Re: [PATCH v4 1/2] dt-bindings: leds-lp55xx: add ti,charge-pump-mode

[Krzysztof Kozlowski]

On 02/02/2023 14:35, Maarten Zanders wrote:
> First off, bear with me here, I only recently started upstreaming 
> patches to kernel. It still feels like navigating an extremely busy 
> shipping lane... Either way, your feedback is highly valued.
> 
> On 2/2/23 14:15, Krzysztof Kozlowski wrote:
>>
>>> diff --git a/include/dt-bindings/leds/leds-lp55xx.h b/include/dt-bindings/leds/leds-lp55xx.h
>>> new file mode 100644
>>> index 000000000000..8f59c1c12dee
>>> --- /dev/null
>>> +++ b/include/dt-bindings/leds/leds-lp55xx.h
>>> @@ -0,0 +1,10 @@
>>> +/* SPDX-License-Identifier: GPL-2.0 */
>>> +#ifndef _DT_BINDINGS_LEDS_LP55XX_H
>>> +#define _DT_BINDINGS_LEDS_LP55XX_H
>>> +
>>> +#define LP55XX_CP_OFF		0
>>> +#define LP55XX_CP_BYPASS	1
>>> +#define LP55XX_CP_BOOST		2
>>> +#define LP55XX_CP_AUTO		3
>> Additionally, these are not used, so it's a dead binding. Drop. Sorry,
>> this is not the approach you should take.
>>
>> Best regards,
>> Krzysztof
>>
> These definitions are intended to be used in the DTS's, so it seems 
> normal to me that most of them go unused in the code? What am I missing?

Bindings mean drivers are using them. Your driver is not using it. It's
a register value, isn't it? Register values are not suitable for
bindings. There is no need for them to be in bindings.

> 
> As for the changes v2 > v3, this was based on input I got on v2 from Lee 
> Jones, maintainer for leds, on the implementation of the parsing of this 
> option:
> 
>>> +	pdata->charge_pump_mode = LP55XX_CP_AUTO;
>>> +	ret = of_property_read_string(np, "ti,charge-pump-mode", &pm);
>>> +	if (!ret) {
>>> +		for (cp_mode = LP55XX_CP_OFF;
>>> +		     cp_mode < ARRAY_SIZE(charge_pump_modes);
>>> +		     cp_mode++) {
>>> +			if (!strcasecmp(pm, charge_pump_modes[cp_mode])) {
>>> +				pdata->charge_pump_mode = cp_mode;
>>> +				break;
>>> +			}
>>> +		}
>>> +	}
>> A little over-engineered, no?
>>
>> Why not make the property a numerical value, then simply:
>>
>>    ret = of_property_read_u32(np, "ti,charge-pump-mode", &pdata->charge_pump_mode);
>>    if (ret)
>>            data->charge_pump_mode = LP55XX_CP_AUTO;
> 
> I found examples of similar configuration options of both types with 
> other drivers in the kernel tree (ie string & uint8). I can appreciate 
> both viewpoints but unfortunately cannot comply with both.

Strings in DTS are usually easier to for humans to read, but it's not a
requirement to use them. The problem of storing register values is that
binding is tied/coupled with hardware programming model, so you cannot
add a new device if the register value is a bit different (e.g.
LP55XX_CP_OFF is 0x1). You need entire new binding for such case. With
string - no need. With binding constants (IDs) also no need, so was this
the intention? Just to be clear - it is then ID or binding constant, not
a value for hardware register.

Best regards,
Krzysztof