1 //===- Reader.cpp - Code to read bytecode files ---------------------------===//
3 // This library implements the functionality defined in llvm/Bytecode/Reader.h
5 // Note that this library should be as fast as possible, reentrant, and
8 // TODO: Make error message outputs be configurable depending on an option?
9 // TODO: Allow passing in an option to ignore the symbol table
11 //===----------------------------------------------------------------------===//
13 #include "ReaderInternals.h"
14 #include "llvm/Bytecode/Reader.h"
15 #include "llvm/Bytecode/Format.h"
16 #include "llvm/GlobalVariable.h"
17 #include "llvm/Module.h"
18 #include "llvm/BasicBlock.h"
19 #include "llvm/ConstantVals.h"
20 #include "llvm/iPHINode.h"
21 #include "llvm/iOther.h"
22 #include <sys/types.h>
29 bool BytecodeParser::getTypeSlot(const Type *Ty, unsigned &Slot) {
30 if (Ty->isPrimitiveType()) {
31 Slot = Ty->getPrimitiveID();
33 // Check the method level types first...
34 TypeValuesListTy::iterator I = find(MethodTypeValues.begin(),
35 MethodTypeValues.end(), Ty);
36 if (I != MethodTypeValues.end()) {
37 Slot = FirstDerivedTyID+ModuleTypeValues.size()+
38 (&*I - &MethodTypeValues[0]);
40 I = find(ModuleTypeValues.begin(), ModuleTypeValues.end(), Ty);
41 if (I == ModuleTypeValues.end()) return true; // Didn't find type!
42 Slot = FirstDerivedTyID + (&*I - &ModuleTypeValues[0]);
45 //cerr << "getTypeSlot '" << Ty->getName() << "' = " << Slot << endl;
49 const Type *BytecodeParser::getType(unsigned ID) {
50 const Type *T = Type::getPrimitiveType((Type::PrimitiveID)ID);
53 //cerr << "Looking up Type ID: " << ID << endl;
55 const Value *D = getValue(Type::TypeTy, ID, false);
56 if (D == 0) return failure<const Type*>(0);
61 int BytecodeParser::insertValue(Value *Val, vector<ValueList> &ValueTab) {
63 if (getTypeSlot(Val->getType(), type)) return failure<int>(-1);
64 assert(type != Type::TypeTyID && "Types should never be insertValue'd!");
66 if (ValueTab.size() <= type)
67 ValueTab.resize(type+1, ValueList());
69 //cerr << "insertValue Values[" << type << "][" << ValueTab[type].size()
70 // << "] = " << Val << endl;
71 ValueTab[type].push_back(Val);
73 return ValueTab[type].size()-1;
76 Value *BytecodeParser::getValue(const Type *Ty, unsigned oNum, bool Create) {
78 unsigned type; // The type plane it lives in...
80 if (getTypeSlot(Ty, type)) return failure<Value*>(0); // TODO: true
82 if (type == Type::TypeTyID) { // The 'type' plane has implicit values
83 assert(Create == false);
84 const Type *T = Type::getPrimitiveType((Type::PrimitiveID)Num);
85 if (T) return (Value*)T; // Asked for a primitive type...
87 // Otherwise, derived types need offset...
88 Num -= FirstDerivedTyID;
90 // Is it a module level type?
91 if (Num < ModuleTypeValues.size())
92 return (Value*)ModuleTypeValues[Num].get();
94 // Nope, is it a method level type?
95 Num -= ModuleTypeValues.size();
96 if (Num < MethodTypeValues.size())
97 return (Value*)MethodTypeValues[Num].get();
102 if (type < ModuleValues.size()) {
103 if (Num < ModuleValues[type].size())
104 return ModuleValues[type][Num];
105 Num -= ModuleValues[type].size();
108 if (Values.size() > type && Values[type].size() > Num)
109 return Values[type][Num];
111 if (!Create) return failure<Value*>(0); // Do not create a placeholder?
114 switch (Ty->getPrimitiveID()) {
115 case Type::LabelTyID: d = new BBPHolder(Ty, oNum); break;
116 case Type::MethodTyID:
117 cerr << "Creating method pholder! : " << type << ":" << oNum << " "
118 << Ty->getName() << endl;
119 d = new MethPHolder(Ty, oNum);
120 if (insertValue(d, LateResolveModuleValues) ==-1) return failure<Value*>(0);
122 default: d = new DefPHolder(Ty, oNum); break;
125 assert(d != 0 && "How did we not make something?");
126 if (insertValue(d, LateResolveValues) == -1) return failure<Value*>(0);
130 bool BytecodeParser::postResolveValues(ValueTable &ValTab) {
132 for (unsigned ty = 0; ty < ValTab.size(); ++ty) {
133 ValueList &DL = ValTab[ty];
135 while ((Size = DL.size())) {
136 unsigned IDNumber = getValueIDNumberFromPlaceHolder(DL[Size-1]);
138 Value *D = DL[Size-1];
141 Value *NewDef = getValue(D->getType(), IDNumber, false);
143 Error = true; // Unresolved thinger
144 cerr << "Unresolvable reference found: <"
145 << D->getType()->getDescription() << ">:" << IDNumber << "!\n";
147 // Fixup all of the uses of this placeholder def...
148 D->replaceAllUsesWith(NewDef);
150 // Now that all the uses are gone, delete the placeholder...
151 // If we couldn't find a def (error case), then leak a little
152 delete D; // memory, 'cause otherwise we can't remove all uses!
160 bool BytecodeParser::ParseBasicBlock(const uchar *&Buf, const uchar *EndBuf,
162 BB = new BasicBlock();
164 while (Buf < EndBuf) {
166 if (ParseInstruction(Buf, EndBuf, Inst)) {
168 return failure(true);
171 if (Inst == 0) { delete BB; return failure(true); }
172 if (insertValue(Inst, Values) == -1) { delete BB; return failure(true); }
174 BB->getInstList().push_back(Inst);
182 bool BytecodeParser::ParseSymbolTable(const uchar *&Buf, const uchar *EndBuf,
184 while (Buf < EndBuf) {
185 // Symtab block header: [num entries][type id number]
186 unsigned NumEntries, Typ;
187 if (read_vbr(Buf, EndBuf, NumEntries) ||
188 read_vbr(Buf, EndBuf, Typ)) return failure(true);
189 const Type *Ty = getType(Typ);
190 if (Ty == 0) return failure(true);
192 BCR_TRACE(3, "Plane Type: '" << Ty << "' with " << NumEntries <<
195 for (unsigned i = 0; i < NumEntries; ++i) {
196 // Symtab entry: [def slot #][name]
198 if (read_vbr(Buf, EndBuf, slot)) return failure(true);
200 if (read(Buf, EndBuf, Name, false)) // Not aligned...
201 return failure(true);
203 Value *D = getValue(Ty, slot, false); // Find mapping...
205 BCR_TRACE(3, "FAILED LOOKUP: Slot #" << slot << endl);
206 return failure(true);
208 BCR_TRACE(4, "Map: '" << Name << "' to #" << slot << ":" << D;
209 if (!isa<Instruction>(D)) cerr << endl);
211 D->setName(Name, ST);
215 if (Buf > EndBuf) return failure(true);
219 // DeclareNewGlobalValue - Patch up forward references to global values in the
220 // form of ConstantPointerRef.
222 void BytecodeParser::DeclareNewGlobalValue(GlobalValue *GV, unsigned Slot) {
223 // Check to see if there is a forward reference to this global variable...
224 // if there is, eliminate it and patch the reference to use the new def'n.
225 GlobalRefsType::iterator I = GlobalRefs.find(make_pair(GV->getType(), Slot));
227 if (I != GlobalRefs.end()) {
228 GlobalVariable *OldGV = I->second; // Get the placeholder...
229 BCR_TRACE(3, "Mutating CPPR Forward Ref!\n");
231 // Loop over all of the uses of the GlobalValue. The only thing they are
232 // allowed to be at this point is ConstantPointerRef's.
233 assert(OldGV->use_size() == 1 && "Only one reference should exist!");
234 while (!OldGV->use_empty()) {
235 User *U = OldGV->use_back(); // Must be a ConstantPointerRef...
236 ConstantPointerRef *CPPR = cast<ConstantPointerRef>(U);
237 assert(CPPR->getValue() == OldGV && "Something isn't happy");
239 BCR_TRACE(4, "Mutating Forward Ref!\n");
241 // Change the const pool reference to point to the real global variable
242 // now. This should drop a use from the OldGV.
243 CPPR->mutateReference(GV);
246 // Remove GV from the module...
247 GV->getParent()->getGlobalList().remove(OldGV);
248 delete OldGV; // Delete the old placeholder
250 // Remove the map entry for the global now that it has been created...
255 bool BytecodeParser::ParseMethod(const uchar *&Buf, const uchar *EndBuf,
257 // Clear out the local values table...
259 if (MethodSignatureList.empty()) {
260 Error = "Method found, but MethodSignatureList empty!";
261 return failure(true); // Unexpected method!
264 const PointerType *PMTy = MethodSignatureList.front().first; // PtrMeth
265 const MethodType *MTy = dyn_cast<const MethodType>(PMTy->getElementType());
266 if (MTy == 0) return failure(true); // Not ptr to method!
269 if (read_vbr(Buf, EndBuf, isInternal)) return failure(true);
271 unsigned MethSlot = MethodSignatureList.front().second;
272 MethodSignatureList.pop_front();
273 Method *M = new Method(MTy, isInternal != 0);
275 BCR_TRACE(2, "METHOD TYPE: " << MTy << endl);
277 const MethodType::ParamTypes &Params = MTy->getParamTypes();
278 for (MethodType::ParamTypes::const_iterator It = Params.begin();
279 It != Params.end(); ++It) {
280 MethodArgument *MA = new MethodArgument(*It);
281 if (insertValue(MA, Values) == -1) {
282 Error = "Error reading method arguments!\n";
283 delete M; return failure(true);
285 M->getArgumentList().push_back(MA);
288 while (Buf < EndBuf) {
290 const uchar *OldBuf = Buf;
291 if (readBlock(Buf, EndBuf, Type, Size)) {
292 Error = "Error reading Method level block!";
293 delete M; return failure(true);
297 case BytecodeFormat::ConstantPool:
298 BCR_TRACE(2, "BLOCK BytecodeFormat::ConstantPool: {\n");
299 if (ParseConstantPool(Buf, Buf+Size, Values, MethodTypeValues)) {
300 delete M; return failure(true);
304 case BytecodeFormat::BasicBlock: {
305 BCR_TRACE(2, "BLOCK BytecodeFormat::BasicBlock: {\n");
307 if (ParseBasicBlock(Buf, Buf+Size, BB) ||
308 insertValue(BB, Values) == -1) {
309 delete M; return failure(true); // Parse error... :(
312 M->getBasicBlocks().push_back(BB);
316 case BytecodeFormat::SymbolTable:
317 BCR_TRACE(2, "BLOCK BytecodeFormat::SymbolTable: {\n");
318 if (ParseSymbolTable(Buf, Buf+Size, M->getSymbolTableSure())) {
319 delete M; return failure(true);
324 BCR_TRACE(2, "BLOCK <unknown>:ignored! {\n");
326 if (OldBuf > Buf) return failure(true); // Wrap around!
329 BCR_TRACE(2, "} end block\n");
331 if (align32(Buf, EndBuf)) {
332 Error = "Error aligning Method level block!";
333 delete M; // Malformed bc file, read past end of block.
334 return failure(true);
338 if (postResolveValues(LateResolveValues) ||
339 postResolveValues(LateResolveModuleValues)) {
340 Error = "Error resolving method values!";
341 delete M; return failure(true); // Unresolvable references!
344 Value *MethPHolder = getValue(PMTy, MethSlot, false);
345 assert(MethPHolder && "Something is broken no placeholder found!");
346 assert(isa<Method>(MethPHolder) && "Not a method?");
348 unsigned type; // Type slot
349 assert(!getTypeSlot(MTy, type) && "How can meth type not exist?");
350 getTypeSlot(PMTy, type);
352 C->getMethodList().push_back(M);
354 // Replace placeholder with the real method pointer...
355 ModuleValues[type][MethSlot] = M;
357 // Clear out method level types...
358 MethodTypeValues.clear();
360 // If anyone is using the placeholder make them use the real method instead
361 MethPHolder->replaceAllUsesWith(M);
363 // We don't need the placeholder anymore!
366 // If the method is empty, we don't need the method argument entries...
368 M->getArgumentList().delete_all();
370 DeclareNewGlobalValue(M, MethSlot);
375 bool BytecodeParser::ParseModuleGlobalInfo(const uchar *&Buf, const uchar *End,
377 if (!MethodSignatureList.empty()) {
378 Error = "Two ModuleGlobalInfo packets found!";
379 return failure(true); // Two ModuleGlobal blocks?
382 // Read global variables...
384 if (read_vbr(Buf, End, VarType)) return failure(true);
385 while (VarType != Type::VoidTyID) { // List is terminated by Void
386 // VarType Fields: bit0 = isConstant, bit1 = hasInitializer,
387 // bit2 = isInternal, bit3+ = slot#
388 const Type *Ty = getType(VarType >> 3);
389 if (!Ty || !Ty->isPointerType()) {
390 Error = "Global not pointer type! Ty = " + Ty->getDescription();
391 return failure(true);
394 const PointerType *PTy = cast<const PointerType>(Ty);
395 const Type *ElTy = PTy->getElementType();
397 Constant *Initializer = 0;
398 if (VarType & 2) { // Does it have an initalizer?
399 // Do not improvise... values must have been stored in the constant pool,
400 // which should have been read before now.
403 if (read_vbr(Buf, End, InitSlot)) return failure(true);
405 Value *V = getValue(ElTy, InitSlot, false);
406 if (V == 0) return failure(true);
407 Initializer = cast<Constant>(V);
410 // Create the global variable...
411 GlobalVariable *GV = new GlobalVariable(ElTy, VarType & 1, VarType & 4,
413 int DestSlot = insertValue(GV, ModuleValues);
414 if (DestSlot == -1) return failure(true);
416 Mod->getGlobalList().push_back(GV);
418 DeclareNewGlobalValue(GV, unsigned(DestSlot));
420 BCR_TRACE(2, "Global Variable of type: " << PTy->getDescription()
421 << " into slot #" << DestSlot << endl);
423 if (read_vbr(Buf, End, VarType)) return failure(true);
426 // Read the method signatures for all of the methods that are coming, and
427 // create fillers in the Value tables.
428 unsigned MethSignature;
429 if (read_vbr(Buf, End, MethSignature)) return failure(true);
430 while (MethSignature != Type::VoidTyID) { // List is terminated by Void
431 const Type *Ty = getType(MethSignature);
432 if (!Ty || !isa<PointerType>(Ty) ||
433 !isa<MethodType>(cast<PointerType>(Ty)->getElementType())) {
434 Error = "Method not ptr to meth type! Ty = " + Ty->getDescription();
435 return failure(true);
438 // We create methods by passing the underlying MethodType to create...
439 Ty = cast<PointerType>(Ty)->getElementType();
441 // When the ModuleGlobalInfo section is read, we load the type of each
442 // method and the 'ModuleValues' slot that it lands in. We then load a
443 // placeholder into its slot to reserve it. When the method is loaded, this
444 // placeholder is replaced.
446 // Insert the placeholder...
447 Value *Val = new MethPHolder(Ty, 0);
448 if (insertValue(Val, ModuleValues) == -1) return failure(true);
450 // Figure out which entry of its typeslot it went into...
452 if (getTypeSlot(Val->getType(), TypeSlot)) return failure(true);
454 unsigned SlotNo = ModuleValues[TypeSlot].size()-1;
456 // Keep track of this information in a linked list that is emptied as
457 // methods are loaded...
459 MethodSignatureList.push_back(
460 make_pair(cast<const PointerType>(Val->getType()), SlotNo));
461 if (read_vbr(Buf, End, MethSignature)) return failure(true);
462 BCR_TRACE(2, "Method of type: " << Ty << endl);
465 if (align32(Buf, End)) return failure(true);
467 // This is for future proofing... in the future extra fields may be added that
468 // we don't understand, so we transparently ignore them.
474 bool BytecodeParser::ParseModule(const uchar *Buf, const uchar *EndBuf,
478 if (readBlock(Buf, EndBuf, Type, Size)) return failure(true);
479 if (Type != BytecodeFormat::Module || Buf+Size != EndBuf) {
480 Error = "Expected Module packet!";
481 return failure(true); // Hrm, not a class?
484 BCR_TRACE(0, "BLOCK BytecodeFormat::Module: {\n");
485 MethodSignatureList.clear(); // Just in case...
487 // Read into instance variables...
488 if (read_vbr(Buf, EndBuf, FirstDerivedTyID)) return failure(true);
489 if (align32(Buf, EndBuf)) return failure(true);
490 BCR_TRACE(1, "FirstDerivedTyID = " << FirstDerivedTyID << "\n");
492 TheModule = C = new Module();
493 while (Buf < EndBuf) {
494 const uchar *OldBuf = Buf;
495 if (readBlock(Buf, EndBuf, Type, Size)) { delete C; return failure(true); }
497 case BytecodeFormat::ConstantPool:
498 BCR_TRACE(1, "BLOCK BytecodeFormat::ConstantPool: {\n");
499 if (ParseConstantPool(Buf, Buf+Size, ModuleValues, ModuleTypeValues)) {
500 delete C; return failure(true);
504 case BytecodeFormat::ModuleGlobalInfo:
505 BCR_TRACE(1, "BLOCK BytecodeFormat::ModuleGlobalInfo: {\n");
507 if (ParseModuleGlobalInfo(Buf, Buf+Size, C)) {
508 delete C; return failure(true);
512 case BytecodeFormat::Method: {
513 BCR_TRACE(1, "BLOCK BytecodeFormat::Method: {\n");
514 if (ParseMethod(Buf, Buf+Size, C)) {
515 delete C; return failure(true); // Error parsing method
520 case BytecodeFormat::SymbolTable:
521 BCR_TRACE(1, "BLOCK BytecodeFormat::SymbolTable: {\n");
522 if (ParseSymbolTable(Buf, Buf+Size, C->getSymbolTableSure())) {
523 delete C; return failure(true);
528 Error = "Expected Module Block!";
530 if (OldBuf > Buf) return failure(true); // Wrap around!
533 BCR_TRACE(1, "} end block\n");
534 if (align32(Buf, EndBuf)) { delete C; return failure(true); }
537 if (!MethodSignatureList.empty()) { // Expected more methods!
538 Error = "Method expected, but bytecode stream at end!";
539 return failure(true);
542 BCR_TRACE(0, "} end block\n\n");
546 Module *BytecodeParser::ParseBytecode(const uchar *Buf, const uchar *EndBuf) {
547 LateResolveValues.clear();
549 // Read and check signature...
550 if (read(Buf, EndBuf, Sig) ||
551 Sig != ('l' | ('l' << 8) | ('v' << 16) | 'm' << 24)) {
552 Error = "Invalid bytecode signature!";
553 return failure<Module*>(0); // Invalid signature!
557 if (ParseModule(Buf, EndBuf, Result)) return 0;
562 Module *ParseBytecodeBuffer(const uchar *Buffer, unsigned Length) {
563 BytecodeParser Parser;
564 return Parser.ParseBytecode(Buffer, Buffer+Length);
567 // Parse and return a class file...
569 Module *ParseBytecodeFile(const string &Filename, string *ErrorStr) {
573 if (Filename != string("-")) { // Read from a file...
574 int FD = open(Filename.c_str(), O_RDONLY);
576 if (ErrorStr) *ErrorStr = "Error opening file!";
577 return failure<Module*>(0);
580 if (fstat(FD, &StatBuf) == -1) { close(FD); return failure<Module*>(0); }
582 int Length = StatBuf.st_size;
584 if (ErrorStr) *ErrorStr = "Error stat'ing file!";
585 close(FD); return failure<Module*>(0);
587 uchar *Buffer = (uchar*)mmap(0, Length, PROT_READ,
589 if (Buffer == (uchar*)-1) {
590 if (ErrorStr) *ErrorStr = "Error mmapping file!";
591 close(FD); return failure<Module*>(0);
594 BytecodeParser Parser;
595 Result = Parser.ParseBytecode(Buffer, Buffer+Length);
597 munmap((char*)Buffer, Length);
599 if (ErrorStr) *ErrorStr = Parser.getError();
600 } else { // Read from stdin
603 uchar Buffer[4096], *FileData = 0;
604 while ((BlockSize = read(0, Buffer, 4))) {
605 if (BlockSize == -1) { free(FileData); return failure<Module*>(0); }
607 FileData = (uchar*)realloc(FileData, FileSize+BlockSize);
608 memcpy(FileData+FileSize, Buffer, BlockSize);
609 FileSize += BlockSize;
613 if (ErrorStr) *ErrorStr = "Standard Input empty!";
614 free(FileData); return failure<Module*>(0);
619 uchar *Buf = (uchar*)mmap(0, FileSize, PROT_READ|PROT_WRITE,
620 MAP_PRIVATE|MAP_ANONYMOUS, -1, 0);
621 assert((Buf != (uchar*)-1) && "mmap returned error!");
622 memcpy(Buf, FileData, FileSize);
625 uchar *Buf = FileData;
628 BytecodeParser Parser;
629 Result = Parser.ParseBytecode(Buf, Buf+FileSize);
632 munmap((char*)Buf, FileSize); // Free mmap'd data area
634 free(FileData); // Free realloc'd block of memory
637 if (ErrorStr) *ErrorStr = Parser.getError();