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"
20#include "llvm/ADT/PointerUnion.h"
21#include "llvm/ADT/SetVector.h"
22#include "llvm/ADT/SmallVector.h"
23#include "llvm/CodeGen/DbgEntityHistoryCalculator.h"
24#include "llvm/CodeGen/DebugHandlerBase.h"
25#include "llvm/DebugInfo/CodeView/CodeView.h"
26#include "llvm/DebugInfo/CodeView/GlobalTypeTableBuilder.h"
27#include "llvm/DebugInfo/CodeView/TypeIndex.h"
28#include "llvm/IR/DebugLoc.h"
29#include "llvm/Support/Allocator.h"
30#include "llvm/Support/Compiler.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.
52class LLVM_LIBRARY_VISIBILITY CodeViewDebug : public DebugHandlerBase {
53  MCStreamer &OS;
54  BumpPtrAllocator Allocator;
55  codeview::GlobalTypeTableBuilder TypeTable;
56
57  /// Whether to emit type record hashes into .debug$H.
58  bool EmitDebugGlobalHashes = false;
59
60  /// The codeview CPU type used by the translation unit.
61  codeview::CPUType TheCPU;
62
63  /// Represents the most general definition range.
64  struct LocalVarDefRange {
65    /// Indicates that variable data is stored in memory relative to the
66    /// specified register.
67    int InMemory : 1;
68
69    /// Offset of variable data in memory.
70    int DataOffset : 31;
71
72    /// Non-zero if this is a piece of an aggregate.
73    uint16_t IsSubfield : 1;
74
75    /// Offset into aggregate.
76    uint16_t StructOffset : 15;
77
78    /// Register containing the data or the register base of the memory
79    /// location containing the data.
80    uint16_t CVRegister;
81
82    /// Compares all location fields. This includes all fields except the label
83    /// ranges.
84    bool isDifferentLocation(LocalVarDefRange &O) {
85      return InMemory != O.InMemory || DataOffset != O.DataOffset ||
86             IsSubfield != O.IsSubfield || StructOffset != O.StructOffset ||
87             CVRegister != O.CVRegister;
88    }
89
90    SmallVector<std::pair<const MCSymbol *, const MCSymbol *>, 1> Ranges;
91  };
92
93  static LocalVarDefRange createDefRangeMem(uint16_t CVRegister, int Offset);
94
95  /// Similar to DbgVariable in DwarfDebug, but not dwarf-specific.
96  struct LocalVariable {
97    const DILocalVariable *DIVar = nullptr;
98    SmallVector<LocalVarDefRange, 1> DefRanges;
99    bool UseReferenceType = false;
100  };
101
102  struct CVGlobalVariable {
103    const DIGlobalVariable *DIGV;
104    PointerUnion<const GlobalVariable *, const DIExpression *> GVInfo;
105  };
106
107  struct InlineSite {
108    SmallVector<LocalVariable, 1> InlinedLocals;
109    SmallVector<const DILocation *, 1> ChildSites;
110    const DISubprogram *Inlinee = nullptr;
111
112    /// The ID of the inline site or function used with .cv_loc. Not a type
113    /// index.
114    unsigned SiteFuncId = 0;
115  };
116
117  // Combines information from DILexicalBlock and LexicalScope.
118  struct LexicalBlock {
119    SmallVector<LocalVariable, 1> Locals;
120    SmallVector<CVGlobalVariable, 1> Globals;
121    SmallVector<LexicalBlock *, 1> Children;
122    const MCSymbol *Begin;
123    const MCSymbol *End;
124    StringRef Name;
125  };
126
127  // For each function, store a vector of labels to its instructions, as well as
128  // to the end of the function.
129  struct FunctionInfo {
130    FunctionInfo() = default;
131
132    // Uncopyable.
133    FunctionInfo(const FunctionInfo &FI) = delete;
134
135    /// Map from inlined call site to inlined instructions and child inlined
136    /// call sites. Listed in program order.
137    std::unordered_map<const DILocation *, InlineSite> InlineSites;
138
139    /// Ordered list of top-level inlined call sites.
140    SmallVector<const DILocation *, 1> ChildSites;
141
142    SmallVector<LocalVariable, 1> Locals;
143    SmallVector<CVGlobalVariable, 1> Globals;
144
145    std::unordered_map<const DILexicalBlockBase*, LexicalBlock> LexicalBlocks;
146
147    // Lexical blocks containing local variables.
148    SmallVector<LexicalBlock *, 1> ChildBlocks;
149
150    std::vector<std::pair<MCSymbol *, MDNode *>> Annotations;
151    std::vector<std::tuple<const MCSymbol *, const MCSymbol *, const DIType *>>
152        HeapAllocSites;
153
154    const MCSymbol *Begin = nullptr;
155    const MCSymbol *End = nullptr;
156    unsigned FuncId = 0;
157    unsigned LastFileId = 0;
158
159    /// Number of bytes allocated in the prologue for all local stack objects.
160    unsigned FrameSize = 0;
161
162    /// Number of bytes of parameters on the stack.
163    unsigned ParamSize = 0;
164
165    /// Number of bytes pushed to save CSRs.
166    unsigned CSRSize = 0;
167
168    /// Adjustment to apply on x86 when using the VFRAME frame pointer.
169    int OffsetAdjustment = 0;
170
171    /// Two-bit value indicating which register is the designated frame pointer
172    /// register for local variables. Included in S_FRAMEPROC.
173    codeview::EncodedFramePtrReg EncodedLocalFramePtrReg =
174        codeview::EncodedFramePtrReg::None;
175
176    /// Two-bit value indicating which register is the designated frame pointer
177    /// register for stack parameters. Included in S_FRAMEPROC.
178    codeview::EncodedFramePtrReg EncodedParamFramePtrReg =
179        codeview::EncodedFramePtrReg::None;
180
181    codeview::FrameProcedureOptions FrameProcOpts;
182
183    bool HasStackRealignment = false;
184
185    bool HaveLineInfo = false;
186  };
187  FunctionInfo *CurFn = nullptr;
188
189  // Map used to seperate variables according to the lexical scope they belong
190  // in.  This is populated by recordLocalVariable() before
191  // collectLexicalBlocks() separates the variables between the FunctionInfo
192  // and LexicalBlocks.
193  DenseMap<const LexicalScope *, SmallVector<LocalVariable, 1>> ScopeVariables;
194
195  // Map to separate global variables according to the lexical scope they
196  // belong in. A null local scope represents the global scope.
197  typedef SmallVector<CVGlobalVariable, 1> GlobalVariableList;
198  DenseMap<const DIScope*, std::unique_ptr<GlobalVariableList> > ScopeGlobals;
199
200  // Array of global variables which  need to be emitted into a COMDAT section.
201  SmallVector<CVGlobalVariable, 1> ComdatVariables;
202
203  // Array of non-COMDAT global variables.
204  SmallVector<CVGlobalVariable, 1> GlobalVariables;
205
206  /// The set of comdat .debug$S sections that we've seen so far. Each section
207  /// must start with a magic version number that must only be emitted once.
208  /// This set tracks which sections we've already opened.
209  DenseSet<MCSectionCOFF *> ComdatDebugSections;
210
211  /// Switch to the appropriate .debug$S section for GVSym. If GVSym, the symbol
212  /// of an emitted global value, is in a comdat COFF section, this will switch
213  /// to a new .debug$S section in that comdat. This method ensures that the
214  /// section starts with the magic version number on first use. If GVSym is
215  /// null, uses the main .debug$S section.
216  void switchToDebugSectionForSymbol(const MCSymbol *GVSym);
217
218  /// The next available function index for use with our .cv_* directives. Not
219  /// to be confused with type indices for LF_FUNC_ID records.
220  unsigned NextFuncId = 0;
221
222  InlineSite &getInlineSite(const DILocation *InlinedAt,
223                            const DISubprogram *Inlinee);
224
225  codeview::TypeIndex getFuncIdForSubprogram(const DISubprogram *SP);
226
227  void calculateRanges(LocalVariable &Var,
228                       const DbgValueHistoryMap::Entries &Entries);
229
230  static void collectInlineSiteChildren(SmallVectorImpl<unsigned> &Children,
231                                        const FunctionInfo &FI,
232                                        const InlineSite &Site);
233
234  /// Remember some debug info about each function. Keep it in a stable order to
235  /// emit at the end of the TU.
236  MapVector<const Function *, std::unique_ptr<FunctionInfo>> FnDebugInfo;
237
238  /// Map from full file path to .cv_file id. Full paths are built from DIFiles
239  /// and are stored in FileToFilepathMap;
240  DenseMap<StringRef, unsigned> FileIdMap;
241
242  /// All inlined subprograms in the order they should be emitted.
243  SmallSetVector<const DISubprogram *, 4> InlinedSubprograms;
244
245  /// Map from a pair of DI metadata nodes and its DI type (or scope) that can
246  /// be nullptr, to CodeView type indices. Primarily indexed by
247  /// {DIType*, DIType*} and {DISubprogram*, DIType*}.
248  ///
249  /// The second entry in the key is needed for methods as DISubroutineType
250  /// representing static method type are shared with non-method function type.
251  DenseMap<std::pair<const DINode *, const DIType *>, codeview::TypeIndex>
252      TypeIndices;
253
254  /// Map from DICompositeType* to complete type index. Non-record types are
255  /// always looked up in the normal TypeIndices map.
256  DenseMap<const DICompositeType *, codeview::TypeIndex> CompleteTypeIndices;
257
258  /// Complete record types to emit after all active type lowerings are
259  /// finished.
260  SmallVector<const DICompositeType *, 4> DeferredCompleteTypes;
261
262  /// Number of type lowering frames active on the stack.
263  unsigned TypeEmissionLevel = 0;
264
265  codeview::TypeIndex VBPType;
266
267  const DISubprogram *CurrentSubprogram = nullptr;
268
269  // The UDTs we have seen while processing types; each entry is a pair of type
270  // index and type name.
271  std::vector<std::pair<std::string, const DIType *>> LocalUDTs;
272  std::vector<std::pair<std::string, const DIType *>> GlobalUDTs;
273
274  using FileToFilepathMapTy = std::map<const DIFile *, std::string>;
275  FileToFilepathMapTy FileToFilepathMap;
276
277  StringRef getFullFilepath(const DIFile *File);
278
279  unsigned maybeRecordFile(const DIFile *F);
280
281  void maybeRecordLocation(const DebugLoc &DL, const MachineFunction *MF);
282
283  void clear();
284
285  void setCurrentSubprogram(const DISubprogram *SP) {
286    CurrentSubprogram = SP;
287    LocalUDTs.clear();
288  }
289
290  /// Emit the magic version number at the start of a CodeView type or symbol
291  /// section. Appears at the front of every .debug$S or .debug$T or .debug$P
292  /// section.
293  void emitCodeViewMagicVersion();
294
295  void emitTypeInformation();
296
297  void emitTypeGlobalHashes();
298
299  void emitCompilerInformation();
300
301  void emitBuildInfo();
302
303  void emitInlineeLinesSubsection();
304
305  void emitDebugInfoForThunk(const Function *GV,
306                             FunctionInfo &FI,
307                             const MCSymbol *Fn);
308
309  void emitDebugInfoForFunction(const Function *GV, FunctionInfo &FI);
310
311  void emitDebugInfoForRetainedTypes();
312
313  void
314  emitDebugInfoForUDTs(ArrayRef<std::pair<std::string, const DIType *>> UDTs);
315
316  void emitDebugInfoForGlobals();
317  void emitGlobalVariableList(ArrayRef<CVGlobalVariable> Globals);
318  void emitDebugInfoForGlobal(const CVGlobalVariable &CVGV);
319
320  /// Opens a subsection of the given kind in a .debug$S codeview section.
321  /// Returns an end label for use with endCVSubsection when the subsection is
322  /// finished.
323  MCSymbol *beginCVSubsection(codeview::DebugSubsectionKind Kind);
324  void endCVSubsection(MCSymbol *EndLabel);
325
326  /// Opens a symbol record of the given kind. Returns an end label for use with
327  /// endSymbolRecord.
328  MCSymbol *beginSymbolRecord(codeview::SymbolKind Kind);
329  void endSymbolRecord(MCSymbol *SymEnd);
330
331  /// Emits an S_END, S_INLINESITE_END, or S_PROC_ID_END record. These records
332  /// are empty, so we emit them with a simpler assembly sequence that doesn't
333  /// involve labels.
334  void emitEndSymbolRecord(codeview::SymbolKind EndKind);
335
336  void emitInlinedCallSite(const FunctionInfo &FI, const DILocation *InlinedAt,
337                           const InlineSite &Site);
338
339  using InlinedEntity = DbgValueHistoryMap::InlinedEntity;
340
341  void collectGlobalVariableInfo();
342  void collectVariableInfo(const DISubprogram *SP);
343
344  void collectVariableInfoFromMFTable(DenseSet<InlinedEntity> &Processed);
345
346  // Construct the lexical block tree for a routine, pruning emptpy lexical
347  // scopes, and populate it with local variables.
348  void collectLexicalBlockInfo(SmallVectorImpl<LexicalScope *> &Scopes,
349                               SmallVectorImpl<LexicalBlock *> &Blocks,
350                               SmallVectorImpl<LocalVariable> &Locals,
351                               SmallVectorImpl<CVGlobalVariable> &Globals);
352  void collectLexicalBlockInfo(LexicalScope &Scope,
353                               SmallVectorImpl<LexicalBlock *> &ParentBlocks,
354                               SmallVectorImpl<LocalVariable> &ParentLocals,
355                               SmallVectorImpl<CVGlobalVariable> &ParentGlobals);
356
357  /// Records information about a local variable in the appropriate scope. In
358  /// particular, locals from inlined code live inside the inlining site.
359  void recordLocalVariable(LocalVariable &&Var, const LexicalScope *LS);
360
361  /// Emits local variables in the appropriate order.
362  void emitLocalVariableList(const FunctionInfo &FI,
363                             ArrayRef<LocalVariable> Locals);
364
365  /// Emits an S_LOCAL record and its associated defined ranges.
366  void emitLocalVariable(const FunctionInfo &FI, const LocalVariable &Var);
367
368  /// Emits a sequence of lexical block scopes and their children.
369  void emitLexicalBlockList(ArrayRef<LexicalBlock *> Blocks,
370                            const FunctionInfo& FI);
371
372  /// Emit a lexical block scope and its children.
373  void emitLexicalBlock(const LexicalBlock &Block, const FunctionInfo& FI);
374
375  /// Translates the DIType to codeview if necessary and returns a type index
376  /// for it.
377  codeview::TypeIndex getTypeIndex(const DIType *Ty,
378                                   const DIType *ClassTy = nullptr);
379
380  codeview::TypeIndex
381  getTypeIndexForThisPtr(const DIDerivedType *PtrTy,
382                         const DISubroutineType *SubroutineTy);
383
384  codeview::TypeIndex getTypeIndexForReferenceTo(const DIType *Ty);
385
386  codeview::TypeIndex getMemberFunctionType(const DISubprogram *SP,
387                                            const DICompositeType *Class);
388
389  codeview::TypeIndex getScopeIndex(const DIScope *Scope);
390
391  codeview::TypeIndex getVBPTypeIndex();
392
393  void addToUDTs(const DIType *Ty);
394
395  void addUDTSrcLine(const DIType *Ty, codeview::TypeIndex TI);
396
397  codeview::TypeIndex lowerType(const DIType *Ty, const DIType *ClassTy);
398  codeview::TypeIndex lowerTypeAlias(const DIDerivedType *Ty);
399  codeview::TypeIndex lowerTypeArray(const DICompositeType *Ty);
400  codeview::TypeIndex lowerTypeBasic(const DIBasicType *Ty);
401  codeview::TypeIndex lowerTypePointer(
402      const DIDerivedType *Ty,
403      codeview::PointerOptions PO = codeview::PointerOptions::None);
404  codeview::TypeIndex lowerTypeMemberPointer(
405      const DIDerivedType *Ty,
406      codeview::PointerOptions PO = codeview::PointerOptions::None);
407  codeview::TypeIndex lowerTypeModifier(const DIDerivedType *Ty);
408  codeview::TypeIndex lowerTypeFunction(const DISubroutineType *Ty);
409  codeview::TypeIndex lowerTypeVFTableShape(const DIDerivedType *Ty);
410  codeview::TypeIndex lowerTypeMemberFunction(
411      const DISubroutineType *Ty, const DIType *ClassTy, int ThisAdjustment,
412      bool IsStaticMethod,
413      codeview::FunctionOptions FO = codeview::FunctionOptions::None);
414  codeview::TypeIndex lowerTypeEnum(const DICompositeType *Ty);
415  codeview::TypeIndex lowerTypeClass(const DICompositeType *Ty);
416  codeview::TypeIndex lowerTypeUnion(const DICompositeType *Ty);
417
418  /// Symbol records should point to complete types, but type records should
419  /// always point to incomplete types to avoid cycles in the type graph. Only
420  /// use this entry point when generating symbol records. The complete and
421  /// incomplete type indices only differ for record types. All other types use
422  /// the same index.
423  codeview::TypeIndex getCompleteTypeIndex(const DIType *Ty);
424
425  codeview::TypeIndex lowerCompleteTypeClass(const DICompositeType *Ty);
426  codeview::TypeIndex lowerCompleteTypeUnion(const DICompositeType *Ty);
427
428  struct TypeLoweringScope;
429
430  void emitDeferredCompleteTypes();
431
432  void collectMemberInfo(ClassInfo &Info, const DIDerivedType *DDTy);
433  ClassInfo collectClassInfo(const DICompositeType *Ty);
434
435  /// Common record member lowering functionality for record types, which are
436  /// structs, classes, and unions. Returns the field list index and the member
437  /// count.
438  std::tuple<codeview::TypeIndex, codeview::TypeIndex, unsigned, bool>
439  lowerRecordFieldList(const DICompositeType *Ty);
440
441  /// Inserts {{Node, ClassTy}, TI} into TypeIndices and checks for duplicates.
442  codeview::TypeIndex recordTypeIndexForDINode(const DINode *Node,
443                                               codeview::TypeIndex TI,
444                                               const DIType *ClassTy = nullptr);
445
446  unsigned getPointerSizeInBytes();
447
448protected:
449  /// Gather pre-function debug information.
450  void beginFunctionImpl(const MachineFunction *MF) override;
451
452  /// Gather post-function debug information.
453  void endFunctionImpl(const MachineFunction *) override;
454
455public:
456  CodeViewDebug(AsmPrinter *AP);
457
458  void setSymbolSize(const MCSymbol *, uint64_t) override {}
459
460  /// Emit the COFF section that holds the line table information.
461  void endModule() override;
462
463  /// Process beginning of an instruction.
464  void beginInstruction(const MachineInstr *MI) override;
465};
466
467} // end namespace llvm
468
469#endif // LLVM_LIB_CODEGEN_ASMPRINTER_CODEVIEWDEBUG_H
470