On Tue, 12 Feb 2019, Dmitry Torokhov wrote:

> Hi Lee,
>
> On Tue, Feb 12, 2019 at 12:34 PM Lee Jones <[email protected]> wrote:
> >
> > > > From: Bartosz Golaszewski <[email protected]>
> > > >
> > > > Add support for the push- and slide-button events for max77650.
> > > >
> > > > Signed-off-by: Bartosz Golaszewski <[email protected]>
> > > > ---
> > > > drivers/input/misc/Kconfig | 9 ++
> > > > drivers/input/misc/Makefile | 1 +
> > > > drivers/input/misc/max77650-onkey.c | 135 ++++++++++++++++++++++++++++
> > > > 3 files changed, 145 insertions(+)
> > > > create mode 100644 drivers/input/misc/max77650-onkey.c
> >
> > [...]
> >
> > Moving things around a bit:
> >
> > > > +static int max77650_onkey_probe(struct platform_device *pdev)
> > > > +{
> >
> > > > + irq_f = regmap_irq_get_virq(irq_data, MAX77650_INT_nEN_F);
> > > > + if (irq_f <= 0)
> > > > + return -EINVAL;
> > > > +
> > > > + irq_r = regmap_irq_get_virq(irq_data, MAX77650_INT_nEN_R);
> > > > + if (irq_r <= 0)
> > > > + return -EINVAL;
> > >
> > > Ugh, it would be better if you handled IRQ mapping in the MFD piece and
> > > passed it as resources of platform device. Then you'd simply call
> > > platform_get_irq() here and did not have to reach into parent device for
> > > "irq_dara".
> >
> > These device IRQs were defined and registered with the Regmap *set*
> > (actually init()) APIs and thus should be pulled out using the
> > appropriate reverse Regmap *get* APIs here in the device.
> >
> > Registering them with Regmap *and* pulling them back out again in the
> > same (MFD in this case) driver, only to register them as platform data
> > is certainly not how I see the API being designed/used.
> >
> > > > + struct regmap_irq_chip_data *irq_data;
> > > > + struct device *dev, *parent;
> >
> > > > + dev = &pdev->dev;
> > > > + parent = dev->parent;
> > > > + i2c = to_i2c_client(parent);
> > > > + irq_data = i2c_get_clientdata(i2c);
> > > > +
> > > > + map = dev_get_regmap(parent, NULL);
> > > > + if (!map)
> > > > + return -ENODEV;
> >
> > However, this hoop jumping is a bit crazy and for the most part
> > superfluous. Instead, create a struct of attributes you wish to share
> > with the child devices (containing both regmap (which you should call
> > regmap and not map by the way) and irq_data) and pass it as either
> > platform data (preferred) or as device data.
> >
> > If you choose the latter, you do not need to convert from 'device' to
> > 'i2c' to do the look-up. Since this function (probe()) is provided
> > with a platform_device, you can just use platform_get_drvdata() to
> > achieve the same as above.
> >
> > If you go the preferred (platform data) route, then you should use
> > dev_get_platdata() instead.
>
> By doing what you are suggesting (introducing platform data) you
> introducing strong dependency between MFD and input piece for no
> different reason. With the current implementation the parent can be
> reworked completely without involving onkey driver.
>
> I find such clean separation desirable. We are trying to move away
> form platform data where it makes sense.

Never mind. We found a way forward where everyone wins!

Thanks for your time.

--
Lee Jones [李琼斯]
Linaro Services Technical Lead
Linaro.org │ Open source software for ARM SoCs
Follow Linaro: Facebook | Twitter | Blog