1 //===- tools/dsymutil/DwarfLinker.cpp - Dwarf debug info linker -----------===//
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 #include "BinaryHolder.h"
13 #include "llvm/ADT/IntervalMap.h"
14 #include "llvm/ADT/StringMap.h"
15 #include "llvm/ADT/STLExtras.h"
16 #include "llvm/CodeGen/AsmPrinter.h"
17 #include "llvm/CodeGen/DIE.h"
18 #include "llvm/DebugInfo/DWARF/DWARFContext.h"
19 #include "llvm/DebugInfo/DWARF/DWARFDebugInfoEntry.h"
20 #include "llvm/DebugInfo/DWARF/DWARFFormValue.h"
21 #include "llvm/MC/MCAsmBackend.h"
22 #include "llvm/MC/MCAsmInfo.h"
23 #include "llvm/MC/MCContext.h"
24 #include "llvm/MC/MCCodeEmitter.h"
25 #include "llvm/MC/MCDwarf.h"
26 #include "llvm/MC/MCInstrInfo.h"
27 #include "llvm/MC/MCObjectFileInfo.h"
28 #include "llvm/MC/MCRegisterInfo.h"
29 #include "llvm/MC/MCStreamer.h"
30 #include "llvm/Object/MachO.h"
31 #include "llvm/Support/Dwarf.h"
32 #include "llvm/Support/LEB128.h"
33 #include "llvm/Support/TargetRegistry.h"
34 #include "llvm/Target/TargetMachine.h"
35 #include "llvm/Target/TargetOptions.h"
44 void warn(const Twine &Warning, const Twine &Context) {
45 errs() << Twine("while processing ") + Context + ":\n";
46 errs() << Twine("warning: ") + Warning + "\n";
49 bool error(const Twine &Error, const Twine &Context) {
50 errs() << Twine("while processing ") + Context + ":\n";
51 errs() << Twine("error: ") + Error + "\n";
55 template <typename KeyT, typename ValT>
56 using HalfOpenIntervalMap =
57 IntervalMap<KeyT, ValT, IntervalMapImpl::NodeSizer<KeyT, ValT>::LeafSize,
58 IntervalMapHalfOpenInfo<KeyT>>;
60 typedef HalfOpenIntervalMap<uint64_t, int64_t> FunctionIntervals;
62 /// \brief Stores all information relating to a compile unit, be it in
63 /// its original instance in the object file to its brand new cloned
64 /// and linked DIE tree.
67 /// \brief Information gathered about a DIE in the object file.
69 int64_t AddrAdjust; ///< Address offset to apply to the described entity.
70 DIE *Clone; ///< Cloned version of that DIE.
71 uint32_t ParentIdx; ///< The index of this DIE's parent.
72 bool Keep; ///< Is the DIE part of the linked output?
73 bool InDebugMap; ///< Was this DIE's entity found in the map?
76 CompileUnit(DWARFUnit &OrigUnit, unsigned ID)
77 : OrigUnit(OrigUnit), ID(ID), LowPc(UINT64_MAX), HighPc(0), RangeAlloc(),
78 Ranges(RangeAlloc), UnitRangeAttribute(nullptr) {
79 Info.resize(OrigUnit.getNumDIEs());
82 CompileUnit(CompileUnit &&RHS)
83 : OrigUnit(RHS.OrigUnit), Info(std::move(RHS.Info)),
84 CUDie(std::move(RHS.CUDie)), StartOffset(RHS.StartOffset),
85 NextUnitOffset(RHS.NextUnitOffset), RangeAlloc(), Ranges(RangeAlloc) {
86 // The CompileUnit container has been 'reserve()'d with the right
87 // size. We cannot move the IntervalMap anyway.
88 llvm_unreachable("CompileUnits should not be moved.");
91 DWARFUnit &getOrigUnit() const { return OrigUnit; }
93 unsigned getUniqueID() const { return ID; }
95 DIE *getOutputUnitDIE() const { return CUDie.get(); }
96 void setOutputUnitDIE(DIE *Die) { CUDie.reset(Die); }
98 DIEInfo &getInfo(unsigned Idx) { return Info[Idx]; }
99 const DIEInfo &getInfo(unsigned Idx) const { return Info[Idx]; }
101 uint64_t getStartOffset() const { return StartOffset; }
102 uint64_t getNextUnitOffset() const { return NextUnitOffset; }
103 void setStartOffset(uint64_t DebugInfoSize) { StartOffset = DebugInfoSize; }
105 uint64_t getLowPc() const { return LowPc; }
106 uint64_t getHighPc() const { return HighPc; }
108 DIEInteger *getUnitRangesAttribute() const { return UnitRangeAttribute; }
109 const FunctionIntervals &getFunctionRanges() const { return Ranges; }
110 const std::vector<DIEInteger *> &getRangesAttributes() const {
111 return RangeAttributes;
114 const std::vector<std::pair<DIEInteger *, int64_t>> &
115 getLocationAttributes() const {
116 return LocationAttributes;
119 /// \brief Compute the end offset for this unit. Must be
120 /// called after the CU's DIEs have been cloned.
121 /// \returns the next unit offset (which is also the current
122 /// debug_info section size).
123 uint64_t computeNextUnitOffset();
125 /// \brief Keep track of a forward reference to DIE \p Die in \p
126 /// RefUnit by \p Attr. The attribute should be fixed up later to
127 /// point to the absolute offset of \p Die in the debug_info section.
128 void noteForwardReference(DIE *Die, const CompileUnit *RefUnit,
131 /// \brief Apply all fixups recored by noteForwardReference().
132 void fixupForwardReferences();
134 /// \brief Add a function range [\p LowPC, \p HighPC) that is
135 /// relocatad by applying offset \p PCOffset.
136 void addFunctionRange(uint64_t LowPC, uint64_t HighPC, int64_t PCOffset);
138 /// \brief Keep track of a DW_AT_range attribute that we will need to
140 void noteRangeAttribute(const DIE &Die, DIEInteger *Attr);
142 /// \brief Keep track of a location attribute pointing to a location
143 /// list in the debug_loc section.
144 void noteLocationAttribute(DIEInteger *Attr, int64_t PcOffset);
146 /// \brief Add a name accelerator entry for \p Die with \p Name
147 /// which is stored in the string table at \p Offset.
148 void addNameAccelerator(const DIE *Die, const char *Name, uint32_t Offset,
149 bool SkipPubnamesSection = false);
151 /// \brief Add a type accelerator entry for \p Die with \p Name
152 /// which is stored in the string table at \p Offset.
153 void addTypeAccelerator(const DIE *Die, const char *Name, uint32_t Offset);
156 StringRef Name; ///< Name of the entry.
157 const DIE *Die; ///< DIE this entry describes.
158 uint32_t NameOffset; ///< Offset of Name in the string pool.
159 bool SkipPubSection; ///< Emit this entry only in the apple_* sections.
161 AccelInfo(StringRef Name, const DIE *Die, uint32_t NameOffset,
162 bool SkipPubSection = false)
163 : Name(Name), Die(Die), NameOffset(NameOffset),
164 SkipPubSection(SkipPubSection) {}
167 const std::vector<AccelInfo> &getPubnames() const { return Pubnames; }
168 const std::vector<AccelInfo> &getPubtypes() const { return Pubtypes; }
173 std::vector<DIEInfo> Info; ///< DIE info indexed by DIE index.
174 std::unique_ptr<DIE> CUDie; ///< Root of the linked DIE tree.
176 uint64_t StartOffset;
177 uint64_t NextUnitOffset;
182 /// \brief A list of attributes to fixup with the absolute offset of
183 /// a DIE in the debug_info section.
185 /// The offsets for the attributes in this array couldn't be set while
186 /// cloning because for cross-cu forward refences the target DIE's
187 /// offset isn't known you emit the reference attribute.
188 std::vector<std::tuple<DIE *, const CompileUnit *, DIEInteger *>>
189 ForwardDIEReferences;
191 FunctionIntervals::Allocator RangeAlloc;
192 /// \brief The ranges in that interval map are the PC ranges for
193 /// functions in this unit, associated with the PC offset to apply
194 /// to the addresses to get the linked address.
195 FunctionIntervals Ranges;
197 /// \brief DW_AT_ranges attributes to patch after we have gathered
198 /// all the unit's function addresses.
200 std::vector<DIEInteger *> RangeAttributes;
201 DIEInteger *UnitRangeAttribute;
204 /// \brief Location attributes that need to be transfered from th
205 /// original debug_loc section to the liked one. They are stored
206 /// along with the PC offset that is to be applied to their
207 /// function's address.
208 std::vector<std::pair<DIEInteger *, int64_t>> LocationAttributes;
210 /// \brief Accelerator entries for the unit, both for the pub*
211 /// sections and the apple* ones.
213 std::vector<AccelInfo> Pubnames;
214 std::vector<AccelInfo> Pubtypes;
218 uint64_t CompileUnit::computeNextUnitOffset() {
219 NextUnitOffset = StartOffset + 11 /* Header size */;
220 // The root DIE might be null, meaning that the Unit had nothing to
221 // contribute to the linked output. In that case, we will emit the
222 // unit header without any actual DIE.
224 NextUnitOffset += CUDie->getSize();
225 return NextUnitOffset;
228 /// \brief Keep track of a forward cross-cu reference from this unit
229 /// to \p Die that lives in \p RefUnit.
230 void CompileUnit::noteForwardReference(DIE *Die, const CompileUnit *RefUnit,
232 ForwardDIEReferences.emplace_back(Die, RefUnit, Attr);
235 /// \brief Apply all fixups recorded by noteForwardReference().
236 void CompileUnit::fixupForwardReferences() {
237 for (const auto &Ref : ForwardDIEReferences) {
239 const CompileUnit *RefUnit;
241 std::tie(RefDie, RefUnit, Attr) = Ref;
242 Attr->setValue(RefDie->getOffset() + RefUnit->getStartOffset());
246 void CompileUnit::addFunctionRange(uint64_t FuncLowPc, uint64_t FuncHighPc,
248 Ranges.insert(FuncLowPc, FuncHighPc, PcOffset);
249 this->LowPc = std::min(LowPc, FuncLowPc + PcOffset);
250 this->HighPc = std::max(HighPc, FuncHighPc + PcOffset);
253 void CompileUnit::noteRangeAttribute(const DIE &Die, DIEInteger *Attr) {
254 if (Die.getTag() != dwarf::DW_TAG_compile_unit)
255 RangeAttributes.push_back(Attr);
257 UnitRangeAttribute = Attr;
260 void CompileUnit::noteLocationAttribute(DIEInteger *Attr, int64_t PcOffset) {
261 LocationAttributes.emplace_back(Attr, PcOffset);
264 /// \brief Add a name accelerator entry for \p Die with \p Name
265 /// which is stored in the string table at \p Offset.
266 void CompileUnit::addNameAccelerator(const DIE *Die, const char *Name,
267 uint32_t Offset, bool SkipPubSection) {
268 Pubnames.emplace_back(Name, Die, Offset, SkipPubSection);
271 /// \brief Add a type accelerator entry for \p Die with \p Name
272 /// which is stored in the string table at \p Offset.
273 void CompileUnit::addTypeAccelerator(const DIE *Die, const char *Name,
275 Pubtypes.emplace_back(Name, Die, Offset, false);
278 /// \brief A string table that doesn't need relocations.
280 /// We are doing a final link, no need for a string table that
281 /// has relocation entries for every reference to it. This class
282 /// provides this ablitity by just associating offsets with
284 class NonRelocatableStringpool {
286 /// \brief Entries are stored into the StringMap and simply linked
287 /// together through the second element of this pair in order to
288 /// keep track of insertion order.
289 typedef StringMap<std::pair<uint32_t, StringMapEntryBase *>, BumpPtrAllocator>
292 NonRelocatableStringpool()
293 : CurrentEndOffset(0), Sentinel(0), Last(&Sentinel) {
294 // Legacy dsymutil puts an empty string at the start of the line
299 /// \brief Get the offset of string \p S in the string table. This
300 /// can insert a new element or return the offset of a preexisitng
302 uint32_t getStringOffset(StringRef S);
304 /// \brief Get permanent storage for \p S (but do not necessarily
305 /// emit \p S in the output section).
306 /// \returns The StringRef that points to permanent storage to use
307 /// in place of \p S.
308 StringRef internString(StringRef S);
310 // \brief Return the first entry of the string table.
311 const MapTy::MapEntryTy *getFirstEntry() const {
312 return getNextEntry(&Sentinel);
315 // \brief Get the entry following \p E in the string table or null
316 // if \p E was the last entry.
317 const MapTy::MapEntryTy *getNextEntry(const MapTy::MapEntryTy *E) const {
318 return static_cast<const MapTy::MapEntryTy *>(E->getValue().second);
321 uint64_t getSize() { return CurrentEndOffset; }
325 uint32_t CurrentEndOffset;
326 MapTy::MapEntryTy Sentinel, *Last;
329 /// \brief Get the offset of string \p S in the string table. This
330 /// can insert a new element or return the offset of a preexisitng
332 uint32_t NonRelocatableStringpool::getStringOffset(StringRef S) {
333 if (S.empty() && !Strings.empty())
336 std::pair<uint32_t, StringMapEntryBase *> Entry(0, nullptr);
340 // A non-empty string can't be at offset 0, so if we have an entry
341 // with a 0 offset, it must be a previously interned string.
342 std::tie(It, Inserted) = Strings.insert(std::make_pair(S, Entry));
343 if (Inserted || It->getValue().first == 0) {
344 // Set offset and chain at the end of the entries list.
345 It->getValue().first = CurrentEndOffset;
346 CurrentEndOffset += S.size() + 1; // +1 for the '\0'.
347 Last->getValue().second = &*It;
350 return It->getValue().first;
353 /// \brief Put \p S into the StringMap so that it gets permanent
354 /// storage, but do not actually link it in the chain of elements
355 /// that go into the output section. A latter call to
356 /// getStringOffset() with the same string will chain it though.
357 StringRef NonRelocatableStringpool::internString(StringRef S) {
358 std::pair<uint32_t, StringMapEntryBase *> Entry(0, nullptr);
359 auto InsertResult = Strings.insert(std::make_pair(S, Entry));
360 return InsertResult.first->getKey();
363 /// \brief The Dwarf streaming logic
365 /// All interactions with the MC layer that is used to build the debug
366 /// information binary representation are handled in this class.
367 class DwarfStreamer {
368 /// \defgroup MCObjects MC layer objects constructed by the streamer
370 std::unique_ptr<MCRegisterInfo> MRI;
371 std::unique_ptr<MCAsmInfo> MAI;
372 std::unique_ptr<MCObjectFileInfo> MOFI;
373 std::unique_ptr<MCContext> MC;
374 MCAsmBackend *MAB; // Owned by MCStreamer
375 std::unique_ptr<MCInstrInfo> MII;
376 std::unique_ptr<MCSubtargetInfo> MSTI;
377 MCCodeEmitter *MCE; // Owned by MCStreamer
378 MCStreamer *MS; // Owned by AsmPrinter
379 std::unique_ptr<TargetMachine> TM;
380 std::unique_ptr<AsmPrinter> Asm;
383 /// \brief the file we stream the linked Dwarf to.
384 std::unique_ptr<raw_fd_ostream> OutFile;
386 uint32_t RangesSectionSize;
387 uint32_t LocSectionSize;
388 uint32_t LineSectionSize;
390 /// \brief Emit the pubnames or pubtypes section contribution for \p
391 /// Unit into \p Sec. The data is provided in \p Names.
392 void emitPubSectionForUnit(const MCSection *Sec, StringRef Name,
393 const CompileUnit &Unit,
394 const std::vector<CompileUnit::AccelInfo> &Names);
397 /// \brief Actually create the streamer and the ouptut file.
399 /// This could be done directly in the constructor, but it feels
400 /// more natural to handle errors through return value.
401 bool init(Triple TheTriple, StringRef OutputFilename);
403 /// \brief Dump the file to the disk.
406 AsmPrinter &getAsmPrinter() const { return *Asm; }
408 /// \brief Set the current output section to debug_info and change
409 /// the MC Dwarf version to \p DwarfVersion.
410 void switchToDebugInfoSection(unsigned DwarfVersion);
412 /// \brief Emit the compilation unit header for \p Unit in the
413 /// debug_info section.
415 /// As a side effect, this also switches the current Dwarf version
416 /// of the MC layer to the one of U.getOrigUnit().
417 void emitCompileUnitHeader(CompileUnit &Unit);
419 /// \brief Recursively emit the DIE tree rooted at \p Die.
420 void emitDIE(DIE &Die);
422 /// \brief Emit the abbreviation table \p Abbrevs to the
423 /// debug_abbrev section.
424 void emitAbbrevs(const std::vector<DIEAbbrev *> &Abbrevs);
426 /// \brief Emit the string table described by \p Pool.
427 void emitStrings(const NonRelocatableStringpool &Pool);
429 /// \brief Emit debug_ranges for \p FuncRange by translating the
430 /// original \p Entries.
431 void emitRangesEntries(
432 int64_t UnitPcOffset, uint64_t OrigLowPc,
433 FunctionIntervals::const_iterator FuncRange,
434 const std::vector<DWARFDebugRangeList::RangeListEntry> &Entries,
435 unsigned AddressSize);
437 /// \brief Emit debug_aranges entries for \p Unit and if \p
438 /// DoRangesSection is true, also emit the debug_ranges entries for
439 /// the DW_TAG_compile_unit's DW_AT_ranges attribute.
440 void emitUnitRangesEntries(CompileUnit &Unit, bool DoRangesSection);
442 uint32_t getRangesSectionSize() const { return RangesSectionSize; }
444 /// \brief Emit the debug_loc contribution for \p Unit by copying
445 /// the entries from \p Dwarf and offseting them. Update the
446 /// location attributes to point to the new entries.
447 void emitLocationsForUnit(const CompileUnit &Unit, DWARFContext &Dwarf);
449 /// \brief Emit the line table described in \p Rows into the
450 /// debug_line section.
451 void emitLineTableForUnit(StringRef PrologueBytes, unsigned MinInstLength,
452 std::vector<DWARFDebugLine::Row> &Rows,
453 unsigned AdddressSize);
455 uint32_t getLineSectionSize() const { return LineSectionSize; }
457 /// \brief Emit the .debug_pubnames contribution for \p Unit.
458 void emitPubNamesForUnit(const CompileUnit &Unit);
460 /// \brief Emit the .debug_pubtypes contribution for \p Unit.
461 void emitPubTypesForUnit(const CompileUnit &Unit);
464 bool DwarfStreamer::init(Triple TheTriple, StringRef OutputFilename) {
465 std::string ErrorStr;
466 std::string TripleName;
467 StringRef Context = "dwarf streamer init";
470 const Target *TheTarget =
471 TargetRegistry::lookupTarget(TripleName, TheTriple, ErrorStr);
473 return error(ErrorStr, Context);
474 TripleName = TheTriple.getTriple();
476 // Create all the MC Objects.
477 MRI.reset(TheTarget->createMCRegInfo(TripleName));
479 return error(Twine("no register info for target ") + TripleName, Context);
481 MAI.reset(TheTarget->createMCAsmInfo(*MRI, TripleName));
483 return error("no asm info for target " + TripleName, Context);
485 MOFI.reset(new MCObjectFileInfo);
486 MC.reset(new MCContext(MAI.get(), MRI.get(), MOFI.get()));
487 MOFI->InitMCObjectFileInfo(TripleName, Reloc::Default, CodeModel::Default,
490 MAB = TheTarget->createMCAsmBackend(*MRI, TripleName, "");
492 return error("no asm backend for target " + TripleName, Context);
494 MII.reset(TheTarget->createMCInstrInfo());
496 return error("no instr info info for target " + TripleName, Context);
498 MSTI.reset(TheTarget->createMCSubtargetInfo(TripleName, "", ""));
500 return error("no subtarget info for target " + TripleName, Context);
502 MCE = TheTarget->createMCCodeEmitter(*MII, *MRI, *MC);
504 return error("no code emitter for target " + TripleName, Context);
506 // Create the output file.
509 llvm::make_unique<raw_fd_ostream>(OutputFilename, EC, sys::fs::F_None);
511 return error(Twine(OutputFilename) + ": " + EC.message(), Context);
513 MS = TheTarget->createMCObjectStreamer(TheTriple, *MC, *MAB, *OutFile, MCE,
515 /*DWARFMustBeAtTheEnd*/ false);
517 return error("no object streamer for target " + TripleName, Context);
519 // Finally create the AsmPrinter we'll use to emit the DIEs.
520 TM.reset(TheTarget->createTargetMachine(TripleName, "", "", TargetOptions()));
522 return error("no target machine for target " + TripleName, Context);
524 Asm.reset(TheTarget->createAsmPrinter(*TM, std::unique_ptr<MCStreamer>(MS)));
526 return error("no asm printer for target " + TripleName, Context);
528 RangesSectionSize = 0;
535 bool DwarfStreamer::finish() {
540 /// \brief Set the current output section to debug_info and change
541 /// the MC Dwarf version to \p DwarfVersion.
542 void DwarfStreamer::switchToDebugInfoSection(unsigned DwarfVersion) {
543 MS->SwitchSection(MOFI->getDwarfInfoSection());
544 MC->setDwarfVersion(DwarfVersion);
547 /// \brief Emit the compilation unit header for \p Unit in the
548 /// debug_info section.
550 /// A Dwarf scetion header is encoded as:
551 /// uint32_t Unit length (omiting this field)
553 /// uint32_t Abbreviation table offset
554 /// uint8_t Address size
556 /// Leading to a total of 11 bytes.
557 void DwarfStreamer::emitCompileUnitHeader(CompileUnit &Unit) {
558 unsigned Version = Unit.getOrigUnit().getVersion();
559 switchToDebugInfoSection(Version);
561 // Emit size of content not including length itself. The size has
562 // already been computed in CompileUnit::computeOffsets(). Substract
563 // 4 to that size to account for the length field.
564 Asm->EmitInt32(Unit.getNextUnitOffset() - Unit.getStartOffset() - 4);
565 Asm->EmitInt16(Version);
566 // We share one abbreviations table across all units so it's always at the
567 // start of the section.
569 Asm->EmitInt8(Unit.getOrigUnit().getAddressByteSize());
572 /// \brief Emit the \p Abbrevs array as the shared abbreviation table
573 /// for the linked Dwarf file.
574 void DwarfStreamer::emitAbbrevs(const std::vector<DIEAbbrev *> &Abbrevs) {
575 MS->SwitchSection(MOFI->getDwarfAbbrevSection());
576 Asm->emitDwarfAbbrevs(Abbrevs);
579 /// \brief Recursively emit the DIE tree rooted at \p Die.
580 void DwarfStreamer::emitDIE(DIE &Die) {
581 MS->SwitchSection(MOFI->getDwarfInfoSection());
582 Asm->emitDwarfDIE(Die);
585 /// \brief Emit the debug_str section stored in \p Pool.
586 void DwarfStreamer::emitStrings(const NonRelocatableStringpool &Pool) {
587 Asm->OutStreamer->SwitchSection(MOFI->getDwarfStrSection());
588 for (auto *Entry = Pool.getFirstEntry(); Entry;
589 Entry = Pool.getNextEntry(Entry))
590 Asm->OutStreamer->EmitBytes(
591 StringRef(Entry->getKey().data(), Entry->getKey().size() + 1));
594 /// \brief Emit the debug_range section contents for \p FuncRange by
595 /// translating the original \p Entries. The debug_range section
596 /// format is totally trivial, consisting just of pairs of address
597 /// sized addresses describing the ranges.
598 void DwarfStreamer::emitRangesEntries(
599 int64_t UnitPcOffset, uint64_t OrigLowPc,
600 FunctionIntervals::const_iterator FuncRange,
601 const std::vector<DWARFDebugRangeList::RangeListEntry> &Entries,
602 unsigned AddressSize) {
603 MS->SwitchSection(MC->getObjectFileInfo()->getDwarfRangesSection());
605 // Offset each range by the right amount.
606 int64_t PcOffset = FuncRange.value() + UnitPcOffset;
607 for (const auto &Range : Entries) {
608 if (Range.isBaseAddressSelectionEntry(AddressSize)) {
609 warn("unsupported base address selection operation",
610 "emitting debug_ranges");
613 // Do not emit empty ranges.
614 if (Range.StartAddress == Range.EndAddress)
617 // All range entries should lie in the function range.
618 if (!(Range.StartAddress + OrigLowPc >= FuncRange.start() &&
619 Range.EndAddress + OrigLowPc <= FuncRange.stop()))
620 warn("inconsistent range data.", "emitting debug_ranges");
621 MS->EmitIntValue(Range.StartAddress + PcOffset, AddressSize);
622 MS->EmitIntValue(Range.EndAddress + PcOffset, AddressSize);
623 RangesSectionSize += 2 * AddressSize;
626 // Add the terminator entry.
627 MS->EmitIntValue(0, AddressSize);
628 MS->EmitIntValue(0, AddressSize);
629 RangesSectionSize += 2 * AddressSize;
632 /// \brief Emit the debug_aranges contribution of a unit and
633 /// if \p DoDebugRanges is true the debug_range contents for a
634 /// compile_unit level DW_AT_ranges attribute (Which are basically the
635 /// same thing with a different base address).
636 /// Just aggregate all the ranges gathered inside that unit.
637 void DwarfStreamer::emitUnitRangesEntries(CompileUnit &Unit,
638 bool DoDebugRanges) {
639 unsigned AddressSize = Unit.getOrigUnit().getAddressByteSize();
640 // Gather the ranges in a vector, so that we can simplify them. The
641 // IntervalMap will have coalesced the non-linked ranges, but here
642 // we want to coalesce the linked addresses.
643 std::vector<std::pair<uint64_t, uint64_t>> Ranges;
644 const auto &FunctionRanges = Unit.getFunctionRanges();
645 for (auto Range = FunctionRanges.begin(), End = FunctionRanges.end();
646 Range != End; ++Range)
647 Ranges.push_back(std::make_pair(Range.start() + Range.value(),
648 Range.stop() + Range.value()));
650 // The object addresses where sorted, but again, the linked
651 // addresses might end up in a different order.
652 std::sort(Ranges.begin(), Ranges.end());
654 if (!Ranges.empty()) {
655 MS->SwitchSection(MC->getObjectFileInfo()->getDwarfARangesSection());
657 MCSymbol *BeginLabel = Asm->createTempSymbol("Barange");
658 MCSymbol *EndLabel = Asm->createTempSymbol("Earange");
660 unsigned HeaderSize =
661 sizeof(int32_t) + // Size of contents (w/o this field
662 sizeof(int16_t) + // DWARF ARange version number
663 sizeof(int32_t) + // Offset of CU in the .debug_info section
664 sizeof(int8_t) + // Pointer Size (in bytes)
665 sizeof(int8_t); // Segment Size (in bytes)
667 unsigned TupleSize = AddressSize * 2;
668 unsigned Padding = OffsetToAlignment(HeaderSize, TupleSize);
670 Asm->EmitLabelDifference(EndLabel, BeginLabel, 4); // Arange length
671 Asm->OutStreamer->EmitLabel(BeginLabel);
672 Asm->EmitInt16(dwarf::DW_ARANGES_VERSION); // Version number
673 Asm->EmitInt32(Unit.getStartOffset()); // Corresponding unit's offset
674 Asm->EmitInt8(AddressSize); // Address size
675 Asm->EmitInt8(0); // Segment size
677 Asm->OutStreamer->EmitFill(Padding, 0x0);
679 for (auto Range = Ranges.begin(), End = Ranges.end(); Range != End;
681 uint64_t RangeStart = Range->first;
682 MS->EmitIntValue(RangeStart, AddressSize);
683 while ((Range + 1) != End && Range->second == (Range + 1)->first)
685 MS->EmitIntValue(Range->second - RangeStart, AddressSize);
689 Asm->OutStreamer->EmitIntValue(0, AddressSize);
690 Asm->OutStreamer->EmitIntValue(0, AddressSize);
691 Asm->OutStreamer->EmitLabel(EndLabel);
697 MS->SwitchSection(MC->getObjectFileInfo()->getDwarfRangesSection());
698 // Offset each range by the right amount.
699 int64_t PcOffset = -Unit.getLowPc();
700 // Emit coalesced ranges.
701 for (auto Range = Ranges.begin(), End = Ranges.end(); Range != End; ++Range) {
702 MS->EmitIntValue(Range->first + PcOffset, AddressSize);
703 while (Range + 1 != End && Range->second == (Range + 1)->first)
705 MS->EmitIntValue(Range->second + PcOffset, AddressSize);
706 RangesSectionSize += 2 * AddressSize;
709 // Add the terminator entry.
710 MS->EmitIntValue(0, AddressSize);
711 MS->EmitIntValue(0, AddressSize);
712 RangesSectionSize += 2 * AddressSize;
715 /// \brief Emit location lists for \p Unit and update attribtues to
716 /// point to the new entries.
717 void DwarfStreamer::emitLocationsForUnit(const CompileUnit &Unit,
718 DWARFContext &Dwarf) {
719 const std::vector<std::pair<DIEInteger *, int64_t>> &Attributes =
720 Unit.getLocationAttributes();
722 if (Attributes.empty())
725 MS->SwitchSection(MC->getObjectFileInfo()->getDwarfLocSection());
727 unsigned AddressSize = Unit.getOrigUnit().getAddressByteSize();
728 const DWARFSection &InputSec = Dwarf.getLocSection();
729 DataExtractor Data(InputSec.Data, Dwarf.isLittleEndian(), AddressSize);
730 DWARFUnit &OrigUnit = Unit.getOrigUnit();
731 const auto *OrigUnitDie = OrigUnit.getCompileUnitDIE(false);
732 int64_t UnitPcOffset = 0;
733 uint64_t OrigLowPc = OrigUnitDie->getAttributeValueAsAddress(
734 &OrigUnit, dwarf::DW_AT_low_pc, -1ULL);
735 if (OrigLowPc != -1ULL)
736 UnitPcOffset = int64_t(OrigLowPc) - Unit.getLowPc();
738 for (const auto &Attr : Attributes) {
739 uint32_t Offset = Attr.first->getValue();
740 Attr.first->setValue(LocSectionSize);
741 // This is the quantity to add to the old location address to get
742 // the correct address for the new one.
743 int64_t LocPcOffset = Attr.second + UnitPcOffset;
744 while (Data.isValidOffset(Offset)) {
745 uint64_t Low = Data.getUnsigned(&Offset, AddressSize);
746 uint64_t High = Data.getUnsigned(&Offset, AddressSize);
747 LocSectionSize += 2 * AddressSize;
748 if (Low == 0 && High == 0) {
749 Asm->OutStreamer->EmitIntValue(0, AddressSize);
750 Asm->OutStreamer->EmitIntValue(0, AddressSize);
753 Asm->OutStreamer->EmitIntValue(Low + LocPcOffset, AddressSize);
754 Asm->OutStreamer->EmitIntValue(High + LocPcOffset, AddressSize);
755 uint64_t Length = Data.getU16(&Offset);
756 Asm->OutStreamer->EmitIntValue(Length, 2);
757 // Just copy the bytes over.
758 Asm->OutStreamer->EmitBytes(
759 StringRef(InputSec.Data.substr(Offset, Length)));
761 LocSectionSize += Length + 2;
766 void DwarfStreamer::emitLineTableForUnit(StringRef PrologueBytes,
767 unsigned MinInstLength,
768 std::vector<DWARFDebugLine::Row> &Rows,
769 unsigned PointerSize) {
770 // Switch to the section where the table will be emitted into.
771 MS->SwitchSection(MC->getObjectFileInfo()->getDwarfLineSection());
772 MCSymbol *LineStartSym = MC->CreateTempSymbol();
773 MCSymbol *LineEndSym = MC->CreateTempSymbol();
775 // The first 4 bytes is the total length of the information for this
776 // compilation unit (not including these 4 bytes for the length).
777 Asm->EmitLabelDifference(LineEndSym, LineStartSym, 4);
778 Asm->OutStreamer->EmitLabel(LineStartSym);
780 MS->EmitBytes(PrologueBytes);
781 LineSectionSize += PrologueBytes.size() + 4;
783 SmallString<128> EncodingBuffer;
784 raw_svector_ostream EncodingOS(EncodingBuffer);
787 // We only have the dummy entry, dsymutil emits an entry with a 0
788 // address in that case.
789 MCDwarfLineAddr::Encode(*MC, INT64_MAX, 0, EncodingOS);
790 MS->EmitBytes(EncodingOS.str());
791 LineSectionSize += EncodingBuffer.size();
792 MS->EmitLabel(LineEndSym);
796 // Line table state machine fields
797 unsigned FileNum = 1;
798 unsigned LastLine = 1;
800 unsigned IsStatement = 1;
802 uint64_t Address = -1ULL;
804 unsigned RowsSinceLastSequence = 0;
806 for (unsigned Idx = 0; Idx < Rows.size(); ++Idx) {
807 auto &Row = Rows[Idx];
809 int64_t AddressDelta;
810 if (Address == -1ULL) {
811 MS->EmitIntValue(dwarf::DW_LNS_extended_op, 1);
812 MS->EmitULEB128IntValue(PointerSize + 1);
813 MS->EmitIntValue(dwarf::DW_LNE_set_address, 1);
814 MS->EmitIntValue(Row.Address, PointerSize);
815 LineSectionSize += 2 + PointerSize + getULEB128Size(PointerSize + 1);
818 AddressDelta = (Row.Address - Address) / MinInstLength;
821 // FIXME: code copied and transfromed from
822 // MCDwarf.cpp::EmitDwarfLineTable. We should find a way to share
823 // this code, but the current compatibility requirement with
824 // classic dsymutil makes it hard. Revisit that once this
825 // requirement is dropped.
827 if (FileNum != Row.File) {
829 MS->EmitIntValue(dwarf::DW_LNS_set_file, 1);
830 MS->EmitULEB128IntValue(FileNum);
831 LineSectionSize += 1 + getULEB128Size(FileNum);
833 if (Column != Row.Column) {
835 MS->EmitIntValue(dwarf::DW_LNS_set_column, 1);
836 MS->EmitULEB128IntValue(Column);
837 LineSectionSize += 1 + getULEB128Size(Column);
840 // FIXME: We should handle the discriminator here, but dsymutil
841 // doesn' consider it, thus ignore it for now.
843 if (Isa != Row.Isa) {
845 MS->EmitIntValue(dwarf::DW_LNS_set_isa, 1);
846 MS->EmitULEB128IntValue(Isa);
847 LineSectionSize += 1 + getULEB128Size(Isa);
849 if (IsStatement != Row.IsStmt) {
850 IsStatement = Row.IsStmt;
851 MS->EmitIntValue(dwarf::DW_LNS_negate_stmt, 1);
852 LineSectionSize += 1;
854 if (Row.BasicBlock) {
855 MS->EmitIntValue(dwarf::DW_LNS_set_basic_block, 1);
856 LineSectionSize += 1;
859 if (Row.PrologueEnd) {
860 MS->EmitIntValue(dwarf::DW_LNS_set_prologue_end, 1);
861 LineSectionSize += 1;
864 if (Row.EpilogueBegin) {
865 MS->EmitIntValue(dwarf::DW_LNS_set_epilogue_begin, 1);
866 LineSectionSize += 1;
869 int64_t LineDelta = int64_t(Row.Line) - LastLine;
870 if (!Row.EndSequence) {
871 MCDwarfLineAddr::Encode(*MC, LineDelta, AddressDelta, EncodingOS);
872 MS->EmitBytes(EncodingOS.str());
873 LineSectionSize += EncodingBuffer.size();
874 EncodingBuffer.resize(0);
876 Address = Row.Address;
878 RowsSinceLastSequence++;
881 MS->EmitIntValue(dwarf::DW_LNS_advance_line, 1);
882 MS->EmitSLEB128IntValue(LineDelta);
883 LineSectionSize += 1 + getSLEB128Size(LineDelta);
886 MS->EmitIntValue(dwarf::DW_LNS_advance_pc, 1);
887 MS->EmitULEB128IntValue(AddressDelta);
888 LineSectionSize += 1 + getULEB128Size(AddressDelta);
890 MCDwarfLineAddr::Encode(*MC, INT64_MAX, 0, EncodingOS);
891 MS->EmitBytes(EncodingOS.str());
892 LineSectionSize += EncodingBuffer.size();
893 EncodingBuffer.resize(0);
896 LastLine = FileNum = IsStatement = 1;
897 RowsSinceLastSequence = Column = Isa = 0;
901 if (RowsSinceLastSequence) {
902 MCDwarfLineAddr::Encode(*MC, INT64_MAX, 0, EncodingOS);
903 MS->EmitBytes(EncodingOS.str());
904 LineSectionSize += EncodingBuffer.size();
905 EncodingBuffer.resize(0);
909 MS->EmitLabel(LineEndSym);
912 /// \brief Emit the pubnames or pubtypes section contribution for \p
913 /// Unit into \p Sec. The data is provided in \p Names.
914 void DwarfStreamer::emitPubSectionForUnit(
915 const MCSection *Sec, StringRef SecName, const CompileUnit &Unit,
916 const std::vector<CompileUnit::AccelInfo> &Names) {
920 // Start the dwarf pubnames section.
921 Asm->OutStreamer->SwitchSection(Sec);
922 MCSymbol *BeginLabel = Asm->createTempSymbol("pub" + SecName + "_begin");
923 MCSymbol *EndLabel = Asm->createTempSymbol("pub" + SecName + "_end");
925 bool HeaderEmitted = false;
926 // Emit the pubnames for this compilation unit.
927 for (const auto &Name : Names) {
928 if (Name.SkipPubSection)
931 if (!HeaderEmitted) {
933 Asm->EmitLabelDifference(EndLabel, BeginLabel, 4); // Length
934 Asm->OutStreamer->EmitLabel(BeginLabel);
935 Asm->EmitInt16(dwarf::DW_PUBNAMES_VERSION); // Version
936 Asm->EmitInt32(Unit.getStartOffset()); // Unit offset
937 Asm->EmitInt32(Unit.getNextUnitOffset() - Unit.getStartOffset()); // Size
938 HeaderEmitted = true;
940 Asm->EmitInt32(Name.Die->getOffset());
941 Asm->OutStreamer->EmitBytes(
942 StringRef(Name.Name.data(), Name.Name.size() + 1));
947 Asm->EmitInt32(0); // End marker.
948 Asm->OutStreamer->EmitLabel(EndLabel);
951 /// \brief Emit .debug_pubnames for \p Unit.
952 void DwarfStreamer::emitPubNamesForUnit(const CompileUnit &Unit) {
953 emitPubSectionForUnit(MC->getObjectFileInfo()->getDwarfPubNamesSection(),
954 "names", Unit, Unit.getPubnames());
957 /// \brief Emit .debug_pubtypes for \p Unit.
958 void DwarfStreamer::emitPubTypesForUnit(const CompileUnit &Unit) {
959 emitPubSectionForUnit(MC->getObjectFileInfo()->getDwarfPubTypesSection(),
960 "types", Unit, Unit.getPubtypes());
963 /// \brief The core of the Dwarf linking logic.
965 /// The link of the dwarf information from the object files will be
966 /// driven by the selection of 'root DIEs', which are DIEs that
967 /// describe variables or functions that are present in the linked
968 /// binary (and thus have entries in the debug map). All the debug
969 /// information that will be linked (the DIEs, but also the line
970 /// tables, ranges, ...) is derived from that set of root DIEs.
972 /// The root DIEs are identified because they contain relocations that
973 /// correspond to a debug map entry at specific places (the low_pc for
974 /// a function, the location for a variable). These relocations are
975 /// called ValidRelocs in the DwarfLinker and are gathered as a very
976 /// first step when we start processing a DebugMapObject.
979 DwarfLinker(StringRef OutputFilename, const LinkOptions &Options)
980 : OutputFilename(OutputFilename), Options(Options),
981 BinHolder(Options.Verbose) {}
984 for (auto *Abbrev : Abbreviations)
988 /// \brief Link the contents of the DebugMap.
989 bool link(const DebugMap &);
992 /// \brief Called at the start of a debug object link.
993 void startDebugObject(DWARFContext &, DebugMapObject &);
995 /// \brief Called at the end of a debug object link.
996 void endDebugObject();
998 /// \defgroup FindValidRelocations Translate debug map into a list
999 /// of relevant relocations
1006 const DebugMapObject::DebugMapEntry *Mapping;
1008 ValidReloc(uint32_t Offset, uint32_t Size, uint64_t Addend,
1009 const DebugMapObject::DebugMapEntry *Mapping)
1010 : Offset(Offset), Size(Size), Addend(Addend), Mapping(Mapping) {}
1012 bool operator<(const ValidReloc &RHS) const { return Offset < RHS.Offset; }
1015 /// \brief The valid relocations for the current DebugMapObject.
1016 /// This vector is sorted by relocation offset.
1017 std::vector<ValidReloc> ValidRelocs;
1019 /// \brief Index into ValidRelocs of the next relocation to
1020 /// consider. As we walk the DIEs in acsending file offset and as
1021 /// ValidRelocs is sorted by file offset, keeping this index
1022 /// uptodate is all we have to do to have a cheap lookup during the
1023 /// root DIE selection and during DIE cloning.
1024 unsigned NextValidReloc;
1026 bool findValidRelocsInDebugInfo(const object::ObjectFile &Obj,
1027 const DebugMapObject &DMO);
1029 bool findValidRelocs(const object::SectionRef &Section,
1030 const object::ObjectFile &Obj,
1031 const DebugMapObject &DMO);
1033 void findValidRelocsMachO(const object::SectionRef &Section,
1034 const object::MachOObjectFile &Obj,
1035 const DebugMapObject &DMO);
1038 /// \defgroup FindRootDIEs Find DIEs corresponding to debug map entries.
1041 /// \brief Recursively walk the \p DIE tree and look for DIEs to
1042 /// keep. Store that information in \p CU's DIEInfo.
1043 void lookForDIEsToKeep(const DWARFDebugInfoEntryMinimal &DIE,
1044 const DebugMapObject &DMO, CompileUnit &CU,
1047 /// \brief Flags passed to DwarfLinker::lookForDIEsToKeep
1048 enum TravesalFlags {
1049 TF_Keep = 1 << 0, ///< Mark the traversed DIEs as kept.
1050 TF_InFunctionScope = 1 << 1, ///< Current scope is a fucntion scope.
1051 TF_DependencyWalk = 1 << 2, ///< Walking the dependencies of a kept DIE.
1052 TF_ParentWalk = 1 << 3, ///< Walking up the parents of a kept DIE.
1055 /// \brief Mark the passed DIE as well as all the ones it depends on
1057 void keepDIEAndDenpendencies(const DWARFDebugInfoEntryMinimal &DIE,
1058 CompileUnit::DIEInfo &MyInfo,
1059 const DebugMapObject &DMO, CompileUnit &CU,
1062 unsigned shouldKeepDIE(const DWARFDebugInfoEntryMinimal &DIE,
1063 CompileUnit &Unit, CompileUnit::DIEInfo &MyInfo,
1066 unsigned shouldKeepVariableDIE(const DWARFDebugInfoEntryMinimal &DIE,
1068 CompileUnit::DIEInfo &MyInfo, unsigned Flags);
1070 unsigned shouldKeepSubprogramDIE(const DWARFDebugInfoEntryMinimal &DIE,
1072 CompileUnit::DIEInfo &MyInfo,
1075 bool hasValidRelocation(uint32_t StartOffset, uint32_t EndOffset,
1076 CompileUnit::DIEInfo &Info);
1079 /// \defgroup Linking Methods used to link the debug information
1082 /// \brief Recursively clone \p InputDIE into an tree of DIE objects
1083 /// where useless (as decided by lookForDIEsToKeep()) bits have been
1084 /// stripped out and addresses have been rewritten according to the
1087 /// \param OutOffset is the offset the cloned DIE in the output
1089 /// \param PCOffset (while cloning a function scope) is the offset
1090 /// applied to the entry point of the function to get the linked address.
1092 /// \returns the root of the cloned tree.
1093 DIE *cloneDIE(const DWARFDebugInfoEntryMinimal &InputDIE, CompileUnit &U,
1094 int64_t PCOffset, uint32_t OutOffset);
1096 typedef DWARFAbbreviationDeclaration::AttributeSpec AttributeSpec;
1098 /// \brief Information gathered and exchanged between the various
1099 /// clone*Attributes helpers about the attributes of a particular DIE.
1100 struct AttributesInfo {
1101 const char *Name, *MangledName; ///< Names.
1102 uint32_t NameOffset, MangledNameOffset; ///< Offsets in the string pool.
1104 uint64_t OrigHighPc; ///< Value of AT_high_pc in the input DIE
1105 int64_t PCOffset; ///< Offset to apply to PC addresses inside a function.
1107 bool HasLowPc; ///< Does the DIE have a low_pc attribute?
1108 bool IsDeclaration; ///< Is this DIE only a declaration?
1111 : Name(nullptr), MangledName(nullptr), NameOffset(0),
1112 MangledNameOffset(0), OrigHighPc(0), PCOffset(0), HasLowPc(false),
1113 IsDeclaration(false) {}
1116 /// \brief Helper for cloneDIE.
1117 unsigned cloneAttribute(DIE &Die, const DWARFDebugInfoEntryMinimal &InputDIE,
1118 CompileUnit &U, const DWARFFormValue &Val,
1119 const AttributeSpec AttrSpec, unsigned AttrSize,
1120 AttributesInfo &AttrInfo);
1122 /// \brief Helper for cloneDIE.
1123 unsigned cloneStringAttribute(DIE &Die, AttributeSpec AttrSpec,
1124 const DWARFFormValue &Val, const DWARFUnit &U);
1126 /// \brief Helper for cloneDIE.
1128 cloneDieReferenceAttribute(DIE &Die,
1129 const DWARFDebugInfoEntryMinimal &InputDIE,
1130 AttributeSpec AttrSpec, unsigned AttrSize,
1131 const DWARFFormValue &Val, CompileUnit &Unit);
1133 /// \brief Helper for cloneDIE.
1134 unsigned cloneBlockAttribute(DIE &Die, AttributeSpec AttrSpec,
1135 const DWARFFormValue &Val, unsigned AttrSize);
1137 /// \brief Helper for cloneDIE.
1138 unsigned cloneAddressAttribute(DIE &Die, AttributeSpec AttrSpec,
1139 const DWARFFormValue &Val,
1140 const CompileUnit &Unit, AttributesInfo &Info);
1142 /// \brief Helper for cloneDIE.
1143 unsigned cloneScalarAttribute(DIE &Die,
1144 const DWARFDebugInfoEntryMinimal &InputDIE,
1145 CompileUnit &U, AttributeSpec AttrSpec,
1146 const DWARFFormValue &Val, unsigned AttrSize,
1147 AttributesInfo &Info);
1149 /// \brief Helper for cloneDIE.
1150 bool applyValidRelocs(MutableArrayRef<char> Data, uint32_t BaseOffset,
1151 bool isLittleEndian);
1153 /// \brief Assign an abbreviation number to \p Abbrev
1154 void AssignAbbrev(DIEAbbrev &Abbrev);
1156 /// \brief FoldingSet that uniques the abbreviations.
1157 FoldingSet<DIEAbbrev> AbbreviationsSet;
1158 /// \brief Storage for the unique Abbreviations.
1159 /// This is passed to AsmPrinter::emitDwarfAbbrevs(), thus it cannot
1160 /// be changed to a vecot of unique_ptrs.
1161 std::vector<DIEAbbrev *> Abbreviations;
1163 /// \brief Compute and emit debug_ranges section for \p Unit, and
1164 /// patch the attributes referencing it.
1165 void patchRangesForUnit(const CompileUnit &Unit, DWARFContext &Dwarf) const;
1167 /// \brief Generate and emit the DW_AT_ranges attribute for a
1168 /// compile_unit if it had one.
1169 void generateUnitRanges(CompileUnit &Unit) const;
1171 /// \brief Extract the line tables fromt he original dwarf, extract
1172 /// the relevant parts according to the linked function ranges and
1173 /// emit the result in the debug_line section.
1174 void patchLineTableForUnit(CompileUnit &Unit, DWARFContext &OrigDwarf);
1176 /// \brief Emit the accelerator entries for \p Unit.
1177 void emitAcceleratorEntriesForUnit(CompileUnit &Unit);
1179 /// \brief DIELoc objects that need to be destructed (but not freed!).
1180 std::vector<DIELoc *> DIELocs;
1181 /// \brief DIEBlock objects that need to be destructed (but not freed!).
1182 std::vector<DIEBlock *> DIEBlocks;
1183 /// \brief Allocator used for all the DIEValue objects.
1184 BumpPtrAllocator DIEAlloc;
1187 /// \defgroup Helpers Various helper methods.
1190 const DWARFDebugInfoEntryMinimal *
1191 resolveDIEReference(DWARFFormValue &RefValue, const DWARFUnit &Unit,
1192 const DWARFDebugInfoEntryMinimal &DIE,
1193 CompileUnit *&ReferencedCU);
1195 CompileUnit *getUnitForOffset(unsigned Offset);
1197 bool getDIENames(const DWARFDebugInfoEntryMinimal &Die, DWARFUnit &U,
1198 AttributesInfo &Info);
1200 void reportWarning(const Twine &Warning, const DWARFUnit *Unit = nullptr,
1201 const DWARFDebugInfoEntryMinimal *DIE = nullptr) const;
1203 bool createStreamer(Triple TheTriple, StringRef OutputFilename);
1207 std::string OutputFilename;
1208 LinkOptions Options;
1209 BinaryHolder BinHolder;
1210 std::unique_ptr<DwarfStreamer> Streamer;
1212 /// The units of the current debug map object.
1213 std::vector<CompileUnit> Units;
1215 /// The debug map object curently under consideration.
1216 DebugMapObject *CurrentDebugObject;
1218 /// \brief The Dwarf string pool
1219 NonRelocatableStringpool StringPool;
1221 /// \brief This map is keyed by the entry PC of functions in that
1222 /// debug object and the associated value is a pair storing the
1223 /// corresponding end PC and the offset to apply to get the linked
1226 /// See startDebugObject() for a more complete description of its use.
1227 std::map<uint64_t, std::pair<uint64_t, int64_t>> Ranges;
1230 /// \brief Similar to DWARFUnitSection::getUnitForOffset(), but
1231 /// returning our CompileUnit object instead.
1232 CompileUnit *DwarfLinker::getUnitForOffset(unsigned Offset) {
1234 std::upper_bound(Units.begin(), Units.end(), Offset,
1235 [](uint32_t LHS, const CompileUnit &RHS) {
1236 return LHS < RHS.getOrigUnit().getNextUnitOffset();
1238 return CU != Units.end() ? &*CU : nullptr;
1241 /// \brief Resolve the DIE attribute reference that has been
1242 /// extracted in \p RefValue. The resulting DIE migh be in another
1243 /// CompileUnit which is stored into \p ReferencedCU.
1244 /// \returns null if resolving fails for any reason.
1245 const DWARFDebugInfoEntryMinimal *DwarfLinker::resolveDIEReference(
1246 DWARFFormValue &RefValue, const DWARFUnit &Unit,
1247 const DWARFDebugInfoEntryMinimal &DIE, CompileUnit *&RefCU) {
1248 assert(RefValue.isFormClass(DWARFFormValue::FC_Reference));
1249 uint64_t RefOffset = *RefValue.getAsReference(&Unit);
1251 if ((RefCU = getUnitForOffset(RefOffset)))
1252 if (const auto *RefDie = RefCU->getOrigUnit().getDIEForOffset(RefOffset))
1255 reportWarning("could not find referenced DIE", &Unit, &DIE);
1259 /// \brief Get the potential name and mangled name for the entity
1260 /// described by \p Die and store them in \Info if they are not
1262 /// \returns is a name was found.
1263 bool DwarfLinker::getDIENames(const DWARFDebugInfoEntryMinimal &Die,
1264 DWARFUnit &U, AttributesInfo &Info) {
1265 // FIXME: a bit wastefull as the first getName might return the
1267 if (!Info.MangledName &&
1268 (Info.MangledName = Die.getName(&U, DINameKind::LinkageName)))
1269 Info.MangledNameOffset = StringPool.getStringOffset(Info.MangledName);
1271 if (!Info.Name && (Info.Name = Die.getName(&U, DINameKind::ShortName)))
1272 Info.NameOffset = StringPool.getStringOffset(Info.Name);
1274 return Info.Name || Info.MangledName;
1277 /// \brief Report a warning to the user, optionaly including
1278 /// information about a specific \p DIE related to the warning.
1279 void DwarfLinker::reportWarning(const Twine &Warning, const DWARFUnit *Unit,
1280 const DWARFDebugInfoEntryMinimal *DIE) const {
1281 StringRef Context = "<debug map>";
1282 if (CurrentDebugObject)
1283 Context = CurrentDebugObject->getObjectFilename();
1284 warn(Warning, Context);
1286 if (!Options.Verbose || !DIE)
1289 errs() << " in DIE:\n";
1290 DIE->dump(errs(), const_cast<DWARFUnit *>(Unit), 0 /* RecurseDepth */,
1294 bool DwarfLinker::createStreamer(Triple TheTriple, StringRef OutputFilename) {
1295 if (Options.NoOutput)
1298 Streamer = llvm::make_unique<DwarfStreamer>();
1299 return Streamer->init(TheTriple, OutputFilename);
1302 /// \brief Recursive helper to gather the child->parent relationships in the
1303 /// original compile unit.
1304 static void gatherDIEParents(const DWARFDebugInfoEntryMinimal *DIE,
1305 unsigned ParentIdx, CompileUnit &CU) {
1306 unsigned MyIdx = CU.getOrigUnit().getDIEIndex(DIE);
1307 CU.getInfo(MyIdx).ParentIdx = ParentIdx;
1309 if (DIE->hasChildren())
1310 for (auto *Child = DIE->getFirstChild(); Child && !Child->isNULL();
1311 Child = Child->getSibling())
1312 gatherDIEParents(Child, MyIdx, CU);
1315 static bool dieNeedsChildrenToBeMeaningful(uint32_t Tag) {
1319 case dwarf::DW_TAG_subprogram:
1320 case dwarf::DW_TAG_lexical_block:
1321 case dwarf::DW_TAG_subroutine_type:
1322 case dwarf::DW_TAG_structure_type:
1323 case dwarf::DW_TAG_class_type:
1324 case dwarf::DW_TAG_union_type:
1327 llvm_unreachable("Invalid Tag");
1330 void DwarfLinker::startDebugObject(DWARFContext &Dwarf, DebugMapObject &Obj) {
1331 Units.reserve(Dwarf.getNumCompileUnits());
1333 // Iterate over the debug map entries and put all the ones that are
1334 // functions (because they have a size) into the Ranges map. This
1335 // map is very similar to the FunctionRanges that are stored in each
1336 // unit, with 2 notable differences:
1337 // - obviously this one is global, while the other ones are per-unit.
1338 // - this one contains not only the functions described in the DIE
1339 // tree, but also the ones that are only in the debug map.
1340 // The latter information is required to reproduce dsymutil's logic
1341 // while linking line tables. The cases where this information
1342 // matters look like bugs that need to be investigated, but for now
1343 // we need to reproduce dsymutil's behavior.
1344 // FIXME: Once we understood exactly if that information is needed,
1345 // maybe totally remove this (or try to use it to do a real
1346 // -gline-tables-only on Darwin.
1347 for (const auto &Entry : Obj.symbols()) {
1348 const auto &Mapping = Entry.getValue();
1350 Ranges[Mapping.ObjectAddress] = std::make_pair(
1351 Mapping.ObjectAddress + Mapping.Size,
1352 int64_t(Mapping.BinaryAddress) - Mapping.ObjectAddress);
1356 void DwarfLinker::endDebugObject() {
1358 ValidRelocs.clear();
1361 for (auto *Block : DIEBlocks)
1363 for (auto *Loc : DIELocs)
1371 /// \brief Iterate over the relocations of the given \p Section and
1372 /// store the ones that correspond to debug map entries into the
1373 /// ValidRelocs array.
1374 void DwarfLinker::findValidRelocsMachO(const object::SectionRef &Section,
1375 const object::MachOObjectFile &Obj,
1376 const DebugMapObject &DMO) {
1378 Section.getContents(Contents);
1379 DataExtractor Data(Contents, Obj.isLittleEndian(), 0);
1381 for (const object::RelocationRef &Reloc : Section.relocations()) {
1382 object::DataRefImpl RelocDataRef = Reloc.getRawDataRefImpl();
1383 MachO::any_relocation_info MachOReloc = Obj.getRelocation(RelocDataRef);
1384 unsigned RelocSize = 1 << Obj.getAnyRelocationLength(MachOReloc);
1386 if ((RelocSize != 4 && RelocSize != 8) || Reloc.getOffset(Offset64)) {
1387 reportWarning(" unsupported relocation in debug_info section.");
1390 uint32_t Offset = Offset64;
1391 // Mach-o uses REL relocations, the addend is at the relocation offset.
1392 uint64_t Addend = Data.getUnsigned(&Offset, RelocSize);
1394 auto Sym = Reloc.getSymbol();
1395 if (Sym != Obj.symbol_end()) {
1396 StringRef SymbolName;
1397 if (Sym->getName(SymbolName)) {
1398 reportWarning("error getting relocation symbol name.");
1401 if (const auto *Mapping = DMO.lookupSymbol(SymbolName))
1402 ValidRelocs.emplace_back(Offset64, RelocSize, Addend, Mapping);
1403 } else if (const auto *Mapping = DMO.lookupObjectAddress(Addend)) {
1404 // Do not store the addend. The addend was the address of the
1405 // symbol in the object file, the address in the binary that is
1406 // stored in the debug map doesn't need to be offseted.
1407 ValidRelocs.emplace_back(Offset64, RelocSize, 0, Mapping);
1412 /// \brief Dispatch the valid relocation finding logic to the
1413 /// appropriate handler depending on the object file format.
1414 bool DwarfLinker::findValidRelocs(const object::SectionRef &Section,
1415 const object::ObjectFile &Obj,
1416 const DebugMapObject &DMO) {
1417 // Dispatch to the right handler depending on the file type.
1418 if (auto *MachOObj = dyn_cast<object::MachOObjectFile>(&Obj))
1419 findValidRelocsMachO(Section, *MachOObj, DMO);
1421 reportWarning(Twine("unsupported object file type: ") + Obj.getFileName());
1423 if (ValidRelocs.empty())
1426 // Sort the relocations by offset. We will walk the DIEs linearly in
1427 // the file, this allows us to just keep an index in the relocation
1428 // array that we advance during our walk, rather than resorting to
1429 // some associative container. See DwarfLinker::NextValidReloc.
1430 std::sort(ValidRelocs.begin(), ValidRelocs.end());
1434 /// \brief Look for relocations in the debug_info section that match
1435 /// entries in the debug map. These relocations will drive the Dwarf
1436 /// link by indicating which DIEs refer to symbols present in the
1438 /// \returns wether there are any valid relocations in the debug info.
1439 bool DwarfLinker::findValidRelocsInDebugInfo(const object::ObjectFile &Obj,
1440 const DebugMapObject &DMO) {
1441 // Find the debug_info section.
1442 for (const object::SectionRef &Section : Obj.sections()) {
1443 StringRef SectionName;
1444 Section.getName(SectionName);
1445 SectionName = SectionName.substr(SectionName.find_first_not_of("._"));
1446 if (SectionName != "debug_info")
1448 return findValidRelocs(Section, Obj, DMO);
1453 /// \brief Checks that there is a relocation against an actual debug
1454 /// map entry between \p StartOffset and \p NextOffset.
1456 /// This function must be called with offsets in strictly ascending
1457 /// order because it never looks back at relocations it already 'went past'.
1458 /// \returns true and sets Info.InDebugMap if it is the case.
1459 bool DwarfLinker::hasValidRelocation(uint32_t StartOffset, uint32_t EndOffset,
1460 CompileUnit::DIEInfo &Info) {
1461 assert(NextValidReloc == 0 ||
1462 StartOffset > ValidRelocs[NextValidReloc - 1].Offset);
1463 if (NextValidReloc >= ValidRelocs.size())
1466 uint64_t RelocOffset = ValidRelocs[NextValidReloc].Offset;
1468 // We might need to skip some relocs that we didn't consider. For
1469 // example the high_pc of a discarded DIE might contain a reloc that
1470 // is in the list because it actually corresponds to the start of a
1471 // function that is in the debug map.
1472 while (RelocOffset < StartOffset && NextValidReloc < ValidRelocs.size() - 1)
1473 RelocOffset = ValidRelocs[++NextValidReloc].Offset;
1475 if (RelocOffset < StartOffset || RelocOffset >= EndOffset)
1478 const auto &ValidReloc = ValidRelocs[NextValidReloc++];
1479 if (Options.Verbose)
1480 outs() << "Found valid debug map entry: " << ValidReloc.Mapping->getKey()
1481 << " " << format("\t%016" PRIx64 " => %016" PRIx64,
1482 ValidReloc.Mapping->getValue().ObjectAddress,
1483 ValidReloc.Mapping->getValue().BinaryAddress);
1485 Info.AddrAdjust = int64_t(ValidReloc.Mapping->getValue().BinaryAddress) +
1487 ValidReloc.Mapping->getValue().ObjectAddress;
1488 Info.InDebugMap = true;
1492 /// \brief Get the starting and ending (exclusive) offset for the
1493 /// attribute with index \p Idx descibed by \p Abbrev. \p Offset is
1494 /// supposed to point to the position of the first attribute described
1496 /// \return [StartOffset, EndOffset) as a pair.
1497 static std::pair<uint32_t, uint32_t>
1498 getAttributeOffsets(const DWARFAbbreviationDeclaration *Abbrev, unsigned Idx,
1499 unsigned Offset, const DWARFUnit &Unit) {
1500 DataExtractor Data = Unit.getDebugInfoExtractor();
1502 for (unsigned i = 0; i < Idx; ++i)
1503 DWARFFormValue::skipValue(Abbrev->getFormByIndex(i), Data, &Offset, &Unit);
1505 uint32_t End = Offset;
1506 DWARFFormValue::skipValue(Abbrev->getFormByIndex(Idx), Data, &End, &Unit);
1508 return std::make_pair(Offset, End);
1511 /// \brief Check if a variable describing DIE should be kept.
1512 /// \returns updated TraversalFlags.
1513 unsigned DwarfLinker::shouldKeepVariableDIE(
1514 const DWARFDebugInfoEntryMinimal &DIE, CompileUnit &Unit,
1515 CompileUnit::DIEInfo &MyInfo, unsigned Flags) {
1516 const auto *Abbrev = DIE.getAbbreviationDeclarationPtr();
1518 // Global variables with constant value can always be kept.
1519 if (!(Flags & TF_InFunctionScope) &&
1520 Abbrev->findAttributeIndex(dwarf::DW_AT_const_value) != -1U) {
1521 MyInfo.InDebugMap = true;
1522 return Flags | TF_Keep;
1525 uint32_t LocationIdx = Abbrev->findAttributeIndex(dwarf::DW_AT_location);
1526 if (LocationIdx == -1U)
1529 uint32_t Offset = DIE.getOffset() + getULEB128Size(Abbrev->getCode());
1530 const DWARFUnit &OrigUnit = Unit.getOrigUnit();
1531 uint32_t LocationOffset, LocationEndOffset;
1532 std::tie(LocationOffset, LocationEndOffset) =
1533 getAttributeOffsets(Abbrev, LocationIdx, Offset, OrigUnit);
1535 // See if there is a relocation to a valid debug map entry inside
1536 // this variable's location. The order is important here. We want to
1537 // always check in the variable has a valid relocation, so that the
1538 // DIEInfo is filled. However, we don't want a static variable in a
1539 // function to force us to keep the enclosing function.
1540 if (!hasValidRelocation(LocationOffset, LocationEndOffset, MyInfo) ||
1541 (Flags & TF_InFunctionScope))
1544 if (Options.Verbose)
1545 DIE.dump(outs(), const_cast<DWARFUnit *>(&OrigUnit), 0, 8 /* Indent */);
1547 return Flags | TF_Keep;
1550 /// \brief Check if a function describing DIE should be kept.
1551 /// \returns updated TraversalFlags.
1552 unsigned DwarfLinker::shouldKeepSubprogramDIE(
1553 const DWARFDebugInfoEntryMinimal &DIE, CompileUnit &Unit,
1554 CompileUnit::DIEInfo &MyInfo, unsigned Flags) {
1555 const auto *Abbrev = DIE.getAbbreviationDeclarationPtr();
1557 Flags |= TF_InFunctionScope;
1559 uint32_t LowPcIdx = Abbrev->findAttributeIndex(dwarf::DW_AT_low_pc);
1560 if (LowPcIdx == -1U)
1563 uint32_t Offset = DIE.getOffset() + getULEB128Size(Abbrev->getCode());
1564 const DWARFUnit &OrigUnit = Unit.getOrigUnit();
1565 uint32_t LowPcOffset, LowPcEndOffset;
1566 std::tie(LowPcOffset, LowPcEndOffset) =
1567 getAttributeOffsets(Abbrev, LowPcIdx, Offset, OrigUnit);
1570 DIE.getAttributeValueAsAddress(&OrigUnit, dwarf::DW_AT_low_pc, -1ULL);
1571 assert(LowPc != -1ULL && "low_pc attribute is not an address.");
1572 if (LowPc == -1ULL ||
1573 !hasValidRelocation(LowPcOffset, LowPcEndOffset, MyInfo))
1576 if (Options.Verbose)
1577 DIE.dump(outs(), const_cast<DWARFUnit *>(&OrigUnit), 0, 8 /* Indent */);
1581 DWARFFormValue HighPcValue;
1582 if (!DIE.getAttributeValue(&OrigUnit, dwarf::DW_AT_high_pc, HighPcValue)) {
1583 reportWarning("Function without high_pc. Range will be discarded.\n",
1589 if (HighPcValue.isFormClass(DWARFFormValue::FC_Address)) {
1590 HighPc = *HighPcValue.getAsAddress(&OrigUnit);
1592 assert(HighPcValue.isFormClass(DWARFFormValue::FC_Constant));
1593 HighPc = LowPc + *HighPcValue.getAsUnsignedConstant();
1596 // Replace the debug map range with a more accurate one.
1597 Ranges[LowPc] = std::make_pair(HighPc, MyInfo.AddrAdjust);
1598 Unit.addFunctionRange(LowPc, HighPc, MyInfo.AddrAdjust);
1602 /// \brief Check if a DIE should be kept.
1603 /// \returns updated TraversalFlags.
1604 unsigned DwarfLinker::shouldKeepDIE(const DWARFDebugInfoEntryMinimal &DIE,
1606 CompileUnit::DIEInfo &MyInfo,
1608 switch (DIE.getTag()) {
1609 case dwarf::DW_TAG_constant:
1610 case dwarf::DW_TAG_variable:
1611 return shouldKeepVariableDIE(DIE, Unit, MyInfo, Flags);
1612 case dwarf::DW_TAG_subprogram:
1613 return shouldKeepSubprogramDIE(DIE, Unit, MyInfo, Flags);
1614 case dwarf::DW_TAG_module:
1615 case dwarf::DW_TAG_imported_module:
1616 case dwarf::DW_TAG_imported_declaration:
1617 case dwarf::DW_TAG_imported_unit:
1618 // We always want to keep these.
1619 return Flags | TF_Keep;
1625 /// \brief Mark the passed DIE as well as all the ones it depends on
1628 /// This function is called by lookForDIEsToKeep on DIEs that are
1629 /// newly discovered to be needed in the link. It recursively calls
1630 /// back to lookForDIEsToKeep while adding TF_DependencyWalk to the
1631 /// TraversalFlags to inform it that it's not doing the primary DIE
1633 void DwarfLinker::keepDIEAndDenpendencies(const DWARFDebugInfoEntryMinimal &DIE,
1634 CompileUnit::DIEInfo &MyInfo,
1635 const DebugMapObject &DMO,
1636 CompileUnit &CU, unsigned Flags) {
1637 const DWARFUnit &Unit = CU.getOrigUnit();
1640 // First mark all the parent chain as kept.
1641 unsigned AncestorIdx = MyInfo.ParentIdx;
1642 while (!CU.getInfo(AncestorIdx).Keep) {
1643 lookForDIEsToKeep(*Unit.getDIEAtIndex(AncestorIdx), DMO, CU,
1644 TF_ParentWalk | TF_Keep | TF_DependencyWalk);
1645 AncestorIdx = CU.getInfo(AncestorIdx).ParentIdx;
1648 // Then we need to mark all the DIEs referenced by this DIE's
1649 // attributes as kept.
1650 DataExtractor Data = Unit.getDebugInfoExtractor();
1651 const auto *Abbrev = DIE.getAbbreviationDeclarationPtr();
1652 uint32_t Offset = DIE.getOffset() + getULEB128Size(Abbrev->getCode());
1654 // Mark all DIEs referenced through atttributes as kept.
1655 for (const auto &AttrSpec : Abbrev->attributes()) {
1656 DWARFFormValue Val(AttrSpec.Form);
1658 if (!Val.isFormClass(DWARFFormValue::FC_Reference)) {
1659 DWARFFormValue::skipValue(AttrSpec.Form, Data, &Offset, &Unit);
1663 Val.extractValue(Data, &Offset, &Unit);
1664 CompileUnit *ReferencedCU;
1665 if (const auto *RefDIE = resolveDIEReference(Val, Unit, DIE, ReferencedCU))
1666 lookForDIEsToKeep(*RefDIE, DMO, *ReferencedCU,
1667 TF_Keep | TF_DependencyWalk);
1671 /// \brief Recursively walk the \p DIE tree and look for DIEs to
1672 /// keep. Store that information in \p CU's DIEInfo.
1674 /// This function is the entry point of the DIE selection
1675 /// algorithm. It is expected to walk the DIE tree in file order and
1676 /// (though the mediation of its helper) call hasValidRelocation() on
1677 /// each DIE that might be a 'root DIE' (See DwarfLinker class
1679 /// While walking the dependencies of root DIEs, this function is
1680 /// also called, but during these dependency walks the file order is
1681 /// not respected. The TF_DependencyWalk flag tells us which kind of
1682 /// traversal we are currently doing.
1683 void DwarfLinker::lookForDIEsToKeep(const DWARFDebugInfoEntryMinimal &DIE,
1684 const DebugMapObject &DMO, CompileUnit &CU,
1686 unsigned Idx = CU.getOrigUnit().getDIEIndex(&DIE);
1687 CompileUnit::DIEInfo &MyInfo = CU.getInfo(Idx);
1688 bool AlreadyKept = MyInfo.Keep;
1690 // If the Keep flag is set, we are marking a required DIE's
1691 // dependencies. If our target is already marked as kept, we're all
1693 if ((Flags & TF_DependencyWalk) && AlreadyKept)
1696 // We must not call shouldKeepDIE while called from keepDIEAndDenpendencies,
1697 // because it would screw up the relocation finding logic.
1698 if (!(Flags & TF_DependencyWalk))
1699 Flags = shouldKeepDIE(DIE, CU, MyInfo, Flags);
1701 // If it is a newly kept DIE mark it as well as all its dependencies as kept.
1702 if (!AlreadyKept && (Flags & TF_Keep))
1703 keepDIEAndDenpendencies(DIE, MyInfo, DMO, CU, Flags);
1705 // The TF_ParentWalk flag tells us that we are currently walking up
1706 // the parent chain of a required DIE, and we don't want to mark all
1707 // the children of the parents as kept (consider for example a
1708 // DW_TAG_namespace node in the parent chain). There are however a
1709 // set of DIE types for which we want to ignore that directive and still
1710 // walk their children.
1711 if (dieNeedsChildrenToBeMeaningful(DIE.getTag()))
1712 Flags &= ~TF_ParentWalk;
1714 if (!DIE.hasChildren() || (Flags & TF_ParentWalk))
1717 for (auto *Child = DIE.getFirstChild(); Child && !Child->isNULL();
1718 Child = Child->getSibling())
1719 lookForDIEsToKeep(*Child, DMO, CU, Flags);
1722 /// \brief Assign an abbreviation numer to \p Abbrev.
1724 /// Our DIEs get freed after every DebugMapObject has been processed,
1725 /// thus the FoldingSet we use to unique DIEAbbrevs cannot refer to
1726 /// the instances hold by the DIEs. When we encounter an abbreviation
1727 /// that we don't know, we create a permanent copy of it.
1728 void DwarfLinker::AssignAbbrev(DIEAbbrev &Abbrev) {
1729 // Check the set for priors.
1730 FoldingSetNodeID ID;
1733 DIEAbbrev *InSet = AbbreviationsSet.FindNodeOrInsertPos(ID, InsertToken);
1735 // If it's newly added.
1737 // Assign existing abbreviation number.
1738 Abbrev.setNumber(InSet->getNumber());
1740 // Add to abbreviation list.
1741 Abbreviations.push_back(
1742 new DIEAbbrev(Abbrev.getTag(), Abbrev.hasChildren()));
1743 for (const auto &Attr : Abbrev.getData())
1744 Abbreviations.back()->AddAttribute(Attr.getAttribute(), Attr.getForm());
1745 AbbreviationsSet.InsertNode(Abbreviations.back(), InsertToken);
1746 // Assign the unique abbreviation number.
1747 Abbrev.setNumber(Abbreviations.size());
1748 Abbreviations.back()->setNumber(Abbreviations.size());
1752 /// \brief Clone a string attribute described by \p AttrSpec and add
1754 /// \returns the size of the new attribute.
1755 unsigned DwarfLinker::cloneStringAttribute(DIE &Die, AttributeSpec AttrSpec,
1756 const DWARFFormValue &Val,
1757 const DWARFUnit &U) {
1758 // Switch everything to out of line strings.
1759 const char *String = *Val.getAsCString(&U);
1760 unsigned Offset = StringPool.getStringOffset(String);
1761 Die.addValue(dwarf::Attribute(AttrSpec.Attr), dwarf::DW_FORM_strp,
1762 new (DIEAlloc) DIEInteger(Offset));
1766 /// \brief Clone an attribute referencing another DIE and add
1768 /// \returns the size of the new attribute.
1769 unsigned DwarfLinker::cloneDieReferenceAttribute(
1770 DIE &Die, const DWARFDebugInfoEntryMinimal &InputDIE,
1771 AttributeSpec AttrSpec, unsigned AttrSize, const DWARFFormValue &Val,
1772 CompileUnit &Unit) {
1773 uint32_t Ref = *Val.getAsReference(&Unit.getOrigUnit());
1774 DIE *NewRefDie = nullptr;
1775 CompileUnit *RefUnit = nullptr;
1776 const DWARFDebugInfoEntryMinimal *RefDie = nullptr;
1778 if (!(RefUnit = getUnitForOffset(Ref)) ||
1779 !(RefDie = RefUnit->getOrigUnit().getDIEForOffset(Ref))) {
1780 const char *AttributeString = dwarf::AttributeString(AttrSpec.Attr);
1781 if (!AttributeString)
1782 AttributeString = "DW_AT_???";
1783 reportWarning(Twine("Missing DIE for ref in attribute ") + AttributeString +
1785 &Unit.getOrigUnit(), &InputDIE);
1789 unsigned Idx = RefUnit->getOrigUnit().getDIEIndex(RefDie);
1790 CompileUnit::DIEInfo &RefInfo = RefUnit->getInfo(Idx);
1791 if (!RefInfo.Clone) {
1792 assert(Ref > InputDIE.getOffset());
1793 // We haven't cloned this DIE yet. Just create an empty one and
1794 // store it. It'll get really cloned when we process it.
1795 RefInfo.Clone = new DIE(dwarf::Tag(RefDie->getTag()));
1797 NewRefDie = RefInfo.Clone;
1799 if (AttrSpec.Form == dwarf::DW_FORM_ref_addr) {
1800 // We cannot currently rely on a DIEEntry to emit ref_addr
1801 // references, because the implementation calls back to DwarfDebug
1802 // to find the unit offset. (We don't have a DwarfDebug)
1803 // FIXME: we should be able to design DIEEntry reliance on
1806 if (Ref < InputDIE.getOffset()) {
1807 // We must have already cloned that DIE.
1808 uint32_t NewRefOffset =
1809 RefUnit->getStartOffset() + NewRefDie->getOffset();
1810 Attr = new (DIEAlloc) DIEInteger(NewRefOffset);
1812 // A forward reference. Note and fixup later.
1813 Attr = new (DIEAlloc) DIEInteger(0xBADDEF);
1814 Unit.noteForwardReference(NewRefDie, RefUnit, Attr);
1816 Die.addValue(dwarf::Attribute(AttrSpec.Attr), dwarf::DW_FORM_ref_addr,
1821 Die.addValue(dwarf::Attribute(AttrSpec.Attr), dwarf::Form(AttrSpec.Form),
1822 new (DIEAlloc) DIEEntry(*NewRefDie));
1826 /// \brief Clone an attribute of block form (locations, constants) and add
1828 /// \returns the size of the new attribute.
1829 unsigned DwarfLinker::cloneBlockAttribute(DIE &Die, AttributeSpec AttrSpec,
1830 const DWARFFormValue &Val,
1831 unsigned AttrSize) {
1834 DIELoc *Loc = nullptr;
1835 DIEBlock *Block = nullptr;
1836 // Just copy the block data over.
1837 if (AttrSpec.Form == dwarf::DW_FORM_exprloc) {
1838 Loc = new (DIEAlloc) DIELoc();
1839 DIELocs.push_back(Loc);
1841 Block = new (DIEAlloc) DIEBlock();
1842 DIEBlocks.push_back(Block);
1844 Attr = Loc ? static_cast<DIE *>(Loc) : static_cast<DIE *>(Block);
1845 Value = Loc ? static_cast<DIEValue *>(Loc) : static_cast<DIEValue *>(Block);
1846 ArrayRef<uint8_t> Bytes = *Val.getAsBlock();
1847 for (auto Byte : Bytes)
1848 Attr->addValue(static_cast<dwarf::Attribute>(0), dwarf::DW_FORM_data1,
1849 new (DIEAlloc) DIEInteger(Byte));
1850 // FIXME: If DIEBlock and DIELoc just reuses the Size field of
1851 // the DIE class, this if could be replaced by
1852 // Attr->setSize(Bytes.size()).
1855 Loc->ComputeSize(&Streamer->getAsmPrinter());
1857 Block->ComputeSize(&Streamer->getAsmPrinter());
1859 Die.addValue(dwarf::Attribute(AttrSpec.Attr), dwarf::Form(AttrSpec.Form),
1864 /// \brief Clone an address attribute and add it to \p Die.
1865 /// \returns the size of the new attribute.
1866 unsigned DwarfLinker::cloneAddressAttribute(DIE &Die, AttributeSpec AttrSpec,
1867 const DWARFFormValue &Val,
1868 const CompileUnit &Unit,
1869 AttributesInfo &Info) {
1870 uint64_t Addr = *Val.getAsAddress(&Unit.getOrigUnit());
1871 if (AttrSpec.Attr == dwarf::DW_AT_low_pc) {
1872 if (Die.getTag() == dwarf::DW_TAG_inlined_subroutine ||
1873 Die.getTag() == dwarf::DW_TAG_lexical_block)
1874 Addr += Info.PCOffset;
1875 else if (Die.getTag() == dwarf::DW_TAG_compile_unit) {
1876 Addr = Unit.getLowPc();
1877 if (Addr == UINT64_MAX)
1880 Info.HasLowPc = true;
1881 } else if (AttrSpec.Attr == dwarf::DW_AT_high_pc) {
1882 if (Die.getTag() == dwarf::DW_TAG_compile_unit) {
1883 if (uint64_t HighPc = Unit.getHighPc())
1888 // If we have a high_pc recorded for the input DIE, use
1889 // it. Otherwise (when no relocations where applied) just use the
1890 // one we just decoded.
1891 Addr = (Info.OrigHighPc ? Info.OrigHighPc : Addr) + Info.PCOffset;
1894 Die.addValue(static_cast<dwarf::Attribute>(AttrSpec.Attr),
1895 static_cast<dwarf::Form>(AttrSpec.Form),
1896 new (DIEAlloc) DIEInteger(Addr));
1897 return Unit.getOrigUnit().getAddressByteSize();
1900 /// \brief Clone a scalar attribute and add it to \p Die.
1901 /// \returns the size of the new attribute.
1902 unsigned DwarfLinker::cloneScalarAttribute(
1903 DIE &Die, const DWARFDebugInfoEntryMinimal &InputDIE, CompileUnit &Unit,
1904 AttributeSpec AttrSpec, const DWARFFormValue &Val, unsigned AttrSize,
1905 AttributesInfo &Info) {
1907 if (AttrSpec.Attr == dwarf::DW_AT_high_pc &&
1908 Die.getTag() == dwarf::DW_TAG_compile_unit) {
1909 if (Unit.getLowPc() == -1ULL)
1911 // Dwarf >= 4 high_pc is an size, not an address.
1912 Value = Unit.getHighPc() - Unit.getLowPc();
1913 } else if (AttrSpec.Form == dwarf::DW_FORM_sec_offset)
1914 Value = *Val.getAsSectionOffset();
1915 else if (AttrSpec.Form == dwarf::DW_FORM_sdata)
1916 Value = *Val.getAsSignedConstant();
1917 else if (auto OptionalValue = Val.getAsUnsignedConstant())
1918 Value = *OptionalValue;
1920 reportWarning("Unsupported scalar attribute form. Dropping attribute.",
1921 &Unit.getOrigUnit(), &InputDIE);
1924 DIEInteger *Attr = new (DIEAlloc) DIEInteger(Value);
1925 if (AttrSpec.Attr == dwarf::DW_AT_ranges)
1926 Unit.noteRangeAttribute(Die, Attr);
1927 // A more generic way to check for location attributes would be
1928 // nice, but it's very unlikely that any other attribute needs a
1930 else if (AttrSpec.Attr == dwarf::DW_AT_location ||
1931 AttrSpec.Attr == dwarf::DW_AT_frame_base)
1932 Unit.noteLocationAttribute(Attr, Info.PCOffset);
1933 else if (AttrSpec.Attr == dwarf::DW_AT_declaration && Value)
1934 Info.IsDeclaration = true;
1936 Die.addValue(dwarf::Attribute(AttrSpec.Attr), dwarf::Form(AttrSpec.Form),
1941 /// \brief Clone \p InputDIE's attribute described by \p AttrSpec with
1942 /// value \p Val, and add it to \p Die.
1943 /// \returns the size of the cloned attribute.
1944 unsigned DwarfLinker::cloneAttribute(DIE &Die,
1945 const DWARFDebugInfoEntryMinimal &InputDIE,
1947 const DWARFFormValue &Val,
1948 const AttributeSpec AttrSpec,
1949 unsigned AttrSize, AttributesInfo &Info) {
1950 const DWARFUnit &U = Unit.getOrigUnit();
1952 switch (AttrSpec.Form) {
1953 case dwarf::DW_FORM_strp:
1954 case dwarf::DW_FORM_string:
1955 return cloneStringAttribute(Die, AttrSpec, Val, U);
1956 case dwarf::DW_FORM_ref_addr:
1957 case dwarf::DW_FORM_ref1:
1958 case dwarf::DW_FORM_ref2:
1959 case dwarf::DW_FORM_ref4:
1960 case dwarf::DW_FORM_ref8:
1961 return cloneDieReferenceAttribute(Die, InputDIE, AttrSpec, AttrSize, Val,
1963 case dwarf::DW_FORM_block:
1964 case dwarf::DW_FORM_block1:
1965 case dwarf::DW_FORM_block2:
1966 case dwarf::DW_FORM_block4:
1967 case dwarf::DW_FORM_exprloc:
1968 return cloneBlockAttribute(Die, AttrSpec, Val, AttrSize);
1969 case dwarf::DW_FORM_addr:
1970 return cloneAddressAttribute(Die, AttrSpec, Val, Unit, Info);
1971 case dwarf::DW_FORM_data1:
1972 case dwarf::DW_FORM_data2:
1973 case dwarf::DW_FORM_data4:
1974 case dwarf::DW_FORM_data8:
1975 case dwarf::DW_FORM_udata:
1976 case dwarf::DW_FORM_sdata:
1977 case dwarf::DW_FORM_sec_offset:
1978 case dwarf::DW_FORM_flag:
1979 case dwarf::DW_FORM_flag_present:
1980 return cloneScalarAttribute(Die, InputDIE, Unit, AttrSpec, Val, AttrSize,
1983 reportWarning("Unsupported attribute form in cloneAttribute. Dropping.", &U,
1990 /// \brief Apply the valid relocations found by findValidRelocs() to
1991 /// the buffer \p Data, taking into account that Data is at \p BaseOffset
1992 /// in the debug_info section.
1994 /// Like for findValidRelocs(), this function must be called with
1995 /// monotonic \p BaseOffset values.
1997 /// \returns wether any reloc has been applied.
1998 bool DwarfLinker::applyValidRelocs(MutableArrayRef<char> Data,
1999 uint32_t BaseOffset, bool isLittleEndian) {
2000 assert((NextValidReloc == 0 ||
2001 BaseOffset > ValidRelocs[NextValidReloc - 1].Offset) &&
2002 "BaseOffset should only be increasing.");
2003 if (NextValidReloc >= ValidRelocs.size())
2006 // Skip relocs that haven't been applied.
2007 while (NextValidReloc < ValidRelocs.size() &&
2008 ValidRelocs[NextValidReloc].Offset < BaseOffset)
2011 bool Applied = false;
2012 uint64_t EndOffset = BaseOffset + Data.size();
2013 while (NextValidReloc < ValidRelocs.size() &&
2014 ValidRelocs[NextValidReloc].Offset >= BaseOffset &&
2015 ValidRelocs[NextValidReloc].Offset < EndOffset) {
2016 const auto &ValidReloc = ValidRelocs[NextValidReloc++];
2017 assert(ValidReloc.Offset - BaseOffset < Data.size());
2018 assert(ValidReloc.Offset - BaseOffset + ValidReloc.Size <= Data.size());
2020 uint64_t Value = ValidReloc.Mapping->getValue().BinaryAddress;
2021 Value += ValidReloc.Addend;
2022 for (unsigned i = 0; i != ValidReloc.Size; ++i) {
2023 unsigned Index = isLittleEndian ? i : (ValidReloc.Size - i - 1);
2024 Buf[i] = uint8_t(Value >> (Index * 8));
2026 assert(ValidReloc.Size <= sizeof(Buf));
2027 memcpy(&Data[ValidReloc.Offset - BaseOffset], Buf, ValidReloc.Size);
2034 static bool isTypeTag(uint16_t Tag) {
2036 case dwarf::DW_TAG_array_type:
2037 case dwarf::DW_TAG_class_type:
2038 case dwarf::DW_TAG_enumeration_type:
2039 case dwarf::DW_TAG_pointer_type:
2040 case dwarf::DW_TAG_reference_type:
2041 case dwarf::DW_TAG_string_type:
2042 case dwarf::DW_TAG_structure_type:
2043 case dwarf::DW_TAG_subroutine_type:
2044 case dwarf::DW_TAG_typedef:
2045 case dwarf::DW_TAG_union_type:
2046 case dwarf::DW_TAG_ptr_to_member_type:
2047 case dwarf::DW_TAG_set_type:
2048 case dwarf::DW_TAG_subrange_type:
2049 case dwarf::DW_TAG_base_type:
2050 case dwarf::DW_TAG_const_type:
2051 case dwarf::DW_TAG_constant:
2052 case dwarf::DW_TAG_file_type:
2053 case dwarf::DW_TAG_namelist:
2054 case dwarf::DW_TAG_packed_type:
2055 case dwarf::DW_TAG_volatile_type:
2056 case dwarf::DW_TAG_restrict_type:
2057 case dwarf::DW_TAG_interface_type:
2058 case dwarf::DW_TAG_unspecified_type:
2059 case dwarf::DW_TAG_shared_type:
2067 /// \brief Recursively clone \p InputDIE's subtrees that have been
2068 /// selected to appear in the linked output.
2070 /// \param OutOffset is the Offset where the newly created DIE will
2071 /// lie in the linked compile unit.
2073 /// \returns the cloned DIE object or null if nothing was selected.
2074 DIE *DwarfLinker::cloneDIE(const DWARFDebugInfoEntryMinimal &InputDIE,
2075 CompileUnit &Unit, int64_t PCOffset,
2076 uint32_t OutOffset) {
2077 DWARFUnit &U = Unit.getOrigUnit();
2078 unsigned Idx = U.getDIEIndex(&InputDIE);
2079 CompileUnit::DIEInfo &Info = Unit.getInfo(Idx);
2081 // Should the DIE appear in the output?
2082 if (!Unit.getInfo(Idx).Keep)
2085 uint32_t Offset = InputDIE.getOffset();
2086 // The DIE might have been already created by a forward reference
2087 // (see cloneDieReferenceAttribute()).
2088 DIE *Die = Info.Clone;
2090 Die = Info.Clone = new DIE(dwarf::Tag(InputDIE.getTag()));
2091 assert(Die->getTag() == InputDIE.getTag());
2092 Die->setOffset(OutOffset);
2094 // Extract and clone every attribute.
2095 DataExtractor Data = U.getDebugInfoExtractor();
2096 uint32_t NextOffset = U.getDIEAtIndex(Idx + 1)->getOffset();
2097 AttributesInfo AttrInfo;
2099 // We could copy the data only if we need to aply a relocation to
2100 // it. After testing, it seems there is no performance downside to
2101 // doing the copy unconditionally, and it makes the code simpler.
2102 SmallString<40> DIECopy(Data.getData().substr(Offset, NextOffset - Offset));
2103 Data = DataExtractor(DIECopy, Data.isLittleEndian(), Data.getAddressSize());
2104 // Modify the copy with relocated addresses.
2105 if (applyValidRelocs(DIECopy, Offset, Data.isLittleEndian())) {
2106 // If we applied relocations, we store the value of high_pc that was
2107 // potentially stored in the input DIE. If high_pc is an address
2108 // (Dwarf version == 2), then it might have been relocated to a
2109 // totally unrelated value (because the end address in the object
2110 // file might be start address of another function which got moved
2111 // independantly by the linker). The computation of the actual
2112 // high_pc value is done in cloneAddressAttribute().
2113 AttrInfo.OrigHighPc =
2114 InputDIE.getAttributeValueAsAddress(&U, dwarf::DW_AT_high_pc, 0);
2117 // Reset the Offset to 0 as we will be working on the local copy of
2121 const auto *Abbrev = InputDIE.getAbbreviationDeclarationPtr();
2122 Offset += getULEB128Size(Abbrev->getCode());
2124 // We are entering a subprogram. Get and propagate the PCOffset.
2125 if (Die->getTag() == dwarf::DW_TAG_subprogram)
2126 PCOffset = Info.AddrAdjust;
2127 AttrInfo.PCOffset = PCOffset;
2129 for (const auto &AttrSpec : Abbrev->attributes()) {
2130 DWARFFormValue Val(AttrSpec.Form);
2131 uint32_t AttrSize = Offset;
2132 Val.extractValue(Data, &Offset, &U);
2133 AttrSize = Offset - AttrSize;
2136 cloneAttribute(*Die, InputDIE, Unit, Val, AttrSpec, AttrSize, AttrInfo);
2139 // Look for accelerator entries.
2140 uint16_t Tag = InputDIE.getTag();
2141 // FIXME: This is slightly wrong. An inline_subroutine without a
2142 // low_pc, but with AT_ranges might be interesting to get into the
2143 // accelerator tables too. For now stick with dsymutil's behavior.
2144 if ((Info.InDebugMap || AttrInfo.HasLowPc) &&
2145 Tag != dwarf::DW_TAG_compile_unit &&
2146 getDIENames(InputDIE, Unit.getOrigUnit(), AttrInfo)) {
2147 if (AttrInfo.MangledName && AttrInfo.MangledName != AttrInfo.Name)
2148 Unit.addNameAccelerator(Die, AttrInfo.MangledName,
2149 AttrInfo.MangledNameOffset,
2150 Tag == dwarf::DW_TAG_inlined_subroutine);
2152 Unit.addNameAccelerator(Die, AttrInfo.Name, AttrInfo.NameOffset,
2153 Tag == dwarf::DW_TAG_inlined_subroutine);
2154 } else if (isTypeTag(Tag) && !AttrInfo.IsDeclaration &&
2155 getDIENames(InputDIE, Unit.getOrigUnit(), AttrInfo)) {
2156 Unit.addTypeAccelerator(Die, AttrInfo.Name, AttrInfo.NameOffset);
2159 DIEAbbrev &NewAbbrev = Die->getAbbrev();
2160 // If a scope DIE is kept, we must have kept at least one child. If
2161 // it's not the case, we'll just be emitting one wasteful end of
2162 // children marker, but things won't break.
2163 if (InputDIE.hasChildren())
2164 NewAbbrev.setChildrenFlag(dwarf::DW_CHILDREN_yes);
2165 // Assign a permanent abbrev number
2166 AssignAbbrev(Die->getAbbrev());
2168 // Add the size of the abbreviation number to the output offset.
2169 OutOffset += getULEB128Size(Die->getAbbrevNumber());
2171 if (!Abbrev->hasChildren()) {
2173 Die->setSize(OutOffset - Die->getOffset());
2177 // Recursively clone children.
2178 for (auto *Child = InputDIE.getFirstChild(); Child && !Child->isNULL();
2179 Child = Child->getSibling()) {
2180 if (DIE *Clone = cloneDIE(*Child, Unit, PCOffset, OutOffset)) {
2181 Die->addChild(std::unique_ptr<DIE>(Clone));
2182 OutOffset = Clone->getOffset() + Clone->getSize();
2186 // Account for the end of children marker.
2187 OutOffset += sizeof(int8_t);
2189 Die->setSize(OutOffset - Die->getOffset());
2193 /// \brief Patch the input object file relevant debug_ranges entries
2194 /// and emit them in the output file. Update the relevant attributes
2195 /// to point at the new entries.
2196 void DwarfLinker::patchRangesForUnit(const CompileUnit &Unit,
2197 DWARFContext &OrigDwarf) const {
2198 DWARFDebugRangeList RangeList;
2199 const auto &FunctionRanges = Unit.getFunctionRanges();
2200 unsigned AddressSize = Unit.getOrigUnit().getAddressByteSize();
2201 DataExtractor RangeExtractor(OrigDwarf.getRangeSection(),
2202 OrigDwarf.isLittleEndian(), AddressSize);
2203 auto InvalidRange = FunctionRanges.end(), CurrRange = InvalidRange;
2204 DWARFUnit &OrigUnit = Unit.getOrigUnit();
2205 const auto *OrigUnitDie = OrigUnit.getCompileUnitDIE(false);
2206 uint64_t OrigLowPc = OrigUnitDie->getAttributeValueAsAddress(
2207 &OrigUnit, dwarf::DW_AT_low_pc, -1ULL);
2208 // Ranges addresses are based on the unit's low_pc. Compute the
2209 // offset we need to apply to adapt to the the new unit's low_pc.
2210 int64_t UnitPcOffset = 0;
2211 if (OrigLowPc != -1ULL)
2212 UnitPcOffset = int64_t(OrigLowPc) - Unit.getLowPc();
2214 for (const auto &RangeAttribute : Unit.getRangesAttributes()) {
2215 uint32_t Offset = RangeAttribute->getValue();
2216 RangeAttribute->setValue(Streamer->getRangesSectionSize());
2217 RangeList.extract(RangeExtractor, &Offset);
2218 const auto &Entries = RangeList.getEntries();
2219 const DWARFDebugRangeList::RangeListEntry &First = Entries.front();
2221 if (CurrRange == InvalidRange || First.StartAddress < CurrRange.start() ||
2222 First.StartAddress >= CurrRange.stop()) {
2223 CurrRange = FunctionRanges.find(First.StartAddress + OrigLowPc);
2224 if (CurrRange == InvalidRange ||
2225 CurrRange.start() > First.StartAddress + OrigLowPc) {
2226 reportWarning("no mapping for range.");
2231 Streamer->emitRangesEntries(UnitPcOffset, OrigLowPc, CurrRange, Entries,
2236 /// \brief Generate the debug_aranges entries for \p Unit and if the
2237 /// unit has a DW_AT_ranges attribute, also emit the debug_ranges
2238 /// contribution for this attribute.
2239 /// FIXME: this could actually be done right in patchRangesForUnit,
2240 /// but for the sake of initial bit-for-bit compatibility with legacy
2241 /// dsymutil, we have to do it in a delayed pass.
2242 void DwarfLinker::generateUnitRanges(CompileUnit &Unit) const {
2243 DIEInteger *Attr = Unit.getUnitRangesAttribute();
2245 Attr->setValue(Streamer->getRangesSectionSize());
2246 Streamer->emitUnitRangesEntries(Unit, Attr != nullptr);
2249 /// \brief Insert the new line info sequence \p Seq into the current
2250 /// set of already linked line info \p Rows.
2251 static void insertLineSequence(std::vector<DWARFDebugLine::Row> &Seq,
2252 std::vector<DWARFDebugLine::Row> &Rows) {
2256 if (!Rows.empty() && Rows.back().Address < Seq.front().Address) {
2257 Rows.insert(Rows.end(), Seq.begin(), Seq.end());
2262 auto InsertPoint = std::lower_bound(
2263 Rows.begin(), Rows.end(), Seq.front(),
2264 [](const DWARFDebugLine::Row &LHS, const DWARFDebugLine::Row &RHS) {
2265 return LHS.Address < RHS.Address;
2268 // FIXME: this only removes the unneeded end_sequence if the
2269 // sequences have been inserted in order. using a global sort like
2270 // described in patchLineTableForUnit() and delaying the end_sequene
2271 // elimination to emitLineTableForUnit() we can get rid of all of them.
2272 if (InsertPoint != Rows.end() &&
2273 InsertPoint->Address == Seq.front().Address && InsertPoint->EndSequence) {
2274 *InsertPoint = Seq.front();
2275 Rows.insert(InsertPoint + 1, Seq.begin() + 1, Seq.end());
2277 Rows.insert(InsertPoint, Seq.begin(), Seq.end());
2283 /// \brief Extract the line table for \p Unit from \p OrigDwarf, and
2284 /// recreate a relocated version of these for the address ranges that
2285 /// are present in the binary.
2286 void DwarfLinker::patchLineTableForUnit(CompileUnit &Unit,
2287 DWARFContext &OrigDwarf) {
2288 const DWARFDebugInfoEntryMinimal *CUDie =
2289 Unit.getOrigUnit().getCompileUnitDIE();
2290 uint64_t StmtList = CUDie->getAttributeValueAsSectionOffset(
2291 &Unit.getOrigUnit(), dwarf::DW_AT_stmt_list, -1ULL);
2292 if (StmtList == -1ULL)
2295 // Update the cloned DW_AT_stmt_list with the correct debug_line offset.
2296 if (auto *OutputDIE = Unit.getOutputUnitDIE()) {
2297 const auto &Abbrev = OutputDIE->getAbbrev().getData();
2298 auto Stmt = std::find_if(
2299 Abbrev.begin(), Abbrev.end(), [](const DIEAbbrevData &AbbrevData) {
2300 return AbbrevData.getAttribute() == dwarf::DW_AT_stmt_list;
2302 assert(Stmt < Abbrev.end() && "Didn't find DW_AT_stmt_list in cloned DIE!");
2303 DIEInteger *StmtAttr =
2304 cast<DIEInteger>(OutputDIE->getValues()[Stmt - Abbrev.begin()]);
2305 StmtAttr->setValue(Streamer->getLineSectionSize());
2308 // Parse the original line info for the unit.
2309 DWARFDebugLine::LineTable LineTable;
2310 uint32_t StmtOffset = StmtList;
2311 StringRef LineData = OrigDwarf.getLineSection().Data;
2312 DataExtractor LineExtractor(LineData, OrigDwarf.isLittleEndian(),
2313 Unit.getOrigUnit().getAddressByteSize());
2314 LineTable.parse(LineExtractor, &OrigDwarf.getLineSection().Relocs,
2317 // This vector is the output line table.
2318 std::vector<DWARFDebugLine::Row> NewRows;
2319 NewRows.reserve(LineTable.Rows.size());
2321 // Current sequence of rows being extracted, before being inserted
2323 std::vector<DWARFDebugLine::Row> Seq;
2324 const auto &FunctionRanges = Unit.getFunctionRanges();
2325 auto InvalidRange = FunctionRanges.end(), CurrRange = InvalidRange;
2327 // FIXME: This logic is meant to generate exactly the same output as
2328 // Darwin's classic dsynutil. There is a nicer way to implement this
2329 // by simply putting all the relocated line info in NewRows and simply
2330 // sorting NewRows before passing it to emitLineTableForUnit. This
2331 // should be correct as sequences for a function should stay
2332 // together in the sorted output. There are a few corner cases that
2333 // look suspicious though, and that required to implement the logic
2334 // this way. Revisit that once initial validation is finished.
2336 // Iterate over the object file line info and extract the sequences
2337 // that correspond to linked functions.
2338 for (auto &Row : LineTable.Rows) {
2339 // Check wether we stepped out of the range. The range is
2340 // half-open, but consider accept the end address of the range if
2341 // it is marked as end_sequence in the input (because in that
2342 // case, the relocation offset is accurate and that entry won't
2343 // serve as the start of another function).
2344 if (CurrRange == InvalidRange || Row.Address < CurrRange.start() ||
2345 Row.Address > CurrRange.stop() ||
2346 (Row.Address == CurrRange.stop() && !Row.EndSequence)) {
2347 // We just stepped out of a known range. Insert a end_sequence
2348 // corresponding to the end of the range.
2349 uint64_t StopAddress = CurrRange != InvalidRange
2350 ? CurrRange.stop() + CurrRange.value()
2352 CurrRange = FunctionRanges.find(Row.Address);
2353 bool CurrRangeValid =
2354 CurrRange != InvalidRange && CurrRange.start() <= Row.Address;
2355 if (!CurrRangeValid) {
2356 CurrRange = InvalidRange;
2357 if (StopAddress != -1ULL) {
2358 // Try harder by looking in the DebugMapObject function
2359 // ranges map. There are corner cases where this finds a
2360 // valid entry. It's unclear if this is right or wrong, but
2361 // for now do as dsymutil.
2362 // FIXME: Understand exactly what cases this addresses and
2363 // potentially remove it along with the Ranges map.
2364 auto Range = Ranges.lower_bound(Row.Address);
2365 if (Range != Ranges.begin() && Range != Ranges.end())
2368 if (Range != Ranges.end() && Range->first <= Row.Address &&
2369 Range->second.first >= Row.Address) {
2370 StopAddress = Row.Address + Range->second.second;
2374 if (StopAddress != -1ULL && !Seq.empty()) {
2375 // Insert end sequence row with the computed end address, but
2376 // the same line as the previous one.
2377 Seq.emplace_back(Seq.back());
2378 Seq.back().Address = StopAddress;
2379 Seq.back().EndSequence = 1;
2380 Seq.back().PrologueEnd = 0;
2381 Seq.back().BasicBlock = 0;
2382 Seq.back().EpilogueBegin = 0;
2383 insertLineSequence(Seq, NewRows);
2386 if (!CurrRangeValid)
2390 // Ignore empty sequences.
2391 if (Row.EndSequence && Seq.empty())
2394 // Relocate row address and add it to the current sequence.
2395 Row.Address += CurrRange.value();
2396 Seq.emplace_back(Row);
2398 if (Row.EndSequence)
2399 insertLineSequence(Seq, NewRows);
2402 // Finished extracting, now emit the line tables.
2403 uint32_t PrologueEnd = StmtList + 10 + LineTable.Prologue.PrologueLength;
2404 // FIXME: LLVM hardcodes it's prologue values. We just copy the
2405 // prologue over and that works because we act as both producer and
2406 // consumer. It would be nicer to have a real configurable line
2408 if (LineTable.Prologue.Version != 2 ||
2409 LineTable.Prologue.DefaultIsStmt != DWARF2_LINE_DEFAULT_IS_STMT ||
2410 LineTable.Prologue.LineBase != -5 || LineTable.Prologue.LineRange != 14 ||
2411 LineTable.Prologue.OpcodeBase != 13)
2412 reportWarning("line table paramters mismatch. Cannot emit.");
2414 Streamer->emitLineTableForUnit(LineData.slice(StmtList + 4, PrologueEnd),
2415 LineTable.Prologue.MinInstLength, NewRows,
2416 Unit.getOrigUnit().getAddressByteSize());
2419 void DwarfLinker::emitAcceleratorEntriesForUnit(CompileUnit &Unit) {
2420 Streamer->emitPubNamesForUnit(Unit);
2421 Streamer->emitPubTypesForUnit(Unit);
2424 bool DwarfLinker::link(const DebugMap &Map) {
2426 if (Map.begin() == Map.end()) {
2427 errs() << "Empty debug map.\n";
2431 if (!createStreamer(Map.getTriple(), OutputFilename))
2434 // Size of the DIEs (and headers) generated for the linked output.
2435 uint64_t OutputDebugInfoSize = 0;
2436 // A unique ID that identifies each compile unit.
2437 unsigned UnitID = 0;
2438 for (const auto &Obj : Map.objects()) {
2439 CurrentDebugObject = Obj.get();
2441 if (Options.Verbose)
2442 outs() << "DEBUG MAP OBJECT: " << Obj->getObjectFilename() << "\n";
2443 auto ErrOrObj = BinHolder.GetObjectFile(Obj->getObjectFilename());
2444 if (std::error_code EC = ErrOrObj.getError()) {
2445 reportWarning(Twine(Obj->getObjectFilename()) + ": " + EC.message());
2449 // Look for relocations that correspond to debug map entries.
2450 if (!findValidRelocsInDebugInfo(*ErrOrObj, *Obj)) {
2451 if (Options.Verbose)
2452 outs() << "No valid relocations found. Skipping.\n";
2456 // Setup access to the debug info.
2457 DWARFContextInMemory DwarfContext(*ErrOrObj);
2458 startDebugObject(DwarfContext, *Obj);
2460 // In a first phase, just read in the debug info and store the DIE
2461 // parent links that we will use during the next phase.
2462 for (const auto &CU : DwarfContext.compile_units()) {
2463 auto *CUDie = CU->getCompileUnitDIE(false);
2464 if (Options.Verbose) {
2465 outs() << "Input compilation unit:";
2466 CUDie->dump(outs(), CU.get(), 0);
2468 Units.emplace_back(*CU, UnitID++);
2469 gatherDIEParents(CUDie, 0, Units.back());
2472 // Then mark all the DIEs that need to be present in the linked
2473 // output and collect some information about them. Note that this
2474 // loop can not be merged with the previous one becaue cross-cu
2475 // references require the ParentIdx to be setup for every CU in
2476 // the object file before calling this.
2477 for (auto &CurrentUnit : Units)
2478 lookForDIEsToKeep(*CurrentUnit.getOrigUnit().getCompileUnitDIE(), *Obj,
2481 // The calls to applyValidRelocs inside cloneDIE will walk the
2482 // reloc array again (in the same way findValidRelocsInDebugInfo()
2483 // did). We need to reset the NextValidReloc index to the beginning.
2486 // Construct the output DIE tree by cloning the DIEs we chose to
2487 // keep above. If there are no valid relocs, then there's nothing
2489 if (!ValidRelocs.empty())
2490 for (auto &CurrentUnit : Units) {
2491 const auto *InputDIE = CurrentUnit.getOrigUnit().getCompileUnitDIE();
2492 CurrentUnit.setStartOffset(OutputDebugInfoSize);
2493 DIE *OutputDIE = cloneDIE(*InputDIE, CurrentUnit, 0 /* PCOffset */,
2494 11 /* Unit Header size */);
2495 CurrentUnit.setOutputUnitDIE(OutputDIE);
2496 OutputDebugInfoSize = CurrentUnit.computeNextUnitOffset();
2497 if (Options.NoOutput)
2499 // FIXME: for compatibility with the classic dsymutil, we emit
2500 // an empty line table for the unit, even if the unit doesn't
2501 // actually exist in the DIE tree.
2502 patchLineTableForUnit(CurrentUnit, DwarfContext);
2505 patchRangesForUnit(CurrentUnit, DwarfContext);
2506 Streamer->emitLocationsForUnit(CurrentUnit, DwarfContext);
2507 emitAcceleratorEntriesForUnit(CurrentUnit);
2510 // Emit all the compile unit's debug information.
2511 if (!ValidRelocs.empty() && !Options.NoOutput)
2512 for (auto &CurrentUnit : Units) {
2513 generateUnitRanges(CurrentUnit);
2514 CurrentUnit.fixupForwardReferences();
2515 Streamer->emitCompileUnitHeader(CurrentUnit);
2516 if (!CurrentUnit.getOutputUnitDIE())
2518 Streamer->emitDIE(*CurrentUnit.getOutputUnitDIE());
2521 // Clean-up before starting working on the next object.
2525 // Emit everything that's global.
2526 if (!Options.NoOutput) {
2527 Streamer->emitAbbrevs(Abbreviations);
2528 Streamer->emitStrings(StringPool);
2531 return Options.NoOutput ? true : Streamer->finish();
2535 bool linkDwarf(StringRef OutputFilename, const DebugMap &DM,
2536 const LinkOptions &Options) {
2537 DwarfLinker Linker(OutputFilename, Options);
2538 return Linker.link(DM);