[BUG] nvidia: suspend fails due to missing power management systemd services

[pyndys]

Full Raw Package Name (e.g. anda-0.4.14-1.fcrawhide.x86_64)

akmod-nvidia-610.43.02-1.fc44.x86_64 nvidia-driver-610.43.02-1.fc44.x86_64 nvidia-kmod-common-610.43.02-1.fc44.noarch

Release Version

f44

Describe the Bug

The nvidia package ships /usr/lib/modprobe.d/nvidia.conf with NVreg_PreserveVideoMemoryAllocations=1 set, but does not include the required systemd power management services (nvidia-suspend.service, nvidia-resume.service, nvidia-hibernate.service).

The issue is confirmed to be Terra-specific: switching to RPM Fusion drivers (which ship xorg-x11-drv-nvidia-power containing the missing services) fixes suspend completely.

To Reproduce

1. Install nvidia drivers from Terra on Fedora 44
2. Attempt to suspend the system (systemctl suspend or via DE)
3. System immediately wakes up after ~2 seconds

Expected Behavior

Either:

1. Ship nvidia-suspend.service, nvidia-resume.service, nvidia-hibernate.service (like RPM Fusion does via xorg-x11-drv-nvidia-power)
2. Or remove NVreg_PreserveVideoMemoryAllocations=1 from modprobe config if the required services are not provided

Relevant Log Output

NVRM: GPU 0000:01:00.0: PreserveVideoMemoryAllocations module parameter is set. System Power Management attempted without driver procfs suspend interface.
nvidia 0000:01:00.0: PM: pci_pm_suspend(): nv_pmops_suspend [nvidia] returns -5
nvidia 0000:01:00.0: PM: failed to suspend async: error -5
PM: Some devices failed to suspend, or early wake event detected
systemd-sleep: Failed to put system to sleep. System resumed again: Input/output error

[hamachitan]

🛑 The package akmod-nvidia-610.43.02-1.fc44.x86_64 nvidia-driver-610.43.02-1.fc44.x86_64 nvidia-kmod-common-610.43.02 cannot be found in terra44 [mg 404].

[pyndys]

My system (fastfetch):
pyndys@cv01
OS: Fedora Linux 44 (Forty Four) x86_64
Host: MS-7D46 (1.0)
Kernel: Linux 7.0.12-cachyos1.lto.fc44.x86_64
Uptime: 23 mins
Packages: 143 (nix-user), 1556 (rpm)
Shell: fish 4.6.0
Display (MSI30D4): 1920x1080 in 25", 120 Hz [External]
Display (1970W): 1366x768 in 19", 60 Hz [External]
WM: niri 26.04 (Wayland)
Icons: Colloid-Dark [GTK3/4]
Cursor: miku-cursor-linux (32px)
Terminal: ghostty 1.3.1-1.fc44
Terminal Font: JetBrainsMono Nerd Font (12pt)
CPU: 12th Gen Intel(R) Core(TM) i5-12400F (12) @ 4.40 GHz
GPU: NVIDIA GeForce RTX 3050 8GB [Discrete]
Memory: 3.97 GiB / 15.47 GiB (26%)
Swap: 0 B / 8.00 GiB (0%)
Disk (/): 68.89 GiB / 467.40 GiB (15%) - ext4
Local IP (wlp0s20f0u1): 192.168.31.5/24
Locale: en_US.UTF-8

[madonuko]

I've looked into the issue a bit, the service files in question do exist in nvidia-driver-580.spec but not in nvidia-driver.spec.

[GildedRoach]

I've looked into the issue a bit, the service files in question do exist in nvidia-driver-580.spec but not in nvidia-driver.spec.

I copied Negativo in removing these, I wonder now if I should not have...

[msmafra]

As @GildedRoach said and quoting Negativo17:"Kernel suspend notifiers can be enabled for the open modules , which allows the removal of all the SystemD units and scripts related to suspend and hibernate. Considering I’m shipping only open modules, this is for a nice simplification" . If the NVreg_UseKernelSuspendNotifiers is enabled those service may be disabled, masked etc.
Chapter 21. Configuring Power Management Support

BUT it seems, that for now in both branches, Recommended and New Feature, 595.84 (595 series in general) and 610.43.02/610.43.03 respectively to solve some suspend/sleep problems, probably having them can be helpful since some reported NVreg_UseKernelSuspendNotifiers with kernel 7.1 and 7.0 are causing problems with suspend/sleep.
I'm yet to do more tests on Fedora/OpenSuse my machine since I'm having problems with both methods on CachyOS.