Auto merge of #113892 - RalfJung:uninit-undef-poison, r=wesleywiser

clarify MIR uninit vs LLVM undef/poison

In [this LLVM discussion](https://discourse.llvm.org/t/rfc-load-instruction-uninitialized-memory-semantics/67481) I learned that mapping our uninitialized memory in MIR to poison in LLVM would be quite problematic due to the lack of a byte type. I am not sure where to write down this insight but this seems like a reasonable start.

Author: bors

compiler/rustc_codegen_ssa/src/traits/consts.rs

@@ -5,7 +5,13 @@ use rustc_target::abi;
 pub trait ConstMethods<'tcx>: BackendTypes {
     // Constant constructors
     fn const_null(&self, t: Self::Type) -> Self::Value;
+    /// Generate an uninitialized value (matching uninitialized memory in MIR).
+    /// Whether memory is initialized or not is tracked byte-for-byte.
     fn const_undef(&self, t: Self::Type) -> Self::Value;
+    /// Generate a fake value. Poison always affects the entire value, even if just a single byte is
+    /// poison. This can only be used in codepaths that are already UB, i.e., UB-free Rust code
+    /// (including code that e.g. copies uninit memory with `MaybeUninit`) can never encounter a
+    /// poison value.
     fn const_poison(&self, t: Self::Type) -> Self::Value;
     fn const_int(&self, t: Self::Type, i: i64) -> Self::Value;
     fn const_uint(&self, t: Self::Type, i: u64) -> Self::Value;

compiler/rustc_const_eval/src/interpret/operand.rs

@@ -31,7 +31,7 @@ pub enum Immediate<Prov: Provenance = AllocId> {
     /// A pair of two scalar value (must have `ScalarPair` ABI where both fields are
     /// `Scalar::Initialized`).
     ScalarPair(Scalar<Prov>, Scalar<Prov>),
-    /// A value of fully uninitialized memory. Can have and size and layout.
+    /// A value of fully uninitialized memory. Can have arbitrary size and layout.
     Uninit,
 }