[llvm] Support IFuncs on Darwin platforms (#73686)

... by lowering them as lazy resolve-on-first-use symbol resolvers. Note that this is subtly different timing than on ELF platforms, where ifunc resolution happens at load time.

Since ld64 and ld-prime don't support all the cases we need for these, we lower them manually in the AsmPrinter.

Author: jroelofs

llvm/docs/LangRef.rst

@@ -959,10 +959,11 @@ IFuncs
 -------
 
 IFuncs, like as aliases, don't create any new data or func. They are just a new
-symbol that dynamic linker resolves at runtime by calling a resolver function.
+symbol that is resolved at runtime by calling a resolver function.
 
-IFuncs have a name and a resolver that is a function called by dynamic linker
-that returns address of another function associated with the name.
+On ELF platforms, IFuncs are resolved by the dynamic linker at load time. On
+Mach-O platforms, they are lowered in terms of ``.symbol_resolver`` functions,
+which lazily resolve the callee the first time they are called.
 
 IFunc may have an optional :ref:`linkage type <linkage>` and an optional
 :ref:`visibility style <visibility>`.

llvm/include/llvm/CodeGen/AsmPrinter.h

@@ -599,6 +599,26 @@ class AsmPrinter : public MachineFunctionPass {
   /// instructions in verbose mode.
   virtual void emitImplicitDef(const MachineInstr *MI) const;
 
+  /// getSubtargetInfo() cannot be used where this is needed because we don't
+  /// have a MachineFunction when we're lowering a GlobalIFunc, and
+  /// getSubtargetInfo requires one. Override the implementation in targets
+  /// that support the Mach-O IFunc lowering.
+  virtual const MCSubtargetInfo *getIFuncMCSubtargetInfo() const {
+    return nullptr;
+  }
+
+  virtual void emitMachOIFuncStubBody(Module &M, const GlobalIFunc &GI,
+                                      MCSymbol *LazyPointer) {
+    llvm_unreachable(
+        "Mach-O IFunc lowering is not yet supported on this target");
+  }
+
+  virtual void emitMachOIFuncStubHelperBody(Module &M, const GlobalIFunc &GI,
+                                            MCSymbol *LazyPointer) {
+    llvm_unreachable(
+        "Mach-O IFunc lowering is not yet supported on this target");
+  }
+
   /// Emit N NOP instructions.
   void emitNops(unsigned N);
 
@@ -614,7 +634,7 @@ class AsmPrinter : public MachineFunctionPass {
                                          StringRef Suffix) const;
 
   /// Return the MCSymbol for the specified ExternalSymbol.
-  MCSymbol *GetExternalSymbolSymbol(StringRef Sym) const;
+  MCSymbol *GetExternalSymbolSymbol(Twine Sym) const;
 
   /// Return the symbol for the specified jump table entry.
   MCSymbol *GetJTISymbol(unsigned JTID, bool isLinkerPrivate = false) const;
@@ -884,6 +904,7 @@ class AsmPrinter : public MachineFunctionPass {
   void emitGlobalAlias(Module &M, const GlobalAlias &GA);
   void emitGlobalIFunc(Module &M, const GlobalIFunc &GI);
 
+private:
   /// This method decides whether the specified basic block requires a label.
   bool shouldEmitLabelForBasicBlock(const MachineBasicBlock &MBB) const;
 

llvm/lib/CodeGen/AsmPrinter/AsmPrinter.cpp

@@ -59,6 +59,7 @@
 #include "llvm/CodeGen/TargetLowering.h"
 #include "llvm/CodeGen/TargetOpcodes.h"
 #include "llvm/CodeGen/TargetRegisterInfo.h"
+#include "llvm/CodeGen/TargetSubtargetInfo.h"
 #include "llvm/Config/config.h"
 #include "llvm/IR/BasicBlock.h"
 #include "llvm/IR/Comdat.h"
@@ -2147,24 +2148,80 @@ void AsmPrinter::emitGlobalIFunc(Module &M, const GlobalIFunc &GI) {
   assert(!TM.getTargetTriple().isOSBinFormatXCOFF() &&
          "IFunc is not supported on AIX.");
 
-  MCSymbol *Name = getSymbol(&GI);
+  auto EmitLinkage = [&](MCSymbol *Sym) {
+    if (GI.hasExternalLinkage() || !MAI->getWeakRefDirective())
+      OutStreamer->emitSymbolAttribute(Sym, MCSA_Global);
+    else if (GI.hasWeakLinkage() || GI.hasLinkOnceLinkage())
+      OutStreamer->emitSymbolAttribute(Sym, MCSA_WeakReference);
+    else
+      assert(GI.hasLocalLinkage() && "Invalid ifunc linkage");
+  };
 
-  if (GI.hasExternalLinkage() || !MAI->getWeakRefDirective())
-    OutStreamer->emitSymbolAttribute(Name, MCSA_Global);
-  else if (GI.hasWeakLinkage() || GI.hasLinkOnceLinkage())
-    OutStreamer->emitSymbolAttribute(Name, MCSA_WeakReference);
-  else
-    assert(GI.hasLocalLinkage() && "Invalid ifunc linkage");
+  if (TM.getTargetTriple().isOSBinFormatELF()) {
+    MCSymbol *Name = getSymbol(&GI);
+    EmitLinkage(Name);
+    OutStreamer->emitSymbolAttribute(Name, MCSA_ELF_TypeIndFunction);
+    emitVisibility(Name, GI.getVisibility());
+
+    // Emit the directives as assignments aka .set:
+    const MCExpr *Expr = lowerConstant(GI.getResolver());
+    OutStreamer->emitAssignment(Name, Expr);
+    MCSymbol *LocalAlias = getSymbolPreferLocal(GI);
+    if (LocalAlias != Name)
+      OutStreamer->emitAssignment(LocalAlias, Expr);
+
+    return;
+  }
 
-  OutStreamer->emitSymbolAttribute(Name, MCSA_ELF_TypeIndFunction);
-  emitVisibility(Name, GI.getVisibility());
+  if (!TM.getTargetTriple().isOSBinFormatMachO() || !getIFuncMCSubtargetInfo())
+    llvm::report_fatal_error("IFuncs are not supported on this platform");
 
-  // Emit the directives as assignments aka .set:
-  const MCExpr *Expr = lowerConstant(GI.getResolver());
-  OutStreamer->emitAssignment(Name, Expr);
-  MCSymbol *LocalAlias = getSymbolPreferLocal(GI);
-  if (LocalAlias != Name)
-    OutStreamer->emitAssignment(LocalAlias, Expr);
+  // On Darwin platforms, emit a manually-constructed .symbol_resolver that
+  // implements the symbol resolution duties of the IFunc.
+  //
+  // Normally, this would be handled by linker magic, but unfortunately there
+  // are a few limitations in ld64 and ld-prime's implementation of
+  // .symbol_resolver that mean we can't always use them:
+  //
+  //    *  resolvers cannot be the target of an alias
+  //    *  resolvers cannot have private linkage
+  //    *  resolvers cannot have linkonce linkage
+  //    *  resolvers cannot appear in executables
+  //    *  resolvers cannot appear in bundles
+  //
+  // This works around that by emitting a close approximation of what the
+  // linker would have done.
+
+  MCSymbol *LazyPointer =
+      GetExternalSymbolSymbol(GI.getName() + ".lazy_pointer");
+  MCSymbol *StubHelper = GetExternalSymbolSymbol(GI.getName() + ".stub_helper");
+
+  OutStreamer->switchSection(OutContext.getObjectFileInfo()->getDataSection());
+
+  const DataLayout &DL = M.getDataLayout();
+  emitAlignment(Align(DL.getPointerSize()));
+  OutStreamer->emitLabel(LazyPointer);
+  emitVisibility(LazyPointer, GI.getVisibility());
+  OutStreamer->emitValue(MCSymbolRefExpr::create(StubHelper, OutContext), 8);
+
+  OutStreamer->switchSection(OutContext.getObjectFileInfo()->getTextSection());
+
+  const TargetSubtargetInfo *STI =
+      TM.getSubtargetImpl(*GI.getResolverFunction());
+  const TargetLowering *TLI = STI->getTargetLowering();
+  Align TextAlign(TLI->getMinFunctionAlignment());
+
+  MCSymbol *Stub = getSymbol(&GI);
+  EmitLinkage(Stub);
+  OutStreamer->emitCodeAlignment(TextAlign, getIFuncMCSubtargetInfo());
+  OutStreamer->emitLabel(Stub);
+  emitVisibility(Stub, GI.getVisibility());
+  emitMachOIFuncStubBody(M, GI, LazyPointer);
+
+  OutStreamer->emitCodeAlignment(TextAlign, getIFuncMCSubtargetInfo());
+  OutStreamer->emitLabel(StubHelper);
+  emitVisibility(StubHelper, GI.getVisibility());
+  emitMachOIFuncStubHelperBody(M, GI, LazyPointer);
 }
 
 void AsmPrinter::emitRemarksSection(remarks::RemarkStreamer &RS) {
@@ -2311,6 +2368,32 @@ bool AsmPrinter::doFinalization(Module &M) {
   // through user plugins.
   emitStackMaps();
 
+  // Print aliases in topological order, that is, for each alias a = b,
+  // b must be printed before a.
+  // This is because on some targets (e.g. PowerPC) linker expects aliases in
+  // such an order to generate correct TOC information.
+  SmallVector<const GlobalAlias *, 16> AliasStack;
+  SmallPtrSet<const GlobalAlias *, 16> AliasVisited;
+  for (const auto &Alias : M.aliases()) {
+    if (Alias.hasAvailableExternallyLinkage())
+      continue;
+    for (const GlobalAlias *Cur = &Alias; Cur;
+         Cur = dyn_cast<GlobalAlias>(Cur->getAliasee())) {
+      if (!AliasVisited.insert(Cur).second)
+        break;
+      AliasStack.push_back(Cur);
+    }
+    for (const GlobalAlias *AncestorAlias : llvm::reverse(AliasStack))
+      emitGlobalAlias(M, *AncestorAlias);
+    AliasStack.clear();
+  }
+
+  // IFuncs must come before deubginfo in case the backend decides to emit them
+  // as actual functions, since on Mach-O targets, we cannot create regular
+  // sections after DWARF.
+  for (const auto &IFunc : M.ifuncs())
+    emitGlobalIFunc(M, IFunc);
+
   // Finalize debug and EH information.
   for (const HandlerInfo &HI : Handlers) {
     NamedRegionTimer T(HI.TimerName, HI.TimerDescription, HI.TimerGroupName,
@@ -2350,28 +2433,6 @@ bool AsmPrinter::doFinalization(Module &M) {
     }
   }
 
-  // Print aliases in topological order, that is, for each alias a = b,
-  // b must be printed before a.
-  // This is because on some targets (e.g. PowerPC) linker expects aliases in
-  // such an order to generate correct TOC information.
-  SmallVector<const GlobalAlias *, 16> AliasStack;
-  SmallPtrSet<const GlobalAlias *, 16> AliasVisited;
-  for (const auto &Alias : M.aliases()) {
-    if (Alias.hasAvailableExternallyLinkage())
-      continue;
-    for (const GlobalAlias *Cur = &Alias; Cur;
-         Cur = dyn_cast<GlobalAlias>(Cur->getAliasee())) {
-      if (!AliasVisited.insert(Cur).second)
-        break;
-      AliasStack.push_back(Cur);
-    }
-    for (const GlobalAlias *AncestorAlias : llvm::reverse(AliasStack))
-      emitGlobalAlias(M, *AncestorAlias);
-    AliasStack.clear();
-  }
-  for (const auto &IFunc : M.ifuncs())
-    emitGlobalIFunc(M, IFunc);
-
   GCModuleInfo *MI = getAnalysisIfAvailable<GCModuleInfo>();
   assert(MI && "AsmPrinter didn't require GCModuleInfo?");
   for (GCModuleInfo::iterator I = MI->end(), E = MI->begin(); I != E; )
@@ -3745,7 +3806,7 @@ MCSymbol *AsmPrinter::getSymbolWithGlobalValueBase(const GlobalValue *GV,
 }
 
 /// Return the MCSymbol for the specified ExternalSymbol.
-MCSymbol *AsmPrinter::GetExternalSymbolSymbol(StringRef Sym) const {
+MCSymbol *AsmPrinter::GetExternalSymbolSymbol(Twine Sym) const {
   SmallString<60> NameStr;
   Mangler::getNameWithPrefix(NameStr, Sym, getDataLayout());
   return OutContext.getOrCreateSymbol(NameStr);

llvm/lib/IR/Verifier.cpp

@@ -2229,13 +2229,11 @@ void Verifier::verifyFunctionAttrs(FunctionType *FT, AttributeList Attrs,
   }
 
   // Check EVEX512 feature.
-  if (MaxParameterWidth >= 512 && Attrs.hasFnAttr("target-features")) {
-    Triple T(M.getTargetTriple());
-    if (T.isX86()) {
-      StringRef TF = Attrs.getFnAttr("target-features").getValueAsString();
-      Check(!TF.contains("+avx512f") || !TF.contains("-evex512"),
-            "512-bit vector arguments require 'evex512' for AVX512", V);
-    }
+  if (MaxParameterWidth >= 512 && Attrs.hasFnAttr("target-features") &&
+      TT.isX86()) {
+    StringRef TF = Attrs.getFnAttr("target-features").getValueAsString();
+    Check(!TF.contains("+avx512f") || !TF.contains("-evex512"),
+          "512-bit vector arguments require 'evex512' for AVX512", V);
   }
 
   checkUnsignedBaseTenFuncAttr(Attrs, "patchable-function-prefix", V);