1//===-- DWARFUnit.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/DWARFUnit.h"
11#include "llvm/DebugInfo/DWARF/DWARFContext.h"
12#include "llvm/DebugInfo/DWARF/DWARFFormValue.h"
13#include "llvm/Support/Dwarf.h"
14#include "llvm/Support/Path.h"
15#include <cstdio>
16
17using namespace llvm;
18using namespace dwarf;
19
20void DWARFUnitSectionBase::parse(DWARFContext &C, const DWARFSection &Section) {
21  parseImpl(C, Section, C.getDebugAbbrev(), C.getRangeSection(),
22            C.getStringSection(), StringRef(), C.getAddrSection(),
23            C.isLittleEndian());
24}
25
26void DWARFUnitSectionBase::parseDWO(DWARFContext &C,
27                                    const DWARFSection &DWOSection) {
28  parseImpl(C, DWOSection, C.getDebugAbbrevDWO(), C.getRangeDWOSection(),
29            C.getStringDWOSection(), C.getStringOffsetDWOSection(),
30            C.getAddrSection(), C.isLittleEndian());
31}
32
33DWARFUnit::DWARFUnit(DWARFContext &DC, const DWARFSection &Section,
34                     const DWARFDebugAbbrev *DA, StringRef RS, StringRef SS,
35                     StringRef SOS, StringRef AOS, bool LE,
36                     const DWARFUnitSectionBase &UnitSection)
37    : Context(DC), InfoSection(Section), Abbrev(DA), RangeSection(RS),
38      StringSection(SS), StringOffsetSection(SOS), AddrOffsetSection(AOS),
39      isLittleEndian(LE), UnitSection(UnitSection) {
40  clear();
41}
42
43DWARFUnit::~DWARFUnit() {
44}
45
46bool DWARFUnit::getAddrOffsetSectionItem(uint32_t Index,
47                                                uint64_t &Result) const {
48  uint32_t Offset = AddrOffsetSectionBase + Index * AddrSize;
49  if (AddrOffsetSection.size() < Offset + AddrSize)
50    return false;
51  DataExtractor DA(AddrOffsetSection, isLittleEndian, AddrSize);
52  Result = DA.getAddress(&Offset);
53  return true;
54}
55
56bool DWARFUnit::getStringOffsetSectionItem(uint32_t Index,
57                                                  uint32_t &Result) const {
58  // FIXME: string offset section entries are 8-byte for DWARF64.
59  const uint32_t ItemSize = 4;
60  uint32_t Offset = Index * ItemSize;
61  if (StringOffsetSection.size() < Offset + ItemSize)
62    return false;
63  DataExtractor DA(StringOffsetSection, isLittleEndian, 0);
64  Result = DA.getU32(&Offset);
65  return true;
66}
67
68bool DWARFUnit::extractImpl(DataExtractor debug_info, uint32_t *offset_ptr) {
69  Length = debug_info.getU32(offset_ptr);
70  Version = debug_info.getU16(offset_ptr);
71  uint64_t AbbrOffset = debug_info.getU32(offset_ptr);
72  AddrSize = debug_info.getU8(offset_ptr);
73
74  bool LengthOK = debug_info.isValidOffset(getNextUnitOffset() - 1);
75  bool VersionOK = DWARFContext::isSupportedVersion(Version);
76  bool AddrSizeOK = AddrSize == 4 || AddrSize == 8;
77
78  if (!LengthOK || !VersionOK || !AddrSizeOK)
79    return false;
80
81  Abbrevs = Abbrev->getAbbreviationDeclarationSet(AbbrOffset);
82  if (Abbrevs == nullptr)
83    return false;
84
85  return true;
86}
87
88bool DWARFUnit::extract(DataExtractor debug_info, uint32_t *offset_ptr) {
89  clear();
90
91  Offset = *offset_ptr;
92
93  if (debug_info.isValidOffset(*offset_ptr)) {
94    if (extractImpl(debug_info, offset_ptr))
95      return true;
96
97    // reset the offset to where we tried to parse from if anything went wrong
98    *offset_ptr = Offset;
99  }
100
101  return false;
102}
103
104bool DWARFUnit::extractRangeList(uint32_t RangeListOffset,
105                                        DWARFDebugRangeList &RangeList) const {
106  // Require that compile unit is extracted.
107  assert(DieArray.size() > 0);
108  DataExtractor RangesData(RangeSection, isLittleEndian, AddrSize);
109  uint32_t ActualRangeListOffset = RangeSectionBase + RangeListOffset;
110  return RangeList.extract(RangesData, &ActualRangeListOffset);
111}
112
113void DWARFUnit::clear() {
114  Offset = 0;
115  Length = 0;
116  Version = 0;
117  Abbrevs = nullptr;
118  AddrSize = 0;
119  BaseAddr = 0;
120  RangeSectionBase = 0;
121  AddrOffsetSectionBase = 0;
122  clearDIEs(false);
123  DWO.reset();
124}
125
126const char *DWARFUnit::getCompilationDir() {
127  extractDIEsIfNeeded(true);
128  if (DieArray.empty())
129    return nullptr;
130  return DieArray[0].getAttributeValueAsString(this, DW_AT_comp_dir, nullptr);
131}
132
133uint64_t DWARFUnit::getDWOId() {
134  extractDIEsIfNeeded(true);
135  const uint64_t FailValue = -1ULL;
136  if (DieArray.empty())
137    return FailValue;
138  return DieArray[0]
139      .getAttributeValueAsUnsignedConstant(this, DW_AT_GNU_dwo_id, FailValue);
140}
141
142void DWARFUnit::setDIERelations() {
143  if (DieArray.size() <= 1)
144    return;
145
146  std::vector<DWARFDebugInfoEntryMinimal *> ParentChain;
147  DWARFDebugInfoEntryMinimal *SiblingChain = nullptr;
148  for (auto &DIE : DieArray) {
149    if (SiblingChain) {
150      SiblingChain->setSibling(&DIE);
151    }
152    if (const DWARFAbbreviationDeclaration *AbbrDecl =
153            DIE.getAbbreviationDeclarationPtr()) {
154      // Normal DIE.
155      if (AbbrDecl->hasChildren()) {
156        ParentChain.push_back(&DIE);
157        SiblingChain = nullptr;
158      } else {
159        SiblingChain = &DIE;
160      }
161    } else {
162      // NULL entry terminates the sibling chain.
163      SiblingChain = ParentChain.back();
164      ParentChain.pop_back();
165    }
166  }
167  assert(SiblingChain == nullptr || SiblingChain == &DieArray[0]);
168  assert(ParentChain.empty());
169}
170
171void DWARFUnit::extractDIEsToVector(
172    bool AppendCUDie, bool AppendNonCUDies,
173    std::vector<DWARFDebugInfoEntryMinimal> &Dies) const {
174  if (!AppendCUDie && !AppendNonCUDies)
175    return;
176
177  // Set the offset to that of the first DIE and calculate the start of the
178  // next compilation unit header.
179  uint32_t DIEOffset = Offset + getHeaderSize();
180  uint32_t NextCUOffset = getNextUnitOffset();
181  DWARFDebugInfoEntryMinimal DIE;
182  uint32_t Depth = 0;
183  bool IsCUDie = true;
184
185  while (DIEOffset < NextCUOffset && DIE.extractFast(this, &DIEOffset)) {
186    if (IsCUDie) {
187      if (AppendCUDie)
188        Dies.push_back(DIE);
189      if (!AppendNonCUDies)
190        break;
191      // The average bytes per DIE entry has been seen to be
192      // around 14-20 so let's pre-reserve the needed memory for
193      // our DIE entries accordingly.
194      Dies.reserve(Dies.size() + getDebugInfoSize() / 14);
195      IsCUDie = false;
196    } else {
197      Dies.push_back(DIE);
198    }
199
200    if (const DWARFAbbreviationDeclaration *AbbrDecl =
201            DIE.getAbbreviationDeclarationPtr()) {
202      // Normal DIE
203      if (AbbrDecl->hasChildren())
204        ++Depth;
205    } else {
206      // NULL DIE.
207      if (Depth > 0)
208        --Depth;
209      if (Depth == 0)
210        break;  // We are done with this compile unit!
211    }
212  }
213
214  // Give a little bit of info if we encounter corrupt DWARF (our offset
215  // should always terminate at or before the start of the next compilation
216  // unit header).
217  if (DIEOffset > NextCUOffset)
218    fprintf(stderr, "warning: DWARF compile unit extends beyond its "
219                    "bounds cu 0x%8.8x at 0x%8.8x'\n", getOffset(), DIEOffset);
220}
221
222size_t DWARFUnit::extractDIEsIfNeeded(bool CUDieOnly) {
223  if ((CUDieOnly && DieArray.size() > 0) ||
224      DieArray.size() > 1)
225    return 0; // Already parsed.
226
227  bool HasCUDie = DieArray.size() > 0;
228  extractDIEsToVector(!HasCUDie, !CUDieOnly, DieArray);
229
230  if (DieArray.empty())
231    return 0;
232
233  // If CU DIE was just parsed, copy several attribute values from it.
234  if (!HasCUDie) {
235    uint64_t BaseAddr =
236        DieArray[0].getAttributeValueAsAddress(this, DW_AT_low_pc, -1ULL);
237    if (BaseAddr == -1ULL)
238      BaseAddr = DieArray[0].getAttributeValueAsAddress(this, DW_AT_entry_pc, 0);
239    setBaseAddress(BaseAddr);
240    AddrOffsetSectionBase = DieArray[0].getAttributeValueAsSectionOffset(
241        this, DW_AT_GNU_addr_base, 0);
242    RangeSectionBase = DieArray[0].getAttributeValueAsSectionOffset(
243        this, DW_AT_ranges_base, 0);
244    // Don't fall back to DW_AT_GNU_ranges_base: it should be ignored for
245    // skeleton CU DIE, so that DWARF users not aware of it are not broken.
246  }
247
248  setDIERelations();
249  return DieArray.size();
250}
251
252DWARFUnit::DWOHolder::DWOHolder(StringRef DWOPath)
253    : DWOFile(), DWOContext(), DWOU(nullptr) {
254  auto Obj = object::ObjectFile::createObjectFile(DWOPath);
255  if (!Obj)
256    return;
257  DWOFile = std::move(Obj.get());
258  DWOContext.reset(
259      cast<DWARFContext>(DIContext::getDWARFContext(*DWOFile.getBinary())));
260  if (DWOContext->getNumDWOCompileUnits() > 0)
261    DWOU = DWOContext->getDWOCompileUnitAtIndex(0);
262}
263
264bool DWARFUnit::parseDWO() {
265  if (DWO.get())
266    return false;
267  extractDIEsIfNeeded(true);
268  if (DieArray.empty())
269    return false;
270  const char *DWOFileName =
271      DieArray[0].getAttributeValueAsString(this, DW_AT_GNU_dwo_name, nullptr);
272  if (!DWOFileName)
273    return false;
274  const char *CompilationDir =
275      DieArray[0].getAttributeValueAsString(this, DW_AT_comp_dir, nullptr);
276  SmallString<16> AbsolutePath;
277  if (sys::path::is_relative(DWOFileName) && CompilationDir != nullptr) {
278    sys::path::append(AbsolutePath, CompilationDir);
279  }
280  sys::path::append(AbsolutePath, DWOFileName);
281  DWO = llvm::make_unique<DWOHolder>(AbsolutePath);
282  DWARFUnit *DWOCU = DWO->getUnit();
283  // Verify that compile unit in .dwo file is valid.
284  if (!DWOCU || DWOCU->getDWOId() != getDWOId()) {
285    DWO.reset();
286    return false;
287  }
288  // Share .debug_addr and .debug_ranges section with compile unit in .dwo
289  DWOCU->setAddrOffsetSection(AddrOffsetSection, AddrOffsetSectionBase);
290  uint32_t DWORangesBase = DieArray[0].getRangesBaseAttribute(this, 0);
291  DWOCU->setRangesSection(RangeSection, DWORangesBase);
292  return true;
293}
294
295void DWARFUnit::clearDIEs(bool KeepCUDie) {
296  if (DieArray.size() > (unsigned)KeepCUDie) {
297    // std::vectors never get any smaller when resized to a smaller size,
298    // or when clear() or erase() are called, the size will report that it
299    // is smaller, but the memory allocated remains intact (call capacity()
300    // to see this). So we need to create a temporary vector and swap the
301    // contents which will cause just the internal pointers to be swapped
302    // so that when temporary vector goes out of scope, it will destroy the
303    // contents.
304    std::vector<DWARFDebugInfoEntryMinimal> TmpArray;
305    DieArray.swap(TmpArray);
306    // Save at least the compile unit DIE
307    if (KeepCUDie)
308      DieArray.push_back(TmpArray.front());
309  }
310}
311
312void DWARFUnit::collectAddressRanges(DWARFAddressRangesVector &CURanges) {
313  // First, check if CU DIE describes address ranges for the unit.
314  const auto &CUDIERanges = getCompileUnitDIE()->getAddressRanges(this);
315  if (!CUDIERanges.empty()) {
316    CURanges.insert(CURanges.end(), CUDIERanges.begin(), CUDIERanges.end());
317    return;
318  }
319
320  // This function is usually called if there in no .debug_aranges section
321  // in order to produce a compile unit level set of address ranges that
322  // is accurate. If the DIEs weren't parsed, then we don't want all dies for
323  // all compile units to stay loaded when they weren't needed. So we can end
324  // up parsing the DWARF and then throwing them all away to keep memory usage
325  // down.
326  const bool ClearDIEs = extractDIEsIfNeeded(false) > 1;
327  DieArray[0].collectChildrenAddressRanges(this, CURanges);
328
329  // Collect address ranges from DIEs in .dwo if necessary.
330  bool DWOCreated = parseDWO();
331  if (DWO.get())
332    DWO->getUnit()->collectAddressRanges(CURanges);
333  if (DWOCreated)
334    DWO.reset();
335
336  // Keep memory down by clearing DIEs if this generate function
337  // caused them to be parsed.
338  if (ClearDIEs)
339    clearDIEs(true);
340}
341
342const DWARFDebugInfoEntryMinimal *
343DWARFUnit::getSubprogramForAddress(uint64_t Address) {
344  extractDIEsIfNeeded(false);
345  for (const DWARFDebugInfoEntryMinimal &DIE : DieArray) {
346    if (DIE.isSubprogramDIE() &&
347        DIE.addressRangeContainsAddress(this, Address)) {
348      return &DIE;
349    }
350  }
351  return nullptr;
352}
353
354DWARFDebugInfoEntryInlinedChain
355DWARFUnit::getInlinedChainForAddress(uint64_t Address) {
356  // First, find a subprogram that contains the given address (the root
357  // of inlined chain).
358  const DWARFUnit *ChainCU = nullptr;
359  const DWARFDebugInfoEntryMinimal *SubprogramDIE =
360      getSubprogramForAddress(Address);
361  if (SubprogramDIE) {
362    ChainCU = this;
363  } else {
364    // Try to look for subprogram DIEs in the DWO file.
365    parseDWO();
366    if (DWO.get()) {
367      SubprogramDIE = DWO->getUnit()->getSubprogramForAddress(Address);
368      if (SubprogramDIE)
369        ChainCU = DWO->getUnit();
370    }
371  }
372
373  // Get inlined chain rooted at this subprogram DIE.
374  if (!SubprogramDIE)
375    return DWARFDebugInfoEntryInlinedChain();
376  return SubprogramDIE->getInlinedChainForAddress(ChainCU, Address);
377}
378