| 1 | //===-- LTOModule.cpp - LLVM Link Time Optimizer --------------------------===// |
| 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 | // This file implements the Link Time Optimization library. This library is |
| 10 | // intended to be used by linker to optimize code at link time. |
| 11 | // |
| 12 | //===----------------------------------------------------------------------===// |
| 13 | |
| 14 | #include "llvm/LTO/legacy/LTOModule.h" |
| 15 | #include "llvm/Bitcode/BitcodeReader.h" |
| 16 | #include "llvm/CodeGen/TargetSubtargetInfo.h" |
| 17 | #include "llvm/IR/Constants.h" |
| 18 | #include "llvm/IR/LLVMContext.h" |
| 19 | #include "llvm/IR/Mangler.h" |
| 20 | #include "llvm/IR/Metadata.h" |
| 21 | #include "llvm/IR/Module.h" |
| 22 | #include "llvm/MC/MCExpr.h" |
| 23 | #include "llvm/MC/MCInst.h" |
| 24 | #include "llvm/MC/MCSection.h" |
| 25 | #include "llvm/MC/MCSymbol.h" |
| 26 | #include "llvm/MC/TargetRegistry.h" |
| 27 | #include "llvm/Object/IRObjectFile.h" |
| 28 | #include "llvm/Object/MachO.h" |
| 29 | #include "llvm/Object/ObjectFile.h" |
| 30 | #include "llvm/Support/FileSystem.h" |
| 31 | #include "llvm/Support/MemoryBuffer.h" |
| 32 | #include "llvm/Support/SourceMgr.h" |
| 33 | #include "llvm/Target/TargetLoweringObjectFile.h" |
| 34 | #include "llvm/TargetParser/Host.h" |
| 35 | #include "llvm/TargetParser/SubtargetFeature.h" |
| 36 | #include "llvm/TargetParser/Triple.h" |
| 37 | #include "llvm/Transforms/Utils/GlobalStatus.h" |
| 38 | #include <system_error> |
| 39 | using namespace llvm; |
| 40 | using namespace llvm::object; |
| 41 | |
| 42 | LTOModule::LTOModule(std::unique_ptr<Module> M, MemoryBufferRef MBRef, |
| 43 | llvm::TargetMachine *TM) |
| 44 | : Mod(std::move(M)), MBRef(MBRef), _target(TM) { |
| 45 | assert(_target && "target machine is null" ); |
| 46 | SymTab.addModule(M: Mod.get()); |
| 47 | } |
| 48 | |
| 49 | LTOModule::~LTOModule() = default; |
| 50 | |
| 51 | /// isBitcodeFile - Returns 'true' if the file (or memory contents) is LLVM |
| 52 | /// bitcode. |
| 53 | bool LTOModule::isBitcodeFile(const void *Mem, size_t Length) { |
| 54 | Expected<MemoryBufferRef> BCData = IRObjectFile::findBitcodeInMemBuffer( |
| 55 | Object: MemoryBufferRef(StringRef((const char *)Mem, Length), "<mem>" )); |
| 56 | return !errorToBool(Err: BCData.takeError()); |
| 57 | } |
| 58 | |
| 59 | bool LTOModule::isBitcodeFile(StringRef Path) { |
| 60 | ErrorOr<std::unique_ptr<MemoryBuffer>> BufferOrErr = |
| 61 | MemoryBuffer::getFile(Filename: Path); |
| 62 | if (!BufferOrErr) |
| 63 | return false; |
| 64 | |
| 65 | Expected<MemoryBufferRef> BCData = IRObjectFile::findBitcodeInMemBuffer( |
| 66 | Object: BufferOrErr.get()->getMemBufferRef()); |
| 67 | return !errorToBool(Err: BCData.takeError()); |
| 68 | } |
| 69 | |
| 70 | bool LTOModule::isThinLTO() { |
| 71 | Expected<BitcodeLTOInfo> Result = getBitcodeLTOInfo(Buffer: MBRef); |
| 72 | if (!Result) { |
| 73 | logAllUnhandledErrors(E: Result.takeError(), OS&: errs()); |
| 74 | return false; |
| 75 | } |
| 76 | return Result->IsThinLTO; |
| 77 | } |
| 78 | |
| 79 | bool LTOModule::isBitcodeForTarget(MemoryBuffer *Buffer, |
| 80 | StringRef TriplePrefix) { |
| 81 | Expected<MemoryBufferRef> BCOrErr = |
| 82 | IRObjectFile::findBitcodeInMemBuffer(Object: Buffer->getMemBufferRef()); |
| 83 | if (errorToBool(Err: BCOrErr.takeError())) |
| 84 | return false; |
| 85 | LLVMContext Context; |
| 86 | ErrorOr<std::string> TripleOrErr = |
| 87 | expectedToErrorOrAndEmitErrors(Ctx&: Context, Val: getBitcodeTargetTriple(Buffer: *BCOrErr)); |
| 88 | if (!TripleOrErr) |
| 89 | return false; |
| 90 | return StringRef(*TripleOrErr).starts_with(Prefix: TriplePrefix); |
| 91 | } |
| 92 | |
| 93 | std::string LTOModule::getProducerString(MemoryBuffer *Buffer) { |
| 94 | Expected<MemoryBufferRef> BCOrErr = |
| 95 | IRObjectFile::findBitcodeInMemBuffer(Object: Buffer->getMemBufferRef()); |
| 96 | if (errorToBool(Err: BCOrErr.takeError())) |
| 97 | return "" ; |
| 98 | LLVMContext Context; |
| 99 | ErrorOr<std::string> ProducerOrErr = expectedToErrorOrAndEmitErrors( |
| 100 | Ctx&: Context, Val: getBitcodeProducerString(Buffer: *BCOrErr)); |
| 101 | if (!ProducerOrErr) |
| 102 | return "" ; |
| 103 | return *ProducerOrErr; |
| 104 | } |
| 105 | |
| 106 | ErrorOr<std::unique_ptr<LTOModule>> |
| 107 | LTOModule::createFromFile(LLVMContext &Context, StringRef path, |
| 108 | const TargetOptions &options) { |
| 109 | ErrorOr<std::unique_ptr<MemoryBuffer>> BufferOrErr = |
| 110 | MemoryBuffer::getFile(Filename: path); |
| 111 | if (std::error_code EC = BufferOrErr.getError()) { |
| 112 | Context.emitError(ErrorStr: EC.message()); |
| 113 | return EC; |
| 114 | } |
| 115 | std::unique_ptr<MemoryBuffer> Buffer = std::move(BufferOrErr.get()); |
| 116 | return makeLTOModule(Buffer: Buffer->getMemBufferRef(), options, Context, |
| 117 | /* ShouldBeLazy*/ false); |
| 118 | } |
| 119 | |
| 120 | ErrorOr<std::unique_ptr<LTOModule>> |
| 121 | LTOModule::createFromOpenFile(LLVMContext &Context, int fd, StringRef path, |
| 122 | size_t size, const TargetOptions &options) { |
| 123 | return createFromOpenFileSlice(Context, fd, path, map_size: size, offset: 0, options); |
| 124 | } |
| 125 | |
| 126 | ErrorOr<std::unique_ptr<LTOModule>> |
| 127 | LTOModule::createFromOpenFileSlice(LLVMContext &Context, int fd, StringRef path, |
| 128 | size_t map_size, off_t offset, |
| 129 | const TargetOptions &options) { |
| 130 | ErrorOr<std::unique_ptr<MemoryBuffer>> BufferOrErr = |
| 131 | MemoryBuffer::getOpenFileSlice(FD: sys::fs::convertFDToNativeFile(FD: fd), Filename: path, |
| 132 | MapSize: map_size, Offset: offset); |
| 133 | if (std::error_code EC = BufferOrErr.getError()) { |
| 134 | Context.emitError(ErrorStr: EC.message()); |
| 135 | return EC; |
| 136 | } |
| 137 | std::unique_ptr<MemoryBuffer> Buffer = std::move(BufferOrErr.get()); |
| 138 | return makeLTOModule(Buffer: Buffer->getMemBufferRef(), options, Context, |
| 139 | /* ShouldBeLazy */ false); |
| 140 | } |
| 141 | |
| 142 | ErrorOr<std::unique_ptr<LTOModule>> |
| 143 | LTOModule::createFromBuffer(LLVMContext &Context, const void *mem, |
| 144 | size_t length, const TargetOptions &options, |
| 145 | StringRef path) { |
| 146 | StringRef Data((const char *)mem, length); |
| 147 | MemoryBufferRef Buffer(Data, path); |
| 148 | return makeLTOModule(Buffer, options, Context, /* ShouldBeLazy */ false); |
| 149 | } |
| 150 | |
| 151 | ErrorOr<std::unique_ptr<LTOModule>> |
| 152 | LTOModule::createInLocalContext(std::unique_ptr<LLVMContext> Context, |
| 153 | const void *mem, size_t length, |
| 154 | const TargetOptions &options, StringRef path) { |
| 155 | StringRef Data((const char *)mem, length); |
| 156 | MemoryBufferRef Buffer(Data, path); |
| 157 | // If we own a context, we know this is being used only for symbol extraction, |
| 158 | // not linking. Be lazy in that case. |
| 159 | ErrorOr<std::unique_ptr<LTOModule>> Ret = |
| 160 | makeLTOModule(Buffer, options, Context&: *Context, /* ShouldBeLazy */ true); |
| 161 | if (Ret) |
| 162 | (*Ret)->OwnedContext = std::move(Context); |
| 163 | return Ret; |
| 164 | } |
| 165 | |
| 166 | static ErrorOr<std::unique_ptr<Module>> |
| 167 | parseBitcodeFileImpl(MemoryBufferRef Buffer, LLVMContext &Context, |
| 168 | bool ShouldBeLazy) { |
| 169 | // Find the buffer. |
| 170 | Expected<MemoryBufferRef> MBOrErr = |
| 171 | IRObjectFile::findBitcodeInMemBuffer(Object: Buffer); |
| 172 | if (Error E = MBOrErr.takeError()) { |
| 173 | std::error_code EC = errorToErrorCode(Err: std::move(E)); |
| 174 | Context.emitError(ErrorStr: EC.message()); |
| 175 | return EC; |
| 176 | } |
| 177 | |
| 178 | if (!ShouldBeLazy) { |
| 179 | // Parse the full file. |
| 180 | return expectedToErrorOrAndEmitErrors(Ctx&: Context, |
| 181 | Val: parseBitcodeFile(Buffer: *MBOrErr, Context)); |
| 182 | } |
| 183 | |
| 184 | // Parse lazily. |
| 185 | return expectedToErrorOrAndEmitErrors( |
| 186 | Ctx&: Context, |
| 187 | Val: getLazyBitcodeModule(Buffer: *MBOrErr, Context, ShouldLazyLoadMetadata: true /*ShouldLazyLoadMetadata*/)); |
| 188 | } |
| 189 | |
| 190 | ErrorOr<std::unique_ptr<LTOModule>> |
| 191 | LTOModule::makeLTOModule(MemoryBufferRef Buffer, const TargetOptions &options, |
| 192 | LLVMContext &Context, bool ShouldBeLazy) { |
| 193 | ErrorOr<std::unique_ptr<Module>> MOrErr = |
| 194 | parseBitcodeFileImpl(Buffer, Context, ShouldBeLazy); |
| 195 | if (std::error_code EC = MOrErr.getError()) |
| 196 | return EC; |
| 197 | std::unique_ptr<Module> &M = *MOrErr; |
| 198 | |
| 199 | llvm::Triple Triple = M->getTargetTriple(); |
| 200 | if (Triple.empty()) |
| 201 | Triple = llvm::Triple(sys::getDefaultTargetTriple()); |
| 202 | |
| 203 | // find machine architecture for this module |
| 204 | std::string errMsg; |
| 205 | const Target *march = TargetRegistry::lookupTarget(TheTriple: Triple, Error&: errMsg); |
| 206 | if (!march) |
| 207 | return make_error_code(e: object::object_error::arch_not_found); |
| 208 | |
| 209 | // construct LTOModule, hand over ownership of module and target |
| 210 | SubtargetFeatures Features; |
| 211 | Features.getDefaultSubtargetFeatures(Triple); |
| 212 | std::string FeatureStr = Features.getString(); |
| 213 | // Set a default CPU for Darwin triples. |
| 214 | std::string CPU; |
| 215 | if (Triple.isOSDarwin()) { |
| 216 | if (Triple.getArch() == llvm::Triple::x86_64) |
| 217 | CPU = "core2" ; |
| 218 | else if (Triple.getArch() == llvm::Triple::x86) |
| 219 | CPU = "yonah" ; |
| 220 | else if (Triple.isArm64e()) |
| 221 | CPU = "apple-a12" ; |
| 222 | else if (Triple.getArch() == llvm::Triple::aarch64 || |
| 223 | Triple.getArch() == llvm::Triple::aarch64_32) |
| 224 | CPU = "cyclone" ; |
| 225 | } |
| 226 | |
| 227 | TargetMachine *target = march->createTargetMachine(TT: Triple, CPU, Features: FeatureStr, |
| 228 | Options: options, RM: std::nullopt); |
| 229 | |
| 230 | std::unique_ptr<LTOModule> Ret(new LTOModule(std::move(M), Buffer, target)); |
| 231 | Ret->parseSymbols(); |
| 232 | Ret->parseMetadata(); |
| 233 | |
| 234 | return std::move(Ret); |
| 235 | } |
| 236 | |
| 237 | /// Create a MemoryBuffer from a memory range with an optional name. |
| 238 | std::unique_ptr<MemoryBuffer> |
| 239 | LTOModule::makeBuffer(const void *mem, size_t length, StringRef name) { |
| 240 | const char *startPtr = (const char*)mem; |
| 241 | return MemoryBuffer::getMemBuffer(InputData: StringRef(startPtr, length), BufferName: name, RequiresNullTerminator: false); |
| 242 | } |
| 243 | |
| 244 | /// objcClassNameFromExpression - Get string that the data pointer points to. |
| 245 | bool |
| 246 | LTOModule::objcClassNameFromExpression(const Constant *c, std::string &name) { |
| 247 | if (const ConstantExpr *ce = dyn_cast<ConstantExpr>(Val: c)) { |
| 248 | Constant *op = ce->getOperand(i_nocapture: 0); |
| 249 | if (GlobalVariable *gvn = dyn_cast<GlobalVariable>(Val: op)) { |
| 250 | Constant *cn = gvn->getInitializer(); |
| 251 | if (ConstantDataArray *ca = dyn_cast<ConstantDataArray>(Val: cn)) { |
| 252 | if (ca->isCString()) { |
| 253 | name = (".objc_class_name_" + ca->getAsCString()).str(); |
| 254 | return true; |
| 255 | } |
| 256 | } |
| 257 | } |
| 258 | } |
| 259 | return false; |
| 260 | } |
| 261 | |
| 262 | /// addObjCClass - Parse i386/ppc ObjC class data structure. |
| 263 | void LTOModule::addObjCClass(const GlobalVariable *clgv) { |
| 264 | const ConstantStruct *c = dyn_cast<ConstantStruct>(Val: clgv->getInitializer()); |
| 265 | if (!c) return; |
| 266 | |
| 267 | // second slot in __OBJC,__class is pointer to superclass name |
| 268 | std::string superclassName; |
| 269 | if (objcClassNameFromExpression(c: c->getOperand(i_nocapture: 1), name&: superclassName)) { |
| 270 | auto IterBool = _undefines.try_emplace(Key: superclassName); |
| 271 | if (IterBool.second) { |
| 272 | NameAndAttributes &info = IterBool.first->second; |
| 273 | info.name = IterBool.first->first(); |
| 274 | info.attributes = LTO_SYMBOL_DEFINITION_UNDEFINED; |
| 275 | info.isFunction = false; |
| 276 | info.symbol = clgv; |
| 277 | } |
| 278 | } |
| 279 | |
| 280 | // third slot in __OBJC,__class is pointer to class name |
| 281 | std::string className; |
| 282 | if (objcClassNameFromExpression(c: c->getOperand(i_nocapture: 2), name&: className)) { |
| 283 | auto Iter = _defines.insert(key: className).first; |
| 284 | |
| 285 | NameAndAttributes info; |
| 286 | info.name = Iter->first(); |
| 287 | info.attributes = LTO_SYMBOL_PERMISSIONS_DATA | |
| 288 | LTO_SYMBOL_DEFINITION_REGULAR | LTO_SYMBOL_SCOPE_DEFAULT; |
| 289 | info.isFunction = false; |
| 290 | info.symbol = clgv; |
| 291 | _symbols.push_back(x: info); |
| 292 | } |
| 293 | } |
| 294 | |
| 295 | /// addObjCCategory - Parse i386/ppc ObjC category data structure. |
| 296 | void LTOModule::addObjCCategory(const GlobalVariable *clgv) { |
| 297 | const ConstantStruct *c = dyn_cast<ConstantStruct>(Val: clgv->getInitializer()); |
| 298 | if (!c) return; |
| 299 | |
| 300 | // second slot in __OBJC,__category is pointer to target class name |
| 301 | std::string targetclassName; |
| 302 | if (!objcClassNameFromExpression(c: c->getOperand(i_nocapture: 1), name&: targetclassName)) |
| 303 | return; |
| 304 | |
| 305 | auto IterBool = _undefines.try_emplace(Key: targetclassName); |
| 306 | |
| 307 | if (!IterBool.second) |
| 308 | return; |
| 309 | |
| 310 | NameAndAttributes &info = IterBool.first->second; |
| 311 | info.name = IterBool.first->first(); |
| 312 | info.attributes = LTO_SYMBOL_DEFINITION_UNDEFINED; |
| 313 | info.isFunction = false; |
| 314 | info.symbol = clgv; |
| 315 | } |
| 316 | |
| 317 | /// addObjCClassRef - Parse i386/ppc ObjC class list data structure. |
| 318 | void LTOModule::addObjCClassRef(const GlobalVariable *clgv) { |
| 319 | std::string targetclassName; |
| 320 | if (!objcClassNameFromExpression(c: clgv->getInitializer(), name&: targetclassName)) |
| 321 | return; |
| 322 | |
| 323 | auto IterBool = _undefines.try_emplace(Key: targetclassName); |
| 324 | |
| 325 | if (!IterBool.second) |
| 326 | return; |
| 327 | |
| 328 | NameAndAttributes &info = IterBool.first->second; |
| 329 | info.name = IterBool.first->first(); |
| 330 | info.attributes = LTO_SYMBOL_DEFINITION_UNDEFINED; |
| 331 | info.isFunction = false; |
| 332 | info.symbol = clgv; |
| 333 | } |
| 334 | |
| 335 | void LTOModule::addDefinedDataSymbol(ModuleSymbolTable::Symbol Sym) { |
| 336 | SmallString<64> Buffer; |
| 337 | { |
| 338 | raw_svector_ostream OS(Buffer); |
| 339 | SymTab.printSymbolName(OS, S: Sym); |
| 340 | Buffer.c_str(); |
| 341 | } |
| 342 | |
| 343 | const GlobalValue *V = cast<GlobalValue *>(Val&: Sym); |
| 344 | addDefinedDataSymbol(Name: Buffer, v: V); |
| 345 | } |
| 346 | |
| 347 | void LTOModule::addDefinedDataSymbol(StringRef Name, const GlobalValue *v) { |
| 348 | // Add to list of defined symbols. |
| 349 | addDefinedSymbol(Name, def: v, isFunction: false); |
| 350 | |
| 351 | if (!v->hasSection() /* || !isTargetDarwin */) |
| 352 | return; |
| 353 | |
| 354 | // Special case i386/ppc ObjC data structures in magic sections: |
| 355 | // The issue is that the old ObjC object format did some strange |
| 356 | // contortions to avoid real linker symbols. For instance, the |
| 357 | // ObjC class data structure is allocated statically in the executable |
| 358 | // that defines that class. That data structures contains a pointer to |
| 359 | // its superclass. But instead of just initializing that part of the |
| 360 | // struct to the address of its superclass, and letting the static and |
| 361 | // dynamic linkers do the rest, the runtime works by having that field |
| 362 | // instead point to a C-string that is the name of the superclass. |
| 363 | // At runtime the objc initialization updates that pointer and sets |
| 364 | // it to point to the actual super class. As far as the linker |
| 365 | // knows it is just a pointer to a string. But then someone wanted the |
| 366 | // linker to issue errors at build time if the superclass was not found. |
| 367 | // So they figured out a way in mach-o object format to use an absolute |
| 368 | // symbols (.objc_class_name_Foo = 0) and a floating reference |
| 369 | // (.reference .objc_class_name_Bar) to cause the linker into erroring when |
| 370 | // a class was missing. |
| 371 | // The following synthesizes the implicit .objc_* symbols for the linker |
| 372 | // from the ObjC data structures generated by the front end. |
| 373 | |
| 374 | // special case if this data blob is an ObjC class definition |
| 375 | if (const GlobalVariable *GV = dyn_cast<GlobalVariable>(Val: v)) { |
| 376 | StringRef Section = GV->getSection(); |
| 377 | if (Section.starts_with(Prefix: "__OBJC,__class," )) { |
| 378 | addObjCClass(clgv: GV); |
| 379 | } |
| 380 | |
| 381 | // special case if this data blob is an ObjC category definition |
| 382 | else if (Section.starts_with(Prefix: "__OBJC,__category," )) { |
| 383 | addObjCCategory(clgv: GV); |
| 384 | } |
| 385 | |
| 386 | // special case if this data blob is the list of referenced classes |
| 387 | else if (Section.starts_with(Prefix: "__OBJC,__cls_refs," )) { |
| 388 | addObjCClassRef(clgv: GV); |
| 389 | } |
| 390 | } |
| 391 | } |
| 392 | |
| 393 | void LTOModule::addDefinedFunctionSymbol(ModuleSymbolTable::Symbol Sym) { |
| 394 | SmallString<64> Buffer; |
| 395 | { |
| 396 | raw_svector_ostream OS(Buffer); |
| 397 | SymTab.printSymbolName(OS, S: Sym); |
| 398 | Buffer.c_str(); |
| 399 | } |
| 400 | |
| 401 | auto *GV = cast<GlobalValue *>(Val&: Sym); |
| 402 | assert((isa<Function>(GV) || |
| 403 | (isa<GlobalAlias>(GV) && |
| 404 | isa<Function>(cast<GlobalAlias>(GV)->getAliasee()))) && |
| 405 | "Not function or function alias" ); |
| 406 | |
| 407 | addDefinedFunctionSymbol(Name: Buffer, F: GV); |
| 408 | } |
| 409 | |
| 410 | void LTOModule::addDefinedFunctionSymbol(StringRef Name, const GlobalValue *F) { |
| 411 | // add to list of defined symbols |
| 412 | addDefinedSymbol(Name, def: F, isFunction: true); |
| 413 | } |
| 414 | |
| 415 | void LTOModule::addDefinedSymbol(StringRef Name, const GlobalValue *def, |
| 416 | bool isFunction) { |
| 417 | uint32_t attr = 0; |
| 418 | if (auto *gv = dyn_cast<GlobalVariable>(Val: def)) |
| 419 | attr = Log2(A: gv->getAlign().valueOrOne()); |
| 420 | else if (auto *f = dyn_cast<Function>(Val: def)) |
| 421 | attr = Log2(A: f->getAlign().valueOrOne()); |
| 422 | |
| 423 | // set permissions part |
| 424 | if (isFunction) { |
| 425 | attr |= LTO_SYMBOL_PERMISSIONS_CODE; |
| 426 | } else { |
| 427 | const GlobalVariable *gv = dyn_cast<GlobalVariable>(Val: def); |
| 428 | if (gv && gv->isConstant()) |
| 429 | attr |= LTO_SYMBOL_PERMISSIONS_RODATA; |
| 430 | else |
| 431 | attr |= LTO_SYMBOL_PERMISSIONS_DATA; |
| 432 | } |
| 433 | |
| 434 | // set definition part |
| 435 | if (def->hasWeakLinkage() || def->hasLinkOnceLinkage()) |
| 436 | attr |= LTO_SYMBOL_DEFINITION_WEAK; |
| 437 | else if (def->hasCommonLinkage()) |
| 438 | attr |= LTO_SYMBOL_DEFINITION_TENTATIVE; |
| 439 | else |
| 440 | attr |= LTO_SYMBOL_DEFINITION_REGULAR; |
| 441 | |
| 442 | // set scope part |
| 443 | if (def->hasLocalLinkage()) |
| 444 | // Ignore visibility if linkage is local. |
| 445 | attr |= LTO_SYMBOL_SCOPE_INTERNAL; |
| 446 | else if (def->hasHiddenVisibility()) |
| 447 | attr |= LTO_SYMBOL_SCOPE_HIDDEN; |
| 448 | else if (def->hasProtectedVisibility()) |
| 449 | attr |= LTO_SYMBOL_SCOPE_PROTECTED; |
| 450 | else if (def->canBeOmittedFromSymbolTable()) |
| 451 | attr |= LTO_SYMBOL_SCOPE_DEFAULT_CAN_BE_HIDDEN; |
| 452 | else |
| 453 | attr |= LTO_SYMBOL_SCOPE_DEFAULT; |
| 454 | |
| 455 | if (def->hasComdat()) |
| 456 | attr |= LTO_SYMBOL_COMDAT; |
| 457 | |
| 458 | if (isa<GlobalAlias>(Val: def)) |
| 459 | attr |= LTO_SYMBOL_ALIAS; |
| 460 | |
| 461 | auto Iter = _defines.insert(key: Name).first; |
| 462 | |
| 463 | // fill information structure |
| 464 | NameAndAttributes info; |
| 465 | StringRef NameRef = Iter->first(); |
| 466 | info.name = NameRef; |
| 467 | assert(NameRef.data()[NameRef.size()] == '\0'); |
| 468 | info.attributes = attr; |
| 469 | info.isFunction = isFunction; |
| 470 | info.symbol = def; |
| 471 | |
| 472 | // add to table of symbols |
| 473 | _symbols.push_back(x: info); |
| 474 | } |
| 475 | |
| 476 | /// addAsmGlobalSymbol - Add a global symbol from module-level ASM to the |
| 477 | /// defined list. |
| 478 | void LTOModule::addAsmGlobalSymbol(StringRef name, |
| 479 | lto_symbol_attributes scope) { |
| 480 | auto IterBool = _defines.insert(key: name); |
| 481 | |
| 482 | // only add new define if not already defined |
| 483 | if (!IterBool.second) |
| 484 | return; |
| 485 | |
| 486 | NameAndAttributes &info = _undefines[IterBool.first->first()]; |
| 487 | |
| 488 | if (info.symbol == nullptr) { |
| 489 | // FIXME: This is trying to take care of module ASM like this: |
| 490 | // |
| 491 | // module asm ".zerofill __FOO, __foo, _bar_baz_qux, 0" |
| 492 | // |
| 493 | // but is gross and its mother dresses it funny. Have the ASM parser give us |
| 494 | // more details for this type of situation so that we're not guessing so |
| 495 | // much. |
| 496 | |
| 497 | // fill information structure |
| 498 | info.name = IterBool.first->first(); |
| 499 | info.attributes = |
| 500 | LTO_SYMBOL_PERMISSIONS_DATA | LTO_SYMBOL_DEFINITION_REGULAR | scope; |
| 501 | info.isFunction = false; |
| 502 | info.symbol = nullptr; |
| 503 | |
| 504 | // add to table of symbols |
| 505 | _symbols.push_back(x: info); |
| 506 | return; |
| 507 | } |
| 508 | |
| 509 | if (info.isFunction) |
| 510 | addDefinedFunctionSymbol(Name: info.name, F: cast<Function>(Val: info.symbol)); |
| 511 | else |
| 512 | addDefinedDataSymbol(Name: info.name, v: info.symbol); |
| 513 | |
| 514 | _symbols.back().attributes &= ~LTO_SYMBOL_SCOPE_MASK; |
| 515 | _symbols.back().attributes |= scope; |
| 516 | } |
| 517 | |
| 518 | /// addAsmGlobalSymbolUndef - Add a global symbol from module-level ASM to the |
| 519 | /// undefined list. |
| 520 | void LTOModule::addAsmGlobalSymbolUndef(StringRef name) { |
| 521 | auto IterBool = _undefines.try_emplace(Key: name); |
| 522 | |
| 523 | _asm_undefines.push_back(x: IterBool.first->first()); |
| 524 | |
| 525 | // we already have the symbol |
| 526 | if (!IterBool.second) |
| 527 | return; |
| 528 | |
| 529 | uint32_t attr = LTO_SYMBOL_DEFINITION_UNDEFINED; |
| 530 | attr |= LTO_SYMBOL_SCOPE_DEFAULT; |
| 531 | NameAndAttributes &info = IterBool.first->second; |
| 532 | info.name = IterBool.first->first(); |
| 533 | info.attributes = attr; |
| 534 | info.isFunction = false; |
| 535 | info.symbol = nullptr; |
| 536 | } |
| 537 | |
| 538 | /// Add a symbol which isn't defined just yet to a list to be resolved later. |
| 539 | void LTOModule::addPotentialUndefinedSymbol(ModuleSymbolTable::Symbol Sym, |
| 540 | bool isFunc) { |
| 541 | SmallString<64> name; |
| 542 | { |
| 543 | raw_svector_ostream OS(name); |
| 544 | SymTab.printSymbolName(OS, S: Sym); |
| 545 | name.c_str(); |
| 546 | } |
| 547 | |
| 548 | auto IterBool = _undefines.try_emplace(Key: name.str()); |
| 549 | |
| 550 | // we already have the symbol |
| 551 | if (!IterBool.second) |
| 552 | return; |
| 553 | |
| 554 | NameAndAttributes &info = IterBool.first->second; |
| 555 | |
| 556 | info.name = IterBool.first->first(); |
| 557 | |
| 558 | const GlobalValue *decl = dyn_cast_if_present<GlobalValue *>(Val&: Sym); |
| 559 | |
| 560 | if (decl->hasExternalWeakLinkage()) |
| 561 | info.attributes = LTO_SYMBOL_DEFINITION_WEAKUNDEF; |
| 562 | else |
| 563 | info.attributes = LTO_SYMBOL_DEFINITION_UNDEFINED; |
| 564 | |
| 565 | info.isFunction = isFunc; |
| 566 | info.symbol = decl; |
| 567 | } |
| 568 | |
| 569 | void LTOModule::parseSymbols() { |
| 570 | for (auto Sym : SymTab.symbols()) { |
| 571 | auto *GV = dyn_cast_if_present<GlobalValue *>(Val&: Sym); |
| 572 | uint32_t Flags = SymTab.getSymbolFlags(S: Sym); |
| 573 | if (Flags & object::BasicSymbolRef::SF_FormatSpecific) |
| 574 | continue; |
| 575 | |
| 576 | bool IsUndefined = Flags & object::BasicSymbolRef::SF_Undefined; |
| 577 | |
| 578 | if (!GV) { |
| 579 | SmallString<64> Buffer; |
| 580 | { |
| 581 | raw_svector_ostream OS(Buffer); |
| 582 | SymTab.printSymbolName(OS, S: Sym); |
| 583 | Buffer.c_str(); |
| 584 | } |
| 585 | StringRef Name = Buffer; |
| 586 | |
| 587 | if (IsUndefined) |
| 588 | addAsmGlobalSymbolUndef(name: Name); |
| 589 | else if (Flags & object::BasicSymbolRef::SF_Global) |
| 590 | addAsmGlobalSymbol(name: Name, scope: LTO_SYMBOL_SCOPE_DEFAULT); |
| 591 | else |
| 592 | addAsmGlobalSymbol(name: Name, scope: LTO_SYMBOL_SCOPE_INTERNAL); |
| 593 | continue; |
| 594 | } |
| 595 | |
| 596 | auto *F = dyn_cast<Function>(Val: GV); |
| 597 | if (IsUndefined) { |
| 598 | addPotentialUndefinedSymbol(Sym, isFunc: F != nullptr); |
| 599 | continue; |
| 600 | } |
| 601 | |
| 602 | if (F) { |
| 603 | addDefinedFunctionSymbol(Sym); |
| 604 | continue; |
| 605 | } |
| 606 | |
| 607 | if (isa<GlobalVariable>(Val: GV)) { |
| 608 | addDefinedDataSymbol(Sym); |
| 609 | continue; |
| 610 | } |
| 611 | |
| 612 | assert(isa<GlobalAlias>(GV)); |
| 613 | |
| 614 | if (isa<Function>(Val: cast<GlobalAlias>(Val: GV)->getAliasee())) |
| 615 | addDefinedFunctionSymbol(Sym); |
| 616 | else |
| 617 | addDefinedDataSymbol(Sym); |
| 618 | } |
| 619 | |
| 620 | // make symbols for all undefines |
| 621 | for (StringMap<NameAndAttributes>::iterator u =_undefines.begin(), |
| 622 | e = _undefines.end(); u != e; ++u) { |
| 623 | // If this symbol also has a definition, then don't make an undefine because |
| 624 | // it is a tentative definition. |
| 625 | if (_defines.count(Key: u->getKey())) continue; |
| 626 | NameAndAttributes info = u->getValue(); |
| 627 | _symbols.push_back(x: info); |
| 628 | } |
| 629 | } |
| 630 | |
| 631 | /// parseMetadata - Parse metadata from the module |
| 632 | void LTOModule::parseMetadata() { |
| 633 | raw_string_ostream OS(LinkerOpts); |
| 634 | |
| 635 | // Linker Options |
| 636 | if (NamedMDNode *LinkerOptions = |
| 637 | getModule().getNamedMetadata(Name: "llvm.linker.options" )) { |
| 638 | for (unsigned i = 0, e = LinkerOptions->getNumOperands(); i != e; ++i) { |
| 639 | MDNode *MDOptions = LinkerOptions->getOperand(i); |
| 640 | for (unsigned ii = 0, ie = MDOptions->getNumOperands(); ii != ie; ++ii) { |
| 641 | MDString *MDOption = cast<MDString>(Val: MDOptions->getOperand(I: ii)); |
| 642 | OS << " " << MDOption->getString(); |
| 643 | } |
| 644 | } |
| 645 | } |
| 646 | |
| 647 | // Globals - we only need to do this for COFF. |
| 648 | const Triple TT(_target->getTargetTriple()); |
| 649 | if (!TT.isOSBinFormatCOFF()) |
| 650 | return; |
| 651 | Mangler M; |
| 652 | for (const NameAndAttributes &Sym : _symbols) { |
| 653 | if (!Sym.symbol) |
| 654 | continue; |
| 655 | emitLinkerFlagsForGlobalCOFF(OS, GV: Sym.symbol, TT, Mangler&: M); |
| 656 | } |
| 657 | } |
| 658 | |
| 659 | lto::InputFile *LTOModule::createInputFile(const void *buffer, |
| 660 | size_t buffer_size, const char *path, |
| 661 | std::string &outErr) { |
| 662 | StringRef Data((const char *)buffer, buffer_size); |
| 663 | MemoryBufferRef BufferRef(Data, path); |
| 664 | |
| 665 | Expected<std::unique_ptr<lto::InputFile>> ObjOrErr = |
| 666 | lto::InputFile::create(Object: BufferRef); |
| 667 | |
| 668 | if (ObjOrErr) |
| 669 | return ObjOrErr->release(); |
| 670 | |
| 671 | outErr = std::string(path) + |
| 672 | ": Could not read LTO input file: " + toString(E: ObjOrErr.takeError()); |
| 673 | return nullptr; |
| 674 | } |
| 675 | |
| 676 | size_t LTOModule::getDependentLibraryCount(lto::InputFile *input) { |
| 677 | return input->getDependentLibraries().size(); |
| 678 | } |
| 679 | |
| 680 | const char *LTOModule::getDependentLibrary(lto::InputFile *input, size_t index, |
| 681 | size_t *size) { |
| 682 | StringRef S = input->getDependentLibraries()[index]; |
| 683 | *size = S.size(); |
| 684 | return S.data(); |
| 685 | } |
| 686 | |
| 687 | Expected<uint32_t> LTOModule::getMachOCPUType() const { |
| 688 | return MachO::getCPUType(T: Mod->getTargetTriple()); |
| 689 | } |
| 690 | |
| 691 | Expected<uint32_t> LTOModule::getMachOCPUSubType() const { |
| 692 | return MachO::getCPUSubType(T: Mod->getTargetTriple()); |
| 693 | } |
| 694 | |
| 695 | bool LTOModule::hasCtorDtor() const { |
| 696 | for (auto Sym : SymTab.symbols()) { |
| 697 | if (auto *GV = dyn_cast_if_present<GlobalValue *>(Val&: Sym)) { |
| 698 | StringRef Name = GV->getName(); |
| 699 | if (Name.consume_front(Prefix: "llvm.global_" )) { |
| 700 | if (Name == "ctors" || Name == "dtors" ) |
| 701 | return true; |
| 702 | } |
| 703 | } |
| 704 | } |
| 705 | return false; |
| 706 | } |
| 707 | |