diff options
Diffstat (limited to 'results/classifier/gemma3:12b/network/2951')
| -rw-r--r-- | results/classifier/gemma3:12b/network/2951 | 22 |
1 files changed, 22 insertions, 0 deletions
diff --git a/results/classifier/gemma3:12b/network/2951 b/results/classifier/gemma3:12b/network/2951 new file mode 100644 index 000000000..966893643 --- /dev/null +++ b/results/classifier/gemma3:12b/network/2951 @@ -0,0 +1,22 @@ + +First byte of USB NIC is hardcoded to 0x40 +Description of problem: +Incus recently added support for USB attached network interfaces. +As with any network device, we generate a MAC address (using our MAC OUI) and allow the user to override that to a value of their choice. + +That's when we noticed that no matter what MAC address we set, the resulting MAC always has the prefix swapped to "40:". Looking into the code, this is done on purpose here: + +https://gitlab.com/qemu-project/qemu/-/blob/master/hw/usb/dev-network.c?ref_type=heads#L1386 + +Unfortunately there is no comment in the code or in any of the commits touching that code as far as why that is. + +We've also looked at the libvirt code handling those devices and that code seems to also assume that a user provided MAC will be correctly passed through to the guest, no mention of the odd prefix override. + +This is a bit concerning as there are valid IEEE OUI with the "40:" prefix. +So this means that QEMU may be generating collisions with actual physical MAC addresses... + +For a few months now, I've been applying this small patch to my own Incus packages (which bundle QEMU) and haven't heard or seen any obvious issue from the change. + +https://github.com/zabbly/incus/blob/daily/patches/qemu-0001-usb-net-mac.patch + +Does anyone know why this hardcoded MAC address prefix exists? |