GPU falls of the bus then the kernel deadlocks in nvidia-modeset

When the GPU crashes by falling of the bus, the driver doesn’t cleanup properly user processes which are using the Nvidia GPU, so the last process which is using the GPU is getting stuck in the kernel forever leaving the system in a inconsistent state where I can’t do a clean shutdown/reboot, kill this process (even with kill -9) or unload the Nvidia modules.

GPU: GTX 1060
Kernel: 6.18.9-arch1-2 (Arch Linux)
Nvidia drivers: 580.126.18

nvidia-bug-report.log.gz (1.4 MB)


Any updates?

the NV driver throwing a tantrum after Xid 79, have been a known issue for at least like a decade: just search this forum.
The only workaround is to not make a mistake of buying any NV anymore in the future.

Regarding Xid 79, in the words of an NV eng:

The main issue is not about GPU falling off the bus, but about the behavior of the driver when a GPU lost occurs.

yes, we all known this. As I wrote, it’s like at least a decade old problem.