1//===- SymbolizableObjectFile.cpp -----------------------------------------===//
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// Implementation of SymbolizableObjectFile class.
10//
11//===----------------------------------------------------------------------===//
12
13#include "llvm/DebugInfo/Symbolize/SymbolizableObjectFile.h"
14#include "llvm/ADT/STLExtras.h"
15#include "llvm/BinaryFormat/COFF.h"
16#include "llvm/DebugInfo/DWARF/DWARFContext.h"
17#include "llvm/Object/COFF.h"
18#include "llvm/Object/ELFObjectFile.h"
19#include "llvm/Object/ObjectFile.h"
20#include "llvm/Object/SymbolSize.h"
21#include "llvm/Support/Casting.h"
22#include "llvm/Support/DataExtractor.h"
23#include "llvm/TargetParser/Triple.h"
24#include <algorithm>
25
26using namespace llvm;
27using namespace object;
28using namespace symbolize;
29
30Expected<std::unique_ptr<SymbolizableObjectFile>>
31SymbolizableObjectFile::create(const object::ObjectFile *Obj,
32 std::unique_ptr<DIContext> DICtx,
33 bool UntagAddresses) {
34 assert(DICtx);
35 std::unique_ptr<SymbolizableObjectFile> res(
36 new SymbolizableObjectFile(Obj, std::move(DICtx), UntagAddresses));
37 std::unique_ptr<DataExtractor> OpdExtractor;
38 uint64_t OpdAddress = 0;
39 // Find the .opd (function descriptor) section if any, for big-endian
40 // PowerPC64 ELF.
41 if (Obj->getArch() == Triple::ppc64) {
42 for (section_iterator Section : Obj->sections()) {
43 Expected<StringRef> NameOrErr = Section->getName();
44 if (!NameOrErr)
45 return NameOrErr.takeError();
46
47 if (*NameOrErr == ".opd") {
48 Expected<StringRef> E = Section->getContents();
49 if (!E)
50 return E.takeError();
51 OpdExtractor.reset(p: new DataExtractor(*E, Obj->isLittleEndian(),
52 Obj->getBytesInAddress()));
53 OpdAddress = Section->getAddress();
54 break;
55 }
56 }
57 }
58 std::vector<std::pair<SymbolRef, uint64_t>> Symbols =
59 computeSymbolSizes(O: *Obj);
60 for (auto &P : Symbols)
61 if (Error E =
62 res->addSymbol(Symbol: P.first, SymbolSize: P.second, OpdExtractor: OpdExtractor.get(), OpdAddress))
63 return std::move(E);
64
65 // If this is a COFF object and we didn't find any symbols, try the export
66 // table.
67 if (Symbols.empty()) {
68 if (auto *CoffObj = dyn_cast<COFFObjectFile>(Val: Obj))
69 if (Error E = res->addCoffExportSymbols(CoffObj))
70 return std::move(E);
71 }
72
73 std::vector<SymbolDesc> &SS = res->Symbols;
74 // Sort by (Addr,Size,Name). If several SymbolDescs share the same Addr,
75 // pick the one with the largest Size. This helps us avoid symbols with no
76 // size information (Size=0).
77 llvm::stable_sort(Range&: SS);
78 auto I = SS.begin(), E = SS.end(), J = SS.begin();
79 while (I != E) {
80 auto OI = I;
81 while (++I != E && OI->Addr == I->Addr) {
82 }
83 *J++ = I[-1];
84 }
85 SS.erase(first: J, last: SS.end());
86
87 return std::move(res);
88}
89
90SymbolizableObjectFile::SymbolizableObjectFile(const ObjectFile *Obj,
91 std::unique_ptr<DIContext> DICtx,
92 bool UntagAddresses)
93 : Module(Obj), DebugInfoContext(std::move(DICtx)),
94 UntagAddresses(UntagAddresses) {}
95
96namespace {
97
98struct OffsetNamePair {
99 uint32_t Offset;
100 StringRef Name;
101
102 bool operator<(const OffsetNamePair &R) const {
103 return Offset < R.Offset;
104 }
105};
106
107} // end anonymous namespace
108
109Error SymbolizableObjectFile::addCoffExportSymbols(
110 const COFFObjectFile *CoffObj) {
111 // Get all export names and offsets.
112 std::vector<OffsetNamePair> ExportSyms;
113 for (const ExportDirectoryEntryRef &Ref : CoffObj->export_directories()) {
114 StringRef Name;
115 uint32_t Offset;
116 if (auto EC = Ref.getSymbolName(Result&: Name))
117 return EC;
118 if (auto EC = Ref.getExportRVA(Result&: Offset))
119 return EC;
120 ExportSyms.push_back(x: OffsetNamePair{.Offset: Offset, .Name: Name});
121 }
122 if (ExportSyms.empty())
123 return Error::success();
124
125 // Sort by ascending offset.
126 array_pod_sort(Start: ExportSyms.begin(), End: ExportSyms.end());
127
128 // Approximate the symbol sizes by assuming they run to the next symbol.
129 // FIXME: This assumes all exports are functions.
130 uint64_t ImageBase = CoffObj->getImageBase();
131 for (auto I = ExportSyms.begin(), E = ExportSyms.end(); I != E; ++I) {
132 OffsetNamePair &Export = *I;
133 // FIXME: The last export has a one byte size now.
134 uint32_t NextOffset = I != E ? I->Offset : Export.Offset + 1;
135 uint64_t SymbolStart = ImageBase + Export.Offset;
136 uint64_t SymbolSize = NextOffset - Export.Offset;
137 Symbols.push_back(x: {.Addr: SymbolStart, .Size: SymbolSize, .Name: Export.Name, .ELFLocalSymIdx: 0});
138 }
139 return Error::success();
140}
141
142Error SymbolizableObjectFile::addSymbol(const SymbolRef &Symbol,
143 uint64_t SymbolSize,
144 DataExtractor *OpdExtractor,
145 uint64_t OpdAddress) {
146 // Avoid adding symbols from an unknown/undefined section.
147 const ObjectFile &Obj = *Symbol.getObject();
148 Expected<StringRef> SymbolNameOrErr = Symbol.getName();
149 if (!SymbolNameOrErr)
150 return SymbolNameOrErr.takeError();
151 StringRef SymbolName = *SymbolNameOrErr;
152
153 uint32_t ELFSymIdx =
154 Obj.isELF() ? ELFSymbolRef(Symbol).getRawDataRefImpl().d.b : 0;
155 Expected<section_iterator> Sec = Symbol.getSection();
156 if (!Sec || Obj.section_end() == *Sec) {
157 if (Obj.isELF()) {
158 // Store the (index, filename) pair for a file symbol.
159 ELFSymbolRef ESym(Symbol);
160 if (ESym.getELFType() == ELF::STT_FILE)
161 FileSymbols.emplace_back(args&: ELFSymIdx, args&: SymbolName);
162 }
163 return Error::success();
164 }
165
166 Expected<SymbolRef::Type> SymbolTypeOrErr = Symbol.getType();
167 if (!SymbolTypeOrErr)
168 return SymbolTypeOrErr.takeError();
169 SymbolRef::Type SymbolType = *SymbolTypeOrErr;
170 if (Obj.isELF()) {
171 // Ignore any symbols coming from sections that don't have runtime
172 // allocated memory.
173 if ((elf_section_iterator(*Sec)->getFlags() & ELF::SHF_ALLOC) == 0)
174 return Error::success();
175
176 // Allow function and data symbols. Additionally allow STT_NONE, which are
177 // common for functions defined in assembly.
178 uint8_t Type = ELFSymbolRef(Symbol).getELFType();
179 if (Type != ELF::STT_NOTYPE && Type != ELF::STT_FUNC &&
180 Type != ELF::STT_OBJECT && Type != ELF::STT_GNU_IFUNC)
181 return Error::success();
182 // Some STT_NOTYPE symbols are not desired. This excludes STT_SECTION and
183 // ARM mapping symbols.
184 uint32_t Flags = cantFail(ValOrErr: Symbol.getFlags());
185 if (Flags & SymbolRef::SF_FormatSpecific)
186 return Error::success();
187 } else if (SymbolType != SymbolRef::ST_Function &&
188 SymbolType != SymbolRef::ST_Data) {
189 return Error::success();
190 }
191
192 Expected<uint64_t> SymbolAddressOrErr = Symbol.getAddress();
193 if (!SymbolAddressOrErr)
194 return SymbolAddressOrErr.takeError();
195 uint64_t SymbolAddress = *SymbolAddressOrErr;
196 if (UntagAddresses) {
197 // For kernel addresses, bits 56-63 need to be set, so we sign extend bit 55
198 // into bits 56-63 instead of masking them out.
199 SymbolAddress &= (1ull << 56) - 1;
200 SymbolAddress = (int64_t(SymbolAddress) << 8) >> 8;
201 }
202 if (OpdExtractor) {
203 // For big-endian PowerPC64 ELF, symbols in the .opd section refer to
204 // function descriptors. The first word of the descriptor is a pointer to
205 // the function's code.
206 // For the purposes of symbolization, pretend the symbol's address is that
207 // of the function's code, not the descriptor.
208 uint64_t OpdOffset = SymbolAddress - OpdAddress;
209 if (OpdExtractor->isValidOffsetForAddress(offset: OpdOffset))
210 SymbolAddress = OpdExtractor->getAddress(offset_ptr: &OpdOffset);
211 }
212 // Mach-O symbol table names have leading underscore, skip it.
213 if (Module->isMachO())
214 SymbolName.consume_front(Prefix: "_");
215
216 if (Obj.isELF() && ELFSymbolRef(Symbol).getBinding() != ELF::STB_LOCAL)
217 ELFSymIdx = 0;
218 Symbols.push_back(x: {.Addr: SymbolAddress, .Size: SymbolSize, .Name: SymbolName, .ELFLocalSymIdx: ELFSymIdx});
219 return Error::success();
220}
221
222// Return true if this is a 32-bit x86 PE COFF module.
223bool SymbolizableObjectFile::isWin32Module() const {
224 auto *CoffObject = dyn_cast<COFFObjectFile>(Val: Module);
225 return CoffObject && CoffObject->getMachine() == COFF::IMAGE_FILE_MACHINE_I386;
226}
227
228uint64_t SymbolizableObjectFile::getModulePreferredBase() const {
229 if (auto *CoffObject = dyn_cast<COFFObjectFile>(Val: Module))
230 return CoffObject->getImageBase();
231 return 0;
232}
233
234bool SymbolizableObjectFile::getNameFromSymbolTable(
235 uint64_t Address, std::string &Name, uint64_t &Addr, uint64_t &Size,
236 std::string &FileName) const {
237 SymbolDesc SD{.Addr: Address, UINT64_C(-1), .Name: StringRef(), .ELFLocalSymIdx: 0};
238 auto SymbolIterator = llvm::upper_bound(Range: Symbols, Value&: SD);
239 if (SymbolIterator == Symbols.begin())
240 return false;
241 --SymbolIterator;
242 if (SymbolIterator->Size != 0 &&
243 SymbolIterator->Addr + SymbolIterator->Size <= Address)
244 return false;
245 Name = SymbolIterator->Name.str();
246 Addr = SymbolIterator->Addr;
247 Size = SymbolIterator->Size;
248
249 if (SymbolIterator->ELFLocalSymIdx != 0) {
250 // If this is an ELF local symbol, find the STT_FILE symbol preceding
251 // SymbolIterator to get the filename. The ELF spec requires the STT_FILE
252 // symbol (if present) precedes the other STB_LOCAL symbols for the file.
253 assert(Module->isELF());
254 auto It = llvm::upper_bound(
255 Range: FileSymbols,
256 Value: std::make_pair(x: SymbolIterator->ELFLocalSymIdx, y: StringRef()));
257 if (It != FileSymbols.begin())
258 FileName = It[-1].second.str();
259 }
260 return true;
261}
262
263bool SymbolizableObjectFile::shouldOverrideWithSymbolTable(
264 FunctionNameKind FNKind, bool UseSymbolTable) const {
265 // When DWARF is used with -gline-tables-only / -gmlt, the symbol table gives
266 // better answers for linkage names than the DIContext. Otherwise, we are
267 // probably using PEs and PDBs, and we shouldn't do the override. PE files
268 // generally only contain the names of exported symbols.
269 return FNKind == FunctionNameKind::LinkageName && UseSymbolTable &&
270 isa<DWARFContext>(Val: DebugInfoContext.get());
271}
272
273DILineInfo
274SymbolizableObjectFile::symbolizeCode(object::SectionedAddress ModuleOffset,
275 DILineInfoSpecifier LineInfoSpecifier,
276 bool UseSymbolTable) const {
277 if (ModuleOffset.SectionIndex == object::SectionedAddress::UndefSection)
278 ModuleOffset.SectionIndex =
279 getModuleSectionIndexForAddress(Address: ModuleOffset.Address);
280 DILineInfo LineInfo =
281 DebugInfoContext->getLineInfoForAddress(Address: ModuleOffset, Specifier: LineInfoSpecifier);
282
283 // Override function name from symbol table if necessary.
284 if (shouldOverrideWithSymbolTable(FNKind: LineInfoSpecifier.FNKind, UseSymbolTable)) {
285 std::string FunctionName, FileName;
286 uint64_t Start, Size;
287 if (getNameFromSymbolTable(Address: ModuleOffset.Address, Name&: FunctionName, Addr&: Start, Size,
288 FileName)) {
289 LineInfo.FunctionName = FunctionName;
290 LineInfo.StartAddress = Start;
291 if (LineInfo.FileName == DILineInfo::BadString && !FileName.empty())
292 LineInfo.FileName = FileName;
293 }
294 }
295 return LineInfo;
296}
297
298DIInliningInfo SymbolizableObjectFile::symbolizeInlinedCode(
299 object::SectionedAddress ModuleOffset,
300 DILineInfoSpecifier LineInfoSpecifier, bool UseSymbolTable) const {
301 if (ModuleOffset.SectionIndex == object::SectionedAddress::UndefSection)
302 ModuleOffset.SectionIndex =
303 getModuleSectionIndexForAddress(Address: ModuleOffset.Address);
304 DIInliningInfo InlinedContext = DebugInfoContext->getInliningInfoForAddress(
305 Address: ModuleOffset, Specifier: LineInfoSpecifier);
306
307 // Make sure there is at least one frame in context.
308 if (InlinedContext.getNumberOfFrames() == 0)
309 InlinedContext.addFrame(Frame: DILineInfo());
310
311 // Override the function name in lower frame with name from symbol table.
312 if (shouldOverrideWithSymbolTable(FNKind: LineInfoSpecifier.FNKind, UseSymbolTable)) {
313 std::string FunctionName, FileName;
314 uint64_t Start, Size;
315 if (getNameFromSymbolTable(Address: ModuleOffset.Address, Name&: FunctionName, Addr&: Start, Size,
316 FileName)) {
317 DILineInfo *LI = InlinedContext.getMutableFrame(
318 Index: InlinedContext.getNumberOfFrames() - 1);
319 LI->FunctionName = FunctionName;
320 LI->StartAddress = Start;
321 if (LI->FileName == DILineInfo::BadString && !FileName.empty())
322 LI->FileName = FileName;
323 }
324 }
325
326 return InlinedContext;
327}
328
329DIGlobal SymbolizableObjectFile::symbolizeData(
330 object::SectionedAddress ModuleOffset) const {
331 DIGlobal Res;
332 std::string FileName;
333 getNameFromSymbolTable(Address: ModuleOffset.Address, Name&: Res.Name, Addr&: Res.Start, Size&: Res.Size,
334 FileName);
335 Res.DeclFile = FileName;
336
337 // Try and get a better filename:lineno pair from the debuginfo, if present.
338 DILineInfo DL = DebugInfoContext->getLineInfoForDataAddress(Address: ModuleOffset);
339 if (DL.Line != 0) {
340 Res.DeclFile = DL.FileName;
341 Res.DeclLine = DL.Line;
342 }
343 return Res;
344}
345
346std::vector<DILocal> SymbolizableObjectFile::symbolizeFrame(
347 object::SectionedAddress ModuleOffset) const {
348 if (ModuleOffset.SectionIndex == object::SectionedAddress::UndefSection)
349 ModuleOffset.SectionIndex =
350 getModuleSectionIndexForAddress(Address: ModuleOffset.Address);
351 return DebugInfoContext->getLocalsForAddress(Address: ModuleOffset);
352}
353
354std::vector<object::SectionedAddress>
355SymbolizableObjectFile::findSymbol(StringRef Symbol, uint64_t Offset) const {
356 std::vector<object::SectionedAddress> Result;
357 for (const SymbolDesc &Sym : Symbols) {
358 if (Sym.Name == Symbol) {
359 uint64_t Addr = Sym.Addr;
360 if (Offset < Sym.Size)
361 Addr += Offset;
362 object::SectionedAddress A{.Address: Addr, .SectionIndex: getModuleSectionIndexForAddress(Address: Addr)};
363 Result.push_back(x: A);
364 }
365 }
366 return Result;
367}
368
369/// Search for the first occurence of specified Address in ObjectFile.
370uint64_t SymbolizableObjectFile::getModuleSectionIndexForAddress(
371 uint64_t Address) const {
372
373 for (SectionRef Sec : Module->sections()) {
374 if (!Sec.isText() || Sec.isVirtual())
375 continue;
376
377 if (Address >= Sec.getAddress() &&
378 Address < Sec.getAddress() + Sec.getSize())
379 return Sec.getIndex();
380 }
381
382 return object::SectionedAddress::UndefSection;
383}
384