1//===- MCSymbol.h - Machine Code Symbols ------------------------*- 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 the declaration of the MCSymbol class.
10//
11//===----------------------------------------------------------------------===//
12
13#ifndef LLVM_MC_MCSYMBOL_H
14#define LLVM_MC_MCSYMBOL_H
15
16#include "llvm/ADT/PointerIntPair.h"
17#include "llvm/ADT/StringMap.h"
18#include "llvm/ADT/StringRef.h"
19#include "llvm/MC/MCExpr.h"
20#include "llvm/MC/MCFragment.h"
21#include "llvm/Support/ErrorHandling.h"
22#include "llvm/Support/MathExtras.h"
23#include <cassert>
24#include <cstddef>
25#include <cstdint>
26
27namespace llvm {
28
29class MCAsmInfo;
30class MCContext;
31class MCSection;
32class raw_ostream;
33
34/// MCSymbol - Instances of this class represent a symbol name in the MC file,
35/// and MCSymbols are created and uniqued by the MCContext class. MCSymbols
36/// should only be constructed with valid names for the object file.
37///
38/// If the symbol is defined/emitted into the current translation unit, the
39/// Section member is set to indicate what section it lives in. Otherwise, if
40/// it is a reference to an external entity, it has a null section.
41class MCSymbol {
42protected:
43 /// The kind of the symbol. If it is any value other than unset then this
44 /// class is actually one of the appropriate subclasses of MCSymbol.
45 enum SymbolKind {
46 SymbolKindUnset,
47 SymbolKindCOFF,
48 SymbolKindELF,
49 SymbolKindMachO,
50 SymbolKindWasm,
51 SymbolKindXCOFF,
52 };
53
54 /// A symbol can contain an Offset, or Value, or be Common, but never more
55 /// than one of these.
56 enum Contents : uint8_t {
57 SymContentsUnset,
58 SymContentsOffset,
59 SymContentsVariable,
60 SymContentsCommon,
61 SymContentsTargetCommon, // Index stores the section index
62 };
63
64 // Special sentinal value for the absolute pseudo fragment.
65 static MCFragment *AbsolutePseudoFragment;
66
67 /// If a symbol has a Fragment, the section is implied, so we only need
68 /// one pointer.
69 /// The special AbsolutePseudoFragment value is for absolute symbols.
70 /// If this is a variable symbol, this caches the variable value's fragment.
71 /// FIXME: We might be able to simplify this by having the asm streamer create
72 /// dummy fragments.
73 /// If this is a section, then it gives the symbol is defined in. This is null
74 /// for undefined symbols.
75 ///
76 /// If this is a fragment, then it gives the fragment this symbol's value is
77 /// relative to, if any.
78 ///
79 /// For the 'HasName' integer, this is true if this symbol is named.
80 /// A named symbol will have a pointer to the name allocated in the bytes
81 /// immediately prior to the MCSymbol.
82 mutable PointerIntPair<MCFragment *, 1> FragmentAndHasName;
83
84 /// IsTemporary - True if this is an assembler temporary label, which
85 /// typically does not survive in the .o file's symbol table. Usually
86 /// "Lfoo" or ".foo".
87 unsigned IsTemporary : 1;
88
89 /// True if this symbol can be redefined.
90 unsigned IsRedefinable : 1;
91
92 /// IsUsed - True if this symbol has been used.
93 mutable unsigned IsUsed : 1;
94
95 mutable unsigned IsRegistered : 1;
96
97 /// True if this symbol is visible outside this translation unit. Note: ELF
98 /// uses binding instead of this bit.
99 mutable unsigned IsExternal : 1;
100
101 /// This symbol is private extern.
102 mutable unsigned IsPrivateExtern : 1;
103
104 /// LLVM RTTI discriminator. This is actually a SymbolKind enumerator, but is
105 /// unsigned to avoid sign extension and achieve better bitpacking with MSVC.
106 unsigned Kind : 3;
107
108 /// True if we have created a relocation that uses this symbol.
109 mutable unsigned IsUsedInReloc : 1;
110
111 /// This is actually a Contents enumerator, but is unsigned to avoid sign
112 /// extension and achieve better bitpacking with MSVC.
113 unsigned SymbolContents : 3;
114
115 /// The alignment of the symbol, if it is 'common', or -1.
116 ///
117 /// The alignment is stored as log2(align) + 1. This allows all values from
118 /// 0 to 2^31 to be stored which is every power of 2 representable by an
119 /// unsigned.
120 enum : unsigned { NumCommonAlignmentBits = 5 };
121 unsigned CommonAlignLog2 : NumCommonAlignmentBits;
122
123 /// The Flags field is used by object file implementations to store
124 /// additional per symbol information which is not easily classified.
125 enum : unsigned { NumFlagsBits = 16 };
126 mutable uint32_t Flags : NumFlagsBits;
127
128 /// Index field, for use by the object file implementation.
129 mutable uint32_t Index = 0;
130
131 union {
132 /// The offset to apply to the fragment address to form this symbol's value.
133 uint64_t Offset;
134
135 /// The size of the symbol, if it is 'common'.
136 uint64_t CommonSize;
137
138 /// If non-null, the value for a variable symbol.
139 const MCExpr *Value;
140 };
141
142 // MCContext creates and uniques these.
143 friend class MCExpr;
144 friend class MCContext;
145
146 /// The name for a symbol.
147 /// MCSymbol contains a uint64_t so is probably aligned to 8. On a 32-bit
148 /// system, the name is a pointer so isn't going to satisfy the 8 byte
149 /// alignment of uint64_t. Account for that here.
150 using NameEntryStorageTy = union {
151 const StringMapEntry<bool> *NameEntry;
152 uint64_t AlignmentPadding;
153 };
154
155 MCSymbol(SymbolKind Kind, const StringMapEntry<bool> *Name, bool isTemporary)
156 : IsTemporary(isTemporary), IsRedefinable(false), IsUsed(false),
157 IsRegistered(false), IsExternal(false), IsPrivateExtern(false),
158 Kind(Kind), IsUsedInReloc(false), SymbolContents(SymContentsUnset),
159 CommonAlignLog2(0), Flags(0) {
160 Offset = 0;
161 FragmentAndHasName.setInt(!!Name);
162 if (Name)
163 getNameEntryPtr() = Name;
164 }
165
166 // Provide custom new/delete as we will only allocate space for a name
167 // if we need one.
168 void *operator new(size_t s, const StringMapEntry<bool> *Name,
169 MCContext &Ctx);
170
171private:
172 void operator delete(void *);
173 /// Placement delete - required by std, but never called.
174 void operator delete(void*, unsigned) {
175 llvm_unreachable("Constructor throws?");
176 }
177 /// Placement delete - required by std, but never called.
178 void operator delete(void*, unsigned, bool) {
179 llvm_unreachable("Constructor throws?");
180 }
181
182 /// Get a reference to the name field. Requires that we have a name
183 const StringMapEntry<bool> *&getNameEntryPtr() {
184 assert(FragmentAndHasName.getInt() && "Name is required");
185 NameEntryStorageTy *Name = reinterpret_cast<NameEntryStorageTy *>(this);
186 return (*(Name - 1)).NameEntry;
187 }
188 const StringMapEntry<bool> *&getNameEntryPtr() const {
189 return const_cast<MCSymbol*>(this)->getNameEntryPtr();
190 }
191
192public:
193 MCSymbol(const MCSymbol &) = delete;
194 MCSymbol &operator=(const MCSymbol &) = delete;
195
196 /// getName - Get the symbol name.
197 StringRef getName() const {
198 if (!FragmentAndHasName.getInt())
199 return StringRef();
200
201 return getNameEntryPtr()->first();
202 }
203
204 bool isRegistered() const { return IsRegistered; }
205 void setIsRegistered(bool Value) const { IsRegistered = Value; }
206
207 void setUsedInReloc() const { IsUsedInReloc = true; }
208 bool isUsedInReloc() const { return IsUsedInReloc; }
209
210 /// \name Accessors
211 /// @{
212
213 /// isTemporary - Check if this is an assembler temporary symbol.
214 bool isTemporary() const { return IsTemporary; }
215
216 /// isUsed - Check if this is used.
217 bool isUsed() const { return IsUsed; }
218
219 /// Check if this symbol is redefinable.
220 bool isRedefinable() const { return IsRedefinable; }
221 /// Mark this symbol as redefinable.
222 void setRedefinable(bool Value) { IsRedefinable = Value; }
223 /// Prepare this symbol to be redefined.
224 void redefineIfPossible() {
225 if (IsRedefinable) {
226 if (SymbolContents == SymContentsVariable) {
227 Value = nullptr;
228 SymbolContents = SymContentsUnset;
229 }
230 setUndefined();
231 IsRedefinable = false;
232 }
233 }
234
235 /// @}
236 /// \name Associated Sections
237 /// @{
238
239 /// isDefined - Check if this symbol is defined (i.e., it has an address).
240 ///
241 /// Defined symbols are either absolute or in some section.
242 bool isDefined() const { return !isUndefined(); }
243
244 /// isInSection - Check if this symbol is defined in some section (i.e., it
245 /// is defined but not absolute).
246 bool isInSection() const {
247 return isDefined() && !isAbsolute();
248 }
249
250 /// isUndefined - Check if this symbol undefined (i.e., implicitly defined).
251 bool isUndefined(bool SetUsed = true) const {
252 return getFragment(SetUsed) == nullptr;
253 }
254
255 /// isAbsolute - Check if this is an absolute symbol.
256 bool isAbsolute() const {
257 return getFragment() == AbsolutePseudoFragment;
258 }
259
260 /// Get the section associated with a defined, non-absolute symbol.
261 MCSection &getSection() const {
262 assert(isInSection() && "Invalid accessor!");
263 return *getFragment()->getParent();
264 }
265
266 /// Mark the symbol as defined in the fragment \p F.
267 void setFragment(MCFragment *F) const {
268 assert(!isVariable() && "Cannot set fragment of variable");
269 FragmentAndHasName.setPointer(F);
270 }
271
272 /// Mark the symbol as undefined.
273 void setUndefined() { FragmentAndHasName.setPointer(nullptr); }
274
275 bool isELF() const { return Kind == SymbolKindELF; }
276
277 bool isCOFF() const { return Kind == SymbolKindCOFF; }
278
279 bool isMachO() const { return Kind == SymbolKindMachO; }
280
281 bool isWasm() const { return Kind == SymbolKindWasm; }
282
283 bool isXCOFF() const { return Kind == SymbolKindXCOFF; }
284
285 /// @}
286 /// \name Variable Symbols
287 /// @{
288
289 /// isVariable - Check if this is a variable symbol.
290 bool isVariable() const {
291 return SymbolContents == SymContentsVariable;
292 }
293
294 /// getVariableValue - Get the value for variable symbols.
295 const MCExpr *getVariableValue(bool SetUsed = true) const {
296 assert(isVariable() && "Invalid accessor!");
297 IsUsed |= SetUsed;
298 return Value;
299 }
300
301 void setVariableValue(const MCExpr *Value);
302
303 /// @}
304
305 /// Get the (implementation defined) index.
306 uint32_t getIndex() const {
307 return Index;
308 }
309
310 /// Set the (implementation defined) index.
311 void setIndex(uint32_t Value) const {
312 Index = Value;
313 }
314
315 bool isUnset() const { return SymbolContents == SymContentsUnset; }
316
317 uint64_t getOffset() const {
318 assert((SymbolContents == SymContentsUnset ||
319 SymbolContents == SymContentsOffset) &&
320 "Cannot get offset for a common/variable symbol");
321 return Offset;
322 }
323 void setOffset(uint64_t Value) {
324 assert((SymbolContents == SymContentsUnset ||
325 SymbolContents == SymContentsOffset) &&
326 "Cannot set offset for a common/variable symbol");
327 Offset = Value;
328 SymbolContents = SymContentsOffset;
329 }
330
331 /// Return the size of a 'common' symbol.
332 uint64_t getCommonSize() const {
333 assert(isCommon() && "Not a 'common' symbol!");
334 return CommonSize;
335 }
336
337 /// Mark this symbol as being 'common'.
338 ///
339 /// \param Size - The size of the symbol.
340 /// \param Align - The alignment of the symbol.
341 /// \param Target - Is the symbol a target-specific common-like symbol.
342 void setCommon(uint64_t Size, unsigned Align, bool Target = false) {
343 assert(getOffset() == 0);
344 CommonSize = Size;
345 SymbolContents = Target ? SymContentsTargetCommon : SymContentsCommon;
346
347 assert((!Align || isPowerOf2_32(Align)) &&
348 "Alignment must be a power of 2");
349 unsigned Log2Align = Log2_32(Align) + 1;
350 assert(Log2Align < (1U << NumCommonAlignmentBits) &&
351 "Out of range alignment");
352 CommonAlignLog2 = Log2Align;
353 }
354
355 /// Return the alignment of a 'common' symbol.
356 unsigned getCommonAlignment() const {
357 assert(isCommon() && "Not a 'common' symbol!");
358 return CommonAlignLog2 ? (1U << (CommonAlignLog2 - 1)) : 0;
359 }
360
361 /// Declare this symbol as being 'common'.
362 ///
363 /// \param Size - The size of the symbol.
364 /// \param Align - The alignment of the symbol.
365 /// \param Target - Is the symbol a target-specific common-like symbol.
366 /// \return True if symbol was already declared as a different type
367 bool declareCommon(uint64_t Size, unsigned Align, bool Target = false) {
368 assert(isCommon() || getOffset() == 0);
369 if(isCommon()) {
370 if (CommonSize != Size || getCommonAlignment() != Align ||
371 isTargetCommon() != Target)
372 return true;
373 } else
374 setCommon(Size, Align, Target);
375 return false;
376 }
377
378 /// Is this a 'common' symbol.
379 bool isCommon() const {
380 return SymbolContents == SymContentsCommon ||
381 SymbolContents == SymContentsTargetCommon;
382 }
383
384 /// Is this a target-specific common-like symbol.
385 bool isTargetCommon() const {
386 return SymbolContents == SymContentsTargetCommon;
387 }
388
389 MCFragment *getFragment(bool SetUsed = true) const {
390 MCFragment *Fragment = FragmentAndHasName.getPointer();
391 if (Fragment || !isVariable())
392 return Fragment;
393 Fragment = getVariableValue(SetUsed)->findAssociatedFragment();
394 FragmentAndHasName.setPointer(Fragment);
395 return Fragment;
396 }
397
398 bool isExternal() const { return IsExternal; }
399 void setExternal(bool Value) const { IsExternal = Value; }
400
401 bool isPrivateExtern() const { return IsPrivateExtern; }
402 void setPrivateExtern(bool Value) { IsPrivateExtern = Value; }
403
404 /// print - Print the value to the stream \p OS.
405 void print(raw_ostream &OS, const MCAsmInfo *MAI) const;
406
407 /// dump - Print the value to stderr.
408 void dump() const;
409
410protected:
411 /// Get the (implementation defined) symbol flags.
412 uint32_t getFlags() const { return Flags; }
413
414 /// Set the (implementation defined) symbol flags.
415 void setFlags(uint32_t Value) const {
416 assert(Value < (1U << NumFlagsBits) && "Out of range flags");
417 Flags = Value;
418 }
419
420 /// Modify the flags via a mask
421 void modifyFlags(uint32_t Value, uint32_t Mask) const {
422 assert(Value < (1U << NumFlagsBits) && "Out of range flags");
423 Flags = (Flags & ~Mask) | Value;
424 }
425};
426
427inline raw_ostream &operator<<(raw_ostream &OS, const MCSymbol &Sym) {
428 Sym.print(OS, nullptr);
429 return OS;
430}
431
432} // end namespace llvm
433
434#endif // LLVM_MC_MCSYMBOL_H
435