Auto merge of #108089 - Zoxc:windows-tls, r=bjorn3

Support TLS access into dylibs on Windows

This allows access to `#[thread_local]`  in upstream dylibs on Windows by introducing a MIR shim to return the address of the thread local. Accesses that go into an upstream dylib will call the MIR shim to get the address of it.

`convert_tls_rvalues` is introduced in `rustc_codegen_ssa` which rewrites MIR TLS accesses to dummy calls which are replaced with calls to the MIR shims when the dummy calls are lowered to backend calls.

A new `dll_tls_export` target option enables this behavior with a `false` value which is set for Windows platforms.

This fixes https://github.com/rust-lang/rust/issues/84933.

compiler/rustc_codegen_cranelift/src/constant.rs

@@ -54,12 +54,22 @@ pub(crate) fn codegen_tls_ref<'tcx>(
     def_id: DefId,
     layout: TyAndLayout<'tcx>,
 ) -> CValue<'tcx> {
+    let tls_ptr = if !def_id.is_local() && fx.tcx.needs_thread_local_shim(def_id) {
+        let instance = ty::Instance {
+            def: ty::InstanceDef::ThreadLocalShim(def_id),
+            substs: ty::InternalSubsts::empty(),
+        };
+        let func_ref = fx.get_function_ref(instance);
+        let call = fx.bcx.ins().call(func_ref, &[]);
+        fx.bcx.func.dfg.first_result(call)
+    } else {
         let data_id = data_id_for_static(fx.tcx, fx.module, def_id, false);
         let local_data_id = fx.module.declare_data_in_func(data_id, &mut fx.bcx.func);
         if fx.clif_comments.enabled() {
             fx.add_comment(local_data_id, format!("tls {:?}", def_id));
         }
-    let tls_ptr = fx.bcx.ins().tls_value(fx.pointer_type, local_data_id);
+        fx.bcx.ins().tls_value(fx.pointer_type, local_data_id)
+    };
     CValue::by_val(tls_ptr, layout)
 }
 

compiler/rustc_codegen_ssa/src/back/symbol_export.rs

@@ -177,14 +177,29 @@ fn exported_symbols_provider_local(
 
     // FIXME: Sorting this is unnecessary since we are sorting later anyway.
     //        Can we skip the later sorting?
-    let mut symbols: Vec<_> = tcx.with_stable_hashing_context(|hcx| {
+    let sorted = tcx.with_stable_hashing_context(|hcx| {
-        tcx.reachable_non_generics(LOCAL_CRATE)
+        tcx.reachable_non_generics(LOCAL_CRATE).to_sorted(&hcx, true)
-            .to_sorted(&hcx, true)
-            .into_iter()
-            .map(|(&def_id, &info)| (ExportedSymbol::NonGeneric(def_id), info))
-            .collect()
     });
 
+    let mut symbols: Vec<_> =
+        sorted.iter().map(|(&def_id, &info)| (ExportedSymbol::NonGeneric(def_id), info)).collect();
+
+    // Export TLS shims
+    if !tcx.sess.target.dll_tls_export {
+        symbols.extend(sorted.iter().filter_map(|(&def_id, &info)| {
+            tcx.needs_thread_local_shim(def_id).then(|| {
+                (
+                    ExportedSymbol::ThreadLocalShim(def_id),
+                    SymbolExportInfo {
+                        level: info.level,
+                        kind: SymbolExportKind::Text,
+                        used: info.used,
+                    },
+                )
+            })
+        }))
+    }
+
     if tcx.entry_fn(()).is_some() {
         let exported_symbol =
             ExportedSymbol::NoDefId(SymbolName::new(tcx, tcx.sess.target.entry_name.as_ref()));
@@ -380,7 +395,9 @@ fn upstream_monomorphizations_provider(
                         continue;
                     }
                 }
-                ExportedSymbol::NonGeneric(..) | ExportedSymbol::NoDefId(..) => {
+                ExportedSymbol::NonGeneric(..)
+                | ExportedSymbol::ThreadLocalShim(..)
+                | ExportedSymbol::NoDefId(..) => {
                     // These are no monomorphizations
                     continue;
                 }
@@ -500,6 +517,16 @@ pub fn symbol_name_for_instance_in_crate<'tcx>(
                 instantiating_crate,
             )
         }
+        ExportedSymbol::ThreadLocalShim(def_id) => {
+            rustc_symbol_mangling::symbol_name_for_instance_in_crate(
+                tcx,
+                ty::Instance {
+                    def: ty::InstanceDef::ThreadLocalShim(def_id),
+                    substs: ty::InternalSubsts::empty(),
+                },
+                instantiating_crate,
+            )
+        }
         ExportedSymbol::DropGlue(ty) => rustc_symbol_mangling::symbol_name_for_instance_in_crate(
             tcx,
             Instance::resolve_drop_in_place(tcx, ty),
@@ -548,6 +575,8 @@ pub fn linking_symbol_name_for_instance_in_crate<'tcx>(
         ExportedSymbol::DropGlue(..) => None,
         // NoDefId always follow the target's default symbol decoration scheme.
         ExportedSymbol::NoDefId(..) => None,
+        // ThreadLocalShim always follow the target's default symbol decoration scheme.
+        ExportedSymbol::ThreadLocalShim(..) => None,
     };
 
     let (conv, args) = instance

compiler/rustc_codegen_ssa/src/mir/rvalue.rs

@@ -516,8 +516,20 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
 
             mir::Rvalue::ThreadLocalRef(def_id) => {
                 assert!(bx.cx().tcx().is_static(def_id));
-                let static_ = bx.get_static(def_id);
                 let layout = bx.layout_of(bx.cx().tcx().static_ptr_ty(def_id));
+                let static_ = if !def_id.is_local() && bx.cx().tcx().needs_thread_local_shim(def_id)
+                {
+                    let instance = ty::Instance {
+                        def: ty::InstanceDef::ThreadLocalShim(def_id),
+                        substs: ty::InternalSubsts::empty(),
+                    };
+                    let fn_ptr = bx.get_fn_addr(instance);
+                    let fn_abi = bx.fn_abi_of_instance(instance, ty::List::empty());
+                    let fn_ty = bx.fn_decl_backend_type(&fn_abi);
+                    bx.call(fn_ty, Some(fn_abi), fn_ptr, &[], None)
+                } else {
+                    bx.get_static(def_id)
+                };
                 OperandRef { val: OperandValue::Immediate(static_), layout }
             }
             mir::Rvalue::Use(ref operand) => self.codegen_operand(bx, operand),

compiler/rustc_const_eval/src/interpret/terminator.rs

@@ -383,6 +383,7 @@ impl<'mir, 'tcx: 'mir, M: Machine<'mir, 'tcx>> InterpCx<'mir, 'tcx, M> {
             | ty::InstanceDef::DropGlue(..)
             | ty::InstanceDef::CloneShim(..)
             | ty::InstanceDef::FnPtrAddrShim(..)
+            | ty::InstanceDef::ThreadLocalShim(..)
             | ty::InstanceDef::Item(_) => {
                 // We need MIR for this fn
                 let Some((body, instance)) =

compiler/rustc_middle/src/middle/exported_symbols.rs

@@ -43,6 +43,7 @@ pub enum ExportedSymbol<'tcx> {
     NonGeneric(DefId),
     Generic(DefId, SubstsRef<'tcx>),
     DropGlue(Ty<'tcx>),
+    ThreadLocalShim(DefId),
     NoDefId(ty::SymbolName<'tcx>),
 }
 
@@ -58,6 +59,10 @@ impl<'tcx> ExportedSymbol<'tcx> {
             ExportedSymbol::DropGlue(ty) => {
                 tcx.symbol_name(ty::Instance::resolve_drop_in_place(tcx, ty))
             }
+            ExportedSymbol::ThreadLocalShim(def_id) => tcx.symbol_name(ty::Instance {
+                def: ty::InstanceDef::ThreadLocalShim(def_id),
+                substs: ty::InternalSubsts::empty(),
+            }),
             ExportedSymbol::NoDefId(symbol_name) => symbol_name,
         }
     }

compiler/rustc_middle/src/mir/mono.rs

@@ -382,6 +382,7 @@ impl<'tcx> CodegenUnit<'tcx> {
                             | InstanceDef::ClosureOnceShim { .. }
                             | InstanceDef::DropGlue(..)
                             | InstanceDef::CloneShim(..)
+                            | InstanceDef::ThreadLocalShim(..)
                             | InstanceDef::FnPtrAddrShim(..) => None,
                         }
                     }

compiler/rustc_middle/src/mir/tcx.rs

@@ -164,17 +164,7 @@ impl<'tcx> Rvalue<'tcx> {
             Rvalue::Repeat(ref operand, count) => {
                 tcx.mk_array_with_const_len(operand.ty(local_decls, tcx), count)
             }
-            Rvalue::ThreadLocalRef(did) => {
+            Rvalue::ThreadLocalRef(did) => tcx.thread_local_ptr_ty(did),
-                let static_ty = tcx.type_of(did).subst_identity();
-                if tcx.is_mutable_static(did) {
-                    tcx.mk_mut_ptr(static_ty)
-                } else if tcx.is_foreign_item(did) {
-                    tcx.mk_imm_ptr(static_ty)
-                } else {
-                    // FIXME: These things don't *really* have 'static lifetime.
-                    tcx.mk_imm_ref(tcx.lifetimes.re_static, static_ty)
-                }
-            }
             Rvalue::Ref(reg, bk, ref place) => {
                 let place_ty = place.ty(local_decls, tcx).ty;
                 tcx.mk_ref(reg, ty::TypeAndMut { ty: place_ty, mutbl: bk.to_mutbl_lossy() })

compiler/rustc_middle/src/mir/visit.rs

@@ -335,6 +335,7 @@ macro_rules! make_mir_visitor {
                         ty::InstanceDef::VTableShim(_def_id) |
                         ty::InstanceDef::ReifyShim(_def_id) |
                         ty::InstanceDef::Virtual(_def_id, _) |
+                        ty::InstanceDef::ThreadLocalShim(_def_id) |
                         ty::InstanceDef::ClosureOnceShim { call_once: _def_id, track_caller: _ } |
                         ty::InstanceDef::DropGlue(_def_id, None) => {}
 

compiler/rustc_middle/src/ty/instance.rs

@@ -82,6 +82,11 @@ pub enum InstanceDef<'tcx> {
     /// The `DefId` is the ID of the `call_once` method in `FnOnce`.
     ClosureOnceShim { call_once: DefId, track_caller: bool },
 
+    /// Compiler-generated accessor for thread locals which returns a reference to the thread local
+    /// the `DefId` defines. This is used to export thread locals from dylibs on platforms lacking
+    /// native support.
+    ThreadLocalShim(DefId),
+
     /// `core::ptr::drop_in_place::<T>`.
     ///
     /// The `DefId` is for `core::ptr::drop_in_place`.
@@ -156,6 +161,7 @@ impl<'tcx> InstanceDef<'tcx> {
             | InstanceDef::FnPtrShim(def_id, _)
             | InstanceDef::Virtual(def_id, _)
             | InstanceDef::Intrinsic(def_id)
+            | InstanceDef::ThreadLocalShim(def_id)
             | InstanceDef::ClosureOnceShim { call_once: def_id, track_caller: _ }
             | InstanceDef::DropGlue(def_id, _)
             | InstanceDef::CloneShim(def_id, _)
@@ -167,7 +173,9 @@ impl<'tcx> InstanceDef<'tcx> {
     pub fn def_id_if_not_guaranteed_local_codegen(self) -> Option<DefId> {
         match self {
             ty::InstanceDef::Item(def) => Some(def.did),
-            ty::InstanceDef::DropGlue(def_id, Some(_)) => Some(def_id),
+            ty::InstanceDef::DropGlue(def_id, Some(_)) | InstanceDef::ThreadLocalShim(def_id) => {
+                Some(def_id)
+            }
             InstanceDef::VTableShim(..)
             | InstanceDef::ReifyShim(..)
             | InstanceDef::FnPtrShim(..)
@@ -192,6 +200,7 @@ impl<'tcx> InstanceDef<'tcx> {
             | InstanceDef::ClosureOnceShim { call_once: def_id, track_caller: _ }
             | InstanceDef::DropGlue(def_id, _)
             | InstanceDef::CloneShim(def_id, _)
+            | InstanceDef::ThreadLocalShim(def_id)
             | InstanceDef::FnPtrAddrShim(def_id, _) => ty::WithOptConstParam::unknown(def_id),
         }
     }
@@ -215,6 +224,7 @@ impl<'tcx> InstanceDef<'tcx> {
         let def_id = match *self {
             ty::InstanceDef::Item(def) => def.did,
             ty::InstanceDef::DropGlue(_, Some(_)) => return false,
+            ty::InstanceDef::ThreadLocalShim(_) => return false,
             _ => return true,
         };
         matches!(
@@ -255,6 +265,9 @@ impl<'tcx> InstanceDef<'tcx> {
                 )
             });
         }
+        if let ty::InstanceDef::ThreadLocalShim(..) = *self {
+            return false;
+        }
         tcx.codegen_fn_attrs(self.def_id()).requests_inline()
     }
 
@@ -278,6 +291,7 @@ impl<'tcx> InstanceDef<'tcx> {
     pub fn has_polymorphic_mir_body(&self) -> bool {
         match *self {
             InstanceDef::CloneShim(..)
+            | InstanceDef::ThreadLocalShim(..)
             | InstanceDef::FnPtrAddrShim(..)
             | InstanceDef::FnPtrShim(..)
             | InstanceDef::DropGlue(_, Some(_)) => false,
@@ -310,6 +324,7 @@ fn fmt_instance(
         InstanceDef::Item(_) => Ok(()),
         InstanceDef::VTableShim(_) => write!(f, " - shim(vtable)"),
         InstanceDef::ReifyShim(_) => write!(f, " - shim(reify)"),
+        InstanceDef::ThreadLocalShim(_) => write!(f, " - shim(tls)"),
         InstanceDef::Intrinsic(_) => write!(f, " - intrinsic"),
         InstanceDef::Virtual(_, num) => write!(f, " - virtual#{}", num),
         InstanceDef::FnPtrShim(_, ty) => write!(f, " - shim({})", ty),

compiler/rustc_middle/src/ty/mod.rs

@@ -2386,6 +2386,7 @@ impl<'tcx> TyCtxt<'tcx> {
             | ty::InstanceDef::ClosureOnceShim { .. }
             | ty::InstanceDef::DropGlue(..)
             | ty::InstanceDef::CloneShim(..)
+            | ty::InstanceDef::ThreadLocalShim(..)
             | ty::InstanceDef::FnPtrAddrShim(..) => self.mir_shims(instance),
         }
     }

compiler/rustc_middle/src/ty/util.rs

@@ -597,6 +597,28 @@ impl<'tcx> TyCtxt<'tcx> {
         self.static_mutability(def_id) == Some(hir::Mutability::Mut)
     }
 
+    /// Returns `true` if the item pointed to by `def_id` is a thread local which needs a
+    /// thread local shim generated.
+    #[inline]
+    pub fn needs_thread_local_shim(self, def_id: DefId) -> bool {
+        !self.sess.target.dll_tls_export
+            && self.is_thread_local_static(def_id)
+            && !self.is_foreign_item(def_id)
+    }
+
+    /// Returns the type a reference to the thread local takes in MIR.
+    pub fn thread_local_ptr_ty(self, def_id: DefId) -> Ty<'tcx> {
+        let static_ty = self.type_of(def_id).subst_identity();
+        if self.is_mutable_static(def_id) {
+            self.mk_mut_ptr(static_ty)
+        } else if self.is_foreign_item(def_id) {
+            self.mk_imm_ptr(static_ty)
+        } else {
+            // FIXME: These things don't *really* have 'static lifetime.
+            self.mk_imm_ref(self.lifetimes.re_static, static_ty)
+        }
+    }
+
     /// Get the type of the pointer to the static that we use in MIR.
     pub fn static_ptr_ty(self, def_id: DefId) -> Ty<'tcx> {
         // Make sure that any constants in the static's type are evaluated.

compiler/rustc_mir_transform/src/inline.rs

@@ -271,6 +271,7 @@ impl<'tcx> Inliner<'tcx> {
             | InstanceDef::ClosureOnceShim { .. }
             | InstanceDef::DropGlue(..)
             | InstanceDef::CloneShim(..)
+            | InstanceDef::ThreadLocalShim(..)
             | InstanceDef::FnPtrAddrShim(..) => return Ok(()),
         }
 

compiler/rustc_mir_transform/src/inline/cycle.rs

@@ -83,6 +83,7 @@ pub(crate) fn mir_callgraph_reachable<'tcx>(
                 | InstanceDef::ReifyShim(_)
                 | InstanceDef::FnPtrShim(..)
                 | InstanceDef::ClosureOnceShim { .. }
+                | InstanceDef::ThreadLocalShim { .. }
                 | InstanceDef::CloneShim(..) => {}
 
                 // This shim does not call any other functions, thus there can be no recursion.

compiler/rustc_mir_transform/src/shim.rs

@@ -76,6 +76,7 @@ fn make_shim<'tcx>(tcx: TyCtxt<'tcx>, instance: ty::InstanceDef<'tcx>) -> Body<'
 
             build_drop_shim(tcx, def_id, ty)
         }
+        ty::InstanceDef::ThreadLocalShim(..) => build_thread_local_shim(tcx, instance),
         ty::InstanceDef::CloneShim(def_id, ty) => build_clone_shim(tcx, def_id, ty),
         ty::InstanceDef::FnPtrAddrShim(def_id, ty) => build_fn_ptr_addr_shim(tcx, def_id, ty),
         ty::InstanceDef::Virtual(..) => {
@@ -322,6 +323,34 @@ impl<'a, 'tcx> DropElaborator<'a, 'tcx> for DropShimElaborator<'a, 'tcx> {
     }
 }
 
+fn build_thread_local_shim<'tcx>(tcx: TyCtxt<'tcx>, instance: ty::InstanceDef<'tcx>) -> Body<'tcx> {
+    let def_id = instance.def_id();
+
+    let span = tcx.def_span(def_id);
+    let source_info = SourceInfo::outermost(span);
+
+    let mut blocks = IndexVec::with_capacity(1);
+    blocks.push(BasicBlockData {
+        statements: vec![Statement {
+            source_info,
+            kind: StatementKind::Assign(Box::new((
+                Place::return_place(),
+                Rvalue::ThreadLocalRef(def_id),
+            ))),
+        }],
+        terminator: Some(Terminator { source_info, kind: TerminatorKind::Return }),
+        is_cleanup: false,
+    });
+
+    new_body(
+        MirSource::from_instance(instance),
+        blocks,
+        IndexVec::from_raw(vec![LocalDecl::new(tcx.thread_local_ptr_ty(def_id), span)]),
+        0,
+        span,
+    )
+}
+
 /// Builds a `Clone::clone` shim for `self_ty`. Here, `def_id` is `Clone::clone`.
 fn build_clone_shim<'tcx>(tcx: TyCtxt<'tcx>, def_id: DefId, self_ty: Ty<'tcx>) -> Body<'tcx> {
     debug!("build_clone_shim(def_id={:?})", def_id);

compiler/rustc_monomorphize/src/collector.rs

@@ -190,7 +190,8 @@ use rustc_middle::ty::print::with_no_trimmed_paths;
 use rustc_middle::ty::query::TyCtxtAt;
 use rustc_middle::ty::subst::{GenericArgKind, InternalSubsts};
 use rustc_middle::ty::{
-    self, GenericParamDefKind, Instance, Ty, TyCtxt, TypeFoldable, TypeVisitableExt, VtblEntry,
+    self, GenericParamDefKind, Instance, InstanceDef, Ty, TyCtxt, TypeFoldable, TypeVisitableExt,
+    VtblEntry,
 };
 use rustc_middle::{middle::codegen_fn_attrs::CodegenFnAttrFlags, mir::visit::TyContext};
 use rustc_session::config::EntryFnType;
@@ -462,6 +463,16 @@ fn collect_items_rec<'tcx>(
                     collect_miri(tcx, id, &mut neighbors);
                 }
             }
+
+            if tcx.needs_thread_local_shim(def_id) {
+                neighbors.push(respan(
+                    starting_point.span,
+                    MonoItem::Fn(Instance {
+                        def: InstanceDef::ThreadLocalShim(def_id),
+                        substs: InternalSubsts::empty(),
+                    }),
+                ));
+            }
         }
         MonoItem::Fn(instance) => {
             // Sanity check whether this ended up being collected accidentally
@@ -962,6 +973,9 @@ fn visit_instance_use<'tcx>(
                 bug!("{:?} being reified", instance);
             }
         }
+        ty::InstanceDef::ThreadLocalShim(..) => {
+            bug!("{:?} being reified", instance);
+        }
         ty::InstanceDef::DropGlue(_, None) => {
             // Don't need to emit noop drop glue if we are calling directly.
             if !is_direct_call {
@@ -1210,11 +1224,9 @@ impl<'v> RootCollector<'_, 'v> {
                 self.output.push(dummy_spanned(MonoItem::GlobalAsm(id)));
             }
             DefKind::Static(..) => {
-                debug!(
+                let def_id = id.owner_id.to_def_id();
-                    "RootCollector: ItemKind::Static({})",
+                debug!("RootCollector: ItemKind::Static({})", self.tcx.def_path_str(def_id));
-                    self.tcx.def_path_str(id.owner_id.to_def_id())
+                self.output.push(dummy_spanned(MonoItem::Static(def_id)));
-                );
-                self.output.push(dummy_spanned(MonoItem::Static(id.owner_id.to_def_id())));
             }
             DefKind::Const => {
                 // const items only generate mono items if they are

compiler/rustc_monomorphize/src/partitioning/default.rs

@@ -279,6 +279,7 @@ fn characteristic_def_id_of_mono_item<'tcx>(
                 | ty::InstanceDef::DropGlue(..)
                 | ty::InstanceDef::Virtual(..)
                 | ty::InstanceDef::CloneShim(..)
+                | ty::InstanceDef::ThreadLocalShim(..)
                 | ty::InstanceDef::FnPtrAddrShim(..) => return None,
             };
 
@@ -392,6 +393,19 @@ fn mono_item_linkage_and_visibility<'tcx>(
 
 type CguNameCache = FxHashMap<(DefId, bool), Symbol>;
 
+fn static_visibility<'tcx>(
+    tcx: TyCtxt<'tcx>,
+    can_be_internalized: &mut bool,
+    def_id: DefId,
+) -> Visibility {
+    if tcx.is_reachable_non_generic(def_id) {
+        *can_be_internalized = false;
+        default_visibility(tcx, def_id, false)
+    } else {
+        Visibility::Hidden
+    }
+}
+
 fn mono_item_visibility<'tcx>(
     tcx: TyCtxt<'tcx>,
     mono_item: &MonoItem<'tcx>,
@@ -403,21 +417,9 @@ fn mono_item_visibility<'tcx>(
         MonoItem::Fn(instance) => instance,
 
         // Misc handling for generics and such, but otherwise:
-        MonoItem::Static(def_id) => {
+        MonoItem::Static(def_id) => return static_visibility(tcx, can_be_internalized, *def_id),
-            return if tcx.is_reachable_non_generic(*def_id) {
-                *can_be_internalized = false;
-                default_visibility(tcx, *def_id, false)
-            } else {
-                Visibility::Hidden
-            };
-        }
         MonoItem::GlobalAsm(item_id) => {
-            return if tcx.is_reachable_non_generic(item_id.owner_id) {
+            return static_visibility(tcx, can_be_internalized, item_id.owner_id.to_def_id());
-                *can_be_internalized = false;
-                default_visibility(tcx, item_id.owner_id.to_def_id(), false)
-            } else {
-                Visibility::Hidden
-            };
         }
     };
 
@@ -425,6 +427,11 @@ fn mono_item_visibility<'tcx>(
         InstanceDef::Item(def) => def.did,
         InstanceDef::DropGlue(def_id, Some(_)) => def_id,
 
+        // We match the visiblity of statics here
+        InstanceDef::ThreadLocalShim(def_id) => {
+            return static_visibility(tcx, can_be_internalized, def_id);
+        }
+
         // These are all compiler glue and such, never exported, always hidden.
         InstanceDef::VTableShim(..)
         | InstanceDef::ReifyShim(..)

compiler/rustc_symbol_mangling/src/legacy.rs

@@ -65,6 +65,10 @@ pub(super) fn mangle<'tcx>(
         )
         .unwrap();
 
+    if let ty::InstanceDef::ThreadLocalShim(..) = instance.def {
+        let _ = printer.write_str("{{tls-shim}}");
+    }
+
     if let ty::InstanceDef::VTableShim(..) = instance.def {
         let _ = printer.write_str("{{vtable-shim}}");
     }

compiler/rustc_symbol_mangling/src/v0.rs

@@ -42,6 +42,7 @@ pub(super) fn mangle<'tcx>(
 
     // Append `::{shim:...#0}` to shims that can coexist with a non-shim instance.
     let shim_kind = match instance.def {
+        ty::InstanceDef::ThreadLocalShim(_) => Some("tls"),
         ty::InstanceDef::VTableShim(_) => Some("vtable"),
         ty::InstanceDef::ReifyShim(_) => Some("reify"),
 

compiler/rustc_target/src/spec/mod.rs

@@ -1471,6 +1471,8 @@ pub struct TargetOptions {
     pub features: StaticCow<str>,
     /// Whether dynamic linking is available on this target. Defaults to false.
     pub dynamic_linking: bool,
+    /// Whether dynamic linking can export TLS globals. Defaults to true.
+    pub dll_tls_export: bool,
     /// If dynamic linking is available, whether only cdylibs are supported.
     pub only_cdylib: bool,
     /// Whether executables are available on this target. Defaults to true.
@@ -1865,6 +1867,7 @@ impl Default for TargetOptions {
             cpu: "generic".into(),
             features: "".into(),
             dynamic_linking: false,
+            dll_tls_export: true,
             only_cdylib: false,
             executables: true,
             relocation_model: RelocModel::Pic,
@@ -2537,6 +2540,7 @@ impl Target {
         key!(cpu);
         key!(features);
         key!(dynamic_linking, bool);
+        key!(dll_tls_export, bool);
         key!(only_cdylib, bool);
         key!(executables, bool);
         key!(relocation_model, RelocModel)?;
@@ -2791,6 +2795,7 @@ impl ToJson for Target {
         target_option_val!(cpu);
         target_option_val!(features);
         target_option_val!(dynamic_linking);
+        target_option_val!(dll_tls_export);
         target_option_val!(only_cdylib);
         target_option_val!(executables);
         target_option_val!(relocation_model);

compiler/rustc_target/src/spec/msvc_base.rs

@@ -8,6 +8,7 @@ pub fn opts() -> TargetOptions {
 
     TargetOptions {
         linker_flavor: LinkerFlavor::Msvc(Lld::No),
+        dll_tls_export: false,
         is_like_windows: true,
         is_like_msvc: true,
         pre_link_args,

compiler/rustc_target/src/spec/windows_gnu_base.rs

@@ -78,6 +78,7 @@ pub fn opts() -> TargetOptions {
         function_sections: false,
         linker: Some("gcc".into()),
         dynamic_linking: true,
+        dll_tls_export: false,
         dll_prefix: "".into(),
         dll_suffix: ".dll".into(),
         exe_suffix: ".exe".into(),

compiler/rustc_target/src/spec/windows_gnullvm_base.rs

@@ -23,6 +23,7 @@ pub fn opts() -> TargetOptions {
         abi: "llvm".into(),
         linker: Some("clang".into()),
         dynamic_linking: true,
+        dll_tls_export: false,
         dll_prefix: "".into(),
         dll_suffix: ".dll".into(),
         exe_suffix: ".exe".into(),

compiler/rustc_ty_utils/src/abi.rs

@@ -3,7 +3,7 @@ use rustc_hir::lang_items::LangItem;
 use rustc_middle::ty::layout::{
     fn_can_unwind, FnAbiError, HasParamEnv, HasTyCtxt, LayoutCx, LayoutOf, TyAndLayout,
 };
-use rustc_middle::ty::{self, Ty, TyCtxt};
+use rustc_middle::ty::{self, InstanceDef, Ty, TyCtxt};
 use rustc_session::config::OptLevel;
 use rustc_span::def_id::DefId;
 use rustc_target::abi::call::{
@@ -29,6 +29,16 @@ fn fn_sig_for_fn_abi<'tcx>(
     instance: ty::Instance<'tcx>,
     param_env: ty::ParamEnv<'tcx>,
 ) -> ty::PolyFnSig<'tcx> {
+    if let InstanceDef::ThreadLocalShim(..) = instance.def {
+        return ty::Binder::dummy(tcx.mk_fn_sig(
+            [],
+            tcx.thread_local_ptr_ty(instance.def_id()),
+            false,
+            hir::Unsafety::Normal,
+            rustc_target::spec::abi::Abi::Unadjusted,
+        ));
+    }
+
     let ty = instance.ty(tcx, param_env);
     match *ty.kind() {
         ty::FnDef(..) => {

library/std/src/sys/common/thread_local/fast_local.rs

@@ -10,7 +10,7 @@
 macro_rules! __thread_local_inner {
     // used to generate the `LocalKey` value for const-initialized thread locals
     (@key $t:ty, const $init:expr) => {{
-        #[cfg_attr(not(windows), inline)] // see comments below
+        #[cfg_attr(not(bootstrap), inline)]
         #[deny(unsafe_op_in_unsafe_fn)]
         unsafe fn __getit(
             _init: $crate::option::Option<&mut $crate::option::Option<$t>>,
@@ -77,29 +77,7 @@ macro_rules! __thread_local_inner {
             #[inline]
             fn __init() -> $t { $init }
 
-            // When reading this function you might ask "why is this inlined
+            #[cfg_attr(not(bootstrap), inline)]
-            // everywhere other than Windows?", and that's a very reasonable
-            // question to ask. The short story is that it segfaults rustc if
-            // this function is inlined. The longer story is that Windows looks
-            // to not support `extern` references to thread locals across DLL
-            // boundaries. This appears to at least not be supported in the ABI
-            // that LLVM implements.
-            //
-            // Because of this we never inline on Windows, but we do inline on
-            // other platforms (where external references to thread locals
-            // across DLLs are supported). A better fix for this would be to
-            // inline this function on Windows, but only for "statically linked"
-            // components. For example if two separately compiled rlibs end up
-            // getting linked into a DLL then it's fine to inline this function
-            // across that boundary. It's only not fine to inline this function
-            // across a DLL boundary. Unfortunately rustc doesn't currently
-            // have this sort of logic available in an attribute, and it's not
-            // clear that rustc is even equipped to answer this (it's more of a
-            // Cargo question kinda). This means that, unfortunately, Windows
-            // gets the pessimistic path for now where it's never inlined.
-            //
-            // The issue of "should enable on Windows sometimes" is #84933
-            #[cfg_attr(not(windows), inline)]
             unsafe fn __getit(
                 init: $crate::option::Option<&mut $crate::option::Option<$t>>,
             ) -> $crate::option::Option<&'static $t> {

library/std/src/sys/common/thread_local/os_local.rs

@@ -10,7 +10,7 @@
 macro_rules! __thread_local_inner {
     // used to generate the `LocalKey` value for const-initialized thread locals
     (@key $t:ty, const $init:expr) => {{
-        #[cfg_attr(not(windows), inline)] // see comments below
+        #[cfg_attr(not(bootstrap), inline)]
         #[deny(unsafe_op_in_unsafe_fn)]
         unsafe fn __getit(
             _init: $crate::option::Option<&mut $crate::option::Option<$t>>,
@@ -49,29 +49,7 @@ macro_rules! __thread_local_inner {
             #[inline]
             fn __init() -> $t { $init }
 
-            // When reading this function you might ask "why is this inlined
+            #[cfg_attr(not(bootstrap), inline)]
-            // everywhere other than Windows?", and that's a very reasonable
-            // question to ask. The short story is that it segfaults rustc if
-            // this function is inlined. The longer story is that Windows looks
-            // to not support `extern` references to thread locals across DLL
-            // boundaries. This appears to at least not be supported in the ABI
-            // that LLVM implements.
-            //
-            // Because of this we never inline on Windows, but we do inline on
-            // other platforms (where external references to thread locals
-            // across DLLs are supported). A better fix for this would be to
-            // inline this function on Windows, but only for "statically linked"
-            // components. For example if two separately compiled rlibs end up
-            // getting linked into a DLL then it's fine to inline this function
-            // across that boundary. It's only not fine to inline this function
-            // across a DLL boundary. Unfortunately rustc doesn't currently
-            // have this sort of logic available in an attribute, and it's not
-            // clear that rustc is even equipped to answer this (it's more of a
-            // Cargo question kinda). This means that, unfortunately, Windows
-            // gets the pessimistic path for now where it's never inlined.
-            //
-            // The issue of "should enable on Windows sometimes" is #84933
-            #[cfg_attr(not(windows), inline)]
             unsafe fn __getit(
                 init: $crate::option::Option<&mut $crate::option::Option<$t>>,
             ) -> $crate::option::Option<&'static $t> {

tests/ui/thread-local/auxiliary/tls-export.rs

@@ -0,0 +1,17 @@
+#![crate_type = "dylib"]
+#![feature(thread_local)]
+#![feature(cfg_target_thread_local)]
+
+extern crate tls_rlib;
+
+pub use tls_rlib::*;
+
+#[cfg(target_thread_local)]
+#[thread_local]
+pub static FOO: bool = true;
+
+#[cfg(target_thread_local)]
+#[inline(never)]
+pub fn foo_addr() -> usize {
+    &FOO as *const bool as usize
+}

tests/ui/thread-local/auxiliary/tls-rlib.rs

@@ -0,0 +1,15 @@
+// no-prefer-dynamic
+
+#![crate_type = "rlib"]
+#![feature(thread_local)]
+#![feature(cfg_target_thread_local)]
+
+#[cfg(target_thread_local)]
+#[thread_local]
+pub static BAR: bool = true;
+
+#[cfg(target_thread_local)]
+#[inline(never)]
+pub fn bar_addr() -> usize {
+    &BAR as *const bool as usize
+}

tests/ui/thread-local/tls-dylib-access.rs

@@ -0,0 +1,19 @@
+// aux-build: tls-rlib.rs
+// aux-build: tls-export.rs
+// run-pass
+
+#![feature(cfg_target_thread_local)]
+
+#[cfg(target_thread_local)]
+extern crate tls_export;
+
+fn main() {
+    #[cfg(target_thread_local)]
+    {
+        // Check that we get the real address of the `FOO` TLS in the dylib
+        assert_eq!(&tls_export::FOO as *const bool as usize, tls_export::foo_addr());
+
+        // Check that we get the real address of the `BAR` TLS in the rlib linked into the dylib
+        assert_eq!(&tls_export::BAR as *const bool as usize, tls_export::bar_addr());
+    }
+}