Message-ID: <20250819192851.6179e598@jic23-huawei>
Date: Tue, 19 Aug 2025 19:28:51 +0100
From: Jonathan Cameron <jic23@...nel.org>
To: Ben Collins <bcollins@...nel.org>
Cc: David Lechner <dlechner@...libre.com>, Nuno Sá
 <nuno.sa@...log.com>, Andy Shevchenko <andy@...nel.org>,
 linux-iio@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH v5 5/5] iio: mcp9600: Add support for IIR filter

On Mon, 18 Aug 2025 16:00:20 -0400
Ben Collins <bcollins@...nel.org> wrote:

> On Mon, Aug 18, 2025 at 08:10:35PM -0500, Jonathan Cameron wrote:
> >   
> > > > >  	case IIO_CHAN_INFO_SCALE:
> > > > >  		*val = 62;
> > > > >  		*val2 = 500000;
> > > > >  		return IIO_VAL_INT_PLUS_MICRO;
> > > > > +    
> > > > If you want the extra space put it in previous patch.
> > > >     
> > > > >  	case IIO_CHAN_INFO_THERMOCOUPLE_TYPE:
> > > > >  		*val = mcp9600_tc_types[data->thermocouple_type];
> > > > >  		return IIO_VAL_CHAR;
> > > > > +
> > > > > +	case IIO_CHAN_INFO_LOW_PASS_FILTER_3DB_FREQUENCY:
> > > > > +		if (data->filter_level == 0)    
> > > > 
> > > > Return the current requested value. An error is just going to confuse
> > > > someone who tried to write this before enabling the filter and then
> > > > checked to see if the write was successful.    
> > > 
> > > I could not get a concensus on this. On the one hand, if a user sets a
> > > value here, would they not assume that the filter was enabled? What
> > > about cases where a filter_type can be more than one valid type with
> > > different available coefficients for each? What should it show then?  
> > 
> > So I was thinking of this like other things with 'enables' such as events.
> > For those you always set the value first.  They don't really have a type
> > field though (well they do but the ABI allows multiple at once unlike filters
> > so we end up with a quite different looking ABI).
> > 
> > Agreed it gets challenging with multiple filter types. If it weren't for
> > advertising the range I'd suggest just stashing whatever was written and
> > then mapping it to nearest possible when the filter type is set.
> > That's what the ad7124 does for changing between filters anyway
> > though oddly it doesn't seem to have a control for filter type.
> > 
> > This is a good argument against the whole 'none' value for filter type
> > - that's not much used so we could deprecate it for new drivers.
> > 
> > I'm not particularly keen on filter_enable but seems we are coming back
> > around to that option to avoid this corner case.  Alternative being what
> > you have here which isn't great for ease of use.  
> 
> I'm somewhat wondering if the filter frequency and frequency_available
> attributes should not even show in sysfs unless the filter_type was
> something other than "none".
> 
I'm not keen on that and trying to bolt is_visible into the mess of how
we generate attributes would be hard and actual add and remove of attributes
is horrible for races with userspace.

> > So for next version let's go for that. Make sure to include Documentation
> > in a separate patch though so it's easy to see an poke holes in.  
> 
> Just to make sure I understand, you'd like to see a filter_enable
> attribute and filter_type would not contain "none", then frequency and
> frequency_available would always show something for whatever was in
> filter_type?

Yes.  I think that is best way forwards.  If we want to retrofit the one
user of none to support the new ABI as well it should be easy to do.

> 
> > ABI design is a pain sometimes.  
> 
> The epitome of being able to paint yourself into a corner.
> 
Yup.

J

Powered by blists - more mailing lists