Re: [PATCH v1 2/2] gpio: ds4520: Add ADI DS4520 Regulator Support

[Michael Walle <michael@walle.cc>]

Am 2023-04-24 17:39, schrieb Sahin, Okan:
>> Am 2023-04-09 16:25, schrieb Sahin, Okan:
>>>> Fri, Apr 07, 2023 at 03:48:25PM +0200, Linus Walleij kirjoitti:
>>>>> On Wed, Apr 5, 2023 at 3:57 PM Michael Walle <michael@walle.cc>
>>>>> wrote:
>>>>> 
>>>>> > OTOH I'm not sure the driver is doing it correctly, because it also
>>>>> > seems to switch the pullup resisters together with the direction.
>>>>> > I'm not sure that is correct. So there might be just one register
>>>>> > involved after all and the GPIO_REGMAP should work again.
>>>>> 
>>>>> I'm pretty sure that should be in the .set_config() callback.
>>>>> 
>>>>> > Also, according to the datasheet this has some nv memory (to set the
>>>>> > initial state of the GPIOs [?]). So it should really be a
>>>>> > multi-function device. I'm not sure if this has to be considered
>>>>> > right from the beginning or if the device support can start with
>>>>> > GPIO only and later be transitioned to a full featured MFD (probably with
>> nvmem
>>>> support).
>>>>> 
>>>>> That's a bit of a soft definition.
>>>>> 
>>>>> If the chip is *only* doing GPIO and nvram it can be a GPIO-only
>>>>> device I think.
>>>>> 
>>>>> The precedent is a ton of ethernet drivers with nvram for storing
>>>>> e.g.
>>>>> the MAC address. We don't make all of those into MFDs, as the nvram
>>>>> is
>>>>> closely tied to the one and only function of the block.
>>>> 
>>>> I agree with Linus. This should be part of the actual (main) driver
>>>> for the chip as many
>>>> do (like USB to serial adapters that have GPIO capability).
>> 
>> You mean the gpio driver is calling nvmem_register()? Yeah I agree, 
>> that
>> should work.
>> 
>>> I think gpio_regmap is not suitable for this driver as Michael 
>>> stated.
>>> https://www.analog.com/media/en/technical-documentation/data-
>> sheets/ds4520.pdf
>>> Please check block diagram. There are two input registers that 
>>> control
>>> gpio state
>>> so gpio_regmap does not look ok for this. Am I missing something?
>> 
>> You mean F8/F9? That will work as they are for different GPIOs. What
>> doesn't work with gpio-regmap is when you need to modify two different
>> registers for one GPIO. Have a look at gpio_regmap_get() and
>> gpio_regmap_set(). If the default gpio_regmap_simple_xlate() doesn't
>> work
>> you can use your own .xlate() op.
>> 
> 
> Actually, I checked the functions that you suggested, but as far as I 
> understand
> they might work if there would be one bit to set direction or value. 
> However,
> this is not the case for ds4520. In other words, if I want to set the
> gpio direction
> as output, I need to set a corresponding bit for both F0 and F1 
> registers.

I can't follow. F0/F1 is for the pull-up. That was actually my initial
question and Linus said, that should probably be done in a seperate
.set_config operation not together with a direction change.

> In the document, you can see block diagram. I do not know why, but 
> design is
> not standard that’s why I think I can not use gpio-regmap.
> 
>>> Also, at this point I am not planning to add nvmem support.
>> 
>> That is a pity, because that is the whole use case for this gpio
>> expander,
>> no? "Programmable Replacement for Mechanical Jumpers and Switches"
> 
> I can set "SEE" bit as "0" in the Configuration Register to write
> EEPROM so it might solve
> issue.

If you do that unconditionally, that might wear out the EEPROM,
though.

-michael