1 //===- IRSymtab.h - data definitions for IR symbol tables -------*- C++ -*-===// 2 // 3 // The LLVM Compiler Infrastructure 4 // 5 // This file is distributed under the University of Illinois Open Source 6 // License. See LICENSE.TXT for details. 7 // 8 //===----------------------------------------------------------------------===// 9 // 10 // This file contains data definitions and a reader and builder for a symbol 11 // table for LLVM IR. Its purpose is to allow linkers and other consumers of 12 // bitcode files to efficiently read the symbol table for symbol resolution 13 // purposes without needing to construct a module in memory. 14 // 15 // As with most object files the symbol table has two parts: the symbol table 16 // itself and a string table which is referenced by the symbol table. 17 // 18 // A symbol table corresponds to a single bitcode file, which may consist of 19 // multiple modules, so symbol tables may likewise contain symbols for multiple 20 // modules. 21 // 22 //===----------------------------------------------------------------------===// 23 24 #ifndef LLVM_OBJECT_IRSYMTAB_H 25 #define LLVM_OBJECT_IRSYMTAB_H 26 27 #include "llvm/ADT/ArrayRef.h" 28 #include "llvm/ADT/StringRef.h" 29 #include "llvm/ADT/iterator_range.h" 30 #include "llvm/IR/GlobalValue.h" 31 #include "llvm/Object/SymbolicFile.h" 32 #include "llvm/Support/Endian.h" 33 #include "llvm/Support/Error.h" 34 #include <cassert> 35 #include <cstdint> 36 #include <vector> 37 38 namespace llvm { 39 40 struct BitcodeFileContents; 41 class StringTableBuilder; 42 43 namespace irsymtab { 44 45 namespace storage { 46 47 // The data structures in this namespace define the low-level serialization 48 // format. Clients that just want to read a symbol table should use the 49 // irsymtab::Reader class. 50 51 using Word = support::ulittle32_t; 52 53 /// A reference to a string in the string table. 54 struct Str { 55 Word Offset, Size; 56 57 StringRef get(StringRef Strtab) const { 58 return {Strtab.data() + Offset, Size}; 59 } 60 }; 61 62 /// A reference to a range of objects in the symbol table. 63 template <typename T> struct Range { 64 Word Offset, Size; 65 66 ArrayRef<T> get(StringRef Symtab) const { 67 return {reinterpret_cast<const T *>(Symtab.data() + Offset), Size}; 68 } 69 }; 70 71 /// Describes the range of a particular module's symbols within the symbol 72 /// table. 73 struct Module { 74 Word Begin, End; 75 76 /// The index of the first Uncommon for this Module. 77 Word UncBegin; 78 }; 79 80 /// This is equivalent to an IR comdat. 81 struct Comdat { 82 Str Name; 83 }; 84 85 /// Contains the information needed by linkers for symbol resolution, as well as 86 /// by the LTO implementation itself. 87 struct Symbol { 88 /// The mangled symbol name. 89 Str Name; 90 91 /// The unmangled symbol name, or the empty string if this is not an IR 92 /// symbol. 93 Str IRName; 94 95 /// The index into Header::Comdats, or -1 if not a comdat member. 96 Word ComdatIndex; 97 98 Word Flags; 99 enum FlagBits { 100 FB_visibility, // 2 bits 101 FB_has_uncommon = FB_visibility + 2, 102 FB_undefined, 103 FB_weak, 104 FB_common, 105 FB_indirect, 106 FB_used, 107 FB_tls, 108 FB_may_omit, 109 FB_global, 110 FB_format_specific, 111 FB_unnamed_addr, 112 FB_executable, 113 }; 114 }; 115 116 /// This data structure contains rarely used symbol fields and is optionally 117 /// referenced by a Symbol. 118 struct Uncommon { 119 Word CommonSize, CommonAlign; 120 121 /// COFF-specific: the name of the symbol that a weak external resolves to 122 /// if not defined. 123 Str COFFWeakExternFallbackName; 124 125 /// Specified section name, if any. 126 Str SectionName; 127 }; 128 129 struct Header { 130 /// Version number of the symtab format. This number should be incremented 131 /// when the format changes, but it does not need to be incremented if a 132 /// change to LLVM would cause it to create a different symbol table. 133 Word Version; 134 enum { kCurrentVersion = 1 }; 135 136 /// The producer's version string (LLVM_VERSION_STRING " " LLVM_REVISION). 137 /// Consumers should rebuild the symbol table from IR if the producer's 138 /// version does not match the consumer's version due to potential differences 139 /// in symbol table format, symbol enumeration order and so on. 140 Str Producer; 141 142 Range<Module> Modules; 143 Range<Comdat> Comdats; 144 Range<Symbol> Symbols; 145 Range<Uncommon> Uncommons; 146 147 Str TargetTriple, SourceFileName; 148 149 /// COFF-specific: linker directives. 150 Str COFFLinkerOpts; 151 }; 152 153 } // end namespace storage 154 155 /// Fills in Symtab and StrtabBuilder with a valid symbol and string table for 156 /// Mods. 157 Error build(ArrayRef<Module *> Mods, SmallVector<char, 0> &Symtab, 158 StringTableBuilder &StrtabBuilder, BumpPtrAllocator &Alloc); 159 160 /// This represents a symbol that has been read from a storage::Symbol and 161 /// possibly a storage::Uncommon. 162 struct Symbol { 163 // Copied from storage::Symbol. 164 StringRef Name, IRName; 165 int ComdatIndex; 166 uint32_t Flags; 167 168 // Copied from storage::Uncommon. 169 uint32_t CommonSize, CommonAlign; 170 StringRef COFFWeakExternFallbackName; 171 StringRef SectionName; 172 173 /// Returns the mangled symbol name. 174 StringRef getName() const { return Name; } 175 176 /// Returns the unmangled symbol name, or the empty string if this is not an 177 /// IR symbol. 178 StringRef getIRName() const { return IRName; } 179 180 /// Returns the index into the comdat table (see Reader::getComdatTable()), or 181 /// -1 if not a comdat member. 182 int getComdatIndex() const { return ComdatIndex; } 183 184 using S = storage::Symbol; 185 186 GlobalValue::VisibilityTypes getVisibility() const { 187 return GlobalValue::VisibilityTypes((Flags >> S::FB_visibility) & 3); 188 } 189 190 bool isUndefined() const { return (Flags >> S::FB_undefined) & 1; } 191 bool isWeak() const { return (Flags >> S::FB_weak) & 1; } 192 bool isCommon() const { return (Flags >> S::FB_common) & 1; } 193 bool isIndirect() const { return (Flags >> S::FB_indirect) & 1; } 194 bool isUsed() const { return (Flags >> S::FB_used) & 1; } 195 bool isTLS() const { return (Flags >> S::FB_tls) & 1; } 196 197 bool canBeOmittedFromSymbolTable() const { 198 return (Flags >> S::FB_may_omit) & 1; 199 } 200 201 bool isGlobal() const { return (Flags >> S::FB_global) & 1; } 202 bool isFormatSpecific() const { return (Flags >> S::FB_format_specific) & 1; } 203 bool isUnnamedAddr() const { return (Flags >> S::FB_unnamed_addr) & 1; } 204 bool isExecutable() const { return (Flags >> S::FB_executable) & 1; } 205 206 uint64_t getCommonSize() const { 207 assert(isCommon()); 208 return CommonSize; 209 } 210 211 uint32_t getCommonAlignment() const { 212 assert(isCommon()); 213 return CommonAlign; 214 } 215 216 /// COFF-specific: for weak externals, returns the name of the symbol that is 217 /// used as a fallback if the weak external remains undefined. 218 StringRef getCOFFWeakExternalFallback() const { 219 assert(isWeak() && isIndirect()); 220 return COFFWeakExternFallbackName; 221 } 222 223 StringRef getSectionName() const { return SectionName; } 224 }; 225 226 /// This class can be used to read a Symtab and Strtab produced by 227 /// irsymtab::build. 228 class Reader { 229 StringRef Symtab, Strtab; 230 231 ArrayRef<storage::Module> Modules; 232 ArrayRef<storage::Comdat> Comdats; 233 ArrayRef<storage::Symbol> Symbols; 234 ArrayRef<storage::Uncommon> Uncommons; 235 236 StringRef str(storage::Str S) const { return S.get(Strtab); } 237 238 template <typename T> ArrayRef<T> range(storage::Range<T> R) const { 239 return R.get(Symtab); 240 } 241 242 const storage::Header &header() const { 243 return *reinterpret_cast<const storage::Header *>(Symtab.data()); 244 } 245 246 public: 247 class SymbolRef; 248 249 Reader() = default; 250 Reader(StringRef Symtab, StringRef Strtab) : Symtab(Symtab), Strtab(Strtab) { 251 Modules = range(header().Modules); 252 Comdats = range(header().Comdats); 253 Symbols = range(header().Symbols); 254 Uncommons = range(header().Uncommons); 255 } 256 257 using symbol_range = iterator_range<object::content_iterator<SymbolRef>>; 258 259 /// Returns the symbol table for the entire bitcode file. 260 /// The symbols enumerated by this method are ephemeral, but they can be 261 /// copied into an irsymtab::Symbol object. 262 symbol_range symbols() const; 263 264 size_t getNumModules() const { return Modules.size(); } 265 266 /// Returns a slice of the symbol table for the I'th module in the file. 267 /// The symbols enumerated by this method are ephemeral, but they can be 268 /// copied into an irsymtab::Symbol object. 269 symbol_range module_symbols(unsigned I) const; 270 271 StringRef getTargetTriple() const { return str(header().TargetTriple); } 272 273 /// Returns the source file path specified at compile time. 274 StringRef getSourceFileName() const { return str(header().SourceFileName); } 275 276 /// Returns a table with all the comdats used by this file. 277 std::vector<StringRef> getComdatTable() const { 278 std::vector<StringRef> ComdatTable; 279 ComdatTable.reserve(Comdats.size()); 280 for (auto C : Comdats) 281 ComdatTable.push_back(str(C.Name)); 282 return ComdatTable; 283 } 284 285 /// COFF-specific: returns linker options specified in the input file. 286 StringRef getCOFFLinkerOpts() const { return str(header().COFFLinkerOpts); } 287 }; 288 289 /// Ephemeral symbols produced by Reader::symbols() and 290 /// Reader::module_symbols(). 291 class Reader::SymbolRef : public Symbol { 292 const storage::Symbol *SymI, *SymE; 293 const storage::Uncommon *UncI; 294 const Reader *R; 295 296 void read() { 297 if (SymI == SymE) 298 return; 299 300 Name = R->str(SymI->Name); 301 IRName = R->str(SymI->IRName); 302 ComdatIndex = SymI->ComdatIndex; 303 Flags = SymI->Flags; 304 305 if (Flags & (1 << storage::Symbol::FB_has_uncommon)) { 306 CommonSize = UncI->CommonSize; 307 CommonAlign = UncI->CommonAlign; 308 COFFWeakExternFallbackName = R->str(UncI->COFFWeakExternFallbackName); 309 SectionName = R->str(UncI->SectionName); 310 } else 311 // Reset this field so it can be queried unconditionally for all symbols. 312 SectionName = ""; 313 } 314 315 public: 316 SymbolRef(const storage::Symbol *SymI, const storage::Symbol *SymE, 317 const storage::Uncommon *UncI, const Reader *R) 318 : SymI(SymI), SymE(SymE), UncI(UncI), R(R) { 319 read(); 320 } 321 322 void moveNext() { 323 ++SymI; 324 if (Flags & (1 << storage::Symbol::FB_has_uncommon)) 325 ++UncI; 326 read(); 327 } 328 329 bool operator==(const SymbolRef &Other) const { return SymI == Other.SymI; } 330 }; 331 332 inline Reader::symbol_range Reader::symbols() const { 333 return {SymbolRef(Symbols.begin(), Symbols.end(), Uncommons.begin(), this), 334 SymbolRef(Symbols.end(), Symbols.end(), nullptr, this)}; 335 } 336 337 inline Reader::symbol_range Reader::module_symbols(unsigned I) const { 338 const storage::Module &M = Modules[I]; 339 const storage::Symbol *MBegin = Symbols.begin() + M.Begin, 340 *MEnd = Symbols.begin() + M.End; 341 return {SymbolRef(MBegin, MEnd, Uncommons.begin() + M.UncBegin, this), 342 SymbolRef(MEnd, MEnd, nullptr, this)}; 343 } 344 345 /// The contents of the irsymtab in a bitcode file. Any underlying data for the 346 /// irsymtab are owned by Symtab and Strtab. 347 struct FileContents { 348 SmallVector<char, 0> Symtab, Strtab; 349 Reader TheReader; 350 }; 351 352 /// Reads the contents of a bitcode file, creating its irsymtab if necessary. 353 Expected<FileContents> readBitcode(const BitcodeFileContents &BFC); 354 355 } // end namespace irsymtab 356 } // end namespace llvm 357 358 #endif // LLVM_OBJECT_IRSYMTAB_H 359