Description of problem:

Our existing documentation [1] and the upstream Nova documentation [2] describes how to configure PCI passthrough in general.  Unfortunately, neither source of documentation references the additional steps necessary to properly configure passthrough of a GPU device.

[1]: https://access.redhat.com/solutions/3260061
[2]: https://docs.openstack.org/nova/pike/admin/pci-passthrough.html

Our standard kernel package ships with the "nouveau" driver, which will automatically bind to any available nvidia GPUs when the system boots.  This can cause conflicts with guests that attempt to access the GPU at the same time.

The proper instructions can actually be found in our "Virtualization Deployment and Administration Guide" [3], but they are buried in the section on the vGPU feature when in fact that are required for any GPU passhthrough.

[3]: https://access.redhat.com/documentation/en-us/red_hat_enterprise_linux/7/html/virtualization_deployment_and_administration_guide/sect-device-gpu#sect-device-GPU-vGPU

Our OpenStack documentation should provide a summary of these instructions for people who wish to configure GPU passthrough for their Nova servers.

Version-Release number of selected component (if applicable):

RHEL-OSP 12 (and later)