1 //===- DWARFDebugLine.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/DWARFDebugLine.h"
10 #include "llvm/ADT/Optional.h"
11 #include "llvm/ADT/SmallString.h"
12 #include "llvm/ADT/SmallVector.h"
13 #include "llvm/ADT/StringRef.h"
14 #include "llvm/BinaryFormat/Dwarf.h"
15 #include "llvm/DebugInfo/DWARF/DWARFDataExtractor.h"
16 #include "llvm/DebugInfo/DWARF/DWARFDie.h"
17 #include "llvm/DebugInfo/DWARF/DWARFFormValue.h"
18 #include "llvm/Support/Errc.h"
19 #include "llvm/Support/Format.h"
20 #include "llvm/Support/FormatVariadic.h"
21 #include "llvm/Support/raw_ostream.h"
22 #include <algorithm>
23 #include <cassert>
24 #include <cinttypes>
25 #include <cstdint>
26 #include <cstdio>
27 #include <utility>
28 
29 using namespace llvm;
30 using namespace dwarf;
31 
32 namespace llvm {
33 class DwarfContext;
34 }
35 
36 using FileLineInfoKind = DILineInfoSpecifier::FileLineInfoKind;
37 
38 namespace {
39 
40 struct ContentDescriptor {
41   dwarf::LineNumberEntryFormat Type;
42   dwarf::Form Form;
43 };
44 
45 using ContentDescriptors = SmallVector<ContentDescriptor, 4>;
46 
47 } // end anonymous namespace
48 
49 static bool versionIsSupported(uint16_t Version) {
50   return Version >= 2 && Version <= 5;
51 }
52 
53 void DWARFDebugLine::ContentTypeTracker::trackContentType(
54     dwarf::LineNumberEntryFormat ContentType) {
55   switch (ContentType) {
56   case dwarf::DW_LNCT_timestamp:
57     HasModTime = true;
58     break;
59   case dwarf::DW_LNCT_size:
60     HasLength = true;
61     break;
62   case dwarf::DW_LNCT_MD5:
63     HasMD5 = true;
64     break;
65   case dwarf::DW_LNCT_LLVM_source:
66     HasSource = true;
67     break;
68   default:
69     // We only care about values we consider optional, and new values may be
70     // added in the vendor extension range, so we do not match exhaustively.
71     break;
72   }
73 }
74 
75 DWARFDebugLine::Prologue::Prologue() { clear(); }
76 
77 bool DWARFDebugLine::Prologue::hasFileAtIndex(uint64_t FileIndex) const {
78   uint16_t DwarfVersion = getVersion();
79   assert(DwarfVersion != 0 &&
80          "line table prologue has no dwarf version information");
81   if (DwarfVersion >= 5)
82     return FileIndex < FileNames.size();
83   return FileIndex != 0 && FileIndex <= FileNames.size();
84 }
85 
86 Optional<uint64_t> DWARFDebugLine::Prologue::getLastValidFileIndex() const {
87   if (FileNames.empty())
88     return None;
89   uint16_t DwarfVersion = getVersion();
90   assert(DwarfVersion != 0 &&
91          "line table prologue has no dwarf version information");
92   // In DWARF v5 the file names are 0-indexed.
93   if (DwarfVersion >= 5)
94     return FileNames.size() - 1;
95   return FileNames.size();
96 }
97 
98 const llvm::DWARFDebugLine::FileNameEntry &
99 DWARFDebugLine::Prologue::getFileNameEntry(uint64_t Index) const {
100   uint16_t DwarfVersion = getVersion();
101   assert(DwarfVersion != 0 &&
102          "line table prologue has no dwarf version information");
103   // In DWARF v5 the file names are 0-indexed.
104   if (DwarfVersion >= 5)
105     return FileNames[Index];
106   return FileNames[Index - 1];
107 }
108 
109 void DWARFDebugLine::Prologue::clear() {
110   TotalLength = PrologueLength = 0;
111   SegSelectorSize = 0;
112   MinInstLength = MaxOpsPerInst = DefaultIsStmt = LineBase = LineRange = 0;
113   OpcodeBase = 0;
114   FormParams = dwarf::FormParams({0, 0, DWARF32});
115   ContentTypes = ContentTypeTracker();
116   StandardOpcodeLengths.clear();
117   IncludeDirectories.clear();
118   FileNames.clear();
119 }
120 
121 void DWARFDebugLine::Prologue::dump(raw_ostream &OS,
122                                     DIDumpOptions DumpOptions) const {
123   if (!totalLengthIsValid())
124     return;
125   int OffsetDumpWidth = 2 * dwarf::getDwarfOffsetByteSize(FormParams.Format);
126   OS << "Line table prologue:\n"
127      << format("    total_length: 0x%0*" PRIx64 "\n", OffsetDumpWidth,
128                TotalLength)
129      << "          format: " << dwarf::FormatString(FormParams.Format) << "\n"
130      << format("         version: %u\n", getVersion());
131   if (!versionIsSupported(getVersion()))
132     return;
133   if (getVersion() >= 5)
134     OS << format("    address_size: %u\n", getAddressSize())
135        << format(" seg_select_size: %u\n", SegSelectorSize);
136   OS << format(" prologue_length: 0x%0*" PRIx64 "\n", OffsetDumpWidth,
137                PrologueLength)
138      << format(" min_inst_length: %u\n", MinInstLength)
139      << format(getVersion() >= 4 ? "max_ops_per_inst: %u\n" : "", MaxOpsPerInst)
140      << format(" default_is_stmt: %u\n", DefaultIsStmt)
141      << format("       line_base: %i\n", LineBase)
142      << format("      line_range: %u\n", LineRange)
143      << format("     opcode_base: %u\n", OpcodeBase);
144 
145   for (uint32_t I = 0; I != StandardOpcodeLengths.size(); ++I)
146     OS << formatv("standard_opcode_lengths[{0}] = {1}\n",
147                   static_cast<dwarf::LineNumberOps>(I + 1),
148                   StandardOpcodeLengths[I]);
149 
150   if (!IncludeDirectories.empty()) {
151     // DWARF v5 starts directory indexes at 0.
152     uint32_t DirBase = getVersion() >= 5 ? 0 : 1;
153     for (uint32_t I = 0; I != IncludeDirectories.size(); ++I) {
154       OS << format("include_directories[%3u] = ", I + DirBase);
155       IncludeDirectories[I].dump(OS, DumpOptions);
156       OS << '\n';
157     }
158   }
159 
160   if (!FileNames.empty()) {
161     // DWARF v5 starts file indexes at 0.
162     uint32_t FileBase = getVersion() >= 5 ? 0 : 1;
163     for (uint32_t I = 0; I != FileNames.size(); ++I) {
164       const FileNameEntry &FileEntry = FileNames[I];
165       OS <<   format("file_names[%3u]:\n", I + FileBase);
166       OS <<          "           name: ";
167       FileEntry.Name.dump(OS, DumpOptions);
168       OS << '\n'
169          <<   format("      dir_index: %" PRIu64 "\n", FileEntry.DirIdx);
170       if (ContentTypes.HasMD5)
171         OS <<        "   md5_checksum: " << FileEntry.Checksum.digest() << '\n';
172       if (ContentTypes.HasModTime)
173         OS << format("       mod_time: 0x%8.8" PRIx64 "\n", FileEntry.ModTime);
174       if (ContentTypes.HasLength)
175         OS << format("         length: 0x%8.8" PRIx64 "\n", FileEntry.Length);
176       if (ContentTypes.HasSource) {
177         OS <<        "         source: ";
178         FileEntry.Source.dump(OS, DumpOptions);
179         OS << '\n';
180       }
181     }
182   }
183 }
184 
185 // Parse v2-v4 directory and file tables.
186 static Error
187 parseV2DirFileTables(const DWARFDataExtractor &DebugLineData,
188                      uint64_t *OffsetPtr,
189                      DWARFDebugLine::ContentTypeTracker &ContentTypes,
190                      std::vector<DWARFFormValue> &IncludeDirectories,
191                      std::vector<DWARFDebugLine::FileNameEntry> &FileNames) {
192   while (true) {
193     Error Err = Error::success();
194     StringRef S = DebugLineData.getCStrRef(OffsetPtr, &Err);
195     if (Err) {
196       consumeError(std::move(Err));
197       return createStringError(errc::invalid_argument,
198                                "include directories table was not null "
199                                "terminated before the end of the prologue");
200     }
201     if (S.empty())
202       break;
203     DWARFFormValue Dir =
204         DWARFFormValue::createFromPValue(dwarf::DW_FORM_string, S.data());
205     IncludeDirectories.push_back(Dir);
206   }
207 
208   ContentTypes.HasModTime = true;
209   ContentTypes.HasLength = true;
210 
211   while (true) {
212     Error Err = Error::success();
213     StringRef Name = DebugLineData.getCStrRef(OffsetPtr, &Err);
214     if (!Err && Name.empty())
215       break;
216 
217     DWARFDebugLine::FileNameEntry FileEntry;
218     FileEntry.Name =
219         DWARFFormValue::createFromPValue(dwarf::DW_FORM_string, Name.data());
220     FileEntry.DirIdx = DebugLineData.getULEB128(OffsetPtr, &Err);
221     FileEntry.ModTime = DebugLineData.getULEB128(OffsetPtr, &Err);
222     FileEntry.Length = DebugLineData.getULEB128(OffsetPtr, &Err);
223 
224     if (Err) {
225       consumeError(std::move(Err));
226       return createStringError(
227           errc::invalid_argument,
228           "file names table was not null terminated before "
229           "the end of the prologue");
230     }
231     FileNames.push_back(FileEntry);
232   }
233 
234   return Error::success();
235 }
236 
237 // Parse v5 directory/file entry content descriptions.
238 // Returns the descriptors, or an error if we did not find a path or ran off
239 // the end of the prologue.
240 static llvm::Expected<ContentDescriptors>
241 parseV5EntryFormat(const DWARFDataExtractor &DebugLineData, uint64_t *OffsetPtr,
242                    DWARFDebugLine::ContentTypeTracker *ContentTypes) {
243   Error Err = Error::success();
244   ContentDescriptors Descriptors;
245   int FormatCount = DebugLineData.getU8(OffsetPtr, &Err);
246   bool HasPath = false;
247   for (int I = 0; I != FormatCount && !Err; ++I) {
248     ContentDescriptor Descriptor;
249     Descriptor.Type =
250         dwarf::LineNumberEntryFormat(DebugLineData.getULEB128(OffsetPtr, &Err));
251     Descriptor.Form = dwarf::Form(DebugLineData.getULEB128(OffsetPtr, &Err));
252     if (Descriptor.Type == dwarf::DW_LNCT_path)
253       HasPath = true;
254     if (ContentTypes)
255       ContentTypes->trackContentType(Descriptor.Type);
256     Descriptors.push_back(Descriptor);
257   }
258 
259   if (Err)
260     return createStringError(errc::invalid_argument,
261                              "failed to parse entry content descriptors: %s",
262                              toString(std::move(Err)).c_str());
263 
264   if (!HasPath)
265     return createStringError(errc::invalid_argument,
266                              "failed to parse entry content descriptions"
267                              " because no path was found");
268   return Descriptors;
269 }
270 
271 static Error
272 parseV5DirFileTables(const DWARFDataExtractor &DebugLineData,
273                      uint64_t *OffsetPtr, const dwarf::FormParams &FormParams,
274                      const DWARFContext &Ctx, const DWARFUnit *U,
275                      DWARFDebugLine::ContentTypeTracker &ContentTypes,
276                      std::vector<DWARFFormValue> &IncludeDirectories,
277                      std::vector<DWARFDebugLine::FileNameEntry> &FileNames) {
278   // Get the directory entry description.
279   llvm::Expected<ContentDescriptors> DirDescriptors =
280       parseV5EntryFormat(DebugLineData, OffsetPtr, nullptr);
281   if (!DirDescriptors)
282     return DirDescriptors.takeError();
283 
284   // Get the directory entries, according to the format described above.
285   uint64_t DirEntryCount = DebugLineData.getULEB128(OffsetPtr);
286   for (uint64_t I = 0; I != DirEntryCount; ++I) {
287     for (auto Descriptor : *DirDescriptors) {
288       DWARFFormValue Value(Descriptor.Form);
289       switch (Descriptor.Type) {
290       case DW_LNCT_path:
291         if (!Value.extractValue(DebugLineData, OffsetPtr, FormParams, &Ctx, U))
292           return createStringError(errc::invalid_argument,
293                                    "failed to parse directory entry because "
294                                    "extracting the form value failed");
295         IncludeDirectories.push_back(Value);
296         break;
297       default:
298         if (!Value.skipValue(DebugLineData, OffsetPtr, FormParams))
299           return createStringError(errc::invalid_argument,
300                                    "failed to parse directory entry because "
301                                    "skipping the form value failed");
302       }
303     }
304   }
305 
306   // Get the file entry description.
307   llvm::Expected<ContentDescriptors> FileDescriptors =
308       parseV5EntryFormat(DebugLineData, OffsetPtr, &ContentTypes);
309   if (!FileDescriptors)
310     return FileDescriptors.takeError();
311 
312   // Get the file entries, according to the format described above.
313   uint64_t FileEntryCount = DebugLineData.getULEB128(OffsetPtr);
314   for (uint64_t I = 0; I != FileEntryCount; ++I) {
315     DWARFDebugLine::FileNameEntry FileEntry;
316     for (auto Descriptor : *FileDescriptors) {
317       DWARFFormValue Value(Descriptor.Form);
318       if (!Value.extractValue(DebugLineData, OffsetPtr, FormParams, &Ctx, U))
319         return createStringError(errc::invalid_argument,
320                                  "failed to parse file entry because "
321                                  "extracting the form value failed");
322       switch (Descriptor.Type) {
323       case DW_LNCT_path:
324         FileEntry.Name = Value;
325         break;
326       case DW_LNCT_LLVM_source:
327         FileEntry.Source = Value;
328         break;
329       case DW_LNCT_directory_index:
330         FileEntry.DirIdx = Value.getAsUnsignedConstant().value();
331         break;
332       case DW_LNCT_timestamp:
333         FileEntry.ModTime = Value.getAsUnsignedConstant().value();
334         break;
335       case DW_LNCT_size:
336         FileEntry.Length = Value.getAsUnsignedConstant().value();
337         break;
338       case DW_LNCT_MD5:
339         if (!Value.getAsBlock() || Value.getAsBlock().value().size() != 16)
340           return createStringError(
341               errc::invalid_argument,
342               "failed to parse file entry because the MD5 hash is invalid");
343         std::uninitialized_copy_n(Value.getAsBlock().value().begin(), 16,
344                                   FileEntry.Checksum.begin());
345         break;
346       default:
347         break;
348       }
349     }
350     FileNames.push_back(FileEntry);
351   }
352   return Error::success();
353 }
354 
355 uint64_t DWARFDebugLine::Prologue::getLength() const {
356   uint64_t Length = PrologueLength + sizeofTotalLength() +
357                     sizeof(getVersion()) + sizeofPrologueLength();
358   if (getVersion() >= 5)
359     Length += 2; // Address + Segment selector sizes.
360   return Length;
361 }
362 
363 Error DWARFDebugLine::Prologue::parse(
364     DWARFDataExtractor DebugLineData, uint64_t *OffsetPtr,
365     function_ref<void(Error)> RecoverableErrorHandler, const DWARFContext &Ctx,
366     const DWARFUnit *U) {
367   const uint64_t PrologueOffset = *OffsetPtr;
368 
369   clear();
370   DataExtractor::Cursor Cursor(*OffsetPtr);
371   std::tie(TotalLength, FormParams.Format) =
372       DebugLineData.getInitialLength(Cursor);
373 
374   DebugLineData =
375       DWARFDataExtractor(DebugLineData, Cursor.tell() + TotalLength);
376   FormParams.Version = DebugLineData.getU16(Cursor);
377   if (Cursor && !versionIsSupported(getVersion())) {
378     // Treat this error as unrecoverable - we cannot be sure what any of
379     // the data represents including the length field, so cannot skip it or make
380     // any reasonable assumptions.
381     *OffsetPtr = Cursor.tell();
382     return createStringError(
383         errc::not_supported,
384         "parsing line table prologue at offset 0x%8.8" PRIx64
385         ": unsupported version %" PRIu16,
386         PrologueOffset, getVersion());
387   }
388 
389   if (getVersion() >= 5) {
390     FormParams.AddrSize = DebugLineData.getU8(Cursor);
391     assert((!Cursor || DebugLineData.getAddressSize() == 0 ||
392             DebugLineData.getAddressSize() == getAddressSize()) &&
393            "Line table header and data extractor disagree");
394     SegSelectorSize = DebugLineData.getU8(Cursor);
395   }
396 
397   PrologueLength =
398       DebugLineData.getRelocatedValue(Cursor, sizeofPrologueLength());
399   const uint64_t EndPrologueOffset = PrologueLength + Cursor.tell();
400   DebugLineData = DWARFDataExtractor(DebugLineData, EndPrologueOffset);
401   MinInstLength = DebugLineData.getU8(Cursor);
402   if (getVersion() >= 4)
403     MaxOpsPerInst = DebugLineData.getU8(Cursor);
404   DefaultIsStmt = DebugLineData.getU8(Cursor);
405   LineBase = DebugLineData.getU8(Cursor);
406   LineRange = DebugLineData.getU8(Cursor);
407   OpcodeBase = DebugLineData.getU8(Cursor);
408 
409   if (Cursor && OpcodeBase == 0) {
410     // If the opcode base is 0, we cannot read the standard opcode lengths (of
411     // which there are supposed to be one fewer than the opcode base). Assume
412     // there are no standard opcodes and continue parsing.
413     RecoverableErrorHandler(createStringError(
414         errc::invalid_argument,
415         "parsing line table prologue at offset 0x%8.8" PRIx64
416         " found opcode base of 0. Assuming no standard opcodes",
417         PrologueOffset));
418   } else if (Cursor) {
419     StandardOpcodeLengths.reserve(OpcodeBase - 1);
420     for (uint32_t I = 1; I < OpcodeBase; ++I) {
421       uint8_t OpLen = DebugLineData.getU8(Cursor);
422       StandardOpcodeLengths.push_back(OpLen);
423     }
424   }
425 
426   *OffsetPtr = Cursor.tell();
427   // A corrupt file name or directory table does not prevent interpretation of
428   // the main line program, so check the cursor state now so that its errors can
429   // be handled separately.
430   if (!Cursor)
431     return createStringError(
432         errc::invalid_argument,
433         "parsing line table prologue at offset 0x%8.8" PRIx64 ": %s",
434         PrologueOffset, toString(Cursor.takeError()).c_str());
435 
436   Error E =
437       getVersion() >= 5
438           ? parseV5DirFileTables(DebugLineData, OffsetPtr, FormParams, Ctx, U,
439                                  ContentTypes, IncludeDirectories, FileNames)
440           : parseV2DirFileTables(DebugLineData, OffsetPtr, ContentTypes,
441                                  IncludeDirectories, FileNames);
442   if (E) {
443     RecoverableErrorHandler(joinErrors(
444         createStringError(
445             errc::invalid_argument,
446             "parsing line table prologue at 0x%8.8" PRIx64
447             " found an invalid directory or file table description at"
448             " 0x%8.8" PRIx64,
449             PrologueOffset, *OffsetPtr),
450         std::move(E)));
451     return Error::success();
452   }
453 
454   assert(*OffsetPtr <= EndPrologueOffset);
455   if (*OffsetPtr != EndPrologueOffset) {
456     RecoverableErrorHandler(createStringError(
457         errc::invalid_argument,
458         "unknown data in line table prologue at offset 0x%8.8" PRIx64
459         ": parsing ended (at offset 0x%8.8" PRIx64
460         ") before reaching the prologue end at offset 0x%8.8" PRIx64,
461         PrologueOffset, *OffsetPtr, EndPrologueOffset));
462   }
463   return Error::success();
464 }
465 
466 DWARFDebugLine::Row::Row(bool DefaultIsStmt) { reset(DefaultIsStmt); }
467 
468 void DWARFDebugLine::Row::postAppend() {
469   Discriminator = 0;
470   BasicBlock = false;
471   PrologueEnd = false;
472   EpilogueBegin = false;
473 }
474 
475 void DWARFDebugLine::Row::reset(bool DefaultIsStmt) {
476   Address.Address = 0;
477   Address.SectionIndex = object::SectionedAddress::UndefSection;
478   Line = 1;
479   Column = 0;
480   File = 1;
481   Isa = 0;
482   Discriminator = 0;
483   IsStmt = DefaultIsStmt;
484   BasicBlock = false;
485   EndSequence = false;
486   PrologueEnd = false;
487   EpilogueBegin = false;
488 }
489 
490 void DWARFDebugLine::Row::dumpTableHeader(raw_ostream &OS, unsigned Indent) {
491   OS.indent(Indent)
492       << "Address            Line   Column File   ISA Discriminator Flags\n";
493   OS.indent(Indent)
494       << "------------------ ------ ------ ------ --- ------------- "
495          "-------------\n";
496 }
497 
498 void DWARFDebugLine::Row::dump(raw_ostream &OS) const {
499   OS << format("0x%16.16" PRIx64 " %6u %6u", Address.Address, Line, Column)
500      << format(" %6u %3u %13u ", File, Isa, Discriminator)
501      << (IsStmt ? " is_stmt" : "") << (BasicBlock ? " basic_block" : "")
502      << (PrologueEnd ? " prologue_end" : "")
503      << (EpilogueBegin ? " epilogue_begin" : "")
504      << (EndSequence ? " end_sequence" : "") << '\n';
505 }
506 
507 DWARFDebugLine::Sequence::Sequence() { reset(); }
508 
509 void DWARFDebugLine::Sequence::reset() {
510   LowPC = 0;
511   HighPC = 0;
512   SectionIndex = object::SectionedAddress::UndefSection;
513   FirstRowIndex = 0;
514   LastRowIndex = 0;
515   Empty = true;
516 }
517 
518 DWARFDebugLine::LineTable::LineTable() { clear(); }
519 
520 void DWARFDebugLine::LineTable::dump(raw_ostream &OS,
521                                      DIDumpOptions DumpOptions) const {
522   Prologue.dump(OS, DumpOptions);
523 
524   if (!Rows.empty()) {
525     OS << '\n';
526     Row::dumpTableHeader(OS, 0);
527     for (const Row &R : Rows) {
528       R.dump(OS);
529     }
530   }
531 
532   // Terminate the table with a final blank line to clearly delineate it from
533   // later dumps.
534   OS << '\n';
535 }
536 
537 void DWARFDebugLine::LineTable::clear() {
538   Prologue.clear();
539   Rows.clear();
540   Sequences.clear();
541 }
542 
543 DWARFDebugLine::ParsingState::ParsingState(
544     struct LineTable *LT, uint64_t TableOffset,
545     function_ref<void(Error)> ErrorHandler)
546     : LineTable(LT), LineTableOffset(TableOffset), ErrorHandler(ErrorHandler) {
547   resetRowAndSequence();
548 }
549 
550 void DWARFDebugLine::ParsingState::resetRowAndSequence() {
551   Row.reset(LineTable->Prologue.DefaultIsStmt);
552   Sequence.reset();
553 }
554 
555 void DWARFDebugLine::ParsingState::appendRowToMatrix() {
556   unsigned RowNumber = LineTable->Rows.size();
557   if (Sequence.Empty) {
558     // Record the beginning of instruction sequence.
559     Sequence.Empty = false;
560     Sequence.LowPC = Row.Address.Address;
561     Sequence.FirstRowIndex = RowNumber;
562   }
563   LineTable->appendRow(Row);
564   if (Row.EndSequence) {
565     // Record the end of instruction sequence.
566     Sequence.HighPC = Row.Address.Address;
567     Sequence.LastRowIndex = RowNumber + 1;
568     Sequence.SectionIndex = Row.Address.SectionIndex;
569     if (Sequence.isValid())
570       LineTable->appendSequence(Sequence);
571     Sequence.reset();
572   }
573   Row.postAppend();
574 }
575 
576 const DWARFDebugLine::LineTable *
577 DWARFDebugLine::getLineTable(uint64_t Offset) const {
578   LineTableConstIter Pos = LineTableMap.find(Offset);
579   if (Pos != LineTableMap.end())
580     return &Pos->second;
581   return nullptr;
582 }
583 
584 Expected<const DWARFDebugLine::LineTable *> DWARFDebugLine::getOrParseLineTable(
585     DWARFDataExtractor &DebugLineData, uint64_t Offset, const DWARFContext &Ctx,
586     const DWARFUnit *U, function_ref<void(Error)> RecoverableErrorHandler) {
587   if (!DebugLineData.isValidOffset(Offset))
588     return createStringError(errc::invalid_argument, "offset 0x%8.8" PRIx64
589                        " is not a valid debug line section offset",
590                        Offset);
591 
592   std::pair<LineTableIter, bool> Pos =
593       LineTableMap.insert(LineTableMapTy::value_type(Offset, LineTable()));
594   LineTable *LT = &Pos.first->second;
595   if (Pos.second) {
596     if (Error Err =
597             LT->parse(DebugLineData, &Offset, Ctx, U, RecoverableErrorHandler))
598       return std::move(Err);
599     return LT;
600   }
601   return LT;
602 }
603 
604 void DWARFDebugLine::clearLineTable(uint64_t Offset) {
605   LineTableMap.erase(Offset);
606 }
607 
608 static StringRef getOpcodeName(uint8_t Opcode, uint8_t OpcodeBase) {
609   assert(Opcode != 0);
610   if (Opcode < OpcodeBase)
611     return LNStandardString(Opcode);
612   return "special";
613 }
614 
615 uint64_t DWARFDebugLine::ParsingState::advanceAddr(uint64_t OperationAdvance,
616                                                    uint8_t Opcode,
617                                                    uint64_t OpcodeOffset) {
618   StringRef OpcodeName = getOpcodeName(Opcode, LineTable->Prologue.OpcodeBase);
619   // For versions less than 4, the MaxOpsPerInst member is set to 0, as the
620   // maximum_operations_per_instruction field wasn't introduced until DWARFv4.
621   // Don't warn about bad values in this situation.
622   if (ReportAdvanceAddrProblem && LineTable->Prologue.getVersion() >= 4 &&
623       LineTable->Prologue.MaxOpsPerInst != 1)
624     ErrorHandler(createStringError(
625         errc::not_supported,
626         "line table program at offset 0x%8.8" PRIx64
627         " contains a %s opcode at offset 0x%8.8" PRIx64
628         ", but the prologue maximum_operations_per_instruction value is %" PRId8
629         ", which is unsupported. Assuming a value of 1 instead",
630         LineTableOffset, OpcodeName.data(), OpcodeOffset,
631         LineTable->Prologue.MaxOpsPerInst));
632   if (ReportAdvanceAddrProblem && LineTable->Prologue.MinInstLength == 0)
633     ErrorHandler(
634         createStringError(errc::invalid_argument,
635                           "line table program at offset 0x%8.8" PRIx64
636                           " contains a %s opcode at offset 0x%8.8" PRIx64
637                           ", but the prologue minimum_instruction_length value "
638                           "is 0, which prevents any address advancing",
639                           LineTableOffset, OpcodeName.data(), OpcodeOffset));
640   ReportAdvanceAddrProblem = false;
641   uint64_t AddrOffset = OperationAdvance * LineTable->Prologue.MinInstLength;
642   Row.Address.Address += AddrOffset;
643   return AddrOffset;
644 }
645 
646 DWARFDebugLine::ParsingState::AddrAndAdjustedOpcode
647 DWARFDebugLine::ParsingState::advanceAddrForOpcode(uint8_t Opcode,
648                                                    uint64_t OpcodeOffset) {
649   assert(Opcode == DW_LNS_const_add_pc ||
650          Opcode >= LineTable->Prologue.OpcodeBase);
651   if (ReportBadLineRange && LineTable->Prologue.LineRange == 0) {
652     StringRef OpcodeName =
653         getOpcodeName(Opcode, LineTable->Prologue.OpcodeBase);
654     ErrorHandler(
655         createStringError(errc::not_supported,
656                           "line table program at offset 0x%8.8" PRIx64
657                           " contains a %s opcode at offset 0x%8.8" PRIx64
658                           ", but the prologue line_range value is 0. The "
659                           "address and line will not be adjusted",
660                           LineTableOffset, OpcodeName.data(), OpcodeOffset));
661     ReportBadLineRange = false;
662   }
663 
664   uint8_t OpcodeValue = Opcode;
665   if (Opcode == DW_LNS_const_add_pc)
666     OpcodeValue = 255;
667   uint8_t AdjustedOpcode = OpcodeValue - LineTable->Prologue.OpcodeBase;
668   uint64_t OperationAdvance =
669       LineTable->Prologue.LineRange != 0
670           ? AdjustedOpcode / LineTable->Prologue.LineRange
671           : 0;
672   uint64_t AddrOffset = advanceAddr(OperationAdvance, Opcode, OpcodeOffset);
673   return {AddrOffset, AdjustedOpcode};
674 }
675 
676 DWARFDebugLine::ParsingState::AddrAndLineDelta
677 DWARFDebugLine::ParsingState::handleSpecialOpcode(uint8_t Opcode,
678                                                   uint64_t OpcodeOffset) {
679   // A special opcode value is chosen based on the amount that needs
680   // to be added to the line and address registers. The maximum line
681   // increment for a special opcode is the value of the line_base
682   // field in the header, plus the value of the line_range field,
683   // minus 1 (line base + line range - 1). If the desired line
684   // increment is greater than the maximum line increment, a standard
685   // opcode must be used instead of a special opcode. The "address
686   // advance" is calculated by dividing the desired address increment
687   // by the minimum_instruction_length field from the header. The
688   // special opcode is then calculated using the following formula:
689   //
690   //  opcode = (desired line increment - line_base) +
691   //           (line_range * address advance) + opcode_base
692   //
693   // If the resulting opcode is greater than 255, a standard opcode
694   // must be used instead.
695   //
696   // To decode a special opcode, subtract the opcode_base from the
697   // opcode itself to give the adjusted opcode. The amount to
698   // increment the address register is the result of the adjusted
699   // opcode divided by the line_range multiplied by the
700   // minimum_instruction_length field from the header. That is:
701   //
702   //  address increment = (adjusted opcode / line_range) *
703   //                      minimum_instruction_length
704   //
705   // The amount to increment the line register is the line_base plus
706   // the result of the adjusted opcode modulo the line_range. That is:
707   //
708   // line increment = line_base + (adjusted opcode % line_range)
709 
710   DWARFDebugLine::ParsingState::AddrAndAdjustedOpcode AddrAdvanceResult =
711       advanceAddrForOpcode(Opcode, OpcodeOffset);
712   int32_t LineOffset = 0;
713   if (LineTable->Prologue.LineRange != 0)
714     LineOffset =
715         LineTable->Prologue.LineBase +
716         (AddrAdvanceResult.AdjustedOpcode % LineTable->Prologue.LineRange);
717   Row.Line += LineOffset;
718   return {AddrAdvanceResult.AddrDelta, LineOffset};
719 }
720 
721 /// Parse a ULEB128 using the specified \p Cursor. \returns the parsed value on
722 /// success, or None if \p Cursor is in a failing state.
723 template <typename T>
724 static Optional<T> parseULEB128(DWARFDataExtractor &Data,
725                                 DataExtractor::Cursor &Cursor) {
726   T Value = Data.getULEB128(Cursor);
727   if (Cursor)
728     return Value;
729   return None;
730 }
731 
732 Error DWARFDebugLine::LineTable::parse(
733     DWARFDataExtractor &DebugLineData, uint64_t *OffsetPtr,
734     const DWARFContext &Ctx, const DWARFUnit *U,
735     function_ref<void(Error)> RecoverableErrorHandler, raw_ostream *OS,
736     bool Verbose) {
737   assert((OS || !Verbose) && "cannot have verbose output without stream");
738   const uint64_t DebugLineOffset = *OffsetPtr;
739 
740   clear();
741 
742   Error PrologueErr =
743       Prologue.parse(DebugLineData, OffsetPtr, RecoverableErrorHandler, Ctx, U);
744 
745   if (OS) {
746     DIDumpOptions DumpOptions;
747     DumpOptions.Verbose = Verbose;
748     Prologue.dump(*OS, DumpOptions);
749   }
750 
751   if (PrologueErr) {
752     // Ensure there is a blank line after the prologue to clearly delineate it
753     // from later dumps.
754     if (OS)
755       *OS << "\n";
756     return PrologueErr;
757   }
758 
759   uint64_t ProgramLength = Prologue.TotalLength + Prologue.sizeofTotalLength();
760   if (!DebugLineData.isValidOffsetForDataOfSize(DebugLineOffset,
761                                                 ProgramLength)) {
762     assert(DebugLineData.size() > DebugLineOffset &&
763            "prologue parsing should handle invalid offset");
764     uint64_t BytesRemaining = DebugLineData.size() - DebugLineOffset;
765     RecoverableErrorHandler(
766         createStringError(errc::invalid_argument,
767                           "line table program with offset 0x%8.8" PRIx64
768                           " has length 0x%8.8" PRIx64 " but only 0x%8.8" PRIx64
769                           " bytes are available",
770                           DebugLineOffset, ProgramLength, BytesRemaining));
771     // Continue by capping the length at the number of remaining bytes.
772     ProgramLength = BytesRemaining;
773   }
774 
775   // Create a DataExtractor which can only see the data up to the end of the
776   // table, to prevent reading past the end.
777   const uint64_t EndOffset = DebugLineOffset + ProgramLength;
778   DWARFDataExtractor TableData(DebugLineData, EndOffset);
779 
780   // See if we should tell the data extractor the address size.
781   if (TableData.getAddressSize() == 0)
782     TableData.setAddressSize(Prologue.getAddressSize());
783   else
784     assert(Prologue.getAddressSize() == 0 ||
785            Prologue.getAddressSize() == TableData.getAddressSize());
786 
787   ParsingState State(this, DebugLineOffset, RecoverableErrorHandler);
788 
789   *OffsetPtr = DebugLineOffset + Prologue.getLength();
790   if (OS && *OffsetPtr < EndOffset) {
791     *OS << '\n';
792     Row::dumpTableHeader(*OS, /*Indent=*/Verbose ? 12 : 0);
793   }
794   bool TombstonedAddress = false;
795   auto EmitRow = [&] {
796     if (!TombstonedAddress) {
797       if (Verbose) {
798         *OS << "\n";
799         OS->indent(12);
800       }
801       if (OS)
802         State.Row.dump(*OS);
803       State.appendRowToMatrix();
804     }
805   };
806   while (*OffsetPtr < EndOffset) {
807     DataExtractor::Cursor Cursor(*OffsetPtr);
808 
809     if (Verbose)
810       *OS << format("0x%08.08" PRIx64 ": ", *OffsetPtr);
811 
812     uint64_t OpcodeOffset = *OffsetPtr;
813     uint8_t Opcode = TableData.getU8(Cursor);
814     size_t RowCount = Rows.size();
815 
816     if (Cursor && Verbose)
817       *OS << format("%02.02" PRIx8 " ", Opcode);
818 
819     if (Opcode == 0) {
820       // Extended Opcodes always start with a zero opcode followed by
821       // a uleb128 length so you can skip ones you don't know about
822       uint64_t Len = TableData.getULEB128(Cursor);
823       uint64_t ExtOffset = Cursor.tell();
824 
825       // Tolerate zero-length; assume length is correct and soldier on.
826       if (Len == 0) {
827         if (Cursor && Verbose)
828           *OS << "Badly formed extended line op (length 0)\n";
829         if (!Cursor) {
830           if (Verbose)
831             *OS << "\n";
832           RecoverableErrorHandler(Cursor.takeError());
833         }
834         *OffsetPtr = Cursor.tell();
835         continue;
836       }
837 
838       uint8_t SubOpcode = TableData.getU8(Cursor);
839       // OperandOffset will be the same as ExtOffset, if it was not possible to
840       // read the SubOpcode.
841       uint64_t OperandOffset = Cursor.tell();
842       if (Verbose)
843         *OS << LNExtendedString(SubOpcode);
844       switch (SubOpcode) {
845       case DW_LNE_end_sequence:
846         // Set the end_sequence register of the state machine to true and
847         // append a row to the matrix using the current values of the
848         // state-machine registers. Then reset the registers to the initial
849         // values specified above. Every statement program sequence must end
850         // with a DW_LNE_end_sequence instruction which creates a row whose
851         // address is that of the byte after the last target machine instruction
852         // of the sequence.
853         State.Row.EndSequence = true;
854         // No need to test the Cursor is valid here, since it must be to get
855         // into this code path - if it were invalid, the default case would be
856         // followed.
857         EmitRow();
858         State.resetRowAndSequence();
859         break;
860 
861       case DW_LNE_set_address:
862         // Takes a single relocatable address as an operand. The size of the
863         // operand is the size appropriate to hold an address on the target
864         // machine. Set the address register to the value given by the
865         // relocatable address. All of the other statement program opcodes
866         // that affect the address register add a delta to it. This instruction
867         // stores a relocatable value into it instead.
868         //
869         // Make sure the extractor knows the address size.  If not, infer it
870         // from the size of the operand.
871         {
872           uint8_t ExtractorAddressSize = TableData.getAddressSize();
873           uint64_t OpcodeAddressSize = Len - 1;
874           if (ExtractorAddressSize != OpcodeAddressSize &&
875               ExtractorAddressSize != 0)
876             RecoverableErrorHandler(createStringError(
877                 errc::invalid_argument,
878                 "mismatching address size at offset 0x%8.8" PRIx64
879                 " expected 0x%2.2" PRIx8 " found 0x%2.2" PRIx64,
880                 ExtOffset, ExtractorAddressSize, Len - 1));
881 
882           // Assume that the line table is correct and temporarily override the
883           // address size. If the size is unsupported, give up trying to read
884           // the address and continue to the next opcode.
885           if (OpcodeAddressSize != 1 && OpcodeAddressSize != 2 &&
886               OpcodeAddressSize != 4 && OpcodeAddressSize != 8) {
887             RecoverableErrorHandler(createStringError(
888                 errc::invalid_argument,
889                 "address size 0x%2.2" PRIx64
890                 " of DW_LNE_set_address opcode at offset 0x%8.8" PRIx64
891                 " is unsupported",
892                 OpcodeAddressSize, ExtOffset));
893             TableData.skip(Cursor, OpcodeAddressSize);
894           } else {
895             TableData.setAddressSize(OpcodeAddressSize);
896             State.Row.Address.Address = TableData.getRelocatedAddress(
897                 Cursor, &State.Row.Address.SectionIndex);
898 
899             uint64_t Tombstone =
900                 dwarf::computeTombstoneAddress(OpcodeAddressSize);
901             TombstonedAddress = State.Row.Address.Address == Tombstone;
902 
903             // Restore the address size if the extractor already had it.
904             if (ExtractorAddressSize != 0)
905               TableData.setAddressSize(ExtractorAddressSize);
906           }
907 
908           if (Cursor && Verbose) {
909             *OS << " (";
910             DWARFFormValue::dumpAddress(*OS, OpcodeAddressSize, State.Row.Address.Address);
911             *OS << ')';
912           }
913         }
914         break;
915 
916       case DW_LNE_define_file:
917         // Takes 4 arguments. The first is a null terminated string containing
918         // a source file name. The second is an unsigned LEB128 number
919         // representing the directory index of the directory in which the file
920         // was found. The third is an unsigned LEB128 number representing the
921         // time of last modification of the file. The fourth is an unsigned
922         // LEB128 number representing the length in bytes of the file. The time
923         // and length fields may contain LEB128(0) if the information is not
924         // available.
925         //
926         // The directory index represents an entry in the include_directories
927         // section of the statement program prologue. The index is LEB128(0)
928         // if the file was found in the current directory of the compilation,
929         // LEB128(1) if it was found in the first directory in the
930         // include_directories section, and so on. The directory index is
931         // ignored for file names that represent full path names.
932         //
933         // The files are numbered, starting at 1, in the order in which they
934         // appear; the names in the prologue come before names defined by
935         // the DW_LNE_define_file instruction. These numbers are used in the
936         // the file register of the state machine.
937         {
938           FileNameEntry FileEntry;
939           const char *Name = TableData.getCStr(Cursor);
940           FileEntry.Name =
941               DWARFFormValue::createFromPValue(dwarf::DW_FORM_string, Name);
942           FileEntry.DirIdx = TableData.getULEB128(Cursor);
943           FileEntry.ModTime = TableData.getULEB128(Cursor);
944           FileEntry.Length = TableData.getULEB128(Cursor);
945           Prologue.FileNames.push_back(FileEntry);
946           if (Cursor && Verbose)
947             *OS << " (" << Name << ", dir=" << FileEntry.DirIdx << ", mod_time="
948                 << format("(0x%16.16" PRIx64 ")", FileEntry.ModTime)
949                 << ", length=" << FileEntry.Length << ")";
950         }
951         break;
952 
953       case DW_LNE_set_discriminator:
954         State.Row.Discriminator = TableData.getULEB128(Cursor);
955         if (Cursor && Verbose)
956           *OS << " (" << State.Row.Discriminator << ")";
957         break;
958 
959       default:
960         if (Cursor && Verbose)
961           *OS << format("Unrecognized extended op 0x%02.02" PRIx8, SubOpcode)
962               << format(" length %" PRIx64, Len);
963         // Len doesn't include the zero opcode byte or the length itself, but
964         // it does include the sub_opcode, so we have to adjust for that.
965         TableData.skip(Cursor, Len - 1);
966         break;
967       }
968       // Make sure the length as recorded in the table and the standard length
969       // for the opcode match. If they don't, continue from the end as claimed
970       // by the table. Similarly, continue from the claimed end in the event of
971       // a parsing error.
972       uint64_t End = ExtOffset + Len;
973       if (Cursor && Cursor.tell() != End)
974         RecoverableErrorHandler(createStringError(
975             errc::illegal_byte_sequence,
976             "unexpected line op length at offset 0x%8.8" PRIx64
977             " expected 0x%2.2" PRIx64 " found 0x%2.2" PRIx64,
978             ExtOffset, Len, Cursor.tell() - ExtOffset));
979       if (!Cursor && Verbose) {
980         DWARFDataExtractor::Cursor ByteCursor(OperandOffset);
981         uint8_t Byte = TableData.getU8(ByteCursor);
982         if (ByteCursor) {
983           *OS << " (<parsing error>";
984           do {
985             *OS << format(" %2.2" PRIx8, Byte);
986             Byte = TableData.getU8(ByteCursor);
987           } while (ByteCursor);
988           *OS << ")";
989         }
990 
991         // The only parse failure in this case should be if the end was reached.
992         // In that case, throw away the error, as the main Cursor's error will
993         // be sufficient.
994         consumeError(ByteCursor.takeError());
995       }
996       *OffsetPtr = End;
997     } else if (Opcode < Prologue.OpcodeBase) {
998       if (Verbose)
999         *OS << LNStandardString(Opcode);
1000       switch (Opcode) {
1001       // Standard Opcodes
1002       case DW_LNS_copy:
1003         // Takes no arguments. Append a row to the matrix using the
1004         // current values of the state-machine registers.
1005         EmitRow();
1006         break;
1007 
1008       case DW_LNS_advance_pc:
1009         // Takes a single unsigned LEB128 operand, multiplies it by the
1010         // min_inst_length field of the prologue, and adds the
1011         // result to the address register of the state machine.
1012         if (Optional<uint64_t> Operand =
1013                 parseULEB128<uint64_t>(TableData, Cursor)) {
1014           uint64_t AddrOffset =
1015               State.advanceAddr(*Operand, Opcode, OpcodeOffset);
1016           if (Verbose)
1017             *OS << " (" << AddrOffset << ")";
1018         }
1019         break;
1020 
1021       case DW_LNS_advance_line:
1022         // Takes a single signed LEB128 operand and adds that value to
1023         // the line register of the state machine.
1024         {
1025           int64_t LineDelta = TableData.getSLEB128(Cursor);
1026           if (Cursor) {
1027             State.Row.Line += LineDelta;
1028             if (Verbose)
1029               *OS << " (" << State.Row.Line << ")";
1030           }
1031         }
1032         break;
1033 
1034       case DW_LNS_set_file:
1035         // Takes a single unsigned LEB128 operand and stores it in the file
1036         // register of the state machine.
1037         if (Optional<uint16_t> File =
1038                 parseULEB128<uint16_t>(TableData, Cursor)) {
1039           State.Row.File = *File;
1040           if (Verbose)
1041             *OS << " (" << State.Row.File << ")";
1042         }
1043         break;
1044 
1045       case DW_LNS_set_column:
1046         // Takes a single unsigned LEB128 operand and stores it in the
1047         // column register of the state machine.
1048         if (Optional<uint16_t> Column =
1049                 parseULEB128<uint16_t>(TableData, Cursor)) {
1050           State.Row.Column = *Column;
1051           if (Verbose)
1052             *OS << " (" << State.Row.Column << ")";
1053         }
1054         break;
1055 
1056       case DW_LNS_negate_stmt:
1057         // Takes no arguments. Set the is_stmt register of the state
1058         // machine to the logical negation of its current value.
1059         State.Row.IsStmt = !State.Row.IsStmt;
1060         break;
1061 
1062       case DW_LNS_set_basic_block:
1063         // Takes no arguments. Set the basic_block register of the
1064         // state machine to true
1065         State.Row.BasicBlock = true;
1066         break;
1067 
1068       case DW_LNS_const_add_pc:
1069         // Takes no arguments. Add to the address register of the state
1070         // machine the address increment value corresponding to special
1071         // opcode 255. The motivation for DW_LNS_const_add_pc is this:
1072         // when the statement program needs to advance the address by a
1073         // small amount, it can use a single special opcode, which occupies
1074         // a single byte. When it needs to advance the address by up to
1075         // twice the range of the last special opcode, it can use
1076         // DW_LNS_const_add_pc followed by a special opcode, for a total
1077         // of two bytes. Only if it needs to advance the address by more
1078         // than twice that range will it need to use both DW_LNS_advance_pc
1079         // and a special opcode, requiring three or more bytes.
1080         {
1081           uint64_t AddrOffset =
1082               State.advanceAddrForOpcode(Opcode, OpcodeOffset).AddrDelta;
1083           if (Verbose)
1084             *OS << format(" (0x%16.16" PRIx64 ")", AddrOffset);
1085         }
1086         break;
1087 
1088       case DW_LNS_fixed_advance_pc:
1089         // Takes a single uhalf operand. Add to the address register of
1090         // the state machine the value of the (unencoded) operand. This
1091         // is the only extended opcode that takes an argument that is not
1092         // a variable length number. The motivation for DW_LNS_fixed_advance_pc
1093         // is this: existing assemblers cannot emit DW_LNS_advance_pc or
1094         // special opcodes because they cannot encode LEB128 numbers or
1095         // judge when the computation of a special opcode overflows and
1096         // requires the use of DW_LNS_advance_pc. Such assemblers, however,
1097         // can use DW_LNS_fixed_advance_pc instead, sacrificing compression.
1098         {
1099           uint16_t PCOffset =
1100               TableData.getRelocatedValue(Cursor, 2);
1101           if (Cursor) {
1102             State.Row.Address.Address += PCOffset;
1103             if (Verbose)
1104               *OS << format(" (0x%4.4" PRIx16 ")", PCOffset);
1105           }
1106         }
1107         break;
1108 
1109       case DW_LNS_set_prologue_end:
1110         // Takes no arguments. Set the prologue_end register of the
1111         // state machine to true
1112         State.Row.PrologueEnd = true;
1113         break;
1114 
1115       case DW_LNS_set_epilogue_begin:
1116         // Takes no arguments. Set the basic_block register of the
1117         // state machine to true
1118         State.Row.EpilogueBegin = true;
1119         break;
1120 
1121       case DW_LNS_set_isa:
1122         // Takes a single unsigned LEB128 operand and stores it in the
1123         // ISA register of the state machine.
1124         if (Optional<uint8_t> Isa = parseULEB128<uint8_t>(TableData, Cursor)) {
1125           State.Row.Isa = *Isa;
1126           if (Verbose)
1127             *OS << " (" << (uint64_t)State.Row.Isa << ")";
1128         }
1129         break;
1130 
1131       default:
1132         // Handle any unknown standard opcodes here. We know the lengths
1133         // of such opcodes because they are specified in the prologue
1134         // as a multiple of LEB128 operands for each opcode.
1135         {
1136           assert(Opcode - 1U < Prologue.StandardOpcodeLengths.size());
1137           if (Verbose)
1138             *OS << "Unrecognized standard opcode";
1139           uint8_t OpcodeLength = Prologue.StandardOpcodeLengths[Opcode - 1];
1140           std::vector<uint64_t> Operands;
1141           for (uint8_t I = 0; I < OpcodeLength; ++I) {
1142             if (Optional<uint64_t> Value =
1143                     parseULEB128<uint64_t>(TableData, Cursor))
1144               Operands.push_back(*Value);
1145             else
1146               break;
1147           }
1148           if (Verbose && !Operands.empty()) {
1149             *OS << " (operands: ";
1150             bool First = true;
1151             for (uint64_t Value : Operands) {
1152               if (!First)
1153                 *OS << ", ";
1154               First = false;
1155               *OS << format("0x%16.16" PRIx64, Value);
1156             }
1157             if (Verbose)
1158               *OS << ')';
1159           }
1160         }
1161         break;
1162       }
1163 
1164       *OffsetPtr = Cursor.tell();
1165     } else {
1166       // Special Opcodes.
1167       ParsingState::AddrAndLineDelta Delta =
1168           State.handleSpecialOpcode(Opcode, OpcodeOffset);
1169 
1170       if (Verbose)
1171         *OS << "address += " << Delta.Address << ",  line += " << Delta.Line;
1172       EmitRow();
1173       *OffsetPtr = Cursor.tell();
1174     }
1175 
1176     // When a row is added to the matrix, it is also dumped, which includes a
1177     // new line already, so don't add an extra one.
1178     if (Verbose && Rows.size() == RowCount)
1179       *OS << "\n";
1180 
1181     // Most parse failures other than when parsing extended opcodes are due to
1182     // failures to read ULEBs. Bail out of parsing, since we don't know where to
1183     // continue reading from as there is no stated length for such byte
1184     // sequences. Print the final trailing new line if needed before doing so.
1185     if (!Cursor && Opcode != 0) {
1186       if (Verbose)
1187         *OS << "\n";
1188       return Cursor.takeError();
1189     }
1190 
1191     if (!Cursor)
1192       RecoverableErrorHandler(Cursor.takeError());
1193   }
1194 
1195   if (!State.Sequence.Empty)
1196     RecoverableErrorHandler(createStringError(
1197         errc::illegal_byte_sequence,
1198         "last sequence in debug line table at offset 0x%8.8" PRIx64
1199         " is not terminated",
1200         DebugLineOffset));
1201 
1202   // Sort all sequences so that address lookup will work faster.
1203   if (!Sequences.empty()) {
1204     llvm::sort(Sequences, Sequence::orderByHighPC);
1205     // Note: actually, instruction address ranges of sequences should not
1206     // overlap (in shared objects and executables). If they do, the address
1207     // lookup would still work, though, but result would be ambiguous.
1208     // We don't report warning in this case. For example,
1209     // sometimes .so compiled from multiple object files contains a few
1210     // rudimentary sequences for address ranges [0x0, 0xsomething).
1211   }
1212 
1213   // Terminate the table with a final blank line to clearly delineate it from
1214   // later dumps.
1215   if (OS)
1216     *OS << "\n";
1217 
1218   return Error::success();
1219 }
1220 
1221 uint32_t DWARFDebugLine::LineTable::findRowInSeq(
1222     const DWARFDebugLine::Sequence &Seq,
1223     object::SectionedAddress Address) const {
1224   if (!Seq.containsPC(Address))
1225     return UnknownRowIndex;
1226   assert(Seq.SectionIndex == Address.SectionIndex);
1227   // In some cases, e.g. first instruction in a function, the compiler generates
1228   // two entries, both with the same address. We want the last one.
1229   //
1230   // In general we want a non-empty range: the last row whose address is less
1231   // than or equal to Address. This can be computed as upper_bound - 1.
1232   DWARFDebugLine::Row Row;
1233   Row.Address = Address;
1234   RowIter FirstRow = Rows.begin() + Seq.FirstRowIndex;
1235   RowIter LastRow = Rows.begin() + Seq.LastRowIndex;
1236   assert(FirstRow->Address.Address <= Row.Address.Address &&
1237          Row.Address.Address < LastRow[-1].Address.Address);
1238   RowIter RowPos = std::upper_bound(FirstRow + 1, LastRow - 1, Row,
1239                                     DWARFDebugLine::Row::orderByAddress) -
1240                    1;
1241   assert(Seq.SectionIndex == RowPos->Address.SectionIndex);
1242   return RowPos - Rows.begin();
1243 }
1244 
1245 uint32_t DWARFDebugLine::LineTable::lookupAddress(
1246     object::SectionedAddress Address) const {
1247 
1248   // Search for relocatable addresses
1249   uint32_t Result = lookupAddressImpl(Address);
1250 
1251   if (Result != UnknownRowIndex ||
1252       Address.SectionIndex == object::SectionedAddress::UndefSection)
1253     return Result;
1254 
1255   // Search for absolute addresses
1256   Address.SectionIndex = object::SectionedAddress::UndefSection;
1257   return lookupAddressImpl(Address);
1258 }
1259 
1260 uint32_t DWARFDebugLine::LineTable::lookupAddressImpl(
1261     object::SectionedAddress Address) const {
1262   // First, find an instruction sequence containing the given address.
1263   DWARFDebugLine::Sequence Sequence;
1264   Sequence.SectionIndex = Address.SectionIndex;
1265   Sequence.HighPC = Address.Address;
1266   SequenceIter It = llvm::upper_bound(Sequences, Sequence,
1267                                       DWARFDebugLine::Sequence::orderByHighPC);
1268   if (It == Sequences.end() || It->SectionIndex != Address.SectionIndex)
1269     return UnknownRowIndex;
1270   return findRowInSeq(*It, Address);
1271 }
1272 
1273 bool DWARFDebugLine::LineTable::lookupAddressRange(
1274     object::SectionedAddress Address, uint64_t Size,
1275     std::vector<uint32_t> &Result) const {
1276 
1277   // Search for relocatable addresses
1278   if (lookupAddressRangeImpl(Address, Size, Result))
1279     return true;
1280 
1281   if (Address.SectionIndex == object::SectionedAddress::UndefSection)
1282     return false;
1283 
1284   // Search for absolute addresses
1285   Address.SectionIndex = object::SectionedAddress::UndefSection;
1286   return lookupAddressRangeImpl(Address, Size, Result);
1287 }
1288 
1289 bool DWARFDebugLine::LineTable::lookupAddressRangeImpl(
1290     object::SectionedAddress Address, uint64_t Size,
1291     std::vector<uint32_t> &Result) const {
1292   if (Sequences.empty())
1293     return false;
1294   uint64_t EndAddr = Address.Address + Size;
1295   // First, find an instruction sequence containing the given address.
1296   DWARFDebugLine::Sequence Sequence;
1297   Sequence.SectionIndex = Address.SectionIndex;
1298   Sequence.HighPC = Address.Address;
1299   SequenceIter LastSeq = Sequences.end();
1300   SequenceIter SeqPos = llvm::upper_bound(
1301       Sequences, Sequence, DWARFDebugLine::Sequence::orderByHighPC);
1302   if (SeqPos == LastSeq || !SeqPos->containsPC(Address))
1303     return false;
1304 
1305   SequenceIter StartPos = SeqPos;
1306 
1307   // Add the rows from the first sequence to the vector, starting with the
1308   // index we just calculated
1309 
1310   while (SeqPos != LastSeq && SeqPos->LowPC < EndAddr) {
1311     const DWARFDebugLine::Sequence &CurSeq = *SeqPos;
1312     // For the first sequence, we need to find which row in the sequence is the
1313     // first in our range.
1314     uint32_t FirstRowIndex = CurSeq.FirstRowIndex;
1315     if (SeqPos == StartPos)
1316       FirstRowIndex = findRowInSeq(CurSeq, Address);
1317 
1318     // Figure out the last row in the range.
1319     uint32_t LastRowIndex =
1320         findRowInSeq(CurSeq, {EndAddr - 1, Address.SectionIndex});
1321     if (LastRowIndex == UnknownRowIndex)
1322       LastRowIndex = CurSeq.LastRowIndex - 1;
1323 
1324     assert(FirstRowIndex != UnknownRowIndex);
1325     assert(LastRowIndex != UnknownRowIndex);
1326 
1327     for (uint32_t I = FirstRowIndex; I <= LastRowIndex; ++I) {
1328       Result.push_back(I);
1329     }
1330 
1331     ++SeqPos;
1332   }
1333 
1334   return true;
1335 }
1336 
1337 Optional<StringRef> DWARFDebugLine::LineTable::getSourceByIndex(uint64_t FileIndex,
1338                                                                 FileLineInfoKind Kind) const {
1339   if (Kind == FileLineInfoKind::None || !Prologue.hasFileAtIndex(FileIndex))
1340     return None;
1341   const FileNameEntry &Entry = Prologue.getFileNameEntry(FileIndex);
1342   if (auto E = dwarf::toString(Entry.Source))
1343     return StringRef(*E);
1344   return None;
1345 }
1346 
1347 static bool isPathAbsoluteOnWindowsOrPosix(const Twine &Path) {
1348   // Debug info can contain paths from any OS, not necessarily
1349   // an OS we're currently running on. Moreover different compilation units can
1350   // be compiled on different operating systems and linked together later.
1351   return sys::path::is_absolute(Path, sys::path::Style::posix) ||
1352          sys::path::is_absolute(Path, sys::path::Style::windows);
1353 }
1354 
1355 bool DWARFDebugLine::Prologue::getFileNameByIndex(
1356     uint64_t FileIndex, StringRef CompDir, FileLineInfoKind Kind,
1357     std::string &Result, sys::path::Style Style) const {
1358   if (Kind == FileLineInfoKind::None || !hasFileAtIndex(FileIndex))
1359     return false;
1360   const FileNameEntry &Entry = getFileNameEntry(FileIndex);
1361   auto E = dwarf::toString(Entry.Name);
1362   if (!E)
1363     return false;
1364   StringRef FileName = *E;
1365   if (Kind == FileLineInfoKind::RawValue ||
1366       isPathAbsoluteOnWindowsOrPosix(FileName)) {
1367     Result = std::string(FileName);
1368     return true;
1369   }
1370   if (Kind == FileLineInfoKind::BaseNameOnly) {
1371     Result = std::string(llvm::sys::path::filename(FileName));
1372     return true;
1373   }
1374 
1375   SmallString<16> FilePath;
1376   StringRef IncludeDir;
1377   // Be defensive about the contents of Entry.
1378   if (getVersion() >= 5) {
1379     // DirIdx 0 is the compilation directory, so don't include it for
1380     // relative names.
1381     if ((Entry.DirIdx != 0 || Kind != FileLineInfoKind::RelativeFilePath) &&
1382         Entry.DirIdx < IncludeDirectories.size())
1383       IncludeDir = dwarf::toStringRef(IncludeDirectories[Entry.DirIdx]);
1384   } else {
1385     if (0 < Entry.DirIdx && Entry.DirIdx <= IncludeDirectories.size())
1386       IncludeDir = dwarf::toStringRef(IncludeDirectories[Entry.DirIdx - 1]);
1387   }
1388 
1389   // For absolute paths only, include the compilation directory of compile unit.
1390   // We know that FileName is not absolute, the only way to have an absolute
1391   // path at this point would be if IncludeDir is absolute.
1392   if (Kind == FileLineInfoKind::AbsoluteFilePath && !CompDir.empty() &&
1393       !isPathAbsoluteOnWindowsOrPosix(IncludeDir))
1394     sys::path::append(FilePath, Style, CompDir);
1395 
1396   assert((Kind == FileLineInfoKind::AbsoluteFilePath ||
1397           Kind == FileLineInfoKind::RelativeFilePath) &&
1398          "invalid FileLineInfo Kind");
1399 
1400   // sys::path::append skips empty strings.
1401   sys::path::append(FilePath, Style, IncludeDir, FileName);
1402   Result = std::string(FilePath.str());
1403   return true;
1404 }
1405 
1406 bool DWARFDebugLine::LineTable::getFileLineInfoForAddress(
1407     object::SectionedAddress Address, const char *CompDir,
1408     FileLineInfoKind Kind, DILineInfo &Result) const {
1409   // Get the index of row we're looking for in the line table.
1410   uint32_t RowIndex = lookupAddress(Address);
1411   if (RowIndex == -1U)
1412     return false;
1413   // Take file number and line/column from the row.
1414   const auto &Row = Rows[RowIndex];
1415   if (!getFileNameByIndex(Row.File, CompDir, Kind, Result.FileName))
1416     return false;
1417   Result.Line = Row.Line;
1418   Result.Column = Row.Column;
1419   Result.Discriminator = Row.Discriminator;
1420   Result.Source = getSourceByIndex(Row.File, Kind);
1421   return true;
1422 }
1423 
1424 // We want to supply the Unit associated with a .debug_line[.dwo] table when
1425 // we dump it, if possible, but still dump the table even if there isn't a Unit.
1426 // Therefore, collect up handles on all the Units that point into the
1427 // line-table section.
1428 static DWARFDebugLine::SectionParser::LineToUnitMap
1429 buildLineToUnitMap(DWARFUnitVector::iterator_range Units) {
1430   DWARFDebugLine::SectionParser::LineToUnitMap LineToUnit;
1431   for (const auto &U : Units)
1432     if (auto CUDIE = U->getUnitDIE())
1433       if (auto StmtOffset = toSectionOffset(CUDIE.find(DW_AT_stmt_list)))
1434         LineToUnit.insert(std::make_pair(*StmtOffset, &*U));
1435   return LineToUnit;
1436 }
1437 
1438 DWARFDebugLine::SectionParser::SectionParser(
1439     DWARFDataExtractor &Data, const DWARFContext &C,
1440     DWARFUnitVector::iterator_range Units)
1441     : DebugLineData(Data), Context(C) {
1442   LineToUnit = buildLineToUnitMap(Units);
1443   if (!DebugLineData.isValidOffset(Offset))
1444     Done = true;
1445 }
1446 
1447 bool DWARFDebugLine::Prologue::totalLengthIsValid() const {
1448   return TotalLength != 0u;
1449 }
1450 
1451 DWARFDebugLine::LineTable DWARFDebugLine::SectionParser::parseNext(
1452     function_ref<void(Error)> RecoverableErrorHandler,
1453     function_ref<void(Error)> UnrecoverableErrorHandler, raw_ostream *OS,
1454     bool Verbose) {
1455   assert(DebugLineData.isValidOffset(Offset) &&
1456          "parsing should have terminated");
1457   DWARFUnit *U = prepareToParse(Offset);
1458   uint64_t OldOffset = Offset;
1459   LineTable LT;
1460   if (Error Err = LT.parse(DebugLineData, &Offset, Context, U,
1461                            RecoverableErrorHandler, OS, Verbose))
1462     UnrecoverableErrorHandler(std::move(Err));
1463   moveToNextTable(OldOffset, LT.Prologue);
1464   return LT;
1465 }
1466 
1467 void DWARFDebugLine::SectionParser::skip(
1468     function_ref<void(Error)> RecoverableErrorHandler,
1469     function_ref<void(Error)> UnrecoverableErrorHandler) {
1470   assert(DebugLineData.isValidOffset(Offset) &&
1471          "parsing should have terminated");
1472   DWARFUnit *U = prepareToParse(Offset);
1473   uint64_t OldOffset = Offset;
1474   LineTable LT;
1475   if (Error Err = LT.Prologue.parse(DebugLineData, &Offset,
1476                                     RecoverableErrorHandler, Context, U))
1477     UnrecoverableErrorHandler(std::move(Err));
1478   moveToNextTable(OldOffset, LT.Prologue);
1479 }
1480 
1481 DWARFUnit *DWARFDebugLine::SectionParser::prepareToParse(uint64_t Offset) {
1482   DWARFUnit *U = nullptr;
1483   auto It = LineToUnit.find(Offset);
1484   if (It != LineToUnit.end())
1485     U = It->second;
1486   DebugLineData.setAddressSize(U ? U->getAddressByteSize() : 0);
1487   return U;
1488 }
1489 
1490 void DWARFDebugLine::SectionParser::moveToNextTable(uint64_t OldOffset,
1491                                                     const Prologue &P) {
1492   // If the length field is not valid, we don't know where the next table is, so
1493   // cannot continue to parse. Mark the parser as done, and leave the Offset
1494   // value as it currently is. This will be the end of the bad length field.
1495   if (!P.totalLengthIsValid()) {
1496     Done = true;
1497     return;
1498   }
1499 
1500   Offset = OldOffset + P.TotalLength + P.sizeofTotalLength();
1501   if (!DebugLineData.isValidOffset(Offset)) {
1502     Done = true;
1503   }
1504 }
1505