[PATCH v4] gpio: mvebu: new gpio driver for mvebu platforms

Hello,

This patch set adds a new GPIO driver for Marvell EBU SoCs, which aims
at replacing the existing GPIO driver from plat-orion/gpio.c. The
driver comes with the necessary DT changes to make it work on Armada
370 and Armada XP, and it has been tested successfully by Sebastian
Hesselbarth on Dove. It should also work flawlessly on other Marvell
EBU SoCs (Kirkwood, Orion, Discovery), but this remains to be tested
by the relevant maintainers. This isn't however a problem, as those
older platforms need to explictly be converted over to the new pinctrl
driver and the new gpio driver. Until that, they will happily continue
to use the old, known-working gpio driver. So this patch set has no
chance to break any of the existing Marvell platforms.

This new driver reuses a lot of code from the existing
plat-orion/gpio.c driver, but has a number of advantages:

 *) Support for Armada 370 and Armada XP
 *) It is integrated with the mvebu pinctrl driver so that GPIO pins
    are properly muxed, and the GPIO driver knows which GPIO pins are
    output-only or input-only. This driver has been submitted by
    Sebastian Hesselbarth today on the LAKML.
 *) Properly placed in drivers/gpio
 *) More extensible mechanism to support platform differences. The
    plat-orion driver uses a simple mask-offset DT property, which
    works fine for Discovery MV78200 but not for Armada XP. The new
    driver uses different compatible strings to identify the different
    variants of the GPIO controllers.

This patch set should be applied on top of the pinctrl-mvebu patch set
from Sebastian Hesselbarth, as it depends on it.

Changes since v3:
 * Added #interrupt-cells in both the DT binding documentation
   example, and in the DT data for Armada 370 and XP (noticed by
   Stephen Warren)
 * Fixed a typo on the #interrupt-cells description in the DT binding
   description (noticed by Stephen Warren).
 * Fixed a bug in gpio_mvebu_irq_set_type() when the GPIO used was not
   from the first bank (noticed, debugged and fix provided by Andrew
   Lunn)
 * Added the Acked-by or Reviewed-by that were received from Rob
   Herring and Linus Walleij on certain patches.

Changes since v2:
 * Use inline functions for the register accessors, and rename them
   lower-case since they are functions, as suggested by
   Jean-Christophe Plagnol-Villard.
 * Use _relaxed() variants of readl/writel, as suggested by
   Jean-Christophe Plagnol-Villard.
 * Use switch() instead of long list of if() conditions in
   mvebu_gpio_irq_set_type(), as suggested by Jean-Christophe
   Plagnol-Villard.
 * Add DT aliases gpio0, gpio1, gpio2, and then use of_alias_get_id()
   to get the number of GPIO bank, and deduce the GPIO base from
   it. We don't use automatic allocation because we really want the
   internal SoC GPIOs to have 0-31, 32-63, 64-... numbers (while
   automatic allocation allocates 224... for the first bank 202... for
   the second bank, etc.). This strategy was recommended by Uwe
   Kleine-Koenig and is already used in gpio-mxs.c.
 * Use gpio-cells = <2> instead of gpio-cells = <1>, as suggested by
   Olof Johansson. This allows to re-use the default .xlate()
   function, and therefore make the GPIO DT binding actually work (it
   was broken in previous versions of this patch set).
 * Use dev_name() to fill the gpio_chip->name field, instead of
   manually computing a name. This allows to get rid of the global
   variable counting GPIO chips.
 * Rebase on top of the latest pinctrl work from Sebastian
   Hesselbarth.
 * Added the Acked-by from Linus Walleij on the DT binding
   documentation patch.

Changes since v1:
 * Added fixes for Dove from Sebastian Hesselbarth, and his Tested-by
   on the driver patch.

Thanks,

Thomas