Add support for creating PCIe root ports in QEMU #442
Conversation
| "pcie-root-port,bus=${bus},id=${id},chassis=${toString chassis}" | ||
| ] | ||
| ) pcieRootPorts | ||
| ) |
There was a problem hiding this comment.
The lines above now require two spaces of indentation.
|
|
||
| qemu.pcieRootPorts = mkOption { | ||
| description = '' | ||
| A list of PCIe root ports that can be used for device hot-plugging or |
There was a problem hiding this comment.
Interesting feature. Can we add a link that explains how hot-plugging works?
Could we even generate some helper scripts for the user based on this configuration?
There was a problem hiding this comment.
hotplug main feature
https://github.com/tiiuae/vhotplug
and the downstream PR, that uses this patch
tiiuae/ghaf#1624
overall we are quite heavy users of passthrough with microvm.
https://github.com/tiiuae/ghaf/blob/main/modules/hardware/common/devices.nix
https://github.com/tiiuae/ghaf/blob/main/modules/hardware/common/qemu.nix
https://github.com/tiiuae/ghaf/tree/main/modules/microvm
https://github.com/tiiuae/ghaf/tree/main/modules/hardware/passthrough
if there is appetite. I think that it could be great to move some of the functionality from Ghaf to microvm, so that it would be readily available to more folks.
3 participants
This change adds the ability to define PCIe root ports for QEMU virtual machines. These ports can be used for device hot-plugging, allowing vfio-pci devices to be attached dynamically.
In the current version, it's possible to create ports via extraArgs and set them in deviceExtraArgs but with recent changes (#437), this approach produces "bus not found" errors. There is likely a race condition between creating vfio-pci devices which are now placed at the beginning of the QEMU command and the PCIe ports that were appended at the end.
With this change, PCIe root ports are explicitly created before the vfio-pci devices that may use them.