GPIO hint meanings

[Ian Lepore]

On Fri, 2013-09-06 at 16:13 -0600, Warner Losh wrote:
> On Sep 6, 2013, at 4:00 PM, Ian Lepore wrote:
> 
> > On Fri, 2013-09-06 at 13:42 -0600, Warner Losh wrote:
> >> On Sep 6, 2013, at 12:31 PM, Luiz Otavio O Souza wrote:
> >> 
> >>> On 6 September 2013 14:22, Sean Bruno <sean_bruno at yahoo.com> wrote:
> >>> 
> >>>> I think I have a fairly firm grasp on what some of the mips/gpio hints
> >>>> mean, e.g.:
> >>>> 
> >>>> hint.gpio.0.pinmask
> >>>> hint.gpioled.0.at
> >>>> hint.gpioled.0.name
> >>>> hint.gpioled.0.pins
> >>>> 
> >>>> Fairly straightforward.
> >>>> 
> >>>> Now, what do these mean/do:
> >>>> 
> >>>> hint.gpio.0.function_set
> >>>> hint.gpio.0.function_clear
> >>>> 
> >>>> ?
> >>>> 
> >>>> Sean
> >>>> 
> >>>> p.s. I think I'll take this and thrash together a gpioled(4) and gpio(4)
> >>>> man page if I can understand better.
> >>>> 
> >>> 
> >>> 
> >>> Hi Sean,
> >>> 
> >>> Some of the GPIO pins on this SoC family (ar724x, ar71xx and ar9xxx) can be
> >>> set between GPIO and an alternate function. So adding a pin to function_set
> >>> enables this alternate function and the function_clear disables it
> >>> (sometimes the bootloader doesn't initialize properly those pins).
> >>> 
> >>> Each SoC has its own set of pins and functions.
> >>> 
> >>> For ar71xx the pins 0 and 1 can be used as additional SPI chip select
> >>> outputs, pins 9 and 10 are used for UART and there are also reserved pins
> >>> for a SLIC/I2S interface.
> >> 
> >> 
> >> We really need a pinmux/pinctl type interface for this which is standard across drivers/platforms.
> >> 
> > 
> > The more ARM SoCs I look at, the less I think we could design a single
> > pinmux api that works for all of them.  The number of things that can be
> > controlled varies from almost-nothing to chips that let you select from
> > one of a dozen different resistor strengths for pullup or pulldown per
> > pin.  And that's not to mention really crazy things like daisy-chaining
> > pins so the signal also goes to another pin which can be forced as an
> > input even though it's normally a device output.
> 
> Linux is able to have one, although I'm not sure how they handle the daisy-chain... that's a new one on me...
> 

Maybe they just don't, since it's a weird enough thing that probably
nothing uses it.  I only discovered it because the datasheet said it was
a potential workaround for an erratum that had to do with a device not
handling a pin properly.

The semi-related thing I've been pondering lately is clock and power
management.  I don't even care about dynamic stuff, just a simple common
way for a driver to figure out what clock(s) and/or power need to be on
for it to run, and a common api for turning them on would be nice.
(Whether clocks and power should be two separate APIs or not is a basic
question.)

-- Ian