qemu fails when starting with more than one virtual NIC

[Jonas Eriksson]

I just upgraded my VM-host from 14.04 (trusty) to 16.04 (xenial) with qemu version 1:2.5+dfsg-5ubuntu10.2, and noticed that VMs with more than one NIC would not start. I'm controlling my VMs using libvirt, and noticed this message in the log:

libvirtd[6335]: internal error: process exited while connecting to monitor: 2016-07-08T11:07:10.097780Z qemu-system-x86_64: -device virtio-net-pci,netdev=hostnet1,id=net1,mac=<redacted>,bus=pci.0,addr=0x4: duplicate fw_cfg file name: genroms/pxe-virtio.rom

Experimenting using the qemu command line logged by libvirt, I determined that removing all NICs but one would let me start VMs. I created this minimal example to replicate the behaviour:
    $ sudo /usr/bin/kvm -nographic -machine pc-0.11,accel=kvm,usb=off -netdev tap,id=hostnet0,vhost=on -device virtio-net-pci,netdev=hostnet0,id=net0,bus=pci.0,addr=0x3 -netdev tap,id=hostnet1,vhost=on -device virtio-net-pci,netdev=hostnet1,id=net1,bus=pci.0,addr=0x4
    qemu-system-x86_64: -device virtio-net-pci,netdev=hostnet1,id=net1,bus=pci.0,addr=0x4: duplicate fw_cfg file name: genroms/pxe-virtio.rom

Removing one NIC makes this setup work just fine:
    $ sudo /usr/bin/kvm -nographic -machine pc-0.11,accel=kvm,usb=off -netdev tap,id=hostnet0,vhost=on -device virtio-net-pci,netdev=hostnet0,id=net0,bus=pci.0,addr=0x3
[abort using 'ctrl-a', followed by 'x']

To verify that this was not due to the upgrade process, I have also booted a Live CD of the Desktop version of Xenial, and reproduced the error there successfully.

To try and mitigate the issue, I did download the source for the qemu package and did a dirty workaround by replacing the exit(1) below the error message with return, but the real issue seems to be that qemu tries to re-add the fw_cfgs for each NIC defined on the command line.

Thanks,
/Jonas

[Jonas Eriksson (7-ubuncuone-1)]

Update:
The issue seems tied to the pc-0.11 machine type. Using the latest available pc-* (pc-0.15) does not show the same issue. Perhaps this would warrent a wontfix, but at least the issue is now indexable by google :)

[Serge Hallyn (serge-hallyn)]

Thanks for the information. In the past, we've provided upgrade scripts when problems developed with older machine types, but 0.11 is old enough that it probably isn't worth it.