Reimplement `r0`, since it is deprecated

Question

Reimplement `r0`, since it is deprecated

coastalwhite opened this issue a year ago · 7 comments

This repository should find an alternative to r0, since that crate is now archived and deprecated.

Answer 1 · 2023-09-21T09:27:57.000Z

Thanks for opening this issue! I'll take a look. Of course, PRs are more than welcome :)

Answer 2 · 2023-09-21T09:48:24.000Z

I’ll have a look at this. Is there a preference for a separate crate again or should it just become a part of the riscv-rt crate?

Answer 3 · 2023-09-21T09:55:33.000Z

I prefer to reduce the number of dependencies, but if you find a handy crate that eases the work, then use it.

If I were you, I'd implement it in assembly.

Answer 4 · 2023-09-21T22:40:00.000Z

The conclusion when we deprecated r0 was that this can only soundly be implemented in assembly; cortex-m-rt does so here. I think it's best put into your existing startup routine (here).

Answer 5 · 2023-09-22T08:09:02.000Z

I see you feature-gate the RAM initialization. Would you recommend us to do the same for riscv-rt?

Answer 6 · 2023-09-22T09:33:55.000Z

We can get some inspiration from the esp-hal folks: https://github.com/esp-rs/esp-hal/blob/eeb2d1db6f8ffd250b212bd1bebaddc83d0ea8b0/esp-riscv-rt/src/lib.rs#L294C8-L294C8

Answer 7 · 2023-09-23T01:41:33.000Z

I see you feature-gate the RAM initialization. Would you recommend us to do the same for riscv-rt?

That was a later addition from users with MCUs that have things like ECC RAM where you have to write to the entire RAM at startup to initialise the memory controller; usually it's enough to just initialise the .data and .bss sections and leave unused RAM alone. Since there's a startup time cost, it made sense to feature-gate this just for users who need it.

I think you could reasonably support the same feature, but just doing .data and .bss is enough for almost all use cases.