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/Pass.h"
44 #include "llvm/Module.h"
45 #include "llvm/DerivedTypes.h"
46 #include "llvm/iPHINode.h"
47 #include "llvm/iTerminators.h"
48 #include "llvm/iOther.h"
49 #include "llvm/iOperators.h"
50 #include "llvm/iMemory.h"
51 #include "llvm/SymbolTable.h"
52 #include "llvm/PassManager.h"
53 #include "llvm/Intrinsics.h"
54 #include "llvm/Analysis/Dominators.h"
55 #include "llvm/Support/CFG.h"
56 #include "llvm/Support/InstVisitor.h"
57 #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?
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) {
78 verifySymbolTable(M.getSymbolTable());
80 // If this is a real pass, in a pass manager, we must abort before
81 // returning back to the pass manager, or else the pass manager may try to
82 // run other passes on the broken module.
89 bool runOnFunction(Function &F) {
90 // Get dominator information if we are being run by PassManager
91 if (RealPass) DS = &getAnalysis<DominatorSet>();
94 // If this is a real pass, in a pass manager, we must abort before
95 // returning back to the pass manager, or else the pass manager may try to
96 // run other passes on the broken module.
104 bool doFinalization(Module &M) {
105 // Scan through, checking all of the external function's linkage now...
106 for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I)
107 visitGlobalValue(*I);
109 for (Module::giterator I = M.gbegin(), E = M.gend(); I != E; ++I)
110 if (I->isExternal() && I->hasInternalLinkage())
111 CheckFailed("Global Variable is external with internal linkage!", 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);
156 // CheckFailed - A check failed, so print out the condition and the message
157 // that failed. This provides a nice place to put a breakpoint if you want
158 // to see why something is not correct.
160 inline void CheckFailed(const std::string &Message,
161 const Value *V1 = 0, const Value *V2 = 0,
162 const Value *V3 = 0, const Value *V4 = 0) {
163 std::cerr << Message << "\n";
164 if (V1) std::cerr << *V1 << "\n";
165 if (V2) std::cerr << *V2 << "\n";
166 if (V3) std::cerr << *V3 << "\n";
167 if (V4) std::cerr << *V4 << "\n";
172 RegisterPass<Verifier> X("verify", "Module Verifier");
174 // Assert - We know that cond should be true, if not print an error message.
175 #define Assert(C, M) \
176 do { if (!(C)) { CheckFailed(M); return; } } while (0)
177 #define Assert1(C, M, V1) \
178 do { if (!(C)) { CheckFailed(M, V1); return; } } while (0)
179 #define Assert2(C, M, V1, V2) \
180 do { if (!(C)) { CheckFailed(M, V1, V2); return; } } while (0)
181 #define Assert3(C, M, V1, V2, V3) \
182 do { if (!(C)) { CheckFailed(M, V1, V2, V3); return; } } while (0)
183 #define Assert4(C, M, V1, V2, V3, V4) \
184 do { if (!(C)) { CheckFailed(M, V1, V2, V3, V4); return; } } while (0)
187 void Verifier::visitGlobalValue(GlobalValue &GV) {
188 Assert1(!GV.isExternal() || GV.hasExternalLinkage(),
189 "Global value has Internal Linkage!", &GV);
190 Assert1(!GV.hasAppendingLinkage() || isa<GlobalVariable>(GV),
191 "Only global variables can have appending linkage!", &GV);
193 if (GV.hasAppendingLinkage()) {
194 GlobalVariable &GVar = cast<GlobalVariable>(GV);
195 Assert1(isa<ArrayType>(GVar.getType()->getElementType()),
196 "Only global arrays can have appending linkage!", &GV);
200 // verifySymbolTable - Verify that a function or module symbol table is ok
202 void Verifier::verifySymbolTable(SymbolTable &ST) {
203 // Loop over all of the types in the symbol table...
204 for (SymbolTable::iterator TI = ST.begin(), TE = ST.end(); TI != TE; ++TI)
205 for (SymbolTable::type_iterator I = TI->second.begin(),
206 E = TI->second.end(); I != E; ++I) {
207 Value *V = I->second;
209 // Check that there are no void typed values in the symbol table. Values
210 // with a void type cannot be put into symbol tables because they cannot
212 Assert1(V->getType() != Type::VoidTy,
213 "Values with void type are not allowed to have names!", V);
218 // visitFunction - Verify that a function is ok.
220 void Verifier::visitFunction(Function &F) {
221 // Check function arguments...
222 const FunctionType *FT = F.getFunctionType();
223 unsigned NumArgs = F.getArgumentList().size();
225 Assert2(FT->getNumParams() == NumArgs,
226 "# formal arguments must match # of arguments for function type!",
229 // Check that the argument values match the function type for this function...
231 for (Function::aiterator I = F.abegin(), E = F.aend(); I != E; ++I, ++i)
232 Assert2(I->getType() == FT->getParamType(i),
233 "Argument value does not match function argument type!",
234 I, FT->getParamType(i));
236 if (!F.isExternal()) {
237 verifySymbolTable(F.getSymbolTable());
239 // Check the entry node
240 BasicBlock *Entry = &F.getEntryBlock();
241 Assert1(pred_begin(Entry) == pred_end(Entry),
242 "Entry block to function must not have predecessors!", Entry);
247 // verifyBasicBlock - Verify that a basic block is well formed...
249 void Verifier::visitBasicBlock(BasicBlock &BB) {
250 // Check constraints that this basic block imposes on all of the PHI nodes in
252 if (isa<PHINode>(BB.front())) {
253 std::vector<BasicBlock*> Preds(pred_begin(&BB), pred_end(&BB));
254 std::sort(Preds.begin(), Preds.end());
256 for (BasicBlock::iterator I = BB.begin();
257 PHINode *PN = dyn_cast<PHINode>(I); ++I) {
259 // Ensure that PHI nodes have at least one entry!
260 Assert1(PN->getNumIncomingValues() != 0,
261 "PHI nodes must have at least one entry. If the block is dead, "
262 "the PHI should be removed!", PN);
263 Assert1(PN->getNumIncomingValues() >= Preds.size(),
264 "PHINode has more entries than the basic block has predecessors!",
266 Assert1(PN->getNumIncomingValues() <= Preds.size(),
267 "PHINode has less entries than the basic block has predecessors!",
270 // Get and sort all incoming values in the PHI node...
271 std::vector<std::pair<BasicBlock*, Value*> > Values;
272 Values.reserve(PN->getNumIncomingValues());
273 for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
274 Values.push_back(std::make_pair(PN->getIncomingBlock(i),
275 PN->getIncomingValue(i)));
276 std::sort(Values.begin(), Values.end());
278 for (unsigned i = 0, e = Values.size(); i != e; ++i) {
279 // Check to make sure that if there is more than one entry for a
280 // particular basic block in this PHI node, that the incoming values are
283 Assert4(i == 0 || Values[i].first != Values[i-1].first ||
284 Values[i].second == Values[i-1].second,
285 "PHI node has multiple entries for the same basic block with "
286 "different incoming values!", PN, Values[i].first,
287 Values[i].second, Values[i-1].second);
289 // Check to make sure that the predecessors and PHI node entries are
291 Assert3(Values[i].first == Preds[i],
292 "PHI node entries do not match predecessors!", PN,
293 Values[i].first, Preds[i]);
298 // Ensure that basic blocks have terminators!
299 Assert1(BB.getTerminator(), "Basic Block does not have terminator!", &BB);
302 void Verifier::visitTerminatorInst(TerminatorInst &I) {
303 // Ensure that terminators only exist at the end of the basic block.
304 Assert1(&I == I.getParent()->getTerminator(),
305 "Terminator found in the middle of a basic block!", I.getParent());
309 void Verifier::visitReturnInst(ReturnInst &RI) {
310 Function *F = RI.getParent()->getParent();
311 if (RI.getNumOperands() == 0)
312 Assert1(F->getReturnType() == Type::VoidTy,
313 "Function returns no value, but ret instruction found that does!",
316 Assert2(F->getReturnType() == RI.getOperand(0)->getType(),
317 "Function return type does not match operand "
318 "type of return inst!", &RI, F->getReturnType());
320 // Check to make sure that the return value has necessary properties for
322 visitTerminatorInst(RI);
325 // visitUserOp1 - User defined operators shouldn't live beyond the lifetime of a
326 // pass, if any exist, it's an error.
328 void Verifier::visitUserOp1(Instruction &I) {
329 Assert1(0, "User-defined operators should not live outside of a pass!",
333 // visitPHINode - Ensure that a PHI node is well formed.
334 void Verifier::visitPHINode(PHINode &PN) {
335 // Ensure that the PHI nodes are all grouped together at the top of the block.
336 // This can be tested by checking whether the instruction before this is
337 // either nonexistent (because this is begin()) or is a PHI node. If not,
338 // then there is some other instruction before a PHI.
339 Assert2(&PN.getParent()->front() == &PN || isa<PHINode>(PN.getPrev()),
340 "PHI nodes not grouped at top of basic block!",
341 &PN, PN.getParent());
343 // Check that all of the operands of the PHI node have the same type as the
345 for (unsigned i = 0, e = PN.getNumIncomingValues(); i != e; ++i)
346 Assert1(PN.getType() == PN.getIncomingValue(i)->getType(),
347 "PHI node operands are not the same type as the result!", &PN);
349 // All other PHI node constraints are checked in the visitBasicBlock method.
351 visitInstruction(PN);
354 void Verifier::visitCallInst(CallInst &CI) {
355 Assert1(isa<PointerType>(CI.getOperand(0)->getType()),
356 "Called function must be a pointer!", &CI);
357 const PointerType *FPTy = cast<PointerType>(CI.getOperand(0)->getType());
358 Assert1(isa<FunctionType>(FPTy->getElementType()),
359 "Called function is not pointer to function type!", &CI);
361 const FunctionType *FTy = cast<FunctionType>(FPTy->getElementType());
363 // Verify that the correct number of arguments are being passed
365 Assert1(CI.getNumOperands()-1 >= FTy->getNumParams(),
366 "Called function requires more parameters than were provided!",&CI);
368 Assert1(CI.getNumOperands()-1 == FTy->getNumParams(),
369 "Incorrect number of arguments passed to called function!", &CI);
371 // Verify that all arguments to the call match the function type...
372 for (unsigned i = 0, e = FTy->getNumParams(); i != e; ++i)
373 Assert2(CI.getOperand(i+1)->getType() == FTy->getParamType(i),
374 "Call parameter type does not match function signature!",
375 CI.getOperand(i+1), FTy->getParamType(i));
377 if (Function *F = CI.getCalledFunction())
378 if (Intrinsic::ID ID = (Intrinsic::ID)F->getIntrinsicID())
379 visitIntrinsicFunctionCall(ID, CI);
381 visitInstruction(CI);
384 // visitBinaryOperator - Check that both arguments to the binary operator are
387 void Verifier::visitBinaryOperator(BinaryOperator &B) {
388 Assert1(B.getOperand(0)->getType() == B.getOperand(1)->getType(),
389 "Both operands to a binary operator are not of the same type!", &B);
391 // Check that logical operators are only used with integral operands.
392 if (B.getOpcode() == Instruction::And || B.getOpcode() == Instruction::Or ||
393 B.getOpcode() == Instruction::Xor) {
394 Assert1(B.getType()->isIntegral(),
395 "Logical operators only work with integral types!", &B);
396 Assert1(B.getType() == B.getOperand(0)->getType(),
397 "Logical operators must have same type for operands and result!",
399 } else if (isa<SetCondInst>(B)) {
400 // Check that setcc instructions return bool
401 Assert1(B.getType() == Type::BoolTy,
402 "setcc instructions must return boolean values!", &B);
404 // Arithmetic operators only work on integer or fp values
405 Assert1(B.getType() == B.getOperand(0)->getType(),
406 "Arithmetic operators must have same type for operands and result!",
408 Assert1(B.getType()->isInteger() || B.getType()->isFloatingPoint(),
409 "Arithmetic operators must have integer or fp type!", &B);
415 void Verifier::visitShiftInst(ShiftInst &SI) {
416 Assert1(SI.getType()->isInteger(),
417 "Shift must return an integer result!", &SI);
418 Assert1(SI.getType() == SI.getOperand(0)->getType(),
419 "Shift return type must be same as first operand!", &SI);
420 Assert1(SI.getOperand(1)->getType() == Type::UByteTy,
421 "Second operand to shift must be ubyte type!", &SI);
422 visitInstruction(SI);
425 void Verifier::visitGetElementPtrInst(GetElementPtrInst &GEP) {
427 GetElementPtrInst::getIndexedType(GEP.getOperand(0)->getType(),
428 std::vector<Value*>(GEP.idx_begin(), GEP.idx_end()), true);
429 Assert1(ElTy, "Invalid indices for GEP pointer type!", &GEP);
430 Assert2(PointerType::get(ElTy) == GEP.getType(),
431 "GEP is not of right type for indices!", &GEP, ElTy);
432 visitInstruction(GEP);
435 void Verifier::visitLoadInst(LoadInst &LI) {
437 cast<PointerType>(LI.getOperand(0)->getType())->getElementType();
438 Assert2(ElTy == LI.getType(),
439 "Load is not of right type for indices!", &LI, ElTy);
440 visitInstruction(LI);
443 void Verifier::visitStoreInst(StoreInst &SI) {
445 cast<PointerType>(SI.getOperand(1)->getType())->getElementType();
446 Assert2(ElTy == SI.getOperand(0)->getType(),
447 "Stored value is not of right type for indices!", &SI, ElTy);
448 visitInstruction(SI);
452 // verifyInstruction - Verify that an instruction is well formed.
454 void Verifier::visitInstruction(Instruction &I) {
455 BasicBlock *BB = I.getParent();
456 Assert1(BB, "Instruction not embedded in basic block!", &I);
458 if (!isa<PHINode>(I)) { // Check that non-phi nodes are not self referential
459 for (Value::use_iterator UI = I.use_begin(), UE = I.use_end();
461 Assert1(*UI != (User*)&I,
462 "Only PHI nodes may reference their own value!", &I);
465 // Check that void typed values don't have names
466 Assert1(I.getType() != Type::VoidTy || !I.hasName(),
467 "Instruction has a name, but provides a void value!", &I);
469 // Check that all uses of the instruction, if they are instructions
470 // themselves, actually have parent basic blocks. If the use is not an
471 // instruction, it is an error!
473 for (User::use_iterator UI = I.use_begin(), UE = I.use_end();
475 Assert1(isa<Instruction>(*UI), "Use of instruction is not an instruction!",
477 Instruction *Used = cast<Instruction>(*UI);
478 Assert2(Used->getParent() != 0, "Instruction referencing instruction not"
479 " embeded in a basic block!", &I, Used);
482 for (unsigned i = 0, e = I.getNumOperands(); i != e; ++i) {
483 // Check to make sure that the "address of" an intrinsic function is never
485 if (Function *F = dyn_cast<Function>(I.getOperand(i)))
486 Assert1(!F->isIntrinsic() || (i == 0 && isa<CallInst>(I)),
487 "Cannot take the address of an intrinsic!", &I);
489 else if (Instruction *Op = dyn_cast<Instruction>(I.getOperand(i))) {
490 // Check that a definition dominates all of its uses.
492 if (!isa<PHINode>(I)) {
493 // Definition must dominate use unless use is unreachable!
494 Assert2(DS->dominates(Op->getParent(), BB) ||
495 !DS->dominates(&BB->getParent()->getEntryBlock(), BB),
496 "Instruction does not dominate all uses!", Op, &I);
498 // PHI nodes are more difficult than other nodes because they actually
499 // "use" the value in the predecessor basic blocks they correspond to.
500 BasicBlock *PredBB = cast<BasicBlock>(I.getOperand(i+1));
501 Assert2(DS->dominates(Op->getParent(), PredBB) ||
502 !DS->dominates(&BB->getParent()->getEntryBlock(), PredBB),
503 "Instruction does not dominate all uses!", Op, &I);
509 /// visitIntrinsicFunction - Allow intrinsics to be verified in different ways.
510 void Verifier::visitIntrinsicFunctionCall(Intrinsic::ID ID, CallInst &CI) {
511 Function *IF = CI.getCalledFunction();
512 const FunctionType *FT = IF->getFunctionType();
513 Assert1(IF->isExternal(), "Intrinsic functions should never be defined!", IF);
514 unsigned NumArgs = 0;
516 // FIXME: this should check the return type of each intrinsic as well, also
519 case Intrinsic::va_start:
520 Assert1(CI.getParent()->getParent()->getFunctionType()->isVarArg(),
521 "llvm.va_start intrinsic may only occur in function with variable"
525 case Intrinsic::va_end: NumArgs = 1; break;
526 case Intrinsic::va_copy: NumArgs = 1; break;
528 case Intrinsic::setjmp: NumArgs = 1; break;
529 case Intrinsic::longjmp: NumArgs = 2; break;
530 case Intrinsic::sigsetjmp: NumArgs = 2; break;
531 case Intrinsic::siglongjmp: NumArgs = 2; break;
533 case Intrinsic::alpha_ctlz: NumArgs = 1; break;
534 case Intrinsic::alpha_cttz: NumArgs = 1; break;
535 case Intrinsic::alpha_ctpop: NumArgs = 1; break;
536 case Intrinsic::alpha_umulh: NumArgs = 2; break;
537 case Intrinsic::alpha_vecop: NumArgs = 4; break;
538 case Intrinsic::alpha_pup: NumArgs = 3; break;
539 case Intrinsic::alpha_bytezap: NumArgs = 2; break;
540 case Intrinsic::alpha_bytemanip: NumArgs = 3; break;
541 case Intrinsic::alpha_dfpbop: NumArgs = 3; break;
542 case Intrinsic::alpha_dfpuop: NumArgs = 2; break;
543 case Intrinsic::alpha_unordered: NumArgs = 2; break;
544 case Intrinsic::alpha_uqtodfp: NumArgs = 2; break;
545 case Intrinsic::alpha_uqtosfp: NumArgs = 2; break;
546 case Intrinsic::alpha_dfptosq: NumArgs = 2; break;
547 case Intrinsic::alpha_sfptosq: NumArgs = 2; break;
549 case Intrinsic::not_intrinsic:
550 assert(0 && "Invalid intrinsic!"); NumArgs = 0; break;
553 Assert1(FT->getNumParams() == NumArgs || (FT->getNumParams() < NumArgs &&
555 "Illegal # arguments for intrinsic function!", IF);
558 } // End anonymous namespace
560 //===----------------------------------------------------------------------===//
561 // Implement the public interfaces to this file...
562 //===----------------------------------------------------------------------===//
564 FunctionPass *createVerifierPass() {
565 return new Verifier();
569 // verifyFunction - Create
570 bool verifyFunction(const Function &f) {
571 Function &F = (Function&)f;
572 assert(!F.isExternal() && "Cannot verify external functions");
575 DS.doInitialization(*F.getParent());
581 DS.doFinalization(*F.getParent());
586 // verifyModule - Check a module for errors, printing messages on stderr.
587 // Return true if the module is corrupt.
589 bool verifyModule(const Module &M) {
591 Verifier *V = new Verifier();
597 } // End llvm namespace