Re: [PATCH 2/2] iio: light: Add support for ADUX1020 sensor

[Jonathan Cameron <jic23@kernel.org>]

On Wed, 9 Oct 2019 10:21:27 +0000
"Ardelean, Alexandru" <alexandru.Ardelean@analog.com> wrote:

> On Wed, 2019-10-09 at 15:15 +0530, Manivannan Sadhasivam wrote:
> > [External]
> > 
> > Hi Ardelean,

For some reason, my email client decided not to filter this thread
correctly so I didn't realise so much discussion had gone on when
I applied the newer version earlier today.  Oops.  Hopefully
there was nothing major outstanding.  Let me know if there was
as it's not yet in a non rebasing tree...

I've cropped to just where my name got mentioned ;)

..

> >   
> > > - Just curios here: there is gesture mode as well; will that be
> > > implemented
> > > later? Or will there be other modes implemented?  
> > 
> > Currently only proximity mode is implemented. There are gesture and
> > sample
> > modes and I left those as a TODO. But I'm not sure whether IIO is
> > supporting
> > gesture mode properly or not.  
> 
> I don't have any input on this at the moment [about gesture support & IIO].
> I'd have to investigate.
> Maybe Jonathan has some thoughts.

Properly is a hard term for gesture support.  The issue has always
been that every device does it slightly differently.  There are
way too many types of gesture that a device 'might' use.

We do have some drivers (IIRC) doing some gesture sensing, but you may
well find places where things need to expand!

...
> > > > +static int adux1020_read_raw(struct iio_dev *indio_dev,
> > > > +			     struct iio_chan_spec const *chan,
> > > > +			     int *val, int *val2, long mask)
> > > > +{
> > > > +	struct adux1020_data *data = iio_priv(indio_dev);
> > > > +	u16 buf[3];  
> > > 
> > > This buffer looks a bit weird. [8]
> > > It's 3 elements-wide and passed without any information about size.
> > > And only the first element is used.
> > > So, maybe just convert u16 buf[3] -> u16 buf?
> > >   
> > 
> > The buffer declaration is based on the hardware buffer available. It
> > is 3 elements wide since the remaining 2 elements will be used by other
> > modes. The idea here is to reuse the adux1020_measure() API for all 3
> > modes (which has varying buffer sizes).  
> 
> The only thought I have left about this buffer [and forgot to mention it
> earlier], is whether this should be cacheline aligned [or not].
> If it has to be, then maybe it shouldn't be stored on the stack and moved
> to a malloc-ed buffer [on "struct adux1020_data"].
> Cacheline aligned stuff typically deals with potential DMA issues. The DMA
> issues [in this case] could be coming from i2c controllers that can do DMA.
> 
> Jonathan may have more input here.
> 
The i2c subsystem in general doesn't assume that buffers are dma safe
though it would like to ;)

Wolfram did a good presentation on his efforts to sort that out at
ELCE 2018

https://www.youtube.com/watch?v=JDwaMClvV-s