Re: [PATCH] net: uclass: Save ethernet MAC address when generated

[Tom Rini <trini@konsulko.com>]

[-- Attachment #1: Type: text/plain, Size: 2406 bytes --]

On Tue, Nov 16, 2021 at 03:56:45PM +0100, Wolfgang Denk wrote:
> Dear Tom,
> 
> In message <20211116141855.GD24579@bill-the-cat> you wrote:
> > 
> > So, to quote lib/Kconfig:
> > config NET_RANDOM_ETHADDR
> >         bool "Random ethaddr if unset"
> >         help
> >           Selecting this will allow the Ethernet interface to function
> >           even when the ethaddr variable for that interface is unset.
> >           A new MAC address will be generated on every boot and it will
> >           not be added to the environment.
> 
> This description is at least incomplete, because it makes no
> difference between the persistent copy of the environment and it's
> in-memory copy.  For network to function, I think the MAC address
> must be stored in the in-memory copy of the environment.

Well, networking has been working with NET_RANDOM_ETHADDR not updating
the environment, with caveats that user display information that reads
from the environment rather than ->enetaddr.  In so far as the
networking stack cares, ->enetaddr is what matters.  We look at the
environment for a MAC, and if it doesn't match the HW we use environment
MAC, but warn.

Now, I think RFC7042 is the current correct and relevant RFC here and it
doesn't mention persistence.  So I don't see a clear external authority
on if locally administered MAC addresses like this should be persistent
or not in this manner.

> > We need either a re-spin or follow-up as we're changing the documented
> > behavior.  And as I mentioned in the other thread related on-going
> > thread, perhaps "ethmacskip" should play a role in preserving existing
> > behavior?
> 
> We have way too many ways to do the same thing - nearly, just a
> little different :-(

Well, in this case I'm not sure that's the right problem to point at.
We can just set ethmacskip as a bit of corner-case functionality and
move on.

Because honestly, the more I read this, the more I think
https://patchwork.ozlabs.org/project/uboot/patch/20211115121152.3470910-1-michael@walle.cc/
is essentially the right direction.  There's no reason for 'net list' to
be using the environment here when ->enetaddr is what's being used by
the stack.  The use case of "I want to make my locally administered MAC
persist because my USB ethernet adapter lacks a MAC address" is solved
via the environment already.

-- 
Tom

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 659 bytes --]