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/BinaryFormat/Dwarf.h"
13#include "llvm/DebugInfo/DWARF/DWARFAbbreviationDeclaration.h"
14#include "llvm/DebugInfo/DWARF/DWARFCompileUnit.h"
15#include "llvm/DebugInfo/DWARF/DWARFContext.h"
16#include "llvm/DebugInfo/DWARF/DWARFDebugAbbrev.h"
17#include "llvm/DebugInfo/DWARF/DWARFDebugInfoEntry.h"
18#include "llvm/DebugInfo/DWARF/DWARFDebugLoc.h"
19#include "llvm/DebugInfo/DWARF/DWARFDebugRangeList.h"
20#include "llvm/DebugInfo/DWARF/DWARFDebugRnglists.h"
21#include "llvm/DebugInfo/DWARF/DWARFDie.h"
22#include "llvm/DebugInfo/DWARF/DWARFFormValue.h"
23#include "llvm/DebugInfo/DWARF/DWARFListTable.h"
24#include "llvm/DebugInfo/DWARF/DWARFObject.h"
25#include "llvm/DebugInfo/DWARF/DWARFSection.h"
26#include "llvm/DebugInfo/DWARF/DWARFTypeUnit.h"
27#include "llvm/DebugInfo/DWARF/LowLevel/DWARFExpression.h"
28#include "llvm/Object/ObjectFile.h"
29#include "llvm/Support/DataExtractor.h"
30#include "llvm/Support/Errc.h"
31#include "llvm/Support/Path.h"
32#include <algorithm>
33#include <cassert>
34#include <cstddef>
35#include <cstdint>
36#include <utility>
37#include <vector>
38
39using namespace llvm;
40using namespace dwarf;
41
42void DWARFUnitVector::addUnitsForSection(DWARFContext &C,
43 const DWARFSection &Section,
44 DWARFSectionKind SectionKind) {
45 const DWARFObject &D = C.getDWARFObj();
46 addUnitsImpl(Context&: C, Obj: D, Section, DA: C.getDebugAbbrev(), RS: &D.getRangesSection(),
47 LocSection: &D.getLocSection(), SS: D.getStrSection(),
48 SOS: D.getStrOffsetsSection(), AOS: &D.getAddrSection(),
49 LS: D.getLineSection(), LE: D.isLittleEndian(), IsDWO: false, Lazy: false,
50 SectionKind);
51}
52
53void DWARFUnitVector::addUnitsForDWOSection(DWARFContext &C,
54 const DWARFSection &DWOSection,
55 DWARFSectionKind SectionKind,
56 bool Lazy) {
57 const DWARFObject &D = C.getDWARFObj();
58 addUnitsImpl(Context&: C, Obj: D, Section: DWOSection, DA: C.getDebugAbbrevDWO(), RS: &D.getRangesDWOSection(),
59 LocSection: &D.getLocDWOSection(), SS: D.getStrDWOSection(),
60 SOS: D.getStrOffsetsDWOSection(), AOS: &D.getAddrSection(),
61 LS: D.getLineDWOSection(), LE: C.isLittleEndian(), IsDWO: true, Lazy,
62 SectionKind);
63}
64
65void DWARFUnitVector::addUnitsImpl(
66 DWARFContext &Context, const DWARFObject &Obj, const DWARFSection &Section,
67 const DWARFDebugAbbrev *DA, const DWARFSection *RS,
68 const DWARFSection *LocSection, StringRef SS, const DWARFSection &SOS,
69 const DWARFSection *AOS, const DWARFSection &LS, bool LE, bool IsDWO,
70 bool Lazy, DWARFSectionKind SectionKind) {
71 DWARFDataExtractor Data(Obj, Section, LE, 0);
72 // Lazy initialization of Parser, now that we have all section info.
73 if (!Parser) {
74 Parser = [=, &Context, &Obj, &Section, &SOS,
75 &LS](uint64_t Offset, DWARFSectionKind SectionKind,
76 const DWARFSection *CurSection,
77 const DWARFUnitIndex::Entry *IndexEntry)
78 -> std::unique_ptr<DWARFUnit> {
79 const DWARFSection &InfoSection = CurSection ? *CurSection : Section;
80 DWARFDataExtractor Data(Obj, InfoSection, LE, 0);
81 if (!Data.isValidOffset(offset: Offset))
82 return nullptr;
83 DWARFUnitHeader Header;
84 if (Error ExtractErr =
85 Header.extract(Context, debug_info: Data, offset_ptr: &Offset, SectionKind)) {
86 Context.getWarningHandler()(std::move(ExtractErr));
87 return nullptr;
88 }
89 if (!IndexEntry && IsDWO) {
90 const DWARFUnitIndex &Index = getDWARFUnitIndex(
91 Context, Kind: Header.isTypeUnit() ? DW_SECT_EXT_TYPES : DW_SECT_INFO);
92 if (Index) {
93 if (Header.isTypeUnit())
94 IndexEntry = Index.getFromHash(Offset: Header.getTypeHash());
95 else if (auto DWOId = Header.getDWOId())
96 IndexEntry = Index.getFromHash(Offset: *DWOId);
97 }
98 if (!IndexEntry)
99 IndexEntry = Index.getFromOffset(Offset: Header.getOffset());
100 }
101 if (IndexEntry) {
102 if (Error ApplicationErr = Header.applyIndexEntry(Entry: IndexEntry)) {
103 Context.getWarningHandler()(std::move(ApplicationErr));
104 return nullptr;
105 }
106 }
107 std::unique_ptr<DWARFUnit> U;
108 if (Header.isTypeUnit())
109 U = std::make_unique<DWARFTypeUnit>(args&: Context, args: InfoSection, args&: Header, args: DA,
110 args: RS, args: LocSection, args: SS, args: SOS, args: AOS, args: LS,
111 args: LE, args: IsDWO, args&: *this);
112 else
113 U = std::make_unique<DWARFCompileUnit>(args&: Context, args: InfoSection, args&: Header,
114 args: DA, args: RS, args: LocSection, args: SS, args: SOS,
115 args: AOS, args: LS, args: LE, args: IsDWO, args&: *this);
116 return U;
117 };
118 }
119 if (Lazy)
120 return;
121 // Find a reasonable insertion point within the vector. We skip over
122 // (a) units from a different section, (b) units from the same section
123 // but with lower offset-within-section. This keeps units in order
124 // within a section, although not necessarily within the object file,
125 // even if we do lazy parsing.
126 auto I = this->begin();
127 uint64_t Offset = 0;
128 while (Data.isValidOffset(offset: Offset)) {
129 if (I != this->end() &&
130 (&(*I)->getInfoSection() != &Section || (*I)->getOffset() == Offset)) {
131 ++I;
132 continue;
133 }
134 auto U = Parser(Offset, SectionKind, &Section, nullptr);
135 // If parsing failed, we're done with this section.
136 if (!U)
137 break;
138 Offset = U->getNextUnitOffset();
139 I = std::next(x: this->insert(I, Elt: std::move(U)));
140 }
141}
142
143DWARFUnit *DWARFUnitVector::addUnit(std::unique_ptr<DWARFUnit> Unit) {
144 auto I = llvm::upper_bound(Range&: *this, Value&: Unit,
145 C: [](const std::unique_ptr<DWARFUnit> &LHS,
146 const std::unique_ptr<DWARFUnit> &RHS) {
147 return LHS->getOffset() < RHS->getOffset();
148 });
149 return this->insert(I, Elt: std::move(Unit))->get();
150}
151
152DWARFUnit *DWARFUnitVector::getUnitForOffset(uint64_t Offset) const {
153 auto end = begin() + getNumInfoUnits();
154 auto *CU =
155 std::upper_bound(first: begin(), last: end, val: Offset,
156 comp: [](uint64_t LHS, const std::unique_ptr<DWARFUnit> &RHS) {
157 return LHS < RHS->getNextUnitOffset();
158 });
159 if (CU != end && (*CU)->getOffset() <= Offset)
160 return CU->get();
161 return nullptr;
162}
163
164DWARFUnit *
165DWARFUnitVector::getUnitForIndexEntry(const DWARFUnitIndex::Entry &E) {
166 const auto *CUOff = E.getContribution(Sec: DW_SECT_INFO);
167 if (!CUOff)
168 return nullptr;
169
170 uint64_t Offset = CUOff->getOffset();
171 auto end = begin() + getNumInfoUnits();
172
173 auto *CU =
174 std::upper_bound(first: begin(), last: end, val: CUOff->getOffset(),
175 comp: [](uint64_t LHS, const std::unique_ptr<DWARFUnit> &RHS) {
176 return LHS < RHS->getNextUnitOffset();
177 });
178 if (CU != end && (*CU)->getOffset() <= Offset)
179 return CU->get();
180
181 if (!Parser)
182 return nullptr;
183
184 auto U = Parser(Offset, DW_SECT_INFO, nullptr, &E);
185 if (!U)
186 return nullptr;
187
188 auto *NewCU = U.get();
189 this->insert(I: CU, Elt: std::move(U));
190 ++NumInfoUnits;
191 return NewCU;
192}
193
194DWARFUnit::DWARFUnit(DWARFContext &DC, const DWARFSection &Section,
195 const DWARFUnitHeader &Header, const DWARFDebugAbbrev *DA,
196 const DWARFSection *RS, const DWARFSection *LocSection,
197 StringRef SS, const DWARFSection &SOS,
198 const DWARFSection *AOS, const DWARFSection &LS, bool LE,
199 bool IsDWO, const DWARFUnitVector &UnitVector)
200 : Context(DC), InfoSection(Section), Header(Header), Abbrev(DA),
201 RangeSection(RS), LineSection(LS), StringSection(SS),
202 StringOffsetSection(SOS), AddrOffsetSection(AOS), IsLittleEndian(LE),
203 IsDWO(IsDWO), UnitVector(UnitVector) {
204 clear();
205}
206
207DWARFUnit::~DWARFUnit() = default;
208
209DWARFDataExtractor DWARFUnit::getDebugInfoExtractor() const {
210 return DWARFDataExtractor(Context.getDWARFObj(), InfoSection, IsLittleEndian,
211 getAddressByteSize());
212}
213
214std::optional<object::SectionedAddress>
215DWARFUnit::getAddrOffsetSectionItem(uint32_t Index) const {
216 if (!AddrOffsetSectionBase) {
217 auto R = Context.info_section_units();
218 // Surprising if a DWO file has more than one skeleton unit in it - this
219 // probably shouldn't be valid, but if a use case is found, here's where to
220 // support it (probably have to linearly search for the matching skeleton CU
221 // here)
222 if (IsDWO && hasSingleElement(C&: R))
223 return (*R.begin())->getAddrOffsetSectionItem(Index);
224
225 return std::nullopt;
226 }
227
228 uint64_t Offset = *AddrOffsetSectionBase + Index * getAddressByteSize();
229 if (AddrOffsetSection->Data.size() < Offset + getAddressByteSize())
230 return std::nullopt;
231 DWARFDataExtractor DA(Context.getDWARFObj(), *AddrOffsetSection,
232 IsLittleEndian, getAddressByteSize());
233 uint64_t Section;
234 uint64_t Address = DA.getRelocatedAddress(Off: &Offset, SecIx: &Section);
235 return {{.Address: Address, .SectionIndex: Section}};
236}
237
238Expected<uint64_t> DWARFUnit::getStringOffsetSectionItem(uint32_t Index) const {
239 if (!StringOffsetsTableContribution)
240 return make_error<StringError>(
241 Args: "DW_FORM_strx used without a valid string offsets table",
242 Args: inconvertibleErrorCode());
243 unsigned ItemSize = getDwarfStringOffsetsByteSize();
244 uint64_t Offset = getStringOffsetsBase() + Index * ItemSize;
245 if (StringOffsetSection.Data.size() < Offset + ItemSize)
246 return make_error<StringError>(Args: "DW_FORM_strx uses index " + Twine(Index) +
247 ", which is too large",
248 Args: inconvertibleErrorCode());
249 DWARFDataExtractor DA(Context.getDWARFObj(), StringOffsetSection,
250 IsLittleEndian, 0);
251 return DA.getRelocatedValue(Size: ItemSize, Off: &Offset);
252}
253
254Error DWARFUnitHeader::extract(DWARFContext &Context,
255 const DWARFDataExtractor &debug_info,
256 uint64_t *offset_ptr,
257 DWARFSectionKind SectionKind) {
258 Offset = *offset_ptr;
259 Error Err = Error::success();
260 IndexEntry = nullptr;
261 std::tie(args&: Length, args&: FormParams.Format) =
262 debug_info.getInitialLength(Off: offset_ptr, Err: &Err);
263 FormParams.Version = debug_info.getU16(offset_ptr, Err: &Err);
264 if (FormParams.Version >= 5) {
265 UnitType = debug_info.getU8(offset_ptr, Err: &Err);
266 FormParams.AddrSize = debug_info.getU8(offset_ptr, Err: &Err);
267 AbbrOffset = debug_info.getRelocatedValue(
268 Size: FormParams.getDwarfOffsetByteSize(), Off: offset_ptr, SectionIndex: nullptr, Err: &Err);
269 } else {
270 AbbrOffset = debug_info.getRelocatedValue(
271 Size: FormParams.getDwarfOffsetByteSize(), Off: offset_ptr, SectionIndex: nullptr, Err: &Err);
272 FormParams.AddrSize = debug_info.getU8(offset_ptr, Err: &Err);
273 // Fake a unit type based on the section type. This isn't perfect,
274 // but distinguishing compile and type units is generally enough.
275 if (SectionKind == DW_SECT_EXT_TYPES)
276 UnitType = DW_UT_type;
277 else
278 UnitType = DW_UT_compile;
279 }
280 if (isTypeUnit()) {
281 TypeHash = debug_info.getU64(offset_ptr, Err: &Err);
282 TypeOffset = debug_info.getUnsigned(
283 offset_ptr, byte_size: FormParams.getDwarfOffsetByteSize(), Err: &Err);
284 } else if (UnitType == DW_UT_split_compile || UnitType == DW_UT_skeleton)
285 DWOId = debug_info.getU64(offset_ptr, Err: &Err);
286
287 if (Err)
288 return joinErrors(
289 E1: createStringError(
290 EC: errc::invalid_argument,
291 Fmt: "DWARF unit at 0x%8.8" PRIx64 " cannot be parsed:", Vals: Offset),
292 E2: std::move(Err));
293
294 // Header fields all parsed, capture the size of this unit header.
295 assert(*offset_ptr - Offset <= 255 && "unexpected header size");
296 Size = uint8_t(*offset_ptr - Offset);
297 uint64_t NextCUOffset = Offset + getUnitLengthFieldByteSize() + getLength();
298
299 if (!debug_info.isValidOffset(offset: getNextUnitOffset() - 1))
300 return createStringError(EC: errc::invalid_argument,
301 Fmt: "DWARF unit from offset 0x%8.8" PRIx64 " incl. "
302 "to offset 0x%8.8" PRIx64 " excl. "
303 "extends past section size 0x%8.8zx",
304 Vals: Offset, Vals: NextCUOffset, Vals: debug_info.size());
305
306 if (!DWARFContext::isSupportedVersion(version: getVersion()))
307 return createStringError(
308 EC: errc::invalid_argument,
309 Fmt: "DWARF unit at offset 0x%8.8" PRIx64 " "
310 "has unsupported version %" PRIu16 ", supported are 2-%u",
311 Vals: Offset, Vals: getVersion(), Vals: DWARFContext::getMaxSupportedVersion());
312
313 // Type offset is unit-relative; should be after the header and before
314 // the end of the current unit.
315 if (isTypeUnit() && TypeOffset < Size)
316 return createStringError(EC: errc::invalid_argument,
317 Fmt: "DWARF type unit at offset "
318 "0x%8.8" PRIx64 " "
319 "has its relocated type_offset 0x%8.8" PRIx64 " "
320 "pointing inside the header",
321 Vals: Offset, Vals: Offset + TypeOffset);
322
323 if (isTypeUnit() && TypeOffset >= getUnitLengthFieldByteSize() + getLength())
324 return createStringError(
325 EC: errc::invalid_argument,
326 Fmt: "DWARF type unit from offset 0x%8.8" PRIx64 " incl. "
327 "to offset 0x%8.8" PRIx64 " excl. has its "
328 "relocated type_offset 0x%8.8" PRIx64 " pointing past the unit end",
329 Vals: Offset, Vals: NextCUOffset, Vals: Offset + TypeOffset);
330
331 if (Error SizeErr = DWARFContext::checkAddressSizeSupported(
332 AddressSize: getAddressByteSize(), EC: errc::invalid_argument,
333 Fmt: "DWARF unit at offset 0x%8.8" PRIx64, Vals: Offset))
334 return SizeErr;
335
336 // Keep track of the highest DWARF version we encounter across all units.
337 Context.setMaxVersionIfGreater(getVersion());
338 return Error::success();
339}
340
341Error DWARFUnitHeader::applyIndexEntry(const DWARFUnitIndex::Entry *Entry) {
342 assert(Entry);
343 assert(!IndexEntry);
344 IndexEntry = Entry;
345 if (AbbrOffset)
346 return createStringError(EC: errc::invalid_argument,
347 Fmt: "DWARF package unit at offset 0x%8.8" PRIx64
348 " has a non-zero abbreviation offset",
349 Vals: Offset);
350
351 auto *UnitContrib = IndexEntry->getContribution();
352 if (!UnitContrib)
353 return createStringError(EC: errc::invalid_argument,
354 Fmt: "DWARF package unit at offset 0x%8.8" PRIx64
355 " has no contribution index",
356 Vals: Offset);
357
358 uint64_t IndexLength = getLength() + getUnitLengthFieldByteSize();
359 if (UnitContrib->getLength() != IndexLength)
360 return createStringError(EC: errc::invalid_argument,
361 Fmt: "DWARF package unit at offset 0x%8.8" PRIx64
362 " has an inconsistent index (expected: %" PRIu64
363 ", actual: %" PRIu64 ")",
364 Vals: Offset, Vals: UnitContrib->getLength(), Vals: IndexLength);
365
366 auto *AbbrEntry = IndexEntry->getContribution(Sec: DW_SECT_ABBREV);
367 if (!AbbrEntry)
368 return createStringError(EC: errc::invalid_argument,
369 Fmt: "DWARF package unit at offset 0x%8.8" PRIx64
370 " missing abbreviation column",
371 Vals: Offset);
372
373 AbbrOffset = AbbrEntry->getOffset();
374 return Error::success();
375}
376
377Error DWARFUnit::extractRangeList(uint64_t RangeListOffset,
378 DWARFDebugRangeList &RangeList) const {
379 // Require that compile unit is extracted.
380 assert(!DieArray.empty());
381 DWARFDataExtractor RangesData(Context.getDWARFObj(), *RangeSection,
382 IsLittleEndian, getAddressByteSize());
383 uint64_t ActualRangeListOffset = RangeSectionBase + RangeListOffset;
384 return RangeList.extract(data: RangesData, offset_ptr: &ActualRangeListOffset);
385}
386
387void DWARFUnit::clear() {
388 Abbrevs = nullptr;
389 BaseAddr.reset();
390 RangeSectionBase = 0;
391 LocSectionBase = 0;
392 AddrOffsetSectionBase = std::nullopt;
393 SU = nullptr;
394 clearDIEs(KeepCUDie: false);
395 AddrDieMap.clear();
396 if (DWO)
397 DWO->clear();
398 DWO.reset();
399}
400
401const char *DWARFUnit::getCompilationDir() {
402 return dwarf::toString(V: getUnitDIE().find(Attr: DW_AT_comp_dir), Default: nullptr);
403}
404
405void DWARFUnit::extractDIEsToVector(
406 bool AppendCUDie, bool AppendNonCUDies,
407 std::vector<DWARFDebugInfoEntry> &Dies) const {
408 if (!AppendCUDie && !AppendNonCUDies)
409 return;
410
411 // Set the offset to that of the first DIE and calculate the start of the
412 // next compilation unit header.
413 uint64_t DIEOffset = getOffset() + getHeaderSize();
414 uint64_t NextCUOffset = getNextUnitOffset();
415 DWARFDebugInfoEntry DIE;
416 DWARFDataExtractor DebugInfoData = getDebugInfoExtractor();
417 // The end offset has been already checked by DWARFUnitHeader::extract.
418 assert(DebugInfoData.isValidOffset(NextCUOffset - 1));
419 std::vector<uint32_t> Parents;
420 std::vector<uint32_t> PrevSiblings;
421 bool IsCUDie = true;
422
423 assert(
424 ((AppendCUDie && Dies.empty()) || (!AppendCUDie && Dies.size() == 1)) &&
425 "Dies array is not empty");
426
427 // Fill Parents and Siblings stacks with initial value.
428 Parents.push_back(UINT32_MAX);
429 if (!AppendCUDie)
430 Parents.push_back(x: 0);
431 PrevSiblings.push_back(x: 0);
432
433 // Start to extract dies.
434 do {
435 assert(Parents.size() > 0 && "Empty parents stack");
436 assert((Parents.back() == UINT32_MAX || Parents.back() <= Dies.size()) &&
437 "Wrong parent index");
438
439 // Extract die. Stop if any error occurred.
440 if (!DIE.extractFast(U: *this, OffsetPtr: &DIEOffset, DebugInfoData, UEndOffset: NextCUOffset,
441 ParentIdx: Parents.back()))
442 break;
443
444 // If previous sibling is remembered then update it`s SiblingIdx field.
445 if (PrevSiblings.back() > 0) {
446 assert(PrevSiblings.back() < Dies.size() &&
447 "Previous sibling index is out of Dies boundaries");
448 Dies[PrevSiblings.back()].setSiblingIdx(Dies.size());
449 }
450
451 // Store die into the Dies vector.
452 if (IsCUDie) {
453 if (AppendCUDie)
454 Dies.push_back(x: DIE);
455 if (!AppendNonCUDies)
456 break;
457 // The average bytes per DIE entry has been seen to be
458 // around 14-20 so let's pre-reserve the needed memory for
459 // our DIE entries accordingly.
460 Dies.reserve(n: Dies.size() + getDebugInfoSize() / 14);
461 } else {
462 // Remember last previous sibling.
463 PrevSiblings.back() = Dies.size();
464
465 Dies.push_back(x: DIE);
466 }
467
468 // Check for new children scope.
469 if (const DWARFAbbreviationDeclaration *AbbrDecl =
470 DIE.getAbbreviationDeclarationPtr()) {
471 if (AbbrDecl->hasChildren()) {
472 if (AppendCUDie || !IsCUDie) {
473 assert(Dies.size() > 0 && "Dies does not contain any die");
474 Parents.push_back(x: Dies.size() - 1);
475 PrevSiblings.push_back(x: 0);
476 }
477 } else if (IsCUDie)
478 // Stop if we have single compile unit die w/o children.
479 break;
480 } else {
481 // NULL DIE: finishes current children scope.
482 Parents.pop_back();
483 PrevSiblings.pop_back();
484 }
485
486 if (IsCUDie)
487 IsCUDie = false;
488
489 // Stop when compile unit die is removed from the parents stack.
490 } while (Parents.size() > 1);
491}
492
493void DWARFUnit::extractDIEsIfNeeded(bool CUDieOnly) {
494 if (Error e = tryExtractDIEsIfNeeded(CUDieOnly))
495 Context.getRecoverableErrorHandler()(std::move(e));
496}
497
498Error DWARFUnit::tryExtractDIEsIfNeeded(bool CUDieOnly) {
499 if ((CUDieOnly && !DieArray.empty()) || DieArray.size() > 1)
500 return Error::success(); // Already parsed.
501
502 bool HasCUDie = !DieArray.empty();
503 extractDIEsToVector(AppendCUDie: !HasCUDie, AppendNonCUDies: !CUDieOnly, Dies&: DieArray);
504
505 if (DieArray.empty())
506 return Error::success();
507
508 // If CU DIE was just parsed, copy several attribute values from it.
509 if (HasCUDie)
510 return Error::success();
511
512 DWARFDie UnitDie(this, &DieArray[0]);
513 if (std::optional<uint64_t> DWOId =
514 toUnsigned(V: UnitDie.find(Attr: DW_AT_GNU_dwo_id)))
515 Header.setDWOId(*DWOId);
516 if (!IsDWO) {
517 assert(AddrOffsetSectionBase == std::nullopt);
518 assert(RangeSectionBase == 0);
519 assert(LocSectionBase == 0);
520 AddrOffsetSectionBase = toSectionOffset(V: UnitDie.find(Attr: DW_AT_addr_base));
521 if (!AddrOffsetSectionBase)
522 AddrOffsetSectionBase =
523 toSectionOffset(V: UnitDie.find(Attr: DW_AT_GNU_addr_base));
524 RangeSectionBase = toSectionOffset(V: UnitDie.find(Attr: DW_AT_rnglists_base), Default: 0);
525 LocSectionBase = toSectionOffset(V: UnitDie.find(Attr: DW_AT_loclists_base), Default: 0);
526 }
527
528 // In general, in DWARF v5 and beyond we derive the start of the unit's
529 // contribution to the string offsets table from the unit DIE's
530 // DW_AT_str_offsets_base attribute. Split DWARF units do not use this
531 // attribute, so we assume that there is a contribution to the string
532 // offsets table starting at offset 0 of the debug_str_offsets.dwo section.
533 // In both cases we need to determine the format of the contribution,
534 // which may differ from the unit's format.
535 DWARFDataExtractor DA(Context.getDWARFObj(), StringOffsetSection,
536 IsLittleEndian, 0);
537 if (IsDWO || getVersion() >= 5) {
538 auto StringOffsetOrError =
539 IsDWO ? determineStringOffsetsTableContributionDWO(DA)
540 : determineStringOffsetsTableContribution(DA);
541 if (!StringOffsetOrError)
542 return createStringError(EC: errc::invalid_argument,
543 S: "invalid reference to or invalid content in "
544 ".debug_str_offsets[.dwo]: " +
545 toString(E: StringOffsetOrError.takeError()));
546
547 StringOffsetsTableContribution = *StringOffsetOrError;
548 }
549
550 // DWARF v5 uses the .debug_rnglists and .debug_rnglists.dwo sections to
551 // describe address ranges.
552 if (getVersion() >= 5) {
553 // In case of DWP, the base offset from the index has to be added.
554 if (IsDWO) {
555 uint64_t ContributionBaseOffset = 0;
556 if (auto *IndexEntry = Header.getIndexEntry())
557 if (auto *Contrib = IndexEntry->getContribution(Sec: DW_SECT_RNGLISTS))
558 ContributionBaseOffset = Contrib->getOffset();
559 setRangesSection(
560 RS: &Context.getDWARFObj().getRnglistsDWOSection(),
561 Base: ContributionBaseOffset +
562 DWARFListTableHeader::getHeaderSize(Format: Header.getFormat()));
563 } else
564 setRangesSection(RS: &Context.getDWARFObj().getRnglistsSection(),
565 Base: toSectionOffset(V: UnitDie.find(Attr: DW_AT_rnglists_base),
566 Default: DWARFListTableHeader::getHeaderSize(
567 Format: Header.getFormat())));
568 }
569
570 if (IsDWO) {
571 // If we are reading a package file, we need to adjust the location list
572 // data based on the index entries.
573 StringRef Data = Header.getVersion() >= 5
574 ? Context.getDWARFObj().getLoclistsDWOSection().Data
575 : Context.getDWARFObj().getLocDWOSection().Data;
576 if (auto *IndexEntry = Header.getIndexEntry())
577 if (const auto *C = IndexEntry->getContribution(
578 Sec: Header.getVersion() >= 5 ? DW_SECT_LOCLISTS : DW_SECT_EXT_LOC))
579 Data = Data.substr(Start: C->getOffset(), N: C->getLength());
580
581 DWARFDataExtractor DWARFData(Data, IsLittleEndian, getAddressByteSize());
582 LocTable =
583 std::make_unique<DWARFDebugLoclists>(args&: DWARFData, args: Header.getVersion());
584 LocSectionBase = DWARFListTableHeader::getHeaderSize(Format: Header.getFormat());
585 } else if (getVersion() >= 5) {
586 LocTable = std::make_unique<DWARFDebugLoclists>(
587 args: DWARFDataExtractor(Context.getDWARFObj(),
588 Context.getDWARFObj().getLoclistsSection(),
589 IsLittleEndian, getAddressByteSize()),
590 args: getVersion());
591 } else {
592 LocTable = std::make_unique<DWARFDebugLoc>(args: DWARFDataExtractor(
593 Context.getDWARFObj(), Context.getDWARFObj().getLocSection(),
594 IsLittleEndian, getAddressByteSize()));
595 }
596
597 // Don't fall back to DW_AT_GNU_ranges_base: it should be ignored for
598 // skeleton CU DIE, so that DWARF users not aware of it are not broken.
599 return Error::success();
600}
601
602bool DWARFUnit::parseDWO(StringRef DWOAlternativeLocation) {
603 if (IsDWO)
604 return false;
605 if (DWO)
606 return false;
607 DWARFDie UnitDie = getUnitDIE();
608 if (!UnitDie)
609 return false;
610 auto DWOFileName = getVersion() >= 5
611 ? dwarf::toString(V: UnitDie.find(Attr: DW_AT_dwo_name))
612 : dwarf::toString(V: UnitDie.find(Attr: DW_AT_GNU_dwo_name));
613 if (!DWOFileName)
614 return false;
615 auto CompilationDir = dwarf::toString(V: UnitDie.find(Attr: DW_AT_comp_dir));
616 SmallString<16> AbsolutePath;
617 if (sys::path::is_relative(path: *DWOFileName) && CompilationDir &&
618 *CompilationDir) {
619 sys::path::append(path&: AbsolutePath, a: *CompilationDir);
620 }
621 sys::path::append(path&: AbsolutePath, a: *DWOFileName);
622 auto DWOId = getDWOId();
623 if (!DWOId)
624 return false;
625 auto DWOContext = Context.getDWOContext(AbsolutePath);
626 if (!DWOContext) {
627 // Use the alternative location to get the DWARF context for the DWO object.
628 if (DWOAlternativeLocation.empty())
629 return false;
630 // If the alternative context does not correspond to the original DWO object
631 // (different hashes), the below 'getDWOCompileUnitForHash' call will catch
632 // the issue, with a returned null context.
633 DWOContext = Context.getDWOContext(AbsolutePath: DWOAlternativeLocation);
634 if (!DWOContext)
635 return false;
636 }
637
638 DWARFCompileUnit *DWOCU = DWOContext->getDWOCompileUnitForHash(Hash: *DWOId);
639 if (!DWOCU)
640 return false;
641 DWO = std::shared_ptr<DWARFCompileUnit>(std::move(DWOContext), DWOCU);
642 DWO->setSkeletonUnit(this);
643 // Share .debug_addr and .debug_ranges section with compile unit in .dwo
644 if (AddrOffsetSectionBase)
645 DWO->setAddrOffsetSection(AOS: AddrOffsetSection, Base: *AddrOffsetSectionBase);
646 if (getVersion() == 4) {
647 auto DWORangesBase = UnitDie.getRangesBaseAttribute();
648 DWO->setRangesSection(RS: RangeSection, Base: DWORangesBase.value_or(u: 0));
649 }
650
651 return true;
652}
653
654void DWARFUnit::clearDIEs(bool KeepCUDie) {
655 // Do not use resize() + shrink_to_fit() to free memory occupied by dies.
656 // shrink_to_fit() is a *non-binding* request to reduce capacity() to size().
657 // It depends on the implementation whether the request is fulfilled.
658 // Create a new vector with a small capacity and assign it to the DieArray to
659 // have previous contents freed.
660 DieArray = (KeepCUDie && !DieArray.empty())
661 ? std::vector<DWARFDebugInfoEntry>({DieArray[0]})
662 : std::vector<DWARFDebugInfoEntry>();
663}
664
665Expected<DWARFAddressRangesVector>
666DWARFUnit::findRnglistFromOffset(uint64_t Offset) {
667 if (getVersion() <= 4) {
668 DWARFDebugRangeList RangeList;
669 if (Error E = extractRangeList(RangeListOffset: Offset, RangeList))
670 return std::move(E);
671 return RangeList.getAbsoluteRanges(BaseAddr: getBaseAddress());
672 }
673 DWARFDataExtractor RangesData(Context.getDWARFObj(), *RangeSection,
674 IsLittleEndian, Header.getAddressByteSize());
675 DWARFDebugRnglistTable RnglistTable;
676 auto RangeListOrError = RnglistTable.findList(Data: RangesData, Offset);
677 if (RangeListOrError)
678 return RangeListOrError.get().getAbsoluteRanges(BaseAddr: getBaseAddress(), U&: *this);
679 return RangeListOrError.takeError();
680}
681
682Expected<DWARFAddressRangesVector>
683DWARFUnit::findRnglistFromIndex(uint32_t Index) {
684 if (auto Offset = getRnglistOffset(Index))
685 return findRnglistFromOffset(Offset: *Offset);
686
687 return createStringError(EC: errc::invalid_argument,
688 Fmt: "invalid range list table index %d (possibly "
689 "missing the entire range list table)",
690 Vals: Index);
691}
692
693Expected<DWARFAddressRangesVector> DWARFUnit::collectAddressRanges() {
694 DWARFDie UnitDie = getUnitDIE();
695 if (!UnitDie)
696 return createStringError(EC: errc::invalid_argument, S: "No unit DIE");
697
698 // First, check if unit DIE describes address ranges for the whole unit.
699 auto CUDIERangesOrError = UnitDie.getAddressRanges();
700 if (!CUDIERangesOrError)
701 return createStringError(EC: errc::invalid_argument,
702 Fmt: "decoding address ranges: %s",
703 Vals: toString(E: CUDIERangesOrError.takeError()).c_str());
704 return *CUDIERangesOrError;
705}
706
707Expected<DWARFLocationExpressionsVector>
708DWARFUnit::findLoclistFromOffset(uint64_t Offset) {
709 DWARFLocationExpressionsVector Result;
710
711 Error InterpretationError = Error::success();
712
713 Error ParseError = getLocationTable().visitAbsoluteLocationList(
714 Offset, BaseAddr: getBaseAddress(),
715 LookupAddr: [this](uint32_t Index) { return getAddrOffsetSectionItem(Index); },
716 Callback: [&](Expected<DWARFLocationExpression> L) {
717 if (L)
718 Result.push_back(x: std::move(*L));
719 else
720 InterpretationError =
721 joinErrors(E1: L.takeError(), E2: std::move(InterpretationError));
722 return !InterpretationError;
723 });
724
725 if (ParseError || InterpretationError)
726 return joinErrors(E1: std::move(ParseError), E2: std::move(InterpretationError));
727
728 return Result;
729}
730
731void DWARFUnit::updateAddressDieMap(DWARFDie Die) {
732 if (Die.isSubroutineDIE()) {
733 auto DIERangesOrError = Die.getAddressRanges();
734 if (DIERangesOrError) {
735 for (const auto &R : DIERangesOrError.get()) {
736 // Ignore 0-sized ranges.
737 if (R.LowPC == R.HighPC)
738 continue;
739 auto B = AddrDieMap.upper_bound(x: R.LowPC);
740 if (B != AddrDieMap.begin() && R.LowPC < (--B)->second.first) {
741 // The range is a sub-range of existing ranges, we need to split the
742 // existing range.
743 if (R.HighPC < B->second.first)
744 AddrDieMap[R.HighPC] = B->second;
745 if (R.LowPC > B->first)
746 AddrDieMap[B->first].first = R.LowPC;
747 }
748 AddrDieMap[R.LowPC] = std::make_pair(x: R.HighPC, y&: Die);
749 }
750 } else
751 llvm::consumeError(Err: DIERangesOrError.takeError());
752 }
753 // Parent DIEs are added to the AddrDieMap prior to the Children DIEs to
754 // simplify the logic to update AddrDieMap. The child's range will always
755 // be equal or smaller than the parent's range. With this assumption, when
756 // adding one range into the map, it will at most split a range into 3
757 // sub-ranges.
758 for (DWARFDie Child = Die.getFirstChild(); Child; Child = Child.getSibling())
759 updateAddressDieMap(Die: Child);
760}
761
762DWARFDie DWARFUnit::getSubroutineForAddress(uint64_t Address) {
763 extractDIEsIfNeeded(CUDieOnly: false);
764 if (AddrDieMap.empty())
765 updateAddressDieMap(Die: getUnitDIE());
766 auto R = AddrDieMap.upper_bound(x: Address);
767 if (R == AddrDieMap.begin())
768 return DWARFDie();
769 // upper_bound's previous item contains Address.
770 --R;
771 if (Address >= R->second.first)
772 return DWARFDie();
773 return R->second.second;
774}
775
776void DWARFUnit::updateVariableDieMap(DWARFDie Die) {
777 for (DWARFDie Child : Die) {
778 if (isType(T: Child.getTag()))
779 continue;
780 updateVariableDieMap(Die: Child);
781 }
782
783 if (Die.getTag() != DW_TAG_variable)
784 return;
785
786 Expected<DWARFLocationExpressionsVector> Locations =
787 Die.getLocations(Attr: DW_AT_location);
788 if (!Locations) {
789 // Missing DW_AT_location is fine here.
790 consumeError(Err: Locations.takeError());
791 return;
792 }
793
794 uint64_t Address = UINT64_MAX;
795
796 for (const DWARFLocationExpression &Location : *Locations) {
797 uint8_t AddressSize = getAddressByteSize();
798 DataExtractor Data(Location.Expr, isLittleEndian(), AddressSize);
799 DWARFExpression Expr(Data, AddressSize);
800 auto It = Expr.begin();
801 if (It == Expr.end())
802 continue;
803
804 // Match exactly the main sequence used to describe global variables:
805 // `DW_OP_addr[x] [+ DW_OP_plus_uconst]`. Currently, this is the sequence
806 // that LLVM produces for DILocalVariables and DIGlobalVariables. If, in
807 // future, the DWARF producer (`DwarfCompileUnit::addLocationAttribute()` is
808 // a good starting point) is extended to use further expressions, this code
809 // needs to be updated.
810 uint64_t LocationAddr;
811 if (It->getCode() == dwarf::DW_OP_addr) {
812 LocationAddr = It->getRawOperand(Idx: 0);
813 } else if (It->getCode() == dwarf::DW_OP_addrx) {
814 uint64_t DebugAddrOffset = It->getRawOperand(Idx: 0);
815 if (auto Pointer = getAddrOffsetSectionItem(Index: DebugAddrOffset)) {
816 LocationAddr = Pointer->Address;
817 }
818 } else {
819 continue;
820 }
821
822 // Read the optional 2nd operand, a DW_OP_plus_uconst.
823 if (++It != Expr.end()) {
824 if (It->getCode() != dwarf::DW_OP_plus_uconst)
825 continue;
826
827 LocationAddr += It->getRawOperand(Idx: 0);
828
829 // Probe for a 3rd operand, if it exists, bail.
830 if (++It != Expr.end())
831 continue;
832 }
833
834 Address = LocationAddr;
835 break;
836 }
837
838 // Get the size of the global variable. If all else fails (i.e. the global has
839 // no type), then we use a size of one to still allow symbolization of the
840 // exact address.
841 uint64_t GVSize = 1;
842 if (Die.getAttributeValueAsReferencedDie(Attr: DW_AT_type))
843 if (std::optional<uint64_t> Size = Die.getTypeSize(PointerSize: getAddressByteSize()))
844 GVSize = *Size;
845
846 if (Address != UINT64_MAX)
847 VariableDieMap[Address] = {Address + GVSize, Die};
848}
849
850DWARFDie DWARFUnit::getVariableForAddress(uint64_t Address) {
851 extractDIEsIfNeeded(CUDieOnly: false);
852
853 auto RootDie = getUnitDIE();
854
855 auto RootLookup = RootsParsedForVariables.insert(V: RootDie.getOffset());
856 if (RootLookup.second)
857 updateVariableDieMap(Die: RootDie);
858
859 auto R = VariableDieMap.upper_bound(x: Address);
860 if (R == VariableDieMap.begin())
861 return DWARFDie();
862
863 // upper_bound's previous item contains Address.
864 --R;
865 if (Address >= R->second.first)
866 return DWARFDie();
867 return R->second.second;
868}
869
870void
871DWARFUnit::getInlinedChainForAddress(uint64_t Address,
872 SmallVectorImpl<DWARFDie> &InlinedChain) {
873 assert(InlinedChain.empty());
874 // Try to look for subprogram DIEs in the DWO file.
875 parseDWO();
876 // First, find the subroutine that contains the given address (the leaf
877 // of inlined chain).
878 DWARFDie SubroutineDIE =
879 (DWO ? *DWO : *this).getSubroutineForAddress(Address);
880
881 while (SubroutineDIE) {
882 if (SubroutineDIE.isSubprogramDIE()) {
883 InlinedChain.push_back(Elt: SubroutineDIE);
884 return;
885 }
886 if (SubroutineDIE.getTag() == DW_TAG_inlined_subroutine)
887 InlinedChain.push_back(Elt: SubroutineDIE);
888 SubroutineDIE = SubroutineDIE.getParent();
889 }
890}
891
892const DWARFUnitIndex &llvm::getDWARFUnitIndex(DWARFContext &Context,
893 DWARFSectionKind Kind) {
894 if (Kind == DW_SECT_INFO)
895 return Context.getCUIndex();
896 assert(Kind == DW_SECT_EXT_TYPES);
897 return Context.getTUIndex();
898}
899
900DWARFDie DWARFUnit::getParent(const DWARFDebugInfoEntry *Die) {
901 if (const DWARFDebugInfoEntry *Entry = getParentEntry(Die))
902 return DWARFDie(this, Entry);
903
904 return DWARFDie();
905}
906
907const DWARFDebugInfoEntry *
908DWARFUnit::getParentEntry(const DWARFDebugInfoEntry *Die) const {
909 if (!Die)
910 return nullptr;
911 assert(Die >= DieArray.data() && Die < DieArray.data() + DieArray.size());
912
913 if (std::optional<uint32_t> ParentIdx = Die->getParentIdx()) {
914 assert(*ParentIdx < DieArray.size() &&
915 "ParentIdx is out of DieArray boundaries");
916 return getDebugInfoEntry(Index: *ParentIdx);
917 }
918
919 return nullptr;
920}
921
922DWARFDie DWARFUnit::getSibling(const DWARFDebugInfoEntry *Die) {
923 if (const DWARFDebugInfoEntry *Sibling = getSiblingEntry(Die))
924 return DWARFDie(this, Sibling);
925
926 return DWARFDie();
927}
928
929const DWARFDebugInfoEntry *
930DWARFUnit::getSiblingEntry(const DWARFDebugInfoEntry *Die) const {
931 if (!Die)
932 return nullptr;
933 assert(Die >= DieArray.data() && Die < DieArray.data() + DieArray.size());
934
935 if (std::optional<uint32_t> SiblingIdx = Die->getSiblingIdx()) {
936 assert(*SiblingIdx < DieArray.size() &&
937 "SiblingIdx is out of DieArray boundaries");
938 return &DieArray[*SiblingIdx];
939 }
940
941 return nullptr;
942}
943
944DWARFDie DWARFUnit::getPreviousSibling(const DWARFDebugInfoEntry *Die) {
945 if (const DWARFDebugInfoEntry *Sibling = getPreviousSiblingEntry(Die))
946 return DWARFDie(this, Sibling);
947
948 return DWARFDie();
949}
950
951const DWARFDebugInfoEntry *
952DWARFUnit::getPreviousSiblingEntry(const DWARFDebugInfoEntry *Die) const {
953 if (!Die)
954 return nullptr;
955 assert(Die >= DieArray.data() && Die < DieArray.data() + DieArray.size());
956
957 std::optional<uint32_t> ParentIdx = Die->getParentIdx();
958 if (!ParentIdx)
959 // Die is a root die, there is no previous sibling.
960 return nullptr;
961
962 assert(*ParentIdx < DieArray.size() &&
963 "ParentIdx is out of DieArray boundaries");
964 assert(getDIEIndex(Die) > 0 && "Die is a root die");
965
966 uint32_t PrevDieIdx = getDIEIndex(Die) - 1;
967 if (PrevDieIdx == *ParentIdx)
968 // Immediately previous node is parent, there is no previous sibling.
969 return nullptr;
970
971 while (DieArray[PrevDieIdx].getParentIdx() != *ParentIdx) {
972 PrevDieIdx = *DieArray[PrevDieIdx].getParentIdx();
973
974 assert(PrevDieIdx < DieArray.size() &&
975 "PrevDieIdx is out of DieArray boundaries");
976 assert(PrevDieIdx >= *ParentIdx &&
977 "PrevDieIdx is not a child of parent of Die");
978 }
979
980 return &DieArray[PrevDieIdx];
981}
982
983DWARFDie DWARFUnit::getFirstChild(const DWARFDebugInfoEntry *Die) {
984 if (const DWARFDebugInfoEntry *Child = getFirstChildEntry(Die))
985 return DWARFDie(this, Child);
986
987 return DWARFDie();
988}
989
990const DWARFDebugInfoEntry *
991DWARFUnit::getFirstChildEntry(const DWARFDebugInfoEntry *Die) const {
992 if (!Die)
993 return nullptr;
994 assert(Die >= DieArray.data() && Die < DieArray.data() + DieArray.size());
995
996 if (!Die->hasChildren())
997 return nullptr;
998
999 // TODO: Instead of checking here for invalid die we might reject
1000 // invalid dies at parsing stage(DWARFUnit::extractDIEsToVector).
1001 // We do not want access out of bounds when parsing corrupted debug data.
1002 size_t I = getDIEIndex(Die) + 1;
1003 if (I >= DieArray.size())
1004 return nullptr;
1005 return &DieArray[I];
1006}
1007
1008DWARFDie DWARFUnit::getLastChild(const DWARFDebugInfoEntry *Die) {
1009 if (const DWARFDebugInfoEntry *Child = getLastChildEntry(Die))
1010 return DWARFDie(this, Child);
1011
1012 return DWARFDie();
1013}
1014
1015const DWARFDebugInfoEntry *
1016DWARFUnit::getLastChildEntry(const DWARFDebugInfoEntry *Die) const {
1017 if (!Die)
1018 return nullptr;
1019 assert(Die >= DieArray.data() && Die < DieArray.data() + DieArray.size());
1020
1021 if (!Die->hasChildren())
1022 return nullptr;
1023
1024 if (std::optional<uint32_t> SiblingIdx = Die->getSiblingIdx()) {
1025 assert(*SiblingIdx < DieArray.size() &&
1026 "SiblingIdx is out of DieArray boundaries");
1027 assert(DieArray[*SiblingIdx - 1].getTag() == dwarf::DW_TAG_null &&
1028 "Bad end of children marker");
1029 return &DieArray[*SiblingIdx - 1];
1030 }
1031
1032 // If SiblingIdx is set for non-root dies we could be sure that DWARF is
1033 // correct and "end of children marker" must be found. For root die we do not
1034 // have such a guarantee(parsing root die might be stopped if "end of children
1035 // marker" is missing, SiblingIdx is always zero for root die). That is why we
1036 // do not use assertion for checking for "end of children marker" for root
1037 // die.
1038
1039 // TODO: Instead of checking here for invalid die we might reject
1040 // invalid dies at parsing stage(DWARFUnit::extractDIEsToVector).
1041 if (getDIEIndex(Die) == 0 && DieArray.size() > 1 &&
1042 DieArray.back().getTag() == dwarf::DW_TAG_null) {
1043 // For the unit die we might take last item from DieArray.
1044 assert(getDIEIndex(Die) ==
1045 getDIEIndex(const_cast<DWARFUnit *>(this)->getUnitDIE()) &&
1046 "Bad unit die");
1047 return &DieArray.back();
1048 }
1049
1050 return nullptr;
1051}
1052
1053const DWARFAbbreviationDeclarationSet *DWARFUnit::getAbbreviations() const {
1054 if (!Abbrevs) {
1055 Expected<const DWARFAbbreviationDeclarationSet *> AbbrevsOrError =
1056 Abbrev->getAbbreviationDeclarationSet(CUAbbrOffset: getAbbreviationsOffset());
1057 if (!AbbrevsOrError) {
1058 // FIXME: We should propagate this error upwards.
1059 consumeError(Err: AbbrevsOrError.takeError());
1060 return nullptr;
1061 }
1062 Abbrevs = *AbbrevsOrError;
1063 }
1064 return Abbrevs;
1065}
1066
1067std::optional<object::SectionedAddress> DWARFUnit::getBaseAddress() {
1068 if (BaseAddr)
1069 return BaseAddr;
1070
1071 DWARFDie UnitDie = (SU ? SU : this)->getUnitDIE();
1072 std::optional<DWARFFormValue> PC =
1073 UnitDie.find(Attrs: {DW_AT_low_pc, DW_AT_entry_pc});
1074 BaseAddr = toSectionedAddress(V: PC);
1075 return BaseAddr;
1076}
1077
1078Expected<StrOffsetsContributionDescriptor>
1079StrOffsetsContributionDescriptor::validateContributionSize(
1080 DWARFDataExtractor &DA) {
1081 uint8_t EntrySize = getDwarfOffsetByteSize();
1082 // In order to ensure that we don't read a partial record at the end of
1083 // the section we validate for a multiple of the entry size.
1084 uint64_t ValidationSize = alignTo(Value: Size, Align: EntrySize);
1085 // Guard against overflow.
1086 if (ValidationSize >= Size)
1087 if (DA.isValidOffsetForDataOfSize(offset: (uint32_t)Base, length: ValidationSize))
1088 return *this;
1089 return createStringError(EC: errc::invalid_argument, S: "length exceeds section size");
1090}
1091
1092// Look for a DWARF64-formatted contribution to the string offsets table
1093// starting at a given offset and record it in a descriptor.
1094static Expected<StrOffsetsContributionDescriptor>
1095parseDWARF64StringOffsetsTableHeader(DWARFDataExtractor &DA, uint64_t Offset) {
1096 if (!DA.isValidOffsetForDataOfSize(offset: Offset, length: 16))
1097 return createStringError(EC: errc::invalid_argument, S: "section offset exceeds section size");
1098
1099 if (DA.getU32(offset_ptr: &Offset) != dwarf::DW_LENGTH_DWARF64)
1100 return createStringError(EC: errc::invalid_argument, S: "32 bit contribution referenced from a 64 bit unit");
1101
1102 uint64_t Size = DA.getU64(offset_ptr: &Offset);
1103 uint8_t Version = DA.getU16(offset_ptr: &Offset);
1104 (void)DA.getU16(offset_ptr: &Offset); // padding
1105 // The encoded length includes the 2-byte version field and the 2-byte
1106 // padding, so we need to subtract them out when we populate the descriptor.
1107 return StrOffsetsContributionDescriptor(Offset, Size - 4, Version, DWARF64);
1108}
1109
1110// Look for a DWARF32-formatted contribution to the string offsets table
1111// starting at a given offset and record it in a descriptor.
1112static Expected<StrOffsetsContributionDescriptor>
1113parseDWARF32StringOffsetsTableHeader(DWARFDataExtractor &DA, uint64_t Offset) {
1114 if (!DA.isValidOffsetForDataOfSize(offset: Offset, length: 8))
1115 return createStringError(EC: errc::invalid_argument, S: "section offset exceeds section size");
1116
1117 uint32_t ContributionSize = DA.getU32(offset_ptr: &Offset);
1118 if (ContributionSize >= dwarf::DW_LENGTH_lo_reserved)
1119 return createStringError(EC: errc::invalid_argument, S: "invalid length");
1120
1121 uint8_t Version = DA.getU16(offset_ptr: &Offset);
1122 (void)DA.getU16(offset_ptr: &Offset); // padding
1123 // The encoded length includes the 2-byte version field and the 2-byte
1124 // padding, so we need to subtract them out when we populate the descriptor.
1125 return StrOffsetsContributionDescriptor(Offset, ContributionSize - 4, Version,
1126 DWARF32);
1127}
1128
1129static Expected<StrOffsetsContributionDescriptor>
1130parseDWARFStringOffsetsTableHeader(DWARFDataExtractor &DA,
1131 llvm::dwarf::DwarfFormat Format,
1132 uint64_t Offset) {
1133 StrOffsetsContributionDescriptor Desc;
1134 switch (Format) {
1135 case dwarf::DwarfFormat::DWARF64: {
1136 if (Offset < 16)
1137 return createStringError(EC: errc::invalid_argument, S: "insufficient space for 64 bit header prefix");
1138 auto DescOrError = parseDWARF64StringOffsetsTableHeader(DA, Offset: Offset - 16);
1139 if (!DescOrError)
1140 return DescOrError.takeError();
1141 Desc = *DescOrError;
1142 break;
1143 }
1144 case dwarf::DwarfFormat::DWARF32: {
1145 if (Offset < 8)
1146 return createStringError(EC: errc::invalid_argument, S: "insufficient space for 32 bit header prefix");
1147 auto DescOrError = parseDWARF32StringOffsetsTableHeader(DA, Offset: Offset - 8);
1148 if (!DescOrError)
1149 return DescOrError.takeError();
1150 Desc = *DescOrError;
1151 break;
1152 }
1153 }
1154 return Desc.validateContributionSize(DA);
1155}
1156
1157Expected<std::optional<StrOffsetsContributionDescriptor>>
1158DWARFUnit::determineStringOffsetsTableContribution(DWARFDataExtractor &DA) {
1159 assert(!IsDWO);
1160 auto OptOffset = toSectionOffset(V: getUnitDIE().find(Attr: DW_AT_str_offsets_base));
1161 if (!OptOffset)
1162 return std::nullopt;
1163 auto DescOrError =
1164 parseDWARFStringOffsetsTableHeader(DA, Format: Header.getFormat(), Offset: *OptOffset);
1165 if (!DescOrError)
1166 return DescOrError.takeError();
1167 return *DescOrError;
1168}
1169
1170Expected<std::optional<StrOffsetsContributionDescriptor>>
1171DWARFUnit::determineStringOffsetsTableContributionDWO(DWARFDataExtractor &DA) {
1172 assert(IsDWO);
1173 uint64_t Offset = 0;
1174 auto IndexEntry = Header.getIndexEntry();
1175 const auto *C =
1176 IndexEntry ? IndexEntry->getContribution(Sec: DW_SECT_STR_OFFSETS) : nullptr;
1177 if (C)
1178 Offset = C->getOffset();
1179 if (getVersion() >= 5) {
1180 if (DA.getData().data() == nullptr)
1181 return std::nullopt;
1182 Offset += Header.getFormat() == dwarf::DwarfFormat::DWARF32 ? 8 : 16;
1183 // Look for a valid contribution at the given offset.
1184 auto DescOrError = parseDWARFStringOffsetsTableHeader(DA, Format: Header.getFormat(), Offset);
1185 if (!DescOrError)
1186 return DescOrError.takeError();
1187 return *DescOrError;
1188 }
1189 // Prior to DWARF v5, we derive the contribution size from the
1190 // index table (in a package file). In a .dwo file it is simply
1191 // the length of the string offsets section.
1192 StrOffsetsContributionDescriptor Desc;
1193 if (C)
1194 Desc = StrOffsetsContributionDescriptor(C->getOffset(), C->getLength(), 4,
1195 Header.getFormat());
1196 else if (!IndexEntry && !StringOffsetSection.Data.empty())
1197 Desc = StrOffsetsContributionDescriptor(0, StringOffsetSection.Data.size(),
1198 4, Header.getFormat());
1199 else
1200 return std::nullopt;
1201 auto DescOrError = Desc.validateContributionSize(DA);
1202 if (!DescOrError)
1203 return DescOrError.takeError();
1204 return *DescOrError;
1205}
1206
1207std::optional<uint64_t> DWARFUnit::getRnglistOffset(uint32_t Index) {
1208 DataExtractor RangesData(RangeSection->Data, IsLittleEndian,
1209 getAddressByteSize());
1210 DWARFDataExtractor RangesDA(Context.getDWARFObj(), *RangeSection,
1211 IsLittleEndian, 0);
1212 if (std::optional<uint64_t> Off = llvm::DWARFListTableHeader::getOffsetEntry(
1213 Data: RangesData, OffsetTableOffset: RangeSectionBase, Format: getFormat(), Index))
1214 return *Off + RangeSectionBase;
1215 return std::nullopt;
1216}
1217
1218std::optional<uint64_t> DWARFUnit::getLoclistOffset(uint32_t Index) {
1219 if (std::optional<uint64_t> Off = llvm::DWARFListTableHeader::getOffsetEntry(
1220 Data: LocTable->getData(), OffsetTableOffset: LocSectionBase, Format: getFormat(), Index))
1221 return *Off + LocSectionBase;
1222 return std::nullopt;
1223}
1224