1 //===-- MCJIT.h - Class definition for the MCJIT ----------------*- 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 #ifndef LLVM_LIB_EXECUTIONENGINE_MCJIT_MCJIT_H 10 #define LLVM_LIB_EXECUTIONENGINE_MCJIT_MCJIT_H 11 12 #include "llvm/ADT/SmallPtrSet.h" 13 #include "llvm/ADT/SmallVector.h" 14 #include "llvm/ExecutionEngine/ExecutionEngine.h" 15 #include "llvm/ExecutionEngine/ObjectCache.h" 16 #include "llvm/ExecutionEngine/RTDyldMemoryManager.h" 17 #include "llvm/ExecutionEngine/RuntimeDyld.h" 18 #include "llvm/IR/Module.h" 19 #include "llvm/Support/SmallVectorMemoryBuffer.h" 20 21 namespace llvm { 22 class MCJIT; 23 24 // This is a helper class that the MCJIT execution engine uses for linking 25 // functions across modules that it owns. It aggregates the memory manager 26 // that is passed in to the MCJIT constructor and defers most functionality 27 // to that object. 28 class LinkingSymbolResolver : public LegacyJITSymbolResolver { 29 public: LinkingSymbolResolver(MCJIT & Parent,std::shared_ptr<LegacyJITSymbolResolver> Resolver)30 LinkingSymbolResolver(MCJIT &Parent, 31 std::shared_ptr<LegacyJITSymbolResolver> Resolver) 32 : ParentEngine(Parent), ClientResolver(std::move(Resolver)) {} 33 34 JITSymbol findSymbol(const std::string &Name) override; 35 36 // MCJIT doesn't support logical dylibs. findSymbolInLogicalDylib(const std::string & Name)37 JITSymbol findSymbolInLogicalDylib(const std::string &Name) override { 38 return nullptr; 39 } 40 41 private: 42 MCJIT &ParentEngine; 43 std::shared_ptr<LegacyJITSymbolResolver> ClientResolver; 44 void anchor() override; 45 }; 46 47 // About Module states: added->loaded->finalized. 48 // 49 // The purpose of the "added" state is having modules in standby. (added=known 50 // but not compiled). The idea is that you can add a module to provide function 51 // definitions but if nothing in that module is referenced by a module in which 52 // a function is executed (note the wording here because it's not exactly the 53 // ideal case) then the module never gets compiled. This is sort of lazy 54 // compilation. 55 // 56 // The purpose of the "loaded" state (loaded=compiled and required sections 57 // copied into local memory but not yet ready for execution) is to have an 58 // intermediate state wherein clients can remap the addresses of sections, using 59 // MCJIT::mapSectionAddress, (in preparation for later copying to a new location 60 // or an external process) before relocations and page permissions are applied. 61 // 62 // It might not be obvious at first glance, but the "remote-mcjit" case in the 63 // lli tool does this. In that case, the intermediate action is taken by the 64 // RemoteMemoryManager in response to the notifyObjectLoaded function being 65 // called. 66 67 class MCJIT : public ExecutionEngine { 68 MCJIT(std::unique_ptr<Module> M, std::unique_ptr<TargetMachine> tm, 69 std::shared_ptr<MCJITMemoryManager> MemMgr, 70 std::shared_ptr<LegacyJITSymbolResolver> Resolver); 71 72 typedef llvm::SmallPtrSet<Module *, 4> ModulePtrSet; 73 74 class OwningModuleContainer { 75 public: OwningModuleContainer()76 OwningModuleContainer() { 77 } ~OwningModuleContainer()78 ~OwningModuleContainer() { 79 freeModulePtrSet(AddedModules); 80 freeModulePtrSet(LoadedModules); 81 freeModulePtrSet(FinalizedModules); 82 } 83 begin_added()84 ModulePtrSet::iterator begin_added() { return AddedModules.begin(); } end_added()85 ModulePtrSet::iterator end_added() { return AddedModules.end(); } added()86 iterator_range<ModulePtrSet::iterator> added() { 87 return make_range(begin_added(), end_added()); 88 } 89 begin_loaded()90 ModulePtrSet::iterator begin_loaded() { return LoadedModules.begin(); } end_loaded()91 ModulePtrSet::iterator end_loaded() { return LoadedModules.end(); } 92 begin_finalized()93 ModulePtrSet::iterator begin_finalized() { return FinalizedModules.begin(); } end_finalized()94 ModulePtrSet::iterator end_finalized() { return FinalizedModules.end(); } 95 addModule(std::unique_ptr<Module> M)96 void addModule(std::unique_ptr<Module> M) { 97 AddedModules.insert(M.release()); 98 } 99 removeModule(Module * M)100 bool removeModule(Module *M) { 101 return AddedModules.erase(M) || LoadedModules.erase(M) || 102 FinalizedModules.erase(M); 103 } 104 hasModuleBeenAddedButNotLoaded(Module * M)105 bool hasModuleBeenAddedButNotLoaded(Module *M) { 106 return AddedModules.count(M) != 0; 107 } 108 hasModuleBeenLoaded(Module * M)109 bool hasModuleBeenLoaded(Module *M) { 110 // If the module is in either the "loaded" or "finalized" sections it 111 // has been loaded. 112 return (LoadedModules.count(M) != 0 ) || (FinalizedModules.count(M) != 0); 113 } 114 hasModuleBeenFinalized(Module * M)115 bool hasModuleBeenFinalized(Module *M) { 116 return FinalizedModules.count(M) != 0; 117 } 118 ownsModule(Module * M)119 bool ownsModule(Module* M) { 120 return (AddedModules.count(M) != 0) || (LoadedModules.count(M) != 0) || 121 (FinalizedModules.count(M) != 0); 122 } 123 markModuleAsLoaded(Module * M)124 void markModuleAsLoaded(Module *M) { 125 // This checks against logic errors in the MCJIT implementation. 126 // This function should never be called with either a Module that MCJIT 127 // does not own or a Module that has already been loaded and/or finalized. 128 assert(AddedModules.count(M) && 129 "markModuleAsLoaded: Module not found in AddedModules"); 130 131 // Remove the module from the "Added" set. 132 AddedModules.erase(M); 133 134 // Add the Module to the "Loaded" set. 135 LoadedModules.insert(M); 136 } 137 markModuleAsFinalized(Module * M)138 void markModuleAsFinalized(Module *M) { 139 // This checks against logic errors in the MCJIT implementation. 140 // This function should never be called with either a Module that MCJIT 141 // does not own, a Module that has not been loaded or a Module that has 142 // already been finalized. 143 assert(LoadedModules.count(M) && 144 "markModuleAsFinalized: Module not found in LoadedModules"); 145 146 // Remove the module from the "Loaded" section of the list. 147 LoadedModules.erase(M); 148 149 // Add the Module to the "Finalized" section of the list by inserting it 150 // before the 'end' iterator. 151 FinalizedModules.insert(M); 152 } 153 markAllLoadedModulesAsFinalized()154 void markAllLoadedModulesAsFinalized() { 155 for (ModulePtrSet::iterator I = LoadedModules.begin(), 156 E = LoadedModules.end(); 157 I != E; ++I) { 158 Module *M = *I; 159 FinalizedModules.insert(M); 160 } 161 LoadedModules.clear(); 162 } 163 164 private: 165 ModulePtrSet AddedModules; 166 ModulePtrSet LoadedModules; 167 ModulePtrSet FinalizedModules; 168 freeModulePtrSet(ModulePtrSet & MPS)169 void freeModulePtrSet(ModulePtrSet& MPS) { 170 // Go through the module set and delete everything. 171 for (ModulePtrSet::iterator I = MPS.begin(), E = MPS.end(); I != E; ++I) { 172 Module *M = *I; 173 delete M; 174 } 175 MPS.clear(); 176 } 177 }; 178 179 std::unique_ptr<TargetMachine> TM; 180 MCContext *Ctx; 181 std::shared_ptr<MCJITMemoryManager> MemMgr; 182 LinkingSymbolResolver Resolver; 183 RuntimeDyld Dyld; 184 std::vector<JITEventListener*> EventListeners; 185 186 OwningModuleContainer OwnedModules; 187 188 SmallVector<object::OwningBinary<object::Archive>, 2> Archives; 189 SmallVector<std::unique_ptr<MemoryBuffer>, 2> Buffers; 190 191 SmallVector<std::unique_ptr<object::ObjectFile>, 2> LoadedObjects; 192 193 // An optional ObjectCache to be notified of compiled objects and used to 194 // perform lookup of pre-compiled code to avoid re-compilation. 195 ObjectCache *ObjCache; 196 197 Function *FindFunctionNamedInModulePtrSet(StringRef FnName, 198 ModulePtrSet::iterator I, 199 ModulePtrSet::iterator E); 200 201 GlobalVariable *FindGlobalVariableNamedInModulePtrSet(StringRef Name, 202 bool AllowInternal, 203 ModulePtrSet::iterator I, 204 ModulePtrSet::iterator E); 205 206 void runStaticConstructorsDestructorsInModulePtrSet(bool isDtors, 207 ModulePtrSet::iterator I, 208 ModulePtrSet::iterator E); 209 210 public: 211 ~MCJIT() override; 212 213 /// @name ExecutionEngine interface implementation 214 /// @{ 215 void addModule(std::unique_ptr<Module> M) override; 216 void addObjectFile(std::unique_ptr<object::ObjectFile> O) override; 217 void addObjectFile(object::OwningBinary<object::ObjectFile> O) override; 218 void addArchive(object::OwningBinary<object::Archive> O) override; 219 bool removeModule(Module *M) override; 220 221 /// FindFunctionNamed - Search all of the active modules to find the function that 222 /// defines FnName. This is very slow operation and shouldn't be used for 223 /// general code. 224 Function *FindFunctionNamed(StringRef FnName) override; 225 226 /// FindGlobalVariableNamed - Search all of the active modules to find the 227 /// global variable that defines Name. This is very slow operation and 228 /// shouldn't be used for general code. 229 GlobalVariable *FindGlobalVariableNamed(StringRef Name, 230 bool AllowInternal = false) override; 231 232 /// Sets the object manager that MCJIT should use to avoid compilation. 233 void setObjectCache(ObjectCache *manager) override; 234 setProcessAllSections(bool ProcessAllSections)235 void setProcessAllSections(bool ProcessAllSections) override { 236 Dyld.setProcessAllSections(ProcessAllSections); 237 } 238 239 void generateCodeForModule(Module *M) override; 240 241 /// finalizeObject - ensure the module is fully processed and is usable. 242 /// 243 /// It is the user-level function for completing the process of making the 244 /// object usable for execution. It should be called after sections within an 245 /// object have been relocated using mapSectionAddress. When this method is 246 /// called the MCJIT execution engine will reapply relocations for a loaded 247 /// object. 248 /// Is it OK to finalize a set of modules, add modules and finalize again. 249 // FIXME: Do we really need both of these? 250 void finalizeObject() override; 251 virtual void finalizeModule(Module *); 252 void finalizeLoadedModules(); 253 254 /// runStaticConstructorsDestructors - This method is used to execute all of 255 /// the static constructors or destructors for a program. 256 /// 257 /// \param isDtors - Run the destructors instead of constructors. 258 void runStaticConstructorsDestructors(bool isDtors) override; 259 260 void *getPointerToFunction(Function *F) override; 261 262 GenericValue runFunction(Function *F, 263 ArrayRef<GenericValue> ArgValues) override; 264 265 /// getPointerToNamedFunction - This method returns the address of the 266 /// specified function by using the dlsym function call. As such it is only 267 /// useful for resolving library symbols, not code generated symbols. 268 /// 269 /// If AbortOnFailure is false and no function with the given name is 270 /// found, this function silently returns a null pointer. Otherwise, 271 /// it prints a message to stderr and aborts. 272 /// 273 void *getPointerToNamedFunction(StringRef Name, 274 bool AbortOnFailure = true) override; 275 276 /// mapSectionAddress - map a section to its target address space value. 277 /// Map the address of a JIT section as returned from the memory manager 278 /// to the address in the target process as the running code will see it. 279 /// This is the address which will be used for relocation resolution. mapSectionAddress(const void * LocalAddress,uint64_t TargetAddress)280 void mapSectionAddress(const void *LocalAddress, 281 uint64_t TargetAddress) override { 282 Dyld.mapSectionAddress(LocalAddress, TargetAddress); 283 } 284 void RegisterJITEventListener(JITEventListener *L) override; 285 void UnregisterJITEventListener(JITEventListener *L) override; 286 287 // If successful, these function will implicitly finalize all loaded objects. 288 // To get a function address within MCJIT without causing a finalize, use 289 // getSymbolAddress. 290 uint64_t getGlobalValueAddress(const std::string &Name) override; 291 uint64_t getFunctionAddress(const std::string &Name) override; 292 getTargetMachine()293 TargetMachine *getTargetMachine() override { return TM.get(); } 294 295 /// @} 296 /// @name (Private) Registration Interfaces 297 /// @{ 298 Register()299 static void Register() { 300 MCJITCtor = createJIT; 301 } 302 303 static ExecutionEngine * 304 createJIT(std::unique_ptr<Module> M, std::string *ErrorStr, 305 std::shared_ptr<MCJITMemoryManager> MemMgr, 306 std::shared_ptr<LegacyJITSymbolResolver> Resolver, 307 std::unique_ptr<TargetMachine> TM); 308 309 // @} 310 311 // Takes a mangled name and returns the corresponding JITSymbol (if a 312 // definition of that mangled name has been added to the JIT). 313 JITSymbol findSymbol(const std::string &Name, bool CheckFunctionsOnly); 314 315 // DEPRECATED - Please use findSymbol instead. 316 // 317 // This is not directly exposed via the ExecutionEngine API, but it is 318 // used by the LinkingMemoryManager. 319 // 320 // getSymbolAddress takes an unmangled name and returns the corresponding 321 // JITSymbol if a definition of the name has been added to the JIT. 322 uint64_t getSymbolAddress(const std::string &Name, 323 bool CheckFunctionsOnly); 324 325 protected: 326 /// emitObject -- Generate a JITed object in memory from the specified module 327 /// Currently, MCJIT only supports a single module and the module passed to 328 /// this function call is expected to be the contained module. The module 329 /// is passed as a parameter here to prepare for multiple module support in 330 /// the future. 331 std::unique_ptr<MemoryBuffer> emitObject(Module *M); 332 333 void notifyObjectLoaded(const object::ObjectFile &Obj, 334 const RuntimeDyld::LoadedObjectInfo &L); 335 void notifyFreeingObject(const object::ObjectFile &Obj); 336 337 JITSymbol findExistingSymbol(const std::string &Name); 338 Module *findModuleForSymbol(const std::string &Name, bool CheckFunctionsOnly); 339 }; 340 341 } // end llvm namespace 342 343 #endif // LLVM_LIB_EXECUTIONENGINE_MCJIT_MCJIT_H 344