Re: [RFC PATH 2/2] gpio: starfive-jh7100: Add StarFive JH7100 GPIO driver

[Emil Renner Berthing <kernel@esmil.dk>]

On Wed, 28 Jul 2021 at 13:19, Michael Walle <michael@walle.cc> wrote:
> Am 2021-07-28 12:59, schrieb Emil Renner Berthing:
> > On Wed, 28 Jul 2021 at 11:49, Michael Walle <michael@walle.cc> wrote:
> >> Hi Drew,
> >> Am 2021-07-27 07:28, schrieb Drew Fustini:
> >> [..]
> >> >> > > Drew please look at drivers/gpio/gpio-ftgpio010.c for an example
> >> >> > > of GPIO_GENERIC calling bgpio_init() in probe().
> >> >> >
> >> >> > Thank you for the suggestion. However, I am not sure that will work for
> >> >> > this SoC.
> >> >> >
> >> >> > The GPIO registers are described in section 12 of JH7100 datasheet [1]
> >> >> > and I don't think they fit the expectation of gpio-mmio.c because there
> >> >> > is a seperate register for each GPIO line for output data value and
> >> >> > output enable.
> >> >> >
> >> >> > There are 64 output data config registers which are 4 bytes wide. There
> >> >> > are 64 output enable config registers which are 4 bytes wide too. Output
> >> >> > data and output enable registers for a given GPIO pad are contiguous.
> >> >> > GPIO0_DOUT_CFG is 0x50 and GPIO0_DOEN_CFG is 0x54 while GPIO1_DOUT_CFG
> >> >> > is 0x58 and GPIO1_DOEN_CFG is 0x5C. The stride between GPIO pads is
> >> >> > effectively 8, which yields the formula: GPIOn_DOUT_CFG is 0x50+8n.
> >> >> > Similarly, GPIO0_DOEN_CFG is 0x54 and thus GPIOn_DOEN_CFG is 0x54+8n.
> >> >> >
> >> >> > However, GPIO input data does use just one bit for each line. GPIODIN_0
> >> >> > at 0x48 covers GPIO[31:0] and GPIODIN_1 at 0x4c covers GPIO[63:32].
> >>
> >> Mh, I'm not sure I'm understanding the datasheet/registers. _DOUT_CFG
> >> and _DOEN_CFG seem to specify the pad where this GPIO is mapped to.
> >> Shouldn't this be some kind of pinctrl then? Apparently you can map
> >> any GPIO number to any output pad, no? Or at least to all pads
> >> which are described in Table 11-2. What happens if two different GPIOs
> >> are mapped to the same pad? Bit 31 in these _CFG seems to be an invert
> >> bit, but what does it invert?
> >>
> >> Similar, the input GPIOs are connected to an output pad by all the
> >> GPI_*_CFG registers.
> >>
> >> To me it seems, that there two multiplexers for each GPIO, where
> >> you can connect any GPIOn to any input pad and output pad. Sound
> >> like a huge overkill. I must be missing something here.
> >>
> >> But what puzzles me the most, where do I set the actual GPIO output
> >> value?
> >
> > Yeah, it's a little confusing. The DOUT registers choose between a
> > number of
> > signals from various peripherals to control the output value of the
> > pin. Similarly
> > the DOEN registers chose between a number of signals to control the
> > output
> > enable of the pin. However, two of those signals are special in that
> > they are
> > constant 0 or constant 1. This is how you control the output value and
> > output
> > enable from software like a regular GPIO.
> >
> > You're completely right though. This ought to be managed by a proper
> > pinctrl
> > driver, and I'm working on one here:
> > https://github.com/esmil/linux/commits/beaglev-pinctrl
>
> Ahh, I see. So for the non-gpio function you have to set a value other
> than 0 or 1, correct?
>
> And as an implementation detail you have to set the corresponding OE
> pin if the non-gpio function will need a tristate pin (or whatever).
>
> So, the _DOUT_CFG will actually be shared between the pinctrl and the
> gpio driver, right? (I haven't done anything with pinctrl, so this might
> be a stupid question).

No, not a stupid question. You've got that exactly right.

/Emil