January 14, 20197 yr I am new to unRaid so I apologize if I have made a newbie mistake. Here's what I have done so far - Created array and setup docker services for Minecraft and Plex (success) - Attempted to create Windows 10 VM with passthrough to GPU (failed) - Error message: internal error: Unknown PCI header type '127' - Researched online and found that vbios file may be needed (not resolved issue) - Tried various combinations of bios: SEABIOS/OVFM and machine_type: Q35/i440fx (not resolved issue) - Tried spinning without GPU as a VNC (successfully spins but does not meet requirements) My specs: - Ryzen Threadripper 1900X - ASUS X399 Zenith Extreme - 2x XFX Radeon 580 - unRaid v6.6.6 Edited January 14, 20197 yr by littleman11186
January 14, 20197 yr Author Is that a setting or do I have to reboot unraid and choose it during startup?
January 14, 20197 yr yea uncheck it and reboot then try again. Also i would add "kvm.ignore_msrs=1" to the syslinux config. Edited January 14, 20197 yr by david279
January 14, 20197 yr Author The instance appears to spin but does not appear either screen, just the command line unraid login.
January 14, 20197 yr Author Error in the log: 2019-01-14T19:03:45.819066Z qemu-system-x86_64: vfio_err_notifier_handler(0000:09:00.0) Unrecoverable error detected. Please collect any data possible and then kill the guest
January 14, 20197 yr Author I have killed that Windows VM and created a new one, now back to the original error
January 14, 20197 yr Must be something with the amd cards. I have two nvidia cards and both passthru without issues. I asked about the uefi mode because i have had issues passing thru my cards in that mode. Have you tried using a vbios file with the cards?
January 14, 20197 yr Author Yes but I think the PCI header error got in front of my results there. I remember reading that the passthrough can fail after the first attempt, and reboot is required. I'm going to re-try with vbios file after another reboot.
January 14, 20197 yr Author Been through a couple cycles of fully rebooting after the initial VM creation. It does appear that I cannot start even the same VM after initial setup without encountering the PCI header error. Here's what I'm getting from the VM's that do start up: vfio_err_notifier_handler(0000:09:00.0) Unrecoverable error detected. Please collect any data possible and then kill the guest
January 14, 20197 yr A couple of things, I don't have AMD or Threadripper myself, but some things to keep in mind, GPU passthrough usually requires UEFI boot process... That means for the host you should not enable "Legacy Boot", since turning that on disables UEFI boot (for extra measure I even disable the "Compatibility Support Module" in my motherboard bios that even enables it to legacy boot since it just causes problems usually)... For the VM, you should use OVMF, not SeaBios to enable UEFI in the VM... And use the Q35 machine type since i440fx has to emulate the older PCI to be able to passthrough PCIe... Try those, and if that doesn't work, then we will need more details, like the Diagnostics files from the system, and the VM XML file... Edit: Also, unless you have a specific reason to have it enabled (and there may be one for AMD motherboards, don't know), you should disable the "allow_unsafe_interrupts" option you added to your boot config Edited January 14, 20197 yr by Warrentheo
Archived
This topic is now archived and is closed to further replies.