Re: [PATCH 1/2] gpio: regmap: Support few IC specific operations

[Vaittinen, Matti]

On Thu, 2021-05-20 at 14:22 +0200, Michael Walle wrote:
> Am 2021-05-20 14:00, schrieb Matti Vaittinen:
> > On Thu, 2021-05-20 at 13:42 +0200, Michael Walle wrote:
> > > Am 2021-05-20 13:28, schrieb Matti Vaittinen:
> > > > The set_config and init_valid_mask GPIO operations are usually
> > > > very
> > > > IC
> > > > specific. Allow IC drivers to provide these custom operations
> > > > at
> > > > gpio-regmap registration.
> > > > 
> > > > Signed-off-by: Matti Vaittinen <
> > > > matti.vaittinen@xxxxxxxxxxxxxxxxx>
> > > > ---
> > > >  drivers/gpio/gpio-regmap.c  | 49
> > > > +++++++++++++++++++++++++++++++++++++
> > > >  include/linux/gpio/regmap.h | 13 ++++++++++
> > > >  2 files changed, 62 insertions(+)
> > > > 
> > > > diff --git a/drivers/gpio/gpio-regmap.c b/drivers/gpio/gpio-
> > > > regmap.c
> > > > index 134cedf151a7..315285cacd3f 100644
> > > > --- a/drivers/gpio/gpio-regmap.c
> > > > +++ b/drivers/gpio/gpio-regmap.c
> > > > @@ -27,6 +27,10 @@ struct gpio_regmap {
> > > >  	int (*reg_mask_xlate)(struct gpio_regmap *gpio,
> > > > unsigned int
> > > > base,
> > > >  			      unsigned int offset, unsigned int
> > > > *reg,
> > > >  			      unsigned int *mask);
> > > > +	int (*set_config)(struct regmap *regmap, void *drvdata,
> > > > +			  unsigned int offset, unsigned long
> > > > config);
> > > > +	int (*init_valid_mask)(struct regmap *regmap, void
> > > > *drvdata,
> > > > +				unsigned long *valid_mask,
> > > > unsigned int
> > > > ngpios);
> > > 
> > > Maybe we should also make the first argument a "struct
> > > gpio_regmap"
> > > and provide a new gpio_regmap_get_regmap(struct gpio_regmap).
> > > Thus
> > > having a similar api as for the reg_mask_xlate(). Andy?
> > 
> > I don't really see the reason of making this any more complicated
> > for
> > IC drivers. If we don't open the struct gpio_regmap to IC drivers -
> > then they never need the struct gpio_regmap pointer itself but each
> > IC
> > driver would need to do some unnecessary function call
> > (gpio_regmap_get_regmap() in this case). I'd say that would be
> > unnecessary bloat.
> 
> If there is ever the need of additional parameters, you'll have to
> modify that parameter list. Otherwise you'll just have to add a new
> function. Thus might be more future proof.

I do hope the "void *drvdata" allows enough flexibility so that there
is no need to add new parameters. And if there is, then I don't see how
the struct gpio_regmap pointer would have saved us - unless we open the
contents of struct gpio_regmap to IC drivers. (Which might make sense
because that already contains plenty of register details which may need
to be duplicated to drvdata for some IC-specific callbacks. Here we
again have analogy to regulator_desc - which I have often used also in
IC-specific custom callbacks. But as long as we hope to keep the struct
gpio_regmap private I would not add it in arguments).

> But I won't object to it.
> > > > @@ -39,6 +43,43 @@ static unsigned int
> > > > gpio_regmap_addr(unsigned
> > > > int
> > > > addr)
> > > >  	return addr;
> > > >  }
> > > > 
> > > > +static int regmap_gpio_init_valid_mask(struct gpio_chip *gc,
> > > > +					unsigned long
> > > > *valid_mask,
> > > > +					unsigned int ngpios)
> > > > +{
> > > > +	struct gpio_regmap *gpio;
> > > > +	void *drvdata;
> > > > +
> > > > +	gpio = gpiochip_get_data(gc);
> > > > +
> > > > +	if (!gpio->init_valid_mask) {
> > > > +		WARN_ON(!gpio->init_valid_mask);
> > > > +		return -EINVAL;
> > > > +	}
> > > 
> > > Why not the following?
> > > 
> > > if (!gpio->init_valid_mask)
> > >      return 0;
> > 
> > It just feels like an error if regmap_gpio_init_valid_mask() is
> > ever
> > called by core without having the gpio->init_valid_mask set.
> > Probably
> > this would mean that the someone has errorneously modified the
> > gpio-
> > > init_valid_mask set after gpio_regmap registration - whih smells
> > > like
> > a problem. Thus the WARN() sounds like a correct course of action
> > to
> > me. (I may be wrong though - see below)
> > 
> > > Thus copying the behavior of gpiolib.
> > 
> > I must admit I didn't check how this is implemented in gpiolib. But
> > the
> > gpio_chip's init_valid_mask should not be set if regmap_gpio_config
> > does not have valid init_valid_mask pointer at registration. Thus
> > it
> > smells like an error to me if the GPIO core calls the
> > regmap_gpio_init_valid_mask() and regmap_gpio has not set the
> > init_valid_mask pointer. But as I said, I haven't looked in gpiolib
> > for
> > this so I may be wrong.
> 
> Oh, I missed that you only set it when it is set in the
> gpio_regmap_config. Thus, I'd say drop it entirely? It is only within
> this module where things might go wrong.

I have no strong opinion on this. I can drop it if it's not needed.

> > > > +
> > > > +	drvdata = gpio_regmap_get_drvdata(gpio);
> > > > +
> > > > +	return gpio->init_valid_mask(gpio->regmap, drvdata,
> > > > valid_mask,
> > > > ngpios);
> > > > +}
> > > > +
> > > > +static int gpio_regmap_set_config(struct gpio_chip *gc,
> > > > unsigned
> > > > int
> > > > offset,
> > > > +				  unsigned long config)
> > > > +{
> > > > +	struct gpio_regmap *gpio;
> > > > +	void *drvdata;
> > > > +
> > > > +	gpio = gpiochip_get_data(gc);
> > > > +
> > > > +	if (!gpio->set_config) {
> > > > +		WARN_ON(!gpio->set_config);
> > > > +		return -EINVAL;
> > > > +	}
> > > 
> > > same here, return -ENOTSUPP.
> > 
> > As above -
> > if (!gpio->set_config) {
> > 	the gpio-core should never call gpio_regmap_set_config() if the
> > }
> > 
> > Maybe I should add a comment to clarify the WARN() ?
> > > > +
> > > > +	drvdata = gpio_regmap_get_drvdata(gpio);
> > > > +
> > > > +	return gpio->set_config(gpio->regmap, drvdata, offset,
> > > > config);
> > > > +}
> > > > +
> > > >  static int gpio_regmap_simple_xlate(struct gpio_regmap *gpio,
> > > >  				    unsigned int base, unsigned
> > > > int
> > > > offset,
> > > >  				    unsigned int *reg, unsigned
> > > > int
> > > > *mask)
> > > > @@ -235,6 +276,8 @@ struct gpio_regmap
> > > > *gpio_regmap_register(const
> > > > struct gpio_regmap_config *config
> > > >  	gpio->reg_clr_base = config->reg_clr_base;
> > > >  	gpio->reg_dir_in_base = config->reg_dir_in_base;
> > > >  	gpio->reg_dir_out_base = config->reg_dir_out_base;
> > > > +	gpio->set_config = config->set_config;
> > > > +	gpio->init_valid_mask = config->init_valid_mask;
> > > > 
> > > >  	/* if not set, assume there is only one register */
> > > >  	if (!gpio->ngpio_per_reg)
> > > > @@ -253,6 +296,10 @@ struct gpio_regmap
> > > > *gpio_regmap_register(const
> > > > struct gpio_regmap_config *config
> > > >  	chip->ngpio = config->ngpio;
> > > >  	chip->names = config->names;
> > > >  	chip->label = config->label ?: dev_name(config-
> > > > > parent);
> > > > +	if (gpio->set_config)
> > > > +		chip->set_config = gpio_regmap_set_config;
> > > > +	if (gpio->init_valid_mask)
> > > > +		chip->init_valid_mask =
> > > > regmap_gpio_init_valid_mask;
> > > > 
> > > >  #if defined(CONFIG_OF_GPIO)
> > > >  	/* gpiolib will use of_node of the parent if chip-
> > > > > of_node is
> > > > NULL */
> > > > @@ -280,6 +327,8 @@ struct gpio_regmap
> > > > *gpio_regmap_register(const
> > > > struct gpio_regmap_config *config
> > > >  		chip->direction_output =
> > > > gpio_regmap_direction_output;
> > > >  	}
> > > > 
> > > > +	gpio_regmap_set_drvdata(gpio, config->drvdata);
> > > 
> > > I'm wondering if we need the gpio_regmap_set_drvdata() anymore or
> > > if
> > > we can just drop it entirely.
> > 
> > I wouldn't drop it. I think there _may_ be cases where the drvdata
> > is
> > set only after the registration. (Just my gut-feeling, I may be
> > wrong
> > though)
> 
> Ok, but as you already mentioned on IRC, it could be a bit of a trap
> because it might already be used after gpiochip_add_data() and thus
> be NULL if not provided by gpio_regmap_config().

Hmm.. I think you are right. Setting the drvdata after registration is
a call for a race. After that reasoning I agree with you that this
should be dropped.

Best Regards
	Matti Vaittinen