Networking

You can up set up communications between the host and guests, between guests, and between guests and the outside world.

When you design a network interface for a guest, you can use:

a virtio-net vdev connected to another virtio-net vdev, usually in another guest
a virtio-net vdev connected to a devnp-vdevpeer.so driver instance running in the hypervisor host
a pass-through device

The iopkt-* devnp-vdevpeer.so driver runs in the hypervisor host. It enables the iopkt-* instance that is running in the hypervisor host to communicate with guests through appropriately configured virtual devices in the qvm process instances hosting guests (see devnp-vdevpeer.so in the “Utilities and Drivers Reference” chapter).

A virtio-net vdev offers peer-to-peer communication, either to another virtio-net in a guest, or to the devnp-vdevpeer.so driver running in the hypervisor host (see “vdev virtio-net”).

Transmit Segmentation Offload (TSO)

When setting up your peer-to-peer network, consider using Transmit Segmentation Offload (TSO), which allows your NIC to manage the packets. For example, the ifconfig startup in your guest might look something like this:

ifconfig vt1 tcp4csum udp4csum tcp6csum udp6csum tso4 tso6

If you use TSO, you should use the virtio-net vdev peerfeats option. For example:

vdev virtio-net
    ...
    peerfeats 0x00007fc3

See “vdev virtio-net” in the “Configuration” chapter.

For more information about TSO, see “Transmit Segmentation Offload (TSO)” in the Core Networking Stack User's Guide, and ifconfig in the Utilities Reference.

MAC addresses in a hypervisor system

Note the following about assigning MAC addresses:

MAC addresses must be unique in any given Ethernet segment. There can be more than one Ethernet segment on a target board, and a target board may share an Ethernet segment with some external entity.
When assigning MAC addresses, you should set the locally-assigned bit (see IEEE 802.3-2015 Section 1, paragraph 3.2.3, item b).

Enabling peer-to-peer networking

To support connectivity between guests in your hypervisor system, and between guests and the outside world, peer-to-peer networking must be be enabled in the host.

You can do this by starting io-pkt-* in the host, using the vdevpeer option and specifying the the peer interfaces you want to create. You can use the command line to start io-pkt-* after the host has booted, or you can add theio-pkt-* startup instructions to the hypervisor host's startup scripts. This is already implemented in the QNX Hypervisor reference images.

For reference, the following creates three interfaces, one for each guest in a hypervisor system with three guests:

io-pkt-v6-hc -d e1000 -d vdevpeer \
peer=/dev/qvm/qnx7-guest1/p2p,bind=/dev/vdevpeers/vp0,\
peer=/dev/qvm/qnx7-guest2/p2p,bind=/dev/vdevpeers/vp1,\
peer=/dev/qvm/linux-guest/p2p,bind=/dev/vdevpeers/vp2

For more information about configuring peer interfaces, see “vdev virtio-net” in the “Configuration” chapter. See also, “devnp-vdevpeer.so” in the “Utilities and Drivers Reference” chapter.