Merge tag 'pwm/duty_offset-for-6.13-rc1' of https://git.kernel.org/pub/scm/linux/kernel/git/ukleinek/linux

pwm: Support for duty_offset

Support a new abstraction for pwm configuration that allows to specify
the time between start of period and the raising edge of the signal
("duty offset").

This is used in a patch series by Trevor Gamblin for triggering an ADC
conversion and afterwards read out the result. See
https://lore.kernel.org/linux-iio/20240909-ad7625_r1-v5-0-60a397768b25@baylibre.com/
for more details.

3 files changed, 1201 insertions, 416 deletions

diff --git a/drivers/pwm/core.c b/drivers/pwm/core.c
index 6e752e148b98..634be56e204b 100644
--- a/drivers/pwm/core.c
+++ b/drivers/pwm/core.c
@@ -31,6 +31,397 @@ static DEFINE_MUTEX(pwm_lock);
 
 static DEFINE_IDR(pwm_chips);
 
+static void pwmchip_lock(struct pwm_chip *chip)
+{
+	if (chip->atomic)
+		spin_lock(&chip->atomic_lock);
+	else
+		mutex_lock(&chip->nonatomic_lock);
+}
+
+static void pwmchip_unlock(struct pwm_chip *chip)
+{
+	if (chip->atomic)
+		spin_unlock(&chip->atomic_lock);
+	else
+		mutex_unlock(&chip->nonatomic_lock);
+}
+
+DEFINE_GUARD(pwmchip, struct pwm_chip *, pwmchip_lock(_T), pwmchip_unlock(_T))
+
+static bool pwm_wf_valid(const struct pwm_waveform *wf)
+{
+	/*
+	 * For now restrict waveforms to period_length_ns <= S64_MAX to provide
+	 * some space for future extensions. One possibility is to simplify
+	 * representing waveforms with inverted polarity using negative values
+	 * somehow.
+	 */
+	if (wf->period_length_ns > S64_MAX)
+		return false;
+
+	if (wf->duty_length_ns > wf->period_length_ns)
+		return false;
+
+	/*
+	 * .duty_offset_ns is supposed to be smaller than .period_length_ns, apart
+	 * from the corner case .duty_offset_ns == 0 && .period_length_ns == 0.
+	 */
+	if (wf->duty_offset_ns && wf->duty_offset_ns >= wf->period_length_ns)
+		return false;
+
+	return true;
+}
+
+static void pwm_wf2state(const struct pwm_waveform *wf, struct pwm_state *state)
+{
+	if (wf->period_length_ns) {
+		if (wf->duty_length_ns + wf->duty_offset_ns < wf->period_length_ns)
+			*state = (struct pwm_state){
+				.enabled = true,
+				.polarity = PWM_POLARITY_NORMAL,
+				.period = wf->period_length_ns,
+				.duty_cycle = wf->duty_length_ns,
+			};
+		else
+			*state = (struct pwm_state){
+				.enabled = true,
+				.polarity = PWM_POLARITY_INVERSED,
+				.period = wf->period_length_ns,
+				.duty_cycle = wf->period_length_ns - wf->duty_length_ns,
+			};
+	} else {
+		*state = (struct pwm_state){
+			.enabled = false,
+		};
+	}
+}
+
+static void pwm_state2wf(const struct pwm_state *state, struct pwm_waveform *wf)
+{
+	if (state->enabled) {
+		if (state->polarity == PWM_POLARITY_NORMAL)
+			*wf = (struct pwm_waveform){
+				.period_length_ns = state->period,
+				.duty_length_ns = state->duty_cycle,
+				.duty_offset_ns = 0,
+			};
+		else
+			*wf = (struct pwm_waveform){
+				.period_length_ns = state->period,
+				.duty_length_ns = state->period - state->duty_cycle,
+				.duty_offset_ns = state->duty_cycle,
+			};
+	} else {
+		*wf = (struct pwm_waveform){
+			.period_length_ns = 0,
+		};
+	}
+}
+
+static int pwmwfcmp(const struct pwm_waveform *a, const struct pwm_waveform *b)
+{
+	if (a->period_length_ns > b->period_length_ns)
+		return 1;
+
+	if (a->period_length_ns < b->period_length_ns)
+		return -1;
+
+	if (a->duty_length_ns > b->duty_length_ns)
+		return 1;
+
+	if (a->duty_length_ns < b->duty_length_ns)
+		return -1;
+
+	if (a->duty_offset_ns > b->duty_offset_ns)
+		return 1;
+
+	if (a->duty_offset_ns < b->duty_offset_ns)
+		return -1;
+
+	return 0;
+}
+
+static bool pwm_check_rounding(const struct pwm_waveform *wf,
+			       const struct pwm_waveform *wf_rounded)
+{
+	if (!wf->period_length_ns)
+		return true;
+
+	if (wf->period_length_ns < wf_rounded->period_length_ns)
+		return false;
+
+	if (wf->duty_length_ns < wf_rounded->duty_length_ns)
+		return false;
+
+	if (wf->duty_offset_ns < wf_rounded->duty_offset_ns)
+		return false;
+
+	return true;
+}
+
+static int __pwm_round_waveform_tohw(struct pwm_chip *chip, struct pwm_device *pwm,
+				     const struct pwm_waveform *wf, void *wfhw)
+{
+	const struct pwm_ops *ops = chip->ops;
+	int ret;
+
+	ret = ops->round_waveform_tohw(chip, pwm, wf, wfhw);
+	trace_pwm_round_waveform_tohw(pwm, wf, wfhw, ret);
+
+	return ret;
+}
+
+static int __pwm_round_waveform_fromhw(struct pwm_chip *chip, struct pwm_device *pwm,
+				       const void *wfhw, struct pwm_waveform *wf)
+{
+	const struct pwm_ops *ops = chip->ops;
+	int ret;
+
+	ret = ops->round_waveform_fromhw(chip, pwm, wfhw, wf);
+	trace_pwm_round_waveform_fromhw(pwm, wfhw, wf, ret);
+
+	return ret;
+}
+
+static int __pwm_read_waveform(struct pwm_chip *chip, struct pwm_device *pwm, void *wfhw)
+{
+	const struct pwm_ops *ops = chip->ops;
+	int ret;
+
+	ret = ops->read_waveform(chip, pwm, wfhw);
+	trace_pwm_read_waveform(pwm, wfhw, ret);
+
+	return ret;
+}
+
+static int __pwm_write_waveform(struct pwm_chip *chip, struct pwm_device *pwm, const void *wfhw)
+{
+	const struct pwm_ops *ops = chip->ops;
+	int ret;
+
+	ret = ops->write_waveform(chip, pwm, wfhw);
+	trace_pwm_write_waveform(pwm, wfhw, ret);
+
+	return ret;
+}
+
+#define WFHWSIZE 20
+
+/**
+ * pwm_round_waveform_might_sleep - Query hardware capabilities
+ * Cannot be used in atomic context.
+ * @pwm: PWM device
+ * @wf: waveform to round and output parameter
+ *
+ * Typically a given waveform cannot be implemented exactly by hardware, e.g.
+ * because hardware only supports coarse period resolution or no duty_offset.
+ * This function returns the actually implemented waveform if you pass wf to
+ * pwm_set_waveform_might_sleep now.
+ *
+ * Note however that the world doesn't stop turning when you call it, so when
+ * doing
+ *
+ * 	pwm_round_waveform_might_sleep(mypwm, &wf);
+ * 	pwm_set_waveform_might_sleep(mypwm, &wf, true);
+ *
+ * the latter might fail, e.g. because an input clock changed its rate between
+ * these two calls and the waveform determined by
+ * pwm_round_waveform_might_sleep() cannot be implemented any more.
+ *
+ * Returns 0 on success, 1 if there is no valid hardware configuration matching
+ * the input waveform under the PWM rounding rules or a negative errno.
+ */
+int pwm_round_waveform_might_sleep(struct pwm_device *pwm, struct pwm_waveform *wf)
+{
+	struct pwm_chip *chip = pwm->chip;
+	const struct pwm_ops *ops = chip->ops;
+	struct pwm_waveform wf_req = *wf;
+	char wfhw[WFHWSIZE];
+	int ret_tohw, ret_fromhw;
+
+	BUG_ON(WFHWSIZE < ops->sizeof_wfhw);
+
+	if (!pwm_wf_valid(wf))
+		return -EINVAL;
+
+	guard(pwmchip)(chip);
+
+	if (!chip->operational)
+		return -ENODEV;
+
+	ret_tohw = __pwm_round_waveform_tohw(chip, pwm, wf, wfhw);
+	if (ret_tohw < 0)
+		return ret_tohw;
+
+	if (IS_ENABLED(CONFIG_PWM_DEBUG) && ret_tohw > 1)
+		dev_err(&chip->dev, "Unexpected return value from __pwm_round_waveform_tohw: requested %llu/%llu [+%llu], return value %d\n",
+			wf_req.duty_length_ns, wf_req.period_length_ns, wf_req.duty_offset_ns, ret_tohw);
+
+	ret_fromhw = __pwm_round_waveform_fromhw(chip, pwm, wfhw, wf);
+	if (ret_fromhw < 0)
+		return ret_fromhw;
+
+	if (IS_ENABLED(CONFIG_PWM_DEBUG) && ret_fromhw > 0)
+		dev_err(&chip->dev, "Unexpected return value from __pwm_round_waveform_fromhw: requested %llu/%llu [+%llu], return value %d\n",
+			wf_req.duty_length_ns, wf_req.period_length_ns, wf_req.duty_offset_ns, ret_tohw);
+
+	if (IS_ENABLED(CONFIG_PWM_DEBUG) &&
+	    ret_tohw == 0 && !pwm_check_rounding(&wf_req, wf))
+		dev_err(&chip->dev, "Wrong rounding: requested %llu/%llu [+%llu], result %llu/%llu [+%llu]\n",
+			wf_req.duty_length_ns, wf_req.period_length_ns, wf_req.duty_offset_ns,
+			wf->duty_length_ns, wf->period_length_ns, wf->duty_offset_ns);
+
+	return ret_tohw;
+}
+EXPORT_SYMBOL_GPL(pwm_round_waveform_might_sleep);
+
+/**
+ * pwm_get_waveform_might_sleep - Query hardware about current configuration
+ * Cannot be used in atomic context.
+ * @pwm: PWM device
+ * @wf: output parameter
+ *
+ * Stores the current configuration of the PWM in @wf. Note this is the
+ * equivalent of pwm_get_state_hw() (and not pwm_get_state()) for pwm_waveform.
+ */
+int pwm_get_waveform_might_sleep(struct pwm_device *pwm, struct pwm_waveform *wf)
+{
+	struct pwm_chip *chip = pwm->chip;
+	const struct pwm_ops *ops = chip->ops;
+	char wfhw[WFHWSIZE];
+	int err;
+
+	BUG_ON(WFHWSIZE < ops->sizeof_wfhw);
+
+	guard(pwmchip)(chip);
+
+	if (!chip->operational)
+		return -ENODEV;
+
+	err = __pwm_read_waveform(chip, pwm, &wfhw);
+	if (err)
+		return err;
+
+	return __pwm_round_waveform_fromhw(chip, pwm, &wfhw, wf);
+}
+EXPORT_SYMBOL_GPL(pwm_get_waveform_might_sleep);
+
+/* Called with the pwmchip lock held */
+static int __pwm_set_waveform(struct pwm_device *pwm,
+			      const struct pwm_waveform *wf,
+			      bool exact)
+{
+	struct pwm_chip *chip = pwm->chip;
+	const struct pwm_ops *ops = chip->ops;
+	char wfhw[WFHWSIZE];
+	struct pwm_waveform wf_rounded;
+	int err;
+
+	BUG_ON(WFHWSIZE < ops->sizeof_wfhw);
+
+	if (!pwm_wf_valid(wf))
+		return -EINVAL;
+
+	err = __pwm_round_waveform_tohw(chip, pwm, wf, &wfhw);
+	if (err)
+		return err;
+
+	if ((IS_ENABLED(CONFIG_PWM_DEBUG) || exact) && wf->period_length_ns) {
+		err = __pwm_round_waveform_fromhw(chip, pwm, &wfhw, &wf_rounded);
+		if (err)
+			return err;
+
+		if (IS_ENABLED(CONFIG_PWM_DEBUG) && !pwm_check_rounding(wf, &wf_rounded))
+			dev_err(&chip->dev, "Wrong rounding: requested %llu/%llu [+%llu], result %llu/%llu [+%llu]\n",
+				wf->duty_length_ns, wf->period_length_ns, wf->duty_offset_ns,
+				wf_rounded.duty_length_ns, wf_rounded.period_length_ns, wf_rounded.duty_offset_ns);
+
+		if (exact && pwmwfcmp(wf, &wf_rounded)) {
+			dev_dbg(&chip->dev, "Requested no rounding, but %llu/%llu [+%llu] -> %llu/%llu [+%llu]\n",
+				wf->duty_length_ns, wf->period_length_ns, wf->duty_offset_ns,
+				wf_rounded.duty_length_ns, wf_rounded.period_length_ns, wf_rounded.duty_offset_ns);
+
+			return 1;
+		}
+	}
+
+	err = __pwm_write_waveform(chip, pwm, &wfhw);
+	if (err)
+		return err;
+
+	/* update .state */
+	pwm_wf2state(wf, &pwm->state);
+
+	if (IS_ENABLED(CONFIG_PWM_DEBUG) && ops->read_waveform && wf->period_length_ns) {
+		struct pwm_waveform wf_set;
+
+		err = __pwm_read_waveform(chip, pwm, &wfhw);
+		if (err)
+			/* maybe ignore? */
+			return err;
+
+		err = __pwm_round_waveform_fromhw(chip, pwm, &wfhw, &wf_set);
+		if (err)
+			/* maybe ignore? */
+			return err;
+
+		if (pwmwfcmp(&wf_set, &wf_rounded) != 0)
+			dev_err(&chip->dev,
+				"Unexpected setting: requested %llu/%llu [+%llu], expected %llu/%llu [+%llu], set %llu/%llu [+%llu]\n",
+				wf->duty_length_ns, wf->period_length_ns, wf->duty_offset_ns,
+				wf_rounded.duty_length_ns, wf_rounded.period_length_ns, wf_rounded.duty_offset_ns,
+				wf_set.duty_length_ns, wf_set.period_length_ns, wf_set.duty_offset_ns);
+	}
+	return 0;
+}
+
+/**
+ * pwm_set_waveform_might_sleep - Apply a new waveform
+ * Cannot be used in atomic context.
+ * @pwm: PWM device
+ * @wf: The waveform to apply
+ * @exact: If true no rounding is allowed
+ *
+ * Typically a requested waveform cannot be implemented exactly, e.g. because
+ * you requested .period_length_ns = 100 ns, but the hardware can only set
+ * periods that are a multiple of 8.5 ns. With that hardware passing exact =
+ * true results in pwm_set_waveform_might_sleep() failing and returning 1. If
+ * exact = false you get a period of 93.5 ns (i.e. the biggest period not bigger
+ * than the requested value).
+ * Note that even with exact = true, some rounding by less than 1 is
+ * possible/needed. In the above example requesting .period_length_ns = 94 and
+ * exact = true, you get the hardware configured with period = 93.5 ns.
+ */
+int pwm_set_waveform_might_sleep(struct pwm_device *pwm,
+				 const struct pwm_waveform *wf, bool exact)
+{
+	struct pwm_chip *chip = pwm->chip;
+	int err;
+
+	might_sleep();
+
+	guard(pwmchip)(chip);
+
+	if (!chip->operational)
+		return -ENODEV;
+
+	if (IS_ENABLED(CONFIG_PWM_DEBUG) && chip->atomic) {
+		/*
+		 * Catch any drivers that have been marked as atomic but
+		 * that will sleep anyway.
+		 */
+		non_block_start();
+		err = __pwm_set_waveform(pwm, wf, exact);
+		non_block_end();
+	} else {
+		err = __pwm_set_waveform(pwm, wf, exact);
+	}
+
+	return err;
+}
+EXPORT_SYMBOL_GPL(pwm_set_waveform_might_sleep);
+
 static void pwm_apply_debug(struct pwm_device *pwm,
 			    const struct pwm_state *state)
 {
@@ -164,6 +555,7 @@ static bool pwm_state_valid(const struct pwm_state *state)
 static int __pwm_apply(struct pwm_device *pwm, const struct pwm_state *state)
 {
 	struct pwm_chip *chip;
+	const struct pwm_ops *ops;
 	int err;
 
 	if (!pwm || !state)
@@ -187,6 +579,7 @@ static int __pwm_apply(struct pwm_device *pwm, const struct pwm_state *state)
 	}
 
 	chip = pwm->chip;
+	ops = chip->ops;
 
 	if (state->period == pwm->state.period &&
 	    state->duty_cycle == pwm->state.duty_cycle &&
@@ -195,18 +588,69 @@ static int __pwm_apply(struct pwm_device *pwm, const struct pwm_state *state)
 	    state->usage_power == pwm->state.usage_power)
 		return 0;
 
-	err = chip->ops->apply(chip, pwm, state);
-	trace_pwm_apply(pwm, state, err);
-	if (err)
-		return err;
+	if (ops->write_waveform) {
+		struct pwm_waveform wf;
+		char wfhw[WFHWSIZE];
 
-	pwm->state = *state;
+		BUG_ON(WFHWSIZE < ops->sizeof_wfhw);
 
-	/*
-	 * only do this after pwm->state was applied as some
-	 * implementations of .get_state depend on this
-	 */
-	pwm_apply_debug(pwm, state);
+		pwm_state2wf(state, &wf);
+
+		/*
+		 * The rounding is wrong here for states with inverted polarity.
+		 * While .apply() rounds down duty_cycle (which represents the
+		 * time from the start of the period to the inner edge),
+		 * .round_waveform_tohw() rounds down the time the PWM is high.
+		 * Can be fixed if the need arises, until reported otherwise
+		 * let's assume that consumers don't care.
+		 */
+
+		err = __pwm_round_waveform_tohw(chip, pwm, &wf, &wfhw);
+		if (err) {
+			if (err > 0)
+				/*
+				 * This signals an invalid request, typically
+				 * the requested period (or duty_offset) is
+				 * smaller than possible with the hardware.
+				 */
+				return -EINVAL;
+
+			return err;
+		}
+
+		if (IS_ENABLED(CONFIG_PWM_DEBUG)) {
+			struct pwm_waveform wf_rounded;
+
+			err = __pwm_round_waveform_fromhw(chip, pwm, &wfhw, &wf_rounded);
+			if (err)
+				return err;
+
+			if (!pwm_check_rounding(&wf, &wf_rounded))
+				dev_err(&chip->dev, "Wrong rounding: requested %llu/%llu [+%llu], result %llu/%llu [+%llu]\n",
+					wf.duty_length_ns, wf.period_length_ns, wf.duty_offset_ns,
+					wf_rounded.duty_length_ns, wf_rounded.period_length_ns, wf_rounded.duty_offset_ns);
+		}
+
+		err = __pwm_write_waveform(chip, pwm, &wfhw);
+		if (err)
+			return err;
+
+		pwm->state = *state;
+
+	} else {
+		err = ops->apply(chip, pwm, state);
+		trace_pwm_apply(pwm, state, err);
+		if (err)
+			return err;
+
+		pwm->state = *state;
+
+		/*
+		 * only do this after pwm->state was applied as some
+		 * implementations of .get_state() depend on this
+		 */
+		pwm_apply_debug(pwm, state);
+	}
 
 	return 0;
 }
@@ -220,6 +664,7 @@ static int __pwm_apply(struct pwm_device *pwm, const struct pwm_state *state)
 int pwm_apply_might_sleep(struct pwm_device *pwm, const struct pwm_state *state)
 {
 	int err;
+	struct pwm_chip *chip = pwm->chip;
 
 	/*
 	 * Some lowlevel driver's implementations of .apply() make use of
@@ -230,7 +675,12 @@ int pwm_apply_might_sleep(struct pwm_device *pwm, const struct pwm_state *state)
 	 */
 	might_sleep();
 
-	if (IS_ENABLED(CONFIG_PWM_DEBUG) && pwm->chip->atomic) {
+	guard(pwmchip)(chip);
+
+	if (!chip->operational)
+		return -ENODEV;
+
+	if (IS_ENABLED(CONFIG_PWM_DEBUG) && chip->atomic) {
 		/*
 		 * Catch any drivers that have been marked as atomic but
 		 * that will sleep anyway.
@@ -254,13 +704,55 @@ EXPORT_SYMBOL_GPL(pwm_apply_might_sleep);
  */
 int pwm_apply_atomic(struct pwm_device *pwm, const struct pwm_state *state)
 {
-	WARN_ONCE(!pwm->chip->atomic,
+	struct pwm_chip *chip = pwm->chip;
+
+	WARN_ONCE(!chip->atomic,
 		  "sleeping PWM driver used in atomic context\n");
 
+	guard(pwmchip)(chip);
+
+	if (!chip->operational)
+		return -ENODEV;
+
 	return __pwm_apply(pwm, state);
 }
 EXPORT_SYMBOL_GPL(pwm_apply_atomic);
 
+static int pwm_get_state_hw(struct pwm_device *pwm, struct pwm_state *state)
+{
+	struct pwm_chip *chip = pwm->chip;
+	const struct pwm_ops *ops = chip->ops;
+	int ret = -EOPNOTSUPP;
+
+	if (ops->read_waveform) {
+		char wfhw[WFHWSIZE];
+		struct pwm_waveform wf;
+
+		BUG_ON(WFHWSIZE < ops->sizeof_wfhw);
+
+		scoped_guard(pwmchip, chip) {
+
+			ret = __pwm_read_waveform(chip, pwm, &wfhw);
+			if (ret)
+				return ret;
+
+			ret = __pwm_round_waveform_fromhw(chip, pwm, &wfhw, &wf);
+			if (ret)
+				return ret;
+		}
+
+		pwm_wf2state(&wf, state);
+
+	} else if (ops->get_state) {
+		scoped_guard(pwmchip, chip)
+			ret = ops->get_state(chip, pwm, state);
+
+		trace_pwm_get(pwm, state, ret);
+	}
+
+	return ret;
+}
+
 /**
  * pwm_adjust_config() - adjust the current PWM config to the PWM arguments
  * @pwm: PWM device
@@ -334,8 +826,18 @@ static int pwm_capture(struct pwm_device *pwm, struct pwm_capture *result,
 	if (!ops->capture)
 		return -ENOSYS;
 
+	/*
+	 * Holding the pwm_lock is probably not needed. If you use pwm_capture()
+	 * and you're interested to speed it up, please convince yourself it's
+	 * really not needed, test and then suggest a patch on the mailing list.
+	 */
 	guard(mutex)(&pwm_lock);
 
+	guard(pwmchip)(chip);
+
+	if (!chip->operational)
+		return -ENODEV;
+
 	return ops->capture(chip, pwm, result, timeout);
 }
 
@@ -368,6 +870,14 @@ static int pwm_device_request(struct pwm_device *pwm, const char *label)
 	if (test_bit(PWMF_REQUESTED, &pwm->flags))
 		return -EBUSY;
 
+	/*
+	 * This function is called while holding pwm_lock. As .operational only
+	 * changes while holding this lock, checking it here without holding the
+	 * chip lock is fine.
+	 */
+	if (!chip->operational)
+		return -ENODEV;
+
 	if (!try_module_get(chip->owner))
 		return -ENODEV;
 
@@ -386,7 +896,7 @@ err_get_device:
 		}
 	}
 
-	if (ops->get_state) {
+	if (ops->read_waveform || ops->get_state) {
 		/*
 		 * Zero-initialize state because most drivers are unaware of
 		 * .usage_power. The other members of state are supposed to be
@@ -396,9 +906,7 @@ err_get_device:
 		 */
 		struct pwm_state state = { 0, };
 
-		err = ops->get_state(chip, pwm, &state);
-		trace_pwm_get(pwm, &state, err);
-
+		err = pwm_get_state_hw(pwm, &state);
 		if (!err)
 			pwm->state = state;
 
@@ -1020,6 +1528,7 @@ struct pwm_chip *pwmchip_alloc(struct device *parent, unsigned int npwm, size_t
 
 	chip->npwm = npwm;
 	chip->uses_pwmchip_alloc = true;
+	chip->operational = false;
 
 	pwmchip_dev = &chip->dev;
 	device_initialize(pwmchip_dev);
@@ -1084,114 +1593,27 @@ static bool pwm_ops_check(const struct pwm_chip *chip)
 {
 	const struct pwm_ops *ops = chip->ops;
 
-	if (!ops->apply)
-		return false;
-
-	if (IS_ENABLED(CONFIG_PWM_DEBUG) && !ops->get_state)
-		dev_warn(pwmchip_parent(chip),
-			 "Please implement the .get_state() callback\n");
-
-	return true;
-}
-
-/**
- * __pwmchip_add() - register a new PWM chip
- * @chip: the PWM chip to add
- * @owner: reference to the module providing the chip.
- *
- * Register a new PWM chip. @owner is supposed to be THIS_MODULE, use the
- * pwmchip_add wrapper to do this right.
- *
- * Returns: 0 on success or a negative error code on failure.
- */
-int __pwmchip_add(struct pwm_chip *chip, struct module *owner)
-{
-	int ret;
-
-	if (!chip || !pwmchip_parent(chip) || !chip->ops || !chip->npwm)
-		return -EINVAL;
-
-	/*
-	 * a struct pwm_chip must be allocated using (devm_)pwmchip_alloc,
-	 * otherwise the embedded struct device might disappear too early
-	 * resulting in memory corruption.
-	 * Catch drivers that were not converted appropriately.
-	 */
-	if (!chip->uses_pwmchip_alloc)
-		return -EINVAL;
-
-	if (!pwm_ops_check(chip))
-		return -EINVAL;
-
-	chip->owner = owner;
-
-	guard(mutex)(&pwm_lock);
-
-	ret = idr_alloc(&pwm_chips, chip, 0, 0, GFP_KERNEL);
-	if (ret < 0)
-		return ret;
-
-	chip->id = ret;
-
-	dev_set_name(&chip->dev, "pwmchip%u", chip->id);
+	if (ops->write_waveform) {
+		if (!ops->round_waveform_tohw ||
+		    !ops->round_waveform_fromhw ||
+		    !ops->write_waveform)
+			return false;
 
-	if (IS_ENABLED(CONFIG_OF))
-		of_pwmchip_add(chip);
-
-	ret = device_add(&chip->dev);
-	if (ret)
-		goto err_device_add;
-
-	return 0;
-
-err_device_add:
-	if (IS_ENABLED(CONFIG_OF))
-		of_pwmchip_remove(chip);
-
-	idr_remove(&pwm_chips, chip->id);
-
-	return ret;
-}
-EXPORT_SYMBOL_GPL(__pwmchip_add);
-
-/**
- * pwmchip_remove() - remove a PWM chip
- * @chip: the PWM chip to remove
- *
- * Removes a PWM chip.
- */
-void pwmchip_remove(struct pwm_chip *chip)
-{
-	pwmchip_sysfs_unexport(chip);
-
-	if (IS_ENABLED(CONFIG_OF))
-		of_pwmchip_remove(chip);
-
-	scoped_guard(mutex, &pwm_lock)
-		idr_remove(&pwm_chips, chip->id);
-
-	device_del(&chip->dev);
-}
-EXPORT_SYMBOL_GPL(pwmchip_remove);
-
-static void devm_pwmchip_remove(void *data)
-{
-	struct pwm_chip *chip = data;
-
-	pwmchip_remove(chip);
-}
-
-int __devm_pwmchip_add(struct device *dev, struct pwm_chip *chip, struct module *owner)
-{
-	int ret;
+		if (WFHWSIZE < ops->sizeof_wfhw) {
+			dev_warn(pwmchip_parent(chip), "WFHWSIZE < %zu\n", ops->sizeof_wfhw);
+			return false;
+		}
+	} else {
+		if (!ops->apply)
+			return false;
 
-	ret = __pwmchip_add(chip, owner);
-	if (ret)
-		return ret;
+		if (IS_ENABLED(CONFIG_PWM_DEBUG) && !ops->get_state)
+			dev_warn(pwmchip_parent(chip),
+				 "Please implement the .get_state() callback\n");
+	}
 
-	return devm_add_action_or_reset(dev, devm_pwmchip_remove, chip);
+	return true;
 }
-EXPORT_SYMBOL_GPL(__devm_pwmchip_add);
 
 static struct device_link *pwm_device_link_add(struct device *dev,
 					       struct pwm_device *pwm)
@@ -1371,36 +1793,6 @@ static DEFINE_MUTEX(pwm_lookup_lock);
 static LIST_HEAD(pwm_lookup_list);
 
 /**
- * pwm_add_table() - register PWM device consumers
- * @table: array of consumers to register
- * @num: number of consumers in table
- */
-void pwm_add_table(struct pwm_lookup *table, size_t num)
-{
-	guard(mutex)(&pwm_lookup_lock);
-
-	while (num--) {
-		list_add_tail(&table->list, &pwm_lookup_list);
-		table++;
-	}
-}
-
-/**
- * pwm_remove_table() - unregister PWM device consumers
- * @table: array of consumers to unregister
- * @num: number of consumers in table
- */
-void pwm_remove_table(struct pwm_lookup *table, size_t num)
-{
-	guard(mutex)(&pwm_lookup_lock);
-
-	while (num--) {
-		list_del(&table->list);
-		table++;
-	}
-}
-
-/**
  * pwm_get() - look up and request a PWM device
  * @dev: device for PWM consumer
  * @con_id: consumer name
@@ -1538,12 +1930,17 @@ void pwm_put(struct pwm_device *pwm)
 
 	guard(mutex)(&pwm_lock);
 
-	if (!test_and_clear_bit(PWMF_REQUESTED, &pwm->flags)) {
+	/*
+	 * Trigger a warning if a consumer called pwm_put() twice.
+	 * If the chip isn't operational, PWMF_REQUESTED was already cleared in
+	 * pwmchip_remove(). So don't warn in this case.
+	 */
+	if (chip->operational && !test_and_clear_bit(PWMF_REQUESTED, &pwm->flags)) {
 		pr_warn("PWM device already freed\n");
 		return;
 	}
 
-	if (chip->ops->free)
+	if (chip->operational && chip->ops->free)
 		pwm->chip->ops->free(pwm->chip, pwm);
 
 	pwm->label = NULL;
@@ -1621,6 +2018,162 @@ struct pwm_device *devm_fwnode_pwm_get(struct device *dev,
 }
 EXPORT_SYMBOL_GPL(devm_fwnode_pwm_get);
 
+/**
+ * __pwmchip_add() - register a new PWM chip
+ * @chip: the PWM chip to add
+ * @owner: reference to the module providing the chip.
+ *
+ * Register a new PWM chip. @owner is supposed to be THIS_MODULE, use the
+ * pwmchip_add wrapper to do this right.
+ *
+ * Returns: 0 on success or a negative error code on failure.
+ */
+int __pwmchip_add(struct pwm_chip *chip, struct module *owner)
+{
+	int ret;
+
+	if (!chip || !pwmchip_parent(chip) || !chip->ops || !chip->npwm)
+		return -EINVAL;
+
+	/*
+	 * a struct pwm_chip must be allocated using (devm_)pwmchip_alloc,
+	 * otherwise the embedded struct device might disappear too early
+	 * resulting in memory corruption.
+	 * Catch drivers that were not converted appropriately.
+	 */
+	if (!chip->uses_pwmchip_alloc)
+		return -EINVAL;
+
+	if (!pwm_ops_check(chip))
+		return -EINVAL;
+
+	chip->owner = owner;
+
+	if (chip->atomic)
+		spin_lock_init(&chip->atomic_lock);
+	else
+		mutex_init(&chip->nonatomic_lock);
+
+	guard(mutex)(&pwm_lock);
+
+	ret = idr_alloc(&pwm_chips, chip, 0, 0, GFP_KERNEL);
+	if (ret < 0)
+		return ret;
+
+	chip->id = ret;
+
+	dev_set_name(&chip->dev, "pwmchip%u", chip->id);
+
+	if (IS_ENABLED(CONFIG_OF))
+		of_pwmchip_add(chip);
+
+	scoped_guard(pwmchip, chip)
+		chip->operational = true;
+
+	ret = device_add(&chip->dev);
+	if (ret)
+		goto err_device_add;
+
+	return 0;
+
+err_device_add:
+	scoped_guard(pwmchip, chip)
+		chip->operational = false;
+
+	if (IS_ENABLED(CONFIG_OF))
+		of_pwmchip_remove(chip);
+
+	idr_remove(&pwm_chips, chip->id);
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(__pwmchip_add);
+
+/**
+ * pwmchip_remove() - remove a PWM chip
+ * @chip: the PWM chip to remove
+ *
+ * Removes a PWM chip.
+ */
+void pwmchip_remove(struct pwm_chip *chip)
+{
+	pwmchip_sysfs_unexport(chip);
+
+	scoped_guard(mutex, &pwm_lock) {
+		unsigned int i;
+
+		scoped_guard(pwmchip, chip)
+			chip->operational = false;
+
+		for (i = 0; i < chip->npwm; ++i) {
+			struct pwm_device *pwm = &chip->pwms[i];
+
+			if (test_and_clear_bit(PWMF_REQUESTED, &pwm->flags)) {
+				dev_warn(&chip->dev, "Freeing requested PWM #%u\n", i);
+				if (pwm->chip->ops->free)
+					pwm->chip->ops->free(pwm->chip, pwm);
+			}
+		}
+
+		if (IS_ENABLED(CONFIG_OF))
+			of_pwmchip_remove(chip);
+
+		idr_remove(&pwm_chips, chip->id);
+	}
+
+	device_del(&chip->dev);
+}
+EXPORT_SYMBOL_GPL(pwmchip_remove);
+
+static void devm_pwmchip_remove(void *data)
+{
+	struct pwm_chip *chip = data;
+
+	pwmchip_remove(chip);
+}
+
+int __devm_pwmchip_add(struct device *dev, struct pwm_chip *chip, struct module *owner)
+{
+	int ret;
+
+	ret = __pwmchip_add(chip, owner);
+	if (ret)
+		return ret;
+
+	return devm_add_action_or_reset(dev, devm_pwmchip_remove, chip);
+}
+EXPORT_SYMBOL_GPL(__devm_pwmchip_add);
+
+/**
+ * pwm_add_table() - register PWM device consumers
+ * @table: array of consumers to register
+ * @num: number of consumers in table
+ */
+void pwm_add_table(struct pwm_lookup *table, size_t num)
+{
+	guard(mutex)(&pwm_lookup_lock);
+
+	while (num--) {
+		list_add_tail(&table->list, &pwm_lookup_list);
+		table++;
+	}
+}
+
+/**
+ * pwm_remove_table() - unregister PWM device consumers
+ * @table: array of consumers to unregister
+ * @num: number of consumers in table
+ */
+void pwm_remove_table(struct pwm_lookup *table, size_t num)
+{
+	guard(mutex)(&pwm_lookup_lock);
+
+	while (num--) {
+		list_del(&table->list);
+		table++;
+	}
+}
+
 static void pwm_dbg_show(struct pwm_chip *chip, struct seq_file *s)
 {
 	unsigned int i;
diff --git a/drivers/pwm/pwm-axi-pwmgen.c b/drivers/pwm/pwm-axi-pwmgen.c
index 5cee5645fe80..6e56ceb23d18 100644
--- a/drivers/pwm/pwm-axi-pwmgen.c
+++ b/drivers/pwm/pwm-axi-pwmgen.c
@@ -23,6 +23,7 @@
 #include <linux/err.h>
 #include <linux/fpga/adi-axi-common.h>
 #include <linux/io.h>
+#include <linux/minmax.h>
 #include <linux/module.h>
 #include <linux/platform_device.h>
 #include <linux/pwm.h>
@@ -53,86 +54,147 @@ static const struct regmap_config axi_pwmgen_regmap_config = {
 	.max_register = 0xFC,
 };
 
+/* This represents a hardware configuration for one channel */
+struct axi_pwmgen_waveform {
+	u32 period_cnt;
+	u32 duty_cycle_cnt;
+	u32 duty_offset_cnt;
+};
+
 static struct axi_pwmgen_ddata *axi_pwmgen_ddata_from_chip(struct pwm_chip *chip)
 {
 	return pwmchip_get_drvdata(chip);
 }
 
-static int axi_pwmgen_apply(struct pwm_chip *chip, struct pwm_device *pwm,
-			    const struct pwm_state *state)
+static int axi_pwmgen_round_waveform_tohw(struct pwm_chip *chip,
+					  struct pwm_device *pwm,
+					  const struct pwm_waveform *wf,
+					  void *_wfhw)
 {
+	struct axi_pwmgen_waveform *wfhw = _wfhw;
 	struct axi_pwmgen_ddata *ddata = axi_pwmgen_ddata_from_chip(chip);
-	unsigned int ch = pwm->hwpwm;
-	struct regmap *regmap = ddata->regmap;
-	u64 period_cnt, duty_cnt;
-	int ret;
 
-	if (state->polarity != PWM_POLARITY_NORMAL)
-		return -EINVAL;
+	if (wf->period_length_ns == 0) {
+		*wfhw = (struct axi_pwmgen_waveform){
+			.period_cnt = 0,
+			.duty_cycle_cnt = 0,
+			.duty_offset_cnt = 0,
+		};
+	} else {
+		/* With ddata->clk_rate_hz < NSEC_PER_SEC this won't overflow. */
+		wfhw->period_cnt = min_t(u64,
+					 mul_u64_u32_div(wf->period_length_ns, ddata->clk_rate_hz, NSEC_PER_SEC),
+					 U32_MAX);
+
+		if (wfhw->period_cnt == 0) {
+			/*
+			 * The specified period is too short for the hardware.
+			 * Let's round .duty_cycle down to 0 to get a (somewhat)
+			 * valid result.
+			 */
+			wfhw->period_cnt = 1;
+			wfhw->duty_cycle_cnt = 0;
+			wfhw->duty_offset_cnt = 0;
+		} else {
+			wfhw->duty_cycle_cnt = min_t(u64,
+						     mul_u64_u32_div(wf->duty_length_ns, ddata->clk_rate_hz, NSEC_PER_SEC),
+						     U32_MAX);
+			wfhw->duty_offset_cnt = min_t(u64,
+						      mul_u64_u32_div(wf->duty_offset_ns, ddata->clk_rate_hz, NSEC_PER_SEC),
+						      U32_MAX);
+		}
+	}
 
-	if (state->enabled) {
-		period_cnt = mul_u64_u64_div_u64(state->period, ddata->clk_rate_hz, NSEC_PER_SEC);
-		if (period_cnt > UINT_MAX)
-			period_cnt = UINT_MAX;
+	dev_dbg(&chip->dev, "pwm#%u: %lld/%lld [+%lld] @%lu -> PERIOD: %08x, DUTY: %08x, OFFSET: %08x\n",
+		pwm->hwpwm, wf->duty_length_ns, wf->period_length_ns, wf->duty_offset_ns,
+		ddata->clk_rate_hz, wfhw->period_cnt, wfhw->duty_cycle_cnt, wfhw->duty_offset_cnt);
 
-		if (period_cnt == 0)
-			return -EINVAL;
+	return 0;
+}
 
-		ret = regmap_write(regmap, AXI_PWMGEN_CHX_PERIOD(ch), period_cnt);
-		if (ret)
-			return ret;
+static int axi_pwmgen_round_waveform_fromhw(struct pwm_chip *chip, struct pwm_device *pwm,
+					     const void *_wfhw, struct pwm_waveform *wf)
+{
+	const struct axi_pwmgen_waveform *wfhw = _wfhw;
+	struct axi_pwmgen_ddata *ddata = axi_pwmgen_ddata_from_chip(chip);
 
-		duty_cnt = mul_u64_u64_div_u64(state->duty_cycle, ddata->clk_rate_hz, NSEC_PER_SEC);
-		if (duty_cnt > UINT_MAX)
-			duty_cnt = UINT_MAX;
+	wf->period_length_ns = DIV64_U64_ROUND_UP((u64)wfhw->period_cnt * NSEC_PER_SEC,
+					ddata->clk_rate_hz);
 
-		ret = regmap_write(regmap, AXI_PWMGEN_CHX_DUTY(ch), duty_cnt);
-		if (ret)
-			return ret;
-	} else {
-		ret = regmap_write(regmap, AXI_PWMGEN_CHX_PERIOD(ch), 0);
-		if (ret)
-			return ret;
+	wf->duty_length_ns = DIV64_U64_ROUND_UP((u64)wfhw->duty_cycle_cnt * NSEC_PER_SEC,
+					    ddata->clk_rate_hz);
 
-		ret = regmap_write(regmap, AXI_PWMGEN_CHX_DUTY(ch), 0);
-		if (ret)
-			return ret;
-	}
+	wf->duty_offset_ns = DIV64_U64_ROUND_UP((u64)wfhw->duty_offset_cnt * NSEC_PER_SEC,
+					     ddata->clk_rate_hz);
 
-	return regmap_write(regmap, AXI_PWMGEN_REG_CONFIG, AXI_PWMGEN_LOAD_CONFIG);
+	return 0;
 }
 
-static int axi_pwmgen_get_state(struct pwm_chip *chip, struct pwm_device *pwm,
-				struct pwm_state *state)
+static int axi_pwmgen_write_waveform(struct pwm_chip *chip,
+				     struct pwm_device *pwm,
+				     const void *_wfhw)
 {
+	const struct axi_pwmgen_waveform *wfhw = _wfhw;
 	struct axi_pwmgen_ddata *ddata = axi_pwmgen_ddata_from_chip(chip);
 	struct regmap *regmap = ddata->regmap;
 	unsigned int ch = pwm->hwpwm;
-	u32 cnt;
 	int ret;
 
-	ret = regmap_read(regmap, AXI_PWMGEN_CHX_PERIOD(ch), &cnt);
+	ret = regmap_write(regmap, AXI_PWMGEN_CHX_PERIOD(ch), wfhw->period_cnt);
 	if (ret)
 		return ret;
 
-	state->enabled = cnt != 0;
+	ret = regmap_write(regmap, AXI_PWMGEN_CHX_DUTY(ch), wfhw->duty_cycle_cnt);
+	if (ret)
+		return ret;
 
-	state->period = DIV_ROUND_UP_ULL((u64)cnt * NSEC_PER_SEC, ddata->clk_rate_hz);
+	ret = regmap_write(regmap, AXI_PWMGEN_CHX_OFFSET(ch), wfhw->duty_offset_cnt);
+	if (ret)
+		return ret;
 
-	ret = regmap_read(regmap, AXI_PWMGEN_CHX_DUTY(ch), &cnt);
+	return regmap_write(regmap, AXI_PWMGEN_REG_CONFIG, AXI_PWMGEN_LOAD_CONFIG);
+}
+
+static int axi_pwmgen_read_waveform(struct pwm_chip *chip,
+				    struct pwm_device *pwm,
+				    void *_wfhw)
+{
+	struct axi_pwmgen_waveform *wfhw = _wfhw;
+	struct axi_pwmgen_ddata *ddata = axi_pwmgen_ddata_from_chip(chip);
+	struct regmap *regmap = ddata->regmap;
+	unsigned int ch = pwm->hwpwm;
+	int ret;
+
+	ret = regmap_read(regmap, AXI_PWMGEN_CHX_PERIOD(ch), &wfhw->period_cnt);
+	if (ret)
+		return ret;
+
+	ret = regmap_read(regmap, AXI_PWMGEN_CHX_DUTY(ch), &wfhw->duty_cycle_cnt);
 	if (ret)
 		return ret;
 
-	state->duty_cycle = DIV_ROUND_UP_ULL((u64)cnt * NSEC_PER_SEC, ddata->clk_rate_hz);
+	ret = regmap_read(regmap, AXI_PWMGEN_CHX_OFFSET(ch), &wfhw->duty_offset_cnt);
+	if (ret)
+		return ret;
 
-	state->polarity = PWM_POLARITY_NORMAL;
+	if (wfhw->duty_cycle_cnt > wfhw->period_cnt)
+		wfhw->duty_cycle_cnt = wfhw->period_cnt;
+
+	/* XXX: is this the actual behaviour of the hardware? */
+	if (wfhw->duty_offset_cnt >= wfhw->period_cnt) {
+		wfhw->duty_cycle_cnt = 0;
+		wfhw->duty_offset_cnt = 0;
+	}
 
 	return 0;
 }
 
 static const struct pwm_ops axi_pwmgen_pwm_ops = {
-	.apply = axi_pwmgen_apply,
-	.get_state = axi_pwmgen_get_state,
+	.sizeof_wfhw = sizeof(struct axi_pwmgen_waveform),
+	.round_waveform_tohw = axi_pwmgen_round_waveform_tohw,
+	.round_waveform_fromhw = axi_pwmgen_round_waveform_fromhw,
+	.read_waveform = axi_pwmgen_read_waveform,
+	.write_waveform = axi_pwmgen_write_waveform,
 };
 
 static int axi_pwmgen_setup(struct regmap *regmap, struct device *dev)
diff --git a/drivers/pwm/pwm-stm32.c b/drivers/pwm/pwm-stm32.c
index eb24054f9729..b889e64522c3 100644
--- a/drivers/pwm/pwm-stm32.c
+++ b/drivers/pwm/pwm-stm32.c
@@ -51,6 +51,391 @@ static u32 active_channels(struct stm32_pwm *dev)
 	return ccer & TIM_CCER_CCXE;
 }
 
+struct stm32_pwm_waveform {
+	u32 ccer;
+	u32 psc;
+	u32 arr;
+	u32 ccr;
+};
+
+static int stm32_pwm_round_waveform_tohw(struct pwm_chip *chip,
+					 struct pwm_device *pwm,
+					 const struct pwm_waveform *wf,
+					 void *_wfhw)
+{
+	struct stm32_pwm_waveform *wfhw = _wfhw;
+	struct stm32_pwm *priv = to_stm32_pwm_dev(chip);
+	unsigned int ch = pwm->hwpwm;
+	unsigned long rate;
+	u64 ccr, duty;
+	int ret;
+
+	if (wf->period_length_ns == 0) {
+		*wfhw = (struct stm32_pwm_waveform){
+			.ccer = 0,
+		};
+
+		return 0;
+	}
+
+	ret = clk_enable(priv->clk);
+	if (ret)
+		return ret;
+
+	wfhw->ccer = TIM_CCER_CCxE(ch + 1);
+	if (priv->have_complementary_output)
+		wfhw->ccer = TIM_CCER_CCxNE(ch + 1);
+
+	rate = clk_get_rate(priv->clk);
+
+	if (active_channels(priv) & ~(1 << ch * 4)) {
+		u64 arr;
+
+		/*
+		 * Other channels are already enabled, so the configured PSC and
+		 * ARR must be used for this channel, too.
+		 */
+		ret = regmap_read(priv->regmap, TIM_PSC, &wfhw->psc);
+		if (ret)
+			goto out;
+
+		ret = regmap_read(priv->regmap, TIM_ARR, &wfhw->arr);
+		if (ret)
+			goto out;
+
+		/*
+		 * calculate the best value for ARR for the given PSC, refuse if
+		 * the resulting period gets bigger than the requested one.
+		 */
+		arr = mul_u64_u64_div_u64(wf->period_length_ns, rate,
+					  (u64)NSEC_PER_SEC * (wfhw->psc + 1));
+		if (arr <= wfhw->arr) {
+			/*
+			 * requested period is small than the currently
+			 * configured and unchangable period, report back the smallest
+			 * possible period, i.e. the current state; Initialize
+			 * ccr to anything valid.
+			 */
+			wfhw->ccr = 0;
+			ret = 1;
+			goto out;
+		}
+
+	} else {
+		/*
+		 * .probe() asserted that clk_get_rate() is not bigger than 1 GHz, so
+		 * the calculations here won't overflow.
+		 * First we need to find the minimal value for prescaler such that
+		 *
+		 *        period_ns * clkrate
+		 *   ------------------------------ < max_arr + 1
+		 *   NSEC_PER_SEC * (prescaler + 1)
+		 *
+		 * This equation is equivalent to
+		 *
+		 *        period_ns * clkrate
+		 *   ---------------------------- < prescaler + 1
+		 *   NSEC_PER_SEC * (max_arr + 1)
+		 *
+		 * Using integer division and knowing that the right hand side is
+		 * integer, this is further equivalent to
+		 *
+		 *   (period_ns * clkrate) // (NSEC_PER_SEC * (max_arr + 1)) ≤ prescaler
+		 */
+		u64 psc = mul_u64_u64_div_u64(wf->period_length_ns, rate,
+					      (u64)NSEC_PER_SEC * ((u64)priv->max_arr + 1));
+		u64 arr;
+
+		wfhw->psc = min_t(u64, psc, MAX_TIM_PSC);
+
+		arr = mul_u64_u64_div_u64(wf->period_length_ns, rate,
+					  (u64)NSEC_PER_SEC * (wfhw->psc + 1));
+		if (!arr) {
+			/*
+			 * requested period is too small, report back the smallest
+			 * possible period, i.e. ARR = 0. The only valid CCR
+			 * value is then zero, too.
+			 */
+			wfhw->arr = 0;
+			wfhw->ccr = 0;
+			ret = 1;
+			goto out;
+		}
+
+		/*
+		 * ARR is limited intentionally to values less than
+		 * priv->max_arr to allow 100% duty cycle.
+		 */
+		wfhw->arr = min_t(u64, arr, priv->max_arr) - 1;
+	}
+
+	duty = mul_u64_u64_div_u64(wf->duty_length_ns, rate,
+				   (u64)NSEC_PER_SEC * (wfhw->psc + 1));
+	duty = min_t(u64, duty, wfhw->arr + 1);
+
+	if (wf->duty_length_ns && wf->duty_offset_ns &&
+	    wf->duty_length_ns + wf->duty_offset_ns >= wf->period_length_ns) {
+		wfhw->ccer |= TIM_CCER_CCxP(ch + 1);
+		if (priv->have_complementary_output)
+			wfhw->ccer |= TIM_CCER_CCxNP(ch + 1);
+
+		ccr = wfhw->arr + 1 - duty;
+	} else {
+		ccr = duty;
+	}
+
+	wfhw->ccr = min_t(u64, ccr, wfhw->arr + 1);
+
+	dev_dbg(&chip->dev, "pwm#%u: %lld/%lld [+%lld] @%lu -> CCER: %08x, PSC: %08x, ARR: %08x, CCR: %08x\n",
+		pwm->hwpwm, wf->duty_length_ns, wf->period_length_ns, wf->duty_offset_ns,
+		rate, wfhw->ccer, wfhw->psc, wfhw->arr, wfhw->ccr);
+
+out:
+	clk_disable(priv->clk);
+
+	return ret;
+}
+
+/*
+ * This should be moved to lib/math/div64.c. Currently there are some changes
+ * pending to mul_u64_u64_div_u64. Uwe will care for that when the dust settles.
+ */
+static u64 stm32_pwm_mul_u64_u64_div_u64_roundup(u64 a, u64 b, u64 c)
+{
+	u64 res = mul_u64_u64_div_u64(a, b, c);
+	/* Those multiplications might overflow but it doesn't matter */
+	u64 rem = a * b - c * res;
+
+	if (rem)
+		res += 1;
+
+	return res;
+}
+
+static int stm32_pwm_round_waveform_fromhw(struct pwm_chip *chip,
+					   struct pwm_device *pwm,
+					   const void *_wfhw,
+					   struct pwm_waveform *wf)
+{
+	const struct stm32_pwm_waveform *wfhw = _wfhw;
+	struct stm32_pwm *priv = to_stm32_pwm_dev(chip);
+	unsigned int ch = pwm->hwpwm;
+
+	if (wfhw->ccer & TIM_CCER_CCxE(ch + 1)) {
+		unsigned long rate = clk_get_rate(priv->clk);
+		u64 ccr_ns;
+
+		/* The result doesn't overflow for rate >= 15259 */
+		wf->period_length_ns = stm32_pwm_mul_u64_u64_div_u64_roundup(((u64)wfhw->psc + 1) * (wfhw->arr + 1),
+									     NSEC_PER_SEC, rate);
+
+		ccr_ns = stm32_pwm_mul_u64_u64_div_u64_roundup(((u64)wfhw->psc + 1) * wfhw->ccr,
+							       NSEC_PER_SEC, rate);
+
+		if (wfhw->ccer & TIM_CCER_CCxP(ch + 1)) {
+			wf->duty_length_ns =
+				stm32_pwm_mul_u64_u64_div_u64_roundup(((u64)wfhw->psc + 1) * (wfhw->arr + 1 - wfhw->ccr),
+								      NSEC_PER_SEC, rate);
+
+			wf->duty_offset_ns = ccr_ns;
+		} else {
+			wf->duty_length_ns = ccr_ns;
+			wf->duty_offset_ns = 0;
+		}
+
+		dev_dbg(&chip->dev, "pwm#%u: CCER: %08x, PSC: %08x, ARR: %08x, CCR: %08x @%lu -> %lld/%lld [+%lld]\n",
+			pwm->hwpwm, wfhw->ccer, wfhw->psc, wfhw->arr, wfhw->ccr, rate,
+			wf->duty_length_ns, wf->period_length_ns, wf->duty_offset_ns);
+
+	} else {
+		*wf = (struct pwm_waveform){
+			.period_length_ns = 0,
+		};
+	}
+
+	return 0;
+}
+
+static int stm32_pwm_read_waveform(struct pwm_chip *chip,
+				     struct pwm_device *pwm,
+				     void *_wfhw)
+{
+	struct stm32_pwm_waveform *wfhw = _wfhw;
+	struct stm32_pwm *priv = to_stm32_pwm_dev(chip);
+	unsigned int ch = pwm->hwpwm;
+	int ret;
+
+	ret = clk_enable(priv->clk);
+	if (ret)
+		return ret;
+
+	ret = regmap_read(priv->regmap, TIM_CCER, &wfhw->ccer);
+	if (ret)
+		goto out;
+
+	if (wfhw->ccer & TIM_CCER_CCxE(ch + 1)) {
+		ret = regmap_read(priv->regmap, TIM_PSC, &wfhw->psc);
+		if (ret)
+			goto out;
+
+		ret = regmap_read(priv->regmap, TIM_ARR, &wfhw->arr);
+		if (ret)
+			goto out;
+
+		if (wfhw->arr == U32_MAX)
+			wfhw->arr -= 1;
+
+		ret = regmap_read(priv->regmap, TIM_CCRx(ch + 1), &wfhw->ccr);
+		if (ret)
+			goto out;
+
+		if (wfhw->ccr > wfhw->arr + 1)
+			wfhw->ccr = wfhw->arr + 1;
+	}
+
+out:
+	clk_disable(priv->clk);
+
+	return ret;
+}
+
+static int stm32_pwm_write_waveform(struct pwm_chip *chip,
+				      struct pwm_device *pwm,
+				      const void *_wfhw)
+{
+	const struct stm32_pwm_waveform *wfhw = _wfhw;
+	struct stm32_pwm *priv = to_stm32_pwm_dev(chip);
+	unsigned int ch = pwm->hwpwm;
+	int ret;
+
+	ret = clk_enable(priv->clk);
+	if (ret)
+		return ret;
+
+	if (wfhw->ccer & TIM_CCER_CCxE(ch + 1)) {
+		u32 ccer, mask;
+		unsigned int shift;
+		u32 ccmr;
+
+		ret = regmap_read(priv->regmap, TIM_CCER, &ccer);
+		if (ret)
+			goto out;
+
+		/* If there are other channels enabled, don't update PSC and ARR */
+		if (ccer & ~TIM_CCER_CCxE(ch + 1) & TIM_CCER_CCXE) {
+			u32 psc, arr;
+
+			ret = regmap_read(priv->regmap, TIM_PSC, &psc);
+			if (ret)
+				goto out;
+
+			if (psc != wfhw->psc) {
+				ret = -EBUSY;
+				goto out;
+			}
+
+			ret = regmap_read(priv->regmap, TIM_ARR, &arr);
+			if (ret)
+				goto out;
+
+			if (arr != wfhw->arr) {
+				ret = -EBUSY;
+				goto out;
+			}
+		} else {
+			ret = regmap_write(priv->regmap, TIM_PSC, wfhw->psc);
+			if (ret)
+				goto out;
+
+			ret = regmap_write(priv->regmap, TIM_ARR, wfhw->arr);
+			if (ret)
+				goto out;
+
+			ret = regmap_set_bits(priv->regmap, TIM_CR1, TIM_CR1_ARPE);
+			if (ret)
+				goto out;
+
+		}
+
+		/* set polarity */
+		mask = TIM_CCER_CCxP(ch + 1) | TIM_CCER_CCxNP(ch + 1);
+		ret = regmap_update_bits(priv->regmap, TIM_CCER, mask, wfhw->ccer);
+		if (ret)
+			goto out;
+
+		ret = regmap_write(priv->regmap, TIM_CCRx(ch + 1), wfhw->ccr);
+		if (ret)
+			goto out;
+
+		/* Configure output mode */
+		shift = (ch & 0x1) * CCMR_CHANNEL_SHIFT;
+		ccmr = (TIM_CCMR_PE | TIM_CCMR_M1) << shift;
+		mask = CCMR_CHANNEL_MASK << shift;
+
+		if (ch < 2)
+			ret = regmap_update_bits(priv->regmap, TIM_CCMR1, mask, ccmr);
+		else
+			ret = regmap_update_bits(priv->regmap, TIM_CCMR2, mask, ccmr);
+		if (ret)
+			goto out;
+
+		ret = regmap_set_bits(priv->regmap, TIM_BDTR, TIM_BDTR_MOE);
+		if (ret)
+			goto out;
+
+		if (!(ccer & TIM_CCER_CCxE(ch + 1))) {
+			mask = TIM_CCER_CCxE(ch + 1) | TIM_CCER_CCxNE(ch + 1);
+
+			ret = clk_enable(priv->clk);
+			if (ret)
+				goto out;
+
+			ccer = (ccer & ~mask) | (wfhw->ccer & mask);
+			regmap_write(priv->regmap, TIM_CCER, ccer);
+
+			/* Make sure that registers are updated */
+			regmap_set_bits(priv->regmap, TIM_EGR, TIM_EGR_UG);
+
+			/* Enable controller */
+			regmap_set_bits(priv->regmap, TIM_CR1, TIM_CR1_CEN);
+		}
+
+	} else {
+		/* disable channel */
+		u32 mask, ccer;
+
+		mask = TIM_CCER_CCxE(ch + 1);
+		if (priv->have_complementary_output)
+			mask |= TIM_CCER_CCxNE(ch + 1);
+
+		ret = regmap_read(priv->regmap, TIM_CCER, &ccer);
+		if (ret)
+			goto out;
+
+		if (ccer & mask) {
+			ccer = ccer & ~mask;
+
+			ret = regmap_write(priv->regmap, TIM_CCER, ccer);
+			if (ret)
+				goto out;
+
+			if (!(ccer & TIM_CCER_CCXE)) {
+				/* When all channels are disabled, we can disable the controller */
+				ret = regmap_clear_bits(priv->regmap, TIM_CR1, TIM_CR1_CEN);
+				if (ret)
+					goto out;
+			}
+
+			clk_disable(priv->clk);
+		}
+	}
+
+out:
+	clk_disable(priv->clk);
+
+	return ret;
+}
+
 #define TIM_CCER_CC12P (TIM_CCER_CC1P | TIM_CCER_CC2P)
 #define TIM_CCER_CC12E (TIM_CCER_CC1E | TIM_CCER_CC2E)
 #define TIM_CCER_CC34P (TIM_CCER_CC3P | TIM_CCER_CC4P)
@@ -308,228 +693,13 @@ unlock:
 	return ret;
 }
 
-static int stm32_pwm_config(struct stm32_pwm *priv, unsigned int ch,
-			    u64 duty_ns, u64 period_ns)
-{
-	unsigned long long prd, dty;
-	unsigned long long prescaler;
-	u32 ccmr, mask, shift;
-
-	/*
-	 * .probe() asserted that clk_get_rate() is not bigger than 1 GHz, so
-	 * the calculations here won't overflow.
-	 * First we need to find the minimal value for prescaler such that
-	 *
-	 *        period_ns * clkrate
-	 *   ------------------------------ < max_arr + 1
-	 *   NSEC_PER_SEC * (prescaler + 1)
-	 *
-	 * This equation is equivalent to
-	 *
-	 *        period_ns * clkrate
-	 *   ---------------------------- < prescaler + 1
-	 *   NSEC_PER_SEC * (max_arr + 1)
-	 *
-	 * Using integer division and knowing that the right hand side is
-	 * integer, this is further equivalent to
-	 *
-	 *   (period_ns * clkrate) // (NSEC_PER_SEC * (max_arr + 1)) ≤ prescaler
-	 */
-
-	prescaler = mul_u64_u64_div_u64(period_ns, clk_get_rate(priv->clk),
-					(u64)NSEC_PER_SEC * ((u64)priv->max_arr + 1));
-	if (prescaler > MAX_TIM_PSC)
-		return -EINVAL;
-
-	prd = mul_u64_u64_div_u64(period_ns, clk_get_rate(priv->clk),
-				  (u64)NSEC_PER_SEC * (prescaler + 1));
-	if (!prd)
-		return -EINVAL;
-
-	/*
-	 * All channels share the same prescaler and counter so when two
-	 * channels are active at the same time we can't change them
-	 */
-	if (active_channels(priv) & ~(1 << ch * 4)) {
-		u32 psc, arr;
-
-		regmap_read(priv->regmap, TIM_PSC, &psc);
-		regmap_read(priv->regmap, TIM_ARR, &arr);
-
-		if ((psc != prescaler) || (arr != prd - 1))
-			return -EBUSY;
-	}
-
-	regmap_write(priv->regmap, TIM_PSC, prescaler);
-	regmap_write(priv->regmap, TIM_ARR, prd - 1);
-	regmap_set_bits(priv->regmap, TIM_CR1, TIM_CR1_ARPE);
-
-	/* Calculate the duty cycles */
-	dty = mul_u64_u64_div_u64(duty_ns, clk_get_rate(priv->clk),
-				  (u64)NSEC_PER_SEC * (prescaler + 1));
-
-	regmap_write(priv->regmap, TIM_CCRx(ch + 1), dty);
-
-	/* Configure output mode */
-	shift = (ch & 0x1) * CCMR_CHANNEL_SHIFT;
-	ccmr = (TIM_CCMR_PE | TIM_CCMR_M1) << shift;
-	mask = CCMR_CHANNEL_MASK << shift;
-
-	if (ch < 2)
-		regmap_update_bits(priv->regmap, TIM_CCMR1, mask, ccmr);
-	else
-		regmap_update_bits(priv->regmap, TIM_CCMR2, mask, ccmr);
-
-	regmap_set_bits(priv->regmap, TIM_BDTR, TIM_BDTR_MOE);
-
-	return 0;
-}
-
-static int stm32_pwm_set_polarity(struct stm32_pwm *priv, unsigned int ch,
-				  enum pwm_polarity polarity)
-{
-	u32 mask;
-
-	mask = TIM_CCER_CCxP(ch + 1);
-	if (priv->have_complementary_output)
-		mask |= TIM_CCER_CCxNP(ch + 1);
-
-	regmap_update_bits(priv->regmap, TIM_CCER, mask,
-			   polarity == PWM_POLARITY_NORMAL ? 0 : mask);
-
-	return 0;
-}
-
-static int stm32_pwm_enable(struct stm32_pwm *priv, unsigned int ch)
-{
-	u32 mask;
-	int ret;
-
-	ret = clk_enable(priv->clk);
-	if (ret)
-		return ret;
-
-	/* Enable channel */
-	mask = TIM_CCER_CCxE(ch + 1);
-	if (priv->have_complementary_output)
-		mask |= TIM_CCER_CCxNE(ch + 1);
-
-	regmap_set_bits(priv->regmap, TIM_CCER, mask);
-
-	/* Make sure that registers are updated */
-	regmap_set_bits(priv->regmap, TIM_EGR, TIM_EGR_UG);
-
-	/* Enable controller */
-	regmap_set_bits(priv->regmap, TIM_CR1, TIM_CR1_CEN);
-
-	return 0;
-}
-
-static void stm32_pwm_disable(struct stm32_pwm *priv, unsigned int ch)
-{
-	u32 mask;
-
-	/* Disable channel */
-	mask = TIM_CCER_CCxE(ch + 1);
-	if (priv->have_complementary_output)
-		mask |= TIM_CCER_CCxNE(ch + 1);
-
-	regmap_clear_bits(priv->regmap, TIM_CCER, mask);
-
-	/* When all channels are disabled, we can disable the controller */
-	if (!active_channels(priv))
-		regmap_clear_bits(priv->regmap, TIM_CR1, TIM_CR1_CEN);
-
-	clk_disable(priv->clk);
-}
-
-static int stm32_pwm_apply(struct pwm_chip *chip, struct pwm_device *pwm,
-			   const struct pwm_state *state)
-{
-	bool enabled;
-	struct stm32_pwm *priv = to_stm32_pwm_dev(chip);
-	int ret;
-
-	enabled = pwm->state.enabled;
-
-	if (!state->enabled) {
-		if (enabled)
-			stm32_pwm_disable(priv, pwm->hwpwm);
-		return 0;
-	}
-
-	if (state->polarity != pwm->state.polarity)
-		stm32_pwm_set_polarity(priv, pwm->hwpwm, state->polarity);
-
-	ret = stm32_pwm_config(priv, pwm->hwpwm,
-			       state->duty_cycle, state->period);
-	if (ret)
-		return ret;
-
-	if (!enabled && state->enabled)
-		ret = stm32_pwm_enable(priv, pwm->hwpwm);
-
-	return ret;
-}
-
-static int stm32_pwm_apply_locked(struct pwm_chip *chip, struct pwm_device *pwm,
-				  const struct pwm_state *state)
-{
-	struct stm32_pwm *priv = to_stm32_pwm_dev(chip);
-	int ret;
-
-	/* protect common prescaler for all active channels */
-	mutex_lock(&priv->lock);
-	ret = stm32_pwm_apply(chip, pwm, state);
-	mutex_unlock(&priv->lock);
-
-	return ret;
-}
-
-static int stm32_pwm_get_state(struct pwm_chip *chip,
-			       struct pwm_device *pwm, struct pwm_state *state)
-{
-	struct stm32_pwm *priv = to_stm32_pwm_dev(chip);
-	int ch = pwm->hwpwm;
-	unsigned long rate;
-	u32 ccer, psc, arr, ccr;
-	u64 dty, prd;
-	int ret;
-
-	mutex_lock(&priv->lock);
-
-	ret = regmap_read(priv->regmap, TIM_CCER, &ccer);
-	if (ret)
-		goto out;
-
-	state->enabled = ccer & TIM_CCER_CCxE(ch + 1);
-	state->polarity = (ccer & TIM_CCER_CCxP(ch + 1)) ?
-			  PWM_POLARITY_INVERSED : PWM_POLARITY_NORMAL;
-	ret = regmap_read(priv->regmap, TIM_PSC, &psc);
-	if (ret)
-		goto out;
-	ret = regmap_read(priv->regmap, TIM_ARR, &arr);
-	if (ret)
-		goto out;
-	ret = regmap_read(priv->regmap, TIM_CCRx(ch + 1), &ccr);
-	if (ret)
-		goto out;
-
-	rate = clk_get_rate(priv->clk);
-
-	prd = (u64)NSEC_PER_SEC * (psc + 1) * (arr + 1);
-	state->period = DIV_ROUND_UP_ULL(prd, rate);
-	dty = (u64)NSEC_PER_SEC * (psc + 1) * ccr;
-	state->duty_cycle = DIV_ROUND_UP_ULL(dty, rate);
-
-out:
-	mutex_unlock(&priv->lock);
-	return ret;
-}
-
 static const struct pwm_ops stm32pwm_ops = {
-	.apply = stm32_pwm_apply_locked,
-	.get_state = stm32_pwm_get_state,
+	.sizeof_wfhw = sizeof(struct stm32_pwm_waveform),
+	.round_waveform_tohw = stm32_pwm_round_waveform_tohw,
+	.round_waveform_fromhw = stm32_pwm_round_waveform_fromhw,
+	.read_waveform = stm32_pwm_read_waveform,
+	.write_waveform = stm32_pwm_write_waveform,
+
 	.capture = IS_ENABLED(CONFIG_DMA_ENGINE) ? stm32_pwm_capture : NULL,
 };