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 (!AddrOffsetSectionBase) {
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 (IsDWO && hasSingleElement(R))
202 return (*R.begin())->getAddrOffsetSectionItem(Index);
203
204 return None;
205 }
206
207 uint64_t Offset = *AddrOffsetSectionBase + Index * getAddressByteSize();
208 if (AddrOffsetSection->Data.size() < Offset + getAddressByteSize())
209 return None;
210 DWARFDataExtractor DA(Context.getDWARFObj(), *AddrOffsetSection,
211 isLittleEndian, getAddressByteSize());
212 uint64_t Section;
213 uint64_t Address = DA.getRelocatedAddress(&Offset, &Section);
214 return {{Address, Section}};
215 }
216
getStringOffsetSectionItem(uint32_t Index) const217 Optional<uint64_t> DWARFUnit::getStringOffsetSectionItem(uint32_t Index) const {
218 if (!StringOffsetsTableContribution)
219 return None;
220 unsigned ItemSize = getDwarfStringOffsetsByteSize();
221 uint64_t Offset = getStringOffsetsBase() + Index * ItemSize;
222 if (StringOffsetSection.Data.size() < Offset + ItemSize)
223 return None;
224 DWARFDataExtractor DA(Context.getDWARFObj(), StringOffsetSection,
225 isLittleEndian, 0);
226 return DA.getRelocatedValue(ItemSize, &Offset);
227 }
228
extract(DWARFContext & Context,const DWARFDataExtractor & debug_info,uint64_t * offset_ptr,DWARFSectionKind SectionKind)229 bool DWARFUnitHeader::extract(DWARFContext &Context,
230 const DWARFDataExtractor &debug_info,
231 uint64_t *offset_ptr,
232 DWARFSectionKind SectionKind) {
233 Offset = *offset_ptr;
234 Error Err = Error::success();
235 IndexEntry = nullptr;
236 std::tie(Length, FormParams.Format) =
237 debug_info.getInitialLength(offset_ptr, &Err);
238 FormParams.Version = debug_info.getU16(offset_ptr, &Err);
239 if (FormParams.Version >= 5) {
240 UnitType = debug_info.getU8(offset_ptr, &Err);
241 FormParams.AddrSize = debug_info.getU8(offset_ptr, &Err);
242 AbbrOffset = debug_info.getRelocatedValue(
243 FormParams.getDwarfOffsetByteSize(), offset_ptr, nullptr, &Err);
244 } else {
245 AbbrOffset = debug_info.getRelocatedValue(
246 FormParams.getDwarfOffsetByteSize(), offset_ptr, nullptr, &Err);
247 FormParams.AddrSize = debug_info.getU8(offset_ptr, &Err);
248 // Fake a unit type based on the section type. This isn't perfect,
249 // but distinguishing compile and type units is generally enough.
250 if (SectionKind == DW_SECT_EXT_TYPES)
251 UnitType = DW_UT_type;
252 else
253 UnitType = DW_UT_compile;
254 }
255 if (isTypeUnit()) {
256 TypeHash = debug_info.getU64(offset_ptr, &Err);
257 TypeOffset = debug_info.getUnsigned(
258 offset_ptr, FormParams.getDwarfOffsetByteSize(), &Err);
259 } else if (UnitType == DW_UT_split_compile || UnitType == DW_UT_skeleton)
260 DWOId = debug_info.getU64(offset_ptr, &Err);
261
262 if (errorToBool(std::move(Err)))
263 return false;
264
265 // Header fields all parsed, capture the size of this unit header.
266 assert(*offset_ptr - Offset <= 255 && "unexpected header size");
267 Size = uint8_t(*offset_ptr - Offset);
268
269 // Type offset is unit-relative; should be after the header and before
270 // the end of the current unit.
271 bool TypeOffsetOK =
272 !isTypeUnit()
273 ? true
274 : TypeOffset >= Size &&
275 TypeOffset < getLength() + getUnitLengthFieldByteSize();
276 bool LengthOK = debug_info.isValidOffset(getNextUnitOffset() - 1);
277 bool VersionOK = DWARFContext::isSupportedVersion(getVersion());
278 bool AddrSizeOK = DWARFContext::isAddressSizeSupported(getAddressByteSize());
279
280 if (!LengthOK || !VersionOK || !AddrSizeOK || !TypeOffsetOK)
281 return false;
282
283 // Keep track of the highest DWARF version we encounter across all units.
284 Context.setMaxVersionIfGreater(getVersion());
285 return true;
286 }
287
applyIndexEntry(const DWARFUnitIndex::Entry * Entry)288 bool DWARFUnitHeader::applyIndexEntry(const DWARFUnitIndex::Entry *Entry) {
289 assert(Entry);
290 assert(!IndexEntry);
291 IndexEntry = Entry;
292 if (AbbrOffset)
293 return false;
294 auto *UnitContrib = IndexEntry->getContribution();
295 if (!UnitContrib ||
296 UnitContrib->Length != (getLength() + getUnitLengthFieldByteSize()))
297 return false;
298 auto *AbbrEntry = IndexEntry->getContribution(DW_SECT_ABBREV);
299 if (!AbbrEntry)
300 return false;
301 AbbrOffset = AbbrEntry->Offset;
302 return true;
303 }
304
305 // Parse the rangelist table header, including the optional array of offsets
306 // following it (DWARF v5 and later).
307 template<typename ListTableType>
308 static Expected<ListTableType>
parseListTableHeader(DWARFDataExtractor & DA,uint64_t Offset,DwarfFormat Format)309 parseListTableHeader(DWARFDataExtractor &DA, uint64_t Offset,
310 DwarfFormat Format) {
311 // We are expected to be called with Offset 0 or pointing just past the table
312 // header. Correct Offset in the latter case so that it points to the start
313 // of the header.
314 if (Offset > 0) {
315 uint64_t HeaderSize = DWARFListTableHeader::getHeaderSize(Format);
316 if (Offset < HeaderSize)
317 return createStringError(errc::invalid_argument, "did not detect a valid"
318 " list table with base = 0x%" PRIx64 "\n",
319 Offset);
320 Offset -= HeaderSize;
321 }
322 ListTableType Table;
323 if (Error E = Table.extractHeaderAndOffsets(DA, &Offset))
324 return std::move(E);
325 return Table;
326 }
327
extractRangeList(uint64_t RangeListOffset,DWARFDebugRangeList & RangeList) const328 Error DWARFUnit::extractRangeList(uint64_t RangeListOffset,
329 DWARFDebugRangeList &RangeList) const {
330 // Require that compile unit is extracted.
331 assert(!DieArray.empty());
332 DWARFDataExtractor RangesData(Context.getDWARFObj(), *RangeSection,
333 isLittleEndian, getAddressByteSize());
334 uint64_t ActualRangeListOffset = RangeSectionBase + RangeListOffset;
335 return RangeList.extract(RangesData, &ActualRangeListOffset);
336 }
337
clear()338 void DWARFUnit::clear() {
339 Abbrevs = nullptr;
340 BaseAddr.reset();
341 RangeSectionBase = 0;
342 LocSectionBase = 0;
343 AddrOffsetSectionBase = None;
344 clearDIEs(false);
345 DWO.reset();
346 }
347
getCompilationDir()348 const char *DWARFUnit::getCompilationDir() {
349 return dwarf::toString(getUnitDIE().find(DW_AT_comp_dir), nullptr);
350 }
351
extractDIEsToVector(bool AppendCUDie,bool AppendNonCUDies,std::vector<DWARFDebugInfoEntry> & Dies) const352 void DWARFUnit::extractDIEsToVector(
353 bool AppendCUDie, bool AppendNonCUDies,
354 std::vector<DWARFDebugInfoEntry> &Dies) const {
355 if (!AppendCUDie && !AppendNonCUDies)
356 return;
357
358 // Set the offset to that of the first DIE and calculate the start of the
359 // next compilation unit header.
360 uint64_t DIEOffset = getOffset() + getHeaderSize();
361 uint64_t NextCUOffset = getNextUnitOffset();
362 DWARFDebugInfoEntry DIE;
363 DWARFDataExtractor DebugInfoData = getDebugInfoExtractor();
364 uint32_t Depth = 0;
365 bool IsCUDie = true;
366
367 while (DIE.extractFast(*this, &DIEOffset, DebugInfoData, NextCUOffset,
368 Depth)) {
369 if (IsCUDie) {
370 if (AppendCUDie)
371 Dies.push_back(DIE);
372 if (!AppendNonCUDies)
373 break;
374 // The average bytes per DIE entry has been seen to be
375 // around 14-20 so let's pre-reserve the needed memory for
376 // our DIE entries accordingly.
377 Dies.reserve(Dies.size() + getDebugInfoSize() / 14);
378 IsCUDie = false;
379 } else {
380 Dies.push_back(DIE);
381 }
382
383 if (const DWARFAbbreviationDeclaration *AbbrDecl =
384 DIE.getAbbreviationDeclarationPtr()) {
385 // Normal DIE
386 if (AbbrDecl->hasChildren())
387 ++Depth;
388 } else {
389 // NULL DIE.
390 if (Depth > 0)
391 --Depth;
392 if (Depth == 0)
393 break; // We are done with this compile unit!
394 }
395 }
396
397 // Give a little bit of info if we encounter corrupt DWARF (our offset
398 // should always terminate at or before the start of the next compilation
399 // unit header).
400 if (DIEOffset > NextCUOffset)
401 Context.getWarningHandler()(
402 createStringError(errc::invalid_argument,
403 "DWARF compile unit extends beyond its "
404 "bounds cu 0x%8.8" PRIx64 " "
405 "at 0x%8.8" PRIx64 "\n",
406 getOffset(), DIEOffset));
407 }
408
extractDIEsIfNeeded(bool CUDieOnly)409 void DWARFUnit::extractDIEsIfNeeded(bool CUDieOnly) {
410 if (Error e = tryExtractDIEsIfNeeded(CUDieOnly))
411 Context.getRecoverableErrorHandler()(std::move(e));
412 }
413
tryExtractDIEsIfNeeded(bool CUDieOnly)414 Error DWARFUnit::tryExtractDIEsIfNeeded(bool CUDieOnly) {
415 if ((CUDieOnly && !DieArray.empty()) ||
416 DieArray.size() > 1)
417 return Error::success(); // Already parsed.
418
419 bool HasCUDie = !DieArray.empty();
420 extractDIEsToVector(!HasCUDie, !CUDieOnly, DieArray);
421
422 if (DieArray.empty())
423 return Error::success();
424
425 // If CU DIE was just parsed, copy several attribute values from it.
426 if (HasCUDie)
427 return Error::success();
428
429 DWARFDie UnitDie(this, &DieArray[0]);
430 if (Optional<uint64_t> DWOId = toUnsigned(UnitDie.find(DW_AT_GNU_dwo_id)))
431 Header.setDWOId(*DWOId);
432 if (!IsDWO) {
433 assert(AddrOffsetSectionBase == None);
434 assert(RangeSectionBase == 0);
435 assert(LocSectionBase == 0);
436 AddrOffsetSectionBase = toSectionOffset(UnitDie.find(DW_AT_addr_base));
437 if (!AddrOffsetSectionBase)
438 AddrOffsetSectionBase =
439 toSectionOffset(UnitDie.find(DW_AT_GNU_addr_base));
440 RangeSectionBase = toSectionOffset(UnitDie.find(DW_AT_rnglists_base), 0);
441 LocSectionBase = toSectionOffset(UnitDie.find(DW_AT_loclists_base), 0);
442 }
443
444 // In general, in DWARF v5 and beyond we derive the start of the unit's
445 // contribution to the string offsets table from the unit DIE's
446 // DW_AT_str_offsets_base attribute. Split DWARF units do not use this
447 // attribute, so we assume that there is a contribution to the string
448 // offsets table starting at offset 0 of the debug_str_offsets.dwo section.
449 // In both cases we need to determine the format of the contribution,
450 // which may differ from the unit's format.
451 DWARFDataExtractor DA(Context.getDWARFObj(), StringOffsetSection,
452 isLittleEndian, 0);
453 if (IsDWO || getVersion() >= 5) {
454 auto StringOffsetOrError =
455 IsDWO ? determineStringOffsetsTableContributionDWO(DA)
456 : determineStringOffsetsTableContribution(DA);
457 if (!StringOffsetOrError)
458 return createStringError(errc::invalid_argument,
459 "invalid reference to or invalid content in "
460 ".debug_str_offsets[.dwo]: " +
461 toString(StringOffsetOrError.takeError()));
462
463 StringOffsetsTableContribution = *StringOffsetOrError;
464 }
465
466 // DWARF v5 uses the .debug_rnglists and .debug_rnglists.dwo sections to
467 // describe address ranges.
468 if (getVersion() >= 5) {
469 // In case of DWP, the base offset from the index has to be added.
470 if (IsDWO) {
471 uint64_t ContributionBaseOffset = 0;
472 if (auto *IndexEntry = Header.getIndexEntry())
473 if (auto *Contrib = IndexEntry->getContribution(DW_SECT_RNGLISTS))
474 ContributionBaseOffset = Contrib->Offset;
475 setRangesSection(
476 &Context.getDWARFObj().getRnglistsDWOSection(),
477 ContributionBaseOffset +
478 DWARFListTableHeader::getHeaderSize(Header.getFormat()));
479 } else
480 setRangesSection(&Context.getDWARFObj().getRnglistsSection(),
481 toSectionOffset(UnitDie.find(DW_AT_rnglists_base),
482 DWARFListTableHeader::getHeaderSize(
483 Header.getFormat())));
484 }
485
486 if (IsDWO) {
487 // If we are reading a package file, we need to adjust the location list
488 // data based on the index entries.
489 StringRef Data = Header.getVersion() >= 5
490 ? Context.getDWARFObj().getLoclistsDWOSection().Data
491 : Context.getDWARFObj().getLocDWOSection().Data;
492 if (auto *IndexEntry = Header.getIndexEntry())
493 if (const auto *C = IndexEntry->getContribution(
494 Header.getVersion() >= 5 ? DW_SECT_LOCLISTS : DW_SECT_EXT_LOC))
495 Data = Data.substr(C->Offset, C->Length);
496
497 DWARFDataExtractor DWARFData(Data, isLittleEndian, getAddressByteSize());
498 LocTable =
499 std::make_unique<DWARFDebugLoclists>(DWARFData, Header.getVersion());
500 LocSectionBase = DWARFListTableHeader::getHeaderSize(Header.getFormat());
501 } else if (getVersion() >= 5) {
502 LocTable = std::make_unique<DWARFDebugLoclists>(
503 DWARFDataExtractor(Context.getDWARFObj(),
504 Context.getDWARFObj().getLoclistsSection(),
505 isLittleEndian, getAddressByteSize()),
506 getVersion());
507 } else {
508 LocTable = std::make_unique<DWARFDebugLoc>(DWARFDataExtractor(
509 Context.getDWARFObj(), Context.getDWARFObj().getLocSection(),
510 isLittleEndian, getAddressByteSize()));
511 }
512
513 // Don't fall back to DW_AT_GNU_ranges_base: it should be ignored for
514 // skeleton CU DIE, so that DWARF users not aware of it are not broken.
515 return Error::success();
516 }
517
parseDWO()518 bool DWARFUnit::parseDWO() {
519 if (IsDWO)
520 return false;
521 if (DWO.get())
522 return false;
523 DWARFDie UnitDie = getUnitDIE();
524 if (!UnitDie)
525 return false;
526 auto DWOFileName = getVersion() >= 5
527 ? dwarf::toString(UnitDie.find(DW_AT_dwo_name))
528 : dwarf::toString(UnitDie.find(DW_AT_GNU_dwo_name));
529 if (!DWOFileName)
530 return false;
531 auto CompilationDir = dwarf::toString(UnitDie.find(DW_AT_comp_dir));
532 SmallString<16> AbsolutePath;
533 if (sys::path::is_relative(*DWOFileName) && CompilationDir &&
534 *CompilationDir) {
535 sys::path::append(AbsolutePath, *CompilationDir);
536 }
537 sys::path::append(AbsolutePath, *DWOFileName);
538 auto DWOId = getDWOId();
539 if (!DWOId)
540 return false;
541 auto DWOContext = Context.getDWOContext(AbsolutePath);
542 if (!DWOContext)
543 return false;
544
545 DWARFCompileUnit *DWOCU = DWOContext->getDWOCompileUnitForHash(*DWOId);
546 if (!DWOCU)
547 return false;
548 DWO = std::shared_ptr<DWARFCompileUnit>(std::move(DWOContext), DWOCU);
549 // Share .debug_addr and .debug_ranges section with compile unit in .dwo
550 if (AddrOffsetSectionBase)
551 DWO->setAddrOffsetSection(AddrOffsetSection, *AddrOffsetSectionBase);
552 if (getVersion() >= 5) {
553 DWO->setRangesSection(&Context.getDWARFObj().getRnglistsDWOSection(),
554 DWARFListTableHeader::getHeaderSize(getFormat()));
555 } else {
556 auto DWORangesBase = UnitDie.getRangesBaseAttribute();
557 DWO->setRangesSection(RangeSection, DWORangesBase ? *DWORangesBase : 0);
558 }
559
560 return true;
561 }
562
clearDIEs(bool KeepCUDie)563 void DWARFUnit::clearDIEs(bool KeepCUDie) {
564 if (DieArray.size() > (unsigned)KeepCUDie) {
565 DieArray.resize((unsigned)KeepCUDie);
566 DieArray.shrink_to_fit();
567 }
568 }
569
570 Expected<DWARFAddressRangesVector>
findRnglistFromOffset(uint64_t Offset)571 DWARFUnit::findRnglistFromOffset(uint64_t Offset) {
572 if (getVersion() <= 4) {
573 DWARFDebugRangeList RangeList;
574 if (Error E = extractRangeList(Offset, RangeList))
575 return std::move(E);
576 return RangeList.getAbsoluteRanges(getBaseAddress());
577 }
578 DWARFDataExtractor RangesData(Context.getDWARFObj(), *RangeSection,
579 isLittleEndian, Header.getAddressByteSize());
580 DWARFDebugRnglistTable RnglistTable;
581 auto RangeListOrError = RnglistTable.findList(RangesData, Offset);
582 if (RangeListOrError)
583 return RangeListOrError.get().getAbsoluteRanges(getBaseAddress(), *this);
584 return RangeListOrError.takeError();
585 }
586
587 Expected<DWARFAddressRangesVector>
findRnglistFromIndex(uint32_t Index)588 DWARFUnit::findRnglistFromIndex(uint32_t Index) {
589 if (auto Offset = getRnglistOffset(Index))
590 return findRnglistFromOffset(*Offset);
591
592 return createStringError(errc::invalid_argument,
593 "invalid range list table index %d (possibly "
594 "missing the entire range list table)",
595 Index);
596 }
597
collectAddressRanges()598 Expected<DWARFAddressRangesVector> DWARFUnit::collectAddressRanges() {
599 DWARFDie UnitDie = getUnitDIE();
600 if (!UnitDie)
601 return createStringError(errc::invalid_argument, "No unit DIE");
602
603 // First, check if unit DIE describes address ranges for the whole unit.
604 auto CUDIERangesOrError = UnitDie.getAddressRanges();
605 if (!CUDIERangesOrError)
606 return createStringError(errc::invalid_argument,
607 "decoding address ranges: %s",
608 toString(CUDIERangesOrError.takeError()).c_str());
609 return *CUDIERangesOrError;
610 }
611
612 Expected<DWARFLocationExpressionsVector>
findLoclistFromOffset(uint64_t Offset)613 DWARFUnit::findLoclistFromOffset(uint64_t Offset) {
614 DWARFLocationExpressionsVector Result;
615
616 Error InterpretationError = Error::success();
617
618 Error ParseError = getLocationTable().visitAbsoluteLocationList(
619 Offset, getBaseAddress(),
620 [this](uint32_t Index) { return getAddrOffsetSectionItem(Index); },
621 [&](Expected<DWARFLocationExpression> L) {
622 if (L)
623 Result.push_back(std::move(*L));
624 else
625 InterpretationError =
626 joinErrors(L.takeError(), std::move(InterpretationError));
627 return !InterpretationError;
628 });
629
630 if (ParseError || InterpretationError)
631 return joinErrors(std::move(ParseError), std::move(InterpretationError));
632
633 return Result;
634 }
635
updateAddressDieMap(DWARFDie Die)636 void DWARFUnit::updateAddressDieMap(DWARFDie Die) {
637 if (Die.isSubroutineDIE()) {
638 auto DIERangesOrError = Die.getAddressRanges();
639 if (DIERangesOrError) {
640 for (const auto &R : DIERangesOrError.get()) {
641 // Ignore 0-sized ranges.
642 if (R.LowPC == R.HighPC)
643 continue;
644 auto B = AddrDieMap.upper_bound(R.LowPC);
645 if (B != AddrDieMap.begin() && R.LowPC < (--B)->second.first) {
646 // The range is a sub-range of existing ranges, we need to split the
647 // existing range.
648 if (R.HighPC < B->second.first)
649 AddrDieMap[R.HighPC] = B->second;
650 if (R.LowPC > B->first)
651 AddrDieMap[B->first].first = R.LowPC;
652 }
653 AddrDieMap[R.LowPC] = std::make_pair(R.HighPC, Die);
654 }
655 } else
656 llvm::consumeError(DIERangesOrError.takeError());
657 }
658 // Parent DIEs are added to the AddrDieMap prior to the Children DIEs to
659 // simplify the logic to update AddrDieMap. The child's range will always
660 // be equal or smaller than the parent's range. With this assumption, when
661 // adding one range into the map, it will at most split a range into 3
662 // sub-ranges.
663 for (DWARFDie Child = Die.getFirstChild(); Child; Child = Child.getSibling())
664 updateAddressDieMap(Child);
665 }
666
getSubroutineForAddress(uint64_t Address)667 DWARFDie DWARFUnit::getSubroutineForAddress(uint64_t Address) {
668 extractDIEsIfNeeded(false);
669 if (AddrDieMap.empty())
670 updateAddressDieMap(getUnitDIE());
671 auto R = AddrDieMap.upper_bound(Address);
672 if (R == AddrDieMap.begin())
673 return DWARFDie();
674 // upper_bound's previous item contains Address.
675 --R;
676 if (Address >= R->second.first)
677 return DWARFDie();
678 return R->second.second;
679 }
680
681 void
getInlinedChainForAddress(uint64_t Address,SmallVectorImpl<DWARFDie> & InlinedChain)682 DWARFUnit::getInlinedChainForAddress(uint64_t Address,
683 SmallVectorImpl<DWARFDie> &InlinedChain) {
684 assert(InlinedChain.empty());
685 // Try to look for subprogram DIEs in the DWO file.
686 parseDWO();
687 // First, find the subroutine that contains the given address (the leaf
688 // of inlined chain).
689 DWARFDie SubroutineDIE =
690 (DWO ? *DWO : *this).getSubroutineForAddress(Address);
691
692 while (SubroutineDIE) {
693 if (SubroutineDIE.isSubprogramDIE()) {
694 InlinedChain.push_back(SubroutineDIE);
695 return;
696 }
697 if (SubroutineDIE.getTag() == DW_TAG_inlined_subroutine)
698 InlinedChain.push_back(SubroutineDIE);
699 SubroutineDIE = SubroutineDIE.getParent();
700 }
701 }
702
getDWARFUnitIndex(DWARFContext & Context,DWARFSectionKind Kind)703 const DWARFUnitIndex &llvm::getDWARFUnitIndex(DWARFContext &Context,
704 DWARFSectionKind Kind) {
705 if (Kind == DW_SECT_INFO)
706 return Context.getCUIndex();
707 assert(Kind == DW_SECT_EXT_TYPES);
708 return Context.getTUIndex();
709 }
710
getParent(const DWARFDebugInfoEntry * Die)711 DWARFDie DWARFUnit::getParent(const DWARFDebugInfoEntry *Die) {
712 if (!Die)
713 return DWARFDie();
714 const uint32_t Depth = Die->getDepth();
715 // Unit DIEs always have a depth of zero and never have parents.
716 if (Depth == 0)
717 return DWARFDie();
718 // Depth of 1 always means parent is the compile/type unit.
719 if (Depth == 1)
720 return getUnitDIE();
721 // Look for previous DIE with a depth that is one less than the Die's depth.
722 const uint32_t ParentDepth = Depth - 1;
723 for (uint32_t I = getDIEIndex(Die) - 1; I > 0; --I) {
724 if (DieArray[I].getDepth() == ParentDepth)
725 return DWARFDie(this, &DieArray[I]);
726 }
727 return DWARFDie();
728 }
729
getSibling(const DWARFDebugInfoEntry * Die)730 DWARFDie DWARFUnit::getSibling(const DWARFDebugInfoEntry *Die) {
731 if (!Die)
732 return DWARFDie();
733 uint32_t Depth = Die->getDepth();
734 // Unit DIEs always have a depth of zero and never have siblings.
735 if (Depth == 0)
736 return DWARFDie();
737 // NULL DIEs don't have siblings.
738 if (Die->getAbbreviationDeclarationPtr() == nullptr)
739 return DWARFDie();
740
741 // Find the next DIE whose depth is the same as the Die's depth.
742 for (size_t I = getDIEIndex(Die) + 1, EndIdx = DieArray.size(); I < EndIdx;
743 ++I) {
744 if (DieArray[I].getDepth() == Depth)
745 return DWARFDie(this, &DieArray[I]);
746 }
747 return DWARFDie();
748 }
749
getPreviousSibling(const DWARFDebugInfoEntry * Die)750 DWARFDie DWARFUnit::getPreviousSibling(const DWARFDebugInfoEntry *Die) {
751 if (!Die)
752 return DWARFDie();
753 uint32_t Depth = Die->getDepth();
754 // Unit DIEs always have a depth of zero and never have siblings.
755 if (Depth == 0)
756 return DWARFDie();
757
758 // Find the previous DIE whose depth is the same as the Die's depth.
759 for (size_t I = getDIEIndex(Die); I > 0;) {
760 --I;
761 if (DieArray[I].getDepth() == Depth - 1)
762 return DWARFDie();
763 if (DieArray[I].getDepth() == Depth)
764 return DWARFDie(this, &DieArray[I]);
765 }
766 return DWARFDie();
767 }
768
getFirstChild(const DWARFDebugInfoEntry * Die)769 DWARFDie DWARFUnit::getFirstChild(const DWARFDebugInfoEntry *Die) {
770 if (!Die->hasChildren())
771 return DWARFDie();
772
773 // We do not want access out of bounds when parsing corrupted debug data.
774 size_t I = getDIEIndex(Die) + 1;
775 if (I >= DieArray.size())
776 return DWARFDie();
777 return DWARFDie(this, &DieArray[I]);
778 }
779
getLastChild(const DWARFDebugInfoEntry * Die)780 DWARFDie DWARFUnit::getLastChild(const DWARFDebugInfoEntry *Die) {
781 if (!Die->hasChildren())
782 return DWARFDie();
783
784 uint32_t Depth = Die->getDepth();
785 for (size_t I = getDIEIndex(Die) + 1, EndIdx = DieArray.size(); I < EndIdx;
786 ++I) {
787 if (DieArray[I].getDepth() == Depth + 1 &&
788 DieArray[I].getTag() == dwarf::DW_TAG_null)
789 return DWARFDie(this, &DieArray[I]);
790 assert(DieArray[I].getDepth() > Depth && "Not processing children?");
791 }
792 return DWARFDie();
793 }
794
getAbbreviations() const795 const DWARFAbbreviationDeclarationSet *DWARFUnit::getAbbreviations() const {
796 if (!Abbrevs)
797 Abbrevs = Abbrev->getAbbreviationDeclarationSet(Header.getAbbrOffset());
798 return Abbrevs;
799 }
800
getBaseAddress()801 llvm::Optional<object::SectionedAddress> DWARFUnit::getBaseAddress() {
802 if (BaseAddr)
803 return BaseAddr;
804
805 DWARFDie UnitDie = getUnitDIE();
806 Optional<DWARFFormValue> PC = UnitDie.find({DW_AT_low_pc, DW_AT_entry_pc});
807 BaseAddr = toSectionedAddress(PC);
808 return BaseAddr;
809 }
810
811 Expected<StrOffsetsContributionDescriptor>
validateContributionSize(DWARFDataExtractor & DA)812 StrOffsetsContributionDescriptor::validateContributionSize(
813 DWARFDataExtractor &DA) {
814 uint8_t EntrySize = getDwarfOffsetByteSize();
815 // In order to ensure that we don't read a partial record at the end of
816 // the section we validate for a multiple of the entry size.
817 uint64_t ValidationSize = alignTo(Size, EntrySize);
818 // Guard against overflow.
819 if (ValidationSize >= Size)
820 if (DA.isValidOffsetForDataOfSize((uint32_t)Base, ValidationSize))
821 return *this;
822 return createStringError(errc::invalid_argument, "length exceeds section size");
823 }
824
825 // Look for a DWARF64-formatted contribution to the string offsets table
826 // starting at a given offset and record it in a descriptor.
827 static Expected<StrOffsetsContributionDescriptor>
parseDWARF64StringOffsetsTableHeader(DWARFDataExtractor & DA,uint64_t Offset)828 parseDWARF64StringOffsetsTableHeader(DWARFDataExtractor &DA, uint64_t Offset) {
829 if (!DA.isValidOffsetForDataOfSize(Offset, 16))
830 return createStringError(errc::invalid_argument, "section offset exceeds section size");
831
832 if (DA.getU32(&Offset) != dwarf::DW_LENGTH_DWARF64)
833 return createStringError(errc::invalid_argument, "32 bit contribution referenced from a 64 bit unit");
834
835 uint64_t Size = DA.getU64(&Offset);
836 uint8_t Version = DA.getU16(&Offset);
837 (void)DA.getU16(&Offset); // padding
838 // The encoded length includes the 2-byte version field and the 2-byte
839 // padding, so we need to subtract them out when we populate the descriptor.
840 return StrOffsetsContributionDescriptor(Offset, Size - 4, Version, DWARF64);
841 }
842
843 // Look for a DWARF32-formatted contribution to the string offsets table
844 // starting at a given offset and record it in a descriptor.
845 static Expected<StrOffsetsContributionDescriptor>
parseDWARF32StringOffsetsTableHeader(DWARFDataExtractor & DA,uint64_t Offset)846 parseDWARF32StringOffsetsTableHeader(DWARFDataExtractor &DA, uint64_t Offset) {
847 if (!DA.isValidOffsetForDataOfSize(Offset, 8))
848 return createStringError(errc::invalid_argument, "section offset exceeds section size");
849
850 uint32_t ContributionSize = DA.getU32(&Offset);
851 if (ContributionSize >= dwarf::DW_LENGTH_lo_reserved)
852 return createStringError(errc::invalid_argument, "invalid length");
853
854 uint8_t Version = DA.getU16(&Offset);
855 (void)DA.getU16(&Offset); // padding
856 // The encoded length includes the 2-byte version field and the 2-byte
857 // padding, so we need to subtract them out when we populate the descriptor.
858 return StrOffsetsContributionDescriptor(Offset, ContributionSize - 4, Version,
859 DWARF32);
860 }
861
862 static Expected<StrOffsetsContributionDescriptor>
parseDWARFStringOffsetsTableHeader(DWARFDataExtractor & DA,llvm::dwarf::DwarfFormat Format,uint64_t Offset)863 parseDWARFStringOffsetsTableHeader(DWARFDataExtractor &DA,
864 llvm::dwarf::DwarfFormat Format,
865 uint64_t Offset) {
866 StrOffsetsContributionDescriptor Desc;
867 switch (Format) {
868 case dwarf::DwarfFormat::DWARF64: {
869 if (Offset < 16)
870 return createStringError(errc::invalid_argument, "insufficient space for 64 bit header prefix");
871 auto DescOrError = parseDWARF64StringOffsetsTableHeader(DA, Offset - 16);
872 if (!DescOrError)
873 return DescOrError.takeError();
874 Desc = *DescOrError;
875 break;
876 }
877 case dwarf::DwarfFormat::DWARF32: {
878 if (Offset < 8)
879 return createStringError(errc::invalid_argument, "insufficient space for 32 bit header prefix");
880 auto DescOrError = parseDWARF32StringOffsetsTableHeader(DA, Offset - 8);
881 if (!DescOrError)
882 return DescOrError.takeError();
883 Desc = *DescOrError;
884 break;
885 }
886 }
887 return Desc.validateContributionSize(DA);
888 }
889
890 Expected<Optional<StrOffsetsContributionDescriptor>>
determineStringOffsetsTableContribution(DWARFDataExtractor & DA)891 DWARFUnit::determineStringOffsetsTableContribution(DWARFDataExtractor &DA) {
892 assert(!IsDWO);
893 auto OptOffset = toSectionOffset(getUnitDIE().find(DW_AT_str_offsets_base));
894 if (!OptOffset)
895 return None;
896 auto DescOrError =
897 parseDWARFStringOffsetsTableHeader(DA, Header.getFormat(), *OptOffset);
898 if (!DescOrError)
899 return DescOrError.takeError();
900 return *DescOrError;
901 }
902
903 Expected<Optional<StrOffsetsContributionDescriptor>>
determineStringOffsetsTableContributionDWO(DWARFDataExtractor & DA)904 DWARFUnit::determineStringOffsetsTableContributionDWO(DWARFDataExtractor & DA) {
905 assert(IsDWO);
906 uint64_t Offset = 0;
907 auto IndexEntry = Header.getIndexEntry();
908 const auto *C =
909 IndexEntry ? IndexEntry->getContribution(DW_SECT_STR_OFFSETS) : nullptr;
910 if (C)
911 Offset = C->Offset;
912 if (getVersion() >= 5) {
913 if (DA.getData().data() == nullptr)
914 return None;
915 Offset += Header.getFormat() == dwarf::DwarfFormat::DWARF32 ? 8 : 16;
916 // Look for a valid contribution at the given offset.
917 auto DescOrError = parseDWARFStringOffsetsTableHeader(DA, Header.getFormat(), Offset);
918 if (!DescOrError)
919 return DescOrError.takeError();
920 return *DescOrError;
921 }
922 // Prior to DWARF v5, we derive the contribution size from the
923 // index table (in a package file). In a .dwo file it is simply
924 // the length of the string offsets section.
925 StrOffsetsContributionDescriptor Desc;
926 if (C)
927 Desc = StrOffsetsContributionDescriptor(C->Offset, C->Length, 4,
928 Header.getFormat());
929 else if (!IndexEntry && !StringOffsetSection.Data.empty())
930 Desc = StrOffsetsContributionDescriptor(0, StringOffsetSection.Data.size(),
931 4, Header.getFormat());
932 else
933 return None;
934 auto DescOrError = Desc.validateContributionSize(DA);
935 if (!DescOrError)
936 return DescOrError.takeError();
937 return *DescOrError;
938 }
939
getRnglistOffset(uint32_t Index)940 Optional<uint64_t> DWARFUnit::getRnglistOffset(uint32_t Index) {
941 DataExtractor RangesData(RangeSection->Data, isLittleEndian,
942 getAddressByteSize());
943 DWARFDataExtractor RangesDA(Context.getDWARFObj(), *RangeSection,
944 isLittleEndian, 0);
945 if (Optional<uint64_t> Off = llvm::DWARFListTableHeader::getOffsetEntry(
946 RangesData, RangeSectionBase, getFormat(), Index))
947 return *Off + RangeSectionBase;
948 return None;
949 }
950
getLoclistOffset(uint32_t Index)951 Optional<uint64_t> DWARFUnit::getLoclistOffset(uint32_t Index) {
952 if (Optional<uint64_t> Off = llvm::DWARFListTableHeader::getOffsetEntry(
953 LocTable->getData(), LocSectionBase, getFormat(), Index))
954 return *Off + LocSectionBase;
955 return None;
956 }
957