernel/drivers/platform/x86/wmi.ko.xz
updates/dkms/nvidia-modeset.ko.xz: updates/dkms/nvidia.ko.xz kernel/drivers/gpu/drm/drm.ko.xz kernel/drivers/acpi/video.ko.xz kernel/drivers/platform/x86/wmi.ko.xz
updates/dkms/nvidia-peermem.ko.xz:
updates/dkms/nvidia-drm.ko.xz: updates/dkms/nvidia-modeset.ko.xz updates/dkms/nvidia.ko.xz kernel/drivers/gpu/drm/drm_ttm_helper.ko.xz kernel/drivers/gpu/drm/ttm/ttm.ko.xz kernel/drivers/gpu/drm/clients/drm_client_lib.ko.xz kernel/drivers/gpu/drm/drm_kms_helper.ko.xz kernel/drivers/gpu/drm/drm.ko.xz kernel/drivers/acpi/video.ko.xz kernel/drivers/platform/x86/wmi.ko.xz
updates/dkms/nvidia-uvm.ko.xz: updates/dkms/nvidia.ko.xz kernel/drivers/gpu/drm/drm.ko.xz
updates/dkms/nvidia.ko.xz: kernel/drivers/gpu/drm/drm.ko.xz

The last module looks like the required match, except that the colon does not immediately follow the .ko suffix.

Calling nvidia-ctk cdi generate manually and chmodding the output files works perfectly, but has to be repeated on every boot.

[zeroepoch]

I'm seeing this same problem on Fedora 42 and for the same reason as @ohadlb. It's nvidia.ko.xz: rather than nvidia.ko:. I see @elezar made this change in 1f5ce73. Any reason we need the colon now? Is it to ensure the primary nvidia module is available and not check for others? If so then .xz (and maybe other extensions like .gz) needs to be optional.

[EdSwarthout]

I see the same issue on Ubuntu which uses zstd compression.
Here is a pull request to support any suffix: #1518

[elezar]

Any reason we need the colon now? Is it to ensure the primary nvidia module is available and not check for others?

The intent of the colon was to check for the primary module itself and not trigger on modules that depend on the primary module. This was probably not needed, since it's unlikely that a module with an nvidia dependency is present without the nvidia module also being present.