Re: [PATCH 0/3] em28xx: clean up end extend the GPIO port handling

[Antti Palosaari <crope@iki.fi>]

On 04/13/2013 05:25 PM, Mauro Carvalho Chehab wrote:
> Em Sat, 13 Apr 2013 16:15:39 +0300
> Antti Palosaari <crope@iki.fi> escreveu:
>
>> On 04/13/2013 12:48 PM, Frank Schäfer wrote:
>>> Patch 1 removes the unneeded and broken gpio register caching code.
>>> Patch 2 adds the gpio register defintions for the em25xx/em276x/7x/8x
>>> and patch 3 finally adds a new helper function for gpio ports with separate
>>> registers for read and write access.
>>
>>
>> I have nothing to say directly about those patches - they looked good at
>> the quick check. But I wonder if you have any idea if it is possible to
>> use some existing Kernel GPIO functionality in order to provide standard
>> interface (interface like I2C). I did some work last summer in order to
>> use GPIOLIB and it is used between em28xx-dvb and cxd2820r for LNA
>> control. Anyhow, I was a little bit disappointed as GPIOLIB is disabled
>> by default and due to that there is macros to disable LNA when GPIOLIB
>> is not compiled.
>> I noticed recently there is some ongoing development for Kernel GPIO. I
>> haven't looked yet if it makes use of GPIO interface more common...
>
> I have conflicting opinions myself weather we should use gpiolib or not.
>
> I don't mind with the fact that GPIOLIB is disabled by default. If all
> media drivers start depending on it, distros will enable it to keep
> media support on it.
>
> I never took the time to take a look on what methods gpiolib provides.
> Maybe it will bring some benefits. I dunno.

Compare to benefits of I2C bus. It offers standard interface. Also it 
offers userspace debug interface - like I2C also does.

> Just looking at the existing drivers (almost all has some sort of GPIO
> config), GPIO is just a single register bitmask read/write. Most drivers
> need already bitmask read/write operations. So, in principle, I can't
> foresee any code simplification by using a library.

Use of lib interface is not very practical inside of module, however it 
could be used. Again, as compared to I2C there is some bridge drivers 
which do some I2C access using I2C interface, even bridge could do it 
directly (as it offers I2C adapter). I think it is most common to do it 
directly to simplify things.

> Also, from a very pragmatic view, changing (almost) all existing drivers
> to use gpiolib is a big effort.

It is not needed to implement for all driver as one go.

> However, for that to happen, one question should be answered: what
> benefits would be obtained by using gpiolib?

Obtain GPIO access between modules using standard interface and offer 
handy debug interface to switch GPIOs from userspace.

You could ask why we use Kernel I2C library as we could do it directly 
:) Or clock framework. Or SPI, is there SPI bus modeled yet?

regards
Antti

-- 
http://palosaari.fi/