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 <algorithm>
25 #include <cassert>
26 #include <cstddef>
27 #include <cstdint>
28 #include <cstdio>
29 #include <utility>
30 #include <vector>
31
32 using namespace llvm;
33 using namespace dwarf;
34
addUnitsForSection(DWARFContext & C,const DWARFSection & Section,DWARFSectionKind SectionKind)35 void DWARFUnitVector::addUnitsForSection(DWARFContext &C,
36 const DWARFSection &Section,
37 DWARFSectionKind SectionKind) {
38 const DWARFObject &D = C.getDWARFObj();
39 addUnitsImpl(C, D, Section, C.getDebugAbbrev(), &D.getRangesSection(),
40 &D.getLocSection(), D.getStrSection(),
41 D.getStrOffsetsSection(), &D.getAddrSection(),
42 D.getLineSection(), D.isLittleEndian(), false, false,
43 SectionKind);
44 }
45
addUnitsForDWOSection(DWARFContext & C,const DWARFSection & DWOSection,DWARFSectionKind SectionKind,bool Lazy)46 void DWARFUnitVector::addUnitsForDWOSection(DWARFContext &C,
47 const DWARFSection &DWOSection,
48 DWARFSectionKind SectionKind,
49 bool Lazy) {
50 const DWARFObject &D = C.getDWARFObj();
51 addUnitsImpl(C, D, DWOSection, C.getDebugAbbrevDWO(), &D.getRangesDWOSection(),
52 &D.getLocDWOSection(), D.getStrDWOSection(),
53 D.getStrOffsetsDWOSection(), &D.getAddrSection(),
54 D.getLineDWOSection(), C.isLittleEndian(), true, Lazy,
55 SectionKind);
56 }
57
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)58 void DWARFUnitVector::addUnitsImpl(
59 DWARFContext &Context, const DWARFObject &Obj, const DWARFSection &Section,
60 const DWARFDebugAbbrev *DA, const DWARFSection *RS,
61 const DWARFSection *LocSection, StringRef SS, const DWARFSection &SOS,
62 const DWARFSection *AOS, const DWARFSection &LS, bool LE, bool IsDWO,
63 bool Lazy, DWARFSectionKind SectionKind) {
64 DWARFDataExtractor Data(Obj, Section, LE, 0);
65 // Lazy initialization of Parser, now that we have all section info.
66 if (!Parser) {
67 Parser = [=, &Context, &Obj, &Section, &SOS,
68 &LS](uint64_t Offset, DWARFSectionKind SectionKind,
69 const DWARFSection *CurSection,
70 const DWARFUnitIndex::Entry *IndexEntry)
71 -> std::unique_ptr<DWARFUnit> {
72 const DWARFSection &InfoSection = CurSection ? *CurSection : Section;
73 DWARFDataExtractor Data(Obj, InfoSection, LE, 0);
74 if (!Data.isValidOffset(Offset))
75 return nullptr;
76 DWARFUnitHeader Header;
77 if (!Header.extract(Context, Data, &Offset, SectionKind))
78 return nullptr;
79 if (!IndexEntry && IsDWO) {
80 const DWARFUnitIndex &Index = getDWARFUnitIndex(
81 Context, Header.isTypeUnit() ? DW_SECT_EXT_TYPES : DW_SECT_INFO);
82 IndexEntry = Index.getFromOffset(Header.getOffset());
83 }
84 if (IndexEntry && !Header.applyIndexEntry(IndexEntry))
85 return nullptr;
86 std::unique_ptr<DWARFUnit> U;
87 if (Header.isTypeUnit())
88 U = std::make_unique<DWARFTypeUnit>(Context, InfoSection, Header, DA,
89 RS, LocSection, SS, SOS, AOS, LS,
90 LE, IsDWO, *this);
91 else
92 U = std::make_unique<DWARFCompileUnit>(Context, InfoSection, Header,
93 DA, RS, LocSection, SS, SOS,
94 AOS, LS, LE, IsDWO, *this);
95 return U;
96 };
97 }
98 if (Lazy)
99 return;
100 // Find a reasonable insertion point within the vector. We skip over
101 // (a) units from a different section, (b) units from the same section
102 // but with lower offset-within-section. This keeps units in order
103 // within a section, although not necessarily within the object file,
104 // even if we do lazy parsing.
105 auto I = this->begin();
106 uint64_t Offset = 0;
107 while (Data.isValidOffset(Offset)) {
108 if (I != this->end() &&
109 (&(*I)->getInfoSection() != &Section || (*I)->getOffset() == Offset)) {
110 ++I;
111 continue;
112 }
113 auto U = Parser(Offset, SectionKind, &Section, nullptr);
114 // If parsing failed, we're done with this section.
115 if (!U)
116 break;
117 Offset = U->getNextUnitOffset();
118 I = std::next(this->insert(I, std::move(U)));
119 }
120 }
121
addUnit(std::unique_ptr<DWARFUnit> Unit)122 DWARFUnit *DWARFUnitVector::addUnit(std::unique_ptr<DWARFUnit> Unit) {
123 auto I = std::upper_bound(begin(), end(), Unit,
124 [](const std::unique_ptr<DWARFUnit> &LHS,
125 const std::unique_ptr<DWARFUnit> &RHS) {
126 return LHS->getOffset() < RHS->getOffset();
127 });
128 return this->insert(I, std::move(Unit))->get();
129 }
130
getUnitForOffset(uint64_t Offset) const131 DWARFUnit *DWARFUnitVector::getUnitForOffset(uint64_t Offset) const {
132 auto end = begin() + getNumInfoUnits();
133 auto *CU =
134 std::upper_bound(begin(), end, Offset,
135 [](uint64_t LHS, const std::unique_ptr<DWARFUnit> &RHS) {
136 return LHS < RHS->getNextUnitOffset();
137 });
138 if (CU != end && (*CU)->getOffset() <= Offset)
139 return CU->get();
140 return nullptr;
141 }
142
143 DWARFUnit *
getUnitForIndexEntry(const DWARFUnitIndex::Entry & E)144 DWARFUnitVector::getUnitForIndexEntry(const DWARFUnitIndex::Entry &E) {
145 const auto *CUOff = E.getContribution(DW_SECT_INFO);
146 if (!CUOff)
147 return nullptr;
148
149 auto Offset = CUOff->Offset;
150 auto end = begin() + getNumInfoUnits();
151
152 auto *CU =
153 std::upper_bound(begin(), end, CUOff->Offset,
154 [](uint64_t LHS, const std::unique_ptr<DWARFUnit> &RHS) {
155 return LHS < RHS->getNextUnitOffset();
156 });
157 if (CU != end && (*CU)->getOffset() <= Offset)
158 return CU->get();
159
160 if (!Parser)
161 return nullptr;
162
163 auto U = Parser(Offset, DW_SECT_INFO, nullptr, &E);
164 if (!U)
165 U = nullptr;
166
167 auto *NewCU = U.get();
168 this->insert(CU, std::move(U));
169 ++NumInfoUnits;
170 return NewCU;
171 }
172
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)173 DWARFUnit::DWARFUnit(DWARFContext &DC, const DWARFSection &Section,
174 const DWARFUnitHeader &Header, const DWARFDebugAbbrev *DA,
175 const DWARFSection *RS, const DWARFSection *LocSection,
176 StringRef SS, const DWARFSection &SOS,
177 const DWARFSection *AOS, const DWARFSection &LS, bool LE,
178 bool IsDWO, const DWARFUnitVector &UnitVector)
179 : Context(DC), InfoSection(Section), Header(Header), Abbrev(DA),
180 RangeSection(RS), LineSection(LS), StringSection(SS),
181 StringOffsetSection(SOS), AddrOffsetSection(AOS), isLittleEndian(LE),
182 IsDWO(IsDWO), UnitVector(UnitVector) {
183 clear();
184 }
185
186 DWARFUnit::~DWARFUnit() = default;
187
getDebugInfoExtractor() const188 DWARFDataExtractor DWARFUnit::getDebugInfoExtractor() const {
189 return DWARFDataExtractor(Context.getDWARFObj(), InfoSection, isLittleEndian,
190 getAddressByteSize());
191 }
192
193 Optional<object::SectionedAddress>
getAddrOffsetSectionItem(uint32_t Index) const194 DWARFUnit::getAddrOffsetSectionItem(uint32_t Index) const {
195 if (IsDWO) {
196 auto R = Context.info_section_units();
197 // Surprising if a DWO file has more than one skeleton unit in it - this
198 // probably shouldn't be valid, but if a use case is found, here's where to
199 // support it (probably have to linearly search for the matching skeleton CU
200 // here)
201 if (hasSingleElement(R))
202 return (*R.begin())->getAddrOffsetSectionItem(Index);
203 }
204 if (!AddrOffsetSectionBase)
205 return None;
206 uint64_t Offset = *AddrOffsetSectionBase + Index * getAddressByteSize();
207 if (AddrOffsetSection->Data.size() < Offset + getAddressByteSize())
208 return None;
209 DWARFDataExtractor DA(Context.getDWARFObj(), *AddrOffsetSection,
210 isLittleEndian, getAddressByteSize());
211 uint64_t Section;
212 uint64_t Address = DA.getRelocatedAddress(&Offset, &Section);
213 return {{Address, Section}};
214 }
215
getStringOffsetSectionItem(uint32_t Index) const216 Optional<uint64_t> DWARFUnit::getStringOffsetSectionItem(uint32_t Index) const {
217 if (!StringOffsetsTableContribution)
218 return None;
219 unsigned ItemSize = getDwarfStringOffsetsByteSize();
220 uint64_t Offset = getStringOffsetsBase() + Index * ItemSize;
221 if (StringOffsetSection.Data.size() < Offset + ItemSize)
222 return None;
223 DWARFDataExtractor DA(Context.getDWARFObj(), StringOffsetSection,
224 isLittleEndian, 0);
225 return DA.getRelocatedValue(ItemSize, &Offset);
226 }
227
extract(DWARFContext & Context,const DWARFDataExtractor & debug_info,uint64_t * offset_ptr,DWARFSectionKind SectionKind)228 bool DWARFUnitHeader::extract(DWARFContext &Context,
229 const DWARFDataExtractor &debug_info,
230 uint64_t *offset_ptr,
231 DWARFSectionKind SectionKind) {
232 Offset = *offset_ptr;
233 Error Err = Error::success();
234 IndexEntry = nullptr;
235 std::tie(Length, FormParams.Format) =
236 debug_info.getInitialLength(offset_ptr, &Err);
237 FormParams.Version = debug_info.getU16(offset_ptr, &Err);
238 if (FormParams.Version >= 5) {
239 UnitType = debug_info.getU8(offset_ptr, &Err);
240 FormParams.AddrSize = debug_info.getU8(offset_ptr, &Err);
241 AbbrOffset = debug_info.getRelocatedValue(
242 FormParams.getDwarfOffsetByteSize(), offset_ptr, nullptr, &Err);
243 } else {
244 AbbrOffset = debug_info.getRelocatedValue(
245 FormParams.getDwarfOffsetByteSize(), offset_ptr, nullptr, &Err);
246 FormParams.AddrSize = debug_info.getU8(offset_ptr, &Err);
247 // Fake a unit type based on the section type. This isn't perfect,
248 // but distinguishing compile and type units is generally enough.
249 if (SectionKind == DW_SECT_EXT_TYPES)
250 UnitType = DW_UT_type;
251 else
252 UnitType = DW_UT_compile;
253 }
254 if (isTypeUnit()) {
255 TypeHash = debug_info.getU64(offset_ptr, &Err);
256 TypeOffset = debug_info.getUnsigned(
257 offset_ptr, FormParams.getDwarfOffsetByteSize(), &Err);
258 } else if (UnitType == DW_UT_split_compile || UnitType == DW_UT_skeleton)
259 DWOId = debug_info.getU64(offset_ptr, &Err);
260
261 if (errorToBool(std::move(Err)))
262 return false;
263
264 // Header fields all parsed, capture the size of this unit header.
265 assert(*offset_ptr - Offset <= 255 && "unexpected header size");
266 Size = uint8_t(*offset_ptr - Offset);
267
268 // Type offset is unit-relative; should be after the header and before
269 // the end of the current unit.
270 bool TypeOffsetOK =
271 !isTypeUnit()
272 ? true
273 : TypeOffset >= Size &&
274 TypeOffset < getLength() + getUnitLengthFieldByteSize();
275 bool LengthOK = debug_info.isValidOffset(getNextUnitOffset() - 1);
276 bool VersionOK = DWARFContext::isSupportedVersion(getVersion());
277 bool AddrSizeOK = DWARFContext::isAddressSizeSupported(getAddressByteSize());
278
279 if (!LengthOK || !VersionOK || !AddrSizeOK || !TypeOffsetOK)
280 return false;
281
282 // Keep track of the highest DWARF version we encounter across all units.
283 Context.setMaxVersionIfGreater(getVersion());
284 return true;
285 }
286
applyIndexEntry(const DWARFUnitIndex::Entry * Entry)287 bool DWARFUnitHeader::applyIndexEntry(const DWARFUnitIndex::Entry *Entry) {
288 assert(Entry);
289 assert(!IndexEntry);
290 IndexEntry = Entry;
291 if (AbbrOffset)
292 return false;
293 auto *UnitContrib = IndexEntry->getContribution();
294 if (!UnitContrib ||
295 UnitContrib->Length != (getLength() + getUnitLengthFieldByteSize()))
296 return false;
297 auto *AbbrEntry = IndexEntry->getContribution(DW_SECT_ABBREV);
298 if (!AbbrEntry)
299 return false;
300 AbbrOffset = AbbrEntry->Offset;
301 return true;
302 }
303
304 // Parse the rangelist table header, including the optional array of offsets
305 // following it (DWARF v5 and later).
306 template<typename ListTableType>
307 static Expected<ListTableType>
parseListTableHeader(DWARFDataExtractor & DA,uint64_t Offset,DwarfFormat Format)308 parseListTableHeader(DWARFDataExtractor &DA, uint64_t Offset,
309 DwarfFormat Format) {
310 // We are expected to be called with Offset 0 or pointing just past the table
311 // header. Correct Offset in the latter case so that it points to the start
312 // of the header.
313 if (Offset > 0) {
314 uint64_t HeaderSize = DWARFListTableHeader::getHeaderSize(Format);
315 if (Offset < HeaderSize)
316 return createStringError(errc::invalid_argument, "did not detect a valid"
317 " list table with base = 0x%" PRIx64 "\n",
318 Offset);
319 Offset -= HeaderSize;
320 }
321 ListTableType Table;
322 if (Error E = Table.extractHeaderAndOffsets(DA, &Offset))
323 return std::move(E);
324 return Table;
325 }
326
extractRangeList(uint64_t RangeListOffset,DWARFDebugRangeList & RangeList) const327 Error DWARFUnit::extractRangeList(uint64_t RangeListOffset,
328 DWARFDebugRangeList &RangeList) const {
329 // Require that compile unit is extracted.
330 assert(!DieArray.empty());
331 DWARFDataExtractor RangesData(Context.getDWARFObj(), *RangeSection,
332 isLittleEndian, getAddressByteSize());
333 uint64_t ActualRangeListOffset = RangeSectionBase + RangeListOffset;
334 return RangeList.extract(RangesData, &ActualRangeListOffset);
335 }
336
clear()337 void DWARFUnit::clear() {
338 Abbrevs = nullptr;
339 BaseAddr.reset();
340 RangeSectionBase = 0;
341 LocSectionBase = 0;
342 AddrOffsetSectionBase = None;
343 clearDIEs(false);
344 DWO.reset();
345 }
346
getCompilationDir()347 const char *DWARFUnit::getCompilationDir() {
348 return dwarf::toString(getUnitDIE().find(DW_AT_comp_dir), nullptr);
349 }
350
extractDIEsToVector(bool AppendCUDie,bool AppendNonCUDies,std::vector<DWARFDebugInfoEntry> & Dies) const351 void DWARFUnit::extractDIEsToVector(
352 bool AppendCUDie, bool AppendNonCUDies,
353 std::vector<DWARFDebugInfoEntry> &Dies) const {
354 if (!AppendCUDie && !AppendNonCUDies)
355 return;
356
357 // Set the offset to that of the first DIE and calculate the start of the
358 // next compilation unit header.
359 uint64_t DIEOffset = getOffset() + getHeaderSize();
360 uint64_t NextCUOffset = getNextUnitOffset();
361 DWARFDebugInfoEntry DIE;
362 DWARFDataExtractor DebugInfoData = getDebugInfoExtractor();
363 uint32_t Depth = 0;
364 bool IsCUDie = true;
365
366 while (DIE.extractFast(*this, &DIEOffset, DebugInfoData, NextCUOffset,
367 Depth)) {
368 if (IsCUDie) {
369 if (AppendCUDie)
370 Dies.push_back(DIE);
371 if (!AppendNonCUDies)
372 break;
373 // The average bytes per DIE entry has been seen to be
374 // around 14-20 so let's pre-reserve the needed memory for
375 // our DIE entries accordingly.
376 Dies.reserve(Dies.size() + getDebugInfoSize() / 14);
377 IsCUDie = false;
378 } else {
379 Dies.push_back(DIE);
380 }
381
382 if (const DWARFAbbreviationDeclaration *AbbrDecl =
383 DIE.getAbbreviationDeclarationPtr()) {
384 // Normal DIE
385 if (AbbrDecl->hasChildren())
386 ++Depth;
387 } else {
388 // NULL DIE.
389 if (Depth > 0)
390 --Depth;
391 if (Depth == 0)
392 break; // We are done with this compile unit!
393 }
394 }
395
396 // Give a little bit of info if we encounter corrupt DWARF (our offset
397 // should always terminate at or before the start of the next compilation
398 // unit header).
399 if (DIEOffset > NextCUOffset)
400 Context.getWarningHandler()(
401 createStringError(errc::invalid_argument,
402 "DWARF compile unit extends beyond its "
403 "bounds cu 0x%8.8" PRIx64 " "
404 "at 0x%8.8" PRIx64 "\n",
405 getOffset(), DIEOffset));
406 }
407
extractDIEsIfNeeded(bool CUDieOnly)408 void DWARFUnit::extractDIEsIfNeeded(bool CUDieOnly) {
409 if (Error e = tryExtractDIEsIfNeeded(CUDieOnly))
410 Context.getRecoverableErrorHandler()(std::move(e));
411 }
412
tryExtractDIEsIfNeeded(bool CUDieOnly)413 Error DWARFUnit::tryExtractDIEsIfNeeded(bool CUDieOnly) {
414 if ((CUDieOnly && !DieArray.empty()) ||
415 DieArray.size() > 1)
416 return Error::success(); // Already parsed.
417
418 bool HasCUDie = !DieArray.empty();
419 extractDIEsToVector(!HasCUDie, !CUDieOnly, DieArray);
420
421 if (DieArray.empty())
422 return Error::success();
423
424 // If CU DIE was just parsed, copy several attribute values from it.
425 if (HasCUDie)
426 return Error::success();
427
428 DWARFDie UnitDie(this, &DieArray[0]);
429 if (Optional<uint64_t> DWOId = toUnsigned(UnitDie.find(DW_AT_GNU_dwo_id)))
430 Header.setDWOId(*DWOId);
431 if (!IsDWO) {
432 assert(AddrOffsetSectionBase == None);
433 assert(RangeSectionBase == 0);
434 assert(LocSectionBase == 0);
435 AddrOffsetSectionBase = toSectionOffset(UnitDie.find(DW_AT_addr_base));
436 if (!AddrOffsetSectionBase)
437 AddrOffsetSectionBase =
438 toSectionOffset(UnitDie.find(DW_AT_GNU_addr_base));
439 RangeSectionBase = toSectionOffset(UnitDie.find(DW_AT_rnglists_base), 0);
440 LocSectionBase = toSectionOffset(UnitDie.find(DW_AT_loclists_base), 0);
441 }
442
443 // In general, in DWARF v5 and beyond we derive the start of the unit's
444 // contribution to the string offsets table from the unit DIE's
445 // DW_AT_str_offsets_base attribute. Split DWARF units do not use this
446 // attribute, so we assume that there is a contribution to the string
447 // offsets table starting at offset 0 of the debug_str_offsets.dwo section.
448 // In both cases we need to determine the format of the contribution,
449 // which may differ from the unit's format.
450 DWARFDataExtractor DA(Context.getDWARFObj(), StringOffsetSection,
451 isLittleEndian, 0);
452 if (IsDWO || getVersion() >= 5) {
453 auto StringOffsetOrError =
454 IsDWO ? determineStringOffsetsTableContributionDWO(DA)
455 : determineStringOffsetsTableContribution(DA);
456 if (!StringOffsetOrError)
457 return createStringError(errc::invalid_argument,
458 "invalid reference to or invalid content in "
459 ".debug_str_offsets[.dwo]: " +
460 toString(StringOffsetOrError.takeError()));
461
462 StringOffsetsTableContribution = *StringOffsetOrError;
463 }
464
465 // DWARF v5 uses the .debug_rnglists and .debug_rnglists.dwo sections to
466 // describe address ranges.
467 if (getVersion() >= 5) {
468 // In case of DWP, the base offset from the index has to be added.
469 if (IsDWO) {
470 uint64_t ContributionBaseOffset = 0;
471 if (auto *IndexEntry = Header.getIndexEntry())
472 if (auto *Contrib = IndexEntry->getContribution(DW_SECT_RNGLISTS))
473 ContributionBaseOffset = Contrib->Offset;
474 setRangesSection(
475 &Context.getDWARFObj().getRnglistsDWOSection(),
476 ContributionBaseOffset +
477 DWARFListTableHeader::getHeaderSize(Header.getFormat()));
478 } else
479 setRangesSection(&Context.getDWARFObj().getRnglistsSection(),
480 toSectionOffset(UnitDie.find(DW_AT_rnglists_base),
481 DWARFListTableHeader::getHeaderSize(
482 Header.getFormat())));
483 }
484
485 if (IsDWO) {
486 // If we are reading a package file, we need to adjust the location list
487 // data based on the index entries.
488 StringRef Data = Header.getVersion() >= 5
489 ? Context.getDWARFObj().getLoclistsDWOSection().Data
490 : Context.getDWARFObj().getLocDWOSection().Data;
491 if (auto *IndexEntry = Header.getIndexEntry())
492 if (const auto *C = IndexEntry->getContribution(
493 Header.getVersion() >= 5 ? DW_SECT_LOCLISTS : DW_SECT_EXT_LOC))
494 Data = Data.substr(C->Offset, C->Length);
495
496 DWARFDataExtractor DWARFData(Data, isLittleEndian, getAddressByteSize());
497 LocTable =
498 std::make_unique<DWARFDebugLoclists>(DWARFData, Header.getVersion());
499 LocSectionBase = DWARFListTableHeader::getHeaderSize(Header.getFormat());
500 } else if (getVersion() >= 5) {
501 LocTable = std::make_unique<DWARFDebugLoclists>(
502 DWARFDataExtractor(Context.getDWARFObj(),
503 Context.getDWARFObj().getLoclistsSection(),
504 isLittleEndian, getAddressByteSize()),
505 getVersion());
506 } else {
507 LocTable = std::make_unique<DWARFDebugLoc>(DWARFDataExtractor(
508 Context.getDWARFObj(), Context.getDWARFObj().getLocSection(),
509 isLittleEndian, getAddressByteSize()));
510 }
511
512 // Don't fall back to DW_AT_GNU_ranges_base: it should be ignored for
513 // skeleton CU DIE, so that DWARF users not aware of it are not broken.
514 return Error::success();
515 }
516
parseDWO()517 bool DWARFUnit::parseDWO() {
518 if (IsDWO)
519 return false;
520 if (DWO.get())
521 return false;
522 DWARFDie UnitDie = getUnitDIE();
523 if (!UnitDie)
524 return false;
525 auto DWOFileName = getVersion() >= 5
526 ? dwarf::toString(UnitDie.find(DW_AT_dwo_name))
527 : dwarf::toString(UnitDie.find(DW_AT_GNU_dwo_name));
528 if (!DWOFileName)
529 return false;
530 auto CompilationDir = dwarf::toString(UnitDie.find(DW_AT_comp_dir));
531 SmallString<16> AbsolutePath;
532 if (sys::path::is_relative(*DWOFileName) && CompilationDir &&
533 *CompilationDir) {
534 sys::path::append(AbsolutePath, *CompilationDir);
535 }
536 sys::path::append(AbsolutePath, *DWOFileName);
537 auto DWOId = getDWOId();
538 if (!DWOId)
539 return false;
540 auto DWOContext = Context.getDWOContext(AbsolutePath);
541 if (!DWOContext)
542 return false;
543
544 DWARFCompileUnit *DWOCU = DWOContext->getDWOCompileUnitForHash(*DWOId);
545 if (!DWOCU)
546 return false;
547 DWO = std::shared_ptr<DWARFCompileUnit>(std::move(DWOContext), DWOCU);
548 // Share .debug_addr and .debug_ranges section with compile unit in .dwo
549 if (AddrOffsetSectionBase)
550 DWO->setAddrOffsetSection(AddrOffsetSection, *AddrOffsetSectionBase);
551 if (getVersion() >= 5) {
552 DWO->setRangesSection(&Context.getDWARFObj().getRnglistsDWOSection(),
553 DWARFListTableHeader::getHeaderSize(getFormat()));
554 } else {
555 auto DWORangesBase = UnitDie.getRangesBaseAttribute();
556 DWO->setRangesSection(RangeSection, DWORangesBase ? *DWORangesBase : 0);
557 }
558
559 return true;
560 }
561
clearDIEs(bool KeepCUDie)562 void DWARFUnit::clearDIEs(bool KeepCUDie) {
563 if (DieArray.size() > (unsigned)KeepCUDie) {
564 DieArray.resize((unsigned)KeepCUDie);
565 DieArray.shrink_to_fit();
566 }
567 }
568
569 Expected<DWARFAddressRangesVector>
findRnglistFromOffset(uint64_t Offset)570 DWARFUnit::findRnglistFromOffset(uint64_t Offset) {
571 if (getVersion() <= 4) {
572 DWARFDebugRangeList RangeList;
573 if (Error E = extractRangeList(Offset, RangeList))
574 return std::move(E);
575 return RangeList.getAbsoluteRanges(getBaseAddress());
576 }
577 DWARFDataExtractor RangesData(Context.getDWARFObj(), *RangeSection,
578 isLittleEndian, Header.getAddressByteSize());
579 DWARFDebugRnglistTable RnglistTable;
580 auto RangeListOrError = RnglistTable.findList(RangesData, Offset);
581 if (RangeListOrError)
582 return RangeListOrError.get().getAbsoluteRanges(getBaseAddress(), *this);
583 return RangeListOrError.takeError();
584 }
585
586 Expected<DWARFAddressRangesVector>
findRnglistFromIndex(uint32_t Index)587 DWARFUnit::findRnglistFromIndex(uint32_t Index) {
588 if (auto Offset = getRnglistOffset(Index))
589 return findRnglistFromOffset(*Offset);
590
591 return createStringError(errc::invalid_argument,
592 "invalid range list table index %d (possibly "
593 "missing the entire range list table)",
594 Index);
595 }
596
collectAddressRanges()597 Expected<DWARFAddressRangesVector> DWARFUnit::collectAddressRanges() {
598 DWARFDie UnitDie = getUnitDIE();
599 if (!UnitDie)
600 return createStringError(errc::invalid_argument, "No unit DIE");
601
602 // First, check if unit DIE describes address ranges for the whole unit.
603 auto CUDIERangesOrError = UnitDie.getAddressRanges();
604 if (!CUDIERangesOrError)
605 return createStringError(errc::invalid_argument,
606 "decoding address ranges: %s",
607 toString(CUDIERangesOrError.takeError()).c_str());
608 return *CUDIERangesOrError;
609 }
610
611 Expected<DWARFLocationExpressionsVector>
findLoclistFromOffset(uint64_t Offset)612 DWARFUnit::findLoclistFromOffset(uint64_t Offset) {
613 DWARFLocationExpressionsVector Result;
614
615 Error InterpretationError = Error::success();
616
617 Error ParseError = getLocationTable().visitAbsoluteLocationList(
618 Offset, getBaseAddress(),
619 [this](uint32_t Index) { return getAddrOffsetSectionItem(Index); },
620 [&](Expected<DWARFLocationExpression> L) {
621 if (L)
622 Result.push_back(std::move(*L));
623 else
624 InterpretationError =
625 joinErrors(L.takeError(), std::move(InterpretationError));
626 return !InterpretationError;
627 });
628
629 if (ParseError || InterpretationError)
630 return joinErrors(std::move(ParseError), std::move(InterpretationError));
631
632 return Result;
633 }
634
updateAddressDieMap(DWARFDie Die)635 void DWARFUnit::updateAddressDieMap(DWARFDie Die) {
636 if (Die.isSubroutineDIE()) {
637 auto DIERangesOrError = Die.getAddressRanges();
638 if (DIERangesOrError) {
639 for (const auto &R : DIERangesOrError.get()) {
640 // Ignore 0-sized ranges.
641 if (R.LowPC == R.HighPC)
642 continue;
643 auto B = AddrDieMap.upper_bound(R.LowPC);
644 if (B != AddrDieMap.begin() && R.LowPC < (--B)->second.first) {
645 // The range is a sub-range of existing ranges, we need to split the
646 // existing range.
647 if (R.HighPC < B->second.first)
648 AddrDieMap[R.HighPC] = B->second;
649 if (R.LowPC > B->first)
650 AddrDieMap[B->first].first = R.LowPC;
651 }
652 AddrDieMap[R.LowPC] = std::make_pair(R.HighPC, Die);
653 }
654 } else
655 llvm::consumeError(DIERangesOrError.takeError());
656 }
657 // Parent DIEs are added to the AddrDieMap prior to the Children DIEs to
658 // simplify the logic to update AddrDieMap. The child's range will always
659 // be equal or smaller than the parent's range. With this assumption, when
660 // adding one range into the map, it will at most split a range into 3
661 // sub-ranges.
662 for (DWARFDie Child = Die.getFirstChild(); Child; Child = Child.getSibling())
663 updateAddressDieMap(Child);
664 }
665
getSubroutineForAddress(uint64_t Address)666 DWARFDie DWARFUnit::getSubroutineForAddress(uint64_t Address) {
667 extractDIEsIfNeeded(false);
668 if (AddrDieMap.empty())
669 updateAddressDieMap(getUnitDIE());
670 auto R = AddrDieMap.upper_bound(Address);
671 if (R == AddrDieMap.begin())
672 return DWARFDie();
673 // upper_bound's previous item contains Address.
674 --R;
675 if (Address >= R->second.first)
676 return DWARFDie();
677 return R->second.second;
678 }
679
680 void
getInlinedChainForAddress(uint64_t Address,SmallVectorImpl<DWARFDie> & InlinedChain)681 DWARFUnit::getInlinedChainForAddress(uint64_t Address,
682 SmallVectorImpl<DWARFDie> &InlinedChain) {
683 assert(InlinedChain.empty());
684 // Try to look for subprogram DIEs in the DWO file.
685 parseDWO();
686 // First, find the subroutine that contains the given address (the leaf
687 // of inlined chain).
688 DWARFDie SubroutineDIE =
689 (DWO ? *DWO : *this).getSubroutineForAddress(Address);
690
691 if (!SubroutineDIE)
692 return;
693
694 while (!SubroutineDIE.isSubprogramDIE()) {
695 if (SubroutineDIE.getTag() == DW_TAG_inlined_subroutine)
696 InlinedChain.push_back(SubroutineDIE);
697 SubroutineDIE = SubroutineDIE.getParent();
698 }
699 InlinedChain.push_back(SubroutineDIE);
700 }
701
getDWARFUnitIndex(DWARFContext & Context,DWARFSectionKind Kind)702 const DWARFUnitIndex &llvm::getDWARFUnitIndex(DWARFContext &Context,
703 DWARFSectionKind Kind) {
704 if (Kind == DW_SECT_INFO)
705 return Context.getCUIndex();
706 assert(Kind == DW_SECT_EXT_TYPES);
707 return Context.getTUIndex();
708 }
709
getParent(const DWARFDebugInfoEntry * Die)710 DWARFDie DWARFUnit::getParent(const DWARFDebugInfoEntry *Die) {
711 if (!Die)
712 return DWARFDie();
713 const uint32_t Depth = Die->getDepth();
714 // Unit DIEs always have a depth of zero and never have parents.
715 if (Depth == 0)
716 return DWARFDie();
717 // Depth of 1 always means parent is the compile/type unit.
718 if (Depth == 1)
719 return getUnitDIE();
720 // Look for previous DIE with a depth that is one less than the Die's depth.
721 const uint32_t ParentDepth = Depth - 1;
722 for (uint32_t I = getDIEIndex(Die) - 1; I > 0; --I) {
723 if (DieArray[I].getDepth() == ParentDepth)
724 return DWARFDie(this, &DieArray[I]);
725 }
726 return DWARFDie();
727 }
728
getSibling(const DWARFDebugInfoEntry * Die)729 DWARFDie DWARFUnit::getSibling(const DWARFDebugInfoEntry *Die) {
730 if (!Die)
731 return DWARFDie();
732 uint32_t Depth = Die->getDepth();
733 // Unit DIEs always have a depth of zero and never have siblings.
734 if (Depth == 0)
735 return DWARFDie();
736 // NULL DIEs don't have siblings.
737 if (Die->getAbbreviationDeclarationPtr() == nullptr)
738 return DWARFDie();
739
740 // Find the next DIE whose depth is the same as the Die's depth.
741 for (size_t I = getDIEIndex(Die) + 1, EndIdx = DieArray.size(); I < EndIdx;
742 ++I) {
743 if (DieArray[I].getDepth() == Depth)
744 return DWARFDie(this, &DieArray[I]);
745 }
746 return DWARFDie();
747 }
748
getPreviousSibling(const DWARFDebugInfoEntry * Die)749 DWARFDie DWARFUnit::getPreviousSibling(const DWARFDebugInfoEntry *Die) {
750 if (!Die)
751 return DWARFDie();
752 uint32_t Depth = Die->getDepth();
753 // Unit DIEs always have a depth of zero and never have siblings.
754 if (Depth == 0)
755 return DWARFDie();
756
757 // Find the previous DIE whose depth is the same as the Die's depth.
758 for (size_t I = getDIEIndex(Die); I > 0;) {
759 --I;
760 if (DieArray[I].getDepth() == Depth - 1)
761 return DWARFDie();
762 if (DieArray[I].getDepth() == Depth)
763 return DWARFDie(this, &DieArray[I]);
764 }
765 return DWARFDie();
766 }
767
getFirstChild(const DWARFDebugInfoEntry * Die)768 DWARFDie DWARFUnit::getFirstChild(const DWARFDebugInfoEntry *Die) {
769 if (!Die->hasChildren())
770 return DWARFDie();
771
772 // We do not want access out of bounds when parsing corrupted debug data.
773 size_t I = getDIEIndex(Die) + 1;
774 if (I >= DieArray.size())
775 return DWARFDie();
776 return DWARFDie(this, &DieArray[I]);
777 }
778
getLastChild(const DWARFDebugInfoEntry * Die)779 DWARFDie DWARFUnit::getLastChild(const DWARFDebugInfoEntry *Die) {
780 if (!Die->hasChildren())
781 return DWARFDie();
782
783 uint32_t Depth = Die->getDepth();
784 for (size_t I = getDIEIndex(Die) + 1, EndIdx = DieArray.size(); I < EndIdx;
785 ++I) {
786 if (DieArray[I].getDepth() == Depth + 1 &&
787 DieArray[I].getTag() == dwarf::DW_TAG_null)
788 return DWARFDie(this, &DieArray[I]);
789 assert(DieArray[I].getDepth() > Depth && "Not processing children?");
790 }
791 return DWARFDie();
792 }
793
getAbbreviations() const794 const DWARFAbbreviationDeclarationSet *DWARFUnit::getAbbreviations() const {
795 if (!Abbrevs)
796 Abbrevs = Abbrev->getAbbreviationDeclarationSet(Header.getAbbrOffset());
797 return Abbrevs;
798 }
799
getBaseAddress()800 llvm::Optional<object::SectionedAddress> DWARFUnit::getBaseAddress() {
801 if (BaseAddr)
802 return BaseAddr;
803
804 DWARFDie UnitDie = getUnitDIE();
805 Optional<DWARFFormValue> PC = UnitDie.find({DW_AT_low_pc, DW_AT_entry_pc});
806 BaseAddr = toSectionedAddress(PC);
807 return BaseAddr;
808 }
809
810 Expected<StrOffsetsContributionDescriptor>
validateContributionSize(DWARFDataExtractor & DA)811 StrOffsetsContributionDescriptor::validateContributionSize(
812 DWARFDataExtractor &DA) {
813 uint8_t EntrySize = getDwarfOffsetByteSize();
814 // In order to ensure that we don't read a partial record at the end of
815 // the section we validate for a multiple of the entry size.
816 uint64_t ValidationSize = alignTo(Size, EntrySize);
817 // Guard against overflow.
818 if (ValidationSize >= Size)
819 if (DA.isValidOffsetForDataOfSize((uint32_t)Base, ValidationSize))
820 return *this;
821 return createStringError(errc::invalid_argument, "length exceeds section size");
822 }
823
824 // Look for a DWARF64-formatted contribution to the string offsets table
825 // starting at a given offset and record it in a descriptor.
826 static Expected<StrOffsetsContributionDescriptor>
parseDWARF64StringOffsetsTableHeader(DWARFDataExtractor & DA,uint64_t Offset)827 parseDWARF64StringOffsetsTableHeader(DWARFDataExtractor &DA, uint64_t Offset) {
828 if (!DA.isValidOffsetForDataOfSize(Offset, 16))
829 return createStringError(errc::invalid_argument, "section offset exceeds section size");
830
831 if (DA.getU32(&Offset) != dwarf::DW_LENGTH_DWARF64)
832 return createStringError(errc::invalid_argument, "32 bit contribution referenced from a 64 bit unit");
833
834 uint64_t Size = DA.getU64(&Offset);
835 uint8_t Version = DA.getU16(&Offset);
836 (void)DA.getU16(&Offset); // padding
837 // The encoded length includes the 2-byte version field and the 2-byte
838 // padding, so we need to subtract them out when we populate the descriptor.
839 return StrOffsetsContributionDescriptor(Offset, Size - 4, Version, DWARF64);
840 }
841
842 // Look for a DWARF32-formatted contribution to the string offsets table
843 // starting at a given offset and record it in a descriptor.
844 static Expected<StrOffsetsContributionDescriptor>
parseDWARF32StringOffsetsTableHeader(DWARFDataExtractor & DA,uint64_t Offset)845 parseDWARF32StringOffsetsTableHeader(DWARFDataExtractor &DA, uint64_t Offset) {
846 if (!DA.isValidOffsetForDataOfSize(Offset, 8))
847 return createStringError(errc::invalid_argument, "section offset exceeds section size");
848
849 uint32_t ContributionSize = DA.getU32(&Offset);
850 if (ContributionSize >= dwarf::DW_LENGTH_lo_reserved)
851 return createStringError(errc::invalid_argument, "invalid length");
852
853 uint8_t Version = DA.getU16(&Offset);
854 (void)DA.getU16(&Offset); // padding
855 // The encoded length includes the 2-byte version field and the 2-byte
856 // padding, so we need to subtract them out when we populate the descriptor.
857 return StrOffsetsContributionDescriptor(Offset, ContributionSize - 4, Version,
858 DWARF32);
859 }
860
861 static Expected<StrOffsetsContributionDescriptor>
parseDWARFStringOffsetsTableHeader(DWARFDataExtractor & DA,llvm::dwarf::DwarfFormat Format,uint64_t Offset)862 parseDWARFStringOffsetsTableHeader(DWARFDataExtractor &DA,
863 llvm::dwarf::DwarfFormat Format,
864 uint64_t Offset) {
865 StrOffsetsContributionDescriptor Desc;
866 switch (Format) {
867 case dwarf::DwarfFormat::DWARF64: {
868 if (Offset < 16)
869 return createStringError(errc::invalid_argument, "insufficient space for 64 bit header prefix");
870 auto DescOrError = parseDWARF64StringOffsetsTableHeader(DA, Offset - 16);
871 if (!DescOrError)
872 return DescOrError.takeError();
873 Desc = *DescOrError;
874 break;
875 }
876 case dwarf::DwarfFormat::DWARF32: {
877 if (Offset < 8)
878 return createStringError(errc::invalid_argument, "insufficient space for 32 bit header prefix");
879 auto DescOrError = parseDWARF32StringOffsetsTableHeader(DA, Offset - 8);
880 if (!DescOrError)
881 return DescOrError.takeError();
882 Desc = *DescOrError;
883 break;
884 }
885 }
886 return Desc.validateContributionSize(DA);
887 }
888
889 Expected<Optional<StrOffsetsContributionDescriptor>>
determineStringOffsetsTableContribution(DWARFDataExtractor & DA)890 DWARFUnit::determineStringOffsetsTableContribution(DWARFDataExtractor &DA) {
891 assert(!IsDWO);
892 auto OptOffset = toSectionOffset(getUnitDIE().find(DW_AT_str_offsets_base));
893 if (!OptOffset)
894 return None;
895 auto DescOrError =
896 parseDWARFStringOffsetsTableHeader(DA, Header.getFormat(), *OptOffset);
897 if (!DescOrError)
898 return DescOrError.takeError();
899 return *DescOrError;
900 }
901
902 Expected<Optional<StrOffsetsContributionDescriptor>>
determineStringOffsetsTableContributionDWO(DWARFDataExtractor & DA)903 DWARFUnit::determineStringOffsetsTableContributionDWO(DWARFDataExtractor & DA) {
904 assert(IsDWO);
905 uint64_t Offset = 0;
906 auto IndexEntry = Header.getIndexEntry();
907 const auto *C =
908 IndexEntry ? IndexEntry->getContribution(DW_SECT_STR_OFFSETS) : nullptr;
909 if (C)
910 Offset = C->Offset;
911 if (getVersion() >= 5) {
912 if (DA.getData().data() == nullptr)
913 return None;
914 Offset += Header.getFormat() == dwarf::DwarfFormat::DWARF32 ? 8 : 16;
915 // Look for a valid contribution at the given offset.
916 auto DescOrError = parseDWARFStringOffsetsTableHeader(DA, Header.getFormat(), Offset);
917 if (!DescOrError)
918 return DescOrError.takeError();
919 return *DescOrError;
920 }
921 // Prior to DWARF v5, we derive the contribution size from the
922 // index table (in a package file). In a .dwo file it is simply
923 // the length of the string offsets section.
924 StrOffsetsContributionDescriptor Desc;
925 if (C)
926 Desc = StrOffsetsContributionDescriptor(C->Offset, C->Length, 4,
927 Header.getFormat());
928 else if (!IndexEntry && !StringOffsetSection.Data.empty())
929 Desc = StrOffsetsContributionDescriptor(0, StringOffsetSection.Data.size(),
930 4, Header.getFormat());
931 else
932 return None;
933 auto DescOrError = Desc.validateContributionSize(DA);
934 if (!DescOrError)
935 return DescOrError.takeError();
936 return *DescOrError;
937 }
938
getRnglistOffset(uint32_t Index)939 Optional<uint64_t> DWARFUnit::getRnglistOffset(uint32_t Index) {
940 DataExtractor RangesData(RangeSection->Data, isLittleEndian,
941 getAddressByteSize());
942 DWARFDataExtractor RangesDA(Context.getDWARFObj(), *RangeSection,
943 isLittleEndian, 0);
944 if (Optional<uint64_t> Off = llvm::DWARFListTableHeader::getOffsetEntry(
945 RangesData, RangeSectionBase, getFormat(), Index))
946 return *Off + RangeSectionBase;
947 return None;
948 }
949
getLoclistOffset(uint32_t Index)950 Optional<uint64_t> DWARFUnit::getLoclistOffset(uint32_t Index) {
951 if (Optional<uint64_t> Off = llvm::DWARFListTableHeader::getOffsetEntry(
952 LocTable->getData(), LocSectionBase, getFormat(), Index))
953 return *Off + LocSectionBase;
954 return None;
955 }
956