diff options
Diffstat (limited to 'contrib/llvm/lib/ExecutionEngine')
17 files changed, 1206 insertions, 127 deletions
diff --git a/contrib/llvm/lib/ExecutionEngine/ExecutionEngine.cpp b/contrib/llvm/lib/ExecutionEngine/ExecutionEngine.cpp index f286975..2b1e878 100644 --- a/contrib/llvm/lib/ExecutionEngine/ExecutionEngine.cpp +++ b/contrib/llvm/lib/ExecutionEngine/ExecutionEngine.cpp @@ -79,9 +79,10 @@ ExecutionEngine::~ExecutionEngine() { void ExecutionEngine::DeregisterAllTables() { if (ExceptionTableDeregister) { - for (std::vector<void*>::iterator it = AllExceptionTables.begin(), - ie = AllExceptionTables.end(); it != ie; ++it) - ExceptionTableDeregister(*it); + DenseMap<const Function*, void*>::iterator it = AllExceptionTables.begin(); + DenseMap<const Function*, void*>::iterator ite = AllExceptionTables.end(); + for (; it != ite; ++it) + ExceptionTableDeregister(it->second); AllExceptionTables.clear(); } } @@ -310,19 +311,19 @@ void ExecutionEngine::runStaticConstructorsDestructors(Module *module, // it. if (!GV || GV->isDeclaration() || GV->hasLocalLinkage()) return; - // Should be an array of '{ int, void ()* }' structs. The first value is + // Should be an array of '{ i32, void ()* }' structs. The first value is // the init priority, which we ignore. - ConstantArray *InitList = dyn_cast<ConstantArray>(GV->getInitializer()); - if (!InitList) return; + if (isa<ConstantAggregateZero>(GV->getInitializer())) + return; + ConstantArray *InitList = cast<ConstantArray>(GV->getInitializer()); for (unsigned i = 0, e = InitList->getNumOperands(); i != e; ++i) { - ConstantStruct *CS = - dyn_cast<ConstantStruct>(InitList->getOperand(i)); - if (!CS) continue; - if (CS->getNumOperands() != 2) return; // Not array of 2-element structs. + if (isa<ConstantAggregateZero>(InitList->getOperand(i))) + continue; + ConstantStruct *CS = cast<ConstantStruct>(InitList->getOperand(i)); Constant *FP = CS->getOperand(1); if (FP->isNullValue()) - break; // Found a null terminator, exit. + continue; // Found a sentinal value, ignore. // Strip off constant expression casts. if (ConstantExpr *CE = dyn_cast<ConstantExpr>(FP)) @@ -838,7 +839,7 @@ void ExecutionEngine::StoreValueToMemory(const GenericValue &Val, case Type::PointerTyID: // Ensure 64 bit target pointers are fully initialized on 32 bit hosts. if (StoreBytes != sizeof(PointerTy)) - memset(Ptr, 0, StoreBytes); + memset(&(Ptr->PointerVal), 0, StoreBytes); *((PointerTy*)Ptr) = Val.PointerVal; break; diff --git a/contrib/llvm/lib/ExecutionEngine/JIT/Intercept.cpp b/contrib/llvm/lib/ExecutionEngine/JIT/Intercept.cpp index 169e1ba..fa8bee4 100644 --- a/contrib/llvm/lib/ExecutionEngine/JIT/Intercept.cpp +++ b/contrib/llvm/lib/ExecutionEngine/JIT/Intercept.cpp @@ -52,8 +52,8 @@ static void runAtExitHandlers() { #include <sys/stat.h> #endif #include <fcntl.h> -/* stat functions are redirecting to __xstat with a version number. On x86-64 - * linking with libc_nonshared.a and -Wl,--export-dynamic doesn't make 'stat' +/* stat functions are redirecting to __xstat with a version number. On x86-64 + * linking with libc_nonshared.a and -Wl,--export-dynamic doesn't make 'stat' * available as an exported symbol, so we have to add it explicitly. */ namespace { @@ -119,18 +119,18 @@ void *JIT::getPointerToNamedFunction(const std::string &Name, const char *NameStr = Name.c_str(); // If this is an asm specifier, skip the sentinal. if (NameStr[0] == 1) ++NameStr; - + // If it's an external function, look it up in the process image... void *Ptr = sys::DynamicLibrary::SearchForAddressOfSymbol(NameStr); if (Ptr) return Ptr; - + // If it wasn't found and if it starts with an underscore ('_') character, // and has an asm specifier, try again without the underscore. if (Name[0] == 1 && NameStr[0] == '_') { Ptr = sys::DynamicLibrary::SearchForAddressOfSymbol(NameStr+1); if (Ptr) return Ptr; } - + // Darwin/PPC adds $LDBLStub suffixes to various symbols like printf. These // are references to hidden visibility symbols that dlsym cannot resolve. // If we have one of these, strip off $LDBLStub and try again. @@ -147,7 +147,7 @@ void *JIT::getPointerToNamedFunction(const std::string &Name, } #endif } - + /// If a LazyFunctionCreator is installed, use it to get/create the function. if (LazyFunctionCreator) if (void *RP = LazyFunctionCreator(Name)) diff --git a/contrib/llvm/lib/ExecutionEngine/JIT/JIT.cpp b/contrib/llvm/lib/ExecutionEngine/JIT/JIT.cpp index cc76b13..d1f87ac 100644 --- a/contrib/llvm/lib/ExecutionEngine/JIT/JIT.cpp +++ b/contrib/llvm/lib/ExecutionEngine/JIT/JIT.cpp @@ -35,7 +35,7 @@ using namespace llvm; -#ifdef __APPLE__ +#ifdef __APPLE__ // Apple gcc defaults to -fuse-cxa-atexit (i.e. calls __cxa_atexit instead // of atexit). It passes the address of linker generated symbol __dso_handle // to the function. @@ -75,7 +75,7 @@ extern "C" void LLVMLinkInJIT() { #endif #if HAVE_EHTABLE_SUPPORT - + // libgcc defines the __register_frame function to dynamically register new // dwarf frames for exception handling. This functionality is not portable // across compilers and is only provided by GCC. We use the __register_frame @@ -113,10 +113,10 @@ struct LibgccObject { void *unused1; void *unused2; void *unused3; - + /// frame - Pointer to the exception table. void *frame; - + /// encoding - The encoding of the object? union { struct { @@ -124,15 +124,15 @@ struct LibgccObject { unsigned long from_array : 1; unsigned long mixed_encoding : 1; unsigned long encoding : 8; - unsigned long count : 21; + unsigned long count : 21; } b; size_t i; } encoding; - + /// fde_end - libgcc defines this field only if some macro is defined. We /// include this field even if it may not there, to make libgcc happy. char *fde_end; - + /// next - At least we know it's a chained list! struct LibgccObject *next; }; @@ -153,7 +153,7 @@ struct LibgccObjectInfo { /// unseenObjects - LibgccObjects not parsed yet by the unwinding runtime. /// struct LibgccObject* unseenObjects; - + unsigned unused[2]; }; @@ -165,32 +165,32 @@ void DarwinRegisterFrame(void* FrameBegin) { LibgccObjectInfo* LOI = (struct LibgccObjectInfo*) _keymgr_get_and_lock_processwide_ptr(KEYMGR_GCC3_DW2_OBJ_LIST); assert(LOI && "This should be preallocated by the runtime"); - + // Allocate a new LibgccObject to represent this frame. Deallocation of this // object may be impossible: since darwin code in libgcc was written after // the ability to dynamically register frames, things may crash if we // deallocate it. struct LibgccObject* ob = (struct LibgccObject*) malloc(sizeof(struct LibgccObject)); - + // Do like libgcc for the values of the field. ob->unused1 = (void *)-1; ob->unused2 = 0; ob->unused3 = 0; ob->frame = FrameBegin; - ob->encoding.i = 0; + ob->encoding.i = 0; ob->encoding.b.encoding = llvm::dwarf::DW_EH_PE_omit; - + // Put the info on both places, as libgcc uses the first or the second // field. Note that we rely on having two pointers here. If fde_end was a // char, things would get complicated. ob->fde_end = (char*)LOI->unseenObjects; ob->next = LOI->unseenObjects; - + // Update the key's unseenObjects list. LOI->unseenObjects = ob; - - // Finally update the "key". Apparently, libgcc requires it. + + // Finally update the "key". Apparently, libgcc requires it. _keymgr_set_and_unlock_processwide_ptr(KEYMGR_GCC3_DW2_OBJ_LIST, LOI); @@ -312,18 +312,18 @@ JIT::JIT(Module *M, TargetMachine &tm, TargetJITInfo &tji, if (TM.addPassesToEmitMachineCode(PM, *JCE, OptLevel)) { report_fatal_error("Target does not support machine code emission!"); } - + // Register routine for informing unwinding runtime about new EH frames #if HAVE_EHTABLE_SUPPORT #if USE_KEYMGR struct LibgccObjectInfo* LOI = (struct LibgccObjectInfo*) _keymgr_get_and_lock_processwide_ptr(KEYMGR_GCC3_DW2_OBJ_LIST); - + // The key is created on demand, and libgcc creates it the first time an // exception occurs. Since we need the key to register frames, we create // it now. if (!LOI) - LOI = (LibgccObjectInfo*)calloc(sizeof(struct LibgccObjectInfo), 1); + LOI = (LibgccObjectInfo*)calloc(sizeof(struct LibgccObjectInfo), 1); _keymgr_set_and_unlock_processwide_ptr(KEYMGR_GCC3_DW2_OBJ_LIST, LOI); InstallExceptionTableRegister(DarwinRegisterFrame); // Not sure about how to deregister on Darwin. @@ -332,7 +332,7 @@ JIT::JIT(Module *M, TargetMachine &tm, TargetJITInfo &tji, InstallExceptionTableDeregister(__deregister_frame); #endif // __APPLE__ #endif // HAVE_EHTABLE_SUPPORT - + // Initialize passes. PM.doInitialization(); } @@ -365,11 +365,11 @@ void JIT::addModule(Module *M) { if (TM.addPassesToEmitMachineCode(PM, *JCE, CodeGenOpt::Default)) { report_fatal_error("Target does not support machine code emission!"); } - + // Initialize passes. PM.doInitialization(); } - + ExecutionEngine::addModule(M); } @@ -377,29 +377,29 @@ void JIT::addModule(Module *M) { /// since the PassManager it contains references a released Module. bool JIT::removeModule(Module *M) { bool result = ExecutionEngine::removeModule(M); - + MutexGuard locked(lock); - + if (jitstate->getModule() == M) { delete jitstate; jitstate = 0; } - + if (!jitstate && !Modules.empty()) { jitstate = new JITState(Modules[0]); FunctionPassManager &PM = jitstate->getPM(locked); PM.add(new TargetData(*TM.getTargetData())); - + // Turn the machine code intermediate representation into bytes in memory // that may be executed. if (TM.addPassesToEmitMachineCode(PM, *JCE, CodeGenOpt::Default)) { report_fatal_error("Target does not support machine code emission!"); } - + // Initialize passes. PM.doInitialization(); - } + } return result; } @@ -433,7 +433,7 @@ GenericValue JIT::runFunction(Function *F, // Call the function. GenericValue rv; - rv.IntVal = APInt(32, PF(ArgValues[0].IntVal.getZExtValue(), + rv.IntVal = APInt(32, PF(ArgValues[0].IntVal.getZExtValue(), (char **)GVTOP(ArgValues[1]), (const char **)GVTOP(ArgValues[2]))); return rv; @@ -446,7 +446,7 @@ GenericValue JIT::runFunction(Function *F, // Call the function. GenericValue rv; - rv.IntVal = APInt(32, PF(ArgValues[0].IntVal.getZExtValue(), + rv.IntVal = APInt(32, PF(ArgValues[0].IntVal.getZExtValue(), (char **)GVTOP(ArgValues[1]))); return rv; } @@ -480,7 +480,7 @@ GenericValue JIT::runFunction(Function *F, rv.IntVal = APInt(BitWidth, ((int(*)())(intptr_t)FPtr)()); else if (BitWidth <= 64) rv.IntVal = APInt(BitWidth, ((int64_t(*)())(intptr_t)FPtr)()); - else + else llvm_unreachable("Integer types > 64 bits not supported"); return rv; } @@ -542,7 +542,7 @@ GenericValue JIT::runFunction(Function *F, case Type::PointerTyID: void *ArgPtr = GVTOP(AV); if (sizeof(void*) == 4) - C = ConstantInt::get(Type::getInt32Ty(F->getContext()), + C = ConstantInt::get(Type::getInt32Ty(F->getContext()), (int)(intptr_t)ArgPtr); else C = ConstantInt::get(Type::getInt64Ty(F->getContext()), @@ -649,7 +649,7 @@ void JIT::runJITOnFunctionUnlocked(Function *F, const MutexGuard &locked) { "Externally-defined function should not be in pending list."); jitTheFunction(PF, locked); - + // Now that the function has been jitted, ask the JITEmitter to rewrite // the stub with real address of the function. updateFunctionStub(PF); @@ -666,7 +666,7 @@ void JIT::jitTheFunction(Function *F, const MutexGuard &locked) { } /// getPointerToFunction - This method is used to get the address of the -/// specified function, compiling it if neccesary. +/// specified function, compiling it if necessary. /// void *JIT::getPointerToFunction(Function *F) { @@ -703,7 +703,7 @@ void *JIT::getPointerToFunction(Function *F) { void JIT::addPointerToBasicBlock(const BasicBlock *BB, void *Addr) { MutexGuard locked(lock); - + BasicBlockAddressMapTy::iterator I = getBasicBlockAddressMap(locked).find(BB); if (I == getBasicBlockAddressMap(locked).end()) { @@ -724,7 +724,7 @@ void *JIT::getPointerToBasicBlock(BasicBlock *BB) { // resolve basic block address MutexGuard locked(lock); - + BasicBlockAddressMapTy::iterator I = getBasicBlockAddressMap(locked).find(BB); if (I != getBasicBlockAddressMap(locked).end()) { diff --git a/contrib/llvm/lib/ExecutionEngine/JIT/JIT.h b/contrib/llvm/lib/ExecutionEngine/JIT/JIT.h index 1d1763e..b576c16 100644 --- a/contrib/llvm/lib/ExecutionEngine/JIT/JIT.h +++ b/contrib/llvm/lib/ExecutionEngine/JIT/JIT.h @@ -42,7 +42,7 @@ public: FunctionPassManager &getPM(const MutexGuard &L) { return PM; } - + Module *getModule() const { return M; } std::vector<AssertingVH<Function> > &getPendingFunctions(const MutexGuard &L){ return PendingFunctions; @@ -86,7 +86,7 @@ public: static void Register() { JITCtor = createJIT; } - + /// getJITInfo - Return the target JIT information structure. /// TargetJITInfo &getJITInfo() const { return TJI; } @@ -106,7 +106,7 @@ public: } virtual void addModule(Module *M); - + /// removeModule - Remove a Module from the list of modules. Returns true if /// M is found. virtual bool removeModule(Module *M); @@ -146,7 +146,7 @@ public: /// getPointerToBasicBlock - This returns the address of the specified basic /// block, assuming function is compiled. void *getPointerToBasicBlock(BasicBlock *BB); - + /// getOrEmitGlobalVariable - Return the address of the specified global /// variable, possibly emitting it to memory if needed. This is used by the /// Emitter. @@ -172,7 +172,7 @@ public: void freeMachineCodeForFunction(Function *F); /// addPendingFunction - while jitting non-lazily, a called but non-codegen'd - /// function was encountered. Add it to a pending list to be processed after + /// function was encountered. Add it to a pending list to be processed after /// the current function. /// void addPendingFunction(Function *F); diff --git a/contrib/llvm/lib/ExecutionEngine/JIT/JITDebugRegisterer.cpp b/contrib/llvm/lib/ExecutionEngine/JIT/JITDebugRegisterer.cpp index 3b5acb7..e71c20b 100644 --- a/contrib/llvm/lib/ExecutionEngine/JIT/JITDebugRegisterer.cpp +++ b/contrib/llvm/lib/ExecutionEngine/JIT/JITDebugRegisterer.cpp @@ -27,7 +27,6 @@ #include "llvm/Support/raw_ostream.h" #include "llvm/Support/Mutex.h" #include <string> -#include <vector> namespace llvm { @@ -143,7 +142,7 @@ void JITDebugRegisterer::RegisterFunction(const Function *F, DebugInfo &I) { // Add a mapping from F to the entry and buffer, so we can delete this // info later. - FnMap[F] = std::make_pair<std::string, jit_code_entry*>(Buffer, JITCodeEntry); + FnMap[F] = std::make_pair(Buffer, JITCodeEntry); // Acquire the lock and do the registration. { diff --git a/contrib/llvm/lib/ExecutionEngine/JIT/JITDwarfEmitter.cpp b/contrib/llvm/lib/ExecutionEngine/JIT/JITDwarfEmitter.cpp index f54ccca..ddb0d54 100644 --- a/contrib/llvm/lib/ExecutionEngine/JIT/JITDwarfEmitter.cpp +++ b/contrib/llvm/lib/ExecutionEngine/JIT/JITDwarfEmitter.cpp @@ -34,7 +34,7 @@ using namespace llvm; JITDwarfEmitter::JITDwarfEmitter(JIT& theJit) : MMI(0), Jit(theJit) {} -unsigned char* JITDwarfEmitter::EmitDwarfTable(MachineFunction& F, +unsigned char* JITDwarfEmitter::EmitDwarfTable(MachineFunction& F, JITCodeEmitter& jce, unsigned char* StartFunction, unsigned char* EndFunction, @@ -47,10 +47,10 @@ unsigned char* JITDwarfEmitter::EmitDwarfTable(MachineFunction& F, RI = TM.getRegisterInfo(); TFI = TM.getFrameLowering(); JCE = &jce; - + unsigned char* ExceptionTable = EmitExceptionTable(&F, StartFunction, EndFunction); - + unsigned char* Result = 0; const std::vector<const Function *> Personalities = MMI->getPersonalities(); @@ -63,7 +63,7 @@ unsigned char* JITDwarfEmitter::EmitDwarfTable(MachineFunction& F, } -void +void JITDwarfEmitter::EmitFrameMoves(intptr_t BaseLabelPtr, const std::vector<MachineMove> &Moves) const { unsigned PointerSize = TD->getPointerSize(); @@ -74,26 +74,26 @@ JITDwarfEmitter::EmitFrameMoves(intptr_t BaseLabelPtr, for (unsigned i = 0, N = Moves.size(); i < N; ++i) { const MachineMove &Move = Moves[i]; MCSymbol *Label = Move.getLabel(); - + // Throw out move if the label is invalid. if (Label && (*JCE->getLabelLocations())[Label] == 0) continue; - + intptr_t LabelPtr = 0; if (Label) LabelPtr = JCE->getLabelAddress(Label); const MachineLocation &Dst = Move.getDestination(); const MachineLocation &Src = Move.getSource(); - + // Advance row if new location. if (BaseLabelPtr && Label && BaseLabel != Label) { JCE->emitByte(dwarf::DW_CFA_advance_loc4); JCE->emitInt32(LabelPtr - BaseLabelPtr); - - BaseLabel = Label; + + BaseLabel = Label; BaseLabelPtr = LabelPtr; } - + // If advancing cfa. if (Dst.isReg() && Dst.getReg() == MachineLocation::VirtualFP) { if (!Src.isReg()) { @@ -103,7 +103,7 @@ JITDwarfEmitter::EmitFrameMoves(intptr_t BaseLabelPtr, JCE->emitByte(dwarf::DW_CFA_def_cfa); JCE->emitULEB128Bytes(RI->getDwarfRegNum(Src.getReg(), true)); } - + JCE->emitULEB128Bytes(-Src.getOffset()); } else { llvm_unreachable("Machine move not supported yet."); @@ -119,7 +119,7 @@ JITDwarfEmitter::EmitFrameMoves(intptr_t BaseLabelPtr, } else { unsigned Reg = RI->getDwarfRegNum(Src.getReg(), true); int Offset = Dst.getOffset() / stackGrowth; - + if (Offset < 0) { JCE->emitByte(dwarf::DW_CFA_offset_extended_sf); JCE->emitULEB128Bytes(Reg); @@ -382,7 +382,7 @@ unsigned char* JITDwarfEmitter::EmitExceptionTable(MachineFunction* MF, unsigned TypeOffset = sizeof(int8_t) + // Call site format // Call-site table length - MCAsmInfo::getULEB128Size(SizeSites) + + MCAsmInfo::getULEB128Size(SizeSites) + SizeSites + SizeActions + SizeTypes; // Begin the exception table. @@ -452,7 +452,7 @@ unsigned char* JITDwarfEmitter::EmitExceptionTable(MachineFunction* MF, // Emit the type ids. for (unsigned M = TypeInfos.size(); M; --M) { const GlobalVariable *GV = TypeInfos[M - 1]; - + if (GV) { if (TD->getPointerSize() == sizeof(int32_t)) JCE->emitInt32((intptr_t)Jit.getOrEmitGlobalVariable(GV)); @@ -484,7 +484,7 @@ JITDwarfEmitter::EmitCommonEHFrame(const Function* Personality) const { unsigned PointerSize = TD->getPointerSize(); int stackGrowth = stackGrowthDirection == TargetFrameLowering::StackGrowsUp ? PointerSize : -PointerSize; - + unsigned char* StartCommonPtr = (unsigned char*)JCE->getCurrentPCValue(); // EH Common Frame header JCE->allocateSpace(4, 0); @@ -499,13 +499,13 @@ JITDwarfEmitter::EmitCommonEHFrame(const Function* Personality) const { if (Personality) { // Augmentation Size: 3 small ULEBs of one byte each, and the personality // function which size is PointerSize. - JCE->emitULEB128Bytes(3 + PointerSize); - + JCE->emitULEB128Bytes(3 + PointerSize); + // We set the encoding of the personality as direct encoding because we use // the function pointer. The encoding is not relative because the current // PC value may be bigger than the personality function pointer. if (PointerSize == 4) { - JCE->emitByte(dwarf::DW_EH_PE_sdata4); + JCE->emitByte(dwarf::DW_EH_PE_sdata4); JCE->emitInt32(((intptr_t)Jit.getPointerToGlobal(Personality))); } else { JCE->emitByte(dwarf::DW_EH_PE_sdata8); @@ -540,11 +540,11 @@ JITDwarfEmitter::EmitCommonEHFrame(const Function* Personality) const { unsigned char* JITDwarfEmitter::EmitEHFrame(const Function* Personality, unsigned char* StartCommonPtr, - unsigned char* StartFunction, + unsigned char* StartFunction, unsigned char* EndFunction, unsigned char* ExceptionTable) const { unsigned PointerSize = TD->getPointerSize(); - + // EH frame header. unsigned char* StartEHPtr = (unsigned char*)JCE->getCurrentPCValue(); JCE->allocateSpace(4, 0); @@ -558,7 +558,7 @@ JITDwarfEmitter::EmitEHFrame(const Function* Personality, // specific data area in the exception table. if (Personality) { JCE->emitULEB128Bytes(PointerSize == 4 ? 4 : 8); - + if (PointerSize == 4) { if (!MMI->getLandingPads().empty()) JCE->emitInt32(ExceptionTable-(unsigned char*)JCE->getCurrentPCValue()); @@ -573,7 +573,7 @@ JITDwarfEmitter::EmitEHFrame(const Function* Personality, } else { JCE->emitULEB128Bytes(0); } - + // Indicate locations of function specific callee saved registers in // frame. EmitFrameMoves((intptr_t)StartFunction, MMI->getFrameMoves()); @@ -593,6 +593,6 @@ JITDwarfEmitter::EmitEHFrame(const Function* Personality, JCE->emitInt32(0); JCE->emitInt32(0); } - + return StartEHPtr; } diff --git a/contrib/llvm/lib/ExecutionEngine/JIT/JITDwarfEmitter.h b/contrib/llvm/lib/ExecutionEngine/JIT/JITDwarfEmitter.h index 9495697..e1d0045 100644 --- a/contrib/llvm/lib/ExecutionEngine/JIT/JITDwarfEmitter.h +++ b/contrib/llvm/lib/ExecutionEngine/JIT/JITDwarfEmitter.h @@ -35,33 +35,33 @@ class JITDwarfEmitter { MachineModuleInfo* MMI; JIT& Jit; bool stackGrowthDirection; - + unsigned char* EmitExceptionTable(MachineFunction* MF, - unsigned char* StartFunction, + unsigned char* StartFunction, unsigned char* EndFunction) const; - void EmitFrameMoves(intptr_t BaseLabelPtr, + void EmitFrameMoves(intptr_t BaseLabelPtr, const std::vector<MachineMove> &Moves) const; - + unsigned char* EmitCommonEHFrame(const Function* Personality) const; - unsigned char* EmitEHFrame(const Function* Personality, + unsigned char* EmitEHFrame(const Function* Personality, unsigned char* StartBufferPtr, - unsigned char* StartFunction, + unsigned char* StartFunction, unsigned char* EndFunction, unsigned char* ExceptionTable) const; - + public: - + JITDwarfEmitter(JIT& jit); - - unsigned char* EmitDwarfTable(MachineFunction& F, + + unsigned char* EmitDwarfTable(MachineFunction& F, JITCodeEmitter& JCE, unsigned char* StartFunction, unsigned char* EndFunction, unsigned char* &EHFramePtr); - - + + void setModuleInfo(MachineModuleInfo* Info) { MMI = Info; } diff --git a/contrib/llvm/lib/ExecutionEngine/JIT/JITEmitter.cpp b/contrib/llvm/lib/ExecutionEngine/JIT/JITEmitter.cpp index 4cd8757..d046b8a 100644 --- a/contrib/llvm/lib/ExecutionEngine/JIT/JITEmitter.cpp +++ b/contrib/llvm/lib/ExecutionEngine/JIT/JITEmitter.cpp @@ -123,17 +123,18 @@ namespace { return FunctionToLazyStubMap; } - GlobalToIndirectSymMapTy& getGlobalToIndirectSymMap(const MutexGuard& locked) { - assert(locked.holds(TheJIT->lock)); + GlobalToIndirectSymMapTy& getGlobalToIndirectSymMap(const MutexGuard& lck) { + assert(lck.holds(TheJIT->lock)); return GlobalToIndirectSymMap; } - pair<void *, Function *> LookupFunctionFromCallSite( + std::pair<void *, Function *> LookupFunctionFromCallSite( const MutexGuard &locked, void *CallSite) const { assert(locked.holds(TheJIT->lock)); - // The address given to us for the stub may not be exactly right, it might be - // a little bit after the stub. As such, use upper_bound to find it. + // The address given to us for the stub may not be exactly right, it + // might be a little bit after the stub. As such, use upper_bound to + // find it. CallSiteToFunctionMapTy::const_iterator I = CallSiteToFunctionMap.upper_bound(CallSite); assert(I != CallSiteToFunctionMap.begin() && @@ -645,7 +646,7 @@ void *JITResolver::JITCompilerFn(void *Stub) { // The address given to us for the stub may not be exactly right, it might // be a little bit after the stub. As such, use upper_bound to find it. - pair<void*, Function*> I = + std::pair<void*, Function*> I = JR->state.LookupFunctionFromCallSite(locked, Stub); F = I.second; ActualPtr = I.first; @@ -659,7 +660,8 @@ void *JITResolver::JITCompilerFn(void *Stub) { // If lazy compilation is disabled, emit a useful error message and abort. if (!JR->TheJIT->isCompilingLazily()) { - report_fatal_error("LLVM JIT requested to do lazy compilation of function '" + report_fatal_error("LLVM JIT requested to do lazy compilation of" + " function '" + F->getName() + "' when lazy compiles are disabled!"); } @@ -745,7 +747,7 @@ void *JITEmitter::getPointerToGVIndirectSym(GlobalValue *V, void *Reference) { void JITEmitter::processDebugLoc(DebugLoc DL, bool BeforePrintingInsn) { if (DL.isUnknown()) return; if (!BeforePrintingInsn) return; - + const LLVMContext &Context = EmissionDetails.MF->getFunction()->getContext(); if (DL.getScope(Context) != 0 && PrevDL != DL) { @@ -781,7 +783,7 @@ void JITEmitter::startFunction(MachineFunction &F) { uintptr_t ActualSize = 0; // Set the memory writable, if it's not already MemMgr->setMemoryWritable(); - + if (SizeEstimate > 0) { // SizeEstimate will be non-zero on reallocation attempts. ActualSize = SizeEstimate; @@ -859,7 +861,8 @@ bool JITEmitter::finishFunction(MachineFunction &F) { } else if (MR.isBasicBlock()) { ResultPtr = (void*)getMachineBasicBlockAddress(MR.getBasicBlock()); } else if (MR.isConstantPoolIndex()) { - ResultPtr = (void*)getConstantPoolEntryAddress(MR.getConstantPoolIndex()); + ResultPtr = + (void*)getConstantPoolEntryAddress(MR.getConstantPoolIndex()); } else { assert(MR.isJumpTableIndex()); ResultPtr=(void*)getJumpTableEntryAddress(MR.getJumpTableIndex()); @@ -985,7 +988,7 @@ bool JITEmitter::finishFunction(MachineFunction &F) { CurBufferPtr = SavedCurBufferPtr; if (JITExceptionHandling) { - TheJIT->RegisterTable(FrameRegister); + TheJIT->RegisterTable(F.getFunction(), FrameRegister); } if (JITEmitDebugInfo) { @@ -1033,8 +1036,9 @@ void JITEmitter::deallocateMemForFunction(const Function *F) { EmittedFunctions.erase(Emitted); } - // TODO: Do we need to unregister exception handling information from libgcc - // here? + if(JITExceptionHandling) { + TheJIT->DeregisterTable(F); + } if (JITEmitDebugInfo) { DR->UnregisterFunction(F); @@ -1129,7 +1133,7 @@ void JITEmitter::emitJumpTableInfo(MachineJumpTableInfo *MJTI) { const std::vector<MachineJumpTableEntry> &JT = MJTI->getJumpTables(); if (JT.empty() || JumpTableBase == 0) return; - + switch (MJTI->getEntryKind()) { case MachineJumpTableInfo::EK_Inline: return; @@ -1138,11 +1142,11 @@ void JITEmitter::emitJumpTableInfo(MachineJumpTableInfo *MJTI) { // .word LBB123 assert(MJTI->getEntrySize(*TheJIT->getTargetData()) == sizeof(void*) && "Cross JIT'ing?"); - + // For each jump table, map each target in the jump table to the address of // an emitted MachineBasicBlock. intptr_t *SlotPtr = (intptr_t*)JumpTableBase; - + for (unsigned i = 0, e = JT.size(); i != e; ++i) { const std::vector<MachineBasicBlock*> &MBBs = JT[i].MBBs; // Store the address of the basic block for this jump table slot in the @@ -1152,7 +1156,7 @@ void JITEmitter::emitJumpTableInfo(MachineJumpTableInfo *MJTI) { } break; } - + case MachineJumpTableInfo::EK_Custom32: case MachineJumpTableInfo::EK_GPRel32BlockAddress: case MachineJumpTableInfo::EK_LabelDifference32: { diff --git a/contrib/llvm/lib/ExecutionEngine/JIT/OProfileJITEventListener.cpp b/contrib/llvm/lib/ExecutionEngine/JIT/OProfileJITEventListener.cpp index 670fa7d..9a9ed6d 100644 --- a/contrib/llvm/lib/ExecutionEngine/JIT/OProfileJITEventListener.cpp +++ b/contrib/llvm/lib/ExecutionEngine/JIT/OProfileJITEventListener.cpp @@ -108,8 +108,8 @@ void OProfileJITEventListener::NotifyFunctionEmitted( if (op_write_native_code(Agent, F.getName().data(), reinterpret_cast<uint64_t>(FnStart), FnStart, FnSize) == -1) { - DEBUG(dbgs() << "Failed to tell OProfile about native function " - << F.getName() << " at [" + DEBUG(dbgs() << "Failed to tell OProfile about native function " + << F.getName() << " at [" << FnStart << "-" << ((char*)FnStart + FnSize) << "]\n"); return; } @@ -153,9 +153,9 @@ void OProfileJITEventListener::NotifyFunctionEmitted( if (op_write_debug_line_info(Agent, FnStart, LineInfo.size(), &*LineInfo.begin()) == -1) { - DEBUG(dbgs() + DEBUG(dbgs() << "Failed to tell OProfile about line numbers for native function " - << F.getName() << " at [" + << F.getName() << " at [" << FnStart << "-" << ((char*)FnStart + FnSize) << "]\n"); } } diff --git a/contrib/llvm/lib/ExecutionEngine/JIT/TargetSelect.cpp b/contrib/llvm/lib/ExecutionEngine/JIT/TargetSelect.cpp index 6b7173c..8d92ab0 100644 --- a/contrib/llvm/lib/ExecutionEngine/JIT/TargetSelect.cpp +++ b/contrib/llvm/lib/ExecutionEngine/JIT/TargetSelect.cpp @@ -84,7 +84,7 @@ TargetMachine *JIT::selectTarget(Module *Mod, } // Allocate a target... - TargetMachine *Target = + TargetMachine *Target = TheTarget->createTargetMachine(TheTriple.getTriple(), FeaturesStr); assert(Target && "Could not allocate target machine!"); return Target; diff --git a/contrib/llvm/lib/ExecutionEngine/MCJIT/Intercept.cpp b/contrib/llvm/lib/ExecutionEngine/MCJIT/Intercept.cpp new file mode 100644 index 0000000..e431c84 --- /dev/null +++ b/contrib/llvm/lib/ExecutionEngine/MCJIT/Intercept.cpp @@ -0,0 +1,161 @@ +//===-- Intercept.cpp - System function interception routines -------------===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// If a function call occurs to an external function, the JIT is designed to use +// the dynamic loader interface to find a function to call. This is useful for +// calling system calls and library functions that are not available in LLVM. +// Some system calls, however, need to be handled specially. For this reason, +// we intercept some of them here and use our own stubs to handle them. +// +//===----------------------------------------------------------------------===// + +#include "MCJIT.h" +#include "llvm/Support/ErrorHandling.h" +#include "llvm/Support/DynamicLibrary.h" +#include "llvm/Config/config.h" +using namespace llvm; + +// AtExitHandlers - List of functions to call when the program exits, +// registered with the atexit() library function. +static std::vector<void (*)()> AtExitHandlers; + +/// runAtExitHandlers - Run any functions registered by the program's +/// calls to atexit(3), which we intercept and store in +/// AtExitHandlers. +/// +static void runAtExitHandlers() { + while (!AtExitHandlers.empty()) { + void (*Fn)() = AtExitHandlers.back(); + AtExitHandlers.pop_back(); + Fn(); + } +} + +//===----------------------------------------------------------------------===// +// Function stubs that are invoked instead of certain library calls +//===----------------------------------------------------------------------===// + +// Force the following functions to be linked in to anything that uses the +// JIT. This is a hack designed to work around the all-too-clever Glibc +// strategy of making these functions work differently when inlined vs. when +// not inlined, and hiding their real definitions in a separate archive file +// that the dynamic linker can't see. For more info, search for +// 'libc_nonshared.a' on Google, or read http://llvm.org/PR274. +#if defined(__linux__) +#if defined(HAVE_SYS_STAT_H) +#include <sys/stat.h> +#endif +#include <fcntl.h> +/* stat functions are redirecting to __xstat with a version number. On x86-64 + * linking with libc_nonshared.a and -Wl,--export-dynamic doesn't make 'stat' + * available as an exported symbol, so we have to add it explicitly. + */ +namespace { +class StatSymbols { +public: + StatSymbols() { + sys::DynamicLibrary::AddSymbol("stat", (void*)(intptr_t)stat); + sys::DynamicLibrary::AddSymbol("fstat", (void*)(intptr_t)fstat); + sys::DynamicLibrary::AddSymbol("lstat", (void*)(intptr_t)lstat); + sys::DynamicLibrary::AddSymbol("stat64", (void*)(intptr_t)stat64); + sys::DynamicLibrary::AddSymbol("\x1stat64", (void*)(intptr_t)stat64); + sys::DynamicLibrary::AddSymbol("\x1open64", (void*)(intptr_t)open64); + sys::DynamicLibrary::AddSymbol("\x1lseek64", (void*)(intptr_t)lseek64); + sys::DynamicLibrary::AddSymbol("fstat64", (void*)(intptr_t)fstat64); + sys::DynamicLibrary::AddSymbol("lstat64", (void*)(intptr_t)lstat64); + sys::DynamicLibrary::AddSymbol("atexit", (void*)(intptr_t)atexit); + sys::DynamicLibrary::AddSymbol("mknod", (void*)(intptr_t)mknod); + } +}; +} +static StatSymbols initStatSymbols; +#endif // __linux__ + +// jit_exit - Used to intercept the "exit" library call. +static void jit_exit(int Status) { + runAtExitHandlers(); // Run atexit handlers... + exit(Status); +} + +// jit_atexit - Used to intercept the "atexit" library call. +static int jit_atexit(void (*Fn)()) { + AtExitHandlers.push_back(Fn); // Take note of atexit handler... + return 0; // Always successful +} + +static int jit_noop() { + return 0; +} + +//===----------------------------------------------------------------------===// +// +/// getPointerToNamedFunction - This method returns the address of the specified +/// function by using the dynamic loader interface. As such it is only useful +/// for resolving library symbols, not code generated symbols. +/// +void *MCJIT::getPointerToNamedFunction(const std::string &Name, + bool AbortOnFailure) { + if (!isSymbolSearchingDisabled()) { + // Check to see if this is one of the functions we want to intercept. Note, + // we cast to intptr_t here to silence a -pedantic warning that complains + // about casting a function pointer to a normal pointer. + if (Name == "exit") return (void*)(intptr_t)&jit_exit; + if (Name == "atexit") return (void*)(intptr_t)&jit_atexit; + + // We should not invoke parent's ctors/dtors from generated main()! + // On Mingw and Cygwin, the symbol __main is resolved to + // callee's(eg. tools/lli) one, to invoke wrong duplicated ctors + // (and register wrong callee's dtors with atexit(3)). + // We expect ExecutionEngine::runStaticConstructorsDestructors() + // is called before ExecutionEngine::runFunctionAsMain() is called. + if (Name == "__main") return (void*)(intptr_t)&jit_noop; + + const char *NameStr = Name.c_str(); + // If this is an asm specifier, skip the sentinal. + if (NameStr[0] == 1) ++NameStr; + + // If it's an external function, look it up in the process image... + void *Ptr = sys::DynamicLibrary::SearchForAddressOfSymbol(NameStr); + if (Ptr) return Ptr; + + // If it wasn't found and if it starts with an underscore ('_') character, + // and has an asm specifier, try again without the underscore. + if (Name[0] == 1 && NameStr[0] == '_') { + Ptr = sys::DynamicLibrary::SearchForAddressOfSymbol(NameStr+1); + if (Ptr) return Ptr; + } + + // Darwin/PPC adds $LDBLStub suffixes to various symbols like printf. These + // are references to hidden visibility symbols that dlsym cannot resolve. + // If we have one of these, strip off $LDBLStub and try again. +#if defined(__APPLE__) && defined(__ppc__) + if (Name.size() > 9 && Name[Name.size()-9] == '$' && + memcmp(&Name[Name.size()-8], "LDBLStub", 8) == 0) { + // First try turning $LDBLStub into $LDBL128. If that fails, strip it off. + // This mirrors logic in libSystemStubs.a. + std::string Prefix = std::string(Name.begin(), Name.end()-9); + if (void *Ptr = getPointerToNamedFunction(Prefix+"$LDBL128", false)) + return Ptr; + if (void *Ptr = getPointerToNamedFunction(Prefix, false)) + return Ptr; + } +#endif + } + + /// If a LazyFunctionCreator is installed, use it to get/create the function. + if (LazyFunctionCreator) + if (void *RP = LazyFunctionCreator(Name)) + return RP; + + if (AbortOnFailure) { + report_fatal_error("Program used external function '"+Name+ + "' which could not be resolved!"); + } + return 0; +} diff --git a/contrib/llvm/lib/ExecutionEngine/MCJIT/MCJIT.cpp b/contrib/llvm/lib/ExecutionEngine/MCJIT/MCJIT.cpp index f1e9dab..3d4ee36 100644 --- a/contrib/llvm/lib/ExecutionEngine/MCJIT/MCJIT.cpp +++ b/contrib/llvm/lib/ExecutionEngine/MCJIT/MCJIT.cpp @@ -1,4 +1,4 @@ -//===-- JIT.cpp - MC-based Just-in-Time Compiler --------------------------===// +//===-- MCJIT.cpp - MC-based Just-in-Time Compiler ------------------------===// // // The LLVM Compiler Infrastructure // @@ -8,10 +8,17 @@ //===----------------------------------------------------------------------===// #include "MCJIT.h" +#include "MCJITMemoryManager.h" +#include "llvm/DerivedTypes.h" +#include "llvm/Function.h" #include "llvm/ExecutionEngine/GenericValue.h" #include "llvm/ExecutionEngine/MCJIT.h" +#include "llvm/ExecutionEngine/JITMemoryManager.h" +#include "llvm/MC/MCAsmInfo.h" #include "llvm/Support/ErrorHandling.h" #include "llvm/Support/DynamicLibrary.h" +#include "llvm/Support/MemoryBuffer.h" +#include "llvm/Target/TargetData.h" using namespace llvm; @@ -51,20 +58,47 @@ ExecutionEngine *MCJIT::createJIT(Module *M, // If the target supports JIT code generation, create the JIT. if (TargetJITInfo *TJ = TM->getJITInfo()) - return new MCJIT(M, *TM, *TJ, JMM, OptLevel, GVsWithCode); + return new MCJIT(M, TM, *TJ, new MCJITMemoryManager(JMM), OptLevel, + GVsWithCode); if (ErrorStr) *ErrorStr = "target does not support JIT code generation"; return 0; } -MCJIT::MCJIT(Module *M, TargetMachine &tm, TargetJITInfo &tji, - JITMemoryManager *JMM, CodeGenOpt::Level OptLevel, +MCJIT::MCJIT(Module *m, TargetMachine *tm, TargetJITInfo &tji, + RTDyldMemoryManager *MM, CodeGenOpt::Level OptLevel, bool AllocateGVsWithCode) - : ExecutionEngine(M) { + : ExecutionEngine(m), TM(tm), MemMgr(MM), M(m), OS(Buffer), Dyld(MM) { + + PM.add(new TargetData(*TM->getTargetData())); + + // Turn the machine code intermediate representation into bytes in memory + // that may be executed. + if (TM->addPassesToEmitMC(PM, Ctx, OS, CodeGenOpt::Default, false)) { + report_fatal_error("Target does not support MC emission!"); + } + + // Initialize passes. + // FIXME: When we support multiple modules, we'll want to move the code + // gen and finalization out of the constructor here and do it more + // on-demand as part of getPointerToFunction(). + PM.run(*M); + // Flush the output buffer so the SmallVector gets its data. + OS.flush(); + + // Load the object into the dynamic linker. + // FIXME: It would be nice to avoid making yet another copy. + MemoryBuffer *MB = MemoryBuffer::getMemBufferCopy(StringRef(Buffer.data(), + Buffer.size())); + if (Dyld.loadObject(MB)) + report_fatal_error(Dyld.getErrorString()); + // Resolve any relocations. + Dyld.resolveRelocations(); } MCJIT::~MCJIT() { + delete MemMgr; } void *MCJIT::getPointerToBasicBlock(BasicBlock *BB) { @@ -73,8 +107,15 @@ void *MCJIT::getPointerToBasicBlock(BasicBlock *BB) { } void *MCJIT::getPointerToFunction(Function *F) { - report_fatal_error("not yet implemented"); - return 0; + if (F->isDeclaration() || F->hasAvailableExternallyLinkage()) { + bool AbortOnFailure = !F->hasExternalWeakLinkage(); + void *Addr = getPointerToNamedFunction(F->getName(), AbortOnFailure); + addGlobalMapping(F, Addr); + return Addr; + } + + Twine Name = TM->getMCAsmInfo()->getGlobalPrefix() + F->getName(); + return (void*)Dyld.getSymbolAddress(Name.str()); } void *MCJIT::recompileAndRelinkFunction(Function *F) { @@ -87,6 +128,102 @@ void MCJIT::freeMachineCodeForFunction(Function *F) { GenericValue MCJIT::runFunction(Function *F, const std::vector<GenericValue> &ArgValues) { - report_fatal_error("not yet implemented"); + assert(F && "Function *F was null at entry to run()"); + + void *FPtr = getPointerToFunction(F); + assert(FPtr && "Pointer to fn's code was null after getPointerToFunction"); + const FunctionType *FTy = F->getFunctionType(); + const Type *RetTy = FTy->getReturnType(); + + assert((FTy->getNumParams() == ArgValues.size() || + (FTy->isVarArg() && FTy->getNumParams() <= ArgValues.size())) && + "Wrong number of arguments passed into function!"); + assert(FTy->getNumParams() == ArgValues.size() && + "This doesn't support passing arguments through varargs (yet)!"); + + // Handle some common cases first. These cases correspond to common `main' + // prototypes. + if (RetTy->isIntegerTy(32) || RetTy->isVoidTy()) { + switch (ArgValues.size()) { + case 3: + if (FTy->getParamType(0)->isIntegerTy(32) && + FTy->getParamType(1)->isPointerTy() && + FTy->getParamType(2)->isPointerTy()) { + int (*PF)(int, char **, const char **) = + (int(*)(int, char **, const char **))(intptr_t)FPtr; + + // Call the function. + GenericValue rv; + rv.IntVal = APInt(32, PF(ArgValues[0].IntVal.getZExtValue(), + (char **)GVTOP(ArgValues[1]), + (const char **)GVTOP(ArgValues[2]))); + return rv; + } + break; + case 2: + if (FTy->getParamType(0)->isIntegerTy(32) && + FTy->getParamType(1)->isPointerTy()) { + int (*PF)(int, char **) = (int(*)(int, char **))(intptr_t)FPtr; + + // Call the function. + GenericValue rv; + rv.IntVal = APInt(32, PF(ArgValues[0].IntVal.getZExtValue(), + (char **)GVTOP(ArgValues[1]))); + return rv; + } + break; + case 1: + if (FTy->getNumParams() == 1 && + FTy->getParamType(0)->isIntegerTy(32)) { + GenericValue rv; + int (*PF)(int) = (int(*)(int))(intptr_t)FPtr; + rv.IntVal = APInt(32, PF(ArgValues[0].IntVal.getZExtValue())); + return rv; + } + break; + } + } + + // Handle cases where no arguments are passed first. + if (ArgValues.empty()) { + GenericValue rv; + switch (RetTy->getTypeID()) { + default: llvm_unreachable("Unknown return type for function call!"); + case Type::IntegerTyID: { + unsigned BitWidth = cast<IntegerType>(RetTy)->getBitWidth(); + if (BitWidth == 1) + rv.IntVal = APInt(BitWidth, ((bool(*)())(intptr_t)FPtr)()); + else if (BitWidth <= 8) + rv.IntVal = APInt(BitWidth, ((char(*)())(intptr_t)FPtr)()); + else if (BitWidth <= 16) + rv.IntVal = APInt(BitWidth, ((short(*)())(intptr_t)FPtr)()); + else if (BitWidth <= 32) + rv.IntVal = APInt(BitWidth, ((int(*)())(intptr_t)FPtr)()); + else if (BitWidth <= 64) + rv.IntVal = APInt(BitWidth, ((int64_t(*)())(intptr_t)FPtr)()); + else + llvm_unreachable("Integer types > 64 bits not supported"); + return rv; + } + case Type::VoidTyID: + rv.IntVal = APInt(32, ((int(*)())(intptr_t)FPtr)()); + return rv; + case Type::FloatTyID: + rv.FloatVal = ((float(*)())(intptr_t)FPtr)(); + return rv; + case Type::DoubleTyID: + rv.DoubleVal = ((double(*)())(intptr_t)FPtr)(); + return rv; + case Type::X86_FP80TyID: + case Type::FP128TyID: + case Type::PPC_FP128TyID: + llvm_unreachable("long double not supported yet"); + return rv; + case Type::PointerTyID: + return PTOGV(((void*(*)())(intptr_t)FPtr)()); + } + } + + assert("Full-featured argument passing not supported yet!"); return GenericValue(); } diff --git a/contrib/llvm/lib/ExecutionEngine/MCJIT/MCJIT.h b/contrib/llvm/lib/ExecutionEngine/MCJIT/MCJIT.h index cd1f989..1b50766 100644 --- a/contrib/llvm/lib/ExecutionEngine/MCJIT/MCJIT.h +++ b/contrib/llvm/lib/ExecutionEngine/MCJIT/MCJIT.h @@ -10,14 +10,37 @@ #ifndef LLVM_LIB_EXECUTIONENGINE_MCJIT_H #define LLVM_LIB_EXECUTIONENGINE_MCJIT_H +#include "llvm/PassManager.h" #include "llvm/ExecutionEngine/ExecutionEngine.h" +#include "llvm/ExecutionEngine/RuntimeDyld.h" +#include "llvm/ADT/SmallVector.h" +#include "llvm/Support/raw_ostream.h" namespace llvm { +// FIXME: This makes all kinds of horrible assumptions for the time being, +// like only having one module, not needing to worry about multi-threading, +// blah blah. Purely in get-it-up-and-limping mode for now. + class MCJIT : public ExecutionEngine { - MCJIT(Module *M, TargetMachine &tm, TargetJITInfo &tji, - JITMemoryManager *JMM, CodeGenOpt::Level OptLevel, + MCJIT(Module *M, TargetMachine *tm, TargetJITInfo &tji, + RTDyldMemoryManager *MemMgr, CodeGenOpt::Level OptLevel, bool AllocateGVsWithCode); + + TargetMachine *TM; + MCContext *Ctx; + RTDyldMemoryManager *MemMgr; + + // FIXME: These may need moved to a separate 'jitstate' member like the + // non-MC JIT does for multithreading and such. Just keep them here for now. + PassManager PM; + Module *M; + // FIXME: This really doesn't belong here. + SmallVector<char, 4096> Buffer; // Working buffer into which we JIT. + raw_svector_ostream OS; + + RuntimeDyld Dyld; + public: ~MCJIT(); @@ -35,6 +58,16 @@ public: virtual GenericValue runFunction(Function *F, const std::vector<GenericValue> &ArgValues); + /// getPointerToNamedFunction - This method returns the address of the + /// specified function by using the dlsym function call. As such it is only + /// useful for resolving library symbols, not code generated symbols. + /// + /// If AbortOnFailure is false and no function with the given name is + /// found, this function silently returns a null pointer. Otherwise, + /// it prints a message to stderr and aborts. + /// + void *getPointerToNamedFunction(const std::string &Name, + bool AbortOnFailure = true); /// @} /// @name (Private) Registration Interfaces /// @{ diff --git a/contrib/llvm/lib/ExecutionEngine/MCJIT/MCJITMemoryManager.h b/contrib/llvm/lib/ExecutionEngine/MCJIT/MCJITMemoryManager.h new file mode 100644 index 0000000..e3c6fda --- /dev/null +++ b/contrib/llvm/lib/ExecutionEngine/MCJIT/MCJITMemoryManager.h @@ -0,0 +1,59 @@ +//===-- MCJITMemoryManager.h - Definition for the Memory Manager ---C++ -*-===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// + +#ifndef LLVM_LIB_EXECUTIONENGINE_MCJITMEMORYMANAGER_H +#define LLVM_LIB_EXECUTIONENGINE_MCJITMEMORYMANAGER_H + +#include "llvm/Module.h" +#include "llvm/ExecutionEngine/JITMemoryManager.h" +#include "llvm/ExecutionEngine/RuntimeDyld.h" +#include <assert.h> + +namespace llvm { + +// The MCJIT memory manager is a layer between the standard JITMemoryManager +// and the RuntimeDyld interface that maps objects, by name, onto their +// matching LLVM IR counterparts in the module(s) being compiled. +class MCJITMemoryManager : public RTDyldMemoryManager { + JITMemoryManager *JMM; + + // FIXME: Multiple modules. + Module *M; +public: + MCJITMemoryManager(JITMemoryManager *jmm) : JMM(jmm) {} + + // Allocate ActualSize bytes, or more, for the named function. Return + // a pointer to the allocated memory and update Size to reflect how much + // memory was acutally allocated. + uint8_t *startFunctionBody(const char *Name, uintptr_t &Size) { + // FIXME: This should really reference the MCAsmInfo to get the global + // prefix. + if (Name[0] == '_') ++Name; + Function *F = M->getFunction(Name); + assert(F && "No matching function in JIT IR Module!"); + return JMM->startFunctionBody(F, Size); + } + + // Mark the end of the function, including how much of the allocated + // memory was actually used. + void endFunctionBody(const char *Name, uint8_t *FunctionStart, + uint8_t *FunctionEnd) { + // FIXME: This should really reference the MCAsmInfo to get the global + // prefix. + if (Name[0] == '_') ++Name; + Function *F = M->getFunction(Name); + assert(F && "No matching function in JIT IR Module!"); + JMM->endFunctionBody(F, FunctionStart, FunctionEnd); + } + +}; + +} // End llvm namespace + +#endif diff --git a/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/CMakeLists.txt b/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/CMakeLists.txt new file mode 100644 index 0000000..9e53f87 --- /dev/null +++ b/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/CMakeLists.txt @@ -0,0 +1,3 @@ +add_llvm_library(LLVMRuntimeDyld + RuntimeDyld.cpp + ) diff --git a/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/Makefile b/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/Makefile new file mode 100644 index 0000000..5d6f26d --- /dev/null +++ b/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/Makefile @@ -0,0 +1,13 @@ +##===- lib/ExecutionEngine/MCJIT/Makefile ------------------*- Makefile -*-===## +# +# The LLVM Compiler Infrastructure +# +# This file is distributed under the University of Illinois Open Source +# License. See LICENSE.TXT for details. +# +##===----------------------------------------------------------------------===## + +LEVEL = ../../.. +LIBRARYNAME = LLVMRuntimeDyld + +include $(LEVEL)/Makefile.common diff --git a/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyld.cpp b/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyld.cpp new file mode 100644 index 0000000..065e5e3 --- /dev/null +++ b/contrib/llvm/lib/ExecutionEngine/RuntimeDyld/RuntimeDyld.cpp @@ -0,0 +1,669 @@ +//===-- RuntimeDyld.h - Run-time dynamic linker for MC-JIT ------*- C++ -*-===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// Implementation of the MC-JIT runtime dynamic linker. +// +//===----------------------------------------------------------------------===// + +#define DEBUG_TYPE "dyld" +#include "llvm/ADT/OwningPtr.h" +#include "llvm/ADT/SmallVector.h" +#include "llvm/ADT/StringMap.h" +#include "llvm/ADT/StringRef.h" +#include "llvm/ADT/STLExtras.h" +#include "llvm/ADT/Twine.h" +#include "llvm/ExecutionEngine/RuntimeDyld.h" +#include "llvm/Object/MachOObject.h" +#include "llvm/Support/Debug.h" +#include "llvm/Support/ErrorHandling.h" +#include "llvm/Support/Format.h" +#include "llvm/Support/Memory.h" +#include "llvm/Support/MemoryBuffer.h" +#include "llvm/Support/system_error.h" +#include "llvm/Support/raw_ostream.h" +using namespace llvm; +using namespace llvm::object; + +// Empty out-of-line virtual destructor as the key function. +RTDyldMemoryManager::~RTDyldMemoryManager() {} + +namespace llvm { +class RuntimeDyldImpl { + unsigned CPUType; + unsigned CPUSubtype; + + // The MemoryManager to load objects into. + RTDyldMemoryManager *MemMgr; + + // FIXME: This all assumes we're dealing with external symbols for anything + // explicitly referenced. I.e., we can index by name and things + // will work out. In practice, this may not be the case, so we + // should find a way to effectively generalize. + + // For each function, we have a MemoryBlock of it's instruction data. + StringMap<sys::MemoryBlock> Functions; + + // Master symbol table. As modules are loaded and external symbols are + // resolved, their addresses are stored here. + StringMap<uint8_t*> SymbolTable; + + // For each symbol, keep a list of relocations based on it. Anytime + // its address is reassigned (the JIT re-compiled the function, e.g.), + // the relocations get re-resolved. + struct RelocationEntry { + std::string Target; // Object this relocation is contained in. + uint64_t Offset; // Offset into the object for the relocation. + uint32_t Data; // Second word of the raw macho relocation entry. + int64_t Addend; // Addend encoded in the instruction itself, if any. + bool isResolved; // Has this relocation been resolved previously? + + RelocationEntry(StringRef t, uint64_t offset, uint32_t data, int64_t addend) + : Target(t), Offset(offset), Data(data), Addend(addend), + isResolved(false) {} + }; + typedef SmallVector<RelocationEntry, 4> RelocationList; + StringMap<RelocationList> Relocations; + + // FIXME: Also keep a map of all the relocations contained in an object. Use + // this to dynamically answer whether all of the relocations in it have + // been resolved or not. + + bool HasError; + std::string ErrorStr; + + // Set the error state and record an error string. + bool Error(const Twine &Msg) { + ErrorStr = Msg.str(); + HasError = true; + return true; + } + + void extractFunction(StringRef Name, uint8_t *StartAddress, + uint8_t *EndAddress); + bool resolveRelocation(uint8_t *Address, uint8_t *Value, bool isPCRel, + unsigned Type, unsigned Size); + bool resolveX86_64Relocation(uintptr_t Address, uintptr_t Value, bool isPCRel, + unsigned Type, unsigned Size); + bool resolveARMRelocation(uintptr_t Address, uintptr_t Value, bool isPCRel, + unsigned Type, unsigned Size); + + bool loadSegment32(const MachOObject *Obj, + const MachOObject::LoadCommandInfo *SegmentLCI, + const InMemoryStruct<macho::SymtabLoadCommand> &SymtabLC); + bool loadSegment64(const MachOObject *Obj, + const MachOObject::LoadCommandInfo *SegmentLCI, + const InMemoryStruct<macho::SymtabLoadCommand> &SymtabLC); + +public: + RuntimeDyldImpl(RTDyldMemoryManager *mm) : MemMgr(mm), HasError(false) {} + + bool loadObject(MemoryBuffer *InputBuffer); + + void *getSymbolAddress(StringRef Name) { + // FIXME: Just look up as a function for now. Overly simple of course. + // Work in progress. + return SymbolTable.lookup(Name); + } + + void resolveRelocations(); + + void reassignSymbolAddress(StringRef Name, uint8_t *Addr); + + // Is the linker in an error state? + bool hasError() { return HasError; } + + // Mark the error condition as handled and continue. + void clearError() { HasError = false; } + + // Get the error message. + StringRef getErrorString() { return ErrorStr; } +}; + +void RuntimeDyldImpl::extractFunction(StringRef Name, uint8_t *StartAddress, + uint8_t *EndAddress) { + // Allocate memory for the function via the memory manager. + uintptr_t Size = EndAddress - StartAddress + 1; + uint8_t *Mem = MemMgr->startFunctionBody(Name.data(), Size); + assert(Size >= (uint64_t)(EndAddress - StartAddress + 1) && + "Memory manager failed to allocate enough memory!"); + // Copy the function payload into the memory block. + memcpy(Mem, StartAddress, EndAddress - StartAddress + 1); + MemMgr->endFunctionBody(Name.data(), Mem, Mem + Size); + // Remember where we put it. + Functions[Name] = sys::MemoryBlock(Mem, Size); + // Default the assigned address for this symbol to wherever this + // allocated it. + SymbolTable[Name] = Mem; + DEBUG(dbgs() << " allocated to " << Mem << "\n"); +} + +bool RuntimeDyldImpl:: +resolveRelocation(uint8_t *Address, uint8_t *Value, bool isPCRel, + unsigned Type, unsigned Size) { + // This just dispatches to the proper target specific routine. + switch (CPUType) { + default: assert(0 && "Unsupported CPU type!"); + case mach::CTM_x86_64: + return resolveX86_64Relocation((uintptr_t)Address, (uintptr_t)Value, + isPCRel, Type, Size); + case mach::CTM_ARM: + return resolveARMRelocation((uintptr_t)Address, (uintptr_t)Value, + isPCRel, Type, Size); + } + llvm_unreachable(""); +} + +bool RuntimeDyldImpl:: +resolveX86_64Relocation(uintptr_t Address, uintptr_t Value, + bool isPCRel, unsigned Type, + unsigned Size) { + // If the relocation is PC-relative, the value to be encoded is the + // pointer difference. + if (isPCRel) + // FIXME: It seems this value needs to be adjusted by 4 for an effective PC + // address. Is that expected? Only for branches, perhaps? + Value -= Address + 4; + + switch(Type) { + default: + llvm_unreachable("Invalid relocation type!"); + case macho::RIT_X86_64_Unsigned: + case macho::RIT_X86_64_Branch: { + // Mask in the target value a byte at a time (we don't have an alignment + // guarantee for the target address, so this is safest). + uint8_t *p = (uint8_t*)Address; + for (unsigned i = 0; i < Size; ++i) { + *p++ = (uint8_t)Value; + Value >>= 8; + } + return false; + } + case macho::RIT_X86_64_Signed: + case macho::RIT_X86_64_GOTLoad: + case macho::RIT_X86_64_GOT: + case macho::RIT_X86_64_Subtractor: + case macho::RIT_X86_64_Signed1: + case macho::RIT_X86_64_Signed2: + case macho::RIT_X86_64_Signed4: + case macho::RIT_X86_64_TLV: + return Error("Relocation type not implemented yet!"); + } + return false; +} + +bool RuntimeDyldImpl::resolveARMRelocation(uintptr_t Address, uintptr_t Value, + bool isPCRel, unsigned Type, + unsigned Size) { + // If the relocation is PC-relative, the value to be encoded is the + // pointer difference. + if (isPCRel) { + Value -= Address; + // ARM PCRel relocations have an effective-PC offset of two instructions + // (four bytes in Thumb mode, 8 bytes in ARM mode). + // FIXME: For now, assume ARM mode. + Value -= 8; + } + + switch(Type) { + default: + llvm_unreachable("Invalid relocation type!"); + case macho::RIT_Vanilla: { + llvm_unreachable("Invalid relocation type!"); + // Mask in the target value a byte at a time (we don't have an alignment + // guarantee for the target address, so this is safest). + uint8_t *p = (uint8_t*)Address; + for (unsigned i = 0; i < Size; ++i) { + *p++ = (uint8_t)Value; + Value >>= 8; + } + break; + } + case macho::RIT_ARM_Branch24Bit: { + // Mask the value into the target address. We know instructions are + // 32-bit aligned, so we can do it all at once. + uint32_t *p = (uint32_t*)Address; + // The low two bits of the value are not encoded. + Value >>= 2; + // Mask the value to 24 bits. + Value &= 0xffffff; + // FIXME: If the destination is a Thumb function (and the instruction + // is a non-predicated BL instruction), we need to change it to a BLX + // instruction instead. + + // Insert the value into the instruction. + *p = (*p & ~0xffffff) | Value; + break; + } + case macho::RIT_ARM_ThumbBranch22Bit: + case macho::RIT_ARM_ThumbBranch32Bit: + case macho::RIT_ARM_Half: + case macho::RIT_ARM_HalfDifference: + case macho::RIT_Pair: + case macho::RIT_Difference: + case macho::RIT_ARM_LocalDifference: + case macho::RIT_ARM_PreboundLazyPointer: + return Error("Relocation type not implemented yet!"); + } + return false; +} + +bool RuntimeDyldImpl:: +loadSegment32(const MachOObject *Obj, + const MachOObject::LoadCommandInfo *SegmentLCI, + const InMemoryStruct<macho::SymtabLoadCommand> &SymtabLC) { + InMemoryStruct<macho::SegmentLoadCommand> SegmentLC; + Obj->ReadSegmentLoadCommand(*SegmentLCI, SegmentLC); + if (!SegmentLC) + return Error("unable to load segment load command"); + + for (unsigned SectNum = 0; SectNum != SegmentLC->NumSections; ++SectNum) { + InMemoryStruct<macho::Section> Sect; + Obj->ReadSection(*SegmentLCI, SectNum, Sect); + if (!Sect) + return Error("unable to load section: '" + Twine(SectNum) + "'"); + + // FIXME: Improve check. + if (Sect->Flags != 0x80000400) + return Error("unsupported section type!"); + + // Address and names of symbols in the section. + typedef std::pair<uint64_t, StringRef> SymbolEntry; + SmallVector<SymbolEntry, 64> Symbols; + // Index of all the names, in this section or not. Used when we're + // dealing with relocation entries. + SmallVector<StringRef, 64> SymbolNames; + for (unsigned i = 0; i != SymtabLC->NumSymbolTableEntries; ++i) { + InMemoryStruct<macho::SymbolTableEntry> STE; + Obj->ReadSymbolTableEntry(SymtabLC->SymbolTableOffset, i, STE); + if (!STE) + return Error("unable to read symbol: '" + Twine(i) + "'"); + if (STE->SectionIndex > SegmentLC->NumSections) + return Error("invalid section index for symbol: '" + Twine(i) + "'"); + // Get the symbol name. + StringRef Name = Obj->getStringAtIndex(STE->StringIndex); + SymbolNames.push_back(Name); + + // Just skip symbols not defined in this section. + if ((unsigned)STE->SectionIndex - 1 != SectNum) + continue; + + // FIXME: Check the symbol type and flags. + if (STE->Type != 0xF) // external, defined in this section. + return Error("unexpected symbol type!"); + // Flags == 0x8 marks a thumb function for ARM, which is fine as it + // doesn't require any special handling here. + if (STE->Flags != 0x0 && STE->Flags != 0x8) + return Error("unexpected symbol type!"); + + // Remember the symbol. + Symbols.push_back(SymbolEntry(STE->Value, Name)); + + DEBUG(dbgs() << "Function sym: '" << Name << "' @ " << + (Sect->Address + STE->Value) << "\n"); + } + // Sort the symbols by address, just in case they didn't come in that way. + array_pod_sort(Symbols.begin(), Symbols.end()); + + // Extract the function data. + uint8_t *Base = (uint8_t*)Obj->getData(SegmentLC->FileOffset, + SegmentLC->FileSize).data(); + for (unsigned i = 0, e = Symbols.size() - 1; i != e; ++i) { + uint64_t StartOffset = Sect->Address + Symbols[i].first; + uint64_t EndOffset = Symbols[i + 1].first - 1; + DEBUG(dbgs() << "Extracting function: " << Symbols[i].second + << " from [" << StartOffset << ", " << EndOffset << "]\n"); + extractFunction(Symbols[i].second, Base + StartOffset, Base + EndOffset); + } + // The last symbol we do after since the end address is calculated + // differently because there is no next symbol to reference. + uint64_t StartOffset = Symbols[Symbols.size() - 1].first; + uint64_t EndOffset = Sect->Size - 1; + DEBUG(dbgs() << "Extracting function: " << Symbols[Symbols.size()-1].second + << " from [" << StartOffset << ", " << EndOffset << "]\n"); + extractFunction(Symbols[Symbols.size()-1].second, + Base + StartOffset, Base + EndOffset); + + // Now extract the relocation information for each function and process it. + for (unsigned j = 0; j != Sect->NumRelocationTableEntries; ++j) { + InMemoryStruct<macho::RelocationEntry> RE; + Obj->ReadRelocationEntry(Sect->RelocationTableOffset, j, RE); + if (RE->Word0 & macho::RF_Scattered) + return Error("NOT YET IMPLEMENTED: scattered relocations."); + // Word0 of the relocation is the offset into the section where the + // relocation should be applied. We need to translate that into an + // offset into a function since that's our atom. + uint32_t Offset = RE->Word0; + // Look for the function containing the address. This is used for JIT + // code, so the number of functions in section is almost always going + // to be very small (usually just one), so until we have use cases + // where that's not true, just use a trivial linear search. + unsigned SymbolNum; + unsigned NumSymbols = Symbols.size(); + assert(NumSymbols > 0 && Symbols[0].first <= Offset && + "No symbol containing relocation!"); + for (SymbolNum = 0; SymbolNum < NumSymbols - 1; ++SymbolNum) + if (Symbols[SymbolNum + 1].first > Offset) + break; + // Adjust the offset to be relative to the symbol. + Offset -= Symbols[SymbolNum].first; + // Get the name of the symbol containing the relocation. + StringRef TargetName = SymbolNames[SymbolNum]; + + bool isExtern = (RE->Word1 >> 27) & 1; + // Figure out the source symbol of the relocation. If isExtern is true, + // this relocation references the symbol table, otherwise it references + // a section in the same object, numbered from 1 through NumSections + // (SectionBases is [0, NumSections-1]). + // FIXME: Some targets (ARM) use internal relocations even for + // externally visible symbols, if the definition is in the same + // file as the reference. We need to convert those back to by-name + // references. We can resolve the address based on the section + // offset and see if we have a symbol at that address. If we do, + // use that; otherwise, puke. + if (!isExtern) + return Error("Internal relocations not supported."); + uint32_t SourceNum = RE->Word1 & 0xffffff; // 24-bit value + StringRef SourceName = SymbolNames[SourceNum]; + + // FIXME: Get the relocation addend from the target address. + + // Now store the relocation information. Associate it with the source + // symbol. + Relocations[SourceName].push_back(RelocationEntry(TargetName, + Offset, + RE->Word1, + 0 /*Addend*/)); + DEBUG(dbgs() << "Relocation at '" << TargetName << "' + " << Offset + << " from '" << SourceName << "(Word1: " + << format("0x%x", RE->Word1) << ")\n"); + } + } + return false; +} + + +bool RuntimeDyldImpl:: +loadSegment64(const MachOObject *Obj, + const MachOObject::LoadCommandInfo *SegmentLCI, + const InMemoryStruct<macho::SymtabLoadCommand> &SymtabLC) { + InMemoryStruct<macho::Segment64LoadCommand> Segment64LC; + Obj->ReadSegment64LoadCommand(*SegmentLCI, Segment64LC); + if (!Segment64LC) + return Error("unable to load segment load command"); + + for (unsigned SectNum = 0; SectNum != Segment64LC->NumSections; ++SectNum) { + InMemoryStruct<macho::Section64> Sect; + Obj->ReadSection64(*SegmentLCI, SectNum, Sect); + if (!Sect) + return Error("unable to load section: '" + Twine(SectNum) + "'"); + + // FIXME: Improve check. + if (Sect->Flags != 0x80000400) + return Error("unsupported section type!"); + + // Address and names of symbols in the section. + typedef std::pair<uint64_t, StringRef> SymbolEntry; + SmallVector<SymbolEntry, 64> Symbols; + // Index of all the names, in this section or not. Used when we're + // dealing with relocation entries. + SmallVector<StringRef, 64> SymbolNames; + for (unsigned i = 0; i != SymtabLC->NumSymbolTableEntries; ++i) { + InMemoryStruct<macho::Symbol64TableEntry> STE; + Obj->ReadSymbol64TableEntry(SymtabLC->SymbolTableOffset, i, STE); + if (!STE) + return Error("unable to read symbol: '" + Twine(i) + "'"); + if (STE->SectionIndex > Segment64LC->NumSections) + return Error("invalid section index for symbol: '" + Twine(i) + "'"); + // Get the symbol name. + StringRef Name = Obj->getStringAtIndex(STE->StringIndex); + SymbolNames.push_back(Name); + + // Just skip symbols not defined in this section. + if ((unsigned)STE->SectionIndex - 1 != SectNum) + continue; + + // FIXME: Check the symbol type and flags. + if (STE->Type != 0xF) // external, defined in this section. + return Error("unexpected symbol type!"); + if (STE->Flags != 0x0) + return Error("unexpected symbol type!"); + + // Remember the symbol. + Symbols.push_back(SymbolEntry(STE->Value, Name)); + + DEBUG(dbgs() << "Function sym: '" << Name << "' @ " << + (Sect->Address + STE->Value) << "\n"); + } + // Sort the symbols by address, just in case they didn't come in that way. + array_pod_sort(Symbols.begin(), Symbols.end()); + + // Extract the function data. + uint8_t *Base = (uint8_t*)Obj->getData(Segment64LC->FileOffset, + Segment64LC->FileSize).data(); + for (unsigned i = 0, e = Symbols.size() - 1; i != e; ++i) { + uint64_t StartOffset = Sect->Address + Symbols[i].first; + uint64_t EndOffset = Symbols[i + 1].first - 1; + DEBUG(dbgs() << "Extracting function: " << Symbols[i].second + << " from [" << StartOffset << ", " << EndOffset << "]\n"); + extractFunction(Symbols[i].second, Base + StartOffset, Base + EndOffset); + } + // The last symbol we do after since the end address is calculated + // differently because there is no next symbol to reference. + uint64_t StartOffset = Symbols[Symbols.size() - 1].first; + uint64_t EndOffset = Sect->Size - 1; + DEBUG(dbgs() << "Extracting function: " << Symbols[Symbols.size()-1].second + << " from [" << StartOffset << ", " << EndOffset << "]\n"); + extractFunction(Symbols[Symbols.size()-1].second, + Base + StartOffset, Base + EndOffset); + + // Now extract the relocation information for each function and process it. + for (unsigned j = 0; j != Sect->NumRelocationTableEntries; ++j) { + InMemoryStruct<macho::RelocationEntry> RE; + Obj->ReadRelocationEntry(Sect->RelocationTableOffset, j, RE); + if (RE->Word0 & macho::RF_Scattered) + return Error("NOT YET IMPLEMENTED: scattered relocations."); + // Word0 of the relocation is the offset into the section where the + // relocation should be applied. We need to translate that into an + // offset into a function since that's our atom. + uint32_t Offset = RE->Word0; + // Look for the function containing the address. This is used for JIT + // code, so the number of functions in section is almost always going + // to be very small (usually just one), so until we have use cases + // where that's not true, just use a trivial linear search. + unsigned SymbolNum; + unsigned NumSymbols = Symbols.size(); + assert(NumSymbols > 0 && Symbols[0].first <= Offset && + "No symbol containing relocation!"); + for (SymbolNum = 0; SymbolNum < NumSymbols - 1; ++SymbolNum) + if (Symbols[SymbolNum + 1].first > Offset) + break; + // Adjust the offset to be relative to the symbol. + Offset -= Symbols[SymbolNum].first; + // Get the name of the symbol containing the relocation. + StringRef TargetName = SymbolNames[SymbolNum]; + + bool isExtern = (RE->Word1 >> 27) & 1; + // Figure out the source symbol of the relocation. If isExtern is true, + // this relocation references the symbol table, otherwise it references + // a section in the same object, numbered from 1 through NumSections + // (SectionBases is [0, NumSections-1]). + if (!isExtern) + return Error("Internal relocations not supported."); + uint32_t SourceNum = RE->Word1 & 0xffffff; // 24-bit value + StringRef SourceName = SymbolNames[SourceNum]; + + // FIXME: Get the relocation addend from the target address. + + // Now store the relocation information. Associate it with the source + // symbol. + Relocations[SourceName].push_back(RelocationEntry(TargetName, + Offset, + RE->Word1, + 0 /*Addend*/)); + DEBUG(dbgs() << "Relocation at '" << TargetName << "' + " << Offset + << " from '" << SourceName << "(Word1: " + << format("0x%x", RE->Word1) << ")\n"); + } + } + return false; +} + +bool RuntimeDyldImpl::loadObject(MemoryBuffer *InputBuffer) { + // If the linker is in an error state, don't do anything. + if (hasError()) + return true; + // Load the Mach-O wrapper object. + std::string ErrorStr; + OwningPtr<MachOObject> Obj( + MachOObject::LoadFromBuffer(InputBuffer, &ErrorStr)); + if (!Obj) + return Error("unable to load object: '" + ErrorStr + "'"); + + // Get the CPU type information from the header. + const macho::Header &Header = Obj->getHeader(); + + // FIXME: Error checking that the loaded object is compatible with + // the system we're running on. + CPUType = Header.CPUType; + CPUSubtype = Header.CPUSubtype; + + // Validate that the load commands match what we expect. + const MachOObject::LoadCommandInfo *SegmentLCI = 0, *SymtabLCI = 0, + *DysymtabLCI = 0; + for (unsigned i = 0; i != Header.NumLoadCommands; ++i) { + const MachOObject::LoadCommandInfo &LCI = Obj->getLoadCommandInfo(i); + switch (LCI.Command.Type) { + case macho::LCT_Segment: + case macho::LCT_Segment64: + if (SegmentLCI) + return Error("unexpected input object (multiple segments)"); + SegmentLCI = &LCI; + break; + case macho::LCT_Symtab: + if (SymtabLCI) + return Error("unexpected input object (multiple symbol tables)"); + SymtabLCI = &LCI; + break; + case macho::LCT_Dysymtab: + if (DysymtabLCI) + return Error("unexpected input object (multiple symbol tables)"); + DysymtabLCI = &LCI; + break; + default: + return Error("unexpected input object (unexpected load command"); + } + } + + if (!SymtabLCI) + return Error("no symbol table found in object"); + if (!SegmentLCI) + return Error("no symbol table found in object"); + + // Read and register the symbol table data. + InMemoryStruct<macho::SymtabLoadCommand> SymtabLC; + Obj->ReadSymtabLoadCommand(*SymtabLCI, SymtabLC); + if (!SymtabLC) + return Error("unable to load symbol table load command"); + Obj->RegisterStringTable(*SymtabLC); + + // Read the dynamic link-edit information, if present (not present in static + // objects). + if (DysymtabLCI) { + InMemoryStruct<macho::DysymtabLoadCommand> DysymtabLC; + Obj->ReadDysymtabLoadCommand(*DysymtabLCI, DysymtabLC); + if (!DysymtabLC) + return Error("unable to load dynamic link-exit load command"); + + // FIXME: We don't support anything interesting yet. +// if (DysymtabLC->LocalSymbolsIndex != 0) +// return Error("NOT YET IMPLEMENTED: local symbol entries"); +// if (DysymtabLC->ExternalSymbolsIndex != 0) +// return Error("NOT YET IMPLEMENTED: non-external symbol entries"); +// if (DysymtabLC->UndefinedSymbolsIndex != SymtabLC->NumSymbolTableEntries) +// return Error("NOT YET IMPLEMENTED: undefined symbol entries"); + } + + // Load the segment load command. + if (SegmentLCI->Command.Type == macho::LCT_Segment) { + if (loadSegment32(Obj.get(), SegmentLCI, SymtabLC)) + return true; + } else { + if (loadSegment64(Obj.get(), SegmentLCI, SymtabLC)) + return true; + } + + return false; +} + +// Resolve the relocations for all symbols we currently know about. +void RuntimeDyldImpl::resolveRelocations() { + // Just iterate over the symbols in our symbol table and assign their + // addresses. + StringMap<uint8_t*>::iterator i = SymbolTable.begin(); + StringMap<uint8_t*>::iterator e = SymbolTable.end(); + for (;i != e; ++i) + reassignSymbolAddress(i->getKey(), i->getValue()); +} + +// Assign an address to a symbol name and resolve all the relocations +// associated with it. +void RuntimeDyldImpl::reassignSymbolAddress(StringRef Name, uint8_t *Addr) { + // Assign the address in our symbol table. + SymbolTable[Name] = Addr; + + RelocationList &Relocs = Relocations[Name]; + for (unsigned i = 0, e = Relocs.size(); i != e; ++i) { + RelocationEntry &RE = Relocs[i]; + uint8_t *Target = SymbolTable[RE.Target] + RE.Offset; + bool isPCRel = (RE.Data >> 24) & 1; + unsigned Type = (RE.Data >> 28) & 0xf; + unsigned Size = 1 << ((RE.Data >> 25) & 3); + + DEBUG(dbgs() << "Resolving relocation at '" << RE.Target + << "' + " << RE.Offset << " (" << format("%p", Target) << ")" + << " from '" << Name << " (" << format("%p", Addr) << ")" + << "(" << (isPCRel ? "pcrel" : "absolute") + << ", type: " << Type << ", Size: " << Size << ").\n"); + + resolveRelocation(Target, Addr, isPCRel, Type, Size); + RE.isResolved = true; + } +} + +//===----------------------------------------------------------------------===// +// RuntimeDyld class implementation +RuntimeDyld::RuntimeDyld(RTDyldMemoryManager *MM) { + Dyld = new RuntimeDyldImpl(MM); +} + +RuntimeDyld::~RuntimeDyld() { + delete Dyld; +} + +bool RuntimeDyld::loadObject(MemoryBuffer *InputBuffer) { + return Dyld->loadObject(InputBuffer); +} + +void *RuntimeDyld::getSymbolAddress(StringRef Name) { + return Dyld->getSymbolAddress(Name); +} + +void RuntimeDyld::resolveRelocations() { + Dyld->resolveRelocations(); +} + +void RuntimeDyld::reassignSymbolAddress(StringRef Name, uint8_t *Addr) { + Dyld->reassignSymbolAddress(Name, Addr); +} + +StringRef RuntimeDyld::getErrorString() { + return Dyld->getErrorString(); +} + +} // end namespace llvm |
