On Sun, Jan 15, 2017 at 07:04:09PM -0600, David Lechner wrote:

On 01/15/2017 06:34 PM, Dmitry Torokhov wrote:

quoted

On Sun, Jan 15, 2017 at 06:12:29PM -0600, David Lechner wrote:

quoted

On 01/14/2017 01:19 PM, Dmitry Torokhov wrote:

quoted

On Wed, Jan 11, 2017 at 02:02:01PM -0600, David Lechner wrote:

quoted

This adds an optional regulator to the pwm-beeper device. This regulator
acts as an amplifier. The amplifier is only enabled while beeping in order
to reduce power consumption.

Tested on LEGO MINDSTORMS EV3, which has a speaker connected to PWM through
an amplifier.

Signed-off-by: David Lechner <david@lechnology.com>
---
drivers/input/misc/pwm-beeper.c | 29 ++++++++++++++++++++++++++++-
1 file changed, 28 insertions(+), 1 deletion(-)

diff --git a/drivers/input/misc/pwm-beeper.c b/drivers/input/misc/pwm-beeper.c
index 30ac227..708e88e 100644
--- a/drivers/input/misc/pwm-beeper.c
+++ b/drivers/input/misc/pwm-beeper.c

@@ -14,6 +14,7 @@

*/

#include <linux/input.h>
+#include <linux/regulator/consumer.h>
#include <linux/module.h>
#include <linux/kernel.h>
#include <linux/of.h>

@@ -25,8 +26,10 @@

struct pwm_beeper {
struct input_dev *input;
struct pwm_device *pwm;
+	struct regulator *reg;
struct work_struct work;
unsigned long period;
+	bool reg_enabled;
};

#define HZ_TO_NANOSECONDS(x) (1000000000UL/(x))

@@ -38,8 +41,20 @@ static void __pwm_beeper_set(struct pwm_beeper *beeper)

if (period) {
	pwm_config(beeper->pwm, period / 2, period);
	pwm_enable(beeper->pwm);
-	} else
+		if (beeper->reg) {
+			int error;
+
+			error = regulator_enable(beeper->reg);
+			if (!error)
+				beeper->reg_enabled = true;
+		}
+	} else {
+		if (beeper->reg_enabled) {
+			regulator_disable(beeper->reg);
+			beeper->reg_enabled = false;
+		}
	pwm_disable(beeper->pwm);
+	}
}

static void pwm_beeper_work(struct work_struct *work)

@@ -82,6 +97,10 @@ static void pwm_beeper_stop(struct pwm_beeper *beeper)

{
cancel_work_sync(&beeper->work);

+	if (beeper->reg_enabled) {
+		regulator_disable(beeper->reg);
+		beeper->reg_enabled = false;
+	}
if (beeper->period)
	pwm_disable(beeper->pwm);
}

@@ -111,6 +130,14 @@ static int pwm_beeper_probe(struct platform_device *pdev)

	return error;
}

+	beeper->reg = devm_regulator_get_optional(&pdev->dev, "amp");

If you do not use optional regulator then you will not have to check if
you have it or not everywhere: regulator core will give you a dummy that
you can toggle to your heart's content.

Some months ago, I learned that if you are not using device tree and
you do not call regulator_has_full_constraints(), then you do not
get a dummy regulator. And here, we are only checking if the
regulator exists in one place. We will still need the checks for
beeper->reg_enabled to keep calls to regulator_enable() and
regulator_disable() balanced.

Why? You do not have checks for calls to pwm_enable() and pwm_disable(),
(or rather beeper->period is used as such flag) why regulator would be
any different?

regulator_enable() has a __must_check attribute on it, so we get
compiler warnings if we do not check the return value. Also, if
enabling the regulator fails and returns an error, then calling
regulator_disable() later would cause an imbalance.

pwm_enable() and pwm_disable() work differently because they don't
count how many times they have been called. regulator_enable() and
regulator_disable(), on the other hand, work like reference
counting.

Ah, you are right, but it is more than that. It is possible to receive
multiple SND_BELL/SND_TONE events with non-0 value. You need to check if
regulator is already enabled before trying to enable it second time, or
your counting will be off.

Thanks.

-- 
Dmitry