The period value stored in the register is the number of clock cycles - 1.

Account for the offset when calculating the period in nanoseconds. This
fixes the discrepancy reported by the pwm core when CONFIG_PWM_DEBUG=1:

    mediatek-disp-pwm 1100e000.pwm: .apply is not idempotent
    	(ena=1 pol=0 281500/499875) -> (ena=1 pol=0 281500/499750)

Calculation discrepancy in the driver was found by AI run by a colleague
while debugging another PWM related issue. Check against SoC datasheet
and fix done by the author.

Fixes: 3f2b16734914 ("pwm: mtk-disp: Implement atomic API .get_state()")
Cc: Jitao Shi <jitao.shi@mediatek.com>
Signed-off-by: Chen-Yu Tsai <wenst@chromium.org>
---
 drivers/pwm/pwm-mtk-disp.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/pwm/pwm-mtk-disp.c b/drivers/pwm/pwm-mtk-disp.c
index b5409e15dac8..68a4512e821a 100644
--- a/drivers/pwm/pwm-mtk-disp.c
+++ b/drivers/pwm/pwm-mtk-disp.c
@@ -212,7 +212,7 @@ static int mtk_disp_pwm_get_state(struct pwm_chip *chip,
 	 * period has 12 bits, clk_div 11 and NSEC_PER_SEC has 30,
 	 * so period * (clk_div + 1) * NSEC_PER_SEC doesn't overflow.
 	 */
-	state->period = DIV64_U64_ROUND_UP(period * (clk_div + 1) * NSEC_PER_SEC, rate);
+	state->period = DIV64_U64_ROUND_UP((period + 1) * (clk_div + 1) * NSEC_PER_SEC, rate);
 	high_width = FIELD_GET(PWM_HIGH_WIDTH_MASK, con1);
 	state->duty_cycle = DIV64_U64_ROUND_UP(high_width * (clk_div + 1) * NSEC_PER_SEC,
 					       rate);
-- 
2.55.0.887.g758fc8c411-goog

Hello,

On Thu, Aug 27, 2026 at 06:57:36PM +0800, Chen-Yu Tsai wrote:
> The period value stored in the register is the number of clock cycles - 1.
> 
> Account for the offset when calculating the period in nanoseconds. This
> fixes the discrepancy reported by the pwm core when CONFIG_PWM_DEBUG=1:
> 
>     mediatek-disp-pwm 1100e000.pwm: .apply is not idempotent
>     	(ena=1 pol=0 281500/499875) -> (ena=1 pol=0 281500/499750)
> 
> Calculation discrepancy in the driver was found by AI run by a colleague
> while debugging another PWM related issue. Check against SoC datasheet
> and fix done by the author.
> 
> Fixes: 3f2b16734914 ("pwm: mtk-disp: Implement atomic API .get_state()")
> Cc: Jitao Shi <jitao.shi@mediatek.com>
> Signed-off-by: Chen-Yu Tsai <wenst@chromium.org>

Applied with the R-b tags by AngeloGioacchino Del Regno and Fei Shao and

diff --git a/drivers/pwm/pwm-mtk-disp.c b/drivers/pwm/pwm-mtk-disp.c
index 68a4512e821a..56d10cde5b46 100644
--- a/drivers/pwm/pwm-mtk-disp.c
+++ b/drivers/pwm/pwm-mtk-disp.c
@@ -210,7 +210,7 @@ static int mtk_disp_pwm_get_state(struct pwm_chip *chip,
 	period = FIELD_GET(PWM_PERIOD_MASK, con1);
 	/*
 	 * period has 12 bits, clk_div 11 and NSEC_PER_SEC has 30,
-	 * so period * (clk_div + 1) * NSEC_PER_SEC doesn't overflow.
+	 * so (period + 1) * (clk_div + 1) * NSEC_PER_SEC doesn't overflow.
 	 */
 	state->period = DIV64_U64_ROUND_UP((period + 1) * (clk_div + 1) * NSEC_PER_SEC, rate);
 	high_width = FIELD_GET(PWM_HIGH_WIDTH_MASK, con1);

squashed in to

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

.

Thanks
Uwe

On 8/27/26 12:57, Chen-Yu Tsai wrote:
> The period value stored in the register is the number of clock cycles - 1.
> 
> Account for the offset when calculating the period in nanoseconds. This
> fixes the discrepancy reported by the pwm core when CONFIG_PWM_DEBUG=1:
> 
>      mediatek-disp-pwm 1100e000.pwm: .apply is not idempotent
>      	(ena=1 pol=0 281500/499875) -> (ena=1 pol=0 281500/499750)
> 
> Calculation discrepancy in the driver was found by AI run by a colleague
> while debugging another PWM related issue. Check against SoC datasheet
> and fix done by the author.
> 
> Fixes: 3f2b16734914 ("pwm: mtk-disp: Implement atomic API .get_state()")
> Cc: Jitao Shi <jitao.shi@mediatek.com>
> Signed-off-by: Chen-Yu Tsai <wenst@chromium.org>

Reviewed-by: AngeloGioacchino Del Regno <angelogioacchino.delregno@collabora.com>

On Thu, Aug 27, 2026 at 6:57 PM Chen-Yu Tsai <wenst@chromium.org> wrote:
>
> The period value stored in the register is the number of clock cycles - 1.
>
> Account for the offset when calculating the period in nanoseconds. This
> fixes the discrepancy reported by the pwm core when CONFIG_PWM_DEBUG=1:
>
>     mediatek-disp-pwm 1100e000.pwm: .apply is not idempotent
>         (ena=1 pol=0 281500/499875) -> (ena=1 pol=0 281500/499750)
>
> Calculation discrepancy in the driver was found by AI run by a colleague
> while debugging another PWM related issue. Check against SoC datasheet
> and fix done by the author.
>
> Fixes: 3f2b16734914 ("pwm: mtk-disp: Implement atomic API .get_state()")
> Cc: Jitao Shi <jitao.shi@mediatek.com>
> Signed-off-by: Chen-Yu Tsai <wenst@chromium.org>

Reviewed-by: Fei Shao <fshao@chromium.org>