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
25using namespace llvm;
26using namespace object;
27using namespace symbolize;
28
29Expected<std::unique_ptr<SymbolizableObjectFile>>
30SymbolizableObjectFile::create(const object::ObjectFile *Obj,
31 std::unique_ptr<DIContext> DICtx,
32 bool UntagAddresses) {
33 assert(DICtx);
34 std::unique_ptr<SymbolizableObjectFile> res(
35 new SymbolizableObjectFile(Obj, std::move(DICtx), UntagAddresses));
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 Expected<StringRef> NameOrErr = Section->getName();
43 if (!NameOrErr)
44 return NameOrErr.takeError();
45
46 if (*NameOrErr == ".opd") {
47 Expected<StringRef> E = Section->getContents();
48 if (!E)
49 return E.takeError();
50 OpdExtractor.reset(p: new DataExtractor(*E, Obj->isLittleEndian(),
51 Obj->getBytesInAddress()));
52 OpdAddress = Section->getAddress();
53 break;
54 }
55 }
56 }
57 std::vector<std::pair<SymbolRef, uint64_t>> Symbols =
58 computeSymbolSizes(O: *Obj);
59 for (auto &P : Symbols)
60 if (Error E =
61 res->addSymbol(Symbol: P.first, SymbolSize: P.second, OpdExtractor: OpdExtractor.get(), OpdAddress))
62 return std::move(E);
63
64 // If this is a COFF object and we didn't find any symbols, try the export
65 // table.
66 if (Symbols.empty()) {
67 if (auto *CoffObj = dyn_cast<COFFObjectFile>(Val: Obj))
68 if (Error E = res->addCoffExportSymbols(CoffObj))
69 return std::move(E);
70 }
71
72 std::vector<SymbolDesc> &SS = res->Symbols;
73 // Sort by (Addr,Size,Name). If several SymbolDescs share the same Addr,
74 // pick the one with the largest Size. This helps us avoid symbols with no
75 // size information (Size=0).
76 llvm::stable_sort(Range&: SS);
77 auto I = SS.begin(), E = SS.end(), J = SS.begin();
78 while (I != E) {
79 auto OI = I;
80 while (++I != E && OI->Addr == I->Addr) {
81 }
82 *J++ = I[-1];
83 }
84 SS.erase(first: J, last: SS.end());
85
86 return std::move(res);
87}
88
89SymbolizableObjectFile::SymbolizableObjectFile(const ObjectFile *Obj,
90 std::unique_ptr<DIContext> DICtx,
91 bool UntagAddresses)
92 : Module(Obj), DebugInfoContext(std::move(DICtx)),
93 UntagAddresses(UntagAddresses) {}
94
95namespace {
96
97struct OffsetNamePair {
98 uint32_t Offset;
99 StringRef Name;
100
101 bool operator<(const OffsetNamePair &R) const {
102 return Offset < R.Offset;
103 }
104};
105
106} // end anonymous namespace
107
108Error SymbolizableObjectFile::addCoffExportSymbols(
109 const COFFObjectFile *CoffObj) {
110 // Get all export names and offsets.
111 std::vector<OffsetNamePair> ExportSyms;
112 for (const ExportDirectoryEntryRef &Ref : CoffObj->export_directories()) {
113 StringRef Name;
114 uint32_t Offset;
115 if (auto EC = Ref.getSymbolName(Result&: Name))
116 return EC;
117 if (auto EC = Ref.getExportRVA(Result&: Offset))
118 return EC;
119 ExportSyms.push_back(x: OffsetNamePair{.Offset: Offset, .Name: Name});
120 }
121 if (ExportSyms.empty())
122 return Error::success();
123
124 // Sort by ascending offset.
125 array_pod_sort(Start: ExportSyms.begin(), End: ExportSyms.end());
126
127 // Approximate the symbol sizes by assuming they run to the next symbol.
128 // FIXME: This assumes all exports are functions.
129 uint64_t ImageBase = CoffObj->getImageBase();
130 for (auto I = ExportSyms.begin(), E = ExportSyms.end(); I != E; ++I) {
131 OffsetNamePair &Export = *I;
132 // FIXME: The last export has a one byte size now.
133 uint32_t NextOffset = I != E ? I->Offset : Export.Offset + 1;
134 uint64_t SymbolStart = ImageBase + Export.Offset;
135 uint64_t SymbolSize = NextOffset - Export.Offset;
136 Symbols.push_back(x: {.Addr: SymbolStart, .Size: SymbolSize, .Name: Export.Name, .ELFLocalSymIdx: 0});
137 }
138 return Error::success();
139}
140
141Error SymbolizableObjectFile::addSymbol(const SymbolRef &Symbol,
142 uint64_t SymbolSize,
143 DataExtractor *OpdExtractor,
144 uint64_t OpdAddress) {
145 // Avoid adding symbols from an unknown/undefined section.
146 const ObjectFile &Obj = *Symbol.getObject();
147 Expected<StringRef> SymbolNameOrErr = Symbol.getName();
148 if (!SymbolNameOrErr)
149 return SymbolNameOrErr.takeError();
150 StringRef SymbolName = *SymbolNameOrErr;
151
152 uint32_t ELFSymIdx =
153 Obj.isELF() ? ELFSymbolRef(Symbol).getRawDataRefImpl().d.b : 0;
154 Expected<section_iterator> Sec = Symbol.getSection();
155 if (!Sec || Obj.section_end() == *Sec) {
156 if (Obj.isELF()) {
157 // Store the (index, filename) pair for a file symbol.
158 ELFSymbolRef ESym(Symbol);
159 if (ESym.getELFType() == ELF::STT_FILE)
160 FileSymbols.emplace_back(args&: ELFSymIdx, args&: SymbolName);
161 }
162 return Error::success();
163 }
164
165 Expected<SymbolRef::Type> SymbolTypeOrErr = Symbol.getType();
166 if (!SymbolTypeOrErr)
167 return SymbolTypeOrErr.takeError();
168 SymbolRef::Type SymbolType = *SymbolTypeOrErr;
169 if (Obj.isELF()) {
170 // Ignore any symbols coming from sections that don't have runtime
171 // allocated memory.
172 if ((elf_section_iterator(*Sec)->getFlags() & ELF::SHF_ALLOC) == 0)
173 return Error::success();
174
175 // Allow function and data symbols. Additionally allow STT_NONE, which are
176 // common for functions defined in assembly.
177 uint8_t Type = ELFSymbolRef(Symbol).getELFType();
178 if (Type != ELF::STT_NOTYPE && Type != ELF::STT_FUNC &&
179 Type != ELF::STT_OBJECT && Type != ELF::STT_GNU_IFUNC)
180 return Error::success();
181 // Some STT_NOTYPE symbols are not desired. This excludes STT_SECTION and
182 // ARM mapping symbols.
183 uint32_t Flags = cantFail(ValOrErr: Symbol.getFlags());
184 if (Flags & SymbolRef::SF_FormatSpecific)
185 return Error::success();
186 } else if (SymbolType != SymbolRef::ST_Function &&
187 SymbolType != SymbolRef::ST_Data) {
188 return Error::success();
189 }
190
191 Expected<uint64_t> SymbolAddressOrErr = Symbol.getAddress();
192 if (!SymbolAddressOrErr)
193 return SymbolAddressOrErr.takeError();
194 uint64_t SymbolAddress = *SymbolAddressOrErr;
195 if (UntagAddresses) {
196 // For kernel addresses, bits 56-63 need to be set, so we sign extend bit 55
197 // into bits 56-63 instead of masking them out.
198 SymbolAddress &= (1ull << 56) - 1;
199 SymbolAddress = (int64_t(SymbolAddress) << 8) >> 8;
200 }
201 if (OpdExtractor) {
202 // For big-endian PowerPC64 ELF, symbols in the .opd section refer to
203 // function descriptors. The first word of the descriptor is a pointer to
204 // the function's code.
205 // For the purposes of symbolization, pretend the symbol's address is that
206 // of the function's code, not the descriptor.
207 uint64_t OpdOffset = SymbolAddress - OpdAddress;
208 if (OpdExtractor->isValidOffsetForAddress(offset: OpdOffset))
209 SymbolAddress = OpdExtractor->getAddress(offset_ptr: &OpdOffset);
210 }
211 // Mach-O symbol table names have leading underscore, skip it.
212 if (Module->isMachO())
213 SymbolName.consume_front(Prefix: "_");
214
215 if (Obj.isELF() && ELFSymbolRef(Symbol).getBinding() != ELF::STB_LOCAL)
216 ELFSymIdx = 0;
217 Symbols.push_back(x: {.Addr: SymbolAddress, .Size: SymbolSize, .Name: SymbolName, .ELFLocalSymIdx: ELFSymIdx});
218 return Error::success();
219}
220
221// Return true if this is a 32-bit x86 PE COFF module.
222bool SymbolizableObjectFile::isWin32Module() const {
223 auto *CoffObject = dyn_cast<COFFObjectFile>(Val: Module);
224 return CoffObject && CoffObject->getMachine() == COFF::IMAGE_FILE_MACHINE_I386;
225}
226
227uint64_t SymbolizableObjectFile::getModulePreferredBase() const {
228 if (auto *CoffObject = dyn_cast<COFFObjectFile>(Val: Module))
229 return CoffObject->getImageBase();
230 return 0;
231}
232
233bool SymbolizableObjectFile::getNameFromSymbolTable(
234 uint64_t Address, std::string &Name, uint64_t &Addr, uint64_t &Size,
235 std::string &FileName) const {
236 SymbolDesc SD{.Addr: Address, UINT64_C(-1), .Name: StringRef(), .ELFLocalSymIdx: 0};
237 auto SymbolIterator = llvm::upper_bound(Range: Symbols, Value&: SD);
238 if (SymbolIterator == Symbols.begin())
239 return false;
240 --SymbolIterator;
241 if (SymbolIterator->Size != 0 &&
242 SymbolIterator->Addr + SymbolIterator->Size <= Address)
243 return false;
244 Name = SymbolIterator->Name.str();
245 Addr = SymbolIterator->Addr;
246 Size = SymbolIterator->Size;
247
248 if (SymbolIterator->ELFLocalSymIdx != 0) {
249 // If this is an ELF local symbol, find the STT_FILE symbol preceding
250 // SymbolIterator to get the filename. The ELF spec requires the STT_FILE
251 // symbol (if present) precedes the other STB_LOCAL symbols for the file.
252 assert(Module->isELF());
253 auto It = llvm::upper_bound(
254 Range: FileSymbols,
255 Value: std::make_pair(x: SymbolIterator->ELFLocalSymIdx, y: StringRef()));
256 if (It != FileSymbols.begin())
257 FileName = It[-1].second.str();
258 }
259 return true;
260}
261
262bool SymbolizableObjectFile::shouldOverrideWithSymbolTable(
263 FunctionNameKind FNKind, bool UseSymbolTable) const {
264 // When DWARF is used with -gline-tables-only / -gmlt, the symbol table gives
265 // better answers for linkage names than the DIContext. Otherwise, we are
266 // probably using PEs and PDBs, and we shouldn't do the override. PE files
267 // generally only contain the names of exported symbols.
268 return FNKind == FunctionNameKind::LinkageName && UseSymbolTable &&
269 isa<DWARFContext>(Val: DebugInfoContext.get());
270}
271
272DILineInfo
273SymbolizableObjectFile::symbolizeCode(object::SectionedAddress ModuleOffset,
274 DILineInfoSpecifier LineInfoSpecifier,
275 bool UseSymbolTable) const {
276 if (ModuleOffset.SectionIndex == object::SectionedAddress::UndefSection)
277 ModuleOffset.SectionIndex =
278 getModuleSectionIndexForAddress(Address: ModuleOffset.Address);
279 DILineInfo LineInfo;
280 std::optional<DILineInfo> DBGLineInfo =
281 DebugInfoContext->getLineInfoForAddress(Address: ModuleOffset, Specifier: LineInfoSpecifier);
282 if (DBGLineInfo)
283 LineInfo = *DBGLineInfo;
284
285 // Override function name from symbol table if necessary.
286 if (shouldOverrideWithSymbolTable(FNKind: LineInfoSpecifier.FNKind, UseSymbolTable)) {
287 std::string FunctionName, FileName;
288 uint64_t Start, Size;
289 if (getNameFromSymbolTable(Address: ModuleOffset.Address, Name&: FunctionName, Addr&: Start, Size,
290 FileName)) {
291 LineInfo.FunctionName = FunctionName;
292 LineInfo.StartAddress = Start;
293 // Only use the filename from symbol table if the debug info for the
294 // address is missing.
295 if (!DBGLineInfo && !FileName.empty())
296 LineInfo.FileName = FileName;
297 }
298 }
299 return LineInfo;
300}
301
302DIInliningInfo SymbolizableObjectFile::symbolizeInlinedCode(
303 object::SectionedAddress ModuleOffset,
304 DILineInfoSpecifier LineInfoSpecifier, bool UseSymbolTable) const {
305 if (ModuleOffset.SectionIndex == object::SectionedAddress::UndefSection)
306 ModuleOffset.SectionIndex =
307 getModuleSectionIndexForAddress(Address: ModuleOffset.Address);
308 DIInliningInfo InlinedContext = DebugInfoContext->getInliningInfoForAddress(
309 Address: ModuleOffset, Specifier: LineInfoSpecifier);
310
311 // Make sure there is at least one frame in context.
312 bool EmptyFrameAdded = false;
313 if (InlinedContext.getNumberOfFrames() == 0) {
314 EmptyFrameAdded = true;
315 InlinedContext.addFrame(Frame: DILineInfo());
316 }
317
318 // Override the function name in lower frame with name from symbol table.
319 if (shouldOverrideWithSymbolTable(FNKind: LineInfoSpecifier.FNKind, UseSymbolTable)) {
320 std::string FunctionName, FileName;
321 uint64_t Start, Size;
322 if (getNameFromSymbolTable(Address: ModuleOffset.Address, Name&: FunctionName, Addr&: Start, Size,
323 FileName)) {
324 DILineInfo *LI = InlinedContext.getMutableFrame(
325 Index: InlinedContext.getNumberOfFrames() - 1);
326 LI->FunctionName = FunctionName;
327 LI->StartAddress = Start;
328 // Only use the filename from symbol table if the debug info for the
329 // address is missing.
330 if (EmptyFrameAdded && !FileName.empty())
331 LI->FileName = FileName;
332 }
333 }
334
335 return InlinedContext;
336}
337
338DIGlobal SymbolizableObjectFile::symbolizeData(
339 object::SectionedAddress ModuleOffset) const {
340 DIGlobal Res;
341 std::string FileName;
342 getNameFromSymbolTable(Address: ModuleOffset.Address, Name&: Res.Name, Addr&: Res.Start, Size&: Res.Size,
343 FileName);
344 Res.DeclFile = FileName;
345
346 // Try and get a better filename:lineno pair from the debuginfo, if present.
347 std::optional<DILineInfo> DL =
348 DebugInfoContext->getLineInfoForDataAddress(Address: ModuleOffset);
349 if (DL && DL->Line != 0) {
350 Res.DeclFile = DL->FileName;
351 Res.DeclLine = DL->Line;
352 }
353 return Res;
354}
355
356std::vector<DILocal> SymbolizableObjectFile::symbolizeFrame(
357 object::SectionedAddress ModuleOffset) const {
358 if (ModuleOffset.SectionIndex == object::SectionedAddress::UndefSection)
359 ModuleOffset.SectionIndex =
360 getModuleSectionIndexForAddress(Address: ModuleOffset.Address);
361 return DebugInfoContext->getLocalsForAddress(Address: ModuleOffset);
362}
363
364std::vector<object::SectionedAddress>
365SymbolizableObjectFile::findSymbol(StringRef Symbol, uint64_t Offset) const {
366 std::vector<object::SectionedAddress> Result;
367 for (const SymbolDesc &Sym : Symbols) {
368 if (Sym.Name == Symbol) {
369 uint64_t Addr = Sym.Addr;
370 if (Offset < Sym.Size)
371 Addr += Offset;
372 object::SectionedAddress A{.Address: Addr, .SectionIndex: getModuleSectionIndexForAddress(Address: Addr)};
373 Result.push_back(x: A);
374 }
375 }
376 return Result;
377}
378
379/// Search for the first occurence of specified Address in ObjectFile.
380uint64_t SymbolizableObjectFile::getModuleSectionIndexForAddress(
381 uint64_t Address) const {
382
383 for (SectionRef Sec : Module->sections()) {
384 if (!Sec.isText() || Sec.isVirtual())
385 continue;
386
387 if (Address >= Sec.getAddress() &&
388 Address < Sec.getAddress() + Sec.getSize())
389 return Sec.getIndex();
390 }
391
392 return object::SectionedAddress::UndefSection;
393}
394