Re: [RFC PATCH net-next v2 0/7] net: mdio: an MDIO device driver for the Airoha EN8811H

[Aleksei Sviridkin]

On Wed, Sep 09, 2026 at 09:01:14PM +0200, Andrew Lunn wrote:
> I think because you are sending this as RFC, the AI bots are ignoring
> it. I also think it is not getting build testing etc.

Next round goes without the tag. One bot did answer 7/7, but by private
mail only, so that side is not consistent either. The build testing is
what I want back.

> One way to argue it is that the MDIO bus it connected to the MCU. The
> MCU, once booted, forwards the MDIO bus through to the PHY part of the
> package.

That is the argument the binding was missing. It will say that, instead
of describing a driver model choice.

> Consider a normal setup, what happens when you remove the MDIO
> controller when the PHY is directly connected?  And look at what MDIO
> mux drivers do, which is not so different a use case.

It faults. On this board a bus going away under an attached PHY ends in
a NULL phydev->drv out of phy_stop(), directly connected or not.
mdio_mux_uninit_children() unregisters each child bus and frees it,
which is what this driver does. So nothing new to invent here; the hole
is in phylib.

> It is fine. I would also add a select or depends on the PHY driver
> Kconfig stanza to link it to the MDIO_AIROHA_EN8811H driver.

Will do.

> Prose is enough. And C code to catch DT authors who get it wrong. But
> you also need to be careful to not break backwards compatibility when
> the PHY driver is used alone.

The child bus will refuse an address that is not the parent's and name
the node that got it wrong. Boards that describe the chip as a plain PHY
keep working through the PHY driver as before.

> The location is fine. MAINTAINERS need to list it. What you might want
> to do is list it twice, once in the PHY driver MAINTAINERS record, and
> again in the MDIO driver section.

Both records then.