[PATCH v6 0/4] Improve pwm-ir-tx precision

[PATCH v6 0/4] Improve pwm-ir-tx precision

[Sean Young]

The pwm-ir-tx driver has to turn the pwm signal on and off, and suffers
from delays as this is done in process context. Make this work in atomic
context.

changes:

v6:
 - Rename pwm_apply_cansleep() to pwm_apply_might_sleep()
 - rebase

v5:
 - Use dev_err_probe() as Uwe suggested
 - rebase

v4:
 - Split out the pwm_apply_state() rename into seperate commit
 - Atomic pwm function is called pwm_apply_atomic() and sleeping
   counterpart pwm_apply_cansleep()
 - Fix commit messages

v3:
 - Renamed pwm_apply_state() to pwm_apply_cansleep()
 - Moved pwm atomic to struct pwm_chip
 - Fixed pwm docs
 - Other minor code review comments

v2:
 - Simplified the pwm ir tx based on Ivaylo Dimitrov suggestions
 - Renamed pwm_can_sleep() to pwm_is_atomic
 - Fix sleeping issue in pwm-bcm2835 (clk_get_rate() can sleep)
 - use non_block_start() in pwm to check atomic-ness


Sean Young (4):
  pwm: rename pwm_apply_state() to pwm_apply_might_sleep()
  pwm: make it possible to apply pwm changes in atomic context
  pwm: bcm2835: allow pwm driver to be used in atomic context
  media: pwm-ir-tx: trigger edges from hrtimer interrupt context

 Documentation/driver-api/pwm.rst              | 17 +++-
 .../gpu/drm/i915/display/intel_backlight.c    |  6 +-
 drivers/gpu/drm/solomon/ssd130x.c             |  2 +-
 drivers/hwmon/pwm-fan.c                       |  8 +-
 drivers/input/misc/da7280.c                   |  4 +-
 drivers/input/misc/pwm-beeper.c               |  4 +-
 drivers/input/misc/pwm-vibra.c                |  8 +-
 drivers/leds/leds-pwm.c                       |  2 +-
 drivers/leds/rgb/leds-pwm-multicolor.c        |  4 +-
 drivers/media/rc/pwm-ir-tx.c                  | 83 +++++++++++++++++--
 drivers/platform/x86/lenovo-yogabook.c        |  2 +-
 drivers/pwm/core.c                            | 75 +++++++++++++----
 drivers/pwm/pwm-bcm2835.c                     | 31 ++++---
 drivers/pwm/pwm-renesas-tpu.c                 |  1 -
 drivers/pwm/pwm-twl-led.c                     |  2 +-
 drivers/pwm/pwm-vt8500.c                      |  2 +-
 drivers/pwm/sysfs.c                           | 10 +--
 drivers/regulator/pwm-regulator.c             |  4 +-
 drivers/video/backlight/lm3630a_bl.c          |  2 +-
 drivers/video/backlight/lp855x_bl.c           |  2 +-
 drivers/video/backlight/pwm_bl.c              | 12 +--
 drivers/video/fbdev/ssd1307fb.c               |  2 +-
 include/linux/pwm.h                           | 57 +++++++++----
 23 files changed, 248 insertions(+), 92 deletions(-)

-- 
2.43.0

[PATCH v6 1/4] pwm: rename pwm_apply_state() to pwm_apply_might_sleep()

[Sean Young]

In order to introduce a pwm api which can be used from atomic context,
we will need two functions for applying pwm changes:

	int pwm_apply_might_sleep(struct pwm *, struct pwm_state *);
	int pwm_apply_atomic(struct pwm *, struct pwm_state *);

This commit just deals with renaming pwm_apply_state(), a following
commit will introduce the pwm_apply_atomic() function.

Acked-by: Hans de Goede <hdegoede@redhat.com>
Acked-by: Jani Nikula <jani.nikula@intel.com>
Acked-by: Lee Jones <lee@kernel.org>
Signed-off-by: Sean Young <sean@mess.org>
---
 Documentation/driver-api/pwm.rst              |  8 +++---
 .../gpu/drm/i915/display/intel_backlight.c    |  6 ++--
 drivers/gpu/drm/solomon/ssd130x.c             |  2 +-
 drivers/hwmon/pwm-fan.c                       |  8 +++---
 drivers/input/misc/da7280.c                   |  4 +--
 drivers/input/misc/pwm-beeper.c               |  4 +--
 drivers/input/misc/pwm-vibra.c                |  8 +++---
 drivers/leds/leds-pwm.c                       |  2 +-
 drivers/leds/rgb/leds-pwm-multicolor.c        |  4 +--
 drivers/media/rc/pwm-ir-tx.c                  |  4 +--
 drivers/platform/x86/lenovo-yogabook.c        |  2 +-
 drivers/pwm/core.c                            | 18 ++++++------
 drivers/pwm/pwm-twl-led.c                     |  2 +-
 drivers/pwm/pwm-vt8500.c                      |  2 +-
 drivers/pwm/sysfs.c                           | 10 +++----
 drivers/regulator/pwm-regulator.c             |  4 +--
 drivers/video/backlight/lm3630a_bl.c          |  2 +-
 drivers/video/backlight/lp855x_bl.c           |  2 +-
 drivers/video/backlight/pwm_bl.c              | 12 ++++----
 drivers/video/fbdev/ssd1307fb.c               |  2 +-
 include/linux/pwm.h                           | 28 +++++++++----------
 21 files changed, 67 insertions(+), 67 deletions(-)

diff --git a/Documentation/driver-api/pwm.rst b/Documentation/driver-api/pwm.rst
index bb264490a87a..f1d8197c8c43 100644
--- a/Documentation/driver-api/pwm.rst
+++ b/Documentation/driver-api/pwm.rst
@@ -41,7 +41,7 @@ the getter, devm_pwm_get() and devm_fwnode_pwm_get(), also exist.
 
 After being requested, a PWM has to be configured using::
 
-	int pwm_apply_state(struct pwm_device *pwm, struct pwm_state *state);
+	int pwm_apply_might_sleep(struct pwm_device *pwm, struct pwm_state *state);
 
 This API controls both the PWM period/duty_cycle config and the
 enable/disable state.
@@ -57,13 +57,13 @@ If supported by the driver, the signal can be optimized, for example to improve
 EMI by phase shifting the individual channels of a chip.
 
 The pwm_config(), pwm_enable() and pwm_disable() functions are just wrappers
-around pwm_apply_state() and should not be used if the user wants to change
+around pwm_apply_might_sleep() and should not be used if the user wants to change
 several parameter at once. For example, if you see pwm_config() and
 pwm_{enable,disable}() calls in the same function, this probably means you
-should switch to pwm_apply_state().
+should switch to pwm_apply_might_sleep().
 
 The PWM user API also allows one to query the PWM state that was passed to the
-last invocation of pwm_apply_state() using pwm_get_state(). Note this is
+last invocation of pwm_apply_might_sleep() using pwm_get_state(). Note this is
 different to what the driver has actually implemented if the request cannot be
 satisfied exactly with the hardware in use. There is currently no way for
 consumers to get the actually implemented settings.
diff --git a/drivers/gpu/drm/i915/display/intel_backlight.c b/drivers/gpu/drm/i915/display/intel_backlight.c
index 2e8f17c04522..ff9b9918b0a1 100644
--- a/drivers/gpu/drm/i915/display/intel_backlight.c
+++ b/drivers/gpu/drm/i915/display/intel_backlight.c
@@ -274,7 +274,7 @@ static void ext_pwm_set_backlight(const struct drm_connector_state *conn_state,
 	struct intel_panel *panel = &to_intel_connector(conn_state->connector)->panel;
 
 	pwm_set_relative_duty_cycle(&panel->backlight.pwm_state, level, 100);
-	pwm_apply_state(panel->backlight.pwm, &panel->backlight.pwm_state);
+	pwm_apply_might_sleep(panel->backlight.pwm, &panel->backlight.pwm_state);
 }
 
 static void
@@ -427,7 +427,7 @@ static void ext_pwm_disable_backlight(const struct drm_connector_state *old_conn
 	intel_backlight_set_pwm_level(old_conn_state, level);
 
 	panel->backlight.pwm_state.enabled = false;
-	pwm_apply_state(panel->backlight.pwm, &panel->backlight.pwm_state);
+	pwm_apply_might_sleep(panel->backlight.pwm, &panel->backlight.pwm_state);
 }
 
 void intel_backlight_disable(const struct drm_connector_state *old_conn_state)
@@ -749,7 +749,7 @@ static void ext_pwm_enable_backlight(const struct intel_crtc_state *crtc_state,
 
 	pwm_set_relative_duty_cycle(&panel->backlight.pwm_state, level, 100);
 	panel->backlight.pwm_state.enabled = true;
-	pwm_apply_state(panel->backlight.pwm, &panel->backlight.pwm_state);
+	pwm_apply_might_sleep(panel->backlight.pwm, &panel->backlight.pwm_state);
 }
 
 static void __intel_backlight_enable(const struct intel_crtc_state *crtc_state,
diff --git a/drivers/gpu/drm/solomon/ssd130x.c b/drivers/gpu/drm/solomon/ssd130x.c
index e0174f82e353..cce043a4a1dc 100644
--- a/drivers/gpu/drm/solomon/ssd130x.c
+++ b/drivers/gpu/drm/solomon/ssd130x.c
@@ -319,7 +319,7 @@ static int ssd130x_pwm_enable(struct ssd130x_device *ssd130x)
 
 	pwm_init_state(ssd130x->pwm, &pwmstate);
 	pwm_set_relative_duty_cycle(&pwmstate, 50, 100);
-	pwm_apply_state(ssd130x->pwm, &pwmstate);
+	pwm_apply_might_sleep(ssd130x->pwm, &pwmstate);
 
 	/* Enable the PWM */
 	pwm_enable(ssd130x->pwm);
diff --git a/drivers/hwmon/pwm-fan.c b/drivers/hwmon/pwm-fan.c
index 6e4516c2ab89..b67bc9e833c0 100644
--- a/drivers/hwmon/pwm-fan.c
+++ b/drivers/hwmon/pwm-fan.c
@@ -151,7 +151,7 @@ static int pwm_fan_power_on(struct pwm_fan_ctx *ctx)
 	}
 
 	state->enabled = true;
-	ret = pwm_apply_state(ctx->pwm, state);
+	ret = pwm_apply_might_sleep(ctx->pwm, state);
 	if (ret) {
 		dev_err(ctx->dev, "failed to enable PWM\n");
 		goto disable_regulator;
@@ -181,7 +181,7 @@ static int pwm_fan_power_off(struct pwm_fan_ctx *ctx)
 
 	state->enabled = false;
 	state->duty_cycle = 0;
-	ret = pwm_apply_state(ctx->pwm, state);
+	ret = pwm_apply_might_sleep(ctx->pwm, state);
 	if (ret) {
 		dev_err(ctx->dev, "failed to disable PWM\n");
 		return ret;
@@ -207,7 +207,7 @@ static int  __set_pwm(struct pwm_fan_ctx *ctx, unsigned long pwm)
 
 		period = state->period;
 		state->duty_cycle = DIV_ROUND_UP(pwm * (period - 1), MAX_PWM);
-		ret = pwm_apply_state(ctx->pwm, state);
+		ret = pwm_apply_might_sleep(ctx->pwm, state);
 		if (ret)
 			return ret;
 		ret = pwm_fan_power_on(ctx);
@@ -278,7 +278,7 @@ static int pwm_fan_update_enable(struct pwm_fan_ctx *ctx, long val)
 						    state,
 						    &enable_regulator);
 
-			pwm_apply_state(ctx->pwm, state);
+			pwm_apply_might_sleep(ctx->pwm, state);
 			pwm_fan_switch_power(ctx, enable_regulator);
 			pwm_fan_update_state(ctx, 0);
 		}
diff --git a/drivers/input/misc/da7280.c b/drivers/input/misc/da7280.c
index ce82548916bb..c1fa75c0f970 100644
--- a/drivers/input/misc/da7280.c
+++ b/drivers/input/misc/da7280.c
@@ -352,7 +352,7 @@ static int da7280_haptic_set_pwm(struct da7280_haptic *haptics, bool enabled)
 		state.duty_cycle = period_mag_multi;
 	}
 
-	error = pwm_apply_state(haptics->pwm_dev, &state);
+	error = pwm_apply_might_sleep(haptics->pwm_dev, &state);
 	if (error)
 		dev_err(haptics->dev, "Failed to apply pwm state: %d\n", error);
 
@@ -1175,7 +1175,7 @@ static int da7280_probe(struct i2c_client *client)
 		/* Sync up PWM state and ensure it is off. */
 		pwm_init_state(haptics->pwm_dev, &state);
 		state.enabled = false;
-		error = pwm_apply_state(haptics->pwm_dev, &state);
+		error = pwm_apply_might_sleep(haptics->pwm_dev, &state);
 		if (error) {
 			dev_err(dev, "Failed to apply PWM state: %d\n", error);
 			return error;
diff --git a/drivers/input/misc/pwm-beeper.c b/drivers/input/misc/pwm-beeper.c
index 1e731d8397c6..5b9aedf4362f 100644
--- a/drivers/input/misc/pwm-beeper.c
+++ b/drivers/input/misc/pwm-beeper.c
@@ -39,7 +39,7 @@ static int pwm_beeper_on(struct pwm_beeper *beeper, unsigned long period)
 	state.period = period;
 	pwm_set_relative_duty_cycle(&state, 50, 100);
 
-	error = pwm_apply_state(beeper->pwm, &state);
+	error = pwm_apply_might_sleep(beeper->pwm, &state);
 	if (error)
 		return error;
 
@@ -138,7 +138,7 @@ static int pwm_beeper_probe(struct platform_device *pdev)
 	/* Sync up PWM state and ensure it is off. */
 	pwm_init_state(beeper->pwm, &state);
 	state.enabled = false;
-	error = pwm_apply_state(beeper->pwm, &state);
+	error = pwm_apply_might_sleep(beeper->pwm, &state);
 	if (error) {
 		dev_err(dev, "failed to apply initial PWM state: %d\n",
 			error);
diff --git a/drivers/input/misc/pwm-vibra.c b/drivers/input/misc/pwm-vibra.c
index acac79c488aa..3e5ed685ed8f 100644
--- a/drivers/input/misc/pwm-vibra.c
+++ b/drivers/input/misc/pwm-vibra.c
@@ -56,7 +56,7 @@ static int pwm_vibrator_start(struct pwm_vibrator *vibrator)
 	pwm_set_relative_duty_cycle(&state, vibrator->level, 0xffff);
 	state.enabled = true;
 
-	err = pwm_apply_state(vibrator->pwm, &state);
+	err = pwm_apply_might_sleep(vibrator->pwm, &state);
 	if (err) {
 		dev_err(pdev, "failed to apply pwm state: %d\n", err);
 		return err;
@@ -67,7 +67,7 @@ static int pwm_vibrator_start(struct pwm_vibrator *vibrator)
 		state.duty_cycle = vibrator->direction_duty_cycle;
 		state.enabled = true;
 
-		err = pwm_apply_state(vibrator->pwm_dir, &state);
+		err = pwm_apply_might_sleep(vibrator->pwm_dir, &state);
 		if (err) {
 			dev_err(pdev, "failed to apply dir-pwm state: %d\n", err);
 			pwm_disable(vibrator->pwm);
@@ -160,7 +160,7 @@ static int pwm_vibrator_probe(struct platform_device *pdev)
 	/* Sync up PWM state and ensure it is off. */
 	pwm_init_state(vibrator->pwm, &state);
 	state.enabled = false;
-	err = pwm_apply_state(vibrator->pwm, &state);
+	err = pwm_apply_might_sleep(vibrator->pwm, &state);
 	if (err) {
 		dev_err(&pdev->dev, "failed to apply initial PWM state: %d\n",
 			err);
@@ -174,7 +174,7 @@ static int pwm_vibrator_probe(struct platform_device *pdev)
 		/* Sync up PWM state and ensure it is off. */
 		pwm_init_state(vibrator->pwm_dir, &state);
 		state.enabled = false;
-		err = pwm_apply_state(vibrator->pwm_dir, &state);
+		err = pwm_apply_might_sleep(vibrator->pwm_dir, &state);
 		if (err) {
 			dev_err(&pdev->dev, "failed to apply initial PWM state: %d\n",
 				err);
diff --git a/drivers/leds/leds-pwm.c b/drivers/leds/leds-pwm.c
index 2b3bf1353b70..4e3936a39d0e 100644
--- a/drivers/leds/leds-pwm.c
+++ b/drivers/leds/leds-pwm.c
@@ -54,7 +54,7 @@ static int led_pwm_set(struct led_classdev *led_cdev,
 
 	led_dat->pwmstate.duty_cycle = duty;
 	led_dat->pwmstate.enabled = true;
-	return pwm_apply_state(led_dat->pwm, &led_dat->pwmstate);
+	return pwm_apply_might_sleep(led_dat->pwm, &led_dat->pwmstate);
 }
 
 __attribute__((nonnull))
diff --git a/drivers/leds/rgb/leds-pwm-multicolor.c b/drivers/leds/rgb/leds-pwm-multicolor.c
index 46cd062b8b24..e1a81e0109e8 100644
--- a/drivers/leds/rgb/leds-pwm-multicolor.c
+++ b/drivers/leds/rgb/leds-pwm-multicolor.c
@@ -51,8 +51,8 @@ static int led_pwm_mc_set(struct led_classdev *cdev,
 
 		priv->leds[i].state.duty_cycle = duty;
 		priv->leds[i].state.enabled = duty > 0;
-		ret = pwm_apply_state(priv->leds[i].pwm,
-				      &priv->leds[i].state);
+		ret = pwm_apply_might_sleep(priv->leds[i].pwm,
+					    &priv->leds[i].state);
 		if (ret)
 			break;
 	}
diff --git a/drivers/media/rc/pwm-ir-tx.c b/drivers/media/rc/pwm-ir-tx.c
index c5f37c03af9c..cf51e2760975 100644
--- a/drivers/media/rc/pwm-ir-tx.c
+++ b/drivers/media/rc/pwm-ir-tx.c
@@ -68,7 +68,7 @@ static int pwm_ir_tx(struct rc_dev *dev, unsigned int *txbuf,
 
 	for (i = 0; i < count; i++) {
 		state.enabled = !(i % 2);
-		pwm_apply_state(pwm, &state);
+		pwm_apply_might_sleep(pwm, &state);
 
 		edge = ktime_add_us(edge, txbuf[i]);
 		delta = ktime_us_delta(edge, ktime_get());
@@ -77,7 +77,7 @@ static int pwm_ir_tx(struct rc_dev *dev, unsigned int *txbuf,
 	}
 
 	state.enabled = false;
-	pwm_apply_state(pwm, &state);
+	pwm_apply_might_sleep(pwm, &state);
 
 	return count;
 }
diff --git a/drivers/platform/x86/lenovo-yogabook.c b/drivers/platform/x86/lenovo-yogabook.c
index b8d0239192cb..fd62bf746ebd 100644
--- a/drivers/platform/x86/lenovo-yogabook.c
+++ b/drivers/platform/x86/lenovo-yogabook.c
@@ -435,7 +435,7 @@ static int yogabook_pdev_set_kbd_backlight(struct yogabook_data *data, u8 level)
 		.enabled = level,
 	};
 
-	pwm_apply_state(data->kbd_bl_pwm, &state);
+	pwm_apply_might_sleep(data->kbd_bl_pwm, &state);
 	gpiod_set_value(data->kbd_bl_led_enable, level ? 1 : 0);
 	return 0;
 }
diff --git a/drivers/pwm/core.c b/drivers/pwm/core.c
index 29078486534d..fc92ba622e56 100644
--- a/drivers/pwm/core.c
+++ b/drivers/pwm/core.c
@@ -356,8 +356,8 @@ struct pwm_device *pwm_request_from_chip(struct pwm_chip *chip,
 }
 EXPORT_SYMBOL_GPL(pwm_request_from_chip);
 
-static void pwm_apply_state_debug(struct pwm_device *pwm,
-				  const struct pwm_state *state)
+static void pwm_apply_debug(struct pwm_device *pwm,
+			    const struct pwm_state *state)
 {
 	struct pwm_state *last = &pwm->last;
 	struct pwm_chip *chip = pwm->chip;
@@ -463,11 +463,11 @@ static void pwm_apply_state_debug(struct pwm_device *pwm,
 }
 
 /**
- * pwm_apply_state() - atomically apply a new state to a PWM device
+ * pwm_apply_might_sleep() - atomically apply a new state to a PWM device
  * @pwm: PWM device
  * @state: new state to apply
  */
-int pwm_apply_state(struct pwm_device *pwm, const struct pwm_state *state)
+int pwm_apply_might_sleep(struct pwm_device *pwm, const struct pwm_state *state)
 {
 	struct pwm_chip *chip;
 	int err;
@@ -475,7 +475,7 @@ int pwm_apply_state(struct pwm_device *pwm, const struct pwm_state *state)
 	/*
 	 * Some lowlevel driver's implementations of .apply() make use of
 	 * mutexes, also with some drivers only returning when the new
-	 * configuration is active calling pwm_apply_state() from atomic context
+	 * configuration is active calling pwm_apply_might_sleep() from atomic context
 	 * is a bad idea. So make it explicit that calling this function might
 	 * sleep.
 	 */
@@ -505,11 +505,11 @@ int pwm_apply_state(struct pwm_device *pwm, const struct pwm_state *state)
 	 * only do this after pwm->state was applied as some
 	 * implementations of .get_state depend on this
 	 */
-	pwm_apply_state_debug(pwm, state);
+	pwm_apply_debug(pwm, state);
 
 	return 0;
 }
-EXPORT_SYMBOL_GPL(pwm_apply_state);
+EXPORT_SYMBOL_GPL(pwm_apply_might_sleep);
 
 /**
  * pwm_capture() - capture and report a PWM signal
@@ -567,7 +567,7 @@ int pwm_adjust_config(struct pwm_device *pwm)
 		state.period = pargs.period;
 		state.polarity = pargs.polarity;
 
-		return pwm_apply_state(pwm, &state);
+		return pwm_apply_might_sleep(pwm, &state);
 	}
 
 	/*
@@ -590,7 +590,7 @@ int pwm_adjust_config(struct pwm_device *pwm)
 		state.duty_cycle = state.period - state.duty_cycle;
 	}
 
-	return pwm_apply_state(pwm, &state);
+	return pwm_apply_might_sleep(pwm, &state);
 }
 EXPORT_SYMBOL_GPL(pwm_adjust_config);
 
diff --git a/drivers/pwm/pwm-twl-led.c b/drivers/pwm/pwm-twl-led.c
index 625233f4703a..70eec1c66747 100644
--- a/drivers/pwm/pwm-twl-led.c
+++ b/drivers/pwm/pwm-twl-led.c
@@ -172,7 +172,7 @@ static int twl4030_pwmled_apply(struct pwm_chip *chip, struct pwm_device *pwm,
 	 * We cannot skip calling ->config even if state->period ==
 	 * pwm->state.period && state->duty_cycle == pwm->state.duty_cycle
 	 * because we might have exited early in the last call to
-	 * pwm_apply_state because of !state->enabled and so the two values in
+	 * pwm_apply_might_sleep because of !state->enabled and so the two values in
 	 * pwm->state might not be configured in hardware.
 	 */
 	ret = twl4030_pwmled_config(pwm->chip, pwm,
diff --git a/drivers/pwm/pwm-vt8500.c b/drivers/pwm/pwm-vt8500.c
index 5568d5312d3c..3d1d11062558 100644
--- a/drivers/pwm/pwm-vt8500.c
+++ b/drivers/pwm/pwm-vt8500.c
@@ -206,7 +206,7 @@ static int vt8500_pwm_apply(struct pwm_chip *chip, struct pwm_device *pwm,
 	 * We cannot skip calling ->config even if state->period ==
 	 * pwm->state.period && state->duty_cycle == pwm->state.duty_cycle
 	 * because we might have exited early in the last call to
-	 * pwm_apply_state because of !state->enabled and so the two values in
+	 * pwm_apply_might_sleep because of !state->enabled and so the two values in
 	 * pwm->state might not be configured in hardware.
 	 */
 	err = vt8500_pwm_config(pwm->chip, pwm, state->duty_cycle, state->period);
diff --git a/drivers/pwm/sysfs.c b/drivers/pwm/sysfs.c
index 8d1254761e4d..052ccadbdabf 100644
--- a/drivers/pwm/sysfs.c
+++ b/drivers/pwm/sysfs.c
@@ -62,7 +62,7 @@ static ssize_t period_store(struct device *child,
 	mutex_lock(&export->lock);
 	pwm_get_state(pwm, &state);
 	state.period = val;
-	ret = pwm_apply_state(pwm, &state);
+	ret = pwm_apply_might_sleep(pwm, &state);
 	mutex_unlock(&export->lock);
 
 	return ret ? : size;
@@ -97,7 +97,7 @@ static ssize_t duty_cycle_store(struct device *child,
 	mutex_lock(&export->lock);
 	pwm_get_state(pwm, &state);
 	state.duty_cycle = val;
-	ret = pwm_apply_state(pwm, &state);
+	ret = pwm_apply_might_sleep(pwm, &state);
 	mutex_unlock(&export->lock);
 
 	return ret ? : size;
@@ -144,7 +144,7 @@ static ssize_t enable_store(struct device *child,
 		goto unlock;
 	}
 
-	ret = pwm_apply_state(pwm, &state);
+	ret = pwm_apply_might_sleep(pwm, &state);
 
 unlock:
 	mutex_unlock(&export->lock);
@@ -194,7 +194,7 @@ static ssize_t polarity_store(struct device *child,
 	mutex_lock(&export->lock);
 	pwm_get_state(pwm, &state);
 	state.polarity = polarity;
-	ret = pwm_apply_state(pwm, &state);
+	ret = pwm_apply_might_sleep(pwm, &state);
 	mutex_unlock(&export->lock);
 
 	return ret ? : size;
@@ -401,7 +401,7 @@ static int pwm_class_apply_state(struct pwm_export *export,
 				 struct pwm_device *pwm,
 				 struct pwm_state *state)
 {
-	int ret = pwm_apply_state(pwm, state);
+	int ret = pwm_apply_might_sleep(pwm, state);
 
 	/* release lock taken in pwm_class_get_state */
 	mutex_unlock(&export->lock);
diff --git a/drivers/regulator/pwm-regulator.c b/drivers/regulator/pwm-regulator.c
index 2aff6db748e2..698c420e0869 100644
--- a/drivers/regulator/pwm-regulator.c
+++ b/drivers/regulator/pwm-regulator.c
@@ -90,7 +90,7 @@ static int pwm_regulator_set_voltage_sel(struct regulator_dev *rdev,
 	pwm_set_relative_duty_cycle(&pstate,
 			drvdata->duty_cycle_table[selector].dutycycle, 100);
 
-	ret = pwm_apply_state(drvdata->pwm, &pstate);
+	ret = pwm_apply_might_sleep(drvdata->pwm, &pstate);
 	if (ret) {
 		dev_err(&rdev->dev, "Failed to configure PWM: %d\n", ret);
 		return ret;
@@ -216,7 +216,7 @@ static int pwm_regulator_set_voltage(struct regulator_dev *rdev,
 
 	pwm_set_relative_duty_cycle(&pstate, dutycycle, duty_unit);
 
-	ret = pwm_apply_state(drvdata->pwm, &pstate);
+	ret = pwm_apply_might_sleep(drvdata->pwm, &pstate);
 	if (ret) {
 		dev_err(&rdev->dev, "Failed to configure PWM: %d\n", ret);
 		return ret;
diff --git a/drivers/video/backlight/lm3630a_bl.c b/drivers/video/backlight/lm3630a_bl.c
index 8fcb62be597b..a3412c936ca2 100644
--- a/drivers/video/backlight/lm3630a_bl.c
+++ b/drivers/video/backlight/lm3630a_bl.c
@@ -180,7 +180,7 @@ static int lm3630a_pwm_ctrl(struct lm3630a_chip *pchip, int br, int br_max)
 
 	pchip->pwmd_state.enabled = pchip->pwmd_state.duty_cycle ? true : false;
 
-	return pwm_apply_state(pchip->pwmd, &pchip->pwmd_state);
+	return pwm_apply_might_sleep(pchip->pwmd, &pchip->pwmd_state);
 }
 
 /* update and get brightness */
diff --git a/drivers/video/backlight/lp855x_bl.c b/drivers/video/backlight/lp855x_bl.c
index da1f124db69c..7075bfab59c4 100644
--- a/drivers/video/backlight/lp855x_bl.c
+++ b/drivers/video/backlight/lp855x_bl.c
@@ -234,7 +234,7 @@ static int lp855x_pwm_ctrl(struct lp855x *lp, int br, int max_br)
 	state.duty_cycle = div_u64(br * state.period, max_br);
 	state.enabled = state.duty_cycle;
 
-	return pwm_apply_state(lp->pwm, &state);
+	return pwm_apply_might_sleep(lp->pwm, &state);
 }
 
 static int lp855x_bl_update_status(struct backlight_device *bl)
diff --git a/drivers/video/backlight/pwm_bl.c b/drivers/video/backlight/pwm_bl.c
index 289bd9ce4d36..35c716e9043c 100644
--- a/drivers/video/backlight/pwm_bl.c
+++ b/drivers/video/backlight/pwm_bl.c
@@ -103,7 +103,7 @@ static int pwm_backlight_update_status(struct backlight_device *bl)
 		pwm_get_state(pb->pwm, &state);
 		state.duty_cycle = compute_duty_cycle(pb, brightness, &state);
 		state.enabled = true;
-		pwm_apply_state(pb->pwm, &state);
+		pwm_apply_might_sleep(pb->pwm, &state);
 
 		pwm_backlight_power_on(pb);
 	} else {
@@ -120,7 +120,7 @@ static int pwm_backlight_update_status(struct backlight_device *bl)
 		 * inactive output.
 		 */
 		state.enabled = !pb->power_supply && !pb->enable_gpio;
-		pwm_apply_state(pb->pwm, &state);
+		pwm_apply_might_sleep(pb->pwm, &state);
 	}
 
 	if (pb->notify_after)
@@ -528,7 +528,7 @@ static int pwm_backlight_probe(struct platform_device *pdev)
 	if (!state.period && (data->pwm_period_ns > 0))
 		state.period = data->pwm_period_ns;
 
-	ret = pwm_apply_state(pb->pwm, &state);
+	ret = pwm_apply_might_sleep(pb->pwm, &state);
 	if (ret) {
 		dev_err(&pdev->dev, "failed to apply initial PWM state: %d\n",
 			ret);
@@ -633,7 +633,7 @@ static void pwm_backlight_remove(struct platform_device *pdev)
 	pwm_get_state(pb->pwm, &state);
 	state.duty_cycle = 0;
 	state.enabled = false;
-	pwm_apply_state(pb->pwm, &state);
+	pwm_apply_might_sleep(pb->pwm, &state);
 
 	if (pb->exit)
 		pb->exit(&pdev->dev);
@@ -649,7 +649,7 @@ static void pwm_backlight_shutdown(struct platform_device *pdev)
 	pwm_get_state(pb->pwm, &state);
 	state.duty_cycle = 0;
 	state.enabled = false;
-	pwm_apply_state(pb->pwm, &state);
+	pwm_apply_might_sleep(pb->pwm, &state);
 }
 
 #ifdef CONFIG_PM_SLEEP
@@ -673,7 +673,7 @@ static int pwm_backlight_suspend(struct device *dev)
 	pwm_get_state(pb->pwm, &state);
 	state.duty_cycle = 0;
 	state.enabled = false;
-	pwm_apply_state(pb->pwm, &state);
+	pwm_apply_might_sleep(pb->pwm, &state);
 
 	if (pb->notify_after)
 		pb->notify_after(pb->dev, 0);
diff --git a/drivers/video/fbdev/ssd1307fb.c b/drivers/video/fbdev/ssd1307fb.c
index 5ae48e36fccb..1a4f90ea7d5a 100644
--- a/drivers/video/fbdev/ssd1307fb.c
+++ b/drivers/video/fbdev/ssd1307fb.c
@@ -347,7 +347,7 @@ static int ssd1307fb_init(struct ssd1307fb_par *par)
 
 		pwm_init_state(par->pwm, &pwmstate);
 		pwm_set_relative_duty_cycle(&pwmstate, 50, 100);
-		pwm_apply_state(par->pwm, &pwmstate);
+		pwm_apply_might_sleep(par->pwm, &pwmstate);
 
 		/* Enable the PWM */
 		pwm_enable(par->pwm);
diff --git a/include/linux/pwm.h b/include/linux/pwm.h
index cda3597b84f2..5c5c456948a4 100644
--- a/include/linux/pwm.h
+++ b/include/linux/pwm.h
@@ -93,8 +93,8 @@ struct pwm_device {
  * @state: state to fill with the current PWM state
  *
  * The returned PWM state represents the state that was applied by a previous call to
- * pwm_apply_state(). Drivers may have to slightly tweak that state before programming it to
- * hardware. If pwm_apply_state() was never called, this returns either the current hardware
+ * pwm_apply_might_sleep(). Drivers may have to slightly tweak that state before programming it to
+ * hardware. If pwm_apply_might_sleep() was never called, this returns either the current hardware
  * state (if supported) or the default settings.
  */
 static inline void pwm_get_state(const struct pwm_device *pwm,
@@ -158,20 +158,20 @@ static inline void pwm_get_args(const struct pwm_device *pwm,
 }
 
 /**
- * pwm_init_state() - prepare a new state to be applied with pwm_apply_state()
+ * pwm_init_state() - prepare a new state to be applied with pwm_apply_might_sleep()
  * @pwm: PWM device
  * @state: state to fill with the prepared PWM state
  *
  * This functions prepares a state that can later be tweaked and applied
- * to the PWM device with pwm_apply_state(). This is a convenient function
+ * to the PWM device with pwm_apply_might_sleep(). This is a convenient function
  * that first retrieves the current PWM state and the replaces the period
  * and polarity fields with the reference values defined in pwm->args.
  * Once the function returns, you can adjust the ->enabled and ->duty_cycle
- * fields according to your needs before calling pwm_apply_state().
+ * fields according to your needs before calling pwm_apply_might_sleep().
  *
  * ->duty_cycle is initially set to zero to avoid cases where the current
  * ->duty_cycle value exceed the pwm_args->period one, which would trigger
- * an error if the user calls pwm_apply_state() without adjusting ->duty_cycle
+ * an error if the user calls pwm_apply_might_sleep() without adjusting ->duty_cycle
  * first.
  */
 static inline void pwm_init_state(const struct pwm_device *pwm,
@@ -227,7 +227,7 @@ pwm_get_relative_duty_cycle(const struct pwm_state *state, unsigned int scale)
  *
  * pwm_init_state(pwm, &state);
  * pwm_set_relative_duty_cycle(&state, 50, 100);
- * pwm_apply_state(pwm, &state);
+ * pwm_apply_might_sleep(pwm, &state);
  *
  * This functions returns -EINVAL if @duty_cycle and/or @scale are
  * inconsistent (@scale == 0 or @duty_cycle > @scale).
@@ -307,7 +307,7 @@ struct pwm_chip {
 
 #if IS_ENABLED(CONFIG_PWM)
 /* PWM user APIs */
-int pwm_apply_state(struct pwm_device *pwm, const struct pwm_state *state);
+int pwm_apply_might_sleep(struct pwm_device *pwm, const struct pwm_state *state);
 int pwm_adjust_config(struct pwm_device *pwm);
 
 /**
@@ -335,7 +335,7 @@ static inline int pwm_config(struct pwm_device *pwm, int duty_ns,
 
 	state.duty_cycle = duty_ns;
 	state.period = period_ns;
-	return pwm_apply_state(pwm, &state);
+	return pwm_apply_might_sleep(pwm, &state);
 }
 
 /**
@@ -356,7 +356,7 @@ static inline int pwm_enable(struct pwm_device *pwm)
 		return 0;
 
 	state.enabled = true;
-	return pwm_apply_state(pwm, &state);
+	return pwm_apply_might_sleep(pwm, &state);
 }
 
 /**
@@ -375,7 +375,7 @@ static inline void pwm_disable(struct pwm_device *pwm)
 		return;
 
 	state.enabled = false;
-	pwm_apply_state(pwm, &state);
+	pwm_apply_might_sleep(pwm, &state);
 }
 
 /* PWM provider APIs */
@@ -406,8 +406,8 @@ struct pwm_device *devm_fwnode_pwm_get(struct device *dev,
 				       struct fwnode_handle *fwnode,
 				       const char *con_id);
 #else
-static inline int pwm_apply_state(struct pwm_device *pwm,
-				  const struct pwm_state *state)
+static inline int pwm_apply_might_sleep(struct pwm_device *pwm,
+					const struct pwm_state *state)
 {
 	might_sleep();
 	return -ENOTSUPP;
@@ -524,7 +524,7 @@ static inline void pwm_apply_args(struct pwm_device *pwm)
 	state.period = pwm->args.period;
 	state.usage_power = false;
 
-	pwm_apply_state(pwm, &state);
+	pwm_apply_might_sleep(pwm, &state);
 }
 
 struct pwm_lookup {
-- 
2.43.0

Re: [PATCH v6 1/4] pwm: rename pwm_apply_state() to pwm_apply_might_sleep()

[Uwe Kleine-König]

[-- Attachment #1: Type: text/plain, Size: 976 bytes --]

On Wed, Nov 29, 2023 at 09:13:34AM +0000, Sean Young wrote:
> In order to introduce a pwm api which can be used from atomic context,
> we will need two functions for applying pwm changes:
> 
> 	int pwm_apply_might_sleep(struct pwm *, struct pwm_state *);
> 	int pwm_apply_atomic(struct pwm *, struct pwm_state *);
> 
> This commit just deals with renaming pwm_apply_state(), a following
> commit will introduce the pwm_apply_atomic() function.
> 
> Acked-by: Hans de Goede <hdegoede@redhat.com>
> Acked-by: Jani Nikula <jani.nikula@intel.com>
> Acked-by: Lee Jones <lee@kernel.org>
> Signed-off-by: Sean Young <sean@mess.org>

Not a in-detail-review, but I just noticed again, that we have

	K:      pwm_(config|apply_state|ops)

in MAINTAINERS. That one needs adaption, too.

Best regards
Uwe

-- 
Pengutronix e.K.                           | Uwe Kleine-König            |
Industrial Linux Solutions                 | https://www.pengutronix.de/ |

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]

Re: [PATCH v6 1/4] pwm: rename pwm_apply_state() to pwm_apply_might_sleep()

[Sean Young]

On Sat, Dec 09, 2023 at 02:57:42PM +0100, Uwe Kleine-König wrote:
> On Wed, Nov 29, 2023 at 09:13:34AM +0000, Sean Young wrote:
> > In order to introduce a pwm api which can be used from atomic context,
> > we will need two functions for applying pwm changes:
> > 
> > 	int pwm_apply_might_sleep(struct pwm *, struct pwm_state *);
> > 	int pwm_apply_atomic(struct pwm *, struct pwm_state *);
> > 
> > This commit just deals with renaming pwm_apply_state(), a following
> > commit will introduce the pwm_apply_atomic() function.
> > 
> > Acked-by: Hans de Goede <hdegoede@redhat.com>
> > Acked-by: Jani Nikula <jani.nikula@intel.com>
> > Acked-by: Lee Jones <lee@kernel.org>
> > Signed-off-by: Sean Young <sean@mess.org>
> 
> Not a in-detail-review, but I just noticed again, that we have
> 
> 	K:      pwm_(config|apply_state|ops)
> 
> in MAINTAINERS. That one needs adaption, too.

Fixed in v7.

Thanks,

Sean

Re: [PATCH v6 1/4] pwm: rename pwm_apply_state() to pwm_apply_might_sleep()

[Dmitry Torokhov]

On Wed, Nov 29, 2023 at 09:13:34AM +0000, Sean Young wrote:
>  drivers/input/misc/da7280.c                   |  4 +--
>  drivers/input/misc/pwm-beeper.c               |  4 +--
>  drivers/input/misc/pwm-vibra.c                |  8 +++---

Acked-by: Dmitry Torokhov <dmitry.torokhov@gmail.com> # for input

Thanks.

-- 
Dmitry

[PATCH v6 2/4] pwm: make it possible to apply pwm changes in atomic context

[Sean Young]

Some pwm devices require sleeping, for example if the pwm device is
connected over i2c. However, many pwm devices could be used from atomic
context, e.g. memmory mapped pwm. This is useful for, for example, the
pwm-ir-tx driver which requires precise timing. Sleeping causes havoc
with the generated IR signal.

Since not all pmw devices can support atomic context, we also add a
pwm_is_atomic() function to check if it is supported.

Signed-off-by: Sean Young <sean@mess.org>
---
 Documentation/driver-api/pwm.rst |  9 +++++
 drivers/pwm/core.c               | 63 ++++++++++++++++++++++++++------
 drivers/pwm/pwm-renesas-tpu.c    |  1 -
 include/linux/pwm.h              | 29 ++++++++++++++-
 4 files changed, 87 insertions(+), 15 deletions(-)

diff --git a/Documentation/driver-api/pwm.rst b/Documentation/driver-api/pwm.rst
index f1d8197c8c43..1d4536fdf47c 100644
--- a/Documentation/driver-api/pwm.rst
+++ b/Documentation/driver-api/pwm.rst
@@ -43,6 +43,15 @@ After being requested, a PWM has to be configured using::
 
 	int pwm_apply_might_sleep(struct pwm_device *pwm, struct pwm_state *state);
 
+Some PWM devices can be used from atomic context. You can check if this is
+supported with::
+
+        bool pwm_is_atomic(struct pwm_device *pwm);
+
+If true, the PWM can be configured from atomic context with::
+
+	int pwm_apply_atomic(struct pwm_device *pwm, struct pwm_state *state);
+
 This API controls both the PWM period/duty_cycle config and the
 enable/disable state.
 
diff --git a/drivers/pwm/core.c b/drivers/pwm/core.c
index fc92ba622e56..63174e207400 100644
--- a/drivers/pwm/core.c
+++ b/drivers/pwm/core.c
@@ -463,24 +463,15 @@ static void pwm_apply_debug(struct pwm_device *pwm,
 }
 
 /**
- * pwm_apply_might_sleep() - atomically apply a new state to a PWM device
+ * pwm_apply_unchecked() - atomically apply a new state to a PWM device
  * @pwm: PWM device
  * @state: new state to apply
  */
-int pwm_apply_might_sleep(struct pwm_device *pwm, const struct pwm_state *state)
+static int pwm_apply_unchecked(struct pwm_device *pwm, const struct pwm_state *state)
 {
 	struct pwm_chip *chip;
 	int err;
 
-	/*
-	 * Some lowlevel driver's implementations of .apply() make use of
-	 * mutexes, also with some drivers only returning when the new
-	 * configuration is active calling pwm_apply_might_sleep() from atomic context
-	 * is a bad idea. So make it explicit that calling this function might
-	 * sleep.
-	 */
-	might_sleep();
-
 	if (!pwm || !state || !state->period ||
 	    state->duty_cycle > state->period)
 		return -EINVAL;
@@ -501,16 +492,64 @@ int pwm_apply_might_sleep(struct pwm_device *pwm, const struct pwm_state *state)
 
 	pwm->state = *state;
 
+	return 0;
+}
+
+/**
+ * pwm_apply_might_sleep() - atomically apply a new state to a PWM device
+ * Cannot be used in atomic context.
+ * @pwm: PWM device
+ * @state: new state to apply
+ */
+int pwm_apply_might_sleep(struct pwm_device *pwm, const struct pwm_state *state)
+{
+	int err;
+
+	/*
+	 * Some lowlevel driver's implementations of .apply() make use of
+	 * mutexes, also with some drivers only returning when the new
+	 * configuration is active calling pwm_apply_might_sleep() from atomic context
+	 * is a bad idea. So make it explicit that calling this function might
+	 * sleep.
+	 */
+	might_sleep();
+
+	if (IS_ENABLED(CONFIG_PWM_DEBUG) && pwm->chip->atomic) {
+		/*
+		 * Catch any sleeping drivers when atomic is set.
+		 */
+		non_block_start();
+		err = pwm_apply_unchecked(pwm, state);
+		non_block_end();
+	} else {
+		err = pwm_apply_unchecked(pwm, 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;
+	return err;
 }
 EXPORT_SYMBOL_GPL(pwm_apply_might_sleep);
 
+/**
+ * pwm_apply_atomic() - apply a new state to a PWM device from atomic context
+ * Not all pwm devices support this function, check with pwm_is_atomic().
+ * @pwm: PWM device
+ * @state: new state to apply
+ */
+int pwm_apply_atomic(struct pwm_device *pwm, const struct pwm_state *state)
+{
+	WARN_ONCE(!pwm->chip->atomic,
+		  "sleeping pwm driver used in atomic context");
+
+	return pwm_apply_unchecked(pwm, state);
+}
+EXPORT_SYMBOL_GPL(pwm_apply_atomic);
+
 /**
  * pwm_capture() - capture and report a PWM signal
  * @pwm: PWM device
diff --git a/drivers/pwm/pwm-renesas-tpu.c b/drivers/pwm/pwm-renesas-tpu.c
index 4239f2c3e8b2..47ea92cd8c67 100644
--- a/drivers/pwm/pwm-renesas-tpu.c
+++ b/drivers/pwm/pwm-renesas-tpu.c
@@ -11,7 +11,6 @@
 #include <linux/init.h>
 #include <linux/ioport.h>
 #include <linux/module.h>
-#include <linux/mutex.h>
 #include <linux/of.h>
 #include <linux/platform_device.h>
 #include <linux/pm_runtime.h>
diff --git a/include/linux/pwm.h b/include/linux/pwm.h
index 5c5c456948a4..f1fa1243e95a 100644
--- a/include/linux/pwm.h
+++ b/include/linux/pwm.h
@@ -286,6 +286,7 @@ struct pwm_ops {
  * @npwm: number of PWMs controlled by this chip
  * @of_xlate: request a PWM device given a device tree PWM specifier
  * @of_pwm_n_cells: number of cells expected in the device tree PWM specifier
+ * @atomic: can the driver call pwm_apply_atomic in atomic context
  * @list: list node for internal use
  * @pwms: array of PWM devices allocated by the framework
  */
@@ -299,6 +300,7 @@ struct pwm_chip {
 	struct pwm_device * (*of_xlate)(struct pwm_chip *chip,
 					const struct of_phandle_args *args);
 	unsigned int of_pwm_n_cells;
+	bool atomic;
 
 	/* only used internally by the PWM framework */
 	struct list_head list;
@@ -308,6 +310,7 @@ struct pwm_chip {
 #if IS_ENABLED(CONFIG_PWM)
 /* PWM user APIs */
 int pwm_apply_might_sleep(struct pwm_device *pwm, const struct pwm_state *state);
+int pwm_apply_atomic(struct pwm_device *pwm, const struct pwm_state *state);
 int pwm_adjust_config(struct pwm_device *pwm);
 
 /**
@@ -378,6 +381,17 @@ static inline void pwm_disable(struct pwm_device *pwm)
 	pwm_apply_might_sleep(pwm, &state);
 }
 
+/**
+ * pwm_is_atomic() - is pwm_apply_atomic() supported?
+ * @pwm: PWM device
+ *
+ * Returns: true pwm_apply_atomic() can be called from atomic context.
+ */
+static inline bool pwm_is_atomic(struct pwm_device *pwm)
+{
+	return pwm->chip->atomic;
+}
+
 /* PWM provider APIs */
 int pwm_capture(struct pwm_device *pwm, struct pwm_capture *result,
 		unsigned long timeout);
@@ -406,16 +420,27 @@ struct pwm_device *devm_fwnode_pwm_get(struct device *dev,
 				       struct fwnode_handle *fwnode,
 				       const char *con_id);
 #else
+static inline bool pwm_is_atomic(struct pwm_device *pwm)
+{
+	return false;
+}
+
 static inline int pwm_apply_might_sleep(struct pwm_device *pwm,
 					const struct pwm_state *state)
 {
 	might_sleep();
-	return -ENOTSUPP;
+	return -EOPNOTSUPP;
+}
+
+static inline int pwm_apply_atomic(struct pwm_device *pwm,
+				   const struct pwm_state *state)
+{
+	return -EOPNOTSUPP;
 }
 
 static inline int pwm_adjust_config(struct pwm_device *pwm)
 {
-	return -ENOTSUPP;
+	return -EOPNOTSUPP;
 }
 
 static inline int pwm_config(struct pwm_device *pwm, int duty_ns,
-- 
2.43.0

Re: [PATCH v6 2/4] pwm: make it possible to apply pwm changes in atomic context

[Thierry Reding]

[-- Attachment #1: Type: text/plain, Size: 8261 bytes --]

On Wed, Nov 29, 2023 at 09:13:35AM +0000, Sean Young wrote:
> Some pwm devices require sleeping, for example if the pwm device is
> connected over i2c. However, many pwm devices could be used from atomic
> context, e.g. memmory mapped pwm. This is useful for, for example, the
> pwm-ir-tx driver which requires precise timing. Sleeping causes havoc
> with the generated IR signal.
> 
> Since not all pmw devices can support atomic context, we also add a
> pwm_is_atomic() function to check if it is supported.

s/i2c/I2C/ and s/pwm/PWM/ in the above. Also, s/memmory/memory/

> 
> Signed-off-by: Sean Young <sean@mess.org>
> ---
>  Documentation/driver-api/pwm.rst |  9 +++++
>  drivers/pwm/core.c               | 63 ++++++++++++++++++++++++++------
>  drivers/pwm/pwm-renesas-tpu.c    |  1 -
>  include/linux/pwm.h              | 29 ++++++++++++++-
>  4 files changed, 87 insertions(+), 15 deletions(-)
> 
> diff --git a/Documentation/driver-api/pwm.rst b/Documentation/driver-api/pwm.rst
> index f1d8197c8c43..1d4536fdf47c 100644
> --- a/Documentation/driver-api/pwm.rst
> +++ b/Documentation/driver-api/pwm.rst
> @@ -43,6 +43,15 @@ After being requested, a PWM has to be configured using::
>  
>  	int pwm_apply_might_sleep(struct pwm_device *pwm, struct pwm_state *state);
>  
> +Some PWM devices can be used from atomic context. You can check if this is
> +supported with::
> +
> +        bool pwm_is_atomic(struct pwm_device *pwm);

This is now going to look a bit odd. I think it'd be best to change this
to pwm_might_sleep() for consistency with the pwm_apply_might_sleep()
function. Fine to keep the actual member variable as atomic, though, so
we don't have to change the default for all drivers.

> +
> +If true, the PWM can be configured from atomic context with::
> +
> +	int pwm_apply_atomic(struct pwm_device *pwm, struct pwm_state *state);
> +
>  This API controls both the PWM period/duty_cycle config and the
>  enable/disable state.
>  
> diff --git a/drivers/pwm/core.c b/drivers/pwm/core.c
> index fc92ba622e56..63174e207400 100644
> --- a/drivers/pwm/core.c
> +++ b/drivers/pwm/core.c
> @@ -463,24 +463,15 @@ static void pwm_apply_debug(struct pwm_device *pwm,
>  }
>  
>  /**
> - * pwm_apply_might_sleep() - atomically apply a new state to a PWM device
> + * pwm_apply_unchecked() - atomically apply a new state to a PWM device
>   * @pwm: PWM device
>   * @state: new state to apply
>   */
> -int pwm_apply_might_sleep(struct pwm_device *pwm, const struct pwm_state *state)
> +static int pwm_apply_unchecked(struct pwm_device *pwm, const struct pwm_state *state)
>  {
>  	struct pwm_chip *chip;
>  	int err;
>  
> -	/*
> -	 * Some lowlevel driver's implementations of .apply() make use of
> -	 * mutexes, also with some drivers only returning when the new
> -	 * configuration is active calling pwm_apply_might_sleep() from atomic context
> -	 * is a bad idea. So make it explicit that calling this function might
> -	 * sleep.
> -	 */
> -	might_sleep();
> -
>  	if (!pwm || !state || !state->period ||
>  	    state->duty_cycle > state->period)
>  		return -EINVAL;
> @@ -501,16 +492,64 @@ int pwm_apply_might_sleep(struct pwm_device *pwm, const struct pwm_state *state)
>  
>  	pwm->state = *state;
>  
> +	return 0;
> +}
> +
> +/**
> + * pwm_apply_might_sleep() - atomically apply a new state to a PWM device
> + * Cannot be used in atomic context.
> + * @pwm: PWM device
> + * @state: new state to apply
> + */
> +int pwm_apply_might_sleep(struct pwm_device *pwm, const struct pwm_state *state)
> +{
> +	int err;
> +
> +	/*
> +	 * Some lowlevel driver's implementations of .apply() make use of
> +	 * mutexes, also with some drivers only returning when the new
> +	 * configuration is active calling pwm_apply_might_sleep() from atomic context
> +	 * is a bad idea. So make it explicit that calling this function might
> +	 * sleep.
> +	 */
> +	might_sleep();
> +
> +	if (IS_ENABLED(CONFIG_PWM_DEBUG) && pwm->chip->atomic) {
> +		/*
> +		 * Catch any sleeping drivers when atomic is set.

I think this is slightly confusing. What we're really trying to catch
here is drivers that are marked as "atomic" but which still end up
sleeping during ->apply(), right? So maybe something like this would be
a bit more explicit:

		/*
		 * Catch any drivers that have been marked as atomic but
		 * that will sleep anyway.
		 */

> +/**
> + * pwm_apply_atomic() - apply a new state to a PWM device from atomic context
> + * Not all pwm devices support this function, check with pwm_is_atomic().

s/pwm/PWM/ here...

> + * @pwm: PWM device
> + * @state: new state to apply
> + */
> +int pwm_apply_atomic(struct pwm_device *pwm, const struct pwm_state *state)
> +{
> +	WARN_ONCE(!pwm->chip->atomic,
> +		  "sleeping pwm driver used in atomic context");

... and in the warning message as well.

> +
> +	return pwm_apply_unchecked(pwm, state);
> +}
> +EXPORT_SYMBOL_GPL(pwm_apply_atomic);
> +
>  /**
>   * pwm_capture() - capture and report a PWM signal
>   * @pwm: PWM device
> diff --git a/drivers/pwm/pwm-renesas-tpu.c b/drivers/pwm/pwm-renesas-tpu.c
> index 4239f2c3e8b2..47ea92cd8c67 100644
> --- a/drivers/pwm/pwm-renesas-tpu.c
> +++ b/drivers/pwm/pwm-renesas-tpu.c
> @@ -11,7 +11,6 @@
>  #include <linux/init.h>
>  #include <linux/ioport.h>
>  #include <linux/module.h>
> -#include <linux/mutex.h>
>  #include <linux/of.h>
>  #include <linux/platform_device.h>
>  #include <linux/pm_runtime.h>
> diff --git a/include/linux/pwm.h b/include/linux/pwm.h
> index 5c5c456948a4..f1fa1243e95a 100644
> --- a/include/linux/pwm.h
> +++ b/include/linux/pwm.h
> @@ -286,6 +286,7 @@ struct pwm_ops {
>   * @npwm: number of PWMs controlled by this chip
>   * @of_xlate: request a PWM device given a device tree PWM specifier
>   * @of_pwm_n_cells: number of cells expected in the device tree PWM specifier
> + * @atomic: can the driver call pwm_apply_atomic in atomic context

Maybe: "can the driver's ->apply() be called in atomic context"?

>   * @list: list node for internal use
>   * @pwms: array of PWM devices allocated by the framework
>   */
> @@ -299,6 +300,7 @@ struct pwm_chip {
>  	struct pwm_device * (*of_xlate)(struct pwm_chip *chip,
>  					const struct of_phandle_args *args);
>  	unsigned int of_pwm_n_cells;
> +	bool atomic;
>  
>  	/* only used internally by the PWM framework */
>  	struct list_head list;
> @@ -308,6 +310,7 @@ struct pwm_chip {
>  #if IS_ENABLED(CONFIG_PWM)
>  /* PWM user APIs */
>  int pwm_apply_might_sleep(struct pwm_device *pwm, const struct pwm_state *state);
> +int pwm_apply_atomic(struct pwm_device *pwm, const struct pwm_state *state);
>  int pwm_adjust_config(struct pwm_device *pwm);
>  
>  /**
> @@ -378,6 +381,17 @@ static inline void pwm_disable(struct pwm_device *pwm)
>  	pwm_apply_might_sleep(pwm, &state);
>  }
>  
> +/**
> + * pwm_is_atomic() - is pwm_apply_atomic() supported?
> + * @pwm: PWM device
> + *
> + * Returns: true pwm_apply_atomic() can be called from atomic context.
> + */
> +static inline bool pwm_is_atomic(struct pwm_device *pwm)

As I mentioned above, I think it'd be more consistent to call this
pwm_might_sleep().

> +{
> +	return pwm->chip->atomic;
> +}
> +
>  /* PWM provider APIs */
>  int pwm_capture(struct pwm_device *pwm, struct pwm_capture *result,
>  		unsigned long timeout);
> @@ -406,16 +420,27 @@ struct pwm_device *devm_fwnode_pwm_get(struct device *dev,
>  				       struct fwnode_handle *fwnode,
>  				       const char *con_id);
>  #else
> +static inline bool pwm_is_atomic(struct pwm_device *pwm)
> +{
> +	return false;
> +}
> +
>  static inline int pwm_apply_might_sleep(struct pwm_device *pwm,
>  					const struct pwm_state *state)
>  {
>  	might_sleep();
> -	return -ENOTSUPP;
> +	return -EOPNOTSUPP;
> +}
> +
> +static inline int pwm_apply_atomic(struct pwm_device *pwm,
> +				   const struct pwm_state *state)
> +{
> +	return -EOPNOTSUPP;
>  }
>  
>  static inline int pwm_adjust_config(struct pwm_device *pwm)
>  {
> -	return -ENOTSUPP;
> +	return -EOPNOTSUPP;
>  }

What's wrong with ENOTSUPP?

Thierry

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH v6 2/4] pwm: make it possible to apply pwm changes in atomic context

[Sean Young]

Hi Thierry,


On Fri, Dec 08, 2023 at 05:19:52PM +0100, Thierry Reding wrote:
> On Wed, Nov 29, 2023 at 09:13:35AM +0000, Sean Young wrote:
> > Some pwm devices require sleeping, for example if the pwm device is
> > connected over i2c. However, many pwm devices could be used from atomic
> > context, e.g. memmory mapped pwm. This is useful for, for example, the
> > pwm-ir-tx driver which requires precise timing. Sleeping causes havoc
> > with the generated IR signal.
> > 
> > Since not all pmw devices can support atomic context, we also add a
> > pwm_is_atomic() function to check if it is supported.
> 
> s/i2c/I2C/ and s/pwm/PWM/ in the above. Also, s/memmory/memory/

Thanks for your detailed review. I agree with all your points, they are
all nice improvements. Just a question at the bottom:

> 
> > 
> > Signed-off-by: Sean Young <sean@mess.org>
> > ---
> >  Documentation/driver-api/pwm.rst |  9 +++++
> >  drivers/pwm/core.c               | 63 ++++++++++++++++++++++++++------
> >  drivers/pwm/pwm-renesas-tpu.c    |  1 -
> >  include/linux/pwm.h              | 29 ++++++++++++++-
> >  4 files changed, 87 insertions(+), 15 deletions(-)
> > 
> > diff --git a/Documentation/driver-api/pwm.rst b/Documentation/driver-api/pwm.rst
> > index f1d8197c8c43..1d4536fdf47c 100644
> > --- a/Documentation/driver-api/pwm.rst
> > +++ b/Documentation/driver-api/pwm.rst
> > @@ -43,6 +43,15 @@ After being requested, a PWM has to be configured using::
> >  
> >  	int pwm_apply_might_sleep(struct pwm_device *pwm, struct pwm_state *state);
> >  
> > +Some PWM devices can be used from atomic context. You can check if this is
> > +supported with::
> > +
> > +        bool pwm_is_atomic(struct pwm_device *pwm);
> 
> This is now going to look a bit odd. I think it'd be best to change this
> to pwm_might_sleep() for consistency with the pwm_apply_might_sleep()
> function. Fine to keep the actual member variable as atomic, though, so
> we don't have to change the default for all drivers.

Agreed, I was struggling with finding good name and yours is much better,
thanks.

 > +{
> > +	return pwm->chip->atomic;
> > +}
> > +
> >  /* PWM provider APIs */
> >  int pwm_capture(struct pwm_device *pwm, struct pwm_capture *result,
> >  		unsigned long timeout);
> > @@ -406,16 +420,27 @@ struct pwm_device *devm_fwnode_pwm_get(struct device *dev,
> >  				       struct fwnode_handle *fwnode,
> >  				       const char *con_id);
> >  #else
> > +static inline bool pwm_is_atomic(struct pwm_device *pwm)
> > +{
> > +	return false;
> > +}
> > +
> >  static inline int pwm_apply_might_sleep(struct pwm_device *pwm,
> >  					const struct pwm_state *state)
> >  {
> >  	might_sleep();
> > -	return -ENOTSUPP;
> > +	return -EOPNOTSUPP;
> > +}
> > +
> > +static inline int pwm_apply_atomic(struct pwm_device *pwm,
> > +				   const struct pwm_state *state)
> > +{
> > +	return -EOPNOTSUPP;
> >  }
> >  
> >  static inline int pwm_adjust_config(struct pwm_device *pwm)
> >  {
> > -	return -ENOTSUPP;
> > +	return -EOPNOTSUPP;
> >  }
> 
> What's wrong with ENOTSUPP?

This was found by checkpatch, see

https://github.com/torvalds/linux/blob/master/scripts/checkpatch.pl#L4891-L4892

# ENOTSUPP is not a standard error code and should be avoided in new patches.
# Folks usually mean EOPNOTSUPP (also called ENOTSUP), when they type ENOTSUPP.

ENOTSUPP is not really widely used in the tree.

So it was really done to keep checkpatch happy, please let me know what you
would like me to do here.

Thanks,

Sean

[PATCH v6 3/4] pwm: bcm2835: allow pwm driver to be used in atomic context

[Sean Young]

clk_get_rate() may do a mutex lock. Fetch the clock rate once, and prevent
rate changes using clk_rate_exclusive_get().

Signed-off-by: Sean Young <sean@mess.org>
---
 drivers/pwm/pwm-bcm2835.c | 31 +++++++++++++++++++++----------
 1 file changed, 21 insertions(+), 10 deletions(-)

diff --git a/drivers/pwm/pwm-bcm2835.c b/drivers/pwm/pwm-bcm2835.c
index 9777babd5b95..52748194a3da 100644
--- a/drivers/pwm/pwm-bcm2835.c
+++ b/drivers/pwm/pwm-bcm2835.c
@@ -28,6 +28,7 @@ struct bcm2835_pwm {
 	struct device *dev;
 	void __iomem *base;
 	struct clk *clk;
+	unsigned long rate;
 };
 
 static inline struct bcm2835_pwm *to_bcm2835_pwm(struct pwm_chip *chip)
@@ -63,17 +64,11 @@ static int bcm2835_pwm_apply(struct pwm_chip *chip, struct pwm_device *pwm,
 {
 
 	struct bcm2835_pwm *pc = to_bcm2835_pwm(chip);
-	unsigned long rate = clk_get_rate(pc->clk);
 	unsigned long long period_cycles;
 	u64 max_period;
 
 	u32 val;
 
-	if (!rate) {
-		dev_err(pc->dev, "failed to get clock rate\n");
-		return -EINVAL;
-	}
-
 	/*
 	 * period_cycles must be a 32 bit value, so period * rate / NSEC_PER_SEC
 	 * must be <= U32_MAX. As U32_MAX * NSEC_PER_SEC < U64_MAX the
@@ -88,13 +83,13 @@ static int bcm2835_pwm_apply(struct pwm_chip *chip, struct pwm_device *pwm,
 	 * <=> period < ((U32_MAX * NSEC_PER_SEC + NSEC_PER_SEC/2) / rate
 	 * <=> period <= ceil((U32_MAX * NSEC_PER_SEC + NSEC_PER_SEC/2) / rate) - 1
 	 */
-	max_period = DIV_ROUND_UP_ULL((u64)U32_MAX * NSEC_PER_SEC + NSEC_PER_SEC / 2, rate) - 1;
+	max_period = DIV_ROUND_UP_ULL((u64)U32_MAX * NSEC_PER_SEC + NSEC_PER_SEC / 2, pc->rate) - 1;
 
 	if (state->period > max_period)
 		return -EINVAL;
 
 	/* set period */
-	period_cycles = DIV_ROUND_CLOSEST_ULL(state->period * rate, NSEC_PER_SEC);
+	period_cycles = DIV_ROUND_CLOSEST_ULL(state->period * pc->rate, NSEC_PER_SEC);
 
 	/* don't accept a period that is too small */
 	if (period_cycles < PERIOD_MIN)
@@ -103,7 +98,7 @@ static int bcm2835_pwm_apply(struct pwm_chip *chip, struct pwm_device *pwm,
 	writel(period_cycles, pc->base + PERIOD(pwm->hwpwm));
 
 	/* set duty cycle */
-	val = DIV_ROUND_CLOSEST_ULL(state->duty_cycle * rate, NSEC_PER_SEC);
+	val = DIV_ROUND_CLOSEST_ULL(state->duty_cycle * pc->rate, NSEC_PER_SEC);
 	writel(val, pc->base + DUTY(pwm->hwpwm));
 
 	/* set polarity */
@@ -151,14 +146,29 @@ static int bcm2835_pwm_probe(struct platform_device *pdev)
 		return dev_err_probe(&pdev->dev, PTR_ERR(pc->clk),
 				     "clock not found\n");
 
+	ret = clk_rate_exclusive_get(pc->clk);
+	if (ret)
+		return dev_err_probe(&pdev->dev, ret,
+				     "fail to get exclusive rate\n");
+
+	pc->rate = clk_get_rate(pc->clk);
+	if (!pc->rate) {
+		clk_rate_exclusive_put(pc->clk);
+		return dev_err_probe(&pdev->dev, -EINVAL,
+				     "failed to get clock rate\n");
+	}
+
 	pc->chip.dev = &pdev->dev;
 	pc->chip.ops = &bcm2835_pwm_ops;
+	pc->chip.atomic = true;
 	pc->chip.npwm = 2;
 
 	ret = devm_pwmchip_add(&pdev->dev, &pc->chip);
-	if (ret < 0)
+	if (ret < 0) {
+		clk_rate_exclusive_put(pc->clk);
 		return dev_err_probe(&pdev->dev, ret,
 				     "failed to add pwmchip\n");
+	}
 
 	return 0;
 }
@@ -167,6 +177,7 @@ static int bcm2835_pwm_suspend(struct device *dev)
 {
 	struct bcm2835_pwm *pc = dev_get_drvdata(dev);
 
+	clk_rate_exclusive_put(pc->clk);
 	clk_disable_unprepare(pc->clk);
 
 	return 0;
-- 
2.43.0

Re: [PATCH v6 3/4] pwm: bcm2835: allow pwm driver to be used in atomic context

[Florian Fainelli]

[-- Attachment #1: Type: text/plain, Size: 284 bytes --]

On 11/29/23 01:13, Sean Young wrote:
> clk_get_rate() may do a mutex lock. Fetch the clock rate once, and prevent
> rate changes using clk_rate_exclusive_get().
> 
> Signed-off-by: Sean Young <sean@mess.org>

Reviewed-by: Florian Fainelli <florian.fainelli@broadcom.com>
-- 
Florian


[-- Attachment #2: S/MIME Cryptographic Signature --]
[-- Type: application/pkcs7-signature, Size: 4221 bytes --]

Re: [PATCH v6 3/4] pwm: bcm2835: allow pwm driver to be used in atomic context

[Thierry Reding]

[-- Attachment #1: Type: text/plain, Size: 573 bytes --]

On Wed, Nov 29, 2023 at 09:13:36AM +0000, Sean Young wrote:
> clk_get_rate() may do a mutex lock. Fetch the clock rate once, and prevent
> rate changes using clk_rate_exclusive_get().
> 
> Signed-off-by: Sean Young <sean@mess.org>
> ---
>  drivers/pwm/pwm-bcm2835.c | 31 +++++++++++++++++++++----------
>  1 file changed, 21 insertions(+), 10 deletions(-)

s/pwm/PWM/ in the subject. Although, I guess you could just drop the
"PWM" altogether because the subject prefix implies that this is for
PWM. Also, please capitalize after the subject prefix.

Thierry

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH v6 3/4] pwm: bcm2835: allow pwm driver to be used in atomic context

[Sean Young]

On Fri, Dec 08, 2023 at 05:22:52PM +0100, Thierry Reding wrote:
> On Wed, Nov 29, 2023 at 09:13:36AM +0000, Sean Young wrote:
> > clk_get_rate() may do a mutex lock. Fetch the clock rate once, and prevent
> > rate changes using clk_rate_exclusive_get().
> > 
> > Signed-off-by: Sean Young <sean@mess.org>
> > ---
> >  drivers/pwm/pwm-bcm2835.c | 31 +++++++++++++++++++++----------
> >  1 file changed, 21 insertions(+), 10 deletions(-)
> 
> s/pwm/PWM/ in the subject. Although, I guess you could just drop the
> "PWM" altogether because the subject prefix implies that this is for
> PWM.

$ git log --no-merges --oneline drivers/pwm/ | sed -r 's/^\w* ([^:]+): .*/\1/' | sort | uniq -c
   1197 pwm
      1 PWM
  ...

The vast majority of the commits use pwm: as a prefix, only one uses PWM:. 
In fact if you look across the tree almost everywhere lower case is used
for the prefix.

I'm just trying to follow convention.

Having said that, I think the prefix is totally redundant, it is clear from
the commit files what they are affecting. I am not sure what it really adds.

> Also, please capitalize after the subject prefix.

$ git log --no-merges --oneline drivers/pwm/ | grep -E '^\w* ([^:]+): [A-Z]' | wc -l
217
$ git log --no-merges --oneline drivers/pwm/ | grep -E '^\w* ([^:]+): [a-z]' | wc -l
1069

Although not as clear, convention seems to be lower case for commits. The
first line of a commit is not really a sentence, there is no trailing
period.

I am happy to oblige, just wanted to point this out. Sorry if this starts
a bikeshed discussion.

Thanks,

Sean

Re: [PATCH v6 3/4] pwm: bcm2835: allow pwm driver to be used in atomic context

[Uwe Kleine-König]

[-- Attachment #1: Type: text/plain, Size: 2576 bytes --]

On Fri, Dec 08, 2023 at 05:01:26PM +0000, Sean Young wrote:
> On Fri, Dec 08, 2023 at 05:22:52PM +0100, Thierry Reding wrote:
> > On Wed, Nov 29, 2023 at 09:13:36AM +0000, Sean Young wrote:
> > > clk_get_rate() may do a mutex lock. Fetch the clock rate once, and prevent
> > > rate changes using clk_rate_exclusive_get().
> > > 
> > > Signed-off-by: Sean Young <sean@mess.org>
> > > ---
> > >  drivers/pwm/pwm-bcm2835.c | 31 +++++++++++++++++++++----------
> > >  1 file changed, 21 insertions(+), 10 deletions(-)
> > 
> > s/pwm/PWM/ in the subject. Although, I guess you could just drop the
> > "PWM" altogether because the subject prefix implies that this is for
> > PWM.
> 
> $ git log --no-merges --oneline drivers/pwm/ | sed -r 's/^\w* ([^:]+): .*/\1/' | sort | uniq -c
>    1197 pwm
>       1 PWM
>   ...
> 
> The vast majority of the commits use pwm: as a prefix, only one uses PWM:. 
> In fact if you look across the tree almost everywhere lower case is used
> for the prefix.

Thierry doesn't want you to change the subject prefix, but only the
second "pwm" to make it read:

	pwm: bcm2835: allow PWM driver to be used in atomic context

While I understand Thierry here, I'm fine with a lowercase pwm here,
too. In my book a PWM in all uppercase is the type of hardware and pwm
in all lowercase is the framework's name. If you use "PWM driver" or
"pwm driver" then doesn't matter much.

	
> 
> I'm just trying to follow convention.
> 
> Having said that, I think the prefix is totally redundant, it is clear from
> the commit files what they are affecting. I am not sure what it really adds.
> 
> > Also, please capitalize after the subject prefix.
> 
> $ git log --no-merges --oneline drivers/pwm/ | grep -E '^\w* ([^:]+): [A-Z]' | wc -l
> 217
> $ git log --no-merges --oneline drivers/pwm/ | grep -E '^\w* ([^:]+): [a-z]' | wc -l
> 1069

Your matching things like:

	pwm: pwm-tiehrpwm: Add support for configuring polarity of PWM

with the second command. These are perfectly fine as pwm-tiehrpwm is the
driver name and so shouldn't be capitalized. With a bit more care here,
we get:

	$ git log --no-merges --oneline drivers/pwm/ | grep -E '^.+: [a-z][^:]*$' | wc -l
	114
	$ git log --no-merges --oneline drivers/pwm/ | grep -E '^.+: [A-Z][^:]*$' | wc -l
	1167

And the newest of the 114 with a small letter is from 2013.

Best regards
Uwe

-- 
Pengutronix e.K.                           | Uwe Kleine-König            |
Industrial Linux Solutions                 | https://www.pengutronix.de/ |

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]

Re: [PATCH v6 3/4] pwm: bcm2835: allow pwm driver to be used in atomic context

[Sean Young]

On Fri, Dec 08, 2023 at 06:20:40PM +0100, Uwe Kleine-König wrote:
> On Fri, Dec 08, 2023 at 05:01:26PM +0000, Sean Young wrote:
> > On Fri, Dec 08, 2023 at 05:22:52PM +0100, Thierry Reding wrote:
> > > On Wed, Nov 29, 2023 at 09:13:36AM +0000, Sean Young wrote:
> > > > clk_get_rate() may do a mutex lock. Fetch the clock rate once, and prevent
> > > > rate changes using clk_rate_exclusive_get().
> > > > 
> > > > Signed-off-by: Sean Young <sean@mess.org>
> > > > ---
> > > >  drivers/pwm/pwm-bcm2835.c | 31 +++++++++++++++++++++----------
> > > >  1 file changed, 21 insertions(+), 10 deletions(-)
> > > 
> > > s/pwm/PWM/ in the subject. Although, I guess you could just drop the
> > > "PWM" altogether because the subject prefix implies that this is for
> > > PWM.
> > 
> > $ git log --no-merges --oneline drivers/pwm/ | sed -r 's/^\w* ([^:]+): .*/\1/' | sort | uniq -c
> >    1197 pwm
> >       1 PWM
> >   ...
> > 
> > The vast majority of the commits use pwm: as a prefix, only one uses PWM:. 
> > In fact if you look across the tree almost everywhere lower case is used
> > for the prefix.
> 
> Thierry doesn't want you to change the subject prefix, but only the
> second "pwm" to make it read:
> 
> 	pwm: bcm2835: allow PWM driver to be used in atomic context

Ah of course, my bad.

> While I understand Thierry here, I'm fine with a lowercase pwm here,
> too. In my book a PWM in all uppercase is the type of hardware and pwm
> in all lowercase is the framework's name. If you use "PWM driver" or
> "pwm driver" then doesn't matter much.
> 
> 	
> > 
> > I'm just trying to follow convention.
> > 
> > Having said that, I think the prefix is totally redundant, it is clear from
> > the commit files what they are affecting. I am not sure what it really adds.
> > 
> > > Also, please capitalize after the subject prefix.
> > 
> > $ git log --no-merges --oneline drivers/pwm/ | grep -E '^\w* ([^:]+): [A-Z]' | wc -l
> > 217
> > $ git log --no-merges --oneline drivers/pwm/ | grep -E '^\w* ([^:]+): [a-z]' | wc -l
> > 1069
> 
> Your matching things like:
> 
> 	pwm: pwm-tiehrpwm: Add support for configuring polarity of PWM
> 
> with the second command. These are perfectly fine as pwm-tiehrpwm is the
> driver name and so shouldn't be capitalized. With a bit more care here,
> we get:
> 
> 	$ git log --no-merges --oneline drivers/pwm/ | grep -E '^.+: [a-z][^:]*$' | wc -l
> 	114
> 	$ git log --no-merges --oneline drivers/pwm/ | grep -E '^.+: [A-Z][^:]*$' | wc -l
> 	1167
> 
> And the newest of the 114 with a small letter is from 2013.

Again, I stand corrected.

Thanks for pointing it out, I will fix in the next version.


Sean

Re: [PATCH v6 3/4] pwm: bcm2835: allow pwm driver to be used in atomic context

[Thierry Reding]

[-- Attachment #1: Type: text/plain, Size: 1938 bytes --]

On Fri, Dec 08, 2023 at 06:20:40PM +0100, Uwe Kleine-König wrote:
> On Fri, Dec 08, 2023 at 05:01:26PM +0000, Sean Young wrote:
> > On Fri, Dec 08, 2023 at 05:22:52PM +0100, Thierry Reding wrote:
> > > On Wed, Nov 29, 2023 at 09:13:36AM +0000, Sean Young wrote:
> > > > clk_get_rate() may do a mutex lock. Fetch the clock rate once, and prevent
> > > > rate changes using clk_rate_exclusive_get().
> > > > 
> > > > Signed-off-by: Sean Young <sean@mess.org>
> > > > ---
> > > >  drivers/pwm/pwm-bcm2835.c | 31 +++++++++++++++++++++----------
> > > >  1 file changed, 21 insertions(+), 10 deletions(-)
> > > 
> > > s/pwm/PWM/ in the subject. Although, I guess you could just drop the
> > > "PWM" altogether because the subject prefix implies that this is for
> > > PWM.
> > 
> > $ git log --no-merges --oneline drivers/pwm/ | sed -r 's/^\w* ([^:]+): .*/\1/' | sort | uniq -c
> >    1197 pwm
> >       1 PWM
> >   ...
> > 
> > The vast majority of the commits use pwm: as a prefix, only one uses PWM:. 
> > In fact if you look across the tree almost everywhere lower case is used
> > for the prefix.
> 
> Thierry doesn't want you to change the subject prefix, but only the
> second "pwm" to make it read:
> 
> 	pwm: bcm2835: allow PWM driver to be used in atomic context
> 
> While I understand Thierry here, I'm fine with a lowercase pwm here,
> too. In my book a PWM in all uppercase is the type of hardware and pwm
> in all lowercase is the framework's name. If you use "PWM driver" or
> "pwm driver" then doesn't matter much.

I'm not fine with a lowercase "pwm" in what is clearly text. Text should
be grammatically correct and PWM being an abbreviation it should be all
caps. The framework name is also PWM, not pwm. We use the lowercase pwm
as prefix because it represents the directory where the subsystem lives
and we usually don't use capitalization in file and directory names.

Thierry

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

[PATCH v6 4/4] media: pwm-ir-tx: trigger edges from hrtimer interrupt context

[Sean Young]

This makes the generated IR much more precise. Before this change, the
driver is unreliable and many users opted to use gpio-ir-tx instead.

Signed-off-by: Sean Young <sean@mess.org>
---
 drivers/media/rc/pwm-ir-tx.c | 79 ++++++++++++++++++++++++++++++++++--
 1 file changed, 76 insertions(+), 3 deletions(-)

diff --git a/drivers/media/rc/pwm-ir-tx.c b/drivers/media/rc/pwm-ir-tx.c
index cf51e2760975..8575c4596d7b 100644
--- a/drivers/media/rc/pwm-ir-tx.c
+++ b/drivers/media/rc/pwm-ir-tx.c
@@ -10,6 +10,8 @@
 #include <linux/slab.h>
 #include <linux/of.h>
 #include <linux/platform_device.h>
+#include <linux/hrtimer.h>
+#include <linux/completion.h>
 #include <media/rc-core.h>
 
 #define DRIVER_NAME	"pwm-ir-tx"
@@ -17,8 +19,14 @@
 
 struct pwm_ir {
 	struct pwm_device *pwm;
-	unsigned int carrier;
-	unsigned int duty_cycle;
+	struct hrtimer timer;
+	struct completion tx_done;
+	struct pwm_state *state;
+	u32 carrier;
+	u32 duty_cycle;
+	uint *txbuf;
+	uint txbuf_len;
+	uint txbuf_index;
 };
 
 static const struct of_device_id pwm_ir_of_match[] = {
@@ -82,6 +90,62 @@ static int pwm_ir_tx(struct rc_dev *dev, unsigned int *txbuf,
 	return count;
 }
 
+static int pwm_ir_tx_atomic(struct rc_dev *dev, unsigned int *txbuf,
+			    unsigned int count)
+{
+	struct pwm_ir *pwm_ir = dev->priv;
+	struct pwm_device *pwm = pwm_ir->pwm;
+	struct pwm_state state;
+
+	pwm_init_state(pwm, &state);
+
+	state.period = DIV_ROUND_CLOSEST(NSEC_PER_SEC, pwm_ir->carrier);
+	pwm_set_relative_duty_cycle(&state, pwm_ir->duty_cycle, 100);
+
+	pwm_ir->txbuf = txbuf;
+	pwm_ir->txbuf_len = count;
+	pwm_ir->txbuf_index = 0;
+	pwm_ir->state = &state;
+
+	hrtimer_start(&pwm_ir->timer, 0, HRTIMER_MODE_REL);
+
+	wait_for_completion(&pwm_ir->tx_done);
+
+	return count;
+}
+
+static enum hrtimer_restart pwm_ir_timer(struct hrtimer *timer)
+{
+	struct pwm_ir *pwm_ir = container_of(timer, struct pwm_ir, timer);
+	ktime_t now;
+
+	/*
+	 * If we happen to hit an odd latency spike, loop through the
+	 * pulses until we catch up.
+	 */
+	do {
+		u64 ns;
+
+		pwm_ir->state->enabled = !(pwm_ir->txbuf_index % 2);
+		pwm_apply_atomic(pwm_ir->pwm, pwm_ir->state);
+
+		if (pwm_ir->txbuf_index >= pwm_ir->txbuf_len) {
+			complete(&pwm_ir->tx_done);
+
+			return HRTIMER_NORESTART;
+		}
+
+		ns = US_TO_NS(pwm_ir->txbuf[pwm_ir->txbuf_index]);
+		hrtimer_add_expires_ns(timer, ns);
+
+		pwm_ir->txbuf_index++;
+
+		now = timer->base->get_time();
+	} while (hrtimer_get_expires_tv64(timer) < now);
+
+	return HRTIMER_RESTART;
+}
+
 static int pwm_ir_probe(struct platform_device *pdev)
 {
 	struct pwm_ir *pwm_ir;
@@ -103,10 +167,19 @@ static int pwm_ir_probe(struct platform_device *pdev)
 	if (!rcdev)
 		return -ENOMEM;
 
+	if (pwm_is_atomic(pwm_ir->pwm)) {
+		init_completion(&pwm_ir->tx_done);
+		hrtimer_init(&pwm_ir->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
+		pwm_ir->timer.function = pwm_ir_timer;
+		rcdev->tx_ir = pwm_ir_tx_atomic;
+	} else {
+		dev_info(&pdev->dev, "tx will not be accurate as pwm device does not support atomic mode");
+		rcdev->tx_ir = pwm_ir_tx;
+	}
+
 	rcdev->priv = pwm_ir;
 	rcdev->driver_name = DRIVER_NAME;
 	rcdev->device_name = DEVICE_NAME;
-	rcdev->tx_ir = pwm_ir_tx;
 	rcdev->s_tx_duty_cycle = pwm_ir_set_duty_cycle;
 	rcdev->s_tx_carrier = pwm_ir_set_carrier;
 
-- 
2.43.0

Re: [PATCH v6 4/4] media: pwm-ir-tx: trigger edges from hrtimer interrupt context

[Thierry Reding]

[-- Attachment #1: Type: text/plain, Size: 4043 bytes --]

On Wed, Nov 29, 2023 at 09:13:37AM +0000, Sean Young wrote:
> This makes the generated IR much more precise. Before this change, the
> driver is unreliable and many users opted to use gpio-ir-tx instead.
> 
> Signed-off-by: Sean Young <sean@mess.org>
> ---
>  drivers/media/rc/pwm-ir-tx.c | 79 ++++++++++++++++++++++++++++++++++--
>  1 file changed, 76 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/media/rc/pwm-ir-tx.c b/drivers/media/rc/pwm-ir-tx.c
> index cf51e2760975..8575c4596d7b 100644
> --- a/drivers/media/rc/pwm-ir-tx.c
> +++ b/drivers/media/rc/pwm-ir-tx.c
> @@ -10,6 +10,8 @@
>  #include <linux/slab.h>
>  #include <linux/of.h>
>  #include <linux/platform_device.h>
> +#include <linux/hrtimer.h>
> +#include <linux/completion.h>
>  #include <media/rc-core.h>
>  
>  #define DRIVER_NAME	"pwm-ir-tx"
> @@ -17,8 +19,14 @@
>  
>  struct pwm_ir {
>  	struct pwm_device *pwm;
> -	unsigned int carrier;
> -	unsigned int duty_cycle;
> +	struct hrtimer timer;
> +	struct completion tx_done;
> +	struct pwm_state *state;
> +	u32 carrier;
> +	u32 duty_cycle;
> +	uint *txbuf;

Maybe mark this as const to signal that it's not going to get modified?

> +	uint txbuf_len;
> +	uint txbuf_index;

uint is rather rare. Or so I thought. There seem to be quite a few
occurrences throughout the kernel. I'd still prefer unsigned int over
this abbreviated form, but ultimately up to you and Mauro to decide.

>  };
>  
>  static const struct of_device_id pwm_ir_of_match[] = {
> @@ -82,6 +90,62 @@ static int pwm_ir_tx(struct rc_dev *dev, unsigned int *txbuf,
>  	return count;
>  }
>  
> +static int pwm_ir_tx_atomic(struct rc_dev *dev, unsigned int *txbuf,
> +			    unsigned int count)
> +{
> +	struct pwm_ir *pwm_ir = dev->priv;
> +	struct pwm_device *pwm = pwm_ir->pwm;
> +	struct pwm_state state;
> +
> +	pwm_init_state(pwm, &state);
> +
> +	state.period = DIV_ROUND_CLOSEST(NSEC_PER_SEC, pwm_ir->carrier);
> +	pwm_set_relative_duty_cycle(&state, pwm_ir->duty_cycle, 100);
> +
> +	pwm_ir->txbuf = txbuf;
> +	pwm_ir->txbuf_len = count;
> +	pwm_ir->txbuf_index = 0;
> +	pwm_ir->state = &state;
> +
> +	hrtimer_start(&pwm_ir->timer, 0, HRTIMER_MODE_REL);
> +
> +	wait_for_completion(&pwm_ir->tx_done);
> +
> +	return count;
> +}
> +
> +static enum hrtimer_restart pwm_ir_timer(struct hrtimer *timer)
> +{
> +	struct pwm_ir *pwm_ir = container_of(timer, struct pwm_ir, timer);
> +	ktime_t now;
> +
> +	/*
> +	 * If we happen to hit an odd latency spike, loop through the
> +	 * pulses until we catch up.
> +	 */
> +	do {
> +		u64 ns;
> +
> +		pwm_ir->state->enabled = !(pwm_ir->txbuf_index % 2);
> +		pwm_apply_atomic(pwm_ir->pwm, pwm_ir->state);
> +
> +		if (pwm_ir->txbuf_index >= pwm_ir->txbuf_len) {
> +			complete(&pwm_ir->tx_done);
> +
> +			return HRTIMER_NORESTART;
> +		}
> +
> +		ns = US_TO_NS(pwm_ir->txbuf[pwm_ir->txbuf_index]);
> +		hrtimer_add_expires_ns(timer, ns);
> +
> +		pwm_ir->txbuf_index++;
> +
> +		now = timer->base->get_time();
> +	} while (hrtimer_get_expires_tv64(timer) < now);
> +
> +	return HRTIMER_RESTART;
> +}
> +
>  static int pwm_ir_probe(struct platform_device *pdev)
>  {
>  	struct pwm_ir *pwm_ir;
> @@ -103,10 +167,19 @@ static int pwm_ir_probe(struct platform_device *pdev)
>  	if (!rcdev)
>  		return -ENOMEM;
>  
> +	if (pwm_is_atomic(pwm_ir->pwm)) {
> +		init_completion(&pwm_ir->tx_done);
> +		hrtimer_init(&pwm_ir->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
> +		pwm_ir->timer.function = pwm_ir_timer;
> +		rcdev->tx_ir = pwm_ir_tx_atomic;
> +	} else {
> +		dev_info(&pdev->dev, "tx will not be accurate as pwm device does not support atomic mode");

s/tx/TX and s/pwm/PWM/? Also, I'm a bit unhappy about "atomic mode" here
because the term is overloaded in PWM. If you call pwm_appy_*() then by
definition it's going to be "atomic" in the "atomic state" sense. So
maybe switch to something like:

	"TX will not be accurate as PWM device might sleep"

?

Thierry

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [PATCH v6 4/4] media: pwm-ir-tx: trigger edges from hrtimer interrupt context

[Sean Young]

On Fri, Dec 08, 2023 at 05:29:55PM +0100, Thierry Reding wrote:
> On Wed, Nov 29, 2023 at 09:13:37AM +0000, Sean Young wrote:
> > This makes the generated IR much more precise. Before this change, the
> > driver is unreliable and many users opted to use gpio-ir-tx instead.
> > 
> > Signed-off-by: Sean Young <sean@mess.org>
> > ---
> >  drivers/media/rc/pwm-ir-tx.c | 79 ++++++++++++++++++++++++++++++++++--
> >  1 file changed, 76 insertions(+), 3 deletions(-)
> > 
> > diff --git a/drivers/media/rc/pwm-ir-tx.c b/drivers/media/rc/pwm-ir-tx.c
> > index cf51e2760975..8575c4596d7b 100644
> > --- a/drivers/media/rc/pwm-ir-tx.c
> > +++ b/drivers/media/rc/pwm-ir-tx.c
> > @@ -10,6 +10,8 @@
> >  #include <linux/slab.h>
> >  #include <linux/of.h>
> >  #include <linux/platform_device.h>
> > +#include <linux/hrtimer.h>
> > +#include <linux/completion.h>
> >  #include <media/rc-core.h>
> >  
> >  #define DRIVER_NAME	"pwm-ir-tx"
> > @@ -17,8 +19,14 @@
> >  
> >  struct pwm_ir {
> >  	struct pwm_device *pwm;
> > -	unsigned int carrier;
> > -	unsigned int duty_cycle;
> > +	struct hrtimer timer;
> > +	struct completion tx_done;
> > +	struct pwm_state *state;
> > +	u32 carrier;
> > +	u32 duty_cycle;
> > +	uint *txbuf;
> 
> Maybe mark this as const to signal that it's not going to get modified?

Ah nice, I usually forget const. 

> > +	uint txbuf_len;
> > +	uint txbuf_index;
> 
> uint is rather rare. Or so I thought. There seem to be quite a few
> occurrences throughout the kernel. I'd still prefer unsigned int over
> this abbreviated form, but ultimately up to you and Mauro to decide.

Yes, unsigned int is used a lot more. Changed.

> >  static int pwm_ir_probe(struct platform_device *pdev)
> >  {
> >  	struct pwm_ir *pwm_ir;
> > @@ -103,10 +167,19 @@ static int pwm_ir_probe(struct platform_device *pdev)
> >  	if (!rcdev)
> >  		return -ENOMEM;
> >  
> > +	if (pwm_is_atomic(pwm_ir->pwm)) {
> > +		init_completion(&pwm_ir->tx_done);
> > +		hrtimer_init(&pwm_ir->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
> > +		pwm_ir->timer.function = pwm_ir_timer;
> > +		rcdev->tx_ir = pwm_ir_tx_atomic;
> > +	} else {
> > +		dev_info(&pdev->dev, "tx will not be accurate as pwm device does not support atomic mode");
> 
> s/tx/TX and s/pwm/PWM/? Also, I'm a bit unhappy about "atomic mode" here
> because the term is overloaded in PWM. If you call pwm_appy_*() then by
> definition it's going to be "atomic" in the "atomic state" sense. So
> maybe switch to something like:
> 
> 	"TX will not be accurate as PWM device might sleep"
> 
> ?

Very nice, changed.

Thanks
Sean