Re: [PATCH net-next v6 11/17] dt-bindings: net: pse-pd: Add another way of describing several PSE PIs

[Kory Maincent <kory.maincent@bootlin.com>]

On Thu, 4 Apr 2024 10:38:54 +0200
Kory Maincent <kory.maincent@bootlin.com> wrote:

> On Wed, 3 Apr 2024 09:31:42 -0500
> Rob Herring <robh@kernel.org> wrote:
> 
> > On Wed, Apr 03, 2024 at 11:15:48AM +0200, Kory Maincent wrote:  
> > > On Tue, 2 Apr 2024 08:26:37 -0500
> > > Rob Herring <robh@kernel.org> wrote:
> > >     
> > > > > +          pairset-names:
> > > > > +            $ref: /schemas/types.yaml#/definitions/string-array
> > > > > +            description:
> > > > > +              Names of the pairsets as per IEEE 802.3-2022, Section
> > > > > 145.2.4.
> > > > > +              Valid values are "alternative-a" and "alternative-b".
> > > > > Each name      
> > > > 
> > > > Don't state constraints in prose which are defined as schema 
> > > > constraints.    
> > > 
> > > Ok, I will remove the line.
> > >     
> > > > > +          pairsets:
> > > > > +            $ref: /schemas/types.yaml#/definitions/phandle-array
> > > > > +            description:
> > > > > +              List of phandles, each pointing to the power supply for
> > > > > the
> > > > > +              corresponding pairset named in 'pairset-names'. This
> > > > > property
> > > > > +              aligns with IEEE 802.3-2022, Section 33.2.3 and
> > > > > 145.2.4.
> > > > > +              PSE Pinout Alternatives (as per IEEE 802.3-2022 Table
> > > > > 145\u20133)
> > > > > +
> > > > > |-----------|---------------|---------------|---------------|---------------|
> > > > > +              | Conductor | Alternative A | Alternative A |
> > > > > Alternative B | Alternative B |
> > > > > +              |           |    (MDI-X)    |     (MDI)     |      (X)
> > > > > |      (S)      |
> > > > > +
> > > > > |-----------|---------------|---------------|---------------|---------------|
> > > > > +              | 1         | Negative VPSE | Positive VPSE | \u2014
> > > > >     | \u2014             |
> > > > > +              | 2         | Negative VPSE | Positive VPSE | \u2014
> > > > >     | \u2014             |
> > > > > +              | 3         | Positive VPSE | Negative VPSE | \u2014
> > > > >     | \u2014             |
> > > > > +              | 4         | \u2014             | \u2014             |
> > > > > Negative VPSE | Positive VPSE |
> > > > > +              | 5         | \u2014             | \u2014             |
> > > > > Negative VPSE | Positive VPSE |
> > > > > +              | 6         | Positive VPSE | Negative VPSE | \u2014
> > > > >     | \u2014             |
> > > > > +              | 7         | \u2014             | \u2014             |
> > > > > Positive VPSE | Negative VPSE |
> > > > > +              | 8         | \u2014             | \u2014             |
> > > > > Positive VPSE | Negative VPSE |
> > > > > +            minItems: 1
> > > > > +            maxItems: 2      
> > > > 
> > > > "pairsets" does not follow the normal design pattern of foos,
> > > > foo-names, and #foo-cells. You could add #foo-cells I suppose, but what
> > > > would cells convey? I don't think it's a good fit for what you need.
> > > > 
> > > > The other oddity is the number of entries and the names are fixed. That 
> > > > is usually defined per consumer.     
> > > 
> > > Theoretically if the RJ45 port binding was supported it would make more
> > > sense, but in reality it's not feasible as the PSE controller need this
> > > information in its init process.
> > > The PSE controller reset all its port to apply a configuration so we can't
> > > do it when the consumer (RJ45) probe. It would reset the other ports if
> > > one consumer is probed later in the process.    
> > 
> > There is no reason other than convenience that all information some 
> > driver needs has to be in one node or one hierarchy of nodes. You can 
> > fetch anything from anywhere in the DT. It does feel like some of this 
> > belongs in a connector node. We often haven't described connectors in DT 
> > and stick connector properties in the controller node associated with 
> > the connector. Then as things get more complicated, it becomes a mess.   
> 
> Right, we could indeed put all the informations of the pse_pi node in the
> future RJ45 port abstraction node. Then, this series will be put aside until
> we manage to have the port abstraction get merged.
> I am not glad about this as it will stuck my work until then, but indeed
> removing this pse_pi wrapper node which is between the pse_controller node and
> the connector node seems cleaner.

After some new thought, I thinks it is quite similar on the devicetree side to
have it in a pse_pi node or in the connector node.
Here are my agruments to continue using this pse_pi binding description:
- The connector abstraction is in its early work and won't really see a v1 soon
  while the PoE series got mainly all reviewed-by thanks to Andrew.
  This would stuck the PoE series until maybe one or two Linux version.
- It allows to use the "Power Interface" name like described in the standards.
- Even if this is in the PSE controller node, it is generic to all PSEs so it
  shouldn't become a mess.
- It allows to have the PSE controller and Power Interfaces parameters grouped
  together and it will be easier to read. May not really be an argument! ;)
- It will keep the logic of PoDL with the PHY using a single reference to the
  PSE PI through the pses parameter. 

Is it okay for you to continue with it?

Regards,
-- 
Köry Maincent, Bootlin
Embedded Linux and kernel engineering
https://bootlin.com