Re: [PATCH v1 2/6] net: mv88e61xx: Configure PHY ports to also pass packets between them

[Vladimir Oltean <olteanv@gmail.com>]

Hi Lukasz,

On Thu, Jun 01, 2023 at 01:44:30PM +0200, Marek Vasut wrote:
> I think after two years, it would be good to drop the RB tags and do another
> round of reviews.

To expand on Marek's point.

In those past 2 years, Tim Harvey has put in a considerable amount of
effort to add another driver for mv88e6xxx that uses DM_DSA. I believe
the current "PHY" driver for the same hardware should be considered
obsolete until all platforms are converted to DM_DSA, then it can be
deleted. So, no new features for it.

Then, there's also the question of the sanity of the proposed change itself.

I believe that we need to be humble enough to recognize that the U-Boot
network stack is not competent enough to handle the switching capabilities
of a switch, not even enough for it to be safe. It doesn't handle STP
(Spanning Tree Protocol), for one thing. So it will never be capable of
detecting switching loops, such as to block one of its ports in order to
not kill the network.

In principle, I would say: as long as there is no plan to handle STP,
there should be no plan to allow autonomous packet forwarding from U-Boot.
The U-Boot network stack is there so that you can TFTP a kernel and boot it,
which is also the only use case behind DM_DSA.

But you may say: I'm never going to allow packet forwarding from U-Boot
in a network with loops!

Okay, but your patch suggests otherwise. Which ports allow forwarding is
a compile-time option, which... is by definition contrary to any runtime
network topology determinations.

Maybe enabling forwarding between switch ports through a CLI command
that communicates with DM_DSA would be tolerable - assuming that users
are smart enough to not use it in a network with STP. But again, I'm not
really sure what's the use case.