arch: sw_isr: store device info in the table and add funtions to access

[ycsin]

Change the internal function to get_parent_entry, which returns the entire entry of table.

Store the parent interrupt controller device in the irq_parent_offset table, and added 2 helper functions to:

1. determine the parent interrupt controller based on the IRQ
2. determine the IRQ of the parent interrupt controller

Declare the struct _irq_parent_entry in the header and added so that it can be used to test the functions in testsuites.

Note

This change will increase the LUT size for build with multi-level enabled by CONFIG_NUM_x_LEVEL_AGGREGATORS * sizeof(struct device)

---

Note

This PR is spined off from #62907, to hopefully make it easier to review, and is required to fix #63465

Important

Blocking #62907

[galak]

We should NOT add a property for has-multi-instance to the devicetree, but scan the tree in some way to get the count of instances to determine its greater than 1.

[ycsin]

@cfriedt @galak the has-multi-instance prop was misleading, I was using it as a shortcut to get the controller that this implementation supports.

This implementation doesn't support INTMUX, i.e. intc_rv32m1_intmux, a single instance of it has multiple interrupts as child:

zephyr/dts/riscv/openisa/rv32m1.dtsi

Lines 88 to 177 in b4da11f

	intmux0: intmux@4004f000 {
	compatible = "openisa,rv32m1-intmux";
	reg = <0x4004f000 0x200>;
	clocks = <&pcc0 0x13c>;
	status = "disabled";
	#address-cells = <1>;
	#size-cells = <1>;
	ranges = <0x0 0x4004f000 0x200>;
	intmux0_ch0: interrupt-controller@0 {
	compatible = "openisa,rv32m1-intmux-ch";
	#address-cells = <0>;
	#interrupt-cells = <1>;
	interrupt-controller;
	interrupts = <INTMUX_CH0_IRQ>;
	reg = <0x0 0x40>;
	status = "disabled";
	};
	intmux0_ch1: interrupt-controller@40 {
	compatible = "openisa,rv32m1-intmux-ch";
	#address-cells = <0>;
	#interrupt-cells = <1>;
	interrupt-controller;
	interrupts = <INTMUX_CH1_IRQ>;
	reg = <0x40 0x40>;
	status = "disabled";
	};
	intmux0_ch2: interrupt-controller@80 {
	compatible = "openisa,rv32m1-intmux-ch";
	#address-cells = <0>;
	#interrupt-cells = <1>;
	interrupt-controller;
	interrupts = <INTMUX_CH2_IRQ>;
	reg = <0x80 0x40>;
	status = "disabled";
	};
	intmux0_ch3: interrupt-controller@c0 {
	compatible = "openisa,rv32m1-intmux-ch";
	#address-cells = <0>;
	#interrupt-cells = <1>;
	interrupt-controller;
	interrupts = <INTMUX_CH3_IRQ>;
	reg = <0xc0 0x40>;
	status = "disabled";
	};
	intmux0_ch4: interrupt-controller@100 {
	compatible = "openisa,rv32m1-intmux-ch";
	#address-cells = <0>;
	#interrupt-cells = <1>;
	interrupt-controller;
	interrupts = <INTMUX_CH4_IRQ>;
	reg = <0x100 0x40>;
	status = "disabled";
	};
	intmux0_ch5: interrupt-controller@140 {
	compatible = "openisa,rv32m1-intmux-ch";
	#address-cells = <0>;
	#interrupt-cells = <1>;
	interrupt-controller;
	interrupts = <INTMUX_CH5_IRQ>;
	reg = <0x140 0x40>;
	status = "disabled";
	};
	intmux0_ch6: interrupt-controller@180 {
	compatible = "openisa,rv32m1-intmux-ch";
	#address-cells = <0>;
	#interrupt-cells = <1>;
	interrupt-controller;
	interrupts = <INTMUX_CH6_IRQ>;
	reg = <0x180 0x40>;
	status = "disabled";
	};
	intmux0_ch7: interrupt-controller@1c0 {
	compatible = "openisa,rv32m1-intmux-ch";
	#address-cells = <0>;
	#interrupt-cells = <1>;
	interrupt-controller;
	interrupts = <INTMUX_CH7_IRQ>;
	reg = <0x1c0 0x40>;
	status = "disabled";
	};
	};

I've updated the filter to grab the interrupt controller that is the direct child of /soc instead, to just get the root interrupt controllers

[cfriedt]

Does this need to be conditionally defined?

E.g. can we remove the ifdef above, use if (IS_ENABLED()) elsewhere, and let the linker discard unused code and data?

[ycsin]

yeah it fails to compile as the initialization used some undefined macros when 3rd level isn't enabled

[fkokosinski]

LGTM from the perspective of #62907.

[carlescufi]

@dcpleung can you take a look?

[dcpleung]

TBH... we should probably start looking into DAG or tree-like hierarchy for multi-level interrupts so we can avoid making this even complicated in the future.

[ycsin]

TBH... we should probably start looking into DAG or tree-like hierarchy for multi-level interrupts so we can avoid making this even complicated in the future.

Where can I read more about this? Any existing implementation?

[dcpleung]

Maybe the Linux kernel? The idea is (probably) simple, where each interrupt controller is a node in a tree. For multi-level interrupts, each node is a child of some other node (except the root node). So you can have arbitrary levels of interrupts, and each controller can trace back to the root controller. Though, this is an overhaul of interrupt handling and is going to be a huge project... for soul(s) brave enough to tackle it.

[cfriedt]

Maybe the Linux kernel?

Yep. Linux would suffice. Of course, it's mostly about the model. The model will also drastically simplify shared interrupts with a marginal increase in complexity. It's also more stateful.

Zephyr's multi-level model (maybe surprisingly) started out simple too, of course 🤪