1321369Sdim//===- SymbolizableObjectFile.cpp -----------------------------------------===// 2292915Sdim// 3353358Sdim// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4353358Sdim// See https://llvm.org/LICENSE.txt for license information. 5353358Sdim// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6292915Sdim// 7292915Sdim//===----------------------------------------------------------------------===// 8292915Sdim// 9292915Sdim// Implementation of SymbolizableObjectFile class. 10292915Sdim// 11292915Sdim//===----------------------------------------------------------------------===// 12292915Sdim 13292915Sdim#include "SymbolizableObjectFile.h" 14321369Sdim#include "llvm/ADT/STLExtras.h" 15321369Sdim#include "llvm/ADT/StringRef.h" 16321369Sdim#include "llvm/ADT/Triple.h" 17321369Sdim#include "llvm/BinaryFormat/COFF.h" 18321369Sdim#include "llvm/DebugInfo/DWARF/DWARFContext.h" 19321369Sdim#include "llvm/DebugInfo/Symbolize/SymbolizableModule.h" 20309124Sdim#include "llvm/Object/COFF.h" 21321369Sdim#include "llvm/Object/ObjectFile.h" 22292915Sdim#include "llvm/Object/SymbolSize.h" 23321369Sdim#include "llvm/Support/Casting.h" 24292915Sdim#include "llvm/Support/DataExtractor.h" 25321369Sdim#include "llvm/Support/Error.h" 26321369Sdim#include <algorithm> 27321369Sdim#include <cstdint> 28321369Sdim#include <memory> 29321369Sdim#include <string> 30321369Sdim#include <system_error> 31321369Sdim#include <utility> 32321369Sdim#include <vector> 33292915Sdim 34321369Sdimusing namespace llvm; 35292915Sdimusing namespace object; 36321369Sdimusing namespace symbolize; 37292915Sdim 38292915Sdimstatic DILineInfoSpecifier 39292915SdimgetDILineInfoSpecifier(FunctionNameKind FNKind) { 40292915Sdim return DILineInfoSpecifier( 41292915Sdim DILineInfoSpecifier::FileLineInfoKind::AbsoluteFilePath, FNKind); 42292915Sdim} 43292915Sdim 44292915SdimErrorOr<std::unique_ptr<SymbolizableObjectFile>> 45353358SdimSymbolizableObjectFile::create(const object::ObjectFile *Obj, 46360784Sdim std::unique_ptr<DIContext> DICtx, 47360784Sdim bool UntagAddresses) { 48353358Sdim assert(DICtx); 49292915Sdim std::unique_ptr<SymbolizableObjectFile> res( 50360784Sdim new SymbolizableObjectFile(Obj, std::move(DICtx), UntagAddresses)); 51292915Sdim std::unique_ptr<DataExtractor> OpdExtractor; 52292915Sdim uint64_t OpdAddress = 0; 53292915Sdim // Find the .opd (function descriptor) section if any, for big-endian 54292915Sdim // PowerPC64 ELF. 55292915Sdim if (Obj->getArch() == Triple::ppc64) { 56292915Sdim for (section_iterator Section : Obj->sections()) { 57360784Sdim Expected<StringRef> NameOrErr = Section->getName(); 58360784Sdim if (!NameOrErr) 59360784Sdim return errorToErrorCode(NameOrErr.takeError()); 60360784Sdim 61360784Sdim if (*NameOrErr == ".opd") { 62353358Sdim Expected<StringRef> E = Section->getContents(); 63353358Sdim if (!E) 64353358Sdim return errorToErrorCode(E.takeError()); 65353358Sdim OpdExtractor.reset(new DataExtractor(*E, Obj->isLittleEndian(), 66292915Sdim Obj->getBytesInAddress())); 67292915Sdim OpdAddress = Section->getAddress(); 68292915Sdim break; 69292915Sdim } 70292915Sdim } 71292915Sdim } 72292915Sdim std::vector<std::pair<SymbolRef, uint64_t>> Symbols = 73292915Sdim computeSymbolSizes(*Obj); 74292915Sdim for (auto &P : Symbols) 75292915Sdim res->addSymbol(P.first, P.second, OpdExtractor.get(), OpdAddress); 76292915Sdim 77292915Sdim // If this is a COFF object and we didn't find any symbols, try the export 78292915Sdim // table. 79292915Sdim if (Symbols.empty()) { 80292915Sdim if (auto *CoffObj = dyn_cast<COFFObjectFile>(Obj)) 81292915Sdim if (auto EC = res->addCoffExportSymbols(CoffObj)) 82292915Sdim return EC; 83292915Sdim } 84353358Sdim 85353358Sdim std::vector<std::pair<SymbolDesc, StringRef>> &Fs = res->Functions, 86353358Sdim &Os = res->Objects; 87353358Sdim auto Uniquify = [](std::vector<std::pair<SymbolDesc, StringRef>> &S) { 88353358Sdim // Sort by (Addr,Size,Name). If several SymbolDescs share the same Addr, 89353358Sdim // pick the one with the largest Size. This helps us avoid symbols with no 90353358Sdim // size information (Size=0). 91353358Sdim llvm::sort(S); 92353358Sdim auto I = S.begin(), E = S.end(), J = S.begin(); 93353358Sdim while (I != E) { 94353358Sdim auto OI = I; 95353358Sdim while (++I != E && OI->first.Addr == I->first.Addr) { 96353358Sdim } 97353358Sdim *J++ = I[-1]; 98353358Sdim } 99353358Sdim S.erase(J, S.end()); 100353358Sdim }; 101353358Sdim Uniquify(Fs); 102353358Sdim Uniquify(Os); 103353358Sdim 104292915Sdim return std::move(res); 105292915Sdim} 106292915Sdim 107353358SdimSymbolizableObjectFile::SymbolizableObjectFile(const ObjectFile *Obj, 108360784Sdim std::unique_ptr<DIContext> DICtx, 109360784Sdim bool UntagAddresses) 110360784Sdim : Module(Obj), DebugInfoContext(std::move(DICtx)), 111360784Sdim UntagAddresses(UntagAddresses) {} 112292915Sdim 113292915Sdimnamespace { 114321369Sdim 115292915Sdimstruct OffsetNamePair { 116292915Sdim uint32_t Offset; 117292915Sdim StringRef Name; 118321369Sdim 119292915Sdim bool operator<(const OffsetNamePair &R) const { 120292915Sdim return Offset < R.Offset; 121292915Sdim } 122292915Sdim}; 123292915Sdim 124321369Sdim} // end anonymous namespace 125321369Sdim 126292915Sdimstd::error_code SymbolizableObjectFile::addCoffExportSymbols( 127292915Sdim const COFFObjectFile *CoffObj) { 128292915Sdim // Get all export names and offsets. 129292915Sdim std::vector<OffsetNamePair> ExportSyms; 130292915Sdim for (const ExportDirectoryEntryRef &Ref : CoffObj->export_directories()) { 131292915Sdim StringRef Name; 132292915Sdim uint32_t Offset; 133292915Sdim if (auto EC = Ref.getSymbolName(Name)) 134292915Sdim return EC; 135292915Sdim if (auto EC = Ref.getExportRVA(Offset)) 136292915Sdim return EC; 137292915Sdim ExportSyms.push_back(OffsetNamePair{Offset, Name}); 138292915Sdim } 139292915Sdim if (ExportSyms.empty()) 140292915Sdim return std::error_code(); 141292915Sdim 142292915Sdim // Sort by ascending offset. 143292915Sdim array_pod_sort(ExportSyms.begin(), ExportSyms.end()); 144292915Sdim 145292915Sdim // Approximate the symbol sizes by assuming they run to the next symbol. 146292915Sdim // FIXME: This assumes all exports are functions. 147292915Sdim uint64_t ImageBase = CoffObj->getImageBase(); 148292915Sdim for (auto I = ExportSyms.begin(), E = ExportSyms.end(); I != E; ++I) { 149292915Sdim OffsetNamePair &Export = *I; 150292915Sdim // FIXME: The last export has a one byte size now. 151292915Sdim uint32_t NextOffset = I != E ? I->Offset : Export.Offset + 1; 152292915Sdim uint64_t SymbolStart = ImageBase + Export.Offset; 153292915Sdim uint64_t SymbolSize = NextOffset - Export.Offset; 154292915Sdim SymbolDesc SD = {SymbolStart, SymbolSize}; 155353358Sdim Functions.emplace_back(SD, Export.Name); 156292915Sdim } 157292915Sdim return std::error_code(); 158292915Sdim} 159292915Sdim 160292915Sdimstd::error_code SymbolizableObjectFile::addSymbol(const SymbolRef &Symbol, 161292915Sdim uint64_t SymbolSize, 162292915Sdim DataExtractor *OpdExtractor, 163292915Sdim uint64_t OpdAddress) { 164353358Sdim // Avoid adding symbols from an unknown/undefined section. 165353358Sdim const ObjectFile *Obj = Symbol.getObject(); 166353358Sdim Expected<section_iterator> Sec = Symbol.getSection(); 167353358Sdim if (!Sec || (Obj && Obj->section_end() == *Sec)) 168353358Sdim return std::error_code(); 169309124Sdim Expected<SymbolRef::Type> SymbolTypeOrErr = Symbol.getType(); 170309124Sdim if (!SymbolTypeOrErr) 171309124Sdim return errorToErrorCode(SymbolTypeOrErr.takeError()); 172309124Sdim SymbolRef::Type SymbolType = *SymbolTypeOrErr; 173292915Sdim if (SymbolType != SymbolRef::ST_Function && SymbolType != SymbolRef::ST_Data) 174292915Sdim return std::error_code(); 175309124Sdim Expected<uint64_t> SymbolAddressOrErr = Symbol.getAddress(); 176309124Sdim if (!SymbolAddressOrErr) 177309124Sdim return errorToErrorCode(SymbolAddressOrErr.takeError()); 178292915Sdim uint64_t SymbolAddress = *SymbolAddressOrErr; 179360784Sdim if (UntagAddresses) { 180360784Sdim // For kernel addresses, bits 56-63 need to be set, so we sign extend bit 55 181360784Sdim // into bits 56-63 instead of masking them out. 182360784Sdim SymbolAddress &= (1ull << 56) - 1; 183360784Sdim SymbolAddress = (int64_t(SymbolAddress) << 8) >> 8; 184360784Sdim } 185292915Sdim if (OpdExtractor) { 186292915Sdim // For big-endian PowerPC64 ELF, symbols in the .opd section refer to 187292915Sdim // function descriptors. The first word of the descriptor is a pointer to 188292915Sdim // the function's code. 189292915Sdim // For the purposes of symbolization, pretend the symbol's address is that 190292915Sdim // of the function's code, not the descriptor. 191292915Sdim uint64_t OpdOffset = SymbolAddress - OpdAddress; 192360784Sdim if (OpdExtractor->isValidOffsetForAddress(OpdOffset)) 193360784Sdim SymbolAddress = OpdExtractor->getAddress(&OpdOffset); 194292915Sdim } 195309124Sdim Expected<StringRef> SymbolNameOrErr = Symbol.getName(); 196309124Sdim if (!SymbolNameOrErr) 197309124Sdim return errorToErrorCode(SymbolNameOrErr.takeError()); 198292915Sdim StringRef SymbolName = *SymbolNameOrErr; 199292915Sdim // Mach-O symbol table names have leading underscore, skip it. 200321369Sdim if (Module->isMachO() && !SymbolName.empty() && SymbolName[0] == '_') 201292915Sdim SymbolName = SymbolName.drop_front(); 202292915Sdim // FIXME: If a function has alias, there are two entries in symbol table 203292915Sdim // with same address size. Make sure we choose the correct one. 204292915Sdim auto &M = SymbolType == SymbolRef::ST_Function ? Functions : Objects; 205292915Sdim SymbolDesc SD = { SymbolAddress, SymbolSize }; 206353358Sdim M.emplace_back(SD, SymbolName); 207292915Sdim return std::error_code(); 208292915Sdim} 209292915Sdim 210292915Sdim// Return true if this is a 32-bit x86 PE COFF module. 211292915Sdimbool SymbolizableObjectFile::isWin32Module() const { 212292915Sdim auto *CoffObject = dyn_cast<COFFObjectFile>(Module); 213292915Sdim return CoffObject && CoffObject->getMachine() == COFF::IMAGE_FILE_MACHINE_I386; 214292915Sdim} 215292915Sdim 216292915Sdimuint64_t SymbolizableObjectFile::getModulePreferredBase() const { 217292915Sdim if (auto *CoffObject = dyn_cast<COFFObjectFile>(Module)) 218292915Sdim return CoffObject->getImageBase(); 219292915Sdim return 0; 220292915Sdim} 221292915Sdim 222292915Sdimbool SymbolizableObjectFile::getNameFromSymbolTable(SymbolRef::Type Type, 223292915Sdim uint64_t Address, 224292915Sdim std::string &Name, 225292915Sdim uint64_t &Addr, 226292915Sdim uint64_t &Size) const { 227353358Sdim const auto &Symbols = Type == SymbolRef::ST_Function ? Functions : Objects; 228353358Sdim std::pair<SymbolDesc, StringRef> SD{{Address, UINT64_C(-1)}, StringRef()}; 229353358Sdim auto SymbolIterator = llvm::upper_bound(Symbols, SD); 230353358Sdim if (SymbolIterator == Symbols.begin()) 231292915Sdim return false; 232292915Sdim --SymbolIterator; 233292915Sdim if (SymbolIterator->first.Size != 0 && 234292915Sdim SymbolIterator->first.Addr + SymbolIterator->first.Size <= Address) 235292915Sdim return false; 236292915Sdim Name = SymbolIterator->second.str(); 237292915Sdim Addr = SymbolIterator->first.Addr; 238292915Sdim Size = SymbolIterator->first.Size; 239292915Sdim return true; 240292915Sdim} 241292915Sdim 242292915Sdimbool SymbolizableObjectFile::shouldOverrideWithSymbolTable( 243292915Sdim FunctionNameKind FNKind, bool UseSymbolTable) const { 244292915Sdim // When DWARF is used with -gline-tables-only / -gmlt, the symbol table gives 245292915Sdim // better answers for linkage names than the DIContext. Otherwise, we are 246292915Sdim // probably using PEs and PDBs, and we shouldn't do the override. PE files 247292915Sdim // generally only contain the names of exported symbols. 248292915Sdim return FNKind == FunctionNameKind::LinkageName && UseSymbolTable && 249292915Sdim isa<DWARFContext>(DebugInfoContext.get()); 250292915Sdim} 251292915Sdim 252353358SdimDILineInfo 253353358SdimSymbolizableObjectFile::symbolizeCode(object::SectionedAddress ModuleOffset, 254353358Sdim FunctionNameKind FNKind, 255353358Sdim bool UseSymbolTable) const { 256353358Sdim if (ModuleOffset.SectionIndex == object::SectionedAddress::UndefSection) 257353358Sdim ModuleOffset.SectionIndex = 258353358Sdim getModuleSectionIndexForAddress(ModuleOffset.Address); 259353358Sdim DILineInfo LineInfo = DebugInfoContext->getLineInfoForAddress( 260353358Sdim ModuleOffset, getDILineInfoSpecifier(FNKind)); 261353358Sdim 262292915Sdim // Override function name from symbol table if necessary. 263292915Sdim if (shouldOverrideWithSymbolTable(FNKind, UseSymbolTable)) { 264292915Sdim std::string FunctionName; 265292915Sdim uint64_t Start, Size; 266353358Sdim if (getNameFromSymbolTable(SymbolRef::ST_Function, ModuleOffset.Address, 267292915Sdim FunctionName, Start, Size)) { 268292915Sdim LineInfo.FunctionName = FunctionName; 269292915Sdim } 270292915Sdim } 271292915Sdim return LineInfo; 272292915Sdim} 273292915Sdim 274292915SdimDIInliningInfo SymbolizableObjectFile::symbolizeInlinedCode( 275353358Sdim object::SectionedAddress ModuleOffset, FunctionNameKind FNKind, 276353358Sdim bool UseSymbolTable) const { 277353358Sdim if (ModuleOffset.SectionIndex == object::SectionedAddress::UndefSection) 278353358Sdim ModuleOffset.SectionIndex = 279353358Sdim getModuleSectionIndexForAddress(ModuleOffset.Address); 280353358Sdim DIInliningInfo InlinedContext = DebugInfoContext->getInliningInfoForAddress( 281353358Sdim ModuleOffset, getDILineInfoSpecifier(FNKind)); 282292915Sdim 283292915Sdim // Make sure there is at least one frame in context. 284292915Sdim if (InlinedContext.getNumberOfFrames() == 0) 285292915Sdim InlinedContext.addFrame(DILineInfo()); 286292915Sdim 287292915Sdim // Override the function name in lower frame with name from symbol table. 288292915Sdim if (shouldOverrideWithSymbolTable(FNKind, UseSymbolTable)) { 289292915Sdim std::string FunctionName; 290292915Sdim uint64_t Start, Size; 291353358Sdim if (getNameFromSymbolTable(SymbolRef::ST_Function, ModuleOffset.Address, 292292915Sdim FunctionName, Start, Size)) { 293292915Sdim InlinedContext.getMutableFrame(InlinedContext.getNumberOfFrames() - 1) 294292915Sdim ->FunctionName = FunctionName; 295292915Sdim } 296292915Sdim } 297292915Sdim 298292915Sdim return InlinedContext; 299292915Sdim} 300292915Sdim 301353358SdimDIGlobal SymbolizableObjectFile::symbolizeData( 302353358Sdim object::SectionedAddress ModuleOffset) const { 303292915Sdim DIGlobal Res; 304353358Sdim getNameFromSymbolTable(SymbolRef::ST_Data, ModuleOffset.Address, Res.Name, 305353358Sdim Res.Start, Res.Size); 306292915Sdim return Res; 307292915Sdim} 308353358Sdim 309353358Sdimstd::vector<DILocal> SymbolizableObjectFile::symbolizeFrame( 310353358Sdim object::SectionedAddress ModuleOffset) const { 311353358Sdim if (ModuleOffset.SectionIndex == object::SectionedAddress::UndefSection) 312353358Sdim ModuleOffset.SectionIndex = 313353358Sdim getModuleSectionIndexForAddress(ModuleOffset.Address); 314353358Sdim return DebugInfoContext->getLocalsForAddress(ModuleOffset); 315353358Sdim} 316353358Sdim 317353358Sdim/// Search for the first occurence of specified Address in ObjectFile. 318353358Sdimuint64_t SymbolizableObjectFile::getModuleSectionIndexForAddress( 319353358Sdim uint64_t Address) const { 320353358Sdim 321353358Sdim for (SectionRef Sec : Module->sections()) { 322353358Sdim if (!Sec.isText() || Sec.isVirtual()) 323353358Sdim continue; 324353358Sdim 325353358Sdim if (Address >= Sec.getAddress() && 326353358Sdim Address < Sec.getAddress() + Sec.getSize()) 327353358Sdim return Sec.getIndex(); 328353358Sdim } 329353358Sdim 330353358Sdim return object::SectionedAddress::UndefSection; 331353358Sdim} 332