LLVM  16.0.0git
MachOLinkGraphBuilder.cpp
Go to the documentation of this file.
1 //=--------- MachOLinkGraphBuilder.cpp - MachO LinkGraph builder ----------===//
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 // Generic MachO LinkGraph buliding code.
10 //
11 //===----------------------------------------------------------------------===//
12 
13 #include "MachOLinkGraphBuilder.h"
14 
15 #define DEBUG_TYPE "jitlink"
16 
17 static const char *CommonSectionName = "__common";
18 
19 namespace llvm {
20 namespace jitlink {
21 
23 
25 
26  // We only operate on relocatable objects.
27  if (!Obj.isRelocatableObject())
28  return make_error<JITLinkError>("Object is not a relocatable MachO");
29 
30  if (auto Err = createNormalizedSections())
31  return std::move(Err);
32 
33  if (auto Err = createNormalizedSymbols())
34  return std::move(Err);
35 
36  if (auto Err = graphifyRegularSymbols())
37  return std::move(Err);
38 
39  if (auto Err = graphifySectionsWithCustomParsers())
40  return std::move(Err);
41 
42  if (auto Err = addRelocations())
43  return std::move(Err);
44 
45  return std::move(G);
46 }
47 
49  const object::MachOObjectFile &Obj, Triple TT,
50  LinkGraph::GetEdgeKindNameFunction GetEdgeKindName)
51  : Obj(Obj),
52  G(std::make_unique<LinkGraph>(
53  std::string(Obj.getFileName()), std::move(TT), getPointerSize(Obj),
54  getEndianness(Obj), std::move(GetEdgeKindName))) {
55  auto &MachHeader = Obj.getHeader64();
56  SubsectionsViaSymbols = MachHeader.flags & MachO::MH_SUBSECTIONS_VIA_SYMBOLS;
57 }
58 
61  assert(!CustomSectionParserFunctions.count(SectionName) &&
62  "Custom parser for this section already exists");
63  CustomSectionParserFunctions[SectionName] = std::move(Parser);
64 }
65 
67  if ((Desc & MachO::N_WEAK_DEF) || (Desc & MachO::N_WEAK_REF))
68  return Linkage::Weak;
69  return Linkage::Strong;
70 }
71 
73  if (Type & MachO::N_EXT) {
74  if ((Type & MachO::N_PEXT) || Name.startswith("l"))
75  return Scope::Hidden;
76  else
77  return Scope::Default;
78  }
79  return Scope::Local;
80 }
81 
83  return NSym.Desc & MachO::N_ALT_ENTRY;
84 }
85 
87  return (NSec.Flags & MachO::S_ATTR_DEBUG &&
88  strcmp(NSec.SegName, "__DWARF") == 0);
89 }
90 
92  switch (NSec.Flags & MachO::SECTION_TYPE) {
93  case MachO::S_ZEROFILL:
96  return true;
97  default:
98  return false;
99  }
100 }
101 
102 unsigned
103 MachOLinkGraphBuilder::getPointerSize(const object::MachOObjectFile &Obj) {
104  return Obj.is64Bit() ? 8 : 4;
105 }
106 
108 MachOLinkGraphBuilder::getEndianness(const object::MachOObjectFile &Obj) {
110 }
111 
112 Section &MachOLinkGraphBuilder::getCommonSection() {
113  if (!CommonSection)
114  CommonSection = &G->createSection(CommonSectionName,
116  return *CommonSection;
117 }
118 
119 Error MachOLinkGraphBuilder::createNormalizedSections() {
120  // Build normalized sections. Verifies that section data is in-range (for
121  // sections with content) and that address ranges are non-overlapping.
122 
123  LLVM_DEBUG(dbgs() << "Creating normalized sections...\n");
124 
125  for (auto &SecRef : Obj.sections()) {
126  NormalizedSection NSec;
127  uint32_t DataOffset = 0;
128 
129  auto SecIndex = Obj.getSectionIndex(SecRef.getRawDataRefImpl());
130 
131  if (Obj.is64Bit()) {
132  const MachO::section_64 &Sec64 =
133  Obj.getSection64(SecRef.getRawDataRefImpl());
134 
135  memcpy(&NSec.SectName, &Sec64.sectname, 16);
136  NSec.SectName[16] = '\0';
137  memcpy(&NSec.SegName, Sec64.segname, 16);
138  NSec.SegName[16] = '\0';
139 
140  NSec.Address = orc::ExecutorAddr(Sec64.addr);
141  NSec.Size = Sec64.size;
142  NSec.Alignment = 1ULL << Sec64.align;
143  NSec.Flags = Sec64.flags;
144  DataOffset = Sec64.offset;
145  } else {
146  const MachO::section &Sec32 = Obj.getSection(SecRef.getRawDataRefImpl());
147 
148  memcpy(&NSec.SectName, &Sec32.sectname, 16);
149  NSec.SectName[16] = '\0';
150  memcpy(&NSec.SegName, Sec32.segname, 16);
151  NSec.SegName[16] = '\0';
152 
153  NSec.Address = orc::ExecutorAddr(Sec32.addr);
154  NSec.Size = Sec32.size;
155  NSec.Alignment = 1ULL << Sec32.align;
156  NSec.Flags = Sec32.flags;
157  DataOffset = Sec32.offset;
158  }
159 
160  LLVM_DEBUG({
161  dbgs() << " " << NSec.SegName << "," << NSec.SectName << ": "
162  << formatv("{0:x16}", NSec.Address) << " -- "
163  << formatv("{0:x16}", NSec.Address + NSec.Size)
164  << ", align: " << NSec.Alignment << ", index: " << SecIndex
165  << "\n";
166  });
167 
168  // Get the section data if any.
169  if (!isZeroFillSection(NSec)) {
170  if (DataOffset + NSec.Size > Obj.getData().size())
171  return make_error<JITLinkError>(
172  "Section data extends past end of file");
173 
174  NSec.Data = Obj.getData().data() + DataOffset;
175  }
176 
177  // Get prot flags.
178  // FIXME: Make sure this test is correct (it's probably missing cases
179  // as-is).
180  orc::MemProt Prot;
181  if (NSec.Flags & MachO::S_ATTR_PURE_INSTRUCTIONS)
183  else
185 
186  auto FullyQualifiedName =
187  G->allocateString(StringRef(NSec.SegName) + "," + NSec.SectName);
188  NSec.GraphSection = &G->createSection(
189  StringRef(FullyQualifiedName.data(), FullyQualifiedName.size()), Prot);
190 
191  IndexToSection.insert(std::make_pair(SecIndex, std::move(NSec)));
192  }
193 
194  std::vector<NormalizedSection *> Sections;
195  Sections.reserve(IndexToSection.size());
196  for (auto &KV : IndexToSection)
197  Sections.push_back(&KV.second);
198 
199  // If we didn't end up creating any sections then bail out. The code below
200  // assumes that we have at least one section.
201  if (Sections.empty())
202  return Error::success();
203 
204  llvm::sort(Sections,
205  [](const NormalizedSection *LHS, const NormalizedSection *RHS) {
206  assert(LHS && RHS && "Null section?");
207  if (LHS->Address != RHS->Address)
208  return LHS->Address < RHS->Address;
209  return LHS->Size < RHS->Size;
210  });
211 
212  for (unsigned I = 0, E = Sections.size() - 1; I != E; ++I) {
213  auto &Cur = *Sections[I];
214  auto &Next = *Sections[I + 1];
215  if (Next.Address < Cur.Address + Cur.Size)
216  return make_error<JITLinkError>(
217  "Address range for section " +
218  formatv("\"{0}/{1}\" [ {2:x16} -- {3:x16} ] ", Cur.SegName,
219  Cur.SectName, Cur.Address, Cur.Address + Cur.Size) +
220  "overlaps section \"" + Next.SegName + "/" + Next.SectName + "\"" +
221  formatv("\"{0}/{1}\" [ {2:x16} -- {3:x16} ] ", Next.SegName,
222  Next.SectName, Next.Address, Next.Address + Next.Size));
223  }
224 
225  return Error::success();
226 }
227 
228 Error MachOLinkGraphBuilder::createNormalizedSymbols() {
229  LLVM_DEBUG(dbgs() << "Creating normalized symbols...\n");
230 
231  for (auto &SymRef : Obj.symbols()) {
232 
233  unsigned SymbolIndex = Obj.getSymbolIndex(SymRef.getRawDataRefImpl());
234  uint64_t Value;
235  uint32_t NStrX;
236  uint8_t Type;
237  uint8_t Sect;
238  uint16_t Desc;
239 
240  if (Obj.is64Bit()) {
241  const MachO::nlist_64 &NL64 =
242  Obj.getSymbol64TableEntry(SymRef.getRawDataRefImpl());
243  Value = NL64.n_value;
244  NStrX = NL64.n_strx;
245  Type = NL64.n_type;
246  Sect = NL64.n_sect;
247  Desc = NL64.n_desc;
248  } else {
249  const MachO::nlist &NL32 =
250  Obj.getSymbolTableEntry(SymRef.getRawDataRefImpl());
251  Value = NL32.n_value;
252  NStrX = NL32.n_strx;
253  Type = NL32.n_type;
254  Sect = NL32.n_sect;
255  Desc = NL32.n_desc;
256  }
257 
258  // Skip stabs.
259  // FIXME: Are there other symbols we should be skipping?
260  if (Type & MachO::N_STAB)
261  continue;
262 
263  Optional<StringRef> Name;
264  if (NStrX) {
265  if (auto NameOrErr = SymRef.getName())
266  Name = *NameOrErr;
267  else
268  return NameOrErr.takeError();
269  }
270 
271  LLVM_DEBUG({
272  dbgs() << " ";
273  if (!Name)
274  dbgs() << "<anonymous symbol>";
275  else
276  dbgs() << *Name;
277  dbgs() << ": value = " << formatv("{0:x16}", Value)
278  << ", type = " << formatv("{0:x2}", Type)
279  << ", desc = " << formatv("{0:x4}", Desc) << ", sect = ";
280  if (Sect)
281  dbgs() << static_cast<unsigned>(Sect - 1);
282  else
283  dbgs() << "none";
284  dbgs() << "\n";
285  });
286 
287  // If this symbol has a section, verify that the addresses line up.
288  if (Sect != 0) {
289  auto NSec = findSectionByIndex(Sect - 1);
290  if (!NSec)
291  return NSec.takeError();
292 
293  if (orc::ExecutorAddr(Value) < NSec->Address ||
294  orc::ExecutorAddr(Value) > NSec->Address + NSec->Size)
295  return make_error<JITLinkError>("Address " + formatv("{0:x}", Value) +
296  " for symbol " + *Name +
297  " does not fall within section");
298 
299  if (!NSec->GraphSection) {
300  LLVM_DEBUG({
301  dbgs() << " Skipping: Symbol is in section " << NSec->SegName << "/"
302  << NSec->SectName
303  << " which has no associated graph section.\n";
304  });
305  continue;
306  }
307  }
308 
309  IndexToSymbol[SymbolIndex] =
310  &createNormalizedSymbol(*Name, Value, Type, Sect, Desc,
311  getLinkage(Desc), getScope(*Name, Type));
312  }
313 
314  return Error::success();
315 }
316 
317 void MachOLinkGraphBuilder::addSectionStartSymAndBlock(
318  unsigned SecIndex, Section &GraphSec, orc::ExecutorAddr Address,
319  const char *Data, orc::ExecutorAddrDiff Size, uint32_t Alignment,
320  bool IsLive) {
321  Block &B =
322  Data ? G->createContentBlock(GraphSec, ArrayRef<char>(Data, Size),
323  Address, Alignment, 0)
324  : G->createZeroFillBlock(GraphSec, Size, Address, Alignment, 0);
325  auto &Sym = G->addAnonymousSymbol(B, 0, Size, false, IsLive);
326  auto SecI = IndexToSection.find(SecIndex);
327  assert(SecI != IndexToSection.end() && "SecIndex invalid");
328  auto &NSec = SecI->second;
329  assert(!NSec.CanonicalSymbols.count(Sym.getAddress()) &&
330  "Anonymous block start symbol clashes with existing symbol address");
331  NSec.CanonicalSymbols[Sym.getAddress()] = &Sym;
332 }
333 
334 Error MachOLinkGraphBuilder::graphifyRegularSymbols() {
335 
336  LLVM_DEBUG(dbgs() << "Creating graph symbols...\n");
337 
338  /// We only have 256 section indexes: Use a vector rather than a map.
339  std::vector<std::vector<NormalizedSymbol *>> SecIndexToSymbols;
340  SecIndexToSymbols.resize(256);
341 
342  // Create commons, externs, and absolutes, and partition all other symbols by
343  // section.
344  for (auto &KV : IndexToSymbol) {
345  auto &NSym = *KV.second;
346 
347  switch (NSym.Type & MachO::N_TYPE) {
348  case MachO::N_UNDF:
349  if (NSym.Value) {
350  if (!NSym.Name)
351  return make_error<JITLinkError>("Anonymous common symbol at index " +
352  Twine(KV.first));
353  NSym.GraphSymbol = &G->addDefinedSymbol(
354  G->createZeroFillBlock(getCommonSection(),
355  orc::ExecutorAddrDiff(NSym.Value),
356  orc::ExecutorAddr(),
357  1ull << MachO::GET_COMM_ALIGN(NSym.Desc), 0),
358  0, *NSym.Name, orc::ExecutorAddrDiff(NSym.Value), Linkage::Strong,
359  NSym.S, false, NSym.Desc & MachO::N_NO_DEAD_STRIP);
360  } else {
361  if (!NSym.Name)
362  return make_error<JITLinkError>("Anonymous external symbol at "
363  "index " +
364  Twine(KV.first));
365  NSym.GraphSymbol = &G->addExternalSymbol(
366  *NSym.Name, 0, (NSym.Desc & MachO::N_WEAK_REF) != 0);
367  }
368  break;
369  case MachO::N_ABS:
370  if (!NSym.Name)
371  return make_error<JITLinkError>("Anonymous absolute symbol at index " +
372  Twine(KV.first));
373  NSym.GraphSymbol = &G->addAbsoluteSymbol(
374  *NSym.Name, orc::ExecutorAddr(NSym.Value), 0, Linkage::Strong,
375  getScope(*NSym.Name, NSym.Type), NSym.Desc & MachO::N_NO_DEAD_STRIP);
376  break;
377  case MachO::N_SECT:
378  SecIndexToSymbols[NSym.Sect - 1].push_back(&NSym);
379  break;
380  case MachO::N_PBUD:
381  return make_error<JITLinkError>(
382  "Unupported N_PBUD symbol " +
383  (NSym.Name ? ("\"" + *NSym.Name + "\"") : Twine("<anon>")) +
384  " at index " + Twine(KV.first));
385  case MachO::N_INDR:
386  return make_error<JITLinkError>(
387  "Unupported N_INDR symbol " +
388  (NSym.Name ? ("\"" + *NSym.Name + "\"") : Twine("<anon>")) +
389  " at index " + Twine(KV.first));
390  default:
391  return make_error<JITLinkError>(
392  "Unrecognized symbol type " + Twine(NSym.Type & MachO::N_TYPE) +
393  " for symbol " +
394  (NSym.Name ? ("\"" + *NSym.Name + "\"") : Twine("<anon>")) +
395  " at index " + Twine(KV.first));
396  }
397  }
398 
399  // Loop over sections performing regular graphification for those that
400  // don't have custom parsers.
401  for (auto &KV : IndexToSection) {
402  auto SecIndex = KV.first;
403  auto &NSec = KV.second;
404 
405  if (!NSec.GraphSection) {
406  LLVM_DEBUG({
407  dbgs() << " " << NSec.SegName << "/" << NSec.SectName
408  << " has no graph section. Skipping.\n";
409  });
410  continue;
411  }
412 
413  // Skip sections with custom parsers.
414  if (CustomSectionParserFunctions.count(NSec.GraphSection->getName())) {
415  LLVM_DEBUG({
416  dbgs() << " Skipping section " << NSec.GraphSection->getName()
417  << " as it has a custom parser.\n";
418  });
419  continue;
420  } else if ((NSec.Flags & MachO::SECTION_TYPE) ==
422  if (auto Err = graphifyCStringSection(
423  NSec, std::move(SecIndexToSymbols[SecIndex])))
424  return Err;
425  continue;
426  } else
427  LLVM_DEBUG({
428  dbgs() << " Graphifying regular section "
429  << NSec.GraphSection->getName() << "...\n";
430  });
431 
432  bool SectionIsNoDeadStrip = NSec.Flags & MachO::S_ATTR_NO_DEAD_STRIP;
433  bool SectionIsText = NSec.Flags & MachO::S_ATTR_PURE_INSTRUCTIONS;
434 
435  auto &SecNSymStack = SecIndexToSymbols[SecIndex];
436 
437  // If this section is non-empty but there are no symbols covering it then
438  // create one block and anonymous symbol to cover the entire section.
439  if (SecNSymStack.empty()) {
440  if (NSec.Size > 0) {
441  LLVM_DEBUG({
442  dbgs() << " Section non-empty, but contains no symbols. "
443  "Creating anonymous block to cover "
444  << formatv("{0:x16}", NSec.Address) << " -- "
445  << formatv("{0:x16}", NSec.Address + NSec.Size) << "\n";
446  });
447  addSectionStartSymAndBlock(SecIndex, *NSec.GraphSection, NSec.Address,
448  NSec.Data, NSec.Size, NSec.Alignment,
449  SectionIsNoDeadStrip);
450  } else
451  LLVM_DEBUG({
452  dbgs() << " Section empty and contains no symbols. Skipping.\n";
453  });
454  continue;
455  }
456 
457  // Sort the symbol stack in by address, alt-entry status, scope, and name.
458  // We sort in reverse order so that symbols will be visited in the right
459  // order when we pop off the stack below.
460  llvm::sort(SecNSymStack, [](const NormalizedSymbol *LHS,
461  const NormalizedSymbol *RHS) {
462  if (LHS->Value != RHS->Value)
463  return LHS->Value > RHS->Value;
464  if (isAltEntry(*LHS) != isAltEntry(*RHS))
465  return isAltEntry(*RHS);
466  if (LHS->S != RHS->S)
467  return static_cast<uint8_t>(LHS->S) < static_cast<uint8_t>(RHS->S);
468  return LHS->Name < RHS->Name;
469  });
470 
471  // The first symbol in a section can not be an alt-entry symbol.
472  if (!SecNSymStack.empty() && isAltEntry(*SecNSymStack.back()))
473  return make_error<JITLinkError>(
474  "First symbol in " + NSec.GraphSection->getName() + " is alt-entry");
475 
476  // If the section is non-empty but there is no symbol covering the start
477  // address then add an anonymous one.
478  if (orc::ExecutorAddr(SecNSymStack.back()->Value) != NSec.Address) {
479  auto AnonBlockSize =
480  orc::ExecutorAddr(SecNSymStack.back()->Value) - NSec.Address;
481  LLVM_DEBUG({
482  dbgs() << " Section start not covered by symbol. "
483  << "Creating anonymous block to cover [ " << NSec.Address
484  << " -- " << (NSec.Address + AnonBlockSize) << " ]\n";
485  });
486  addSectionStartSymAndBlock(SecIndex, *NSec.GraphSection, NSec.Address,
487  NSec.Data, AnonBlockSize, NSec.Alignment,
488  SectionIsNoDeadStrip);
489  }
490 
491  // Visit section symbols in order by popping off the reverse-sorted stack,
492  // building graph symbols as we go.
493  //
494  // If MH_SUBSECTIONS_VIA_SYMBOLS is set we'll build a block for each
495  // alt-entry chain.
496  //
497  // If MH_SUBSECTIONS_VIA_SYMBOLS is not set then we'll just build one block
498  // for the whole section.
499  while (!SecNSymStack.empty()) {
500  SmallVector<NormalizedSymbol *, 8> BlockSyms;
501 
502  // Get the symbols in this alt-entry chain, or the whole section (if
503  // !SubsectionsViaSymbols).
504  BlockSyms.push_back(SecNSymStack.back());
505  SecNSymStack.pop_back();
506  while (!SecNSymStack.empty() &&
507  (isAltEntry(*SecNSymStack.back()) ||
508  SecNSymStack.back()->Value == BlockSyms.back()->Value ||
509  !SubsectionsViaSymbols)) {
510  BlockSyms.push_back(SecNSymStack.back());
511  SecNSymStack.pop_back();
512  }
513 
514  // BlockNSyms now contains the block symbols in reverse canonical order.
515  auto BlockStart = orc::ExecutorAddr(BlockSyms.front()->Value);
516  orc::ExecutorAddr BlockEnd =
517  SecNSymStack.empty() ? NSec.Address + NSec.Size
518  : orc::ExecutorAddr(SecNSymStack.back()->Value);
519  orc::ExecutorAddrDiff BlockOffset = BlockStart - NSec.Address;
520  orc::ExecutorAddrDiff BlockSize = BlockEnd - BlockStart;
521 
522  LLVM_DEBUG({
523  dbgs() << " Creating block for " << formatv("{0:x16}", BlockStart)
524  << " -- " << formatv("{0:x16}", BlockEnd) << ": "
525  << NSec.GraphSection->getName() << " + "
526  << formatv("{0:x16}", BlockOffset) << " with "
527  << BlockSyms.size() << " symbol(s)...\n";
528  });
529 
530  Block &B =
531  NSec.Data
532  ? G->createContentBlock(
533  *NSec.GraphSection,
534  ArrayRef<char>(NSec.Data + BlockOffset, BlockSize),
535  BlockStart, NSec.Alignment, BlockStart % NSec.Alignment)
536  : G->createZeroFillBlock(*NSec.GraphSection, BlockSize,
537  BlockStart, NSec.Alignment,
538  BlockStart % NSec.Alignment);
539 
540  Optional<orc::ExecutorAddr> LastCanonicalAddr;
541  auto SymEnd = BlockEnd;
542  while (!BlockSyms.empty()) {
543  auto &NSym = *BlockSyms.back();
544  BlockSyms.pop_back();
545 
546  bool SymLive =
547  (NSym.Desc & MachO::N_NO_DEAD_STRIP) || SectionIsNoDeadStrip;
548 
549  auto &Sym = createStandardGraphSymbol(
550  NSym, B, SymEnd - orc::ExecutorAddr(NSym.Value), SectionIsText,
551  SymLive, LastCanonicalAddr != orc::ExecutorAddr(NSym.Value));
552 
553  if (LastCanonicalAddr != Sym.getAddress()) {
554  if (LastCanonicalAddr)
555  SymEnd = *LastCanonicalAddr;
556  LastCanonicalAddr = Sym.getAddress();
557  }
558  }
559  }
560  }
561 
562  return Error::success();
563 }
564 
565 Symbol &MachOLinkGraphBuilder::createStandardGraphSymbol(NormalizedSymbol &NSym,
566  Block &B, size_t Size,
567  bool IsText,
568  bool IsNoDeadStrip,
569  bool IsCanonical) {
570 
571  LLVM_DEBUG({
572  dbgs() << " " << formatv("{0:x16}", NSym.Value) << " -- "
573  << formatv("{0:x16}", NSym.Value + Size) << ": ";
574  if (!NSym.Name)
575  dbgs() << "<anonymous symbol>";
576  else
577  dbgs() << NSym.Name;
578  if (IsText)
579  dbgs() << " [text]";
580  if (IsNoDeadStrip)
581  dbgs() << " [no-dead-strip]";
582  if (!IsCanonical)
583  dbgs() << " [non-canonical]";
584  dbgs() << "\n";
585  });
586 
587  auto SymOffset = orc::ExecutorAddr(NSym.Value) - B.getAddress();
588  auto &Sym =
589  NSym.Name
590  ? G->addDefinedSymbol(B, SymOffset, *NSym.Name, Size, NSym.L, NSym.S,
591  IsText, IsNoDeadStrip)
592  : G->addAnonymousSymbol(B, SymOffset, Size, IsText, IsNoDeadStrip);
593  NSym.GraphSymbol = &Sym;
594 
595  if (IsCanonical)
596  setCanonicalSymbol(getSectionByIndex(NSym.Sect - 1), Sym);
597 
598  return Sym;
599 }
600 
601 Error MachOLinkGraphBuilder::graphifySectionsWithCustomParsers() {
602  // Graphify special sections.
603  for (auto &KV : IndexToSection) {
604  auto &NSec = KV.second;
605 
606  // Skip non-graph sections.
607  if (!NSec.GraphSection)
608  continue;
609 
610  auto HI = CustomSectionParserFunctions.find(NSec.GraphSection->getName());
611  if (HI != CustomSectionParserFunctions.end()) {
612  auto &Parse = HI->second;
613  if (auto Err = Parse(NSec))
614  return Err;
615  }
616  }
617 
618  return Error::success();
619 }
620 
621 Error MachOLinkGraphBuilder::graphifyCStringSection(
622  NormalizedSection &NSec, std::vector<NormalizedSymbol *> NSyms) {
623  assert(NSec.GraphSection && "C string literal section missing graph section");
624  assert(NSec.Data && "C string literal section has no data");
625 
626  LLVM_DEBUG({
627  dbgs() << " Graphifying C-string literal section "
628  << NSec.GraphSection->getName() << "\n";
629  });
630 
631  if (NSec.Data[NSec.Size - 1] != '\0')
632  return make_error<JITLinkError>("C string literal section " +
633  NSec.GraphSection->getName() +
634  " does not end with null terminator");
635 
636  /// Sort into reverse order to use as a stack.
637  llvm::sort(NSyms,
638  [](const NormalizedSymbol *LHS, const NormalizedSymbol *RHS) {
639  if (LHS->Value != RHS->Value)
640  return LHS->Value > RHS->Value;
641  if (LHS->L != RHS->L)
642  return LHS->L > RHS->L;
643  if (LHS->S != RHS->S)
644  return LHS->S > RHS->S;
645  if (RHS->Name) {
646  if (!LHS->Name)
647  return true;
648  return *LHS->Name > *RHS->Name;
649  }
650  return false;
651  });
652 
653  bool SectionIsNoDeadStrip = NSec.Flags & MachO::S_ATTR_NO_DEAD_STRIP;
654  bool SectionIsText = NSec.Flags & MachO::S_ATTR_PURE_INSTRUCTIONS;
655  orc::ExecutorAddrDiff BlockStart = 0;
656 
657  // Scan section for null characters.
658  for (size_t I = 0; I != NSec.Size; ++I)
659  if (NSec.Data[I] == '\0') {
660  size_t BlockSize = I + 1 - BlockStart;
661  // Create a block for this null terminated string.
662  auto &B = G->createContentBlock(*NSec.GraphSection,
663  {NSec.Data + BlockStart, BlockSize},
664  NSec.Address + BlockStart, NSec.Alignment,
665  BlockStart % NSec.Alignment);
666 
667  LLVM_DEBUG({
668  dbgs() << " Created block " << B.getRange()
669  << ", align = " << B.getAlignment()
670  << ", align-ofs = " << B.getAlignmentOffset() << " for \"";
671  for (size_t J = 0; J != std::min(B.getSize(), size_t(16)); ++J)
672  switch (B.getContent()[J]) {
673  case '\0': break;
674  case '\n': dbgs() << "\\n"; break;
675  case '\t': dbgs() << "\\t"; break;
676  default: dbgs() << B.getContent()[J]; break;
677  }
678  if (B.getSize() > 16)
679  dbgs() << "...";
680  dbgs() << "\"\n";
681  });
682 
683  // If there's no symbol at the start of this block then create one.
684  if (NSyms.empty() ||
685  orc::ExecutorAddr(NSyms.back()->Value) != B.getAddress()) {
686  auto &S = G->addAnonymousSymbol(B, 0, BlockSize, false, false);
687  setCanonicalSymbol(NSec, S);
688  LLVM_DEBUG({
689  dbgs() << " Adding symbol for c-string block " << B.getRange()
690  << ": <anonymous symbol> at offset 0\n";
691  });
692  }
693 
694  // Process any remaining symbols that point into this block.
695  auto LastCanonicalAddr = B.getAddress() + BlockSize;
696  while (!NSyms.empty() && orc::ExecutorAddr(NSyms.back()->Value) <
697  B.getAddress() + BlockSize) {
698  auto &NSym = *NSyms.back();
699  size_t SymSize = (B.getAddress() + BlockSize) -
700  orc::ExecutorAddr(NSyms.back()->Value);
701  bool SymLive =
702  (NSym.Desc & MachO::N_NO_DEAD_STRIP) || SectionIsNoDeadStrip;
703 
704  bool IsCanonical = false;
705  if (LastCanonicalAddr != orc::ExecutorAddr(NSym.Value)) {
706  IsCanonical = true;
707  LastCanonicalAddr = orc::ExecutorAddr(NSym.Value);
708  }
709 
710  auto &Sym = createStandardGraphSymbol(NSym, B, SymSize, SectionIsText,
711  SymLive, IsCanonical);
712  (void)Sym;
713  LLVM_DEBUG({
714  dbgs() << " Adding symbol for c-string block " << B.getRange()
715  << ": "
716  << (Sym.hasName() ? Sym.getName() : "<anonymous symbol>")
717  << " at offset " << formatv("{0:x}", Sym.getOffset()) << "\n";
718  });
719 
720  NSyms.pop_back();
721  }
722 
723  BlockStart += BlockSize;
724  }
725 
726  return Error::success();
727 }
728 
730  auto *CUSec = G.findSectionByName(CompactUnwindSectionName);
731  if (!CUSec)
732  return Error::success();
733 
734  if (!G.getTargetTriple().isOSBinFormatMachO())
735  return make_error<JITLinkError>(
736  "Error linking " + G.getName() +
737  ": compact unwind splitting not supported on non-macho target " +
738  G.getTargetTriple().str());
739 
740  unsigned CURecordSize = 0;
741  unsigned PersonalityEdgeOffset = 0;
742  unsigned LSDAEdgeOffset = 0;
743  switch (G.getTargetTriple().getArch()) {
744  case Triple::aarch64:
745  case Triple::x86_64:
746  // 64-bit compact-unwind record format:
747  // Range start: 8 bytes.
748  // Range size: 4 bytes.
749  // CU encoding: 4 bytes.
750  // Personality: 8 bytes.
751  // LSDA: 8 bytes.
752  CURecordSize = 32;
753  PersonalityEdgeOffset = 16;
754  LSDAEdgeOffset = 24;
755  break;
756  default:
757  return make_error<JITLinkError>(
758  "Error linking " + G.getName() +
759  ": compact unwind splitting not supported on " +
760  G.getTargetTriple().getArchName());
761  }
762 
763  std::vector<Block *> OriginalBlocks(CUSec->blocks().begin(),
764  CUSec->blocks().end());
765  LLVM_DEBUG({
766  dbgs() << "In " << G.getName() << " splitting compact unwind section "
767  << CompactUnwindSectionName << " containing "
768  << OriginalBlocks.size() << " initial blocks...\n";
769  });
770 
771  while (!OriginalBlocks.empty()) {
772  auto *B = OriginalBlocks.back();
773  OriginalBlocks.pop_back();
774 
775  if (B->getSize() == 0) {
776  LLVM_DEBUG({
777  dbgs() << " Skipping empty block at "
778  << formatv("{0:x16}", B->getAddress()) << "\n";
779  });
780  continue;
781  }
782 
783  LLVM_DEBUG({
784  dbgs() << " Splitting block at " << formatv("{0:x16}", B->getAddress())
785  << " into " << (B->getSize() / CURecordSize)
786  << " compact unwind record(s)\n";
787  });
788 
789  if (B->getSize() % CURecordSize)
790  return make_error<JITLinkError>(
791  "Error splitting compact unwind record in " + G.getName() +
792  ": block at " + formatv("{0:x}", B->getAddress()) + " has size " +
793  formatv("{0:x}", B->getSize()) +
794  " (not a multiple of CU record size of " +
795  formatv("{0:x}", CURecordSize) + ")");
796 
797  unsigned NumBlocks = B->getSize() / CURecordSize;
799 
800  for (unsigned I = 0; I != NumBlocks; ++I) {
801  auto &CURec = G.splitBlock(*B, CURecordSize, &C);
802  bool AddedKeepAlive = false;
803 
804  for (auto &E : CURec.edges()) {
805  if (E.getOffset() == 0) {
806  LLVM_DEBUG({
807  dbgs() << " Updating compact unwind record at "
808  << formatv("{0:x16}", CURec.getAddress()) << " to point to "
809  << (E.getTarget().hasName() ? E.getTarget().getName()
810  : StringRef())
811  << " (at " << formatv("{0:x16}", E.getTarget().getAddress())
812  << ")\n";
813  });
814 
815  if (E.getTarget().isExternal())
816  return make_error<JITLinkError>(
817  "Error adding keep-alive edge for compact unwind record at " +
818  formatv("{0:x}", CURec.getAddress()) + ": target " +
819  E.getTarget().getName() + " is an external symbol");
820  auto &TgtBlock = E.getTarget().getBlock();
821  auto &CURecSym =
822  G.addAnonymousSymbol(CURec, 0, CURecordSize, false, false);
823  TgtBlock.addEdge(Edge::KeepAlive, 0, CURecSym, 0);
824  AddedKeepAlive = true;
825  } else if (E.getOffset() != PersonalityEdgeOffset &&
826  E.getOffset() != LSDAEdgeOffset)
827  return make_error<JITLinkError>("Unexpected edge at offset " +
828  formatv("{0:x}", E.getOffset()) +
829  " in compact unwind record at " +
830  formatv("{0:x}", CURec.getAddress()));
831  }
832 
833  if (!AddedKeepAlive)
834  return make_error<JITLinkError>(
835  "Error adding keep-alive edge for compact unwind record at " +
836  formatv("{0:x}", CURec.getAddress()) +
837  ": no outgoing target edge at offset 0");
838  }
839  }
840  return Error::success();
841 }
842 
843 } // end namespace jitlink
844 } // end namespace llvm
llvm::MachO::S_CSTRING_LITERALS
@ S_CSTRING_LITERALS
S_CSTRING_LITERALS - Section with literal C strings.
Definition: MachO.h:131
llvm::MachO::S_ATTR_NO_DEAD_STRIP
@ S_ATTR_NO_DEAD_STRIP
S_ATTR_NO_DEAD_STRIP - No dead stripping.
Definition: MachO.h:200
getPointerSize
static uint64_t getPointerSize(const Value *V, const DataLayout &DL, const TargetLibraryInfo &TLI, const Function *F)
Definition: DeadStoreElimination.cpp:209
llvm::AArch64CC::HI
@ HI
Definition: AArch64BaseInfo.h:263
llvm
This is an optimization pass for GlobalISel generic memory operations.
Definition: AddressRanges.h:18
BlockSize
static const int BlockSize
Definition: TarWriter.cpp:33
llvm::lltok::Error
@ Error
Definition: LLToken.h:21
llvm::MachO::N_EXT
@ N_EXT
Definition: MachO.h:310
llvm::object::MachOObjectFile::getSectionIndex
uint64_t getSectionIndex(DataRefImpl Sec) const override
Definition: MachOObjectFile.cpp:1932
llvm::object::Binary::isLittleEndian
bool isLittleEndian() const
Definition: Binary.h:152
llvm::Error::success
static ErrorSuccess success()
Create a success value.
Definition: Error.h:329
llvm::Triple
Triple - Helper class for working with autoconf configuration names.
Definition: Triple.h:44
llvm::MachO::S_GB_ZEROFILL
@ S_GB_ZEROFILL
S_GB_ZEROFILL - Zero fill on demand section (that can be larger than 4 gigabytes).
Definition: MachO.h:155
llvm::StringMap::end
iterator end()
Definition: StringMap.h:204
llvm::MachO::GET_COMM_ALIGN
uint8_t GET_COMM_ALIGN(uint16_t n_desc)
Definition: MachO.h:1543
llvm::Type
The instances of the Type class are immutable: once they are created, they are never changed.
Definition: Type.h:45
llvm::Triple::x86_64
@ x86_64
Definition: Triple.h:86
llvm::Optional
Definition: APInt.h:33
llvm::MachO::N_WEAK_DEF
@ N_WEAK_DEF
Definition: MachO.h:346
llvm::MachO::S_ZEROFILL
@ S_ZEROFILL
S_ZEROFILL - Zero fill on demand section.
Definition: MachO.h:129
llvm::MachO::N_UNDF
@ N_UNDF
Definition: MachO.h:316
llvm::Expected
Tagged union holding either a T or a Error.
Definition: APFloat.h:41
RHS
Value * RHS
Definition: X86PartialReduction.cpp:76
llvm::MachO::SECTION_TYPE
@ SECTION_TYPE
Definition: MachO.h:114
llvm::StringMap::find
iterator find(StringRef Key)
Definition: StringMap.h:217
llvm::Data
@ Data
Definition: SIMachineScheduler.h:55
llvm::object::MachOObjectFile::getSymbolIndex
uint64_t getSymbolIndex(DataRefImpl Symb) const
Definition: MachOObjectFile.cpp:2618
LLVM_DEBUG
#define LLVM_DEBUG(X)
Definition: Debug.h:101
llvm::object::MachOObjectFile::getSection
Expected< SectionRef > getSection(unsigned SectionIndex) const
Definition: MachOObjectFile.cpp:2001
llvm::object::ObjectFile::symbols
symbol_iterator_range symbols() const
Definition: ObjectFile.h:319
llvm::dbgs
raw_ostream & dbgs()
dbgs() - This returns a reference to a raw_ostream for debugging messages.
Definition: Debug.cpp:163
LHS
Value * LHS
Definition: X86PartialReduction.cpp:75
llvm::formatv
auto formatv(const char *Fmt, Ts &&... Vals) -> formatv_object< decltype(std::make_tuple(detail::build_format_adapter(std::forward< Ts >(Vals))...))>
Definition: FormatVariadic.h:251
E
static GCRegistry::Add< CoreCLRGC > E("coreclr", "CoreCLR-compatible GC")
llvm::support::little
@ little
Definition: Endian.h:27
C
(vector float) vec_cmpeq(*A, *B) C
Definition: README_ALTIVEC.txt:86
llvm::orc::MemProt::Read
@ Read
llvm::MachO::N_TYPE
@ N_TYPE
Definition: MachO.h:309
llvm::MachO::N_STAB
@ N_STAB
Definition: MachO.h:307
llvm::MachO::N_PBUD
@ N_PBUD
Definition: MachO.h:319
getFileName
static Expected< StringRef > getFileName(const DebugStringTableSubsectionRef &Strings, const DebugChecksumsSubsectionRef &Checksums, uint32_t FileID)
Definition: CodeViewYAMLDebugSections.cpp:555
llvm::MachO::S_ATTR_DEBUG
@ S_ATTR_DEBUG
S_ATTR_DEBUG - A debug section.
Definition: MachO.h:207
B
static GCRegistry::Add< OcamlGC > B("ocaml", "ocaml 3.10-compatible GC")
llvm::orc::ExecutorAddrDiff
uint64_t ExecutorAddrDiff
Definition: ExecutorAddress.h:28
llvm::AMDGPU::ElfNote::SectionName
const char SectionName[]
Definition: AMDGPUPTNote.h:24
llvm::StringRef::data
const char * data() const
data - Get a pointer to the start of the string (which may not be null terminated).
Definition: StringRef.h:131
llvm::MachO::N_WEAK_REF
@ N_WEAK_REF
Definition: MachO.h:345
llvm::object::MachOObjectFile::getSection64
MachO::section_64 getSection64(DataRefImpl DRI) const
Definition: MachOObjectFile.cpp:4601
llvm::sort
void sort(IteratorTy Start, IteratorTy End)
Definition: STLExtras.h:1538
llvm::object::MachOObjectFile
Definition: MachO.h:406
G
const DataFlowGraph & G
Definition: RDFGraph.cpp:200
llvm::object::MachOObjectFile::is64Bit
bool is64Bit() const
Definition: MachOObjectFile.cpp:5254
llvm::MachO::N_PEXT
@ N_PEXT
Definition: MachO.h:308
llvm::MachO::S_ATTR_PURE_INSTRUCTIONS
@ S_ATTR_PURE_INSTRUCTIONS
S_ATTR_PURE_INSTRUCTIONS - Section contains only true machine instructions.
Definition: MachO.h:192
llvm::object::MachOObjectFile::getSymbol64TableEntry
MachO::nlist_64 getSymbol64TableEntry(DataRefImpl DRI) const
Definition: MachOObjectFile.cpp:4625
llvm::orc::MemProt
MemProt
Describes Read/Write/Exec permissions for memory.
Definition: MemoryFlags.h:27
uint64_t
llvm::MachO::N_INDR
@ N_INDR
Definition: MachO.h:320
move
compiles ldr LCPI1_0 ldr ldr mov lsr tst moveq r1 ldr LCPI1_1 and r0 bx lr It would be better to do something like to fold the shift into the conditional move
Definition: README.txt:546
llvm::MachO::MH_SUBSECTIONS_VIA_SYMBOLS
@ MH_SUBSECTIONS_VIA_SYMBOLS
Definition: MachO.h:73
I
#define I(x, y, z)
Definition: MD5.cpp:58
assert
assert(ImpDefSCC.getReg()==AMDGPU::SCC &&ImpDefSCC.isDef())
llvm::move
OutputIt move(R &&Range, OutputIt Out)
Provide wrappers to std::move which take ranges instead of having to pass begin/end explicitly.
Definition: STLExtras.h:1666
memcpy
<%struct.s * > cast struct s *S to sbyte *< sbyte * > sbyte uint cast struct s *agg result to sbyte *< sbyte * > sbyte uint cast struct s *memtmp to sbyte *< sbyte * > sbyte uint ret void llc ends up issuing two memcpy or custom lower memcpy(of small size) to be ldmia/stmia. I think option 2 is better but the current register allocator cannot allocate a chunk of registers at a time. A feasible temporary solution is to use specific physical registers at the lowering time for small(<
llvm::MachO::N_ALT_ENTRY
@ N_ALT_ENTRY
Definition: MachO.h:348
CommonSectionName
static const char * CommonSectionName
Definition: MachOLinkGraphBuilder.cpp:17
llvm::min
Expected< ExpressionValue > min(const ExpressionValue &Lhs, const ExpressionValue &Rhs)
Definition: FileCheck.cpp:357
llvm::StringRef
StringRef - Represent a constant reference to a string, i.e.
Definition: StringRef.h:50
llvm::MachO::N_SECT
@ N_SECT
Definition: MachO.h:318
if
if(llvm_vc STREQUAL "") set(fake_version_inc "$
Definition: CMakeLists.txt:14
uint32_t
S
add sub stmia L5 ldr r0 bl L_printf $stub Instead of a and a wouldn t it be better to do three moves *Return an aggregate type is even return S
Definition: README.txt:210
llvm::MachO::N_ABS
@ N_ABS
Definition: MachO.h:317
llvm::StringRef::size
constexpr size_t size() const
size - Get the string size.
Definition: StringRef.h:137
llvm::object::MachOObjectFile::getSymbolTableEntry
MachO::nlist getSymbolTableEntry(DataRefImpl DRI) const
Definition: MachOObjectFile.cpp:4619
llvm::GraphProgram::Name
Name
Definition: GraphWriter.h:50
std
Definition: BitVector.h:851
uint16_t
llvm::SectionName
Definition: DWARFSection.h:21
llvm::Error
Lightweight error class with error context and mandatory checking.
Definition: Error.h:155
llvm::object::MachOObjectFile::getHeader64
const MachO::mach_header_64 & getHeader64() const
Definition: MachOObjectFile.cpp:4787
llvm::TargetStackID::Value
Value
Definition: TargetFrameLowering.h:27
llvm::StringMap::count
size_type count(StringRef Key) const
count - Return 1 if the element is in the map, 0 otherwise.
Definition: StringMap.h:245
llvm::MachO::S_THREAD_LOCAL_ZEROFILL
@ S_THREAD_LOCAL_ZEROFILL
S_THREAD_LOCAL_ZEROFILL - Thread local zerofill section.
Definition: MachO.h:169
llvm::ARMBuildAttrs::Symbol
@ Symbol
Definition: ARMBuildAttributes.h:83
llvm::object::ObjectFile::sections
section_iterator_range sections() const
Definition: ObjectFile.h:327
llvm::support::endianness
endianness
Definition: Endian.h:27
llvm::object::Binary::getData
StringRef getData() const
Definition: Binary.cpp:39
llvm::msgpack::Type
Type
MessagePack types as defined in the standard, with the exception of Integer being divided into a sign...
Definition: MsgPackReader.h:48
llvm::orc::MemProt::Exec
@ Exec
MachOLinkGraphBuilder.h
llvm::pdb::PDB_SymType::Block
@ Block
llvm::Triple::aarch64
@ aarch64
Definition: Triple.h:51
llvm::MachO::N_NO_DEAD_STRIP
@ N_NO_DEAD_STRIP
Definition: MachO.h:344
llvm::support::big
@ big
Definition: Endian.h:27
SpecialSubKind::string
@ string
llvm::object::MachOObjectFile::isRelocatableObject
bool isRelocatableObject() const override
True if this is a relocatable object (.o/.obj).
Definition: MachOObjectFile.cpp:5271
llvm::orc::MemProt::Write
@ Write