1 //===-- RuntimeDyldImpl.h - Run-time dynamic linker for MC-JIT --*- 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 // Interface for the implementations of runtime dynamic linker facilities.
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #ifndef LLVM_LIB_EXECUTIONENGINE_RUNTIMEDYLD_RUNTIMEDYLDIMPL_H
15 #define LLVM_LIB_EXECUTIONENGINE_RUNTIMEDYLD_RUNTIMEDYLDIMPL_H
16 
17 #include "llvm/ADT/DenseMap.h"
18 #include "llvm/ADT/SmallVector.h"
19 #include "llvm/ADT/StringMap.h"
20 #include "llvm/ADT/Triple.h"
21 #include "llvm/ExecutionEngine/RTDyldMemoryManager.h"
22 #include "llvm/ExecutionEngine/RuntimeDyld.h"
23 #include "llvm/ExecutionEngine/RuntimeDyldChecker.h"
24 #include "llvm/Object/ObjectFile.h"
25 #include "llvm/Support/Debug.h"
26 #include "llvm/Support/ErrorHandling.h"
27 #include "llvm/Support/Format.h"
28 #include "llvm/Support/Host.h"
29 #include "llvm/Support/Mutex.h"
30 #include "llvm/Support/SwapByteOrder.h"
31 #include "llvm/Support/raw_ostream.h"
32 #include <map>
33 #include <system_error>
34 
35 using namespace llvm;
36 using namespace llvm::object;
37 
38 namespace llvm {
39 
40   // Helper for extensive error checking in debug builds.
Check(std::error_code Err)41 inline std::error_code Check(std::error_code Err) {
42   if (Err) {
43     report_fatal_error(Err.message());
44   }
45   return Err;
46 }
47 
48 class Twine;
49 
50 /// SectionEntry - represents a section emitted into memory by the dynamic
51 /// linker.
52 class SectionEntry {
53 public:
54   /// Name - section name.
55   std::string Name;
56 
57   /// Address - address in the linker's memory where the section resides.
58   uint8_t *Address;
59 
60   /// Size - section size. Doesn't include the stubs.
61   size_t Size;
62 
63   /// LoadAddress - the address of the section in the target process's memory.
64   /// Used for situations in which JIT-ed code is being executed in the address
65   /// space of a separate process.  If the code executes in the same address
66   /// space where it was JIT-ed, this just equals Address.
67   uint64_t LoadAddress;
68 
69   /// StubOffset - used for architectures with stub functions for far
70   /// relocations (like ARM).
71   uintptr_t StubOffset;
72 
73   /// ObjAddress - address of the section in the in-memory object file.  Used
74   /// for calculating relocations in some object formats (like MachO).
75   uintptr_t ObjAddress;
76 
SectionEntry(StringRef name,uint8_t * address,size_t size,uintptr_t objAddress)77   SectionEntry(StringRef name, uint8_t *address, size_t size,
78                uintptr_t objAddress)
79       : Name(name), Address(address), Size(size),
80         LoadAddress(reinterpret_cast<uintptr_t>(address)), StubOffset(size),
81         ObjAddress(objAddress) {}
82 };
83 
84 /// RelocationEntry - used to represent relocations internally in the dynamic
85 /// linker.
86 class RelocationEntry {
87 public:
88   /// SectionID - the section this relocation points to.
89   unsigned SectionID;
90 
91   /// Offset - offset into the section.
92   uint64_t Offset;
93 
94   /// RelType - relocation type.
95   uint32_t RelType;
96 
97   /// Addend - the relocation addend encoded in the instruction itself.  Also
98   /// used to make a relocation section relative instead of symbol relative.
99   int64_t Addend;
100 
101   struct SectionPair {
102       uint32_t SectionA;
103       uint32_t SectionB;
104   };
105 
106   /// SymOffset - Section offset of the relocation entry's symbol (used for GOT
107   /// lookup).
108   union {
109     uint64_t SymOffset;
110     SectionPair Sections;
111   };
112 
113   /// True if this is a PCRel relocation (MachO specific).
114   bool IsPCRel;
115 
116   /// The size of this relocation (MachO specific).
117   unsigned Size;
118 
RelocationEntry(unsigned id,uint64_t offset,uint32_t type,int64_t addend)119   RelocationEntry(unsigned id, uint64_t offset, uint32_t type, int64_t addend)
120       : SectionID(id), Offset(offset), RelType(type), Addend(addend),
121         SymOffset(0), IsPCRel(false), Size(0) {}
122 
RelocationEntry(unsigned id,uint64_t offset,uint32_t type,int64_t addend,uint64_t symoffset)123   RelocationEntry(unsigned id, uint64_t offset, uint32_t type, int64_t addend,
124                   uint64_t symoffset)
125       : SectionID(id), Offset(offset), RelType(type), Addend(addend),
126         SymOffset(symoffset), IsPCRel(false), Size(0) {}
127 
RelocationEntry(unsigned id,uint64_t offset,uint32_t type,int64_t addend,bool IsPCRel,unsigned Size)128   RelocationEntry(unsigned id, uint64_t offset, uint32_t type, int64_t addend,
129                   bool IsPCRel, unsigned Size)
130       : SectionID(id), Offset(offset), RelType(type), Addend(addend),
131         SymOffset(0), IsPCRel(IsPCRel), Size(Size) {}
132 
RelocationEntry(unsigned id,uint64_t offset,uint32_t type,int64_t addend,unsigned SectionA,uint64_t SectionAOffset,unsigned SectionB,uint64_t SectionBOffset,bool IsPCRel,unsigned Size)133   RelocationEntry(unsigned id, uint64_t offset, uint32_t type, int64_t addend,
134                   unsigned SectionA, uint64_t SectionAOffset, unsigned SectionB,
135                   uint64_t SectionBOffset, bool IsPCRel, unsigned Size)
136       : SectionID(id), Offset(offset), RelType(type),
137         Addend(SectionAOffset - SectionBOffset + addend), IsPCRel(IsPCRel),
138         Size(Size) {
139     Sections.SectionA = SectionA;
140     Sections.SectionB = SectionB;
141   }
142 };
143 
144 class RelocationValueRef {
145 public:
146   unsigned SectionID;
147   uint64_t Offset;
148   int64_t Addend;
149   const char *SymbolName;
RelocationValueRef()150   RelocationValueRef() : SectionID(0), Offset(0), Addend(0),
151                          SymbolName(nullptr) {}
152 
153   inline bool operator==(const RelocationValueRef &Other) const {
154     return SectionID == Other.SectionID && Offset == Other.Offset &&
155            Addend == Other.Addend && SymbolName == Other.SymbolName;
156   }
157   inline bool operator<(const RelocationValueRef &Other) const {
158     if (SectionID != Other.SectionID)
159       return SectionID < Other.SectionID;
160     if (Offset != Other.Offset)
161       return Offset < Other.Offset;
162     if (Addend != Other.Addend)
163       return Addend < Other.Addend;
164     return SymbolName < Other.SymbolName;
165   }
166 };
167 
168 /// @brief Symbol info for RuntimeDyld.
169 class SymbolTableEntry : public JITSymbolBase {
170 public:
SymbolTableEntry()171   SymbolTableEntry()
172     : JITSymbolBase(JITSymbolFlags::None), Offset(0), SectionID(0) {}
173 
SymbolTableEntry(unsigned SectionID,uint64_t Offset,JITSymbolFlags Flags)174   SymbolTableEntry(unsigned SectionID, uint64_t Offset, JITSymbolFlags Flags)
175     : JITSymbolBase(Flags), Offset(Offset), SectionID(SectionID) {}
176 
getSectionID()177   unsigned getSectionID() const { return SectionID; }
getOffset()178   uint64_t getOffset() const { return Offset; }
179 
180 private:
181   uint64_t Offset;
182   unsigned SectionID;
183 };
184 
185 typedef StringMap<SymbolTableEntry> RTDyldSymbolTable;
186 
187 class RuntimeDyldImpl {
188   friend class RuntimeDyld::LoadedObjectInfo;
189   friend class RuntimeDyldCheckerImpl;
190 protected:
191   // The MemoryManager to load objects into.
192   RuntimeDyld::MemoryManager &MemMgr;
193 
194   // The symbol resolver to use for external symbols.
195   RuntimeDyld::SymbolResolver &Resolver;
196 
197   // Attached RuntimeDyldChecker instance. Null if no instance attached.
198   RuntimeDyldCheckerImpl *Checker;
199 
200   // A list of all sections emitted by the dynamic linker.  These sections are
201   // referenced in the code by means of their index in this list - SectionID.
202   typedef SmallVector<SectionEntry, 64> SectionList;
203   SectionList Sections;
204 
205   typedef unsigned SID; // Type for SectionIDs
206 #define RTDYLD_INVALID_SECTION_ID ((SID)(-1))
207 
208   // Keep a map of sections from object file to the SectionID which
209   // references it.
210   typedef std::map<SectionRef, unsigned> ObjSectionToIDMap;
211 
212   // A global symbol table for symbols from all loaded modules.
213   RTDyldSymbolTable GlobalSymbolTable;
214 
215   // Keep a map of common symbols to their info pairs
216   typedef std::vector<SymbolRef> CommonSymbolList;
217 
218   // For each symbol, keep a list of relocations based on it. Anytime
219   // its address is reassigned (the JIT re-compiled the function, e.g.),
220   // the relocations get re-resolved.
221   // The symbol (or section) the relocation is sourced from is the Key
222   // in the relocation list where it's stored.
223   typedef SmallVector<RelocationEntry, 64> RelocationList;
224   // Relocations to sections already loaded. Indexed by SectionID which is the
225   // source of the address. The target where the address will be written is
226   // SectionID/Offset in the relocation itself.
227   DenseMap<unsigned, RelocationList> Relocations;
228 
229   // Relocations to external symbols that are not yet resolved.  Symbols are
230   // external when they aren't found in the global symbol table of all loaded
231   // modules.  This map is indexed by symbol name.
232   StringMap<RelocationList> ExternalSymbolRelocations;
233 
234 
235   typedef std::map<RelocationValueRef, uintptr_t> StubMap;
236 
237   Triple::ArchType Arch;
238   bool IsTargetLittleEndian;
239 
240   // True if all sections should be passed to the memory manager, false if only
241   // sections containing relocations should be. Defaults to 'false'.
242   bool ProcessAllSections;
243 
244   // This mutex prevents simultaneously loading objects from two different
245   // threads.  This keeps us from having to protect individual data structures
246   // and guarantees that section allocation requests to the memory manager
247   // won't be interleaved between modules.  It is also used in mapSectionAddress
248   // and resolveRelocations to protect write access to internal data structures.
249   //
250   // loadObject may be called on the same thread during the handling of of
251   // processRelocations, and that's OK.  The handling of the relocation lists
252   // is written in such a way as to work correctly if new elements are added to
253   // the end of the list while the list is being processed.
254   sys::Mutex lock;
255 
256   virtual unsigned getMaxStubSize() = 0;
257   virtual unsigned getStubAlignment() = 0;
258 
259   bool HasError;
260   std::string ErrorStr;
261 
262   // Set the error state and record an error string.
Error(const Twine & Msg)263   bool Error(const Twine &Msg) {
264     ErrorStr = Msg.str();
265     HasError = true;
266     return true;
267   }
268 
getSectionLoadAddress(unsigned SectionID)269   uint64_t getSectionLoadAddress(unsigned SectionID) const {
270     return Sections[SectionID].LoadAddress;
271   }
272 
getSectionAddress(unsigned SectionID)273   uint8_t *getSectionAddress(unsigned SectionID) const {
274     return (uint8_t *)Sections[SectionID].Address;
275   }
276 
writeInt16BE(uint8_t * Addr,uint16_t Value)277   void writeInt16BE(uint8_t *Addr, uint16_t Value) {
278     if (IsTargetLittleEndian)
279       sys::swapByteOrder(Value);
280     *Addr       = (Value >> 8) & 0xFF;
281     *(Addr + 1) = Value & 0xFF;
282   }
283 
writeInt32BE(uint8_t * Addr,uint32_t Value)284   void writeInt32BE(uint8_t *Addr, uint32_t Value) {
285     if (IsTargetLittleEndian)
286       sys::swapByteOrder(Value);
287     *Addr       = (Value >> 24) & 0xFF;
288     *(Addr + 1) = (Value >> 16) & 0xFF;
289     *(Addr + 2) = (Value >> 8) & 0xFF;
290     *(Addr + 3) = Value & 0xFF;
291   }
292 
writeInt64BE(uint8_t * Addr,uint64_t Value)293   void writeInt64BE(uint8_t *Addr, uint64_t Value) {
294     if (IsTargetLittleEndian)
295       sys::swapByteOrder(Value);
296     *Addr       = (Value >> 56) & 0xFF;
297     *(Addr + 1) = (Value >> 48) & 0xFF;
298     *(Addr + 2) = (Value >> 40) & 0xFF;
299     *(Addr + 3) = (Value >> 32) & 0xFF;
300     *(Addr + 4) = (Value >> 24) & 0xFF;
301     *(Addr + 5) = (Value >> 16) & 0xFF;
302     *(Addr + 6) = (Value >> 8) & 0xFF;
303     *(Addr + 7) = Value & 0xFF;
304   }
305 
306   /// Endian-aware read Read the least significant Size bytes from Src.
307   uint64_t readBytesUnaligned(uint8_t *Src, unsigned Size) const;
308 
309   /// Endian-aware write. Write the least significant Size bytes from Value to
310   /// Dst.
311   void writeBytesUnaligned(uint64_t Value, uint8_t *Dst, unsigned Size) const;
312 
313   /// \brief Given the common symbols discovered in the object file, emit a
314   /// new section for them and update the symbol mappings in the object and
315   /// symbol table.
316   void emitCommonSymbols(const ObjectFile &Obj, CommonSymbolList &CommonSymbols);
317 
318   /// \brief Emits section data from the object file to the MemoryManager.
319   /// \param IsCode if it's true then allocateCodeSection() will be
320   ///        used for emits, else allocateDataSection() will be used.
321   /// \return SectionID.
322   unsigned emitSection(const ObjectFile &Obj, const SectionRef &Section,
323                        bool IsCode);
324 
325   /// \brief Find Section in LocalSections. If the secton is not found - emit
326   ///        it and store in LocalSections.
327   /// \param IsCode if it's true then allocateCodeSection() will be
328   ///        used for emmits, else allocateDataSection() will be used.
329   /// \return SectionID.
330   unsigned findOrEmitSection(const ObjectFile &Obj, const SectionRef &Section,
331                              bool IsCode, ObjSectionToIDMap &LocalSections);
332 
333   // \brief Add a relocation entry that uses the given section.
334   void addRelocationForSection(const RelocationEntry &RE, unsigned SectionID);
335 
336   // \brief Add a relocation entry that uses the given symbol.  This symbol may
337   // be found in the global symbol table, or it may be external.
338   void addRelocationForSymbol(const RelocationEntry &RE, StringRef SymbolName);
339 
340   /// \brief Emits long jump instruction to Addr.
341   /// \return Pointer to the memory area for emitting target address.
342   uint8_t *createStubFunction(uint8_t *Addr, unsigned AbiVariant = 0);
343 
344   /// \brief Resolves relocations from Relocs list with address from Value.
345   void resolveRelocationList(const RelocationList &Relocs, uint64_t Value);
346 
347   /// \brief A object file specific relocation resolver
348   /// \param RE The relocation to be resolved
349   /// \param Value Target symbol address to apply the relocation action
350   virtual void resolveRelocation(const RelocationEntry &RE, uint64_t Value) = 0;
351 
352   /// \brief Parses one or more object file relocations (some object files use
353   ///        relocation pairs) and stores it to Relocations or SymbolRelocations
354   ///        (this depends on the object file type).
355   /// \return Iterator to the next relocation that needs to be parsed.
356   virtual relocation_iterator
357   processRelocationRef(unsigned SectionID, relocation_iterator RelI,
358                        const ObjectFile &Obj, ObjSectionToIDMap &ObjSectionToID,
359                        StubMap &Stubs) = 0;
360 
361   /// \brief Resolve relocations to external symbols.
362   void resolveExternalSymbols();
363 
364   // \brief Compute an upper bound of the memory that is required to load all
365   // sections
366   void computeTotalAllocSize(const ObjectFile &Obj, uint64_t &CodeSize,
367                              uint64_t &DataSizeRO, uint64_t &DataSizeRW);
368 
369   // \brief Compute the stub buffer size required for a section
370   unsigned computeSectionStubBufSize(const ObjectFile &Obj,
371                                      const SectionRef &Section);
372 
373   // \brief Implementation of the generic part of the loadObject algorithm.
374   std::pair<unsigned, unsigned> loadObjectImpl(const object::ObjectFile &Obj);
375 
376 public:
RuntimeDyldImpl(RuntimeDyld::MemoryManager & MemMgr,RuntimeDyld::SymbolResolver & Resolver)377   RuntimeDyldImpl(RuntimeDyld::MemoryManager &MemMgr,
378                   RuntimeDyld::SymbolResolver &Resolver)
379     : MemMgr(MemMgr), Resolver(Resolver), Checker(nullptr),
380       ProcessAllSections(false), HasError(false) {
381   }
382 
383   virtual ~RuntimeDyldImpl();
384 
setProcessAllSections(bool ProcessAllSections)385   void setProcessAllSections(bool ProcessAllSections) {
386     this->ProcessAllSections = ProcessAllSections;
387   }
388 
setRuntimeDyldChecker(RuntimeDyldCheckerImpl * Checker)389   void setRuntimeDyldChecker(RuntimeDyldCheckerImpl *Checker) {
390     this->Checker = Checker;
391   }
392 
393   virtual std::unique_ptr<RuntimeDyld::LoadedObjectInfo>
394   loadObject(const object::ObjectFile &Obj) = 0;
395 
getSymbolLocalAddress(StringRef Name)396   uint8_t* getSymbolLocalAddress(StringRef Name) const {
397     // FIXME: Just look up as a function for now. Overly simple of course.
398     // Work in progress.
399     RTDyldSymbolTable::const_iterator pos = GlobalSymbolTable.find(Name);
400     if (pos == GlobalSymbolTable.end())
401       return nullptr;
402     const auto &SymInfo = pos->second;
403     return getSectionAddress(SymInfo.getSectionID()) + SymInfo.getOffset();
404   }
405 
getSymbol(StringRef Name)406   RuntimeDyld::SymbolInfo getSymbol(StringRef Name) const {
407     // FIXME: Just look up as a function for now. Overly simple of course.
408     // Work in progress.
409     RTDyldSymbolTable::const_iterator pos = GlobalSymbolTable.find(Name);
410     if (pos == GlobalSymbolTable.end())
411       return nullptr;
412     const auto &SymEntry = pos->second;
413     uint64_t TargetAddr =
414       getSectionLoadAddress(SymEntry.getSectionID()) + SymEntry.getOffset();
415     return RuntimeDyld::SymbolInfo(TargetAddr, SymEntry.getFlags());
416   }
417 
418   void resolveRelocations();
419 
420   void reassignSectionAddress(unsigned SectionID, uint64_t Addr);
421 
422   void mapSectionAddress(const void *LocalAddress, uint64_t TargetAddress);
423 
424   // Is the linker in an error state?
hasError()425   bool hasError() { return HasError; }
426 
427   // Mark the error condition as handled and continue.
clearError()428   void clearError() { HasError = false; }
429 
430   // Get the error message.
getErrorString()431   StringRef getErrorString() { return ErrorStr; }
432 
433   virtual bool isCompatibleFile(const ObjectFile &Obj) const = 0;
434 
435   virtual void registerEHFrames();
436 
437   virtual void deregisterEHFrames();
438 
finalizeLoad(const ObjectFile & ObjImg,ObjSectionToIDMap & SectionMap)439   virtual void finalizeLoad(const ObjectFile &ObjImg,
440                             ObjSectionToIDMap &SectionMap) {}
441 };
442 
443 } // end namespace llvm
444 
445 #endif
446