Re: [PATCH 1/2] of: net: Add helper function of_get_ethdev_label()

[Pali Rohár]

On Thursday 13 January 2022 19:27:19 Pali Rohár wrote:
> On Friday 07 January 2022 17:55:25 Andrew Lunn wrote:
> > On Fri, Jan 07, 2022 at 04:29:31PM +0000, Russell King (Oracle) wrote:
> > > On Fri, Jan 07, 2022 at 05:12:21PM +0100, Pali Rohár wrote:
> > > > Adds a new helper function of_get_ethdev_label() which sets initial name of
> > > > specified netdev interface based on DT "label" property. It is same what is
> > > > doing DSA function dsa_port_parse_of() for DSA ports.
> > > > 
> > > > This helper function can be useful for drivers to make consistency between
> > > > DSA and netdev interface names.
> > > > 
> > > > Signed-off-by: Pali Rohár <pali@xxxxxxxxxx>
> > > 
> > > Doesn't this also need a patch to update the DT binding document
> > > Documentation/devicetree/bindings/net/ethernet-controller.yaml ?
> > > 
> > > Also it needs a covering message for the series, and a well thought
> > > out argument why this is required. Consistency with DSA probably
> > > isn't a good enough reason.
> > > 
> > > >From what I remember, there have been a number of network interface
> > > naming proposals over the years, and as you can see, none of them have
> > > been successful... but who knows what will happen this time.
> > 
> > I agree with Russell here. I doubt this is going to be accepted.
> > 
> > DSA is special because DSA is very old, much older than DT, and maybe
> > older than udev. The old DSA platform drivers had a mechanism to
> > supply the interface name to the DSA core. When we added a DT binding
> > to DSA we kept that mechanism, since that mechanism had been used for
> > a long time.
> > 
> > Even if you could show there was a generic old mechanism, from before
> > the days of DT, that allowed interface names to be set from platform
> > drivers, i doubt it would be accepted because there is no continuity,
> > which DSA has.
> 
> Well, DT should universally describe HW board wiring. From HW point of
> view, it is really does not matter if RJ45 port is connected to embedded
> PHY on SoC itself or to the external PHY chip, or to the switch chip
> with embedded PHY. And if board has mix of these options, also labels
> (as printed on product box) should be in DTS described in the same way,
> independently of which software solution / driver is used for particular
> chip. It really should not matter for DTS if kernel is using for
> particular HW part DSA driver or ethernet driver.
> 
> So there really should be some common way. And if the one which DSA is
> using is the old mechanism, what is the new mechanism then?

Hello! Any comments on this?