Im able to get the vm to see the GPU, its saying Error 43: Driver failed to
load

>From what Im finding out I should be able to edit the XML file and trick
the NVIDIA drivers into not knowing its a part of a VM.

First you have to setup the libvirt user:
saslpasswd2 -f /etc/libvirt/passwd.db -c root

then you edit the xml:
cd /etc/libvirt/qemu
virsh edit Win-A

Add kvm hidden state and such as shown here:
https://wiki.archlinux.org/index.php/PCI_passthrough_via_OVMF#Troubleshooting


Then I try to save it but then it complains about:

 Requested operation is not valid: domain has assigned non-USB host devices

This is as far as I have made it.


On Sat, Mar 16, 2019 at 12:43 PM Wesley Stewart <[email protected]> wrote:

> I actually use an oVirt VM as my desktop at home.  Why?  Because I can and
> I didn't want to have to buy another box for my desktop.  However there are
> a few caveats to this and some lessons I learned.
>
> The consumer grade GTX cards are going to have issues being passed
> through.  NVIDIA blocks this in their drivers and RedHat honors NVIDIA's
> requirements for this.  You can enter an agreement with NVIDIA, which I am
> sure costs some money somewhere, to get special drivers.  Or you can use
> their enterprise Quadro cards, however I tested a P2000 and could not get
> it to work on oVIrt.  But I did not try very hard.
>
> At home, I am using an AMD Radeon RX 480/580.  AMD doesn't care if you
> pass it through and it works almost right out of the box.  You just want to
> make sure to blacklist the AMD drivers for the host, and put them in the
> PCI-Stubs group.  It's really easy with some basic bash/terminal knowledge:
>
> https://ovirt.org/develop/release-management/features/virt/hostdev-passthrough.html
> Check the physical GPU portion.
>
> You can, HOWEVER, trick the VM into not thinking it's a VM in KVM and
> QEMU.  To prove this, I built a proxmox box at work to test it and I was
> able to successfully pass through a GTX 1060 without too much trouble.  I
> Have some notes here:
>
> https://docs.google.com/document/d/1vlOIbJ4iXE7ypdSJ0OWQZfgcG9dEx2_PxXh6oRjl878/edit?usp=sharing
> However these are pretty rough notes.  I just wanted to prove it could be
> done.
>
> For oVirt a lot of the QEMU/KVM settings and commands are going to be
> stored in databases and files.  and I had trouble getting these to pick up
> on the oVirt side.  I even tried creating a VDSM hook, as mentioned in this
> thread:
> https://lists.ovirt.org/pipermail/users/2017-March/080905.html
> But I never got it working.
>
> I am definitely no expert, but I have done a lot of tinkering with this.
> But you will get pushback from oVirt/Redhat as this is not officially
> supported.  If you want it to work with consumer grade cards, use AMD.  If
> you want to do it "right" you will have to buy the enterprise quadro
> cards.  But these get quite pricey.
>
> On Sat, Mar 16, 2019 at 12:15 PM Darin Schmidt <[email protected]>
> wrote:
>
>> If I'm reading through the documents correctly, you can't use consumer
>> grade gpus such as gtx 1080's and assign them to a VM?
>>
>> I'm trying to do something similar to how Linus used untaid to build
>> several gaming machines from one.
>>
>> Any help would be appreciated.
>>
>> Thanks.
>> _______________________________________________
>> Users mailing list -- [email protected]
>> To unsubscribe send an email to [email protected]
>> Privacy Statement: https://www.ovirt.org/site/privacy-policy/
>> oVirt Code of Conduct:
>> https://www.ovirt.org/community/about/community-guidelines/
>> List Archives:
>> https://lists.ovirt.org/archives/list/[email protected]/message/DJBX3NMFJRP3L5DRFMJAD7FWA3WAX4PM/
>>
>
_______________________________________________
Users mailing list -- [email protected]
To unsubscribe send an email to [email protected]
Privacy Statement: https://www.ovirt.org/site/privacy-policy/
oVirt Code of Conduct: 
https://www.ovirt.org/community/about/community-guidelines/
List Archives: 
https://lists.ovirt.org/archives/list/[email protected]/message/ZRON32TALP7SXTISRFR5TYKIG3EGUAVN/

Reply via email to