1 //===- llvm/CodeGen/DwarfExpression.h - Dwarf Compile Unit ------*- 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 dwarf compile unit.
10 //
11 //===----------------------------------------------------------------------===//
12 
13 #ifndef LLVM_LIB_CODEGEN_ASMPRINTER_DWARFEXPRESSION_H
14 #define LLVM_LIB_CODEGEN_ASMPRINTER_DWARFEXPRESSION_H
15 
16 #include "ByteStreamer.h"
17 #include "llvm/ADT/ArrayRef.h"
18 #include "llvm/ADT/None.h"
19 #include "llvm/ADT/Optional.h"
20 #include "llvm/ADT/SmallVector.h"
21 #include "llvm/IR/DebugInfoMetadata.h"
22 #include <cassert>
23 #include <cstdint>
24 #include <iterator>
25 
26 namespace llvm {
27 
28 class AsmPrinter;
29 class APInt;
30 class DwarfCompileUnit;
31 class DIELoc;
32 class TargetRegisterInfo;
33 class MachineLocation;
34 
35 /// Holds a DIExpression and keeps track of how many operands have been consumed
36 /// so far.
37 class DIExpressionCursor {
38   DIExpression::expr_op_iterator Start, End;
39 
40 public:
41   DIExpressionCursor(const DIExpression *Expr) {
42     if (!Expr) {
43       assert(Start == End);
44       return;
45     }
46     Start = Expr->expr_op_begin();
47     End = Expr->expr_op_end();
48   }
49 
50   DIExpressionCursor(ArrayRef<uint64_t> Expr)
51       : Start(Expr.begin()), End(Expr.end()) {}
52 
53   DIExpressionCursor(const DIExpressionCursor &) = default;
54 
55   /// Consume one operation.
56   Optional<DIExpression::ExprOperand> take() {
57     if (Start == End)
58       return None;
59     return *(Start++);
60   }
61 
62   /// Consume N operations.
63   void consume(unsigned N) { std::advance(Start, N); }
64 
65   /// Return the current operation.
66   Optional<DIExpression::ExprOperand> peek() const {
67     if (Start == End)
68       return None;
69     return *(Start);
70   }
71 
72   /// Return the next operation.
73   Optional<DIExpression::ExprOperand> peekNext() const {
74     if (Start == End)
75       return None;
76 
77     auto Next = Start.getNext();
78     if (Next == End)
79       return None;
80 
81     return *Next;
82   }
83 
84   /// Determine whether there are any operations left in this expression.
85   operator bool() const { return Start != End; }
86 
87   DIExpression::expr_op_iterator begin() const { return Start; }
88   DIExpression::expr_op_iterator end() const { return End; }
89 
90   /// Retrieve the fragment information, if any.
91   Optional<DIExpression::FragmentInfo> getFragmentInfo() const {
92     return DIExpression::getFragmentInfo(Start, End);
93   }
94 };
95 
96 /// Base class containing the logic for constructing DWARF expressions
97 /// independently of whether they are emitted into a DIE or into a .debug_loc
98 /// entry.
99 ///
100 /// Some DWARF operations, e.g. DW_OP_entry_value, need to calculate the size
101 /// of a succeeding DWARF block before the latter is emitted to the output.
102 /// To handle such cases, data can conditionally be emitted to a temporary
103 /// buffer, which can later on be committed to the main output. The size of the
104 /// temporary buffer is queryable, allowing for the size of the data to be
105 /// emitted before the data is committed.
106 class DwarfExpression {
107 protected:
108   /// Holds information about all subregisters comprising a register location.
109   struct Register {
110     int DwarfRegNo;
111     unsigned SubRegSize;
112     const char *Comment;
113 
114     /// Create a full register, no extra DW_OP_piece operators necessary.
115     static Register createRegister(int RegNo, const char *Comment) {
116       return {RegNo, 0, Comment};
117     }
118 
119     /// Create a subregister that needs a DW_OP_piece operator with SizeInBits.
120     static Register createSubRegister(int RegNo, unsigned SizeInBits,
121                                       const char *Comment) {
122       return {RegNo, SizeInBits, Comment};
123     }
124 
125     bool isSubRegister() const { return SubRegSize; }
126   };
127 
128   /// Whether we are currently emitting an entry value operation.
129   bool IsEmittingEntryValue = false;
130 
131   DwarfCompileUnit &CU;
132 
133   /// The register location, if any.
134   SmallVector<Register, 2> DwarfRegs;
135 
136   /// Current Fragment Offset in Bits.
137   uint64_t OffsetInBits = 0;
138 
139   /// Sometimes we need to add a DW_OP_bit_piece to describe a subregister.
140   unsigned SubRegisterSizeInBits : 16;
141   unsigned SubRegisterOffsetInBits : 16;
142 
143   /// The kind of location description being produced.
144   enum { Unknown = 0, Register, Memory, Implicit };
145 
146   /// Additional location flags which may be combined with any location kind.
147   /// Currently, entry values are not supported for the Memory location kind.
148   enum { EntryValue = 1 << 0, Indirect = 1 << 1, CallSiteParamValue = 1 << 2 };
149 
150   unsigned LocationKind : 3;
151   unsigned SavedLocationKind : 3;
152   unsigned LocationFlags : 3;
153   unsigned DwarfVersion : 4;
154 
155 public:
156   /// Set the location (\p Loc) and \ref DIExpression (\p DIExpr) to describe.
157   void setLocation(const MachineLocation &Loc, const DIExpression *DIExpr);
158 
159   bool isUnknownLocation() const { return LocationKind == Unknown; }
160 
161   bool isMemoryLocation() const { return LocationKind == Memory; }
162 
163   bool isRegisterLocation() const { return LocationKind == Register; }
164 
165   bool isImplicitLocation() const { return LocationKind == Implicit; }
166 
167   bool isEntryValue() const { return LocationFlags & EntryValue; }
168 
169   bool isIndirect() const { return LocationFlags & Indirect; }
170 
171   bool isParameterValue() { return LocationFlags & CallSiteParamValue; }
172 
173   Optional<uint8_t> TagOffset;
174 
175 protected:
176   /// Push a DW_OP_piece / DW_OP_bit_piece for emitting later, if one is needed
177   /// to represent a subregister.
178   void setSubRegisterPiece(unsigned SizeInBits, unsigned OffsetInBits) {
179     assert(SizeInBits < 65536 && OffsetInBits < 65536);
180     SubRegisterSizeInBits = SizeInBits;
181     SubRegisterOffsetInBits = OffsetInBits;
182   }
183 
184   /// Add masking operations to stencil out a subregister.
185   void maskSubRegister();
186 
187   /// Output a dwarf operand and an optional assembler comment.
188   virtual void emitOp(uint8_t Op, const char *Comment = nullptr) = 0;
189 
190   /// Emit a raw signed value.
191   virtual void emitSigned(int64_t Value) = 0;
192 
193   /// Emit a raw unsigned value.
194   virtual void emitUnsigned(uint64_t Value) = 0;
195 
196   virtual void emitData1(uint8_t Value) = 0;
197 
198   virtual void emitBaseTypeRef(uint64_t Idx) = 0;
199 
200   /// Start emitting data to the temporary buffer. The data stored in the
201   /// temporary buffer can be committed to the main output using
202   /// commitTemporaryBuffer().
203   virtual void enableTemporaryBuffer() = 0;
204 
205   /// Disable emission to the temporary buffer. This does not commit data
206   /// in the temporary buffer to the main output.
207   virtual void disableTemporaryBuffer() = 0;
208 
209   /// Return the emitted size, in number of bytes, for the data stored in the
210   /// temporary buffer.
211   virtual unsigned getTemporaryBufferSize() = 0;
212 
213   /// Commit the data stored in the temporary buffer to the main output.
214   virtual void commitTemporaryBuffer() = 0;
215 
216   /// Emit a normalized unsigned constant.
217   void emitConstu(uint64_t Value);
218 
219   /// Return whether the given machine register is the frame register in the
220   /// current function.
221   virtual bool isFrameRegister(const TargetRegisterInfo &TRI,
222                                llvm::Register MachineReg) = 0;
223 
224   /// Emit a DW_OP_reg operation. Note that this is only legal inside a DWARF
225   /// register location description.
226   void addReg(int DwarfReg, const char *Comment = nullptr);
227 
228   /// Emit a DW_OP_breg operation.
229   void addBReg(int DwarfReg, int Offset);
230 
231   /// Emit DW_OP_fbreg <Offset>.
232   void addFBReg(int Offset);
233 
234   /// Emit a partial DWARF register operation.
235   ///
236   /// \param MachineReg           The register number.
237   /// \param MaxSize              If the register must be composed from
238   ///                             sub-registers this is an upper bound
239   ///                             for how many bits the emitted DW_OP_piece
240   ///                             may cover.
241   ///
242   /// If size and offset is zero an operation for the entire register is
243   /// emitted: Some targets do not provide a DWARF register number for every
244   /// register.  If this is the case, this function will attempt to emit a DWARF
245   /// register by emitting a fragment of a super-register or by piecing together
246   /// multiple subregisters that alias the register.
247   ///
248   /// \return false if no DWARF register exists for MachineReg.
249   bool addMachineReg(const TargetRegisterInfo &TRI, llvm::Register MachineReg,
250                      unsigned MaxSize = ~1U);
251 
252   /// Emit a DW_OP_piece or DW_OP_bit_piece operation for a variable fragment.
253   /// \param OffsetInBits    This is an optional offset into the location that
254   /// is at the top of the DWARF stack.
255   void addOpPiece(unsigned SizeInBits, unsigned OffsetInBits = 0);
256 
257   /// Emit a shift-right dwarf operation.
258   void addShr(unsigned ShiftBy);
259 
260   /// Emit a bitwise and dwarf operation.
261   void addAnd(unsigned Mask);
262 
263   /// Emit a DW_OP_stack_value, if supported.
264   ///
265   /// The proper way to describe a constant value is DW_OP_constu <const>,
266   /// DW_OP_stack_value.  Unfortunately, DW_OP_stack_value was not available
267   /// until DWARF 4, so we will continue to generate DW_OP_constu <const> for
268   /// DWARF 2 and DWARF 3. Technically, this is incorrect since DW_OP_const
269   /// <const> actually describes a value at a constant address, not a constant
270   /// value.  However, in the past there was no better way to describe a
271   /// constant value, so the producers and consumers started to rely on
272   /// heuristics to disambiguate the value vs. location status of the
273   /// expression.  See PR21176 for more details.
274   void addStackValue();
275 
276   /// Finalize an entry value by emitting its size operand, and committing the
277   /// DWARF block which has been emitted to the temporary buffer.
278   void finalizeEntryValue();
279 
280   /// Cancel the emission of an entry value.
281   void cancelEntryValue();
282 
283   ~DwarfExpression() = default;
284 
285 public:
286   DwarfExpression(unsigned DwarfVersion, DwarfCompileUnit &CU)
287       : CU(CU), SubRegisterSizeInBits(0), SubRegisterOffsetInBits(0),
288         LocationKind(Unknown), SavedLocationKind(Unknown),
289         LocationFlags(Unknown), DwarfVersion(DwarfVersion) {}
290 
291   /// This needs to be called last to commit any pending changes.
292   void finalize();
293 
294   /// Emit a signed constant.
295   void addSignedConstant(int64_t Value);
296 
297   /// Emit an unsigned constant.
298   void addUnsignedConstant(uint64_t Value);
299 
300   /// Emit an unsigned constant.
301   void addUnsignedConstant(const APInt &Value);
302 
303   /// Emit an floating point constant.
304   void addConstantFP(const APFloat &Value, const AsmPrinter &AP);
305 
306   /// Lock this down to become a memory location description.
307   void setMemoryLocationKind() {
308     assert(isUnknownLocation());
309     LocationKind = Memory;
310   }
311 
312   /// Lock this down to become an entry value location.
313   void setEntryValueFlags(const MachineLocation &Loc);
314 
315   /// Lock this down to become a call site parameter location.
316   void setCallSiteParamValueFlag() { LocationFlags |= CallSiteParamValue; }
317 
318   /// Emit a machine register location. As an optimization this may also consume
319   /// the prefix of a DwarfExpression if a more efficient representation for
320   /// combining the register location and the first operation exists.
321   ///
322   /// \param FragmentOffsetInBits     If this is one fragment out of a
323   /// fragmented
324   ///                                 location, this is the offset of the
325   ///                                 fragment inside the entire variable.
326   /// \return                         false if no DWARF register exists
327   ///                                 for MachineReg.
328   bool addMachineRegExpression(const TargetRegisterInfo &TRI,
329                                DIExpressionCursor &Expr,
330                                llvm::Register MachineReg,
331                                unsigned FragmentOffsetInBits = 0);
332 
333   /// Begin emission of an entry value dwarf operation. The entry value's
334   /// first operand is the size of the DWARF block (its second operand),
335   /// which needs to be calculated at time of emission, so we don't emit
336   /// any operands here.
337   void beginEntryValueExpression(DIExpressionCursor &ExprCursor);
338 
339   /// Return the index of a base type with the given properties and
340   /// create one if necessary.
341   unsigned getOrCreateBaseType(unsigned BitSize, dwarf::TypeKind Encoding);
342 
343   /// Emit all remaining operations in the DIExpressionCursor. The
344   /// cursor must not contain any DW_OP_LLVM_arg operations.
345   void addExpression(DIExpressionCursor &&Expr);
346 
347   /// Emit all remaining operations in the DIExpressionCursor.
348   /// DW_OP_LLVM_arg operations are resolved by calling (\p InsertArg).
349   //
350   /// \return false if any call to (\p InsertArg) returns false.
351   bool addExpression(
352       DIExpressionCursor &&Expr,
353       llvm::function_ref<bool(unsigned, DIExpressionCursor &)> InsertArg);
354 
355   /// If applicable, emit an empty DW_OP_piece / DW_OP_bit_piece to advance to
356   /// the fragment described by \c Expr.
357   void addFragmentOffset(const DIExpression *Expr);
358 
359   void emitLegacySExt(unsigned FromBits);
360   void emitLegacyZExt(unsigned FromBits);
361 
362   /// Emit location information expressed via WebAssembly location + offset
363   /// The Index is an identifier for locals, globals or operand stack.
364   void addWasmLocation(unsigned Index, uint64_t Offset);
365 };
366 
367 /// DwarfExpression implementation for .debug_loc entries.
368 class DebugLocDwarfExpression final : public DwarfExpression {
369 
370   struct TempBuffer {
371     SmallString<32> Bytes;
372     std::vector<std::string> Comments;
373     BufferByteStreamer BS;
374 
375     TempBuffer(bool GenerateComments) : BS(Bytes, Comments, GenerateComments) {}
376   };
377 
378   std::unique_ptr<TempBuffer> TmpBuf;
379   BufferByteStreamer &OutBS;
380   bool IsBuffering = false;
381 
382   /// Return the byte streamer that currently is being emitted to.
383   ByteStreamer &getActiveStreamer() { return IsBuffering ? TmpBuf->BS : OutBS; }
384 
385   void emitOp(uint8_t Op, const char *Comment = nullptr) override;
386   void emitSigned(int64_t Value) override;
387   void emitUnsigned(uint64_t Value) override;
388   void emitData1(uint8_t Value) override;
389   void emitBaseTypeRef(uint64_t Idx) override;
390 
391   void enableTemporaryBuffer() override;
392   void disableTemporaryBuffer() override;
393   unsigned getTemporaryBufferSize() override;
394   void commitTemporaryBuffer() override;
395 
396   bool isFrameRegister(const TargetRegisterInfo &TRI,
397                        llvm::Register MachineReg) override;
398 
399 public:
400   DebugLocDwarfExpression(unsigned DwarfVersion, BufferByteStreamer &BS,
401                           DwarfCompileUnit &CU)
402       : DwarfExpression(DwarfVersion, CU), OutBS(BS) {}
403 };
404 
405 /// DwarfExpression implementation for singular DW_AT_location.
406 class DIEDwarfExpression final : public DwarfExpression {
407   const AsmPrinter &AP;
408   DIELoc &OutDIE;
409   DIELoc TmpDIE;
410   bool IsBuffering = false;
411 
412   /// Return the DIE that currently is being emitted to.
413   DIELoc &getActiveDIE() { return IsBuffering ? TmpDIE : OutDIE; }
414 
415   void emitOp(uint8_t Op, const char *Comment = nullptr) override;
416   void emitSigned(int64_t Value) override;
417   void emitUnsigned(uint64_t Value) override;
418   void emitData1(uint8_t Value) override;
419   void emitBaseTypeRef(uint64_t Idx) override;
420 
421   void enableTemporaryBuffer() override;
422   void disableTemporaryBuffer() override;
423   unsigned getTemporaryBufferSize() override;
424   void commitTemporaryBuffer() override;
425 
426   bool isFrameRegister(const TargetRegisterInfo &TRI,
427                        llvm::Register MachineReg) override;
428 
429 public:
430   DIEDwarfExpression(const AsmPrinter &AP, DwarfCompileUnit &CU, DIELoc &DIE);
431 
432   DIELoc *finalize() {
433     DwarfExpression::finalize();
434     return &OutDIE;
435   }
436 };
437 
438 } // end namespace llvm
439 
440 #endif // LLVM_LIB_CODEGEN_ASMPRINTER_DWARFEXPRESSION_H
441