February 2, 201412 yr Having spent the evening playing with unRAID 6.0b3, I managed to get graphics passthrough to a windows VM working. It turned out to be a good deal easier than I expected and works pretty much the same as I had it working in KVM on Manjaro but at least now we have something approaching the unRAID way. I've written it up in a blog post; http://mediaserver8.blogspot.ie/2014/02/unraid-60-xen-guest-gpu-passthrough.html#more In essence, the steps are; [*]Set up bridged networking (via unRAID web UI) [*]Identify the device id(s) to be passed through (using lspci) [*]Add the device IDs to xen-pciback (via syslinux.cfg) [*]Confirm devices as assignable (usig xl pci-assignable-list) [*]Add the device ids to the VM config file [*]Boot, install drivers, enjoy! I also established that it's possible to make the devices assignable (add to xen-pciback) via the go script. Some people might be more comfortable editing this than their syslinux.cfg file but I'm sure it's not a great idea for a number of reasons. For reference, here's how it can be done; #!/bin/bash # Start the Management Utility /usr/local/sbin/emhttp & xl pci-assignable-add 05:00.0 xl pci-assignable-add 05:00.1
February 2, 201412 yr Nice guide! I will be doing this in next few months after move to have server for me and desktop for the wife all in one!
February 2, 201412 yr I also established that it's possible to make the devices assignable (add to xen-pciback) via the go script. Some people might be more comfortable editing this than their syslinux.cfg file but I'm sure it's not a great idea for a number of reasons. For reference, here's how it can be done; #!/bin/bash # Start the Management Utility /usr/local/sbin/emhttp & xl pci-assignable-add 05:00.0 xl pci-assignable-add 05:00.1 Depending on the device / motherboard / bios / Intel / AMD / Etc... this is will not work reliably and start / stopping VMs will hose the PCI Device from working until reboot and sometimes a poweroff is required. It's best to have the syslinux assign the PCI Devices before the Linux Kernel "grabs" the device and loads a module for it. There are also some other commands you will want to add to the Xen Kernel and the Linux Kernel (both in Syslinux) to make PCI Passthrough more reliable / compatible with various Motherboard too. AMD and Intel CPUs each have their own. I will post those tomorrow if you do not find them first.
February 2, 201412 yr Author I also established that it's possible to make the devices assignable (add to xen-pciback) via the go script. Some people might be more comfortable editing this than their syslinux.cfg file but I'm sure it's not a great idea for a number of reasons. For reference, here's how it can be done; #!/bin/bash # Start the Management Utility /usr/local/sbin/emhttp & xl pci-assignable-add 05:00.0 xl pci-assignable-add 05:00.1 Depending on the device / motherboard / bios / Intel / AMD / Etc... this is will not work reliably and start / stopping VMs will hose the PCI Device from working until reboot and sometimes a poweroff is required. It's best to have the syslinux assign the PCI Devices before the Linux Kernel "grabs" the device and loads a module for it. There are also some other commands you will want to add to the Xen Kernel and the Linux Kernel (both in Syslinux) to make PCI Passthrough more reliable / compatible with various Motherboard too. AMD and Intel CPUs each have their own. I will post those tomorrow if you do not find them first. No chance of me finding them as I'm off to bed now - it's past 3.00AM here! Would appreciate anything you can add - I'll update the guide accordingly.
February 2, 201412 yr Great job with the guide! I didn't want to write one and yours looks better than what I could have done. Question: Why are you assigning so much memory to Dom0? Mine is running on less than 1GB but can ballon up to 2GB. I am running XBMC, Mysql, unRAID and a ton of other tools / apps too. The only reason you would assign more 2GB is if you are running the cache directories or plugins (but most of those are nothing anyway). Run the following commands: free -m and xl info What is memory usage? Another tool you want to check out too... xl top
February 2, 201412 yr Nvidia GPU cards are still a no go on passthru unless "Quadrilized", right? Sent from my LG-VS980 using Tapatalk
February 2, 201412 yr Nvidia GPU cards are still a no go on passthru unless "Quadrilized", right? Correct
February 2, 201412 yr Author Question: Why are you assigning so much memory to Dom0? No reason whatsoever. That value is the default in syslinux.cfg as installed and I simply didn't change it. I'm sure as I continue to build up a few vms, I'll be releasing more memory from dom0. Thanks. Peter.
February 3, 201412 yr I also established that it's possible to make the devices assignable (add to xen-pciback) via the go script. Some people might be more comfortable editing this than their syslinux.cfg file but I'm sure it's not a great idea for a number of reasons. For reference, here's how it can be done; #!/bin/bash # Start the Management Utility /usr/local/sbin/emhttp & xl pci-assignable-add 05:00.0 xl pci-assignable-add 05:00.1 Depending on the device / motherboard / bios / Intel / AMD / Etc... this is will not work reliably and start / stopping VMs will hose the PCI Device from working until reboot and sometimes a poweroff is required. It's best to have the syslinux assign the PCI Devices before the Linux Kernel "grabs" the device and loads a module for it. There are also some other commands you will want to add to the Xen Kernel and the Linux Kernel (both in Syslinux) to make PCI Passthrough more reliable / compatible with various Motherboard too. AMD and Intel CPUs each have their own. I will post those tomorrow if you do not find them first. Looks like nobody found out. iommu=1? I got my radeon r9 290 passed through working "okay", though no reboots without host reboot, and it hung up completely at some point. Xen also complains about the kernel not being able to reset the card through sys, to which I attribute the rebooting problem. Could you enlighten us?
February 3, 201412 yr Author Looks like nobody found out. iommu=1? I got my radeon r9 290 passed through working "okay", though no reboots without host reboot, and it hung up completely at some point. Xen also complains about the kernel not being able to reset the card through sys, to which I attribute the rebooting problem. Could you enlighten us? Did you follow the instructions on setting up rebooting in IronicBadgers blog post? http://blog.ktz.me/?p=219
February 3, 201412 yr Not yet, though that seems more like a workaround than a solution. Gonna give it a whirl.
Archived
This topic is now archived and is closed to further replies.