1 //===-- Verifier.cpp - Implement the Module Verifier -------------*- C++ -*-==//
3 // The LLVM Compiler Infrastructure
5 // This file was developed by the LLVM research group and is distributed under
6 // the University of Illinois Open Source License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file defines the function verifier interface, that can be used for some
11 // sanity checking of input to the system.
13 // Note that this does not provide full 'java style' security and verifications,
14 // instead it just tries to ensure that code is well formed.
16 // * Both of a binary operator's parameters are the same type
17 // * Verify that the indices of mem access instructions match other operands
18 // * Verify that arithmetic and other things are only performed on first class
19 // types. Verify that shifts & logicals only happen on integrals f.e.
20 // . All of the constants in a switch statement are of the correct type
21 // * The code is in valid SSA form
22 // . It should be illegal to put a label into any other type (like a structure)
23 // or to return one. [except constant arrays!]
24 // * Only phi nodes can be self referential: 'add int %0, %0 ; <int>:0' is bad
25 // * PHI nodes must have an entry for each predecessor, with no extras.
26 // * PHI nodes must be the first thing in a basic block, all grouped together
27 // * PHI nodes must have at least one entry
28 // * All basic blocks should only end with terminator insts, not contain them
29 // * The entry node to a function must not have predecessors
30 // * All Instructions must be embedded into a basic block
31 // . Function's cannot take a void typed parameter
32 // * Verify that a function's argument list agrees with it's declared type.
33 // * It is illegal to specify a name for a void value.
34 // * It is illegal to have a internal global value with no initializer
35 // * It is illegal to have a ret instruction that returns a value that does not
36 // agree with the function return value type.
37 // * Function call argument types match the function prototype
38 // * All other things that are tested by asserts spread about the code...
40 //===----------------------------------------------------------------------===//
42 #include "llvm/Analysis/Verifier.h"
43 #include "llvm/Assembly/Writer.h"
44 #include "llvm/Pass.h"
45 #include "llvm/Module.h"
46 #include "llvm/DerivedTypes.h"
47 #include "llvm/iPHINode.h"
48 #include "llvm/iTerminators.h"
49 #include "llvm/iOther.h"
50 #include "llvm/iOperators.h"
51 #include "llvm/iMemory.h"
52 #include "llvm/SymbolTable.h"
53 #include "llvm/PassManager.h"
54 #include "llvm/Intrinsics.h"
55 #include "llvm/Analysis/Dominators.h"
56 #include "llvm/Support/CFG.h"
57 #include "llvm/Support/InstVisitor.h"
58 #include "Support/STLExtras.h"
62 namespace { // Anonymous namespace for class
64 struct Verifier : public FunctionPass, InstVisitor<Verifier> {
65 bool Broken; // Is this module found to be broken?
66 bool RealPass; // Are we not being run by a PassManager?
67 bool AbortBroken; // If broken, should it or should it not abort?
68 Module *Mod; // Module we are verifying right now
69 DominatorSet *DS; // Dominator set, caution can be null!
71 Verifier() : Broken(false), RealPass(true), AbortBroken(true), DS(0) {}
72 Verifier(bool AB) : Broken(false), RealPass(true), AbortBroken(AB), DS(0) {}
73 Verifier(DominatorSet &ds)
74 : Broken(false), RealPass(false), AbortBroken(false), DS(&ds) {}
77 bool doInitialization(Module &M) {
79 verifySymbolTable(M.getSymbolTable());
81 // If this is a real pass, in a pass manager, we must abort before
82 // returning back to the pass manager, or else the pass manager may try to
83 // run other passes on the broken module.
90 bool runOnFunction(Function &F) {
91 // Get dominator information if we are being run by PassManager
92 if (RealPass) DS = &getAnalysis<DominatorSet>();
95 // If this is a real pass, in a pass manager, we must abort before
96 // returning back to the pass manager, or else the pass manager may try to
97 // run other passes on the broken module.
105 bool doFinalization(Module &M) {
106 // Scan through, checking all of the external function's linkage now...
107 for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I)
108 visitGlobalValue(*I);
110 for (Module::giterator I = M.gbegin(), E = M.gend(); I != E; ++I)
111 visitGlobalValue(*I);
113 // If the module is broken, abort at this time.
118 virtual void getAnalysisUsage(AnalysisUsage &AU) const {
119 AU.setPreservesAll();
121 AU.addRequired<DominatorSet>();
124 // abortIfBroken - If the module is broken and we are supposed to abort on
125 // this condition, do so.
127 void abortIfBroken() const {
128 if (Broken && AbortBroken) {
129 std::cerr << "Broken module found, compilation aborted!\n";
135 // Verification methods...
136 void verifySymbolTable(SymbolTable &ST);
137 void visitGlobalValue(GlobalValue &GV);
138 void visitFunction(Function &F);
139 void visitBasicBlock(BasicBlock &BB);
140 void visitPHINode(PHINode &PN);
141 void visitBinaryOperator(BinaryOperator &B);
142 void visitShiftInst(ShiftInst &SI);
143 void visitVANextInst(VANextInst &VAN) { visitInstruction(VAN); }
144 void visitVAArgInst(VAArgInst &VAA) { visitInstruction(VAA); }
145 void visitCallInst(CallInst &CI);
146 void visitGetElementPtrInst(GetElementPtrInst &GEP);
147 void visitLoadInst(LoadInst &LI);
148 void visitStoreInst(StoreInst &SI);
149 void visitInstruction(Instruction &I);
150 void visitTerminatorInst(TerminatorInst &I);
151 void visitReturnInst(ReturnInst &RI);
152 void visitUserOp1(Instruction &I);
153 void visitUserOp2(Instruction &I) { visitUserOp1(I); }
154 void visitIntrinsicFunctionCall(Intrinsic::ID ID, CallInst &CI);
157 void WriteValue(const Value *V) {
159 if (isa<Instruction>(V)) {
161 } else if (const Type *Ty = dyn_cast<Type>(V)) {
162 WriteTypeSymbolic(std::cerr, Ty, Mod);
164 WriteAsOperand (std::cerr, V, true, true, Mod);
170 // CheckFailed - A check failed, so print out the condition and the message
171 // that failed. This provides a nice place to put a breakpoint if you want
172 // to see why something is not correct.
174 void CheckFailed(const std::string &Message,
175 const Value *V1 = 0, const Value *V2 = 0,
176 const Value *V3 = 0, const Value *V4 = 0) {
177 std::cerr << Message << "\n";
186 RegisterOpt<Verifier> X("verify", "Module Verifier");
187 } // End anonymous namespace
190 // Assert - We know that cond should be true, if not print an error message.
191 #define Assert(C, M) \
192 do { if (!(C)) { CheckFailed(M); return; } } while (0)
193 #define Assert1(C, M, V1) \
194 do { if (!(C)) { CheckFailed(M, V1); return; } } while (0)
195 #define Assert2(C, M, V1, V2) \
196 do { if (!(C)) { CheckFailed(M, V1, V2); return; } } while (0)
197 #define Assert3(C, M, V1, V2, V3) \
198 do { if (!(C)) { CheckFailed(M, V1, V2, V3); return; } } while (0)
199 #define Assert4(C, M, V1, V2, V3, V4) \
200 do { if (!(C)) { CheckFailed(M, V1, V2, V3, V4); return; } } while (0)
203 void Verifier::visitGlobalValue(GlobalValue &GV) {
204 Assert1(!GV.isExternal() || GV.hasExternalLinkage(),
205 "Global is external, but doesn't have external linkage!", &GV);
206 Assert1(!GV.hasAppendingLinkage() || isa<GlobalVariable>(GV),
207 "Only global variables can have appending linkage!", &GV);
209 if (GV.hasAppendingLinkage()) {
210 GlobalVariable &GVar = cast<GlobalVariable>(GV);
211 Assert1(isa<ArrayType>(GVar.getType()->getElementType()),
212 "Only global arrays can have appending linkage!", &GV);
216 // verifySymbolTable - Verify that a function or module symbol table is ok
218 void Verifier::verifySymbolTable(SymbolTable &ST) {
219 // Loop over all of the types in the symbol table...
220 for (SymbolTable::iterator TI = ST.begin(), TE = ST.end(); TI != TE; ++TI)
221 for (SymbolTable::type_iterator I = TI->second.begin(),
222 E = TI->second.end(); I != E; ++I) {
223 Value *V = I->second;
225 // Check that there are no void typed values in the symbol table. Values
226 // with a void type cannot be put into symbol tables because they cannot
228 Assert1(V->getType() != Type::VoidTy,
229 "Values with void type are not allowed to have names!", V);
234 // visitFunction - Verify that a function is ok.
236 void Verifier::visitFunction(Function &F) {
237 // Check function arguments...
238 const FunctionType *FT = F.getFunctionType();
239 unsigned NumArgs = F.getArgumentList().size();
241 Assert2(FT->getNumParams() == NumArgs,
242 "# formal arguments must match # of arguments for function type!",
245 // Check that the argument values match the function type for this function...
247 for (Function::aiterator I = F.abegin(), E = F.aend(); I != E; ++I, ++i)
248 Assert2(I->getType() == FT->getParamType(i),
249 "Argument value does not match function argument type!",
250 I, FT->getParamType(i));
252 if (!F.isExternal()) {
253 verifySymbolTable(F.getSymbolTable());
255 // Check the entry node
256 BasicBlock *Entry = &F.getEntryBlock();
257 Assert1(pred_begin(Entry) == pred_end(Entry),
258 "Entry block to function must not have predecessors!", Entry);
263 // verifyBasicBlock - Verify that a basic block is well formed...
265 void Verifier::visitBasicBlock(BasicBlock &BB) {
266 // Check constraints that this basic block imposes on all of the PHI nodes in
268 if (isa<PHINode>(BB.front())) {
269 std::vector<BasicBlock*> Preds(pred_begin(&BB), pred_end(&BB));
270 std::sort(Preds.begin(), Preds.end());
272 for (BasicBlock::iterator I = BB.begin();
273 PHINode *PN = dyn_cast<PHINode>(I); ++I) {
275 // Ensure that PHI nodes have at least one entry!
276 Assert1(PN->getNumIncomingValues() != 0,
277 "PHI nodes must have at least one entry. If the block is dead, "
278 "the PHI should be removed!", PN);
279 Assert1(PN->getNumIncomingValues() >= Preds.size(),
280 "PHINode has more entries than the basic block has predecessors!",
282 Assert1(PN->getNumIncomingValues() <= Preds.size(),
283 "PHINode has less entries than the basic block has predecessors!",
286 // Get and sort all incoming values in the PHI node...
287 std::vector<std::pair<BasicBlock*, Value*> > Values;
288 Values.reserve(PN->getNumIncomingValues());
289 for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
290 Values.push_back(std::make_pair(PN->getIncomingBlock(i),
291 PN->getIncomingValue(i)));
292 std::sort(Values.begin(), Values.end());
294 for (unsigned i = 0, e = Values.size(); i != e; ++i) {
295 // Check to make sure that if there is more than one entry for a
296 // particular basic block in this PHI node, that the incoming values are
299 Assert4(i == 0 || Values[i].first != Values[i-1].first ||
300 Values[i].second == Values[i-1].second,
301 "PHI node has multiple entries for the same basic block with "
302 "different incoming values!", PN, Values[i].first,
303 Values[i].second, Values[i-1].second);
305 // Check to make sure that the predecessors and PHI node entries are
307 Assert3(Values[i].first == Preds[i],
308 "PHI node entries do not match predecessors!", PN,
309 Values[i].first, Preds[i]);
314 // Ensure that basic blocks have terminators!
315 Assert1(BB.getTerminator(), "Basic Block does not have terminator!", &BB);
318 void Verifier::visitTerminatorInst(TerminatorInst &I) {
319 // Ensure that terminators only exist at the end of the basic block.
320 Assert1(&I == I.getParent()->getTerminator(),
321 "Terminator found in the middle of a basic block!", I.getParent());
325 void Verifier::visitReturnInst(ReturnInst &RI) {
326 Function *F = RI.getParent()->getParent();
327 if (RI.getNumOperands() == 0)
328 Assert1(F->getReturnType() == Type::VoidTy,
329 "Function returns no value, but ret instruction found that does!",
332 Assert2(F->getReturnType() == RI.getOperand(0)->getType(),
333 "Function return type does not match operand "
334 "type of return inst!", &RI, F->getReturnType());
336 // Check to make sure that the return value has necessary properties for
338 visitTerminatorInst(RI);
341 // visitUserOp1 - User defined operators shouldn't live beyond the lifetime of a
342 // pass, if any exist, it's an error.
344 void Verifier::visitUserOp1(Instruction &I) {
345 Assert1(0, "User-defined operators should not live outside of a pass!",
349 // visitPHINode - Ensure that a PHI node is well formed.
350 void Verifier::visitPHINode(PHINode &PN) {
351 // Ensure that the PHI nodes are all grouped together at the top of the block.
352 // This can be tested by checking whether the instruction before this is
353 // either nonexistent (because this is begin()) or is a PHI node. If not,
354 // then there is some other instruction before a PHI.
355 Assert2(&PN.getParent()->front() == &PN || isa<PHINode>(PN.getPrev()),
356 "PHI nodes not grouped at top of basic block!",
357 &PN, PN.getParent());
359 // Check that all of the operands of the PHI node have the same type as the
361 for (unsigned i = 0, e = PN.getNumIncomingValues(); i != e; ++i)
362 Assert1(PN.getType() == PN.getIncomingValue(i)->getType(),
363 "PHI node operands are not the same type as the result!", &PN);
365 // All other PHI node constraints are checked in the visitBasicBlock method.
367 visitInstruction(PN);
370 void Verifier::visitCallInst(CallInst &CI) {
371 Assert1(isa<PointerType>(CI.getOperand(0)->getType()),
372 "Called function must be a pointer!", &CI);
373 const PointerType *FPTy = cast<PointerType>(CI.getOperand(0)->getType());
374 Assert1(isa<FunctionType>(FPTy->getElementType()),
375 "Called function is not pointer to function type!", &CI);
377 const FunctionType *FTy = cast<FunctionType>(FPTy->getElementType());
379 // Verify that the correct number of arguments are being passed
381 Assert1(CI.getNumOperands()-1 >= FTy->getNumParams(),
382 "Called function requires more parameters than were provided!",&CI);
384 Assert1(CI.getNumOperands()-1 == FTy->getNumParams(),
385 "Incorrect number of arguments passed to called function!", &CI);
387 // Verify that all arguments to the call match the function type...
388 for (unsigned i = 0, e = FTy->getNumParams(); i != e; ++i)
389 Assert2(CI.getOperand(i+1)->getType() == FTy->getParamType(i),
390 "Call parameter type does not match function signature!",
391 CI.getOperand(i+1), FTy->getParamType(i));
393 if (Function *F = CI.getCalledFunction())
394 if (Intrinsic::ID ID = (Intrinsic::ID)F->getIntrinsicID())
395 visitIntrinsicFunctionCall(ID, CI);
397 visitInstruction(CI);
400 // visitBinaryOperator - Check that both arguments to the binary operator are
403 void Verifier::visitBinaryOperator(BinaryOperator &B) {
404 Assert1(B.getOperand(0)->getType() == B.getOperand(1)->getType(),
405 "Both operands to a binary operator are not of the same type!", &B);
407 // Check that logical operators are only used with integral operands.
408 if (B.getOpcode() == Instruction::And || B.getOpcode() == Instruction::Or ||
409 B.getOpcode() == Instruction::Xor) {
410 Assert1(B.getType()->isIntegral(),
411 "Logical operators only work with integral types!", &B);
412 Assert1(B.getType() == B.getOperand(0)->getType(),
413 "Logical operators must have same type for operands and result!",
415 } else if (isa<SetCondInst>(B)) {
416 // Check that setcc instructions return bool
417 Assert1(B.getType() == Type::BoolTy,
418 "setcc instructions must return boolean values!", &B);
420 // Arithmetic operators only work on integer or fp values
421 Assert1(B.getType() == B.getOperand(0)->getType(),
422 "Arithmetic operators must have same type for operands and result!",
424 Assert1(B.getType()->isInteger() || B.getType()->isFloatingPoint(),
425 "Arithmetic operators must have integer or fp type!", &B);
431 void Verifier::visitShiftInst(ShiftInst &SI) {
432 Assert1(SI.getType()->isInteger(),
433 "Shift must return an integer result!", &SI);
434 Assert1(SI.getType() == SI.getOperand(0)->getType(),
435 "Shift return type must be same as first operand!", &SI);
436 Assert1(SI.getOperand(1)->getType() == Type::UByteTy,
437 "Second operand to shift must be ubyte type!", &SI);
438 visitInstruction(SI);
441 void Verifier::visitGetElementPtrInst(GetElementPtrInst &GEP) {
443 GetElementPtrInst::getIndexedType(GEP.getOperand(0)->getType(),
444 std::vector<Value*>(GEP.idx_begin(), GEP.idx_end()), true);
445 Assert1(ElTy, "Invalid indices for GEP pointer type!", &GEP);
446 Assert2(PointerType::get(ElTy) == GEP.getType(),
447 "GEP is not of right type for indices!", &GEP, ElTy);
448 visitInstruction(GEP);
451 void Verifier::visitLoadInst(LoadInst &LI) {
453 cast<PointerType>(LI.getOperand(0)->getType())->getElementType();
454 Assert2(ElTy == LI.getType(),
455 "Load result type does not match pointer operand type!", &LI, ElTy);
456 visitInstruction(LI);
459 void Verifier::visitStoreInst(StoreInst &SI) {
461 cast<PointerType>(SI.getOperand(1)->getType())->getElementType();
462 Assert2(ElTy == SI.getOperand(0)->getType(),
463 "Stored value type does not match pointer operand type!", &SI, ElTy);
464 visitInstruction(SI);
468 // verifyInstruction - Verify that an instruction is well formed.
470 void Verifier::visitInstruction(Instruction &I) {
471 BasicBlock *BB = I.getParent();
472 Assert1(BB, "Instruction not embedded in basic block!", &I);
474 if (!isa<PHINode>(I)) { // Check that non-phi nodes are not self referential
475 for (Value::use_iterator UI = I.use_begin(), UE = I.use_end();
477 Assert1(*UI != (User*)&I,
478 "Only PHI nodes may reference their own value!", &I);
481 // Check that void typed values don't have names
482 Assert1(I.getType() != Type::VoidTy || !I.hasName(),
483 "Instruction has a name, but provides a void value!", &I);
485 // Check that all uses of the instruction, if they are instructions
486 // themselves, actually have parent basic blocks. If the use is not an
487 // instruction, it is an error!
489 for (User::use_iterator UI = I.use_begin(), UE = I.use_end();
491 Assert1(isa<Instruction>(*UI), "Use of instruction is not an instruction!",
493 Instruction *Used = cast<Instruction>(*UI);
494 Assert2(Used->getParent() != 0, "Instruction referencing instruction not"
495 " embeded in a basic block!", &I, Used);
498 for (unsigned i = 0, e = I.getNumOperands(); i != e; ++i) {
499 // Check to make sure that the "address of" an intrinsic function is never
501 if (Function *F = dyn_cast<Function>(I.getOperand(i)))
502 Assert1(!F->isIntrinsic() || (i == 0 && isa<CallInst>(I)),
503 "Cannot take the address of an intrinsic!", &I);
505 else if (Instruction *Op = dyn_cast<Instruction>(I.getOperand(i))) {
506 // Check that a definition dominates all of its uses.
508 if (!isa<PHINode>(I)) {
509 // Definition must dominate use unless use is unreachable!
510 Assert2(DS->dominates(Op->getParent(), BB) ||
511 !DS->dominates(&BB->getParent()->getEntryBlock(), BB),
512 "Instruction does not dominate all uses!", Op, &I);
514 // PHI nodes are more difficult than other nodes because they actually
515 // "use" the value in the predecessor basic blocks they correspond to.
516 BasicBlock *PredBB = cast<BasicBlock>(I.getOperand(i+1));
517 Assert2(DS->dominates(Op->getParent(), PredBB) ||
518 !DS->dominates(&BB->getParent()->getEntryBlock(), PredBB),
519 "Instruction does not dominate all uses!", Op, &I);
525 /// visitIntrinsicFunction - Allow intrinsics to be verified in different ways.
526 void Verifier::visitIntrinsicFunctionCall(Intrinsic::ID ID, CallInst &CI) {
527 Function *IF = CI.getCalledFunction();
528 const FunctionType *FT = IF->getFunctionType();
529 Assert1(IF->isExternal(), "Intrinsic functions should never be defined!", IF);
530 unsigned NumArgs = 0;
532 // FIXME: this should check the return type of each intrinsic as well, also
535 case Intrinsic::va_start:
536 Assert1(CI.getParent()->getParent()->getFunctionType()->isVarArg(),
537 "llvm.va_start intrinsic may only occur in function with variable"
541 case Intrinsic::va_end: NumArgs = 1; break;
542 case Intrinsic::va_copy: NumArgs = 1; break;
544 case Intrinsic::setjmp: NumArgs = 1; break;
545 case Intrinsic::longjmp: NumArgs = 2; break;
546 case Intrinsic::sigsetjmp: NumArgs = 2; break;
547 case Intrinsic::siglongjmp: NumArgs = 2; break;
549 case Intrinsic::alpha_ctlz: NumArgs = 1; break;
550 case Intrinsic::alpha_cttz: NumArgs = 1; break;
551 case Intrinsic::alpha_ctpop: NumArgs = 1; break;
552 case Intrinsic::alpha_umulh: NumArgs = 2; break;
553 case Intrinsic::alpha_vecop: NumArgs = 4; break;
554 case Intrinsic::alpha_pup: NumArgs = 3; break;
555 case Intrinsic::alpha_bytezap: NumArgs = 2; break;
556 case Intrinsic::alpha_bytemanip: NumArgs = 3; break;
557 case Intrinsic::alpha_dfpbop: NumArgs = 3; break;
558 case Intrinsic::alpha_dfpuop: NumArgs = 2; break;
559 case Intrinsic::alpha_unordered: NumArgs = 2; break;
560 case Intrinsic::alpha_uqtodfp: NumArgs = 2; break;
561 case Intrinsic::alpha_uqtosfp: NumArgs = 2; break;
562 case Intrinsic::alpha_dfptosq: NumArgs = 2; break;
563 case Intrinsic::alpha_sfptosq: NumArgs = 2; break;
565 case Intrinsic::not_intrinsic:
566 assert(0 && "Invalid intrinsic!"); NumArgs = 0; break;
569 Assert1(FT->getNumParams() == NumArgs || (FT->getNumParams() < NumArgs &&
571 "Illegal # arguments for intrinsic function!", IF);
575 //===----------------------------------------------------------------------===//
576 // Implement the public interfaces to this file...
577 //===----------------------------------------------------------------------===//
579 FunctionPass *llvm::createVerifierPass() {
580 return new Verifier();
584 // verifyFunction - Create
585 bool llvm::verifyFunction(const Function &f) {
586 Function &F = (Function&)f;
587 assert(!F.isExternal() && "Cannot verify external functions");
590 DS.doInitialization(*F.getParent());
596 DS.doFinalization(*F.getParent());
601 // verifyModule - Check a module for errors, printing messages on stderr.
602 // Return true if the module is corrupt.
604 bool llvm::verifyModule(const Module &M) {
606 Verifier *V = new Verifier();