#include "llvm/Support/MathExtras.h"
#include "llvm/Support/MemoryBuffer.h"
#include "llvm/Support/raw_ostream.h"
+#include "llvm/Support/ManagedStatic.h"
+
using namespace llvm;
enum {
SWITCH_INST_MAGIC = 0x4B5 // May 2012 => 1205 => Hex
};
-void BitcodeReader::materializeForwardReferencedFunctions() {
- while (!BlockAddrFwdRefs.empty()) {
- Function *F = BlockAddrFwdRefs.begin()->first;
- F->Materialize();
+std::error_code BitcodeReader::materializeForwardReferencedFunctions() {
+ if (WillMaterializeAllForwardRefs)
+ return std::error_code();
+
+ // Prevent recursion.
+ WillMaterializeAllForwardRefs = true;
+
+ while (!BasicBlockFwdRefQueue.empty()) {
+ Function *F = BasicBlockFwdRefQueue.front();
+ BasicBlockFwdRefQueue.pop_front();
+ assert(F && "Expected valid function");
+ if (!BasicBlockFwdRefs.count(F))
+ // Already materialized.
+ continue;
+
+ // Check for a function that isn't materializable to prevent an infinite
+ // loop. When parsing a blockaddress stored in a global variable, there
+ // isn't a trivial way to check if a function will have a body without a
+ // linear search through FunctionsWithBodies, so just check it here.
+ if (!F->isMaterializable())
+ return Error(BitcodeError::NeverResolvedFunctionFromBlockAddress);
+
+ // Try to materialize F.
+ if (std::error_code EC = Materialize(F))
+ return EC;
}
+ assert(BasicBlockFwdRefs.empty() && "Function missing from queue");
+
+ // Reset state.
+ WillMaterializeAllForwardRefs = false;
+ return std::error_code();
}
void BitcodeReader::FreeState() {
DeferredFunctionInfo.clear();
MDKindMap.clear();
- assert(BlockAddrFwdRefs.empty() && "Unresolved blockaddress fwd references");
+ assert(BasicBlockFwdRefs.empty() && "Unresolved blockaddress fwd references");
+ BasicBlockFwdRefQueue.clear();
}
//===----------------------------------------------------------------------===//
if (!Fn)
return Error(BitcodeError::InvalidRecord);
+ // Don't let Fn get dematerialized.
+ BlockAddressesTaken.insert(Fn);
+
// If the function is already parsed we can insert the block address right
// away.
+ BasicBlock *BB;
+ unsigned BBID = Record[2];
+ if (!BBID)
+ // Invalid reference to entry block.
+ return Error(BitcodeError::InvalidID);
if (!Fn->empty()) {
Function::iterator BBI = Fn->begin(), BBE = Fn->end();
- for (size_t I = 0, E = Record[2]; I != E; ++I) {
+ for (size_t I = 0, E = BBID; I != E; ++I) {
if (BBI == BBE)
return Error(BitcodeError::InvalidID);
++BBI;
}
- V = BlockAddress::get(Fn, BBI);
+ BB = BBI;
} else {
// Otherwise insert a placeholder and remember it so it can be inserted
// when the function is parsed.
- GlobalVariable *FwdRef = new GlobalVariable(*Fn->getParent(),
- Type::getInt8Ty(Context),
- false, GlobalValue::InternalLinkage,
- nullptr, "");
- BlockAddrFwdRefs[Fn].push_back(std::make_pair(Record[2], FwdRef));
- V = FwdRef;
+ auto &FwdBBs = BasicBlockFwdRefs[Fn];
+ if (FwdBBs.empty())
+ BasicBlockFwdRefQueue.push_back(Fn);
+ if (FwdBBs.size() < BBID + 1)
+ FwdBBs.resize(BBID + 1);
+ if (!FwdBBs[BBID])
+ FwdBBs[BBID] = BasicBlock::Create(Context);
+ BB = FwdBBs[BBID];
}
+ V = BlockAddress::get(Fn, BB);
break;
}
}
unsigned NumUses = 0;
SmallDenseMap<const Use *, unsigned, 16> Order;
for (const Use &U : V->uses()) {
- if (NumUses > Record.size())
+ if (++NumUses > Record.size())
break;
- Order[&U] = Record[NumUses++];
+ Order[&U] = Record[NumUses - 1];
}
if (Order.size() != Record.size() || NumUses > Record.size())
// Mismatches can happen if the functions are being materialized lazily
// creating now, so that we can match up the body with them later.
if (!isProto) {
FunctionsWithBodies.push_back(Func);
- if (LazyStreamer) DeferredFunctionInfo[Func] = 0;
+ if (LazyStreamer)
+ DeferredFunctionInfo[Func] = 0;
}
break;
}
switch (BitCode) {
default: // Default behavior: reject
return Error(BitcodeError::InvalidValue);
- case bitc::FUNC_CODE_DECLAREBLOCKS: // DECLAREBLOCKS: [nblocks]
+ case bitc::FUNC_CODE_DECLAREBLOCKS: { // DECLAREBLOCKS: [nblocks]
if (Record.size() < 1 || Record[0] == 0)
return Error(BitcodeError::InvalidRecord);
// Create all the basic blocks for the function.
FunctionBBs.resize(Record[0]);
- for (unsigned i = 0, e = FunctionBBs.size(); i != e; ++i)
- FunctionBBs[i] = BasicBlock::Create(Context, "", F);
+
+ // See if anything took the address of blocks in this function.
+ auto BBFRI = BasicBlockFwdRefs.find(F);
+ if (BBFRI == BasicBlockFwdRefs.end()) {
+ for (unsigned i = 0, e = FunctionBBs.size(); i != e; ++i)
+ FunctionBBs[i] = BasicBlock::Create(Context, "", F);
+ } else {
+ auto &BBRefs = BBFRI->second;
+ // Check for invalid basic block references.
+ if (BBRefs.size() > FunctionBBs.size())
+ return Error(BitcodeError::InvalidID);
+ assert(!BBRefs.empty() && "Unexpected empty array");
+ assert(!BBRefs.front() && "Invalid reference to entry block");
+ for (unsigned I = 0, E = FunctionBBs.size(), RE = BBRefs.size(); I != E;
+ ++I)
+ if (I < RE && BBRefs[I]) {
+ BBRefs[I]->insertInto(F);
+ FunctionBBs[I] = BBRefs[I];
+ } else {
+ FunctionBBs[I] = BasicBlock::Create(Context, "", F);
+ }
+
+ // Erase from the table.
+ BasicBlockFwdRefs.erase(BBFRI);
+ }
+
CurBB = FunctionBBs[0];
continue;
+ }
case bitc::FUNC_CODE_DEBUG_LOC_AGAIN: // DEBUG_LOC_AGAIN
// This record indicates that the last instruction is at the same
// FIXME: Check for unresolved forward-declared metadata references
// and clean up leaks.
- // See if anything took the address of blocks in this function. If so,
- // resolve them now.
- DenseMap<Function*, std::vector<BlockAddrRefTy> >::iterator BAFRI =
- BlockAddrFwdRefs.find(F);
- if (BAFRI != BlockAddrFwdRefs.end()) {
- std::vector<BlockAddrRefTy> &RefList = BAFRI->second;
- for (unsigned i = 0, e = RefList.size(); i != e; ++i) {
- unsigned BlockIdx = RefList[i].first;
- if (BlockIdx >= FunctionBBs.size())
- return Error(BitcodeError::InvalidID);
-
- GlobalVariable *FwdRef = RefList[i].second;
- FwdRef->replaceAllUsesWith(BlockAddress::get(F, FunctionBBs[BlockIdx]));
- FwdRef->eraseFromParent();
- }
-
- BlockAddrFwdRefs.erase(BAFRI);
- }
-
// Trim the value list down to the size it was before we parsed this function.
ValueList.shrinkTo(ModuleValueListSize);
MDValueList.shrinkTo(ModuleMDValueListSize);
}
}
- return std::error_code();
+ // Bring in any functions that this function forward-referenced via
+ // blockaddresses.
+ return materializeForwardReferencedFunctions();
}
bool BitcodeReader::isDematerializable(const GlobalValue *GV) const {
const Function *F = dyn_cast<Function>(GV);
if (!F || F->isDeclaration())
return false;
+
+ // Dematerializing F would leave dangling references that wouldn't be
+ // reconnected on re-materialization.
+ if (BlockAddressesTaken.count(F))
+ return false;
+
return DeferredFunctionInfo.count(const_cast<Function*>(F));
}
assert(DeferredFunctionInfo.count(F) && "No info to read function later?");
// Just forget the function body, we can remat it later.
- F->deleteBody();
+ F->dropAllReferences();
}
std::error_code BitcodeReader::MaterializeModule(Module *M) {
assert(M == TheModule &&
"Can only Materialize the Module this BitcodeReader is attached to.");
+
+ // Promise to materialize all forward references.
+ WillMaterializeAllForwardRefs = true;
+
// Iterate over the module, deserializing any functions that are still on
// disk.
for (Module::iterator F = TheModule->begin(), E = TheModule->end();
if (NextUnreadBit)
ParseModule(true);
+ // Check that all block address forward references got resolved (as we
+ // promised above).
+ if (!BasicBlockFwdRefs.empty())
+ return Error(BitcodeError::NeverResolvedFunctionFromBlockAddress);
+
// Upgrade any intrinsic calls that slipped through (should not happen!) and
// delete the old functions to clean up. We can't do this unless the entire
// module is materialized because there could always be another function body
const unsigned char *BufPtr = (const unsigned char*)Buffer->getBufferStart();
const unsigned char *BufEnd = BufPtr+Buffer->getBufferSize();
- if (Buffer->getBufferSize() & 3) {
- if (!isRawBitcode(BufPtr, BufEnd) && !isBitcodeWrapper(BufPtr, BufEnd))
- return Error(BitcodeError::InvalidBitcodeSignature);
- else
- return Error(BitcodeError::BitcodeStreamInvalidSize);
- }
+ if (Buffer->getBufferSize() & 3)
+ return Error(BitcodeError::InvalidBitcodeSignature);
// If we have a wrapper header, parse it and ignore the non-bc file contents.
// The magic number is 0x0B17C0DE stored in little endian.
unsigned char buf[16];
if (Bytes->readBytes(0, 16, buf) == -1)
- return Error(BitcodeError::BitcodeStreamInvalidSize);
+ return Error(BitcodeError::InvalidBitcodeSignature);
if (!isBitcode(buf, buf + 16))
return Error(BitcodeError::InvalidBitcodeSignature);
std::string message(int IE) const override {
BitcodeError E = static_cast<BitcodeError>(IE);
switch (E) {
- case BitcodeError::BitcodeStreamInvalidSize:
- return "Bitcode stream length should be >= 16 bytes and a multiple of 4";
case BitcodeError::ConflictingMETADATA_KINDRecords:
return "Conflicting METADATA_KIND records";
case BitcodeError::CouldNotFindFunctionInStream:
return "Invalid multiple blocks";
case BitcodeError::NeverResolvedValueFoundInFunction:
return "Never resolved value found in function";
+ case BitcodeError::NeverResolvedFunctionFromBlockAddress:
+ return "Never resolved function from blockaddress";
case BitcodeError::InvalidValue:
return "Invalid value";
}
};
}
+static ManagedStatic<BitcodeErrorCategoryType> ErrorCategory;
+
const std::error_category &llvm::BitcodeErrorCategory() {
- static BitcodeErrorCategoryType O;
- return O;
+ return *ErrorCategory;
}
//===----------------------------------------------------------------------===//
// External interface
//===----------------------------------------------------------------------===//
-/// getLazyBitcodeModule - lazy function-at-a-time loading from a file.
+/// \brief Get a lazy one-at-time loading module from bitcode.
+///
+/// This isn't always used in a lazy context. In particular, it's also used by
+/// \a parseBitcodeFile(). If this is truly lazy, then we need to eagerly pull
+/// in forward-referenced functions from block address references.
///
-ErrorOr<Module *> llvm::getLazyBitcodeModule(MemoryBuffer *Buffer,
- LLVMContext &Context) {
+/// \param[in] WillMaterializeAll Set to \c true if the caller promises to
+/// materialize everything -- in particular, if this isn't truly lazy.
+static ErrorOr<Module *>
+getLazyBitcodeModuleImpl(std::unique_ptr<MemoryBuffer> &&Buffer,
+ LLVMContext &Context, bool WillMaterializeAll) {
Module *M = new Module(Buffer->getBufferIdentifier(), Context);
- BitcodeReader *R = new BitcodeReader(Buffer, Context);
+ BitcodeReader *R = new BitcodeReader(Buffer.get(), Context);
M->setMaterializer(R);
- if (std::error_code EC = R->ParseBitcodeInto(M)) {
+
+ auto cleanupOnError = [&](std::error_code EC) {
R->releaseBuffer(); // Never take ownership on error.
delete M; // Also deletes R.
return EC;
- }
+ };
- R->materializeForwardReferencedFunctions();
+ if (std::error_code EC = R->ParseBitcodeInto(M))
+ return cleanupOnError(EC);
+ if (!WillMaterializeAll)
+ // Resolve forward references from blockaddresses.
+ if (std::error_code EC = R->materializeForwardReferencedFunctions())
+ return cleanupOnError(EC);
+
+ Buffer.release(); // The BitcodeReader owns it now.
return M;
}
+ErrorOr<Module *>
+llvm::getLazyBitcodeModule(std::unique_ptr<MemoryBuffer> &&Buffer,
+ LLVMContext &Context) {
+ return getLazyBitcodeModuleImpl(std::move(Buffer), Context, false);
+}
Module *llvm::getStreamedBitcodeModule(const std::string &name,
DataStreamer *streamer,
return M;
}
-ErrorOr<Module *> llvm::parseBitcodeFile(MemoryBuffer *Buffer,
+ErrorOr<Module *> llvm::parseBitcodeFile(MemoryBufferRef Buffer,
LLVMContext &Context) {
- ErrorOr<Module *> ModuleOrErr = getLazyBitcodeModule(Buffer, Context);
+ std::unique_ptr<MemoryBuffer> Buf = MemoryBuffer::getMemBuffer(Buffer, false);
+ ErrorOr<Module *> ModuleOrErr =
+ getLazyBitcodeModuleImpl(std::move(Buf), Context, true);
if (!ModuleOrErr)
return ModuleOrErr;
Module *M = ModuleOrErr.get();
// Read in the entire module, and destroy the BitcodeReader.
- if (std::error_code EC = M->materializeAllPermanently(true)) {
+ if (std::error_code EC = M->materializeAllPermanently()) {
delete M;
return EC;
}
return M;
}
-std::string llvm::getBitcodeTargetTriple(MemoryBuffer *Buffer,
+std::string llvm::getBitcodeTargetTriple(MemoryBufferRef Buffer,
LLVMContext &Context) {
- BitcodeReader *R = new BitcodeReader(Buffer, Context);
+ std::unique_ptr<MemoryBuffer> Buf = MemoryBuffer::getMemBuffer(Buffer, false);
+ auto R = llvm::make_unique<BitcodeReader>(Buf.release(), Context);
ErrorOr<std::string> Triple = R->parseTriple();
- R->releaseBuffer();
- delete R;
if (Triple.getError())
return "";
return Triple.get();