Re: [PATCH v4 2/2] iio: (bma400) add driver for the BMA400

[Dan Robertson]

Sorry for the incredibly late reply. Before I submit the next patchset version,
I have a question from the last set of reviews.

On Mon, Oct 21, 2019 at 04:20:16PM +0100, Jonathan Cameron wrote:
> On Sat, 19 Oct 2019 02:43:51 +0000
> Dan Robertson <dan@xxxxxxxxxxxxxxx> wrote:
> > On Fri, Oct 18, 2019 at 10:23:38AM +0300, Andy Shevchenko wrote:
> > > On Fri, Oct 18, 2019 at 6:44 AM Dan Robertson <dan@xxxxxxxxxxxxxxx> wrote:  
> > > > +static const int bma400_osr_table[] = { 0, 1, 3 };  
> > >   
> > > > +/* See the ACC_CONFIG1 section of the datasheet */
> > > > +static const int bma400_sample_freqs[] = {
> > > > +       12,  500000,
> > > > +       25,  0,
> > > > +       50,  0,
> > > > +       100, 0,
> > > > +       200, 0,
> > > > +       400, 0,
> > > > +       800, 0,
> > > > +};  
> > > 
> > > This can be replaced by a formula(s).  
> > 
> > Yeah I think I can implement the get, set, and read functions for sample_freq
> > with a formula, but the scale and sample frequency tables are needed by the
> > implementation of read_avail. A implementation of read_avail with a range and
> > a step would be ideal, but I couldn't find any documentation on implementing
> > read_avail where the step value of the range is a multiple. Please correct
> > me if I've missed something.
> 
> Indeed. We've only defined it as being fixed intervals.
> I'm not keen to expand the options for the userspace interface any
> further.  
> 
> You could compute the values at startup and store it in your state structure
> I think (or compute them on demand, but you'd need to have the space somewhere
> non volatile).
> 

I ended up writing an implementation that uses a formula for the get/set
functions of the sample frequency and scale, but uses a table for the
implementation of the read_avail function. While it does work, I worry
that this makes the driver less maintainable and would make it harder to
add support for a new hypothetical future BMA4xx device. Also, the majority
of drivers seem to use a table for the raw value to user input conversion,
so a move from this might make the code less "familiar".

If we do stick with the translation table, would it be better to have two
tables (a translation table and a read_avail table) so that we do not have
a step distance of two? This would mean we would need to maintain two
tables, but would simplify the code.

Random workflow question:

The sampling ratio, frequency, etc code seems to be the most complicated part
of the driver. Is it typically recommended to upstream a more minimal driver
that might assume the defaults?

Cheers,

 - Dan