The PWM currently functions, however if we try to utilise the pwn in a
device tree, for example as a pwm-backlight:

        lcd_bl: backlight {
                compatible = "pwm-backlight";
                pwms = <&mtu3 3 833333>;
		...

This fails:

[   15.603948] platform backlight: deferred probe pending: pwm-backlight: unable to request PWM

The PWM driver forms part of the Renesas Multi-Function Timer Pulse Unit
3. The PWM does not have a DT node of it's own. Share the DT node of the
parent MFD device, so that the PWM channels can be referenced via phandles.

Co-developed-by: Sebastian Reichel <sebastian.reichel@collabora.com>
Signed-off-by: Sebastian Reichel <sebastian.reichel@collabora.com>
Signed-off-by: Martyn Welch <martyn.welch@collabora.com>
---
 drivers/pwm/pwm-rz-mtu3.c | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/drivers/pwm/pwm-rz-mtu3.c b/drivers/pwm/pwm-rz-mtu3.c
index ab39bd37edafc..5825875fa0128 100644
--- a/drivers/pwm/pwm-rz-mtu3.c
+++ b/drivers/pwm/pwm-rz-mtu3.c
@@ -523,6 +523,12 @@ static int rz_mtu3_pwm_probe(struct platform_device *pdev)
 	if (ret < 0)
 		return ret;
 
+	/*
+	 * There is only one DT node, get it from the parent MFD device, so
+	 * that the PWM channels can be referenced via phandles
+	 */
+	dev->of_node = dev->parent->of_node;
+
 	chip->ops = &rz_mtu3_pwm_ops;
 	ret = devm_pwmchip_add(&pdev->dev, chip);
 	if (ret)
-- 
2.51.0

Hello,

[adding maintainers of drivers/of and Krzysztof to Cc:]

On Thu, Oct 09, 2025 at 05:24:44PM +0100, Martyn Welch wrote:
> The PWM currently functions, however if we try to utilise the pwn in a
> device tree, for example as a pwm-backlight:
> 
>         lcd_bl: backlight {
>                 compatible = "pwm-backlight";
>                 pwms = <&mtu3 3 833333>;
> 		...
> 
> This fails:
> 
> [   15.603948] platform backlight: deferred probe pending: pwm-backlight: unable to request PWM
> 
> The PWM driver forms part of the Renesas Multi-Function Timer Pulse Unit
> 3. The PWM does not have a DT node of it's own. Share the DT node of the
> parent MFD device, so that the PWM channels can be referenced via phandles.
> 
> Co-developed-by: Sebastian Reichel <sebastian.reichel@collabora.com>
> Signed-off-by: Sebastian Reichel <sebastian.reichel@collabora.com>
> Signed-off-by: Martyn Welch <martyn.welch@collabora.com>
> ---
>  drivers/pwm/pwm-rz-mtu3.c | 6 ++++++
>  1 file changed, 6 insertions(+)
> 
> diff --git a/drivers/pwm/pwm-rz-mtu3.c b/drivers/pwm/pwm-rz-mtu3.c
> index ab39bd37edafc..5825875fa0128 100644
> --- a/drivers/pwm/pwm-rz-mtu3.c
> +++ b/drivers/pwm/pwm-rz-mtu3.c
> @@ -523,6 +523,12 @@ static int rz_mtu3_pwm_probe(struct platform_device *pdev)
>  	if (ret < 0)
>  		return ret;
>  
> +	/*
> +	 * There is only one DT node, get it from the parent MFD device, so
> +	 * that the PWM channels can be referenced via phandles
> +	 */
> +	dev->of_node = dev->parent->of_node;
> +

I (very quickly) talked to Krzysztof about this. He said that
of_node_get() should probably be used here. I wonder if
device_add_of_node() is the right function to use (which uses
of_node_get(), also handles fwnode and implements some safeguards).

>  	chip->ops = &rz_mtu3_pwm_ops;
>  	ret = devm_pwmchip_add(&pdev->dev, chip);
>  	if (ret)

Best regards
Uwe

On 21/10/2025 12:19, Uwe Kleine-König wrote:
>> diff --git a/drivers/pwm/pwm-rz-mtu3.c b/drivers/pwm/pwm-rz-mtu3.c
>> index ab39bd37edafc..5825875fa0128 100644
>> --- a/drivers/pwm/pwm-rz-mtu3.c
>> +++ b/drivers/pwm/pwm-rz-mtu3.c
>> @@ -523,6 +523,12 @@ static int rz_mtu3_pwm_probe(struct platform_device *pdev)
>>  	if (ret < 0)
>>  		return ret;
>>  
>> +	/*
>> +	 * There is only one DT node, get it from the parent MFD device, so
>> +	 * that the PWM channels can be referenced via phandles
>> +	 */
>> +	dev->of_node = dev->parent->of_node;
>> +
> 
> I (very quickly) talked to Krzysztof about this. He said that
> of_node_get() should probably be used here. I wonder if
> device_add_of_node() is the right function to use (which uses
> of_node_get(), also handles fwnode and implements some safeguards).


I am not so sure about device_add_of_node(). You do not need to
get_device(), because reference is already hold. Although setting
dev->fwnode might make sense... But, not that important I think, works
with me.


Best regards,
Krzysztof

Hello,

On Tue, Oct 21, 2025 at 12:47:52PM +0200, Krzysztof Kozlowski wrote:
> On 21/10/2025 12:19, Uwe Kleine-König wrote:
> >> diff --git a/drivers/pwm/pwm-rz-mtu3.c b/drivers/pwm/pwm-rz-mtu3.c
> >> index ab39bd37edafc..5825875fa0128 100644
> >> --- a/drivers/pwm/pwm-rz-mtu3.c
> >> +++ b/drivers/pwm/pwm-rz-mtu3.c
> >> @@ -523,6 +523,12 @@ static int rz_mtu3_pwm_probe(struct platform_device *pdev)
> >>  	if (ret < 0)
> >>  		return ret;
> >>  
> >> +	/*
> >> +	 * There is only one DT node, get it from the parent MFD device, so
> >> +	 * that the PWM channels can be referenced via phandles
> >> +	 */
> >> +	dev->of_node = dev->parent->of_node;
> >> +
> > 
> > I (very quickly) talked to Krzysztof about this. He said that
> > of_node_get() should probably be used here. I wonder if
> > device_add_of_node() is the right function to use (which uses
> > of_node_get(), also handles fwnode and implements some safeguards).
> 
> 
> I am not so sure about device_add_of_node(). You do not need to
> get_device(), because reference is already hold. Although setting
> dev->fwnode might make sense... But, not that important I think, works
> with me.

Note that device_add_of_node() only holds the get_device() reference
temporarily as it calls put_device() before returning. So that's only to
assert that the device doesn't disappear in-flight. That "locking" might
not be needed here, but it also doesn't do any harm (apart from a minor
runtime overhead).

Best regards
Uwe