On Mon, Oct 24, 2011 at 03:48, Linus Walleij wrote:

On Mon, Oct 24, 2011 at 9:36 AM, Grant Likely wrote:

quoted

On Mon, Oct 24, 2011 at 09:26:38AM +0200, Linus Walleij wrote:

(...)

quoted

quoted

I was more thinking along the lines of one device per GPIO controller,
then you ioctl() to ask /dev/gpio0 how many pins it has or so.

And there is also the question of whether it is even a good idea to
export pinctrl manipulation to userspace.

The application I've seen is in automatic control.

I think people do things like connect they GPIO pins to electrical
relays, plus on top of that they use all the stuff in drivers/staging/iio.

All that from userspace. Controlling entire factories and industrial
robots, weapon systems too, I'm afraid.

The control of these dangerous things runs on a realtime-patched
kernel, in a single userspace app with a few threads and they have
done some realtime-tetris scheduling the beast more or less
manually with SCHED_FIFO. Basically that app is all that runs on
the board, and its threads take precedence over everything else
on the system.

That is the typical beast that is poking around on the GPIO sysfs
interfaces...

we all agree that GPIO from userspace makes sense.  the only complaint
i've seen so far against the GPIO sysfs interface that should be
addressed is the performance overhead.

but the question here is about pinctrl.  does userspace really need to
manipulate the pinmapping ?  if we agree on that, then the question is
on the userspace interface.

assuming we want this, i can't see the performance argument being made
here for pinctrl.  which means doing a sysfs interface here like we
already have with GPIO makes the most sense.  GPIO deals in "binary"
data for the most part (reading/writing 0/1 ints) so the string-based
sysfs parsing is a bit weird, but pinctrl deals with strings
everywhere for selecting mapping groups, so sysfs is the natural
answer.
-mike