On Aug 18, Andrew Lunn wrote:

On Sun, Aug 18, 2024 at 02:48:05PM +0200, Benjamin Larsson wrote:

quoted

On 17/08/2024 23:39, Andrew Lunn wrote:

quoted

How messy are the GPIO and PWM registers? Are there N blocks of
independent GPIO registers? and M blocks of independent PWM registers?
By that, does one block of GPIO registers contain all you need for one
GPIO controller? One block of PWM registers give you all you need for
one PWM controller? Or are the registers for one GPIO controller
scattered all over the place?

Could you point at a public datasheet?

       Andrew

Hi, per my understanding there is no public datasheet/register reference
manual.

But here is the division of regions of the registers in the gpio block and
how it is currently divided between the drivers (according to my current
understanding).

1FBF0200, gpio/pinctrl
1FBF0204, gpio/pinctrl
1FBF0208, gpio/pinctrl
1FBF020C, gpio/pinctrl
1FBF0210, gpio/pinctrl
1FBF0214, gpio/pinctrl

A typical SoC has multiple instances of a GPIO controller. Each GPIO
controller typically has 4 or 5 registers: In, Out, Direction,
Interrupt Enable, Interrupt Status. If these 4 or 5 registers are
contiguous, you could have one DT node per controller, rather than one
node for all GPIO controllers.

it is the same for en7581 pinctrl too. I think we can squash most of the
gpio/irq registers into "bigger" io-regions (just keeping a couple of holes
for pwm and leds). It is just a matter of moving the logic from the dts to
the driver. I am currently working on it. I will post v2 soon.

If the hardware designer has really messed up and fully interleaved
GPIO and PWM, it might be better to have an MFD. The MFD node has a
single reg covering the entire range. The MFD would then map the whole
range, and provide accessors to the child devices. Hard code the
knowledge of what registers are where. Given how badly the hardware is
designed, it is unlikely it will get reused in the future, so there is
no point putting lots of stuff into DT. Hard code it.

I am not sure it is possible/feasible to implement a MFD device here since
the mapped region is huge and sparse.

Regards,
Lorenzo

	Andrew