raw_access

[burrbull]

Related to rust-embedded/wg#791

cc @adamgreig

[adamgreig]

I'm not sure if there's any advantage to skipping volatile_cell but still having the registers be themselves memory-mapped (with &self cast to a pointer). We still have the problem that Rust can dereference &self and perform an invalid read. Probably we have to do something similar to chiptool where the peripheral struct owns a pointer and the register methods offset it - the total memory cost is the same because instead of having reference-to-a-ZST like now, we would have an owned-pointer.

It might be hard to do this and keep vcell too, but maybe we can just remove vcell support at the same time?

[burrbull]

It might be hard to do this and keep vcell too, but maybe we can just remove vcell support at the same time?

It is not hard. svd2rust code is flexible enough.

Review yet one time, please.

P.S. Let's merge #842 first

[burrbull]

/ci diff pr --current " --raw-access"

[github-actions]

Diff for comment

• pr --current " --raw-access"

[romancardenas]

Any news on this? May I help somehow?

[burrbull]

@romancardenas I've rebased but I'm not sure there is nothing missed.

[Emilgardis]

/ci diff pr --current " --raw-access"

[github-actions]

Diff for comment

• pr --current " --raw-access"

[Emilgardis]

Do we not want &#ty here? Same for the other accessors

[romancardenas]

I think that, with raw access, peripherals return pointers to registers instead of a reference to one of the fields of their register block. Thus, instead of borrowing access, users are "owners" of these registers.