Avoid unnecessary buffer allocation and memcpy for compressed sections.

[ruiu]

Summary

Previously, we uncompress all compressed sections before doing anything.
That works, and that is conceptually simple, but that could results in
a waste of CPU time and memory if uncompressed sections are then
discarded or just copied to the output buffer.

In particular, if .debug_gnu_pub{names,types} are compressed and if no
-gdb-index option is given, we wasted CPU and memory because we
uncompress them into newly allocated bufers and then memcpy the buffers
to the output buffer. That temporary buffer was redundant.

This patch changes how to uncompress sections. Now, compressed sections
are uncompressed lazily. To do that, Data member of InputSectionBase
is now hidden from outside, and data() accessor automatically expands
an compressed buffer if necessary.

If no one calls data(), then writeTo() directly uncompresses
compressed data into the output buffer. That eliminates the redundant
memory allocation and redundant memcpy.

This patch significantly reduces memory consumption (20 GiB max RSS to
15 Gib) for an executable whose .debug_gnu_pub{names,types} are in total
5 GiB in an uncompressed form.

[dblaikie]

Out of curiosity - there's one relocation in each pubnames contribution. Does this solution (or future intended work) manage to avoid decompressing all the input compressed sections into memory, to then apply the relocation, and then write out the buffer (instead either streaming compressed bytes from input to output (either recompressing along the way, or not, as needed))?

Would be great to hear that this generalizes/handles relocations - to help with other compressed debug sections as well.

Though there are some debug sections which may have no relocations (debug_loclists and debug_rnglists maybe, debug_abbrev for sure) which would benefit even if this doesn't help sections with relocs.

[ruiu]

In lld, we do apply relocations after we copy input section contents to the output buffer, so this patch doesn't affect how we apply relocations. If a compressed section contains relocations, they are processed fine like before.

[dblaikie]

In D52917#1256109, @ruiu wrote:

In lld, we do apply relocations after we copy input section contents to the output buffer, so this patch doesn't affect how we apply relocations. If a compressed section contains relocations, they are processed fine like before.

Ah, awesome - should've guessed that might be the case. This sounds like it could be really helpful for all debug sections, then!

I see this uncompresses directly into the output, which is also great - no need to have the whole uncompressed contents in memory at any one time, then.

Out of curiosity - does LLD support compressed output? How/where does it do that? It'd be super nifty if it'd stream in the compressed data and stream it out through compression without having to have it all decompressed in memory. But that might hit problems with relocations that use the bytes in the section as part of the relocation computation :/

lld/ELF/InputSection.cpp

148

Did you intend to zero out this allocation? (the '()' after the [] will cause the array to be zero initialized, I think - which might be unnecessary if you're about to write to all the bytes anyway?)

[ruiu]

Out of curiosity - does LLD support compressed output? How/where does it do that? It'd be super nifty if it'd stream in the compressed data and stream it out through compression without having to have it all decompressed in memory. But that might hit problems with relocations that use the bytes in the section as part of the relocation computation :/

lld supports compressed output. When an output section compression is enabled, we allocate a temporary buffer and pass that buffer to writeTo() as if it were a real output buffer. Then, writeTo() copies input sections to a given buffer and apply relocations. We then compress it and write to a real output buffer.

As you guessed, I don't think we can stream-compress output section contents because relocation application is essentially a random access to the output buffer.

lld/ELF/InputSection.cpp

148

Ah thanks. We don't have to initialize the buffer.

[MaskRay]

lld/ELF/InputSection.cpp

148

How about omitting the initializer (new char[Size]) to save value initialization of each char (filling with zeroes)

[MaskRay]

lld/test/ELF/relocatable-compressed-input.s

10

rLLD324944 did s/uncompress/decompress/g. If you now feel that uncompress is better, you may make these changes in a separate commit.

[ruiu]

lld/test/ELF/relocatable-compressed-input.s

10

I understand the point but this is sort of related to this change because now I'm using zlib::uncompress instead of Decompressor. I think Decompressor should have been named Uncompressor in the first place, but I'm following the local convention (which is now "uncompress").

[ruiu]

• do not zero-initialize a temporary buffer
• inline data()

[MaskRay]

lld/ELF/InputSection.cpp

152

It seems uncompress() is only called once. What do you think if uncompress() is inlined here and these mutable member variables are marked as mutable?

[ruiu]

lld/ELF/InputSection.cpp

152

I could, but I'm not sure if it is a good thing to mark RawData and UncompressedBuf mutable throughout this class. At least in this way, only this function mutates these members.

[grimar]

Idea by itself LG. My suggestions about implementation details are below.

lld/ELF/InputSection.cpp

266

What do you think about still using the Decompressor here for parsing the
headers and checking zlib availability?
Like below.

Decompressor Dec = check(Decompressor::create(Name, toStringRef(RawData),
                                              Config->IsLE, Config->Is64));
UncompressedSize = Dec.getDecompressedSize();
if (Name.startswith(".zdebug")) {
  RawData = RawData.slice(12);
  Name = Saver.save("." + Name.substr(2));
  return;
}
assert(Flags & SHF_COMPRESSED);
Flags &= ~(uint64_t)SHF_COMPRESSED;

// New-style 64-bit header
if (Config->Is64)
  RawData = RawData.slice(sizeof(Elf64_Chdr));
else
  RawData = RawData.slice(sizeof(Elf32_Chdr));

or, if we add getCompressedData method to Decompressor class, it can be:

Decompressor Dec = check(Decompressor::create(Name, toStringRef(RawData),
                                              Config->IsLE, Config->Is64));
UncompressedSize = Dec.getDecompressedSize();
RawData = Dec.getCompressedData();

if (Name.startswith(".zdebug")) {
  Name = Saver.save("." + Name.substr(2));
  return;
}

assert(Flags & SHF_COMPRESSED);
Flags &= ~(uint64_t)SHF_COMPRESSED;

lld/test/ELF/strip-debug.s

3

If you remove --strip-debug, this will crash the linker because of
llvm_unreachable("zlib::uncompress is unavailable"); on windows, where zlib is not available.

You may want to check zlib::isAvailable() or use Decompressor.

[ruiu]

• use zlib::isAvailable

lld/ELF/InputSection.cpp

266

I found that Decompressor hides too many details and is not easy to use. I feel more comfortable to do this explicitly in this file.

lld/test/ELF/strip-debug.s

3

Done.

[dblaikie]

lld/ELF/InputSection.cpp

152

Personally I'd vote for using mutable - using const_cast places a "spooky action at a distance" to the use of this type (it means you can't make a const instance of this type if you use this function - because const_casting away const on an actually const object is UB). While that might not be an issue in this particular class (it might be quite unlikely anyone would make a const instance of this type) - I'd still lean that way out of stylistic consistency with other use cases, etc.

Not a huge deal, though.

[ruiu]

• make member variables mutable instead of using const_cast to call uncompress()

lld/ELF/InputSection.cpp

152

I don't have a strong opinion, so I changed the code to not use const_cast but mutable.

lld/ELF/AArch64ErrataFix.cpp

static uint64_t scanCortexA53Errata843419(InputSection *IS, uint64_t &Off,
   bool OptionalAllowed = Limit - Off > 12;
   if (Off >= Limit || Limit - Off < 12) {
     // Need at least 3 4-byte sized instructions to trigger erratum.
     Off = Limit;
     return 0;
   }
 
   uint64_t PatchOff = 0;
-  const uint8_t *Buf = IS->Data.begin();
+  const uint8_t *Buf = IS->data().begin();
   const ulittle32_t *InstBuf = reinterpret_cast<const ulittle32_t *>(Buf + Off);
   uint32_t Instr1 = *InstBuf++;
   uint32_t Instr2 = *InstBuf++;
   uint32_t Instr3 = *InstBuf++;
   if (is843419ErratumSequence(Instr1, Instr2, Instr3)) {
     PatchOff = Off + 8;
   } else if (OptionalAllowed && !isBranch(Instr3)) {
     uint32_t Instr4 = *InstBuf++;

 
 uint64_t lld::elf::Patch843419Section::getLDSTAddr() const {
   return Patchee->getVA(PatcheeOffset);
 }
 
 void lld::elf::Patch843419Section::writeTo(uint8_t *Buf) {
   // Copy the instruction that we will be replacing with a branch in the
   // Patchee Section.
-  write32le(Buf, read32le(Patchee->Data.begin() + PatcheeOffset));
+  write32le(Buf, read32le(Patchee->data().begin() + PatcheeOffset));
 
   // Apply any relocation transferred from the original PatcheeSection.
   // For a SyntheticSection Buf already has OutSecOff added, but relocateAlloc
   // also adds OutSecOff so we need to subtract to avoid double counting.
   this->relocateAlloc(Buf - OutSecOff, Buf - OutSecOff + getSize());
 
   // Return address is the next instruction after the one we have just copied.
   uint64_t S = getLDSTAddr() + 4;
for (InputSection *IS : ISD.Sections) {
     auto CodeSym = llvm::find_if(MapSyms, [&](const Defined *MS) {
       return MS->getName().startswith("$x");
     });
 
     while (CodeSym != MapSyms.end()) {
       auto DataSym = std::next(CodeSym);
       uint64_t Off = (*CodeSym)->Value;
       uint64_t Limit =
-          (DataSym == MapSyms.end()) ? IS->Data.size() : (*DataSym)->Value;
+          (DataSym == MapSyms.end()) ? IS->data().size() : (*DataSym)->Value;
 
       while (Off < Limit) {
         uint64_t StartAddr = IS->getVA(Off);
         if (uint64_t PatcheeOffset = scanCortexA53Errata843419(IS, Off, Limit))
           implementPatch(StartAddr, PatcheeOffset, IS, Patches);
       }
       if (DataSym == MapSyms.end())
         break;

lld/ELF/DWARF.cpp

 using namespace llvm::object;
 using namespace lld;
 using namespace lld::elf;
 
 template <class ELFT> LLDDwarfObj<ELFT>::LLDDwarfObj(ObjFile<ELFT> *Obj) {
   for (InputSectionBase *Sec : Obj->getSections()) {
     if (!Sec)
       continue;
+
     if (LLDDWARFSection *M = StringSwitch<LLDDWARFSection *>(Sec->Name)
                                  .Case(".debug_info", &InfoSection)
                                  .Case(".debug_ranges", &RangeSection)
                                  .Case(".debug_line", &LineSection)
                                  .Default(nullptr)) {
-      Sec->maybeDecompress();
-      M->Data = toStringRef(Sec->Data);
+      M->Data = toStringRef(Sec->data());
       M->Sec = Sec;
       continue;
     }
+
     if (Sec->Name == ".debug_abbrev")
-      AbbrevSection = toStringRef(Sec->Data);
+      AbbrevSection = toStringRef(Sec->data());
     else if (Sec->Name == ".debug_gnu_pubnames")
-      GnuPubNamesSection = toStringRef(Sec->Data);
+      GnuPubNamesSection = toStringRef(Sec->data());
     else if (Sec->Name == ".debug_gnu_pubtypes")
-      GnuPubTypesSection = toStringRef(Sec->Data);
+      GnuPubTypesSection = toStringRef(Sec->data());
     else if (Sec->Name == ".debug_str")
-      StrSection = toStringRef(Sec->Data);
-    if (Sec->Name == ".debug_line_str")
-      LineStringSection = toStringRef(Sec->Data);
+      StrSection = toStringRef(Sec->data());
+    else if (Sec->Name == ".debug_line_str")
+      LineStringSection = toStringRef(Sec->data());
   }
 }
 
 // Find if there is a relocation at Pos in Sec.  The code is a bit
 // more complicated than usual because we need to pass a section index
 // to llvm since it has no idea about InputSection.
 template <class ELFT>
 template <class RelTy>

lld/ELF/Driver.cpp

if (Config->EMachine == EM_ARM) {
     // when the input objects have been compiled with an architecture that
     // supports them.
     if (Config->ARMHasBlx == false)
       warn("lld uses blx instruction, no object with architecture supporting "
            "feature detected.");
   }
 
   // This adds a .comment section containing a version string. We have to add it
-  // before decompressAndMergeSections because the .comment section is a
+  // before uncompressAndMergeSections because the .comment section is a
   // mergeable section.
   if (!Config->Relocatable)
     InputSections.push_back(createCommentSection());
 
   // Do size optimizations: garbage collection, merging of SHF_MERGE sections
   // and identical code folding.
-  decompressSections();
   splitSections<ELFT>();
   markLive<ELFT>();
   demoteSharedSymbols<ELFT>();
   mergeSections();
   if (Config->ICF != ICFLevel::None) {
     findKeepUniqueSections<ELFT>(Args);
     doIcf<ELFT>();
   }

lld/ELF/EhFrame.cpp

 public:
   EhReader(InputSectionBase *S, ArrayRef<uint8_t> D) : IS(S), D(D) {}
   size_t readEhRecordSize();
   uint8_t getFdeEncoding();
 
 private:
   template <class P> void failOn(const P *Loc, const Twine &Msg) {
     fatal("corrupted .eh_frame: " + Msg + "\n>>> defined in " +
-          IS->getObjMsg((const uint8_t *)Loc - IS->Data.data()));
+          IS->getObjMsg((const uint8_t *)Loc - IS->data().data()));
   }
 
   uint8_t readByte();
   void skipBytes(size_t Count);
   StringRef readString();
   void skipLeb128();
   void skipAugP();
 
   InputSectionBase *IS;
   ArrayRef<uint8_t> D;
 };
 }
 
 size_t elf::readEhRecordSize(InputSectionBase *S, size_t Off) {
-  return EhReader(S, S->Data.slice(Off)).readEhRecordSize();
+  return EhReader(S, S->data().slice(Off)).readEhRecordSize();
 }
 
 // .eh_frame section is a sequence of records. Each record starts with
 // a 4 byte length field. This function reads the length.
 size_t EhReader::readEhRecordSize() {
   if (D.size() < 4)
     failOn(D.data(), "CIE/FDE too small");
 

lld/ELF/ICF.cpp

bool ICF<ELFT>::constantEq(const InputSection *SecA, ArrayRef<RelTy> RA,
   return true;
 }
 
 // Compare "non-moving" part of two InputSections, namely everything
 // except relocation targets.
 template <class ELFT>
 bool ICF<ELFT>::equalsConstant(const InputSection *A, const InputSection *B) {
   if (A->NumRelocations != B->NumRelocations || A->Flags != B->Flags ||
-      A->getSize() != B->getSize() || A->Data != B->Data)
+      A->getSize() != B->getSize() || A->data() != B->data())
     return false;
 
   // If two sections have different output sections, we cannot merge them.
   // FIXME: This doesn't do the right thing in the case where there is a linker
   // script. We probably need to move output section assignment before ICF to
   // get the correct behaviour here.
   if (getOutputSectionName(A) != getOutputSectionName(B))
     return false;
template <class ELFT> void ICF<ELFT>::run() {
   for (InputSectionBase *Sec : InputSections)
     if (auto *S = dyn_cast<InputSection>(Sec))
       if (isEligible(S))
         Sections.push_back(S);
 
   // Initially, we use hash values to partition sections.
   parallelForEach(Sections, [&](InputSection *S) {
     // Set MSB to 1 to avoid collisions with non-hash IDs.
-    S->Class[0] = xxHash64(S->Data) | (1U << 31);
+    S->Class[0] = xxHash64(S->data()) | (1U << 31);
   });
 
   // From now on, sections in Sections vector are ordered so that sections
   // in the same equivalence class are consecutive in the vector.
   std::stable_sort(Sections.begin(), Sections.end(),
                    [](InputSection *A, InputSection *B) {
                      return A->Class[0] < B->Class[0];
                    });

lld/ELF/InputFiles.cpp

if (!Target)
     fatal(toString(this) + ": unsupported relocation reference");
   return Target;
 }
 
 // Create a regular InputSection class that has the same contents
 // as a given section.
 static InputSection *toRegularSection(MergeInputSection *Sec) {
   return make<InputSection>(Sec->File, Sec->Flags, Sec->Type, Sec->Alignment,
-                            Sec->Data, Sec->Name);
+                            Sec->data(), Sec->Name);
 }
 
 template <class ELFT>
 InputSectionBase *ObjFile<ELFT>::createInputSection(const Elf_Shdr &Sec) {
   StringRef Name = getSectionName(Sec);
 
   switch (Sec.sh_type) {
   case SHT_ARM_ATTRIBUTES: {

lld/ELF/InputSection.h

public:
   // ObjFile<ELFT>, but in order to avoid ELFT, we use InputFile as
   // its static type.
   InputFile *File;
 
   template <class ELFT> ObjFile<ELFT> *getFile() const {
     return cast_or_null<ObjFile<ELFT>>(File);
   }
 
-  ArrayRef<uint8_t> Data;
+  ArrayRef<uint8_t> data() const {
+    if (UncompressedSize >= 0 && !UncompressedBuf)
+      uncompress();
+    return RawData;
+  }
+
   uint64_t getOffsetInFile() const;
 
   // True if this section has already been placed to a linker script
   // output section. This is needed because, in a linker script, you
   // can refer to the same section more than once. For example, in
   // the following linker script,
   //
   //   .foo : { *(.text) }
public:
 
   InputSection *getLinkOrderDep() const;
 
   // Get the function symbol that encloses this offset from within the
   // section.
   template <class ELFT>
   Defined *getEnclosingFunction(uint64_t Offset);
 
-  // Compilers emit zlib-compressed debug sections if the -gz option
-  // is given. This function checks if this section is compressed, and
-  // if so, decompress in memory.
-  void maybeDecompress();
-
   // Returns a source location string. Used to construct an error message.
   template <class ELFT> std::string getLocation(uint64_t Offset);
   std::string getSrcMsg(const Symbol &Sym, uint64_t Offset);
   std::string getObjMsg(uint64_t Offset);
 
   // Each section knows how to relocate itself. These functions apply
   // relocations, assuming that Buf points to this section's copy in
   // the mmap'ed output buffer.
public:
   // such functions and adjust their prologues.  This is very similar
   // to relocation. See https://gcc.gnu.org/wiki/SplitStacks for more
   // information.
   template <typename ELFT>
   void adjustSplitStackFunctionPrologues(uint8_t *Buf, uint8_t *End);
 
 
   template <typename T> llvm::ArrayRef<T> getDataAs() const {
-    size_t S = Data.size();
+    size_t S = data().size();
     assert(S % sizeof(T) == 0);
-    return llvm::makeArrayRef<T>((const T *)Data.data(), S / sizeof(T));
+    return llvm::makeArrayRef<T>((const T *)data().data(), S / sizeof(T));
   }
 
-private:
-  // A pointer that owns decompressed data if a section is compressed by zlib.
+protected:
+  void parseCompressedHeader();
+  void uncompress() const;
+
+  mutable ArrayRef<uint8_t> RawData;
+
+  // A pointer that owns uncompressed data if a section is compressed by zlib.
   // Since the feature is not used often, this is usually a nullptr.
-  std::unique_ptr<char[]> DecompressBuf;
+  mutable std::unique_ptr<char[]> UncompressedBuf;
+  int64_t UncompressedSize = -1;
 };
 
 // SectionPiece represents a piece of splittable section contents.
 // We allocate a lot of these and binary search on them. This means that they
 // have to be as compact as possible, which is why we don't store the size (can
 // be found by looking at the next one).
 struct SectionPiece {
   SectionPiece(size_t Off, uint32_t Hash, bool Live)
public:
   llvm::DenseMap<uint32_t, uint32_t> OffsetMap;
 
   // Returns I'th piece's data. This function is very hot when
   // string merging is enabled, so we want to inline.
   LLVM_ATTRIBUTE_ALWAYS_INLINE
   llvm::CachedHashStringRef getData(size_t I) const {
     size_t Begin = Pieces[I].InputOff;
     size_t End =
-        (Pieces.size() - 1 == I) ? Data.size() : Pieces[I + 1].InputOff;
-    return {toStringRef(Data.slice(Begin, End - Begin)), Pieces[I].Hash};
+        (Pieces.size() - 1 == I) ? data().size() : Pieces[I + 1].InputOff;
+    return {toStringRef(data().slice(Begin, End - Begin)), Pieces[I].Hash};
   }
 
   // Returns the SectionPiece at a given input section offset.
   SectionPiece *getSectionPiece(uint64_t Offset);
   const SectionPiece *getSectionPiece(uint64_t Offset) const {
     return const_cast<MergeInputSection *>(this)->getSectionPiece(Offset);
   }
 
   SyntheticSection *getParent() const;
 
 private:
   void splitStrings(ArrayRef<uint8_t> A, size_t Size);
   void splitNonStrings(ArrayRef<uint8_t> A, size_t Size);
 };
 
 struct EhSectionPiece {
   EhSectionPiece(size_t Off, InputSectionBase *Sec, uint32_t Size,
                  unsigned FirstRelocation)
       : InputOff(Off), Sec(Sec), Size(Size), FirstRelocation(FirstRelocation) {}
 
-  ArrayRef<uint8_t> data() { return {Sec->Data.data() + this->InputOff, Size}; }
+  ArrayRef<uint8_t> data() {
+    return {Sec->data().data() + this->InputOff, Size};
+  }
 
   size_t InputOff;
   ssize_t OutputOff = -1;
   InputSectionBase *Sec;
   uint32_t Size;
   unsigned FirstRelocation;
 };
 

lld/ELF/InputSection.cpp

 #include "Relocations.h"
 #include "SymbolTable.h"
 #include "Symbols.h"
 #include "SyntheticSections.h"
 #include "Target.h"
 #include "Thunks.h"
 #include "lld/Common/ErrorHandler.h"
 #include "lld/Common/Memory.h"
-#include "llvm/Object/Decompressor.h"
 #include "llvm/Support/Compiler.h"
 #include "llvm/Support/Compression.h"
 #include "llvm/Support/Endian.h"
 #include "llvm/Support/Threading.h"
 #include "llvm/Support/xxhash.h"
 #include <algorithm>
 #include <mutex>
 #include <set>

 
 InputSectionBase::InputSectionBase(InputFile *File, uint64_t Flags,
                                    uint32_t Type, uint64_t Entsize,
                                    uint32_t Link, uint32_t Info,
                                    uint32_t Alignment, ArrayRef<uint8_t> Data,
                                    StringRef Name, Kind SectionKind)
     : SectionBase(SectionKind, Name, Flags, Entsize, Alignment, Type, Info,
                   Link),
-      File(File), Data(Data) {
+      File(File), RawData(Data) {
   // In order to reduce memory allocation, we assume that mergeable
   // sections are smaller than 4 GiB, which is not an unreasonable
   // assumption as of 2017.
-  if (SectionKind == SectionBase::Merge && Data.size() > UINT32_MAX)
+  if (SectionKind == SectionBase::Merge && RawData.size() > UINT32_MAX)
     error(toString(this) + ": section too large");
 
   NumRelocations = 0;
   AreRelocsRela = false;
 
   // The ELF spec states that a value of 0 means the section has
   // no alignment constraits.
   uint32_t V = std::max<uint64_t>(Alignment, 1);
   if (!isPowerOf2_64(V))
     fatal(toString(File) + ": section sh_addralign is not a power of 2");
   this->Alignment = V;
+
+  // In ELF, each section can be compressed by zlib, and if compressed,
+  // section name may be mangled by appending "z" (e.g. ".zdebug_info").
+  // If that's the case, demangle section name so that we can handle a
+  // section as if it weren't compressed.
+  if ((Flags & SHF_COMPRESSED) || Name.startswith(".zdebug")) {
+    if (!zlib::isAvailable())
+      error(toString(File) + ": contains a compressed section, "
+            + "but zlib is not available");
+    parseCompressedHeader();
+  }
 }
 
 // Drop SHF_GROUP bit unless we are producing a re-linkable object file.
 // SHF_GROUP is a marker that a section belongs to some comdat group.
 // That flag doesn't make sense in an executable.
 static uint64_t getFlags(uint64_t Flags) {
   Flags &= ~(uint64_t)SHF_INFO_LINK;
   if (!Config->Relocatable)
InputSectionBase::InputSectionBase(ObjFile<ELFT> &File,
   // We might want to relax this in the future.
   if (Hdr.sh_addralign > UINT32_MAX)
     fatal(toString(&File) + ": section sh_addralign is too large");
 }
 
 size_t InputSectionBase::getSize() const {
   if (auto *S = dyn_cast<SyntheticSection>(this))
     return S->getSize();
+  if (UncompressedSize >= 0)
+    return UncompressedSize;
+  return RawData.size();
+}
 
-  return Data.size();
+void InputSectionBase::uncompress() const {
+  size_t Size = UncompressedSize;
+  UncompressedBuf.reset(new char[Size]);
+
+  if (Error E =
+          zlib::uncompress(toStringRef(RawData), UncompressedBuf.get(), Size))
+    fatal(toString(this) +
+          ": uncompress failed: " + llvm::toString(std::move(E)));
+  RawData = makeArrayRef((uint8_t *)UncompressedBuf.get(), Size);
 }
 
 uint64_t InputSectionBase::getOffsetInFile() const {
   const uint8_t *FileStart = (const uint8_t *)File->MB.getBufferStart();
-  const uint8_t *SecStart = Data.begin();
+  const uint8_t *SecStart = data().begin();
   return SecStart - FileStart;
 }
 
 uint64_t SectionBase::getOffset(uint64_t Offset) const {
   switch (kind()) {
   case Output: {
     auto *OS = cast<OutputSection>(this);
     // For output sections we treat offset -1 as the end of the section.
else if (auto *MS = dyn_cast<MergeInputSection>(this))
     Sec = MS->getParent();
   else if (auto *EH = dyn_cast<EhInputSection>(this))
     Sec = EH->getParent();
   else
     return cast<OutputSection>(this);
   return Sec ? Sec->getParent() : nullptr;
 }
 
-// Decompress section contents if required. Note that this function
-// is called from parallelForEach, so it must be thread-safe.
-void InputSectionBase::maybeDecompress() {
-  if (DecompressBuf)
+// When a section is compressed, `RawData` consists with a header followed
+// by zlib-compressed data. This function parses a header to initialize
+// `UncompressedSize` member and remove the header from `RawData`.
+void InputSectionBase::parseCompressedHeader() {
+  // Old-style header
+  if (Name.startswith(".zdebug")) {
+    if (!toStringRef(RawData).startswith("ZLIB")) {
+      error(toString(this) + ": corrupted compressed section header");
       return;
-  if (!(Flags & SHF_COMPRESSED) && !Name.startswith(".zdebug"))
+    }
+    RawData = RawData.slice(4);
+
+    if (RawData.size() < 8) {
+      error(toString(this) + ": corrupted compressed section header");
       return;
+    }
 
-  // Decompress a section.
-  Decompressor Dec = check(Decompressor::create(Name, toStringRef(Data),
-                                                Config->IsLE, Config->Is64));
-
-  size_t Size = Dec.getDecompressedSize();
-  DecompressBuf.reset(new char[Size + Name.size()]());
-  if (Error E = Dec.decompress({DecompressBuf.get(), Size}))
-    fatal(toString(this) +
-          ": decompress failed: " + llvm::toString(std::move(E)));
+    UncompressedSize = read64be(RawData.data());
+    RawData = RawData.slice(8);
+
+    // Restore the original section name.
+    // (e.g. ".zdebug_info" -> ".debug_info")
+    Name = Saver.save("." + Name.substr(2));
+    return;
+  }
 
-  Data = makeArrayRef((uint8_t *)DecompressBuf.get(), Size);
+  assert(Flags & SHF_COMPRESSED);
   Flags &= ~(uint64_t)SHF_COMPRESSED;
 
-  // A section name may have been altered if compressed. If that's
-  // the case, restore the original name. (i.e. ".zdebug_" -> ".debug_")
-  if (Name.startswith(".zdebug")) {
-    DecompressBuf[Size] = '.';
-    memcpy(&DecompressBuf[Size + 1], Name.data() + 2, Name.size() - 2);
-    Name = StringRef(&DecompressBuf[Size], Name.size() - 1);
+  // New-style 64-bit header
+  if (Config->Is64) {
+    if (RawData.size() < sizeof(Elf64_Chdr)) {
+      error(toString(this) + ": corrupted compressed section");
+      return;
     }
+
+    auto *Hdr = reinterpret_cast<const Elf64_Chdr *>(RawData.data());
+    if (read32(&Hdr->ch_type) != ELFCOMPRESS_ZLIB) {
+      error(toString(this) + ": unsupported compression type");
+      return;
+    }
+
+    UncompressedSize = read64(&Hdr->ch_size);
+    RawData = RawData.slice(sizeof(*Hdr));
+    return;
+  }
+
+  // New-style 32-bit header
+  if (RawData.size() < sizeof(Elf32_Chdr)) {
+    error(toString(this) + ": corrupted compressed section");
+    return;
+  }
+
+  auto *Hdr = reinterpret_cast<const Elf32_Chdr *>(RawData.data());
+  if (read32(&Hdr->ch_type) != ELFCOMPRESS_ZLIB) {
+    error(toString(this) + ": unsupported compression type");
+    return;
+  }
+
+  UncompressedSize = read32(&Hdr->ch_size);
+  RawData = RawData.slice(sizeof(*Hdr));
 }
 
 InputSection *InputSectionBase::getLinkOrderDep() const {
   assert(Link);
   assert(Flags & SHF_LINK_ORDER);
   return cast<InputSection>(File->getSections()[Link]);
 }
 
 // Find a function symbol that encloses a given location.
if (Sym.Type == STT_SECTION) {
       }
       SectionBase *Section = D->Section;
       if (Section == &InputSection::Discarded) {
         P->setSymbolAndType(0, 0, false);
         continue;
       }
 
       int64_t Addend = getAddend<ELFT>(Rel);
-      const uint8_t *BufLoc = Sec->Data.begin() + Rel.r_offset;
+      const uint8_t *BufLoc = Sec->data().begin() + Rel.r_offset;
       if (!RelTy::IsRela)
         Addend = Target->getImplicitAddend(BufLoc, Type);
 
       if (Config->EMachine == EM_MIPS && Config->Relocatable &&
           Target->getRelExpr(Type, Sym, BufLoc) == R_MIPS_GOTREL) {
         // Some MIPS relocations depend on "gp" value. By default,
         // this value has 0x7ff0 offset from a .got section. But
         // relocatable files produced by a complier or a linker
template <class ELFT> void InputSection::writeTo(uint8_t *Buf) {
   }
 
   // If -r is given, we may have a SHT_GROUP section.
   if (Type == SHT_GROUP) {
     copyShtGroup<ELFT>(Buf + OutSecOff);
     return;
   }
 
+  // If this is a compressed section, uncompress section contents directly
+  // to the buffer.
+  if (UncompressedSize >= 0 && !UncompressedBuf) {
+    size_t Size = UncompressedSize;
+    if (Error E = zlib::uncompress(toStringRef(RawData),
+                                   (char *)(Buf + OutSecOff), Size))
+      fatal(toString(this) +
+            ": uncompress failed: " + llvm::toString(std::move(E)));
+    uint8_t *BufEnd = Buf + OutSecOff + Size;
+    relocate<ELFT>(Buf, BufEnd);
+    return;
+  }
+
   // Copy section contents from source object file to output file
   // and then apply relocations.
-  memcpy(Buf + OutSecOff, Data.data(), Data.size());
-  uint8_t *BufEnd = Buf + OutSecOff + Data.size();
+  memcpy(Buf + OutSecOff, data().data(), data().size());
+  uint8_t *BufEnd = Buf + OutSecOff + data().size();
   relocate<ELFT>(Buf, BufEnd);
 }
 
 void InputSection::replace(InputSection *Other) {
   Alignment = std::max(Alignment, Other->Alignment);
   Other->Repl = Repl;
   Other->Live = false;
 }
if (AreRelocsRela)
     split<ELFT>(relas<ELFT>());
   else
     split<ELFT>(rels<ELFT>());
 }
 
 template <class ELFT, class RelTy>
 void EhInputSection::split(ArrayRef<RelTy> Rels) {
   unsigned RelI = 0;
-  for (size_t Off = 0, End = Data.size(); Off != End;) {
+  for (size_t Off = 0, End = data().size(); Off != End;) {
     size_t Size = readEhRecordSize(this, Off);
     Pieces.emplace_back(Off, this, Size, getReloc(Off, Size, Rels, RelI));
     // The empty record is the end marker.
     if (Size == 4)
       break;
     Off += Size;
   }
 }

 // into small chunks for further processing.
 //
 // Note that this function is called from parallelForEach. This must be
 // thread-safe (i.e. no memory allocation from the pools).
 void MergeInputSection::splitIntoPieces() {
   assert(Pieces.empty());
 
   if (Flags & SHF_STRINGS)
-    splitStrings(Data, Entsize);
+    splitStrings(data(), Entsize);
   else
-    splitNonStrings(Data, Entsize);
+    splitNonStrings(data(), Entsize);
 
   OffsetMap.reserve(Pieces.size());
   for (size_t I = 0, E = Pieces.size(); I != E; ++I)
     OffsetMap[Pieces[I].InputOff] = I;
 }
 
 template <class It, class T, class Compare>
 static It fastUpperBound(It First, It Last, const T &Value, Compare Comp) {
   size_t Size = std::distance(First, Last);
   assert(Size != 0);
   while (Size != 1) {
     size_t H = Size / 2;
     const It MI = First + H;
     Size -= H;
     First = Comp(Value, *MI) ? First : First + H;
   }
   return Comp(Value, *First) ? First : First + 1;
 }
 
 SectionPiece *MergeInputSection::getSectionPiece(uint64_t Offset) {
-  if (this->Data.size() <= Offset)
+  if (this->data().size() <= Offset)
     fatal(toString(this) + ": offset is outside the section");
 
   // Find a piece starting at a given offset.
   auto It = OffsetMap.find(Offset);
   if (It != OffsetMap.end())
     return &Pieces[It->second];
 
   // If Offset is not at beginning of a section piece, it is not in the map.

lld/ELF/MarkLive.cpp

 using namespace llvm::support::endian;
 
 using namespace lld;
 using namespace lld::elf;
 
 template <class ELFT>
 static typename ELFT::uint getAddend(InputSectionBase &Sec,
                                      const typename ELFT::Rel &Rel) {
-  return Target->getImplicitAddend(Sec.Data.begin() + Rel.r_offset,
+  return Target->getImplicitAddend(Sec.data().begin() + Rel.r_offset,
                                    Rel.getType(Config->IsMips64EL));
 }
 
 template <class ELFT>
 static typename ELFT::uint getAddend(InputSectionBase &Sec,
                                      const typename ELFT::Rela &Rel) {
   return Rel.r_addend;
 }

lld/ELF/Relocations.cpp

static int64_t computeMipsAddend(const RelTy &Rel, const RelTy *End,
   if (RelTy::IsRela)
     return 0;
 
   RelType Type = Rel.getType(Config->IsMips64EL);
   uint32_t PairTy = getMipsPairType(Type, IsLocal);
   if (PairTy == R_MIPS_NONE)
     return 0;
 
-  const uint8_t *Buf = Sec.Data.data();
+  const uint8_t *Buf = Sec.data().data();
   uint32_t SymIndex = Rel.getSymbol(Config->IsMips64EL);
 
   // To make things worse, paired relocations might not be contiguous in
   // the relocation table, so we need to do linear search. *sigh*
   for (const RelTy *RI = &Rel; RI != End; ++RI)
     if (RI->getType(Config->IsMips64EL) == PairTy &&
         RI->getSymbol(Config->IsMips64EL) == SymIndex)
       return Target->getImplicitAddend(Buf + RI->r_offset, PairTy);
static int64_t computeAddend(const RelTy &Rel, const RelTy *End,
                              InputSectionBase &Sec, RelExpr Expr,
                              bool IsLocal) {
   int64_t Addend;
   RelType Type = Rel.getType(Config->IsMips64EL);
 
   if (RelTy::IsRela) {
     Addend = getAddend<ELFT>(Rel);
   } else {
-    const uint8_t *Buf = Sec.Data.data();
+    const uint8_t *Buf = Sec.data().data();
     Addend = Target->getImplicitAddend(Buf + Rel.r_offset, Type);
   }
 
   if (Config->EMachine == EM_PPC64 && Config->Pic && Type == R_PPC64_TOC)
     Addend += getPPC64TocBase();
   if (Config->EMachine == EM_MIPS)
     Addend += computeMipsAddend<ELFT>(Rel, End, Sec, Expr, IsLocal);
 
static void scanReloc(InputSectionBase &Sec, OffsetGetter &GetOffset, RelTy *&I,
   uint64_t Offset = GetOffset.get(Rel.r_offset);
   if (Offset == uint64_t(-1))
     return;
 
   // Skip if the target symbol is an erroneous undefined symbol.
   if (maybeReportUndefined(Sym, Sec, Rel.r_offset))
     return;
 
-  const uint8_t *RelocatedAddr = Sec.Data.begin() + Rel.r_offset;
+  const uint8_t *RelocatedAddr = Sec.data().begin() + Rel.r_offset;
   RelExpr Expr = Target->getRelExpr(Type, Sym, RelocatedAddr);
 
   // Ignore "hint" relocations because they are only markers for relaxation.
   if (isRelExprOneOf<R_HINT, R_NONE>(Expr))
     return;
 
   // Strenghten or relax relocations.
   //

lld/ELF/SyntheticSections.h

 
 private:
   std::vector<Thunk *> Thunks;
   size_t Size = 0;
 };
 
 InputSection *createInterpSection();
 MergeInputSection *createCommentSection();
-void decompressSections();
 template <class ELFT> void splitSections();
 void mergeSections();
 
 Defined *addSyntheticLocal(StringRef Name, uint8_t Type, uint64_t Value,
                            uint64_t Size, InputSectionBase &Section);
 
 // Linker generated sections which can be used as inputs.
 struct InStruct {

lld/ELF/SyntheticSections.cpp

 #include "lld/Common/Memory.h"
 #include "lld/Common/Strings.h"
 #include "lld/Common/Threads.h"
 #include "lld/Common/Version.h"
 #include "llvm/ADT/SetOperations.h"
 #include "llvm/ADT/StringExtras.h"
 #include "llvm/BinaryFormat/Dwarf.h"
 #include "llvm/DebugInfo/DWARF/DWARFDebugPubTable.h"
-#include "llvm/Object/Decompressor.h"
 #include "llvm/Object/ELFObjectFile.h"
+#include "llvm/Support/Compression.h"
 #include "llvm/Support/Endian.h"
 #include "llvm/Support/LEB128.h"
 #include "llvm/Support/MD5.h"
 #include "llvm/Support/RandomNumberGenerator.h"
 #include "llvm/Support/SHA1.h"
 #include "llvm/Support/xxhash.h"
 #include <cstdlib>
 #include <thread>
MipsAbiFlagsSection<ELFT> *MipsAbiFlagsSection<ELFT>::create() {
 
   for (InputSectionBase *Sec : InputSections) {
     if (Sec->Type != SHT_MIPS_ABIFLAGS)
       continue;
     Sec->Live = false;
     Create = true;
 
     std::string Filename = toString(Sec->File);
-    const size_t Size = Sec->Data.size();
+    const size_t Size = Sec->data().size();
     // Older version of BFD (such as the default FreeBSD linker) concatenate
     // .MIPS.abiflags instead of merging. To allow for this case (or potential
     // zero padding) we ignore everything after the first Elf_Mips_ABIFlags
     if (Size < sizeof(Elf_Mips_ABIFlags)) {
       error(Filename + ": invalid size of .MIPS.abiflags section: got " +
             Twine(Size) + " instead of " + Twine(sizeof(Elf_Mips_ABIFlags)));
       return nullptr;
     }
-    auto *S = reinterpret_cast<const Elf_Mips_ABIFlags *>(Sec->Data.data());
+    auto *S = reinterpret_cast<const Elf_Mips_ABIFlags *>(Sec->data().data());
     if (S->version != 0) {
       error(Filename + ": unexpected .MIPS.abiflags version " +
             Twine(S->version));
       return nullptr;
     }
 
     // LLD checks ISA compatibility in calcMipsEFlags(). Here we just
     // select the highest number of ISA/Rev/Ext.
MipsOptionsSection<ELFT> *MipsOptionsSection<ELFT>::create() {
   if (Sections.empty())
     return nullptr;
 
   Elf_Mips_RegInfo Reginfo = {};
   for (InputSectionBase *Sec : Sections) {
     Sec->Live = false;
 
     std::string Filename = toString(Sec->File);
-    ArrayRef<uint8_t> D = Sec->Data;
+    ArrayRef<uint8_t> D = Sec->data();
 
     while (!D.empty()) {
       if (D.size() < sizeof(Elf_Mips_Options)) {
         error(Filename + ": invalid size of .MIPS.options section");
         break;
       }
 
       auto *Opt = reinterpret_cast<const Elf_Mips_Options *>(D.data());
MipsReginfoSection<ELFT> *MipsReginfoSection<ELFT>::create() {
 
   if (Sections.empty())
     return nullptr;
 
   Elf_Mips_RegInfo Reginfo = {};
   for (InputSectionBase *Sec : Sections) {
     Sec->Live = false;
 
-    if (Sec->Data.size() != sizeof(Elf_Mips_RegInfo)) {
+    if (Sec->data().size() != sizeof(Elf_Mips_RegInfo)) {
       error(toString(Sec->File) + ": invalid size of .reginfo section");
       return nullptr;
     }
 
-    auto *R = reinterpret_cast<const Elf_Mips_RegInfo *>(Sec->Data.data());
+    auto *R = reinterpret_cast<const Elf_Mips_RegInfo *>(Sec->data().data());
     Reginfo.ri_gprmask |= R->ri_gprmask;
     Sec->getFile<ELFT>()->MipsGp0 = R->ri_gp_value;
   };
 
   return make<MipsReginfoSection<ELFT>>(Reginfo);
 }
 
 InputSection *elf::createInterpSection() {
static MergeSyntheticSection *createMergeSynthetic(StringRef Name,
                                                    uint64_t Flags,
                                                    uint32_t Alignment) {
   bool ShouldTailMerge = (Flags & SHF_STRINGS) && Config->Optimize >= 2;
   if (ShouldTailMerge)
     return make<MergeTailSection>(Name, Type, Flags, Alignment);
   return make<MergeNoTailSection>(Name, Type, Flags, Alignment);
 }
 
-// Debug sections may be compressed by zlib. Decompress if exists.
-void elf::decompressSections() {
-  parallelForEach(InputSections,
-                  [](InputSectionBase *Sec) { Sec->maybeDecompress(); });
-}
-
 template <class ELFT> void elf::splitSections() {
   // splitIntoPieces needs to be called on each MergeInputSection
   // before calling finalizeContents().
   parallelForEach(InputSections, [](InputSectionBase *Sec) {
     if (auto *S = dyn_cast<MergeInputSection>(Sec))
       S->splitIntoPieces();
     else if (auto *Eh = dyn_cast<EhInputSection>(Sec))
       Eh->split<ELFT>();

lld/test/ELF/compressed-debug-input-err.s

 # REQUIRES: zlib, x86
 
 # RUN: llvm-mc -filetype=obj -triple=x86_64-unknown-linux %s -o %t.o
 # RUN: not ld.lld %t.o -o %t.so -shared 2>&1 | FileCheck %s
 
-## Check we are able to report zlib decompressor errors.
-# CHECK: error: {{.*}}.o:(.zdebug_str): decompress failed: zlib error: Z_DATA_ERROR
+## Check we are able to report zlib uncompress errors.
+# CHECK: error: {{.*}}.o:(.debug_str): uncompress failed: zlib error: Z_DATA_ERROR
 
 .section .zdebug_str,"MS",@progbits,1
  .ascii "ZLIB"
  .byte 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x1

lld/test/ELF/relocatable-compressed-input.s

 # REQUIRES: x86, zlib
 
 # RUN: llvm-mc -compress-debug-sections=zlib-gnu -filetype=obj -triple=x86_64-unknown-linux %s -o %t1
 # RUN: llvm-readobj -sections %t1 | FileCheck -check-prefix=GNU %s
 # GNU: Name: .zdebug_str
 
 # RUN: ld.lld %t1 -o %t2 -r
 # RUN: llvm-readobj -sections -section-data %t2 | FileCheck %s
 
-## Check we decompress section and remove ".z" prefix specific for zlib-gnu compression.
+## Check we uncompress section and remove ".z" prefix specific for zlib-gnu compression.
 # CHECK:      Section {
 # CHECK:        Index:
 # CHECK:        Name: .debug_str
 # CHECK-NEXT:   Type: SHT_PROGBITS
 # CHECK-NEXT:   Flags [
 # CHECK-NEXT:     SHF_MERGE
 # CHECK-NEXT:     SHF_STRINGS
 # CHECK-NEXT:   ]

lld/test/ELF/strip-debug.s

 # REQUIRES: x86
 # RUN: llvm-mc -filetype=obj -triple=x86_64-unknown-linux %s -o %t
 # RUN: ld.lld %t -o %t2 --strip-debug
 # RUN: llvm-readobj -sections %t2 | FileCheck %s
 # RUN: ld.lld %t -o %t2 -S
 # RUN: llvm-readobj -sections %t2 | FileCheck %s
 # RUN: ld.lld %t -o %t2 --strip-all
 # RUN: llvm-readobj -sections %t2 | FileCheck %s
 
 # CHECK-NOT: Foo
 # CHECK-NOT: Bar
 
 .section .debug_Foo,"",@progbits
 .section .zdebug_Bar,"",@progbits
+.ascii "ZLIB"
+.quad 0