1 //===-- Verifier.cpp - Implement the Module Verifier -------------*- C++ -*-==//
3 // This file defines the function verifier interface, that can be used for some
4 // sanity checking of input to the system.
6 // Note that this does not provide full 'java style' security and verifications,
7 // instead it just tries to ensure that code is well formed.
9 // * Both of a binary operator's parameters are the same type
10 // * Verify that the indices of mem access instructions match other operands
11 // * Verify that arithmetic and other things are only performed on first class
12 // types. Verify that shifts & logicals only happen on integrals f.e.
13 // . All of the constants in a switch statement are of the correct type
14 // * The code is in valid SSA form
15 // . It should be illegal to put a label into any other type (like a structure)
16 // or to return one. [except constant arrays!]
17 // * Only phi nodes can be self referential: 'add int %0, %0 ; <int>:0' is bad
18 // * PHI nodes must have an entry for each predecessor, with no extras.
19 // * PHI nodes must be the first thing in a basic block, all grouped together
20 // * PHI nodes must have at least one entry
21 // * All basic blocks should only end with terminator insts, not contain them
22 // * The entry node to a function must not have predecessors
23 // * All Instructions must be embeded into a basic block
24 // . Function's cannot take a void typed parameter
25 // * Verify that a function's argument list agrees with it's declared type.
26 // * It is illegal to specify a name for a void value.
27 // * It is illegal to have a internal global value with no intitalizer
28 // * It is illegal to have a ret instruction that returns a value that does not
29 // agree with the function return value type.
30 // * Function call argument types match the function prototype
31 // * All other things that are tested by asserts spread about the code...
33 //===----------------------------------------------------------------------===//
35 #include "llvm/Analysis/Verifier.h"
36 #include "llvm/Pass.h"
37 #include "llvm/Module.h"
38 #include "llvm/DerivedTypes.h"
39 #include "llvm/iPHINode.h"
40 #include "llvm/iTerminators.h"
41 #include "llvm/iOther.h"
42 #include "llvm/iOperators.h"
43 #include "llvm/iMemory.h"
44 #include "llvm/SymbolTable.h"
45 #include "llvm/PassManager.h"
46 #include "llvm/Intrinsics.h"
47 #include "llvm/Analysis/Dominators.h"
48 #include "llvm/Support/CFG.h"
49 #include "llvm/Support/InstVisitor.h"
50 #include "Support/STLExtras.h"
53 namespace { // Anonymous namespace for class
55 struct Verifier : public FunctionPass, InstVisitor<Verifier> {
56 bool Broken; // Is this module found to be broken?
57 bool RealPass; // Are we not being run by a PassManager?
58 bool AbortBroken; // If broken, should it or should it not abort?
60 DominatorSet *DS; // Dominator set, caution can be null!
62 Verifier() : Broken(false), RealPass(true), AbortBroken(true), DS(0) {}
63 Verifier(bool AB) : Broken(false), RealPass(true), AbortBroken(AB), DS(0) {}
64 Verifier(DominatorSet &ds)
65 : Broken(false), RealPass(false), AbortBroken(false), DS(&ds) {}
68 bool doInitialization(Module &M) {
69 verifySymbolTable(M.getSymbolTable());
71 // If this is a real pass, in a pass manager, we must abort before
72 // returning back to the pass manager, or else the pass manager may try to
73 // run other passes on the broken module.
80 bool runOnFunction(Function &F) {
81 // Get dominator information if we are being run by PassManager
82 if (RealPass) DS = &getAnalysis<DominatorSet>();
85 // If this is a real pass, in a pass manager, we must abort before
86 // returning back to the pass manager, or else the pass manager may try to
87 // run other passes on the broken module.
95 bool doFinalization(Module &M) {
96 // Scan through, checking all of the external function's linkage now...
97 for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I)
100 for (Module::giterator I = M.gbegin(), E = M.gend(); I != E; ++I)
101 if (I->isExternal() && I->hasInternalLinkage())
102 CheckFailed("Global Variable is external with internal linkage!", I);
104 // If the module is broken, abort at this time.
109 virtual void getAnalysisUsage(AnalysisUsage &AU) const {
110 AU.setPreservesAll();
112 AU.addRequired<DominatorSet>();
115 // abortIfBroken - If the module is broken and we are supposed to abort on
116 // this condition, do so.
118 void abortIfBroken() const {
119 if (Broken && AbortBroken) {
120 std::cerr << "Broken module found, compilation aborted!\n";
126 // Verification methods...
127 void verifySymbolTable(SymbolTable &ST);
128 void visitGlobalValue(GlobalValue &GV);
129 void visitFunction(Function &F);
130 void visitBasicBlock(BasicBlock &BB);
131 void visitPHINode(PHINode &PN);
132 void visitBinaryOperator(BinaryOperator &B);
133 void visitShiftInst(ShiftInst &SI);
134 void visitVarArgInst(VarArgInst &VAI) { visitInstruction(VAI); }
135 void visitCallInst(CallInst &CI);
136 void visitGetElementPtrInst(GetElementPtrInst &GEP);
137 void visitLoadInst(LoadInst &LI);
138 void visitStoreInst(StoreInst &SI);
139 void visitInstruction(Instruction &I);
140 void visitTerminatorInst(TerminatorInst &I);
141 void visitReturnInst(ReturnInst &RI);
142 void visitUserOp1(Instruction &I);
143 void visitUserOp2(Instruction &I) { visitUserOp1(I); }
144 void visitIntrinsicFunctionCall(LLVMIntrinsic::ID ID, CallInst &CI);
146 // CheckFailed - A check failed, so print out the condition and the message
147 // that failed. This provides a nice place to put a breakpoint if you want
148 // to see why something is not correct.
150 inline void CheckFailed(const std::string &Message,
151 const Value *V1 = 0, const Value *V2 = 0,
152 const Value *V3 = 0, const Value *V4 = 0) {
153 std::cerr << Message << "\n";
154 if (V1) std::cerr << *V1 << "\n";
155 if (V2) std::cerr << *V2 << "\n";
156 if (V3) std::cerr << *V3 << "\n";
157 if (V4) std::cerr << *V4 << "\n";
162 RegisterPass<Verifier> X("verify", "Module Verifier");
165 // Assert - We know that cond should be true, if not print an error message.
166 #define Assert(C, M) \
167 do { if (!(C)) { CheckFailed(M); return; } } while (0)
168 #define Assert1(C, M, V1) \
169 do { if (!(C)) { CheckFailed(M, V1); return; } } while (0)
170 #define Assert2(C, M, V1, V2) \
171 do { if (!(C)) { CheckFailed(M, V1, V2); return; } } while (0)
172 #define Assert3(C, M, V1, V2, V3) \
173 do { if (!(C)) { CheckFailed(M, V1, V2, V3); return; } } while (0)
174 #define Assert4(C, M, V1, V2, V3, V4) \
175 do { if (!(C)) { CheckFailed(M, V1, V2, V3, V4); return; } } while (0)
178 void Verifier::visitGlobalValue(GlobalValue &GV) {
179 Assert1(!GV.isExternal() || GV.hasExternalLinkage(),
180 "Global value has Internal Linkage!", &GV);
181 Assert1(!GV.hasAppendingLinkage() || isa<GlobalVariable>(GV),
182 "Only global variables can have appending linkage!", &GV);
184 if (GV.hasAppendingLinkage()) {
185 GlobalVariable &GVar = cast<GlobalVariable>(GV);
186 Assert1(isa<ArrayType>(GVar.getType()->getElementType()),
187 "Only global arrays can have appending linkage!", &GV);
191 // verifySymbolTable - Verify that a function or module symbol table is ok
193 void Verifier::verifySymbolTable(SymbolTable &ST) {
194 // Loop over all of the types in the symbol table...
195 for (SymbolTable::iterator TI = ST.begin(), TE = ST.end(); TI != TE; ++TI)
196 for (SymbolTable::type_iterator I = TI->second.begin(),
197 E = TI->second.end(); I != E; ++I) {
198 Value *V = I->second;
200 // Check that there are no void typed values in the symbol table. Values
201 // with a void type cannot be put into symbol tables because they cannot
203 Assert1(V->getType() != Type::VoidTy,
204 "Values with void type are not allowed to have names!", V);
209 // visitFunction - Verify that a function is ok.
211 void Verifier::visitFunction(Function &F) {
212 // Check function arguments...
213 const FunctionType *FT = F.getFunctionType();
214 unsigned NumArgs = F.getArgumentList().size();
216 Assert2(FT->getNumParams() == NumArgs,
217 "# formal arguments must match # of arguments for function type!",
220 // Check that the argument values match the function type for this function...
222 for (Function::aiterator I = F.abegin(), E = F.aend(); I != E; ++I, ++i)
223 Assert2(I->getType() == FT->getParamType(i),
224 "Argument value does not match function argument type!",
225 I, FT->getParamType(i));
227 if (!F.isExternal()) {
228 verifySymbolTable(F.getSymbolTable());
230 // Check the entry node
231 BasicBlock *Entry = &F.getEntryNode();
232 Assert1(pred_begin(Entry) == pred_end(Entry),
233 "Entry block to function must not have predecessors!", Entry);
238 // verifyBasicBlock - Verify that a basic block is well formed...
240 void Verifier::visitBasicBlock(BasicBlock &BB) {
241 // Ensure that basic blocks have terminators!
242 Assert1(BB.getTerminator(), "Basic Block does not have terminator!", &BB);
245 void Verifier::visitTerminatorInst(TerminatorInst &I) {
246 // Ensure that terminators only exist at the end of the basic block.
247 Assert1(&I == I.getParent()->getTerminator(),
248 "Terminator found in the middle of a basic block!", I.getParent());
252 void Verifier::visitReturnInst(ReturnInst &RI) {
253 Function *F = RI.getParent()->getParent();
254 if (RI.getNumOperands() == 0)
255 Assert1(F->getReturnType() == Type::VoidTy,
256 "Function returns no value, but ret instruction found that does!",
259 Assert2(F->getReturnType() == RI.getOperand(0)->getType(),
260 "Function return type does not match operand "
261 "type of return inst!", &RI, F->getReturnType());
263 // Check to make sure that the return value has necessary properties for
265 visitTerminatorInst(RI);
268 // visitUserOp1 - User defined operators shouldn't live beyond the lifetime of a
269 // pass, if any exist, it's an error.
271 void Verifier::visitUserOp1(Instruction &I) {
272 Assert1(0, "User-defined operators should not live outside of a pass!",
276 // visitPHINode - Ensure that a PHI node is well formed.
277 void Verifier::visitPHINode(PHINode &PN) {
278 // Ensure that the PHI nodes are all grouped together at the top of the block.
279 // This can be tested by checking whether the instruction before this is
280 // either nonexistant (because this is begin()) or is a PHI node. If not,
281 // then there is some other instruction before a PHI.
282 Assert2(PN.getPrev() == 0 || isa<PHINode>(PN.getPrev()),
283 "PHI nodes not grouped at top of basic block!",
284 &PN, PN.getParent());
286 // Ensure that PHI nodes have at least one entry!
287 Assert1(PN.getNumIncomingValues() != 0,
288 "PHI nodes must have at least one entry. If the block is dead, "
289 "the PHI should be removed!",
292 std::vector<BasicBlock*> Preds(pred_begin(PN.getParent()),
293 pred_end(PN.getParent()));
294 // Loop over all of the incoming values, make sure that there are
295 // predecessors for each one...
297 for (unsigned i = 0, e = PN.getNumIncomingValues(); i != e; ++i) {
298 // Make sure all of the incoming values are the right types...
299 Assert2(PN.getType() == PN.getIncomingValue(i)->getType(),
300 "PHI node argument type does not agree with PHI node type!",
301 &PN, PN.getIncomingValue(i));
303 BasicBlock *BB = PN.getIncomingBlock(i);
304 std::vector<BasicBlock*>::iterator PI =
305 find(Preds.begin(), Preds.end(), BB);
306 Assert2(PI != Preds.end(), "PHI node has entry for basic block that"
307 " is not a predecessor!", &PN, BB);
311 // There should be no entries left in the predecessor list...
312 for (std::vector<BasicBlock*>::iterator I = Preds.begin(),
313 E = Preds.end(); I != E; ++I)
314 Assert2(0, "PHI node does not have entry for a predecessor basic block!",
317 // Now we go through and check to make sure that if there is more than one
318 // entry for a particular basic block in this PHI node, that the incoming
319 // values are all identical.
321 std::vector<std::pair<BasicBlock*, Value*> > Values;
322 Values.reserve(PN.getNumIncomingValues());
323 for (unsigned i = 0, e = PN.getNumIncomingValues(); i != e; ++i)
324 Values.push_back(std::make_pair(PN.getIncomingBlock(i),
325 PN.getIncomingValue(i)));
327 // Sort the Values vector so that identical basic block entries are adjacent.
328 std::sort(Values.begin(), Values.end());
330 // Check for identical basic blocks with differing incoming values...
331 for (unsigned i = 1, e = PN.getNumIncomingValues(); i < e; ++i)
332 Assert4(Values[i].first != Values[i-1].first ||
333 Values[i].second == Values[i-1].second,
334 "PHI node has multiple entries for the same basic block with "
335 "different incoming values!", &PN, Values[i].first,
336 Values[i].second, Values[i-1].second);
338 visitInstruction(PN);
341 void Verifier::visitCallInst(CallInst &CI) {
342 Assert1(isa<PointerType>(CI.getOperand(0)->getType()),
343 "Called function must be a pointer!", &CI);
344 const PointerType *FPTy = cast<PointerType>(CI.getOperand(0)->getType());
345 Assert1(isa<FunctionType>(FPTy->getElementType()),
346 "Called function is not pointer to function type!", &CI);
348 const FunctionType *FTy = cast<FunctionType>(FPTy->getElementType());
350 // Verify that the correct number of arguments are being passed
352 Assert1(CI.getNumOperands()-1 >= FTy->getNumParams(),
353 "Called function requires more parameters than were provided!",&CI);
355 Assert1(CI.getNumOperands()-1 == FTy->getNumParams(),
356 "Incorrect number of arguments passed to called function!", &CI);
358 // Verify that all arguments to the call match the function type...
359 for (unsigned i = 0, e = FTy->getNumParams(); i != e; ++i)
360 Assert2(CI.getOperand(i+1)->getType() == FTy->getParamType(i),
361 "Call parameter type does not match function signature!",
362 CI.getOperand(i+1), FTy->getParamType(i));
364 if (Function *F = CI.getCalledFunction())
365 if (LLVMIntrinsic::ID ID = (LLVMIntrinsic::ID)F->getIntrinsicID())
366 visitIntrinsicFunctionCall(ID, CI);
368 visitInstruction(CI);
371 // visitBinaryOperator - Check that both arguments to the binary operator are
374 void Verifier::visitBinaryOperator(BinaryOperator &B) {
375 Assert1(B.getOperand(0)->getType() == B.getOperand(1)->getType(),
376 "Both operands to a binary operator are not of the same type!", &B);
378 // Check that logical operators are only used with integral operands.
379 if (B.getOpcode() == Instruction::And || B.getOpcode() == Instruction::Or ||
380 B.getOpcode() == Instruction::Xor) {
381 Assert1(B.getType()->isIntegral(),
382 "Logical operators only work with integral types!", &B);
383 Assert1(B.getType() == B.getOperand(0)->getType(),
384 "Logical operators must have same type for operands and result!",
386 } else if (isa<SetCondInst>(B)) {
387 // Check that setcc instructions return bool
388 Assert1(B.getType() == Type::BoolTy,
389 "setcc instructions must return boolean values!", &B);
391 // Arithmetic operators only work on integer or fp values
392 Assert1(B.getType() == B.getOperand(0)->getType(),
393 "Arithmetic operators must have same type for operands and result!",
395 Assert1(B.getType()->isInteger() || B.getType()->isFloatingPoint(),
396 "Arithmetic operators must have integer or fp type!", &B);
402 void Verifier::visitShiftInst(ShiftInst &SI) {
403 Assert1(SI.getType()->isInteger(),
404 "Shift must return an integer result!", &SI);
405 Assert1(SI.getType() == SI.getOperand(0)->getType(),
406 "Shift return type must be same as first operand!", &SI);
407 Assert1(SI.getOperand(1)->getType() == Type::UByteTy,
408 "Second operand to shift must be ubyte type!", &SI);
409 visitInstruction(SI);
412 void Verifier::visitGetElementPtrInst(GetElementPtrInst &GEP) {
414 GetElementPtrInst::getIndexedType(GEP.getOperand(0)->getType(),
415 std::vector<Value*>(GEP.idx_begin(), GEP.idx_end()), true);
416 Assert1(ElTy, "Invalid indices for GEP pointer type!", &GEP);
417 Assert2(PointerType::get(ElTy) == GEP.getType(),
418 "GEP is not of right type for indices!", &GEP, ElTy);
419 visitInstruction(GEP);
422 void Verifier::visitLoadInst(LoadInst &LI) {
424 cast<PointerType>(LI.getOperand(0)->getType())->getElementType();
425 Assert2(ElTy == LI.getType(),
426 "Load is not of right type for indices!", &LI, ElTy);
427 visitInstruction(LI);
430 void Verifier::visitStoreInst(StoreInst &SI) {
432 cast<PointerType>(SI.getOperand(1)->getType())->getElementType();
433 Assert2(ElTy == SI.getOperand(0)->getType(),
434 "Stored value is not of right type for indices!", &SI, ElTy);
435 visitInstruction(SI);
439 // verifyInstruction - Verify that an instruction is well formed.
441 void Verifier::visitInstruction(Instruction &I) {
442 BasicBlock *BB = I.getParent();
443 Assert1(BB, "Instruction not embedded in basic block!", &I);
445 // Check that all uses of the instruction, if they are instructions
446 // themselves, actually have parent basic blocks. If the use is not an
447 // instruction, it is an error!
449 for (User::use_iterator UI = I.use_begin(), UE = I.use_end();
451 Assert1(isa<Instruction>(*UI), "Use of instruction is not an instruction!",
453 Instruction *Used = cast<Instruction>(*UI);
454 Assert2(Used->getParent() != 0, "Instruction referencing instruction not"
455 " embeded in a basic block!", &I, Used);
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 a definition dominates all of its uses.
471 for (User::use_iterator UI = I.use_begin(), UE = I.use_end();
473 Instruction *Use = cast<Instruction>(*UI);
475 // PHI nodes are more difficult than other nodes because they actually
476 // "use" the value in the predecessor basic blocks they correspond to.
477 if (PHINode *PN = dyn_cast<PHINode>(Use)) {
478 for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
479 if (&I == PN->getIncomingValue(i)) {
480 // Make sure that I dominates the end of pred(i)
481 BasicBlock *Pred = PN->getIncomingBlock(i);
483 // Use must be dominated by by definition unless use is unreachable!
484 Assert2(DS->dominates(BB, Pred) ||
485 !DS->dominates(&BB->getParent()->getEntryNode(), Pred),
486 "Instruction does not dominate all uses!",
491 // Use must be dominated by by definition unless use is unreachable!
492 Assert2(DS->dominates(&I, Use) ||
493 !DS->dominates(&BB->getParent()->getEntryNode(),Use->getParent()),
494 "Instruction does not dominate all uses!", &I, Use);
498 // Check to make sure that the "address of" an intrinsic function is never
500 for (unsigned i = 0, e = I.getNumOperands(); i != e; ++i)
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);
506 /// visitIntrinsicFunction - Allow intrinsics to be verified in different ways.
507 void Verifier::visitIntrinsicFunctionCall(LLVMIntrinsic::ID ID, CallInst &CI) {
508 Function *IF = CI.getCalledFunction();
509 const FunctionType *FT = IF->getFunctionType();
510 Assert1(IF->isExternal(), "Intrinsic functions should never be defined!", IF);
511 unsigned NumArgs = 0;
513 // FIXME: this should check the return type of each intrinsic as well, also
516 case LLVMIntrinsic::va_start:
517 Assert1(CI.getParent()->getParent()->getFunctionType()->isVarArg(),
518 "llvm.va_start intrinsic may only occur in function with variable"
522 case LLVMIntrinsic::va_end: NumArgs = 1; break;
523 case LLVMIntrinsic::va_copy: NumArgs = 2; break;
525 case LLVMIntrinsic::unwind: NumArgs = 0; break;
527 case LLVMIntrinsic::setjmp: NumArgs = 1; break;
528 case LLVMIntrinsic::longjmp: NumArgs = 2; break;
529 case LLVMIntrinsic::sigsetjmp: NumArgs = 2; break;
530 case LLVMIntrinsic::siglongjmp: NumArgs = 2; break;
532 case LLVMIntrinsic::alpha_ctlz: NumArgs = 1; break;
533 case LLVMIntrinsic::alpha_cttz: NumArgs = 1; break;
534 case LLVMIntrinsic::alpha_ctpop: NumArgs = 1; break;
535 case LLVMIntrinsic::alpha_umulh: NumArgs = 2; break;
536 case LLVMIntrinsic::alpha_vecop: NumArgs = 4; break;
537 case LLVMIntrinsic::alpha_pup: NumArgs = 3; break;
538 case LLVMIntrinsic::alpha_bytezap: NumArgs = 2; break;
539 case LLVMIntrinsic::alpha_bytemanip: NumArgs = 3; break;
540 case LLVMIntrinsic::alpha_dfpbop: NumArgs = 3; break;
541 case LLVMIntrinsic::alpha_dfpuop: NumArgs = 2; break;
542 case LLVMIntrinsic::alpha_unordered: NumArgs = 2; break;
543 case LLVMIntrinsic::alpha_uqtodfp: NumArgs = 2; break;
544 case LLVMIntrinsic::alpha_uqtosfp: NumArgs = 2; break;
545 case LLVMIntrinsic::alpha_dfptosq: NumArgs = 2; break;
546 case LLVMIntrinsic::alpha_sfptosq: NumArgs = 2; break;
548 case LLVMIntrinsic::not_intrinsic:
549 assert(0 && "Invalid intrinsic!"); NumArgs = 0; break;
552 Assert1(FT->getNumParams() == NumArgs || (FT->getNumParams() < NumArgs &&
554 "Illegal # arguments for intrinsic function!", IF);
558 //===----------------------------------------------------------------------===//
559 // Implement the public interfaces to this file...
560 //===----------------------------------------------------------------------===//
562 Pass *createVerifierPass() {
563 return new Verifier();
567 // verifyFunction - Create
568 bool verifyFunction(const Function &f) {
569 Function &F = (Function&)f;
570 assert(!F.isExternal() && "Cannot verify external functions");
573 DS.doInitialization(*F.getParent());
579 DS.doFinalization(*F.getParent());
584 // verifyModule - Check a module for errors, printing messages on stderr.
585 // Return true if the module is corrupt.
587 bool verifyModule(const Module &M) {
589 Verifier *V = new Verifier();