1 //===- DWARFDebugLine.cpp -------------------------------------------------===//
2 //
3 //                     The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 
10 #include "llvm/DebugInfo/DWARF/DWARFDebugLine.h"
11 #include "llvm/ADT/Optional.h"
12 #include "llvm/ADT/SmallString.h"
13 #include "llvm/ADT/SmallVector.h"
14 #include "llvm/ADT/StringRef.h"
15 #include "llvm/BinaryFormat/Dwarf.h"
16 #include "llvm/DebugInfo/DWARF/DWARFFormValue.h"
17 #include "llvm/DebugInfo/DWARF/DWARFRelocMap.h"
18 #include "llvm/Support/Format.h"
19 #include "llvm/Support/Path.h"
20 #include "llvm/Support/raw_ostream.h"
21 #include <algorithm>
22 #include <cassert>
23 #include <cinttypes>
24 #include <cstdint>
25 #include <cstdio>
26 #include <utility>
27 
28 using namespace llvm;
29 using namespace dwarf;
30 
31 using FileLineInfoKind = DILineInfoSpecifier::FileLineInfoKind;
32 
33 namespace {
34 
35 struct ContentDescriptor {
36   dwarf::LineNumberEntryFormat Type;
37   dwarf::Form Form;
38 };
39 
40 using ContentDescriptors = SmallVector<ContentDescriptor, 4>;
41 
42 } // end anonmyous namespace
43 
44 void DWARFDebugLine::ContentTypeTracker::trackContentType(
45     dwarf::LineNumberEntryFormat ContentType) {
46   switch (ContentType) {
47   case dwarf::DW_LNCT_timestamp:
48     HasModTime = true;
49     break;
50   case dwarf::DW_LNCT_size:
51     HasLength = true;
52     break;
53   case dwarf::DW_LNCT_MD5:
54     HasMD5 = true;
55     break;
56   case dwarf::DW_LNCT_LLVM_source:
57     HasSource = true;
58     break;
59   default:
60     // We only care about values we consider optional, and new values may be
61     // added in the vendor extension range, so we do not match exhaustively.
62     break;
63   }
64 }
65 
66 DWARFDebugLine::Prologue::Prologue() { clear(); }
67 
68 void DWARFDebugLine::Prologue::clear() {
69   TotalLength = PrologueLength = 0;
70   SegSelectorSize = 0;
71   MinInstLength = MaxOpsPerInst = DefaultIsStmt = LineBase = LineRange = 0;
72   OpcodeBase = 0;
73   FormParams = dwarf::FormParams({0, 0, DWARF32});
74   ContentTypes = ContentTypeTracker();
75   StandardOpcodeLengths.clear();
76   IncludeDirectories.clear();
77   FileNames.clear();
78 }
79 
80 void DWARFDebugLine::Prologue::dump(raw_ostream &OS,
81                                     DIDumpOptions DumpOptions) const {
82   OS << "Line table prologue:\n"
83      << format("    total_length: 0x%8.8" PRIx64 "\n", TotalLength)
84      << format("         version: %u\n", getVersion());
85   if (getVersion() >= 5)
86     OS << format("    address_size: %u\n", getAddressSize())
87        << format(" seg_select_size: %u\n", SegSelectorSize);
88   OS << format(" prologue_length: 0x%8.8" PRIx64 "\n", PrologueLength)
89      << format(" min_inst_length: %u\n", MinInstLength)
90      << format(getVersion() >= 4 ? "max_ops_per_inst: %u\n" : "", MaxOpsPerInst)
91      << format(" default_is_stmt: %u\n", DefaultIsStmt)
92      << format("       line_base: %i\n", LineBase)
93      << format("      line_range: %u\n", LineRange)
94      << format("     opcode_base: %u\n", OpcodeBase);
95 
96   for (uint32_t I = 0; I != StandardOpcodeLengths.size(); ++I)
97     OS << format("standard_opcode_lengths[%s] = %u\n",
98                  LNStandardString(I + 1).data(), StandardOpcodeLengths[I]);
99 
100   if (!IncludeDirectories.empty()) {
101     // DWARF v5 starts directory indexes at 0.
102     uint32_t DirBase = getVersion() >= 5 ? 0 : 1;
103     for (uint32_t I = 0; I != IncludeDirectories.size(); ++I) {
104       OS << format("include_directories[%3u] = ", I + DirBase);
105       IncludeDirectories[I].dump(OS, DumpOptions);
106       OS << '\n';
107     }
108   }
109 
110   if (!FileNames.empty()) {
111     // DWARF v5 starts file indexes at 0.
112     uint32_t FileBase = getVersion() >= 5 ? 0 : 1;
113     for (uint32_t I = 0; I != FileNames.size(); ++I) {
114       const FileNameEntry &FileEntry = FileNames[I];
115       OS <<   format("file_names[%3u]:\n", I + FileBase);
116       OS <<          "           name: ";
117       FileEntry.Name.dump(OS, DumpOptions);
118       OS << '\n'
119          <<   format("      dir_index: %" PRIu64 "\n", FileEntry.DirIdx);
120       if (ContentTypes.HasMD5)
121         OS <<        "   md5_checksum: " << FileEntry.Checksum.digest() << '\n';
122       if (ContentTypes.HasModTime)
123         OS << format("       mod_time: 0x%8.8" PRIx64 "\n", FileEntry.ModTime);
124       if (ContentTypes.HasLength)
125         OS << format("         length: 0x%8.8" PRIx64 "\n", FileEntry.Length);
126       if (ContentTypes.HasSource) {
127         OS <<        "         source: ";
128         FileEntry.Source.dump(OS, DumpOptions);
129         OS << '\n';
130       }
131     }
132   }
133 }
134 
135 // Parse v2-v4 directory and file tables.
136 static void
137 parseV2DirFileTables(const DWARFDataExtractor &DebugLineData,
138                      uint32_t *OffsetPtr, uint64_t EndPrologueOffset,
139                      DWARFDebugLine::ContentTypeTracker &ContentTypes,
140                      std::vector<DWARFFormValue> &IncludeDirectories,
141                      std::vector<DWARFDebugLine::FileNameEntry> &FileNames) {
142   while (*OffsetPtr < EndPrologueOffset) {
143     StringRef S = DebugLineData.getCStrRef(OffsetPtr);
144     if (S.empty())
145       break;
146     DWARFFormValue Dir(dwarf::DW_FORM_string);
147     Dir.setPValue(S.data());
148     IncludeDirectories.push_back(Dir);
149   }
150 
151   while (*OffsetPtr < EndPrologueOffset) {
152     StringRef Name = DebugLineData.getCStrRef(OffsetPtr);
153     if (Name.empty())
154       break;
155     DWARFDebugLine::FileNameEntry FileEntry;
156     FileEntry.Name.setForm(dwarf::DW_FORM_string);
157     FileEntry.Name.setPValue(Name.data());
158     FileEntry.DirIdx = DebugLineData.getULEB128(OffsetPtr);
159     FileEntry.ModTime = DebugLineData.getULEB128(OffsetPtr);
160     FileEntry.Length = DebugLineData.getULEB128(OffsetPtr);
161     FileNames.push_back(FileEntry);
162   }
163 
164   ContentTypes.HasModTime = true;
165   ContentTypes.HasLength = true;
166 }
167 
168 // Parse v5 directory/file entry content descriptions.
169 // Returns the descriptors, or an empty vector if we did not find a path or
170 // ran off the end of the prologue.
171 static ContentDescriptors
172 parseV5EntryFormat(const DWARFDataExtractor &DebugLineData, uint32_t
173     *OffsetPtr, uint64_t EndPrologueOffset, DWARFDebugLine::ContentTypeTracker
174     *ContentTypes) {
175   ContentDescriptors Descriptors;
176   int FormatCount = DebugLineData.getU8(OffsetPtr);
177   bool HasPath = false;
178   for (int I = 0; I != FormatCount; ++I) {
179     if (*OffsetPtr >= EndPrologueOffset)
180       return ContentDescriptors();
181     ContentDescriptor Descriptor;
182     Descriptor.Type =
183       dwarf::LineNumberEntryFormat(DebugLineData.getULEB128(OffsetPtr));
184     Descriptor.Form = dwarf::Form(DebugLineData.getULEB128(OffsetPtr));
185     if (Descriptor.Type == dwarf::DW_LNCT_path)
186       HasPath = true;
187     if (ContentTypes)
188       ContentTypes->trackContentType(Descriptor.Type);
189     Descriptors.push_back(Descriptor);
190   }
191   return HasPath ? Descriptors : ContentDescriptors();
192 }
193 
194 static bool
195 parseV5DirFileTables(const DWARFDataExtractor &DebugLineData,
196                      uint32_t *OffsetPtr, uint64_t EndPrologueOffset,
197                      const dwarf::FormParams &FormParams,
198                      const DWARFContext &Ctx, const DWARFUnit *U,
199                      DWARFDebugLine::ContentTypeTracker &ContentTypes,
200                      std::vector<DWARFFormValue> &IncludeDirectories,
201                      std::vector<DWARFDebugLine::FileNameEntry> &FileNames) {
202   // Get the directory entry description.
203   ContentDescriptors DirDescriptors =
204       parseV5EntryFormat(DebugLineData, OffsetPtr, EndPrologueOffset, nullptr);
205   if (DirDescriptors.empty())
206     return false;
207 
208   // Get the directory entries, according to the format described above.
209   int DirEntryCount = DebugLineData.getU8(OffsetPtr);
210   for (int I = 0; I != DirEntryCount; ++I) {
211     if (*OffsetPtr >= EndPrologueOffset)
212       return false;
213     for (auto Descriptor : DirDescriptors) {
214       DWARFFormValue Value(Descriptor.Form);
215       switch (Descriptor.Type) {
216       case DW_LNCT_path:
217         if (!Value.extractValue(DebugLineData, OffsetPtr, FormParams, &Ctx, U))
218           return false;
219         IncludeDirectories.push_back(Value);
220         break;
221       default:
222         if (!Value.skipValue(DebugLineData, OffsetPtr, FormParams))
223           return false;
224       }
225     }
226   }
227 
228   // Get the file entry description.
229   ContentDescriptors FileDescriptors =
230       parseV5EntryFormat(DebugLineData, OffsetPtr, EndPrologueOffset,
231           &ContentTypes);
232   if (FileDescriptors.empty())
233     return false;
234 
235   // Get the file entries, according to the format described above.
236   int FileEntryCount = DebugLineData.getU8(OffsetPtr);
237   for (int I = 0; I != FileEntryCount; ++I) {
238     if (*OffsetPtr >= EndPrologueOffset)
239       return false;
240     DWARFDebugLine::FileNameEntry FileEntry;
241     for (auto Descriptor : FileDescriptors) {
242       DWARFFormValue Value(Descriptor.Form);
243       if (!Value.extractValue(DebugLineData, OffsetPtr, FormParams, &Ctx, U))
244         return false;
245       switch (Descriptor.Type) {
246       case DW_LNCT_path:
247         FileEntry.Name = Value;
248         break;
249       case DW_LNCT_LLVM_source:
250         FileEntry.Source = Value;
251         break;
252       case DW_LNCT_directory_index:
253         FileEntry.DirIdx = Value.getAsUnsignedConstant().getValue();
254         break;
255       case DW_LNCT_timestamp:
256         FileEntry.ModTime = Value.getAsUnsignedConstant().getValue();
257         break;
258       case DW_LNCT_size:
259         FileEntry.Length = Value.getAsUnsignedConstant().getValue();
260         break;
261       case DW_LNCT_MD5:
262         assert(Value.getAsBlock().getValue().size() == 16);
263         std::uninitialized_copy_n(Value.getAsBlock().getValue().begin(), 16,
264                                   FileEntry.Checksum.Bytes.begin());
265         break;
266       default:
267         break;
268       }
269     }
270     FileNames.push_back(FileEntry);
271   }
272   return true;
273 }
274 
275 bool DWARFDebugLine::Prologue::parse(const DWARFDataExtractor &DebugLineData,
276                                      uint32_t *OffsetPtr,
277                                      const DWARFContext &Ctx,
278                                      const DWARFUnit *U) {
279   const uint64_t PrologueOffset = *OffsetPtr;
280 
281   clear();
282   TotalLength = DebugLineData.getU32(OffsetPtr);
283   if (TotalLength == UINT32_MAX) {
284     FormParams.Format = dwarf::DWARF64;
285     TotalLength = DebugLineData.getU64(OffsetPtr);
286   } else if (TotalLength >= 0xffffff00) {
287     return false;
288   }
289   FormParams.Version = DebugLineData.getU16(OffsetPtr);
290   if (getVersion() < 2)
291     return false;
292 
293   if (getVersion() >= 5) {
294     FormParams.AddrSize = DebugLineData.getU8(OffsetPtr);
295     assert((DebugLineData.getAddressSize() == 0 ||
296             DebugLineData.getAddressSize() == getAddressSize()) &&
297            "Line table header and data extractor disagree");
298     SegSelectorSize = DebugLineData.getU8(OffsetPtr);
299   }
300 
301   PrologueLength = DebugLineData.getUnsigned(OffsetPtr, sizeofPrologueLength());
302   const uint64_t EndPrologueOffset = PrologueLength + *OffsetPtr;
303   MinInstLength = DebugLineData.getU8(OffsetPtr);
304   if (getVersion() >= 4)
305     MaxOpsPerInst = DebugLineData.getU8(OffsetPtr);
306   DefaultIsStmt = DebugLineData.getU8(OffsetPtr);
307   LineBase = DebugLineData.getU8(OffsetPtr);
308   LineRange = DebugLineData.getU8(OffsetPtr);
309   OpcodeBase = DebugLineData.getU8(OffsetPtr);
310 
311   StandardOpcodeLengths.reserve(OpcodeBase - 1);
312   for (uint32_t I = 1; I < OpcodeBase; ++I) {
313     uint8_t OpLen = DebugLineData.getU8(OffsetPtr);
314     StandardOpcodeLengths.push_back(OpLen);
315   }
316 
317   if (getVersion() >= 5) {
318     if (!parseV5DirFileTables(DebugLineData, OffsetPtr, EndPrologueOffset,
319                               FormParams, Ctx, U, ContentTypes,
320                               IncludeDirectories, FileNames)) {
321       fprintf(stderr,
322               "warning: parsing line table prologue at 0x%8.8" PRIx64
323               " found an invalid directory or file table description at"
324               " 0x%8.8" PRIx64 "\n", PrologueOffset, (uint64_t)*OffsetPtr);
325       return false;
326     }
327   } else
328     parseV2DirFileTables(DebugLineData, OffsetPtr, EndPrologueOffset,
329                          ContentTypes, IncludeDirectories, FileNames);
330 
331   if (*OffsetPtr != EndPrologueOffset) {
332     fprintf(stderr,
333             "warning: parsing line table prologue at 0x%8.8" PRIx64
334             " should have ended at 0x%8.8" PRIx64
335             " but it ended at 0x%8.8" PRIx64 "\n",
336             PrologueOffset, EndPrologueOffset, (uint64_t)*OffsetPtr);
337     return false;
338   }
339   return true;
340 }
341 
342 DWARFDebugLine::Row::Row(bool DefaultIsStmt) { reset(DefaultIsStmt); }
343 
344 void DWARFDebugLine::Row::postAppend() {
345   BasicBlock = false;
346   PrologueEnd = false;
347   EpilogueBegin = false;
348 }
349 
350 void DWARFDebugLine::Row::reset(bool DefaultIsStmt) {
351   Address = 0;
352   Line = 1;
353   Column = 0;
354   File = 1;
355   Isa = 0;
356   Discriminator = 0;
357   IsStmt = DefaultIsStmt;
358   BasicBlock = false;
359   EndSequence = false;
360   PrologueEnd = false;
361   EpilogueBegin = false;
362 }
363 
364 void DWARFDebugLine::Row::dumpTableHeader(raw_ostream &OS) {
365   OS << "Address            Line   Column File   ISA Discriminator Flags\n"
366      << "------------------ ------ ------ ------ --- ------------- "
367         "-------------\n";
368 }
369 
370 void DWARFDebugLine::Row::dump(raw_ostream &OS) const {
371   OS << format("0x%16.16" PRIx64 " %6u %6u", Address, Line, Column)
372      << format(" %6u %3u %13u ", File, Isa, Discriminator)
373      << (IsStmt ? " is_stmt" : "") << (BasicBlock ? " basic_block" : "")
374      << (PrologueEnd ? " prologue_end" : "")
375      << (EpilogueBegin ? " epilogue_begin" : "")
376      << (EndSequence ? " end_sequence" : "") << '\n';
377 }
378 
379 DWARFDebugLine::Sequence::Sequence() { reset(); }
380 
381 void DWARFDebugLine::Sequence::reset() {
382   LowPC = 0;
383   HighPC = 0;
384   FirstRowIndex = 0;
385   LastRowIndex = 0;
386   Empty = true;
387 }
388 
389 DWARFDebugLine::LineTable::LineTable() { clear(); }
390 
391 void DWARFDebugLine::LineTable::dump(raw_ostream &OS,
392                                      DIDumpOptions DumpOptions) const {
393   Prologue.dump(OS, DumpOptions);
394   OS << '\n';
395 
396   if (!Rows.empty()) {
397     Row::dumpTableHeader(OS);
398     for (const Row &R : Rows) {
399       R.dump(OS);
400     }
401   }
402 }
403 
404 void DWARFDebugLine::LineTable::clear() {
405   Prologue.clear();
406   Rows.clear();
407   Sequences.clear();
408 }
409 
410 DWARFDebugLine::ParsingState::ParsingState(struct LineTable *LT)
411     : LineTable(LT) {
412   resetRowAndSequence();
413 }
414 
415 void DWARFDebugLine::ParsingState::resetRowAndSequence() {
416   Row.reset(LineTable->Prologue.DefaultIsStmt);
417   Sequence.reset();
418 }
419 
420 void DWARFDebugLine::ParsingState::appendRowToMatrix(uint32_t Offset) {
421   if (Sequence.Empty) {
422     // Record the beginning of instruction sequence.
423     Sequence.Empty = false;
424     Sequence.LowPC = Row.Address;
425     Sequence.FirstRowIndex = RowNumber;
426   }
427   ++RowNumber;
428   LineTable->appendRow(Row);
429   if (Row.EndSequence) {
430     // Record the end of instruction sequence.
431     Sequence.HighPC = Row.Address;
432     Sequence.LastRowIndex = RowNumber;
433     if (Sequence.isValid())
434       LineTable->appendSequence(Sequence);
435     Sequence.reset();
436   }
437   Row.postAppend();
438 }
439 
440 const DWARFDebugLine::LineTable *
441 DWARFDebugLine::getLineTable(uint32_t Offset) const {
442   LineTableConstIter Pos = LineTableMap.find(Offset);
443   if (Pos != LineTableMap.end())
444     return &Pos->second;
445   return nullptr;
446 }
447 
448 const DWARFDebugLine::LineTable *
449 DWARFDebugLine::getOrParseLineTable(DWARFDataExtractor &DebugLineData,
450                                     uint32_t Offset, const DWARFContext &Ctx,
451                                     const DWARFUnit *U) {
452   if (!DebugLineData.isValidOffset(Offset))
453     return nullptr;
454 
455   std::pair<LineTableIter, bool> Pos =
456       LineTableMap.insert(LineTableMapTy::value_type(Offset, LineTable()));
457   LineTable *LT = &Pos.first->second;
458   if (Pos.second) {
459     if (!LT->parse(DebugLineData, &Offset, Ctx, U))
460       return nullptr;
461   }
462   return LT;
463 }
464 
465 bool DWARFDebugLine::LineTable::parse(DWARFDataExtractor &DebugLineData,
466                                       uint32_t *OffsetPtr,
467                                       const DWARFContext &Ctx,
468                                       const DWARFUnit *U, raw_ostream *OS) {
469   const uint32_t DebugLineOffset = *OffsetPtr;
470 
471   clear();
472 
473   if (!Prologue.parse(DebugLineData, OffsetPtr, Ctx, U)) {
474     // Restore our offset and return false to indicate failure!
475     *OffsetPtr = DebugLineOffset;
476     return false;
477   }
478 
479   if (OS) {
480     // The presence of OS signals verbose dumping.
481     DIDumpOptions DumpOptions;
482     DumpOptions.Verbose = true;
483     Prologue.dump(*OS, DumpOptions);
484   }
485 
486   const uint32_t EndOffset =
487       DebugLineOffset + Prologue.TotalLength + Prologue.sizeofTotalLength();
488 
489   // See if we should tell the data extractor the address size.
490   if (DebugLineData.getAddressSize() == 0)
491     DebugLineData.setAddressSize(Prologue.getAddressSize());
492   else
493     assert(Prologue.getAddressSize() == 0 ||
494            Prologue.getAddressSize() == DebugLineData.getAddressSize());
495 
496   ParsingState State(this);
497 
498   while (*OffsetPtr < EndOffset) {
499     if (OS)
500       *OS << format("0x%08.08" PRIx32 ": ", *OffsetPtr);
501 
502     uint8_t Opcode = DebugLineData.getU8(OffsetPtr);
503 
504     if (OS)
505       *OS << format("%02.02" PRIx8 " ", Opcode);
506 
507     if (Opcode == 0) {
508       // Extended Opcodes always start with a zero opcode followed by
509       // a uleb128 length so you can skip ones you don't know about
510       uint64_t Len = DebugLineData.getULEB128(OffsetPtr);
511       uint32_t ExtOffset = *OffsetPtr;
512 
513       // Tolerate zero-length; assume length is correct and soldier on.
514       if (Len == 0) {
515         if (OS)
516           *OS << "Badly formed extended line op (length 0)\n";
517         continue;
518       }
519 
520       uint8_t SubOpcode = DebugLineData.getU8(OffsetPtr);
521       if (OS)
522         *OS << LNExtendedString(SubOpcode);
523       switch (SubOpcode) {
524       case DW_LNE_end_sequence:
525         // Set the end_sequence register of the state machine to true and
526         // append a row to the matrix using the current values of the
527         // state-machine registers. Then reset the registers to the initial
528         // values specified above. Every statement program sequence must end
529         // with a DW_LNE_end_sequence instruction which creates a row whose
530         // address is that of the byte after the last target machine instruction
531         // of the sequence.
532         State.Row.EndSequence = true;
533         State.appendRowToMatrix(*OffsetPtr);
534         if (OS) {
535           *OS << "\n";
536           OS->indent(12);
537           State.Row.dump(*OS);
538         }
539         State.resetRowAndSequence();
540         break;
541 
542       case DW_LNE_set_address:
543         // Takes a single relocatable address as an operand. The size of the
544         // operand is the size appropriate to hold an address on the target
545         // machine. Set the address register to the value given by the
546         // relocatable address. All of the other statement program opcodes
547         // that affect the address register add a delta to it. This instruction
548         // stores a relocatable value into it instead.
549         //
550         // Make sure the extractor knows the address size.  If not, infer it
551         // from the size of the operand.
552         if (DebugLineData.getAddressSize() == 0)
553           DebugLineData.setAddressSize(Len - 1);
554         else if (DebugLineData.getAddressSize() != Len - 1) {
555           fprintf(stderr, "Mismatching address size at offset 0x%8.8" PRIx32
556                   " expected 0x%2.2" PRIx8 " found 0x%2.2" PRIx64 "\n",
557                   ExtOffset, DebugLineData.getAddressSize(), Len - 1);
558           // Skip the rest of the line-number program.
559           *OffsetPtr = EndOffset;
560           return false;
561         }
562         State.Row.Address = DebugLineData.getRelocatedAddress(OffsetPtr);
563         if (OS)
564           *OS << format(" (0x%16.16" PRIx64 ")", State.Row.Address);
565         break;
566 
567       case DW_LNE_define_file:
568         // Takes 4 arguments. The first is a null terminated string containing
569         // a source file name. The second is an unsigned LEB128 number
570         // representing the directory index of the directory in which the file
571         // was found. The third is an unsigned LEB128 number representing the
572         // time of last modification of the file. The fourth is an unsigned
573         // LEB128 number representing the length in bytes of the file. The time
574         // and length fields may contain LEB128(0) if the information is not
575         // available.
576         //
577         // The directory index represents an entry in the include_directories
578         // section of the statement program prologue. The index is LEB128(0)
579         // if the file was found in the current directory of the compilation,
580         // LEB128(1) if it was found in the first directory in the
581         // include_directories section, and so on. The directory index is
582         // ignored for file names that represent full path names.
583         //
584         // The files are numbered, starting at 1, in the order in which they
585         // appear; the names in the prologue come before names defined by
586         // the DW_LNE_define_file instruction. These numbers are used in the
587         // the file register of the state machine.
588         {
589           FileNameEntry FileEntry;
590           const char *Name = DebugLineData.getCStr(OffsetPtr);
591           FileEntry.Name.setForm(dwarf::DW_FORM_string);
592           FileEntry.Name.setPValue(Name);
593           FileEntry.DirIdx = DebugLineData.getULEB128(OffsetPtr);
594           FileEntry.ModTime = DebugLineData.getULEB128(OffsetPtr);
595           FileEntry.Length = DebugLineData.getULEB128(OffsetPtr);
596           Prologue.FileNames.push_back(FileEntry);
597           if (OS)
598             *OS << " (" << Name << ", dir=" << FileEntry.DirIdx << ", mod_time="
599                 << format("(0x%16.16" PRIx64 ")", FileEntry.ModTime)
600                 << ", length=" << FileEntry.Length << ")";
601         }
602         break;
603 
604       case DW_LNE_set_discriminator:
605         State.Row.Discriminator = DebugLineData.getULEB128(OffsetPtr);
606         if (OS)
607           *OS << " (" << State.Row.Discriminator << ")";
608         break;
609 
610       default:
611         if (OS)
612           *OS << format("Unrecognized extended op 0x%02.02" PRIx8, SubOpcode)
613               << format(" length %" PRIx64, Len);
614         // Len doesn't include the zero opcode byte or the length itself, but
615         // it does include the sub_opcode, so we have to adjust for that.
616         (*OffsetPtr) += Len - 1;
617         break;
618       }
619       // Make sure the stated and parsed lengths are the same.
620       // Otherwise we have an unparseable line-number program.
621       if (*OffsetPtr - ExtOffset != Len) {
622         fprintf(stderr, "Unexpected line op length at offset 0x%8.8" PRIx32
623                 " expected 0x%2.2" PRIx64 " found 0x%2.2" PRIx32 "\n",
624                 ExtOffset, Len, *OffsetPtr - ExtOffset);
625         // Skip the rest of the line-number program.
626         *OffsetPtr = EndOffset;
627         return false;
628       }
629     } else if (Opcode < Prologue.OpcodeBase) {
630       if (OS)
631         *OS << LNStandardString(Opcode);
632       switch (Opcode) {
633       // Standard Opcodes
634       case DW_LNS_copy:
635         // Takes no arguments. Append a row to the matrix using the
636         // current values of the state-machine registers. Then set
637         // the basic_block register to false.
638         State.appendRowToMatrix(*OffsetPtr);
639         if (OS) {
640           *OS << "\n";
641           OS->indent(12);
642           State.Row.dump(*OS);
643           *OS << "\n";
644         }
645         break;
646 
647       case DW_LNS_advance_pc:
648         // Takes a single unsigned LEB128 operand, multiplies it by the
649         // min_inst_length field of the prologue, and adds the
650         // result to the address register of the state machine.
651         {
652           uint64_t AddrOffset =
653               DebugLineData.getULEB128(OffsetPtr) * Prologue.MinInstLength;
654           State.Row.Address += AddrOffset;
655           if (OS)
656             *OS << " (" << AddrOffset << ")";
657         }
658         break;
659 
660       case DW_LNS_advance_line:
661         // Takes a single signed LEB128 operand and adds that value to
662         // the line register of the state machine.
663         State.Row.Line += DebugLineData.getSLEB128(OffsetPtr);
664         if (OS)
665           *OS << " (" << State.Row.Line << ")";
666         break;
667 
668       case DW_LNS_set_file:
669         // Takes a single unsigned LEB128 operand and stores it in the file
670         // register of the state machine.
671         State.Row.File = DebugLineData.getULEB128(OffsetPtr);
672         if (OS)
673           *OS << " (" << State.Row.File << ")";
674         break;
675 
676       case DW_LNS_set_column:
677         // Takes a single unsigned LEB128 operand and stores it in the
678         // column register of the state machine.
679         State.Row.Column = DebugLineData.getULEB128(OffsetPtr);
680         if (OS)
681           *OS << " (" << State.Row.Column << ")";
682         break;
683 
684       case DW_LNS_negate_stmt:
685         // Takes no arguments. Set the is_stmt register of the state
686         // machine to the logical negation of its current value.
687         State.Row.IsStmt = !State.Row.IsStmt;
688         break;
689 
690       case DW_LNS_set_basic_block:
691         // Takes no arguments. Set the basic_block register of the
692         // state machine to true
693         State.Row.BasicBlock = true;
694         break;
695 
696       case DW_LNS_const_add_pc:
697         // Takes no arguments. Add to the address register of the state
698         // machine the address increment value corresponding to special
699         // opcode 255. The motivation for DW_LNS_const_add_pc is this:
700         // when the statement program needs to advance the address by a
701         // small amount, it can use a single special opcode, which occupies
702         // a single byte. When it needs to advance the address by up to
703         // twice the range of the last special opcode, it can use
704         // DW_LNS_const_add_pc followed by a special opcode, for a total
705         // of two bytes. Only if it needs to advance the address by more
706         // than twice that range will it need to use both DW_LNS_advance_pc
707         // and a special opcode, requiring three or more bytes.
708         {
709           uint8_t AdjustOpcode = 255 - Prologue.OpcodeBase;
710           uint64_t AddrOffset =
711               (AdjustOpcode / Prologue.LineRange) * Prologue.MinInstLength;
712           State.Row.Address += AddrOffset;
713           if (OS)
714             *OS
715                 << format(" (0x%16.16" PRIx64 ")", AddrOffset);
716         }
717         break;
718 
719       case DW_LNS_fixed_advance_pc:
720         // Takes a single uhalf operand. Add to the address register of
721         // the state machine the value of the (unencoded) operand. This
722         // is the only extended opcode that takes an argument that is not
723         // a variable length number. The motivation for DW_LNS_fixed_advance_pc
724         // is this: existing assemblers cannot emit DW_LNS_advance_pc or
725         // special opcodes because they cannot encode LEB128 numbers or
726         // judge when the computation of a special opcode overflows and
727         // requires the use of DW_LNS_advance_pc. Such assemblers, however,
728         // can use DW_LNS_fixed_advance_pc instead, sacrificing compression.
729         {
730           uint16_t PCOffset = DebugLineData.getU16(OffsetPtr);
731           State.Row.Address += PCOffset;
732           if (OS)
733             *OS
734                 << format(" (0x%16.16" PRIx64 ")", PCOffset);
735         }
736         break;
737 
738       case DW_LNS_set_prologue_end:
739         // Takes no arguments. Set the prologue_end register of the
740         // state machine to true
741         State.Row.PrologueEnd = true;
742         break;
743 
744       case DW_LNS_set_epilogue_begin:
745         // Takes no arguments. Set the basic_block register of the
746         // state machine to true
747         State.Row.EpilogueBegin = true;
748         break;
749 
750       case DW_LNS_set_isa:
751         // Takes a single unsigned LEB128 operand and stores it in the
752         // column register of the state machine.
753         State.Row.Isa = DebugLineData.getULEB128(OffsetPtr);
754         if (OS)
755           *OS << " (" << State.Row.Isa << ")";
756         break;
757 
758       default:
759         // Handle any unknown standard opcodes here. We know the lengths
760         // of such opcodes because they are specified in the prologue
761         // as a multiple of LEB128 operands for each opcode.
762         {
763           assert(Opcode - 1U < Prologue.StandardOpcodeLengths.size());
764           uint8_t OpcodeLength = Prologue.StandardOpcodeLengths[Opcode - 1];
765           for (uint8_t I = 0; I < OpcodeLength; ++I) {
766             uint64_t Value = DebugLineData.getULEB128(OffsetPtr);
767             if (OS)
768               *OS << format("Skipping ULEB128 value: 0x%16.16" PRIx64 ")\n",
769                             Value);
770           }
771         }
772         break;
773       }
774     } else {
775       // Special Opcodes
776 
777       // A special opcode value is chosen based on the amount that needs
778       // to be added to the line and address registers. The maximum line
779       // increment for a special opcode is the value of the line_base
780       // field in the header, plus the value of the line_range field,
781       // minus 1 (line base + line range - 1). If the desired line
782       // increment is greater than the maximum line increment, a standard
783       // opcode must be used instead of a special opcode. The "address
784       // advance" is calculated by dividing the desired address increment
785       // by the minimum_instruction_length field from the header. The
786       // special opcode is then calculated using the following formula:
787       //
788       //  opcode = (desired line increment - line_base) +
789       //           (line_range * address advance) + opcode_base
790       //
791       // If the resulting opcode is greater than 255, a standard opcode
792       // must be used instead.
793       //
794       // To decode a special opcode, subtract the opcode_base from the
795       // opcode itself to give the adjusted opcode. The amount to
796       // increment the address register is the result of the adjusted
797       // opcode divided by the line_range multiplied by the
798       // minimum_instruction_length field from the header. That is:
799       //
800       //  address increment = (adjusted opcode / line_range) *
801       //                      minimum_instruction_length
802       //
803       // The amount to increment the line register is the line_base plus
804       // the result of the adjusted opcode modulo the line_range. That is:
805       //
806       // line increment = line_base + (adjusted opcode % line_range)
807 
808       uint8_t AdjustOpcode = Opcode - Prologue.OpcodeBase;
809       uint64_t AddrOffset =
810           (AdjustOpcode / Prologue.LineRange) * Prologue.MinInstLength;
811       int32_t LineOffset =
812           Prologue.LineBase + (AdjustOpcode % Prologue.LineRange);
813       State.Row.Line += LineOffset;
814       State.Row.Address += AddrOffset;
815 
816       if (OS) {
817         *OS << "address += " << ((uint32_t)AdjustOpcode)
818             << ",  line += " << LineOffset << "\n";
819         OS->indent(12);
820         State.Row.dump(*OS);
821       }
822 
823       State.appendRowToMatrix(*OffsetPtr);
824       // Reset discriminator to 0.
825       State.Row.Discriminator = 0;
826     }
827     if(OS)
828       *OS << "\n";
829   }
830 
831   if (!State.Sequence.Empty) {
832     fprintf(stderr, "warning: last sequence in debug line table is not"
833                     "terminated!\n");
834   }
835 
836   // Sort all sequences so that address lookup will work faster.
837   if (!Sequences.empty()) {
838     llvm::sort(Sequences.begin(), Sequences.end(), Sequence::orderByLowPC);
839     // Note: actually, instruction address ranges of sequences should not
840     // overlap (in shared objects and executables). If they do, the address
841     // lookup would still work, though, but result would be ambiguous.
842     // We don't report warning in this case. For example,
843     // sometimes .so compiled from multiple object files contains a few
844     // rudimentary sequences for address ranges [0x0, 0xsomething).
845   }
846 
847   return EndOffset;
848 }
849 
850 uint32_t
851 DWARFDebugLine::LineTable::findRowInSeq(const DWARFDebugLine::Sequence &Seq,
852                                         uint64_t Address) const {
853   if (!Seq.containsPC(Address))
854     return UnknownRowIndex;
855   // Search for instruction address in the rows describing the sequence.
856   // Rows are stored in a vector, so we may use arithmetical operations with
857   // iterators.
858   DWARFDebugLine::Row Row;
859   Row.Address = Address;
860   RowIter FirstRow = Rows.begin() + Seq.FirstRowIndex;
861   RowIter LastRow = Rows.begin() + Seq.LastRowIndex;
862   LineTable::RowIter RowPos = std::lower_bound(
863       FirstRow, LastRow, Row, DWARFDebugLine::Row::orderByAddress);
864   if (RowPos == LastRow) {
865     return Seq.LastRowIndex - 1;
866   }
867   uint32_t Index = Seq.FirstRowIndex + (RowPos - FirstRow);
868   if (RowPos->Address > Address) {
869     if (RowPos == FirstRow)
870       return UnknownRowIndex;
871     else
872       Index--;
873   }
874   return Index;
875 }
876 
877 uint32_t DWARFDebugLine::LineTable::lookupAddress(uint64_t Address) const {
878   if (Sequences.empty())
879     return UnknownRowIndex;
880   // First, find an instruction sequence containing the given address.
881   DWARFDebugLine::Sequence Sequence;
882   Sequence.LowPC = Address;
883   SequenceIter FirstSeq = Sequences.begin();
884   SequenceIter LastSeq = Sequences.end();
885   SequenceIter SeqPos = std::lower_bound(
886       FirstSeq, LastSeq, Sequence, DWARFDebugLine::Sequence::orderByLowPC);
887   DWARFDebugLine::Sequence FoundSeq;
888   if (SeqPos == LastSeq) {
889     FoundSeq = Sequences.back();
890   } else if (SeqPos->LowPC == Address) {
891     FoundSeq = *SeqPos;
892   } else {
893     if (SeqPos == FirstSeq)
894       return UnknownRowIndex;
895     FoundSeq = *(SeqPos - 1);
896   }
897   return findRowInSeq(FoundSeq, Address);
898 }
899 
900 bool DWARFDebugLine::LineTable::lookupAddressRange(
901     uint64_t Address, uint64_t Size, std::vector<uint32_t> &Result) const {
902   if (Sequences.empty())
903     return false;
904   uint64_t EndAddr = Address + Size;
905   // First, find an instruction sequence containing the given address.
906   DWARFDebugLine::Sequence Sequence;
907   Sequence.LowPC = Address;
908   SequenceIter FirstSeq = Sequences.begin();
909   SequenceIter LastSeq = Sequences.end();
910   SequenceIter SeqPos = std::lower_bound(
911       FirstSeq, LastSeq, Sequence, DWARFDebugLine::Sequence::orderByLowPC);
912   if (SeqPos == LastSeq || SeqPos->LowPC != Address) {
913     if (SeqPos == FirstSeq)
914       return false;
915     SeqPos--;
916   }
917   if (!SeqPos->containsPC(Address))
918     return false;
919 
920   SequenceIter StartPos = SeqPos;
921 
922   // Add the rows from the first sequence to the vector, starting with the
923   // index we just calculated
924 
925   while (SeqPos != LastSeq && SeqPos->LowPC < EndAddr) {
926     const DWARFDebugLine::Sequence &CurSeq = *SeqPos;
927     // For the first sequence, we need to find which row in the sequence is the
928     // first in our range.
929     uint32_t FirstRowIndex = CurSeq.FirstRowIndex;
930     if (SeqPos == StartPos)
931       FirstRowIndex = findRowInSeq(CurSeq, Address);
932 
933     // Figure out the last row in the range.
934     uint32_t LastRowIndex = findRowInSeq(CurSeq, EndAddr - 1);
935     if (LastRowIndex == UnknownRowIndex)
936       LastRowIndex = CurSeq.LastRowIndex - 1;
937 
938     assert(FirstRowIndex != UnknownRowIndex);
939     assert(LastRowIndex != UnknownRowIndex);
940 
941     for (uint32_t I = FirstRowIndex; I <= LastRowIndex; ++I) {
942       Result.push_back(I);
943     }
944 
945     ++SeqPos;
946   }
947 
948   return true;
949 }
950 
951 bool DWARFDebugLine::LineTable::hasFileAtIndex(uint64_t FileIndex) const {
952   return FileIndex != 0 && FileIndex <= Prologue.FileNames.size();
953 }
954 
955 Optional<StringRef> DWARFDebugLine::LineTable::getSourceByIndex(uint64_t FileIndex,
956                                                                 FileLineInfoKind Kind) const {
957   if (Kind == FileLineInfoKind::None || !hasFileAtIndex(FileIndex))
958     return None;
959   const FileNameEntry &Entry = Prologue.FileNames[FileIndex - 1];
960   if (Optional<const char *> source = Entry.Source.getAsCString())
961     return StringRef(*source);
962   return None;
963 }
964 
965 static bool isPathAbsoluteOnWindowsOrPosix(const Twine &Path) {
966   // Debug info can contain paths from any OS, not necessarily
967   // an OS we're currently running on. Moreover different compilation units can
968   // be compiled on different operating systems and linked together later.
969   return sys::path::is_absolute(Path, sys::path::Style::posix) ||
970          sys::path::is_absolute(Path, sys::path::Style::windows);
971 }
972 
973 bool DWARFDebugLine::LineTable::getFileNameByIndex(uint64_t FileIndex,
974                                                    const char *CompDir,
975                                                    FileLineInfoKind Kind,
976                                                    std::string &Result) const {
977   if (Kind == FileLineInfoKind::None || !hasFileAtIndex(FileIndex))
978     return false;
979   const FileNameEntry &Entry = Prologue.FileNames[FileIndex - 1];
980   StringRef FileName = Entry.Name.getAsCString().getValue();
981   if (Kind != FileLineInfoKind::AbsoluteFilePath ||
982       isPathAbsoluteOnWindowsOrPosix(FileName)) {
983     Result = FileName;
984     return true;
985   }
986 
987   SmallString<16> FilePath;
988   uint64_t IncludeDirIndex = Entry.DirIdx;
989   StringRef IncludeDir;
990   // Be defensive about the contents of Entry.
991   if (IncludeDirIndex > 0 &&
992       IncludeDirIndex <= Prologue.IncludeDirectories.size())
993     IncludeDir = Prologue.IncludeDirectories[IncludeDirIndex - 1]
994                      .getAsCString()
995                      .getValue();
996 
997   // We may still need to append compilation directory of compile unit.
998   // We know that FileName is not absolute, the only way to have an
999   // absolute path at this point would be if IncludeDir is absolute.
1000   if (CompDir && Kind == FileLineInfoKind::AbsoluteFilePath &&
1001       !isPathAbsoluteOnWindowsOrPosix(IncludeDir))
1002     sys::path::append(FilePath, CompDir);
1003 
1004   // sys::path::append skips empty strings.
1005   sys::path::append(FilePath, IncludeDir, FileName);
1006   Result = FilePath.str();
1007   return true;
1008 }
1009 
1010 bool DWARFDebugLine::LineTable::getFileLineInfoForAddress(
1011     uint64_t Address, const char *CompDir, FileLineInfoKind Kind,
1012     DILineInfo &Result) const {
1013   // Get the index of row we're looking for in the line table.
1014   uint32_t RowIndex = lookupAddress(Address);
1015   if (RowIndex == -1U)
1016     return false;
1017   // Take file number and line/column from the row.
1018   const auto &Row = Rows[RowIndex];
1019   if (!getFileNameByIndex(Row.File, CompDir, Kind, Result.FileName))
1020     return false;
1021   Result.Line = Row.Line;
1022   Result.Column = Row.Column;
1023   Result.Discriminator = Row.Discriminator;
1024   Result.Source = getSourceByIndex(Row.File, Kind);
1025   return true;
1026 }
1027