Re: [RFC PATCH 0/3] Add converter framework

["Nuno Sá" <noname.nuno@gmail.com>]

On Wed, 2023-08-30 at 17:29 +0100, Jonathan Cameron wrote:
> On Fri, 4 Aug 2023 16:53:38 +0200
> Nuno Sa <nuno.sa@analog.com> wrote:
> 
> > This is the initial RFC following the discussion in [1]. I'm aware this is
> > by no means ready for inclusion and it's not even compilable since in
> > the RFC I did not included the patch to add component_compare_fwnode()
> > and component_release_fwnode(). 
> 
> Whilst I haven't read this through yet, I suspect Olivier will be able to
> offer some insight on some of this and likewise you may be able to
> point out pitfalls etc in his series (I see you did some review already :)
> 
> https://lore.kernel.org/linux-iio/20230727150324.1157933-1-olivier.moysan@foss.st.com/
> 
> Both are about multiple interacting components of an overall datapath.
> Whether there is commonality isn't yet clear to me.
> 

I made a very general comment in that series but I need to look better at it. Not
sure if we can merge them together but let's see...

> > 
> > The goal is to have a first feel on the
> > direction of the framework so that if I  need to drastically change it,
> > better do it now. The RFC also brings the ad9647 and the axi_adc core to
> > the same functionality we have now upstream with one extra fundamental
> > feature that is calibrating the digital interface. This would be very
> > difficult to do with the current design. Note that I don't expect any
> > review on those drivers (rather than things related to the framework). 
> > 
> > I also want to bring up a couple of things that I've
> > been thinking that I'm yet not sure about (so some feedback might make
> > mind in one direction or another).
> > 
> > 1) Im yet not sure if I should have different compatibles in the
> > axi-adc-core driver. Note this soft core is a generic core and for every
> > design (where the frontend device changes or has subtle changes like
> > different number of data paths) there are subtle changes. So, the number
> > of channels might be different, the available test patterns might be
> > different, some ops might be available for some designs but not for
> > others, etc... 
> 
> I don't suppose there is any chance Analog can make at least some of this
> discoverable from the hardware?  Capability registers etc in the long
> run. Can't fix what is already out there.
> 

Well, it is a soft core so my naive assumption is that it's doable if some HDL guy is
willing to implement it. But yes, it might get supported only for new designs.

> > With a different compatible we could fine tune
> > those differences (with a chip_info like structure) and pass some const
> > converter_config to the framework that would allow it to do more safety
> > checks and potentially reduce the number of converter_ops.
> > OTOH, starting to add all of these compatibles might become messy in the
> > long run and will likely mean that we'll always have to change both
> > drivers in order to support a new frontend. And the frontend devices
> > should really be the ones having all the "knowledge" to configure the
> > soft core even if it means more converter_ops (though devicetree might
> > help as some features are really HW dependent). I more inclined to just
> > leave things as-is in the RFC.
> 
> I'm fine with putting this stuff in DT where possible.
> 
> > 
> > 2) There are some IIO attributes (like scale, frequency, etc) that might
> > be implemented in the soft cores. I still didn't made my mind if I should just
> > have a catch all read_raw() and write_raw() converter_ops or more fine
> > tuned ops. Having the catch all reduces the number of ops but also makes
> > it more easier to add stuff that ends up being not used anymore and then
> > forgotten. There are also cases (eg: setting sampling frequency) where
> > we might need to apply settings in both the frontend and the backend
> > devices which means having the catch all write_raw() would be more
> > awkward in these case. I'm a bit more inclined to the more specific ops. 
> 
> It's the kernel - we can always change the internal API later as long as we
> don't touch the user space part.  Go with your gut feeling today and
> if it changes this sort of refactor usually isn't that bad.
> 

Agreed...

> 
- Nuno Sá