[PATCH v1 1/1] gpio: of: provide option generic property for initial line states

Vladimir Zapolskiy vladimir_zapolskiy at mentor.com
Thu Jun 15 02:48:28 PDT 2017


Hi Oleksij,

On 06/15/2017 08:56 AM, Oleksij Rempel wrote:
> Hi Vladimir,
> 
> On Sun, Jun 11, 2017 at 10:57:18PM +0300, Vladimir Zapolskiy wrote:
>> Hi Oleksij,
>>
>> On 06/11/2017 09:19 AM, Oleksij Rempel wrote:
>>> Hi Rob,
>>>
>>> On Fri, Jun 09, 2017 at 10:41:30AM -0500, Rob Herring wrote:
>>>> On Fri, Jun 9, 2017 at 9:16 AM, Oleksij Rempel <o.rempel at pengutronix.de> wrote:
>>>>> This property is needed for barebox or may be other bootloaders
>>>>> to set proper gpio configuration as early as possible.
>>>>>
>>>>> This example is mainly based on "lines-initial-states" property
>>>>> from this binding documentation:
>>>>> Documentation/devicetree/bindings/gpio/gpio-pcf857x.txt
>>>>
>>>> Don't we have a binding for this already based on gpio-hogs.
>>>
>>> Markus Pargmann was working on this topic before and suddenly was not
>>> able to finish it.
>>
>> that's fine, but the proposed extension is supposedly not wanted.
> 
> :(

No worries, we're in the discussion and 'supposedly' was a meaningful
word :)

>>> I found only some not finished discussions and no code or documentation
>>> in linux kernel upstream.
>>>
>>> So far, I was able to find fallowing patches and discussions:
>>> patch 23.08.2015, last comment 11.05.2017:
>>> https://patchwork.kernel.org/patch/7096721/
>>>
>>> patch 23.02.2016, last comment 09.03.2016:
>>> https://patchwork.ozlabs.org/patch/586743/
>>>
>>> Since there was objections about node-like design for gpio-initval, I
>>> provided an array property, which is similar to initval properties already used
>>> for different other device nodes. For example:
>>>
>>> Documentation/devicetree/bindings/net/wireless/qcom,ath10k.txt
>>> qcom,ath10k-calibration-data
>>>
>>> Documentation/devicetree/bindings/gpio/gpio-pcf857x.txt
>>> lines-initial-states
>>
>> It'd rather be better to fix the drivers and remove the properties
>> from these found device tree binding descriptions.
> 
> no sure if it is possible. WiFi calibration data is kind of initval too.

Sorry, I have no idea about the ath10k calibration data and its
connection to the GPIO framework.

>>>
>>> Please fixme if I miss something.
>>
>> Why is GPIO hogging mechanism not good enough for your purpose?
> 
> Becouse it should not permanently hog a gpio. The "Hogging" should be
> released as soon as some driver will request it.
> 
> Right now, if I define gpio-hog, boot loader will configure it and start
> linux. But linux will never let it free.
> 
> If you have other suggestions which fit to this use case, please tell
> me. I assume, right now I'm just blind.

Ok, I'll meditate on it.

> Other way, I can imagine is to create real device nodes which use this
> gpios. Compatible for this devices will be like this:
> some_gpio_holder: gpio at 0 {
> 	compatible = "gpio-for-userspace";
> 	gpio = <&gpio1 5 GPIO_ACTIVE_HIGH>;
> };

Generally it might be a good option, because if GPIOs are essentially
needed to be preset, I can conclude there are some ICs to which these
GPIOs are connected, thus drivers of these ICs can become GPIO consumers
and get a proper description in a board DTB.

> 
> Boot loader will configure it as needed, an linux will do nothing with
> it so it can be used over /dev/gpio interface.

As I read it from Documentation/gpio/sysfs.txt generic interface(s)
to control GPIOs is a last resort, it is not clear if you want a new
feature for all possible consumers or just userspace consumers.

--
With best wishes,
Vladimir



More information about the barebox mailing list