The other day my unRaid Sandisk USB died, and I used the MyServers plugin to restore onto a Samsung USB instead. Everything basically returned to its working state - but VM GPU passthrough. The error logs imply that the GPU is being utilized - it's not(or at the very least it shouldn't be). The only other process I use the GPU for is Tdarr, and I keep it in a off-state majority of the time. I even disabled Docker, rebooted with Docker in an Off state, and still the GPU passthrough is being "used". For whatever reason the VMs think it's being utilized? I re-enable docker, and turn on Tdarr manually to see if it'll start encoding my library - and yep, the GPU is working inside the Docker container, and it's being done by NVenc, not cpu.

 

I also tried making a new Win10 VM. Without the GPU passthrough first, and it worked fine. I installed Splashtop. Then I shutdown the VM and tried to do passthrough. But it had same problem as my pre-existing GPU-passthrough VM. Tried changing PCIe ACS override from Downstream to Both with no appreciable difference.

 

For the curious, this is basically the log flood:

Quote

[timestamp] qemu-system-x86_64: vfio_region_write(0000:01:00.0:region1+0xb41a9, 0x0,1) failed: Device or resource busy

 

 

Not really putting this in the bug section in MyServers plugin, because, for all I know I'm missing something very obvious. Any help would be appreciated. 

On 2/19/2022 at 12:03 AM, ecx said:

The other day my unRaid Sandisk USB died, and I used the MyServers plugin to restore onto a Samsung USB instead

Probably in use by efifb, maybe before you were booting unraid in legacy and now you are booting in uefi?

Search in this section of the forum, you will find a lot of discussions relating vfio_region_write, BAR X: can't reserve mem, video=efifb:off, and similar.

 

On 2/19/2022 at 12:03 AM, ecx said:

The error logs imply that the GPU is being utilized - it's not

It is, trust it