Re: [PATCH v8 1/2] dt-bindings: adc: add AD7173

[Jonathan Cameron <jic23@kernel.org>]

On Sun, 17 Dec 2023 19:00:32 -0600
David Lechner <dlechner@baylibre.com> wrote:

> On Sun, Dec 17, 2023 at 7:50 AM Jonathan Cameron <jic23@kernel.org> wrote:
> >
> > On Thu, 14 Dec 2023 19:03:28 +0200
> > Ceclan Dumitru <mitrutzceclan@gmail.com> wrote:
> >  
> > > On 12/14/23 18:12, David Lechner wrote:  
> > > > On Thu, Dec 14, 2023 at 1:43 PM Ceclan Dumitru <mitrutzceclan@gmail.com> wrote:  
> > > >> On 12/12/23 17:09, David Lechner wrote:  
> > > >>> On Tue, Dec 12, 2023 at 11:45 AM Dumitru Ceclan <mitrutzceclan@gmail.com> wrote:  
> > >  
> > > >> ...
> > > >>  
> > > >>>> +  interrupts:
> > > >>>> +    maxItems: 1  
> > > >>>
> > > >>> Shouldn't this be 2? The datasheet says there is a "Data Output Ready"
> > > >>> signal on the DOUT/RDY pin and an "Error Output" on the SYNC/ERROR
> > > >>> pin. Although I could see how RDY could be considered part of the SPI
> > > >>> bus. In any case, a description explaining what the interrupt is would
> > > >>> be useful.
> > > >>>  
> > > >>
> > > >> I do not see how there could be 2 interrupts. DOUT/RDY is used as an
> > > >> interrupt when waiting for a conversion to finalize.
> > > >>
> > > >> Sync and Error are sepparate pins, Sync(if enabled) works only as an
> > > >> input that resets the modulator and the digital filter.  
> > > >
> > > > I only looked at the AD7172-2 datasheet and pin 15 is labeled
> > > > SYNC/ERROR. Maybe they are separate pins on other chips?  
> > >
> > > Yep, sorry, missed that. All other supported models have them separate.  
> >
> >  
> > > >  
> > > >>
> > > >> Error can be configured as input, output or ERROR output (OR between all
> > > >> internal error sources).
> > > >>
> > > >> Would this be alright
> > > >>   interrupts:
> > > >>
> > > >>     description: Conversion completion interrupt.
> > > >>                  Pin is shared with SPI DOUT.
> > > >>     maxItems: 1  
> > > >
> > > > Since ERROR is an output, I would expect it to be an interrupt. The
> > > > RDY output, on the other hand, would be wired to a SPI controller with
> > > > the SPI_READY feature (I use the Linux flag name here because I'm not
> > > > aware of a corresponding devicetree flag). So I don't think the RDY
> > > > signal would be an interrupt.
> > > >  
> > >
> > > Error does not have the purpose to be an interrupt. The only interrupt
> > > used from this chip is the one from the DOUT/~RDY pin. Sure, it is wired
> > > to the SPI controller, but when you can't also receive interrupts on
> > > that very same CPU pin an issue arises. So that pin is also wired to
> > > another GPIO with interrupt support.  
> >
> > You've lost me.  It's a pin that has a state change when an error condition
> > occurs.  Why not an interrupt?  Doesn't matter that the driver doesn't
> > use this functionality. dt-bindings should be as comprehensive as possible.
> > Given it's a multipurpose pin you'd also want to support it as a gpio to be
> > complete alongside the other GPIOs.
> >  
> > >
> > > This is the same way that ad4130.yaml is written for example (with the
> > > exception that ad4130 supports configuring where the interrupt is routed).
> > >
> > > In regards to SPI_READY _BITUL(7) /* slave pulls low to pause */: the
> > > ad_sigma_delta framework (if it can be called that) is written to expect
> > > a pin interrupt, not to use SPI_READY controller feature.  
> >
> > SPI_READY is supported by only a couple of controllers. I'm not even that
> > sure exactly how it is defined and whether that lines up with this usecase.
> > From some old asci art. https://lore.kernel.org/all/1456747459-8559-1-git-send-email-linux@rempel-privat.de/
> >
> > Flow control: Ready Sequence
> > Master CS   |-----1\_______________________|
> > Slave  FC   |--------2\____________________|
> > DATA        |-----------3\_________________|
> >
> > So you set master and then wait for a flow control pin (the ready signal) before
> > you can actually talk to the device.
> >
> > Here we are indicating data is ready to be be read out.
> >
> > So I don't 'think' SPI_READY applies.
> >
> > Jonathan
> >  
> 
> I'm not arguing that SPI_READY applies in this particular case, but
> FWIW it does seem to me like...
> 
> 1) SPI_READY could be implemented in any SPI driver using a GPIO
> interrupt (similar to how we already have GPIO CS)
> 2) In cases where the SPI controller does have actual hardware support
> for SPI_READY and the ADC chip A) uses CS to trigger a conversion and
> B) has a "busy" signal that goes low when the conversion is complete,
> then the SPI_READY feature could be used to make reading sample data
> more efficient by avoiding any CPU intervention between CS assertion
> and starting the data xfer due to waiting for the conversion to
> complete either by waiting for an interrupt or sleeping for a fixed
> amount of time.

Agreed that SPI_READY is possible if an SPI controller uses GPIOs for
CS and that signal.  If not a GPIO for CS then the SPI_READY should
also be hardware managed.

I could potentially be adapted to this sort of case if conditions
like the CS being active before the ready is set is taking into account.

This is a bit like SPI in general - far too many things that could
be built and no particular standards for them.

Jonathan