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
addUnitsForSection(DWARFContext & C,const DWARFSection & Section,DWARFSectionKind SectionKind)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.getRangesSection(),
41 &D.getLocSection(), D.getStrSection(),
42 D.getStrOffsetsSection(), &D.getAddrSection(),
43 D.getLineSection(), D.isLittleEndian(), false, false,
44 SectionKind);
45 }
46
addUnitsForDWOSection(DWARFContext & C,const DWARFSection & DWOSection,DWARFSectionKind SectionKind,bool Lazy)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.getRangesDWOSection(),
53 &D.getLocDWOSection(), D.getStrDWOSection(),
54 D.getStrOffsetsDWOSection(), &D.getAddrSection(),
55 D.getLineDWOSection(), C.isLittleEndian(), true, Lazy,
56 SectionKind);
57 }
58
addUnitsImpl(DWARFContext & Context,const DWARFObject & Obj,const DWARFSection & Section,const DWARFDebugAbbrev * DA,const DWARFSection * RS,const DWARFSection * LocSection,StringRef SS,const DWARFSection & SOS,const DWARFSection * AOS,const DWARFSection & LS,bool LE,bool IsDWO,bool Lazy,DWARFSectionKind SectionKind)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](uint64_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 = std::make_unique<DWARFTypeUnit>(Context, InfoSection, Header, DA,
87 RS, LocSection, SS, SOS, AOS, LS,
88 LE, IsDWO, *this);
89 else
90 U = std::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 uint64_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
addUnit(std::unique_ptr<DWARFUnit> Unit)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
getUnitForOffset(uint64_t Offset) const129 DWARFUnit *DWARFUnitVector::getUnitForOffset(uint64_t Offset) const {
130 auto end = begin() + getNumInfoUnits();
131 auto *CU =
132 std::upper_bound(begin(), end, Offset,
133 [](uint64_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 *
getUnitForIndexEntry(const DWARFUnitIndex::Entry & E)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 [](uint64_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
DWARFUnit(DWARFContext & DC,const DWARFSection & Section,const DWARFUnitHeader & Header,const DWARFDebugAbbrev * DA,const DWARFSection * RS,const DWARFSection * LocSection,StringRef SS,const DWARFSection & SOS,const DWARFSection * AOS,const DWARFSection & LS,bool LE,bool IsDWO,const DWARFUnitVector & UnitVector)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), LineSection(LS), StringSection(SS),
179 StringOffsetSection(SOS), AddrOffsetSection(AOS), isLittleEndian(LE),
180 IsDWO(IsDWO), UnitVector(UnitVector) {
181 clear();
182 if (IsDWO) {
183 // If we are reading a package file, we need to adjust the location list
184 // data based on the index entries.
185 StringRef Data = LocSection->Data;
186 if (auto *IndexEntry = Header.getIndexEntry())
187 if (const auto *C = IndexEntry->getOffset(DW_SECT_LOC))
188 Data = Data.substr(C->Offset, C->Length);
189
190 DWARFDataExtractor DWARFData =
191 Header.getVersion() >= 5
192 ? DWARFDataExtractor(Context.getDWARFObj(),
193 Context.getDWARFObj().getLoclistsDWOSection(),
194 isLittleEndian, getAddressByteSize())
195 : DWARFDataExtractor(Data, isLittleEndian, getAddressByteSize());
196 LocTable =
197 std::make_unique<DWARFDebugLoclists>(DWARFData, Header.getVersion());
198
199 } else if (Header.getVersion() >= 5) {
200 LocTable = std::make_unique<DWARFDebugLoclists>(
201 DWARFDataExtractor(Context.getDWARFObj(),
202 Context.getDWARFObj().getLoclistsSection(),
203 isLittleEndian, getAddressByteSize()),
204 Header.getVersion());
205 } else {
206 LocTable = std::make_unique<DWARFDebugLoc>(
207 DWARFDataExtractor(Context.getDWARFObj(), *LocSection, isLittleEndian,
208 getAddressByteSize()));
209 }
210 }
211
212 DWARFUnit::~DWARFUnit() = default;
213
getDebugInfoExtractor() const214 DWARFDataExtractor DWARFUnit::getDebugInfoExtractor() const {
215 return DWARFDataExtractor(Context.getDWARFObj(), InfoSection, isLittleEndian,
216 getAddressByteSize());
217 }
218
219 Optional<object::SectionedAddress>
getAddrOffsetSectionItem(uint32_t Index) const220 DWARFUnit::getAddrOffsetSectionItem(uint32_t Index) const {
221 if (IsDWO) {
222 auto R = Context.info_section_units();
223 auto I = R.begin();
224 // Surprising if a DWO file has more than one skeleton unit in it - this
225 // probably shouldn't be valid, but if a use case is found, here's where to
226 // support it (probably have to linearly search for the matching skeleton CU
227 // here)
228 if (I != R.end() && std::next(I) == R.end())
229 return (*I)->getAddrOffsetSectionItem(Index);
230 }
231 if (!AddrOffsetSectionBase)
232 return None;
233 uint64_t Offset = *AddrOffsetSectionBase + Index * getAddressByteSize();
234 if (AddrOffsetSection->Data.size() < Offset + getAddressByteSize())
235 return None;
236 DWARFDataExtractor DA(Context.getDWARFObj(), *AddrOffsetSection,
237 isLittleEndian, getAddressByteSize());
238 uint64_t Section;
239 uint64_t Address = DA.getRelocatedAddress(&Offset, &Section);
240 return {{Address, Section}};
241 }
242
getStringOffsetSectionItem(uint32_t Index) const243 Optional<uint64_t> DWARFUnit::getStringOffsetSectionItem(uint32_t Index) const {
244 if (!StringOffsetsTableContribution)
245 return None;
246 unsigned ItemSize = getDwarfStringOffsetsByteSize();
247 uint64_t Offset = getStringOffsetsBase() + Index * ItemSize;
248 if (StringOffsetSection.Data.size() < Offset + ItemSize)
249 return None;
250 DWARFDataExtractor DA(Context.getDWARFObj(), StringOffsetSection,
251 isLittleEndian, 0);
252 return DA.getRelocatedValue(ItemSize, &Offset);
253 }
254
extract(DWARFContext & Context,const DWARFDataExtractor & debug_info,uint64_t * offset_ptr,DWARFSectionKind SectionKind,const DWARFUnitIndex * Index,const DWARFUnitIndex::Entry * Entry)255 bool DWARFUnitHeader::extract(DWARFContext &Context,
256 const DWARFDataExtractor &debug_info,
257 uint64_t *offset_ptr,
258 DWARFSectionKind SectionKind,
259 const DWARFUnitIndex *Index,
260 const DWARFUnitIndex::Entry *Entry) {
261 Offset = *offset_ptr;
262 Error Err = Error::success();
263 IndexEntry = Entry;
264 if (!IndexEntry && Index)
265 IndexEntry = Index->getFromOffset(*offset_ptr);
266 Length = debug_info.getRelocatedValue(4, offset_ptr, nullptr, &Err);
267 FormParams.Format = DWARF32;
268 if (Length == dwarf::DW_LENGTH_DWARF64) {
269 Length = debug_info.getU64(offset_ptr, &Err);
270 FormParams.Format = DWARF64;
271 }
272 FormParams.Version = debug_info.getU16(offset_ptr, &Err);
273 if (FormParams.Version >= 5) {
274 UnitType = debug_info.getU8(offset_ptr, &Err);
275 FormParams.AddrSize = debug_info.getU8(offset_ptr, &Err);
276 AbbrOffset = debug_info.getRelocatedValue(
277 FormParams.getDwarfOffsetByteSize(), offset_ptr, nullptr, &Err);
278 } else {
279 AbbrOffset = debug_info.getRelocatedValue(
280 FormParams.getDwarfOffsetByteSize(), offset_ptr, nullptr, &Err);
281 FormParams.AddrSize = debug_info.getU8(offset_ptr, &Err);
282 // Fake a unit type based on the section type. This isn't perfect,
283 // but distinguishing compile and type units is generally enough.
284 if (SectionKind == DW_SECT_TYPES)
285 UnitType = DW_UT_type;
286 else
287 UnitType = DW_UT_compile;
288 }
289 if (IndexEntry) {
290 if (AbbrOffset)
291 return false;
292 auto *UnitContrib = IndexEntry->getOffset();
293 if (!UnitContrib || UnitContrib->Length != (Length + 4))
294 return false;
295 auto *AbbrEntry = IndexEntry->getOffset(DW_SECT_ABBREV);
296 if (!AbbrEntry)
297 return false;
298 AbbrOffset = AbbrEntry->Offset;
299 }
300 if (isTypeUnit()) {
301 TypeHash = debug_info.getU64(offset_ptr, &Err);
302 TypeOffset = debug_info.getUnsigned(
303 offset_ptr, FormParams.getDwarfOffsetByteSize(), &Err);
304 } else if (UnitType == DW_UT_split_compile || UnitType == DW_UT_skeleton)
305 DWOId = debug_info.getU64(offset_ptr, &Err);
306
307 if (errorToBool(std::move(Err)))
308 return false;
309
310 // Header fields all parsed, capture the size of this unit header.
311 assert(*offset_ptr - Offset <= 255 && "unexpected header size");
312 Size = uint8_t(*offset_ptr - Offset);
313
314 // Type offset is unit-relative; should be after the header and before
315 // the end of the current unit.
316 bool TypeOffsetOK =
317 !isTypeUnit()
318 ? true
319 : TypeOffset >= Size &&
320 TypeOffset < getLength() + getUnitLengthFieldByteSize();
321 bool LengthOK = debug_info.isValidOffset(getNextUnitOffset() - 1);
322 bool VersionOK = DWARFContext::isSupportedVersion(getVersion());
323 bool AddrSizeOK = getAddressByteSize() == 4 || getAddressByteSize() == 8;
324
325 if (!LengthOK || !VersionOK || !AddrSizeOK || !TypeOffsetOK)
326 return false;
327
328 // Keep track of the highest DWARF version we encounter across all units.
329 Context.setMaxVersionIfGreater(getVersion());
330 return true;
331 }
332
333 // Parse the rangelist table header, including the optional array of offsets
334 // following it (DWARF v5 and later).
335 template<typename ListTableType>
336 static Expected<ListTableType>
parseListTableHeader(DWARFDataExtractor & DA,uint64_t Offset,DwarfFormat Format)337 parseListTableHeader(DWARFDataExtractor &DA, uint64_t Offset,
338 DwarfFormat Format) {
339 // We are expected to be called with Offset 0 or pointing just past the table
340 // header. Correct Offset in the latter case so that it points to the start
341 // of the header.
342 if (Offset > 0) {
343 uint64_t HeaderSize = DWARFListTableHeader::getHeaderSize(Format);
344 if (Offset < HeaderSize)
345 return createStringError(errc::invalid_argument, "did not detect a valid"
346 " list table with base = 0x%" PRIx64 "\n",
347 Offset);
348 Offset -= HeaderSize;
349 }
350 ListTableType Table;
351 if (Error E = Table.extractHeaderAndOffsets(DA, &Offset))
352 return std::move(E);
353 return Table;
354 }
355
extractRangeList(uint64_t RangeListOffset,DWARFDebugRangeList & RangeList) const356 Error DWARFUnit::extractRangeList(uint64_t RangeListOffset,
357 DWARFDebugRangeList &RangeList) const {
358 // Require that compile unit is extracted.
359 assert(!DieArray.empty());
360 DWARFDataExtractor RangesData(Context.getDWARFObj(), *RangeSection,
361 isLittleEndian, getAddressByteSize());
362 uint64_t ActualRangeListOffset = RangeSectionBase + RangeListOffset;
363 return RangeList.extract(RangesData, &ActualRangeListOffset);
364 }
365
clear()366 void DWARFUnit::clear() {
367 Abbrevs = nullptr;
368 BaseAddr.reset();
369 RangeSectionBase = 0;
370 LocSectionBase = 0;
371 AddrOffsetSectionBase = None;
372 clearDIEs(false);
373 DWO.reset();
374 }
375
getCompilationDir()376 const char *DWARFUnit::getCompilationDir() {
377 return dwarf::toString(getUnitDIE().find(DW_AT_comp_dir), nullptr);
378 }
379
extractDIEsToVector(bool AppendCUDie,bool AppendNonCUDies,std::vector<DWARFDebugInfoEntry> & Dies) const380 void DWARFUnit::extractDIEsToVector(
381 bool AppendCUDie, bool AppendNonCUDies,
382 std::vector<DWARFDebugInfoEntry> &Dies) const {
383 if (!AppendCUDie && !AppendNonCUDies)
384 return;
385
386 // Set the offset to that of the first DIE and calculate the start of the
387 // next compilation unit header.
388 uint64_t DIEOffset = getOffset() + getHeaderSize();
389 uint64_t NextCUOffset = getNextUnitOffset();
390 DWARFDebugInfoEntry DIE;
391 DWARFDataExtractor DebugInfoData = getDebugInfoExtractor();
392 uint32_t Depth = 0;
393 bool IsCUDie = true;
394
395 while (DIE.extractFast(*this, &DIEOffset, DebugInfoData, NextCUOffset,
396 Depth)) {
397 if (IsCUDie) {
398 if (AppendCUDie)
399 Dies.push_back(DIE);
400 if (!AppendNonCUDies)
401 break;
402 // The average bytes per DIE entry has been seen to be
403 // around 14-20 so let's pre-reserve the needed memory for
404 // our DIE entries accordingly.
405 Dies.reserve(Dies.size() + getDebugInfoSize() / 14);
406 IsCUDie = false;
407 } else {
408 Dies.push_back(DIE);
409 }
410
411 if (const DWARFAbbreviationDeclaration *AbbrDecl =
412 DIE.getAbbreviationDeclarationPtr()) {
413 // Normal DIE
414 if (AbbrDecl->hasChildren())
415 ++Depth;
416 } else {
417 // NULL DIE.
418 if (Depth > 0)
419 --Depth;
420 if (Depth == 0)
421 break; // We are done with this compile unit!
422 }
423 }
424
425 // Give a little bit of info if we encounter corrupt DWARF (our offset
426 // should always terminate at or before the start of the next compilation
427 // unit header).
428 if (DIEOffset > NextCUOffset)
429 WithColor::warning() << format("DWARF compile unit extends beyond its "
430 "bounds cu 0x%8.8" PRIx64 " "
431 "at 0x%8.8" PRIx64 "\n",
432 getOffset(), DIEOffset);
433 }
434
extractDIEsIfNeeded(bool CUDieOnly)435 void DWARFUnit::extractDIEsIfNeeded(bool CUDieOnly) {
436 if (Error e = tryExtractDIEsIfNeeded(CUDieOnly))
437 WithColor::error() << toString(std::move(e));
438 }
439
tryExtractDIEsIfNeeded(bool CUDieOnly)440 Error DWARFUnit::tryExtractDIEsIfNeeded(bool CUDieOnly) {
441 if ((CUDieOnly && !DieArray.empty()) ||
442 DieArray.size() > 1)
443 return Error::success(); // Already parsed.
444
445 bool HasCUDie = !DieArray.empty();
446 extractDIEsToVector(!HasCUDie, !CUDieOnly, DieArray);
447
448 if (DieArray.empty())
449 return Error::success();
450
451 // If CU DIE was just parsed, copy several attribute values from it.
452 if (HasCUDie)
453 return Error::success();
454
455 DWARFDie UnitDie(this, &DieArray[0]);
456 if (Optional<uint64_t> DWOId = toUnsigned(UnitDie.find(DW_AT_GNU_dwo_id)))
457 Header.setDWOId(*DWOId);
458 if (!IsDWO) {
459 assert(AddrOffsetSectionBase == None);
460 assert(RangeSectionBase == 0);
461 assert(LocSectionBase == 0);
462 AddrOffsetSectionBase = toSectionOffset(UnitDie.find(DW_AT_addr_base));
463 if (!AddrOffsetSectionBase)
464 AddrOffsetSectionBase =
465 toSectionOffset(UnitDie.find(DW_AT_GNU_addr_base));
466 RangeSectionBase = toSectionOffset(UnitDie.find(DW_AT_rnglists_base), 0);
467 LocSectionBase = toSectionOffset(UnitDie.find(DW_AT_loclists_base), 0);
468 }
469
470 // In general, in DWARF v5 and beyond we derive the start of the unit's
471 // contribution to the string offsets table from the unit DIE's
472 // DW_AT_str_offsets_base attribute. Split DWARF units do not use this
473 // attribute, so we assume that there is a contribution to the string
474 // offsets table starting at offset 0 of the debug_str_offsets.dwo section.
475 // In both cases we need to determine the format of the contribution,
476 // which may differ from the unit's format.
477 DWARFDataExtractor DA(Context.getDWARFObj(), StringOffsetSection,
478 isLittleEndian, 0);
479 if (IsDWO || getVersion() >= 5) {
480 auto StringOffsetOrError =
481 IsDWO ? determineStringOffsetsTableContributionDWO(DA)
482 : determineStringOffsetsTableContribution(DA);
483 if (!StringOffsetOrError)
484 return createStringError(errc::invalid_argument,
485 "invalid reference to or invalid content in "
486 ".debug_str_offsets[.dwo]: " +
487 toString(StringOffsetOrError.takeError()));
488
489 StringOffsetsTableContribution = *StringOffsetOrError;
490 }
491
492 // DWARF v5 uses the .debug_rnglists and .debug_rnglists.dwo sections to
493 // describe address ranges.
494 if (getVersion() >= 5) {
495 if (IsDWO)
496 setRangesSection(&Context.getDWARFObj().getRnglistsDWOSection(), 0);
497 else
498 setRangesSection(&Context.getDWARFObj().getRnglistsSection(),
499 toSectionOffset(UnitDie.find(DW_AT_rnglists_base), 0));
500 if (RangeSection->Data.size()) {
501 // Parse the range list table header. Individual range lists are
502 // extracted lazily.
503 DWARFDataExtractor RangesDA(Context.getDWARFObj(), *RangeSection,
504 isLittleEndian, 0);
505 auto TableOrError = parseListTableHeader<DWARFDebugRnglistTable>(
506 RangesDA, RangeSectionBase, Header.getFormat());
507 if (!TableOrError)
508 return createStringError(errc::invalid_argument,
509 "parsing a range list table: " +
510 toString(TableOrError.takeError()));
511
512 RngListTable = TableOrError.get();
513
514 // In a split dwarf unit, there is no DW_AT_rnglists_base attribute.
515 // Adjust RangeSectionBase to point past the table header.
516 if (IsDWO && RngListTable)
517 RangeSectionBase = RngListTable->getHeaderSize();
518 }
519
520 // In a split dwarf unit, there is no DW_AT_loclists_base attribute.
521 // Setting LocSectionBase to point past the table header.
522 if (IsDWO)
523 setLocSection(&Context.getDWARFObj().getLoclistsDWOSection(),
524 DWARFListTableHeader::getHeaderSize(Header.getFormat()));
525 else
526 setLocSection(&Context.getDWARFObj().getLoclistsSection(),
527 toSectionOffset(UnitDie.find(DW_AT_loclists_base), 0));
528
529 if (LocSection->Data.size()) {
530 if (IsDWO)
531 LoclistTableHeader.emplace(".debug_loclists.dwo", "locations");
532 else
533 LoclistTableHeader.emplace(".debug_loclists", "locations");
534
535 uint64_t HeaderSize = DWARFListTableHeader::getHeaderSize(Header.getFormat());
536 uint64_t Offset = getLocSectionBase();
537 DWARFDataExtractor Data(Context.getDWARFObj(), *LocSection,
538 isLittleEndian, getAddressByteSize());
539 if (Offset < HeaderSize)
540 return createStringError(errc::invalid_argument,
541 "did not detect a valid"
542 " list table with base = 0x%" PRIx64 "\n",
543 Offset);
544 Offset -= HeaderSize;
545 if (Error E = LoclistTableHeader->extract(Data, &Offset))
546 return createStringError(errc::invalid_argument,
547 "parsing a loclist table: " +
548 toString(std::move(E)));
549 }
550 }
551
552 // Don't fall back to DW_AT_GNU_ranges_base: it should be ignored for
553 // skeleton CU DIE, so that DWARF users not aware of it are not broken.
554 return Error::success();
555 }
556
parseDWO()557 bool DWARFUnit::parseDWO() {
558 if (IsDWO)
559 return false;
560 if (DWO.get())
561 return false;
562 DWARFDie UnitDie = getUnitDIE();
563 if (!UnitDie)
564 return false;
565 auto DWOFileName = getVersion() >= 5
566 ? dwarf::toString(UnitDie.find(DW_AT_dwo_name))
567 : dwarf::toString(UnitDie.find(DW_AT_GNU_dwo_name));
568 if (!DWOFileName)
569 return false;
570 auto CompilationDir = dwarf::toString(UnitDie.find(DW_AT_comp_dir));
571 SmallString<16> AbsolutePath;
572 if (sys::path::is_relative(*DWOFileName) && CompilationDir &&
573 *CompilationDir) {
574 sys::path::append(AbsolutePath, *CompilationDir);
575 }
576 sys::path::append(AbsolutePath, *DWOFileName);
577 auto DWOId = getDWOId();
578 if (!DWOId)
579 return false;
580 auto DWOContext = Context.getDWOContext(AbsolutePath);
581 if (!DWOContext)
582 return false;
583
584 DWARFCompileUnit *DWOCU = DWOContext->getDWOCompileUnitForHash(*DWOId);
585 if (!DWOCU)
586 return false;
587 DWO = std::shared_ptr<DWARFCompileUnit>(std::move(DWOContext), DWOCU);
588 // Share .debug_addr and .debug_ranges section with compile unit in .dwo
589 if (AddrOffsetSectionBase)
590 DWO->setAddrOffsetSection(AddrOffsetSection, *AddrOffsetSectionBase);
591 if (getVersion() >= 5) {
592 DWO->setRangesSection(&Context.getDWARFObj().getRnglistsDWOSection(), 0);
593 DWARFDataExtractor RangesDA(Context.getDWARFObj(), *RangeSection,
594 isLittleEndian, 0);
595 if (auto TableOrError = parseListTableHeader<DWARFDebugRnglistTable>(
596 RangesDA, RangeSectionBase, Header.getFormat()))
597 DWO->RngListTable = TableOrError.get();
598 else
599 WithColor::error() << "parsing a range list table: "
600 << toString(TableOrError.takeError())
601 << '\n';
602 if (DWO->RngListTable)
603 DWO->RangeSectionBase = DWO->RngListTable->getHeaderSize();
604 } else {
605 auto DWORangesBase = UnitDie.getRangesBaseAttribute();
606 DWO->setRangesSection(RangeSection, DWORangesBase ? *DWORangesBase : 0);
607 }
608
609 return true;
610 }
611
clearDIEs(bool KeepCUDie)612 void DWARFUnit::clearDIEs(bool KeepCUDie) {
613 if (DieArray.size() > (unsigned)KeepCUDie) {
614 DieArray.resize((unsigned)KeepCUDie);
615 DieArray.shrink_to_fit();
616 }
617 }
618
619 Expected<DWARFAddressRangesVector>
findRnglistFromOffset(uint64_t Offset)620 DWARFUnit::findRnglistFromOffset(uint64_t Offset) {
621 if (getVersion() <= 4) {
622 DWARFDebugRangeList RangeList;
623 if (Error E = extractRangeList(Offset, RangeList))
624 return std::move(E);
625 return RangeList.getAbsoluteRanges(getBaseAddress());
626 }
627 if (RngListTable) {
628 DWARFDataExtractor RangesData(Context.getDWARFObj(), *RangeSection,
629 isLittleEndian, RngListTable->getAddrSize());
630 auto RangeListOrError = RngListTable->findList(RangesData, Offset);
631 if (RangeListOrError)
632 return RangeListOrError.get().getAbsoluteRanges(getBaseAddress(), *this);
633 return RangeListOrError.takeError();
634 }
635
636 return createStringError(errc::invalid_argument,
637 "missing or invalid range list table");
638 }
639
640 Expected<DWARFAddressRangesVector>
findRnglistFromIndex(uint32_t Index)641 DWARFUnit::findRnglistFromIndex(uint32_t Index) {
642 if (auto Offset = getRnglistOffset(Index))
643 return findRnglistFromOffset(*Offset);
644
645 if (RngListTable)
646 return createStringError(errc::invalid_argument,
647 "invalid range list table index %d", Index);
648
649 return createStringError(errc::invalid_argument,
650 "missing or invalid range list table");
651 }
652
collectAddressRanges()653 Expected<DWARFAddressRangesVector> DWARFUnit::collectAddressRanges() {
654 DWARFDie UnitDie = getUnitDIE();
655 if (!UnitDie)
656 return createStringError(errc::invalid_argument, "No unit DIE");
657
658 // First, check if unit DIE describes address ranges for the whole unit.
659 auto CUDIERangesOrError = UnitDie.getAddressRanges();
660 if (!CUDIERangesOrError)
661 return createStringError(errc::invalid_argument,
662 "decoding address ranges: %s",
663 toString(CUDIERangesOrError.takeError()).c_str());
664 return *CUDIERangesOrError;
665 }
666
667 Expected<DWARFLocationExpressionsVector>
findLoclistFromOffset(uint64_t Offset)668 DWARFUnit::findLoclistFromOffset(uint64_t Offset) {
669 DWARFLocationExpressionsVector Result;
670
671 Error InterpretationError = Error::success();
672
673 Error ParseError = getLocationTable().visitAbsoluteLocationList(
674 Offset, getBaseAddress(),
675 [this](uint32_t Index) { return getAddrOffsetSectionItem(Index); },
676 [&](Expected<DWARFLocationExpression> L) {
677 if (L)
678 Result.push_back(std::move(*L));
679 else
680 InterpretationError =
681 joinErrors(L.takeError(), std::move(InterpretationError));
682 return !InterpretationError;
683 });
684
685 if (ParseError || InterpretationError)
686 return joinErrors(std::move(ParseError), std::move(InterpretationError));
687
688 return Result;
689 }
690
updateAddressDieMap(DWARFDie Die)691 void DWARFUnit::updateAddressDieMap(DWARFDie Die) {
692 if (Die.isSubroutineDIE()) {
693 auto DIERangesOrError = Die.getAddressRanges();
694 if (DIERangesOrError) {
695 for (const auto &R : DIERangesOrError.get()) {
696 // Ignore 0-sized ranges.
697 if (R.LowPC == R.HighPC)
698 continue;
699 auto B = AddrDieMap.upper_bound(R.LowPC);
700 if (B != AddrDieMap.begin() && R.LowPC < (--B)->second.first) {
701 // The range is a sub-range of existing ranges, we need to split the
702 // existing range.
703 if (R.HighPC < B->second.first)
704 AddrDieMap[R.HighPC] = B->second;
705 if (R.LowPC > B->first)
706 AddrDieMap[B->first].first = R.LowPC;
707 }
708 AddrDieMap[R.LowPC] = std::make_pair(R.HighPC, Die);
709 }
710 } else
711 llvm::consumeError(DIERangesOrError.takeError());
712 }
713 // Parent DIEs are added to the AddrDieMap prior to the Children DIEs to
714 // simplify the logic to update AddrDieMap. The child's range will always
715 // be equal or smaller than the parent's range. With this assumption, when
716 // adding one range into the map, it will at most split a range into 3
717 // sub-ranges.
718 for (DWARFDie Child = Die.getFirstChild(); Child; Child = Child.getSibling())
719 updateAddressDieMap(Child);
720 }
721
getSubroutineForAddress(uint64_t Address)722 DWARFDie DWARFUnit::getSubroutineForAddress(uint64_t Address) {
723 extractDIEsIfNeeded(false);
724 if (AddrDieMap.empty())
725 updateAddressDieMap(getUnitDIE());
726 auto R = AddrDieMap.upper_bound(Address);
727 if (R == AddrDieMap.begin())
728 return DWARFDie();
729 // upper_bound's previous item contains Address.
730 --R;
731 if (Address >= R->second.first)
732 return DWARFDie();
733 return R->second.second;
734 }
735
736 void
getInlinedChainForAddress(uint64_t Address,SmallVectorImpl<DWARFDie> & InlinedChain)737 DWARFUnit::getInlinedChainForAddress(uint64_t Address,
738 SmallVectorImpl<DWARFDie> &InlinedChain) {
739 assert(InlinedChain.empty());
740 // Try to look for subprogram DIEs in the DWO file.
741 parseDWO();
742 // First, find the subroutine that contains the given address (the leaf
743 // of inlined chain).
744 DWARFDie SubroutineDIE =
745 (DWO ? *DWO : *this).getSubroutineForAddress(Address);
746
747 if (!SubroutineDIE)
748 return;
749
750 while (!SubroutineDIE.isSubprogramDIE()) {
751 if (SubroutineDIE.getTag() == DW_TAG_inlined_subroutine)
752 InlinedChain.push_back(SubroutineDIE);
753 SubroutineDIE = SubroutineDIE.getParent();
754 }
755 InlinedChain.push_back(SubroutineDIE);
756 }
757
getDWARFUnitIndex(DWARFContext & Context,DWARFSectionKind Kind)758 const DWARFUnitIndex &llvm::getDWARFUnitIndex(DWARFContext &Context,
759 DWARFSectionKind Kind) {
760 if (Kind == DW_SECT_INFO)
761 return Context.getCUIndex();
762 assert(Kind == DW_SECT_TYPES);
763 return Context.getTUIndex();
764 }
765
getParent(const DWARFDebugInfoEntry * Die)766 DWARFDie DWARFUnit::getParent(const DWARFDebugInfoEntry *Die) {
767 if (!Die)
768 return DWARFDie();
769 const uint32_t Depth = Die->getDepth();
770 // Unit DIEs always have a depth of zero and never have parents.
771 if (Depth == 0)
772 return DWARFDie();
773 // Depth of 1 always means parent is the compile/type unit.
774 if (Depth == 1)
775 return getUnitDIE();
776 // Look for previous DIE with a depth that is one less than the Die's depth.
777 const uint32_t ParentDepth = Depth - 1;
778 for (uint32_t I = getDIEIndex(Die) - 1; I > 0; --I) {
779 if (DieArray[I].getDepth() == ParentDepth)
780 return DWARFDie(this, &DieArray[I]);
781 }
782 return DWARFDie();
783 }
784
getSibling(const DWARFDebugInfoEntry * Die)785 DWARFDie DWARFUnit::getSibling(const DWARFDebugInfoEntry *Die) {
786 if (!Die)
787 return DWARFDie();
788 uint32_t Depth = Die->getDepth();
789 // Unit DIEs always have a depth of zero and never have siblings.
790 if (Depth == 0)
791 return DWARFDie();
792 // NULL DIEs don't have siblings.
793 if (Die->getAbbreviationDeclarationPtr() == nullptr)
794 return DWARFDie();
795
796 // Find the next DIE whose depth is the same as the Die's depth.
797 for (size_t I = getDIEIndex(Die) + 1, EndIdx = DieArray.size(); I < EndIdx;
798 ++I) {
799 if (DieArray[I].getDepth() == Depth)
800 return DWARFDie(this, &DieArray[I]);
801 }
802 return DWARFDie();
803 }
804
getPreviousSibling(const DWARFDebugInfoEntry * Die)805 DWARFDie DWARFUnit::getPreviousSibling(const DWARFDebugInfoEntry *Die) {
806 if (!Die)
807 return DWARFDie();
808 uint32_t Depth = Die->getDepth();
809 // Unit DIEs always have a depth of zero and never have siblings.
810 if (Depth == 0)
811 return DWARFDie();
812
813 // Find the previous DIE whose depth is the same as the Die's depth.
814 for (size_t I = getDIEIndex(Die); I > 0;) {
815 --I;
816 if (DieArray[I].getDepth() == Depth - 1)
817 return DWARFDie();
818 if (DieArray[I].getDepth() == Depth)
819 return DWARFDie(this, &DieArray[I]);
820 }
821 return DWARFDie();
822 }
823
getFirstChild(const DWARFDebugInfoEntry * Die)824 DWARFDie DWARFUnit::getFirstChild(const DWARFDebugInfoEntry *Die) {
825 if (!Die->hasChildren())
826 return DWARFDie();
827
828 // We do not want access out of bounds when parsing corrupted debug data.
829 size_t I = getDIEIndex(Die) + 1;
830 if (I >= DieArray.size())
831 return DWARFDie();
832 return DWARFDie(this, &DieArray[I]);
833 }
834
getLastChild(const DWARFDebugInfoEntry * Die)835 DWARFDie DWARFUnit::getLastChild(const DWARFDebugInfoEntry *Die) {
836 if (!Die->hasChildren())
837 return DWARFDie();
838
839 uint32_t Depth = Die->getDepth();
840 for (size_t I = getDIEIndex(Die) + 1, EndIdx = DieArray.size(); I < EndIdx;
841 ++I) {
842 if (DieArray[I].getDepth() == Depth + 1 &&
843 DieArray[I].getTag() == dwarf::DW_TAG_null)
844 return DWARFDie(this, &DieArray[I]);
845 assert(DieArray[I].getDepth() > Depth && "Not processing children?");
846 }
847 return DWARFDie();
848 }
849
getAbbreviations() const850 const DWARFAbbreviationDeclarationSet *DWARFUnit::getAbbreviations() const {
851 if (!Abbrevs)
852 Abbrevs = Abbrev->getAbbreviationDeclarationSet(Header.getAbbrOffset());
853 return Abbrevs;
854 }
855
getBaseAddress()856 llvm::Optional<object::SectionedAddress> DWARFUnit::getBaseAddress() {
857 if (BaseAddr)
858 return BaseAddr;
859
860 DWARFDie UnitDie = getUnitDIE();
861 Optional<DWARFFormValue> PC = UnitDie.find({DW_AT_low_pc, DW_AT_entry_pc});
862 BaseAddr = toSectionedAddress(PC);
863 return BaseAddr;
864 }
865
866 Expected<StrOffsetsContributionDescriptor>
validateContributionSize(DWARFDataExtractor & DA)867 StrOffsetsContributionDescriptor::validateContributionSize(
868 DWARFDataExtractor &DA) {
869 uint8_t EntrySize = getDwarfOffsetByteSize();
870 // In order to ensure that we don't read a partial record at the end of
871 // the section we validate for a multiple of the entry size.
872 uint64_t ValidationSize = alignTo(Size, EntrySize);
873 // Guard against overflow.
874 if (ValidationSize >= Size)
875 if (DA.isValidOffsetForDataOfSize((uint32_t)Base, ValidationSize))
876 return *this;
877 return createStringError(errc::invalid_argument, "length exceeds section size");
878 }
879
880 // Look for a DWARF64-formatted contribution to the string offsets table
881 // starting at a given offset and record it in a descriptor.
882 static Expected<StrOffsetsContributionDescriptor>
parseDWARF64StringOffsetsTableHeader(DWARFDataExtractor & DA,uint64_t Offset)883 parseDWARF64StringOffsetsTableHeader(DWARFDataExtractor &DA, uint64_t Offset) {
884 if (!DA.isValidOffsetForDataOfSize(Offset, 16))
885 return createStringError(errc::invalid_argument, "section offset exceeds section size");
886
887 if (DA.getU32(&Offset) != dwarf::DW_LENGTH_DWARF64)
888 return createStringError(errc::invalid_argument, "32 bit contribution referenced from a 64 bit unit");
889
890 uint64_t Size = DA.getU64(&Offset);
891 uint8_t Version = DA.getU16(&Offset);
892 (void)DA.getU16(&Offset); // padding
893 // The encoded length includes the 2-byte version field and the 2-byte
894 // padding, so we need to subtract them out when we populate the descriptor.
895 return StrOffsetsContributionDescriptor(Offset, Size - 4, Version, DWARF64);
896 }
897
898 // Look for a DWARF32-formatted contribution to the string offsets table
899 // starting at a given offset and record it in a descriptor.
900 static Expected<StrOffsetsContributionDescriptor>
parseDWARF32StringOffsetsTableHeader(DWARFDataExtractor & DA,uint64_t Offset)901 parseDWARF32StringOffsetsTableHeader(DWARFDataExtractor &DA, uint64_t Offset) {
902 if (!DA.isValidOffsetForDataOfSize(Offset, 8))
903 return createStringError(errc::invalid_argument, "section offset exceeds section size");
904
905 uint32_t ContributionSize = DA.getU32(&Offset);
906 if (ContributionSize >= dwarf::DW_LENGTH_lo_reserved)
907 return createStringError(errc::invalid_argument, "invalid length");
908
909 uint8_t Version = DA.getU16(&Offset);
910 (void)DA.getU16(&Offset); // padding
911 // The encoded length includes the 2-byte version field and the 2-byte
912 // padding, so we need to subtract them out when we populate the descriptor.
913 return StrOffsetsContributionDescriptor(Offset, ContributionSize - 4, Version,
914 DWARF32);
915 }
916
917 static Expected<StrOffsetsContributionDescriptor>
parseDWARFStringOffsetsTableHeader(DWARFDataExtractor & DA,llvm::dwarf::DwarfFormat Format,uint64_t Offset)918 parseDWARFStringOffsetsTableHeader(DWARFDataExtractor &DA,
919 llvm::dwarf::DwarfFormat Format,
920 uint64_t Offset) {
921 StrOffsetsContributionDescriptor Desc;
922 switch (Format) {
923 case dwarf::DwarfFormat::DWARF64: {
924 if (Offset < 16)
925 return createStringError(errc::invalid_argument, "insufficient space for 64 bit header prefix");
926 auto DescOrError = parseDWARF64StringOffsetsTableHeader(DA, Offset - 16);
927 if (!DescOrError)
928 return DescOrError.takeError();
929 Desc = *DescOrError;
930 break;
931 }
932 case dwarf::DwarfFormat::DWARF32: {
933 if (Offset < 8)
934 return createStringError(errc::invalid_argument, "insufficient space for 32 bit header prefix");
935 auto DescOrError = parseDWARF32StringOffsetsTableHeader(DA, Offset - 8);
936 if (!DescOrError)
937 return DescOrError.takeError();
938 Desc = *DescOrError;
939 break;
940 }
941 }
942 return Desc.validateContributionSize(DA);
943 }
944
945 Expected<Optional<StrOffsetsContributionDescriptor>>
determineStringOffsetsTableContribution(DWARFDataExtractor & DA)946 DWARFUnit::determineStringOffsetsTableContribution(DWARFDataExtractor &DA) {
947 uint64_t Offset;
948 if (IsDWO) {
949 Offset = 0;
950 if (DA.getData().data() == nullptr)
951 return None;
952 } else {
953 auto OptOffset = toSectionOffset(getUnitDIE().find(DW_AT_str_offsets_base));
954 if (!OptOffset)
955 return None;
956 Offset = *OptOffset;
957 }
958 auto DescOrError = parseDWARFStringOffsetsTableHeader(DA, Header.getFormat(), Offset);
959 if (!DescOrError)
960 return DescOrError.takeError();
961 return *DescOrError;
962 }
963
964 Expected<Optional<StrOffsetsContributionDescriptor>>
determineStringOffsetsTableContributionDWO(DWARFDataExtractor & DA)965 DWARFUnit::determineStringOffsetsTableContributionDWO(DWARFDataExtractor & DA) {
966 uint64_t Offset = 0;
967 auto IndexEntry = Header.getIndexEntry();
968 const auto *C =
969 IndexEntry ? IndexEntry->getOffset(DW_SECT_STR_OFFSETS) : nullptr;
970 if (C)
971 Offset = C->Offset;
972 if (getVersion() >= 5) {
973 if (DA.getData().data() == nullptr)
974 return None;
975 Offset += Header.getFormat() == dwarf::DwarfFormat::DWARF32 ? 8 : 16;
976 // Look for a valid contribution at the given offset.
977 auto DescOrError = parseDWARFStringOffsetsTableHeader(DA, Header.getFormat(), Offset);
978 if (!DescOrError)
979 return DescOrError.takeError();
980 return *DescOrError;
981 }
982 // Prior to DWARF v5, we derive the contribution size from the
983 // index table (in a package file). In a .dwo file it is simply
984 // the length of the string offsets section.
985 if (!IndexEntry)
986 return {
987 Optional<StrOffsetsContributionDescriptor>(
988 {0, StringOffsetSection.Data.size(), 4, DWARF32})};
989 if (C)
990 return {Optional<StrOffsetsContributionDescriptor>(
991 {C->Offset, C->Length, 4, DWARF32})};
992 return None;
993 }
994