Screwed up something working on GPU passthrough

ajzride

I finally got my FirePro S7150x2 working, but was getting absolutely terrible performance on Windows machines (>80% GPU at idle), so I was attempting to remove the mxGPU drives and passthrough the whole card to see if that would resolve the issue.

I ran the commands:

rmmod gim
yum remove mxgpu-4.19.0+1-modules.x86_64

Which for sure removed the MxGPU drivers, but now I can't see the AMD card at all, it doesn't even show up in lspci (only the onboard Matrox shows up).

Running
mod probe GIM
returns
FATAL: Module GIM not found.

Should GIM still be there? Am I chasing ghosts with the card not showing up in lspci? I did not open the case after removing mxgpu, so I don't know why the card would disappear from lspci.

I tried doing an in-place upgrade but that did not restore the ability to see the card in lspci.

ajzride

@ajzride

Furthermore, attempting to re-install mxGPU gives the following result:

[15:41 ajzXCP-NG01 tmp]# xe-install-supplemental-pack mxgpu-2.0.0.amd.iso
Error code: SR_BACKEND_FAILURE_78
Error parameters: , VDI Creation failed [opterr=Command ['/sbin/lvremove', '-f', '/dev/VG_XenStorage-e07d2043-7496-7ddd-c2fa-b9ab74a04002/VHD-e84b446b-ce81-4af5-ab88-679304dfe750'] failed (Couldn't find device with uuid kaAIxv-4Qo2-IfKO-2hx1-aAm0-GWjx-GlkCPR.
  Failed to find logical volume "VG_XenStorage-e07d2043-7496-7ddd-c2fa-b9ab74a04002/VHD-e84b446b-ce81-4af5-ab88-679304dfe750"): Input/output error],

planedrop

Maybe try removing the GPU, booting, then shutting down and adding the GPU back and see if it shows up?

I'm not quite as familiar with the intricacies of setting up vGPU so don't really know how it would have broken something, but I do know my way around GPU passthrough and as long as it shows up in lspci it should be really easy to do.

Not that you don't already know this, but first step is getting it to show up in lspci then things should be easy.

ajzride

@planedrop

I had to wipe and reinstall XCP, never did figure it out. Don’t have time to troubleshoot any further.

planedrop

@ajzride Gotcha, at least restoring is pretty easy.

Passthrough is pretty straightforward, just be sure to passthrough the entire device including audio sub devices and it should "just work", been very reliable for me.