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 "MachOUtils.h"
14 #include "NonRelocatableStringpool.h"
15 #include "llvm/ADT/IntervalMap.h"
16 #include "llvm/ADT/StringMap.h"
17 #include "llvm/ADT/STLExtras.h"
18 #include "llvm/CodeGen/AsmPrinter.h"
19 #include "llvm/CodeGen/DIE.h"
20 #include "llvm/Config/config.h"
21 #include "llvm/DebugInfo/DWARF/DWARFContext.h"
22 #include "llvm/DebugInfo/DWARF/DWARFDebugInfoEntry.h"
23 #include "llvm/DebugInfo/DWARF/DWARFFormValue.h"
24 #include "llvm/MC/MCAsmBackend.h"
25 #include "llvm/MC/MCAsmInfo.h"
26 #include "llvm/MC/MCContext.h"
27 #include "llvm/MC/MCCodeEmitter.h"
28 #include "llvm/MC/MCDwarf.h"
29 #include "llvm/MC/MCInstrInfo.h"
30 #include "llvm/MC/MCObjectFileInfo.h"
31 #include "llvm/MC/MCRegisterInfo.h"
32 #include "llvm/MC/MCStreamer.h"
33 #include "llvm/MC/MCSubtargetInfo.h"
34 #include "llvm/Object/MachO.h"
35 #include "llvm/Support/Dwarf.h"
36 #include "llvm/Support/LEB128.h"
37 #include "llvm/Support/TargetRegistry.h"
38 #include "llvm/Target/TargetMachine.h"
39 #include "llvm/Target/TargetOptions.h"
48 template <typename KeyT, typename ValT>
49 using HalfOpenIntervalMap =
50 IntervalMap<KeyT, ValT, IntervalMapImpl::NodeSizer<KeyT, ValT>::LeafSize,
51 IntervalMapHalfOpenInfo<KeyT>>;
53 typedef HalfOpenIntervalMap<uint64_t, int64_t> FunctionIntervals;
55 // FIXME: Delete this structure.
56 struct PatchLocation {
57 DIE::value_iterator I;
59 PatchLocation() = default;
60 PatchLocation(DIE::value_iterator I) : I(I) {}
62 void set(uint64_t New) const {
65 assert(Old.getType() == DIEValue::isInteger);
66 *I = DIEValue(Old.getAttribute(), Old.getForm(), DIEInteger(New));
69 uint64_t get() const {
71 return I->getDIEInteger().getValue();
78 /// A DeclContext is a named program scope that is used for ODR
79 /// uniquing of types.
80 /// The set of DeclContext for the ODR-subject parts of a Dwarf link
81 /// is expanded (and uniqued) with each new object file processed. We
82 /// need to determine the context of each DIE in an linked object file
83 /// to see if the corresponding type has already been emitted.
85 /// The contexts are conceptually organised as a tree (eg. a function
86 /// scope is contained in a namespace scope that contains other
87 /// scopes), but storing/accessing them in an actual tree is too
88 /// inefficient: we need to be able to very quickly query a context
89 /// for a given child context by name. Storing a StringMap in each
90 /// DeclContext would be too space inefficient.
91 /// The solution here is to give each DeclContext a link to its parent
92 /// (this allows to walk up the tree), but to query the existance of a
93 /// specific DeclContext using a separate DenseMap keyed on the hash
94 /// of the fully qualified name of the context.
96 unsigned QualifiedNameHash;
102 const DeclContext &Parent;
103 const DWARFDebugInfoEntryMinimal *LastSeenDIE;
104 uint32_t LastSeenCompileUnitID;
105 uint32_t CanonicalDIEOffset;
110 typedef DenseSet<DeclContext *, DeclMapInfo> Map;
113 : QualifiedNameHash(0), Line(0), ByteSize(0),
114 Tag(dwarf::DW_TAG_compile_unit), Name(), File(), Parent(*this),
115 LastSeenDIE(nullptr), LastSeenCompileUnitID(0), CanonicalDIEOffset(0) {}
117 DeclContext(unsigned Hash, uint32_t Line, uint32_t ByteSize, uint16_t Tag,
118 StringRef Name, StringRef File, const DeclContext &Parent,
119 const DWARFDebugInfoEntryMinimal *LastSeenDIE = nullptr,
121 : QualifiedNameHash(Hash), Line(Line), ByteSize(ByteSize), Tag(Tag),
122 Name(Name), File(File), Parent(Parent), LastSeenDIE(LastSeenDIE),
123 LastSeenCompileUnitID(CUId), CanonicalDIEOffset(0) {}
125 uint32_t getQualifiedNameHash() const { return QualifiedNameHash; }
127 bool setLastSeenDIE(CompileUnit &U, const DWARFDebugInfoEntryMinimal *Die);
129 uint32_t getCanonicalDIEOffset() const { return CanonicalDIEOffset; }
130 void setCanonicalDIEOffset(uint32_t Offset) { CanonicalDIEOffset = Offset; }
132 uint16_t getTag() const { return Tag; }
133 StringRef getName() const { return Name; }
136 /// Info type for the DenseMap storing the DeclContext pointers.
137 struct DeclMapInfo : private DenseMapInfo<DeclContext *> {
138 using DenseMapInfo<DeclContext *>::getEmptyKey;
139 using DenseMapInfo<DeclContext *>::getTombstoneKey;
141 static unsigned getHashValue(const DeclContext *Ctxt) {
142 return Ctxt->QualifiedNameHash;
145 static bool isEqual(const DeclContext *LHS, const DeclContext *RHS) {
146 if (RHS == getEmptyKey() || RHS == getTombstoneKey())
148 return LHS->QualifiedNameHash == RHS->QualifiedNameHash &&
149 LHS->Line == RHS->Line && LHS->ByteSize == RHS->ByteSize &&
150 LHS->Name.data() == RHS->Name.data() &&
151 LHS->File.data() == RHS->File.data() &&
152 LHS->Parent.QualifiedNameHash == RHS->Parent.QualifiedNameHash;
156 /// This class gives a tree-like API to the DenseMap that stores the
157 /// DeclContext objects. It also holds the BumpPtrAllocator where
158 /// these objects will be allocated.
159 class DeclContextTree {
160 BumpPtrAllocator Allocator;
162 DeclContext::Map Contexts;
165 /// Get the child of \a Context described by \a DIE in \a Unit. The
166 /// required strings will be interned in \a StringPool.
167 /// \returns The child DeclContext along with one bit that is set if
168 /// this context is invalid.
169 /// FIXME: the invalid bit along the return value is to emulate some
170 /// dsymutil-classic functionality. See the fucntion definition for
171 /// a more thorough discussion of its use.
172 PointerIntPair<DeclContext *, 1>
173 getChildDeclContext(DeclContext &Context,
174 const DWARFDebugInfoEntryMinimal *DIE, CompileUnit &Unit,
175 NonRelocatableStringpool &StringPool);
177 DeclContext &getRoot() { return Root; }
180 /// \brief Stores all information relating to a compile unit, be it in
181 /// its original instance in the object file to its brand new cloned
182 /// and linked DIE tree.
185 /// \brief Information gathered about a DIE in the object file.
187 int64_t AddrAdjust; ///< Address offset to apply to the described entity.
188 DeclContext *Ctxt; ///< ODR Declaration context.
189 DIE *Clone; ///< Cloned version of that DIE.
190 uint32_t ParentIdx; ///< The index of this DIE's parent.
191 bool Keep; ///< Is the DIE part of the linked output?
192 bool InDebugMap; ///< Was this DIE's entity found in the map?
195 CompileUnit(DWARFUnit &OrigUnit, unsigned ID, bool CanUseODR)
196 : OrigUnit(OrigUnit), ID(ID), LowPc(UINT64_MAX), HighPc(0), RangeAlloc(),
198 Info.resize(OrigUnit.getNumDIEs());
200 const auto *CUDie = OrigUnit.getUnitDIE(false);
201 unsigned Lang = CUDie->getAttributeValueAsUnsignedConstant(
202 &OrigUnit, dwarf::DW_AT_language, 0);
203 HasODR = CanUseODR && (Lang == dwarf::DW_LANG_C_plus_plus ||
204 Lang == dwarf::DW_LANG_C_plus_plus_03 ||
205 Lang == dwarf::DW_LANG_C_plus_plus_11 ||
206 Lang == dwarf::DW_LANG_C_plus_plus_14 ||
207 Lang == dwarf::DW_LANG_ObjC_plus_plus);
210 CompileUnit(CompileUnit &&RHS)
211 : OrigUnit(RHS.OrigUnit), Info(std::move(RHS.Info)),
212 CUDie(std::move(RHS.CUDie)), StartOffset(RHS.StartOffset),
213 NextUnitOffset(RHS.NextUnitOffset), RangeAlloc(), Ranges(RangeAlloc) {
214 // The CompileUnit container has been 'reserve()'d with the right
215 // size. We cannot move the IntervalMap anyway.
216 llvm_unreachable("CompileUnits should not be moved.");
219 DWARFUnit &getOrigUnit() const { return OrigUnit; }
221 unsigned getUniqueID() const { return ID; }
223 DIE *getOutputUnitDIE() const { return CUDie; }
224 void setOutputUnitDIE(DIE *Die) { CUDie = Die; }
226 bool hasODR() const { return HasODR; }
228 DIEInfo &getInfo(unsigned Idx) { return Info[Idx]; }
229 const DIEInfo &getInfo(unsigned Idx) const { return Info[Idx]; }
231 uint64_t getStartOffset() const { return StartOffset; }
232 uint64_t getNextUnitOffset() const { return NextUnitOffset; }
233 void setStartOffset(uint64_t DebugInfoSize) { StartOffset = DebugInfoSize; }
235 uint64_t getLowPc() const { return LowPc; }
236 uint64_t getHighPc() const { return HighPc; }
238 Optional<PatchLocation> getUnitRangesAttribute() const {
239 return UnitRangeAttribute;
241 const FunctionIntervals &getFunctionRanges() const { return Ranges; }
242 const std::vector<PatchLocation> &getRangesAttributes() const {
243 return RangeAttributes;
246 const std::vector<std::pair<PatchLocation, int64_t>> &
247 getLocationAttributes() const {
248 return LocationAttributes;
251 /// \brief Compute the end offset for this unit. Must be
252 /// called after the CU's DIEs have been cloned.
253 /// \returns the next unit offset (which is also the current
254 /// debug_info section size).
255 uint64_t computeNextUnitOffset();
257 /// \brief Keep track of a forward reference to DIE \p Die in \p
258 /// RefUnit by \p Attr. The attribute should be fixed up later to
259 /// point to the absolute offset of \p Die in the debug_info section
260 /// or to the canonical offset of \p Ctxt if it is non-null.
261 void noteForwardReference(DIE *Die, const CompileUnit *RefUnit,
262 DeclContext *Ctxt, PatchLocation Attr);
264 /// \brief Apply all fixups recored by noteForwardReference().
265 void fixupForwardReferences();
267 /// \brief Add a function range [\p LowPC, \p HighPC) that is
268 /// relocatad by applying offset \p PCOffset.
269 void addFunctionRange(uint64_t LowPC, uint64_t HighPC, int64_t PCOffset);
271 /// \brief Keep track of a DW_AT_range attribute that we will need to
273 void noteRangeAttribute(const DIE &Die, PatchLocation Attr);
275 /// \brief Keep track of a location attribute pointing to a location
276 /// list in the debug_loc section.
277 void noteLocationAttribute(PatchLocation Attr, int64_t PcOffset);
279 /// \brief Add a name accelerator entry for \p Die with \p Name
280 /// which is stored in the string table at \p Offset.
281 void addNameAccelerator(const DIE *Die, const char *Name, uint32_t Offset,
282 bool SkipPubnamesSection = false);
284 /// \brief Add a type accelerator entry for \p Die with \p Name
285 /// which is stored in the string table at \p Offset.
286 void addTypeAccelerator(const DIE *Die, const char *Name, uint32_t Offset);
289 StringRef Name; ///< Name of the entry.
290 const DIE *Die; ///< DIE this entry describes.
291 uint32_t NameOffset; ///< Offset of Name in the string pool.
292 bool SkipPubSection; ///< Emit this entry only in the apple_* sections.
294 AccelInfo(StringRef Name, const DIE *Die, uint32_t NameOffset,
295 bool SkipPubSection = false)
296 : Name(Name), Die(Die), NameOffset(NameOffset),
297 SkipPubSection(SkipPubSection) {}
300 const std::vector<AccelInfo> &getPubnames() const { return Pubnames; }
301 const std::vector<AccelInfo> &getPubtypes() const { return Pubtypes; }
303 /// Get the full path for file \a FileNum in the line table
304 const char *getResolvedPath(unsigned FileNum) {
305 if (FileNum >= ResolvedPaths.size())
307 return ResolvedPaths[FileNum].size() ? ResolvedPaths[FileNum].c_str()
311 /// Set the fully resolved path for the line-table's file \a FileNum
313 void setResolvedPath(unsigned FileNum, const std::string &Path) {
314 if (ResolvedPaths.size() <= FileNum)
315 ResolvedPaths.resize(FileNum + 1);
316 ResolvedPaths[FileNum] = Path;
322 std::vector<DIEInfo> Info; ///< DIE info indexed by DIE index.
323 DIE *CUDie; ///< Root of the linked DIE tree.
325 uint64_t StartOffset;
326 uint64_t NextUnitOffset;
331 /// \brief A list of attributes to fixup with the absolute offset of
332 /// a DIE in the debug_info section.
334 /// The offsets for the attributes in this array couldn't be set while
335 /// cloning because for cross-cu forward refences the target DIE's
336 /// offset isn't known you emit the reference attribute.
337 std::vector<std::tuple<DIE *, const CompileUnit *, DeclContext *,
338 PatchLocation>> ForwardDIEReferences;
340 FunctionIntervals::Allocator RangeAlloc;
341 /// \brief The ranges in that interval map are the PC ranges for
342 /// functions in this unit, associated with the PC offset to apply
343 /// to the addresses to get the linked address.
344 FunctionIntervals Ranges;
346 /// \brief DW_AT_ranges attributes to patch after we have gathered
347 /// all the unit's function addresses.
349 std::vector<PatchLocation> RangeAttributes;
350 Optional<PatchLocation> UnitRangeAttribute;
353 /// \brief Location attributes that need to be transfered from th
354 /// original debug_loc section to the liked one. They are stored
355 /// along with the PC offset that is to be applied to their
356 /// function's address.
357 std::vector<std::pair<PatchLocation, int64_t>> LocationAttributes;
359 /// \brief Accelerator entries for the unit, both for the pub*
360 /// sections and the apple* ones.
362 std::vector<AccelInfo> Pubnames;
363 std::vector<AccelInfo> Pubtypes;
366 /// Cached resolved paths from the line table.
367 std::vector<std::string> ResolvedPaths;
369 /// Is this unit subject to the ODR rule?
373 uint64_t CompileUnit::computeNextUnitOffset() {
374 NextUnitOffset = StartOffset + 11 /* Header size */;
375 // The root DIE might be null, meaning that the Unit had nothing to
376 // contribute to the linked output. In that case, we will emit the
377 // unit header without any actual DIE.
379 NextUnitOffset += CUDie->getSize();
380 return NextUnitOffset;
383 /// \brief Keep track of a forward cross-cu reference from this unit
384 /// to \p Die that lives in \p RefUnit.
385 void CompileUnit::noteForwardReference(DIE *Die, const CompileUnit *RefUnit,
386 DeclContext *Ctxt, PatchLocation Attr) {
387 ForwardDIEReferences.emplace_back(Die, RefUnit, Ctxt, Attr);
390 /// \brief Apply all fixups recorded by noteForwardReference().
391 void CompileUnit::fixupForwardReferences() {
392 for (const auto &Ref : ForwardDIEReferences) {
394 const CompileUnit *RefUnit;
397 std::tie(RefDie, RefUnit, Ctxt, Attr) = Ref;
398 if (Ctxt && Ctxt->getCanonicalDIEOffset())
399 Attr.set(Ctxt->getCanonicalDIEOffset());
401 Attr.set(RefDie->getOffset() + RefUnit->getStartOffset());
405 void CompileUnit::addFunctionRange(uint64_t FuncLowPc, uint64_t FuncHighPc,
407 Ranges.insert(FuncLowPc, FuncHighPc, PcOffset);
408 this->LowPc = std::min(LowPc, FuncLowPc + PcOffset);
409 this->HighPc = std::max(HighPc, FuncHighPc + PcOffset);
412 void CompileUnit::noteRangeAttribute(const DIE &Die, PatchLocation Attr) {
413 if (Die.getTag() != dwarf::DW_TAG_compile_unit)
414 RangeAttributes.push_back(Attr);
416 UnitRangeAttribute = Attr;
419 void CompileUnit::noteLocationAttribute(PatchLocation Attr, int64_t PcOffset) {
420 LocationAttributes.emplace_back(Attr, PcOffset);
423 /// \brief Add a name accelerator entry for \p Die with \p Name
424 /// which is stored in the string table at \p Offset.
425 void CompileUnit::addNameAccelerator(const DIE *Die, const char *Name,
426 uint32_t Offset, bool SkipPubSection) {
427 Pubnames.emplace_back(Name, Die, Offset, SkipPubSection);
430 /// \brief Add a type accelerator entry for \p Die with \p Name
431 /// which is stored in the string table at \p Offset.
432 void CompileUnit::addTypeAccelerator(const DIE *Die, const char *Name,
434 Pubtypes.emplace_back(Name, Die, Offset, false);
437 /// \brief The Dwarf streaming logic
439 /// All interactions with the MC layer that is used to build the debug
440 /// information binary representation are handled in this class.
441 class DwarfStreamer {
442 /// \defgroup MCObjects MC layer objects constructed by the streamer
444 std::unique_ptr<MCRegisterInfo> MRI;
445 std::unique_ptr<MCAsmInfo> MAI;
446 std::unique_ptr<MCObjectFileInfo> MOFI;
447 std::unique_ptr<MCContext> MC;
448 MCAsmBackend *MAB; // Owned by MCStreamer
449 std::unique_ptr<MCInstrInfo> MII;
450 std::unique_ptr<MCSubtargetInfo> MSTI;
451 MCCodeEmitter *MCE; // Owned by MCStreamer
452 MCStreamer *MS; // Owned by AsmPrinter
453 std::unique_ptr<TargetMachine> TM;
454 std::unique_ptr<AsmPrinter> Asm;
457 /// \brief the file we stream the linked Dwarf to.
458 std::unique_ptr<raw_fd_ostream> OutFile;
460 uint32_t RangesSectionSize;
461 uint32_t LocSectionSize;
462 uint32_t LineSectionSize;
463 uint32_t FrameSectionSize;
465 /// \brief Emit the pubnames or pubtypes section contribution for \p
466 /// Unit into \p Sec. The data is provided in \p Names.
467 void emitPubSectionForUnit(MCSection *Sec, StringRef Name,
468 const CompileUnit &Unit,
469 const std::vector<CompileUnit::AccelInfo> &Names);
472 /// \brief Actually create the streamer and the ouptut file.
474 /// This could be done directly in the constructor, but it feels
475 /// more natural to handle errors through return value.
476 bool init(Triple TheTriple, StringRef OutputFilename);
478 /// \brief Dump the file to the disk.
479 bool finish(const DebugMap &);
481 AsmPrinter &getAsmPrinter() const { return *Asm; }
483 /// \brief Set the current output section to debug_info and change
484 /// the MC Dwarf version to \p DwarfVersion.
485 void switchToDebugInfoSection(unsigned DwarfVersion);
487 /// \brief Emit the compilation unit header for \p Unit in the
488 /// debug_info section.
490 /// As a side effect, this also switches the current Dwarf version
491 /// of the MC layer to the one of U.getOrigUnit().
492 void emitCompileUnitHeader(CompileUnit &Unit);
494 /// \brief Recursively emit the DIE tree rooted at \p Die.
495 void emitDIE(DIE &Die);
497 /// \brief Emit the abbreviation table \p Abbrevs to the
498 /// debug_abbrev section.
499 void emitAbbrevs(const std::vector<DIEAbbrev *> &Abbrevs);
501 /// \brief Emit the string table described by \p Pool.
502 void emitStrings(const NonRelocatableStringpool &Pool);
504 /// \brief Emit debug_ranges for \p FuncRange by translating the
505 /// original \p Entries.
506 void emitRangesEntries(
507 int64_t UnitPcOffset, uint64_t OrigLowPc,
508 FunctionIntervals::const_iterator FuncRange,
509 const std::vector<DWARFDebugRangeList::RangeListEntry> &Entries,
510 unsigned AddressSize);
512 /// \brief Emit debug_aranges entries for \p Unit and if \p
513 /// DoRangesSection is true, also emit the debug_ranges entries for
514 /// the DW_TAG_compile_unit's DW_AT_ranges attribute.
515 void emitUnitRangesEntries(CompileUnit &Unit, bool DoRangesSection);
517 uint32_t getRangesSectionSize() const { return RangesSectionSize; }
519 /// \brief Emit the debug_loc contribution for \p Unit by copying
520 /// the entries from \p Dwarf and offseting them. Update the
521 /// location attributes to point to the new entries.
522 void emitLocationsForUnit(const CompileUnit &Unit, DWARFContext &Dwarf);
524 /// \brief Emit the line table described in \p Rows into the
525 /// debug_line section.
526 void emitLineTableForUnit(MCDwarfLineTableParams Params,
527 StringRef PrologueBytes, unsigned MinInstLength,
528 std::vector<DWARFDebugLine::Row> &Rows,
529 unsigned AdddressSize);
531 uint32_t getLineSectionSize() const { return LineSectionSize; }
533 /// \brief Emit the .debug_pubnames contribution for \p Unit.
534 void emitPubNamesForUnit(const CompileUnit &Unit);
536 /// \brief Emit the .debug_pubtypes contribution for \p Unit.
537 void emitPubTypesForUnit(const CompileUnit &Unit);
539 /// \brief Emit a CIE.
540 void emitCIE(StringRef CIEBytes);
542 /// \brief Emit an FDE with data \p Bytes.
543 void emitFDE(uint32_t CIEOffset, uint32_t AddreSize, uint32_t Address,
546 uint32_t getFrameSectionSize() const { return FrameSectionSize; }
549 bool DwarfStreamer::init(Triple TheTriple, StringRef OutputFilename) {
550 std::string ErrorStr;
551 std::string TripleName;
552 StringRef Context = "dwarf streamer init";
555 const Target *TheTarget =
556 TargetRegistry::lookupTarget(TripleName, TheTriple, ErrorStr);
558 return error(ErrorStr, Context);
559 TripleName = TheTriple.getTriple();
561 // Create all the MC Objects.
562 MRI.reset(TheTarget->createMCRegInfo(TripleName));
564 return error(Twine("no register info for target ") + TripleName, Context);
566 MAI.reset(TheTarget->createMCAsmInfo(*MRI, TripleName));
568 return error("no asm info for target " + TripleName, Context);
570 MOFI.reset(new MCObjectFileInfo);
571 MC.reset(new MCContext(MAI.get(), MRI.get(), MOFI.get()));
572 MOFI->InitMCObjectFileInfo(TheTriple, Reloc::Default, CodeModel::Default,
575 MAB = TheTarget->createMCAsmBackend(*MRI, TripleName, "");
577 return error("no asm backend for target " + TripleName, Context);
579 MII.reset(TheTarget->createMCInstrInfo());
581 return error("no instr info info for target " + TripleName, Context);
583 MSTI.reset(TheTarget->createMCSubtargetInfo(TripleName, "", ""));
585 return error("no subtarget info for target " + TripleName, Context);
587 MCE = TheTarget->createMCCodeEmitter(*MII, *MRI, *MC);
589 return error("no code emitter for target " + TripleName, Context);
591 // Create the output file.
594 llvm::make_unique<raw_fd_ostream>(OutputFilename, EC, sys::fs::F_None);
596 return error(Twine(OutputFilename) + ": " + EC.message(), Context);
598 MS = TheTarget->createMCObjectStreamer(TheTriple, *MC, *MAB, *OutFile, MCE,
600 /*DWARFMustBeAtTheEnd*/ false);
602 return error("no object streamer for target " + TripleName, Context);
604 // Finally create the AsmPrinter we'll use to emit the DIEs.
605 TM.reset(TheTarget->createTargetMachine(TripleName, "", "", TargetOptions()));
607 return error("no target machine for target " + TripleName, Context);
609 Asm.reset(TheTarget->createAsmPrinter(*TM, std::unique_ptr<MCStreamer>(MS)));
611 return error("no asm printer for target " + TripleName, Context);
613 RangesSectionSize = 0;
616 FrameSectionSize = 0;
621 bool DwarfStreamer::finish(const DebugMap &DM) {
622 if (DM.getTriple().isOSDarwin() && !DM.getBinaryPath().empty())
623 return MachOUtils::generateDsymCompanion(DM, *MS, *OutFile);
629 /// \brief Set the current output section to debug_info and change
630 /// the MC Dwarf version to \p DwarfVersion.
631 void DwarfStreamer::switchToDebugInfoSection(unsigned DwarfVersion) {
632 MS->SwitchSection(MOFI->getDwarfInfoSection());
633 MC->setDwarfVersion(DwarfVersion);
636 /// \brief Emit the compilation unit header for \p Unit in the
637 /// debug_info section.
639 /// A Dwarf scetion header is encoded as:
640 /// uint32_t Unit length (omiting this field)
642 /// uint32_t Abbreviation table offset
643 /// uint8_t Address size
645 /// Leading to a total of 11 bytes.
646 void DwarfStreamer::emitCompileUnitHeader(CompileUnit &Unit) {
647 unsigned Version = Unit.getOrigUnit().getVersion();
648 switchToDebugInfoSection(Version);
650 // Emit size of content not including length itself. The size has
651 // already been computed in CompileUnit::computeOffsets(). Substract
652 // 4 to that size to account for the length field.
653 Asm->EmitInt32(Unit.getNextUnitOffset() - Unit.getStartOffset() - 4);
654 Asm->EmitInt16(Version);
655 // We share one abbreviations table across all units so it's always at the
656 // start of the section.
658 Asm->EmitInt8(Unit.getOrigUnit().getAddressByteSize());
661 /// \brief Emit the \p Abbrevs array as the shared abbreviation table
662 /// for the linked Dwarf file.
663 void DwarfStreamer::emitAbbrevs(const std::vector<DIEAbbrev *> &Abbrevs) {
664 MS->SwitchSection(MOFI->getDwarfAbbrevSection());
665 Asm->emitDwarfAbbrevs(Abbrevs);
668 /// \brief Recursively emit the DIE tree rooted at \p Die.
669 void DwarfStreamer::emitDIE(DIE &Die) {
670 MS->SwitchSection(MOFI->getDwarfInfoSection());
671 Asm->emitDwarfDIE(Die);
674 /// \brief Emit the debug_str section stored in \p Pool.
675 void DwarfStreamer::emitStrings(const NonRelocatableStringpool &Pool) {
676 Asm->OutStreamer->SwitchSection(MOFI->getDwarfStrSection());
677 for (auto *Entry = Pool.getFirstEntry(); Entry;
678 Entry = Pool.getNextEntry(Entry))
679 Asm->OutStreamer->EmitBytes(
680 StringRef(Entry->getKey().data(), Entry->getKey().size() + 1));
683 /// \brief Emit the debug_range section contents for \p FuncRange by
684 /// translating the original \p Entries. The debug_range section
685 /// format is totally trivial, consisting just of pairs of address
686 /// sized addresses describing the ranges.
687 void DwarfStreamer::emitRangesEntries(
688 int64_t UnitPcOffset, uint64_t OrigLowPc,
689 FunctionIntervals::const_iterator FuncRange,
690 const std::vector<DWARFDebugRangeList::RangeListEntry> &Entries,
691 unsigned AddressSize) {
692 MS->SwitchSection(MC->getObjectFileInfo()->getDwarfRangesSection());
694 // Offset each range by the right amount.
695 int64_t PcOffset = Entries.empty() ? 0 : FuncRange.value() + UnitPcOffset;
696 for (const auto &Range : Entries) {
697 if (Range.isBaseAddressSelectionEntry(AddressSize)) {
698 warn("unsupported base address selection operation",
699 "emitting debug_ranges");
702 // Do not emit empty ranges.
703 if (Range.StartAddress == Range.EndAddress)
706 // All range entries should lie in the function range.
707 if (!(Range.StartAddress + OrigLowPc >= FuncRange.start() &&
708 Range.EndAddress + OrigLowPc <= FuncRange.stop()))
709 warn("inconsistent range data.", "emitting debug_ranges");
710 MS->EmitIntValue(Range.StartAddress + PcOffset, AddressSize);
711 MS->EmitIntValue(Range.EndAddress + PcOffset, AddressSize);
712 RangesSectionSize += 2 * AddressSize;
715 // Add the terminator entry.
716 MS->EmitIntValue(0, AddressSize);
717 MS->EmitIntValue(0, AddressSize);
718 RangesSectionSize += 2 * AddressSize;
721 /// \brief Emit the debug_aranges contribution of a unit and
722 /// if \p DoDebugRanges is true the debug_range contents for a
723 /// compile_unit level DW_AT_ranges attribute (Which are basically the
724 /// same thing with a different base address).
725 /// Just aggregate all the ranges gathered inside that unit.
726 void DwarfStreamer::emitUnitRangesEntries(CompileUnit &Unit,
727 bool DoDebugRanges) {
728 unsigned AddressSize = Unit.getOrigUnit().getAddressByteSize();
729 // Gather the ranges in a vector, so that we can simplify them. The
730 // IntervalMap will have coalesced the non-linked ranges, but here
731 // we want to coalesce the linked addresses.
732 std::vector<std::pair<uint64_t, uint64_t>> Ranges;
733 const auto &FunctionRanges = Unit.getFunctionRanges();
734 for (auto Range = FunctionRanges.begin(), End = FunctionRanges.end();
735 Range != End; ++Range)
736 Ranges.push_back(std::make_pair(Range.start() + Range.value(),
737 Range.stop() + Range.value()));
739 // The object addresses where sorted, but again, the linked
740 // addresses might end up in a different order.
741 std::sort(Ranges.begin(), Ranges.end());
743 if (!Ranges.empty()) {
744 MS->SwitchSection(MC->getObjectFileInfo()->getDwarfARangesSection());
746 MCSymbol *BeginLabel = Asm->createTempSymbol("Barange");
747 MCSymbol *EndLabel = Asm->createTempSymbol("Earange");
749 unsigned HeaderSize =
750 sizeof(int32_t) + // Size of contents (w/o this field
751 sizeof(int16_t) + // DWARF ARange version number
752 sizeof(int32_t) + // Offset of CU in the .debug_info section
753 sizeof(int8_t) + // Pointer Size (in bytes)
754 sizeof(int8_t); // Segment Size (in bytes)
756 unsigned TupleSize = AddressSize * 2;
757 unsigned Padding = OffsetToAlignment(HeaderSize, TupleSize);
759 Asm->EmitLabelDifference(EndLabel, BeginLabel, 4); // Arange length
760 Asm->OutStreamer->EmitLabel(BeginLabel);
761 Asm->EmitInt16(dwarf::DW_ARANGES_VERSION); // Version number
762 Asm->EmitInt32(Unit.getStartOffset()); // Corresponding unit's offset
763 Asm->EmitInt8(AddressSize); // Address size
764 Asm->EmitInt8(0); // Segment size
766 Asm->OutStreamer->EmitFill(Padding, 0x0);
768 for (auto Range = Ranges.begin(), End = Ranges.end(); Range != End;
770 uint64_t RangeStart = Range->first;
771 MS->EmitIntValue(RangeStart, AddressSize);
772 while ((Range + 1) != End && Range->second == (Range + 1)->first)
774 MS->EmitIntValue(Range->second - RangeStart, AddressSize);
778 Asm->OutStreamer->EmitIntValue(0, AddressSize);
779 Asm->OutStreamer->EmitIntValue(0, AddressSize);
780 Asm->OutStreamer->EmitLabel(EndLabel);
786 MS->SwitchSection(MC->getObjectFileInfo()->getDwarfRangesSection());
787 // Offset each range by the right amount.
788 int64_t PcOffset = -Unit.getLowPc();
789 // Emit coalesced ranges.
790 for (auto Range = Ranges.begin(), End = Ranges.end(); Range != End; ++Range) {
791 MS->EmitIntValue(Range->first + PcOffset, AddressSize);
792 while (Range + 1 != End && Range->second == (Range + 1)->first)
794 MS->EmitIntValue(Range->second + PcOffset, AddressSize);
795 RangesSectionSize += 2 * AddressSize;
798 // Add the terminator entry.
799 MS->EmitIntValue(0, AddressSize);
800 MS->EmitIntValue(0, AddressSize);
801 RangesSectionSize += 2 * AddressSize;
804 /// \brief Emit location lists for \p Unit and update attribtues to
805 /// point to the new entries.
806 void DwarfStreamer::emitLocationsForUnit(const CompileUnit &Unit,
807 DWARFContext &Dwarf) {
808 const auto &Attributes = Unit.getLocationAttributes();
810 if (Attributes.empty())
813 MS->SwitchSection(MC->getObjectFileInfo()->getDwarfLocSection());
815 unsigned AddressSize = Unit.getOrigUnit().getAddressByteSize();
816 const DWARFSection &InputSec = Dwarf.getLocSection();
817 DataExtractor Data(InputSec.Data, Dwarf.isLittleEndian(), AddressSize);
818 DWARFUnit &OrigUnit = Unit.getOrigUnit();
819 const auto *OrigUnitDie = OrigUnit.getUnitDIE(false);
820 int64_t UnitPcOffset = 0;
821 uint64_t OrigLowPc = OrigUnitDie->getAttributeValueAsAddress(
822 &OrigUnit, dwarf::DW_AT_low_pc, -1ULL);
823 if (OrigLowPc != -1ULL)
824 UnitPcOffset = int64_t(OrigLowPc) - Unit.getLowPc();
826 for (const auto &Attr : Attributes) {
827 uint32_t Offset = Attr.first.get();
828 Attr.first.set(LocSectionSize);
829 // This is the quantity to add to the old location address to get
830 // the correct address for the new one.
831 int64_t LocPcOffset = Attr.second + UnitPcOffset;
832 while (Data.isValidOffset(Offset)) {
833 uint64_t Low = Data.getUnsigned(&Offset, AddressSize);
834 uint64_t High = Data.getUnsigned(&Offset, AddressSize);
835 LocSectionSize += 2 * AddressSize;
836 if (Low == 0 && High == 0) {
837 Asm->OutStreamer->EmitIntValue(0, AddressSize);
838 Asm->OutStreamer->EmitIntValue(0, AddressSize);
841 Asm->OutStreamer->EmitIntValue(Low + LocPcOffset, AddressSize);
842 Asm->OutStreamer->EmitIntValue(High + LocPcOffset, AddressSize);
843 uint64_t Length = Data.getU16(&Offset);
844 Asm->OutStreamer->EmitIntValue(Length, 2);
845 // Just copy the bytes over.
846 Asm->OutStreamer->EmitBytes(
847 StringRef(InputSec.Data.substr(Offset, Length)));
849 LocSectionSize += Length + 2;
854 void DwarfStreamer::emitLineTableForUnit(MCDwarfLineTableParams Params,
855 StringRef PrologueBytes,
856 unsigned MinInstLength,
857 std::vector<DWARFDebugLine::Row> &Rows,
858 unsigned PointerSize) {
859 // Switch to the section where the table will be emitted into.
860 MS->SwitchSection(MC->getObjectFileInfo()->getDwarfLineSection());
861 MCSymbol *LineStartSym = MC->createTempSymbol();
862 MCSymbol *LineEndSym = MC->createTempSymbol();
864 // The first 4 bytes is the total length of the information for this
865 // compilation unit (not including these 4 bytes for the length).
866 Asm->EmitLabelDifference(LineEndSym, LineStartSym, 4);
867 Asm->OutStreamer->EmitLabel(LineStartSym);
869 MS->EmitBytes(PrologueBytes);
870 LineSectionSize += PrologueBytes.size() + 4;
872 SmallString<128> EncodingBuffer;
873 raw_svector_ostream EncodingOS(EncodingBuffer);
876 // We only have the dummy entry, dsymutil emits an entry with a 0
877 // address in that case.
878 MCDwarfLineAddr::Encode(*MC, Params, INT64_MAX, 0, EncodingOS);
879 MS->EmitBytes(EncodingOS.str());
880 LineSectionSize += EncodingBuffer.size();
881 MS->EmitLabel(LineEndSym);
885 // Line table state machine fields
886 unsigned FileNum = 1;
887 unsigned LastLine = 1;
889 unsigned IsStatement = 1;
891 uint64_t Address = -1ULL;
893 unsigned RowsSinceLastSequence = 0;
895 for (unsigned Idx = 0; Idx < Rows.size(); ++Idx) {
896 auto &Row = Rows[Idx];
898 int64_t AddressDelta;
899 if (Address == -1ULL) {
900 MS->EmitIntValue(dwarf::DW_LNS_extended_op, 1);
901 MS->EmitULEB128IntValue(PointerSize + 1);
902 MS->EmitIntValue(dwarf::DW_LNE_set_address, 1);
903 MS->EmitIntValue(Row.Address, PointerSize);
904 LineSectionSize += 2 + PointerSize + getULEB128Size(PointerSize + 1);
907 AddressDelta = (Row.Address - Address) / MinInstLength;
910 // FIXME: code copied and transfromed from
911 // MCDwarf.cpp::EmitDwarfLineTable. We should find a way to share
912 // this code, but the current compatibility requirement with
913 // classic dsymutil makes it hard. Revisit that once this
914 // requirement is dropped.
916 if (FileNum != Row.File) {
918 MS->EmitIntValue(dwarf::DW_LNS_set_file, 1);
919 MS->EmitULEB128IntValue(FileNum);
920 LineSectionSize += 1 + getULEB128Size(FileNum);
922 if (Column != Row.Column) {
924 MS->EmitIntValue(dwarf::DW_LNS_set_column, 1);
925 MS->EmitULEB128IntValue(Column);
926 LineSectionSize += 1 + getULEB128Size(Column);
929 // FIXME: We should handle the discriminator here, but dsymutil
930 // doesn' consider it, thus ignore it for now.
932 if (Isa != Row.Isa) {
934 MS->EmitIntValue(dwarf::DW_LNS_set_isa, 1);
935 MS->EmitULEB128IntValue(Isa);
936 LineSectionSize += 1 + getULEB128Size(Isa);
938 if (IsStatement != Row.IsStmt) {
939 IsStatement = Row.IsStmt;
940 MS->EmitIntValue(dwarf::DW_LNS_negate_stmt, 1);
941 LineSectionSize += 1;
943 if (Row.BasicBlock) {
944 MS->EmitIntValue(dwarf::DW_LNS_set_basic_block, 1);
945 LineSectionSize += 1;
948 if (Row.PrologueEnd) {
949 MS->EmitIntValue(dwarf::DW_LNS_set_prologue_end, 1);
950 LineSectionSize += 1;
953 if (Row.EpilogueBegin) {
954 MS->EmitIntValue(dwarf::DW_LNS_set_epilogue_begin, 1);
955 LineSectionSize += 1;
958 int64_t LineDelta = int64_t(Row.Line) - LastLine;
959 if (!Row.EndSequence) {
960 MCDwarfLineAddr::Encode(*MC, Params, LineDelta, AddressDelta, EncodingOS);
961 MS->EmitBytes(EncodingOS.str());
962 LineSectionSize += EncodingBuffer.size();
963 EncodingBuffer.resize(0);
964 Address = Row.Address;
966 RowsSinceLastSequence++;
969 MS->EmitIntValue(dwarf::DW_LNS_advance_line, 1);
970 MS->EmitSLEB128IntValue(LineDelta);
971 LineSectionSize += 1 + getSLEB128Size(LineDelta);
974 MS->EmitIntValue(dwarf::DW_LNS_advance_pc, 1);
975 MS->EmitULEB128IntValue(AddressDelta);
976 LineSectionSize += 1 + getULEB128Size(AddressDelta);
978 MCDwarfLineAddr::Encode(*MC, Params, INT64_MAX, 0, EncodingOS);
979 MS->EmitBytes(EncodingOS.str());
980 LineSectionSize += EncodingBuffer.size();
981 EncodingBuffer.resize(0);
983 LastLine = FileNum = IsStatement = 1;
984 RowsSinceLastSequence = Column = Isa = 0;
988 if (RowsSinceLastSequence) {
989 MCDwarfLineAddr::Encode(*MC, Params, INT64_MAX, 0, EncodingOS);
990 MS->EmitBytes(EncodingOS.str());
991 LineSectionSize += EncodingBuffer.size();
992 EncodingBuffer.resize(0);
995 MS->EmitLabel(LineEndSym);
998 /// \brief Emit the pubnames or pubtypes section contribution for \p
999 /// Unit into \p Sec. The data is provided in \p Names.
1000 void DwarfStreamer::emitPubSectionForUnit(
1001 MCSection *Sec, StringRef SecName, const CompileUnit &Unit,
1002 const std::vector<CompileUnit::AccelInfo> &Names) {
1006 // Start the dwarf pubnames section.
1007 Asm->OutStreamer->SwitchSection(Sec);
1008 MCSymbol *BeginLabel = Asm->createTempSymbol("pub" + SecName + "_begin");
1009 MCSymbol *EndLabel = Asm->createTempSymbol("pub" + SecName + "_end");
1011 bool HeaderEmitted = false;
1012 // Emit the pubnames for this compilation unit.
1013 for (const auto &Name : Names) {
1014 if (Name.SkipPubSection)
1017 if (!HeaderEmitted) {
1019 Asm->EmitLabelDifference(EndLabel, BeginLabel, 4); // Length
1020 Asm->OutStreamer->EmitLabel(BeginLabel);
1021 Asm->EmitInt16(dwarf::DW_PUBNAMES_VERSION); // Version
1022 Asm->EmitInt32(Unit.getStartOffset()); // Unit offset
1023 Asm->EmitInt32(Unit.getNextUnitOffset() - Unit.getStartOffset()); // Size
1024 HeaderEmitted = true;
1026 Asm->EmitInt32(Name.Die->getOffset());
1027 Asm->OutStreamer->EmitBytes(
1028 StringRef(Name.Name.data(), Name.Name.size() + 1));
1033 Asm->EmitInt32(0); // End marker.
1034 Asm->OutStreamer->EmitLabel(EndLabel);
1037 /// \brief Emit .debug_pubnames for \p Unit.
1038 void DwarfStreamer::emitPubNamesForUnit(const CompileUnit &Unit) {
1039 emitPubSectionForUnit(MC->getObjectFileInfo()->getDwarfPubNamesSection(),
1040 "names", Unit, Unit.getPubnames());
1043 /// \brief Emit .debug_pubtypes for \p Unit.
1044 void DwarfStreamer::emitPubTypesForUnit(const CompileUnit &Unit) {
1045 emitPubSectionForUnit(MC->getObjectFileInfo()->getDwarfPubTypesSection(),
1046 "types", Unit, Unit.getPubtypes());
1049 /// \brief Emit a CIE into the debug_frame section.
1050 void DwarfStreamer::emitCIE(StringRef CIEBytes) {
1051 MS->SwitchSection(MC->getObjectFileInfo()->getDwarfFrameSection());
1053 MS->EmitBytes(CIEBytes);
1054 FrameSectionSize += CIEBytes.size();
1057 /// \brief Emit a FDE into the debug_frame section. \p FDEBytes
1058 /// contains the FDE data without the length, CIE offset and address
1059 /// which will be replaced with the paramter values.
1060 void DwarfStreamer::emitFDE(uint32_t CIEOffset, uint32_t AddrSize,
1061 uint32_t Address, StringRef FDEBytes) {
1062 MS->SwitchSection(MC->getObjectFileInfo()->getDwarfFrameSection());
1064 MS->EmitIntValue(FDEBytes.size() + 4 + AddrSize, 4);
1065 MS->EmitIntValue(CIEOffset, 4);
1066 MS->EmitIntValue(Address, AddrSize);
1067 MS->EmitBytes(FDEBytes);
1068 FrameSectionSize += FDEBytes.size() + 8 + AddrSize;
1071 /// \brief The core of the Dwarf linking logic.
1073 /// The link of the dwarf information from the object files will be
1074 /// driven by the selection of 'root DIEs', which are DIEs that
1075 /// describe variables or functions that are present in the linked
1076 /// binary (and thus have entries in the debug map). All the debug
1077 /// information that will be linked (the DIEs, but also the line
1078 /// tables, ranges, ...) is derived from that set of root DIEs.
1080 /// The root DIEs are identified because they contain relocations that
1081 /// correspond to a debug map entry at specific places (the low_pc for
1082 /// a function, the location for a variable). These relocations are
1083 /// called ValidRelocs in the DwarfLinker and are gathered as a very
1084 /// first step when we start processing a DebugMapObject.
1087 DwarfLinker(StringRef OutputFilename, const LinkOptions &Options)
1088 : OutputFilename(OutputFilename), Options(Options),
1089 BinHolder(Options.Verbose), LastCIEOffset(0) {}
1092 for (auto *Abbrev : Abbreviations)
1096 /// \brief Link the contents of the DebugMap.
1097 bool link(const DebugMap &);
1100 /// \brief Called at the start of a debug object link.
1101 void startDebugObject(DWARFContext &, DebugMapObject &);
1103 /// \brief Called at the end of a debug object link.
1104 void endDebugObject();
1106 /// Keeps track of relocations.
1107 class RelocationManager {
1112 const DebugMapObject::DebugMapEntry *Mapping;
1114 ValidReloc(uint32_t Offset, uint32_t Size, uint64_t Addend,
1115 const DebugMapObject::DebugMapEntry *Mapping)
1116 : Offset(Offset), Size(Size), Addend(Addend), Mapping(Mapping) {}
1118 bool operator<(const ValidReloc &RHS) const {
1119 return Offset < RHS.Offset;
1123 DwarfLinker &Linker;
1125 /// \brief The valid relocations for the current DebugMapObject.
1126 /// This vector is sorted by relocation offset.
1127 std::vector<ValidReloc> ValidRelocs;
1129 /// \brief Index into ValidRelocs of the next relocation to
1130 /// consider. As we walk the DIEs in acsending file offset and as
1131 /// ValidRelocs is sorted by file offset, keeping this index
1132 /// uptodate is all we have to do to have a cheap lookup during the
1133 /// root DIE selection and during DIE cloning.
1134 unsigned NextValidReloc;
1137 RelocationManager(DwarfLinker &Linker)
1138 : Linker(Linker), NextValidReloc(0) {}
1140 bool hasValidRelocs() const { return !ValidRelocs.empty(); }
1141 /// \brief Reset the NextValidReloc counter.
1142 void resetValidRelocs() { NextValidReloc = 0; }
1144 /// \defgroup FindValidRelocations Translate debug map into a list
1145 /// of relevant relocations
1148 bool findValidRelocsInDebugInfo(const object::ObjectFile &Obj,
1149 const DebugMapObject &DMO);
1151 bool findValidRelocs(const object::SectionRef &Section,
1152 const object::ObjectFile &Obj,
1153 const DebugMapObject &DMO);
1155 void findValidRelocsMachO(const object::SectionRef &Section,
1156 const object::MachOObjectFile &Obj,
1157 const DebugMapObject &DMO);
1160 bool hasValidRelocation(uint32_t StartOffset, uint32_t EndOffset,
1161 CompileUnit::DIEInfo &Info);
1163 bool applyValidRelocs(MutableArrayRef<char> Data, uint32_t BaseOffset,
1164 bool isLittleEndian);
1167 /// \defgroup FindRootDIEs Find DIEs corresponding to debug map entries.
1170 /// \brief Recursively walk the \p DIE tree and look for DIEs to
1171 /// keep. Store that information in \p CU's DIEInfo.
1172 void lookForDIEsToKeep(RelocationManager &RelocMgr,
1173 const DWARFDebugInfoEntryMinimal &DIE,
1174 const DebugMapObject &DMO, CompileUnit &CU,
1177 /// \brief Flags passed to DwarfLinker::lookForDIEsToKeep
1178 enum TravesalFlags {
1179 TF_Keep = 1 << 0, ///< Mark the traversed DIEs as kept.
1180 TF_InFunctionScope = 1 << 1, ///< Current scope is a fucntion scope.
1181 TF_DependencyWalk = 1 << 2, ///< Walking the dependencies of a kept DIE.
1182 TF_ParentWalk = 1 << 3, ///< Walking up the parents of a kept DIE.
1183 TF_ODR = 1 << 4, ///< Use the ODR whhile keeping dependants.
1184 TF_SkipPC = 1 << 5, ///< Skip all location attributes.
1187 /// \brief Mark the passed DIE as well as all the ones it depends on
1189 void keepDIEAndDenpendencies(RelocationManager &RelocMgr,
1190 const DWARFDebugInfoEntryMinimal &DIE,
1191 CompileUnit::DIEInfo &MyInfo,
1192 const DebugMapObject &DMO, CompileUnit &CU,
1195 unsigned shouldKeepDIE(RelocationManager &RelocMgr,
1196 const DWARFDebugInfoEntryMinimal &DIE,
1197 CompileUnit &Unit, CompileUnit::DIEInfo &MyInfo,
1200 unsigned shouldKeepVariableDIE(RelocationManager &RelocMgr,
1201 const DWARFDebugInfoEntryMinimal &DIE,
1203 CompileUnit::DIEInfo &MyInfo, unsigned Flags);
1205 unsigned shouldKeepSubprogramDIE(RelocationManager &RelocMgr,
1206 const DWARFDebugInfoEntryMinimal &DIE,
1208 CompileUnit::DIEInfo &MyInfo,
1211 bool hasValidRelocation(uint32_t StartOffset, uint32_t EndOffset,
1212 CompileUnit::DIEInfo &Info);
1215 /// \defgroup Linking Methods used to link the debug information
1220 DwarfLinker &Linker;
1221 RelocationManager &RelocMgr;
1222 /// Allocator used for all the DIEValue objects.
1223 BumpPtrAllocator &DIEAlloc;
1224 MutableArrayRef<CompileUnit> CompileUnits;
1225 LinkOptions Options;
1228 DIECloner(DwarfLinker &Linker, RelocationManager &RelocMgr,
1229 BumpPtrAllocator &DIEAlloc,
1230 MutableArrayRef<CompileUnit> CompileUnits, LinkOptions &Options)
1231 : Linker(Linker), RelocMgr(RelocMgr), DIEAlloc(DIEAlloc),
1232 CompileUnits(CompileUnits), Options(Options) {}
1234 /// Recursively clone \p InputDIE into an tree of DIE objects
1235 /// where useless (as decided by lookForDIEsToKeep()) bits have been
1236 /// stripped out and addresses have been rewritten according to the
1239 /// \param OutOffset is the offset the cloned DIE in the output
1241 /// \param PCOffset (while cloning a function scope) is the offset
1242 /// applied to the entry point of the function to get the linked address.
1244 /// \returns the root of the cloned tree or null if nothing was selected.
1245 DIE *cloneDIE(RelocationManager &RelocMgr,
1246 const DWARFDebugInfoEntryMinimal &InputDIE, CompileUnit &U,
1247 int64_t PCOffset, uint32_t OutOffset, unsigned Flags);
1249 /// Construct the output DIE tree by cloning the DIEs we
1250 /// chose to keep above. If there are no valid relocs, then there's
1251 /// nothing to clone/emit.
1252 void cloneAllCompileUnits(DWARFContextInMemory &DwarfContext);
1255 typedef DWARFAbbreviationDeclaration::AttributeSpec AttributeSpec;
1257 /// Information gathered and exchanged between the various
1258 /// clone*Attributes helpers about the attributes of a particular DIE.
1259 struct AttributesInfo {
1260 const char *Name, *MangledName; ///< Names.
1261 uint32_t NameOffset, MangledNameOffset; ///< Offsets in the string pool.
1263 uint64_t OrigLowPc; ///< Value of AT_low_pc in the input DIE
1264 uint64_t OrigHighPc; ///< Value of AT_high_pc in the input DIE
1265 int64_t PCOffset; ///< Offset to apply to PC addresses inside a function.
1267 bool HasLowPc; ///< Does the DIE have a low_pc attribute?
1268 bool IsDeclaration; ///< Is this DIE only a declaration?
1271 : Name(nullptr), MangledName(nullptr), NameOffset(0),
1272 MangledNameOffset(0), OrigLowPc(UINT64_MAX), OrigHighPc(0),
1273 PCOffset(0), HasLowPc(false), IsDeclaration(false) {}
1276 /// Helper for cloneDIE.
1277 unsigned cloneAttribute(DIE &Die,
1278 const DWARFDebugInfoEntryMinimal &InputDIE,
1279 CompileUnit &U, const DWARFFormValue &Val,
1280 const AttributeSpec AttrSpec, unsigned AttrSize,
1281 AttributesInfo &AttrInfo);
1283 /// Clone a string attribute described by \p AttrSpec and add
1285 /// \returns the size of the new attribute.
1286 unsigned cloneStringAttribute(DIE &Die, AttributeSpec AttrSpec,
1287 const DWARFFormValue &Val,
1288 const DWARFUnit &U);
1290 /// Clone an attribute referencing another DIE and add
1292 /// \returns the size of the new attribute.
1294 cloneDieReferenceAttribute(DIE &Die,
1295 const DWARFDebugInfoEntryMinimal &InputDIE,
1296 AttributeSpec AttrSpec, unsigned AttrSize,
1297 const DWARFFormValue &Val, CompileUnit &Unit);
1299 /// Clone an attribute referencing another DIE and add
1301 /// \returns the size of the new attribute.
1302 unsigned cloneBlockAttribute(DIE &Die, AttributeSpec AttrSpec,
1303 const DWARFFormValue &Val, unsigned AttrSize);
1305 /// Clone an attribute referencing another DIE and add
1307 /// \returns the size of the new attribute.
1308 unsigned cloneAddressAttribute(DIE &Die, AttributeSpec AttrSpec,
1309 const DWARFFormValue &Val,
1310 const CompileUnit &Unit,
1311 AttributesInfo &Info);
1313 /// Clone a scalar attribute and add it to \p Die.
1314 /// \returns the size of the new attribute.
1315 unsigned cloneScalarAttribute(DIE &Die,
1316 const DWARFDebugInfoEntryMinimal &InputDIE,
1317 CompileUnit &U, AttributeSpec AttrSpec,
1318 const DWARFFormValue &Val, unsigned AttrSize,
1319 AttributesInfo &Info);
1321 /// Get the potential name and mangled name for the entity
1322 /// described by \p Die and store them in \Info if they are not
1324 /// \returns is a name was found.
1325 bool getDIENames(const DWARFDebugInfoEntryMinimal &Die, DWARFUnit &U,
1326 AttributesInfo &Info);
1328 /// Create a copy of abbreviation Abbrev.
1329 void copyAbbrev(const DWARFAbbreviationDeclaration &Abbrev, bool hasODR);
1332 /// \brief Assign an abbreviation number to \p Abbrev
1333 void AssignAbbrev(DIEAbbrev &Abbrev);
1335 /// \brief FoldingSet that uniques the abbreviations.
1336 FoldingSet<DIEAbbrev> AbbreviationsSet;
1337 /// \brief Storage for the unique Abbreviations.
1338 /// This is passed to AsmPrinter::emitDwarfAbbrevs(), thus it cannot
1339 /// be changed to a vecot of unique_ptrs.
1340 std::vector<DIEAbbrev *> Abbreviations;
1342 /// \brief Compute and emit debug_ranges section for \p Unit, and
1343 /// patch the attributes referencing it.
1344 void patchRangesForUnit(const CompileUnit &Unit, DWARFContext &Dwarf) const;
1346 /// \brief Generate and emit the DW_AT_ranges attribute for a
1347 /// compile_unit if it had one.
1348 void generateUnitRanges(CompileUnit &Unit) const;
1350 /// \brief Extract the line tables fromt he original dwarf, extract
1351 /// the relevant parts according to the linked function ranges and
1352 /// emit the result in the debug_line section.
1353 void patchLineTableForUnit(CompileUnit &Unit, DWARFContext &OrigDwarf);
1355 /// \brief Emit the accelerator entries for \p Unit.
1356 void emitAcceleratorEntriesForUnit(CompileUnit &Unit);
1358 /// \brief Patch the frame info for an object file and emit it.
1359 void patchFrameInfoForObject(const DebugMapObject &, DWARFContext &,
1360 unsigned AddressSize);
1362 /// \brief DIELoc objects that need to be destructed (but not freed!).
1363 std::vector<DIELoc *> DIELocs;
1364 /// \brief DIEBlock objects that need to be destructed (but not freed!).
1365 std::vector<DIEBlock *> DIEBlocks;
1366 /// \brief Allocator used for all the DIEValue objects.
1367 BumpPtrAllocator DIEAlloc;
1370 /// ODR Contexts for that link.
1371 DeclContextTree ODRContexts;
1373 /// \defgroup Helpers Various helper methods.
1376 const DWARFDebugInfoEntryMinimal *
1377 resolveDIEReference(const DWARFFormValue &RefValue, const DWARFUnit &Unit,
1378 const DWARFDebugInfoEntryMinimal &DIE,
1379 CompileUnit *&ReferencedCU);
1381 CompileUnit *getUnitForOffset(unsigned Offset);
1383 void reportWarning(const Twine &Warning, const DWARFUnit *Unit = nullptr,
1384 const DWARFDebugInfoEntryMinimal *DIE = nullptr) const;
1386 bool createStreamer(Triple TheTriple, StringRef OutputFilename);
1388 /// \brief Attempt to load a debug object from disk.
1389 ErrorOr<const object::ObjectFile &> loadObject(BinaryHolder &BinaryHolder,
1390 DebugMapObject &Obj,
1391 const DebugMap &Map);
1395 std::string OutputFilename;
1396 LinkOptions Options;
1397 BinaryHolder BinHolder;
1398 std::unique_ptr<DwarfStreamer> Streamer;
1399 uint64_t OutputDebugInfoSize;
1401 /// The units of the current debug map object.
1402 std::vector<CompileUnit> Units;
1404 /// The debug map object curently under consideration.
1405 DebugMapObject *CurrentDebugObject;
1407 /// \brief The Dwarf string pool
1408 NonRelocatableStringpool StringPool;
1410 /// \brief This map is keyed by the entry PC of functions in that
1411 /// debug object and the associated value is a pair storing the
1412 /// corresponding end PC and the offset to apply to get the linked
1415 /// See startDebugObject() for a more complete description of its use.
1416 std::map<uint64_t, std::pair<uint64_t, int64_t>> Ranges;
1418 /// \brief The CIEs that have been emitted in the output
1419 /// section. The actual CIE data serves a the key to this StringMap,
1420 /// this takes care of comparing the semantics of CIEs defined in
1421 /// different object files.
1422 StringMap<uint32_t> EmittedCIEs;
1424 /// Offset of the last CIE that has been emitted in the output
1425 /// debug_frame section.
1426 uint32_t LastCIEOffset;
1429 /// \brief Similar to DWARFUnitSection::getUnitForOffset(), but
1430 /// returning our CompileUnit object instead.
1431 CompileUnit *DwarfLinker::getUnitForOffset(unsigned Offset) {
1433 std::upper_bound(Units.begin(), Units.end(), Offset,
1434 [](uint32_t LHS, const CompileUnit &RHS) {
1435 return LHS < RHS.getOrigUnit().getNextUnitOffset();
1437 return CU != Units.end() ? &*CU : nullptr;
1440 /// \brief Resolve the DIE attribute reference that has been
1441 /// extracted in \p RefValue. The resulting DIE migh be in another
1442 /// CompileUnit which is stored into \p ReferencedCU.
1443 /// \returns null if resolving fails for any reason.
1444 const DWARFDebugInfoEntryMinimal *DwarfLinker::resolveDIEReference(
1445 const DWARFFormValue &RefValue, const DWARFUnit &Unit,
1446 const DWARFDebugInfoEntryMinimal &DIE, CompileUnit *&RefCU) {
1447 assert(RefValue.isFormClass(DWARFFormValue::FC_Reference));
1448 uint64_t RefOffset = *RefValue.getAsReference(&Unit);
1450 if ((RefCU = getUnitForOffset(RefOffset)))
1451 if (const auto *RefDie = RefCU->getOrigUnit().getDIEForOffset(RefOffset))
1454 reportWarning("could not find referenced DIE", &Unit, &DIE);
1458 /// \returns whether the passed \a Attr type might contain a DIE
1459 /// reference suitable for ODR uniquing.
1460 static bool isODRAttribute(uint16_t Attr) {
1464 case dwarf::DW_AT_type:
1465 case dwarf::DW_AT_containing_type:
1466 case dwarf::DW_AT_specification:
1467 case dwarf::DW_AT_abstract_origin:
1468 case dwarf::DW_AT_import:
1471 llvm_unreachable("Improper attribute.");
1474 /// Set the last DIE/CU a context was seen in and, possibly invalidate
1475 /// the context if it is ambiguous.
1477 /// In the current implementation, we don't handle overloaded
1478 /// functions well, because the argument types are not taken into
1479 /// account when computing the DeclContext tree.
1481 /// Some of this is mitigated byt using mangled names that do contain
1482 /// the arguments types, but sometimes (eg. with function templates)
1483 /// we don't have that. In that case, just do not unique anything that
1484 /// refers to the contexts we are not able to distinguish.
1486 /// If a context that is not a namespace appears twice in the same CU,
1487 /// we know it is ambiguous. Make it invalid.
1488 bool DeclContext::setLastSeenDIE(CompileUnit &U,
1489 const DWARFDebugInfoEntryMinimal *Die) {
1490 if (LastSeenCompileUnitID == U.getUniqueID()) {
1491 DWARFUnit &OrigUnit = U.getOrigUnit();
1492 uint32_t FirstIdx = OrigUnit.getDIEIndex(LastSeenDIE);
1493 U.getInfo(FirstIdx).Ctxt = nullptr;
1497 LastSeenCompileUnitID = U.getUniqueID();
1502 /// Get the child context of \a Context corresponding to \a DIE.
1504 /// \returns the child context or null if we shouldn't track children
1505 /// contexts. It also returns an additional bit meaning 'invalid'. An
1506 /// invalid context means it shouldn't be considered for uniquing, but
1507 /// its not returning null, because some children of that context
1508 /// might be uniquing candidates.
1509 /// FIXME: this is for dsymutil-classic compatibility, I don't think
1510 /// it buys us much.
1511 PointerIntPair<DeclContext *, 1> DeclContextTree::getChildDeclContext(
1512 DeclContext &Context, const DWARFDebugInfoEntryMinimal *DIE, CompileUnit &U,
1513 NonRelocatableStringpool &StringPool) {
1514 unsigned Tag = DIE->getTag();
1516 // FIXME: dsymutil-classic compat: We should bail out here if we
1517 // have a specification or an abstract_origin. We will get the
1518 // parent context wrong here.
1522 // By default stop gathering child contexts.
1523 return PointerIntPair<DeclContext *, 1>(nullptr);
1524 case dwarf::DW_TAG_compile_unit:
1525 // FIXME: Add support for DW_TAG_module.
1526 return PointerIntPair<DeclContext *, 1>(&Context);
1527 case dwarf::DW_TAG_subprogram:
1528 // Do not unique anything inside CU local functions.
1529 if ((Context.getTag() == dwarf::DW_TAG_namespace ||
1530 Context.getTag() == dwarf::DW_TAG_compile_unit) &&
1531 !DIE->getAttributeValueAsUnsignedConstant(&U.getOrigUnit(),
1532 dwarf::DW_AT_external, 0))
1533 return PointerIntPair<DeclContext *, 1>(nullptr);
1535 case dwarf::DW_TAG_member:
1536 case dwarf::DW_TAG_namespace:
1537 case dwarf::DW_TAG_structure_type:
1538 case dwarf::DW_TAG_class_type:
1539 case dwarf::DW_TAG_union_type:
1540 case dwarf::DW_TAG_enumeration_type:
1541 case dwarf::DW_TAG_typedef:
1542 // Artificial things might be ambiguous, because they might be
1543 // created on demand. For example implicitely defined constructors
1544 // are ambiguous because of the way we identify contexts, and they
1545 // won't be generated everytime everywhere.
1546 if (DIE->getAttributeValueAsUnsignedConstant(&U.getOrigUnit(),
1547 dwarf::DW_AT_artificial, 0))
1548 return PointerIntPair<DeclContext *, 1>(nullptr);
1552 const char *Name = DIE->getName(&U.getOrigUnit(), DINameKind::LinkageName);
1553 const char *ShortName = DIE->getName(&U.getOrigUnit(), DINameKind::ShortName);
1555 StringRef ShortNameRef;
1559 NameRef = StringPool.internString(Name);
1560 else if (Tag == dwarf::DW_TAG_namespace)
1561 // FIXME: For dsymutil-classic compatibility. I think uniquing
1562 // within anonymous namespaces is wrong. There is no ODR guarantee
1564 NameRef = StringPool.internString("(anonymous namespace)");
1566 if (ShortName && ShortName != Name)
1567 ShortNameRef = StringPool.internString(ShortName);
1569 ShortNameRef = NameRef;
1571 if (Tag != dwarf::DW_TAG_class_type && Tag != dwarf::DW_TAG_structure_type &&
1572 Tag != dwarf::DW_TAG_union_type &&
1573 Tag != dwarf::DW_TAG_enumeration_type && NameRef.empty())
1574 return PointerIntPair<DeclContext *, 1>(nullptr);
1578 unsigned ByteSize = 0;
1580 // Gather some discriminating data about the DeclContext we will be
1581 // creating: File, line number and byte size. This shouldn't be
1582 // necessary, because the ODR is just about names, but given that we
1583 // do some approximations with overloaded functions and anonymous
1584 // namespaces, use these additional data points to make the process safer.
1585 ByteSize = DIE->getAttributeValueAsUnsignedConstant(
1586 &U.getOrigUnit(), dwarf::DW_AT_byte_size, UINT64_MAX);
1587 if (Tag != dwarf::DW_TAG_namespace || !Name) {
1588 if (unsigned FileNum = DIE->getAttributeValueAsUnsignedConstant(
1589 &U.getOrigUnit(), dwarf::DW_AT_decl_file, 0)) {
1590 if (const auto *LT = U.getOrigUnit().getContext().getLineTableForUnit(
1591 &U.getOrigUnit())) {
1592 // FIXME: dsymutil-classic compatibility. I'd rather not
1593 // unique anything in anonymous namespaces, but if we do, then
1594 // verify that the file and line correspond.
1595 if (!Name && Tag == dwarf::DW_TAG_namespace)
1598 // FIXME: Passing U.getOrigUnit().getCompilationDir()
1599 // instead of "" would allow more uniquing, but for now, do
1600 // it this way to match dsymutil-classic.
1601 if (LT->getFileNameByIndex(
1603 DILineInfoSpecifier::FileLineInfoKind::AbsoluteFilePath,
1605 Line = DIE->getAttributeValueAsUnsignedConstant(
1606 &U.getOrigUnit(), dwarf::DW_AT_decl_line, 0);
1607 #ifdef HAVE_REALPATH
1608 // Cache the resolved paths, because calling realpath is expansive.
1609 if (const char *ResolvedPath = U.getResolvedPath(FileNum)) {
1610 File = ResolvedPath;
1612 char RealPath[PATH_MAX + 1];
1613 RealPath[PATH_MAX] = 0;
1614 if (::realpath(File.c_str(), RealPath))
1616 U.setResolvedPath(FileNum, File);
1619 FileRef = StringPool.internString(File);
1625 if (!Line && NameRef.empty())
1626 return PointerIntPair<DeclContext *, 1>(nullptr);
1628 // FIXME: dsymutil-classic compat won't unique the same type
1629 // presented once as a struct and once as a class. Use the Tag in
1630 // the fully qualified name hash to get the same effect.
1631 // We hash NameRef, which is the mangled name, in order to get most
1632 // overloaded functions resolvec correctly.
1633 unsigned Hash = hash_combine(Context.getQualifiedNameHash(), Tag, NameRef);
1635 // FIXME: dsymutil-classic compatibility: when we don't have a name,
1636 // use the filename.
1637 if (Tag == dwarf::DW_TAG_namespace && NameRef == "(anonymous namespace)")
1638 Hash = hash_combine(Hash, FileRef);
1640 // Now look if this context already exists.
1641 DeclContext Key(Hash, Line, ByteSize, Tag, NameRef, FileRef, Context);
1642 auto ContextIter = Contexts.find(&Key);
1644 if (ContextIter == Contexts.end()) {
1645 // The context wasn't found.
1647 DeclContext *NewContext =
1648 new (Allocator) DeclContext(Hash, Line, ByteSize, Tag, NameRef, FileRef,
1649 Context, DIE, U.getUniqueID());
1650 std::tie(ContextIter, Inserted) = Contexts.insert(NewContext);
1651 assert(Inserted && "Failed to insert DeclContext");
1653 } else if (Tag != dwarf::DW_TAG_namespace &&
1654 !(*ContextIter)->setLastSeenDIE(U, DIE)) {
1655 // The context was found, but it is ambiguous with another context
1656 // in the same file. Mark it invalid.
1657 return PointerIntPair<DeclContext *, 1>(*ContextIter, /* Invalid= */ 1);
1660 assert(ContextIter != Contexts.end());
1661 // FIXME: dsymutil-classic compatibility. Union types aren't
1662 // uniques, but their children might be.
1663 if ((Tag == dwarf::DW_TAG_subprogram &&
1664 Context.getTag() != dwarf::DW_TAG_structure_type &&
1665 Context.getTag() != dwarf::DW_TAG_class_type) ||
1666 (Tag == dwarf::DW_TAG_union_type))
1667 return PointerIntPair<DeclContext *, 1>(*ContextIter, /* Invalid= */ 1);
1669 return PointerIntPair<DeclContext *, 1>(*ContextIter);
1672 bool DwarfLinker::DIECloner::getDIENames(const DWARFDebugInfoEntryMinimal &Die,
1673 DWARFUnit &U, AttributesInfo &Info) {
1674 // FIXME: a bit wasteful as the first getName might return the
1676 if (!Info.MangledName &&
1677 (Info.MangledName = Die.getName(&U, DINameKind::LinkageName)))
1678 Info.MangledNameOffset =
1679 Linker.StringPool.getStringOffset(Info.MangledName);
1681 if (!Info.Name && (Info.Name = Die.getName(&U, DINameKind::ShortName)))
1682 Info.NameOffset = Linker.StringPool.getStringOffset(Info.Name);
1684 return Info.Name || Info.MangledName;
1687 /// \brief Report a warning to the user, optionaly including
1688 /// information about a specific \p DIE related to the warning.
1689 void DwarfLinker::reportWarning(const Twine &Warning, const DWARFUnit *Unit,
1690 const DWARFDebugInfoEntryMinimal *DIE) const {
1691 StringRef Context = "<debug map>";
1692 if (CurrentDebugObject)
1693 Context = CurrentDebugObject->getObjectFilename();
1694 warn(Warning, Context);
1696 if (!Options.Verbose || !DIE)
1699 errs() << " in DIE:\n";
1700 DIE->dump(errs(), const_cast<DWARFUnit *>(Unit), 0 /* RecurseDepth */,
1704 bool DwarfLinker::createStreamer(Triple TheTriple, StringRef OutputFilename) {
1705 if (Options.NoOutput)
1708 Streamer = llvm::make_unique<DwarfStreamer>();
1709 return Streamer->init(TheTriple, OutputFilename);
1712 /// \brief Recursive helper to gather the child->parent relationships in the
1713 /// original compile unit.
1714 static void gatherDIEParents(const DWARFDebugInfoEntryMinimal *DIE,
1715 unsigned ParentIdx, CompileUnit &CU,
1716 DeclContext *CurrentDeclContext,
1717 NonRelocatableStringpool &StringPool,
1718 DeclContextTree &Contexts) {
1719 unsigned MyIdx = CU.getOrigUnit().getDIEIndex(DIE);
1720 CompileUnit::DIEInfo &Info = CU.getInfo(MyIdx);
1722 Info.ParentIdx = ParentIdx;
1724 if (CurrentDeclContext) {
1725 auto PtrInvalidPair = Contexts.getChildDeclContext(*CurrentDeclContext,
1726 DIE, CU, StringPool);
1727 CurrentDeclContext = PtrInvalidPair.getPointer();
1729 PtrInvalidPair.getInt() ? nullptr : PtrInvalidPair.getPointer();
1731 Info.Ctxt = CurrentDeclContext = nullptr;
1734 if (DIE->hasChildren())
1735 for (auto *Child = DIE->getFirstChild(); Child && !Child->isNULL();
1736 Child = Child->getSibling())
1737 gatherDIEParents(Child, MyIdx, CU, CurrentDeclContext, StringPool,
1741 static bool dieNeedsChildrenToBeMeaningful(uint32_t Tag) {
1745 case dwarf::DW_TAG_subprogram:
1746 case dwarf::DW_TAG_lexical_block:
1747 case dwarf::DW_TAG_subroutine_type:
1748 case dwarf::DW_TAG_structure_type:
1749 case dwarf::DW_TAG_class_type:
1750 case dwarf::DW_TAG_union_type:
1753 llvm_unreachable("Invalid Tag");
1756 static unsigned getRefAddrSize(const DWARFUnit &U) {
1757 if (U.getVersion() == 2)
1758 return U.getAddressByteSize();
1762 void DwarfLinker::startDebugObject(DWARFContext &Dwarf, DebugMapObject &Obj) {
1763 Units.reserve(Dwarf.getNumCompileUnits());
1764 // Iterate over the debug map entries and put all the ones that are
1765 // functions (because they have a size) into the Ranges map. This
1766 // map is very similar to the FunctionRanges that are stored in each
1767 // unit, with 2 notable differences:
1768 // - obviously this one is global, while the other ones are per-unit.
1769 // - this one contains not only the functions described in the DIE
1770 // tree, but also the ones that are only in the debug map.
1771 // The latter information is required to reproduce dsymutil's logic
1772 // while linking line tables. The cases where this information
1773 // matters look like bugs that need to be investigated, but for now
1774 // we need to reproduce dsymutil's behavior.
1775 // FIXME: Once we understood exactly if that information is needed,
1776 // maybe totally remove this (or try to use it to do a real
1777 // -gline-tables-only on Darwin.
1778 for (const auto &Entry : Obj.symbols()) {
1779 const auto &Mapping = Entry.getValue();
1781 Ranges[Mapping.ObjectAddress] = std::make_pair(
1782 Mapping.ObjectAddress + Mapping.Size,
1783 int64_t(Mapping.BinaryAddress) - Mapping.ObjectAddress);
1787 void DwarfLinker::endDebugObject() {
1791 for (auto I = DIEBlocks.begin(), E = DIEBlocks.end(); I != E; ++I)
1793 for (auto I = DIELocs.begin(), E = DIELocs.end(); I != E; ++I)
1801 /// \brief Iterate over the relocations of the given \p Section and
1802 /// store the ones that correspond to debug map entries into the
1803 /// ValidRelocs array.
1804 void DwarfLinker::RelocationManager::
1805 findValidRelocsMachO(const object::SectionRef &Section,
1806 const object::MachOObjectFile &Obj,
1807 const DebugMapObject &DMO) {
1809 Section.getContents(Contents);
1810 DataExtractor Data(Contents, Obj.isLittleEndian(), 0);
1812 for (const object::RelocationRef &Reloc : Section.relocations()) {
1813 object::DataRefImpl RelocDataRef = Reloc.getRawDataRefImpl();
1814 MachO::any_relocation_info MachOReloc = Obj.getRelocation(RelocDataRef);
1815 unsigned RelocSize = 1 << Obj.getAnyRelocationLength(MachOReloc);
1816 uint64_t Offset64 = Reloc.getOffset();
1817 if ((RelocSize != 4 && RelocSize != 8)) {
1818 Linker.reportWarning(" unsupported relocation in debug_info section.");
1821 uint32_t Offset = Offset64;
1822 // Mach-o uses REL relocations, the addend is at the relocation offset.
1823 uint64_t Addend = Data.getUnsigned(&Offset, RelocSize);
1825 auto Sym = Reloc.getSymbol();
1826 if (Sym != Obj.symbol_end()) {
1827 ErrorOr<StringRef> SymbolName = Sym->getName();
1829 Linker.reportWarning("error getting relocation symbol name.");
1832 if (const auto *Mapping = DMO.lookupSymbol(*SymbolName))
1833 ValidRelocs.emplace_back(Offset64, RelocSize, Addend, Mapping);
1834 } else if (const auto *Mapping = DMO.lookupObjectAddress(Addend)) {
1835 // Do not store the addend. The addend was the address of the
1836 // symbol in the object file, the address in the binary that is
1837 // stored in the debug map doesn't need to be offseted.
1838 ValidRelocs.emplace_back(Offset64, RelocSize, 0, Mapping);
1843 /// \brief Dispatch the valid relocation finding logic to the
1844 /// appropriate handler depending on the object file format.
1845 bool DwarfLinker::RelocationManager::findValidRelocs(
1846 const object::SectionRef &Section, const object::ObjectFile &Obj,
1847 const DebugMapObject &DMO) {
1848 // Dispatch to the right handler depending on the file type.
1849 if (auto *MachOObj = dyn_cast<object::MachOObjectFile>(&Obj))
1850 findValidRelocsMachO(Section, *MachOObj, DMO);
1852 Linker.reportWarning(Twine("unsupported object file type: ") +
1855 if (ValidRelocs.empty())
1858 // Sort the relocations by offset. We will walk the DIEs linearly in
1859 // the file, this allows us to just keep an index in the relocation
1860 // array that we advance during our walk, rather than resorting to
1861 // some associative container. See DwarfLinker::NextValidReloc.
1862 std::sort(ValidRelocs.begin(), ValidRelocs.end());
1866 /// \brief Look for relocations in the debug_info section that match
1867 /// entries in the debug map. These relocations will drive the Dwarf
1868 /// link by indicating which DIEs refer to symbols present in the
1870 /// \returns wether there are any valid relocations in the debug info.
1871 bool DwarfLinker::RelocationManager::
1872 findValidRelocsInDebugInfo(const object::ObjectFile &Obj,
1873 const DebugMapObject &DMO) {
1874 // Find the debug_info section.
1875 for (const object::SectionRef &Section : Obj.sections()) {
1876 StringRef SectionName;
1877 Section.getName(SectionName);
1878 SectionName = SectionName.substr(SectionName.find_first_not_of("._"));
1879 if (SectionName != "debug_info")
1881 return findValidRelocs(Section, Obj, DMO);
1886 /// \brief Checks that there is a relocation against an actual debug
1887 /// map entry between \p StartOffset and \p NextOffset.
1889 /// This function must be called with offsets in strictly ascending
1890 /// order because it never looks back at relocations it already 'went past'.
1891 /// \returns true and sets Info.InDebugMap if it is the case.
1892 bool DwarfLinker::RelocationManager::
1893 hasValidRelocation(uint32_t StartOffset, uint32_t EndOffset,
1894 CompileUnit::DIEInfo &Info) {
1895 assert(NextValidReloc == 0 ||
1896 StartOffset > ValidRelocs[NextValidReloc - 1].Offset);
1897 if (NextValidReloc >= ValidRelocs.size())
1900 uint64_t RelocOffset = ValidRelocs[NextValidReloc].Offset;
1902 // We might need to skip some relocs that we didn't consider. For
1903 // example the high_pc of a discarded DIE might contain a reloc that
1904 // is in the list because it actually corresponds to the start of a
1905 // function that is in the debug map.
1906 while (RelocOffset < StartOffset && NextValidReloc < ValidRelocs.size() - 1)
1907 RelocOffset = ValidRelocs[++NextValidReloc].Offset;
1909 if (RelocOffset < StartOffset || RelocOffset >= EndOffset)
1912 const auto &ValidReloc = ValidRelocs[NextValidReloc++];
1913 const auto &Mapping = ValidReloc.Mapping->getValue();
1914 if (Linker.Options.Verbose)
1915 outs() << "Found valid debug map entry: " << ValidReloc.Mapping->getKey()
1916 << " " << format("\t%016" PRIx64 " => %016" PRIx64,
1917 uint64_t(Mapping.ObjectAddress),
1918 uint64_t(Mapping.BinaryAddress));
1920 Info.AddrAdjust = int64_t(Mapping.BinaryAddress) + ValidReloc.Addend -
1921 Mapping.ObjectAddress;
1922 Info.InDebugMap = true;
1926 /// \brief Get the starting and ending (exclusive) offset for the
1927 /// attribute with index \p Idx descibed by \p Abbrev. \p Offset is
1928 /// supposed to point to the position of the first attribute described
1930 /// \return [StartOffset, EndOffset) as a pair.
1931 static std::pair<uint32_t, uint32_t>
1932 getAttributeOffsets(const DWARFAbbreviationDeclaration *Abbrev, unsigned Idx,
1933 unsigned Offset, const DWARFUnit &Unit) {
1934 DataExtractor Data = Unit.getDebugInfoExtractor();
1936 for (unsigned i = 0; i < Idx; ++i)
1937 DWARFFormValue::skipValue(Abbrev->getFormByIndex(i), Data, &Offset, &Unit);
1939 uint32_t End = Offset;
1940 DWARFFormValue::skipValue(Abbrev->getFormByIndex(Idx), Data, &End, &Unit);
1942 return std::make_pair(Offset, End);
1945 /// \brief Check if a variable describing DIE should be kept.
1946 /// \returns updated TraversalFlags.
1947 unsigned DwarfLinker::shouldKeepVariableDIE(RelocationManager &RelocMgr,
1948 const DWARFDebugInfoEntryMinimal &DIE,
1950 CompileUnit::DIEInfo &MyInfo,
1952 const auto *Abbrev = DIE.getAbbreviationDeclarationPtr();
1954 // Global variables with constant value can always be kept.
1955 if (!(Flags & TF_InFunctionScope) &&
1956 Abbrev->findAttributeIndex(dwarf::DW_AT_const_value) != -1U) {
1957 MyInfo.InDebugMap = true;
1958 return Flags | TF_Keep;
1961 uint32_t LocationIdx = Abbrev->findAttributeIndex(dwarf::DW_AT_location);
1962 if (LocationIdx == -1U)
1965 uint32_t Offset = DIE.getOffset() + getULEB128Size(Abbrev->getCode());
1966 const DWARFUnit &OrigUnit = Unit.getOrigUnit();
1967 uint32_t LocationOffset, LocationEndOffset;
1968 std::tie(LocationOffset, LocationEndOffset) =
1969 getAttributeOffsets(Abbrev, LocationIdx, Offset, OrigUnit);
1971 // See if there is a relocation to a valid debug map entry inside
1972 // this variable's location. The order is important here. We want to
1973 // always check in the variable has a valid relocation, so that the
1974 // DIEInfo is filled. However, we don't want a static variable in a
1975 // function to force us to keep the enclosing function.
1976 if (!RelocMgr.hasValidRelocation(LocationOffset, LocationEndOffset, MyInfo) ||
1977 (Flags & TF_InFunctionScope))
1980 if (Options.Verbose)
1981 DIE.dump(outs(), const_cast<DWARFUnit *>(&OrigUnit), 0, 8 /* Indent */);
1983 return Flags | TF_Keep;
1986 /// \brief Check if a function describing DIE should be kept.
1987 /// \returns updated TraversalFlags.
1988 unsigned DwarfLinker::shouldKeepSubprogramDIE(
1989 RelocationManager &RelocMgr,
1990 const DWARFDebugInfoEntryMinimal &DIE, CompileUnit &Unit,
1991 CompileUnit::DIEInfo &MyInfo, unsigned Flags) {
1992 const auto *Abbrev = DIE.getAbbreviationDeclarationPtr();
1994 Flags |= TF_InFunctionScope;
1996 uint32_t LowPcIdx = Abbrev->findAttributeIndex(dwarf::DW_AT_low_pc);
1997 if (LowPcIdx == -1U)
2000 uint32_t Offset = DIE.getOffset() + getULEB128Size(Abbrev->getCode());
2001 const DWARFUnit &OrigUnit = Unit.getOrigUnit();
2002 uint32_t LowPcOffset, LowPcEndOffset;
2003 std::tie(LowPcOffset, LowPcEndOffset) =
2004 getAttributeOffsets(Abbrev, LowPcIdx, Offset, OrigUnit);
2007 DIE.getAttributeValueAsAddress(&OrigUnit, dwarf::DW_AT_low_pc, -1ULL);
2008 assert(LowPc != -1ULL && "low_pc attribute is not an address.");
2009 if (LowPc == -1ULL ||
2010 !RelocMgr.hasValidRelocation(LowPcOffset, LowPcEndOffset, MyInfo))
2013 if (Options.Verbose)
2014 DIE.dump(outs(), const_cast<DWARFUnit *>(&OrigUnit), 0, 8 /* Indent */);
2018 DWARFFormValue HighPcValue;
2019 if (!DIE.getAttributeValue(&OrigUnit, dwarf::DW_AT_high_pc, HighPcValue)) {
2020 reportWarning("Function without high_pc. Range will be discarded.\n",
2026 if (HighPcValue.isFormClass(DWARFFormValue::FC_Address)) {
2027 HighPc = *HighPcValue.getAsAddress(&OrigUnit);
2029 assert(HighPcValue.isFormClass(DWARFFormValue::FC_Constant));
2030 HighPc = LowPc + *HighPcValue.getAsUnsignedConstant();
2033 // Replace the debug map range with a more accurate one.
2034 Ranges[LowPc] = std::make_pair(HighPc, MyInfo.AddrAdjust);
2035 Unit.addFunctionRange(LowPc, HighPc, MyInfo.AddrAdjust);
2039 /// \brief Check if a DIE should be kept.
2040 /// \returns updated TraversalFlags.
2041 unsigned DwarfLinker::shouldKeepDIE(RelocationManager &RelocMgr,
2042 const DWARFDebugInfoEntryMinimal &DIE,
2044 CompileUnit::DIEInfo &MyInfo,
2046 switch (DIE.getTag()) {
2047 case dwarf::DW_TAG_constant:
2048 case dwarf::DW_TAG_variable:
2049 return shouldKeepVariableDIE(RelocMgr, DIE, Unit, MyInfo, Flags);
2050 case dwarf::DW_TAG_subprogram:
2051 return shouldKeepSubprogramDIE(RelocMgr, DIE, Unit, MyInfo, Flags);
2052 case dwarf::DW_TAG_module:
2053 case dwarf::DW_TAG_imported_module:
2054 case dwarf::DW_TAG_imported_declaration:
2055 case dwarf::DW_TAG_imported_unit:
2056 // We always want to keep these.
2057 return Flags | TF_Keep;
2063 /// \brief Mark the passed DIE as well as all the ones it depends on
2066 /// This function is called by lookForDIEsToKeep on DIEs that are
2067 /// newly discovered to be needed in the link. It recursively calls
2068 /// back to lookForDIEsToKeep while adding TF_DependencyWalk to the
2069 /// TraversalFlags to inform it that it's not doing the primary DIE
2071 void DwarfLinker::keepDIEAndDenpendencies(RelocationManager &RelocMgr,
2072 const DWARFDebugInfoEntryMinimal &Die,
2073 CompileUnit::DIEInfo &MyInfo,
2074 const DebugMapObject &DMO,
2075 CompileUnit &CU, bool UseODR) {
2076 const DWARFUnit &Unit = CU.getOrigUnit();
2079 // First mark all the parent chain as kept.
2080 unsigned AncestorIdx = MyInfo.ParentIdx;
2081 while (!CU.getInfo(AncestorIdx).Keep) {
2082 unsigned ODRFlag = UseODR ? TF_ODR : 0;
2083 lookForDIEsToKeep(RelocMgr, *Unit.getDIEAtIndex(AncestorIdx), DMO, CU,
2084 TF_ParentWalk | TF_Keep | TF_DependencyWalk | ODRFlag);
2085 AncestorIdx = CU.getInfo(AncestorIdx).ParentIdx;
2088 // Then we need to mark all the DIEs referenced by this DIE's
2089 // attributes as kept.
2090 DataExtractor Data = Unit.getDebugInfoExtractor();
2091 const auto *Abbrev = Die.getAbbreviationDeclarationPtr();
2092 uint32_t Offset = Die.getOffset() + getULEB128Size(Abbrev->getCode());
2094 // Mark all DIEs referenced through atttributes as kept.
2095 for (const auto &AttrSpec : Abbrev->attributes()) {
2096 DWARFFormValue Val(AttrSpec.Form);
2098 if (!Val.isFormClass(DWARFFormValue::FC_Reference)) {
2099 DWARFFormValue::skipValue(AttrSpec.Form, Data, &Offset, &Unit);
2103 Val.extractValue(Data, &Offset, &Unit);
2104 CompileUnit *ReferencedCU;
2105 if (const auto *RefDIE =
2106 resolveDIEReference(Val, Unit, Die, ReferencedCU)) {
2107 uint32_t RefIdx = ReferencedCU->getOrigUnit().getDIEIndex(RefDIE);
2108 CompileUnit::DIEInfo &Info = ReferencedCU->getInfo(RefIdx);
2109 // If the referenced DIE has a DeclContext that has already been
2110 // emitted, then do not keep the one in this CU. We'll link to
2111 // the canonical DIE in cloneDieReferenceAttribute.
2112 // FIXME: compatibility with dsymutil-classic. UseODR shouldn't
2113 // be necessary and could be advantageously replaced by
2114 // ReferencedCU->hasODR() && CU.hasODR().
2115 // FIXME: compatibility with dsymutil-classic. There is no
2116 // reason not to unique ref_addr references.
2117 if (AttrSpec.Form != dwarf::DW_FORM_ref_addr && UseODR && Info.Ctxt &&
2118 Info.Ctxt != ReferencedCU->getInfo(Info.ParentIdx).Ctxt &&
2119 Info.Ctxt->getCanonicalDIEOffset() && isODRAttribute(AttrSpec.Attr))
2122 unsigned ODRFlag = UseODR ? TF_ODR : 0;
2123 lookForDIEsToKeep(RelocMgr, *RefDIE, DMO, *ReferencedCU,
2124 TF_Keep | TF_DependencyWalk | ODRFlag);
2129 /// \brief Recursively walk the \p DIE tree and look for DIEs to
2130 /// keep. Store that information in \p CU's DIEInfo.
2132 /// This function is the entry point of the DIE selection
2133 /// algorithm. It is expected to walk the DIE tree in file order and
2134 /// (though the mediation of its helper) call hasValidRelocation() on
2135 /// each DIE that might be a 'root DIE' (See DwarfLinker class
2137 /// While walking the dependencies of root DIEs, this function is
2138 /// also called, but during these dependency walks the file order is
2139 /// not respected. The TF_DependencyWalk flag tells us which kind of
2140 /// traversal we are currently doing.
2141 void DwarfLinker::lookForDIEsToKeep(RelocationManager &RelocMgr,
2142 const DWARFDebugInfoEntryMinimal &Die,
2143 const DebugMapObject &DMO, CompileUnit &CU,
2145 unsigned Idx = CU.getOrigUnit().getDIEIndex(&Die);
2146 CompileUnit::DIEInfo &MyInfo = CU.getInfo(Idx);
2147 bool AlreadyKept = MyInfo.Keep;
2149 // If the Keep flag is set, we are marking a required DIE's
2150 // dependencies. If our target is already marked as kept, we're all
2152 if ((Flags & TF_DependencyWalk) && AlreadyKept)
2155 // We must not call shouldKeepDIE while called from keepDIEAndDenpendencies,
2156 // because it would screw up the relocation finding logic.
2157 if (!(Flags & TF_DependencyWalk))
2158 Flags = shouldKeepDIE(RelocMgr, Die, CU, MyInfo, Flags);
2160 // If it is a newly kept DIE mark it as well as all its dependencies as kept.
2161 if (!AlreadyKept && (Flags & TF_Keep)) {
2162 bool UseOdr = (Flags & TF_DependencyWalk) ? (Flags & TF_ODR) : CU.hasODR();
2163 keepDIEAndDenpendencies(RelocMgr, Die, MyInfo, DMO, CU, UseOdr);
2165 // The TF_ParentWalk flag tells us that we are currently walking up
2166 // the parent chain of a required DIE, and we don't want to mark all
2167 // the children of the parents as kept (consider for example a
2168 // DW_TAG_namespace node in the parent chain). There are however a
2169 // set of DIE types for which we want to ignore that directive and still
2170 // walk their children.
2171 if (dieNeedsChildrenToBeMeaningful(Die.getTag()))
2172 Flags &= ~TF_ParentWalk;
2174 if (!Die.hasChildren() || (Flags & TF_ParentWalk))
2177 for (auto *Child = Die.getFirstChild(); Child && !Child->isNULL();
2178 Child = Child->getSibling())
2179 lookForDIEsToKeep(RelocMgr, *Child, DMO, CU, Flags);
2182 /// \brief Assign an abbreviation numer to \p Abbrev.
2184 /// Our DIEs get freed after every DebugMapObject has been processed,
2185 /// thus the FoldingSet we use to unique DIEAbbrevs cannot refer to
2186 /// the instances hold by the DIEs. When we encounter an abbreviation
2187 /// that we don't know, we create a permanent copy of it.
2188 void DwarfLinker::AssignAbbrev(DIEAbbrev &Abbrev) {
2189 // Check the set for priors.
2190 FoldingSetNodeID ID;
2193 DIEAbbrev *InSet = AbbreviationsSet.FindNodeOrInsertPos(ID, InsertToken);
2195 // If it's newly added.
2197 // Assign existing abbreviation number.
2198 Abbrev.setNumber(InSet->getNumber());
2200 // Add to abbreviation list.
2201 Abbreviations.push_back(
2202 new DIEAbbrev(Abbrev.getTag(), Abbrev.hasChildren()));
2203 for (const auto &Attr : Abbrev.getData())
2204 Abbreviations.back()->AddAttribute(Attr.getAttribute(), Attr.getForm());
2205 AbbreviationsSet.InsertNode(Abbreviations.back(), InsertToken);
2206 // Assign the unique abbreviation number.
2207 Abbrev.setNumber(Abbreviations.size());
2208 Abbreviations.back()->setNumber(Abbreviations.size());
2212 unsigned DwarfLinker::DIECloner::cloneStringAttribute(DIE &Die,
2213 AttributeSpec AttrSpec,
2214 const DWARFFormValue &Val,
2215 const DWARFUnit &U) {
2216 // Switch everything to out of line strings.
2217 const char *String = *Val.getAsCString(&U);
2218 unsigned Offset = Linker.StringPool.getStringOffset(String);
2219 Die.addValue(DIEAlloc, dwarf::Attribute(AttrSpec.Attr), dwarf::DW_FORM_strp,
2220 DIEInteger(Offset));
2224 unsigned DwarfLinker::DIECloner::cloneDieReferenceAttribute(
2225 DIE &Die, const DWARFDebugInfoEntryMinimal &InputDIE,
2226 AttributeSpec AttrSpec, unsigned AttrSize, const DWARFFormValue &Val,
2227 CompileUnit &Unit) {
2228 const DWARFUnit &U = Unit.getOrigUnit();
2229 uint32_t Ref = *Val.getAsReference(&U);
2230 DIE *NewRefDie = nullptr;
2231 CompileUnit *RefUnit = nullptr;
2232 DeclContext *Ctxt = nullptr;
2234 const DWARFDebugInfoEntryMinimal *RefDie =
2235 Linker.resolveDIEReference(Val, U, InputDIE, RefUnit);
2237 // If the referenced DIE is not found, drop the attribute.
2241 unsigned Idx = RefUnit->getOrigUnit().getDIEIndex(RefDie);
2242 CompileUnit::DIEInfo &RefInfo = RefUnit->getInfo(Idx);
2244 // If we already have emitted an equivalent DeclContext, just point
2246 if (isODRAttribute(AttrSpec.Attr)) {
2247 Ctxt = RefInfo.Ctxt;
2248 if (Ctxt && Ctxt->getCanonicalDIEOffset()) {
2249 DIEInteger Attr(Ctxt->getCanonicalDIEOffset());
2250 Die.addValue(DIEAlloc, dwarf::Attribute(AttrSpec.Attr),
2251 dwarf::DW_FORM_ref_addr, Attr);
2252 return getRefAddrSize(U);
2256 if (!RefInfo.Clone) {
2257 assert(Ref > InputDIE.getOffset());
2258 // We haven't cloned this DIE yet. Just create an empty one and
2259 // store it. It'll get really cloned when we process it.
2260 RefInfo.Clone = DIE::get(DIEAlloc, dwarf::Tag(RefDie->getTag()));
2262 NewRefDie = RefInfo.Clone;
2264 if (AttrSpec.Form == dwarf::DW_FORM_ref_addr ||
2265 (Unit.hasODR() && isODRAttribute(AttrSpec.Attr))) {
2266 // We cannot currently rely on a DIEEntry to emit ref_addr
2267 // references, because the implementation calls back to DwarfDebug
2268 // to find the unit offset. (We don't have a DwarfDebug)
2269 // FIXME: we should be able to design DIEEntry reliance on
2272 if (Ref < InputDIE.getOffset()) {
2273 // We must have already cloned that DIE.
2274 uint32_t NewRefOffset =
2275 RefUnit->getStartOffset() + NewRefDie->getOffset();
2276 Attr = NewRefOffset;
2277 Die.addValue(DIEAlloc, dwarf::Attribute(AttrSpec.Attr),
2278 dwarf::DW_FORM_ref_addr, DIEInteger(Attr));
2280 // A forward reference. Note and fixup later.
2282 Unit.noteForwardReference(
2283 NewRefDie, RefUnit, Ctxt,
2284 Die.addValue(DIEAlloc, dwarf::Attribute(AttrSpec.Attr),
2285 dwarf::DW_FORM_ref_addr, DIEInteger(Attr)));
2287 return getRefAddrSize(U);
2290 Die.addValue(DIEAlloc, dwarf::Attribute(AttrSpec.Attr),
2291 dwarf::Form(AttrSpec.Form), DIEEntry(*NewRefDie));
2295 unsigned DwarfLinker::DIECloner::cloneBlockAttribute(DIE &Die,
2296 AttributeSpec AttrSpec,
2297 const DWARFFormValue &Val,
2298 unsigned AttrSize) {
2301 DIELoc *Loc = nullptr;
2302 DIEBlock *Block = nullptr;
2303 // Just copy the block data over.
2304 if (AttrSpec.Form == dwarf::DW_FORM_exprloc) {
2305 Loc = new (DIEAlloc) DIELoc;
2306 Linker.DIELocs.push_back(Loc);
2308 Block = new (DIEAlloc) DIEBlock;
2309 Linker.DIEBlocks.push_back(Block);
2311 Attr = Loc ? static_cast<DIEValueList *>(Loc)
2312 : static_cast<DIEValueList *>(Block);
2315 Value = DIEValue(dwarf::Attribute(AttrSpec.Attr),
2316 dwarf::Form(AttrSpec.Form), Loc);
2318 Value = DIEValue(dwarf::Attribute(AttrSpec.Attr),
2319 dwarf::Form(AttrSpec.Form), Block);
2320 ArrayRef<uint8_t> Bytes = *Val.getAsBlock();
2321 for (auto Byte : Bytes)
2322 Attr->addValue(DIEAlloc, static_cast<dwarf::Attribute>(0),
2323 dwarf::DW_FORM_data1, DIEInteger(Byte));
2324 // FIXME: If DIEBlock and DIELoc just reuses the Size field of
2325 // the DIE class, this if could be replaced by
2326 // Attr->setSize(Bytes.size()).
2327 if (Linker.Streamer) {
2328 auto *AsmPrinter = &Linker.Streamer->getAsmPrinter();
2330 Loc->ComputeSize(AsmPrinter);
2332 Block->ComputeSize(AsmPrinter);
2334 Die.addValue(DIEAlloc, Value);
2338 unsigned DwarfLinker::DIECloner::cloneAddressAttribute(
2339 DIE &Die, AttributeSpec AttrSpec, const DWARFFormValue &Val,
2340 const CompileUnit &Unit, AttributesInfo &Info) {
2341 uint64_t Addr = *Val.getAsAddress(&Unit.getOrigUnit());
2342 if (AttrSpec.Attr == dwarf::DW_AT_low_pc) {
2343 if (Die.getTag() == dwarf::DW_TAG_inlined_subroutine ||
2344 Die.getTag() == dwarf::DW_TAG_lexical_block)
2345 // The low_pc of a block or inline subroutine might get
2346 // relocated because it happens to match the low_pc of the
2347 // enclosing subprogram. To prevent issues with that, always use
2348 // the low_pc from the input DIE if relocations have been applied.
2349 Addr = (Info.OrigLowPc != UINT64_MAX ? Info.OrigLowPc : Addr) +
2351 else if (Die.getTag() == dwarf::DW_TAG_compile_unit) {
2352 Addr = Unit.getLowPc();
2353 if (Addr == UINT64_MAX)
2356 Info.HasLowPc = true;
2357 } else if (AttrSpec.Attr == dwarf::DW_AT_high_pc) {
2358 if (Die.getTag() == dwarf::DW_TAG_compile_unit) {
2359 if (uint64_t HighPc = Unit.getHighPc())
2364 // If we have a high_pc recorded for the input DIE, use
2365 // it. Otherwise (when no relocations where applied) just use the
2366 // one we just decoded.
2367 Addr = (Info.OrigHighPc ? Info.OrigHighPc : Addr) + Info.PCOffset;
2370 Die.addValue(DIEAlloc, static_cast<dwarf::Attribute>(AttrSpec.Attr),
2371 static_cast<dwarf::Form>(AttrSpec.Form), DIEInteger(Addr));
2372 return Unit.getOrigUnit().getAddressByteSize();
2375 unsigned DwarfLinker::DIECloner::cloneScalarAttribute(
2376 DIE &Die, const DWARFDebugInfoEntryMinimal &InputDIE, CompileUnit &Unit,
2377 AttributeSpec AttrSpec, const DWARFFormValue &Val, unsigned AttrSize,
2378 AttributesInfo &Info) {
2380 if (AttrSpec.Attr == dwarf::DW_AT_high_pc &&
2381 Die.getTag() == dwarf::DW_TAG_compile_unit) {
2382 if (Unit.getLowPc() == -1ULL)
2384 // Dwarf >= 4 high_pc is an size, not an address.
2385 Value = Unit.getHighPc() - Unit.getLowPc();
2386 } else if (AttrSpec.Form == dwarf::DW_FORM_sec_offset)
2387 Value = *Val.getAsSectionOffset();
2388 else if (AttrSpec.Form == dwarf::DW_FORM_sdata)
2389 Value = *Val.getAsSignedConstant();
2390 else if (auto OptionalValue = Val.getAsUnsignedConstant())
2391 Value = *OptionalValue;
2393 Linker.reportWarning(
2394 "Unsupported scalar attribute form. Dropping attribute.",
2395 &Unit.getOrigUnit(), &InputDIE);
2398 PatchLocation Patch =
2399 Die.addValue(DIEAlloc, dwarf::Attribute(AttrSpec.Attr),
2400 dwarf::Form(AttrSpec.Form), DIEInteger(Value));
2401 if (AttrSpec.Attr == dwarf::DW_AT_ranges)
2402 Unit.noteRangeAttribute(Die, Patch);
2404 // A more generic way to check for location attributes would be
2405 // nice, but it's very unlikely that any other attribute needs a
2407 else if (AttrSpec.Attr == dwarf::DW_AT_location ||
2408 AttrSpec.Attr == dwarf::DW_AT_frame_base)
2409 Unit.noteLocationAttribute(Patch, Info.PCOffset);
2410 else if (AttrSpec.Attr == dwarf::DW_AT_declaration && Value)
2411 Info.IsDeclaration = true;
2416 /// \brief Clone \p InputDIE's attribute described by \p AttrSpec with
2417 /// value \p Val, and add it to \p Die.
2418 /// \returns the size of the cloned attribute.
2419 unsigned DwarfLinker::DIECloner::cloneAttribute(
2420 DIE &Die, const DWARFDebugInfoEntryMinimal &InputDIE, CompileUnit &Unit,
2421 const DWARFFormValue &Val, const AttributeSpec AttrSpec, unsigned AttrSize,
2422 AttributesInfo &Info) {
2423 const DWARFUnit &U = Unit.getOrigUnit();
2425 switch (AttrSpec.Form) {
2426 case dwarf::DW_FORM_strp:
2427 case dwarf::DW_FORM_string:
2428 return cloneStringAttribute(Die, AttrSpec, Val, U);
2429 case dwarf::DW_FORM_ref_addr:
2430 case dwarf::DW_FORM_ref1:
2431 case dwarf::DW_FORM_ref2:
2432 case dwarf::DW_FORM_ref4:
2433 case dwarf::DW_FORM_ref8:
2434 return cloneDieReferenceAttribute(Die, InputDIE, AttrSpec, AttrSize, Val,
2436 case dwarf::DW_FORM_block:
2437 case dwarf::DW_FORM_block1:
2438 case dwarf::DW_FORM_block2:
2439 case dwarf::DW_FORM_block4:
2440 case dwarf::DW_FORM_exprloc:
2441 return cloneBlockAttribute(Die, AttrSpec, Val, AttrSize);
2442 case dwarf::DW_FORM_addr:
2443 return cloneAddressAttribute(Die, AttrSpec, Val, Unit, Info);
2444 case dwarf::DW_FORM_data1:
2445 case dwarf::DW_FORM_data2:
2446 case dwarf::DW_FORM_data4:
2447 case dwarf::DW_FORM_data8:
2448 case dwarf::DW_FORM_udata:
2449 case dwarf::DW_FORM_sdata:
2450 case dwarf::DW_FORM_sec_offset:
2451 case dwarf::DW_FORM_flag:
2452 case dwarf::DW_FORM_flag_present:
2453 return cloneScalarAttribute(Die, InputDIE, Unit, AttrSpec, Val, AttrSize,
2456 Linker.reportWarning(
2457 "Unsupported attribute form in cloneAttribute. Dropping.", &U,
2464 /// \brief Apply the valid relocations found by findValidRelocs() to
2465 /// the buffer \p Data, taking into account that Data is at \p BaseOffset
2466 /// in the debug_info section.
2468 /// Like for findValidRelocs(), this function must be called with
2469 /// monotonic \p BaseOffset values.
2471 /// \returns wether any reloc has been applied.
2472 bool DwarfLinker::RelocationManager::
2473 applyValidRelocs(MutableArrayRef<char> Data, uint32_t BaseOffset,
2474 bool isLittleEndian) {
2475 assert((NextValidReloc == 0 ||
2476 BaseOffset > ValidRelocs[NextValidReloc - 1].Offset) &&
2477 "BaseOffset should only be increasing.");
2478 if (NextValidReloc >= ValidRelocs.size())
2481 // Skip relocs that haven't been applied.
2482 while (NextValidReloc < ValidRelocs.size() &&
2483 ValidRelocs[NextValidReloc].Offset < BaseOffset)
2486 bool Applied = false;
2487 uint64_t EndOffset = BaseOffset + Data.size();
2488 while (NextValidReloc < ValidRelocs.size() &&
2489 ValidRelocs[NextValidReloc].Offset >= BaseOffset &&
2490 ValidRelocs[NextValidReloc].Offset < EndOffset) {
2491 const auto &ValidReloc = ValidRelocs[NextValidReloc++];
2492 assert(ValidReloc.Offset - BaseOffset < Data.size());
2493 assert(ValidReloc.Offset - BaseOffset + ValidReloc.Size <= Data.size());
2495 uint64_t Value = ValidReloc.Mapping->getValue().BinaryAddress;
2496 Value += ValidReloc.Addend;
2497 for (unsigned i = 0; i != ValidReloc.Size; ++i) {
2498 unsigned Index = isLittleEndian ? i : (ValidReloc.Size - i - 1);
2499 Buf[i] = uint8_t(Value >> (Index * 8));
2501 assert(ValidReloc.Size <= sizeof(Buf));
2502 memcpy(&Data[ValidReloc.Offset - BaseOffset], Buf, ValidReloc.Size);
2509 static bool isTypeTag(uint16_t Tag) {
2511 case dwarf::DW_TAG_array_type:
2512 case dwarf::DW_TAG_class_type:
2513 case dwarf::DW_TAG_enumeration_type:
2514 case dwarf::DW_TAG_pointer_type:
2515 case dwarf::DW_TAG_reference_type:
2516 case dwarf::DW_TAG_string_type:
2517 case dwarf::DW_TAG_structure_type:
2518 case dwarf::DW_TAG_subroutine_type:
2519 case dwarf::DW_TAG_typedef:
2520 case dwarf::DW_TAG_union_type:
2521 case dwarf::DW_TAG_ptr_to_member_type:
2522 case dwarf::DW_TAG_set_type:
2523 case dwarf::DW_TAG_subrange_type:
2524 case dwarf::DW_TAG_base_type:
2525 case dwarf::DW_TAG_const_type:
2526 case dwarf::DW_TAG_constant:
2527 case dwarf::DW_TAG_file_type:
2528 case dwarf::DW_TAG_namelist:
2529 case dwarf::DW_TAG_packed_type:
2530 case dwarf::DW_TAG_volatile_type:
2531 case dwarf::DW_TAG_restrict_type:
2532 case dwarf::DW_TAG_interface_type:
2533 case dwarf::DW_TAG_unspecified_type:
2534 case dwarf::DW_TAG_shared_type:
2543 shouldSkipAttribute(DWARFAbbreviationDeclaration::AttributeSpec AttrSpec,
2544 uint16_t Tag, bool InDebugMap, bool SkipPC,
2545 bool InFunctionScope) {
2546 switch (AttrSpec.Attr) {
2549 case dwarf::DW_AT_low_pc:
2550 case dwarf::DW_AT_high_pc:
2551 case dwarf::DW_AT_ranges:
2553 case dwarf::DW_AT_location:
2554 case dwarf::DW_AT_frame_base:
2555 // FIXME: for some reason dsymutil-classic keeps the location
2556 // attributes when they are of block type (ie. not location
2557 // lists). This is totally wrong for globals where we will keep a
2558 // wrong address. It is mostly harmless for locals, but there is
2559 // no point in keeping these anyway when the function wasn't linked.
2560 return (SkipPC || (!InFunctionScope && Tag == dwarf::DW_TAG_variable &&
2562 !DWARFFormValue(AttrSpec.Form).isFormClass(DWARFFormValue::FC_Block);
2566 DIE *DwarfLinker::DIECloner::cloneDIE(
2567 RelocationManager &RelocMgr, const DWARFDebugInfoEntryMinimal &InputDIE,
2568 CompileUnit &Unit, int64_t PCOffset, uint32_t OutOffset, unsigned Flags) {
2569 DWARFUnit &U = Unit.getOrigUnit();
2570 unsigned Idx = U.getDIEIndex(&InputDIE);
2571 CompileUnit::DIEInfo &Info = Unit.getInfo(Idx);
2573 // Should the DIE appear in the output?
2574 if (!Unit.getInfo(Idx).Keep)
2577 uint32_t Offset = InputDIE.getOffset();
2578 // The DIE might have been already created by a forward reference
2579 // (see cloneDieReferenceAttribute()).
2580 DIE *Die = Info.Clone;
2582 Die = Info.Clone = DIE::get(DIEAlloc, dwarf::Tag(InputDIE.getTag()));
2583 assert(Die->getTag() == InputDIE.getTag());
2584 Die->setOffset(OutOffset);
2585 if (Unit.hasODR() && Die->getTag() != dwarf::DW_TAG_namespace && Info.Ctxt &&
2586 Info.Ctxt != Unit.getInfo(Info.ParentIdx).Ctxt &&
2587 !Info.Ctxt->getCanonicalDIEOffset()) {
2588 // We are about to emit a DIE that is the root of its own valid
2589 // DeclContext tree. Make the current offset the canonical offset
2590 // for this context.
2591 Info.Ctxt->setCanonicalDIEOffset(OutOffset + Unit.getStartOffset());
2594 // Extract and clone every attribute.
2595 DataExtractor Data = U.getDebugInfoExtractor();
2596 uint32_t NextOffset = U.getDIEAtIndex(Idx + 1)->getOffset();
2597 AttributesInfo AttrInfo;
2599 // We could copy the data only if we need to aply a relocation to
2600 // it. After testing, it seems there is no performance downside to
2601 // doing the copy unconditionally, and it makes the code simpler.
2602 SmallString<40> DIECopy(Data.getData().substr(Offset, NextOffset - Offset));
2603 Data = DataExtractor(DIECopy, Data.isLittleEndian(), Data.getAddressSize());
2604 // Modify the copy with relocated addresses.
2605 if (RelocMgr.applyValidRelocs(DIECopy, Offset, Data.isLittleEndian())) {
2606 // If we applied relocations, we store the value of high_pc that was
2607 // potentially stored in the input DIE. If high_pc is an address
2608 // (Dwarf version == 2), then it might have been relocated to a
2609 // totally unrelated value (because the end address in the object
2610 // file might be start address of another function which got moved
2611 // independantly by the linker). The computation of the actual
2612 // high_pc value is done in cloneAddressAttribute().
2613 AttrInfo.OrigHighPc =
2614 InputDIE.getAttributeValueAsAddress(&U, dwarf::DW_AT_high_pc, 0);
2615 // Also store the low_pc. It might get relocated in an
2616 // inline_subprogram that happens at the beginning of its
2617 // inlining function.
2618 AttrInfo.OrigLowPc =
2619 InputDIE.getAttributeValueAsAddress(&U, dwarf::DW_AT_low_pc, UINT64_MAX);
2622 // Reset the Offset to 0 as we will be working on the local copy of
2626 const auto *Abbrev = InputDIE.getAbbreviationDeclarationPtr();
2627 Offset += getULEB128Size(Abbrev->getCode());
2629 // We are entering a subprogram. Get and propagate the PCOffset.
2630 if (Die->getTag() == dwarf::DW_TAG_subprogram)
2631 PCOffset = Info.AddrAdjust;
2632 AttrInfo.PCOffset = PCOffset;
2634 if (Abbrev->getTag() == dwarf::DW_TAG_subprogram) {
2635 Flags |= TF_InFunctionScope;
2636 if (!Info.InDebugMap)
2640 bool Copied = false;
2641 for (const auto &AttrSpec : Abbrev->attributes()) {
2642 if (shouldSkipAttribute(AttrSpec, Die->getTag(), Info.InDebugMap,
2643 Flags & TF_SkipPC, Flags & TF_InFunctionScope)) {
2644 DWARFFormValue::skipValue(AttrSpec.Form, Data, &Offset, &U);
2645 // FIXME: dsymutil-classic keeps the old abbreviation around
2646 // even if it's not used. We can remove this (and the copyAbbrev
2647 // helper) as soon as bit-for-bit compatibility is not a goal anymore.
2649 copyAbbrev(*InputDIE.getAbbreviationDeclarationPtr(), Unit.hasODR());
2655 DWARFFormValue Val(AttrSpec.Form);
2656 uint32_t AttrSize = Offset;
2657 Val.extractValue(Data, &Offset, &U);
2658 AttrSize = Offset - AttrSize;
2661 cloneAttribute(*Die, InputDIE, Unit, Val, AttrSpec, AttrSize, AttrInfo);
2664 // Look for accelerator entries.
2665 uint16_t Tag = InputDIE.getTag();
2666 // FIXME: This is slightly wrong. An inline_subroutine without a
2667 // low_pc, but with AT_ranges might be interesting to get into the
2668 // accelerator tables too. For now stick with dsymutil's behavior.
2669 if ((Info.InDebugMap || AttrInfo.HasLowPc) &&
2670 Tag != dwarf::DW_TAG_compile_unit &&
2671 getDIENames(InputDIE, Unit.getOrigUnit(), AttrInfo)) {
2672 if (AttrInfo.MangledName && AttrInfo.MangledName != AttrInfo.Name)
2673 Unit.addNameAccelerator(Die, AttrInfo.MangledName,
2674 AttrInfo.MangledNameOffset,
2675 Tag == dwarf::DW_TAG_inlined_subroutine);
2677 Unit.addNameAccelerator(Die, AttrInfo.Name, AttrInfo.NameOffset,
2678 Tag == dwarf::DW_TAG_inlined_subroutine);
2679 } else if (isTypeTag(Tag) && !AttrInfo.IsDeclaration &&
2680 getDIENames(InputDIE, Unit.getOrigUnit(), AttrInfo)) {
2681 Unit.addTypeAccelerator(Die, AttrInfo.Name, AttrInfo.NameOffset);
2684 DIEAbbrev NewAbbrev = Die->generateAbbrev();
2685 // If a scope DIE is kept, we must have kept at least one child. If
2686 // it's not the case, we'll just be emitting one wasteful end of
2687 // children marker, but things won't break.
2688 if (InputDIE.hasChildren())
2689 NewAbbrev.setChildrenFlag(dwarf::DW_CHILDREN_yes);
2690 // Assign a permanent abbrev number
2691 Linker.AssignAbbrev(NewAbbrev);
2692 Die->setAbbrevNumber(NewAbbrev.getNumber());
2694 // Add the size of the abbreviation number to the output offset.
2695 OutOffset += getULEB128Size(Die->getAbbrevNumber());
2697 if (!Abbrev->hasChildren()) {
2699 Die->setSize(OutOffset - Die->getOffset());
2703 // Recursively clone children.
2704 for (auto *Child = InputDIE.getFirstChild(); Child && !Child->isNULL();
2705 Child = Child->getSibling()) {
2707 cloneDIE(RelocMgr, *Child, Unit, PCOffset, OutOffset, Flags)) {
2708 Die->addChild(Clone);
2709 OutOffset = Clone->getOffset() + Clone->getSize();
2713 // Account for the end of children marker.
2714 OutOffset += sizeof(int8_t);
2716 Die->setSize(OutOffset - Die->getOffset());
2720 /// \brief Patch the input object file relevant debug_ranges entries
2721 /// and emit them in the output file. Update the relevant attributes
2722 /// to point at the new entries.
2723 void DwarfLinker::patchRangesForUnit(const CompileUnit &Unit,
2724 DWARFContext &OrigDwarf) const {
2725 DWARFDebugRangeList RangeList;
2726 const auto &FunctionRanges = Unit.getFunctionRanges();
2727 unsigned AddressSize = Unit.getOrigUnit().getAddressByteSize();
2728 DataExtractor RangeExtractor(OrigDwarf.getRangeSection(),
2729 OrigDwarf.isLittleEndian(), AddressSize);
2730 auto InvalidRange = FunctionRanges.end(), CurrRange = InvalidRange;
2731 DWARFUnit &OrigUnit = Unit.getOrigUnit();
2732 const auto *OrigUnitDie = OrigUnit.getUnitDIE(false);
2733 uint64_t OrigLowPc = OrigUnitDie->getAttributeValueAsAddress(
2734 &OrigUnit, dwarf::DW_AT_low_pc, -1ULL);
2735 // Ranges addresses are based on the unit's low_pc. Compute the
2736 // offset we need to apply to adapt to the the new unit's low_pc.
2737 int64_t UnitPcOffset = 0;
2738 if (OrigLowPc != -1ULL)
2739 UnitPcOffset = int64_t(OrigLowPc) - Unit.getLowPc();
2741 for (const auto &RangeAttribute : Unit.getRangesAttributes()) {
2742 uint32_t Offset = RangeAttribute.get();
2743 RangeAttribute.set(Streamer->getRangesSectionSize());
2744 RangeList.extract(RangeExtractor, &Offset);
2745 const auto &Entries = RangeList.getEntries();
2746 if (!Entries.empty()) {
2747 const DWARFDebugRangeList::RangeListEntry &First = Entries.front();
2749 if (CurrRange == InvalidRange ||
2750 First.StartAddress + OrigLowPc < CurrRange.start() ||
2751 First.StartAddress + OrigLowPc >= CurrRange.stop()) {
2752 CurrRange = FunctionRanges.find(First.StartAddress + OrigLowPc);
2753 if (CurrRange == InvalidRange ||
2754 CurrRange.start() > First.StartAddress + OrigLowPc) {
2755 reportWarning("no mapping for range.");
2761 Streamer->emitRangesEntries(UnitPcOffset, OrigLowPc, CurrRange, Entries,
2766 /// \brief Generate the debug_aranges entries for \p Unit and if the
2767 /// unit has a DW_AT_ranges attribute, also emit the debug_ranges
2768 /// contribution for this attribute.
2769 /// FIXME: this could actually be done right in patchRangesForUnit,
2770 /// but for the sake of initial bit-for-bit compatibility with legacy
2771 /// dsymutil, we have to do it in a delayed pass.
2772 void DwarfLinker::generateUnitRanges(CompileUnit &Unit) const {
2773 auto Attr = Unit.getUnitRangesAttribute();
2775 Attr->set(Streamer->getRangesSectionSize());
2776 Streamer->emitUnitRangesEntries(Unit, static_cast<bool>(Attr));
2779 /// \brief Insert the new line info sequence \p Seq into the current
2780 /// set of already linked line info \p Rows.
2781 static void insertLineSequence(std::vector<DWARFDebugLine::Row> &Seq,
2782 std::vector<DWARFDebugLine::Row> &Rows) {
2786 if (!Rows.empty() && Rows.back().Address < Seq.front().Address) {
2787 Rows.insert(Rows.end(), Seq.begin(), Seq.end());
2792 auto InsertPoint = std::lower_bound(
2793 Rows.begin(), Rows.end(), Seq.front(),
2794 [](const DWARFDebugLine::Row &LHS, const DWARFDebugLine::Row &RHS) {
2795 return LHS.Address < RHS.Address;
2798 // FIXME: this only removes the unneeded end_sequence if the
2799 // sequences have been inserted in order. using a global sort like
2800 // described in patchLineTableForUnit() and delaying the end_sequene
2801 // elimination to emitLineTableForUnit() we can get rid of all of them.
2802 if (InsertPoint != Rows.end() &&
2803 InsertPoint->Address == Seq.front().Address && InsertPoint->EndSequence) {
2804 *InsertPoint = Seq.front();
2805 Rows.insert(InsertPoint + 1, Seq.begin() + 1, Seq.end());
2807 Rows.insert(InsertPoint, Seq.begin(), Seq.end());
2813 static void patchStmtList(DIE &Die, DIEInteger Offset) {
2814 for (auto &V : Die.values())
2815 if (V.getAttribute() == dwarf::DW_AT_stmt_list) {
2816 V = DIEValue(V.getAttribute(), V.getForm(), Offset);
2820 llvm_unreachable("Didn't find DW_AT_stmt_list in cloned DIE!");
2823 /// \brief Extract the line table for \p Unit from \p OrigDwarf, and
2824 /// recreate a relocated version of these for the address ranges that
2825 /// are present in the binary.
2826 void DwarfLinker::patchLineTableForUnit(CompileUnit &Unit,
2827 DWARFContext &OrigDwarf) {
2828 const DWARFDebugInfoEntryMinimal *CUDie = Unit.getOrigUnit().getUnitDIE();
2829 uint64_t StmtList = CUDie->getAttributeValueAsSectionOffset(
2830 &Unit.getOrigUnit(), dwarf::DW_AT_stmt_list, -1ULL);
2831 if (StmtList == -1ULL)
2834 // Update the cloned DW_AT_stmt_list with the correct debug_line offset.
2835 if (auto *OutputDIE = Unit.getOutputUnitDIE())
2836 patchStmtList(*OutputDIE, DIEInteger(Streamer->getLineSectionSize()));
2838 // Parse the original line info for the unit.
2839 DWARFDebugLine::LineTable LineTable;
2840 uint32_t StmtOffset = StmtList;
2841 StringRef LineData = OrigDwarf.getLineSection().Data;
2842 DataExtractor LineExtractor(LineData, OrigDwarf.isLittleEndian(),
2843 Unit.getOrigUnit().getAddressByteSize());
2844 LineTable.parse(LineExtractor, &OrigDwarf.getLineSection().Relocs,
2847 // This vector is the output line table.
2848 std::vector<DWARFDebugLine::Row> NewRows;
2849 NewRows.reserve(LineTable.Rows.size());
2851 // Current sequence of rows being extracted, before being inserted
2853 std::vector<DWARFDebugLine::Row> Seq;
2854 const auto &FunctionRanges = Unit.getFunctionRanges();
2855 auto InvalidRange = FunctionRanges.end(), CurrRange = InvalidRange;
2857 // FIXME: This logic is meant to generate exactly the same output as
2858 // Darwin's classic dsynutil. There is a nicer way to implement this
2859 // by simply putting all the relocated line info in NewRows and simply
2860 // sorting NewRows before passing it to emitLineTableForUnit. This
2861 // should be correct as sequences for a function should stay
2862 // together in the sorted output. There are a few corner cases that
2863 // look suspicious though, and that required to implement the logic
2864 // this way. Revisit that once initial validation is finished.
2866 // Iterate over the object file line info and extract the sequences
2867 // that correspond to linked functions.
2868 for (auto &Row : LineTable.Rows) {
2869 // Check wether we stepped out of the range. The range is
2870 // half-open, but consider accept the end address of the range if
2871 // it is marked as end_sequence in the input (because in that
2872 // case, the relocation offset is accurate and that entry won't
2873 // serve as the start of another function).
2874 if (CurrRange == InvalidRange || Row.Address < CurrRange.start() ||
2875 Row.Address > CurrRange.stop() ||
2876 (Row.Address == CurrRange.stop() && !Row.EndSequence)) {
2877 // We just stepped out of a known range. Insert a end_sequence
2878 // corresponding to the end of the range.
2879 uint64_t StopAddress = CurrRange != InvalidRange
2880 ? CurrRange.stop() + CurrRange.value()
2882 CurrRange = FunctionRanges.find(Row.Address);
2883 bool CurrRangeValid =
2884 CurrRange != InvalidRange && CurrRange.start() <= Row.Address;
2885 if (!CurrRangeValid) {
2886 CurrRange = InvalidRange;
2887 if (StopAddress != -1ULL) {
2888 // Try harder by looking in the DebugMapObject function
2889 // ranges map. There are corner cases where this finds a
2890 // valid entry. It's unclear if this is right or wrong, but
2891 // for now do as dsymutil.
2892 // FIXME: Understand exactly what cases this addresses and
2893 // potentially remove it along with the Ranges map.
2894 auto Range = Ranges.lower_bound(Row.Address);
2895 if (Range != Ranges.begin() && Range != Ranges.end())
2898 if (Range != Ranges.end() && Range->first <= Row.Address &&
2899 Range->second.first >= Row.Address) {
2900 StopAddress = Row.Address + Range->second.second;
2904 if (StopAddress != -1ULL && !Seq.empty()) {
2905 // Insert end sequence row with the computed end address, but
2906 // the same line as the previous one.
2907 auto NextLine = Seq.back();
2908 NextLine.Address = StopAddress;
2909 NextLine.EndSequence = 1;
2910 NextLine.PrologueEnd = 0;
2911 NextLine.BasicBlock = 0;
2912 NextLine.EpilogueBegin = 0;
2913 Seq.push_back(NextLine);
2914 insertLineSequence(Seq, NewRows);
2917 if (!CurrRangeValid)
2921 // Ignore empty sequences.
2922 if (Row.EndSequence && Seq.empty())
2925 // Relocate row address and add it to the current sequence.
2926 Row.Address += CurrRange.value();
2927 Seq.emplace_back(Row);
2929 if (Row.EndSequence)
2930 insertLineSequence(Seq, NewRows);
2933 // Finished extracting, now emit the line tables.
2934 uint32_t PrologueEnd = StmtList + 10 + LineTable.Prologue.PrologueLength;
2935 // FIXME: LLVM hardcodes it's prologue values. We just copy the
2936 // prologue over and that works because we act as both producer and
2937 // consumer. It would be nicer to have a real configurable line
2939 if (LineTable.Prologue.Version != 2 ||
2940 LineTable.Prologue.DefaultIsStmt != DWARF2_LINE_DEFAULT_IS_STMT ||
2941 LineTable.Prologue.OpcodeBase > 13)
2942 reportWarning("line table paramters mismatch. Cannot emit.");
2944 MCDwarfLineTableParams Params;
2945 Params.DWARF2LineOpcodeBase = LineTable.Prologue.OpcodeBase;
2946 Params.DWARF2LineBase = LineTable.Prologue.LineBase;
2947 Params.DWARF2LineRange = LineTable.Prologue.LineRange;
2948 Streamer->emitLineTableForUnit(Params,
2949 LineData.slice(StmtList + 4, PrologueEnd),
2950 LineTable.Prologue.MinInstLength, NewRows,
2951 Unit.getOrigUnit().getAddressByteSize());
2955 void DwarfLinker::emitAcceleratorEntriesForUnit(CompileUnit &Unit) {
2956 Streamer->emitPubNamesForUnit(Unit);
2957 Streamer->emitPubTypesForUnit(Unit);
2960 /// \brief Read the frame info stored in the object, and emit the
2961 /// patched frame descriptions for the linked binary.
2963 /// This is actually pretty easy as the data of the CIEs and FDEs can
2964 /// be considered as black boxes and moved as is. The only thing to do
2965 /// is to patch the addresses in the headers.
2966 void DwarfLinker::patchFrameInfoForObject(const DebugMapObject &DMO,
2967 DWARFContext &OrigDwarf,
2968 unsigned AddrSize) {
2969 StringRef FrameData = OrigDwarf.getDebugFrameSection();
2970 if (FrameData.empty())
2973 DataExtractor Data(FrameData, OrigDwarf.isLittleEndian(), 0);
2974 uint32_t InputOffset = 0;
2976 // Store the data of the CIEs defined in this object, keyed by their
2978 DenseMap<uint32_t, StringRef> LocalCIES;
2980 while (Data.isValidOffset(InputOffset)) {
2981 uint32_t EntryOffset = InputOffset;
2982 uint32_t InitialLength = Data.getU32(&InputOffset);
2983 if (InitialLength == 0xFFFFFFFF)
2984 return reportWarning("Dwarf64 bits no supported");
2986 uint32_t CIEId = Data.getU32(&InputOffset);
2987 if (CIEId == 0xFFFFFFFF) {
2988 // This is a CIE, store it.
2989 StringRef CIEData = FrameData.substr(EntryOffset, InitialLength + 4);
2990 LocalCIES[EntryOffset] = CIEData;
2991 // The -4 is to account for the CIEId we just read.
2992 InputOffset += InitialLength - 4;
2996 uint32_t Loc = Data.getUnsigned(&InputOffset, AddrSize);
2998 // Some compilers seem to emit frame info that doesn't start at
2999 // the function entry point, thus we can't just lookup the address
3000 // in the debug map. Use the linker's range map to see if the FDE
3001 // describes something that we can relocate.
3002 auto Range = Ranges.upper_bound(Loc);
3003 if (Range != Ranges.begin())
3005 if (Range == Ranges.end() || Range->first > Loc ||
3006 Range->second.first <= Loc) {
3007 // The +4 is to account for the size of the InitialLength field itself.
3008 InputOffset = EntryOffset + InitialLength + 4;
3012 // This is an FDE, and we have a mapping.
3013 // Have we already emitted a corresponding CIE?
3014 StringRef CIEData = LocalCIES[CIEId];
3015 if (CIEData.empty())
3016 return reportWarning("Inconsistent debug_frame content. Dropping.");
3018 // Look if we already emitted a CIE that corresponds to the
3019 // referenced one (the CIE data is the key of that lookup).
3020 auto IteratorInserted = EmittedCIEs.insert(
3021 std::make_pair(CIEData, Streamer->getFrameSectionSize()));
3022 // If there is no CIE yet for this ID, emit it.
3023 if (IteratorInserted.second ||
3024 // FIXME: dsymutil-classic only caches the last used CIE for
3025 // reuse. Mimic that behavior for now. Just removing that
3026 // second half of the condition and the LastCIEOffset variable
3027 // makes the code DTRT.
3028 LastCIEOffset != IteratorInserted.first->getValue()) {
3029 LastCIEOffset = Streamer->getFrameSectionSize();
3030 IteratorInserted.first->getValue() = LastCIEOffset;
3031 Streamer->emitCIE(CIEData);
3034 // Emit the FDE with updated address and CIE pointer.
3035 // (4 + AddrSize) is the size of the CIEId + initial_location
3036 // fields that will get reconstructed by emitFDE().
3037 unsigned FDERemainingBytes = InitialLength - (4 + AddrSize);
3038 Streamer->emitFDE(IteratorInserted.first->getValue(), AddrSize,
3039 Loc + Range->second.second,
3040 FrameData.substr(InputOffset, FDERemainingBytes));
3041 InputOffset += FDERemainingBytes;
3045 void DwarfLinker::DIECloner::copyAbbrev(
3046 const DWARFAbbreviationDeclaration &Abbrev, bool hasODR) {
3047 DIEAbbrev Copy(dwarf::Tag(Abbrev.getTag()),
3048 dwarf::Form(Abbrev.hasChildren()));
3050 for (const auto &Attr : Abbrev.attributes()) {
3051 uint16_t Form = Attr.Form;
3052 if (hasODR && isODRAttribute(Attr.Attr))
3053 Form = dwarf::DW_FORM_ref_addr;
3054 Copy.AddAttribute(dwarf::Attribute(Attr.Attr), dwarf::Form(Form));
3057 Linker.AssignAbbrev(Copy);
3060 ErrorOr<const object::ObjectFile &>
3061 DwarfLinker::loadObject(BinaryHolder &BinaryHolder, DebugMapObject &Obj,
3062 const DebugMap &Map) {
3064 BinaryHolder.GetObjectFiles(Obj.getObjectFilename(), Obj.getTimestamp());
3065 if (std::error_code EC = ErrOrObjs.getError()) {
3066 reportWarning(Twine(Obj.getObjectFilename()) + ": " + EC.message());
3069 auto ErrOrObj = BinaryHolder.Get(Map.getTriple());
3070 if (std::error_code EC = ErrOrObj.getError())
3071 reportWarning(Twine(Obj.getObjectFilename()) + ": " + EC.message());
3075 void DwarfLinker::DIECloner::cloneAllCompileUnits(
3076 DWARFContextInMemory &DwarfContext) {
3077 if (!Linker.Streamer)
3080 for (auto &CurrentUnit : CompileUnits) {
3081 const auto *InputDIE = CurrentUnit.getOrigUnit().getUnitDIE();
3082 CurrentUnit.setStartOffset(Linker.OutputDebugInfoSize);
3083 DIE *OutputDIE = cloneDIE(RelocMgr, *InputDIE, CurrentUnit,
3084 0 /* PC offset */, 11 /* Unit Header size */, 0);
3085 CurrentUnit.setOutputUnitDIE(OutputDIE);
3086 Linker.OutputDebugInfoSize = CurrentUnit.computeNextUnitOffset();
3087 if (Linker.Options.NoOutput)
3089 // FIXME: for compatibility with the classic dsymutil, we emit
3090 // an empty line table for the unit, even if the unit doesn't
3091 // actually exist in the DIE tree.
3092 Linker.patchLineTableForUnit(CurrentUnit, DwarfContext);
3095 Linker.patchRangesForUnit(CurrentUnit, DwarfContext);
3096 Linker.Streamer->emitLocationsForUnit(CurrentUnit, DwarfContext);
3097 Linker.emitAcceleratorEntriesForUnit(CurrentUnit);
3100 if (Linker.Options.NoOutput)
3103 // Emit all the compile unit's debug information.
3104 for (auto &CurrentUnit : CompileUnits) {
3105 Linker.generateUnitRanges(CurrentUnit);
3106 CurrentUnit.fixupForwardReferences();
3107 Linker.Streamer->emitCompileUnitHeader(CurrentUnit);
3108 if (!CurrentUnit.getOutputUnitDIE())
3110 Linker.Streamer->emitDIE(*CurrentUnit.getOutputUnitDIE());
3114 bool DwarfLinker::link(const DebugMap &Map) {
3116 if (!createStreamer(Map.getTriple(), OutputFilename))
3119 // Size of the DIEs (and headers) generated for the linked output.
3120 OutputDebugInfoSize = 0;
3121 // A unique ID that identifies each compile unit.
3122 unsigned UnitID = 0;
3123 for (const auto &Obj : Map.objects()) {
3124 CurrentDebugObject = Obj.get();
3126 if (Options.Verbose)
3127 outs() << "DEBUG MAP OBJECT: " << Obj->getObjectFilename() << "\n";
3128 auto ErrOrObj = loadObject(BinHolder, *Obj, Map);
3132 // Look for relocations that correspond to debug map entries.
3133 RelocationManager RelocMgr(*this);
3134 if (!RelocMgr.findValidRelocsInDebugInfo(*ErrOrObj, *Obj)) {
3135 if (Options.Verbose)
3136 outs() << "No valid relocations found. Skipping.\n";
3140 // Setup access to the debug info.
3141 DWARFContextInMemory DwarfContext(*ErrOrObj);
3142 startDebugObject(DwarfContext, *Obj);
3144 // In a first phase, just read in the debug info and store the DIE
3145 // parent links that we will use during the next phase.
3146 for (const auto &CU : DwarfContext.compile_units()) {
3147 auto *CUDie = CU->getUnitDIE(false);
3148 if (Options.Verbose) {
3149 outs() << "Input compilation unit:";
3150 CUDie->dump(outs(), CU.get(), 0);
3152 Units.emplace_back(*CU, UnitID++, !Options.NoODR);
3153 gatherDIEParents(CUDie, 0, Units.back(), &ODRContexts.getRoot(),
3154 StringPool, ODRContexts);
3157 // Then mark all the DIEs that need to be present in the linked
3158 // output and collect some information about them. Note that this
3159 // loop can not be merged with the previous one becaue cross-cu
3160 // references require the ParentIdx to be setup for every CU in
3161 // the object file before calling this.
3162 for (auto &CurrentUnit : Units)
3163 lookForDIEsToKeep(RelocMgr, *CurrentUnit.getOrigUnit().getUnitDIE(), *Obj,
3166 // The calls to applyValidRelocs inside cloneDIE will walk the
3167 // reloc array again (in the same way findValidRelocsInDebugInfo()
3168 // did). We need to reset the NextValidReloc index to the beginning.
3169 RelocMgr.resetValidRelocs();
3170 if (RelocMgr.hasValidRelocs())
3171 DIECloner(*this, RelocMgr, DIEAlloc, Units, Options)
3172 .cloneAllCompileUnits(DwarfContext);
3173 if (!Options.NoOutput && !Units.empty())
3174 patchFrameInfoForObject(*Obj, DwarfContext,
3175 Units[0].getOrigUnit().getAddressByteSize());
3177 // Clean-up before starting working on the next object.
3181 // Emit everything that's global.
3182 if (!Options.NoOutput) {
3183 Streamer->emitAbbrevs(Abbreviations);
3184 Streamer->emitStrings(StringPool);
3187 return Options.NoOutput ? true : Streamer->finish(Map);
3191 /// \brief Get the offset of string \p S in the string table. This
3192 /// can insert a new element or return the offset of a preexisitng
3194 uint32_t NonRelocatableStringpool::getStringOffset(StringRef S) {
3195 if (S.empty() && !Strings.empty())
3198 std::pair<uint32_t, StringMapEntryBase *> Entry(0, nullptr);
3202 // A non-empty string can't be at offset 0, so if we have an entry
3203 // with a 0 offset, it must be a previously interned string.
3204 std::tie(It, Inserted) = Strings.insert(std::make_pair(S, Entry));
3205 if (Inserted || It->getValue().first == 0) {
3206 // Set offset and chain at the end of the entries list.
3207 It->getValue().first = CurrentEndOffset;
3208 CurrentEndOffset += S.size() + 1; // +1 for the '\0'.
3209 Last->getValue().second = &*It;
3212 return It->getValue().first;
3215 /// \brief Put \p S into the StringMap so that it gets permanent
3216 /// storage, but do not actually link it in the chain of elements
3217 /// that go into the output section. A latter call to
3218 /// getStringOffset() with the same string will chain it though.
3219 StringRef NonRelocatableStringpool::internString(StringRef S) {
3220 std::pair<uint32_t, StringMapEntryBase *> Entry(0, nullptr);
3221 auto InsertResult = Strings.insert(std::make_pair(S, Entry));
3222 return InsertResult.first->getKey();
3225 void warn(const Twine &Warning, const Twine &Context) {
3226 errs() << Twine("while processing ") + Context + ":\n";
3227 errs() << Twine("warning: ") + Warning + "\n";
3230 bool error(const Twine &Error, const Twine &Context) {
3231 errs() << Twine("while processing ") + Context + ":\n";
3232 errs() << Twine("error: ") + Error + "\n";
3236 bool linkDwarf(StringRef OutputFilename, const DebugMap &DM,
3237 const LinkOptions &Options) {
3238 DwarfLinker Linker(OutputFilename, Options);
3239 return Linker.link(DM);