From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <SRS0=KP4v=CG=linux.intel.com=mika.westerberg@kernel.org>
Date: Wed, 8 Nov 2017 19:03:57 +0200
From: Mika Westerberg <mika.westerberg@linux.intel.com>
To: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Cc: Jonathan Cameron <Jonathan.Cameron@huawei.com>,
	Jonathan Cameron <jic23@kernel.org>, linux-iio@vger.kernel.org,
	Hartmut Knaack <knaack.h@gmx.de>,
	Lars-Peter Clausen <lars@metafoo.de>,
	Peter Meerwald-Stadler <pmeerw@pmeerw.net>,
	Linus Walleij <linus.walleij@linaro.org>
Subject: Re: [PATCH v3 3/5] iio: proximity: sx9500: Set IRQ pin to
 direction-input if necessary
Message-ID: <20171108170357.GN18997@lahna.fi.intel.com>
References: <20171103130340.42459-1-andriy.shevchenko@linux.intel.com>
 <20171103130340.42459-3-andriy.shevchenko@linux.intel.com>
 <20171104032059.0000049a@huawei.com>
 <1510158946.25007.112.camel@linux.intel.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
In-Reply-To: <1510158946.25007.112.camel@linux.intel.com>
List-ID: <linux-iio@vger.kernel.org>

On Wed, Nov 08, 2017 at 06:35:46PM +0200, Andy Shevchenko wrote:
> +Cc: Mika
> 
> On Sat, 2017-11-04 at 03:20 +0000, Jonathan Cameron wrote:
> > On Fri, 3 Nov 2017 15:03:38 +0200
> > Andy Shevchenko <andriy.shevchenko@linux.intel.com> wrote:
> > 
> > > With the new more strict ACPI gpio code the DSDT's IoRestriction
> > > flags
> > > are honored on gpiod_get(), but in some DSDT's it is wrong, so
> > > explicitly call gpiod_direction_input() on the IRQ GPIO if
> > > necessary.
> > > 
> > > Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
> > 
> > Again, I really really don't like filling driver code with fixes
> > for broken firmware.  I appreciate we have to cope with this, but
> > it does rather seem like this should be moved into the core code
> > for say gpiod_get_irq.
> 
> I would love to fix in general, though it looks not so trivial:
> 
> - gpiod_get() doesn't know if GPIO is going to be used as IRQ
> - gpiod_to_irq() doesn't know if descriptor in question comes from
> GpioIo() ACPI resource

One idea is to allow this strict mode to be relaxed by drivers perhaps
by passing quirks through struct acpi_gpio_mapping:

static const struct acpi_gpio_mapping acpi_foo_gpios[] = {
	/*
	 * This platform has a bug in ACPI GPIO description making IRQ
	 * GPIO to be output only. Ask the GPIO core to ignore this
	 * limit.
	 */
	{ "foobar-gpios", &foobar_gpios, 1, ACPI_QUIRK_IGNORE_IO_RESTRICTION },
	{},
};

or something like that. Not sure if I missed something obvious, though.