[systemd-devel] when/where was support for assigning "ethX" names removed?

[Chris Friesen]

On 05/27/2016 09:43 AM, Michael Biebl wrote:
> 2016-05-27 17:14 GMT+02:00 Chris Friesen <cbf123 at mail.usask.ca>:
>> And the annoying thing is that if I turn off the new naming scheme there
>> seems to be less determinism than there used to be.  I assume this is due to
>> the effort to extract more parallelism at boot, but it's causing me grief.
>
> The old network interface naming scheme (which was bound to MAC
> addresses) used the same name space as the kernel.
> This was/is inherently racy. To make it more likely to succeed, udev
> tried the renaming several times.
>
> This hack was removed in upstream udev [1]. So this mean if you now
> use the old scheme it's much more likely that it fails.

The issue is not that the renaming fails (I actually modified the kernel to 
start at eth1000 so there is no possibility of collision).

The problem is that the kernel-assigned "ethX" names are not deterministic.  If 
I take two identical systems and boot the same OS on both, I can get different 
"ethX" ordering due to the fact that multiple drivers are modprobed in parallel 
and they race against each other to obtain the next eth device number.

All I'm looking for is a way to remove this raciness.  I don't care if it takes 
longer to boot.  I *think* that if I can serialize the modprobe then I should 
get deterministic numbering.

Chris