Re: [PATCH v2 1/2] mfd: rohm-bd71828: Use software nodes for gpio-keys

[Dmitry Torokhov <dmitry.torokhov@gmail.com>]

On Tue, Mar 24, 2026 at 08:45:15AM +0200, Matti Vaittinen wrote:
> On 23/03/2026 03:37, Dmitry Torokhov wrote:
> > Refactor the rohm-bd71828 MFD driver to use software nodes for
> > instantiating the gpio-keys child device, replacing the old
> > platform_data mechanism.
> > 
> > The power key's properties are now defined using software nodes and
> > property entries. The IRQ is passed as a resource attached to the
> > platform device.
> > 
> > This will allow dropping support for using platform data for configuring
> > gpio-keys in the future.
> > 
> > Signed-off-by: Dmitry Torokhov <dmitry.torokhov@gmail.com>
> 
> Thanks a lot Dmitry for converting this to the swnodes. I like the idea very
> much :) A few minor, (mostly styling related as I am a bit old-fashioned)
> comments.
> 
> > ---
> >   drivers/mfd/rohm-bd71828.c | 117 ++++++++++++++++++++++++++++++++-------------
> >   1 file changed, 85 insertions(+), 32 deletions(-)
> > 
> > diff --git a/drivers/mfd/rohm-bd71828.c b/drivers/mfd/rohm-bd71828.c
> > index a79f354bf5cb..20b7910e7f63 100644
> > --- a/drivers/mfd/rohm-bd71828.c
> > +++ b/drivers/mfd/rohm-bd71828.c
> > @@ -5,7 +5,8 @@
> >    * ROHM BD718[15/28/79] and BD72720 PMIC driver
> >    */
> > -#include <linux/gpio_keys.h>
> > +#include <linux/device/devres.h>
> > +#include <linux/gfp_types.h>
> >   #include <linux/i2c.h>
> >   #include <linux/input.h>
> >   #include <linux/interrupt.h>
> > @@ -18,6 +19,7 @@
> >   #include <linux/mfd/rohm-generic.h>
> >   #include <linux/module.h>
> >   #include <linux/of.h>
> > +#include <linux/property.h>
> >   #include <linux/regmap.h>
> >   #include <linux/types.h>
> > @@ -37,19 +39,6 @@
> >   		},							   \
> >   	}
> > -static struct gpio_keys_button button = {
> > -	.code = KEY_POWER,
> > -	.gpio = -1,
> > -	.type = EV_KEY,
> > -	.wakeup = 1,
> > -};
> > -
> > -static const struct gpio_keys_platform_data bd71828_powerkey_data = {
> > -	.buttons = &button,
> > -	.nbuttons = 1,
> > -	.name = "bd71828-pwrkey",
> > -};
> > -
> >   static const struct resource bd71815_rtc_irqs[] = {
> >   	DEFINE_RES_IRQ_NAMED(BD71815_INT_RTC0, "bd70528-rtc-alm-0"),
> >   	DEFINE_RES_IRQ_NAMED(BD71815_INT_RTC1, "bd70528-rtc-alm-1"),
> > @@ -174,11 +163,8 @@ static struct mfd_cell bd71828_mfd_cells[] = {
> >   		.name = "bd71828-rtc",
> >   		.resources = bd71828_rtc_irqs,
> >   		.num_resources = ARRAY_SIZE(bd71828_rtc_irqs),
> > -	}, {
> > -		.name = "gpio-keys",
> > -		.platform_data = &bd71828_powerkey_data,
> > -		.pdata_size = sizeof(bd71828_powerkey_data),
> >   	},
> > +	/* Power button is registered separately */
> >   };
> >   static const struct resource bd72720_power_irqs[] = {
> > @@ -242,11 +228,8 @@ static const struct mfd_cell bd72720_mfd_cells[] = {
> >   		.name = "bd72720-rtc",
> >   		.resources = bd72720_rtc_irqs,
> >   		.num_resources = ARRAY_SIZE(bd72720_rtc_irqs),
> > -	}, {
> > -		.name = "gpio-keys",
> > -		.platform_data = &bd71828_powerkey_data,
> > -		.pdata_size = sizeof(bd71828_powerkey_data),
> >   	},
> > +	/* Power button is registered separately */
> >   };
> >   static const struct regmap_range bd71815_volatile_ranges[] = {
> > @@ -877,6 +860,75 @@ static int set_clk_mode(struct device *dev, struct regmap *regmap,
> >   				  OUT32K_MODE_CMOS);
> >   }
> > +static void bd71828_i2c_unregister_swnodes(void *data)
> > +{
> > +	const struct software_node *nodes = data;
> > +
> > +	software_node_unregister_node_group((const struct software_node *[]){
> > +		&nodes[0],
> > +		&nodes[1],
> > +		NULL
> > +	});
> 
> Perhaps it was possible to use a temporary variable for the software_node
> pointer array? It would allow also old-school fellows like me to pick the
> meaning by a glance :)

Done.

> 
> > +}
> > +
> > +static int bd71828_i2c_register_pwrbutton(struct device *dev, int button_irq,
> > +					  struct irq_domain *irq_domain)
> > +{
> > +	static const struct property_entry bd71828_powerkey_parent_props[] = {
> > +		PROPERTY_ENTRY_STRING("label", "bd71828-pwrkey"),
> > +		{ }
> > +	};
> > +	static const struct property_entry bd71828_powerkey_props[] = {
> > +		PROPERTY_ENTRY_U32("linux,code", KEY_POWER),
> > +		PROPERTY_ENTRY_BOOL("wakeup-source"),
> > +		{ }
> > +	};
> > +	struct software_node *nodes;
> > +	int error;
> > +
> > +	nodes = devm_kcalloc(dev, 2, sizeof(*nodes), GFP_KERNEL);
> > +	if (!nodes)
> > +		return -ENOMEM;
> > +
> > +	/* Node corresponding to gpio-keys device itself */
> > +	nodes[0].name = devm_kasprintf(dev, GFP_KERNEL, "%s-power-key", dev_name(dev));
> > +	if (!nodes[0].name)
> > +		return -ENOMEM;
> > +
> > +	nodes[0].properties = bd71828_powerkey_parent_props;
> > +
> > +	/* Node representing power button within gpio-keys device */
> > +	nodes[1].parent = &nodes[0];
> > +	nodes[1].properties = bd71828_powerkey_props;
> > +
> > +	error = software_node_register_node_group((const struct software_node *[]){
> > +		&nodes[0],
> > +		&nodes[1],
> > +		NULL
> > +	});
> 
> I think having a temporary variable might make this to look more familiar.

Done.

> 
> > +	if (error)
> > +		return error;
> > +
> > +	error = devm_add_action_or_reset(dev, bd71828_i2c_unregister_swnodes, nodes);
> > +	if (error)
> > +		return error;
> > +
> > +	const struct mfd_cell gpio_keys_cell = {
> > +		.name = "gpio-keys",
> > +		.resources = (const struct resource[]) {
> > +			DEFINE_RES_IRQ_NAMED(button_irq, "bd71828-pwrkey"),
> > +		},
> > +		.num_resources = 1,
> > +		.swnode = &nodes[0],
> > +	};
> 
> I don't really love seeing variables declared in the middle of a block.

I normally try not to declare in the middle of the code but in this case
I opted for it because swnode is not known so we have to split
initialization. 

> Perhaps consider splitting the software-node preparation in own function and
> doing something like (completely untested thought):
> 
> ret = alloc_and_prepare_the_swnode_stuff(...);
> if (!ret) {
> 	const struct software_node *nodes[] = { ... };
> 	const struct mfd_cell gpio_keys_cell = { ... };
> 	
> 	...
> }

That is an option but then spreads dealing with nodes across multiple
functions. I moved the declaration and partial initialization to the
beginning of the function, and assign swnode before calling
devm_mfd_add_devices().

> 
> or alternatively, split the software-node and MFD device registration into
> own function? Do you think it would work?
> 
> > +	error = devm_mfd_add_devices(dev, PLATFORM_DEVID_AUTO, &gpio_keys_cell, 1,
> > +				     NULL, 0, irq_domain);
> > +	if (error)
> > +		return dev_err_probe(dev, error, "Failed to create power button subdevice");
> > +
> > +	return 0;
> > +}
> > +
> >   static struct i2c_client *bd71828_dev;
> >   static void bd71828_power_off(void)
> >   {
> > @@ -929,6 +981,7 @@ static struct regmap *bd72720_do_regmaps(struct i2c_client *i2c)
> >   static int bd71828_i2c_probe(struct i2c_client *i2c)
> >   {
> >   	struct regmap_irq_chip_data *irq_data;
> > +	struct irq_domain *irq_domain;
> >   	int ret;
> >   	struct regmap *regmap = NULL;
> >   	const struct regmap_config *regmap_config;
> > @@ -1008,6 +1061,8 @@ static int bd71828_i2c_probe(struct i2c_client *i2c)
> >   	dev_dbg(&i2c->dev, "Registered %d IRQs for chip\n",
> >   		irqchip->num_irqs);
> > +	irq_domain = regmap_irq_get_domain(irq_data);
> 
> nit: Maybe move this call closer to where the irq_domain is actually needed
> for the first time.

Done.

Thanks.

-- 
Dmitry