Wrong VM unbinding
Philippe wrote a IGT test to trigger the issue, see #77 (comment 1513878)
##################
During the tear-down of a Vulkan sample renderheadless, Xe throws a "Engine memory cat error:" and then several errors and leave driver in a state that no other application can use it, not even unload the driver works. Requiring a hard-reboot.
The operation that causes this is a VM unbind, that is pointed in the renderheadless_logs.txt in the line with <hangs for a few seconds here. this is the point that dmesg errors shows up>
.
Here the kernel logs: dmesg.txt
I'm reviewing what is possible wrong in user-space but anyways Xe driver should properly reset in such cases and allow other applications to run.
Kernel version: commit e996a2d6 Merge branch 'tip-xe-wa' into 'xe'