Re: [PATCH v5 06/10] dt-bindings: iio: accel: add interrupt-names

[Jonathan Cameron <jic23@kernel.org>]

On Fri, 6 Dec 2024 18:29:48 +0100
Lothar Rubusch <l.rubusch@gmail.com> wrote:

> On Fri, Dec 6, 2024 at 6:08 PM Conor Dooley <conor@kernel.org> wrote:
> >
> > On Thu, Dec 05, 2024 at 08:41:52PM +0100, Lothar Rubusch wrote:  
> > > On Thu, Dec 5, 2024 at 6:54 PM Conor Dooley <conor@kernel.org> wrote:  
> > > >
> > > > On Thu, Dec 05, 2024 at 05:13:39PM +0000, Lothar Rubusch wrote:  
> > > > > Add interrupt-names INT1 and INT2 for the two interrupt lines of the
> > > > > sensor. Only one line will be connected for incoming events. The driver
> > > > > needs to be configured accordingly. If no interrupt line is set up, the
> > > > > sensor will still measure, but no events are possible.
> > > > >
> > > > > Signed-off-by: Lothar Rubusch <l.rubusch@gmail.com>
> > > > > ---
> > > > >  .../devicetree/bindings/iio/accel/adi,adxl345.yaml         | 7 +++++++
> > > > >  1 file changed, 7 insertions(+)
> > > > >
> > > > > diff --git a/Documentation/devicetree/bindings/iio/accel/adi,adxl345.yaml b/Documentation/devicetree/bindings/iio/accel/adi,adxl345.yaml
Side note, but patch name must include what device it is!

dt-bindings: iio: accel: adxl345: ...

> > > > > index 280ed479ef5..67e2c029a6c 100644
> > > > > --- a/Documentation/devicetree/bindings/iio/accel/adi,adxl345.yaml
> > > > > +++ b/Documentation/devicetree/bindings/iio/accel/adi,adxl345.yaml
> > > > > @@ -37,6 +37,11 @@ properties:
> > > > >    interrupts:
> > > > >      maxItems: 1
> > > > >
> > > > > +  interrupt-names:
> > > > > +    description: Use either INT1 or INT2 for events, or ignore events.
> > > > > +    items:
> > > > > +      - enum: [INT1, INT2]  
> > > >
> > > > The description for this ", or ignore events" does not make sense. Just
> > > > drop it, it's clear what happens if you don't provide interrupts.
> > > >
> > > > However, interrupts is a required property but interrupt-names is not.
> > > > Seems rather pointless not making interrupt-names a required property
> > > > (in the binding!) since if you only add interrupts and not
> > > > interrupt-names you can't even use the interrupt as you do not know
> > > > whether or not it is INT1 or INT2?  
> > >
> > > What I meant is, yes, the sensor needs an interrupt line.
> > > Interrupt-names is optional. The sensor always can measure. When
> > > interrupt-names is specified, though, the sensor will setup a FIFO and
> > > can use events, such as data ready, watermark, single tap, freefall,
> > > etc. Without the interrupt-names, the sensor goes into a "FIFO bypass
> > > mode" without its specific events.  
> >
> > What I'm talking about here is how it is ultimately pointless for
> > interrupts to be a required property if it can never be used without
> > interrupt-names as you cannot know which interrupt is in use. I think
> > both should be made mandatory or neither.
> >  
> 
> Ah, now I can see your point. I agree that it should be equally
> mandatory as the interrupt. Legacy implementations used simply always
> just INT1. I'd like to make it configurable in the IIO driver but
> tried to avoid the DT topic for now (which was not a smart decision
> either). Hence, I added the interrupt-names.
> I'm unsure should I make "interrupt-names" a required property now?
> What about the existing DTS files using this sensor? There are no
> interrupt-names specified, so if made required, the missing
> interrupt-names there would break binding check, or not?

Neither should be required.  The driver isn't currently using interrupts
and I presume it is functional?  So I'd just drop the required on interrupts.
Now a condition that says interrupt-names is needed if interrupts is supplied
would be a useful addition (IIRC there are examples of that in tree).

So interrupts being required is a bug that we should fix by just
dropping that.  

Jonathan



> 
> > > Hence, I better drop the description entirely, since it rather seems
> > > to be confusing.