Re: [PATCH V4 3/3] hwmon: pwm-fan: Add RPM support via external interrupt

[Robin Murphy <robin.murphy@xxxxxxx>]

On 03/04/2019 10:55, Stefan Wahren wrote:
> Hi Guenter,
>
> Am 02.04.19 um 22:55 schrieb Guenter Roeck:
> > On Tue, Apr 02, 2019 at 04:21:50PM +0200, Stefan Wahren wrote:
> > > This adds RPM support to the pwm-fan driver in order to use with
> > > fancontrol/pwmconfig. This feature is intended for fans with a tachometer
> > > output signal, which generate a defined number of pulses per revolution.
> > >
> > > Signed-off-by: Stefan Wahren <stefan.wahren@xxxxxxxx>
> > > ---
> > >   drivers/hwmon/pwm-fan.c | 111 ++++++++++++++++++++++++++++++++++++++++++++++--
> > >   1 file changed, 107 insertions(+), 4 deletions(-)
> > >
> > > diff --git a/drivers/hwmon/pwm-fan.c b/drivers/hwmon/pwm-fan.c
> > > index 167221c..3245a49 100644
> > > --- a/drivers/hwmon/pwm-fan.c
> > > +++ b/drivers/hwmon/pwm-fan.c
> > > @@ -18,6 +18,7 @@
> > >   
> > >   #include <linux/hwmon.h>
> > >   #include <linux/hwmon-sysfs.h>
> > > +#include <linux/interrupt.h>
> > >   #include <linux/module.h>
> > >   #include <linux/mutex.h>
> > >   #include <linux/of.h>
> > > @@ -26,6 +27,7 @@
> > >   #include <linux/regulator/consumer.h>
> > >   #include <linux/sysfs.h>
> > >   #include <linux/thermal.h>
> > > +#include <linux/timer.h>
> > >   
> > >   #define MAX_PWM 255
> > >   
> > > @@ -33,6 +35,14 @@ struct pwm_fan_ctx {
> > >   	struct mutex lock;
> > >   	struct pwm_device *pwm;
> > >   	struct regulator *reg_en;
> > > +
> > > +	int irq;
> > > +	atomic_t pulses;
> > > +	unsigned int rpm;
> > > +	u8 pulses_per_revolution;
> > > +	ktime_t sample_start;
> > > +	struct timer_list rpm_timer;
> > > +
> > >   	unsigned int pwm_value;
> > >   	unsigned int pwm_fan_state;
> > >   	unsigned int pwm_fan_max_state;
> > > @@ -40,6 +50,32 @@ struct pwm_fan_ctx {
> > >   	struct thermal_cooling_device *cdev;
> > >   };
> > >   
> > > +/* This handler assumes self resetting edge triggered interrupt. */
> > > +static irqreturn_t pulse_handler(int irq, void *dev_id)
> > > +{
> > > +	struct pwm_fan_ctx *ctx = dev_id;
> > > +
> > > +	atomic_inc(&ctx->pulses);
> > > +
> > > +	return IRQ_HANDLED;
> > > +}
> > > +
> > > +static void sample_timer(struct timer_list *t)
> > > +{
> > > +	struct pwm_fan_ctx *ctx = from_timer(ctx, t, rpm_timer);
> > > +	int pulses;
> > > +	u64 tmp;
> > > +
> > > +	pulses = atomic_read(&ctx->pulses);
> > > +	atomic_sub(pulses, &ctx->pulses);
> > > +	tmp = (u64)pulses * ktime_ms_delta(ktime_get(), ctx->sample_start) * 60;
> > > +	do_div(tmp, ctx->pulses_per_revolution * 1000);
> > > +	ctx->rpm = tmp;
> > > +
> > > +	ctx->sample_start = ktime_get();
> > > +	mod_timer(&ctx->rpm_timer, jiffies + HZ);
> > > +}
> > > +
> > >   static int  __set_pwm(struct pwm_fan_ctx *ctx, unsigned long pwm)
> > >   {
> > >   	unsigned long period;
> > > @@ -100,15 +136,49 @@ static ssize_t pwm_show(struct device *dev, struct device_attribute *attr,
> > >   	return sprintf(buf, "%u\n", ctx->pwm_value);
> > >   }
> > >   
> > > +static ssize_t rpm_show(struct device *dev,
> > > +			struct device_attribute *attr, char *buf)
> > > +{
> > > +	struct pwm_fan_ctx *ctx = dev_get_drvdata(dev);
> > > +
> > > +	return sprintf(buf, "%u\n", ctx->rpm);
> > > +}
> > >   
> > >   static SENSOR_DEVICE_ATTR_RW(pwm1, pwm, 0);
> > > +static SENSOR_DEVICE_ATTR_RO(fan1_input, rpm, 0);
> > >   
> > >   static struct attribute *pwm_fan_attrs[] = {
> > >   	&sensor_dev_attr_pwm1.dev_attr.attr,
> > > +	&sensor_dev_attr_fan1_input.dev_attr.attr,
> > >   	NULL,
> > >   };
> > >   
> > > -ATTRIBUTE_GROUPS(pwm_fan);
> > > +static umode_t pwm_fan_attrs_visible(struct kobject *kobj, struct attribute *a,
> > > +				     int n)
> > > +{
> > > +	struct device *dev = container_of(kobj, struct device, kobj);
> > > +	struct pwm_fan_ctx *ctx = dev_get_drvdata(dev);
> > > +	struct device_attribute *devattr;
> > > +
> > > +	/* Hide fan_input in case no interrupt is available  */
> > > +	devattr = container_of(a, struct device_attribute, attr);
> > > +	if (devattr == &sensor_dev_attr_fan1_input.dev_attr) {
> > > +		if (ctx->irq <= 0)
> > > +			return 0;
> > > +	}
> > Side note: This can be easier written as
> > 	if (n == 1 && ctx->irq <= 0)
> > 		return 0;
> >
> > Not that it matters much.
> >
> > > +
> > > +	return a->mode;
> > > +}
> > > +
> > > +static const struct attribute_group pwm_fan_group = {
> > > +	.attrs = pwm_fan_attrs,
> > > +	.is_visible = pwm_fan_attrs_visible,
> > > +};
> > > +
> > > +static const struct attribute_group *pwm_fan_groups[] = {
> > > +	&pwm_fan_group,
> > > +	NULL,
> > > +};
> > >   
> > >   /* thermal cooling device callbacks */
> > >   static int pwm_fan_get_max_state(struct thermal_cooling_device *cdev,
> > > @@ -261,17 +331,45 @@ static int pwm_fan_probe(struct platform_device *pdev)
> > >   		goto err_reg_disable;
> > >   	}
> > >   
> > > +	timer_setup(&ctx->rpm_timer, sample_timer, 0);
> > > +
> > > +	if (of_property_read_u8(pdev->dev.of_node, "pulses-per-revolution",
> > This does not work: The property is not defined as u8. You have to either
> > use of_property_read_u32() or declare the property as u8.
> pulses_per_revolution is defined as u8 since this version

The variable might be, but the "pulses-per-revolution" property itself 
is not being defined with the appropriate DT type ("/bits/ 8") in the 
binding, and thus will be stored as a regular 32-bit cell, for which 
reading it as a u8 array may or may not work correctly depending on 
endianness.

TBH, unless there's a real need for a specific binary format in the FDT, 
I don't think it's usually worth the bother of using irregular DT types, 
especially when the practical impact amounts to possibly saving up to 3 
bytes for a property which usually won't need to be specified anyway. 
I'd just do something like:

	u32 ppr = 2;

	of_property_read_u32(np, "pulses-per-revolution", &ppr);
	ctx->pulses_per_revolution = ppr;

> > [ Sorry, I didn't know until recently that this is necessary ]
> >
> > > +				&ctx->pulses_per_revolution)) {
> > > +		ctx->pulses_per_revolution = 2;
> > > +	}
> > > +
> > > +	if (!ctx->pulses_per_revolution) {
> > > +		dev_err(&pdev->dev, "pulses-per-revolution can't be zero.\n");
> > > +		ret = -EINVAL;
> > > +		goto err_pwm_disable;
> > > +	}
> > > +
> > > +	ctx->irq = platform_get_irq(pdev, 0);
> > > +	if (ctx->irq == -EPROBE_DEFER) {
> > > +		ret = ctx->irq;
> > > +		goto err_pwm_disable;
> > It might be better to call platform_get_irq() and to do do this check
> > first, before enabling the regulator (in practice before calling
> > devm_regulator_get_optional). It doesn't make sense to enable the
> > regulator only to disable it because the irq is not yet available.
> >
> > > +	} else if (ctx->irq > 0) {
> > As written, this else is unnecessary, and static checkers will complain
> > about it.
> >
> > > +		ret = devm_request_irq(&pdev->dev, ctx->irq, pulse_handler, 0,
> > > +				       pdev->name, ctx);
> > > +		if (ret) {
> > > +			dev_err(&pdev->dev, "Can't get interrupt working.\n");
> > > +			goto err_pwm_disable;

We could still continue without RPM support at this point, couldn't we? 
Or is this a deliberate "if that failed, then who knows how messed up 
the system is..." kind of thing?

Robin.

> > > +		}
> > > +		ctx->sample_start = ktime_get();
> > > +		mod_timer(&ctx->rpm_timer, jiffies + HZ);
> > > +	}
> > > +
> > >   	hwmon = devm_hwmon_device_register_with_groups(&pdev->dev, "pwmfan",
> > >   						       ctx, pwm_fan_groups);
> > >   	if (IS_ERR(hwmon)) {
> > >   		dev_err(&pdev->dev, "Failed to register hwmon device\n");
> > >   		ret = PTR_ERR(hwmon);
> > > -		goto err_pwm_disable;
> > > +		goto err_del_timer;
> > >   	}
> > >   
> > >   	ret = pwm_fan_of_get_cooling_data(&pdev->dev, ctx);
> > >   	if (ret)
> > > -		return ret;
> > > +		goto err_del_timer;
> > Outch. This is buggy and should have been "goto err_pwm_disable;".
> > It needs to be fixed with a separate patch, and first, so we can
> > backport it. Can you do that ?
>
> Sure
>
> Stefan