Re: [PATCH v2 1/1] counter: cros_ec: Add synchronization sensor

[Gwendal Grignou]

I resend a counter driver for the EC at
https://patchwork.kernel.org/patch/11479437/

I tried a timestamp only IIO sensor, but this is not allowed, as the
timestamp channel is very specific: no extended parameters can be
added.
I did not add a COUNTER_COUNT_TALLY type, as a newer function
COUNTER_COUNT_FUNCTION_INCREASE, fits the counter better.
I am still using IIO_COUNT (inspired by the st counter driver) to
gather the event in real time on the iio side.

Gwendal.


On Mon, Nov 11, 2019 at 5:16 PM William Breathitt Gray
<vilhelm.gray@xxxxxxxxx> wrote:
>
> On Mon, Nov 11, 2019 at 11:49:55AM +0000, Jonathan Cameron wrote:
> > On Sun, 10 Nov 2019 10:14:08 -0500
> > William Breathitt Gray <vilhelm.gray@xxxxxxxxx> wrote:
> >
> > > On Tue, Sep 24, 2019 at 04:20:51PM +0200, Fabien Lahoudere wrote:
> > > > Hi all,
> > > >
> > > > After some discussions and investigation, the timestamp is very
> > > > important for that sync driver.
> > > > Google team uses that timestamp to compare with gyroscope timestamp.
> > > >
> > > > So the important data is timestamp and counter value is useless.
> > > > Just the event of counter increment is important to get a timestamp.
> > > >
> > > > In that case, my idea was to just use an IIO driver with a single
> > > > channel with IIO_TIMESTAMP. We discuss this here and it seems
> > > > controversial.
> > > >
> > > > So my question to Jonathan is if we have a timestamp coming from the EC
> > > > itself, can we consider this timestamp as a good IIO driver?
> > > >
> > > > Any other idea is welcome, however Google team would like to manage
> > > > only IIO drivers if possible.
> > > >
> > > > Thanks
> > >
> > > Jonathan,
> > >
> > > Should the the timestamp from the EC be introduced as an IIO driver
> > > using IIO_TIMESTAMP?
> >
> > It is is a rather odd driver but I suppose it would be fine with lots
> > of clear docs on why it is how it is...
> >
> > >
> > > Since there is no corresponding EC Counter driver in the baseline right
> > > now we don't have a conflict yet. If the EC timestamp is introduced as
> > > an IIO driver then we should make any future EC Counter driver mutually
> > > exclusive with the IIO driver in order to prevent any memory space
> > > conflict. At that point we may deprecate the IIO driver and move the
> > > timestamp functionality to the corresponding Counter driver.
> >
> > That route does become somewhat of a mess so I suspect we'd have to have
> > a single driver supporting both userspace interfaces.  If you are happy
> > that we'd be adding a bit of legacy to support for ever then we can go
> > that way.
>
> Generally I'd prefer all components of a device to be supported, but
> if this is as Fabien suggests that due to the nature of this particular
> device the counter value is of no interest, then a Counter driver is of
> little practical use here. In this particular case, it seems better to
> restrict the driver support to just the timestamp functionality that
> will be used, rather than introduce extra code to expose values that
> will likely be ignored and risk adding code to the kernel that becomes
> unmaintained due to lack of exposure or interest.
>
> William Breathitt Gray
>
> >
> > >
> > > That's assuming someone is interested in the Count component enough to
> > > implement an EC Counter driver; otherwise, the IIO driver will serve
> > > just fine if timestamp is the only data desired from this device.
> > >
> > > William Breathitt Gray
> >
> >