xref: /freebsd/contrib/llvm-project/llvm/lib/DWARFLinker/Classic/DWARFLinker.cpp (revision 0fca6ea1d4eea4c934cfff25ac9ee8ad6fe95583)
1 //=== DWARFLinker.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/DWARFLinker/Classic/DWARFLinker.h"
10 #include "llvm/ADT/ArrayRef.h"
11 #include "llvm/ADT/BitVector.h"
12 #include "llvm/ADT/STLExtras.h"
13 #include "llvm/ADT/StringExtras.h"
14 #include "llvm/CodeGen/NonRelocatableStringpool.h"
15 #include "llvm/DWARFLinker/Classic/DWARFLinkerDeclContext.h"
16 #include "llvm/DWARFLinker/Classic/DWARFStreamer.h"
17 #include "llvm/DWARFLinker/Utils.h"
18 #include "llvm/DebugInfo/DWARF/DWARFAbbreviationDeclaration.h"
19 #include "llvm/DebugInfo/DWARF/DWARFAcceleratorTable.h"
20 #include "llvm/DebugInfo/DWARF/DWARFContext.h"
21 #include "llvm/DebugInfo/DWARF/DWARFDataExtractor.h"
22 #include "llvm/DebugInfo/DWARF/DWARFDebugLine.h"
23 #include "llvm/DebugInfo/DWARF/DWARFDebugMacro.h"
24 #include "llvm/DebugInfo/DWARF/DWARFDebugRangeList.h"
25 #include "llvm/DebugInfo/DWARF/DWARFDie.h"
26 #include "llvm/DebugInfo/DWARF/DWARFExpression.h"
27 #include "llvm/DebugInfo/DWARF/DWARFFormValue.h"
28 #include "llvm/DebugInfo/DWARF/DWARFSection.h"
29 #include "llvm/DebugInfo/DWARF/DWARFUnit.h"
30 #include "llvm/MC/MCDwarf.h"
31 #include "llvm/Support/DataExtractor.h"
32 #include "llvm/Support/Error.h"
33 #include "llvm/Support/ErrorHandling.h"
34 #include "llvm/Support/ErrorOr.h"
35 #include "llvm/Support/FormatVariadic.h"
36 #include "llvm/Support/LEB128.h"
37 #include "llvm/Support/Path.h"
38 #include "llvm/Support/ThreadPool.h"
39 #include <vector>
40 
41 namespace llvm {
42 
43 using namespace dwarf_linker;
44 using namespace dwarf_linker::classic;
45 
46 /// Hold the input and output of the debug info size in bytes.
47 struct DebugInfoSize {
48   uint64_t Input;
49   uint64_t Output;
50 };
51 
52 /// Compute the total size of the debug info.
getDebugInfoSize(DWARFContext & Dwarf)53 static uint64_t getDebugInfoSize(DWARFContext &Dwarf) {
54   uint64_t Size = 0;
55   for (auto &Unit : Dwarf.compile_units()) {
56     Size += Unit->getLength();
57   }
58   return Size;
59 }
60 
61 /// Similar to DWARFUnitSection::getUnitForOffset(), but returning our
62 /// CompileUnit object instead.
getUnitForOffset(const UnitListTy & Units,uint64_t Offset)63 static CompileUnit *getUnitForOffset(const UnitListTy &Units, uint64_t Offset) {
64   auto CU = llvm::upper_bound(
65       Units, Offset, [](uint64_t LHS, const std::unique_ptr<CompileUnit> &RHS) {
66         return LHS < RHS->getOrigUnit().getNextUnitOffset();
67       });
68   return CU != Units.end() ? CU->get() : nullptr;
69 }
70 
71 /// Resolve the DIE attribute reference that has been extracted in \p RefValue.
72 /// The resulting DIE might be in another CompileUnit which is stored into \p
73 /// ReferencedCU. \returns null if resolving fails for any reason.
resolveDIEReference(const DWARFFile & File,const UnitListTy & Units,const DWARFFormValue & RefValue,const DWARFDie & DIE,CompileUnit * & RefCU)74 DWARFDie DWARFLinker::resolveDIEReference(const DWARFFile &File,
75                                           const UnitListTy &Units,
76                                           const DWARFFormValue &RefValue,
77                                           const DWARFDie &DIE,
78                                           CompileUnit *&RefCU) {
79   assert(RefValue.isFormClass(DWARFFormValue::FC_Reference));
80   uint64_t RefOffset;
81   if (std::optional<uint64_t> Off = RefValue.getAsRelativeReference()) {
82     RefOffset = RefValue.getUnit()->getOffset() + *Off;
83   } else if (Off = RefValue.getAsDebugInfoReference(); Off) {
84     RefOffset = *Off;
85   } else {
86     reportWarning("Unsupported reference type", File, &DIE);
87     return DWARFDie();
88   }
89   if ((RefCU = getUnitForOffset(Units, RefOffset)))
90     if (const auto RefDie = RefCU->getOrigUnit().getDIEForOffset(RefOffset)) {
91       // In a file with broken references, an attribute might point to a NULL
92       // DIE.
93       if (!RefDie.isNULL())
94         return RefDie;
95     }
96 
97   reportWarning("could not find referenced DIE", File, &DIE);
98   return DWARFDie();
99 }
100 
101 /// \returns whether the passed \a Attr type might contain a DIE reference
102 /// suitable for ODR uniquing.
isODRAttribute(uint16_t Attr)103 static bool isODRAttribute(uint16_t Attr) {
104   switch (Attr) {
105   default:
106     return false;
107   case dwarf::DW_AT_type:
108   case dwarf::DW_AT_containing_type:
109   case dwarf::DW_AT_specification:
110   case dwarf::DW_AT_abstract_origin:
111   case dwarf::DW_AT_import:
112     return true;
113   }
114   llvm_unreachable("Improper attribute.");
115 }
116 
isTypeTag(uint16_t Tag)117 static bool isTypeTag(uint16_t Tag) {
118   switch (Tag) {
119   case dwarf::DW_TAG_array_type:
120   case dwarf::DW_TAG_class_type:
121   case dwarf::DW_TAG_enumeration_type:
122   case dwarf::DW_TAG_pointer_type:
123   case dwarf::DW_TAG_reference_type:
124   case dwarf::DW_TAG_string_type:
125   case dwarf::DW_TAG_structure_type:
126   case dwarf::DW_TAG_subroutine_type:
127   case dwarf::DW_TAG_template_alias:
128   case dwarf::DW_TAG_typedef:
129   case dwarf::DW_TAG_union_type:
130   case dwarf::DW_TAG_ptr_to_member_type:
131   case dwarf::DW_TAG_set_type:
132   case dwarf::DW_TAG_subrange_type:
133   case dwarf::DW_TAG_base_type:
134   case dwarf::DW_TAG_const_type:
135   case dwarf::DW_TAG_constant:
136   case dwarf::DW_TAG_file_type:
137   case dwarf::DW_TAG_namelist:
138   case dwarf::DW_TAG_packed_type:
139   case dwarf::DW_TAG_volatile_type:
140   case dwarf::DW_TAG_restrict_type:
141   case dwarf::DW_TAG_atomic_type:
142   case dwarf::DW_TAG_interface_type:
143   case dwarf::DW_TAG_unspecified_type:
144   case dwarf::DW_TAG_shared_type:
145   case dwarf::DW_TAG_immutable_type:
146     return true;
147   default:
148     break;
149   }
150   return false;
151 }
152 
getDIENames(const DWARFDie & Die,AttributesInfo & Info,OffsetsStringPool & StringPool,bool StripTemplate)153 bool DWARFLinker::DIECloner::getDIENames(const DWARFDie &Die,
154                                          AttributesInfo &Info,
155                                          OffsetsStringPool &StringPool,
156                                          bool StripTemplate) {
157   // This function will be called on DIEs having low_pcs and
158   // ranges. As getting the name might be more expansive, filter out
159   // blocks directly.
160   if (Die.getTag() == dwarf::DW_TAG_lexical_block)
161     return false;
162 
163   if (!Info.MangledName)
164     if (const char *MangledName = Die.getLinkageName())
165       Info.MangledName = StringPool.getEntry(MangledName);
166 
167   if (!Info.Name)
168     if (const char *Name = Die.getShortName())
169       Info.Name = StringPool.getEntry(Name);
170 
171   if (!Info.MangledName)
172     Info.MangledName = Info.Name;
173 
174   if (StripTemplate && Info.Name && Info.MangledName != Info.Name) {
175     StringRef Name = Info.Name.getString();
176     if (std::optional<StringRef> StrippedName = StripTemplateParameters(Name))
177       Info.NameWithoutTemplate = StringPool.getEntry(*StrippedName);
178   }
179 
180   return Info.Name || Info.MangledName;
181 }
182 
183 /// Resolve the relative path to a build artifact referenced by DWARF by
184 /// applying DW_AT_comp_dir.
resolveRelativeObjectPath(SmallVectorImpl<char> & Buf,DWARFDie CU)185 static void resolveRelativeObjectPath(SmallVectorImpl<char> &Buf, DWARFDie CU) {
186   sys::path::append(Buf, dwarf::toString(CU.find(dwarf::DW_AT_comp_dir), ""));
187 }
188 
189 /// Collect references to parseable Swift interfaces in imported
190 /// DW_TAG_module blocks.
analyzeImportedModule(const DWARFDie & DIE,CompileUnit & CU,DWARFLinkerBase::SwiftInterfacesMapTy * ParseableSwiftInterfaces,std::function<void (const Twine &,const DWARFDie &)> ReportWarning)191 static void analyzeImportedModule(
192     const DWARFDie &DIE, CompileUnit &CU,
193     DWARFLinkerBase::SwiftInterfacesMapTy *ParseableSwiftInterfaces,
194     std::function<void(const Twine &, const DWARFDie &)> ReportWarning) {
195   if (CU.getLanguage() != dwarf::DW_LANG_Swift)
196     return;
197 
198   if (!ParseableSwiftInterfaces)
199     return;
200 
201   StringRef Path = dwarf::toStringRef(DIE.find(dwarf::DW_AT_LLVM_include_path));
202   if (!Path.ends_with(".swiftinterface"))
203     return;
204   // Don't track interfaces that are part of the SDK.
205   StringRef SysRoot = dwarf::toStringRef(DIE.find(dwarf::DW_AT_LLVM_sysroot));
206   if (SysRoot.empty())
207     SysRoot = CU.getSysRoot();
208   if (!SysRoot.empty() && Path.starts_with(SysRoot))
209     return;
210   // Don't track interfaces that are part of the toolchain.
211   // For example: Swift, _Concurrency, ...
212   StringRef DeveloperDir = guessDeveloperDir(SysRoot);
213   if (!DeveloperDir.empty() && Path.starts_with(DeveloperDir))
214     return;
215   if (isInToolchainDir(Path))
216     return;
217   std::optional<const char *> Name =
218       dwarf::toString(DIE.find(dwarf::DW_AT_name));
219   if (!Name)
220     return;
221   auto &Entry = (*ParseableSwiftInterfaces)[*Name];
222   // The prepend path is applied later when copying.
223   DWARFDie CUDie = CU.getOrigUnit().getUnitDIE();
224   SmallString<128> ResolvedPath;
225   if (sys::path::is_relative(Path))
226     resolveRelativeObjectPath(ResolvedPath, CUDie);
227   sys::path::append(ResolvedPath, Path);
228   if (!Entry.empty() && Entry != ResolvedPath)
229     ReportWarning(Twine("Conflicting parseable interfaces for Swift Module ") +
230                       *Name + ": " + Entry + " and " + Path,
231                   DIE);
232   Entry = std::string(ResolvedPath);
233 }
234 
235 /// The distinct types of work performed by the work loop in
236 /// analyzeContextInfo.
237 enum class ContextWorklistItemType : uint8_t {
238   AnalyzeContextInfo,
239   UpdateChildPruning,
240   UpdatePruning,
241 };
242 
243 /// This class represents an item in the work list. The type defines what kind
244 /// of work needs to be performed when processing the current item. Everything
245 /// but the Type and Die fields are optional based on the type.
246 struct ContextWorklistItem {
247   DWARFDie Die;
248   unsigned ParentIdx;
249   union {
250     CompileUnit::DIEInfo *OtherInfo;
251     DeclContext *Context;
252   };
253   ContextWorklistItemType Type;
254   bool InImportedModule;
255 
ContextWorklistItemllvm::ContextWorklistItem256   ContextWorklistItem(DWARFDie Die, ContextWorklistItemType T,
257                       CompileUnit::DIEInfo *OtherInfo = nullptr)
258       : Die(Die), ParentIdx(0), OtherInfo(OtherInfo), Type(T),
259         InImportedModule(false) {}
260 
ContextWorklistItemllvm::ContextWorklistItem261   ContextWorklistItem(DWARFDie Die, DeclContext *Context, unsigned ParentIdx,
262                       bool InImportedModule)
263       : Die(Die), ParentIdx(ParentIdx), Context(Context),
264         Type(ContextWorklistItemType::AnalyzeContextInfo),
265         InImportedModule(InImportedModule) {}
266 };
267 
updatePruning(const DWARFDie & Die,CompileUnit & CU,uint64_t ModulesEndOffset)268 static bool updatePruning(const DWARFDie &Die, CompileUnit &CU,
269                           uint64_t ModulesEndOffset) {
270   CompileUnit::DIEInfo &Info = CU.getInfo(Die);
271 
272   // Prune this DIE if it is either a forward declaration inside a
273   // DW_TAG_module or a DW_TAG_module that contains nothing but
274   // forward declarations.
275   Info.Prune &= (Die.getTag() == dwarf::DW_TAG_module) ||
276                 (isTypeTag(Die.getTag()) &&
277                  dwarf::toUnsigned(Die.find(dwarf::DW_AT_declaration), 0));
278 
279   // Only prune forward declarations inside a DW_TAG_module for which a
280   // definition exists elsewhere.
281   if (ModulesEndOffset == 0)
282     Info.Prune &= Info.Ctxt && Info.Ctxt->getCanonicalDIEOffset();
283   else
284     Info.Prune &= Info.Ctxt && Info.Ctxt->getCanonicalDIEOffset() > 0 &&
285                   Info.Ctxt->getCanonicalDIEOffset() <= ModulesEndOffset;
286 
287   return Info.Prune;
288 }
289 
updateChildPruning(const DWARFDie & Die,CompileUnit & CU,CompileUnit::DIEInfo & ChildInfo)290 static void updateChildPruning(const DWARFDie &Die, CompileUnit &CU,
291                                CompileUnit::DIEInfo &ChildInfo) {
292   CompileUnit::DIEInfo &Info = CU.getInfo(Die);
293   Info.Prune &= ChildInfo.Prune;
294 }
295 
296 /// Recursive helper to build the global DeclContext information and
297 /// gather the child->parent relationships in the original compile unit.
298 ///
299 /// This function uses the same work list approach as lookForDIEsToKeep.
300 ///
301 /// \return true when this DIE and all of its children are only
302 /// forward declarations to types defined in external clang modules
303 /// (i.e., forward declarations that are children of a DW_TAG_module).
analyzeContextInfo(const DWARFDie & DIE,unsigned ParentIdx,CompileUnit & CU,DeclContext * CurrentDeclContext,DeclContextTree & Contexts,uint64_t ModulesEndOffset,DWARFLinkerBase::SwiftInterfacesMapTy * ParseableSwiftInterfaces,std::function<void (const Twine &,const DWARFDie &)> ReportWarning)304 static void analyzeContextInfo(
305     const DWARFDie &DIE, unsigned ParentIdx, CompileUnit &CU,
306     DeclContext *CurrentDeclContext, DeclContextTree &Contexts,
307     uint64_t ModulesEndOffset,
308     DWARFLinkerBase::SwiftInterfacesMapTy *ParseableSwiftInterfaces,
309     std::function<void(const Twine &, const DWARFDie &)> ReportWarning) {
310   // LIFO work list.
311   std::vector<ContextWorklistItem> Worklist;
312   Worklist.emplace_back(DIE, CurrentDeclContext, ParentIdx, false);
313 
314   while (!Worklist.empty()) {
315     ContextWorklistItem Current = Worklist.back();
316     Worklist.pop_back();
317 
318     switch (Current.Type) {
319     case ContextWorklistItemType::UpdatePruning:
320       updatePruning(Current.Die, CU, ModulesEndOffset);
321       continue;
322     case ContextWorklistItemType::UpdateChildPruning:
323       updateChildPruning(Current.Die, CU, *Current.OtherInfo);
324       continue;
325     case ContextWorklistItemType::AnalyzeContextInfo:
326       break;
327     }
328 
329     unsigned Idx = CU.getOrigUnit().getDIEIndex(Current.Die);
330     CompileUnit::DIEInfo &Info = CU.getInfo(Idx);
331 
332     // Clang imposes an ODR on modules(!) regardless of the language:
333     //  "The module-id should consist of only a single identifier,
334     //   which provides the name of the module being defined. Each
335     //   module shall have a single definition."
336     //
337     // This does not extend to the types inside the modules:
338     //  "[I]n C, this implies that if two structs are defined in
339     //   different submodules with the same name, those two types are
340     //   distinct types (but may be compatible types if their
341     //   definitions match)."
342     //
343     // We treat non-C++ modules like namespaces for this reason.
344     if (Current.Die.getTag() == dwarf::DW_TAG_module &&
345         Current.ParentIdx == 0 &&
346         dwarf::toString(Current.Die.find(dwarf::DW_AT_name), "") !=
347             CU.getClangModuleName()) {
348       Current.InImportedModule = true;
349       analyzeImportedModule(Current.Die, CU, ParseableSwiftInterfaces,
350                             ReportWarning);
351     }
352 
353     Info.ParentIdx = Current.ParentIdx;
354     Info.InModuleScope = CU.isClangModule() || Current.InImportedModule;
355     if (CU.hasODR() || Info.InModuleScope) {
356       if (Current.Context) {
357         auto PtrInvalidPair = Contexts.getChildDeclContext(
358             *Current.Context, Current.Die, CU, Info.InModuleScope);
359         Current.Context = PtrInvalidPair.getPointer();
360         Info.Ctxt =
361             PtrInvalidPair.getInt() ? nullptr : PtrInvalidPair.getPointer();
362         if (Info.Ctxt)
363           Info.Ctxt->setDefinedInClangModule(Info.InModuleScope);
364       } else
365         Info.Ctxt = Current.Context = nullptr;
366     }
367 
368     Info.Prune = Current.InImportedModule;
369     // Add children in reverse order to the worklist to effectively process
370     // them in order.
371     Worklist.emplace_back(Current.Die, ContextWorklistItemType::UpdatePruning);
372     for (auto Child : reverse(Current.Die.children())) {
373       CompileUnit::DIEInfo &ChildInfo = CU.getInfo(Child);
374       Worklist.emplace_back(
375           Current.Die, ContextWorklistItemType::UpdateChildPruning, &ChildInfo);
376       Worklist.emplace_back(Child, Current.Context, Idx,
377                             Current.InImportedModule);
378     }
379   }
380 }
381 
dieNeedsChildrenToBeMeaningful(uint32_t Tag)382 static bool dieNeedsChildrenToBeMeaningful(uint32_t Tag) {
383   switch (Tag) {
384   default:
385     return false;
386   case dwarf::DW_TAG_class_type:
387   case dwarf::DW_TAG_common_block:
388   case dwarf::DW_TAG_lexical_block:
389   case dwarf::DW_TAG_structure_type:
390   case dwarf::DW_TAG_subprogram:
391   case dwarf::DW_TAG_subroutine_type:
392   case dwarf::DW_TAG_union_type:
393     return true;
394   }
395   llvm_unreachable("Invalid Tag");
396 }
397 
cleanupAuxiliarryData(LinkContext & Context)398 void DWARFLinker::cleanupAuxiliarryData(LinkContext &Context) {
399   Context.clear();
400 
401   for (DIEBlock *I : DIEBlocks)
402     I->~DIEBlock();
403   for (DIELoc *I : DIELocs)
404     I->~DIELoc();
405 
406   DIEBlocks.clear();
407   DIELocs.clear();
408   DIEAlloc.Reset();
409 }
410 
isTlsAddressCode(uint8_t DW_OP_Code)411 static bool isTlsAddressCode(uint8_t DW_OP_Code) {
412   return DW_OP_Code == dwarf::DW_OP_form_tls_address ||
413          DW_OP_Code == dwarf::DW_OP_GNU_push_tls_address;
414 }
415 
416 std::pair<bool, std::optional<int64_t>>
getVariableRelocAdjustment(AddressesMap & RelocMgr,const DWARFDie & DIE)417 DWARFLinker::getVariableRelocAdjustment(AddressesMap &RelocMgr,
418                                         const DWARFDie &DIE) {
419   assert((DIE.getTag() == dwarf::DW_TAG_variable ||
420           DIE.getTag() == dwarf::DW_TAG_constant) &&
421          "Wrong type of input die");
422 
423   const auto *Abbrev = DIE.getAbbreviationDeclarationPtr();
424 
425   // Check if DIE has DW_AT_location attribute.
426   DWARFUnit *U = DIE.getDwarfUnit();
427   std::optional<uint32_t> LocationIdx =
428       Abbrev->findAttributeIndex(dwarf::DW_AT_location);
429   if (!LocationIdx)
430     return std::make_pair(false, std::nullopt);
431 
432   // Get offset to the DW_AT_location attribute.
433   uint64_t AttrOffset =
434       Abbrev->getAttributeOffsetFromIndex(*LocationIdx, DIE.getOffset(), *U);
435 
436   // Get value of the DW_AT_location attribute.
437   std::optional<DWARFFormValue> LocationValue =
438       Abbrev->getAttributeValueFromOffset(*LocationIdx, AttrOffset, *U);
439   if (!LocationValue)
440     return std::make_pair(false, std::nullopt);
441 
442   // Check that DW_AT_location attribute is of 'exprloc' class.
443   // Handling value of location expressions for attributes of 'loclist'
444   // class is not implemented yet.
445   std::optional<ArrayRef<uint8_t>> Expr = LocationValue->getAsBlock();
446   if (!Expr)
447     return std::make_pair(false, std::nullopt);
448 
449   // Parse 'exprloc' expression.
450   DataExtractor Data(toStringRef(*Expr), U->getContext().isLittleEndian(),
451                      U->getAddressByteSize());
452   DWARFExpression Expression(Data, U->getAddressByteSize(),
453                              U->getFormParams().Format);
454 
455   bool HasLocationAddress = false;
456   uint64_t CurExprOffset = 0;
457   for (DWARFExpression::iterator It = Expression.begin();
458        It != Expression.end(); ++It) {
459     DWARFExpression::iterator NextIt = It;
460     ++NextIt;
461 
462     const DWARFExpression::Operation &Op = *It;
463     switch (Op.getCode()) {
464     case dwarf::DW_OP_const2u:
465     case dwarf::DW_OP_const4u:
466     case dwarf::DW_OP_const8u:
467     case dwarf::DW_OP_const2s:
468     case dwarf::DW_OP_const4s:
469     case dwarf::DW_OP_const8s:
470       if (NextIt == Expression.end() || !isTlsAddressCode(NextIt->getCode()))
471         break;
472       [[fallthrough]];
473     case dwarf::DW_OP_addr: {
474       HasLocationAddress = true;
475       // Check relocation for the address.
476       if (std::optional<int64_t> RelocAdjustment =
477               RelocMgr.getExprOpAddressRelocAdjustment(
478                   *U, Op, AttrOffset + CurExprOffset,
479                   AttrOffset + Op.getEndOffset(), Options.Verbose))
480         return std::make_pair(HasLocationAddress, *RelocAdjustment);
481     } break;
482     case dwarf::DW_OP_constx:
483     case dwarf::DW_OP_addrx: {
484       HasLocationAddress = true;
485       if (std::optional<uint64_t> AddressOffset =
486               DIE.getDwarfUnit()->getIndexedAddressOffset(
487                   Op.getRawOperand(0))) {
488         // Check relocation for the address.
489         if (std::optional<int64_t> RelocAdjustment =
490                 RelocMgr.getExprOpAddressRelocAdjustment(
491                     *U, Op, *AddressOffset,
492                     *AddressOffset + DIE.getDwarfUnit()->getAddressByteSize(),
493                     Options.Verbose))
494           return std::make_pair(HasLocationAddress, *RelocAdjustment);
495       }
496     } break;
497     default: {
498       // Nothing to do.
499     } break;
500     }
501     CurExprOffset = Op.getEndOffset();
502   }
503 
504   return std::make_pair(HasLocationAddress, std::nullopt);
505 }
506 
507 /// Check if a variable describing DIE should be kept.
508 /// \returns updated TraversalFlags.
shouldKeepVariableDIE(AddressesMap & RelocMgr,const DWARFDie & DIE,CompileUnit::DIEInfo & MyInfo,unsigned Flags)509 unsigned DWARFLinker::shouldKeepVariableDIE(AddressesMap &RelocMgr,
510                                             const DWARFDie &DIE,
511                                             CompileUnit::DIEInfo &MyInfo,
512                                             unsigned Flags) {
513   const auto *Abbrev = DIE.getAbbreviationDeclarationPtr();
514 
515   // Global variables with constant value can always be kept.
516   if (!(Flags & TF_InFunctionScope) &&
517       Abbrev->findAttributeIndex(dwarf::DW_AT_const_value)) {
518     MyInfo.InDebugMap = true;
519     return Flags | TF_Keep;
520   }
521 
522   // See if there is a relocation to a valid debug map entry inside this
523   // variable's location. The order is important here. We want to always check
524   // if the variable has a valid relocation, so that the DIEInfo is filled.
525   // However, we don't want a static variable in a function to force us to keep
526   // the enclosing function, unless requested explicitly.
527   std::pair<bool, std::optional<int64_t>> LocExprAddrAndRelocAdjustment =
528       getVariableRelocAdjustment(RelocMgr, DIE);
529 
530   if (LocExprAddrAndRelocAdjustment.first)
531     MyInfo.HasLocationExpressionAddr = true;
532 
533   if (!LocExprAddrAndRelocAdjustment.second)
534     return Flags;
535 
536   MyInfo.AddrAdjust = *LocExprAddrAndRelocAdjustment.second;
537   MyInfo.InDebugMap = true;
538 
539   if (((Flags & TF_InFunctionScope) &&
540        !LLVM_UNLIKELY(Options.KeepFunctionForStatic)))
541     return Flags;
542 
543   if (Options.Verbose) {
544     outs() << "Keeping variable DIE:";
545     DIDumpOptions DumpOpts;
546     DumpOpts.ChildRecurseDepth = 0;
547     DumpOpts.Verbose = Options.Verbose;
548     DIE.dump(outs(), 8 /* Indent */, DumpOpts);
549   }
550 
551   return Flags | TF_Keep;
552 }
553 
554 /// Check if a function describing DIE should be kept.
555 /// \returns updated TraversalFlags.
shouldKeepSubprogramDIE(AddressesMap & RelocMgr,const DWARFDie & DIE,const DWARFFile & File,CompileUnit & Unit,CompileUnit::DIEInfo & MyInfo,unsigned Flags)556 unsigned DWARFLinker::shouldKeepSubprogramDIE(
557     AddressesMap &RelocMgr, const DWARFDie &DIE, const DWARFFile &File,
558     CompileUnit &Unit, CompileUnit::DIEInfo &MyInfo, unsigned Flags) {
559   Flags |= TF_InFunctionScope;
560 
561   auto LowPc = dwarf::toAddress(DIE.find(dwarf::DW_AT_low_pc));
562   if (!LowPc)
563     return Flags;
564 
565   assert(LowPc && "low_pc attribute is not an address.");
566   std::optional<int64_t> RelocAdjustment =
567       RelocMgr.getSubprogramRelocAdjustment(DIE, Options.Verbose);
568   if (!RelocAdjustment)
569     return Flags;
570 
571   MyInfo.AddrAdjust = *RelocAdjustment;
572   MyInfo.InDebugMap = true;
573 
574   if (Options.Verbose) {
575     outs() << "Keeping subprogram DIE:";
576     DIDumpOptions DumpOpts;
577     DumpOpts.ChildRecurseDepth = 0;
578     DumpOpts.Verbose = Options.Verbose;
579     DIE.dump(outs(), 8 /* Indent */, DumpOpts);
580   }
581 
582   if (DIE.getTag() == dwarf::DW_TAG_label) {
583     if (Unit.hasLabelAt(*LowPc))
584       return Flags;
585 
586     DWARFUnit &OrigUnit = Unit.getOrigUnit();
587     // FIXME: dsymutil-classic compat. dsymutil-classic doesn't consider labels
588     // that don't fall into the CU's aranges. This is wrong IMO. Debug info
589     // generation bugs aside, this is really wrong in the case of labels, where
590     // a label marking the end of a function will have a PC == CU's high_pc.
591     if (dwarf::toAddress(OrigUnit.getUnitDIE().find(dwarf::DW_AT_high_pc))
592             .value_or(UINT64_MAX) <= LowPc)
593       return Flags;
594     Unit.addLabelLowPc(*LowPc, MyInfo.AddrAdjust);
595     return Flags | TF_Keep;
596   }
597 
598   Flags |= TF_Keep;
599 
600   std::optional<uint64_t> HighPc = DIE.getHighPC(*LowPc);
601   if (!HighPc) {
602     reportWarning("Function without high_pc. Range will be discarded.\n", File,
603                   &DIE);
604     return Flags;
605   }
606   if (*LowPc > *HighPc) {
607     reportWarning("low_pc greater than high_pc. Range will be discarded.\n",
608                   File, &DIE);
609     return Flags;
610   }
611 
612   // Replace the debug map range with a more accurate one.
613   Unit.addFunctionRange(*LowPc, *HighPc, MyInfo.AddrAdjust);
614   return Flags;
615 }
616 
617 /// Check if a DIE should be kept.
618 /// \returns updated TraversalFlags.
shouldKeepDIE(AddressesMap & RelocMgr,const DWARFDie & DIE,const DWARFFile & File,CompileUnit & Unit,CompileUnit::DIEInfo & MyInfo,unsigned Flags)619 unsigned DWARFLinker::shouldKeepDIE(AddressesMap &RelocMgr, const DWARFDie &DIE,
620                                     const DWARFFile &File, CompileUnit &Unit,
621                                     CompileUnit::DIEInfo &MyInfo,
622                                     unsigned Flags) {
623   switch (DIE.getTag()) {
624   case dwarf::DW_TAG_constant:
625   case dwarf::DW_TAG_variable:
626     return shouldKeepVariableDIE(RelocMgr, DIE, MyInfo, Flags);
627   case dwarf::DW_TAG_subprogram:
628   case dwarf::DW_TAG_label:
629     return shouldKeepSubprogramDIE(RelocMgr, DIE, File, Unit, MyInfo, Flags);
630   case dwarf::DW_TAG_base_type:
631     // DWARF Expressions may reference basic types, but scanning them
632     // is expensive. Basic types are tiny, so just keep all of them.
633   case dwarf::DW_TAG_imported_module:
634   case dwarf::DW_TAG_imported_declaration:
635   case dwarf::DW_TAG_imported_unit:
636     // We always want to keep these.
637     return Flags | TF_Keep;
638   default:
639     break;
640   }
641 
642   return Flags;
643 }
644 
645 /// Helper that updates the completeness of the current DIE based on the
646 /// completeness of one of its children. It depends on the incompleteness of
647 /// the children already being computed.
updateChildIncompleteness(const DWARFDie & Die,CompileUnit & CU,CompileUnit::DIEInfo & ChildInfo)648 static void updateChildIncompleteness(const DWARFDie &Die, CompileUnit &CU,
649                                       CompileUnit::DIEInfo &ChildInfo) {
650   switch (Die.getTag()) {
651   case dwarf::DW_TAG_structure_type:
652   case dwarf::DW_TAG_class_type:
653   case dwarf::DW_TAG_union_type:
654     break;
655   default:
656     return;
657   }
658 
659   CompileUnit::DIEInfo &MyInfo = CU.getInfo(Die);
660 
661   if (ChildInfo.Incomplete || ChildInfo.Prune)
662     MyInfo.Incomplete = true;
663 }
664 
665 /// Helper that updates the completeness of the current DIE based on the
666 /// completeness of the DIEs it references. It depends on the incompleteness of
667 /// the referenced DIE already being computed.
updateRefIncompleteness(const DWARFDie & Die,CompileUnit & CU,CompileUnit::DIEInfo & RefInfo)668 static void updateRefIncompleteness(const DWARFDie &Die, CompileUnit &CU,
669                                     CompileUnit::DIEInfo &RefInfo) {
670   switch (Die.getTag()) {
671   case dwarf::DW_TAG_typedef:
672   case dwarf::DW_TAG_member:
673   case dwarf::DW_TAG_reference_type:
674   case dwarf::DW_TAG_ptr_to_member_type:
675   case dwarf::DW_TAG_pointer_type:
676     break;
677   default:
678     return;
679   }
680 
681   CompileUnit::DIEInfo &MyInfo = CU.getInfo(Die);
682 
683   if (MyInfo.Incomplete)
684     return;
685 
686   if (RefInfo.Incomplete)
687     MyInfo.Incomplete = true;
688 }
689 
690 /// Look at the children of the given DIE and decide whether they should be
691 /// kept.
lookForChildDIEsToKeep(const DWARFDie & Die,CompileUnit & CU,unsigned Flags,SmallVectorImpl<WorklistItem> & Worklist)692 void DWARFLinker::lookForChildDIEsToKeep(
693     const DWARFDie &Die, CompileUnit &CU, unsigned Flags,
694     SmallVectorImpl<WorklistItem> &Worklist) {
695   // The TF_ParentWalk flag tells us that we are currently walking up the
696   // parent chain of a required DIE, and we don't want to mark all the children
697   // of the parents as kept (consider for example a DW_TAG_namespace node in
698   // the parent chain). There are however a set of DIE types for which we want
699   // to ignore that directive and still walk their children.
700   if (dieNeedsChildrenToBeMeaningful(Die.getTag()))
701     Flags &= ~DWARFLinker::TF_ParentWalk;
702 
703   // We're finished if this DIE has no children or we're walking the parent
704   // chain.
705   if (!Die.hasChildren() || (Flags & DWARFLinker::TF_ParentWalk))
706     return;
707 
708   // Add children in reverse order to the worklist to effectively process them
709   // in order.
710   for (auto Child : reverse(Die.children())) {
711     // Add a worklist item before every child to calculate incompleteness right
712     // after the current child is processed.
713     CompileUnit::DIEInfo &ChildInfo = CU.getInfo(Child);
714     Worklist.emplace_back(Die, CU, WorklistItemType::UpdateChildIncompleteness,
715                           &ChildInfo);
716     Worklist.emplace_back(Child, CU, Flags);
717   }
718 }
719 
isODRCanonicalCandidate(const DWARFDie & Die,CompileUnit & CU)720 static bool isODRCanonicalCandidate(const DWARFDie &Die, CompileUnit &CU) {
721   CompileUnit::DIEInfo &Info = CU.getInfo(Die);
722 
723   if (!Info.Ctxt || (Die.getTag() == dwarf::DW_TAG_namespace))
724     return false;
725 
726   if (!CU.hasODR() && !Info.InModuleScope)
727     return false;
728 
729   return !Info.Incomplete && Info.Ctxt != CU.getInfo(Info.ParentIdx).Ctxt;
730 }
731 
markODRCanonicalDie(const DWARFDie & Die,CompileUnit & CU)732 void DWARFLinker::markODRCanonicalDie(const DWARFDie &Die, CompileUnit &CU) {
733   CompileUnit::DIEInfo &Info = CU.getInfo(Die);
734 
735   Info.ODRMarkingDone = true;
736   if (Info.Keep && isODRCanonicalCandidate(Die, CU) &&
737       !Info.Ctxt->hasCanonicalDIE())
738     Info.Ctxt->setHasCanonicalDIE();
739 }
740 
741 /// Look at DIEs referenced by the given DIE and decide whether they should be
742 /// kept. All DIEs referenced though attributes should be kept.
lookForRefDIEsToKeep(const DWARFDie & Die,CompileUnit & CU,unsigned Flags,const UnitListTy & Units,const DWARFFile & File,SmallVectorImpl<WorklistItem> & Worklist)743 void DWARFLinker::lookForRefDIEsToKeep(
744     const DWARFDie &Die, CompileUnit &CU, unsigned Flags,
745     const UnitListTy &Units, const DWARFFile &File,
746     SmallVectorImpl<WorklistItem> &Worklist) {
747   bool UseOdr = (Flags & DWARFLinker::TF_DependencyWalk)
748                     ? (Flags & DWARFLinker::TF_ODR)
749                     : CU.hasODR();
750   DWARFUnit &Unit = CU.getOrigUnit();
751   DWARFDataExtractor Data = Unit.getDebugInfoExtractor();
752   const auto *Abbrev = Die.getAbbreviationDeclarationPtr();
753   uint64_t Offset = Die.getOffset() + getULEB128Size(Abbrev->getCode());
754 
755   SmallVector<std::pair<DWARFDie, CompileUnit &>, 4> ReferencedDIEs;
756   for (const auto &AttrSpec : Abbrev->attributes()) {
757     DWARFFormValue Val(AttrSpec.Form);
758     if (!Val.isFormClass(DWARFFormValue::FC_Reference) ||
759         AttrSpec.Attr == dwarf::DW_AT_sibling) {
760       DWARFFormValue::skipValue(AttrSpec.Form, Data, &Offset,
761                                 Unit.getFormParams());
762       continue;
763     }
764 
765     Val.extractValue(Data, &Offset, Unit.getFormParams(), &Unit);
766     CompileUnit *ReferencedCU;
767     if (auto RefDie =
768             resolveDIEReference(File, Units, Val, Die, ReferencedCU)) {
769       CompileUnit::DIEInfo &Info = ReferencedCU->getInfo(RefDie);
770       // If the referenced DIE has a DeclContext that has already been
771       // emitted, then do not keep the one in this CU. We'll link to
772       // the canonical DIE in cloneDieReferenceAttribute.
773       //
774       // FIXME: compatibility with dsymutil-classic. UseODR shouldn't
775       // be necessary and could be advantageously replaced by
776       // ReferencedCU->hasODR() && CU.hasODR().
777       //
778       // FIXME: compatibility with dsymutil-classic. There is no
779       // reason not to unique ref_addr references.
780       if (AttrSpec.Form != dwarf::DW_FORM_ref_addr &&
781           isODRAttribute(AttrSpec.Attr) && Info.Ctxt &&
782           Info.Ctxt->hasCanonicalDIE())
783         continue;
784 
785       // Keep a module forward declaration if there is no definition.
786       if (!(isODRAttribute(AttrSpec.Attr) && Info.Ctxt &&
787             Info.Ctxt->hasCanonicalDIE()))
788         Info.Prune = false;
789       ReferencedDIEs.emplace_back(RefDie, *ReferencedCU);
790     }
791   }
792 
793   unsigned ODRFlag = UseOdr ? DWARFLinker::TF_ODR : 0;
794 
795   // Add referenced DIEs in reverse order to the worklist to effectively
796   // process them in order.
797   for (auto &P : reverse(ReferencedDIEs)) {
798     // Add a worklist item before every child to calculate incompleteness right
799     // after the current child is processed.
800     CompileUnit::DIEInfo &Info = P.second.getInfo(P.first);
801     Worklist.emplace_back(Die, CU, WorklistItemType::UpdateRefIncompleteness,
802                           &Info);
803     Worklist.emplace_back(P.first, P.second,
804                           DWARFLinker::TF_Keep |
805                               DWARFLinker::TF_DependencyWalk | ODRFlag);
806   }
807 }
808 
809 /// Look at the parent of the given DIE and decide whether they should be kept.
lookForParentDIEsToKeep(unsigned AncestorIdx,CompileUnit & CU,unsigned Flags,SmallVectorImpl<WorklistItem> & Worklist)810 void DWARFLinker::lookForParentDIEsToKeep(
811     unsigned AncestorIdx, CompileUnit &CU, unsigned Flags,
812     SmallVectorImpl<WorklistItem> &Worklist) {
813   // Stop if we encounter an ancestor that's already marked as kept.
814   if (CU.getInfo(AncestorIdx).Keep)
815     return;
816 
817   DWARFUnit &Unit = CU.getOrigUnit();
818   DWARFDie ParentDIE = Unit.getDIEAtIndex(AncestorIdx);
819   Worklist.emplace_back(CU.getInfo(AncestorIdx).ParentIdx, CU, Flags);
820   Worklist.emplace_back(ParentDIE, CU, Flags);
821 }
822 
823 /// Recursively walk the \p DIE tree and look for DIEs to keep. Store that
824 /// information in \p CU's DIEInfo.
825 ///
826 /// This function is the entry point of the DIE selection algorithm. It is
827 /// expected to walk the DIE tree in file order and (though the mediation of
828 /// its helper) call hasValidRelocation() on each DIE that might be a 'root
829 /// DIE' (See DwarfLinker class comment).
830 ///
831 /// While walking the dependencies of root DIEs, this function is also called,
832 /// but during these dependency walks the file order is not respected. The
833 /// TF_DependencyWalk flag tells us which kind of traversal we are currently
834 /// doing.
835 ///
836 /// The recursive algorithm is implemented iteratively as a work list because
837 /// very deep recursion could exhaust the stack for large projects. The work
838 /// list acts as a scheduler for different types of work that need to be
839 /// performed.
840 ///
841 /// The recursive nature of the algorithm is simulated by running the "main"
842 /// algorithm (LookForDIEsToKeep) followed by either looking at more DIEs
843 /// (LookForChildDIEsToKeep, LookForRefDIEsToKeep, LookForParentDIEsToKeep) or
844 /// fixing up a computed property (UpdateChildIncompleteness,
845 /// UpdateRefIncompleteness).
846 ///
847 /// The return value indicates whether the DIE is incomplete.
lookForDIEsToKeep(AddressesMap & AddressesMap,const UnitListTy & Units,const DWARFDie & Die,const DWARFFile & File,CompileUnit & Cu,unsigned Flags)848 void DWARFLinker::lookForDIEsToKeep(AddressesMap &AddressesMap,
849                                     const UnitListTy &Units,
850                                     const DWARFDie &Die, const DWARFFile &File,
851                                     CompileUnit &Cu, unsigned Flags) {
852   // LIFO work list.
853   SmallVector<WorklistItem, 4> Worklist;
854   Worklist.emplace_back(Die, Cu, Flags);
855 
856   while (!Worklist.empty()) {
857     WorklistItem Current = Worklist.pop_back_val();
858 
859     // Look at the worklist type to decide what kind of work to perform.
860     switch (Current.Type) {
861     case WorklistItemType::UpdateChildIncompleteness:
862       updateChildIncompleteness(Current.Die, Current.CU, *Current.OtherInfo);
863       continue;
864     case WorklistItemType::UpdateRefIncompleteness:
865       updateRefIncompleteness(Current.Die, Current.CU, *Current.OtherInfo);
866       continue;
867     case WorklistItemType::LookForChildDIEsToKeep:
868       lookForChildDIEsToKeep(Current.Die, Current.CU, Current.Flags, Worklist);
869       continue;
870     case WorklistItemType::LookForRefDIEsToKeep:
871       lookForRefDIEsToKeep(Current.Die, Current.CU, Current.Flags, Units, File,
872                            Worklist);
873       continue;
874     case WorklistItemType::LookForParentDIEsToKeep:
875       lookForParentDIEsToKeep(Current.AncestorIdx, Current.CU, Current.Flags,
876                               Worklist);
877       continue;
878     case WorklistItemType::MarkODRCanonicalDie:
879       markODRCanonicalDie(Current.Die, Current.CU);
880       continue;
881     case WorklistItemType::LookForDIEsToKeep:
882       break;
883     }
884 
885     unsigned Idx = Current.CU.getOrigUnit().getDIEIndex(Current.Die);
886     CompileUnit::DIEInfo &MyInfo = Current.CU.getInfo(Idx);
887 
888     if (MyInfo.Prune) {
889       // We're walking the dependencies of a module forward declaration that was
890       // kept because there is no definition.
891       if (Current.Flags & TF_DependencyWalk)
892         MyInfo.Prune = false;
893       else
894         continue;
895     }
896 
897     // If the Keep flag is set, we are marking a required DIE's dependencies.
898     // If our target is already marked as kept, we're all set.
899     bool AlreadyKept = MyInfo.Keep;
900     if ((Current.Flags & TF_DependencyWalk) && AlreadyKept)
901       continue;
902 
903     if (!(Current.Flags & TF_DependencyWalk))
904       Current.Flags = shouldKeepDIE(AddressesMap, Current.Die, File, Current.CU,
905                                     MyInfo, Current.Flags);
906 
907     // We need to mark context for the canonical die in the end of normal
908     // traversing(not TF_DependencyWalk) or after normal traversing if die
909     // was not marked as kept.
910     if (!(Current.Flags & TF_DependencyWalk) ||
911         (MyInfo.ODRMarkingDone && !MyInfo.Keep)) {
912       if (Current.CU.hasODR() || MyInfo.InModuleScope)
913         Worklist.emplace_back(Current.Die, Current.CU,
914                               WorklistItemType::MarkODRCanonicalDie);
915     }
916 
917     // Finish by looking for child DIEs. Because of the LIFO worklist we need
918     // to schedule that work before any subsequent items are added to the
919     // worklist.
920     Worklist.emplace_back(Current.Die, Current.CU, Current.Flags,
921                           WorklistItemType::LookForChildDIEsToKeep);
922 
923     if (AlreadyKept || !(Current.Flags & TF_Keep))
924       continue;
925 
926     // If it is a newly kept DIE mark it as well as all its dependencies as
927     // kept.
928     MyInfo.Keep = true;
929 
930     // We're looking for incomplete types.
931     MyInfo.Incomplete =
932         Current.Die.getTag() != dwarf::DW_TAG_subprogram &&
933         Current.Die.getTag() != dwarf::DW_TAG_member &&
934         dwarf::toUnsigned(Current.Die.find(dwarf::DW_AT_declaration), 0);
935 
936     // After looking at the parent chain, look for referenced DIEs. Because of
937     // the LIFO worklist we need to schedule that work before any subsequent
938     // items are added to the worklist.
939     Worklist.emplace_back(Current.Die, Current.CU, Current.Flags,
940                           WorklistItemType::LookForRefDIEsToKeep);
941 
942     bool UseOdr = (Current.Flags & TF_DependencyWalk) ? (Current.Flags & TF_ODR)
943                                                       : Current.CU.hasODR();
944     unsigned ODRFlag = UseOdr ? TF_ODR : 0;
945     unsigned ParFlags = TF_ParentWalk | TF_Keep | TF_DependencyWalk | ODRFlag;
946 
947     // Now schedule the parent walk.
948     Worklist.emplace_back(MyInfo.ParentIdx, Current.CU, ParFlags);
949   }
950 }
951 
952 #ifndef NDEBUG
953 /// A broken link in the keep chain. By recording both the parent and the child
954 /// we can show only broken links for DIEs with multiple children.
955 struct BrokenLink {
BrokenLinkllvm::BrokenLink956   BrokenLink(DWARFDie Parent, DWARFDie Child) : Parent(Parent), Child(Child) {}
957   DWARFDie Parent;
958   DWARFDie Child;
959 };
960 
961 /// Verify the keep chain by looking for DIEs that are kept but who's parent
962 /// isn't.
verifyKeepChain(CompileUnit & CU)963 static void verifyKeepChain(CompileUnit &CU) {
964   std::vector<DWARFDie> Worklist;
965   Worklist.push_back(CU.getOrigUnit().getUnitDIE());
966 
967   // List of broken links.
968   std::vector<BrokenLink> BrokenLinks;
969 
970   while (!Worklist.empty()) {
971     const DWARFDie Current = Worklist.back();
972     Worklist.pop_back();
973 
974     const bool CurrentDieIsKept = CU.getInfo(Current).Keep;
975 
976     for (DWARFDie Child : reverse(Current.children())) {
977       Worklist.push_back(Child);
978 
979       const bool ChildDieIsKept = CU.getInfo(Child).Keep;
980       if (!CurrentDieIsKept && ChildDieIsKept)
981         BrokenLinks.emplace_back(Current, Child);
982     }
983   }
984 
985   if (!BrokenLinks.empty()) {
986     for (BrokenLink Link : BrokenLinks) {
987       WithColor::error() << formatv(
988           "Found invalid link in keep chain between {0:x} and {1:x}\n",
989           Link.Parent.getOffset(), Link.Child.getOffset());
990 
991       errs() << "Parent:";
992       Link.Parent.dump(errs(), 0, {});
993       CU.getInfo(Link.Parent).dump();
994 
995       errs() << "Child:";
996       Link.Child.dump(errs(), 2, {});
997       CU.getInfo(Link.Child).dump();
998     }
999     report_fatal_error("invalid keep chain");
1000   }
1001 }
1002 #endif
1003 
1004 /// Assign an abbreviation number to \p Abbrev.
1005 ///
1006 /// Our DIEs get freed after every DebugMapObject has been processed,
1007 /// thus the FoldingSet we use to unique DIEAbbrevs cannot refer to
1008 /// the instances hold by the DIEs. When we encounter an abbreviation
1009 /// that we don't know, we create a permanent copy of it.
assignAbbrev(DIEAbbrev & Abbrev)1010 void DWARFLinker::assignAbbrev(DIEAbbrev &Abbrev) {
1011   // Check the set for priors.
1012   FoldingSetNodeID ID;
1013   Abbrev.Profile(ID);
1014   void *InsertToken;
1015   DIEAbbrev *InSet = AbbreviationsSet.FindNodeOrInsertPos(ID, InsertToken);
1016 
1017   // If it's newly added.
1018   if (InSet) {
1019     // Assign existing abbreviation number.
1020     Abbrev.setNumber(InSet->getNumber());
1021   } else {
1022     // Add to abbreviation list.
1023     Abbreviations.push_back(
1024         std::make_unique<DIEAbbrev>(Abbrev.getTag(), Abbrev.hasChildren()));
1025     for (const auto &Attr : Abbrev.getData())
1026       Abbreviations.back()->AddAttribute(Attr);
1027     AbbreviationsSet.InsertNode(Abbreviations.back().get(), InsertToken);
1028     // Assign the unique abbreviation number.
1029     Abbrev.setNumber(Abbreviations.size());
1030     Abbreviations.back()->setNumber(Abbreviations.size());
1031   }
1032 }
1033 
cloneStringAttribute(DIE & Die,AttributeSpec AttrSpec,const DWARFFormValue & Val,const DWARFUnit & U,AttributesInfo & Info)1034 unsigned DWARFLinker::DIECloner::cloneStringAttribute(DIE &Die,
1035                                                       AttributeSpec AttrSpec,
1036                                                       const DWARFFormValue &Val,
1037                                                       const DWARFUnit &U,
1038                                                       AttributesInfo &Info) {
1039   std::optional<const char *> String = dwarf::toString(Val);
1040   if (!String)
1041     return 0;
1042   DwarfStringPoolEntryRef StringEntry;
1043   if (AttrSpec.Form == dwarf::DW_FORM_line_strp) {
1044     StringEntry = DebugLineStrPool.getEntry(*String);
1045   } else {
1046     StringEntry = DebugStrPool.getEntry(*String);
1047 
1048     if (AttrSpec.Attr == dwarf::DW_AT_APPLE_origin) {
1049       Info.HasAppleOrigin = true;
1050       if (std::optional<StringRef> FileName =
1051               ObjFile.Addresses->getLibraryInstallName()) {
1052         StringEntry = DebugStrPool.getEntry(*FileName);
1053       }
1054     }
1055 
1056     // Update attributes info.
1057     if (AttrSpec.Attr == dwarf::DW_AT_name)
1058       Info.Name = StringEntry;
1059     else if (AttrSpec.Attr == dwarf::DW_AT_MIPS_linkage_name ||
1060              AttrSpec.Attr == dwarf::DW_AT_linkage_name)
1061       Info.MangledName = StringEntry;
1062     if (U.getVersion() >= 5) {
1063       // Switch everything to DW_FORM_strx strings.
1064       auto StringOffsetIndex =
1065           StringOffsetPool.getValueIndex(StringEntry.getOffset());
1066       return Die
1067           .addValue(DIEAlloc, dwarf::Attribute(AttrSpec.Attr),
1068                     dwarf::DW_FORM_strx, DIEInteger(StringOffsetIndex))
1069           ->sizeOf(U.getFormParams());
1070     }
1071     // Switch everything to out of line strings.
1072     AttrSpec.Form = dwarf::DW_FORM_strp;
1073   }
1074   Die.addValue(DIEAlloc, dwarf::Attribute(AttrSpec.Attr), AttrSpec.Form,
1075                DIEInteger(StringEntry.getOffset()));
1076   return 4;
1077 }
1078 
cloneDieReferenceAttribute(DIE & Die,const DWARFDie & InputDIE,AttributeSpec AttrSpec,unsigned AttrSize,const DWARFFormValue & Val,const DWARFFile & File,CompileUnit & Unit)1079 unsigned DWARFLinker::DIECloner::cloneDieReferenceAttribute(
1080     DIE &Die, const DWARFDie &InputDIE, AttributeSpec AttrSpec,
1081     unsigned AttrSize, const DWARFFormValue &Val, const DWARFFile &File,
1082     CompileUnit &Unit) {
1083   const DWARFUnit &U = Unit.getOrigUnit();
1084   uint64_t Ref;
1085   if (std::optional<uint64_t> Off = Val.getAsRelativeReference())
1086     Ref = Val.getUnit()->getOffset() + *Off;
1087   else if (Off = Val.getAsDebugInfoReference(); Off)
1088     Ref = *Off;
1089   else
1090     return 0;
1091 
1092   DIE *NewRefDie = nullptr;
1093   CompileUnit *RefUnit = nullptr;
1094 
1095   DWARFDie RefDie =
1096       Linker.resolveDIEReference(File, CompileUnits, Val, InputDIE, RefUnit);
1097 
1098   // If the referenced DIE is not found,  drop the attribute.
1099   if (!RefDie || AttrSpec.Attr == dwarf::DW_AT_sibling)
1100     return 0;
1101 
1102   CompileUnit::DIEInfo &RefInfo = RefUnit->getInfo(RefDie);
1103 
1104   // If we already have emitted an equivalent DeclContext, just point
1105   // at it.
1106   if (isODRAttribute(AttrSpec.Attr) && RefInfo.Ctxt &&
1107       RefInfo.Ctxt->getCanonicalDIEOffset()) {
1108     assert(RefInfo.Ctxt->hasCanonicalDIE() &&
1109            "Offset to canonical die is set, but context is not marked");
1110     DIEInteger Attr(RefInfo.Ctxt->getCanonicalDIEOffset());
1111     Die.addValue(DIEAlloc, dwarf::Attribute(AttrSpec.Attr),
1112                  dwarf::DW_FORM_ref_addr, Attr);
1113     return U.getRefAddrByteSize();
1114   }
1115 
1116   if (!RefInfo.Clone) {
1117     // We haven't cloned this DIE yet. Just create an empty one and
1118     // store it. It'll get really cloned when we process it.
1119     RefInfo.UnclonedReference = true;
1120     RefInfo.Clone = DIE::get(DIEAlloc, dwarf::Tag(RefDie.getTag()));
1121   }
1122   NewRefDie = RefInfo.Clone;
1123 
1124   if (AttrSpec.Form == dwarf::DW_FORM_ref_addr ||
1125       (Unit.hasODR() && isODRAttribute(AttrSpec.Attr))) {
1126     // We cannot currently rely on a DIEEntry to emit ref_addr
1127     // references, because the implementation calls back to DwarfDebug
1128     // to find the unit offset. (We don't have a DwarfDebug)
1129     // FIXME: we should be able to design DIEEntry reliance on
1130     // DwarfDebug away.
1131     uint64_t Attr;
1132     if (Ref < InputDIE.getOffset() && !RefInfo.UnclonedReference) {
1133       // We have already cloned that DIE.
1134       uint32_t NewRefOffset =
1135           RefUnit->getStartOffset() + NewRefDie->getOffset();
1136       Attr = NewRefOffset;
1137       Die.addValue(DIEAlloc, dwarf::Attribute(AttrSpec.Attr),
1138                    dwarf::DW_FORM_ref_addr, DIEInteger(Attr));
1139     } else {
1140       // A forward reference. Note and fixup later.
1141       Attr = 0xBADDEF;
1142       Unit.noteForwardReference(
1143           NewRefDie, RefUnit, RefInfo.Ctxt,
1144           Die.addValue(DIEAlloc, dwarf::Attribute(AttrSpec.Attr),
1145                        dwarf::DW_FORM_ref_addr, DIEInteger(Attr)));
1146     }
1147     return U.getRefAddrByteSize();
1148   }
1149 
1150   Die.addValue(DIEAlloc, dwarf::Attribute(AttrSpec.Attr),
1151                dwarf::Form(AttrSpec.Form), DIEEntry(*NewRefDie));
1152 
1153   return AttrSize;
1154 }
1155 
cloneExpression(DataExtractor & Data,DWARFExpression Expression,const DWARFFile & File,CompileUnit & Unit,SmallVectorImpl<uint8_t> & OutputBuffer,int64_t AddrRelocAdjustment,bool IsLittleEndian)1156 void DWARFLinker::DIECloner::cloneExpression(
1157     DataExtractor &Data, DWARFExpression Expression, const DWARFFile &File,
1158     CompileUnit &Unit, SmallVectorImpl<uint8_t> &OutputBuffer,
1159     int64_t AddrRelocAdjustment, bool IsLittleEndian) {
1160   using Encoding = DWARFExpression::Operation::Encoding;
1161 
1162   uint8_t OrigAddressByteSize = Unit.getOrigUnit().getAddressByteSize();
1163 
1164   uint64_t OpOffset = 0;
1165   for (auto &Op : Expression) {
1166     auto Desc = Op.getDescription();
1167     // DW_OP_const_type is variable-length and has 3
1168     // operands. Thus far we only support 2.
1169     if ((Desc.Op.size() == 2 && Desc.Op[0] == Encoding::BaseTypeRef) ||
1170         (Desc.Op.size() == 2 && Desc.Op[1] == Encoding::BaseTypeRef &&
1171          Desc.Op[0] != Encoding::Size1))
1172       Linker.reportWarning("Unsupported DW_OP encoding.", File);
1173 
1174     if ((Desc.Op.size() == 1 && Desc.Op[0] == Encoding::BaseTypeRef) ||
1175         (Desc.Op.size() == 2 && Desc.Op[1] == Encoding::BaseTypeRef &&
1176          Desc.Op[0] == Encoding::Size1)) {
1177       // This code assumes that the other non-typeref operand fits into 1 byte.
1178       assert(OpOffset < Op.getEndOffset());
1179       uint32_t ULEBsize = Op.getEndOffset() - OpOffset - 1;
1180       assert(ULEBsize <= 16);
1181 
1182       // Copy over the operation.
1183       assert(!Op.getSubCode() && "SubOps not yet supported");
1184       OutputBuffer.push_back(Op.getCode());
1185       uint64_t RefOffset;
1186       if (Desc.Op.size() == 1) {
1187         RefOffset = Op.getRawOperand(0);
1188       } else {
1189         OutputBuffer.push_back(Op.getRawOperand(0));
1190         RefOffset = Op.getRawOperand(1);
1191       }
1192       uint32_t Offset = 0;
1193       // Look up the base type. For DW_OP_convert, the operand may be 0 to
1194       // instead indicate the generic type. The same holds for
1195       // DW_OP_reinterpret, which is currently not supported.
1196       if (RefOffset > 0 || Op.getCode() != dwarf::DW_OP_convert) {
1197         RefOffset += Unit.getOrigUnit().getOffset();
1198         auto RefDie = Unit.getOrigUnit().getDIEForOffset(RefOffset);
1199         CompileUnit::DIEInfo &Info = Unit.getInfo(RefDie);
1200         if (DIE *Clone = Info.Clone)
1201           Offset = Clone->getOffset();
1202         else
1203           Linker.reportWarning(
1204               "base type ref doesn't point to DW_TAG_base_type.", File);
1205       }
1206       uint8_t ULEB[16];
1207       unsigned RealSize = encodeULEB128(Offset, ULEB, ULEBsize);
1208       if (RealSize > ULEBsize) {
1209         // Emit the generic type as a fallback.
1210         RealSize = encodeULEB128(0, ULEB, ULEBsize);
1211         Linker.reportWarning("base type ref doesn't fit.", File);
1212       }
1213       assert(RealSize == ULEBsize && "padding failed");
1214       ArrayRef<uint8_t> ULEBbytes(ULEB, ULEBsize);
1215       OutputBuffer.append(ULEBbytes.begin(), ULEBbytes.end());
1216     } else if (!Linker.Options.Update && Op.getCode() == dwarf::DW_OP_addrx) {
1217       if (std::optional<object::SectionedAddress> SA =
1218               Unit.getOrigUnit().getAddrOffsetSectionItem(
1219                   Op.getRawOperand(0))) {
1220         // DWARFLinker does not use addrx forms since it generates relocated
1221         // addresses. Replace DW_OP_addrx with DW_OP_addr here.
1222         // Argument of DW_OP_addrx should be relocated here as it is not
1223         // processed by applyValidRelocs.
1224         OutputBuffer.push_back(dwarf::DW_OP_addr);
1225         uint64_t LinkedAddress = SA->Address + AddrRelocAdjustment;
1226         if (IsLittleEndian != sys::IsLittleEndianHost)
1227           sys::swapByteOrder(LinkedAddress);
1228         ArrayRef<uint8_t> AddressBytes(
1229             reinterpret_cast<const uint8_t *>(&LinkedAddress),
1230             OrigAddressByteSize);
1231         OutputBuffer.append(AddressBytes.begin(), AddressBytes.end());
1232       } else
1233         Linker.reportWarning("cannot read DW_OP_addrx operand.", File);
1234     } else if (!Linker.Options.Update && Op.getCode() == dwarf::DW_OP_constx) {
1235       if (std::optional<object::SectionedAddress> SA =
1236               Unit.getOrigUnit().getAddrOffsetSectionItem(
1237                   Op.getRawOperand(0))) {
1238         // DWARFLinker does not use constx forms since it generates relocated
1239         // addresses. Replace DW_OP_constx with DW_OP_const[*]u here.
1240         // Argument of DW_OP_constx should be relocated here as it is not
1241         // processed by applyValidRelocs.
1242         std::optional<uint8_t> OutOperandKind;
1243         switch (OrigAddressByteSize) {
1244         case 4:
1245           OutOperandKind = dwarf::DW_OP_const4u;
1246           break;
1247         case 8:
1248           OutOperandKind = dwarf::DW_OP_const8u;
1249           break;
1250         default:
1251           Linker.reportWarning(
1252               formatv(("unsupported address size: {0}."), OrigAddressByteSize),
1253               File);
1254           break;
1255         }
1256 
1257         if (OutOperandKind) {
1258           OutputBuffer.push_back(*OutOperandKind);
1259           uint64_t LinkedAddress = SA->Address + AddrRelocAdjustment;
1260           if (IsLittleEndian != sys::IsLittleEndianHost)
1261             sys::swapByteOrder(LinkedAddress);
1262           ArrayRef<uint8_t> AddressBytes(
1263               reinterpret_cast<const uint8_t *>(&LinkedAddress),
1264               OrigAddressByteSize);
1265           OutputBuffer.append(AddressBytes.begin(), AddressBytes.end());
1266         }
1267       } else
1268         Linker.reportWarning("cannot read DW_OP_constx operand.", File);
1269     } else {
1270       // Copy over everything else unmodified.
1271       StringRef Bytes = Data.getData().slice(OpOffset, Op.getEndOffset());
1272       OutputBuffer.append(Bytes.begin(), Bytes.end());
1273     }
1274     OpOffset = Op.getEndOffset();
1275   }
1276 }
1277 
cloneBlockAttribute(DIE & Die,const DWARFDie & InputDIE,const DWARFFile & File,CompileUnit & Unit,AttributeSpec AttrSpec,const DWARFFormValue & Val,bool IsLittleEndian)1278 unsigned DWARFLinker::DIECloner::cloneBlockAttribute(
1279     DIE &Die, const DWARFDie &InputDIE, const DWARFFile &File,
1280     CompileUnit &Unit, AttributeSpec AttrSpec, const DWARFFormValue &Val,
1281     bool IsLittleEndian) {
1282   DIEValueList *Attr;
1283   DIEValue Value;
1284   DIELoc *Loc = nullptr;
1285   DIEBlock *Block = nullptr;
1286   if (AttrSpec.Form == dwarf::DW_FORM_exprloc) {
1287     Loc = new (DIEAlloc) DIELoc;
1288     Linker.DIELocs.push_back(Loc);
1289   } else {
1290     Block = new (DIEAlloc) DIEBlock;
1291     Linker.DIEBlocks.push_back(Block);
1292   }
1293   Attr = Loc ? static_cast<DIEValueList *>(Loc)
1294              : static_cast<DIEValueList *>(Block);
1295 
1296   DWARFUnit &OrigUnit = Unit.getOrigUnit();
1297   // If the block is a DWARF Expression, clone it into the temporary
1298   // buffer using cloneExpression(), otherwise copy the data directly.
1299   SmallVector<uint8_t, 32> Buffer;
1300   ArrayRef<uint8_t> Bytes = *Val.getAsBlock();
1301   if (DWARFAttribute::mayHaveLocationExpr(AttrSpec.Attr) &&
1302       (Val.isFormClass(DWARFFormValue::FC_Block) ||
1303        Val.isFormClass(DWARFFormValue::FC_Exprloc))) {
1304     DataExtractor Data(StringRef((const char *)Bytes.data(), Bytes.size()),
1305                        IsLittleEndian, OrigUnit.getAddressByteSize());
1306     DWARFExpression Expr(Data, OrigUnit.getAddressByteSize(),
1307                          OrigUnit.getFormParams().Format);
1308     cloneExpression(Data, Expr, File, Unit, Buffer,
1309                     Unit.getInfo(InputDIE).AddrAdjust, IsLittleEndian);
1310     Bytes = Buffer;
1311   }
1312   for (auto Byte : Bytes)
1313     Attr->addValue(DIEAlloc, static_cast<dwarf::Attribute>(0),
1314                    dwarf::DW_FORM_data1, DIEInteger(Byte));
1315 
1316   // FIXME: If DIEBlock and DIELoc just reuses the Size field of
1317   // the DIE class, this "if" could be replaced by
1318   // Attr->setSize(Bytes.size()).
1319   if (Loc)
1320     Loc->setSize(Bytes.size());
1321   else
1322     Block->setSize(Bytes.size());
1323 
1324   if (Loc)
1325     Value = DIEValue(dwarf::Attribute(AttrSpec.Attr),
1326                      dwarf::Form(AttrSpec.Form), Loc);
1327   else {
1328     // The expression location data might be updated and exceed the original
1329     // size. Check whether the new data fits into the original form.
1330     if ((AttrSpec.Form == dwarf::DW_FORM_block1 &&
1331          (Bytes.size() > UINT8_MAX)) ||
1332         (AttrSpec.Form == dwarf::DW_FORM_block2 &&
1333          (Bytes.size() > UINT16_MAX)) ||
1334         (AttrSpec.Form == dwarf::DW_FORM_block4 && (Bytes.size() > UINT32_MAX)))
1335       AttrSpec.Form = dwarf::DW_FORM_block;
1336 
1337     Value = DIEValue(dwarf::Attribute(AttrSpec.Attr),
1338                      dwarf::Form(AttrSpec.Form), Block);
1339   }
1340 
1341   return Die.addValue(DIEAlloc, Value)->sizeOf(OrigUnit.getFormParams());
1342 }
1343 
cloneAddressAttribute(DIE & Die,const DWARFDie & InputDIE,AttributeSpec AttrSpec,unsigned AttrSize,const DWARFFormValue & Val,const CompileUnit & Unit,AttributesInfo & Info)1344 unsigned DWARFLinker::DIECloner::cloneAddressAttribute(
1345     DIE &Die, const DWARFDie &InputDIE, AttributeSpec AttrSpec,
1346     unsigned AttrSize, const DWARFFormValue &Val, const CompileUnit &Unit,
1347     AttributesInfo &Info) {
1348   if (AttrSpec.Attr == dwarf::DW_AT_low_pc)
1349     Info.HasLowPc = true;
1350 
1351   if (LLVM_UNLIKELY(Linker.Options.Update)) {
1352     Die.addValue(DIEAlloc, dwarf::Attribute(AttrSpec.Attr),
1353                  dwarf::Form(AttrSpec.Form), DIEInteger(Val.getRawUValue()));
1354     return AttrSize;
1355   }
1356 
1357   // Cloned Die may have address attributes relocated to a
1358   // totally unrelated value. This can happen:
1359   //   - If high_pc is an address (Dwarf version == 2), then it might have been
1360   //     relocated to a totally unrelated value (because the end address in the
1361   //     object file might be start address of another function which got moved
1362   //     independently by the linker).
1363   //   - If address relocated in an inline_subprogram that happens at the
1364   //     beginning of its inlining function.
1365   //  To avoid above cases and to not apply relocation twice (in
1366   //  applyValidRelocs and here), read address attribute from InputDIE and apply
1367   //  Info.PCOffset here.
1368 
1369   std::optional<DWARFFormValue> AddrAttribute = InputDIE.find(AttrSpec.Attr);
1370   if (!AddrAttribute)
1371     llvm_unreachable("Cann't find attribute.");
1372 
1373   std::optional<uint64_t> Addr = AddrAttribute->getAsAddress();
1374   if (!Addr) {
1375     Linker.reportWarning("Cann't read address attribute value.", ObjFile);
1376     return 0;
1377   }
1378 
1379   if (InputDIE.getTag() == dwarf::DW_TAG_compile_unit &&
1380       AttrSpec.Attr == dwarf::DW_AT_low_pc) {
1381     if (std::optional<uint64_t> LowPC = Unit.getLowPc())
1382       Addr = *LowPC;
1383     else
1384       return 0;
1385   } else if (InputDIE.getTag() == dwarf::DW_TAG_compile_unit &&
1386              AttrSpec.Attr == dwarf::DW_AT_high_pc) {
1387     if (uint64_t HighPc = Unit.getHighPc())
1388       Addr = HighPc;
1389     else
1390       return 0;
1391   } else {
1392     *Addr += Info.PCOffset;
1393   }
1394 
1395   if (AttrSpec.Form == dwarf::DW_FORM_addr) {
1396     Die.addValue(DIEAlloc, static_cast<dwarf::Attribute>(AttrSpec.Attr),
1397                  AttrSpec.Form, DIEInteger(*Addr));
1398     return Unit.getOrigUnit().getAddressByteSize();
1399   }
1400 
1401   auto AddrIndex = AddrPool.getValueIndex(*Addr);
1402 
1403   return Die
1404       .addValue(DIEAlloc, static_cast<dwarf::Attribute>(AttrSpec.Attr),
1405                 dwarf::Form::DW_FORM_addrx, DIEInteger(AddrIndex))
1406       ->sizeOf(Unit.getOrigUnit().getFormParams());
1407 }
1408 
cloneScalarAttribute(DIE & Die,const DWARFDie & InputDIE,const DWARFFile & File,CompileUnit & Unit,AttributeSpec AttrSpec,const DWARFFormValue & Val,unsigned AttrSize,AttributesInfo & Info)1409 unsigned DWARFLinker::DIECloner::cloneScalarAttribute(
1410     DIE &Die, const DWARFDie &InputDIE, const DWARFFile &File,
1411     CompileUnit &Unit, AttributeSpec AttrSpec, const DWARFFormValue &Val,
1412     unsigned AttrSize, AttributesInfo &Info) {
1413   uint64_t Value;
1414 
1415   // Check for the offset to the macro table. If offset is incorrect then we
1416   // need to remove the attribute.
1417   if (AttrSpec.Attr == dwarf::DW_AT_macro_info) {
1418     if (std::optional<uint64_t> Offset = Val.getAsSectionOffset()) {
1419       const llvm::DWARFDebugMacro *Macro = File.Dwarf->getDebugMacinfo();
1420       if (Macro == nullptr || !Macro->hasEntryForOffset(*Offset))
1421         return 0;
1422     }
1423   }
1424 
1425   if (AttrSpec.Attr == dwarf::DW_AT_macros) {
1426     if (std::optional<uint64_t> Offset = Val.getAsSectionOffset()) {
1427       const llvm::DWARFDebugMacro *Macro = File.Dwarf->getDebugMacro();
1428       if (Macro == nullptr || !Macro->hasEntryForOffset(*Offset))
1429         return 0;
1430     }
1431   }
1432 
1433   if (AttrSpec.Attr == dwarf::DW_AT_str_offsets_base) {
1434     // DWARFLinker generates common .debug_str_offsets table used for all
1435     // compile units. The offset to the common .debug_str_offsets table is 8 on
1436     // DWARF32.
1437     Info.AttrStrOffsetBaseSeen = true;
1438     return Die
1439         .addValue(DIEAlloc, dwarf::DW_AT_str_offsets_base,
1440                   dwarf::DW_FORM_sec_offset, DIEInteger(8))
1441         ->sizeOf(Unit.getOrigUnit().getFormParams());
1442   }
1443 
1444   if (LLVM_UNLIKELY(Linker.Options.Update)) {
1445     if (auto OptionalValue = Val.getAsUnsignedConstant())
1446       Value = *OptionalValue;
1447     else if (auto OptionalValue = Val.getAsSignedConstant())
1448       Value = *OptionalValue;
1449     else if (auto OptionalValue = Val.getAsSectionOffset())
1450       Value = *OptionalValue;
1451     else {
1452       Linker.reportWarning(
1453           "Unsupported scalar attribute form. Dropping attribute.", File,
1454           &InputDIE);
1455       return 0;
1456     }
1457     if (AttrSpec.Attr == dwarf::DW_AT_declaration && Value)
1458       Info.IsDeclaration = true;
1459 
1460     if (AttrSpec.Form == dwarf::DW_FORM_loclistx)
1461       Die.addValue(DIEAlloc, dwarf::Attribute(AttrSpec.Attr),
1462                    dwarf::Form(AttrSpec.Form), DIELocList(Value));
1463     else
1464       Die.addValue(DIEAlloc, dwarf::Attribute(AttrSpec.Attr),
1465                    dwarf::Form(AttrSpec.Form), DIEInteger(Value));
1466     return AttrSize;
1467   }
1468 
1469   [[maybe_unused]] dwarf::Form OriginalForm = AttrSpec.Form;
1470   if (AttrSpec.Form == dwarf::DW_FORM_rnglistx) {
1471     // DWARFLinker does not generate .debug_addr table. Thus we need to change
1472     // all "addrx" related forms to "addr" version. Change DW_FORM_rnglistx
1473     // to DW_FORM_sec_offset here.
1474     std::optional<uint64_t> Index = Val.getAsSectionOffset();
1475     if (!Index) {
1476       Linker.reportWarning("Cannot read the attribute. Dropping.", File,
1477                            &InputDIE);
1478       return 0;
1479     }
1480     std::optional<uint64_t> Offset =
1481         Unit.getOrigUnit().getRnglistOffset(*Index);
1482     if (!Offset) {
1483       Linker.reportWarning("Cannot read the attribute. Dropping.", File,
1484                            &InputDIE);
1485       return 0;
1486     }
1487 
1488     Value = *Offset;
1489     AttrSpec.Form = dwarf::DW_FORM_sec_offset;
1490     AttrSize = Unit.getOrigUnit().getFormParams().getDwarfOffsetByteSize();
1491   } else if (AttrSpec.Form == dwarf::DW_FORM_loclistx) {
1492     // DWARFLinker does not generate .debug_addr table. Thus we need to change
1493     // all "addrx" related forms to "addr" version. Change DW_FORM_loclistx
1494     // to DW_FORM_sec_offset here.
1495     std::optional<uint64_t> Index = Val.getAsSectionOffset();
1496     if (!Index) {
1497       Linker.reportWarning("Cannot read the attribute. Dropping.", File,
1498                            &InputDIE);
1499       return 0;
1500     }
1501     std::optional<uint64_t> Offset =
1502         Unit.getOrigUnit().getLoclistOffset(*Index);
1503     if (!Offset) {
1504       Linker.reportWarning("Cannot read the attribute. Dropping.", File,
1505                            &InputDIE);
1506       return 0;
1507     }
1508 
1509     Value = *Offset;
1510     AttrSpec.Form = dwarf::DW_FORM_sec_offset;
1511     AttrSize = Unit.getOrigUnit().getFormParams().getDwarfOffsetByteSize();
1512   } else if (AttrSpec.Attr == dwarf::DW_AT_high_pc &&
1513              Die.getTag() == dwarf::DW_TAG_compile_unit) {
1514     std::optional<uint64_t> LowPC = Unit.getLowPc();
1515     if (!LowPC)
1516       return 0;
1517     // Dwarf >= 4 high_pc is an size, not an address.
1518     Value = Unit.getHighPc() - *LowPC;
1519   } else if (AttrSpec.Form == dwarf::DW_FORM_sec_offset)
1520     Value = *Val.getAsSectionOffset();
1521   else if (AttrSpec.Form == dwarf::DW_FORM_sdata)
1522     Value = *Val.getAsSignedConstant();
1523   else if (auto OptionalValue = Val.getAsUnsignedConstant())
1524     Value = *OptionalValue;
1525   else {
1526     Linker.reportWarning(
1527         "Unsupported scalar attribute form. Dropping attribute.", File,
1528         &InputDIE);
1529     return 0;
1530   }
1531 
1532   DIE::value_iterator Patch =
1533       Die.addValue(DIEAlloc, dwarf::Attribute(AttrSpec.Attr),
1534                    dwarf::Form(AttrSpec.Form), DIEInteger(Value));
1535   if (AttrSpec.Attr == dwarf::DW_AT_ranges ||
1536       AttrSpec.Attr == dwarf::DW_AT_start_scope) {
1537     Unit.noteRangeAttribute(Die, Patch);
1538     Info.HasRanges = true;
1539   } else if (DWARFAttribute::mayHaveLocationList(AttrSpec.Attr) &&
1540              dwarf::doesFormBelongToClass(AttrSpec.Form,
1541                                           DWARFFormValue::FC_SectionOffset,
1542                                           Unit.getOrigUnit().getVersion())) {
1543 
1544     CompileUnit::DIEInfo &LocationDieInfo = Unit.getInfo(InputDIE);
1545     Unit.noteLocationAttribute({Patch, LocationDieInfo.InDebugMap
1546                                            ? LocationDieInfo.AddrAdjust
1547                                            : Info.PCOffset});
1548   } else if (AttrSpec.Attr == dwarf::DW_AT_declaration && Value)
1549     Info.IsDeclaration = true;
1550 
1551   // check that all dwarf::DW_FORM_rnglistx are handled previously.
1552   assert((Info.HasRanges || (OriginalForm != dwarf::DW_FORM_rnglistx)) &&
1553          "Unhandled DW_FORM_rnglistx attribute");
1554 
1555   return AttrSize;
1556 }
1557 
1558 /// Clone \p InputDIE's attribute described by \p AttrSpec with
1559 /// value \p Val, and add it to \p Die.
1560 /// \returns the size of the cloned attribute.
cloneAttribute(DIE & Die,const DWARFDie & InputDIE,const DWARFFile & File,CompileUnit & Unit,const DWARFFormValue & Val,const AttributeSpec AttrSpec,unsigned AttrSize,AttributesInfo & Info,bool IsLittleEndian)1561 unsigned DWARFLinker::DIECloner::cloneAttribute(
1562     DIE &Die, const DWARFDie &InputDIE, const DWARFFile &File,
1563     CompileUnit &Unit, const DWARFFormValue &Val, const AttributeSpec AttrSpec,
1564     unsigned AttrSize, AttributesInfo &Info, bool IsLittleEndian) {
1565   const DWARFUnit &U = Unit.getOrigUnit();
1566 
1567   switch (AttrSpec.Form) {
1568   case dwarf::DW_FORM_strp:
1569   case dwarf::DW_FORM_line_strp:
1570   case dwarf::DW_FORM_string:
1571   case dwarf::DW_FORM_strx:
1572   case dwarf::DW_FORM_strx1:
1573   case dwarf::DW_FORM_strx2:
1574   case dwarf::DW_FORM_strx3:
1575   case dwarf::DW_FORM_strx4:
1576     return cloneStringAttribute(Die, AttrSpec, Val, U, Info);
1577   case dwarf::DW_FORM_ref_addr:
1578   case dwarf::DW_FORM_ref1:
1579   case dwarf::DW_FORM_ref2:
1580   case dwarf::DW_FORM_ref4:
1581   case dwarf::DW_FORM_ref8:
1582     return cloneDieReferenceAttribute(Die, InputDIE, AttrSpec, AttrSize, Val,
1583                                       File, Unit);
1584   case dwarf::DW_FORM_block:
1585   case dwarf::DW_FORM_block1:
1586   case dwarf::DW_FORM_block2:
1587   case dwarf::DW_FORM_block4:
1588   case dwarf::DW_FORM_exprloc:
1589     return cloneBlockAttribute(Die, InputDIE, File, Unit, AttrSpec, Val,
1590                                IsLittleEndian);
1591   case dwarf::DW_FORM_addr:
1592   case dwarf::DW_FORM_addrx:
1593   case dwarf::DW_FORM_addrx1:
1594   case dwarf::DW_FORM_addrx2:
1595   case dwarf::DW_FORM_addrx3:
1596   case dwarf::DW_FORM_addrx4:
1597     return cloneAddressAttribute(Die, InputDIE, AttrSpec, AttrSize, Val, Unit,
1598                                  Info);
1599   case dwarf::DW_FORM_data1:
1600   case dwarf::DW_FORM_data2:
1601   case dwarf::DW_FORM_data4:
1602   case dwarf::DW_FORM_data8:
1603   case dwarf::DW_FORM_udata:
1604   case dwarf::DW_FORM_sdata:
1605   case dwarf::DW_FORM_sec_offset:
1606   case dwarf::DW_FORM_flag:
1607   case dwarf::DW_FORM_flag_present:
1608   case dwarf::DW_FORM_rnglistx:
1609   case dwarf::DW_FORM_loclistx:
1610   case dwarf::DW_FORM_implicit_const:
1611     return cloneScalarAttribute(Die, InputDIE, File, Unit, AttrSpec, Val,
1612                                 AttrSize, Info);
1613   default:
1614     Linker.reportWarning("Unsupported attribute form " +
1615                              dwarf::FormEncodingString(AttrSpec.Form) +
1616                              " in cloneAttribute. Dropping.",
1617                          File, &InputDIE);
1618   }
1619 
1620   return 0;
1621 }
1622 
addObjCAccelerator(CompileUnit & Unit,const DIE * Die,DwarfStringPoolEntryRef Name,OffsetsStringPool & StringPool,bool SkipPubSection)1623 void DWARFLinker::DIECloner::addObjCAccelerator(CompileUnit &Unit,
1624                                                 const DIE *Die,
1625                                                 DwarfStringPoolEntryRef Name,
1626                                                 OffsetsStringPool &StringPool,
1627                                                 bool SkipPubSection) {
1628   std::optional<ObjCSelectorNames> Names =
1629       getObjCNamesIfSelector(Name.getString());
1630   if (!Names)
1631     return;
1632   Unit.addNameAccelerator(Die, StringPool.getEntry(Names->Selector),
1633                           SkipPubSection);
1634   Unit.addObjCAccelerator(Die, StringPool.getEntry(Names->ClassName),
1635                           SkipPubSection);
1636   if (Names->ClassNameNoCategory)
1637     Unit.addObjCAccelerator(
1638         Die, StringPool.getEntry(*Names->ClassNameNoCategory), SkipPubSection);
1639   if (Names->MethodNameNoCategory)
1640     Unit.addNameAccelerator(
1641         Die, StringPool.getEntry(*Names->MethodNameNoCategory), SkipPubSection);
1642 }
1643 
1644 static bool
shouldSkipAttribute(bool Update,DWARFAbbreviationDeclaration::AttributeSpec AttrSpec,bool SkipPC)1645 shouldSkipAttribute(bool Update,
1646                     DWARFAbbreviationDeclaration::AttributeSpec AttrSpec,
1647                     bool SkipPC) {
1648   switch (AttrSpec.Attr) {
1649   default:
1650     return false;
1651   case dwarf::DW_AT_low_pc:
1652   case dwarf::DW_AT_high_pc:
1653   case dwarf::DW_AT_ranges:
1654     return !Update && SkipPC;
1655   case dwarf::DW_AT_rnglists_base:
1656     // In case !Update the .debug_addr table is not generated/preserved.
1657     // Thus instead of DW_FORM_rnglistx the DW_FORM_sec_offset is used.
1658     // Since DW_AT_rnglists_base is used for only DW_FORM_rnglistx the
1659     // DW_AT_rnglists_base is removed.
1660     return !Update;
1661   case dwarf::DW_AT_loclists_base:
1662     // In case !Update the .debug_addr table is not generated/preserved.
1663     // Thus instead of DW_FORM_loclistx the DW_FORM_sec_offset is used.
1664     // Since DW_AT_loclists_base is used for only DW_FORM_loclistx the
1665     // DW_AT_loclists_base is removed.
1666     return !Update;
1667   case dwarf::DW_AT_location:
1668   case dwarf::DW_AT_frame_base:
1669     return !Update && SkipPC;
1670   }
1671 }
1672 
1673 struct AttributeLinkedOffsetFixup {
1674   int64_t LinkedOffsetFixupVal;
1675   uint64_t InputAttrStartOffset;
1676   uint64_t InputAttrEndOffset;
1677 };
1678 
cloneDIE(const DWARFDie & InputDIE,const DWARFFile & File,CompileUnit & Unit,int64_t PCOffset,uint32_t OutOffset,unsigned Flags,bool IsLittleEndian,DIE * Die)1679 DIE *DWARFLinker::DIECloner::cloneDIE(const DWARFDie &InputDIE,
1680                                       const DWARFFile &File, CompileUnit &Unit,
1681                                       int64_t PCOffset, uint32_t OutOffset,
1682                                       unsigned Flags, bool IsLittleEndian,
1683                                       DIE *Die) {
1684   DWARFUnit &U = Unit.getOrigUnit();
1685   unsigned Idx = U.getDIEIndex(InputDIE);
1686   CompileUnit::DIEInfo &Info = Unit.getInfo(Idx);
1687 
1688   // Should the DIE appear in the output?
1689   if (!Unit.getInfo(Idx).Keep)
1690     return nullptr;
1691 
1692   uint64_t Offset = InputDIE.getOffset();
1693   assert(!(Die && Info.Clone) && "Can't supply a DIE and a cloned DIE");
1694   if (!Die) {
1695     // The DIE might have been already created by a forward reference
1696     // (see cloneDieReferenceAttribute()).
1697     if (!Info.Clone)
1698       Info.Clone = DIE::get(DIEAlloc, dwarf::Tag(InputDIE.getTag()));
1699     Die = Info.Clone;
1700   }
1701 
1702   assert(Die->getTag() == InputDIE.getTag());
1703   Die->setOffset(OutOffset);
1704   if (isODRCanonicalCandidate(InputDIE, Unit) && Info.Ctxt &&
1705       (Info.Ctxt->getCanonicalDIEOffset() == 0)) {
1706     if (!Info.Ctxt->hasCanonicalDIE())
1707       Info.Ctxt->setHasCanonicalDIE();
1708     // We are about to emit a DIE that is the root of its own valid
1709     // DeclContext tree. Make the current offset the canonical offset
1710     // for this context.
1711     Info.Ctxt->setCanonicalDIEOffset(OutOffset + Unit.getStartOffset());
1712   }
1713 
1714   // Extract and clone every attribute.
1715   DWARFDataExtractor Data = U.getDebugInfoExtractor();
1716   // Point to the next DIE (generally there is always at least a NULL
1717   // entry after the current one). If this is a lone
1718   // DW_TAG_compile_unit without any children, point to the next unit.
1719   uint64_t NextOffset = (Idx + 1 < U.getNumDIEs())
1720                             ? U.getDIEAtIndex(Idx + 1).getOffset()
1721                             : U.getNextUnitOffset();
1722   AttributesInfo AttrInfo;
1723 
1724   // We could copy the data only if we need to apply a relocation to it. After
1725   // testing, it seems there is no performance downside to doing the copy
1726   // unconditionally, and it makes the code simpler.
1727   SmallString<40> DIECopy(Data.getData().substr(Offset, NextOffset - Offset));
1728   Data =
1729       DWARFDataExtractor(DIECopy, Data.isLittleEndian(), Data.getAddressSize());
1730 
1731   // Modify the copy with relocated addresses.
1732   ObjFile.Addresses->applyValidRelocs(DIECopy, Offset, Data.isLittleEndian());
1733 
1734   // Reset the Offset to 0 as we will be working on the local copy of
1735   // the data.
1736   Offset = 0;
1737 
1738   const auto *Abbrev = InputDIE.getAbbreviationDeclarationPtr();
1739   Offset += getULEB128Size(Abbrev->getCode());
1740 
1741   // We are entering a subprogram. Get and propagate the PCOffset.
1742   if (Die->getTag() == dwarf::DW_TAG_subprogram)
1743     PCOffset = Info.AddrAdjust;
1744   AttrInfo.PCOffset = PCOffset;
1745 
1746   if (Abbrev->getTag() == dwarf::DW_TAG_subprogram) {
1747     Flags |= TF_InFunctionScope;
1748     if (!Info.InDebugMap && LLVM_LIKELY(!Update))
1749       Flags |= TF_SkipPC;
1750   } else if (Abbrev->getTag() == dwarf::DW_TAG_variable) {
1751     // Function-local globals could be in the debug map even when the function
1752     // is not, e.g., inlined functions.
1753     if ((Flags & TF_InFunctionScope) && Info.InDebugMap)
1754       Flags &= ~TF_SkipPC;
1755     // Location expressions referencing an address which is not in debug map
1756     // should be deleted.
1757     else if (!Info.InDebugMap && Info.HasLocationExpressionAddr &&
1758              LLVM_LIKELY(!Update))
1759       Flags |= TF_SkipPC;
1760   }
1761 
1762   std::optional<StringRef> LibraryInstallName =
1763       ObjFile.Addresses->getLibraryInstallName();
1764   SmallVector<AttributeLinkedOffsetFixup> AttributesFixups;
1765   for (const auto &AttrSpec : Abbrev->attributes()) {
1766     if (shouldSkipAttribute(Update, AttrSpec, Flags & TF_SkipPC)) {
1767       DWARFFormValue::skipValue(AttrSpec.Form, Data, &Offset,
1768                                 U.getFormParams());
1769       continue;
1770     }
1771 
1772     AttributeLinkedOffsetFixup CurAttrFixup;
1773     CurAttrFixup.InputAttrStartOffset = InputDIE.getOffset() + Offset;
1774     CurAttrFixup.LinkedOffsetFixupVal =
1775         Unit.getStartOffset() + OutOffset - CurAttrFixup.InputAttrStartOffset;
1776 
1777     DWARFFormValue Val = AttrSpec.getFormValue();
1778     uint64_t AttrSize = Offset;
1779     Val.extractValue(Data, &Offset, U.getFormParams(), &U);
1780     CurAttrFixup.InputAttrEndOffset = InputDIE.getOffset() + Offset;
1781     AttrSize = Offset - AttrSize;
1782 
1783     uint64_t FinalAttrSize =
1784         cloneAttribute(*Die, InputDIE, File, Unit, Val, AttrSpec, AttrSize,
1785                        AttrInfo, IsLittleEndian);
1786     if (FinalAttrSize != 0 && ObjFile.Addresses->needToSaveValidRelocs())
1787       AttributesFixups.push_back(CurAttrFixup);
1788 
1789     OutOffset += FinalAttrSize;
1790   }
1791 
1792   uint16_t Tag = InputDIE.getTag();
1793   // Add the DW_AT_APPLE_origin attribute to Compile Unit die if we have
1794   // an install name and the DWARF doesn't have the attribute yet.
1795   const bool NeedsAppleOrigin = (Tag == dwarf::DW_TAG_compile_unit) &&
1796                                 LibraryInstallName.has_value() &&
1797                                 !AttrInfo.HasAppleOrigin;
1798   if (NeedsAppleOrigin) {
1799     auto StringEntry = DebugStrPool.getEntry(LibraryInstallName.value());
1800     Die->addValue(DIEAlloc, dwarf::Attribute(dwarf::DW_AT_APPLE_origin),
1801                   dwarf::DW_FORM_strp, DIEInteger(StringEntry.getOffset()));
1802     AttrInfo.Name = StringEntry;
1803     OutOffset += 4;
1804   }
1805 
1806   // Look for accelerator entries.
1807   // FIXME: This is slightly wrong. An inline_subroutine without a
1808   // low_pc, but with AT_ranges might be interesting to get into the
1809   // accelerator tables too. For now stick with dsymutil's behavior.
1810   if ((Info.InDebugMap || AttrInfo.HasLowPc || AttrInfo.HasRanges) &&
1811       Tag != dwarf::DW_TAG_compile_unit &&
1812       getDIENames(InputDIE, AttrInfo, DebugStrPool,
1813                   Tag != dwarf::DW_TAG_inlined_subroutine)) {
1814     if (AttrInfo.MangledName && AttrInfo.MangledName != AttrInfo.Name)
1815       Unit.addNameAccelerator(Die, AttrInfo.MangledName,
1816                               Tag == dwarf::DW_TAG_inlined_subroutine);
1817     if (AttrInfo.Name) {
1818       if (AttrInfo.NameWithoutTemplate)
1819         Unit.addNameAccelerator(Die, AttrInfo.NameWithoutTemplate,
1820                                 /* SkipPubSection */ true);
1821       Unit.addNameAccelerator(Die, AttrInfo.Name,
1822                               Tag == dwarf::DW_TAG_inlined_subroutine);
1823     }
1824     if (AttrInfo.Name)
1825       addObjCAccelerator(Unit, Die, AttrInfo.Name, DebugStrPool,
1826                          /* SkipPubSection =*/true);
1827 
1828   } else if (Tag == dwarf::DW_TAG_namespace) {
1829     if (!AttrInfo.Name)
1830       AttrInfo.Name = DebugStrPool.getEntry("(anonymous namespace)");
1831     Unit.addNamespaceAccelerator(Die, AttrInfo.Name);
1832   } else if (Tag == dwarf::DW_TAG_imported_declaration && AttrInfo.Name) {
1833     Unit.addNamespaceAccelerator(Die, AttrInfo.Name);
1834   } else if (isTypeTag(Tag) && !AttrInfo.IsDeclaration &&
1835              getDIENames(InputDIE, AttrInfo, DebugStrPool) && AttrInfo.Name &&
1836              AttrInfo.Name.getString()[0]) {
1837     uint32_t Hash = hashFullyQualifiedName(InputDIE, Unit, File);
1838     uint64_t RuntimeLang =
1839         dwarf::toUnsigned(InputDIE.find(dwarf::DW_AT_APPLE_runtime_class))
1840             .value_or(0);
1841     bool ObjCClassIsImplementation =
1842         (RuntimeLang == dwarf::DW_LANG_ObjC ||
1843          RuntimeLang == dwarf::DW_LANG_ObjC_plus_plus) &&
1844         dwarf::toUnsigned(InputDIE.find(dwarf::DW_AT_APPLE_objc_complete_type))
1845             .value_or(0);
1846     Unit.addTypeAccelerator(Die, AttrInfo.Name, ObjCClassIsImplementation,
1847                             Hash);
1848   }
1849 
1850   // Determine whether there are any children that we want to keep.
1851   bool HasChildren = false;
1852   for (auto Child : InputDIE.children()) {
1853     unsigned Idx = U.getDIEIndex(Child);
1854     if (Unit.getInfo(Idx).Keep) {
1855       HasChildren = true;
1856       break;
1857     }
1858   }
1859 
1860   if (Unit.getOrigUnit().getVersion() >= 5 && !AttrInfo.AttrStrOffsetBaseSeen &&
1861       Die->getTag() == dwarf::DW_TAG_compile_unit) {
1862     // No DW_AT_str_offsets_base seen, add it to the DIE.
1863     Die->addValue(DIEAlloc, dwarf::DW_AT_str_offsets_base,
1864                   dwarf::DW_FORM_sec_offset, DIEInteger(8));
1865     OutOffset += 4;
1866   }
1867 
1868   DIEAbbrev NewAbbrev = Die->generateAbbrev();
1869   if (HasChildren)
1870     NewAbbrev.setChildrenFlag(dwarf::DW_CHILDREN_yes);
1871   // Assign a permanent abbrev number
1872   Linker.assignAbbrev(NewAbbrev);
1873   Die->setAbbrevNumber(NewAbbrev.getNumber());
1874 
1875   uint64_t AbbrevNumberSize = getULEB128Size(Die->getAbbrevNumber());
1876 
1877   // Add the size of the abbreviation number to the output offset.
1878   OutOffset += AbbrevNumberSize;
1879 
1880   // Update fixups with the size of the abbreviation number
1881   for (AttributeLinkedOffsetFixup &F : AttributesFixups)
1882     F.LinkedOffsetFixupVal += AbbrevNumberSize;
1883 
1884   for (AttributeLinkedOffsetFixup &F : AttributesFixups)
1885     ObjFile.Addresses->updateAndSaveValidRelocs(
1886         Unit.getOrigUnit().getVersion() >= 5, Unit.getOrigUnit().getOffset(),
1887         F.LinkedOffsetFixupVal, F.InputAttrStartOffset, F.InputAttrEndOffset);
1888 
1889   if (!HasChildren) {
1890     // Update our size.
1891     Die->setSize(OutOffset - Die->getOffset());
1892     return Die;
1893   }
1894 
1895   // Recursively clone children.
1896   for (auto Child : InputDIE.children()) {
1897     if (DIE *Clone = cloneDIE(Child, File, Unit, PCOffset, OutOffset, Flags,
1898                               IsLittleEndian)) {
1899       Die->addChild(Clone);
1900       OutOffset = Clone->getOffset() + Clone->getSize();
1901     }
1902   }
1903 
1904   // Account for the end of children marker.
1905   OutOffset += sizeof(int8_t);
1906   // Update our size.
1907   Die->setSize(OutOffset - Die->getOffset());
1908   return Die;
1909 }
1910 
1911 /// Patch the input object file relevant debug_ranges or debug_rnglists
1912 /// entries and emit them in the output file. Update the relevant attributes
1913 /// to point at the new entries.
generateUnitRanges(CompileUnit & Unit,const DWARFFile & File,DebugDieValuePool & AddrPool) const1914 void DWARFLinker::generateUnitRanges(CompileUnit &Unit, const DWARFFile &File,
1915                                      DebugDieValuePool &AddrPool) const {
1916   if (LLVM_UNLIKELY(Options.Update))
1917     return;
1918 
1919   const auto &FunctionRanges = Unit.getFunctionRanges();
1920 
1921   // Build set of linked address ranges for unit function ranges.
1922   AddressRanges LinkedFunctionRanges;
1923   for (const AddressRangeValuePair &Range : FunctionRanges)
1924     LinkedFunctionRanges.insert(
1925         {Range.Range.start() + Range.Value, Range.Range.end() + Range.Value});
1926 
1927   // Emit LinkedFunctionRanges into .debug_aranges
1928   if (!LinkedFunctionRanges.empty())
1929     TheDwarfEmitter->emitDwarfDebugArangesTable(Unit, LinkedFunctionRanges);
1930 
1931   RngListAttributesTy AllRngListAttributes = Unit.getRangesAttributes();
1932   std::optional<PatchLocation> UnitRngListAttribute =
1933       Unit.getUnitRangesAttribute();
1934 
1935   if (!AllRngListAttributes.empty() || UnitRngListAttribute) {
1936     std::optional<AddressRangeValuePair> CachedRange;
1937     MCSymbol *EndLabel = TheDwarfEmitter->emitDwarfDebugRangeListHeader(Unit);
1938 
1939     // Read original address ranges, apply relocation value, emit linked address
1940     // ranges.
1941     for (PatchLocation &AttributePatch : AllRngListAttributes) {
1942       // Get ranges from the source DWARF corresponding to the current
1943       // attribute.
1944       AddressRanges LinkedRanges;
1945       if (Expected<DWARFAddressRangesVector> OriginalRanges =
1946               Unit.getOrigUnit().findRnglistFromOffset(AttributePatch.get())) {
1947         // Apply relocation adjustment.
1948         for (const auto &Range : *OriginalRanges) {
1949           if (!CachedRange || !CachedRange->Range.contains(Range.LowPC))
1950             CachedRange = FunctionRanges.getRangeThatContains(Range.LowPC);
1951 
1952           // All range entries should lie in the function range.
1953           if (!CachedRange) {
1954             reportWarning("inconsistent range data.", File);
1955             continue;
1956           }
1957 
1958           // Store range for emiting.
1959           LinkedRanges.insert({Range.LowPC + CachedRange->Value,
1960                                Range.HighPC + CachedRange->Value});
1961         }
1962       } else {
1963         llvm::consumeError(OriginalRanges.takeError());
1964         reportWarning("invalid range list ignored.", File);
1965       }
1966 
1967       // Emit linked ranges.
1968       TheDwarfEmitter->emitDwarfDebugRangeListFragment(
1969           Unit, LinkedRanges, AttributePatch, AddrPool);
1970     }
1971 
1972     // Emit ranges for Unit AT_ranges attribute.
1973     if (UnitRngListAttribute.has_value())
1974       TheDwarfEmitter->emitDwarfDebugRangeListFragment(
1975           Unit, LinkedFunctionRanges, *UnitRngListAttribute, AddrPool);
1976 
1977     // Emit ranges footer.
1978     TheDwarfEmitter->emitDwarfDebugRangeListFooter(Unit, EndLabel);
1979   }
1980 }
1981 
generateUnitLocations(CompileUnit & Unit,const DWARFFile & File,ExpressionHandlerRef ExprHandler)1982 void DWARFLinker::DIECloner::generateUnitLocations(
1983     CompileUnit &Unit, const DWARFFile &File,
1984     ExpressionHandlerRef ExprHandler) {
1985   if (LLVM_UNLIKELY(Linker.Options.Update))
1986     return;
1987 
1988   const LocListAttributesTy &AllLocListAttributes =
1989       Unit.getLocationAttributes();
1990 
1991   if (AllLocListAttributes.empty())
1992     return;
1993 
1994   // Emit locations list table header.
1995   MCSymbol *EndLabel = Emitter->emitDwarfDebugLocListHeader(Unit);
1996 
1997   for (auto &CurLocAttr : AllLocListAttributes) {
1998     // Get location expressions vector corresponding to the current attribute
1999     // from the source DWARF.
2000     Expected<DWARFLocationExpressionsVector> OriginalLocations =
2001         Unit.getOrigUnit().findLoclistFromOffset(CurLocAttr.get());
2002 
2003     if (!OriginalLocations) {
2004       llvm::consumeError(OriginalLocations.takeError());
2005       Linker.reportWarning("Invalid location attribute ignored.", File);
2006       continue;
2007     }
2008 
2009     DWARFLocationExpressionsVector LinkedLocationExpressions;
2010     for (DWARFLocationExpression &CurExpression : *OriginalLocations) {
2011       DWARFLocationExpression LinkedExpression;
2012 
2013       if (CurExpression.Range) {
2014         // Relocate address range.
2015         LinkedExpression.Range = {
2016             CurExpression.Range->LowPC + CurLocAttr.RelocAdjustment,
2017             CurExpression.Range->HighPC + CurLocAttr.RelocAdjustment};
2018       }
2019 
2020       // Clone expression.
2021       LinkedExpression.Expr.reserve(CurExpression.Expr.size());
2022       ExprHandler(CurExpression.Expr, LinkedExpression.Expr,
2023                   CurLocAttr.RelocAdjustment);
2024 
2025       LinkedLocationExpressions.push_back(LinkedExpression);
2026     }
2027 
2028     // Emit locations list table fragment corresponding to the CurLocAttr.
2029     Emitter->emitDwarfDebugLocListFragment(Unit, LinkedLocationExpressions,
2030                                            CurLocAttr, AddrPool);
2031   }
2032 
2033   // Emit locations list table footer.
2034   Emitter->emitDwarfDebugLocListFooter(Unit, EndLabel);
2035 }
2036 
patchAddrBase(DIE & Die,DIEInteger Offset)2037 static void patchAddrBase(DIE &Die, DIEInteger Offset) {
2038   for (auto &V : Die.values())
2039     if (V.getAttribute() == dwarf::DW_AT_addr_base) {
2040       V = DIEValue(V.getAttribute(), V.getForm(), Offset);
2041       return;
2042     }
2043 
2044   llvm_unreachable("Didn't find a DW_AT_addr_base in cloned DIE!");
2045 }
2046 
emitDebugAddrSection(CompileUnit & Unit,const uint16_t DwarfVersion) const2047 void DWARFLinker::DIECloner::emitDebugAddrSection(
2048     CompileUnit &Unit, const uint16_t DwarfVersion) const {
2049 
2050   if (LLVM_UNLIKELY(Linker.Options.Update))
2051     return;
2052 
2053   if (DwarfVersion < 5)
2054     return;
2055 
2056   if (AddrPool.getValues().empty())
2057     return;
2058 
2059   MCSymbol *EndLabel = Emitter->emitDwarfDebugAddrsHeader(Unit);
2060   patchAddrBase(*Unit.getOutputUnitDIE(),
2061                 DIEInteger(Emitter->getDebugAddrSectionSize()));
2062   Emitter->emitDwarfDebugAddrs(AddrPool.getValues(),
2063                                Unit.getOrigUnit().getAddressByteSize());
2064   Emitter->emitDwarfDebugAddrsFooter(Unit, EndLabel);
2065 }
2066 
2067 /// Insert the new line info sequence \p Seq into the current
2068 /// set of already linked line info \p Rows.
insertLineSequence(std::vector<DWARFDebugLine::Row> & Seq,std::vector<DWARFDebugLine::Row> & Rows)2069 static void insertLineSequence(std::vector<DWARFDebugLine::Row> &Seq,
2070                                std::vector<DWARFDebugLine::Row> &Rows) {
2071   if (Seq.empty())
2072     return;
2073 
2074   if (!Rows.empty() && Rows.back().Address < Seq.front().Address) {
2075     llvm::append_range(Rows, Seq);
2076     Seq.clear();
2077     return;
2078   }
2079 
2080   object::SectionedAddress Front = Seq.front().Address;
2081   auto InsertPoint = partition_point(
2082       Rows, [=](const DWARFDebugLine::Row &O) { return O.Address < Front; });
2083 
2084   // FIXME: this only removes the unneeded end_sequence if the
2085   // sequences have been inserted in order. Using a global sort like
2086   // described in generateLineTableForUnit() and delaying the end_sequene
2087   // elimination to emitLineTableForUnit() we can get rid of all of them.
2088   if (InsertPoint != Rows.end() && InsertPoint->Address == Front &&
2089       InsertPoint->EndSequence) {
2090     *InsertPoint = Seq.front();
2091     Rows.insert(InsertPoint + 1, Seq.begin() + 1, Seq.end());
2092   } else {
2093     Rows.insert(InsertPoint, Seq.begin(), Seq.end());
2094   }
2095 
2096   Seq.clear();
2097 }
2098 
patchStmtList(DIE & Die,DIEInteger Offset)2099 static void patchStmtList(DIE &Die, DIEInteger Offset) {
2100   for (auto &V : Die.values())
2101     if (V.getAttribute() == dwarf::DW_AT_stmt_list) {
2102       V = DIEValue(V.getAttribute(), V.getForm(), Offset);
2103       return;
2104     }
2105 
2106   llvm_unreachable("Didn't find DW_AT_stmt_list in cloned DIE!");
2107 }
2108 
rememberUnitForMacroOffset(CompileUnit & Unit)2109 void DWARFLinker::DIECloner::rememberUnitForMacroOffset(CompileUnit &Unit) {
2110   DWARFUnit &OrigUnit = Unit.getOrigUnit();
2111   DWARFDie OrigUnitDie = OrigUnit.getUnitDIE();
2112 
2113   if (std::optional<uint64_t> MacroAttr =
2114           dwarf::toSectionOffset(OrigUnitDie.find(dwarf::DW_AT_macros))) {
2115     UnitMacroMap.insert(std::make_pair(*MacroAttr, &Unit));
2116     return;
2117   }
2118 
2119   if (std::optional<uint64_t> MacroAttr =
2120           dwarf::toSectionOffset(OrigUnitDie.find(dwarf::DW_AT_macro_info))) {
2121     UnitMacroMap.insert(std::make_pair(*MacroAttr, &Unit));
2122     return;
2123   }
2124 }
2125 
generateLineTableForUnit(CompileUnit & Unit)2126 void DWARFLinker::DIECloner::generateLineTableForUnit(CompileUnit &Unit) {
2127   if (LLVM_UNLIKELY(Emitter == nullptr))
2128     return;
2129 
2130   // Check whether DW_AT_stmt_list attribute is presented.
2131   DWARFDie CUDie = Unit.getOrigUnit().getUnitDIE();
2132   auto StmtList = dwarf::toSectionOffset(CUDie.find(dwarf::DW_AT_stmt_list));
2133   if (!StmtList)
2134     return;
2135 
2136   // Update the cloned DW_AT_stmt_list with the correct debug_line offset.
2137   if (auto *OutputDIE = Unit.getOutputUnitDIE())
2138     patchStmtList(*OutputDIE, DIEInteger(Emitter->getLineSectionSize()));
2139 
2140   if (const DWARFDebugLine::LineTable *LT =
2141           ObjFile.Dwarf->getLineTableForUnit(&Unit.getOrigUnit())) {
2142 
2143     DWARFDebugLine::LineTable LineTable;
2144 
2145     // Set Line Table header.
2146     LineTable.Prologue = LT->Prologue;
2147 
2148     // Set Line Table Rows.
2149     if (Linker.Options.Update) {
2150       LineTable.Rows = LT->Rows;
2151       // If all the line table contains is a DW_LNE_end_sequence, clear the line
2152       // table rows, it will be inserted again in the DWARFStreamer.
2153       if (LineTable.Rows.size() == 1 && LineTable.Rows[0].EndSequence)
2154         LineTable.Rows.clear();
2155 
2156       LineTable.Sequences = LT->Sequences;
2157     } else {
2158       // This vector is the output line table.
2159       std::vector<DWARFDebugLine::Row> NewRows;
2160       NewRows.reserve(LT->Rows.size());
2161 
2162       // Current sequence of rows being extracted, before being inserted
2163       // in NewRows.
2164       std::vector<DWARFDebugLine::Row> Seq;
2165 
2166       const auto &FunctionRanges = Unit.getFunctionRanges();
2167       std::optional<AddressRangeValuePair> CurrRange;
2168 
2169       // FIXME: This logic is meant to generate exactly the same output as
2170       // Darwin's classic dsymutil. There is a nicer way to implement this
2171       // by simply putting all the relocated line info in NewRows and simply
2172       // sorting NewRows before passing it to emitLineTableForUnit. This
2173       // should be correct as sequences for a function should stay
2174       // together in the sorted output. There are a few corner cases that
2175       // look suspicious though, and that required to implement the logic
2176       // this way. Revisit that once initial validation is finished.
2177 
2178       // Iterate over the object file line info and extract the sequences
2179       // that correspond to linked functions.
2180       for (DWARFDebugLine::Row Row : LT->Rows) {
2181         // Check whether we stepped out of the range. The range is
2182         // half-open, but consider accept the end address of the range if
2183         // it is marked as end_sequence in the input (because in that
2184         // case, the relocation offset is accurate and that entry won't
2185         // serve as the start of another function).
2186         if (!CurrRange || !CurrRange->Range.contains(Row.Address.Address)) {
2187           // We just stepped out of a known range. Insert a end_sequence
2188           // corresponding to the end of the range.
2189           uint64_t StopAddress =
2190               CurrRange ? CurrRange->Range.end() + CurrRange->Value : -1ULL;
2191           CurrRange = FunctionRanges.getRangeThatContains(Row.Address.Address);
2192           if (StopAddress != -1ULL && !Seq.empty()) {
2193             // Insert end sequence row with the computed end address, but
2194             // the same line as the previous one.
2195             auto NextLine = Seq.back();
2196             NextLine.Address.Address = StopAddress;
2197             NextLine.EndSequence = 1;
2198             NextLine.PrologueEnd = 0;
2199             NextLine.BasicBlock = 0;
2200             NextLine.EpilogueBegin = 0;
2201             Seq.push_back(NextLine);
2202             insertLineSequence(Seq, NewRows);
2203           }
2204 
2205           if (!CurrRange)
2206             continue;
2207         }
2208 
2209         // Ignore empty sequences.
2210         if (Row.EndSequence && Seq.empty())
2211           continue;
2212 
2213         // Relocate row address and add it to the current sequence.
2214         Row.Address.Address += CurrRange->Value;
2215         Seq.emplace_back(Row);
2216 
2217         if (Row.EndSequence)
2218           insertLineSequence(Seq, NewRows);
2219       }
2220 
2221       LineTable.Rows = std::move(NewRows);
2222     }
2223 
2224     Emitter->emitLineTableForUnit(LineTable, Unit, DebugStrPool,
2225                                   DebugLineStrPool);
2226   } else
2227     Linker.reportWarning("Cann't load line table.", ObjFile);
2228 }
2229 
emitAcceleratorEntriesForUnit(CompileUnit & Unit)2230 void DWARFLinker::emitAcceleratorEntriesForUnit(CompileUnit &Unit) {
2231   for (AccelTableKind AccelTableKind : Options.AccelTables) {
2232     switch (AccelTableKind) {
2233     case AccelTableKind::Apple: {
2234       // Add namespaces.
2235       for (const auto &Namespace : Unit.getNamespaces())
2236         AppleNamespaces.addName(Namespace.Name, Namespace.Die->getOffset() +
2237                                                     Unit.getStartOffset());
2238       // Add names.
2239       for (const auto &Pubname : Unit.getPubnames())
2240         AppleNames.addName(Pubname.Name,
2241                            Pubname.Die->getOffset() + Unit.getStartOffset());
2242       // Add types.
2243       for (const auto &Pubtype : Unit.getPubtypes())
2244         AppleTypes.addName(
2245             Pubtype.Name, Pubtype.Die->getOffset() + Unit.getStartOffset(),
2246             Pubtype.Die->getTag(),
2247             Pubtype.ObjcClassImplementation ? dwarf::DW_FLAG_type_implementation
2248                                             : 0,
2249             Pubtype.QualifiedNameHash);
2250       // Add ObjC names.
2251       for (const auto &ObjC : Unit.getObjC())
2252         AppleObjc.addName(ObjC.Name,
2253                           ObjC.Die->getOffset() + Unit.getStartOffset());
2254     } break;
2255     case AccelTableKind::Pub: {
2256       TheDwarfEmitter->emitPubNamesForUnit(Unit);
2257       TheDwarfEmitter->emitPubTypesForUnit(Unit);
2258     } break;
2259     case AccelTableKind::DebugNames: {
2260       for (const auto &Namespace : Unit.getNamespaces())
2261         DebugNames.addName(
2262             Namespace.Name, Namespace.Die->getOffset(),
2263             DWARF5AccelTableData::getDefiningParentDieOffset(*Namespace.Die),
2264             Namespace.Die->getTag(), Unit.getUniqueID(),
2265             Unit.getTag() == dwarf::DW_TAG_type_unit);
2266       for (const auto &Pubname : Unit.getPubnames())
2267         DebugNames.addName(
2268             Pubname.Name, Pubname.Die->getOffset(),
2269             DWARF5AccelTableData::getDefiningParentDieOffset(*Pubname.Die),
2270             Pubname.Die->getTag(), Unit.getUniqueID(),
2271             Unit.getTag() == dwarf::DW_TAG_type_unit);
2272       for (const auto &Pubtype : Unit.getPubtypes())
2273         DebugNames.addName(
2274             Pubtype.Name, Pubtype.Die->getOffset(),
2275             DWARF5AccelTableData::getDefiningParentDieOffset(*Pubtype.Die),
2276             Pubtype.Die->getTag(), Unit.getUniqueID(),
2277             Unit.getTag() == dwarf::DW_TAG_type_unit);
2278     } break;
2279     }
2280   }
2281 }
2282 
2283 /// Read the frame info stored in the object, and emit the
2284 /// patched frame descriptions for the resulting file.
2285 ///
2286 /// This is actually pretty easy as the data of the CIEs and FDEs can
2287 /// be considered as black boxes and moved as is. The only thing to do
2288 /// is to patch the addresses in the headers.
patchFrameInfoForObject(LinkContext & Context)2289 void DWARFLinker::patchFrameInfoForObject(LinkContext &Context) {
2290   DWARFContext &OrigDwarf = *Context.File.Dwarf;
2291   unsigned SrcAddrSize = OrigDwarf.getDWARFObj().getAddressSize();
2292 
2293   StringRef FrameData = OrigDwarf.getDWARFObj().getFrameSection().Data;
2294   if (FrameData.empty())
2295     return;
2296 
2297   RangesTy AllUnitsRanges;
2298   for (std::unique_ptr<CompileUnit> &Unit : Context.CompileUnits) {
2299     for (auto CurRange : Unit->getFunctionRanges())
2300       AllUnitsRanges.insert(CurRange.Range, CurRange.Value);
2301   }
2302 
2303   DataExtractor Data(FrameData, OrigDwarf.isLittleEndian(), 0);
2304   uint64_t InputOffset = 0;
2305 
2306   // Store the data of the CIEs defined in this object, keyed by their
2307   // offsets.
2308   DenseMap<uint64_t, StringRef> LocalCIES;
2309 
2310   while (Data.isValidOffset(InputOffset)) {
2311     uint64_t EntryOffset = InputOffset;
2312     uint32_t InitialLength = Data.getU32(&InputOffset);
2313     if (InitialLength == 0xFFFFFFFF)
2314       return reportWarning("Dwarf64 bits no supported", Context.File);
2315 
2316     uint32_t CIEId = Data.getU32(&InputOffset);
2317     if (CIEId == 0xFFFFFFFF) {
2318       // This is a CIE, store it.
2319       StringRef CIEData = FrameData.substr(EntryOffset, InitialLength + 4);
2320       LocalCIES[EntryOffset] = CIEData;
2321       // The -4 is to account for the CIEId we just read.
2322       InputOffset += InitialLength - 4;
2323       continue;
2324     }
2325 
2326     uint64_t Loc = Data.getUnsigned(&InputOffset, SrcAddrSize);
2327 
2328     // Some compilers seem to emit frame info that doesn't start at
2329     // the function entry point, thus we can't just lookup the address
2330     // in the debug map. Use the AddressInfo's range map to see if the FDE
2331     // describes something that we can relocate.
2332     std::optional<AddressRangeValuePair> Range =
2333         AllUnitsRanges.getRangeThatContains(Loc);
2334     if (!Range) {
2335       // The +4 is to account for the size of the InitialLength field itself.
2336       InputOffset = EntryOffset + InitialLength + 4;
2337       continue;
2338     }
2339 
2340     // This is an FDE, and we have a mapping.
2341     // Have we already emitted a corresponding CIE?
2342     StringRef CIEData = LocalCIES[CIEId];
2343     if (CIEData.empty())
2344       return reportWarning("Inconsistent debug_frame content. Dropping.",
2345                            Context.File);
2346 
2347     // Look if we already emitted a CIE that corresponds to the
2348     // referenced one (the CIE data is the key of that lookup).
2349     auto IteratorInserted = EmittedCIEs.insert(
2350         std::make_pair(CIEData, TheDwarfEmitter->getFrameSectionSize()));
2351     // If there is no CIE yet for this ID, emit it.
2352     if (IteratorInserted.second) {
2353       LastCIEOffset = TheDwarfEmitter->getFrameSectionSize();
2354       IteratorInserted.first->getValue() = LastCIEOffset;
2355       TheDwarfEmitter->emitCIE(CIEData);
2356     }
2357 
2358     // Emit the FDE with updated address and CIE pointer.
2359     // (4 + AddrSize) is the size of the CIEId + initial_location
2360     // fields that will get reconstructed by emitFDE().
2361     unsigned FDERemainingBytes = InitialLength - (4 + SrcAddrSize);
2362     TheDwarfEmitter->emitFDE(IteratorInserted.first->getValue(), SrcAddrSize,
2363                              Loc + Range->Value,
2364                              FrameData.substr(InputOffset, FDERemainingBytes));
2365     InputOffset += FDERemainingBytes;
2366   }
2367 }
2368 
hashFullyQualifiedName(DWARFDie DIE,CompileUnit & U,const DWARFFile & File,int ChildRecurseDepth)2369 uint32_t DWARFLinker::DIECloner::hashFullyQualifiedName(DWARFDie DIE,
2370                                                         CompileUnit &U,
2371                                                         const DWARFFile &File,
2372                                                         int ChildRecurseDepth) {
2373   const char *Name = nullptr;
2374   DWARFUnit *OrigUnit = &U.getOrigUnit();
2375   CompileUnit *CU = &U;
2376   std::optional<DWARFFormValue> Ref;
2377 
2378   while (true) {
2379     if (const char *CurrentName = DIE.getName(DINameKind::ShortName))
2380       Name = CurrentName;
2381 
2382     if (!(Ref = DIE.find(dwarf::DW_AT_specification)) &&
2383         !(Ref = DIE.find(dwarf::DW_AT_abstract_origin)))
2384       break;
2385 
2386     if (!Ref->isFormClass(DWARFFormValue::FC_Reference))
2387       break;
2388 
2389     CompileUnit *RefCU;
2390     if (auto RefDIE =
2391             Linker.resolveDIEReference(File, CompileUnits, *Ref, DIE, RefCU)) {
2392       CU = RefCU;
2393       OrigUnit = &RefCU->getOrigUnit();
2394       DIE = RefDIE;
2395     }
2396   }
2397 
2398   unsigned Idx = OrigUnit->getDIEIndex(DIE);
2399   if (!Name && DIE.getTag() == dwarf::DW_TAG_namespace)
2400     Name = "(anonymous namespace)";
2401 
2402   if (CU->getInfo(Idx).ParentIdx == 0 ||
2403       // FIXME: dsymutil-classic compatibility. Ignore modules.
2404       CU->getOrigUnit().getDIEAtIndex(CU->getInfo(Idx).ParentIdx).getTag() ==
2405           dwarf::DW_TAG_module)
2406     return djbHash(Name ? Name : "", djbHash(ChildRecurseDepth ? "" : "::"));
2407 
2408   DWARFDie Die = OrigUnit->getDIEAtIndex(CU->getInfo(Idx).ParentIdx);
2409   return djbHash(
2410       (Name ? Name : ""),
2411       djbHash((Name ? "::" : ""),
2412               hashFullyQualifiedName(Die, *CU, File, ++ChildRecurseDepth)));
2413 }
2414 
getDwoId(const DWARFDie & CUDie)2415 static uint64_t getDwoId(const DWARFDie &CUDie) {
2416   auto DwoId = dwarf::toUnsigned(
2417       CUDie.find({dwarf::DW_AT_dwo_id, dwarf::DW_AT_GNU_dwo_id}));
2418   if (DwoId)
2419     return *DwoId;
2420   return 0;
2421 }
2422 
2423 static std::string
remapPath(StringRef Path,const DWARFLinkerBase::ObjectPrefixMapTy & ObjectPrefixMap)2424 remapPath(StringRef Path,
2425           const DWARFLinkerBase::ObjectPrefixMapTy &ObjectPrefixMap) {
2426   if (ObjectPrefixMap.empty())
2427     return Path.str();
2428 
2429   SmallString<256> p = Path;
2430   for (const auto &Entry : ObjectPrefixMap)
2431     if (llvm::sys::path::replace_path_prefix(p, Entry.first, Entry.second))
2432       break;
2433   return p.str().str();
2434 }
2435 
2436 static std::string
getPCMFile(const DWARFDie & CUDie,const DWARFLinkerBase::ObjectPrefixMapTy * ObjectPrefixMap)2437 getPCMFile(const DWARFDie &CUDie,
2438            const DWARFLinkerBase::ObjectPrefixMapTy *ObjectPrefixMap) {
2439   std::string PCMFile = dwarf::toString(
2440       CUDie.find({dwarf::DW_AT_dwo_name, dwarf::DW_AT_GNU_dwo_name}), "");
2441 
2442   if (PCMFile.empty())
2443     return PCMFile;
2444 
2445   if (ObjectPrefixMap)
2446     PCMFile = remapPath(PCMFile, *ObjectPrefixMap);
2447 
2448   return PCMFile;
2449 }
2450 
isClangModuleRef(const DWARFDie & CUDie,std::string & PCMFile,LinkContext & Context,unsigned Indent,bool Quiet)2451 std::pair<bool, bool> DWARFLinker::isClangModuleRef(const DWARFDie &CUDie,
2452                                                     std::string &PCMFile,
2453                                                     LinkContext &Context,
2454                                                     unsigned Indent,
2455                                                     bool Quiet) {
2456   if (PCMFile.empty())
2457     return std::make_pair(false, false);
2458 
2459   // Clang module DWARF skeleton CUs abuse this for the path to the module.
2460   uint64_t DwoId = getDwoId(CUDie);
2461 
2462   std::string Name = dwarf::toString(CUDie.find(dwarf::DW_AT_name), "");
2463   if (Name.empty()) {
2464     if (!Quiet)
2465       reportWarning("Anonymous module skeleton CU for " + PCMFile,
2466                     Context.File);
2467     return std::make_pair(true, true);
2468   }
2469 
2470   if (!Quiet && Options.Verbose) {
2471     outs().indent(Indent);
2472     outs() << "Found clang module reference " << PCMFile;
2473   }
2474 
2475   auto Cached = ClangModules.find(PCMFile);
2476   if (Cached != ClangModules.end()) {
2477     // FIXME: Until PR27449 (https://llvm.org/bugs/show_bug.cgi?id=27449) is
2478     // fixed in clang, only warn about DWO_id mismatches in verbose mode.
2479     // ASTFileSignatures will change randomly when a module is rebuilt.
2480     if (!Quiet && Options.Verbose && (Cached->second != DwoId))
2481       reportWarning(Twine("hash mismatch: this object file was built against a "
2482                           "different version of the module ") +
2483                         PCMFile,
2484                     Context.File);
2485     if (!Quiet && Options.Verbose)
2486       outs() << " [cached].\n";
2487     return std::make_pair(true, true);
2488   }
2489 
2490   return std::make_pair(true, false);
2491 }
2492 
registerModuleReference(const DWARFDie & CUDie,LinkContext & Context,ObjFileLoaderTy Loader,CompileUnitHandlerTy OnCUDieLoaded,unsigned Indent)2493 bool DWARFLinker::registerModuleReference(const DWARFDie &CUDie,
2494                                           LinkContext &Context,
2495                                           ObjFileLoaderTy Loader,
2496                                           CompileUnitHandlerTy OnCUDieLoaded,
2497                                           unsigned Indent) {
2498   std::string PCMFile = getPCMFile(CUDie, Options.ObjectPrefixMap);
2499   std::pair<bool, bool> IsClangModuleRef =
2500       isClangModuleRef(CUDie, PCMFile, Context, Indent, false);
2501 
2502   if (!IsClangModuleRef.first)
2503     return false;
2504 
2505   if (IsClangModuleRef.second)
2506     return true;
2507 
2508   if (Options.Verbose)
2509     outs() << " ...\n";
2510 
2511   // Cyclic dependencies are disallowed by Clang, but we still
2512   // shouldn't run into an infinite loop, so mark it as processed now.
2513   ClangModules.insert({PCMFile, getDwoId(CUDie)});
2514 
2515   if (Error E = loadClangModule(Loader, CUDie, PCMFile, Context, OnCUDieLoaded,
2516                                 Indent + 2)) {
2517     consumeError(std::move(E));
2518     return false;
2519   }
2520   return true;
2521 }
2522 
loadClangModule(ObjFileLoaderTy Loader,const DWARFDie & CUDie,const std::string & PCMFile,LinkContext & Context,CompileUnitHandlerTy OnCUDieLoaded,unsigned Indent)2523 Error DWARFLinker::loadClangModule(
2524     ObjFileLoaderTy Loader, const DWARFDie &CUDie, const std::string &PCMFile,
2525     LinkContext &Context, CompileUnitHandlerTy OnCUDieLoaded, unsigned Indent) {
2526 
2527   uint64_t DwoId = getDwoId(CUDie);
2528   std::string ModuleName = dwarf::toString(CUDie.find(dwarf::DW_AT_name), "");
2529 
2530   /// Using a SmallString<0> because loadClangModule() is recursive.
2531   SmallString<0> Path(Options.PrependPath);
2532   if (sys::path::is_relative(PCMFile))
2533     resolveRelativeObjectPath(Path, CUDie);
2534   sys::path::append(Path, PCMFile);
2535   // Don't use the cached binary holder because we have no thread-safety
2536   // guarantee and the lifetime is limited.
2537 
2538   if (Loader == nullptr) {
2539     reportError("Could not load clang module: loader is not specified.\n",
2540                 Context.File);
2541     return Error::success();
2542   }
2543 
2544   auto ErrOrObj = Loader(Context.File.FileName, Path);
2545   if (!ErrOrObj)
2546     return Error::success();
2547 
2548   std::unique_ptr<CompileUnit> Unit;
2549   for (const auto &CU : ErrOrObj->Dwarf->compile_units()) {
2550     OnCUDieLoaded(*CU);
2551     // Recursively get all modules imported by this one.
2552     auto ChildCUDie = CU->getUnitDIE();
2553     if (!ChildCUDie)
2554       continue;
2555     if (!registerModuleReference(ChildCUDie, Context, Loader, OnCUDieLoaded,
2556                                  Indent)) {
2557       if (Unit) {
2558         std::string Err =
2559             (PCMFile +
2560              ": Clang modules are expected to have exactly 1 compile unit.\n");
2561         reportError(Err, Context.File);
2562         return make_error<StringError>(Err, inconvertibleErrorCode());
2563       }
2564       // FIXME: Until PR27449 (https://llvm.org/bugs/show_bug.cgi?id=27449) is
2565       // fixed in clang, only warn about DWO_id mismatches in verbose mode.
2566       // ASTFileSignatures will change randomly when a module is rebuilt.
2567       uint64_t PCMDwoId = getDwoId(ChildCUDie);
2568       if (PCMDwoId != DwoId) {
2569         if (Options.Verbose)
2570           reportWarning(
2571               Twine("hash mismatch: this object file was built against a "
2572                     "different version of the module ") +
2573                   PCMFile,
2574               Context.File);
2575         // Update the cache entry with the DwoId of the module loaded from disk.
2576         ClangModules[PCMFile] = PCMDwoId;
2577       }
2578 
2579       // Add this module.
2580       Unit = std::make_unique<CompileUnit>(*CU, UniqueUnitID++, !Options.NoODR,
2581                                            ModuleName);
2582     }
2583   }
2584 
2585   if (Unit)
2586     Context.ModuleUnits.emplace_back(RefModuleUnit{*ErrOrObj, std::move(Unit)});
2587 
2588   return Error::success();
2589 }
2590 
cloneAllCompileUnits(DWARFContext & DwarfContext,const DWARFFile & File,bool IsLittleEndian)2591 uint64_t DWARFLinker::DIECloner::cloneAllCompileUnits(
2592     DWARFContext &DwarfContext, const DWARFFile &File, bool IsLittleEndian) {
2593   uint64_t OutputDebugInfoSize =
2594       (Emitter == nullptr) ? 0 : Emitter->getDebugInfoSectionSize();
2595   const uint64_t StartOutputDebugInfoSize = OutputDebugInfoSize;
2596 
2597   for (auto &CurrentUnit : CompileUnits) {
2598     const uint16_t DwarfVersion = CurrentUnit->getOrigUnit().getVersion();
2599     const uint32_t UnitHeaderSize = DwarfVersion >= 5 ? 12 : 11;
2600     auto InputDIE = CurrentUnit->getOrigUnit().getUnitDIE();
2601     CurrentUnit->setStartOffset(OutputDebugInfoSize);
2602     if (!InputDIE) {
2603       OutputDebugInfoSize = CurrentUnit->computeNextUnitOffset(DwarfVersion);
2604       continue;
2605     }
2606     if (CurrentUnit->getInfo(0).Keep) {
2607       // Clone the InputDIE into your Unit DIE in our compile unit since it
2608       // already has a DIE inside of it.
2609       CurrentUnit->createOutputDIE();
2610       rememberUnitForMacroOffset(*CurrentUnit);
2611       cloneDIE(InputDIE, File, *CurrentUnit, 0 /* PC offset */, UnitHeaderSize,
2612                0, IsLittleEndian, CurrentUnit->getOutputUnitDIE());
2613     }
2614 
2615     OutputDebugInfoSize = CurrentUnit->computeNextUnitOffset(DwarfVersion);
2616 
2617     if (Emitter != nullptr) {
2618 
2619       generateLineTableForUnit(*CurrentUnit);
2620 
2621       Linker.emitAcceleratorEntriesForUnit(*CurrentUnit);
2622 
2623       if (LLVM_UNLIKELY(Linker.Options.Update))
2624         continue;
2625 
2626       Linker.generateUnitRanges(*CurrentUnit, File, AddrPool);
2627 
2628       auto ProcessExpr = [&](SmallVectorImpl<uint8_t> &SrcBytes,
2629                              SmallVectorImpl<uint8_t> &OutBytes,
2630                              int64_t RelocAdjustment) {
2631         DWARFUnit &OrigUnit = CurrentUnit->getOrigUnit();
2632         DataExtractor Data(SrcBytes, IsLittleEndian,
2633                            OrigUnit.getAddressByteSize());
2634         cloneExpression(Data,
2635                         DWARFExpression(Data, OrigUnit.getAddressByteSize(),
2636                                         OrigUnit.getFormParams().Format),
2637                         File, *CurrentUnit, OutBytes, RelocAdjustment,
2638                         IsLittleEndian);
2639       };
2640       generateUnitLocations(*CurrentUnit, File, ProcessExpr);
2641       emitDebugAddrSection(*CurrentUnit, DwarfVersion);
2642     }
2643     AddrPool.clear();
2644   }
2645 
2646   if (Emitter != nullptr) {
2647     assert(Emitter);
2648     // Emit macro tables.
2649     Emitter->emitMacroTables(File.Dwarf.get(), UnitMacroMap, DebugStrPool);
2650 
2651     // Emit all the compile unit's debug information.
2652     for (auto &CurrentUnit : CompileUnits) {
2653       CurrentUnit->fixupForwardReferences();
2654 
2655       if (!CurrentUnit->getOutputUnitDIE())
2656         continue;
2657 
2658       unsigned DwarfVersion = CurrentUnit->getOrigUnit().getVersion();
2659 
2660       assert(Emitter->getDebugInfoSectionSize() ==
2661              CurrentUnit->getStartOffset());
2662       Emitter->emitCompileUnitHeader(*CurrentUnit, DwarfVersion);
2663       Emitter->emitDIE(*CurrentUnit->getOutputUnitDIE());
2664       assert(Emitter->getDebugInfoSectionSize() ==
2665              CurrentUnit->computeNextUnitOffset(DwarfVersion));
2666     }
2667   }
2668 
2669   return OutputDebugInfoSize - StartOutputDebugInfoSize;
2670 }
2671 
copyInvariantDebugSection(DWARFContext & Dwarf)2672 void DWARFLinker::copyInvariantDebugSection(DWARFContext &Dwarf) {
2673   TheDwarfEmitter->emitSectionContents(Dwarf.getDWARFObj().getLocSection().Data,
2674                                        DebugSectionKind::DebugLoc);
2675   TheDwarfEmitter->emitSectionContents(
2676       Dwarf.getDWARFObj().getRangesSection().Data,
2677       DebugSectionKind::DebugRange);
2678   TheDwarfEmitter->emitSectionContents(
2679       Dwarf.getDWARFObj().getFrameSection().Data, DebugSectionKind::DebugFrame);
2680   TheDwarfEmitter->emitSectionContents(Dwarf.getDWARFObj().getArangesSection(),
2681                                        DebugSectionKind::DebugARanges);
2682   TheDwarfEmitter->emitSectionContents(
2683       Dwarf.getDWARFObj().getAddrSection().Data, DebugSectionKind::DebugAddr);
2684   TheDwarfEmitter->emitSectionContents(
2685       Dwarf.getDWARFObj().getRnglistsSection().Data,
2686       DebugSectionKind::DebugRngLists);
2687   TheDwarfEmitter->emitSectionContents(
2688       Dwarf.getDWARFObj().getLoclistsSection().Data,
2689       DebugSectionKind::DebugLocLists);
2690 }
2691 
addObjectFile(DWARFFile & File,ObjFileLoaderTy Loader,CompileUnitHandlerTy OnCUDieLoaded)2692 void DWARFLinker::addObjectFile(DWARFFile &File, ObjFileLoaderTy Loader,
2693                                 CompileUnitHandlerTy OnCUDieLoaded) {
2694   ObjectContexts.emplace_back(LinkContext(File));
2695 
2696   if (ObjectContexts.back().File.Dwarf) {
2697     for (const std::unique_ptr<DWARFUnit> &CU :
2698          ObjectContexts.back().File.Dwarf->compile_units()) {
2699       DWARFDie CUDie = CU->getUnitDIE();
2700 
2701       if (!CUDie)
2702         continue;
2703 
2704       OnCUDieLoaded(*CU);
2705 
2706       if (!LLVM_UNLIKELY(Options.Update))
2707         registerModuleReference(CUDie, ObjectContexts.back(), Loader,
2708                                 OnCUDieLoaded);
2709     }
2710   }
2711 }
2712 
link()2713 Error DWARFLinker::link() {
2714   assert((Options.TargetDWARFVersion != 0) &&
2715          "TargetDWARFVersion should be set");
2716 
2717   // First populate the data structure we need for each iteration of the
2718   // parallel loop.
2719   unsigned NumObjects = ObjectContexts.size();
2720 
2721   // This Dwarf string pool which is used for emission. It must be used
2722   // serially as the order of calling getStringOffset matters for
2723   // reproducibility.
2724   OffsetsStringPool DebugStrPool(true);
2725   OffsetsStringPool DebugLineStrPool(false);
2726   DebugDieValuePool StringOffsetPool;
2727 
2728   // ODR Contexts for the optimize.
2729   DeclContextTree ODRContexts;
2730 
2731   for (LinkContext &OptContext : ObjectContexts) {
2732     if (Options.Verbose)
2733       outs() << "DEBUG MAP OBJECT: " << OptContext.File.FileName << "\n";
2734 
2735     if (!OptContext.File.Dwarf)
2736       continue;
2737 
2738     if (Options.VerifyInputDWARF)
2739       verifyInput(OptContext.File);
2740 
2741     // Look for relocations that correspond to address map entries.
2742 
2743     // there was findvalidrelocations previously ... probably we need to gather
2744     // info here
2745     if (LLVM_LIKELY(!Options.Update) &&
2746         !OptContext.File.Addresses->hasValidRelocs()) {
2747       if (Options.Verbose)
2748         outs() << "No valid relocations found. Skipping.\n";
2749 
2750       // Set "Skip" flag as a signal to other loops that we should not
2751       // process this iteration.
2752       OptContext.Skip = true;
2753       continue;
2754     }
2755 
2756     // Setup access to the debug info.
2757     if (!OptContext.File.Dwarf)
2758       continue;
2759 
2760     // Check whether type units are presented.
2761     if (!OptContext.File.Dwarf->types_section_units().empty()) {
2762       reportWarning("type units are not currently supported: file will "
2763                     "be skipped",
2764                     OptContext.File);
2765       OptContext.Skip = true;
2766       continue;
2767     }
2768 
2769     // Clone all the clang modules with requires extracting the DIE units. We
2770     // don't need the full debug info until the Analyze phase.
2771     OptContext.CompileUnits.reserve(
2772         OptContext.File.Dwarf->getNumCompileUnits());
2773     for (const auto &CU : OptContext.File.Dwarf->compile_units()) {
2774       auto CUDie = CU->getUnitDIE(/*ExtractUnitDIEOnly=*/true);
2775       if (Options.Verbose) {
2776         outs() << "Input compilation unit:";
2777         DIDumpOptions DumpOpts;
2778         DumpOpts.ChildRecurseDepth = 0;
2779         DumpOpts.Verbose = Options.Verbose;
2780         CUDie.dump(outs(), 0, DumpOpts);
2781       }
2782     }
2783 
2784     for (auto &CU : OptContext.ModuleUnits) {
2785       if (Error Err = cloneModuleUnit(OptContext, CU, ODRContexts, DebugStrPool,
2786                                       DebugLineStrPool, StringOffsetPool))
2787         reportWarning(toString(std::move(Err)), CU.File);
2788     }
2789   }
2790 
2791   // At this point we know how much data we have emitted. We use this value to
2792   // compare canonical DIE offsets in analyzeContextInfo to see if a definition
2793   // is already emitted, without being affected by canonical die offsets set
2794   // later. This prevents undeterminism when analyze and clone execute
2795   // concurrently, as clone set the canonical DIE offset and analyze reads it.
2796   const uint64_t ModulesEndOffset =
2797       (TheDwarfEmitter == nullptr) ? 0
2798                                    : TheDwarfEmitter->getDebugInfoSectionSize();
2799 
2800   // These variables manage the list of processed object files.
2801   // The mutex and condition variable are to ensure that this is thread safe.
2802   std::mutex ProcessedFilesMutex;
2803   std::condition_variable ProcessedFilesConditionVariable;
2804   BitVector ProcessedFiles(NumObjects, false);
2805 
2806   //  Analyzing the context info is particularly expensive so it is executed in
2807   //  parallel with emitting the previous compile unit.
2808   auto AnalyzeLambda = [&](size_t I) {
2809     auto &Context = ObjectContexts[I];
2810 
2811     if (Context.Skip || !Context.File.Dwarf)
2812       return;
2813 
2814     for (const auto &CU : Context.File.Dwarf->compile_units()) {
2815       // Previously we only extracted the unit DIEs. We need the full debug info
2816       // now.
2817       auto CUDie = CU->getUnitDIE(/*ExtractUnitDIEOnly=*/false);
2818       std::string PCMFile = getPCMFile(CUDie, Options.ObjectPrefixMap);
2819 
2820       if (!CUDie || LLVM_UNLIKELY(Options.Update) ||
2821           !isClangModuleRef(CUDie, PCMFile, Context, 0, true).first) {
2822         Context.CompileUnits.push_back(std::make_unique<CompileUnit>(
2823             *CU, UniqueUnitID++, !Options.NoODR && !Options.Update, ""));
2824       }
2825     }
2826 
2827     // Now build the DIE parent links that we will use during the next phase.
2828     for (auto &CurrentUnit : Context.CompileUnits) {
2829       auto CUDie = CurrentUnit->getOrigUnit().getUnitDIE();
2830       if (!CUDie)
2831         continue;
2832       analyzeContextInfo(CurrentUnit->getOrigUnit().getUnitDIE(), 0,
2833                          *CurrentUnit, &ODRContexts.getRoot(), ODRContexts,
2834                          ModulesEndOffset, Options.ParseableSwiftInterfaces,
2835                          [&](const Twine &Warning, const DWARFDie &DIE) {
2836                            reportWarning(Warning, Context.File, &DIE);
2837                          });
2838     }
2839   };
2840 
2841   // For each object file map how many bytes were emitted.
2842   StringMap<DebugInfoSize> SizeByObject;
2843 
2844   // And then the remaining work in serial again.
2845   // Note, although this loop runs in serial, it can run in parallel with
2846   // the analyzeContextInfo loop so long as we process files with indices >=
2847   // than those processed by analyzeContextInfo.
2848   auto CloneLambda = [&](size_t I) {
2849     auto &OptContext = ObjectContexts[I];
2850     if (OptContext.Skip || !OptContext.File.Dwarf)
2851       return;
2852 
2853     // Then mark all the DIEs that need to be present in the generated output
2854     // and collect some information about them.
2855     // Note that this loop can not be merged with the previous one because
2856     // cross-cu references require the ParentIdx to be setup for every CU in
2857     // the object file before calling this.
2858     if (LLVM_UNLIKELY(Options.Update)) {
2859       for (auto &CurrentUnit : OptContext.CompileUnits)
2860         CurrentUnit->markEverythingAsKept();
2861       copyInvariantDebugSection(*OptContext.File.Dwarf);
2862     } else {
2863       for (auto &CurrentUnit : OptContext.CompileUnits) {
2864         lookForDIEsToKeep(*OptContext.File.Addresses, OptContext.CompileUnits,
2865                           CurrentUnit->getOrigUnit().getUnitDIE(),
2866                           OptContext.File, *CurrentUnit, 0);
2867 #ifndef NDEBUG
2868         verifyKeepChain(*CurrentUnit);
2869 #endif
2870       }
2871     }
2872 
2873     // The calls to applyValidRelocs inside cloneDIE will walk the reloc
2874     // array again (in the same way findValidRelocsInDebugInfo() did). We
2875     // need to reset the NextValidReloc index to the beginning.
2876     if (OptContext.File.Addresses->hasValidRelocs() ||
2877         LLVM_UNLIKELY(Options.Update)) {
2878       SizeByObject[OptContext.File.FileName].Input =
2879           getDebugInfoSize(*OptContext.File.Dwarf);
2880       SizeByObject[OptContext.File.FileName].Output =
2881           DIECloner(*this, TheDwarfEmitter, OptContext.File, DIEAlloc,
2882                     OptContext.CompileUnits, Options.Update, DebugStrPool,
2883                     DebugLineStrPool, StringOffsetPool)
2884               .cloneAllCompileUnits(*OptContext.File.Dwarf, OptContext.File,
2885                                     OptContext.File.Dwarf->isLittleEndian());
2886     }
2887     if ((TheDwarfEmitter != nullptr) && !OptContext.CompileUnits.empty() &&
2888         LLVM_LIKELY(!Options.Update))
2889       patchFrameInfoForObject(OptContext);
2890 
2891     // Clean-up before starting working on the next object.
2892     cleanupAuxiliarryData(OptContext);
2893   };
2894 
2895   auto EmitLambda = [&]() {
2896     // Emit everything that's global.
2897     if (TheDwarfEmitter != nullptr) {
2898       TheDwarfEmitter->emitAbbrevs(Abbreviations, Options.TargetDWARFVersion);
2899       TheDwarfEmitter->emitStrings(DebugStrPool);
2900       TheDwarfEmitter->emitStringOffsets(StringOffsetPool.getValues(),
2901                                          Options.TargetDWARFVersion);
2902       TheDwarfEmitter->emitLineStrings(DebugLineStrPool);
2903       for (AccelTableKind TableKind : Options.AccelTables) {
2904         switch (TableKind) {
2905         case AccelTableKind::Apple:
2906           TheDwarfEmitter->emitAppleNamespaces(AppleNamespaces);
2907           TheDwarfEmitter->emitAppleNames(AppleNames);
2908           TheDwarfEmitter->emitAppleTypes(AppleTypes);
2909           TheDwarfEmitter->emitAppleObjc(AppleObjc);
2910           break;
2911         case AccelTableKind::Pub:
2912           // Already emitted by emitAcceleratorEntriesForUnit.
2913           // Already emitted by emitAcceleratorEntriesForUnit.
2914           break;
2915         case AccelTableKind::DebugNames:
2916           TheDwarfEmitter->emitDebugNames(DebugNames);
2917           break;
2918         }
2919       }
2920     }
2921   };
2922 
2923   auto AnalyzeAll = [&]() {
2924     for (unsigned I = 0, E = NumObjects; I != E; ++I) {
2925       AnalyzeLambda(I);
2926 
2927       std::unique_lock<std::mutex> LockGuard(ProcessedFilesMutex);
2928       ProcessedFiles.set(I);
2929       ProcessedFilesConditionVariable.notify_one();
2930     }
2931   };
2932 
2933   auto CloneAll = [&]() {
2934     for (unsigned I = 0, E = NumObjects; I != E; ++I) {
2935       {
2936         std::unique_lock<std::mutex> LockGuard(ProcessedFilesMutex);
2937         if (!ProcessedFiles[I]) {
2938           ProcessedFilesConditionVariable.wait(
2939               LockGuard, [&]() { return ProcessedFiles[I]; });
2940         }
2941       }
2942 
2943       CloneLambda(I);
2944     }
2945     EmitLambda();
2946   };
2947 
2948   // To limit memory usage in the single threaded case, analyze and clone are
2949   // run sequentially so the OptContext is freed after processing each object
2950   // in endDebugObject.
2951   if (Options.Threads == 1) {
2952     for (unsigned I = 0, E = NumObjects; I != E; ++I) {
2953       AnalyzeLambda(I);
2954       CloneLambda(I);
2955     }
2956     EmitLambda();
2957   } else {
2958     DefaultThreadPool Pool(hardware_concurrency(2));
2959     Pool.async(AnalyzeAll);
2960     Pool.async(CloneAll);
2961     Pool.wait();
2962   }
2963 
2964   if (Options.Statistics) {
2965     // Create a vector sorted in descending order by output size.
2966     std::vector<std::pair<StringRef, DebugInfoSize>> Sorted;
2967     for (auto &E : SizeByObject)
2968       Sorted.emplace_back(E.first(), E.second);
2969     llvm::sort(Sorted, [](auto &LHS, auto &RHS) {
2970       return LHS.second.Output > RHS.second.Output;
2971     });
2972 
2973     auto ComputePercentange = [](int64_t Input, int64_t Output) -> float {
2974       const float Difference = Output - Input;
2975       const float Sum = Input + Output;
2976       if (Sum == 0)
2977         return 0;
2978       return (Difference / (Sum / 2));
2979     };
2980 
2981     int64_t InputTotal = 0;
2982     int64_t OutputTotal = 0;
2983     const char *FormatStr = "{0,-45} {1,10}b  {2,10}b {3,8:P}\n";
2984 
2985     // Print header.
2986     outs() << ".debug_info section size (in bytes)\n";
2987     outs() << "----------------------------------------------------------------"
2988               "---------------\n";
2989     outs() << "Filename                                           Object       "
2990               "  dSYM   Change\n";
2991     outs() << "----------------------------------------------------------------"
2992               "---------------\n";
2993 
2994     // Print body.
2995     for (auto &E : Sorted) {
2996       InputTotal += E.second.Input;
2997       OutputTotal += E.second.Output;
2998       llvm::outs() << formatv(
2999           FormatStr, sys::path::filename(E.first).take_back(45), E.second.Input,
3000           E.second.Output, ComputePercentange(E.second.Input, E.second.Output));
3001     }
3002     // Print total and footer.
3003     outs() << "----------------------------------------------------------------"
3004               "---------------\n";
3005     llvm::outs() << formatv(FormatStr, "Total", InputTotal, OutputTotal,
3006                             ComputePercentange(InputTotal, OutputTotal));
3007     outs() << "----------------------------------------------------------------"
3008               "---------------\n\n";
3009   }
3010 
3011   return Error::success();
3012 }
3013 
cloneModuleUnit(LinkContext & Context,RefModuleUnit & Unit,DeclContextTree & ODRContexts,OffsetsStringPool & DebugStrPool,OffsetsStringPool & DebugLineStrPool,DebugDieValuePool & StringOffsetPool,unsigned Indent)3014 Error DWARFLinker::cloneModuleUnit(LinkContext &Context, RefModuleUnit &Unit,
3015                                    DeclContextTree &ODRContexts,
3016                                    OffsetsStringPool &DebugStrPool,
3017                                    OffsetsStringPool &DebugLineStrPool,
3018                                    DebugDieValuePool &StringOffsetPool,
3019                                    unsigned Indent) {
3020   assert(Unit.Unit.get() != nullptr);
3021 
3022   if (!Unit.Unit->getOrigUnit().getUnitDIE().hasChildren())
3023     return Error::success();
3024 
3025   if (Options.Verbose) {
3026     outs().indent(Indent);
3027     outs() << "cloning .debug_info from " << Unit.File.FileName << "\n";
3028   }
3029 
3030   // Analyze context for the module.
3031   analyzeContextInfo(Unit.Unit->getOrigUnit().getUnitDIE(), 0, *(Unit.Unit),
3032                      &ODRContexts.getRoot(), ODRContexts, 0,
3033                      Options.ParseableSwiftInterfaces,
3034                      [&](const Twine &Warning, const DWARFDie &DIE) {
3035                        reportWarning(Warning, Context.File, &DIE);
3036                      });
3037   // Keep everything.
3038   Unit.Unit->markEverythingAsKept();
3039 
3040   // Clone unit.
3041   UnitListTy CompileUnits;
3042   CompileUnits.emplace_back(std::move(Unit.Unit));
3043   assert(TheDwarfEmitter);
3044   DIECloner(*this, TheDwarfEmitter, Unit.File, DIEAlloc, CompileUnits,
3045             Options.Update, DebugStrPool, DebugLineStrPool, StringOffsetPool)
3046       .cloneAllCompileUnits(*Unit.File.Dwarf, Unit.File,
3047                             Unit.File.Dwarf->isLittleEndian());
3048   return Error::success();
3049 }
3050 
verifyInput(const DWARFFile & File)3051 void DWARFLinker::verifyInput(const DWARFFile &File) {
3052   assert(File.Dwarf);
3053 
3054   std::string Buffer;
3055   raw_string_ostream OS(Buffer);
3056   DIDumpOptions DumpOpts;
3057   if (!File.Dwarf->verify(OS, DumpOpts.noImplicitRecursion())) {
3058     if (Options.InputVerificationHandler)
3059       Options.InputVerificationHandler(File, OS.str());
3060   }
3061 }
3062 
3063 } // namespace llvm
3064