// 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
// * 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.
+// * Verify that arithmetic and other things are only performed on first class
+// 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!]
// * Only phi nodes can be self referential: 'add int %0, %0 ; <int>: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
+// * PHI nodes must have at least one entry
// * 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 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
+// * It is illegal to have a internal global value with no intitalizer
// * 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
#include "llvm/iPHINode.h"
#include "llvm/iTerminators.h"
#include "llvm/iOther.h"
+#include "llvm/iOperators.h"
#include "llvm/iMemory.h"
#include "llvm/SymbolTable.h"
+#include "llvm/PassManager.h"
+#include "llvm/Intrinsics.h"
+#include "llvm/Analysis/Dominators.h"
#include "llvm/Support/CFG.h"
#include "llvm/Support/InstVisitor.h"
#include "Support/STLExtras.h"
#include <algorithm>
-#include <iostream>
namespace { // Anonymous namespace for class
struct Verifier : public FunctionPass, InstVisitor<Verifier> {
- bool Broken;
+ 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) {}
+ 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) {}
- virtual const char *getPassName() const { return "Module Verifier"; }
bool doInitialization(Module &M) {
verifySymbolTable(M.getSymbolTable());
+
+ // If this is a real pass, in a pass manager, we must abort before
+ // returning back to the pass manager, or else the pass manager may try to
+ // run other passes on the broken module.
+ //
+ if (RealPass)
+ abortIfBroken();
return false;
}
bool runOnFunction(Function &F) {
+ // Get dominator information if we are being run by PassManager
+ if (RealPass) DS = &getAnalysis<DominatorSet>();
visit(F);
+
+ // If this is a real pass, in a pass manager, we must abort before
+ // returning back to the pass manager, or else the pass manager may try to
+ // run other passes on the broken module.
+ //
+ if (RealPass)
+ abortIfBroken();
+
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)
+ visitGlobalValue(*I);
+
+ for (Module::giterator I = M.gbegin(), E = M.gend(); I != E; ++I)
if (I->isExternal() && I->hasInternalLinkage())
- CheckFailed("Function Declaration has Internal Linkage!", I);
+ CheckFailed("Global Variable is external with internal linkage!", I);
- if (Broken) {
- std::cerr << "Broken module found, compilation aborted!\n";
- abort();
- }
+ // If the module is broken, abort at this time.
+ abortIfBroken();
return false;
}
virtual void getAnalysisUsage(AnalysisUsage &AU) const {
AU.setPreservesAll();
+ if (RealPass)
+ AU.addRequired<DominatorSet>();
+ }
+
+ // abortIfBroken - If the module is broken and we are supposed to abort on
+ // this condition, do so.
+ //
+ void abortIfBroken() const {
+ if (Broken && AbortBroken) {
+ std::cerr << "Broken module found, compilation aborted!\n";
+ abort();
+ }
}
+
// Verification methods...
- void verifySymbolTable(SymbolTable *ST);
+ void verifySymbolTable(SymbolTable &ST);
+ void visitGlobalValue(GlobalValue &GV);
void visitFunction(Function &F);
void visitBasicBlock(BasicBlock &BB);
void visitPHINode(PHINode &PN);
void visitBinaryOperator(BinaryOperator &B);
+ void visitShiftInst(ShiftInst &SI);
+ void visitVarArgInst(VarArgInst &VAI) { visitInstruction(VAI); }
void visitCallInst(CallInst &CI);
void visitGetElementPtrInst(GetElementPtrInst &GEP);
void visitLoadInst(LoadInst &LI);
void visitInstruction(Instruction &I);
void visitTerminatorInst(TerminatorInst &I);
void visitReturnInst(ReturnInst &RI);
+ void visitUserOp1(Instruction &I);
+ void visitUserOp2(Instruction &I) { visitUserOp1(I); }
+ void visitIntrinsicFunctionCall(LLVMIntrinsic::ID ID, CallInst &CI);
// 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
Broken = true;
}
};
+
+ RegisterPass<Verifier> X("verify", "Module Verifier");
}
// Assert - We know that cond should be true, if not print an error message.
do { if (!(C)) { CheckFailed(M, V1, V2, V3, V4); return; } } while (0)
+void Verifier::visitGlobalValue(GlobalValue &GV) {
+ Assert1(!GV.isExternal() || GV.hasExternalLinkage(),
+ "Global value has Internal Linkage!", &GV);
+ Assert1(!GV.hasAppendingLinkage() || isa<GlobalVariable>(GV),
+ "Only global variables can have appending linkage!", &GV);
+
+ if (GV.hasAppendingLinkage()) {
+ GlobalVariable &GVar = cast<GlobalVariable>(GV);
+ Assert1(isa<ArrayType>(GVar.getType()->getElementType()),
+ "Only global arrays can have appending linkage!", &GV);
+ }
+}
+
// verifySymbolTable - Verify that a function or module symbol table is ok
//
-void Verifier::verifySymbolTable(SymbolTable *ST) {
- if (ST == 0) return; // No symbol table to process
-
+void Verifier::verifySymbolTable(SymbolTable &ST) {
// Loop over all of the types in the symbol table...
- for (SymbolTable::iterator TI = ST->begin(), TE = ST->end(); TI != TE; ++TI)
+ 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;
// 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,
+ Assert2(FT->getNumParams() == 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));
+ 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));
+
+ if (!F.isExternal()) {
+ verifySymbolTable(F.getSymbolTable());
+
+ // Check the entry node
+ BasicBlock *Entry = &F.getEntryNode();
+ Assert1(pred_begin(Entry) == pred_end(Entry),
+ "Entry block to function must not have predecessors!", Entry);
}
-
- // Check the entry node
- BasicBlock *Entry = &F.getEntryNode();
- Assert1(pred_begin(Entry) == pred_end(Entry),
- "Entry block to function must not have predecessors!", Entry);
}
// 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);
}
void Verifier::visitReturnInst(ReturnInst &RI) {
"Function return type does not match operand "
"type of return inst!", &RI, F->getReturnType());
- // Check to make sure that the return value has neccesary properties for
+ // Check to make sure that the return value has necessary properties for
// terminators...
visitTerminatorInst(RI);
}
+// visitUserOp1 - User defined operators shouldn't live beyond the lifetime of a
+// pass, if any exist, it's an error.
+//
+void Verifier::visitUserOp1(Instruction &I) {
+ Assert1(0, "User-defined operators should not live outside of a pass!",
+ &I);
+}
// visitPHINode - Ensure that a PHI node is well formed.
void Verifier::visitPHINode(PHINode &PN) {
"PHI nodes not grouped at top of basic block!",
&PN, PN.getParent());
+ // Ensure that PHI nodes have at least one entry!
+ Assert1(PN.getNumIncomingValues() != 0,
+ "PHI nodes must have at least one entry. If the block is dead, "
+ "the PHI should be removed!",
+ &PN);
+
std::vector<BasicBlock*> Preds(pred_begin(PN.getParent()),
pred_end(PN.getParent()));
// Loop over all of the incoming values, make sure that there are
Assert2(CI.getOperand(i+1)->getType() == FTy->getParamType(i),
"Call parameter type does not match function signature!",
CI.getOperand(i+1), FTy->getParamType(i));
+
+ if (Function *F = CI.getCalledFunction())
+ if (LLVMIntrinsic::ID ID = (LLVMIntrinsic::ID)F->getIntrinsicID())
+ visitIntrinsicFunctionCall(ID, CI);
+
+ visitInstruction(CI);
}
// visitBinaryOperator - Check that both arguments to the binary operator are
// of the same type!
//
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(B.getOperand(0)->getType() == B.getOperand(1)->getType(),
+ "Both operands to a binary operator are not of the same type!", &B);
+
+ // Check that logical operators are only used with integral operands.
+ if (B.getOpcode() == Instruction::And || B.getOpcode() == Instruction::Or ||
+ B.getOpcode() == Instruction::Xor) {
+ Assert1(B.getType()->isIntegral(),
+ "Logical operators only work with integral types!", &B);
+ Assert1(B.getType() == B.getOperand(0)->getType(),
+ "Logical operators must have same type for operands and result!",
+ &B);
+ } else if (isa<SetCondInst>(B)) {
+ // Check that setcc instructions return bool
+ Assert1(B.getType() == Type::BoolTy,
+ "setcc instructions must return boolean values!", &B);
+ } else {
+ // Arithmetic operators only work on integer or fp values
+ Assert1(B.getType() == B.getOperand(0)->getType(),
+ "Arithmetic operators must have same type for operands and result!",
+ &B);
+ Assert1(B.getType()->isInteger() || B.getType()->isFloatingPoint(),
+ "Arithmetic operators must have integer or fp type!", &B);
+ }
+
visitInstruction(B);
}
+void Verifier::visitShiftInst(ShiftInst &SI) {
+ Assert1(SI.getType()->isInteger(),
+ "Shift must return an integer result!", &SI);
+ Assert1(SI.getType() == SI.getOperand(0)->getType(),
+ "Shift return type must be same as first operand!", &SI);
+ Assert1(SI.getOperand(1)->getType() == Type::UByteTy,
+ "Second operand to shift must be ubyte type!", &SI);
+ visitInstruction(SI);
+}
+
void Verifier::visitGetElementPtrInst(GetElementPtrInst &GEP) {
- const Type *ElTy = MemAccessInst::getIndexedType(GEP.getOperand(0)->getType(),
- GEP.copyIndices(), true);
+ const Type *ElTy =
+ GetElementPtrInst::getIndexedType(GEP.getOperand(0)->getType(),
+ std::vector<Value*>(GEP.idx_begin(), GEP.idx_end()), 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);
}
void Verifier::visitLoadInst(LoadInst &LI) {
- const Type *ElTy = LoadInst::getIndexedType(LI.getOperand(0)->getType(),
- LI.copyIndices());
- Assert1(ElTy, "Invalid indices for load pointer type!", &LI);
+ const Type *ElTy =
+ cast<PointerType>(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 = StoreInst::getIndexedType(SI.getOperand(1)->getType(),
- SI.copyIndices());
- Assert1(ElTy, "Invalid indices for store pointer type!", &SI);
+ const Type *ElTy =
+ cast<PointerType>(SI.getOperand(1)->getType())->getElementType();
Assert2(ElTy == SI.getOperand(0)->getType(),
"Stored value is not of right type for indices!", &SI, ElTy);
visitInstruction(SI);
}
-// verifyInstruction - Verify that a non-terminator instruction is well formed.
+// verifyInstruction - Verify that an instruction is well formed.
//
void Verifier::visitInstruction(Instruction &I) {
- Assert1(I.getParent(), "Instruction not embedded in basic block!", &I);
+ BasicBlock *BB = I.getParent();
+ Assert1(BB, "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
"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<Instruction>(*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<PHINode>(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);
+
+ // Use must be dominated by by definition unless use is unreachable!
+ Assert2(DS->dominates(BB, Pred) ||
+ !DS->dominates(&BB->getParent()->getEntryNode(), Pred),
+ "Instruction does not dominate all uses!",
+ &I, PN);
+ }
+
+ } else {
+ // Use must be dominated by by definition unless use is unreachable!
+ Assert2(DS->dominates(&I, Use) ||
+ !DS->dominates(&BB->getParent()->getEntryNode(),Use->getParent()),
+ "Instruction does not dominate all uses!", &I, Use);
+ }
+ }
+
+ // Check to make sure that the "address of" an intrinsic function is never
+ // taken.
+ for (unsigned i = 0, e = I.getNumOperands(); i != e; ++i)
+ if (Function *F = dyn_cast<Function>(I.getOperand(i)))
+ Assert1(!F->isIntrinsic() || (i == 0 && isa<CallInst>(I)),
+ "Cannot take the address of an intrinsic!", &I);
+}
+
+/// visitIntrinsicFunction - Allow intrinsics to be verified in different ways.
+void Verifier::visitIntrinsicFunctionCall(LLVMIntrinsic::ID ID, CallInst &CI) {
+ Function *IF = CI.getCalledFunction();
+ const FunctionType *FT = IF->getFunctionType();
+ Assert1(IF->isExternal(), "Intrinsic functions should never be defined!", IF);
+ unsigned NumArgs = 0;
+
+ // FIXME: this should check the return type of each intrinsic as well, also
+ // arguments!
+ switch (ID) {
+ case LLVMIntrinsic::va_start:
+ Assert1(CI.getParent()->getParent()->getFunctionType()->isVarArg(),
+ "llvm.va_start intrinsic may only occur in function with variable"
+ " args!", &CI);
+ NumArgs = 1;
+ break;
+ case LLVMIntrinsic::va_end: NumArgs = 1; break;
+ case LLVMIntrinsic::va_copy: NumArgs = 2; break;
+
+ case LLVMIntrinsic::setjmp: NumArgs = 1; break;
+ case LLVMIntrinsic::longjmp: NumArgs = 2; break;
+ case LLVMIntrinsic::sigsetjmp: NumArgs = 2; break;
+ case LLVMIntrinsic::siglongjmp: NumArgs = 2; break;
+
+ case LLVMIntrinsic::alpha_ctlz: NumArgs = 1; break;
+ case LLVMIntrinsic::alpha_cttz: NumArgs = 1; break;
+ case LLVMIntrinsic::alpha_ctpop: NumArgs = 1; break;
+ case LLVMIntrinsic::alpha_umulh: NumArgs = 2; break;
+ case LLVMIntrinsic::alpha_vecop: NumArgs = 4; break;
+ case LLVMIntrinsic::alpha_pup: NumArgs = 3; break;
+ case LLVMIntrinsic::alpha_bytezap: NumArgs = 2; break;
+ case LLVMIntrinsic::alpha_bytemanip: NumArgs = 3; break;
+ case LLVMIntrinsic::alpha_dfpbop: NumArgs = 3; break;
+ case LLVMIntrinsic::alpha_dfpuop: NumArgs = 2; break;
+ case LLVMIntrinsic::alpha_unordered: NumArgs = 2; break;
+ case LLVMIntrinsic::alpha_uqtodfp: NumArgs = 2; break;
+ case LLVMIntrinsic::alpha_uqtosfp: NumArgs = 2; break;
+ case LLVMIntrinsic::alpha_dfptosq: NumArgs = 2; break;
+ case LLVMIntrinsic::alpha_sfptosq: NumArgs = 2; break;
+
+ case LLVMIntrinsic::not_intrinsic:
+ assert(0 && "Invalid intrinsic!"); NumArgs = 0; break;
+ }
+
+ Assert1(FT->getNumParams() == NumArgs || (FT->getNumParams() < NumArgs &&
+ FT->isVarArg()),
+ "Illegal # arguments for intrinsic function!", IF);
}
// Implement the public interfaces to this file...
//===----------------------------------------------------------------------===//
-Pass *createVerifierPass() {
+FunctionPass *createVerifierPass() {
return new Verifier();
}
-bool verifyFunction(const Function &F) {
- Verifier V;
- V.visit((Function&)F);
+
+// 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;
}
// Return true if the module is corrupt.
//
bool verifyModule(const Module &M) {
- Verifier V;
- V.run((Module&)M);
- return V.Broken;
+ PassManager PM;
+ Verifier *V = new Verifier();
+ PM.add(V);
+ PM.run((Module&)M);
+ return V->Broken;
}