Hello All,

For the last three Unraid upgrades, I have not been able to reboot the system cleanly. Each time, the reboot process gets stuck at “Shutting down array” indefinitely, and after waiting 15 to 20 minutes I eventually have to power cycle the machine manually.

Today, after upgrading to 7.3.1, I noticed it was again stuck at “Shutting Down Services - Shutting down array.” This time I was still able to log into the dashboard and collect some logs. However, as soon as I clicked on the VMs tab, the entire dashboard froze and became completely unresponsive. I checked again about 10 minutes later and the page was still frozen with no response at all.

And no SSH Terminal open, so that wasn't the cause.

Can someone please help in identifying what's causing this? I had no changes, didn't install any new plugins or apps.

I could not generate the diagnostics but did catch some logs within System Log section.

syslog.txt

Disable Docker and VM service and reboot in safe mode, start the array and see if it stops, if yes, retest again after booting in normal mode to rule out a plugin issue, leave the services still disabled, if still OK enable one of the services and retest, then the other, note that if the issue happens with the docker service enabled only, you may need to try enabling one or a few containers at a time to find the culprit.

Just to add this happens with the docker disabled as well. It just won't stop the array to properly reboot on it's own. I haven't tried disabling VM service, and I suspect it has something to do with it since the dashboard will completely lock up when I click on "VMS" tab but only while it's this crippled state of stopping the array.

Just for others out there if they have the same issue:
I was finally able to capture enough logs and feed them thru the AI:

06:58:09–10 — The host started resetting the passed-through GPU via VFIO (the vfio-pci 0000:01:00.0 resets), which is the normal process of returning a GPU from the VM back to the host.

06:58:11 — The Nvidia driver started reloading on the host side to reclaim the GPU.

06:58:13 — Kernel crash. A NULL pointer dereference occurred inside the Nvidia driver at nv_audio_dynamic_power, triggered by the qemu-event process (PID 18970). The crash note confirms: qemu-event[18970] exited with irqs disabled — a very bad state indicating the driver died mid-operation.

The root cause is a bug in the Nvidia driver's GPU audio power management code during the VFIO passthrough teardown/reclaim cycle. When the GPU was handed back from the VM to the host, the Nvidia kernel module tried to manage the HDMI audio component's power state (nv_audio_dynamic_power) and dereferenced a null pointer — likely because the audio device's internal state was left in an inconsistent condition by the VFIO reset.

1. Add video=efifb:off to the Unraid boot parameters to prevent the host from trying to use the GPU's framebuffer at all

2. Use the vendor-reset kernel module, which does a more thorough GPU reset on passthrough teardown