//===- DebugInfo.cpp - Debug Information Helper Classes -------------------===// // // The LLVM Compiler Infrastructure // // This file is distributed under the University of Illinois Open Source // License. See LICENSE.TXT for details. // //===----------------------------------------------------------------------===// // // This file implements the helper classes used to build and interpret debug // information in LLVM IR form. // //===----------------------------------------------------------------------===// #include "llvm-c/DebugInfo.h" #include "llvm/ADT/DenseMap.h" #include "llvm/ADT/DenseSet.h" #include "llvm/ADT/None.h" #include "llvm/ADT/STLExtras.h" #include "llvm/ADT/SmallPtrSet.h" #include "llvm/ADT/SmallVector.h" #include "llvm/ADT/StringRef.h" #include "llvm/IR/BasicBlock.h" #include "llvm/IR/Constants.h" #include "llvm/IR/DebugInfoMetadata.h" #include "llvm/IR/DebugLoc.h" #include "llvm/IR/DebugInfo.h" #include "llvm/IR/DIBuilder.h" #include "llvm/IR/Function.h" #include "llvm/IR/GVMaterializer.h" #include "llvm/IR/Instruction.h" #include "llvm/IR/IntrinsicInst.h" #include "llvm/IR/LLVMContext.h" #include "llvm/IR/Metadata.h" #include "llvm/IR/Module.h" #include "llvm/Support/Casting.h" #include #include #include using namespace llvm; using namespace llvm::dwarf; DISubprogram *llvm::getDISubprogram(const MDNode *Scope) { if (auto *LocalScope = dyn_cast_or_null(Scope)) return LocalScope->getSubprogram(); return nullptr; } //===----------------------------------------------------------------------===// // DebugInfoFinder implementations. //===----------------------------------------------------------------------===// void DebugInfoFinder::reset() { CUs.clear(); SPs.clear(); GVs.clear(); TYs.clear(); Scopes.clear(); NodesSeen.clear(); } void DebugInfoFinder::processModule(const Module &M) { for (auto *CU : M.debug_compile_units()) { addCompileUnit(CU); for (auto DIG : CU->getGlobalVariables()) { if (!addGlobalVariable(DIG)) continue; auto *GV = DIG->getVariable(); processScope(GV->getScope()); processType(GV->getType().resolve()); } for (auto *ET : CU->getEnumTypes()) processType(ET); for (auto *RT : CU->getRetainedTypes()) if (auto *T = dyn_cast(RT)) processType(T); else processSubprogram(cast(RT)); for (auto *Import : CU->getImportedEntities()) { auto *Entity = Import->getEntity().resolve(); if (auto *T = dyn_cast(Entity)) processType(T); else if (auto *SP = dyn_cast(Entity)) processSubprogram(SP); else if (auto *NS = dyn_cast(Entity)) processScope(NS->getScope()); else if (auto *M = dyn_cast(Entity)) processScope(M->getScope()); } } for (auto &F : M.functions()) { if (auto *SP = cast_or_null(F.getSubprogram())) processSubprogram(SP); // There could be subprograms from inlined functions referenced from // instructions only. Walk the function to find them. for (const BasicBlock &BB : F) { for (const Instruction &I : BB) { if (!I.getDebugLoc()) continue; processLocation(M, I.getDebugLoc().get()); } } } } void DebugInfoFinder::processLocation(const Module &M, const DILocation *Loc) { if (!Loc) return; processScope(Loc->getScope()); processLocation(M, Loc->getInlinedAt()); } void DebugInfoFinder::processType(DIType *DT) { if (!addType(DT)) return; processScope(DT->getScope().resolve()); if (auto *ST = dyn_cast(DT)) { for (DITypeRef Ref : ST->getTypeArray()) processType(Ref.resolve()); return; } if (auto *DCT = dyn_cast(DT)) { processType(DCT->getBaseType().resolve()); for (Metadata *D : DCT->getElements()) { if (auto *T = dyn_cast(D)) processType(T); else if (auto *SP = dyn_cast(D)) processSubprogram(SP); } return; } if (auto *DDT = dyn_cast(DT)) { processType(DDT->getBaseType().resolve()); } } void DebugInfoFinder::processScope(DIScope *Scope) { if (!Scope) return; if (auto *Ty = dyn_cast(Scope)) { processType(Ty); return; } if (auto *CU = dyn_cast(Scope)) { addCompileUnit(CU); return; } if (auto *SP = dyn_cast(Scope)) { processSubprogram(SP); return; } if (!addScope(Scope)) return; if (auto *LB = dyn_cast(Scope)) { processScope(LB->getScope()); } else if (auto *NS = dyn_cast(Scope)) { processScope(NS->getScope()); } else if (auto *M = dyn_cast(Scope)) { processScope(M->getScope()); } } void DebugInfoFinder::processSubprogram(DISubprogram *SP) { if (!addSubprogram(SP)) return; processScope(SP->getScope().resolve()); processType(SP->getType()); for (auto *Element : SP->getTemplateParams()) { if (auto *TType = dyn_cast(Element)) { processType(TType->getType().resolve()); } else if (auto *TVal = dyn_cast(Element)) { processType(TVal->getType().resolve()); } } } void DebugInfoFinder::processDeclare(const Module &M, const DbgDeclareInst *DDI) { auto *N = dyn_cast(DDI->getVariable()); if (!N) return; auto *DV = dyn_cast(N); if (!DV) return; if (!NodesSeen.insert(DV).second) return; processScope(DV->getScope()); processType(DV->getType().resolve()); } void DebugInfoFinder::processValue(const Module &M, const DbgValueInst *DVI) { auto *N = dyn_cast(DVI->getVariable()); if (!N) return; auto *DV = dyn_cast(N); if (!DV) return; if (!NodesSeen.insert(DV).second) return; processScope(DV->getScope()); processType(DV->getType().resolve()); } bool DebugInfoFinder::addType(DIType *DT) { if (!DT) return false; if (!NodesSeen.insert(DT).second) return false; TYs.push_back(const_cast(DT)); return true; } bool DebugInfoFinder::addCompileUnit(DICompileUnit *CU) { if (!CU) return false; if (!NodesSeen.insert(CU).second) return false; CUs.push_back(CU); return true; } bool DebugInfoFinder::addGlobalVariable(DIGlobalVariableExpression *DIG) { if (!NodesSeen.insert(DIG).second) return false; GVs.push_back(DIG); return true; } bool DebugInfoFinder::addSubprogram(DISubprogram *SP) { if (!SP) return false; if (!NodesSeen.insert(SP).second) return false; SPs.push_back(SP); return true; } bool DebugInfoFinder::addScope(DIScope *Scope) { if (!Scope) return false; // FIXME: Ocaml binding generates a scope with no content, we treat it // as null for now. if (Scope->getNumOperands() == 0) return false; if (!NodesSeen.insert(Scope).second) return false; Scopes.push_back(Scope); return true; } static MDNode *stripDebugLocFromLoopID(MDNode *N) { assert(N->op_begin() != N->op_end() && "Missing self reference?"); // if there is no debug location, we do not have to rewrite this MDNode. if (std::none_of(N->op_begin() + 1, N->op_end(), [](const MDOperand &Op) { return isa(Op.get()); })) return N; // If there is only the debug location without any actual loop metadata, we // can remove the metadata. if (std::none_of(N->op_begin() + 1, N->op_end(), [](const MDOperand &Op) { return !isa(Op.get()); })) return nullptr; SmallVector Args; // Reserve operand 0 for loop id self reference. auto TempNode = MDNode::getTemporary(N->getContext(), None); Args.push_back(TempNode.get()); // Add all non-debug location operands back. for (auto Op = N->op_begin() + 1; Op != N->op_end(); Op++) { if (!isa(*Op)) Args.push_back(*Op); } // Set the first operand to itself. MDNode *LoopID = MDNode::get(N->getContext(), Args); LoopID->replaceOperandWith(0, LoopID); return LoopID; } bool llvm::stripDebugInfo(Function &F) { bool Changed = false; if (F.getMetadata(LLVMContext::MD_dbg)) { Changed = true; F.setSubprogram(nullptr); } DenseMap LoopIDsMap; for (BasicBlock &BB : F) { for (auto II = BB.begin(), End = BB.end(); II != End;) { Instruction &I = *II++; // We may delete the instruction, increment now. if (isa(&I)) { I.eraseFromParent(); Changed = true; continue; } if (I.getDebugLoc()) { Changed = true; I.setDebugLoc(DebugLoc()); } } auto *TermInst = BB.getTerminator(); if (!TermInst) // This is invalid IR, but we may not have run the verifier yet continue; if (auto *LoopID = TermInst->getMetadata(LLVMContext::MD_loop)) { auto *NewLoopID = LoopIDsMap.lookup(LoopID); if (!NewLoopID) NewLoopID = LoopIDsMap[LoopID] = stripDebugLocFromLoopID(LoopID); if (NewLoopID != LoopID) TermInst->setMetadata(LLVMContext::MD_loop, NewLoopID); } } return Changed; } bool llvm::StripDebugInfo(Module &M) { bool Changed = false; for (Module::named_metadata_iterator NMI = M.named_metadata_begin(), NME = M.named_metadata_end(); NMI != NME;) { NamedMDNode *NMD = &*NMI; ++NMI; // We're stripping debug info, and without them, coverage information // doesn't quite make sense. if (NMD->getName().startswith("llvm.dbg.") || NMD->getName() == "llvm.gcov") { NMD->eraseFromParent(); Changed = true; } } for (Function &F : M) Changed |= stripDebugInfo(F); for (auto &GV : M.globals()) { SmallVector MDs; GV.getMetadata(LLVMContext::MD_dbg, MDs); if (!MDs.empty()) { GV.eraseMetadata(LLVMContext::MD_dbg); Changed = true; } } if (GVMaterializer *Materializer = M.getMaterializer()) Materializer->setStripDebugInfo(); return Changed; } namespace { /// Helper class to downgrade -g metadata to -gline-tables-only metadata. class DebugTypeInfoRemoval { DenseMap Replacements; public: /// The (void)() type. MDNode *EmptySubroutineType; private: /// Remember what linkage name we originally had before stripping. If we end /// up making two subprograms identical who originally had different linkage /// names, then we need to make one of them distinct, to avoid them getting /// uniqued. Maps the new node to the old linkage name. DenseMap NewToLinkageName; // TODO: Remember the distinct subprogram we created for a given linkage name, // so that we can continue to unique whenever possible. Map to the first (possibly distinct) mdsubprogram // created for that combination. This is not strictly needed for correctness, // but can cut down on the number of MDNodes and let us diff cleanly with the // output of -gline-tables-only. public: DebugTypeInfoRemoval(LLVMContext &C) : EmptySubroutineType(DISubroutineType::get(C, DINode::FlagZero, 0, MDNode::get(C, {}))) {} Metadata *map(Metadata *M) { if (!M) return nullptr; auto Replacement = Replacements.find(M); if (Replacement != Replacements.end()) return Replacement->second; return M; } MDNode *mapNode(Metadata *N) { return dyn_cast_or_null(map(N)); } /// Recursively remap N and all its referenced children. Does a DF post-order /// traversal, so as to remap bottoms up. void traverseAndRemap(MDNode *N) { traverse(N); } private: // Create a new DISubprogram, to replace the one given. DISubprogram *getReplacementSubprogram(DISubprogram *MDS) { auto *FileAndScope = cast_or_null(map(MDS->getFile())); StringRef LinkageName = MDS->getName().empty() ? MDS->getLinkageName() : ""; DISubprogram *Declaration = nullptr; auto *Type = cast_or_null(map(MDS->getType())); DITypeRef ContainingType(map(MDS->getContainingType())); auto *Unit = cast_or_null(map(MDS->getUnit())); auto Variables = nullptr; auto TemplateParams = nullptr; // Make a distinct DISubprogram, for situations that warrent it. auto distinctMDSubprogram = [&]() { return DISubprogram::getDistinct( MDS->getContext(), FileAndScope, MDS->getName(), LinkageName, FileAndScope, MDS->getLine(), Type, MDS->isLocalToUnit(), MDS->isDefinition(), MDS->getScopeLine(), ContainingType, MDS->getVirtuality(), MDS->getVirtualIndex(), MDS->getThisAdjustment(), MDS->getFlags(), MDS->isOptimized(), Unit, TemplateParams, Declaration, Variables); }; if (MDS->isDistinct()) return distinctMDSubprogram(); auto *NewMDS = DISubprogram::get( MDS->getContext(), FileAndScope, MDS->getName(), LinkageName, FileAndScope, MDS->getLine(), Type, MDS->isLocalToUnit(), MDS->isDefinition(), MDS->getScopeLine(), ContainingType, MDS->getVirtuality(), MDS->getVirtualIndex(), MDS->getThisAdjustment(), MDS->getFlags(), MDS->isOptimized(), Unit, TemplateParams, Declaration, Variables); StringRef OldLinkageName = MDS->getLinkageName(); // See if we need to make a distinct one. auto OrigLinkage = NewToLinkageName.find(NewMDS); if (OrigLinkage != NewToLinkageName.end()) { if (OrigLinkage->second == OldLinkageName) // We're good. return NewMDS; // Otherwise, need to make a distinct one. // TODO: Query the map to see if we already have one. return distinctMDSubprogram(); } NewToLinkageName.insert({NewMDS, MDS->getLinkageName()}); return NewMDS; } /// Create a new compile unit, to replace the one given DICompileUnit *getReplacementCU(DICompileUnit *CU) { // Drop skeleton CUs. if (CU->getDWOId()) return nullptr; auto *File = cast_or_null(map(CU->getFile())); MDTuple *EnumTypes = nullptr; MDTuple *RetainedTypes = nullptr; MDTuple *GlobalVariables = nullptr; MDTuple *ImportedEntities = nullptr; return DICompileUnit::getDistinct( CU->getContext(), CU->getSourceLanguage(), File, CU->getProducer(), CU->isOptimized(), CU->getFlags(), CU->getRuntimeVersion(), CU->getSplitDebugFilename(), DICompileUnit::LineTablesOnly, EnumTypes, RetainedTypes, GlobalVariables, ImportedEntities, CU->getMacros(), CU->getDWOId(), CU->getSplitDebugInlining(), CU->getDebugInfoForProfiling(), CU->getGnuPubnames()); } DILocation *getReplacementMDLocation(DILocation *MLD) { auto *Scope = map(MLD->getScope()); auto *InlinedAt = map(MLD->getInlinedAt()); if (MLD->isDistinct()) return DILocation::getDistinct(MLD->getContext(), MLD->getLine(), MLD->getColumn(), Scope, InlinedAt); return DILocation::get(MLD->getContext(), MLD->getLine(), MLD->getColumn(), Scope, InlinedAt); } /// Create a new generic MDNode, to replace the one given MDNode *getReplacementMDNode(MDNode *N) { SmallVector Ops; Ops.reserve(N->getNumOperands()); for (auto &I : N->operands()) if (I) Ops.push_back(map(I)); auto *Ret = MDNode::get(N->getContext(), Ops); return Ret; } /// Attempt to re-map N to a newly created node. void remap(MDNode *N) { if (Replacements.count(N)) return; auto doRemap = [&](MDNode *N) -> MDNode * { if (!N) return nullptr; if (auto *MDSub = dyn_cast(N)) { remap(MDSub->getUnit()); return getReplacementSubprogram(MDSub); } if (isa(N)) return EmptySubroutineType; if (auto *CU = dyn_cast(N)) return getReplacementCU(CU); if (isa(N)) return N; if (auto *MDLB = dyn_cast(N)) // Remap to our referenced scope (recursively). return mapNode(MDLB->getScope()); if (auto *MLD = dyn_cast(N)) return getReplacementMDLocation(MLD); // Otherwise, if we see these, just drop them now. Not strictly necessary, // but this speeds things up a little. if (isa(N)) return nullptr; return getReplacementMDNode(N); }; Replacements[N] = doRemap(N); } /// Do the remapping traversal. void traverse(MDNode *); }; } // end anonymous namespace void DebugTypeInfoRemoval::traverse(MDNode *N) { if (!N || Replacements.count(N)) return; // To avoid cycles, as well as for efficiency sake, we will sometimes prune // parts of the graph. auto prune = [](MDNode *Parent, MDNode *Child) { if (auto *MDS = dyn_cast(Parent)) return Child == MDS->getVariables().get(); return false; }; SmallVector ToVisit; DenseSet Opened; // Visit each node starting at N in post order, and map them. ToVisit.push_back(N); while (!ToVisit.empty()) { auto *N = ToVisit.back(); if (!Opened.insert(N).second) { // Close it. remap(N); ToVisit.pop_back(); continue; } for (auto &I : N->operands()) if (auto *MDN = dyn_cast_or_null(I)) if (!Opened.count(MDN) && !Replacements.count(MDN) && !prune(N, MDN) && !isa(MDN)) ToVisit.push_back(MDN); } } bool llvm::stripNonLineTableDebugInfo(Module &M) { bool Changed = false; // First off, delete the debug intrinsics. auto RemoveUses = [&](StringRef Name) { if (auto *DbgVal = M.getFunction(Name)) { while (!DbgVal->use_empty()) cast(DbgVal->user_back())->eraseFromParent(); DbgVal->eraseFromParent(); Changed = true; } }; RemoveUses("llvm.dbg.declare"); RemoveUses("llvm.dbg.value"); // Delete non-CU debug info named metadata nodes. for (auto NMI = M.named_metadata_begin(), NME = M.named_metadata_end(); NMI != NME;) { NamedMDNode *NMD = &*NMI; ++NMI; // Specifically keep dbg.cu around. if (NMD->getName() == "llvm.dbg.cu") continue; } // Drop all dbg attachments from global variables. for (auto &GV : M.globals()) GV.eraseMetadata(LLVMContext::MD_dbg); DebugTypeInfoRemoval Mapper(M.getContext()); auto remap = [&](MDNode *Node) -> MDNode * { if (!Node) return nullptr; Mapper.traverseAndRemap(Node); auto *NewNode = Mapper.mapNode(Node); Changed |= Node != NewNode; Node = NewNode; return NewNode; }; // Rewrite the DebugLocs to be equivalent to what // -gline-tables-only would have created. for (auto &F : M) { if (auto *SP = F.getSubprogram()) { Mapper.traverseAndRemap(SP); auto *NewSP = cast(Mapper.mapNode(SP)); Changed |= SP != NewSP; F.setSubprogram(NewSP); } for (auto &BB : F) { for (auto &I : BB) { auto remapDebugLoc = [&](DebugLoc DL) -> DebugLoc { auto *Scope = DL.getScope(); MDNode *InlinedAt = DL.getInlinedAt(); Scope = remap(Scope); InlinedAt = remap(InlinedAt); return DebugLoc::get(DL.getLine(), DL.getCol(), Scope, InlinedAt); }; if (I.getDebugLoc() != DebugLoc()) I.setDebugLoc(remapDebugLoc(I.getDebugLoc())); // Remap DILocations in untyped MDNodes (e.g., llvm.loop). SmallVector, 2> MDs; I.getAllMetadata(MDs); for (auto Attachment : MDs) if (auto *T = dyn_cast_or_null(Attachment.second)) for (unsigned N = 0; N < T->getNumOperands(); ++N) if (auto *Loc = dyn_cast_or_null(T->getOperand(N))) if (Loc != DebugLoc()) T->replaceOperandWith(N, remapDebugLoc(Loc)); } } } // Create a new llvm.dbg.cu, which is equivalent to the one // -gline-tables-only would have created. for (auto &NMD : M.getNamedMDList()) { SmallVector Ops; for (MDNode *Op : NMD.operands()) Ops.push_back(remap(Op)); if (!Changed) continue; NMD.clearOperands(); for (auto *Op : Ops) if (Op) NMD.addOperand(Op); } return Changed; } unsigned llvm::getDebugMetadataVersionFromModule(const Module &M) { if (auto *Val = mdconst::dyn_extract_or_null( M.getModuleFlag("Debug Info Version"))) return Val->getZExtValue(); return 0; } void Instruction::applyMergedLocation(const DILocation *LocA, const DILocation *LocB) { setDebugLoc(DILocation::getMergedLocation(LocA, LocB, this)); } //===----------------------------------------------------------------------===// // LLVM C API implementations. //===----------------------------------------------------------------------===// static unsigned map_from_llvmDWARFsourcelanguage(LLVMDWARFSourceLanguage lang) { switch (lang) { #define HANDLE_DW_LANG(ID, NAME, VERSION, VENDOR) \ case LLVMDWARFSourceLanguage##NAME: return ID; #include "llvm/BinaryFormat/Dwarf.def" #undef HANDLE_DW_LANG } llvm_unreachable("Unhandled Tag"); } unsigned LLVMDebugMetadataVersion() { return DEBUG_METADATA_VERSION; } LLVMDIBuilderRef LLVMCreateDIBuilderDisallowUnresolved(LLVMModuleRef M) { return wrap(new DIBuilder(*unwrap(M), false)); } LLVMDIBuilderRef LLVMCreateDIBuilder(LLVMModuleRef M) { return wrap(new DIBuilder(*unwrap(M))); } unsigned LLVMGetModuleDebugMetadataVersion(LLVMModuleRef M) { return getDebugMetadataVersionFromModule(*unwrap(M)); } LLVMBool LLVMStripModuleDebugInfo(LLVMModuleRef M) { return StripDebugInfo(*unwrap(M)); } void LLVMDisposeDIBuilder(LLVMDIBuilderRef Builder) { delete unwrap(Builder); } void LLVMDIBuilderFinalize(LLVMDIBuilderRef Builder) { unwrap(Builder)->finalize(); } LLVMMetadataRef LLVMDIBuilderCreateCompileUnit( LLVMDIBuilderRef Builder, LLVMDWARFSourceLanguage Lang, LLVMMetadataRef FileRef, const char *Producer, size_t ProducerLen, LLVMBool isOptimized, const char *Flags, size_t FlagsLen, unsigned RuntimeVer, const char *SplitName, size_t SplitNameLen, LLVMDWARFEmissionKind Kind, unsigned DWOId, LLVMBool SplitDebugInlining, LLVMBool DebugInfoForProfiling) { auto File = unwrap(FileRef); return wrap(unwrap(Builder)->createCompileUnit( map_from_llvmDWARFsourcelanguage(Lang), File, StringRef(Producer, ProducerLen), isOptimized, StringRef(Flags, FlagsLen), RuntimeVer, StringRef(SplitName, SplitNameLen), static_cast(Kind), DWOId, SplitDebugInlining, DebugInfoForProfiling)); } LLVMMetadataRef LLVMDIBuilderCreateFile(LLVMDIBuilderRef Builder, const char *Filename, size_t FilenameLen, const char *Directory, size_t DirectoryLen) { return wrap(unwrap(Builder)->createFile(StringRef(Filename, FilenameLen), StringRef(Directory, DirectoryLen))); } LLVMMetadataRef LLVMDIBuilderCreateDebugLocation(LLVMContextRef Ctx, unsigned Line, unsigned Column, LLVMMetadataRef Scope, LLVMMetadataRef InlinedAt) { return wrap(DILocation::get(*unwrap(Ctx), Line, Column, unwrap(Scope), unwrap(InlinedAt))); }