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