Auto merge of #128742 - RalfJung:miri-vtable-uniqueness, r=saethlin

miri: make vtable addresses not globally unique

Miri currently gives vtables a unique global address. That's not actually matching reality though. So this PR enables Miri to generate different addresses for the same type-trait pair.

To avoid generating an unbounded number of `AllocId` (and consuming unbounded amounts of memory), we use the "salt" technique that we also already use for giving constants non-unique addresses: the cache is keyed on a "salt" value n top of the actually relevant key, and Miri picks a random salt (currently in the range `0..16`) each time it needs to choose an `AllocId` for one of these globals -- that means we'll get up to 16 different addresses for each vtable. The salt scheme is integrated into the global allocation deduplication logic in `tcx`, and also used for functions and string literals. (So this also fixes the problem that casting the same function to a fn ptr over and over will consume unbounded memory.)

r? `@saethlin`
Fixes https://github.com/rust-lang/miri/issues/3737

compiler/rustc_const_eval/src/interpret/cast.rs

@@ -400,6 +400,12 @@ impl<'tcx, M: Machine<'tcx>> InterpCx<'tcx, M> {
             }
             (ty::Dynamic(data_a, _, ty::Dyn), ty::Dynamic(data_b, _, ty::Dyn)) => {
                 let val = self.read_immediate(src)?;
+                // MIR building generates odd NOP casts, prevent them from causing unexpected trouble.
+                // See <https://github.com/rust-lang/rust/issues/128880>.
+                // FIXME: ideally we wouldn't have to do this.
+                if data_a == data_b {
+                    return self.write_immediate(*val, dest);
+                }
                 // Take apart the old pointer, and find the dynamic type.
                 let (old_data, old_vptr) = val.to_scalar_pair();
                 let old_data = old_data.to_pointer(self)?;

compiler/rustc_const_eval/src/interpret/machine.rs

@@ -19,7 +19,7 @@ use rustc_target::spec::abi::Abi as CallAbi;
 use super::{
     throw_unsup, throw_unsup_format, AllocBytes, AllocId, AllocKind, AllocRange, Allocation,
     ConstAllocation, CtfeProvenance, FnArg, Frame, ImmTy, InterpCx, InterpResult, MPlaceTy,
-    MemoryKind, Misalignment, OpTy, PlaceTy, Pointer, Provenance,
+    MemoryKind, Misalignment, OpTy, PlaceTy, Pointer, Provenance, CTFE_ALLOC_SALT,
 };
 
 /// Data returned by [`Machine::after_stack_pop`], and consumed by
@@ -575,6 +575,14 @@ pub trait Machine<'tcx>: Sized {
     {
         eval(ecx, val, span, layout)
     }
+
+    /// Returns the salt to be used for a deduplicated global alloation.
+    /// If the allocation is for a function, the instance is provided as well
+    /// (this lets Miri ensure unique addresses for some functions).
+    fn get_global_alloc_salt(
+        ecx: &InterpCx<'tcx, Self>,
+        instance: Option<ty::Instance<'tcx>>,
+    ) -> usize;
 }
 
 /// A lot of the flexibility above is just needed for `Miri`, but all "compile-time" machines
@@ -677,4 +685,12 @@ pub macro compile_time_machine(<$tcx: lifetime>) {
         let (prov, offset) = ptr.into_parts();
         Some((prov.alloc_id(), offset, prov.immutable()))
     }
+
+    #[inline(always)]
+    fn get_global_alloc_salt(
+        _ecx: &InterpCx<$tcx, Self>,
+        _instance: Option<ty::Instance<$tcx>>,
+    ) -> usize {
+        CTFE_ALLOC_SALT
+    }
 }

compiler/rustc_const_eval/src/interpret/memory.rs

@@ -195,7 +195,10 @@ impl<'tcx, M: Machine<'tcx>> InterpCx<'tcx, M> {
 
     pub fn fn_ptr(&mut self, fn_val: FnVal<'tcx, M::ExtraFnVal>) -> Pointer<M::Provenance> {
         let id = match fn_val {
-            FnVal::Instance(instance) => self.tcx.reserve_and_set_fn_alloc(instance),
+            FnVal::Instance(instance) => {
+                let salt = M::get_global_alloc_salt(self, Some(instance));
+                self.tcx.reserve_and_set_fn_alloc(instance, salt)
+            }
             FnVal::Other(extra) => {
                 // FIXME(RalfJung): Should we have a cache here?
                 let id = self.tcx.reserve_alloc_id();

compiler/rustc_const_eval/src/interpret/place.rs

@@ -1008,7 +1008,8 @@ where
         // Use cache for immutable strings.
         let ptr = if mutbl.is_not() {
             // Use dedup'd allocation function.
-            let id = tcx.allocate_bytes_dedup(str.as_bytes());
+            let salt = M::get_global_alloc_salt(self, None);
+            let id = tcx.allocate_bytes_dedup(str.as_bytes(), salt);
 
             // Turn untagged "global" pointers (obtained via `tcx`) into the machine pointer to the allocation.
             M::adjust_alloc_root_pointer(&self, Pointer::from(id), Some(kind))?

compiler/rustc_const_eval/src/interpret/traits.rs

@@ -28,7 +28,9 @@ impl<'tcx, M: Machine<'tcx>> InterpCx<'tcx, M> {
         ensure_monomorphic_enough(*self.tcx, ty)?;
         ensure_monomorphic_enough(*self.tcx, poly_trait_ref)?;
 
-        let vtable_symbolic_allocation = self.tcx.reserve_and_set_vtable_alloc(ty, poly_trait_ref);
+        let salt = M::get_global_alloc_salt(self, None);
+        let vtable_symbolic_allocation =
+            self.tcx.reserve_and_set_vtable_alloc(ty, poly_trait_ref, salt);
         let vtable_ptr = self.global_root_pointer(Pointer::from(vtable_symbolic_allocation))?;
         Ok(vtable_ptr.into())
     }