Re: Aw: Re: [RFC v1 5/5] arm64: dts: mediatek: Add mt7986 based Bananapi R3 Mini

[Frank Wunderlich]

Am 9. Mai 2024 12:10:59 MESZ schrieb AngeloGioacchino Del Regno <angelogioacchino.delregno at collabora.com>:
>Il 08/05/24 20:25, Frank Wunderlich ha scritto:
>> Hi
>> 
>>> Gesendet: Dienstag, 07. Mai 2024 um 15:35 Uhr
>>> Von: "AngeloGioacchino Del Regno" <angelogioacchino.delregno at collabora.com>
>>> 
>>> Il 06/05/24 18:00, Frank Wunderlich ha scritto:
>> 
>>>>>> +	fan: pwm-fan {
>>>>>> +		compatible = "pwm-fan";
>>>>>> +		#cooling-cells = <2>;
>>>>>> +		/* cooling level (0, 1, 2) - pwm inverted */
>>>>>> +		cooling-levels = <255 96 0>;
>>>>> 
>>>>> Did you try to actually invert the PWM?
>>>>> 
>>>>> Look for PWM_POLARITY_INVERTED ;-)
>>>> 
>>>> Mtk pwm driver does not support it
>>>> 
>>>> https://elixir.bootlin.com/linux/latest/source/drivers/pwm/pwm-mediatek.c#L211
>>>> 
>>> 
>>> You're right, sorry - I confused the general purpose PWM controller with the
>>> rather specific DISP_PWM controller (which does support polarity inversion).
>>> 
>>> It's good - but I'd appreciate if you can please add a comment stating that
>>> the PWM values are inverted in SW because the controller does *not* support
>>> polarity inversion... so that next time someone looks at this will immediately
>>> understand what's going on and why :-)
>> 
>> so i would change comment like this:
>> 
>> 		/* cooling level (0, 1, 2)
>> 		 * signal is inverted on board
>> 		 * mtk pwm driver does not support
>> 		 * PWM_POLARITY_INVERTED */
>> 
>
>There you go:
>
>/*
> * The signal is inverted on this board and the general purpose
> * PWM HW IP in this SoC does not support polarity inversion.
> */
>/* Cooling level < 0  1  2> */
>cooling-levels = <255 96 0>;

Thanks for clearing structure of the comment,but imho actually it is a driver issue (for all mtk SoC). Not sure it is really a hardware limitation. So i would change this to "... and the PWM driver does not support polarity inversion."

>>>>>> +		pwms = <&pwm 0 10000>;
>>>>>> +		status = "okay";
>>>>>> +	};
>>>>>> +
>>>>>> +	phy14: ethernet-phy at 14 {
>> ...
>>>>>> +		interrupts-extended = <&pio 48 IRQ_TYPE_EDGE_FALLING>;
>>>>>> +		reset-gpios = <&pio 49 GPIO_ACTIVE_LOW>;
>>>>>> +		reset-assert-us = <10000>;
>>>>>> +		reset-deassert-us = <20000>;
>>>>>> +		phy-mode = "2500base-x";
>>>>>> +		full-duplex;
>>>>>> +		pause;
>>>>>> +		airoha,pnswap-rx;
>>>>>> +
>>>>>> +		leds {
>>>>>> +			#address-cells = <1>;
>>>>>> +			#size-cells = <0>;
>>>>>> +
>>>>>> +			led at 0 { /* en8811_a_gpio5 */
>>>>>> +				reg = <0>;
>>>>>> +				color = <LED_COLOR_ID_YELLOW>;
>>>>>> +				function = LED_FUNCTION_LAN;
>>>>>> +				function-enumerator = <1>;
>>>>> 
>>>>> Why aren't you simply using a label?
>>>> 
>>>> You mean the comment? I can add it of course like for regulators.
>>>> 
>>> 
>>> I mean in place of the function-enumerator... that's practically used to
>>> distinguish between instances, it's not too common to see it, and usually
>>> "label" replaces exactly that - just that, instead of a different number,
>>> it gets a different name with no (usually) meaningless numbers :-)
>> 
>> as far as i understand using label also makes "function" property useless, after discussing
>> this with eric i would drop both on all 4 places by labels like these:
>> 
>> label = "yellow-lan";
>> label = "green-lan";
>> ...
>> 
>> not sure if we should drop color property too...
>> 
>
>I'm looking at the leds binding (leds/common.yaml) right now.
>
>My suggestion of using 'label' was actually wrong - and your devicetree was
>actually right!!! (apart from the default-trigger that may not work)
>
>Infact, the documentation says, in brief:
>
>- function-enumerator is ignored if label is present
>- function doesn't say that gets ignored
>- color doesn't say that gets ignored
>- label says:
>  - If not present -> get string from node name
>  - function-enumerator ignored
>  - This property is deprecated
>
>...but the 'label' binding does not say 'deprecated: true', which is something
>that must be fixed!

Ok,i can try to add the property to binding (independ of this series). Imho label was cleaner than function and function-enumerator...

>So, I'm sorry for the confusion, the noise and the useless loss of time around
>this - you can keep the LED nodes as they are, and that's a lesson for the future
>me reviewing another node like this one.

Don't worry, we are all humas...i missed looking in linux-next for the other binding-patches.

>P.S.: This shouldn't have been a RFC, as the patches are more than RFC quality!!!

I sent it as RFC because i had not expected to be merged before next is closed.

>Cheers,
>Angelo
>
>>>>>> +				default-state = "keep";
>>>>>> +				linux,default-trigger = "netdev";
>>>>>> +			};
>>>>>> +			led at 1 { /* en8811_a_gpio4 */
>>>>>> +				reg = <1>;
>>>>>> +				color = <LED_COLOR_ID_GREEN>;
>>>>>> +				function = LED_FUNCTION_LAN;
>>>>>> +				function-enumerator = <2>;
>>>>>> +				default-state = "keep";
>>>>>> +				linux,default-trigger = "netdev";
>>>>>> +			};
>>>>>> +		};
>>>>>> +	};
>>>>>> +
>>>>>> +	phy15: ethernet-phy at 15 {
>>>>>> +		reg = <15>;


regards Frank