[PATCH v2 1/2] ARM: at91/dt: describe rgmii ethernet phy connected to sama5d3xek boards

[Boris BREZILLON]

On Thu, 10 Jul 2014 08:35:15 -0700
Florian Fainelli <f.fainelli at gmail.com> wrote:

> 2014-07-10 2:07 GMT-07:00 Boris BREZILLON <boris.brezillon at free-electrons.com>:
> > On Fri, 27 Jun 2014 09:52:56 +0200
> > Nicolas Ferre <nicolas.ferre at atmel.com> wrote:
> >
> >> On 26/06/2014 22:01, Boris BREZILLON :
> >> > Hi Florian,
> >> >
> >> > On 26/06/2014 20:15, Florian Fainelli wrote:
> >> >> Hi Boris,
> >> >>
> >> >> 2014-06-26 3:13 GMT-07:00 Boris BREZILLON <boris.brezillon at free-electrons.com>:
> >> >>> Add ethernet-phy node and specify phy interrupt (connected to pin PB25).
> >> >>>
> >> >>> The PHY address is not specified here because atmel have 2 different
> >> >>> designs
> >> >>> for its CPU modules: one is connecting PHYAD[0-2] pins to pull up resistors
> >> >>> (Embest design) and the other one is connection PHYAD0 to a pull up
> >> >>> resistor and PHYAD[1-2] to pull down resistors (Ronetix design).
> >> >>> As a result, Ronetix design will have its PHY available at address 0x1 and
> >> >>> Embest design at 0x7.
> >> >>> Let the net PHY core automatically detect the PHY address by scanning the
> >> >>> MDIO bus.
> >> >> I though the compatible string was listed as a required property, but
> >> >> it is not. The 'reg' property however is listed as required, although
> >> >> the of_miodbus_register() works just fine without it, although that is
> >> >> a Linux-specific implementation detail.
> >> >
> >> > Indeed, it's listed in the required property list of the DT binding doc,
> >> > but the code implement auto detection if reg is missing.
> >> > However this line [1] clearly shows that specifying the reg property is
> >> > the preferred way of doing things.
> >> >
> >> > I could define 2 different sama5d3xcm.dtsi (sama5d3xcm-ronetix.dtsi and
> >> > sama5d3xcm-embest.dtsi) to avoid this dirty hack,
> >> > but then we would have 2 more dtb and the user would have to determine
> >> > which CPU module he owns to choose the appropriate dtb.
> >> > If at91, arm-soc and DT maintainers agree with this approach I can
> >> > definitely propose something.
> >>
> >> Yes Boris, I definitively prefer not to add another .dtsi file for this
> >> series if we can avoid it.
> >>
> >
> > Okay, now that I don't specify the reg property I have a bunch of
> > noisy logs (which is exactly what the developer of of_mdio.c wanted in
> > order to force people to specify the reg property).
> >
> > It seems to be a problem for atmel users (all these logs make them
> > think there is something wrong with the net device).
> >
> > Apart from the dts/dtsi split solution, which Nicolas wants to avoid, I
> > see two solutions here:
> >
> > 1) remove the logs (or use dev_dbg instead of dev_info) from of_mdio.c.
> > But I'm pretty sure this solution won't be accepted :-).
> 
> I am fine with using dev_dbg() instead of dev_info() for that sort of
> messages, provided that you state the rationale of this change
> (spewing the log console with probing messages) and specify tha the
> 'reg' property is optional.
> 
> >
> > 2) define 2 ethernet phys (one for each possible solution). I tested it
> > and it works fine (only the available PHY is registered and there is no
> > noisy logs anymore).
> 
> One advantage of that solution is that you'll get slightly faster boot
> times since you won't have to auto-probe for the PHYs on the MDIO bus,
> the time savings get bigger as you start using higher PHY addresses.

Yes I prefer this solution too, but is it acceptable to define 2 phy
nodes even if only one is really available ?




-- 
Boris Brezillon, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com