LLVM 17.0.0git
CodeViewDebug.h
Go to the documentation of this file.
1//===- llvm/lib/CodeGen/AsmPrinter/CodeViewDebug.h --------------*- C++ -*-===//
2//
3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4// See https://llvm.org/LICENSE.txt for license information.
5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6//
7//===----------------------------------------------------------------------===//
8//
9// This file contains support for writing Microsoft CodeView debug info.
10//
11//===----------------------------------------------------------------------===//
12
13#ifndef LLVM_LIB_CODEGEN_ASMPRINTER_CODEVIEWDEBUG_H
14#define LLVM_LIB_CODEGEN_ASMPRINTER_CODEVIEWDEBUG_H
15
16#include "llvm/ADT/ArrayRef.h"
17#include "llvm/ADT/DenseMap.h"
18#include "llvm/ADT/DenseSet.h"
19#include "llvm/ADT/MapVector.h"
21#include "llvm/ADT/SetVector.h"
28#include "llvm/IR/DebugLoc.h"
31#include <cstdint>
32#include <map>
33#include <string>
34#include <tuple>
35#include <unordered_map>
36#include <utility>
37#include <vector>
38
39namespace llvm {
40
41struct ClassInfo;
42class StringRef;
43class AsmPrinter;
44class Function;
45class GlobalVariable;
46class MCSectionCOFF;
47class MCStreamer;
48class MCSymbol;
49class MachineFunction;
50
51/// Collects and handles line tables information in a CodeView format.
53public:
54 struct LocalVarDef {
55 /// Indicates that variable data is stored in memory relative to the
56 /// specified register.
57 int InMemory : 1;
58
59 /// Offset of variable data in memory.
60 int DataOffset : 31;
61
62 /// Non-zero if this is a piece of an aggregate.
64
65 /// Offset into aggregate.
67
68 /// Register containing the data or the register base of the memory
69 /// location containing the data.
71
73 uint64_t Val = 0;
74 std::memcpy(&Val, &DR, sizeof(Val));
75 return Val;
76 }
77
79 LocalVarDef DR;
80 std::memcpy(&DR, &Val, sizeof(Val));
81 return DR;
82 }
83 };
84
85 static_assert(sizeof(uint64_t) == sizeof(LocalVarDef));
86
87private:
88 MCStreamer &OS;
91
92 /// Whether to emit type record hashes into .debug$H.
93 bool EmitDebugGlobalHashes = false;
94
95 /// The codeview CPU type used by the translation unit.
96 codeview::CPUType TheCPU;
97
98 static LocalVarDef createDefRangeMem(uint16_t CVRegister, int Offset);
99
100 /// Similar to DbgVariable in DwarfDebug, but not dwarf-specific.
101 struct LocalVariable {
102 const DILocalVariable *DIVar = nullptr;
103 MapVector<LocalVarDef,
105 DefRanges;
106 bool UseReferenceType = false;
107 std::optional<APSInt> ConstantValue;
108 };
109
110 struct CVGlobalVariable {
111 const DIGlobalVariable *DIGV;
112 PointerUnion<const GlobalVariable *, const DIExpression *> GVInfo;
113 };
114
115 struct InlineSite {
116 SmallVector<LocalVariable, 1> InlinedLocals;
117 SmallVector<const DILocation *, 1> ChildSites;
118 const DISubprogram *Inlinee = nullptr;
119
120 /// The ID of the inline site or function used with .cv_loc. Not a type
121 /// index.
122 unsigned SiteFuncId = 0;
123 };
124
125 // Combines information from DILexicalBlock and LexicalScope.
126 struct LexicalBlock {
127 SmallVector<LocalVariable, 1> Locals;
128 SmallVector<CVGlobalVariable, 1> Globals;
129 SmallVector<LexicalBlock *, 1> Children;
130 const MCSymbol *Begin;
131 const MCSymbol *End;
132 StringRef Name;
133 };
134
135 // For each function, store a vector of labels to its instructions, as well as
136 // to the end of the function.
137 struct FunctionInfo {
138 FunctionInfo() = default;
139
140 // Uncopyable.
141 FunctionInfo(const FunctionInfo &FI) = delete;
142
143 /// Map from inlined call site to inlined instructions and child inlined
144 /// call sites. Listed in program order.
145 std::unordered_map<const DILocation *, InlineSite> InlineSites;
146
147 /// Ordered list of top-level inlined call sites.
148 SmallVector<const DILocation *, 1> ChildSites;
149
150 SmallVector<LocalVariable, 1> Locals;
151 SmallVector<CVGlobalVariable, 1> Globals;
152
153 std::unordered_map<const DILexicalBlockBase*, LexicalBlock> LexicalBlocks;
154
155 // Lexical blocks containing local variables.
156 SmallVector<LexicalBlock *, 1> ChildBlocks;
157
158 std::vector<std::pair<MCSymbol *, MDNode *>> Annotations;
159 std::vector<std::tuple<const MCSymbol *, const MCSymbol *, const DIType *>>
160 HeapAllocSites;
161
162 const MCSymbol *Begin = nullptr;
163 const MCSymbol *End = nullptr;
164 unsigned FuncId = 0;
165 unsigned LastFileId = 0;
166
167 /// Number of bytes allocated in the prologue for all local stack objects.
168 unsigned FrameSize = 0;
169
170 /// Number of bytes of parameters on the stack.
171 unsigned ParamSize = 0;
172
173 /// Number of bytes pushed to save CSRs.
174 unsigned CSRSize = 0;
175
176 /// Adjustment to apply on x86 when using the VFRAME frame pointer.
177 int OffsetAdjustment = 0;
178
179 /// Two-bit value indicating which register is the designated frame pointer
180 /// register for local variables. Included in S_FRAMEPROC.
181 codeview::EncodedFramePtrReg EncodedLocalFramePtrReg =
182 codeview::EncodedFramePtrReg::None;
183
184 /// Two-bit value indicating which register is the designated frame pointer
185 /// register for stack parameters. Included in S_FRAMEPROC.
186 codeview::EncodedFramePtrReg EncodedParamFramePtrReg =
187 codeview::EncodedFramePtrReg::None;
188
189 codeview::FrameProcedureOptions FrameProcOpts;
190
191 bool HasStackRealignment = false;
192
193 bool HaveLineInfo = false;
194 };
195 FunctionInfo *CurFn = nullptr;
196
197 codeview::SourceLanguage CurrentSourceLanguage =
199
200 // This map records the constant offset in DIExpression of the
201 // DIGlobalVariableExpression referencing the DIGlobalVariable.
202 DenseMap<const DIGlobalVariable *, uint64_t> CVGlobalVariableOffsets;
203
204 // Map used to seperate variables according to the lexical scope they belong
205 // in. This is populated by recordLocalVariable() before
206 // collectLexicalBlocks() separates the variables between the FunctionInfo
207 // and LexicalBlocks.
208 DenseMap<const LexicalScope *, SmallVector<LocalVariable, 1>> ScopeVariables;
209
210 // Map to separate global variables according to the lexical scope they
211 // belong in. A null local scope represents the global scope.
212 typedef SmallVector<CVGlobalVariable, 1> GlobalVariableList;
213 DenseMap<const DIScope*, std::unique_ptr<GlobalVariableList> > ScopeGlobals;
214
215 // Array of global variables which need to be emitted into a COMDAT section.
216 SmallVector<CVGlobalVariable, 1> ComdatVariables;
217
218 // Array of non-COMDAT global variables.
219 SmallVector<CVGlobalVariable, 1> GlobalVariables;
220
221 /// List of static const data members to be emitted as S_CONSTANTs.
222 SmallVector<const DIDerivedType *, 4> StaticConstMembers;
223
224 /// The set of comdat .debug$S sections that we've seen so far. Each section
225 /// must start with a magic version number that must only be emitted once.
226 /// This set tracks which sections we've already opened.
227 DenseSet<MCSectionCOFF *> ComdatDebugSections;
228
229 /// Switch to the appropriate .debug$S section for GVSym. If GVSym, the symbol
230 /// of an emitted global value, is in a comdat COFF section, this will switch
231 /// to a new .debug$S section in that comdat. This method ensures that the
232 /// section starts with the magic version number on first use. If GVSym is
233 /// null, uses the main .debug$S section.
234 void switchToDebugSectionForSymbol(const MCSymbol *GVSym);
235
236 /// The next available function index for use with our .cv_* directives. Not
237 /// to be confused with type indices for LF_FUNC_ID records.
238 unsigned NextFuncId = 0;
239
240 InlineSite &getInlineSite(const DILocation *InlinedAt,
241 const DISubprogram *Inlinee);
242
243 codeview::TypeIndex getFuncIdForSubprogram(const DISubprogram *SP);
244
245 void calculateRanges(LocalVariable &Var,
246 const DbgValueHistoryMap::Entries &Entries);
247
248 /// Remember some debug info about each function. Keep it in a stable order to
249 /// emit at the end of the TU.
250 MapVector<const Function *, std::unique_ptr<FunctionInfo>> FnDebugInfo;
251
252 /// Map from full file path to .cv_file id. Full paths are built from DIFiles
253 /// and are stored in FileToFilepathMap;
254 DenseMap<StringRef, unsigned> FileIdMap;
255
256 /// All inlined subprograms in the order they should be emitted.
257 SmallSetVector<const DISubprogram *, 4> InlinedSubprograms;
258
259 /// Map from a pair of DI metadata nodes and its DI type (or scope) that can
260 /// be nullptr, to CodeView type indices. Primarily indexed by
261 /// {DIType*, DIType*} and {DISubprogram*, DIType*}.
262 ///
263 /// The second entry in the key is needed for methods as DISubroutineType
264 /// representing static method type are shared with non-method function type.
265 DenseMap<std::pair<const DINode *, const DIType *>, codeview::TypeIndex>
266 TypeIndices;
267
268 /// Map from DICompositeType* to complete type index. Non-record types are
269 /// always looked up in the normal TypeIndices map.
270 DenseMap<const DICompositeType *, codeview::TypeIndex> CompleteTypeIndices;
271
272 /// Complete record types to emit after all active type lowerings are
273 /// finished.
274 SmallVector<const DICompositeType *, 4> DeferredCompleteTypes;
275
276 /// Number of type lowering frames active on the stack.
277 unsigned TypeEmissionLevel = 0;
278
279 codeview::TypeIndex VBPType;
280
281 const DISubprogram *CurrentSubprogram = nullptr;
282
283 // The UDTs we have seen while processing types; each entry is a pair of type
284 // index and type name.
285 std::vector<std::pair<std::string, const DIType *>> LocalUDTs;
286 std::vector<std::pair<std::string, const DIType *>> GlobalUDTs;
287
288 using FileToFilepathMapTy = std::map<const DIFile *, std::string>;
289 FileToFilepathMapTy FileToFilepathMap;
290
291 StringRef getFullFilepath(const DIFile *File);
292
293 unsigned maybeRecordFile(const DIFile *F);
294
295 void maybeRecordLocation(const DebugLoc &DL, const MachineFunction *MF);
296
297 void clear();
298
299 void setCurrentSubprogram(const DISubprogram *SP) {
300 CurrentSubprogram = SP;
301 LocalUDTs.clear();
302 }
303
304 /// Emit the magic version number at the start of a CodeView type or symbol
305 /// section. Appears at the front of every .debug$S or .debug$T or .debug$P
306 /// section.
307 void emitCodeViewMagicVersion();
308
309 void emitTypeInformation();
310
311 void emitTypeGlobalHashes();
312
313 void emitObjName();
314
315 void emitCompilerInformation();
316
317 void emitBuildInfo();
318
319 void emitInlineeLinesSubsection();
320
321 void emitDebugInfoForThunk(const Function *GV,
322 FunctionInfo &FI,
323 const MCSymbol *Fn);
324
325 void emitDebugInfoForFunction(const Function *GV, FunctionInfo &FI);
326
327 void emitDebugInfoForRetainedTypes();
328
329 void emitDebugInfoForUDTs(
330 const std::vector<std::pair<std::string, const DIType *>> &UDTs);
331
332 void collectDebugInfoForGlobals();
333 void emitDebugInfoForGlobals();
334 void emitGlobalVariableList(ArrayRef<CVGlobalVariable> Globals);
335 void emitConstantSymbolRecord(const DIType *DTy, APSInt &Value,
336 const std::string &QualifiedName);
337 void emitDebugInfoForGlobal(const CVGlobalVariable &CVGV);
338 void emitStaticConstMemberList();
339
340 /// Opens a subsection of the given kind in a .debug$S codeview section.
341 /// Returns an end label for use with endCVSubsection when the subsection is
342 /// finished.
343 MCSymbol *beginCVSubsection(codeview::DebugSubsectionKind Kind);
344 void endCVSubsection(MCSymbol *EndLabel);
345
346 /// Opens a symbol record of the given kind. Returns an end label for use with
347 /// endSymbolRecord.
348 MCSymbol *beginSymbolRecord(codeview::SymbolKind Kind);
349 void endSymbolRecord(MCSymbol *SymEnd);
350
351 /// Emits an S_END, S_INLINESITE_END, or S_PROC_ID_END record. These records
352 /// are empty, so we emit them with a simpler assembly sequence that doesn't
353 /// involve labels.
354 void emitEndSymbolRecord(codeview::SymbolKind EndKind);
355
356 void emitInlinedCallSite(const FunctionInfo &FI, const DILocation *InlinedAt,
357 const InlineSite &Site);
358
359 using InlinedEntity = DbgValueHistoryMap::InlinedEntity;
360
361 void collectGlobalVariableInfo();
362 void collectVariableInfo(const DISubprogram *SP);
363
364 void collectVariableInfoFromMFTable(DenseSet<InlinedEntity> &Processed);
365
366 // Construct the lexical block tree for a routine, pruning emptpy lexical
367 // scopes, and populate it with local variables.
368 void collectLexicalBlockInfo(SmallVectorImpl<LexicalScope *> &Scopes,
369 SmallVectorImpl<LexicalBlock *> &Blocks,
370 SmallVectorImpl<LocalVariable> &Locals,
371 SmallVectorImpl<CVGlobalVariable> &Globals);
372 void collectLexicalBlockInfo(LexicalScope &Scope,
373 SmallVectorImpl<LexicalBlock *> &ParentBlocks,
374 SmallVectorImpl<LocalVariable> &ParentLocals,
375 SmallVectorImpl<CVGlobalVariable> &ParentGlobals);
376
377 /// Records information about a local variable in the appropriate scope. In
378 /// particular, locals from inlined code live inside the inlining site.
379 void recordLocalVariable(LocalVariable &&Var, const LexicalScope *LS);
380
381 /// Emits local variables in the appropriate order.
382 void emitLocalVariableList(const FunctionInfo &FI,
383 ArrayRef<LocalVariable> Locals);
384
385 /// Emits an S_LOCAL record and its associated defined ranges.
386 void emitLocalVariable(const FunctionInfo &FI, const LocalVariable &Var);
387
388 /// Emits a sequence of lexical block scopes and their children.
389 void emitLexicalBlockList(ArrayRef<LexicalBlock *> Blocks,
390 const FunctionInfo& FI);
391
392 /// Emit a lexical block scope and its children.
393 void emitLexicalBlock(const LexicalBlock &Block, const FunctionInfo& FI);
394
395 /// Translates the DIType to codeview if necessary and returns a type index
396 /// for it.
397 codeview::TypeIndex getTypeIndex(const DIType *Ty,
398 const DIType *ClassTy = nullptr);
399
400 codeview::TypeIndex
401 getTypeIndexForThisPtr(const DIDerivedType *PtrTy,
402 const DISubroutineType *SubroutineTy);
403
404 codeview::TypeIndex getTypeIndexForReferenceTo(const DIType *Ty);
405
406 codeview::TypeIndex getMemberFunctionType(const DISubprogram *SP,
407 const DICompositeType *Class);
408
409 codeview::TypeIndex getScopeIndex(const DIScope *Scope);
410
411 codeview::TypeIndex getVBPTypeIndex();
412
413 void addToUDTs(const DIType *Ty);
414
415 void addUDTSrcLine(const DIType *Ty, codeview::TypeIndex TI);
416
417 codeview::TypeIndex lowerType(const DIType *Ty, const DIType *ClassTy);
418 codeview::TypeIndex lowerTypeAlias(const DIDerivedType *Ty);
419 codeview::TypeIndex lowerTypeArray(const DICompositeType *Ty);
420 codeview::TypeIndex lowerTypeString(const DIStringType *Ty);
421 codeview::TypeIndex lowerTypeBasic(const DIBasicType *Ty);
422 codeview::TypeIndex lowerTypePointer(
423 const DIDerivedType *Ty,
425 codeview::TypeIndex lowerTypeMemberPointer(
426 const DIDerivedType *Ty,
428 codeview::TypeIndex lowerTypeModifier(const DIDerivedType *Ty);
429 codeview::TypeIndex lowerTypeFunction(const DISubroutineType *Ty);
430 codeview::TypeIndex lowerTypeVFTableShape(const DIDerivedType *Ty);
431 codeview::TypeIndex lowerTypeMemberFunction(
432 const DISubroutineType *Ty, const DIType *ClassTy, int ThisAdjustment,
433 bool IsStaticMethod,
435 codeview::TypeIndex lowerTypeEnum(const DICompositeType *Ty);
436 codeview::TypeIndex lowerTypeClass(const DICompositeType *Ty);
437 codeview::TypeIndex lowerTypeUnion(const DICompositeType *Ty);
438
439 /// Symbol records should point to complete types, but type records should
440 /// always point to incomplete types to avoid cycles in the type graph. Only
441 /// use this entry point when generating symbol records. The complete and
442 /// incomplete type indices only differ for record types. All other types use
443 /// the same index.
444 codeview::TypeIndex getCompleteTypeIndex(const DIType *Ty);
445
446 codeview::TypeIndex lowerCompleteTypeClass(const DICompositeType *Ty);
447 codeview::TypeIndex lowerCompleteTypeUnion(const DICompositeType *Ty);
448
449 struct TypeLoweringScope;
450
451 void emitDeferredCompleteTypes();
452
453 void collectMemberInfo(ClassInfo &Info, const DIDerivedType *DDTy);
454 ClassInfo collectClassInfo(const DICompositeType *Ty);
455
456 /// Common record member lowering functionality for record types, which are
457 /// structs, classes, and unions. Returns the field list index and the member
458 /// count.
459 std::tuple<codeview::TypeIndex, codeview::TypeIndex, unsigned, bool>
460 lowerRecordFieldList(const DICompositeType *Ty);
461
462 /// Inserts {{Node, ClassTy}, TI} into TypeIndices and checks for duplicates.
463 codeview::TypeIndex recordTypeIndexForDINode(const DINode *Node,
464 codeview::TypeIndex TI,
465 const DIType *ClassTy = nullptr);
466
467 /// Collect the names of parent scopes, innermost to outermost. Return the
468 /// innermost subprogram scope if present. Ensure that parent type scopes are
469 /// inserted into the type table.
470 const DISubprogram *
471 collectParentScopeNames(const DIScope *Scope,
472 SmallVectorImpl<StringRef> &ParentScopeNames);
473 std::string getFullyQualifiedName(const DIScope *Scope, StringRef Name);
474 std::string getFullyQualifiedName(const DIScope *Scope);
475
476 unsigned getPointerSizeInBytes();
477
478protected:
479 /// Gather pre-function debug information.
480 void beginFunctionImpl(const MachineFunction *MF) override;
481
482 /// Gather post-function debug information.
483 void endFunctionImpl(const MachineFunction *) override;
484
485 /// Check if the current module is in Fortran.
487 return CurrentSourceLanguage == codeview::SourceLanguage::Fortran;
488 }
489
490public:
492
493 void beginModule(Module *M) override;
494
495 void setSymbolSize(const MCSymbol *, uint64_t) override {}
496
497 /// Emit the COFF section that holds the line table information.
498 void endModule() override;
499
500 /// Process beginning of an instruction.
501 void beginInstruction(const MachineInstr *MI) override;
502};
503
504template <> struct DenseMapInfo<CodeViewDebug::LocalVarDef> {
505
508 }
509
512 }
513
514 static unsigned getHashValue(const CodeViewDebug::LocalVarDef &DR) {
516 }
517
522 }
523};
524
525} // end namespace llvm
526
527#endif // LLVM_LIB_CODEGEN_ASMPRINTER_CODEVIEWDEBUG_H
MachineBasicBlock MachineBasicBlock::iterator DebugLoc DL
This file defines the BumpPtrAllocator interface.
Analysis containing CSE Info
Definition: CSEInfo.cpp:27
#define LLVM_LIBRARY_VISIBILITY
LLVM_LIBRARY_VISIBILITY - If a class marked with this attribute is linked into a shared library,...
Definition: Compiler.h:126
static void clear(coro::Shape &Shape)
Definition: Coroutines.cpp:149
This file defines the DenseMap class.
This file defines the DenseSet and SmallDenseSet classes.
std::string Name
IRTranslator LLVM IR MI
#define F(x, y, z)
Definition: MD5.cpp:55
This file implements a map that provides insertion order iteration.
This file defines the PointerUnion class, which is a discriminated union of pointer types.
Profile::FuncID FuncId
Definition: Profile.cpp:321
Basic Register Allocator
This file implements a set that has insertion order iteration characteristics.
This file defines the SmallVector class.
Value * RHS
Value * LHS
This class is intended to be used as a driving class for all asm writers.
Definition: AsmPrinter.h:84
Allocate memory in an ever growing pool, as if by bump-pointer.
Definition: Allocator.h:66
Collects and handles line tables information in a CodeView format.
Definition: CodeViewDebug.h:52
bool moduleIsInFortran()
Check if the current module is in Fortran.
void setSymbolSize(const MCSymbol *, uint64_t) override
For symbols that have a size designated (e.g.
SmallVector< Entry, 4 > Entries
std::pair< const DINode *, const DILocation * > InlinedEntity
Base class for debug information backends.
Streaming machine code generation interface.
Definition: MCStreamer.h:212
MCSymbol - Instances of this class represent a symbol name in the MC file, and MCSymbols are created ...
Definition: MCSymbol.h:41
Representation of each machine instruction.
Definition: MachineInstr.h:68
This class implements a map that also provides access to all stored values in a deterministic order.
Definition: MapVector.h:37
A Module instance is used to store all the information related to an LLVM module.
Definition: Module.h:65
This is a 'vector' (really, a variable-sized array), optimized for the case when the array is small.
Definition: SmallVector.h:1200
PointerOptions
Equivalent to misc lfPointerAttr bitfields.
Definition: CodeView.h:362
CPUType
These values correspond to the CV_CPU_TYPE_e enumeration, and are documented here: https://msdn....
Definition: CodeView.h:75
SymbolKind
Duplicate copy of the above enum, but using the official CV names.
Definition: CodeView.h:47
SourceLanguage
These values correspond to the CV_CFL_LANG enumeration, and are documented here: https://msdn....
Definition: CodeView.h:143
This is an optimization pass for GlobalISel generic memory operations.
Definition: AddressRanges.h:18
@ Offset
Definition: DWP.cpp:406
std::tuple< uint64_t, uint32_t > InlineSite
Definition: MCPseudoProbe.h:96
int DataOffset
Offset of variable data in memory.
Definition: CodeViewDebug.h:60
static uint64_t toOpaqueValue(const LocalVarDef DR)
Definition: CodeViewDebug.h:72
int InMemory
Indicates that variable data is stored in memory relative to the specified register.
Definition: CodeViewDebug.h:57
static LocalVarDef createFromOpaqueValue(uint64_t Val)
Definition: CodeViewDebug.h:78
uint16_t CVRegister
Register containing the data or the register base of the memory location containing the data.
Definition: CodeViewDebug.h:70
uint16_t StructOffset
Offset into aggregate.
Definition: CodeViewDebug.h:66
uint16_t IsSubfield
Non-zero if this is a piece of an aggregate.
Definition: CodeViewDebug.h:63
static bool isEqual(const CodeViewDebug::LocalVarDef &LHS, const CodeViewDebug::LocalVarDef &RHS)
static CodeViewDebug::LocalVarDef getEmptyKey()
static unsigned getHashValue(const CodeViewDebug::LocalVarDef &DR)
static CodeViewDebug::LocalVarDef getTombstoneKey()
An information struct used to provide DenseMap with the various necessary components for a given valu...
Definition: DenseMapInfo.h:51