/// parameters and return values may be passed via registers, we have a
/// specialized thread-local shadow for return values
/// (__msan_retval_tls) and parameters (__msan_param_tls).
+///
+/// Origin tracking.
+///
+/// MemorySanitizer can track origins (allocation points) of all uninitialized
+/// values. This behavior is controlled with a flag (msan-track-origins) and is
+/// disabled by default.
+///
+/// Origins are 4-byte values created and interpreted by the runtime library.
+/// They are stored in a second shadow mapping, one 4-byte value for 4 bytes
+/// of application memory. Propagation of origins is basically a bunch of
+/// "select" instructions that pick the origin of a dirty argument, if an
+/// instruction has one.
+///
+/// Every 4 aligned, consecutive bytes of application memory have one origin
+/// value associated with them. If these bytes contain uninitialized data
+/// coming from 2 different allocations, the last store wins. Because of this,
+/// MemorySanitizer reports can show unrelated origins, but this is unlikely in
+/// practice.
+///
+/// Origins are meaningless for fully initialized values, so MemorySanitizer
+/// avoids storing origin to memory when a fully initialized value is stored.
+/// This way it avoids needless overwritting origin of the 4-byte region on
+/// a short (i.e. 1 byte) clean store, and it is also good for performance.
//===----------------------------------------------------------------------===//
#define DEBUG_TYPE "msan"
#include "llvm/Transforms/Instrumentation.h"
-#include "BlackList.h"
#include "llvm/ADT/DepthFirstIterator.h"
#include "llvm/ADT/SmallString.h"
#include "llvm/ADT/SmallVector.h"
+#include "llvm/ADT/Triple.h"
#include "llvm/ADT/ValueMap.h"
-#include "llvm/DataLayout.h"
-#include "llvm/Function.h"
-#include "llvm/IRBuilder.h"
-#include "llvm/InlineAsm.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/IRBuilder.h"
+#include "llvm/IR/InlineAsm.h"
+#include "llvm/IR/IntrinsicInst.h"
+#include "llvm/IR/LLVMContext.h"
+#include "llvm/IR/MDBuilder.h"
+#include "llvm/IR/Module.h"
+#include "llvm/IR/Type.h"
#include "llvm/InstVisitor.h"
-#include "llvm/IntrinsicInst.h"
-#include "llvm/LLVMContext.h"
-#include "llvm/MDBuilder.h"
-#include "llvm/Module.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Compiler.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/Transforms/Utils/BasicBlockUtils.h"
+#include "llvm/Transforms/Utils/Local.h"
#include "llvm/Transforms/Utils/ModuleUtils.h"
-#include "llvm/Type.h"
+#include "llvm/Transforms/Utils/SpecialCaseList.h"
using namespace llvm;
static const uint64_t kShadowMask64 = 1ULL << 46;
static const uint64_t kOriginOffset32 = 1ULL << 30;
static const uint64_t kOriginOffset64 = 1ULL << 45;
-static const uint64_t kShadowTLSAlignment = 8;
+static const unsigned kMinOriginAlignment = 4;
+static const unsigned kShadowTLSAlignment = 8;
-// This is an important flag that makes the reports much more
-// informative at the cost of greater slowdown. Not fully implemented
-// yet.
-// FIXME: this should be a top-level clang flag, e.g.
-// -fmemory-sanitizer-full.
+/// \brief Track origins of uninitialized values.
+///
+/// Adds a section to MemorySanitizer report that points to the allocation
+/// (stack or heap) the uninitialized bits came from originally.
static cl::opt<bool> ClTrackOrigins("msan-track-origins",
cl::desc("Track origins (allocation sites) of poisoned memory"),
cl::Hidden, cl::init(false));
static cl::opt<int> ClPoisonStackPattern("msan-poison-stack-pattern",
cl::desc("poison uninitialized stack variables with the given patter"),
cl::Hidden, cl::init(0xff));
+static cl::opt<bool> ClPoisonUndef("msan-poison-undef",
+ cl::desc("poison undef temps"),
+ cl::Hidden, cl::init(true));
static cl::opt<bool> ClHandleICmp("msan-handle-icmp",
cl::desc("propagate shadow through ICmpEQ and ICmpNE"),
cl::Hidden, cl::init(true));
+static cl::opt<bool> ClHandleICmpExact("msan-handle-icmp-exact",
+ cl::desc("exact handling of relational integer ICmp"),
+ cl::Hidden, cl::init(false));
+
static cl::opt<bool> ClStoreCleanOrigin("msan-store-clean-origin",
cl::desc("store origin for clean (fully initialized) values"),
cl::Hidden, cl::init(false));
cl::desc("print out instructions with default strict semantics"),
cl::Hidden, cl::init(false));
-static cl::opt<std::string> ClBlackListFile("msan-blacklist",
+static cl::opt<std::string> ClBlacklistFile("msan-blacklist",
cl::desc("File containing the list of functions where MemorySanitizer "
"should not report bugs"), cl::Hidden);
/// uninitialized reads.
class MemorySanitizer : public FunctionPass {
public:
- MemorySanitizer() : FunctionPass(ID), TD(0), WarningFn(0) { }
+ MemorySanitizer(bool TrackOrigins = false,
+ StringRef BlacklistFile = StringRef())
+ : FunctionPass(ID),
+ TrackOrigins(TrackOrigins || ClTrackOrigins),
+ TD(0),
+ WarningFn(0),
+ BlacklistFile(BlacklistFile.empty() ? ClBlacklistFile
+ : BlacklistFile) { }
const char *getPassName() const { return "MemorySanitizer"; }
bool runOnFunction(Function &F);
bool doInitialization(Module &M);
private:
void initializeCallbacks(Module &M);
+ /// \brief Track origins (allocation points) of uninitialized values.
+ bool TrackOrigins;
+
DataLayout *TD;
LLVMContext *C;
Type *IntptrTy;
MDNode *ColdCallWeights;
/// \brief Branch weights for origin store.
MDNode *OriginStoreWeights;
+ /// \brief Path to blacklist file.
+ SmallString<64> BlacklistFile;
/// \brief The blacklist.
- OwningPtr<BlackList> BL;
+ OwningPtr<SpecialCaseList> BL;
/// \brief An empty volatile inline asm that prevents callback merge.
InlineAsm *EmptyAsm;
"MemorySanitizer: detects uninitialized reads.",
false, false)
-FunctionPass *llvm::createMemorySanitizerPass() {
- return new MemorySanitizer();
+FunctionPass *llvm::createMemorySanitizerPass(bool TrackOrigins,
+ StringRef BlacklistFile) {
+ return new MemorySanitizer(TrackOrigins, BlacklistFile);
}
/// \brief Create a non-const global initialized with the given string.
RetvalTLS = new GlobalVariable(
M, ArrayType::get(IRB.getInt64Ty(), 8), false,
GlobalVariable::ExternalLinkage, 0, "__msan_retval_tls", 0,
- GlobalVariable::GeneralDynamicTLSModel);
+ GlobalVariable::InitialExecTLSModel);
RetvalOriginTLS = new GlobalVariable(
M, OriginTy, false, GlobalVariable::ExternalLinkage, 0,
- "__msan_retval_origin_tls", 0, GlobalVariable::GeneralDynamicTLSModel);
+ "__msan_retval_origin_tls", 0, GlobalVariable::InitialExecTLSModel);
ParamTLS = new GlobalVariable(
M, ArrayType::get(IRB.getInt64Ty(), 1000), false,
GlobalVariable::ExternalLinkage, 0, "__msan_param_tls", 0,
- GlobalVariable::GeneralDynamicTLSModel);
+ GlobalVariable::InitialExecTLSModel);
ParamOriginTLS = new GlobalVariable(
M, ArrayType::get(OriginTy, 1000), false, GlobalVariable::ExternalLinkage,
- 0, "__msan_param_origin_tls", 0, GlobalVariable::GeneralDynamicTLSModel);
+ 0, "__msan_param_origin_tls", 0, GlobalVariable::InitialExecTLSModel);
VAArgTLS = new GlobalVariable(
M, ArrayType::get(IRB.getInt64Ty(), 1000), false,
GlobalVariable::ExternalLinkage, 0, "__msan_va_arg_tls", 0,
- GlobalVariable::GeneralDynamicTLSModel);
+ GlobalVariable::InitialExecTLSModel);
VAArgOverflowSizeTLS = new GlobalVariable(
M, IRB.getInt64Ty(), false, GlobalVariable::ExternalLinkage, 0,
"__msan_va_arg_overflow_size_tls", 0,
- GlobalVariable::GeneralDynamicTLSModel);
+ GlobalVariable::InitialExecTLSModel);
OriginTLS = new GlobalVariable(
M, IRB.getInt32Ty(), false, GlobalVariable::ExternalLinkage, 0,
- "__msan_origin_tls", 0, GlobalVariable::GeneralDynamicTLSModel);
+ "__msan_origin_tls", 0, GlobalVariable::InitialExecTLSModel);
// We insert an empty inline asm after __msan_report* to avoid callback merge.
EmptyAsm = InlineAsm::get(FunctionType::get(IRB.getVoidTy(), false),
TD = getAnalysisIfAvailable<DataLayout>();
if (!TD)
return false;
- BL.reset(new BlackList(ClBlackListFile));
+ BL.reset(SpecialCaseList::createOrDie(BlacklistFile));
C = &(M.getContext());
unsigned PtrSize = TD->getPointerSizeInBits(/* AddressSpace */0);
switch (PtrSize) {
appendToGlobalCtors(M, cast<Function>(M.getOrInsertFunction(
"__msan_init", IRB.getVoidTy(), NULL)), 0);
- new GlobalVariable(M, IRB.getInt32Ty(), true, GlobalValue::WeakODRLinkage,
- IRB.getInt32(ClTrackOrigins), "__msan_track_origins");
+ if (TrackOrigins)
+ new GlobalVariable(M, IRB.getInt32Ty(), true, GlobalValue::WeakODRLinkage,
+ IRB.getInt32(TrackOrigins), "__msan_track_origins");
+
+ if (ClKeepGoing)
+ new GlobalVariable(M, IRB.getInt32Ty(), true, GlobalValue::WeakODRLinkage,
+ IRB.getInt32(ClKeepGoing), "__msan_keep_going");
return true;
}
SmallVector<PHINode *, 16> ShadowPHINodes, OriginPHINodes;
ValueMap<Value*, Value*> ShadowMap, OriginMap;
bool InsertChecks;
+ bool LoadShadow;
+ bool PoisonStack;
+ bool PoisonUndef;
OwningPtr<VarArgHelper> VAHelper;
- // An unfortunate workaround for asymmetric lowering of va_arg stuff.
- // See a comment in visitCallSite for more details.
- static const unsigned AMD64GpEndOffset = 48; // AMD64 ABI Draft 0.99.6 p3.5.7
- static const unsigned AMD64FpEndOffset = 176;
-
struct ShadowOriginAndInsertPoint {
Instruction *Shadow;
Instruction *Origin;
SmallVector<Instruction*, 16> StoreList;
MemorySanitizerVisitor(Function &F, MemorySanitizer &MS)
- : F(F), MS(MS), VAHelper(CreateVarArgHelper(F, MS, *this)) {
- InsertChecks = !MS.BL->isIn(F);
+ : F(F), MS(MS), VAHelper(CreateVarArgHelper(F, MS, *this)) {
+ bool SanitizeFunction = !MS.BL->isIn(F) && F.getAttributes().hasAttribute(
+ AttributeSet::FunctionIndex,
+ Attribute::SanitizeMemory);
+ InsertChecks = SanitizeFunction;
+ LoadShadow = SanitizeFunction;
+ PoisonStack = SanitizeFunction && ClPoisonStack;
+ PoisonUndef = SanitizeFunction && ClPoisonUndef;
+
DEBUG(if (!InsertChecks)
- dbgs() << "MemorySanitizer is not inserting checks into '"
- << F.getName() << "'\n");
+ dbgs() << "MemorySanitizer is not inserting checks into '"
+ << F.getName() << "'\n");
}
void materializeStores() {
IRB.CreateAlignedStore(Shadow, ShadowPtr, I.getAlignment());
DEBUG(dbgs() << " STORE: " << *NewSI << "\n");
(void)NewSI;
- // If the store is volatile, add a check.
- if (I.isVolatile())
- insertCheck(Val, &I);
+
if (ClCheckAccessAddress)
insertCheck(Addr, &I);
- if (ClTrackOrigins) {
+ if (MS.TrackOrigins) {
+ unsigned Alignment = std::max(kMinOriginAlignment, I.getAlignment());
if (ClStoreCleanOrigin || isa<StructType>(Shadow->getType())) {
- IRB.CreateStore(getOrigin(Val), getOriginPtr(Addr, IRB));
+ IRB.CreateAlignedStore(getOrigin(Val), getOriginPtr(Addr, IRB),
+ Alignment);
} else {
Value *ConvertedShadow = convertToShadowTyNoVec(Shadow, IRB);
SplitBlockAndInsertIfThen(cast<Instruction>(Cmp), false,
MS.OriginStoreWeights);
IRBuilder<> IRBNew(CheckTerm);
- IRBNew.CreateStore(getOrigin(Val), getOriginPtr(Addr, IRBNew));
+ IRBNew.CreateAlignedStore(getOrigin(Val), getOriginPtr(Addr, IRBNew),
+ Alignment);
}
}
}
MS.ColdCallWeights);
IRB.SetInsertPoint(CheckTerm);
- if (ClTrackOrigins) {
+ if (MS.TrackOrigins) {
Instruction *Origin = InstrumentationList[i].Origin;
IRB.CreateStore(Origin ? (Value*)Origin : (Value*)IRB.getInt32(0),
MS.OriginTLS);
bool runOnFunction() {
MS.initializeCallbacks(*F.getParent());
if (!MS.TD) return false;
+
+ // In the presence of unreachable blocks, we may see Phi nodes with
+ // incoming nodes from such blocks. Since InstVisitor skips unreachable
+ // blocks, such nodes will not have any shadow value associated with them.
+ // It's easier to remove unreachable blocks than deal with missing shadow.
+ removeUnreachableBlocks(F);
+
// Iterate all BBs in depth-first order and create shadow instructions
// for all instructions (where applicable).
// For PHI nodes we create dummy shadow PHIs which will be finalized later.
for (size_t i = 0, n = ShadowPHINodes.size(); i < n; i++) {
PHINode *PN = ShadowPHINodes[i];
PHINode *PNS = cast<PHINode>(getShadow(PN));
- PHINode *PNO = ClTrackOrigins ? cast<PHINode>(getOrigin(PN)) : 0;
+ PHINode *PNO = MS.TrackOrigins ? cast<PHINode>(getOrigin(PN)) : 0;
size_t NumValues = PN->getNumIncomingValues();
for (size_t v = 0; v < NumValues; v++) {
PNS->addIncoming(getShadow(PN, v), PN->getIncomingBlock(v));
// This may return weird-sized types like i1.
if (IntegerType *IT = dyn_cast<IntegerType>(OrigTy))
return IT;
- if (VectorType *VT = dyn_cast<VectorType>(OrigTy))
- return VectorType::getInteger(VT);
+ if (VectorType *VT = dyn_cast<VectorType>(OrigTy)) {
+ uint32_t EltSize = MS.TD->getTypeSizeInBits(VT->getElementType());
+ return VectorType::get(IntegerType::get(*MS.C, EltSize),
+ VT->getNumElements());
+ }
if (StructType *ST = dyn_cast<StructType>(OrigTy)) {
SmallVector<Type*, 4> Elements;
for (unsigned i = 0, n = ST->getNumElements(); i < n; i++)
DEBUG(dbgs() << "getShadowTy: " << *ST << " ===> " << *Res << "\n");
return Res;
}
- uint32_t TypeSize = MS.TD->getTypeStoreSizeInBits(OrigTy);
+ uint32_t TypeSize = MS.TD->getTypeSizeInBits(OrigTy);
return IntegerType::get(*MS.C, TypeSize);
}
/// \brief Compute the origin address for a given function argument.
Value *getOriginPtrForArgument(Value *A, IRBuilder<> &IRB,
int ArgOffset) {
- if (!ClTrackOrigins) return 0;
+ if (!MS.TrackOrigins) return 0;
Value *Base = IRB.CreatePointerCast(MS.ParamOriginTLS, MS.IntptrTy);
Base = IRB.CreateAdd(Base, ConstantInt::get(MS.IntptrTy, ArgOffset));
return IRB.CreateIntToPtr(Base, PointerType::get(MS.OriginTy, 0),
/// \brief Set Origin to be the origin value for V.
void setOrigin(Value *V, Value *Origin) {
- if (!ClTrackOrigins) return;
+ if (!MS.TrackOrigins) return;
assert(!OriginMap.count(V) && "Values may only have one origin");
DEBUG(dbgs() << "ORIGIN: " << *V << " ==> " << *Origin << "\n");
OriginMap[V] = Origin;
///
/// Clean shadow (all zeroes) means all bits of the value are defined
/// (initialized).
- Value *getCleanShadow(Value *V) {
+ Constant *getCleanShadow(Value *V) {
Type *ShadowTy = getShadowTy(V);
if (!ShadowTy)
return 0;
return ConstantStruct::get(ST, Vals);
}
+ /// \brief Create a dirty shadow for a given value.
+ Constant *getPoisonedShadow(Value *V) {
+ Type *ShadowTy = getShadowTy(V);
+ if (!ShadowTy)
+ return 0;
+ return getPoisonedShadow(ShadowTy);
+ }
+
/// \brief Create a clean (zero) origin.
Value *getCleanOrigin() {
return Constant::getNullValue(MS.OriginTy);
return Shadow;
}
if (UndefValue *U = dyn_cast<UndefValue>(V)) {
- Value *AllOnes = getPoisonedShadow(getShadowTy(V));
+ Value *AllOnes = PoisonUndef ? getPoisonedShadow(V) : getCleanShadow(V);
DEBUG(dbgs() << "Undef: " << *U << " ==> " << *AllOnes << "\n");
(void)U;
return AllOnes;
if (AI->hasByValAttr()) {
// ByVal pointer itself has clean shadow. We copy the actual
// argument shadow to the underlying memory.
+ // Figure out maximal valid memcpy alignment.
+ unsigned ArgAlign = AI->getParamAlignment();
+ if (ArgAlign == 0) {
+ Type *EltType = A->getType()->getPointerElementType();
+ ArgAlign = MS.TD->getABITypeAlignment(EltType);
+ }
+ unsigned CopyAlign = std::min(ArgAlign, kShadowTLSAlignment);
Value *Cpy = EntryIRB.CreateMemCpy(
- getShadowPtr(V, EntryIRB.getInt8Ty(), EntryIRB),
- Base, Size, AI->getParamAlignment());
+ getShadowPtr(V, EntryIRB.getInt8Ty(), EntryIRB), Base, Size,
+ CopyAlign);
DEBUG(dbgs() << " ByValCpy: " << *Cpy << "\n");
(void)Cpy;
*ShadowPtr = getCleanShadow(V);
} else {
- *ShadowPtr = EntryIRB.CreateLoad(Base);
+ *ShadowPtr = EntryIRB.CreateAlignedLoad(Base, kShadowTLSAlignment);
}
DEBUG(dbgs() << " ARG: " << *AI << " ==> " <<
**ShadowPtr << "\n");
- if (ClTrackOrigins) {
+ if (MS.TrackOrigins) {
Value* OriginPtr = getOriginPtrForArgument(AI, EntryIRB, ArgOffset);
setOrigin(A, EntryIRB.CreateLoad(OriginPtr));
}
}
- ArgOffset += DataLayout::RoundUpAlignment(Size, 8);
+ ArgOffset += DataLayout::RoundUpAlignment(Size, kShadowTLSAlignment);
}
assert(*ShadowPtr && "Could not find shadow for an argument");
return *ShadowPtr;
/// \brief Get the origin for a value.
Value *getOrigin(Value *V) {
- if (!ClTrackOrigins) return 0;
+ if (!MS.TrackOrigins) return 0;
if (isa<Instruction>(V) || isa<Argument>(V)) {
Value *Origin = OriginMap[V];
if (!Origin) {
IRBuilder<> IRB(&I);
Type *ShadowTy = getShadowTy(&I);
Value *Addr = I.getPointerOperand();
- Value *ShadowPtr = getShadowPtr(Addr, ShadowTy, IRB);
- setShadow(&I, IRB.CreateAlignedLoad(ShadowPtr, I.getAlignment(), "_msld"));
+ if (LoadShadow) {
+ Value *ShadowPtr = getShadowPtr(Addr, ShadowTy, IRB);
+ setShadow(&I,
+ IRB.CreateAlignedLoad(ShadowPtr, I.getAlignment(), "_msld"));
+ } else {
+ setShadow(&I, getCleanShadow(&I));
+ }
if (ClCheckAccessAddress)
insertCheck(I.getPointerOperand(), &I);
- if (ClTrackOrigins)
- setOrigin(&I, IRB.CreateLoad(getOriginPtr(Addr, IRB)));
+ if (MS.TrackOrigins) {
+ if (LoadShadow) {
+ unsigned Alignment = std::max(kMinOriginAlignment, I.getAlignment());
+ setOrigin(&I,
+ IRB.CreateAlignedLoad(getOriginPtr(Addr, IRB), Alignment));
+ } else {
+ setOrigin(&I, getCleanOrigin());
+ }
+ }
}
/// \brief Instrument StoreInst
///
/// Stores the corresponding shadow and (optionally) origin.
/// Optionally, checks that the store address is fully defined.
- /// Volatile stores check that the value being stored is fully defined.
void visitStoreInst(StoreInst &I) {
StoreList.push_back(&I);
}
}
}
- if (ClTrackOrigins) {
+ if (MSV->MS.TrackOrigins) {
assert(OpOrigin);
if (!Origin) {
Origin = OpOrigin;
/// \brief Add an application value to the mix.
Combiner &Add(Value *V) {
Value *OpShadow = MSV->getShadow(V);
- Value *OpOrigin = ClTrackOrigins ? MSV->getOrigin(V) : 0;
+ Value *OpOrigin = MSV->MS.TrackOrigins ? MSV->getOrigin(V) : 0;
return Add(OpShadow, OpOrigin);
}
Shadow = MSV->CreateShadowCast(IRB, Shadow, MSV->getShadowTy(I));
MSV->setShadow(I, Shadow);
}
- if (ClTrackOrigins) {
+ if (MSV->MS.TrackOrigins) {
assert(Origin);
MSV->setOrigin(I, Origin);
}
/// \brief Propagate origin for arbitrary operation.
void setOriginForNaryOp(Instruction &I) {
- if (!ClTrackOrigins) return;
+ if (!MS.TrackOrigins) return;
IRBuilder<> IRB(&I);
OriginCombiner OC(this, IRB);
for (Instruction::op_iterator OI = I.op_begin(); OI != I.op_end(); ++OI)
}
size_t VectorOrPrimitiveTypeSizeInBits(Type *Ty) {
+ assert(!(Ty->isVectorTy() && Ty->getScalarType()->isPointerTy()) &&
+ "Vector of pointers is not a valid shadow type");
return Ty->isVectorTy() ?
Ty->getVectorNumElements() * Ty->getScalarSizeInBits() :
Ty->getPrimitiveSizeInBits();
Value *B = I.getOperand(1);
Value *Sa = getShadow(A);
Value *Sb = getShadow(B);
- if (A->getType()->isPointerTy())
- A = IRB.CreatePointerCast(A, MS.IntptrTy);
- if (B->getType()->isPointerTy())
- B = IRB.CreatePointerCast(B, MS.IntptrTy);
+
+ // Get rid of pointers and vectors of pointers.
+ // For ints (and vectors of ints), types of A and Sa match,
+ // and this is a no-op.
+ A = IRB.CreatePointerCast(A, Sa->getType());
+ B = IRB.CreatePointerCast(B, Sb->getType());
+
// A == B <==> (C = A^B) == 0
// A != B <==> (C = A^B) != 0
// Sc = Sa | Sb
setOriginForNaryOp(I);
}
+ /// \brief Build the lowest possible value of V, taking into account V's
+ /// uninitialized bits.
+ Value *getLowestPossibleValue(IRBuilder<> &IRB, Value *A, Value *Sa,
+ bool isSigned) {
+ if (isSigned) {
+ // Split shadow into sign bit and other bits.
+ Value *SaOtherBits = IRB.CreateLShr(IRB.CreateShl(Sa, 1), 1);
+ Value *SaSignBit = IRB.CreateXor(Sa, SaOtherBits);
+ // Maximise the undefined shadow bit, minimize other undefined bits.
+ return
+ IRB.CreateOr(IRB.CreateAnd(A, IRB.CreateNot(SaOtherBits)), SaSignBit);
+ } else {
+ // Minimize undefined bits.
+ return IRB.CreateAnd(A, IRB.CreateNot(Sa));
+ }
+ }
+
+ /// \brief Build the highest possible value of V, taking into account V's
+ /// uninitialized bits.
+ Value *getHighestPossibleValue(IRBuilder<> &IRB, Value *A, Value *Sa,
+ bool isSigned) {
+ if (isSigned) {
+ // Split shadow into sign bit and other bits.
+ Value *SaOtherBits = IRB.CreateLShr(IRB.CreateShl(Sa, 1), 1);
+ Value *SaSignBit = IRB.CreateXor(Sa, SaOtherBits);
+ // Minimise the undefined shadow bit, maximise other undefined bits.
+ return
+ IRB.CreateOr(IRB.CreateAnd(A, IRB.CreateNot(SaSignBit)), SaOtherBits);
+ } else {
+ // Maximize undefined bits.
+ return IRB.CreateOr(A, Sa);
+ }
+ }
+
+ /// \brief Instrument relational comparisons.
+ ///
+ /// This function does exact shadow propagation for all relational
+ /// comparisons of integers, pointers and vectors of those.
+ /// FIXME: output seems suboptimal when one of the operands is a constant
+ void handleRelationalComparisonExact(ICmpInst &I) {
+ IRBuilder<> IRB(&I);
+ Value *A = I.getOperand(0);
+ Value *B = I.getOperand(1);
+ Value *Sa = getShadow(A);
+ Value *Sb = getShadow(B);
+
+ // Get rid of pointers and vectors of pointers.
+ // For ints (and vectors of ints), types of A and Sa match,
+ // and this is a no-op.
+ A = IRB.CreatePointerCast(A, Sa->getType());
+ B = IRB.CreatePointerCast(B, Sb->getType());
+
+ // Let [a0, a1] be the interval of possible values of A, taking into account
+ // its undefined bits. Let [b0, b1] be the interval of possible values of B.
+ // Then (A cmp B) is defined iff (a0 cmp b1) == (a1 cmp b0).
+ bool IsSigned = I.isSigned();
+ Value *S1 = IRB.CreateICmp(I.getPredicate(),
+ getLowestPossibleValue(IRB, A, Sa, IsSigned),
+ getHighestPossibleValue(IRB, B, Sb, IsSigned));
+ Value *S2 = IRB.CreateICmp(I.getPredicate(),
+ getHighestPossibleValue(IRB, A, Sa, IsSigned),
+ getLowestPossibleValue(IRB, B, Sb, IsSigned));
+ Value *Si = IRB.CreateXor(S1, S2);
+ setShadow(&I, Si);
+ setOriginForNaryOp(I);
+ }
+
/// \brief Instrument signed relational comparisons.
///
/// Handle (x<0) and (x>=0) comparisons (essentially, sign bit tests) by
}
void visitICmpInst(ICmpInst &I) {
- if (ClHandleICmp && I.isEquality())
+ if (!ClHandleICmp) {
+ handleShadowOr(I);
+ return;
+ }
+ if (I.isEquality()) {
handleEqualityComparison(I);
- else if (ClHandleICmp && I.isSigned() && I.isRelational())
+ return;
+ }
+
+ assert(I.isRelational());
+ if (ClHandleICmpExact) {
+ handleRelationalComparisonExact(I);
+ return;
+ }
+ if (I.isSigned()) {
handleSignedRelationalComparison(I);
- else
- handleShadowOr(I);
+ return;
+ }
+
+ assert(I.isUnsigned());
+ if ((isa<Constant>(I.getOperand(0)) || isa<Constant>(I.getOperand(1)))) {
+ handleRelationalComparisonExact(I);
+ return;
+ }
+
+ handleShadowOr(I);
}
void visitFCmpInst(FCmpInst &I) {
VAHelper->visitVACopyInst(I);
}
+ enum IntrinsicKind {
+ IK_DoesNotAccessMemory,
+ IK_OnlyReadsMemory,
+ IK_WritesMemory
+ };
+
+ static IntrinsicKind getIntrinsicKind(Intrinsic::ID iid) {
+ const int DoesNotAccessMemory = IK_DoesNotAccessMemory;
+ const int OnlyReadsArgumentPointees = IK_OnlyReadsMemory;
+ const int OnlyReadsMemory = IK_OnlyReadsMemory;
+ const int OnlyAccessesArgumentPointees = IK_WritesMemory;
+ const int UnknownModRefBehavior = IK_WritesMemory;
+#define GET_INTRINSIC_MODREF_BEHAVIOR
+#define ModRefBehavior IntrinsicKind
+#include "llvm/IR/Intrinsics.gen"
+#undef ModRefBehavior
+#undef GET_INTRINSIC_MODREF_BEHAVIOR
+ }
+
+ /// \brief Handle vector store-like intrinsics.
+ ///
+ /// Instrument intrinsics that look like a simple SIMD store: writes memory,
+ /// has 1 pointer argument and 1 vector argument, returns void.
+ bool handleVectorStoreIntrinsic(IntrinsicInst &I) {
+ IRBuilder<> IRB(&I);
+ Value* Addr = I.getArgOperand(0);
+ Value *Shadow = getShadow(&I, 1);
+ Value *ShadowPtr = getShadowPtr(Addr, Shadow->getType(), IRB);
+
+ // We don't know the pointer alignment (could be unaligned SSE store!).
+ // Have to assume to worst case.
+ IRB.CreateAlignedStore(Shadow, ShadowPtr, 1);
+
+ if (ClCheckAccessAddress)
+ insertCheck(Addr, &I);
+
+ // FIXME: use ClStoreCleanOrigin
+ // FIXME: factor out common code from materializeStores
+ if (MS.TrackOrigins)
+ IRB.CreateStore(getOrigin(&I, 1), getOriginPtr(Addr, IRB));
+ return true;
+ }
+
+ /// \brief Handle vector load-like intrinsics.
+ ///
+ /// Instrument intrinsics that look like a simple SIMD load: reads memory,
+ /// has 1 pointer argument, returns a vector.
+ bool handleVectorLoadIntrinsic(IntrinsicInst &I) {
+ IRBuilder<> IRB(&I);
+ Value *Addr = I.getArgOperand(0);
+
+ Type *ShadowTy = getShadowTy(&I);
+ if (LoadShadow) {
+ Value *ShadowPtr = getShadowPtr(Addr, ShadowTy, IRB);
+ // We don't know the pointer alignment (could be unaligned SSE load!).
+ // Have to assume to worst case.
+ setShadow(&I, IRB.CreateAlignedLoad(ShadowPtr, 1, "_msld"));
+ } else {
+ setShadow(&I, getCleanShadow(&I));
+ }
+
+
+ if (ClCheckAccessAddress)
+ insertCheck(Addr, &I);
+
+ if (MS.TrackOrigins) {
+ if (LoadShadow)
+ setOrigin(&I, IRB.CreateLoad(getOriginPtr(Addr, IRB)));
+ else
+ setOrigin(&I, getCleanOrigin());
+ }
+ return true;
+ }
+
+ /// \brief Handle (SIMD arithmetic)-like intrinsics.
+ ///
+ /// Instrument intrinsics with any number of arguments of the same type,
+ /// equal to the return type. The type should be simple (no aggregates or
+ /// pointers; vectors are fine).
+ /// Caller guarantees that this intrinsic does not access memory.
+ bool maybeHandleSimpleNomemIntrinsic(IntrinsicInst &I) {
+ Type *RetTy = I.getType();
+ if (!(RetTy->isIntOrIntVectorTy() ||
+ RetTy->isFPOrFPVectorTy() ||
+ RetTy->isX86_MMXTy()))
+ return false;
+
+ unsigned NumArgOperands = I.getNumArgOperands();
+
+ for (unsigned i = 0; i < NumArgOperands; ++i) {
+ Type *Ty = I.getArgOperand(i)->getType();
+ if (Ty != RetTy)
+ return false;
+ }
+
+ IRBuilder<> IRB(&I);
+ ShadowAndOriginCombiner SC(this, IRB);
+ for (unsigned i = 0; i < NumArgOperands; ++i)
+ SC.Add(I.getArgOperand(i));
+ SC.Done(&I);
+
+ return true;
+ }
+
+ /// \brief Heuristically instrument unknown intrinsics.
+ ///
+ /// The main purpose of this code is to do something reasonable with all
+ /// random intrinsics we might encounter, most importantly - SIMD intrinsics.
+ /// We recognize several classes of intrinsics by their argument types and
+ /// ModRefBehaviour and apply special intrumentation when we are reasonably
+ /// sure that we know what the intrinsic does.
+ ///
+ /// We special-case intrinsics where this approach fails. See llvm.bswap
+ /// handling as an example of that.
+ bool handleUnknownIntrinsic(IntrinsicInst &I) {
+ unsigned NumArgOperands = I.getNumArgOperands();
+ if (NumArgOperands == 0)
+ return false;
+
+ Intrinsic::ID iid = I.getIntrinsicID();
+ IntrinsicKind IK = getIntrinsicKind(iid);
+ bool OnlyReadsMemory = IK == IK_OnlyReadsMemory;
+ bool WritesMemory = IK == IK_WritesMemory;
+ assert(!(OnlyReadsMemory && WritesMemory));
+
+ if (NumArgOperands == 2 &&
+ I.getArgOperand(0)->getType()->isPointerTy() &&
+ I.getArgOperand(1)->getType()->isVectorTy() &&
+ I.getType()->isVoidTy() &&
+ WritesMemory) {
+ // This looks like a vector store.
+ return handleVectorStoreIntrinsic(I);
+ }
+
+ if (NumArgOperands == 1 &&
+ I.getArgOperand(0)->getType()->isPointerTy() &&
+ I.getType()->isVectorTy() &&
+ OnlyReadsMemory) {
+ // This looks like a vector load.
+ return handleVectorLoadIntrinsic(I);
+ }
+
+ if (!OnlyReadsMemory && !WritesMemory)
+ if (maybeHandleSimpleNomemIntrinsic(I))
+ return true;
+
+ // FIXME: detect and handle SSE maskstore/maskload
+ return false;
+ }
+
void handleBswap(IntrinsicInst &I) {
IRBuilder<> IRB(&I);
Value *Op = I.getArgOperand(0);
handleBswap(I);
break;
default:
- visitInstruction(I);
+ if (!handleUnknownIntrinsic(I))
+ visitInstruction(I);
break;
}
}
if (Function *Func = Call->getCalledFunction()) {
// Clear out readonly/readnone attributes.
AttrBuilder B;
- B.addAttribute(Attributes::ReadOnly)
- .addAttribute(Attributes::ReadNone);
- Func->removeAttribute(AttributeSet::FunctionIndex,
- Attributes::get(Func->getContext(), B));
+ B.addAttribute(Attribute::ReadOnly)
+ .addAttribute(Attribute::ReadNone);
+ Func->removeAttributes(AttributeSet::FunctionIndex,
+ AttributeSet::get(Func->getContext(),
+ AttributeSet::FunctionIndex,
+ B));
}
}
IRBuilder<> IRB(&I);
Value *ArgShadowBase = getShadowPtrForArgument(A, IRB, ArgOffset);
DEBUG(dbgs() << " Arg#" << i << ": " << *A <<
" Shadow: " << *ArgShadow << "\n");
- if (CS.paramHasAttr(i + 1, Attributes::ByVal)) {
+ if (CS.paramHasAttr(i + 1, Attribute::ByVal)) {
assert(A->getType()->isPointerTy() &&
"ByVal argument is not a pointer!");
Size = MS.TD->getTypeAllocSize(A->getType()->getPointerElementType());
Store = IRB.CreateAlignedStore(ArgShadow, ArgShadowBase,
kShadowTLSAlignment);
}
- if (ClTrackOrigins)
+ if (MS.TrackOrigins)
IRB.CreateStore(getOrigin(A),
getOriginPtrForArgument(A, IRB, ArgOffset));
+ (void)Store;
assert(Size != 0 && Store != 0);
DEBUG(dbgs() << " Param:" << *Store << "\n");
ArgOffset += DataLayout::RoundUpAlignment(Size, 8);
IRBAfter.CreateAlignedLoad(getShadowPtrForRetval(&I, IRBAfter),
kShadowTLSAlignment, "_msret");
setShadow(&I, RetvalShadow);
- if (ClTrackOrigins)
+ if (MS.TrackOrigins)
setOrigin(&I, IRBAfter.CreateLoad(getOriginPtrForRetval(IRBAfter)));
}
Value *ShadowPtr = getShadowPtrForRetval(RetVal, IRB);
DEBUG(dbgs() << "Return: " << *Shadow << "\n" << *ShadowPtr << "\n");
IRB.CreateAlignedStore(Shadow, ShadowPtr, kShadowTLSAlignment);
- if (ClTrackOrigins)
+ if (MS.TrackOrigins)
IRB.CreateStore(getOrigin(RetVal), getOriginPtrForRetval(IRB));
}
}
ShadowPHINodes.push_back(&I);
setShadow(&I, IRB.CreatePHI(getShadowTy(&I), I.getNumIncomingValues(),
"_msphi_s"));
- if (ClTrackOrigins)
+ if (MS.TrackOrigins)
setOrigin(&I, IRB.CreatePHI(MS.OriginTy, I.getNumIncomingValues(),
"_msphi_o"));
}
void visitAllocaInst(AllocaInst &I) {
setShadow(&I, getCleanShadow(&I));
- if (!ClPoisonStack) return;
IRBuilder<> IRB(I.getNextNode());
uint64_t Size = MS.TD->getTypeAllocSize(I.getAllocatedType());
- if (ClPoisonStackWithCall) {
+ if (PoisonStack && ClPoisonStackWithCall) {
IRB.CreateCall2(MS.MsanPoisonStackFn,
IRB.CreatePointerCast(&I, IRB.getInt8PtrTy()),
ConstantInt::get(MS.IntptrTy, Size));
} else {
Value *ShadowBase = getShadowPtr(&I, Type::getInt8PtrTy(*MS.C), IRB);
- IRB.CreateMemSet(ShadowBase, IRB.getInt8(ClPoisonStackPattern),
- Size, I.getAlignment());
+ Value *PoisonValue = IRB.getInt8(PoisonStack ? ClPoisonStackPattern : 0);
+ IRB.CreateMemSet(ShadowBase, PoisonValue, Size, I.getAlignment());
}
- if (ClTrackOrigins) {
+ if (PoisonStack && MS.TrackOrigins) {
setOrigin(&I, getCleanOrigin());
SmallString<2048> StackDescriptionStorage;
raw_svector_ostream StackDescription(StackDescriptionStorage);
setShadow(&I, IRB.CreateSelect(I.getCondition(),
getShadow(I.getTrueValue()), getShadow(I.getFalseValue()),
"_msprop"));
- if (ClTrackOrigins)
- setOrigin(&I, IRB.CreateSelect(I.getCondition(),
+ if (MS.TrackOrigins) {
+ // Origins are always i32, so any vector conditions must be flattened.
+ // FIXME: consider tracking vector origins for app vectors?
+ Value *Cond = I.getCondition();
+ if (Cond->getType()->isVectorTy()) {
+ Value *ConvertedShadow = convertToShadowTyNoVec(Cond, IRB);
+ Cond = IRB.CreateICmpNE(ConvertedShadow,
+ getCleanShadow(ConvertedShadow), "_mso_select");
+ }
+ setOrigin(&I, IRB.CreateSelect(Cond,
getOrigin(I.getTrueValue()), getOrigin(I.getFalseValue())));
+ }
}
void visitLandingPadInst(LandingPadInst &I) {
// Unpoison the whole __va_list_tag.
// FIXME: magic ABI constants.
IRB.CreateMemSet(ShadowPtr, Constant::getNullValue(IRB.getInt8Ty()),
- /* size */24, /* alignment */16, false);
+ /* size */24, /* alignment */8, false);
}
void visitVACopyInst(VACopyInst &I) {
// Unpoison the whole __va_list_tag.
// FIXME: magic ABI constants.
IRB.CreateMemSet(ShadowPtr, Constant::getNullValue(IRB.getInt8Ty()),
- /* size */ 24, /* alignment */ 16, false);
+ /* size */24, /* alignment */8, false);
}
void finalizeInstrumentation() {
Value *OverflowArgAreaPtr = IRB.CreateLoad(OverflowArgAreaPtrPtr);
Value *OverflowArgAreaShadowPtr =
MSV.getShadowPtr(OverflowArgAreaPtr, IRB.getInt8Ty(), IRB);
- Value *SrcPtr =
- getShadowPtrForVAArgument(VAArgTLSCopy, IRB, AMD64FpEndOffset);
+ Value *SrcPtr = IRB.CreateConstGEP1_32(VAArgTLSCopy, AMD64FpEndOffset);
IRB.CreateMemCpy(OverflowArgAreaShadowPtr, SrcPtr, VAArgOverflowSize, 16);
}
}
};
-VarArgHelper* CreateVarArgHelper(Function &Func, MemorySanitizer &Msan,
+/// \brief A no-op implementation of VarArgHelper.
+struct VarArgNoOpHelper : public VarArgHelper {
+ VarArgNoOpHelper(Function &F, MemorySanitizer &MS,
+ MemorySanitizerVisitor &MSV) {}
+
+ void visitCallSite(CallSite &CS, IRBuilder<> &IRB) {}
+
+ void visitVAStartInst(VAStartInst &I) {}
+
+ void visitVACopyInst(VACopyInst &I) {}
+
+ void finalizeInstrumentation() {}
+};
+
+VarArgHelper *CreateVarArgHelper(Function &Func, MemorySanitizer &Msan,
MemorySanitizerVisitor &Visitor) {
- return new VarArgAMD64Helper(Func, Msan, Visitor);
+ // VarArg handling is only implemented on AMD64. False positives are possible
+ // on other platforms.
+ llvm::Triple TargetTriple(Func.getParent()->getTargetTriple());
+ if (TargetTriple.getArch() == llvm::Triple::x86_64)
+ return new VarArgAMD64Helper(Func, Msan, Visitor);
+ else
+ return new VarArgNoOpHelper(Func, Msan, Visitor);
}
} // namespace
// Clear out readonly/readnone attributes.
AttrBuilder B;
- B.addAttribute(Attributes::ReadOnly)
- .addAttribute(Attributes::ReadNone);
- F.removeAttribute(AttributeSet::FunctionIndex,
- Attributes::get(F.getContext(), B));
+ B.addAttribute(Attribute::ReadOnly)
+ .addAttribute(Attribute::ReadNone);
+ F.removeAttributes(AttributeSet::FunctionIndex,
+ AttributeSet::get(F.getContext(),
+ AttributeSet::FunctionIndex, B));
return Visitor.runOnFunction();
}