[PATCH v2 0/7] ARM: berlin: add pinctrl support

Sebastian Hesselbarth sebastian.hesselbarth at gmail.com
Thu Apr 24 06:23:29 PDT 2014


On 04/24/2014 02:52 PM, Linus Walleij wrote:
> On Wed, Apr 23, 2014 at 5:51 PM, Antoine Ténart
> <antoine.tenart at free-electrons.com> wrote:
>
>> This series adds support for the Marvell Berlin pin-controller, allowing
>> to configure the pin muxing from the device tree.
>>
>> The Berlin pin-controller support is divided into 3 drivers, each
>> driving one Berlin SoC. These drivers use a Berlin common part.
>>
>> This series applies on top of patches introducing the Marvell Berlin
>> BG2Q you can find on Sebastian's berlin/for-next branch[1] and the patch
>> allowing not to define the get_group_pins() function[2].
>>
>> Tested on the Berlin BG2Q.
>
> So now I need some advice from the mvebu pinctrl maintainers
> (Thomas, Sebastian etc):
>
> - Is this a totally different pin controller so that drivers/pinctrl/mvebu
>   can not be used?

Unfortunately, yes. Well actually, it _can_ be seen as a subset of
the mvebu pinctrl:

- mvebu (usually) uses 4bit per mux function with 8 functions/register
- berlin uses 1-4bit per mux function with as many functions/register
   as there fit in 32b.

This great feature saves _at least_ one additional address decoding!
It doesn't save registers, because they will never be synthesized, but
at least each SoC has a very different pinmux layout. *sigh*

> - Really?

Yep.

> - OK can you help me review this thing?

Sure thing!

> - Should the base folder really be named "berlin" or is this going to
>    be part of a bigger family of pin controllers so a more neutral name
>    should be sought?

Well, Marvell basically has two groups of SoCs, MVEBU and PXA/MMP. Don't
ask me why but sometimes they share IP, sometimes they don't.

Berlin names Marvell SoCs prefixed 88DExxxx, I *think* it may be derived 
from the PXA/MMP line of SoCs. To make it more confuse,
it also got the marketing name "Armada".

> - Why do hardware engineers seek to reinvent wheels like pin
>    controllers, GPIO and DMA engines all the time :-/

I guess it is: "Look what we found in our IP archives".

Honestly, I can think of making pinctrl/mvebu and pinctrl/berlin
compatible but I don't know if it is worth the pain :P

We have "custom" set/get_function() callbacks in pinctrl/mvebu
so that should fit. Each "group" of pins has a name and a bunch
of "functions", that fits too.

I need some time to think about it, but if you insist on it, I
can possibly make it work.

Sebastian




More information about the linux-arm-kernel mailing list