ST's hardware differentiates between GPIO mode and Pinctrl alternate
functions.  When a pin is in GPIO mode, there are dedicated registers
to set and obtain direction status.  However, If a pin's alternate
function is in use then the direction is set and status is derived
from a bunch of syscon registers.  The issue is; until now there was
a lack of parity between the two.

For example:

Catting the two following information sources could result in
conflicting information (output has been snipped for simplicity):

 $ cat /sys/kernel/debug/gpio
  GPIOs 32-39, platform/961f080.pin-controller-sbc, PIO4:
   gpio-33  (?                   ) out hi

 $ cat /sys/kernel/debug/pinctrl/<pin-controller>/pinconf-pins
  pin 33 (PIO4[1]):[OE:0,PU:0,OD:0]
         [retime:0,invclk:0,clknotdat:0,de:0,rt-clk:0,rt-delay:0]

In this example GPIO-33 is a GPIO controlled LED, which is set for
output, as you'd expect.  However, when the same information is
drafted from Pinctrl, it clearly states that OE (Output Enable) is
not set i.e. the pin is set for input.  This is because OE normally
only represents alternate functions and has no bearing on how the
pin operates when in Alt-0 (GPIO mode).

This patch changes the current semantics and provides a parity link
between the two subsystems.  The get_direction() call-back firstly
determines which function a pin is operating in, then uses the
appropriate helpers for that mode.

Reported-by: Olivier Clergeaud <olivier.clergeaud@st.com>
Acked-by: Maxime Coquelin <maxime.coquelin@st.com>
Signed-off-by: Lee Jones <lee.jones@linaro.org>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>