diff options
Diffstat (limited to 'lib/Transforms/Instrumentation')
-rw-r--r-- | lib/Transforms/Instrumentation/AddressSanitizer.cpp | 291 | ||||
-rw-r--r-- | lib/Transforms/Instrumentation/BoundsChecking.cpp | 209 | ||||
-rw-r--r-- | lib/Transforms/Instrumentation/CMakeLists.txt | 3 | ||||
-rw-r--r-- | lib/Transforms/Instrumentation/GCOVProfiling.cpp | 120 | ||||
-rw-r--r-- | lib/Transforms/Instrumentation/Instrumentation.cpp | 5 | ||||
-rw-r--r-- | lib/Transforms/Instrumentation/PathProfiling.cpp | 2 | ||||
-rw-r--r-- | lib/Transforms/Instrumentation/ThreadSanitizer.cpp | 267 |
7 files changed, 703 insertions, 194 deletions
diff --git a/lib/Transforms/Instrumentation/AddressSanitizer.cpp b/lib/Transforms/Instrumentation/AddressSanitizer.cpp index b43b9e5..bf35eac 100644 --- a/lib/Transforms/Instrumentation/AddressSanitizer.cpp +++ b/lib/Transforms/Instrumentation/AddressSanitizer.cpp @@ -16,20 +16,23 @@ #define DEBUG_TYPE "asan" #include "FunctionBlackList.h" +#include "llvm/Function.h" +#include "llvm/IRBuilder.h" +#include "llvm/InlineAsm.h" +#include "llvm/IntrinsicInst.h" +#include "llvm/LLVMContext.h" +#include "llvm/Module.h" +#include "llvm/Type.h" #include "llvm/ADT/ArrayRef.h" #include "llvm/ADT/OwningPtr.h" #include "llvm/ADT/SmallSet.h" #include "llvm/ADT/SmallString.h" #include "llvm/ADT/SmallVector.h" #include "llvm/ADT/StringExtras.h" -#include "llvm/Function.h" -#include "llvm/IntrinsicInst.h" -#include "llvm/LLVMContext.h" -#include "llvm/Module.h" +#include "llvm/ADT/Triple.h" #include "llvm/Support/CommandLine.h" #include "llvm/Support/DataTypes.h" #include "llvm/Support/Debug.h" -#include "llvm/Support/IRBuilder.h" #include "llvm/Support/raw_ostream.h" #include "llvm/Support/system_error.h" #include "llvm/Target/TargetData.h" @@ -37,7 +40,6 @@ #include "llvm/Transforms/Instrumentation.h" #include "llvm/Transforms/Utils/BasicBlockUtils.h" #include "llvm/Transforms/Utils/ModuleUtils.h" -#include "llvm/Type.h" #include <string> #include <algorithm> @@ -47,6 +49,7 @@ using namespace llvm; static const uint64_t kDefaultShadowScale = 3; static const uint64_t kDefaultShadowOffset32 = 1ULL << 29; static const uint64_t kDefaultShadowOffset64 = 1ULL << 44; +static const uint64_t kDefaultShadowOffsetAndroid = 0; static const size_t kMaxStackMallocSize = 1 << 16; // 64K static const uintptr_t kCurrentStackFrameMagic = 0x41B58AB3; @@ -70,6 +73,9 @@ static const int kAsanStackMidRedzoneMagic = 0xf2; static const int kAsanStackRightRedzoneMagic = 0xf3; static const int kAsanStackPartialRedzoneMagic = 0xf4; +// Accesses sizes are powers of two: 1, 2, 4, 8, 16. +static const size_t kNumberOfAccessSizes = 5; + // Command-line flags. // This flag may need to be replaced with -f[no-]asan-reads. @@ -77,6 +83,17 @@ static cl::opt<bool> ClInstrumentReads("asan-instrument-reads", cl::desc("instrument read instructions"), cl::Hidden, cl::init(true)); static cl::opt<bool> ClInstrumentWrites("asan-instrument-writes", cl::desc("instrument write instructions"), cl::Hidden, cl::init(true)); +static cl::opt<bool> ClInstrumentAtomics("asan-instrument-atomics", + cl::desc("instrument atomic instructions (rmw, cmpxchg)"), + cl::Hidden, cl::init(true)); +// This flag limits the number of instructions to be instrumented +// in any given BB. Normally, this should be set to unlimited (INT_MAX), +// but due to http://llvm.org/bugs/show_bug.cgi?id=12652 we temporary +// set it to 10000. +static cl::opt<int> ClMaxInsnsToInstrumentPerBB("asan-max-ins-per-bb", + cl::init(10000), + cl::desc("maximal number of instructions to instrument in any given BB"), + cl::Hidden); // This flag may need to be replaced with -f[no]asan-stack. static cl::opt<bool> ClStack("asan-stack", cl::desc("Handle stack memory"), cl::Hidden, cl::init(true)); @@ -125,18 +142,29 @@ static cl::opt<int> ClDebugMax("asan-debug-max", cl::desc("Debug man inst"), namespace { +/// An object of this type is created while instrumenting every function. +struct AsanFunctionContext { + AsanFunctionContext(Function &Function) : F(Function) { } + + Function &F; +}; + /// AddressSanitizer: instrument the code in module to find memory bugs. struct AddressSanitizer : public ModulePass { AddressSanitizer(); virtual const char *getPassName() const; - void instrumentMop(Instruction *I); - void instrumentAddress(Instruction *OrigIns, IRBuilder<> &IRB, + void instrumentMop(AsanFunctionContext &AFC, Instruction *I); + void instrumentAddress(AsanFunctionContext &AFC, + Instruction *OrigIns, IRBuilder<> &IRB, Value *Addr, uint32_t TypeSize, bool IsWrite); - Instruction *generateCrashCode(IRBuilder<> &IRB, Value *Addr, - bool IsWrite, uint32_t TypeSize); - bool instrumentMemIntrinsic(MemIntrinsic *MI); - void instrumentMemIntrinsicParam(Instruction *OrigIns, Value *Addr, - Value *Size, + Value *createSlowPathCmp(IRBuilder<> &IRB, Value *AddrLong, + Value *ShadowValue, uint32_t TypeSize); + Instruction *generateCrashCode(BasicBlock *BB, Value *Addr, Value *PC, + bool IsWrite, size_t AccessSizeIndex); + bool instrumentMemIntrinsic(AsanFunctionContext &AFC, MemIntrinsic *MI); + void instrumentMemIntrinsicParam(AsanFunctionContext &AFC, + Instruction *OrigIns, Value *Addr, + Value *Size, Instruction *InsertBefore, bool IsWrite); Value *memToShadow(Value *Shadow, IRBuilder<> &IRB); bool handleFunction(Module &M, Function &F); @@ -144,7 +172,6 @@ struct AddressSanitizer : public ModulePass { bool poisonStackInFunction(Module &M, Function &F); virtual bool runOnModule(Module &M); bool insertGlobalRedzones(Module &M); - BranchInst *splitBlockAndInsertIfThen(Instruction *SplitBefore, Value *Cmp); static char ID; // Pass identification, replacement for typeid private: @@ -163,11 +190,11 @@ struct AddressSanitizer : public ModulePass { return getAlignedSize(SizeInBytes); } + Function *checkInterfaceFunction(Constant *FuncOrBitcast); void PoisonStack(const ArrayRef<AllocaInst*> &AllocaVec, IRBuilder<> IRB, Value *ShadowBase, bool DoPoison); bool LooksLikeCodeInBug11395(Instruction *I); - Module *CurrentModule; LLVMContext *C; TargetData *TD; uint64_t MappingOffset; @@ -180,7 +207,11 @@ struct AddressSanitizer : public ModulePass { Function *AsanInitFunction; Instruction *CtorInsertBefore; OwningPtr<FunctionBlackList> BL; + // This array is indexed by AccessIsWrite and log2(AccessSize). + Function *AsanErrorCallback[2][kNumberOfAccessSizes]; + InlineAsm *EmptyAsm; }; + } // namespace char AddressSanitizer::ID = 0; @@ -196,6 +227,12 @@ const char *AddressSanitizer::getPassName() const { return "AddressSanitizer"; } +static size_t TypeSizeToSizeIndex(uint32_t TypeSize) { + size_t Res = CountTrailingZeros_32(TypeSize / 8); + assert(Res < kNumberOfAccessSizes); + return Res; +} + // Create a constant for Str so that we can pass it to the run-time lib. static GlobalVariable *createPrivateGlobalForString(Module &M, StringRef Str) { Constant *StrConst = ConstantDataArray::getString(M.getContext(), Str); @@ -206,29 +243,32 @@ static GlobalVariable *createPrivateGlobalForString(Module &M, StringRef Str) { // Split the basic block and insert an if-then code. // Before: // Head -// SplitBefore +// Cmp // Tail // After: // Head // if (Cmp) -// NewBasicBlock -// SplitBefore +// ThenBlock // Tail // -// Returns the NewBasicBlock's terminator. -BranchInst *AddressSanitizer::splitBlockAndInsertIfThen( - Instruction *SplitBefore, Value *Cmp) { +// If ThenBlock is zero, a new block is created and its terminator is returned. +// Otherwize 0 is returned. +static BranchInst *splitBlockAndInsertIfThen(Value *Cmp, + BasicBlock *ThenBlock = 0) { + Instruction *SplitBefore = cast<Instruction>(Cmp)->getNextNode(); BasicBlock *Head = SplitBefore->getParent(); BasicBlock *Tail = Head->splitBasicBlock(SplitBefore); TerminatorInst *HeadOldTerm = Head->getTerminator(); - BasicBlock *NewBasicBlock = - BasicBlock::Create(*C, "", Head->getParent()); - BranchInst *HeadNewTerm = BranchInst::Create(/*ifTrue*/NewBasicBlock, - /*ifFalse*/Tail, - Cmp); + BranchInst *CheckTerm = 0; + if (!ThenBlock) { + LLVMContext &C = Head->getParent()->getParent()->getContext(); + ThenBlock = BasicBlock::Create(C, "", Head->getParent(), Tail); + CheckTerm = BranchInst::Create(Tail, ThenBlock); + } + BranchInst *HeadNewTerm = + BranchInst::Create(/*ifTrue*/ThenBlock, /*ifFalse*/Tail, Cmp); ReplaceInstWithInst(HeadOldTerm, HeadNewTerm); - BranchInst *CheckTerm = BranchInst::Create(Tail, NewBasicBlock); return CheckTerm; } @@ -242,12 +282,13 @@ Value *AddressSanitizer::memToShadow(Value *Shadow, IRBuilder<> &IRB) { MappingOffset)); } -void AddressSanitizer::instrumentMemIntrinsicParam(Instruction *OrigIns, +void AddressSanitizer::instrumentMemIntrinsicParam( + AsanFunctionContext &AFC, Instruction *OrigIns, Value *Addr, Value *Size, Instruction *InsertBefore, bool IsWrite) { // Check the first byte. { IRBuilder<> IRB(InsertBefore); - instrumentAddress(OrigIns, IRB, Addr, 8, IsWrite); + instrumentAddress(AFC, OrigIns, IRB, Addr, 8, IsWrite); } // Check the last byte. { @@ -257,15 +298,16 @@ void AddressSanitizer::instrumentMemIntrinsicParam(Instruction *OrigIns, SizeMinusOne = IRB.CreateIntCast(SizeMinusOne, IntptrTy, false); Value *AddrLong = IRB.CreatePointerCast(Addr, IntptrTy); Value *AddrPlusSizeMinisOne = IRB.CreateAdd(AddrLong, SizeMinusOne); - instrumentAddress(OrigIns, IRB, AddrPlusSizeMinisOne, 8, IsWrite); + instrumentAddress(AFC, OrigIns, IRB, AddrPlusSizeMinisOne, 8, IsWrite); } } // Instrument memset/memmove/memcpy -bool AddressSanitizer::instrumentMemIntrinsic(MemIntrinsic *MI) { +bool AddressSanitizer::instrumentMemIntrinsic(AsanFunctionContext &AFC, + MemIntrinsic *MI) { Value *Dst = MI->getDest(); MemTransferInst *MemTran = dyn_cast<MemTransferInst>(MI); - Value *Src = MemTran ? MemTran->getSource() : NULL; + Value *Src = MemTran ? MemTran->getSource() : 0; Value *Length = MI->getLength(); Constant *ConstLength = dyn_cast<Constant>(Length); @@ -277,26 +319,46 @@ bool AddressSanitizer::instrumentMemIntrinsic(MemIntrinsic *MI) { IRBuilder<> IRB(InsertBefore); Value *Cmp = IRB.CreateICmpNE(Length, - Constant::getNullValue(Length->getType())); - InsertBefore = splitBlockAndInsertIfThen(InsertBefore, Cmp); + Constant::getNullValue(Length->getType())); + InsertBefore = splitBlockAndInsertIfThen(Cmp); } - instrumentMemIntrinsicParam(MI, Dst, Length, InsertBefore, true); + instrumentMemIntrinsicParam(AFC, MI, Dst, Length, InsertBefore, true); if (Src) - instrumentMemIntrinsicParam(MI, Src, Length, InsertBefore, false); + instrumentMemIntrinsicParam(AFC, MI, Src, Length, InsertBefore, false); return true; } -static Value *getLDSTOperand(Instruction *I) { +// If I is an interesting memory access, return the PointerOperand +// and set IsWrite. Otherwise return NULL. +static Value *isInterestingMemoryAccess(Instruction *I, bool *IsWrite) { if (LoadInst *LI = dyn_cast<LoadInst>(I)) { + if (!ClInstrumentReads) return NULL; + *IsWrite = false; return LI->getPointerOperand(); } - return cast<StoreInst>(*I).getPointerOperand(); + if (StoreInst *SI = dyn_cast<StoreInst>(I)) { + if (!ClInstrumentWrites) return NULL; + *IsWrite = true; + return SI->getPointerOperand(); + } + if (AtomicRMWInst *RMW = dyn_cast<AtomicRMWInst>(I)) { + if (!ClInstrumentAtomics) return NULL; + *IsWrite = true; + return RMW->getPointerOperand(); + } + if (AtomicCmpXchgInst *XCHG = dyn_cast<AtomicCmpXchgInst>(I)) { + if (!ClInstrumentAtomics) return NULL; + *IsWrite = true; + return XCHG->getPointerOperand(); + } + return NULL; } -void AddressSanitizer::instrumentMop(Instruction *I) { - int IsWrite = isa<StoreInst>(*I); - Value *Addr = getLDSTOperand(I); +void AddressSanitizer::instrumentMop(AsanFunctionContext &AFC, Instruction *I) { + bool IsWrite; + Value *Addr = isInterestingMemoryAccess(I, &IsWrite); + assert(Addr); if (ClOpt && ClOptGlobals && isa<GlobalVariable>(Addr)) { // We are accessing a global scalar variable. Nothing to catch here. return; @@ -314,22 +376,57 @@ void AddressSanitizer::instrumentMop(Instruction *I) { } IRBuilder<> IRB(I); - instrumentAddress(I, IRB, Addr, TypeSize, IsWrite); + instrumentAddress(AFC, I, IRB, Addr, TypeSize, IsWrite); +} + +// Validate the result of Module::getOrInsertFunction called for an interface +// function of AddressSanitizer. If the instrumented module defines a function +// with the same name, their prototypes must match, otherwise +// getOrInsertFunction returns a bitcast. +Function *AddressSanitizer::checkInterfaceFunction(Constant *FuncOrBitcast) { + if (isa<Function>(FuncOrBitcast)) return cast<Function>(FuncOrBitcast); + FuncOrBitcast->dump(); + report_fatal_error("trying to redefine an AddressSanitizer " + "interface function"); } Instruction *AddressSanitizer::generateCrashCode( - IRBuilder<> &IRB, Value *Addr, bool IsWrite, uint32_t TypeSize) { - // IsWrite and TypeSize are encoded in the function name. - std::string FunctionName = std::string(kAsanReportErrorTemplate) + - (IsWrite ? "store" : "load") + itostr(TypeSize / 8); - Value *ReportWarningFunc = CurrentModule->getOrInsertFunction( - FunctionName, IRB.getVoidTy(), IntptrTy, NULL); - CallInst *Call = IRB.CreateCall(ReportWarningFunc, Addr); - Call->setDoesNotReturn(); + BasicBlock *BB, Value *Addr, Value *PC, + bool IsWrite, size_t AccessSizeIndex) { + IRBuilder<> IRB(BB->getFirstNonPHI()); + CallInst *Call; + if (PC) + Call = IRB.CreateCall2(AsanErrorCallback[IsWrite][AccessSizeIndex], + Addr, PC); + else + Call = IRB.CreateCall(AsanErrorCallback[IsWrite][AccessSizeIndex], Addr); + // We don't do Call->setDoesNotReturn() because the BB already has + // UnreachableInst at the end. + // This EmptyAsm is required to avoid callback merge. + IRB.CreateCall(EmptyAsm); return Call; } -void AddressSanitizer::instrumentAddress(Instruction *OrigIns, +Value *AddressSanitizer::createSlowPathCmp(IRBuilder<> &IRB, Value *AddrLong, + Value *ShadowValue, + uint32_t TypeSize) { + size_t Granularity = 1 << MappingScale; + // Addr & (Granularity - 1) + Value *LastAccessedByte = IRB.CreateAnd( + AddrLong, ConstantInt::get(IntptrTy, Granularity - 1)); + // (Addr & (Granularity - 1)) + size - 1 + if (TypeSize / 8 > 1) + LastAccessedByte = IRB.CreateAdd( + LastAccessedByte, ConstantInt::get(IntptrTy, TypeSize / 8 - 1)); + // (uint8_t) ((Addr & (Granularity-1)) + size - 1) + LastAccessedByte = IRB.CreateIntCast( + LastAccessedByte, IRB.getInt8Ty(), false); + // ((uint8_t) ((Addr & (Granularity-1)) + size - 1)) >= ShadowValue + return IRB.CreateICmpSGE(LastAccessedByte, ShadowValue); +} + +void AddressSanitizer::instrumentAddress(AsanFunctionContext &AFC, + Instruction *OrigIns, IRBuilder<> &IRB, Value *Addr, uint32_t TypeSize, bool IsWrite) { Value *AddrLong = IRB.CreatePointerCast(Addr, IntptrTy); @@ -344,31 +441,25 @@ void AddressSanitizer::instrumentAddress(Instruction *OrigIns, Value *Cmp = IRB.CreateICmpNE(ShadowValue, CmpVal); - Instruction *CheckTerm = splitBlockAndInsertIfThen( - cast<Instruction>(Cmp)->getNextNode(), Cmp); - IRBuilder<> IRB2(CheckTerm); + BasicBlock *CrashBlock = BasicBlock::Create(*C, "crash_bb", &AFC.F); + new UnreachableInst(*C, CrashBlock); + size_t AccessSizeIndex = TypeSizeToSizeIndex(TypeSize); + Instruction *Crash = + generateCrashCode(CrashBlock, AddrLong, 0, IsWrite, AccessSizeIndex); + Crash->setDebugLoc(OrigIns->getDebugLoc()); size_t Granularity = 1 << MappingScale; if (TypeSize < 8 * Granularity) { - // Addr & (Granularity - 1) - Value *Lower3Bits = IRB2.CreateAnd( - AddrLong, ConstantInt::get(IntptrTy, Granularity - 1)); - // (Addr & (Granularity - 1)) + size - 1 - Value *LastAccessedByte = IRB2.CreateAdd( - Lower3Bits, ConstantInt::get(IntptrTy, TypeSize / 8 - 1)); - // (uint8_t) ((Addr & (Granularity-1)) + size - 1) - LastAccessedByte = IRB2.CreateIntCast( - LastAccessedByte, IRB.getInt8Ty(), false); - // ((uint8_t) ((Addr & (Granularity-1)) + size - 1)) >= ShadowValue - Value *Cmp2 = IRB2.CreateICmpSGE(LastAccessedByte, ShadowValue); - - CheckTerm = splitBlockAndInsertIfThen(CheckTerm, Cmp2); - } - - IRBuilder<> IRB1(CheckTerm); - Instruction *Crash = generateCrashCode(IRB1, AddrLong, IsWrite, TypeSize); - Crash->setDebugLoc(OrigIns->getDebugLoc()); - ReplaceInstWithInst(CheckTerm, new UnreachableInst(*C)); + BranchInst *CheckTerm = splitBlockAndInsertIfThen(Cmp); + assert(CheckTerm->isUnconditional()); + BasicBlock *NextBB = CheckTerm->getSuccessor(0); + IRB.SetInsertPoint(CheckTerm); + Value *Cmp2 = createSlowPathCmp(IRB, AddrLong, ShadowValue, TypeSize); + BranchInst *NewTerm = BranchInst::Create(CrashBlock, NextBB, Cmp2); + ReplaceInstWithInst(CheckTerm, NewTerm); + } else { + splitBlockAndInsertIfThen(Cmp, CrashBlock); + } } // This function replaces all global variables with new variables that have @@ -473,7 +564,7 @@ bool AddressSanitizer::insertGlobalRedzones(Module &M) { // Create a new global variable with enough space for a redzone. GlobalVariable *NewGlobal = new GlobalVariable( M, NewTy, G->isConstant(), G->getLinkage(), - NewInitializer, "", G, G->isThreadLocal()); + NewInitializer, "", G, G->getThreadLocalMode()); NewGlobal->copyAttributesFrom(G); NewGlobal->setAlignment(RedzoneSize); @@ -501,7 +592,7 @@ bool AddressSanitizer::insertGlobalRedzones(Module &M) { M, ArrayOfGlobalStructTy, false, GlobalVariable::PrivateLinkage, ConstantArray::get(ArrayOfGlobalStructTy, Initializers), ""); - Function *AsanRegisterGlobals = cast<Function>(M.getOrInsertFunction( + Function *AsanRegisterGlobals = checkInterfaceFunction(M.getOrInsertFunction( kAsanRegisterGlobalsName, IRB.getVoidTy(), IntptrTy, IntptrTy, NULL)); AsanRegisterGlobals->setLinkage(Function::ExternalLinkage); @@ -516,8 +607,10 @@ bool AddressSanitizer::insertGlobalRedzones(Module &M) { GlobalValue::InternalLinkage, kAsanModuleDtorName, &M); BasicBlock *AsanDtorBB = BasicBlock::Create(*C, "", AsanDtorFunction); IRBuilder<> IRB_Dtor(ReturnInst::Create(*C, AsanDtorBB)); - Function *AsanUnregisterGlobals = cast<Function>(M.getOrInsertFunction( - kAsanUnregisterGlobalsName, IRB.getVoidTy(), IntptrTy, IntptrTy, NULL)); + Function *AsanUnregisterGlobals = + checkInterfaceFunction(M.getOrInsertFunction( + kAsanUnregisterGlobalsName, + IRB.getVoidTy(), IntptrTy, IntptrTy, NULL)); AsanUnregisterGlobals->setLinkage(Function::ExternalLinkage); IRB_Dtor.CreateCall2(AsanUnregisterGlobals, @@ -537,7 +630,6 @@ bool AddressSanitizer::runOnModule(Module &M) { return false; BL.reset(new FunctionBlackList(ClBlackListFile)); - CurrentModule = &M; C = &(M.getContext()); LongSize = TD->getPointerSizeInBits(); IntptrTy = Type::getIntNTy(*C, LongSize); @@ -551,13 +643,33 @@ bool AddressSanitizer::runOnModule(Module &M) { // call __asan_init in the module ctor. IRBuilder<> IRB(CtorInsertBefore); - AsanInitFunction = cast<Function>( + AsanInitFunction = checkInterfaceFunction( M.getOrInsertFunction(kAsanInitName, IRB.getVoidTy(), NULL)); AsanInitFunction->setLinkage(Function::ExternalLinkage); IRB.CreateCall(AsanInitFunction); - MappingOffset = LongSize == 32 - ? kDefaultShadowOffset32 : kDefaultShadowOffset64; + // Create __asan_report* callbacks. + for (size_t AccessIsWrite = 0; AccessIsWrite <= 1; AccessIsWrite++) { + for (size_t AccessSizeIndex = 0; AccessSizeIndex < kNumberOfAccessSizes; + AccessSizeIndex++) { + // IsWrite and TypeSize are encoded in the function name. + std::string FunctionName = std::string(kAsanReportErrorTemplate) + + (AccessIsWrite ? "store" : "load") + itostr(1 << AccessSizeIndex); + // If we are merging crash callbacks, they have two parameters. + AsanErrorCallback[AccessIsWrite][AccessSizeIndex] = cast<Function>( + M.getOrInsertFunction(FunctionName, IRB.getVoidTy(), IntptrTy, NULL)); + } + } + // We insert an empty inline asm after __asan_report* to avoid callback merge. + EmptyAsm = InlineAsm::get(FunctionType::get(IRB.getVoidTy(), false), + StringRef(""), StringRef(""), + /*hasSideEffects=*/true); + + llvm::Triple targetTriple(M.getTargetTriple()); + bool isAndroid = targetTriple.getEnvironment() == llvm::Triple::ANDROIDEABI; + + MappingOffset = isAndroid ? kDefaultShadowOffsetAndroid : + (LongSize == 32 ? kDefaultShadowOffset32 : kDefaultShadowOffset64); if (ClMappingOffsetLog >= 0) { if (ClMappingOffsetLog == 0) { // special case @@ -640,17 +752,17 @@ bool AddressSanitizer::handleFunction(Module &M, Function &F) { SmallSet<Value*, 16> TempsToInstrument; SmallVector<Instruction*, 16> ToInstrument; SmallVector<Instruction*, 8> NoReturnCalls; + bool IsWrite; // Fill the set of memory operations to instrument. for (Function::iterator FI = F.begin(), FE = F.end(); FI != FE; ++FI) { TempsToInstrument.clear(); + int NumInsnsPerBB = 0; for (BasicBlock::iterator BI = FI->begin(), BE = FI->end(); BI != BE; ++BI) { if (LooksLikeCodeInBug11395(BI)) return false; - if ((isa<LoadInst>(BI) && ClInstrumentReads) || - (isa<StoreInst>(BI) && ClInstrumentWrites)) { - Value *Addr = getLDSTOperand(BI); + if (Value *Addr = isInterestingMemoryAccess(BI, &IsWrite)) { if (ClOpt && ClOptSameTemp) { if (!TempsToInstrument.insert(Addr)) continue; // We've seen this temp in the current BB. @@ -668,19 +780,24 @@ bool AddressSanitizer::handleFunction(Module &M, Function &F) { continue; } ToInstrument.push_back(BI); + NumInsnsPerBB++; + if (NumInsnsPerBB >= ClMaxInsnsToInstrumentPerBB) + break; } } + AsanFunctionContext AFC(F); + // Instrument. int NumInstrumented = 0; for (size_t i = 0, n = ToInstrument.size(); i != n; i++) { Instruction *Inst = ToInstrument[i]; if (ClDebugMin < 0 || ClDebugMax < 0 || (NumInstrumented >= ClDebugMin && NumInstrumented <= ClDebugMax)) { - if (isa<StoreInst>(Inst) || isa<LoadInst>(Inst)) - instrumentMop(Inst); + if (isInterestingMemoryAccess(Inst, &IsWrite)) + instrumentMop(AFC, Inst); else - instrumentMemIntrinsic(cast<MemIntrinsic>(Inst)); + instrumentMemIntrinsic(AFC, cast<MemIntrinsic>(Inst)); } NumInstrumented++; } diff --git a/lib/Transforms/Instrumentation/BoundsChecking.cpp b/lib/Transforms/Instrumentation/BoundsChecking.cpp new file mode 100644 index 0000000..09e0f14 --- /dev/null +++ b/lib/Transforms/Instrumentation/BoundsChecking.cpp @@ -0,0 +1,209 @@ +//===- BoundsChecking.cpp - Instrumentation for run-time bounds checking --===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file implements a pass that instruments the code to perform run-time +// bounds checking on loads, stores, and other memory intrinsics. +// +//===----------------------------------------------------------------------===// + +#define DEBUG_TYPE "bounds-checking" +#include "llvm/IRBuilder.h" +#include "llvm/Intrinsics.h" +#include "llvm/Pass.h" +#include "llvm/ADT/Statistic.h" +#include "llvm/Analysis/MemoryBuiltins.h" +#include "llvm/Support/CommandLine.h" +#include "llvm/Support/Debug.h" +#include "llvm/Support/InstIterator.h" +#include "llvm/Support/TargetFolder.h" +#include "llvm/Support/raw_ostream.h" +#include "llvm/Target/TargetData.h" +#include "llvm/Transforms/Instrumentation.h" +using namespace llvm; + +static cl::opt<bool> SingleTrapBB("bounds-checking-single-trap", + cl::desc("Use one trap block per function")); + +STATISTIC(ChecksAdded, "Bounds checks added"); +STATISTIC(ChecksSkipped, "Bounds checks skipped"); +STATISTIC(ChecksUnable, "Bounds checks unable to add"); + +typedef IRBuilder<true, TargetFolder> BuilderTy; + +namespace { + struct BoundsChecking : public FunctionPass { + static char ID; + + BoundsChecking(unsigned _Penalty = 5) : FunctionPass(ID), Penalty(_Penalty){ + initializeBoundsCheckingPass(*PassRegistry::getPassRegistry()); + } + + virtual bool runOnFunction(Function &F); + + virtual void getAnalysisUsage(AnalysisUsage &AU) const { + AU.addRequired<TargetData>(); + } + + private: + const TargetData *TD; + ObjectSizeOffsetEvaluator *ObjSizeEval; + BuilderTy *Builder; + Instruction *Inst; + BasicBlock *TrapBB; + unsigned Penalty; + + BasicBlock *getTrapBB(); + void emitBranchToTrap(Value *Cmp = 0); + bool computeAllocSize(Value *Ptr, APInt &Offset, Value* &OffsetValue, + APInt &Size, Value* &SizeValue); + bool instrument(Value *Ptr, Value *Val); + }; +} + +char BoundsChecking::ID = 0; +INITIALIZE_PASS(BoundsChecking, "bounds-checking", "Run-time bounds checking", + false, false) + + +/// getTrapBB - create a basic block that traps. All overflowing conditions +/// branch to this block. There's only one trap block per function. +BasicBlock *BoundsChecking::getTrapBB() { + if (TrapBB && SingleTrapBB) + return TrapBB; + + Function *Fn = Inst->getParent()->getParent(); + BasicBlock::iterator PrevInsertPoint = Builder->GetInsertPoint(); + TrapBB = BasicBlock::Create(Fn->getContext(), "trap", Fn); + Builder->SetInsertPoint(TrapBB); + + llvm::Value *F = Intrinsic::getDeclaration(Fn->getParent(), Intrinsic::trap); + CallInst *TrapCall = Builder->CreateCall(F); + TrapCall->setDoesNotReturn(); + TrapCall->setDoesNotThrow(); + TrapCall->setDebugLoc(Inst->getDebugLoc()); + Builder->CreateUnreachable(); + + Builder->SetInsertPoint(PrevInsertPoint); + return TrapBB; +} + + +/// emitBranchToTrap - emit a branch instruction to a trap block. +/// If Cmp is non-null, perform a jump only if its value evaluates to true. +void BoundsChecking::emitBranchToTrap(Value *Cmp) { + // check if the comparison is always false + ConstantInt *C = dyn_cast_or_null<ConstantInt>(Cmp); + if (C) { + ++ChecksSkipped; + if (!C->getZExtValue()) + return; + else + Cmp = 0; // unconditional branch + } + + Instruction *Inst = Builder->GetInsertPoint(); + BasicBlock *OldBB = Inst->getParent(); + BasicBlock *Cont = OldBB->splitBasicBlock(Inst); + OldBB->getTerminator()->eraseFromParent(); + + if (Cmp) + BranchInst::Create(getTrapBB(), Cont, Cmp, OldBB); + else + BranchInst::Create(getTrapBB(), OldBB); +} + + +/// instrument - adds run-time bounds checks to memory accessing instructions. +/// Ptr is the pointer that will be read/written, and InstVal is either the +/// result from the load or the value being stored. It is used to determine the +/// size of memory block that is touched. +/// Returns true if any change was made to the IR, false otherwise. +bool BoundsChecking::instrument(Value *Ptr, Value *InstVal) { + uint64_t NeededSize = TD->getTypeStoreSize(InstVal->getType()); + DEBUG(dbgs() << "Instrument " << *Ptr << " for " << Twine(NeededSize) + << " bytes\n"); + + SizeOffsetEvalType SizeOffset = ObjSizeEval->compute(Ptr); + + if (!ObjSizeEval->bothKnown(SizeOffset)) { + ++ChecksUnable; + return false; + } + + Value *Size = SizeOffset.first; + Value *Offset = SizeOffset.second; + ConstantInt *SizeCI = dyn_cast<ConstantInt>(Size); + + IntegerType *IntTy = TD->getIntPtrType(Inst->getContext()); + Value *NeededSizeVal = ConstantInt::get(IntTy, NeededSize); + + // three checks are required to ensure safety: + // . Offset >= 0 (since the offset is given from the base ptr) + // . Size >= Offset (unsigned) + // . Size - Offset >= NeededSize (unsigned) + // + // optimization: if Size >= 0 (signed), skip 1st check + // FIXME: add NSW/NUW here? -- we dont care if the subtraction overflows + Value *ObjSize = Builder->CreateSub(Size, Offset); + Value *Cmp2 = Builder->CreateICmpULT(Size, Offset); + Value *Cmp3 = Builder->CreateICmpULT(ObjSize, NeededSizeVal); + Value *Or = Builder->CreateOr(Cmp2, Cmp3); + if (!SizeCI || SizeCI->getValue().slt(0)) { + Value *Cmp1 = Builder->CreateICmpSLT(Offset, ConstantInt::get(IntTy, 0)); + Or = Builder->CreateOr(Cmp1, Or); + } + emitBranchToTrap(Or); + + ++ChecksAdded; + return true; +} + +bool BoundsChecking::runOnFunction(Function &F) { + TD = &getAnalysis<TargetData>(); + + TrapBB = 0; + BuilderTy TheBuilder(F.getContext(), TargetFolder(TD)); + Builder = &TheBuilder; + ObjectSizeOffsetEvaluator TheObjSizeEval(TD, F.getContext()); + ObjSizeEval = &TheObjSizeEval; + + // check HANDLE_MEMORY_INST in include/llvm/Instruction.def for memory + // touching instructions + std::vector<Instruction*> WorkList; + for (inst_iterator i = inst_begin(F), e = inst_end(F); i != e; ++i) { + Instruction *I = &*i; + if (isa<LoadInst>(I) || isa<StoreInst>(I) || isa<AtomicCmpXchgInst>(I) || + isa<AtomicRMWInst>(I)) + WorkList.push_back(I); + } + + bool MadeChange = false; + for (std::vector<Instruction*>::iterator i = WorkList.begin(), + e = WorkList.end(); i != e; ++i) { + Inst = *i; + + Builder->SetInsertPoint(Inst); + if (LoadInst *LI = dyn_cast<LoadInst>(Inst)) { + MadeChange |= instrument(LI->getPointerOperand(), LI); + } else if (StoreInst *SI = dyn_cast<StoreInst>(Inst)) { + MadeChange |= instrument(SI->getPointerOperand(), SI->getValueOperand()); + } else if (AtomicCmpXchgInst *AI = dyn_cast<AtomicCmpXchgInst>(Inst)) { + MadeChange |= instrument(AI->getPointerOperand(),AI->getCompareOperand()); + } else if (AtomicRMWInst *AI = dyn_cast<AtomicRMWInst>(Inst)) { + MadeChange |= instrument(AI->getPointerOperand(), AI->getValOperand()); + } else { + llvm_unreachable("unknown Instruction type"); + } + } + return MadeChange; +} + +FunctionPass *llvm::createBoundsCheckingPass(unsigned Penalty) { + return new BoundsChecking(Penalty); +} diff --git a/lib/Transforms/Instrumentation/CMakeLists.txt b/lib/Transforms/Instrumentation/CMakeLists.txt index e4c8cf1..00de882 100644 --- a/lib/Transforms/Instrumentation/CMakeLists.txt +++ b/lib/Transforms/Instrumentation/CMakeLists.txt @@ -1,5 +1,6 @@ add_llvm_library(LLVMInstrumentation AddressSanitizer.cpp + BoundsChecking.cpp EdgeProfiling.cpp FunctionBlackList.cpp GCOVProfiling.cpp @@ -9,3 +10,5 @@ add_llvm_library(LLVMInstrumentation ProfilingUtils.cpp ThreadSanitizer.cpp ) + +add_dependencies(LLVMInstrumentation intrinsics_gen) diff --git a/lib/Transforms/Instrumentation/GCOVProfiling.cpp b/lib/Transforms/Instrumentation/GCOVProfiling.cpp index 96e5d5b..264a6a6 100644 --- a/lib/Transforms/Instrumentation/GCOVProfiling.cpp +++ b/lib/Transforms/Instrumentation/GCOVProfiling.cpp @@ -18,22 +18,23 @@ #include "ProfilingUtils.h" #include "llvm/Transforms/Instrumentation.h" -#include "llvm/Analysis/DebugInfo.h" +#include "llvm/DebugInfo.h" +#include "llvm/IRBuilder.h" +#include "llvm/Instructions.h" #include "llvm/Module.h" #include "llvm/Pass.h" -#include "llvm/Instructions.h" -#include "llvm/Support/raw_ostream.h" -#include "llvm/Support/Debug.h" -#include "llvm/Support/DebugLoc.h" -#include "llvm/Support/InstIterator.h" -#include "llvm/Support/IRBuilder.h" -#include "llvm/Support/PathV2.h" #include "llvm/ADT/DenseMap.h" -#include "llvm/ADT/Statistic.h" #include "llvm/ADT/STLExtras.h" +#include "llvm/ADT/Statistic.h" #include "llvm/ADT/StringExtras.h" #include "llvm/ADT/StringMap.h" #include "llvm/ADT/UniqueVector.h" +#include "llvm/Support/Debug.h" +#include "llvm/Support/DebugLoc.h" +#include "llvm/Support/InstIterator.h" +#include "llvm/Support/PathV2.h" +#include "llvm/Support/raw_ostream.h" +#include "llvm/Transforms/Utils/ModuleUtils.h" #include <string> #include <utility> using namespace llvm; @@ -57,7 +58,6 @@ namespace { virtual const char *getPassName() const { return "GCOV Profiler"; } - private: bool runOnModule(Module &M); @@ -90,6 +90,7 @@ namespace { // list. void insertCounterWriteout(SmallVector<std::pair<GlobalVariable *, MDNode *>, 8> &); + void insertIndirectCounterIncrement(); std::string mangleName(DICompileUnit CU, std::string NewStem); @@ -421,6 +422,7 @@ bool GCOVProfiler::emitProfileArcs() { if (!CU_Nodes) return false; bool Result = false; + bool InsertIndCounterIncrCode = false; for (unsigned i = 0, e = CU_Nodes->getNumOperands(); i != e; ++i) { DICompileUnit CU(CU_Nodes->getOperand(i)); DIArray SPs = CU.getSubprograms(); @@ -446,7 +448,7 @@ bool GCOVProfiler::emitProfileArcs() { new GlobalVariable(*M, CounterTy, false, GlobalValue::InternalLinkage, Constant::getNullValue(CounterTy), - "__llvm_gcov_ctr", 0, false, 0); + "__llvm_gcov_ctr"); CountersBySP.push_back(std::make_pair(Counters, (MDNode*)SP)); UniqueVector<BasicBlock *> ComplexEdgePreds; @@ -507,15 +509,21 @@ bool GCOVProfiler::emitProfileArcs() { Value *CounterPtrArray = Builder.CreateConstInBoundsGEP2_64(EdgeTable, 0, i * ComplexEdgePreds.size()); + + // Build code to increment the counter. + InsertIndCounterIncrCode = true; Builder.CreateCall2(getIncrementIndirectCounterFunc(), EdgeState, CounterPtrArray); - // clear the predecessor number - Builder.CreateStore(ConstantInt::get(Int32Ty, 0xffffffff), EdgeState); } } } + insertCounterWriteout(CountersBySP); } + + if (InsertIndCounterIncrCode) + insertIndirectCounterIncrement(); + return Result; } @@ -574,13 +582,14 @@ Constant *GCOVProfiler::getStartFileFunc() { } Constant *GCOVProfiler::getIncrementIndirectCounterFunc() { + Type *Int32Ty = Type::getInt32Ty(*Ctx); + Type *Int64Ty = Type::getInt64Ty(*Ctx); Type *Args[] = { - Type::getInt32PtrTy(*Ctx), // uint32_t *predecessor - Type::getInt64PtrTy(*Ctx)->getPointerTo(), // uint64_t **state_table_row + Int32Ty->getPointerTo(), // uint32_t *predecessor + Int64Ty->getPointerTo()->getPointerTo() // uint64_t **counters }; - FunctionType *FTy = FunctionType::get(Type::getVoidTy(*Ctx), - Args, false); - return M->getOrInsertFunction("llvm_gcda_increment_indirect_counter", FTy); + FunctionType *FTy = FunctionType::get(Type::getVoidTy(*Ctx), Args, false); + return M->getOrInsertFunction("__llvm_gcov_indirect_counter_increment", FTy); } Constant *GCOVProfiler::getEmitFunctionFunc() { @@ -588,8 +597,7 @@ Constant *GCOVProfiler::getEmitFunctionFunc() { Type::getInt32Ty(*Ctx), // uint32_t ident Type::getInt8PtrTy(*Ctx), // const char *function_name }; - FunctionType *FTy = FunctionType::get(Type::getVoidTy(*Ctx), - Args, false); + FunctionType *FTy = FunctionType::get(Type::getVoidTy(*Ctx), Args, false); return M->getOrInsertFunction("llvm_gcda_emit_function", FTy); } @@ -665,5 +673,75 @@ void GCOVProfiler::insertCounterWriteout( } Builder.CreateRetVoid(); - InsertProfilingShutdownCall(WriteoutF, M); + // Create a small bit of code that registers the "__llvm_gcov_writeout" + // function to be executed at exit. + FunctionType *FTy = FunctionType::get(Type::getVoidTy(*Ctx), false); + Function *F = Function::Create(FTy, GlobalValue::InternalLinkage, + "__llvm_gcov_init", M); + F->setUnnamedAddr(true); + F->setLinkage(GlobalValue::InternalLinkage); + F->addFnAttr(Attribute::NoInline); + + BB = BasicBlock::Create(*Ctx, "entry", F); + Builder.SetInsertPoint(BB); + + FTy = FunctionType::get(Type::getInt32Ty(*Ctx), + PointerType::get(FTy, 0), false); + Constant *AtExitFn = M->getOrInsertFunction("atexit", FTy); + Builder.CreateCall(AtExitFn, WriteoutF); + Builder.CreateRetVoid(); + + appendToGlobalCtors(*M, F, 0); +} + +void GCOVProfiler::insertIndirectCounterIncrement() { + Function *Fn = + cast<Function>(GCOVProfiler::getIncrementIndirectCounterFunc()); + Fn->setUnnamedAddr(true); + Fn->setLinkage(GlobalValue::InternalLinkage); + Fn->addFnAttr(Attribute::NoInline); + + Type *Int32Ty = Type::getInt32Ty(*Ctx); + Type *Int64Ty = Type::getInt64Ty(*Ctx); + Constant *NegOne = ConstantInt::get(Int32Ty, 0xffffffff); + + // Create basic blocks for function. + BasicBlock *BB = BasicBlock::Create(*Ctx, "entry", Fn); + IRBuilder<> Builder(BB); + + BasicBlock *PredNotNegOne = BasicBlock::Create(*Ctx, "", Fn); + BasicBlock *CounterEnd = BasicBlock::Create(*Ctx, "", Fn); + BasicBlock *Exit = BasicBlock::Create(*Ctx, "exit", Fn); + + // uint32_t pred = *predecessor; + // if (pred == 0xffffffff) return; + Argument *Arg = Fn->arg_begin(); + Arg->setName("predecessor"); + Value *Pred = Builder.CreateLoad(Arg, "pred"); + Value *Cond = Builder.CreateICmpEQ(Pred, NegOne); + BranchInst::Create(Exit, PredNotNegOne, Cond, BB); + + Builder.SetInsertPoint(PredNotNegOne); + + // uint64_t *counter = counters[pred]; + // if (!counter) return; + Value *ZExtPred = Builder.CreateZExt(Pred, Int64Ty); + Arg = llvm::next(Fn->arg_begin()); + Arg->setName("counters"); + Value *GEP = Builder.CreateGEP(Arg, ZExtPred); + Value *Counter = Builder.CreateLoad(GEP, "counter"); + Cond = Builder.CreateICmpEQ(Counter, + Constant::getNullValue(Int64Ty->getPointerTo())); + Builder.CreateCondBr(Cond, Exit, CounterEnd); + + // ++*counter; + Builder.SetInsertPoint(CounterEnd); + Value *Add = Builder.CreateAdd(Builder.CreateLoad(Counter), + ConstantInt::get(Int64Ty, 1)); + Builder.CreateStore(Add, Counter); + Builder.CreateBr(Exit); + + // Fill in the exit block. + Builder.SetInsertPoint(Exit); + Builder.CreateRetVoid(); } diff --git a/lib/Transforms/Instrumentation/Instrumentation.cpp b/lib/Transforms/Instrumentation/Instrumentation.cpp index c7266e2..1e0b4a3 100644 --- a/lib/Transforms/Instrumentation/Instrumentation.cpp +++ b/lib/Transforms/Instrumentation/Instrumentation.cpp @@ -20,11 +20,12 @@ using namespace llvm; /// initializeInstrumentation - Initialize all passes in the TransformUtils /// library. void llvm::initializeInstrumentation(PassRegistry &Registry) { + initializeAddressSanitizerPass(Registry); + initializeBoundsCheckingPass(Registry); initializeEdgeProfilerPass(Registry); + initializeGCOVProfilerPass(Registry); initializeOptimalEdgeProfilerPass(Registry); initializePathProfilerPass(Registry); - initializeGCOVProfilerPass(Registry); - initializeAddressSanitizerPass(Registry); initializeThreadSanitizerPass(Registry); } diff --git a/lib/Transforms/Instrumentation/PathProfiling.cpp b/lib/Transforms/Instrumentation/PathProfiling.cpp index b214796..cc27146 100644 --- a/lib/Transforms/Instrumentation/PathProfiling.cpp +++ b/lib/Transforms/Instrumentation/PathProfiling.cpp @@ -55,11 +55,11 @@ #include "llvm/LLVMContext.h" #include "llvm/Module.h" #include "llvm/Pass.h" +#include "llvm/TypeBuilder.h" #include "llvm/Support/Compiler.h" #include "llvm/Support/CFG.h" #include "llvm/Support/CommandLine.h" #include "llvm/Support/Debug.h" -#include "llvm/Support/TypeBuilder.h" #include "llvm/Support/raw_ostream.h" #include "llvm/Transforms/Utils/BasicBlockUtils.h" #include "llvm/Transforms/Instrumentation.h" diff --git a/lib/Transforms/Instrumentation/ThreadSanitizer.cpp b/lib/Transforms/Instrumentation/ThreadSanitizer.cpp index 8bb337e..dc0fa71 100644 --- a/lib/Transforms/Instrumentation/ThreadSanitizer.cpp +++ b/lib/Transforms/Instrumentation/ThreadSanitizer.cpp @@ -22,73 +22,73 @@ #define DEBUG_TYPE "tsan" #include "FunctionBlackList.h" -#include "llvm/ADT/SmallSet.h" -#include "llvm/ADT/SmallString.h" -#include "llvm/ADT/SmallVector.h" -#include "llvm/ADT/StringExtras.h" -#include "llvm/Intrinsics.h" #include "llvm/Function.h" +#include "llvm/IRBuilder.h" +#include "llvm/Intrinsics.h" #include "llvm/LLVMContext.h" #include "llvm/Metadata.h" #include "llvm/Module.h" +#include "llvm/Type.h" +#include "llvm/ADT/SmallSet.h" +#include "llvm/ADT/SmallString.h" +#include "llvm/ADT/SmallVector.h" +#include "llvm/ADT/Statistic.h" +#include "llvm/ADT/StringExtras.h" #include "llvm/Support/CommandLine.h" #include "llvm/Support/Debug.h" -#include "llvm/Support/IRBuilder.h" #include "llvm/Support/MathExtras.h" #include "llvm/Support/raw_ostream.h" #include "llvm/Target/TargetData.h" #include "llvm/Transforms/Instrumentation.h" +#include "llvm/Transforms/Utils/BasicBlockUtils.h" #include "llvm/Transforms/Utils/ModuleUtils.h" -#include "llvm/Type.h" using namespace llvm; static cl::opt<std::string> ClBlackListFile("tsan-blacklist", cl::desc("Blacklist file"), cl::Hidden); -static cl::opt<bool> ClPrintStats("tsan-print-stats", - cl::desc("Print ThreadSanitizer instrumentation stats"), cl::Hidden); +STATISTIC(NumInstrumentedReads, "Number of instrumented reads"); +STATISTIC(NumInstrumentedWrites, "Number of instrumented writes"); +STATISTIC(NumOmittedReadsBeforeWrite, + "Number of reads ignored due to following writes"); +STATISTIC(NumAccessesWithBadSize, "Number of accesses with bad size"); +STATISTIC(NumInstrumentedVtableWrites, "Number of vtable ptr writes"); +STATISTIC(NumOmittedReadsFromConstantGlobals, + "Number of reads from constant globals"); +STATISTIC(NumOmittedReadsFromVtable, "Number of vtable reads"); namespace { -// Stats counters for ThreadSanitizer instrumentation. -struct ThreadSanitizerStats { - size_t NumInstrumentedReads; - size_t NumInstrumentedWrites; - size_t NumOmittedReadsBeforeWrite; - size_t NumAccessesWithBadSize; - size_t NumInstrumentedVtableWrites; - size_t NumOmittedReadsFromConstantGlobals; - size_t NumOmittedReadsFromVtable; -}; - /// ThreadSanitizer: instrument the code in module to find races. struct ThreadSanitizer : public FunctionPass { ThreadSanitizer(); + const char *getPassName() const; bool runOnFunction(Function &F); bool doInitialization(Module &M); - bool doFinalization(Module &M); - bool instrumentLoadOrStore(Instruction *I); static char ID; // Pass identification, replacement for typeid. private: - void choseInstructionsToInstrument(SmallVectorImpl<Instruction*> &Local, - SmallVectorImpl<Instruction*> &All); + bool instrumentLoadOrStore(Instruction *I); + bool instrumentAtomic(Instruction *I); + void chooseInstructionsToInstrument(SmallVectorImpl<Instruction*> &Local, + SmallVectorImpl<Instruction*> &All); bool addrPointsToConstantData(Value *Addr); + int getMemoryAccessFuncIndex(Value *Addr); TargetData *TD; OwningPtr<FunctionBlackList> BL; + IntegerType *OrdTy; // Callbacks to run-time library are computed in doInitialization. - Value *TsanFuncEntry; - Value *TsanFuncExit; + Function *TsanFuncEntry; + Function *TsanFuncExit; // Accesses sizes are powers of two: 1, 2, 4, 8, 16. static const size_t kNumberOfAccessSizes = 5; - Value *TsanRead[kNumberOfAccessSizes]; - Value *TsanWrite[kNumberOfAccessSizes]; - Value *TsanVptrUpdate; - - // Stats are modified w/o synchronization. - ThreadSanitizerStats stats; + Function *TsanRead[kNumberOfAccessSizes]; + Function *TsanWrite[kNumberOfAccessSizes]; + Function *TsanAtomicLoad[kNumberOfAccessSizes]; + Function *TsanAtomicStore[kNumberOfAccessSizes]; + Function *TsanVptrUpdate; }; } // namespace @@ -97,6 +97,10 @@ INITIALIZE_PASS(ThreadSanitizer, "tsan", "ThreadSanitizer: detects data races.", false, false) +const char *ThreadSanitizer::getPassName() const { + return "ThreadSanitizer"; +} + ThreadSanitizer::ThreadSanitizer() : FunctionPass(ID), TD(NULL) { @@ -106,12 +110,18 @@ FunctionPass *llvm::createThreadSanitizerPass() { return new ThreadSanitizer(); } +static Function *checkInterfaceFunction(Constant *FuncOrBitcast) { + if (Function *F = dyn_cast<Function>(FuncOrBitcast)) + return F; + FuncOrBitcast->dump(); + report_fatal_error("ThreadSanitizer interface function redefined"); +} + bool ThreadSanitizer::doInitialization(Module &M) { TD = getAnalysisIfAvailable<TargetData>(); if (!TD) return false; BL.reset(new FunctionBlackList(ClBlackListFile)); - memset(&stats, 0, sizeof(stats)); // Always insert a call to __tsan_init into the module's CTORs. IRBuilder<> IRB(M.getContext()); @@ -120,38 +130,38 @@ bool ThreadSanitizer::doInitialization(Module &M) { appendToGlobalCtors(M, cast<Function>(TsanInit), 0); // Initialize the callbacks. - TsanFuncEntry = M.getOrInsertFunction("__tsan_func_entry", IRB.getVoidTy(), - IRB.getInt8PtrTy(), NULL); - TsanFuncExit = M.getOrInsertFunction("__tsan_func_exit", IRB.getVoidTy(), - NULL); + TsanFuncEntry = checkInterfaceFunction(M.getOrInsertFunction( + "__tsan_func_entry", IRB.getVoidTy(), IRB.getInt8PtrTy(), NULL)); + TsanFuncExit = checkInterfaceFunction(M.getOrInsertFunction( + "__tsan_func_exit", IRB.getVoidTy(), NULL)); + OrdTy = IRB.getInt32Ty(); for (size_t i = 0; i < kNumberOfAccessSizes; ++i) { - SmallString<32> ReadName("__tsan_read"); - ReadName += itostr(1 << i); - TsanRead[i] = M.getOrInsertFunction(ReadName, IRB.getVoidTy(), - IRB.getInt8PtrTy(), NULL); - SmallString<32> WriteName("__tsan_write"); - WriteName += itostr(1 << i); - TsanWrite[i] = M.getOrInsertFunction(WriteName, IRB.getVoidTy(), - IRB.getInt8PtrTy(), NULL); - } - TsanVptrUpdate = M.getOrInsertFunction("__tsan_vptr_update", IRB.getVoidTy(), - IRB.getInt8PtrTy(), IRB.getInt8PtrTy(), - NULL); - return true; -} + const size_t ByteSize = 1 << i; + const size_t BitSize = ByteSize * 8; + SmallString<32> ReadName("__tsan_read" + itostr(ByteSize)); + TsanRead[i] = checkInterfaceFunction(M.getOrInsertFunction( + ReadName, IRB.getVoidTy(), IRB.getInt8PtrTy(), NULL)); -bool ThreadSanitizer::doFinalization(Module &M) { - if (ClPrintStats) { - errs() << "ThreadSanitizerStats " << M.getModuleIdentifier() - << ": wr " << stats.NumInstrumentedWrites - << "; rd " << stats.NumInstrumentedReads - << "; vt " << stats.NumInstrumentedVtableWrites - << "; bs " << stats.NumAccessesWithBadSize - << "; rbw " << stats.NumOmittedReadsBeforeWrite - << "; rcg " << stats.NumOmittedReadsFromConstantGlobals - << "; rvt " << stats.NumOmittedReadsFromVtable - << "\n"; + SmallString<32> WriteName("__tsan_write" + itostr(ByteSize)); + TsanWrite[i] = checkInterfaceFunction(M.getOrInsertFunction( + WriteName, IRB.getVoidTy(), IRB.getInt8PtrTy(), NULL)); + + Type *Ty = Type::getIntNTy(M.getContext(), BitSize); + Type *PtrTy = Ty->getPointerTo(); + SmallString<32> AtomicLoadName("__tsan_atomic" + itostr(BitSize) + + "_load"); + TsanAtomicLoad[i] = checkInterfaceFunction(M.getOrInsertFunction( + AtomicLoadName, Ty, PtrTy, OrdTy, NULL)); + + SmallString<32> AtomicStoreName("__tsan_atomic" + itostr(BitSize) + + "_store"); + TsanAtomicStore[i] = checkInterfaceFunction(M.getOrInsertFunction( + AtomicStoreName, IRB.getVoidTy(), PtrTy, Ty, OrdTy, + NULL)); } + TsanVptrUpdate = checkInterfaceFunction(M.getOrInsertFunction( + "__tsan_vptr_update", IRB.getVoidTy(), IRB.getInt8PtrTy(), + IRB.getInt8PtrTy(), NULL)); return true; } @@ -173,13 +183,13 @@ bool ThreadSanitizer::addrPointsToConstantData(Value *Addr) { if (GlobalVariable *GV = dyn_cast<GlobalVariable>(Addr)) { if (GV->isConstant()) { // Reads from constant globals can not race with any writes. - stats.NumOmittedReadsFromConstantGlobals++; + NumOmittedReadsFromConstantGlobals++; return true; } } else if(LoadInst *L = dyn_cast<LoadInst>(Addr)) { if (isVtableAccess(L)) { // Reads from a vtable pointer can not race with any writes. - stats.NumOmittedReadsFromVtable++; + NumOmittedReadsFromVtable++; return true; } } @@ -197,7 +207,7 @@ bool ThreadSanitizer::addrPointsToConstantData(Value *Addr) { // // 'Local' is a vector of insns within the same BB (no calls between). // 'All' is a vector of insns that will be instrumented. -void ThreadSanitizer::choseInstructionsToInstrument( +void ThreadSanitizer::chooseInstructionsToInstrument( SmallVectorImpl<Instruction*> &Local, SmallVectorImpl<Instruction*> &All) { SmallSet<Value*, 8> WriteTargets; @@ -212,7 +222,7 @@ void ThreadSanitizer::choseInstructionsToInstrument( Value *Addr = Load->getPointerOperand(); if (WriteTargets.count(Addr)) { // We will write to this temp, so no reason to analyze the read. - stats.NumOmittedReadsBeforeWrite++; + NumOmittedReadsBeforeWrite++; continue; } if (addrPointsToConstantData(Addr)) { @@ -225,12 +235,27 @@ void ThreadSanitizer::choseInstructionsToInstrument( Local.clear(); } +static bool isAtomic(Instruction *I) { + if (LoadInst *LI = dyn_cast<LoadInst>(I)) + return LI->isAtomic() && LI->getSynchScope() == CrossThread; + if (StoreInst *SI = dyn_cast<StoreInst>(I)) + return SI->isAtomic() && SI->getSynchScope() == CrossThread; + if (isa<AtomicRMWInst>(I)) + return true; + if (isa<AtomicCmpXchgInst>(I)) + return true; + if (FenceInst *FI = dyn_cast<FenceInst>(I)) + return FI->getSynchScope() == CrossThread; + return false; +} + bool ThreadSanitizer::runOnFunction(Function &F) { if (!TD) return false; if (BL->isIn(F)) return false; SmallVector<Instruction*, 8> RetVec; SmallVector<Instruction*, 8> AllLoadsAndStores; SmallVector<Instruction*, 8> LocalLoadsAndStores; + SmallVector<Instruction*, 8> AtomicAccesses; bool Res = false; bool HasCalls = false; @@ -240,16 +265,18 @@ bool ThreadSanitizer::runOnFunction(Function &F) { BasicBlock &BB = *FI; for (BasicBlock::iterator BI = BB.begin(), BE = BB.end(); BI != BE; ++BI) { - if (isa<LoadInst>(BI) || isa<StoreInst>(BI)) + if (isAtomic(BI)) + AtomicAccesses.push_back(BI); + else if (isa<LoadInst>(BI) || isa<StoreInst>(BI)) LocalLoadsAndStores.push_back(BI); else if (isa<ReturnInst>(BI)) RetVec.push_back(BI); else if (isa<CallInst>(BI) || isa<InvokeInst>(BI)) { HasCalls = true; - choseInstructionsToInstrument(LocalLoadsAndStores, AllLoadsAndStores); + chooseInstructionsToInstrument(LocalLoadsAndStores, AllLoadsAndStores); } } - choseInstructionsToInstrument(LocalLoadsAndStores, AllLoadsAndStores); + chooseInstructionsToInstrument(LocalLoadsAndStores, AllLoadsAndStores); } // We have collected all loads and stores. @@ -261,6 +288,11 @@ bool ThreadSanitizer::runOnFunction(Function &F) { Res |= instrumentLoadOrStore(AllLoadsAndStores[i]); } + // Instrument atomic memory accesses. + for (size_t i = 0, n = AtomicAccesses.size(); i < n; ++i) { + Res |= instrumentAtomic(AtomicAccesses[i]); + } + // Instrument function entry/exit points if there were instrumented accesses. if (Res || HasCalls) { IRBuilder<> IRB(F.getEntryBlock().getFirstNonPHI()); @@ -283,29 +315,98 @@ bool ThreadSanitizer::instrumentLoadOrStore(Instruction *I) { Value *Addr = IsWrite ? cast<StoreInst>(I)->getPointerOperand() : cast<LoadInst>(I)->getPointerOperand(); - Type *OrigPtrTy = Addr->getType(); - Type *OrigTy = cast<PointerType>(OrigPtrTy)->getElementType(); - assert(OrigTy->isSized()); - uint32_t TypeSize = TD->getTypeStoreSizeInBits(OrigTy); - if (TypeSize != 8 && TypeSize != 16 && - TypeSize != 32 && TypeSize != 64 && TypeSize != 128) { - stats.NumAccessesWithBadSize++; - // Ignore all unusual sizes. + int Idx = getMemoryAccessFuncIndex(Addr); + if (Idx < 0) return false; - } if (IsWrite && isVtableAccess(I)) { + DEBUG(dbgs() << " VPTR : " << *I << "\n"); Value *StoredValue = cast<StoreInst>(I)->getValueOperand(); + // StoredValue does not necessary have a pointer type. + if (isa<IntegerType>(StoredValue->getType())) + StoredValue = IRB.CreateIntToPtr(StoredValue, IRB.getInt8PtrTy()); + // Call TsanVptrUpdate. IRB.CreateCall2(TsanVptrUpdate, IRB.CreatePointerCast(Addr, IRB.getInt8PtrTy()), IRB.CreatePointerCast(StoredValue, IRB.getInt8PtrTy())); - stats.NumInstrumentedVtableWrites++; + NumInstrumentedVtableWrites++; return true; } - size_t Idx = CountTrailingZeros_32(TypeSize / 8); - assert(Idx < kNumberOfAccessSizes); Value *OnAccessFunc = IsWrite ? TsanWrite[Idx] : TsanRead[Idx]; IRB.CreateCall(OnAccessFunc, IRB.CreatePointerCast(Addr, IRB.getInt8PtrTy())); - if (IsWrite) stats.NumInstrumentedWrites++; - else stats.NumInstrumentedReads++; + if (IsWrite) NumInstrumentedWrites++; + else NumInstrumentedReads++; + return true; +} + +static ConstantInt *createOrdering(IRBuilder<> *IRB, AtomicOrdering ord) { + uint32_t v = 0; + switch (ord) { + case NotAtomic: assert(false); + case Unordered: // Fall-through. + case Monotonic: v = 1 << 0; break; + // case Consume: v = 1 << 1; break; // Not specified yet. + case Acquire: v = 1 << 2; break; + case Release: v = 1 << 3; break; + case AcquireRelease: v = 1 << 4; break; + case SequentiallyConsistent: v = 1 << 5; break; + } + return IRB->getInt32(v); +} + +bool ThreadSanitizer::instrumentAtomic(Instruction *I) { + IRBuilder<> IRB(I); + if (LoadInst *LI = dyn_cast<LoadInst>(I)) { + Value *Addr = LI->getPointerOperand(); + int Idx = getMemoryAccessFuncIndex(Addr); + if (Idx < 0) + return false; + const size_t ByteSize = 1 << Idx; + const size_t BitSize = ByteSize * 8; + Type *Ty = Type::getIntNTy(IRB.getContext(), BitSize); + Type *PtrTy = Ty->getPointerTo(); + Value *Args[] = {IRB.CreatePointerCast(Addr, PtrTy), + createOrdering(&IRB, LI->getOrdering())}; + CallInst *C = CallInst::Create(TsanAtomicLoad[Idx], + ArrayRef<Value*>(Args)); + ReplaceInstWithInst(I, C); + + } else if (StoreInst *SI = dyn_cast<StoreInst>(I)) { + Value *Addr = SI->getPointerOperand(); + int Idx = getMemoryAccessFuncIndex(Addr); + if (Idx < 0) + return false; + const size_t ByteSize = 1 << Idx; + const size_t BitSize = ByteSize * 8; + Type *Ty = Type::getIntNTy(IRB.getContext(), BitSize); + Type *PtrTy = Ty->getPointerTo(); + Value *Args[] = {IRB.CreatePointerCast(Addr, PtrTy), + IRB.CreateIntCast(SI->getValueOperand(), Ty, false), + createOrdering(&IRB, SI->getOrdering())}; + CallInst *C = CallInst::Create(TsanAtomicStore[Idx], + ArrayRef<Value*>(Args)); + ReplaceInstWithInst(I, C); + } else if (isa<AtomicRMWInst>(I)) { + // FIXME: Not yet supported. + } else if (isa<AtomicCmpXchgInst>(I)) { + // FIXME: Not yet supported. + } else if (isa<FenceInst>(I)) { + // FIXME: Not yet supported. + } return true; } + +int ThreadSanitizer::getMemoryAccessFuncIndex(Value *Addr) { + Type *OrigPtrTy = Addr->getType(); + Type *OrigTy = cast<PointerType>(OrigPtrTy)->getElementType(); + assert(OrigTy->isSized()); + uint32_t TypeSize = TD->getTypeStoreSizeInBits(OrigTy); + if (TypeSize != 8 && TypeSize != 16 && + TypeSize != 32 && TypeSize != 64 && TypeSize != 128) { + NumAccessesWithBadSize++; + // Ignore all unusual sizes. + return -1; + } + size_t Idx = CountTrailingZeros_32(TypeSize / 8); + assert(Idx < kNumberOfAccessSizes); + return Idx; +} |