Support `cloud-hypervisor` by extending init script

[lukasfrank]

Summary

In order to be able to consume the VFs provided by the dpservice, they must be bound to the vfio-pci driver. Since dpservice is owning the VFs it would make sense to prepare them there.

Basic example

The following commands need to be executed for all VFs:

Example for PCI address 0000:06:06.1:

echo 'vfio-pci' | sudo tee /sys/bus/pci/devices/0000:06:06.1/driver_override
echo '0000:06:06.1' | sudo tee /sys/bus/pci/drivers/vfio-pci/bind

TODO:

• @lukasfrank Check if cloud-hypervisor-provider and libvirt-provider can run on the same host

[lukasfrank]

@guvenc Is there a timeline when this will be implemented? Would be great to have this as discussed with a flag since this blocks the cloud-hypervisor-provider development

[guvenc]

@lukasfrank
workaround to unblock:
please run the command below for all VFs in a loop on two compute nodes we have.

echo 'vfio-pci' | sudo tee /sys/bus/pci/devices/0000:06:06.1/driver_override
echo '0000:06:06.1' | sudo tee /sys/bus/pci/drivers/vfio-pci/bind

I will ask @byteocean whether he can help you ASAP
but FYI: this script only runs when the compute node gets restarted and no VFs existed and even if we had the script, we would still need to run the manual step above for a node which is already up and has a running dpservice.
The script would help only on newly on boarded vanilla compute-nodes or if we restart the compute nodes.

[byteocean]

@lukasfrank @guvenc is unbinding from mlx5_core driver is needed as well before the two commands here?