Re: [PATCH 3/3] iio: adc: ti-ads112c14: add support for buffered read

[Jonathan Cameron <jic23@kernel.org>]

On Fri, 24 Jul 2026 09:47:34 -0500
David Lechner <dlechner@baylibre.com> wrote:

> On 7/23/26 6:25 PM, Jonathan Cameron wrote:
> > On Tue, 21 Jul 2026 18:23:58 -0500
> > David Lechner <dlechner@baylibre.com> wrote:
> >   
> >> On 7/19/26 7:55 PM, Jonathan Cameron wrote:  
> >>> On Tue, 14 Jul 2026 19:19:32 -0500
> >>> "David Lechner (TI)" <dlechner@baylibre.com> wrote:
> >>>     
> >>>> Add support for buffered reads using a triggered buffer.
> >>>>
> >>>> The device has a continuous conversion mode, but that can only be used
> >>>> with one channel at a time since there is nothing like a sequencer to
> >>>> support that in hardware. Instead, we use single-shot reads like we do
> >>>> for direct reads to be able to read multiple channels.
> >>>>
> >>>> Support for continuous conversion mode could be added in the future if
> >>>> needed via a 2nd buffer.    
> >>>
> >>> How about enabling that if only one channel is requested?  I vaguely recall
> >>> us doing that for another driver (though I might be dreaming :)  I did
> >>> see your comment in the cover letter about it affecting timing and that
> >>> making life complex.  Fine to leave considering this for another day
> >>> but maybe don't suggest a particular solution here.
> >>>     
> >> Actually, my latest thought it to do it by trigger type. This chip has
> >> a DRDY interrupt that can be used as the trigger for the continuous
> >> conversion mode but we need some software trigger to the single-shot
> >> mode. It think this would take care of the timing issues as well.
> >>
> >> The logic would be that if the self trigger (DRDY) is selected in
> >> the trigger/current_trigger attribute, then buffered read will use
> >> continuous mode and fail if more than one channel is enabled. If
> >> another (software) trigger is enabled, then use single-shot mode
> >> and allow multiple channels.  
> > 
> > Hmm. Bit unintuitive but can't really be helped.
> >   
> >>
> >> The timing issue is that (or will be since I haven't sent the relevant
> >> patches yet) there is a settling delay before the first sample. For
> >> single-shot mode, every sample is considered the first sample (because
> >> it is "single"), so the chip applies this settling time on every sample.
> >> So it might be good enough to just document that as a quirk of software
> >> triggers for this particular chip since they would be using single-shot
> >> mode?
> >>  
> > This is where it gets messy if settling time is reported. I guess acceptable
> > given it doesn't make that much sense for an external trigger anyway.
> > In theory if the settling time has to happen each single shot, we could
> > just merge it into sampling frequency (as no longer depends on whether
> > channel changes or not) but then we have to make that dependent on whether
> > it is the devices own trigger or not. Messy. I guess one to paper over
> > as a quirk / corner case and document as you say.  
> Yes, this is the direction I am leaning as well. The input chopping feature
> that we have been talking about has a similar issue. In addition to doubling
> the sample period, it also includes the settling time (2x as well) minus a
> few clock cycles. Even messier.
> 
For that one I think we need at least a readonly sysfs attribute to indicate
chopping is going on.  Then we tweak the settling time docs to include
a reference to chopping.

Jonathan

>