Shutdown/reboot hangs on Wayland: nv_drm_atomic_commit "Flip event timeout on head 0" (610.43.02, RTX 3060)

Description

On every shutdown/reboot the machine frequently fails to power off completely (~19 out of 20 attempts). systemd finishes the entire shutdown (The system will power off now!), the screen loses signal and USB powers down, but the machine never reaches the final power-off — case fans and RGB stay on and I have to force-off with the power button. It is intermittent (~1 in 20 powers off correctly).

Note: this reproduces with both the open kernel module and the proprietary/closed module (610.43.02), so it may not be strictly specific to the open module — but I'm filing here since this repo is actively monitored. Same issue is tracked on the forum thread linked below.

System

Kernel log at shutdown (always the same pattern)

systemd-logind: The system will power off now!
nvidia-modeset: WARNING: GPU:0: Lost display notification (0:0x00000000); continuing.
[drm:nv_drm_atomic_commit [nvidia_drm]] *ERROR* [nvidia-drm] [GPU ID 0x00000b00] Flip event timeout on head 0
[drm:nv_drm_atomic_commit [nvidia_drm]] *ERROR* [nvidia-drm] [GPU ID 0x00000b00] Flip event timeout on head 0
[drm:nv_drm_atomic_commit [nvidia_drm]] *ERROR* [nvidia-drm] [GPU ID 0x00000b00] Flip event timeout on head 0

Data points that might help isolate the root cause

1. The hang only happens when shutting down from a running Wayland session. Booting to multi-user.target (no graphical session) and running poweroff powers off cleanly 3/3 times. So the trigger is the GPU teardown of the Wayland compositor (kwin_wayland) during shutdown, not the platform/firmware.

2. Setting nvidia_drm.fbdev=0 makes the "Flip event timeout" messages disappear completely, but the shutdown STILL hangs. This strongly suggests the flip timeout is a concurrent symptom, not the root cause — the actual hang is in the GPU teardown itself. At shutdown, kwin_wayland burns ~3 min of CPU and logs atomic commit failed: Permission denied before the hang.

3. Unloading the NVIDIA modules right before poweroff (a /usr/lib/systemd/system-shutdown script running modprobe -r nvidia_drm nvidia_modeset nvidia_uvm nvidia) does not help either.

Ruled out (none of these fixed it)

• BIOS update to latest + ErP Ready (S4+S5) enabled
• acpi=force / ACPI tweaks
• USB wakeup disabled for all controllers (/proc/acpi/wakeup)
• NVMe APST (nvme_core.default_ps_max_latency_us=0)
• KDE session restore disabled (loginMode=emptySession)
• NVreg_PreserveVideoMemoryAllocations=0
• Switching between the open and the proprietary kernel module (both hang identically)
• nvidia_drm.fbdev=0 (removes the timeout messages but not the hang)

Related

• Forum thread (same issue, reported across 545/555/570/590…): https://forums.developer.nvidia.com/t/545-29-06-18-1-flip-event-timeout-error-on-startup-shutdown-and-sometimes-suspend-wayland-unusable/274788/64?u=ayozetorres
• nvidia-bug-report.log.gz attached.

nvidia-bug-report.log.gz

Happy to test patches or gather more logs.

Start by reproducing shutdown from a running Wayland session and compare it with poweroff from multi-user.target. Read the attached nvidia-bug-report.log.gz alongside the nvidia_drm flip-timeout messages, kwin_wayland errors, and the tested system-shutdown module-unload script. Done means identifying and resolving the GPU teardown hang, not just suppressing the timeout logs.