Commit df61b366af26 ("pinctrl: core: Use delayed work for hogs") caused a
regression at least with sh-pfc that is also a GPIO controller as
noted by Geert Uytterhoeven <geert@linux-m68k.org>.

As the original pinctrl_register() has issues calling pin controller
driver functions early before the controller has finished registering,
we can't just revert commit df61b366af26. That would break the drivers
using GENERIC_PINCTRL_GROUPS or GENERIC_PINMUX_FUNCTIONS.

So let's fix the issue with the following steps as a single patch:

1. Revert the late_init parts of commit df61b366af26.

   The late_init clearly won't work and we have to just give up
   on fixing pinctrl_register() for GENERIC_PINCTRL_GROUPS and
   GENERIC_PINMUX_FUNCTIONS.

2. Split pinctrl_register() into two parts

   By splitting pinctrl_register() into pinctrl_init_controller()
   and pinctrl_create_and_start() we have better control over when
   it's safe to call pinctrl_create().

3. Introduce a new pinctrl_register_and_init() function

   As suggested by Linus Walleij <linus.walleij@linaro.org>, we
   can just introduce a new function for the controllers that need
   pinctrl_create() called later.

4. Convert the four known problem cases to use new function

   Let's convert pinctrl-imx, pinctrl-single, sh-pfc and ti-iodelay
   to use the new function to fix the issues. The rest of the drivers
   can be converted later. Let's also update Documentation/pinctrl.txt
   accordingly because of the known issues with pinctrl_register().

Fixes: df61b366af26 ("pinctrl: core: Use delayed work for hogs")
Reported-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Gary Bisson <gary.bisson@boundarydevices.com>
Signed-off-by: Tony Lindgren <tony@atomide.com>
Tested-by: Geert Uytterhoeven <geert+renesas@glider.be>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>