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/MC/MCSubtargetInfo.h"
31 #include "llvm/Object/MachO.h"
32 #include "llvm/Support/Dwarf.h"
33 #include "llvm/Support/LEB128.h"
34 #include "llvm/Support/TargetRegistry.h"
35 #include "llvm/Target/TargetMachine.h"
36 #include "llvm/Target/TargetOptions.h"
45 void warn(const Twine &Warning, const Twine &Context) {
46 errs() << Twine("while processing ") + Context + ":\n";
47 errs() << Twine("warning: ") + Warning + "\n";
50 bool error(const Twine &Error, const Twine &Context) {
51 errs() << Twine("while processing ") + Context + ":\n";
52 errs() << Twine("error: ") + Error + "\n";
56 template <typename KeyT, typename ValT>
57 using HalfOpenIntervalMap =
58 IntervalMap<KeyT, ValT, IntervalMapImpl::NodeSizer<KeyT, ValT>::LeafSize,
59 IntervalMapHalfOpenInfo<KeyT>>;
61 typedef HalfOpenIntervalMap<uint64_t, int64_t> FunctionIntervals;
63 // FIXME: Delete this structure once DIE::Values has a stable iterator we can
65 struct PatchLocation {
69 PatchLocation() : Die(nullptr), Index(0) {}
70 PatchLocation(DIE &Die, unsigned Index) : Die(&Die), Index(Index) {}
71 PatchLocation(DIE &Die)
72 : Die(&Die), Index(std::distance(Die.values_begin(), Die.values_end())) {}
74 void set(uint64_t New) const {
76 assert((signed)Index <
77 std::distance(Die->values_begin(), Die->values_end()));
78 const auto &Old = Die->values_begin()[Index];
79 assert(Old.getType() == DIEValue::isInteger);
81 DIEValue(Old.getAttribute(), Old.getForm(), DIEInteger(New)));
84 uint64_t get() const {
86 assert((signed)Index <
87 std::distance(Die->values_begin(), Die->values_end()));
88 assert(Die->values_begin()[Index].getType() == DIEValue::isInteger);
89 return Die->values_begin()[Index].getDIEInteger().getValue();
93 /// \brief Stores all information relating to a compile unit, be it in
94 /// its original instance in the object file to its brand new cloned
95 /// and linked DIE tree.
98 /// \brief Information gathered about a DIE in the object file.
100 int64_t AddrAdjust; ///< Address offset to apply to the described entity.
101 DIE *Clone; ///< Cloned version of that DIE.
102 uint32_t ParentIdx; ///< The index of this DIE's parent.
103 bool Keep; ///< Is the DIE part of the linked output?
104 bool InDebugMap; ///< Was this DIE's entity found in the map?
107 CompileUnit(DWARFUnit &OrigUnit, unsigned ID)
108 : OrigUnit(OrigUnit), ID(ID), LowPc(UINT64_MAX), HighPc(0), RangeAlloc(),
110 Info.resize(OrigUnit.getNumDIEs());
113 CompileUnit(CompileUnit &&RHS)
114 : OrigUnit(RHS.OrigUnit), Info(std::move(RHS.Info)),
115 CUDie(std::move(RHS.CUDie)), StartOffset(RHS.StartOffset),
116 NextUnitOffset(RHS.NextUnitOffset), RangeAlloc(), Ranges(RangeAlloc) {
117 // The CompileUnit container has been 'reserve()'d with the right
118 // size. We cannot move the IntervalMap anyway.
119 llvm_unreachable("CompileUnits should not be moved.");
122 DWARFUnit &getOrigUnit() const { return OrigUnit; }
124 unsigned getUniqueID() const { return ID; }
126 DIE *getOutputUnitDIE() const { return CUDie.get(); }
127 void setOutputUnitDIE(DIE *Die) { CUDie.reset(Die); }
129 DIEInfo &getInfo(unsigned Idx) { return Info[Idx]; }
130 const DIEInfo &getInfo(unsigned Idx) const { return Info[Idx]; }
132 uint64_t getStartOffset() const { return StartOffset; }
133 uint64_t getNextUnitOffset() const { return NextUnitOffset; }
134 void setStartOffset(uint64_t DebugInfoSize) { StartOffset = DebugInfoSize; }
136 uint64_t getLowPc() const { return LowPc; }
137 uint64_t getHighPc() const { return HighPc; }
139 Optional<PatchLocation> getUnitRangesAttribute() const {
140 return UnitRangeAttribute;
142 const FunctionIntervals &getFunctionRanges() const { return Ranges; }
143 const std::vector<PatchLocation> &getRangesAttributes() const {
144 return RangeAttributes;
147 const std::vector<std::pair<PatchLocation, int64_t>> &
148 getLocationAttributes() const {
149 return LocationAttributes;
152 /// \brief Compute the end offset for this unit. Must be
153 /// called after the CU's DIEs have been cloned.
154 /// \returns the next unit offset (which is also the current
155 /// debug_info section size).
156 uint64_t computeNextUnitOffset();
158 /// \brief Keep track of a forward reference to DIE \p Die in \p
159 /// RefUnit by \p Attr. The attribute should be fixed up later to
160 /// point to the absolute offset of \p Die in the debug_info section.
161 void noteForwardReference(DIE *Die, const CompileUnit *RefUnit,
164 /// \brief Apply all fixups recored by noteForwardReference().
165 void fixupForwardReferences();
167 /// \brief Add a function range [\p LowPC, \p HighPC) that is
168 /// relocatad by applying offset \p PCOffset.
169 void addFunctionRange(uint64_t LowPC, uint64_t HighPC, int64_t PCOffset);
171 /// \brief Keep track of a DW_AT_range attribute that we will need to
173 void noteRangeAttribute(const DIE &Die, PatchLocation Attr);
175 /// \brief Keep track of a location attribute pointing to a location
176 /// list in the debug_loc section.
177 void noteLocationAttribute(PatchLocation Attr, int64_t PcOffset);
179 /// \brief Add a name accelerator entry for \p Die with \p Name
180 /// which is stored in the string table at \p Offset.
181 void addNameAccelerator(const DIE *Die, const char *Name, uint32_t Offset,
182 bool SkipPubnamesSection = false);
184 /// \brief Add a type accelerator entry for \p Die with \p Name
185 /// which is stored in the string table at \p Offset.
186 void addTypeAccelerator(const DIE *Die, const char *Name, uint32_t Offset);
189 StringRef Name; ///< Name of the entry.
190 const DIE *Die; ///< DIE this entry describes.
191 uint32_t NameOffset; ///< Offset of Name in the string pool.
192 bool SkipPubSection; ///< Emit this entry only in the apple_* sections.
194 AccelInfo(StringRef Name, const DIE *Die, uint32_t NameOffset,
195 bool SkipPubSection = false)
196 : Name(Name), Die(Die), NameOffset(NameOffset),
197 SkipPubSection(SkipPubSection) {}
200 const std::vector<AccelInfo> &getPubnames() const { return Pubnames; }
201 const std::vector<AccelInfo> &getPubtypes() const { return Pubtypes; }
206 std::vector<DIEInfo> Info; ///< DIE info indexed by DIE index.
207 std::unique_ptr<DIE> CUDie; ///< Root of the linked DIE tree.
209 uint64_t StartOffset;
210 uint64_t NextUnitOffset;
215 /// \brief A list of attributes to fixup with the absolute offset of
216 /// a DIE in the debug_info section.
218 /// The offsets for the attributes in this array couldn't be set while
219 /// cloning because for cross-cu forward refences the target DIE's
220 /// offset isn't known you emit the reference attribute.
221 std::vector<std::tuple<DIE *, const CompileUnit *, PatchLocation>>
222 ForwardDIEReferences;
224 FunctionIntervals::Allocator RangeAlloc;
225 /// \brief The ranges in that interval map are the PC ranges for
226 /// functions in this unit, associated with the PC offset to apply
227 /// to the addresses to get the linked address.
228 FunctionIntervals Ranges;
230 /// \brief DW_AT_ranges attributes to patch after we have gathered
231 /// all the unit's function addresses.
233 std::vector<PatchLocation> RangeAttributes;
234 Optional<PatchLocation> UnitRangeAttribute;
237 /// \brief Location attributes that need to be transfered from th
238 /// original debug_loc section to the liked one. They are stored
239 /// along with the PC offset that is to be applied to their
240 /// function's address.
241 std::vector<std::pair<PatchLocation, int64_t>> LocationAttributes;
243 /// \brief Accelerator entries for the unit, both for the pub*
244 /// sections and the apple* ones.
246 std::vector<AccelInfo> Pubnames;
247 std::vector<AccelInfo> Pubtypes;
251 uint64_t CompileUnit::computeNextUnitOffset() {
252 NextUnitOffset = StartOffset + 11 /* Header size */;
253 // The root DIE might be null, meaning that the Unit had nothing to
254 // contribute to the linked output. In that case, we will emit the
255 // unit header without any actual DIE.
257 NextUnitOffset += CUDie->getSize();
258 return NextUnitOffset;
261 /// \brief Keep track of a forward cross-cu reference from this unit
262 /// to \p Die that lives in \p RefUnit.
263 void CompileUnit::noteForwardReference(DIE *Die, const CompileUnit *RefUnit,
264 PatchLocation Attr) {
265 ForwardDIEReferences.emplace_back(Die, RefUnit, Attr);
268 /// \brief Apply all fixups recorded by noteForwardReference().
269 void CompileUnit::fixupForwardReferences() {
270 for (const auto &Ref : ForwardDIEReferences) {
272 const CompileUnit *RefUnit;
274 std::tie(RefDie, RefUnit, Attr) = Ref;
275 Attr.set(RefDie->getOffset() + RefUnit->getStartOffset());
279 void CompileUnit::addFunctionRange(uint64_t FuncLowPc, uint64_t FuncHighPc,
281 Ranges.insert(FuncLowPc, FuncHighPc, PcOffset);
282 this->LowPc = std::min(LowPc, FuncLowPc + PcOffset);
283 this->HighPc = std::max(HighPc, FuncHighPc + PcOffset);
286 void CompileUnit::noteRangeAttribute(const DIE &Die, PatchLocation Attr) {
287 if (Die.getTag() != dwarf::DW_TAG_compile_unit)
288 RangeAttributes.push_back(Attr);
290 UnitRangeAttribute = Attr;
293 void CompileUnit::noteLocationAttribute(PatchLocation Attr, int64_t PcOffset) {
294 LocationAttributes.emplace_back(Attr, PcOffset);
297 /// \brief Add a name accelerator entry for \p Die with \p Name
298 /// which is stored in the string table at \p Offset.
299 void CompileUnit::addNameAccelerator(const DIE *Die, const char *Name,
300 uint32_t Offset, bool SkipPubSection) {
301 Pubnames.emplace_back(Name, Die, Offset, SkipPubSection);
304 /// \brief Add a type accelerator entry for \p Die with \p Name
305 /// which is stored in the string table at \p Offset.
306 void CompileUnit::addTypeAccelerator(const DIE *Die, const char *Name,
308 Pubtypes.emplace_back(Name, Die, Offset, false);
311 /// \brief A string table that doesn't need relocations.
313 /// We are doing a final link, no need for a string table that
314 /// has relocation entries for every reference to it. This class
315 /// provides this ablitity by just associating offsets with
317 class NonRelocatableStringpool {
319 /// \brief Entries are stored into the StringMap and simply linked
320 /// together through the second element of this pair in order to
321 /// keep track of insertion order.
322 typedef StringMap<std::pair<uint32_t, StringMapEntryBase *>, BumpPtrAllocator>
325 NonRelocatableStringpool()
326 : CurrentEndOffset(0), Sentinel(0), Last(&Sentinel) {
327 // Legacy dsymutil puts an empty string at the start of the line
332 /// \brief Get the offset of string \p S in the string table. This
333 /// can insert a new element or return the offset of a preexisitng
335 uint32_t getStringOffset(StringRef S);
337 /// \brief Get permanent storage for \p S (but do not necessarily
338 /// emit \p S in the output section).
339 /// \returns The StringRef that points to permanent storage to use
340 /// in place of \p S.
341 StringRef internString(StringRef S);
343 // \brief Return the first entry of the string table.
344 const MapTy::MapEntryTy *getFirstEntry() const {
345 return getNextEntry(&Sentinel);
348 // \brief Get the entry following \p E in the string table or null
349 // if \p E was the last entry.
350 const MapTy::MapEntryTy *getNextEntry(const MapTy::MapEntryTy *E) const {
351 return static_cast<const MapTy::MapEntryTy *>(E->getValue().second);
354 uint64_t getSize() { return CurrentEndOffset; }
358 uint32_t CurrentEndOffset;
359 MapTy::MapEntryTy Sentinel, *Last;
362 /// \brief Get the offset of string \p S in the string table. This
363 /// can insert a new element or return the offset of a preexisitng
365 uint32_t NonRelocatableStringpool::getStringOffset(StringRef S) {
366 if (S.empty() && !Strings.empty())
369 std::pair<uint32_t, StringMapEntryBase *> Entry(0, nullptr);
373 // A non-empty string can't be at offset 0, so if we have an entry
374 // with a 0 offset, it must be a previously interned string.
375 std::tie(It, Inserted) = Strings.insert(std::make_pair(S, Entry));
376 if (Inserted || It->getValue().first == 0) {
377 // Set offset and chain at the end of the entries list.
378 It->getValue().first = CurrentEndOffset;
379 CurrentEndOffset += S.size() + 1; // +1 for the '\0'.
380 Last->getValue().second = &*It;
383 return It->getValue().first;
386 /// \brief Put \p S into the StringMap so that it gets permanent
387 /// storage, but do not actually link it in the chain of elements
388 /// that go into the output section. A latter call to
389 /// getStringOffset() with the same string will chain it though.
390 StringRef NonRelocatableStringpool::internString(StringRef S) {
391 std::pair<uint32_t, StringMapEntryBase *> Entry(0, nullptr);
392 auto InsertResult = Strings.insert(std::make_pair(S, Entry));
393 return InsertResult.first->getKey();
396 /// \brief The Dwarf streaming logic
398 /// All interactions with the MC layer that is used to build the debug
399 /// information binary representation are handled in this class.
400 class DwarfStreamer {
401 /// \defgroup MCObjects MC layer objects constructed by the streamer
403 std::unique_ptr<MCRegisterInfo> MRI;
404 std::unique_ptr<MCAsmInfo> MAI;
405 std::unique_ptr<MCObjectFileInfo> MOFI;
406 std::unique_ptr<MCContext> MC;
407 MCAsmBackend *MAB; // Owned by MCStreamer
408 std::unique_ptr<MCInstrInfo> MII;
409 std::unique_ptr<MCSubtargetInfo> MSTI;
410 MCCodeEmitter *MCE; // Owned by MCStreamer
411 MCStreamer *MS; // Owned by AsmPrinter
412 std::unique_ptr<TargetMachine> TM;
413 std::unique_ptr<AsmPrinter> Asm;
416 /// \brief the file we stream the linked Dwarf to.
417 std::unique_ptr<raw_fd_ostream> OutFile;
419 uint32_t RangesSectionSize;
420 uint32_t LocSectionSize;
421 uint32_t LineSectionSize;
422 uint32_t FrameSectionSize;
424 /// \brief Emit the pubnames or pubtypes section contribution for \p
425 /// Unit into \p Sec. The data is provided in \p Names.
426 void emitPubSectionForUnit(MCSection *Sec, StringRef Name,
427 const CompileUnit &Unit,
428 const std::vector<CompileUnit::AccelInfo> &Names);
431 /// \brief Actually create the streamer and the ouptut file.
433 /// This could be done directly in the constructor, but it feels
434 /// more natural to handle errors through return value.
435 bool init(Triple TheTriple, StringRef OutputFilename);
437 /// \brief Dump the file to the disk.
440 AsmPrinter &getAsmPrinter() const { return *Asm; }
442 /// \brief Set the current output section to debug_info and change
443 /// the MC Dwarf version to \p DwarfVersion.
444 void switchToDebugInfoSection(unsigned DwarfVersion);
446 /// \brief Emit the compilation unit header for \p Unit in the
447 /// debug_info section.
449 /// As a side effect, this also switches the current Dwarf version
450 /// of the MC layer to the one of U.getOrigUnit().
451 void emitCompileUnitHeader(CompileUnit &Unit);
453 /// \brief Recursively emit the DIE tree rooted at \p Die.
454 void emitDIE(DIE &Die);
456 /// \brief Emit the abbreviation table \p Abbrevs to the
457 /// debug_abbrev section.
458 void emitAbbrevs(const std::vector<DIEAbbrev *> &Abbrevs);
460 /// \brief Emit the string table described by \p Pool.
461 void emitStrings(const NonRelocatableStringpool &Pool);
463 /// \brief Emit debug_ranges for \p FuncRange by translating the
464 /// original \p Entries.
465 void emitRangesEntries(
466 int64_t UnitPcOffset, uint64_t OrigLowPc,
467 FunctionIntervals::const_iterator FuncRange,
468 const std::vector<DWARFDebugRangeList::RangeListEntry> &Entries,
469 unsigned AddressSize);
471 /// \brief Emit debug_aranges entries for \p Unit and if \p
472 /// DoRangesSection is true, also emit the debug_ranges entries for
473 /// the DW_TAG_compile_unit's DW_AT_ranges attribute.
474 void emitUnitRangesEntries(CompileUnit &Unit, bool DoRangesSection);
476 uint32_t getRangesSectionSize() const { return RangesSectionSize; }
478 /// \brief Emit the debug_loc contribution for \p Unit by copying
479 /// the entries from \p Dwarf and offseting them. Update the
480 /// location attributes to point to the new entries.
481 void emitLocationsForUnit(const CompileUnit &Unit, DWARFContext &Dwarf);
483 /// \brief Emit the line table described in \p Rows into the
484 /// debug_line section.
485 void emitLineTableForUnit(StringRef PrologueBytes, unsigned MinInstLength,
486 std::vector<DWARFDebugLine::Row> &Rows,
487 unsigned AdddressSize);
489 uint32_t getLineSectionSize() const { return LineSectionSize; }
491 /// \brief Emit the .debug_pubnames contribution for \p Unit.
492 void emitPubNamesForUnit(const CompileUnit &Unit);
494 /// \brief Emit the .debug_pubtypes contribution for \p Unit.
495 void emitPubTypesForUnit(const CompileUnit &Unit);
497 /// \brief Emit a CIE.
498 void emitCIE(StringRef CIEBytes);
500 /// \brief Emit an FDE with data \p Bytes.
501 void emitFDE(uint32_t CIEOffset, uint32_t AddreSize, uint32_t Address,
504 uint32_t getFrameSectionSize() const { return FrameSectionSize; }
507 bool DwarfStreamer::init(Triple TheTriple, StringRef OutputFilename) {
508 std::string ErrorStr;
509 std::string TripleName;
510 StringRef Context = "dwarf streamer init";
513 const Target *TheTarget =
514 TargetRegistry::lookupTarget(TripleName, TheTriple, ErrorStr);
516 return error(ErrorStr, Context);
517 TripleName = TheTriple.getTriple();
519 // Create all the MC Objects.
520 MRI.reset(TheTarget->createMCRegInfo(TripleName));
522 return error(Twine("no register info for target ") + TripleName, Context);
524 MAI.reset(TheTarget->createMCAsmInfo(*MRI, TripleName));
526 return error("no asm info for target " + TripleName, Context);
528 MOFI.reset(new MCObjectFileInfo);
529 MC.reset(new MCContext(MAI.get(), MRI.get(), MOFI.get()));
530 MOFI->InitMCObjectFileInfo(TheTriple, Reloc::Default, CodeModel::Default,
533 MAB = TheTarget->createMCAsmBackend(*MRI, TripleName, "");
535 return error("no asm backend for target " + TripleName, Context);
537 MII.reset(TheTarget->createMCInstrInfo());
539 return error("no instr info info for target " + TripleName, Context);
541 MSTI.reset(TheTarget->createMCSubtargetInfo(TripleName, "", ""));
543 return error("no subtarget info for target " + TripleName, Context);
545 MCE = TheTarget->createMCCodeEmitter(*MII, *MRI, *MC);
547 return error("no code emitter for target " + TripleName, Context);
549 // Create the output file.
552 llvm::make_unique<raw_fd_ostream>(OutputFilename, EC, sys::fs::F_None);
554 return error(Twine(OutputFilename) + ": " + EC.message(), Context);
556 MS = TheTarget->createMCObjectStreamer(TheTriple, *MC, *MAB, *OutFile, MCE,
558 /*DWARFMustBeAtTheEnd*/ false);
560 return error("no object streamer for target " + TripleName, Context);
562 // Finally create the AsmPrinter we'll use to emit the DIEs.
563 TM.reset(TheTarget->createTargetMachine(TripleName, "", "", TargetOptions()));
565 return error("no target machine for target " + TripleName, Context);
567 Asm.reset(TheTarget->createAsmPrinter(*TM, std::unique_ptr<MCStreamer>(MS)));
569 return error("no asm printer for target " + TripleName, Context);
571 RangesSectionSize = 0;
574 FrameSectionSize = 0;
579 bool DwarfStreamer::finish() {
584 /// \brief Set the current output section to debug_info and change
585 /// the MC Dwarf version to \p DwarfVersion.
586 void DwarfStreamer::switchToDebugInfoSection(unsigned DwarfVersion) {
587 MS->SwitchSection(MOFI->getDwarfInfoSection());
588 MC->setDwarfVersion(DwarfVersion);
591 /// \brief Emit the compilation unit header for \p Unit in the
592 /// debug_info section.
594 /// A Dwarf scetion header is encoded as:
595 /// uint32_t Unit length (omiting this field)
597 /// uint32_t Abbreviation table offset
598 /// uint8_t Address size
600 /// Leading to a total of 11 bytes.
601 void DwarfStreamer::emitCompileUnitHeader(CompileUnit &Unit) {
602 unsigned Version = Unit.getOrigUnit().getVersion();
603 switchToDebugInfoSection(Version);
605 // Emit size of content not including length itself. The size has
606 // already been computed in CompileUnit::computeOffsets(). Substract
607 // 4 to that size to account for the length field.
608 Asm->EmitInt32(Unit.getNextUnitOffset() - Unit.getStartOffset() - 4);
609 Asm->EmitInt16(Version);
610 // We share one abbreviations table across all units so it's always at the
611 // start of the section.
613 Asm->EmitInt8(Unit.getOrigUnit().getAddressByteSize());
616 /// \brief Emit the \p Abbrevs array as the shared abbreviation table
617 /// for the linked Dwarf file.
618 void DwarfStreamer::emitAbbrevs(const std::vector<DIEAbbrev *> &Abbrevs) {
619 MS->SwitchSection(MOFI->getDwarfAbbrevSection());
620 Asm->emitDwarfAbbrevs(Abbrevs);
623 /// \brief Recursively emit the DIE tree rooted at \p Die.
624 void DwarfStreamer::emitDIE(DIE &Die) {
625 MS->SwitchSection(MOFI->getDwarfInfoSection());
626 Asm->emitDwarfDIE(Die);
629 /// \brief Emit the debug_str section stored in \p Pool.
630 void DwarfStreamer::emitStrings(const NonRelocatableStringpool &Pool) {
631 Asm->OutStreamer->SwitchSection(MOFI->getDwarfStrSection());
632 for (auto *Entry = Pool.getFirstEntry(); Entry;
633 Entry = Pool.getNextEntry(Entry))
634 Asm->OutStreamer->EmitBytes(
635 StringRef(Entry->getKey().data(), Entry->getKey().size() + 1));
638 /// \brief Emit the debug_range section contents for \p FuncRange by
639 /// translating the original \p Entries. The debug_range section
640 /// format is totally trivial, consisting just of pairs of address
641 /// sized addresses describing the ranges.
642 void DwarfStreamer::emitRangesEntries(
643 int64_t UnitPcOffset, uint64_t OrigLowPc,
644 FunctionIntervals::const_iterator FuncRange,
645 const std::vector<DWARFDebugRangeList::RangeListEntry> &Entries,
646 unsigned AddressSize) {
647 MS->SwitchSection(MC->getObjectFileInfo()->getDwarfRangesSection());
649 // Offset each range by the right amount.
650 int64_t PcOffset = FuncRange.value() + UnitPcOffset;
651 for (const auto &Range : Entries) {
652 if (Range.isBaseAddressSelectionEntry(AddressSize)) {
653 warn("unsupported base address selection operation",
654 "emitting debug_ranges");
657 // Do not emit empty ranges.
658 if (Range.StartAddress == Range.EndAddress)
661 // All range entries should lie in the function range.
662 if (!(Range.StartAddress + OrigLowPc >= FuncRange.start() &&
663 Range.EndAddress + OrigLowPc <= FuncRange.stop()))
664 warn("inconsistent range data.", "emitting debug_ranges");
665 MS->EmitIntValue(Range.StartAddress + PcOffset, AddressSize);
666 MS->EmitIntValue(Range.EndAddress + PcOffset, AddressSize);
667 RangesSectionSize += 2 * AddressSize;
670 // Add the terminator entry.
671 MS->EmitIntValue(0, AddressSize);
672 MS->EmitIntValue(0, AddressSize);
673 RangesSectionSize += 2 * AddressSize;
676 /// \brief Emit the debug_aranges contribution of a unit and
677 /// if \p DoDebugRanges is true the debug_range contents for a
678 /// compile_unit level DW_AT_ranges attribute (Which are basically the
679 /// same thing with a different base address).
680 /// Just aggregate all the ranges gathered inside that unit.
681 void DwarfStreamer::emitUnitRangesEntries(CompileUnit &Unit,
682 bool DoDebugRanges) {
683 unsigned AddressSize = Unit.getOrigUnit().getAddressByteSize();
684 // Gather the ranges in a vector, so that we can simplify them. The
685 // IntervalMap will have coalesced the non-linked ranges, but here
686 // we want to coalesce the linked addresses.
687 std::vector<std::pair<uint64_t, uint64_t>> Ranges;
688 const auto &FunctionRanges = Unit.getFunctionRanges();
689 for (auto Range = FunctionRanges.begin(), End = FunctionRanges.end();
690 Range != End; ++Range)
691 Ranges.push_back(std::make_pair(Range.start() + Range.value(),
692 Range.stop() + Range.value()));
694 // The object addresses where sorted, but again, the linked
695 // addresses might end up in a different order.
696 std::sort(Ranges.begin(), Ranges.end());
698 if (!Ranges.empty()) {
699 MS->SwitchSection(MC->getObjectFileInfo()->getDwarfARangesSection());
701 MCSymbol *BeginLabel = Asm->createTempSymbol("Barange");
702 MCSymbol *EndLabel = Asm->createTempSymbol("Earange");
704 unsigned HeaderSize =
705 sizeof(int32_t) + // Size of contents (w/o this field
706 sizeof(int16_t) + // DWARF ARange version number
707 sizeof(int32_t) + // Offset of CU in the .debug_info section
708 sizeof(int8_t) + // Pointer Size (in bytes)
709 sizeof(int8_t); // Segment Size (in bytes)
711 unsigned TupleSize = AddressSize * 2;
712 unsigned Padding = OffsetToAlignment(HeaderSize, TupleSize);
714 Asm->EmitLabelDifference(EndLabel, BeginLabel, 4); // Arange length
715 Asm->OutStreamer->EmitLabel(BeginLabel);
716 Asm->EmitInt16(dwarf::DW_ARANGES_VERSION); // Version number
717 Asm->EmitInt32(Unit.getStartOffset()); // Corresponding unit's offset
718 Asm->EmitInt8(AddressSize); // Address size
719 Asm->EmitInt8(0); // Segment size
721 Asm->OutStreamer->EmitFill(Padding, 0x0);
723 for (auto Range = Ranges.begin(), End = Ranges.end(); Range != End;
725 uint64_t RangeStart = Range->first;
726 MS->EmitIntValue(RangeStart, AddressSize);
727 while ((Range + 1) != End && Range->second == (Range + 1)->first)
729 MS->EmitIntValue(Range->second - RangeStart, AddressSize);
733 Asm->OutStreamer->EmitIntValue(0, AddressSize);
734 Asm->OutStreamer->EmitIntValue(0, AddressSize);
735 Asm->OutStreamer->EmitLabel(EndLabel);
741 MS->SwitchSection(MC->getObjectFileInfo()->getDwarfRangesSection());
742 // Offset each range by the right amount.
743 int64_t PcOffset = -Unit.getLowPc();
744 // Emit coalesced ranges.
745 for (auto Range = Ranges.begin(), End = Ranges.end(); Range != End; ++Range) {
746 MS->EmitIntValue(Range->first + PcOffset, AddressSize);
747 while (Range + 1 != End && Range->second == (Range + 1)->first)
749 MS->EmitIntValue(Range->second + PcOffset, AddressSize);
750 RangesSectionSize += 2 * AddressSize;
753 // Add the terminator entry.
754 MS->EmitIntValue(0, AddressSize);
755 MS->EmitIntValue(0, AddressSize);
756 RangesSectionSize += 2 * AddressSize;
759 /// \brief Emit location lists for \p Unit and update attribtues to
760 /// point to the new entries.
761 void DwarfStreamer::emitLocationsForUnit(const CompileUnit &Unit,
762 DWARFContext &Dwarf) {
763 const auto &Attributes = Unit.getLocationAttributes();
765 if (Attributes.empty())
768 MS->SwitchSection(MC->getObjectFileInfo()->getDwarfLocSection());
770 unsigned AddressSize = Unit.getOrigUnit().getAddressByteSize();
771 const DWARFSection &InputSec = Dwarf.getLocSection();
772 DataExtractor Data(InputSec.Data, Dwarf.isLittleEndian(), AddressSize);
773 DWARFUnit &OrigUnit = Unit.getOrigUnit();
774 const auto *OrigUnitDie = OrigUnit.getUnitDIE(false);
775 int64_t UnitPcOffset = 0;
776 uint64_t OrigLowPc = OrigUnitDie->getAttributeValueAsAddress(
777 &OrigUnit, dwarf::DW_AT_low_pc, -1ULL);
778 if (OrigLowPc != -1ULL)
779 UnitPcOffset = int64_t(OrigLowPc) - Unit.getLowPc();
781 for (const auto &Attr : Attributes) {
782 uint32_t Offset = Attr.first.get();
783 Attr.first.set(LocSectionSize);
784 // This is the quantity to add to the old location address to get
785 // the correct address for the new one.
786 int64_t LocPcOffset = Attr.second + UnitPcOffset;
787 while (Data.isValidOffset(Offset)) {
788 uint64_t Low = Data.getUnsigned(&Offset, AddressSize);
789 uint64_t High = Data.getUnsigned(&Offset, AddressSize);
790 LocSectionSize += 2 * AddressSize;
791 if (Low == 0 && High == 0) {
792 Asm->OutStreamer->EmitIntValue(0, AddressSize);
793 Asm->OutStreamer->EmitIntValue(0, AddressSize);
796 Asm->OutStreamer->EmitIntValue(Low + LocPcOffset, AddressSize);
797 Asm->OutStreamer->EmitIntValue(High + LocPcOffset, AddressSize);
798 uint64_t Length = Data.getU16(&Offset);
799 Asm->OutStreamer->EmitIntValue(Length, 2);
800 // Just copy the bytes over.
801 Asm->OutStreamer->EmitBytes(
802 StringRef(InputSec.Data.substr(Offset, Length)));
804 LocSectionSize += Length + 2;
809 void DwarfStreamer::emitLineTableForUnit(StringRef PrologueBytes,
810 unsigned MinInstLength,
811 std::vector<DWARFDebugLine::Row> &Rows,
812 unsigned PointerSize) {
813 // Switch to the section where the table will be emitted into.
814 MS->SwitchSection(MC->getObjectFileInfo()->getDwarfLineSection());
815 MCSymbol *LineStartSym = MC->createTempSymbol();
816 MCSymbol *LineEndSym = MC->createTempSymbol();
818 // The first 4 bytes is the total length of the information for this
819 // compilation unit (not including these 4 bytes for the length).
820 Asm->EmitLabelDifference(LineEndSym, LineStartSym, 4);
821 Asm->OutStreamer->EmitLabel(LineStartSym);
823 MS->EmitBytes(PrologueBytes);
824 LineSectionSize += PrologueBytes.size() + 4;
826 SmallString<128> EncodingBuffer;
827 raw_svector_ostream EncodingOS(EncodingBuffer);
830 // We only have the dummy entry, dsymutil emits an entry with a 0
831 // address in that case.
832 MCDwarfLineAddr::Encode(*MC, INT64_MAX, 0, EncodingOS);
833 MS->EmitBytes(EncodingOS.str());
834 LineSectionSize += EncodingBuffer.size();
835 MS->EmitLabel(LineEndSym);
839 // Line table state machine fields
840 unsigned FileNum = 1;
841 unsigned LastLine = 1;
843 unsigned IsStatement = 1;
845 uint64_t Address = -1ULL;
847 unsigned RowsSinceLastSequence = 0;
849 for (unsigned Idx = 0; Idx < Rows.size(); ++Idx) {
850 auto &Row = Rows[Idx];
852 int64_t AddressDelta;
853 if (Address == -1ULL) {
854 MS->EmitIntValue(dwarf::DW_LNS_extended_op, 1);
855 MS->EmitULEB128IntValue(PointerSize + 1);
856 MS->EmitIntValue(dwarf::DW_LNE_set_address, 1);
857 MS->EmitIntValue(Row.Address, PointerSize);
858 LineSectionSize += 2 + PointerSize + getULEB128Size(PointerSize + 1);
861 AddressDelta = (Row.Address - Address) / MinInstLength;
864 // FIXME: code copied and transfromed from
865 // MCDwarf.cpp::EmitDwarfLineTable. We should find a way to share
866 // this code, but the current compatibility requirement with
867 // classic dsymutil makes it hard. Revisit that once this
868 // requirement is dropped.
870 if (FileNum != Row.File) {
872 MS->EmitIntValue(dwarf::DW_LNS_set_file, 1);
873 MS->EmitULEB128IntValue(FileNum);
874 LineSectionSize += 1 + getULEB128Size(FileNum);
876 if (Column != Row.Column) {
878 MS->EmitIntValue(dwarf::DW_LNS_set_column, 1);
879 MS->EmitULEB128IntValue(Column);
880 LineSectionSize += 1 + getULEB128Size(Column);
883 // FIXME: We should handle the discriminator here, but dsymutil
884 // doesn' consider it, thus ignore it for now.
886 if (Isa != Row.Isa) {
888 MS->EmitIntValue(dwarf::DW_LNS_set_isa, 1);
889 MS->EmitULEB128IntValue(Isa);
890 LineSectionSize += 1 + getULEB128Size(Isa);
892 if (IsStatement != Row.IsStmt) {
893 IsStatement = Row.IsStmt;
894 MS->EmitIntValue(dwarf::DW_LNS_negate_stmt, 1);
895 LineSectionSize += 1;
897 if (Row.BasicBlock) {
898 MS->EmitIntValue(dwarf::DW_LNS_set_basic_block, 1);
899 LineSectionSize += 1;
902 if (Row.PrologueEnd) {
903 MS->EmitIntValue(dwarf::DW_LNS_set_prologue_end, 1);
904 LineSectionSize += 1;
907 if (Row.EpilogueBegin) {
908 MS->EmitIntValue(dwarf::DW_LNS_set_epilogue_begin, 1);
909 LineSectionSize += 1;
912 int64_t LineDelta = int64_t(Row.Line) - LastLine;
913 if (!Row.EndSequence) {
914 MCDwarfLineAddr::Encode(*MC, LineDelta, AddressDelta, EncodingOS);
915 MS->EmitBytes(EncodingOS.str());
916 LineSectionSize += EncodingBuffer.size();
917 EncodingBuffer.resize(0);
919 Address = Row.Address;
921 RowsSinceLastSequence++;
924 MS->EmitIntValue(dwarf::DW_LNS_advance_line, 1);
925 MS->EmitSLEB128IntValue(LineDelta);
926 LineSectionSize += 1 + getSLEB128Size(LineDelta);
929 MS->EmitIntValue(dwarf::DW_LNS_advance_pc, 1);
930 MS->EmitULEB128IntValue(AddressDelta);
931 LineSectionSize += 1 + getULEB128Size(AddressDelta);
933 MCDwarfLineAddr::Encode(*MC, INT64_MAX, 0, EncodingOS);
934 MS->EmitBytes(EncodingOS.str());
935 LineSectionSize += EncodingBuffer.size();
936 EncodingBuffer.resize(0);
939 LastLine = FileNum = IsStatement = 1;
940 RowsSinceLastSequence = Column = Isa = 0;
944 if (RowsSinceLastSequence) {
945 MCDwarfLineAddr::Encode(*MC, INT64_MAX, 0, EncodingOS);
946 MS->EmitBytes(EncodingOS.str());
947 LineSectionSize += EncodingBuffer.size();
948 EncodingBuffer.resize(0);
952 MS->EmitLabel(LineEndSym);
955 /// \brief Emit the pubnames or pubtypes section contribution for \p
956 /// Unit into \p Sec. The data is provided in \p Names.
957 void DwarfStreamer::emitPubSectionForUnit(
958 MCSection *Sec, StringRef SecName, const CompileUnit &Unit,
959 const std::vector<CompileUnit::AccelInfo> &Names) {
963 // Start the dwarf pubnames section.
964 Asm->OutStreamer->SwitchSection(Sec);
965 MCSymbol *BeginLabel = Asm->createTempSymbol("pub" + SecName + "_begin");
966 MCSymbol *EndLabel = Asm->createTempSymbol("pub" + SecName + "_end");
968 bool HeaderEmitted = false;
969 // Emit the pubnames for this compilation unit.
970 for (const auto &Name : Names) {
971 if (Name.SkipPubSection)
974 if (!HeaderEmitted) {
976 Asm->EmitLabelDifference(EndLabel, BeginLabel, 4); // Length
977 Asm->OutStreamer->EmitLabel(BeginLabel);
978 Asm->EmitInt16(dwarf::DW_PUBNAMES_VERSION); // Version
979 Asm->EmitInt32(Unit.getStartOffset()); // Unit offset
980 Asm->EmitInt32(Unit.getNextUnitOffset() - Unit.getStartOffset()); // Size
981 HeaderEmitted = true;
983 Asm->EmitInt32(Name.Die->getOffset());
984 Asm->OutStreamer->EmitBytes(
985 StringRef(Name.Name.data(), Name.Name.size() + 1));
990 Asm->EmitInt32(0); // End marker.
991 Asm->OutStreamer->EmitLabel(EndLabel);
994 /// \brief Emit .debug_pubnames for \p Unit.
995 void DwarfStreamer::emitPubNamesForUnit(const CompileUnit &Unit) {
996 emitPubSectionForUnit(MC->getObjectFileInfo()->getDwarfPubNamesSection(),
997 "names", Unit, Unit.getPubnames());
1000 /// \brief Emit .debug_pubtypes for \p Unit.
1001 void DwarfStreamer::emitPubTypesForUnit(const CompileUnit &Unit) {
1002 emitPubSectionForUnit(MC->getObjectFileInfo()->getDwarfPubTypesSection(),
1003 "types", Unit, Unit.getPubtypes());
1006 /// \brief Emit a CIE into the debug_frame section.
1007 void DwarfStreamer::emitCIE(StringRef CIEBytes) {
1008 MS->SwitchSection(MC->getObjectFileInfo()->getDwarfFrameSection());
1010 MS->EmitBytes(CIEBytes);
1011 FrameSectionSize += CIEBytes.size();
1014 /// \brief Emit a FDE into the debug_frame section. \p FDEBytes
1015 /// contains the FDE data without the length, CIE offset and address
1016 /// which will be replaced with the paramter values.
1017 void DwarfStreamer::emitFDE(uint32_t CIEOffset, uint32_t AddrSize,
1018 uint32_t Address, StringRef FDEBytes) {
1019 MS->SwitchSection(MC->getObjectFileInfo()->getDwarfFrameSection());
1021 MS->EmitIntValue(FDEBytes.size() + 4 + AddrSize, 4);
1022 MS->EmitIntValue(CIEOffset, 4);
1023 MS->EmitIntValue(Address, AddrSize);
1024 MS->EmitBytes(FDEBytes);
1025 FrameSectionSize += FDEBytes.size() + 8 + AddrSize;
1028 /// \brief The core of the Dwarf linking logic.
1030 /// The link of the dwarf information from the object files will be
1031 /// driven by the selection of 'root DIEs', which are DIEs that
1032 /// describe variables or functions that are present in the linked
1033 /// binary (and thus have entries in the debug map). All the debug
1034 /// information that will be linked (the DIEs, but also the line
1035 /// tables, ranges, ...) is derived from that set of root DIEs.
1037 /// The root DIEs are identified because they contain relocations that
1038 /// correspond to a debug map entry at specific places (the low_pc for
1039 /// a function, the location for a variable). These relocations are
1040 /// called ValidRelocs in the DwarfLinker and are gathered as a very
1041 /// first step when we start processing a DebugMapObject.
1044 DwarfLinker(StringRef OutputFilename, const LinkOptions &Options)
1045 : OutputFilename(OutputFilename), Options(Options),
1046 BinHolder(Options.Verbose), LastCIEOffset(0) {}
1049 for (auto *Abbrev : Abbreviations)
1053 /// \brief Link the contents of the DebugMap.
1054 bool link(const DebugMap &);
1057 /// \brief Called at the start of a debug object link.
1058 void startDebugObject(DWARFContext &, DebugMapObject &);
1060 /// \brief Called at the end of a debug object link.
1061 void endDebugObject();
1063 /// \defgroup FindValidRelocations Translate debug map into a list
1064 /// of relevant relocations
1071 const DebugMapObject::DebugMapEntry *Mapping;
1073 ValidReloc(uint32_t Offset, uint32_t Size, uint64_t Addend,
1074 const DebugMapObject::DebugMapEntry *Mapping)
1075 : Offset(Offset), Size(Size), Addend(Addend), Mapping(Mapping) {}
1077 bool operator<(const ValidReloc &RHS) const { return Offset < RHS.Offset; }
1080 /// \brief The valid relocations for the current DebugMapObject.
1081 /// This vector is sorted by relocation offset.
1082 std::vector<ValidReloc> ValidRelocs;
1084 /// \brief Index into ValidRelocs of the next relocation to
1085 /// consider. As we walk the DIEs in acsending file offset and as
1086 /// ValidRelocs is sorted by file offset, keeping this index
1087 /// uptodate is all we have to do to have a cheap lookup during the
1088 /// root DIE selection and during DIE cloning.
1089 unsigned NextValidReloc;
1091 bool findValidRelocsInDebugInfo(const object::ObjectFile &Obj,
1092 const DebugMapObject &DMO);
1094 bool findValidRelocs(const object::SectionRef &Section,
1095 const object::ObjectFile &Obj,
1096 const DebugMapObject &DMO);
1098 void findValidRelocsMachO(const object::SectionRef &Section,
1099 const object::MachOObjectFile &Obj,
1100 const DebugMapObject &DMO);
1103 /// \defgroup FindRootDIEs Find DIEs corresponding to debug map entries.
1106 /// \brief Recursively walk the \p DIE tree and look for DIEs to
1107 /// keep. Store that information in \p CU's DIEInfo.
1108 void lookForDIEsToKeep(const DWARFDebugInfoEntryMinimal &DIE,
1109 const DebugMapObject &DMO, CompileUnit &CU,
1112 /// \brief Flags passed to DwarfLinker::lookForDIEsToKeep
1113 enum TravesalFlags {
1114 TF_Keep = 1 << 0, ///< Mark the traversed DIEs as kept.
1115 TF_InFunctionScope = 1 << 1, ///< Current scope is a fucntion scope.
1116 TF_DependencyWalk = 1 << 2, ///< Walking the dependencies of a kept DIE.
1117 TF_ParentWalk = 1 << 3, ///< Walking up the parents of a kept DIE.
1120 /// \brief Mark the passed DIE as well as all the ones it depends on
1122 void keepDIEAndDenpendencies(const DWARFDebugInfoEntryMinimal &DIE,
1123 CompileUnit::DIEInfo &MyInfo,
1124 const DebugMapObject &DMO, CompileUnit &CU,
1127 unsigned shouldKeepDIE(const DWARFDebugInfoEntryMinimal &DIE,
1128 CompileUnit &Unit, CompileUnit::DIEInfo &MyInfo,
1131 unsigned shouldKeepVariableDIE(const DWARFDebugInfoEntryMinimal &DIE,
1133 CompileUnit::DIEInfo &MyInfo, unsigned Flags);
1135 unsigned shouldKeepSubprogramDIE(const DWARFDebugInfoEntryMinimal &DIE,
1137 CompileUnit::DIEInfo &MyInfo,
1140 bool hasValidRelocation(uint32_t StartOffset, uint32_t EndOffset,
1141 CompileUnit::DIEInfo &Info);
1144 /// \defgroup Linking Methods used to link the debug information
1147 /// \brief Recursively clone \p InputDIE into an tree of DIE objects
1148 /// where useless (as decided by lookForDIEsToKeep()) bits have been
1149 /// stripped out and addresses have been rewritten according to the
1152 /// \param OutOffset is the offset the cloned DIE in the output
1154 /// \param PCOffset (while cloning a function scope) is the offset
1155 /// applied to the entry point of the function to get the linked address.
1157 /// \returns the root of the cloned tree.
1158 DIE *cloneDIE(const DWARFDebugInfoEntryMinimal &InputDIE, CompileUnit &U,
1159 int64_t PCOffset, uint32_t OutOffset);
1161 typedef DWARFAbbreviationDeclaration::AttributeSpec AttributeSpec;
1163 /// \brief Information gathered and exchanged between the various
1164 /// clone*Attributes helpers about the attributes of a particular DIE.
1165 struct AttributesInfo {
1166 const char *Name, *MangledName; ///< Names.
1167 uint32_t NameOffset, MangledNameOffset; ///< Offsets in the string pool.
1169 uint64_t OrigHighPc; ///< Value of AT_high_pc in the input DIE
1170 int64_t PCOffset; ///< Offset to apply to PC addresses inside a function.
1172 bool HasLowPc; ///< Does the DIE have a low_pc attribute?
1173 bool IsDeclaration; ///< Is this DIE only a declaration?
1176 : Name(nullptr), MangledName(nullptr), NameOffset(0),
1177 MangledNameOffset(0), OrigHighPc(0), PCOffset(0), HasLowPc(false),
1178 IsDeclaration(false) {}
1181 /// \brief Helper for cloneDIE.
1182 unsigned cloneAttribute(DIE &Die, const DWARFDebugInfoEntryMinimal &InputDIE,
1183 CompileUnit &U, const DWARFFormValue &Val,
1184 const AttributeSpec AttrSpec, unsigned AttrSize,
1185 AttributesInfo &AttrInfo);
1187 /// \brief Helper for cloneDIE.
1188 unsigned cloneStringAttribute(DIE &Die, AttributeSpec AttrSpec,
1189 const DWARFFormValue &Val, const DWARFUnit &U);
1191 /// \brief Helper for cloneDIE.
1193 cloneDieReferenceAttribute(DIE &Die,
1194 const DWARFDebugInfoEntryMinimal &InputDIE,
1195 AttributeSpec AttrSpec, unsigned AttrSize,
1196 const DWARFFormValue &Val, CompileUnit &Unit);
1198 /// \brief Helper for cloneDIE.
1199 unsigned cloneBlockAttribute(DIE &Die, AttributeSpec AttrSpec,
1200 const DWARFFormValue &Val, unsigned AttrSize);
1202 /// \brief Helper for cloneDIE.
1203 unsigned cloneAddressAttribute(DIE &Die, AttributeSpec AttrSpec,
1204 const DWARFFormValue &Val,
1205 const CompileUnit &Unit, AttributesInfo &Info);
1207 /// \brief Helper for cloneDIE.
1208 unsigned cloneScalarAttribute(DIE &Die,
1209 const DWARFDebugInfoEntryMinimal &InputDIE,
1210 CompileUnit &U, AttributeSpec AttrSpec,
1211 const DWARFFormValue &Val, unsigned AttrSize,
1212 AttributesInfo &Info);
1214 /// \brief Helper for cloneDIE.
1215 bool applyValidRelocs(MutableArrayRef<char> Data, uint32_t BaseOffset,
1216 bool isLittleEndian);
1218 /// \brief Assign an abbreviation number to \p Abbrev
1219 void AssignAbbrev(DIEAbbrev &Abbrev);
1221 /// \brief FoldingSet that uniques the abbreviations.
1222 FoldingSet<DIEAbbrev> AbbreviationsSet;
1223 /// \brief Storage for the unique Abbreviations.
1224 /// This is passed to AsmPrinter::emitDwarfAbbrevs(), thus it cannot
1225 /// be changed to a vecot of unique_ptrs.
1226 std::vector<DIEAbbrev *> Abbreviations;
1228 /// \brief Compute and emit debug_ranges section for \p Unit, and
1229 /// patch the attributes referencing it.
1230 void patchRangesForUnit(const CompileUnit &Unit, DWARFContext &Dwarf) const;
1232 /// \brief Generate and emit the DW_AT_ranges attribute for a
1233 /// compile_unit if it had one.
1234 void generateUnitRanges(CompileUnit &Unit) const;
1236 /// \brief Extract the line tables fromt he original dwarf, extract
1237 /// the relevant parts according to the linked function ranges and
1238 /// emit the result in the debug_line section.
1239 void patchLineTableForUnit(CompileUnit &Unit, DWARFContext &OrigDwarf);
1241 /// \brief Emit the accelerator entries for \p Unit.
1242 void emitAcceleratorEntriesForUnit(CompileUnit &Unit);
1244 /// \brief Patch the frame info for an object file and emit it.
1245 void patchFrameInfoForObject(const DebugMapObject &, DWARFContext &,
1246 unsigned AddressSize);
1248 /// \brief DIELoc objects that need to be destructed (but not freed!).
1249 std::vector<DIELoc *> DIELocs;
1250 /// \brief DIEBlock objects that need to be destructed (but not freed!).
1251 std::vector<DIEBlock *> DIEBlocks;
1252 /// \brief Allocator used for all the DIEValue objects.
1253 BumpPtrAllocator DIEAlloc;
1256 /// \defgroup Helpers Various helper methods.
1259 const DWARFDebugInfoEntryMinimal *
1260 resolveDIEReference(DWARFFormValue &RefValue, const DWARFUnit &Unit,
1261 const DWARFDebugInfoEntryMinimal &DIE,
1262 CompileUnit *&ReferencedCU);
1264 CompileUnit *getUnitForOffset(unsigned Offset);
1266 bool getDIENames(const DWARFDebugInfoEntryMinimal &Die, DWARFUnit &U,
1267 AttributesInfo &Info);
1269 void reportWarning(const Twine &Warning, const DWARFUnit *Unit = nullptr,
1270 const DWARFDebugInfoEntryMinimal *DIE = nullptr) const;
1272 bool createStreamer(Triple TheTriple, StringRef OutputFilename);
1276 std::string OutputFilename;
1277 LinkOptions Options;
1278 BinaryHolder BinHolder;
1279 std::unique_ptr<DwarfStreamer> Streamer;
1281 /// The units of the current debug map object.
1282 std::vector<CompileUnit> Units;
1284 /// The debug map object curently under consideration.
1285 DebugMapObject *CurrentDebugObject;
1287 /// \brief The Dwarf string pool
1288 NonRelocatableStringpool StringPool;
1290 /// \brief This map is keyed by the entry PC of functions in that
1291 /// debug object and the associated value is a pair storing the
1292 /// corresponding end PC and the offset to apply to get the linked
1295 /// See startDebugObject() for a more complete description of its use.
1296 std::map<uint64_t, std::pair<uint64_t, int64_t>> Ranges;
1298 /// \brief The CIEs that have been emitted in the output
1299 /// section. The actual CIE data serves a the key to this StringMap,
1300 /// this takes care of comparing the semantics of CIEs defined in
1301 /// different object files.
1302 StringMap<uint32_t> EmittedCIEs;
1304 /// Offset of the last CIE that has been emitted in the output
1305 /// debug_frame section.
1306 uint32_t LastCIEOffset;
1309 /// \brief Similar to DWARFUnitSection::getUnitForOffset(), but
1310 /// returning our CompileUnit object instead.
1311 CompileUnit *DwarfLinker::getUnitForOffset(unsigned Offset) {
1313 std::upper_bound(Units.begin(), Units.end(), Offset,
1314 [](uint32_t LHS, const CompileUnit &RHS) {
1315 return LHS < RHS.getOrigUnit().getNextUnitOffset();
1317 return CU != Units.end() ? &*CU : nullptr;
1320 /// \brief Resolve the DIE attribute reference that has been
1321 /// extracted in \p RefValue. The resulting DIE migh be in another
1322 /// CompileUnit which is stored into \p ReferencedCU.
1323 /// \returns null if resolving fails for any reason.
1324 const DWARFDebugInfoEntryMinimal *DwarfLinker::resolveDIEReference(
1325 DWARFFormValue &RefValue, const DWARFUnit &Unit,
1326 const DWARFDebugInfoEntryMinimal &DIE, CompileUnit *&RefCU) {
1327 assert(RefValue.isFormClass(DWARFFormValue::FC_Reference));
1328 uint64_t RefOffset = *RefValue.getAsReference(&Unit);
1330 if ((RefCU = getUnitForOffset(RefOffset)))
1331 if (const auto *RefDie = RefCU->getOrigUnit().getDIEForOffset(RefOffset))
1334 reportWarning("could not find referenced DIE", &Unit, &DIE);
1338 /// \brief Get the potential name and mangled name for the entity
1339 /// described by \p Die and store them in \Info if they are not
1341 /// \returns is a name was found.
1342 bool DwarfLinker::getDIENames(const DWARFDebugInfoEntryMinimal &Die,
1343 DWARFUnit &U, AttributesInfo &Info) {
1344 // FIXME: a bit wastefull as the first getName might return the
1346 if (!Info.MangledName &&
1347 (Info.MangledName = Die.getName(&U, DINameKind::LinkageName)))
1348 Info.MangledNameOffset = StringPool.getStringOffset(Info.MangledName);
1350 if (!Info.Name && (Info.Name = Die.getName(&U, DINameKind::ShortName)))
1351 Info.NameOffset = StringPool.getStringOffset(Info.Name);
1353 return Info.Name || Info.MangledName;
1356 /// \brief Report a warning to the user, optionaly including
1357 /// information about a specific \p DIE related to the warning.
1358 void DwarfLinker::reportWarning(const Twine &Warning, const DWARFUnit *Unit,
1359 const DWARFDebugInfoEntryMinimal *DIE) const {
1360 StringRef Context = "<debug map>";
1361 if (CurrentDebugObject)
1362 Context = CurrentDebugObject->getObjectFilename();
1363 warn(Warning, Context);
1365 if (!Options.Verbose || !DIE)
1368 errs() << " in DIE:\n";
1369 DIE->dump(errs(), const_cast<DWARFUnit *>(Unit), 0 /* RecurseDepth */,
1373 bool DwarfLinker::createStreamer(Triple TheTriple, StringRef OutputFilename) {
1374 if (Options.NoOutput)
1377 Streamer = llvm::make_unique<DwarfStreamer>();
1378 return Streamer->init(TheTriple, OutputFilename);
1381 /// \brief Recursive helper to gather the child->parent relationships in the
1382 /// original compile unit.
1383 static void gatherDIEParents(const DWARFDebugInfoEntryMinimal *DIE,
1384 unsigned ParentIdx, CompileUnit &CU) {
1385 unsigned MyIdx = CU.getOrigUnit().getDIEIndex(DIE);
1386 CU.getInfo(MyIdx).ParentIdx = ParentIdx;
1388 if (DIE->hasChildren())
1389 for (auto *Child = DIE->getFirstChild(); Child && !Child->isNULL();
1390 Child = Child->getSibling())
1391 gatherDIEParents(Child, MyIdx, CU);
1394 static bool dieNeedsChildrenToBeMeaningful(uint32_t Tag) {
1398 case dwarf::DW_TAG_subprogram:
1399 case dwarf::DW_TAG_lexical_block:
1400 case dwarf::DW_TAG_subroutine_type:
1401 case dwarf::DW_TAG_structure_type:
1402 case dwarf::DW_TAG_class_type:
1403 case dwarf::DW_TAG_union_type:
1406 llvm_unreachable("Invalid Tag");
1409 void DwarfLinker::startDebugObject(DWARFContext &Dwarf, DebugMapObject &Obj) {
1410 Units.reserve(Dwarf.getNumCompileUnits());
1412 // Iterate over the debug map entries and put all the ones that are
1413 // functions (because they have a size) into the Ranges map. This
1414 // map is very similar to the FunctionRanges that are stored in each
1415 // unit, with 2 notable differences:
1416 // - obviously this one is global, while the other ones are per-unit.
1417 // - this one contains not only the functions described in the DIE
1418 // tree, but also the ones that are only in the debug map.
1419 // The latter information is required to reproduce dsymutil's logic
1420 // while linking line tables. The cases where this information
1421 // matters look like bugs that need to be investigated, but for now
1422 // we need to reproduce dsymutil's behavior.
1423 // FIXME: Once we understood exactly if that information is needed,
1424 // maybe totally remove this (or try to use it to do a real
1425 // -gline-tables-only on Darwin.
1426 for (const auto &Entry : Obj.symbols()) {
1427 const auto &Mapping = Entry.getValue();
1429 Ranges[Mapping.ObjectAddress] = std::make_pair(
1430 Mapping.ObjectAddress + Mapping.Size,
1431 int64_t(Mapping.BinaryAddress) - Mapping.ObjectAddress);
1435 void DwarfLinker::endDebugObject() {
1437 ValidRelocs.clear();
1440 for (auto *Block : DIEBlocks)
1442 for (auto *Loc : DIELocs)
1450 /// \brief Iterate over the relocations of the given \p Section and
1451 /// store the ones that correspond to debug map entries into the
1452 /// ValidRelocs array.
1453 void DwarfLinker::findValidRelocsMachO(const object::SectionRef &Section,
1454 const object::MachOObjectFile &Obj,
1455 const DebugMapObject &DMO) {
1457 Section.getContents(Contents);
1458 DataExtractor Data(Contents, Obj.isLittleEndian(), 0);
1460 for (const object::RelocationRef &Reloc : Section.relocations()) {
1461 object::DataRefImpl RelocDataRef = Reloc.getRawDataRefImpl();
1462 MachO::any_relocation_info MachOReloc = Obj.getRelocation(RelocDataRef);
1463 unsigned RelocSize = 1 << Obj.getAnyRelocationLength(MachOReloc);
1465 if ((RelocSize != 4 && RelocSize != 8) || Reloc.getOffset(Offset64)) {
1466 reportWarning(" unsupported relocation in debug_info section.");
1469 uint32_t Offset = Offset64;
1470 // Mach-o uses REL relocations, the addend is at the relocation offset.
1471 uint64_t Addend = Data.getUnsigned(&Offset, RelocSize);
1473 auto Sym = Reloc.getSymbol();
1474 if (Sym != Obj.symbol_end()) {
1475 StringRef SymbolName;
1476 if (Sym->getName(SymbolName)) {
1477 reportWarning("error getting relocation symbol name.");
1480 if (const auto *Mapping = DMO.lookupSymbol(SymbolName))
1481 ValidRelocs.emplace_back(Offset64, RelocSize, Addend, Mapping);
1482 } else if (const auto *Mapping = DMO.lookupObjectAddress(Addend)) {
1483 // Do not store the addend. The addend was the address of the
1484 // symbol in the object file, the address in the binary that is
1485 // stored in the debug map doesn't need to be offseted.
1486 ValidRelocs.emplace_back(Offset64, RelocSize, 0, Mapping);
1491 /// \brief Dispatch the valid relocation finding logic to the
1492 /// appropriate handler depending on the object file format.
1493 bool DwarfLinker::findValidRelocs(const object::SectionRef &Section,
1494 const object::ObjectFile &Obj,
1495 const DebugMapObject &DMO) {
1496 // Dispatch to the right handler depending on the file type.
1497 if (auto *MachOObj = dyn_cast<object::MachOObjectFile>(&Obj))
1498 findValidRelocsMachO(Section, *MachOObj, DMO);
1500 reportWarning(Twine("unsupported object file type: ") + Obj.getFileName());
1502 if (ValidRelocs.empty())
1505 // Sort the relocations by offset. We will walk the DIEs linearly in
1506 // the file, this allows us to just keep an index in the relocation
1507 // array that we advance during our walk, rather than resorting to
1508 // some associative container. See DwarfLinker::NextValidReloc.
1509 std::sort(ValidRelocs.begin(), ValidRelocs.end());
1513 /// \brief Look for relocations in the debug_info section that match
1514 /// entries in the debug map. These relocations will drive the Dwarf
1515 /// link by indicating which DIEs refer to symbols present in the
1517 /// \returns wether there are any valid relocations in the debug info.
1518 bool DwarfLinker::findValidRelocsInDebugInfo(const object::ObjectFile &Obj,
1519 const DebugMapObject &DMO) {
1520 // Find the debug_info section.
1521 for (const object::SectionRef &Section : Obj.sections()) {
1522 StringRef SectionName;
1523 Section.getName(SectionName);
1524 SectionName = SectionName.substr(SectionName.find_first_not_of("._"));
1525 if (SectionName != "debug_info")
1527 return findValidRelocs(Section, Obj, DMO);
1532 /// \brief Checks that there is a relocation against an actual debug
1533 /// map entry between \p StartOffset and \p NextOffset.
1535 /// This function must be called with offsets in strictly ascending
1536 /// order because it never looks back at relocations it already 'went past'.
1537 /// \returns true and sets Info.InDebugMap if it is the case.
1538 bool DwarfLinker::hasValidRelocation(uint32_t StartOffset, uint32_t EndOffset,
1539 CompileUnit::DIEInfo &Info) {
1540 assert(NextValidReloc == 0 ||
1541 StartOffset > ValidRelocs[NextValidReloc - 1].Offset);
1542 if (NextValidReloc >= ValidRelocs.size())
1545 uint64_t RelocOffset = ValidRelocs[NextValidReloc].Offset;
1547 // We might need to skip some relocs that we didn't consider. For
1548 // example the high_pc of a discarded DIE might contain a reloc that
1549 // is in the list because it actually corresponds to the start of a
1550 // function that is in the debug map.
1551 while (RelocOffset < StartOffset && NextValidReloc < ValidRelocs.size() - 1)
1552 RelocOffset = ValidRelocs[++NextValidReloc].Offset;
1554 if (RelocOffset < StartOffset || RelocOffset >= EndOffset)
1557 const auto &ValidReloc = ValidRelocs[NextValidReloc++];
1558 const auto &Mapping = ValidReloc.Mapping->getValue();
1559 if (Options.Verbose)
1560 outs() << "Found valid debug map entry: " << ValidReloc.Mapping->getKey()
1561 << " " << format("\t%016" PRIx64 " => %016" PRIx64,
1562 uint64_t(Mapping.ObjectAddress),
1563 uint64_t(Mapping.BinaryAddress));
1565 Info.AddrAdjust = int64_t(Mapping.BinaryAddress) + ValidReloc.Addend -
1566 Mapping.ObjectAddress;
1567 Info.InDebugMap = true;
1571 /// \brief Get the starting and ending (exclusive) offset for the
1572 /// attribute with index \p Idx descibed by \p Abbrev. \p Offset is
1573 /// supposed to point to the position of the first attribute described
1575 /// \return [StartOffset, EndOffset) as a pair.
1576 static std::pair<uint32_t, uint32_t>
1577 getAttributeOffsets(const DWARFAbbreviationDeclaration *Abbrev, unsigned Idx,
1578 unsigned Offset, const DWARFUnit &Unit) {
1579 DataExtractor Data = Unit.getDebugInfoExtractor();
1581 for (unsigned i = 0; i < Idx; ++i)
1582 DWARFFormValue::skipValue(Abbrev->getFormByIndex(i), Data, &Offset, &Unit);
1584 uint32_t End = Offset;
1585 DWARFFormValue::skipValue(Abbrev->getFormByIndex(Idx), Data, &End, &Unit);
1587 return std::make_pair(Offset, End);
1590 /// \brief Check if a variable describing DIE should be kept.
1591 /// \returns updated TraversalFlags.
1592 unsigned DwarfLinker::shouldKeepVariableDIE(
1593 const DWARFDebugInfoEntryMinimal &DIE, CompileUnit &Unit,
1594 CompileUnit::DIEInfo &MyInfo, unsigned Flags) {
1595 const auto *Abbrev = DIE.getAbbreviationDeclarationPtr();
1597 // Global variables with constant value can always be kept.
1598 if (!(Flags & TF_InFunctionScope) &&
1599 Abbrev->findAttributeIndex(dwarf::DW_AT_const_value) != -1U) {
1600 MyInfo.InDebugMap = true;
1601 return Flags | TF_Keep;
1604 uint32_t LocationIdx = Abbrev->findAttributeIndex(dwarf::DW_AT_location);
1605 if (LocationIdx == -1U)
1608 uint32_t Offset = DIE.getOffset() + getULEB128Size(Abbrev->getCode());
1609 const DWARFUnit &OrigUnit = Unit.getOrigUnit();
1610 uint32_t LocationOffset, LocationEndOffset;
1611 std::tie(LocationOffset, LocationEndOffset) =
1612 getAttributeOffsets(Abbrev, LocationIdx, Offset, OrigUnit);
1614 // See if there is a relocation to a valid debug map entry inside
1615 // this variable's location. The order is important here. We want to
1616 // always check in the variable has a valid relocation, so that the
1617 // DIEInfo is filled. However, we don't want a static variable in a
1618 // function to force us to keep the enclosing function.
1619 if (!hasValidRelocation(LocationOffset, LocationEndOffset, MyInfo) ||
1620 (Flags & TF_InFunctionScope))
1623 if (Options.Verbose)
1624 DIE.dump(outs(), const_cast<DWARFUnit *>(&OrigUnit), 0, 8 /* Indent */);
1626 return Flags | TF_Keep;
1629 /// \brief Check if a function describing DIE should be kept.
1630 /// \returns updated TraversalFlags.
1631 unsigned DwarfLinker::shouldKeepSubprogramDIE(
1632 const DWARFDebugInfoEntryMinimal &DIE, CompileUnit &Unit,
1633 CompileUnit::DIEInfo &MyInfo, unsigned Flags) {
1634 const auto *Abbrev = DIE.getAbbreviationDeclarationPtr();
1636 Flags |= TF_InFunctionScope;
1638 uint32_t LowPcIdx = Abbrev->findAttributeIndex(dwarf::DW_AT_low_pc);
1639 if (LowPcIdx == -1U)
1642 uint32_t Offset = DIE.getOffset() + getULEB128Size(Abbrev->getCode());
1643 const DWARFUnit &OrigUnit = Unit.getOrigUnit();
1644 uint32_t LowPcOffset, LowPcEndOffset;
1645 std::tie(LowPcOffset, LowPcEndOffset) =
1646 getAttributeOffsets(Abbrev, LowPcIdx, Offset, OrigUnit);
1649 DIE.getAttributeValueAsAddress(&OrigUnit, dwarf::DW_AT_low_pc, -1ULL);
1650 assert(LowPc != -1ULL && "low_pc attribute is not an address.");
1651 if (LowPc == -1ULL ||
1652 !hasValidRelocation(LowPcOffset, LowPcEndOffset, MyInfo))
1655 if (Options.Verbose)
1656 DIE.dump(outs(), const_cast<DWARFUnit *>(&OrigUnit), 0, 8 /* Indent */);
1660 DWARFFormValue HighPcValue;
1661 if (!DIE.getAttributeValue(&OrigUnit, dwarf::DW_AT_high_pc, HighPcValue)) {
1662 reportWarning("Function without high_pc. Range will be discarded.\n",
1668 if (HighPcValue.isFormClass(DWARFFormValue::FC_Address)) {
1669 HighPc = *HighPcValue.getAsAddress(&OrigUnit);
1671 assert(HighPcValue.isFormClass(DWARFFormValue::FC_Constant));
1672 HighPc = LowPc + *HighPcValue.getAsUnsignedConstant();
1675 // Replace the debug map range with a more accurate one.
1676 Ranges[LowPc] = std::make_pair(HighPc, MyInfo.AddrAdjust);
1677 Unit.addFunctionRange(LowPc, HighPc, MyInfo.AddrAdjust);
1681 /// \brief Check if a DIE should be kept.
1682 /// \returns updated TraversalFlags.
1683 unsigned DwarfLinker::shouldKeepDIE(const DWARFDebugInfoEntryMinimal &DIE,
1685 CompileUnit::DIEInfo &MyInfo,
1687 switch (DIE.getTag()) {
1688 case dwarf::DW_TAG_constant:
1689 case dwarf::DW_TAG_variable:
1690 return shouldKeepVariableDIE(DIE, Unit, MyInfo, Flags);
1691 case dwarf::DW_TAG_subprogram:
1692 return shouldKeepSubprogramDIE(DIE, Unit, MyInfo, Flags);
1693 case dwarf::DW_TAG_module:
1694 case dwarf::DW_TAG_imported_module:
1695 case dwarf::DW_TAG_imported_declaration:
1696 case dwarf::DW_TAG_imported_unit:
1697 // We always want to keep these.
1698 return Flags | TF_Keep;
1704 /// \brief Mark the passed DIE as well as all the ones it depends on
1707 /// This function is called by lookForDIEsToKeep on DIEs that are
1708 /// newly discovered to be needed in the link. It recursively calls
1709 /// back to lookForDIEsToKeep while adding TF_DependencyWalk to the
1710 /// TraversalFlags to inform it that it's not doing the primary DIE
1712 void DwarfLinker::keepDIEAndDenpendencies(const DWARFDebugInfoEntryMinimal &DIE,
1713 CompileUnit::DIEInfo &MyInfo,
1714 const DebugMapObject &DMO,
1715 CompileUnit &CU, unsigned Flags) {
1716 const DWARFUnit &Unit = CU.getOrigUnit();
1719 // First mark all the parent chain as kept.
1720 unsigned AncestorIdx = MyInfo.ParentIdx;
1721 while (!CU.getInfo(AncestorIdx).Keep) {
1722 lookForDIEsToKeep(*Unit.getDIEAtIndex(AncestorIdx), DMO, CU,
1723 TF_ParentWalk | TF_Keep | TF_DependencyWalk);
1724 AncestorIdx = CU.getInfo(AncestorIdx).ParentIdx;
1727 // Then we need to mark all the DIEs referenced by this DIE's
1728 // attributes as kept.
1729 DataExtractor Data = Unit.getDebugInfoExtractor();
1730 const auto *Abbrev = DIE.getAbbreviationDeclarationPtr();
1731 uint32_t Offset = DIE.getOffset() + getULEB128Size(Abbrev->getCode());
1733 // Mark all DIEs referenced through atttributes as kept.
1734 for (const auto &AttrSpec : Abbrev->attributes()) {
1735 DWARFFormValue Val(AttrSpec.Form);
1737 if (!Val.isFormClass(DWARFFormValue::FC_Reference)) {
1738 DWARFFormValue::skipValue(AttrSpec.Form, Data, &Offset, &Unit);
1742 Val.extractValue(Data, &Offset, &Unit);
1743 CompileUnit *ReferencedCU;
1744 if (const auto *RefDIE = resolveDIEReference(Val, Unit, DIE, ReferencedCU))
1745 lookForDIEsToKeep(*RefDIE, DMO, *ReferencedCU,
1746 TF_Keep | TF_DependencyWalk);
1750 /// \brief Recursively walk the \p DIE tree and look for DIEs to
1751 /// keep. Store that information in \p CU's DIEInfo.
1753 /// This function is the entry point of the DIE selection
1754 /// algorithm. It is expected to walk the DIE tree in file order and
1755 /// (though the mediation of its helper) call hasValidRelocation() on
1756 /// each DIE that might be a 'root DIE' (See DwarfLinker class
1758 /// While walking the dependencies of root DIEs, this function is
1759 /// also called, but during these dependency walks the file order is
1760 /// not respected. The TF_DependencyWalk flag tells us which kind of
1761 /// traversal we are currently doing.
1762 void DwarfLinker::lookForDIEsToKeep(const DWARFDebugInfoEntryMinimal &DIE,
1763 const DebugMapObject &DMO, CompileUnit &CU,
1765 unsigned Idx = CU.getOrigUnit().getDIEIndex(&DIE);
1766 CompileUnit::DIEInfo &MyInfo = CU.getInfo(Idx);
1767 bool AlreadyKept = MyInfo.Keep;
1769 // If the Keep flag is set, we are marking a required DIE's
1770 // dependencies. If our target is already marked as kept, we're all
1772 if ((Flags & TF_DependencyWalk) && AlreadyKept)
1775 // We must not call shouldKeepDIE while called from keepDIEAndDenpendencies,
1776 // because it would screw up the relocation finding logic.
1777 if (!(Flags & TF_DependencyWalk))
1778 Flags = shouldKeepDIE(DIE, CU, MyInfo, Flags);
1780 // If it is a newly kept DIE mark it as well as all its dependencies as kept.
1781 if (!AlreadyKept && (Flags & TF_Keep))
1782 keepDIEAndDenpendencies(DIE, MyInfo, DMO, CU, Flags);
1784 // The TF_ParentWalk flag tells us that we are currently walking up
1785 // the parent chain of a required DIE, and we don't want to mark all
1786 // the children of the parents as kept (consider for example a
1787 // DW_TAG_namespace node in the parent chain). There are however a
1788 // set of DIE types for which we want to ignore that directive and still
1789 // walk their children.
1790 if (dieNeedsChildrenToBeMeaningful(DIE.getTag()))
1791 Flags &= ~TF_ParentWalk;
1793 if (!DIE.hasChildren() || (Flags & TF_ParentWalk))
1796 for (auto *Child = DIE.getFirstChild(); Child && !Child->isNULL();
1797 Child = Child->getSibling())
1798 lookForDIEsToKeep(*Child, DMO, CU, Flags);
1801 /// \brief Assign an abbreviation numer to \p Abbrev.
1803 /// Our DIEs get freed after every DebugMapObject has been processed,
1804 /// thus the FoldingSet we use to unique DIEAbbrevs cannot refer to
1805 /// the instances hold by the DIEs. When we encounter an abbreviation
1806 /// that we don't know, we create a permanent copy of it.
1807 void DwarfLinker::AssignAbbrev(DIEAbbrev &Abbrev) {
1808 // Check the set for priors.
1809 FoldingSetNodeID ID;
1812 DIEAbbrev *InSet = AbbreviationsSet.FindNodeOrInsertPos(ID, InsertToken);
1814 // If it's newly added.
1816 // Assign existing abbreviation number.
1817 Abbrev.setNumber(InSet->getNumber());
1819 // Add to abbreviation list.
1820 Abbreviations.push_back(
1821 new DIEAbbrev(Abbrev.getTag(), Abbrev.hasChildren()));
1822 for (const auto &Attr : Abbrev.getData())
1823 Abbreviations.back()->AddAttribute(Attr.getAttribute(), Attr.getForm());
1824 AbbreviationsSet.InsertNode(Abbreviations.back(), InsertToken);
1825 // Assign the unique abbreviation number.
1826 Abbrev.setNumber(Abbreviations.size());
1827 Abbreviations.back()->setNumber(Abbreviations.size());
1831 /// \brief Clone a string attribute described by \p AttrSpec and add
1833 /// \returns the size of the new attribute.
1834 unsigned DwarfLinker::cloneStringAttribute(DIE &Die, AttributeSpec AttrSpec,
1835 const DWARFFormValue &Val,
1836 const DWARFUnit &U) {
1837 // Switch everything to out of line strings.
1838 const char *String = *Val.getAsCString(&U);
1839 unsigned Offset = StringPool.getStringOffset(String);
1840 Die.addValue(dwarf::Attribute(AttrSpec.Attr), dwarf::DW_FORM_strp,
1841 DIEInteger(Offset));
1845 /// \brief Clone an attribute referencing another DIE and add
1847 /// \returns the size of the new attribute.
1848 unsigned DwarfLinker::cloneDieReferenceAttribute(
1849 DIE &Die, const DWARFDebugInfoEntryMinimal &InputDIE,
1850 AttributeSpec AttrSpec, unsigned AttrSize, const DWARFFormValue &Val,
1851 CompileUnit &Unit) {
1852 uint32_t Ref = *Val.getAsReference(&Unit.getOrigUnit());
1853 DIE *NewRefDie = nullptr;
1854 CompileUnit *RefUnit = nullptr;
1855 const DWARFDebugInfoEntryMinimal *RefDie = nullptr;
1857 if (!(RefUnit = getUnitForOffset(Ref)) ||
1858 !(RefDie = RefUnit->getOrigUnit().getDIEForOffset(Ref))) {
1859 const char *AttributeString = dwarf::AttributeString(AttrSpec.Attr);
1860 if (!AttributeString)
1861 AttributeString = "DW_AT_???";
1862 reportWarning(Twine("Missing DIE for ref in attribute ") + AttributeString +
1864 &Unit.getOrigUnit(), &InputDIE);
1868 unsigned Idx = RefUnit->getOrigUnit().getDIEIndex(RefDie);
1869 CompileUnit::DIEInfo &RefInfo = RefUnit->getInfo(Idx);
1870 if (!RefInfo.Clone) {
1871 assert(Ref > InputDIE.getOffset());
1872 // We haven't cloned this DIE yet. Just create an empty one and
1873 // store it. It'll get really cloned when we process it.
1874 RefInfo.Clone = new DIE(dwarf::Tag(RefDie->getTag()));
1876 NewRefDie = RefInfo.Clone;
1878 if (AttrSpec.Form == dwarf::DW_FORM_ref_addr) {
1879 // We cannot currently rely on a DIEEntry to emit ref_addr
1880 // references, because the implementation calls back to DwarfDebug
1881 // to find the unit offset. (We don't have a DwarfDebug)
1882 // FIXME: we should be able to design DIEEntry reliance on
1885 if (Ref < InputDIE.getOffset()) {
1886 // We must have already cloned that DIE.
1887 uint32_t NewRefOffset =
1888 RefUnit->getStartOffset() + NewRefDie->getOffset();
1889 Attr = NewRefOffset;
1891 // A forward reference. Note and fixup later.
1893 Unit.noteForwardReference(NewRefDie, RefUnit, PatchLocation(Die));
1895 Die.addValue(dwarf::Attribute(AttrSpec.Attr), dwarf::DW_FORM_ref_addr,
1900 Die.addValue(dwarf::Attribute(AttrSpec.Attr), dwarf::Form(AttrSpec.Form),
1901 DIEEntry(*NewRefDie));
1905 /// \brief Clone an attribute of block form (locations, constants) and add
1907 /// \returns the size of the new attribute.
1908 unsigned DwarfLinker::cloneBlockAttribute(DIE &Die, AttributeSpec AttrSpec,
1909 const DWARFFormValue &Val,
1910 unsigned AttrSize) {
1913 DIELoc *Loc = nullptr;
1914 DIEBlock *Block = nullptr;
1915 // Just copy the block data over.
1916 if (AttrSpec.Form == dwarf::DW_FORM_exprloc) {
1917 Loc = new (DIEAlloc) DIELoc;
1918 DIELocs.push_back(Loc);
1920 Block = new (DIEAlloc) DIEBlock;
1921 DIEBlocks.push_back(Block);
1923 Attr = Loc ? static_cast<DIE *>(Loc) : static_cast<DIE *>(Block);
1926 Value = DIEValue(dwarf::Attribute(AttrSpec.Attr),
1927 dwarf::Form(AttrSpec.Form), Loc);
1929 Value = DIEValue(dwarf::Attribute(AttrSpec.Attr),
1930 dwarf::Form(AttrSpec.Form), Block);
1931 ArrayRef<uint8_t> Bytes = *Val.getAsBlock();
1932 for (auto Byte : Bytes)
1933 Attr->addValue(static_cast<dwarf::Attribute>(0), dwarf::DW_FORM_data1,
1935 // FIXME: If DIEBlock and DIELoc just reuses the Size field of
1936 // the DIE class, this if could be replaced by
1937 // Attr->setSize(Bytes.size()).
1940 Loc->ComputeSize(&Streamer->getAsmPrinter());
1942 Block->ComputeSize(&Streamer->getAsmPrinter());
1944 Die.addValue(Value);
1948 /// \brief Clone an address attribute and add it to \p Die.
1949 /// \returns the size of the new attribute.
1950 unsigned DwarfLinker::cloneAddressAttribute(DIE &Die, AttributeSpec AttrSpec,
1951 const DWARFFormValue &Val,
1952 const CompileUnit &Unit,
1953 AttributesInfo &Info) {
1954 uint64_t Addr = *Val.getAsAddress(&Unit.getOrigUnit());
1955 if (AttrSpec.Attr == dwarf::DW_AT_low_pc) {
1956 if (Die.getTag() == dwarf::DW_TAG_inlined_subroutine ||
1957 Die.getTag() == dwarf::DW_TAG_lexical_block)
1958 Addr += Info.PCOffset;
1959 else if (Die.getTag() == dwarf::DW_TAG_compile_unit) {
1960 Addr = Unit.getLowPc();
1961 if (Addr == UINT64_MAX)
1964 Info.HasLowPc = true;
1965 } else if (AttrSpec.Attr == dwarf::DW_AT_high_pc) {
1966 if (Die.getTag() == dwarf::DW_TAG_compile_unit) {
1967 if (uint64_t HighPc = Unit.getHighPc())
1972 // If we have a high_pc recorded for the input DIE, use
1973 // it. Otherwise (when no relocations where applied) just use the
1974 // one we just decoded.
1975 Addr = (Info.OrigHighPc ? Info.OrigHighPc : Addr) + Info.PCOffset;
1978 Die.addValue(static_cast<dwarf::Attribute>(AttrSpec.Attr),
1979 static_cast<dwarf::Form>(AttrSpec.Form), DIEInteger(Addr));
1980 return Unit.getOrigUnit().getAddressByteSize();
1983 /// \brief Clone a scalar attribute and add it to \p Die.
1984 /// \returns the size of the new attribute.
1985 unsigned DwarfLinker::cloneScalarAttribute(
1986 DIE &Die, const DWARFDebugInfoEntryMinimal &InputDIE, CompileUnit &Unit,
1987 AttributeSpec AttrSpec, const DWARFFormValue &Val, unsigned AttrSize,
1988 AttributesInfo &Info) {
1990 if (AttrSpec.Attr == dwarf::DW_AT_high_pc &&
1991 Die.getTag() == dwarf::DW_TAG_compile_unit) {
1992 if (Unit.getLowPc() == -1ULL)
1994 // Dwarf >= 4 high_pc is an size, not an address.
1995 Value = Unit.getHighPc() - Unit.getLowPc();
1996 } else if (AttrSpec.Form == dwarf::DW_FORM_sec_offset)
1997 Value = *Val.getAsSectionOffset();
1998 else if (AttrSpec.Form == dwarf::DW_FORM_sdata)
1999 Value = *Val.getAsSignedConstant();
2000 else if (auto OptionalValue = Val.getAsUnsignedConstant())
2001 Value = *OptionalValue;
2003 reportWarning("Unsupported scalar attribute form. Dropping attribute.",
2004 &Unit.getOrigUnit(), &InputDIE);
2007 DIEInteger Attr(Value);
2008 if (AttrSpec.Attr == dwarf::DW_AT_ranges)
2009 Unit.noteRangeAttribute(Die, PatchLocation(Die));
2010 // A more generic way to check for location attributes would be
2011 // nice, but it's very unlikely that any other attribute needs a
2013 else if (AttrSpec.Attr == dwarf::DW_AT_location ||
2014 AttrSpec.Attr == dwarf::DW_AT_frame_base)
2015 Unit.noteLocationAttribute(PatchLocation(Die), Info.PCOffset);
2016 else if (AttrSpec.Attr == dwarf::DW_AT_declaration && Value)
2017 Info.IsDeclaration = true;
2019 Die.addValue(dwarf::Attribute(AttrSpec.Attr), dwarf::Form(AttrSpec.Form),
2024 /// \brief Clone \p InputDIE's attribute described by \p AttrSpec with
2025 /// value \p Val, and add it to \p Die.
2026 /// \returns the size of the cloned attribute.
2027 unsigned DwarfLinker::cloneAttribute(DIE &Die,
2028 const DWARFDebugInfoEntryMinimal &InputDIE,
2030 const DWARFFormValue &Val,
2031 const AttributeSpec AttrSpec,
2032 unsigned AttrSize, AttributesInfo &Info) {
2033 const DWARFUnit &U = Unit.getOrigUnit();
2035 switch (AttrSpec.Form) {
2036 case dwarf::DW_FORM_strp:
2037 case dwarf::DW_FORM_string:
2038 return cloneStringAttribute(Die, AttrSpec, Val, U);
2039 case dwarf::DW_FORM_ref_addr:
2040 case dwarf::DW_FORM_ref1:
2041 case dwarf::DW_FORM_ref2:
2042 case dwarf::DW_FORM_ref4:
2043 case dwarf::DW_FORM_ref8:
2044 return cloneDieReferenceAttribute(Die, InputDIE, AttrSpec, AttrSize, Val,
2046 case dwarf::DW_FORM_block:
2047 case dwarf::DW_FORM_block1:
2048 case dwarf::DW_FORM_block2:
2049 case dwarf::DW_FORM_block4:
2050 case dwarf::DW_FORM_exprloc:
2051 return cloneBlockAttribute(Die, AttrSpec, Val, AttrSize);
2052 case dwarf::DW_FORM_addr:
2053 return cloneAddressAttribute(Die, AttrSpec, Val, Unit, Info);
2054 case dwarf::DW_FORM_data1:
2055 case dwarf::DW_FORM_data2:
2056 case dwarf::DW_FORM_data4:
2057 case dwarf::DW_FORM_data8:
2058 case dwarf::DW_FORM_udata:
2059 case dwarf::DW_FORM_sdata:
2060 case dwarf::DW_FORM_sec_offset:
2061 case dwarf::DW_FORM_flag:
2062 case dwarf::DW_FORM_flag_present:
2063 return cloneScalarAttribute(Die, InputDIE, Unit, AttrSpec, Val, AttrSize,
2066 reportWarning("Unsupported attribute form in cloneAttribute. Dropping.", &U,
2073 /// \brief Apply the valid relocations found by findValidRelocs() to
2074 /// the buffer \p Data, taking into account that Data is at \p BaseOffset
2075 /// in the debug_info section.
2077 /// Like for findValidRelocs(), this function must be called with
2078 /// monotonic \p BaseOffset values.
2080 /// \returns wether any reloc has been applied.
2081 bool DwarfLinker::applyValidRelocs(MutableArrayRef<char> Data,
2082 uint32_t BaseOffset, bool isLittleEndian) {
2083 assert((NextValidReloc == 0 ||
2084 BaseOffset > ValidRelocs[NextValidReloc - 1].Offset) &&
2085 "BaseOffset should only be increasing.");
2086 if (NextValidReloc >= ValidRelocs.size())
2089 // Skip relocs that haven't been applied.
2090 while (NextValidReloc < ValidRelocs.size() &&
2091 ValidRelocs[NextValidReloc].Offset < BaseOffset)
2094 bool Applied = false;
2095 uint64_t EndOffset = BaseOffset + Data.size();
2096 while (NextValidReloc < ValidRelocs.size() &&
2097 ValidRelocs[NextValidReloc].Offset >= BaseOffset &&
2098 ValidRelocs[NextValidReloc].Offset < EndOffset) {
2099 const auto &ValidReloc = ValidRelocs[NextValidReloc++];
2100 assert(ValidReloc.Offset - BaseOffset < Data.size());
2101 assert(ValidReloc.Offset - BaseOffset + ValidReloc.Size <= Data.size());
2103 uint64_t Value = ValidReloc.Mapping->getValue().BinaryAddress;
2104 Value += ValidReloc.Addend;
2105 for (unsigned i = 0; i != ValidReloc.Size; ++i) {
2106 unsigned Index = isLittleEndian ? i : (ValidReloc.Size - i - 1);
2107 Buf[i] = uint8_t(Value >> (Index * 8));
2109 assert(ValidReloc.Size <= sizeof(Buf));
2110 memcpy(&Data[ValidReloc.Offset - BaseOffset], Buf, ValidReloc.Size);
2117 static bool isTypeTag(uint16_t Tag) {
2119 case dwarf::DW_TAG_array_type:
2120 case dwarf::DW_TAG_class_type:
2121 case dwarf::DW_TAG_enumeration_type:
2122 case dwarf::DW_TAG_pointer_type:
2123 case dwarf::DW_TAG_reference_type:
2124 case dwarf::DW_TAG_string_type:
2125 case dwarf::DW_TAG_structure_type:
2126 case dwarf::DW_TAG_subroutine_type:
2127 case dwarf::DW_TAG_typedef:
2128 case dwarf::DW_TAG_union_type:
2129 case dwarf::DW_TAG_ptr_to_member_type:
2130 case dwarf::DW_TAG_set_type:
2131 case dwarf::DW_TAG_subrange_type:
2132 case dwarf::DW_TAG_base_type:
2133 case dwarf::DW_TAG_const_type:
2134 case dwarf::DW_TAG_constant:
2135 case dwarf::DW_TAG_file_type:
2136 case dwarf::DW_TAG_namelist:
2137 case dwarf::DW_TAG_packed_type:
2138 case dwarf::DW_TAG_volatile_type:
2139 case dwarf::DW_TAG_restrict_type:
2140 case dwarf::DW_TAG_interface_type:
2141 case dwarf::DW_TAG_unspecified_type:
2142 case dwarf::DW_TAG_shared_type:
2150 /// \brief Recursively clone \p InputDIE's subtrees that have been
2151 /// selected to appear in the linked output.
2153 /// \param OutOffset is the Offset where the newly created DIE will
2154 /// lie in the linked compile unit.
2156 /// \returns the cloned DIE object or null if nothing was selected.
2157 DIE *DwarfLinker::cloneDIE(const DWARFDebugInfoEntryMinimal &InputDIE,
2158 CompileUnit &Unit, int64_t PCOffset,
2159 uint32_t OutOffset) {
2160 DWARFUnit &U = Unit.getOrigUnit();
2161 unsigned Idx = U.getDIEIndex(&InputDIE);
2162 CompileUnit::DIEInfo &Info = Unit.getInfo(Idx);
2164 // Should the DIE appear in the output?
2165 if (!Unit.getInfo(Idx).Keep)
2168 uint32_t Offset = InputDIE.getOffset();
2169 // The DIE might have been already created by a forward reference
2170 // (see cloneDieReferenceAttribute()).
2171 DIE *Die = Info.Clone;
2173 Die = Info.Clone = new DIE(dwarf::Tag(InputDIE.getTag()));
2174 assert(Die->getTag() == InputDIE.getTag());
2175 Die->setOffset(OutOffset);
2177 // Extract and clone every attribute.
2178 DataExtractor Data = U.getDebugInfoExtractor();
2179 uint32_t NextOffset = U.getDIEAtIndex(Idx + 1)->getOffset();
2180 AttributesInfo AttrInfo;
2182 // We could copy the data only if we need to aply a relocation to
2183 // it. After testing, it seems there is no performance downside to
2184 // doing the copy unconditionally, and it makes the code simpler.
2185 SmallString<40> DIECopy(Data.getData().substr(Offset, NextOffset - Offset));
2186 Data = DataExtractor(DIECopy, Data.isLittleEndian(), Data.getAddressSize());
2187 // Modify the copy with relocated addresses.
2188 if (applyValidRelocs(DIECopy, Offset, Data.isLittleEndian())) {
2189 // If we applied relocations, we store the value of high_pc that was
2190 // potentially stored in the input DIE. If high_pc is an address
2191 // (Dwarf version == 2), then it might have been relocated to a
2192 // totally unrelated value (because the end address in the object
2193 // file might be start address of another function which got moved
2194 // independantly by the linker). The computation of the actual
2195 // high_pc value is done in cloneAddressAttribute().
2196 AttrInfo.OrigHighPc =
2197 InputDIE.getAttributeValueAsAddress(&U, dwarf::DW_AT_high_pc, 0);
2200 // Reset the Offset to 0 as we will be working on the local copy of
2204 const auto *Abbrev = InputDIE.getAbbreviationDeclarationPtr();
2205 Offset += getULEB128Size(Abbrev->getCode());
2207 // We are entering a subprogram. Get and propagate the PCOffset.
2208 if (Die->getTag() == dwarf::DW_TAG_subprogram)
2209 PCOffset = Info.AddrAdjust;
2210 AttrInfo.PCOffset = PCOffset;
2212 for (const auto &AttrSpec : Abbrev->attributes()) {
2213 DWARFFormValue Val(AttrSpec.Form);
2214 uint32_t AttrSize = Offset;
2215 Val.extractValue(Data, &Offset, &U);
2216 AttrSize = Offset - AttrSize;
2219 cloneAttribute(*Die, InputDIE, Unit, Val, AttrSpec, AttrSize, AttrInfo);
2222 // Look for accelerator entries.
2223 uint16_t Tag = InputDIE.getTag();
2224 // FIXME: This is slightly wrong. An inline_subroutine without a
2225 // low_pc, but with AT_ranges might be interesting to get into the
2226 // accelerator tables too. For now stick with dsymutil's behavior.
2227 if ((Info.InDebugMap || AttrInfo.HasLowPc) &&
2228 Tag != dwarf::DW_TAG_compile_unit &&
2229 getDIENames(InputDIE, Unit.getOrigUnit(), AttrInfo)) {
2230 if (AttrInfo.MangledName && AttrInfo.MangledName != AttrInfo.Name)
2231 Unit.addNameAccelerator(Die, AttrInfo.MangledName,
2232 AttrInfo.MangledNameOffset,
2233 Tag == dwarf::DW_TAG_inlined_subroutine);
2235 Unit.addNameAccelerator(Die, AttrInfo.Name, AttrInfo.NameOffset,
2236 Tag == dwarf::DW_TAG_inlined_subroutine);
2237 } else if (isTypeTag(Tag) && !AttrInfo.IsDeclaration &&
2238 getDIENames(InputDIE, Unit.getOrigUnit(), AttrInfo)) {
2239 Unit.addTypeAccelerator(Die, AttrInfo.Name, AttrInfo.NameOffset);
2242 DIEAbbrev NewAbbrev = Die->generateAbbrev();
2243 // If a scope DIE is kept, we must have kept at least one child. If
2244 // it's not the case, we'll just be emitting one wasteful end of
2245 // children marker, but things won't break.
2246 if (InputDIE.hasChildren())
2247 NewAbbrev.setChildrenFlag(dwarf::DW_CHILDREN_yes);
2248 // Assign a permanent abbrev number
2249 AssignAbbrev(NewAbbrev);
2250 Die->setAbbrevNumber(NewAbbrev.getNumber());
2252 // Add the size of the abbreviation number to the output offset.
2253 OutOffset += getULEB128Size(Die->getAbbrevNumber());
2255 if (!Abbrev->hasChildren()) {
2257 Die->setSize(OutOffset - Die->getOffset());
2261 // Recursively clone children.
2262 for (auto *Child = InputDIE.getFirstChild(); Child && !Child->isNULL();
2263 Child = Child->getSibling()) {
2264 if (DIE *Clone = cloneDIE(*Child, Unit, PCOffset, OutOffset)) {
2265 Die->addChild(std::unique_ptr<DIE>(Clone));
2266 OutOffset = Clone->getOffset() + Clone->getSize();
2270 // Account for the end of children marker.
2271 OutOffset += sizeof(int8_t);
2273 Die->setSize(OutOffset - Die->getOffset());
2277 /// \brief Patch the input object file relevant debug_ranges entries
2278 /// and emit them in the output file. Update the relevant attributes
2279 /// to point at the new entries.
2280 void DwarfLinker::patchRangesForUnit(const CompileUnit &Unit,
2281 DWARFContext &OrigDwarf) const {
2282 DWARFDebugRangeList RangeList;
2283 const auto &FunctionRanges = Unit.getFunctionRanges();
2284 unsigned AddressSize = Unit.getOrigUnit().getAddressByteSize();
2285 DataExtractor RangeExtractor(OrigDwarf.getRangeSection(),
2286 OrigDwarf.isLittleEndian(), AddressSize);
2287 auto InvalidRange = FunctionRanges.end(), CurrRange = InvalidRange;
2288 DWARFUnit &OrigUnit = Unit.getOrigUnit();
2289 const auto *OrigUnitDie = OrigUnit.getUnitDIE(false);
2290 uint64_t OrigLowPc = OrigUnitDie->getAttributeValueAsAddress(
2291 &OrigUnit, dwarf::DW_AT_low_pc, -1ULL);
2292 // Ranges addresses are based on the unit's low_pc. Compute the
2293 // offset we need to apply to adapt to the the new unit's low_pc.
2294 int64_t UnitPcOffset = 0;
2295 if (OrigLowPc != -1ULL)
2296 UnitPcOffset = int64_t(OrigLowPc) - Unit.getLowPc();
2298 for (const auto &RangeAttribute : Unit.getRangesAttributes()) {
2299 uint32_t Offset = RangeAttribute.get();
2300 RangeAttribute.set(Streamer->getRangesSectionSize());
2301 RangeList.extract(RangeExtractor, &Offset);
2302 const auto &Entries = RangeList.getEntries();
2303 const DWARFDebugRangeList::RangeListEntry &First = Entries.front();
2305 if (CurrRange == InvalidRange || First.StartAddress < CurrRange.start() ||
2306 First.StartAddress >= CurrRange.stop()) {
2307 CurrRange = FunctionRanges.find(First.StartAddress + OrigLowPc);
2308 if (CurrRange == InvalidRange ||
2309 CurrRange.start() > First.StartAddress + OrigLowPc) {
2310 reportWarning("no mapping for range.");
2315 Streamer->emitRangesEntries(UnitPcOffset, OrigLowPc, CurrRange, Entries,
2320 /// \brief Generate the debug_aranges entries for \p Unit and if the
2321 /// unit has a DW_AT_ranges attribute, also emit the debug_ranges
2322 /// contribution for this attribute.
2323 /// FIXME: this could actually be done right in patchRangesForUnit,
2324 /// but for the sake of initial bit-for-bit compatibility with legacy
2325 /// dsymutil, we have to do it in a delayed pass.
2326 void DwarfLinker::generateUnitRanges(CompileUnit &Unit) const {
2327 auto Attr = Unit.getUnitRangesAttribute();
2329 Attr->set(Streamer->getRangesSectionSize());
2330 Streamer->emitUnitRangesEntries(Unit, static_cast<bool>(Attr));
2333 /// \brief Insert the new line info sequence \p Seq into the current
2334 /// set of already linked line info \p Rows.
2335 static void insertLineSequence(std::vector<DWARFDebugLine::Row> &Seq,
2336 std::vector<DWARFDebugLine::Row> &Rows) {
2340 if (!Rows.empty() && Rows.back().Address < Seq.front().Address) {
2341 Rows.insert(Rows.end(), Seq.begin(), Seq.end());
2346 auto InsertPoint = std::lower_bound(
2347 Rows.begin(), Rows.end(), Seq.front(),
2348 [](const DWARFDebugLine::Row &LHS, const DWARFDebugLine::Row &RHS) {
2349 return LHS.Address < RHS.Address;
2352 // FIXME: this only removes the unneeded end_sequence if the
2353 // sequences have been inserted in order. using a global sort like
2354 // described in patchLineTableForUnit() and delaying the end_sequene
2355 // elimination to emitLineTableForUnit() we can get rid of all of them.
2356 if (InsertPoint != Rows.end() &&
2357 InsertPoint->Address == Seq.front().Address && InsertPoint->EndSequence) {
2358 *InsertPoint = Seq.front();
2359 Rows.insert(InsertPoint + 1, Seq.begin() + 1, Seq.end());
2361 Rows.insert(InsertPoint, Seq.begin(), Seq.end());
2367 /// \brief Extract the line table for \p Unit from \p OrigDwarf, and
2368 /// recreate a relocated version of these for the address ranges that
2369 /// are present in the binary.
2370 void DwarfLinker::patchLineTableForUnit(CompileUnit &Unit,
2371 DWARFContext &OrigDwarf) {
2372 const DWARFDebugInfoEntryMinimal *CUDie = Unit.getOrigUnit().getUnitDIE();
2373 uint64_t StmtList = CUDie->getAttributeValueAsSectionOffset(
2374 &Unit.getOrigUnit(), dwarf::DW_AT_stmt_list, -1ULL);
2375 if (StmtList == -1ULL)
2378 // Update the cloned DW_AT_stmt_list with the correct debug_line offset.
2379 if (auto *OutputDIE = Unit.getOutputUnitDIE()) {
2381 std::find_if(OutputDIE->values_begin(), OutputDIE->values_end(),
2382 [](const DIEValue &Value) {
2383 return Value.getAttribute() == dwarf::DW_AT_stmt_list;
2385 assert(Stmt != OutputDIE->values_end() &&
2386 "Didn't find DW_AT_stmt_list in cloned DIE!");
2387 OutputDIE->setValue(Stmt - OutputDIE->values_begin(),
2388 DIEValue(Stmt->getAttribute(), Stmt->getForm(),
2389 DIEInteger(Streamer->getLineSectionSize())));
2392 // Parse the original line info for the unit.
2393 DWARFDebugLine::LineTable LineTable;
2394 uint32_t StmtOffset = StmtList;
2395 StringRef LineData = OrigDwarf.getLineSection().Data;
2396 DataExtractor LineExtractor(LineData, OrigDwarf.isLittleEndian(),
2397 Unit.getOrigUnit().getAddressByteSize());
2398 LineTable.parse(LineExtractor, &OrigDwarf.getLineSection().Relocs,
2401 // This vector is the output line table.
2402 std::vector<DWARFDebugLine::Row> NewRows;
2403 NewRows.reserve(LineTable.Rows.size());
2405 // Current sequence of rows being extracted, before being inserted
2407 std::vector<DWARFDebugLine::Row> Seq;
2408 const auto &FunctionRanges = Unit.getFunctionRanges();
2409 auto InvalidRange = FunctionRanges.end(), CurrRange = InvalidRange;
2411 // FIXME: This logic is meant to generate exactly the same output as
2412 // Darwin's classic dsynutil. There is a nicer way to implement this
2413 // by simply putting all the relocated line info in NewRows and simply
2414 // sorting NewRows before passing it to emitLineTableForUnit. This
2415 // should be correct as sequences for a function should stay
2416 // together in the sorted output. There are a few corner cases that
2417 // look suspicious though, and that required to implement the logic
2418 // this way. Revisit that once initial validation is finished.
2420 // Iterate over the object file line info and extract the sequences
2421 // that correspond to linked functions.
2422 for (auto &Row : LineTable.Rows) {
2423 // Check wether we stepped out of the range. The range is
2424 // half-open, but consider accept the end address of the range if
2425 // it is marked as end_sequence in the input (because in that
2426 // case, the relocation offset is accurate and that entry won't
2427 // serve as the start of another function).
2428 if (CurrRange == InvalidRange || Row.Address < CurrRange.start() ||
2429 Row.Address > CurrRange.stop() ||
2430 (Row.Address == CurrRange.stop() && !Row.EndSequence)) {
2431 // We just stepped out of a known range. Insert a end_sequence
2432 // corresponding to the end of the range.
2433 uint64_t StopAddress = CurrRange != InvalidRange
2434 ? CurrRange.stop() + CurrRange.value()
2436 CurrRange = FunctionRanges.find(Row.Address);
2437 bool CurrRangeValid =
2438 CurrRange != InvalidRange && CurrRange.start() <= Row.Address;
2439 if (!CurrRangeValid) {
2440 CurrRange = InvalidRange;
2441 if (StopAddress != -1ULL) {
2442 // Try harder by looking in the DebugMapObject function
2443 // ranges map. There are corner cases where this finds a
2444 // valid entry. It's unclear if this is right or wrong, but
2445 // for now do as dsymutil.
2446 // FIXME: Understand exactly what cases this addresses and
2447 // potentially remove it along with the Ranges map.
2448 auto Range = Ranges.lower_bound(Row.Address);
2449 if (Range != Ranges.begin() && Range != Ranges.end())
2452 if (Range != Ranges.end() && Range->first <= Row.Address &&
2453 Range->second.first >= Row.Address) {
2454 StopAddress = Row.Address + Range->second.second;
2458 if (StopAddress != -1ULL && !Seq.empty()) {
2459 // Insert end sequence row with the computed end address, but
2460 // the same line as the previous one.
2461 Seq.emplace_back(Seq.back());
2462 Seq.back().Address = StopAddress;
2463 Seq.back().EndSequence = 1;
2464 Seq.back().PrologueEnd = 0;
2465 Seq.back().BasicBlock = 0;
2466 Seq.back().EpilogueBegin = 0;
2467 insertLineSequence(Seq, NewRows);
2470 if (!CurrRangeValid)
2474 // Ignore empty sequences.
2475 if (Row.EndSequence && Seq.empty())
2478 // Relocate row address and add it to the current sequence.
2479 Row.Address += CurrRange.value();
2480 Seq.emplace_back(Row);
2482 if (Row.EndSequence)
2483 insertLineSequence(Seq, NewRows);
2486 // Finished extracting, now emit the line tables.
2487 uint32_t PrologueEnd = StmtList + 10 + LineTable.Prologue.PrologueLength;
2488 // FIXME: LLVM hardcodes it's prologue values. We just copy the
2489 // prologue over and that works because we act as both producer and
2490 // consumer. It would be nicer to have a real configurable line
2492 if (LineTable.Prologue.Version != 2 ||
2493 LineTable.Prologue.DefaultIsStmt != DWARF2_LINE_DEFAULT_IS_STMT ||
2494 LineTable.Prologue.LineBase != -5 || LineTable.Prologue.LineRange != 14 ||
2495 LineTable.Prologue.OpcodeBase != 13)
2496 reportWarning("line table paramters mismatch. Cannot emit.");
2498 Streamer->emitLineTableForUnit(LineData.slice(StmtList + 4, PrologueEnd),
2499 LineTable.Prologue.MinInstLength, NewRows,
2500 Unit.getOrigUnit().getAddressByteSize());
2503 void DwarfLinker::emitAcceleratorEntriesForUnit(CompileUnit &Unit) {
2504 Streamer->emitPubNamesForUnit(Unit);
2505 Streamer->emitPubTypesForUnit(Unit);
2508 /// \brief Read the frame info stored in the object, and emit the
2509 /// patched frame descriptions for the linked binary.
2511 /// This is actually pretty easy as the data of the CIEs and FDEs can
2512 /// be considered as black boxes and moved as is. The only thing to do
2513 /// is to patch the addresses in the headers.
2514 void DwarfLinker::patchFrameInfoForObject(const DebugMapObject &DMO,
2515 DWARFContext &OrigDwarf,
2516 unsigned AddrSize) {
2517 StringRef FrameData = OrigDwarf.getDebugFrameSection();
2518 if (FrameData.empty())
2521 DataExtractor Data(FrameData, OrigDwarf.isLittleEndian(), 0);
2522 uint32_t InputOffset = 0;
2524 // Store the data of the CIEs defined in this object, keyed by their
2526 DenseMap<uint32_t, StringRef> LocalCIES;
2528 while (Data.isValidOffset(InputOffset)) {
2529 uint32_t EntryOffset = InputOffset;
2530 uint32_t InitialLength = Data.getU32(&InputOffset);
2531 if (InitialLength == 0xFFFFFFFF)
2532 return reportWarning("Dwarf64 bits no supported");
2534 uint32_t CIEId = Data.getU32(&InputOffset);
2535 if (CIEId == 0xFFFFFFFF) {
2536 // This is a CIE, store it.
2537 StringRef CIEData = FrameData.substr(EntryOffset, InitialLength + 4);
2538 LocalCIES[EntryOffset] = CIEData;
2539 // The -4 is to account for the CIEId we just read.
2540 InputOffset += InitialLength - 4;
2544 uint32_t Loc = Data.getUnsigned(&InputOffset, AddrSize);
2546 // Some compilers seem to emit frame info that doesn't start at
2547 // the function entry point, thus we can't just lookup the address
2548 // in the debug map. Use the linker's range map to see if the FDE
2549 // describes something that we can relocate.
2550 auto Range = Ranges.upper_bound(Loc);
2551 if (Range != Ranges.begin())
2553 if (Range == Ranges.end() || Range->first > Loc ||
2554 Range->second.first <= Loc) {
2555 // The +4 is to account for the size of the InitialLength field itself.
2556 InputOffset = EntryOffset + InitialLength + 4;
2560 // This is an FDE, and we have a mapping.
2561 // Have we already emitted a corresponding CIE?
2562 StringRef CIEData = LocalCIES[CIEId];
2563 if (CIEData.empty())
2564 return reportWarning("Inconsistent debug_frame content. Dropping.");
2566 // Look if we already emitted a CIE that corresponds to the
2567 // referenced one (the CIE data is the key of that lookup).
2568 auto IteratorInserted = EmittedCIEs.insert(
2569 std::make_pair(CIEData, Streamer->getFrameSectionSize()));
2570 // If there is no CIE yet for this ID, emit it.
2571 if (IteratorInserted.second ||
2572 // FIXME: dsymutil-classic only caches the last used CIE for
2573 // reuse. Mimic that behavior for now. Just removing that
2574 // second half of the condition and the LastCIEOffset variable
2575 // makes the code DTRT.
2576 LastCIEOffset != IteratorInserted.first->getValue()) {
2577 LastCIEOffset = Streamer->getFrameSectionSize();
2578 IteratorInserted.first->getValue() = LastCIEOffset;
2579 Streamer->emitCIE(CIEData);
2582 // Emit the FDE with updated address and CIE pointer.
2583 // (4 + AddrSize) is the size of the CIEId + initial_location
2584 // fields that will get reconstructed by emitFDE().
2585 unsigned FDERemainingBytes = InitialLength - (4 + AddrSize);
2586 Streamer->emitFDE(IteratorInserted.first->getValue(), AddrSize,
2587 Loc + Range->second.second,
2588 FrameData.substr(InputOffset, FDERemainingBytes));
2589 InputOffset += FDERemainingBytes;
2593 bool DwarfLinker::link(const DebugMap &Map) {
2595 if (Map.begin() == Map.end()) {
2596 errs() << "Empty debug map.\n";
2600 if (!createStreamer(Map.getTriple(), OutputFilename))
2603 // Size of the DIEs (and headers) generated for the linked output.
2604 uint64_t OutputDebugInfoSize = 0;
2605 // A unique ID that identifies each compile unit.
2606 unsigned UnitID = 0;
2607 for (const auto &Obj : Map.objects()) {
2608 CurrentDebugObject = Obj.get();
2610 if (Options.Verbose)
2611 outs() << "DEBUG MAP OBJECT: " << Obj->getObjectFilename() << "\n";
2612 auto ErrOrObj = BinHolder.GetObjectFile(Obj->getObjectFilename());
2613 if (std::error_code EC = ErrOrObj.getError()) {
2614 reportWarning(Twine(Obj->getObjectFilename()) + ": " + EC.message());
2618 // Look for relocations that correspond to debug map entries.
2619 if (!findValidRelocsInDebugInfo(*ErrOrObj, *Obj)) {
2620 if (Options.Verbose)
2621 outs() << "No valid relocations found. Skipping.\n";
2625 // Setup access to the debug info.
2626 DWARFContextInMemory DwarfContext(*ErrOrObj);
2627 startDebugObject(DwarfContext, *Obj);
2629 // In a first phase, just read in the debug info and store the DIE
2630 // parent links that we will use during the next phase.
2631 for (const auto &CU : DwarfContext.compile_units()) {
2632 auto *CUDie = CU->getUnitDIE(false);
2633 if (Options.Verbose) {
2634 outs() << "Input compilation unit:";
2635 CUDie->dump(outs(), CU.get(), 0);
2637 Units.emplace_back(*CU, UnitID++);
2638 gatherDIEParents(CUDie, 0, Units.back());
2641 // Then mark all the DIEs that need to be present in the linked
2642 // output and collect some information about them. Note that this
2643 // loop can not be merged with the previous one becaue cross-cu
2644 // references require the ParentIdx to be setup for every CU in
2645 // the object file before calling this.
2646 for (auto &CurrentUnit : Units)
2647 lookForDIEsToKeep(*CurrentUnit.getOrigUnit().getUnitDIE(), *Obj,
2650 // The calls to applyValidRelocs inside cloneDIE will walk the
2651 // reloc array again (in the same way findValidRelocsInDebugInfo()
2652 // did). We need to reset the NextValidReloc index to the beginning.
2655 // Construct the output DIE tree by cloning the DIEs we chose to
2656 // keep above. If there are no valid relocs, then there's nothing
2658 if (!ValidRelocs.empty())
2659 for (auto &CurrentUnit : Units) {
2660 const auto *InputDIE = CurrentUnit.getOrigUnit().getUnitDIE();
2661 CurrentUnit.setStartOffset(OutputDebugInfoSize);
2662 DIE *OutputDIE = cloneDIE(*InputDIE, CurrentUnit, 0 /* PCOffset */,
2663 11 /* Unit Header size */);
2664 CurrentUnit.setOutputUnitDIE(OutputDIE);
2665 OutputDebugInfoSize = CurrentUnit.computeNextUnitOffset();
2666 if (Options.NoOutput)
2668 // FIXME: for compatibility with the classic dsymutil, we emit
2669 // an empty line table for the unit, even if the unit doesn't
2670 // actually exist in the DIE tree.
2671 patchLineTableForUnit(CurrentUnit, DwarfContext);
2674 patchRangesForUnit(CurrentUnit, DwarfContext);
2675 Streamer->emitLocationsForUnit(CurrentUnit, DwarfContext);
2676 emitAcceleratorEntriesForUnit(CurrentUnit);
2679 // Emit all the compile unit's debug information.
2680 if (!ValidRelocs.empty() && !Options.NoOutput)
2681 for (auto &CurrentUnit : Units) {
2682 generateUnitRanges(CurrentUnit);
2683 CurrentUnit.fixupForwardReferences();
2684 Streamer->emitCompileUnitHeader(CurrentUnit);
2685 if (!CurrentUnit.getOutputUnitDIE())
2687 Streamer->emitDIE(*CurrentUnit.getOutputUnitDIE());
2690 if (!ValidRelocs.empty() && !Options.NoOutput && !Units.empty())
2691 patchFrameInfoForObject(*Obj, DwarfContext,
2692 Units[0].getOrigUnit().getAddressByteSize());
2694 // Clean-up before starting working on the next object.
2698 // Emit everything that's global.
2699 if (!Options.NoOutput) {
2700 Streamer->emitAbbrevs(Abbreviations);
2701 Streamer->emitStrings(StringPool);
2704 return Options.NoOutput ? true : Streamer->finish();
2708 bool linkDwarf(StringRef OutputFilename, const DebugMap &DM,
2709 const LinkOptions &Options) {
2710 DwarfLinker Linker(OutputFilename, Options);
2711 return Linker.link(DM);