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/CodeGen/AsmPrinter.h"
14 #include "llvm/CodeGen/DIE.h"
15 #include "llvm/DebugInfo/DWARF/DWARFContext.h"
16 #include "llvm/DebugInfo/DWARF/DWARFDebugInfoEntry.h"
17 #include "llvm/DebugInfo/DWARF/DWARFFormValue.h"
18 #include "llvm/MC/MCAsmBackend.h"
19 #include "llvm/MC/MCAsmInfo.h"
20 #include "llvm/MC/MCContext.h"
21 #include "llvm/MC/MCCodeEmitter.h"
22 #include "llvm/MC/MCInstrInfo.h"
23 #include "llvm/MC/MCObjectFileInfo.h"
24 #include "llvm/MC/MCRegisterInfo.h"
25 #include "llvm/MC/MCStreamer.h"
26 #include "llvm/Object/MachO.h"
27 #include "llvm/Support/Dwarf.h"
28 #include "llvm/Support/LEB128.h"
29 #include "llvm/Support/TargetRegistry.h"
30 #include "llvm/Target/TargetMachine.h"
31 #include "llvm/Target/TargetOptions.h"
39 void warn(const Twine &Warning, const Twine &Context) {
40 errs() << Twine("while processing ") + Context + ":\n";
41 errs() << Twine("warning: ") + Warning + "\n";
44 bool error(const Twine &Error, const Twine &Context) {
45 errs() << Twine("while processing ") + Context + ":\n";
46 errs() << Twine("error: ") + Error + "\n";
50 /// \brief Stores all information relating to a compile unit, be it in
51 /// its original instance in the object file to its brand new cloned
52 /// and linked DIE tree.
55 /// \brief Information gathered about a DIE in the object file.
57 int64_t AddrAdjust; ///< Address offset to apply to the described entity.
58 DIE *Clone; ///< Cloned version of that DIE.
59 uint32_t ParentIdx; ///< The index of this DIE's parent.
60 bool Keep; ///< Is the DIE part of the linked output?
61 bool InDebugMap; ///< Was this DIE's entity found in the map?
64 CompileUnit(DWARFUnit &OrigUnit) : OrigUnit(OrigUnit) {
65 Info.resize(OrigUnit.getNumDIEs());
68 // Workaround MSVC not supporting implicit move ops
69 CompileUnit(CompileUnit &&RHS)
70 : OrigUnit(RHS.OrigUnit), Info(std::move(RHS.Info)),
71 CUDie(std::move(RHS.CUDie)), StartOffset(RHS.StartOffset),
72 NextUnitOffset(RHS.NextUnitOffset) {}
74 DWARFUnit &getOrigUnit() const { return OrigUnit; }
76 DIE *getOutputUnitDIE() const { return CUDie.get(); }
77 void setOutputUnitDIE(DIE *Die) { CUDie.reset(Die); }
79 DIEInfo &getInfo(unsigned Idx) { return Info[Idx]; }
80 const DIEInfo &getInfo(unsigned Idx) const { return Info[Idx]; }
82 uint64_t getStartOffset() const { return StartOffset; }
83 uint64_t getNextUnitOffset() const { return NextUnitOffset; }
85 void setStartOffset(uint64_t DebugInfoSize) { StartOffset = DebugInfoSize; }
87 /// \brief Compute the end offset for this unit. Must be
88 /// called after the CU's DIEs have been cloned.
89 /// \returns the next unit offset (which is also the current
90 /// debug_info section size).
91 uint64_t computeNextUnitOffset();
93 /// \brief Keep track of a forward reference to DIE \p Die by
94 /// \p Attr. The attribute should be fixed up later to point to the
95 /// absolute offset of \p Die in the debug_info section.
96 void noteForwardReference(DIE *Die, DIEInteger *Attr);
98 /// \brief Apply all fixups recored by noteForwardReference().
99 void fixupForwardReferences();
103 std::vector<DIEInfo> Info; ///< DIE info indexed by DIE index.
104 std::unique_ptr<DIE> CUDie; ///< Root of the linked DIE tree.
106 uint64_t StartOffset;
107 uint64_t NextUnitOffset;
109 /// \brief A list of attributes to fixup with the absolute offset of
110 /// a DIE in the debug_info section.
112 /// The offsets for the attributes in this array couldn't be set while
113 /// cloning because for forward refences the target DIE's offset isn't
114 /// known you emit the reference attribute.
115 std::vector<std::pair<DIE *, DIEInteger *>> ForwardDIEReferences;
118 uint64_t CompileUnit::computeNextUnitOffset() {
119 NextUnitOffset = StartOffset + 11 /* Header size */;
120 // The root DIE might be null, meaning that the Unit had nothing to
121 // contribute to the linked output. In that case, we will emit the
122 // unit header without any actual DIE.
124 NextUnitOffset += CUDie->getSize();
125 return NextUnitOffset;
128 /// \brief Keep track of a forward reference to \p Die.
129 void CompileUnit::noteForwardReference(DIE *Die, DIEInteger *Attr) {
130 ForwardDIEReferences.emplace_back(Die, Attr);
133 /// \brief Apply all fixups recorded by noteForwardReference().
134 void CompileUnit::fixupForwardReferences() {
135 for (const auto &Ref : ForwardDIEReferences)
136 Ref.second->setValue(Ref.first->getOffset() + getStartOffset());
139 /// \brief A string table that doesn't need relocations.
141 /// We are doing a final link, no need for a string table that
142 /// has relocation entries for every reference to it. This class
143 /// provides this ablitity by just associating offsets with
145 class NonRelocatableStringpool {
147 /// \brief Entries are stored into the StringMap and simply linked
148 /// together through the second element of this pair in order to
149 /// keep track of insertion order.
150 typedef StringMap<std::pair<uint32_t, StringMapEntryBase *>, BumpPtrAllocator>
153 NonRelocatableStringpool()
154 : CurrentEndOffset(0), Sentinel(0), Last(&Sentinel) {
155 // Legacy dsymutil puts an empty string at the start of the line
160 /// \brief Get the offset of string \p S in the string table. This
161 /// can insert a new element or return the offset of a preexisitng
163 uint32_t getStringOffset(StringRef S);
165 /// \brief Get permanent storage for \p S (but do not necessarily
166 /// emit \p S in the output section).
167 /// \returns The StringRef that points to permanent storage to use
168 /// in place of \p S.
169 StringRef internString(StringRef S);
171 // \brief Return the first entry of the string table.
172 const MapTy::MapEntryTy *getFirstEntry() const {
173 return getNextEntry(&Sentinel);
176 // \brief Get the entry following \p E in the string table or null
177 // if \p E was the last entry.
178 const MapTy::MapEntryTy *getNextEntry(const MapTy::MapEntryTy *E) const {
179 return static_cast<const MapTy::MapEntryTy *>(E->getValue().second);
182 uint64_t getSize() { return CurrentEndOffset; }
186 uint32_t CurrentEndOffset;
187 MapTy::MapEntryTy Sentinel, *Last;
190 /// \brief Get the offset of string \p S in the string table. This
191 /// can insert a new element or return the offset of a preexisitng
193 uint32_t NonRelocatableStringpool::getStringOffset(StringRef S) {
194 if (S.empty() && !Strings.empty())
197 std::pair<uint32_t, StringMapEntryBase *> Entry(0, nullptr);
201 // A non-empty string can't be at offset 0, so if we have an entry
202 // with a 0 offset, it must be a previously interned string.
203 std::tie(It, Inserted) = Strings.insert(std::make_pair(S, Entry));
204 if (Inserted || It->getValue().first == 0) {
205 // Set offset and chain at the end of the entries list.
206 It->getValue().first = CurrentEndOffset;
207 CurrentEndOffset += S.size() + 1; // +1 for the '\0'.
208 Last->getValue().second = &*It;
211 return It->getValue().first;
214 /// \brief Put \p S into the StringMap so that it gets permanent
215 /// storage, but do not actually link it in the chain of elements
216 /// that go into the output section. A latter call to
217 /// getStringOffset() with the same string will chain it though.
218 StringRef NonRelocatableStringpool::internString(StringRef S) {
219 std::pair<uint32_t, StringMapEntryBase *> Entry(0, nullptr);
220 auto InsertResult = Strings.insert(std::make_pair(S, Entry));
221 return InsertResult.first->getKey();
224 /// \brief The Dwarf streaming logic
226 /// All interactions with the MC layer that is used to build the debug
227 /// information binary representation are handled in this class.
228 class DwarfStreamer {
229 /// \defgroup MCObjects MC layer objects constructed by the streamer
231 std::unique_ptr<MCRegisterInfo> MRI;
232 std::unique_ptr<MCAsmInfo> MAI;
233 std::unique_ptr<MCObjectFileInfo> MOFI;
234 std::unique_ptr<MCContext> MC;
235 MCAsmBackend *MAB; // Owned by MCStreamer
236 std::unique_ptr<MCInstrInfo> MII;
237 std::unique_ptr<MCSubtargetInfo> MSTI;
238 MCCodeEmitter *MCE; // Owned by MCStreamer
239 MCStreamer *MS; // Owned by AsmPrinter
240 std::unique_ptr<TargetMachine> TM;
241 std::unique_ptr<AsmPrinter> Asm;
244 /// \brief the file we stream the linked Dwarf to.
245 std::unique_ptr<raw_fd_ostream> OutFile;
248 /// \brief Actually create the streamer and the ouptut file.
250 /// This could be done directly in the constructor, but it feels
251 /// more natural to handle errors through return value.
252 bool init(Triple TheTriple, StringRef OutputFilename);
254 /// \brief Dump the file to the disk.
257 AsmPrinter &getAsmPrinter() const { return *Asm; }
259 /// \brief Set the current output section to debug_info and change
260 /// the MC Dwarf version to \p DwarfVersion.
261 void switchToDebugInfoSection(unsigned DwarfVersion);
263 /// \brief Emit the compilation unit header for \p Unit in the
264 /// debug_info section.
266 /// As a side effect, this also switches the current Dwarf version
267 /// of the MC layer to the one of U.getOrigUnit().
268 void emitCompileUnitHeader(CompileUnit &Unit);
270 /// \brief Recursively emit the DIE tree rooted at \p Die.
271 void emitDIE(DIE &Die);
273 /// \brief Emit the abbreviation table \p Abbrevs to the
274 /// debug_abbrev section.
275 void emitAbbrevs(const std::vector<DIEAbbrev *> &Abbrevs);
277 /// \brief Emit the string table described by \p Pool.
278 void emitStrings(const NonRelocatableStringpool &Pool);
281 bool DwarfStreamer::init(Triple TheTriple, StringRef OutputFilename) {
282 std::string ErrorStr;
283 std::string TripleName;
284 StringRef Context = "dwarf streamer init";
287 const Target *TheTarget =
288 TargetRegistry::lookupTarget(TripleName, TheTriple, ErrorStr);
290 return error(ErrorStr, Context);
291 TripleName = TheTriple.getTriple();
293 // Create all the MC Objects.
294 MRI.reset(TheTarget->createMCRegInfo(TripleName));
296 return error(Twine("no register info for target ") + TripleName, Context);
298 MAI.reset(TheTarget->createMCAsmInfo(*MRI, TripleName));
300 return error("no asm info for target " + TripleName, Context);
302 MOFI.reset(new MCObjectFileInfo);
303 MC.reset(new MCContext(MAI.get(), MRI.get(), MOFI.get()));
304 MOFI->InitMCObjectFileInfo(TripleName, Reloc::Default, CodeModel::Default,
307 MAB = TheTarget->createMCAsmBackend(*MRI, TripleName, "");
309 return error("no asm backend for target " + TripleName, Context);
311 MII.reset(TheTarget->createMCInstrInfo());
313 return error("no instr info info for target " + TripleName, Context);
315 MSTI.reset(TheTarget->createMCSubtargetInfo(TripleName, "", ""));
317 return error("no subtarget info for target " + TripleName, Context);
319 MCE = TheTarget->createMCCodeEmitter(*MII, *MRI, *MC);
321 return error("no code emitter for target " + TripleName, Context);
323 // Create the output file.
326 llvm::make_unique<raw_fd_ostream>(OutputFilename, EC, sys::fs::F_None);
328 return error(Twine(OutputFilename) + ": " + EC.message(), Context);
330 MS = TheTarget->createMCObjectStreamer(TripleName, *MC, *MAB, *OutFile, MCE,
333 return error("no object streamer for target " + TripleName, Context);
335 // Finally create the AsmPrinter we'll use to emit the DIEs.
336 TM.reset(TheTarget->createTargetMachine(TripleName, "", "", TargetOptions()));
338 return error("no target machine for target " + TripleName, Context);
340 Asm.reset(TheTarget->createAsmPrinter(*TM, std::unique_ptr<MCStreamer>(MS)));
342 return error("no asm printer for target " + TripleName, Context);
347 bool DwarfStreamer::finish() {
352 /// \brief Set the current output section to debug_info and change
353 /// the MC Dwarf version to \p DwarfVersion.
354 void DwarfStreamer::switchToDebugInfoSection(unsigned DwarfVersion) {
355 MS->SwitchSection(MOFI->getDwarfInfoSection());
356 MC->setDwarfVersion(DwarfVersion);
359 /// \brief Emit the compilation unit header for \p Unit in the
360 /// debug_info section.
362 /// A Dwarf scetion header is encoded as:
363 /// uint32_t Unit length (omiting this field)
365 /// uint32_t Abbreviation table offset
366 /// uint8_t Address size
368 /// Leading to a total of 11 bytes.
369 void DwarfStreamer::emitCompileUnitHeader(CompileUnit &Unit) {
370 unsigned Version = Unit.getOrigUnit().getVersion();
371 switchToDebugInfoSection(Version);
373 // Emit size of content not including length itself. The size has
374 // already been computed in CompileUnit::computeOffsets(). Substract
375 // 4 to that size to account for the length field.
376 Asm->EmitInt32(Unit.getNextUnitOffset() - Unit.getStartOffset() - 4);
377 Asm->EmitInt16(Version);
378 // We share one abbreviations table across all units so it's always at the
379 // start of the section.
381 Asm->EmitInt8(Unit.getOrigUnit().getAddressByteSize());
384 /// \brief Emit the \p Abbrevs array as the shared abbreviation table
385 /// for the linked Dwarf file.
386 void DwarfStreamer::emitAbbrevs(const std::vector<DIEAbbrev *> &Abbrevs) {
387 MS->SwitchSection(MOFI->getDwarfAbbrevSection());
388 Asm->emitDwarfAbbrevs(Abbrevs);
391 /// \brief Recursively emit the DIE tree rooted at \p Die.
392 void DwarfStreamer::emitDIE(DIE &Die) {
393 MS->SwitchSection(MOFI->getDwarfInfoSection());
394 Asm->emitDwarfDIE(Die);
397 /// \brief Emit the debug_str section stored in \p Pool.
398 void DwarfStreamer::emitStrings(const NonRelocatableStringpool &Pool) {
399 Asm->OutStreamer.SwitchSection(MOFI->getDwarfStrSection());
400 for (auto *Entry = Pool.getFirstEntry(); Entry;
401 Entry = Pool.getNextEntry(Entry))
402 Asm->OutStreamer.EmitBytes(
403 StringRef(Entry->getKey().data(), Entry->getKey().size() + 1));
406 /// \brief The core of the Dwarf linking logic.
408 /// The link of the dwarf information from the object files will be
409 /// driven by the selection of 'root DIEs', which are DIEs that
410 /// describe variables or functions that are present in the linked
411 /// binary (and thus have entries in the debug map). All the debug
412 /// information that will be linked (the DIEs, but also the line
413 /// tables, ranges, ...) is derived from that set of root DIEs.
415 /// The root DIEs are identified because they contain relocations that
416 /// correspond to a debug map entry at specific places (the low_pc for
417 /// a function, the location for a variable). These relocations are
418 /// called ValidRelocs in the DwarfLinker and are gathered as a very
419 /// first step when we start processing a DebugMapObject.
422 DwarfLinker(StringRef OutputFilename, const LinkOptions &Options)
423 : OutputFilename(OutputFilename), Options(Options),
424 BinHolder(Options.Verbose) {}
427 for (auto *Abbrev : Abbreviations)
431 /// \brief Link the contents of the DebugMap.
432 bool link(const DebugMap &);
435 /// \brief Called at the start of a debug object link.
436 void startDebugObject(DWARFContext &);
438 /// \brief Called at the end of a debug object link.
439 void endDebugObject();
441 /// \defgroup FindValidRelocations Translate debug map into a list
442 /// of relevant relocations
449 const DebugMapObject::DebugMapEntry *Mapping;
451 ValidReloc(uint32_t Offset, uint32_t Size, uint64_t Addend,
452 const DebugMapObject::DebugMapEntry *Mapping)
453 : Offset(Offset), Size(Size), Addend(Addend), Mapping(Mapping) {}
455 bool operator<(const ValidReloc &RHS) const { return Offset < RHS.Offset; }
458 /// \brief The valid relocations for the current DebugMapObject.
459 /// This vector is sorted by relocation offset.
460 std::vector<ValidReloc> ValidRelocs;
462 /// \brief Index into ValidRelocs of the next relocation to
463 /// consider. As we walk the DIEs in acsending file offset and as
464 /// ValidRelocs is sorted by file offset, keeping this index
465 /// uptodate is all we have to do to have a cheap lookup during the
466 /// root DIE selection and during DIE cloning.
467 unsigned NextValidReloc;
469 bool findValidRelocsInDebugInfo(const object::ObjectFile &Obj,
470 const DebugMapObject &DMO);
472 bool findValidRelocs(const object::SectionRef &Section,
473 const object::ObjectFile &Obj,
474 const DebugMapObject &DMO);
476 void findValidRelocsMachO(const object::SectionRef &Section,
477 const object::MachOObjectFile &Obj,
478 const DebugMapObject &DMO);
481 /// \defgroup FindRootDIEs Find DIEs corresponding to debug map entries.
484 /// \brief Recursively walk the \p DIE tree and look for DIEs to
485 /// keep. Store that information in \p CU's DIEInfo.
486 void lookForDIEsToKeep(const DWARFDebugInfoEntryMinimal &DIE,
487 const DebugMapObject &DMO, CompileUnit &CU,
490 /// \brief Flags passed to DwarfLinker::lookForDIEsToKeep
492 TF_Keep = 1 << 0, ///< Mark the traversed DIEs as kept.
493 TF_InFunctionScope = 1 << 1, ///< Current scope is a fucntion scope.
494 TF_DependencyWalk = 1 << 2, ///< Walking the dependencies of a kept DIE.
495 TF_ParentWalk = 1 << 3, ///< Walking up the parents of a kept DIE.
498 /// \brief Mark the passed DIE as well as all the ones it depends on
500 void keepDIEAndDenpendencies(const DWARFDebugInfoEntryMinimal &DIE,
501 CompileUnit::DIEInfo &MyInfo,
502 const DebugMapObject &DMO, CompileUnit &CU,
505 unsigned shouldKeepDIE(const DWARFDebugInfoEntryMinimal &DIE,
506 CompileUnit &Unit, CompileUnit::DIEInfo &MyInfo,
509 unsigned shouldKeepVariableDIE(const DWARFDebugInfoEntryMinimal &DIE,
511 CompileUnit::DIEInfo &MyInfo, unsigned Flags);
513 unsigned shouldKeepSubprogramDIE(const DWARFDebugInfoEntryMinimal &DIE,
515 CompileUnit::DIEInfo &MyInfo,
518 bool hasValidRelocation(uint32_t StartOffset, uint32_t EndOffset,
519 CompileUnit::DIEInfo &Info);
522 /// \defgroup Linking Methods used to link the debug information
525 /// \brief Recursively clone \p InputDIE into an tree of DIE objects
526 /// where useless (as decided by lookForDIEsToKeep()) bits have been
527 /// stripped out and addresses have been rewritten according to the
530 /// \param OutOffset is the offset the cloned DIE in the output
532 /// \param PCOffset (while cloning a function scope) is the offset
533 /// applied to the entry point of the function to get the linked address.
535 /// \returns the root of the cloned tree.
536 DIE *cloneDIE(const DWARFDebugInfoEntryMinimal &InputDIE, CompileUnit &U,
537 int64_t PCOffset, uint32_t OutOffset);
539 typedef DWARFAbbreviationDeclaration::AttributeSpec AttributeSpec;
541 /// \brief Information gathered and exchanged between the various
542 /// clone*Attributes helpers about the attributes of a particular DIE.
543 struct AttributesInfo {
544 uint64_t OrigHighPc; ///< Value of AT_high_pc in the input DIE
545 int64_t PCOffset; ///< Offset to apply to PC addresses inside a function.
547 AttributesInfo() : OrigHighPc(0), PCOffset(0) {}
550 /// \brief Helper for cloneDIE.
551 unsigned cloneAttribute(DIE &Die, const DWARFDebugInfoEntryMinimal &InputDIE,
552 CompileUnit &U, const DWARFFormValue &Val,
553 const AttributeSpec AttrSpec, unsigned AttrSize,
554 AttributesInfo &AttrInfo);
556 /// \brief Helper for cloneDIE.
557 unsigned cloneStringAttribute(DIE &Die, AttributeSpec AttrSpec,
558 const DWARFFormValue &Val, const DWARFUnit &U);
560 /// \brief Helper for cloneDIE.
562 cloneDieReferenceAttribute(DIE &Die,
563 const DWARFDebugInfoEntryMinimal &InputDIE,
564 AttributeSpec AttrSpec, unsigned AttrSize,
565 const DWARFFormValue &Val, const DWARFUnit &U);
567 /// \brief Helper for cloneDIE.
568 unsigned cloneBlockAttribute(DIE &Die, AttributeSpec AttrSpec,
569 const DWARFFormValue &Val, unsigned AttrSize);
571 /// \brief Helper for cloneDIE.
572 unsigned cloneAddressAttribute(DIE &Die, AttributeSpec AttrSpec,
573 const DWARFFormValue &Val,
574 const CompileUnit &Unit, AttributesInfo &Info);
576 /// \brief Helper for cloneDIE.
577 unsigned cloneScalarAttribute(DIE &Die,
578 const DWARFDebugInfoEntryMinimal &InputDIE,
579 const DWARFUnit &U, AttributeSpec AttrSpec,
580 const DWARFFormValue &Val, unsigned AttrSize);
582 /// \brief Helper for cloneDIE.
583 bool applyValidRelocs(MutableArrayRef<char> Data, uint32_t BaseOffset,
584 bool isLittleEndian);
586 /// \brief Assign an abbreviation number to \p Abbrev
587 void AssignAbbrev(DIEAbbrev &Abbrev);
589 /// \brief FoldingSet that uniques the abbreviations.
590 FoldingSet<DIEAbbrev> AbbreviationsSet;
591 /// \brief Storage for the unique Abbreviations.
592 /// This is passed to AsmPrinter::emitDwarfAbbrevs(), thus it cannot
593 /// be changed to a vecot of unique_ptrs.
594 std::vector<DIEAbbrev *> Abbreviations;
596 /// \brief DIELoc objects that need to be destructed (but not freed!).
597 std::vector<DIELoc *> DIELocs;
598 /// \brief DIEBlock objects that need to be destructed (but not freed!).
599 std::vector<DIEBlock *> DIEBlocks;
600 /// \brief Allocator used for all the DIEValue objects.
601 BumpPtrAllocator DIEAlloc;
604 /// \defgroup Helpers Various helper methods.
607 const DWARFDebugInfoEntryMinimal *
608 resolveDIEReference(DWARFFormValue &RefValue, const DWARFUnit &Unit,
609 const DWARFDebugInfoEntryMinimal &DIE,
610 CompileUnit *&ReferencedCU);
612 CompileUnit *getUnitForOffset(unsigned Offset);
614 void reportWarning(const Twine &Warning, const DWARFUnit *Unit = nullptr,
615 const DWARFDebugInfoEntryMinimal *DIE = nullptr);
617 bool createStreamer(Triple TheTriple, StringRef OutputFilename);
621 std::string OutputFilename;
623 BinaryHolder BinHolder;
624 std::unique_ptr<DwarfStreamer> Streamer;
626 /// The units of the current debug map object.
627 std::vector<CompileUnit> Units;
629 /// The debug map object curently under consideration.
630 DebugMapObject *CurrentDebugObject;
632 /// \brief The Dwarf string pool
633 NonRelocatableStringpool StringPool;
636 /// \brief Similar to DWARFUnitSection::getUnitForOffset(), but
637 /// returning our CompileUnit object instead.
638 CompileUnit *DwarfLinker::getUnitForOffset(unsigned Offset) {
640 std::upper_bound(Units.begin(), Units.end(), Offset,
641 [](uint32_t LHS, const CompileUnit &RHS) {
642 return LHS < RHS.getOrigUnit().getNextUnitOffset();
644 return CU != Units.end() ? &*CU : nullptr;
647 /// \brief Resolve the DIE attribute reference that has been
648 /// extracted in \p RefValue. The resulting DIE migh be in another
649 /// CompileUnit which is stored into \p ReferencedCU.
650 /// \returns null if resolving fails for any reason.
651 const DWARFDebugInfoEntryMinimal *DwarfLinker::resolveDIEReference(
652 DWARFFormValue &RefValue, const DWARFUnit &Unit,
653 const DWARFDebugInfoEntryMinimal &DIE, CompileUnit *&RefCU) {
654 assert(RefValue.isFormClass(DWARFFormValue::FC_Reference));
655 uint64_t RefOffset = *RefValue.getAsReference(&Unit);
657 if ((RefCU = getUnitForOffset(RefOffset)))
658 if (const auto *RefDie = RefCU->getOrigUnit().getDIEForOffset(RefOffset))
661 reportWarning("could not find referenced DIE", &Unit, &DIE);
665 /// \brief Report a warning to the user, optionaly including
666 /// information about a specific \p DIE related to the warning.
667 void DwarfLinker::reportWarning(const Twine &Warning, const DWARFUnit *Unit,
668 const DWARFDebugInfoEntryMinimal *DIE) {
669 StringRef Context = "<debug map>";
670 if (CurrentDebugObject)
671 Context = CurrentDebugObject->getObjectFilename();
672 warn(Warning, Context);
674 if (!Options.Verbose || !DIE)
677 errs() << " in DIE:\n";
678 DIE->dump(errs(), const_cast<DWARFUnit *>(Unit), 0 /* RecurseDepth */,
682 bool DwarfLinker::createStreamer(Triple TheTriple, StringRef OutputFilename) {
683 if (Options.NoOutput)
686 Streamer = llvm::make_unique<DwarfStreamer>();
687 return Streamer->init(TheTriple, OutputFilename);
690 /// \brief Recursive helper to gather the child->parent relationships in the
691 /// original compile unit.
692 static void gatherDIEParents(const DWARFDebugInfoEntryMinimal *DIE,
693 unsigned ParentIdx, CompileUnit &CU) {
694 unsigned MyIdx = CU.getOrigUnit().getDIEIndex(DIE);
695 CU.getInfo(MyIdx).ParentIdx = ParentIdx;
697 if (DIE->hasChildren())
698 for (auto *Child = DIE->getFirstChild(); Child && !Child->isNULL();
699 Child = Child->getSibling())
700 gatherDIEParents(Child, MyIdx, CU);
703 static bool dieNeedsChildrenToBeMeaningful(uint32_t Tag) {
707 case dwarf::DW_TAG_subprogram:
708 case dwarf::DW_TAG_lexical_block:
709 case dwarf::DW_TAG_subroutine_type:
710 case dwarf::DW_TAG_structure_type:
711 case dwarf::DW_TAG_class_type:
712 case dwarf::DW_TAG_union_type:
715 llvm_unreachable("Invalid Tag");
718 void DwarfLinker::startDebugObject(DWARFContext &Dwarf) {
719 Units.reserve(Dwarf.getNumCompileUnits());
723 void DwarfLinker::endDebugObject() {
727 for (auto *Block : DIEBlocks)
729 for (auto *Loc : DIELocs)
737 /// \brief Iterate over the relocations of the given \p Section and
738 /// store the ones that correspond to debug map entries into the
739 /// ValidRelocs array.
740 void DwarfLinker::findValidRelocsMachO(const object::SectionRef &Section,
741 const object::MachOObjectFile &Obj,
742 const DebugMapObject &DMO) {
744 Section.getContents(Contents);
745 DataExtractor Data(Contents, Obj.isLittleEndian(), 0);
747 for (const object::RelocationRef &Reloc : Section.relocations()) {
748 object::DataRefImpl RelocDataRef = Reloc.getRawDataRefImpl();
749 MachO::any_relocation_info MachOReloc = Obj.getRelocation(RelocDataRef);
750 unsigned RelocSize = 1 << Obj.getAnyRelocationLength(MachOReloc);
752 if ((RelocSize != 4 && RelocSize != 8) || Reloc.getOffset(Offset64)) {
753 reportWarning(" unsupported relocation in debug_info section.");
756 uint32_t Offset = Offset64;
757 // Mach-o uses REL relocations, the addend is at the relocation offset.
758 uint64_t Addend = Data.getUnsigned(&Offset, RelocSize);
760 auto Sym = Reloc.getSymbol();
761 if (Sym != Obj.symbol_end()) {
762 StringRef SymbolName;
763 if (Sym->getName(SymbolName)) {
764 reportWarning("error getting relocation symbol name.");
767 if (const auto *Mapping = DMO.lookupSymbol(SymbolName))
768 ValidRelocs.emplace_back(Offset64, RelocSize, Addend, Mapping);
769 } else if (const auto *Mapping = DMO.lookupObjectAddress(Addend)) {
770 // Do not store the addend. The addend was the address of the
771 // symbol in the object file, the address in the binary that is
772 // stored in the debug map doesn't need to be offseted.
773 ValidRelocs.emplace_back(Offset64, RelocSize, 0, Mapping);
778 /// \brief Dispatch the valid relocation finding logic to the
779 /// appropriate handler depending on the object file format.
780 bool DwarfLinker::findValidRelocs(const object::SectionRef &Section,
781 const object::ObjectFile &Obj,
782 const DebugMapObject &DMO) {
783 // Dispatch to the right handler depending on the file type.
784 if (auto *MachOObj = dyn_cast<object::MachOObjectFile>(&Obj))
785 findValidRelocsMachO(Section, *MachOObj, DMO);
787 reportWarning(Twine("unsupported object file type: ") + Obj.getFileName());
789 if (ValidRelocs.empty())
792 // Sort the relocations by offset. We will walk the DIEs linearly in
793 // the file, this allows us to just keep an index in the relocation
794 // array that we advance during our walk, rather than resorting to
795 // some associative container. See DwarfLinker::NextValidReloc.
796 std::sort(ValidRelocs.begin(), ValidRelocs.end());
800 /// \brief Look for relocations in the debug_info section that match
801 /// entries in the debug map. These relocations will drive the Dwarf
802 /// link by indicating which DIEs refer to symbols present in the
804 /// \returns wether there are any valid relocations in the debug info.
805 bool DwarfLinker::findValidRelocsInDebugInfo(const object::ObjectFile &Obj,
806 const DebugMapObject &DMO) {
807 // Find the debug_info section.
808 for (const object::SectionRef &Section : Obj.sections()) {
809 StringRef SectionName;
810 Section.getName(SectionName);
811 SectionName = SectionName.substr(SectionName.find_first_not_of("._"));
812 if (SectionName != "debug_info")
814 return findValidRelocs(Section, Obj, DMO);
819 /// \brief Checks that there is a relocation against an actual debug
820 /// map entry between \p StartOffset and \p NextOffset.
822 /// This function must be called with offsets in strictly ascending
823 /// order because it never looks back at relocations it already 'went past'.
824 /// \returns true and sets Info.InDebugMap if it is the case.
825 bool DwarfLinker::hasValidRelocation(uint32_t StartOffset, uint32_t EndOffset,
826 CompileUnit::DIEInfo &Info) {
827 assert(NextValidReloc == 0 ||
828 StartOffset > ValidRelocs[NextValidReloc - 1].Offset);
829 if (NextValidReloc >= ValidRelocs.size())
832 uint64_t RelocOffset = ValidRelocs[NextValidReloc].Offset;
834 // We might need to skip some relocs that we didn't consider. For
835 // example the high_pc of a discarded DIE might contain a reloc that
836 // is in the list because it actually corresponds to the start of a
837 // function that is in the debug map.
838 while (RelocOffset < StartOffset && NextValidReloc < ValidRelocs.size() - 1)
839 RelocOffset = ValidRelocs[++NextValidReloc].Offset;
841 if (RelocOffset < StartOffset || RelocOffset >= EndOffset)
844 const auto &ValidReloc = ValidRelocs[NextValidReloc++];
846 outs() << "Found valid debug map entry: " << ValidReloc.Mapping->getKey()
847 << " " << format("\t%016" PRIx64 " => %016" PRIx64,
848 ValidReloc.Mapping->getValue().ObjectAddress,
849 ValidReloc.Mapping->getValue().BinaryAddress);
851 Info.AddrAdjust = int64_t(ValidReloc.Mapping->getValue().BinaryAddress) +
853 ValidReloc.Mapping->getValue().ObjectAddress;
854 Info.InDebugMap = true;
858 /// \brief Get the starting and ending (exclusive) offset for the
859 /// attribute with index \p Idx descibed by \p Abbrev. \p Offset is
860 /// supposed to point to the position of the first attribute described
862 /// \return [StartOffset, EndOffset) as a pair.
863 static std::pair<uint32_t, uint32_t>
864 getAttributeOffsets(const DWARFAbbreviationDeclaration *Abbrev, unsigned Idx,
865 unsigned Offset, const DWARFUnit &Unit) {
866 DataExtractor Data = Unit.getDebugInfoExtractor();
868 for (unsigned i = 0; i < Idx; ++i)
869 DWARFFormValue::skipValue(Abbrev->getFormByIndex(i), Data, &Offset, &Unit);
871 uint32_t End = Offset;
872 DWARFFormValue::skipValue(Abbrev->getFormByIndex(Idx), Data, &End, &Unit);
874 return std::make_pair(Offset, End);
877 /// \brief Check if a variable describing DIE should be kept.
878 /// \returns updated TraversalFlags.
879 unsigned DwarfLinker::shouldKeepVariableDIE(
880 const DWARFDebugInfoEntryMinimal &DIE, CompileUnit &Unit,
881 CompileUnit::DIEInfo &MyInfo, unsigned Flags) {
882 const auto *Abbrev = DIE.getAbbreviationDeclarationPtr();
884 // Global variables with constant value can always be kept.
885 if (!(Flags & TF_InFunctionScope) &&
886 Abbrev->findAttributeIndex(dwarf::DW_AT_const_value) != -1U) {
887 MyInfo.InDebugMap = true;
888 return Flags | TF_Keep;
891 uint32_t LocationIdx = Abbrev->findAttributeIndex(dwarf::DW_AT_location);
892 if (LocationIdx == -1U)
895 uint32_t Offset = DIE.getOffset() + getULEB128Size(Abbrev->getCode());
896 const DWARFUnit &OrigUnit = Unit.getOrigUnit();
897 uint32_t LocationOffset, LocationEndOffset;
898 std::tie(LocationOffset, LocationEndOffset) =
899 getAttributeOffsets(Abbrev, LocationIdx, Offset, OrigUnit);
901 // See if there is a relocation to a valid debug map entry inside
902 // this variable's location. The order is important here. We want to
903 // always check in the variable has a valid relocation, so that the
904 // DIEInfo is filled. However, we don't want a static variable in a
905 // function to force us to keep the enclosing function.
906 if (!hasValidRelocation(LocationOffset, LocationEndOffset, MyInfo) ||
907 (Flags & TF_InFunctionScope))
911 DIE.dump(outs(), const_cast<DWARFUnit *>(&OrigUnit), 0, 8 /* Indent */);
913 return Flags | TF_Keep;
916 /// \brief Check if a function describing DIE should be kept.
917 /// \returns updated TraversalFlags.
918 unsigned DwarfLinker::shouldKeepSubprogramDIE(
919 const DWARFDebugInfoEntryMinimal &DIE, CompileUnit &Unit,
920 CompileUnit::DIEInfo &MyInfo, unsigned Flags) {
921 const auto *Abbrev = DIE.getAbbreviationDeclarationPtr();
923 Flags |= TF_InFunctionScope;
925 uint32_t LowPcIdx = Abbrev->findAttributeIndex(dwarf::DW_AT_low_pc);
929 uint32_t Offset = DIE.getOffset() + getULEB128Size(Abbrev->getCode());
930 const DWARFUnit &OrigUnit = Unit.getOrigUnit();
931 uint32_t LowPcOffset, LowPcEndOffset;
932 std::tie(LowPcOffset, LowPcEndOffset) =
933 getAttributeOffsets(Abbrev, LowPcIdx, Offset, OrigUnit);
936 DIE.getAttributeValueAsAddress(&OrigUnit, dwarf::DW_AT_low_pc, -1ULL);
937 assert(LowPc != -1ULL && "low_pc attribute is not an address.");
938 if (LowPc == -1ULL ||
939 !hasValidRelocation(LowPcOffset, LowPcEndOffset, MyInfo))
943 DIE.dump(outs(), const_cast<DWARFUnit *>(&OrigUnit), 0, 8 /* Indent */);
945 return Flags | TF_Keep;
948 /// \brief Check if a DIE should be kept.
949 /// \returns updated TraversalFlags.
950 unsigned DwarfLinker::shouldKeepDIE(const DWARFDebugInfoEntryMinimal &DIE,
952 CompileUnit::DIEInfo &MyInfo,
954 switch (DIE.getTag()) {
955 case dwarf::DW_TAG_constant:
956 case dwarf::DW_TAG_variable:
957 return shouldKeepVariableDIE(DIE, Unit, MyInfo, Flags);
958 case dwarf::DW_TAG_subprogram:
959 return shouldKeepSubprogramDIE(DIE, Unit, MyInfo, Flags);
960 case dwarf::DW_TAG_module:
961 case dwarf::DW_TAG_imported_module:
962 case dwarf::DW_TAG_imported_declaration:
963 case dwarf::DW_TAG_imported_unit:
964 // We always want to keep these.
965 return Flags | TF_Keep;
971 /// \brief Mark the passed DIE as well as all the ones it depends on
974 /// This function is called by lookForDIEsToKeep on DIEs that are
975 /// newly discovered to be needed in the link. It recursively calls
976 /// back to lookForDIEsToKeep while adding TF_DependencyWalk to the
977 /// TraversalFlags to inform it that it's not doing the primary DIE
979 void DwarfLinker::keepDIEAndDenpendencies(const DWARFDebugInfoEntryMinimal &DIE,
980 CompileUnit::DIEInfo &MyInfo,
981 const DebugMapObject &DMO,
982 CompileUnit &CU, unsigned Flags) {
983 const DWARFUnit &Unit = CU.getOrigUnit();
986 // First mark all the parent chain as kept.
987 unsigned AncestorIdx = MyInfo.ParentIdx;
988 while (!CU.getInfo(AncestorIdx).Keep) {
989 lookForDIEsToKeep(*Unit.getDIEAtIndex(AncestorIdx), DMO, CU,
990 TF_ParentWalk | TF_Keep | TF_DependencyWalk);
991 AncestorIdx = CU.getInfo(AncestorIdx).ParentIdx;
994 // Then we need to mark all the DIEs referenced by this DIE's
995 // attributes as kept.
996 DataExtractor Data = Unit.getDebugInfoExtractor();
997 const auto *Abbrev = DIE.getAbbreviationDeclarationPtr();
998 uint32_t Offset = DIE.getOffset() + getULEB128Size(Abbrev->getCode());
1000 // Mark all DIEs referenced through atttributes as kept.
1001 for (const auto &AttrSpec : Abbrev->attributes()) {
1002 DWARFFormValue Val(AttrSpec.Form);
1004 if (!Val.isFormClass(DWARFFormValue::FC_Reference)) {
1005 DWARFFormValue::skipValue(AttrSpec.Form, Data, &Offset, &Unit);
1009 Val.extractValue(Data, &Offset, &Unit);
1010 CompileUnit *ReferencedCU;
1011 if (const auto *RefDIE = resolveDIEReference(Val, Unit, DIE, ReferencedCU))
1012 lookForDIEsToKeep(*RefDIE, DMO, *ReferencedCU,
1013 TF_Keep | TF_DependencyWalk);
1017 /// \brief Recursively walk the \p DIE tree and look for DIEs to
1018 /// keep. Store that information in \p CU's DIEInfo.
1020 /// This function is the entry point of the DIE selection
1021 /// algorithm. It is expected to walk the DIE tree in file order and
1022 /// (though the mediation of its helper) call hasValidRelocation() on
1023 /// each DIE that might be a 'root DIE' (See DwarfLinker class
1025 /// While walking the dependencies of root DIEs, this function is
1026 /// also called, but during these dependency walks the file order is
1027 /// not respected. The TF_DependencyWalk flag tells us which kind of
1028 /// traversal we are currently doing.
1029 void DwarfLinker::lookForDIEsToKeep(const DWARFDebugInfoEntryMinimal &DIE,
1030 const DebugMapObject &DMO, CompileUnit &CU,
1032 unsigned Idx = CU.getOrigUnit().getDIEIndex(&DIE);
1033 CompileUnit::DIEInfo &MyInfo = CU.getInfo(Idx);
1034 bool AlreadyKept = MyInfo.Keep;
1036 // If the Keep flag is set, we are marking a required DIE's
1037 // dependencies. If our target is already marked as kept, we're all
1039 if ((Flags & TF_DependencyWalk) && AlreadyKept)
1042 // We must not call shouldKeepDIE while called from keepDIEAndDenpendencies,
1043 // because it would screw up the relocation finding logic.
1044 if (!(Flags & TF_DependencyWalk))
1045 Flags = shouldKeepDIE(DIE, CU, MyInfo, Flags);
1047 // If it is a newly kept DIE mark it as well as all its dependencies as kept.
1048 if (!AlreadyKept && (Flags & TF_Keep))
1049 keepDIEAndDenpendencies(DIE, MyInfo, DMO, CU, Flags);
1051 // The TF_ParentWalk flag tells us that we are currently walking up
1052 // the parent chain of a required DIE, and we don't want to mark all
1053 // the children of the parents as kept (consider for example a
1054 // DW_TAG_namespace node in the parent chain). There are however a
1055 // set of DIE types for which we want to ignore that directive and still
1056 // walk their children.
1057 if (dieNeedsChildrenToBeMeaningful(DIE.getTag()))
1058 Flags &= ~TF_ParentWalk;
1060 if (!DIE.hasChildren() || (Flags & TF_ParentWalk))
1063 for (auto *Child = DIE.getFirstChild(); Child && !Child->isNULL();
1064 Child = Child->getSibling())
1065 lookForDIEsToKeep(*Child, DMO, CU, Flags);
1068 /// \brief Assign an abbreviation numer to \p Abbrev.
1070 /// Our DIEs get freed after every DebugMapObject has been processed,
1071 /// thus the FoldingSet we use to unique DIEAbbrevs cannot refer to
1072 /// the instances hold by the DIEs. When we encounter an abbreviation
1073 /// that we don't know, we create a permanent copy of it.
1074 void DwarfLinker::AssignAbbrev(DIEAbbrev &Abbrev) {
1075 // Check the set for priors.
1076 FoldingSetNodeID ID;
1079 DIEAbbrev *InSet = AbbreviationsSet.FindNodeOrInsertPos(ID, InsertToken);
1081 // If it's newly added.
1083 // Assign existing abbreviation number.
1084 Abbrev.setNumber(InSet->getNumber());
1086 // Add to abbreviation list.
1087 Abbreviations.push_back(
1088 new DIEAbbrev(Abbrev.getTag(), Abbrev.hasChildren()));
1089 for (const auto &Attr : Abbrev.getData())
1090 Abbreviations.back()->AddAttribute(Attr.getAttribute(), Attr.getForm());
1091 AbbreviationsSet.InsertNode(Abbreviations.back(), InsertToken);
1092 // Assign the unique abbreviation number.
1093 Abbrev.setNumber(Abbreviations.size());
1094 Abbreviations.back()->setNumber(Abbreviations.size());
1098 /// \brief Clone a string attribute described by \p AttrSpec and add
1100 /// \returns the size of the new attribute.
1101 unsigned DwarfLinker::cloneStringAttribute(DIE &Die, AttributeSpec AttrSpec,
1102 const DWARFFormValue &Val,
1103 const DWARFUnit &U) {
1104 // Switch everything to out of line strings.
1105 const char *String = *Val.getAsCString(&U);
1106 unsigned Offset = StringPool.getStringOffset(String);
1107 Die.addValue(dwarf::Attribute(AttrSpec.Attr), dwarf::DW_FORM_strp,
1108 new (DIEAlloc) DIEInteger(Offset));
1112 /// \brief Clone an attribute referencing another DIE and add
1114 /// \returns the size of the new attribute.
1115 unsigned DwarfLinker::cloneDieReferenceAttribute(
1116 DIE &Die, const DWARFDebugInfoEntryMinimal &InputDIE,
1117 AttributeSpec AttrSpec, unsigned AttrSize, const DWARFFormValue &Val,
1118 const DWARFUnit &U) {
1119 uint32_t Ref = *Val.getAsReference(&U);
1120 DIE *NewRefDie = nullptr;
1121 CompileUnit *RefUnit = nullptr;
1122 const DWARFDebugInfoEntryMinimal *RefDie = nullptr;
1124 if (!(RefUnit = getUnitForOffset(Ref)) ||
1125 !(RefDie = RefUnit->getOrigUnit().getDIEForOffset(Ref))) {
1126 const char *AttributeString = dwarf::AttributeString(AttrSpec.Attr);
1127 if (!AttributeString)
1128 AttributeString = "DW_AT_???";
1129 reportWarning(Twine("Missing DIE for ref in attribute ") + AttributeString +
1135 unsigned Idx = RefUnit->getOrigUnit().getDIEIndex(RefDie);
1136 CompileUnit::DIEInfo &RefInfo = RefUnit->getInfo(Idx);
1137 if (!RefInfo.Clone) {
1138 assert(Ref > InputDIE.getOffset());
1139 // We haven't cloned this DIE yet. Just create an empty one and
1140 // store it. It'll get really cloned when we process it.
1141 RefInfo.Clone = new DIE(dwarf::Tag(RefDie->getTag()));
1143 NewRefDie = RefInfo.Clone;
1145 if (AttrSpec.Form == dwarf::DW_FORM_ref_addr) {
1146 // We cannot currently rely on a DIEEntry to emit ref_addr
1147 // references, because the implementation calls back to DwarfDebug
1148 // to find the unit offset. (We don't have a DwarfDebug)
1149 // FIXME: we should be able to design DIEEntry reliance on
1152 if (Ref < InputDIE.getOffset()) {
1153 // We must have already cloned that DIE.
1154 uint32_t NewRefOffset =
1155 RefUnit->getStartOffset() + NewRefDie->getOffset();
1156 Attr = new (DIEAlloc) DIEInteger(NewRefOffset);
1158 // A forward reference. Note and fixup later.
1159 Attr = new (DIEAlloc) DIEInteger(0xBADDEF);
1160 RefUnit->noteForwardReference(NewRefDie, Attr);
1162 Die.addValue(dwarf::Attribute(AttrSpec.Attr), dwarf::DW_FORM_ref_addr,
1167 Die.addValue(dwarf::Attribute(AttrSpec.Attr), dwarf::Form(AttrSpec.Form),
1168 new (DIEAlloc) DIEEntry(*NewRefDie));
1172 /// \brief Clone an attribute of block form (locations, constants) and add
1174 /// \returns the size of the new attribute.
1175 unsigned DwarfLinker::cloneBlockAttribute(DIE &Die, AttributeSpec AttrSpec,
1176 const DWARFFormValue &Val,
1177 unsigned AttrSize) {
1180 DIELoc *Loc = nullptr;
1181 DIEBlock *Block = nullptr;
1182 // Just copy the block data over.
1183 if (AttrSpec.Attr == dwarf::DW_FORM_exprloc) {
1184 Loc = new (DIEAlloc) DIELoc();
1185 DIELocs.push_back(Loc);
1187 Block = new (DIEAlloc) DIEBlock();
1188 DIEBlocks.push_back(Block);
1190 Attr = Loc ? static_cast<DIE *>(Loc) : static_cast<DIE *>(Block);
1191 Value = Loc ? static_cast<DIEValue *>(Loc) : static_cast<DIEValue *>(Block);
1192 ArrayRef<uint8_t> Bytes = *Val.getAsBlock();
1193 for (auto Byte : Bytes)
1194 Attr->addValue(static_cast<dwarf::Attribute>(0), dwarf::DW_FORM_data1,
1195 new (DIEAlloc) DIEInteger(Byte));
1196 // FIXME: If DIEBlock and DIELoc just reuses the Size field of
1197 // the DIE class, this if could be replaced by
1198 // Attr->setSize(Bytes.size()).
1201 Loc->ComputeSize(&Streamer->getAsmPrinter());
1203 Block->ComputeSize(&Streamer->getAsmPrinter());
1205 Die.addValue(dwarf::Attribute(AttrSpec.Attr), dwarf::Form(AttrSpec.Form),
1210 /// \brief Clone an address attribute and add it to \p Die.
1211 /// \returns the size of the new attribute.
1212 unsigned DwarfLinker::cloneAddressAttribute(DIE &Die, AttributeSpec AttrSpec,
1213 const DWARFFormValue &Val,
1214 const CompileUnit &Unit,
1215 AttributesInfo &Info) {
1216 int64_t Addr = *Val.getAsAddress(&Unit.getOrigUnit());
1217 if (AttrSpec.Attr == dwarf::DW_AT_low_pc) {
1218 if (Die.getTag() == dwarf::DW_TAG_inlined_subroutine ||
1219 Die.getTag() == dwarf::DW_TAG_lexical_block)
1220 Addr += Info.PCOffset;
1221 } else if (AttrSpec.Attr == dwarf::DW_AT_high_pc) {
1222 // If we have a high_pc recorded for the input DIE, use
1223 // it. Otherwise (when no relocations where applied) just use the
1224 // one we just decoded.
1225 Addr = (Info.OrigHighPc ? Info.OrigHighPc : Addr) + Info.PCOffset;
1228 Die.addValue(static_cast<dwarf::Attribute>(AttrSpec.Attr),
1229 static_cast<dwarf::Form>(AttrSpec.Form),
1230 new (DIEAlloc) DIEInteger(Addr));
1231 return Unit.getOrigUnit().getAddressByteSize();
1234 /// \brief Clone a scalar attribute and add it to \p Die.
1235 /// \returns the size of the new attribute.
1236 unsigned DwarfLinker::cloneScalarAttribute(
1237 DIE &Die, const DWARFDebugInfoEntryMinimal &InputDIE, const DWARFUnit &U,
1238 AttributeSpec AttrSpec, const DWARFFormValue &Val, unsigned AttrSize) {
1240 if (AttrSpec.Form == dwarf::DW_FORM_sec_offset)
1241 Value = *Val.getAsSectionOffset();
1242 else if (AttrSpec.Form == dwarf::DW_FORM_sdata)
1243 Value = *Val.getAsSignedConstant();
1244 else if (auto OptionalValue = Val.getAsUnsignedConstant())
1245 Value = *OptionalValue;
1247 reportWarning("Unsupported scalar attribute form. Dropping attribute.", &U,
1251 Die.addValue(dwarf::Attribute(AttrSpec.Attr), dwarf::Form(AttrSpec.Form),
1252 new (DIEAlloc) DIEInteger(Value));
1256 /// \brief Clone \p InputDIE's attribute described by \p AttrSpec with
1257 /// value \p Val, and add it to \p Die.
1258 /// \returns the size of the cloned attribute.
1259 unsigned DwarfLinker::cloneAttribute(DIE &Die,
1260 const DWARFDebugInfoEntryMinimal &InputDIE,
1262 const DWARFFormValue &Val,
1263 const AttributeSpec AttrSpec,
1264 unsigned AttrSize, AttributesInfo &Info) {
1265 const DWARFUnit &U = Unit.getOrigUnit();
1267 switch (AttrSpec.Form) {
1268 case dwarf::DW_FORM_strp:
1269 case dwarf::DW_FORM_string:
1270 return cloneStringAttribute(Die, AttrSpec, Val, U);
1271 case dwarf::DW_FORM_ref_addr:
1272 case dwarf::DW_FORM_ref1:
1273 case dwarf::DW_FORM_ref2:
1274 case dwarf::DW_FORM_ref4:
1275 case dwarf::DW_FORM_ref8:
1276 return cloneDieReferenceAttribute(Die, InputDIE, AttrSpec, AttrSize, Val,
1278 case dwarf::DW_FORM_block:
1279 case dwarf::DW_FORM_block1:
1280 case dwarf::DW_FORM_block2:
1281 case dwarf::DW_FORM_block4:
1282 case dwarf::DW_FORM_exprloc:
1283 return cloneBlockAttribute(Die, AttrSpec, Val, AttrSize);
1284 case dwarf::DW_FORM_addr:
1285 return cloneAddressAttribute(Die, AttrSpec, Val, Unit, Info);
1286 case dwarf::DW_FORM_data1:
1287 case dwarf::DW_FORM_data2:
1288 case dwarf::DW_FORM_data4:
1289 case dwarf::DW_FORM_data8:
1290 case dwarf::DW_FORM_udata:
1291 case dwarf::DW_FORM_sdata:
1292 case dwarf::DW_FORM_sec_offset:
1293 case dwarf::DW_FORM_flag:
1294 case dwarf::DW_FORM_flag_present:
1295 return cloneScalarAttribute(Die, InputDIE, U, AttrSpec, Val, AttrSize);
1297 reportWarning("Unsupported attribute form in cloneAttribute. Dropping.", &U,
1304 /// \brief Apply the valid relocations found by findValidRelocs() to
1305 /// the buffer \p Data, taking into account that Data is at \p BaseOffset
1306 /// in the debug_info section.
1308 /// Like for findValidRelocs(), this function must be called with
1309 /// monotonic \p BaseOffset values.
1311 /// \returns wether any reloc has been applied.
1312 bool DwarfLinker::applyValidRelocs(MutableArrayRef<char> Data,
1313 uint32_t BaseOffset, bool isLittleEndian) {
1314 assert((NextValidReloc == 0 ||
1315 BaseOffset > ValidRelocs[NextValidReloc - 1].Offset) &&
1316 "BaseOffset should only be increasing.");
1317 if (NextValidReloc >= ValidRelocs.size())
1320 // Skip relocs that haven't been applied.
1321 while (NextValidReloc < ValidRelocs.size() &&
1322 ValidRelocs[NextValidReloc].Offset < BaseOffset)
1325 bool Applied = false;
1326 uint64_t EndOffset = BaseOffset + Data.size();
1327 while (NextValidReloc < ValidRelocs.size() &&
1328 ValidRelocs[NextValidReloc].Offset >= BaseOffset &&
1329 ValidRelocs[NextValidReloc].Offset < EndOffset) {
1330 const auto &ValidReloc = ValidRelocs[NextValidReloc++];
1331 assert(ValidReloc.Offset - BaseOffset < Data.size());
1332 assert(ValidReloc.Offset - BaseOffset + ValidReloc.Size <= Data.size());
1334 uint64_t Value = ValidReloc.Mapping->getValue().BinaryAddress;
1335 Value += ValidReloc.Addend;
1336 for (unsigned i = 0; i != ValidReloc.Size; ++i) {
1337 unsigned Index = isLittleEndian ? i : (ValidReloc.Size - i - 1);
1338 Buf[i] = uint8_t(Value >> (Index * 8));
1340 assert(ValidReloc.Size <= sizeof(Buf));
1341 memcpy(&Data[ValidReloc.Offset - BaseOffset], Buf, ValidReloc.Size);
1348 /// \brief Recursively clone \p InputDIE's subtrees that have been
1349 /// selected to appear in the linked output.
1351 /// \param OutOffset is the Offset where the newly created DIE will
1352 /// lie in the linked compile unit.
1354 /// \returns the cloned DIE object or null if nothing was selected.
1355 DIE *DwarfLinker::cloneDIE(const DWARFDebugInfoEntryMinimal &InputDIE,
1356 CompileUnit &Unit, int64_t PCOffset,
1357 uint32_t OutOffset) {
1358 DWARFUnit &U = Unit.getOrigUnit();
1359 unsigned Idx = U.getDIEIndex(&InputDIE);
1360 CompileUnit::DIEInfo &Info = Unit.getInfo(Idx);
1362 // Should the DIE appear in the output?
1363 if (!Unit.getInfo(Idx).Keep)
1366 uint32_t Offset = InputDIE.getOffset();
1367 // The DIE might have been already created by a forward reference
1368 // (see cloneDieReferenceAttribute()).
1369 DIE *Die = Info.Clone;
1371 Die = Info.Clone = new DIE(dwarf::Tag(InputDIE.getTag()));
1372 assert(Die->getTag() == InputDIE.getTag());
1373 Die->setOffset(OutOffset);
1375 // Extract and clone every attribute.
1376 DataExtractor Data = U.getDebugInfoExtractor();
1377 uint32_t NextOffset = U.getDIEAtIndex(Idx + 1)->getOffset();
1378 AttributesInfo AttrInfo;
1380 // We could copy the data only if we need to aply a relocation to
1381 // it. After testing, it seems there is no performance downside to
1382 // doing the copy unconditionally, and it makes the code simpler.
1383 SmallString<40> DIECopy(Data.getData().substr(Offset, NextOffset - Offset));
1384 Data = DataExtractor(DIECopy, Data.isLittleEndian(), Data.getAddressSize());
1385 // Modify the copy with relocated addresses.
1386 if (applyValidRelocs(DIECopy, Offset, Data.isLittleEndian())) {
1387 // If we applied relocations, we store the value of high_pc that was
1388 // potentially stored in the input DIE. If high_pc is an address
1389 // (Dwarf version == 2), then it might have been relocated to a
1390 // totally unrelated value (because the end address in the object
1391 // file might be start address of another function which got moved
1392 // independantly by the linker). The computation of the actual
1393 // high_pc value is done in cloneAddressAttribute().
1394 AttrInfo.OrigHighPc =
1395 InputDIE.getAttributeValueAsAddress(&U, dwarf::DW_AT_high_pc, 0);
1398 // Reset the Offset to 0 as we will be working on the local copy of
1402 const auto *Abbrev = InputDIE.getAbbreviationDeclarationPtr();
1403 Offset += getULEB128Size(Abbrev->getCode());
1405 // We are entering a subprogram. Get and propagate the PCOffset.
1406 if (Die->getTag() == dwarf::DW_TAG_subprogram)
1407 PCOffset = Info.AddrAdjust;
1408 AttrInfo.PCOffset = PCOffset;
1410 for (const auto &AttrSpec : Abbrev->attributes()) {
1411 DWARFFormValue Val(AttrSpec.Form);
1412 uint32_t AttrSize = Offset;
1413 Val.extractValue(Data, &Offset, &U);
1414 AttrSize = Offset - AttrSize;
1417 cloneAttribute(*Die, InputDIE, Unit, Val, AttrSpec, AttrSize, AttrInfo);
1420 DIEAbbrev &NewAbbrev = Die->getAbbrev();
1421 // If a scope DIE is kept, we must have kept at least one child. If
1422 // it's not the case, we'll just be emitting one wasteful end of
1423 // children marker, but things won't break.
1424 if (InputDIE.hasChildren())
1425 NewAbbrev.setChildrenFlag(dwarf::DW_CHILDREN_yes);
1426 // Assign a permanent abbrev number
1427 AssignAbbrev(Die->getAbbrev());
1429 // Add the size of the abbreviation number to the output offset.
1430 OutOffset += getULEB128Size(Die->getAbbrevNumber());
1432 if (!Abbrev->hasChildren()) {
1434 Die->setSize(OutOffset - Die->getOffset());
1438 // Recursively clone children.
1439 for (auto *Child = InputDIE.getFirstChild(); Child && !Child->isNULL();
1440 Child = Child->getSibling()) {
1441 if (DIE *Clone = cloneDIE(*Child, Unit, PCOffset, OutOffset)) {
1442 Die->addChild(std::unique_ptr<DIE>(Clone));
1443 OutOffset = Clone->getOffset() + Clone->getSize();
1447 // Account for the end of children marker.
1448 OutOffset += sizeof(int8_t);
1450 Die->setSize(OutOffset - Die->getOffset());
1454 bool DwarfLinker::link(const DebugMap &Map) {
1456 if (Map.begin() == Map.end()) {
1457 errs() << "Empty debug map.\n";
1461 if (!createStreamer(Map.getTriple(), OutputFilename))
1464 // Size of the DIEs (and headers) generated for the linked output.
1465 uint64_t OutputDebugInfoSize = 0;
1467 for (const auto &Obj : Map.objects()) {
1468 CurrentDebugObject = Obj.get();
1470 if (Options.Verbose)
1471 outs() << "DEBUG MAP OBJECT: " << Obj->getObjectFilename() << "\n";
1472 auto ErrOrObj = BinHolder.GetObjectFile(Obj->getObjectFilename());
1473 if (std::error_code EC = ErrOrObj.getError()) {
1474 reportWarning(Twine(Obj->getObjectFilename()) + ": " + EC.message());
1478 // Look for relocations that correspond to debug map entries.
1479 if (!findValidRelocsInDebugInfo(*ErrOrObj, *Obj)) {
1480 if (Options.Verbose)
1481 outs() << "No valid relocations found. Skipping.\n";
1485 // Setup access to the debug info.
1486 DWARFContextInMemory DwarfContext(*ErrOrObj);
1487 startDebugObject(DwarfContext);
1489 // In a first phase, just read in the debug info and store the DIE
1490 // parent links that we will use during the next phase.
1491 for (const auto &CU : DwarfContext.compile_units()) {
1492 auto *CUDie = CU->getCompileUnitDIE(false);
1493 if (Options.Verbose) {
1494 outs() << "Input compilation unit:";
1495 CUDie->dump(outs(), CU.get(), 0);
1497 Units.emplace_back(*CU);
1498 gatherDIEParents(CUDie, 0, Units.back());
1501 // Then mark all the DIEs that need to be present in the linked
1502 // output and collect some information about them. Note that this
1503 // loop can not be merged with the previous one becaue cross-cu
1504 // references require the ParentIdx to be setup for every CU in
1505 // the object file before calling this.
1506 for (auto &CurrentUnit : Units)
1507 lookForDIEsToKeep(*CurrentUnit.getOrigUnit().getCompileUnitDIE(), *Obj,
1510 // The calls to applyValidRelocs inside cloneDIE will walk the
1511 // reloc array again (in the same way findValidRelocsInDebugInfo()
1512 // did). We need to reset the NextValidReloc index to the beginning.
1515 // Construct the output DIE tree by cloning the DIEs we chose to
1516 // keep above. If there are no valid relocs, then there's nothing
1518 if (!ValidRelocs.empty())
1519 for (auto &CurrentUnit : Units) {
1520 const auto *InputDIE = CurrentUnit.getOrigUnit().getCompileUnitDIE();
1521 CurrentUnit.setStartOffset(OutputDebugInfoSize);
1522 DIE *OutputDIE = cloneDIE(*InputDIE, CurrentUnit, 0 /* PCOffset */,
1523 11 /* Unit Header size */);
1524 CurrentUnit.setOutputUnitDIE(OutputDIE);
1525 OutputDebugInfoSize = CurrentUnit.computeNextUnitOffset();
1528 // Emit all the compile unit's debug information.
1529 if (!ValidRelocs.empty() && !Options.NoOutput)
1530 for (auto &CurrentUnit : Units) {
1531 CurrentUnit.fixupForwardReferences();
1532 Streamer->emitCompileUnitHeader(CurrentUnit);
1533 if (!CurrentUnit.getOutputUnitDIE())
1535 Streamer->emitDIE(*CurrentUnit.getOutputUnitDIE());
1538 // Clean-up before starting working on the next object.
1542 // Emit everything that's global.
1543 if (!Options.NoOutput) {
1544 Streamer->emitAbbrevs(Abbreviations);
1545 Streamer->emitStrings(StringPool);
1548 return Options.NoOutput ? true : Streamer->finish();
1552 bool linkDwarf(StringRef OutputFilename, const DebugMap &DM,
1553 const LinkOptions &Options) {
1554 DwarfLinker Linker(OutputFilename, Options);
1555 return Linker.link(DM);