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/StringMap.h"
14 #include "llvm/ADT/STLExtras.h"
15 #include "llvm/CodeGen/AsmPrinter.h"
16 #include "llvm/CodeGen/DIE.h"
17 #include "llvm/DebugInfo/DWARF/DWARFContext.h"
18 #include "llvm/DebugInfo/DWARF/DWARFDebugInfoEntry.h"
19 #include "llvm/DebugInfo/DWARF/DWARFFormValue.h"
20 #include "llvm/MC/MCAsmBackend.h"
21 #include "llvm/MC/MCAsmInfo.h"
22 #include "llvm/MC/MCContext.h"
23 #include "llvm/MC/MCCodeEmitter.h"
24 #include "llvm/MC/MCInstrInfo.h"
25 #include "llvm/MC/MCObjectFileInfo.h"
26 #include "llvm/MC/MCRegisterInfo.h"
27 #include "llvm/MC/MCStreamer.h"
28 #include "llvm/Object/MachO.h"
29 #include "llvm/Support/Dwarf.h"
30 #include "llvm/Support/LEB128.h"
31 #include "llvm/Support/TargetRegistry.h"
32 #include "llvm/Target/TargetMachine.h"
33 #include "llvm/Target/TargetOptions.h"
41 void warn(const Twine &Warning, const Twine &Context) {
42 errs() << Twine("while processing ") + Context + ":\n";
43 errs() << Twine("warning: ") + Warning + "\n";
46 bool error(const Twine &Error, const Twine &Context) {
47 errs() << Twine("while processing ") + Context + ":\n";
48 errs() << Twine("error: ") + Error + "\n";
52 /// \brief Stores all information relating to a compile unit, be it in
53 /// its original instance in the object file to its brand new cloned
54 /// and linked DIE tree.
57 /// \brief Information gathered about a DIE in the object file.
59 int64_t AddrAdjust; ///< Address offset to apply to the described entity.
60 DIE *Clone; ///< Cloned version of that DIE.
61 uint32_t ParentIdx; ///< The index of this DIE's parent.
62 bool Keep; ///< Is the DIE part of the linked output?
63 bool InDebugMap; ///< Was this DIE's entity found in the map?
66 CompileUnit(DWARFUnit &OrigUnit) : OrigUnit(OrigUnit) {
67 Info.resize(OrigUnit.getNumDIEs());
70 // Workaround MSVC not supporting implicit move ops
71 CompileUnit(CompileUnit &&RHS)
72 : OrigUnit(RHS.OrigUnit), Info(std::move(RHS.Info)),
73 CUDie(std::move(RHS.CUDie)), StartOffset(RHS.StartOffset),
74 NextUnitOffset(RHS.NextUnitOffset) {}
76 DWARFUnit &getOrigUnit() const { return OrigUnit; }
78 DIE *getOutputUnitDIE() const { return CUDie.get(); }
79 void setOutputUnitDIE(DIE *Die) { CUDie.reset(Die); }
81 DIEInfo &getInfo(unsigned Idx) { return Info[Idx]; }
82 const DIEInfo &getInfo(unsigned Idx) const { return Info[Idx]; }
84 uint64_t getStartOffset() const { return StartOffset; }
85 uint64_t getNextUnitOffset() const { return NextUnitOffset; }
87 void setStartOffset(uint64_t DebugInfoSize) { StartOffset = DebugInfoSize; }
89 /// \brief Compute the end offset for this unit. Must be
90 /// called after the CU's DIEs have been cloned.
91 /// \returns the next unit offset (which is also the current
92 /// debug_info section size).
93 uint64_t computeNextUnitOffset();
95 /// \brief Keep track of a forward reference to DIE \p Die by
96 /// \p Attr. The attribute should be fixed up later to point to the
97 /// absolute offset of \p Die in the debug_info section.
98 void noteForwardReference(DIE *Die, DIEInteger *Attr);
100 /// \brief Apply all fixups recored by noteForwardReference().
101 void fixupForwardReferences();
105 std::vector<DIEInfo> Info; ///< DIE info indexed by DIE index.
106 std::unique_ptr<DIE> CUDie; ///< Root of the linked DIE tree.
108 uint64_t StartOffset;
109 uint64_t NextUnitOffset;
111 /// \brief A list of attributes to fixup with the absolute offset of
112 /// a DIE in the debug_info section.
114 /// The offsets for the attributes in this array couldn't be set while
115 /// cloning because for forward refences the target DIE's offset isn't
116 /// known you emit the reference attribute.
117 std::vector<std::pair<DIE *, DIEInteger *>> ForwardDIEReferences;
120 uint64_t CompileUnit::computeNextUnitOffset() {
121 NextUnitOffset = StartOffset + 11 /* Header size */;
122 // The root DIE might be null, meaning that the Unit had nothing to
123 // contribute to the linked output. In that case, we will emit the
124 // unit header without any actual DIE.
126 NextUnitOffset += CUDie->getSize();
127 return NextUnitOffset;
130 /// \brief Keep track of a forward reference to \p Die.
131 void CompileUnit::noteForwardReference(DIE *Die, DIEInteger *Attr) {
132 ForwardDIEReferences.emplace_back(Die, Attr);
135 /// \brief Apply all fixups recorded by noteForwardReference().
136 void CompileUnit::fixupForwardReferences() {
137 for (const auto &Ref : ForwardDIEReferences)
138 Ref.second->setValue(Ref.first->getOffset() + getStartOffset());
141 /// \brief A string table that doesn't need relocations.
143 /// We are doing a final link, no need for a string table that
144 /// has relocation entries for every reference to it. This class
145 /// provides this ablitity by just associating offsets with
147 class NonRelocatableStringpool {
149 /// \brief Entries are stored into the StringMap and simply linked
150 /// together through the second element of this pair in order to
151 /// keep track of insertion order.
152 typedef StringMap<std::pair<uint32_t, StringMapEntryBase *>, BumpPtrAllocator>
155 NonRelocatableStringpool()
156 : CurrentEndOffset(0), Sentinel(0), Last(&Sentinel) {
157 // Legacy dsymutil puts an empty string at the start of the line
162 /// \brief Get the offset of string \p S in the string table. This
163 /// can insert a new element or return the offset of a preexisitng
165 uint32_t getStringOffset(StringRef S);
167 /// \brief Get permanent storage for \p S (but do not necessarily
168 /// emit \p S in the output section).
169 /// \returns The StringRef that points to permanent storage to use
170 /// in place of \p S.
171 StringRef internString(StringRef S);
173 // \brief Return the first entry of the string table.
174 const MapTy::MapEntryTy *getFirstEntry() const {
175 return getNextEntry(&Sentinel);
178 // \brief Get the entry following \p E in the string table or null
179 // if \p E was the last entry.
180 const MapTy::MapEntryTy *getNextEntry(const MapTy::MapEntryTy *E) const {
181 return static_cast<const MapTy::MapEntryTy *>(E->getValue().second);
184 uint64_t getSize() { return CurrentEndOffset; }
188 uint32_t CurrentEndOffset;
189 MapTy::MapEntryTy Sentinel, *Last;
192 /// \brief Get the offset of string \p S in the string table. This
193 /// can insert a new element or return the offset of a preexisitng
195 uint32_t NonRelocatableStringpool::getStringOffset(StringRef S) {
196 if (S.empty() && !Strings.empty())
199 std::pair<uint32_t, StringMapEntryBase *> Entry(0, nullptr);
203 // A non-empty string can't be at offset 0, so if we have an entry
204 // with a 0 offset, it must be a previously interned string.
205 std::tie(It, Inserted) = Strings.insert(std::make_pair(S, Entry));
206 if (Inserted || It->getValue().first == 0) {
207 // Set offset and chain at the end of the entries list.
208 It->getValue().first = CurrentEndOffset;
209 CurrentEndOffset += S.size() + 1; // +1 for the '\0'.
210 Last->getValue().second = &*It;
213 return It->getValue().first;
216 /// \brief Put \p S into the StringMap so that it gets permanent
217 /// storage, but do not actually link it in the chain of elements
218 /// that go into the output section. A latter call to
219 /// getStringOffset() with the same string will chain it though.
220 StringRef NonRelocatableStringpool::internString(StringRef S) {
221 std::pair<uint32_t, StringMapEntryBase *> Entry(0, nullptr);
222 auto InsertResult = Strings.insert(std::make_pair(S, Entry));
223 return InsertResult.first->getKey();
226 /// \brief The Dwarf streaming logic
228 /// All interactions with the MC layer that is used to build the debug
229 /// information binary representation are handled in this class.
230 class DwarfStreamer {
231 /// \defgroup MCObjects MC layer objects constructed by the streamer
233 std::unique_ptr<MCRegisterInfo> MRI;
234 std::unique_ptr<MCAsmInfo> MAI;
235 std::unique_ptr<MCObjectFileInfo> MOFI;
236 std::unique_ptr<MCContext> MC;
237 MCAsmBackend *MAB; // Owned by MCStreamer
238 std::unique_ptr<MCInstrInfo> MII;
239 std::unique_ptr<MCSubtargetInfo> MSTI;
240 MCCodeEmitter *MCE; // Owned by MCStreamer
241 MCStreamer *MS; // Owned by AsmPrinter
242 std::unique_ptr<TargetMachine> TM;
243 std::unique_ptr<AsmPrinter> Asm;
246 /// \brief the file we stream the linked Dwarf to.
247 std::unique_ptr<raw_fd_ostream> OutFile;
250 /// \brief Actually create the streamer and the ouptut file.
252 /// This could be done directly in the constructor, but it feels
253 /// more natural to handle errors through return value.
254 bool init(Triple TheTriple, StringRef OutputFilename);
256 /// \brief Dump the file to the disk.
259 AsmPrinter &getAsmPrinter() const { return *Asm; }
261 /// \brief Set the current output section to debug_info and change
262 /// the MC Dwarf version to \p DwarfVersion.
263 void switchToDebugInfoSection(unsigned DwarfVersion);
265 /// \brief Emit the compilation unit header for \p Unit in the
266 /// debug_info section.
268 /// As a side effect, this also switches the current Dwarf version
269 /// of the MC layer to the one of U.getOrigUnit().
270 void emitCompileUnitHeader(CompileUnit &Unit);
272 /// \brief Recursively emit the DIE tree rooted at \p Die.
273 void emitDIE(DIE &Die);
275 /// \brief Emit the abbreviation table \p Abbrevs to the
276 /// debug_abbrev section.
277 void emitAbbrevs(const std::vector<DIEAbbrev *> &Abbrevs);
279 /// \brief Emit the string table described by \p Pool.
280 void emitStrings(const NonRelocatableStringpool &Pool);
283 bool DwarfStreamer::init(Triple TheTriple, StringRef OutputFilename) {
284 std::string ErrorStr;
285 std::string TripleName;
286 StringRef Context = "dwarf streamer init";
289 const Target *TheTarget =
290 TargetRegistry::lookupTarget(TripleName, TheTriple, ErrorStr);
292 return error(ErrorStr, Context);
293 TripleName = TheTriple.getTriple();
295 // Create all the MC Objects.
296 MRI.reset(TheTarget->createMCRegInfo(TripleName));
298 return error(Twine("no register info for target ") + TripleName, Context);
300 MAI.reset(TheTarget->createMCAsmInfo(*MRI, TripleName));
302 return error("no asm info for target " + TripleName, Context);
304 MOFI.reset(new MCObjectFileInfo);
305 MC.reset(new MCContext(MAI.get(), MRI.get(), MOFI.get()));
306 MOFI->InitMCObjectFileInfo(TripleName, Reloc::Default, CodeModel::Default,
309 MAB = TheTarget->createMCAsmBackend(*MRI, TripleName, "");
311 return error("no asm backend for target " + TripleName, Context);
313 MII.reset(TheTarget->createMCInstrInfo());
315 return error("no instr info info for target " + TripleName, Context);
317 MSTI.reset(TheTarget->createMCSubtargetInfo(TripleName, "", ""));
319 return error("no subtarget info for target " + TripleName, Context);
321 MCE = TheTarget->createMCCodeEmitter(*MII, *MRI, *MC);
323 return error("no code emitter for target " + TripleName, Context);
325 // Create the output file.
328 llvm::make_unique<raw_fd_ostream>(OutputFilename, EC, sys::fs::F_None);
330 return error(Twine(OutputFilename) + ": " + EC.message(), Context);
332 MS = TheTarget->createMCObjectStreamer(TripleName, *MC, *MAB, *OutFile, MCE,
335 return error("no object streamer for target " + TripleName, Context);
337 // Finally create the AsmPrinter we'll use to emit the DIEs.
338 TM.reset(TheTarget->createTargetMachine(TripleName, "", "", TargetOptions()));
340 return error("no target machine for target " + TripleName, Context);
342 Asm.reset(TheTarget->createAsmPrinter(*TM, std::unique_ptr<MCStreamer>(MS)));
344 return error("no asm printer for target " + TripleName, Context);
349 bool DwarfStreamer::finish() {
354 /// \brief Set the current output section to debug_info and change
355 /// the MC Dwarf version to \p DwarfVersion.
356 void DwarfStreamer::switchToDebugInfoSection(unsigned DwarfVersion) {
357 MS->SwitchSection(MOFI->getDwarfInfoSection());
358 MC->setDwarfVersion(DwarfVersion);
361 /// \brief Emit the compilation unit header for \p Unit in the
362 /// debug_info section.
364 /// A Dwarf scetion header is encoded as:
365 /// uint32_t Unit length (omiting this field)
367 /// uint32_t Abbreviation table offset
368 /// uint8_t Address size
370 /// Leading to a total of 11 bytes.
371 void DwarfStreamer::emitCompileUnitHeader(CompileUnit &Unit) {
372 unsigned Version = Unit.getOrigUnit().getVersion();
373 switchToDebugInfoSection(Version);
375 // Emit size of content not including length itself. The size has
376 // already been computed in CompileUnit::computeOffsets(). Substract
377 // 4 to that size to account for the length field.
378 Asm->EmitInt32(Unit.getNextUnitOffset() - Unit.getStartOffset() - 4);
379 Asm->EmitInt16(Version);
380 // We share one abbreviations table across all units so it's always at the
381 // start of the section.
383 Asm->EmitInt8(Unit.getOrigUnit().getAddressByteSize());
386 /// \brief Emit the \p Abbrevs array as the shared abbreviation table
387 /// for the linked Dwarf file.
388 void DwarfStreamer::emitAbbrevs(const std::vector<DIEAbbrev *> &Abbrevs) {
389 MS->SwitchSection(MOFI->getDwarfAbbrevSection());
390 Asm->emitDwarfAbbrevs(Abbrevs);
393 /// \brief Recursively emit the DIE tree rooted at \p Die.
394 void DwarfStreamer::emitDIE(DIE &Die) {
395 MS->SwitchSection(MOFI->getDwarfInfoSection());
396 Asm->emitDwarfDIE(Die);
399 /// \brief Emit the debug_str section stored in \p Pool.
400 void DwarfStreamer::emitStrings(const NonRelocatableStringpool &Pool) {
401 Asm->OutStreamer.SwitchSection(MOFI->getDwarfStrSection());
402 for (auto *Entry = Pool.getFirstEntry(); Entry;
403 Entry = Pool.getNextEntry(Entry))
404 Asm->OutStreamer.EmitBytes(
405 StringRef(Entry->getKey().data(), Entry->getKey().size() + 1));
408 /// \brief The core of the Dwarf linking logic.
410 /// The link of the dwarf information from the object files will be
411 /// driven by the selection of 'root DIEs', which are DIEs that
412 /// describe variables or functions that are present in the linked
413 /// binary (and thus have entries in the debug map). All the debug
414 /// information that will be linked (the DIEs, but also the line
415 /// tables, ranges, ...) is derived from that set of root DIEs.
417 /// The root DIEs are identified because they contain relocations that
418 /// correspond to a debug map entry at specific places (the low_pc for
419 /// a function, the location for a variable). These relocations are
420 /// called ValidRelocs in the DwarfLinker and are gathered as a very
421 /// first step when we start processing a DebugMapObject.
424 DwarfLinker(StringRef OutputFilename, const LinkOptions &Options)
425 : OutputFilename(OutputFilename), Options(Options),
426 BinHolder(Options.Verbose) {}
429 for (auto *Abbrev : Abbreviations)
433 /// \brief Link the contents of the DebugMap.
434 bool link(const DebugMap &);
437 /// \brief Called at the start of a debug object link.
438 void startDebugObject(DWARFContext &);
440 /// \brief Called at the end of a debug object link.
441 void endDebugObject();
443 /// \defgroup FindValidRelocations Translate debug map into a list
444 /// of relevant relocations
451 const DebugMapObject::DebugMapEntry *Mapping;
453 ValidReloc(uint32_t Offset, uint32_t Size, uint64_t Addend,
454 const DebugMapObject::DebugMapEntry *Mapping)
455 : Offset(Offset), Size(Size), Addend(Addend), Mapping(Mapping) {}
457 bool operator<(const ValidReloc &RHS) const { return Offset < RHS.Offset; }
460 /// \brief The valid relocations for the current DebugMapObject.
461 /// This vector is sorted by relocation offset.
462 std::vector<ValidReloc> ValidRelocs;
464 /// \brief Index into ValidRelocs of the next relocation to
465 /// consider. As we walk the DIEs in acsending file offset and as
466 /// ValidRelocs is sorted by file offset, keeping this index
467 /// uptodate is all we have to do to have a cheap lookup during the
468 /// root DIE selection and during DIE cloning.
469 unsigned NextValidReloc;
471 bool findValidRelocsInDebugInfo(const object::ObjectFile &Obj,
472 const DebugMapObject &DMO);
474 bool findValidRelocs(const object::SectionRef &Section,
475 const object::ObjectFile &Obj,
476 const DebugMapObject &DMO);
478 void findValidRelocsMachO(const object::SectionRef &Section,
479 const object::MachOObjectFile &Obj,
480 const DebugMapObject &DMO);
483 /// \defgroup FindRootDIEs Find DIEs corresponding to debug map entries.
486 /// \brief Recursively walk the \p DIE tree and look for DIEs to
487 /// keep. Store that information in \p CU's DIEInfo.
488 void lookForDIEsToKeep(const DWARFDebugInfoEntryMinimal &DIE,
489 const DebugMapObject &DMO, CompileUnit &CU,
492 /// \brief Flags passed to DwarfLinker::lookForDIEsToKeep
494 TF_Keep = 1 << 0, ///< Mark the traversed DIEs as kept.
495 TF_InFunctionScope = 1 << 1, ///< Current scope is a fucntion scope.
496 TF_DependencyWalk = 1 << 2, ///< Walking the dependencies of a kept DIE.
497 TF_ParentWalk = 1 << 3, ///< Walking up the parents of a kept DIE.
500 /// \brief Mark the passed DIE as well as all the ones it depends on
502 void keepDIEAndDenpendencies(const DWARFDebugInfoEntryMinimal &DIE,
503 CompileUnit::DIEInfo &MyInfo,
504 const DebugMapObject &DMO, CompileUnit &CU,
507 unsigned shouldKeepDIE(const DWARFDebugInfoEntryMinimal &DIE,
508 CompileUnit &Unit, CompileUnit::DIEInfo &MyInfo,
511 unsigned shouldKeepVariableDIE(const DWARFDebugInfoEntryMinimal &DIE,
513 CompileUnit::DIEInfo &MyInfo, unsigned Flags);
515 unsigned shouldKeepSubprogramDIE(const DWARFDebugInfoEntryMinimal &DIE,
517 CompileUnit::DIEInfo &MyInfo,
520 bool hasValidRelocation(uint32_t StartOffset, uint32_t EndOffset,
521 CompileUnit::DIEInfo &Info);
524 /// \defgroup Linking Methods used to link the debug information
527 /// \brief Recursively clone \p InputDIE into an tree of DIE objects
528 /// where useless (as decided by lookForDIEsToKeep()) bits have been
529 /// stripped out and addresses have been rewritten according to the
532 /// \param OutOffset is the offset the cloned DIE in the output
534 /// \param PCOffset (while cloning a function scope) is the offset
535 /// applied to the entry point of the function to get the linked address.
537 /// \returns the root of the cloned tree.
538 DIE *cloneDIE(const DWARFDebugInfoEntryMinimal &InputDIE, CompileUnit &U,
539 int64_t PCOffset, uint32_t OutOffset);
541 typedef DWARFAbbreviationDeclaration::AttributeSpec AttributeSpec;
543 /// \brief Information gathered and exchanged between the various
544 /// clone*Attributes helpers about the attributes of a particular DIE.
545 struct AttributesInfo {
546 uint64_t OrigHighPc; ///< Value of AT_high_pc in the input DIE
547 int64_t PCOffset; ///< Offset to apply to PC addresses inside a function.
549 AttributesInfo() : OrigHighPc(0), PCOffset(0) {}
552 /// \brief Helper for cloneDIE.
553 unsigned cloneAttribute(DIE &Die, const DWARFDebugInfoEntryMinimal &InputDIE,
554 CompileUnit &U, const DWARFFormValue &Val,
555 const AttributeSpec AttrSpec, unsigned AttrSize,
556 AttributesInfo &AttrInfo);
558 /// \brief Helper for cloneDIE.
559 unsigned cloneStringAttribute(DIE &Die, AttributeSpec AttrSpec,
560 const DWARFFormValue &Val, const DWARFUnit &U);
562 /// \brief Helper for cloneDIE.
564 cloneDieReferenceAttribute(DIE &Die,
565 const DWARFDebugInfoEntryMinimal &InputDIE,
566 AttributeSpec AttrSpec, unsigned AttrSize,
567 const DWARFFormValue &Val, const DWARFUnit &U);
569 /// \brief Helper for cloneDIE.
570 unsigned cloneBlockAttribute(DIE &Die, AttributeSpec AttrSpec,
571 const DWARFFormValue &Val, unsigned AttrSize);
573 /// \brief Helper for cloneDIE.
574 unsigned cloneAddressAttribute(DIE &Die, AttributeSpec AttrSpec,
575 const DWARFFormValue &Val,
576 const CompileUnit &Unit, AttributesInfo &Info);
578 /// \brief Helper for cloneDIE.
579 unsigned cloneScalarAttribute(DIE &Die,
580 const DWARFDebugInfoEntryMinimal &InputDIE,
581 const DWARFUnit &U, AttributeSpec AttrSpec,
582 const DWARFFormValue &Val, unsigned AttrSize);
584 /// \brief Helper for cloneDIE.
585 bool applyValidRelocs(MutableArrayRef<char> Data, uint32_t BaseOffset,
586 bool isLittleEndian);
588 /// \brief Assign an abbreviation number to \p Abbrev
589 void AssignAbbrev(DIEAbbrev &Abbrev);
591 /// \brief FoldingSet that uniques the abbreviations.
592 FoldingSet<DIEAbbrev> AbbreviationsSet;
593 /// \brief Storage for the unique Abbreviations.
594 /// This is passed to AsmPrinter::emitDwarfAbbrevs(), thus it cannot
595 /// be changed to a vecot of unique_ptrs.
596 std::vector<DIEAbbrev *> Abbreviations;
598 /// \brief DIELoc objects that need to be destructed (but not freed!).
599 std::vector<DIELoc *> DIELocs;
600 /// \brief DIEBlock objects that need to be destructed (but not freed!).
601 std::vector<DIEBlock *> DIEBlocks;
602 /// \brief Allocator used for all the DIEValue objects.
603 BumpPtrAllocator DIEAlloc;
606 /// \defgroup Helpers Various helper methods.
609 const DWARFDebugInfoEntryMinimal *
610 resolveDIEReference(DWARFFormValue &RefValue, const DWARFUnit &Unit,
611 const DWARFDebugInfoEntryMinimal &DIE,
612 CompileUnit *&ReferencedCU);
614 CompileUnit *getUnitForOffset(unsigned Offset);
616 void reportWarning(const Twine &Warning, const DWARFUnit *Unit = nullptr,
617 const DWARFDebugInfoEntryMinimal *DIE = nullptr);
619 bool createStreamer(Triple TheTriple, StringRef OutputFilename);
623 std::string OutputFilename;
625 BinaryHolder BinHolder;
626 std::unique_ptr<DwarfStreamer> Streamer;
628 /// The units of the current debug map object.
629 std::vector<CompileUnit> Units;
631 /// The debug map object curently under consideration.
632 DebugMapObject *CurrentDebugObject;
634 /// \brief The Dwarf string pool
635 NonRelocatableStringpool StringPool;
638 /// \brief Similar to DWARFUnitSection::getUnitForOffset(), but
639 /// returning our CompileUnit object instead.
640 CompileUnit *DwarfLinker::getUnitForOffset(unsigned Offset) {
642 std::upper_bound(Units.begin(), Units.end(), Offset,
643 [](uint32_t LHS, const CompileUnit &RHS) {
644 return LHS < RHS.getOrigUnit().getNextUnitOffset();
646 return CU != Units.end() ? &*CU : nullptr;
649 /// \brief Resolve the DIE attribute reference that has been
650 /// extracted in \p RefValue. The resulting DIE migh be in another
651 /// CompileUnit which is stored into \p ReferencedCU.
652 /// \returns null if resolving fails for any reason.
653 const DWARFDebugInfoEntryMinimal *DwarfLinker::resolveDIEReference(
654 DWARFFormValue &RefValue, const DWARFUnit &Unit,
655 const DWARFDebugInfoEntryMinimal &DIE, CompileUnit *&RefCU) {
656 assert(RefValue.isFormClass(DWARFFormValue::FC_Reference));
657 uint64_t RefOffset = *RefValue.getAsReference(&Unit);
659 if ((RefCU = getUnitForOffset(RefOffset)))
660 if (const auto *RefDie = RefCU->getOrigUnit().getDIEForOffset(RefOffset))
663 reportWarning("could not find referenced DIE", &Unit, &DIE);
667 /// \brief Report a warning to the user, optionaly including
668 /// information about a specific \p DIE related to the warning.
669 void DwarfLinker::reportWarning(const Twine &Warning, const DWARFUnit *Unit,
670 const DWARFDebugInfoEntryMinimal *DIE) {
671 StringRef Context = "<debug map>";
672 if (CurrentDebugObject)
673 Context = CurrentDebugObject->getObjectFilename();
674 warn(Warning, Context);
676 if (!Options.Verbose || !DIE)
679 errs() << " in DIE:\n";
680 DIE->dump(errs(), const_cast<DWARFUnit *>(Unit), 0 /* RecurseDepth */,
684 bool DwarfLinker::createStreamer(Triple TheTriple, StringRef OutputFilename) {
685 if (Options.NoOutput)
688 Streamer = llvm::make_unique<DwarfStreamer>();
689 return Streamer->init(TheTriple, OutputFilename);
692 /// \brief Recursive helper to gather the child->parent relationships in the
693 /// original compile unit.
694 static void gatherDIEParents(const DWARFDebugInfoEntryMinimal *DIE,
695 unsigned ParentIdx, CompileUnit &CU) {
696 unsigned MyIdx = CU.getOrigUnit().getDIEIndex(DIE);
697 CU.getInfo(MyIdx).ParentIdx = ParentIdx;
699 if (DIE->hasChildren())
700 for (auto *Child = DIE->getFirstChild(); Child && !Child->isNULL();
701 Child = Child->getSibling())
702 gatherDIEParents(Child, MyIdx, CU);
705 static bool dieNeedsChildrenToBeMeaningful(uint32_t Tag) {
709 case dwarf::DW_TAG_subprogram:
710 case dwarf::DW_TAG_lexical_block:
711 case dwarf::DW_TAG_subroutine_type:
712 case dwarf::DW_TAG_structure_type:
713 case dwarf::DW_TAG_class_type:
714 case dwarf::DW_TAG_union_type:
717 llvm_unreachable("Invalid Tag");
720 void DwarfLinker::startDebugObject(DWARFContext &Dwarf) {
721 Units.reserve(Dwarf.getNumCompileUnits());
725 void DwarfLinker::endDebugObject() {
729 for (auto *Block : DIEBlocks)
731 for (auto *Loc : DIELocs)
739 /// \brief Iterate over the relocations of the given \p Section and
740 /// store the ones that correspond to debug map entries into the
741 /// ValidRelocs array.
742 void DwarfLinker::findValidRelocsMachO(const object::SectionRef &Section,
743 const object::MachOObjectFile &Obj,
744 const DebugMapObject &DMO) {
746 Section.getContents(Contents);
747 DataExtractor Data(Contents, Obj.isLittleEndian(), 0);
749 for (const object::RelocationRef &Reloc : Section.relocations()) {
750 object::DataRefImpl RelocDataRef = Reloc.getRawDataRefImpl();
751 MachO::any_relocation_info MachOReloc = Obj.getRelocation(RelocDataRef);
752 unsigned RelocSize = 1 << Obj.getAnyRelocationLength(MachOReloc);
754 if ((RelocSize != 4 && RelocSize != 8) || Reloc.getOffset(Offset64)) {
755 reportWarning(" unsupported relocation in debug_info section.");
758 uint32_t Offset = Offset64;
759 // Mach-o uses REL relocations, the addend is at the relocation offset.
760 uint64_t Addend = Data.getUnsigned(&Offset, RelocSize);
762 auto Sym = Reloc.getSymbol();
763 if (Sym != Obj.symbol_end()) {
764 StringRef SymbolName;
765 if (Sym->getName(SymbolName)) {
766 reportWarning("error getting relocation symbol name.");
769 if (const auto *Mapping = DMO.lookupSymbol(SymbolName))
770 ValidRelocs.emplace_back(Offset64, RelocSize, Addend, Mapping);
771 } else if (const auto *Mapping = DMO.lookupObjectAddress(Addend)) {
772 // Do not store the addend. The addend was the address of the
773 // symbol in the object file, the address in the binary that is
774 // stored in the debug map doesn't need to be offseted.
775 ValidRelocs.emplace_back(Offset64, RelocSize, 0, Mapping);
780 /// \brief Dispatch the valid relocation finding logic to the
781 /// appropriate handler depending on the object file format.
782 bool DwarfLinker::findValidRelocs(const object::SectionRef &Section,
783 const object::ObjectFile &Obj,
784 const DebugMapObject &DMO) {
785 // Dispatch to the right handler depending on the file type.
786 if (auto *MachOObj = dyn_cast<object::MachOObjectFile>(&Obj))
787 findValidRelocsMachO(Section, *MachOObj, DMO);
789 reportWarning(Twine("unsupported object file type: ") + Obj.getFileName());
791 if (ValidRelocs.empty())
794 // Sort the relocations by offset. We will walk the DIEs linearly in
795 // the file, this allows us to just keep an index in the relocation
796 // array that we advance during our walk, rather than resorting to
797 // some associative container. See DwarfLinker::NextValidReloc.
798 std::sort(ValidRelocs.begin(), ValidRelocs.end());
802 /// \brief Look for relocations in the debug_info section that match
803 /// entries in the debug map. These relocations will drive the Dwarf
804 /// link by indicating which DIEs refer to symbols present in the
806 /// \returns wether there are any valid relocations in the debug info.
807 bool DwarfLinker::findValidRelocsInDebugInfo(const object::ObjectFile &Obj,
808 const DebugMapObject &DMO) {
809 // Find the debug_info section.
810 for (const object::SectionRef &Section : Obj.sections()) {
811 StringRef SectionName;
812 Section.getName(SectionName);
813 SectionName = SectionName.substr(SectionName.find_first_not_of("._"));
814 if (SectionName != "debug_info")
816 return findValidRelocs(Section, Obj, DMO);
821 /// \brief Checks that there is a relocation against an actual debug
822 /// map entry between \p StartOffset and \p NextOffset.
824 /// This function must be called with offsets in strictly ascending
825 /// order because it never looks back at relocations it already 'went past'.
826 /// \returns true and sets Info.InDebugMap if it is the case.
827 bool DwarfLinker::hasValidRelocation(uint32_t StartOffset, uint32_t EndOffset,
828 CompileUnit::DIEInfo &Info) {
829 assert(NextValidReloc == 0 ||
830 StartOffset > ValidRelocs[NextValidReloc - 1].Offset);
831 if (NextValidReloc >= ValidRelocs.size())
834 uint64_t RelocOffset = ValidRelocs[NextValidReloc].Offset;
836 // We might need to skip some relocs that we didn't consider. For
837 // example the high_pc of a discarded DIE might contain a reloc that
838 // is in the list because it actually corresponds to the start of a
839 // function that is in the debug map.
840 while (RelocOffset < StartOffset && NextValidReloc < ValidRelocs.size() - 1)
841 RelocOffset = ValidRelocs[++NextValidReloc].Offset;
843 if (RelocOffset < StartOffset || RelocOffset >= EndOffset)
846 const auto &ValidReloc = ValidRelocs[NextValidReloc++];
848 outs() << "Found valid debug map entry: " << ValidReloc.Mapping->getKey()
849 << " " << format("\t%016" PRIx64 " => %016" PRIx64,
850 ValidReloc.Mapping->getValue().ObjectAddress,
851 ValidReloc.Mapping->getValue().BinaryAddress);
853 Info.AddrAdjust = int64_t(ValidReloc.Mapping->getValue().BinaryAddress) +
855 ValidReloc.Mapping->getValue().ObjectAddress;
856 Info.InDebugMap = true;
860 /// \brief Get the starting and ending (exclusive) offset for the
861 /// attribute with index \p Idx descibed by \p Abbrev. \p Offset is
862 /// supposed to point to the position of the first attribute described
864 /// \return [StartOffset, EndOffset) as a pair.
865 static std::pair<uint32_t, uint32_t>
866 getAttributeOffsets(const DWARFAbbreviationDeclaration *Abbrev, unsigned Idx,
867 unsigned Offset, const DWARFUnit &Unit) {
868 DataExtractor Data = Unit.getDebugInfoExtractor();
870 for (unsigned i = 0; i < Idx; ++i)
871 DWARFFormValue::skipValue(Abbrev->getFormByIndex(i), Data, &Offset, &Unit);
873 uint32_t End = Offset;
874 DWARFFormValue::skipValue(Abbrev->getFormByIndex(Idx), Data, &End, &Unit);
876 return std::make_pair(Offset, End);
879 /// \brief Check if a variable describing DIE should be kept.
880 /// \returns updated TraversalFlags.
881 unsigned DwarfLinker::shouldKeepVariableDIE(
882 const DWARFDebugInfoEntryMinimal &DIE, CompileUnit &Unit,
883 CompileUnit::DIEInfo &MyInfo, unsigned Flags) {
884 const auto *Abbrev = DIE.getAbbreviationDeclarationPtr();
886 // Global variables with constant value can always be kept.
887 if (!(Flags & TF_InFunctionScope) &&
888 Abbrev->findAttributeIndex(dwarf::DW_AT_const_value) != -1U) {
889 MyInfo.InDebugMap = true;
890 return Flags | TF_Keep;
893 uint32_t LocationIdx = Abbrev->findAttributeIndex(dwarf::DW_AT_location);
894 if (LocationIdx == -1U)
897 uint32_t Offset = DIE.getOffset() + getULEB128Size(Abbrev->getCode());
898 const DWARFUnit &OrigUnit = Unit.getOrigUnit();
899 uint32_t LocationOffset, LocationEndOffset;
900 std::tie(LocationOffset, LocationEndOffset) =
901 getAttributeOffsets(Abbrev, LocationIdx, Offset, OrigUnit);
903 // See if there is a relocation to a valid debug map entry inside
904 // this variable's location. The order is important here. We want to
905 // always check in the variable has a valid relocation, so that the
906 // DIEInfo is filled. However, we don't want a static variable in a
907 // function to force us to keep the enclosing function.
908 if (!hasValidRelocation(LocationOffset, LocationEndOffset, MyInfo) ||
909 (Flags & TF_InFunctionScope))
913 DIE.dump(outs(), const_cast<DWARFUnit *>(&OrigUnit), 0, 8 /* Indent */);
915 return Flags | TF_Keep;
918 /// \brief Check if a function describing DIE should be kept.
919 /// \returns updated TraversalFlags.
920 unsigned DwarfLinker::shouldKeepSubprogramDIE(
921 const DWARFDebugInfoEntryMinimal &DIE, CompileUnit &Unit,
922 CompileUnit::DIEInfo &MyInfo, unsigned Flags) {
923 const auto *Abbrev = DIE.getAbbreviationDeclarationPtr();
925 Flags |= TF_InFunctionScope;
927 uint32_t LowPcIdx = Abbrev->findAttributeIndex(dwarf::DW_AT_low_pc);
931 uint32_t Offset = DIE.getOffset() + getULEB128Size(Abbrev->getCode());
932 const DWARFUnit &OrigUnit = Unit.getOrigUnit();
933 uint32_t LowPcOffset, LowPcEndOffset;
934 std::tie(LowPcOffset, LowPcEndOffset) =
935 getAttributeOffsets(Abbrev, LowPcIdx, Offset, OrigUnit);
938 DIE.getAttributeValueAsAddress(&OrigUnit, dwarf::DW_AT_low_pc, -1ULL);
939 assert(LowPc != -1ULL && "low_pc attribute is not an address.");
940 if (LowPc == -1ULL ||
941 !hasValidRelocation(LowPcOffset, LowPcEndOffset, MyInfo))
945 DIE.dump(outs(), const_cast<DWARFUnit *>(&OrigUnit), 0, 8 /* Indent */);
947 return Flags | TF_Keep;
950 /// \brief Check if a DIE should be kept.
951 /// \returns updated TraversalFlags.
952 unsigned DwarfLinker::shouldKeepDIE(const DWARFDebugInfoEntryMinimal &DIE,
954 CompileUnit::DIEInfo &MyInfo,
956 switch (DIE.getTag()) {
957 case dwarf::DW_TAG_constant:
958 case dwarf::DW_TAG_variable:
959 return shouldKeepVariableDIE(DIE, Unit, MyInfo, Flags);
960 case dwarf::DW_TAG_subprogram:
961 return shouldKeepSubprogramDIE(DIE, Unit, MyInfo, Flags);
962 case dwarf::DW_TAG_module:
963 case dwarf::DW_TAG_imported_module:
964 case dwarf::DW_TAG_imported_declaration:
965 case dwarf::DW_TAG_imported_unit:
966 // We always want to keep these.
967 return Flags | TF_Keep;
973 /// \brief Mark the passed DIE as well as all the ones it depends on
976 /// This function is called by lookForDIEsToKeep on DIEs that are
977 /// newly discovered to be needed in the link. It recursively calls
978 /// back to lookForDIEsToKeep while adding TF_DependencyWalk to the
979 /// TraversalFlags to inform it that it's not doing the primary DIE
981 void DwarfLinker::keepDIEAndDenpendencies(const DWARFDebugInfoEntryMinimal &DIE,
982 CompileUnit::DIEInfo &MyInfo,
983 const DebugMapObject &DMO,
984 CompileUnit &CU, unsigned Flags) {
985 const DWARFUnit &Unit = CU.getOrigUnit();
988 // First mark all the parent chain as kept.
989 unsigned AncestorIdx = MyInfo.ParentIdx;
990 while (!CU.getInfo(AncestorIdx).Keep) {
991 lookForDIEsToKeep(*Unit.getDIEAtIndex(AncestorIdx), DMO, CU,
992 TF_ParentWalk | TF_Keep | TF_DependencyWalk);
993 AncestorIdx = CU.getInfo(AncestorIdx).ParentIdx;
996 // Then we need to mark all the DIEs referenced by this DIE's
997 // attributes as kept.
998 DataExtractor Data = Unit.getDebugInfoExtractor();
999 const auto *Abbrev = DIE.getAbbreviationDeclarationPtr();
1000 uint32_t Offset = DIE.getOffset() + getULEB128Size(Abbrev->getCode());
1002 // Mark all DIEs referenced through atttributes as kept.
1003 for (const auto &AttrSpec : Abbrev->attributes()) {
1004 DWARFFormValue Val(AttrSpec.Form);
1006 if (!Val.isFormClass(DWARFFormValue::FC_Reference)) {
1007 DWARFFormValue::skipValue(AttrSpec.Form, Data, &Offset, &Unit);
1011 Val.extractValue(Data, &Offset, &Unit);
1012 CompileUnit *ReferencedCU;
1013 if (const auto *RefDIE = resolveDIEReference(Val, Unit, DIE, ReferencedCU))
1014 lookForDIEsToKeep(*RefDIE, DMO, *ReferencedCU,
1015 TF_Keep | TF_DependencyWalk);
1019 /// \brief Recursively walk the \p DIE tree and look for DIEs to
1020 /// keep. Store that information in \p CU's DIEInfo.
1022 /// This function is the entry point of the DIE selection
1023 /// algorithm. It is expected to walk the DIE tree in file order and
1024 /// (though the mediation of its helper) call hasValidRelocation() on
1025 /// each DIE that might be a 'root DIE' (See DwarfLinker class
1027 /// While walking the dependencies of root DIEs, this function is
1028 /// also called, but during these dependency walks the file order is
1029 /// not respected. The TF_DependencyWalk flag tells us which kind of
1030 /// traversal we are currently doing.
1031 void DwarfLinker::lookForDIEsToKeep(const DWARFDebugInfoEntryMinimal &DIE,
1032 const DebugMapObject &DMO, CompileUnit &CU,
1034 unsigned Idx = CU.getOrigUnit().getDIEIndex(&DIE);
1035 CompileUnit::DIEInfo &MyInfo = CU.getInfo(Idx);
1036 bool AlreadyKept = MyInfo.Keep;
1038 // If the Keep flag is set, we are marking a required DIE's
1039 // dependencies. If our target is already marked as kept, we're all
1041 if ((Flags & TF_DependencyWalk) && AlreadyKept)
1044 // We must not call shouldKeepDIE while called from keepDIEAndDenpendencies,
1045 // because it would screw up the relocation finding logic.
1046 if (!(Flags & TF_DependencyWalk))
1047 Flags = shouldKeepDIE(DIE, CU, MyInfo, Flags);
1049 // If it is a newly kept DIE mark it as well as all its dependencies as kept.
1050 if (!AlreadyKept && (Flags & TF_Keep))
1051 keepDIEAndDenpendencies(DIE, MyInfo, DMO, CU, Flags);
1053 // The TF_ParentWalk flag tells us that we are currently walking up
1054 // the parent chain of a required DIE, and we don't want to mark all
1055 // the children of the parents as kept (consider for example a
1056 // DW_TAG_namespace node in the parent chain). There are however a
1057 // set of DIE types for which we want to ignore that directive and still
1058 // walk their children.
1059 if (dieNeedsChildrenToBeMeaningful(DIE.getTag()))
1060 Flags &= ~TF_ParentWalk;
1062 if (!DIE.hasChildren() || (Flags & TF_ParentWalk))
1065 for (auto *Child = DIE.getFirstChild(); Child && !Child->isNULL();
1066 Child = Child->getSibling())
1067 lookForDIEsToKeep(*Child, DMO, CU, Flags);
1070 /// \brief Assign an abbreviation numer to \p Abbrev.
1072 /// Our DIEs get freed after every DebugMapObject has been processed,
1073 /// thus the FoldingSet we use to unique DIEAbbrevs cannot refer to
1074 /// the instances hold by the DIEs. When we encounter an abbreviation
1075 /// that we don't know, we create a permanent copy of it.
1076 void DwarfLinker::AssignAbbrev(DIEAbbrev &Abbrev) {
1077 // Check the set for priors.
1078 FoldingSetNodeID ID;
1081 DIEAbbrev *InSet = AbbreviationsSet.FindNodeOrInsertPos(ID, InsertToken);
1083 // If it's newly added.
1085 // Assign existing abbreviation number.
1086 Abbrev.setNumber(InSet->getNumber());
1088 // Add to abbreviation list.
1089 Abbreviations.push_back(
1090 new DIEAbbrev(Abbrev.getTag(), Abbrev.hasChildren()));
1091 for (const auto &Attr : Abbrev.getData())
1092 Abbreviations.back()->AddAttribute(Attr.getAttribute(), Attr.getForm());
1093 AbbreviationsSet.InsertNode(Abbreviations.back(), InsertToken);
1094 // Assign the unique abbreviation number.
1095 Abbrev.setNumber(Abbreviations.size());
1096 Abbreviations.back()->setNumber(Abbreviations.size());
1100 /// \brief Clone a string attribute described by \p AttrSpec and add
1102 /// \returns the size of the new attribute.
1103 unsigned DwarfLinker::cloneStringAttribute(DIE &Die, AttributeSpec AttrSpec,
1104 const DWARFFormValue &Val,
1105 const DWARFUnit &U) {
1106 // Switch everything to out of line strings.
1107 const char *String = *Val.getAsCString(&U);
1108 unsigned Offset = StringPool.getStringOffset(String);
1109 Die.addValue(dwarf::Attribute(AttrSpec.Attr), dwarf::DW_FORM_strp,
1110 new (DIEAlloc) DIEInteger(Offset));
1114 /// \brief Clone an attribute referencing another DIE and add
1116 /// \returns the size of the new attribute.
1117 unsigned DwarfLinker::cloneDieReferenceAttribute(
1118 DIE &Die, const DWARFDebugInfoEntryMinimal &InputDIE,
1119 AttributeSpec AttrSpec, unsigned AttrSize, const DWARFFormValue &Val,
1120 const DWARFUnit &U) {
1121 uint32_t Ref = *Val.getAsReference(&U);
1122 DIE *NewRefDie = nullptr;
1123 CompileUnit *RefUnit = nullptr;
1124 const DWARFDebugInfoEntryMinimal *RefDie = nullptr;
1126 if (!(RefUnit = getUnitForOffset(Ref)) ||
1127 !(RefDie = RefUnit->getOrigUnit().getDIEForOffset(Ref))) {
1128 const char *AttributeString = dwarf::AttributeString(AttrSpec.Attr);
1129 if (!AttributeString)
1130 AttributeString = "DW_AT_???";
1131 reportWarning(Twine("Missing DIE for ref in attribute ") + AttributeString +
1137 unsigned Idx = RefUnit->getOrigUnit().getDIEIndex(RefDie);
1138 CompileUnit::DIEInfo &RefInfo = RefUnit->getInfo(Idx);
1139 if (!RefInfo.Clone) {
1140 assert(Ref > InputDIE.getOffset());
1141 // We haven't cloned this DIE yet. Just create an empty one and
1142 // store it. It'll get really cloned when we process it.
1143 RefInfo.Clone = new DIE(dwarf::Tag(RefDie->getTag()));
1145 NewRefDie = RefInfo.Clone;
1147 if (AttrSpec.Form == dwarf::DW_FORM_ref_addr) {
1148 // We cannot currently rely on a DIEEntry to emit ref_addr
1149 // references, because the implementation calls back to DwarfDebug
1150 // to find the unit offset. (We don't have a DwarfDebug)
1151 // FIXME: we should be able to design DIEEntry reliance on
1154 if (Ref < InputDIE.getOffset()) {
1155 // We must have already cloned that DIE.
1156 uint32_t NewRefOffset =
1157 RefUnit->getStartOffset() + NewRefDie->getOffset();
1158 Attr = new (DIEAlloc) DIEInteger(NewRefOffset);
1160 // A forward reference. Note and fixup later.
1161 Attr = new (DIEAlloc) DIEInteger(0xBADDEF);
1162 RefUnit->noteForwardReference(NewRefDie, Attr);
1164 Die.addValue(dwarf::Attribute(AttrSpec.Attr), dwarf::DW_FORM_ref_addr,
1169 Die.addValue(dwarf::Attribute(AttrSpec.Attr), dwarf::Form(AttrSpec.Form),
1170 new (DIEAlloc) DIEEntry(*NewRefDie));
1174 /// \brief Clone an attribute of block form (locations, constants) and add
1176 /// \returns the size of the new attribute.
1177 unsigned DwarfLinker::cloneBlockAttribute(DIE &Die, AttributeSpec AttrSpec,
1178 const DWARFFormValue &Val,
1179 unsigned AttrSize) {
1182 DIELoc *Loc = nullptr;
1183 DIEBlock *Block = nullptr;
1184 // Just copy the block data over.
1185 if (AttrSpec.Attr == dwarf::DW_FORM_exprloc) {
1186 Loc = new (DIEAlloc) DIELoc();
1187 DIELocs.push_back(Loc);
1189 Block = new (DIEAlloc) DIEBlock();
1190 DIEBlocks.push_back(Block);
1192 Attr = Loc ? static_cast<DIE *>(Loc) : static_cast<DIE *>(Block);
1193 Value = Loc ? static_cast<DIEValue *>(Loc) : static_cast<DIEValue *>(Block);
1194 ArrayRef<uint8_t> Bytes = *Val.getAsBlock();
1195 for (auto Byte : Bytes)
1196 Attr->addValue(static_cast<dwarf::Attribute>(0), dwarf::DW_FORM_data1,
1197 new (DIEAlloc) DIEInteger(Byte));
1198 // FIXME: If DIEBlock and DIELoc just reuses the Size field of
1199 // the DIE class, this if could be replaced by
1200 // Attr->setSize(Bytes.size()).
1203 Loc->ComputeSize(&Streamer->getAsmPrinter());
1205 Block->ComputeSize(&Streamer->getAsmPrinter());
1207 Die.addValue(dwarf::Attribute(AttrSpec.Attr), dwarf::Form(AttrSpec.Form),
1212 /// \brief Clone an address attribute and add it to \p Die.
1213 /// \returns the size of the new attribute.
1214 unsigned DwarfLinker::cloneAddressAttribute(DIE &Die, AttributeSpec AttrSpec,
1215 const DWARFFormValue &Val,
1216 const CompileUnit &Unit,
1217 AttributesInfo &Info) {
1218 int64_t Addr = *Val.getAsAddress(&Unit.getOrigUnit());
1219 if (AttrSpec.Attr == dwarf::DW_AT_low_pc) {
1220 if (Die.getTag() == dwarf::DW_TAG_inlined_subroutine ||
1221 Die.getTag() == dwarf::DW_TAG_lexical_block)
1222 Addr += Info.PCOffset;
1223 } else if (AttrSpec.Attr == dwarf::DW_AT_high_pc) {
1224 // If we have a high_pc recorded for the input DIE, use
1225 // it. Otherwise (when no relocations where applied) just use the
1226 // one we just decoded.
1227 Addr = (Info.OrigHighPc ? Info.OrigHighPc : Addr) + Info.PCOffset;
1230 Die.addValue(static_cast<dwarf::Attribute>(AttrSpec.Attr),
1231 static_cast<dwarf::Form>(AttrSpec.Form),
1232 new (DIEAlloc) DIEInteger(Addr));
1233 return Unit.getOrigUnit().getAddressByteSize();
1236 /// \brief Clone a scalar attribute and add it to \p Die.
1237 /// \returns the size of the new attribute.
1238 unsigned DwarfLinker::cloneScalarAttribute(
1239 DIE &Die, const DWARFDebugInfoEntryMinimal &InputDIE, const DWARFUnit &U,
1240 AttributeSpec AttrSpec, const DWARFFormValue &Val, unsigned AttrSize) {
1242 if (AttrSpec.Form == dwarf::DW_FORM_sec_offset)
1243 Value = *Val.getAsSectionOffset();
1244 else if (AttrSpec.Form == dwarf::DW_FORM_sdata)
1245 Value = *Val.getAsSignedConstant();
1246 else if (auto OptionalValue = Val.getAsUnsignedConstant())
1247 Value = *OptionalValue;
1249 reportWarning("Unsupported scalar attribute form. Dropping attribute.", &U,
1253 Die.addValue(dwarf::Attribute(AttrSpec.Attr), dwarf::Form(AttrSpec.Form),
1254 new (DIEAlloc) DIEInteger(Value));
1258 /// \brief Clone \p InputDIE's attribute described by \p AttrSpec with
1259 /// value \p Val, and add it to \p Die.
1260 /// \returns the size of the cloned attribute.
1261 unsigned DwarfLinker::cloneAttribute(DIE &Die,
1262 const DWARFDebugInfoEntryMinimal &InputDIE,
1264 const DWARFFormValue &Val,
1265 const AttributeSpec AttrSpec,
1266 unsigned AttrSize, AttributesInfo &Info) {
1267 const DWARFUnit &U = Unit.getOrigUnit();
1269 switch (AttrSpec.Form) {
1270 case dwarf::DW_FORM_strp:
1271 case dwarf::DW_FORM_string:
1272 return cloneStringAttribute(Die, AttrSpec, Val, U);
1273 case dwarf::DW_FORM_ref_addr:
1274 case dwarf::DW_FORM_ref1:
1275 case dwarf::DW_FORM_ref2:
1276 case dwarf::DW_FORM_ref4:
1277 case dwarf::DW_FORM_ref8:
1278 return cloneDieReferenceAttribute(Die, InputDIE, AttrSpec, AttrSize, Val,
1280 case dwarf::DW_FORM_block:
1281 case dwarf::DW_FORM_block1:
1282 case dwarf::DW_FORM_block2:
1283 case dwarf::DW_FORM_block4:
1284 case dwarf::DW_FORM_exprloc:
1285 return cloneBlockAttribute(Die, AttrSpec, Val, AttrSize);
1286 case dwarf::DW_FORM_addr:
1287 return cloneAddressAttribute(Die, AttrSpec, Val, Unit, Info);
1288 case dwarf::DW_FORM_data1:
1289 case dwarf::DW_FORM_data2:
1290 case dwarf::DW_FORM_data4:
1291 case dwarf::DW_FORM_data8:
1292 case dwarf::DW_FORM_udata:
1293 case dwarf::DW_FORM_sdata:
1294 case dwarf::DW_FORM_sec_offset:
1295 case dwarf::DW_FORM_flag:
1296 case dwarf::DW_FORM_flag_present:
1297 return cloneScalarAttribute(Die, InputDIE, U, AttrSpec, Val, AttrSize);
1299 reportWarning("Unsupported attribute form in cloneAttribute. Dropping.", &U,
1306 /// \brief Apply the valid relocations found by findValidRelocs() to
1307 /// the buffer \p Data, taking into account that Data is at \p BaseOffset
1308 /// in the debug_info section.
1310 /// Like for findValidRelocs(), this function must be called with
1311 /// monotonic \p BaseOffset values.
1313 /// \returns wether any reloc has been applied.
1314 bool DwarfLinker::applyValidRelocs(MutableArrayRef<char> Data,
1315 uint32_t BaseOffset, bool isLittleEndian) {
1316 assert((NextValidReloc == 0 ||
1317 BaseOffset > ValidRelocs[NextValidReloc - 1].Offset) &&
1318 "BaseOffset should only be increasing.");
1319 if (NextValidReloc >= ValidRelocs.size())
1322 // Skip relocs that haven't been applied.
1323 while (NextValidReloc < ValidRelocs.size() &&
1324 ValidRelocs[NextValidReloc].Offset < BaseOffset)
1327 bool Applied = false;
1328 uint64_t EndOffset = BaseOffset + Data.size();
1329 while (NextValidReloc < ValidRelocs.size() &&
1330 ValidRelocs[NextValidReloc].Offset >= BaseOffset &&
1331 ValidRelocs[NextValidReloc].Offset < EndOffset) {
1332 const auto &ValidReloc = ValidRelocs[NextValidReloc++];
1333 assert(ValidReloc.Offset - BaseOffset < Data.size());
1334 assert(ValidReloc.Offset - BaseOffset + ValidReloc.Size <= Data.size());
1336 uint64_t Value = ValidReloc.Mapping->getValue().BinaryAddress;
1337 Value += ValidReloc.Addend;
1338 for (unsigned i = 0; i != ValidReloc.Size; ++i) {
1339 unsigned Index = isLittleEndian ? i : (ValidReloc.Size - i - 1);
1340 Buf[i] = uint8_t(Value >> (Index * 8));
1342 assert(ValidReloc.Size <= sizeof(Buf));
1343 memcpy(&Data[ValidReloc.Offset - BaseOffset], Buf, ValidReloc.Size);
1350 /// \brief Recursively clone \p InputDIE's subtrees that have been
1351 /// selected to appear in the linked output.
1353 /// \param OutOffset is the Offset where the newly created DIE will
1354 /// lie in the linked compile unit.
1356 /// \returns the cloned DIE object or null if nothing was selected.
1357 DIE *DwarfLinker::cloneDIE(const DWARFDebugInfoEntryMinimal &InputDIE,
1358 CompileUnit &Unit, int64_t PCOffset,
1359 uint32_t OutOffset) {
1360 DWARFUnit &U = Unit.getOrigUnit();
1361 unsigned Idx = U.getDIEIndex(&InputDIE);
1362 CompileUnit::DIEInfo &Info = Unit.getInfo(Idx);
1364 // Should the DIE appear in the output?
1365 if (!Unit.getInfo(Idx).Keep)
1368 uint32_t Offset = InputDIE.getOffset();
1369 // The DIE might have been already created by a forward reference
1370 // (see cloneDieReferenceAttribute()).
1371 DIE *Die = Info.Clone;
1373 Die = Info.Clone = new DIE(dwarf::Tag(InputDIE.getTag()));
1374 assert(Die->getTag() == InputDIE.getTag());
1375 Die->setOffset(OutOffset);
1377 // Extract and clone every attribute.
1378 DataExtractor Data = U.getDebugInfoExtractor();
1379 uint32_t NextOffset = U.getDIEAtIndex(Idx + 1)->getOffset();
1380 AttributesInfo AttrInfo;
1382 // We could copy the data only if we need to aply a relocation to
1383 // it. After testing, it seems there is no performance downside to
1384 // doing the copy unconditionally, and it makes the code simpler.
1385 SmallString<40> DIECopy(Data.getData().substr(Offset, NextOffset - Offset));
1386 Data = DataExtractor(DIECopy, Data.isLittleEndian(), Data.getAddressSize());
1387 // Modify the copy with relocated addresses.
1388 if (applyValidRelocs(DIECopy, Offset, Data.isLittleEndian())) {
1389 // If we applied relocations, we store the value of high_pc that was
1390 // potentially stored in the input DIE. If high_pc is an address
1391 // (Dwarf version == 2), then it might have been relocated to a
1392 // totally unrelated value (because the end address in the object
1393 // file might be start address of another function which got moved
1394 // independantly by the linker). The computation of the actual
1395 // high_pc value is done in cloneAddressAttribute().
1396 AttrInfo.OrigHighPc =
1397 InputDIE.getAttributeValueAsAddress(&U, dwarf::DW_AT_high_pc, 0);
1400 // Reset the Offset to 0 as we will be working on the local copy of
1404 const auto *Abbrev = InputDIE.getAbbreviationDeclarationPtr();
1405 Offset += getULEB128Size(Abbrev->getCode());
1407 // We are entering a subprogram. Get and propagate the PCOffset.
1408 if (Die->getTag() == dwarf::DW_TAG_subprogram)
1409 PCOffset = Info.AddrAdjust;
1410 AttrInfo.PCOffset = PCOffset;
1412 for (const auto &AttrSpec : Abbrev->attributes()) {
1413 DWARFFormValue Val(AttrSpec.Form);
1414 uint32_t AttrSize = Offset;
1415 Val.extractValue(Data, &Offset, &U);
1416 AttrSize = Offset - AttrSize;
1419 cloneAttribute(*Die, InputDIE, Unit, Val, AttrSpec, AttrSize, AttrInfo);
1422 DIEAbbrev &NewAbbrev = Die->getAbbrev();
1423 // If a scope DIE is kept, we must have kept at least one child. If
1424 // it's not the case, we'll just be emitting one wasteful end of
1425 // children marker, but things won't break.
1426 if (InputDIE.hasChildren())
1427 NewAbbrev.setChildrenFlag(dwarf::DW_CHILDREN_yes);
1428 // Assign a permanent abbrev number
1429 AssignAbbrev(Die->getAbbrev());
1431 // Add the size of the abbreviation number to the output offset.
1432 OutOffset += getULEB128Size(Die->getAbbrevNumber());
1434 if (!Abbrev->hasChildren()) {
1436 Die->setSize(OutOffset - Die->getOffset());
1440 // Recursively clone children.
1441 for (auto *Child = InputDIE.getFirstChild(); Child && !Child->isNULL();
1442 Child = Child->getSibling()) {
1443 if (DIE *Clone = cloneDIE(*Child, Unit, PCOffset, OutOffset)) {
1444 Die->addChild(std::unique_ptr<DIE>(Clone));
1445 OutOffset = Clone->getOffset() + Clone->getSize();
1449 // Account for the end of children marker.
1450 OutOffset += sizeof(int8_t);
1452 Die->setSize(OutOffset - Die->getOffset());
1456 bool DwarfLinker::link(const DebugMap &Map) {
1458 if (Map.begin() == Map.end()) {
1459 errs() << "Empty debug map.\n";
1463 if (!createStreamer(Map.getTriple(), OutputFilename))
1466 // Size of the DIEs (and headers) generated for the linked output.
1467 uint64_t OutputDebugInfoSize = 0;
1469 for (const auto &Obj : Map.objects()) {
1470 CurrentDebugObject = Obj.get();
1472 if (Options.Verbose)
1473 outs() << "DEBUG MAP OBJECT: " << Obj->getObjectFilename() << "\n";
1474 auto ErrOrObj = BinHolder.GetObjectFile(Obj->getObjectFilename());
1475 if (std::error_code EC = ErrOrObj.getError()) {
1476 reportWarning(Twine(Obj->getObjectFilename()) + ": " + EC.message());
1480 // Look for relocations that correspond to debug map entries.
1481 if (!findValidRelocsInDebugInfo(*ErrOrObj, *Obj)) {
1482 if (Options.Verbose)
1483 outs() << "No valid relocations found. Skipping.\n";
1487 // Setup access to the debug info.
1488 DWARFContextInMemory DwarfContext(*ErrOrObj);
1489 startDebugObject(DwarfContext);
1491 // In a first phase, just read in the debug info and store the DIE
1492 // parent links that we will use during the next phase.
1493 for (const auto &CU : DwarfContext.compile_units()) {
1494 auto *CUDie = CU->getCompileUnitDIE(false);
1495 if (Options.Verbose) {
1496 outs() << "Input compilation unit:";
1497 CUDie->dump(outs(), CU.get(), 0);
1499 Units.emplace_back(*CU);
1500 gatherDIEParents(CUDie, 0, Units.back());
1503 // Then mark all the DIEs that need to be present in the linked
1504 // output and collect some information about them. Note that this
1505 // loop can not be merged with the previous one becaue cross-cu
1506 // references require the ParentIdx to be setup for every CU in
1507 // the object file before calling this.
1508 for (auto &CurrentUnit : Units)
1509 lookForDIEsToKeep(*CurrentUnit.getOrigUnit().getCompileUnitDIE(), *Obj,
1512 // The calls to applyValidRelocs inside cloneDIE will walk the
1513 // reloc array again (in the same way findValidRelocsInDebugInfo()
1514 // did). We need to reset the NextValidReloc index to the beginning.
1517 // Construct the output DIE tree by cloning the DIEs we chose to
1518 // keep above. If there are no valid relocs, then there's nothing
1520 if (!ValidRelocs.empty())
1521 for (auto &CurrentUnit : Units) {
1522 const auto *InputDIE = CurrentUnit.getOrigUnit().getCompileUnitDIE();
1523 CurrentUnit.setStartOffset(OutputDebugInfoSize);
1524 DIE *OutputDIE = cloneDIE(*InputDIE, CurrentUnit, 0 /* PCOffset */,
1525 11 /* Unit Header size */);
1526 CurrentUnit.setOutputUnitDIE(OutputDIE);
1527 OutputDebugInfoSize = CurrentUnit.computeNextUnitOffset();
1530 // Emit all the compile unit's debug information.
1531 if (!ValidRelocs.empty() && !Options.NoOutput)
1532 for (auto &CurrentUnit : Units) {
1533 CurrentUnit.fixupForwardReferences();
1534 Streamer->emitCompileUnitHeader(CurrentUnit);
1535 if (!CurrentUnit.getOutputUnitDIE())
1537 Streamer->emitDIE(*CurrentUnit.getOutputUnitDIE());
1540 // Clean-up before starting working on the next object.
1544 // Emit everything that's global.
1545 if (!Options.NoOutput) {
1546 Streamer->emitAbbrevs(Abbreviations);
1547 Streamer->emitStrings(StringPool);
1550 return Options.NoOutput ? true : Streamer->finish();
1554 bool linkDwarf(StringRef OutputFilename, const DebugMap &DM,
1555 const LinkOptions &Options) {
1556 DwarfLinker Linker(OutputFilename, Options);
1557 return Linker.link(DM);