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/SmallString.h"
11#include "llvm/ADT/SmallVector.h"
12#include "llvm/ADT/StringRef.h"
13#include "llvm/BinaryFormat/Dwarf.h"
14#include "llvm/DebugInfo/DWARF/DWARFDataExtractor.h"
15#include "llvm/DebugInfo/DWARF/DWARFDie.h"
16#include "llvm/DebugInfo/DWARF/DWARFFormValue.h"
17#include "llvm/Support/Errc.h"
18#include "llvm/Support/Format.h"
19#include "llvm/Support/FormatVariadic.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
28using namespace llvm;
29using namespace dwarf;
30
31using FileLineInfoKind = DILineInfoSpecifier::FileLineInfoKind;
32
33namespace {
34
35struct ContentDescriptor {
36 dwarf::LineNumberEntryFormat Type;
37 dwarf::Form Form;
38};
39
40using ContentDescriptors = SmallVector<ContentDescriptor, 4>;
41
42} // end anonymous namespace
43
44static bool versionIsSupported(uint16_t Version) {
45 return Version >= 2 && Version <= 5;
46}
47
48void DWARFDebugLine::ContentTypeTracker::trackContentType(
49 dwarf::LineNumberEntryFormat ContentType) {
50 switch (ContentType) {
51 case dwarf::DW_LNCT_timestamp:
52 HasModTime = true;
53 break;
54 case dwarf::DW_LNCT_size:
55 HasLength = true;
56 break;
57 case dwarf::DW_LNCT_MD5:
58 HasMD5 = true;
59 break;
60 case dwarf::DW_LNCT_LLVM_source:
61 HasSource = true;
62 break;
63 default:
64 // We only care about values we consider optional, and new values may be
65 // added in the vendor extension range, so we do not match exhaustively.
66 break;
67 }
68}
69
70DWARFDebugLine::Prologue::Prologue() { clear(); }
71
72bool DWARFDebugLine::Prologue::hasFileAtIndex(uint64_t FileIndex) const {
73 uint16_t DwarfVersion = getVersion();
74 assert(DwarfVersion != 0 &&
75 "line table prologue has no dwarf version information");
76 if (DwarfVersion >= 5)
77 return FileIndex < FileNames.size();
78 return FileIndex != 0 && FileIndex <= FileNames.size();
79}
80
81std::optional<uint64_t>
82DWARFDebugLine::Prologue::getLastValidFileIndex() const {
83 if (FileNames.empty())
84 return std::nullopt;
85 uint16_t DwarfVersion = getVersion();
86 assert(DwarfVersion != 0 &&
87 "line table prologue has no dwarf version information");
88 // In DWARF v5 the file names are 0-indexed.
89 if (DwarfVersion >= 5)
90 return FileNames.size() - 1;
91 return FileNames.size();
92}
93
94const llvm::DWARFDebugLine::FileNameEntry &
95DWARFDebugLine::Prologue::getFileNameEntry(uint64_t Index) const {
96 uint16_t DwarfVersion = getVersion();
97 assert(DwarfVersion != 0 &&
98 "line table prologue has no dwarf version information");
99 // In DWARF v5 the file names are 0-indexed.
100 if (DwarfVersion >= 5)
101 return FileNames[Index];
102 return FileNames[Index - 1];
103}
104
105void DWARFDebugLine::Prologue::clear() {
106 TotalLength = PrologueLength = 0;
107 SegSelectorSize = 0;
108 MinInstLength = MaxOpsPerInst = DefaultIsStmt = LineBase = LineRange = 0;
109 OpcodeBase = 0;
110 FormParams = dwarf::FormParams({.Version: 0, .AddrSize: 0, .Format: DWARF32});
111 ContentTypes = ContentTypeTracker();
112 StandardOpcodeLengths.clear();
113 IncludeDirectories.clear();
114 FileNames.clear();
115}
116
117void DWARFDebugLine::Prologue::dump(raw_ostream &OS,
118 DIDumpOptions DumpOptions) const {
119 if (!totalLengthIsValid())
120 return;
121 int OffsetDumpWidth = 2 * dwarf::getDwarfOffsetByteSize(Format: FormParams.Format);
122 OS << "Line table prologue:\n"
123 << format(Fmt: " total_length: 0x%0*" PRIx64 "\n", Vals: OffsetDumpWidth,
124 Vals: TotalLength)
125 << " format: " << dwarf::FormatString(Format: FormParams.Format) << "\n"
126 << format(Fmt: " version: %u\n", Vals: getVersion());
127 if (!versionIsSupported(Version: getVersion()))
128 return;
129 if (getVersion() >= 5)
130 OS << format(Fmt: " address_size: %u\n", Vals: getAddressSize())
131 << format(Fmt: " seg_select_size: %u\n", Vals: SegSelectorSize);
132 OS << format(Fmt: " prologue_length: 0x%0*" PRIx64 "\n", Vals: OffsetDumpWidth,
133 Vals: PrologueLength)
134 << format(Fmt: " min_inst_length: %u\n", Vals: MinInstLength)
135 << format(Fmt: getVersion() >= 4 ? "max_ops_per_inst: %u\n" : "", Vals: MaxOpsPerInst)
136 << format(Fmt: " default_is_stmt: %u\n", Vals: DefaultIsStmt)
137 << format(Fmt: " line_base: %i\n", Vals: LineBase)
138 << format(Fmt: " line_range: %u\n", Vals: LineRange)
139 << format(Fmt: " opcode_base: %u\n", Vals: OpcodeBase);
140
141 for (uint32_t I = 0; I != StandardOpcodeLengths.size(); ++I)
142 OS << formatv(Fmt: "standard_opcode_lengths[{0}] = {1}\n",
143 Vals: static_cast<dwarf::LineNumberOps>(I + 1),
144 Vals: StandardOpcodeLengths[I]);
145
146 if (!IncludeDirectories.empty()) {
147 // DWARF v5 starts directory indexes at 0.
148 uint32_t DirBase = getVersion() >= 5 ? 0 : 1;
149 for (uint32_t I = 0; I != IncludeDirectories.size(); ++I) {
150 OS << format(Fmt: "include_directories[%3u] = ", Vals: I + DirBase);
151 IncludeDirectories[I].dump(OS, DumpOpts: DumpOptions);
152 OS << '\n';
153 }
154 }
155
156 if (!FileNames.empty()) {
157 // DWARF v5 starts file indexes at 0.
158 uint32_t FileBase = getVersion() >= 5 ? 0 : 1;
159 for (uint32_t I = 0; I != FileNames.size(); ++I) {
160 const FileNameEntry &FileEntry = FileNames[I];
161 OS << format(Fmt: "file_names[%3u]:\n", Vals: I + FileBase);
162 OS << " name: ";
163 FileEntry.Name.dump(OS, DumpOpts: DumpOptions);
164 OS << '\n'
165 << format(Fmt: " dir_index: %" PRIu64 "\n", Vals: FileEntry.DirIdx);
166 if (ContentTypes.HasMD5)
167 OS << " md5_checksum: " << FileEntry.Checksum.digest() << '\n';
168 if (ContentTypes.HasModTime)
169 OS << format(Fmt: " mod_time: 0x%8.8" PRIx64 "\n", Vals: FileEntry.ModTime);
170 if (ContentTypes.HasLength)
171 OS << format(Fmt: " length: 0x%8.8" PRIx64 "\n", Vals: FileEntry.Length);
172 if (ContentTypes.HasSource) {
173 auto Source = FileEntry.Source.getAsCString();
174 if (!Source)
175 consumeError(Err: Source.takeError());
176 else if ((*Source)[0]) {
177 OS << " source: ";
178 FileEntry.Source.dump(OS, DumpOpts: DumpOptions);
179 OS << '\n';
180 }
181 }
182 }
183 }
184}
185
186// Parse v2-v4 directory and file tables.
187static Error
188parseV2DirFileTables(const DWARFDataExtractor &DebugLineData,
189 uint64_t *OffsetPtr,
190 DWARFDebugLine::ContentTypeTracker &ContentTypes,
191 std::vector<DWARFFormValue> &IncludeDirectories,
192 std::vector<DWARFDebugLine::FileNameEntry> &FileNames) {
193 while (true) {
194 Error Err = Error::success();
195 StringRef S = DebugLineData.getCStrRef(OffsetPtr, Err: &Err);
196 if (Err) {
197 consumeError(Err: std::move(Err));
198 return createStringError(EC: errc::invalid_argument,
199 S: "include directories table was not null "
200 "terminated before the end of the prologue");
201 }
202 if (S.empty())
203 break;
204 DWARFFormValue Dir =
205 DWARFFormValue::createFromPValue(F: dwarf::DW_FORM_string, V: S.data());
206 IncludeDirectories.push_back(x: Dir);
207 }
208
209 ContentTypes.HasModTime = true;
210 ContentTypes.HasLength = true;
211
212 while (true) {
213 Error Err = Error::success();
214 StringRef Name = DebugLineData.getCStrRef(OffsetPtr, Err: &Err);
215 if (!Err && Name.empty())
216 break;
217
218 DWARFDebugLine::FileNameEntry FileEntry;
219 FileEntry.Name =
220 DWARFFormValue::createFromPValue(F: dwarf::DW_FORM_string, V: Name.data());
221 FileEntry.DirIdx = DebugLineData.getULEB128(offset_ptr: OffsetPtr, Err: &Err);
222 FileEntry.ModTime = DebugLineData.getULEB128(offset_ptr: OffsetPtr, Err: &Err);
223 FileEntry.Length = DebugLineData.getULEB128(offset_ptr: OffsetPtr, Err: &Err);
224
225 if (Err) {
226 consumeError(Err: std::move(Err));
227 return createStringError(
228 EC: errc::invalid_argument,
229 S: "file names table was not null terminated before "
230 "the end of the prologue");
231 }
232 FileNames.push_back(x: FileEntry);
233 }
234
235 return Error::success();
236}
237
238// Parse v5 directory/file entry content descriptions.
239// Returns the descriptors, or an error if we did not find a path or ran off
240// the end of the prologue.
241static llvm::Expected<ContentDescriptors>
242parseV5EntryFormat(const DWARFDataExtractor &DebugLineData, uint64_t *OffsetPtr,
243 DWARFDebugLine::ContentTypeTracker *ContentTypes) {
244 Error Err = Error::success();
245 ContentDescriptors Descriptors;
246 int FormatCount = DebugLineData.getU8(offset_ptr: OffsetPtr, Err: &Err);
247 bool HasPath = false;
248 for (int I = 0; I != FormatCount && !Err; ++I) {
249 ContentDescriptor Descriptor;
250 Descriptor.Type =
251 dwarf::LineNumberEntryFormat(DebugLineData.getULEB128(offset_ptr: OffsetPtr, Err: &Err));
252 Descriptor.Form = dwarf::Form(DebugLineData.getULEB128(offset_ptr: OffsetPtr, Err: &Err));
253 if (Descriptor.Type == dwarf::DW_LNCT_path)
254 HasPath = true;
255 if (ContentTypes)
256 ContentTypes->trackContentType(ContentType: Descriptor.Type);
257 Descriptors.push_back(Elt: Descriptor);
258 }
259
260 if (Err)
261 return createStringError(EC: errc::invalid_argument,
262 Fmt: "failed to parse entry content descriptors: %s",
263 Vals: toString(E: std::move(Err)).c_str());
264
265 if (!HasPath)
266 return createStringError(EC: errc::invalid_argument,
267 S: "failed to parse entry content descriptions"
268 " because no path was found");
269 return Descriptors;
270}
271
272static Error
273parseV5DirFileTables(const DWARFDataExtractor &DebugLineData,
274 uint64_t *OffsetPtr, const dwarf::FormParams &FormParams,
275 const DWARFContext &Ctx, const DWARFUnit *U,
276 DWARFDebugLine::ContentTypeTracker &ContentTypes,
277 std::vector<DWARFFormValue> &IncludeDirectories,
278 std::vector<DWARFDebugLine::FileNameEntry> &FileNames) {
279 // Get the directory entry description.
280 llvm::Expected<ContentDescriptors> DirDescriptors =
281 parseV5EntryFormat(DebugLineData, OffsetPtr, ContentTypes: nullptr);
282 if (!DirDescriptors)
283 return DirDescriptors.takeError();
284
285 // Get the directory entries, according to the format described above.
286 uint64_t DirEntryCount = DebugLineData.getULEB128(offset_ptr: OffsetPtr);
287 for (uint64_t I = 0; I != DirEntryCount; ++I) {
288 for (auto Descriptor : *DirDescriptors) {
289 DWARFFormValue Value(Descriptor.Form);
290 switch (Descriptor.Type) {
291 case DW_LNCT_path:
292 if (!Value.extractValue(Data: DebugLineData, OffsetPtr, FormParams, Context: &Ctx, Unit: U))
293 return createStringError(EC: errc::invalid_argument,
294 S: "failed to parse directory entry because "
295 "extracting the form value failed");
296 IncludeDirectories.push_back(x: Value);
297 break;
298 default:
299 if (!Value.skipValue(DebugInfoData: DebugLineData, OffsetPtr, Params: FormParams))
300 return createStringError(EC: errc::invalid_argument,
301 S: "failed to parse directory entry because "
302 "skipping the form value failed");
303 }
304 }
305 }
306
307 // Get the file entry description.
308 llvm::Expected<ContentDescriptors> FileDescriptors =
309 parseV5EntryFormat(DebugLineData, OffsetPtr, ContentTypes: &ContentTypes);
310 if (!FileDescriptors)
311 return FileDescriptors.takeError();
312
313 // Get the file entries, according to the format described above.
314 uint64_t FileEntryCount = DebugLineData.getULEB128(offset_ptr: OffsetPtr);
315 for (uint64_t I = 0; I != FileEntryCount; ++I) {
316 DWARFDebugLine::FileNameEntry FileEntry;
317 for (auto Descriptor : *FileDescriptors) {
318 DWARFFormValue Value(Descriptor.Form);
319 if (!Value.extractValue(Data: DebugLineData, OffsetPtr, FormParams, Context: &Ctx, Unit: U))
320 return createStringError(EC: errc::invalid_argument,
321 S: "failed to parse file entry because "
322 "extracting the form value failed");
323 switch (Descriptor.Type) {
324 case DW_LNCT_path:
325 FileEntry.Name = Value;
326 break;
327 case DW_LNCT_LLVM_source:
328 FileEntry.Source = Value;
329 break;
330 case DW_LNCT_directory_index:
331 FileEntry.DirIdx = *Value.getAsUnsignedConstant();
332 break;
333 case DW_LNCT_timestamp:
334 FileEntry.ModTime = *Value.getAsUnsignedConstant();
335 break;
336 case DW_LNCT_size:
337 FileEntry.Length = *Value.getAsUnsignedConstant();
338 break;
339 case DW_LNCT_MD5:
340 if (!Value.getAsBlock() || Value.getAsBlock()->size() != 16)
341 return createStringError(
342 EC: errc::invalid_argument,
343 S: "failed to parse file entry because the MD5 hash is invalid");
344 std::uninitialized_copy_n(first: Value.getAsBlock()->begin(), n: 16,
345 result: FileEntry.Checksum.begin());
346 break;
347 default:
348 break;
349 }
350 }
351 FileNames.push_back(x: FileEntry);
352 }
353 return Error::success();
354}
355
356uint64_t DWARFDebugLine::Prologue::getLength() const {
357 uint64_t Length = PrologueLength + sizeofTotalLength() +
358 sizeof(getVersion()) + sizeofPrologueLength();
359 if (getVersion() >= 5)
360 Length += 2; // Address + Segment selector sizes.
361 return Length;
362}
363
364Error DWARFDebugLine::Prologue::parse(
365 DWARFDataExtractor DebugLineData, uint64_t *OffsetPtr,
366 function_ref<void(Error)> RecoverableErrorHandler, const DWARFContext &Ctx,
367 const DWARFUnit *U) {
368 const uint64_t PrologueOffset = *OffsetPtr;
369
370 clear();
371 DataExtractor::Cursor Cursor(*OffsetPtr);
372 std::tie(args&: TotalLength, args&: FormParams.Format) =
373 DebugLineData.getInitialLength(C&: Cursor);
374
375 DebugLineData =
376 DWARFDataExtractor(DebugLineData, Cursor.tell() + TotalLength);
377 FormParams.Version = DebugLineData.getU16(C&: Cursor);
378 if (Cursor && !versionIsSupported(Version: getVersion())) {
379 // Treat this error as unrecoverable - we cannot be sure what any of
380 // the data represents including the length field, so cannot skip it or make
381 // any reasonable assumptions.
382 *OffsetPtr = Cursor.tell();
383 return createStringError(
384 EC: errc::not_supported,
385 Fmt: "parsing line table prologue at offset 0x%8.8" PRIx64
386 ": unsupported version %" PRIu16,
387 Vals: PrologueOffset, Vals: getVersion());
388 }
389
390 if (getVersion() >= 5) {
391 FormParams.AddrSize = DebugLineData.getU8(C&: Cursor);
392 const uint8_t DataAddrSize = DebugLineData.getAddressSize();
393 const uint8_t PrologueAddrSize = getAddressSize();
394 if (Cursor) {
395 if (DataAddrSize == 0) {
396 if (PrologueAddrSize != 4 && PrologueAddrSize != 8) {
397 RecoverableErrorHandler(createStringError(
398 EC: errc::not_supported,
399 Fmt: "parsing line table prologue at offset 0x%8.8" PRIx64
400 ": invalid address size %" PRIu8,
401 Vals: PrologueOffset, Vals: PrologueAddrSize));
402 }
403 } else if (DataAddrSize != PrologueAddrSize) {
404 RecoverableErrorHandler(createStringError(
405 EC: errc::not_supported,
406 Fmt: "parsing line table prologue at offset 0x%8.8" PRIx64 ": address "
407 "size %" PRIu8 " doesn't match architecture address size %" PRIu8,
408 Vals: PrologueOffset, Vals: PrologueAddrSize, Vals: DataAddrSize));
409 }
410 }
411 SegSelectorSize = DebugLineData.getU8(C&: Cursor);
412 }
413
414 PrologueLength =
415 DebugLineData.getRelocatedValue(C&: Cursor, Size: sizeofPrologueLength());
416 const uint64_t EndPrologueOffset = PrologueLength + Cursor.tell();
417 DebugLineData = DWARFDataExtractor(DebugLineData, EndPrologueOffset);
418 MinInstLength = DebugLineData.getU8(C&: Cursor);
419 if (getVersion() >= 4)
420 MaxOpsPerInst = DebugLineData.getU8(C&: Cursor);
421 DefaultIsStmt = DebugLineData.getU8(C&: Cursor);
422 LineBase = DebugLineData.getU8(C&: Cursor);
423 LineRange = DebugLineData.getU8(C&: Cursor);
424 OpcodeBase = DebugLineData.getU8(C&: Cursor);
425
426 if (Cursor && OpcodeBase == 0) {
427 // If the opcode base is 0, we cannot read the standard opcode lengths (of
428 // which there are supposed to be one fewer than the opcode base). Assume
429 // there are no standard opcodes and continue parsing.
430 RecoverableErrorHandler(createStringError(
431 EC: errc::invalid_argument,
432 Fmt: "parsing line table prologue at offset 0x%8.8" PRIx64
433 " found opcode base of 0. Assuming no standard opcodes",
434 Vals: PrologueOffset));
435 } else if (Cursor) {
436 StandardOpcodeLengths.reserve(n: OpcodeBase - 1);
437 for (uint32_t I = 1; I < OpcodeBase; ++I) {
438 uint8_t OpLen = DebugLineData.getU8(C&: Cursor);
439 StandardOpcodeLengths.push_back(x: OpLen);
440 }
441 }
442
443 *OffsetPtr = Cursor.tell();
444 // A corrupt file name or directory table does not prevent interpretation of
445 // the main line program, so check the cursor state now so that its errors can
446 // be handled separately.
447 if (!Cursor)
448 return createStringError(
449 EC: errc::invalid_argument,
450 Fmt: "parsing line table prologue at offset 0x%8.8" PRIx64 ": %s",
451 Vals: PrologueOffset, Vals: toString(E: Cursor.takeError()).c_str());
452
453 Error E =
454 getVersion() >= 5
455 ? parseV5DirFileTables(DebugLineData, OffsetPtr, FormParams, Ctx, U,
456 ContentTypes, IncludeDirectories, FileNames)
457 : parseV2DirFileTables(DebugLineData, OffsetPtr, ContentTypes,
458 IncludeDirectories, FileNames);
459 if (E) {
460 RecoverableErrorHandler(joinErrors(
461 E1: createStringError(
462 EC: errc::invalid_argument,
463 Fmt: "parsing line table prologue at 0x%8.8" PRIx64
464 " found an invalid directory or file table description at"
465 " 0x%8.8" PRIx64,
466 Vals: PrologueOffset, Vals: *OffsetPtr),
467 E2: std::move(E)));
468 return Error::success();
469 }
470
471 assert(*OffsetPtr <= EndPrologueOffset);
472 if (*OffsetPtr != EndPrologueOffset) {
473 RecoverableErrorHandler(createStringError(
474 EC: errc::invalid_argument,
475 Fmt: "unknown data in line table prologue at offset 0x%8.8" PRIx64
476 ": parsing ended (at offset 0x%8.8" PRIx64
477 ") before reaching the prologue end at offset 0x%8.8" PRIx64,
478 Vals: PrologueOffset, Vals: *OffsetPtr, Vals: EndPrologueOffset));
479 }
480 return Error::success();
481}
482
483DWARFDebugLine::Row::Row(bool DefaultIsStmt) { reset(DefaultIsStmt); }
484
485void DWARFDebugLine::Row::postAppend() {
486 Discriminator = 0;
487 BasicBlock = false;
488 PrologueEnd = false;
489 EpilogueBegin = false;
490}
491
492void DWARFDebugLine::Row::reset(bool DefaultIsStmt) {
493 Address.Address = 0;
494 Address.SectionIndex = object::SectionedAddress::UndefSection;
495 Line = 1;
496 Column = 0;
497 File = 1;
498 Isa = 0;
499 Discriminator = 0;
500 IsStmt = DefaultIsStmt;
501 OpIndex = 0;
502 BasicBlock = false;
503 EndSequence = false;
504 PrologueEnd = false;
505 EpilogueBegin = false;
506}
507
508void DWARFDebugLine::Row::dumpTableHeader(raw_ostream &OS, unsigned Indent) {
509 OS.indent(NumSpaces: Indent)
510 << "Address Line Column File ISA Discriminator OpIndex "
511 "Flags\n";
512 OS.indent(NumSpaces: Indent)
513 << "------------------ ------ ------ ------ --- ------------- ------- "
514 "-------------\n";
515}
516
517void DWARFDebugLine::Row::dump(raw_ostream &OS) const {
518 OS << format(Fmt: "0x%16.16" PRIx64 " %6u %6u", Vals: Address.Address, Vals: Line, Vals: Column)
519 << format(Fmt: " %6u %3u %13u %7u ", Vals: File, Vals: Isa, Vals: Discriminator, Vals: OpIndex)
520 << (IsStmt ? " is_stmt" : "") << (BasicBlock ? " basic_block" : "")
521 << (PrologueEnd ? " prologue_end" : "")
522 << (EpilogueBegin ? " epilogue_begin" : "")
523 << (EndSequence ? " end_sequence" : "") << '\n';
524}
525
526DWARFDebugLine::Sequence::Sequence() { reset(); }
527
528void DWARFDebugLine::Sequence::reset() {
529 LowPC = 0;
530 HighPC = 0;
531 SectionIndex = object::SectionedAddress::UndefSection;
532 FirstRowIndex = 0;
533 LastRowIndex = 0;
534 Empty = true;
535}
536
537DWARFDebugLine::LineTable::LineTable() { clear(); }
538
539void DWARFDebugLine::LineTable::dump(raw_ostream &OS,
540 DIDumpOptions DumpOptions) const {
541 Prologue.dump(OS, DumpOptions);
542
543 if (!Rows.empty()) {
544 OS << '\n';
545 Row::dumpTableHeader(OS, Indent: 0);
546 for (const Row &R : Rows) {
547 R.dump(OS);
548 }
549 }
550
551 // Terminate the table with a final blank line to clearly delineate it from
552 // later dumps.
553 OS << '\n';
554}
555
556void DWARFDebugLine::LineTable::clear() {
557 Prologue.clear();
558 Rows.clear();
559 Sequences.clear();
560}
561
562DWARFDebugLine::ParsingState::ParsingState(
563 struct LineTable *LT, uint64_t TableOffset,
564 function_ref<void(Error)> ErrorHandler)
565 : LineTable(LT), LineTableOffset(TableOffset), ErrorHandler(ErrorHandler) {
566 resetRowAndSequence();
567}
568
569void DWARFDebugLine::ParsingState::resetRowAndSequence() {
570 Row.reset(DefaultIsStmt: LineTable->Prologue.DefaultIsStmt);
571 Sequence.reset();
572}
573
574void DWARFDebugLine::ParsingState::appendRowToMatrix() {
575 unsigned RowNumber = LineTable->Rows.size();
576 if (Sequence.Empty) {
577 // Record the beginning of instruction sequence.
578 Sequence.Empty = false;
579 Sequence.LowPC = Row.Address.Address;
580 Sequence.FirstRowIndex = RowNumber;
581 }
582 LineTable->appendRow(R: Row);
583 if (Row.EndSequence) {
584 // Record the end of instruction sequence.
585 Sequence.HighPC = Row.Address.Address;
586 Sequence.LastRowIndex = RowNumber + 1;
587 Sequence.SectionIndex = Row.Address.SectionIndex;
588 if (Sequence.isValid())
589 LineTable->appendSequence(S: Sequence);
590 Sequence.reset();
591 }
592 Row.postAppend();
593}
594
595const DWARFDebugLine::LineTable *
596DWARFDebugLine::getLineTable(uint64_t Offset) const {
597 LineTableConstIter Pos = LineTableMap.find(x: Offset);
598 if (Pos != LineTableMap.end())
599 return &Pos->second;
600 return nullptr;
601}
602
603Expected<const DWARFDebugLine::LineTable *> DWARFDebugLine::getOrParseLineTable(
604 DWARFDataExtractor &DebugLineData, uint64_t Offset, const DWARFContext &Ctx,
605 const DWARFUnit *U, function_ref<void(Error)> RecoverableErrorHandler) {
606 if (!DebugLineData.isValidOffset(offset: Offset))
607 return createStringError(EC: errc::invalid_argument, Fmt: "offset 0x%8.8" PRIx64
608 " is not a valid debug line section offset",
609 Vals: Offset);
610
611 std::pair<LineTableIter, bool> Pos =
612 LineTableMap.insert(x: LineTableMapTy::value_type(Offset, LineTable()));
613 LineTable *LT = &Pos.first->second;
614 if (Pos.second) {
615 if (Error Err =
616 LT->parse(DebugLineData, OffsetPtr: &Offset, Ctx, U, RecoverableErrorHandler))
617 return std::move(Err);
618 return LT;
619 }
620 return LT;
621}
622
623void DWARFDebugLine::clearLineTable(uint64_t Offset) {
624 LineTableMap.erase(x: Offset);
625}
626
627static StringRef getOpcodeName(uint8_t Opcode, uint8_t OpcodeBase) {
628 assert(Opcode != 0);
629 if (Opcode < OpcodeBase)
630 return LNStandardString(Standard: Opcode);
631 return "special";
632}
633
634DWARFDebugLine::ParsingState::AddrOpIndexDelta
635DWARFDebugLine::ParsingState::advanceAddrOpIndex(uint64_t OperationAdvance,
636 uint8_t Opcode,
637 uint64_t OpcodeOffset) {
638 StringRef OpcodeName = getOpcodeName(Opcode, OpcodeBase: LineTable->Prologue.OpcodeBase);
639 // For versions less than 4, the MaxOpsPerInst member is set to 0, as the
640 // maximum_operations_per_instruction field wasn't introduced until DWARFv4.
641 // Don't warn about bad values in this situation.
642 if (ReportAdvanceAddrProblem && LineTable->Prologue.getVersion() >= 4 &&
643 LineTable->Prologue.MaxOpsPerInst == 0)
644 ErrorHandler(createStringError(
645 EC: errc::invalid_argument,
646 Fmt: "line table program at offset 0x%8.8" PRIx64
647 " contains a %s opcode at offset 0x%8.8" PRIx64
648 ", but the prologue maximum_operations_per_instruction value is 0"
649 ", which is invalid. Assuming a value of 1 instead",
650 Vals: LineTableOffset, Vals: OpcodeName.data(), Vals: OpcodeOffset));
651 // Although we are able to correctly parse line number programs with
652 // MaxOpsPerInst > 1, the rest of DWARFDebugLine and its
653 // users have not been updated to handle line information for all operations
654 // in a multi-operation instruction, so warn about potentially incorrect
655 // results.
656 if (ReportAdvanceAddrProblem && LineTable->Prologue.MaxOpsPerInst > 1)
657 ErrorHandler(createStringError(
658 EC: errc::not_supported,
659 Fmt: "line table program at offset 0x%8.8" PRIx64
660 " contains a %s opcode at offset 0x%8.8" PRIx64
661 ", but the prologue maximum_operations_per_instruction value is %" PRId8
662 ", which is experimentally supported, so line number information "
663 "may be incorrect",
664 Vals: LineTableOffset, Vals: OpcodeName.data(), Vals: OpcodeOffset,
665 Vals: LineTable->Prologue.MaxOpsPerInst));
666 if (ReportAdvanceAddrProblem && LineTable->Prologue.MinInstLength == 0)
667 ErrorHandler(
668 createStringError(EC: errc::invalid_argument,
669 Fmt: "line table program at offset 0x%8.8" PRIx64
670 " contains a %s opcode at offset 0x%8.8" PRIx64
671 ", but the prologue minimum_instruction_length value "
672 "is 0, which prevents any address advancing",
673 Vals: LineTableOffset, Vals: OpcodeName.data(), Vals: OpcodeOffset));
674 ReportAdvanceAddrProblem = false;
675
676 // Advances the address and op_index according to DWARFv5, section 6.2.5.1:
677 //
678 // new address = address +
679 // minimum_instruction_length *
680 // ((op_index + operation advance) / maximum_operations_per_instruction)
681 //
682 // new op_index =
683 // (op_index + operation advance) % maximum_operations_per_instruction
684
685 // For versions less than 4, the MaxOpsPerInst member is set to 0, as the
686 // maximum_operations_per_instruction field wasn't introduced until DWARFv4.
687 uint8_t MaxOpsPerInst =
688 std::max(a: LineTable->Prologue.MaxOpsPerInst, b: uint8_t{1});
689
690 uint64_t AddrOffset = ((Row.OpIndex + OperationAdvance) / MaxOpsPerInst) *
691 LineTable->Prologue.MinInstLength;
692 Row.Address.Address += AddrOffset;
693
694 uint8_t PrevOpIndex = Row.OpIndex;
695 Row.OpIndex = (Row.OpIndex + OperationAdvance) % MaxOpsPerInst;
696 int16_t OpIndexDelta = static_cast<int16_t>(Row.OpIndex) - PrevOpIndex;
697
698 return {.AddrOffset: AddrOffset, .OpIndexDelta: OpIndexDelta};
699}
700
701DWARFDebugLine::ParsingState::OpcodeAdvanceResults
702DWARFDebugLine::ParsingState::advanceForOpcode(uint8_t Opcode,
703 uint64_t OpcodeOffset) {
704 assert(Opcode == DW_LNS_const_add_pc ||
705 Opcode >= LineTable->Prologue.OpcodeBase);
706 if (ReportBadLineRange && LineTable->Prologue.LineRange == 0) {
707 StringRef OpcodeName =
708 getOpcodeName(Opcode, OpcodeBase: LineTable->Prologue.OpcodeBase);
709 ErrorHandler(
710 createStringError(EC: errc::not_supported,
711 Fmt: "line table program at offset 0x%8.8" PRIx64
712 " contains a %s opcode at offset 0x%8.8" PRIx64
713 ", but the prologue line_range value is 0. The "
714 "address and line will not be adjusted",
715 Vals: LineTableOffset, Vals: OpcodeName.data(), Vals: OpcodeOffset));
716 ReportBadLineRange = false;
717 }
718
719 uint8_t OpcodeValue = Opcode;
720 if (Opcode == DW_LNS_const_add_pc)
721 OpcodeValue = 255;
722 uint8_t AdjustedOpcode = OpcodeValue - LineTable->Prologue.OpcodeBase;
723 uint64_t OperationAdvance =
724 LineTable->Prologue.LineRange != 0
725 ? AdjustedOpcode / LineTable->Prologue.LineRange
726 : 0;
727 AddrOpIndexDelta Advance =
728 advanceAddrOpIndex(OperationAdvance, Opcode, OpcodeOffset);
729 return {.AddrDelta: Advance.AddrOffset, .OpIndexDelta: Advance.OpIndexDelta, .AdjustedOpcode: AdjustedOpcode};
730}
731
732DWARFDebugLine::ParsingState::SpecialOpcodeDelta
733DWARFDebugLine::ParsingState::handleSpecialOpcode(uint8_t Opcode,
734 uint64_t OpcodeOffset) {
735 // A special opcode value is chosen based on the amount that needs
736 // to be added to the line and address registers. The maximum line
737 // increment for a special opcode is the value of the line_base
738 // field in the header, plus the value of the line_range field,
739 // minus 1 (line base + line range - 1). If the desired line
740 // increment is greater than the maximum line increment, a standard
741 // opcode must be used instead of a special opcode. The "address
742 // advance" is calculated by dividing the desired address increment
743 // by the minimum_instruction_length field from the header. The
744 // special opcode is then calculated using the following formula:
745 //
746 // opcode = (desired line increment - line_base) +
747 // (line_range * address advance) + opcode_base
748 //
749 // If the resulting opcode is greater than 255, a standard opcode
750 // must be used instead.
751 //
752 // To decode a special opcode, subtract the opcode_base from the
753 // opcode itself to give the adjusted opcode. The amount to
754 // increment the address register is the result of the adjusted
755 // opcode divided by the line_range multiplied by the
756 // minimum_instruction_length field from the header. That is:
757 //
758 // address increment = (adjusted opcode / line_range) *
759 // minimum_instruction_length
760 //
761 // The amount to increment the line register is the line_base plus
762 // the result of the adjusted opcode modulo the line_range. That is:
763 //
764 // line increment = line_base + (adjusted opcode % line_range)
765
766 DWARFDebugLine::ParsingState::OpcodeAdvanceResults AddrAdvanceResult =
767 advanceForOpcode(Opcode, OpcodeOffset);
768 int32_t LineOffset = 0;
769 if (LineTable->Prologue.LineRange != 0)
770 LineOffset =
771 LineTable->Prologue.LineBase +
772 (AddrAdvanceResult.AdjustedOpcode % LineTable->Prologue.LineRange);
773 Row.Line += LineOffset;
774 return {.Address: AddrAdvanceResult.AddrDelta, .Line: LineOffset,
775 .OpIndex: AddrAdvanceResult.OpIndexDelta};
776}
777
778/// Parse a ULEB128 using the specified \p Cursor. \returns the parsed value on
779/// success, or std::nullopt if \p Cursor is in a failing state.
780template <typename T>
781static std::optional<T> parseULEB128(DWARFDataExtractor &Data,
782 DataExtractor::Cursor &Cursor) {
783 T Value = Data.getULEB128(C&: Cursor);
784 if (Cursor)
785 return Value;
786 return std::nullopt;
787}
788
789Error DWARFDebugLine::LineTable::parse(
790 DWARFDataExtractor &DebugLineData, uint64_t *OffsetPtr,
791 const DWARFContext &Ctx, const DWARFUnit *U,
792 function_ref<void(Error)> RecoverableErrorHandler, raw_ostream *OS,
793 bool Verbose) {
794 assert((OS || !Verbose) && "cannot have verbose output without stream");
795 const uint64_t DebugLineOffset = *OffsetPtr;
796
797 clear();
798
799 Error PrologueErr =
800 Prologue.parse(DebugLineData, OffsetPtr, RecoverableErrorHandler, Ctx, U);
801
802 if (OS) {
803 DIDumpOptions DumpOptions;
804 DumpOptions.Verbose = Verbose;
805 Prologue.dump(OS&: *OS, DumpOptions);
806 }
807
808 if (PrologueErr) {
809 // Ensure there is a blank line after the prologue to clearly delineate it
810 // from later dumps.
811 if (OS)
812 *OS << "\n";
813 return PrologueErr;
814 }
815
816 uint64_t ProgramLength = Prologue.TotalLength + Prologue.sizeofTotalLength();
817 if (!DebugLineData.isValidOffsetForDataOfSize(offset: DebugLineOffset,
818 length: ProgramLength)) {
819 assert(DebugLineData.size() > DebugLineOffset &&
820 "prologue parsing should handle invalid offset");
821 uint64_t BytesRemaining = DebugLineData.size() - DebugLineOffset;
822 RecoverableErrorHandler(
823 createStringError(EC: errc::invalid_argument,
824 Fmt: "line table program with offset 0x%8.8" PRIx64
825 " has length 0x%8.8" PRIx64 " but only 0x%8.8" PRIx64
826 " bytes are available",
827 Vals: DebugLineOffset, Vals: ProgramLength, Vals: BytesRemaining));
828 // Continue by capping the length at the number of remaining bytes.
829 ProgramLength = BytesRemaining;
830 }
831
832 // Create a DataExtractor which can only see the data up to the end of the
833 // table, to prevent reading past the end.
834 const uint64_t EndOffset = DebugLineOffset + ProgramLength;
835 DWARFDataExtractor TableData(DebugLineData, EndOffset);
836
837 // See if we should tell the data extractor the address size.
838 if (TableData.getAddressSize() == 0)
839 TableData.setAddressSize(Prologue.getAddressSize());
840 else
841 assert(Prologue.getAddressSize() == 0 ||
842 Prologue.getAddressSize() == TableData.getAddressSize());
843
844 ParsingState State(this, DebugLineOffset, RecoverableErrorHandler);
845
846 *OffsetPtr = DebugLineOffset + Prologue.getLength();
847 if (OS && *OffsetPtr < EndOffset) {
848 *OS << '\n';
849 Row::dumpTableHeader(OS&: *OS, /*Indent=*/Verbose ? 12 : 0);
850 }
851 bool TombstonedAddress = false;
852 auto EmitRow = [&] {
853 if (!TombstonedAddress) {
854 if (Verbose) {
855 *OS << "\n";
856 OS->indent(NumSpaces: 12);
857 }
858 if (OS)
859 State.Row.dump(OS&: *OS);
860 State.appendRowToMatrix();
861 }
862 };
863 while (*OffsetPtr < EndOffset) {
864 DataExtractor::Cursor Cursor(*OffsetPtr);
865
866 if (Verbose)
867 *OS << format(Fmt: "0x%08.08" PRIx64 ": ", Vals: *OffsetPtr);
868
869 uint64_t OpcodeOffset = *OffsetPtr;
870 uint8_t Opcode = TableData.getU8(C&: Cursor);
871 size_t RowCount = Rows.size();
872
873 if (Cursor && Verbose)
874 *OS << format(Fmt: "%02.02" PRIx8 " ", Vals: Opcode);
875
876 if (Opcode == 0) {
877 // Extended Opcodes always start with a zero opcode followed by
878 // a uleb128 length so you can skip ones you don't know about
879 uint64_t Len = TableData.getULEB128(C&: Cursor);
880 uint64_t ExtOffset = Cursor.tell();
881
882 // Tolerate zero-length; assume length is correct and soldier on.
883 if (Len == 0) {
884 if (Cursor && Verbose)
885 *OS << "Badly formed extended line op (length 0)\n";
886 if (!Cursor) {
887 if (Verbose)
888 *OS << "\n";
889 RecoverableErrorHandler(Cursor.takeError());
890 }
891 *OffsetPtr = Cursor.tell();
892 continue;
893 }
894
895 uint8_t SubOpcode = TableData.getU8(C&: Cursor);
896 // OperandOffset will be the same as ExtOffset, if it was not possible to
897 // read the SubOpcode.
898 uint64_t OperandOffset = Cursor.tell();
899 if (Verbose)
900 *OS << LNExtendedString(Encoding: SubOpcode);
901 switch (SubOpcode) {
902 case DW_LNE_end_sequence:
903 // Set the end_sequence register of the state machine to true and
904 // append a row to the matrix using the current values of the
905 // state-machine registers. Then reset the registers to the initial
906 // values specified above. Every statement program sequence must end
907 // with a DW_LNE_end_sequence instruction which creates a row whose
908 // address is that of the byte after the last target machine instruction
909 // of the sequence.
910 State.Row.EndSequence = true;
911 // No need to test the Cursor is valid here, since it must be to get
912 // into this code path - if it were invalid, the default case would be
913 // followed.
914 EmitRow();
915 State.resetRowAndSequence();
916 break;
917
918 case DW_LNE_set_address:
919 // Takes a single relocatable address as an operand. The size of the
920 // operand is the size appropriate to hold an address on the target
921 // machine. Set the address register to the value given by the
922 // relocatable address and set the op_index register to 0. All of the
923 // other statement program opcodes that affect the address register
924 // add a delta to it. This instruction stores a relocatable value into
925 // it instead.
926 //
927 // Make sure the extractor knows the address size. If not, infer it
928 // from the size of the operand.
929 {
930 uint8_t ExtractorAddressSize = TableData.getAddressSize();
931 uint64_t OpcodeAddressSize = Len - 1;
932 if (ExtractorAddressSize != OpcodeAddressSize &&
933 ExtractorAddressSize != 0)
934 RecoverableErrorHandler(createStringError(
935 EC: errc::invalid_argument,
936 Fmt: "mismatching address size at offset 0x%8.8" PRIx64
937 " expected 0x%2.2" PRIx8 " found 0x%2.2" PRIx64,
938 Vals: ExtOffset, Vals: ExtractorAddressSize, Vals: Len - 1));
939
940 // Assume that the line table is correct and temporarily override the
941 // address size. If the size is unsupported, give up trying to read
942 // the address and continue to the next opcode.
943 if (OpcodeAddressSize != 1 && OpcodeAddressSize != 2 &&
944 OpcodeAddressSize != 4 && OpcodeAddressSize != 8) {
945 RecoverableErrorHandler(createStringError(
946 EC: errc::invalid_argument,
947 Fmt: "address size 0x%2.2" PRIx64
948 " of DW_LNE_set_address opcode at offset 0x%8.8" PRIx64
949 " is unsupported",
950 Vals: OpcodeAddressSize, Vals: ExtOffset));
951 TableData.skip(C&: Cursor, Length: OpcodeAddressSize);
952 } else {
953 TableData.setAddressSize(OpcodeAddressSize);
954 State.Row.Address.Address = TableData.getRelocatedAddress(
955 C&: Cursor, SecIx: &State.Row.Address.SectionIndex);
956 State.Row.OpIndex = 0;
957
958 uint64_t Tombstone =
959 dwarf::computeTombstoneAddress(AddressByteSize: OpcodeAddressSize);
960 TombstonedAddress = State.Row.Address.Address == Tombstone;
961
962 // Restore the address size if the extractor already had it.
963 if (ExtractorAddressSize != 0)
964 TableData.setAddressSize(ExtractorAddressSize);
965 }
966
967 if (Cursor && Verbose) {
968 *OS << " (";
969 DWARFFormValue::dumpAddress(OS&: *OS, AddressSize: OpcodeAddressSize, Address: State.Row.Address.Address);
970 *OS << ')';
971 }
972 }
973 break;
974
975 case DW_LNE_define_file:
976 // Takes 4 arguments. The first is a null terminated string containing
977 // a source file name. The second is an unsigned LEB128 number
978 // representing the directory index of the directory in which the file
979 // was found. The third is an unsigned LEB128 number representing the
980 // time of last modification of the file. The fourth is an unsigned
981 // LEB128 number representing the length in bytes of the file. The time
982 // and length fields may contain LEB128(0) if the information is not
983 // available.
984 //
985 // The directory index represents an entry in the include_directories
986 // section of the statement program prologue. The index is LEB128(0)
987 // if the file was found in the current directory of the compilation,
988 // LEB128(1) if it was found in the first directory in the
989 // include_directories section, and so on. The directory index is
990 // ignored for file names that represent full path names.
991 //
992 // The files are numbered, starting at 1, in the order in which they
993 // appear; the names in the prologue come before names defined by
994 // the DW_LNE_define_file instruction. These numbers are used in the
995 // the file register of the state machine.
996 {
997 FileNameEntry FileEntry;
998 const char *Name = TableData.getCStr(C&: Cursor);
999 FileEntry.Name =
1000 DWARFFormValue::createFromPValue(F: dwarf::DW_FORM_string, V: Name);
1001 FileEntry.DirIdx = TableData.getULEB128(C&: Cursor);
1002 FileEntry.ModTime = TableData.getULEB128(C&: Cursor);
1003 FileEntry.Length = TableData.getULEB128(C&: Cursor);
1004 Prologue.FileNames.push_back(x: FileEntry);
1005 if (Cursor && Verbose)
1006 *OS << " (" << Name << ", dir=" << FileEntry.DirIdx << ", mod_time="
1007 << format(Fmt: "(0x%16.16" PRIx64 ")", Vals: FileEntry.ModTime)
1008 << ", length=" << FileEntry.Length << ")";
1009 }
1010 break;
1011
1012 case DW_LNE_set_discriminator:
1013 State.Row.Discriminator = TableData.getULEB128(C&: Cursor);
1014 if (Cursor && Verbose)
1015 *OS << " (" << State.Row.Discriminator << ")";
1016 break;
1017
1018 default:
1019 if (Cursor && Verbose)
1020 *OS << format(Fmt: "Unrecognized extended op 0x%02.02" PRIx8, Vals: SubOpcode)
1021 << format(Fmt: " length %" PRIx64, Vals: Len);
1022 // Len doesn't include the zero opcode byte or the length itself, but
1023 // it does include the sub_opcode, so we have to adjust for that.
1024 TableData.skip(C&: Cursor, Length: Len - 1);
1025 break;
1026 }
1027 // Make sure the length as recorded in the table and the standard length
1028 // for the opcode match. If they don't, continue from the end as claimed
1029 // by the table. Similarly, continue from the claimed end in the event of
1030 // a parsing error.
1031 uint64_t End = ExtOffset + Len;
1032 if (Cursor && Cursor.tell() != End)
1033 RecoverableErrorHandler(createStringError(
1034 EC: errc::illegal_byte_sequence,
1035 Fmt: "unexpected line op length at offset 0x%8.8" PRIx64
1036 " expected 0x%2.2" PRIx64 " found 0x%2.2" PRIx64,
1037 Vals: ExtOffset, Vals: Len, Vals: Cursor.tell() - ExtOffset));
1038 if (!Cursor && Verbose) {
1039 DWARFDataExtractor::Cursor ByteCursor(OperandOffset);
1040 uint8_t Byte = TableData.getU8(C&: ByteCursor);
1041 if (ByteCursor) {
1042 *OS << " (<parsing error>";
1043 do {
1044 *OS << format(Fmt: " %2.2" PRIx8, Vals: Byte);
1045 Byte = TableData.getU8(C&: ByteCursor);
1046 } while (ByteCursor);
1047 *OS << ")";
1048 }
1049
1050 // The only parse failure in this case should be if the end was reached.
1051 // In that case, throw away the error, as the main Cursor's error will
1052 // be sufficient.
1053 consumeError(Err: ByteCursor.takeError());
1054 }
1055 *OffsetPtr = End;
1056 } else if (Opcode < Prologue.OpcodeBase) {
1057 if (Verbose)
1058 *OS << LNStandardString(Standard: Opcode);
1059 switch (Opcode) {
1060 // Standard Opcodes
1061 case DW_LNS_copy:
1062 // Takes no arguments. Append a row to the matrix using the
1063 // current values of the state-machine registers.
1064 EmitRow();
1065 break;
1066
1067 case DW_LNS_advance_pc:
1068 // Takes a single unsigned LEB128 operand as the operation advance
1069 // and modifies the address and op_index registers of the state machine
1070 // according to that.
1071 if (std::optional<uint64_t> Operand =
1072 parseULEB128<uint64_t>(Data&: TableData, Cursor)) {
1073 ParsingState::AddrOpIndexDelta Advance =
1074 State.advanceAddrOpIndex(OperationAdvance: *Operand, Opcode, OpcodeOffset);
1075 if (Verbose)
1076 *OS << " (addr += " << Advance.AddrOffset
1077 << ", op-index += " << Advance.OpIndexDelta << ")";
1078 }
1079 break;
1080
1081 case DW_LNS_advance_line:
1082 // Takes a single signed LEB128 operand and adds that value to
1083 // the line register of the state machine.
1084 {
1085 int64_t LineDelta = TableData.getSLEB128(C&: Cursor);
1086 if (Cursor) {
1087 State.Row.Line += LineDelta;
1088 if (Verbose)
1089 *OS << " (" << State.Row.Line << ")";
1090 }
1091 }
1092 break;
1093
1094 case DW_LNS_set_file:
1095 // Takes a single unsigned LEB128 operand and stores it in the file
1096 // register of the state machine.
1097 if (std::optional<uint16_t> File =
1098 parseULEB128<uint16_t>(Data&: TableData, Cursor)) {
1099 State.Row.File = *File;
1100 if (Verbose)
1101 *OS << " (" << State.Row.File << ")";
1102 }
1103 break;
1104
1105 case DW_LNS_set_column:
1106 // Takes a single unsigned LEB128 operand and stores it in the
1107 // column register of the state machine.
1108 if (std::optional<uint16_t> Column =
1109 parseULEB128<uint16_t>(Data&: TableData, Cursor)) {
1110 State.Row.Column = *Column;
1111 if (Verbose)
1112 *OS << " (" << State.Row.Column << ")";
1113 }
1114 break;
1115
1116 case DW_LNS_negate_stmt:
1117 // Takes no arguments. Set the is_stmt register of the state
1118 // machine to the logical negation of its current value.
1119 State.Row.IsStmt = !State.Row.IsStmt;
1120 break;
1121
1122 case DW_LNS_set_basic_block:
1123 // Takes no arguments. Set the basic_block register of the
1124 // state machine to true
1125 State.Row.BasicBlock = true;
1126 break;
1127
1128 case DW_LNS_const_add_pc:
1129 // Takes no arguments. Advance the address and op_index registers of
1130 // the state machine by the increments corresponding to special
1131 // opcode 255. The motivation for DW_LNS_const_add_pc is this:
1132 // when the statement program needs to advance the address by a
1133 // small amount, it can use a single special opcode, which occupies
1134 // a single byte. When it needs to advance the address by up to
1135 // twice the range of the last special opcode, it can use
1136 // DW_LNS_const_add_pc followed by a special opcode, for a total
1137 // of two bytes. Only if it needs to advance the address by more
1138 // than twice that range will it need to use both DW_LNS_advance_pc
1139 // and a special opcode, requiring three or more bytes.
1140 {
1141 ParsingState::OpcodeAdvanceResults Advance =
1142 State.advanceForOpcode(Opcode, OpcodeOffset);
1143 if (Verbose)
1144 *OS << format(Fmt: " (addr += 0x%16.16" PRIx64 ", op-index += %" PRIu8
1145 ")",
1146 Vals: Advance.AddrDelta, Vals: Advance.OpIndexDelta);
1147 }
1148 break;
1149
1150 case DW_LNS_fixed_advance_pc:
1151 // Takes a single uhalf operand. Add to the address register of
1152 // the state machine the value of the (unencoded) operand and set
1153 // the op_index register to 0. This is the only extended opcode that
1154 // takes an argument that is not a variable length number.
1155 // The motivation for DW_LNS_fixed_advance_pc is this: existing
1156 // assemblers cannot emit DW_LNS_advance_pc or special opcodes because
1157 // they cannot encode LEB128 numbers or judge when the computation
1158 // of a special opcode overflows and requires the use of
1159 // DW_LNS_advance_pc. Such assemblers, however, can use
1160 // DW_LNS_fixed_advance_pc instead, sacrificing compression.
1161 {
1162 uint16_t PCOffset =
1163 TableData.getRelocatedValue(C&: Cursor, Size: 2);
1164 if (Cursor) {
1165 State.Row.Address.Address += PCOffset;
1166 State.Row.OpIndex = 0;
1167 if (Verbose)
1168 *OS << format(Fmt: " (addr += 0x%4.4" PRIx16 ", op-index = 0)",
1169 Vals: PCOffset);
1170 }
1171 }
1172 break;
1173
1174 case DW_LNS_set_prologue_end:
1175 // Takes no arguments. Set the prologue_end register of the
1176 // state machine to true
1177 State.Row.PrologueEnd = true;
1178 break;
1179
1180 case DW_LNS_set_epilogue_begin:
1181 // Takes no arguments. Set the basic_block register of the
1182 // state machine to true
1183 State.Row.EpilogueBegin = true;
1184 break;
1185
1186 case DW_LNS_set_isa:
1187 // Takes a single unsigned LEB128 operand and stores it in the
1188 // ISA register of the state machine.
1189 if (std::optional<uint8_t> Isa =
1190 parseULEB128<uint8_t>(Data&: TableData, Cursor)) {
1191 State.Row.Isa = *Isa;
1192 if (Verbose)
1193 *OS << " (" << (uint64_t)State.Row.Isa << ")";
1194 }
1195 break;
1196
1197 default:
1198 // Handle any unknown standard opcodes here. We know the lengths
1199 // of such opcodes because they are specified in the prologue
1200 // as a multiple of LEB128 operands for each opcode.
1201 {
1202 assert(Opcode - 1U < Prologue.StandardOpcodeLengths.size());
1203 if (Verbose)
1204 *OS << "Unrecognized standard opcode";
1205 uint8_t OpcodeLength = Prologue.StandardOpcodeLengths[Opcode - 1];
1206 std::vector<uint64_t> Operands;
1207 for (uint8_t I = 0; I < OpcodeLength; ++I) {
1208 if (std::optional<uint64_t> Value =
1209 parseULEB128<uint64_t>(Data&: TableData, Cursor))
1210 Operands.push_back(x: *Value);
1211 else
1212 break;
1213 }
1214 if (Verbose && !Operands.empty()) {
1215 *OS << " (operands: ";
1216 bool First = true;
1217 for (uint64_t Value : Operands) {
1218 if (!First)
1219 *OS << ", ";
1220 First = false;
1221 *OS << format(Fmt: "0x%16.16" PRIx64, Vals: Value);
1222 }
1223 if (Verbose)
1224 *OS << ')';
1225 }
1226 }
1227 break;
1228 }
1229
1230 *OffsetPtr = Cursor.tell();
1231 } else {
1232 // Special Opcodes.
1233 ParsingState::SpecialOpcodeDelta Delta =
1234 State.handleSpecialOpcode(Opcode, OpcodeOffset);
1235
1236 if (Verbose)
1237 *OS << "address += " << Delta.Address << ", line += " << Delta.Line
1238 << ", op-index += " << Delta.OpIndex;
1239 EmitRow();
1240 *OffsetPtr = Cursor.tell();
1241 }
1242
1243 // When a row is added to the matrix, it is also dumped, which includes a
1244 // new line already, so don't add an extra one.
1245 if (Verbose && Rows.size() == RowCount)
1246 *OS << "\n";
1247
1248 // Most parse failures other than when parsing extended opcodes are due to
1249 // failures to read ULEBs. Bail out of parsing, since we don't know where to
1250 // continue reading from as there is no stated length for such byte
1251 // sequences. Print the final trailing new line if needed before doing so.
1252 if (!Cursor && Opcode != 0) {
1253 if (Verbose)
1254 *OS << "\n";
1255 return Cursor.takeError();
1256 }
1257
1258 if (!Cursor)
1259 RecoverableErrorHandler(Cursor.takeError());
1260 }
1261
1262 if (!State.Sequence.Empty)
1263 RecoverableErrorHandler(createStringError(
1264 EC: errc::illegal_byte_sequence,
1265 Fmt: "last sequence in debug line table at offset 0x%8.8" PRIx64
1266 " is not terminated",
1267 Vals: DebugLineOffset));
1268
1269 // Sort all sequences so that address lookup will work faster.
1270 if (!Sequences.empty()) {
1271 llvm::sort(C&: Sequences, Comp: Sequence::orderByHighPC);
1272 // Note: actually, instruction address ranges of sequences should not
1273 // overlap (in shared objects and executables). If they do, the address
1274 // lookup would still work, though, but result would be ambiguous.
1275 // We don't report warning in this case. For example,
1276 // sometimes .so compiled from multiple object files contains a few
1277 // rudimentary sequences for address ranges [0x0, 0xsomething).
1278 }
1279
1280 // Terminate the table with a final blank line to clearly delineate it from
1281 // later dumps.
1282 if (OS)
1283 *OS << "\n";
1284
1285 return Error::success();
1286}
1287
1288uint32_t DWARFDebugLine::LineTable::findRowInSeq(
1289 const DWARFDebugLine::Sequence &Seq,
1290 object::SectionedAddress Address) const {
1291 if (!Seq.containsPC(PC: Address))
1292 return UnknownRowIndex;
1293 assert(Seq.SectionIndex == Address.SectionIndex);
1294 // In some cases, e.g. first instruction in a function, the compiler generates
1295 // two entries, both with the same address. We want the last one.
1296 //
1297 // In general we want a non-empty range: the last row whose address is less
1298 // than or equal to Address. This can be computed as upper_bound - 1.
1299 //
1300 // TODO: This function, and its users, needs to be update to return multiple
1301 // rows for bundles with multiple op-indexes.
1302 DWARFDebugLine::Row Row;
1303 Row.Address = Address;
1304 RowIter FirstRow = Rows.begin() + Seq.FirstRowIndex;
1305 RowIter LastRow = Rows.begin() + Seq.LastRowIndex;
1306 assert(FirstRow->Address.Address <= Row.Address.Address &&
1307 Row.Address.Address < LastRow[-1].Address.Address);
1308 RowIter RowPos = std::upper_bound(first: FirstRow + 1, last: LastRow - 1, val: Row,
1309 comp: DWARFDebugLine::Row::orderByAddress) -
1310 1;
1311 assert(Seq.SectionIndex == RowPos->Address.SectionIndex);
1312 return RowPos - Rows.begin();
1313}
1314
1315uint32_t DWARFDebugLine::LineTable::lookupAddress(
1316 object::SectionedAddress Address) const {
1317
1318 // Search for relocatable addresses
1319 uint32_t Result = lookupAddressImpl(Address);
1320
1321 if (Result != UnknownRowIndex ||
1322 Address.SectionIndex == object::SectionedAddress::UndefSection)
1323 return Result;
1324
1325 // Search for absolute addresses
1326 Address.SectionIndex = object::SectionedAddress::UndefSection;
1327 return lookupAddressImpl(Address);
1328}
1329
1330uint32_t DWARFDebugLine::LineTable::lookupAddressImpl(
1331 object::SectionedAddress Address) const {
1332 // First, find an instruction sequence containing the given address.
1333 DWARFDebugLine::Sequence Sequence;
1334 Sequence.SectionIndex = Address.SectionIndex;
1335 Sequence.HighPC = Address.Address;
1336 SequenceIter It = llvm::upper_bound(Range: Sequences, Value&: Sequence,
1337 C: DWARFDebugLine::Sequence::orderByHighPC);
1338 if (It == Sequences.end() || It->SectionIndex != Address.SectionIndex)
1339 return UnknownRowIndex;
1340 return findRowInSeq(Seq: *It, Address);
1341}
1342
1343bool DWARFDebugLine::LineTable::lookupAddressRange(
1344 object::SectionedAddress Address, uint64_t Size,
1345 std::vector<uint32_t> &Result) const {
1346
1347 // Search for relocatable addresses
1348 if (lookupAddressRangeImpl(Address, Size, Result))
1349 return true;
1350
1351 if (Address.SectionIndex == object::SectionedAddress::UndefSection)
1352 return false;
1353
1354 // Search for absolute addresses
1355 Address.SectionIndex = object::SectionedAddress::UndefSection;
1356 return lookupAddressRangeImpl(Address, Size, Result);
1357}
1358
1359bool DWARFDebugLine::LineTable::lookupAddressRangeImpl(
1360 object::SectionedAddress Address, uint64_t Size,
1361 std::vector<uint32_t> &Result) const {
1362 if (Sequences.empty())
1363 return false;
1364 uint64_t EndAddr = Address.Address + Size;
1365 // First, find an instruction sequence containing the given address.
1366 DWARFDebugLine::Sequence Sequence;
1367 Sequence.SectionIndex = Address.SectionIndex;
1368 Sequence.HighPC = Address.Address;
1369 SequenceIter LastSeq = Sequences.end();
1370 SequenceIter SeqPos = llvm::upper_bound(
1371 Range: Sequences, Value&: Sequence, C: DWARFDebugLine::Sequence::orderByHighPC);
1372 if (SeqPos == LastSeq || !SeqPos->containsPC(PC: Address))
1373 return false;
1374
1375 SequenceIter StartPos = SeqPos;
1376
1377 // Add the rows from the first sequence to the vector, starting with the
1378 // index we just calculated
1379
1380 while (SeqPos != LastSeq && SeqPos->LowPC < EndAddr) {
1381 const DWARFDebugLine::Sequence &CurSeq = *SeqPos;
1382 // For the first sequence, we need to find which row in the sequence is the
1383 // first in our range.
1384 uint32_t FirstRowIndex = CurSeq.FirstRowIndex;
1385 if (SeqPos == StartPos)
1386 FirstRowIndex = findRowInSeq(Seq: CurSeq, Address);
1387
1388 // Figure out the last row in the range.
1389 uint32_t LastRowIndex =
1390 findRowInSeq(Seq: CurSeq, Address: {.Address: EndAddr - 1, .SectionIndex: Address.SectionIndex});
1391 if (LastRowIndex == UnknownRowIndex)
1392 LastRowIndex = CurSeq.LastRowIndex - 1;
1393
1394 assert(FirstRowIndex != UnknownRowIndex);
1395 assert(LastRowIndex != UnknownRowIndex);
1396
1397 for (uint32_t I = FirstRowIndex; I <= LastRowIndex; ++I) {
1398 Result.push_back(x: I);
1399 }
1400
1401 ++SeqPos;
1402 }
1403
1404 return true;
1405}
1406
1407std::optional<StringRef>
1408DWARFDebugLine::LineTable::getSourceByIndex(uint64_t FileIndex,
1409 FileLineInfoKind Kind) const {
1410 if (Kind == FileLineInfoKind::None || !Prologue.hasFileAtIndex(FileIndex))
1411 return std::nullopt;
1412 const FileNameEntry &Entry = Prologue.getFileNameEntry(Index: FileIndex);
1413 if (auto E = dwarf::toString(V: Entry.Source))
1414 return StringRef(*E);
1415 return std::nullopt;
1416}
1417
1418static bool isPathAbsoluteOnWindowsOrPosix(const Twine &Path) {
1419 // Debug info can contain paths from any OS, not necessarily
1420 // an OS we're currently running on. Moreover different compilation units can
1421 // be compiled on different operating systems and linked together later.
1422 return sys::path::is_absolute(path: Path, style: sys::path::Style::posix) ||
1423 sys::path::is_absolute(path: Path, style: sys::path::Style::windows);
1424}
1425
1426bool DWARFDebugLine::Prologue::getFileNameByIndex(
1427 uint64_t FileIndex, StringRef CompDir, FileLineInfoKind Kind,
1428 std::string &Result, sys::path::Style Style) const {
1429 if (Kind == FileLineInfoKind::None || !hasFileAtIndex(FileIndex))
1430 return false;
1431 const FileNameEntry &Entry = getFileNameEntry(Index: FileIndex);
1432 auto E = dwarf::toString(V: Entry.Name);
1433 if (!E)
1434 return false;
1435 StringRef FileName = *E;
1436 if (Kind == FileLineInfoKind::RawValue ||
1437 isPathAbsoluteOnWindowsOrPosix(Path: FileName)) {
1438 Result = std::string(FileName);
1439 return true;
1440 }
1441 if (Kind == FileLineInfoKind::BaseNameOnly) {
1442 Result = std::string(llvm::sys::path::filename(path: FileName));
1443 return true;
1444 }
1445
1446 SmallString<16> FilePath;
1447 StringRef IncludeDir;
1448 // Be defensive about the contents of Entry.
1449 if (getVersion() >= 5) {
1450 // DirIdx 0 is the compilation directory, so don't include it for
1451 // relative names.
1452 if ((Entry.DirIdx != 0 || Kind != FileLineInfoKind::RelativeFilePath) &&
1453 Entry.DirIdx < IncludeDirectories.size())
1454 IncludeDir = dwarf::toStringRef(V: IncludeDirectories[Entry.DirIdx]);
1455 } else {
1456 if (0 < Entry.DirIdx && Entry.DirIdx <= IncludeDirectories.size())
1457 IncludeDir = dwarf::toStringRef(V: IncludeDirectories[Entry.DirIdx - 1]);
1458 }
1459
1460 // For absolute paths only, include the compilation directory of compile unit,
1461 // unless v5 DirIdx == 0 (IncludeDir indicates the compilation directory). We
1462 // know that FileName is not absolute, the only way to have an absolute path
1463 // at this point would be if IncludeDir is absolute.
1464 if (Kind == FileLineInfoKind::AbsoluteFilePath &&
1465 (getVersion() < 5 || Entry.DirIdx != 0) && !CompDir.empty() &&
1466 !isPathAbsoluteOnWindowsOrPosix(Path: IncludeDir))
1467 sys::path::append(path&: FilePath, style: Style, a: CompDir);
1468
1469 assert((Kind == FileLineInfoKind::AbsoluteFilePath ||
1470 Kind == FileLineInfoKind::RelativeFilePath) &&
1471 "invalid FileLineInfo Kind");
1472
1473 // sys::path::append skips empty strings.
1474 sys::path::append(path&: FilePath, style: Style, a: IncludeDir, b: FileName);
1475 Result = std::string(FilePath);
1476 return true;
1477}
1478
1479bool DWARFDebugLine::LineTable::getFileLineInfoForAddress(
1480 object::SectionedAddress Address, const char *CompDir,
1481 FileLineInfoKind Kind, DILineInfo &Result) const {
1482 // Get the index of row we're looking for in the line table.
1483 uint32_t RowIndex = lookupAddress(Address);
1484 if (RowIndex == -1U)
1485 return false;
1486 // Take file number and line/column from the row.
1487 const auto &Row = Rows[RowIndex];
1488 if (!getFileNameByIndex(FileIndex: Row.File, CompDir, Kind, Result&: Result.FileName))
1489 return false;
1490 Result.Line = Row.Line;
1491 Result.Column = Row.Column;
1492 Result.Discriminator = Row.Discriminator;
1493 Result.Source = getSourceByIndex(FileIndex: Row.File, Kind);
1494 return true;
1495}
1496
1497bool DWARFDebugLine::LineTable::getDirectoryForEntry(
1498 const FileNameEntry &Entry, std::string &Directory) const {
1499 if (Prologue.getVersion() >= 5) {
1500 if (Entry.DirIdx < Prologue.IncludeDirectories.size()) {
1501 Directory =
1502 dwarf::toString(V: Prologue.IncludeDirectories[Entry.DirIdx], Default: "");
1503 return true;
1504 }
1505 return false;
1506 }
1507 if (0 < Entry.DirIdx && Entry.DirIdx <= Prologue.IncludeDirectories.size()) {
1508 Directory =
1509 dwarf::toString(V: Prologue.IncludeDirectories[Entry.DirIdx - 1], Default: "");
1510 return true;
1511 }
1512 return false;
1513}
1514
1515// We want to supply the Unit associated with a .debug_line[.dwo] table when
1516// we dump it, if possible, but still dump the table even if there isn't a Unit.
1517// Therefore, collect up handles on all the Units that point into the
1518// line-table section.
1519static DWARFDebugLine::SectionParser::LineToUnitMap
1520buildLineToUnitMap(DWARFUnitVector::iterator_range Units) {
1521 DWARFDebugLine::SectionParser::LineToUnitMap LineToUnit;
1522 for (const auto &U : Units)
1523 if (auto CUDIE = U->getUnitDIE())
1524 if (auto StmtOffset = toSectionOffset(V: CUDIE.find(Attr: DW_AT_stmt_list)))
1525 LineToUnit.insert(x: std::make_pair(x&: *StmtOffset, y: &*U));
1526 return LineToUnit;
1527}
1528
1529DWARFDebugLine::SectionParser::SectionParser(
1530 DWARFDataExtractor &Data, const DWARFContext &C,
1531 DWARFUnitVector::iterator_range Units)
1532 : DebugLineData(Data), Context(C) {
1533 LineToUnit = buildLineToUnitMap(Units);
1534 if (!DebugLineData.isValidOffset(offset: Offset))
1535 Done = true;
1536}
1537
1538bool DWARFDebugLine::Prologue::totalLengthIsValid() const {
1539 return TotalLength != 0u;
1540}
1541
1542DWARFDebugLine::LineTable DWARFDebugLine::SectionParser::parseNext(
1543 function_ref<void(Error)> RecoverableErrorHandler,
1544 function_ref<void(Error)> UnrecoverableErrorHandler, raw_ostream *OS,
1545 bool Verbose) {
1546 assert(DebugLineData.isValidOffset(Offset) &&
1547 "parsing should have terminated");
1548 DWARFUnit *U = prepareToParse(Offset);
1549 uint64_t OldOffset = Offset;
1550 LineTable LT;
1551 if (Error Err = LT.parse(DebugLineData, OffsetPtr: &Offset, Ctx: Context, U,
1552 RecoverableErrorHandler, OS, Verbose))
1553 UnrecoverableErrorHandler(std::move(Err));
1554 moveToNextTable(OldOffset, P: LT.Prologue);
1555 return LT;
1556}
1557
1558void DWARFDebugLine::SectionParser::skip(
1559 function_ref<void(Error)> RecoverableErrorHandler,
1560 function_ref<void(Error)> UnrecoverableErrorHandler) {
1561 assert(DebugLineData.isValidOffset(Offset) &&
1562 "parsing should have terminated");
1563 DWARFUnit *U = prepareToParse(Offset);
1564 uint64_t OldOffset = Offset;
1565 LineTable LT;
1566 if (Error Err = LT.Prologue.parse(DebugLineData, OffsetPtr: &Offset,
1567 RecoverableErrorHandler, Ctx: Context, U))
1568 UnrecoverableErrorHandler(std::move(Err));
1569 moveToNextTable(OldOffset, P: LT.Prologue);
1570}
1571
1572DWARFUnit *DWARFDebugLine::SectionParser::prepareToParse(uint64_t Offset) {
1573 DWARFUnit *U = nullptr;
1574 auto It = LineToUnit.find(x: Offset);
1575 if (It != LineToUnit.end())
1576 U = It->second;
1577 DebugLineData.setAddressSize(U ? U->getAddressByteSize() : 0);
1578 return U;
1579}
1580
1581bool DWARFDebugLine::SectionParser::hasValidVersion(uint64_t Offset) {
1582 DataExtractor::Cursor Cursor(Offset);
1583 auto [TotalLength, _] = DebugLineData.getInitialLength(C&: Cursor);
1584 DWARFDataExtractor HeaderData(DebugLineData, Cursor.tell() + TotalLength);
1585 uint16_t Version = HeaderData.getU16(C&: Cursor);
1586 if (!Cursor) {
1587 // Ignore any error here.
1588 // If this is not the end of the section parseNext() will still be
1589 // attempted, where this error will occur again (and can be handled).
1590 consumeError(Err: Cursor.takeError());
1591 return false;
1592 }
1593 return versionIsSupported(Version);
1594}
1595
1596void DWARFDebugLine::SectionParser::moveToNextTable(uint64_t OldOffset,
1597 const Prologue &P) {
1598 // If the length field is not valid, we don't know where the next table is, so
1599 // cannot continue to parse. Mark the parser as done, and leave the Offset
1600 // value as it currently is. This will be the end of the bad length field.
1601 if (!P.totalLengthIsValid()) {
1602 Done = true;
1603 return;
1604 }
1605
1606 Offset = OldOffset + P.TotalLength + P.sizeofTotalLength();
1607 if (!DebugLineData.isValidOffset(offset: Offset)) {
1608 Done = true;
1609 return;
1610 }
1611
1612 // Heuristic: If the version is valid, then this is probably a line table.
1613 // Otherwise, the offset might need alignment (to a 4 or 8 byte boundary).
1614 if (hasValidVersion(Offset))
1615 return;
1616
1617 // ARM C/C++ Compiler aligns each line table to word boundaries and pads out
1618 // the .debug_line section to a word multiple. Note that in the specification
1619 // this does not seem forbidden since each unit has a DW_AT_stmt_list.
1620 for (unsigned Align : {4, 8}) {
1621 uint64_t AlignedOffset = alignTo(Value: Offset, Align);
1622 if (!DebugLineData.isValidOffset(offset: AlignedOffset)) {
1623 // This is almost certainly not another line table but some alignment
1624 // padding. This assumes the alignments tested are ordered, and are
1625 // smaller than the header size (which is true for 4 and 8).
1626 Done = true;
1627 return;
1628 }
1629 if (hasValidVersion(Offset: AlignedOffset)) {
1630 Offset = AlignedOffset;
1631 break;
1632 }
1633 }
1634}
1635