[strict provenance] make the compiler's tagged_ptr stuff conform

[Gankra]

This issue is part of the Strict Provenance Experiment - #95228

tagged_ptr and everything that uses it is entirely provenance crimes

rust/compiler/rustc_data_structures/src/tagged_ptr.rs

Lines 52 to 60 in bb5c437

	fn into_usize(self) -> usize;
	/// # Safety
	///
	/// The passed `ptr` must be returned from `into_usize`.
	///
	/// This acts as `ptr::read` semantically, it should not be called more than
	/// once on non-`Copy` `Pointer`s.
	unsafe fn from_usize(ptr: usize) -> Self;

This is the precise point in updating std+compiler to conform to strict-provenance that I ran out of energy and just started marking the compiler as WONTFIX for the MVP. I believe @eddyb started working on this, but got a bit sidetracked because they ran into the fact that we're shifting pointers down so that we can high-bit-tag instead of low-bit-tag, and while this isn't per-se a problem for strict-provenance, it wasn't obviously portable to CHERI, which sucks.

Since then it's been clarified that e.g. Morello (CHERI on ARM) actually has allowances for high-bit tagging and pointer shifting, so it's possible that the code can just be changed to properly use pointers and with_addr and friends.

[HKalbasi]

Fixed by #110243 ?

[RalfJung]

Yes seems like it. :)

@WaffleLapkin do you know other places in the compiler that are still using "fuzzy provenance casts"?

[WaffleLapkin]

@RalfJung There are still

rust/compiler/rustc_middle/src/ty/generic_args.rs

Line 34 in 92ad4b4

ptr: NonZeroUsize,

rust/compiler/rustc_middle/src/ty/mod.rs

Line 846 in 92ad4b4

ptr: NonZeroUsize,

I swear I was meaning to fix them... Maybe I forgot to open a PR? Will try to dig for it/write a new one.

Either way besides those two I don't remember and can't find any int2ptr casts in the compiler. Closing this as fixed by #110243.

[RalfJung]

@WaffleLapkin are we tracking those last two somewhere?

[WaffleLapkin]

@RalfJung not to my knowledge, I can open an issue though

[WaffleLapkin]

For the record I've opened #119217