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 if (I->isExternal() && I->hasInternalLinkage())
112 CheckFailed("Global Variable is external with internal linkage!", I);
114 // If the module is broken, abort at this time.
119 virtual void getAnalysisUsage(AnalysisUsage &AU) const {
120 AU.setPreservesAll();
122 AU.addRequired<DominatorSet>();
125 // abortIfBroken - If the module is broken and we are supposed to abort on
126 // this condition, do so.
128 void abortIfBroken() const {
129 if (Broken && AbortBroken) {
130 std::cerr << "Broken module found, compilation aborted!\n";
136 // Verification methods...
137 void verifySymbolTable(SymbolTable &ST);
138 void visitGlobalValue(GlobalValue &GV);
139 void visitFunction(Function &F);
140 void visitBasicBlock(BasicBlock &BB);
141 void visitPHINode(PHINode &PN);
142 void visitBinaryOperator(BinaryOperator &B);
143 void visitShiftInst(ShiftInst &SI);
144 void visitVANextInst(VANextInst &VAN) { visitInstruction(VAN); }
145 void visitVAArgInst(VAArgInst &VAA) { visitInstruction(VAA); }
146 void visitCallInst(CallInst &CI);
147 void visitGetElementPtrInst(GetElementPtrInst &GEP);
148 void visitLoadInst(LoadInst &LI);
149 void visitStoreInst(StoreInst &SI);
150 void visitInstruction(Instruction &I);
151 void visitTerminatorInst(TerminatorInst &I);
152 void visitReturnInst(ReturnInst &RI);
153 void visitUserOp1(Instruction &I);
154 void visitUserOp2(Instruction &I) { visitUserOp1(I); }
155 void visitIntrinsicFunctionCall(Intrinsic::ID ID, CallInst &CI);
158 void WriteValue(const Value *V) {
160 if (isa<Instruction>(V)) {
162 } else if (const Type *Ty = dyn_cast<Type>(V)) {
163 WriteTypeSymbolic(std::cerr, Ty, Mod);
165 WriteAsOperand (std::cerr, V, true, true, Mod);
171 // CheckFailed - A check failed, so print out the condition and the message
172 // that failed. This provides a nice place to put a breakpoint if you want
173 // to see why something is not correct.
175 void CheckFailed(const std::string &Message,
176 const Value *V1 = 0, const Value *V2 = 0,
177 const Value *V3 = 0, const Value *V4 = 0) {
178 std::cerr << Message << "\n";
187 RegisterOpt<Verifier> X("verify", "Module Verifier");
188 } // End anonymous namespace
191 // Assert - We know that cond should be true, if not print an error message.
192 #define Assert(C, M) \
193 do { if (!(C)) { CheckFailed(M); return; } } while (0)
194 #define Assert1(C, M, V1) \
195 do { if (!(C)) { CheckFailed(M, V1); return; } } while (0)
196 #define Assert2(C, M, V1, V2) \
197 do { if (!(C)) { CheckFailed(M, V1, V2); return; } } while (0)
198 #define Assert3(C, M, V1, V2, V3) \
199 do { if (!(C)) { CheckFailed(M, V1, V2, V3); return; } } while (0)
200 #define Assert4(C, M, V1, V2, V3, V4) \
201 do { if (!(C)) { CheckFailed(M, V1, V2, V3, V4); return; } } while (0)
204 void Verifier::visitGlobalValue(GlobalValue &GV) {
205 Assert1(!GV.isExternal() || GV.hasExternalLinkage(),
206 "Global value has Internal Linkage!", &GV);
207 Assert1(!GV.hasAppendingLinkage() || isa<GlobalVariable>(GV),
208 "Only global variables can have appending linkage!", &GV);
210 if (GV.hasAppendingLinkage()) {
211 GlobalVariable &GVar = cast<GlobalVariable>(GV);
212 Assert1(isa<ArrayType>(GVar.getType()->getElementType()),
213 "Only global arrays can have appending linkage!", &GV);
217 // verifySymbolTable - Verify that a function or module symbol table is ok
219 void Verifier::verifySymbolTable(SymbolTable &ST) {
220 // Loop over all of the types in the symbol table...
221 for (SymbolTable::iterator TI = ST.begin(), TE = ST.end(); TI != TE; ++TI)
222 for (SymbolTable::type_iterator I = TI->second.begin(),
223 E = TI->second.end(); I != E; ++I) {
224 Value *V = I->second;
226 // Check that there are no void typed values in the symbol table. Values
227 // with a void type cannot be put into symbol tables because they cannot
229 Assert1(V->getType() != Type::VoidTy,
230 "Values with void type are not allowed to have names!", V);
235 // visitFunction - Verify that a function is ok.
237 void Verifier::visitFunction(Function &F) {
238 // Check function arguments...
239 const FunctionType *FT = F.getFunctionType();
240 unsigned NumArgs = F.getArgumentList().size();
242 Assert2(FT->getNumParams() == NumArgs,
243 "# formal arguments must match # of arguments for function type!",
246 // Check that the argument values match the function type for this function...
248 for (Function::aiterator I = F.abegin(), E = F.aend(); I != E; ++I, ++i)
249 Assert2(I->getType() == FT->getParamType(i),
250 "Argument value does not match function argument type!",
251 I, FT->getParamType(i));
253 if (!F.isExternal()) {
254 verifySymbolTable(F.getSymbolTable());
256 // Check the entry node
257 BasicBlock *Entry = &F.getEntryBlock();
258 Assert1(pred_begin(Entry) == pred_end(Entry),
259 "Entry block to function must not have predecessors!", Entry);
264 // verifyBasicBlock - Verify that a basic block is well formed...
266 void Verifier::visitBasicBlock(BasicBlock &BB) {
267 // Check constraints that this basic block imposes on all of the PHI nodes in
269 if (isa<PHINode>(BB.front())) {
270 std::vector<BasicBlock*> Preds(pred_begin(&BB), pred_end(&BB));
271 std::sort(Preds.begin(), Preds.end());
273 for (BasicBlock::iterator I = BB.begin();
274 PHINode *PN = dyn_cast<PHINode>(I); ++I) {
276 // Ensure that PHI nodes have at least one entry!
277 Assert1(PN->getNumIncomingValues() != 0,
278 "PHI nodes must have at least one entry. If the block is dead, "
279 "the PHI should be removed!", PN);
280 Assert1(PN->getNumIncomingValues() >= Preds.size(),
281 "PHINode has more entries than the basic block has predecessors!",
283 Assert1(PN->getNumIncomingValues() <= Preds.size(),
284 "PHINode has less entries than the basic block has predecessors!",
287 // Get and sort all incoming values in the PHI node...
288 std::vector<std::pair<BasicBlock*, Value*> > Values;
289 Values.reserve(PN->getNumIncomingValues());
290 for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
291 Values.push_back(std::make_pair(PN->getIncomingBlock(i),
292 PN->getIncomingValue(i)));
293 std::sort(Values.begin(), Values.end());
295 for (unsigned i = 0, e = Values.size(); i != e; ++i) {
296 // Check to make sure that if there is more than one entry for a
297 // particular basic block in this PHI node, that the incoming values are
300 Assert4(i == 0 || Values[i].first != Values[i-1].first ||
301 Values[i].second == Values[i-1].second,
302 "PHI node has multiple entries for the same basic block with "
303 "different incoming values!", PN, Values[i].first,
304 Values[i].second, Values[i-1].second);
306 // Check to make sure that the predecessors and PHI node entries are
308 Assert3(Values[i].first == Preds[i],
309 "PHI node entries do not match predecessors!", PN,
310 Values[i].first, Preds[i]);
315 // Ensure that basic blocks have terminators!
316 Assert1(BB.getTerminator(), "Basic Block does not have terminator!", &BB);
319 void Verifier::visitTerminatorInst(TerminatorInst &I) {
320 // Ensure that terminators only exist at the end of the basic block.
321 Assert1(&I == I.getParent()->getTerminator(),
322 "Terminator found in the middle of a basic block!", I.getParent());
326 void Verifier::visitReturnInst(ReturnInst &RI) {
327 Function *F = RI.getParent()->getParent();
328 if (RI.getNumOperands() == 0)
329 Assert1(F->getReturnType() == Type::VoidTy,
330 "Function returns no value, but ret instruction found that does!",
333 Assert2(F->getReturnType() == RI.getOperand(0)->getType(),
334 "Function return type does not match operand "
335 "type of return inst!", &RI, F->getReturnType());
337 // Check to make sure that the return value has necessary properties for
339 visitTerminatorInst(RI);
342 // visitUserOp1 - User defined operators shouldn't live beyond the lifetime of a
343 // pass, if any exist, it's an error.
345 void Verifier::visitUserOp1(Instruction &I) {
346 Assert1(0, "User-defined operators should not live outside of a pass!",
350 // visitPHINode - Ensure that a PHI node is well formed.
351 void Verifier::visitPHINode(PHINode &PN) {
352 // Ensure that the PHI nodes are all grouped together at the top of the block.
353 // This can be tested by checking whether the instruction before this is
354 // either nonexistent (because this is begin()) or is a PHI node. If not,
355 // then there is some other instruction before a PHI.
356 Assert2(&PN.getParent()->front() == &PN || isa<PHINode>(PN.getPrev()),
357 "PHI nodes not grouped at top of basic block!",
358 &PN, PN.getParent());
360 // Check that all of the operands of the PHI node have the same type as the
362 for (unsigned i = 0, e = PN.getNumIncomingValues(); i != e; ++i)
363 Assert1(PN.getType() == PN.getIncomingValue(i)->getType(),
364 "PHI node operands are not the same type as the result!", &PN);
366 // All other PHI node constraints are checked in the visitBasicBlock method.
368 visitInstruction(PN);
371 void Verifier::visitCallInst(CallInst &CI) {
372 Assert1(isa<PointerType>(CI.getOperand(0)->getType()),
373 "Called function must be a pointer!", &CI);
374 const PointerType *FPTy = cast<PointerType>(CI.getOperand(0)->getType());
375 Assert1(isa<FunctionType>(FPTy->getElementType()),
376 "Called function is not pointer to function type!", &CI);
378 const FunctionType *FTy = cast<FunctionType>(FPTy->getElementType());
380 // Verify that the correct number of arguments are being passed
382 Assert1(CI.getNumOperands()-1 >= FTy->getNumParams(),
383 "Called function requires more parameters than were provided!",&CI);
385 Assert1(CI.getNumOperands()-1 == FTy->getNumParams(),
386 "Incorrect number of arguments passed to called function!", &CI);
388 // Verify that all arguments to the call match the function type...
389 for (unsigned i = 0, e = FTy->getNumParams(); i != e; ++i)
390 Assert2(CI.getOperand(i+1)->getType() == FTy->getParamType(i),
391 "Call parameter type does not match function signature!",
392 CI.getOperand(i+1), FTy->getParamType(i));
394 if (Function *F = CI.getCalledFunction())
395 if (Intrinsic::ID ID = (Intrinsic::ID)F->getIntrinsicID())
396 visitIntrinsicFunctionCall(ID, CI);
398 visitInstruction(CI);
401 // visitBinaryOperator - Check that both arguments to the binary operator are
404 void Verifier::visitBinaryOperator(BinaryOperator &B) {
405 Assert1(B.getOperand(0)->getType() == B.getOperand(1)->getType(),
406 "Both operands to a binary operator are not of the same type!", &B);
408 // Check that logical operators are only used with integral operands.
409 if (B.getOpcode() == Instruction::And || B.getOpcode() == Instruction::Or ||
410 B.getOpcode() == Instruction::Xor) {
411 Assert1(B.getType()->isIntegral(),
412 "Logical operators only work with integral types!", &B);
413 Assert1(B.getType() == B.getOperand(0)->getType(),
414 "Logical operators must have same type for operands and result!",
416 } else if (isa<SetCondInst>(B)) {
417 // Check that setcc instructions return bool
418 Assert1(B.getType() == Type::BoolTy,
419 "setcc instructions must return boolean values!", &B);
421 // Arithmetic operators only work on integer or fp values
422 Assert1(B.getType() == B.getOperand(0)->getType(),
423 "Arithmetic operators must have same type for operands and result!",
425 Assert1(B.getType()->isInteger() || B.getType()->isFloatingPoint(),
426 "Arithmetic operators must have integer or fp type!", &B);
432 void Verifier::visitShiftInst(ShiftInst &SI) {
433 Assert1(SI.getType()->isInteger(),
434 "Shift must return an integer result!", &SI);
435 Assert1(SI.getType() == SI.getOperand(0)->getType(),
436 "Shift return type must be same as first operand!", &SI);
437 Assert1(SI.getOperand(1)->getType() == Type::UByteTy,
438 "Second operand to shift must be ubyte type!", &SI);
439 visitInstruction(SI);
442 void Verifier::visitGetElementPtrInst(GetElementPtrInst &GEP) {
444 GetElementPtrInst::getIndexedType(GEP.getOperand(0)->getType(),
445 std::vector<Value*>(GEP.idx_begin(), GEP.idx_end()), true);
446 Assert1(ElTy, "Invalid indices for GEP pointer type!", &GEP);
447 Assert2(PointerType::get(ElTy) == GEP.getType(),
448 "GEP is not of right type for indices!", &GEP, ElTy);
449 visitInstruction(GEP);
452 void Verifier::visitLoadInst(LoadInst &LI) {
454 cast<PointerType>(LI.getOperand(0)->getType())->getElementType();
455 Assert2(ElTy == LI.getType(),
456 "Load result type does not match pointer operand type!", &LI, ElTy);
457 visitInstruction(LI);
460 void Verifier::visitStoreInst(StoreInst &SI) {
462 cast<PointerType>(SI.getOperand(1)->getType())->getElementType();
463 Assert2(ElTy == SI.getOperand(0)->getType(),
464 "Stored value type does not match pointer operand type!", &SI, ElTy);
465 visitInstruction(SI);
469 // verifyInstruction - Verify that an instruction is well formed.
471 void Verifier::visitInstruction(Instruction &I) {
472 BasicBlock *BB = I.getParent();
473 Assert1(BB, "Instruction not embedded in basic block!", &I);
475 if (!isa<PHINode>(I)) { // Check that non-phi nodes are not self referential
476 for (Value::use_iterator UI = I.use_begin(), UE = I.use_end();
478 Assert1(*UI != (User*)&I,
479 "Only PHI nodes may reference their own value!", &I);
482 // Check that void typed values don't have names
483 Assert1(I.getType() != Type::VoidTy || !I.hasName(),
484 "Instruction has a name, but provides a void value!", &I);
486 // Check that all uses of the instruction, if they are instructions
487 // themselves, actually have parent basic blocks. If the use is not an
488 // instruction, it is an error!
490 for (User::use_iterator UI = I.use_begin(), UE = I.use_end();
492 Assert1(isa<Instruction>(*UI), "Use of instruction is not an instruction!",
494 Instruction *Used = cast<Instruction>(*UI);
495 Assert2(Used->getParent() != 0, "Instruction referencing instruction not"
496 " embeded in a basic block!", &I, Used);
499 for (unsigned i = 0, e = I.getNumOperands(); i != e; ++i) {
500 // Check to make sure that the "address of" an intrinsic function is never
502 if (Function *F = dyn_cast<Function>(I.getOperand(i)))
503 Assert1(!F->isIntrinsic() || (i == 0 && isa<CallInst>(I)),
504 "Cannot take the address of an intrinsic!", &I);
506 else if (Instruction *Op = dyn_cast<Instruction>(I.getOperand(i))) {
507 // Check that a definition dominates all of its uses.
509 if (!isa<PHINode>(I)) {
510 // Definition must dominate use unless use is unreachable!
511 Assert2(DS->dominates(Op->getParent(), BB) ||
512 !DS->dominates(&BB->getParent()->getEntryBlock(), BB),
513 "Instruction does not dominate all uses!", Op, &I);
515 // PHI nodes are more difficult than other nodes because they actually
516 // "use" the value in the predecessor basic blocks they correspond to.
517 BasicBlock *PredBB = cast<BasicBlock>(I.getOperand(i+1));
518 Assert2(DS->dominates(Op->getParent(), PredBB) ||
519 !DS->dominates(&BB->getParent()->getEntryBlock(), PredBB),
520 "Instruction does not dominate all uses!", Op, &I);
526 /// visitIntrinsicFunction - Allow intrinsics to be verified in different ways.
527 void Verifier::visitIntrinsicFunctionCall(Intrinsic::ID ID, CallInst &CI) {
528 Function *IF = CI.getCalledFunction();
529 const FunctionType *FT = IF->getFunctionType();
530 Assert1(IF->isExternal(), "Intrinsic functions should never be defined!", IF);
531 unsigned NumArgs = 0;
533 // FIXME: this should check the return type of each intrinsic as well, also
536 case Intrinsic::va_start:
537 Assert1(CI.getParent()->getParent()->getFunctionType()->isVarArg(),
538 "llvm.va_start intrinsic may only occur in function with variable"
542 case Intrinsic::va_end: NumArgs = 1; break;
543 case Intrinsic::va_copy: NumArgs = 1; break;
545 case Intrinsic::setjmp: NumArgs = 1; break;
546 case Intrinsic::longjmp: NumArgs = 2; break;
547 case Intrinsic::sigsetjmp: NumArgs = 2; break;
548 case Intrinsic::siglongjmp: NumArgs = 2; break;
550 case Intrinsic::alpha_ctlz: NumArgs = 1; break;
551 case Intrinsic::alpha_cttz: NumArgs = 1; break;
552 case Intrinsic::alpha_ctpop: NumArgs = 1; break;
553 case Intrinsic::alpha_umulh: NumArgs = 2; break;
554 case Intrinsic::alpha_vecop: NumArgs = 4; break;
555 case Intrinsic::alpha_pup: NumArgs = 3; break;
556 case Intrinsic::alpha_bytezap: NumArgs = 2; break;
557 case Intrinsic::alpha_bytemanip: NumArgs = 3; break;
558 case Intrinsic::alpha_dfpbop: NumArgs = 3; break;
559 case Intrinsic::alpha_dfpuop: NumArgs = 2; break;
560 case Intrinsic::alpha_unordered: NumArgs = 2; break;
561 case Intrinsic::alpha_uqtodfp: NumArgs = 2; break;
562 case Intrinsic::alpha_uqtosfp: NumArgs = 2; break;
563 case Intrinsic::alpha_dfptosq: NumArgs = 2; break;
564 case Intrinsic::alpha_sfptosq: NumArgs = 2; break;
566 case Intrinsic::not_intrinsic:
567 assert(0 && "Invalid intrinsic!"); NumArgs = 0; break;
570 Assert1(FT->getNumParams() == NumArgs || (FT->getNumParams() < NumArgs &&
572 "Illegal # arguments for intrinsic function!", IF);
576 //===----------------------------------------------------------------------===//
577 // Implement the public interfaces to this file...
578 //===----------------------------------------------------------------------===//
580 FunctionPass *llvm::createVerifierPass() {
581 return new Verifier();
585 // verifyFunction - Create
586 bool llvm::verifyFunction(const Function &f) {
587 Function &F = (Function&)f;
588 assert(!F.isExternal() && "Cannot verify external functions");
591 DS.doInitialization(*F.getParent());
597 DS.doFinalization(*F.getParent());
602 // verifyModule - Check a module for errors, printing messages on stderr.
603 // Return true if the module is corrupt.
605 bool llvm::verifyModule(const Module &M) {
607 Verifier *V = new Verifier();