1//===-- SymbolizableObjectFile.cpp ----------------------------------------===//
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// Implementation of SymbolizableObjectFile class.
11//
12//===----------------------------------------------------------------------===//
13
14#include "SymbolizableObjectFile.h"
15#include "llvm/Object/COFF.h"
16#include "llvm/Object/SymbolSize.h"
17#include "llvm/Support/DataExtractor.h"
18#include "llvm/DebugInfo/DWARF/DWARFContext.h"
19
20namespace llvm {
21namespace symbolize {
22
23using namespace object;
24
25static DILineInfoSpecifier
26getDILineInfoSpecifier(FunctionNameKind FNKind) {
27  return DILineInfoSpecifier(
28      DILineInfoSpecifier::FileLineInfoKind::AbsoluteFilePath, FNKind);
29}
30
31ErrorOr<std::unique_ptr<SymbolizableObjectFile>>
32SymbolizableObjectFile::create(object::ObjectFile *Obj,
33                               std::unique_ptr<DIContext> DICtx) {
34  std::unique_ptr<SymbolizableObjectFile> res(
35      new SymbolizableObjectFile(Obj, std::move(DICtx)));
36  std::unique_ptr<DataExtractor> OpdExtractor;
37  uint64_t OpdAddress = 0;
38  // Find the .opd (function descriptor) section if any, for big-endian
39  // PowerPC64 ELF.
40  if (Obj->getArch() == Triple::ppc64) {
41    for (section_iterator Section : Obj->sections()) {
42      StringRef Name;
43      StringRef Data;
44      if (auto EC = Section->getName(Name))
45        return EC;
46      if (Name == ".opd") {
47        if (auto EC = Section->getContents(Data))
48          return EC;
49        OpdExtractor.reset(new DataExtractor(Data, Obj->isLittleEndian(),
50                                             Obj->getBytesInAddress()));
51        OpdAddress = Section->getAddress();
52        break;
53      }
54    }
55  }
56  std::vector<std::pair<SymbolRef, uint64_t>> Symbols =
57      computeSymbolSizes(*Obj);
58  for (auto &P : Symbols)
59    res->addSymbol(P.first, P.second, OpdExtractor.get(), OpdAddress);
60
61  // If this is a COFF object and we didn't find any symbols, try the export
62  // table.
63  if (Symbols.empty()) {
64    if (auto *CoffObj = dyn_cast<COFFObjectFile>(Obj))
65      if (auto EC = res->addCoffExportSymbols(CoffObj))
66        return EC;
67  }
68  return std::move(res);
69}
70
71SymbolizableObjectFile::SymbolizableObjectFile(ObjectFile *Obj,
72                                               std::unique_ptr<DIContext> DICtx)
73    : Module(Obj), DebugInfoContext(std::move(DICtx)) {}
74
75namespace {
76struct OffsetNamePair {
77  uint32_t Offset;
78  StringRef Name;
79  bool operator<(const OffsetNamePair &R) const {
80    return Offset < R.Offset;
81  }
82};
83}
84
85std::error_code SymbolizableObjectFile::addCoffExportSymbols(
86    const COFFObjectFile *CoffObj) {
87  // Get all export names and offsets.
88  std::vector<OffsetNamePair> ExportSyms;
89  for (const ExportDirectoryEntryRef &Ref : CoffObj->export_directories()) {
90    StringRef Name;
91    uint32_t Offset;
92    if (auto EC = Ref.getSymbolName(Name))
93      return EC;
94    if (auto EC = Ref.getExportRVA(Offset))
95      return EC;
96    ExportSyms.push_back(OffsetNamePair{Offset, Name});
97  }
98  if (ExportSyms.empty())
99    return std::error_code();
100
101  // Sort by ascending offset.
102  array_pod_sort(ExportSyms.begin(), ExportSyms.end());
103
104  // Approximate the symbol sizes by assuming they run to the next symbol.
105  // FIXME: This assumes all exports are functions.
106  uint64_t ImageBase = CoffObj->getImageBase();
107  for (auto I = ExportSyms.begin(), E = ExportSyms.end(); I != E; ++I) {
108    OffsetNamePair &Export = *I;
109    // FIXME: The last export has a one byte size now.
110    uint32_t NextOffset = I != E ? I->Offset : Export.Offset + 1;
111    uint64_t SymbolStart = ImageBase + Export.Offset;
112    uint64_t SymbolSize = NextOffset - Export.Offset;
113    SymbolDesc SD = {SymbolStart, SymbolSize};
114    Functions.insert(std::make_pair(SD, Export.Name));
115  }
116  return std::error_code();
117}
118
119std::error_code SymbolizableObjectFile::addSymbol(const SymbolRef &Symbol,
120                                                  uint64_t SymbolSize,
121                                                  DataExtractor *OpdExtractor,
122                                                  uint64_t OpdAddress) {
123  Expected<SymbolRef::Type> SymbolTypeOrErr = Symbol.getType();
124  if (!SymbolTypeOrErr)
125    return errorToErrorCode(SymbolTypeOrErr.takeError());
126  SymbolRef::Type SymbolType = *SymbolTypeOrErr;
127  if (SymbolType != SymbolRef::ST_Function && SymbolType != SymbolRef::ST_Data)
128    return std::error_code();
129  Expected<uint64_t> SymbolAddressOrErr = Symbol.getAddress();
130  if (!SymbolAddressOrErr)
131    return errorToErrorCode(SymbolAddressOrErr.takeError());
132  uint64_t SymbolAddress = *SymbolAddressOrErr;
133  if (OpdExtractor) {
134    // For big-endian PowerPC64 ELF, symbols in the .opd section refer to
135    // function descriptors. The first word of the descriptor is a pointer to
136    // the function's code.
137    // For the purposes of symbolization, pretend the symbol's address is that
138    // of the function's code, not the descriptor.
139    uint64_t OpdOffset = SymbolAddress - OpdAddress;
140    uint32_t OpdOffset32 = OpdOffset;
141    if (OpdOffset == OpdOffset32 &&
142        OpdExtractor->isValidOffsetForAddress(OpdOffset32))
143      SymbolAddress = OpdExtractor->getAddress(&OpdOffset32);
144  }
145  Expected<StringRef> SymbolNameOrErr = Symbol.getName();
146  if (!SymbolNameOrErr)
147    return errorToErrorCode(SymbolNameOrErr.takeError());
148  StringRef SymbolName = *SymbolNameOrErr;
149  // Mach-O symbol table names have leading underscore, skip it.
150  if (Module->isMachO() && SymbolName.size() > 0 && SymbolName[0] == '_')
151    SymbolName = SymbolName.drop_front();
152  // FIXME: If a function has alias, there are two entries in symbol table
153  // with same address size. Make sure we choose the correct one.
154  auto &M = SymbolType == SymbolRef::ST_Function ? Functions : Objects;
155  SymbolDesc SD = { SymbolAddress, SymbolSize };
156  M.insert(std::make_pair(SD, SymbolName));
157  return std::error_code();
158}
159
160// Return true if this is a 32-bit x86 PE COFF module.
161bool SymbolizableObjectFile::isWin32Module() const {
162  auto *CoffObject = dyn_cast<COFFObjectFile>(Module);
163  return CoffObject && CoffObject->getMachine() == COFF::IMAGE_FILE_MACHINE_I386;
164}
165
166uint64_t SymbolizableObjectFile::getModulePreferredBase() const {
167  if (auto *CoffObject = dyn_cast<COFFObjectFile>(Module))
168    return CoffObject->getImageBase();
169  return 0;
170}
171
172bool SymbolizableObjectFile::getNameFromSymbolTable(SymbolRef::Type Type,
173                                                    uint64_t Address,
174                                                    std::string &Name,
175                                                    uint64_t &Addr,
176                                                    uint64_t &Size) const {
177  const auto &SymbolMap = Type == SymbolRef::ST_Function ? Functions : Objects;
178  if (SymbolMap.empty())
179    return false;
180  SymbolDesc SD = { Address, Address };
181  auto SymbolIterator = SymbolMap.upper_bound(SD);
182  if (SymbolIterator == SymbolMap.begin())
183    return false;
184  --SymbolIterator;
185  if (SymbolIterator->first.Size != 0 &&
186      SymbolIterator->first.Addr + SymbolIterator->first.Size <= Address)
187    return false;
188  Name = SymbolIterator->second.str();
189  Addr = SymbolIterator->first.Addr;
190  Size = SymbolIterator->first.Size;
191  return true;
192}
193
194bool SymbolizableObjectFile::shouldOverrideWithSymbolTable(
195    FunctionNameKind FNKind, bool UseSymbolTable) const {
196  // When DWARF is used with -gline-tables-only / -gmlt, the symbol table gives
197  // better answers for linkage names than the DIContext. Otherwise, we are
198  // probably using PEs and PDBs, and we shouldn't do the override. PE files
199  // generally only contain the names of exported symbols.
200  return FNKind == FunctionNameKind::LinkageName && UseSymbolTable &&
201         isa<DWARFContext>(DebugInfoContext.get());
202}
203
204DILineInfo SymbolizableObjectFile::symbolizeCode(uint64_t ModuleOffset,
205                                                 FunctionNameKind FNKind,
206                                                 bool UseSymbolTable) const {
207  DILineInfo LineInfo;
208  if (DebugInfoContext) {
209    LineInfo = DebugInfoContext->getLineInfoForAddress(
210        ModuleOffset, getDILineInfoSpecifier(FNKind));
211  }
212  // Override function name from symbol table if necessary.
213  if (shouldOverrideWithSymbolTable(FNKind, UseSymbolTable)) {
214    std::string FunctionName;
215    uint64_t Start, Size;
216    if (getNameFromSymbolTable(SymbolRef::ST_Function, ModuleOffset,
217                               FunctionName, Start, Size)) {
218      LineInfo.FunctionName = FunctionName;
219    }
220  }
221  return LineInfo;
222}
223
224DIInliningInfo SymbolizableObjectFile::symbolizeInlinedCode(
225    uint64_t ModuleOffset, FunctionNameKind FNKind, bool UseSymbolTable) const {
226  DIInliningInfo InlinedContext;
227
228  if (DebugInfoContext)
229    InlinedContext = DebugInfoContext->getInliningInfoForAddress(
230        ModuleOffset, getDILineInfoSpecifier(FNKind));
231  // Make sure there is at least one frame in context.
232  if (InlinedContext.getNumberOfFrames() == 0)
233    InlinedContext.addFrame(DILineInfo());
234
235  // Override the function name in lower frame with name from symbol table.
236  if (shouldOverrideWithSymbolTable(FNKind, UseSymbolTable)) {
237    std::string FunctionName;
238    uint64_t Start, Size;
239    if (getNameFromSymbolTable(SymbolRef::ST_Function, ModuleOffset,
240                               FunctionName, Start, Size)) {
241      InlinedContext.getMutableFrame(InlinedContext.getNumberOfFrames() - 1)
242          ->FunctionName = FunctionName;
243    }
244  }
245
246  return InlinedContext;
247}
248
249DIGlobal SymbolizableObjectFile::symbolizeData(uint64_t ModuleOffset) const {
250  DIGlobal Res;
251  getNameFromSymbolTable(SymbolRef::ST_Data, ModuleOffset, Res.Name, Res.Start,
252                         Res.Size);
253  return Res;
254}
255
256}  // namespace symbolize
257}  // namespace llvm
258
259