Add a -lint pass which checks for common sources of undefined or likely

unintended behavior.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@100798 91177308-0d34-0410-b5e6-96231b3b80d8

4 files changed, 453 insertions, 0 deletions

diff --git a/include/llvm/Analysis/Lint.h b/include/llvm/Analysis/Lint.h
new file mode 100644
index 0000000..2f01366
--- /dev/null
+++ b/include/llvm/Analysis/Lint.h
@@ -0,0 +1,52 @@
+//===-- llvm/Analysis/Lint.h - LLVM IR Lint ---------------------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file defines lint interfaces that can be used for some sanity checking
+// of input to the system, and for checking that transformations
+// haven't done something bad. In contrast to the Verifier, the Lint checker
+// checks for undefined behavior or constructions with likely unintended
+// behavior.
+//
+// To see what specifically is checked, look at Lint.cpp
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_ANALYSIS_LINT_H
+#define LLVM_ANALYSIS_LINT_H
+
+#include <string>
+
+namespace llvm {
+
+class FunctionPass;
+class Module;
+class Function;
+
+/// @brief Create a lint pass.
+///
+/// Check a module or function.
+FunctionPass *createLintPass();
+
+/// @brief Check a module.
+///
+/// This should only be used for debugging, because it plays games with
+/// PassManagers and stuff.
+void lintModule(
+  const Module &M,  ///< The module to be checked
+  std::string *ErrorInfo = 0      ///< Information about failures.
+);
+
+// lintFunction - Check a function.
+void lintFunction(
+  const Function &F  ///< The function to be checked
+);
+
+} // End llvm namespace
+
+#endif
diff --git a/include/llvm/LinkAllPasses.h b/include/llvm/LinkAllPasses.h
index ae53851..f6ba39c 100644
--- a/include/llvm/LinkAllPasses.h
+++ b/include/llvm/LinkAllPasses.h
@@ -23,6 +23,7 @@
 #include "llvm/Analysis/PointerTracking.h"
 #include "llvm/Analysis/PostDominators.h"
 #include "llvm/Analysis/ScalarEvolution.h"
+#include "llvm/Analysis/Lint.h"
 #include "llvm/Assembly/PrintModulePass.h"
 #include "llvm/CodeGen/Passes.h"
 #include "llvm/Function.h"
@@ -137,6 +138,7 @@ namespace {
       (void) llvm::createGEPSplitterPass();
       (void) llvm::createSCCVNPass();
       (void) llvm::createABCDPass();
+      (void) llvm::createLintPass();
 
       (void)new llvm::IntervalPartition();
       (void)new llvm::FindUsedTypes();
diff --git a/lib/Analysis/Lint.cpp b/lib/Analysis/Lint.cpp
new file mode 100644
index 0000000..1994517
--- /dev/null
+++ b/lib/Analysis/Lint.cpp
@@ -0,0 +1,368 @@
+//===-- Lint.cpp - Check for common errors in LLVM IR ---------------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This pass statically checks for common and easily-identified constructs
+// which produce undefined or likely unintended behavior in LLVM IR.
+//
+// It is not a guarantee of correctness, in two ways. First, it isn't
+// comprehensive. There are checks which could be done statically which are
+// not yet implemented. Some of these are indicated by TODO comments, but
+// those aren't comprehensive either. Second, many conditions cannot be
+// checked statically. This pass does no dynamic instrumentation, so it
+// can't check for all possible problems.
+// 
+// Another limitation is that it assumes all code will be executed. A store
+// through a null pointer in a basic block which is never reached is harmless,
+// but this pass will warn about it anyway.
+
+// Optimization passes may make conditions that this pass checks for more or
+// less obvious. If an optimization pass appears to be introducing a warning,
+// it may be that the optimization pass is merely exposing an existing
+// condition in the code.
+// 
+// This code may be run before instcombine. In many cases, instcombine checks
+// for the same kinds of things and turns instructions with undefined behavior
+// into unreachable (or equivalent). Because of this, this pass makes some
+// effort to look through bitcasts and so on.
+// 
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Analysis/Passes.h"
+#include "llvm/Analysis/AliasAnalysis.h"
+#include "llvm/Analysis/Lint.h"
+#include "llvm/Analysis/ValueTracking.h"
+#include "llvm/Assembly/Writer.h"
+#include "llvm/Target/TargetData.h"
+#include "llvm/Pass.h"
+#include "llvm/PassManager.h"
+#include "llvm/IntrinsicInst.h"
+#include "llvm/Function.h"
+#include "llvm/Support/CallSite.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/InstVisitor.h"
+#include "llvm/Support/raw_ostream.h"
+using namespace llvm;
+
+namespace {
+  class Lint : public FunctionPass, public InstVisitor<Lint> {
+    friend class InstVisitor<Lint>;
+
+    void visitCallSite(CallSite CS);
+    void visitMemoryReference(Instruction &I, Value *Ptr, unsigned Align,
+                              const Type *Ty);
+
+    void visitInstruction(Instruction &I);
+    void visitCallInst(CallInst &I);
+    void visitInvokeInst(InvokeInst &I);
+    void visitReturnInst(ReturnInst &I);
+    void visitLoadInst(LoadInst &I);
+    void visitStoreInst(StoreInst &I);
+    void visitSDiv(BinaryOperator &I);
+    void visitUDiv(BinaryOperator &I);
+    void visitSRem(BinaryOperator &I);
+    void visitURem(BinaryOperator &I);
+    void visitAllocaInst(AllocaInst &I);
+    void visitVAArgInst(VAArgInst &I);
+    void visitIndirectBrInst(IndirectBrInst &I);
+
+  public:
+    Module *Mod;
+    AliasAnalysis *AA;
+    TargetData *TD;
+
+    std::string Messages;
+    raw_string_ostream MessagesStr;
+
+    static char ID; // Pass identification, replacement for typeid
+    Lint() : FunctionPass(&ID), MessagesStr(Messages) {}
+
+    virtual bool runOnFunction(Function &F);
+
+    virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+      AU.setPreservesAll();
+      AU.addRequired<AliasAnalysis>();
+    }
+    virtual void print(raw_ostream &O, const Module *M) const {}
+
+    void WriteValue(const Value *V) {
+      if (!V) return;
+      if (isa<Instruction>(V)) {
+        MessagesStr << *V << '\n';
+      } else {
+        WriteAsOperand(MessagesStr, V, true, Mod);
+        MessagesStr << '\n';
+      }
+    }
+
+    void WriteType(const Type *T) {
+      if (!T) return;
+      MessagesStr << ' ';
+      WriteTypeSymbolic(MessagesStr, T, Mod);
+    }
+
+    // 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 = 0, const Value *V2 = 0,
+                     const Value *V3 = 0, const Value *V4 = 0) {
+      MessagesStr << Message.str() << "\n";
+      WriteValue(V1);
+      WriteValue(V2);
+      WriteValue(V3);
+      WriteValue(V4);
+    }
+
+    void CheckFailed(const Twine &Message, const Value *V1,
+                     const Type *T2, const Value *V3 = 0) {
+      MessagesStr << Message.str() << "\n";
+      WriteValue(V1);
+      WriteType(T2);
+      WriteValue(V3);
+    }
+
+    void CheckFailed(const Twine &Message, const Type *T1,
+                     const Type *T2 = 0, const Type *T3 = 0) {
+      MessagesStr << Message.str() << "\n";
+      WriteType(T1);
+      WriteType(T2);
+      WriteType(T3);
+    }
+  };
+}
+
+char Lint::ID = 0;
+static RegisterPass<Lint>
+X("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)
+
+// Lint::run - This is the main Analysis entry point for a
+// function.
+//
+bool Lint::runOnFunction(Function &F) {
+  Mod = F.getParent();
+  AA = &getAnalysis<AliasAnalysis>();
+  TD = getAnalysisIfAvailable<TargetData>();
+  visit(F);
+  dbgs() << MessagesStr.str();
+  return false;
+}
+
+void Lint::visitInstruction(Instruction &I) {
+}
+
+void Lint::visitCallSite(CallSite CS) {
+  Instruction &I = *CS.getInstruction();
+  Value *Callee = CS.getCalledValue();
+
+  // TODO: Check function alignment?
+  visitMemoryReference(I, Callee, 0, 0);
+
+  if (Function *F = dyn_cast<Function>(Callee->stripPointerCasts())) {
+    Assert1(CS.getCallingConv() == F->getCallingConv(),
+            "Caller and callee calling convention differ", &I);
+
+    const FunctionType *FT = F->getFunctionType();
+    unsigned NumActualArgs = unsigned(CS.arg_end()-CS.arg_begin());
+
+    Assert1(FT->isVarArg() ?
+              FT->getNumParams() <= NumActualArgs :
+              FT->getNumParams() == NumActualArgs,
+            "Call argument count mismatches callee argument count", &I);
+      
+    // TODO: Check argument types (in case the callee was casted)
+
+    // TODO: Check ABI-significant attributes.
+
+    // TODO: Check noalias attribute.
+
+    // TODO: Check sret attribute.
+  }
+
+  // TODO: Check the "tail" keyword constraints.
+
+  if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(&I))
+    switch (II->getIntrinsicID()) {
+    default: break;
+
+    // TODO: Check more intrinsics
+
+    case Intrinsic::memcpy: {
+      MemCpyInst *MCI = cast<MemCpyInst>(&I);
+      visitMemoryReference(I, MCI->getSource(), MCI->getAlignment(), 0);
+      visitMemoryReference(I, MCI->getDest(), MCI->getAlignment(), 0);
+
+      unsigned Size = 0;
+      if (const ConstantInt *Len =
+            dyn_cast<ConstantInt>(MCI->getLength()->stripPointerCasts()))
+        if (Len->getValue().isIntN(32))
+          Size = Len->getValue().getZExtValue();
+      Assert1(AA->alias(MCI->getSource(), Size, MCI->getDest(), Size) !=
+              AliasAnalysis::MustAlias,
+              "memcpy source and destination overlap", &I);
+      break;
+    }
+    case Intrinsic::memmove: {
+      MemMoveInst *MMI = cast<MemMoveInst>(&I);
+      visitMemoryReference(I, MMI->getSource(), MMI->getAlignment(), 0);
+      visitMemoryReference(I, MMI->getDest(), MMI->getAlignment(), 0);
+      break;
+    }
+    case Intrinsic::memset: {
+      MemSetInst *MSI = cast<MemSetInst>(&I);
+      visitMemoryReference(I, MSI->getDest(), MSI->getAlignment(), 0);
+      break;
+    }
+
+    case Intrinsic::vastart:
+      visitMemoryReference(I, CS.getArgument(0), 0, 0);
+      break;
+    case Intrinsic::vacopy:
+      visitMemoryReference(I, CS.getArgument(0), 0, 0);
+      visitMemoryReference(I, CS.getArgument(1), 0, 0);
+      break;
+    case Intrinsic::vaend:
+      visitMemoryReference(I, CS.getArgument(0), 0, 0);
+      break;
+
+    case Intrinsic::stackrestore:
+      visitMemoryReference(I, CS.getArgument(0), 0, 0);
+      break;
+    }
+}
+
+void Lint::visitCallInst(CallInst &I) {
+  return visitCallSite(&I);
+}
+
+void Lint::visitInvokeInst(InvokeInst &I) {
+  return visitCallSite(&I);
+}
+
+void Lint::visitReturnInst(ReturnInst &I) {
+  Function *F = I.getParent()->getParent();
+  Assert1(!F->doesNotReturn(),
+          "Return statement in function with noreturn attribute", &I);
+}
+
+// TODO: Add a length argument and check that the reference is in bounds
+// TODO: Add read/write/execute flags and check for writing to read-only
+//       memory or jumping to suspicious writeable memory
+void Lint::visitMemoryReference(Instruction &I,
+                                Value *Ptr, unsigned Align, const Type *Ty) {
+  Assert1(!isa<ConstantPointerNull>(Ptr->getUnderlyingObject()),
+          "Null pointer dereference", &I);
+  Assert1(!isa<UndefValue>(Ptr->getUnderlyingObject()),
+          "Undef pointer dereference", &I);
+
+  if (TD) {
+    if (Align == 0 && Ty) Align = TD->getABITypeAlignment(Ty);
+
+    if (Align != 0) {
+      unsigned BitWidth = TD->getTypeSizeInBits(Ptr->getType());
+      APInt Mask = APInt::getAllOnesValue(BitWidth),
+                   KnownZero(BitWidth, 0), KnownOne(BitWidth, 0);
+      ComputeMaskedBits(Ptr, Mask, KnownZero, KnownOne, TD);
+      Assert1(!(KnownOne & APInt::getLowBitsSet(BitWidth, Log2_32(Align))),
+              "Memory reference address is misaligned", &I);
+    }
+  }
+}
+
+void Lint::visitLoadInst(LoadInst &I) {
+  visitMemoryReference(I, I.getPointerOperand(), I.getAlignment(), I.getType());
+}
+
+void Lint::visitStoreInst(StoreInst &I) {
+  visitMemoryReference(I, I.getPointerOperand(), I.getAlignment(),
+                  I.getOperand(0)->getType());
+}
+
+static bool isZero(Value *V, TargetData *TD) {
+  unsigned BitWidth = cast<IntegerType>(V->getType())->getBitWidth();
+  APInt Mask = APInt::getAllOnesValue(BitWidth),
+               KnownZero(BitWidth, 0), KnownOne(BitWidth, 0);
+  ComputeMaskedBits(V, Mask, KnownZero, KnownOne, TD);
+  return KnownZero.isAllOnesValue();
+}
+
+void Lint::visitSDiv(BinaryOperator &I) {
+  Assert1(!isZero(I.getOperand(1), TD), "Division by zero", &I);
+}
+
+void Lint::visitUDiv(BinaryOperator &I) {
+  Assert1(!isZero(I.getOperand(1), TD), "Division by zero", &I);
+}
+
+void Lint::visitSRem(BinaryOperator &I) {
+  Assert1(!isZero(I.getOperand(1), TD), "Division by zero", &I);
+}
+
+void Lint::visitURem(BinaryOperator &I) {
+  Assert1(!isZero(I.getOperand(1), TD), "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(),
+            "Static alloca outside of entry block", &I);
+}
+
+void Lint::visitVAArgInst(VAArgInst &I) {
+  visitMemoryReference(I, I.getOperand(0), 0, 0);
+}
+
+void Lint::visitIndirectBrInst(IndirectBrInst &I) {
+  visitMemoryReference(I, I.getAddress(), 0, 0);
+}
+
+//===----------------------------------------------------------------------===//
+//  Implement the public interfaces to this file...
+//===----------------------------------------------------------------------===//
+
+FunctionPass *llvm::createLintPass() {
+  return new Lint();
+}
+
+/// lintFunction - Check a function for errors, printing messages on stderr.
+///
+void llvm::lintFunction(const Function &f) {
+  Function &F = const_cast<Function&>(f);
+  assert(!F.isDeclaration() && "Cannot lint external functions");
+
+  FunctionPassManager FPM(F.getParent());
+  Lint *V = new Lint();
+  FPM.add(V);
+  FPM.run(F);
+}
+
+/// lintModule - Check a module for errors, printing messages on stderr.
+/// Return true if the module is corrupt.
+///
+void llvm::lintModule(const Module &M, std::string *ErrorInfo) {
+  PassManager PM;
+  Lint *V = new Lint();
+  PM.add(V);
+  PM.run(const_cast<Module&>(M));
+
+  if (ErrorInfo)
+    *ErrorInfo = V->MessagesStr.str();
+}
diff --git a/test/Other/lint.ll b/test/Other/lint.ll
new file mode 100644
index 0000000..6ccaa6f
--- /dev/null
+++ b/test/Other/lint.ll
@@ -0,0 +1,31 @@
+; RUN: opt -lint -disable-output < %s |& FileCheck %s
+target datalayout = "e-p:64:64:64"
+
+declare fastcc void @bar()
+
+define i32 @foo() noreturn {
+; CHECK: Caller and callee calling convention differ
+  call void @bar()
+; CHECK: Null pointer dereference
+  store i32 0, i32* null
+; CHECK: Null pointer dereference
+  %t = load i32* null
+; CHECK: Memory reference address is misaligned
+  %x = inttoptr i32 1 to i32*
+  load i32* %x, align 4
+; CHECK: Division by zero
+  %sd = sdiv i32 2, 0
+; CHECK: Division by zero
+  %ud = udiv i32 2, 0
+; CHECK: Division by zero
+  %sr = srem i32 2, 0
+; CHECK: Division by zero
+  %ur = urem i32 2, 0
+  br label %next
+
+next:
+; CHECK: Static alloca outside of entry block
+  %a = alloca i32
+; CHECK: Return statement in function with noreturn attribute
+  ret i32 0
+}