support default property values in devicetree bindings

[pabigot]

I have a device for which the binding should include a property like this:

startup-delay-us:
    type: int
    category: optional
    description: >
        Startup time, in microseconds.

        The driver will busy-wait for this period before returning
        from a call that enables the regulator.

where absence of the property means that no delay is required.

There may be multiple of these devices, so I'd like the driver implementation file to use a trick like this[*]:

#ifdef DT_REGULATOR_GPIO_0
REGULATOR_GPIO_DEVICE(0);
#endif
#ifdef DT_REGULATOR_GPIO_1
REGULATOR_GPIO_DEVICE(1);
#endif
/* ... */

The killer is that REGULATOR_GPIO_DEVICE(id) needs to populate a config structure field like this:

    .startup_us = CPP_MAGIC_WRAPPER(DT_INST_##id##_REGULATOR_GPIO_STARTUP_DELAY_US, 0),

DT_INST_0_REGULATOR_GPIO_STARTUP_DELAY_US will either be undefined, or will be defined to a value that is unlikely to be 1, so preprocessor tricks like COND_CODE() won't work for CPP_MAGIC_WRAPPER.

If I could add this to the yaml:

    default: 0

and have the Zephyr scripting provide the default if there isn't one in the node that would make life easier. (Because there are actually multiple properties of this type, and having to do:

#ifdef REGULATOR_GPIO_0
#ifndef DT_INST_0_REGULATOR_GPIO_STARTUP_DELAY_US
#define DT_INST_0_REGULATOR_GPIO_STARTUP_DELAY_US 0
#endif
...
REGULATOR_GPIO_DEVICE(0)

rapidly becomes tedious.)

Is there another way to solve this?

[*] well, no, I'd like to have something that does that without having to unroll all the instances up to an arbitrary maximum, but I don't know how to make that happen, either.

[ulfalizer]

Only skimmed, but would making startup-delay-us obligatory work? The description string could mention that 0 works and means no delay.

[pabigot]

That's what I'm going to do, but it's not ideal since the common case is there is no delay and so no need for the property to be specified. It just clutters the devicetree source.

This is at least the third time I've had a situation where I want a property to be optional in the device tree because I can infer a reasonable default in the driver from other information. This is the first time I've had to handle multiple instances of the device, and C simply doesn't provide the preprocessor capability to abstract the conditional into the generator macro.

[ulfalizer]

Yeah... just spent a while trying to make an IS_ENABLED()-like macro for it, before I realized why it's tricky (can't tell whether something is the result of a macro replacement or not). :P

Seems like a reasonable feature now, even if some #ifdefs might not be horrible in that case.

I'll add it to my todo list.

[galak]

I'm not a fan of this approach, if a default is desired, just have the value set in the .dts file.

[pabigot]

@galak is that a "reject" or merely "I don't like it but won't block it"? I was about to start using this via #17928, but if it's not acceptable I won't waste my time.