[embassy-rp] Enable interrupts in init function

[sourcebox]

Fix for #4982

[Dirbaio]

This is risky, the user might have good reasons to run init with interrupts disabled.

• It might be inside a critical section, we'd be "prematurely ending" it which can cause unsoundness.
• RTIC's init() runs with interrupts disabled because the resource sharing stuff isn't fully initialized yet, forcing enabling interrupts can probably cause unsoundness too.

[sourcebox]

So what would be the alternative to make it independent of the bootloader? Because the risks you describe are probably also present when using a bootloader that enables them by default.

[Dirbaio]

So what would be the alternative to make it independent of the bootloader?

I don't think it's possible to fix. The current convention is irqs are enabled at boot. Libraries can't enable irqs without risking unsoundness. If you have a broken bootloader that jumps to the app with irqs disabled, it's up to you, the user, to unsafely enable irqs at start of main when you know it can't cause unsoundness.

Because the risks you describe are probably also present when using a bootloader that enables them by default.

No, because the firmware will disable irqs if it's necessary for soundness, it doesn't rely on irqs being disabled on boot.

[Dirbaio]

Maybe we can do it in pre_init. That shouldn't cause any unsoundness.

Disable+unpend all irqs individually in NVIC, then enable global interrupts.

[sourcebox]

Ok, I close this PR so we can check a better solution.