More pinmux feedback, and GPIO vs. pinmux interaction

[Stephen Warren]

Linus (W),

Some more thoughts on pinmux:

========== GPIO/pinmux API interactions

I'd like to understand how the gpio and pinmux subsystems are intended
to interact.

Quoting from Documentation/gpio.txt:

  Note that requesting a GPIO does NOT cause it to be configured in any
  way; it just marks that GPIO as in use.  Separate code must handle any
  pin setup (e.g. controlling which pin the GPIO uses, pullup/pulldown).

Due to this, the current Tegra GPIO driver contains additional non-
standard functions:

void tegra_gpio_enable(int gpio);
void tegra_gpio_disable(int gpio);

In some cases, those functions are called by board files at boot time,
and in other cases by drivers themselves, right before gpio_request().

Is it your intention that such custom functions be replaced by
pinmux_request_gpio(), and the Tegra pinmux driver's gpio_request_enable
op should call tegra_gpio_enable?

That seems like the right idea to me.

========== pinmux calling GPIO or vice-versa?

Having pinmux call gpio appears to conflict with a couple of things from
your recent pinmux patchset:

> diff --git a/drivers/Kconfig b/drivers/Kconfig
> +# pinctrl before gpio - gpio drivers may need it
> +
> +source "drivers/pinctrl/Kconfig"
> +
>  source "drivers/gpio/Kconfig"
>
> diff --git a/drivers/Makefile b/drivers/Makefile
>...
> +# GPIO must come after pinctrl as gpios may need to mux pins etc
> +obj-y				+= pinctrl/

Don't those patches imply that the GPIO controller code is calling into
the pinmux code to perform muxing, not the other way around?

========== When pins get marked as in-used

There seems to be a discrepancy in the way the client APIs work: For
special functions, the client calls pinmux_get, then later pinmux_enable,
whereas for GPIOs, the client just calls pinmux_request_gpio. I'm not
sure of the benefit of having separate pinmux_get and pinmux_enable
calls; I thought the idea was that a client could:

a = pinmux_get(dev, "funca");
b = pinmux_get(dev, "funcb");
pinmux_enable(a);
...
pinmux_disable(a);
pinmux_enable(b);
...
pinmux_disable(b);
pinmux_put(b);
pinmux_put(a);

However, since the pins are marked as reserved/in-use when pinmux_get
is called rather than pinmux_enable, the code above won't work; it'd
need to be:

a = pinmux_get(dev, "funca");
pinmux_enable(a);
...
pinmux_disable(a);
pinmux_put(a);
b = pinmux_get(dev, "funcb");
pinmux_enable(b);
...
pinmux_disable(b);
pinmux_put(b);

Perhaps pin usage marking should be moved into enable/disable?

========== Matched request/free calls for GPIOs

A couple more comments on your recent pinmux patches in light of this: