Date:	Mon, 26 Jan 2015 16:39:30 +0000
From:	Mark Rutland <mark.rutland@....com>
To:	Mika Westerberg <mika.westerberg@...ux.intel.com>
Cc:	Jiri Kosina <jkosina@...e.cz>,
	Benjamin Tissoires <benjamin.tissoires@...hat.com>,
	Rob Herring <robh+dt@...nel.org>,
	Pawel Moll <Pawel.Moll@....com>,
	Ian Campbell <ijc+devicetree@...lion.org.uk>,
	Kumar Gala <galak@...eaurora.org>,
	Jarkko Nikula <jarkko.nikula@...ux.intel.com>,
	"linux-input@...r.kernel.org" <linux-input@...r.kernel.org>,
	"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH 2/2] HID: i2c-hid: Add support for GPIO interrupts

On Mon, Jan 26, 2015 at 04:13:56PM +0000, Mika Westerberg wrote:
> On Mon, Jan 26, 2015 at 04:01:20PM +0000, Mark Rutland wrote:
> > On Mon, Jan 26, 2015 at 03:16:37PM +0000, Mika Westerberg wrote:
> > > On Mon, Jan 26, 2015 at 02:50:01PM +0000, Mark Rutland wrote:
> > > > On Mon, Jan 26, 2015 at 02:47:29PM +0000, Mika Westerberg wrote:
> > > > > On Mon, Jan 26, 2015 at 02:37:24PM +0000, Mark Rutland wrote:
> > > > > > On Mon, Jan 26, 2015 at 02:29:33PM +0000, Mika Westerberg wrote:
> > > > > > > The HID over I2C specification allows to have the interrupt for a HID
> > > > > > > device to be GPIO instead of directly connected to the IO-APIC.
> > > > > > > 
> > > > > > > Add support for this so that when the driver does not find proper interrupt
> > > > > > > number from the I2C client structure we check if the device has property
> > > > > > > named "gpios". This is then assumed to be the GPIO that serves as an
> > > > > > > interrupt for the device.
> > > > > > > 
> > > > > > > Signed-off-by: Mika Westerberg <mika.westerberg@...ux.intel.com>
> > > > > > > ---
> > > > > > >  .../devicetree/bindings/hid/hid-over-i2c.txt       |  5 +-
> > > > > > >  drivers/hid/i2c-hid/i2c-hid.c                      | 70 ++++++++++++++++------
> > > > > > >  2 files changed, 56 insertions(+), 19 deletions(-)
> > > > > > > 
> > > > > > > diff --git a/Documentation/devicetree/bindings/hid/hid-over-i2c.txt b/Documentation/devicetree/bindings/hid/hid-over-i2c.txt
> > > > > > > index 488edcb264c4..8f4a99dad3b9 100644
> > > > > > > --- a/Documentation/devicetree/bindings/hid/hid-over-i2c.txt
> > > > > > > +++ b/Documentation/devicetree/bindings/hid/hid-over-i2c.txt
> > > > > > > @@ -15,7 +15,10 @@ Required properties:
> > > > > > >  - reg: i2c slave address
> > > > > > >  - hid-descr-addr: HID descriptor address
> > > > > > >  - interrupt-parent: the phandle for the interrupt controller
> > > > > > > -- interrupts: interrupt line
> > > > > > > +- interrupts: interrupt line if the device uses IO-APIC interrupts
> > > > > > > +
> > > > > > > +Optional properties:
> > > > > > > +- gpios: GPIO used as an interrupt if the device uses GPIO interrupts
> > > > > > 
> > > > > > Elsewhere we've said that for a GPIO acting as an interrupt line, GPIO
> > > > > > controller should be marked as an interrupt-controller, and the GPIO
> > > > > > described as an interrupt line. That also gets you the appropriate
> > > > > > configuration for the GPIO as an interrupt.
> > > > > > 
> > > > > > Does this GPIO serve any other purpose than an ersatz interrupt line?
> > > > > 
> > > > > It is just an interrupt.
> > > > > 
> > > > > > If not, it should probably be described as an interrupt. From the PoV of
> > > > > > this device, it's just an interrupt controller hooked up to the
> > > > > > interrupt pin.
> > > > > 
> > > > > What I'm trying to do is to get a GPIO that is described in ACPI (as
> > > > > GpioInt() in _CRS) to be supported in this driver using gpiolib like
> > > > > this:
> > > > > 
> > > > > 	desc = gpiod_get(&client->dev, NULL);
> > > > > 
> > > > > This calls to find "gpios" property which ends up finding the GpioInt()
> > > > > in _CRS.
> > > > 
> > > > I understand what you are trying to do, but I disagree on the principle.
> > > > If it's logically an interrupt, it should be described as an interrupt.
> > > 
> > > It is a GPIO line that is used as interrupt. It is not IO-APIC interrupt
> > > or anything like that. It will be handled through a GPIO driver.
> > 
> > I understand that the interrupt line is wired up to a GPIO controller,
> > where the GPIO controller is able to raise an interrupt as required.
> > 
> > However, from the PoV of the I2C device, this doesn't matter. From it's
> > PoV it raises an interrupt, and that's all. It has no idea what happens
> > to be wired up to its IRQ pin, and nor should it.
> > 
> > > > If ACPI lacks the ability to describe things in that fashion, it's yet
> > > > another reason that we shouldn't be pretending that DT and ACPI are the
> > > > same thing.
> > > 
> > > I'm not saying they are the same thing (they're not). I'm trying to get
> > > a GPIO from ACPI translated to an interrupt so that the driver can use
> > > it. Preferably so that the DT description does not prevent people from
> > > using the same on non-ACPI platforms.
> > 
> > If you're following the Microsoft HID over I2C ACPI spec, why is this DT
> > binding document relevant. Assuming you're following the spec, you won't
> > be using _DSD with "hid-over-i2c". If you're not following the spec then
> > you aren't following the spec, so the spec is irrelevant.
> > 
> > > ACPI can desribe Interrupt(), GpioInt() and GpioIo() just fine. It is
> > > the Microsoft HID over I2C specification that says it should be
> > > GpioInt() even though we have seen Interrupt() used there as well.
> > 
> > Ok, so if the HID over I2C spec says that for ACPI, do that for ACPI.
> > 
> > I don't follow why the DT binding should do this.
> 
> I added it there because I thought it's the right thing to do. After all
> it is assumed that there exists property "gpios" now for both ACPI and
> DT if I pass NULL in gpiod_get().

That's not great. Now we have ACPI and DT concerns leaking into each
other when they shouldn't necessarily.

> Obviously I was wrong.

Well, not quite. We shouldn't accept non-documented properties in DT,
and documenting the property is certainly appreciated. The issue here
seems to be an inconsistency between how we handle GPIO-backed
interrupts described in DT vs ACPI.

As far as I can tell from the specs I've looked at, GpioInt describes an
interrupt that is backed by a GPIO. That sounds similar to what we would
do in DT in that we are describing the resource logically as an
interrupt.

What I don't follow is why GpioInt seems to be translated as a GPIO
rather than as an interrupt which happens to be backed by a GPIO. Were
it not for that, the DT and ACPI cases would align better.

> I'll remove the binding documentation from the next revision and call it
> "ACPI only" if it makes this work for you.

I would prefer not if that still leaves the property possible in the DT
case. We should figure out how we cater for these differences more
generally; I'm sure more are going to appear.

Mark.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists