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
38namespace llvm {
39
40struct BitcodeFileContents;
41
42namespace irsymtab {
43
44namespace storage {
45
46// The data structures in this namespace define the low-level serialization
47// format. Clients that just want to read a symbol table should use the
48// irsymtab::Reader class.
49
50using Word = support::ulittle32_t;
51
52/// A reference to a string in the string table.
53struct Str {
54  Word Offset, Size;
55
56  StringRef get(StringRef Strtab) const {
57    return {Strtab.data() + Offset, Size};
58  }
59};
60
61/// A reference to a range of objects in the symbol table.
62template <typename T> struct Range {
63  Word Offset, Size;
64
65  ArrayRef<T> get(StringRef Symtab) const {
66    return {reinterpret_cast<const T *>(Symtab.data() + Offset), Size};
67  }
68};
69
70/// Describes the range of a particular module's symbols within the symbol
71/// table.
72struct Module {
73  Word Begin, End;
74
75  /// The index of the first Uncommon for this Module.
76  Word UncBegin;
77};
78
79/// This is equivalent to an IR comdat.
80struct Comdat {
81  Str Name;
82};
83
84/// Contains the information needed by linkers for symbol resolution, as well as
85/// by the LTO implementation itself.
86struct Symbol {
87  /// The mangled symbol name.
88  Str Name;
89
90  /// The unmangled symbol name, or the empty string if this is not an IR
91  /// symbol.
92  Str IRName;
93
94  /// The index into Header::Comdats, or -1 if not a comdat member.
95  Word ComdatIndex;
96
97  Word Flags;
98  enum FlagBits {
99    FB_visibility, // 2 bits
100    FB_has_uncommon = FB_visibility + 2,
101    FB_undefined,
102    FB_weak,
103    FB_common,
104    FB_indirect,
105    FB_used,
106    FB_tls,
107    FB_may_omit,
108    FB_global,
109    FB_format_specific,
110    FB_unnamed_addr,
111    FB_executable,
112  };
113};
114
115/// This data structure contains rarely used symbol fields and is optionally
116/// referenced by a Symbol.
117struct Uncommon {
118  Word CommonSize, CommonAlign;
119
120  /// COFF-specific: the name of the symbol that a weak external resolves to
121  /// if not defined.
122  Str COFFWeakExternFallbackName;
123};
124
125struct Header {
126  Range<Module> Modules;
127  Range<Comdat> Comdats;
128  Range<Symbol> Symbols;
129  Range<Uncommon> Uncommons;
130
131  Str TargetTriple, SourceFileName;
132
133  /// COFF-specific: linker directives.
134  Str COFFLinkerOpts;
135};
136
137} // end namespace storage
138
139/// Fills in Symtab and Strtab with a valid symbol and string table for Mods.
140Error build(ArrayRef<Module *> Mods, SmallVector<char, 0> &Symtab,
141            SmallVector<char, 0> &Strtab);
142
143/// This represents a symbol that has been read from a storage::Symbol and
144/// possibly a storage::Uncommon.
145struct Symbol {
146  // Copied from storage::Symbol.
147  StringRef Name, IRName;
148  int ComdatIndex;
149  uint32_t Flags;
150
151  // Copied from storage::Uncommon.
152  uint32_t CommonSize, CommonAlign;
153  StringRef COFFWeakExternFallbackName;
154
155  /// Returns the mangled symbol name.
156  StringRef getName() const { return Name; }
157
158  /// Returns the unmangled symbol name, or the empty string if this is not an
159  /// IR symbol.
160  StringRef getIRName() const { return IRName; }
161
162  /// Returns the index into the comdat table (see Reader::getComdatTable()), or
163  /// -1 if not a comdat member.
164  int getComdatIndex() const { return ComdatIndex; }
165
166  using S = storage::Symbol;
167
168  GlobalValue::VisibilityTypes getVisibility() const {
169    return GlobalValue::VisibilityTypes((Flags >> S::FB_visibility) & 3);
170  }
171
172  bool isUndefined() const { return (Flags >> S::FB_undefined) & 1; }
173  bool isWeak() const { return (Flags >> S::FB_weak) & 1; }
174  bool isCommon() const { return (Flags >> S::FB_common) & 1; }
175  bool isIndirect() const { return (Flags >> S::FB_indirect) & 1; }
176  bool isUsed() const { return (Flags >> S::FB_used) & 1; }
177  bool isTLS() const { return (Flags >> S::FB_tls) & 1; }
178
179  bool canBeOmittedFromSymbolTable() const {
180    return (Flags >> S::FB_may_omit) & 1;
181  }
182
183  bool isGlobal() const { return (Flags >> S::FB_global) & 1; }
184  bool isFormatSpecific() const { return (Flags >> S::FB_format_specific) & 1; }
185  bool isUnnamedAddr() const { return (Flags >> S::FB_unnamed_addr) & 1; }
186  bool isExecutable() const { return (Flags >> S::FB_executable) & 1; }
187
188  uint64_t getCommonSize() const {
189    assert(isCommon());
190    return CommonSize;
191  }
192
193  uint32_t getCommonAlignment() const {
194    assert(isCommon());
195    return CommonAlign;
196  }
197
198  /// COFF-specific: for weak externals, returns the name of the symbol that is
199  /// used as a fallback if the weak external remains undefined.
200  StringRef getCOFFWeakExternalFallback() const {
201    assert(isWeak() && isIndirect());
202    return COFFWeakExternFallbackName;
203  }
204};
205
206/// This class can be used to read a Symtab and Strtab produced by
207/// irsymtab::build.
208class Reader {
209  StringRef Symtab, Strtab;
210
211  ArrayRef<storage::Module> Modules;
212  ArrayRef<storage::Comdat> Comdats;
213  ArrayRef<storage::Symbol> Symbols;
214  ArrayRef<storage::Uncommon> Uncommons;
215
216  StringRef str(storage::Str S) const { return S.get(Strtab); }
217
218  template <typename T> ArrayRef<T> range(storage::Range<T> R) const {
219    return R.get(Symtab);
220  }
221
222  const storage::Header &header() const {
223    return *reinterpret_cast<const storage::Header *>(Symtab.data());
224  }
225
226public:
227  class SymbolRef;
228
229  Reader() = default;
230  Reader(StringRef Symtab, StringRef Strtab) : Symtab(Symtab), Strtab(Strtab) {
231    Modules = range(header().Modules);
232    Comdats = range(header().Comdats);
233    Symbols = range(header().Symbols);
234    Uncommons = range(header().Uncommons);
235  }
236
237  using symbol_range = iterator_range<object::content_iterator<SymbolRef>>;
238
239  /// Returns the symbol table for the entire bitcode file.
240  /// The symbols enumerated by this method are ephemeral, but they can be
241  /// copied into an irsymtab::Symbol object.
242  symbol_range symbols() const;
243
244  /// Returns a slice of the symbol table for the I'th module in the file.
245  /// The symbols enumerated by this method are ephemeral, but they can be
246  /// copied into an irsymtab::Symbol object.
247  symbol_range module_symbols(unsigned I) const;
248
249  StringRef getTargetTriple() const { return str(header().TargetTriple); }
250
251  /// Returns the source file path specified at compile time.
252  StringRef getSourceFileName() const { return str(header().SourceFileName); }
253
254  /// Returns a table with all the comdats used by this file.
255  std::vector<StringRef> getComdatTable() const {
256    std::vector<StringRef> ComdatTable;
257    ComdatTable.reserve(Comdats.size());
258    for (auto C : Comdats)
259      ComdatTable.push_back(str(C.Name));
260    return ComdatTable;
261  }
262
263  /// COFF-specific: returns linker options specified in the input file.
264  StringRef getCOFFLinkerOpts() const { return str(header().COFFLinkerOpts); }
265};
266
267/// Ephemeral symbols produced by Reader::symbols() and
268/// Reader::module_symbols().
269class Reader::SymbolRef : public Symbol {
270  const storage::Symbol *SymI, *SymE;
271  const storage::Uncommon *UncI;
272  const Reader *R;
273
274  void read() {
275    if (SymI == SymE)
276      return;
277
278    Name = R->str(SymI->Name);
279    IRName = R->str(SymI->IRName);
280    ComdatIndex = SymI->ComdatIndex;
281    Flags = SymI->Flags;
282
283    if (Flags & (1 << storage::Symbol::FB_has_uncommon)) {
284      CommonSize = UncI->CommonSize;
285      CommonAlign = UncI->CommonAlign;
286      COFFWeakExternFallbackName = R->str(UncI->COFFWeakExternFallbackName);
287    }
288  }
289
290public:
291  SymbolRef(const storage::Symbol *SymI, const storage::Symbol *SymE,
292            const storage::Uncommon *UncI, const Reader *R)
293      : SymI(SymI), SymE(SymE), UncI(UncI), R(R) {
294    read();
295  }
296
297  void moveNext() {
298    ++SymI;
299    if (Flags & (1 << storage::Symbol::FB_has_uncommon))
300      ++UncI;
301    read();
302  }
303
304  bool operator==(const SymbolRef &Other) const { return SymI == Other.SymI; }
305};
306
307inline Reader::symbol_range Reader::symbols() const {
308  return {SymbolRef(Symbols.begin(), Symbols.end(), Uncommons.begin(), this),
309          SymbolRef(Symbols.end(), Symbols.end(), nullptr, this)};
310}
311
312inline Reader::symbol_range Reader::module_symbols(unsigned I) const {
313  const storage::Module &M = Modules[I];
314  const storage::Symbol *MBegin = Symbols.begin() + M.Begin,
315                        *MEnd = Symbols.begin() + M.End;
316  return {SymbolRef(MBegin, MEnd, Uncommons.begin() + M.UncBegin, this),
317          SymbolRef(MEnd, MEnd, nullptr, this)};
318}
319
320/// The contents of the irsymtab in a bitcode file. Any underlying data for the
321/// irsymtab are owned by Symtab and Strtab.
322struct FileContents {
323  SmallVector<char, 0> Symtab, Strtab;
324  Reader TheReader;
325};
326
327/// Reads the contents of a bitcode file, creating its irsymtab if necessary.
328Expected<FileContents> readBitcode(const BitcodeFileContents &BFC);
329
330} // end namespace irsymtab
331} // end namespace llvm
332
333#endif // LLVM_OBJECT_IRSYMTAB_H
334