/// 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/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/BlackList.h"
+#include "llvm/Transforms/Utils/Local.h"
#include "llvm/Transforms/Utils/ModuleUtils.h"
-#include "llvm/Type.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 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));
cl::desc("propagate shadow through ICmpEQ and ICmpNE"),
cl::Hidden, cl::init(true));
+static cl::opt<bool> ClStoreCleanOrigin("msan-store-clean-origin",
+ cl::desc("store origin for clean (fully initialized) values"),
+ cl::Hidden, cl::init(false));
+
// This flag controls whether we check the shadow of the address
// operand of load or store. Such bugs are very rare, since load from
// a garbage address typically results in SEGV, but still happen
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);
/// MemorySanitizer: instrument the code in module to find
/// uninitialized reads.
class MemorySanitizer : public FunctionPass {
-public:
- MemorySanitizer() : FunctionPass(ID), TD(0) { }
+ public:
+ 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);
static char ID; // Pass identification, replacement for typeid.
-private:
+ private:
+ void initializeCallbacks(Module &M);
+
+ /// \brief Track origins (allocation points) of uninitialized values.
+ bool TrackOrigins;
+
DataLayout *TD;
LLVMContext *C;
Type *IntptrTy;
uint64_t OriginOffset;
/// \brief Branch weights for error reporting.
MDNode *ColdCallWeights;
+ /// \brief Branch weights for origin store.
+ MDNode *OriginStoreWeights;
+ /// \bried Path to blacklist file.
+ SmallString<64> BlacklistFile;
/// \brief The blacklist.
OwningPtr<BlackList> BL;
/// \brief An empty volatile inline asm that prevents callback merge.
"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.
GlobalValue::PrivateLinkage, StrConst, "");
}
-/// \brief Module-level initialization.
-///
-/// Obtains pointers to the required runtime library functions, and
-/// inserts a call to __msan_init to the module's constructor list.
-bool MemorySanitizer::doInitialization(Module &M) {
- TD = getAnalysisIfAvailable<DataLayout>();
- if (!TD)
- return false;
- BL.reset(new BlackList(ClBlackListFile));
- C = &(M.getContext());
- unsigned PtrSize = TD->getPointerSizeInBits(/* AddressSpace */0);
- switch (PtrSize) {
- case 64:
- ShadowMask = kShadowMask64;
- OriginOffset = kOriginOffset64;
- break;
- case 32:
- ShadowMask = kShadowMask32;
- OriginOffset = kOriginOffset32;
- break;
- default:
- report_fatal_error("unsupported pointer size");
- break;
- }
- IRBuilder<> IRB(*C);
- IntptrTy = IRB.getIntPtrTy(TD);
- OriginTy = IRB.getInt32Ty();
-
- ColdCallWeights = MDBuilder(*C).createBranchWeights(1, 1000);
-
- // Insert a call to __msan_init/__msan_track_origins into the module's CTORs.
- appendToGlobalCtors(M, cast<Function>(M.getOrInsertFunction(
- "__msan_init", IRB.getVoidTy(), NULL)), 0);
-
- new GlobalVariable(M, IRB.getInt32Ty(), true, GlobalValue::LinkOnceODRLinkage,
- IRB.getInt32(ClTrackOrigins), "__msan_track_origins");
+/// \brief Insert extern declaration of runtime-provided functions and globals.
+void MemorySanitizer::initializeCallbacks(Module &M) {
+ // Only do this once.
+ if (WarningFn)
+ return;
+ IRBuilder<> IRB(*C);
// Create the callback.
// FIXME: this function should have "Cold" calling conv,
// which is not yet implemented.
MsanPoisonStackFn = M.getOrInsertFunction(
"__msan_poison_stack", IRB.getVoidTy(), IRB.getInt8PtrTy(), IntptrTy, NULL);
MemmoveFn = M.getOrInsertFunction(
- "__msan_memmove", IRB.getInt8PtrTy(), IRB.getInt8PtrTy(), IRB.getInt8PtrTy(),
- IntptrTy, NULL);
+ "__msan_memmove", IRB.getInt8PtrTy(), IRB.getInt8PtrTy(),
+ IRB.getInt8PtrTy(), IntptrTy, NULL);
MemcpyFn = M.getOrInsertFunction(
"__msan_memcpy", IRB.getInt8PtrTy(), IRB.getInt8PtrTy(), IRB.getInt8PtrTy(),
IntptrTy, NULL);
EmptyAsm = InlineAsm::get(FunctionType::get(IRB.getVoidTy(), false),
StringRef(""), StringRef(""),
/*hasSideEffects=*/true);
+}
+
+/// \brief Module-level initialization.
+///
+/// inserts a call to __msan_init to the module's constructor list.
+bool MemorySanitizer::doInitialization(Module &M) {
+ TD = getAnalysisIfAvailable<DataLayout>();
+ if (!TD)
+ return false;
+ BL.reset(new BlackList(BlacklistFile));
+ C = &(M.getContext());
+ unsigned PtrSize = TD->getPointerSizeInBits(/* AddressSpace */0);
+ switch (PtrSize) {
+ case 64:
+ ShadowMask = kShadowMask64;
+ OriginOffset = kOriginOffset64;
+ break;
+ case 32:
+ ShadowMask = kShadowMask32;
+ OriginOffset = kOriginOffset32;
+ break;
+ default:
+ report_fatal_error("unsupported pointer size");
+ break;
+ }
+
+ IRBuilder<> IRB(*C);
+ IntptrTy = IRB.getIntPtrTy(TD);
+ OriginTy = IRB.getInt32Ty();
+
+ ColdCallWeights = MDBuilder(*C).createBranchWeights(1, 1000);
+ OriginStoreWeights = MDBuilder(*C).createBranchWeights(1, 1000);
+
+ // Insert a call to __msan_init/__msan_track_origins into the module's CTORs.
+ appendToGlobalCtors(M, cast<Function>(M.getOrInsertFunction(
+ "__msan_init", IRB.getVoidTy(), NULL)), 0);
+
+ new GlobalVariable(M, IRB.getInt32Ty(), true, GlobalValue::WeakODRLinkage,
+ IRB.getInt32(TrackOrigins), "__msan_track_origins");
+
return true;
}
// 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 AMD64GpEndOffset = 48; // AMD64 ABI Draft 0.99.6 p3.5.7
static const unsigned AMD64FpEndOffset = 176;
struct ShadowOriginAndInsertPoint {
ShadowOriginAndInsertPoint() : Shadow(0), Origin(0), OrigIns(0) { }
};
SmallVector<ShadowOriginAndInsertPoint, 16> InstrumentationList;
+ SmallVector<Instruction*, 16> StoreList;
MemorySanitizerVisitor(Function &F, MemorySanitizer &MS)
: F(F), MS(MS), VAHelper(CreateVarArgHelper(F, MS, *this)) {
<< F.getName() << "'\n");
}
+ void materializeStores() {
+ for (size_t i = 0, n = StoreList.size(); i < n; i++) {
+ StoreInst& I = *dyn_cast<StoreInst>(StoreList[i]);
+
+ IRBuilder<> IRB(&I);
+ Value *Val = I.getValueOperand();
+ Value *Addr = I.getPointerOperand();
+ Value *Shadow = getShadow(Val);
+ Value *ShadowPtr = getShadowPtr(Addr, Shadow->getType(), IRB);
+
+ StoreInst *NewSI =
+ IRB.CreateAlignedStore(Shadow, ShadowPtr, I.getAlignment());
+ DEBUG(dbgs() << " STORE: " << *NewSI << "\n");
+ (void)NewSI;
+
+ if (ClCheckAccessAddress)
+ insertCheck(Addr, &I);
+
+ if (MS.TrackOrigins) {
+ unsigned Alignment = std::max(kMinOriginAlignment, I.getAlignment());
+ if (ClStoreCleanOrigin || isa<StructType>(Shadow->getType())) {
+ IRB.CreateAlignedStore(getOrigin(Val), getOriginPtr(Addr, IRB),
+ Alignment);
+ } else {
+ Value *ConvertedShadow = convertToShadowTyNoVec(Shadow, IRB);
+
+ Constant *Cst = dyn_cast_or_null<Constant>(ConvertedShadow);
+ // TODO(eugenis): handle non-zero constant shadow by inserting an
+ // unconditional check (can not simply fail compilation as this could
+ // be in the dead code).
+ if (Cst)
+ continue;
+
+ Value *Cmp = IRB.CreateICmpNE(ConvertedShadow,
+ getCleanShadow(ConvertedShadow), "_mscmp");
+ Instruction *CheckTerm =
+ SplitBlockAndInsertIfThen(cast<Instruction>(Cmp), false,
+ MS.OriginStoreWeights);
+ IRBuilder<> IRBNew(CheckTerm);
+ IRBNew.CreateAlignedStore(getOrigin(Val), getOriginPtr(Addr, IRBNew),
+ Alignment);
+ }
+ }
+ }
+ }
+
void materializeChecks() {
for (size_t i = 0, n = InstrumentationList.size(); i < n; i++) {
Instruction *Shadow = InstrumentationList[i].Shadow;
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);
/// \brief Add MemorySanitizer instrumentation to a function.
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));
VAHelper->finalizeInstrumentation();
+ // Delayed instrumentation of StoreInst.
+ // This may add new checks to be inserted later.
+ materializeStores();
+
+ // Insert shadow value checks.
materializeChecks();
return true;
// 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;
}
DEBUG(dbgs() << " ARG: " << *AI << " ==> " <<
**ShadowPtr << "\n");
- if (ClTrackOrigins) {
+ if (MS.TrackOrigins) {
Value* OriginPtr = getOriginPtrForArgument(AI, EntryIRB, ArgOffset);
setOrigin(A, EntryIRB.CreateLoad(OriginPtr));
}
/// \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) {
ShadowOriginAndInsertPoint(Shadow, Origin, OrigIns));
}
- //------------------- Visitors.
+ // ------------------- Visitors.
/// \brief Instrument LoadInst
///
if (ClCheckAccessAddress)
insertCheck(I.getPointerOperand(), &I);
- if (ClTrackOrigins)
- setOrigin(&I, IRB.CreateAlignedLoad(getOriginPtr(Addr, IRB), I.getAlignment()));
+ if (MS.TrackOrigins) {
+ unsigned Alignment = std::max(kMinOriginAlignment, I.getAlignment());
+ setOrigin(&I, IRB.CreateAlignedLoad(getOriginPtr(Addr, IRB), Alignment));
+ }
}
/// \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) {
- IRBuilder<> IRB(&I);
- Value *Val = I.getValueOperand();
- Value *Addr = I.getPointerOperand();
- Value *Shadow = getShadow(Val);
- Value *ShadowPtr = getShadowPtr(Addr, Shadow->getType(), IRB);
-
- StoreInst *NewSI = 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)
- IRB.CreateAlignedStore(getOrigin(Val), getOriginPtr(Addr, IRB), I.getAlignment());
+ StoreList.push_back(&I);
}
// Vector manipulation.
setOriginForNaryOp(I);
}
- /// \brief Propagate origin for an instruction.
+ /// \brief Default propagation of shadow and/or origin.
///
- /// This is a general case of origin propagation. For an Nary operation,
- /// is set to the origin of an argument that is not entirely initialized.
- /// If there is more than one such arguments, the rightmost of them is picked.
- /// It does not matter which one is picked if all arguments are initialized.
- void setOriginForNaryOp(Instruction &I) {
- if (!ClTrackOrigins) return;
- IRBuilder<> IRB(&I);
- Value *Origin = getOrigin(&I, 0);
- for (unsigned Op = 1, n = I.getNumOperands(); Op < n; ++Op) {
- Value *S = convertToShadowTyNoVec(getShadow(&I, Op), IRB);
- Origin = IRB.CreateSelect(IRB.CreateICmpNE(S, getCleanShadow(S)),
- getOrigin(&I, Op), Origin);
+ /// This class implements the general case of shadow propagation, used in all
+ /// cases where we don't know and/or don't care about what the operation
+ /// actually does. It converts all input shadow values to a common type
+ /// (extending or truncating as necessary), and bitwise OR's them.
+ ///
+ /// This is much cheaper than inserting checks (i.e. requiring inputs to be
+ /// fully initialized), and less prone to false positives.
+ ///
+ /// This class also implements the general case of origin propagation. For a
+ /// Nary operation, result origin is set to the origin of an argument that is
+ /// not entirely initialized. If there is more than one such arguments, the
+ /// rightmost of them is picked. It does not matter which one is picked if all
+ /// arguments are initialized.
+ template <bool CombineShadow>
+ class Combiner {
+ Value *Shadow;
+ Value *Origin;
+ IRBuilder<> &IRB;
+ MemorySanitizerVisitor *MSV;
+
+ public:
+ Combiner(MemorySanitizerVisitor *MSV, IRBuilder<> &IRB) :
+ Shadow(0), Origin(0), IRB(IRB), MSV(MSV) {}
+
+ /// \brief Add a pair of shadow and origin values to the mix.
+ Combiner &Add(Value *OpShadow, Value *OpOrigin) {
+ if (CombineShadow) {
+ assert(OpShadow);
+ if (!Shadow)
+ Shadow = OpShadow;
+ else {
+ OpShadow = MSV->CreateShadowCast(IRB, OpShadow, Shadow->getType());
+ Shadow = IRB.CreateOr(Shadow, OpShadow, "_msprop");
+ }
+ }
+
+ if (MSV->MS.TrackOrigins) {
+ assert(OpOrigin);
+ if (!Origin) {
+ Origin = OpOrigin;
+ } else {
+ Value *FlatShadow = MSV->convertToShadowTyNoVec(OpShadow, IRB);
+ Value *Cond = IRB.CreateICmpNE(FlatShadow,
+ MSV->getCleanShadow(FlatShadow));
+ Origin = IRB.CreateSelect(Cond, OpOrigin, Origin);
+ }
+ }
+ return *this;
}
- setOrigin(&I, Origin);
- }
- /// \brief Propagate shadow for a binary operation.
- ///
- /// Shadow = Shadow0 | Shadow1, all 3 must have the same type.
- /// Bitwise OR is selected as an operation that will never lose even a bit of
- /// poison.
- void handleShadowOrBinary(Instruction &I) {
+ /// \brief Add an application value to the mix.
+ Combiner &Add(Value *V) {
+ Value *OpShadow = MSV->getShadow(V);
+ Value *OpOrigin = MSV->MS.TrackOrigins ? MSV->getOrigin(V) : 0;
+ return Add(OpShadow, OpOrigin);
+ }
+
+ /// \brief Set the current combined values as the given instruction's shadow
+ /// and origin.
+ void Done(Instruction *I) {
+ if (CombineShadow) {
+ assert(Shadow);
+ Shadow = MSV->CreateShadowCast(IRB, Shadow, MSV->getShadowTy(I));
+ MSV->setShadow(I, Shadow);
+ }
+ if (MSV->MS.TrackOrigins) {
+ assert(Origin);
+ MSV->setOrigin(I, Origin);
+ }
+ }
+ };
+
+ typedef Combiner<true> ShadowAndOriginCombiner;
+ typedef Combiner<false> OriginCombiner;
+
+ /// \brief Propagate origin for arbitrary operation.
+ void setOriginForNaryOp(Instruction &I) {
+ if (!MS.TrackOrigins) return;
IRBuilder<> IRB(&I);
- Value *Shadow0 = getShadow(&I, 0);
- Value *Shadow1 = getShadow(&I, 1);
- setShadow(&I, IRB.CreateOr(Shadow0, Shadow1, "_msprop"));
- setOriginForNaryOp(I);
+ OriginCombiner OC(this, IRB);
+ for (Instruction::op_iterator OI = I.op_begin(); OI != I.op_end(); ++OI)
+ OC.Add(OI->get());
+ OC.Done(&I);
+ }
+
+ 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();
+ }
+
+ /// \brief Cast between two shadow types, extending or truncating as
+ /// necessary.
+ Value *CreateShadowCast(IRBuilder<> &IRB, Value *V, Type *dstTy) {
+ Type *srcTy = V->getType();
+ if (dstTy->isIntegerTy() && srcTy->isIntegerTy())
+ return IRB.CreateIntCast(V, dstTy, false);
+ if (dstTy->isVectorTy() && srcTy->isVectorTy() &&
+ dstTy->getVectorNumElements() == srcTy->getVectorNumElements())
+ return IRB.CreateIntCast(V, dstTy, false);
+ size_t srcSizeInBits = VectorOrPrimitiveTypeSizeInBits(srcTy);
+ size_t dstSizeInBits = VectorOrPrimitiveTypeSizeInBits(dstTy);
+ Value *V1 = IRB.CreateBitCast(V, Type::getIntNTy(*MS.C, srcSizeInBits));
+ Value *V2 =
+ IRB.CreateIntCast(V1, Type::getIntNTy(*MS.C, dstSizeInBits), false);
+ return IRB.CreateBitCast(V2, dstTy);
+ // TODO: handle struct types.
}
/// \brief Propagate shadow for arbitrary operation.
- ///
- /// This is a general case of shadow propagation, used in all cases where we
- /// don't know and/or care about what the operation actually does.
- /// It converts all input shadow values to a common type (extending or
- /// truncating as necessary), and bitwise OR's them.
- ///
- /// This is much cheaper than inserting checks (i.e. requiring inputs to be
- /// fully initialized), and less prone to false positives.
- // FIXME: is the casting actually correct?
- // FIXME: merge this with handleShadowOrBinary.
void handleShadowOr(Instruction &I) {
IRBuilder<> IRB(&I);
- Value *Shadow = getShadow(&I, 0);
- for (unsigned Op = 1, n = I.getNumOperands(); Op < n; ++Op)
- Shadow = IRB.CreateOr(
- Shadow, IRB.CreateIntCast(getShadow(&I, Op), Shadow->getType(), false),
- "_msprop");
- Shadow = IRB.CreateIntCast(Shadow, getShadowTy(&I), false);
- setShadow(&I, Shadow);
- setOriginForNaryOp(I);
+ ShadowAndOriginCombiner SC(this, IRB);
+ for (Instruction::op_iterator OI = I.op_begin(); OI != I.op_end(); ++OI)
+ SC.Add(OI->get());
+ SC.Done(&I);
}
- void visitFAdd(BinaryOperator &I) { handleShadowOrBinary(I); }
- void visitFSub(BinaryOperator &I) { handleShadowOrBinary(I); }
- void visitFMul(BinaryOperator &I) { handleShadowOrBinary(I); }
- void visitAdd(BinaryOperator &I) { handleShadowOrBinary(I); }
- void visitSub(BinaryOperator &I) { handleShadowOrBinary(I); }
- void visitXor(BinaryOperator &I) { handleShadowOrBinary(I); }
- void visitMul(BinaryOperator &I) { handleShadowOrBinary(I); }
+ void visitFAdd(BinaryOperator &I) { handleShadowOr(I); }
+ void visitFSub(BinaryOperator &I) { handleShadowOr(I); }
+ void visitFMul(BinaryOperator &I) { handleShadowOr(I); }
+ void visitAdd(BinaryOperator &I) { handleShadowOr(I); }
+ void visitSub(BinaryOperator &I) { handleShadowOr(I); }
+ void visitXor(BinaryOperator &I) { handleShadowOr(I); }
+ void visitMul(BinaryOperator &I) { handleShadowOr(I); }
void handleDiv(Instruction &I) {
IRBuilder<> IRB(&I);
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
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);
+ 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"));
+
+ if (ClCheckAccessAddress)
+ insertCheck(Addr, &I);
+
+ if (MS.TrackOrigins)
+ setOrigin(&I, IRB.CreateLoad(getOriginPtr(Addr, IRB)));
+ 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);
+ Type *OpType = Op->getType();
+ Function *BswapFunc = Intrinsic::getDeclaration(
+ F.getParent(), Intrinsic::bswap, ArrayRef<Type*>(&OpType, 1));
+ setShadow(&I, IRB.CreateCall(BswapFunc, getShadow(Op)));
+ setOrigin(&I, getOrigin(Op));
+ }
+
+ void visitIntrinsicInst(IntrinsicInst &I) {
+ switch (I.getIntrinsicID()) {
+ case llvm::Intrinsic::bswap:
+ handleBswap(I);
+ break;
+ default:
+ if (!handleUnknownIntrinsic(I))
+ visitInstruction(I);
+ break;
+ }
+ }
+
void visitCallSite(CallSite CS) {
Instruction &I = *CS.getInstruction();
assert((CS.isCall() || CS.isInvoke()) && "Unknown type of CallSite");
// will get propagated to a void RetVal.
if (Call->isTailCall() && Call->getType() != Call->getParent()->getType())
Call->setTailCall(false);
- if (isa<IntrinsicInst>(&I)) {
- // All intrinsics we care about are handled in corresponding visit*
- // methods. Add checks for the arguments, mark retval as clean.
- visitInstruction(I);
- return;
+
+ assert(!isa<IntrinsicInst>(&I) && "intrinsics are handled elsewhere");
+
+ // We are going to insert code that relies on the fact that the callee
+ // will become a non-readonly function after it is instrumented by us. To
+ // prevent this code from being optimized out, mark that function
+ // non-readonly in advance.
+ if (Function *Func = Call->getCalledFunction()) {
+ // Clear out readonly/readnone attributes.
+ AttrBuilder B;
+ B.addAttribute(Attribute::ReadOnly)
+ .addAttribute(Attribute::ReadNone);
+ Func->removeAttribute(AttributeSet::FunctionIndex,
+ Attribute::get(Func->getContext(), 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());
Size, Alignment);
} else {
Size = MS.TD->getTypeAllocSize(A->getType());
- Store = IRB.CreateStore(ArgShadow, ArgShadowBase);
+ Store = IRB.CreateAlignedStore(ArgShadow, ArgShadowBase,
+ kShadowTLSAlignment);
}
- if (ClTrackOrigins)
+ if (MS.TrackOrigins)
IRB.CreateStore(getOrigin(A),
getOriginPtrForArgument(A, IRB, ArgOffset));
assert(Size != 0 && Store != 0);
IRBuilder<> IRBBefore(&I);
// Untill we have full dynamic coverage, make sure the retval shadow is 0.
Value *Base = getShadowPtrForRetval(&I, IRBBefore);
- IRBBefore.CreateStore(getCleanShadow(&I), Base);
+ IRBBefore.CreateAlignedStore(getCleanShadow(&I), Base, kShadowTLSAlignment);
Instruction *NextInsn = 0;
if (CS.isCall()) {
NextInsn = I.getNextNode();
"Could not find insertion point for retval shadow load");
}
IRBuilder<> IRBAfter(NextInsn);
- setShadow(&I, IRBAfter.CreateLoad(getShadowPtrForRetval(&I, IRBAfter),
- "_msret"));
- if (ClTrackOrigins)
+ Value *RetvalShadow =
+ IRBAfter.CreateAlignedLoad(getShadowPtrForRetval(&I, IRBAfter),
+ kShadowTLSAlignment, "_msret");
+ setShadow(&I, RetvalShadow);
+ if (MS.TrackOrigins)
setOrigin(&I, IRBAfter.CreateLoad(getOriginPtrForRetval(IRBAfter)));
}
Value *Shadow = getShadow(RetVal);
Value *ShadowPtr = getShadowPtrForRetval(RetVal, IRB);
DEBUG(dbgs() << "Return: " << *Shadow << "\n" << *ShadowPtr << "\n");
- IRB.CreateStore(Shadow, ShadowPtr);
- if (ClTrackOrigins)
+ IRB.CreateAlignedStore(Shadow, ShadowPtr, kShadowTLSAlignment);
+ 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"));
}
Size, I.getAlignment());
}
- if (ClTrackOrigins) {
+ if (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) {
struct VarArgAMD64Helper : public VarArgHelper {
// 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 AMD64GpEndOffset = 48; // AMD64 ABI Draft 0.99.6 p3.5.7
static const unsigned AMD64FpEndOffset = 176;
Function &F;
Base = getShadowPtrForVAArgument(A, IRB, OverflowOffset);
OverflowOffset += DataLayout::RoundUpAlignment(ArgSize, 8);
}
- IRB.CreateStore(MSV.getShadow(A), Base);
+ IRB.CreateAlignedStore(MSV.getShadow(A), Base, kShadowTLSAlignment);
}
Constant *OverflowSize =
ConstantInt::get(IRB.getInt64Ty(), OverflowOffset - AMD64FpEndOffset);
// 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() {
// Clear out readonly/readnone attributes.
AttrBuilder B;
- B.addAttribute(Attributes::ReadOnly)
- .addAttribute(Attributes::ReadNone);
- F.removeAttribute(AttrListPtr::FunctionIndex,
- Attributes::get(F.getContext(), B));
+ B.addAttribute(Attribute::ReadOnly)
+ .addAttribute(Attribute::ReadNone);
+ F.removeAttribute(AttributeSet::FunctionIndex,
+ Attribute::get(F.getContext(), B));
return Visitor.runOnFunction();
}
projects / oota-llvm.git / blobdiff