scripts: kconfig: Add dt_compat_get_inst_prop method

[nandojve]

The dt_str_val method has been deprecated. This add an alternative
that uses compat and instance number to find a node and than
return the string representation or a passed default value.

This is necessary becase path is valid only at SoC directory. For
instance, a driver for a SPI bus need search for a compat based on
the instance to determine the proper configuration.

depends on: #22215

Example: how determine the right interface name from the device tree
node label.

at DT:

&foo0 {
	compatible = "manufacturer,device"
	status = "disabled";
	label = "FOO_0";
};
&foo1 {
	compatible = "manufacturer,device"
	status = "okay";
	label = "FOO_1";
};

at Kconfig:

DT_FOO      := manufacturer,device
DT_FOO_INST := 0
DT_FOO_DEF  := "default string on not match/error - optional empty"

config FOO
	string "text"
	default "$(dt_compat_get_str, \
		 label, \
		 $(DT_FOO), \
		 $(DT_FOO_INST), \
		 $(DT_FOO_DEF))"

The result is CONFIG_FOO="FOO_1"

An example of use was delivered too.

[nandojve]

Rebase and Fix 364 line.

        if prop not in node.props:
            continue

to

        if prop not in node.props:
            break

[ulfalizer]

Some nits and random thoughts.

Having some generic property lookup functions like these seems nice.

[nandojve]

I assumed that instance_no will be enough to check but testing in deep I saw inconsistencies.

&foo0 {
	status = "disabled";
	bar@0 {
		compatible = "man,bar";
		reg = <0x0>;
		label = "WPAN_0";
		status = "disabled";
	};
	bar@1 {
		compatible = "man,bar";
		reg = <0x1>;
		label = "WPAN_1";
		status = "okay";
	};
};

&foo1 {
	status = "okay";
	bar@0 {
		compatible = "man,bar";
		reg = <0x0>;
		label = "RF2XX_2";
		status = "okay";
	};
}; 

On this example the first instance should be /../foo1/bar@0 and must be the only one. The current version returns /../foo0/bar@0. At generated_dts_board.conf file I saw two instances /../foo0/bar@1 and /../foo1/bar@0.

Added scripts: dts: edtlib.py: Fix DT enabled node property to address the issue and fix unnecessary instances at generated_dts_board.conf

[ulfalizer]

Having status = "disabled" disable all the child nodes as well makes sense to me. Thanks for that.

I wonder how common it is to have nested status properties, but if they show up, it probably makes sense for the parent status to take precedence.

Feels like the old _set_instance_no() code could be cleaned up a bit. It's pretty subtle how it works (depends on edt.nodes only including the "previous" nodes when the function is called).

I'm thinking it might be nice to have an edt.compat2nodes property, where e.g. edt.compat2nodes["foo"] would be a list with all enabled nodes that have compatible = ..., "foo", .... That would simplify dt_compat_get_str() as well.

I'll make a separate PR for that.

[ulfalizer]

PR that turns Node.instance_no into EDT.compat2nodes: #21592

[ulfalizer]

Seems alright to me. Nice to have a function like that. Could be simplified a bit once #21592 is in too. Suspect most people are vacation at the moment.

Wonder if the function could just be called dt_compat_get too, since it also works for integers. Maybe there could be a dt_compat_get_array later if needed, that also takes an array index.

[nandojve]

Seems alright to me. Nice to have a function like that. Could be simplified a bit once #21592 is in too. Suspect most people are vacation at the moment.

Wonder if the function could just be called dt_compat_get too, since it also works for integers. Maybe there could be a dt_compat_get_array later if needed, that also takes an array index.

I agreed with you, make sense.

[nandojve]

Rebase on top of dts: edtlib: Turn Node.instance_no into more flexible EDT.compat2nodes #21592

[zephyrbot]

All checks passed.

Tip: The bot edits this comment instead of posting a new one, so you can check the comment's history to see earlier messages.

[tbursztyka]

Mixing up DTS into Kconfig is, imo, a mistake.
After almost 3 years we have been using DTS, we are still not yet where we want to be (too many instance-centric options are still in Kconfig, etc..)
And that PR makes it worse by mixing both. Definitely not a good idea.

[nandojve]

Mixing up DTS into Kconfig is, imo, a mistake.
After almost 3 years we have been using DTS, we are still not yet where we want to be (too many instance-centric options are still in Kconfig, etc..)
And that PR makes it worse by mixing both. Definitely not a good idea.

The initial motivation of this PR was an open discussion at #21402. I agreed with you after weeks looking around DT, fixups, Kconfig etc. I believe any driver definition/configuration must be on that place and not around.

I would like mention that there are DT improvements on #21592 and bug fix at 'scripts: dts: edtlib.py: Fix DT enabled node property' that probably need move forward.

[nandojve]

Created "scripts: dts: edtlib.py: Fix DT enabled node property" #22215 dependency to separate DT fix from the current PR.

[ulfalizer]

@tbursztyka

Mixing up DTS into Kconfig is, imo, a mistake.
After almost 3 years we have been using DTS, we are still not yet where we want to be (too many instance-centric options are still in Kconfig, etc..)
And that PR makes it worse by mixing both. Definitely not a good idea.

I think the problem is mostly Kconfig options that allow you to change stuff that's already known from selecting the board.

Think of Kconfig like an options menu in a game. You wouldn't show an option for selecting the operating system there, since it's already known. Kconfig is supposed to be simple like that, but the concepts got lost.

Having devicetree information available to Kconfig is something else though. It allows you to tweak the options based on what the devicetree says (e.g. to only show Windows-specific options if the game runs on Windows, just to extend that example).

This PR might help with getting rid of devicetree.conf too, which would be nice.

[nandojve]

Since there isn't movement here I'm closing.