1 //===- GNULDBackend.h -----------------------------------------------------===// 2 // 3 // The MCLinker Project 4 // 5 // This file is distributed under the University of Illinois Open Source 6 // License. See LICENSE.TXT for details. 7 // 8 //===----------------------------------------------------------------------===// 9 #ifndef MCLD_TARGET_GNU_LDBACKEND_H 10 #define MCLD_TARGET_GNU_LDBACKEND_H 11 #ifdef ENABLE_UNITTEST 12 #include <gtest.h> 13 #endif 14 #include <mcld/Target/TargetLDBackend.h> 15 16 #include <llvm/Support/ELF.h> 17 #include <mcld/ADT/HashTable.h> 18 #include <mcld/ADT/HashEntry.h> 19 #include <mcld/LD/ELFDynObjFileFormat.h> 20 #include <mcld/LD/ELFExecFileFormat.h> 21 #include <mcld/LD/ELFObjectFileFormat.h> 22 #include <mcld/LD/GNUArchiveReader.h> 23 #include <mcld/LD/ELFObjectReader.h> 24 #include <mcld/LD/ELFDynObjReader.h> 25 #include <mcld/LD/ELFBinaryReader.h> 26 #include <mcld/LD/ELFObjectWriter.h> 27 #include <mcld/LD/ELFSegment.h> 28 #include <mcld/LD/ELFSegmentFactory.h> 29 #include <mcld/Target/ELFDynamic.h> 30 #include <mcld/Target/GNUInfo.h> 31 32 #include <mcld/Support/GCFactory.h> 33 #include <mcld/Module.h> 34 35 namespace mcld { 36 37 class Module; 38 class LinkerConfig; 39 class IRBuilder; 40 class Layout; 41 class EhFrameHdr; 42 class BranchIslandFactory; 43 class StubFactory; 44 class GNUInfo; 45 46 /** \class GNULDBackend 47 * \brief GNULDBackend provides a common interface for all GNU Unix-OS 48 * LDBackend. 49 */ 50 class GNULDBackend : public TargetLDBackend 51 { 52 protected: 53 GNULDBackend(const LinkerConfig& pConfig, GNUInfo* pInfo); 54 55 public: 56 virtual ~GNULDBackend(); 57 58 // ----- readers/writers ----- // 59 GNUArchiveReader* createArchiveReader(Module& pModule); 60 ELFObjectReader* createObjectReader(IRBuilder& pBuilder); 61 ELFDynObjReader* createDynObjReader(IRBuilder& pBuilder); 62 ELFBinaryReader* createBinaryReader(IRBuilder& pBuilder); 63 ELFObjectWriter* createWriter(); 64 65 // ----- output sections ----- // 66 /// initStdSections - initialize standard sections of the output file. 67 bool initStdSections(ObjectBuilder& pBuilder); 68 69 /// getOutputFormat - get the sections of the output file. 70 const ELFFileFormat* getOutputFormat() const; 71 ELFFileFormat* getOutputFormat(); 72 73 // ----- target symbols ----- // 74 /// initStandardSymbols - initialize standard symbols. 75 /// Some section symbols is undefined in input object, and linkers must set 76 /// up its value. Take __init_array_begin for example. This symbol is an 77 /// undefined symbol in input objects. ObjectLinker must finalize its value 78 /// to the begin of the .init_array section, then relocation enties to 79 /// __init_array_begin can be applied without emission of "undefined 80 /// reference to `__init_array_begin'". 81 bool initStandardSymbols(IRBuilder& pBuilder, Module& pModule); 82 83 /// finalizeSymbol - Linker checks pSymbol.reserved() if it's not zero, 84 /// then it will ask backend to finalize the symbol value. 85 /// @return ture - if backend set the symbol value sucessfully 86 /// @return false - if backend do not recognize the symbol finalizeSymbols()87 bool finalizeSymbols() { 88 return (finalizeStandardSymbols() && 89 finalizeTargetSymbols()); 90 } 91 92 /// finalizeStandardSymbols - set the value of standard symbols 93 virtual bool finalizeStandardSymbols(); 94 95 /// finalizeTargetSymbols - set the value of target symbols 96 virtual bool finalizeTargetSymbols() = 0; 97 98 /// finalizeTLSSymbol - set the value of a TLS symbol 99 virtual bool finalizeTLSSymbol(LDSymbol& pSymbol); 100 101 size_t sectionStartOffset() const; 102 getInfo()103 const GNUInfo& getInfo() const { return *m_pInfo; } getInfo()104 GNUInfo& getInfo() { return *m_pInfo; } 105 hasTextRel()106 bool hasTextRel() const { return m_bHasTextRel; } 107 hasStaticTLS()108 bool hasStaticTLS() const { return m_bHasStaticTLS; } 109 110 /// segmentStartAddr - this function returns the start address of the segment 111 uint64_t segmentStartAddr() const; 112 113 /// partialScanRelocation - When doing partial linking, fix the relocation 114 /// offset after section merge 115 void partialScanRelocation(Relocation& pReloc, 116 Module& pModule, 117 const LDSection& pSection); 118 119 /// sizeNamePools - compute the size of regular name pools 120 /// In ELF executable files, regular name pools are .symtab, .strtab., 121 /// .dynsym, .dynstr, and .hash 122 virtual void sizeNamePools(Module& pModule, bool pIsStaticLink); 123 124 /// emitSectionData - emit target-dependent section data 125 virtual uint64_t emitSectionData(const LDSection& pSection, 126 MemoryRegion& pRegion) const = 0; 127 128 /// emitRegNamePools - emit regular name pools - .symtab, .strtab 129 virtual void emitRegNamePools(const Module& pModule, MemoryArea& pOutput); 130 131 /// emitNamePools - emit dynamic name pools - .dyntab, .dynstr, .hash 132 virtual void emitDynNamePools(Module& pModule, MemoryArea& pOutput); 133 134 /// emitELFHashTab - emit .hash 135 virtual void emitELFHashTab(const Module::SymbolTable& pSymtab, 136 MemoryArea& pOutput); 137 138 /// emitGNUHashTab - emit .gnu.hash 139 virtual void emitGNUHashTab(Module::SymbolTable& pSymtab, 140 MemoryArea& pOutput); 141 142 /// sizeInterp - compute the size of program interpreter's name 143 /// In ELF executables, this is the length of dynamic linker's path name 144 virtual void sizeInterp(); 145 146 /// emitInterp - emit the .interp 147 virtual void emitInterp(MemoryArea& pOutput); 148 149 /// getSectionOrder - compute the layout order of the section 150 /// Layout calls this function to get the default order of the pSectHdr. 151 /// If the pSectHdr.type() is LDFileFormat::Target, then getSectionOrder() 152 /// will call getTargetSectionOrder(). 153 /// 154 /// If targets favors certain order for general sections, please override 155 /// this function. 156 /// 157 /// @see getTargetSectionOrder 158 virtual unsigned int getSectionOrder(const LDSection& pSectHdr) const; 159 160 /// getTargetSectionOrder - compute the layout order of target section 161 /// If the target favors certain order for the given gSectHdr, please 162 /// override this function. 163 /// 164 /// By default, this function returns the maximun order, and pSectHdr 165 /// will be the last section to be laid out. getTargetSectionOrder(const LDSection & pSectHdr)166 virtual unsigned int getTargetSectionOrder(const LDSection& pSectHdr) const 167 { return (unsigned int)-1; } 168 169 /// numOfSegments - return the number of segments 170 /// if the target favors other ways to emit program header, please override 171 /// this function numOfSegments()172 size_t numOfSegments() const { return m_ELFSegmentTable.size(); } 173 174 /// elfSegmentTable - return the reference of the elf segment table elfSegmentTable()175 ELFSegmentFactory& elfSegmentTable() { return m_ELFSegmentTable; } 176 177 /// elfSegmentTable - return the reference of the elf segment table elfSegmentTable()178 const ELFSegmentFactory& elfSegmentTable() const { return m_ELFSegmentTable; } 179 180 /// commonPageSize - the common page size of the target machine 181 uint64_t commonPageSize() const; 182 183 /// abiPageSize - the abi page size of the target machine 184 uint64_t abiPageSize() const; 185 186 /// getSymbolIdx - get the symbol index of ouput symbol table 187 size_t getSymbolIdx(const LDSymbol* pSymbol) const; 188 189 /// allocateCommonSymbols - allocate common symbols in the corresponding 190 /// sections. 191 /// Different concrete target backend may overlap this function. 192 virtual bool allocateCommonSymbols(Module& pModule); 193 194 /// updateSectionFlags - update pTo's flags when merging pFrom 195 /// update the output section flags based on input section flags. 196 virtual bool updateSectionFlags(LDSection& pTo, const LDSection& pFrom); 197 198 /// isSymbolPreemtible - whether the symbol can be preemted by other 199 /// link unit 200 /// @ref Google gold linker, symtab.h:551 201 bool isSymbolPreemptible(const ResolveInfo& pSym) const; 202 getSymDesc(uint16_t pShndx)203 virtual ResolveInfo::Desc getSymDesc(uint16_t pShndx) const { 204 return ResolveInfo::Define; 205 } 206 207 /// symbolNeedsDynRel - return whether the symbol needs a dynamic relocation 208 /// @ref Google gold linker, symtab.h:645 209 bool symbolNeedsDynRel(const ResolveInfo& pSym, 210 bool pSymHasPLT, 211 bool isAbsReloc) const; 212 213 // getTDATASymbol - get section symbol of .tdata 214 LDSymbol& getTDATASymbol(); 215 const LDSymbol& getTDATASymbol() const; 216 217 /// getTBSSSymbol - get section symbol of .tbss 218 LDSymbol& getTBSSSymbol(); 219 const LDSymbol& getTBSSSymbol() const; 220 221 // ----- relaxation ----- // 222 /// initBRIslandFactory - initialize the branch island factory for relaxation 223 bool initBRIslandFactory(); 224 225 /// initStubFactory - initialize the stub factory for relaxation 226 bool initStubFactory(); 227 228 /// getBRIslandFactory getBRIslandFactory()229 BranchIslandFactory* getBRIslandFactory() { return m_pBRIslandFactory; } 230 231 /// getStubFactory getStubFactory()232 StubFactory* getStubFactory() { return m_pStubFactory; } 233 234 /// maxBranchOffset - return the max (forward) branch offset of the backend. 235 /// Target can override this function if needed. maxBranchOffset()236 virtual uint64_t maxBranchOffset() { return (uint64_t)-1; } 237 238 protected: 239 uint64_t getSymbolSize(const LDSymbol& pSymbol) const; 240 241 uint64_t getSymbolInfo(const LDSymbol& pSymbol) const; 242 243 uint64_t getSymbolValue(const LDSymbol& pSymbol) const; 244 245 uint64_t getSymbolShndx(const LDSymbol& pSymbol) const; 246 247 /// isTemporary - Whether pSymbol is a local label. 248 virtual bool isTemporary(const LDSymbol& pSymbol) const; 249 250 /// getHashBucketCount - calculate hash bucket count. 251 /// @ref Google gold linker, dynobj.cc:791 252 static unsigned getHashBucketCount(unsigned pNumOfSymbols, bool pIsGNUStyle); 253 254 /// getGNUHashMaskbitslog2 - calculate the number of mask bits in log2 255 /// @ref binutils gold, dynobj.cc:1165 256 unsigned getGNUHashMaskbitslog2(unsigned pNumOfSymbols) const; 257 258 /// isDynamicSymbol 259 /// @ref Google gold linker: symtab.cc:311 260 bool isDynamicSymbol(const LDSymbol& pSymbol); 261 262 /// isDynamicSymbol 263 /// @ref Google gold linker: symtab.cc:311 264 bool isDynamicSymbol(const ResolveInfo& pResolveInfo); 265 266 /// symbolNeedsPLT - return whether the symbol needs a PLT entry 267 /// @ref Google gold linker, symtab.h:596 268 bool symbolNeedsPLT(const ResolveInfo& pSym) const; 269 270 /// symbolNeedsCopyReloc - return whether the symbol needs a copy relocation 271 bool symbolNeedsCopyReloc(const Relocation& pReloc, 272 const ResolveInfo& pSym) const; 273 274 /// symbolHasFinalValue - return true if the symbol's value can be decided at 275 /// link time 276 bool symbolFinalValueIsKnown(const ResolveInfo& pSym) const; 277 278 /// emitSymbol32 - emit an ELF32 symbol 279 void emitSymbol32(llvm::ELF::Elf32_Sym& pSym32, 280 LDSymbol& pSymbol, 281 char* pStrtab, 282 size_t pStrtabsize, 283 size_t pSymtabIdx); 284 285 /// emitSymbol64 - emit an ELF64 symbol 286 void emitSymbol64(llvm::ELF::Elf64_Sym& pSym64, 287 LDSymbol& pSymbol, 288 char* pStrtab, 289 size_t pStrtabsize, 290 size_t pSymtabIdx); 291 292 /// checkAndSetHasTextRel - check pSection flag to set HasTextRel 293 void checkAndSetHasTextRel(const LDSection& pSection); 294 295 void setHasStaticTLS(bool pVal = true) { m_bHasStaticTLS = pVal; } 296 297 private: 298 /// createProgramHdrs - base on output sections to create the program headers 299 void createProgramHdrs(Module& pModule); 300 301 /// doCreateProgramHdrs - backend can implement this function to create the 302 /// target-dependent segments 303 virtual void doCreateProgramHdrs(Module& pModule) = 0; 304 305 /// setupProgramHdrs - set up the attributes of segments 306 /// (i.e., offset, addresses, file/mem size, flag, and alignment) 307 void setupProgramHdrs(); 308 309 /// getSegmentFlag - give a section flag and return the corresponding segment 310 /// flag getSegmentFlag(const uint32_t pSectionFlag)311 inline uint32_t getSegmentFlag(const uint32_t pSectionFlag) 312 { 313 uint32_t flag = llvm::ELF::PF_R; 314 if (0 != (pSectionFlag & llvm::ELF::SHF_WRITE)) 315 flag |= llvm::ELF::PF_W; 316 if (0 != (pSectionFlag & llvm::ELF::SHF_EXECINSTR)) 317 flag |= llvm::ELF::PF_X; 318 return flag; 319 } 320 321 /// setupGNUStackInfo - setup the section flag of .note.GNU-stack in output 322 void setupGNUStackInfo(Module& pModule); 323 324 /// setupRelro - setup the offset constraint of PT_RELRO 325 void setupRelro(Module& pModule); 326 327 /// setOutputSectionOffset - helper function to set a group of output sections' 328 /// offset, and set pSectBegin to pStartOffset if pStartOffset is not -1U. 329 void setOutputSectionOffset(Module& pModule, 330 Module::iterator pSectBegin, 331 Module::iterator pSectEnd, 332 uint64_t pStartOffset = -1U); 333 334 /// setOutputSectionOffset - helper function to set output sections' address. 335 void setOutputSectionAddress(Module& pModule, 336 Module::iterator pSectBegin, 337 Module::iterator pSectEnd); 338 339 /// layout - layout method 340 void layout(Module& pModule); 341 342 /// preLayout - Backend can do any needed modification before layout 343 void preLayout(Module& pModule, IRBuilder& pBuilder); 344 345 /// postLayout -Backend can do any needed modification after layout 346 void postLayout(Module& pModule, IRBuilder& pBuilder); 347 348 /// preLayout - Backend can do any needed modification before layout 349 virtual void doPreLayout(IRBuilder& pBuilder) = 0; 350 351 /// postLayout -Backend can do any needed modification after layout 352 virtual void doPostLayout(Module& pModule, IRBuilder& pLinker) = 0; 353 354 /// postProcessing - Backend can do any needed modification in the final stage 355 void postProcessing(MemoryArea& pOutput); 356 357 /// dynamic - the dynamic section of the target machine. 358 virtual ELFDynamic& dynamic() = 0; 359 360 /// dynamic - the dynamic section of the target machine. 361 virtual const ELFDynamic& dynamic() const = 0; 362 363 /// relax - the relaxation pass 364 bool relax(Module& pModule, IRBuilder& pBuilder); 365 366 /// mayRelax - Backends should override this function if they need relaxation mayRelax()367 virtual bool mayRelax() { return false; } 368 369 /// doRelax - Backend can orevride this function to add its relaxation 370 /// implementation. Return true if the output (e.g., .text) is "relaxed" 371 /// (i.e. layout is changed), and set pFinished to true if everything is fit, 372 /// otherwise set it to false. doRelax(Module & pModule,IRBuilder & pBuilder,bool & pFinished)373 virtual bool doRelax(Module& pModule, IRBuilder& pBuilder, bool& pFinished) 374 { return false; } 375 376 /// getRelEntrySize - the size in BYTE of rel type relocation 377 virtual size_t getRelEntrySize() = 0; 378 379 /// getRelEntrySize - the size in BYTE of rela type relocation 380 virtual size_t getRelaEntrySize() = 0; 381 382 protected: 383 // Based on Kind in LDFileFormat to define basic section orders for ELF, and 384 // refer gold linker to add more enumerations to handle Regular and BSS kind 385 enum SectionOrder { 386 SHO_INTERP = 1, // .interp 387 SHO_RO_NOTE, // .note.ABI-tag, .note.gnu.build-id 388 SHO_NAMEPOOL, // *.hash, .dynsym, .dynstr 389 SHO_RELOCATION, // .rel.*, .rela.* 390 SHO_REL_PLT, // .rel.plt should come after other .rel.* 391 SHO_INIT, // .init 392 SHO_PLT, // .plt 393 SHO_TEXT, // .text 394 SHO_FINI, // .fini 395 SHO_RO, // .rodata 396 SHO_EXCEPTION, // .eh_frame_hdr, .eh_frame, .gcc_except_table 397 SHO_TLS_DATA, // .tdata 398 SHO_TLS_BSS, // .tbss 399 SHO_RELRO_LOCAL, // .data.rel.ro.local 400 SHO_RELRO, // .data.rel.ro, 401 SHO_RELRO_LAST, // for x86 to adjust .got if needed 402 SHO_NON_RELRO_FIRST, // for x86 to adjust .got.plt if needed 403 SHO_DATA, // .data 404 SHO_LARGE_DATA, // .ldata 405 SHO_RW_NOTE, // 406 SHO_SMALL_DATA, // .sdata 407 SHO_SMALL_BSS, // .sbss 408 SHO_BSS, // .bss 409 SHO_LARGE_BSS, // .lbss 410 SHO_UNDEFINED, // default order 411 SHO_STRTAB // .strtab 412 }; 413 414 typedef std::pair<LDSection*, unsigned int> SHOEntry; 415 416 struct SHOCompare 417 { operatorSHOCompare418 bool operator()(const SHOEntry& X, const SHOEntry& Y) const 419 { return X.second < Y.second; } 420 }; 421 422 struct SymCompare 423 { operatorSymCompare424 bool operator()(const LDSymbol* X, const LDSymbol* Y) const 425 { return (X==Y); } 426 }; 427 428 // for gnu style hash table 429 struct DynsymCompare 430 { 431 bool needGNUHash(const LDSymbol& X) const; 432 433 bool operator()(const LDSymbol* X, const LDSymbol* Y) const; 434 }; 435 436 struct SymPtrHash 437 { operatorSymPtrHash438 size_t operator()(const LDSymbol* pKey) const 439 { 440 return (unsigned((uintptr_t)pKey) >> 4) ^ 441 (unsigned((uintptr_t)pKey) >> 9); 442 } 443 }; 444 445 typedef HashEntry<LDSymbol*, size_t, SymCompare> SymHashEntryType; 446 typedef HashTable<SymHashEntryType, 447 SymPtrHash, 448 EntryFactory<SymHashEntryType> > HashTableType; 449 450 451 protected: 452 ELFObjectReader* m_pObjectReader; 453 454 // ----- file formats ----- // 455 ELFDynObjFileFormat* m_pDynObjFileFormat; 456 ELFExecFileFormat* m_pExecFileFormat; 457 ELFObjectFileFormat* m_pObjectFileFormat; 458 459 // GNUInfo 460 GNUInfo* m_pInfo; 461 462 // ELF segment factory 463 ELFSegmentFactory m_ELFSegmentTable; 464 465 // branch island factory 466 BranchIslandFactory* m_pBRIslandFactory; 467 468 // stub factory 469 StubFactory* m_pStubFactory; 470 471 // map the LDSymbol to its index in the output symbol table 472 HashTableType* m_pSymIndexMap; 473 474 // section .eh_frame_hdr 475 EhFrameHdr* m_pEhFrameHdr; 476 477 // ----- dynamic flags ----- // 478 // DF_TEXTREL of DT_FLAGS 479 bool m_bHasTextRel; 480 481 // DF_STATIC_TLS of DT_FLAGS 482 bool m_bHasStaticTLS; 483 484 // ----- standard symbols ----- // 485 // section symbols 486 LDSymbol* f_pPreInitArrayStart; 487 LDSymbol* f_pPreInitArrayEnd; 488 LDSymbol* f_pInitArrayStart; 489 LDSymbol* f_pInitArrayEnd; 490 LDSymbol* f_pFiniArrayStart; 491 LDSymbol* f_pFiniArrayEnd; 492 LDSymbol* f_pStack; 493 LDSymbol* f_pDynamic; 494 495 // section symbols for .tdata and .tbss 496 LDSymbol* f_pTDATA; 497 LDSymbol* f_pTBSS; 498 499 // segment symbols 500 LDSymbol* f_pExecutableStart; 501 LDSymbol* f_pEText; 502 LDSymbol* f_p_EText; 503 LDSymbol* f_p__EText; 504 LDSymbol* f_pEData; 505 LDSymbol* f_p_EData; 506 LDSymbol* f_pBSSStart; 507 LDSymbol* f_pEnd; 508 LDSymbol* f_p_End; 509 }; 510 511 } // namespace of mcld 512 513 #endif 514 515