X-Git-Url: http://plrg.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FVMCore%2FVerifier.cpp;h=da65b1ce30ae1c19aebec39bac73e717e957ac9c;hb=24ea74eb9a47b81c1557926acd83e0fbe6d7594e;hp=f99b5ab5aa31811120241021d897f5fd478fde56;hpb=25d15a7ff75f04a26c9643e849807895ab67babd;p=oota-llvm.git diff --git a/lib/VMCore/Verifier.cpp b/lib/VMCore/Verifier.cpp index f99b5ab5aa3..da65b1ce30a 100644 --- a/lib/VMCore/Verifier.cpp +++ b/lib/VMCore/Verifier.cpp @@ -6,206 +6,435 @@ // Note that this does not provide full 'java style' security and verifications, // instead it just tries to ensure that code is well formed. // -// . There are no duplicated names in a symbol table... ie there !exist a val -// with the same name as something in the symbol table, but with a different -// address as what is in the symbol table... -// . Both of a binary operator's parameters are the same type +// * Both of a binary operator's parameters are the same type +// * Verify that the indices of mem access instructions match other operands // . Verify that arithmetic and other things are only performed on first class -// types. No adding structures or arrays. +// types. Verify that shifts & logicals only happen on integrals f.e. // . All of the constants in a switch statement are of the correct type -// . The code is in valid SSA form +// * The code is in valid SSA form // . It should be illegal to put a label into any other type (like a structure) // or to return one. [except constant arrays!] -// . Right now 'add bool 0, 0' is valid. This isn't particularly good. // * Only phi nodes can be self referential: 'add int %0, %0 ; :0' is bad // * PHI nodes must have an entry for each predecessor, with no extras. +// * PHI nodes must be the first thing in a basic block, all grouped together // * All basic blocks should only end with terminator insts, not contain them // * The entry node to a function must not have predecessors // * All Instructions must be embeded into a basic block -// . Verify that none of the Value getType()'s are null. // . Function's cannot take a void typed parameter -// . Verify that a function's argument list agrees with it's declared type. +// * Verify that a function's argument list agrees with it's declared type. // . Verify that arrays and structures have fixed elements: No unsized arrays. // * It is illegal to specify a name for a void value. // * It is illegal to have a internal function that is just a declaration -// . All other things that are tested by asserts spread about the code... +// * It is illegal to have a ret instruction that returns a value that does not +// agree with the function return value type. +// * Function call argument types match the function prototype +// * All other things that are tested by asserts spread about the code... // //===----------------------------------------------------------------------===// #include "llvm/Analysis/Verifier.h" #include "llvm/Pass.h" -#include "llvm/Function.h" #include "llvm/Module.h" -#include "llvm/BasicBlock.h" -#include "llvm/Type.h" +#include "llvm/DerivedTypes.h" #include "llvm/iPHINode.h" +#include "llvm/iTerminators.h" +#include "llvm/iOther.h" +#include "llvm/iMemory.h" #include "llvm/SymbolTable.h" +#include "llvm/PassManager.h" +#include "llvm/Analysis/Dominators.h" #include "llvm/Support/CFG.h" +#include "llvm/Support/InstVisitor.h" #include "Support/STLExtras.h" #include +#include -#if 0 -#define t(x) (1 << (unsigned)Type::x) -#define SignedIntegralTypes (t(SByteTyID) | t(ShortTyID) | \ - t(IntTyID) | t(LongTyID)) -static long UnsignedIntegralTypes = t(UByteTyID) | t(UShortTyID) | - t(UIntTyID) | t(ULongTyID); -static const long FloatingPointTypes = t(FloatTyID) | t(DoubleTyID); - -static const long IntegralTypes = SignedIntegralTypes | UnsignedIntegralTypes; - -static long ValidTypes[Type::FirstDerivedTyID] = { - [(unsigned)Instruction::UnaryOps::Not] t(BoolTyID), - //[Instruction::UnaryOps::Add] = IntegralTypes, - // [Instruction::Sub] = IntegralTypes, -}; -#undef t -#endif - -// CheckFailed - A check failed, so print out the condition and the message that -// failed. This provides a nice place to put a breakpoint if you want to see -// why something is not correct. -// -static inline void CheckFailed(const char *Cond, const std::string &Message, - const Value *V1 = 0, const Value *V2 = 0) { - std::cerr << Message << "\n"; - if (V1) { V1->dump(); std::cerr << "\n"; } - if (V2) { V2->dump(); std::cerr << "\n"; } +namespace { // Anonymous namespace for class + + struct Verifier : public FunctionPass, InstVisitor { + bool Broken; // Is this module found to be broken? + bool RealPass; // Are we not being run by a PassManager? + bool AbortBroken; // If broken, should it or should it not abort? + + DominatorSet *DS; // Dominator set, caution can be null! + + Verifier() : Broken(false), RealPass(true), AbortBroken(true), DS(0) {} + Verifier(bool AB) : Broken(false), RealPass(true), AbortBroken(AB), DS(0) {} + Verifier(DominatorSet &ds) + : Broken(false), RealPass(false), AbortBroken(false), DS(&ds) {} + + + bool doInitialization(Module &M) { + verifySymbolTable(M.getSymbolTable()); + return false; + } + + bool runOnFunction(Function &F) { + // Get dominator information if we are being run by PassManager + if (RealPass) DS = &getAnalysis(); + visit(F); + return false; + } + + bool doFinalization(Module &M) { + // Scan through, checking all of the external function's linkage now... + for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I) + if (I->isExternal() && I->hasInternalLinkage()) + CheckFailed("Function Declaration has Internal Linkage!", I); + + if (Broken && AbortBroken) { + std::cerr << "Broken module found, compilation aborted!\n"; + abort(); + } + return false; + } + + virtual void getAnalysisUsage(AnalysisUsage &AU) const { + AU.setPreservesAll(); + if (RealPass) + AU.addRequired(); + } + + // Verification methods... + void verifySymbolTable(SymbolTable *ST); + void visitFunction(Function &F); + void visitBasicBlock(BasicBlock &BB); + void visitPHINode(PHINode &PN); + void visitBinaryOperator(BinaryOperator &B); + void visitCallInst(CallInst &CI); + void visitGetElementPtrInst(GetElementPtrInst &GEP); + void visitLoadInst(LoadInst &LI); + void visitStoreInst(StoreInst &SI); + void visitInstruction(Instruction &I); + void visitTerminatorInst(TerminatorInst &I); + void visitReturnInst(ReturnInst &RI); + + // CheckFailed - A check failed, so print out the condition and the message + // that failed. This provides a nice place to put a breakpoint if you want + // to see why something is not correct. + // + inline void CheckFailed(const std::string &Message, + const Value *V1 = 0, const Value *V2 = 0, + const Value *V3 = 0, const Value *V4 = 0) { + std::cerr << Message << "\n"; + if (V1) std::cerr << *V1 << "\n"; + if (V2) std::cerr << *V2 << "\n"; + if (V3) std::cerr << *V3 << "\n"; + if (V4) std::cerr << *V4 << "\n"; + Broken = true; + } + }; + + RegisterPass X("verify", "Module Verifier"); } // Assert - We know that cond should be true, if not print an error message. #define Assert(C, M) \ - do { if (!(C)) { CheckFailed(#C, M); Broken = true; } } while (0) + do { if (!(C)) { CheckFailed(M); return; } } while (0) #define Assert1(C, M, V1) \ - do { if (!(C)) { CheckFailed(#C, M, V1); Broken = true; } } while (0) + do { if (!(C)) { CheckFailed(M, V1); return; } } while (0) #define Assert2(C, M, V1, V2) \ - do { if (!(C)) { CheckFailed(#C, M, V1, V2); Broken = true; } } while (0) + do { if (!(C)) { CheckFailed(M, V1, V2); return; } } while (0) +#define Assert3(C, M, V1, V2, V3) \ + do { if (!(C)) { CheckFailed(M, V1, V2, V3); return; } } while (0) +#define Assert4(C, M, V1, V2, V3, V4) \ + do { if (!(C)) { CheckFailed(M, V1, V2, V3, V4); return; } } while (0) -// verifyInstruction - Verify that a non-terminator instruction is well formed. +// verifySymbolTable - Verify that a function or module symbol table is ok // -static bool verifyInstruction(const Instruction *I) { - bool Broken = false; - assert(I->getParent() && "Instruction not embedded in basic block!"); - Assert1(!isa(I), - "Terminator instruction found embedded in basic block!\n", I); +void Verifier::verifySymbolTable(SymbolTable *ST) { + if (ST == 0) return; // No symbol table to process - // Check that all uses of the instruction, if they are instructions - // themselves, actually have parent basic blocks. - // - for (User::use_const_iterator UI = I->use_begin(), UE = I->use_end(); - UI != UE; ++UI) { - if (Instruction *Used = dyn_cast(*UI)) - Assert2(Used->getParent() != 0, "Instruction referencing instruction not" - " embeded in a basic block!", I, Used); - } + // Loop over all of the types in the symbol table... + for (SymbolTable::iterator TI = ST->begin(), TE = ST->end(); TI != TE; ++TI) + for (SymbolTable::type_iterator I = TI->second.begin(), + E = TI->second.end(); I != E; ++I) { + Value *V = I->second; - // Check that PHI nodes look ok - if (const PHINode *PN = dyn_cast(I)) { - std::vector Preds(pred_begin(I->getParent()), - pred_end(I->getParent())); - // Loop over all of the incoming values, make sure that there are - // predecessors for each one... - // - for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) { - const BasicBlock *BB = PN->getIncomingBlock(i); - std::vector::iterator PI = - find(Preds.begin(), Preds.end(), BB); - Assert2(PI != Preds.end(), "PHI node has entry for basic block that" - " is not a predecessor!", PN, BB); - if (PI != Preds.end()) Preds.erase(PI); + // Check that there are no void typed values in the symbol table. Values + // with a void type cannot be put into symbol tables because they cannot + // have names! + Assert1(V->getType() != Type::VoidTy, + "Values with void type are not allowed to have names!", V); } +} - // There should be no entries left in the predecessor list... - for (std::vector::iterator I = Preds.begin(), - E = Preds.end(); I != E; ++I) - Assert2(0, "PHI node does not have entry for a predecessor basic block!", - PN, *I); - } else { - // Check that non-phi nodes are not self referential... - for (Value::use_const_iterator UI = I->use_begin(), UE = I->use_end(); - UI != UE; ++UI) - Assert1(*UI != (const User*)I, - "Only PHI nodes may reference their own value!", I); + +// visitFunction - Verify that a function is ok. +// +void Verifier::visitFunction(Function &F) { + if (F.isExternal()) return; + + verifySymbolTable(F.getSymbolTable()); + + // Check function arguments... + const FunctionType *FT = F.getFunctionType(); + unsigned NumArgs = F.getArgumentList().size(); + + Assert2(!FT->isVarArg(), "Cannot define varargs functions in LLVM!", &F, FT); + Assert2(FT->getParamTypes().size() == NumArgs, + "# formal arguments must match # of arguments for function type!", + &F, FT); + + // Check that the argument values match the function type for this function... + if (FT->getParamTypes().size() == NumArgs) { + unsigned i = 0; + for (Function::aiterator I = F.abegin(), E = F.aend(); I != E; ++I, ++i) + Assert2(I->getType() == FT->getParamType(i), + "Argument value does not match function argument type!", + I, FT->getParamType(i)); } - return Broken; + // Check the entry node + BasicBlock *Entry = &F.getEntryNode(); + Assert1(pred_begin(Entry) == pred_end(Entry), + "Entry block to function must not have predecessors!", Entry); } // verifyBasicBlock - Verify that a basic block is well formed... // -static bool verifyBasicBlock(const BasicBlock *BB) { - bool Broken = false; - Assert1(BB->getTerminator(), "Basic Block does not have terminator!\n", BB); +void Verifier::visitBasicBlock(BasicBlock &BB) { + // Ensure that basic blocks have terminators! + Assert1(BB.getTerminator(), "Basic Block does not have terminator!", &BB); +} - // Verify all instructions, except the terminator... - Broken |= reduce_apply_bool(BB->begin(), BB->end()-1, verifyInstruction); - return Broken; +void Verifier::visitTerminatorInst(TerminatorInst &I) { + // Ensure that terminators only exist at the end of the basic block. + Assert1(&I == I.getParent()->getTerminator(), + "Terminator found in the middle of a basic block!", I.getParent()); + visitInstruction(I); } -// verifySymbolTable - Verify that a method or module symbol table is ok -// -static bool verifySymbolTable(const SymbolTable *ST) { - if (ST == 0) return false; - bool Broken = false; +void Verifier::visitReturnInst(ReturnInst &RI) { + Function *F = RI.getParent()->getParent(); + if (RI.getNumOperands() == 0) + Assert1(F->getReturnType() == Type::VoidTy, + "Function returns no value, but ret instruction found that does!", + &RI); + else + Assert2(F->getReturnType() == RI.getOperand(0)->getType(), + "Function return type does not match operand " + "type of return inst!", &RI, F->getReturnType()); - // Loop over all of the types in the symbol table... - for (SymbolTable::const_iterator TI = ST->begin(), TE = ST->end(); - TI != TE; ++TI) - for (SymbolTable::type_const_iterator I = TI->second.begin(), - E = TI->second.end(); I != E; ++I) { - Value *V = I->second; + // Check to make sure that the return value has neccesary properties for + // terminators... + visitTerminatorInst(RI); +} - // Check that there are no void typed values in the symbol table. Values - // with a void type cannot be put into symbol tables because they cannot - // have names! - Assert1(V->getType() != Type::VoidTy, - "Values with void type are not allowed to have names!\n", V); - } - return Broken; +// visitPHINode - Ensure that a PHI node is well formed. +void Verifier::visitPHINode(PHINode &PN) { + // Ensure that the PHI nodes are all grouped together at the top of the block. + // This can be tested by checking whether the instruction before this is + // either nonexistant (because this is begin()) or is a PHI node. If not, + // then there is some other instruction before a PHI. + Assert2(PN.getPrev() == 0 || isa(PN.getPrev()), + "PHI nodes not grouped at top of basic block!", + &PN, PN.getParent()); + + std::vector Preds(pred_begin(PN.getParent()), + pred_end(PN.getParent())); + // Loop over all of the incoming values, make sure that there are + // predecessors for each one... + // + for (unsigned i = 0, e = PN.getNumIncomingValues(); i != e; ++i) { + // Make sure all of the incoming values are the right types... + Assert2(PN.getType() == PN.getIncomingValue(i)->getType(), + "PHI node argument type does not agree with PHI node type!", + &PN, PN.getIncomingValue(i)); + + BasicBlock *BB = PN.getIncomingBlock(i); + std::vector::iterator PI = + find(Preds.begin(), Preds.end(), BB); + Assert2(PI != Preds.end(), "PHI node has entry for basic block that" + " is not a predecessor!", &PN, BB); + Preds.erase(PI); + } + + // There should be no entries left in the predecessor list... + for (std::vector::iterator I = Preds.begin(), + E = Preds.end(); I != E; ++I) + Assert2(0, "PHI node does not have entry for a predecessor basic block!", + &PN, *I); + + // Now we go through and check to make sure that if there is more than one + // entry for a particular basic block in this PHI node, that the incoming + // values are all identical. + // + std::vector > Values; + Values.reserve(PN.getNumIncomingValues()); + for (unsigned i = 0, e = PN.getNumIncomingValues(); i != e; ++i) + Values.push_back(std::make_pair(PN.getIncomingBlock(i), + PN.getIncomingValue(i))); + + // Sort the Values vector so that identical basic block entries are adjacent. + std::sort(Values.begin(), Values.end()); + + // Check for identical basic blocks with differing incoming values... + for (unsigned i = 1, e = PN.getNumIncomingValues(); i < e; ++i) + Assert4(Values[i].first != Values[i-1].first || + Values[i].second == Values[i-1].second, + "PHI node has multiple entries for the same basic block with " + "different incoming values!", &PN, Values[i].first, + Values[i].second, Values[i-1].second); + + visitInstruction(PN); +} + +void Verifier::visitCallInst(CallInst &CI) { + Assert1(isa(CI.getOperand(0)->getType()), + "Called function must be a pointer!", &CI); + const PointerType *FPTy = cast(CI.getOperand(0)->getType()); + Assert1(isa(FPTy->getElementType()), + "Called function is not pointer to function type!", &CI); + + const FunctionType *FTy = cast(FPTy->getElementType()); + + // Verify that the correct number of arguments are being passed + if (FTy->isVarArg()) + Assert1(CI.getNumOperands()-1 >= FTy->getNumParams(), + "Called function requires more parameters than were provided!",&CI); + else + Assert1(CI.getNumOperands()-1 == FTy->getNumParams(), + "Incorrect number of arguments passed to called function!", &CI); + + // Verify that all arguments to the call match the function type... + for (unsigned i = 0, e = FTy->getNumParams(); i != e; ++i) + Assert2(CI.getOperand(i+1)->getType() == FTy->getParamType(i), + "Call parameter type does not match function signature!", + CI.getOperand(i+1), FTy->getParamType(i)); + + visitInstruction(CI); } -// verifyMethod - Verify that a method is ok. Return true if not so that -// verifyModule and direct clients of the verifyMethod function are correctly -// informed. +// visitBinaryOperator - Check that both arguments to the binary operator are +// of the same type! // -bool verifyMethod(const Function *F) { - if (F->isExternal()) return false; // Can happen if called by verifyModule - bool Broken = verifySymbolTable(F->getSymbolTable()); +void Verifier::visitBinaryOperator(BinaryOperator &B) { + Assert2(B.getOperand(0)->getType() == B.getOperand(1)->getType(), + "Both operands to a binary operator are not of the same type!", + B.getOperand(0), B.getOperand(1)); - Assert1(!F->isExternal() || F->hasExternalLinkage(), - "Function cannot be an 'internal' 'declare'ation!", F); + visitInstruction(B); +} - const BasicBlock *Entry = F->getEntryNode(); - Assert1(pred_begin(Entry) == pred_end(Entry), - "Entry block to method must not have predecessors!", Entry); +void Verifier::visitGetElementPtrInst(GetElementPtrInst &GEP) { + const Type *ElTy = MemAccessInst::getIndexedType(GEP.getOperand(0)->getType(), + GEP.copyIndices(), true); + Assert1(ElTy, "Invalid indices for GEP pointer type!", &GEP); + Assert2(PointerType::get(ElTy) == GEP.getType(), + "GEP is not of right type for indices!", &GEP, ElTy); + visitInstruction(GEP); +} - Broken |= reduce_apply_bool(F->begin(), F->end(), verifyBasicBlock); - return Broken; +void Verifier::visitLoadInst(LoadInst &LI) { + const Type *ElTy = + cast(LI.getOperand(0)->getType())->getElementType(); + Assert2(ElTy == LI.getType(), + "Load is not of right type for indices!", &LI, ElTy); + visitInstruction(LI); } +void Verifier::visitStoreInst(StoreInst &SI) { + const Type *ElTy = + cast(SI.getOperand(1)->getType())->getElementType(); + Assert2(ElTy == SI.getOperand(0)->getType(), + "Stored value is not of right type for indices!", &SI, ElTy); + visitInstruction(SI); +} -namespace { // Anonymous namespace for class - struct VerifierPass : public MethodPass { - bool doInitialization(Module *M) { - verifySymbolTable(M->getSymbolTable()); - return false; +// verifyInstruction - Verify that an instruction is well formed. +// +void Verifier::visitInstruction(Instruction &I) { + Assert1(I.getParent(), "Instruction not embedded in basic block!", &I); + + // Check that all uses of the instruction, if they are instructions + // themselves, actually have parent basic blocks. If the use is not an + // instruction, it is an error! + // + for (User::use_iterator UI = I.use_begin(), UE = I.use_end(); + UI != UE; ++UI) { + Assert1(isa(*UI), "Use of instruction is not an instruction!", + *UI); + Instruction *Used = cast(*UI); + Assert2(Used->getParent() != 0, "Instruction referencing instruction not" + " embeded in a basic block!", &I, Used); + } + + if (!isa(I)) { // Check that non-phi nodes are not self referential + for (Value::use_iterator UI = I.use_begin(), UE = I.use_end(); + UI != UE; ++UI) + Assert1(*UI != (User*)&I, + "Only PHI nodes may reference their own value!", &I); + } + + // Check that void typed values don't have names + Assert1(I.getType() != Type::VoidTy || !I.hasName(), + "Instruction has a name, but provides a void value!", &I); + + // Check that a definition dominates all of its uses. + // + for (User::use_iterator UI = I.use_begin(), UE = I.use_end(); + UI != UE; ++UI) { + Instruction *Use = cast(*UI); + + // PHI nodes are more difficult than other nodes because they actually + // "use" the value in the predecessor basic blocks they correspond to. + if (PHINode *PN = dyn_cast(Use)) { + for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) + if (&I == PN->getIncomingValue(i)) { + // Make sure that I dominates the end of pred(i) + BasicBlock *Pred = PN->getIncomingBlock(i); + + Assert2(DS->dominates(I.getParent(), Pred), + "Instruction does not dominate all uses!", + &I, PN); + } + + } else { + Assert2(DS->dominates(&I, Use), + "Instruction does not dominate all uses!", &I, Use); } - bool runOnMethod(Method *M) { verifyMethod(M); return false; } - }; + } } + +//===----------------------------------------------------------------------===// +// Implement the public interfaces to this file... +//===----------------------------------------------------------------------===// + Pass *createVerifierPass() { - return new VerifierPass(); + return new Verifier(); +} + + +// verifyFunction - Create +bool verifyFunction(const Function &f) { + Function &F = (Function&)f; + assert(!F.isExternal() && "Cannot verify external functions"); + + DominatorSet DS; + DS.doInitialization(*F.getParent()); + DS.runOnFunction(F); + + Verifier V(DS); + V.runOnFunction(F); + + DS.doFinalization(*F.getParent()); + + return V.Broken; } // verifyModule - Check a module for errors, printing messages on stderr. // Return true if the module is corrupt. // -bool verifyModule(const Module *M) { - return verifySymbolTable(M->getSymbolTable()) | - reduce_apply_bool(M->begin(), M->end(), verifyMethod); +bool verifyModule(const Module &M) { + PassManager PM; + Verifier *V = new Verifier(); + PM.add(V); + PM.run((Module&)M); + return V->Broken; }