Re: [PATCH v4 1/2] gpio: Add support for IDT 79RC3243x GPIO controller

[Andy Shevchenko <andy.shevchenko@gmail.com>]

On Wed, Apr 28, 2021 at 5:48 PM Michael Walle <michael@walle.cc> wrote:
>
> [Adding Mark here, too]
>
> Am 2021-04-28 16:32, schrieb Andy Shevchenko:
> > On Wed, Apr 28, 2021 at 5:04 PM Michael Walle <michael@walle.cc> wrote:
> >> Am 2021-04-28 15:44, schrieb Andy Shevchenko:
> >> > On Wed, Apr 28, 2021 at 2:57 PM Michael Walle <michael@walle.cc> wrote:
> >> >>
> >> >> Am 2021-04-28 13:07, schrieb Andy Shevchenko:
> >> >> > On Wed, Apr 28, 2021 at 1:51 AM Michael Walle <michael@walle.cc> wrote:
> >> >> >> Am 2021-04-26 12:29, schrieb Andy Shevchenko:
> >> >> >> > On Mon, Apr 26, 2021 at 12:55 PM Thomas Bogendoerfer
> >> >> >> > <tsbogend@alpha.franken.de> wrote:
> >> >> >> >
> >> >> >> > 2) there is gpio-regmap generic code, that may be worth
> >> >> >> > considering.
> >> >> >>
> >> >> >> This driver uses memory mapped registers. While that is
> >> >> >> also possible with gpio-regmap, there is one drawback:
> >> >> >> it assumes gpiochip->can_sleep = true for now, see [1].
> >> >> >> Unfortunately, there is no easy way to ask the regmap
> >> >> >> if its mmio/fastio.
> >> >> >
> >> >> > I don't see how it is an impediment.
> >> >>
> >> >> You'd have to use the *_cansleep() variants with the gpios,
> >> >> which cannot be used everywhere, no?
> >> >
> >> > *can* sleep means that it requires a sleeping context to run, if your
> >> > controller is fine with that, there are no worries. OTOH if you want
> >> > to run this in an atomic context, then consumers can't do with that
> >> > kind of controller.
> >>
> >> Ok, then we are on the same track.
> >>
> >> > What I meant above (and you stripped it here) is
> >> > to add a patch that will fix that and set it based on
> >> > gpio_regmap_config.
> >>
> >> Yes, but ideally, it would ask the regmap. Otherwise that
> >> information is redundant and might mismatch, i.e. gpio_regmap_config
> >> tell can_sleep=false but the regmap is an I2C type for example. Also
> >> if a driver wants to support both regmap types, we are no step
> >> further.
> >
> > Yeah, I agree that is a band aid, but you are free to fix it actually
> > on regmap level.
> > I don't think it will require an enormous amount of work there.
>
> I'd love to fix that, but Mark was against exposing that property
> outside of regmap. So it it what it is for now ;) Maybe he'll change
> his mind or someone has another idea.

Then let's go to ugly variant with duplicating it in gpio-regmap
config. with a FIXME note or so. I don't think we should allow new
drivers be based on bgpio_init().

-- 
With Best Regards,
Andy Shevchenko