[RFC PATCH 00/12] arm/tegra: Initialize GPIO & pinmux from DT

[Stephen Warren <swarren@nvidia.com>]

This patch modifies Tegra's device tree support to remove the dependency
on harmony_pinmux_init(), thus making it completely board-independent.

Some notes:

* This series is built on top of linux-next with a bunch of patches
  applied, in particular the removal of irq_to_gpio and custom gpio_to_irq
  that I'm in the process of sending to Russell. I haven't yet thought
  through how/where to merge this without causing all kinds of conflicts.

* I took care to preserve bisectability of Tegra DT support. However,
  linux-next doesn't yet have entirely useful Tegra DT support; some stuff
  from Grant's devicetree/next hasn't been pushed into linux-next yet. If
  we don't care about bisectability, I can remove a couple commits and
  possibly squash some others.

* The approach taken here is to have a custom semantic SoC-specific
  binding for each the gpio and pinmux drivers. Other alternatives
  suggested included:

  1) A generic "list of register writes" to be performed at boot. This has
     the advantage of reusability across different SoCs. However, this
     approach isn't semantic, and requires detailed knowledge of pinmux
     registers and potentially fiddly calculations when constructing the
     device tree.

  2) The ability to define disabled child nodes of the pinmux controller
     that are not processed by tegra_pinmux_probe_dt(). Other devices may
     refer to those using phandles, and later enable/disable them, thus
     representing dynamic pinmuxing in the device tree. I wasn't convinced
     whether we should represent dynamic pinmuxing using phandles.

  I discussed in more detail why I prefer the current proposal in various
  email threads.

* tegra_pinmux_probe_dt() enumerates all legal pingroup names, and searches
  for a pinmux controller subnode of that name, then processes each one
  that is found. An alternative that some may prefer would be to enumerate
  each child node of the pinmux controller, and have each node contain an
  explicit pingroup name property instead. Does anyone have any preference
  here? I suppose the latter option would obviate the need to add
  of_find_child_node_by_name().

Thanks for reading!

Stephen Warren (12):
  dt: Add of_find_child_node_by_name()
  arm/tegra: Prep boards for gpio/pinmux conversion to pdevs
  arm/tegra: Avoid duplicate gpio/pinmux devices with dt
  arm/tegra: board-dt: Add AUXDATA for tegra-gpio and tegra-pinmux
  arm/dt: Tegra: Add nvidia,gpios property to GPIO controller
  arm/dt: Tegra: Add pinmux node
  gpio/tegra: Convert to a platform device
  gpio/tegra: Add device tree support
  arm/tegra: Convert pinmux driver to a platform device
  arm/tegra: Add device tree support to pinmux driver
  arm/tegra: board-dt: Remove dependency on non-dt pinmux functions
  arm/tegra: Remove temporary gpio/pinmux registration workaround

 arch/arm/boot/dts/tegra-harmony.dts          |  479 ++++++++++++++++++++++++++
 arch/arm/boot/dts/tegra-seaboard.dts         |  409 ++++++++++++++++++++++
 arch/arm/boot/dts/tegra20.dtsi               |    5 +
 arch/arm/mach-tegra/Makefile                 |    1 -
 arch/arm/mach-tegra/board-dt.c               |   12 +-
 arch/arm/mach-tegra/board-harmony-pinmux.c   |    8 +
 arch/arm/mach-tegra/board-paz00-pinmux.c     |    8 +
 arch/arm/mach-tegra/board-seaboard-pinmux.c  |    9 +-
 arch/arm/mach-tegra/board-trimslice-pinmux.c |    7 +
 arch/arm/mach-tegra/devices.c                |   10 +
 arch/arm/mach-tegra/devices.h                |    2 +
 arch/arm/mach-tegra/pinmux.c                 |  136 ++++++++
 drivers/gpio/gpio-tegra.c                    |   56 +++-
 drivers/of/base.c                            |   18 +
 include/linux/of.h                           |    2 +
 15 files changed, 1138 insertions(+), 24 deletions(-)