Message-ID: <53FDE9B2.4040001@collabora.co.uk>
Date:	Wed, 27 Aug 2014 16:22:42 +0200
From:	Javier Martinez Canillas <javier.martinez@...labora.co.uk>
To:	Andreas Färber <afaerber@...e.de>
CC:	Kukjin Kim <kgene.kim@...sung.com>,
	Doug Anderson <dianders@...omium.org>,
	Olof Johansson <olof@...om.net>,
	Nick Dyer <nick.dyer@...ev.co.uk>,
	Yufeng Shen <miletus@...omium.org>,
	linux-samsung-soc@...r.kernel.org,
	linux-arm-kernel@...ts.infradead.org, linux-kernel@...r.kernel.org,
	Sjoerd Simons <sjoerd.simons@...labora.co.uk>,
	Tomasz Figa <tomasz.figa@...il.com>,
	Daniel Stone <daniels@...labora.com>
Subject: Re: [PATCH v2 1/3] ARM: dts: Add Peach Pit dts entry for Atmel touchpad

Hello Andreas,

On 08/27/2014 03:11 PM, Andreas Färber wrote:
> Hi Javier,
>>>
>>> +	trackpad@4b {
>>> +		compatible = "atmel,maxtouch";
>>> +		reg = <0x4b>;
>>> +		interrupt-parent = <&gpx1>;
>>> +		interrupts = <2 IRQ_TYPE_NONE>;
>>> +		pinctrl-names = "default";
>>> +		pinctrl-0 = <&trackpad_irq>;
>>> +		linux,gpio-keymap = <KEY_RESERVED
>>> +		                     KEY_RESERVED
>>> +		                     KEY_RESERVED
>>> +		                     KEY_RESERVED
>>> +		                     KEY_RESERVED
>>> +		                     BTN_LEFT>;
>>> +		wakeup-source;
>>> +	};
>>>
>>> 0 == KEY_RESERVED, so you can consistently use it for GPIO 0-2, too. :)
>>>
>> 
>> I know that the value of KEY_RESERVED is 0 but I didn't use KEY_RESERVED
>> for the GPIO on purpose.
>> 
>> What I understood is that the SPT_GPIOPWN_T19 object sends messages using
>> a status byte so you have a maximum of 8 GPIO but not every maXTouch
>> devices use all of them. So in the particular case of the device in the
>> Peach Pit, from the 8 possible GPIO only 4 can be used and these are pins
>> 2-5. So in theory you could connect 3 more GPIO in case you had more
>> buttons (e.g: BTN_RIGHT, BTN_MIDDLE) but only 1 is used since the
>> Chromebook just have BTN_LEFT.
> 
> FWIW when I press to the bottom right of my touchpad, I do get
> right-click functionality even with just BTN_LEFT specified in the
> keymap. Magic. :)
>

Right, I'm not an expert in input but after asking a colleague he
explained to me that user-space input drivers deal with the MT tracking
for you, but there is still a single event code reported (BTN_LEFT) as an
indication that the touchpad has been physically clicked. That's why you
are having right click even when the hardware is not reporting it.

You can confirm that by running plain evdev or by clicking on the touchpad
using an object (e.g: a pencil) so ABS_MT_* events won't be reported.

>> Nick sent a patch [0] that extend the atmel touchpad DT binding and the
>> doc says "Use KEY_RESERVED for unused padding values". But is not clear
>> what value you should use for GPIO that are actually supported by the
>> device but have no keycode associated.
>> 
>> So by using 0 instead of KEY_RESERVED I wanted to document that pins 2-4
>> are actually supported and not reserved by the device but there is no
>> keycode associated with that GPIO.
> 
> You already documented that via comments though.
>

Yes but still I don't like to use KEY_RESERVED for all the pins since some
of them are not really reserved. Can we agree on disagree here and after
Nick answer I can post as a follow-up patch after this series gets merged?

>> If there was a BTN_NONE or KEY_UNUSED it would had been better but I think
>> that making a distinction between these two cases (reserved pin vs GPIO
>> available but not used) is useful.
> 
> Maybe Nick can comment here.
> 
>>> I probably should add the two trailing _RESERVEDs, too?
>>>
>> 
>> I see that is used for properties that are arrays, for example
>> "linux,keymap" in Documentation/devicetree/bindings/input/matrix-keymap.txt.
> 
> That does not answer my question: Do all maxTouch touchpads (or
> specifically that in Spring) need eight entries, padded with said
> KEY_RESERVED? In my experiments using just six entries (i.e., until the
> non-zero entry) worked okay - so does this T19 message have specifically
> eight bits? Tegra used just four entries iirc.
> 

I guess there are other touchpads that have a different offset (e.g: using
the last 4 pins) so I think that is good as a way of documenting the
layout even if not strictly necessary for the driver.

But again, without proper documentation is hard to say so I'll let Nick to
answer that.

>>> With my above snippet I got an awful lot of "Interrupt triggered but
>>> zero messages" warnings (which I simply commented out as quickfix).
>>> Is that why you are using _EDGE_FALLING? Or pin-function 0xf?
>>> (In my case the ChromeOS DT had IRQ_TYPE_NONE and pin-function 0x0.)
>>>
>> 
>> These are two separate but related things:
>> 
>> a) IRQF_TRIGGER_FALLING:
>> 
>> Yes, the Chrome OS DT for Peach Pit also has IRQ_TYPE_NONE but the DTS is
>> not correct.
>> 
>> If you look at the Chrome OS atmel driver
>> (drivers/input/touchscreen/atmel_mxt_ts.c), it passes IRQF_TRIGGER_FALLING
>> to request_threaded_irq():
>> 
>> /* Default to falling edge if no platform data provided */
>> irqflags = data->pdata ? data->pdata->irqflags : IRQF_TRIGGER_FALLING;
>> error = request_threaded_irq(client->irq, NULL, mxt_interrupt,
>> 			     irqflags | IRQF_ONESHOT,
>> 			     client->name, data);
>> 
>> The above code is wrong since is overwriting the edge/level type flags set
>> by OF when parsing the "interrupts" property so you have to use the
>> expected IRQ flags in your DTS.
>> 
>> b) pin-function 0xf instead of 0x0:
>> 
>> The pin-function 0x0 is GPIO input while 0xf is GPIO IRQ. Usually on other
>> SoCs to use a GPIO IRQ you just configure the pad as GPIO input and then
>> enable the pin as an interrupt but on Exynos SoC these are really two
>> different functions. So if you configure the pin as GPIO input and this
>> happens after the pin is configured as GPIO IRQ, interrupts are not triggered.
>> 
>> I faced that issue before [1] and was solved with Tomasz's commit:
>> 
>> f6a8249 pinctrl: exynos: Lock GPIOs as interrupts when used as EINTs
>> 
>> which changes the pinctrl-exynos driver to setup a pin as GPIO IRQ on
>> .irq_request_resources instead of .irq_set_type. So, with that patch even
>> when pin-function re-configures the function to GPIO input, is then
>> configured as GPIO IRQ when request_threaded_irq() is called.
>> 
>> So probably is working for you just because you tested on linux-next that
>> already has Tomasz's changes but still the correct thing to do is to setup
>> the pin as 0xf. This change probably is needed on other pins used as GPIO
>> IRQ that are using 0x0 now.
>> 
>> Sorry, the email became longer than I wanted but I hope is helpful to you.
> 
> Thanks for the explanations, I'll test those settings on Spring then.
> 
> Could you point me to what ChromeOS tree and branch I should be looking
> at? For instance, the linux-next.git chromeos-3.8 branch did not have
> any DT for Spring. Therefore my series is based on /proc/device-tree
> rather than any ChromeOS source code.
> 

I use as a reference
https://chromium.googlesource.com/chromiumos/third_party/kernel branch
chromeos-3.8

> Thanks,
> Andreas
> 

Best regards,
Javier
--
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