Re: [PATCH v3 3/5] iio: accel: mma9551: Add support to get irqs directly from fwnode

[Jonathan Cameron <jic23@jic23.retrosnub.co.uk>]

On Sun, 5 Dec 2021 22:39:07 +0200
Andy Shevchenko <andy.shevchenko@gmail.com> wrote:

> On Sun, Dec 5, 2021 at 9:57 PM Jonathan Cameron <jic23@kernel.org> wrote:
> >
> > From: Jonathan Cameron <Jonathan.Cameron@huawei.com>
> >
> > Note the interrupt type should be specified by firmware, not the driver
> > so that is also dropped.
> >
> > Drop previous gpio based retrieval method. Whilst in theory this
> > might cause problems with direction if anyone is using ACPI GioIo().  
> 
> GpioIo()

My proofreading fails me again...

> 
> > As Andy described in v1, such a situation would typically reflect
> > a pin that is actually used in both directions (not true here)
> > or missdesigned ACPI tables.  
> 
> ...
> 
> > -               gpio = devm_gpiod_get_index(dev, NULL, i, GPIOD_IN);
> > -               if (IS_ERR(gpio)) {
> > -                       dev_err(dev, "acpi gpio get index failed\n");
> > -                       return PTR_ERR(gpio);
> > -               }
> > -
> > -               ret = gpiod_to_irq(gpio);
> > -               if (ret < 0)  
> 
> > +               ret = fwnode_irq_get(dev_fwnode(dev), i);
> > +               if (ret)
> >                         return ret;  
> 
> I don't remember why we decided that this gonna work, because
> fwnode_irq_get() is not an equivalent to the above, more precisely in
> ACPI case it only covers the GSIs (Global System Interrupts) which in
> such case may or may not be GPIOs. On x86 it's usually direct IOxAPIC
> ones.

This isn't accessed as a GPIO, but I kind of assume someone was using
it with one given this code.

I think I misread what you'd written in reply to v1.
Thanks for the explanation.

> 
> So, this conversion would probably make it impossible to use this
> device in the ACPI case.

oops :)


> 
> See also this discussion:
> https://lore.kernel.org/lkml/20211109200840.135019-1-puranjay12@gmail.com/T/#u
> 

I'd not made the connection that the non named one was also missing most
of the ways it could be specified.  Ah well, this will take a little while
to get a test framework in place to poke at it properly.  Maybe one for
a bored moment over the festive season.

Jonathan