[Nouveau] Regression: Kernel/GPU crash with "Asynchronous wait on fence"

[Owen T. Heisler]

Hi,

I am experiencing a kernel/GPU crash once every few days on an Nvidia 
Optimus system with a secondary display connected to an Nvidia output. 
The secondary display turns off suddenly, X freezes, and in most cases 
the kernel hangs.

Module parameter `nouveau.config=NvClkMode=15` is in use, but I get the 
same behavior without it.

I have captured a variety of log data, but I find these two errors 
consistently:

- Asynchronous wait on fence nouveau:systemd-logind
- nouveau 0000:01:00.0: tmr: stalled at ffffffffffffffff
- Fixing recursive fault but reboot is needed!

This is a regression; nouveau was stable with Debian 10 buster (Linux 
v4.19). The crashes started after upgrading to Debian 11 bullseye. I 
have tested Linux v4.14.290, v4.19.255, and the latest nouveau-next 
commit 9622bcb7c72b230d64b7f7d2f9505e17214f3597; all exhibit the same 
behavior (with some variation in log output).

Is a userspace change causing a kernel crash? Do I need to try different 
versions of libdrm and xf86-video-nouveau userspace components?

I posted more information and log data at:
<https://gitlab.freedesktop.org/drm/nouveau/-/issues/180>

Thanks,
Owen