1 //===- MCSymbol.h - Machine Code Symbols ------------------------*- C++ -*-===//
2 //
3 //                     The LLVM Compiler Infrastructure
4 //
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
7 //
8 //===----------------------------------------------------------------------===//
9 //
10 // This file contains the declaration of the MCSymbol class.
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #ifndef LLVM_MC_MCSYMBOL_H
15 #define LLVM_MC_MCSYMBOL_H
16 
17 #include "llvm/ADT/PointerIntPair.h"
18 #include "llvm/ADT/PointerUnion.h"
19 #include "llvm/ADT/StringMap.h"
20 #include "llvm/MC/MCAssembler.h"
21 #include "llvm/Support/Compiler.h"
22 
23 namespace llvm {
24 class MCAsmInfo;
25 class MCExpr;
26 class MCSymbol;
27 class MCFragment;
28 class MCSection;
29 class MCContext;
30 class raw_ostream;
31 
32 /// MCSymbol - Instances of this class represent a symbol name in the MC file,
33 /// and MCSymbols are created and uniqued by the MCContext class.  MCSymbols
34 /// should only be constructed with valid names for the object file.
35 ///
36 /// If the symbol is defined/emitted into the current translation unit, the
37 /// Section member is set to indicate what section it lives in.  Otherwise, if
38 /// it is a reference to an external entity, it has a null section.
39 class MCSymbol {
40 protected:
41   /// The kind of the symbol.  If it is any value other than unset then this
42   /// class is actually one of the appropriate subclasses of MCSymbol.
43   enum SymbolKind {
44     SymbolKindUnset,
45     SymbolKindCOFF,
46     SymbolKindELF,
47     SymbolKindMachO,
48   };
49 
50   /// A symbol can contain an Offset, or Value, or be Common, but never more
51   /// than one of these.
52   enum Contents : uint8_t {
53     SymContentsUnset,
54     SymContentsOffset,
55     SymContentsVariable,
56     SymContentsCommon,
57   };
58 
59   // Special sentinal value for the absolute pseudo fragment.
60   static MCFragment *AbsolutePseudoFragment;
61 
62   /// If a symbol has a Fragment, the section is implied, so we only need
63   /// one pointer.
64   /// The special AbsolutePseudoFragment value is for absolute symbols.
65   /// If this is a variable symbol, this caches the variable value's fragment.
66   /// FIXME: We might be able to simplify this by having the asm streamer create
67   /// dummy fragments.
68   /// If this is a section, then it gives the symbol is defined in. This is null
69   /// for undefined symbols.
70   ///
71   /// If this is a fragment, then it gives the fragment this symbol's value is
72   /// relative to, if any.
73   ///
74   /// For the 'HasName' integer, this is true if this symbol is named.
75   /// A named symbol will have a pointer to the name allocated in the bytes
76   /// immediately prior to the MCSymbol.
77   mutable PointerIntPair<MCFragment *, 1> FragmentAndHasName;
78 
79   /// IsTemporary - True if this is an assembler temporary label, which
80   /// typically does not survive in the .o file's symbol table.  Usually
81   /// "Lfoo" or ".foo".
82   unsigned IsTemporary : 1;
83 
84   /// \brief True if this symbol can be redefined.
85   unsigned IsRedefinable : 1;
86 
87   /// IsUsed - True if this symbol has been used.
88   mutable unsigned IsUsed : 1;
89 
90   mutable bool IsRegistered : 1;
91 
92   /// This symbol is visible outside this translation unit.
93   mutable unsigned IsExternal : 1;
94 
95   /// This symbol is private extern.
96   mutable unsigned IsPrivateExtern : 1;
97 
98   /// LLVM RTTI discriminator. This is actually a SymbolKind enumerator, but is
99   /// unsigned to avoid sign extension and achieve better bitpacking with MSVC.
100   unsigned Kind : 2;
101 
102   /// True if we have created a relocation that uses this symbol.
103   mutable unsigned IsUsedInReloc : 1;
104 
105   /// This is actually a Contents enumerator, but is unsigned to avoid sign
106   /// extension and achieve better bitpacking with MSVC.
107   unsigned SymbolContents : 2;
108 
109   /// The alignment of the symbol, if it is 'common', or -1.
110   ///
111   /// The alignment is stored as log2(align) + 1.  This allows all values from
112   /// 0 to 2^31 to be stored which is every power of 2 representable by an
113   /// unsigned.
114   enum : unsigned { NumCommonAlignmentBits = 5 };
115   unsigned CommonAlignLog2 : NumCommonAlignmentBits;
116 
117   /// The Flags field is used by object file implementations to store
118   /// additional per symbol information which is not easily classified.
119   enum : unsigned { NumFlagsBits = 16 };
120   mutable uint32_t Flags : NumFlagsBits;
121 
122   /// Index field, for use by the object file implementation.
123   mutable uint32_t Index = 0;
124 
125   union {
126     /// The offset to apply to the fragment address to form this symbol's value.
127     uint64_t Offset;
128 
129     /// The size of the symbol, if it is 'common'.
130     uint64_t CommonSize;
131 
132     /// If non-null, the value for a variable symbol.
133     const MCExpr *Value;
134   };
135 
136 protected: // MCContext creates and uniques these.
137   friend class MCExpr;
138   friend class MCContext;
139 
140   /// \brief The name for a symbol.
141   /// MCSymbol contains a uint64_t so is probably aligned to 8.  On a 32-bit
142   /// system, the name is a pointer so isn't going to satisfy the 8 byte
143   /// alignment of uint64_t.  Account for that here.
144   typedef union {
145     const StringMapEntry<bool> *NameEntry;
146     uint64_t AlignmentPadding;
147   } NameEntryStorageTy;
148 
MCSymbol(SymbolKind Kind,const StringMapEntry<bool> * Name,bool isTemporary)149   MCSymbol(SymbolKind Kind, const StringMapEntry<bool> *Name, bool isTemporary)
150       : IsTemporary(isTemporary), IsRedefinable(false), IsUsed(false),
151         IsRegistered(false), IsExternal(false), IsPrivateExtern(false),
152         Kind(Kind), IsUsedInReloc(false), SymbolContents(SymContentsUnset),
153         CommonAlignLog2(0), Flags(0) {
154     Offset = 0;
155     FragmentAndHasName.setInt(!!Name);
156     if (Name)
157       getNameEntryPtr() = Name;
158   }
159 
160   // Provide custom new/delete as we will only allocate space for a name
161   // if we need one.
162   void *operator new(size_t s, const StringMapEntry<bool> *Name,
163                      MCContext &Ctx);
164 
165 private:
166 
167   void operator delete(void *);
168   /// \brief Placement delete - required by std, but never called.
delete(void *,unsigned)169   void operator delete(void*, unsigned) {
170     llvm_unreachable("Constructor throws?");
171   }
172   /// \brief Placement delete - required by std, but never called.
delete(void *,unsigned,bool)173   void operator delete(void*, unsigned, bool) {
174     llvm_unreachable("Constructor throws?");
175   }
176 
177   MCSymbol(const MCSymbol &) = delete;
178   void operator=(const MCSymbol &) = delete;
179   MCSection *getSectionPtr(bool SetUsed = true) const {
180     if (MCFragment *F = getFragment(SetUsed)) {
181       assert(F != AbsolutePseudoFragment);
182       return F->getParent();
183     }
184     return nullptr;
185   }
186 
187   /// \brief Get a reference to the name field.  Requires that we have a name
getNameEntryPtr()188   const StringMapEntry<bool> *&getNameEntryPtr() {
189     assert(FragmentAndHasName.getInt() && "Name is required");
190     NameEntryStorageTy *Name = reinterpret_cast<NameEntryStorageTy *>(this);
191     return (*(Name - 1)).NameEntry;
192   }
getNameEntryPtr()193   const StringMapEntry<bool> *&getNameEntryPtr() const {
194     return const_cast<MCSymbol*>(this)->getNameEntryPtr();
195   }
196 
197 public:
198   /// getName - Get the symbol name.
getName()199   StringRef getName() const {
200     if (!FragmentAndHasName.getInt())
201       return StringRef();
202 
203     return getNameEntryPtr()->first();
204   }
205 
isRegistered()206   bool isRegistered() const { return IsRegistered; }
setIsRegistered(bool Value)207   void setIsRegistered(bool Value) const { IsRegistered = Value; }
208 
setUsedInReloc()209   void setUsedInReloc() const { IsUsedInReloc = true; }
isUsedInReloc()210   bool isUsedInReloc() const { return IsUsedInReloc; }
211 
212   /// \name Accessors
213   /// @{
214 
215   /// isTemporary - Check if this is an assembler temporary symbol.
isTemporary()216   bool isTemporary() const { return IsTemporary; }
217 
218   /// isUsed - Check if this is used.
isUsed()219   bool isUsed() const { return IsUsed; }
setUsed(bool Value)220   void setUsed(bool Value) const { IsUsed |= Value; }
221 
222   /// \brief Check if this symbol is redefinable.
isRedefinable()223   bool isRedefinable() const { return IsRedefinable; }
224   /// \brief Mark this symbol as redefinable.
setRedefinable(bool Value)225   void setRedefinable(bool Value) { IsRedefinable = Value; }
226   /// \brief Prepare this symbol to be redefined.
redefineIfPossible()227   void redefineIfPossible() {
228     if (IsRedefinable) {
229       if (SymbolContents == SymContentsVariable) {
230         Value = nullptr;
231         SymbolContents = SymContentsUnset;
232       }
233       setUndefined();
234       IsRedefinable = false;
235     }
236   }
237 
238   /// @}
239   /// \name Associated Sections
240   /// @{
241 
242   /// isDefined - Check if this symbol is defined (i.e., it has an address).
243   ///
244   /// Defined symbols are either absolute or in some section.
245   bool isDefined(bool SetUsed = true) const {
246     return getFragment(SetUsed) != nullptr;
247   }
248 
249   /// isInSection - Check if this symbol is defined in some section (i.e., it
250   /// is defined but not absolute).
251   bool isInSection(bool SetUsed = true) const {
252     return isDefined(SetUsed) && !isAbsolute(SetUsed);
253   }
254 
255   /// isUndefined - Check if this symbol undefined (i.e., implicitly defined).
256   bool isUndefined(bool SetUsed = true) const { return !isDefined(SetUsed); }
257 
258   /// isAbsolute - Check if this is an absolute symbol.
259   bool isAbsolute(bool SetUsed = true) const {
260     return getFragment(SetUsed) == AbsolutePseudoFragment;
261   }
262 
263   /// Get the section associated with a defined, non-absolute symbol.
264   MCSection &getSection(bool SetUsed = true) const {
265     assert(isInSection(SetUsed) && "Invalid accessor!");
266     return *getSectionPtr(SetUsed);
267   }
268 
269   /// Mark the symbol as defined in the fragment \p F.
setFragment(MCFragment * F)270   void setFragment(MCFragment *F) const {
271     assert(!isVariable() && "Cannot set fragment of variable");
272     FragmentAndHasName.setPointer(F);
273   }
274 
275   /// Mark the symbol as undefined.
setUndefined()276   void setUndefined() { FragmentAndHasName.setPointer(nullptr); }
277 
isELF()278   bool isELF() const { return Kind == SymbolKindELF; }
279 
isCOFF()280   bool isCOFF() const { return Kind == SymbolKindCOFF; }
281 
isMachO()282   bool isMachO() const { return Kind == SymbolKindMachO; }
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 
getOffset()314   uint64_t getOffset() const {
315     assert((SymbolContents == SymContentsUnset ||
316             SymbolContents == SymContentsOffset) &&
317            "Cannot get offset for a common/variable symbol");
318     return Offset;
319   }
setOffset(uint64_t Value)320   void setOffset(uint64_t Value) {
321     assert((SymbolContents == SymContentsUnset ||
322             SymbolContents == SymContentsOffset) &&
323            "Cannot set offset for a common/variable symbol");
324     Offset = Value;
325     SymbolContents = SymContentsOffset;
326   }
327 
328   /// Return the size of a 'common' symbol.
getCommonSize()329   uint64_t getCommonSize() const {
330     assert(isCommon() && "Not a 'common' symbol!");
331     return CommonSize;
332   }
333 
334   /// Mark this symbol as being 'common'.
335   ///
336   /// \param Size - The size of the symbol.
337   /// \param Align - The alignment of the symbol.
setCommon(uint64_t Size,unsigned Align)338   void setCommon(uint64_t Size, unsigned Align) {
339     assert(getOffset() == 0);
340     CommonSize = Size;
341     SymbolContents = SymContentsCommon;
342 
343     assert((!Align || isPowerOf2_32(Align)) &&
344            "Alignment must be a power of 2");
345     unsigned Log2Align = Log2_32(Align) + 1;
346     assert(Log2Align < (1U << NumCommonAlignmentBits) &&
347            "Out of range alignment");
348     CommonAlignLog2 = Log2Align;
349   }
350 
351   ///  Return the alignment of a 'common' symbol.
getCommonAlignment()352   unsigned getCommonAlignment() const {
353     assert(isCommon() && "Not a 'common' symbol!");
354     return CommonAlignLog2 ? (1U << (CommonAlignLog2 - 1)) : 0;
355   }
356 
357   /// Declare this symbol as being 'common'.
358   ///
359   /// \param Size - The size of the symbol.
360   /// \param Align - The alignment of the symbol.
361   /// \return True if symbol was already declared as a different type
declareCommon(uint64_t Size,unsigned Align)362   bool declareCommon(uint64_t Size, unsigned Align) {
363     assert(isCommon() || getOffset() == 0);
364     if(isCommon()) {
365       if(CommonSize != Size || getCommonAlignment() != Align)
366        return true;
367     } else
368       setCommon(Size, Align);
369     return false;
370   }
371 
372   /// Is this a 'common' symbol.
isCommon()373   bool isCommon() const {
374     return SymbolContents == SymContentsCommon;
375   }
376 
377   MCFragment *getFragment(bool SetUsed = true) const {
378     MCFragment *Fragment = FragmentAndHasName.getPointer();
379     if (Fragment || !isVariable())
380       return Fragment;
381     Fragment = getVariableValue(SetUsed)->findAssociatedFragment();
382     FragmentAndHasName.setPointer(Fragment);
383     return Fragment;
384   }
385 
isExternal()386   bool isExternal() const { return IsExternal; }
setExternal(bool Value)387   void setExternal(bool Value) const { IsExternal = Value; }
388 
isPrivateExtern()389   bool isPrivateExtern() const { return IsPrivateExtern; }
setPrivateExtern(bool Value)390   void setPrivateExtern(bool Value) { IsPrivateExtern = Value; }
391 
392   /// print - Print the value to the stream \p OS.
393   void print(raw_ostream &OS, const MCAsmInfo *MAI) const;
394 
395   /// dump - Print the value to stderr.
396   void dump() const;
397 
398 protected:
399   /// Get the (implementation defined) symbol flags.
getFlags()400   uint32_t getFlags() const { return Flags; }
401 
402   /// Set the (implementation defined) symbol flags.
setFlags(uint32_t Value)403   void setFlags(uint32_t Value) const {
404     assert(Value < (1U << NumFlagsBits) && "Out of range flags");
405     Flags = Value;
406   }
407 
408   /// Modify the flags via a mask
modifyFlags(uint32_t Value,uint32_t Mask)409   void modifyFlags(uint32_t Value, uint32_t Mask) const {
410     assert(Value < (1U << NumFlagsBits) && "Out of range flags");
411     Flags = (Flags & ~Mask) | Value;
412   }
413 };
414 
415 inline raw_ostream &operator<<(raw_ostream &OS, const MCSymbol &Sym) {
416   Sym.print(OS, nullptr);
417   return OS;
418 }
419 } // end namespace llvm
420 
421 #endif
422