1 //===- DWARFAcceleratorTable.cpp ------------------------------------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 
9 #include "llvm/DebugInfo/DWARF/DWARFAcceleratorTable.h"
10 
11 #include "llvm/ADT/SmallVector.h"
12 #include "llvm/BinaryFormat/Dwarf.h"
13 #include "llvm/DebugInfo/DWARF/DWARFRelocMap.h"
14 #include "llvm/Support/Compiler.h"
15 #include "llvm/Support/DJB.h"
16 #include "llvm/Support/Errc.h"
17 #include "llvm/Support/Format.h"
18 #include "llvm/Support/FormatVariadic.h"
19 #include "llvm/Support/ScopedPrinter.h"
20 #include "llvm/Support/raw_ostream.h"
21 #include <cstddef>
22 #include <cstdint>
23 #include <utility>
24 
25 using namespace llvm;
26 
27 namespace {
28 struct Atom {
29   unsigned Value;
30 };
31 
32 static raw_ostream &operator<<(raw_ostream &OS, const Atom &A) {
33   StringRef Str = dwarf::AtomTypeString(A.Value);
34   if (!Str.empty())
35     return OS << Str;
36   return OS << "DW_ATOM_unknown_" << format("%x", A.Value);
37 }
38 } // namespace
39 
40 static Atom formatAtom(unsigned Atom) { return {Atom}; }
41 
42 DWARFAcceleratorTable::~DWARFAcceleratorTable() = default;
43 
44 Error AppleAcceleratorTable::extract() {
45   uint64_t Offset = 0;
46 
47   // Check that we can at least read the header.
48   if (!AccelSection.isValidOffset(offsetof(Header, HeaderDataLength) + 4))
49     return createStringError(errc::illegal_byte_sequence,
50                              "Section too small: cannot read header.");
51 
52   Hdr.Magic = AccelSection.getU32(&Offset);
53   Hdr.Version = AccelSection.getU16(&Offset);
54   Hdr.HashFunction = AccelSection.getU16(&Offset);
55   Hdr.BucketCount = AccelSection.getU32(&Offset);
56   Hdr.HashCount = AccelSection.getU32(&Offset);
57   Hdr.HeaderDataLength = AccelSection.getU32(&Offset);
58 
59   // Check that we can read all the hashes and offsets from the
60   // section (see SourceLevelDebugging.rst for the structure of the index).
61   // We need to substract one because we're checking for an *offset* which is
62   // equal to the size for an empty table and hence pointer after the section.
63   if (!AccelSection.isValidOffset(sizeof(Hdr) + Hdr.HeaderDataLength +
64                                   Hdr.BucketCount * 4 + Hdr.HashCount * 8 - 1))
65     return createStringError(
66         errc::illegal_byte_sequence,
67         "Section too small: cannot read buckets and hashes.");
68 
69   HdrData.DIEOffsetBase = AccelSection.getU32(&Offset);
70   uint32_t NumAtoms = AccelSection.getU32(&Offset);
71 
72   for (unsigned i = 0; i < NumAtoms; ++i) {
73     uint16_t AtomType = AccelSection.getU16(&Offset);
74     auto AtomForm = static_cast<dwarf::Form>(AccelSection.getU16(&Offset));
75     HdrData.Atoms.push_back(std::make_pair(AtomType, AtomForm));
76   }
77 
78   IsValid = true;
79   return Error::success();
80 }
81 
82 uint32_t AppleAcceleratorTable::getNumBuckets() { return Hdr.BucketCount; }
83 uint32_t AppleAcceleratorTable::getNumHashes() { return Hdr.HashCount; }
84 uint32_t AppleAcceleratorTable::getSizeHdr() { return sizeof(Hdr); }
85 uint32_t AppleAcceleratorTable::getHeaderDataLength() {
86   return Hdr.HeaderDataLength;
87 }
88 
89 ArrayRef<std::pair<AppleAcceleratorTable::HeaderData::AtomType,
90                    AppleAcceleratorTable::HeaderData::Form>>
91 AppleAcceleratorTable::getAtomsDesc() {
92   return HdrData.Atoms;
93 }
94 
95 bool AppleAcceleratorTable::validateForms() {
96   for (auto Atom : getAtomsDesc()) {
97     DWARFFormValue FormValue(Atom.second);
98     switch (Atom.first) {
99     case dwarf::DW_ATOM_die_offset:
100     case dwarf::DW_ATOM_die_tag:
101     case dwarf::DW_ATOM_type_flags:
102       if ((!FormValue.isFormClass(DWARFFormValue::FC_Constant) &&
103            !FormValue.isFormClass(DWARFFormValue::FC_Flag)) ||
104           FormValue.getForm() == dwarf::DW_FORM_sdata)
105         return false;
106       break;
107     default:
108       break;
109     }
110   }
111   return true;
112 }
113 
114 std::pair<uint64_t, dwarf::Tag>
115 AppleAcceleratorTable::readAtoms(uint64_t *HashDataOffset) {
116   uint64_t DieOffset = dwarf::DW_INVALID_OFFSET;
117   dwarf::Tag DieTag = dwarf::DW_TAG_null;
118   dwarf::FormParams FormParams = {Hdr.Version, 0, dwarf::DwarfFormat::DWARF32};
119 
120   for (auto Atom : getAtomsDesc()) {
121     DWARFFormValue FormValue(Atom.second);
122     FormValue.extractValue(AccelSection, HashDataOffset, FormParams);
123     switch (Atom.first) {
124     case dwarf::DW_ATOM_die_offset:
125       DieOffset = *FormValue.getAsUnsignedConstant();
126       break;
127     case dwarf::DW_ATOM_die_tag:
128       DieTag = (dwarf::Tag)*FormValue.getAsUnsignedConstant();
129       break;
130     default:
131       break;
132     }
133   }
134   return {DieOffset, DieTag};
135 }
136 
137 void AppleAcceleratorTable::Header::dump(ScopedPrinter &W) const {
138   DictScope HeaderScope(W, "Header");
139   W.printHex("Magic", Magic);
140   W.printHex("Version", Version);
141   W.printHex("Hash function", HashFunction);
142   W.printNumber("Bucket count", BucketCount);
143   W.printNumber("Hashes count", HashCount);
144   W.printNumber("HeaderData length", HeaderDataLength);
145 }
146 
147 Optional<uint64_t> AppleAcceleratorTable::HeaderData::extractOffset(
148     Optional<DWARFFormValue> Value) const {
149   if (!Value)
150     return None;
151 
152   switch (Value->getForm()) {
153   case dwarf::DW_FORM_ref1:
154   case dwarf::DW_FORM_ref2:
155   case dwarf::DW_FORM_ref4:
156   case dwarf::DW_FORM_ref8:
157   case dwarf::DW_FORM_ref_udata:
158     return Value->getRawUValue() + DIEOffsetBase;
159   default:
160     return Value->getAsSectionOffset();
161   }
162 }
163 
164 bool AppleAcceleratorTable::dumpName(ScopedPrinter &W,
165                                      SmallVectorImpl<DWARFFormValue> &AtomForms,
166                                      uint64_t *DataOffset) const {
167   dwarf::FormParams FormParams = {Hdr.Version, 0, dwarf::DwarfFormat::DWARF32};
168   uint64_t NameOffset = *DataOffset;
169   if (!AccelSection.isValidOffsetForDataOfSize(*DataOffset, 4)) {
170     W.printString("Incorrectly terminated list.");
171     return false;
172   }
173   uint64_t StringOffset = AccelSection.getRelocatedValue(4, DataOffset);
174   if (!StringOffset)
175     return false; // End of list
176 
177   DictScope NameScope(W, ("Name@0x" + Twine::utohexstr(NameOffset)).str());
178   W.startLine() << format("String: 0x%08" PRIx64, StringOffset);
179   W.getOStream() << " \"" << StringSection.getCStr(&StringOffset) << "\"\n";
180 
181   unsigned NumData = AccelSection.getU32(DataOffset);
182   for (unsigned Data = 0; Data < NumData; ++Data) {
183     ListScope DataScope(W, ("Data " + Twine(Data)).str());
184     unsigned i = 0;
185     for (auto &Atom : AtomForms) {
186       W.startLine() << format("Atom[%d]: ", i);
187       if (Atom.extractValue(AccelSection, DataOffset, FormParams)) {
188         Atom.dump(W.getOStream());
189         if (Optional<uint64_t> Val = Atom.getAsUnsignedConstant()) {
190           StringRef Str = dwarf::AtomValueString(HdrData.Atoms[i].first, *Val);
191           if (!Str.empty())
192             W.getOStream() << " (" << Str << ")";
193         }
194       } else
195         W.getOStream() << "Error extracting the value";
196       W.getOStream() << "\n";
197       i++;
198     }
199   }
200   return true; // more entries follow
201 }
202 
203 LLVM_DUMP_METHOD void AppleAcceleratorTable::dump(raw_ostream &OS) const {
204   if (!IsValid)
205     return;
206 
207   ScopedPrinter W(OS);
208 
209   Hdr.dump(W);
210 
211   W.printNumber("DIE offset base", HdrData.DIEOffsetBase);
212   W.printNumber("Number of atoms", uint64_t(HdrData.Atoms.size()));
213   SmallVector<DWARFFormValue, 3> AtomForms;
214   {
215     ListScope AtomsScope(W, "Atoms");
216     unsigned i = 0;
217     for (const auto &Atom : HdrData.Atoms) {
218       DictScope AtomScope(W, ("Atom " + Twine(i++)).str());
219       W.startLine() << "Type: " << formatAtom(Atom.first) << '\n';
220       W.startLine() << "Form: " << formatv("{0}", Atom.second) << '\n';
221       AtomForms.push_back(DWARFFormValue(Atom.second));
222     }
223   }
224 
225   // Now go through the actual tables and dump them.
226   uint64_t Offset = sizeof(Hdr) + Hdr.HeaderDataLength;
227   uint64_t HashesBase = Offset + Hdr.BucketCount * 4;
228   uint64_t OffsetsBase = HashesBase + Hdr.HashCount * 4;
229 
230   for (unsigned Bucket = 0; Bucket < Hdr.BucketCount; ++Bucket) {
231     unsigned Index = AccelSection.getU32(&Offset);
232 
233     ListScope BucketScope(W, ("Bucket " + Twine(Bucket)).str());
234     if (Index == UINT32_MAX) {
235       W.printString("EMPTY");
236       continue;
237     }
238 
239     for (unsigned HashIdx = Index; HashIdx < Hdr.HashCount; ++HashIdx) {
240       uint64_t HashOffset = HashesBase + HashIdx*4;
241       uint64_t OffsetsOffset = OffsetsBase + HashIdx*4;
242       uint32_t Hash = AccelSection.getU32(&HashOffset);
243 
244       if (Hash % Hdr.BucketCount != Bucket)
245         break;
246 
247       uint64_t DataOffset = AccelSection.getU32(&OffsetsOffset);
248       ListScope HashScope(W, ("Hash 0x" + Twine::utohexstr(Hash)).str());
249       if (!AccelSection.isValidOffset(DataOffset)) {
250         W.printString("Invalid section offset");
251         continue;
252       }
253       while (dumpName(W, AtomForms, &DataOffset))
254         /*empty*/;
255     }
256   }
257 }
258 
259 AppleAcceleratorTable::Entry::Entry(
260     const AppleAcceleratorTable::HeaderData &HdrData)
261     : HdrData(&HdrData) {
262   Values.reserve(HdrData.Atoms.size());
263   for (const auto &Atom : HdrData.Atoms)
264     Values.push_back(DWARFFormValue(Atom.second));
265 }
266 
267 void AppleAcceleratorTable::Entry::extract(
268     const AppleAcceleratorTable &AccelTable, uint64_t *Offset) {
269 
270   dwarf::FormParams FormParams = {AccelTable.Hdr.Version, 0,
271                                   dwarf::DwarfFormat::DWARF32};
272   for (auto &Atom : Values)
273     Atom.extractValue(AccelTable.AccelSection, Offset, FormParams);
274 }
275 
276 Optional<DWARFFormValue>
277 AppleAcceleratorTable::Entry::lookup(HeaderData::AtomType Atom) const {
278   assert(HdrData && "Dereferencing end iterator?");
279   assert(HdrData->Atoms.size() == Values.size());
280   for (auto Tuple : zip_first(HdrData->Atoms, Values)) {
281     if (std::get<0>(Tuple).first == Atom)
282       return std::get<1>(Tuple);
283   }
284   return None;
285 }
286 
287 Optional<uint64_t> AppleAcceleratorTable::Entry::getDIESectionOffset() const {
288   return HdrData->extractOffset(lookup(dwarf::DW_ATOM_die_offset));
289 }
290 
291 Optional<uint64_t> AppleAcceleratorTable::Entry::getCUOffset() const {
292   return HdrData->extractOffset(lookup(dwarf::DW_ATOM_cu_offset));
293 }
294 
295 Optional<dwarf::Tag> AppleAcceleratorTable::Entry::getTag() const {
296   Optional<DWARFFormValue> Tag = lookup(dwarf::DW_ATOM_die_tag);
297   if (!Tag)
298     return None;
299   if (Optional<uint64_t> Value = Tag->getAsUnsignedConstant())
300     return dwarf::Tag(*Value);
301   return None;
302 }
303 
304 AppleAcceleratorTable::ValueIterator::ValueIterator(
305     const AppleAcceleratorTable &AccelTable, uint64_t Offset)
306     : AccelTable(&AccelTable), Current(AccelTable.HdrData), DataOffset(Offset) {
307   if (!AccelTable.AccelSection.isValidOffsetForDataOfSize(DataOffset, 4))
308     return;
309 
310   // Read the first entry.
311   NumData = AccelTable.AccelSection.getU32(&DataOffset);
312   Next();
313 }
314 
315 void AppleAcceleratorTable::ValueIterator::Next() {
316   assert(NumData > 0 && "attempted to increment iterator past the end");
317   auto &AccelSection = AccelTable->AccelSection;
318   if (Data >= NumData ||
319       !AccelSection.isValidOffsetForDataOfSize(DataOffset, 4)) {
320     NumData = 0;
321     DataOffset = 0;
322     return;
323   }
324   Current.extract(*AccelTable, &DataOffset);
325   ++Data;
326 }
327 
328 iterator_range<AppleAcceleratorTable::ValueIterator>
329 AppleAcceleratorTable::equal_range(StringRef Key) const {
330   if (!IsValid)
331     return make_range(ValueIterator(), ValueIterator());
332 
333   // Find the bucket.
334   unsigned HashValue = djbHash(Key);
335   unsigned Bucket = HashValue % Hdr.BucketCount;
336   uint64_t BucketBase = sizeof(Hdr) + Hdr.HeaderDataLength;
337   uint64_t HashesBase = BucketBase + Hdr.BucketCount * 4;
338   uint64_t OffsetsBase = HashesBase + Hdr.HashCount * 4;
339 
340   uint64_t BucketOffset = BucketBase + Bucket * 4;
341   unsigned Index = AccelSection.getU32(&BucketOffset);
342 
343   // Search through all hashes in the bucket.
344   for (unsigned HashIdx = Index; HashIdx < Hdr.HashCount; ++HashIdx) {
345     uint64_t HashOffset = HashesBase + HashIdx * 4;
346     uint64_t OffsetsOffset = OffsetsBase + HashIdx * 4;
347     uint32_t Hash = AccelSection.getU32(&HashOffset);
348 
349     if (Hash % Hdr.BucketCount != Bucket)
350       // We are already in the next bucket.
351       break;
352 
353     uint64_t DataOffset = AccelSection.getU32(&OffsetsOffset);
354     uint64_t StringOffset = AccelSection.getRelocatedValue(4, &DataOffset);
355     if (!StringOffset)
356       break;
357 
358     // Finally, compare the key.
359     if (Key == StringSection.getCStr(&StringOffset))
360       return make_range({*this, DataOffset}, ValueIterator());
361   }
362   return make_range(ValueIterator(), ValueIterator());
363 }
364 
365 void DWARFDebugNames::Header::dump(ScopedPrinter &W) const {
366   DictScope HeaderScope(W, "Header");
367   W.printHex("Length", UnitLength);
368   W.printNumber("Version", Version);
369   W.printNumber("CU count", CompUnitCount);
370   W.printNumber("Local TU count", LocalTypeUnitCount);
371   W.printNumber("Foreign TU count", ForeignTypeUnitCount);
372   W.printNumber("Bucket count", BucketCount);
373   W.printNumber("Name count", NameCount);
374   W.printHex("Abbreviations table size", AbbrevTableSize);
375   W.startLine() << "Augmentation: '" << AugmentationString << "'\n";
376 }
377 
378 Error DWARFDebugNames::Header::extract(const DWARFDataExtractor &AS,
379                                              uint64_t *Offset) {
380   uint64_t StartingOffset = *Offset;
381   // Check that we can read the unit length field.
382   if (!AS.isValidOffsetForDataOfSize(StartingOffset, 4))
383     return createStringError(errc::illegal_byte_sequence,
384                              "Section too small: cannot read header.");
385   UnitLength = AS.getU32(Offset);
386   if (UnitLength >= dwarf::DW_LENGTH_lo_reserved &&
387       UnitLength != dwarf::DW_LENGTH_DWARF64)
388     return createStringError(errc::illegal_byte_sequence,
389                              "Unsupported reserved unit length value");
390   Format = (UnitLength == dwarf::DW_LENGTH_DWARF64) ? dwarf::DWARF64
391                                                     : dwarf::DWARF32;
392 
393   // These fields are the same for 32-bit and 64-bit DWARF formats.
394   constexpr unsigned CommonHeaderSize = 2 + // Version
395                                         2 + // Padding
396                                         4 + // CU count
397                                         4 + // Local TU count
398                                         4 + // Foreign TU count
399                                         4 + // Bucket count
400                                         4 + // Name count
401                                         4 + // Abbreviations table size
402                                         4;  // Augmentation string size
403   // Check that we can read the fixed-size part.
404   if (!AS.isValidOffsetForDataOfSize(
405           StartingOffset,
406           CommonHeaderSize + dwarf::getUnitLengthFieldByteSize(Format)))
407     return createStringError(errc::illegal_byte_sequence,
408                              "Section too small: cannot read header.");
409   if (Format == dwarf::DWARF64)
410     UnitLength = AS.getU64(Offset);
411   Version = AS.getU16(Offset);
412   // Skip padding
413   *Offset += 2;
414   CompUnitCount = AS.getU32(Offset);
415   LocalTypeUnitCount = AS.getU32(Offset);
416   ForeignTypeUnitCount = AS.getU32(Offset);
417   BucketCount = AS.getU32(Offset);
418   NameCount = AS.getU32(Offset);
419   AbbrevTableSize = AS.getU32(Offset);
420   AugmentationStringSize = alignTo(AS.getU32(Offset), 4);
421 
422   if (!AS.isValidOffsetForDataOfSize(*Offset, AugmentationStringSize))
423     return createStringError(
424         errc::illegal_byte_sequence,
425         "Section too small: cannot read header augmentation.");
426   AugmentationString.resize(AugmentationStringSize);
427   AS.getU8(Offset, reinterpret_cast<uint8_t *>(AugmentationString.data()),
428            AugmentationStringSize);
429   return Error::success();
430 }
431 
432 void DWARFDebugNames::Abbrev::dump(ScopedPrinter &W) const {
433   DictScope AbbrevScope(W, ("Abbreviation 0x" + Twine::utohexstr(Code)).str());
434   W.startLine() << formatv("Tag: {0}\n", Tag);
435 
436   for (const auto &Attr : Attributes)
437     W.startLine() << formatv("{0}: {1}\n", Attr.Index, Attr.Form);
438 }
439 
440 static constexpr DWARFDebugNames::AttributeEncoding sentinelAttrEnc() {
441   return {dwarf::Index(0), dwarf::Form(0)};
442 }
443 
444 static bool isSentinel(const DWARFDebugNames::AttributeEncoding &AE) {
445   return AE == sentinelAttrEnc();
446 }
447 
448 static DWARFDebugNames::Abbrev sentinelAbbrev() {
449   return DWARFDebugNames::Abbrev(0, dwarf::Tag(0), {});
450 }
451 
452 static bool isSentinel(const DWARFDebugNames::Abbrev &Abbr) {
453   return Abbr.Code == 0;
454 }
455 
456 DWARFDebugNames::Abbrev DWARFDebugNames::AbbrevMapInfo::getEmptyKey() {
457   return sentinelAbbrev();
458 }
459 
460 DWARFDebugNames::Abbrev DWARFDebugNames::AbbrevMapInfo::getTombstoneKey() {
461   return DWARFDebugNames::Abbrev(~0, dwarf::Tag(0), {});
462 }
463 
464 Expected<DWARFDebugNames::AttributeEncoding>
465 DWARFDebugNames::NameIndex::extractAttributeEncoding(uint64_t *Offset) {
466   if (*Offset >= EntriesBase) {
467     return createStringError(errc::illegal_byte_sequence,
468                              "Incorrectly terminated abbreviation table.");
469   }
470 
471   uint32_t Index = Section.AccelSection.getULEB128(Offset);
472   uint32_t Form = Section.AccelSection.getULEB128(Offset);
473   return AttributeEncoding(dwarf::Index(Index), dwarf::Form(Form));
474 }
475 
476 Expected<std::vector<DWARFDebugNames::AttributeEncoding>>
477 DWARFDebugNames::NameIndex::extractAttributeEncodings(uint64_t *Offset) {
478   std::vector<AttributeEncoding> Result;
479   for (;;) {
480     auto AttrEncOr = extractAttributeEncoding(Offset);
481     if (!AttrEncOr)
482       return AttrEncOr.takeError();
483     if (isSentinel(*AttrEncOr))
484       return std::move(Result);
485 
486     Result.emplace_back(*AttrEncOr);
487   }
488 }
489 
490 Expected<DWARFDebugNames::Abbrev>
491 DWARFDebugNames::NameIndex::extractAbbrev(uint64_t *Offset) {
492   if (*Offset >= EntriesBase) {
493     return createStringError(errc::illegal_byte_sequence,
494                              "Incorrectly terminated abbreviation table.");
495   }
496 
497   uint32_t Code = Section.AccelSection.getULEB128(Offset);
498   if (Code == 0)
499     return sentinelAbbrev();
500 
501   uint32_t Tag = Section.AccelSection.getULEB128(Offset);
502   auto AttrEncOr = extractAttributeEncodings(Offset);
503   if (!AttrEncOr)
504     return AttrEncOr.takeError();
505   return Abbrev(Code, dwarf::Tag(Tag), std::move(*AttrEncOr));
506 }
507 
508 Error DWARFDebugNames::NameIndex::extract() {
509   const DWARFDataExtractor &AS = Section.AccelSection;
510   uint64_t Offset = Base;
511   if (Error E = Hdr.extract(AS, &Offset))
512     return E;
513 
514   const unsigned SectionOffsetSize = dwarf::getDwarfOffsetByteSize(Hdr.Format);
515   CUsBase = Offset;
516   Offset += Hdr.CompUnitCount * SectionOffsetSize;
517   Offset += Hdr.LocalTypeUnitCount * SectionOffsetSize;
518   Offset += Hdr.ForeignTypeUnitCount * 8;
519   BucketsBase = Offset;
520   Offset += Hdr.BucketCount * 4;
521   HashesBase = Offset;
522   if (Hdr.BucketCount > 0)
523     Offset += Hdr.NameCount * 4;
524   StringOffsetsBase = Offset;
525   Offset += Hdr.NameCount * SectionOffsetSize;
526   EntryOffsetsBase = Offset;
527   Offset += Hdr.NameCount * SectionOffsetSize;
528 
529   if (!AS.isValidOffsetForDataOfSize(Offset, Hdr.AbbrevTableSize))
530     return createStringError(errc::illegal_byte_sequence,
531                              "Section too small: cannot read abbreviations.");
532 
533   EntriesBase = Offset + Hdr.AbbrevTableSize;
534 
535   for (;;) {
536     auto AbbrevOr = extractAbbrev(&Offset);
537     if (!AbbrevOr)
538       return AbbrevOr.takeError();
539     if (isSentinel(*AbbrevOr))
540       return Error::success();
541 
542     if (!Abbrevs.insert(std::move(*AbbrevOr)).second)
543       return createStringError(errc::invalid_argument,
544                                "Duplicate abbreviation code.");
545   }
546 }
547 
548 DWARFDebugNames::Entry::Entry(const NameIndex &NameIdx, const Abbrev &Abbr)
549     : NameIdx(&NameIdx), Abbr(&Abbr) {
550   // This merely creates form values. It is up to the caller
551   // (NameIndex::getEntry) to populate them.
552   Values.reserve(Abbr.Attributes.size());
553   for (const auto &Attr : Abbr.Attributes)
554     Values.emplace_back(Attr.Form);
555 }
556 
557 Optional<DWARFFormValue>
558 DWARFDebugNames::Entry::lookup(dwarf::Index Index) const {
559   assert(Abbr->Attributes.size() == Values.size());
560   for (auto Tuple : zip_first(Abbr->Attributes, Values)) {
561     if (std::get<0>(Tuple).Index == Index)
562       return std::get<1>(Tuple);
563   }
564   return None;
565 }
566 
567 Optional<uint64_t> DWARFDebugNames::Entry::getDIEUnitOffset() const {
568   if (Optional<DWARFFormValue> Off = lookup(dwarf::DW_IDX_die_offset))
569     return Off->getAsReferenceUVal();
570   return None;
571 }
572 
573 Optional<uint64_t> DWARFDebugNames::Entry::getCUIndex() const {
574   if (Optional<DWARFFormValue> Off = lookup(dwarf::DW_IDX_compile_unit))
575     return Off->getAsUnsignedConstant();
576   // In a per-CU index, the entries without a DW_IDX_compile_unit attribute
577   // implicitly refer to the single CU.
578   if (NameIdx->getCUCount() == 1)
579     return 0;
580   return None;
581 }
582 
583 Optional<uint64_t> DWARFDebugNames::Entry::getCUOffset() const {
584   Optional<uint64_t> Index = getCUIndex();
585   if (!Index || *Index >= NameIdx->getCUCount())
586     return None;
587   return NameIdx->getCUOffset(*Index);
588 }
589 
590 void DWARFDebugNames::Entry::dump(ScopedPrinter &W) const {
591   W.printHex("Abbrev", Abbr->Code);
592   W.startLine() << formatv("Tag: {0}\n", Abbr->Tag);
593   assert(Abbr->Attributes.size() == Values.size());
594   for (auto Tuple : zip_first(Abbr->Attributes, Values)) {
595     W.startLine() << formatv("{0}: ", std::get<0>(Tuple).Index);
596     std::get<1>(Tuple).dump(W.getOStream());
597     W.getOStream() << '\n';
598   }
599 }
600 
601 char DWARFDebugNames::SentinelError::ID;
602 std::error_code DWARFDebugNames::SentinelError::convertToErrorCode() const {
603   return inconvertibleErrorCode();
604 }
605 
606 uint64_t DWARFDebugNames::NameIndex::getCUOffset(uint32_t CU) const {
607   assert(CU < Hdr.CompUnitCount);
608   const unsigned SectionOffsetSize = dwarf::getDwarfOffsetByteSize(Hdr.Format);
609   uint64_t Offset = CUsBase + SectionOffsetSize * CU;
610   return Section.AccelSection.getRelocatedValue(SectionOffsetSize, &Offset);
611 }
612 
613 uint64_t DWARFDebugNames::NameIndex::getLocalTUOffset(uint32_t TU) const {
614   assert(TU < Hdr.LocalTypeUnitCount);
615   const unsigned SectionOffsetSize = dwarf::getDwarfOffsetByteSize(Hdr.Format);
616   uint64_t Offset = CUsBase + SectionOffsetSize * (Hdr.CompUnitCount + TU);
617   return Section.AccelSection.getRelocatedValue(SectionOffsetSize, &Offset);
618 }
619 
620 uint64_t DWARFDebugNames::NameIndex::getForeignTUSignature(uint32_t TU) const {
621   assert(TU < Hdr.ForeignTypeUnitCount);
622   const unsigned SectionOffsetSize = dwarf::getDwarfOffsetByteSize(Hdr.Format);
623   uint64_t Offset =
624       CUsBase +
625       SectionOffsetSize * (Hdr.CompUnitCount + Hdr.LocalTypeUnitCount) + 8 * TU;
626   return Section.AccelSection.getU64(&Offset);
627 }
628 
629 Expected<DWARFDebugNames::Entry>
630 DWARFDebugNames::NameIndex::getEntry(uint64_t *Offset) const {
631   const DWARFDataExtractor &AS = Section.AccelSection;
632   if (!AS.isValidOffset(*Offset))
633     return createStringError(errc::illegal_byte_sequence,
634                              "Incorrectly terminated entry list.");
635 
636   uint32_t AbbrevCode = AS.getULEB128(Offset);
637   if (AbbrevCode == 0)
638     return make_error<SentinelError>();
639 
640   const auto AbbrevIt = Abbrevs.find_as(AbbrevCode);
641   if (AbbrevIt == Abbrevs.end())
642     return createStringError(errc::invalid_argument, "Invalid abbreviation.");
643 
644   Entry E(*this, *AbbrevIt);
645 
646   dwarf::FormParams FormParams = {Hdr.Version, 0, Hdr.Format};
647   for (auto &Value : E.Values) {
648     if (!Value.extractValue(AS, Offset, FormParams))
649       return createStringError(errc::io_error,
650                                "Error extracting index attribute values.");
651   }
652   return std::move(E);
653 }
654 
655 DWARFDebugNames::NameTableEntry
656 DWARFDebugNames::NameIndex::getNameTableEntry(uint32_t Index) const {
657   assert(0 < Index && Index <= Hdr.NameCount);
658   const unsigned SectionOffsetSize = dwarf::getDwarfOffsetByteSize(Hdr.Format);
659   uint64_t StringOffsetOffset =
660       StringOffsetsBase + SectionOffsetSize * (Index - 1);
661   uint64_t EntryOffsetOffset =
662       EntryOffsetsBase + SectionOffsetSize * (Index - 1);
663   const DWARFDataExtractor &AS = Section.AccelSection;
664 
665   uint64_t StringOffset =
666       AS.getRelocatedValue(SectionOffsetSize, &StringOffsetOffset);
667   uint64_t EntryOffset = AS.getUnsigned(&EntryOffsetOffset, SectionOffsetSize);
668   EntryOffset += EntriesBase;
669   return {Section.StringSection, Index, StringOffset, EntryOffset};
670 }
671 
672 uint32_t
673 DWARFDebugNames::NameIndex::getBucketArrayEntry(uint32_t Bucket) const {
674   assert(Bucket < Hdr.BucketCount);
675   uint64_t BucketOffset = BucketsBase + 4 * Bucket;
676   return Section.AccelSection.getU32(&BucketOffset);
677 }
678 
679 uint32_t DWARFDebugNames::NameIndex::getHashArrayEntry(uint32_t Index) const {
680   assert(0 < Index && Index <= Hdr.NameCount);
681   uint64_t HashOffset = HashesBase + 4 * (Index - 1);
682   return Section.AccelSection.getU32(&HashOffset);
683 }
684 
685 // Returns true if we should continue scanning for entries, false if this is the
686 // last (sentinel) entry). In case of a parsing error we also return false, as
687 // it's not possible to recover this entry list (but the other lists may still
688 // parse OK).
689 bool DWARFDebugNames::NameIndex::dumpEntry(ScopedPrinter &W,
690                                            uint64_t *Offset) const {
691   uint64_t EntryId = *Offset;
692   auto EntryOr = getEntry(Offset);
693   if (!EntryOr) {
694     handleAllErrors(EntryOr.takeError(), [](const SentinelError &) {},
695                     [&W](const ErrorInfoBase &EI) { EI.log(W.startLine()); });
696     return false;
697   }
698 
699   DictScope EntryScope(W, ("Entry @ 0x" + Twine::utohexstr(EntryId)).str());
700   EntryOr->dump(W);
701   return true;
702 }
703 
704 void DWARFDebugNames::NameIndex::dumpName(ScopedPrinter &W,
705                                           const NameTableEntry &NTE,
706                                           Optional<uint32_t> Hash) const {
707   DictScope NameScope(W, ("Name " + Twine(NTE.getIndex())).str());
708   if (Hash)
709     W.printHex("Hash", *Hash);
710 
711   W.startLine() << format("String: 0x%08" PRIx64, NTE.getStringOffset());
712   W.getOStream() << " \"" << NTE.getString() << "\"\n";
713 
714   uint64_t EntryOffset = NTE.getEntryOffset();
715   while (dumpEntry(W, &EntryOffset))
716     /*empty*/;
717 }
718 
719 void DWARFDebugNames::NameIndex::dumpCUs(ScopedPrinter &W) const {
720   ListScope CUScope(W, "Compilation Unit offsets");
721   for (uint32_t CU = 0; CU < Hdr.CompUnitCount; ++CU)
722     W.startLine() << format("CU[%u]: 0x%08" PRIx64 "\n", CU, getCUOffset(CU));
723 }
724 
725 void DWARFDebugNames::NameIndex::dumpLocalTUs(ScopedPrinter &W) const {
726   if (Hdr.LocalTypeUnitCount == 0)
727     return;
728 
729   ListScope TUScope(W, "Local Type Unit offsets");
730   for (uint32_t TU = 0; TU < Hdr.LocalTypeUnitCount; ++TU)
731     W.startLine() << format("LocalTU[%u]: 0x%08" PRIx64 "\n", TU,
732                             getLocalTUOffset(TU));
733 }
734 
735 void DWARFDebugNames::NameIndex::dumpForeignTUs(ScopedPrinter &W) const {
736   if (Hdr.ForeignTypeUnitCount == 0)
737     return;
738 
739   ListScope TUScope(W, "Foreign Type Unit signatures");
740   for (uint32_t TU = 0; TU < Hdr.ForeignTypeUnitCount; ++TU) {
741     W.startLine() << format("ForeignTU[%u]: 0x%016" PRIx64 "\n", TU,
742                             getForeignTUSignature(TU));
743   }
744 }
745 
746 void DWARFDebugNames::NameIndex::dumpAbbreviations(ScopedPrinter &W) const {
747   ListScope AbbrevsScope(W, "Abbreviations");
748   for (const auto &Abbr : Abbrevs)
749     Abbr.dump(W);
750 }
751 
752 void DWARFDebugNames::NameIndex::dumpBucket(ScopedPrinter &W,
753                                             uint32_t Bucket) const {
754   ListScope BucketScope(W, ("Bucket " + Twine(Bucket)).str());
755   uint32_t Index = getBucketArrayEntry(Bucket);
756   if (Index == 0) {
757     W.printString("EMPTY");
758     return;
759   }
760   if (Index > Hdr.NameCount) {
761     W.printString("Name index is invalid");
762     return;
763   }
764 
765   for (; Index <= Hdr.NameCount; ++Index) {
766     uint32_t Hash = getHashArrayEntry(Index);
767     if (Hash % Hdr.BucketCount != Bucket)
768       break;
769 
770     dumpName(W, getNameTableEntry(Index), Hash);
771   }
772 }
773 
774 LLVM_DUMP_METHOD void DWARFDebugNames::NameIndex::dump(ScopedPrinter &W) const {
775   DictScope UnitScope(W, ("Name Index @ 0x" + Twine::utohexstr(Base)).str());
776   Hdr.dump(W);
777   dumpCUs(W);
778   dumpLocalTUs(W);
779   dumpForeignTUs(W);
780   dumpAbbreviations(W);
781 
782   if (Hdr.BucketCount > 0) {
783     for (uint32_t Bucket = 0; Bucket < Hdr.BucketCount; ++Bucket)
784       dumpBucket(W, Bucket);
785     return;
786   }
787 
788   W.startLine() << "Hash table not present\n";
789   for (NameTableEntry NTE : *this)
790     dumpName(W, NTE, None);
791 }
792 
793 Error DWARFDebugNames::extract() {
794   uint64_t Offset = 0;
795   while (AccelSection.isValidOffset(Offset)) {
796     NameIndex Next(*this, Offset);
797     if (Error E = Next.extract())
798       return E;
799     Offset = Next.getNextUnitOffset();
800     NameIndices.push_back(std::move(Next));
801   }
802   return Error::success();
803 }
804 
805 iterator_range<DWARFDebugNames::ValueIterator>
806 DWARFDebugNames::NameIndex::equal_range(StringRef Key) const {
807   return make_range(ValueIterator(*this, Key), ValueIterator());
808 }
809 
810 LLVM_DUMP_METHOD void DWARFDebugNames::dump(raw_ostream &OS) const {
811   ScopedPrinter W(OS);
812   for (const NameIndex &NI : NameIndices)
813     NI.dump(W);
814 }
815 
816 Optional<uint64_t>
817 DWARFDebugNames::ValueIterator::findEntryOffsetInCurrentIndex() {
818   const Header &Hdr = CurrentIndex->Hdr;
819   if (Hdr.BucketCount == 0) {
820     // No Hash Table, We need to search through all names in the Name Index.
821     for (NameTableEntry NTE : *CurrentIndex) {
822       if (NTE.getString() == Key)
823         return NTE.getEntryOffset();
824     }
825     return None;
826   }
827 
828   // The Name Index has a Hash Table, so use that to speed up the search.
829   // Compute the Key Hash, if it has not been done already.
830   if (!Hash)
831     Hash = caseFoldingDjbHash(Key);
832   uint32_t Bucket = *Hash % Hdr.BucketCount;
833   uint32_t Index = CurrentIndex->getBucketArrayEntry(Bucket);
834   if (Index == 0)
835     return None; // Empty bucket
836 
837   for (; Index <= Hdr.NameCount; ++Index) {
838     uint32_t Hash = CurrentIndex->getHashArrayEntry(Index);
839     if (Hash % Hdr.BucketCount != Bucket)
840       return None; // End of bucket
841 
842     NameTableEntry NTE = CurrentIndex->getNameTableEntry(Index);
843     if (NTE.getString() == Key)
844       return NTE.getEntryOffset();
845   }
846   return None;
847 }
848 
849 bool DWARFDebugNames::ValueIterator::getEntryAtCurrentOffset() {
850   auto EntryOr = CurrentIndex->getEntry(&DataOffset);
851   if (!EntryOr) {
852     consumeError(EntryOr.takeError());
853     return false;
854   }
855   CurrentEntry = std::move(*EntryOr);
856   return true;
857 }
858 
859 bool DWARFDebugNames::ValueIterator::findInCurrentIndex() {
860   Optional<uint64_t> Offset = findEntryOffsetInCurrentIndex();
861   if (!Offset)
862     return false;
863   DataOffset = *Offset;
864   return getEntryAtCurrentOffset();
865 }
866 
867 void DWARFDebugNames::ValueIterator::searchFromStartOfCurrentIndex() {
868   for (const NameIndex *End = CurrentIndex->Section.NameIndices.end();
869        CurrentIndex != End; ++CurrentIndex) {
870     if (findInCurrentIndex())
871       return;
872   }
873   setEnd();
874 }
875 
876 void DWARFDebugNames::ValueIterator::next() {
877   assert(CurrentIndex && "Incrementing an end() iterator?");
878 
879   // First try the next entry in the current Index.
880   if (getEntryAtCurrentOffset())
881     return;
882 
883   // If we're a local iterator or we have reached the last Index, we're done.
884   if (IsLocal || CurrentIndex == &CurrentIndex->Section.NameIndices.back()) {
885     setEnd();
886     return;
887   }
888 
889   // Otherwise, try the next index.
890   ++CurrentIndex;
891   searchFromStartOfCurrentIndex();
892 }
893 
894 DWARFDebugNames::ValueIterator::ValueIterator(const DWARFDebugNames &AccelTable,
895                                               StringRef Key)
896     : CurrentIndex(AccelTable.NameIndices.begin()), IsLocal(false),
897       Key(std::string(Key)) {
898   searchFromStartOfCurrentIndex();
899 }
900 
901 DWARFDebugNames::ValueIterator::ValueIterator(
902     const DWARFDebugNames::NameIndex &NI, StringRef Key)
903     : CurrentIndex(&NI), IsLocal(true), Key(std::string(Key)) {
904   if (!findInCurrentIndex())
905     setEnd();
906 }
907 
908 iterator_range<DWARFDebugNames::ValueIterator>
909 DWARFDebugNames::equal_range(StringRef Key) const {
910   if (NameIndices.empty())
911     return make_range(ValueIterator(), ValueIterator());
912   return make_range(ValueIterator(*this, Key), ValueIterator());
913 }
914 
915 const DWARFDebugNames::NameIndex *
916 DWARFDebugNames::getCUNameIndex(uint64_t CUOffset) {
917   if (CUToNameIndex.size() == 0 && NameIndices.size() > 0) {
918     for (const auto &NI : *this) {
919       for (uint32_t CU = 0; CU < NI.getCUCount(); ++CU)
920         CUToNameIndex.try_emplace(NI.getCUOffset(CU), &NI);
921     }
922   }
923   return CUToNameIndex.lookup(CUOffset);
924 }
925