Testing PCI passthrough and SR-IOV with emulated PCI NIC

Testing nova PCI and SR-IOV device handling in devstack in general requires special hardware available in the devstack machine or VM.

Since libvirt 9.3.0 and qemu 8.0.0 the situation is a lot simpler as qemu is now capable of emulating and intel igb NIC that supports SR-IOV. So devstack can be run in a VM that has one or more SR-IOV capable NIC while the host running the devstack VM does not require any special hardware just recent libvirt and qemu installed.

Note

The emulated igb device used in this doc only useful for testing purposes. While network connectivity will work through both the PF and the VF interfaces the networking performance will be limited by the qemu emulation.

Add SR-IOV capable igb NIC to a devstack VM

You can add an igb NIC to the devstack VM definition:

virsh attach-interface --type network --source <libvirt network name> --model igb --config <devstack-domain>

The SR-IOV capability also requires an IOMMU device and split APIC defined in the domain as well as the q35 machine type. So make sure you have the following sections:

<devices>
    ...
    <iommu model='intel'>
        <driver intremap='on' iotlb='on'/>
    </iommu>
</devices>

<features>
    ..
    <ioapic driver='qemu'/>
</features>

<os>
    ...
    <type arch='x86_64' machine='pc-q35'>hvm</type>
</os>

Then enable IOMMU in the devstack VM by adding intel_iommu=on to the kernel command line.

Using nova to get a VM with SR-IOV capable igb NICs

If you are using nova 2025.1 (Epoxy) or newer then you can use the hw_machine_type=q35, hw_vif_model=igb, and hw_iommu_model=intel image properties to request a VM with igb NICs.

Configuring your devstack VM for SR-IOV

Follow the guide;