diff options
Diffstat (limited to 'contrib/llvm/lib/Analysis/Lint.cpp')
-rw-r--r-- | contrib/llvm/lib/Analysis/Lint.cpp | 560 |
1 files changed, 375 insertions, 185 deletions
diff --git a/contrib/llvm/lib/Analysis/Lint.cpp b/contrib/llvm/lib/Analysis/Lint.cpp index b5c7245..65a90d7 100644 --- a/contrib/llvm/lib/Analysis/Lint.cpp +++ b/contrib/llvm/lib/Analysis/Lint.cpp @@ -36,12 +36,14 @@ #include "llvm/Analysis/Lint.h" #include "llvm/ADT/STLExtras.h" +#include "llvm/ADT/SmallSet.h" #include "llvm/Analysis/AliasAnalysis.h" #include "llvm/Analysis/AssumptionCache.h" #include "llvm/Analysis/ConstantFolding.h" #include "llvm/Analysis/InstructionSimplify.h" #include "llvm/Analysis/Loads.h" #include "llvm/Analysis/Passes.h" +#include "llvm/Analysis/TargetLibraryInfo.h" #include "llvm/Analysis/ValueTracking.h" #include "llvm/IR/CallSite.h" #include "llvm/IR/DataLayout.h" @@ -49,19 +51,18 @@ #include "llvm/IR/Function.h" #include "llvm/IR/InstVisitor.h" #include "llvm/IR/IntrinsicInst.h" +#include "llvm/IR/LegacyPassManager.h" #include "llvm/Pass.h" -#include "llvm/PassManager.h" #include "llvm/Support/Debug.h" #include "llvm/Support/raw_ostream.h" -#include "llvm/Target/TargetLibraryInfo.h" using namespace llvm; namespace { namespace MemRef { - static unsigned Read = 1; - static unsigned Write = 2; - static unsigned Callee = 4; - static unsigned Branchee = 8; + static const unsigned Read = 1; + static const unsigned Write = 2; + static const unsigned Callee = 4; + static const unsigned Branchee = 8; } class Lint : public FunctionPass, public InstVisitor<Lint> { @@ -73,6 +74,8 @@ namespace { void visitMemoryReference(Instruction &I, Value *Ptr, uint64_t Size, unsigned Align, Type *Ty, unsigned Flags); + void visitEHBeginCatch(IntrinsicInst *II); + void visitEHEndCatch(IntrinsicInst *II); void visitCallInst(CallInst &I); void visitInvokeInst(InvokeInst &I); @@ -95,8 +98,8 @@ namespace { void visitInsertElementInst(InsertElementInst &I); void visitUnreachableInst(UnreachableInst &I); - Value *findValue(Value *V, bool OffsetOk) const; - Value *findValueImpl(Value *V, bool OffsetOk, + Value *findValue(Value *V, const DataLayout &DL, bool OffsetOk) const; + Value *findValueImpl(Value *V, const DataLayout &DL, bool OffsetOk, SmallPtrSetImpl<Value *> &Visited) const; public: @@ -104,7 +107,6 @@ namespace { AliasAnalysis *AA; AssumptionCache *AC; DominatorTree *DT; - const DataLayout *DL; TargetLibraryInfo *TLI; std::string Messages; @@ -121,32 +123,38 @@ namespace { AU.setPreservesAll(); AU.addRequired<AliasAnalysis>(); AU.addRequired<AssumptionCacheTracker>(); - AU.addRequired<TargetLibraryInfo>(); + AU.addRequired<TargetLibraryInfoWrapperPass>(); AU.addRequired<DominatorTreeWrapperPass>(); } void print(raw_ostream &O, const Module *M) const override {} - void WriteValue(const Value *V) { - if (!V) return; - if (isa<Instruction>(V)) { - MessagesStr << *V << '\n'; - } else { - V->printAsOperand(MessagesStr, true, Mod); - MessagesStr << '\n'; + void WriteValues(ArrayRef<const Value *> Vs) { + for (const Value *V : Vs) { + if (!V) + continue; + if (isa<Instruction>(V)) { + MessagesStr << *V << '\n'; + } else { + V->printAsOperand(MessagesStr, true, Mod); + MessagesStr << '\n'; + } } } - // CheckFailed - A check failed, so print out the condition and the message - // that failed. This provides a nice place to put a breakpoint if you want - // to see why something is not correct. - void CheckFailed(const Twine &Message, - const Value *V1 = nullptr, const Value *V2 = nullptr, - const Value *V3 = nullptr, const Value *V4 = nullptr) { - MessagesStr << Message.str() << "\n"; - WriteValue(V1); - WriteValue(V2); - WriteValue(V3); - WriteValue(V4); + /// \brief A check failed, so printout out the condition and the message. + /// + /// This provides a nice place to put a breakpoint if you want to see why + /// something is not correct. + void CheckFailed(const Twine &Message) { MessagesStr << Message << '\n'; } + + /// \brief A check failed (with values to print). + /// + /// This calls the Message-only version so that the above is easier to set + /// a breakpoint on. + template <typename T1, typename... Ts> + void CheckFailed(const Twine &Message, const T1 &V1, const Ts &...Vs) { + CheckFailed(Message); + WriteValues({V1, Vs...}); } }; } @@ -155,23 +163,15 @@ char Lint::ID = 0; INITIALIZE_PASS_BEGIN(Lint, "lint", "Statically lint-checks LLVM IR", false, true) INITIALIZE_PASS_DEPENDENCY(AssumptionCacheTracker) -INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfo) +INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfoWrapperPass) INITIALIZE_PASS_DEPENDENCY(DominatorTreeWrapperPass) INITIALIZE_AG_DEPENDENCY(AliasAnalysis) INITIALIZE_PASS_END(Lint, "lint", "Statically lint-checks LLVM IR", false, true) // Assert - We know that cond should be true, if not print an error message. -#define Assert(C, M) \ - do { if (!(C)) { CheckFailed(M); return; } } while (0) -#define Assert1(C, M, V1) \ - do { if (!(C)) { CheckFailed(M, V1); return; } } while (0) -#define Assert2(C, M, V1, V2) \ - do { if (!(C)) { CheckFailed(M, V1, V2); return; } } while (0) -#define Assert3(C, M, V1, V2, V3) \ - do { if (!(C)) { CheckFailed(M, V1, V2, V3); return; } } while (0) -#define Assert4(C, M, V1, V2, V3, V4) \ - do { if (!(C)) { CheckFailed(M, V1, V2, V3, V4); return; } } while (0) +#define Assert(C, ...) \ + do { if (!(C)) { CheckFailed(__VA_ARGS__); return; } } while (0) // Lint::run - This is the main Analysis entry point for a // function. @@ -181,9 +181,7 @@ bool Lint::runOnFunction(Function &F) { AA = &getAnalysis<AliasAnalysis>(); AC = &getAnalysis<AssumptionCacheTracker>().getAssumptionCache(F); DT = &getAnalysis<DominatorTreeWrapperPass>().getDomTree(); - DataLayoutPass *DLP = getAnalysisIfAvailable<DataLayoutPass>(); - DL = DLP ? &DLP->getDataLayout() : nullptr; - TLI = &getAnalysis<TargetLibraryInfo>(); + TLI = &getAnalysis<TargetLibraryInfoWrapperPass>().getTLI(); visit(F); dbgs() << MessagesStr.str(); Messages.clear(); @@ -193,8 +191,8 @@ bool Lint::runOnFunction(Function &F) { void Lint::visitFunction(Function &F) { // This isn't undefined behavior, it's just a little unusual, and it's a // fairly common mistake to neglect to name a function. - Assert1(F.hasName() || F.hasLocalLinkage(), - "Unusual: Unnamed function with non-local linkage", &F); + Assert(F.hasName() || F.hasLocalLinkage(), + "Unusual: Unnamed function with non-local linkage", &F); // TODO: Check for irreducible control flow. } @@ -202,27 +200,30 @@ void Lint::visitFunction(Function &F) { void Lint::visitCallSite(CallSite CS) { Instruction &I = *CS.getInstruction(); Value *Callee = CS.getCalledValue(); + const DataLayout &DL = CS->getModule()->getDataLayout(); visitMemoryReference(I, Callee, AliasAnalysis::UnknownSize, 0, nullptr, MemRef::Callee); - if (Function *F = dyn_cast<Function>(findValue(Callee, /*OffsetOk=*/false))) { - Assert1(CS.getCallingConv() == F->getCallingConv(), - "Undefined behavior: Caller and callee calling convention differ", - &I); + if (Function *F = dyn_cast<Function>(findValue(Callee, DL, + /*OffsetOk=*/false))) { + Assert(CS.getCallingConv() == F->getCallingConv(), + "Undefined behavior: Caller and callee calling convention differ", + &I); FunctionType *FT = F->getFunctionType(); unsigned NumActualArgs = CS.arg_size(); - Assert1(FT->isVarArg() ? - FT->getNumParams() <= NumActualArgs : - FT->getNumParams() == NumActualArgs, - "Undefined behavior: Call argument count mismatches callee " - "argument count", &I); + Assert(FT->isVarArg() ? FT->getNumParams() <= NumActualArgs + : FT->getNumParams() == NumActualArgs, + "Undefined behavior: Call argument count mismatches callee " + "argument count", + &I); - Assert1(FT->getReturnType() == I.getType(), - "Undefined behavior: Call return type mismatches " - "callee return type", &I); + Assert(FT->getReturnType() == I.getType(), + "Undefined behavior: Call return type mismatches " + "callee return type", + &I); // Check argument types (in case the callee was casted) and attributes. // TODO: Verify that caller and callee attributes are compatible. @@ -232,9 +233,10 @@ void Lint::visitCallSite(CallSite CS) { Value *Actual = *AI; if (PI != PE) { Argument *Formal = PI++; - Assert1(Formal->getType() == Actual->getType(), - "Undefined behavior: Call argument type mismatches " - "callee parameter type", &I); + Assert(Formal->getType() == Actual->getType(), + "Undefined behavior: Call argument type mismatches " + "callee parameter type", + &I); // Check that noalias arguments don't alias other arguments. This is // not fully precise because we don't know the sizes of the dereferenced @@ -243,9 +245,9 @@ void Lint::visitCallSite(CallSite CS) { for (CallSite::arg_iterator BI = CS.arg_begin(); BI != AE; ++BI) if (AI != BI && (*BI)->getType()->isPointerTy()) { AliasAnalysis::AliasResult Result = AA->alias(*AI, *BI); - Assert1(Result != AliasAnalysis::MustAlias && - Result != AliasAnalysis::PartialAlias, - "Unusual: noalias argument aliases another argument", &I); + Assert(Result != AliasAnalysis::MustAlias && + Result != AliasAnalysis::PartialAlias, + "Unusual: noalias argument aliases another argument", &I); } // Check that an sret argument points to valid memory. @@ -253,8 +255,8 @@ void Lint::visitCallSite(CallSite CS) { Type *Ty = cast<PointerType>(Formal->getType())->getElementType(); visitMemoryReference(I, Actual, AA->getTypeStoreSize(Ty), - DL ? DL->getABITypeAlignment(Ty) : 0, - Ty, MemRef::Read | MemRef::Write); + DL.getABITypeAlignment(Ty), Ty, + MemRef::Read | MemRef::Write); } } } @@ -263,10 +265,11 @@ void Lint::visitCallSite(CallSite CS) { if (CS.isCall() && cast<CallInst>(CS.getInstruction())->isTailCall()) for (CallSite::arg_iterator AI = CS.arg_begin(), AE = CS.arg_end(); AI != AE; ++AI) { - Value *Obj = findValue(*AI, /*OffsetOk=*/true); - Assert1(!isa<AllocaInst>(Obj), - "Undefined behavior: Call with \"tail\" keyword references " - "alloca", &I); + Value *Obj = findValue(*AI, DL, /*OffsetOk=*/true); + Assert(!isa<AllocaInst>(Obj), + "Undefined behavior: Call with \"tail\" keyword references " + "alloca", + &I); } @@ -291,13 +294,13 @@ void Lint::visitCallSite(CallSite CS) { // overlap is not distinguished from the case where nothing is known. uint64_t Size = 0; if (const ConstantInt *Len = - dyn_cast<ConstantInt>(findValue(MCI->getLength(), - /*OffsetOk=*/false))) + dyn_cast<ConstantInt>(findValue(MCI->getLength(), DL, + /*OffsetOk=*/false))) if (Len->getValue().isIntN(32)) Size = Len->getValue().getZExtValue(); - Assert1(AA->alias(MCI->getSource(), Size, MCI->getDest(), Size) != - AliasAnalysis::MustAlias, - "Undefined behavior: memcpy source and destination overlap", &I); + Assert(AA->alias(MCI->getSource(), Size, MCI->getDest(), Size) != + AliasAnalysis::MustAlias, + "Undefined behavior: memcpy source and destination overlap", &I); break; } case Intrinsic::memmove: { @@ -321,9 +324,9 @@ void Lint::visitCallSite(CallSite CS) { } case Intrinsic::vastart: - Assert1(I.getParent()->getParent()->isVarArg(), - "Undefined behavior: va_start called in a non-varargs function", - &I); + Assert(I.getParent()->getParent()->isVarArg(), + "Undefined behavior: va_start called in a non-varargs function", + &I); visitMemoryReference(I, CS.getArgument(0), AliasAnalysis::UnknownSize, 0, nullptr, MemRef::Read | MemRef::Write); @@ -346,6 +349,13 @@ void Lint::visitCallSite(CallSite CS) { visitMemoryReference(I, CS.getArgument(0), AliasAnalysis::UnknownSize, 0, nullptr, MemRef::Read | MemRef::Write); break; + + case Intrinsic::eh_begincatch: + visitEHBeginCatch(II); + break; + case Intrinsic::eh_endcatch: + visitEHEndCatch(II); + break; } } @@ -359,14 +369,13 @@ void Lint::visitInvokeInst(InvokeInst &I) { void Lint::visitReturnInst(ReturnInst &I) { Function *F = I.getParent()->getParent(); - Assert1(!F->doesNotReturn(), - "Unusual: Return statement in function with noreturn attribute", - &I); + Assert(!F->doesNotReturn(), + "Unusual: Return statement in function with noreturn attribute", &I); if (Value *V = I.getReturnValue()) { - Value *Obj = findValue(V, /*OffsetOk=*/true); - Assert1(!isa<AllocaInst>(Obj), - "Unusual: Returning alloca value", &I); + Value *Obj = + findValue(V, F->getParent()->getDataLayout(), /*OffsetOk=*/true); + Assert(!isa<AllocaInst>(Obj), "Unusual: Returning alloca value", &I); } } @@ -380,45 +389,47 @@ void Lint::visitMemoryReference(Instruction &I, if (Size == 0) return; - Value *UnderlyingObject = findValue(Ptr, /*OffsetOk=*/true); - Assert1(!isa<ConstantPointerNull>(UnderlyingObject), - "Undefined behavior: Null pointer dereference", &I); - Assert1(!isa<UndefValue>(UnderlyingObject), - "Undefined behavior: Undef pointer dereference", &I); - Assert1(!isa<ConstantInt>(UnderlyingObject) || - !cast<ConstantInt>(UnderlyingObject)->isAllOnesValue(), - "Unusual: All-ones pointer dereference", &I); - Assert1(!isa<ConstantInt>(UnderlyingObject) || - !cast<ConstantInt>(UnderlyingObject)->isOne(), - "Unusual: Address one pointer dereference", &I); + Value *UnderlyingObject = + findValue(Ptr, I.getModule()->getDataLayout(), /*OffsetOk=*/true); + Assert(!isa<ConstantPointerNull>(UnderlyingObject), + "Undefined behavior: Null pointer dereference", &I); + Assert(!isa<UndefValue>(UnderlyingObject), + "Undefined behavior: Undef pointer dereference", &I); + Assert(!isa<ConstantInt>(UnderlyingObject) || + !cast<ConstantInt>(UnderlyingObject)->isAllOnesValue(), + "Unusual: All-ones pointer dereference", &I); + Assert(!isa<ConstantInt>(UnderlyingObject) || + !cast<ConstantInt>(UnderlyingObject)->isOne(), + "Unusual: Address one pointer dereference", &I); if (Flags & MemRef::Write) { if (const GlobalVariable *GV = dyn_cast<GlobalVariable>(UnderlyingObject)) - Assert1(!GV->isConstant(), - "Undefined behavior: Write to read-only memory", &I); - Assert1(!isa<Function>(UnderlyingObject) && - !isa<BlockAddress>(UnderlyingObject), - "Undefined behavior: Write to text section", &I); + Assert(!GV->isConstant(), "Undefined behavior: Write to read-only memory", + &I); + Assert(!isa<Function>(UnderlyingObject) && + !isa<BlockAddress>(UnderlyingObject), + "Undefined behavior: Write to text section", &I); } if (Flags & MemRef::Read) { - Assert1(!isa<Function>(UnderlyingObject), - "Unusual: Load from function body", &I); - Assert1(!isa<BlockAddress>(UnderlyingObject), - "Undefined behavior: Load from block address", &I); + Assert(!isa<Function>(UnderlyingObject), "Unusual: Load from function body", + &I); + Assert(!isa<BlockAddress>(UnderlyingObject), + "Undefined behavior: Load from block address", &I); } if (Flags & MemRef::Callee) { - Assert1(!isa<BlockAddress>(UnderlyingObject), - "Undefined behavior: Call to block address", &I); + Assert(!isa<BlockAddress>(UnderlyingObject), + "Undefined behavior: Call to block address", &I); } if (Flags & MemRef::Branchee) { - Assert1(!isa<Constant>(UnderlyingObject) || - isa<BlockAddress>(UnderlyingObject), - "Undefined behavior: Branch to non-blockaddress", &I); + Assert(!isa<Constant>(UnderlyingObject) || + isa<BlockAddress>(UnderlyingObject), + "Undefined behavior: Branch to non-blockaddress", &I); } // Check for buffer overflows and misalignment. // Only handles memory references that read/write something simple like an // alloca instruction or a global variable. + auto &DL = I.getModule()->getDataLayout(); int64_t Offset = 0; if (Value *Base = GetPointerBaseWithConstantOffset(Ptr, Offset, DL)) { // OK, so the access is to a constant offset from Ptr. Check that Ptr is @@ -429,37 +440,37 @@ void Lint::visitMemoryReference(Instruction &I, if (AllocaInst *AI = dyn_cast<AllocaInst>(Base)) { Type *ATy = AI->getAllocatedType(); - if (DL && !AI->isArrayAllocation() && ATy->isSized()) - BaseSize = DL->getTypeAllocSize(ATy); + if (!AI->isArrayAllocation() && ATy->isSized()) + BaseSize = DL.getTypeAllocSize(ATy); BaseAlign = AI->getAlignment(); - if (DL && BaseAlign == 0 && ATy->isSized()) - BaseAlign = DL->getABITypeAlignment(ATy); + if (BaseAlign == 0 && ATy->isSized()) + BaseAlign = DL.getABITypeAlignment(ATy); } else if (GlobalVariable *GV = dyn_cast<GlobalVariable>(Base)) { // If the global may be defined differently in another compilation unit // then don't warn about funky memory accesses. if (GV->hasDefinitiveInitializer()) { Type *GTy = GV->getType()->getElementType(); - if (DL && GTy->isSized()) - BaseSize = DL->getTypeAllocSize(GTy); + if (GTy->isSized()) + BaseSize = DL.getTypeAllocSize(GTy); BaseAlign = GV->getAlignment(); - if (DL && BaseAlign == 0 && GTy->isSized()) - BaseAlign = DL->getABITypeAlignment(GTy); + if (BaseAlign == 0 && GTy->isSized()) + BaseAlign = DL.getABITypeAlignment(GTy); } } // Accesses from before the start or after the end of the object are not // defined. - Assert1(Size == AliasAnalysis::UnknownSize || - BaseSize == AliasAnalysis::UnknownSize || - (Offset >= 0 && Offset + Size <= BaseSize), - "Undefined behavior: Buffer overflow", &I); + Assert(Size == AliasAnalysis::UnknownSize || + BaseSize == AliasAnalysis::UnknownSize || + (Offset >= 0 && Offset + Size <= BaseSize), + "Undefined behavior: Buffer overflow", &I); // Accesses that say that the memory is more aligned than it is are not // defined. - if (DL && Align == 0 && Ty && Ty->isSized()) - Align = DL->getABITypeAlignment(Ty); - Assert1(!BaseAlign || Align <= MinAlign(BaseAlign, Offset), - "Undefined behavior: Memory reference address is misaligned", &I); + if (Align == 0 && Ty && Ty->isSized()) + Align = DL.getABITypeAlignment(Ty); + Assert(!BaseAlign || Align <= MinAlign(BaseAlign, Offset), + "Undefined behavior: Memory reference address is misaligned", &I); } } @@ -477,39 +488,219 @@ void Lint::visitStoreInst(StoreInst &I) { } void Lint::visitXor(BinaryOperator &I) { - Assert1(!isa<UndefValue>(I.getOperand(0)) || - !isa<UndefValue>(I.getOperand(1)), - "Undefined result: xor(undef, undef)", &I); + Assert(!isa<UndefValue>(I.getOperand(0)) || !isa<UndefValue>(I.getOperand(1)), + "Undefined result: xor(undef, undef)", &I); } void Lint::visitSub(BinaryOperator &I) { - Assert1(!isa<UndefValue>(I.getOperand(0)) || - !isa<UndefValue>(I.getOperand(1)), - "Undefined result: sub(undef, undef)", &I); + Assert(!isa<UndefValue>(I.getOperand(0)) || !isa<UndefValue>(I.getOperand(1)), + "Undefined result: sub(undef, undef)", &I); } void Lint::visitLShr(BinaryOperator &I) { - if (ConstantInt *CI = - dyn_cast<ConstantInt>(findValue(I.getOperand(1), /*OffsetOk=*/false))) - Assert1(CI->getValue().ult(cast<IntegerType>(I.getType())->getBitWidth()), - "Undefined result: Shift count out of range", &I); + if (ConstantInt *CI = dyn_cast<ConstantInt>( + findValue(I.getOperand(1), I.getModule()->getDataLayout(), + /*OffsetOk=*/false))) + Assert(CI->getValue().ult(cast<IntegerType>(I.getType())->getBitWidth()), + "Undefined result: Shift count out of range", &I); } void Lint::visitAShr(BinaryOperator &I) { - if (ConstantInt *CI = - dyn_cast<ConstantInt>(findValue(I.getOperand(1), /*OffsetOk=*/false))) - Assert1(CI->getValue().ult(cast<IntegerType>(I.getType())->getBitWidth()), - "Undefined result: Shift count out of range", &I); + if (ConstantInt *CI = dyn_cast<ConstantInt>(findValue( + I.getOperand(1), I.getModule()->getDataLayout(), /*OffsetOk=*/false))) + Assert(CI->getValue().ult(cast<IntegerType>(I.getType())->getBitWidth()), + "Undefined result: Shift count out of range", &I); } void Lint::visitShl(BinaryOperator &I) { - if (ConstantInt *CI = - dyn_cast<ConstantInt>(findValue(I.getOperand(1), /*OffsetOk=*/false))) - Assert1(CI->getValue().ult(cast<IntegerType>(I.getType())->getBitWidth()), - "Undefined result: Shift count out of range", &I); + if (ConstantInt *CI = dyn_cast<ConstantInt>(findValue( + I.getOperand(1), I.getModule()->getDataLayout(), /*OffsetOk=*/false))) + Assert(CI->getValue().ult(cast<IntegerType>(I.getType())->getBitWidth()), + "Undefined result: Shift count out of range", &I); +} + +static bool +allPredsCameFromLandingPad(BasicBlock *BB, + SmallSet<BasicBlock *, 4> &VisitedBlocks) { + VisitedBlocks.insert(BB); + if (BB->isLandingPad()) + return true; + // If we find a block with no predecessors, the search failed. + if (pred_empty(BB)) + return false; + for (BasicBlock *Pred : predecessors(BB)) { + if (VisitedBlocks.count(Pred)) + continue; + if (!allPredsCameFromLandingPad(Pred, VisitedBlocks)) + return false; + } + return true; +} + +static bool +allSuccessorsReachEndCatch(BasicBlock *BB, BasicBlock::iterator InstBegin, + IntrinsicInst **SecondBeginCatch, + SmallSet<BasicBlock *, 4> &VisitedBlocks) { + VisitedBlocks.insert(BB); + for (BasicBlock::iterator I = InstBegin, E = BB->end(); I != E; ++I) { + IntrinsicInst *IC = dyn_cast<IntrinsicInst>(I); + if (IC && IC->getIntrinsicID() == Intrinsic::eh_endcatch) + return true; + // If we find another begincatch while looking for an endcatch, + // that's also an error. + if (IC && IC->getIntrinsicID() == Intrinsic::eh_begincatch) { + *SecondBeginCatch = IC; + return false; + } + } + + // If we reach a block with no successors while searching, the + // search has failed. + if (succ_empty(BB)) + return false; + // Otherwise, search all of the successors. + for (BasicBlock *Succ : successors(BB)) { + if (VisitedBlocks.count(Succ)) + continue; + if (!allSuccessorsReachEndCatch(Succ, Succ->begin(), SecondBeginCatch, + VisitedBlocks)) + return false; + } + return true; +} + +void Lint::visitEHBeginCatch(IntrinsicInst *II) { + // The checks in this function make a potentially dubious assumption about + // the CFG, namely that any block involved in a catch is only used for the + // catch. This will very likely be true of IR generated by a front end, + // but it may cease to be true, for example, if the IR is run through a + // pass which combines similar blocks. + // + // In general, if we encounter a block the isn't dominated by the catch + // block while we are searching the catch block's successors for a call + // to end catch intrinsic, then it is possible that it will be legal for + // a path through this block to never reach a call to llvm.eh.endcatch. + // An analogous statement could be made about our search for a landing + // pad among the catch block's predecessors. + // + // What is actually required is that no path is possible at runtime that + // reaches a call to llvm.eh.begincatch without having previously visited + // a landingpad instruction and that no path is possible at runtime that + // calls llvm.eh.begincatch and does not subsequently call llvm.eh.endcatch + // (mentally adjusting for the fact that in reality these calls will be + // removed before code generation). + // + // Because this is a lint check, we take a pessimistic approach and warn if + // the control flow is potentially incorrect. + + SmallSet<BasicBlock *, 4> VisitedBlocks; + BasicBlock *CatchBB = II->getParent(); + + // The begin catch must occur in a landing pad block or all paths + // to it must have come from a landing pad. + Assert(allPredsCameFromLandingPad(CatchBB, VisitedBlocks), + "llvm.eh.begincatch may be reachable without passing a landingpad", + II); + + // Reset the visited block list. + VisitedBlocks.clear(); + + IntrinsicInst *SecondBeginCatch = nullptr; + + // This has to be called before it is asserted. Otherwise, the first assert + // below can never be hit. + bool EndCatchFound = allSuccessorsReachEndCatch( + CatchBB, std::next(static_cast<BasicBlock::iterator>(II)), + &SecondBeginCatch, VisitedBlocks); + Assert( + SecondBeginCatch == nullptr, + "llvm.eh.begincatch may be called a second time before llvm.eh.endcatch", + II, SecondBeginCatch); + Assert(EndCatchFound, + "Some paths from llvm.eh.begincatch may not reach llvm.eh.endcatch", + II); +} + +static bool allPredCameFromBeginCatch( + BasicBlock *BB, BasicBlock::reverse_iterator InstRbegin, + IntrinsicInst **SecondEndCatch, SmallSet<BasicBlock *, 4> &VisitedBlocks) { + VisitedBlocks.insert(BB); + // Look for a begincatch in this block. + for (BasicBlock::reverse_iterator RI = InstRbegin, RE = BB->rend(); RI != RE; + ++RI) { + IntrinsicInst *IC = dyn_cast<IntrinsicInst>(&*RI); + if (IC && IC->getIntrinsicID() == Intrinsic::eh_begincatch) + return true; + // If we find another end catch before we find a begin catch, that's + // an error. + if (IC && IC->getIntrinsicID() == Intrinsic::eh_endcatch) { + *SecondEndCatch = IC; + return false; + } + // If we encounter a landingpad instruction, the search failed. + if (isa<LandingPadInst>(*RI)) + return false; + } + // If while searching we find a block with no predeccesors, + // the search failed. + if (pred_empty(BB)) + return false; + // Search any predecessors we haven't seen before. + for (BasicBlock *Pred : predecessors(BB)) { + if (VisitedBlocks.count(Pred)) + continue; + if (!allPredCameFromBeginCatch(Pred, Pred->rbegin(), SecondEndCatch, + VisitedBlocks)) + return false; + } + return true; +} + +void Lint::visitEHEndCatch(IntrinsicInst *II) { + // The check in this function makes a potentially dubious assumption about + // the CFG, namely that any block involved in a catch is only used for the + // catch. This will very likely be true of IR generated by a front end, + // but it may cease to be true, for example, if the IR is run through a + // pass which combines similar blocks. + // + // In general, if we encounter a block the isn't post-dominated by the + // end catch block while we are searching the end catch block's predecessors + // for a call to the begin catch intrinsic, then it is possible that it will + // be legal for a path to reach the end catch block without ever having + // called llvm.eh.begincatch. + // + // What is actually required is that no path is possible at runtime that + // reaches a call to llvm.eh.endcatch without having previously visited + // a call to llvm.eh.begincatch (mentally adjusting for the fact that in + // reality these calls will be removed before code generation). + // + // Because this is a lint check, we take a pessimistic approach and warn if + // the control flow is potentially incorrect. + + BasicBlock *EndCatchBB = II->getParent(); + + // Alls paths to the end catch call must pass through a begin catch call. + + // If llvm.eh.begincatch wasn't called in the current block, we'll use this + // lambda to recursively look for it in predecessors. + SmallSet<BasicBlock *, 4> VisitedBlocks; + IntrinsicInst *SecondEndCatch = nullptr; + + // This has to be called before it is asserted. Otherwise, the first assert + // below can never be hit. + bool BeginCatchFound = + allPredCameFromBeginCatch(EndCatchBB, BasicBlock::reverse_iterator(II), + &SecondEndCatch, VisitedBlocks); + Assert( + SecondEndCatch == nullptr, + "llvm.eh.endcatch may be called a second time after llvm.eh.begincatch", + II, SecondEndCatch); + Assert(BeginCatchFound, + "llvm.eh.endcatch may be reachable without passing llvm.eh.begincatch", + II); } -static bool isZero(Value *V, const DataLayout *DL, DominatorTree *DT, +static bool isZero(Value *V, const DataLayout &DL, DominatorTree *DT, AssumptionCache *AC) { // Assume undef could be zero. if (isa<UndefValue>(V)) @@ -550,30 +741,30 @@ static bool isZero(Value *V, const DataLayout *DL, DominatorTree *DT, } void Lint::visitSDiv(BinaryOperator &I) { - Assert1(!isZero(I.getOperand(1), DL, DT, AC), - "Undefined behavior: Division by zero", &I); + Assert(!isZero(I.getOperand(1), I.getModule()->getDataLayout(), DT, AC), + "Undefined behavior: Division by zero", &I); } void Lint::visitUDiv(BinaryOperator &I) { - Assert1(!isZero(I.getOperand(1), DL, DT, AC), - "Undefined behavior: Division by zero", &I); + Assert(!isZero(I.getOperand(1), I.getModule()->getDataLayout(), DT, AC), + "Undefined behavior: Division by zero", &I); } void Lint::visitSRem(BinaryOperator &I) { - Assert1(!isZero(I.getOperand(1), DL, DT, AC), - "Undefined behavior: Division by zero", &I); + Assert(!isZero(I.getOperand(1), I.getModule()->getDataLayout(), DT, AC), + "Undefined behavior: Division by zero", &I); } void Lint::visitURem(BinaryOperator &I) { - Assert1(!isZero(I.getOperand(1), DL, DT, AC), - "Undefined behavior: Division by zero", &I); + Assert(!isZero(I.getOperand(1), I.getModule()->getDataLayout(), DT, AC), + "Undefined behavior: Division by zero", &I); } void Lint::visitAllocaInst(AllocaInst &I) { if (isa<ConstantInt>(I.getArraySize())) // This isn't undefined behavior, it's just an obvious pessimization. - Assert1(&I.getParent()->getParent()->getEntryBlock() == I.getParent(), - "Pessimization: Static alloca outside of entry block", &I); + Assert(&I.getParent()->getParent()->getEntryBlock() == I.getParent(), + "Pessimization: Static alloca outside of entry block", &I); // TODO: Check for an unusual size (MSB set?) } @@ -587,32 +778,33 @@ void Lint::visitIndirectBrInst(IndirectBrInst &I) { visitMemoryReference(I, I.getAddress(), AliasAnalysis::UnknownSize, 0, nullptr, MemRef::Branchee); - Assert1(I.getNumDestinations() != 0, - "Undefined behavior: indirectbr with no destinations", &I); + Assert(I.getNumDestinations() != 0, + "Undefined behavior: indirectbr with no destinations", &I); } void Lint::visitExtractElementInst(ExtractElementInst &I) { - if (ConstantInt *CI = - dyn_cast<ConstantInt>(findValue(I.getIndexOperand(), - /*OffsetOk=*/false))) - Assert1(CI->getValue().ult(I.getVectorOperandType()->getNumElements()), - "Undefined result: extractelement index out of range", &I); + if (ConstantInt *CI = dyn_cast<ConstantInt>( + findValue(I.getIndexOperand(), I.getModule()->getDataLayout(), + /*OffsetOk=*/false))) + Assert(CI->getValue().ult(I.getVectorOperandType()->getNumElements()), + "Undefined result: extractelement index out of range", &I); } void Lint::visitInsertElementInst(InsertElementInst &I) { - if (ConstantInt *CI = - dyn_cast<ConstantInt>(findValue(I.getOperand(2), - /*OffsetOk=*/false))) - Assert1(CI->getValue().ult(I.getType()->getNumElements()), - "Undefined result: insertelement index out of range", &I); + if (ConstantInt *CI = dyn_cast<ConstantInt>( + findValue(I.getOperand(2), I.getModule()->getDataLayout(), + /*OffsetOk=*/false))) + Assert(CI->getValue().ult(I.getType()->getNumElements()), + "Undefined result: insertelement index out of range", &I); } void Lint::visitUnreachableInst(UnreachableInst &I) { // This isn't undefined behavior, it's merely suspicious. - Assert1(&I == I.getParent()->begin() || - std::prev(BasicBlock::iterator(&I))->mayHaveSideEffects(), - "Unusual: unreachable immediately preceded by instruction without " - "side effects", &I); + Assert(&I == I.getParent()->begin() || + std::prev(BasicBlock::iterator(&I))->mayHaveSideEffects(), + "Unusual: unreachable immediately preceded by instruction without " + "side effects", + &I); } /// findValue - Look through bitcasts and simple memory reference patterns @@ -622,13 +814,13 @@ void Lint::visitUnreachableInst(UnreachableInst &I) { /// Most analysis passes don't require this logic, because instcombine /// will simplify most of these kinds of things away. But it's a goal of /// this Lint pass to be useful even on non-optimized IR. -Value *Lint::findValue(Value *V, bool OffsetOk) const { +Value *Lint::findValue(Value *V, const DataLayout &DL, bool OffsetOk) const { SmallPtrSet<Value *, 4> Visited; - return findValueImpl(V, OffsetOk, Visited); + return findValueImpl(V, DL, OffsetOk, Visited); } /// findValueImpl - Implementation helper for findValue. -Value *Lint::findValueImpl(Value *V, bool OffsetOk, +Value *Lint::findValueImpl(Value *V, const DataLayout &DL, bool OffsetOk, SmallPtrSetImpl<Value *> &Visited) const { // Detect self-referential values. if (!Visited.insert(V).second) @@ -649,7 +841,7 @@ Value *Lint::findValueImpl(Value *V, bool OffsetOk, break; if (Value *U = FindAvailableLoadedValue(L->getPointerOperand(), BB, BBI, 6, AA)) - return findValueImpl(U, OffsetOk, Visited); + return findValueImpl(U, DL, OffsetOk, Visited); if (BBI != BB->begin()) break; BB = BB->getUniquePredecessor(); if (!BB) break; @@ -658,40 +850,38 @@ Value *Lint::findValueImpl(Value *V, bool OffsetOk, } else if (PHINode *PN = dyn_cast<PHINode>(V)) { if (Value *W = PN->hasConstantValue()) if (W != V) - return findValueImpl(W, OffsetOk, Visited); + return findValueImpl(W, DL, OffsetOk, Visited); } else if (CastInst *CI = dyn_cast<CastInst>(V)) { if (CI->isNoopCast(DL)) - return findValueImpl(CI->getOperand(0), OffsetOk, Visited); + return findValueImpl(CI->getOperand(0), DL, OffsetOk, Visited); } else if (ExtractValueInst *Ex = dyn_cast<ExtractValueInst>(V)) { if (Value *W = FindInsertedValue(Ex->getAggregateOperand(), Ex->getIndices())) if (W != V) - return findValueImpl(W, OffsetOk, Visited); + return findValueImpl(W, DL, OffsetOk, Visited); } else if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V)) { // Same as above, but for ConstantExpr instead of Instruction. if (Instruction::isCast(CE->getOpcode())) { if (CastInst::isNoopCast(Instruction::CastOps(CE->getOpcode()), - CE->getOperand(0)->getType(), - CE->getType(), - DL ? DL->getIntPtrType(V->getType()) : - Type::getInt64Ty(V->getContext()))) - return findValueImpl(CE->getOperand(0), OffsetOk, Visited); + CE->getOperand(0)->getType(), CE->getType(), + DL.getIntPtrType(V->getType()))) + return findValueImpl(CE->getOperand(0), DL, OffsetOk, Visited); } else if (CE->getOpcode() == Instruction::ExtractValue) { ArrayRef<unsigned> Indices = CE->getIndices(); if (Value *W = FindInsertedValue(CE->getOperand(0), Indices)) if (W != V) - return findValueImpl(W, OffsetOk, Visited); + return findValueImpl(W, DL, OffsetOk, Visited); } } // As a last resort, try SimplifyInstruction or constant folding. if (Instruction *Inst = dyn_cast<Instruction>(V)) { if (Value *W = SimplifyInstruction(Inst, DL, TLI, DT, AC)) - return findValueImpl(W, OffsetOk, Visited); + return findValueImpl(W, DL, OffsetOk, Visited); } else if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V)) { if (Value *W = ConstantFoldConstantExpression(CE, DL, TLI)) if (W != V) - return findValueImpl(W, OffsetOk, Visited); + return findValueImpl(W, DL, OffsetOk, Visited); } return V; @@ -711,7 +901,7 @@ void llvm::lintFunction(const Function &f) { Function &F = const_cast<Function&>(f); assert(!F.isDeclaration() && "Cannot lint external functions"); - FunctionPassManager FPM(F.getParent()); + legacy::FunctionPassManager FPM(F.getParent()); Lint *V = new Lint(); FPM.add(V); FPM.run(F); @@ -720,7 +910,7 @@ void llvm::lintFunction(const Function &f) { /// lintModule - Check a module for errors, printing messages on stderr. /// void llvm::lintModule(const Module &M) { - PassManager PM; + legacy::PassManager PM; Lint *V = new Lint(); PM.add(V); PM.run(const_cast<Module&>(M)); |