Re: [RFC PATCH net-next 1/9] dt-bindings: net: dsa: convert ksz bindings document to yaml

[Rob Herring <robh@kernel.org>]

On Thu, Oct 22, 2020 at 12:17:26PM -0700, Florian Fainelli wrote:
> On 10/22/20 5:37 AM, Vladimir Oltean wrote:
> > On Thu, Oct 22, 2020 at 12:54:52PM +0200, Kurt Kanzenbach wrote:
> >> On Wed Oct 21 2020, Florian Fainelli wrote:
> >>> On 10/21/2020 5:16 PM, Vladimir Oltean wrote:
> >>>> On Wed, Oct 21, 2020 at 08:52:01AM +0200, Kurt Kanzenbach wrote:
> >>>>> On Mon Oct 19 2020, Christian Eggers wrote:
> >>>>> The node names should be switch. See dsa.yaml.
> >>>>>
> >>>>>> +            compatible = "microchip,ksz9477";
> >>>>>> +            reg = <0>;
> >>>>>> +            reset-gpios = <&gpio5 0 GPIO_ACTIVE_LOW>;
> >>>>>> +
> >>>>>> +            spi-max-frequency = <44000000>;
> >>>>>> +            spi-cpha;
> >>>>>> +            spi-cpol;
> >>>>>> +
> >>>>>> +            ports {
> >>>>>
> >>>>> ethernet-ports are preferred.
> >>>>
> >>>> This is backwards to me, instead of an 'ethernet-switch' with 'ports',
> >>>> we have a 'switch' with 'ethernet-ports'. Whatever.
> >>>
> >>> The rationale AFAIR was that dual Ethernet port controllers like TI's 
> >>> CPSW needed to describe each port as a pseudo Ethernet MAC and using 
> >>> 'ethernet-ports' as a contained allowed to disambiguate with the 'ports' 
> >>> container used in display subsystem descriptions.
> >>
> >> Yes, that was the outcome of previous discussions.
> > 
> > And why would that disambiguation be necessary in the first place? My
> > understanding is that the whole node path provides the necessary
> > namespacing to avoid the confusion. For example, the 'reg' property
> > means 100 things to 100 buses, and no one has an issue with that. I am
> > not expecting an Ethernet switch to have an HDMI port, I might be wrong
> > though.
> 
> The disambiguation is more of a hint given to DT analysis tools to
> validate a given node with little to no knowledge of the containing
> node. I don't really have a dog in the fight here.

A node name should mean 1 and only 1 type/class of node. 'ports' (and 
'port') is for the graph binding.

Rob