xref: /freebsd/contrib/llvm-project/llvm/lib/DebugInfo/DWARF/DWARFUnit.cpp (revision 6966ac055c3b7a39266fb982493330df7a097997)
1 //===- DWARFUnit.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/DWARFUnit.h"
10 #include "llvm/ADT/SmallString.h"
11 #include "llvm/ADT/StringRef.h"
12 #include "llvm/DebugInfo/DWARF/DWARFAbbreviationDeclaration.h"
13 #include "llvm/DebugInfo/DWARF/DWARFCompileUnit.h"
14 #include "llvm/DebugInfo/DWARF/DWARFContext.h"
15 #include "llvm/DebugInfo/DWARF/DWARFDebugAbbrev.h"
16 #include "llvm/DebugInfo/DWARF/DWARFDebugInfoEntry.h"
17 #include "llvm/DebugInfo/DWARF/DWARFDebugRnglists.h"
18 #include "llvm/DebugInfo/DWARF/DWARFDie.h"
19 #include "llvm/DebugInfo/DWARF/DWARFFormValue.h"
20 #include "llvm/DebugInfo/DWARF/DWARFTypeUnit.h"
21 #include "llvm/Support/DataExtractor.h"
22 #include "llvm/Support/Errc.h"
23 #include "llvm/Support/Path.h"
24 #include "llvm/Support/WithColor.h"
25 #include <algorithm>
26 #include <cassert>
27 #include <cstddef>
28 #include <cstdint>
29 #include <cstdio>
30 #include <utility>
31 #include <vector>
32 
33 using namespace llvm;
34 using namespace dwarf;
35 
36 void DWARFUnitVector::addUnitsForSection(DWARFContext &C,
37                                          const DWARFSection &Section,
38                                          DWARFSectionKind SectionKind) {
39   const DWARFObject &D = C.getDWARFObj();
40   addUnitsImpl(C, D, Section, C.getDebugAbbrev(), &D.getRangeSection(),
41                &D.getLocSection(), D.getStringSection(),
42                D.getStringOffsetSection(), &D.getAddrSection(),
43                D.getLineSection(), D.isLittleEndian(), false, false,
44                SectionKind);
45 }
46 
47 void DWARFUnitVector::addUnitsForDWOSection(DWARFContext &C,
48                                             const DWARFSection &DWOSection,
49                                             DWARFSectionKind SectionKind,
50                                             bool Lazy) {
51   const DWARFObject &D = C.getDWARFObj();
52   addUnitsImpl(C, D, DWOSection, C.getDebugAbbrevDWO(), &D.getRangeDWOSection(),
53                &D.getLocDWOSection(), D.getStringDWOSection(),
54                D.getStringOffsetDWOSection(), &D.getAddrSection(),
55                D.getLineDWOSection(), C.isLittleEndian(), true, Lazy,
56                SectionKind);
57 }
58 
59 void DWARFUnitVector::addUnitsImpl(
60     DWARFContext &Context, const DWARFObject &Obj, const DWARFSection &Section,
61     const DWARFDebugAbbrev *DA, const DWARFSection *RS,
62     const DWARFSection *LocSection, StringRef SS, const DWARFSection &SOS,
63     const DWARFSection *AOS, const DWARFSection &LS, bool LE, bool IsDWO,
64     bool Lazy, DWARFSectionKind SectionKind) {
65   DWARFDataExtractor Data(Obj, Section, LE, 0);
66   // Lazy initialization of Parser, now that we have all section info.
67   if (!Parser) {
68     Parser = [=, &Context, &Obj, &Section, &SOS,
69               &LS](uint32_t Offset, DWARFSectionKind SectionKind,
70                    const DWARFSection *CurSection,
71                    const DWARFUnitIndex::Entry *IndexEntry)
72         -> std::unique_ptr<DWARFUnit> {
73       const DWARFSection &InfoSection = CurSection ? *CurSection : Section;
74       DWARFDataExtractor Data(Obj, InfoSection, LE, 0);
75       if (!Data.isValidOffset(Offset))
76         return nullptr;
77       const DWARFUnitIndex *Index = nullptr;
78       if (IsDWO)
79         Index = &getDWARFUnitIndex(Context, SectionKind);
80       DWARFUnitHeader Header;
81       if (!Header.extract(Context, Data, &Offset, SectionKind, Index,
82                           IndexEntry))
83         return nullptr;
84       std::unique_ptr<DWARFUnit> U;
85       if (Header.isTypeUnit())
86         U = llvm::make_unique<DWARFTypeUnit>(Context, InfoSection, Header, DA,
87                                              RS, LocSection, SS, SOS, AOS, LS,
88                                              LE, IsDWO, *this);
89       else
90         U = llvm::make_unique<DWARFCompileUnit>(Context, InfoSection, Header,
91                                                 DA, RS, LocSection, SS, SOS,
92                                                 AOS, LS, LE, IsDWO, *this);
93       return U;
94     };
95   }
96   if (Lazy)
97     return;
98   // Find a reasonable insertion point within the vector.  We skip over
99   // (a) units from a different section, (b) units from the same section
100   // but with lower offset-within-section.  This keeps units in order
101   // within a section, although not necessarily within the object file,
102   // even if we do lazy parsing.
103   auto I = this->begin();
104   uint32_t Offset = 0;
105   while (Data.isValidOffset(Offset)) {
106     if (I != this->end() &&
107         (&(*I)->getInfoSection() != &Section || (*I)->getOffset() == Offset)) {
108       ++I;
109       continue;
110     }
111     auto U = Parser(Offset, SectionKind, &Section, nullptr);
112     // If parsing failed, we're done with this section.
113     if (!U)
114       break;
115     Offset = U->getNextUnitOffset();
116     I = std::next(this->insert(I, std::move(U)));
117   }
118 }
119 
120 DWARFUnit *DWARFUnitVector::addUnit(std::unique_ptr<DWARFUnit> Unit) {
121   auto I = std::upper_bound(begin(), end(), Unit,
122                             [](const std::unique_ptr<DWARFUnit> &LHS,
123                                const std::unique_ptr<DWARFUnit> &RHS) {
124                               return LHS->getOffset() < RHS->getOffset();
125                             });
126   return this->insert(I, std::move(Unit))->get();
127 }
128 
129 DWARFUnit *DWARFUnitVector::getUnitForOffset(uint32_t Offset) const {
130   auto end = begin() + getNumInfoUnits();
131   auto *CU =
132       std::upper_bound(begin(), end, Offset,
133                        [](uint32_t LHS, const std::unique_ptr<DWARFUnit> &RHS) {
134                          return LHS < RHS->getNextUnitOffset();
135                        });
136   if (CU != end && (*CU)->getOffset() <= Offset)
137     return CU->get();
138   return nullptr;
139 }
140 
141 DWARFUnit *
142 DWARFUnitVector::getUnitForIndexEntry(const DWARFUnitIndex::Entry &E) {
143   const auto *CUOff = E.getOffset(DW_SECT_INFO);
144   if (!CUOff)
145     return nullptr;
146 
147   auto Offset = CUOff->Offset;
148   auto end = begin() + getNumInfoUnits();
149 
150   auto *CU =
151       std::upper_bound(begin(), end, CUOff->Offset,
152                        [](uint32_t LHS, const std::unique_ptr<DWARFUnit> &RHS) {
153                          return LHS < RHS->getNextUnitOffset();
154                        });
155   if (CU != end && (*CU)->getOffset() <= Offset)
156     return CU->get();
157 
158   if (!Parser)
159     return nullptr;
160 
161   auto U = Parser(Offset, DW_SECT_INFO, nullptr, &E);
162   if (!U)
163     U = nullptr;
164 
165   auto *NewCU = U.get();
166   this->insert(CU, std::move(U));
167   ++NumInfoUnits;
168   return NewCU;
169 }
170 
171 DWARFUnit::DWARFUnit(DWARFContext &DC, const DWARFSection &Section,
172                      const DWARFUnitHeader &Header, const DWARFDebugAbbrev *DA,
173                      const DWARFSection *RS, const DWARFSection *LocSection,
174                      StringRef SS, const DWARFSection &SOS,
175                      const DWARFSection *AOS, const DWARFSection &LS, bool LE,
176                      bool IsDWO, const DWARFUnitVector &UnitVector)
177     : Context(DC), InfoSection(Section), Header(Header), Abbrev(DA),
178       RangeSection(RS), LocSection(LocSection), LineSection(LS),
179       StringSection(SS), StringOffsetSection(SOS), AddrOffsetSection(AOS),
180       isLittleEndian(LE), IsDWO(IsDWO), UnitVector(UnitVector) {
181   clear();
182   // For split DWARF we only need to keep track of the location list section's
183   // data (no relocations), and if we are reading a package file, we need to
184   // adjust the location list data based on the index entries.
185   if (IsDWO) {
186     LocSectionData = LocSection->Data;
187     if (auto *IndexEntry = Header.getIndexEntry())
188       if (const auto *C = IndexEntry->getOffset(DW_SECT_LOC))
189         LocSectionData = LocSectionData.substr(C->Offset, C->Length);
190   }
191 }
192 
193 DWARFUnit::~DWARFUnit() = default;
194 
195 DWARFDataExtractor DWARFUnit::getDebugInfoExtractor() const {
196   return DWARFDataExtractor(Context.getDWARFObj(), InfoSection, isLittleEndian,
197                             getAddressByteSize());
198 }
199 
200 Optional<object::SectionedAddress>
201 DWARFUnit::getAddrOffsetSectionItem(uint32_t Index) const {
202   if (IsDWO) {
203     auto R = Context.info_section_units();
204     auto I = R.begin();
205     // Surprising if a DWO file has more than one skeleton unit in it - this
206     // probably shouldn't be valid, but if a use case is found, here's where to
207     // support it (probably have to linearly search for the matching skeleton CU
208     // here)
209     if (I != R.end() && std::next(I) == R.end())
210       return (*I)->getAddrOffsetSectionItem(Index);
211   }
212   uint32_t Offset = AddrOffsetSectionBase + Index * getAddressByteSize();
213   if (AddrOffsetSection->Data.size() < Offset + getAddressByteSize())
214     return None;
215   DWARFDataExtractor DA(Context.getDWARFObj(), *AddrOffsetSection,
216                         isLittleEndian, getAddressByteSize());
217   uint64_t Section;
218   uint64_t Address = DA.getRelocatedAddress(&Offset, &Section);
219   return {{Address, Section}};
220 }
221 
222 Optional<uint64_t> DWARFUnit::getStringOffsetSectionItem(uint32_t Index) const {
223   if (!StringOffsetsTableContribution)
224     return None;
225   unsigned ItemSize = getDwarfStringOffsetsByteSize();
226   uint32_t Offset = getStringOffsetsBase() + Index * ItemSize;
227   if (StringOffsetSection.Data.size() < Offset + ItemSize)
228     return None;
229   DWARFDataExtractor DA(Context.getDWARFObj(), StringOffsetSection,
230                         isLittleEndian, 0);
231   return DA.getRelocatedValue(ItemSize, &Offset);
232 }
233 
234 bool DWARFUnitHeader::extract(DWARFContext &Context,
235                               const DWARFDataExtractor &debug_info,
236                               uint32_t *offset_ptr,
237                               DWARFSectionKind SectionKind,
238                               const DWARFUnitIndex *Index,
239                               const DWARFUnitIndex::Entry *Entry) {
240   Offset = *offset_ptr;
241   IndexEntry = Entry;
242   if (!IndexEntry && Index)
243     IndexEntry = Index->getFromOffset(*offset_ptr);
244   Length = debug_info.getRelocatedValue(4, offset_ptr);
245   FormParams.Format = DWARF32;
246   unsigned SizeOfLength = 4;
247   if (Length == 0xffffffff) {
248     Length = debug_info.getU64(offset_ptr);
249     FormParams.Format = DWARF64;
250     SizeOfLength = 8;
251   }
252   FormParams.Version = debug_info.getU16(offset_ptr);
253   if (FormParams.Version >= 5) {
254     UnitType = debug_info.getU8(offset_ptr);
255     FormParams.AddrSize = debug_info.getU8(offset_ptr);
256     AbbrOffset = debug_info.getRelocatedValue(FormParams.getDwarfOffsetByteSize(), offset_ptr);
257   } else {
258     AbbrOffset = debug_info.getRelocatedValue(FormParams.getDwarfOffsetByteSize(), offset_ptr);
259     FormParams.AddrSize = debug_info.getU8(offset_ptr);
260     // Fake a unit type based on the section type.  This isn't perfect,
261     // but distinguishing compile and type units is generally enough.
262     if (SectionKind == DW_SECT_TYPES)
263       UnitType = DW_UT_type;
264     else
265       UnitType = DW_UT_compile;
266   }
267   if (IndexEntry) {
268     if (AbbrOffset)
269       return false;
270     auto *UnitContrib = IndexEntry->getOffset();
271     if (!UnitContrib || UnitContrib->Length != (Length + 4))
272       return false;
273     auto *AbbrEntry = IndexEntry->getOffset(DW_SECT_ABBREV);
274     if (!AbbrEntry)
275       return false;
276     AbbrOffset = AbbrEntry->Offset;
277   }
278   if (isTypeUnit()) {
279     TypeHash = debug_info.getU64(offset_ptr);
280     TypeOffset = debug_info.getU32(offset_ptr);
281   } else if (UnitType == DW_UT_split_compile || UnitType == DW_UT_skeleton)
282     DWOId = debug_info.getU64(offset_ptr);
283 
284   // Header fields all parsed, capture the size of this unit header.
285   assert(*offset_ptr - Offset <= 255 && "unexpected header size");
286   Size = uint8_t(*offset_ptr - Offset);
287 
288   // Type offset is unit-relative; should be after the header and before
289   // the end of the current unit.
290   bool TypeOffsetOK =
291       !isTypeUnit()
292           ? true
293           : TypeOffset >= Size && TypeOffset < getLength() + SizeOfLength;
294   bool LengthOK = debug_info.isValidOffset(getNextUnitOffset() - 1);
295   bool VersionOK = DWARFContext::isSupportedVersion(getVersion());
296   bool AddrSizeOK = getAddressByteSize() == 4 || getAddressByteSize() == 8;
297 
298   if (!LengthOK || !VersionOK || !AddrSizeOK || !TypeOffsetOK)
299     return false;
300 
301   // Keep track of the highest DWARF version we encounter across all units.
302   Context.setMaxVersionIfGreater(getVersion());
303   return true;
304 }
305 
306 // Parse the rangelist table header, including the optional array of offsets
307 // following it (DWARF v5 and later).
308 static Expected<DWARFDebugRnglistTable>
309 parseRngListTableHeader(DWARFDataExtractor &DA, uint32_t Offset) {
310   // TODO: Support DWARF64
311   // We are expected to be called with Offset 0 or pointing just past the table
312   // header, which is 12 bytes long for DWARF32.
313   if (Offset > 0) {
314     if (Offset < 12U)
315       return createStringError(errc::invalid_argument, "Did not detect a valid"
316                                " range list table with base = 0x%" PRIu32,
317                                Offset);
318     Offset -= 12U;
319   }
320   llvm::DWARFDebugRnglistTable Table;
321   if (Error E = Table.extractHeaderAndOffsets(DA, &Offset))
322     return std::move(E);
323   return Table;
324 }
325 
326 Error DWARFUnit::extractRangeList(uint32_t RangeListOffset,
327                                   DWARFDebugRangeList &RangeList) const {
328   // Require that compile unit is extracted.
329   assert(!DieArray.empty());
330   DWARFDataExtractor RangesData(Context.getDWARFObj(), *RangeSection,
331                                 isLittleEndian, getAddressByteSize());
332   uint32_t ActualRangeListOffset = RangeSectionBase + RangeListOffset;
333   return RangeList.extract(RangesData, &ActualRangeListOffset);
334 }
335 
336 void DWARFUnit::clear() {
337   Abbrevs = nullptr;
338   BaseAddr.reset();
339   RangeSectionBase = 0;
340   AddrOffsetSectionBase = 0;
341   clearDIEs(false);
342   DWO.reset();
343 }
344 
345 const char *DWARFUnit::getCompilationDir() {
346   return dwarf::toString(getUnitDIE().find(DW_AT_comp_dir), nullptr);
347 }
348 
349 void DWARFUnit::extractDIEsToVector(
350     bool AppendCUDie, bool AppendNonCUDies,
351     std::vector<DWARFDebugInfoEntry> &Dies) const {
352   if (!AppendCUDie && !AppendNonCUDies)
353     return;
354 
355   // Set the offset to that of the first DIE and calculate the start of the
356   // next compilation unit header.
357   uint32_t DIEOffset = getOffset() + getHeaderSize();
358   uint32_t NextCUOffset = getNextUnitOffset();
359   DWARFDebugInfoEntry DIE;
360   DWARFDataExtractor DebugInfoData = getDebugInfoExtractor();
361   uint32_t Depth = 0;
362   bool IsCUDie = true;
363 
364   while (DIE.extractFast(*this, &DIEOffset, DebugInfoData, NextCUOffset,
365                          Depth)) {
366     if (IsCUDie) {
367       if (AppendCUDie)
368         Dies.push_back(DIE);
369       if (!AppendNonCUDies)
370         break;
371       // The average bytes per DIE entry has been seen to be
372       // around 14-20 so let's pre-reserve the needed memory for
373       // our DIE entries accordingly.
374       Dies.reserve(Dies.size() + getDebugInfoSize() / 14);
375       IsCUDie = false;
376     } else {
377       Dies.push_back(DIE);
378     }
379 
380     if (const DWARFAbbreviationDeclaration *AbbrDecl =
381             DIE.getAbbreviationDeclarationPtr()) {
382       // Normal DIE
383       if (AbbrDecl->hasChildren())
384         ++Depth;
385     } else {
386       // NULL DIE.
387       if (Depth > 0)
388         --Depth;
389       if (Depth == 0)
390         break;  // We are done with this compile unit!
391     }
392   }
393 
394   // Give a little bit of info if we encounter corrupt DWARF (our offset
395   // should always terminate at or before the start of the next compilation
396   // unit header).
397   if (DIEOffset > NextCUOffset)
398     WithColor::warning() << format("DWARF compile unit extends beyond its "
399                                    "bounds cu 0x%8.8x at 0x%8.8x\n",
400                                    getOffset(), DIEOffset);
401 }
402 
403 size_t DWARFUnit::extractDIEsIfNeeded(bool CUDieOnly) {
404   if ((CUDieOnly && !DieArray.empty()) ||
405       DieArray.size() > 1)
406     return 0; // Already parsed.
407 
408   bool HasCUDie = !DieArray.empty();
409   extractDIEsToVector(!HasCUDie, !CUDieOnly, DieArray);
410 
411   if (DieArray.empty())
412     return 0;
413 
414   // If CU DIE was just parsed, copy several attribute values from it.
415   if (!HasCUDie) {
416     DWARFDie UnitDie = getUnitDIE();
417     if (Optional<uint64_t> DWOId = toUnsigned(UnitDie.find(DW_AT_GNU_dwo_id)))
418       Header.setDWOId(*DWOId);
419     if (!IsDWO) {
420       assert(AddrOffsetSectionBase == 0);
421       assert(RangeSectionBase == 0);
422       AddrOffsetSectionBase = toSectionOffset(UnitDie.find(DW_AT_addr_base), 0);
423       if (!AddrOffsetSectionBase)
424         AddrOffsetSectionBase =
425             toSectionOffset(UnitDie.find(DW_AT_GNU_addr_base), 0);
426       RangeSectionBase = toSectionOffset(UnitDie.find(DW_AT_rnglists_base), 0);
427     }
428 
429     // In general, in DWARF v5 and beyond we derive the start of the unit's
430     // contribution to the string offsets table from the unit DIE's
431     // DW_AT_str_offsets_base attribute. Split DWARF units do not use this
432     // attribute, so we assume that there is a contribution to the string
433     // offsets table starting at offset 0 of the debug_str_offsets.dwo section.
434     // In both cases we need to determine the format of the contribution,
435     // which may differ from the unit's format.
436     DWARFDataExtractor DA(Context.getDWARFObj(), StringOffsetSection,
437                           isLittleEndian, 0);
438     if (IsDWO || getVersion() >= 5) {
439       auto StringOffsetOrError =
440           IsDWO ? determineStringOffsetsTableContributionDWO(DA)
441                 : determineStringOffsetsTableContribution(DA);
442       if (!StringOffsetOrError) {
443         WithColor::error() << "invalid contribution to string offsets table in section .debug_str_offsets[.dwo]: "
444                            << toString(StringOffsetOrError.takeError()) << '\n';
445       } else {
446         StringOffsetsTableContribution = *StringOffsetOrError;
447       }
448     }
449 
450     // DWARF v5 uses the .debug_rnglists and .debug_rnglists.dwo sections to
451     // describe address ranges.
452     if (getVersion() >= 5) {
453       if (IsDWO)
454         setRangesSection(&Context.getDWARFObj().getRnglistsDWOSection(), 0);
455       else
456         setRangesSection(&Context.getDWARFObj().getRnglistsSection(),
457                          toSectionOffset(UnitDie.find(DW_AT_rnglists_base), 0));
458       if (RangeSection->Data.size()) {
459         // Parse the range list table header. Individual range lists are
460         // extracted lazily.
461         DWARFDataExtractor RangesDA(Context.getDWARFObj(), *RangeSection,
462                                     isLittleEndian, 0);
463         if (auto TableOrError =
464                 parseRngListTableHeader(RangesDA, RangeSectionBase))
465           RngListTable = TableOrError.get();
466         else
467           WithColor::error() << "parsing a range list table: "
468                              << toString(TableOrError.takeError())
469                              << '\n';
470 
471         // In a split dwarf unit, there is no DW_AT_rnglists_base attribute.
472         // Adjust RangeSectionBase to point past the table header.
473         if (IsDWO && RngListTable)
474           RangeSectionBase = RngListTable->getHeaderSize();
475       }
476     }
477 
478     // Don't fall back to DW_AT_GNU_ranges_base: it should be ignored for
479     // skeleton CU DIE, so that DWARF users not aware of it are not broken.
480     }
481 
482   return DieArray.size();
483 }
484 
485 bool DWARFUnit::parseDWO() {
486   if (IsDWO)
487     return false;
488   if (DWO.get())
489     return false;
490   DWARFDie UnitDie = getUnitDIE();
491   if (!UnitDie)
492     return false;
493   auto DWOFileName = dwarf::toString(UnitDie.find(DW_AT_GNU_dwo_name));
494   if (!DWOFileName)
495     return false;
496   auto CompilationDir = dwarf::toString(UnitDie.find(DW_AT_comp_dir));
497   SmallString<16> AbsolutePath;
498   if (sys::path::is_relative(*DWOFileName) && CompilationDir &&
499       *CompilationDir) {
500     sys::path::append(AbsolutePath, *CompilationDir);
501   }
502   sys::path::append(AbsolutePath, *DWOFileName);
503   auto DWOId = getDWOId();
504   if (!DWOId)
505     return false;
506   auto DWOContext = Context.getDWOContext(AbsolutePath);
507   if (!DWOContext)
508     return false;
509 
510   DWARFCompileUnit *DWOCU = DWOContext->getDWOCompileUnitForHash(*DWOId);
511   if (!DWOCU)
512     return false;
513   DWO = std::shared_ptr<DWARFCompileUnit>(std::move(DWOContext), DWOCU);
514   // Share .debug_addr and .debug_ranges section with compile unit in .dwo
515   DWO->setAddrOffsetSection(AddrOffsetSection, AddrOffsetSectionBase);
516   if (getVersion() >= 5) {
517     DWO->setRangesSection(&Context.getDWARFObj().getRnglistsDWOSection(), 0);
518     DWARFDataExtractor RangesDA(Context.getDWARFObj(), *RangeSection,
519                                 isLittleEndian, 0);
520     if (auto TableOrError = parseRngListTableHeader(RangesDA, RangeSectionBase))
521       DWO->RngListTable = TableOrError.get();
522     else
523       WithColor::error() << "parsing a range list table: "
524                          << toString(TableOrError.takeError())
525                          << '\n';
526     if (DWO->RngListTable)
527       DWO->RangeSectionBase = DWO->RngListTable->getHeaderSize();
528   } else {
529     auto DWORangesBase = UnitDie.getRangesBaseAttribute();
530     DWO->setRangesSection(RangeSection, DWORangesBase ? *DWORangesBase : 0);
531   }
532 
533   return true;
534 }
535 
536 void DWARFUnit::clearDIEs(bool KeepCUDie) {
537   if (DieArray.size() > (unsigned)KeepCUDie) {
538     DieArray.resize((unsigned)KeepCUDie);
539     DieArray.shrink_to_fit();
540   }
541 }
542 
543 Expected<DWARFAddressRangesVector>
544 DWARFUnit::findRnglistFromOffset(uint32_t Offset) {
545   if (getVersion() <= 4) {
546     DWARFDebugRangeList RangeList;
547     if (Error E = extractRangeList(Offset, RangeList))
548       return std::move(E);
549     return RangeList.getAbsoluteRanges(getBaseAddress());
550   }
551   if (RngListTable) {
552     DWARFDataExtractor RangesData(Context.getDWARFObj(), *RangeSection,
553                                   isLittleEndian, RngListTable->getAddrSize());
554     auto RangeListOrError = RngListTable->findList(RangesData, Offset);
555     if (RangeListOrError)
556       return RangeListOrError.get().getAbsoluteRanges(getBaseAddress(), *this);
557     return RangeListOrError.takeError();
558   }
559 
560   return createStringError(errc::invalid_argument,
561                            "missing or invalid range list table");
562 }
563 
564 Expected<DWARFAddressRangesVector>
565 DWARFUnit::findRnglistFromIndex(uint32_t Index) {
566   if (auto Offset = getRnglistOffset(Index))
567     return findRnglistFromOffset(*Offset + RangeSectionBase);
568 
569   if (RngListTable)
570     return createStringError(errc::invalid_argument,
571                              "invalid range list table index %d", Index);
572   else
573     return createStringError(errc::invalid_argument,
574                              "missing or invalid range list table");
575 }
576 
577 Expected<DWARFAddressRangesVector> DWARFUnit::collectAddressRanges() {
578   DWARFDie UnitDie = getUnitDIE();
579   if (!UnitDie)
580     return createStringError(errc::invalid_argument, "No unit DIE");
581 
582   // First, check if unit DIE describes address ranges for the whole unit.
583   auto CUDIERangesOrError = UnitDie.getAddressRanges();
584   if (!CUDIERangesOrError)
585     return createStringError(errc::invalid_argument,
586                              "decoding address ranges: %s",
587                              toString(CUDIERangesOrError.takeError()).c_str());
588   return *CUDIERangesOrError;
589 }
590 
591 void DWARFUnit::updateAddressDieMap(DWARFDie Die) {
592   if (Die.isSubroutineDIE()) {
593     auto DIERangesOrError = Die.getAddressRanges();
594     if (DIERangesOrError) {
595       for (const auto &R : DIERangesOrError.get()) {
596         // Ignore 0-sized ranges.
597         if (R.LowPC == R.HighPC)
598           continue;
599         auto B = AddrDieMap.upper_bound(R.LowPC);
600         if (B != AddrDieMap.begin() && R.LowPC < (--B)->second.first) {
601           // The range is a sub-range of existing ranges, we need to split the
602           // existing range.
603           if (R.HighPC < B->second.first)
604             AddrDieMap[R.HighPC] = B->second;
605           if (R.LowPC > B->first)
606             AddrDieMap[B->first].first = R.LowPC;
607         }
608         AddrDieMap[R.LowPC] = std::make_pair(R.HighPC, Die);
609       }
610     } else
611       llvm::consumeError(DIERangesOrError.takeError());
612   }
613   // Parent DIEs are added to the AddrDieMap prior to the Children DIEs to
614   // simplify the logic to update AddrDieMap. The child's range will always
615   // be equal or smaller than the parent's range. With this assumption, when
616   // adding one range into the map, it will at most split a range into 3
617   // sub-ranges.
618   for (DWARFDie Child = Die.getFirstChild(); Child; Child = Child.getSibling())
619     updateAddressDieMap(Child);
620 }
621 
622 DWARFDie DWARFUnit::getSubroutineForAddress(uint64_t Address) {
623   extractDIEsIfNeeded(false);
624   if (AddrDieMap.empty())
625     updateAddressDieMap(getUnitDIE());
626   auto R = AddrDieMap.upper_bound(Address);
627   if (R == AddrDieMap.begin())
628     return DWARFDie();
629   // upper_bound's previous item contains Address.
630   --R;
631   if (Address >= R->second.first)
632     return DWARFDie();
633   return R->second.second;
634 }
635 
636 void
637 DWARFUnit::getInlinedChainForAddress(uint64_t Address,
638                                      SmallVectorImpl<DWARFDie> &InlinedChain) {
639   assert(InlinedChain.empty());
640   // Try to look for subprogram DIEs in the DWO file.
641   parseDWO();
642   // First, find the subroutine that contains the given address (the leaf
643   // of inlined chain).
644   DWARFDie SubroutineDIE =
645       (DWO ? *DWO : *this).getSubroutineForAddress(Address);
646 
647   if (!SubroutineDIE)
648     return;
649 
650   while (!SubroutineDIE.isSubprogramDIE()) {
651     if (SubroutineDIE.getTag() == DW_TAG_inlined_subroutine)
652       InlinedChain.push_back(SubroutineDIE);
653     SubroutineDIE  = SubroutineDIE.getParent();
654   }
655   InlinedChain.push_back(SubroutineDIE);
656 }
657 
658 const DWARFUnitIndex &llvm::getDWARFUnitIndex(DWARFContext &Context,
659                                               DWARFSectionKind Kind) {
660   if (Kind == DW_SECT_INFO)
661     return Context.getCUIndex();
662   assert(Kind == DW_SECT_TYPES);
663   return Context.getTUIndex();
664 }
665 
666 DWARFDie DWARFUnit::getParent(const DWARFDebugInfoEntry *Die) {
667   if (!Die)
668     return DWARFDie();
669   const uint32_t Depth = Die->getDepth();
670   // Unit DIEs always have a depth of zero and never have parents.
671   if (Depth == 0)
672     return DWARFDie();
673   // Depth of 1 always means parent is the compile/type unit.
674   if (Depth == 1)
675     return getUnitDIE();
676   // Look for previous DIE with a depth that is one less than the Die's depth.
677   const uint32_t ParentDepth = Depth - 1;
678   for (uint32_t I = getDIEIndex(Die) - 1; I > 0; --I) {
679     if (DieArray[I].getDepth() == ParentDepth)
680       return DWARFDie(this, &DieArray[I]);
681   }
682   return DWARFDie();
683 }
684 
685 DWARFDie DWARFUnit::getSibling(const DWARFDebugInfoEntry *Die) {
686   if (!Die)
687     return DWARFDie();
688   uint32_t Depth = Die->getDepth();
689   // Unit DIEs always have a depth of zero and never have siblings.
690   if (Depth == 0)
691     return DWARFDie();
692   // NULL DIEs don't have siblings.
693   if (Die->getAbbreviationDeclarationPtr() == nullptr)
694     return DWARFDie();
695 
696   // Find the next DIE whose depth is the same as the Die's depth.
697   for (size_t I = getDIEIndex(Die) + 1, EndIdx = DieArray.size(); I < EndIdx;
698        ++I) {
699     if (DieArray[I].getDepth() == Depth)
700       return DWARFDie(this, &DieArray[I]);
701   }
702   return DWARFDie();
703 }
704 
705 DWARFDie DWARFUnit::getPreviousSibling(const DWARFDebugInfoEntry *Die) {
706   if (!Die)
707     return DWARFDie();
708   uint32_t Depth = Die->getDepth();
709   // Unit DIEs always have a depth of zero and never have siblings.
710   if (Depth == 0)
711     return DWARFDie();
712 
713   // Find the previous DIE whose depth is the same as the Die's depth.
714   for (size_t I = getDIEIndex(Die); I > 0;) {
715     --I;
716     if (DieArray[I].getDepth() == Depth - 1)
717       return DWARFDie();
718     if (DieArray[I].getDepth() == Depth)
719       return DWARFDie(this, &DieArray[I]);
720   }
721   return DWARFDie();
722 }
723 
724 DWARFDie DWARFUnit::getFirstChild(const DWARFDebugInfoEntry *Die) {
725   if (!Die->hasChildren())
726     return DWARFDie();
727 
728   // We do not want access out of bounds when parsing corrupted debug data.
729   size_t I = getDIEIndex(Die) + 1;
730   if (I >= DieArray.size())
731     return DWARFDie();
732   return DWARFDie(this, &DieArray[I]);
733 }
734 
735 DWARFDie DWARFUnit::getLastChild(const DWARFDebugInfoEntry *Die) {
736   if (!Die->hasChildren())
737     return DWARFDie();
738 
739   uint32_t Depth = Die->getDepth();
740   for (size_t I = getDIEIndex(Die) + 1, EndIdx = DieArray.size(); I < EndIdx;
741        ++I) {
742     if (DieArray[I].getDepth() == Depth + 1 &&
743         DieArray[I].getTag() == dwarf::DW_TAG_null)
744       return DWARFDie(this, &DieArray[I]);
745     assert(DieArray[I].getDepth() > Depth && "Not processing children?");
746   }
747   return DWARFDie();
748 }
749 
750 const DWARFAbbreviationDeclarationSet *DWARFUnit::getAbbreviations() const {
751   if (!Abbrevs)
752     Abbrevs = Abbrev->getAbbreviationDeclarationSet(Header.getAbbrOffset());
753   return Abbrevs;
754 }
755 
756 llvm::Optional<object::SectionedAddress> DWARFUnit::getBaseAddress() {
757   if (BaseAddr)
758     return BaseAddr;
759 
760   DWARFDie UnitDie = getUnitDIE();
761   Optional<DWARFFormValue> PC = UnitDie.find({DW_AT_low_pc, DW_AT_entry_pc});
762   BaseAddr = toSectionedAddress(PC);
763   return BaseAddr;
764 }
765 
766 Expected<StrOffsetsContributionDescriptor>
767 StrOffsetsContributionDescriptor::validateContributionSize(
768     DWARFDataExtractor &DA) {
769   uint8_t EntrySize = getDwarfOffsetByteSize();
770   // In order to ensure that we don't read a partial record at the end of
771   // the section we validate for a multiple of the entry size.
772   uint64_t ValidationSize = alignTo(Size, EntrySize);
773   // Guard against overflow.
774   if (ValidationSize >= Size)
775     if (DA.isValidOffsetForDataOfSize((uint32_t)Base, ValidationSize))
776       return *this;
777   return createStringError(errc::invalid_argument, "length exceeds section size");
778 }
779 
780 // Look for a DWARF64-formatted contribution to the string offsets table
781 // starting at a given offset and record it in a descriptor.
782 static Expected<StrOffsetsContributionDescriptor>
783 parseDWARF64StringOffsetsTableHeader(DWARFDataExtractor &DA, uint32_t Offset) {
784   if (!DA.isValidOffsetForDataOfSize(Offset, 16))
785     return createStringError(errc::invalid_argument, "section offset exceeds section size");
786 
787   if (DA.getU32(&Offset) != 0xffffffff)
788     return createStringError(errc::invalid_argument, "32 bit contribution referenced from a 64 bit unit");
789 
790   uint64_t Size = DA.getU64(&Offset);
791   uint8_t Version = DA.getU16(&Offset);
792   (void)DA.getU16(&Offset); // padding
793   // The encoded length includes the 2-byte version field and the 2-byte
794   // padding, so we need to subtract them out when we populate the descriptor.
795   return StrOffsetsContributionDescriptor(Offset, Size - 4, Version, DWARF64);
796 }
797 
798 // Look for a DWARF32-formatted contribution to the string offsets table
799 // starting at a given offset and record it in a descriptor.
800 static Expected<StrOffsetsContributionDescriptor>
801 parseDWARF32StringOffsetsTableHeader(DWARFDataExtractor &DA, uint32_t Offset) {
802   if (!DA.isValidOffsetForDataOfSize(Offset, 8))
803     return createStringError(errc::invalid_argument, "section offset exceeds section size");
804 
805   uint32_t ContributionSize = DA.getU32(&Offset);
806   if (ContributionSize >= 0xfffffff0)
807     return createStringError(errc::invalid_argument, "invalid length");
808 
809   uint8_t Version = DA.getU16(&Offset);
810   (void)DA.getU16(&Offset); // padding
811   // The encoded length includes the 2-byte version field and the 2-byte
812   // padding, so we need to subtract them out when we populate the descriptor.
813   return StrOffsetsContributionDescriptor(Offset, ContributionSize - 4, Version,
814                                           DWARF32);
815 }
816 
817 static Expected<StrOffsetsContributionDescriptor>
818 parseDWARFStringOffsetsTableHeader(DWARFDataExtractor &DA,
819                                    llvm::dwarf::DwarfFormat Format,
820                                    uint64_t Offset) {
821   StrOffsetsContributionDescriptor Desc;
822   switch (Format) {
823   case dwarf::DwarfFormat::DWARF64: {
824     if (Offset < 16)
825       return createStringError(errc::invalid_argument, "insufficient space for 64 bit header prefix");
826     auto DescOrError = parseDWARF64StringOffsetsTableHeader(DA, (uint32_t)Offset - 16);
827     if (!DescOrError)
828       return DescOrError.takeError();
829     Desc = *DescOrError;
830     break;
831   }
832   case dwarf::DwarfFormat::DWARF32: {
833     if (Offset < 8)
834       return createStringError(errc::invalid_argument, "insufficient space for 32 bit header prefix");
835     auto DescOrError = parseDWARF32StringOffsetsTableHeader(DA, (uint32_t)Offset - 8);
836     if (!DescOrError)
837       return DescOrError.takeError();
838     Desc = *DescOrError;
839     break;
840   }
841   }
842   return Desc.validateContributionSize(DA);
843 }
844 
845 Expected<Optional<StrOffsetsContributionDescriptor>>
846 DWARFUnit::determineStringOffsetsTableContribution(DWARFDataExtractor &DA) {
847   uint64_t Offset;
848   if (IsDWO) {
849     Offset = 0;
850     if (DA.getData().data() == nullptr)
851       return None;
852   } else {
853     auto OptOffset = toSectionOffset(getUnitDIE().find(DW_AT_str_offsets_base));
854     if (!OptOffset)
855       return None;
856     Offset = *OptOffset;
857   }
858   auto DescOrError = parseDWARFStringOffsetsTableHeader(DA, Header.getFormat(), Offset);
859   if (!DescOrError)
860     return DescOrError.takeError();
861   return *DescOrError;
862 }
863 
864 Expected<Optional<StrOffsetsContributionDescriptor>>
865 DWARFUnit::determineStringOffsetsTableContributionDWO(DWARFDataExtractor & DA) {
866   uint64_t Offset = 0;
867   auto IndexEntry = Header.getIndexEntry();
868   const auto *C =
869       IndexEntry ? IndexEntry->getOffset(DW_SECT_STR_OFFSETS) : nullptr;
870   if (C)
871     Offset = C->Offset;
872   if (getVersion() >= 5) {
873     if (DA.getData().data() == nullptr)
874       return None;
875     Offset += Header.getFormat() == dwarf::DwarfFormat::DWARF32 ? 8 : 16;
876     // Look for a valid contribution at the given offset.
877     auto DescOrError = parseDWARFStringOffsetsTableHeader(DA, Header.getFormat(), Offset);
878     if (!DescOrError)
879       return DescOrError.takeError();
880     return *DescOrError;
881   }
882   // Prior to DWARF v5, we derive the contribution size from the
883   // index table (in a package file). In a .dwo file it is simply
884   // the length of the string offsets section.
885   if (!IndexEntry)
886     return {
887         Optional<StrOffsetsContributionDescriptor>(
888             {0, StringOffsetSection.Data.size(), 4, DWARF32})};
889   if (C)
890     return {Optional<StrOffsetsContributionDescriptor>(
891         {C->Offset, C->Length, 4, DWARF32})};
892   return None;
893 }
894