1 //===--- ObjectFilePCHContainerOperations.cpp -----------------------------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 9 #include "clang/CodeGen/ObjectFilePCHContainerOperations.h" 10 #include "CGDebugInfo.h" 11 #include "CodeGenModule.h" 12 #include "clang/AST/ASTContext.h" 13 #include "clang/AST/DeclObjC.h" 14 #include "clang/AST/Expr.h" 15 #include "clang/AST/RecursiveASTVisitor.h" 16 #include "clang/Basic/CodeGenOptions.h" 17 #include "clang/Basic/Diagnostic.h" 18 #include "clang/Basic/TargetInfo.h" 19 #include "clang/CodeGen/BackendUtil.h" 20 #include "clang/Frontend/CompilerInstance.h" 21 #include "clang/Lex/HeaderSearch.h" 22 #include "clang/Lex/Preprocessor.h" 23 #include "llvm/ADT/StringRef.h" 24 #include "llvm/Bitstream/BitstreamReader.h" 25 #include "llvm/DebugInfo/DWARF/DWARFContext.h" 26 #include "llvm/IR/Constants.h" 27 #include "llvm/IR/DataLayout.h" 28 #include "llvm/IR/LLVMContext.h" 29 #include "llvm/IR/Module.h" 30 #include "llvm/MC/TargetRegistry.h" 31 #include "llvm/Object/COFF.h" 32 #include "llvm/Object/ObjectFile.h" 33 #include "llvm/Support/Path.h" 34 #include <memory> 35 #include <utility> 36 37 using namespace clang; 38 39 #define DEBUG_TYPE "pchcontainer" 40 41 namespace { 42 class PCHContainerGenerator : public ASTConsumer { 43 DiagnosticsEngine &Diags; 44 const std::string MainFileName; 45 const std::string OutputFileName; 46 ASTContext *Ctx; 47 ModuleMap &MMap; 48 const HeaderSearchOptions &HeaderSearchOpts; 49 const PreprocessorOptions &PreprocessorOpts; 50 CodeGenOptions CodeGenOpts; 51 const TargetOptions TargetOpts; 52 LangOptions LangOpts; 53 std::unique_ptr<llvm::LLVMContext> VMContext; 54 std::unique_ptr<llvm::Module> M; 55 std::unique_ptr<CodeGen::CodeGenModule> Builder; 56 std::unique_ptr<raw_pwrite_stream> OS; 57 std::shared_ptr<PCHBuffer> Buffer; 58 59 /// Visit every type and emit debug info for it. 60 struct DebugTypeVisitor : public RecursiveASTVisitor<DebugTypeVisitor> { 61 clang::CodeGen::CGDebugInfo &DI; 62 ASTContext &Ctx; 63 DebugTypeVisitor(clang::CodeGen::CGDebugInfo &DI, ASTContext &Ctx) 64 : DI(DI), Ctx(Ctx) {} 65 66 /// Determine whether this type can be represented in DWARF. 67 static bool CanRepresent(const Type *Ty) { 68 return !Ty->isDependentType() && !Ty->isUndeducedType(); 69 } 70 71 bool VisitImportDecl(ImportDecl *D) { 72 if (!D->getImportedOwningModule()) 73 DI.EmitImportDecl(*D); 74 return true; 75 } 76 77 bool VisitTypeDecl(TypeDecl *D) { 78 // TagDecls may be deferred until after all decls have been merged and we 79 // know the complete type. Pure forward declarations will be skipped, but 80 // they don't need to be emitted into the module anyway. 81 if (auto *TD = dyn_cast<TagDecl>(D)) 82 if (!TD->isCompleteDefinition()) 83 return true; 84 85 QualType QualTy = Ctx.getTypeDeclType(D); 86 if (!QualTy.isNull() && CanRepresent(QualTy.getTypePtr())) 87 DI.getOrCreateStandaloneType(QualTy, D->getLocation()); 88 return true; 89 } 90 91 bool VisitObjCInterfaceDecl(ObjCInterfaceDecl *D) { 92 QualType QualTy(D->getTypeForDecl(), 0); 93 if (!QualTy.isNull() && CanRepresent(QualTy.getTypePtr())) 94 DI.getOrCreateStandaloneType(QualTy, D->getLocation()); 95 return true; 96 } 97 98 bool VisitFunctionDecl(FunctionDecl *D) { 99 if (isa<CXXMethodDecl>(D)) 100 // This is not yet supported. Constructing the `this' argument 101 // mandates a CodeGenFunction. 102 return true; 103 104 SmallVector<QualType, 16> ArgTypes; 105 for (auto i : D->parameters()) 106 ArgTypes.push_back(i->getType()); 107 QualType RetTy = D->getReturnType(); 108 QualType FnTy = Ctx.getFunctionType(RetTy, ArgTypes, 109 FunctionProtoType::ExtProtoInfo()); 110 if (CanRepresent(FnTy.getTypePtr())) 111 DI.EmitFunctionDecl(D, D->getLocation(), FnTy); 112 return true; 113 } 114 115 bool VisitObjCMethodDecl(ObjCMethodDecl *D) { 116 if (!D->getClassInterface()) 117 return true; 118 119 bool selfIsPseudoStrong, selfIsConsumed; 120 SmallVector<QualType, 16> ArgTypes; 121 ArgTypes.push_back(D->getSelfType(Ctx, D->getClassInterface(), 122 selfIsPseudoStrong, selfIsConsumed)); 123 ArgTypes.push_back(Ctx.getObjCSelType()); 124 for (auto i : D->parameters()) 125 ArgTypes.push_back(i->getType()); 126 QualType RetTy = D->getReturnType(); 127 QualType FnTy = Ctx.getFunctionType(RetTy, ArgTypes, 128 FunctionProtoType::ExtProtoInfo()); 129 if (CanRepresent(FnTy.getTypePtr())) 130 DI.EmitFunctionDecl(D, D->getLocation(), FnTy); 131 return true; 132 } 133 }; 134 135 public: 136 PCHContainerGenerator(CompilerInstance &CI, const std::string &MainFileName, 137 const std::string &OutputFileName, 138 std::unique_ptr<raw_pwrite_stream> OS, 139 std::shared_ptr<PCHBuffer> Buffer) 140 : Diags(CI.getDiagnostics()), MainFileName(MainFileName), 141 OutputFileName(OutputFileName), Ctx(nullptr), 142 MMap(CI.getPreprocessor().getHeaderSearchInfo().getModuleMap()), 143 HeaderSearchOpts(CI.getHeaderSearchOpts()), 144 PreprocessorOpts(CI.getPreprocessorOpts()), 145 TargetOpts(CI.getTargetOpts()), LangOpts(CI.getLangOpts()), 146 OS(std::move(OS)), Buffer(std::move(Buffer)) { 147 // The debug info output isn't affected by CodeModel and 148 // ThreadModel, but the backend expects them to be nonempty. 149 CodeGenOpts.CodeModel = "default"; 150 LangOpts.setThreadModel(LangOptions::ThreadModelKind::Single); 151 CodeGenOpts.DebugTypeExtRefs = true; 152 // When building a module MainFileName is the name of the modulemap file. 153 CodeGenOpts.MainFileName = 154 LangOpts.CurrentModule.empty() ? MainFileName : LangOpts.CurrentModule; 155 CodeGenOpts.setDebugInfo(codegenoptions::FullDebugInfo); 156 CodeGenOpts.setDebuggerTuning(CI.getCodeGenOpts().getDebuggerTuning()); 157 CodeGenOpts.DebugPrefixMap = 158 CI.getInvocation().getCodeGenOpts().DebugPrefixMap; 159 CodeGenOpts.DebugStrictDwarf = CI.getCodeGenOpts().DebugStrictDwarf; 160 } 161 162 ~PCHContainerGenerator() override = default; 163 164 void Initialize(ASTContext &Context) override { 165 assert(!Ctx && "initialized multiple times"); 166 167 Ctx = &Context; 168 VMContext.reset(new llvm::LLVMContext()); 169 M.reset(new llvm::Module(MainFileName, *VMContext)); 170 M->setDataLayout(Ctx->getTargetInfo().getDataLayoutString()); 171 Builder.reset(new CodeGen::CodeGenModule( 172 *Ctx, HeaderSearchOpts, PreprocessorOpts, CodeGenOpts, *M, Diags)); 173 174 // Prepare CGDebugInfo to emit debug info for a clang module. 175 auto *DI = Builder->getModuleDebugInfo(); 176 StringRef ModuleName = llvm::sys::path::filename(MainFileName); 177 DI->setPCHDescriptor( 178 {ModuleName, "", OutputFileName, ASTFileSignature::createDISentinel()}); 179 DI->setModuleMap(MMap); 180 } 181 182 bool HandleTopLevelDecl(DeclGroupRef D) override { 183 if (Diags.hasErrorOccurred()) 184 return true; 185 186 // Collect debug info for all decls in this group. 187 for (auto *I : D) 188 if (!I->isFromASTFile()) { 189 DebugTypeVisitor DTV(*Builder->getModuleDebugInfo(), *Ctx); 190 DTV.TraverseDecl(I); 191 } 192 return true; 193 } 194 195 void HandleTopLevelDeclInObjCContainer(DeclGroupRef D) override { 196 HandleTopLevelDecl(D); 197 } 198 199 void HandleTagDeclDefinition(TagDecl *D) override { 200 if (Diags.hasErrorOccurred()) 201 return; 202 203 if (D->isFromASTFile()) 204 return; 205 206 // Anonymous tag decls are deferred until we are building their declcontext. 207 if (D->getName().empty()) 208 return; 209 210 // Defer tag decls until their declcontext is complete. 211 auto *DeclCtx = D->getDeclContext(); 212 while (DeclCtx) { 213 if (auto *D = dyn_cast<TagDecl>(DeclCtx)) 214 if (!D->isCompleteDefinition()) 215 return; 216 DeclCtx = DeclCtx->getParent(); 217 } 218 219 DebugTypeVisitor DTV(*Builder->getModuleDebugInfo(), *Ctx); 220 DTV.TraverseDecl(D); 221 Builder->UpdateCompletedType(D); 222 } 223 224 void HandleTagDeclRequiredDefinition(const TagDecl *D) override { 225 if (Diags.hasErrorOccurred()) 226 return; 227 228 if (const RecordDecl *RD = dyn_cast<RecordDecl>(D)) 229 Builder->getModuleDebugInfo()->completeRequiredType(RD); 230 } 231 232 void HandleImplicitImportDecl(ImportDecl *D) override { 233 if (!D->getImportedOwningModule()) 234 Builder->getModuleDebugInfo()->EmitImportDecl(*D); 235 } 236 237 /// Emit a container holding the serialized AST. 238 void HandleTranslationUnit(ASTContext &Ctx) override { 239 assert(M && VMContext && Builder); 240 // Delete these on function exit. 241 std::unique_ptr<llvm::LLVMContext> VMContext = std::move(this->VMContext); 242 std::unique_ptr<llvm::Module> M = std::move(this->M); 243 std::unique_ptr<CodeGen::CodeGenModule> Builder = std::move(this->Builder); 244 245 if (Diags.hasErrorOccurred()) 246 return; 247 248 M->setTargetTriple(Ctx.getTargetInfo().getTriple().getTriple()); 249 M->setDataLayout(Ctx.getTargetInfo().getDataLayoutString()); 250 251 // PCH files don't have a signature field in the control block, 252 // but LLVM detects DWO CUs by looking for a non-zero DWO id. 253 // We use the lower 64 bits for debug info. 254 255 uint64_t Signature = 256 Buffer->Signature ? Buffer->Signature.truncatedValue() : ~1ULL; 257 258 Builder->getModuleDebugInfo()->setDwoId(Signature); 259 260 // Finalize the Builder. 261 if (Builder) 262 Builder->Release(); 263 264 // Ensure the target exists. 265 std::string Error; 266 auto Triple = Ctx.getTargetInfo().getTriple(); 267 if (!llvm::TargetRegistry::lookupTarget(Triple.getTriple(), Error)) 268 llvm::report_fatal_error(llvm::Twine(Error)); 269 270 // Emit the serialized Clang AST into its own section. 271 assert(Buffer->IsComplete && "serialization did not complete"); 272 auto &SerializedAST = Buffer->Data; 273 auto Size = SerializedAST.size(); 274 275 if (Triple.isOSBinFormatWasm()) { 276 // Emit __clangast in custom section instead of named data segment 277 // to find it while iterating sections. 278 // This could be avoided if all data segements (the wasm sense) were 279 // represented as their own sections (in the llvm sense). 280 // TODO: https://github.com/WebAssembly/tool-conventions/issues/138 281 llvm::NamedMDNode *MD = 282 M->getOrInsertNamedMetadata("wasm.custom_sections"); 283 llvm::Metadata *Ops[2] = { 284 llvm::MDString::get(*VMContext, "__clangast"), 285 llvm::MDString::get(*VMContext, 286 StringRef(SerializedAST.data(), Size))}; 287 auto *NameAndContent = llvm::MDTuple::get(*VMContext, Ops); 288 MD->addOperand(NameAndContent); 289 } else { 290 auto Int8Ty = llvm::Type::getInt8Ty(*VMContext); 291 auto *Ty = llvm::ArrayType::get(Int8Ty, Size); 292 auto *Data = llvm::ConstantDataArray::getString( 293 *VMContext, StringRef(SerializedAST.data(), Size), 294 /*AddNull=*/false); 295 auto *ASTSym = new llvm::GlobalVariable( 296 *M, Ty, /*constant*/ true, llvm::GlobalVariable::InternalLinkage, 297 Data, "__clang_ast"); 298 // The on-disk hashtable needs to be aligned. 299 ASTSym->setAlignment(llvm::Align(8)); 300 301 // Mach-O also needs a segment name. 302 if (Triple.isOSBinFormatMachO()) 303 ASTSym->setSection("__CLANG,__clangast"); 304 // COFF has an eight character length limit. 305 else if (Triple.isOSBinFormatCOFF()) 306 ASTSym->setSection("clangast"); 307 else 308 ASTSym->setSection("__clangast"); 309 } 310 311 LLVM_DEBUG({ 312 // Print the IR for the PCH container to the debug output. 313 llvm::SmallString<0> Buffer; 314 clang::EmitBackendOutput( 315 Diags, HeaderSearchOpts, CodeGenOpts, TargetOpts, LangOpts, 316 Ctx.getTargetInfo().getDataLayoutString(), M.get(), 317 BackendAction::Backend_EmitLL, 318 std::make_unique<llvm::raw_svector_ostream>(Buffer)); 319 llvm::dbgs() << Buffer; 320 }); 321 322 // Use the LLVM backend to emit the pch container. 323 clang::EmitBackendOutput(Diags, HeaderSearchOpts, CodeGenOpts, TargetOpts, 324 LangOpts, 325 Ctx.getTargetInfo().getDataLayoutString(), M.get(), 326 BackendAction::Backend_EmitObj, std::move(OS)); 327 328 // Free the memory for the temporary buffer. 329 llvm::SmallVector<char, 0> Empty; 330 SerializedAST = std::move(Empty); 331 } 332 }; 333 334 } // anonymous namespace 335 336 std::unique_ptr<ASTConsumer> 337 ObjectFilePCHContainerWriter::CreatePCHContainerGenerator( 338 CompilerInstance &CI, const std::string &MainFileName, 339 const std::string &OutputFileName, 340 std::unique_ptr<llvm::raw_pwrite_stream> OS, 341 std::shared_ptr<PCHBuffer> Buffer) const { 342 return std::make_unique<PCHContainerGenerator>( 343 CI, MainFileName, OutputFileName, std::move(OS), Buffer); 344 } 345 346 StringRef 347 ObjectFilePCHContainerReader::ExtractPCH(llvm::MemoryBufferRef Buffer) const { 348 StringRef PCH; 349 auto OFOrErr = llvm::object::ObjectFile::createObjectFile(Buffer); 350 if (OFOrErr) { 351 auto &OF = OFOrErr.get(); 352 bool IsCOFF = isa<llvm::object::COFFObjectFile>(*OF); 353 // Find the clang AST section in the container. 354 for (auto &Section : OF->sections()) { 355 StringRef Name; 356 if (Expected<StringRef> NameOrErr = Section.getName()) 357 Name = *NameOrErr; 358 else 359 consumeError(NameOrErr.takeError()); 360 361 if ((!IsCOFF && Name == "__clangast") || (IsCOFF && Name == "clangast")) { 362 if (Expected<StringRef> E = Section.getContents()) 363 return *E; 364 else { 365 handleAllErrors(E.takeError(), [&](const llvm::ErrorInfoBase &EIB) { 366 EIB.log(llvm::errs()); 367 }); 368 return ""; 369 } 370 } 371 } 372 } 373 handleAllErrors(OFOrErr.takeError(), [&](const llvm::ErrorInfoBase &EIB) { 374 if (EIB.convertToErrorCode() == 375 llvm::object::object_error::invalid_file_type) 376 // As a fallback, treat the buffer as a raw AST. 377 PCH = Buffer.getBuffer(); 378 else 379 EIB.log(llvm::errs()); 380 }); 381 return PCH; 382 } 383