(In reply to Takashi Iwai from comment #2) Depends on the point of view. Currently the device a) nameing scheme and its b) renamings in the kernel is not a) valid nor b) race free. If a) would be valid which is that is would be deterministic then b) wouldn't be required from the point of view from the side of the user space. For b) I can remember a bug (AFAICR bsc#961908 in comment https://bugzilla.suse.com/show_bug.cgi?id=961908#c16) where loading a kernel module for a 4 port and 2 port card the renaming had broken down ... which had required a hand crafted 70-persistent-net.rules to get the MAC address order into the order of the device names.