/// AddCatchInfo - Extract the personality and type infos from an eh.selector
/// call, and add them to the specified machine basic block.
void llvm::AddCatchInfo(CallInst &I, MachineModuleInfo *MMI,
MachineBasicBlock *MBB) {
// Inform the MachineModuleInfo of the personality for this landing pad.
ConstantExpr *CE = cast<ConstantExpr>(I.getOperand(2));
assert(CE->getOpcode() == Instruction::BitCast &&
isa<Function>(CE->getOperand(0)) &&
"Personality should be a function");
MMI->addPersonality(MBB, cast<Function>(CE->getOperand(0)));
// Gather all the type infos for this landing pad and pass them along to
// MachineModuleInfo.
std::vector<GlobalVariable *> TyInfo;
unsigned N = I.getNumOperands();
for (unsigned i = N - 1; i > 2; --i) {
if (ConstantInt *CI = dyn_cast<ConstantInt>(I.getOperand(i))) {
unsigned FilterLength = CI->getZExtValue();
unsigned FirstCatch = i + FilterLength + !FilterLength;
assert (FirstCatch <= N && "Invalid filter length");
if (FirstCatch < N) {
TyInfo.reserve(N - FirstCatch);
for (unsigned j = FirstCatch; j < N; ++j)
TyInfo.push_back(ExtractTypeInfo(I.getOperand(j)));
MMI->addCatchTypeInfo(MBB, TyInfo);
TyInfo.clear();
}
if (!FilterLength) {
// Cleanup.
MMI->addCleanup(MBB);
} else {
// Filter.
TyInfo.reserve(FilterLength - 1);
for (unsigned j = i + 1; j < FirstCatch; ++j)
TyInfo.push_back(ExtractTypeInfo(I.getOperand(j)));
MMI->addFilterTypeInfo(MBB, TyInfo);
TyInfo.clear();
}
N = i;
}
}
if (N > 3) {
TyInfo.reserve(N - 3);
for (unsigned j = 3; j < N; ++j)
TyInfo.push_back(ExtractTypeInfo(I.getOperand(j)));
MMI->addCatchTypeInfo(MBB, TyInfo);
}
}
/// IsConstantOffsetFromGlobal - If this constant is actually a constant offset
/// from a global, return the global and the constant. Because of
/// constantexprs, this function is recursive.
static bool IsConstantOffsetFromGlobal(Constant *C, GlobalValue *&GV,
int64_t &Offset, const TargetData &TD) {
// Trivial case, constant is the global.
if ((GV = dyn_cast<GlobalValue>(C))) {
Offset = 0;
return true;
}
// Otherwise, if this isn't a constant expr, bail out.
ConstantExpr *CE = dyn_cast<ConstantExpr>(C);
if (!CE) return false;
// Look through ptr->int and ptr->ptr casts.
if (CE->getOpcode() == Instruction::PtrToInt ||
CE->getOpcode() == Instruction::BitCast)
return IsConstantOffsetFromGlobal(CE->getOperand(0), GV, Offset, TD);
// i32* getelementptr ([5 x i32]* @a, i32 0, i32 5)
if (CE->getOpcode() == Instruction::GetElementPtr) {
// Cannot compute this if the element type of the pointer is missing size
// info.
if (!cast<PointerType>(CE->getOperand(0)->getType())->getElementType()->isSized())
return false;
// If the base isn't a global+constant, we aren't either.
if (!IsConstantOffsetFromGlobal(CE->getOperand(0), GV, Offset, TD))
return false;
// Otherwise, add any offset that our operands provide.
gep_type_iterator GTI = gep_type_begin(CE);
for (unsigned i = 1, e = CE->getNumOperands(); i != e; ++i, ++GTI) {
ConstantInt *CI = dyn_cast<ConstantInt>(CE->getOperand(i));
if (!CI) return false; // Index isn't a simple constant?
if (CI->getZExtValue() == 0) continue; // Not adding anything.
if (const StructType *ST = dyn_cast<StructType>(*GTI)) {
// N = N + Offset
Offset += TD.getStructLayout(ST)->getElementOffset(CI->getZExtValue());
} else {
const SequentialType *SQT = cast<SequentialType>(*GTI);
Offset += TD.getTypeSize(SQT->getElementType())*CI->getSExtValue();
}
}
return true;
}
return false;
}
void ClassHierarchyUtils::findClassHierarchy(Module& M) {
// extract call hierarchy using std::type_info structures rather
// than debug info. The former works even for when there are anonymous
// namespaces. type_info structs can be obtained from vtable globals.
// (for more info, see http://mentorembedded.github.io/cxx-abi/abi.html)
for (Module::global_iterator I=M.global_begin(), E=M.global_end(); I != E; I++) {
GlobalVariable* G = &*I;
if (G->getName().startswith("_ZTV")) {
if (G->hasInitializer()) {
SDEBUG("soaap.util.classhierarchy", 3, G->dump());
ConstantArray* Ginit = cast<ConstantArray>(G->getInitializer());
// Ginit[1] is the type_info global for this vtable's type
bool typeInfoFound = false;
bool primaryVTable = true;
for (int i=0; i<Ginit->getNumOperands(); i++) {
// typeinfo will be the first global variable in the array.
// It's not always at a fixed index so we have to search for it...
if (GlobalVariable* TI = dyn_cast<GlobalVariable>(Ginit->getOperand(i)->stripPointerCasts())) {
//TI->dump();
typeInfoToVTable[TI] = G;
vTableToTypeInfo[G] = TI;
processTypeInfo(TI); // process TI recursively (it contains refs to super-class TIs)
typeInfoFound = true;
if (primaryVTable) {
primaryVTable = false;
vTableToSecondaryVTableMaps[G][0] = i+1; // i+1 is also the size of the vtable header
}
else {
// offset_to_top is at the previous index
ConstantExpr* offsetValCast = cast<ConstantExpr>(Ginit->getOperand(i-1)->stripPointerCasts());
ConstantInt* offsetVal = cast<ConstantInt>(offsetValCast->getOperand(0));
int offsetToTop = offsetVal->getSExtValue();
if (offsetToTop > 0) {
dbgs() << "ERROR: offsetToTop is positive!\n";
G->dump();
}
else {
offsetToTop *= -1;
}
SDEBUG("soaap.util.classhierarchy", 3, dbgs() << "offsetToTop: " << offsetToTop << "\n");
vTableToSecondaryVTableMaps[G][offsetToTop] = i+1;
}
}
}
if (!typeInfoFound) {
dbgs() << "ERROR: vtable initializer is not a global variable...\n";
dbgs() << *G << " = " << *Ginit << "\n";
}
}
}
}
SDEBUG("soaap.util.classhierarchy", 3, ppClassHierarchy(classToSubclasses));
}
void ControlFlowIntegrity::PatchDtorFunctions(void) {
GlobalVariable *global_dtors = _M.getNamedGlobal("llvm.global_dtors");
// check for dtor section
if (!global_dtors || !global_dtors->getOperand(0)) {
return;
}
Constant *c = global_dtors->getInitializer();
if (!c)
report_fatal_error("llvm.global_dtors without initializer!", false);
ConstantArray *CA = dyn_cast<ConstantArray>(c);
if (!CA)
report_fatal_error("Cast to ConstantArray failed", true);
for (Value *Op : ValueOpRange(*CA)) {
ConstantStruct *CS = dyn_cast<ConstantStruct>(Op);
if (!CS)
report_fatal_error("Cast to ConstantStruct failed", true);
Constant *FP = CS->getOperand(1);
if (FP->isNullValue())
break; // found a NULL termintator, stop here
Function *F = dyn_cast_or_null<Function>(FP);
if (F == NULL) {
// Strip off constant expression cast
ConstantExpr *CE = dyn_cast<ConstantExpr>(FP);
if (!CE)
report_fatal_error("Cast to ConstantExpr failed", true);
if (CE->isCast()) {
FP = CE->getOperand(0);
}
F = dyn_cast_or_null<Function>(FP);
}
if (!F)
report_fatal_error("Cast to Function failed", true);
// set visibility to hidden (do not export), unless it is already
// local (for ex. static), in which case we have to make it non-static
if (F->hasLocalLinkage()) {
F->setLinkage(llvm::GlobalValue::ExternalLinkage);
}
F->setVisibility(GlobalValue::HiddenVisibility);
_FiniFunctions.insert(F->getName());
}
if (global_dtors->getNumUses() > 0)
report_fatal_error("llvm.global_dtors uses count is > 0!", false);
global_dtors->removeFromParent();
}
/// Return true if the specified constant can be handled by the code generator.
/// We don't want to generate something like:
/// void *X = &X/42;
/// because the code generator doesn't have a relocation that can handle that.
///
/// This function should be called if C was not found (but just got inserted)
/// in SimpleConstants to avoid having to rescan the same constants all the
/// time.
static bool
isSimpleEnoughValueToCommitHelper(Constant *C,
SmallPtrSetImpl<Constant *> &SimpleConstants,
const DataLayout &DL) {
// Simple global addresses are supported, do not allow dllimport or
// thread-local globals.
if (auto *GV = dyn_cast<GlobalValue>(C))
return !GV->hasDLLImportStorageClass() && !GV->isThreadLocal();
// Simple integer, undef, constant aggregate zero, etc are all supported.
if (C->getNumOperands() == 0 || isa<BlockAddress>(C))
return true;
// Aggregate values are safe if all their elements are.
if (isa<ConstantArray>(C) || isa<ConstantStruct>(C) ||
isa<ConstantVector>(C)) {
for (Value *Op : C->operands())
if (!isSimpleEnoughValueToCommit(cast<Constant>(Op), SimpleConstants, DL))
return false;
return true;
}
// We don't know exactly what relocations are allowed in constant expressions,
// so we allow &global+constantoffset, which is safe and uniformly supported
// across targets.
ConstantExpr *CE = cast<ConstantExpr>(C);
switch (CE->getOpcode()) {
case Instruction::BitCast:
// Bitcast is fine if the casted value is fine.
return isSimpleEnoughValueToCommit(CE->getOperand(0), SimpleConstants, DL);
case Instruction::IntToPtr:
case Instruction::PtrToInt:
// int <=> ptr is fine if the int type is the same size as the
// pointer type.
if (DL.getTypeSizeInBits(CE->getType()) !=
DL.getTypeSizeInBits(CE->getOperand(0)->getType()))
return false;
return isSimpleEnoughValueToCommit(CE->getOperand(0), SimpleConstants, DL);
// GEP is fine if it is simple + constant offset.
case Instruction::GetElementPtr:
for (unsigned i = 1, e = CE->getNumOperands(); i != e; ++i)
if (!isa<ConstantInt>(CE->getOperand(i)))
return false;
return isSimpleEnoughValueToCommit(CE->getOperand(0), SimpleConstants, DL);
case Instruction::Add:
// We allow simple+cst.
if (!isa<ConstantInt>(CE->getOperand(1)))
return false;
return isSimpleEnoughValueToCommit(CE->getOperand(0), SimpleConstants, DL);
}
return false;
}
static GlobalValue *getAliaseeGV(GlobalAlias *GA) {
Constant *C = GA->getAliasee();
assert(C && "Must alias something");
if (GlobalValue *GV = dyn_cast<GlobalValue>(C))
return GV;
ConstantExpr *CE = cast<ConstantExpr>(C);
assert((CE->getOpcode() == Instruction::BitCast ||
CE->getOpcode() == Instruction::AddrSpaceCast ||
CE->getOpcode() == Instruction::GetElementPtr) &&
"Unsupported aliasee");
return cast<GlobalValue>(CE->getOperand(0));
}
GlobalValue *GlobalAlias::getAliasedGlobal() {
Constant *C = getAliasee();
if (C == 0) return 0;
if (GlobalValue *GV = dyn_cast<GlobalValue>(C))
return GV;
ConstantExpr *CE = cast<ConstantExpr>(C);
assert((CE->getOpcode() == Instruction::BitCast ||
CE->getOpcode() == Instruction::AddrSpaceCast ||
CE->getOpcode() == Instruction::GetElementPtr) &&
"Unsupported aliasee");
return cast<GlobalValue>(CE->getOperand(0));
}
static Constant*
sd_getDestructorFunction(Constant* vtblElement) {
ConstantExpr* bcExpr = NULL;
// if this a constant bitcast expression, this might be a vthunk
if ((bcExpr = dyn_cast<ConstantExpr>(vtblElement)) && bcExpr->getOpcode() == BITCAST_OPCODE) {
Constant* operand = bcExpr->getOperand(0);
// this is a vthunk
if (sd_isDestructorName(operand->getName())) {
return operand;
}
}
return NULL;
}
static Constant*
sd_isRTTI(Constant* vtblElement) {
ConstantExpr* bcExpr = NULL;
// if this a constant bitcast expression, this might be a vthunk
if ((bcExpr = dyn_cast<ConstantExpr>(vtblElement)) &&
bcExpr->getOpcode() == Instruction::BitCast) {
Constant* operand = bcExpr->getOperand(0);
// this is a vthunk
if (operand->getName().startswith("_ZTI")) {
return operand;
}
}
return NULL;
}
std::string readAnnotate(Function *f) {
std::string annotation = "";
// Get annotation variable
GlobalVariable *glob =
f->getParent()->getGlobalVariable("llvm.global.annotations");
if (glob != NULL) {
// Get the array
if (ConstantArray *ca = dyn_cast<ConstantArray>(glob->getInitializer())) {
for (unsigned i = 0; i < ca->getNumOperands(); ++i) {
// Get the struct
if (ConstantStruct *structAn =
dyn_cast<ConstantStruct>(ca->getOperand(i))) {
if (ConstantExpr *expr =
dyn_cast<ConstantExpr>(structAn->getOperand(0))) {
// If it's a bitcast we can check if the annotation is concerning
// the current function
if (expr->getOpcode() == Instruction::BitCast &&
expr->getOperand(0) == f) {
ConstantExpr *note = cast<ConstantExpr>(structAn->getOperand(1));
// If it's a GetElementPtr, that means we found the variable
// containing the annotations
if (note->getOpcode() == Instruction::GetElementPtr) {
if (GlobalVariable *annoteStr =
dyn_cast<GlobalVariable>(note->getOperand(0))) {
if (ConstantDataSequential *data =
dyn_cast<ConstantDataSequential>(
annoteStr->getInitializer())) {
if (data->isString()) {
annotation += data->getAsString().lower() + " ";
}
}
}
}
}
}
}
}
}
}
return annotation;
}
void ControlFlowIntegrity::ParseAnnotations(void) {
GlobalVariable *global_ctors = _M.getNamedGlobal("llvm.global.annotations");
// check for ctor section
if (!global_ctors || !global_ctors->getOperand(0)) {
return;
}
Constant *c = global_ctors->getInitializer();
if (!c)
report_fatal_error("llvm.global.annotations without initializer!", false);
ConstantArray *CA = dyn_cast<ConstantArray>(c);
if (!CA)
report_fatal_error("Cast to ConstantArray failed", true);
for (Value *Op : ValueOpRange(*CA)) {
ConstantStruct *CS = dyn_cast<ConstantStruct>(Op);
if (!CS)
report_fatal_error("Cast to ConstantStruct failed", true);
Constant *FP = CS->getOperand(0);
if (FP->isNullValue())
break; // found a NULL termintator, stop here
ConstantExpr *CE;
Function *F = dyn_cast_or_null<Function>(FP);
if (F == NULL) {
// Strip off constant expression cast
CE = dyn_cast<ConstantExpr>(FP);
if (!CE)
report_fatal_error("Cast to ConstantExpr failed", true);
if (CE->isCast()) {
FP = CE->getOperand(0);
F = dyn_cast_or_null<Function>(FP);
}
}
if (!F)
report_fatal_error("Cast to Function failed", true);
Constant *SP = CS->getOperand(1);
if (SP->isNullValue())
break; // found a NULL termintator, stop here
// Strip off constant expression cast
CE = dyn_cast<ConstantExpr>(SP);
if (!CE)
report_fatal_error("Cast to ConstantExpr failed", true);
if (CE->isCast()) {
SP = CE->getOperand(0);
}
Value *V = SP->getOperand(0);
GlobalVariable *GV = dyn_cast_or_null<GlobalVariable>(V);
if (!GV)
report_fatal_error("Cast to GlobalVariable failed", false);
assert(GV && "cast to GlobalVariable failed");
Constant *cval = GV->getInitializer();
const StringRef s = (cast<ConstantDataArray>(cval))->getAsCString();
if (s == ANNOTATION_IGNORE_BLOCK) {
_IgnoredBlocksFunctions.insert(F->getName());
}
}
if (global_ctors->getNumUses() > 0)
report_fatal_error("llvm.global.annotations uses count is > 0", false);
}
//
// Method: runOnModule()
//
// Description:
// Entry point for this LLVM pass.
// Search for all call sites to casted functions.
// Check if they only differ in an argument type
// Cast the argument, and call the original function
//
// Inputs:
// M - A reference to the LLVM module to transform
//
// Outputs:
// M - The transformed LLVM module.
//
// Return value:
// true - The module was modified.
// false - The module was not modified.
//
bool ArgCast::runOnModule(Module& M) {
std::vector<CallInst*> worklist;
for (Module::iterator I = M.begin(); I != M.end(); ++I) {
if (I->mayBeOverridden())
continue;
// Find all uses of this function
for(Value::user_iterator ui = I->user_begin(), ue = I->user_end(); ui != ue; ) {
// check if is ever casted to a different function type
ConstantExpr *CE = dyn_cast<ConstantExpr>(*ui++);
if(!CE)
continue;
if (CE->getOpcode() != Instruction::BitCast)
continue;
if(CE->getOperand(0) != I)
continue;
const PointerType *PTy = dyn_cast<PointerType>(CE->getType());
if (!PTy)
continue;
const Type *ETy = PTy->getElementType();
const FunctionType *FTy = dyn_cast<FunctionType>(ETy);
if(!FTy)
continue;
// casting to a varargs funtion
// or function with same number of arguments
// possibly varying types of arguments
if(FTy->getNumParams() != I->arg_size() && !FTy->isVarArg())
continue;
for(Value::user_iterator uii = CE->user_begin(),
uee = CE->user_end(); uii != uee; ++uii) {
// Find all uses of the casted value, and check if it is
// used in a Call Instruction
if (CallInst* CI = dyn_cast<CallInst>(*uii)) {
// Check that it is the called value, and not an argument
if(CI->getCalledValue() != CE)
continue;
// Check that the number of arguments passed, and expected
// by the function are the same.
if(!I->isVarArg()) {
if(CI->getNumOperands() != I->arg_size() + 1)
continue;
} else {
if(CI->getNumOperands() < I->arg_size() + 1)
continue;
}
// If so, add to worklist
worklist.push_back(CI);
}
}
}
}
// Proces the worklist of potential call sites to transform
while(!worklist.empty()) {
CallInst *CI = worklist.back();
worklist.pop_back();
// Get the called Function
Function *F = cast<Function>(CI->getCalledValue()->stripPointerCasts());
const FunctionType *FTy = F->getFunctionType();
SmallVector<Value*, 8> Args;
unsigned i =0;
for(i =0; i< FTy->getNumParams(); ++i) {
Type *ArgType = CI->getOperand(i+1)->getType();
Type *FormalType = FTy->getParamType(i);
// If the types for this argument match, just add it to the
// parameter list. No cast needs to be inserted.
if(ArgType == FormalType) {
Args.push_back(CI->getOperand(i+1));
}
else if(ArgType->isPointerTy() && FormalType->isPointerTy()) {
CastInst *CastI = CastInst::CreatePointerCast(CI->getOperand(i+1),
FormalType, "", CI);
Args.push_back(CastI);
} else if (ArgType->isIntegerTy() && FormalType->isIntegerTy()) {
unsigned SrcBits = ArgType->getScalarSizeInBits();
unsigned DstBits = FormalType->getScalarSizeInBits();
if(SrcBits > DstBits) {
CastInst *CastI = CastInst::CreateIntegerCast(CI->getOperand(i+1),
FormalType, true, "", CI);
Args.push_back(CastI);
} else {
if (F->getAttributes().hasAttribute(i+1, Attribute::SExt)) {
CastInst *CastI = CastInst::CreateIntegerCast(CI->getOperand(i+1),
FormalType, true, "", CI);
Args.push_back(CastI);
} else if (F->getAttributes().hasAttribute(i+1, Attribute::ZExt)) {
CastInst *CastI = CastInst::CreateIntegerCast(CI->getOperand(i+1),
FormalType, false, "", CI);
Args.push_back(CastI);
} else {
// Use ZExt in default case.
// Derived from InstCombine. Also, the only reason this should happen
// is mismatched prototypes.
// Seen in case of integer constants which get interpreted as i32,
// even if being used as i64.
// TODO: is this correct?
CastInst *CastI = CastInst::CreateIntegerCast(CI->getOperand(i+1),
FormalType, false, "", CI);
Args.push_back(CastI);
}
}
} else {
DEBUG(ArgType->dump());
DEBUG(FormalType->dump());
break;
}
}
// If we found an argument we could not cast, try the next instruction
if(i != FTy->getNumParams()) {
continue;
}
if(FTy->isVarArg()) {
for(; i< CI->getNumOperands() - 1 ;i++) {
Args.push_back(CI->getOperand(i+1));
}
}
// else replace the call instruction
CallInst *CINew = CallInst::Create(F, Args, "", CI);
CINew->setCallingConv(CI->getCallingConv());
CINew->setAttributes(CI->getAttributes());
if(!CI->use_empty()) {
CastInst *RetCast;
if(CI->getType() != CINew->getType()) {
if(CI->getType()->isPointerTy() && CINew->getType()->isPointerTy())
RetCast = CastInst::CreatePointerCast(CINew, CI->getType(), "", CI);
else if(CI->getType()->isIntOrIntVectorTy() && CINew->getType()->isIntOrIntVectorTy())
RetCast = CastInst::CreateIntegerCast(CINew, CI->getType(), false, "", CI);
else if(CI->getType()->isIntOrIntVectorTy() && CINew->getType()->isPointerTy())
RetCast = CastInst::CreatePointerCast(CINew, CI->getType(), "", CI);
else if(CI->getType()->isPointerTy() && CINew->getType()->isIntOrIntVectorTy())
RetCast = new IntToPtrInst(CINew, CI->getType(), "", CI);
else {
// TODO: I'm not sure what right behavior is here, but this case should be handled.
llvm_unreachable("Unexpected type conversion in call!");
abort();
}
CI->replaceAllUsesWith(RetCast);
} else {
CI->replaceAllUsesWith(CINew);
}
}
// Debug printing
DEBUG(errs() << "ARGCAST:");
DEBUG(errs() << "ERASE:");
DEBUG(CI->dump());
DEBUG(errs() << "ARGCAST:");
DEBUG(errs() << "ADDED:");
DEBUG(CINew->dump());
CI->eraseFromParent();
numChanged++;
}
return true;
}
virtual bool runOnFunction(Function &F) {
//F.dump();
bool changed = false;
for (inst_iterator inst_it = inst_begin(F), _inst_end = inst_end(F); inst_it != _inst_end; ++inst_it) {
LoadInst *li = dyn_cast<LoadInst>(&*inst_it);
if (!li) continue;
ConstantExpr *ce = dyn_cast<ConstantExpr>(li->getOperand(0));
// Not 100% sure what the isGEPWithNoNotionalOverIndexing() means, but
// at least it checks if it's a gep:
if (ce && ce->isGEPWithNoNotionalOverIndexing() && ce->getOperand(0)->getType() == g.llvm_flavor_type_ptr) {
changed = handleFlavor(li, ce);
}
GlobalVariable *gv = dyn_cast<GlobalVariable>(li->getOperand(0));
if (!gv) continue;
llvm::Type* gv_t = gv->getType();
if (gv_t == g.llvm_bool_type_ptr->getPointerTo()) {
changed = handleBool(li, gv) || changed;
continue;
}
if (gv_t == g.llvm_class_type_ptr->getPointerTo()) {
changed = handleCls(li, gv) || changed;
continue;
}
}
return changed;
}
