[PATCH 0/3] iio: cleanup masklength usage

[Jonathan Cameron]

On Fri, 26 Apr 2024 10:26:31 -0500
David Lechner <dlechner at baylibre.com> wrote:

> On Fri, Apr 26, 2024 at 2:13 AM Nuno Sá <noname.nuno at gmail.com> wrote:
> >
> > On Thu, 2024-04-25 at 10:03 -0500, David Lechner wrote:  
> > > While working on other patches I noticed that a few drivers are setting
> > > the masklength field of struct iio_dev even though it is marked as
> > > [INTERN]. It looks like maybe this was not always the case, but we can
> > > safely clean it up now without breaking anything.
> > >
> > > ---
> > > David Lechner (3):
> > >       iio: adc: ad7266: don't set masklength
> > >       iio: adc: mxs-lradc-adc: don't set masklength
> > >       iio: buffer: initialize masklength accumulator to 0
> > >
> > >  drivers/iio/adc/ad7266.c          | 1 -
> > >  drivers/iio/adc/mxs-lradc-adc.c   | 1 -
> > >  drivers/iio/industrialio-buffer.c | 2 +-
> > >  3 files changed, 1 insertion(+), 3 deletions(-)
> > > ---
> > > base-commit: b80ad8e3cd2712b78b98804d1f59199680d8ed91
> > > change-id: 20240425-b4-iio-masklength-cleanup-86b632b19901
> > >  
> >
> > Hi David,
> >
> > Nice cleanup. The patches look good to me but there's one thing missing :). As you
> > correctly noted, the field should be internal to the IIO core and drivers should not
> > touch it. Hence, you need to make sure not driver is using it so we can move it into
> > struct iio_dev_opaque [1]. That's the place all the intern fields should, eventually,
> > end up.
> >
> > Now, quite some drivers in the trigger handler will read the masklength for looping
> > with for_each_set_bit(). Hence, the straight thing would be an helper to get it.
> > Maybe there's a clever way...
> >
> > I know this is more work than what you had in mind but I think it should be fairly
> > simple (hopefully) and since you started it :), maybe we can get the whole thing done
> > and remove another [INTERN] member from the iio_dev struct.
> >
> > [1]: https://elixir.bootlin.com/linux/latest/source/include/linux/iio/iio-opaque.h#L42
> >
> > - Nuno Sá  
> 
> Sounds like fun. :-p
> 
> I will look into it.

I think this one might be miss marked as [INTERN]. It should be constant from the driver
point of view, but given active_scan_masks is meant to be visible to the driver,
it's length should probably be as well.

Sure every driver should be able to trivially work this out for themselves, but
do we care about stopping them using this?

It might be worth some nice iterator wrappers with names like
iio_for_each_active_channel() though I'd expect those to still be accessing these
fields directly as this is a high performance path so we don't want to to bounce
though a core function to get them.

Jonathan