Auto merge of #66874 - RalfJung:miri-assert-panic, r=oli-obk

Miri engine: proper support for `Assert` MIR terminators

This puts down the basis for rust-lang/miri#1070, and I also did some clean-up. The Miri side of this is at rust-lang/miri#1084.

r? @oli-obk

Author: bors

src/librustc/mir/interpret/value.rs

@@ -471,6 +471,13 @@ impl<Tag> From<Scalar<Tag>> for ScalarMaybeUndef<Tag> {
     }
 }
 
+impl<Tag> From<Pointer<Tag>> for ScalarMaybeUndef<Tag> {
+    #[inline(always)]
+    fn from(s: Pointer<Tag>) -> Self {
+        ScalarMaybeUndef::Scalar(s.into())
+    }
+}
+
 impl<Tag: fmt::Debug, Id: fmt::Debug> fmt::Debug for ScalarMaybeUndef<Tag, Id> {
     fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
         match self {

src/librustc_mir/const_eval.rs

@@ -23,7 +23,7 @@ use syntax::{source_map::{Span, DUMMY_SP}, symbol::Symbol};
 use crate::interpret::{self,
     PlaceTy, MPlaceTy, OpTy, ImmTy, Immediate, Scalar, Pointer,
     RawConst, ConstValue, Machine,
-    InterpResult, InterpErrorInfo, GlobalId, InterpCx, StackPopCleanup,
+    InterpResult, InterpErrorInfo, GlobalId, InterpCx, StackPopCleanup, AssertMessage,
     Allocation, AllocId, MemoryKind, Memory,
     snapshot, RefTracking, intern_const_alloc_recursive,
 };
@@ -395,6 +395,40 @@ impl<'mir, 'tcx> interpret::Machine<'mir, 'tcx> for CompileTimeInterpreter<'mir,
         )
     }
 
+    fn assert_panic(
+        ecx: &mut InterpCx<'mir, 'tcx, Self>,
+        _span: Span,
+        msg: &AssertMessage<'tcx>,
+        _unwind: Option<mir::BasicBlock>,
+    ) -> InterpResult<'tcx> {
+        use rustc::mir::interpret::PanicInfo::*;
+        Err(match msg {
+            BoundsCheck { ref len, ref index } => {
+                let len = ecx
+                    .read_immediate(ecx.eval_operand(len, None)?)
+                    .expect("can't eval len")
+                    .to_scalar()?
+                    .to_machine_usize(&*ecx)?;
+                let index = ecx
+                    .read_immediate(ecx.eval_operand(index, None)?)
+                    .expect("can't eval index")
+                    .to_scalar()?
+                    .to_machine_usize(&*ecx)?;
+                err_panic!(BoundsCheck { len, index })
+            }
+            Overflow(op) => err_panic!(Overflow(*op)),
+            OverflowNeg => err_panic!(OverflowNeg),
+            DivisionByZero => err_panic!(DivisionByZero),
+            RemainderByZero => err_panic!(RemainderByZero),
+            ResumedAfterReturn(generator_kind)
+                => err_panic!(ResumedAfterReturn(*generator_kind)),
+            ResumedAfterPanic(generator_kind)
+                => err_panic!(ResumedAfterPanic(*generator_kind)),
+            Panic { .. } => bug!("`Panic` variant cannot occur in MIR"),
+        }
+        .into())
+    }
+
     fn ptr_to_int(
         _mem: &Memory<'mir, 'tcx, Self>,
         _ptr: Pointer,
@@ -423,7 +457,7 @@ impl<'mir, 'tcx> interpret::Machine<'mir, 'tcx> for CompileTimeInterpreter<'mir,
     }
 
     #[inline(always)]
-    fn tag_allocation<'b>(
+    fn init_allocation_extra<'b>(
         _memory_extra: &(),
         _id: AllocId,
         alloc: Cow<'b, Allocation>,
@@ -518,7 +552,7 @@ pub fn const_caller_location<'tcx>(
         tcx.type_of(tcx.require_lang_item(PanicLocationLangItem, None))
             .subst(tcx, tcx.mk_substs([tcx.lifetimes.re_static.into()].iter())),
     );
-    let loc_place = ecx.alloc_caller_location(file, line, col).unwrap();
+    let loc_place = ecx.alloc_caller_location(file, line, col);
     intern_const_alloc_recursive(&mut ecx, None, loc_place).unwrap();
     let loc_const = ty::Const {
         ty: loc_ty,

src/librustc_mir/interpret/cast.rs

@@ -55,7 +55,7 @@ impl<'mir, 'tcx, M: Machine<'mir, 'tcx>> InterpCx<'mir, 'tcx, M> {
                         ).ok_or_else(|| err_inval!(TooGeneric))?;
 
                         let fn_ptr = self.memory.create_fn_alloc(FnVal::Instance(instance));
-                        self.write_scalar(Scalar::Ptr(fn_ptr.into()), dest)?;
+                        self.write_scalar(fn_ptr, dest)?;
                     }
                     _ => bug!("reify fn pointer on {:?}", src.layout.ty),
                 }
@@ -88,8 +88,7 @@ impl<'mir, 'tcx, M: Machine<'mir, 'tcx>> InterpCx<'mir, 'tcx, M> {
                             ty::ClosureKind::FnOnce,
                         );
                         let fn_ptr = self.memory.create_fn_alloc(FnVal::Instance(instance));
-                        let val = Immediate::Scalar(Scalar::Ptr(fn_ptr.into()).into());
-                        self.write_immediate(val, dest)?;
+                        self.write_scalar(fn_ptr, dest)?;
                     }
                     _ => bug!("closure fn pointer on {:?}", src.layout.ty),
                 }

src/librustc_mir/interpret/eval_context.rs

@@ -164,6 +164,20 @@ impl<'tcx, Tag: Copy + 'static> LocalState<'tcx, Tag> {
     }
 }
 
+impl<'mir, 'tcx, Tag, Extra> Frame<'mir, 'tcx, Tag, Extra> {
+    /// Return the `SourceInfo` of the current instruction.
+    pub fn current_source_info(&self) -> Option<mir::SourceInfo> {
+        self.block.map(|block| {
+            let block = &self.body.basic_blocks()[block];
+            if self.stmt < block.statements.len() {
+                block.statements[self.stmt].source_info
+            } else {
+                block.terminator().source_info
+            }
+        })
+    }
+}
+
 impl<'mir, 'tcx, M: Machine<'mir, 'tcx>> HasDataLayout for InterpCx<'mir, 'tcx, M> {
     #[inline]
     fn data_layout(&self) -> &layout::TargetDataLayout {
@@ -236,6 +250,12 @@ impl<'mir, 'tcx, M: Machine<'mir, 'tcx>> InterpCx<'mir, 'tcx, M> {
         self.memory.force_bits(scalar, size)
     }
 
+    /// Call this to turn untagged "global" pointers (obtained via `tcx`) into
+    /// the *canonical* machine pointer to the allocation.  Must never be used
+    /// for any other pointers!
+    ///
+    /// This represents a *direct* access to that memory, as opposed to access
+    /// through a pointer that was created by the program.
     #[inline(always)]
     pub fn tag_static_base_pointer(&self, ptr: Pointer) -> Pointer<M::PointerTag> {
         self.memory.tag_static_base_pointer(ptr)
@@ -828,34 +848,28 @@ impl<'mir, 'tcx, M: Machine<'mir, 'tcx>> InterpCx<'mir, 'tcx, M> {
     pub fn generate_stacktrace(&self, explicit_span: Option<Span>) -> Vec<FrameInfo<'tcx>> {
         let mut last_span = None;
         let mut frames = Vec::new();
-        for &Frame { instance, span, body, block, stmt, .. } in self.stack().iter().rev() {
+        for frame in self.stack().iter().rev() {
             // make sure we don't emit frames that are duplicates of the previous
-            if explicit_span == Some(span) {
-                last_span = Some(span);
+            if explicit_span == Some(frame.span) {
+                last_span = Some(frame.span);
                 continue;
             }
             if let Some(last) = last_span {
-                if last == span {
+                if last == frame.span {
                     continue;
                 }
             } else {
-                last_span = Some(span);
+                last_span = Some(frame.span);
             }
 
-            let lint_root = block.and_then(|block| {
-                let block = &body.basic_blocks()[block];
-                let source_info = if stmt < block.statements.len() {
-                    block.statements[stmt].source_info
-                } else {
-                    block.terminator().source_info
-                };
-                match &body.source_scopes[source_info.scope].local_data {
+            let lint_root = frame.current_source_info().and_then(|source_info| {
+                match &frame.body.source_scopes[source_info.scope].local_data {
                     mir::ClearCrossCrate::Set(data) => Some(data.lint_root),
                     mir::ClearCrossCrate::Clear => None,
                 }
             });
 
-            frames.push(FrameInfo { call_site: span, instance, lint_root });
+            frames.push(FrameInfo { call_site: frame.span, instance: frame.instance, lint_root });
         }
         trace!("generate stacktrace: {:#?}, {:?}", frames, explicit_span);
         frames

src/librustc_mir/interpret/intrinsics.rs

@@ -110,13 +110,7 @@ impl<'mir, 'tcx, M: Machine<'mir, 'tcx>> InterpCx<'mir, 'tcx, M> {
 
         match intrinsic_name {
             "caller_location" => {
-                let topmost = span.ctxt().outer_expn().expansion_cause().unwrap_or(span);
-                let caller = self.tcx.sess.source_map().lookup_char_pos(topmost.lo());
-                let location = self.alloc_caller_location(
-                    Symbol::intern(&caller.file.name.to_string()),
-                    caller.line as u32,
-                    caller.col_display as u32 + 1,
-                )?;
+                let location = self.alloc_caller_location_for_span(span);
                 self.write_scalar(location.ptr, dest)?;
             }
 

src/librustc_mir/interpret/intrinsics/caller_location.rs

@@ -1,50 +1,48 @@
 use rustc::middle::lang_items::PanicLocationLangItem;
-use rustc::mir::interpret::{Pointer, PointerArithmetic, Scalar};
 use rustc::ty::subst::Subst;
-use rustc_target::abi::{LayoutOf, Size};
-use syntax_pos::Symbol;
+use rustc_target::abi::LayoutOf;
+use syntax_pos::{Symbol, Span};
 
-use crate::interpret::{MemoryKind, MPlaceTy, intrinsics::{InterpCx, InterpResult, Machine}};
+use crate::interpret::{Scalar, MemoryKind, MPlaceTy, intrinsics::{InterpCx, Machine}};
 
 impl<'mir, 'tcx, M: Machine<'mir, 'tcx>> InterpCx<'mir, 'tcx, M> {
-    pub fn alloc_caller_location(
+    crate fn alloc_caller_location(
         &mut self,
         filename: Symbol,
         line: u32,
         col: u32,
-    ) -> InterpResult<'tcx, MPlaceTy<'tcx, M::PointerTag>> {
+    ) -> MPlaceTy<'tcx, M::PointerTag> {
+        let file = self.allocate_str(&filename.as_str(), MemoryKind::CallerLocation);
         let line = Scalar::from_u32(line);
         let col = Scalar::from_u32(col);
 
-        let ptr_size = self.pointer_size();
-        let u32_size = Size::from_bits(32);
-
+        // Allocate memory for `CallerLocation` struct.
         let loc_ty = self.tcx.type_of(self.tcx.require_lang_item(PanicLocationLangItem, None))
             .subst(*self.tcx, self.tcx.mk_substs([self.tcx.lifetimes.re_static.into()].iter()));
-        let loc_layout = self.layout_of(loc_ty)?;
-
-        let file_alloc = self.tcx.allocate_bytes(filename.as_str().as_bytes());
-        let file_ptr = Pointer::new(file_alloc, Size::ZERO);
-        let file = Scalar::Ptr(self.tag_static_base_pointer(file_ptr));
-        let file_len = Scalar::from_uint(filename.as_str().len() as u128, ptr_size);
-
+        let loc_layout = self.layout_of(loc_ty).unwrap();
         let location = self.allocate(loc_layout, MemoryKind::CallerLocation);
 
-        let file_out = self.mplace_field(location, 0)?;
-        let file_ptr_out = self.force_ptr(self.mplace_field(file_out, 0)?.ptr)?;
-        let file_len_out = self.force_ptr(self.mplace_field(file_out, 1)?.ptr)?;
-        let line_out = self.force_ptr(self.mplace_field(location, 1)?.ptr)?;
-        let col_out = self.force_ptr(self.mplace_field(location, 2)?.ptr)?;
+        // Initialize fields.
+        self.write_immediate(file.to_ref(), self.mplace_field(location, 0).unwrap().into())
+            .expect("writing to memory we just allocated cannot fail");
+        self.write_scalar(line, self.mplace_field(location, 1).unwrap().into())
+            .expect("writing to memory we just allocated cannot fail");
+        self.write_scalar(col, self.mplace_field(location, 2).unwrap().into())
+            .expect("writing to memory we just allocated cannot fail");
 
-        let layout = &self.tcx.data_layout;
-        // We just allocated this, so we can skip the bounds checks.
-        let alloc = self.memory.get_raw_mut(file_ptr_out.alloc_id)?;
-
-        alloc.write_scalar(layout, file_ptr_out, file.into(), ptr_size)?;
-        alloc.write_scalar(layout, file_len_out, file_len.into(), ptr_size)?;
-        alloc.write_scalar(layout, line_out, line.into(), u32_size)?;
-        alloc.write_scalar(layout, col_out, col.into(), u32_size)?;
+        location
+    }
 
-        Ok(location)
+    pub fn alloc_caller_location_for_span(
+        &mut self,
+        span: Span,
+    ) -> MPlaceTy<'tcx, M::PointerTag> {
+        let topmost = span.ctxt().outer_expn().expansion_cause().unwrap_or(span);
+        let caller = self.tcx.sess.source_map().lookup_char_pos(topmost.lo());
+        self.alloc_caller_location(
+            Symbol::intern(&caller.file.name.to_string()),
+            caller.line as u32,
+            caller.col_display as u32 + 1,
+        )
     }
 }

src/librustc_mir/interpret/machine.rs

@@ -11,7 +11,7 @@ use rustc::ty::{self, Ty, TyCtxt};
 use syntax_pos::Span;
 
 use super::{
-    Allocation, AllocId, InterpResult, Scalar, AllocationExtra,
+    Allocation, AllocId, InterpResult, Scalar, AllocationExtra, AssertMessage,
     InterpCx, PlaceTy, OpTy, ImmTy, MemoryKind, Pointer, Memory,
     Frame, Operand,
 };
@@ -175,6 +175,14 @@ pub trait Machine<'mir, 'tcx>: Sized {
         unwind: Option<mir::BasicBlock>,
     ) -> InterpResult<'tcx>;
 
+    /// Called to evaluate `Assert` MIR terminators that trigger a panic.
+    fn assert_panic(
+        ecx: &mut InterpCx<'mir, 'tcx, Self>,
+        span: Span,
+        msg: &AssertMessage<'tcx>,
+        unwind: Option<mir::BasicBlock>,
+    ) -> InterpResult<'tcx>;
+
     /// Called for read access to a foreign static item.
     ///
     /// This will only be called once per static and machine; the result is cached in
@@ -233,20 +241,19 @@ pub trait Machine<'mir, 'tcx>: Sized {
     /// cache the result. (This relies on `AllocMap::get_or` being able to add the
     /// owned allocation to the map even when the map is shared.)
     ///
-    /// For static allocations, the tag returned must be the same as the one returned by
-    /// `tag_static_base_pointer`.
-    fn tag_allocation<'b>(
+    /// Also return the "base" tag to use for this allocation: the one that is used for direct
+    /// accesses to this allocation. If `kind == STATIC_KIND`, this tag must be consistent
+    /// with `tag_static_base_pointer`.
+    fn init_allocation_extra<'b>(
         memory_extra: &Self::MemoryExtra,
         id: AllocId,
         alloc: Cow<'b, Allocation>,
         kind: Option<MemoryKind<Self::MemoryKinds>>,
     ) -> (Cow<'b, Allocation<Self::PointerTag, Self::AllocExtra>>, Self::PointerTag);
 
-    /// Return the "base" tag for the given static allocation: the one that is used for direct
-    /// accesses to this static/const/fn allocation.
-    ///
-    /// Be aware that requesting the `Allocation` for that `id` will lead to cycles
-    /// for cyclic statics!
+    /// Return the "base" tag for the given *static* allocation: the one that is used for direct
+    /// accesses to this static/const/fn allocation. If `id` is not a static allocation,
+    /// this will return an unusable tag (i.e., accesses will be UB)!
     fn tag_static_base_pointer(
         memory_extra: &Self::MemoryExtra,
         id: AllocId,

src/librustc_mir/interpret/memory.rs

@@ -143,6 +143,12 @@ impl<'mir, 'tcx, M: Machine<'mir, 'tcx>> Memory<'mir, 'tcx, M> {
         }
     }
 
+    /// Call this to turn untagged "global" pointers (obtained via `tcx`) into
+    /// the *canonical* machine pointer to the allocation.  Must never be used
+    /// for any other pointers!
+    ///
+    /// This represents a *direct* access to that memory, as opposed to access
+    /// through a pointer that was created by the program.
     #[inline]
     pub fn tag_static_base_pointer(&self, ptr: Pointer) -> Pointer<M::PointerTag> {
         ptr.with_tag(M::tag_static_base_pointer(&self.extra, ptr.alloc_id))
@@ -191,7 +197,9 @@ impl<'mir, 'tcx, M: Machine<'mir, 'tcx>> Memory<'mir, 'tcx, M> {
         kind: MemoryKind<M::MemoryKinds>,
     ) -> Pointer<M::PointerTag> {
         let id = self.tcx.alloc_map.lock().reserve();
-        let (alloc, tag) = M::tag_allocation(&self.extra, id, Cow::Owned(alloc), Some(kind));
+        debug_assert_ne!(Some(kind), M::STATIC_KIND.map(MemoryKind::Machine),
+            "dynamically allocating static memory");
+        let (alloc, tag) = M::init_allocation_extra(&self.extra, id, Cow::Owned(alloc), Some(kind));
         self.alloc_map.insert(id, (kind, alloc.into_owned()));
         Pointer::from(id).with_tag(tag)
     }
@@ -350,7 +358,7 @@ impl<'mir, 'tcx, M: Machine<'mir, 'tcx>> Memory<'mir, 'tcx, M> {
             sptr
         } else {
             // A "real" access, we must get a pointer.
-            Scalar::Ptr(self.force_ptr(sptr)?)
+            Scalar::from(self.force_ptr(sptr)?)
         };
         Ok(match normalized.to_bits_or_ptr(self.pointer_size(), self) {
             Ok(bits) => {
@@ -473,14 +481,15 @@ impl<'mir, 'tcx, M: Machine<'mir, 'tcx>> Memory<'mir, 'tcx, M> {
                 }
             }
         };
-        // We got tcx memory. Let the machine figure out whether and how to
-        // turn that into memory with the right pointer tag.
-        Ok(M::tag_allocation(
+        // We got tcx memory. Let the machine initialize its "extra" stuff.
+        let (alloc, tag) = M::init_allocation_extra(
             memory_extra,
             id, // always use the ID we got as input, not the "hidden" one.
             alloc,
             M::STATIC_KIND.map(MemoryKind::Machine),
-        ).0)
+        );
+        debug_assert_eq!(tag, M::tag_static_base_pointer(memory_extra, id));
+        Ok(alloc)
     }
 
     /// Gives raw access to the `Allocation`, without bounds or alignment checks.