Auto merge of #113128 - WaffleLapkin:become_trully_unuwuable, r=oli-obk,RalfJung

Support tail calls in mir via `TerminatorKind::TailCall`

This is one of the interesting bits in tail call implementation — MIR support.

This adds a new `TerminatorKind` which represents a tail call:
```rust
    TailCall {
        func: Operand<'tcx>,
        args: Vec<Operand<'tcx>>,
        fn_span: Span,
    },
```

*Structurally* this is very similar to a normal `Call` but is missing a few fields:
- `destination` — tail calls don't write to destination, instead they pass caller's destination to the callee (such that eventual `return` will write to the caller of the function that used tail call)
- `target` — similarly to `destination` tail calls pass the caller's return address to the callee, so there is nothing to do
- `unwind` — I _think_ this is applicable too, although it's a bit confusing
- `call_source` — `become` forbids operators and is not created as a lowering of something else; tail calls always come from HIR (at least for now)

It might be helpful to read the interpreter implementation to understand what `TailCall` means exactly, although I've tried documenting it too.

-----

There are a few `FIXME`-questions still left, ideally we'd be able to answer them during review ':)

-----

r? `@oli-obk`
cc `@scottmcm` `@DrMeepster` `@JakobDegen`

compiler/rustc_const_eval/src/check_consts/check.rs

@@ -135,6 +135,8 @@ impl<'mir, 'tcx> Qualifs<'mir, 'tcx> {
         ccx: &'mir ConstCx<'mir, 'tcx>,
         tainted_by_errors: Option<ErrorGuaranteed>,
     ) -> ConstQualifs {
+        // FIXME(explicit_tail_calls): uhhhh I think we can return without return now, does it change anything
+
         // Find the `Return` terminator if one exists.
         //
         // If no `Return` terminator exists, this MIR is divergent. Just return the conservative
@@ -711,7 +713,14 @@ impl<'tcx> Visitor<'tcx> for Checker<'_, 'tcx> {
         self.super_terminator(terminator, location);
 
         match &terminator.kind {
-            TerminatorKind::Call { func, args, fn_span, call_source, .. } => {
+            TerminatorKind::Call { func, args, fn_span, .. }
+            | TerminatorKind::TailCall { func, args, fn_span, .. } => {
+                let call_source = match terminator.kind {
+                    TerminatorKind::Call { call_source, .. } => call_source,
+                    TerminatorKind::TailCall { .. } => CallSource::Normal,
+                    _ => unreachable!(),
+                };
+
                 let ConstCx { tcx, body, param_env, .. } = *self.ccx;
                 let caller = self.def_id();
 
@@ -783,7 +792,7 @@ impl<'tcx> Visitor<'tcx> for Checker<'_, 'tcx> {
                             callee,
                             args: fn_args,
                             span: *fn_span,
-                            call_source: *call_source,
+                            call_source,
                             feature: Some(if tcx.features().const_trait_impl {
                                 sym::effects
                             } else {
@@ -830,7 +839,7 @@ impl<'tcx> Visitor<'tcx> for Checker<'_, 'tcx> {
                         callee,
                         args: fn_args,
                         span: *fn_span,
-                        call_source: *call_source,
+                        call_source,
                         feature: None,
                     });
                     return;

compiler/rustc_const_eval/src/check_consts/post_drop_elaboration.rs

@@ -108,6 +108,7 @@ impl<'tcx> Visitor<'tcx> for CheckLiveDrops<'_, 'tcx> {
 
             mir::TerminatorKind::UnwindTerminate(_)
             | mir::TerminatorKind::Call { .. }
+            | mir::TerminatorKind::TailCall { .. }
             | mir::TerminatorKind::Assert { .. }
             | mir::TerminatorKind::FalseEdge { .. }
             | mir::TerminatorKind::FalseUnwind { .. }

compiler/rustc_const_eval/src/interpret/eval_context.rs

@@ -26,8 +26,8 @@ use rustc_target::abi::{call::FnAbi, Align, HasDataLayout, Size, TargetDataLayou
 use super::{
     err_inval, throw_inval, throw_ub, throw_ub_custom, throw_unsup, GlobalId, Immediate,
     InterpErrorInfo, InterpResult, MPlaceTy, Machine, MemPlace, MemPlaceMeta, Memory, MemoryKind,
-    OpTy, Operand, Place, PlaceTy, Pointer, PointerArithmetic, Projectable, Provenance, Scalar,
-    StackPopJump,
+    OpTy, Operand, Place, PlaceTy, Pointer, PointerArithmetic, Projectable, Provenance,
+    ReturnAction, Scalar,
 };
 use crate::errors;
 use crate::util;
@@ -159,6 +159,19 @@ pub enum StackPopCleanup {
     Root { cleanup: bool },
 }
 
+/// Return type of [`InterpCx::pop_stack_frame`].
+pub struct StackPopInfo<'tcx, Prov: Provenance> {
+    /// Additional information about the action to be performed when returning from the popped
+    /// stack frame.
+    pub return_action: ReturnAction,
+
+    /// [`return_to_block`](Frame::return_to_block) of the popped stack frame.
+    pub return_to_block: StackPopCleanup,
+
+    /// [`return_place`](Frame::return_place) of the popped stack frame.
+    pub return_place: MPlaceTy<'tcx, Prov>,
+}
+
 /// State of a local variable including a memoized layout
 #[derive(Clone)]
 pub struct LocalState<'tcx, Prov: Provenance = CtfeProvenance> {
@@ -803,14 +816,31 @@ impl<'tcx, M: Machine<'tcx>> InterpCx<'tcx, M> {
         return_to_block: StackPopCleanup,
     ) -> InterpResult<'tcx> {
         trace!("body: {:#?}", body);
+
+        // First push a stack frame so we have access to the local args
+        self.push_new_stack_frame(instance, body, return_to_block, return_place.clone())?;
+
+        self.after_stack_frame_push(instance, body)?;
+
+        Ok(())
+    }
+
+    /// Creates a new stack frame, initializes it and pushes it onto the stack.
+    /// A private helper for [`push_stack_frame`](InterpCx::push_stack_frame).
+    fn push_new_stack_frame(
+        &mut self,
+        instance: ty::Instance<'tcx>,
+        body: &'tcx mir::Body<'tcx>,
+        return_to_block: StackPopCleanup,
+        return_place: MPlaceTy<'tcx, M::Provenance>,
+    ) -> InterpResult<'tcx> {
         let dead_local = LocalState { value: LocalValue::Dead, layout: Cell::new(None) };
         let locals = IndexVec::from_elem(dead_local, &body.local_decls);
-        // First push a stack frame so we have access to the local args
         let pre_frame = Frame {
             body,
             loc: Right(body.span), // Span used for errors caused during preamble.
             return_to_block,
-            return_place: return_place.clone(),
+            return_place,
             locals,
             instance,
             tracing_span: SpanGuard::new(),
@@ -819,6 +849,15 @@ impl<'tcx, M: Machine<'tcx>> InterpCx<'tcx, M> {
         let frame = M::init_frame(self, pre_frame)?;
         self.stack_mut().push(frame);
 
+        Ok(())
+    }
+
+    /// A private helper for [`push_stack_frame`](InterpCx::push_stack_frame).
+    fn after_stack_frame_push(
+        &mut self,
+        instance: ty::Instance<'tcx>,
+        body: &'tcx mir::Body<'tcx>,
+    ) -> InterpResult<'tcx> {
         // Make sure all the constants required by this frame evaluate successfully (post-monomorphization check).
         for &const_ in &body.required_consts {
             let c =
@@ -839,6 +878,61 @@ impl<'tcx, M: Machine<'tcx>> InterpCx<'tcx, M> {
         Ok(())
     }
 
+    /// Pops a stack frame from the stack and returns some information about it.
+    ///
+    /// This also deallocates locals, if necessary.
+    ///
+    /// [`M::before_stack_pop`] should be called before calling this function.
+    /// [`M::after_stack_pop`] is called by this function automatically.
+    ///
+    /// [`M::before_stack_pop`]: Machine::before_stack_pop
+    /// [`M::after_stack_pop`]: Machine::after_stack_pop
+    pub fn pop_stack_frame(
+        &mut self,
+        unwinding: bool,
+    ) -> InterpResult<'tcx, StackPopInfo<'tcx, M::Provenance>> {
+        let cleanup = self.cleanup_current_frame_locals()?;
+
+        let frame =
+            self.stack_mut().pop().expect("tried to pop a stack frame, but there were none");
+
+        let return_to_block = frame.return_to_block;
+        let return_place = frame.return_place.clone();
+
+        let return_action;
+        if cleanup {
+            return_action = M::after_stack_pop(self, frame, unwinding)?;
+            assert_ne!(return_action, ReturnAction::NoCleanup);
+        } else {
+            return_action = ReturnAction::NoCleanup;
+        };
+
+        Ok(StackPopInfo { return_action, return_to_block, return_place })
+    }
+
+    /// A private helper for [`pop_stack_frame`](InterpCx::pop_stack_frame).
+    /// Returns `true` if cleanup has been done, `false` otherwise.
+    fn cleanup_current_frame_locals(&mut self) -> InterpResult<'tcx, bool> {
+        // Cleanup: deallocate locals.
+        // Usually we want to clean up (deallocate locals), but in a few rare cases we don't.
+        // We do this while the frame is still on the stack, so errors point to the callee.
+        let return_to_block = self.frame().return_to_block;
+        let cleanup = match return_to_block {
+            StackPopCleanup::Goto { .. } => true,
+            StackPopCleanup::Root { cleanup, .. } => cleanup,
+        };
+
+        if cleanup {
+            // We need to take the locals out, since we need to mutate while iterating.
+            let locals = mem::take(&mut self.frame_mut().locals);
+            for local in &locals {
+                self.deallocate_local(local.value)?;
+            }
+        }
+
+        Ok(cleanup)
+    }
+
     /// Jump to the given block.
     #[inline]
     pub fn go_to_block(&mut self, target: mir::BasicBlock) {
@@ -886,7 +980,7 @@ impl<'tcx, M: Machine<'tcx>> InterpCx<'tcx, M> {
     }
 
     /// Pops the current frame from the stack, deallocating the
-    /// memory for allocated locals.
+    /// memory for allocated locals, and jumps to an appropriate place.
     ///
     /// If `unwinding` is `false`, then we are performing a normal return
     /// from a function. In this case, we jump back into the frame of the caller,
@@ -899,7 +993,10 @@ impl<'tcx, M: Machine<'tcx>> InterpCx<'tcx, M> {
     /// The cleanup block ends with a special `Resume` terminator, which will
     /// cause us to continue unwinding.
     #[instrument(skip(self), level = "debug")]
-    pub(super) fn pop_stack_frame(&mut self, unwinding: bool) -> InterpResult<'tcx> {
+    pub(super) fn return_from_current_stack_frame(
+        &mut self,
+        unwinding: bool,
+    ) -> InterpResult<'tcx> {
         info!(
             "popping stack frame ({})",
             if unwinding { "during unwinding" } else { "returning from function" }
@@ -947,45 +1044,31 @@ impl<'tcx, M: Machine<'tcx>> InterpCx<'tcx, M> {
             Ok(())
         };
 
-        // Cleanup: deallocate locals.
-        // Usually we want to clean up (deallocate locals), but in a few rare cases we don't.
-        // We do this while the frame is still on the stack, so errors point to the callee.
-        let return_to_block = self.frame().return_to_block;
-        let cleanup = match return_to_block {
-            StackPopCleanup::Goto { .. } => true,
-            StackPopCleanup::Root { cleanup, .. } => cleanup,
-        };
-        if cleanup {
-            // We need to take the locals out, since we need to mutate while iterating.
-            let locals = mem::take(&mut self.frame_mut().locals);
-            for local in &locals {
-                self.deallocate_local(local.value)?;
-            }
-        }
-
         // All right, now it is time to actually pop the frame.
-        // Note that its locals are gone already, but that's fine.
-        let frame =
-            self.stack_mut().pop().expect("tried to pop a stack frame, but there were none");
+        let stack_pop_info = self.pop_stack_frame(unwinding)?;
+
         // Report error from return value copy, if any.
         copy_ret_result?;
 
-        // If we are not doing cleanup, also skip everything else.
-        if !cleanup {
-            assert!(self.stack().is_empty(), "only the topmost frame should ever be leaked");
-            assert!(!unwinding, "tried to skip cleanup during unwinding");
-            // Skip machine hook.
-            return Ok(());
-        }
-        if M::after_stack_pop(self, frame, unwinding)? == StackPopJump::NoJump {
-            // The hook already did everything.
-            return Ok(());
+        match stack_pop_info.return_action {
+            ReturnAction::Normal => {}
+            ReturnAction::NoJump => {
+                // The hook already did everything.
+                return Ok(());
+            }
+            ReturnAction::NoCleanup => {
+                // If we are not doing cleanup, also skip everything else.
+                assert!(self.stack().is_empty(), "only the topmost frame should ever be leaked");
+                assert!(!unwinding, "tried to skip cleanup during unwinding");
+                // Skip machine hook.
+                return Ok(());
+            }
         }
 
         // Normal return, figure out where to jump.
         if unwinding {
             // Follow the unwind edge.
-            let unwind = match return_to_block {
+            let unwind = match stack_pop_info.return_to_block {
                 StackPopCleanup::Goto { unwind, .. } => unwind,
                 StackPopCleanup::Root { .. } => {
                     panic!("encountered StackPopCleanup::Root when unwinding!")
@@ -995,7 +1078,7 @@ impl<'tcx, M: Machine<'tcx>> InterpCx<'tcx, M> {
             self.unwind_to_block(unwind)
         } else {
             // Follow the normal return edge.
-            match return_to_block {
+            match stack_pop_info.return_to_block {
                 StackPopCleanup::Goto { ret, .. } => self.return_to_block(ret),
                 StackPopCleanup::Root { .. } => {
                     assert!(

compiler/rustc_const_eval/src/interpret/machine.rs

@@ -23,10 +23,11 @@ use super::{
     MemoryKind, Misalignment, OpTy, PlaceTy, Pointer, Provenance,
 };
 
-/// Data returned by Machine::stack_pop,
-/// to provide further control over the popping of the stack frame
+/// Data returned by [`Machine::after_stack_pop`], and consumed by
+/// [`InterpCx::return_from_current_stack_frame`] to determine what actions should be done when
+/// returning from a stack frame.
 #[derive(Eq, PartialEq, Debug, Copy, Clone)]
-pub enum StackPopJump {
+pub enum ReturnAction {
     /// Indicates that no special handling should be
     /// done - we'll either return normally or unwind
     /// based on the terminator for the function
@@ -36,6 +37,9 @@ pub enum StackPopJump {
     /// Indicates that we should *not* jump to the return/unwind address, as the callback already
     /// took care of everything.
     NoJump,
+
+    /// Returned by [`InterpCx::pop_stack_frame`] when no cleanup should be done.
+    NoCleanup,
 }
 
 /// Whether this kind of memory is allowed to leak
@@ -522,10 +526,10 @@ pub trait Machine<'tcx>: Sized {
         _ecx: &mut InterpCx<'tcx, Self>,
         _frame: Frame<'tcx, Self::Provenance, Self::FrameExtra>,
         unwinding: bool,
-    ) -> InterpResult<'tcx, StackPopJump> {
+    ) -> InterpResult<'tcx, ReturnAction> {
         // By default, we do not support unwinding from panics
         assert!(!unwinding);
-        Ok(StackPopJump::Normal)
+        Ok(ReturnAction::Normal)
     }
 
     /// Called immediately after actual memory was allocated for a local

compiler/rustc_const_eval/src/interpret/mod.rs

@@ -26,7 +26,7 @@ pub use self::intern::{
     intern_const_alloc_for_constprop, intern_const_alloc_recursive, HasStaticRootDefId, InternKind,
     InternResult,
 };
-pub use self::machine::{compile_time_machine, AllocMap, Machine, MayLeak, StackPopJump};
+pub use self::machine::{compile_time_machine, AllocMap, Machine, MayLeak, ReturnAction};
 pub use self::memory::{AllocKind, AllocRef, AllocRefMut, FnVal, Memory, MemoryKind};
 pub use self::operand::{ImmTy, Immediate, OpTy, Readable};
 pub use self::place::{MPlaceTy, MemPlaceMeta, PlaceTy, Writeable};

compiler/rustc_const_eval/src/interpret/step.rs

@@ -32,7 +32,7 @@ impl<'tcx, M: Machine<'tcx>> InterpCx<'tcx, M> {
             // We are unwinding and this fn has no cleanup code.
             // Just go on unwinding.
             trace!("unwinding: skipping frame");
-            self.pop_stack_frame(/* unwinding */ true)?;
+            self.return_from_current_stack_frame(/* unwinding */ true)?;
             return Ok(true);
         };
         let basic_block = &self.body().basic_blocks[loc.block];

compiler/rustc_const_eval/src/interpret/terminator.rs

@@ -4,9 +4,8 @@ use either::Either;
 use rustc_middle::ty::TyCtxt;
 use tracing::trace;
 
-use rustc_middle::span_bug;
 use rustc_middle::{
-    mir,
+    bug, mir, span_bug,
     ty::{
         self,
         layout::{FnAbiOf, IntegerExt, LayoutOf, TyAndLayout},
@@ -26,7 +25,10 @@ use super::{
     InterpResult, MPlaceTy, Machine, OpTy, PlaceTy, Projectable, Provenance, Scalar,
     StackPopCleanup,
 };
-use crate::fluent_generated as fluent;
+use crate::{
+    fluent_generated as fluent,
+    interpret::{eval_context::StackPopInfo, ReturnAction},
+};
 
 /// An argment passed to a function.
 #[derive(Clone, Debug)]
@@ -47,6 +49,15 @@ impl<'tcx, Prov: Provenance> FnArg<'tcx, Prov> {
     }
 }
 
+struct EvaluatedCalleeAndArgs<'tcx, M: Machine<'tcx>> {
+    callee: FnVal<'tcx, M::ExtraFnVal>,
+    args: Vec<FnArg<'tcx, M::Provenance>>,
+    fn_sig: ty::FnSig<'tcx>,
+    fn_abi: &'tcx FnAbi<'tcx, Ty<'tcx>>,
+    /// True if the function is marked as `#[track_caller]` ([`ty::InstanceKind::requires_caller_location`])
+    with_caller_location: bool,
+}
+
 impl<'tcx, M: Machine<'tcx>> InterpCx<'tcx, M> {
     /// Make a copy of the given fn_arg. Any `InPlace` are degenerated to copies, no protection of the
     /// original memory occurs.
@@ -84,7 +95,7 @@ impl<'tcx, M: Machine<'tcx>> InterpCx<'tcx, M> {
         use rustc_middle::mir::TerminatorKind::*;
         match terminator.kind {
             Return => {
-                self.pop_stack_frame(/* unwinding */ false)?
+                self.return_from_current_stack_frame(/* unwinding */ false)?
             }
 
             Goto { target } => self.go_to_block(target),
@@ -124,40 +135,13 @@ impl<'tcx, M: Machine<'tcx>> InterpCx<'tcx, M> {
             } => {
                 let old_stack = self.frame_idx();
                 let old_loc = self.frame().loc;
-                let func = self.eval_operand(func, None)?;
-                let args = self.eval_fn_call_arguments(args)?;
 
-                let fn_sig_binder = func.layout.ty.fn_sig(*self.tcx);
-                let fn_sig =
-                    self.tcx.normalize_erasing_late_bound_regions(self.param_env, fn_sig_binder);
-                let extra_args = &args[fn_sig.inputs().len()..];
-                let extra_args =
-                    self.tcx.mk_type_list_from_iter(extra_args.iter().map(|arg| arg.layout().ty));
-
-                let (fn_val, fn_abi, with_caller_location) = match *func.layout.ty.kind() {
-                    ty::FnPtr(_sig) => {
-                        let fn_ptr = self.read_pointer(&func)?;
-                        let fn_val = self.get_ptr_fn(fn_ptr)?;
-                        (fn_val, self.fn_abi_of_fn_ptr(fn_sig_binder, extra_args)?, false)
-                    }
-                    ty::FnDef(def_id, args) => {
-                        let instance = self.resolve(def_id, args)?;
-                        (
-                            FnVal::Instance(instance),
-                            self.fn_abi_of_instance(instance, extra_args)?,
-                            instance.def.requires_caller_location(*self.tcx),
-                        )
-                    }
-                    _ => span_bug!(
-                        terminator.source_info.span,
-                        "invalid callee of type {}",
-                        func.layout.ty
-                    ),
-                };
+                let EvaluatedCalleeAndArgs { callee, args, fn_sig, fn_abi, with_caller_location } =
+                    self.eval_callee_and_args(terminator, func, args)?;
 
                 let destination = self.force_allocation(&self.eval_place(destination)?)?;
                 self.eval_fn_call(
-                    fn_val,
+                    callee,
                     (fn_sig.abi, fn_abi),
                     &args,
                     with_caller_location,
@@ -172,6 +156,22 @@ impl<'tcx, M: Machine<'tcx>> InterpCx<'tcx, M> {
                 }
             }
 
+            TailCall { ref func, ref args, fn_span: _ } => {
+                let old_frame_idx = self.frame_idx();
+
+                let EvaluatedCalleeAndArgs { callee, args, fn_sig, fn_abi, with_caller_location } =
+                    self.eval_callee_and_args(terminator, func, args)?;
+
+                self.eval_fn_tail_call(callee, (fn_sig.abi, fn_abi), &args, with_caller_location)?;
+
+                if self.frame_idx() != old_frame_idx {
+                    span_bug!(
+                        terminator.source_info.span,
+                        "evaluating this tail call pushed a new stack frame"
+                    );
+                }
+            }
+
             Drop { place, target, unwind, replace: _ } => {
                 let place = self.eval_place(place)?;
                 let instance = Instance::resolve_drop_in_place(*self.tcx, place.layout.ty);
@@ -209,7 +209,7 @@ impl<'tcx, M: Machine<'tcx>> InterpCx<'tcx, M> {
                 trace!("unwinding: resuming from cleanup");
                 // By definition, a Resume terminator means
                 // that we're unwinding
-                self.pop_stack_frame(/* unwinding */ true)?;
+                self.return_from_current_stack_frame(/* unwinding */ true)?;
                 return Ok(());
             }
 
@@ -514,6 +514,45 @@ impl<'tcx, M: Machine<'tcx>> InterpCx<'tcx, M> {
         Ok(())
     }
 
+    /// Shared part of `Call` and `TailCall` implementation — finding and evaluating all the
+    /// necessary information about callee and arguments to make a call.
+    fn eval_callee_and_args(
+        &self,
+        terminator: &mir::Terminator<'tcx>,
+        func: &mir::Operand<'tcx>,
+        args: &[Spanned<mir::Operand<'tcx>>],
+    ) -> InterpResult<'tcx, EvaluatedCalleeAndArgs<'tcx, M>> {
+        let func = self.eval_operand(func, None)?;
+        let args = self.eval_fn_call_arguments(args)?;
+
+        let fn_sig_binder = func.layout.ty.fn_sig(*self.tcx);
+        let fn_sig = self.tcx.normalize_erasing_late_bound_regions(self.param_env, fn_sig_binder);
+        let extra_args = &args[fn_sig.inputs().len()..];
+        let extra_args =
+            self.tcx.mk_type_list_from_iter(extra_args.iter().map(|arg| arg.layout().ty));
+
+        let (callee, fn_abi, with_caller_location) = match *func.layout.ty.kind() {
+            ty::FnPtr(_sig) => {
+                let fn_ptr = self.read_pointer(&func)?;
+                let fn_val = self.get_ptr_fn(fn_ptr)?;
+                (fn_val, self.fn_abi_of_fn_ptr(fn_sig_binder, extra_args)?, false)
+            }
+            ty::FnDef(def_id, args) => {
+                let instance = self.resolve(def_id, args)?;
+                (
+                    FnVal::Instance(instance),
+                    self.fn_abi_of_instance(instance, extra_args)?,
+                    instance.def.requires_caller_location(*self.tcx),
+                )
+            }
+            _ => {
+                span_bug!(terminator.source_info.span, "invalid callee of type {}", func.layout.ty)
+            }
+        };
+
+        Ok(EvaluatedCalleeAndArgs { callee, args, fn_sig, fn_abi, with_caller_location })
+    }
+
     /// Call this function -- pushing the stack frame and initializing the arguments.
     ///
     /// `caller_fn_abi` is used to determine if all the arguments are passed the proper way.
@@ -924,6 +963,49 @@ impl<'tcx, M: Machine<'tcx>> InterpCx<'tcx, M> {
         }
     }
 
+    pub(crate) fn eval_fn_tail_call(
+        &mut self,
+        fn_val: FnVal<'tcx, M::ExtraFnVal>,
+        (caller_abi, caller_fn_abi): (Abi, &FnAbi<'tcx, Ty<'tcx>>),
+        args: &[FnArg<'tcx, M::Provenance>],
+        with_caller_location: bool,
+    ) -> InterpResult<'tcx> {
+        trace!("eval_fn_call: {:#?}", fn_val);
+
+        // This is the "canonical" implementation of tails calls,
+        // a pop of the current stack frame, followed by a normal call
+        // which pushes a new stack frame, with the return address from
+        // the popped stack frame.
+        //
+        // Note that we are using `pop_stack_frame` and not `return_from_current_stack_frame`,
+        // as the latter "executes" the goto to the return block, but we don't want to,
+        // only the tail called function should return to the current return block.
+        M::before_stack_pop(self, self.frame())?;
+
+        let StackPopInfo { return_action, return_to_block, return_place } =
+            self.pop_stack_frame(false)?;
+
+        assert_eq!(return_action, ReturnAction::Normal);
+
+        let StackPopCleanup::Goto { ret, unwind } = return_to_block else {
+            bug!("can't tailcall as root");
+        };
+
+        // FIXME(explicit_tail_calls):
+        //   we should check if both caller&callee can/n't unwind,
+        //   see <https://github.com/rust-lang/rust/pull/113128#issuecomment-1614979803>
+
+        self.eval_fn_call(
+            fn_val,
+            (caller_abi, caller_fn_abi),
+            args,
+            with_caller_location,
+            &return_place,
+            ret,
+            unwind,
+        )
+    }
+
     fn check_fn_target_features(&self, instance: ty::Instance<'tcx>) -> InterpResult<'tcx, ()> {
         // Calling functions with `#[target_feature]` is not unsafe on WASM, see #84988
         let attrs = self.tcx.codegen_fn_attrs(instance.def_id());