1 //===- tools/dsymutil/DwarfLinker.cpp - Dwarf debug info linker -----------===//
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 #include "BinaryHolder.h"
13 #include "llvm/ADT/IntervalMap.h"
14 #include "llvm/ADT/StringMap.h"
15 #include "llvm/ADT/STLExtras.h"
16 #include "llvm/CodeGen/AsmPrinter.h"
17 #include "llvm/CodeGen/DIE.h"
18 #include "llvm/DebugInfo/DWARF/DWARFContext.h"
19 #include "llvm/DebugInfo/DWARF/DWARFDebugInfoEntry.h"
20 #include "llvm/DebugInfo/DWARF/DWARFFormValue.h"
21 #include "llvm/MC/MCAsmBackend.h"
22 #include "llvm/MC/MCAsmInfo.h"
23 #include "llvm/MC/MCContext.h"
24 #include "llvm/MC/MCCodeEmitter.h"
25 #include "llvm/MC/MCDwarf.h"
26 #include "llvm/MC/MCInstrInfo.h"
27 #include "llvm/MC/MCObjectFileInfo.h"
28 #include "llvm/MC/MCRegisterInfo.h"
29 #include "llvm/MC/MCStreamer.h"
30 #include "llvm/Object/MachO.h"
31 #include "llvm/Support/Dwarf.h"
32 #include "llvm/Support/LEB128.h"
33 #include "llvm/Support/TargetRegistry.h"
34 #include "llvm/Target/TargetMachine.h"
35 #include "llvm/Target/TargetOptions.h"
44 void warn(const Twine &Warning, const Twine &Context) {
45 errs() << Twine("while processing ") + Context + ":\n";
46 errs() << Twine("warning: ") + Warning + "\n";
49 bool error(const Twine &Error, const Twine &Context) {
50 errs() << Twine("while processing ") + Context + ":\n";
51 errs() << Twine("error: ") + Error + "\n";
55 template <typename KeyT, typename ValT>
56 using HalfOpenIntervalMap =
57 IntervalMap<KeyT, ValT, IntervalMapImpl::NodeSizer<KeyT, ValT>::LeafSize,
58 IntervalMapHalfOpenInfo<KeyT>>;
60 typedef HalfOpenIntervalMap<uint64_t, int64_t> FunctionIntervals;
62 /// \brief Stores all information relating to a compile unit, be it in
63 /// its original instance in the object file to its brand new cloned
64 /// and linked DIE tree.
67 /// \brief Information gathered about a DIE in the object file.
69 int64_t AddrAdjust; ///< Address offset to apply to the described entity.
70 DIE *Clone; ///< Cloned version of that DIE.
71 uint32_t ParentIdx; ///< The index of this DIE's parent.
72 bool Keep; ///< Is the DIE part of the linked output?
73 bool InDebugMap; ///< Was this DIE's entity found in the map?
76 CompileUnit(DWARFUnit &OrigUnit, unsigned ID)
77 : OrigUnit(OrigUnit), ID(ID), LowPc(UINT64_MAX), HighPc(0), RangeAlloc(),
78 Ranges(RangeAlloc), UnitRangeAttribute(nullptr) {
79 Info.resize(OrigUnit.getNumDIEs());
82 CompileUnit(CompileUnit &&RHS)
83 : OrigUnit(RHS.OrigUnit), Info(std::move(RHS.Info)),
84 CUDie(std::move(RHS.CUDie)), StartOffset(RHS.StartOffset),
85 NextUnitOffset(RHS.NextUnitOffset), RangeAlloc(), Ranges(RangeAlloc) {
86 // The CompileUnit container has been 'reserve()'d with the right
87 // size. We cannot move the IntervalMap anyway.
88 llvm_unreachable("CompileUnits should not be moved.");
91 DWARFUnit &getOrigUnit() const { return OrigUnit; }
93 unsigned getUniqueID() const { return ID; }
95 DIE *getOutputUnitDIE() const { return CUDie.get(); }
96 void setOutputUnitDIE(DIE *Die) { CUDie.reset(Die); }
98 DIEInfo &getInfo(unsigned Idx) { return Info[Idx]; }
99 const DIEInfo &getInfo(unsigned Idx) const { return Info[Idx]; }
101 uint64_t getStartOffset() const { return StartOffset; }
102 uint64_t getNextUnitOffset() const { return NextUnitOffset; }
103 void setStartOffset(uint64_t DebugInfoSize) { StartOffset = DebugInfoSize; }
105 uint64_t getLowPc() const { return LowPc; }
106 uint64_t getHighPc() const { return HighPc; }
108 DIEInteger *getUnitRangesAttribute() const { return UnitRangeAttribute; }
109 const FunctionIntervals &getFunctionRanges() const { return Ranges; }
110 const std::vector<DIEInteger *> &getRangesAttributes() const {
111 return RangeAttributes;
114 const std::vector<std::pair<DIEInteger *, int64_t>> &
115 getLocationAttributes() const {
116 return LocationAttributes;
119 /// \brief Compute the end offset for this unit. Must be
120 /// called after the CU's DIEs have been cloned.
121 /// \returns the next unit offset (which is also the current
122 /// debug_info section size).
123 uint64_t computeNextUnitOffset();
125 /// \brief Keep track of a forward reference to DIE \p Die in \p
126 /// RefUnit by \p Attr. The attribute should be fixed up later to
127 /// point to the absolute offset of \p Die in the debug_info section.
128 void noteForwardReference(DIE *Die, const CompileUnit *RefUnit,
131 /// \brief Apply all fixups recored by noteForwardReference().
132 void fixupForwardReferences();
134 /// \brief Add a function range [\p LowPC, \p HighPC) that is
135 /// relocatad by applying offset \p PCOffset.
136 void addFunctionRange(uint64_t LowPC, uint64_t HighPC, int64_t PCOffset);
138 /// \brief Keep track of a DW_AT_range attribute that we will need to
140 void noteRangeAttribute(const DIE &Die, DIEInteger *Attr);
142 /// \brief Keep track of a location attribute pointing to a location
143 /// list in the debug_loc section.
144 void noteLocationAttribute(DIEInteger *Attr, int64_t PcOffset);
146 /// \brief Add a name accelerator entry for \p Die with \p Name
147 /// which is stored in the string table at \p Offset.
148 void addNameAccelerator(const DIE *Die, const char *Name, uint32_t Offset,
149 bool SkipPubnamesSection = false);
151 /// \brief Add a type accelerator entry for \p Die with \p Name
152 /// which is stored in the string table at \p Offset.
153 void addTypeAccelerator(const DIE *Die, const char *Name, uint32_t Offset);
156 StringRef Name; ///< Name of the entry.
157 const DIE *Die; ///< DIE this entry describes.
158 uint32_t NameOffset; ///< Offset of Name in the string pool.
159 bool SkipPubSection; ///< Emit this entry only in the apple_* sections.
161 AccelInfo(StringRef Name, const DIE *Die, uint32_t NameOffset,
162 bool SkipPubSection = false)
163 : Name(Name), Die(Die), NameOffset(NameOffset),
164 SkipPubSection(SkipPubSection) {}
167 const std::vector<AccelInfo> &getPubnames() const { return Pubnames; }
168 const std::vector<AccelInfo> &getPubtypes() const { return Pubtypes; }
173 std::vector<DIEInfo> Info; ///< DIE info indexed by DIE index.
174 std::unique_ptr<DIE> CUDie; ///< Root of the linked DIE tree.
176 uint64_t StartOffset;
177 uint64_t NextUnitOffset;
182 /// \brief A list of attributes to fixup with the absolute offset of
183 /// a DIE in the debug_info section.
185 /// The offsets for the attributes in this array couldn't be set while
186 /// cloning because for cross-cu forward refences the target DIE's
187 /// offset isn't known you emit the reference attribute.
188 std::vector<std::tuple<DIE *, const CompileUnit *, DIEInteger *>>
189 ForwardDIEReferences;
191 FunctionIntervals::Allocator RangeAlloc;
192 /// \brief The ranges in that interval map are the PC ranges for
193 /// functions in this unit, associated with the PC offset to apply
194 /// to the addresses to get the linked address.
195 FunctionIntervals Ranges;
197 /// \brief DW_AT_ranges attributes to patch after we have gathered
198 /// all the unit's function addresses.
200 std::vector<DIEInteger *> RangeAttributes;
201 DIEInteger *UnitRangeAttribute;
204 /// \brief Location attributes that need to be transfered from th
205 /// original debug_loc section to the liked one. They are stored
206 /// along with the PC offset that is to be applied to their
207 /// function's address.
208 std::vector<std::pair<DIEInteger *, int64_t>> LocationAttributes;
210 /// \brief Accelerator entries for the unit, both for the pub*
211 /// sections and the apple* ones.
213 std::vector<AccelInfo> Pubnames;
214 std::vector<AccelInfo> Pubtypes;
218 uint64_t CompileUnit::computeNextUnitOffset() {
219 NextUnitOffset = StartOffset + 11 /* Header size */;
220 // The root DIE might be null, meaning that the Unit had nothing to
221 // contribute to the linked output. In that case, we will emit the
222 // unit header without any actual DIE.
224 NextUnitOffset += CUDie->getSize();
225 return NextUnitOffset;
228 /// \brief Keep track of a forward cross-cu reference from this unit
229 /// to \p Die that lives in \p RefUnit.
230 void CompileUnit::noteForwardReference(DIE *Die, const CompileUnit *RefUnit,
232 ForwardDIEReferences.emplace_back(Die, RefUnit, Attr);
235 /// \brief Apply all fixups recorded by noteForwardReference().
236 void CompileUnit::fixupForwardReferences() {
237 for (const auto &Ref : ForwardDIEReferences) {
239 const CompileUnit *RefUnit;
241 std::tie(RefDie, RefUnit, Attr) = Ref;
242 Attr->setValue(RefDie->getOffset() + RefUnit->getStartOffset());
246 void CompileUnit::addFunctionRange(uint64_t FuncLowPc, uint64_t FuncHighPc,
248 Ranges.insert(FuncLowPc, FuncHighPc, PcOffset);
249 this->LowPc = std::min(LowPc, FuncLowPc + PcOffset);
250 this->HighPc = std::max(HighPc, FuncHighPc + PcOffset);
253 void CompileUnit::noteRangeAttribute(const DIE &Die, DIEInteger *Attr) {
254 if (Die.getTag() != dwarf::DW_TAG_compile_unit)
255 RangeAttributes.push_back(Attr);
257 UnitRangeAttribute = Attr;
260 void CompileUnit::noteLocationAttribute(DIEInteger *Attr, int64_t PcOffset) {
261 LocationAttributes.emplace_back(Attr, PcOffset);
264 /// \brief Add a name accelerator entry for \p Die with \p Name
265 /// which is stored in the string table at \p Offset.
266 void CompileUnit::addNameAccelerator(const DIE *Die, const char *Name,
267 uint32_t Offset, bool SkipPubSection) {
268 Pubnames.emplace_back(Name, Die, Offset, SkipPubSection);
271 /// \brief Add a type accelerator entry for \p Die with \p Name
272 /// which is stored in the string table at \p Offset.
273 void CompileUnit::addTypeAccelerator(const DIE *Die, const char *Name,
275 Pubtypes.emplace_back(Name, Die, Offset, false);
278 /// \brief A string table that doesn't need relocations.
280 /// We are doing a final link, no need for a string table that
281 /// has relocation entries for every reference to it. This class
282 /// provides this ablitity by just associating offsets with
284 class NonRelocatableStringpool {
286 /// \brief Entries are stored into the StringMap and simply linked
287 /// together through the second element of this pair in order to
288 /// keep track of insertion order.
289 typedef StringMap<std::pair<uint32_t, StringMapEntryBase *>, BumpPtrAllocator>
292 NonRelocatableStringpool()
293 : CurrentEndOffset(0), Sentinel(0), Last(&Sentinel) {
294 // Legacy dsymutil puts an empty string at the start of the line
299 /// \brief Get the offset of string \p S in the string table. This
300 /// can insert a new element or return the offset of a preexisitng
302 uint32_t getStringOffset(StringRef S);
304 /// \brief Get permanent storage for \p S (but do not necessarily
305 /// emit \p S in the output section).
306 /// \returns The StringRef that points to permanent storage to use
307 /// in place of \p S.
308 StringRef internString(StringRef S);
310 // \brief Return the first entry of the string table.
311 const MapTy::MapEntryTy *getFirstEntry() const {
312 return getNextEntry(&Sentinel);
315 // \brief Get the entry following \p E in the string table or null
316 // if \p E was the last entry.
317 const MapTy::MapEntryTy *getNextEntry(const MapTy::MapEntryTy *E) const {
318 return static_cast<const MapTy::MapEntryTy *>(E->getValue().second);
321 uint64_t getSize() { return CurrentEndOffset; }
325 uint32_t CurrentEndOffset;
326 MapTy::MapEntryTy Sentinel, *Last;
329 /// \brief Get the offset of string \p S in the string table. This
330 /// can insert a new element or return the offset of a preexisitng
332 uint32_t NonRelocatableStringpool::getStringOffset(StringRef S) {
333 if (S.empty() && !Strings.empty())
336 std::pair<uint32_t, StringMapEntryBase *> Entry(0, nullptr);
340 // A non-empty string can't be at offset 0, so if we have an entry
341 // with a 0 offset, it must be a previously interned string.
342 std::tie(It, Inserted) = Strings.insert(std::make_pair(S, Entry));
343 if (Inserted || It->getValue().first == 0) {
344 // Set offset and chain at the end of the entries list.
345 It->getValue().first = CurrentEndOffset;
346 CurrentEndOffset += S.size() + 1; // +1 for the '\0'.
347 Last->getValue().second = &*It;
350 return It->getValue().first;
353 /// \brief Put \p S into the StringMap so that it gets permanent
354 /// storage, but do not actually link it in the chain of elements
355 /// that go into the output section. A latter call to
356 /// getStringOffset() with the same string will chain it though.
357 StringRef NonRelocatableStringpool::internString(StringRef S) {
358 std::pair<uint32_t, StringMapEntryBase *> Entry(0, nullptr);
359 auto InsertResult = Strings.insert(std::make_pair(S, Entry));
360 return InsertResult.first->getKey();
363 /// \brief The Dwarf streaming logic
365 /// All interactions with the MC layer that is used to build the debug
366 /// information binary representation are handled in this class.
367 class DwarfStreamer {
368 /// \defgroup MCObjects MC layer objects constructed by the streamer
370 std::unique_ptr<MCRegisterInfo> MRI;
371 std::unique_ptr<MCAsmInfo> MAI;
372 std::unique_ptr<MCObjectFileInfo> MOFI;
373 std::unique_ptr<MCContext> MC;
374 MCAsmBackend *MAB; // Owned by MCStreamer
375 std::unique_ptr<MCInstrInfo> MII;
376 std::unique_ptr<MCSubtargetInfo> MSTI;
377 MCCodeEmitter *MCE; // Owned by MCStreamer
378 MCStreamer *MS; // Owned by AsmPrinter
379 std::unique_ptr<TargetMachine> TM;
380 std::unique_ptr<AsmPrinter> Asm;
383 /// \brief the file we stream the linked Dwarf to.
384 std::unique_ptr<raw_fd_ostream> OutFile;
386 uint32_t RangesSectionSize;
387 uint32_t LocSectionSize;
388 uint32_t LineSectionSize;
390 /// \brief Emit the pubnames or pubtypes section contribution for \p
391 /// Unit into \p Sec. The data is provided in \p Names.
392 void emitPubSectionForUnit(const MCSection *Sec, StringRef Name,
393 const CompileUnit &Unit,
394 const std::vector<CompileUnit::AccelInfo> &Names);
397 /// \brief Actually create the streamer and the ouptut file.
399 /// This could be done directly in the constructor, but it feels
400 /// more natural to handle errors through return value.
401 bool init(Triple TheTriple, StringRef OutputFilename);
403 /// \brief Dump the file to the disk.
406 AsmPrinter &getAsmPrinter() const { return *Asm; }
408 /// \brief Set the current output section to debug_info and change
409 /// the MC Dwarf version to \p DwarfVersion.
410 void switchToDebugInfoSection(unsigned DwarfVersion);
412 /// \brief Emit the compilation unit header for \p Unit in the
413 /// debug_info section.
415 /// As a side effect, this also switches the current Dwarf version
416 /// of the MC layer to the one of U.getOrigUnit().
417 void emitCompileUnitHeader(CompileUnit &Unit);
419 /// \brief Recursively emit the DIE tree rooted at \p Die.
420 void emitDIE(DIE &Die);
422 /// \brief Emit the abbreviation table \p Abbrevs to the
423 /// debug_abbrev section.
424 void emitAbbrevs(const std::vector<DIEAbbrev *> &Abbrevs);
426 /// \brief Emit the string table described by \p Pool.
427 void emitStrings(const NonRelocatableStringpool &Pool);
429 /// \brief Emit debug_ranges for \p FuncRange by translating the
430 /// original \p Entries.
431 void emitRangesEntries(
432 int64_t UnitPcOffset, uint64_t OrigLowPc,
433 FunctionIntervals::const_iterator FuncRange,
434 const std::vector<DWARFDebugRangeList::RangeListEntry> &Entries,
435 unsigned AddressSize);
437 /// \brief Emit debug_aranges entries for \p Unit and if \p
438 /// DoRangesSection is true, also emit the debug_ranges entries for
439 /// the DW_TAG_compile_unit's DW_AT_ranges attribute.
440 void emitUnitRangesEntries(CompileUnit &Unit, bool DoRangesSection);
442 uint32_t getRangesSectionSize() const { return RangesSectionSize; }
444 /// \brief Emit the debug_loc contribution for \p Unit by copying
445 /// the entries from \p Dwarf and offseting them. Update the
446 /// location attributes to point to the new entries.
447 void emitLocationsForUnit(const CompileUnit &Unit, DWARFContext &Dwarf);
449 /// \brief Emit the line table described in \p Rows into the
450 /// debug_line section.
451 void emitLineTableForUnit(StringRef PrologueBytes, unsigned MinInstLength,
452 std::vector<DWARFDebugLine::Row> &Rows,
453 unsigned AdddressSize);
455 uint32_t getLineSectionSize() const { return LineSectionSize; }
457 /// \brief Emit the .debug_pubnames contribution for \p Unit.
458 void emitPubNamesForUnit(const CompileUnit &Unit);
460 /// \brief Emit the .debug_pubtypes contribution for \p Unit.
461 void emitPubTypesForUnit(const CompileUnit &Unit);
464 bool DwarfStreamer::init(Triple TheTriple, StringRef OutputFilename) {
465 std::string ErrorStr;
466 std::string TripleName;
467 StringRef Context = "dwarf streamer init";
470 const Target *TheTarget =
471 TargetRegistry::lookupTarget(TripleName, TheTriple, ErrorStr);
473 return error(ErrorStr, Context);
474 TripleName = TheTriple.getTriple();
476 // Create all the MC Objects.
477 MRI.reset(TheTarget->createMCRegInfo(TripleName));
479 return error(Twine("no register info for target ") + TripleName, Context);
481 MAI.reset(TheTarget->createMCAsmInfo(*MRI, TripleName));
483 return error("no asm info for target " + TripleName, Context);
485 MOFI.reset(new MCObjectFileInfo);
486 MC.reset(new MCContext(MAI.get(), MRI.get(), MOFI.get()));
487 MOFI->InitMCObjectFileInfo(TripleName, Reloc::Default, CodeModel::Default,
490 MAB = TheTarget->createMCAsmBackend(*MRI, TripleName, "");
492 return error("no asm backend for target " + TripleName, Context);
494 MII.reset(TheTarget->createMCInstrInfo());
496 return error("no instr info info for target " + TripleName, Context);
498 MSTI.reset(TheTarget->createMCSubtargetInfo(TripleName, "", ""));
500 return error("no subtarget info for target " + TripleName, Context);
502 MCE = TheTarget->createMCCodeEmitter(*MII, *MRI, *MC);
504 return error("no code emitter for target " + TripleName, Context);
506 // Create the output file.
509 llvm::make_unique<raw_fd_ostream>(OutputFilename, EC, sys::fs::F_None);
511 return error(Twine(OutputFilename) + ": " + EC.message(), Context);
513 MS = TheTarget->createMCObjectStreamer(TheTriple, *MC, *MAB, *OutFile, MCE,
516 return error("no object streamer for target " + TripleName, Context);
518 // Finally create the AsmPrinter we'll use to emit the DIEs.
519 TM.reset(TheTarget->createTargetMachine(TripleName, "", "", TargetOptions()));
521 return error("no target machine for target " + TripleName, Context);
523 Asm.reset(TheTarget->createAsmPrinter(*TM, std::unique_ptr<MCStreamer>(MS)));
525 return error("no asm printer for target " + TripleName, Context);
527 RangesSectionSize = 0;
534 bool DwarfStreamer::finish() {
539 /// \brief Set the current output section to debug_info and change
540 /// the MC Dwarf version to \p DwarfVersion.
541 void DwarfStreamer::switchToDebugInfoSection(unsigned DwarfVersion) {
542 MS->SwitchSection(MOFI->getDwarfInfoSection());
543 MC->setDwarfVersion(DwarfVersion);
546 /// \brief Emit the compilation unit header for \p Unit in the
547 /// debug_info section.
549 /// A Dwarf scetion header is encoded as:
550 /// uint32_t Unit length (omiting this field)
552 /// uint32_t Abbreviation table offset
553 /// uint8_t Address size
555 /// Leading to a total of 11 bytes.
556 void DwarfStreamer::emitCompileUnitHeader(CompileUnit &Unit) {
557 unsigned Version = Unit.getOrigUnit().getVersion();
558 switchToDebugInfoSection(Version);
560 // Emit size of content not including length itself. The size has
561 // already been computed in CompileUnit::computeOffsets(). Substract
562 // 4 to that size to account for the length field.
563 Asm->EmitInt32(Unit.getNextUnitOffset() - Unit.getStartOffset() - 4);
564 Asm->EmitInt16(Version);
565 // We share one abbreviations table across all units so it's always at the
566 // start of the section.
568 Asm->EmitInt8(Unit.getOrigUnit().getAddressByteSize());
571 /// \brief Emit the \p Abbrevs array as the shared abbreviation table
572 /// for the linked Dwarf file.
573 void DwarfStreamer::emitAbbrevs(const std::vector<DIEAbbrev *> &Abbrevs) {
574 MS->SwitchSection(MOFI->getDwarfAbbrevSection());
575 Asm->emitDwarfAbbrevs(Abbrevs);
578 /// \brief Recursively emit the DIE tree rooted at \p Die.
579 void DwarfStreamer::emitDIE(DIE &Die) {
580 MS->SwitchSection(MOFI->getDwarfInfoSection());
581 Asm->emitDwarfDIE(Die);
584 /// \brief Emit the debug_str section stored in \p Pool.
585 void DwarfStreamer::emitStrings(const NonRelocatableStringpool &Pool) {
586 Asm->OutStreamer.SwitchSection(MOFI->getDwarfStrSection());
587 for (auto *Entry = Pool.getFirstEntry(); Entry;
588 Entry = Pool.getNextEntry(Entry))
589 Asm->OutStreamer.EmitBytes(
590 StringRef(Entry->getKey().data(), Entry->getKey().size() + 1));
593 /// \brief Emit the debug_range section contents for \p FuncRange by
594 /// translating the original \p Entries. The debug_range section
595 /// format is totally trivial, consisting just of pairs of address
596 /// sized addresses describing the ranges.
597 void DwarfStreamer::emitRangesEntries(
598 int64_t UnitPcOffset, uint64_t OrigLowPc,
599 FunctionIntervals::const_iterator FuncRange,
600 const std::vector<DWARFDebugRangeList::RangeListEntry> &Entries,
601 unsigned AddressSize) {
602 MS->SwitchSection(MC->getObjectFileInfo()->getDwarfRangesSection());
604 // Offset each range by the right amount.
605 int64_t PcOffset = FuncRange.value() + UnitPcOffset;
606 for (const auto &Range : Entries) {
607 if (Range.isBaseAddressSelectionEntry(AddressSize)) {
608 warn("unsupported base address selection operation",
609 "emitting debug_ranges");
612 // Do not emit empty ranges.
613 if (Range.StartAddress == Range.EndAddress)
616 // All range entries should lie in the function range.
617 if (!(Range.StartAddress + OrigLowPc >= FuncRange.start() &&
618 Range.EndAddress + OrigLowPc <= FuncRange.stop()))
619 warn("inconsistent range data.", "emitting debug_ranges");
620 MS->EmitIntValue(Range.StartAddress + PcOffset, AddressSize);
621 MS->EmitIntValue(Range.EndAddress + PcOffset, AddressSize);
622 RangesSectionSize += 2 * AddressSize;
625 // Add the terminator entry.
626 MS->EmitIntValue(0, AddressSize);
627 MS->EmitIntValue(0, AddressSize);
628 RangesSectionSize += 2 * AddressSize;
631 /// \brief Emit the debug_aranges contribution of a unit and
632 /// if \p DoDebugRanges is true the debug_range contents for a
633 /// compile_unit level DW_AT_ranges attribute (Which are basically the
634 /// same thing with a different base address).
635 /// Just aggregate all the ranges gathered inside that unit.
636 void DwarfStreamer::emitUnitRangesEntries(CompileUnit &Unit,
637 bool DoDebugRanges) {
638 unsigned AddressSize = Unit.getOrigUnit().getAddressByteSize();
639 // Gather the ranges in a vector, so that we can simplify them. The
640 // IntervalMap will have coalesced the non-linked ranges, but here
641 // we want to coalesce the linked addresses.
642 std::vector<std::pair<uint64_t, uint64_t>> Ranges;
643 const auto &FunctionRanges = Unit.getFunctionRanges();
644 for (auto Range = FunctionRanges.begin(), End = FunctionRanges.end();
645 Range != End; ++Range)
646 Ranges.push_back(std::make_pair(Range.start() + Range.value(),
647 Range.stop() + Range.value()));
649 // The object addresses where sorted, but again, the linked
650 // addresses might end up in a different order.
651 std::sort(Ranges.begin(), Ranges.end());
653 if (!Ranges.empty()) {
654 MS->SwitchSection(MC->getObjectFileInfo()->getDwarfARangesSection());
656 MCSymbol *BeginLabel = Asm->createTempSymbol("Barange");
657 MCSymbol *EndLabel = Asm->createTempSymbol("Earange");
659 unsigned HeaderSize =
660 sizeof(int32_t) + // Size of contents (w/o this field
661 sizeof(int16_t) + // DWARF ARange version number
662 sizeof(int32_t) + // Offset of CU in the .debug_info section
663 sizeof(int8_t) + // Pointer Size (in bytes)
664 sizeof(int8_t); // Segment Size (in bytes)
666 unsigned TupleSize = AddressSize * 2;
667 unsigned Padding = OffsetToAlignment(HeaderSize, TupleSize);
669 Asm->EmitLabelDifference(EndLabel, BeginLabel, 4); // Arange length
670 Asm->OutStreamer.EmitLabel(BeginLabel);
671 Asm->EmitInt16(dwarf::DW_ARANGES_VERSION); // Version number
672 Asm->EmitInt32(Unit.getStartOffset()); // Corresponding unit's offset
673 Asm->EmitInt8(AddressSize); // Address size
674 Asm->EmitInt8(0); // Segment size
676 Asm->OutStreamer.EmitFill(Padding, 0x0);
678 for (auto Range = Ranges.begin(), End = Ranges.end(); Range != End;
680 uint64_t RangeStart = Range->first;
681 MS->EmitIntValue(RangeStart, AddressSize);
682 while ((Range + 1) != End && Range->second == (Range + 1)->first)
684 MS->EmitIntValue(Range->second - RangeStart, AddressSize);
688 Asm->OutStreamer.EmitIntValue(0, AddressSize);
689 Asm->OutStreamer.EmitIntValue(0, AddressSize);
690 Asm->OutStreamer.EmitLabel(EndLabel);
696 MS->SwitchSection(MC->getObjectFileInfo()->getDwarfRangesSection());
697 // Offset each range by the right amount.
698 int64_t PcOffset = -Unit.getLowPc();
699 // Emit coalesced ranges.
700 for (auto Range = Ranges.begin(), End = Ranges.end(); Range != End; ++Range) {
701 MS->EmitIntValue(Range->first + PcOffset, AddressSize);
702 while (Range + 1 != End && Range->second == (Range + 1)->first)
704 MS->EmitIntValue(Range->second + PcOffset, AddressSize);
705 RangesSectionSize += 2 * AddressSize;
708 // Add the terminator entry.
709 MS->EmitIntValue(0, AddressSize);
710 MS->EmitIntValue(0, AddressSize);
711 RangesSectionSize += 2 * AddressSize;
714 /// \brief Emit location lists for \p Unit and update attribtues to
715 /// point to the new entries.
716 void DwarfStreamer::emitLocationsForUnit(const CompileUnit &Unit,
717 DWARFContext &Dwarf) {
718 const std::vector<std::pair<DIEInteger *, int64_t>> &Attributes =
719 Unit.getLocationAttributes();
721 if (Attributes.empty())
724 MS->SwitchSection(MC->getObjectFileInfo()->getDwarfLocSection());
726 unsigned AddressSize = Unit.getOrigUnit().getAddressByteSize();
727 const DWARFSection &InputSec = Dwarf.getLocSection();
728 DataExtractor Data(InputSec.Data, Dwarf.isLittleEndian(), AddressSize);
729 DWARFUnit &OrigUnit = Unit.getOrigUnit();
730 const auto *OrigUnitDie = OrigUnit.getCompileUnitDIE(false);
731 int64_t UnitPcOffset = 0;
732 uint64_t OrigLowPc = OrigUnitDie->getAttributeValueAsAddress(
733 &OrigUnit, dwarf::DW_AT_low_pc, -1ULL);
734 if (OrigLowPc != -1ULL)
735 UnitPcOffset = int64_t(OrigLowPc) - Unit.getLowPc();
737 for (const auto &Attr : Attributes) {
738 uint32_t Offset = Attr.first->getValue();
739 Attr.first->setValue(LocSectionSize);
740 // This is the quantity to add to the old location address to get
741 // the correct address for the new one.
742 int64_t LocPcOffset = Attr.second + UnitPcOffset;
743 while (Data.isValidOffset(Offset)) {
744 uint64_t Low = Data.getUnsigned(&Offset, AddressSize);
745 uint64_t High = Data.getUnsigned(&Offset, AddressSize);
746 LocSectionSize += 2 * AddressSize;
747 if (Low == 0 && High == 0) {
748 Asm->OutStreamer.EmitIntValue(0, AddressSize);
749 Asm->OutStreamer.EmitIntValue(0, AddressSize);
752 Asm->OutStreamer.EmitIntValue(Low + LocPcOffset, AddressSize);
753 Asm->OutStreamer.EmitIntValue(High + LocPcOffset, AddressSize);
754 uint64_t Length = Data.getU16(&Offset);
755 Asm->OutStreamer.EmitIntValue(Length, 2);
756 // Just copy the bytes over.
757 Asm->OutStreamer.EmitBytes(
758 StringRef(InputSec.Data.substr(Offset, Length)));
760 LocSectionSize += Length + 2;
765 void DwarfStreamer::emitLineTableForUnit(StringRef PrologueBytes,
766 unsigned MinInstLength,
767 std::vector<DWARFDebugLine::Row> &Rows,
768 unsigned PointerSize) {
769 // Switch to the section where the table will be emitted into.
770 MS->SwitchSection(MC->getObjectFileInfo()->getDwarfLineSection());
771 MCSymbol *LineStartSym = MC->CreateTempSymbol();
772 MCSymbol *LineEndSym = MC->CreateTempSymbol();
774 // The first 4 bytes is the total length of the information for this
775 // compilation unit (not including these 4 bytes for the length).
776 Asm->EmitLabelDifference(LineEndSym, LineStartSym, 4);
777 Asm->OutStreamer.EmitLabel(LineStartSym);
779 MS->EmitBytes(PrologueBytes);
780 LineSectionSize += PrologueBytes.size() + 4;
782 SmallString<128> EncodingBuffer;
783 raw_svector_ostream EncodingOS(EncodingBuffer);
786 // We only have the dummy entry, dsymutil emits an entry with a 0
787 // address in that case.
788 MCDwarfLineAddr::Encode(*MC, INT64_MAX, 0, EncodingOS);
789 MS->EmitBytes(EncodingOS.str());
790 LineSectionSize += EncodingBuffer.size();
791 MS->EmitLabel(LineEndSym);
795 // Line table state machine fields
796 unsigned FileNum = 1;
797 unsigned LastLine = 1;
799 unsigned IsStatement = 1;
801 uint64_t Address = -1ULL;
803 unsigned RowsSinceLastSequence = 0;
805 for (unsigned Idx = 0; Idx < Rows.size(); ++Idx) {
806 auto &Row = Rows[Idx];
808 int64_t AddressDelta;
809 if (Address == -1ULL) {
810 MS->EmitIntValue(dwarf::DW_LNS_extended_op, 1);
811 MS->EmitULEB128IntValue(PointerSize + 1);
812 MS->EmitIntValue(dwarf::DW_LNE_set_address, 1);
813 MS->EmitIntValue(Row.Address, PointerSize);
814 LineSectionSize += 2 + PointerSize + getULEB128Size(PointerSize + 1);
817 AddressDelta = (Row.Address - Address) / MinInstLength;
820 // FIXME: code copied and transfromed from
821 // MCDwarf.cpp::EmitDwarfLineTable. We should find a way to share
822 // this code, but the current compatibility requirement with
823 // classic dsymutil makes it hard. Revisit that once this
824 // requirement is dropped.
826 if (FileNum != Row.File) {
828 MS->EmitIntValue(dwarf::DW_LNS_set_file, 1);
829 MS->EmitULEB128IntValue(FileNum);
830 LineSectionSize += 1 + getULEB128Size(FileNum);
832 if (Column != Row.Column) {
834 MS->EmitIntValue(dwarf::DW_LNS_set_column, 1);
835 MS->EmitULEB128IntValue(Column);
836 LineSectionSize += 1 + getULEB128Size(Column);
839 // FIXME: We should handle the discriminator here, but dsymutil
840 // doesn' consider it, thus ignore it for now.
842 if (Isa != Row.Isa) {
844 MS->EmitIntValue(dwarf::DW_LNS_set_isa, 1);
845 MS->EmitULEB128IntValue(Isa);
846 LineSectionSize += 1 + getULEB128Size(Isa);
848 if (IsStatement != Row.IsStmt) {
849 IsStatement = Row.IsStmt;
850 MS->EmitIntValue(dwarf::DW_LNS_negate_stmt, 1);
851 LineSectionSize += 1;
853 if (Row.BasicBlock) {
854 MS->EmitIntValue(dwarf::DW_LNS_set_basic_block, 1);
855 LineSectionSize += 1;
858 if (Row.PrologueEnd) {
859 MS->EmitIntValue(dwarf::DW_LNS_set_prologue_end, 1);
860 LineSectionSize += 1;
863 if (Row.EpilogueBegin) {
864 MS->EmitIntValue(dwarf::DW_LNS_set_epilogue_begin, 1);
865 LineSectionSize += 1;
868 int64_t LineDelta = int64_t(Row.Line) - LastLine;
869 if (!Row.EndSequence) {
870 MCDwarfLineAddr::Encode(*MC, LineDelta, AddressDelta, EncodingOS);
871 MS->EmitBytes(EncodingOS.str());
872 LineSectionSize += EncodingBuffer.size();
873 EncodingBuffer.resize(0);
875 Address = Row.Address;
877 RowsSinceLastSequence++;
880 MS->EmitIntValue(dwarf::DW_LNS_advance_line, 1);
881 MS->EmitSLEB128IntValue(LineDelta);
882 LineSectionSize += 1 + getSLEB128Size(LineDelta);
885 MS->EmitIntValue(dwarf::DW_LNS_advance_pc, 1);
886 MS->EmitULEB128IntValue(AddressDelta);
887 LineSectionSize += 1 + getULEB128Size(AddressDelta);
889 MCDwarfLineAddr::Encode(*MC, INT64_MAX, 0, EncodingOS);
890 MS->EmitBytes(EncodingOS.str());
891 LineSectionSize += EncodingBuffer.size();
892 EncodingBuffer.resize(0);
895 LastLine = FileNum = IsStatement = 1;
896 RowsSinceLastSequence = Column = Isa = 0;
900 if (RowsSinceLastSequence) {
901 MCDwarfLineAddr::Encode(*MC, INT64_MAX, 0, EncodingOS);
902 MS->EmitBytes(EncodingOS.str());
903 LineSectionSize += EncodingBuffer.size();
904 EncodingBuffer.resize(0);
908 MS->EmitLabel(LineEndSym);
911 /// \brief Emit the pubnames or pubtypes section contribution for \p
912 /// Unit into \p Sec. The data is provided in \p Names.
913 void DwarfStreamer::emitPubSectionForUnit(
914 const MCSection *Sec, StringRef SecName, const CompileUnit &Unit,
915 const std::vector<CompileUnit::AccelInfo> &Names) {
919 // Start the dwarf pubnames section.
920 Asm->OutStreamer.SwitchSection(Sec);
921 MCSymbol *BeginLabel = Asm->createTempSymbol("pub" + SecName + "_begin");
922 MCSymbol *EndLabel = Asm->createTempSymbol("pub" + SecName + "_end");
924 bool HeaderEmitted = false;
925 // Emit the pubnames for this compilation unit.
926 for (const auto &Name : Names) {
927 if (Name.SkipPubSection)
930 if (!HeaderEmitted) {
932 Asm->EmitLabelDifference(EndLabel, BeginLabel, 4); // Length
933 Asm->OutStreamer.EmitLabel(BeginLabel);
934 Asm->EmitInt16(dwarf::DW_PUBNAMES_VERSION); // Version
935 Asm->EmitInt32(Unit.getStartOffset()); // Unit offset
936 Asm->EmitInt32(Unit.getNextUnitOffset() - Unit.getStartOffset()); // Size
937 HeaderEmitted = true;
939 Asm->EmitInt32(Name.Die->getOffset());
940 Asm->OutStreamer.EmitBytes(
941 StringRef(Name.Name.data(), Name.Name.size() + 1));
946 Asm->EmitInt32(0); // End marker.
947 Asm->OutStreamer.EmitLabel(EndLabel);
950 /// \brief Emit .debug_pubnames for \p Unit.
951 void DwarfStreamer::emitPubNamesForUnit(const CompileUnit &Unit) {
952 emitPubSectionForUnit(MC->getObjectFileInfo()->getDwarfPubNamesSection(),
953 "names", Unit, Unit.getPubnames());
956 /// \brief Emit .debug_pubtypes for \p Unit.
957 void DwarfStreamer::emitPubTypesForUnit(const CompileUnit &Unit) {
958 emitPubSectionForUnit(MC->getObjectFileInfo()->getDwarfPubTypesSection(),
959 "types", Unit, Unit.getPubtypes());
962 /// \brief The core of the Dwarf linking logic.
964 /// The link of the dwarf information from the object files will be
965 /// driven by the selection of 'root DIEs', which are DIEs that
966 /// describe variables or functions that are present in the linked
967 /// binary (and thus have entries in the debug map). All the debug
968 /// information that will be linked (the DIEs, but also the line
969 /// tables, ranges, ...) is derived from that set of root DIEs.
971 /// The root DIEs are identified because they contain relocations that
972 /// correspond to a debug map entry at specific places (the low_pc for
973 /// a function, the location for a variable). These relocations are
974 /// called ValidRelocs in the DwarfLinker and are gathered as a very
975 /// first step when we start processing a DebugMapObject.
978 DwarfLinker(StringRef OutputFilename, const LinkOptions &Options)
979 : OutputFilename(OutputFilename), Options(Options),
980 BinHolder(Options.Verbose) {}
983 for (auto *Abbrev : Abbreviations)
987 /// \brief Link the contents of the DebugMap.
988 bool link(const DebugMap &);
991 /// \brief Called at the start of a debug object link.
992 void startDebugObject(DWARFContext &, DebugMapObject &);
994 /// \brief Called at the end of a debug object link.
995 void endDebugObject();
997 /// \defgroup FindValidRelocations Translate debug map into a list
998 /// of relevant relocations
1005 const DebugMapObject::DebugMapEntry *Mapping;
1007 ValidReloc(uint32_t Offset, uint32_t Size, uint64_t Addend,
1008 const DebugMapObject::DebugMapEntry *Mapping)
1009 : Offset(Offset), Size(Size), Addend(Addend), Mapping(Mapping) {}
1011 bool operator<(const ValidReloc &RHS) const { return Offset < RHS.Offset; }
1014 /// \brief The valid relocations for the current DebugMapObject.
1015 /// This vector is sorted by relocation offset.
1016 std::vector<ValidReloc> ValidRelocs;
1018 /// \brief Index into ValidRelocs of the next relocation to
1019 /// consider. As we walk the DIEs in acsending file offset and as
1020 /// ValidRelocs is sorted by file offset, keeping this index
1021 /// uptodate is all we have to do to have a cheap lookup during the
1022 /// root DIE selection and during DIE cloning.
1023 unsigned NextValidReloc;
1025 bool findValidRelocsInDebugInfo(const object::ObjectFile &Obj,
1026 const DebugMapObject &DMO);
1028 bool findValidRelocs(const object::SectionRef &Section,
1029 const object::ObjectFile &Obj,
1030 const DebugMapObject &DMO);
1032 void findValidRelocsMachO(const object::SectionRef &Section,
1033 const object::MachOObjectFile &Obj,
1034 const DebugMapObject &DMO);
1037 /// \defgroup FindRootDIEs Find DIEs corresponding to debug map entries.
1040 /// \brief Recursively walk the \p DIE tree and look for DIEs to
1041 /// keep. Store that information in \p CU's DIEInfo.
1042 void lookForDIEsToKeep(const DWARFDebugInfoEntryMinimal &DIE,
1043 const DebugMapObject &DMO, CompileUnit &CU,
1046 /// \brief Flags passed to DwarfLinker::lookForDIEsToKeep
1047 enum TravesalFlags {
1048 TF_Keep = 1 << 0, ///< Mark the traversed DIEs as kept.
1049 TF_InFunctionScope = 1 << 1, ///< Current scope is a fucntion scope.
1050 TF_DependencyWalk = 1 << 2, ///< Walking the dependencies of a kept DIE.
1051 TF_ParentWalk = 1 << 3, ///< Walking up the parents of a kept DIE.
1054 /// \brief Mark the passed DIE as well as all the ones it depends on
1056 void keepDIEAndDenpendencies(const DWARFDebugInfoEntryMinimal &DIE,
1057 CompileUnit::DIEInfo &MyInfo,
1058 const DebugMapObject &DMO, CompileUnit &CU,
1061 unsigned shouldKeepDIE(const DWARFDebugInfoEntryMinimal &DIE,
1062 CompileUnit &Unit, CompileUnit::DIEInfo &MyInfo,
1065 unsigned shouldKeepVariableDIE(const DWARFDebugInfoEntryMinimal &DIE,
1067 CompileUnit::DIEInfo &MyInfo, unsigned Flags);
1069 unsigned shouldKeepSubprogramDIE(const DWARFDebugInfoEntryMinimal &DIE,
1071 CompileUnit::DIEInfo &MyInfo,
1074 bool hasValidRelocation(uint32_t StartOffset, uint32_t EndOffset,
1075 CompileUnit::DIEInfo &Info);
1078 /// \defgroup Linking Methods used to link the debug information
1081 /// \brief Recursively clone \p InputDIE into an tree of DIE objects
1082 /// where useless (as decided by lookForDIEsToKeep()) bits have been
1083 /// stripped out and addresses have been rewritten according to the
1086 /// \param OutOffset is the offset the cloned DIE in the output
1088 /// \param PCOffset (while cloning a function scope) is the offset
1089 /// applied to the entry point of the function to get the linked address.
1091 /// \returns the root of the cloned tree.
1092 DIE *cloneDIE(const DWARFDebugInfoEntryMinimal &InputDIE, CompileUnit &U,
1093 int64_t PCOffset, uint32_t OutOffset);
1095 typedef DWARFAbbreviationDeclaration::AttributeSpec AttributeSpec;
1097 /// \brief Information gathered and exchanged between the various
1098 /// clone*Attributes helpers about the attributes of a particular DIE.
1099 struct AttributesInfo {
1100 const char *Name, *MangledName; ///< Names.
1101 uint32_t NameOffset, MangledNameOffset; ///< Offsets in the string pool.
1103 uint64_t OrigHighPc; ///< Value of AT_high_pc in the input DIE
1104 int64_t PCOffset; ///< Offset to apply to PC addresses inside a function.
1106 bool HasLowPc; ///< Does the DIE have a low_pc attribute?
1107 bool IsDeclaration; ///< Is this DIE only a declaration?
1110 : Name(nullptr), MangledName(nullptr), NameOffset(0),
1111 MangledNameOffset(0), OrigHighPc(0), PCOffset(0), HasLowPc(false),
1112 IsDeclaration(false) {}
1115 /// \brief Helper for cloneDIE.
1116 unsigned cloneAttribute(DIE &Die, const DWARFDebugInfoEntryMinimal &InputDIE,
1117 CompileUnit &U, const DWARFFormValue &Val,
1118 const AttributeSpec AttrSpec, unsigned AttrSize,
1119 AttributesInfo &AttrInfo);
1121 /// \brief Helper for cloneDIE.
1122 unsigned cloneStringAttribute(DIE &Die, AttributeSpec AttrSpec,
1123 const DWARFFormValue &Val, const DWARFUnit &U);
1125 /// \brief Helper for cloneDIE.
1127 cloneDieReferenceAttribute(DIE &Die,
1128 const DWARFDebugInfoEntryMinimal &InputDIE,
1129 AttributeSpec AttrSpec, unsigned AttrSize,
1130 const DWARFFormValue &Val, CompileUnit &Unit);
1132 /// \brief Helper for cloneDIE.
1133 unsigned cloneBlockAttribute(DIE &Die, AttributeSpec AttrSpec,
1134 const DWARFFormValue &Val, unsigned AttrSize);
1136 /// \brief Helper for cloneDIE.
1137 unsigned cloneAddressAttribute(DIE &Die, AttributeSpec AttrSpec,
1138 const DWARFFormValue &Val,
1139 const CompileUnit &Unit, AttributesInfo &Info);
1141 /// \brief Helper for cloneDIE.
1142 unsigned cloneScalarAttribute(DIE &Die,
1143 const DWARFDebugInfoEntryMinimal &InputDIE,
1144 CompileUnit &U, AttributeSpec AttrSpec,
1145 const DWARFFormValue &Val, unsigned AttrSize,
1146 AttributesInfo &Info);
1148 /// \brief Helper for cloneDIE.
1149 bool applyValidRelocs(MutableArrayRef<char> Data, uint32_t BaseOffset,
1150 bool isLittleEndian);
1152 /// \brief Assign an abbreviation number to \p Abbrev
1153 void AssignAbbrev(DIEAbbrev &Abbrev);
1155 /// \brief FoldingSet that uniques the abbreviations.
1156 FoldingSet<DIEAbbrev> AbbreviationsSet;
1157 /// \brief Storage for the unique Abbreviations.
1158 /// This is passed to AsmPrinter::emitDwarfAbbrevs(), thus it cannot
1159 /// be changed to a vecot of unique_ptrs.
1160 std::vector<DIEAbbrev *> Abbreviations;
1162 /// \brief Compute and emit debug_ranges section for \p Unit, and
1163 /// patch the attributes referencing it.
1164 void patchRangesForUnit(const CompileUnit &Unit, DWARFContext &Dwarf) const;
1166 /// \brief Generate and emit the DW_AT_ranges attribute for a
1167 /// compile_unit if it had one.
1168 void generateUnitRanges(CompileUnit &Unit) const;
1170 /// \brief Extract the line tables fromt he original dwarf, extract
1171 /// the relevant parts according to the linked function ranges and
1172 /// emit the result in the debug_line section.
1173 void patchLineTableForUnit(CompileUnit &Unit, DWARFContext &OrigDwarf);
1175 /// \brief Emit the accelerator entries for \p Unit.
1176 void emitAcceleratorEntriesForUnit(CompileUnit &Unit);
1178 /// \brief DIELoc objects that need to be destructed (but not freed!).
1179 std::vector<DIELoc *> DIELocs;
1180 /// \brief DIEBlock objects that need to be destructed (but not freed!).
1181 std::vector<DIEBlock *> DIEBlocks;
1182 /// \brief Allocator used for all the DIEValue objects.
1183 BumpPtrAllocator DIEAlloc;
1186 /// \defgroup Helpers Various helper methods.
1189 const DWARFDebugInfoEntryMinimal *
1190 resolveDIEReference(DWARFFormValue &RefValue, const DWARFUnit &Unit,
1191 const DWARFDebugInfoEntryMinimal &DIE,
1192 CompileUnit *&ReferencedCU);
1194 CompileUnit *getUnitForOffset(unsigned Offset);
1196 bool getDIENames(const DWARFDebugInfoEntryMinimal &Die, DWARFUnit &U,
1197 AttributesInfo &Info);
1199 void reportWarning(const Twine &Warning, const DWARFUnit *Unit = nullptr,
1200 const DWARFDebugInfoEntryMinimal *DIE = nullptr) const;
1202 bool createStreamer(Triple TheTriple, StringRef OutputFilename);
1206 std::string OutputFilename;
1207 LinkOptions Options;
1208 BinaryHolder BinHolder;
1209 std::unique_ptr<DwarfStreamer> Streamer;
1211 /// The units of the current debug map object.
1212 std::vector<CompileUnit> Units;
1214 /// The debug map object curently under consideration.
1215 DebugMapObject *CurrentDebugObject;
1217 /// \brief The Dwarf string pool
1218 NonRelocatableStringpool StringPool;
1220 /// \brief This map is keyed by the entry PC of functions in that
1221 /// debug object and the associated value is a pair storing the
1222 /// corresponding end PC and the offset to apply to get the linked
1225 /// See startDebugObject() for a more complete description of its use.
1226 std::map<uint64_t, std::pair<uint64_t, int64_t>> Ranges;
1229 /// \brief Similar to DWARFUnitSection::getUnitForOffset(), but
1230 /// returning our CompileUnit object instead.
1231 CompileUnit *DwarfLinker::getUnitForOffset(unsigned Offset) {
1233 std::upper_bound(Units.begin(), Units.end(), Offset,
1234 [](uint32_t LHS, const CompileUnit &RHS) {
1235 return LHS < RHS.getOrigUnit().getNextUnitOffset();
1237 return CU != Units.end() ? &*CU : nullptr;
1240 /// \brief Resolve the DIE attribute reference that has been
1241 /// extracted in \p RefValue. The resulting DIE migh be in another
1242 /// CompileUnit which is stored into \p ReferencedCU.
1243 /// \returns null if resolving fails for any reason.
1244 const DWARFDebugInfoEntryMinimal *DwarfLinker::resolveDIEReference(
1245 DWARFFormValue &RefValue, const DWARFUnit &Unit,
1246 const DWARFDebugInfoEntryMinimal &DIE, CompileUnit *&RefCU) {
1247 assert(RefValue.isFormClass(DWARFFormValue::FC_Reference));
1248 uint64_t RefOffset = *RefValue.getAsReference(&Unit);
1250 if ((RefCU = getUnitForOffset(RefOffset)))
1251 if (const auto *RefDie = RefCU->getOrigUnit().getDIEForOffset(RefOffset))
1254 reportWarning("could not find referenced DIE", &Unit, &DIE);
1258 /// \brief Get the potential name and mangled name for the entity
1259 /// described by \p Die and store them in \Info if they are not
1261 /// \returns is a name was found.
1262 bool DwarfLinker::getDIENames(const DWARFDebugInfoEntryMinimal &Die,
1263 DWARFUnit &U, AttributesInfo &Info) {
1264 // FIXME: a bit wastefull as the first getName might return the
1266 if (!Info.MangledName &&
1267 (Info.MangledName = Die.getName(&U, DINameKind::LinkageName)))
1268 Info.MangledNameOffset = StringPool.getStringOffset(Info.MangledName);
1270 if (!Info.Name && (Info.Name = Die.getName(&U, DINameKind::ShortName)))
1271 Info.NameOffset = StringPool.getStringOffset(Info.Name);
1273 return Info.Name || Info.MangledName;
1276 /// \brief Report a warning to the user, optionaly including
1277 /// information about a specific \p DIE related to the warning.
1278 void DwarfLinker::reportWarning(const Twine &Warning, const DWARFUnit *Unit,
1279 const DWARFDebugInfoEntryMinimal *DIE) const {
1280 StringRef Context = "<debug map>";
1281 if (CurrentDebugObject)
1282 Context = CurrentDebugObject->getObjectFilename();
1283 warn(Warning, Context);
1285 if (!Options.Verbose || !DIE)
1288 errs() << " in DIE:\n";
1289 DIE->dump(errs(), const_cast<DWARFUnit *>(Unit), 0 /* RecurseDepth */,
1293 bool DwarfLinker::createStreamer(Triple TheTriple, StringRef OutputFilename) {
1294 if (Options.NoOutput)
1297 Streamer = llvm::make_unique<DwarfStreamer>();
1298 return Streamer->init(TheTriple, OutputFilename);
1301 /// \brief Recursive helper to gather the child->parent relationships in the
1302 /// original compile unit.
1303 static void gatherDIEParents(const DWARFDebugInfoEntryMinimal *DIE,
1304 unsigned ParentIdx, CompileUnit &CU) {
1305 unsigned MyIdx = CU.getOrigUnit().getDIEIndex(DIE);
1306 CU.getInfo(MyIdx).ParentIdx = ParentIdx;
1308 if (DIE->hasChildren())
1309 for (auto *Child = DIE->getFirstChild(); Child && !Child->isNULL();
1310 Child = Child->getSibling())
1311 gatherDIEParents(Child, MyIdx, CU);
1314 static bool dieNeedsChildrenToBeMeaningful(uint32_t Tag) {
1318 case dwarf::DW_TAG_subprogram:
1319 case dwarf::DW_TAG_lexical_block:
1320 case dwarf::DW_TAG_subroutine_type:
1321 case dwarf::DW_TAG_structure_type:
1322 case dwarf::DW_TAG_class_type:
1323 case dwarf::DW_TAG_union_type:
1326 llvm_unreachable("Invalid Tag");
1329 void DwarfLinker::startDebugObject(DWARFContext &Dwarf, DebugMapObject &Obj) {
1330 Units.reserve(Dwarf.getNumCompileUnits());
1332 // Iterate over the debug map entries and put all the ones that are
1333 // functions (because they have a size) into the Ranges map. This
1334 // map is very similar to the FunctionRanges that are stored in each
1335 // unit, with 2 notable differences:
1336 // - obviously this one is global, while the other ones are per-unit.
1337 // - this one contains not only the functions described in the DIE
1338 // tree, but also the ones that are only in the debug map.
1339 // The latter information is required to reproduce dsymutil's logic
1340 // while linking line tables. The cases where this information
1341 // matters look like bugs that need to be investigated, but for now
1342 // we need to reproduce dsymutil's behavior.
1343 // FIXME: Once we understood exactly if that information is needed,
1344 // maybe totally remove this (or try to use it to do a real
1345 // -gline-tables-only on Darwin.
1346 for (const auto &Entry : Obj.symbols()) {
1347 const auto &Mapping = Entry.getValue();
1349 Ranges[Mapping.ObjectAddress] = std::make_pair(
1350 Mapping.ObjectAddress + Mapping.Size,
1351 int64_t(Mapping.BinaryAddress) - Mapping.ObjectAddress);
1355 void DwarfLinker::endDebugObject() {
1357 ValidRelocs.clear();
1360 for (auto *Block : DIEBlocks)
1362 for (auto *Loc : DIELocs)
1370 /// \brief Iterate over the relocations of the given \p Section and
1371 /// store the ones that correspond to debug map entries into the
1372 /// ValidRelocs array.
1373 void DwarfLinker::findValidRelocsMachO(const object::SectionRef &Section,
1374 const object::MachOObjectFile &Obj,
1375 const DebugMapObject &DMO) {
1377 Section.getContents(Contents);
1378 DataExtractor Data(Contents, Obj.isLittleEndian(), 0);
1380 for (const object::RelocationRef &Reloc : Section.relocations()) {
1381 object::DataRefImpl RelocDataRef = Reloc.getRawDataRefImpl();
1382 MachO::any_relocation_info MachOReloc = Obj.getRelocation(RelocDataRef);
1383 unsigned RelocSize = 1 << Obj.getAnyRelocationLength(MachOReloc);
1385 if ((RelocSize != 4 && RelocSize != 8) || Reloc.getOffset(Offset64)) {
1386 reportWarning(" unsupported relocation in debug_info section.");
1389 uint32_t Offset = Offset64;
1390 // Mach-o uses REL relocations, the addend is at the relocation offset.
1391 uint64_t Addend = Data.getUnsigned(&Offset, RelocSize);
1393 auto Sym = Reloc.getSymbol();
1394 if (Sym != Obj.symbol_end()) {
1395 StringRef SymbolName;
1396 if (Sym->getName(SymbolName)) {
1397 reportWarning("error getting relocation symbol name.");
1400 if (const auto *Mapping = DMO.lookupSymbol(SymbolName))
1401 ValidRelocs.emplace_back(Offset64, RelocSize, Addend, Mapping);
1402 } else if (const auto *Mapping = DMO.lookupObjectAddress(Addend)) {
1403 // Do not store the addend. The addend was the address of the
1404 // symbol in the object file, the address in the binary that is
1405 // stored in the debug map doesn't need to be offseted.
1406 ValidRelocs.emplace_back(Offset64, RelocSize, 0, Mapping);
1411 /// \brief Dispatch the valid relocation finding logic to the
1412 /// appropriate handler depending on the object file format.
1413 bool DwarfLinker::findValidRelocs(const object::SectionRef &Section,
1414 const object::ObjectFile &Obj,
1415 const DebugMapObject &DMO) {
1416 // Dispatch to the right handler depending on the file type.
1417 if (auto *MachOObj = dyn_cast<object::MachOObjectFile>(&Obj))
1418 findValidRelocsMachO(Section, *MachOObj, DMO);
1420 reportWarning(Twine("unsupported object file type: ") + Obj.getFileName());
1422 if (ValidRelocs.empty())
1425 // Sort the relocations by offset. We will walk the DIEs linearly in
1426 // the file, this allows us to just keep an index in the relocation
1427 // array that we advance during our walk, rather than resorting to
1428 // some associative container. See DwarfLinker::NextValidReloc.
1429 std::sort(ValidRelocs.begin(), ValidRelocs.end());
1433 /// \brief Look for relocations in the debug_info section that match
1434 /// entries in the debug map. These relocations will drive the Dwarf
1435 /// link by indicating which DIEs refer to symbols present in the
1437 /// \returns wether there are any valid relocations in the debug info.
1438 bool DwarfLinker::findValidRelocsInDebugInfo(const object::ObjectFile &Obj,
1439 const DebugMapObject &DMO) {
1440 // Find the debug_info section.
1441 for (const object::SectionRef &Section : Obj.sections()) {
1442 StringRef SectionName;
1443 Section.getName(SectionName);
1444 SectionName = SectionName.substr(SectionName.find_first_not_of("._"));
1445 if (SectionName != "debug_info")
1447 return findValidRelocs(Section, Obj, DMO);
1452 /// \brief Checks that there is a relocation against an actual debug
1453 /// map entry between \p StartOffset and \p NextOffset.
1455 /// This function must be called with offsets in strictly ascending
1456 /// order because it never looks back at relocations it already 'went past'.
1457 /// \returns true and sets Info.InDebugMap if it is the case.
1458 bool DwarfLinker::hasValidRelocation(uint32_t StartOffset, uint32_t EndOffset,
1459 CompileUnit::DIEInfo &Info) {
1460 assert(NextValidReloc == 0 ||
1461 StartOffset > ValidRelocs[NextValidReloc - 1].Offset);
1462 if (NextValidReloc >= ValidRelocs.size())
1465 uint64_t RelocOffset = ValidRelocs[NextValidReloc].Offset;
1467 // We might need to skip some relocs that we didn't consider. For
1468 // example the high_pc of a discarded DIE might contain a reloc that
1469 // is in the list because it actually corresponds to the start of a
1470 // function that is in the debug map.
1471 while (RelocOffset < StartOffset && NextValidReloc < ValidRelocs.size() - 1)
1472 RelocOffset = ValidRelocs[++NextValidReloc].Offset;
1474 if (RelocOffset < StartOffset || RelocOffset >= EndOffset)
1477 const auto &ValidReloc = ValidRelocs[NextValidReloc++];
1478 if (Options.Verbose)
1479 outs() << "Found valid debug map entry: " << ValidReloc.Mapping->getKey()
1480 << " " << format("\t%016" PRIx64 " => %016" PRIx64,
1481 ValidReloc.Mapping->getValue().ObjectAddress,
1482 ValidReloc.Mapping->getValue().BinaryAddress);
1484 Info.AddrAdjust = int64_t(ValidReloc.Mapping->getValue().BinaryAddress) +
1486 ValidReloc.Mapping->getValue().ObjectAddress;
1487 Info.InDebugMap = true;
1491 /// \brief Get the starting and ending (exclusive) offset for the
1492 /// attribute with index \p Idx descibed by \p Abbrev. \p Offset is
1493 /// supposed to point to the position of the first attribute described
1495 /// \return [StartOffset, EndOffset) as a pair.
1496 static std::pair<uint32_t, uint32_t>
1497 getAttributeOffsets(const DWARFAbbreviationDeclaration *Abbrev, unsigned Idx,
1498 unsigned Offset, const DWARFUnit &Unit) {
1499 DataExtractor Data = Unit.getDebugInfoExtractor();
1501 for (unsigned i = 0; i < Idx; ++i)
1502 DWARFFormValue::skipValue(Abbrev->getFormByIndex(i), Data, &Offset, &Unit);
1504 uint32_t End = Offset;
1505 DWARFFormValue::skipValue(Abbrev->getFormByIndex(Idx), Data, &End, &Unit);
1507 return std::make_pair(Offset, End);
1510 /// \brief Check if a variable describing DIE should be kept.
1511 /// \returns updated TraversalFlags.
1512 unsigned DwarfLinker::shouldKeepVariableDIE(
1513 const DWARFDebugInfoEntryMinimal &DIE, CompileUnit &Unit,
1514 CompileUnit::DIEInfo &MyInfo, unsigned Flags) {
1515 const auto *Abbrev = DIE.getAbbreviationDeclarationPtr();
1517 // Global variables with constant value can always be kept.
1518 if (!(Flags & TF_InFunctionScope) &&
1519 Abbrev->findAttributeIndex(dwarf::DW_AT_const_value) != -1U) {
1520 MyInfo.InDebugMap = true;
1521 return Flags | TF_Keep;
1524 uint32_t LocationIdx = Abbrev->findAttributeIndex(dwarf::DW_AT_location);
1525 if (LocationIdx == -1U)
1528 uint32_t Offset = DIE.getOffset() + getULEB128Size(Abbrev->getCode());
1529 const DWARFUnit &OrigUnit = Unit.getOrigUnit();
1530 uint32_t LocationOffset, LocationEndOffset;
1531 std::tie(LocationOffset, LocationEndOffset) =
1532 getAttributeOffsets(Abbrev, LocationIdx, Offset, OrigUnit);
1534 // See if there is a relocation to a valid debug map entry inside
1535 // this variable's location. The order is important here. We want to
1536 // always check in the variable has a valid relocation, so that the
1537 // DIEInfo is filled. However, we don't want a static variable in a
1538 // function to force us to keep the enclosing function.
1539 if (!hasValidRelocation(LocationOffset, LocationEndOffset, MyInfo) ||
1540 (Flags & TF_InFunctionScope))
1543 if (Options.Verbose)
1544 DIE.dump(outs(), const_cast<DWARFUnit *>(&OrigUnit), 0, 8 /* Indent */);
1546 return Flags | TF_Keep;
1549 /// \brief Check if a function describing DIE should be kept.
1550 /// \returns updated TraversalFlags.
1551 unsigned DwarfLinker::shouldKeepSubprogramDIE(
1552 const DWARFDebugInfoEntryMinimal &DIE, CompileUnit &Unit,
1553 CompileUnit::DIEInfo &MyInfo, unsigned Flags) {
1554 const auto *Abbrev = DIE.getAbbreviationDeclarationPtr();
1556 Flags |= TF_InFunctionScope;
1558 uint32_t LowPcIdx = Abbrev->findAttributeIndex(dwarf::DW_AT_low_pc);
1559 if (LowPcIdx == -1U)
1562 uint32_t Offset = DIE.getOffset() + getULEB128Size(Abbrev->getCode());
1563 const DWARFUnit &OrigUnit = Unit.getOrigUnit();
1564 uint32_t LowPcOffset, LowPcEndOffset;
1565 std::tie(LowPcOffset, LowPcEndOffset) =
1566 getAttributeOffsets(Abbrev, LowPcIdx, Offset, OrigUnit);
1569 DIE.getAttributeValueAsAddress(&OrigUnit, dwarf::DW_AT_low_pc, -1ULL);
1570 assert(LowPc != -1ULL && "low_pc attribute is not an address.");
1571 if (LowPc == -1ULL ||
1572 !hasValidRelocation(LowPcOffset, LowPcEndOffset, MyInfo))
1575 if (Options.Verbose)
1576 DIE.dump(outs(), const_cast<DWARFUnit *>(&OrigUnit), 0, 8 /* Indent */);
1580 DWARFFormValue HighPcValue;
1581 if (!DIE.getAttributeValue(&OrigUnit, dwarf::DW_AT_high_pc, HighPcValue)) {
1582 reportWarning("Function without high_pc. Range will be discarded.\n",
1588 if (HighPcValue.isFormClass(DWARFFormValue::FC_Address)) {
1589 HighPc = *HighPcValue.getAsAddress(&OrigUnit);
1591 assert(HighPcValue.isFormClass(DWARFFormValue::FC_Constant));
1592 HighPc = LowPc + *HighPcValue.getAsUnsignedConstant();
1595 // Replace the debug map range with a more accurate one.
1596 Ranges[LowPc] = std::make_pair(HighPc, MyInfo.AddrAdjust);
1597 Unit.addFunctionRange(LowPc, HighPc, MyInfo.AddrAdjust);
1601 /// \brief Check if a DIE should be kept.
1602 /// \returns updated TraversalFlags.
1603 unsigned DwarfLinker::shouldKeepDIE(const DWARFDebugInfoEntryMinimal &DIE,
1605 CompileUnit::DIEInfo &MyInfo,
1607 switch (DIE.getTag()) {
1608 case dwarf::DW_TAG_constant:
1609 case dwarf::DW_TAG_variable:
1610 return shouldKeepVariableDIE(DIE, Unit, MyInfo, Flags);
1611 case dwarf::DW_TAG_subprogram:
1612 return shouldKeepSubprogramDIE(DIE, Unit, MyInfo, Flags);
1613 case dwarf::DW_TAG_module:
1614 case dwarf::DW_TAG_imported_module:
1615 case dwarf::DW_TAG_imported_declaration:
1616 case dwarf::DW_TAG_imported_unit:
1617 // We always want to keep these.
1618 return Flags | TF_Keep;
1624 /// \brief Mark the passed DIE as well as all the ones it depends on
1627 /// This function is called by lookForDIEsToKeep on DIEs that are
1628 /// newly discovered to be needed in the link. It recursively calls
1629 /// back to lookForDIEsToKeep while adding TF_DependencyWalk to the
1630 /// TraversalFlags to inform it that it's not doing the primary DIE
1632 void DwarfLinker::keepDIEAndDenpendencies(const DWARFDebugInfoEntryMinimal &DIE,
1633 CompileUnit::DIEInfo &MyInfo,
1634 const DebugMapObject &DMO,
1635 CompileUnit &CU, unsigned Flags) {
1636 const DWARFUnit &Unit = CU.getOrigUnit();
1639 // First mark all the parent chain as kept.
1640 unsigned AncestorIdx = MyInfo.ParentIdx;
1641 while (!CU.getInfo(AncestorIdx).Keep) {
1642 lookForDIEsToKeep(*Unit.getDIEAtIndex(AncestorIdx), DMO, CU,
1643 TF_ParentWalk | TF_Keep | TF_DependencyWalk);
1644 AncestorIdx = CU.getInfo(AncestorIdx).ParentIdx;
1647 // Then we need to mark all the DIEs referenced by this DIE's
1648 // attributes as kept.
1649 DataExtractor Data = Unit.getDebugInfoExtractor();
1650 const auto *Abbrev = DIE.getAbbreviationDeclarationPtr();
1651 uint32_t Offset = DIE.getOffset() + getULEB128Size(Abbrev->getCode());
1653 // Mark all DIEs referenced through atttributes as kept.
1654 for (const auto &AttrSpec : Abbrev->attributes()) {
1655 DWARFFormValue Val(AttrSpec.Form);
1657 if (!Val.isFormClass(DWARFFormValue::FC_Reference)) {
1658 DWARFFormValue::skipValue(AttrSpec.Form, Data, &Offset, &Unit);
1662 Val.extractValue(Data, &Offset, &Unit);
1663 CompileUnit *ReferencedCU;
1664 if (const auto *RefDIE = resolveDIEReference(Val, Unit, DIE, ReferencedCU))
1665 lookForDIEsToKeep(*RefDIE, DMO, *ReferencedCU,
1666 TF_Keep | TF_DependencyWalk);
1670 /// \brief Recursively walk the \p DIE tree and look for DIEs to
1671 /// keep. Store that information in \p CU's DIEInfo.
1673 /// This function is the entry point of the DIE selection
1674 /// algorithm. It is expected to walk the DIE tree in file order and
1675 /// (though the mediation of its helper) call hasValidRelocation() on
1676 /// each DIE that might be a 'root DIE' (See DwarfLinker class
1678 /// While walking the dependencies of root DIEs, this function is
1679 /// also called, but during these dependency walks the file order is
1680 /// not respected. The TF_DependencyWalk flag tells us which kind of
1681 /// traversal we are currently doing.
1682 void DwarfLinker::lookForDIEsToKeep(const DWARFDebugInfoEntryMinimal &DIE,
1683 const DebugMapObject &DMO, CompileUnit &CU,
1685 unsigned Idx = CU.getOrigUnit().getDIEIndex(&DIE);
1686 CompileUnit::DIEInfo &MyInfo = CU.getInfo(Idx);
1687 bool AlreadyKept = MyInfo.Keep;
1689 // If the Keep flag is set, we are marking a required DIE's
1690 // dependencies. If our target is already marked as kept, we're all
1692 if ((Flags & TF_DependencyWalk) && AlreadyKept)
1695 // We must not call shouldKeepDIE while called from keepDIEAndDenpendencies,
1696 // because it would screw up the relocation finding logic.
1697 if (!(Flags & TF_DependencyWalk))
1698 Flags = shouldKeepDIE(DIE, CU, MyInfo, Flags);
1700 // If it is a newly kept DIE mark it as well as all its dependencies as kept.
1701 if (!AlreadyKept && (Flags & TF_Keep))
1702 keepDIEAndDenpendencies(DIE, MyInfo, DMO, CU, Flags);
1704 // The TF_ParentWalk flag tells us that we are currently walking up
1705 // the parent chain of a required DIE, and we don't want to mark all
1706 // the children of the parents as kept (consider for example a
1707 // DW_TAG_namespace node in the parent chain). There are however a
1708 // set of DIE types for which we want to ignore that directive and still
1709 // walk their children.
1710 if (dieNeedsChildrenToBeMeaningful(DIE.getTag()))
1711 Flags &= ~TF_ParentWalk;
1713 if (!DIE.hasChildren() || (Flags & TF_ParentWalk))
1716 for (auto *Child = DIE.getFirstChild(); Child && !Child->isNULL();
1717 Child = Child->getSibling())
1718 lookForDIEsToKeep(*Child, DMO, CU, Flags);
1721 /// \brief Assign an abbreviation numer to \p Abbrev.
1723 /// Our DIEs get freed after every DebugMapObject has been processed,
1724 /// thus the FoldingSet we use to unique DIEAbbrevs cannot refer to
1725 /// the instances hold by the DIEs. When we encounter an abbreviation
1726 /// that we don't know, we create a permanent copy of it.
1727 void DwarfLinker::AssignAbbrev(DIEAbbrev &Abbrev) {
1728 // Check the set for priors.
1729 FoldingSetNodeID ID;
1732 DIEAbbrev *InSet = AbbreviationsSet.FindNodeOrInsertPos(ID, InsertToken);
1734 // If it's newly added.
1736 // Assign existing abbreviation number.
1737 Abbrev.setNumber(InSet->getNumber());
1739 // Add to abbreviation list.
1740 Abbreviations.push_back(
1741 new DIEAbbrev(Abbrev.getTag(), Abbrev.hasChildren()));
1742 for (const auto &Attr : Abbrev.getData())
1743 Abbreviations.back()->AddAttribute(Attr.getAttribute(), Attr.getForm());
1744 AbbreviationsSet.InsertNode(Abbreviations.back(), InsertToken);
1745 // Assign the unique abbreviation number.
1746 Abbrev.setNumber(Abbreviations.size());
1747 Abbreviations.back()->setNumber(Abbreviations.size());
1751 /// \brief Clone a string attribute described by \p AttrSpec and add
1753 /// \returns the size of the new attribute.
1754 unsigned DwarfLinker::cloneStringAttribute(DIE &Die, AttributeSpec AttrSpec,
1755 const DWARFFormValue &Val,
1756 const DWARFUnit &U) {
1757 // Switch everything to out of line strings.
1758 const char *String = *Val.getAsCString(&U);
1759 unsigned Offset = StringPool.getStringOffset(String);
1760 Die.addValue(dwarf::Attribute(AttrSpec.Attr), dwarf::DW_FORM_strp,
1761 new (DIEAlloc) DIEInteger(Offset));
1765 /// \brief Clone an attribute referencing another DIE and add
1767 /// \returns the size of the new attribute.
1768 unsigned DwarfLinker::cloneDieReferenceAttribute(
1769 DIE &Die, const DWARFDebugInfoEntryMinimal &InputDIE,
1770 AttributeSpec AttrSpec, unsigned AttrSize, const DWARFFormValue &Val,
1771 CompileUnit &Unit) {
1772 uint32_t Ref = *Val.getAsReference(&Unit.getOrigUnit());
1773 DIE *NewRefDie = nullptr;
1774 CompileUnit *RefUnit = nullptr;
1775 const DWARFDebugInfoEntryMinimal *RefDie = nullptr;
1777 if (!(RefUnit = getUnitForOffset(Ref)) ||
1778 !(RefDie = RefUnit->getOrigUnit().getDIEForOffset(Ref))) {
1779 const char *AttributeString = dwarf::AttributeString(AttrSpec.Attr);
1780 if (!AttributeString)
1781 AttributeString = "DW_AT_???";
1782 reportWarning(Twine("Missing DIE for ref in attribute ") + AttributeString +
1784 &Unit.getOrigUnit(), &InputDIE);
1788 unsigned Idx = RefUnit->getOrigUnit().getDIEIndex(RefDie);
1789 CompileUnit::DIEInfo &RefInfo = RefUnit->getInfo(Idx);
1790 if (!RefInfo.Clone) {
1791 assert(Ref > InputDIE.getOffset());
1792 // We haven't cloned this DIE yet. Just create an empty one and
1793 // store it. It'll get really cloned when we process it.
1794 RefInfo.Clone = new DIE(dwarf::Tag(RefDie->getTag()));
1796 NewRefDie = RefInfo.Clone;
1798 if (AttrSpec.Form == dwarf::DW_FORM_ref_addr) {
1799 // We cannot currently rely on a DIEEntry to emit ref_addr
1800 // references, because the implementation calls back to DwarfDebug
1801 // to find the unit offset. (We don't have a DwarfDebug)
1802 // FIXME: we should be able to design DIEEntry reliance on
1805 if (Ref < InputDIE.getOffset()) {
1806 // We must have already cloned that DIE.
1807 uint32_t NewRefOffset =
1808 RefUnit->getStartOffset() + NewRefDie->getOffset();
1809 Attr = new (DIEAlloc) DIEInteger(NewRefOffset);
1811 // A forward reference. Note and fixup later.
1812 Attr = new (DIEAlloc) DIEInteger(0xBADDEF);
1813 Unit.noteForwardReference(NewRefDie, RefUnit, Attr);
1815 Die.addValue(dwarf::Attribute(AttrSpec.Attr), dwarf::DW_FORM_ref_addr,
1820 Die.addValue(dwarf::Attribute(AttrSpec.Attr), dwarf::Form(AttrSpec.Form),
1821 new (DIEAlloc) DIEEntry(*NewRefDie));
1825 /// \brief Clone an attribute of block form (locations, constants) and add
1827 /// \returns the size of the new attribute.
1828 unsigned DwarfLinker::cloneBlockAttribute(DIE &Die, AttributeSpec AttrSpec,
1829 const DWARFFormValue &Val,
1830 unsigned AttrSize) {
1833 DIELoc *Loc = nullptr;
1834 DIEBlock *Block = nullptr;
1835 // Just copy the block data over.
1836 if (AttrSpec.Form == dwarf::DW_FORM_exprloc) {
1837 Loc = new (DIEAlloc) DIELoc();
1838 DIELocs.push_back(Loc);
1840 Block = new (DIEAlloc) DIEBlock();
1841 DIEBlocks.push_back(Block);
1843 Attr = Loc ? static_cast<DIE *>(Loc) : static_cast<DIE *>(Block);
1844 Value = Loc ? static_cast<DIEValue *>(Loc) : static_cast<DIEValue *>(Block);
1845 ArrayRef<uint8_t> Bytes = *Val.getAsBlock();
1846 for (auto Byte : Bytes)
1847 Attr->addValue(static_cast<dwarf::Attribute>(0), dwarf::DW_FORM_data1,
1848 new (DIEAlloc) DIEInteger(Byte));
1849 // FIXME: If DIEBlock and DIELoc just reuses the Size field of
1850 // the DIE class, this if could be replaced by
1851 // Attr->setSize(Bytes.size()).
1854 Loc->ComputeSize(&Streamer->getAsmPrinter());
1856 Block->ComputeSize(&Streamer->getAsmPrinter());
1858 Die.addValue(dwarf::Attribute(AttrSpec.Attr), dwarf::Form(AttrSpec.Form),
1863 /// \brief Clone an address attribute and add it to \p Die.
1864 /// \returns the size of the new attribute.
1865 unsigned DwarfLinker::cloneAddressAttribute(DIE &Die, AttributeSpec AttrSpec,
1866 const DWARFFormValue &Val,
1867 const CompileUnit &Unit,
1868 AttributesInfo &Info) {
1869 uint64_t Addr = *Val.getAsAddress(&Unit.getOrigUnit());
1870 if (AttrSpec.Attr == dwarf::DW_AT_low_pc) {
1871 if (Die.getTag() == dwarf::DW_TAG_inlined_subroutine ||
1872 Die.getTag() == dwarf::DW_TAG_lexical_block)
1873 Addr += Info.PCOffset;
1874 else if (Die.getTag() == dwarf::DW_TAG_compile_unit) {
1875 Addr = Unit.getLowPc();
1876 if (Addr == UINT64_MAX)
1879 Info.HasLowPc = true;
1880 } else if (AttrSpec.Attr == dwarf::DW_AT_high_pc) {
1881 if (Die.getTag() == dwarf::DW_TAG_compile_unit) {
1882 if (uint64_t HighPc = Unit.getHighPc())
1887 // If we have a high_pc recorded for the input DIE, use
1888 // it. Otherwise (when no relocations where applied) just use the
1889 // one we just decoded.
1890 Addr = (Info.OrigHighPc ? Info.OrigHighPc : Addr) + Info.PCOffset;
1893 Die.addValue(static_cast<dwarf::Attribute>(AttrSpec.Attr),
1894 static_cast<dwarf::Form>(AttrSpec.Form),
1895 new (DIEAlloc) DIEInteger(Addr));
1896 return Unit.getOrigUnit().getAddressByteSize();
1899 /// \brief Clone a scalar attribute and add it to \p Die.
1900 /// \returns the size of the new attribute.
1901 unsigned DwarfLinker::cloneScalarAttribute(
1902 DIE &Die, const DWARFDebugInfoEntryMinimal &InputDIE, CompileUnit &Unit,
1903 AttributeSpec AttrSpec, const DWARFFormValue &Val, unsigned AttrSize,
1904 AttributesInfo &Info) {
1906 if (AttrSpec.Attr == dwarf::DW_AT_high_pc &&
1907 Die.getTag() == dwarf::DW_TAG_compile_unit) {
1908 if (Unit.getLowPc() == -1ULL)
1910 // Dwarf >= 4 high_pc is an size, not an address.
1911 Value = Unit.getHighPc() - Unit.getLowPc();
1912 } else if (AttrSpec.Form == dwarf::DW_FORM_sec_offset)
1913 Value = *Val.getAsSectionOffset();
1914 else if (AttrSpec.Form == dwarf::DW_FORM_sdata)
1915 Value = *Val.getAsSignedConstant();
1916 else if (auto OptionalValue = Val.getAsUnsignedConstant())
1917 Value = *OptionalValue;
1919 reportWarning("Unsupported scalar attribute form. Dropping attribute.",
1920 &Unit.getOrigUnit(), &InputDIE);
1923 DIEInteger *Attr = new (DIEAlloc) DIEInteger(Value);
1924 if (AttrSpec.Attr == dwarf::DW_AT_ranges)
1925 Unit.noteRangeAttribute(Die, Attr);
1926 // A more generic way to check for location attributes would be
1927 // nice, but it's very unlikely that any other attribute needs a
1929 else if (AttrSpec.Attr == dwarf::DW_AT_location ||
1930 AttrSpec.Attr == dwarf::DW_AT_frame_base)
1931 Unit.noteLocationAttribute(Attr, Info.PCOffset);
1932 else if (AttrSpec.Attr == dwarf::DW_AT_declaration && Value)
1933 Info.IsDeclaration = true;
1935 Die.addValue(dwarf::Attribute(AttrSpec.Attr), dwarf::Form(AttrSpec.Form),
1940 /// \brief Clone \p InputDIE's attribute described by \p AttrSpec with
1941 /// value \p Val, and add it to \p Die.
1942 /// \returns the size of the cloned attribute.
1943 unsigned DwarfLinker::cloneAttribute(DIE &Die,
1944 const DWARFDebugInfoEntryMinimal &InputDIE,
1946 const DWARFFormValue &Val,
1947 const AttributeSpec AttrSpec,
1948 unsigned AttrSize, AttributesInfo &Info) {
1949 const DWARFUnit &U = Unit.getOrigUnit();
1951 switch (AttrSpec.Form) {
1952 case dwarf::DW_FORM_strp:
1953 case dwarf::DW_FORM_string:
1954 return cloneStringAttribute(Die, AttrSpec, Val, U);
1955 case dwarf::DW_FORM_ref_addr:
1956 case dwarf::DW_FORM_ref1:
1957 case dwarf::DW_FORM_ref2:
1958 case dwarf::DW_FORM_ref4:
1959 case dwarf::DW_FORM_ref8:
1960 return cloneDieReferenceAttribute(Die, InputDIE, AttrSpec, AttrSize, Val,
1962 case dwarf::DW_FORM_block:
1963 case dwarf::DW_FORM_block1:
1964 case dwarf::DW_FORM_block2:
1965 case dwarf::DW_FORM_block4:
1966 case dwarf::DW_FORM_exprloc:
1967 return cloneBlockAttribute(Die, AttrSpec, Val, AttrSize);
1968 case dwarf::DW_FORM_addr:
1969 return cloneAddressAttribute(Die, AttrSpec, Val, Unit, Info);
1970 case dwarf::DW_FORM_data1:
1971 case dwarf::DW_FORM_data2:
1972 case dwarf::DW_FORM_data4:
1973 case dwarf::DW_FORM_data8:
1974 case dwarf::DW_FORM_udata:
1975 case dwarf::DW_FORM_sdata:
1976 case dwarf::DW_FORM_sec_offset:
1977 case dwarf::DW_FORM_flag:
1978 case dwarf::DW_FORM_flag_present:
1979 return cloneScalarAttribute(Die, InputDIE, Unit, AttrSpec, Val, AttrSize,
1982 reportWarning("Unsupported attribute form in cloneAttribute. Dropping.", &U,
1989 /// \brief Apply the valid relocations found by findValidRelocs() to
1990 /// the buffer \p Data, taking into account that Data is at \p BaseOffset
1991 /// in the debug_info section.
1993 /// Like for findValidRelocs(), this function must be called with
1994 /// monotonic \p BaseOffset values.
1996 /// \returns wether any reloc has been applied.
1997 bool DwarfLinker::applyValidRelocs(MutableArrayRef<char> Data,
1998 uint32_t BaseOffset, bool isLittleEndian) {
1999 assert((NextValidReloc == 0 ||
2000 BaseOffset > ValidRelocs[NextValidReloc - 1].Offset) &&
2001 "BaseOffset should only be increasing.");
2002 if (NextValidReloc >= ValidRelocs.size())
2005 // Skip relocs that haven't been applied.
2006 while (NextValidReloc < ValidRelocs.size() &&
2007 ValidRelocs[NextValidReloc].Offset < BaseOffset)
2010 bool Applied = false;
2011 uint64_t EndOffset = BaseOffset + Data.size();
2012 while (NextValidReloc < ValidRelocs.size() &&
2013 ValidRelocs[NextValidReloc].Offset >= BaseOffset &&
2014 ValidRelocs[NextValidReloc].Offset < EndOffset) {
2015 const auto &ValidReloc = ValidRelocs[NextValidReloc++];
2016 assert(ValidReloc.Offset - BaseOffset < Data.size());
2017 assert(ValidReloc.Offset - BaseOffset + ValidReloc.Size <= Data.size());
2019 uint64_t Value = ValidReloc.Mapping->getValue().BinaryAddress;
2020 Value += ValidReloc.Addend;
2021 for (unsigned i = 0; i != ValidReloc.Size; ++i) {
2022 unsigned Index = isLittleEndian ? i : (ValidReloc.Size - i - 1);
2023 Buf[i] = uint8_t(Value >> (Index * 8));
2025 assert(ValidReloc.Size <= sizeof(Buf));
2026 memcpy(&Data[ValidReloc.Offset - BaseOffset], Buf, ValidReloc.Size);
2033 static bool isTypeTag(uint16_t Tag) {
2035 case dwarf::DW_TAG_array_type:
2036 case dwarf::DW_TAG_class_type:
2037 case dwarf::DW_TAG_enumeration_type:
2038 case dwarf::DW_TAG_pointer_type:
2039 case dwarf::DW_TAG_reference_type:
2040 case dwarf::DW_TAG_string_type:
2041 case dwarf::DW_TAG_structure_type:
2042 case dwarf::DW_TAG_subroutine_type:
2043 case dwarf::DW_TAG_typedef:
2044 case dwarf::DW_TAG_union_type:
2045 case dwarf::DW_TAG_ptr_to_member_type:
2046 case dwarf::DW_TAG_set_type:
2047 case dwarf::DW_TAG_subrange_type:
2048 case dwarf::DW_TAG_base_type:
2049 case dwarf::DW_TAG_const_type:
2050 case dwarf::DW_TAG_constant:
2051 case dwarf::DW_TAG_file_type:
2052 case dwarf::DW_TAG_namelist:
2053 case dwarf::DW_TAG_packed_type:
2054 case dwarf::DW_TAG_volatile_type:
2055 case dwarf::DW_TAG_restrict_type:
2056 case dwarf::DW_TAG_interface_type:
2057 case dwarf::DW_TAG_unspecified_type:
2058 case dwarf::DW_TAG_shared_type:
2066 /// \brief Recursively clone \p InputDIE's subtrees that have been
2067 /// selected to appear in the linked output.
2069 /// \param OutOffset is the Offset where the newly created DIE will
2070 /// lie in the linked compile unit.
2072 /// \returns the cloned DIE object or null if nothing was selected.
2073 DIE *DwarfLinker::cloneDIE(const DWARFDebugInfoEntryMinimal &InputDIE,
2074 CompileUnit &Unit, int64_t PCOffset,
2075 uint32_t OutOffset) {
2076 DWARFUnit &U = Unit.getOrigUnit();
2077 unsigned Idx = U.getDIEIndex(&InputDIE);
2078 CompileUnit::DIEInfo &Info = Unit.getInfo(Idx);
2080 // Should the DIE appear in the output?
2081 if (!Unit.getInfo(Idx).Keep)
2084 uint32_t Offset = InputDIE.getOffset();
2085 // The DIE might have been already created by a forward reference
2086 // (see cloneDieReferenceAttribute()).
2087 DIE *Die = Info.Clone;
2089 Die = Info.Clone = new DIE(dwarf::Tag(InputDIE.getTag()));
2090 assert(Die->getTag() == InputDIE.getTag());
2091 Die->setOffset(OutOffset);
2093 // Extract and clone every attribute.
2094 DataExtractor Data = U.getDebugInfoExtractor();
2095 uint32_t NextOffset = U.getDIEAtIndex(Idx + 1)->getOffset();
2096 AttributesInfo AttrInfo;
2098 // We could copy the data only if we need to aply a relocation to
2099 // it. After testing, it seems there is no performance downside to
2100 // doing the copy unconditionally, and it makes the code simpler.
2101 SmallString<40> DIECopy(Data.getData().substr(Offset, NextOffset - Offset));
2102 Data = DataExtractor(DIECopy, Data.isLittleEndian(), Data.getAddressSize());
2103 // Modify the copy with relocated addresses.
2104 if (applyValidRelocs(DIECopy, Offset, Data.isLittleEndian())) {
2105 // If we applied relocations, we store the value of high_pc that was
2106 // potentially stored in the input DIE. If high_pc is an address
2107 // (Dwarf version == 2), then it might have been relocated to a
2108 // totally unrelated value (because the end address in the object
2109 // file might be start address of another function which got moved
2110 // independantly by the linker). The computation of the actual
2111 // high_pc value is done in cloneAddressAttribute().
2112 AttrInfo.OrigHighPc =
2113 InputDIE.getAttributeValueAsAddress(&U, dwarf::DW_AT_high_pc, 0);
2116 // Reset the Offset to 0 as we will be working on the local copy of
2120 const auto *Abbrev = InputDIE.getAbbreviationDeclarationPtr();
2121 Offset += getULEB128Size(Abbrev->getCode());
2123 // We are entering a subprogram. Get and propagate the PCOffset.
2124 if (Die->getTag() == dwarf::DW_TAG_subprogram)
2125 PCOffset = Info.AddrAdjust;
2126 AttrInfo.PCOffset = PCOffset;
2128 for (const auto &AttrSpec : Abbrev->attributes()) {
2129 DWARFFormValue Val(AttrSpec.Form);
2130 uint32_t AttrSize = Offset;
2131 Val.extractValue(Data, &Offset, &U);
2132 AttrSize = Offset - AttrSize;
2135 cloneAttribute(*Die, InputDIE, Unit, Val, AttrSpec, AttrSize, AttrInfo);
2138 // Look for accelerator entries.
2139 uint16_t Tag = InputDIE.getTag();
2140 // FIXME: This is slightly wrong. An inline_subroutine without a
2141 // low_pc, but with AT_ranges might be interesting to get into the
2142 // accelerator tables too. For now stick with dsymutil's behavior.
2143 if ((Info.InDebugMap || AttrInfo.HasLowPc) &&
2144 Tag != dwarf::DW_TAG_compile_unit &&
2145 getDIENames(InputDIE, Unit.getOrigUnit(), AttrInfo)) {
2146 if (AttrInfo.MangledName && AttrInfo.MangledName != AttrInfo.Name)
2147 Unit.addNameAccelerator(Die, AttrInfo.MangledName,
2148 AttrInfo.MangledNameOffset,
2149 Tag == dwarf::DW_TAG_inlined_subroutine);
2151 Unit.addNameAccelerator(Die, AttrInfo.Name, AttrInfo.NameOffset,
2152 Tag == dwarf::DW_TAG_inlined_subroutine);
2153 } else if (isTypeTag(Tag) && !AttrInfo.IsDeclaration &&
2154 getDIENames(InputDIE, Unit.getOrigUnit(), AttrInfo)) {
2155 Unit.addTypeAccelerator(Die, AttrInfo.Name, AttrInfo.NameOffset);
2158 DIEAbbrev &NewAbbrev = Die->getAbbrev();
2159 // If a scope DIE is kept, we must have kept at least one child. If
2160 // it's not the case, we'll just be emitting one wasteful end of
2161 // children marker, but things won't break.
2162 if (InputDIE.hasChildren())
2163 NewAbbrev.setChildrenFlag(dwarf::DW_CHILDREN_yes);
2164 // Assign a permanent abbrev number
2165 AssignAbbrev(Die->getAbbrev());
2167 // Add the size of the abbreviation number to the output offset.
2168 OutOffset += getULEB128Size(Die->getAbbrevNumber());
2170 if (!Abbrev->hasChildren()) {
2172 Die->setSize(OutOffset - Die->getOffset());
2176 // Recursively clone children.
2177 for (auto *Child = InputDIE.getFirstChild(); Child && !Child->isNULL();
2178 Child = Child->getSibling()) {
2179 if (DIE *Clone = cloneDIE(*Child, Unit, PCOffset, OutOffset)) {
2180 Die->addChild(std::unique_ptr<DIE>(Clone));
2181 OutOffset = Clone->getOffset() + Clone->getSize();
2185 // Account for the end of children marker.
2186 OutOffset += sizeof(int8_t);
2188 Die->setSize(OutOffset - Die->getOffset());
2192 /// \brief Patch the input object file relevant debug_ranges entries
2193 /// and emit them in the output file. Update the relevant attributes
2194 /// to point at the new entries.
2195 void DwarfLinker::patchRangesForUnit(const CompileUnit &Unit,
2196 DWARFContext &OrigDwarf) const {
2197 DWARFDebugRangeList RangeList;
2198 const auto &FunctionRanges = Unit.getFunctionRanges();
2199 unsigned AddressSize = Unit.getOrigUnit().getAddressByteSize();
2200 DataExtractor RangeExtractor(OrigDwarf.getRangeSection(),
2201 OrigDwarf.isLittleEndian(), AddressSize);
2202 auto InvalidRange = FunctionRanges.end(), CurrRange = InvalidRange;
2203 DWARFUnit &OrigUnit = Unit.getOrigUnit();
2204 const auto *OrigUnitDie = OrigUnit.getCompileUnitDIE(false);
2205 uint64_t OrigLowPc = OrigUnitDie->getAttributeValueAsAddress(
2206 &OrigUnit, dwarf::DW_AT_low_pc, -1ULL);
2207 // Ranges addresses are based on the unit's low_pc. Compute the
2208 // offset we need to apply to adapt to the the new unit's low_pc.
2209 int64_t UnitPcOffset = 0;
2210 if (OrigLowPc != -1ULL)
2211 UnitPcOffset = int64_t(OrigLowPc) - Unit.getLowPc();
2213 for (const auto &RangeAttribute : Unit.getRangesAttributes()) {
2214 uint32_t Offset = RangeAttribute->getValue();
2215 RangeAttribute->setValue(Streamer->getRangesSectionSize());
2216 RangeList.extract(RangeExtractor, &Offset);
2217 const auto &Entries = RangeList.getEntries();
2218 const DWARFDebugRangeList::RangeListEntry &First = Entries.front();
2220 if (CurrRange == InvalidRange || First.StartAddress < CurrRange.start() ||
2221 First.StartAddress >= CurrRange.stop()) {
2222 CurrRange = FunctionRanges.find(First.StartAddress + OrigLowPc);
2223 if (CurrRange == InvalidRange ||
2224 CurrRange.start() > First.StartAddress + OrigLowPc) {
2225 reportWarning("no mapping for range.");
2230 Streamer->emitRangesEntries(UnitPcOffset, OrigLowPc, CurrRange, Entries,
2235 /// \brief Generate the debug_aranges entries for \p Unit and if the
2236 /// unit has a DW_AT_ranges attribute, also emit the debug_ranges
2237 /// contribution for this attribute.
2238 /// FIXME: this could actually be done right in patchRangesForUnit,
2239 /// but for the sake of initial bit-for-bit compatibility with legacy
2240 /// dsymutil, we have to do it in a delayed pass.
2241 void DwarfLinker::generateUnitRanges(CompileUnit &Unit) const {
2242 DIEInteger *Attr = Unit.getUnitRangesAttribute();
2244 Attr->setValue(Streamer->getRangesSectionSize());
2245 Streamer->emitUnitRangesEntries(Unit, Attr != nullptr);
2248 /// \brief Insert the new line info sequence \p Seq into the current
2249 /// set of already linked line info \p Rows.
2250 static void insertLineSequence(std::vector<DWARFDebugLine::Row> &Seq,
2251 std::vector<DWARFDebugLine::Row> &Rows) {
2255 if (!Rows.empty() && Rows.back().Address < Seq.front().Address) {
2256 Rows.insert(Rows.end(), Seq.begin(), Seq.end());
2261 auto InsertPoint = std::lower_bound(
2262 Rows.begin(), Rows.end(), Seq.front(),
2263 [](const DWARFDebugLine::Row &LHS, const DWARFDebugLine::Row &RHS) {
2264 return LHS.Address < RHS.Address;
2267 // FIXME: this only removes the unneeded end_sequence if the
2268 // sequences have been inserted in order. using a global sort like
2269 // described in patchLineTableForUnit() and delaying the end_sequene
2270 // elimination to emitLineTableForUnit() we can get rid of all of them.
2271 if (InsertPoint != Rows.end() &&
2272 InsertPoint->Address == Seq.front().Address && InsertPoint->EndSequence) {
2273 *InsertPoint = Seq.front();
2274 Rows.insert(InsertPoint + 1, Seq.begin() + 1, Seq.end());
2276 Rows.insert(InsertPoint, Seq.begin(), Seq.end());
2282 /// \brief Extract the line table for \p Unit from \p OrigDwarf, and
2283 /// recreate a relocated version of these for the address ranges that
2284 /// are present in the binary.
2285 void DwarfLinker::patchLineTableForUnit(CompileUnit &Unit,
2286 DWARFContext &OrigDwarf) {
2287 const DWARFDebugInfoEntryMinimal *CUDie =
2288 Unit.getOrigUnit().getCompileUnitDIE();
2289 uint64_t StmtList = CUDie->getAttributeValueAsSectionOffset(
2290 &Unit.getOrigUnit(), dwarf::DW_AT_stmt_list, -1ULL);
2291 if (StmtList == -1ULL)
2294 // Update the cloned DW_AT_stmt_list with the correct debug_line offset.
2295 if (auto *OutputDIE = Unit.getOutputUnitDIE()) {
2296 const auto &Abbrev = OutputDIE->getAbbrev().getData();
2297 auto Stmt = std::find_if(
2298 Abbrev.begin(), Abbrev.end(), [](const DIEAbbrevData &AbbrevData) {
2299 return AbbrevData.getAttribute() == dwarf::DW_AT_stmt_list;
2301 assert(Stmt < Abbrev.end() && "Didn't find DW_AT_stmt_list in cloned DIE!");
2302 DIEInteger *StmtAttr =
2303 cast<DIEInteger>(OutputDIE->getValues()[Stmt - Abbrev.begin()]);
2304 StmtAttr->setValue(Streamer->getLineSectionSize());
2307 // Parse the original line info for the unit.
2308 DWARFDebugLine::LineTable LineTable;
2309 uint32_t StmtOffset = StmtList;
2310 StringRef LineData = OrigDwarf.getLineSection().Data;
2311 DataExtractor LineExtractor(LineData, OrigDwarf.isLittleEndian(),
2312 Unit.getOrigUnit().getAddressByteSize());
2313 LineTable.parse(LineExtractor, &OrigDwarf.getLineSection().Relocs,
2316 // This vector is the output line table.
2317 std::vector<DWARFDebugLine::Row> NewRows;
2318 NewRows.reserve(LineTable.Rows.size());
2320 // Current sequence of rows being extracted, before being inserted
2322 std::vector<DWARFDebugLine::Row> Seq;
2323 const auto &FunctionRanges = Unit.getFunctionRanges();
2324 auto InvalidRange = FunctionRanges.end(), CurrRange = InvalidRange;
2326 // FIXME: This logic is meant to generate exactly the same output as
2327 // Darwin's classic dsynutil. There is a nicer way to implement this
2328 // by simply putting all the relocated line info in NewRows and simply
2329 // sorting NewRows before passing it to emitLineTableForUnit. This
2330 // should be correct as sequences for a function should stay
2331 // together in the sorted output. There are a few corner cases that
2332 // look suspicious though, and that required to implement the logic
2333 // this way. Revisit that once initial validation is finished.
2335 // Iterate over the object file line info and extract the sequences
2336 // that correspond to linked functions.
2337 for (auto &Row : LineTable.Rows) {
2338 // Check wether we stepped out of the range. The range is
2339 // half-open, but consider accept the end address of the range if
2340 // it is marked as end_sequence in the input (because in that
2341 // case, the relocation offset is accurate and that entry won't
2342 // serve as the start of another function).
2343 if (CurrRange == InvalidRange || Row.Address < CurrRange.start() ||
2344 Row.Address > CurrRange.stop() ||
2345 (Row.Address == CurrRange.stop() && !Row.EndSequence)) {
2346 // We just stepped out of a known range. Insert a end_sequence
2347 // corresponding to the end of the range.
2348 uint64_t StopAddress = CurrRange != InvalidRange
2349 ? CurrRange.stop() + CurrRange.value()
2351 CurrRange = FunctionRanges.find(Row.Address);
2352 bool CurrRangeValid =
2353 CurrRange != InvalidRange && CurrRange.start() <= Row.Address;
2354 if (!CurrRangeValid) {
2355 CurrRange = InvalidRange;
2356 if (StopAddress != -1ULL) {
2357 // Try harder by looking in the DebugMapObject function
2358 // ranges map. There are corner cases where this finds a
2359 // valid entry. It's unclear if this is right or wrong, but
2360 // for now do as dsymutil.
2361 // FIXME: Understand exactly what cases this addresses and
2362 // potentially remove it along with the Ranges map.
2363 auto Range = Ranges.lower_bound(Row.Address);
2364 if (Range != Ranges.begin() && Range != Ranges.end())
2367 if (Range != Ranges.end() && Range->first <= Row.Address &&
2368 Range->second.first >= Row.Address) {
2369 StopAddress = Row.Address + Range->second.second;
2373 if (StopAddress != -1ULL && !Seq.empty()) {
2374 // Insert end sequence row with the computed end address, but
2375 // the same line as the previous one.
2376 Seq.emplace_back(Seq.back());
2377 Seq.back().Address = StopAddress;
2378 Seq.back().EndSequence = 1;
2379 Seq.back().PrologueEnd = 0;
2380 Seq.back().BasicBlock = 0;
2381 Seq.back().EpilogueBegin = 0;
2382 insertLineSequence(Seq, NewRows);
2385 if (!CurrRangeValid)
2389 // Ignore empty sequences.
2390 if (Row.EndSequence && Seq.empty())
2393 // Relocate row address and add it to the current sequence.
2394 Row.Address += CurrRange.value();
2395 Seq.emplace_back(Row);
2397 if (Row.EndSequence)
2398 insertLineSequence(Seq, NewRows);
2401 // Finished extracting, now emit the line tables.
2402 uint32_t PrologueEnd = StmtList + 10 + LineTable.Prologue.PrologueLength;
2403 // FIXME: LLVM hardcodes it's prologue values. We just copy the
2404 // prologue over and that works because we act as both producer and
2405 // consumer. It would be nicer to have a real configurable line
2407 if (LineTable.Prologue.Version != 2 ||
2408 LineTable.Prologue.DefaultIsStmt != DWARF2_LINE_DEFAULT_IS_STMT ||
2409 LineTable.Prologue.LineBase != -5 || LineTable.Prologue.LineRange != 14 ||
2410 LineTable.Prologue.OpcodeBase != 13)
2411 reportWarning("line table paramters mismatch. Cannot emit.");
2413 Streamer->emitLineTableForUnit(LineData.slice(StmtList + 4, PrologueEnd),
2414 LineTable.Prologue.MinInstLength, NewRows,
2415 Unit.getOrigUnit().getAddressByteSize());
2418 void DwarfLinker::emitAcceleratorEntriesForUnit(CompileUnit &Unit) {
2419 Streamer->emitPubNamesForUnit(Unit);
2420 Streamer->emitPubTypesForUnit(Unit);
2423 bool DwarfLinker::link(const DebugMap &Map) {
2425 if (Map.begin() == Map.end()) {
2426 errs() << "Empty debug map.\n";
2430 if (!createStreamer(Map.getTriple(), OutputFilename))
2433 // Size of the DIEs (and headers) generated for the linked output.
2434 uint64_t OutputDebugInfoSize = 0;
2435 // A unique ID that identifies each compile unit.
2436 unsigned UnitID = 0;
2437 for (const auto &Obj : Map.objects()) {
2438 CurrentDebugObject = Obj.get();
2440 if (Options.Verbose)
2441 outs() << "DEBUG MAP OBJECT: " << Obj->getObjectFilename() << "\n";
2442 auto ErrOrObj = BinHolder.GetObjectFile(Obj->getObjectFilename());
2443 if (std::error_code EC = ErrOrObj.getError()) {
2444 reportWarning(Twine(Obj->getObjectFilename()) + ": " + EC.message());
2448 // Look for relocations that correspond to debug map entries.
2449 if (!findValidRelocsInDebugInfo(*ErrOrObj, *Obj)) {
2450 if (Options.Verbose)
2451 outs() << "No valid relocations found. Skipping.\n";
2455 // Setup access to the debug info.
2456 DWARFContextInMemory DwarfContext(*ErrOrObj);
2457 startDebugObject(DwarfContext, *Obj);
2459 // In a first phase, just read in the debug info and store the DIE
2460 // parent links that we will use during the next phase.
2461 for (const auto &CU : DwarfContext.compile_units()) {
2462 auto *CUDie = CU->getCompileUnitDIE(false);
2463 if (Options.Verbose) {
2464 outs() << "Input compilation unit:";
2465 CUDie->dump(outs(), CU.get(), 0);
2467 Units.emplace_back(*CU, UnitID++);
2468 gatherDIEParents(CUDie, 0, Units.back());
2471 // Then mark all the DIEs that need to be present in the linked
2472 // output and collect some information about them. Note that this
2473 // loop can not be merged with the previous one becaue cross-cu
2474 // references require the ParentIdx to be setup for every CU in
2475 // the object file before calling this.
2476 for (auto &CurrentUnit : Units)
2477 lookForDIEsToKeep(*CurrentUnit.getOrigUnit().getCompileUnitDIE(), *Obj,
2480 // The calls to applyValidRelocs inside cloneDIE will walk the
2481 // reloc array again (in the same way findValidRelocsInDebugInfo()
2482 // did). We need to reset the NextValidReloc index to the beginning.
2485 // Construct the output DIE tree by cloning the DIEs we chose to
2486 // keep above. If there are no valid relocs, then there's nothing
2488 if (!ValidRelocs.empty())
2489 for (auto &CurrentUnit : Units) {
2490 const auto *InputDIE = CurrentUnit.getOrigUnit().getCompileUnitDIE();
2491 CurrentUnit.setStartOffset(OutputDebugInfoSize);
2492 DIE *OutputDIE = cloneDIE(*InputDIE, CurrentUnit, 0 /* PCOffset */,
2493 11 /* Unit Header size */);
2494 CurrentUnit.setOutputUnitDIE(OutputDIE);
2495 OutputDebugInfoSize = CurrentUnit.computeNextUnitOffset();
2496 if (Options.NoOutput)
2498 // FIXME: for compatibility with the classic dsymutil, we emit
2499 // an empty line table for the unit, even if the unit doesn't
2500 // actually exist in the DIE tree.
2501 patchLineTableForUnit(CurrentUnit, DwarfContext);
2504 patchRangesForUnit(CurrentUnit, DwarfContext);
2505 Streamer->emitLocationsForUnit(CurrentUnit, DwarfContext);
2506 emitAcceleratorEntriesForUnit(CurrentUnit);
2509 // Emit all the compile unit's debug information.
2510 if (!ValidRelocs.empty() && !Options.NoOutput)
2511 for (auto &CurrentUnit : Units) {
2512 generateUnitRanges(CurrentUnit);
2513 CurrentUnit.fixupForwardReferences();
2514 Streamer->emitCompileUnitHeader(CurrentUnit);
2515 if (!CurrentUnit.getOutputUnitDIE())
2517 Streamer->emitDIE(*CurrentUnit.getOutputUnitDIE());
2520 // Clean-up before starting working on the next object.
2524 // Emit everything that's global.
2525 if (!Options.NoOutput) {
2526 Streamer->emitAbbrevs(Abbreviations);
2527 Streamer->emitStrings(StringPool);
2530 return Options.NoOutput ? true : Streamer->finish();
2534 bool linkDwarf(StringRef OutputFilename, const DebugMap &DM,
2535 const LinkOptions &Options) {
2536 DwarfLinker Linker(OutputFilename, Options);
2537 return Linker.link(DM);