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"
63 namespace { // Anonymous namespace for class
65 struct Verifier : public FunctionPass, InstVisitor<Verifier> {
66 bool Broken; // Is this module found to be broken?
67 bool RealPass; // Are we not being run by a PassManager?
68 bool AbortBroken; // If broken, should it or should it not abort?
69 Module *Mod; // Module we are verifying right now
70 DominatorSet *DS; // Dominator set, caution can be null!
72 Verifier() : Broken(false), RealPass(true), AbortBroken(true), DS(0) {}
73 Verifier(bool AB) : Broken(false), RealPass(true), AbortBroken(AB), DS(0) {}
74 Verifier(DominatorSet &ds)
75 : Broken(false), RealPass(false), AbortBroken(false), DS(&ds) {}
78 bool doInitialization(Module &M) {
80 verifySymbolTable(M.getSymbolTable());
82 // If this is a real pass, in a pass manager, we must abort before
83 // returning back to the pass manager, or else the pass manager may try to
84 // run other passes on the broken module.
91 bool runOnFunction(Function &F) {
92 // Get dominator information if we are being run by PassManager
93 if (RealPass) DS = &getAnalysis<DominatorSet>();
96 // If this is a real pass, in a pass manager, we must abort before
97 // returning back to the pass manager, or else the pass manager may try to
98 // run other passes on the broken module.
106 bool doFinalization(Module &M) {
107 // Scan through, checking all of the external function's linkage now...
108 for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I)
109 visitGlobalValue(*I);
111 for (Module::giterator I = M.gbegin(), E = M.gend(); I != E; ++I)
112 if (I->isExternal() && I->hasInternalLinkage())
113 CheckFailed("Global Variable is external with internal linkage!", I);
115 // If the module is broken, abort at this time.
120 virtual void getAnalysisUsage(AnalysisUsage &AU) const {
121 AU.setPreservesAll();
123 AU.addRequired<DominatorSet>();
126 // abortIfBroken - If the module is broken and we are supposed to abort on
127 // this condition, do so.
129 void abortIfBroken() const {
130 if (Broken && AbortBroken) {
131 std::cerr << "Broken module found, compilation aborted!\n";
137 // Verification methods...
138 void verifySymbolTable(SymbolTable &ST);
139 void visitGlobalValue(GlobalValue &GV);
140 void visitFunction(Function &F);
141 void visitBasicBlock(BasicBlock &BB);
142 void visitPHINode(PHINode &PN);
143 void visitBinaryOperator(BinaryOperator &B);
144 void visitShiftInst(ShiftInst &SI);
145 void visitVANextInst(VANextInst &VAN) { visitInstruction(VAN); }
146 void visitVAArgInst(VAArgInst &VAA) { visitInstruction(VAA); }
147 void visitCallInst(CallInst &CI);
148 void visitGetElementPtrInst(GetElementPtrInst &GEP);
149 void visitLoadInst(LoadInst &LI);
150 void visitStoreInst(StoreInst &SI);
151 void visitInstruction(Instruction &I);
152 void visitTerminatorInst(TerminatorInst &I);
153 void visitReturnInst(ReturnInst &RI);
154 void visitUserOp1(Instruction &I);
155 void visitUserOp2(Instruction &I) { visitUserOp1(I); }
156 void visitIntrinsicFunctionCall(Intrinsic::ID ID, CallInst &CI);
158 // CheckFailed - A check failed, so print out the condition and the message
159 // that failed. This provides a nice place to put a breakpoint if you want
160 // to see why something is not correct.
162 inline void CheckFailed(const std::string &Message,
163 const Value *V1 = 0, const Value *V2 = 0,
164 const Value *V3 = 0, const Value *V4 = 0) {
165 std::cerr << Message << "\n";
166 if (V1) { WriteAsOperand (std::cerr, V1, true, true, Mod); std::cerr << "\n"; }
167 if (V2) { WriteAsOperand (std::cerr, V2, true, true, Mod); std::cerr << "\n"; }
168 if (V3) { WriteAsOperand (std::cerr, V3, true, true, Mod); std::cerr << "\n"; }
169 if (V4) { WriteAsOperand (std::cerr, V4, true, true, Mod); std::cerr << "\n"; }
174 RegisterOpt<Verifier> X("verify", "Module Verifier");
176 // Assert - We know that cond should be true, if not print an error message.
177 #define Assert(C, M) \
178 do { if (!(C)) { CheckFailed(M); return; } } while (0)
179 #define Assert1(C, M, V1) \
180 do { if (!(C)) { CheckFailed(M, V1); return; } } while (0)
181 #define Assert2(C, M, V1, V2) \
182 do { if (!(C)) { CheckFailed(M, V1, V2); return; } } while (0)
183 #define Assert3(C, M, V1, V2, V3) \
184 do { if (!(C)) { CheckFailed(M, V1, V2, V3); return; } } while (0)
185 #define Assert4(C, M, V1, V2, V3, V4) \
186 do { if (!(C)) { CheckFailed(M, V1, V2, V3, V4); return; } } while (0)
189 void Verifier::visitGlobalValue(GlobalValue &GV) {
190 Assert1(!GV.isExternal() || GV.hasExternalLinkage(),
191 "Global value has Internal Linkage!", &GV);
192 Assert1(!GV.hasAppendingLinkage() || isa<GlobalVariable>(GV),
193 "Only global variables can have appending linkage!", &GV);
195 if (GV.hasAppendingLinkage()) {
196 GlobalVariable &GVar = cast<GlobalVariable>(GV);
197 Assert1(isa<ArrayType>(GVar.getType()->getElementType()),
198 "Only global arrays can have appending linkage!", &GV);
202 // verifySymbolTable - Verify that a function or module symbol table is ok
204 void Verifier::verifySymbolTable(SymbolTable &ST) {
205 // Loop over all of the types in the symbol table...
206 for (SymbolTable::iterator TI = ST.begin(), TE = ST.end(); TI != TE; ++TI)
207 for (SymbolTable::type_iterator I = TI->second.begin(),
208 E = TI->second.end(); I != E; ++I) {
209 Value *V = I->second;
211 // Check that there are no void typed values in the symbol table. Values
212 // with a void type cannot be put into symbol tables because they cannot
214 Assert1(V->getType() != Type::VoidTy,
215 "Values with void type are not allowed to have names!", V);
220 // visitFunction - Verify that a function is ok.
222 void Verifier::visitFunction(Function &F) {
223 // Check function arguments...
224 const FunctionType *FT = F.getFunctionType();
225 unsigned NumArgs = F.getArgumentList().size();
227 Assert2(FT->getNumParams() == NumArgs,
228 "# formal arguments must match # of arguments for function type!",
231 // Check that the argument values match the function type for this function...
233 for (Function::aiterator I = F.abegin(), E = F.aend(); I != E; ++I, ++i)
234 Assert2(I->getType() == FT->getParamType(i),
235 "Argument value does not match function argument type!",
236 I, FT->getParamType(i));
238 if (!F.isExternal()) {
239 verifySymbolTable(F.getSymbolTable());
241 // Check the entry node
242 BasicBlock *Entry = &F.getEntryBlock();
243 Assert1(pred_begin(Entry) == pred_end(Entry),
244 "Entry block to function must not have predecessors!", Entry);
249 // verifyBasicBlock - Verify that a basic block is well formed...
251 void Verifier::visitBasicBlock(BasicBlock &BB) {
252 // Check constraints that this basic block imposes on all of the PHI nodes in
254 if (isa<PHINode>(BB.front())) {
255 std::vector<BasicBlock*> Preds(pred_begin(&BB), pred_end(&BB));
256 std::sort(Preds.begin(), Preds.end());
258 for (BasicBlock::iterator I = BB.begin();
259 PHINode *PN = dyn_cast<PHINode>(I); ++I) {
261 // Ensure that PHI nodes have at least one entry!
262 Assert1(PN->getNumIncomingValues() != 0,
263 "PHI nodes must have at least one entry. If the block is dead, "
264 "the PHI should be removed!", PN);
265 Assert1(PN->getNumIncomingValues() >= Preds.size(),
266 "PHINode has more entries than the basic block has predecessors!",
268 Assert1(PN->getNumIncomingValues() <= Preds.size(),
269 "PHINode has less entries than the basic block has predecessors!",
272 // Get and sort all incoming values in the PHI node...
273 std::vector<std::pair<BasicBlock*, Value*> > Values;
274 Values.reserve(PN->getNumIncomingValues());
275 for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
276 Values.push_back(std::make_pair(PN->getIncomingBlock(i),
277 PN->getIncomingValue(i)));
278 std::sort(Values.begin(), Values.end());
280 for (unsigned i = 0, e = Values.size(); i != e; ++i) {
281 // Check to make sure that if there is more than one entry for a
282 // particular basic block in this PHI node, that the incoming values are
285 Assert4(i == 0 || Values[i].first != Values[i-1].first ||
286 Values[i].second == Values[i-1].second,
287 "PHI node has multiple entries for the same basic block with "
288 "different incoming values!", PN, Values[i].first,
289 Values[i].second, Values[i-1].second);
291 // Check to make sure that the predecessors and PHI node entries are
293 Assert3(Values[i].first == Preds[i],
294 "PHI node entries do not match predecessors!", PN,
295 Values[i].first, Preds[i]);
300 // Ensure that basic blocks have terminators!
301 Assert1(BB.getTerminator(), "Basic Block does not have terminator!", &BB);
304 void Verifier::visitTerminatorInst(TerminatorInst &I) {
305 // Ensure that terminators only exist at the end of the basic block.
306 Assert1(&I == I.getParent()->getTerminator(),
307 "Terminator found in the middle of a basic block!", I.getParent());
311 void Verifier::visitReturnInst(ReturnInst &RI) {
312 Function *F = RI.getParent()->getParent();
313 if (RI.getNumOperands() == 0)
314 Assert1(F->getReturnType() == Type::VoidTy,
315 "Function returns no value, but ret instruction found that does!",
318 Assert2(F->getReturnType() == RI.getOperand(0)->getType(),
319 "Function return type does not match operand "
320 "type of return inst!", &RI, F->getReturnType());
322 // Check to make sure that the return value has necessary properties for
324 visitTerminatorInst(RI);
327 // visitUserOp1 - User defined operators shouldn't live beyond the lifetime of a
328 // pass, if any exist, it's an error.
330 void Verifier::visitUserOp1(Instruction &I) {
331 Assert1(0, "User-defined operators should not live outside of a pass!",
335 // visitPHINode - Ensure that a PHI node is well formed.
336 void Verifier::visitPHINode(PHINode &PN) {
337 // Ensure that the PHI nodes are all grouped together at the top of the block.
338 // This can be tested by checking whether the instruction before this is
339 // either nonexistent (because this is begin()) or is a PHI node. If not,
340 // then there is some other instruction before a PHI.
341 Assert2(&PN.getParent()->front() == &PN || isa<PHINode>(PN.getPrev()),
342 "PHI nodes not grouped at top of basic block!",
343 &PN, PN.getParent());
345 // Check that all of the operands of the PHI node have the same type as the
347 for (unsigned i = 0, e = PN.getNumIncomingValues(); i != e; ++i)
348 Assert1(PN.getType() == PN.getIncomingValue(i)->getType(),
349 "PHI node operands are not the same type as the result!", &PN);
351 // All other PHI node constraints are checked in the visitBasicBlock method.
353 visitInstruction(PN);
356 void Verifier::visitCallInst(CallInst &CI) {
357 Assert1(isa<PointerType>(CI.getOperand(0)->getType()),
358 "Called function must be a pointer!", &CI);
359 const PointerType *FPTy = cast<PointerType>(CI.getOperand(0)->getType());
360 Assert1(isa<FunctionType>(FPTy->getElementType()),
361 "Called function is not pointer to function type!", &CI);
363 const FunctionType *FTy = cast<FunctionType>(FPTy->getElementType());
365 // Verify that the correct number of arguments are being passed
367 Assert1(CI.getNumOperands()-1 >= FTy->getNumParams(),
368 "Called function requires more parameters than were provided!",&CI);
370 Assert1(CI.getNumOperands()-1 == FTy->getNumParams(),
371 "Incorrect number of arguments passed to called function!", &CI);
373 // Verify that all arguments to the call match the function type...
374 for (unsigned i = 0, e = FTy->getNumParams(); i != e; ++i)
375 Assert2(CI.getOperand(i+1)->getType() == FTy->getParamType(i),
376 "Call parameter type does not match function signature!",
377 CI.getOperand(i+1), FTy->getParamType(i));
379 if (Function *F = CI.getCalledFunction())
380 if (Intrinsic::ID ID = (Intrinsic::ID)F->getIntrinsicID())
381 visitIntrinsicFunctionCall(ID, CI);
383 visitInstruction(CI);
386 // visitBinaryOperator - Check that both arguments to the binary operator are
389 void Verifier::visitBinaryOperator(BinaryOperator &B) {
390 Assert1(B.getOperand(0)->getType() == B.getOperand(1)->getType(),
391 "Both operands to a binary operator are not of the same type!", &B);
393 // Check that logical operators are only used with integral operands.
394 if (B.getOpcode() == Instruction::And || B.getOpcode() == Instruction::Or ||
395 B.getOpcode() == Instruction::Xor) {
396 Assert1(B.getType()->isIntegral(),
397 "Logical operators only work with integral types!", &B);
398 Assert1(B.getType() == B.getOperand(0)->getType(),
399 "Logical operators must have same type for operands and result!",
401 } else if (isa<SetCondInst>(B)) {
402 // Check that setcc instructions return bool
403 Assert1(B.getType() == Type::BoolTy,
404 "setcc instructions must return boolean values!", &B);
406 // Arithmetic operators only work on integer or fp values
407 Assert1(B.getType() == B.getOperand(0)->getType(),
408 "Arithmetic operators must have same type for operands and result!",
410 Assert1(B.getType()->isInteger() || B.getType()->isFloatingPoint(),
411 "Arithmetic operators must have integer or fp type!", &B);
417 void Verifier::visitShiftInst(ShiftInst &SI) {
418 Assert1(SI.getType()->isInteger(),
419 "Shift must return an integer result!", &SI);
420 Assert1(SI.getType() == SI.getOperand(0)->getType(),
421 "Shift return type must be same as first operand!", &SI);
422 Assert1(SI.getOperand(1)->getType() == Type::UByteTy,
423 "Second operand to shift must be ubyte type!", &SI);
424 visitInstruction(SI);
427 void Verifier::visitGetElementPtrInst(GetElementPtrInst &GEP) {
429 GetElementPtrInst::getIndexedType(GEP.getOperand(0)->getType(),
430 std::vector<Value*>(GEP.idx_begin(), GEP.idx_end()), true);
431 Assert1(ElTy, "Invalid indices for GEP pointer type!", &GEP);
432 Assert2(PointerType::get(ElTy) == GEP.getType(),
433 "GEP is not of right type for indices!", &GEP, ElTy);
434 visitInstruction(GEP);
437 void Verifier::visitLoadInst(LoadInst &LI) {
439 cast<PointerType>(LI.getOperand(0)->getType())->getElementType();
440 Assert2(ElTy == LI.getType(),
441 "Load result type does not match pointer operand type!", &LI, ElTy);
442 visitInstruction(LI);
445 void Verifier::visitStoreInst(StoreInst &SI) {
447 cast<PointerType>(SI.getOperand(1)->getType())->getElementType();
448 Assert2(ElTy == SI.getOperand(0)->getType(),
449 "Stored value type does not match pointer operand type!", &SI, ElTy);
450 visitInstruction(SI);
454 // verifyInstruction - Verify that an instruction is well formed.
456 void Verifier::visitInstruction(Instruction &I) {
457 BasicBlock *BB = I.getParent();
458 Assert1(BB, "Instruction not embedded in basic block!", &I);
460 if (!isa<PHINode>(I)) { // Check that non-phi nodes are not self referential
461 for (Value::use_iterator UI = I.use_begin(), UE = I.use_end();
463 Assert1(*UI != (User*)&I,
464 "Only PHI nodes may reference their own value!", &I);
467 // Check that void typed values don't have names
468 Assert1(I.getType() != Type::VoidTy || !I.hasName(),
469 "Instruction has a name, but provides a void value!", &I);
471 // Check that all uses of the instruction, if they are instructions
472 // themselves, actually have parent basic blocks. If the use is not an
473 // instruction, it is an error!
475 for (User::use_iterator UI = I.use_begin(), UE = I.use_end();
477 Assert1(isa<Instruction>(*UI), "Use of instruction is not an instruction!",
479 Instruction *Used = cast<Instruction>(*UI);
480 Assert2(Used->getParent() != 0, "Instruction referencing instruction not"
481 " embeded in a basic block!", &I, Used);
484 for (unsigned i = 0, e = I.getNumOperands(); i != e; ++i) {
485 // Check to make sure that the "address of" an intrinsic function is never
487 if (Function *F = dyn_cast<Function>(I.getOperand(i)))
488 Assert1(!F->isIntrinsic() || (i == 0 && isa<CallInst>(I)),
489 "Cannot take the address of an intrinsic!", &I);
491 else if (Instruction *Op = dyn_cast<Instruction>(I.getOperand(i))) {
492 // Check that a definition dominates all of its uses.
494 if (!isa<PHINode>(I)) {
495 // Definition must dominate use unless use is unreachable!
496 Assert2(DS->dominates(Op->getParent(), BB) ||
497 !DS->dominates(&BB->getParent()->getEntryBlock(), BB),
498 "Instruction does not dominate all uses!", Op, &I);
500 // PHI nodes are more difficult than other nodes because they actually
501 // "use" the value in the predecessor basic blocks they correspond to.
502 BasicBlock *PredBB = cast<BasicBlock>(I.getOperand(i+1));
503 Assert2(DS->dominates(Op->getParent(), PredBB) ||
504 !DS->dominates(&BB->getParent()->getEntryBlock(), PredBB),
505 "Instruction does not dominate all uses!", Op, &I);
511 /// visitIntrinsicFunction - Allow intrinsics to be verified in different ways.
512 void Verifier::visitIntrinsicFunctionCall(Intrinsic::ID ID, CallInst &CI) {
513 Function *IF = CI.getCalledFunction();
514 const FunctionType *FT = IF->getFunctionType();
515 Assert1(IF->isExternal(), "Intrinsic functions should never be defined!", IF);
516 unsigned NumArgs = 0;
518 // FIXME: this should check the return type of each intrinsic as well, also
521 case Intrinsic::va_start:
522 Assert1(CI.getParent()->getParent()->getFunctionType()->isVarArg(),
523 "llvm.va_start intrinsic may only occur in function with variable"
527 case Intrinsic::va_end: NumArgs = 1; break;
528 case Intrinsic::va_copy: NumArgs = 1; break;
530 case Intrinsic::setjmp: NumArgs = 1; break;
531 case Intrinsic::longjmp: NumArgs = 2; break;
532 case Intrinsic::sigsetjmp: NumArgs = 2; break;
533 case Intrinsic::siglongjmp: NumArgs = 2; break;
535 case Intrinsic::alpha_ctlz: NumArgs = 1; break;
536 case Intrinsic::alpha_cttz: NumArgs = 1; break;
537 case Intrinsic::alpha_ctpop: NumArgs = 1; break;
538 case Intrinsic::alpha_umulh: NumArgs = 2; break;
539 case Intrinsic::alpha_vecop: NumArgs = 4; break;
540 case Intrinsic::alpha_pup: NumArgs = 3; break;
541 case Intrinsic::alpha_bytezap: NumArgs = 2; break;
542 case Intrinsic::alpha_bytemanip: NumArgs = 3; break;
543 case Intrinsic::alpha_dfpbop: NumArgs = 3; break;
544 case Intrinsic::alpha_dfpuop: NumArgs = 2; break;
545 case Intrinsic::alpha_unordered: NumArgs = 2; break;
546 case Intrinsic::alpha_uqtodfp: NumArgs = 2; break;
547 case Intrinsic::alpha_uqtosfp: NumArgs = 2; break;
548 case Intrinsic::alpha_dfptosq: NumArgs = 2; break;
549 case Intrinsic::alpha_sfptosq: NumArgs = 2; break;
551 case Intrinsic::not_intrinsic:
552 assert(0 && "Invalid intrinsic!"); NumArgs = 0; break;
555 Assert1(FT->getNumParams() == NumArgs || (FT->getNumParams() < NumArgs &&
557 "Illegal # arguments for intrinsic function!", IF);
560 } // End anonymous namespace
562 //===----------------------------------------------------------------------===//
563 // Implement the public interfaces to this file...
564 //===----------------------------------------------------------------------===//
566 FunctionPass *createVerifierPass() {
567 return new Verifier();
571 // verifyFunction - Create
572 bool verifyFunction(const Function &f) {
573 Function &F = (Function&)f;
574 assert(!F.isExternal() && "Cannot verify external functions");
577 DS.doInitialization(*F.getParent());
583 DS.doFinalization(*F.getParent());
588 // verifyModule - Check a module for errors, printing messages on stderr.
589 // Return true if the module is corrupt.
591 bool verifyModule(const Module &M) {
593 Verifier *V = new Verifier();
599 } // End llvm namespace