1 //===- tools/dsymutil/DwarfLinker.cpp - Dwarf debug info linker -----------===//
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 #include "BinaryHolder.h"
13 #include "llvm/ADT/IntervalMap.h"
14 #include "llvm/ADT/StringMap.h"
15 #include "llvm/ADT/STLExtras.h"
16 #include "llvm/CodeGen/AsmPrinter.h"
17 #include "llvm/CodeGen/DIE.h"
18 #include "llvm/DebugInfo/DWARF/DWARFContext.h"
19 #include "llvm/DebugInfo/DWARF/DWARFDebugInfoEntry.h"
20 #include "llvm/DebugInfo/DWARF/DWARFFormValue.h"
21 #include "llvm/MC/MCAsmBackend.h"
22 #include "llvm/MC/MCAsmInfo.h"
23 #include "llvm/MC/MCContext.h"
24 #include "llvm/MC/MCCodeEmitter.h"
25 #include "llvm/MC/MCInstrInfo.h"
26 #include "llvm/MC/MCObjectFileInfo.h"
27 #include "llvm/MC/MCRegisterInfo.h"
28 #include "llvm/MC/MCStreamer.h"
29 #include "llvm/Object/MachO.h"
30 #include "llvm/Support/Dwarf.h"
31 #include "llvm/Support/LEB128.h"
32 #include "llvm/Support/TargetRegistry.h"
33 #include "llvm/Target/TargetMachine.h"
34 #include "llvm/Target/TargetOptions.h"
42 void warn(const Twine &Warning, const Twine &Context) {
43 errs() << Twine("while processing ") + Context + ":\n";
44 errs() << Twine("warning: ") + Warning + "\n";
47 bool error(const Twine &Error, const Twine &Context) {
48 errs() << Twine("while processing ") + Context + ":\n";
49 errs() << Twine("error: ") + Error + "\n";
53 template <typename KeyT, typename ValT>
54 using HalfOpenIntervalMap =
55 IntervalMap<KeyT, ValT, IntervalMapImpl::NodeSizer<KeyT, ValT>::LeafSize,
56 IntervalMapHalfOpenInfo<KeyT>>;
58 /// \brief Stores all information relating to a compile unit, be it in
59 /// its original instance in the object file to its brand new cloned
60 /// and linked DIE tree.
63 /// \brief Information gathered about a DIE in the object file.
65 int64_t AddrAdjust; ///< Address offset to apply to the described entity.
66 DIE *Clone; ///< Cloned version of that DIE.
67 uint32_t ParentIdx; ///< The index of this DIE's parent.
68 bool Keep; ///< Is the DIE part of the linked output?
69 bool InDebugMap; ///< Was this DIE's entity found in the map?
72 CompileUnit(DWARFUnit &OrigUnit)
73 : OrigUnit(OrigUnit), LowPc(UINT64_MAX), HighPc(0), RangeAlloc(),
75 Info.resize(OrigUnit.getNumDIEs());
78 CompileUnit(CompileUnit &&RHS)
79 : OrigUnit(RHS.OrigUnit), Info(std::move(RHS.Info)),
80 CUDie(std::move(RHS.CUDie)), StartOffset(RHS.StartOffset),
81 NextUnitOffset(RHS.NextUnitOffset), RangeAlloc(), Ranges(RangeAlloc) {
82 // The CompileUnit container has been 'reserve()'d with the right
83 // size. We cannot move the IntervalMap anyway.
84 llvm_unreachable("CompileUnits should not be moved.");
87 DWARFUnit &getOrigUnit() const { return OrigUnit; }
89 DIE *getOutputUnitDIE() const { return CUDie.get(); }
90 void setOutputUnitDIE(DIE *Die) { CUDie.reset(Die); }
92 DIEInfo &getInfo(unsigned Idx) { return Info[Idx]; }
93 const DIEInfo &getInfo(unsigned Idx) const { return Info[Idx]; }
95 uint64_t getStartOffset() const { return StartOffset; }
96 uint64_t getNextUnitOffset() const { return NextUnitOffset; }
98 uint64_t getLowPc() const { return LowPc; }
99 uint64_t getHighPc() const { return HighPc; }
101 void setStartOffset(uint64_t DebugInfoSize) { StartOffset = DebugInfoSize; }
103 /// \brief Compute the end offset for this unit. Must be
104 /// called after the CU's DIEs have been cloned.
105 /// \returns the next unit offset (which is also the current
106 /// debug_info section size).
107 uint64_t computeNextUnitOffset();
109 /// \brief Keep track of a forward reference to DIE \p Die by
110 /// \p Attr. The attribute should be fixed up later to point to the
111 /// absolute offset of \p Die in the debug_info section.
112 void noteForwardReference(DIE *Die, DIEInteger *Attr);
114 /// \brief Apply all fixups recored by noteForwardReference().
115 void fixupForwardReferences();
117 /// \brief Add a function range [\p LowPC, \p HighPC) that is
118 /// relocatad by applying offset \p PCOffset.
119 void addFunctionRange(uint64_t LowPC, uint64_t HighPC, int64_t PCOffset);
123 std::vector<DIEInfo> Info; ///< DIE info indexed by DIE index.
124 std::unique_ptr<DIE> CUDie; ///< Root of the linked DIE tree.
126 uint64_t StartOffset;
127 uint64_t NextUnitOffset;
132 /// \brief A list of attributes to fixup with the absolute offset of
133 /// a DIE in the debug_info section.
135 /// The offsets for the attributes in this array couldn't be set while
136 /// cloning because for forward refences the target DIE's offset isn't
137 /// known you emit the reference attribute.
138 std::vector<std::pair<DIE *, DIEInteger *>> ForwardDIEReferences;
140 HalfOpenIntervalMap<uint64_t, int64_t>::Allocator RangeAlloc;
141 /// \brief The ranges in that interval map are the PC ranges for
142 /// functions in this unit, associated with the PC offset to apply
143 /// to the addresses to get the linked address.
144 HalfOpenIntervalMap<uint64_t, int64_t> Ranges;
147 uint64_t CompileUnit::computeNextUnitOffset() {
148 NextUnitOffset = StartOffset + 11 /* Header size */;
149 // The root DIE might be null, meaning that the Unit had nothing to
150 // contribute to the linked output. In that case, we will emit the
151 // unit header without any actual DIE.
153 NextUnitOffset += CUDie->getSize();
154 return NextUnitOffset;
157 /// \brief Keep track of a forward reference to \p Die.
158 void CompileUnit::noteForwardReference(DIE *Die, DIEInteger *Attr) {
159 ForwardDIEReferences.emplace_back(Die, Attr);
162 /// \brief Apply all fixups recorded by noteForwardReference().
163 void CompileUnit::fixupForwardReferences() {
164 for (const auto &Ref : ForwardDIEReferences)
165 Ref.second->setValue(Ref.first->getOffset() + getStartOffset());
168 void CompileUnit::addFunctionRange(uint64_t FuncLowPc, uint64_t FuncHighPc,
170 Ranges.insert(FuncLowPc, FuncHighPc, PcOffset);
171 this->LowPc = std::min(LowPc, FuncLowPc + PcOffset);
172 this->HighPc = std::max(HighPc, FuncHighPc + PcOffset);
175 /// \brief A string table that doesn't need relocations.
177 /// We are doing a final link, no need for a string table that
178 /// has relocation entries for every reference to it. This class
179 /// provides this ablitity by just associating offsets with
181 class NonRelocatableStringpool {
183 /// \brief Entries are stored into the StringMap and simply linked
184 /// together through the second element of this pair in order to
185 /// keep track of insertion order.
186 typedef StringMap<std::pair<uint32_t, StringMapEntryBase *>, BumpPtrAllocator>
189 NonRelocatableStringpool()
190 : CurrentEndOffset(0), Sentinel(0), Last(&Sentinel) {
191 // Legacy dsymutil puts an empty string at the start of the line
196 /// \brief Get the offset of string \p S in the string table. This
197 /// can insert a new element or return the offset of a preexisitng
199 uint32_t getStringOffset(StringRef S);
201 /// \brief Get permanent storage for \p S (but do not necessarily
202 /// emit \p S in the output section).
203 /// \returns The StringRef that points to permanent storage to use
204 /// in place of \p S.
205 StringRef internString(StringRef S);
207 // \brief Return the first entry of the string table.
208 const MapTy::MapEntryTy *getFirstEntry() const {
209 return getNextEntry(&Sentinel);
212 // \brief Get the entry following \p E in the string table or null
213 // if \p E was the last entry.
214 const MapTy::MapEntryTy *getNextEntry(const MapTy::MapEntryTy *E) const {
215 return static_cast<const MapTy::MapEntryTy *>(E->getValue().second);
218 uint64_t getSize() { return CurrentEndOffset; }
222 uint32_t CurrentEndOffset;
223 MapTy::MapEntryTy Sentinel, *Last;
226 /// \brief Get the offset of string \p S in the string table. This
227 /// can insert a new element or return the offset of a preexisitng
229 uint32_t NonRelocatableStringpool::getStringOffset(StringRef S) {
230 if (S.empty() && !Strings.empty())
233 std::pair<uint32_t, StringMapEntryBase *> Entry(0, nullptr);
237 // A non-empty string can't be at offset 0, so if we have an entry
238 // with a 0 offset, it must be a previously interned string.
239 std::tie(It, Inserted) = Strings.insert(std::make_pair(S, Entry));
240 if (Inserted || It->getValue().first == 0) {
241 // Set offset and chain at the end of the entries list.
242 It->getValue().first = CurrentEndOffset;
243 CurrentEndOffset += S.size() + 1; // +1 for the '\0'.
244 Last->getValue().second = &*It;
247 return It->getValue().first;
250 /// \brief Put \p S into the StringMap so that it gets permanent
251 /// storage, but do not actually link it in the chain of elements
252 /// that go into the output section. A latter call to
253 /// getStringOffset() with the same string will chain it though.
254 StringRef NonRelocatableStringpool::internString(StringRef S) {
255 std::pair<uint32_t, StringMapEntryBase *> Entry(0, nullptr);
256 auto InsertResult = Strings.insert(std::make_pair(S, Entry));
257 return InsertResult.first->getKey();
260 /// \brief The Dwarf streaming logic
262 /// All interactions with the MC layer that is used to build the debug
263 /// information binary representation are handled in this class.
264 class DwarfStreamer {
265 /// \defgroup MCObjects MC layer objects constructed by the streamer
267 std::unique_ptr<MCRegisterInfo> MRI;
268 std::unique_ptr<MCAsmInfo> MAI;
269 std::unique_ptr<MCObjectFileInfo> MOFI;
270 std::unique_ptr<MCContext> MC;
271 MCAsmBackend *MAB; // Owned by MCStreamer
272 std::unique_ptr<MCInstrInfo> MII;
273 std::unique_ptr<MCSubtargetInfo> MSTI;
274 MCCodeEmitter *MCE; // Owned by MCStreamer
275 MCStreamer *MS; // Owned by AsmPrinter
276 std::unique_ptr<TargetMachine> TM;
277 std::unique_ptr<AsmPrinter> Asm;
280 /// \brief the file we stream the linked Dwarf to.
281 std::unique_ptr<raw_fd_ostream> OutFile;
284 /// \brief Actually create the streamer and the ouptut file.
286 /// This could be done directly in the constructor, but it feels
287 /// more natural to handle errors through return value.
288 bool init(Triple TheTriple, StringRef OutputFilename);
290 /// \brief Dump the file to the disk.
293 AsmPrinter &getAsmPrinter() const { return *Asm; }
295 /// \brief Set the current output section to debug_info and change
296 /// the MC Dwarf version to \p DwarfVersion.
297 void switchToDebugInfoSection(unsigned DwarfVersion);
299 /// \brief Emit the compilation unit header for \p Unit in the
300 /// debug_info section.
302 /// As a side effect, this also switches the current Dwarf version
303 /// of the MC layer to the one of U.getOrigUnit().
304 void emitCompileUnitHeader(CompileUnit &Unit);
306 /// \brief Recursively emit the DIE tree rooted at \p Die.
307 void emitDIE(DIE &Die);
309 /// \brief Emit the abbreviation table \p Abbrevs to the
310 /// debug_abbrev section.
311 void emitAbbrevs(const std::vector<DIEAbbrev *> &Abbrevs);
313 /// \brief Emit the string table described by \p Pool.
314 void emitStrings(const NonRelocatableStringpool &Pool);
317 bool DwarfStreamer::init(Triple TheTriple, StringRef OutputFilename) {
318 std::string ErrorStr;
319 std::string TripleName;
320 StringRef Context = "dwarf streamer init";
323 const Target *TheTarget =
324 TargetRegistry::lookupTarget(TripleName, TheTriple, ErrorStr);
326 return error(ErrorStr, Context);
327 TripleName = TheTriple.getTriple();
329 // Create all the MC Objects.
330 MRI.reset(TheTarget->createMCRegInfo(TripleName));
332 return error(Twine("no register info for target ") + TripleName, Context);
334 MAI.reset(TheTarget->createMCAsmInfo(*MRI, TripleName));
336 return error("no asm info for target " + TripleName, Context);
338 MOFI.reset(new MCObjectFileInfo);
339 MC.reset(new MCContext(MAI.get(), MRI.get(), MOFI.get()));
340 MOFI->InitMCObjectFileInfo(TripleName, Reloc::Default, CodeModel::Default,
343 MAB = TheTarget->createMCAsmBackend(*MRI, TripleName, "");
345 return error("no asm backend for target " + TripleName, Context);
347 MII.reset(TheTarget->createMCInstrInfo());
349 return error("no instr info info for target " + TripleName, Context);
351 MSTI.reset(TheTarget->createMCSubtargetInfo(TripleName, "", ""));
353 return error("no subtarget info for target " + TripleName, Context);
355 MCE = TheTarget->createMCCodeEmitter(*MII, *MRI, *MC);
357 return error("no code emitter for target " + TripleName, Context);
359 // Create the output file.
362 llvm::make_unique<raw_fd_ostream>(OutputFilename, EC, sys::fs::F_None);
364 return error(Twine(OutputFilename) + ": " + EC.message(), Context);
366 MS = TheTarget->createMCObjectStreamer(TripleName, *MC, *MAB, *OutFile, MCE,
369 return error("no object streamer for target " + TripleName, Context);
371 // Finally create the AsmPrinter we'll use to emit the DIEs.
372 TM.reset(TheTarget->createTargetMachine(TripleName, "", "", TargetOptions()));
374 return error("no target machine for target " + TripleName, Context);
376 Asm.reset(TheTarget->createAsmPrinter(*TM, std::unique_ptr<MCStreamer>(MS)));
378 return error("no asm printer for target " + TripleName, Context);
383 bool DwarfStreamer::finish() {
388 /// \brief Set the current output section to debug_info and change
389 /// the MC Dwarf version to \p DwarfVersion.
390 void DwarfStreamer::switchToDebugInfoSection(unsigned DwarfVersion) {
391 MS->SwitchSection(MOFI->getDwarfInfoSection());
392 MC->setDwarfVersion(DwarfVersion);
395 /// \brief Emit the compilation unit header for \p Unit in the
396 /// debug_info section.
398 /// A Dwarf scetion header is encoded as:
399 /// uint32_t Unit length (omiting this field)
401 /// uint32_t Abbreviation table offset
402 /// uint8_t Address size
404 /// Leading to a total of 11 bytes.
405 void DwarfStreamer::emitCompileUnitHeader(CompileUnit &Unit) {
406 unsigned Version = Unit.getOrigUnit().getVersion();
407 switchToDebugInfoSection(Version);
409 // Emit size of content not including length itself. The size has
410 // already been computed in CompileUnit::computeOffsets(). Substract
411 // 4 to that size to account for the length field.
412 Asm->EmitInt32(Unit.getNextUnitOffset() - Unit.getStartOffset() - 4);
413 Asm->EmitInt16(Version);
414 // We share one abbreviations table across all units so it's always at the
415 // start of the section.
417 Asm->EmitInt8(Unit.getOrigUnit().getAddressByteSize());
420 /// \brief Emit the \p Abbrevs array as the shared abbreviation table
421 /// for the linked Dwarf file.
422 void DwarfStreamer::emitAbbrevs(const std::vector<DIEAbbrev *> &Abbrevs) {
423 MS->SwitchSection(MOFI->getDwarfAbbrevSection());
424 Asm->emitDwarfAbbrevs(Abbrevs);
427 /// \brief Recursively emit the DIE tree rooted at \p Die.
428 void DwarfStreamer::emitDIE(DIE &Die) {
429 MS->SwitchSection(MOFI->getDwarfInfoSection());
430 Asm->emitDwarfDIE(Die);
433 /// \brief Emit the debug_str section stored in \p Pool.
434 void DwarfStreamer::emitStrings(const NonRelocatableStringpool &Pool) {
435 Asm->OutStreamer.SwitchSection(MOFI->getDwarfStrSection());
436 for (auto *Entry = Pool.getFirstEntry(); Entry;
437 Entry = Pool.getNextEntry(Entry))
438 Asm->OutStreamer.EmitBytes(
439 StringRef(Entry->getKey().data(), Entry->getKey().size() + 1));
442 /// \brief The core of the Dwarf linking logic.
444 /// The link of the dwarf information from the object files will be
445 /// driven by the selection of 'root DIEs', which are DIEs that
446 /// describe variables or functions that are present in the linked
447 /// binary (and thus have entries in the debug map). All the debug
448 /// information that will be linked (the DIEs, but also the line
449 /// tables, ranges, ...) is derived from that set of root DIEs.
451 /// The root DIEs are identified because they contain relocations that
452 /// correspond to a debug map entry at specific places (the low_pc for
453 /// a function, the location for a variable). These relocations are
454 /// called ValidRelocs in the DwarfLinker and are gathered as a very
455 /// first step when we start processing a DebugMapObject.
458 DwarfLinker(StringRef OutputFilename, const LinkOptions &Options)
459 : OutputFilename(OutputFilename), Options(Options),
460 BinHolder(Options.Verbose) {}
463 for (auto *Abbrev : Abbreviations)
467 /// \brief Link the contents of the DebugMap.
468 bool link(const DebugMap &);
471 /// \brief Called at the start of a debug object link.
472 void startDebugObject(DWARFContext &);
474 /// \brief Called at the end of a debug object link.
475 void endDebugObject();
477 /// \defgroup FindValidRelocations Translate debug map into a list
478 /// of relevant relocations
485 const DebugMapObject::DebugMapEntry *Mapping;
487 ValidReloc(uint32_t Offset, uint32_t Size, uint64_t Addend,
488 const DebugMapObject::DebugMapEntry *Mapping)
489 : Offset(Offset), Size(Size), Addend(Addend), Mapping(Mapping) {}
491 bool operator<(const ValidReloc &RHS) const { return Offset < RHS.Offset; }
494 /// \brief The valid relocations for the current DebugMapObject.
495 /// This vector is sorted by relocation offset.
496 std::vector<ValidReloc> ValidRelocs;
498 /// \brief Index into ValidRelocs of the next relocation to
499 /// consider. As we walk the DIEs in acsending file offset and as
500 /// ValidRelocs is sorted by file offset, keeping this index
501 /// uptodate is all we have to do to have a cheap lookup during the
502 /// root DIE selection and during DIE cloning.
503 unsigned NextValidReloc;
505 bool findValidRelocsInDebugInfo(const object::ObjectFile &Obj,
506 const DebugMapObject &DMO);
508 bool findValidRelocs(const object::SectionRef &Section,
509 const object::ObjectFile &Obj,
510 const DebugMapObject &DMO);
512 void findValidRelocsMachO(const object::SectionRef &Section,
513 const object::MachOObjectFile &Obj,
514 const DebugMapObject &DMO);
517 /// \defgroup FindRootDIEs Find DIEs corresponding to debug map entries.
520 /// \brief Recursively walk the \p DIE tree and look for DIEs to
521 /// keep. Store that information in \p CU's DIEInfo.
522 void lookForDIEsToKeep(const DWARFDebugInfoEntryMinimal &DIE,
523 const DebugMapObject &DMO, CompileUnit &CU,
526 /// \brief Flags passed to DwarfLinker::lookForDIEsToKeep
528 TF_Keep = 1 << 0, ///< Mark the traversed DIEs as kept.
529 TF_InFunctionScope = 1 << 1, ///< Current scope is a fucntion scope.
530 TF_DependencyWalk = 1 << 2, ///< Walking the dependencies of a kept DIE.
531 TF_ParentWalk = 1 << 3, ///< Walking up the parents of a kept DIE.
534 /// \brief Mark the passed DIE as well as all the ones it depends on
536 void keepDIEAndDenpendencies(const DWARFDebugInfoEntryMinimal &DIE,
537 CompileUnit::DIEInfo &MyInfo,
538 const DebugMapObject &DMO, CompileUnit &CU,
541 unsigned shouldKeepDIE(const DWARFDebugInfoEntryMinimal &DIE,
542 CompileUnit &Unit, CompileUnit::DIEInfo &MyInfo,
545 unsigned shouldKeepVariableDIE(const DWARFDebugInfoEntryMinimal &DIE,
547 CompileUnit::DIEInfo &MyInfo, unsigned Flags);
549 unsigned shouldKeepSubprogramDIE(const DWARFDebugInfoEntryMinimal &DIE,
551 CompileUnit::DIEInfo &MyInfo,
554 bool hasValidRelocation(uint32_t StartOffset, uint32_t EndOffset,
555 CompileUnit::DIEInfo &Info);
558 /// \defgroup Linking Methods used to link the debug information
561 /// \brief Recursively clone \p InputDIE into an tree of DIE objects
562 /// where useless (as decided by lookForDIEsToKeep()) bits have been
563 /// stripped out and addresses have been rewritten according to the
566 /// \param OutOffset is the offset the cloned DIE in the output
568 /// \param PCOffset (while cloning a function scope) is the offset
569 /// applied to the entry point of the function to get the linked address.
571 /// \returns the root of the cloned tree.
572 DIE *cloneDIE(const DWARFDebugInfoEntryMinimal &InputDIE, CompileUnit &U,
573 int64_t PCOffset, uint32_t OutOffset);
575 typedef DWARFAbbreviationDeclaration::AttributeSpec AttributeSpec;
577 /// \brief Information gathered and exchanged between the various
578 /// clone*Attributes helpers about the attributes of a particular DIE.
579 struct AttributesInfo {
580 uint64_t OrigHighPc; ///< Value of AT_high_pc in the input DIE
581 int64_t PCOffset; ///< Offset to apply to PC addresses inside a function.
583 AttributesInfo() : OrigHighPc(0), PCOffset(0) {}
586 /// \brief Helper for cloneDIE.
587 unsigned cloneAttribute(DIE &Die, const DWARFDebugInfoEntryMinimal &InputDIE,
588 CompileUnit &U, const DWARFFormValue &Val,
589 const AttributeSpec AttrSpec, unsigned AttrSize,
590 AttributesInfo &AttrInfo);
592 /// \brief Helper for cloneDIE.
593 unsigned cloneStringAttribute(DIE &Die, AttributeSpec AttrSpec,
594 const DWARFFormValue &Val, const DWARFUnit &U);
596 /// \brief Helper for cloneDIE.
598 cloneDieReferenceAttribute(DIE &Die,
599 const DWARFDebugInfoEntryMinimal &InputDIE,
600 AttributeSpec AttrSpec, unsigned AttrSize,
601 const DWARFFormValue &Val, const DWARFUnit &U);
603 /// \brief Helper for cloneDIE.
604 unsigned cloneBlockAttribute(DIE &Die, AttributeSpec AttrSpec,
605 const DWARFFormValue &Val, unsigned AttrSize);
607 /// \brief Helper for cloneDIE.
608 unsigned cloneAddressAttribute(DIE &Die, AttributeSpec AttrSpec,
609 const DWARFFormValue &Val,
610 const CompileUnit &Unit, AttributesInfo &Info);
612 /// \brief Helper for cloneDIE.
613 unsigned cloneScalarAttribute(DIE &Die,
614 const DWARFDebugInfoEntryMinimal &InputDIE,
615 const CompileUnit &U, AttributeSpec AttrSpec,
616 const DWARFFormValue &Val, unsigned AttrSize);
618 /// \brief Helper for cloneDIE.
619 bool applyValidRelocs(MutableArrayRef<char> Data, uint32_t BaseOffset,
620 bool isLittleEndian);
622 /// \brief Assign an abbreviation number to \p Abbrev
623 void AssignAbbrev(DIEAbbrev &Abbrev);
625 /// \brief FoldingSet that uniques the abbreviations.
626 FoldingSet<DIEAbbrev> AbbreviationsSet;
627 /// \brief Storage for the unique Abbreviations.
628 /// This is passed to AsmPrinter::emitDwarfAbbrevs(), thus it cannot
629 /// be changed to a vecot of unique_ptrs.
630 std::vector<DIEAbbrev *> Abbreviations;
632 /// \brief DIELoc objects that need to be destructed (but not freed!).
633 std::vector<DIELoc *> DIELocs;
634 /// \brief DIEBlock objects that need to be destructed (but not freed!).
635 std::vector<DIEBlock *> DIEBlocks;
636 /// \brief Allocator used for all the DIEValue objects.
637 BumpPtrAllocator DIEAlloc;
640 /// \defgroup Helpers Various helper methods.
643 const DWARFDebugInfoEntryMinimal *
644 resolveDIEReference(DWARFFormValue &RefValue, const DWARFUnit &Unit,
645 const DWARFDebugInfoEntryMinimal &DIE,
646 CompileUnit *&ReferencedCU);
648 CompileUnit *getUnitForOffset(unsigned Offset);
650 void reportWarning(const Twine &Warning, const DWARFUnit *Unit = nullptr,
651 const DWARFDebugInfoEntryMinimal *DIE = nullptr);
653 bool createStreamer(Triple TheTriple, StringRef OutputFilename);
657 std::string OutputFilename;
659 BinaryHolder BinHolder;
660 std::unique_ptr<DwarfStreamer> Streamer;
662 /// The units of the current debug map object.
663 std::vector<CompileUnit> Units;
665 /// The debug map object curently under consideration.
666 DebugMapObject *CurrentDebugObject;
668 /// \brief The Dwarf string pool
669 NonRelocatableStringpool StringPool;
672 /// \brief Similar to DWARFUnitSection::getUnitForOffset(), but
673 /// returning our CompileUnit object instead.
674 CompileUnit *DwarfLinker::getUnitForOffset(unsigned Offset) {
676 std::upper_bound(Units.begin(), Units.end(), Offset,
677 [](uint32_t LHS, const CompileUnit &RHS) {
678 return LHS < RHS.getOrigUnit().getNextUnitOffset();
680 return CU != Units.end() ? &*CU : nullptr;
683 /// \brief Resolve the DIE attribute reference that has been
684 /// extracted in \p RefValue. The resulting DIE migh be in another
685 /// CompileUnit which is stored into \p ReferencedCU.
686 /// \returns null if resolving fails for any reason.
687 const DWARFDebugInfoEntryMinimal *DwarfLinker::resolveDIEReference(
688 DWARFFormValue &RefValue, const DWARFUnit &Unit,
689 const DWARFDebugInfoEntryMinimal &DIE, CompileUnit *&RefCU) {
690 assert(RefValue.isFormClass(DWARFFormValue::FC_Reference));
691 uint64_t RefOffset = *RefValue.getAsReference(&Unit);
693 if ((RefCU = getUnitForOffset(RefOffset)))
694 if (const auto *RefDie = RefCU->getOrigUnit().getDIEForOffset(RefOffset))
697 reportWarning("could not find referenced DIE", &Unit, &DIE);
701 /// \brief Report a warning to the user, optionaly including
702 /// information about a specific \p DIE related to the warning.
703 void DwarfLinker::reportWarning(const Twine &Warning, const DWARFUnit *Unit,
704 const DWARFDebugInfoEntryMinimal *DIE) {
705 StringRef Context = "<debug map>";
706 if (CurrentDebugObject)
707 Context = CurrentDebugObject->getObjectFilename();
708 warn(Warning, Context);
710 if (!Options.Verbose || !DIE)
713 errs() << " in DIE:\n";
714 DIE->dump(errs(), const_cast<DWARFUnit *>(Unit), 0 /* RecurseDepth */,
718 bool DwarfLinker::createStreamer(Triple TheTriple, StringRef OutputFilename) {
719 if (Options.NoOutput)
722 Streamer = llvm::make_unique<DwarfStreamer>();
723 return Streamer->init(TheTriple, OutputFilename);
726 /// \brief Recursive helper to gather the child->parent relationships in the
727 /// original compile unit.
728 static void gatherDIEParents(const DWARFDebugInfoEntryMinimal *DIE,
729 unsigned ParentIdx, CompileUnit &CU) {
730 unsigned MyIdx = CU.getOrigUnit().getDIEIndex(DIE);
731 CU.getInfo(MyIdx).ParentIdx = ParentIdx;
733 if (DIE->hasChildren())
734 for (auto *Child = DIE->getFirstChild(); Child && !Child->isNULL();
735 Child = Child->getSibling())
736 gatherDIEParents(Child, MyIdx, CU);
739 static bool dieNeedsChildrenToBeMeaningful(uint32_t Tag) {
743 case dwarf::DW_TAG_subprogram:
744 case dwarf::DW_TAG_lexical_block:
745 case dwarf::DW_TAG_subroutine_type:
746 case dwarf::DW_TAG_structure_type:
747 case dwarf::DW_TAG_class_type:
748 case dwarf::DW_TAG_union_type:
751 llvm_unreachable("Invalid Tag");
754 void DwarfLinker::startDebugObject(DWARFContext &Dwarf) {
755 Units.reserve(Dwarf.getNumCompileUnits());
759 void DwarfLinker::endDebugObject() {
763 for (auto *Block : DIEBlocks)
765 for (auto *Loc : DIELocs)
773 /// \brief Iterate over the relocations of the given \p Section and
774 /// store the ones that correspond to debug map entries into the
775 /// ValidRelocs array.
776 void DwarfLinker::findValidRelocsMachO(const object::SectionRef &Section,
777 const object::MachOObjectFile &Obj,
778 const DebugMapObject &DMO) {
780 Section.getContents(Contents);
781 DataExtractor Data(Contents, Obj.isLittleEndian(), 0);
783 for (const object::RelocationRef &Reloc : Section.relocations()) {
784 object::DataRefImpl RelocDataRef = Reloc.getRawDataRefImpl();
785 MachO::any_relocation_info MachOReloc = Obj.getRelocation(RelocDataRef);
786 unsigned RelocSize = 1 << Obj.getAnyRelocationLength(MachOReloc);
788 if ((RelocSize != 4 && RelocSize != 8) || Reloc.getOffset(Offset64)) {
789 reportWarning(" unsupported relocation in debug_info section.");
792 uint32_t Offset = Offset64;
793 // Mach-o uses REL relocations, the addend is at the relocation offset.
794 uint64_t Addend = Data.getUnsigned(&Offset, RelocSize);
796 auto Sym = Reloc.getSymbol();
797 if (Sym != Obj.symbol_end()) {
798 StringRef SymbolName;
799 if (Sym->getName(SymbolName)) {
800 reportWarning("error getting relocation symbol name.");
803 if (const auto *Mapping = DMO.lookupSymbol(SymbolName))
804 ValidRelocs.emplace_back(Offset64, RelocSize, Addend, Mapping);
805 } else if (const auto *Mapping = DMO.lookupObjectAddress(Addend)) {
806 // Do not store the addend. The addend was the address of the
807 // symbol in the object file, the address in the binary that is
808 // stored in the debug map doesn't need to be offseted.
809 ValidRelocs.emplace_back(Offset64, RelocSize, 0, Mapping);
814 /// \brief Dispatch the valid relocation finding logic to the
815 /// appropriate handler depending on the object file format.
816 bool DwarfLinker::findValidRelocs(const object::SectionRef &Section,
817 const object::ObjectFile &Obj,
818 const DebugMapObject &DMO) {
819 // Dispatch to the right handler depending on the file type.
820 if (auto *MachOObj = dyn_cast<object::MachOObjectFile>(&Obj))
821 findValidRelocsMachO(Section, *MachOObj, DMO);
823 reportWarning(Twine("unsupported object file type: ") + Obj.getFileName());
825 if (ValidRelocs.empty())
828 // Sort the relocations by offset. We will walk the DIEs linearly in
829 // the file, this allows us to just keep an index in the relocation
830 // array that we advance during our walk, rather than resorting to
831 // some associative container. See DwarfLinker::NextValidReloc.
832 std::sort(ValidRelocs.begin(), ValidRelocs.end());
836 /// \brief Look for relocations in the debug_info section that match
837 /// entries in the debug map. These relocations will drive the Dwarf
838 /// link by indicating which DIEs refer to symbols present in the
840 /// \returns wether there are any valid relocations in the debug info.
841 bool DwarfLinker::findValidRelocsInDebugInfo(const object::ObjectFile &Obj,
842 const DebugMapObject &DMO) {
843 // Find the debug_info section.
844 for (const object::SectionRef &Section : Obj.sections()) {
845 StringRef SectionName;
846 Section.getName(SectionName);
847 SectionName = SectionName.substr(SectionName.find_first_not_of("._"));
848 if (SectionName != "debug_info")
850 return findValidRelocs(Section, Obj, DMO);
855 /// \brief Checks that there is a relocation against an actual debug
856 /// map entry between \p StartOffset and \p NextOffset.
858 /// This function must be called with offsets in strictly ascending
859 /// order because it never looks back at relocations it already 'went past'.
860 /// \returns true and sets Info.InDebugMap if it is the case.
861 bool DwarfLinker::hasValidRelocation(uint32_t StartOffset, uint32_t EndOffset,
862 CompileUnit::DIEInfo &Info) {
863 assert(NextValidReloc == 0 ||
864 StartOffset > ValidRelocs[NextValidReloc - 1].Offset);
865 if (NextValidReloc >= ValidRelocs.size())
868 uint64_t RelocOffset = ValidRelocs[NextValidReloc].Offset;
870 // We might need to skip some relocs that we didn't consider. For
871 // example the high_pc of a discarded DIE might contain a reloc that
872 // is in the list because it actually corresponds to the start of a
873 // function that is in the debug map.
874 while (RelocOffset < StartOffset && NextValidReloc < ValidRelocs.size() - 1)
875 RelocOffset = ValidRelocs[++NextValidReloc].Offset;
877 if (RelocOffset < StartOffset || RelocOffset >= EndOffset)
880 const auto &ValidReloc = ValidRelocs[NextValidReloc++];
882 outs() << "Found valid debug map entry: " << ValidReloc.Mapping->getKey()
883 << " " << format("\t%016" PRIx64 " => %016" PRIx64,
884 ValidReloc.Mapping->getValue().ObjectAddress,
885 ValidReloc.Mapping->getValue().BinaryAddress);
887 Info.AddrAdjust = int64_t(ValidReloc.Mapping->getValue().BinaryAddress) +
889 ValidReloc.Mapping->getValue().ObjectAddress;
890 Info.InDebugMap = true;
894 /// \brief Get the starting and ending (exclusive) offset for the
895 /// attribute with index \p Idx descibed by \p Abbrev. \p Offset is
896 /// supposed to point to the position of the first attribute described
898 /// \return [StartOffset, EndOffset) as a pair.
899 static std::pair<uint32_t, uint32_t>
900 getAttributeOffsets(const DWARFAbbreviationDeclaration *Abbrev, unsigned Idx,
901 unsigned Offset, const DWARFUnit &Unit) {
902 DataExtractor Data = Unit.getDebugInfoExtractor();
904 for (unsigned i = 0; i < Idx; ++i)
905 DWARFFormValue::skipValue(Abbrev->getFormByIndex(i), Data, &Offset, &Unit);
907 uint32_t End = Offset;
908 DWARFFormValue::skipValue(Abbrev->getFormByIndex(Idx), Data, &End, &Unit);
910 return std::make_pair(Offset, End);
913 /// \brief Check if a variable describing DIE should be kept.
914 /// \returns updated TraversalFlags.
915 unsigned DwarfLinker::shouldKeepVariableDIE(
916 const DWARFDebugInfoEntryMinimal &DIE, CompileUnit &Unit,
917 CompileUnit::DIEInfo &MyInfo, unsigned Flags) {
918 const auto *Abbrev = DIE.getAbbreviationDeclarationPtr();
920 // Global variables with constant value can always be kept.
921 if (!(Flags & TF_InFunctionScope) &&
922 Abbrev->findAttributeIndex(dwarf::DW_AT_const_value) != -1U) {
923 MyInfo.InDebugMap = true;
924 return Flags | TF_Keep;
927 uint32_t LocationIdx = Abbrev->findAttributeIndex(dwarf::DW_AT_location);
928 if (LocationIdx == -1U)
931 uint32_t Offset = DIE.getOffset() + getULEB128Size(Abbrev->getCode());
932 const DWARFUnit &OrigUnit = Unit.getOrigUnit();
933 uint32_t LocationOffset, LocationEndOffset;
934 std::tie(LocationOffset, LocationEndOffset) =
935 getAttributeOffsets(Abbrev, LocationIdx, Offset, OrigUnit);
937 // See if there is a relocation to a valid debug map entry inside
938 // this variable's location. The order is important here. We want to
939 // always check in the variable has a valid relocation, so that the
940 // DIEInfo is filled. However, we don't want a static variable in a
941 // function to force us to keep the enclosing function.
942 if (!hasValidRelocation(LocationOffset, LocationEndOffset, MyInfo) ||
943 (Flags & TF_InFunctionScope))
947 DIE.dump(outs(), const_cast<DWARFUnit *>(&OrigUnit), 0, 8 /* Indent */);
949 return Flags | TF_Keep;
952 /// \brief Check if a function describing DIE should be kept.
953 /// \returns updated TraversalFlags.
954 unsigned DwarfLinker::shouldKeepSubprogramDIE(
955 const DWARFDebugInfoEntryMinimal &DIE, CompileUnit &Unit,
956 CompileUnit::DIEInfo &MyInfo, unsigned Flags) {
957 const auto *Abbrev = DIE.getAbbreviationDeclarationPtr();
959 Flags |= TF_InFunctionScope;
961 uint32_t LowPcIdx = Abbrev->findAttributeIndex(dwarf::DW_AT_low_pc);
965 uint32_t Offset = DIE.getOffset() + getULEB128Size(Abbrev->getCode());
966 const DWARFUnit &OrigUnit = Unit.getOrigUnit();
967 uint32_t LowPcOffset, LowPcEndOffset;
968 std::tie(LowPcOffset, LowPcEndOffset) =
969 getAttributeOffsets(Abbrev, LowPcIdx, Offset, OrigUnit);
972 DIE.getAttributeValueAsAddress(&OrigUnit, dwarf::DW_AT_low_pc, -1ULL);
973 assert(LowPc != -1ULL && "low_pc attribute is not an address.");
974 if (LowPc == -1ULL ||
975 !hasValidRelocation(LowPcOffset, LowPcEndOffset, MyInfo))
979 DIE.dump(outs(), const_cast<DWARFUnit *>(&OrigUnit), 0, 8 /* Indent */);
983 DWARFFormValue HighPcValue;
984 if (!DIE.getAttributeValue(&OrigUnit, dwarf::DW_AT_high_pc, HighPcValue)) {
985 reportWarning("Function without high_pc. Range will be discarded.\n",
991 if (HighPcValue.isFormClass(DWARFFormValue::FC_Address)) {
992 HighPc = *HighPcValue.getAsAddress(&OrigUnit);
994 assert(HighPcValue.isFormClass(DWARFFormValue::FC_Constant));
995 HighPc = LowPc + *HighPcValue.getAsUnsignedConstant();
998 Unit.addFunctionRange(LowPc, HighPc, MyInfo.AddrAdjust);
1002 /// \brief Check if a DIE should be kept.
1003 /// \returns updated TraversalFlags.
1004 unsigned DwarfLinker::shouldKeepDIE(const DWARFDebugInfoEntryMinimal &DIE,
1006 CompileUnit::DIEInfo &MyInfo,
1008 switch (DIE.getTag()) {
1009 case dwarf::DW_TAG_constant:
1010 case dwarf::DW_TAG_variable:
1011 return shouldKeepVariableDIE(DIE, Unit, MyInfo, Flags);
1012 case dwarf::DW_TAG_subprogram:
1013 return shouldKeepSubprogramDIE(DIE, Unit, MyInfo, Flags);
1014 case dwarf::DW_TAG_module:
1015 case dwarf::DW_TAG_imported_module:
1016 case dwarf::DW_TAG_imported_declaration:
1017 case dwarf::DW_TAG_imported_unit:
1018 // We always want to keep these.
1019 return Flags | TF_Keep;
1025 /// \brief Mark the passed DIE as well as all the ones it depends on
1028 /// This function is called by lookForDIEsToKeep on DIEs that are
1029 /// newly discovered to be needed in the link. It recursively calls
1030 /// back to lookForDIEsToKeep while adding TF_DependencyWalk to the
1031 /// TraversalFlags to inform it that it's not doing the primary DIE
1033 void DwarfLinker::keepDIEAndDenpendencies(const DWARFDebugInfoEntryMinimal &DIE,
1034 CompileUnit::DIEInfo &MyInfo,
1035 const DebugMapObject &DMO,
1036 CompileUnit &CU, unsigned Flags) {
1037 const DWARFUnit &Unit = CU.getOrigUnit();
1040 // First mark all the parent chain as kept.
1041 unsigned AncestorIdx = MyInfo.ParentIdx;
1042 while (!CU.getInfo(AncestorIdx).Keep) {
1043 lookForDIEsToKeep(*Unit.getDIEAtIndex(AncestorIdx), DMO, CU,
1044 TF_ParentWalk | TF_Keep | TF_DependencyWalk);
1045 AncestorIdx = CU.getInfo(AncestorIdx).ParentIdx;
1048 // Then we need to mark all the DIEs referenced by this DIE's
1049 // attributes as kept.
1050 DataExtractor Data = Unit.getDebugInfoExtractor();
1051 const auto *Abbrev = DIE.getAbbreviationDeclarationPtr();
1052 uint32_t Offset = DIE.getOffset() + getULEB128Size(Abbrev->getCode());
1054 // Mark all DIEs referenced through atttributes as kept.
1055 for (const auto &AttrSpec : Abbrev->attributes()) {
1056 DWARFFormValue Val(AttrSpec.Form);
1058 if (!Val.isFormClass(DWARFFormValue::FC_Reference)) {
1059 DWARFFormValue::skipValue(AttrSpec.Form, Data, &Offset, &Unit);
1063 Val.extractValue(Data, &Offset, &Unit);
1064 CompileUnit *ReferencedCU;
1065 if (const auto *RefDIE = resolveDIEReference(Val, Unit, DIE, ReferencedCU))
1066 lookForDIEsToKeep(*RefDIE, DMO, *ReferencedCU,
1067 TF_Keep | TF_DependencyWalk);
1071 /// \brief Recursively walk the \p DIE tree and look for DIEs to
1072 /// keep. Store that information in \p CU's DIEInfo.
1074 /// This function is the entry point of the DIE selection
1075 /// algorithm. It is expected to walk the DIE tree in file order and
1076 /// (though the mediation of its helper) call hasValidRelocation() on
1077 /// each DIE that might be a 'root DIE' (See DwarfLinker class
1079 /// While walking the dependencies of root DIEs, this function is
1080 /// also called, but during these dependency walks the file order is
1081 /// not respected. The TF_DependencyWalk flag tells us which kind of
1082 /// traversal we are currently doing.
1083 void DwarfLinker::lookForDIEsToKeep(const DWARFDebugInfoEntryMinimal &DIE,
1084 const DebugMapObject &DMO, CompileUnit &CU,
1086 unsigned Idx = CU.getOrigUnit().getDIEIndex(&DIE);
1087 CompileUnit::DIEInfo &MyInfo = CU.getInfo(Idx);
1088 bool AlreadyKept = MyInfo.Keep;
1090 // If the Keep flag is set, we are marking a required DIE's
1091 // dependencies. If our target is already marked as kept, we're all
1093 if ((Flags & TF_DependencyWalk) && AlreadyKept)
1096 // We must not call shouldKeepDIE while called from keepDIEAndDenpendencies,
1097 // because it would screw up the relocation finding logic.
1098 if (!(Flags & TF_DependencyWalk))
1099 Flags = shouldKeepDIE(DIE, CU, MyInfo, Flags);
1101 // If it is a newly kept DIE mark it as well as all its dependencies as kept.
1102 if (!AlreadyKept && (Flags & TF_Keep))
1103 keepDIEAndDenpendencies(DIE, MyInfo, DMO, CU, Flags);
1105 // The TF_ParentWalk flag tells us that we are currently walking up
1106 // the parent chain of a required DIE, and we don't want to mark all
1107 // the children of the parents as kept (consider for example a
1108 // DW_TAG_namespace node in the parent chain). There are however a
1109 // set of DIE types for which we want to ignore that directive and still
1110 // walk their children.
1111 if (dieNeedsChildrenToBeMeaningful(DIE.getTag()))
1112 Flags &= ~TF_ParentWalk;
1114 if (!DIE.hasChildren() || (Flags & TF_ParentWalk))
1117 for (auto *Child = DIE.getFirstChild(); Child && !Child->isNULL();
1118 Child = Child->getSibling())
1119 lookForDIEsToKeep(*Child, DMO, CU, Flags);
1122 /// \brief Assign an abbreviation numer to \p Abbrev.
1124 /// Our DIEs get freed after every DebugMapObject has been processed,
1125 /// thus the FoldingSet we use to unique DIEAbbrevs cannot refer to
1126 /// the instances hold by the DIEs. When we encounter an abbreviation
1127 /// that we don't know, we create a permanent copy of it.
1128 void DwarfLinker::AssignAbbrev(DIEAbbrev &Abbrev) {
1129 // Check the set for priors.
1130 FoldingSetNodeID ID;
1133 DIEAbbrev *InSet = AbbreviationsSet.FindNodeOrInsertPos(ID, InsertToken);
1135 // If it's newly added.
1137 // Assign existing abbreviation number.
1138 Abbrev.setNumber(InSet->getNumber());
1140 // Add to abbreviation list.
1141 Abbreviations.push_back(
1142 new DIEAbbrev(Abbrev.getTag(), Abbrev.hasChildren()));
1143 for (const auto &Attr : Abbrev.getData())
1144 Abbreviations.back()->AddAttribute(Attr.getAttribute(), Attr.getForm());
1145 AbbreviationsSet.InsertNode(Abbreviations.back(), InsertToken);
1146 // Assign the unique abbreviation number.
1147 Abbrev.setNumber(Abbreviations.size());
1148 Abbreviations.back()->setNumber(Abbreviations.size());
1152 /// \brief Clone a string attribute described by \p AttrSpec and add
1154 /// \returns the size of the new attribute.
1155 unsigned DwarfLinker::cloneStringAttribute(DIE &Die, AttributeSpec AttrSpec,
1156 const DWARFFormValue &Val,
1157 const DWARFUnit &U) {
1158 // Switch everything to out of line strings.
1159 const char *String = *Val.getAsCString(&U);
1160 unsigned Offset = StringPool.getStringOffset(String);
1161 Die.addValue(dwarf::Attribute(AttrSpec.Attr), dwarf::DW_FORM_strp,
1162 new (DIEAlloc) DIEInteger(Offset));
1166 /// \brief Clone an attribute referencing another DIE and add
1168 /// \returns the size of the new attribute.
1169 unsigned DwarfLinker::cloneDieReferenceAttribute(
1170 DIE &Die, const DWARFDebugInfoEntryMinimal &InputDIE,
1171 AttributeSpec AttrSpec, unsigned AttrSize, const DWARFFormValue &Val,
1172 const DWARFUnit &U) {
1173 uint32_t Ref = *Val.getAsReference(&U);
1174 DIE *NewRefDie = nullptr;
1175 CompileUnit *RefUnit = nullptr;
1176 const DWARFDebugInfoEntryMinimal *RefDie = nullptr;
1178 if (!(RefUnit = getUnitForOffset(Ref)) ||
1179 !(RefDie = RefUnit->getOrigUnit().getDIEForOffset(Ref))) {
1180 const char *AttributeString = dwarf::AttributeString(AttrSpec.Attr);
1181 if (!AttributeString)
1182 AttributeString = "DW_AT_???";
1183 reportWarning(Twine("Missing DIE for ref in attribute ") + AttributeString +
1189 unsigned Idx = RefUnit->getOrigUnit().getDIEIndex(RefDie);
1190 CompileUnit::DIEInfo &RefInfo = RefUnit->getInfo(Idx);
1191 if (!RefInfo.Clone) {
1192 assert(Ref > InputDIE.getOffset());
1193 // We haven't cloned this DIE yet. Just create an empty one and
1194 // store it. It'll get really cloned when we process it.
1195 RefInfo.Clone = new DIE(dwarf::Tag(RefDie->getTag()));
1197 NewRefDie = RefInfo.Clone;
1199 if (AttrSpec.Form == dwarf::DW_FORM_ref_addr) {
1200 // We cannot currently rely on a DIEEntry to emit ref_addr
1201 // references, because the implementation calls back to DwarfDebug
1202 // to find the unit offset. (We don't have a DwarfDebug)
1203 // FIXME: we should be able to design DIEEntry reliance on
1206 if (Ref < InputDIE.getOffset()) {
1207 // We must have already cloned that DIE.
1208 uint32_t NewRefOffset =
1209 RefUnit->getStartOffset() + NewRefDie->getOffset();
1210 Attr = new (DIEAlloc) DIEInteger(NewRefOffset);
1212 // A forward reference. Note and fixup later.
1213 Attr = new (DIEAlloc) DIEInteger(0xBADDEF);
1214 RefUnit->noteForwardReference(NewRefDie, Attr);
1216 Die.addValue(dwarf::Attribute(AttrSpec.Attr), dwarf::DW_FORM_ref_addr,
1221 Die.addValue(dwarf::Attribute(AttrSpec.Attr), dwarf::Form(AttrSpec.Form),
1222 new (DIEAlloc) DIEEntry(*NewRefDie));
1226 /// \brief Clone an attribute of block form (locations, constants) and add
1228 /// \returns the size of the new attribute.
1229 unsigned DwarfLinker::cloneBlockAttribute(DIE &Die, AttributeSpec AttrSpec,
1230 const DWARFFormValue &Val,
1231 unsigned AttrSize) {
1234 DIELoc *Loc = nullptr;
1235 DIEBlock *Block = nullptr;
1236 // Just copy the block data over.
1237 if (AttrSpec.Form == dwarf::DW_FORM_exprloc) {
1238 Loc = new (DIEAlloc) DIELoc();
1239 DIELocs.push_back(Loc);
1241 Block = new (DIEAlloc) DIEBlock();
1242 DIEBlocks.push_back(Block);
1244 Attr = Loc ? static_cast<DIE *>(Loc) : static_cast<DIE *>(Block);
1245 Value = Loc ? static_cast<DIEValue *>(Loc) : static_cast<DIEValue *>(Block);
1246 ArrayRef<uint8_t> Bytes = *Val.getAsBlock();
1247 for (auto Byte : Bytes)
1248 Attr->addValue(static_cast<dwarf::Attribute>(0), dwarf::DW_FORM_data1,
1249 new (DIEAlloc) DIEInteger(Byte));
1250 // FIXME: If DIEBlock and DIELoc just reuses the Size field of
1251 // the DIE class, this if could be replaced by
1252 // Attr->setSize(Bytes.size()).
1255 Loc->ComputeSize(&Streamer->getAsmPrinter());
1257 Block->ComputeSize(&Streamer->getAsmPrinter());
1259 Die.addValue(dwarf::Attribute(AttrSpec.Attr), dwarf::Form(AttrSpec.Form),
1264 /// \brief Clone an address attribute and add it to \p Die.
1265 /// \returns the size of the new attribute.
1266 unsigned DwarfLinker::cloneAddressAttribute(DIE &Die, AttributeSpec AttrSpec,
1267 const DWARFFormValue &Val,
1268 const CompileUnit &Unit,
1269 AttributesInfo &Info) {
1270 uint64_t Addr = *Val.getAsAddress(&Unit.getOrigUnit());
1271 if (AttrSpec.Attr == dwarf::DW_AT_low_pc) {
1272 if (Die.getTag() == dwarf::DW_TAG_inlined_subroutine ||
1273 Die.getTag() == dwarf::DW_TAG_lexical_block)
1274 Addr += Info.PCOffset;
1275 else if (Die.getTag() == dwarf::DW_TAG_compile_unit) {
1276 Addr = Unit.getLowPc();
1277 if (Addr == UINT64_MAX)
1280 } else if (AttrSpec.Attr == dwarf::DW_AT_high_pc) {
1281 if (Die.getTag() == dwarf::DW_TAG_compile_unit) {
1282 if (uint64_t HighPc = Unit.getHighPc())
1287 // If we have a high_pc recorded for the input DIE, use
1288 // it. Otherwise (when no relocations where applied) just use the
1289 // one we just decoded.
1290 Addr = (Info.OrigHighPc ? Info.OrigHighPc : Addr) + Info.PCOffset;
1293 Die.addValue(static_cast<dwarf::Attribute>(AttrSpec.Attr),
1294 static_cast<dwarf::Form>(AttrSpec.Form),
1295 new (DIEAlloc) DIEInteger(Addr));
1296 return Unit.getOrigUnit().getAddressByteSize();
1299 /// \brief Clone a scalar attribute and add it to \p Die.
1300 /// \returns the size of the new attribute.
1301 unsigned DwarfLinker::cloneScalarAttribute(
1302 DIE &Die, const DWARFDebugInfoEntryMinimal &InputDIE,
1303 const CompileUnit &Unit, AttributeSpec AttrSpec, const DWARFFormValue &Val,
1304 unsigned AttrSize) {
1306 if (AttrSpec.Attr == dwarf::DW_AT_high_pc &&
1307 Die.getTag() == dwarf::DW_TAG_compile_unit) {
1308 if (Unit.getLowPc() == -1ULL)
1310 // Dwarf >= 4 high_pc is an size, not an address.
1311 Value = Unit.getHighPc() - Unit.getLowPc();
1312 } else if (AttrSpec.Form == dwarf::DW_FORM_sec_offset)
1313 Value = *Val.getAsSectionOffset();
1314 else if (AttrSpec.Form == dwarf::DW_FORM_sdata)
1315 Value = *Val.getAsSignedConstant();
1316 else if (auto OptionalValue = Val.getAsUnsignedConstant())
1317 Value = *OptionalValue;
1319 reportWarning("Unsupported scalar attribute form. Dropping attribute.",
1320 &Unit.getOrigUnit(), &InputDIE);
1323 Die.addValue(dwarf::Attribute(AttrSpec.Attr), dwarf::Form(AttrSpec.Form),
1324 new (DIEAlloc) DIEInteger(Value));
1328 /// \brief Clone \p InputDIE's attribute described by \p AttrSpec with
1329 /// value \p Val, and add it to \p Die.
1330 /// \returns the size of the cloned attribute.
1331 unsigned DwarfLinker::cloneAttribute(DIE &Die,
1332 const DWARFDebugInfoEntryMinimal &InputDIE,
1334 const DWARFFormValue &Val,
1335 const AttributeSpec AttrSpec,
1336 unsigned AttrSize, AttributesInfo &Info) {
1337 const DWARFUnit &U = Unit.getOrigUnit();
1339 switch (AttrSpec.Form) {
1340 case dwarf::DW_FORM_strp:
1341 case dwarf::DW_FORM_string:
1342 return cloneStringAttribute(Die, AttrSpec, Val, U);
1343 case dwarf::DW_FORM_ref_addr:
1344 case dwarf::DW_FORM_ref1:
1345 case dwarf::DW_FORM_ref2:
1346 case dwarf::DW_FORM_ref4:
1347 case dwarf::DW_FORM_ref8:
1348 return cloneDieReferenceAttribute(Die, InputDIE, AttrSpec, AttrSize, Val,
1350 case dwarf::DW_FORM_block:
1351 case dwarf::DW_FORM_block1:
1352 case dwarf::DW_FORM_block2:
1353 case dwarf::DW_FORM_block4:
1354 case dwarf::DW_FORM_exprloc:
1355 return cloneBlockAttribute(Die, AttrSpec, Val, AttrSize);
1356 case dwarf::DW_FORM_addr:
1357 return cloneAddressAttribute(Die, AttrSpec, Val, Unit, Info);
1358 case dwarf::DW_FORM_data1:
1359 case dwarf::DW_FORM_data2:
1360 case dwarf::DW_FORM_data4:
1361 case dwarf::DW_FORM_data8:
1362 case dwarf::DW_FORM_udata:
1363 case dwarf::DW_FORM_sdata:
1364 case dwarf::DW_FORM_sec_offset:
1365 case dwarf::DW_FORM_flag:
1366 case dwarf::DW_FORM_flag_present:
1367 return cloneScalarAttribute(Die, InputDIE, Unit, AttrSpec, Val, AttrSize);
1369 reportWarning("Unsupported attribute form in cloneAttribute. Dropping.", &U,
1376 /// \brief Apply the valid relocations found by findValidRelocs() to
1377 /// the buffer \p Data, taking into account that Data is at \p BaseOffset
1378 /// in the debug_info section.
1380 /// Like for findValidRelocs(), this function must be called with
1381 /// monotonic \p BaseOffset values.
1383 /// \returns wether any reloc has been applied.
1384 bool DwarfLinker::applyValidRelocs(MutableArrayRef<char> Data,
1385 uint32_t BaseOffset, bool isLittleEndian) {
1386 assert((NextValidReloc == 0 ||
1387 BaseOffset > ValidRelocs[NextValidReloc - 1].Offset) &&
1388 "BaseOffset should only be increasing.");
1389 if (NextValidReloc >= ValidRelocs.size())
1392 // Skip relocs that haven't been applied.
1393 while (NextValidReloc < ValidRelocs.size() &&
1394 ValidRelocs[NextValidReloc].Offset < BaseOffset)
1397 bool Applied = false;
1398 uint64_t EndOffset = BaseOffset + Data.size();
1399 while (NextValidReloc < ValidRelocs.size() &&
1400 ValidRelocs[NextValidReloc].Offset >= BaseOffset &&
1401 ValidRelocs[NextValidReloc].Offset < EndOffset) {
1402 const auto &ValidReloc = ValidRelocs[NextValidReloc++];
1403 assert(ValidReloc.Offset - BaseOffset < Data.size());
1404 assert(ValidReloc.Offset - BaseOffset + ValidReloc.Size <= Data.size());
1406 uint64_t Value = ValidReloc.Mapping->getValue().BinaryAddress;
1407 Value += ValidReloc.Addend;
1408 for (unsigned i = 0; i != ValidReloc.Size; ++i) {
1409 unsigned Index = isLittleEndian ? i : (ValidReloc.Size - i - 1);
1410 Buf[i] = uint8_t(Value >> (Index * 8));
1412 assert(ValidReloc.Size <= sizeof(Buf));
1413 memcpy(&Data[ValidReloc.Offset - BaseOffset], Buf, ValidReloc.Size);
1420 /// \brief Recursively clone \p InputDIE's subtrees that have been
1421 /// selected to appear in the linked output.
1423 /// \param OutOffset is the Offset where the newly created DIE will
1424 /// lie in the linked compile unit.
1426 /// \returns the cloned DIE object or null if nothing was selected.
1427 DIE *DwarfLinker::cloneDIE(const DWARFDebugInfoEntryMinimal &InputDIE,
1428 CompileUnit &Unit, int64_t PCOffset,
1429 uint32_t OutOffset) {
1430 DWARFUnit &U = Unit.getOrigUnit();
1431 unsigned Idx = U.getDIEIndex(&InputDIE);
1432 CompileUnit::DIEInfo &Info = Unit.getInfo(Idx);
1434 // Should the DIE appear in the output?
1435 if (!Unit.getInfo(Idx).Keep)
1438 uint32_t Offset = InputDIE.getOffset();
1439 // The DIE might have been already created by a forward reference
1440 // (see cloneDieReferenceAttribute()).
1441 DIE *Die = Info.Clone;
1443 Die = Info.Clone = new DIE(dwarf::Tag(InputDIE.getTag()));
1444 assert(Die->getTag() == InputDIE.getTag());
1445 Die->setOffset(OutOffset);
1447 // Extract and clone every attribute.
1448 DataExtractor Data = U.getDebugInfoExtractor();
1449 uint32_t NextOffset = U.getDIEAtIndex(Idx + 1)->getOffset();
1450 AttributesInfo AttrInfo;
1452 // We could copy the data only if we need to aply a relocation to
1453 // it. After testing, it seems there is no performance downside to
1454 // doing the copy unconditionally, and it makes the code simpler.
1455 SmallString<40> DIECopy(Data.getData().substr(Offset, NextOffset - Offset));
1456 Data = DataExtractor(DIECopy, Data.isLittleEndian(), Data.getAddressSize());
1457 // Modify the copy with relocated addresses.
1458 if (applyValidRelocs(DIECopy, Offset, Data.isLittleEndian())) {
1459 // If we applied relocations, we store the value of high_pc that was
1460 // potentially stored in the input DIE. If high_pc is an address
1461 // (Dwarf version == 2), then it might have been relocated to a
1462 // totally unrelated value (because the end address in the object
1463 // file might be start address of another function which got moved
1464 // independantly by the linker). The computation of the actual
1465 // high_pc value is done in cloneAddressAttribute().
1466 AttrInfo.OrigHighPc =
1467 InputDIE.getAttributeValueAsAddress(&U, dwarf::DW_AT_high_pc, 0);
1470 // Reset the Offset to 0 as we will be working on the local copy of
1474 const auto *Abbrev = InputDIE.getAbbreviationDeclarationPtr();
1475 Offset += getULEB128Size(Abbrev->getCode());
1477 // We are entering a subprogram. Get and propagate the PCOffset.
1478 if (Die->getTag() == dwarf::DW_TAG_subprogram)
1479 PCOffset = Info.AddrAdjust;
1480 AttrInfo.PCOffset = PCOffset;
1482 for (const auto &AttrSpec : Abbrev->attributes()) {
1483 DWARFFormValue Val(AttrSpec.Form);
1484 uint32_t AttrSize = Offset;
1485 Val.extractValue(Data, &Offset, &U);
1486 AttrSize = Offset - AttrSize;
1489 cloneAttribute(*Die, InputDIE, Unit, Val, AttrSpec, AttrSize, AttrInfo);
1492 DIEAbbrev &NewAbbrev = Die->getAbbrev();
1493 // If a scope DIE is kept, we must have kept at least one child. If
1494 // it's not the case, we'll just be emitting one wasteful end of
1495 // children marker, but things won't break.
1496 if (InputDIE.hasChildren())
1497 NewAbbrev.setChildrenFlag(dwarf::DW_CHILDREN_yes);
1498 // Assign a permanent abbrev number
1499 AssignAbbrev(Die->getAbbrev());
1501 // Add the size of the abbreviation number to the output offset.
1502 OutOffset += getULEB128Size(Die->getAbbrevNumber());
1504 if (!Abbrev->hasChildren()) {
1506 Die->setSize(OutOffset - Die->getOffset());
1510 // Recursively clone children.
1511 for (auto *Child = InputDIE.getFirstChild(); Child && !Child->isNULL();
1512 Child = Child->getSibling()) {
1513 if (DIE *Clone = cloneDIE(*Child, Unit, PCOffset, OutOffset)) {
1514 Die->addChild(std::unique_ptr<DIE>(Clone));
1515 OutOffset = Clone->getOffset() + Clone->getSize();
1519 // Account for the end of children marker.
1520 OutOffset += sizeof(int8_t);
1522 Die->setSize(OutOffset - Die->getOffset());
1526 bool DwarfLinker::link(const DebugMap &Map) {
1528 if (Map.begin() == Map.end()) {
1529 errs() << "Empty debug map.\n";
1533 if (!createStreamer(Map.getTriple(), OutputFilename))
1536 // Size of the DIEs (and headers) generated for the linked output.
1537 uint64_t OutputDebugInfoSize = 0;
1539 for (const auto &Obj : Map.objects()) {
1540 CurrentDebugObject = Obj.get();
1542 if (Options.Verbose)
1543 outs() << "DEBUG MAP OBJECT: " << Obj->getObjectFilename() << "\n";
1544 auto ErrOrObj = BinHolder.GetObjectFile(Obj->getObjectFilename());
1545 if (std::error_code EC = ErrOrObj.getError()) {
1546 reportWarning(Twine(Obj->getObjectFilename()) + ": " + EC.message());
1550 // Look for relocations that correspond to debug map entries.
1551 if (!findValidRelocsInDebugInfo(*ErrOrObj, *Obj)) {
1552 if (Options.Verbose)
1553 outs() << "No valid relocations found. Skipping.\n";
1557 // Setup access to the debug info.
1558 DWARFContextInMemory DwarfContext(*ErrOrObj);
1559 startDebugObject(DwarfContext);
1561 // In a first phase, just read in the debug info and store the DIE
1562 // parent links that we will use during the next phase.
1563 for (const auto &CU : DwarfContext.compile_units()) {
1564 auto *CUDie = CU->getCompileUnitDIE(false);
1565 if (Options.Verbose) {
1566 outs() << "Input compilation unit:";
1567 CUDie->dump(outs(), CU.get(), 0);
1569 Units.emplace_back(*CU);
1570 gatherDIEParents(CUDie, 0, Units.back());
1573 // Then mark all the DIEs that need to be present in the linked
1574 // output and collect some information about them. Note that this
1575 // loop can not be merged with the previous one becaue cross-cu
1576 // references require the ParentIdx to be setup for every CU in
1577 // the object file before calling this.
1578 for (auto &CurrentUnit : Units)
1579 lookForDIEsToKeep(*CurrentUnit.getOrigUnit().getCompileUnitDIE(), *Obj,
1582 // The calls to applyValidRelocs inside cloneDIE will walk the
1583 // reloc array again (in the same way findValidRelocsInDebugInfo()
1584 // did). We need to reset the NextValidReloc index to the beginning.
1587 // Construct the output DIE tree by cloning the DIEs we chose to
1588 // keep above. If there are no valid relocs, then there's nothing
1590 if (!ValidRelocs.empty())
1591 for (auto &CurrentUnit : Units) {
1592 const auto *InputDIE = CurrentUnit.getOrigUnit().getCompileUnitDIE();
1593 CurrentUnit.setStartOffset(OutputDebugInfoSize);
1594 DIE *OutputDIE = cloneDIE(*InputDIE, CurrentUnit, 0 /* PCOffset */,
1595 11 /* Unit Header size */);
1596 CurrentUnit.setOutputUnitDIE(OutputDIE);
1597 OutputDebugInfoSize = CurrentUnit.computeNextUnitOffset();
1600 // Emit all the compile unit's debug information.
1601 if (!ValidRelocs.empty() && !Options.NoOutput)
1602 for (auto &CurrentUnit : Units) {
1603 CurrentUnit.fixupForwardReferences();
1604 Streamer->emitCompileUnitHeader(CurrentUnit);
1605 if (!CurrentUnit.getOutputUnitDIE())
1607 Streamer->emitDIE(*CurrentUnit.getOutputUnitDIE());
1610 // Clean-up before starting working on the next object.
1614 // Emit everything that's global.
1615 if (!Options.NoOutput) {
1616 Streamer->emitAbbrevs(Abbreviations);
1617 Streamer->emitStrings(StringPool);
1620 return Options.NoOutput ? true : Streamer->finish();
1624 bool linkDwarf(StringRef OutputFilename, const DebugMap &DM,
1625 const LinkOptions &Options) {
1626 DwarfLinker Linker(OutputFilename, Options);
1627 return Linker.link(DM);