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