diff mbox series

[V3,2/2] pwm: Add GPIO PWM driver

Message ID 20240128163630.104725-3-wahrenst@gmx.net
State Superseded
Headers show
Series pwm: Add GPIO PWM driver | expand

Commit Message

Stefan Wahren Jan. 28, 2024, 4:36 p.m. UTC
From: Vincent Whitchurch <vincent.whitchurch@axis.com>

Add a software PWM which toggles a GPIO from a high-resolution timer.

This will naturally not be as accurate or as efficient as a hardware
PWM, but it is useful in some cases.  I have for example used it for
evaluating LED brightness handling (via leds-pwm) on a board where the
LED was just hooked up to a GPIO, and for a simple verification of the
timer frequency on another platform.

Since high-resolution timers are used, sleeping gpio chips are not
supported and are rejected in the probe function.

Signed-off-by: Vincent Whitchurch <vincent.whitchurch@axis.com>
Co-developed-by: Stefan Wahren <wahrenst@gmx.net>
Signed-off-by: Stefan Wahren <wahrenst@gmx.net>
---
 drivers/pwm/Kconfig    |  11 ++
 drivers/pwm/Makefile   |   1 +
 drivers/pwm/pwm-gpio.c | 221 +++++++++++++++++++++++++++++++++++++++++
 3 files changed, 233 insertions(+)
 create mode 100644 drivers/pwm/pwm-gpio.c

--
2.34.1

Comments

Krzysztof Kozlowski Jan. 29, 2024, 9:40 a.m. UTC | #1
On 28/01/2024 17:36, Stefan Wahren wrote:
> From: Vincent Whitchurch <vincent.whitchurch@axis.com>
> 
> Add a software PWM which toggles a GPIO from a high-resolution timer.
> 
> This will naturally not be as accurate or as efficient as a hardware
> PWM, but it is useful in some cases.  I have for example used it for
> evaluating LED brightness handling (via leds-pwm) on a board where the
> LED was just hooked up to a GPIO, and for a simple verification of the
> timer frequency on another platform.
> 

...

> +
> +static struct platform_driver pwm_gpio_driver = {
> +	.driver = {
> +		.name = "pwm-gpio",
> +		.of_match_table = pwm_gpio_dt_ids,
> +	},
> +	.probe = pwm_gpio_probe,
> +	.remove_new = pwm_gpio_remove,
> +};
> +module_platform_driver(pwm_gpio_driver);
> +
> +MODULE_DESCRIPTION("PWM GPIO driver");
> +MODULE_ALIAS("platform:pwm-gpio");

You should not need MODULE_ALIAS() in normal cases. If you need it,
usually it means your device ID table is wrong (e.g. misses either
entries or MODULE_DEVICE_TABLE()). MODULE_ALIAS() is not a substitute
for incomplete ID table.


Best regards,
Krzysztof
Linus Walleij Jan. 29, 2024, 1:55 p.m. UTC | #2
On Mon, Jan 29, 2024 at 10:40 AM Krzysztof Kozlowski <krzk@kernel.org> wrote:
> On 28/01/2024 17:36, Stefan Wahren wrote:

> > +MODULE_ALIAS("platform:pwm-gpio");
>
> You should not need MODULE_ALIAS() in normal cases. If you need it,
> usually it means your device ID table is wrong (e.g. misses either
> entries or MODULE_DEVICE_TABLE()). MODULE_ALIAS() is not a substitute
> for incomplete ID table.

I was under the impression that MODULE_ALIAS
was something people put in to make platforms using
udev behave better by default, e.g. autoprobing these
devices better (on coldplug for "platform:*").

For example:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=4f46d6e7e5ffbce0ee1d1a80767fdf45e56cc863

But I might have got it all wrong :/

Yours,
Linus Walleij
Krzysztof Kozlowski Jan. 30, 2024, 8:36 a.m. UTC | #3
On 29/01/2024 14:55, Linus Walleij wrote:
> On Mon, Jan 29, 2024 at 10:40 AM Krzysztof Kozlowski <krzk@kernel.org> wrote:
>> On 28/01/2024 17:36, Stefan Wahren wrote:
> 
>>> +MODULE_ALIAS("platform:pwm-gpio");
>>
>> You should not need MODULE_ALIAS() in normal cases. If you need it,
>> usually it means your device ID table is wrong (e.g. misses either
>> entries or MODULE_DEVICE_TABLE()). MODULE_ALIAS() is not a substitute
>> for incomplete ID table.
> 
> I was under the impression that MODULE_ALIAS
> was something people put in to make platforms using
> udev behave better by default, e.g. autoprobing these
> devices better (on coldplug for "platform:*").
> 
> For example:
> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=4f46d6e7e5ffbce0ee1d1a80767fdf45e56cc863
> 
> But I might have got it all wrong :/

That was 2008 :), but answering to the problem: people put MODULE_ALIAS
because they (do not have / forgot / don't want to add)* proper id table
for platform devices with MODULE_DEVICE_TABLE.

While it works, it does not offer the same scalability as proper ID
table and is confusing, because it suggests that driver should be
autoloaded by a platform ID, but not matched/bound by platform ID! Such
situation is very rare and would deserve a comment anyway.

* choose appropriate

See also:
https://lore.kernel.org/all/?q=f%3Akozlowski+drop+unneeded+module_alias


Best regards,
Krzysztof
Sean Young Jan. 30, 2024, 9:08 a.m. UTC | #4
On Sun, Jan 28, 2024 at 05:36:30PM +0100, Stefan Wahren wrote:
> From: Vincent Whitchurch <vincent.whitchurch@axis.com>
> 
> Add a software PWM which toggles a GPIO from a high-resolution timer.
> 
> This will naturally not be as accurate or as efficient as a hardware
> PWM, but it is useful in some cases.  I have for example used it for
> evaluating LED brightness handling (via leds-pwm) on a board where the
> LED was just hooked up to a GPIO, and for a simple verification of the
> timer frequency on another platform.
> 
> Since high-resolution timers are used, sleeping gpio chips are not
> supported and are rejected in the probe function.
> 
> Signed-off-by: Vincent Whitchurch <vincent.whitchurch@axis.com>
> Co-developed-by: Stefan Wahren <wahrenst@gmx.net>
> Signed-off-by: Stefan Wahren <wahrenst@gmx.net>
> ---
>  drivers/pwm/Kconfig    |  11 ++
>  drivers/pwm/Makefile   |   1 +
>  drivers/pwm/pwm-gpio.c | 221 +++++++++++++++++++++++++++++++++++++++++
>  3 files changed, 233 insertions(+)
>  create mode 100644 drivers/pwm/pwm-gpio.c
> 
> diff --git a/drivers/pwm/Kconfig b/drivers/pwm/Kconfig
> index 4b956d661755..7cfda2cde130 100644
> --- a/drivers/pwm/Kconfig
> +++ b/drivers/pwm/Kconfig
> @@ -227,6 +227,17 @@ config PWM_FSL_FTM
>  	  To compile this driver as a module, choose M here: the module
>  	  will be called pwm-fsl-ftm.
> 
> +config PWM_GPIO
> +	tristate "GPIO PWM support"
> +	depends on GPIOLIB
> +	depends on HIGH_RES_TIMERS
> +	help
> +	  Generic PWM framework driver for a software PWM toggling a GPIO pin
> +	  from kernel high-resolution timers.
> +
> +	  To compile this driver as a module, choose M here: the module
> +	  will be called pwm-gpio.
> +
>  config PWM_HIBVT
>  	tristate "HiSilicon BVT PWM support"
>  	depends on ARCH_HISI || COMPILE_TEST
> diff --git a/drivers/pwm/Makefile b/drivers/pwm/Makefile
> index c5ec9e168ee7..59d1a46bb1af 100644
> --- a/drivers/pwm/Makefile
> +++ b/drivers/pwm/Makefile
> @@ -19,6 +19,7 @@ obj-$(CONFIG_PWM_DWC_CORE)	+= pwm-dwc-core.o
>  obj-$(CONFIG_PWM_DWC)		+= pwm-dwc.o
>  obj-$(CONFIG_PWM_EP93XX)	+= pwm-ep93xx.o
>  obj-$(CONFIG_PWM_FSL_FTM)	+= pwm-fsl-ftm.o
> +obj-$(CONFIG_PWM_GPIO)		+= pwm-gpio.o
>  obj-$(CONFIG_PWM_HIBVT)		+= pwm-hibvt.o
>  obj-$(CONFIG_PWM_IMG)		+= pwm-img.o
>  obj-$(CONFIG_PWM_IMX1)		+= pwm-imx1.o
> diff --git a/drivers/pwm/pwm-gpio.c b/drivers/pwm/pwm-gpio.c
> new file mode 100644
> index 000000000000..14637abe465b
> --- /dev/null
> +++ b/drivers/pwm/pwm-gpio.c
> @@ -0,0 +1,221 @@
> +// SPDX-License-Identifier: GPL-2.0-only
> +/*
> + * Generic software PWM for modulating GPIOs
> + *
> + * Copyright (C) 2020 Axis Communications AB
> + * Copyright (C) 2020 Nicola Di Lieto
> + * Copyright (C) 2024 Stefan Wahren
> + */
> +
> +#include <linux/device.h>
> +#include <linux/err.h>
> +#include <linux/gpio/consumer.h>
> +#include <linux/hrtimer.h>
> +#include <linux/module.h>
> +#include <linux/mod_devicetable.h>
> +#include <linux/platform_device.h>
> +#include <linux/pwm.h>
> +#include <linux/spinlock.h>
> +
> +struct pwm_gpio {
> +	struct pwm_chip chip;
> +	struct hrtimer gpio_timer;
> +	struct gpio_desc *gpio;
> +	struct pwm_state state;
> +	struct pwm_state next_state;
> +
> +	/* Protect internal state between pwm_ops and hrtimer */
> +	spinlock_t lock;
> +
> +	bool changing;
> +	bool running;
> +	bool level;
> +};
> +
> +static unsigned long pwm_gpio_toggle(struct pwm_gpio *gpwm, bool level)
> +{
> +	const struct pwm_state *state = &gpwm->state;
> +	bool invert = state->polarity == PWM_POLARITY_INVERSED;
> +
> +	gpwm->level = level;
> +	gpiod_set_value(gpwm->gpio, gpwm->level ^ invert);
> +
> +	if (!state->duty_cycle || state->duty_cycle == state->period) {
> +		gpwm->running = false;
> +		return 0;
> +	}
> +
> +	gpwm->running = true;
> +	return level ? state->duty_cycle : state->period - state->duty_cycle;
> +}
> +
> +static enum hrtimer_restart pwm_gpio_timer(struct hrtimer *gpio_timer)
> +{
> +	struct pwm_gpio *gpwm = container_of(gpio_timer, struct pwm_gpio,
> +					     gpio_timer);
> +	unsigned long next_toggle;
> +	unsigned long flags;
> +	bool new_level;
> +
> +	spin_lock_irqsave(&gpwm->lock, flags);
> +
> +	/* Apply new state at end of current period */
> +	if (!gpwm->level && gpwm->changing) {
> +		gpwm->changing = false;
> +		gpwm->state = gpwm->next_state;
> +		new_level = !!gpwm->state.duty_cycle;
> +	} else {
> +		new_level = !gpwm->level;
> +	}
> +
> +	next_toggle = pwm_gpio_toggle(gpwm, new_level);
> +	if (next_toggle) {
> +		hrtimer_forward(gpio_timer, hrtimer_get_expires(gpio_timer),
> +				ns_to_ktime(next_toggle));
> +	}
> +
> +	spin_unlock_irqrestore(&gpwm->lock, flags);
> +
> +	return next_toggle ? HRTIMER_RESTART : HRTIMER_NORESTART;
> +}
> +
> +static int pwm_gpio_apply(struct pwm_chip *chip, struct pwm_device *pwm,
> +			  const struct pwm_state *state)
> +{
> +	struct pwm_gpio *gpwm = container_of(chip, struct pwm_gpio, chip);
> +	bool invert = state->polarity == PWM_POLARITY_INVERSED;
> +	unsigned long flags;

Not sure this is necessary but how about:

	if (state->duty_cycle < hrtimer_resolution ||
	    state->period - state->duty_cycle < hrtimer_resolution)
		return -EINVAL;

> +
> +	if (!state->enabled) {
> +		hrtimer_cancel(&gpwm->gpio_timer);
> +	} else if (!gpwm->running) {
> +		/*
> +		 * This just enables the output, but pwm_gpio_toggle()
> +		 * really starts the duty cycle.
> +		 */
> +		int ret = gpiod_direction_output(gpwm->gpio, invert);
> +
> +		if (ret)
> +			return ret;
> +	}
> +
> +	spin_lock_irqsave(&gpwm->lock, flags);
> +
> +	if (!state->enabled) {
> +		gpwm->state = *state;
> +		gpwm->running = false;
> +		gpwm->changing = false;
> +
> +		gpiod_set_value(gpwm->gpio, invert);
> +	} else if (gpwm->running) {
> +		gpwm->next_state = *state;
> +		gpwm->changing = true;
> +	} else {
> +		unsigned long next_toggle;
> +
> +		gpwm->state = *state;
> +		gpwm->changing = false;
> +
> +		next_toggle = pwm_gpio_toggle(gpwm, !!state->duty_cycle);
> +		if (next_toggle) {
> +			hrtimer_start(&gpwm->gpio_timer, next_toggle,
> +				      HRTIMER_MODE_REL);
> +		}
> +	}
> +
> +	spin_unlock_irqrestore(&gpwm->lock, flags);
> +
> +	return 0;
> +}
> +
> +static int pwm_gpio_get_state(struct pwm_chip *chip, struct pwm_device *pwm,
> +			       struct pwm_state *state)
> +{
> +	struct pwm_gpio *gpwm = container_of(chip, struct pwm_gpio, chip);
> +	unsigned long flags;
> +
> +	spin_lock_irqsave(&gpwm->lock, flags);
> +
> +	if (gpwm->changing)
> +		*state = gpwm->next_state;
> +	else
> +		*state = gpwm->state;
> +
> +	spin_unlock_irqrestore(&gpwm->lock, flags);
> +
> +	return 0;
> +}
> +
> +static const struct pwm_ops pwm_gpio_ops = {
> +	.apply = pwm_gpio_apply,
> +	.get_state = pwm_gpio_get_state,
> +};
> +
> +static int pwm_gpio_probe(struct platform_device *pdev)
> +{
> +	struct device *dev = &pdev->dev;
> +	struct pwm_gpio *gpwm;
> +	int ret;
> +
> +	gpwm = devm_kzalloc(dev, sizeof(*gpwm), GFP_KERNEL);
> +	if (!gpwm)
> +		return -ENOMEM;
> +
> +	spin_lock_init(&gpwm->lock);
> +
> +	gpwm->gpio = devm_gpiod_get(dev, NULL, GPIOD_ASIS);
> +	if (IS_ERR(gpwm->gpio)) {
> +		return dev_err_probe(dev, PTR_ERR(gpwm->gpio),
> +				     "could not get gpio\n");
> +	}
> +
> +	if (gpiod_cansleep(gpwm->gpio)) {
> +		return dev_err_probe(dev, -EINVAL,
> +				     "sleeping GPIO %d not supported\n",
> +				     desc_to_gpio(gpwm->gpio));
> +	}
> +
> +	gpwm->chip.dev = dev;
> +	gpwm->chip.ops = &pwm_gpio_ops;
> +	gpwm->chip.npwm = 1;

	gpwm->chip.atomic = true; ?

> +
> +	hrtimer_init(&gpwm->gpio_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
> +	gpwm->gpio_timer.function = pwm_gpio_timer;
> +
> +	ret = pwmchip_add(&gpwm->chip);
> +	if (ret < 0)
> +		return dev_err_probe(dev, ret, "could not add pwmchip\n");
> +
> +	platform_set_drvdata(pdev, gpwm);
> +
> +	return 0;
> +}
> +
> +static void pwm_gpio_remove(struct platform_device *pdev)
> +{
> +	struct pwm_gpio *gpwm = platform_get_drvdata(pdev);
> +
> +	pwmchip_remove(&gpwm->chip);
> +	hrtimer_cancel(&gpwm->gpio_timer);
> +}
> +
> +static const struct of_device_id pwm_gpio_dt_ids[] = {
> +	{ .compatible = "pwm-gpio" },
> +	{ /* sentinel */ }
> +};
> +MODULE_DEVICE_TABLE(of, pwm_gpio_dt_ids);
> +
> +static struct platform_driver pwm_gpio_driver = {
> +	.driver = {
> +		.name = "pwm-gpio",
> +		.of_match_table = pwm_gpio_dt_ids,
> +	},
> +	.probe = pwm_gpio_probe,
> +	.remove_new = pwm_gpio_remove,
> +};
> +module_platform_driver(pwm_gpio_driver);
> +
> +MODULE_DESCRIPTION("PWM GPIO driver");
> +MODULE_ALIAS("platform:pwm-gpio");
> +MODULE_AUTHOR("Vincent Whitchurch");
> +MODULE_LICENSE("GPL");
> --
> 2.34.1
>
Stefan Wahren Jan. 30, 2024, 11:21 a.m. UTC | #5
Hi Sean,

Am 30.01.24 um 10:08 schrieb Sean Young:
> On Sun, Jan 28, 2024 at 05:36:30PM +0100, Stefan Wahren wrote:
>> From: Vincent Whitchurch <vincent.whitchurch@axis.com>
>>
>> Add a software PWM which toggles a GPIO from a high-resolution timer.
>>
>> This will naturally not be as accurate or as efficient as a hardware
>> PWM, but it is useful in some cases.  I have for example used it for
>> evaluating LED brightness handling (via leds-pwm) on a board where the
>> LED was just hooked up to a GPIO, and for a simple verification of the
>> timer frequency on another platform.
>>
>> Since high-resolution timers are used, sleeping gpio chips are not
>> supported and are rejected in the probe function.
>>
>> Signed-off-by: Vincent Whitchurch <vincent.whitchurch@axis.com>
>> Co-developed-by: Stefan Wahren <wahrenst@gmx.net>
>> Signed-off-by: Stefan Wahren <wahrenst@gmx.net>
>> ---
...
>> +
>> +static int pwm_gpio_apply(struct pwm_chip *chip, struct pwm_device *pwm,
>> +			  const struct pwm_state *state)
>> +{
>> +	struct pwm_gpio *gpwm = container_of(chip, struct pwm_gpio, chip);
>> +	bool invert = state->polarity == PWM_POLARITY_INVERSED;
>> +	unsigned long flags;
> Not sure this is necessary but how about:
>
> 	if (state->duty_cycle < hrtimer_resolution ||
> 	    state->period - state->duty_cycle < hrtimer_resolution)
> 		return -EINVAL;
>
i think i get the idea, but we need to care about corner cases like
duty_cycle = 0
diff mbox series

Patch

diff --git a/drivers/pwm/Kconfig b/drivers/pwm/Kconfig
index 4b956d661755..7cfda2cde130 100644
--- a/drivers/pwm/Kconfig
+++ b/drivers/pwm/Kconfig
@@ -227,6 +227,17 @@  config PWM_FSL_FTM
 	  To compile this driver as a module, choose M here: the module
 	  will be called pwm-fsl-ftm.

+config PWM_GPIO
+	tristate "GPIO PWM support"
+	depends on GPIOLIB
+	depends on HIGH_RES_TIMERS
+	help
+	  Generic PWM framework driver for a software PWM toggling a GPIO pin
+	  from kernel high-resolution timers.
+
+	  To compile this driver as a module, choose M here: the module
+	  will be called pwm-gpio.
+
 config PWM_HIBVT
 	tristate "HiSilicon BVT PWM support"
 	depends on ARCH_HISI || COMPILE_TEST
diff --git a/drivers/pwm/Makefile b/drivers/pwm/Makefile
index c5ec9e168ee7..59d1a46bb1af 100644
--- a/drivers/pwm/Makefile
+++ b/drivers/pwm/Makefile
@@ -19,6 +19,7 @@  obj-$(CONFIG_PWM_DWC_CORE)	+= pwm-dwc-core.o
 obj-$(CONFIG_PWM_DWC)		+= pwm-dwc.o
 obj-$(CONFIG_PWM_EP93XX)	+= pwm-ep93xx.o
 obj-$(CONFIG_PWM_FSL_FTM)	+= pwm-fsl-ftm.o
+obj-$(CONFIG_PWM_GPIO)		+= pwm-gpio.o
 obj-$(CONFIG_PWM_HIBVT)		+= pwm-hibvt.o
 obj-$(CONFIG_PWM_IMG)		+= pwm-img.o
 obj-$(CONFIG_PWM_IMX1)		+= pwm-imx1.o
diff --git a/drivers/pwm/pwm-gpio.c b/drivers/pwm/pwm-gpio.c
new file mode 100644
index 000000000000..14637abe465b
--- /dev/null
+++ b/drivers/pwm/pwm-gpio.c
@@ -0,0 +1,221 @@ 
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Generic software PWM for modulating GPIOs
+ *
+ * Copyright (C) 2020 Axis Communications AB
+ * Copyright (C) 2020 Nicola Di Lieto
+ * Copyright (C) 2024 Stefan Wahren
+ */
+
+#include <linux/device.h>
+#include <linux/err.h>
+#include <linux/gpio/consumer.h>
+#include <linux/hrtimer.h>
+#include <linux/module.h>
+#include <linux/mod_devicetable.h>
+#include <linux/platform_device.h>
+#include <linux/pwm.h>
+#include <linux/spinlock.h>
+
+struct pwm_gpio {
+	struct pwm_chip chip;
+	struct hrtimer gpio_timer;
+	struct gpio_desc *gpio;
+	struct pwm_state state;
+	struct pwm_state next_state;
+
+	/* Protect internal state between pwm_ops and hrtimer */
+	spinlock_t lock;
+
+	bool changing;
+	bool running;
+	bool level;
+};
+
+static unsigned long pwm_gpio_toggle(struct pwm_gpio *gpwm, bool level)
+{
+	const struct pwm_state *state = &gpwm->state;
+	bool invert = state->polarity == PWM_POLARITY_INVERSED;
+
+	gpwm->level = level;
+	gpiod_set_value(gpwm->gpio, gpwm->level ^ invert);
+
+	if (!state->duty_cycle || state->duty_cycle == state->period) {
+		gpwm->running = false;
+		return 0;
+	}
+
+	gpwm->running = true;
+	return level ? state->duty_cycle : state->period - state->duty_cycle;
+}
+
+static enum hrtimer_restart pwm_gpio_timer(struct hrtimer *gpio_timer)
+{
+	struct pwm_gpio *gpwm = container_of(gpio_timer, struct pwm_gpio,
+					     gpio_timer);
+	unsigned long next_toggle;
+	unsigned long flags;
+	bool new_level;
+
+	spin_lock_irqsave(&gpwm->lock, flags);
+
+	/* Apply new state at end of current period */
+	if (!gpwm->level && gpwm->changing) {
+		gpwm->changing = false;
+		gpwm->state = gpwm->next_state;
+		new_level = !!gpwm->state.duty_cycle;
+	} else {
+		new_level = !gpwm->level;
+	}
+
+	next_toggle = pwm_gpio_toggle(gpwm, new_level);
+	if (next_toggle) {
+		hrtimer_forward(gpio_timer, hrtimer_get_expires(gpio_timer),
+				ns_to_ktime(next_toggle));
+	}
+
+	spin_unlock_irqrestore(&gpwm->lock, flags);
+
+	return next_toggle ? HRTIMER_RESTART : HRTIMER_NORESTART;
+}
+
+static int pwm_gpio_apply(struct pwm_chip *chip, struct pwm_device *pwm,
+			  const struct pwm_state *state)
+{
+	struct pwm_gpio *gpwm = container_of(chip, struct pwm_gpio, chip);
+	bool invert = state->polarity == PWM_POLARITY_INVERSED;
+	unsigned long flags;
+
+	if (!state->enabled) {
+		hrtimer_cancel(&gpwm->gpio_timer);
+	} else if (!gpwm->running) {
+		/*
+		 * This just enables the output, but pwm_gpio_toggle()
+		 * really starts the duty cycle.
+		 */
+		int ret = gpiod_direction_output(gpwm->gpio, invert);
+
+		if (ret)
+			return ret;
+	}
+
+	spin_lock_irqsave(&gpwm->lock, flags);
+
+	if (!state->enabled) {
+		gpwm->state = *state;
+		gpwm->running = false;
+		gpwm->changing = false;
+
+		gpiod_set_value(gpwm->gpio, invert);
+	} else if (gpwm->running) {
+		gpwm->next_state = *state;
+		gpwm->changing = true;
+	} else {
+		unsigned long next_toggle;
+
+		gpwm->state = *state;
+		gpwm->changing = false;
+
+		next_toggle = pwm_gpio_toggle(gpwm, !!state->duty_cycle);
+		if (next_toggle) {
+			hrtimer_start(&gpwm->gpio_timer, next_toggle,
+				      HRTIMER_MODE_REL);
+		}
+	}
+
+	spin_unlock_irqrestore(&gpwm->lock, flags);
+
+	return 0;
+}
+
+static int pwm_gpio_get_state(struct pwm_chip *chip, struct pwm_device *pwm,
+			       struct pwm_state *state)
+{
+	struct pwm_gpio *gpwm = container_of(chip, struct pwm_gpio, chip);
+	unsigned long flags;
+
+	spin_lock_irqsave(&gpwm->lock, flags);
+
+	if (gpwm->changing)
+		*state = gpwm->next_state;
+	else
+		*state = gpwm->state;
+
+	spin_unlock_irqrestore(&gpwm->lock, flags);
+
+	return 0;
+}
+
+static const struct pwm_ops pwm_gpio_ops = {
+	.apply = pwm_gpio_apply,
+	.get_state = pwm_gpio_get_state,
+};
+
+static int pwm_gpio_probe(struct platform_device *pdev)
+{
+	struct device *dev = &pdev->dev;
+	struct pwm_gpio *gpwm;
+	int ret;
+
+	gpwm = devm_kzalloc(dev, sizeof(*gpwm), GFP_KERNEL);
+	if (!gpwm)
+		return -ENOMEM;
+
+	spin_lock_init(&gpwm->lock);
+
+	gpwm->gpio = devm_gpiod_get(dev, NULL, GPIOD_ASIS);
+	if (IS_ERR(gpwm->gpio)) {
+		return dev_err_probe(dev, PTR_ERR(gpwm->gpio),
+				     "could not get gpio\n");
+	}
+
+	if (gpiod_cansleep(gpwm->gpio)) {
+		return dev_err_probe(dev, -EINVAL,
+				     "sleeping GPIO %d not supported\n",
+				     desc_to_gpio(gpwm->gpio));
+	}
+
+	gpwm->chip.dev = dev;
+	gpwm->chip.ops = &pwm_gpio_ops;
+	gpwm->chip.npwm = 1;
+
+	hrtimer_init(&gpwm->gpio_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
+	gpwm->gpio_timer.function = pwm_gpio_timer;
+
+	ret = pwmchip_add(&gpwm->chip);
+	if (ret < 0)
+		return dev_err_probe(dev, ret, "could not add pwmchip\n");
+
+	platform_set_drvdata(pdev, gpwm);
+
+	return 0;
+}
+
+static void pwm_gpio_remove(struct platform_device *pdev)
+{
+	struct pwm_gpio *gpwm = platform_get_drvdata(pdev);
+
+	pwmchip_remove(&gpwm->chip);
+	hrtimer_cancel(&gpwm->gpio_timer);
+}
+
+static const struct of_device_id pwm_gpio_dt_ids[] = {
+	{ .compatible = "pwm-gpio" },
+	{ /* sentinel */ }
+};
+MODULE_DEVICE_TABLE(of, pwm_gpio_dt_ids);
+
+static struct platform_driver pwm_gpio_driver = {
+	.driver = {
+		.name = "pwm-gpio",
+		.of_match_table = pwm_gpio_dt_ids,
+	},
+	.probe = pwm_gpio_probe,
+	.remove_new = pwm_gpio_remove,
+};
+module_platform_driver(pwm_gpio_driver);
+
+MODULE_DESCRIPTION("PWM GPIO driver");
+MODULE_ALIAS("platform:pwm-gpio");
+MODULE_AUTHOR("Vincent Whitchurch");
+MODULE_LICENSE("GPL");