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