There is a race condition when two drivers initiate interfaces simultaneously.
This was detected after adding the Realtek driver in T6713 (some debugs and logs we can find in this task)
sudo dmesg | grep -ie "renamed from" [ 0.883572] vmxnet3 0000:0b:00.0 e2: renamed from eth0 [ 1.229786] e1000e 0000:04:00.0 e3: renamed from eth1 [ 1.235255] e1000e 0000:0c:00.0 e4: renamed from eth0 [ 1.240044] e1000e 0000:13:00.0 e5: renamed from eth2 [ 1.245533] e1000e 0000:1b:00.0 e6: renamed from eth3 [ 1.749115] r8152 4-1:1.0 e7: renamed from eth0 [ 18.536107] e1000e 0000:13:00.0 eth1: renamed from e5 [ 18.540849] vmxnet3 0000:0b:00.0 eth0: renamed from e2 [ 18.647673] e1000e 0000:04:00.0 eth3: renamed from e3 [ 18.754589] e1000e 0000:0c:00.0 eth4: renamed from e4 [ 18.761645] e1000e 0000:1b:00.0 eth2: renamed from e6 [ 18.832072] r8152 4-1:1.0 eth5: renamed from e7
Thoughts from @zsdc
The problem is in the logic of our initialization of interfaces.
The only way to solve it correctly is to avoid creating interfaces named ethX by default, if possible, or to delay the initialization of drivers/interfaces so that two drivers never initialize devices at the same time.
Or, rename the CLI and the configured system eth to ether so that they do not overlap with standard names.