Add #[entry], #[irq] and #[exception(...)] macros.

[thejpster]

Adds:

• #[entry]
• #[exception(Undefined)]
• #[exception(SupervisorCall)]
• #[exception(PrefetchAbort)]
• #[exception(DataAbort)]
• #[irq] (or #[exception(Irq)])

You previously had to write:

#[unsafe(no_mangle)]
unsafe extern "C" fn _irq_handler() {
    todo!()
}

Now you can write:

#[irq]
fn irq_handler() {
    todo!()
}

[thejpster]

OK, I guess we can't set a default target for the whole repository. rebased without that.

[thejpster]

Uh, my re-entrant interrupt test is failing in CI, so I dropped that commit. It was unrelated to these changes anyway, so I'll fix it here and add it in another PR.

[robamu]

Maybe exception(IrqHandler) would be better as interrupt or interrupt_handler ?
It is technically an exception, but I associate exception with unexpected error cases.

[thejpster]

Yeah, I take your point. But Arm doesn't distinguish between handling an IRQ and handling an Abort. They are all just jumps from the vector table, and they have the same banked register state.

Is it worth creating a whole new macro just for the interrupt handler?

[thejpster]

I fixed the re-entrant interrupt handling. We need to preserve LR across the call to _irq_handler.

[thejpster]

I added #[interrupt] as an alias for #[exception(IrqHandler)]

[thejpster]

Force pushed with formatting fixed

[robamu]

I tested the interrupt macro on the Zynq7000, seems to work fine.

One more question:
The function signature(s) looks like this according to docs:

#[cortex_a_rt::exception(UndefinedHandler)]
fn my_handler(addr: usize) -> ! {
    todo!();
}

Is it possibly problematic to return the address like in the extern C function signature?

[thejpster]

Uh, the interrupt handler doesn't take an argument and also returns nothing - unlike the exception handlers (which take and return an address) and the SVCall handler (which takes the SVCall number and returns an empty tuple).

I should check the documentation is clear on this.

[thejpster]

Rebased, and then revised again - hopefully adding clarity and readability. In particular, I now make clear there are "Prefetch Aborts" and "Data Aborts", both handled in ABT mode.

I've left the handlers as safe functions, but I could be persuaded that the handlers that fn handler(usize) -> usize should be unsafe because returning a weird integer here will cause undefined behaviour. Basically the only sound thing you can do is return the argument, or the argument + 4.

[thejpster]

OK, here's what it's like if you require unsafe.

I think I prefer it this way?

[thejpster]

Good to go?

[robamu]

Yes, LGTM 👍