Wierd problem with multiple network interfaces on VM

[therealjrd]

This one has me stumped.

I have a virtualization server

gentoo-kernel-6.12.63
libvirt-11.10.0
qemu-10.2.0-r1

The host machine has a 4 port NIC card in it, containing 4 intel 82580s. It's also configured with a bridge config to 8 tun/tap virtual interfaces tied to the main interface on the motherboard.

The four intel nics show up on the host as enp6s0f[0..4]. The bridge is named br0.

I have a number of VMs for various purposes, a couple of which use dedicated ports on the multi-nic card. The guest configuration of those VMs has a bridge type virtual NIC bound to the host br0, and a macvtap device tied to one of the enp6 interfaces. That's been working fine for some time. Recently, I tried to configure up a couple of new VMs with that recipe, using the previously unused enp6 nics. I can configure the VM fine, but when I start it, it comes up with only one nic, the main one tied to host br0.

I've tried a number of things, changed the IO scheme from guest to the nic (e1000, virtio, or Hypervisor Default) but that makes no difference. Reordering the interfaces makes no difference. On the theory that some internal resource was getting exhausted, I shut down the other multi-nic VMs, no difference.

Oddly, if I delete the interface tied to host br0, the guest will come up, using the enp6 device, as its only interface. So I know the basic mechanism works, there's something preventing it from coming up as a second interface.

I'm not sure what else to look at. I have two VMs using this scheme, which work great. I'm reluctant to touch them, as I don't want to break them the same way I seem to have done with my new ones.

What else should I be looking at? I looked at the docs on libvirtd logging, and it has big neon warnings about level 1, so I haven't tried it yet, but I suspect I need to. Anything else? Any other information I could supply here?

Thanks in advance for any hints.

[therealjrd]

I turned on some libvirtd logging on the machine which is having the problem. When I start one of my test VMs I see in the log where it's trying to bring up the direct interface. Immediately after that I see errors of the form

Code: Select all

2026-03-22 14:30:15.512+0000: 5720: debug : virPCIVPDResourceGetKeywordPrefix:76 : keyword='0' key='<null>'
2026-03-22 14:30:15.512+0000: 5720: debug : virPCIVPDParseVPDLargeResourceFields:469 : Could not determine a field value format for keyword: 0
2026-03-22 14:30:15.512+0000: 5720: debug : virPCIVPDParseVPDLargeResourceFields:477 : data field length invalid

and

Code: Select all

2026-03-22 14:30:15.515+0000: 5720: debug : virPCIVPDParse:687 : Encountered an unexpected VPD resource tag: 0x7
2026-03-22 14:30:15.516+0000: 5720: debug : virPCIVPDResourceGetKeywordPrefix:76 : keyword='N' key='<null>'
2026-03-22 14:30:15.516+0000: 5720: debug : virPCIVPDParseVPDLargeResourceFields:469 : Could not determine a field value format for keyword: N

I'm not sure those invalid characters come through when pasting here, the first is ^E, the second is ^K.

I tried setting up a spare machine as a virt server, and trying an analogous test. I get no such errors. I'm now wondering if some set of files used by libvirt or qemu is corrupted. Or maybe more likely, my libvirt is encounterning VPDs from some devices (like my NIC card) which it doesn't understand?

Does this ring any bells with anyone?