Rollup merge of #147744 - RalfJung:miri, r=RalfJung

miri subtree update

Subtree update of `miri` to rust-lang/miri@50ba3a7.

Created using https://github.com/rust-lang/josh-sync.

r? ``````@ghost``````

Author: matthiaskrgr

Cargo.lock

@@ -1164,7 +1164,7 @@ dependencies = [
  "libc",
  "option-ext",
  "redox_users 0.5.2",
- "windows-sys 0.60.2",
+ "windows-sys 0.61.2",
 ]
 
 [[package]]
@@ -2106,19 +2106,19 @@ dependencies = [
 
 [[package]]
 name = "libffi"
-version = "4.1.1"
+version = "5.0.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "e7681c6fab541f799a829e44a445a0666cf8d8a6cfebf89419e6aed52c604e87"
+checksum = "0444124f3ffd67e1b0b0c661a7f81a278a135eb54aaad4078e79fbc8be50c8a5"
 dependencies = [
  "libc",
  "libffi-sys",
 ]
 
 [[package]]
 name = "libffi-sys"
-version = "3.3.2"
+version = "4.0.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "7b0d828d367b4450ed08e7d510dc46636cd660055f50d67ac943bfe788767c29"
+checksum = "3d722da8817ea580d0669da6babe2262d7b86a1af1103da24102b8bb9c101ce7"
 dependencies = [
  "cc",
 ]
@@ -2374,7 +2374,7 @@ version = "0.6.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "536bfad37a309d62069485248eeaba1e8d9853aaf951caaeaed0585a95346f08"
 dependencies = [
- "windows-sys 0.60.2",
+ "windows-sys 0.61.2",
 ]
 
 [[package]]

src/tools/miri/Cargo.lock

@@ -794,19 +794,19 @@ checksum = "1171693293099992e19cddea4e8b849964e9846f4acee11b3948bcc337be8776"
 
 [[package]]
 name = "libffi"
-version = "4.1.1"
+version = "5.0.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "e7681c6fab541f799a829e44a445a0666cf8d8a6cfebf89419e6aed52c604e87"
+checksum = "0444124f3ffd67e1b0b0c661a7f81a278a135eb54aaad4078e79fbc8be50c8a5"
 dependencies = [
  "libc",
  "libffi-sys",
 ]
 
 [[package]]
 name = "libffi-sys"
-version = "3.3.2"
+version = "4.0.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "7b0d828d367b4450ed08e7d510dc46636cd660055f50d67ac943bfe788767c29"
+checksum = "3d722da8817ea580d0669da6babe2262d7b86a1af1103da24102b8bb9c101ce7"
 dependencies = [
  "cc",
 ]

src/tools/miri/Cargo.toml

@@ -39,7 +39,7 @@ features = ['unprefixed_malloc_on_supported_platforms']
 [target.'cfg(unix)'.dependencies]
 libc = "0.2"
 # native-lib dependencies
-libffi = { version = "4.1.1", optional = true }
+libffi = { version = "5.0.0", optional = true }
 libloading = { version = "0.8", optional = true }
 serde = { version = "1.0.219", features = ["derive"], optional = true }
 

src/tools/miri/rust-version

@@ -1 +1 @@
-36e4f5d1fe1d63953a5bf1758ce2b64172623e2e
+28d0a4a205f9e511ad2f51ee79a4aa19a704a455

src/tools/miri/src/concurrency/genmc/global_allocations.rs

@@ -6,7 +6,7 @@ use rand::SeedableRng;
 use rand::rngs::StdRng;
 use rustc_const_eval::interpret::{AllocId, AllocInfo, InterpResult, interp_ok};
 use rustc_data_structures::fx::FxHashMap;
-use tracing::debug;
+use rustc_log::tracing::debug;
 
 use crate::alloc_addresses::AddressGenerator;
 

src/tools/miri/src/concurrency/genmc/run.rs

@@ -3,6 +3,7 @@ use std::sync::Arc;
 use std::time::Instant;
 
 use genmc_sys::EstimationResult;
+use rustc_log::tracing;
 use rustc_middle::ty::TyCtxt;
 
 use super::GlobalState;

src/tools/miri/src/concurrency/genmc/scheduling.rs

@@ -1,4 +1,5 @@
 use genmc_sys::{ActionKind, ExecutionState};
+use rustc_data_structures::either::Either;
 use rustc_middle::mir::TerminatorKind;
 use rustc_middle::ty::{self, Ty};
 
@@ -38,7 +39,7 @@ fn get_next_instruction_kind<'tcx>(
     let Some(frame) = thread_manager.active_thread_stack().last() else {
         return interp_ok(NonAtomic);
     };
-    let either::Either::Left(loc) = frame.current_loc() else {
+    let Either::Left(loc) = frame.current_loc() else {
         // We are unwinding, so the next step is definitely not atomic.
         return interp_ok(NonAtomic);
     };

src/tools/miri/src/diagnostics.rs

@@ -139,6 +139,7 @@ pub enum NonHaltingDiagnostic {
     NativeCallSharedMem {
         tracing: bool,
     },
+    NativeCallFnPtr,
     WeakMemoryOutdatedLoad {
         ptr: Pointer,
     },
@@ -644,6 +645,11 @@ impl<'tcx> MiriMachine<'tcx> {
             Int2Ptr { .. } => ("integer-to-pointer cast".to_string(), DiagLevel::Warning),
             NativeCallSharedMem { .. } =>
                 ("sharing memory with a native function".to_string(), DiagLevel::Warning),
+            NativeCallFnPtr =>
+                (
+                    "sharing a function pointer with a native function".to_string(),
+                    DiagLevel::Warning,
+                ),
             ExternTypeReborrow =>
                 ("reborrow of reference to `extern type`".to_string(), DiagLevel::Warning),
             GenmcCompareExchangeWeak | GenmcCompareExchangeOrderingMismatch { .. } =>
@@ -682,6 +688,8 @@ impl<'tcx> MiriMachine<'tcx> {
             Int2Ptr { .. } => format!("integer-to-pointer cast"),
             NativeCallSharedMem { .. } =>
                 format!("sharing memory with a native function called via FFI"),
+            NativeCallFnPtr =>
+                format!("sharing a function pointer with a native function called via FFI"),
             WeakMemoryOutdatedLoad { ptr } =>
                 format!("weak memory emulation: outdated value returned from load at {ptr}"),
             ExternTypeReborrow =>
@@ -779,6 +787,11 @@ impl<'tcx> MiriMachine<'tcx> {
                         ),
                     ]
                 },
+            NativeCallFnPtr => {
+                vec![note!(
+                    "calling Rust functions from C is not supported and will, in the best case, crash the program"
+                )]
+            }
             ExternTypeReborrow => {
                 assert!(self.borrow_tracker.as_ref().is_some_and(|b| {
                     matches!(

src/tools/miri/src/shims/native_lib/ffi.rs

@@ -9,11 +9,9 @@ use libffi::middle::{Arg as ArgPtr, Cif, Type as FfiType};
 ///
 /// The safety invariants of the foreign function being called must be upheld (if any).
 pub unsafe fn call<R: libffi::high::CType>(fun: CodePtr, args: &mut [OwnedArg]) -> R {
-    let arg_ptrs: Vec<_> = args.iter().map(|arg| arg.ptr()).collect();
     let cif = Cif::new(args.iter_mut().map(|arg| arg.ty.take().unwrap()), R::reify().into_middle());
-    // SAFETY: Caller upholds that the function is safe to call, and since we
-    // were passed a slice reference we know the `OwnedArg`s won't have been
-    // dropped by this point.
+    let arg_ptrs: Vec<_> = args.iter().map(|arg| ArgPtr::new(&*arg.bytes)).collect();
+    // SAFETY: Caller upholds that the function is safe to call.
     unsafe { cif.call(fun, &arg_ptrs) }
 }
 
@@ -31,16 +29,4 @@ impl OwnedArg {
     pub fn new(ty: FfiType, bytes: Box<[u8]>) -> Self {
         Self { ty: Some(ty), bytes }
     }
-
-    /// Creates a libffi argument pointer pointing to this argument's bytes.
-    /// NB: Since `libffi::middle::Arg` ignores the lifetime of the reference
-    /// it's derived from, it is up to the caller to ensure the `OwnedArg` is
-    /// not dropped before unsafely calling `libffi::middle::Cif::call()`!
-    fn ptr(&self) -> ArgPtr {
-        // FIXME: Using `&self.bytes[0]` to reference the whole array is
-        // definitely unsound under SB, but we're waiting on
-        // https://github.com/libffi-rs/libffi-rs/commit/112a37b3b6ffb35bd75241fbcc580de40ba74a73
-        // to land in a release so that we don't need to do this.
-        ArgPtr::new(&self.bytes[0])
-    }
 }

src/tools/miri/src/shims/native_lib/mod.rs

@@ -7,7 +7,7 @@ use libffi::low::CodePtr;
 use libffi::middle::Type as FfiType;
 use rustc_abi::{HasDataLayout, Size};
 use rustc_data_structures::either;
-use rustc_middle::ty::layout::HasTypingEnv;
+use rustc_middle::ty::layout::TyAndLayout;
 use rustc_middle::ty::{self, IntTy, Ty, UintTy};
 use rustc_span::Symbol;
 use serde::{Deserialize, Serialize};
@@ -278,7 +278,7 @@ trait EvalContextExtPriv<'tcx>: crate::MiriInterpCxExt<'tcx> {
 
         // This should go first so that we emit unsupported before doing a bunch
         // of extra work for types that aren't supported yet.
-        let ty = this.ty_to_ffitype(v.layout.ty)?;
+        let ty = this.ty_to_ffitype(v.layout)?;
 
         // Helper to print a warning when a pointer is shared with the native code.
         let expose = |prov: Provenance| -> InterpResult<'tcx> {
@@ -387,34 +387,44 @@ trait EvalContextExtPriv<'tcx>: crate::MiriInterpCxExt<'tcx> {
         let this = self.eval_context_ref();
         let mut fields = vec![];
         for field in &adt_def.non_enum_variant().fields {
-            fields.push(this.ty_to_ffitype(field.ty(*this.tcx, args))?);
+            let layout = this.layout_of(field.ty(*this.tcx, args))?;
+            fields.push(this.ty_to_ffitype(layout)?);
         }
 
         interp_ok(FfiType::structure(fields))
     }
 
     /// Gets the matching libffi type for a given Ty.
-    fn ty_to_ffitype(&self, ty: Ty<'tcx>) -> InterpResult<'tcx, FfiType> {
-        let this = self.eval_context_ref();
-        interp_ok(match ty.kind() {
-            ty::Int(IntTy::I8) => FfiType::i8(),
-            ty::Int(IntTy::I16) => FfiType::i16(),
-            ty::Int(IntTy::I32) => FfiType::i32(),
-            ty::Int(IntTy::I64) => FfiType::i64(),
-            ty::Int(IntTy::Isize) => FfiType::isize(),
-            ty::Uint(UintTy::U8) => FfiType::u8(),
-            ty::Uint(UintTy::U16) => FfiType::u16(),
-            ty::Uint(UintTy::U32) => FfiType::u32(),
-            ty::Uint(UintTy::U64) => FfiType::u64(),
-            ty::Uint(UintTy::Usize) => FfiType::usize(),
-            ty::RawPtr(pointee_ty, _mut) => {
-                if !pointee_ty.is_sized(*this.tcx, this.typing_env()) {
-                    throw_unsup_format!("passing a pointer to an unsized type over FFI: {}", ty);
-                }
-                FfiType::pointer()
-            }
-            ty::Adt(adt_def, args) => self.adt_to_ffitype(ty, *adt_def, args)?,
-            _ => throw_unsup_format!("unsupported argument type for native call: {}", ty),
+    fn ty_to_ffitype(&self, layout: TyAndLayout<'tcx>) -> InterpResult<'tcx, FfiType> {
+        use rustc_abi::{AddressSpace, BackendRepr, Integer, Primitive};
+
+        // `BackendRepr::Scalar` is also a signal to pass this type as a scalar in the ABI. This
+        // matches what codegen does. This does mean that we support some types whose ABI is not
+        // stable, but that's fine -- we are anyway quite conservative in native-lib mode.
+        if let BackendRepr::Scalar(s) = layout.backend_repr {
+            // Simple sanity-check: this cannot be `repr(C)`.
+            assert!(!layout.ty.ty_adt_def().is_some_and(|adt| adt.repr().c()));
+            return interp_ok(match s.primitive() {
+                Primitive::Int(Integer::I8, /* signed */ true) => FfiType::i8(),
+                Primitive::Int(Integer::I16, /* signed */ true) => FfiType::i16(),
+                Primitive::Int(Integer::I32, /* signed */ true) => FfiType::i32(),
+                Primitive::Int(Integer::I64, /* signed */ true) => FfiType::i64(),
+                Primitive::Int(Integer::I8, /* signed */ false) => FfiType::u8(),
+                Primitive::Int(Integer::I16, /* signed */ false) => FfiType::u16(),
+                Primitive::Int(Integer::I32, /* signed */ false) => FfiType::u32(),
+                Primitive::Int(Integer::I64, /* signed */ false) => FfiType::u64(),
+                Primitive::Pointer(AddressSpace::ZERO) => FfiType::pointer(),
+                _ =>
+                    throw_unsup_format!(
+                        "unsupported scalar argument type for native call: {}",
+                        layout.ty
+                    ),
+            });
+        }
+        interp_ok(match layout.ty.kind() {
+            // Scalar types have already been handled above.
+            ty::Adt(adt_def, args) => self.adt_to_ffitype(layout.ty, *adt_def, args)?,
+            _ => throw_unsup_format!("unsupported argument type for native call: {}", layout.ty),
         })
     }
 }
@@ -455,6 +465,13 @@ pub trait EvalContextExt<'tcx>: crate::MiriInterpCxExt<'tcx> {
         // pointer was passed as argument). Uninitialised memory is left as-is, but any data
         // exposed this way is garbage anyway.
         this.visit_reachable_allocs(this.exposed_allocs(), |this, alloc_id, info| {
+            if matches!(info.kind, AllocKind::Function) {
+                static DEDUP: AtomicBool = AtomicBool::new(false);
+                if !DEDUP.swap(true, std::sync::atomic::Ordering::Relaxed) {
+                    // Newly set, so first time we get here.
+                    this.emit_diagnostic(NonHaltingDiagnostic::NativeCallFnPtr);
+                }
+            }
             // If there is no data behind this pointer, skip this.
             if !matches!(info.kind, AllocKind::LiveData) {
                 return interp_ok(());