[PATCH v15 0/5] Change PWM-controlled LED pin active mode and algorithm

[PATCH v15 0/5] Change PWM-controlled LED pin active mode and algorithm

[Nylon Chen]

According to the circuit diagram of User LEDs - RGB described in the
manual hifive-unleashed-a00.pdf[0] and hifive-unmatched-schematics-v3.pdf[1].

The behavior of PWM is acitve-high.

According to the descriptionof PWM for pwmcmp in SiFive FU740-C000 Manual[2].

The pwm algorithm is (PW) pulse active time  = (D) duty * (T) period.
The `frac` variable is pulse "inactive" time so we need to invert it.

So this patchset removes active-low in DTS and adds reverse logic to the driver.

Links:
- [0]: https://sifive.cdn.prismic.io/sifive/c52a8e32-05ce-4aaf-95c8-7bf8453f8698_hifive-unleashed-a00-schematics-1.pdf
- [1]: https://sifive.cdn.prismic.io/sifive/6a06d6c0-6e66-49b5-8e9e-e68ce76f4192_hifive-unmatched-schematics-v3.pdf
- [2]: https://sifive.cdn.prismic.io/sifive/1a82e600-1f93-4f41-b2d8-86ed8b16acba_fu740-c000-manual-v1p6.pdf

Updated patches: 2
New patches: 0
Unchanged patches: 0

Changes in v15:
- Squashed patch 2 and 5 to combine algorithm changes with related comments
- Squashed patch 3 and 4 to unify rounding logic and idempotency fix

Changed in v14:
 - Change `frac` from `u32` to `u64` and cast the constant in `min()` to
   `u64`  so that `do_div(frac, state->period)` no longer triggers type-mismatch

Changed in v13:
 - Fix syntax error: Added missing closing parenthesis in do_div()
   function call.

Changed in v12:
 - Replace division with do_div() to fix __udivdi3 modpost error.

Changed in v11:
 - Fix rounding consistency in apply() and get_state()
 - Add code comments to help clarify Reference Manual errors.

Changed in v10:
 - Add 'inactive' variable in apply() to match pwm_sifive_get_state()
   style
 - Update comment about hardware limitation - it cannot generate 0% duty
   cycle rather than 100% duty cycle

Changed in v9:
 - Fix commit message to adhere to 75 columns rule.
 - Update commit message's subject.
 - Add a variable for inactive logic.

Changed in v8:
 - Fix Signed-off-by and Co-developed-by typo.

Changed in v7:
 - Remove active-low strings from hifive-unleashed-a00.dts file.

Changed in v6:
 - Separate the idempotent test bug fixes into a new patch.
 - Move the reversing the duty before the line checking
   state->enabled.
 - Fix the algorithm and change it to take the minimum value first and
   then reverse it.

Changed in v5:
 - Add the updates to the PWM algorithm based on version 2 back in.
 - Replace div64_ul with DIV_ROUND_UP_ULL to correct the error in the
   period value of the idempotent test in pwm_apply_state_debug.

Changed in v4:
 - Remove previous updates to the PWM algorithm.

Changed in v3:
 - Convert the reference link to standard link.
 - Move the inverted function before taking the minimum value.
 - Change polarity check condition(high and low).
 - Pick the biggest period length possible that is not bigger than the
   requested period.

Changed in v2:
 - Convert the reference link to standard link.
 - Fix typo: s/sifive unmatched:/sifive: unmatched:/.
 - Remove active-low from hifive-unleashed-a00.dts.
 - Include this reference link in the dts and pwm commit messages.

Nylon Chen (5):
  riscv: dts: sifive: unleashed/unmatched: Remove PWM controlled LED's
    active-low properties
  pwm: sifive: change the PWM algorithm
  pwm: sifive: Fix the error in the idempotent test within the
    pwm_apply_state_debug function
  pwm: sifive: Fix rounding issues in apply and get_state functions
  pwm: sifive: clarify inverted compare logic in comments

 .../boot/dts/sifive/hifive-unleashed-a00.dts  | 12 ++---
 .../boot/dts/sifive/hifive-unmatched-a00.dts  | 12 ++---
 drivers/pwm/pwm-sifive.c                      | 52 ++++++++++++++-----
 3 files changed, 47 insertions(+), 29 deletions(-)

-- 
2.34.1


_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv

[PATCH v15 1/3] riscv: dts: sifive: unleashed/unmatched: Remove PWM controlled LED's active-low properties

[Nylon Chen]

This removes the active-low properties of the PWM-controlled LEDs in
the HiFive Unmatched device tree.

The reference is hifive-unleashed-a00.pdf[0] and hifive-unmatched-schematics-v3.pdf[1].

Link: https://sifive.cdn.prismic.io/sifive/c52a8e32-05ce-4aaf-95c8-7bf8453f8698_hifive-unleashed-a00-schematics-1.pdf [0]
Link: https://sifive.cdn.prismic.io/sifive/6a06d6c0-6e66-49b5-8e9e-e68ce76f4192_hifive-unmatched-schematics-v3.pdf [1]

Acked-by: Conor Dooley <conor.dooley@microchip.com>
Reviewed-by: Conor Dooley <conor.dooley@microchip.com>
Signed-off-by: Vincent Chen <vincent.chen@sifive.com>
Signed-off-by: Nylon Chen <nylon.chen@sifive.com>
---
 arch/riscv/boot/dts/sifive/hifive-unleashed-a00.dts | 12 ++++--------
 arch/riscv/boot/dts/sifive/hifive-unmatched-a00.dts | 12 ++++--------
 2 files changed, 8 insertions(+), 16 deletions(-)

diff --git a/arch/riscv/boot/dts/sifive/hifive-unleashed-a00.dts b/arch/riscv/boot/dts/sifive/hifive-unleashed-a00.dts
index 900a50526d77..06731b8c7bc3 100644
--- a/arch/riscv/boot/dts/sifive/hifive-unleashed-a00.dts
+++ b/arch/riscv/boot/dts/sifive/hifive-unleashed-a00.dts
@@ -49,32 +49,28 @@ led-controller {
 		compatible = "pwm-leds";
 
 		led-d1 {
-			pwms = <&pwm0 0 7812500 PWM_POLARITY_INVERTED>;
-			active-low;
+			pwms = <&pwm0 0 7812500 0>;
 			color = <LED_COLOR_ID_GREEN>;
 			max-brightness = <255>;
 			label = "d1";
 		};
 
 		led-d2 {
-			pwms = <&pwm0 1 7812500 PWM_POLARITY_INVERTED>;
-			active-low;
+			pwms = <&pwm0 1 7812500 0>;
 			color = <LED_COLOR_ID_GREEN>;
 			max-brightness = <255>;
 			label = "d2";
 		};
 
 		led-d3 {
-			pwms = <&pwm0 2 7812500 PWM_POLARITY_INVERTED>;
-			active-low;
+			pwms = <&pwm0 2 7812500 0>;
 			color = <LED_COLOR_ID_GREEN>;
 			max-brightness = <255>;
 			label = "d3";
 		};
 
 		led-d4 {
-			pwms = <&pwm0 3 7812500 PWM_POLARITY_INVERTED>;
-			active-low;
+			pwms = <&pwm0 3 7812500 0>;
 			color = <LED_COLOR_ID_GREEN>;
 			max-brightness = <255>;
 			label = "d4";
diff --git a/arch/riscv/boot/dts/sifive/hifive-unmatched-a00.dts b/arch/riscv/boot/dts/sifive/hifive-unmatched-a00.dts
index 72b87b08ab44..03ce2cee4e97 100644
--- a/arch/riscv/boot/dts/sifive/hifive-unmatched-a00.dts
+++ b/arch/riscv/boot/dts/sifive/hifive-unmatched-a00.dts
@@ -51,8 +51,7 @@ led-controller-1 {
 		compatible = "pwm-leds";
 
 		led-d12 {
-			pwms = <&pwm0 0 7812500 PWM_POLARITY_INVERTED>;
-			active-low;
+			pwms = <&pwm0 0 7812500 0>;
 			color = <LED_COLOR_ID_GREEN>;
 			max-brightness = <255>;
 			label = "d12";
@@ -68,20 +67,17 @@ multi-led {
 			label = "d2";
 
 			led-red {
-				pwms = <&pwm0 2 7812500 PWM_POLARITY_INVERTED>;
-				active-low;
+				pwms = <&pwm0 2 7812500 0>;
 				color = <LED_COLOR_ID_RED>;
 			};
 
 			led-green {
-				pwms = <&pwm0 1 7812500 PWM_POLARITY_INVERTED>;
-				active-low;
+				pwms = <&pwm0 1 7812500 0>;
 				color = <LED_COLOR_ID_GREEN>;
 			};
 
 			led-blue {
-				pwms = <&pwm0 3 7812500 PWM_POLARITY_INVERTED>;
-				active-low;
+				pwms = <&pwm0 3 7812500 0>;
 				color = <LED_COLOR_ID_BLUE>;
 			};
 		};
-- 
2.34.1


_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv

[PATCH v15 2/3] pwm: sifive: fix PWM algorithm and clarify inverted compare behavior

[Nylon Chen]

The `frac` variable represents the pulse inactive time, and the result
of this algorithm is the pulse active time. Therefore, we must reverse
the result.

Although the SiFive Reference Manual states "pwms >= pwmcmpX -> HIGH",
the hardware behavior is inverted due to a fixed XNOR with 0. As a result,
the pwmcmp register actually defines the low (inactive) portion of the pulse.

The reference is SiFive FU740-C000 Manual[0]

Link: https://sifive.cdn.prismic.io/sifive/1a82e600-1f93-4f41-b2d8-86ed8b16acba_fu740-c000-manual-v1p6.pdf [0]

Co-developed-by: Zong Li <zong.li@sifive.com>
Signed-off-by: Zong Li <zong.li@sifive.com>
Co-developed-by: Vincent Chen <vincent.chen@sifive.com>
Signed-off-by: Vincent Chen <vincent.chen@sifive.com>
Signed-off-by: Nylon Chen <nylon.chen@sifive.com>
---
 drivers/pwm/pwm-sifive.c | 39 +++++++++++++++++++++++++++++++--------
 1 file changed, 31 insertions(+), 8 deletions(-)

diff --git a/drivers/pwm/pwm-sifive.c b/drivers/pwm/pwm-sifive.c
index d5b647e6be78..f3694801d3ee 100644
--- a/drivers/pwm/pwm-sifive.c
+++ b/drivers/pwm/pwm-sifive.c
@@ -4,11 +4,28 @@
  * For SiFive's PWM IP block documentation please refer Chapter 14 of
  * Reference Manual : https://static.dev.sifive.com/FU540-C000-v1.0.pdf
  *
+ * PWM output inversion: According to the SiFive Reference manual
+ * the output of each comparator is high whenever the value of pwms is
+ * greater than or equal to the corresponding pwmcmpX[Reference Manual].
+ *
+ * Figure 29 in the same manual shows that the pwmcmpXcenter bit is
+ * hard-tied to 0 (XNOR), which effectively inverts the comparison so that
+ * the output goes HIGH when  `pwms < pwmcmpX`.
+ *
+ * In other words, each pwmcmp register actually defines the **inactive**
+ * (low) period of the pulse, not the active time exactly opposite to what
+ * the documentation text implies.
+ *
+ * To compensate, this driver always **inverts** the duty value when reading
+ * or writing pwmcmp registers , so that users interact with a conventional
+ * **active-high** PWM interface.
+ *
+ *
  * Limitations:
  * - When changing both duty cycle and period, we cannot prevent in
  *   software that the output might produce a period with mixed
  *   settings (new period length and old duty cycle).
- * - The hardware cannot generate a 100% duty cycle.
+ * - The hardware cannot generate a 0% duty cycle.
  * - The hardware generates only inverted output.
  */
 #include <linux/clk.h>
@@ -110,9 +127,14 @@ static int pwm_sifive_get_state(struct pwm_chip *chip, struct pwm_device *pwm,
 				struct pwm_state *state)
 {
 	struct pwm_sifive_ddata *ddata = pwm_sifive_chip_to_ddata(chip);
-	u32 duty, val;
+	u32 duty, val, inactive;
 
-	duty = readl(ddata->regs + PWM_SIFIVE_PWMCMP(pwm->hwpwm));
+	inactive = readl(ddata->regs + PWM_SIFIVE_PWMCMP(pwm->hwpwm));
+	/*
+	 * PWM hardware uses 'inactive' counts in pwmcmp, so invert to get actual duty.
+	 * Here, 'inactive' is the low time and we compute duty as max_count - inactive.
+	 */
+	duty = (1U << PWM_SIFIVE_CMPWIDTH) - 1 - inactive;
 
 	state->enabled = duty > 0;
 
@@ -123,7 +145,7 @@ static int pwm_sifive_get_state(struct pwm_chip *chip, struct pwm_device *pwm,
 	state->period = ddata->real_period;
 	state->duty_cycle =
 		(u64)duty * ddata->real_period >> PWM_SIFIVE_CMPWIDTH;
-	state->polarity = PWM_POLARITY_INVERSED;
+	state->polarity = PWM_POLARITY_NORMAL;
 
 	return 0;
 }
@@ -137,9 +159,9 @@ static int pwm_sifive_apply(struct pwm_chip *chip, struct pwm_device *pwm,
 	unsigned long long num;
 	bool enabled;
 	int ret = 0;
-	u32 frac;
+	u32 frac, inactive;
 
-	if (state->polarity != PWM_POLARITY_INVERSED)
+	if (state->polarity != PWM_POLARITY_NORMAL)
 		return -EINVAL;
 
 	cur_state = pwm->state;
@@ -157,8 +179,9 @@ static int pwm_sifive_apply(struct pwm_chip *chip, struct pwm_device *pwm,
 	 */
 	num = (u64)duty_cycle * (1U << PWM_SIFIVE_CMPWIDTH);
 	frac = DIV64_U64_ROUND_CLOSEST(num, state->period);
-	/* The hardware cannot generate a 100% duty cycle */
+	/* The hardware cannot generate a 0% duty cycle */
 	frac = min(frac, (1U << PWM_SIFIVE_CMPWIDTH) - 1);
+	inactive = (1U << PWM_SIFIVE_CMPWIDTH) - 1 - frac;
 
 	mutex_lock(&ddata->lock);
 	if (state->period != ddata->approx_period) {
@@ -190,7 +213,7 @@ static int pwm_sifive_apply(struct pwm_chip *chip, struct pwm_device *pwm,
 		}
 	}
 
-	writel(frac, ddata->regs + PWM_SIFIVE_PWMCMP(pwm->hwpwm));
+	writel(inactive, ddata->regs + PWM_SIFIVE_PWMCMP(pwm->hwpwm));
 
 	if (!state->enabled)
 		clk_disable(ddata->clk);
-- 
2.34.1


_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv

[PATCH v15 3/3] pwm: sifive: fix rounding and idempotency issues in apply and get_state

[Nylon Chen]

This fix ensures consistent rounding and avoids mismatches
between applied and reported PWM values that could trigger false
idempotency failures in debug checks

This change ensures:
- real_period is now calculated using DIV_ROUND_UP_ULL() to avoid underestimation.
- duty_cycle is rounded up to match the fractional computation in apply()
- apply() truncates the result to compensate for get_state's rounding up logic

These fixes resolve issues like:
.apply is supposed to round down duty_cycle (requested: 360/504000, applied: 361/504124)
.apply is not idempotent (ena=1 pol=0 1739692/4032985) -> (ena=1 pol=0 1739630/4032985)

Reported-by: kernel test robot <lkp@intel.com>
Closes: https://lore.kernel.org/oe-kbuild-all/202505080303.dBfU5YMS-lkp@intel.com/
Co-developed-by: Zong Li <zong.li@sifive.com>
Signed-off-by: Zong Li <zong.li@sifive.com>
Signed-off-by: Nylon Chen <nylon.chen@sifive.com>
---
 drivers/pwm/pwm-sifive.c | 15 +++++++++------
 1 file changed, 9 insertions(+), 6 deletions(-)

diff --git a/drivers/pwm/pwm-sifive.c b/drivers/pwm/pwm-sifive.c
index f3694801d3ee..4a07315b0744 100644
--- a/drivers/pwm/pwm-sifive.c
+++ b/drivers/pwm/pwm-sifive.c
@@ -118,7 +118,7 @@ static void pwm_sifive_update_clock(struct pwm_sifive_ddata *ddata,
 
 	/* As scale <= 15 the shift operation cannot overflow. */
 	num = (unsigned long long)NSEC_PER_SEC << (PWM_SIFIVE_CMPWIDTH + scale);
-	ddata->real_period = div64_ul(num, rate);
+	ddata->real_period = DIV_ROUND_UP_ULL(num, rate);
 	dev_dbg(ddata->parent,
 		"New real_period = %u ns\n", ddata->real_period);
 }
@@ -143,8 +143,8 @@ static int pwm_sifive_get_state(struct pwm_chip *chip, struct pwm_device *pwm,
 		state->enabled = false;
 
 	state->period = ddata->real_period;
-	state->duty_cycle =
-		(u64)duty * ddata->real_period >> PWM_SIFIVE_CMPWIDTH;
+	state->duty_cycle = DIV_ROUND_UP_ULL((u64)duty * ddata->real_period,
+					     (1U << PWM_SIFIVE_CMPWIDTH));
 	state->polarity = PWM_POLARITY_NORMAL;
 
 	return 0;
@@ -159,7 +159,8 @@ static int pwm_sifive_apply(struct pwm_chip *chip, struct pwm_device *pwm,
 	unsigned long long num;
 	bool enabled;
 	int ret = 0;
-	u32 frac, inactive;
+	u64 frac;
+	u32 inactive;
 
 	if (state->polarity != PWM_POLARITY_NORMAL)
 		return -EINVAL;
@@ -178,9 +179,11 @@ static int pwm_sifive_apply(struct pwm_chip *chip, struct pwm_device *pwm,
 	 * consecutively
 	 */
 	num = (u64)duty_cycle * (1U << PWM_SIFIVE_CMPWIDTH);
-	frac = DIV64_U64_ROUND_CLOSEST(num, state->period);
+	frac = num;
+	do_div(frac, state->period);
 	/* The hardware cannot generate a 0% duty cycle */
-	frac = min(frac, (1U << PWM_SIFIVE_CMPWIDTH) - 1);
+	frac = min(frac, (u64)(1U << PWM_SIFIVE_CMPWIDTH) - 1);
+	/* pwmcmp register must be loaded with the inactive(invert the duty) */
 	inactive = (1U << PWM_SIFIVE_CMPWIDTH) - 1 - frac;
 
 	mutex_lock(&ddata->lock);
-- 
2.34.1


_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv

Re: [PATCH v15 0/5] Change PWM-controlled LED pin active mode and algorithm

[Uwe Kleine-König]

[-- Attachment #1.1: Type: text/plain, Size: 1022 bytes --]

Hello Nylon,

On Thu, May 29, 2025 at 11:53:38AM +0800, Nylon Chen wrote:
> According to the circuit diagram of User LEDs - RGB described in the
> manual hifive-unleashed-a00.pdf[0] and hifive-unmatched-schematics-v3.pdf[1].
> 
> The behavior of PWM is acitve-high.
> 
> According to the descriptionof PWM for pwmcmp in SiFive FU740-C000 Manual[2].
> 
> The pwm algorithm is (PW) pulse active time  = (D) duty * (T) period.
> The `frac` variable is pulse "inactive" time so we need to invert it.
> 
> So this patchset removes active-low in DTS and adds reverse logic to the driver.

Applied to

	https://git.kernel.org/pub/scm/linux/kernel/git/ukleinek/linux.git pwm/for-nexxt

as 6.17-rc1 material. I pondered if this should go in earlier and/or
should get marked for stable. I'm unsure and willing to consider input
here. The problem I see is that while this is a fix, the change is a
breaking one and you need the dts change and the code change applied
either both or none.

Best regards
Uwe

[-- Attachment #1.2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]

[-- Attachment #2: Type: text/plain, Size: 161 bytes --]

_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv