[TableGen] Fix validateOperandClass for non Phyical Reg (#118146)
[llvm-project.git] / llvm / lib / DWARFLinker / Classic / DWARFLinker.cpp
blobd2b3561ee1c80284e65f6ddf04e941c41c4d9b58
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 //===----------------------------------------------------------------------===//
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>
41 namespace llvm {
43 using namespace dwarf_linker;
44 using namespace dwarf_linker::classic;
46 /// Hold the input and output of the debug info size in bytes.
47 struct DebugInfoSize {
48 uint64_t Input;
49 uint64_t Output;
52 /// Compute the total size of the debug info.
53 static uint64_t getDebugInfoSize(DWARFContext &Dwarf) {
54 uint64_t Size = 0;
55 for (auto &Unit : Dwarf.compile_units()) {
56 Size += Unit->getLength();
58 return Size;
61 /// Similar to DWARFUnitSection::getUnitForOffset(), but returning our
62 /// CompileUnit object instead.
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;
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.
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();
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;
97 reportWarning("could not find referenced DIE", File, &DIE);
98 return DWARFDie();
101 /// \returns whether the passed \a Attr type might contain a DIE reference
102 /// suitable for ODR uniquing.
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;
114 llvm_unreachable("Improper attribute.");
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;
150 return false;
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;
163 if (!Info.MangledName)
164 if (const char *MangledName = Die.getLinkageName())
165 Info.MangledName = StringPool.getEntry(MangledName);
167 if (!Info.Name)
168 if (const char *Name = Die.getShortName())
169 Info.Name = StringPool.getEntry(Name);
171 if (!Info.MangledName)
172 Info.MangledName = Info.Name;
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);
180 return Info.Name || Info.MangledName;
183 /// Resolve the relative path to a build artifact referenced by DWARF by
184 /// applying DW_AT_comp_dir.
185 static void resolveRelativeObjectPath(SmallVectorImpl<char> &Buf, DWARFDie CU) {
186 sys::path::append(Buf, dwarf::toString(CU.find(dwarf::DW_AT_comp_dir), ""));
189 /// Collect references to parseable Swift interfaces in imported
190 /// DW_TAG_module blocks.
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;
198 if (!ParseableSwiftInterfaces)
199 return;
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);
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,
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;
253 ContextWorklistItemType Type;
254 bool InImportedModule;
256 ContextWorklistItem(DWARFDie Die, ContextWorklistItemType T,
257 CompileUnit::DIEInfo *OtherInfo = nullptr)
258 : Die(Die), ParentIdx(0), OtherInfo(OtherInfo), Type(T),
259 InImportedModule(false) {}
261 ContextWorklistItem(DWARFDie Die, DeclContext *Context, unsigned ParentIdx,
262 bool InImportedModule)
263 : Die(Die), ParentIdx(ParentIdx), Context(Context),
264 Type(ContextWorklistItemType::AnalyzeContextInfo),
265 InImportedModule(InImportedModule) {}
268 static bool updatePruning(const DWARFDie &Die, CompileUnit &CU,
269 uint64_t ModulesEndOffset) {
270 CompileUnit::DIEInfo &Info = CU.getInfo(Die);
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));
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;
287 return Info.Prune;
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;
296 /// Recursive helper to build the global DeclContext information and
297 /// gather the child->parent relationships in the original compile unit.
299 /// This function uses the same work list approach as lookForDIEsToKeep.
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).
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);
314 while (!Worklist.empty()) {
315 ContextWorklistItem Current = Worklist.back();
316 Worklist.pop_back();
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;
329 unsigned Idx = CU.getOrigUnit().getDIEIndex(Current.Die);
330 CompileUnit::DIEInfo &Info = CU.getInfo(Idx);
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."
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)."
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);
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;
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);
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;
395 llvm_unreachable("Invalid Tag");
398 void DWARFLinker::cleanupAuxiliarryData(LinkContext &Context) {
399 Context.clear();
401 for (DIEBlock *I : DIEBlocks)
402 I->~DIEBlock();
403 for (DIELoc *I : DIELocs)
404 I->~DIELoc();
406 DIEBlocks.clear();
407 DIELocs.clear();
408 DIEAlloc.Reset();
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;
416 std::pair<bool, std::optional<int64_t>>
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");
423 const auto *Abbrev = DIE.getAbbreviationDeclarationPtr();
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);
432 // Get offset to the DW_AT_location attribute.
433 uint64_t AttrOffset =
434 Abbrev->getAttributeOffsetFromIndex(*LocationIdx, DIE.getOffset(), *U);
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);
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);
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);
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;
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);
496 } break;
497 default: {
498 // Nothing to do.
499 } break;
501 CurExprOffset = Op.getEndOffset();
504 return std::make_pair(HasLocationAddress, std::nullopt);
507 /// Check if a variable describing DIE should be kept.
508 /// \returns updated TraversalFlags.
509 unsigned DWARFLinker::shouldKeepVariableDIE(AddressesMap &RelocMgr,
510 const DWARFDie &DIE,
511 CompileUnit::DIEInfo &MyInfo,
512 unsigned Flags) {
513 const auto *Abbrev = DIE.getAbbreviationDeclarationPtr();
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;
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);
530 if (LocExprAddrAndRelocAdjustment.first)
531 MyInfo.HasLocationExpressionAddr = true;
533 if (!LocExprAddrAndRelocAdjustment.second)
534 return Flags;
536 MyInfo.AddrAdjust = *LocExprAddrAndRelocAdjustment.second;
537 MyInfo.InDebugMap = true;
539 if (((Flags & TF_InFunctionScope) &&
540 !LLVM_UNLIKELY(Options.KeepFunctionForStatic)))
541 return Flags;
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);
551 return Flags | TF_Keep;
554 /// Check if a function describing DIE should be kept.
555 /// \returns updated TraversalFlags.
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;
561 auto LowPc = dwarf::toAddress(DIE.find(dwarf::DW_AT_low_pc));
562 if (!LowPc)
563 return Flags;
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;
571 MyInfo.AddrAdjust = *RelocAdjustment;
572 MyInfo.InDebugMap = true;
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);
582 if (DIE.getTag() == dwarf::DW_TAG_label) {
583 if (Unit.hasLabelAt(*LowPc))
584 return Flags;
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;
598 Flags |= TF_Keep;
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;
606 if (*LowPc > *HighPc) {
607 reportWarning("low_pc greater than high_pc. Range will be discarded.\n",
608 File, &DIE);
609 return Flags;
612 // Replace the debug map range with a more accurate one.
613 Unit.addFunctionRange(*LowPc, *HighPc, MyInfo.AddrAdjust);
614 return Flags;
617 /// Check if a DIE should be kept.
618 /// \returns updated TraversalFlags.
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;
642 return Flags;
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.
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;
659 CompileUnit::DIEInfo &MyInfo = CU.getInfo(Die);
661 if (ChildInfo.Incomplete || ChildInfo.Prune)
662 MyInfo.Incomplete = true;
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.
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;
681 CompileUnit::DIEInfo &MyInfo = CU.getInfo(Die);
683 if (MyInfo.Incomplete)
684 return;
686 if (RefInfo.Incomplete)
687 MyInfo.Incomplete = true;
690 /// Look at the children of the given DIE and decide whether they should be
691 /// kept.
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;
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;
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);
720 static bool isODRCanonicalCandidate(const DWARFDie &Die, CompileUnit &CU) {
721 CompileUnit::DIEInfo &Info = CU.getInfo(Die);
723 if (!Info.Ctxt || (Die.getTag() == dwarf::DW_TAG_namespace))
724 return false;
726 if (!CU.hasODR() && !Info.InModuleScope)
727 return false;
729 return !Info.Incomplete && Info.Ctxt != CU.getInfo(Info.ParentIdx).Ctxt;
732 void DWARFLinker::markODRCanonicalDie(const DWARFDie &Die, CompileUnit &CU) {
733 CompileUnit::DIEInfo &Info = CU.getInfo(Die);
735 Info.ODRMarkingDone = true;
736 if (Info.Keep && isODRCanonicalCandidate(Die, CU) &&
737 !Info.Ctxt->hasCanonicalDIE())
738 Info.Ctxt->setHasCanonicalDIE();
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.
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());
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;
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.
774 // FIXME: compatibility with dsymutil-classic. UseODR shouldn't
775 // be necessary and could be advantageously replaced by
776 // ReferencedCU->hasODR() && CU.hasODR().
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;
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);
793 unsigned ODRFlag = UseOdr ? DWARFLinker::TF_ODR : 0;
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);
809 /// Look at the parent of the given DIE and decide whether they should be kept.
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;
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);
823 /// Recursively walk the \p DIE tree and look for DIEs to keep. Store that
824 /// information in \p CU's DIEInfo.
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).
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.
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.
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).
847 /// The return value indicates whether the DIE is incomplete.
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);
856 while (!Worklist.empty()) {
857 WorklistItem Current = Worklist.pop_back_val();
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;
885 unsigned Idx = Current.CU.getOrigUnit().getDIEIndex(Current.Die);
886 CompileUnit::DIEInfo &MyInfo = Current.CU.getInfo(Idx);
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;
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;
903 if (!(Current.Flags & TF_DependencyWalk))
904 Current.Flags = shouldKeepDIE(AddressesMap, Current.Die, File, Current.CU,
905 MyInfo, Current.Flags);
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);
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);
923 if (AlreadyKept || !(Current.Flags & TF_Keep))
924 continue;
926 // If it is a newly kept DIE mark it as well as all its dependencies as
927 // kept.
928 MyInfo.Keep = true;
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);
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);
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;
947 // Now schedule the parent walk.
948 Worklist.emplace_back(MyInfo.ParentIdx, Current.CU, ParFlags);
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 {
956 BrokenLink(DWARFDie Parent, DWARFDie Child) : Parent(Parent), Child(Child) {}
957 DWARFDie Parent;
958 DWARFDie Child;
961 /// Verify the keep chain by looking for DIEs that are kept but who's parent
962 /// isn't.
963 static void verifyKeepChain(CompileUnit &CU) {
964 std::vector<DWARFDie> Worklist;
965 Worklist.push_back(CU.getOrigUnit().getUnitDIE());
967 // List of broken links.
968 std::vector<BrokenLink> BrokenLinks;
970 while (!Worklist.empty()) {
971 const DWARFDie Current = Worklist.back();
972 Worklist.pop_back();
974 const bool CurrentDieIsKept = CU.getInfo(Current).Keep;
976 for (DWARFDie Child : reverse(Current.children())) {
977 Worklist.push_back(Child);
979 const bool ChildDieIsKept = CU.getInfo(Child).Keep;
980 if (!CurrentDieIsKept && ChildDieIsKept)
981 BrokenLinks.emplace_back(Current, Child);
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());
991 errs() << "Parent:";
992 Link.Parent.dump(errs(), 0, {});
993 CU.getInfo(Link.Parent).dump();
995 errs() << "Child:";
996 Link.Child.dump(errs(), 2, {});
997 CU.getInfo(Link.Child).dump();
999 report_fatal_error("invalid keep chain");
1002 #endif
1004 /// Assign an abbreviation number to \p Abbrev.
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.
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);
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());
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);
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);
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());
1071 // Switch everything to out of line strings.
1072 AttrSpec.Form = dwarf::DW_FORM_strp;
1074 Die.addValue(DIEAlloc, dwarf::Attribute(AttrSpec.Attr), AttrSpec.Form,
1075 DIEInteger(StringEntry.getOffset()));
1076 return 4;
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;
1092 DIE *NewRefDie = nullptr;
1093 CompileUnit *RefUnit = nullptr;
1095 DWARFDie RefDie =
1096 Linker.resolveDIEReference(File, CompileUnits, Val, InputDIE, RefUnit);
1098 // If the referenced DIE is not found, drop the attribute.
1099 if (!RefDie || AttrSpec.Attr == dwarf::DW_AT_sibling)
1100 return 0;
1102 CompileUnit::DIEInfo &RefInfo = RefUnit->getInfo(RefDie);
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();
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()));
1122 NewRefDie = RefInfo.Clone;
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)));
1147 return U.getRefAddrByteSize();
1150 Die.addValue(DIEAlloc, dwarf::Attribute(AttrSpec.Attr),
1151 dwarf::Form(AttrSpec.Form), DIEEntry(*NewRefDie));
1153 return AttrSize;
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;
1162 uint8_t OrigAddressByteSize = Unit.getOrigUnit().getAddressByteSize();
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);
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);
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);
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);
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);
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;
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());
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());
1274 OpOffset = Op.getEndOffset();
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);
1293 Attr = Loc ? static_cast<DIEValueList *>(Loc)
1294 : static_cast<DIEValueList *>(Block);
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;
1312 for (auto Byte : Bytes)
1313 Attr->addValue(DIEAlloc, static_cast<dwarf::Attribute>(0),
1314 dwarf::DW_FORM_data1, DIEInteger(Byte));
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());
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;
1337 Value = DIEValue(dwarf::Attribute(AttrSpec.Attr),
1338 dwarf::Form(AttrSpec.Form), Block);
1341 return Die.addValue(DIEAlloc, Value)->sizeOf(OrigUnit.getFormParams());
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;
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;
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.
1369 std::optional<DWARFFormValue> AddrAttribute = InputDIE.find(AttrSpec.Attr);
1370 if (!AddrAttribute)
1371 llvm_unreachable("Cann't find attribute.");
1373 std::optional<uint64_t> Addr = AddrAttribute->getAsAddress();
1374 if (!Addr) {
1375 Linker.reportWarning("Cann't read address attribute value.", ObjFile);
1376 return 0;
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;
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();
1401 auto AddrIndex = AddrPool.getValueIndex(*Addr);
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());
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;
1415 // We don't emit any skeleton CUs with dsymutil. So avoid emitting
1416 // a redundant DW_AT_GNU_dwo_id on the non-skeleton CU.
1417 if (AttrSpec.Attr == dwarf::DW_AT_GNU_dwo_id ||
1418 AttrSpec.Attr == dwarf::DW_AT_dwo_id)
1419 return 0;
1421 // Check for the offset to the macro table. If offset is incorrect then we
1422 // need to remove the attribute.
1423 if (AttrSpec.Attr == dwarf::DW_AT_macro_info) {
1424 if (std::optional<uint64_t> Offset = Val.getAsSectionOffset()) {
1425 const llvm::DWARFDebugMacro *Macro = File.Dwarf->getDebugMacinfo();
1426 if (Macro == nullptr || !Macro->hasEntryForOffset(*Offset))
1427 return 0;
1431 if (AttrSpec.Attr == dwarf::DW_AT_macros) {
1432 if (std::optional<uint64_t> Offset = Val.getAsSectionOffset()) {
1433 const llvm::DWARFDebugMacro *Macro = File.Dwarf->getDebugMacro();
1434 if (Macro == nullptr || !Macro->hasEntryForOffset(*Offset))
1435 return 0;
1439 if (AttrSpec.Attr == dwarf::DW_AT_str_offsets_base) {
1440 // DWARFLinker generates common .debug_str_offsets table used for all
1441 // compile units. The offset to the common .debug_str_offsets table is 8 on
1442 // DWARF32.
1443 Info.AttrStrOffsetBaseSeen = true;
1444 return Die
1445 .addValue(DIEAlloc, dwarf::DW_AT_str_offsets_base,
1446 dwarf::DW_FORM_sec_offset, DIEInteger(8))
1447 ->sizeOf(Unit.getOrigUnit().getFormParams());
1450 if (LLVM_UNLIKELY(Linker.Options.Update)) {
1451 if (auto OptionalValue = Val.getAsUnsignedConstant())
1452 Value = *OptionalValue;
1453 else if (auto OptionalValue = Val.getAsSignedConstant())
1454 Value = *OptionalValue;
1455 else if (auto OptionalValue = Val.getAsSectionOffset())
1456 Value = *OptionalValue;
1457 else {
1458 Linker.reportWarning(
1459 "Unsupported scalar attribute form. Dropping attribute.", File,
1460 &InputDIE);
1461 return 0;
1463 if (AttrSpec.Attr == dwarf::DW_AT_declaration && Value)
1464 Info.IsDeclaration = true;
1466 if (AttrSpec.Form == dwarf::DW_FORM_loclistx)
1467 Die.addValue(DIEAlloc, dwarf::Attribute(AttrSpec.Attr),
1468 dwarf::Form(AttrSpec.Form), DIELocList(Value));
1469 else
1470 Die.addValue(DIEAlloc, dwarf::Attribute(AttrSpec.Attr),
1471 dwarf::Form(AttrSpec.Form), DIEInteger(Value));
1472 return AttrSize;
1475 [[maybe_unused]] dwarf::Form OriginalForm = AttrSpec.Form;
1476 if (AttrSpec.Form == dwarf::DW_FORM_rnglistx) {
1477 // DWARFLinker does not generate .debug_addr table. Thus we need to change
1478 // all "addrx" related forms to "addr" version. Change DW_FORM_rnglistx
1479 // to DW_FORM_sec_offset here.
1480 std::optional<uint64_t> Index = Val.getAsSectionOffset();
1481 if (!Index) {
1482 Linker.reportWarning("Cannot read the attribute. Dropping.", File,
1483 &InputDIE);
1484 return 0;
1486 std::optional<uint64_t> Offset =
1487 Unit.getOrigUnit().getRnglistOffset(*Index);
1488 if (!Offset) {
1489 Linker.reportWarning("Cannot read the attribute. Dropping.", File,
1490 &InputDIE);
1491 return 0;
1494 Value = *Offset;
1495 AttrSpec.Form = dwarf::DW_FORM_sec_offset;
1496 AttrSize = Unit.getOrigUnit().getFormParams().getDwarfOffsetByteSize();
1497 } else if (AttrSpec.Form == dwarf::DW_FORM_loclistx) {
1498 // DWARFLinker does not generate .debug_addr table. Thus we need to change
1499 // all "addrx" related forms to "addr" version. Change DW_FORM_loclistx
1500 // to DW_FORM_sec_offset here.
1501 std::optional<uint64_t> Index = Val.getAsSectionOffset();
1502 if (!Index) {
1503 Linker.reportWarning("Cannot read the attribute. Dropping.", File,
1504 &InputDIE);
1505 return 0;
1507 std::optional<uint64_t> Offset =
1508 Unit.getOrigUnit().getLoclistOffset(*Index);
1509 if (!Offset) {
1510 Linker.reportWarning("Cannot read the attribute. Dropping.", File,
1511 &InputDIE);
1512 return 0;
1515 Value = *Offset;
1516 AttrSpec.Form = dwarf::DW_FORM_sec_offset;
1517 AttrSize = Unit.getOrigUnit().getFormParams().getDwarfOffsetByteSize();
1518 } else if (AttrSpec.Attr == dwarf::DW_AT_high_pc &&
1519 Die.getTag() == dwarf::DW_TAG_compile_unit) {
1520 std::optional<uint64_t> LowPC = Unit.getLowPc();
1521 if (!LowPC)
1522 return 0;
1523 // Dwarf >= 4 high_pc is an size, not an address.
1524 Value = Unit.getHighPc() - *LowPC;
1525 } else if (AttrSpec.Form == dwarf::DW_FORM_sec_offset)
1526 Value = *Val.getAsSectionOffset();
1527 else if (AttrSpec.Form == dwarf::DW_FORM_sdata)
1528 Value = *Val.getAsSignedConstant();
1529 else if (auto OptionalValue = Val.getAsUnsignedConstant())
1530 Value = *OptionalValue;
1531 else {
1532 Linker.reportWarning(
1533 "Unsupported scalar attribute form. Dropping attribute.", File,
1534 &InputDIE);
1535 return 0;
1538 DIE::value_iterator Patch =
1539 Die.addValue(DIEAlloc, dwarf::Attribute(AttrSpec.Attr),
1540 dwarf::Form(AttrSpec.Form), DIEInteger(Value));
1541 if (AttrSpec.Attr == dwarf::DW_AT_ranges ||
1542 AttrSpec.Attr == dwarf::DW_AT_start_scope) {
1543 Unit.noteRangeAttribute(Die, Patch);
1544 Info.HasRanges = true;
1545 } else if (DWARFAttribute::mayHaveLocationList(AttrSpec.Attr) &&
1546 dwarf::doesFormBelongToClass(AttrSpec.Form,
1547 DWARFFormValue::FC_SectionOffset,
1548 Unit.getOrigUnit().getVersion())) {
1550 CompileUnit::DIEInfo &LocationDieInfo = Unit.getInfo(InputDIE);
1551 Unit.noteLocationAttribute({Patch, LocationDieInfo.InDebugMap
1552 ? LocationDieInfo.AddrAdjust
1553 : Info.PCOffset});
1554 } else if (AttrSpec.Attr == dwarf::DW_AT_declaration && Value)
1555 Info.IsDeclaration = true;
1557 // check that all dwarf::DW_FORM_rnglistx are handled previously.
1558 assert((Info.HasRanges || (OriginalForm != dwarf::DW_FORM_rnglistx)) &&
1559 "Unhandled DW_FORM_rnglistx attribute");
1561 return AttrSize;
1564 /// Clone \p InputDIE's attribute described by \p AttrSpec with
1565 /// value \p Val, and add it to \p Die.
1566 /// \returns the size of the cloned attribute.
1567 unsigned DWARFLinker::DIECloner::cloneAttribute(
1568 DIE &Die, const DWARFDie &InputDIE, const DWARFFile &File,
1569 CompileUnit &Unit, const DWARFFormValue &Val, const AttributeSpec AttrSpec,
1570 unsigned AttrSize, AttributesInfo &Info, bool IsLittleEndian) {
1571 const DWARFUnit &U = Unit.getOrigUnit();
1573 switch (AttrSpec.Form) {
1574 case dwarf::DW_FORM_strp:
1575 case dwarf::DW_FORM_line_strp:
1576 case dwarf::DW_FORM_string:
1577 case dwarf::DW_FORM_strx:
1578 case dwarf::DW_FORM_strx1:
1579 case dwarf::DW_FORM_strx2:
1580 case dwarf::DW_FORM_strx3:
1581 case dwarf::DW_FORM_strx4:
1582 return cloneStringAttribute(Die, AttrSpec, Val, U, Info);
1583 case dwarf::DW_FORM_ref_addr:
1584 case dwarf::DW_FORM_ref1:
1585 case dwarf::DW_FORM_ref2:
1586 case dwarf::DW_FORM_ref4:
1587 case dwarf::DW_FORM_ref8:
1588 return cloneDieReferenceAttribute(Die, InputDIE, AttrSpec, AttrSize, Val,
1589 File, Unit);
1590 case dwarf::DW_FORM_block:
1591 case dwarf::DW_FORM_block1:
1592 case dwarf::DW_FORM_block2:
1593 case dwarf::DW_FORM_block4:
1594 case dwarf::DW_FORM_exprloc:
1595 return cloneBlockAttribute(Die, InputDIE, File, Unit, AttrSpec, Val,
1596 IsLittleEndian);
1597 case dwarf::DW_FORM_addr:
1598 case dwarf::DW_FORM_addrx:
1599 case dwarf::DW_FORM_addrx1:
1600 case dwarf::DW_FORM_addrx2:
1601 case dwarf::DW_FORM_addrx3:
1602 case dwarf::DW_FORM_addrx4:
1603 return cloneAddressAttribute(Die, InputDIE, AttrSpec, AttrSize, Val, Unit,
1604 Info);
1605 case dwarf::DW_FORM_data1:
1606 case dwarf::DW_FORM_data2:
1607 case dwarf::DW_FORM_data4:
1608 case dwarf::DW_FORM_data8:
1609 case dwarf::DW_FORM_udata:
1610 case dwarf::DW_FORM_sdata:
1611 case dwarf::DW_FORM_sec_offset:
1612 case dwarf::DW_FORM_flag:
1613 case dwarf::DW_FORM_flag_present:
1614 case dwarf::DW_FORM_rnglistx:
1615 case dwarf::DW_FORM_loclistx:
1616 case dwarf::DW_FORM_implicit_const:
1617 return cloneScalarAttribute(Die, InputDIE, File, Unit, AttrSpec, Val,
1618 AttrSize, Info);
1619 default:
1620 Linker.reportWarning("Unsupported attribute form " +
1621 dwarf::FormEncodingString(AttrSpec.Form) +
1622 " in cloneAttribute. Dropping.",
1623 File, &InputDIE);
1626 return 0;
1629 void DWARFLinker::DIECloner::addObjCAccelerator(CompileUnit &Unit,
1630 const DIE *Die,
1631 DwarfStringPoolEntryRef Name,
1632 OffsetsStringPool &StringPool,
1633 bool SkipPubSection) {
1634 std::optional<ObjCSelectorNames> Names =
1635 getObjCNamesIfSelector(Name.getString());
1636 if (!Names)
1637 return;
1638 Unit.addNameAccelerator(Die, StringPool.getEntry(Names->Selector),
1639 SkipPubSection);
1640 Unit.addObjCAccelerator(Die, StringPool.getEntry(Names->ClassName),
1641 SkipPubSection);
1642 if (Names->ClassNameNoCategory)
1643 Unit.addObjCAccelerator(
1644 Die, StringPool.getEntry(*Names->ClassNameNoCategory), SkipPubSection);
1645 if (Names->MethodNameNoCategory)
1646 Unit.addNameAccelerator(
1647 Die, StringPool.getEntry(*Names->MethodNameNoCategory), SkipPubSection);
1650 static bool
1651 shouldSkipAttribute(bool Update,
1652 DWARFAbbreviationDeclaration::AttributeSpec AttrSpec,
1653 bool SkipPC) {
1654 switch (AttrSpec.Attr) {
1655 default:
1656 return false;
1657 case dwarf::DW_AT_low_pc:
1658 case dwarf::DW_AT_high_pc:
1659 case dwarf::DW_AT_ranges:
1660 return !Update && SkipPC;
1661 case dwarf::DW_AT_rnglists_base:
1662 // In case !Update the .debug_addr table is not generated/preserved.
1663 // Thus instead of DW_FORM_rnglistx the DW_FORM_sec_offset is used.
1664 // Since DW_AT_rnglists_base is used for only DW_FORM_rnglistx the
1665 // DW_AT_rnglists_base is removed.
1666 return !Update;
1667 case dwarf::DW_AT_loclists_base:
1668 // In case !Update the .debug_addr table is not generated/preserved.
1669 // Thus instead of DW_FORM_loclistx the DW_FORM_sec_offset is used.
1670 // Since DW_AT_loclists_base is used for only DW_FORM_loclistx the
1671 // DW_AT_loclists_base is removed.
1672 return !Update;
1673 case dwarf::DW_AT_location:
1674 case dwarf::DW_AT_frame_base:
1675 return !Update && SkipPC;
1679 struct AttributeLinkedOffsetFixup {
1680 int64_t LinkedOffsetFixupVal;
1681 uint64_t InputAttrStartOffset;
1682 uint64_t InputAttrEndOffset;
1685 DIE *DWARFLinker::DIECloner::cloneDIE(const DWARFDie &InputDIE,
1686 const DWARFFile &File, CompileUnit &Unit,
1687 int64_t PCOffset, uint32_t OutOffset,
1688 unsigned Flags, bool IsLittleEndian,
1689 DIE *Die) {
1690 DWARFUnit &U = Unit.getOrigUnit();
1691 unsigned Idx = U.getDIEIndex(InputDIE);
1692 CompileUnit::DIEInfo &Info = Unit.getInfo(Idx);
1694 // Should the DIE appear in the output?
1695 if (!Unit.getInfo(Idx).Keep)
1696 return nullptr;
1698 uint64_t Offset = InputDIE.getOffset();
1699 assert(!(Die && Info.Clone) && "Can't supply a DIE and a cloned DIE");
1700 if (!Die) {
1701 // The DIE might have been already created by a forward reference
1702 // (see cloneDieReferenceAttribute()).
1703 if (!Info.Clone)
1704 Info.Clone = DIE::get(DIEAlloc, dwarf::Tag(InputDIE.getTag()));
1705 Die = Info.Clone;
1708 assert(Die->getTag() == InputDIE.getTag());
1709 Die->setOffset(OutOffset);
1710 if (isODRCanonicalCandidate(InputDIE, Unit) && Info.Ctxt &&
1711 (Info.Ctxt->getCanonicalDIEOffset() == 0)) {
1712 if (!Info.Ctxt->hasCanonicalDIE())
1713 Info.Ctxt->setHasCanonicalDIE();
1714 // We are about to emit a DIE that is the root of its own valid
1715 // DeclContext tree. Make the current offset the canonical offset
1716 // for this context.
1717 Info.Ctxt->setCanonicalDIEOffset(OutOffset + Unit.getStartOffset());
1720 // Extract and clone every attribute.
1721 DWARFDataExtractor Data = U.getDebugInfoExtractor();
1722 // Point to the next DIE (generally there is always at least a NULL
1723 // entry after the current one). If this is a lone
1724 // DW_TAG_compile_unit without any children, point to the next unit.
1725 uint64_t NextOffset = (Idx + 1 < U.getNumDIEs())
1726 ? U.getDIEAtIndex(Idx + 1).getOffset()
1727 : U.getNextUnitOffset();
1728 AttributesInfo AttrInfo;
1730 // We could copy the data only if we need to apply a relocation to it. After
1731 // testing, it seems there is no performance downside to doing the copy
1732 // unconditionally, and it makes the code simpler.
1733 SmallString<40> DIECopy(Data.getData().substr(Offset, NextOffset - Offset));
1734 Data =
1735 DWARFDataExtractor(DIECopy, Data.isLittleEndian(), Data.getAddressSize());
1737 // Modify the copy with relocated addresses.
1738 ObjFile.Addresses->applyValidRelocs(DIECopy, Offset, Data.isLittleEndian());
1740 // Reset the Offset to 0 as we will be working on the local copy of
1741 // the data.
1742 Offset = 0;
1744 const auto *Abbrev = InputDIE.getAbbreviationDeclarationPtr();
1745 Offset += getULEB128Size(Abbrev->getCode());
1747 // We are entering a subprogram. Get and propagate the PCOffset.
1748 if (Die->getTag() == dwarf::DW_TAG_subprogram)
1749 PCOffset = Info.AddrAdjust;
1750 AttrInfo.PCOffset = PCOffset;
1752 if (Abbrev->getTag() == dwarf::DW_TAG_subprogram) {
1753 Flags |= TF_InFunctionScope;
1754 if (!Info.InDebugMap && LLVM_LIKELY(!Update))
1755 Flags |= TF_SkipPC;
1756 } else if (Abbrev->getTag() == dwarf::DW_TAG_variable) {
1757 // Function-local globals could be in the debug map even when the function
1758 // is not, e.g., inlined functions.
1759 if ((Flags & TF_InFunctionScope) && Info.InDebugMap)
1760 Flags &= ~TF_SkipPC;
1761 // Location expressions referencing an address which is not in debug map
1762 // should be deleted.
1763 else if (!Info.InDebugMap && Info.HasLocationExpressionAddr &&
1764 LLVM_LIKELY(!Update))
1765 Flags |= TF_SkipPC;
1768 std::optional<StringRef> LibraryInstallName =
1769 ObjFile.Addresses->getLibraryInstallName();
1770 SmallVector<AttributeLinkedOffsetFixup> AttributesFixups;
1771 for (const auto &AttrSpec : Abbrev->attributes()) {
1772 if (shouldSkipAttribute(Update, AttrSpec, Flags & TF_SkipPC)) {
1773 DWARFFormValue::skipValue(AttrSpec.Form, Data, &Offset,
1774 U.getFormParams());
1775 continue;
1778 AttributeLinkedOffsetFixup CurAttrFixup;
1779 CurAttrFixup.InputAttrStartOffset = InputDIE.getOffset() + Offset;
1780 CurAttrFixup.LinkedOffsetFixupVal =
1781 Unit.getStartOffset() + OutOffset - CurAttrFixup.InputAttrStartOffset;
1783 DWARFFormValue Val = AttrSpec.getFormValue();
1784 uint64_t AttrSize = Offset;
1785 Val.extractValue(Data, &Offset, U.getFormParams(), &U);
1786 CurAttrFixup.InputAttrEndOffset = InputDIE.getOffset() + Offset;
1787 AttrSize = Offset - AttrSize;
1789 uint64_t FinalAttrSize =
1790 cloneAttribute(*Die, InputDIE, File, Unit, Val, AttrSpec, AttrSize,
1791 AttrInfo, IsLittleEndian);
1792 if (FinalAttrSize != 0 && ObjFile.Addresses->needToSaveValidRelocs())
1793 AttributesFixups.push_back(CurAttrFixup);
1795 OutOffset += FinalAttrSize;
1798 uint16_t Tag = InputDIE.getTag();
1799 // Add the DW_AT_APPLE_origin attribute to Compile Unit die if we have
1800 // an install name and the DWARF doesn't have the attribute yet.
1801 const bool NeedsAppleOrigin = (Tag == dwarf::DW_TAG_compile_unit) &&
1802 LibraryInstallName.has_value() &&
1803 !AttrInfo.HasAppleOrigin;
1804 if (NeedsAppleOrigin) {
1805 auto StringEntry = DebugStrPool.getEntry(LibraryInstallName.value());
1806 Die->addValue(DIEAlloc, dwarf::Attribute(dwarf::DW_AT_APPLE_origin),
1807 dwarf::DW_FORM_strp, DIEInteger(StringEntry.getOffset()));
1808 AttrInfo.Name = StringEntry;
1809 OutOffset += 4;
1812 // Look for accelerator entries.
1813 // FIXME: This is slightly wrong. An inline_subroutine without a
1814 // low_pc, but with AT_ranges might be interesting to get into the
1815 // accelerator tables too. For now stick with dsymutil's behavior.
1816 if ((Info.InDebugMap || AttrInfo.HasLowPc || AttrInfo.HasRanges) &&
1817 Tag != dwarf::DW_TAG_compile_unit &&
1818 getDIENames(InputDIE, AttrInfo, DebugStrPool,
1819 Tag != dwarf::DW_TAG_inlined_subroutine)) {
1820 if (AttrInfo.MangledName && AttrInfo.MangledName != AttrInfo.Name)
1821 Unit.addNameAccelerator(Die, AttrInfo.MangledName,
1822 Tag == dwarf::DW_TAG_inlined_subroutine);
1823 if (AttrInfo.Name) {
1824 if (AttrInfo.NameWithoutTemplate)
1825 Unit.addNameAccelerator(Die, AttrInfo.NameWithoutTemplate,
1826 /* SkipPubSection */ true);
1827 Unit.addNameAccelerator(Die, AttrInfo.Name,
1828 Tag == dwarf::DW_TAG_inlined_subroutine);
1830 if (AttrInfo.Name)
1831 addObjCAccelerator(Unit, Die, AttrInfo.Name, DebugStrPool,
1832 /* SkipPubSection =*/true);
1834 } else if (Tag == dwarf::DW_TAG_namespace) {
1835 if (!AttrInfo.Name)
1836 AttrInfo.Name = DebugStrPool.getEntry("(anonymous namespace)");
1837 Unit.addNamespaceAccelerator(Die, AttrInfo.Name);
1838 } else if (Tag == dwarf::DW_TAG_imported_declaration && AttrInfo.Name) {
1839 Unit.addNamespaceAccelerator(Die, AttrInfo.Name);
1840 } else if (isTypeTag(Tag) && !AttrInfo.IsDeclaration) {
1841 bool Success = getDIENames(InputDIE, AttrInfo, DebugStrPool);
1842 uint64_t RuntimeLang =
1843 dwarf::toUnsigned(InputDIE.find(dwarf::DW_AT_APPLE_runtime_class))
1844 .value_or(0);
1845 bool ObjCClassIsImplementation =
1846 (RuntimeLang == dwarf::DW_LANG_ObjC ||
1847 RuntimeLang == dwarf::DW_LANG_ObjC_plus_plus) &&
1848 dwarf::toUnsigned(InputDIE.find(dwarf::DW_AT_APPLE_objc_complete_type))
1849 .value_or(0);
1850 if (Success && AttrInfo.Name && !AttrInfo.Name.getString().empty()) {
1851 uint32_t Hash = hashFullyQualifiedName(InputDIE, Unit, File);
1852 Unit.addTypeAccelerator(Die, AttrInfo.Name, ObjCClassIsImplementation,
1853 Hash);
1856 // For Swift, mangled names are put into DW_AT_linkage_name.
1857 if (Success && AttrInfo.MangledName &&
1858 RuntimeLang == dwarf::DW_LANG_Swift &&
1859 !AttrInfo.MangledName.getString().empty() &&
1860 AttrInfo.MangledName != AttrInfo.Name) {
1861 auto Hash = djbHash(AttrInfo.MangledName.getString().data());
1862 Unit.addTypeAccelerator(Die, AttrInfo.MangledName,
1863 ObjCClassIsImplementation, Hash);
1867 // Determine whether there are any children that we want to keep.
1868 bool HasChildren = false;
1869 for (auto Child : InputDIE.children()) {
1870 unsigned Idx = U.getDIEIndex(Child);
1871 if (Unit.getInfo(Idx).Keep) {
1872 HasChildren = true;
1873 break;
1877 if (Unit.getOrigUnit().getVersion() >= 5 && !AttrInfo.AttrStrOffsetBaseSeen &&
1878 Die->getTag() == dwarf::DW_TAG_compile_unit) {
1879 // No DW_AT_str_offsets_base seen, add it to the DIE.
1880 Die->addValue(DIEAlloc, dwarf::DW_AT_str_offsets_base,
1881 dwarf::DW_FORM_sec_offset, DIEInteger(8));
1882 OutOffset += 4;
1885 DIEAbbrev NewAbbrev = Die->generateAbbrev();
1886 if (HasChildren)
1887 NewAbbrev.setChildrenFlag(dwarf::DW_CHILDREN_yes);
1888 // Assign a permanent abbrev number
1889 Linker.assignAbbrev(NewAbbrev);
1890 Die->setAbbrevNumber(NewAbbrev.getNumber());
1892 uint64_t AbbrevNumberSize = getULEB128Size(Die->getAbbrevNumber());
1894 // Add the size of the abbreviation number to the output offset.
1895 OutOffset += AbbrevNumberSize;
1897 // Update fixups with the size of the abbreviation number
1898 for (AttributeLinkedOffsetFixup &F : AttributesFixups)
1899 F.LinkedOffsetFixupVal += AbbrevNumberSize;
1901 for (AttributeLinkedOffsetFixup &F : AttributesFixups)
1902 ObjFile.Addresses->updateAndSaveValidRelocs(
1903 Unit.getOrigUnit().getVersion() >= 5, Unit.getOrigUnit().getOffset(),
1904 F.LinkedOffsetFixupVal, F.InputAttrStartOffset, F.InputAttrEndOffset);
1906 if (!HasChildren) {
1907 // Update our size.
1908 Die->setSize(OutOffset - Die->getOffset());
1909 return Die;
1912 // Recursively clone children.
1913 for (auto Child : InputDIE.children()) {
1914 if (DIE *Clone = cloneDIE(Child, File, Unit, PCOffset, OutOffset, Flags,
1915 IsLittleEndian)) {
1916 Die->addChild(Clone);
1917 OutOffset = Clone->getOffset() + Clone->getSize();
1921 // Account for the end of children marker.
1922 OutOffset += sizeof(int8_t);
1923 // Update our size.
1924 Die->setSize(OutOffset - Die->getOffset());
1925 return Die;
1928 /// Patch the input object file relevant debug_ranges or debug_rnglists
1929 /// entries and emit them in the output file. Update the relevant attributes
1930 /// to point at the new entries.
1931 void DWARFLinker::generateUnitRanges(CompileUnit &Unit, const DWARFFile &File,
1932 DebugDieValuePool &AddrPool) const {
1933 if (LLVM_UNLIKELY(Options.Update))
1934 return;
1936 const auto &FunctionRanges = Unit.getFunctionRanges();
1938 // Build set of linked address ranges for unit function ranges.
1939 AddressRanges LinkedFunctionRanges;
1940 for (const AddressRangeValuePair &Range : FunctionRanges)
1941 LinkedFunctionRanges.insert(
1942 {Range.Range.start() + Range.Value, Range.Range.end() + Range.Value});
1944 // Emit LinkedFunctionRanges into .debug_aranges
1945 if (!LinkedFunctionRanges.empty())
1946 TheDwarfEmitter->emitDwarfDebugArangesTable(Unit, LinkedFunctionRanges);
1948 RngListAttributesTy AllRngListAttributes = Unit.getRangesAttributes();
1949 std::optional<PatchLocation> UnitRngListAttribute =
1950 Unit.getUnitRangesAttribute();
1952 if (!AllRngListAttributes.empty() || UnitRngListAttribute) {
1953 std::optional<AddressRangeValuePair> CachedRange;
1954 MCSymbol *EndLabel = TheDwarfEmitter->emitDwarfDebugRangeListHeader(Unit);
1956 // Read original address ranges, apply relocation value, emit linked address
1957 // ranges.
1958 for (PatchLocation &AttributePatch : AllRngListAttributes) {
1959 // Get ranges from the source DWARF corresponding to the current
1960 // attribute.
1961 AddressRanges LinkedRanges;
1962 if (Expected<DWARFAddressRangesVector> OriginalRanges =
1963 Unit.getOrigUnit().findRnglistFromOffset(AttributePatch.get())) {
1964 // Apply relocation adjustment.
1965 for (const auto &Range : *OriginalRanges) {
1966 if (!CachedRange || !CachedRange->Range.contains(Range.LowPC))
1967 CachedRange = FunctionRanges.getRangeThatContains(Range.LowPC);
1969 // All range entries should lie in the function range.
1970 if (!CachedRange) {
1971 reportWarning("inconsistent range data.", File);
1972 continue;
1975 // Store range for emiting.
1976 LinkedRanges.insert({Range.LowPC + CachedRange->Value,
1977 Range.HighPC + CachedRange->Value});
1979 } else {
1980 llvm::consumeError(OriginalRanges.takeError());
1981 reportWarning("invalid range list ignored.", File);
1984 // Emit linked ranges.
1985 TheDwarfEmitter->emitDwarfDebugRangeListFragment(
1986 Unit, LinkedRanges, AttributePatch, AddrPool);
1989 // Emit ranges for Unit AT_ranges attribute.
1990 if (UnitRngListAttribute.has_value())
1991 TheDwarfEmitter->emitDwarfDebugRangeListFragment(
1992 Unit, LinkedFunctionRanges, *UnitRngListAttribute, AddrPool);
1994 // Emit ranges footer.
1995 TheDwarfEmitter->emitDwarfDebugRangeListFooter(Unit, EndLabel);
1999 void DWARFLinker::DIECloner::generateUnitLocations(
2000 CompileUnit &Unit, const DWARFFile &File,
2001 ExpressionHandlerRef ExprHandler) {
2002 if (LLVM_UNLIKELY(Linker.Options.Update))
2003 return;
2005 const LocListAttributesTy &AllLocListAttributes =
2006 Unit.getLocationAttributes();
2008 if (AllLocListAttributes.empty())
2009 return;
2011 // Emit locations list table header.
2012 MCSymbol *EndLabel = Emitter->emitDwarfDebugLocListHeader(Unit);
2014 for (auto &CurLocAttr : AllLocListAttributes) {
2015 // Get location expressions vector corresponding to the current attribute
2016 // from the source DWARF.
2017 Expected<DWARFLocationExpressionsVector> OriginalLocations =
2018 Unit.getOrigUnit().findLoclistFromOffset(CurLocAttr.get());
2020 if (!OriginalLocations) {
2021 llvm::consumeError(OriginalLocations.takeError());
2022 Linker.reportWarning("Invalid location attribute ignored.", File);
2023 continue;
2026 DWARFLocationExpressionsVector LinkedLocationExpressions;
2027 for (DWARFLocationExpression &CurExpression : *OriginalLocations) {
2028 DWARFLocationExpression LinkedExpression;
2030 if (CurExpression.Range) {
2031 // Relocate address range.
2032 LinkedExpression.Range = {
2033 CurExpression.Range->LowPC + CurLocAttr.RelocAdjustment,
2034 CurExpression.Range->HighPC + CurLocAttr.RelocAdjustment};
2037 // Clone expression.
2038 LinkedExpression.Expr.reserve(CurExpression.Expr.size());
2039 ExprHandler(CurExpression.Expr, LinkedExpression.Expr,
2040 CurLocAttr.RelocAdjustment);
2042 LinkedLocationExpressions.push_back(LinkedExpression);
2045 // Emit locations list table fragment corresponding to the CurLocAttr.
2046 Emitter->emitDwarfDebugLocListFragment(Unit, LinkedLocationExpressions,
2047 CurLocAttr, AddrPool);
2050 // Emit locations list table footer.
2051 Emitter->emitDwarfDebugLocListFooter(Unit, EndLabel);
2054 static void patchAddrBase(DIE &Die, DIEInteger Offset) {
2055 for (auto &V : Die.values())
2056 if (V.getAttribute() == dwarf::DW_AT_addr_base) {
2057 V = DIEValue(V.getAttribute(), V.getForm(), Offset);
2058 return;
2061 llvm_unreachable("Didn't find a DW_AT_addr_base in cloned DIE!");
2064 void DWARFLinker::DIECloner::emitDebugAddrSection(
2065 CompileUnit &Unit, const uint16_t DwarfVersion) const {
2067 if (LLVM_UNLIKELY(Linker.Options.Update))
2068 return;
2070 if (DwarfVersion < 5)
2071 return;
2073 if (AddrPool.getValues().empty())
2074 return;
2076 MCSymbol *EndLabel = Emitter->emitDwarfDebugAddrsHeader(Unit);
2077 patchAddrBase(*Unit.getOutputUnitDIE(),
2078 DIEInteger(Emitter->getDebugAddrSectionSize()));
2079 Emitter->emitDwarfDebugAddrs(AddrPool.getValues(),
2080 Unit.getOrigUnit().getAddressByteSize());
2081 Emitter->emitDwarfDebugAddrsFooter(Unit, EndLabel);
2084 /// Insert the new line info sequence \p Seq into the current
2085 /// set of already linked line info \p Rows.
2086 static void insertLineSequence(std::vector<DWARFDebugLine::Row> &Seq,
2087 std::vector<DWARFDebugLine::Row> &Rows) {
2088 if (Seq.empty())
2089 return;
2091 if (!Rows.empty() && Rows.back().Address < Seq.front().Address) {
2092 llvm::append_range(Rows, Seq);
2093 Seq.clear();
2094 return;
2097 object::SectionedAddress Front = Seq.front().Address;
2098 auto InsertPoint = partition_point(
2099 Rows, [=](const DWARFDebugLine::Row &O) { return O.Address < Front; });
2101 // FIXME: this only removes the unneeded end_sequence if the
2102 // sequences have been inserted in order. Using a global sort like
2103 // described in generateLineTableForUnit() and delaying the end_sequene
2104 // elimination to emitLineTableForUnit() we can get rid of all of them.
2105 if (InsertPoint != Rows.end() && InsertPoint->Address == Front &&
2106 InsertPoint->EndSequence) {
2107 *InsertPoint = Seq.front();
2108 Rows.insert(InsertPoint + 1, Seq.begin() + 1, Seq.end());
2109 } else {
2110 Rows.insert(InsertPoint, Seq.begin(), Seq.end());
2113 Seq.clear();
2116 static void patchStmtList(DIE &Die, DIEInteger Offset) {
2117 for (auto &V : Die.values())
2118 if (V.getAttribute() == dwarf::DW_AT_stmt_list) {
2119 V = DIEValue(V.getAttribute(), V.getForm(), Offset);
2120 return;
2123 llvm_unreachable("Didn't find DW_AT_stmt_list in cloned DIE!");
2126 void DWARFLinker::DIECloner::rememberUnitForMacroOffset(CompileUnit &Unit) {
2127 DWARFUnit &OrigUnit = Unit.getOrigUnit();
2128 DWARFDie OrigUnitDie = OrigUnit.getUnitDIE();
2130 if (std::optional<uint64_t> MacroAttr =
2131 dwarf::toSectionOffset(OrigUnitDie.find(dwarf::DW_AT_macros))) {
2132 UnitMacroMap.insert(std::make_pair(*MacroAttr, &Unit));
2133 return;
2136 if (std::optional<uint64_t> MacroAttr =
2137 dwarf::toSectionOffset(OrigUnitDie.find(dwarf::DW_AT_macro_info))) {
2138 UnitMacroMap.insert(std::make_pair(*MacroAttr, &Unit));
2139 return;
2143 void DWARFLinker::DIECloner::generateLineTableForUnit(CompileUnit &Unit) {
2144 if (LLVM_UNLIKELY(Emitter == nullptr))
2145 return;
2147 // Check whether DW_AT_stmt_list attribute is presented.
2148 DWARFDie CUDie = Unit.getOrigUnit().getUnitDIE();
2149 auto StmtList = dwarf::toSectionOffset(CUDie.find(dwarf::DW_AT_stmt_list));
2150 if (!StmtList)
2151 return;
2153 // Update the cloned DW_AT_stmt_list with the correct debug_line offset.
2154 if (auto *OutputDIE = Unit.getOutputUnitDIE())
2155 patchStmtList(*OutputDIE, DIEInteger(Emitter->getLineSectionSize()));
2157 if (const DWARFDebugLine::LineTable *LT =
2158 ObjFile.Dwarf->getLineTableForUnit(&Unit.getOrigUnit())) {
2160 DWARFDebugLine::LineTable LineTable;
2162 // Set Line Table header.
2163 LineTable.Prologue = LT->Prologue;
2165 // Set Line Table Rows.
2166 if (Linker.Options.Update) {
2167 LineTable.Rows = LT->Rows;
2168 // If all the line table contains is a DW_LNE_end_sequence, clear the line
2169 // table rows, it will be inserted again in the DWARFStreamer.
2170 if (LineTable.Rows.size() == 1 && LineTable.Rows[0].EndSequence)
2171 LineTable.Rows.clear();
2173 LineTable.Sequences = LT->Sequences;
2174 } else {
2175 // This vector is the output line table.
2176 std::vector<DWARFDebugLine::Row> NewRows;
2177 NewRows.reserve(LT->Rows.size());
2179 // Current sequence of rows being extracted, before being inserted
2180 // in NewRows.
2181 std::vector<DWARFDebugLine::Row> Seq;
2183 const auto &FunctionRanges = Unit.getFunctionRanges();
2184 std::optional<AddressRangeValuePair> CurrRange;
2186 // FIXME: This logic is meant to generate exactly the same output as
2187 // Darwin's classic dsymutil. There is a nicer way to implement this
2188 // by simply putting all the relocated line info in NewRows and simply
2189 // sorting NewRows before passing it to emitLineTableForUnit. This
2190 // should be correct as sequences for a function should stay
2191 // together in the sorted output. There are a few corner cases that
2192 // look suspicious though, and that required to implement the logic
2193 // this way. Revisit that once initial validation is finished.
2195 // Iterate over the object file line info and extract the sequences
2196 // that correspond to linked functions.
2197 for (DWARFDebugLine::Row Row : LT->Rows) {
2198 // Check whether we stepped out of the range. The range is
2199 // half-open, but consider accept the end address of the range if
2200 // it is marked as end_sequence in the input (because in that
2201 // case, the relocation offset is accurate and that entry won't
2202 // serve as the start of another function).
2203 if (!CurrRange || !CurrRange->Range.contains(Row.Address.Address)) {
2204 // We just stepped out of a known range. Insert a end_sequence
2205 // corresponding to the end of the range.
2206 uint64_t StopAddress =
2207 CurrRange ? CurrRange->Range.end() + CurrRange->Value : -1ULL;
2208 CurrRange = FunctionRanges.getRangeThatContains(Row.Address.Address);
2209 if (StopAddress != -1ULL && !Seq.empty()) {
2210 // Insert end sequence row with the computed end address, but
2211 // the same line as the previous one.
2212 auto NextLine = Seq.back();
2213 NextLine.Address.Address = StopAddress;
2214 NextLine.EndSequence = 1;
2215 NextLine.PrologueEnd = 0;
2216 NextLine.BasicBlock = 0;
2217 NextLine.EpilogueBegin = 0;
2218 Seq.push_back(NextLine);
2219 insertLineSequence(Seq, NewRows);
2222 if (!CurrRange)
2223 continue;
2226 // Ignore empty sequences.
2227 if (Row.EndSequence && Seq.empty())
2228 continue;
2230 // Relocate row address and add it to the current sequence.
2231 Row.Address.Address += CurrRange->Value;
2232 Seq.emplace_back(Row);
2234 if (Row.EndSequence)
2235 insertLineSequence(Seq, NewRows);
2238 LineTable.Rows = std::move(NewRows);
2241 Emitter->emitLineTableForUnit(LineTable, Unit, DebugStrPool,
2242 DebugLineStrPool);
2243 } else
2244 Linker.reportWarning("Cann't load line table.", ObjFile);
2247 void DWARFLinker::emitAcceleratorEntriesForUnit(CompileUnit &Unit) {
2248 for (AccelTableKind AccelTableKind : Options.AccelTables) {
2249 switch (AccelTableKind) {
2250 case AccelTableKind::Apple: {
2251 // Add namespaces.
2252 for (const auto &Namespace : Unit.getNamespaces())
2253 AppleNamespaces.addName(Namespace.Name, Namespace.Die->getOffset() +
2254 Unit.getStartOffset());
2255 // Add names.
2256 for (const auto &Pubname : Unit.getPubnames())
2257 AppleNames.addName(Pubname.Name,
2258 Pubname.Die->getOffset() + Unit.getStartOffset());
2259 // Add types.
2260 for (const auto &Pubtype : Unit.getPubtypes())
2261 AppleTypes.addName(
2262 Pubtype.Name, Pubtype.Die->getOffset() + Unit.getStartOffset(),
2263 Pubtype.Die->getTag(),
2264 Pubtype.ObjcClassImplementation ? dwarf::DW_FLAG_type_implementation
2265 : 0,
2266 Pubtype.QualifiedNameHash);
2267 // Add ObjC names.
2268 for (const auto &ObjC : Unit.getObjC())
2269 AppleObjc.addName(ObjC.Name,
2270 ObjC.Die->getOffset() + Unit.getStartOffset());
2271 } break;
2272 case AccelTableKind::Pub: {
2273 TheDwarfEmitter->emitPubNamesForUnit(Unit);
2274 TheDwarfEmitter->emitPubTypesForUnit(Unit);
2275 } break;
2276 case AccelTableKind::DebugNames: {
2277 for (const auto &Namespace : Unit.getNamespaces())
2278 DebugNames.addName(
2279 Namespace.Name, Namespace.Die->getOffset(),
2280 DWARF5AccelTableData::getDefiningParentDieOffset(*Namespace.Die),
2281 Namespace.Die->getTag(), Unit.getUniqueID(),
2282 Unit.getTag() == dwarf::DW_TAG_type_unit);
2283 for (const auto &Pubname : Unit.getPubnames())
2284 DebugNames.addName(
2285 Pubname.Name, Pubname.Die->getOffset(),
2286 DWARF5AccelTableData::getDefiningParentDieOffset(*Pubname.Die),
2287 Pubname.Die->getTag(), Unit.getUniqueID(),
2288 Unit.getTag() == dwarf::DW_TAG_type_unit);
2289 for (const auto &Pubtype : Unit.getPubtypes())
2290 DebugNames.addName(
2291 Pubtype.Name, Pubtype.Die->getOffset(),
2292 DWARF5AccelTableData::getDefiningParentDieOffset(*Pubtype.Die),
2293 Pubtype.Die->getTag(), Unit.getUniqueID(),
2294 Unit.getTag() == dwarf::DW_TAG_type_unit);
2295 } break;
2300 /// Read the frame info stored in the object, and emit the
2301 /// patched frame descriptions for the resulting file.
2303 /// This is actually pretty easy as the data of the CIEs and FDEs can
2304 /// be considered as black boxes and moved as is. The only thing to do
2305 /// is to patch the addresses in the headers.
2306 void DWARFLinker::patchFrameInfoForObject(LinkContext &Context) {
2307 DWARFContext &OrigDwarf = *Context.File.Dwarf;
2308 unsigned SrcAddrSize = OrigDwarf.getDWARFObj().getAddressSize();
2310 StringRef FrameData = OrigDwarf.getDWARFObj().getFrameSection().Data;
2311 if (FrameData.empty())
2312 return;
2314 RangesTy AllUnitsRanges;
2315 for (std::unique_ptr<CompileUnit> &Unit : Context.CompileUnits) {
2316 for (auto CurRange : Unit->getFunctionRanges())
2317 AllUnitsRanges.insert(CurRange.Range, CurRange.Value);
2320 DataExtractor Data(FrameData, OrigDwarf.isLittleEndian(), 0);
2321 uint64_t InputOffset = 0;
2323 // Store the data of the CIEs defined in this object, keyed by their
2324 // offsets.
2325 DenseMap<uint64_t, StringRef> LocalCIES;
2327 while (Data.isValidOffset(InputOffset)) {
2328 uint64_t EntryOffset = InputOffset;
2329 uint32_t InitialLength = Data.getU32(&InputOffset);
2330 if (InitialLength == 0xFFFFFFFF)
2331 return reportWarning("Dwarf64 bits no supported", Context.File);
2333 uint32_t CIEId = Data.getU32(&InputOffset);
2334 if (CIEId == 0xFFFFFFFF) {
2335 // This is a CIE, store it.
2336 StringRef CIEData = FrameData.substr(EntryOffset, InitialLength + 4);
2337 LocalCIES[EntryOffset] = CIEData;
2338 // The -4 is to account for the CIEId we just read.
2339 InputOffset += InitialLength - 4;
2340 continue;
2343 uint64_t Loc = Data.getUnsigned(&InputOffset, SrcAddrSize);
2345 // Some compilers seem to emit frame info that doesn't start at
2346 // the function entry point, thus we can't just lookup the address
2347 // in the debug map. Use the AddressInfo's range map to see if the FDE
2348 // describes something that we can relocate.
2349 std::optional<AddressRangeValuePair> Range =
2350 AllUnitsRanges.getRangeThatContains(Loc);
2351 if (!Range) {
2352 // The +4 is to account for the size of the InitialLength field itself.
2353 InputOffset = EntryOffset + InitialLength + 4;
2354 continue;
2357 // This is an FDE, and we have a mapping.
2358 // Have we already emitted a corresponding CIE?
2359 StringRef CIEData = LocalCIES[CIEId];
2360 if (CIEData.empty())
2361 return reportWarning("Inconsistent debug_frame content. Dropping.",
2362 Context.File);
2364 // Look if we already emitted a CIE that corresponds to the
2365 // referenced one (the CIE data is the key of that lookup).
2366 auto IteratorInserted = EmittedCIEs.insert(
2367 std::make_pair(CIEData, TheDwarfEmitter->getFrameSectionSize()));
2368 // If there is no CIE yet for this ID, emit it.
2369 if (IteratorInserted.second) {
2370 LastCIEOffset = TheDwarfEmitter->getFrameSectionSize();
2371 IteratorInserted.first->getValue() = LastCIEOffset;
2372 TheDwarfEmitter->emitCIE(CIEData);
2375 // Emit the FDE with updated address and CIE pointer.
2376 // (4 + AddrSize) is the size of the CIEId + initial_location
2377 // fields that will get reconstructed by emitFDE().
2378 unsigned FDERemainingBytes = InitialLength - (4 + SrcAddrSize);
2379 TheDwarfEmitter->emitFDE(IteratorInserted.first->getValue(), SrcAddrSize,
2380 Loc + Range->Value,
2381 FrameData.substr(InputOffset, FDERemainingBytes));
2382 InputOffset += FDERemainingBytes;
2386 uint32_t DWARFLinker::DIECloner::hashFullyQualifiedName(DWARFDie DIE,
2387 CompileUnit &U,
2388 const DWARFFile &File,
2389 int ChildRecurseDepth) {
2390 const char *Name = nullptr;
2391 DWARFUnit *OrigUnit = &U.getOrigUnit();
2392 CompileUnit *CU = &U;
2393 std::optional<DWARFFormValue> Ref;
2395 while (true) {
2396 if (const char *CurrentName = DIE.getName(DINameKind::ShortName))
2397 Name = CurrentName;
2399 if (!(Ref = DIE.find(dwarf::DW_AT_specification)) &&
2400 !(Ref = DIE.find(dwarf::DW_AT_abstract_origin)))
2401 break;
2403 if (!Ref->isFormClass(DWARFFormValue::FC_Reference))
2404 break;
2406 CompileUnit *RefCU;
2407 if (auto RefDIE =
2408 Linker.resolveDIEReference(File, CompileUnits, *Ref, DIE, RefCU)) {
2409 CU = RefCU;
2410 OrigUnit = &RefCU->getOrigUnit();
2411 DIE = RefDIE;
2415 unsigned Idx = OrigUnit->getDIEIndex(DIE);
2416 if (!Name && DIE.getTag() == dwarf::DW_TAG_namespace)
2417 Name = "(anonymous namespace)";
2419 if (CU->getInfo(Idx).ParentIdx == 0 ||
2420 // FIXME: dsymutil-classic compatibility. Ignore modules.
2421 CU->getOrigUnit().getDIEAtIndex(CU->getInfo(Idx).ParentIdx).getTag() ==
2422 dwarf::DW_TAG_module)
2423 return djbHash(Name ? Name : "", djbHash(ChildRecurseDepth ? "" : "::"));
2425 DWARFDie Die = OrigUnit->getDIEAtIndex(CU->getInfo(Idx).ParentIdx);
2426 return djbHash(
2427 (Name ? Name : ""),
2428 djbHash((Name ? "::" : ""),
2429 hashFullyQualifiedName(Die, *CU, File, ++ChildRecurseDepth)));
2432 static uint64_t getDwoId(const DWARFDie &CUDie) {
2433 auto DwoId = dwarf::toUnsigned(
2434 CUDie.find({dwarf::DW_AT_dwo_id, dwarf::DW_AT_GNU_dwo_id}));
2435 if (DwoId)
2436 return *DwoId;
2437 return 0;
2440 static std::string
2441 remapPath(StringRef Path,
2442 const DWARFLinkerBase::ObjectPrefixMapTy &ObjectPrefixMap) {
2443 if (ObjectPrefixMap.empty())
2444 return Path.str();
2446 SmallString<256> p = Path;
2447 for (const auto &Entry : ObjectPrefixMap)
2448 if (llvm::sys::path::replace_path_prefix(p, Entry.first, Entry.second))
2449 break;
2450 return p.str().str();
2453 static std::string
2454 getPCMFile(const DWARFDie &CUDie,
2455 const DWARFLinkerBase::ObjectPrefixMapTy *ObjectPrefixMap) {
2456 std::string PCMFile = dwarf::toString(
2457 CUDie.find({dwarf::DW_AT_dwo_name, dwarf::DW_AT_GNU_dwo_name}), "");
2459 if (PCMFile.empty())
2460 return PCMFile;
2462 if (ObjectPrefixMap)
2463 PCMFile = remapPath(PCMFile, *ObjectPrefixMap);
2465 return PCMFile;
2468 std::pair<bool, bool> DWARFLinker::isClangModuleRef(const DWARFDie &CUDie,
2469 std::string &PCMFile,
2470 LinkContext &Context,
2471 unsigned Indent,
2472 bool Quiet) {
2473 if (PCMFile.empty())
2474 return std::make_pair(false, false);
2476 // Clang module DWARF skeleton CUs abuse this for the path to the module.
2477 uint64_t DwoId = getDwoId(CUDie);
2479 std::string Name = dwarf::toString(CUDie.find(dwarf::DW_AT_name), "");
2480 if (Name.empty()) {
2481 if (!Quiet)
2482 reportWarning("Anonymous module skeleton CU for " + PCMFile,
2483 Context.File);
2484 return std::make_pair(true, true);
2487 if (!Quiet && Options.Verbose) {
2488 outs().indent(Indent);
2489 outs() << "Found clang module reference " << PCMFile;
2492 auto Cached = ClangModules.find(PCMFile);
2493 if (Cached != ClangModules.end()) {
2494 // FIXME: Until PR27449 (https://llvm.org/bugs/show_bug.cgi?id=27449) is
2495 // fixed in clang, only warn about DWO_id mismatches in verbose mode.
2496 // ASTFileSignatures will change randomly when a module is rebuilt.
2497 if (!Quiet && Options.Verbose && (Cached->second != DwoId))
2498 reportWarning(Twine("hash mismatch: this object file was built against a "
2499 "different version of the module ") +
2500 PCMFile,
2501 Context.File);
2502 if (!Quiet && Options.Verbose)
2503 outs() << " [cached].\n";
2504 return std::make_pair(true, true);
2507 return std::make_pair(true, false);
2510 bool DWARFLinker::registerModuleReference(const DWARFDie &CUDie,
2511 LinkContext &Context,
2512 ObjFileLoaderTy Loader,
2513 CompileUnitHandlerTy OnCUDieLoaded,
2514 unsigned Indent) {
2515 std::string PCMFile = getPCMFile(CUDie, Options.ObjectPrefixMap);
2516 std::pair<bool, bool> IsClangModuleRef =
2517 isClangModuleRef(CUDie, PCMFile, Context, Indent, false);
2519 if (!IsClangModuleRef.first)
2520 return false;
2522 if (IsClangModuleRef.second)
2523 return true;
2525 if (Options.Verbose)
2526 outs() << " ...\n";
2528 // Cyclic dependencies are disallowed by Clang, but we still
2529 // shouldn't run into an infinite loop, so mark it as processed now.
2530 ClangModules.insert({PCMFile, getDwoId(CUDie)});
2532 if (Error E = loadClangModule(Loader, CUDie, PCMFile, Context, OnCUDieLoaded,
2533 Indent + 2)) {
2534 consumeError(std::move(E));
2535 return false;
2537 return true;
2540 Error DWARFLinker::loadClangModule(
2541 ObjFileLoaderTy Loader, const DWARFDie &CUDie, const std::string &PCMFile,
2542 LinkContext &Context, CompileUnitHandlerTy OnCUDieLoaded, unsigned Indent) {
2544 uint64_t DwoId = getDwoId(CUDie);
2545 std::string ModuleName = dwarf::toString(CUDie.find(dwarf::DW_AT_name), "");
2547 /// Using a SmallString<0> because loadClangModule() is recursive.
2548 SmallString<0> Path(Options.PrependPath);
2549 if (sys::path::is_relative(PCMFile))
2550 resolveRelativeObjectPath(Path, CUDie);
2551 sys::path::append(Path, PCMFile);
2552 // Don't use the cached binary holder because we have no thread-safety
2553 // guarantee and the lifetime is limited.
2555 if (Loader == nullptr) {
2556 reportError("Could not load clang module: loader is not specified.\n",
2557 Context.File);
2558 return Error::success();
2561 auto ErrOrObj = Loader(Context.File.FileName, Path);
2562 if (!ErrOrObj)
2563 return Error::success();
2565 std::unique_ptr<CompileUnit> Unit;
2566 for (const auto &CU : ErrOrObj->Dwarf->compile_units()) {
2567 OnCUDieLoaded(*CU);
2568 // Recursively get all modules imported by this one.
2569 auto ChildCUDie = CU->getUnitDIE();
2570 if (!ChildCUDie)
2571 continue;
2572 if (!registerModuleReference(ChildCUDie, Context, Loader, OnCUDieLoaded,
2573 Indent)) {
2574 if (Unit) {
2575 std::string Err =
2576 (PCMFile +
2577 ": Clang modules are expected to have exactly 1 compile unit.\n");
2578 reportError(Err, Context.File);
2579 return make_error<StringError>(Err, inconvertibleErrorCode());
2581 // FIXME: Until PR27449 (https://llvm.org/bugs/show_bug.cgi?id=27449) is
2582 // fixed in clang, only warn about DWO_id mismatches in verbose mode.
2583 // ASTFileSignatures will change randomly when a module is rebuilt.
2584 uint64_t PCMDwoId = getDwoId(ChildCUDie);
2585 if (PCMDwoId != DwoId) {
2586 if (Options.Verbose)
2587 reportWarning(
2588 Twine("hash mismatch: this object file was built against a "
2589 "different version of the module ") +
2590 PCMFile,
2591 Context.File);
2592 // Update the cache entry with the DwoId of the module loaded from disk.
2593 ClangModules[PCMFile] = PCMDwoId;
2596 // Add this module.
2597 Unit = std::make_unique<CompileUnit>(*CU, UniqueUnitID++, !Options.NoODR,
2598 ModuleName);
2602 if (Unit)
2603 Context.ModuleUnits.emplace_back(RefModuleUnit{*ErrOrObj, std::move(Unit)});
2605 return Error::success();
2608 uint64_t DWARFLinker::DIECloner::cloneAllCompileUnits(
2609 DWARFContext &DwarfContext, const DWARFFile &File, bool IsLittleEndian) {
2610 uint64_t OutputDebugInfoSize =
2611 (Emitter == nullptr) ? 0 : Emitter->getDebugInfoSectionSize();
2612 const uint64_t StartOutputDebugInfoSize = OutputDebugInfoSize;
2614 for (auto &CurrentUnit : CompileUnits) {
2615 const uint16_t DwarfVersion = CurrentUnit->getOrigUnit().getVersion();
2616 const uint32_t UnitHeaderSize = DwarfVersion >= 5 ? 12 : 11;
2617 auto InputDIE = CurrentUnit->getOrigUnit().getUnitDIE();
2618 CurrentUnit->setStartOffset(OutputDebugInfoSize);
2619 if (!InputDIE) {
2620 OutputDebugInfoSize = CurrentUnit->computeNextUnitOffset(DwarfVersion);
2621 continue;
2623 if (CurrentUnit->getInfo(0).Keep) {
2624 // Clone the InputDIE into your Unit DIE in our compile unit since it
2625 // already has a DIE inside of it.
2626 CurrentUnit->createOutputDIE();
2627 rememberUnitForMacroOffset(*CurrentUnit);
2628 cloneDIE(InputDIE, File, *CurrentUnit, 0 /* PC offset */, UnitHeaderSize,
2629 0, IsLittleEndian, CurrentUnit->getOutputUnitDIE());
2632 OutputDebugInfoSize = CurrentUnit->computeNextUnitOffset(DwarfVersion);
2634 if (Emitter != nullptr) {
2636 generateLineTableForUnit(*CurrentUnit);
2638 Linker.emitAcceleratorEntriesForUnit(*CurrentUnit);
2640 if (LLVM_UNLIKELY(Linker.Options.Update))
2641 continue;
2643 Linker.generateUnitRanges(*CurrentUnit, File, AddrPool);
2645 auto ProcessExpr = [&](SmallVectorImpl<uint8_t> &SrcBytes,
2646 SmallVectorImpl<uint8_t> &OutBytes,
2647 int64_t RelocAdjustment) {
2648 DWARFUnit &OrigUnit = CurrentUnit->getOrigUnit();
2649 DataExtractor Data(SrcBytes, IsLittleEndian,
2650 OrigUnit.getAddressByteSize());
2651 cloneExpression(Data,
2652 DWARFExpression(Data, OrigUnit.getAddressByteSize(),
2653 OrigUnit.getFormParams().Format),
2654 File, *CurrentUnit, OutBytes, RelocAdjustment,
2655 IsLittleEndian);
2657 generateUnitLocations(*CurrentUnit, File, ProcessExpr);
2658 emitDebugAddrSection(*CurrentUnit, DwarfVersion);
2660 AddrPool.clear();
2663 if (Emitter != nullptr) {
2664 assert(Emitter);
2665 // Emit macro tables.
2666 Emitter->emitMacroTables(File.Dwarf.get(), UnitMacroMap, DebugStrPool);
2668 // Emit all the compile unit's debug information.
2669 for (auto &CurrentUnit : CompileUnits) {
2670 CurrentUnit->fixupForwardReferences();
2672 if (!CurrentUnit->getOutputUnitDIE())
2673 continue;
2675 unsigned DwarfVersion = CurrentUnit->getOrigUnit().getVersion();
2677 assert(Emitter->getDebugInfoSectionSize() ==
2678 CurrentUnit->getStartOffset());
2679 Emitter->emitCompileUnitHeader(*CurrentUnit, DwarfVersion);
2680 Emitter->emitDIE(*CurrentUnit->getOutputUnitDIE());
2681 assert(Emitter->getDebugInfoSectionSize() ==
2682 CurrentUnit->computeNextUnitOffset(DwarfVersion));
2686 return OutputDebugInfoSize - StartOutputDebugInfoSize;
2689 void DWARFLinker::copyInvariantDebugSection(DWARFContext &Dwarf) {
2690 TheDwarfEmitter->emitSectionContents(Dwarf.getDWARFObj().getLocSection().Data,
2691 DebugSectionKind::DebugLoc);
2692 TheDwarfEmitter->emitSectionContents(
2693 Dwarf.getDWARFObj().getRangesSection().Data,
2694 DebugSectionKind::DebugRange);
2695 TheDwarfEmitter->emitSectionContents(
2696 Dwarf.getDWARFObj().getFrameSection().Data, DebugSectionKind::DebugFrame);
2697 TheDwarfEmitter->emitSectionContents(Dwarf.getDWARFObj().getArangesSection(),
2698 DebugSectionKind::DebugARanges);
2699 TheDwarfEmitter->emitSectionContents(
2700 Dwarf.getDWARFObj().getAddrSection().Data, DebugSectionKind::DebugAddr);
2701 TheDwarfEmitter->emitSectionContents(
2702 Dwarf.getDWARFObj().getRnglistsSection().Data,
2703 DebugSectionKind::DebugRngLists);
2704 TheDwarfEmitter->emitSectionContents(
2705 Dwarf.getDWARFObj().getLoclistsSection().Data,
2706 DebugSectionKind::DebugLocLists);
2709 void DWARFLinker::addObjectFile(DWARFFile &File, ObjFileLoaderTy Loader,
2710 CompileUnitHandlerTy OnCUDieLoaded) {
2711 ObjectContexts.emplace_back(LinkContext(File));
2713 if (ObjectContexts.back().File.Dwarf) {
2714 for (const std::unique_ptr<DWARFUnit> &CU :
2715 ObjectContexts.back().File.Dwarf->compile_units()) {
2716 DWARFDie CUDie = CU->getUnitDIE();
2718 if (!CUDie)
2719 continue;
2721 OnCUDieLoaded(*CU);
2723 if (!LLVM_UNLIKELY(Options.Update))
2724 registerModuleReference(CUDie, ObjectContexts.back(), Loader,
2725 OnCUDieLoaded);
2730 Error DWARFLinker::link() {
2731 assert((Options.TargetDWARFVersion != 0) &&
2732 "TargetDWARFVersion should be set");
2734 // First populate the data structure we need for each iteration of the
2735 // parallel loop.
2736 unsigned NumObjects = ObjectContexts.size();
2738 // This Dwarf string pool which is used for emission. It must be used
2739 // serially as the order of calling getStringOffset matters for
2740 // reproducibility.
2741 OffsetsStringPool DebugStrPool(true);
2742 OffsetsStringPool DebugLineStrPool(false);
2743 DebugDieValuePool StringOffsetPool;
2745 // ODR Contexts for the optimize.
2746 DeclContextTree ODRContexts;
2748 for (LinkContext &OptContext : ObjectContexts) {
2749 if (Options.Verbose)
2750 outs() << "DEBUG MAP OBJECT: " << OptContext.File.FileName << "\n";
2752 if (!OptContext.File.Dwarf)
2753 continue;
2755 if (Options.VerifyInputDWARF)
2756 verifyInput(OptContext.File);
2758 // Look for relocations that correspond to address map entries.
2760 // there was findvalidrelocations previously ... probably we need to gather
2761 // info here
2762 if (LLVM_LIKELY(!Options.Update) &&
2763 !OptContext.File.Addresses->hasValidRelocs()) {
2764 if (Options.Verbose)
2765 outs() << "No valid relocations found. Skipping.\n";
2767 // Set "Skip" flag as a signal to other loops that we should not
2768 // process this iteration.
2769 OptContext.Skip = true;
2770 continue;
2773 // Setup access to the debug info.
2774 if (!OptContext.File.Dwarf)
2775 continue;
2777 // Check whether type units are presented.
2778 if (!OptContext.File.Dwarf->types_section_units().empty()) {
2779 reportWarning("type units are not currently supported: file will "
2780 "be skipped",
2781 OptContext.File);
2782 OptContext.Skip = true;
2783 continue;
2786 // Clone all the clang modules with requires extracting the DIE units. We
2787 // don't need the full debug info until the Analyze phase.
2788 OptContext.CompileUnits.reserve(
2789 OptContext.File.Dwarf->getNumCompileUnits());
2790 for (const auto &CU : OptContext.File.Dwarf->compile_units()) {
2791 auto CUDie = CU->getUnitDIE(/*ExtractUnitDIEOnly=*/true);
2792 if (Options.Verbose) {
2793 outs() << "Input compilation unit:";
2794 DIDumpOptions DumpOpts;
2795 DumpOpts.ChildRecurseDepth = 0;
2796 DumpOpts.Verbose = Options.Verbose;
2797 CUDie.dump(outs(), 0, DumpOpts);
2801 for (auto &CU : OptContext.ModuleUnits) {
2802 if (Error Err = cloneModuleUnit(OptContext, CU, ODRContexts, DebugStrPool,
2803 DebugLineStrPool, StringOffsetPool))
2804 reportWarning(toString(std::move(Err)), CU.File);
2808 // At this point we know how much data we have emitted. We use this value to
2809 // compare canonical DIE offsets in analyzeContextInfo to see if a definition
2810 // is already emitted, without being affected by canonical die offsets set
2811 // later. This prevents undeterminism when analyze and clone execute
2812 // concurrently, as clone set the canonical DIE offset and analyze reads it.
2813 const uint64_t ModulesEndOffset =
2814 (TheDwarfEmitter == nullptr) ? 0
2815 : TheDwarfEmitter->getDebugInfoSectionSize();
2817 // These variables manage the list of processed object files.
2818 // The mutex and condition variable are to ensure that this is thread safe.
2819 std::mutex ProcessedFilesMutex;
2820 std::condition_variable ProcessedFilesConditionVariable;
2821 BitVector ProcessedFiles(NumObjects, false);
2823 // Analyzing the context info is particularly expensive so it is executed in
2824 // parallel with emitting the previous compile unit.
2825 auto AnalyzeLambda = [&](size_t I) {
2826 auto &Context = ObjectContexts[I];
2828 if (Context.Skip || !Context.File.Dwarf)
2829 return;
2831 for (const auto &CU : Context.File.Dwarf->compile_units()) {
2832 // Previously we only extracted the unit DIEs. We need the full debug info
2833 // now.
2834 auto CUDie = CU->getUnitDIE(/*ExtractUnitDIEOnly=*/false);
2835 std::string PCMFile = getPCMFile(CUDie, Options.ObjectPrefixMap);
2837 if (!CUDie || LLVM_UNLIKELY(Options.Update) ||
2838 !isClangModuleRef(CUDie, PCMFile, Context, 0, true).first) {
2839 Context.CompileUnits.push_back(std::make_unique<CompileUnit>(
2840 *CU, UniqueUnitID++, !Options.NoODR && !Options.Update, ""));
2844 // Now build the DIE parent links that we will use during the next phase.
2845 for (auto &CurrentUnit : Context.CompileUnits) {
2846 auto CUDie = CurrentUnit->getOrigUnit().getUnitDIE();
2847 if (!CUDie)
2848 continue;
2849 analyzeContextInfo(CurrentUnit->getOrigUnit().getUnitDIE(), 0,
2850 *CurrentUnit, &ODRContexts.getRoot(), ODRContexts,
2851 ModulesEndOffset, Options.ParseableSwiftInterfaces,
2852 [&](const Twine &Warning, const DWARFDie &DIE) {
2853 reportWarning(Warning, Context.File, &DIE);
2858 // For each object file map how many bytes were emitted.
2859 StringMap<DebugInfoSize> SizeByObject;
2861 // And then the remaining work in serial again.
2862 // Note, although this loop runs in serial, it can run in parallel with
2863 // the analyzeContextInfo loop so long as we process files with indices >=
2864 // than those processed by analyzeContextInfo.
2865 auto CloneLambda = [&](size_t I) {
2866 auto &OptContext = ObjectContexts[I];
2867 if (OptContext.Skip || !OptContext.File.Dwarf)
2868 return;
2870 // Then mark all the DIEs that need to be present in the generated output
2871 // and collect some information about them.
2872 // Note that this loop can not be merged with the previous one because
2873 // cross-cu references require the ParentIdx to be setup for every CU in
2874 // the object file before calling this.
2875 if (LLVM_UNLIKELY(Options.Update)) {
2876 for (auto &CurrentUnit : OptContext.CompileUnits)
2877 CurrentUnit->markEverythingAsKept();
2878 copyInvariantDebugSection(*OptContext.File.Dwarf);
2879 } else {
2880 for (auto &CurrentUnit : OptContext.CompileUnits) {
2881 lookForDIEsToKeep(*OptContext.File.Addresses, OptContext.CompileUnits,
2882 CurrentUnit->getOrigUnit().getUnitDIE(),
2883 OptContext.File, *CurrentUnit, 0);
2884 #ifndef NDEBUG
2885 verifyKeepChain(*CurrentUnit);
2886 #endif
2890 // The calls to applyValidRelocs inside cloneDIE will walk the reloc
2891 // array again (in the same way findValidRelocsInDebugInfo() did). We
2892 // need to reset the NextValidReloc index to the beginning.
2893 if (OptContext.File.Addresses->hasValidRelocs() ||
2894 LLVM_UNLIKELY(Options.Update)) {
2895 SizeByObject[OptContext.File.FileName].Input =
2896 getDebugInfoSize(*OptContext.File.Dwarf);
2897 SizeByObject[OptContext.File.FileName].Output =
2898 DIECloner(*this, TheDwarfEmitter, OptContext.File, DIEAlloc,
2899 OptContext.CompileUnits, Options.Update, DebugStrPool,
2900 DebugLineStrPool, StringOffsetPool)
2901 .cloneAllCompileUnits(*OptContext.File.Dwarf, OptContext.File,
2902 OptContext.File.Dwarf->isLittleEndian());
2904 if ((TheDwarfEmitter != nullptr) && !OptContext.CompileUnits.empty() &&
2905 LLVM_LIKELY(!Options.Update))
2906 patchFrameInfoForObject(OptContext);
2908 // Clean-up before starting working on the next object.
2909 cleanupAuxiliarryData(OptContext);
2912 auto EmitLambda = [&]() {
2913 // Emit everything that's global.
2914 if (TheDwarfEmitter != nullptr) {
2915 TheDwarfEmitter->emitAbbrevs(Abbreviations, Options.TargetDWARFVersion);
2916 TheDwarfEmitter->emitStrings(DebugStrPool);
2917 TheDwarfEmitter->emitStringOffsets(StringOffsetPool.getValues(),
2918 Options.TargetDWARFVersion);
2919 TheDwarfEmitter->emitLineStrings(DebugLineStrPool);
2920 for (AccelTableKind TableKind : Options.AccelTables) {
2921 switch (TableKind) {
2922 case AccelTableKind::Apple:
2923 TheDwarfEmitter->emitAppleNamespaces(AppleNamespaces);
2924 TheDwarfEmitter->emitAppleNames(AppleNames);
2925 TheDwarfEmitter->emitAppleTypes(AppleTypes);
2926 TheDwarfEmitter->emitAppleObjc(AppleObjc);
2927 break;
2928 case AccelTableKind::Pub:
2929 // Already emitted by emitAcceleratorEntriesForUnit.
2930 // Already emitted by emitAcceleratorEntriesForUnit.
2931 break;
2932 case AccelTableKind::DebugNames:
2933 TheDwarfEmitter->emitDebugNames(DebugNames);
2934 break;
2940 auto AnalyzeAll = [&]() {
2941 for (unsigned I = 0, E = NumObjects; I != E; ++I) {
2942 AnalyzeLambda(I);
2944 std::unique_lock<std::mutex> LockGuard(ProcessedFilesMutex);
2945 ProcessedFiles.set(I);
2946 ProcessedFilesConditionVariable.notify_one();
2950 auto CloneAll = [&]() {
2951 for (unsigned I = 0, E = NumObjects; I != E; ++I) {
2953 std::unique_lock<std::mutex> LockGuard(ProcessedFilesMutex);
2954 if (!ProcessedFiles[I]) {
2955 ProcessedFilesConditionVariable.wait(
2956 LockGuard, [&]() { return ProcessedFiles[I]; });
2960 CloneLambda(I);
2962 EmitLambda();
2965 // To limit memory usage in the single threaded case, analyze and clone are
2966 // run sequentially so the OptContext is freed after processing each object
2967 // in endDebugObject.
2968 if (Options.Threads == 1) {
2969 for (unsigned I = 0, E = NumObjects; I != E; ++I) {
2970 AnalyzeLambda(I);
2971 CloneLambda(I);
2973 EmitLambda();
2974 } else {
2975 DefaultThreadPool Pool(hardware_concurrency(2));
2976 Pool.async(AnalyzeAll);
2977 Pool.async(CloneAll);
2978 Pool.wait();
2981 if (Options.Statistics) {
2982 // Create a vector sorted in descending order by output size.
2983 std::vector<std::pair<StringRef, DebugInfoSize>> Sorted;
2984 for (auto &E : SizeByObject)
2985 Sorted.emplace_back(E.first(), E.second);
2986 llvm::sort(Sorted, [](auto &LHS, auto &RHS) {
2987 return LHS.second.Output > RHS.second.Output;
2990 auto ComputePercentange = [](int64_t Input, int64_t Output) -> float {
2991 const float Difference = Output - Input;
2992 const float Sum = Input + Output;
2993 if (Sum == 0)
2994 return 0;
2995 return (Difference / (Sum / 2));
2998 int64_t InputTotal = 0;
2999 int64_t OutputTotal = 0;
3000 const char *FormatStr = "{0,-45} {1,10}b {2,10}b {3,8:P}\n";
3002 // Print header.
3003 outs() << ".debug_info section size (in bytes)\n";
3004 outs() << "----------------------------------------------------------------"
3005 "---------------\n";
3006 outs() << "Filename Object "
3007 " dSYM Change\n";
3008 outs() << "----------------------------------------------------------------"
3009 "---------------\n";
3011 // Print body.
3012 for (auto &E : Sorted) {
3013 InputTotal += E.second.Input;
3014 OutputTotal += E.second.Output;
3015 llvm::outs() << formatv(
3016 FormatStr, sys::path::filename(E.first).take_back(45), E.second.Input,
3017 E.second.Output, ComputePercentange(E.second.Input, E.second.Output));
3019 // Print total and footer.
3020 outs() << "----------------------------------------------------------------"
3021 "---------------\n";
3022 llvm::outs() << formatv(FormatStr, "Total", InputTotal, OutputTotal,
3023 ComputePercentange(InputTotal, OutputTotal));
3024 outs() << "----------------------------------------------------------------"
3025 "---------------\n\n";
3028 return Error::success();
3031 Error DWARFLinker::cloneModuleUnit(LinkContext &Context, RefModuleUnit &Unit,
3032 DeclContextTree &ODRContexts,
3033 OffsetsStringPool &DebugStrPool,
3034 OffsetsStringPool &DebugLineStrPool,
3035 DebugDieValuePool &StringOffsetPool,
3036 unsigned Indent) {
3037 assert(Unit.Unit.get() != nullptr);
3039 if (!Unit.Unit->getOrigUnit().getUnitDIE().hasChildren())
3040 return Error::success();
3042 if (Options.Verbose) {
3043 outs().indent(Indent);
3044 outs() << "cloning .debug_info from " << Unit.File.FileName << "\n";
3047 // Analyze context for the module.
3048 analyzeContextInfo(Unit.Unit->getOrigUnit().getUnitDIE(), 0, *(Unit.Unit),
3049 &ODRContexts.getRoot(), ODRContexts, 0,
3050 Options.ParseableSwiftInterfaces,
3051 [&](const Twine &Warning, const DWARFDie &DIE) {
3052 reportWarning(Warning, Context.File, &DIE);
3054 // Keep everything.
3055 Unit.Unit->markEverythingAsKept();
3057 // Clone unit.
3058 UnitListTy CompileUnits;
3059 CompileUnits.emplace_back(std::move(Unit.Unit));
3060 assert(TheDwarfEmitter);
3061 DIECloner(*this, TheDwarfEmitter, Unit.File, DIEAlloc, CompileUnits,
3062 Options.Update, DebugStrPool, DebugLineStrPool, StringOffsetPool)
3063 .cloneAllCompileUnits(*Unit.File.Dwarf, Unit.File,
3064 Unit.File.Dwarf->isLittleEndian());
3065 return Error::success();
3068 void DWARFLinker::verifyInput(const DWARFFile &File) {
3069 assert(File.Dwarf);
3071 std::string Buffer;
3072 raw_string_ostream OS(Buffer);
3073 DIDumpOptions DumpOpts;
3074 if (!File.Dwarf->verify(OS, DumpOpts.noImplicitRecursion())) {
3075 if (Options.InputVerificationHandler)
3076 Options.InputVerificationHandler(File, OS.str());
3080 } // namespace llvm