Re: [PATCH v1 2/3] Documentation: firmware-guide: gpio-properties: active_low only for GpioIo()

[Andy Shevchenko <andy.shevchenko@xxxxxxxxx>]

On Thu, Oct 29, 2020 at 07:58:39PM +0100, Ricardo Ribalda wrote:
> On Thu, Oct 29, 2020 at 7:13 PM Andy Shevchenko
> <andy.shevchenko@xxxxxxxxx> wrote:
> > On Thu, Oct 29, 2020 at 8:10 PM Ricardo Ribalda <ribalda@xxxxxxxxxx> wrote:
> > > On Thu, Oct 29, 2020 at 6:32 PM Ricardo Ribalda <ribalda@xxxxxxxxxx> wrote:
> > > > On Thu, Oct 29, 2020 at 6:26 PM Andy Shevchenko
> > > > <andy.shevchenko@xxxxxxxxx> wrote:
> > > > > On Thu, Oct 29, 2020 at 7:20 PM Andy Shevchenko
> > > > > <andy.shevchenko@xxxxxxxxx> wrote:
> > > > > > On Thu, Oct 29, 2020 at 7:17 PM Andy Shevchenko
> > > > > > <andy.shevchenko@xxxxxxxxx> wrote:

...

> > > > > > > > Let me try to explain myself again:
> > > > > > > >
> > > > > > > > I have a gpio pin that produces IRQs on both edges. so ActiveLevel is Both
> > > > > > > >
> > > > > > > > The problem is that the value of that pin is inverted: Low means 1 and
> > > > > > > > high means 0.
> > > > > > > >
> > > > > > > > How can I describe that the pin "is inverted" without using the _DSD field?
> > > > > > >
> > > > > > > "Both edges" and "inverted" or "polarity low" in one sentence make no sense.
> > > > > >
> > > > > > To be on the constructive side, I can *imagine* so badly designed
> > > > > > hardware that uses level and edge at the same time, but before I go to
> > > > > > conclusions, can you share relevant (pieces of) datasheet?
> > > > >
> > > > > The [1] is a real example of how GPIO is being used to detect changing
> > > > > of current level of the signal.
> > > > > Note, ACPI tables for that device have problems [2], but I guess you
> > > > > may get the idea.
> > > >
> > > >
> > > > This is exactly what I need to do. Get an IRQ whenever the value
> > > > changes. But the pin is "inverted"
> > > >
> > > > This is the "schematic" :  https://ibb.co/f8GMBbP . I want to pass to
> > > > userspace a "1" when the switch is closed and "0"  when it is open.
> > > >
> > > And there are also other devices where the swith works the other way
> > > around, so the acpi should be verbose enough to describe both
> > > situations.
> > >
> > > With my proposal (use the same active_low field as with GpioIO) we
> > > cover both usecases.
> >
> > Even without your proposal it's feasible.
> > You see, the problem here is that if you describe GPIO as Interrupt,
> > the edge and level together make complete nonsense.
> >
> > Solution: do *not* describe it as Interrupt.
> 
> Now I get my mistake:
> 
> I thought that gpiod_to_irq will not work unless it was a GpioInt()
> but it works fine. So in this case I will just convert it to that.

It's actually that gpio_to_irq() is solely for GPIOs which initially are not IRQs.

> Could we say that doing gpiod_get_value() from a GpioInt() is always
> wrong?

But it's not wrong. Some cases simply make little or no sense, but in principal
why not? Yes, it may be fragile or too much customized.

> Can we modify the code to avoid it?

GpioInt() is orthogonal to GPIO APIs in Linux kernel. It close to be
impossible. Also see above.

> Sorry for the confusion and thanks for your help.

No problem, you're welcome, it's good that you started a discussion!

> > > > > [1]: https://elixir.bootlin.com/linux/latest/source/drivers/extcon/extcon-intel-int3496.c#L138
> > > > > [2]: https://elixir.bootlin.com/linux/latest/source/drivers/extcon/extcon-intel-int3496.c#L45

-- 
With Best Regards,
Andy Shevchenko