Re: [PATCH v12 1/2] dt-bindings: adc: add AD7173

[Conor Dooley <conor@kernel.org>]

[-- Attachment #1: Type: text/plain, Size: 4199 bytes --]

On Sun, Jan 21, 2024 at 03:41:18PM +0000, Jonathan Cameron wrote:
> On Thu, 18 Jan 2024 16:06:29 +0000
> Conor Dooley <conor@kernel.org> wrote:
> > On Thu, Jan 18, 2024 at 05:51:20PM +0200, Ceclan Dumitru wrote:
> > > On 1/18/24 17:23, Conor Dooley wrote:  
> > > > On Thu, Jan 18, 2024 at 02:49:22PM +0200, Dumitru Ceclan wrote:  
> > > >> +  adi,clock-select:
> > > >> +    description: |
> > > >> +      Select the ADC clock source. Valid values are:
> > > >> +      int         : Internal oscillator
> > > >> +      int-out     : Internal oscillator with output on XTAL2 pin
> > > >> +      ext-clk     : External clock input on XTAL2 pin
> > > >> +      xtal        : External crystal on XTAL1 and XTAL2 pins
> > > >> +
> > > >> +    $ref: /schemas/types.yaml#/definitions/string
> > > >> +    enum:
> > > >> +      - int
> > > >> +      - int-out
> > > >> +      - ext-clk
> > > >> +      - xtal
> > > >> +    default: int  
> > > > I am not a fan of properties like this one, that in my view reimplement
> > > > things that are supported by the regular clocks properties. I've got
> > > > some questions for you so I can understand whether or not this custom
> > > > property is required.
> > > > 
> > > > Whether or not the ext-clk or xtal is used is known based on
> > > > clock-names - why is the custom property required to determine that?  
> > 
> > > > If neither of those clocks are present, then the internal clock would be
> > > > used. Choosing to use the internal clock if an external one is provided
> > > > sounds to me like a software policy decision made by the operating
> > > > system.  
> > > 
> > > If there was no int-out, sure. I considered that the choice between int
> > > and int-out could be made here. So better for driver to choose int/int-out?  
> > 
> > This part of my comments was specifically about choosing between use of
> > the internal clock when ext-clk or xtal are provided, which I think
> > excludes the possibility of using int-out, since the XTAL2 pin is an
> > input.
> > 
> > There's 3 situations:
> > - no external clock provided
> > - ext-clk provided
> > - xtal provided
> > 
> > For the former, you know you're in that state when no "clocks" property
> > is present. The latter two you can differentiate based on "clock-names".
> > 
> > Choosing to use the internal clock if an external clock is provided
> > seems to be a software policy decision, unless I am mistaken.
> 
> Agreed, though it rarely makes sense as if someone put down a precision
> clock they normally wanted you to use it!
> 
> So as a general rule we don't both providing policy controls beyond if
> there is extra hardware (external clock source) then use that.
> 
> If someone has a good reason to want to do something else then we can
> probably figure out a reasonable way to control it.

Yah, I figured there'd be few situations, outside of maybe debugging
hardware issues, where that internal clock is more desirable to use.

> > > > Finally, if the ADC has a clock output, why can that not be represented
> > > > by making the ADC a clock-controller?
> > > >   
> > > 
> > > Was not familiar with this/did not cross my mind. So if xtal/ext-clk is
> > > present, the driver should detect it and disable the option for clock
> > > output? (Common pin XTAL2)  
> > 
> > Yeah, if those clocks are provided you would not register as a clock
> > controller. If there is a user of the output clock, it should have its
> > own "clocks" property that references the ADC's output.
> > 
> > Your dt-binding could also make clocks/clock-names & clock-controller
> > mutually exclusive.
> 
> That would indeed be the nicest solution.  How this has been done
> in drivers has somewhat 'evolved' over time, but this is the nicest
> option from point of view of standard bindings and clarity over what
> is going on.

Yeah, I know that this has not really been normal thing to do in some
corners of the kernel (ethernet PHYs in particular I think have been
rolling their own clock controller stuff) and I've been trying to push
back on these kinds of things for new devices.

Thanks,
Conor.

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]