Re: [PATCH 1/2] hwmon: Add devicetree bindings to gpio-fan

[Guenter Roeck <linux@roeck-us.net>]

On Fri, Sep 07, 2012 at 05:34:34PM +0100, Jamie Lentin wrote:
> Allow a gpio-fan to be defined in devicetree, see binding documentation
> for details.
> 
> Signed-off-by: Jamie Lentin <jm@lentin.co.uk>
> ---
>  .../devicetree/bindings/gpio/gpio-fan.txt          |   25 +++++
>  drivers/hwmon/gpio-fan.c                           |  111 ++++++++++++++++++++
>  2 files changed, 136 insertions(+)
>  create mode 100644 Documentation/devicetree/bindings/gpio/gpio-fan.txt
> 
> diff --git a/Documentation/devicetree/bindings/gpio/gpio-fan.txt b/Documentation/devicetree/bindings/gpio/gpio-fan.txt
> new file mode 100644
> index 0000000..2dd457a
> --- /dev/null
> +++ b/Documentation/devicetree/bindings/gpio/gpio-fan.txt
> @@ -0,0 +1,25 @@
> +Bindings for fan connected to GPIO lines
> +
> +Required properties:
> +- compatible : "gpio-fan"
> +- gpios: Specifies the pins that map to bits in the control value,
> +  ordered MSB-->LSB.
> +- gpio-fan,speed-map: A mapping of possible fan RPM speeds and the
> +  control value that should be set to achieve them. This array
> +  must have the RPM values in ascending order.
> +
> +Optional properties:
> +- alarm-gpios: This pin going active indicates something is wrong with
> +  the fan, and a udev event will be fired.
> +
> +Examples:
> +
> +	gpio_fan {
> +		compatible = "gpio-fan";
> +		gpios = <&gpio1 14 1
> +			 &gpio1 13 1>;
> +		gpio-fan,speed-map = <0    0
> +				      3000 1
> +				      6000 2>;
> +		alarm-gpios = <&gpio1 15 1>;
> +	};
> diff --git a/drivers/hwmon/gpio-fan.c b/drivers/hwmon/gpio-fan.c
> index 2f4b01b..876ce41 100644
> --- a/drivers/hwmon/gpio-fan.c
> +++ b/drivers/hwmon/gpio-fan.c
> @@ -31,6 +31,8 @@
>  #include <linux/hwmon.h>
>  #include <linux/gpio.h>
>  #include <linux/gpio-fan.h>
> +#include <linux/of_platform.h>
> +#include <linux/of_gpio.h>
>  
>  struct gpio_fan_data {
>  	struct platform_device	*pdev;
> @@ -400,12 +402,120 @@ static ssize_t show_name(struct device *dev,
>  
>  static DEVICE_ATTR(name, S_IRUGO, show_name, NULL);
>  
> +
> +#ifdef CONFIG_OF
> +/*
> + * Translate OpenFirmware node properties into platform_data
> + */
> +static int gpio_fan_get_of_pdata(struct device *dev,
> +			    struct gpio_fan_platform_data *pdata)
> +{
> +	struct device_node *node;
> +	struct gpio_fan_speed  *speed;
> +	unsigned *ctrl;
> +	unsigned i;
> +	u32 u;
> +	struct property *prop;
> +	const __be32 *p;
> +
> +	node = dev->of_node;
> +
> +	/* Fill GPIO pin array */
> +	pdata->num_ctrl = of_gpio_count(node);
> +	ctrl = devm_kzalloc(dev, pdata->num_ctrl * sizeof(unsigned),
> +				GFP_KERNEL);
> +	if (!ctrl)
> +		return -ENOMEM;
> +	for (i = 0; i < of_gpio_count(node); i++) {
> +		int val;
> +
> +		val = of_get_gpio(node, i);
> +		if (val >= 0)
> +			ctrl[i] = val;
> +		else
> +			return -EINVAL;

I very much prefer not to hide the error code and reverse the logic.

		if (val < 0)
			return val;
		ctrl[i] = val;

> +	}
> +	pdata->ctrl = ctrl;
> +
> +	/* Get speed map array size */
> +	i = 0;
> +	of_property_for_each_u32(node, "gpio-fan,speed-map", prop, p, u)
> +		i++;

Looking through the of code, isn't there a function which returns the
number of elements ?

Something like
	int length;
	struct property *prop;
	
	prop = of_find_property(node, "gpio-fan,speed-map", &length);

> +	if (i & 1) {
> +		dev_err(dev, "gpio-fan,speed-map contains odd number of entries");
> +		return -ENODEV;
> +	}

How about i == 0 ?

> +	pdata->num_speed = i >> 1;
> +	speed = devm_kzalloc(dev,
> +			pdata->num_speed * sizeof(struct gpio_fan_speed),
> +			GFP_KERNEL);
> +	if (!speed)
> +		return -ENOMEM;
> +
> +	/* Populate speed map */
> +	i = 0;
> +	of_property_for_each_u32(node, "gpio-fan,speed-map", prop, p, u) {
> +		if (i & 1)
> +			speed[i >> 1].ctrl_val = (int)u;
> +		else
> +			speed[i >> 1].rpm = (int)u;

Are those type casts necessary ?

> +		i++;
> +	}
> +	pdata->speed = speed;
> +
> +	/* Alarm GPIO if one exists */
> +	if (of_gpio_named_count(node, "alarm-gpios")) {
> +		struct gpio_fan_alarm *alarm;
> +		int val;
> +		enum of_gpio_flags flags;
> +
> +		alarm = devm_kzalloc(dev, sizeof(struct gpio_fan_alarm),
> +					GFP_KERNEL);
> +		if (!alarm)
> +			return -ENOMEM;
> +
> +		val = of_get_named_gpio_flags(node, "alarm-gpios", 0, &flags);
> +		if (val >= 0) {
> +			alarm->gpio = val;
> +			alarm->active_low = flags & OF_GPIO_ACTIVE_LOW;
> +		} else
> +			return -EINVAL;
> +
How about
		if (val < 0)
			return val;
		alarm->gpio = val;
		alarm->active_low = flags & OF_GPIO_ACTIVE_LOW;

> +		pdata->alarm = alarm;
> +	}
> +
> +	return 0;
> +}
> +
> +static const struct of_device_id of_gpio_fan_match[] = {
> +	{ .compatible = "gpio-fan", },
> +	{},
> +};
> +#endif /* CONFIG_OF_GPIO */
> +
>  static int __devinit gpio_fan_probe(struct platform_device *pdev)
>  {
>  	int err;
>  	struct gpio_fan_data *fan_data;
>  	struct gpio_fan_platform_data *pdata = pdev->dev.platform_data;
>  
> +#ifdef CONFIG_OF
> +	if (!pdata) {
> +		struct gpio_fan_platform_data *alt_pdata;
> +
Why not just use pdata ? I don't really see a reason for introducing
a second variable. After all, we know here that pdata is NULL, and that it won't
be used.

> +		alt_pdata = devm_kzalloc(&pdev->dev,
> +					sizeof(struct gpio_fan_platform_data),
> +					GFP_KERNEL);
> +		if (!alt_pdata)
> +			return -ENOMEM;
> +
> +		err = gpio_fan_get_of_pdata(&pdev->dev, alt_pdata);
> +		if (err)
> +			return err;
> +		pdata = alt_pdata;
> +	}
> +#endif /* CONFIG_OF_GPIO */
> +
>  	if (!pdata)
>  		return -EINVAL;

#else case, maybe ?

>  
> @@ -511,6 +621,7 @@ static struct platform_driver gpio_fan_driver = {
>  	.driver	= {
>  		.name	= "gpio-fan",
>  		.pm	= GPIO_FAN_PM,
> +		.of_match_table = of_match_ptr(of_gpio_fan_match),
>  	},
>  };
>  
> -- 
> 1.7.10.4
> 
>