Make thread local accesses an `Rvalue` in MIR

[oli-obk]

#70598 (comment)

Thread local accesses are a runtime operation, even if LLVM shims them to appear like normal static accesses. We should encode this runtime-ness in MIR and translate them back to the llvm static accesses in the llvm backend. The other backends (miri, cranelift) then have an easier life at doing the right thing ^TM.

cc @bjorn3 @eddyb @vakaras

[eddyb]

Copying my #70684 (comment) here
(note "the address of a #[thread_local] static", not "access", the address itself is runtime):

---

As I suggested elsewhere, I believe #[thread_local] statics should have their own separate access mechanism, probably their own Rvalue (or we could start extending what Rvalue::{Ref,RawPtr} can borrow).

That's because getting the address of a #[thread_local] static is a fundamentally runtime operation, and LLVM reusing the same syntax as for regular statics, is in the wrong here.

[vakaras]

@eddyb @oli-obk
Do you have an estimate how much effort this would require? If it is doable in a day or two and you could give me the mentoring instructions, I could try implementing this.

[eddyb]

No idea, cc @matthewjasper @wesleywiser @spastorino (I forget who did the move to constant pointers for statics) thanks @bjorn3!

[bjorn3]

I forget who did the move to constant pointers for statics

That was done in #67000.