Imported Upstream version 6.10.0.49

Former-commit-id: 1d6753294b2993e1fbf92de9366bb9544db4189b

external/llvm-project/lld/lib/ReaderWriter/CMakeLists.txt

@@ -0,0 +1,20 @@
+add_subdirectory(MachO)
+add_subdirectory(YAML)
+
+if (MSVC)
+  add_definitions(-wd4062) # Suppress 'warning C4062: Enumerator has no associated handler in a switch statement.'
+endif()
+
+add_lld_library(lldReaderWriter
+  FileArchive.cpp
+
+  ADDITIONAL_HEADER_DIRS
+  ${LLD_INCLUDE_DIR}/lld/ReaderWriter
+
+  LINK_COMPONENTS
+    Object
+    Support
+
+  LINK_LIBS
+    lldCore
+  )

external/llvm-project/lld/lib/ReaderWriter/FileArchive.cpp

@@ -0,0 +1,228 @@
+//===- lib/ReaderWriter/FileArchive.cpp -----------------------------------===//
+//
+//                             The LLVM Linker
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "lld/Common/LLVM.h"
+#include "lld/Core/ArchiveLibraryFile.h"
+#include "lld/Core/File.h"
+#include "lld/Core/Reader.h"
+#include "llvm/ADT/STLExtras.h"
+#include "llvm/ADT/StringRef.h"
+#include "llvm/BinaryFormat/Magic.h"
+#include "llvm/Object/Archive.h"
+#include "llvm/Object/Error.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/ErrorOr.h"
+#include "llvm/Support/FileSystem.h"
+#include "llvm/Support/Format.h"
+#include "llvm/Support/MemoryBuffer.h"
+#include "llvm/Support/raw_ostream.h"
+#include <memory>
+#include <set>
+#include <string>
+#include <system_error>
+#include <unordered_map>
+#include <utility>
+#include <vector>
+
+using llvm::object::Archive;
+using llvm::file_magic;
+using llvm::identify_magic;
+
+namespace lld {
+
+namespace {
+
+/// \brief The FileArchive class represents an Archive Library file
+class FileArchive : public lld::ArchiveLibraryFile {
+public:
+  FileArchive(std::unique_ptr<MemoryBuffer> mb, const Registry &reg,
+              StringRef path, bool logLoading)
+      : ArchiveLibraryFile(path), _mb(std::shared_ptr<MemoryBuffer>(mb.release())),
+        _registry(reg), _logLoading(logLoading) {}
+
+  /// \brief Check if any member of the archive contains an Atom with the
+  /// specified name and return the File object for that member, or nullptr.
+  File *find(StringRef name) override {
+    auto member = _symbolMemberMap.find(name);
+    if (member == _symbolMemberMap.end())
+      return nullptr;
+    Archive::Child c = member->second;
+
+    // Don't return a member already returned
+    Expected<StringRef> buf = c.getBuffer();
+    if (!buf) {
+      // TODO: Actually report errors helpfully.
+      consumeError(buf.takeError());
+      return nullptr;
+    }
+    const char *memberStart = buf->data();
+    if (_membersInstantiated.count(memberStart))
+      return nullptr;
+    _membersInstantiated.insert(memberStart);
+
+    std::unique_ptr<File> result;
+    if (instantiateMember(c, result))
+      return nullptr;
+
+    File *file = result.get();
+    _filesReturned.push_back(std::move(result));
+
+    // Give up the file pointer. It was stored and will be destroyed with destruction of FileArchive
+    return file;
+  }
+
+  /// \brief parse each member
+  std::error_code
+  parseAllMembers(std::vector<std::unique_ptr<File>> &result) override {
+    if (std::error_code ec = parse())
+      return ec;
+    llvm::Error err = llvm::Error::success();
+    for (auto mf = _archive->child_begin(err), me = _archive->child_end();
+         mf != me; ++mf) {
+      std::unique_ptr<File> file;
+      if (std::error_code ec = instantiateMember(*mf, file)) {
+        // err is Success (or we wouldn't be in the loop body) but we can't
+        // return without testing or consuming it.
+        consumeError(std::move(err));
+        return ec;
+      }
+      result.push_back(std::move(file));
+    }
+    if (err)
+      return errorToErrorCode(std::move(err));
+    return std::error_code();
+  }
+
+  const AtomRange<DefinedAtom> defined() const override {
+    return _noDefinedAtoms;
+  }
+
+  const AtomRange<UndefinedAtom> undefined() const override {
+    return _noUndefinedAtoms;
+  }
+
+  const AtomRange<SharedLibraryAtom> sharedLibrary() const override {
+    return _noSharedLibraryAtoms;
+  }
+
+  const AtomRange<AbsoluteAtom> absolute() const override {
+    return _noAbsoluteAtoms;
+  }
+
+  void clearAtoms() override {
+    _noDefinedAtoms.clear();
+    _noUndefinedAtoms.clear();
+    _noSharedLibraryAtoms.clear();
+    _noAbsoluteAtoms.clear();
+  }
+
+protected:
+  std::error_code doParse() override {
+    // Make Archive object which will be owned by FileArchive object.
+    llvm::Error Err = llvm::Error::success();
+    _archive.reset(new Archive(_mb->getMemBufferRef(), Err));
+    if (Err)
+      return errorToErrorCode(std::move(Err));
+    std::error_code ec;
+    if ((ec = buildTableOfContents()))
+      return ec;
+    return std::error_code();
+  }
+
+private:
+  std::error_code instantiateMember(Archive::Child member,
+                                    std::unique_ptr<File> &result) const {
+    Expected<llvm::MemoryBufferRef> mbOrErr = member.getMemoryBufferRef();
+    if (!mbOrErr)
+      return errorToErrorCode(mbOrErr.takeError());
+    llvm::MemoryBufferRef mb = mbOrErr.get();
+    std::string memberPath = (_archive->getFileName() + "("
+                           + mb.getBufferIdentifier() + ")").str();
+
+    if (_logLoading)
+      llvm::errs() << memberPath << "\n";
+
+    std::unique_ptr<MemoryBuffer> memberMB(MemoryBuffer::getMemBuffer(
+        mb.getBuffer(), mb.getBufferIdentifier(), false));
+
+    ErrorOr<std::unique_ptr<File>> fileOrErr =
+        _registry.loadFile(std::move(memberMB));
+    if (std::error_code ec = fileOrErr.getError())
+      return ec;
+    result = std::move(fileOrErr.get());
+    if (std::error_code ec = result->parse())
+      return ec;
+    result->setArchivePath(_archive->getFileName());
+
+    // The memory buffer is co-owned by the archive file and the children,
+    // so that the bufffer is deallocated when all the members are destructed.
+    result->setSharedMemoryBuffer(_mb);
+    return std::error_code();
+  }
+
+  std::error_code buildTableOfContents() {
+    DEBUG_WITH_TYPE("FileArchive", llvm::dbgs()
+                                       << "Table of contents for archive '"
+                                       << _archive->getFileName() << "':\n");
+    for (const Archive::Symbol &sym : _archive->symbols()) {
+      StringRef name = sym.getName();
+      Expected<Archive::Child> memberOrErr = sym.getMember();
+      if (!memberOrErr)
+        return errorToErrorCode(memberOrErr.takeError());
+      Archive::Child member = memberOrErr.get();
+      DEBUG_WITH_TYPE("FileArchive",
+                      llvm::dbgs()
+                          << llvm::format("0x%08llX ",
+                                          member.getBuffer()->data())
+                          << "'" << name << "'\n");
+      _symbolMemberMap.insert(std::make_pair(name, member));
+    }
+    return std::error_code();
+  }
+
+  typedef std::unordered_map<StringRef, Archive::Child> MemberMap;
+  typedef std::set<const char *> InstantiatedSet;
+
+  std::shared_ptr<MemoryBuffer> _mb;
+  const Registry &_registry;
+  std::unique_ptr<Archive> _archive;
+  MemberMap _symbolMemberMap;
+  InstantiatedSet _membersInstantiated;
+  bool _logLoading;
+  std::vector<std::unique_ptr<MemoryBuffer>> _memberBuffers;
+  std::vector<std::unique_ptr<File>> _filesReturned;
+};
+
+class ArchiveReader : public Reader {
+public:
+  ArchiveReader(bool logLoading) : _logLoading(logLoading) {}
+
+  bool canParse(file_magic magic, MemoryBufferRef) const override {
+    return magic == file_magic::archive;
+  }
+
+  ErrorOr<std::unique_ptr<File>> loadFile(std::unique_ptr<MemoryBuffer> mb,
+                                          const Registry &reg) const override {
+    StringRef path = mb->getBufferIdentifier();
+    std::unique_ptr<File> ret =
+        llvm::make_unique<FileArchive>(std::move(mb), reg, path, _logLoading);
+    return std::move(ret);
+  }
+
+private:
+  bool _logLoading;
+};
+
+} // anonymous namespace
+
+void Registry::addSupportArchives(bool logLoading) {
+  add(std::unique_ptr<Reader>(new ArchiveReader(logLoading)));
+}
+
+} // namespace lld

external/llvm-project/lld/lib/ReaderWriter/MachO/ArchHandler.cpp

@@ -0,0 +1,172 @@
+//===- lib/FileFormat/MachO/ArchHandler.cpp -------------------------------===//
+//
+//                             The LLVM Linker
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+
+#include "ArchHandler.h"
+#include "Atoms.h"
+#include "MachONormalizedFileBinaryUtils.h"
+#include "llvm/ADT/StringRef.h"
+#include "llvm/ADT/StringSwitch.h"
+#include "llvm/ADT/Triple.h"
+#include "llvm/Support/ErrorHandling.h"
+
+using namespace llvm::MachO;
+using namespace lld::mach_o::normalized;
+
+namespace lld {
+namespace mach_o {
+
+
+ArchHandler::ArchHandler() {
+}
+
+ArchHandler::~ArchHandler() {
+}
+
+std::unique_ptr<mach_o::ArchHandler> ArchHandler::create(
+                                               MachOLinkingContext::Arch arch) {
+  switch (arch) {
+  case MachOLinkingContext::arch_x86_64:
+    return create_x86_64();
+  case MachOLinkingContext::arch_x86:
+    return create_x86();
+  case MachOLinkingContext::arch_armv6:
+  case MachOLinkingContext::arch_armv7:
+  case MachOLinkingContext::arch_armv7s:
+    return create_arm();
+  case MachOLinkingContext::arch_arm64:
+    return create_arm64();
+  default:
+    llvm_unreachable("Unknown arch");
+  }
+}
+
+
+bool ArchHandler::isLazyPointer(const Reference &ref) {
+  // A lazy bind entry is needed for a lazy pointer.
+  const StubInfo &info = stubInfo();
+  if (ref.kindNamespace() != Reference::KindNamespace::mach_o)
+    return false;
+  if (ref.kindArch() != info.lazyPointerReferenceToFinal.arch)
+    return false;
+  return (ref.kindValue() == info.lazyPointerReferenceToFinal.kind);
+}
+
+
+ArchHandler::RelocPattern ArchHandler::relocPattern(const Relocation &reloc) {
+  assert((reloc.type & 0xFFF0) == 0);
+  uint16_t result = reloc.type;
+  if (reloc.scattered)
+    result |= rScattered;
+  if (reloc.pcRel)
+    result |= rPcRel;
+  if (reloc.isExtern)
+    result |= rExtern;
+  switch(reloc.length) {
+  case 0:
+    break;
+  case 1:
+    result |= rLength2;
+    break;
+  case 2:
+    result |= rLength4;
+    break;
+  case 3:
+    result |= rLength8;
+    break;
+  default:
+    llvm_unreachable("bad r_length");
+  }
+  return result;
+}
+
+normalized::Relocation
+ArchHandler::relocFromPattern(ArchHandler::RelocPattern pattern) {
+  normalized::Relocation result;
+  result.offset    = 0;
+  result.scattered = (pattern & rScattered);
+  result.type     = (RelocationInfoType)(pattern & 0xF);
+  result.pcRel    = (pattern & rPcRel);
+  result.isExtern = (pattern & rExtern);
+  result.value    = 0;
+  result.symbol    = 0;
+  switch (pattern & 0x300) {
+  case rLength1:
+    result.length = 0;
+    break;
+  case rLength2:
+    result.length = 1;
+    break;
+  case rLength4:
+    result.length = 2;
+    break;
+  case rLength8:
+    result.length = 3;
+    break;
+  }
+  return result;
+}
+
+void ArchHandler::appendReloc(normalized::Relocations &relocs, uint32_t offset,
+                              uint32_t symbol, uint32_t value,
+                              RelocPattern pattern) {
+  normalized::Relocation reloc = relocFromPattern(pattern);
+  reloc.offset = offset;
+  reloc.symbol = symbol;
+  reloc.value  = value;
+  relocs.push_back(reloc);
+}
+
+
+int16_t ArchHandler::readS16(const uint8_t *addr, bool isBig) {
+    return read16(addr, isBig);
+}
+
+int32_t ArchHandler::readS32(const uint8_t *addr, bool isBig) {
+  return read32(addr, isBig);
+}
+
+uint32_t ArchHandler::readU32(const uint8_t *addr, bool isBig) {
+  return read32(addr, isBig);
+}
+
+  int64_t ArchHandler::readS64(const uint8_t *addr, bool isBig) {
+  return read64(addr, isBig);
+}
+
+bool ArchHandler::isDwarfCIE(bool isBig, const DefinedAtom *atom) {
+  assert(atom->contentType() == DefinedAtom::typeCFI);
+  if (atom->rawContent().size() < sizeof(uint32_t))
+    return false;
+  uint32_t size = read32(atom->rawContent().data(), isBig);
+
+  uint32_t idOffset = sizeof(uint32_t);
+  if (size == 0xffffffffU)
+    idOffset += sizeof(uint64_t);
+
+  return read32(atom->rawContent().data() + idOffset, isBig) == 0;
+}
+
+const Atom *ArchHandler::fdeTargetFunction(const DefinedAtom *fde) {
+  for (auto ref : *fde) {
+    if (ref->kindNamespace() == Reference::KindNamespace::mach_o &&
+        ref->kindValue() == unwindRefToFunctionKind()) {
+      assert(ref->kindArch() == kindArch() && "unexpected Reference arch");
+      return ref->target();
+    }
+  }
+
+  return nullptr;
+}
+
+} // namespace mach_o
+} // namespace lld
+
+
+

external/llvm-project/lld/lib/ReaderWriter/MachO/ArchHandler.h

@@ -0,0 +1,323 @@
+//===- lib/FileFormat/MachO/ArchHandler.h ---------------------------------===//
+//
+//                             The LLVM Linker
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLD_READER_WRITER_MACHO_ARCH_HANDLER_H
+#define LLD_READER_WRITER_MACHO_ARCH_HANDLER_H
+
+#include "Atoms.h"
+#include "File.h"
+#include "MachONormalizedFile.h"
+#include "lld/Common/LLVM.h"
+#include "lld/Core/Error.h"
+#include "lld/Core/Reference.h"
+#include "lld/Core/Simple.h"
+#include "lld/ReaderWriter/MachOLinkingContext.h"
+#include "llvm/ADT/Triple.h"
+
+namespace lld {
+namespace mach_o {
+
+///
+/// The ArchHandler class handles all architecture specific aspects of
+/// mach-o linking.
+///
+class ArchHandler {
+public:
+  virtual ~ArchHandler();
+
+  /// There is no public interface to subclasses of ArchHandler, so this
+  /// is the only way to instantiate an ArchHandler.
+  static std::unique_ptr<ArchHandler> create(MachOLinkingContext::Arch arch);
+
+  /// Get (arch specific) kind strings used by Registry.
+  virtual const Registry::KindStrings *kindStrings() = 0;
+
+  /// Convert mach-o Arch to Reference::KindArch.
+  virtual Reference::KindArch kindArch() = 0;
+
+  /// Used by StubPass to update References to shared library functions
+  /// to be references to a stub.
+  virtual bool isCallSite(const Reference &) = 0;
+
+  /// Used by GOTPass to locate GOT References
+  virtual bool isGOTAccess(const Reference &, bool &canBypassGOT) {
+    return false;
+  }
+
+  /// Used by TLVPass to locate TLV References.
+  virtual bool isTLVAccess(const Reference &) const { return false; }
+
+  /// Used by the TLVPass to update TLV References.
+  virtual void updateReferenceToTLV(const Reference *) {}
+
+  /// Used by ShimPass to insert shims in branches that switch mode.
+  virtual bool isNonCallBranch(const Reference &) = 0;
+
+  /// Used by GOTPass to update GOT References
+  virtual void updateReferenceToGOT(const Reference *, bool targetIsNowGOT) {}
+
+  /// Does this architecture make use of __unwind_info sections for exception
+  /// handling? If so, it will need a separate pass to create them.
+  virtual bool needsCompactUnwind() = 0;
+
+  /// Returns the kind of reference to use to synthesize a 32-bit image-offset
+  /// value, used in the __unwind_info section.
+  virtual Reference::KindValue imageOffsetKind() = 0;
+
+  /// Returns the kind of reference to use to synthesize a 32-bit image-offset
+  /// indirect value. Used for personality functions in the __unwind_info
+  /// section.
+  virtual Reference::KindValue imageOffsetKindIndirect() = 0;
+
+  /// Architecture specific compact unwind type that signals __eh_frame should
+  /// actually be used.
+  virtual uint32_t dwarfCompactUnwindType() = 0;
+
+  /// Reference from an __eh_frame CIE atom to its personality function it's
+  /// describing. Usually pointer-sized and PC-relative, but differs in whether
+  /// it needs to be in relocatable objects.
+  virtual Reference::KindValue unwindRefToPersonalityFunctionKind() = 0;
+
+  /// Reference from an __eh_frame FDE to the CIE it's based on.
+  virtual Reference::KindValue unwindRefToCIEKind() = 0;
+
+  /// Reference from an __eh_frame FDE atom to the function it's
+  /// describing. Usually pointer-sized and PC-relative, but differs in whether
+  /// it needs to be in relocatable objects.
+  virtual Reference::KindValue unwindRefToFunctionKind() = 0;
+
+  /// Reference from an __unwind_info entry of dwarfCompactUnwindType to the
+  /// required __eh_frame entry. On current architectures, the low 24 bits
+  /// represent the offset of the function's FDE entry from the start of
+  /// __eh_frame.
+  virtual Reference::KindValue unwindRefToEhFrameKind() = 0;
+
+  /// Returns a pointer sized reference kind.  On 64-bit targets this will
+  /// likely be something like pointer64, and pointer32 on 32-bit targets.
+  virtual Reference::KindValue pointerKind() = 0;
+
+  virtual const Atom *fdeTargetFunction(const DefinedAtom *fde);
+
+  /// Used by normalizedFromAtoms() to know where to generated rebasing and
+  /// binding info in final executables.
+  virtual bool isPointer(const Reference &) = 0;
+
+  /// Used by normalizedFromAtoms() to know where to generated lazy binding
+  /// info in final executables.
+  virtual bool isLazyPointer(const Reference &);
+
+  /// Reference from an __stub_helper entry to the required offset of the
+  /// lazy bind commands.
+  virtual Reference::KindValue lazyImmediateLocationKind() = 0;
+
+  /// Returns true if the specified relocation is paired to the next relocation.
+  virtual bool isPairedReloc(const normalized::Relocation &) = 0;
+
+  /// Prototype for a helper function.  Given a sectionIndex and address,
+  /// finds the atom and offset with that atom of that address.
+  typedef std::function<llvm::Error (uint32_t sectionIndex, uint64_t addr,
+                        const lld::Atom **, Reference::Addend *)>
+                        FindAtomBySectionAndAddress;
+
+  /// Prototype for a helper function.  Given a symbolIndex, finds the atom
+  /// representing that symbol.
+  typedef std::function<llvm::Error (uint32_t symbolIndex,
+                        const lld::Atom **)> FindAtomBySymbolIndex;
+
+  /// Analyzes a relocation from a .o file and returns the info
+  /// (kind, target, addend) needed to instantiate a Reference.
+  /// Two helper functions are passed as parameters to find the target atom
+  /// given a symbol index or address.
+  virtual llvm::Error
+          getReferenceInfo(const normalized::Relocation &reloc,
+                           const DefinedAtom *inAtom,
+                           uint32_t offsetInAtom,
+                           uint64_t fixupAddress, bool isBigEndian,
+                           FindAtomBySectionAndAddress atomFromAddress,
+                           FindAtomBySymbolIndex atomFromSymbolIndex,
+                           Reference::KindValue *kind,
+                           const lld::Atom **target,
+                           Reference::Addend *addend) = 0;
+
+  /// Analyzes a pair of relocations from a .o file and returns the info
+  /// (kind, target, addend) needed to instantiate a Reference.
+  /// Two helper functions are passed as parameters to find the target atom
+  /// given a symbol index or address.
+  virtual llvm::Error
+      getPairReferenceInfo(const normalized::Relocation &reloc1,
+                           const normalized::Relocation &reloc2,
+                           const DefinedAtom *inAtom,
+                           uint32_t offsetInAtom,
+                           uint64_t fixupAddress, bool isBig, bool scatterable,
+                           FindAtomBySectionAndAddress atomFromAddress,
+                           FindAtomBySymbolIndex atomFromSymbolIndex,
+                           Reference::KindValue *kind,
+                           const lld::Atom **target,
+                           Reference::Addend *addend) = 0;
+
+  /// Prototype for a helper function.  Given an atom, finds the symbol table
+  /// index for it in the output file.
+  typedef std::function<uint32_t (const Atom &atom)> FindSymbolIndexForAtom;
+
+  /// Prototype for a helper function.  Given an atom, finds the index
+  /// of the section that will contain the atom.
+  typedef std::function<uint32_t (const Atom &atom)> FindSectionIndexForAtom;
+
+  /// Prototype for a helper function.  Given an atom, finds the address
+  /// assigned to it in the output file.
+  typedef std::function<uint64_t (const Atom &atom)> FindAddressForAtom;
+
+  /// Some architectures require local symbols on anonymous atoms.
+  virtual bool needsLocalSymbolInRelocatableFile(const DefinedAtom *atom) {
+    return false;
+  }
+
+  /// Copy raw content then apply all fixup References on an Atom.
+  virtual void generateAtomContent(const DefinedAtom &atom, bool relocatable,
+                                   FindAddressForAtom findAddress,
+                                   FindAddressForAtom findSectionAddress,
+                                   uint64_t imageBaseAddress,
+                          llvm::MutableArrayRef<uint8_t> atomContentBuffer) = 0;
+
+  /// Used in -r mode to convert a Reference to a mach-o relocation.
+  virtual void appendSectionRelocations(const DefinedAtom &atom,
+                                        uint64_t atomSectionOffset,
+                                        const Reference &ref,
+                                        FindSymbolIndexForAtom,
+                                        FindSectionIndexForAtom,
+                                        FindAddressForAtom,
+                                        normalized::Relocations&) = 0;
+
+  /// Add arch-specific References.
+  virtual void addAdditionalReferences(MachODefinedAtom &atom) { }
+
+  // Add Reference for data-in-code marker.
+  virtual void addDataInCodeReference(MachODefinedAtom &atom, uint32_t atomOff,
+                                      uint16_t length, uint16_t kind) { }
+
+  /// Returns true if the specificed Reference value marks the start or end
+  /// of a data-in-code range in an atom.
+  virtual bool isDataInCodeTransition(Reference::KindValue refKind) {
+    return false;
+  }
+
+  /// Returns the Reference value for a Reference that marks that start of
+  /// a data-in-code range.
+  virtual Reference::KindValue dataInCodeTransitionStart(
+                                                const MachODefinedAtom &atom) {
+    return 0;
+  }
+
+  /// Returns the Reference value for a Reference that marks that end of
+  /// a data-in-code range.
+  virtual Reference::KindValue dataInCodeTransitionEnd(
+                                                const MachODefinedAtom &atom) {
+    return 0;
+  }
+
+  /// Only relevant for 32-bit arm archs.
+  virtual bool isThumbFunction(const DefinedAtom &atom) { return false; }
+
+  /// Only relevant for 32-bit arm archs.
+  virtual const DefinedAtom *createShim(MachOFile &file, bool thumbToArm,
+                                        const DefinedAtom &) {
+    llvm_unreachable("shims only support on arm");
+  }
+
+  /// Does a given unwind-cfi atom represent a CIE (as opposed to an FDE).
+  static bool isDwarfCIE(bool isBig, const DefinedAtom *atom);
+
+  struct ReferenceInfo {
+    Reference::KindArch arch;
+    uint16_t            kind;
+    uint32_t            offset;
+    int32_t             addend;
+  };
+
+  struct OptionalRefInfo {
+    bool                used;
+    uint16_t            kind;
+    uint32_t            offset;
+    int32_t             addend;
+  };
+
+  /// Table of architecture specific information for creating stubs.
+  struct StubInfo {
+    const char*     binderSymbolName;
+    ReferenceInfo   lazyPointerReferenceToHelper;
+    ReferenceInfo   lazyPointerReferenceToFinal;
+    ReferenceInfo   nonLazyPointerReferenceToBinder;
+    uint8_t         codeAlignment;
+
+    uint32_t        stubSize;
+    uint8_t         stubBytes[16];
+    ReferenceInfo   stubReferenceToLP;
+    OptionalRefInfo optStubReferenceToLP;
+
+    uint32_t        stubHelperSize;
+    uint8_t         stubHelperBytes[16];
+    ReferenceInfo   stubHelperReferenceToImm;
+    ReferenceInfo   stubHelperReferenceToHelperCommon;
+
+    DefinedAtom::ContentType stubHelperImageCacheContentType;
+
+    uint32_t        stubHelperCommonSize;
+    uint8_t         stubHelperCommonAlignment;
+    uint8_t         stubHelperCommonBytes[36];
+    ReferenceInfo   stubHelperCommonReferenceToCache;
+    OptionalRefInfo optStubHelperCommonReferenceToCache;
+    ReferenceInfo   stubHelperCommonReferenceToBinder;
+    OptionalRefInfo optStubHelperCommonReferenceToBinder;
+  };
+
+  virtual const StubInfo &stubInfo() = 0;
+
+protected:
+  ArchHandler();
+
+  static std::unique_ptr<mach_o::ArchHandler> create_x86_64();
+  static std::unique_ptr<mach_o::ArchHandler> create_x86();
+  static std::unique_ptr<mach_o::ArchHandler> create_arm();
+  static std::unique_ptr<mach_o::ArchHandler> create_arm64();
+
+  // Handy way to pack mach-o r_type and other bit fields into one 16-bit value.
+  typedef uint16_t RelocPattern;
+  enum {
+    rScattered = 0x8000,
+    rPcRel     = 0x4000,
+    rExtern    = 0x2000,
+    rLength1   = 0x0000,
+    rLength2   = 0x0100,
+    rLength4   = 0x0200,
+    rLength8   = 0x0300,
+    rLenArmLo  = rLength1,
+    rLenArmHi  = rLength2,
+    rLenThmbLo = rLength4,
+    rLenThmbHi = rLength8
+  };
+  /// Extract RelocPattern from normalized mach-o relocation.
+  static RelocPattern relocPattern(const normalized::Relocation &reloc);
+  /// Create normalized Relocation initialized from pattern.
+  static normalized::Relocation relocFromPattern(RelocPattern pattern);
+  /// One liner to add a relocation.
+  static void appendReloc(normalized::Relocations &relocs, uint32_t offset,
+                          uint32_t symbol, uint32_t value,
+                          RelocPattern pattern);
+
+
+  static int16_t  readS16(const uint8_t *addr, bool isBig);
+  static int32_t  readS32(const uint8_t *addr, bool isBig);
+  static uint32_t readU32(const uint8_t *addr, bool isBig);
+  static int64_t  readS64(const uint8_t *addr, bool isBig);
+};
+
+} // namespace mach_o
+} // namespace lld
+
+#endif // LLD_READER_WRITER_MACHO_ARCH_HANDLER_H

external/llvm-project/lld/lib/ReaderWriter/MachO/Atoms.h

@@ -0,0 +1,181 @@
+//===- lib/ReaderWriter/MachO/Atoms.h ---------------------------*- C++ -*-===//
+//
+//                             The LLVM Linker
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLD_READER_WRITER_MACHO_ATOMS_H
+#define LLD_READER_WRITER_MACHO_ATOMS_H
+
+#include "lld/Core/Atom.h"
+#include "lld/Core/DefinedAtom.h"
+#include "lld/Core/SharedLibraryAtom.h"
+#include "lld/Core/Simple.h"
+#include "llvm/ADT/ArrayRef.h"
+#include "llvm/ADT/StringRef.h"
+#include <cstdint>
+#include <string>
+
+namespace lld {
+
+class File;
+
+namespace mach_o {
+
+class MachODefinedAtom : public SimpleDefinedAtom {
+public:
+  MachODefinedAtom(const File &f, const StringRef name, Scope scope,
+                   ContentType type, Merge merge, bool thumb, bool noDeadStrip,
+                   const ArrayRef<uint8_t> content, Alignment align)
+      : SimpleDefinedAtom(f), _name(name), _content(content),
+        _align(align), _contentType(type), _scope(scope), _merge(merge),
+        _thumb(thumb), _noDeadStrip(noDeadStrip) {}
+
+  // Constructor for zero-fill content
+  MachODefinedAtom(const File &f, const StringRef name, Scope scope,
+                   ContentType type, uint64_t size, bool noDeadStrip,
+                   Alignment align)
+      : SimpleDefinedAtom(f), _name(name),
+        _content(ArrayRef<uint8_t>(nullptr, size)), _align(align),
+        _contentType(type), _scope(scope), _merge(mergeNo), _thumb(false),
+        _noDeadStrip(noDeadStrip) {}
+
+  ~MachODefinedAtom() override = default;
+
+  uint64_t size() const override { return _content.size(); }
+
+  ContentType contentType() const override { return _contentType; }
+
+  Alignment alignment() const override { return _align; }
+
+  StringRef name() const override { return _name; }
+
+  Scope scope() const override { return _scope; }
+
+  Merge merge() const override { return _merge; }
+
+  DeadStripKind deadStrip() const override {
+    if (_contentType == DefinedAtom::typeInitializerPtr)
+      return deadStripNever;
+    if (_contentType == DefinedAtom::typeTerminatorPtr)
+      return deadStripNever;
+    if (_noDeadStrip)
+      return deadStripNever;
+    return deadStripNormal;
+  }
+
+  ArrayRef<uint8_t> rawContent() const override {
+    // Note: Zerofill atoms have a content pointer which is null.
+    return _content;
+  }
+
+  bool isThumb() const { return _thumb; }
+
+private:
+  const StringRef _name;
+  const ArrayRef<uint8_t> _content;
+  const DefinedAtom::Alignment _align;
+  const ContentType _contentType;
+  const Scope _scope;
+  const Merge _merge;
+  const bool _thumb;
+  const bool _noDeadStrip;
+};
+
+class MachODefinedCustomSectionAtom : public MachODefinedAtom {
+public:
+  MachODefinedCustomSectionAtom(const File &f, const StringRef name,
+                                Scope scope, ContentType type, Merge merge,
+                                bool thumb, bool noDeadStrip,
+                                const ArrayRef<uint8_t> content,
+                                StringRef sectionName, Alignment align)
+      : MachODefinedAtom(f, name, scope, type, merge, thumb, noDeadStrip,
+                         content, align),
+        _sectionName(sectionName) {}
+
+  ~MachODefinedCustomSectionAtom() override = default;
+
+  SectionChoice sectionChoice() const override {
+    return DefinedAtom::sectionCustomRequired;
+  }
+
+  StringRef customSectionName() const override {
+    return _sectionName;
+  }
+private:
+  StringRef _sectionName;
+};
+
+class MachOTentativeDefAtom : public SimpleDefinedAtom {
+public:
+  MachOTentativeDefAtom(const File &f, const StringRef name, Scope scope,
+                        uint64_t size, DefinedAtom::Alignment align)
+      : SimpleDefinedAtom(f), _name(name), _scope(scope), _size(size),
+        _align(align) {}
+
+  ~MachOTentativeDefAtom() override = default;
+
+  uint64_t size() const override { return _size; }
+
+  Merge merge() const override { return DefinedAtom::mergeAsTentative; }
+
+  ContentType contentType() const override { return DefinedAtom::typeZeroFill; }
+
+  Alignment alignment() const override { return _align; }
+
+  StringRef name() const override { return _name; }
+
+  Scope scope() const override { return _scope; }
+
+  ArrayRef<uint8_t> rawContent() const override { return ArrayRef<uint8_t>(); }
+
+private:
+  const std::string _name;
+  const Scope _scope;
+  const uint64_t _size;
+  const DefinedAtom::Alignment _align;
+};
+
+class MachOSharedLibraryAtom : public SharedLibraryAtom {
+public:
+  MachOSharedLibraryAtom(const File &file, StringRef name,
+                         StringRef dylibInstallName, bool weakDef)
+      : SharedLibraryAtom(), _file(file), _name(name),
+        _dylibInstallName(dylibInstallName) {}
+  ~MachOSharedLibraryAtom() override = default;
+
+  StringRef loadName() const override { return _dylibInstallName; }
+
+  bool canBeNullAtRuntime() const override {
+    // FIXME: this may actually be changeable. For now, all symbols are strongly
+    // defined though.
+    return false;
+  }
+
+  const File &file() const override { return _file; }
+
+  StringRef name() const override { return _name; }
+
+  Type type() const override {
+    // Unused in MachO (I think).
+    return Type::Unknown;
+  }
+
+  uint64_t size() const override {
+    // Unused in MachO (I think)
+    return 0;
+  }
+
+private:
+  const File &_file;
+  StringRef _name;
+  StringRef _dylibInstallName;
+};
+
+} // end namespace mach_o
+} // end namespace lld
+
+#endif // LLD_READER_WRITER_MACHO_ATOMS_H

external/llvm-project/lld/lib/ReaderWriter/MachO/CMakeLists.txt

@@ -0,0 +1,34 @@
+add_lld_library(lldMachO
+  ArchHandler.cpp
+  ArchHandler_arm.cpp
+  ArchHandler_arm64.cpp
+  ArchHandler_x86.cpp
+  ArchHandler_x86_64.cpp
+  CompactUnwindPass.cpp
+  GOTPass.cpp
+  LayoutPass.cpp
+  MachOLinkingContext.cpp
+  MachONormalizedFileBinaryReader.cpp
+  MachONormalizedFileBinaryWriter.cpp
+  MachONormalizedFileFromAtoms.cpp
+  MachONormalizedFileToAtoms.cpp
+  MachONormalizedFileYAML.cpp
+  ObjCPass.cpp
+  ShimPass.cpp
+  StubsPass.cpp
+  TLVPass.cpp
+  WriterMachO.cpp
+
+  LINK_COMPONENTS
+    DebugInfoDWARF
+    Demangle
+    Object
+    Support
+
+  LINK_LIBS
+    lldCore
+    lldYAML
+    ${LLVM_PTHREAD_LIB}
+  )
+
+include_directories(.)

external/llvm-project/lld/lib/ReaderWriter/MachO/DebugInfo.h

@@ -0,0 +1,106 @@
+//===- lib/ReaderWriter/MachO/File.h ----------------------------*- C++ -*-===//
+//
+//                             The LLVM Linker
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLD_READER_WRITER_MACHO_DEBUGINFO_H
+#define LLD_READER_WRITER_MACHO_DEBUGINFO_H
+
+#include "lld/Core/Atom.h"
+#include <vector>
+
+#include "llvm/Support/Format.h"
+#include "llvm/Support/raw_ostream.h"
+
+
+namespace lld {
+namespace mach_o {
+
+class DebugInfo {
+public:
+  enum class Kind {
+    Dwarf,
+    Stabs
+  };
+
+  Kind kind() const { return _kind; }
+
+  void setAllocator(std::unique_ptr<llvm::BumpPtrAllocator> allocator) {
+    _allocator = std::move(allocator);
+  }
+
+protected:
+  DebugInfo(Kind kind) : _kind(kind) {}
+
+private:
+  std::unique_ptr<llvm::BumpPtrAllocator> _allocator;
+  Kind _kind;
+};
+
+struct TranslationUnitSource {
+  StringRef name;
+  StringRef path;
+};
+
+class DwarfDebugInfo : public DebugInfo {
+public:
+  DwarfDebugInfo(TranslationUnitSource tu)
+    : DebugInfo(Kind::Dwarf), _tu(std::move(tu)) {}
+
+  static inline bool classof(const DebugInfo *di) {
+    return di->kind() == Kind::Dwarf;
+  }
+
+  const TranslationUnitSource &translationUnitSource() const { return _tu; }
+
+private:
+  TranslationUnitSource _tu;
+};
+
+struct Stab {
+  Stab(const Atom* atom, uint8_t type, uint8_t other, uint16_t desc,
+       uint32_t value, StringRef str)
+    : atom(atom), type(type), other(other), desc(desc), value(value),
+      str(str) {}
+
+  const class Atom*   atom;
+  uint8_t             type;
+  uint8_t             other;
+  uint16_t            desc;
+  uint32_t            value;
+  StringRef           str;
+};
+
+inline raw_ostream& operator<<(raw_ostream &os, Stab &s) {
+  os << "Stab -- atom: " << llvm::format("%p", s.atom) << ", type: " << (uint32_t)s.type
+     << ", other: " << (uint32_t)s.other << ", desc: " << s.desc << ", value: " << s.value
+     << ", str: '" << s.str << "'";
+  return os;
+}
+
+class StabsDebugInfo : public DebugInfo {
+public:
+
+  typedef std::vector<Stab> StabsList;
+
+  StabsDebugInfo(StabsList stabs)
+    : DebugInfo(Kind::Stabs), _stabs(std::move(stabs)) {}
+
+  static inline bool classof(const DebugInfo *di) {
+    return di->kind() == Kind::Stabs;
+  }
+
+  const StabsList& stabs() const { return _stabs; }
+
+public:
+  StabsList _stabs;
+};
+
+} // end namespace mach_o
+} // end namespace lld
+
+#endif // LLD_READER_WRITER_MACHO_DEBUGINFO_H

external/llvm-project/lld/lib/ReaderWriter/MachO/ExecutableAtoms.h

@@ -0,0 +1,155 @@
+//===- lib/ReaderWriter/MachO/ExecutableAtoms.h ---------------------------===//
+//
+//                             The LLVM Linker
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLD_READER_WRITER_MACHO_EXECUTABLE_ATOMS_H
+#define LLD_READER_WRITER_MACHO_EXECUTABLE_ATOMS_H
+
+#include "Atoms.h"
+#include "File.h"
+
+#include "llvm/BinaryFormat/MachO.h"
+
+#include "lld/Core/DefinedAtom.h"
+#include "lld/Core/File.h"
+#include "lld/Core/LinkingContext.h"
+#include "lld/Core/Reference.h"
+#include "lld/Core/Simple.h"
+#include "lld/Core/UndefinedAtom.h"
+#include "lld/ReaderWriter/MachOLinkingContext.h"
+
+namespace lld {
+namespace mach_o {
+
+
+//
+// CEntryFile adds an UndefinedAtom for "_main" so that the Resolving
+// phase will fail if "_main" is undefined.
+//
+class CEntryFile : public SimpleFile {
+public:
+  CEntryFile(const MachOLinkingContext &context)
+      : SimpleFile("C entry", kindCEntryObject),
+       _undefMain(*this, context.entrySymbolName()) {
+    this->addAtom(_undefMain);
+  }
+
+private:
+  SimpleUndefinedAtom   _undefMain;
+};
+
+
+//
+// StubHelperFile adds an UndefinedAtom for "dyld_stub_binder" so that
+// the Resolveing phase will fail if "dyld_stub_binder" is undefined.
+//
+class StubHelperFile : public SimpleFile {
+public:
+  StubHelperFile(const MachOLinkingContext &context)
+      : SimpleFile("stub runtime", kindStubHelperObject),
+        _undefBinder(*this, context.binderSymbolName()) {
+    this->addAtom(_undefBinder);
+  }
+
+private:
+  SimpleUndefinedAtom   _undefBinder;
+};
+
+
+//
+// MachHeaderAliasFile lazily instantiates the magic symbols that mark the start
+// of the mach_header for final linked images.
+//
+class MachHeaderAliasFile : public SimpleFile {
+public:
+  MachHeaderAliasFile(const MachOLinkingContext &context)
+    : SimpleFile("mach_header symbols", kindHeaderObject) {
+    StringRef machHeaderSymbolName;
+    DefinedAtom::Scope symbolScope = DefinedAtom::scopeLinkageUnit;
+    StringRef dsoHandleName;
+    switch (context.outputMachOType()) {
+    case llvm::MachO::MH_OBJECT:
+      machHeaderSymbolName = "__mh_object_header";
+      break;
+    case llvm::MachO::MH_EXECUTE:
+      machHeaderSymbolName = "__mh_execute_header";
+      symbolScope = DefinedAtom::scopeGlobal;
+      dsoHandleName = "___dso_handle";
+      break;
+    case llvm::MachO::MH_FVMLIB:
+      llvm_unreachable("no mach_header symbol for file type");
+    case llvm::MachO::MH_CORE:
+      llvm_unreachable("no mach_header symbol for file type");
+    case llvm::MachO::MH_PRELOAD:
+      llvm_unreachable("no mach_header symbol for file type");
+    case llvm::MachO::MH_DYLIB:
+      machHeaderSymbolName = "__mh_dylib_header";
+      dsoHandleName = "___dso_handle";
+      break;
+    case llvm::MachO::MH_DYLINKER:
+      machHeaderSymbolName = "__mh_dylinker_header";
+      dsoHandleName = "___dso_handle";
+      break;
+    case llvm::MachO::MH_BUNDLE:
+      machHeaderSymbolName = "__mh_bundle_header";
+      dsoHandleName = "___dso_handle";
+      break;
+    case llvm::MachO::MH_DYLIB_STUB:
+      llvm_unreachable("no mach_header symbol for file type");
+    case llvm::MachO::MH_DSYM:
+      llvm_unreachable("no mach_header symbol for file type");
+    case llvm::MachO::MH_KEXT_BUNDLE:
+      dsoHandleName = "___dso_handle";
+      break;
+    }
+    if (!machHeaderSymbolName.empty())
+      _definedAtoms.push_back(new (allocator()) MachODefinedAtom(
+          *this, machHeaderSymbolName, symbolScope,
+          DefinedAtom::typeMachHeader, DefinedAtom::mergeNo, false,
+          true /* noDeadStrip */,
+          ArrayRef<uint8_t>(), DefinedAtom::Alignment(4096)));
+
+    if (!dsoHandleName.empty())
+      _definedAtoms.push_back(new (allocator()) MachODefinedAtom(
+          *this, dsoHandleName, DefinedAtom::scopeLinkageUnit,
+          DefinedAtom::typeDSOHandle, DefinedAtom::mergeNo, false,
+          true /* noDeadStrip */,
+          ArrayRef<uint8_t>(), DefinedAtom::Alignment(1)));
+  }
+
+  const AtomRange<DefinedAtom> defined() const override {
+    return _definedAtoms;
+  }
+  const AtomRange<UndefinedAtom> undefined() const override {
+    return _noUndefinedAtoms;
+  }
+
+  const AtomRange<SharedLibraryAtom> sharedLibrary() const override {
+    return _noSharedLibraryAtoms;
+  }
+
+  const AtomRange<AbsoluteAtom> absolute() const override {
+    return _noAbsoluteAtoms;
+  }
+
+  void clearAtoms() override {
+    _definedAtoms.clear();
+    _noUndefinedAtoms.clear();
+    _noSharedLibraryAtoms.clear();
+    _noAbsoluteAtoms.clear();
+  }
+
+
+private:
+  mutable AtomVector<DefinedAtom> _definedAtoms;
+};
+
+} // namespace mach_o
+} // namespace lld
+
+#endif // LLD_READER_WRITER_MACHO_EXECUTABLE_ATOMS_H

external/llvm-project/lld/lib/ReaderWriter/MachO/File.h

@@ -0,0 +1,400 @@
+//===- lib/ReaderWriter/MachO/File.h ----------------------------*- C++ -*-===//
+//
+//                             The LLVM Linker
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLD_READER_WRITER_MACHO_FILE_H
+#define LLD_READER_WRITER_MACHO_FILE_H
+
+#include "Atoms.h"
+#include "DebugInfo.h"
+#include "MachONormalizedFile.h"
+#include "lld/Core/SharedLibraryFile.h"
+#include "lld/Core/Simple.h"
+#include "llvm/ADT/DenseMap.h"
+#include "llvm/ADT/StringMap.h"
+#include "llvm/Support/Format.h"
+#include <unordered_map>
+
+namespace lld {
+namespace mach_o {
+
+using lld::mach_o::normalized::Section;
+
+class MachOFile : public SimpleFile {
+public:
+
+  /// Real file constructor - for on-disk files.
+  MachOFile(std::unique_ptr<MemoryBuffer> mb, MachOLinkingContext *ctx)
+    : SimpleFile(mb->getBufferIdentifier(), File::kindMachObject),
+      _mb(std::move(mb)), _ctx(ctx) {}
+
+  /// Dummy file constructor - for virtual files.
+  MachOFile(StringRef path)
+    : SimpleFile(path, File::kindMachObject) {}
+
+  void addDefinedAtom(StringRef name, Atom::Scope scope,
+                      DefinedAtom::ContentType type, DefinedAtom::Merge merge,
+                      uint64_t sectionOffset, uint64_t contentSize, bool thumb,
+                      bool noDeadStrip, bool copyRefs,
+                      const Section *inSection) {
+    assert(sectionOffset+contentSize <= inSection->content.size());
+    ArrayRef<uint8_t> content = inSection->content.slice(sectionOffset,
+                                                        contentSize);
+    if (copyRefs) {
+      // Make a copy of the atom's name and content that is owned by this file.
+      name = name.copy(allocator());
+      content = content.copy(allocator());
+    }
+    DefinedAtom::Alignment align(
+        inSection->alignment,
+        sectionOffset % inSection->alignment);
+    auto *atom =
+        new (allocator()) MachODefinedAtom(*this, name, scope, type, merge,
+                                           thumb, noDeadStrip, content, align);
+    addAtomForSection(inSection, atom, sectionOffset);
+  }
+
+  void addDefinedAtomInCustomSection(StringRef name, Atom::Scope scope,
+                      DefinedAtom::ContentType type, DefinedAtom::Merge merge,
+                      bool thumb, bool noDeadStrip, uint64_t sectionOffset,
+                      uint64_t contentSize, StringRef sectionName,
+                      bool copyRefs, const Section *inSection) {
+    assert(sectionOffset+contentSize <= inSection->content.size());
+    ArrayRef<uint8_t> content = inSection->content.slice(sectionOffset,
+                                                        contentSize);
+   if (copyRefs) {
+      // Make a copy of the atom's name and content that is owned by this file.
+      name = name.copy(allocator());
+      content = content.copy(allocator());
+      sectionName = sectionName.copy(allocator());
+    }
+    DefinedAtom::Alignment align(
+        inSection->alignment,
+        sectionOffset % inSection->alignment);
+    auto *atom =
+        new (allocator()) MachODefinedCustomSectionAtom(*this, name, scope, type,
+                                                        merge, thumb,
+                                                        noDeadStrip, content,
+                                                        sectionName, align);
+    addAtomForSection(inSection, atom, sectionOffset);
+  }
+
+  void addZeroFillDefinedAtom(StringRef name, Atom::Scope scope,
+                              uint64_t sectionOffset, uint64_t size,
+                              bool noDeadStrip, bool copyRefs,
+                              const Section *inSection) {
+    if (copyRefs) {
+      // Make a copy of the atom's name and content that is owned by this file.
+      name = name.copy(allocator());
+    }
+    DefinedAtom::Alignment align(
+        inSection->alignment,
+        sectionOffset % inSection->alignment);
+
+    DefinedAtom::ContentType type = DefinedAtom::typeUnknown;
+    switch (inSection->type) {
+    case llvm::MachO::S_ZEROFILL:
+      type = DefinedAtom::typeZeroFill;
+      break;
+    case llvm::MachO::S_THREAD_LOCAL_ZEROFILL:
+      type = DefinedAtom::typeTLVInitialZeroFill;
+      break;
+    default:
+      llvm_unreachable("Unrecognized zero-fill section");
+    }
+
+    auto *atom =
+        new (allocator()) MachODefinedAtom(*this, name, scope, type, size,
+                                           noDeadStrip, align);
+    addAtomForSection(inSection, atom, sectionOffset);
+  }
+
+  void addUndefinedAtom(StringRef name, bool copyRefs) {
+    if (copyRefs) {
+      // Make a copy of the atom's name that is owned by this file.
+      name = name.copy(allocator());
+    }
+    auto *atom = new (allocator()) SimpleUndefinedAtom(*this, name);
+    addAtom(*atom);
+    _undefAtoms[name] = atom;
+  }
+
+  void addTentativeDefAtom(StringRef name, Atom::Scope scope, uint64_t size,
+                           DefinedAtom::Alignment align, bool copyRefs) {
+    if (copyRefs) {
+      // Make a copy of the atom's name that is owned by this file.
+      name = name.copy(allocator());
+    }
+    auto *atom =
+        new (allocator()) MachOTentativeDefAtom(*this, name, scope, size, align);
+    addAtom(*atom);
+    _undefAtoms[name] = atom;
+  }
+
+  /// Search this file for an the atom from 'section' that covers
+  /// 'offsetInSect'.  Returns nullptr is no atom found.
+  MachODefinedAtom *findAtomCoveringAddress(const Section &section,
+                                            uint64_t offsetInSect,
+                                            uint32_t *foundOffsetAtom=nullptr) {
+    const auto &pos = _sectionAtoms.find(&section);
+    if (pos == _sectionAtoms.end())
+      return nullptr;
+    const auto &vec = pos->second;
+    assert(offsetInSect < section.content.size());
+    // Vector of atoms for section are already sorted, so do binary search.
+    const auto &atomPos = std::lower_bound(vec.begin(), vec.end(), offsetInSect,
+        [offsetInSect](const SectionOffsetAndAtom &ao,
+                       uint64_t targetAddr) -> bool {
+          // Each atom has a start offset of its slice of the
+          // section's content. This compare function must return true
+          // iff the atom's range is before the offset being searched for.
+          uint64_t atomsEndOffset = ao.offset+ao.atom->rawContent().size();
+          return (atomsEndOffset <= offsetInSect);
+        });
+    if (atomPos == vec.end())
+      return nullptr;
+    if (foundOffsetAtom)
+      *foundOffsetAtom = offsetInSect - atomPos->offset;
+    return atomPos->atom;
+  }
+
+  /// Searches this file for an UndefinedAtom named 'name'. Returns
+  /// nullptr is no such atom found.
+  const lld::Atom *findUndefAtom(StringRef name) {
+    auto pos = _undefAtoms.find(name);
+    if (pos == _undefAtoms.end())
+      return nullptr;
+    return pos->second;
+  }
+
+  typedef std::function<void (MachODefinedAtom* atom)> DefinedAtomVisitor;
+
+  void eachDefinedAtom(DefinedAtomVisitor vistor) {
+    for (auto &sectAndAtoms : _sectionAtoms) {
+      for (auto &offAndAtom : sectAndAtoms.second) {
+        vistor(offAndAtom.atom);
+      }
+    }
+  }
+
+  typedef std::function<void(MachODefinedAtom *atom, uint64_t offset)>
+      SectionAtomVisitor;
+
+  void eachAtomInSection(const Section &section, SectionAtomVisitor visitor) {
+    auto pos = _sectionAtoms.find(&section);
+    if (pos == _sectionAtoms.end())
+      return;
+    auto vec = pos->second;
+
+    for (auto &offAndAtom : vec)
+      visitor(offAndAtom.atom, offAndAtom.offset);
+  }
+
+  MachOLinkingContext::Arch arch() const { return _arch; }
+  void setArch(MachOLinkingContext::Arch arch) { _arch = arch; }
+
+  MachOLinkingContext::OS OS() const { return _os; }
+  void setOS(MachOLinkingContext::OS os) { _os = os; }
+
+  MachOLinkingContext::ObjCConstraint objcConstraint() const {
+    return _objcConstraint;
+  }
+  void setObjcConstraint(MachOLinkingContext::ObjCConstraint v) {
+    _objcConstraint = v;
+  }
+
+  uint32_t minVersion() const { return _minVersion; }
+  void setMinVersion(uint32_t v) { _minVersion = v; }
+
+  LoadCommandType minVersionLoadCommandKind() const {
+    return _minVersionLoadCommandKind;
+  }
+  void setMinVersionLoadCommandKind(LoadCommandType v) {
+    _minVersionLoadCommandKind = v;
+  }
+
+  uint32_t swiftVersion() const { return _swiftVersion; }
+  void setSwiftVersion(uint32_t v) { _swiftVersion = v; }
+
+  bool subsectionsViaSymbols() const {
+    return _flags & llvm::MachO::MH_SUBSECTIONS_VIA_SYMBOLS;
+  }
+  void setFlags(normalized::FileFlags v) { _flags = v; }
+
+  /// Methods for support type inquiry through isa, cast, and dyn_cast:
+  static inline bool classof(const File *F) {
+    return F->kind() == File::kindMachObject;
+  }
+
+  void setDebugInfo(std::unique_ptr<DebugInfo> debugInfo) {
+    _debugInfo = std::move(debugInfo);
+  }
+
+  DebugInfo* debugInfo() const { return _debugInfo.get(); }
+  std::unique_ptr<DebugInfo> takeDebugInfo() { return std::move(_debugInfo); }
+
+protected:
+  std::error_code doParse() override {
+    // Convert binary file to normalized mach-o.
+    auto normFile = normalized::readBinary(_mb, _ctx->arch());
+    if (auto ec = normFile.takeError())
+      return llvm::errorToErrorCode(std::move(ec));
+    // Convert normalized mach-o to atoms.
+    if (auto ec = normalized::normalizedObjectToAtoms(this, **normFile, false))
+      return llvm::errorToErrorCode(std::move(ec));
+    return std::error_code();
+  }
+
+private:
+  struct SectionOffsetAndAtom { uint64_t offset;  MachODefinedAtom *atom; };
+
+  void addAtomForSection(const Section *inSection, MachODefinedAtom* atom,
+                         uint64_t sectionOffset) {
+    SectionOffsetAndAtom offAndAtom;
+    offAndAtom.offset = sectionOffset;
+    offAndAtom.atom   = atom;
+     _sectionAtoms[inSection].push_back(offAndAtom);
+    addAtom(*atom);
+  }
+
+  typedef llvm::DenseMap<const normalized::Section *,
+                         std::vector<SectionOffsetAndAtom>>  SectionToAtoms;
+  typedef llvm::StringMap<const lld::Atom *> NameToAtom;
+
+  std::unique_ptr<MemoryBuffer> _mb;
+  MachOLinkingContext          *_ctx;
+  SectionToAtoms                _sectionAtoms;
+  NameToAtom                     _undefAtoms;
+  MachOLinkingContext::Arch      _arch = MachOLinkingContext::arch_unknown;
+  MachOLinkingContext::OS        _os = MachOLinkingContext::OS::unknown;
+  uint32_t                       _minVersion = 0;
+  LoadCommandType               _minVersionLoadCommandKind = (LoadCommandType)0;
+  MachOLinkingContext::ObjCConstraint _objcConstraint =
+      MachOLinkingContext::objc_unknown;
+  uint32_t                       _swiftVersion = 0;
+  normalized::FileFlags        _flags = llvm::MachO::MH_SUBSECTIONS_VIA_SYMBOLS;
+  std::unique_ptr<DebugInfo>   _debugInfo;
+};
+
+class MachODylibFile : public SharedLibraryFile {
+public:
+  MachODylibFile(std::unique_ptr<MemoryBuffer> mb, MachOLinkingContext *ctx)
+      : SharedLibraryFile(mb->getBufferIdentifier()),
+        _mb(std::move(mb)), _ctx(ctx) {}
+
+  MachODylibFile(StringRef path) : SharedLibraryFile(path) {}
+
+  OwningAtomPtr<SharedLibraryAtom> exports(StringRef name) const override {
+    // Pass down _installName so that if this requested symbol
+    // is re-exported through this dylib, the SharedLibraryAtom's loadName()
+    // is this dylib installName and not the implementation dylib's.
+    // NOTE: isData is not needed for dylibs (it matters for static libs).
+    return exports(name, _installName);
+  }
+
+  /// Adds symbol name that this dylib exports. The corresponding
+  /// SharedLibraryAtom is created lazily (since most symbols are not used).
+  void addExportedSymbol(StringRef name, bool weakDef, bool copyRefs) {
+    if (copyRefs) {
+      name = name.copy(allocator());
+    }
+    AtomAndFlags info(weakDef);
+    _nameToAtom[name] = info;
+  }
+
+  void addReExportedDylib(StringRef dylibPath) {
+    _reExportedDylibs.emplace_back(dylibPath);
+  }
+
+  StringRef installName() const { return _installName; }
+  uint32_t currentVersion() { return _currentVersion; }
+  uint32_t compatVersion() { return _compatVersion; }
+
+  void setInstallName(StringRef name) { _installName = name; }
+  void setCompatVersion(uint32_t version) { _compatVersion = version; }
+  void setCurrentVersion(uint32_t version) { _currentVersion = version; }
+
+  typedef std::function<MachODylibFile *(StringRef)> FindDylib;
+
+  void loadReExportedDylibs(FindDylib find) {
+    for (ReExportedDylib &entry : _reExportedDylibs) {
+      entry.file = find(entry.path);
+    }
+  }
+
+  StringRef getDSOName() const override { return _installName; }
+
+  std::error_code doParse() override {
+    // Convert binary file to normalized mach-o.
+    auto normFile = normalized::readBinary(_mb, _ctx->arch());
+    if (auto ec = normFile.takeError())
+      return llvm::errorToErrorCode(std::move(ec));
+    // Convert normalized mach-o to atoms.
+    if (auto ec = normalized::normalizedDylibToAtoms(this, **normFile, false))
+      return llvm::errorToErrorCode(std::move(ec));
+    return std::error_code();
+  }
+
+private:
+  OwningAtomPtr<SharedLibraryAtom> exports(StringRef name,
+                                   StringRef installName) const {
+    // First, check if requested symbol is directly implemented by this dylib.
+    auto entry = _nameToAtom.find(name);
+    if (entry != _nameToAtom.end()) {
+      // FIXME: Make this map a set and only used in assert builds.
+      // Note, its safe to assert here as the resolver is the only client of
+      // this API and it only requests exports for undefined symbols.
+      // If we return from here we are no longer undefined so we should never
+      // get here again.
+      assert(!entry->second.atom && "Duplicate shared library export");
+      bool weakDef = entry->second.weakDef;
+      auto *atom = new (allocator()) MachOSharedLibraryAtom(*this, name,
+                                                            installName,
+                                                            weakDef);
+      entry->second.atom = atom;
+      return atom;
+    }
+
+    // Next, check if symbol is implemented in some re-exported dylib.
+    for (const ReExportedDylib &dylib : _reExportedDylibs) {
+      assert(dylib.file);
+      auto atom = dylib.file->exports(name, installName);
+      if (atom.get())
+        return atom;
+    }
+
+    // Symbol not exported or re-exported by this dylib.
+    return nullptr;
+  }
+
+  struct ReExportedDylib {
+    ReExportedDylib(StringRef p) : path(p), file(nullptr) { }
+    StringRef       path;
+    MachODylibFile *file;
+  };
+
+  struct AtomAndFlags {
+    AtomAndFlags() : atom(nullptr), weakDef(false) { }
+    AtomAndFlags(bool weak) : atom(nullptr), weakDef(weak) { }
+    const SharedLibraryAtom  *atom;
+    bool                      weakDef;
+  };
+
+  std::unique_ptr<MemoryBuffer>              _mb;
+  MachOLinkingContext                       *_ctx;
+  StringRef                                  _installName;
+  uint32_t                                   _currentVersion;
+  uint32_t                                   _compatVersion;
+  std::vector<ReExportedDylib>               _reExportedDylibs;
+  mutable std::unordered_map<StringRef, AtomAndFlags> _nameToAtom;
+};
+
+} // end namespace mach_o
+} // end namespace lld
+
+#endif // LLD_READER_WRITER_MACHO_FILE_H

external/llvm-project/lld/lib/ReaderWriter/MachO/FlatNamespaceFile.h

@@ -0,0 +1,63 @@
+//===- lib/ReaderWriter/MachO/FlatNamespaceFile.h -------------------------===//
+//
+//                             The LLVM Linker
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLD_READER_WRITER_MACHO_FLAT_NAMESPACE_FILE_H
+#define LLD_READER_WRITER_MACHO_FLAT_NAMESPACE_FILE_H
+
+#include "Atoms.h"
+#include "lld/Core/SharedLibraryFile.h"
+#include "lld/ReaderWriter/MachOLinkingContext.h"
+#include "llvm/Support/Debug.h"
+
+namespace lld {
+namespace mach_o {
+
+//
+// A FlateNamespaceFile instance may be added as a resolution source of last
+// resort, depending on how -flat_namespace and -undefined are set.
+//
+class FlatNamespaceFile : public SharedLibraryFile {
+public:
+  FlatNamespaceFile(const MachOLinkingContext &context)
+    : SharedLibraryFile("flat namespace") { }
+
+  OwningAtomPtr<SharedLibraryAtom> exports(StringRef name) const override {
+    return new (allocator()) MachOSharedLibraryAtom(*this, name, getDSOName(),
+                                                    false);
+  }
+
+  StringRef getDSOName() const override { return "flat-namespace"; }
+
+  const AtomRange<DefinedAtom> defined() const override {
+    return _noDefinedAtoms;
+  }
+  const AtomRange<UndefinedAtom> undefined() const override {
+    return _noUndefinedAtoms;
+  }
+
+  const AtomRange<SharedLibraryAtom> sharedLibrary() const override {
+    return _noSharedLibraryAtoms;
+  }
+
+  const AtomRange<AbsoluteAtom> absolute() const override {
+    return _noAbsoluteAtoms;
+  }
+
+  void clearAtoms() override {
+    _noDefinedAtoms.clear();
+    _noUndefinedAtoms.clear();
+    _noSharedLibraryAtoms.clear();
+    _noAbsoluteAtoms.clear();
+  }
+};
+
+} // namespace mach_o
+} // namespace lld
+
+#endif // LLD_READER_WRITER_MACHO_FLAT_NAMESPACE_FILE_H

external/llvm-project/lld/lib/ReaderWriter/MachO/GOTPass.cpp

@@ -0,0 +1,184 @@
+//===- lib/ReaderWriter/MachO/GOTPass.cpp -----------------------*- C++ -*-===//
+//
+//                             The LLVM Linker
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+///
+/// \file
+/// This linker pass transforms all GOT kind references to real references.
+/// That is, in assembly you can write something like:
+///     movq foo@GOTPCREL(%rip), %rax
+/// which means you want to load a pointer to "foo" out of the GOT (global
+/// Offsets Table). In the object file, the Atom containing this instruction
+/// has a Reference whose target is an Atom named "foo" and the Reference
+/// kind is a GOT load.  The linker needs to instantiate a pointer sized
+/// GOT entry.  This is done be creating a GOT Atom to represent that pointer
+/// sized data in this pass, and altering the Atom graph so the Reference now
+/// points to the GOT Atom entry (corresponding to "foo") and changing the
+/// Reference Kind to reflect it is now pointing to a GOT entry (rather
+/// then needing a GOT entry).
+///
+/// There is one optimization the linker can do here.  If the target of the GOT
+/// is in the same linkage unit and does not need to be interposable, and
+/// the GOT use is just a load (not some other operation), this pass can
+/// transform that load into an LEA (add).  This optimizes away one memory load
+/// which at runtime that could stall the pipeline.  This optimization only
+/// works for architectures in which a (GOT) load instruction can be change to
+/// an LEA instruction that is the same size.  The method isGOTAccess() should
+/// only return true for "canBypassGOT" if this optimization is supported.
+///
+//===----------------------------------------------------------------------===//
+
+#include "ArchHandler.h"
+#include "File.h"
+#include "MachOPasses.h"
+#include "lld/Common/LLVM.h"
+#include "lld/Core/DefinedAtom.h"
+#include "lld/Core/File.h"
+#include "lld/Core/Reference.h"
+#include "lld/Core/Simple.h"
+#include "llvm/ADT/DenseMap.h"
+#include "llvm/ADT/STLExtras.h"
+
+namespace lld {
+namespace mach_o {
+
+//
+//  GOT Entry Atom created by the GOT pass.
+//
+class GOTEntryAtom : public SimpleDefinedAtom {
+public:
+  GOTEntryAtom(const File &file, bool is64, StringRef name)
+    : SimpleDefinedAtom(file), _is64(is64), _name(name) { }
+
+  ~GOTEntryAtom() override = default;
+
+  ContentType contentType() const override {
+    return DefinedAtom::typeGOT;
+  }
+
+  Alignment alignment() const override {
+    return _is64 ? 8 : 4;
+  }
+
+  uint64_t size() const override {
+    return _is64 ? 8 : 4;
+  }
+
+  ContentPermissions permissions() const override {
+    return DefinedAtom::permRW_;
+  }
+
+  ArrayRef<uint8_t> rawContent() const override {
+    static const uint8_t zeros[] =
+        { 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00 };
+    return llvm::makeArrayRef(zeros, size());
+  }
+
+  StringRef slotName() const {
+    return _name;
+  }
+
+private:
+  const bool _is64;
+  StringRef _name;
+};
+
+/// Pass for instantiating and optimizing GOT slots.
+///
+class GOTPass : public Pass {
+public:
+  GOTPass(const MachOLinkingContext &context)
+      : _ctx(context), _archHandler(_ctx.archHandler()),
+        _file(*_ctx.make_file<MachOFile>("<mach-o GOT Pass>")) {
+    _file.setOrdinal(_ctx.getNextOrdinalAndIncrement());
+  }
+
+private:
+  llvm::Error perform(SimpleFile &mergedFile) override {
+    // Scan all references in all atoms.
+    for (const DefinedAtom *atom : mergedFile.defined()) {
+      for (const Reference *ref : *atom) {
+        // Look at instructions accessing the GOT.
+        bool canBypassGOT;
+        if (!_archHandler.isGOTAccess(*ref, canBypassGOT))
+          continue;
+        const Atom *target = ref->target();
+        assert(target != nullptr);
+
+        if (!shouldReplaceTargetWithGOTAtom(target, canBypassGOT)) {
+          // Update reference kind to reflect that target is a direct accesss.
+          _archHandler.updateReferenceToGOT(ref, false);
+        } else {
+          // Replace the target with a reference to a GOT entry.
+          const DefinedAtom *gotEntry = makeGOTEntry(target);
+          const_cast<Reference *>(ref)->setTarget(gotEntry);
+          // Update reference kind to reflect that target is now a GOT entry.
+          _archHandler.updateReferenceToGOT(ref, true);
+        }
+      }
+    }
+
+    // Sort and add all created GOT Atoms to master file
+    std::vector<const GOTEntryAtom *> entries;
+    entries.reserve(_targetToGOT.size());
+    for (auto &it : _targetToGOT)
+      entries.push_back(it.second);
+    std::sort(entries.begin(), entries.end(),
+              [](const GOTEntryAtom *left, const GOTEntryAtom *right) {
+      return (left->slotName().compare(right->slotName()) < 0);
+    });
+    for (const GOTEntryAtom *slot : entries)
+      mergedFile.addAtom(*slot);
+
+    return llvm::Error::success();
+  }
+
+  bool shouldReplaceTargetWithGOTAtom(const Atom *target, bool canBypassGOT) {
+    // Accesses to shared library symbols must go through GOT.
+    if (isa<SharedLibraryAtom>(target))
+      return true;
+    // Accesses to interposable symbols in same linkage unit must also go
+    // through GOT.
+    const DefinedAtom *defTarget = dyn_cast<DefinedAtom>(target);
+    if (defTarget != nullptr &&
+        defTarget->interposable() != DefinedAtom::interposeNo) {
+      assert(defTarget->scope() != DefinedAtom::scopeTranslationUnit);
+      return true;
+    }
+    // Target does not require indirection.  So, if instruction allows GOT to be
+    // by-passed, do that optimization and don't create GOT entry.
+    return !canBypassGOT;
+  }
+
+  const DefinedAtom *makeGOTEntry(const Atom *target) {
+    auto pos = _targetToGOT.find(target);
+    if (pos == _targetToGOT.end()) {
+      auto *gotEntry = new (_file.allocator())
+          GOTEntryAtom(_file, _ctx.is64Bit(), target->name());
+      _targetToGOT[target] = gotEntry;
+      const ArchHandler::ReferenceInfo &nlInfo = _archHandler.stubInfo().
+                                                nonLazyPointerReferenceToBinder;
+      gotEntry->addReference(Reference::KindNamespace::mach_o, nlInfo.arch,
+                             nlInfo.kind, 0, target, 0);
+      return gotEntry;
+    }
+    return pos->second;
+  }
+
+  const MachOLinkingContext &_ctx;
+  mach_o::ArchHandler                             &_archHandler;
+  MachOFile                                       &_file;
+  llvm::DenseMap<const Atom*, const GOTEntryAtom*> _targetToGOT;
+};
+
+void addGOTPass(PassManager &pm, const MachOLinkingContext &ctx) {
+  assert(ctx.needsGOTPass());
+  pm.add(llvm::make_unique<GOTPass>(ctx));
+}
+
+} // end namesapce mach_o
+} // end namesapce lld

external/llvm-project/lld/lib/ReaderWriter/MachO/LayoutPass.cpp

@@ -0,0 +1,489 @@
+//===-- ReaderWriter/MachO/LayoutPass.cpp - Layout atoms ------------------===//
+//
+//                             The LLVM Linker
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "LayoutPass.h"
+#include "lld/Core/Instrumentation.h"
+#include "lld/Core/PassManager.h"
+#include "lld/ReaderWriter/MachOLinkingContext.h"
+#include "llvm/ADT/STLExtras.h"
+#include "llvm/ADT/Twine.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/Parallel.h"
+#include <algorithm>
+#include <set>
+#include <utility>
+
+using namespace lld;
+
+#define DEBUG_TYPE "LayoutPass"
+
+namespace lld {
+namespace mach_o {
+
+static bool compareAtoms(const LayoutPass::SortKey &,
+                         const LayoutPass::SortKey &,
+                         LayoutPass::SortOverride customSorter);
+
+#ifndef NDEBUG
+// Return "reason (leftval, rightval)"
+static std::string formatReason(StringRef reason, int leftVal, int rightVal) {
+  return (Twine(reason) + " (" + Twine(leftVal) + ", " + Twine(rightVal) + ")")
+      .str();
+}
+
+// Less-than relationship of two atoms must be transitive, which is, if a < b
+// and b < c, a < c must be true. This function checks the transitivity by
+// checking the sort results.
+static void checkTransitivity(std::vector<LayoutPass::SortKey> &vec,
+                              LayoutPass::SortOverride customSorter) {
+  for (auto i = vec.begin(), e = vec.end(); (i + 1) != e; ++i) {
+    for (auto j = i + 1; j != e; ++j) {
+      assert(compareAtoms(*i, *j, customSorter));
+      assert(!compareAtoms(*j, *i, customSorter));
+    }
+  }
+}
+
+// Helper functions to check follow-on graph.
+typedef llvm::DenseMap<const DefinedAtom *, const DefinedAtom *> AtomToAtomT;
+
+static std::string atomToDebugString(const Atom *atom) {
+  const DefinedAtom *definedAtom = dyn_cast<DefinedAtom>(atom);
+  std::string str;
+  llvm::raw_string_ostream s(str);
+  if (definedAtom->name().empty())
+    s << "<anonymous " << definedAtom << ">";
+  else
+    s << definedAtom->name();
+  s << " in ";
+  if (definedAtom->customSectionName().empty())
+    s << "<anonymous>";
+  else
+    s << definedAtom->customSectionName();
+  s.flush();
+  return str;
+}
+
+static void showCycleDetectedError(const Registry &registry,
+                                   AtomToAtomT &followOnNexts,
+                                   const DefinedAtom *atom) {
+  const DefinedAtom *start = atom;
+  llvm::dbgs() << "There's a cycle in a follow-on chain!\n";
+  do {
+    llvm::dbgs() << "  " << atomToDebugString(atom) << "\n";
+    for (const Reference *ref : *atom) {
+      StringRef kindValStr;
+      if (!registry.referenceKindToString(ref->kindNamespace(), ref->kindArch(),
+                                          ref->kindValue(), kindValStr)) {
+        kindValStr = "<unknown>";
+      }
+      llvm::dbgs() << "    " << kindValStr
+                   << ": " << atomToDebugString(ref->target()) << "\n";
+    }
+    atom = followOnNexts[atom];
+  } while (atom != start);
+  llvm::report_fatal_error("Cycle detected");
+}
+
+/// Exit if there's a cycle in a followon chain reachable from the
+/// given root atom. Uses the tortoise and hare algorithm to detect a
+/// cycle.
+static void checkNoCycleInFollowonChain(const Registry &registry,
+                                        AtomToAtomT &followOnNexts,
+                                        const DefinedAtom *root) {
+  const DefinedAtom *tortoise = root;
+  const DefinedAtom *hare = followOnNexts[root];
+  while (true) {
+    if (!tortoise || !hare)
+      return;
+    if (tortoise == hare)
+      showCycleDetectedError(registry, followOnNexts, tortoise);
+    tortoise = followOnNexts[tortoise];
+    hare = followOnNexts[followOnNexts[hare]];
+  }
+}
+
+static void checkReachabilityFromRoot(AtomToAtomT &followOnRoots,
+                                      const DefinedAtom *atom) {
+  if (!atom) return;
+  auto i = followOnRoots.find(atom);
+  if (i == followOnRoots.end()) {
+    llvm_unreachable(((Twine("Atom <") + atomToDebugString(atom) +
+                       "> has no follow-on root!"))
+                         .str()
+                         .c_str());
+  }
+  const DefinedAtom *ap = i->second;
+  while (true) {
+    const DefinedAtom *next = followOnRoots[ap];
+    if (!next) {
+      llvm_unreachable((Twine("Atom <" + atomToDebugString(atom) +
+                              "> is not reachable from its root!"))
+                           .str()
+                           .c_str());
+    }
+    if (next == ap)
+      return;
+    ap = next;
+  }
+}
+
+static void printDefinedAtoms(const File::AtomRange<DefinedAtom> &atomRange) {
+  for (const DefinedAtom *atom : atomRange) {
+    llvm::dbgs() << "  file=" << atom->file().path()
+                 << ", name=" << atom->name()
+                 << ", size=" << atom->size()
+                 << ", type=" << atom->contentType()
+                 << ", ordinal=" << atom->ordinal()
+                 << "\n";
+  }
+}
+
+/// Verify that the followon chain is sane. Should not be called in
+/// release binary.
+void LayoutPass::checkFollowonChain(const File::AtomRange<DefinedAtom> &range) {
+  ScopedTask task(getDefaultDomain(), "LayoutPass::checkFollowonChain");
+
+  // Verify that there's no cycle in follow-on chain.
+  std::set<const DefinedAtom *> roots;
+  for (const auto &ai : _followOnRoots)
+    roots.insert(ai.second);
+  for (const DefinedAtom *root : roots)
+    checkNoCycleInFollowonChain(_registry, _followOnNexts, root);
+
+  // Verify that all the atoms in followOnNexts have references to
+  // their roots.
+  for (const auto &ai : _followOnNexts) {
+    checkReachabilityFromRoot(_followOnRoots, ai.first);
+    checkReachabilityFromRoot(_followOnRoots, ai.second);
+  }
+}
+#endif // #ifndef NDEBUG
+
+/// The function compares atoms by sorting atoms in the following order
+/// a) Sorts atoms by their ordinal overrides (layout-after/ingroup)
+/// b) Sorts atoms by their permissions
+/// c) Sorts atoms by their content
+/// d) Sorts atoms by custom sorter
+/// e) Sorts atoms on how they appear using File Ordinality
+/// f) Sorts atoms on how they appear within the File
+static bool compareAtomsSub(const LayoutPass::SortKey &lc,
+                            const LayoutPass::SortKey &rc,
+                            LayoutPass::SortOverride customSorter,
+                            std::string &reason) {
+  const DefinedAtom *left = lc._atom.get();
+  const DefinedAtom *right = rc._atom.get();
+  if (left == right) {
+    reason = "same";
+    return false;
+  }
+
+  // Find the root of the chain if it is a part of a follow-on chain.
+  const DefinedAtom *leftRoot = lc._root;
+  const DefinedAtom *rightRoot = rc._root;
+
+  // Sort atoms by their ordinal overrides only if they fall in the same
+  // chain.
+  if (leftRoot == rightRoot) {
+    DEBUG(reason = formatReason("override", lc._override, rc._override));
+    return lc._override < rc._override;
+  }
+
+  // Sort same permissions together.
+  DefinedAtom::ContentPermissions leftPerms = leftRoot->permissions();
+  DefinedAtom::ContentPermissions rightPerms = rightRoot->permissions();
+
+  if (leftPerms != rightPerms) {
+    DEBUG(reason =
+              formatReason("contentPerms", (int)leftPerms, (int)rightPerms));
+    return leftPerms < rightPerms;
+  }
+
+  // Sort same content types together.
+  DefinedAtom::ContentType leftType = leftRoot->contentType();
+  DefinedAtom::ContentType rightType = rightRoot->contentType();
+
+  if (leftType != rightType) {
+    DEBUG(reason = formatReason("contentType", (int)leftType, (int)rightType));
+    return leftType < rightType;
+  }
+
+  // Use custom sorter if supplied.
+  if (customSorter) {
+    bool leftBeforeRight;
+    if (customSorter(leftRoot, rightRoot, leftBeforeRight))
+      return leftBeforeRight;
+  }
+
+  // Sort by .o order.
+  const File *leftFile = &leftRoot->file();
+  const File *rightFile = &rightRoot->file();
+
+  if (leftFile != rightFile) {
+    DEBUG(reason = formatReason(".o order", (int)leftFile->ordinal(),
+                                (int)rightFile->ordinal()));
+    return leftFile->ordinal() < rightFile->ordinal();
+  }
+
+  // Sort by atom order with .o file.
+  uint64_t leftOrdinal = leftRoot->ordinal();
+  uint64_t rightOrdinal = rightRoot->ordinal();
+
+  if (leftOrdinal != rightOrdinal) {
+    DEBUG(reason = formatReason("ordinal", (int)leftRoot->ordinal(),
+                                (int)rightRoot->ordinal()));
+    return leftOrdinal < rightOrdinal;
+  }
+
+  llvm::errs() << "Unordered: <" << left->name() << "> <"
+               << right->name() << ">\n";
+  llvm_unreachable("Atoms with Same Ordinal!");
+}
+
+static bool compareAtoms(const LayoutPass::SortKey &lc,
+                         const LayoutPass::SortKey &rc,
+                         LayoutPass::SortOverride customSorter) {
+  std::string reason;
+  bool result = compareAtomsSub(lc, rc, customSorter, reason);
+  DEBUG({
+    StringRef comp = result ? "<" : ">=";
+    llvm::dbgs() << "Layout: '" << lc._atom.get()->name()
+                 << "' " << comp << " '"
+                 << rc._atom.get()->name() << "' (" << reason << ")\n";
+  });
+  return result;
+}
+
+LayoutPass::LayoutPass(const Registry &registry, SortOverride sorter)
+    : _registry(registry), _customSorter(std::move(sorter)) {}
+
+// Returns the atom immediately followed by the given atom in the followon
+// chain.
+const DefinedAtom *LayoutPass::findAtomFollowedBy(
+    const DefinedAtom *targetAtom) {
+  // Start from the beginning of the chain and follow the chain until
+  // we find the targetChain.
+  const DefinedAtom *atom = _followOnRoots[targetAtom];
+  while (true) {
+    const DefinedAtom *prevAtom = atom;
+    AtomToAtomT::iterator targetFollowOnAtomsIter = _followOnNexts.find(atom);
+    // The target atom must be in the chain of its root.
+    assert(targetFollowOnAtomsIter != _followOnNexts.end());
+    atom = targetFollowOnAtomsIter->second;
+    if (atom == targetAtom)
+      return prevAtom;
+  }
+}
+
+// Check if all the atoms followed by the given target atom are of size zero.
+// When this method is called, an atom being added is not of size zero and
+// will be added to the head of the followon chain. All the atoms between the
+// atom and the targetAtom (specified by layout-after) need to be of size zero
+// in this case. Otherwise the desired layout is impossible.
+bool LayoutPass::checkAllPrevAtomsZeroSize(const DefinedAtom *targetAtom) {
+  const DefinedAtom *atom = _followOnRoots[targetAtom];
+  while (true) {
+    if (atom == targetAtom)
+      return true;
+    if (atom->size() != 0)
+      // TODO: print warning that an impossible layout is being desired by the
+      // user.
+      return false;
+    AtomToAtomT::iterator targetFollowOnAtomsIter = _followOnNexts.find(atom);
+    // The target atom must be in the chain of its root.
+    assert(targetFollowOnAtomsIter != _followOnNexts.end());
+    atom = targetFollowOnAtomsIter->second;
+  }
+}
+
+// Set the root of all atoms in targetAtom's chain to the given root.
+void LayoutPass::setChainRoot(const DefinedAtom *targetAtom,
+                              const DefinedAtom *root) {
+  // Walk through the followon chain and override each node's root.
+  while (true) {
+    _followOnRoots[targetAtom] = root;
+    AtomToAtomT::iterator targetFollowOnAtomsIter =
+        _followOnNexts.find(targetAtom);
+    if (targetFollowOnAtomsIter == _followOnNexts.end())
+      return;
+    targetAtom = targetFollowOnAtomsIter->second;
+  }
+}
+
+/// This pass builds the followon tables described by two DenseMaps
+/// followOnRoots and followonNexts.
+/// The followOnRoots map contains a mapping of a DefinedAtom to its root
+/// The followOnNexts map contains a mapping of what DefinedAtom follows the
+/// current Atom
+/// The algorithm follows a very simple approach
+/// a) If the atom is first seen, then make that as the root atom
+/// b) The targetAtom which this Atom contains, has the root thats set to the
+///    root of the current atom
+/// c) If the targetAtom is part of a different tree and the root of the
+///    targetAtom is itself, Chain all the atoms that are contained in the tree
+///    to the current Tree
+/// d) If the targetAtom is part of a different chain and the root of the
+///    targetAtom until the targetAtom has all atoms of size 0, then chain the
+///    targetAtoms and its tree to the current chain
+void LayoutPass::buildFollowOnTable(const File::AtomRange<DefinedAtom> &range) {
+  ScopedTask task(getDefaultDomain(), "LayoutPass::buildFollowOnTable");
+  // Set the initial size of the followon and the followonNext hash to the
+  // number of atoms that we have.
+  _followOnRoots.reserve(range.size());
+  _followOnNexts.reserve(range.size());
+  for (const DefinedAtom *ai : range) {
+    for (const Reference *r : *ai) {
+      if (r->kindNamespace() != lld::Reference::KindNamespace::all ||
+          r->kindValue() != lld::Reference::kindLayoutAfter)
+        continue;
+      const DefinedAtom *targetAtom = dyn_cast<DefinedAtom>(r->target());
+      _followOnNexts[ai] = targetAtom;
+
+      // If we find a followon for the first time, let's make that atom as the
+      // root atom.
+      if (_followOnRoots.count(ai) == 0)
+        _followOnRoots[ai] = ai;
+
+      auto iter = _followOnRoots.find(targetAtom);
+      if (iter == _followOnRoots.end()) {
+        // If the targetAtom is not a root of any chain, let's make the root of
+        // the targetAtom to the root of the current chain.
+
+        // The expression m[i] = m[j] where m is a DenseMap and i != j is not
+        // safe. m[j] returns a reference, which would be invalidated when a
+        // rehashing occurs. If rehashing occurs to make room for m[i], m[j]
+        // becomes invalid, and that invalid reference would be used as the RHS
+        // value of the expression.
+        // Copy the value to workaround.
+        const DefinedAtom *tmp = _followOnRoots[ai];
+        _followOnRoots[targetAtom] = tmp;
+        continue;
+      }
+      if (iter->second == targetAtom) {
+        // If the targetAtom is the root of a chain, the chain becomes part of
+        // the current chain. Rewrite the subchain's root to the current
+        // chain's root.
+        setChainRoot(targetAtom, _followOnRoots[ai]);
+        continue;
+      }
+      // The targetAtom is already a part of a chain. If the current atom is
+      // of size zero, we can insert it in the middle of the chain just
+      // before the target atom, while not breaking other atom's followon
+      // relationships. If it's not, we can only insert the current atom at
+      // the beginning of the chain. All the atoms followed by the target
+      // atom must be of size zero in that case to satisfy the followon
+      // relationships.
+      size_t currentAtomSize = ai->size();
+      if (currentAtomSize == 0) {
+        const DefinedAtom *targetPrevAtom = findAtomFollowedBy(targetAtom);
+        _followOnNexts[targetPrevAtom] = ai;
+        const DefinedAtom *tmp = _followOnRoots[targetPrevAtom];
+        _followOnRoots[ai] = tmp;
+        continue;
+      }
+      if (!checkAllPrevAtomsZeroSize(targetAtom))
+        break;
+      _followOnNexts[ai] = _followOnRoots[targetAtom];
+      setChainRoot(_followOnRoots[targetAtom], _followOnRoots[ai]);
+    }
+  }
+}
+
+/// Build an ordinal override map by traversing the followon chain, and
+/// assigning ordinals to each atom, if the atoms have their ordinals
+/// already assigned skip the atom and move to the next. This is the
+/// main map thats used to sort the atoms while comparing two atoms together
+void
+LayoutPass::buildOrdinalOverrideMap(const File::AtomRange<DefinedAtom> &range) {
+  ScopedTask task(getDefaultDomain(), "LayoutPass::buildOrdinalOverrideMap");
+  uint64_t index = 0;
+  for (const DefinedAtom *ai : range) {
+    const DefinedAtom *atom = ai;
+    if (_ordinalOverrideMap.find(atom) != _ordinalOverrideMap.end())
+      continue;
+    AtomToAtomT::iterator start = _followOnRoots.find(atom);
+    if (start == _followOnRoots.end())
+      continue;
+    for (const DefinedAtom *nextAtom = start->second; nextAtom;
+         nextAtom = _followOnNexts[nextAtom]) {
+      AtomToOrdinalT::iterator pos = _ordinalOverrideMap.find(nextAtom);
+      if (pos == _ordinalOverrideMap.end())
+        _ordinalOverrideMap[nextAtom] = index++;
+    }
+  }
+}
+
+std::vector<LayoutPass::SortKey>
+LayoutPass::decorate(File::AtomRange<DefinedAtom> &atomRange) const {
+  std::vector<SortKey> ret;
+  for (OwningAtomPtr<DefinedAtom> &atom : atomRange.owning_ptrs()) {
+    auto ri = _followOnRoots.find(atom.get());
+    auto oi = _ordinalOverrideMap.find(atom.get());
+    const auto *root = (ri == _followOnRoots.end()) ? atom.get() : ri->second;
+    uint64_t override = (oi == _ordinalOverrideMap.end()) ? 0 : oi->second;
+    ret.push_back(SortKey(std::move(atom), root, override));
+  }
+  return ret;
+}
+
+void LayoutPass::undecorate(File::AtomRange<DefinedAtom> &atomRange,
+                            std::vector<SortKey> &keys) const {
+  size_t i = 0;
+  for (SortKey &k : keys)
+    atomRange[i++] = std::move(k._atom);
+}
+
+/// Perform the actual pass
+llvm::Error LayoutPass::perform(SimpleFile &mergedFile) {
+  DEBUG(llvm::dbgs() << "******** Laying out atoms:\n");
+  // sort the atoms
+  ScopedTask task(getDefaultDomain(), "LayoutPass");
+  File::AtomRange<DefinedAtom> atomRange = mergedFile.defined();
+
+  // Build follow on tables
+  buildFollowOnTable(atomRange);
+
+  // Check the structure of followon graph if running in debug mode.
+  DEBUG(checkFollowonChain(atomRange));
+
+  // Build override maps
+  buildOrdinalOverrideMap(atomRange);
+
+  DEBUG({
+    llvm::dbgs() << "unsorted atoms:\n";
+    printDefinedAtoms(atomRange);
+  });
+
+  std::vector<LayoutPass::SortKey> vec = decorate(atomRange);
+  sort(llvm::parallel::par, vec.begin(), vec.end(),
+       [&](const LayoutPass::SortKey &l, const LayoutPass::SortKey &r) -> bool {
+         return compareAtoms(l, r, _customSorter);
+       });
+  DEBUG(checkTransitivity(vec, _customSorter));
+  undecorate(atomRange, vec);
+
+  DEBUG({
+    llvm::dbgs() << "sorted atoms:\n";
+    printDefinedAtoms(atomRange);
+  });
+
+  DEBUG(llvm::dbgs() << "******** Finished laying out atoms\n");
+  return llvm::Error::success();
+}
+
+void addLayoutPass(PassManager &pm, const MachOLinkingContext &ctx) {
+  pm.add(llvm::make_unique<LayoutPass>(
+      ctx.registry(), [&](const DefinedAtom * left, const DefinedAtom * right,
+                          bool & leftBeforeRight) ->bool {
+    return ctx.customAtomOrderer(left, right, leftBeforeRight);
+  }));
+}
+
+} // namespace mach_o
+} // namespace lld

external/llvm-project/lld/lib/ReaderWriter/MachO/LayoutPass.h

@@ -0,0 +1,119 @@
+//===------ lib/ReaderWriter/MachO/LayoutPass.h - Handles Layout of atoms -===//
+//
+//                             The LLVM Linker
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLD_READER_WRITER_MACHO_LAYOUT_PASS_H
+#define LLD_READER_WRITER_MACHO_LAYOUT_PASS_H
+
+#include "lld/Core/File.h"
+#include "lld/Core/Pass.h"
+#include "lld/Core/Reader.h"
+#include "lld/Core/Simple.h"
+#include "llvm/ADT/DenseMap.h"
+#include <map>
+#include <string>
+#include <vector>
+
+namespace lld {
+class DefinedAtom;
+class SimpleFile;
+
+namespace mach_o {
+
+/// This linker pass does the layout of the atoms. The pass is done after the
+/// order their .o files were found on the command line, then by order of the
+/// atoms (address) in the .o file.  But some atoms have a preferred location
+/// in their section (such as pinned to the start or end of the section), so
+/// the sort must take that into account too.
+class LayoutPass : public Pass {
+public:
+  struct SortKey {
+    SortKey(OwningAtomPtr<DefinedAtom> &&atom,
+            const DefinedAtom *root, uint64_t override)
+    : _atom(std::move(atom)), _root(root), _override(override) {}
+    OwningAtomPtr<DefinedAtom> _atom;
+    const DefinedAtom *_root;
+    uint64_t _override;
+
+    // Note, these are only here to appease MSVC bots which didn't like
+    // the same methods being implemented/deleted in OwningAtomPtr.
+    SortKey(SortKey &&key) : _atom(std::move(key._atom)), _root(key._root),
+                             _override(key._override) {
+      key._root = nullptr;
+    }
+
+    SortKey &operator=(SortKey &&key) {
+      _atom = std::move(key._atom);
+      _root = key._root;
+      key._root = nullptr;
+      _override = key._override;
+      return *this;
+    }
+
+  private:
+    SortKey(const SortKey &) = delete;
+    void operator=(const SortKey&) = delete;
+  };
+
+  typedef std::function<bool (const DefinedAtom *left, const DefinedAtom *right,
+                              bool &leftBeforeRight)> SortOverride;
+
+  LayoutPass(const Registry &registry, SortOverride sorter);
+
+  /// Sorts atoms in mergedFile by content type then by command line order.
+  llvm::Error perform(SimpleFile &mergedFile) override;
+
+  ~LayoutPass() override = default;
+
+private:
+  // Build the followOn atoms chain as specified by the kindLayoutAfter
+  // reference type
+  void buildFollowOnTable(const File::AtomRange<DefinedAtom> &range);
+
+  // Build a map of Atoms to ordinals for sorting the atoms
+  void buildOrdinalOverrideMap(const File::AtomRange<DefinedAtom> &range);
+
+  const Registry &_registry;
+  SortOverride _customSorter;
+
+  typedef llvm::DenseMap<const DefinedAtom *, const DefinedAtom *> AtomToAtomT;
+  typedef llvm::DenseMap<const DefinedAtom *, uint64_t> AtomToOrdinalT;
+
+  // A map to be used to sort atoms. It represents the order of atoms in the
+  // result; if Atom X is mapped to atom Y in this map, X will be located
+  // immediately before Y in the output file. Y might be mapped to another
+  // atom, constructing a follow-on chain. An atom cannot be mapped to more
+  // than one atom unless all but one atom are of size zero.
+  AtomToAtomT _followOnNexts;
+
+  // A map to be used to sort atoms. It's a map from an atom to its root of
+  // follow-on chain. A root atom is mapped to itself. If an atom is not in
+  // _followOnNexts, the atom is not in this map, and vice versa.
+  AtomToAtomT _followOnRoots;
+
+  AtomToOrdinalT _ordinalOverrideMap;
+
+  // Helper methods for buildFollowOnTable().
+  const DefinedAtom *findAtomFollowedBy(const DefinedAtom *targetAtom);
+  bool checkAllPrevAtomsZeroSize(const DefinedAtom *targetAtom);
+
+  void setChainRoot(const DefinedAtom *targetAtom, const DefinedAtom *root);
+
+  std::vector<SortKey> decorate(File::AtomRange<DefinedAtom> &atomRange) const;
+
+  void undecorate(File::AtomRange<DefinedAtom> &atomRange,
+                  std::vector<SortKey> &keys) const;
+
+  // Check if the follow-on graph is a correct structure. For debugging only.
+  void checkFollowonChain(const File::AtomRange<DefinedAtom> &range);
+};
+
+} // namespace mach_o
+} // namespace lld
+
+#endif // LLD_READER_WRITER_MACHO_LAYOUT_PASS_H

external/llvm-project/lld/lib/ReaderWriter/MachO/MachONormalizedFile.h

@@ -0,0 +1,336 @@
+//===- lib/ReaderWriter/MachO/MachONormalizedFile.h -----------------------===//
+//
+//                             The LLVM Linker
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+///
+/// \file These data structures comprise the "normalized" view of
+/// mach-o object files. The normalized view is an in-memory only data structure
+/// which is always in native endianness and pointer size.
+///
+/// The normalized view easily converts to and from YAML using YAML I/O.
+///
+/// The normalized view converts to and from binary mach-o object files using
+/// the writeBinary() and readBinary() functions.
+///
+/// The normalized view converts to and from lld::Atoms using the
+/// normalizedToAtoms() and normalizedFromAtoms().
+///
+/// Overall, the conversion paths available look like:
+///
+///                 +---------------+
+///                 | binary mach-o |
+///                 +---------------+
+///                        ^
+///                        |
+///                        v
+///                  +------------+         +------+
+///                  | normalized |   <->   | yaml |
+///                  +------------+         +------+
+///                        ^
+///                        |
+///                        v
+///                    +-------+
+///                    | Atoms |
+///                    +-------+
+///
+
+#ifndef LLD_READER_WRITER_MACHO_NORMALIZE_FILE_H
+#define LLD_READER_WRITER_MACHO_NORMALIZE_FILE_H
+
+#include "DebugInfo.h"
+#include "lld/Common/LLVM.h"
+#include "lld/Core/Error.h"
+#include "lld/ReaderWriter/MachOLinkingContext.h"
+#include "llvm/ADT/SmallString.h"
+#include "llvm/ADT/StringRef.h"
+#include "llvm/BinaryFormat/MachO.h"
+#include "llvm/Support/Allocator.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/ErrorOr.h"
+#include "llvm/Support/YAMLTraits.h"
+
+using llvm::BumpPtrAllocator;
+using llvm::yaml::Hex64;
+using llvm::yaml::Hex32;
+using llvm::yaml::Hex16;
+using llvm::yaml::Hex8;
+using llvm::yaml::SequenceTraits;
+using llvm::MachO::HeaderFileType;
+using llvm::MachO::BindType;
+using llvm::MachO::RebaseType;
+using llvm::MachO::NListType;
+using llvm::MachO::RelocationInfoType;
+using llvm::MachO::SectionType;
+using llvm::MachO::LoadCommandType;
+using llvm::MachO::ExportSymbolKind;
+using llvm::MachO::DataRegionType;
+
+namespace lld {
+namespace mach_o {
+namespace normalized {
+
+
+/// The real mach-o relocation record is 8-bytes on disk and is
+/// encoded in one of two different bit-field patterns.  This
+/// normalized form has the union of all possible fields.
+struct Relocation {
+  Relocation() : offset(0), scattered(false),
+                 type(llvm::MachO::GENERIC_RELOC_VANILLA),
+                 length(0), pcRel(false), isExtern(false), value(0),
+                 symbol(0) { }
+
+  Hex32               offset;
+  bool                scattered;
+  RelocationInfoType  type;
+  uint8_t             length;
+  bool                pcRel;
+  bool                isExtern;
+  Hex32               value;
+  uint32_t            symbol;
+};
+
+/// A typedef so that YAML I/O can treat this vector as a sequence.
+typedef std::vector<Relocation> Relocations;
+
+/// A typedef so that YAML I/O can process the raw bytes in a section.
+typedef std::vector<Hex8> ContentBytes;
+
+/// A typedef so that YAML I/O can treat indirect symbols as a flow sequence.
+typedef std::vector<uint32_t> IndirectSymbols;
+
+/// A typedef so that YAML I/O can encode/decode section attributes.
+LLVM_YAML_STRONG_TYPEDEF(uint32_t, SectionAttr)
+
+/// A typedef so that YAML I/O can encode/decode section alignment.
+LLVM_YAML_STRONG_TYPEDEF(uint16_t, SectionAlignment)
+
+/// Mach-O has a 32-bit and 64-bit section record.  This normalized form
+/// can support either kind.
+struct Section {
+  Section() : type(llvm::MachO::S_REGULAR),
+              attributes(0), alignment(1), address(0) { }
+
+  StringRef       segmentName;
+  StringRef       sectionName;
+  SectionType     type;
+  SectionAttr     attributes;
+  SectionAlignment        alignment;
+  Hex64           address;
+  ArrayRef<uint8_t> content;
+  Relocations     relocations;
+  IndirectSymbols indirectSymbols;
+};
+
+
+/// A typedef so that YAML I/O can encode/decode the scope bits of an nlist.
+LLVM_YAML_STRONG_TYPEDEF(uint8_t, SymbolScope)
+
+/// A typedef so that YAML I/O can encode/decode the desc bits of an nlist.
+LLVM_YAML_STRONG_TYPEDEF(uint16_t, SymbolDesc)
+
+/// Mach-O has a 32-bit and 64-bit symbol table entry (nlist), and the symbol
+/// type and scope and mixed in the same n_type field.  This normalized form
+/// works for any pointer size and separates out the type and scope.
+struct Symbol {
+  Symbol() : type(llvm::MachO::N_UNDF), scope(0), sect(0), desc(0), value(0) { }
+
+  StringRef     name;
+  NListType     type;
+  SymbolScope   scope;
+  uint8_t       sect;
+  SymbolDesc    desc;
+  Hex64         value;
+};
+
+/// Check whether the given section type indicates a zero-filled section.
+// FIXME: Utility functions of this kind should probably be moved into
+//        llvm/Support.
+inline bool isZeroFillSection(SectionType T) {
+  return (T == llvm::MachO::S_ZEROFILL ||
+          T == llvm::MachO::S_THREAD_LOCAL_ZEROFILL);
+}
+
+/// A typedef so that YAML I/O can (de/en)code the protection bits of a segment.
+LLVM_YAML_STRONG_TYPEDEF(uint32_t, VMProtect)
+
+/// A typedef to hold verions X.Y.X packed into 32-bit xxxx.yy.zz
+LLVM_YAML_STRONG_TYPEDEF(uint32_t, PackedVersion)
+
+/// Segments are only used in normalized final linked images (not in relocatable
+/// object files). They specify how a range of the file is loaded.
+struct Segment {
+  StringRef     name;
+  Hex64         address;
+  Hex64         size;
+  VMProtect     init_access;
+  VMProtect     max_access;
+};
+
+/// Only used in normalized final linked images to specify on which dylibs
+/// it depends.
+struct DependentDylib {
+  StringRef       path;
+  LoadCommandType kind;
+  PackedVersion   compatVersion;
+  PackedVersion   currentVersion;
+};
+
+/// A normalized rebasing entry.  Only used in normalized final linked images.
+struct RebaseLocation {
+  Hex32         segOffset;
+  uint8_t       segIndex;
+  RebaseType    kind;
+};
+
+/// A normalized binding entry.  Only used in normalized final linked images.
+struct BindLocation {
+  Hex32           segOffset;
+  uint8_t         segIndex;
+  BindType        kind;
+  bool            canBeNull;
+  int             ordinal;
+  StringRef       symbolName;
+  Hex64           addend;
+};
+
+/// A typedef so that YAML I/O can encode/decode export flags.
+LLVM_YAML_STRONG_TYPEDEF(uint32_t, ExportFlags)
+
+/// A normalized export entry.  Only used in normalized final linked images.
+struct Export {
+  StringRef         name;
+  Hex64             offset;
+  ExportSymbolKind  kind;
+  ExportFlags       flags;
+  Hex32             otherOffset;
+  StringRef         otherName;
+};
+
+/// A normalized data-in-code entry.
+struct DataInCode {
+  Hex32           offset;
+  Hex16           length;
+  DataRegionType  kind;
+};
+
+/// A typedef so that YAML I/O can encode/decode mach_header.flags.
+LLVM_YAML_STRONG_TYPEDEF(uint32_t, FileFlags)
+
+///
+struct NormalizedFile {
+  MachOLinkingContext::Arch   arch = MachOLinkingContext::arch_unknown;
+  HeaderFileType              fileType = llvm::MachO::MH_OBJECT;
+  FileFlags                   flags = 0;
+  std::vector<Segment>        segments; // Not used in object files.
+  std::vector<Section>        sections;
+
+  // Symbols sorted by kind.
+  std::vector<Symbol>         localSymbols;
+  std::vector<Symbol>         globalSymbols;
+  std::vector<Symbol>         undefinedSymbols;
+  std::vector<Symbol>         stabsSymbols;
+
+  // Maps to load commands with no LINKEDIT content (final linked images only).
+  std::vector<DependentDylib> dependentDylibs;
+  StringRef                   installName;        // dylibs only
+  PackedVersion               compatVersion = 0;  // dylibs only
+  PackedVersion               currentVersion = 0; // dylibs only
+  bool                        hasUUID = false;
+  bool                        hasMinVersionLoadCommand = false;
+  bool                        generateDataInCodeLoadCommand = false;
+  std::vector<StringRef>      rpaths;
+  Hex64                       entryAddress = 0;
+  Hex64                       stackSize = 0;
+  MachOLinkingContext::OS     os = MachOLinkingContext::OS::unknown;
+  Hex64                       sourceVersion = 0;
+  PackedVersion               minOSverson = 0;
+  PackedVersion               sdkVersion = 0;
+  LoadCommandType             minOSVersionKind = (LoadCommandType)0;
+
+  // Maps to load commands with LINKEDIT content (final linked images only).
+  Hex32                       pageSize = 0;
+  std::vector<RebaseLocation> rebasingInfo;
+  std::vector<BindLocation>   bindingInfo;
+  std::vector<BindLocation>   weakBindingInfo;
+  std::vector<BindLocation>   lazyBindingInfo;
+  std::vector<Export>         exportInfo;
+  std::vector<uint8_t>        functionStarts;
+  std::vector<DataInCode>     dataInCode;
+
+  // TODO:
+  // code-signature
+  // split-seg-info
+  // function-starts
+
+  // For any allocations in this struct which need to be owned by this struct.
+  BumpPtrAllocator            ownedAllocations;
+};
+
+/// Tests if a file is a non-fat mach-o object file.
+bool isThinObjectFile(StringRef path, MachOLinkingContext::Arch &arch);
+
+/// If the buffer is a fat file with the request arch, then this function
+/// returns true with 'offset' and 'size' set to location of the arch slice
+/// within the buffer.  Otherwise returns false;
+bool sliceFromFatFile(MemoryBufferRef mb, MachOLinkingContext::Arch arch,
+                      uint32_t &offset, uint32_t &size);
+
+/// Reads a mach-o file and produces an in-memory normalized view.
+llvm::Expected<std::unique_ptr<NormalizedFile>>
+readBinary(std::unique_ptr<MemoryBuffer> &mb,
+           const MachOLinkingContext::Arch arch);
+
+/// Takes in-memory normalized view and writes a mach-o object file.
+llvm::Error writeBinary(const NormalizedFile &file, StringRef path);
+
+size_t headerAndLoadCommandsSize(const NormalizedFile &file);
+
+
+/// Parses a yaml encoded mach-o file to produce an in-memory normalized view.
+llvm::Expected<std::unique_ptr<NormalizedFile>>
+readYaml(std::unique_ptr<MemoryBuffer> &mb);
+
+/// Writes a yaml encoded mach-o files given an in-memory normalized view.
+std::error_code writeYaml(const NormalizedFile &file, raw_ostream &out);
+
+llvm::Error
+normalizedObjectToAtoms(MachOFile *file,
+                        const NormalizedFile &normalizedFile,
+                        bool copyRefs);
+
+llvm::Error
+normalizedDylibToAtoms(MachODylibFile *file,
+                       const NormalizedFile &normalizedFile,
+                       bool copyRefs);
+
+/// Takes in-memory normalized dylib or object and parses it into lld::File
+llvm::Expected<std::unique_ptr<lld::File>>
+normalizedToAtoms(const NormalizedFile &normalizedFile, StringRef path,
+                  bool copyRefs);
+
+/// Takes atoms and generates a normalized macho-o view.
+llvm::Expected<std::unique_ptr<NormalizedFile>>
+normalizedFromAtoms(const lld::File &atomFile, const MachOLinkingContext &ctxt);
+
+
+} // namespace normalized
+
+/// Class for interfacing mach-o yaml files into generic yaml parsing
+class MachOYamlIOTaggedDocumentHandler : public YamlIOTaggedDocumentHandler {
+public:
+  MachOYamlIOTaggedDocumentHandler(MachOLinkingContext::Arch arch)
+    : _arch(arch) { }
+  bool handledDocTag(llvm::yaml::IO &io, const lld::File *&file) const override;
+private:
+  const MachOLinkingContext::Arch _arch;
+};
+
+} // namespace mach_o
+} // namespace lld
+
+#endif // LLD_READER_WRITER_MACHO_NORMALIZE_FILE_H