bool GlobalMerge::doInitialization(Module &M) {
DenseMap<unsigned, SmallVector<GlobalVariable*, 16> > Globals, ConstGlobals,
BSSGlobals;
const DataLayout *TD = TLI->getDataLayout();
unsigned MaxOffset = TLI->getMaximalGlobalOffset();
bool Changed = false;
// Grab all non-const globals.
for (Module::global_iterator I = M.global_begin(),
E = M.global_end(); I != E; ++I) {
// Merge is safe for "normal" internal globals only
if (!I->hasLocalLinkage() || I->isThreadLocal() || I->hasSection())
continue;
PointerType *PT = dyn_cast<PointerType>(I->getType());
assert(PT && "Global variable is not a pointer!");
unsigned AddressSpace = PT->getAddressSpace();
// Ignore fancy-aligned globals for now.
unsigned Alignment = TD->getPreferredAlignment(I);
Type *Ty = I->getType()->getElementType();
if (Alignment > TD->getABITypeAlignment(Ty))
continue;
// Ignore all 'special' globals.
if (I->getName().startswith("llvm.") ||
I->getName().startswith(".llvm."))
continue;
if (TD->getTypeAllocSize(Ty) < MaxOffset) {
if (TargetLoweringObjectFile::getKindForGlobal(I, TLI->getTargetMachine())
.isBSSLocal())
BSSGlobals[AddressSpace].push_back(I);
else if (I->isConstant())
ConstGlobals[AddressSpace].push_back(I);
else
Globals[AddressSpace].push_back(I);
}
}
for (DenseMap<unsigned, SmallVector<GlobalVariable*, 16> >::iterator
I = Globals.begin(), E = Globals.end(); I != E; ++I)
if (I->second.size() > 1)
Changed |= doMerge(I->second, M, false, I->first);
for (DenseMap<unsigned, SmallVector<GlobalVariable*, 16> >::iterator
I = BSSGlobals.begin(), E = BSSGlobals.end(); I != E; ++I)
if (I->second.size() > 1)
Changed |= doMerge(I->second, M, false, I->first);
// FIXME: This currently breaks the EH processing due to way how the
// typeinfo detection works. We might want to detect the TIs and ignore
// them in the future.
// if (ConstGlobals.size() > 1)
// Changed |= doMerge(ConstGlobals, M, true);
return Changed;
}
bool ARMGlobalMerge::doInitialization(Module &M) {
SmallVector<GlobalVariable*, 16> Globals, ConstGlobals, BSSGlobals;
const TargetData *TD = TLI->getTargetData();
unsigned MaxOffset = TLI->getMaximalGlobalOffset();
bool Changed = false;
// Disable this pass on darwin. The debugger is not yet ready to extract
// variable's info from a merged global.
if (TLI->getTargetMachine().getSubtarget<ARMSubtarget>().isTargetDarwin())
return false;
// Grab all non-const globals.
for (Module::global_iterator I = M.global_begin(),
E = M.global_end(); I != E; ++I) {
// Merge is safe for "normal" internal globals only
if (!I->hasLocalLinkage() || I->isThreadLocal() || I->hasSection())
continue;
// Ignore fancy-aligned globals for now.
if (I->getAlignment() != 0)
continue;
// Ignore all 'special' globals.
if (I->getName().startswith("llvm.") ||
I->getName().startswith(".llvm."))
continue;
if (TD->getTypeAllocSize(I->getType()->getElementType()) < MaxOffset) {
const TargetLoweringObjectFile &TLOF = TLI->getObjFileLowering();
if (TLOF.getKindForGlobal(I, TLI->getTargetMachine()).isBSSLocal())
BSSGlobals.push_back(I);
else if (I->isConstant())
ConstGlobals.push_back(I);
else
Globals.push_back(I);
}
}
if (Globals.size() > 1)
Changed |= doMerge(Globals, M, false);
if (BSSGlobals.size() > 1)
Changed |= doMerge(BSSGlobals, M, false);
// FIXME: This currently breaks the EH processing due to way how the
// typeinfo detection works. We might want to detect the TIs and ignore
// them in the future.
// if (ConstGlobals.size() > 1)
// Changed |= doMerge(ConstGlobals, M, true);
return Changed;
}
/// SplitFunctionsOutOfModule - Given a module and a list of functions in the
/// module, split the functions OUT of the specified module, and place them in
/// the new module.
Module *
llvm::SplitFunctionsOutOfModule(Module *M,
const std::vector<Function*> &F,
DenseMap<const Value*, Value*> &ValueMap) {
// Make sure functions & globals are all external so that linkage
// between the two modules will work.
for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I)
I->setLinkage(GlobalValue::ExternalLinkage);
for (Module::global_iterator I = M->global_begin(), E = M->global_end();
I != E; ++I) {
if (I->hasName() && I->getName()[0] == '\01')
I->setName(I->getName().substr(1));
I->setLinkage(GlobalValue::ExternalLinkage);
}
DenseMap<const Value*, Value*> NewValueMap;
Module *New = CloneModule(M, NewValueMap);
// Make sure global initializers exist only in the safe module (CBE->.so)
for (Module::global_iterator I = New->global_begin(), E = New->global_end();
I != E; ++I)
I->setInitializer(0); // Delete the initializer to make it external
// Remove the Test functions from the Safe module
std::set<Function *> TestFunctions;
for (unsigned i = 0, e = F.size(); i != e; ++i) {
Function *TNOF = cast<Function>(ValueMap[F[i]]);
DEBUG(errs() << "Removing function ");
DEBUG(WriteAsOperand(errs(), TNOF, false));
DEBUG(errs() << "\n");
TestFunctions.insert(cast<Function>(NewValueMap[TNOF]));
DeleteFunctionBody(TNOF); // Function is now external in this module!
}
// Remove the Safe functions from the Test module
for (Module::iterator I = New->begin(), E = New->end(); I != E; ++I)
if (!TestFunctions.count(I))
DeleteFunctionBody(I);
// Make sure that there is a global ctor/dtor array in both halves of the
// module if they both have static ctor/dtor functions.
SplitStaticCtorDtor("llvm.global_ctors", M, New, NewValueMap);
SplitStaticCtorDtor("llvm.global_dtors", M, New, NewValueMap);
return New;
}
/// GetAllUndefinedSymbols - calculates the set of undefined symbols that still
/// exist in an LLVM module. This is a bit tricky because there may be two
/// symbols with the same name but different LLVM types that will be resolved to
/// each other but aren't currently (thus we need to treat it as resolved).
///
/// Inputs:
/// M - The module in which to find undefined symbols.
///
/// Outputs:
/// UndefinedSymbols - A set of C++ strings containing the name of all
/// undefined symbols.
///
static void
GetAllUndefinedSymbols(Module *M, std::set<std::string> &UndefinedSymbols) {
std::set<std::string> DefinedSymbols;
UndefinedSymbols.clear();
// If the program doesn't define a main, try pulling one in from a .a file.
// This is needed for programs where the main function is defined in an
// archive, such f2c'd programs.
Function *Main = M->getFunction("main");
if (Main == 0 || Main->isDeclaration())
UndefinedSymbols.insert("main");
for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I)
if (I->hasName()) {
if (I->isDeclaration())
UndefinedSymbols.insert(I->getName());
else if (!I->hasLocalLinkage()) {
assert(!I->hasDLLImportLinkage()
&& "Found dllimported non-external symbol!");
DefinedSymbols.insert(I->getName());
}
}
for (Module::global_iterator I = M->global_begin(), E = M->global_end();
I != E; ++I)
if (I->hasName()) {
if (I->isDeclaration())
UndefinedSymbols.insert(I->getName());
else if (!I->hasLocalLinkage()) {
assert(!I->hasDLLImportLinkage()
&& "Found dllimported non-external symbol!");
DefinedSymbols.insert(I->getName());
}
}
for (Module::alias_iterator I = M->alias_begin(), E = M->alias_end();
I != E; ++I)
if (I->hasName())
DefinedSymbols.insert(I->getName());
// Prune out any defined symbols from the undefined symbols set...
for (std::set<std::string>::iterator I = UndefinedSymbols.begin();
I != UndefinedSymbols.end(); )
if (DefinedSymbols.count(*I))
UndefinedSymbols.erase(I++); // This symbol really is defined!
else
++I; // Keep this symbol in the undefined symbols list
}
void MemoryInstrumenter::instrumentGlobals(Module &M) {
TargetData &TD = getAnalysis<TargetData>();
IDAssigner &IDA = getAnalysis<IDAssigner>();
// Function HookGlobalsAlloc contains only one basic block.
// The BB iterates through all global variables, and calls HookMemAlloc
// for each of them.
BasicBlock *BB = BasicBlock::Create(M.getContext(), "entry",
GlobalsAllocHook);
Instruction *Ret = ReturnInst::Create(M.getContext(), BB);
for (Module::global_iterator GI = M.global_begin(), E = M.global_end();
GI != E; ++GI) {
// We are going to delete llvm.global_ctors.
// Therefore, don't instrument it.
if (GI->getName() == "llvm.global_ctors")
continue;
// Prevent global variables from sharing the same address, because it
// breaks the assumption that global variables do not alias.
// The same goes to functions.
if (GI->hasUnnamedAddr()) {
GI->setUnnamedAddr(false);
}
uint64_t TypeSize = TD.getTypeStoreSize(GI->getType()->getElementType());
instrumentMemoryAllocation(GI,
ConstantInt::get(LongType, TypeSize),
NULL,
Ret);
instrumentPointer(GI, NULL, Ret);
}
for (Module::iterator F = M.begin(); F != M.end(); ++F) {
// These hooks added by us don't have a value ID.
if (MemAllocHook == F || MainArgsAllocHook == F || TopLevelHook == F ||
AddrTakenHook == F || CallHook == F || ReturnHook == F ||
GlobalsAllocHook == F || MemHooksIniter == F || AfterForkHook == F ||
BeforeForkHook == F) {
continue;
}
// InvalidID: maybe this is inserted by alias checker in hybrid mode.
if (IDA.getValueID(F) == IDAssigner::InvalidID)
continue;
// Ignore intrinsic functions because we cannot take the address of
// an intrinsic. Also, no function pointers will point to instrinsic
// functions.
if (F->isIntrinsic())
continue;
// Prevent functions from sharing the same address.
if (F->hasUnnamedAddr()) {
F->setUnnamedAddr(false);
}
uint64_t TypeSize = TD.getTypeStoreSize(F->getType());
assert(TypeSize == TD.getPointerSize());
instrumentMemoryAllocation(F,
ConstantInt::get(LongType, TypeSize),
NULL,
Ret);
instrumentPointer(F, NULL, Ret);
}
}
/// StripSymbolNames - Strip symbol names.
static bool StripSymbolNames(Module &M, bool PreserveDbgInfo) {
SmallPtrSet<const GlobalValue*, 8> llvmUsedValues;
findUsedValues(M.getGlobalVariable("llvm.used"), llvmUsedValues);
findUsedValues(M.getGlobalVariable("llvm.compiler.used"), llvmUsedValues);
for (Module::global_iterator I = M.global_begin(), E = M.global_end();
I != E; ++I) {
if (I->hasLocalLinkage() && llvmUsedValues.count(I) == 0)
if (!PreserveDbgInfo || !I->getName().startswith("llvm.dbg"))
I->setName(""); // Internal symbols can't participate in linkage
}
for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I) {
if (I->hasLocalLinkage() && llvmUsedValues.count(I) == 0)
if (!PreserveDbgInfo || !I->getName().startswith("llvm.dbg"))
I->setName(""); // Internal symbols can't participate in linkage
StripSymtab(I->getValueSymbolTable(), PreserveDbgInfo);
}
// Remove all names from types.
StripTypeSymtab(M.getTypeSymbolTable(), PreserveDbgInfo);
return true;
}
void LTOCodeGenerator::applyScopeRestrictions()
{
if ( !_scopeRestrictionsDone ) {
Module* mergedModule = _linker.getModule();
// Start off with a verification pass.
PassManager passes;
passes.add(createVerifierPass());
// mark which symbols can not be internalized
if ( !_mustPreserveSymbols.empty() ) {
Mangler mangler(*mergedModule,
_target->getTargetAsmInfo()->getGlobalPrefix());
std::vector<const char*> mustPreserveList;
for (Module::iterator f = mergedModule->begin(),
e = mergedModule->end(); f != e; ++f) {
if ( !f->isDeclaration()
&& _mustPreserveSymbols.count(mangler.getValueName(f)) )
mustPreserveList.push_back(::strdup(f->getName().c_str()));
}
for (Module::global_iterator v = mergedModule->global_begin(),
e = mergedModule->global_end(); v != e; ++v) {
if ( !v->isDeclaration()
&& _mustPreserveSymbols.count(mangler.getValueName(v)) )
mustPreserveList.push_back(::strdup(v->getName().c_str()));
}
passes.add(createInternalizePass(mustPreserveList));
}
// apply scope restrictions
passes.run(*mergedModule);
_scopeRestrictionsDone = true;
}
}
void
AndroidBitcodeLinker::GetAllSymbols(Module *M,
std::set<std::string> &UndefinedSymbols,
std::set<std::string> &DefinedSymbols) {
UndefinedSymbols.clear();
DefinedSymbols.clear();
Function *Main = M->getFunction("main");
if (Main == 0 || Main->isDeclaration())
UndefinedSymbols.insert("main");
for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I)
if (I->hasName()) {
if (I->isDeclaration())
UndefinedSymbols.insert(I->getName());
else if (!I->hasLocalLinkage()) {
assert(!I->hasDLLImportStorageClass()
&& "Found dllimported non-external symbol!");
DefinedSymbols.insert(I->getName());
}
}
for (Module::global_iterator I = M->global_begin(), E = M->global_end();
I != E; ++I)
if (I->hasName()) {
if (I->isDeclaration())
UndefinedSymbols.insert(I->getName());
else if (!I->hasLocalLinkage()) {
assert(!I->hasDLLImportStorageClass()
&& "Found dllimported non-external symbol!");
DefinedSymbols.insert(I->getName());
}
}
for (Module::alias_iterator I = M->alias_begin(), E = M->alias_end();
I != E; ++I)
if (I->hasName())
DefinedSymbols.insert(I->getName());
for (std::set<std::string>::iterator I = UndefinedSymbols.begin();
I != UndefinedSymbols.end(); )
if (DefinedSymbols.count(*I))
UndefinedSymbols.erase(I++);
else
++I;
}
static void getSymbols(Module*M, std::vector<std::string>& symbols) {
// Loop over global variables
for (Module::global_iterator GI = M->global_begin(), GE=M->global_end(); GI != GE; ++GI)
if (!GI->isDeclaration() && !GI->hasLocalLinkage())
if (!GI->getName().empty())
symbols.push_back(GI->getName());
// Loop over functions
for (Module::iterator FI = M->begin(), FE = M->end(); FI != FE; ++FI)
if (!FI->isDeclaration() && !FI->hasLocalLinkage())
if (!FI->getName().empty())
symbols.push_back(FI->getName());
// Loop over aliases
for (Module::alias_iterator AI = M->alias_begin(), AE = M->alias_end();
AI != AE; ++AI) {
if (AI->hasName())
symbols.push_back(AI->getName());
}
}
bool ARMGlobalMerge::doInitialization(Module &M) {
SmallVector<GlobalVariable*, 16> Globals, ConstGlobals;
const TargetData *TD = TLI->getTargetData();
unsigned MaxOffset = TLI->getMaximalGlobalOffset();
bool Changed = false;
// Grab all non-const globals.
for (Module::global_iterator I = M.global_begin(),
E = M.global_end(); I != E; ++I) {
// Merge is safe for "normal" internal globals only
if (!I->hasLocalLinkage() || I->isThreadLocal() || I->hasSection())
continue;
// Ignore fancy-aligned globals for now.
if (I->getAlignment() != 0)
continue;
// Ignore all 'special' globals.
if (I->getName().startswith("llvm.") ||
I->getName().startswith(".llvm."))
continue;
if (TD->getTypeAllocSize(I->getType()) < MaxOffset) {
if (I->isConstant())
ConstGlobals.push_back(I);
else
Globals.push_back(I);
}
}
if (Globals.size() > 1)
Changed |= doMerge(Globals, M, false);
// FIXME: This currently breaks the EH processing due to way how the
// typeinfo detection works. We might want to detect the TIs and ignore
// them in the future.
// if (ConstGlobals.size() > 1)
// Changed |= doMerge(ConstGlobals, M, true);
return Changed;
}
/// DisambiguateGlobalSymbols - Mangle symbols to guarantee uniqueness by
/// modifying predominantly internal symbols rather than external ones.
///
static void DisambiguateGlobalSymbols(Module *M) {
// Try not to cause collisions by minimizing chances of renaming an
// already-external symbol, so take in external globals and functions as-is.
// The code should work correctly without disambiguation (assuming the same
// mangler is used by the two code generators), but having symbols with the
// same name causes warnings to be emitted by the code generator.
Mangler Mang(*M);
// Agree with the CBE on symbol naming
Mang.markCharUnacceptable('.');
for (Module::global_iterator I = M->global_begin(), E = M->global_end();
I != E; ++I) {
// Don't mangle asm names.
if (!I->hasName() || I->getName()[0] != 1)
I->setName(Mang.getMangledName(I));
}
for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I) {
// Don't mangle asm names or intrinsics.
if ((!I->hasName() || I->getName()[0] != 1) &&
I->getIntrinsicID() == 0)
I->setName(Mang.getMangledName(I));
}
}
bool InternalizePass::runOnModule(Module &M) {
CallGraph *CG = getAnalysisIfAvailable<CallGraph>();
CallGraphNode *ExternalNode = CG ? CG->getExternalCallingNode() : 0;
bool Changed = false;
// Never internalize functions which code-gen might insert.
// FIXME: We should probably add this (and the __stack_chk_guard) via some
// type of call-back in CodeGen.
ExternalNames.insert("__stack_chk_fail");
// Mark all functions not in the api as internal.
// FIXME: maybe use private linkage?
for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I)
if (!I->isDeclaration() && // Function must be defined here
// Available externally is really just a "declaration with a body".
!I->hasAvailableExternallyLinkage() &&
!I->hasLocalLinkage() && // Can't already have internal linkage
!ExternalNames.count(I->getName())) {// Not marked to keep external?
I->setLinkage(GlobalValue::InternalLinkage);
// Remove a callgraph edge from the external node to this function.
if (ExternalNode) ExternalNode->removeOneAbstractEdgeTo((*CG)[I]);
Changed = true;
++NumFunctions;
DEBUG(dbgs() << "Internalizing func " << I->getName() << "\n");
}
// Never internalize the llvm.used symbol. It is used to implement
// attribute((used)).
// FIXME: Shouldn't this just filter on llvm.metadata section??
ExternalNames.insert("llvm.used");
ExternalNames.insert("llvm.compiler.used");
// Never internalize anchors used by the machine module info, else the info
// won't find them. (see MachineModuleInfo.)
ExternalNames.insert("llvm.global_ctors");
ExternalNames.insert("llvm.global_dtors");
ExternalNames.insert("llvm.global.annotations");
// Never internalize symbols code-gen inserts.
ExternalNames.insert("__stack_chk_guard");
// Mark all global variables with initializers that are not in the api as
// internal as well.
// FIXME: maybe use private linkage?
for (Module::global_iterator I = M.global_begin(), E = M.global_end();
I != E; ++I)
if (!I->isDeclaration() && !I->hasLocalLinkage() &&
// Available externally is really just a "declaration with a body".
!I->hasAvailableExternallyLinkage() &&
!ExternalNames.count(I->getName())) {
I->setLinkage(GlobalValue::InternalLinkage);
Changed = true;
++NumGlobals;
DEBUG(dbgs() << "Internalized gvar " << I->getName() << "\n");
}
// Mark all aliases that are not in the api as internal as well.
for (Module::alias_iterator I = M.alias_begin(), E = M.alias_end();
I != E; ++I)
if (!I->isDeclaration() && !I->hasInternalLinkage() &&
// Available externally is really just a "declaration with a body".
!I->hasAvailableExternallyLinkage() &&
!ExternalNames.count(I->getName())) {
I->setLinkage(GlobalValue::InternalLinkage);
Changed = true;
++NumAliases;
DEBUG(dbgs() << "Internalized alias " << I->getName() << "\n");
}
return Changed;
}
/// SplitFunctionsOutOfModule - Given a module and a list of functions in the
/// module, split the functions OUT of the specified module, and place them in
/// the new module.
Module *
llvm::SplitFunctionsOutOfModule(Module *M,
const std::vector<Function*> &F,
ValueToValueMapTy &VMap) {
// Make sure functions & globals are all external so that linkage
// between the two modules will work.
for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I)
I->setLinkage(GlobalValue::ExternalLinkage);
for (Module::global_iterator I = M->global_begin(), E = M->global_end();
I != E; ++I) {
if (I->hasName() && I->getName()[0] == '\01')
I->setName(I->getName().substr(1));
I->setLinkage(GlobalValue::ExternalLinkage);
}
ValueToValueMapTy NewVMap;
Module *New = CloneModule(M, NewVMap);
// Remove the Test functions from the Safe module
std::set<Function *> TestFunctions;
for (unsigned i = 0, e = F.size(); i != e; ++i) {
Function *TNOF = cast<Function>(VMap[F[i]]);
DEBUG(errs() << "Removing function ");
DEBUG(WriteAsOperand(errs(), TNOF, false));
DEBUG(errs() << "\n");
TestFunctions.insert(cast<Function>(NewVMap[TNOF]));
DeleteFunctionBody(TNOF); // Function is now external in this module!
}
// Remove the Safe functions from the Test module
for (Module::iterator I = New->begin(), E = New->end(); I != E; ++I)
if (!TestFunctions.count(I))
DeleteFunctionBody(I);
// Try to split the global initializers evenly
for (Module::global_iterator I = M->global_begin(), E = M->global_end();
I != E; ++I) {
GlobalVariable *GV = cast<GlobalVariable>(NewVMap[I]);
if (Function *TestFn = globalInitUsesExternalBA(I)) {
if (Function *SafeFn = globalInitUsesExternalBA(GV)) {
errs() << "*** Error: when reducing functions, encountered "
"the global '";
WriteAsOperand(errs(), GV, false);
errs() << "' with an initializer that references blockaddresses "
"from safe function '" << SafeFn->getName()
<< "' and from test function '" << TestFn->getName() << "'.\n";
exit(1);
}
I->setInitializer(0); // Delete the initializer to make it external
} else {
// If we keep it in the safe module, then delete it in the test module
GV->setInitializer(0);
}
}
// Make sure that there is a global ctor/dtor array in both halves of the
// module if they both have static ctor/dtor functions.
SplitStaticCtorDtor("llvm.global_ctors", M, New, NewVMap);
SplitStaticCtorDtor("llvm.global_dtors", M, New, NewVMap);
return New;
}
int main(int argc, char **argv) {
// Print a stack trace if we signal out.
sys::PrintStackTraceOnErrorSignal();
PrettyStackTraceProgram X(argc, argv);
LLVMContext &Context = getGlobalContext();
llvm_shutdown_obj Y; // Call llvm_shutdown() on exit.
cl::ParseCommandLineOptions(argc, argv, "llvm extractor\n");
// Use lazy loading, since we only care about selected global values.
SMDiagnostic Err;
std::auto_ptr<Module> M;
M.reset(getLazyIRFileModule(InputFilename, Err, Context));
if (M.get() == 0) {
Err.print(argv[0], errs());
return 1;
}
// Use SetVector to avoid duplicates.
SetVector<GlobalValue *> GVs;
// Figure out which globals we should extract.
for (size_t i = 0, e = ExtractGlobals.size(); i != e; ++i) {
GlobalValue *GV = M.get()->getNamedGlobal(ExtractGlobals[i]);
if (!GV) {
errs() << argv[0] << ": program doesn't contain global named '"
<< ExtractGlobals[i] << "'!\n";
return 1;
}
GVs.insert(GV);
}
// Extract globals via regular expression matching.
for (size_t i = 0, e = ExtractRegExpGlobals.size(); i != e; ++i) {
std::string Error;
Regex RegEx(ExtractRegExpGlobals[i]);
if (!RegEx.isValid(Error)) {
errs() << argv[0] << ": '" << ExtractRegExpGlobals[i] << "' "
"invalid regex: " << Error;
}
bool match = false;
for (Module::global_iterator GV = M.get()->global_begin(),
E = M.get()->global_end(); GV != E; GV++) {
if (RegEx.match(GV->getName())) {
GVs.insert(&*GV);
match = true;
}
}
if (!match) {
errs() << argv[0] << ": program doesn't contain global named '"
<< ExtractRegExpGlobals[i] << "'!\n";
return 1;
}
}
// Figure out which functions we should extract.
for (size_t i = 0, e = ExtractFuncs.size(); i != e; ++i) {
GlobalValue *GV = M.get()->getFunction(ExtractFuncs[i]);
if (!GV) {
errs() << argv[0] << ": program doesn't contain function named '"
<< ExtractFuncs[i] << "'!\n";
return 1;
}
GVs.insert(GV);
}
// Extract functions via regular expression matching.
for (size_t i = 0, e = ExtractRegExpFuncs.size(); i != e; ++i) {
std::string Error;
StringRef RegExStr = ExtractRegExpFuncs[i];
Regex RegEx(RegExStr);
if (!RegEx.isValid(Error)) {
errs() << argv[0] << ": '" << ExtractRegExpFuncs[i] << "' "
"invalid regex: " << Error;
}
bool match = false;
for (Module::iterator F = M.get()->begin(), E = M.get()->end(); F != E;
F++) {
if (RegEx.match(F->getName())) {
GVs.insert(&*F);
match = true;
}
}
if (!match) {
errs() << argv[0] << ": program doesn't contain global named '"
<< ExtractRegExpFuncs[i] << "'!\n";
return 1;
}
}
// Materialize requisite global values.
if (!DeleteFn)
for (size_t i = 0, e = GVs.size(); i != e; ++i) {
GlobalValue *GV = GVs[i];
if (GV->isMaterializable()) {
std::string ErrInfo;
if (GV->Materialize(&ErrInfo)) {
errs() << argv[0] << ": error reading input: " << ErrInfo << "\n";
return 1;
}
}
}
else {
// Deleting. Materialize every GV that's *not* in GVs.
SmallPtrSet<GlobalValue *, 8> GVSet(GVs.begin(), GVs.end());
for (Module::global_iterator I = M->global_begin(), E = M->global_end();
I != E; ++I) {
GlobalVariable *G = I;
if (!GVSet.count(G) && G->isMaterializable()) {
std::string ErrInfo;
if (G->Materialize(&ErrInfo)) {
errs() << argv[0] << ": error reading input: " << ErrInfo << "\n";
return 1;
}
}
}
for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I) {
Function *F = I;
if (!GVSet.count(F) && F->isMaterializable()) {
std::string ErrInfo;
if (F->Materialize(&ErrInfo)) {
errs() << argv[0] << ": error reading input: " << ErrInfo << "\n";
return 1;
}
}
}
}
// In addition to deleting all other functions, we also want to spiff it
// up a little bit. Do this now.
PassManager Passes;
Passes.add(new TargetData(M.get())); // Use correct TargetData
std::vector<GlobalValue*> Gvs(GVs.begin(), GVs.end());
Passes.add(createGVExtractionPass(Gvs, DeleteFn));
if (!DeleteFn)
Passes.add(createGlobalDCEPass()); // Delete unreachable globals
Passes.add(createStripDeadDebugInfoPass()); // Remove dead debug info
Passes.add(createStripDeadPrototypesPass()); // Remove dead func decls
std::string ErrorInfo;
tool_output_file Out(OutputFilename.c_str(), ErrorInfo,
raw_fd_ostream::F_Binary);
if (!ErrorInfo.empty()) {
errs() << ErrorInfo << '\n';
return 1;
}
if (OutputAssembly)
Passes.add(createPrintModulePass(&Out.os()));
else if (Force || !CheckBitcodeOutputToConsole(Out.os(), true))
Passes.add(createBitcodeWriterPass(Out.os()));
Passes.run(*M.get());
// Declare success.
Out.keep();
return 0;
}
static PointerType *buildTlsTemplate(Module &M, std::vector<VarInfo> *TlsVars) {
std::vector<Type*> FieldBssTypes;
std::vector<Type*> FieldInitTypes;
std::vector<Constant*> FieldInitValues;
PassState State(&M);
for (Module::global_iterator GV = M.global_begin();
GV != M.global_end();
++GV) {
if (GV->isThreadLocal()) {
if (!GV->hasInitializer()) {
// Since this is a whole-program transformation, "extern" TLS
// variables are not allowed at this point.
report_fatal_error(std::string("TLS variable without an initializer: ")
+ GV->getName());
}
if (!GV->getInitializer()->isNullValue()) {
addVarToTlsTemplate(&State, &FieldInitTypes,
&FieldInitValues, GV);
VarInfo Info;
Info.TlsVar = GV;
Info.IsBss = false;
Info.TemplateIndex = FieldInitTypes.size() - 1;
TlsVars->push_back(Info);
}
}
}
// Handle zero-initialized TLS variables in a second pass, because
// these should follow non-zero-initialized TLS variables.
for (Module::global_iterator GV = M.global_begin();
GV != M.global_end();
++GV) {
if (GV->isThreadLocal() && GV->getInitializer()->isNullValue()) {
addVarToTlsTemplate(&State, &FieldBssTypes, NULL, GV);
VarInfo Info;
Info.TlsVar = GV;
Info.IsBss = true;
Info.TemplateIndex = FieldBssTypes.size() - 1;
TlsVars->push_back(Info);
}
}
// Add final alignment padding so that
// (struct tls_struct *) __nacl_read_tp() - 1
// gives the correct, aligned start of the TLS variables given the
// x86-style layout we are using. This requires some more bytes to
// be memset() to zero at runtime. This wastage doesn't seem
// important gives that we're not trying to optimize packing by
// reordering to put similarly-aligned variables together.
padToAlignment(&State, &FieldBssTypes, NULL, State.Alignment);
// We create the TLS template structs as "packed" because we insert
// alignment padding ourselves, and LLVM's implicit insertion of
// padding would interfere with ours. tls_bss_template can start at
// a non-aligned address immediately following the last field in
// tls_init_template.
StructType *InitTemplateType =
StructType::create(M.getContext(), "tls_init_template");
InitTemplateType->setBody(FieldInitTypes, /*isPacked=*/true);
StructType *BssTemplateType =
StructType::create(M.getContext(), "tls_bss_template");
BssTemplateType->setBody(FieldBssTypes, /*isPacked=*/true);
StructType *TemplateType = StructType::create(M.getContext(), "tls_struct");
SmallVector<Type*, 2> TemplateTopFields;
TemplateTopFields.push_back(InitTemplateType);
TemplateTopFields.push_back(BssTemplateType);
TemplateType->setBody(TemplateTopFields, /*isPacked=*/true);
PointerType *TemplatePtrType = PointerType::get(TemplateType, 0);
// We define the following symbols, which are the same as those
// defined by NaCl's original customized binutils linker scripts:
// __tls_template_start
// __tls_template_tdata_end
// __tls_template_end
// We also define __tls_template_alignment, which was not defined by
// the original linker scripts.
const char *StartSymbol = "__tls_template_start";
Constant *TemplateData = ConstantStruct::get(InitTemplateType,
FieldInitValues);
GlobalVariable *TemplateDataVar =
new GlobalVariable(M, InitTemplateType, /*isConstant=*/true,
GlobalValue::InternalLinkage, TemplateData);
setGlobalVariableValue(M, StartSymbol, TemplateDataVar);
TemplateDataVar->setName(StartSymbol);
Constant *TdataEnd = ConstantExpr::getGetElementPtr(
TemplateDataVar,
ConstantInt::get(M.getContext(), APInt(32, 1)));
setGlobalVariableValue(M, "__tls_template_tdata_end", TdataEnd);
Constant *TotalEnd = ConstantExpr::getGetElementPtr(
ConstantExpr::getBitCast(TemplateDataVar, TemplatePtrType),
ConstantInt::get(M.getContext(), APInt(32, 1)));
setGlobalVariableValue(M, "__tls_template_end", TotalEnd);
const char *AlignmentSymbol = "__tls_template_alignment";
Type *i32 = Type::getInt32Ty(M.getContext());
GlobalVariable *AlignmentVar = new GlobalVariable(
M, i32, /*isConstant=*/true,
GlobalValue::InternalLinkage,
ConstantInt::get(M.getContext(), APInt(32, State.Alignment)));
setGlobalVariableValue(M, AlignmentSymbol, AlignmentVar);
AlignmentVar->setName(AlignmentSymbol);
return TemplatePtrType;
}
bool GlobalMerge::doInitialization(Module &M) {
if (!EnableGlobalMerge)
return false;
auto &DL = M.getDataLayout();
DenseMap<unsigned, SmallVector<GlobalVariable*, 16> > Globals, ConstGlobals,
BSSGlobals;
bool Changed = false;
setMustKeepGlobalVariables(M);
// Grab all non-const globals.
for (Module::global_iterator I = M.global_begin(),
E = M.global_end(); I != E; ++I) {
// Merge is safe for "normal" internal or external globals only
if (I->isDeclaration() || I->isThreadLocal() || I->hasSection())
continue;
if (!(EnableGlobalMergeOnExternal && I->hasExternalLinkage()) &&
!I->hasInternalLinkage())
continue;
PointerType *PT = dyn_cast<PointerType>(I->getType());
assert(PT && "Global variable is not a pointer!");
unsigned AddressSpace = PT->getAddressSpace();
// Ignore fancy-aligned globals for now.
unsigned Alignment = DL.getPreferredAlignment(I);
Type *Ty = I->getType()->getElementType();
if (Alignment > DL.getABITypeAlignment(Ty))
continue;
// Ignore all 'special' globals.
if (I->getName().startswith("llvm.") ||
I->getName().startswith(".llvm."))
continue;
// Ignore all "required" globals:
if (isMustKeepGlobalVariable(I))
continue;
if (DL.getTypeAllocSize(Ty) < MaxOffset) {
if (TargetLoweringObjectFile::getKindForGlobal(I, *TM).isBSSLocal())
BSSGlobals[AddressSpace].push_back(I);
else if (I->isConstant())
ConstGlobals[AddressSpace].push_back(I);
else
Globals[AddressSpace].push_back(I);
}
}
for (DenseMap<unsigned, SmallVector<GlobalVariable*, 16> >::iterator
I = Globals.begin(), E = Globals.end(); I != E; ++I)
if (I->second.size() > 1)
Changed |= doMerge(I->second, M, false, I->first);
for (DenseMap<unsigned, SmallVector<GlobalVariable*, 16> >::iterator
I = BSSGlobals.begin(), E = BSSGlobals.end(); I != E; ++I)
if (I->second.size() > 1)
Changed |= doMerge(I->second, M, false, I->first);
if (EnableGlobalMergeOnConst)
for (DenseMap<unsigned, SmallVector<GlobalVariable*, 16> >::iterator
I = ConstGlobals.begin(), E = ConstGlobals.end(); I != E; ++I)
if (I->second.size() > 1)
Changed |= doMerge(I->second, M, true, I->first);
return Changed;
}
/// InputFilename is a LLVM bitcode file. Read it using bitcode reader.
/// Collect global functions and symbol names in symbols vector.
/// Collect external references in references vector.
/// Return LTO_READ_SUCCESS if there is no error.
enum LTOStatus
LTO::readLLVMObjectFile(const std::string &InputFilename,
NameToSymbolMap &symbols,
std::set<std::string> &references)
{
Module *m = getModule(InputFilename);
if (!m)
return LTO_READ_FAILURE;
// Collect Target info
getTarget(m);
if (!Target)
return LTO_READ_FAILURE;
// Use mangler to add GlobalPrefix to names to match linker names.
// FIXME : Instead of hard coding "-" use GlobalPrefix.
Mangler mangler(*m, Target->getTargetAsmInfo()->getGlobalPrefix());
modules.push_back(m);
for (Module::iterator f = m->begin(), e = m->end(); f != e; ++f) {
LTOLinkageTypes lt = getLTOLinkageType(f);
LTOVisibilityTypes vis = getLTOVisibilityType(f);
if (!f->isDeclaration() && lt != LTOInternalLinkage
&& strncmp (f->getName().c_str(), "llvm.", 5)) {
int alignment = ( 16 > f->getAlignment() ? 16 : f->getAlignment());
LLVMSymbol *newSymbol = new LLVMSymbol(lt, vis, f, f->getName(),
mangler.getValueName(f),
Log2_32(alignment));
symbols[newSymbol->getMangledName()] = newSymbol;
allSymbols[newSymbol->getMangledName()] = newSymbol;
}
// Collect external symbols referenced by this function.
for (Function::iterator b = f->begin(), fe = f->end(); b != fe; ++b)
for (BasicBlock::iterator i = b->begin(), be = b->end();
i != be; ++i) {
for (unsigned count = 0, total = i->getNumOperands();
count != total; ++count)
findExternalRefs(i->getOperand(count), references, mangler);
}
}
for (Module::global_iterator v = m->global_begin(), e = m->global_end();
v != e; ++v) {
LTOLinkageTypes lt = getLTOLinkageType(v);
LTOVisibilityTypes vis = getLTOVisibilityType(v);
if (!v->isDeclaration() && lt != LTOInternalLinkage
&& strncmp (v->getName().c_str(), "llvm.", 5)) {
const TargetData *TD = Target->getTargetData();
LLVMSymbol *newSymbol = new LLVMSymbol(lt, vis, v, v->getName(),
mangler.getValueName(v),
TD->getPreferredAlignmentLog(v));
symbols[newSymbol->getMangledName()] = newSymbol;
allSymbols[newSymbol->getMangledName()] = newSymbol;
for (unsigned count = 0, total = v->getNumOperands();
count != total; ++count)
findExternalRefs(v->getOperand(count), references, mangler);
}
}
return LTO_READ_SUCCESS;
}
/// Based on GetAllUndefinedSymbols() from LLVM3.2
///
/// GetAllUndefinedSymbols - calculates the set of undefined symbols that still
/// exist in an LLVM module. This is a bit tricky because there may be two
/// symbols with the same name but different LLVM types that will be resolved to
/// each other but aren't currently (thus we need to treat it as resolved).
///
/// Inputs:
/// M - The module in which to find undefined symbols.
///
/// Outputs:
/// UndefinedSymbols - A set of C++ strings containing the name of all
/// undefined symbols.
///
static void
GetAllUndefinedSymbols(Module *M, std::set<std::string> &UndefinedSymbols) {
static const std::string llvmIntrinsicPrefix="llvm.";
std::set<std::string> DefinedSymbols;
UndefinedSymbols.clear();
KLEE_DEBUG_WITH_TYPE("klee_linker",
dbgs() << "*** Computing undefined symbols ***\n");
for (Module::iterator I = M->begin(), E = M->end(); I != E; ++I)
if (I->hasName()) {
if (I->isDeclaration())
UndefinedSymbols.insert(I->getName());
else if (!I->hasLocalLinkage()) {
#if LLVM_VERSION_CODE < LLVM_VERSION(3, 5)
assert(!I->hasDLLImportLinkage() && "Found dllimported non-external symbol!");
#else
assert(!I->hasDLLImportStorageClass() && "Found dllimported non-external symbol!");
#endif
DefinedSymbols.insert(I->getName());
}
}
for (Module::global_iterator I = M->global_begin(), E = M->global_end();
I != E; ++I)
if (I->hasName()) {
if (I->isDeclaration())
UndefinedSymbols.insert(I->getName());
else if (!I->hasLocalLinkage()) {
#if LLVM_VERSION_CODE < LLVM_VERSION(3, 5)
assert(!I->hasDLLImportLinkage() && "Found dllimported non-external symbol!");
#else
assert(!I->hasDLLImportStorageClass() && "Found dllimported non-external symbol!");
#endif
DefinedSymbols.insert(I->getName());
}
}
for (Module::alias_iterator I = M->alias_begin(), E = M->alias_end();
I != E; ++I)
if (I->hasName())
DefinedSymbols.insert(I->getName());
// Prune out any defined symbols from the undefined symbols set
// and other symbols we don't want to treat as an undefined symbol
std::vector<std::string> SymbolsToRemove;
for (std::set<std::string>::iterator I = UndefinedSymbols.begin();
I != UndefinedSymbols.end(); ++I )
{
if (DefinedSymbols.count(*I))
{
SymbolsToRemove.push_back(*I);
continue;
}
// Strip out llvm intrinsics
if ( (I->size() >= llvmIntrinsicPrefix.size() ) &&
(I->compare(0, llvmIntrinsicPrefix.size(), llvmIntrinsicPrefix) == 0) )
{
KLEE_DEBUG_WITH_TYPE("klee_linker", dbgs() << "LLVM intrinsic " << *I <<
" has will be removed from undefined symbols"<< "\n");
SymbolsToRemove.push_back(*I);
continue;
}
// Symbol really is undefined
KLEE_DEBUG_WITH_TYPE("klee_linker",
dbgs() << "Symbol " << *I << " is undefined.\n");
}
// Remove KLEE intrinsics from set of undefined symbols
for (SpecialFunctionHandler::const_iterator sf = SpecialFunctionHandler::begin(),
se = SpecialFunctionHandler::end(); sf != se; ++sf)
{
if (UndefinedSymbols.find(sf->name) == UndefinedSymbols.end())
continue;
SymbolsToRemove.push_back(sf->name);
KLEE_DEBUG_WITH_TYPE("klee_linker",
dbgs() << "KLEE intrinsic " << sf->name <<
" has will be removed from undefined symbols"<< "\n");
}
// Now remove the symbols from undefined set.
for (size_t i = 0, j = SymbolsToRemove.size(); i < j; ++i )
UndefinedSymbols.erase(SymbolsToRemove[i]);
KLEE_DEBUG_WITH_TYPE("klee_linker",
dbgs() << "*** Finished computing undefined symbols ***\n");
}
bool InternalizePass::runOnModule(Module &M) {
CallGraph *CG = getAnalysisIfAvailable<CallGraph>();
CallGraphNode *ExternalNode = CG ? CG->getExternalCallingNode() : 0;
if (ExternalNames.empty()) {
// Return if we're not in 'all but main' mode and have no external api
if (!AllButMain)
return false;
// If no list or file of symbols was specified, check to see if there is a
// "main" symbol defined in the module. If so, use it, otherwise do not
// internalize the module, it must be a library or something.
//
Function *MainFunc = M.getFunction("main");
if (MainFunc == 0 || MainFunc->isDeclaration())
return false; // No main found, must be a library...
// Preserve main, internalize all else.
ExternalNames.insert(MainFunc->getName());
}
bool Changed = false;
// Never internalize functions which code-gen might insert.
ExternalNames.insert("__stack_chk_fail");
// Mark all functions not in the api as internal.
// FIXME: maybe use private linkage?
for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I)
if (!I->isDeclaration() && // Function must be defined here
// Available externally is really just a "declaration with a body".
!I->hasAvailableExternallyLinkage() &&
!I->hasLocalLinkage() && // Can't already have internal linkage
!ExternalNames.count(I->getName())) {// Not marked to keep external?
I->setLinkage(GlobalValue::InternalLinkage);
// Remove a callgraph edge from the external node to this function.
if (ExternalNode) ExternalNode->removeOneAbstractEdgeTo((*CG)[I]);
Changed = true;
++NumFunctions;
DEBUG(dbgs() << "Internalizing func " << I->getName() << "\n");
}
// Never internalize the llvm.used symbol. It is used to implement
// attribute((used)).
// FIXME: Shouldn't this just filter on llvm.metadata section??
ExternalNames.insert("llvm.used");
ExternalNames.insert("llvm.compiler.used");
// Never internalize anchors used by the machine module info, else the info
// won't find them. (see MachineModuleInfo.)
ExternalNames.insert("llvm.global_ctors");
ExternalNames.insert("llvm.global_dtors");
ExternalNames.insert("llvm.global.annotations");
// Never internalize symbols code-gen inserts.
ExternalNames.insert("__stack_chk_guard");
// Mark all global variables with initializers that are not in the api as
// internal as well.
// FIXME: maybe use private linkage?
for (Module::global_iterator I = M.global_begin(), E = M.global_end();
I != E; ++I)
if (!I->isDeclaration() && !I->hasLocalLinkage() &&
// Available externally is really just a "declaration with a body".
!I->hasAvailableExternallyLinkage() &&
!ExternalNames.count(I->getName())) {
I->setLinkage(GlobalValue::InternalLinkage);
Changed = true;
++NumGlobals;
DEBUG(dbgs() << "Internalized gvar " << I->getName() << "\n");
}
// Mark all aliases that are not in the api as internal as well.
for (Module::alias_iterator I = M.alias_begin(), E = M.alias_end();
I != E; ++I)
if (!I->isDeclaration() && !I->hasInternalLinkage() &&
// Available externally is really just a "declaration with a body".
!I->hasAvailableExternallyLinkage() &&
!ExternalNames.count(I->getName())) {
I->setLinkage(GlobalValue::InternalLinkage);
Changed = true;
++NumAliases;
DEBUG(dbgs() << "Internalized alias " << I->getName() << "\n");
}
return Changed;
}
bool InternalizePass::runOnModule(Module &M) {
CallGraphWrapperPass *CGPass = getAnalysisIfAvailable<CallGraphWrapperPass>();
CallGraph *CG = CGPass ? &CGPass->getCallGraph() : 0;
CallGraphNode *ExternalNode = CG ? CG->getExternalCallingNode() : 0;
bool Changed = false;
SmallPtrSet<GlobalValue *, 8> Used;
collectUsedGlobalVariables(M, Used, false);
// We must assume that globals in llvm.used have a reference that not even
// the linker can see, so we don't internalize them.
// For llvm.compiler.used the situation is a bit fuzzy. The assembler and
// linker can drop those symbols. If this pass is running as part of LTO,
// one might think that it could just drop llvm.compiler.used. The problem
// is that even in LTO llvm doesn't see every reference. For example,
// we don't see references from function local inline assembly. To be
// conservative, we internalize symbols in llvm.compiler.used, but we
// keep llvm.compiler.used so that the symbol is not deleted by llvm.
for (SmallPtrSet<GlobalValue *, 8>::iterator I = Used.begin(), E = Used.end();
I != E; ++I) {
GlobalValue *V = *I;
ExternalNames.insert(V->getName());
}
// Mark all functions not in the api as internal.
for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I) {
if (!shouldInternalize(*I, ExternalNames, OnlyHidden))
continue;
I->setLinkage(GlobalValue::InternalLinkage);
if (ExternalNode)
// Remove a callgraph edge from the external node to this function.
ExternalNode->removeOneAbstractEdgeTo((*CG)[I]);
Changed = true;
++NumFunctions;
DEBUG(dbgs() << "Internalizing func " << I->getName() << "\n");
}
// Never internalize the llvm.used symbol. It is used to implement
// attribute((used)).
// FIXME: Shouldn't this just filter on llvm.metadata section??
ExternalNames.insert("llvm.used");
ExternalNames.insert("llvm.compiler.used");
// Never internalize anchors used by the machine module info, else the info
// won't find them. (see MachineModuleInfo.)
ExternalNames.insert("llvm.global_ctors");
ExternalNames.insert("llvm.global_dtors");
ExternalNames.insert("llvm.global.annotations");
// Never internalize symbols code-gen inserts.
// FIXME: We should probably add this (and the __stack_chk_guard) via some
// type of call-back in CodeGen.
ExternalNames.insert("__stack_chk_fail");
ExternalNames.insert("__stack_chk_guard");
// Mark all global variables with initializers that are not in the api as
// internal as well.
for (Module::global_iterator I = M.global_begin(), E = M.global_end();
I != E; ++I) {
if (!shouldInternalize(*I, ExternalNames, OnlyHidden))
continue;
I->setLinkage(GlobalValue::InternalLinkage);
Changed = true;
++NumGlobals;
DEBUG(dbgs() << "Internalized gvar " << I->getName() << "\n");
}
// Mark all aliases that are not in the api as internal as well.
for (Module::alias_iterator I = M.alias_begin(), E = M.alias_end();
I != E; ++I) {
if (!shouldInternalize(*I, ExternalNames, OnlyHidden))
continue;
I->setLinkage(GlobalValue::InternalLinkage);
Changed = true;
++NumAliases;
DEBUG(dbgs() << "Internalized alias " << I->getName() << "\n");
}
return Changed;
}