X-Git-Url: http://demsky.eecs.uci.edu/git/?a=blobdiff_plain;f=CDSPass.cpp;h=a33738a96d26880395afc7beaf3b284e1bf1e705;hb=HEAD;hp=354e5bd49b7c00e1845f646c9e98860bbc730486;hpb=c30760be09db5745af0f3bb0f54cc1aa60fd0eb4;p=c11llvm.git diff --git a/CDSPass.cpp b/CDSPass.cpp index 354e5bd..a33738a 100644 --- a/CDSPass.cpp +++ b/CDSPass.cpp @@ -1,7 +1,8 @@ //===-- CDSPass.cpp - xxx -------------------------------===// // -// The LLVM Compiler Infrastructure -// +// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. +// See https://llvm.org/LICENSE.txt for license information. +// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception // This file is distributed under the University of Illinois Open Source // License. See LICENSE.TXT for details. // @@ -24,10 +25,12 @@ #include "llvm/ADT/SmallString.h" #include "llvm/Analysis/ValueTracking.h" #include "llvm/Analysis/CaptureTracking.h" +#include "llvm/Analysis/LoopInfo.h" #include "llvm/IR/BasicBlock.h" #include "llvm/IR/Function.h" #include "llvm/IR/IRBuilder.h" #include "llvm/IR/Instructions.h" +#include "llvm/IR/IntrinsicInst.h" #include "llvm/IR/LLVMContext.h" #include "llvm/IR/LegacyPassManager.h" #include "llvm/IR/Module.h" @@ -38,17 +41,18 @@ #include "llvm/Support/AtomicOrdering.h" #include "llvm/Support/Debug.h" #include "llvm/Transforms/Scalar.h" -#include "llvm/Transforms/Utils/Local.h" #include "llvm/Transforms/Utils/BasicBlockUtils.h" +#include "llvm/Transforms/Utils/EscapeEnumerator.h" #include "llvm/Transforms/IPO/PassManagerBuilder.h" #include -#define DEBUG_TYPE "CDS" using namespace llvm; +#define CDS_DEBUG +#define DEBUG_TYPE "CDS" #include -Value *getPosition( Instruction * I, IRBuilder <> IRB) +static inline Value *getPosition( Instruction * I, IRBuilder <> IRB, bool print = false) { const DebugLoc & debug_location = I->getDebugLoc (); std::string position_string; @@ -57,29 +61,45 @@ Value *getPosition( Instruction * I, IRBuilder <> IRB) debug_location . print (position_stream); } - return IRB . CreateGlobalStringPtr (position_string); + if (print) { + errs() << position_string << "\n"; + } + + return IRB.CreateGlobalStringPtr (position_string); } -#define FUNCARRAYSIZE 4 +static inline bool checkSignature(Function * func, Value * args[]) { + FunctionType * FType = func->getFunctionType(); + for (unsigned i = 0 ; i < FType->getNumParams(); i++) { + if (FType->getParamType(i) != args[i]->getType()) { +#ifdef CDS_DEBUG + errs() << "expects: " << *FType->getParamType(i) + << "\tbut receives: " << *args[i]->getType() << "\n"; +#endif + return false; + } + } + + return true; +} STATISTIC(NumInstrumentedReads, "Number of instrumented reads"); STATISTIC(NumInstrumentedWrites, "Number of instrumented writes"); -// STATISTIC(NumInstrumentedVtableWrites, "Number of vtable ptr writes"); -// STATISTIC(NumInstrumentedVtableReads, "Number of vtable ptr reads"); - 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(NumInstrumentedVtableReads, "Number of vtable ptr reads"); STATISTIC(NumOmittedReadsFromConstantGlobals, "Number of reads from constant globals"); STATISTIC(NumOmittedReadsFromVtable, "Number of vtable reads"); STATISTIC(NumOmittedNonCaptured, "Number of accesses ignored due to capturing"); -Type * Int8Ty; -Type * Int16Ty; -Type * Int32Ty; -Type * Int64Ty; -Type * OrdTy; +// static const char *const kCDSModuleCtorName = "cds.module_ctor"; +// static const char *const kCDSInitName = "cds_init"; +Type * OrdTy; +Type * IntPtrTy; Type * Int8PtrTy; Type * Int16PtrTy; Type * Int32PtrTy; @@ -87,97 +107,134 @@ Type * Int64PtrTy; Type * VoidTy; -Constant * CDSLoad[FUNCARRAYSIZE]; -Constant * CDSStore[FUNCARRAYSIZE]; -Constant * CDSAtomicInit[FUNCARRAYSIZE]; -Constant * CDSAtomicLoad[FUNCARRAYSIZE]; -Constant * CDSAtomicStore[FUNCARRAYSIZE]; -Constant * CDSAtomicRMW[AtomicRMWInst::LAST_BINOP + 1][FUNCARRAYSIZE]; -Constant * CDSAtomicCAS_V1[FUNCARRAYSIZE]; -Constant * CDSAtomicCAS_V2[FUNCARRAYSIZE]; -Constant * CDSAtomicThreadFence; - -bool start = false; - -int getAtomicOrderIndex(AtomicOrdering order){ - switch (order) { - case AtomicOrdering::Monotonic: - return (int)AtomicOrderingCABI::relaxed; -// case AtomicOrdering::Consume: // not specified yet -// return AtomicOrderingCABI::consume; - case AtomicOrdering::Acquire: - return (int)AtomicOrderingCABI::acquire; - case AtomicOrdering::Release: - return (int)AtomicOrderingCABI::release; - case AtomicOrdering::AcquireRelease: - return (int)AtomicOrderingCABI::acq_rel; - case AtomicOrdering::SequentiallyConsistent: - return (int)AtomicOrderingCABI::seq_cst; - default: - // unordered or Not Atomic - return -1; - } +static const size_t kNumberOfAccessSizes = 4; + +int getAtomicOrderIndex(AtomicOrdering order) { + switch (order) { + case AtomicOrdering::Monotonic: + return (int)AtomicOrderingCABI::relaxed; + //case AtomicOrdering::Consume: // not specified yet + // return AtomicOrderingCABI::consume; + case AtomicOrdering::Acquire: + return (int)AtomicOrderingCABI::acquire; + case AtomicOrdering::Release: + return (int)AtomicOrderingCABI::release; + case AtomicOrdering::AcquireRelease: + return (int)AtomicOrderingCABI::acq_rel; + case AtomicOrdering::SequentiallyConsistent: + return (int)AtomicOrderingCABI::seq_cst; + default: + // unordered or Not Atomic + return -1; + } } -int getTypeSize(Type* type) { - if (type == Int8PtrTy) { - return sizeof(char)*8; - } else if (type == Int16PtrTy) { - return sizeof(short)*8; - } else if (type == Int32PtrTy) { - return sizeof(int)*8; - } else if (type == Int64PtrTy) { - return sizeof(long long int)*8; - } else { - return sizeof(void*)*8; - } - - return -1; +AtomicOrderingCABI indexToAtomicOrder(int index) { + switch (index) { + case 0: + return AtomicOrderingCABI::relaxed; + case 1: + return AtomicOrderingCABI::consume; + case 2: + return AtomicOrderingCABI::acquire; + case 3: + return AtomicOrderingCABI::release; + case 4: + return AtomicOrderingCABI::acq_rel; + case 5: + return AtomicOrderingCABI::seq_cst; + default: + errs() << "Bad Atomic index\n"; + return AtomicOrderingCABI::seq_cst; + } +} + +/* According to atomic_base.h: __cmpexch_failure_order */ +int AtomicCasFailureOrderIndex(int index) { + AtomicOrderingCABI succ_order = indexToAtomicOrder(index); + AtomicOrderingCABI fail_order; + if (succ_order == AtomicOrderingCABI::acq_rel) + fail_order = AtomicOrderingCABI::acquire; + else if (succ_order == AtomicOrderingCABI::release) + fail_order = AtomicOrderingCABI::relaxed; + else + fail_order = succ_order; + + return (int) fail_order; } -static int sizetoindex(int size) { - switch(size) { - case 8: return 0; - case 16: return 1; - case 32: return 2; - case 64: return 3; - } - return -1; +/* The original function checkSanitizerInterfaceFunction was defined + * in llvm/Transforms/Utils/ModuleUtils.h + */ +static Function * checkCDSPassInterfaceFunction(Constant *FuncOrBitcast) { + if (isa(FuncOrBitcast)) + return cast(FuncOrBitcast); + FuncOrBitcast->print(errs()); + errs() << "\n"; + std::string Err; + raw_string_ostream Stream(Err); + Stream << "CDSPass interface function redefined: " << *FuncOrBitcast; + report_fatal_error(Err); } namespace { - struct CDSPass : public FunctionPass { - static char ID; - CDSPass() : FunctionPass(ID) {} - bool runOnFunction(Function &F) override; - - private: - void initializeCallbacks(Module &M); - bool instrumentLoadOrStore(Instruction *I, const DataLayout &DL); - bool instrumentAtomic(Instruction *I, const DataLayout &DL); - bool instrumentAtomicCall(CallInst *CI, const DataLayout &DL); - void chooseInstructionsToInstrument(SmallVectorImpl &Local, - SmallVectorImpl &All, - const DataLayout &DL); - bool addrPointsToConstantData(Value *Addr); - int getMemoryAccessFuncIndex(Value *Addr, const DataLayout &DL); - }; + struct CDSPass : public FunctionPass { + CDSPass() : FunctionPass(ID) {} + StringRef getPassName() const override; + bool runOnFunction(Function &F) override; + bool doInitialization(Module &M) override; + static char ID; + + private: + void initializeCallbacks(Module &M); + bool instrumentLoadOrStore(Instruction *I, const DataLayout &DL); + bool instrumentVolatile(Instruction *I, const DataLayout &DL); + bool instrumentMemIntrinsic(Instruction *I); + bool isAtomicCall(Instruction *I); + bool instrumentAtomic(Instruction *I, const DataLayout &DL); + bool instrumentAtomicCall(CallInst *CI, const DataLayout &DL); + bool shouldInstrumentBeforeAtomics(Instruction *I); + void chooseInstructionsToInstrument(SmallVectorImpl &Local, + SmallVectorImpl &All, + const DataLayout &DL); + bool addrPointsToConstantData(Value *Addr); + int getMemoryAccessFuncIndex(Value *Addr, const DataLayout &DL); + bool instrumentLoops(Function &F); + + Function * CDSFuncEntry; + Function * CDSFuncExit; + + Function * CDSLoad[kNumberOfAccessSizes]; + Function * CDSStore[kNumberOfAccessSizes]; + Function * CDSVolatileLoad[kNumberOfAccessSizes]; + Function * CDSVolatileStore[kNumberOfAccessSizes]; + Function * CDSAtomicInit[kNumberOfAccessSizes]; + Function * CDSAtomicLoad[kNumberOfAccessSizes]; + Function * CDSAtomicStore[kNumberOfAccessSizes]; + Function * CDSAtomicRMW[AtomicRMWInst::LAST_BINOP + 1][kNumberOfAccessSizes]; + Function * CDSAtomicCAS_V1[kNumberOfAccessSizes]; + Function * CDSAtomicCAS_V2[kNumberOfAccessSizes]; + Function * CDSAtomicThreadFence; + Function * MemmoveFn, * MemcpyFn, * MemsetFn; + // Function * CDSCtorFunction; + + std::vector AtomicFuncNames; + std::vector PartialAtomicFuncNames; + }; } -static bool isVtableAccess(Instruction *I) { - if (MDNode *Tag = I->getMetadata(LLVMContext::MD_tbaa)) - return Tag->isTBAAVtableAccess(); - return false; +StringRef CDSPass::getPassName() const { + return "CDSPass"; } void CDSPass::initializeCallbacks(Module &M) { LLVMContext &Ctx = M.getContext(); + AttributeList Attr; + Attr = Attr.addAttribute(Ctx, AttributeList::FunctionIndex, + Attribute::NoUnwind); Type * Int1Ty = Type::getInt1Ty(Ctx); - Int8Ty = Type::getInt8Ty(Ctx); - Int16Ty = Type::getInt16Ty(Ctx); - Int32Ty = Type::getInt32Ty(Ctx); - Int64Ty = Type::getInt64Ty(Ctx); + Type * Int32Ty = Type::getInt32Ty(Ctx); OrdTy = Type::getInt32Ty(Ctx); Int8PtrTy = Type::getInt8PtrTy(Ctx); @@ -186,9 +243,16 @@ void CDSPass::initializeCallbacks(Module &M) { Int64PtrTy = Type::getInt64PtrTy(Ctx); VoidTy = Type::getVoidTy(Ctx); - + + CDSFuncEntry = checkCDSPassInterfaceFunction( + M.getOrInsertFunction("cds_func_entry", + Attr, VoidTy, Int8PtrTy)); + CDSFuncExit = checkCDSPassInterfaceFunction( + M.getOrInsertFunction("cds_func_exit", + Attr, VoidTy, Int8PtrTy)); + // Get the function to call from our untime library. - for (unsigned i = 0; i < FUNCARRAYSIZE; i++) { + for (unsigned i = 0; i < kNumberOfAccessSizes; i++) { const unsigned ByteSize = 1U << i; const unsigned BitSize = ByteSize * 8; @@ -202,18 +266,31 @@ void CDSPass::initializeCallbacks(Module &M) { // void cds_atomic_store8 (void * obj, int atomic_index, uint8_t val) SmallString<32> LoadName("cds_load" + BitSizeStr); SmallString<32> StoreName("cds_store" + BitSizeStr); + SmallString<32> VolatileLoadName("cds_volatile_load" + BitSizeStr); + SmallString<32> VolatileStoreName("cds_volatile_store" + BitSizeStr); SmallString<32> AtomicInitName("cds_atomic_init" + BitSizeStr); SmallString<32> AtomicLoadName("cds_atomic_load" + BitSizeStr); SmallString<32> AtomicStoreName("cds_atomic_store" + BitSizeStr); - CDSLoad[i] = M.getOrInsertFunction(LoadName, VoidTy, PtrTy); - CDSStore[i] = M.getOrInsertFunction(StoreName, VoidTy, PtrTy); - CDSAtomicInit[i] = M.getOrInsertFunction(AtomicInitName, - VoidTy, PtrTy, Ty, Int8PtrTy); - CDSAtomicLoad[i] = M.getOrInsertFunction(AtomicLoadName, - Ty, PtrTy, OrdTy, Int8PtrTy); - CDSAtomicStore[i] = M.getOrInsertFunction(AtomicStoreName, - VoidTy, PtrTy, Ty, OrdTy, Int8PtrTy); + CDSLoad[i] = checkCDSPassInterfaceFunction( + M.getOrInsertFunction(LoadName, Attr, VoidTy, Int8PtrTy)); + CDSStore[i] = checkCDSPassInterfaceFunction( + M.getOrInsertFunction(StoreName, Attr, VoidTy, Int8PtrTy)); + CDSVolatileLoad[i] = checkCDSPassInterfaceFunction( + M.getOrInsertFunction(VolatileLoadName, + Attr, Ty, PtrTy, Int8PtrTy)); + CDSVolatileStore[i] = checkCDSPassInterfaceFunction( + M.getOrInsertFunction(VolatileStoreName, + Attr, VoidTy, PtrTy, Ty, Int8PtrTy)); + CDSAtomicInit[i] = checkCDSPassInterfaceFunction( + M.getOrInsertFunction(AtomicInitName, + Attr, VoidTy, PtrTy, Ty, Int8PtrTy)); + CDSAtomicLoad[i] = checkCDSPassInterfaceFunction( + M.getOrInsertFunction(AtomicLoadName, + Attr, Ty, PtrTy, OrdTy, Int8PtrTy)); + CDSAtomicStore[i] = checkCDSPassInterfaceFunction( + M.getOrInsertFunction(AtomicStoreName, + Attr, VoidTy, PtrTy, Ty, OrdTy, Int8PtrTy)); for (int op = AtomicRMWInst::FIRST_BINOP; op <= AtomicRMWInst::LAST_BINOP; ++op) { @@ -236,62 +313,531 @@ void CDSPass::initializeCallbacks(Module &M) { continue; SmallString<32> AtomicRMWName("cds_atomic" + NamePart + BitSizeStr); - CDSAtomicRMW[op][i] = M.getOrInsertFunction(AtomicRMWName, - Ty, PtrTy, Ty, OrdTy, Int8PtrTy); + CDSAtomicRMW[op][i] = checkCDSPassInterfaceFunction( + M.getOrInsertFunction(AtomicRMWName, + Attr, Ty, PtrTy, Ty, OrdTy, Int8PtrTy)); } // only supportes strong version SmallString<32> AtomicCASName_V1("cds_atomic_compare_exchange" + BitSizeStr + "_v1"); SmallString<32> AtomicCASName_V2("cds_atomic_compare_exchange" + BitSizeStr + "_v2"); - CDSAtomicCAS_V1[i] = M.getOrInsertFunction(AtomicCASName_V1, - Ty, PtrTy, Ty, Ty, OrdTy, OrdTy, Int8PtrTy); - CDSAtomicCAS_V2[i] = M.getOrInsertFunction(AtomicCASName_V2, - Int1Ty, PtrTy, PtrTy, Ty, OrdTy, OrdTy, Int8PtrTy); + CDSAtomicCAS_V1[i] = checkCDSPassInterfaceFunction( + M.getOrInsertFunction(AtomicCASName_V1, + Attr, Ty, PtrTy, Ty, Ty, OrdTy, OrdTy, Int8PtrTy)); + CDSAtomicCAS_V2[i] = checkCDSPassInterfaceFunction( + M.getOrInsertFunction(AtomicCASName_V2, + Attr, Int1Ty, PtrTy, PtrTy, Ty, OrdTy, OrdTy, Int8PtrTy)); } - CDSAtomicThreadFence = M.getOrInsertFunction("cds_atomic_thread_fence", - VoidTy, OrdTy, Int8PtrTy); + CDSAtomicThreadFence = checkCDSPassInterfaceFunction( + M.getOrInsertFunction("cds_atomic_thread_fence", Attr, VoidTy, OrdTy, Int8PtrTy)); + + MemmoveFn = checkCDSPassInterfaceFunction( + M.getOrInsertFunction("memmove", Attr, Int8PtrTy, Int8PtrTy, + Int8PtrTy, IntPtrTy)); + MemcpyFn = checkCDSPassInterfaceFunction( + M.getOrInsertFunction("memcpy", Attr, Int8PtrTy, Int8PtrTy, + Int8PtrTy, IntPtrTy)); + MemsetFn = checkCDSPassInterfaceFunction( + M.getOrInsertFunction("memset", Attr, Int8PtrTy, Int8PtrTy, + Int32Ty, IntPtrTy)); } -void printArgs(CallInst *); +bool CDSPass::doInitialization(Module &M) { + const DataLayout &DL = M.getDataLayout(); + IntPtrTy = DL.getIntPtrType(M.getContext()); + + // createSanitizerCtorAndInitFunctions is defined in "llvm/Transforms/Utils/ModuleUtils.h" + // We do not support it yet + /* + std::tie(CDSCtorFunction, std::ignore) = createSanitizerCtorAndInitFunctions( + M, kCDSModuleCtorName, kCDSInitName, {}, {}); + + appendToGlobalCtors(M, CDSCtorFunction, 0); + */ -bool isAtomicCall(Instruction *I) -{ - if ( auto *CI = dyn_cast(I) ) { - Function *fun = CI->getCalledFunction(); - if (fun == NULL) + AtomicFuncNames = + { + "atomic_init", "atomic_load", "atomic_store", + "atomic_fetch_", "atomic_exchange", "atomic_compare_exchange_" + }; + + PartialAtomicFuncNames = + { + "load", "store", "fetch", "exchange", "compare_exchange_" + }; + + return true; +} + +static bool isVtableAccess(Instruction *I) { + if (MDNode *Tag = I->getMetadata(LLVMContext::MD_tbaa)) + return Tag->isTBAAVtableAccess(); + return false; +} + +// Do not instrument known races/"benign races" that come from compiler +// instrumentatin. The user has no way of suppressing them. +static bool shouldInstrumentReadWriteFromAddress(const Module *M, Value *Addr) { + // Peel off GEPs and BitCasts. + Addr = Addr->stripInBoundsOffsets(); + + if (GlobalVariable *GV = dyn_cast(Addr)) { + if (GV->hasSection()) { + StringRef SectionName = GV->getSection(); + // Check if the global is in the PGO counters section. + auto OF = Triple(M->getTargetTriple()).getObjectFormat(); + if (SectionName.endswith( + getInstrProfSectionName(IPSK_cnts, OF, /*AddSegmentInfo=*/false))) + return false; + } + + // Check if the global is private gcov data. + if (GV->getName().startswith("__llvm_gcov") || + GV->getName().startswith("__llvm_gcda")) + return false; + } + + // Do not instrument acesses from different address spaces; we cannot deal + // with them. + if (Addr) { + Type *PtrTy = cast(Addr->getType()->getScalarType()); + if (PtrTy->getPointerAddressSpace() != 0) return false; + } - StringRef funName = fun->getName(); + return true; +} - if ( (CI->isTailCall() && funName.contains("atomic_")) || - funName.contains("atomic_compare_exchange_") ) { - // printArgs(CI); +bool CDSPass::addrPointsToConstantData(Value *Addr) { + // If this is a GEP, just analyze its pointer operand. + if (GetElementPtrInst *GEP = dyn_cast(Addr)) + Addr = GEP->getPointerOperand(); + + if (GlobalVariable *GV = dyn_cast(Addr)) { + if (GV->isConstant()) { + // Reads from constant globals can not race with any writes. + NumOmittedReadsFromConstantGlobals++; return true; } + } else if (LoadInst *L = dyn_cast(Addr)) { + if (isVtableAccess(L)) { + // Reads from a vtable pointer can not race with any writes. + NumOmittedReadsFromVtable++; + return true; + } + } + return false; +} + +bool CDSPass::shouldInstrumentBeforeAtomics(Instruction * Inst) { + if (LoadInst *LI = dyn_cast(Inst)) { + AtomicOrdering ordering = LI->getOrdering(); + if ( isAtLeastOrStrongerThan(ordering, AtomicOrdering::Acquire) ) + return true; + } else if (StoreInst *SI = dyn_cast(Inst)) { + AtomicOrdering ordering = SI->getOrdering(); + if ( isAtLeastOrStrongerThan(ordering, AtomicOrdering::Acquire) ) + return true; + } else if (AtomicRMWInst *RMWI = dyn_cast(Inst)) { + AtomicOrdering ordering = RMWI->getOrdering(); + if ( isAtLeastOrStrongerThan(ordering, AtomicOrdering::Acquire) ) + return true; + } else if (AtomicCmpXchgInst *CASI = dyn_cast(Inst)) { + AtomicOrdering ordering = CASI->getSuccessOrdering(); + if ( isAtLeastOrStrongerThan(ordering, AtomicOrdering::Acquire) ) + return true; + } else if (FenceInst *FI = dyn_cast(Inst)) { + AtomicOrdering ordering = FI->getOrdering(); + if ( isAtLeastOrStrongerThan(ordering, AtomicOrdering::Acquire) ) + return true; } return false; } -void printArgs (CallInst *CI) -{ - Function *fun = CI->getCalledFunction(); - StringRef funName = fun->getName(); +void CDSPass::chooseInstructionsToInstrument( + SmallVectorImpl &Local, SmallVectorImpl &All, + const DataLayout &DL) { + SmallPtrSet WriteTargets; + // Iterate from the end. + for (Instruction *I : reverse(Local)) { + if (StoreInst *Store = dyn_cast(I)) { + Value *Addr = Store->getPointerOperand(); + if (!shouldInstrumentReadWriteFromAddress(I->getModule(), Addr)) + continue; + WriteTargets.insert(Addr); + } else { + LoadInst *Load = cast(I); + Value *Addr = Load->getPointerOperand(); + if (!shouldInstrumentReadWriteFromAddress(I->getModule(), Addr)) + continue; + if (WriteTargets.count(Addr)) { + // We will write to this temp, so no reason to analyze the read. + NumOmittedReadsBeforeWrite++; + continue; + } + if (addrPointsToConstantData(Addr)) { + // Addr points to some constant data -- it can not race with any writes. + continue; + } + } + Value *Addr = isa(*I) + ? cast(I)->getPointerOperand() + : cast(I)->getPointerOperand(); + if (isa(GetUnderlyingObject(Addr, DL)) && + !PointerMayBeCaptured(Addr, true, true)) { + // The variable is addressable but not captured, so it cannot be + // referenced from a different thread and participate in a data race + // (see llvm/Analysis/CaptureTracking.h for details). + NumOmittedNonCaptured++; + continue; + } + All.push_back(I); + } + Local.clear(); +} - User::op_iterator begin = CI->arg_begin(); - User::op_iterator end = CI->arg_end(); +/* Not implemented +void CDSPass::InsertRuntimeIgnores(Function &F) { + IRBuilder<> IRB(F.getEntryBlock().getFirstNonPHI()); + IRB.CreateCall(CDSIgnoreBegin); + EscapeEnumerator EE(F, "cds_ignore_cleanup", ClHandleCxxExceptions); + while (IRBuilder<> *AtExit = EE.Next()) { + AtExit->CreateCall(CDSIgnoreEnd); + } +}*/ + +bool CDSPass::runOnFunction(Function &F) { + initializeCallbacks( *F.getParent() ); + SmallVector AllLoadsAndStores; + SmallVector LocalLoadsAndStores; + SmallVector VolatileLoadsAndStores; + SmallVector AtomicAccesses; + SmallVector MemIntrinCalls; + + bool Res = false; + bool HasAtomic = false; + bool HasVolatile = false; + const DataLayout &DL = F.getParent()->getDataLayout(); + + // instrumentLoops(F); + + for (auto &BB : F) { + for (auto &Inst : BB) { + if ( (&Inst)->isAtomic() ) { + AtomicAccesses.push_back(&Inst); + HasAtomic = true; + + if (shouldInstrumentBeforeAtomics(&Inst)) { + chooseInstructionsToInstrument(LocalLoadsAndStores, AllLoadsAndStores, + DL); + } + } else if (isAtomicCall(&Inst) ) { + AtomicAccesses.push_back(&Inst); + HasAtomic = true; + chooseInstructionsToInstrument(LocalLoadsAndStores, AllLoadsAndStores, + DL); + } else if (isa(Inst) || isa(Inst)) { + LoadInst *LI = dyn_cast(&Inst); + StoreInst *SI = dyn_cast(&Inst); + bool isVolatile = ( LI ? LI->isVolatile() : SI->isVolatile() ); + + if (isVolatile) { + VolatileLoadsAndStores.push_back(&Inst); + HasVolatile = true; + } else + LocalLoadsAndStores.push_back(&Inst); + } else if (isa(Inst) || isa(Inst)) { + if (isa(Inst)) + MemIntrinCalls.push_back(&Inst); + + /*if (CallInst *CI = dyn_cast(&Inst)) + maybeMarkSanitizerLibraryCallNoBuiltin(CI, TLI); + */ + + chooseInstructionsToInstrument(LocalLoadsAndStores, AllLoadsAndStores, + DL); + } + } + + chooseInstructionsToInstrument(LocalLoadsAndStores, AllLoadsAndStores, DL); + } + + for (auto Inst : AllLoadsAndStores) { + Res |= instrumentLoadOrStore(Inst, DL); + } + + for (auto Inst : VolatileLoadsAndStores) { + Res |= instrumentVolatile(Inst, DL); + } + + for (auto Inst : AtomicAccesses) { + Res |= instrumentAtomic(Inst, DL); + } + + for (auto Inst : MemIntrinCalls) { + Res |= instrumentMemIntrinsic(Inst); + } + + // Instrument function entry and exit for functions containing atomics or volatiles + if (Res && ( HasAtomic || HasVolatile) ) { + IRBuilder<> IRB(F.getEntryBlock().getFirstNonPHI()); + /* Unused for now + Value *ReturnAddress = IRB.CreateCall( + Intrinsic::getDeclaration(F.getParent(), Intrinsic::returnaddress), + IRB.getInt32(0)); + */ + + Value * FuncName = IRB.CreateGlobalStringPtr(F.getName()); + IRB.CreateCall(CDSFuncEntry, FuncName); + + EscapeEnumerator EE(F, "cds_cleanup", true); + while (IRBuilder<> *AtExit = EE.Next()) { + AtExit->CreateCall(CDSFuncExit, FuncName); + } + + Res = true; + } + + return false; +} + +bool CDSPass::instrumentLoadOrStore(Instruction *I, + const DataLayout &DL) { + IRBuilder<> IRB(I); + bool IsWrite = isa(*I); + Value *Addr = IsWrite + ? cast(I)->getPointerOperand() + : cast(I)->getPointerOperand(); + + // swifterror memory addresses are mem2reg promoted by instruction selection. + // As such they cannot have regular uses like an instrumentation function and + // it makes no sense to track them as memory. + if (Addr->isSwiftError()) + return false; + + int Idx = getMemoryAccessFuncIndex(Addr, DL); + if (Idx < 0) + return false; + + if (IsWrite && isVtableAccess(I)) { + /* TODO + LLVM_DEBUG(dbgs() << " VPTR : " << *I << "\n"); + Value *StoredValue = cast(I)->getValueOperand(); + // StoredValue may be a vector type if we are storing several vptrs at once. + // In this case, just take the first element of the vector since this is + // enough to find vptr races. + if (isa(StoredValue->getType())) + StoredValue = IRB.CreateExtractElement( + StoredValue, ConstantInt::get(IRB.getInt32Ty(), 0)); + if (StoredValue->getType()->isIntegerTy()) + StoredValue = IRB.CreateIntToPtr(StoredValue, IRB.getInt8PtrTy()); + // Call TsanVptrUpdate. + IRB.CreateCall(TsanVptrUpdate, + {IRB.CreatePointerCast(Addr, IRB.getInt8PtrTy()), + IRB.CreatePointerCast(StoredValue, IRB.getInt8PtrTy())}); + NumInstrumentedVtableWrites++; + */ + return true; + } + + if (!IsWrite && isVtableAccess(I)) { + /* TODO + IRB.CreateCall(TsanVptrLoad, + IRB.CreatePointerCast(Addr, IRB.getInt8PtrTy())); + NumInstrumentedVtableReads++; + */ + return true; + } + + // TODO: unaligned reads and writes + Value *OnAccessFunc = nullptr; + OnAccessFunc = IsWrite ? CDSStore[Idx] : CDSLoad[Idx]; + IRB.CreateCall(OnAccessFunc, IRB.CreatePointerCast(Addr, IRB.getInt8PtrTy())); + if (IsWrite) NumInstrumentedWrites++; + else NumInstrumentedReads++; + return true; +} + +bool CDSPass::instrumentVolatile(Instruction * I, const DataLayout &DL) { + IRBuilder<> IRB(I); + Value *position = getPosition(I, IRB); + + if (LoadInst *LI = dyn_cast(I)) { + Value *Addr = LI->getPointerOperand(); + int Idx=getMemoryAccessFuncIndex(Addr, DL); + if (Idx < 0) + return false; + const unsigned ByteSize = 1U << Idx; + const unsigned BitSize = ByteSize * 8; + Type *Ty = Type::getIntNTy(IRB.getContext(), BitSize); + Type *PtrTy = Ty->getPointerTo(); + Value *Args[] = {IRB.CreatePointerCast(Addr, PtrTy), position}; + + Type *OrigTy = cast(Addr->getType())->getElementType(); + Value *C = IRB.CreateCall(CDSVolatileLoad[Idx], Args); + Value *Cast = IRB.CreateBitOrPointerCast(C, OrigTy); + I->replaceAllUsesWith(Cast); + } else if (StoreInst *SI = dyn_cast(I)) { + Value *Addr = SI->getPointerOperand(); + int Idx=getMemoryAccessFuncIndex(Addr, DL); + if (Idx < 0) + return false; + const unsigned ByteSize = 1U << Idx; + const unsigned BitSize = ByteSize * 8; + Type *Ty = Type::getIntNTy(IRB.getContext(), BitSize); + Type *PtrTy = Ty->getPointerTo(); + Value *Args[] = {IRB.CreatePointerCast(Addr, PtrTy), + IRB.CreateBitOrPointerCast(SI->getValueOperand(), Ty), + position}; + CallInst *C = CallInst::Create(CDSVolatileStore[Idx], Args); + ReplaceInstWithInst(I, C); + } else { + return false; + } + + return true; +} + +bool CDSPass::instrumentMemIntrinsic(Instruction *I) { + IRBuilder<> IRB(I); + if (MemSetInst *M = dyn_cast(I)) { + IRB.CreateCall( + MemsetFn, + {IRB.CreatePointerCast(M->getArgOperand(0), IRB.getInt8PtrTy()), + IRB.CreateIntCast(M->getArgOperand(1), IRB.getInt32Ty(), false), + IRB.CreateIntCast(M->getArgOperand(2), IntPtrTy, false)}); + I->eraseFromParent(); + } else if (MemTransferInst *M = dyn_cast(I)) { + IRB.CreateCall( + isa(M) ? MemcpyFn : MemmoveFn, + {IRB.CreatePointerCast(M->getArgOperand(0), IRB.getInt8PtrTy()), + IRB.CreatePointerCast(M->getArgOperand(1), IRB.getInt8PtrTy()), + IRB.CreateIntCast(M->getArgOperand(2), IntPtrTy, false)}); + I->eraseFromParent(); + } + return false; +} + +bool CDSPass::instrumentAtomic(Instruction * I, const DataLayout &DL) { + IRBuilder<> IRB(I); + + if (auto *CI = dyn_cast(I)) { + return instrumentAtomicCall(CI, DL); + } + + Value *position = getPosition(I, IRB); + if (LoadInst *LI = dyn_cast(I)) { + Value *Addr = LI->getPointerOperand(); + int Idx=getMemoryAccessFuncIndex(Addr, DL); + if (Idx < 0) + return false; + + int atomic_order_index = getAtomicOrderIndex(LI->getOrdering()); + Value *order = ConstantInt::get(OrdTy, atomic_order_index); + Value *Args[] = {Addr, order, position}; + Instruction* funcInst = CallInst::Create(CDSAtomicLoad[Idx], Args); + ReplaceInstWithInst(LI, funcInst); + } else if (StoreInst *SI = dyn_cast(I)) { + Value *Addr = SI->getPointerOperand(); + int Idx=getMemoryAccessFuncIndex(Addr, DL); + if (Idx < 0) + return false; + + int atomic_order_index = getAtomicOrderIndex(SI->getOrdering()); + Value *val = SI->getValueOperand(); + Value *order = ConstantInt::get(OrdTy, atomic_order_index); + Value *Args[] = {Addr, val, order, position}; + Instruction* funcInst = CallInst::Create(CDSAtomicStore[Idx], Args); + ReplaceInstWithInst(SI, funcInst); + } else if (AtomicRMWInst *RMWI = dyn_cast(I)) { + Value *Addr = RMWI->getPointerOperand(); + int Idx=getMemoryAccessFuncIndex(Addr, DL); + if (Idx < 0) + return false; + + int atomic_order_index = getAtomicOrderIndex(RMWI->getOrdering()); + Value *val = RMWI->getValOperand(); + Value *order = ConstantInt::get(OrdTy, atomic_order_index); + Value *Args[] = {Addr, val, order, position}; + Instruction* funcInst = CallInst::Create(CDSAtomicRMW[RMWI->getOperation()][Idx], Args); + ReplaceInstWithInst(RMWI, funcInst); + } else if (AtomicCmpXchgInst *CASI = dyn_cast(I)) { + IRBuilder<> IRB(CASI); + + Value *Addr = CASI->getPointerOperand(); + int Idx=getMemoryAccessFuncIndex(Addr, DL); + if (Idx < 0) + return false; + + const unsigned ByteSize = 1U << Idx; + const unsigned BitSize = ByteSize * 8; + Type *Ty = Type::getIntNTy(IRB.getContext(), BitSize); + Type *PtrTy = Ty->getPointerTo(); + + Value *CmpOperand = IRB.CreateBitOrPointerCast(CASI->getCompareOperand(), Ty); + Value *NewOperand = IRB.CreateBitOrPointerCast(CASI->getNewValOperand(), Ty); + + int atomic_order_index_succ = getAtomicOrderIndex(CASI->getSuccessOrdering()); + int atomic_order_index_fail = getAtomicOrderIndex(CASI->getFailureOrdering()); + Value *order_succ = ConstantInt::get(OrdTy, atomic_order_index_succ); + Value *order_fail = ConstantInt::get(OrdTy, atomic_order_index_fail); + + Value *Args[] = {IRB.CreatePointerCast(Addr, PtrTy), + CmpOperand, NewOperand, + order_succ, order_fail, position}; - if ( funName.contains("atomic_") ) { - std::vector parameters; + CallInst *funcInst = IRB.CreateCall(CDSAtomicCAS_V1[Idx], Args); + Value *Success = IRB.CreateICmpEQ(funcInst, CmpOperand); - for (User::op_iterator it = begin; it != end; ++it) { - Value *param = *it; - parameters.push_back(param); - errs() << *param << " type: " << *param->getType() << "\n"; + Value *OldVal = funcInst; + Type *OrigOldValTy = CASI->getNewValOperand()->getType(); + if (Ty != OrigOldValTy) { + // The value is a pointer, so we need to cast the return value. + OldVal = IRB.CreateIntToPtr(funcInst, OrigOldValTy); } + + Value *Res = + IRB.CreateInsertValue(UndefValue::get(CASI->getType()), OldVal, 0); + Res = IRB.CreateInsertValue(Res, Success, 1); + + I->replaceAllUsesWith(Res); + I->eraseFromParent(); + } else if (FenceInst *FI = dyn_cast(I)) { + int atomic_order_index = getAtomicOrderIndex(FI->getOrdering()); + Value *order = ConstantInt::get(OrdTy, atomic_order_index); + Value *Args[] = {order, position}; + + CallInst *funcInst = CallInst::Create(CDSAtomicThreadFence, Args); + ReplaceInstWithInst(FI, funcInst); + // errs() << "Thread Fences replaced\n"; } + return true; +} +bool CDSPass::isAtomicCall(Instruction *I) { + if ( auto *CI = dyn_cast(I) ) { + Function *fun = CI->getCalledFunction(); + if (fun == NULL) + return false; + + StringRef funName = fun->getName(); + + // TODO: come up with better rules for function name checking + for (StringRef name : AtomicFuncNames) { + if ( funName.contains(name) ) + return true; + } + + for (StringRef PartialName : PartialAtomicFuncNames) { + if (funName.contains(PartialName) && + funName.contains("atomic") ) + return true; + } + } + + return false; } bool CDSPass::instrumentAtomicCall(CallInst *CI, const DataLayout &DL) { @@ -312,6 +858,7 @@ bool CDSPass::instrumentAtomicCall(CallInst *CI, const DataLayout &DL) { // the pointer to the address is always the first argument Value *OrigPtr = parameters[0]; + int Idx = getMemoryAccessFuncIndex(OrigPtr, DL); if (Idx < 0) return false; @@ -323,13 +870,22 @@ bool CDSPass::instrumentAtomicCall(CallInst *CI, const DataLayout &DL) { // atomic_init; args = {obj, order} if (funName.contains("atomic_init")) { + Value *OrigVal = parameters[1]; + Value *ptr = IRB.CreatePointerCast(OrigPtr, PtrTy); - Value *val = IRB.CreateBitOrPointerCast(parameters[1], Ty); + Value *val; + if (OrigVal->getType()->isPtrOrPtrVectorTy()) + val = IRB.CreatePointerCast(OrigVal, Ty); + else + val = IRB.CreateIntCast(OrigVal, Ty, true); + Value *args[] = {ptr, val, position}; - Instruction* funcInst=CallInst::Create(CDSAtomicInit[Idx], args); - ReplaceInstWithInst(CI, funcInst); + if (!checkSignature(CDSAtomicInit[Idx], args)) + return false; + Instruction* funcInst = CallInst::Create(CDSAtomicInit[Idx], args); + ReplaceInstWithInst(CI, funcInst); return true; } @@ -345,10 +901,35 @@ bool CDSPass::instrumentAtomicCall(CallInst *CI, const DataLayout &DL) { order = ConstantInt::get(OrdTy, (int) AtomicOrderingCABI::seq_cst); Value *args[] = {ptr, order, position}; - - Instruction* funcInst=CallInst::Create(CDSAtomicLoad[Idx], args); + + if (!checkSignature(CDSAtomicLoad[Idx], args)) + return false; + + Instruction* funcInst = CallInst::Create(CDSAtomicLoad[Idx], args); ReplaceInstWithInst(CI, funcInst); + return true; + } else if (funName.contains("atomic") && + funName.contains("load") ) { + // does this version of call always have an atomic order as an argument? + Value *ptr = IRB.CreatePointerCast(OrigPtr, PtrTy); + Value *order = IRB.CreateBitOrPointerCast(parameters[1], OrdTy); + Value *args[] = {ptr, order, position}; + + // Without this check, gdax does not compile :( + if (!CI->getType()->isPointerTy()) { + return false; + } + + if (!checkSignature(CDSAtomicLoad[Idx], args)) + return false; + + CallInst *funcInst = IRB.CreateCall(CDSAtomicLoad[Idx], args); + Value *RetVal = IRB.CreateIntToPtr(funcInst, CI->getType()); + + CI->replaceAllUsesWith(RetVal); + CI->eraseFromParent(); + return true; } @@ -366,8 +947,36 @@ bool CDSPass::instrumentAtomicCall(CallInst *CI, const DataLayout &DL) { order = ConstantInt::get(OrdTy, (int) AtomicOrderingCABI::seq_cst); Value *args[] = {ptr, val, order, position}; - - Instruction* funcInst=CallInst::Create(CDSAtomicStore[Idx], args); + + if (!checkSignature(CDSAtomicStore[Idx], args)) + return false; + + Instruction* funcInst = CallInst::Create(CDSAtomicStore[Idx], args); + ReplaceInstWithInst(CI, funcInst); + + return true; + } else if (funName.contains("atomic") && + funName.contains("store") ) { + // Does this version of call always have an atomic order as an argument? + if (parameters.size() < 3) + return false; + + Value *OrigVal = parameters[1]; + Value *ptr = IRB.CreatePointerCast(OrigPtr, PtrTy); + + Value *val; + if (OrigVal->getType()->isPtrOrPtrVectorTy()) + val = IRB.CreatePointerCast(OrigVal, Ty); + else + val = IRB.CreateIntCast(OrigVal, Ty, true); + + Value *order = IRB.CreateBitOrPointerCast(parameters[2], OrdTy); + Value *args[] = {ptr, val, order, position}; + + if (!checkSignature(CDSAtomicStore[Idx], args)) + return false; + + Instruction* funcInst = CallInst::Create(CDSAtomicStore[Idx], args); ReplaceInstWithInst(CI, funcInst); return true; @@ -375,7 +984,8 @@ bool CDSPass::instrumentAtomicCall(CallInst *CI, const DataLayout &DL) { // atomic_fetch_*; args = {obj, val, order} if (funName.contains("atomic_fetch_") || - funName.contains("atomic_exchange") ) { + funName.contains("atomic_exchange")) { + bool isExplicit = funName.contains("_explicit"); Value *OrigVal = parameters[1]; @@ -393,12 +1003,17 @@ bool CDSPass::instrumentAtomicCall(CallInst *CI, const DataLayout &DL) { else if ( funName.contains("atomic_exchange") ) op = AtomicRMWInst::Xchg; else { - errs() << "Unknown atomic read modify write operation\n"; + errs() << "Unknown atomic read-modify-write operation\n"; return false; } Value *ptr = IRB.CreatePointerCast(OrigPtr, PtrTy); - Value *val = IRB.CreatePointerCast(OrigVal, Ty); + Value *val; + if (OrigVal->getType()->isPtrOrPtrVectorTy()) + val = IRB.CreatePointerCast(OrigVal, Ty); + else + val = IRB.CreateIntCast(OrigVal, Ty, true); + Value *order; if (isExplicit) order = IRB.CreateBitOrPointerCast(parameters[2], OrdTy); @@ -406,8 +1021,54 @@ bool CDSPass::instrumentAtomicCall(CallInst *CI, const DataLayout &DL) { order = ConstantInt::get(OrdTy, (int) AtomicOrderingCABI::seq_cst); Value *args[] = {ptr, val, order, position}; - - Instruction* funcInst=CallInst::Create(CDSAtomicRMW[op][Idx], args); + + if (!checkSignature(CDSAtomicRMW[op][Idx], args)) + return false; + + Instruction* funcInst = CallInst::Create(CDSAtomicRMW[op][Idx], args); + ReplaceInstWithInst(CI, funcInst); + + return true; + } else if (funName.contains("fetch")) { + errs() << "atomic fetch captured. Not implemented yet. "; + errs() << "See source file :"; + getPosition(CI, IRB, true); + return false; + } else if (funName.contains("exchange") && + !funName.contains("compare_exchange") ) { + if (CI->getType()->isPointerTy()) { + /** + * TODO: instrument the following case + * mcs-lock.h + * std::atomic m_tail; + * struct T * me; + * struct T * pred = m_tail.exchange(me, memory_order_*); + */ + errs() << "atomic exchange captured. Not implemented yet. "; + errs() << "See source file :"; + getPosition(CI, IRB, true); + + return false; + } + + Value *OrigVal = parameters[1]; + + Value *ptr = IRB.CreatePointerCast(OrigPtr, PtrTy); + Value *val; + if (OrigVal->getType()->isPtrOrPtrVectorTy()) + val = IRB.CreatePointerCast(OrigVal, Ty); + else + val = IRB.CreateIntCast(OrigVal, Ty, true); + + Value *order = IRB.CreateBitOrPointerCast(parameters[2], OrdTy); + Value *args[] = {ptr, val, order, position}; + + int op = AtomicRMWInst::Xchg; + + if (!checkSignature(CDSAtomicRMW[op][Idx], args)) + return false; + + Instruction* funcInst = CallInst::Create(CDSAtomicRMW[op][Idx], args); ReplaceInstWithInst(CI, funcInst); return true; @@ -426,7 +1087,18 @@ bool CDSPass::instrumentAtomicCall(CallInst *CI, const DataLayout &DL) { Value *order_succ, *order_fail; if (isExplicit) { order_succ = IRB.CreateBitOrPointerCast(parameters[3], OrdTy); - order_fail = IRB.CreateBitOrPointerCast(parameters[4], OrdTy); + + if (parameters.size() > 4) { + order_fail = IRB.CreateBitOrPointerCast(parameters[4], OrdTy); + } else { + /* The failure order is not provided */ + order_fail = order_succ; + ConstantInt * order_succ_cast = dyn_cast(order_succ); + int index = order_succ_cast->getSExtValue(); + + order_fail = ConstantInt::get(OrdTy, + AtomicCasFailureOrderIndex(index)); + } } else { order_succ = ConstantInt::get(OrdTy, (int) AtomicOrderingCABI::seq_cst); @@ -436,356 +1108,137 @@ bool CDSPass::instrumentAtomicCall(CallInst *CI, const DataLayout &DL) { Value *args[] = {Addr, CmpOperand, NewOperand, order_succ, order_fail, position}; - - Instruction* funcInst=CallInst::Create(CDSAtomicCAS_V2[Idx], args); - ReplaceInstWithInst(CI, funcInst); - - return true; - } - - return false; -} - -static bool shouldInstrumentReadWriteFromAddress(const Module *M, Value *Addr) { - // Peel off GEPs and BitCasts. - Addr = Addr->stripInBoundsOffsets(); - - if (GlobalVariable *GV = dyn_cast(Addr)) { - if (GV->hasSection()) { - StringRef SectionName = GV->getSection(); - // Check if the global is in the PGO counters section. - auto OF = Triple(M->getTargetTriple()).getObjectFormat(); - if (SectionName.endswith( - getInstrProfSectionName(IPSK_cnts, OF, /*AddSegmentInfo=*/false))) - return false; - } - - // Check if the global is private gcov data. - if (GV->getName().startswith("__llvm_gcov") || - GV->getName().startswith("__llvm_gcda")) - return false; - } - - // Do not instrument acesses from different address spaces; we cannot deal - // with them. - if (Addr) { - Type *PtrTy = cast(Addr->getType()->getScalarType()); - if (PtrTy->getPointerAddressSpace() != 0) - return false; - } - - return true; -} - -bool CDSPass::addrPointsToConstantData(Value *Addr) { - // If this is a GEP, just analyze its pointer operand. - if (GetElementPtrInst *GEP = dyn_cast(Addr)) - Addr = GEP->getPointerOperand(); - - if (GlobalVariable *GV = dyn_cast(Addr)) { - if (GV->isConstant()) { - // Reads from constant globals can not race with any writes. - NumOmittedReadsFromConstantGlobals++; - return true; - } - } else if (LoadInst *L = dyn_cast(Addr)) { - if (isVtableAccess(L)) { - // Reads from a vtable pointer can not race with any writes. - NumOmittedReadsFromVtable++; - return true; - } - } - return false; -} - -bool CDSPass::runOnFunction(Function &F) { - if (F.getName() == "main") { - F.setName("user_main"); - errs() << "main replaced by user_main\n"; - } - - if (true) { - initializeCallbacks( *F.getParent() ); - - SmallVector AllLoadsAndStores; - SmallVector LocalLoadsAndStores; - SmallVector AtomicAccesses; - std::vector worklist; + if (!checkSignature(CDSAtomicCAS_V2[Idx], args)) + return false; - bool Res = false; - const DataLayout &DL = F.getParent()->getDataLayout(); + Instruction* funcInst = CallInst::Create(CDSAtomicCAS_V2[Idx], args); + ReplaceInstWithInst(CI, funcInst); - errs() << "--- " << F.getName() << "---\n"; + return true; + } else if ( funName.contains("compare_exchange_strong") || + funName.contains("compare_exchange_weak") ) { + Value *Addr = IRB.CreatePointerCast(OrigPtr, PtrTy); + Value *CmpOperand = IRB.CreatePointerCast(parameters[1], PtrTy); + Value *NewOperand = IRB.CreateBitOrPointerCast(parameters[2], Ty); - for (auto &B : F) { - for (auto &I : B) { - if ( (&I)->isAtomic() || isAtomicCall(&I) ) { - AtomicAccesses.push_back(&I); - } else if (isa(I) || isa(I)) { - LocalLoadsAndStores.push_back(&I); - } else if (isa(I) || isa(I)) { - // not implemented yet - } - } + Value *order_succ, *order_fail; + order_succ = IRB.CreateBitOrPointerCast(parameters[3], OrdTy); - chooseInstructionsToInstrument(LocalLoadsAndStores, AllLoadsAndStores, DL); - } + if (parameters.size() > 4) { + order_fail = IRB.CreateBitOrPointerCast(parameters[4], OrdTy); + } else { + /* The failure order is not provided */ + order_fail = order_succ; + ConstantInt * order_succ_cast = dyn_cast(order_succ); + int index = order_succ_cast->getSExtValue(); + + order_fail = ConstantInt::get(OrdTy, + AtomicCasFailureOrderIndex(index)); + } - for (auto Inst : AllLoadsAndStores) { -// Res |= instrumentLoadOrStore(Inst, DL); -// errs() << "load and store are replaced\n"; - } + Value *args[] = {Addr, CmpOperand, NewOperand, + order_succ, order_fail, position}; - for (auto Inst : AtomicAccesses) { - Res |= instrumentAtomic(Inst, DL); - } + if (!checkSignature(CDSAtomicCAS_V2[Idx], args)) + return false; - if (F.getName() == "user_main") { - // F.dump(); - } + Instruction* funcInst = CallInst::Create(CDSAtomicCAS_V2[Idx], args); + ReplaceInstWithInst(CI, funcInst); - } + return true; + } - return false; + return false; } -void CDSPass::chooseInstructionsToInstrument( - SmallVectorImpl &Local, SmallVectorImpl &All, - const DataLayout &DL) { - SmallPtrSet WriteTargets; - // Iterate from the end. - for (Instruction *I : reverse(Local)) { - if (StoreInst *Store = dyn_cast(I)) { - Value *Addr = Store->getPointerOperand(); - if (!shouldInstrumentReadWriteFromAddress(I->getModule(), Addr)) - continue; - WriteTargets.insert(Addr); - } else { - LoadInst *Load = cast(I); - Value *Addr = Load->getPointerOperand(); - if (!shouldInstrumentReadWriteFromAddress(I->getModule(), Addr)) - continue; - if (WriteTargets.count(Addr)) { - // We will write to this temp, so no reason to analyze the read. - NumOmittedReadsBeforeWrite++; - continue; - } - if (addrPointsToConstantData(Addr)) { - // Addr points to some constant data -- it can not race with any writes. - continue; - } - } - Value *Addr = isa(*I) - ? cast(I)->getPointerOperand() - : cast(I)->getPointerOperand(); - if (isa(GetUnderlyingObject(Addr, DL)) && - !PointerMayBeCaptured(Addr, true, true)) { - // The variable is addressable but not captured, so it cannot be - // referenced from a different thread and participate in a data race - // (see llvm/Analysis/CaptureTracking.h for details). - NumOmittedNonCaptured++; - continue; - } - All.push_back(I); - } - Local.clear(); +int CDSPass::getMemoryAccessFuncIndex(Value *Addr, + const DataLayout &DL) { + Type *OrigPtrTy = Addr->getType(); + Type *OrigTy = cast(OrigPtrTy)->getElementType(); + assert(OrigTy->isSized()); + uint32_t TypeSize = DL.getTypeStoreSizeInBits(OrigTy); + if (TypeSize != 8 && TypeSize != 16 && + TypeSize != 32 && TypeSize != 64 && TypeSize != 128) { + NumAccessesWithBadSize++; + // Ignore all unusual sizes. + return -1; + } + size_t Idx = countTrailingZeros(TypeSize / 8); + //assert(Idx < kNumberOfAccessSizes); + if (Idx >= kNumberOfAccessSizes) { + return -1; + } + return Idx; } +bool CDSPass::instrumentLoops(Function &F) +{ + DominatorTree DT(F); + LoopInfo LI(DT); + + SmallVector Loops = LI.getLoopsInPreorder(); + bool instrumented = false; + + // Do a post-order traversal of the loops so that counter updates can be + // iteratively hoisted outside the loop nest. + for (auto *Loop : llvm::reverse(Loops)) { + bool instrument_loop = false; + + // Iterator over loop blocks and search for atomics and volatiles + Loop::block_iterator it; + for (it = Loop->block_begin(); it != Loop->block_end(); it++) { + BasicBlock * block = *it; + for (auto &Inst : *block) { + if ( (&Inst)->isAtomic() ) { + instrument_loop = true; + break; + } else if (isAtomicCall(&Inst)) { + instrument_loop = true; + break; + } else if (isa(Inst) || isa(Inst)) { + LoadInst *LI = dyn_cast(&Inst); + StoreInst *SI = dyn_cast(&Inst); + bool isVolatile = ( LI ? LI->isVolatile() : SI->isVolatile() ); + + if (isVolatile) { + instrument_loop = true; + break; + } + } + } + + if (instrument_loop) + break; + } -bool CDSPass::instrumentLoadOrStore(Instruction *I, - const DataLayout &DL) { - IRBuilder<> IRB(I); - bool IsWrite = isa(*I); - Value *Addr = IsWrite - ? cast(I)->getPointerOperand() - : cast(I)->getPointerOperand(); - - // swifterror memory addresses are mem2reg promoted by instruction selection. - // As such they cannot have regular uses like an instrumentation function and - // it makes no sense to track them as memory. - if (Addr->isSwiftError()) - return false; - - int size = getTypeSize(Addr->getType()); - int index = sizetoindex(size); - -// not supported by CDS yet -/* if (IsWrite && isVtableAccess(I)) { - LLVM_DEBUG(dbgs() << " VPTR : " << *I << "\n"); - Value *StoredValue = cast(I)->getValueOperand(); - // StoredValue may be a vector type if we are storing several vptrs at once. - // In this case, just take the first element of the vector since this is - // enough to find vptr races. - if (isa(StoredValue->getType())) - StoredValue = IRB.CreateExtractElement( - StoredValue, ConstantInt::get(IRB.getInt32Ty(), 0)); - if (StoredValue->getType()->isIntegerTy()) - StoredValue = IRB.CreateIntToPtr(StoredValue, IRB.getInt8PtrTy()); - // Call TsanVptrUpdate. - IRB.CreateCall(TsanVptrUpdate, - {IRB.CreatePointerCast(Addr, IRB.getInt8PtrTy()), - IRB.CreatePointerCast(StoredValue, IRB.getInt8PtrTy())}); - NumInstrumentedVtableWrites++; - return true; - } - - if (!IsWrite && isVtableAccess(I)) { - IRB.CreateCall(TsanVptrLoad, - IRB.CreatePointerCast(Addr, IRB.getInt8PtrTy())); - NumInstrumentedVtableReads++; - return true; - } -*/ - - Value *OnAccessFunc = nullptr; - OnAccessFunc = IsWrite ? CDSStore[index] : CDSLoad[index]; - - Type *ArgType = IRB.CreatePointerCast(Addr, Addr->getType())->getType(); - - if ( ArgType != Int8PtrTy && ArgType != Int16PtrTy && - ArgType != Int32PtrTy && ArgType != Int64PtrTy ) { - //errs() << "A load or store of type "; - //errs() << *ArgType; - //errs() << " is passed in\n"; - return false; // if other types of load or stores are passed in - } - IRB.CreateCall(OnAccessFunc, IRB.CreatePointerCast(Addr, Addr->getType())); - if (IsWrite) NumInstrumentedWrites++; - else NumInstrumentedReads++; - return true; -} + if (instrument_loop) { + // TODO: what to instrument? + errs() << "Function: " << F.getName() << "\n"; + BasicBlock * header = Loop->getHeader(); + header->dump(); -// todo: replace getTypeSize with the getMemoryAccessFuncIndex -bool CDSPass::instrumentAtomic(Instruction * I, const DataLayout &DL) { - IRBuilder<> IRB(I); - // LLVMContext &Ctx = IRB.getContext(); - - if (auto *CI = dyn_cast(I)) { - return instrumentAtomicCall(CI, DL); - } - - Value *position = getPosition(I, IRB); - - if (StoreInst *SI = dyn_cast(I)) { - int atomic_order_index = getAtomicOrderIndex(SI->getOrdering()); - - Value *val = SI->getValueOperand(); - Value *ptr = SI->getPointerOperand(); - Value *order = ConstantInt::get(OrdTy, atomic_order_index); - Value *args[] = {ptr, val, order, position}; - - int size=getTypeSize(ptr->getType()); - int index=sizetoindex(size); - - Instruction* funcInst=CallInst::Create(CDSAtomicStore[index], args); - ReplaceInstWithInst(SI, funcInst); -// errs() << "Store replaced\n"; - } else if (LoadInst *LI = dyn_cast(I)) { - int atomic_order_index = getAtomicOrderIndex(LI->getOrdering()); - - Value *ptr = LI->getPointerOperand(); - Value *order = ConstantInt::get(OrdTy, atomic_order_index); - Value *args[] = {ptr, order, position}; - - int size=getTypeSize(ptr->getType()); - int index=sizetoindex(size); - - Instruction* funcInst=CallInst::Create(CDSAtomicLoad[index], args); - ReplaceInstWithInst(LI, funcInst); -// errs() << "Load Replaced\n"; - } else if (AtomicRMWInst *RMWI = dyn_cast(I)) { - int atomic_order_index = getAtomicOrderIndex(RMWI->getOrdering()); - - Value *val = RMWI->getValOperand(); - Value *ptr = RMWI->getPointerOperand(); - Value *order = ConstantInt::get(OrdTy, atomic_order_index); - Value *args[] = {ptr, val, order, position}; - - int size = getTypeSize(ptr->getType()); - int index = sizetoindex(size); - - Instruction* funcInst = CallInst::Create(CDSAtomicRMW[RMWI->getOperation()][index], args); - ReplaceInstWithInst(RMWI, funcInst); -// errs() << RMWI->getOperationName(RMWI->getOperation()); -// errs() << " replaced\n"; - } else if (AtomicCmpXchgInst *CASI = dyn_cast(I)) { - IRBuilder<> IRB(CASI); - - Value *Addr = CASI->getPointerOperand(); - - int size = getTypeSize(Addr->getType()); - int index = sizetoindex(size); - const unsigned ByteSize = 1U << index; - const unsigned BitSize = ByteSize * 8; - Type *Ty = Type::getIntNTy(IRB.getContext(), BitSize); - Type *PtrTy = Ty->getPointerTo(); - - Value *CmpOperand = IRB.CreateBitOrPointerCast(CASI->getCompareOperand(), Ty); - Value *NewOperand = IRB.CreateBitOrPointerCast(CASI->getNewValOperand(), Ty); - - int atomic_order_index_succ = getAtomicOrderIndex(CASI->getSuccessOrdering()); - int atomic_order_index_fail = getAtomicOrderIndex(CASI->getFailureOrdering()); - Value *order_succ = ConstantInt::get(OrdTy, atomic_order_index_succ); - Value *order_fail = ConstantInt::get(OrdTy, atomic_order_index_fail); - - Value *Args[] = {IRB.CreatePointerCast(Addr, PtrTy), - CmpOperand, NewOperand, - order_succ, order_fail, position}; - - CallInst *funcInst = IRB.CreateCall(CDSAtomicCAS_V1[index], Args); - Value *Success = IRB.CreateICmpEQ(funcInst, CmpOperand); - - Value *OldVal = funcInst; - Type *OrigOldValTy = CASI->getNewValOperand()->getType(); - if (Ty != OrigOldValTy) { - // The value is a pointer, so we need to cast the return value. - OldVal = IRB.CreateIntToPtr(funcInst, OrigOldValTy); - } - - Value *Res = - IRB.CreateInsertValue(UndefValue::get(CASI->getType()), OldVal, 0); - Res = IRB.CreateInsertValue(Res, Success, 1); - - I->replaceAllUsesWith(Res); - I->eraseFromParent(); - } else if (FenceInst *FI = dyn_cast(I)) { - int atomic_order_index = getAtomicOrderIndex(FI->getOrdering()); - Value *order = ConstantInt::get(OrdTy, atomic_order_index); - Value *Args[] = {order, position}; - - CallInst *funcInst = CallInst::Create(CDSAtomicThreadFence, Args); - ReplaceInstWithInst(FI, funcInst); -// errs() << "Thread Fences replaced\n"; - } - return true; -} + instrumented = true; + } + } -int CDSPass::getMemoryAccessFuncIndex(Value *Addr, - const DataLayout &DL) { - Type *OrigPtrTy = Addr->getType(); - Type *OrigTy = cast(OrigPtrTy)->getElementType(); - assert(OrigTy->isSized()); - uint32_t TypeSize = DL.getTypeStoreSizeInBits(OrigTy); - if (TypeSize != 8 && TypeSize != 16 && - TypeSize != 32 && TypeSize != 64 && TypeSize != 128) { - // NumAccessesWithBadSize++; - // Ignore all unusual sizes. - return -1; - } - size_t Idx = countTrailingZeros(TypeSize / 8); - // assert(Idx < FUNCARRAYSIZE); - return Idx; + return instrumented; } - char CDSPass::ID = 0; // Automatically enable the pass. static void registerCDSPass(const PassManagerBuilder &, - legacy::PassManagerBase &PM) { - PM.add(new CDSPass()); + legacy::PassManagerBase &PM) { + PM.add(new CDSPass()); } + +/* Enable the pass when opt level is greater than 0 */ +static RegisterStandardPasses + RegisterMyPass1(PassManagerBuilder::EP_OptimizerLast, +registerCDSPass); + +/* Enable the pass when opt level is 0 */ static RegisterStandardPasses - RegisterMyPass(PassManagerBuilder::EP_OptimizerLast, + RegisterMyPass2(PassManagerBuilder::EP_EnabledOnOptLevel0, registerCDSPass);