tools/lld/lib/ReaderWriter/MachO/MachONormalizedFileToAtoms.cpp

reference, declaration → definition
definition → references, declarations, derived classes, virtual overrides
reference to multiple definitions → definitions
unreferenced

//===- lib/ReaderWriter/MachO/MachONormalizedFileToAtoms.cpp --------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//

///
/// \file Converts from in-memory normalized mach-o to in-memory Atoms.
///
///                  +------------+
///                  | normalized |
///                  +------------+
///                        |
///                        |
///                        v
///                    +-------+
///                    | Atoms |
///                    +-------+

#include "ArchHandler.h"
#include "Atoms.h"
#include "File.h"
#include "MachONormalizedFile.h"
#include "MachONormalizedFileBinaryUtils.h"
#include "lld/Common/LLVM.h"
#include "lld/Core/Error.h"
#include "llvm/BinaryFormat/Dwarf.h"
#include "llvm/BinaryFormat/MachO.h"
#include "llvm/DebugInfo/DWARF/DWARFFormValue.h"
#include "llvm/Support/DataExtractor.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/Error.h"
#include "llvm/Support/Format.h"
#include "llvm/Support/LEB128.h"
#include "llvm/Support/raw_ostream.h"

using namespace llvm::MachO;
using namespace lld::mach_o::normalized;

#define DEBUG_TYPE "normalized-file-to-atoms"

namespace lld {
namespace mach_o {


namespace { // anonymous


#define ENTRY(seg, sect, type, atomType) \
  {seg, sect, type, DefinedAtom::atomType }

struct MachORelocatableSectionToAtomType {
  StringRef                 segmentName;
  StringRef                 sectionName;
  SectionType               sectionType;
  DefinedAtom::ContentType  atomType;
};

const MachORelocatableSectionToAtomType sectsToAtomType[] = {
  ENTRY("__TEXT", "__text",           S_REGULAR,          typeCode),
  ENTRY("__TEXT", "__text",           S_REGULAR,          typeResolver),
  ENTRY("__TEXT", "__cstring",        S_CSTRING_LITERALS, typeCString),
  ENTRY("",       "",                 S_CSTRING_LITERALS, typeCString),
  ENTRY("__TEXT", "__ustring",        S_REGULAR,          typeUTF16String),
  ENTRY("__TEXT", "__const",          S_REGULAR,          typeConstant),
  ENTRY("__TEXT", "__const_coal",     S_COALESCED,        typeConstant),
  ENTRY("__TEXT", "__eh_frame",       S_COALESCED,        typeCFI),
  ENTRY("__TEXT", "__eh_frame",       S_REGULAR,          typeCFI),
  ENTRY("__TEXT", "__literal4",       S_4BYTE_LITERALS,   typeLiteral4),
  ENTRY("__TEXT", "__literal8",       S_8BYTE_LITERALS,   typeLiteral8),
  ENTRY("__TEXT", "__literal16",      S_16BYTE_LITERALS,  typeLiteral16),
  ENTRY("__TEXT", "__gcc_except_tab", S_REGULAR,          typeLSDA),
  ENTRY("__DATA", "__data",           S_REGULAR,          typeData),
  ENTRY("__DATA", "__datacoal_nt",    S_COALESCED,        typeData),
  ENTRY("__DATA", "__const",          S_REGULAR,          typeConstData),
  ENTRY("__DATA", "__cfstring",       S_REGULAR,          typeCFString),
  ENTRY("__DATA", "__mod_init_func",  S_MOD_INIT_FUNC_POINTERS,
                                                          typeInitializerPtr),
  ENTRY("__DATA", "__mod_term_func",  S_MOD_TERM_FUNC_POINTERS,
                                                          typeTerminatorPtr),
  ENTRY("__DATA", "__got",            S_NON_LAZY_SYMBOL_POINTERS,
                                                          typeGOT),
  ENTRY("__DATA", "__bss",            S_ZEROFILL,         typeZeroFill),
  ENTRY("",       "",                 S_NON_LAZY_SYMBOL_POINTERS,
                                                          typeGOT),
  ENTRY("__DATA", "__interposing",    S_INTERPOSING,      typeInterposingTuples),
  ENTRY("__DATA", "__thread_vars",    S_THREAD_LOCAL_VARIABLES,
                                                          typeThunkTLV),
  ENTRY("__DATA", "__thread_data", S_THREAD_LOCAL_REGULAR, typeTLVInitialData),
  ENTRY("__DATA", "__thread_bss",     S_THREAD_LOCAL_ZEROFILL,
                                                        typeTLVInitialZeroFill),
  ENTRY("__DATA", "__objc_imageinfo", S_REGULAR,          typeObjCImageInfo),
  ENTRY("__DATA", "__objc_catlist",   S_REGULAR,          typeObjC2CategoryList),
  ENTRY("",       "",                 S_INTERPOSING,      typeInterposingTuples),
  ENTRY("__LD",   "__compact_unwind", S_REGULAR,
                                                         typeCompactUnwindInfo),
  ENTRY("",       "",                 S_REGULAR,          typeUnknown)
};
#undef ENTRY


/// Figures out ContentType of a mach-o section.
DefinedAtom::ContentType atomTypeFromSection(const Section &section,
                                             bool &customSectionName) {
  // First look for match of name and type. Empty names in table are wildcards.
  customSectionName = false;
  for (const MachORelocatableSectionToAtomType *p = sectsToAtomType ;
                                 p->atomType != DefinedAtom::typeUnknown; ++p) {
    if (p->sectionType != section.type)
      continue;
    if (!p->segmentName.equals(section.segmentName) && !p->segmentName.empty())
      continue;
    if (!p->sectionName.equals(section.sectionName) && !p->sectionName.empty())
      continue;
    customSectionName = p->segmentName.empty() && p->sectionName.empty();
    return p->atomType;
  }
  // Look for code denoted by section attributes
  if (section.attributes & S_ATTR_PURE_INSTRUCTIONS)
    return DefinedAtom::typeCode;

  return DefinedAtom::typeUnknown;
}

enum AtomizeModel {
  atomizeAtSymbols,
  atomizeFixedSize,
  atomizePointerSize,
  atomizeUTF8,
  atomizeUTF16,
  atomizeCFI,
  atomizeCU,
  atomizeCFString
};

/// Returns info on how to atomize a section of the specified ContentType.
void sectionParseInfo(DefinedAtom::ContentType atomType,
                      unsigned int &sizeMultiple,
                      DefinedAtom::Scope &scope,
                      DefinedAtom::Merge &merge,
                      AtomizeModel &atomizeModel) {
  struct ParseInfo {
    DefinedAtom::ContentType  atomType;
    unsigned int              sizeMultiple;
    DefinedAtom::Scope        scope;
    DefinedAtom::Merge        merge;
    AtomizeModel              atomizeModel;
  };

  #define ENTRY(type, size, scope, merge, model) \
    {DefinedAtom::type, size, DefinedAtom::scope, DefinedAtom::merge, model }

  static const ParseInfo parseInfo[] = {
    ENTRY(typeCode,              1, scopeGlobal,          mergeNo,
                                                            atomizeAtSymbols),
    ENTRY(typeData,              1, scopeGlobal,          mergeNo,
                                                            atomizeAtSymbols),
    ENTRY(typeConstData,         1, scopeGlobal,          mergeNo,
                                                            atomizeAtSymbols),
    ENTRY(typeZeroFill,          1, scopeGlobal,          mergeNo,
                                                            atomizeAtSymbols),
    ENTRY(typeConstant,          1, scopeGlobal,          mergeNo,
                                                            atomizeAtSymbols),
    ENTRY(typeCString,           1, scopeLinkageUnit,     mergeByContent,
                                                            atomizeUTF8),
    ENTRY(typeUTF16String,       1, scopeLinkageUnit,     mergeByContent,
                                                            atomizeUTF16),
    ENTRY(typeCFI,               4, scopeTranslationUnit, mergeNo,
                                                            atomizeCFI),
    ENTRY(typeLiteral4,          4, scopeLinkageUnit,     mergeByContent,
                                                            atomizeFixedSize),
    ENTRY(typeLiteral8,          8, scopeLinkageUnit,     mergeByContent,
                                                            atomizeFixedSize),
    ENTRY(typeLiteral16,        16, scopeLinkageUnit,     mergeByContent,
                                                            atomizeFixedSize),
    ENTRY(typeCFString,          4, scopeLinkageUnit,     mergeByContent,
                                                            atomizeCFString),
    ENTRY(typeInitializerPtr,    4, scopeTranslationUnit, mergeNo,
                                                            atomizePointerSize),
    ENTRY(typeTerminatorPtr,     4, scopeTranslationUnit, mergeNo,
                                                            atomizePointerSize),
    ENTRY(typeCompactUnwindInfo, 4, scopeTranslationUnit, mergeNo,
                                                            atomizeCU),
    ENTRY(typeGOT,               4, scopeLinkageUnit,     mergeByContent,
                                                            atomizePointerSize),
    ENTRY(typeObjC2CategoryList, 4, scopeTranslationUnit, mergeByContent,
                                                            atomizePointerSize),
    ENTRY(typeUnknown,           1, scopeGlobal,          mergeNo,
                                                            atomizeAtSymbols)
  };
  #undef ENTRY
  const int tableLen = sizeof(parseInfo) / sizeof(ParseInfo);
  for (int i=0; i < tableLen; ++i) {
    if (parseInfo[i].atomType == atomType) {
      sizeMultiple = parseInfo[i].sizeMultiple;
      scope        = parseInfo[i].scope;
      merge        = parseInfo[i].merge;
      atomizeModel = parseInfo[i].atomizeModel;
      return;
    }
  }

  // Unknown type is atomized by symbols.
  sizeMultiple = 1;
  scope = DefinedAtom::scopeGlobal;
  merge = DefinedAtom::mergeNo;
  atomizeModel = atomizeAtSymbols;
}


Atom::Scope atomScope(uint8_t scope) {
  switch (scope) {
  case N_EXT:
    return Atom::scopeGlobal;
  case N_PEXT:
  case N_PEXT | N_EXT:
    return Atom::scopeLinkageUnit;
  case 0:
    return Atom::scopeTranslationUnit;
  }
  llvm_unreachable("unknown scope value!");
}

void appendSymbolsInSection(const std::vector<Symbol> &inSymbols,
                            uint32_t sectionIndex,
                            SmallVector<const Symbol *, 64> &outSyms) {
  for (const Symbol &sym : inSymbols) {
    // Only look at definition symbols.
    if ((sym.type & N_TYPE) != N_SECT)
      continue;
    if (sym.sect != sectionIndex)
      continue;
    outSyms.push_back(&sym);
  }
}

void atomFromSymbol(DefinedAtom::ContentType atomType, const Section &section,
                    MachOFile &file, uint64_t symbolAddr, StringRef symbolName,
                    uint16_t symbolDescFlags, Atom::Scope symbolScope,
                    uint64_t nextSymbolAddr, bool scatterable, bool copyRefs) {
  // Mach-O symbol table does have size in it. Instead the size is the
  // difference between this and the next symbol.
  uint64_t size = nextSymbolAddr - symbolAddr;
  uint64_t offset = symbolAddr - section.address;
  bool noDeadStrip = (symbolDescFlags & N_NO_DEAD_STRIP) || !scatterable;
  if (isZeroFillSection(section.type)) {
    file.addZeroFillDefinedAtom(symbolName, symbolScope, offset, size,
                                noDeadStrip, copyRefs, &section);
  } else {
    DefinedAtom::Merge merge = (symbolDescFlags & N_WEAK_DEF)
                              ? DefinedAtom::mergeAsWeak : DefinedAtom::mergeNo;
    bool thumb = (symbolDescFlags & N_ARM_THUMB_DEF);
    if (atomType == DefinedAtom::typeUnknown) {
      // Mach-O needs a segment and section name.  Concatentate those two
      // with a / separator (e.g. "seg/sect") to fit into the lld model
      // of just a section name.
      std::string segSectName = section.segmentName.str()
                                + "/" + section.sectionName.str();
      file.addDefinedAtomInCustomSection(symbolName, symbolScope, atomType,
                                         merge, thumb, noDeadStrip, offset,
                                         size, segSectName, true, &section);
    } else {
      if ((atomType == lld::DefinedAtom::typeCode) &&
          (symbolDescFlags & N_SYMBOL_RESOLVER)) {
        atomType = lld::DefinedAtom::typeResolver;
      }
      file.addDefinedAtom(symbolName, symbolScope, atomType, merge,
                          offset, size, thumb, noDeadStrip, copyRefs, &section);
    }
  }
}

llvm::Error processSymboledSection(DefinedAtom::ContentType atomType,
                                   const Section &section,
                                   const NormalizedFile &normalizedFile,
                                   MachOFile &file, bool scatterable,
                                   bool copyRefs) {
  // Find section's index.
  uint32_t sectIndex = 1;
  for (auto &sect : normalizedFile.sections) {
    if (&sect == &section)
      break;
    ++sectIndex;
  }

  // Find all symbols in this section.
  SmallVector<const Symbol *, 64> symbols;
  appendSymbolsInSection(normalizedFile.globalSymbols, sectIndex, symbols);
  appendSymbolsInSection(normalizedFile.localSymbols,  sectIndex, symbols);

  // Sort symbols.
  std::sort(symbols.begin(), symbols.end(),
            [](const Symbol *lhs, const Symbol *rhs) -> bool {
              if (lhs == rhs)
                return false;
              // First by address.
              uint64_t lhsAddr = lhs->value;
              uint64_t rhsAddr = rhs->value;
              if (lhsAddr != rhsAddr)
                return lhsAddr < rhsAddr;
               // If same address, one is an alias so sort by scope.
              Atom::Scope lScope = atomScope(lhs->scope);
              Atom::Scope rScope = atomScope(rhs->scope);
              if (lScope != rScope)
                return lScope < rScope;
              // If same address and scope, see if one might be better as
              // the alias.
              bool lPrivate = (lhs->name.front() == 'l');
              bool rPrivate = (rhs->name.front() == 'l');
              if (lPrivate != rPrivate)
                return lPrivate;
              // If same address and scope, sort by name.
              return lhs->name < rhs->name;
            });

  // Debug logging of symbols.
  //for (const Symbol *sym : symbols)
  //  llvm::errs() << "  sym: "
  //    << llvm::format("0x%08llx ", (uint64_t)sym->value)
  //    << ", " << sym->name << "\n";

  // If section has no symbols and no content, there are no atoms.
  if (symbols.empty() && section.content.empty())
    return llvm::Error::success();

  if (symbols.empty()) {
    // Section has no symbols, put all content in one anoymous atom.
    atomFromSymbol(atomType, section, file, section.address, StringRef(),
                  0, Atom::scopeTranslationUnit,
                  section.address + section.content.size(),
                  scatterable, copyRefs);
  }
  else if (symbols.front()->value != section.address) {
    // Section has anonymous content before first symbol.
    atomFromSymbol(atomType, section, file, section.address, StringRef(),
                   0, Atom::scopeTranslationUnit, symbols.front()->value,
                   scatterable, copyRefs);
  }

  const Symbol *lastSym = nullptr;
  for (const Symbol *sym : symbols) {
    if (lastSym != nullptr) {
      // Ignore any assembler added "ltmpNNN" symbol at start of section
      // if there is another symbol at the start.
      if ((lastSym->value != sym->value)
          || lastSym->value != section.address
          || !lastSym->name.startswith("ltmp")) {
        atomFromSymbol(atomType, section, file, lastSym->value, lastSym->name,
                       lastSym->desc, atomScope(lastSym->scope), sym->value,
                       scatterable, copyRefs);
      }
    }
    lastSym = sym;
  }
  if (lastSym != nullptr) {
    atomFromSymbol(atomType, section, file, lastSym->value, lastSym->name,
                   lastSym->desc, atomScope(lastSym->scope),
                   section.address + section.content.size(),
                   scatterable, copyRefs);
  }

  // If object built without .subsections_via_symbols, add reference chain.
  if (!scatterable) {
    MachODefinedAtom *prevAtom = nullptr;
    file.eachAtomInSection(section,
                           [&](MachODefinedAtom *atom, uint64_t offset)->void {
      if (prevAtom)
        prevAtom->addReference(Reference::KindNamespace::all,
                               Reference::KindArch::all,
                               Reference::kindLayoutAfter, 0, atom, 0);
      prevAtom = atom;
    });
  }

  return llvm::Error::success();
}

llvm::Error processSection(DefinedAtom::ContentType atomType,
                           const Section &section,
                           bool customSectionName,
                           const NormalizedFile &normalizedFile,
                           MachOFile &file, bool scatterable,
                           bool copyRefs) {
  const bool is64 = MachOLinkingContext::is64Bit(normalizedFile.arch);
  const bool isBig = MachOLinkingContext::isBigEndian(normalizedFile.arch);

  // Get info on how to atomize section.
  unsigned int       sizeMultiple;
  DefinedAtom::Scope scope;
  DefinedAtom::Merge merge;
  AtomizeModel       atomizeModel;
  sectionParseInfo(atomType, sizeMultiple, scope, merge, atomizeModel);

  // Validate section size.
  if ((section.content.size() % sizeMultiple) != 0)
    return llvm::make_error<GenericError>(Twine("Section ")
                                          + section.segmentName
                                          + "/" + section.sectionName
                                          + " has size ("
                                          + Twine(section.content.size())
                                          + ") which is not a multiple of "
                                          + Twine(sizeMultiple));

  if (atomizeModel == atomizeAtSymbols) {
    // Break section up into atoms each with a fixed size.
    return processSymboledSection(atomType, section, normalizedFile, file,
                                  scatterable, copyRefs);
  } else {
    unsigned int size;
    for (unsigned int offset = 0, e = section.content.size(); offset != e;) {
      switch (atomizeModel) {
      case atomizeFixedSize:
        // Break section up into atoms each with a fixed size.
        size = sizeMultiple;
        break;
      case atomizePointerSize:
        // Break section up into atoms each the size of a pointer.
        size = is64 ? 8 : 4;
        break;
      case atomizeUTF8:
        // Break section up into zero terminated c-strings.
        size = 0;
        for (unsigned int i = offset; i < e; ++i) {
          if (section.content[i] == 0) {
            size = i + 1 - offset;
            break;
          }
        }
        break;
      case atomizeUTF16:
        // Break section up into zero terminated UTF16 strings.
        size = 0;
        for (unsigned int i = offset; i < e; i += 2) {
          if ((section.content[i] == 0) && (section.content[i + 1] == 0)) {
            size = i + 2 - offset;
            break;
          }
        }
        break;
      case atomizeCFI:
        // Break section up into dwarf unwind CFIs (FDE or CIE).
        size = read32(&section.content[offset], isBig) + 4;
        if (offset+size > section.content.size()) {
          return llvm::make_error<GenericError>(Twine("Section ")
                                                + section.segmentName
                                                + "/" + section.sectionName
                                                + " is malformed.  Size of CFI "
                                                "starting at offset ("
                                                + Twine(offset)
                                                + ") is past end of section.");
        }
        break;
      case atomizeCU:
        // Break section up into compact unwind entries.
        size = is64 ? 32 : 20;
        break;
      case atomizeCFString:
        // Break section up into NS/CFString objects.
        size = is64 ? 32 : 16;
        break;
      case atomizeAtSymbols:
        break;
      }
      if (size == 0) {
        return llvm::make_error<GenericError>(Twine("Section ")
                                              + section.segmentName
                                              + "/" + section.sectionName
                                              + " is malformed.  The last atom "
                                              "is not zero terminated.");
      }
      if (customSectionName) {
        // Mach-O needs a segment and section name.  Concatentate those two
        // with a / separator (e.g. "seg/sect") to fit into the lld model
        // of just a section name.
        std::string segSectName = section.segmentName.str()
                                  + "/" + section.sectionName.str();
        file.addDefinedAtomInCustomSection(StringRef(), scope, atomType,
                                           merge, false, false, offset,
                                           size, segSectName, true, &section);
      } else {
        file.addDefinedAtom(StringRef(), scope, atomType, merge, offset, size,
                            false, false, copyRefs, &section);
      }
      offset += size;
    }
  }
  return llvm::Error::success();
}

const Section* findSectionCoveringAddress(const NormalizedFile &normalizedFile,
                                          uint64_t address) {
  for (const Section &s : normalizedFile.sections) {
    uint64_t sAddr = s.address;
    if ((sAddr <= address) && (address < sAddr+s.content.size())) {
      return &s;
    }
  }
  return nullptr;
}

const MachODefinedAtom *
findAtomCoveringAddress(const NormalizedFile &normalizedFile, MachOFile &file,
                        uint64_t addr, Reference::Addend &addend) {
  const Section *sect = nullptr;
  sect = findSectionCoveringAddress(normalizedFile, addr);
  if (!sect)
    return nullptr;

  uint32_t offsetInTarget;
  uint64_t offsetInSect = addr - sect->address;
  auto atom =
      file.findAtomCoveringAddress(*sect, offsetInSect, &offsetInTarget);
  addend = offsetInTarget;
  return atom;
}

// Walks all relocations for a section in a normalized .o file and
// creates corresponding lld::Reference objects.
llvm::Error convertRelocs(const Section &section,
                          const NormalizedFile &normalizedFile,
                          bool scatterable,
                          MachOFile &file,
                          ArchHandler &handler) {
  // Utility function for ArchHandler to find atom by its address.
  auto atomByAddr = [&] (uint32_t sectIndex, uint64_t addr,
                         const lld::Atom **atom, Reference::Addend *addend)
                         -> llvm::Error {
    if (sectIndex > normalizedFile.sections.size())
      return llvm::make_error<GenericError>(Twine("out of range section "
                                     "index (") + Twine(sectIndex) + ")");
    const Section *sect = nullptr;
    if (sectIndex == 0) {
      sect = findSectionCoveringAddress(normalizedFile, addr);
      if (!sect)
        return llvm::make_error<GenericError>(Twine("address (" + Twine(addr)
                                       + ") is not in any section"));
    } else {
      sect = &normalizedFile.sections[sectIndex-1];
    }
    uint32_t offsetInTarget;
    uint64_t offsetInSect = addr - sect->address;
    *atom = file.findAtomCoveringAddress(*sect, offsetInSect, &offsetInTarget);
    *addend = offsetInTarget;
    return llvm::Error::success();
  };

  // Utility function for ArchHandler to find atom by its symbol index.
  auto atomBySymbol = [&] (uint32_t symbolIndex, const lld::Atom **result)
                           -> llvm::Error {
    // Find symbol from index.
    const Symbol *sym = nullptr;
    uint32_t numStabs  = normalizedFile.stabsSymbols.size();
    uint32_t numLocal  = normalizedFile.localSymbols.size();
    uint32_t numGlobal = normalizedFile.globalSymbols.size();
    uint32_t numUndef  = normalizedFile.undefinedSymbols.size();
    assert(symbolIndex >= numStabs && "Searched for stab via atomBySymbol?");
    if (symbolIndex < numStabs+numLocal) {
      sym = &normalizedFile.localSymbols[symbolIndex-numStabs];
    } else if (symbolIndex < numStabs+numLocal+numGlobal) {
      sym = &normalizedFile.globalSymbols[symbolIndex-numStabs-numLocal];
    } else if (symbolIndex < numStabs+numLocal+numGlobal+numUndef) {
      sym = &normalizedFile.undefinedSymbols[symbolIndex-numStabs-numLocal-
                                             numGlobal];
    } else {
      return llvm::make_error<GenericError>(Twine("symbol index (")
                                     + Twine(symbolIndex) + ") out of range");
    }

    // Find atom from symbol.
    if ((sym->type & N_TYPE) == N_SECT) {
      if (sym->sect > normalizedFile.sections.size())
        return llvm::make_error<GenericError>(Twine("symbol section index (")
                                        + Twine(sym->sect) + ") out of range ");
      const Section &symSection = normalizedFile.sections[sym->sect-1];
      uint64_t targetOffsetInSect = sym->value - symSection.address;
      MachODefinedAtom *target = file.findAtomCoveringAddress(symSection,
                                                            targetOffsetInSect);
      if (target) {
        *result = target;
        return llvm::Error::success();
      }
      return llvm::make_error<GenericError>("no atom found for defined symbol");
    } else if ((sym->type & N_TYPE) == N_UNDF) {
      const lld::Atom *target = file.findUndefAtom(sym->name);
      if (target) {
        *result = target;
        return llvm::Error::success();
      }
      return llvm::make_error<GenericError>("no undefined atom found for sym");
    } else {
      // Search undefs
      return llvm::make_error<GenericError>("no atom found for symbol");
    }
  };

  const bool isBig = MachOLinkingContext::isBigEndian(normalizedFile.arch);
  // Use old-school iterator so that paired relocations can be grouped.
  for (auto it=section.relocations.begin(), e=section.relocations.end();
                                                                it != e; ++it) {
    const Relocation &reloc = *it;
    // Find atom this relocation is in.
    if (reloc.offset > section.content.size())
      return llvm::make_error<GenericError>(
                                    Twine("r_address (") + Twine(reloc.offset)
                                    + ") is larger than section size ("
                                    + Twine(section.content.size()) + ")");
    uint32_t offsetInAtom;
    MachODefinedAtom *inAtom = file.findAtomCoveringAddress(section,
                                                            reloc.offset,
                                                            &offsetInAtom);
    assert(inAtom && "r_address in range, should have found atom");
    uint64_t fixupAddress = section.address + reloc.offset;

    const lld::Atom *target = nullptr;
    Reference::Addend addend = 0;
    Reference::KindValue kind;
    if (handler.isPairedReloc(reloc)) {
      // Handle paired relocations together.
      const Relocation &reloc2 = *++it;
      auto relocErr = handler.getPairReferenceInfo(
          reloc, reloc2, inAtom, offsetInAtom, fixupAddress, isBig, scatterable,
          atomByAddr, atomBySymbol, &kind, &target, &addend);
      if (relocErr) {
        return handleErrors(std::move(relocErr),
                            [&](std::unique_ptr<GenericError> GE) {
          return llvm::make_error<GenericError>(
            Twine("bad relocation (") + GE->getMessage()
             + ") in section "
             + section.segmentName + "/" + section.sectionName
             + " (r1_address=" + Twine::utohexstr(reloc.offset)
             + ", r1_type=" + Twine(reloc.type)
             + ", r1_extern=" + Twine(reloc.isExtern)
             + ", r1_length=" + Twine((int)reloc.length)
             + ", r1_pcrel=" + Twine(reloc.pcRel)
             + (!reloc.scattered ? (Twine(", r1_symbolnum=")
                                    + Twine(reloc.symbol))
                                 : (Twine(", r1_scattered=1, r1_value=")
                                    + Twine(reloc.value)))
             + ")"
             + ", (r2_address=" + Twine::utohexstr(reloc2.offset)
             + ", r2_type=" + Twine(reloc2.type)
             + ", r2_extern=" + Twine(reloc2.isExtern)
             + ", r2_length=" + Twine((int)reloc2.length)
             + ", r2_pcrel=" + Twine(reloc2.pcRel)
             + (!reloc2.scattered ? (Twine(", r2_symbolnum=")
                                     + Twine(reloc2.symbol))
                                  : (Twine(", r2_scattered=1, r2_value=")
                                     + Twine(reloc2.value)))
             + ")" );
          });
      }
    }
    else {
      // Use ArchHandler to convert relocation record into information
      // needed to instantiate an lld::Reference object.
      auto relocErr = handler.getReferenceInfo(
          reloc, inAtom, offsetInAtom, fixupAddress, isBig, atomByAddr,
          atomBySymbol, &kind, &target, &addend);
      if (relocErr) {
        return handleErrors(std::move(relocErr),
                            [&](std::unique_ptr<GenericError> GE) {
          return llvm::make_error<GenericError>(
            Twine("bad relocation (") + GE->getMessage()
             + ") in section "
             + section.segmentName + "/" + section.sectionName
             + " (r_address=" + Twine::utohexstr(reloc.offset)
             + ", r_type=" + Twine(reloc.type)
             + ", r_extern=" + Twine(reloc.isExtern)
             + ", r_length=" + Twine((int)reloc.length)
             + ", r_pcrel=" + Twine(reloc.pcRel)
             + (!reloc.scattered ? (Twine(", r_symbolnum=") + Twine(reloc.symbol))
                                 : (Twine(", r_scattered=1, r_value=")
                                    + Twine(reloc.value)))
             + ")" );
          });
      }
    }
    // Instantiate an lld::Reference object and add to its atom.
    inAtom->addReference(Reference::KindNamespace::mach_o,
                         handler.kindArch(),
                         kind, offsetInAtom, target, addend);
  }

  return llvm::Error::success();
}

bool isDebugInfoSection(const Section &section) {
  if ((section.attributes & S_ATTR_DEBUG) == 0)
    return false;
  return section.segmentName.equals("__DWARF");
}

static const Atom* findDefinedAtomByName(MachOFile &file, Twine name) {
  std::string strName = name.str();
  for (auto *atom : file.defined())
    if (atom->name() == strName)
      return atom;
  return nullptr;
}

static StringRef copyDebugString(StringRef str, BumpPtrAllocator &alloc) {
  char *strCopy = alloc.Allocate<char>(str.size() + 1);
  memcpy(strCopy, str.data(), str.size());
  strCopy[str.size()] = '\0';
  return strCopy;
}

llvm::Error parseStabs(MachOFile &file,
                       const NormalizedFile &normalizedFile,
                       bool copyRefs) {

  if (normalizedFile.stabsSymbols.empty())
    return llvm::Error::success();

  // FIXME: Kill this off when we can move to sane yaml parsing.
  std::unique_ptr<BumpPtrAllocator> allocator;
  if (copyRefs)
    allocator = std::make_unique<BumpPtrAllocator>();

  enum { start, inBeginEnd } state = start;

  const Atom *currentAtom = nullptr;
  uint64_t currentAtomAddress = 0;
  StabsDebugInfo::StabsList stabsList;
  for (const auto &stabSym : normalizedFile.stabsSymbols) {
    Stab stab(nullptr, stabSym.type, stabSym.sect, stabSym.desc,
              stabSym.value, stabSym.name);
    switch (state) {
    case start:
      switch (static_cast<StabType>(stabSym.type)) {
      case N_BNSYM:
        state = inBeginEnd;
        currentAtomAddress = stabSym.value;
        Reference::Addend addend;
        currentAtom = findAtomCoveringAddress(normalizedFile, file,
                                              currentAtomAddress, addend);
        if (addend != 0)
          return llvm::make_error<GenericError>(
                   "Non-zero addend for BNSYM '" + stabSym.name + "' in " +
                   file.path());
        if (currentAtom)
          stab.atom = currentAtom;
        else {
          // FIXME: ld64 just issues a warning here - should we match that?
          return llvm::make_error<GenericError>(
                   "can't find atom for stabs BNSYM at " +
                   Twine::utohexstr(stabSym.value) + " in " + file.path());
        }
        break;
      case N_SO:
      case N_OSO:
        // Not associated with an atom, just copy.
        if (copyRefs)
          stab.str = copyDebugString(stabSym.name, *allocator);
        else
          stab.str = stabSym.name;
        break;
      case N_GSYM: {
        auto colonIdx = stabSym.name.find(':');
        if (colonIdx != StringRef::npos) {
          StringRef name = stabSym.name.substr(0, colonIdx);
          currentAtom = findDefinedAtomByName(file, "_" + name);
          stab.atom = currentAtom;
          if (copyRefs)
            stab.str = copyDebugString(stabSym.name, *allocator);
          else
            stab.str = stabSym.name;
        } else {
          currentAtom = findDefinedAtomByName(file, stabSym.name);
          stab.atom = currentAtom;
          if (copyRefs)
            stab.str = copyDebugString(stabSym.name, *allocator);
          else
            stab.str = stabSym.name;
        }
        if (stab.atom == nullptr)
          return llvm::make_error<GenericError>(
                   "can't find atom for N_GSYM stabs" + stabSym.name +
                   " in " + file.path());
        break;
      }
      case N_FUN:
        return llvm::make_error<GenericError>(
                 "old-style N_FUN stab '" + stabSym.name + "' unsupported");
      default:
        return llvm::make_error<GenericError>(
                 "unrecognized stab symbol '" + stabSym.name + "'");
      }
      break;
    case inBeginEnd:
      stab.atom = currentAtom;
      switch (static_cast<StabType>(stabSym.type)) {
      case N_ENSYM:
        state = start;
        currentAtom = nullptr;
        break;
      case N_FUN:
        // Just copy the string.
        if (copyRefs)
          stab.str = copyDebugString(stabSym.name, *allocator);
        else
          stab.str = stabSym.name;
        break;
      default:
        return llvm::make_error<GenericError>(
                 "unrecognized stab symbol '" + stabSym.name + "'");
      }
    }
    llvm::dbgs() << "Adding to stabsList: " << stab << "\n";
    stabsList.push_back(stab);
  }

  file.setDebugInfo(std::make_unique<StabsDebugInfo>(std::move(stabsList)));

  // FIXME: Kill this off when we fix YAML memory ownership.
  file.debugInfo()->setAllocator(std::move(allocator));

  return llvm::Error::success();
}

static llvm::DataExtractor
dataExtractorFromSection(const NormalizedFile &normalizedFile,
                         const Section &S) {
  const bool is64 = MachOLinkingContext::is64Bit(normalizedFile.arch);
  const bool isBig = MachOLinkingContext::isBigEndian(normalizedFile.arch);
  StringRef SecData(reinterpret_cast<const char*>(S.content.data()),
                    S.content.size());
  return llvm::DataExtractor(SecData, !isBig, is64 ? 8 : 4);
}

// FIXME: Cribbed from llvm-dwp -- should share "lightweight CU DIE
//        inspection" code if possible.
static uint64_t getCUAbbrevOffset(llvm::DataExtractor abbrevData,
                                  uint64_t abbrCode) {
  uint64_t curCode;
  uint64_t offset = 0;
  while ((curCode = abbrevData.getULEB128(&offset)) != abbrCode) {
    // Tag
    abbrevData.getULEB128(&offset);
    // DW_CHILDREN
    abbrevData.getU8(&offset);
    // Attributes
    while (abbrevData.getULEB128(&offset) | abbrevData.getULEB128(&offset))
      ;
  }
  return offset;
}

// FIXME: Cribbed from llvm-dwp -- should share "lightweight CU DIE
//        inspection" code if possible.
static Expected<const char *>
getIndexedString(const NormalizedFile &normalizedFile,
                 llvm::dwarf::Form form, llvm::DataExtractor infoData,
                 uint64_t &infoOffset, const Section &stringsSection) {
  if (form == llvm::dwarf::DW_FORM_string)
   return infoData.getCStr(&infoOffset);
  if (form != llvm::dwarf::DW_FORM_strp)
    return llvm::make_error<GenericError>(
        "string field encoded without DW_FORM_strp");
  uint64_t stringOffset = infoData.getU32(&infoOffset);
  llvm::DataExtractor stringsData =
    dataExtractorFromSection(normalizedFile, stringsSection);
  return stringsData.getCStr(&stringOffset);
}

// FIXME: Cribbed from llvm-dwp -- should share "lightweight CU DIE
//        inspection" code if possible.
static llvm::Expected<TranslationUnitSource>
readCompUnit(const NormalizedFile &normalizedFile,
             const Section &info,
             const Section &abbrev,
             const Section &strings,
             StringRef path) {
  // FIXME: Cribbed from llvm-dwp -- should share "lightweight CU DIE
  //        inspection" code if possible.
  uint64_t offset = 0;
  llvm::dwarf::DwarfFormat Format = llvm::dwarf::DwarfFormat::DWARF32;
  auto infoData = dataExtractorFromSection(normalizedFile, info);
  uint32_t length = infoData.getU32(&offset);
  if (length == 0xffffffff) {
    Format = llvm::dwarf::DwarfFormat::DWARF64;
    infoData.getU64(&offset);
  }
  else if (length > 0xffffff00)
    return llvm::make_error<GenericError>("Malformed DWARF in " + path);

  uint16_t version = infoData.getU16(&offset);

  if (version < 2 || version > 4)
    return llvm::make_error<GenericError>("Unsupported DWARF version in " +
                                          path);

  infoData.getU32(&offset); // Abbrev offset (should be zero)
  uint8_t addrSize = infoData.getU8(&offset);

  uint32_t abbrCode = infoData.getULEB128(&offset);
  auto abbrevData = dataExtractorFromSection(normalizedFile, abbrev);
  uint64_t abbrevOffset = getCUAbbrevOffset(abbrevData, abbrCode);
  uint64_t tag = abbrevData.getULEB128(&abbrevOffset);
  if (tag != llvm::dwarf::DW_TAG_compile_unit)
    return llvm::make_error<GenericError>("top level DIE is not a compile unit");
  // DW_CHILDREN
  abbrevData.getU8(&abbrevOffset);
  uint32_t name;
  llvm::dwarf::Form form;
  llvm::dwarf::FormParams formParams = {version, addrSize, Format};
  TranslationUnitSource tu;
  while ((name = abbrevData.getULEB128(&abbrevOffset)) |
         (form = static_cast<llvm::dwarf::Form>(
             abbrevData.getULEB128(&abbrevOffset))) &&
         (name != 0 || form != 0)) {
    switch (name) {
    case llvm::dwarf::DW_AT_name: {
      if (auto eName = getIndexedString(normalizedFile, form, infoData, offset,
                                        strings))
          tu.name = *eName;
      else
        return eName.takeError();
      break;
    }
    case llvm::dwarf::DW_AT_comp_dir: {
      if (auto eName = getIndexedString(normalizedFile, form, infoData, offset,
                                        strings))
        tu.path = *eName;
      else
        return eName.takeError();
      break;
    }
    default:
      llvm::DWARFFormValue::skipValue(form, infoData, &offset, formParams);
    }
  }
  return tu;
}

llvm::Error parseDebugInfo(MachOFile &file,
                           const NormalizedFile &normalizedFile, bool copyRefs) {

  // Find the interesting debug info sections.
  const Section *debugInfo = nullptr;
  const Section *debugAbbrev = nullptr;
  const Section *debugStrings = nullptr;

  for (auto &s : normalizedFile.sections) {
    if (s.segmentName == "__DWARF") {
      if (s.sectionName == "__debug_info")
        debugInfo = &s;
      else if (s.sectionName == "__debug_abbrev")
        debugAbbrev = &s;
      else if (s.sectionName == "__debug_str")
        debugStrings = &s;
    }
  }

  if (!debugInfo)
    return parseStabs(file, normalizedFile, copyRefs);

  if (debugInfo->content.size() == 0)
    return llvm::Error::success();

  if (debugInfo->content.size() < 12)
    return llvm::make_error<GenericError>("Malformed __debug_info section in " +
                                          file.path() + ": too small");

  if (!debugAbbrev)
    return llvm::make_error<GenericError>("Missing __dwarf_abbrev section in " +
                                          file.path());

  if (auto tuOrErr = readCompUnit(normalizedFile, *debugInfo, *debugAbbrev,
                                  *debugStrings, file.path())) {
    // FIXME: Kill of allocator and code under 'copyRefs' when we fix YAML
    //        memory ownership.
    std::unique_ptr<BumpPtrAllocator> allocator;
    if (copyRefs) {
      allocator = std::make_unique<BumpPtrAllocator>();
      tuOrErr->name = copyDebugString(tuOrErr->name, *allocator);
      tuOrErr->path = copyDebugString(tuOrErr->path, *allocator);
    }
    file.setDebugInfo(std::make_unique<DwarfDebugInfo>(std::move(*tuOrErr)));
    if (copyRefs)
      file.debugInfo()->setAllocator(std::move(allocator));
  } else
    return tuOrErr.takeError();

  return llvm::Error::success();
}

static int64_t readSPtr(bool is64, bool isBig, const uint8_t *addr) {
  if (is64)
    return read64(addr, isBig);

  int32_t res = read32(addr, isBig);
  return res;
}

/// --- Augmentation String Processing ---

struct CIEInfo {
  bool _augmentationDataPresent = false;
  bool _mayHaveEH = false;
  uint32_t _offsetOfLSDA = ~0U;
  uint32_t _offsetOfPersonality = ~0U;
  uint32_t _offsetOfFDEPointerEncoding = ~0U;
  uint32_t _augmentationDataLength = ~0U;
};

typedef llvm::DenseMap<const MachODefinedAtom*, CIEInfo> CIEInfoMap;

static llvm::Error processAugmentationString(const uint8_t *augStr,
                                             CIEInfo &cieInfo,
                                             unsigned &len) {

  if (augStr[0] == '\0') {
    len = 1;
    return llvm::Error::success();
  }

  if (augStr[0] != 'z')
    return llvm::make_error<GenericError>("expected 'z' at start of "
                                          "augmentation string");

  cieInfo._augmentationDataPresent = true;
  uint64_t idx = 1;

  uint32_t offsetInAugmentationData = 0;
  while (augStr[idx] != '\0') {
    if (augStr[idx] == 'L') {
      cieInfo._offsetOfLSDA = offsetInAugmentationData;
      // This adds a single byte to the augmentation data.
      ++offsetInAugmentationData;
      ++idx;
      continue;
    }
    if (augStr[idx] == 'P') {
      cieInfo._offsetOfPersonality = offsetInAugmentationData;
      // This adds a single byte to the augmentation data for the encoding,
      // then a number of bytes for the pointer data.
      // FIXME: We are assuming 4 is correct here for the pointer size as we
      // always currently use delta32ToGOT.
      offsetInAugmentationData += 5;
      ++idx;
      continue;
    }
    if (augStr[idx] == 'R') {
      cieInfo._offsetOfFDEPointerEncoding = offsetInAugmentationData;
      // This adds a single byte to the augmentation data.
      ++offsetInAugmentationData;
      ++idx;
      continue;
    }
    if (augStr[idx] == 'e') {
      if (augStr[idx + 1] != 'h')
        return llvm::make_error<GenericError>("expected 'eh' in "
                                              "augmentation string");
      cieInfo._mayHaveEH = true;
      idx += 2;
      continue;
    }
    ++idx;
  }

  cieInfo._augmentationDataLength = offsetInAugmentationData;

  len = idx + 1;
  return llvm::Error::success();
}

static llvm::Error processCIE(const NormalizedFile &normalizedFile,
                              MachOFile &file,
                              mach_o::ArchHandler &handler,
                              const Section *ehFrameSection,
                              MachODefinedAtom *atom,
                              uint64_t offset,
                              CIEInfoMap &cieInfos) {
  const bool isBig = MachOLinkingContext::isBigEndian(normalizedFile.arch);
  const uint8_t *frameData = atom->rawContent().data();

  CIEInfo cieInfo;

  uint32_t size = read32(frameData, isBig);
  uint64_t cieIDField = size == 0xffffffffU
                          ? sizeof(uint32_t) + sizeof(uint64_t)
                          : sizeof(uint32_t);
  uint64_t versionField = cieIDField + sizeof(uint32_t);
  uint64_t augmentationStringField = versionField + sizeof(uint8_t);

  unsigned augmentationStringLength = 0;
  if (auto err = processAugmentationString(frameData + augmentationStringField,
                                           cieInfo, augmentationStringLength))
    return err;

  if (cieInfo._offsetOfPersonality != ~0U) {
    // If we have augmentation data for the personality function, then we may
    // need to implicitly generate its relocation.

    // Parse the EH Data field which is pointer sized.
    uint64_t EHDataField = augmentationStringField + augmentationStringLength;
    const bool is64 = MachOLinkingContext::is64Bit(normalizedFile.arch);
    unsigned EHDataFieldSize = (cieInfo._mayHaveEH ? (is64 ? 8 : 4) : 0);

    // Parse Code Align Factor which is a ULEB128.
    uint64_t CodeAlignField = EHDataField + EHDataFieldSize;
    unsigned lengthFieldSize = 0;
    llvm::decodeULEB128(frameData + CodeAlignField, &lengthFieldSize);

    // Parse Data Align Factor which is a SLEB128.
    uint64_t DataAlignField = CodeAlignField + lengthFieldSize;
    llvm::decodeSLEB128(frameData + DataAlignField, &lengthFieldSize);

    // Parse Return Address Register which is a byte.
    uint64_t ReturnAddressField = DataAlignField + lengthFieldSize;

    // Parse the augmentation length which is a ULEB128.
    uint64_t AugmentationLengthField = ReturnAddressField + 1;
    uint64_t AugmentationLength =
      llvm::decodeULEB128(frameData + AugmentationLengthField,
                          &lengthFieldSize);

    if (AugmentationLength != cieInfo._augmentationDataLength)
      return llvm::make_error<GenericError>("CIE augmentation data length "
                                            "mismatch");

    // Get the start address of the augmentation data.
    uint64_t AugmentationDataField = AugmentationLengthField + lengthFieldSize;

    // Parse the personality function from the augmentation data.
    uint64_t PersonalityField =
      AugmentationDataField + cieInfo._offsetOfPersonality;

    // Parse the personality encoding.
    // FIXME: Verify that this is a 32-bit pcrel offset.
    uint64_t PersonalityFunctionField = PersonalityField + 1;

    if (atom->begin() != atom->end()) {
      // If we have an explicit relocation, then make sure it matches this
      // offset as this is where we'd expect it to be applied to.
      DefinedAtom::reference_iterator CurrentRef = atom->begin();
      if (CurrentRef->offsetInAtom() != PersonalityFunctionField)
        return llvm::make_error<GenericError>("CIE personality reloc at "
                                              "wrong offset");

      if (++CurrentRef != atom->end())
        return llvm::make_error<GenericError>("CIE contains too many relocs");
    } else {
      // Implicitly generate the personality function reloc.  It's assumed to
      // be a delta32 offset to a GOT entry.
      // FIXME: Parse the encoding and check this.
      int32_t funcDelta = read32(frameData + PersonalityFunctionField, isBig);
      uint64_t funcAddress = ehFrameSection->address + offset +
                             PersonalityFunctionField;
      funcAddress += funcDelta;

      const MachODefinedAtom *func = nullptr;
      Reference::Addend addend;
      func = findAtomCoveringAddress(normalizedFile, file, funcAddress,
                                     addend);
      atom->addReference(Reference::KindNamespace::mach_o, handler.kindArch(),
                         handler.unwindRefToPersonalityFunctionKind(),
                         PersonalityFunctionField, func, addend);
    }
  } else if (atom->begin() != atom->end()) {
    // Otherwise, we expect there to be no relocations in this atom as the only
    // relocation would have been to the personality function.
    return llvm::make_error<GenericError>("unexpected relocation in CIE");
  }


  cieInfos[atom] = std::move(cieInfo);

  return llvm::Error::success();
}

static llvm::Error processFDE(const NormalizedFile &normalizedFile,
                              MachOFile &file,
                              mach_o::ArchHandler &handler,
                              const Section *ehFrameSection,
                              MachODefinedAtom *atom,
                              uint64_t offset,
                              const CIEInfoMap &cieInfos) {

  const bool isBig = MachOLinkingContext::isBigEndian(normalizedFile.arch);
  const bool is64 = MachOLinkingContext::is64Bit(normalizedFile.arch);

  // Compiler wasn't lazy and actually told us what it meant.
  // Unfortunately, the compiler may not have generated references for all of
  // [cie, func, lsda] and so we still need to parse the FDE and add references
  // for any the compiler didn't generate.
  if (atom->begin() != atom->end())
    atom->sortReferences();

  DefinedAtom::reference_iterator CurrentRef = atom->begin();

  // This helper returns the reference (if one exists) at the offset we are
  // currently processing.  It automatically increments the ref iterator if we
  // do return a ref, and throws an error if we pass over a ref without
  // comsuming it.
  auto currentRefGetter = [&CurrentRef,
                           &atom](uint64_t Offset)->const Reference* {
    // If there are no more refs found, then we are done.
    if (CurrentRef == atom->end())
      return nullptr;

    const Reference *Ref = *CurrentRef;

    // If we haven't reached the offset for this reference, then return that
    // we don't yet have a reference to process.
    if (Offset < Ref->offsetInAtom())
      return nullptr;

    // If the offset is equal, then we want to process this ref.
    if (Offset == Ref->offsetInAtom()) {
      ++CurrentRef;
      return Ref;
    }

    // The current ref is at an offset which is earlier than the current
    // offset, then we failed to consume it when we should have.  In this case
    // throw an error.
    llvm::report_fatal_error("Skipped reference when processing FDE");
  };

  // Helper to either get the reference at this current location, and verify
  // that it is of the expected type, or add a reference of that type.
  // Returns the reference target.
  auto verifyOrAddReference = [&](uint64_t targetAddress,
                                  Reference::KindValue refKind,
                                  uint64_t refAddress,
                                  bool allowsAddend)->const Atom* {
    if (auto *ref = currentRefGetter(refAddress)) {
      // The compiler already emitted a relocation for the CIE ref.  This should
      // have been converted to the correct type of reference in
      // get[Pair]ReferenceInfo().
      assert(ref->kindValue() == refKind &&
             "Incorrect EHFrame reference kind");
      return ref->target();
    }
    Reference::Addend addend;
    auto *target = findAtomCoveringAddress(normalizedFile, file,
                                           targetAddress, addend);
    atom->addReference(Reference::KindNamespace::mach_o, handler.kindArch(),
                       refKind, refAddress, target, addend);

    if (!allowsAddend)
      assert(!addend && "EHFrame reference cannot have addend");
    return target;
  };

  const uint8_t *startFrameData = atom->rawContent().data();
  const uint8_t *frameData = startFrameData;

  uint32_t size = read32(frameData, isBig);
  uint64_t cieFieldInFDE = size == 0xffffffffU
    ? sizeof(uint32_t) + sizeof(uint64_t)
    : sizeof(uint32_t);

  // Linker needs to fixup a reference from the FDE to its parent CIE (a
  // 32-bit byte offset backwards in the __eh_frame section).
  uint32_t cieDelta = read32(frameData + cieFieldInFDE, isBig);
  uint64_t cieAddress = ehFrameSection->address + offset + cieFieldInFDE;
  cieAddress -= cieDelta;

  auto *cieRefTarget = verifyOrAddReference(cieAddress,
                                            handler.unwindRefToCIEKind(),
                                            cieFieldInFDE, false);
  const MachODefinedAtom *cie = dyn_cast<MachODefinedAtom>(cieRefTarget);
  assert(cie && cie->contentType() == DefinedAtom::typeCFI &&
         "FDE's CIE field does not point at the start of a CIE.");

  const CIEInfo &cieInfo = cieInfos.find(cie)->second;

  // Linker needs to fixup reference from the FDE to the function it's
  // describing. FIXME: there are actually different ways to do this, and the
  // particular method used is specified in the CIE's augmentation fields
  // (hopefully)
  uint64_t rangeFieldInFDE = cieFieldInFDE + sizeof(uint32_t);

  int64_t functionFromFDE = readSPtr(is64, isBig,
                                     frameData + rangeFieldInFDE);
  uint64_t rangeStart = ehFrameSection->address + offset + rangeFieldInFDE;
  rangeStart += functionFromFDE;

  verifyOrAddReference(rangeStart,
                       handler.unwindRefToFunctionKind(),
                       rangeFieldInFDE, true);

  // Handle the augmentation data if there is any.
  if (cieInfo._augmentationDataPresent) {
    // First process the augmentation data length field.
    uint64_t augmentationDataLengthFieldInFDE =
      rangeFieldInFDE + 2 * (is64 ? sizeof(uint64_t) : sizeof(uint32_t));
    unsigned lengthFieldSize = 0;
    uint64_t augmentationDataLength =
      llvm::decodeULEB128(frameData + augmentationDataLengthFieldInFDE,
                          &lengthFieldSize);

    if (cieInfo._offsetOfLSDA != ~0U && augmentationDataLength > 0) {

      // Look at the augmentation data field.
      uint64_t augmentationDataFieldInFDE =
        augmentationDataLengthFieldInFDE + lengthFieldSize;

      int64_t lsdaFromFDE = readSPtr(is64, isBig,
                                     frameData + augmentationDataFieldInFDE);
      uint64_t lsdaStart =
        ehFrameSection->address + offset + augmentationDataFieldInFDE +
        lsdaFromFDE;

      verifyOrAddReference(lsdaStart,
                           handler.unwindRefToFunctionKind(),
                           augmentationDataFieldInFDE, true);
    }
  }

  return llvm::Error::success();
}

llvm::Error addEHFrameReferences(const NormalizedFile &normalizedFile,
                                 MachOFile &file,
                                 mach_o::ArchHandler &handler) {

  const Section *ehFrameSection = nullptr;
  for (auto &section : normalizedFile.sections)
    if (section.segmentName == "__TEXT" &&
        section.sectionName == "__eh_frame") {
      ehFrameSection = &section;
      break;
    }

  // No __eh_frame so nothing to do.
  if (!ehFrameSection)
    return llvm::Error::success();

  llvm::Error ehFrameErr = llvm::Error::success();
  CIEInfoMap cieInfos;

  file.eachAtomInSection(*ehFrameSection,
                         [&](MachODefinedAtom *atom, uint64_t offset) -> void {
    assert(atom->contentType() == DefinedAtom::typeCFI);

    // Bail out if we've encountered an error.
    if (ehFrameErr)
      return;

    const bool isBig = MachOLinkingContext::isBigEndian(normalizedFile.arch);
    if (ArchHandler::isDwarfCIE(isBig, atom))
      ehFrameErr = processCIE(normalizedFile, file, handler, ehFrameSection,
                              atom, offset, cieInfos);
    else
      ehFrameErr = processFDE(normalizedFile, file, handler, ehFrameSection,
                              atom, offset, cieInfos);
  });

  return ehFrameErr;
}

llvm::Error parseObjCImageInfo(const Section &sect,
                               const NormalizedFile &normalizedFile,
                               MachOFile &file) {

  //	struct objc_image_info  {
  //		uint32_t	version;	// initially 0
  //		uint32_t	flags;
  //	};

  ArrayRef<uint8_t> content = sect.content;
  if (content.size() != 8)
    return llvm::make_error<GenericError>(sect.segmentName + "/" +
                                          sect.sectionName +
                                          " in file " + file.path() +
                                          " should be 8 bytes in size");

  const bool isBig = MachOLinkingContext::isBigEndian(normalizedFile.arch);
  uint32_t version = read32(content.data(), isBig);
  if (version)
    return llvm::make_error<GenericError>(sect.segmentName + "/" +
                                          sect.sectionName +
                                          " in file " + file.path() +
                                          " should have version=0");

  uint32_t flags = read32(content.data() + 4, isBig);
  if (flags & (MachOLinkingContext::objc_supports_gc |
               MachOLinkingContext::objc_gc_only))
    return llvm::make_error<GenericError>(sect.segmentName + "/" +
                                          sect.sectionName +
                                          " in file " + file.path() +
                                          " uses GC.  This is not supported");

  if (flags & MachOLinkingContext::objc_retainReleaseForSimulator)
    file.setObjcConstraint(MachOLinkingContext::objc_retainReleaseForSimulator);
  else
    file.setObjcConstraint(MachOLinkingContext::objc_retainRelease);

  file.setSwiftVersion((flags >> 8) & 0xFF);

  return llvm::Error::success();
}

/// Converts normalized mach-o file into an lld::File and lld::Atoms.
llvm::Expected<std::unique_ptr<lld::File>>
objectToAtoms(const NormalizedFile &normalizedFile, StringRef path,
              bool copyRefs) {
  std::unique_ptr<MachOFile> file(new MachOFile(path));
  if (auto ec = normalizedObjectToAtoms(file.get(), normalizedFile, copyRefs))
    return std::move(ec);
  return std::unique_ptr<File>(std::move(file));
}

llvm::Expected<std::unique_ptr<lld::File>>
dylibToAtoms(const NormalizedFile &normalizedFile, StringRef path,
             bool copyRefs) {
  // Instantiate SharedLibraryFile object.
  std::unique_ptr<MachODylibFile> file(new MachODylibFile(path));
  if (auto ec = normalizedDylibToAtoms(file.get(), normalizedFile, copyRefs))
    return std::move(ec);
  return std::unique_ptr<File>(std::move(file));
}

} // anonymous namespace

namespace normalized {

static bool isObjCImageInfo(const Section &sect) {
  return (sect.segmentName == "__OBJC" && sect.sectionName == "__image_info") ||
    (sect.segmentName == "__DATA" && sect.sectionName == "__objc_imageinfo");
}

llvm::Error
normalizedObjectToAtoms(MachOFile *file,
                        const NormalizedFile &normalizedFile,
                        bool copyRefs) {
  LLVM_DEBUG(llvm::dbgs() << "******** Normalizing file to atoms: "
                          << file->path() << "\n");
  bool scatterable = ((normalizedFile.flags & MH_SUBSECTIONS_VIA_SYMBOLS) != 0);

  // Create atoms from each section.
  for (auto &sect : normalizedFile.sections) {

    // If this is a debug-info section parse it specially.
    if (isDebugInfoSection(sect))
      continue;

    // If the file contains an objc_image_info struct, then we should parse the
    // ObjC flags and Swift version.
    if (isObjCImageInfo(sect)) {
      if (auto ec = parseObjCImageInfo(sect, normalizedFile, *file))
        return ec;
      // We then skip adding atoms for this section as we use the ObjCPass to
      // re-emit this data after it has been aggregated for all files.
      continue;
    }

    bool customSectionName;
    DefinedAtom::ContentType atomType = atomTypeFromSection(sect,
                                                            customSectionName);
    if (auto ec =  processSection(atomType, sect, customSectionName,
                                  normalizedFile, *file, scatterable, copyRefs))
      return ec;
  }
  // Create atoms from undefined symbols.
  for (auto &sym : normalizedFile.undefinedSymbols) {
    // Undefinded symbols with n_value != 0 are actually tentative definitions.
    if (sym.value == Hex64(0)) {
      file->addUndefinedAtom(sym.name, copyRefs);
    } else {
      file->addTentativeDefAtom(sym.name, atomScope(sym.scope), sym.value,
                                DefinedAtom::Alignment(1 << (sym.desc >> 8)),
                                copyRefs);
    }
  }

  // Convert mach-o relocations to References
  std::unique_ptr<mach_o::ArchHandler> handler
                                     = ArchHandler::create(normalizedFile.arch);
  for (auto &sect : normalizedFile.sections) {
    if (isDebugInfoSection(sect))
      continue;
    if (llvm::Error ec = convertRelocs(sect, normalizedFile, scatterable,
                                       *file, *handler))
      return ec;
  }

  // Add additional arch-specific References
  file->eachDefinedAtom([&](MachODefinedAtom* atom) -> void {
    handler->addAdditionalReferences(*atom);
  });

  // Each __eh_frame section needs references to both __text (the function we're
  // providing unwind info for) and itself (FDE -> CIE). These aren't
  // represented in the relocations on some architectures, so we have to add
  // them back in manually there.
  if (auto ec = addEHFrameReferences(normalizedFile, *file, *handler))
    return ec;

  // Process mach-o data-in-code regions array. That information is encoded in
  // atoms as References at each transition point.
  unsigned nextIndex = 0;
  for (const DataInCode &entry : normalizedFile.dataInCode) {
    ++nextIndex;
    const Section* s = findSectionCoveringAddress(normalizedFile, entry.offset);
    if (!s) {
      return llvm::make_error<GenericError>(Twine("LC_DATA_IN_CODE address ("
                                                  + Twine(entry.offset)
                                                  + ") is not in any section"));
    }
    uint64_t offsetInSect = entry.offset - s->address;
    uint32_t offsetInAtom;
    MachODefinedAtom *atom = file->findAtomCoveringAddress(*s, offsetInSect,
                                                           &offsetInAtom);
    if (offsetInAtom + entry.length > atom->size()) {
      return llvm::make_error<GenericError>(Twine("LC_DATA_IN_CODE entry "
                                                  "(offset="
                                                  + Twine(entry.offset)
                                                  + ", length="
                                                  + Twine(entry.length)
                                                  + ") crosses atom boundary."));
    }
    // Add reference that marks start of data-in-code.
    atom->addReference(Reference::KindNamespace::mach_o, handler->kindArch(),
                       handler->dataInCodeTransitionStart(*atom),
                       offsetInAtom, atom, entry.kind);

    // Peek at next entry, if it starts where this one ends, skip ending ref.
    if (nextIndex < normalizedFile.dataInCode.size()) {
      const DataInCode &nextEntry = normalizedFile.dataInCode[nextIndex];
      if (nextEntry.offset == (entry.offset + entry.length))
        continue;
    }

    // If data goes to end of function, skip ending ref.
    if ((offsetInAtom + entry.length) == atom->size())
      continue;

    // Add reference that marks end of data-in-code.
    atom->addReference(Reference::KindNamespace::mach_o, handler->kindArch(),
                       handler->dataInCodeTransitionEnd(*atom),
                       offsetInAtom+entry.length, atom, 0);
  }

  // Cache some attributes on the file for use later.
  file->setFlags(normalizedFile.flags);
  file->setArch(normalizedFile.arch);
  file->setOS(normalizedFile.os);
  file->setMinVersion(normalizedFile.minOSverson);
  file->setMinVersionLoadCommandKind(normalizedFile.minOSVersionKind);

  // Sort references in each atom to their canonical order.
  for (const DefinedAtom* defAtom : file->defined()) {
    reinterpret_cast<const SimpleDefinedAtom*>(defAtom)->sortReferences();
  }

  if (auto err = parseDebugInfo(*file, normalizedFile, copyRefs))
    return err;

  return llvm::Error::success();
}

llvm::Error
normalizedDylibToAtoms(MachODylibFile *file,
                       const NormalizedFile &normalizedFile,
                       bool copyRefs) {
  file->setInstallName(normalizedFile.installName);
  file->setCompatVersion(normalizedFile.compatVersion);
  file->setCurrentVersion(normalizedFile.currentVersion);

  // Tell MachODylibFile object about all symbols it exports.
  if (!normalizedFile.exportInfo.empty()) {
    // If exports trie exists, use it instead of traditional symbol table.
    for (const Export &exp : normalizedFile.exportInfo) {
      bool weakDef = (exp.flags & EXPORT_SYMBOL_FLAGS_WEAK_DEFINITION);
      // StringRefs from export iterator are ephemeral, so force copy.
      file->addExportedSymbol(exp.name, weakDef, true);
    }
  } else {
    for (auto &sym : normalizedFile.globalSymbols) {
      assert((sym.scope & N_EXT) && "only expect external symbols here");
      bool weakDef = (sym.desc & N_WEAK_DEF);
      file->addExportedSymbol(sym.name, weakDef, copyRefs);
    }
  }
  // Tell MachODylibFile object about all dylibs it re-exports.
  for (const DependentDylib &dep : normalizedFile.dependentDylibs) {
    if (dep.kind == llvm::MachO::LC_REEXPORT_DYLIB)
      file->addReExportedDylib(dep.path);
  }
  return llvm::Error::success();
}

void relocatableSectionInfoForContentType(DefinedAtom::ContentType atomType,
                                          StringRef &segmentName,
                                          StringRef &sectionName,
                                          SectionType &sectionType,
                                          SectionAttr &sectionAttrs,
                                          bool &relocsToDefinedCanBeImplicit) {

  for (const MachORelocatableSectionToAtomType *p = sectsToAtomType ;
                                 p->atomType != DefinedAtom::typeUnknown; ++p) {
    if (p->atomType != atomType)
      continue;
    // Wild carded entries are ignored for reverse lookups.
    if (p->segmentName.empty() || p->sectionName.empty())
      continue;
    segmentName = p->segmentName;
    sectionName = p->sectionName;
    sectionType = p->sectionType;
    sectionAttrs = 0;
    relocsToDefinedCanBeImplicit = false;
    if (atomType == DefinedAtom::typeCode)
      sectionAttrs = S_ATTR_PURE_INSTRUCTIONS;
    if (atomType == DefinedAtom::typeCFI)
      relocsToDefinedCanBeImplicit = true;
    return;
  }
  llvm_unreachable("content type not yet supported");
}

llvm::Expected<std::unique_ptr<lld::File>>
normalizedToAtoms(const NormalizedFile &normalizedFile, StringRef path,
                  bool copyRefs) {
  switch (normalizedFile.fileType) {
  case MH_DYLIB:
  case MH_DYLIB_STUB:
    return dylibToAtoms(normalizedFile, path, copyRefs);
  case MH_OBJECT:
    return objectToAtoms(normalizedFile, path, copyRefs);
  default:
    llvm_unreachable("unhandled MachO file type!");
  }
}

} // namespace normalized
} // namespace mach_o
} // namespace lld