Re: [PATCH 0/3] Introduce a counter inkernel API

[William Breathitt Gray]

On Sun, Apr 12, 2020 at 01:31:45AM +0200, Alexandre Belloni wrote:
> Hi,
> 
> On 11/04/2020 13:22:59-0400, William Breathitt Gray wrote:
> > I'm not inherently opposed to adding an in-kernel API for the Counter
> > subsystem, but I'm not sure yet if it's necessary for this particular
> > situation.
> > 
> > Is the purpose of this driver to allow users to poll on the rotary
> > encoder position value? If so, perhaps instead of an in-kernel API, the
> > polling functionality should be added as part of the Counter subsystem;
> > I can see this being a useful feature for many counter devices, and
> > it'll keep the code contained to a single subsystem.
> > 
> > By the way, I'm going to be submitting a major update to the Counter
> > subsystem code in the next couple weeks that isolates the sysfs code
> > from the rest of the subsystem -- it'll likely affect the interface and
> > code here -- so I'll probably wait to decide for certain until that
> > patch lands; I anticipate it making things easier for you here after
> > it's merged.
> > 
> > For now, I want to get a better high-level understanding about how users
> > would interact with this driver to use the device (input_setup_polling
> > is a new call for me). That should help me understand whether an
> > in-kernel API is the best choice here.
> > 
> 
> Well, the goal is not really polling the counters but mainly exposing
> the correct userspace interface for the rotary encoders that are
> connected to quadrature decoders.
> 
> The input driver is using polling because this reduces the complexity of
> the patches but the ultimate goal is to also have interrupts working.

Okay, I think understand now. Interrupt support is another feature I
want to get working for counters too, so that development will probably
overlap with this driver as well. Hopefully with interrupts working
you'll be able to signal to the input driver whenever data is ready,
rather than just polling periodically to check.

> I'm pretty sure the in-kernel interface can also have other usages like
> for example iio triggers. I could envision having to read an ADC after x
> turns of a motor to check for the torque.

That's an interesting use case. I can see how an in-kernel interface
would be helpful here.

> I also think that having the sysfs code separate would help as it could
> be considered as one of the in-kernel interface user.
> 
> BTW, do you have plans to add a character device interface?

Yes, actually a character device interface (and the timestamp feature)
is the primary motivation for this refactoring: sysfs code is separated
so that it can share a common core of functionality with the character
device code.

Implementing an in-kernel API should be trivial after these changes
since it will be just a matter of codifying the shared code that forms
the new core of the Counter subsystem. So perhaps this patchset should
wait until the Counter subsystem is updated, since it may be easier to
interact with counter devices once that is complete.

If you're curious about this patch, it's available on my personal iio
tree in the counter_chardev branch:
https://gitlab.com/vilhelmgray/iio/-/commits/counter_chardev/

I still need to add the character device code and respective userspace
API, but I expect to have it completed in the next couple weeks,
assuming no major issues or delays arise.

William Breathitt Gray

Attachment: signature.asc
Description: PGP signature