On 3/4/19 10:03 AM, Erik Skultety wrote:
On Sun, Mar 03, 2019 at 10:23:11AM -0300, Daniel Henrique Barboza wrote:
This series includes Libvirt support for a new QEMU feature for the spapr (PPC64) machine, NVIDIA V100 + P9 passthrough. Refer to [1] for the version 3 of this feature (same version used as a reference for this series).
v2 has a entirely different approach from the first patch [2] that simply grants IPC_LOCK to the QEMU process:
- first patch is a cleanup to make it easier to insert additional logic in the code
- patch 2 contains helper functions that browses the device tree at /proc/device-tree to detect if a given VFIO PCI device is using a NVLink2 bus
- patch 3 includes the passthroughLimit calculation for PPC64 guests that are using NVLink2 passthrough GPUs.
[1] https://patchwork.kernel.org/cover/10831413/ Is there something more (like a doc) apart from the patches that I can use to crosscheck your implementation?
I don't know. I am CCing Piotr Jaroszynski from NVIDIA here to see if he can give us more info. Piotr, is there any public available doc from NVIDIA with NVLink2 passthrough specifications? Thanks, Daniel
Erik