Attaching virtual GPU devices to guests in nova

Thanks. Probably need Sylvain Bauza to take a look at this.

Although, isn't there a workaround for this?

https://review.openstack.org/#/c/459753/

From that patch:

Image hide_hypervisor_id property helps libvirt set an xml instance file property, which will hide on guest host KVM hypervisor signature ("KVMKVMKVM\0\0\0"). According to the commit message in QEMU repository [1]:

"The latest Nvidia driver (337.88) specifically checks for KVM as the hypervisor and reports Code 43 for the driver in a Windows guest when found. Removing or changing the KVM signature is sufficient for the driver to load and work."

Also mentioned here:

https://docs.openstack.org/python-glanceclient/latest/cli/property-keys.html

img_hide_hypervisor_id:

Some hypervisors add a signature to their guests. While the presence of the signature can enable some paravirtualization features on the guest, it can also have the effect of preventing some drivers from loading. Hiding the signature by setting this property to true may allow such drivers to load and work.

While I understand your point, I tried to not provide a documentation for only one specific GPU driver, here NVidia.

Yes, you're right, in order to support virtual GPUs for NVidia devices, you need to have both a supported device ( see http://docs.nvidia.com/grid/5.0/pdf/grid-vgpu-user-guide.pdf ) and an accepted driver (and for example, not the 'nouveau' driver).

Maybe I should just add a note that would point to that documentation.

Matt, https://review.openstack.org/#/c/459753/ is for hiding the hypervisor signature in the guest because the nVidia driver checks it. That's for GPU passthrough, not for virtual GPU given you need to have a specific nVidia driver for your compute node itself (not the guest).

Putting the bug back to Confirmed.