Re: [PATCH v4 02/15] gpio: regmap: set gpio_chip of_node

[Andy Shevchenko <andy.shevchenko@xxxxxxxxx>]

On Fri, Mar 5, 2021 at 9:45 AM Álvaro Fernández Rojas <noltari@xxxxxxxxx> wrote:
> > El 4 mar 2021, a las 17:33, Andy Shevchenko <andy.shevchenko@xxxxxxxxx> escribió:
> > On Thu, Mar 4, 2021 at 5:44 PM Álvaro Fernández Rojas <noltari@xxxxxxxxx> wrote:
> >>> El 4 mar 2021, a las 16:28, Andy Shevchenko <andy.shevchenko@xxxxxxxxx> escribió:
> >>> On Thu, Mar 4, 2021 at 5:24 PM Álvaro Fernández Rojas <noltari@xxxxxxxxx> wrote:
> >>>>> El 4 mar 2021, a las 16:17, Andy Shevchenko <andy.shevchenko@xxxxxxxxx> escribió:
> >>>>> On Thu, Mar 4, 2021 at 5:06 PM Álvaro Fernández Rojas <noltari@xxxxxxxxx> wrote:
> >>>>>>> El 4 mar 2021, a las 11:35, Andy Shevchenko <andy.shevchenko@xxxxxxxxx> escribió:
> >>>>>>> On Thu, Mar 4, 2021 at 10:57 AM Álvaro Fernández Rojas
> >>>>>>> <noltari@xxxxxxxxx> wrote:
> >>>>>
> >>>>>>>> + * @of_node:           (Optional) The device node
> >>>>>>>
> >>>>>>>> +       struct device_node *of_node;
> >>>>>>>
> >>>>>>> Can we use fwnode from day 1, please?
> >>>>>>
> >>>>>> Could you explain this? I haven’t dealt with fwnode never :$
> >>>>>> BTW, this is done to fix this check when parsing gpio ranges:
> >>>>>> https://github.com/torvalds/linux/blob/f69d02e37a85645aa90d18cacfff36dba370f797/drivers/gpio/gpiolib-of.c#L933-L934
> >>>>>
> >>>>> Use struct fwnode_handle pointer instead of OF-specific one.
> >>>>
> >>>> But is that compatible with the current gpiolib-of code? :$
> >>>
> >>> Yes (after a bit of amendment I have sent today as v2:
> >>> https://lore.kernel.org/linux-gpio/20210304150215.80652-1-andriy.shevchenko@xxxxxxxxxxxxxxx/T/#u).
> >>
> >> Well that doesn’t fulfill my definition of “current gpiolib-of code”…
> >> @Linus what should I do about this?
> >
> > Well, fwnode is a generic, and I strongly against spreading
> > OF-specific code when we have fwnode working. But let's hear Linus
> > out, of course!
> >
> > But it seems you are right and the library needs a few more amendments.
>
> Yes, but I’m trying to do as few amendments as possible since I already have quite a large amount of patches :)

I understand your goal.
But as far I can say you don't need to rely on my patch series.

> >>>>> Also here is the question, why do you need to have that field in the
> >>>>> regmap config structure and can't simply use the parent's fwnode?
> >>>>> Also I'm puzzled why it's not working w/o this patch: GPIO library
> >>>>> effectively assigns parent's fwnode (okay, of_node right now).
> >>>>
> >>>> Because gpio regmap a child node of the pin controller, which is the one probed (gpio regmap is probed from the pin controller).
> >>>> Therefore the parent’s fwnode is useless, since the correct gpio_chip node is the child's one (we have pin-ranges declared in the child node, referencing the parent pinctrl node).
> >>>
> >>> I see. Can you point me out to the code where we get the node and
> >>> where it's being retrieved / filled?
> >>
> >> Sure, this is where the child node is searched: https://github.com/Noltari/linux/blob/6d1ebb8ff26ed54592eef1fcd3b58834acb48c04/drivers/pinctrl/bcm/pinctrl-bcm63xx.c#L100-L109
> >> Then the gpio child node is probed and assigned here: https://github.com/Noltari/linux/blob/6d1ebb8ff26ed54592eef1fcd3b58834acb48c04/drivers/pinctrl/bcm/pinctrl-bcm63xx.c#L51
> >
> > So, this is not (*yet) in upstream, correct?
>
> No it’s not, but I've already changed the approach several times and I’m starting to get tired about it...

I feel your pain, although I think that the best way is avoid
spreading OF-specifics over generic code.
Using fwnode API is pretty much straightforward. It has been designed
in a way to make it less invasive for existing code to be converted.
There are plenty of changes in the upstream that show how it looks
like.

You may check drivers under drivers/leds/ which have been switched to
fwnode (and AFAIK new code is usually not OF specific).

> > So, why not to switch to fwnode API in that driver as well?
> >
> > When you do that and supply fwnode thru the regmap configuration, in
> > the gpio-regmap we may assign it to of_node (via to_of_node() API).
> >
> >> Basically, I based that part of the code on the ingenic pin controller: https://github.com/torvalds/linux/blob/f69d02e37a85645aa90d18cacfff36dba370f797/drivers/pinctrl/pinctrl-ingenic.c#L2485-L2491
> >> https://github.com/torvalds/linux/blob/f69d02e37a85645aa90d18cacfff36dba370f797/Documentation/devicetree/bindings/pinctrl/ingenic%2Cpinctrl.yaml#L155-L176
> >
> > This doesn't use remgap GPIO.
>
> Yes, I know, but there aren’t any pinctrl drivers using regmap GPIO right now, so I couldn’t base my code on anything else :)

Be a pioneer!

-- 
With Best Regards,
Andy Shevchenko