Some pinctrl drivers (Tegra at least) program a pin to be a GPIO in a
completely different manner than they select which function to mux out of
that pin. In order to support a single "free" pinmux_op, the driver would
need to maintain a per-pin state of requested-for-gpio vs. requested-for-
function. However, that's a lot of work when the core already has explicit
separate paths for gpio request/free and function request/free.

So, add a gpio_disable_free op to struct pinmux_ops, and make pin_free()
call it when appropriate.

When doing this, I noticed that when calling pin_request():

    !!gpio == (gpio_range != NULL)

... and so I collapsed those two parameters in both pin_request(), and
when adding writing the new code in pin_free().

Also, for pin_free():

    !!free_func == (gpio_range != NULL)

However, I didn't want pin_free() to know about the GPIO function naming
special case, so instead, I reworked pin_free() to always return the pin's
previously requested function, and now pinmux_free_gpio() calls
kfree(function). This is much more balanced with the allocation having
been performed in pinmux_request_gpio().
Signed-off-by: Stephen Warren <swarren@nvidia.com>
Signed-off-by: Linus Walleij <linus.walleij@linaro.org>