1 //===-- Lint.cpp - Check for common errors in LLVM IR ---------------------===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This pass statically checks for common and easily-identified constructs
11 // which produce undefined or likely unintended behavior in LLVM IR.
13 // It is not a guarantee of correctness, in two ways. First, it isn't
14 // comprehensive. There are checks which could be done statically which are
15 // not yet implemented. Some of these are indicated by TODO comments, but
16 // those aren't comprehensive either. Second, many conditions cannot be
17 // checked statically. This pass does no dynamic instrumentation, so it
18 // can't check for all possible problems.
20 // Another limitation is that it assumes all code will be executed. A store
21 // through a null pointer in a basic block which is never reached is harmless,
22 // but this pass will warn about it anyway.
24 // Optimization passes may make conditions that this pass checks for more or
25 // less obvious. If an optimization pass appears to be introducing a warning,
26 // it may be that the optimization pass is merely exposing an existing
27 // condition in the code.
29 // This code may be run before instcombine. In many cases, instcombine checks
30 // for the same kinds of things and turns instructions with undefined behavior
31 // into unreachable (or equivalent). Because of this, this pass makes some
32 // effort to look through bitcasts and so on.
34 //===----------------------------------------------------------------------===//
36 #include "llvm/Analysis/Passes.h"
37 #include "llvm/Analysis/AliasAnalysis.h"
38 #include "llvm/Analysis/Lint.h"
39 #include "llvm/Analysis/ValueTracking.h"
40 #include "llvm/Assembly/Writer.h"
41 #include "llvm/Target/TargetData.h"
42 #include "llvm/Pass.h"
43 #include "llvm/PassManager.h"
44 #include "llvm/IntrinsicInst.h"
45 #include "llvm/Function.h"
46 #include "llvm/Support/CallSite.h"
47 #include "llvm/Support/Debug.h"
48 #include "llvm/Support/InstVisitor.h"
49 #include "llvm/Support/raw_ostream.h"
50 #include "llvm/ADT/STLExtras.h"
55 static unsigned Read = 1;
56 static unsigned Write = 2;
57 static unsigned Callee = 4;
58 static unsigned Branchee = 8;
61 class Lint : public FunctionPass, public InstVisitor<Lint> {
62 friend class InstVisitor<Lint>;
64 void visitFunction(Function &F);
66 void visitCallSite(CallSite CS);
67 void visitMemoryReference(Instruction &I, Value *Ptr, unsigned Align,
68 const Type *Ty, unsigned Flags);
70 void visitCallInst(CallInst &I);
71 void visitInvokeInst(InvokeInst &I);
72 void visitReturnInst(ReturnInst &I);
73 void visitLoadInst(LoadInst &I);
74 void visitStoreInst(StoreInst &I);
75 void visitXor(BinaryOperator &I);
76 void visitSub(BinaryOperator &I);
77 void visitLShr(BinaryOperator &I);
78 void visitAShr(BinaryOperator &I);
79 void visitShl(BinaryOperator &I);
80 void visitSDiv(BinaryOperator &I);
81 void visitUDiv(BinaryOperator &I);
82 void visitSRem(BinaryOperator &I);
83 void visitURem(BinaryOperator &I);
84 void visitAllocaInst(AllocaInst &I);
85 void visitVAArgInst(VAArgInst &I);
86 void visitIndirectBrInst(IndirectBrInst &I);
87 void visitExtractElementInst(ExtractElementInst &I);
88 void visitInsertElementInst(InsertElementInst &I);
89 void visitUnreachableInst(UnreachableInst &I);
97 raw_string_ostream MessagesStr;
99 static char ID; // Pass identification, replacement for typeid
100 Lint() : FunctionPass(&ID), MessagesStr(Messages) {}
102 virtual bool runOnFunction(Function &F);
104 virtual void getAnalysisUsage(AnalysisUsage &AU) const {
105 AU.setPreservesAll();
106 AU.addRequired<AliasAnalysis>();
108 virtual void print(raw_ostream &O, const Module *M) const {}
110 void WriteValue(const Value *V) {
112 if (isa<Instruction>(V)) {
113 MessagesStr << *V << '\n';
115 WriteAsOperand(MessagesStr, V, true, Mod);
120 void WriteType(const Type *T) {
123 WriteTypeSymbolic(MessagesStr, T, Mod);
126 // CheckFailed - A check failed, so print out the condition and the message
127 // that failed. This provides a nice place to put a breakpoint if you want
128 // to see why something is not correct.
129 void CheckFailed(const Twine &Message,
130 const Value *V1 = 0, const Value *V2 = 0,
131 const Value *V3 = 0, const Value *V4 = 0) {
132 MessagesStr << Message.str() << "\n";
139 void CheckFailed(const Twine &Message, const Value *V1,
140 const Type *T2, const Value *V3 = 0) {
141 MessagesStr << Message.str() << "\n";
147 void CheckFailed(const Twine &Message, const Type *T1,
148 const Type *T2 = 0, const Type *T3 = 0) {
149 MessagesStr << Message.str() << "\n";
158 static RegisterPass<Lint>
159 X("lint", "Statically lint-checks LLVM IR", false, true);
161 // Assert - We know that cond should be true, if not print an error message.
162 #define Assert(C, M) \
163 do { if (!(C)) { CheckFailed(M); return; } } while (0)
164 #define Assert1(C, M, V1) \
165 do { if (!(C)) { CheckFailed(M, V1); return; } } while (0)
166 #define Assert2(C, M, V1, V2) \
167 do { if (!(C)) { CheckFailed(M, V1, V2); return; } } while (0)
168 #define Assert3(C, M, V1, V2, V3) \
169 do { if (!(C)) { CheckFailed(M, V1, V2, V3); return; } } while (0)
170 #define Assert4(C, M, V1, V2, V3, V4) \
171 do { if (!(C)) { CheckFailed(M, V1, V2, V3, V4); return; } } while (0)
173 // Lint::run - This is the main Analysis entry point for a
176 bool Lint::runOnFunction(Function &F) {
178 AA = &getAnalysis<AliasAnalysis>();
179 TD = getAnalysisIfAvailable<TargetData>();
181 dbgs() << MessagesStr.str();
185 void Lint::visitFunction(Function &F) {
186 // This isn't undefined behavior, it's just a little unusual, and it's a
187 // fairly common mistake to neglect to name a function.
188 Assert1(F.hasName() || F.hasLocalLinkage(),
189 "Unusual: Unnamed function with non-local linkage", &F);
192 void Lint::visitCallSite(CallSite CS) {
193 Instruction &I = *CS.getInstruction();
194 Value *Callee = CS.getCalledValue();
196 // TODO: Check function alignment?
197 visitMemoryReference(I, Callee, 0, 0, MemRef::Callee);
199 if (Function *F = dyn_cast<Function>(Callee->stripPointerCasts())) {
200 Assert1(CS.getCallingConv() == F->getCallingConv(),
201 "Undefined behavior: Caller and callee calling convention differ",
204 const FunctionType *FT = F->getFunctionType();
205 unsigned NumActualArgs = unsigned(CS.arg_end()-CS.arg_begin());
207 Assert1(FT->isVarArg() ?
208 FT->getNumParams() <= NumActualArgs :
209 FT->getNumParams() == NumActualArgs,
210 "Undefined behavior: Call argument count mismatches callee "
211 "argument count", &I);
213 // TODO: Check argument types (in case the callee was casted)
215 // TODO: Check ABI-significant attributes.
217 // TODO: Check noalias attribute.
219 // TODO: Check sret attribute.
222 // TODO: Check the "tail" keyword constraints.
224 if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(&I))
225 switch (II->getIntrinsicID()) {
228 // TODO: Check more intrinsics
230 case Intrinsic::memcpy: {
231 MemCpyInst *MCI = cast<MemCpyInst>(&I);
232 visitMemoryReference(I, MCI->getSource(), MCI->getAlignment(), 0,
234 visitMemoryReference(I, MCI->getDest(), MCI->getAlignment(), 0,
237 // Check that the memcpy arguments don't overlap. The AliasAnalysis API
238 // isn't expressive enough for what we really want to do. Known partial
239 // overlap is not distinguished from the case where nothing is known.
241 if (const ConstantInt *Len =
242 dyn_cast<ConstantInt>(MCI->getLength()->stripPointerCasts()))
243 if (Len->getValue().isIntN(32))
244 Size = Len->getValue().getZExtValue();
245 Assert1(AA->alias(MCI->getSource(), Size, MCI->getDest(), Size) !=
246 AliasAnalysis::MustAlias,
247 "Undefined behavior: memcpy source and destination overlap", &I);
250 case Intrinsic::memmove: {
251 MemMoveInst *MMI = cast<MemMoveInst>(&I);
252 visitMemoryReference(I, MMI->getSource(), MMI->getAlignment(), 0,
254 visitMemoryReference(I, MMI->getDest(), MMI->getAlignment(), 0,
258 case Intrinsic::memset: {
259 MemSetInst *MSI = cast<MemSetInst>(&I);
260 visitMemoryReference(I, MSI->getDest(), MSI->getAlignment(), 0,
265 case Intrinsic::vastart:
266 Assert1(I.getParent()->getParent()->isVarArg(),
267 "Undefined behavior: va_start called in a non-varargs function",
270 visitMemoryReference(I, CS.getArgument(0), 0, 0,
271 MemRef::Read | MemRef::Write);
273 case Intrinsic::vacopy:
274 visitMemoryReference(I, CS.getArgument(0), 0, 0, MemRef::Write);
275 visitMemoryReference(I, CS.getArgument(1), 0, 0, MemRef::Read);
277 case Intrinsic::vaend:
278 visitMemoryReference(I, CS.getArgument(0), 0, 0,
279 MemRef::Read | MemRef::Write);
282 case Intrinsic::stackrestore:
283 visitMemoryReference(I, CS.getArgument(0), 0, 0,
289 void Lint::visitCallInst(CallInst &I) {
290 return visitCallSite(&I);
293 void Lint::visitInvokeInst(InvokeInst &I) {
294 return visitCallSite(&I);
297 void Lint::visitReturnInst(ReturnInst &I) {
298 Function *F = I.getParent()->getParent();
299 Assert1(!F->doesNotReturn(),
300 "Unusual: Return statement in function with noreturn attribute",
304 // TODO: Add a length argument and check that the reference is in bounds
305 void Lint::visitMemoryReference(Instruction &I,
306 Value *Ptr, unsigned Align, const Type *Ty,
308 Value *UnderlyingObject = Ptr->getUnderlyingObject();
309 Assert1(!isa<ConstantPointerNull>(UnderlyingObject),
310 "Undefined behavior: Null pointer dereference", &I);
311 Assert1(!isa<UndefValue>(UnderlyingObject),
312 "Undefined behavior: Undef pointer dereference", &I);
314 if (Flags & MemRef::Write) {
315 if (const GlobalVariable *GV = dyn_cast<GlobalVariable>(UnderlyingObject))
316 Assert1(!GV->isConstant(),
317 "Undefined behavior: Write to read-only memory", &I);
318 Assert1(!isa<Function>(UnderlyingObject) &&
319 !isa<BlockAddress>(UnderlyingObject),
320 "Undefined behavior: Write to text section", &I);
322 if (Flags & MemRef::Read) {
323 Assert1(!isa<Function>(UnderlyingObject),
324 "Unusual: Load from function body", &I);
325 Assert1(!isa<BlockAddress>(UnderlyingObject),
326 "Undefined behavior: Load from block address", &I);
328 if (Flags & MemRef::Callee) {
329 Assert1(!isa<BlockAddress>(UnderlyingObject),
330 "Undefined behavior: Call to block address", &I);
332 if (Flags & MemRef::Branchee) {
333 Assert1(!isa<Constant>(UnderlyingObject) ||
334 isa<BlockAddress>(UnderlyingObject),
335 "Undefined behavior: Branch to non-blockaddress", &I);
339 if (Align == 0 && Ty) Align = TD->getABITypeAlignment(Ty);
342 unsigned BitWidth = TD->getTypeSizeInBits(Ptr->getType());
343 APInt Mask = APInt::getAllOnesValue(BitWidth),
344 KnownZero(BitWidth, 0), KnownOne(BitWidth, 0);
345 ComputeMaskedBits(Ptr, Mask, KnownZero, KnownOne, TD);
346 Assert1(!(KnownOne & APInt::getLowBitsSet(BitWidth, Log2_32(Align))),
347 "Undefined behavior: Memory reference address is misaligned", &I);
352 void Lint::visitLoadInst(LoadInst &I) {
353 visitMemoryReference(I, I.getPointerOperand(), I.getAlignment(), I.getType(),
357 void Lint::visitStoreInst(StoreInst &I) {
358 visitMemoryReference(I, I.getPointerOperand(), I.getAlignment(),
359 I.getOperand(0)->getType(), MemRef::Write);
362 void Lint::visitXor(BinaryOperator &I) {
363 Assert1(!isa<UndefValue>(I.getOperand(0)) ||
364 !isa<UndefValue>(I.getOperand(1)),
365 "Undefined result: xor(undef, undef)", &I);
368 void Lint::visitSub(BinaryOperator &I) {
369 Assert1(!isa<UndefValue>(I.getOperand(0)) ||
370 !isa<UndefValue>(I.getOperand(1)),
371 "Undefined result: sub(undef, undef)", &I);
374 void Lint::visitLShr(BinaryOperator &I) {
375 if (ConstantInt *CI =
376 dyn_cast<ConstantInt>(I.getOperand(1)->stripPointerCasts()))
377 Assert1(CI->getValue().ult(cast<IntegerType>(I.getType())->getBitWidth()),
378 "Undefined result: Shift count out of range", &I);
381 void Lint::visitAShr(BinaryOperator &I) {
382 if (ConstantInt *CI =
383 dyn_cast<ConstantInt>(I.getOperand(1)->stripPointerCasts()))
384 Assert1(CI->getValue().ult(cast<IntegerType>(I.getType())->getBitWidth()),
385 "Undefined result: Shift count out of range", &I);
388 void Lint::visitShl(BinaryOperator &I) {
389 if (ConstantInt *CI =
390 dyn_cast<ConstantInt>(I.getOperand(1)->stripPointerCasts()))
391 Assert1(CI->getValue().ult(cast<IntegerType>(I.getType())->getBitWidth()),
392 "Undefined result: Shift count out of range", &I);
395 static bool isZero(Value *V, TargetData *TD) {
396 // Assume undef could be zero.
397 if (isa<UndefValue>(V)) return true;
399 unsigned BitWidth = cast<IntegerType>(V->getType())->getBitWidth();
400 APInt Mask = APInt::getAllOnesValue(BitWidth),
401 KnownZero(BitWidth, 0), KnownOne(BitWidth, 0);
402 ComputeMaskedBits(V, Mask, KnownZero, KnownOne, TD);
403 return KnownZero.isAllOnesValue();
406 void Lint::visitSDiv(BinaryOperator &I) {
407 Assert1(!isZero(I.getOperand(1), TD),
408 "Undefined behavior: Division by zero", &I);
411 void Lint::visitUDiv(BinaryOperator &I) {
412 Assert1(!isZero(I.getOperand(1), TD),
413 "Undefined behavior: Division by zero", &I);
416 void Lint::visitSRem(BinaryOperator &I) {
417 Assert1(!isZero(I.getOperand(1), TD),
418 "Undefined behavior: Division by zero", &I);
421 void Lint::visitURem(BinaryOperator &I) {
422 Assert1(!isZero(I.getOperand(1), TD),
423 "Undefined behavior: Division by zero", &I);
426 void Lint::visitAllocaInst(AllocaInst &I) {
427 if (isa<ConstantInt>(I.getArraySize()))
428 // This isn't undefined behavior, it's just an obvious pessimization.
429 Assert1(&I.getParent()->getParent()->getEntryBlock() == I.getParent(),
430 "Pessimization: Static alloca outside of entry block", &I);
433 void Lint::visitVAArgInst(VAArgInst &I) {
434 visitMemoryReference(I, I.getOperand(0), 0, 0,
435 MemRef::Read | MemRef::Write);
438 void Lint::visitIndirectBrInst(IndirectBrInst &I) {
439 visitMemoryReference(I, I.getAddress(), 0, 0, MemRef::Branchee);
442 void Lint::visitExtractElementInst(ExtractElementInst &I) {
443 if (ConstantInt *CI =
444 dyn_cast<ConstantInt>(I.getIndexOperand()->stripPointerCasts()))
445 Assert1(CI->getValue().ult(I.getVectorOperandType()->getNumElements()),
446 "Undefined result: extractelement index out of range", &I);
449 void Lint::visitInsertElementInst(InsertElementInst &I) {
450 if (ConstantInt *CI =
451 dyn_cast<ConstantInt>(I.getOperand(2)->stripPointerCasts()))
452 Assert1(CI->getValue().ult(I.getType()->getNumElements()),
453 "Undefined result: insertelement index out of range", &I);
456 void Lint::visitUnreachableInst(UnreachableInst &I) {
457 // This isn't undefined behavior, it's merely suspicious.
458 Assert1(&I == I.getParent()->begin() ||
459 prior(BasicBlock::iterator(&I))->mayHaveSideEffects(),
460 "Unusual: unreachable immediately preceded by instruction without "
464 //===----------------------------------------------------------------------===//
465 // Implement the public interfaces to this file...
466 //===----------------------------------------------------------------------===//
468 FunctionPass *llvm::createLintPass() {
472 /// lintFunction - Check a function for errors, printing messages on stderr.
474 void llvm::lintFunction(const Function &f) {
475 Function &F = const_cast<Function&>(f);
476 assert(!F.isDeclaration() && "Cannot lint external functions");
478 FunctionPassManager FPM(F.getParent());
479 Lint *V = new Lint();
484 /// lintModule - Check a module for errors, printing messages on stderr.
485 /// Return true if the module is corrupt.
487 void llvm::lintModule(const Module &M, std::string *ErrorInfo) {
489 Lint *V = new Lint();
491 PM.run(const_cast<Module&>(M));
494 *ErrorInfo = V->MessagesStr.str();