Hello,

I am doing software testing on Intel GPU and i915 driver. I am using Intel NUC 9 CoffeeLake, Ubuntu 20.04 LTS and using Intel GVT-g to passthrough iGPU to VM and host.

As during testing, something can crash, I wish to reset the i915 driver and GPU. For this I do the following steps:
1. Remove the vGPU
2. Remove kvmgt
3. stop the X server
4. Unbind the GPU
5. Remove the i915
6. Load the i915 driver (This loads the GPU back as well)
7. start the X server
8. Load kvmgt and create vGPU

Please note: During this process, no VM had been spawned so vGPU was free.

In Kernel 5.4.0-59, I had a non-deterministic issue with modprobe i915 leading to a kernel bug as follows:

[53021.184248] BUG: unable to handle page fault for address: 00000000000fe248
[53021.184251] #PF: supervisor read access in kernel mode
[53021.184252] #PF: error_code(0x0000) - not-present page


In Kernel 5.11.0-rc7 I am having a deterministic issue with the step unbinding the GPU. It leads to "segmentation fault" and the dmesg logs is as shown:

[ 133.006604] general protection fault, probably for non-canonical address 0x3b5a596b612093f7: 0000 [#1] SMP NOPTI
[ 133.006618] CPU: 13 PID: 1920 Comm: bash Tainted: G U 5.11.0-051100rc7-generic #202102072330
[ 133.006630] Hardware name: Intel(R) Client Systems NUC9i9QNX/NUC9i9QNB, BIOS QXCFL579.0044.2020.0617.1537 06/17/2020
[ 133.006638] RIP: 0010:snd_soc_tplg_component_remove+0x8f/0x290 [snd_soc_core]

Any help is appreciated!

Also, Is there any other way as well that I could reset the i915 driver and reset the GPU without rebooting the machine?