Skip to content
Snippets Groups Projects
SMPFunction.cpp 338 KiB
Newer Older
		else if ((this->OutgoingArgsSize == 0) && (!TempEntry.Read) && (!TempEntry.Written)) {
			// We have not started accumulating outgoing args bytes, we have reached the
			//  AllocPointDelta, yet we find space that is neither written nor read. This
			//  empty space at the bottom of the stack frame could just be for stack alignment
			//  purposes, especially in the new x86-64 ABI, so it should not prevent us from
			//  finding outgoing args space above it.
			++AlignmentPadSize;
		}
		else if (TempEntry.Read || TempEntry.EBPRelativeAccess || !TempEntry.Written
			|| !TempEntry.ESPRelativeAccess) {
			OutgoingArgsRegionFinished = true;
			if (!TempEntry.Read && !TempEntry.Written) {
				// Could be stack frame padding.
				++FramePadSize;
			}
			else {
				break; // No padding region
			}
		}
		else {
			this->OutgoingArgsSize++;
			if (TempEntry.IndexedAccess) {
				IndexedOutgoingArgs = true;
			}
		}
	}

	// Add in the alignment padding below the written outargs region.
	if (this->OutgoingArgsSize > 0) {
		this->OutgoingArgsSize += AlignmentPadSize;
	}

	// If any outgoing arg was accessed using an index register, then we don't know how high
	//  the index register value went. It could potentially consume the so-called padding
	//  region, which might be just the region we did not detect direct accesses to because
	//  the accesses were indirect. To be safe, we expand the outgoing args region to fill
	//  the padding region above it in this indexed access case.
	if (IndexedOutgoingArgs) {
		this->OutgoingArgsSize += FramePadSize;
	// Sometimes we encounter unused stack space above the outgoing args. Lump this space
	//  in with the outgoing args. We detect this by noting when the outgoing args space
	//  has only partially used the space assigned to a local var.
	// NOTE: This is usually just stack padding to maintain stack alignment. It could
	//  also be the case that the lowest local variable is accessed indirectly and we missed
	//  seeing its address taken, in which case it would be unsound to lump it into the
	//  outgoing args region. We might want to create a local var called STACKPAD
	//  to occupy this space.
	if ((0 < this->OutgoingArgsSize) && (this->OutgoingArgsSize < this->FuncInfo.frsize)) {
		long MapIndex = (this->AllocPointDelta - this->MinStackDelta);
		assert(0 <= MapIndex);
		MapIndex += (((long) this->OutgoingArgsSize) - 1);
		struct StackFrameEntry TempEntry = this->StackFrameMap.at((size_t) MapIndex);
		if (NULL == TempEntry.VarPtr) { // Gap in stack frame; IDA 6.0
			SMP_msg("Gap in stack frame: %s\n", this->GetFuncName());
		}
		else if (this->OutgoingArgsSize < (TempEntry.VarPtr->offset + TempEntry.VarPtr->size)) {
			SMP_msg("OutGoingArgsSize = %d", this->OutgoingArgsSize);
			this->OutgoingArgsSize = TempEntry.VarPtr->offset + TempEntry.VarPtr->size;
			SMP_msg(" adjusted to %d\n", this->OutgoingArgsSize);
	return;
} // end of SMPFunction::FindOutgoingArgsSize()

// If TempOp reads or writes to a stack location, return the offset (relative to the initial
//  stack pointer value) and the size in bytes of the data access. Also return whether the
//  access was frame-pointer-relative, and whether signedness can be inferred due to a load
//  from the stack being zero-extended or sign-extended.
// NOTE: This function assumes that offsets are already normalized. i.e. the TempOp argument
//  should always come from a DEF or USE that has been normalized to the stack delta at function entry.
// NOTE: TempOp must be of type o_displ or o_phrase, as no other operand type could be a
//  stack memory access.
// BaseValue is either this->MinStackAccessOffset, or this->MinStackDelta (when this->MinStackAccessOffset is still
//  being computed).
// Return true if a stack memory access was found in TempOp, false otherwise.
bool SMPFunction::MDGetStackOffsetAndSize(SMPInstr *Instr, op_t TempOp, sval_t BaseValue, ea_t &offset, size_t &DataSize, bool &FP,
										  bool &Indexed, bool &Signed, bool &Unsigned) {
	sval_t sp_delta = Instr->GetStackPtrOffset();
	ea_t InstAddr = Instr->GetAddr(); // helps debugging
	assert((o_displ == TempOp.type) || (o_phrase == TempOp.type));
	MDExtractAddressFields(TempOp, BaseReg, IndexReg, ScaleFactor, offset);
	if (TempOp.type == o_phrase) {
		assert(offset == 0);  // implicit zero, as in [esp] ==> [esp+0]

	SignedOffset = (int) offset;  // avoid sign errors during adjustment arithmetic

	if ((BaseReg == R_sp) || (IndexReg == R_sp)) {
		// ESP-relative constant offset
		if (!Instr->AreDefsNormalized()) {
			SignedOffset += sp_delta; // base offsets from entry ESP value
		}
		SignedOffset -= BaseValue; // convert to StackFrameMap index
		offset = (ea_t) SignedOffset; // write back to outgoing argument
		// Get size of data written
		DataSize = GetOpDataSize(TempOp);
		FP = false;
		Indexed = ((BaseReg != R_none) && (IndexReg != R_none)); // two regs used
		unsigned short opcode = Instr->GetCmd().itype;
		Unsigned = (opcode == NN_movzx);
		Signed = (opcode == NN_movsx);
		if ((0 > SignedOffset) && (!Indexed) && (BaseValue == this->MinStackAccessOffset)) {
			// Consider asserting here.
			SMP_msg("ERROR: Negative offset in MDGetStackOffsetAndSize for inst dump: \n");
	else if (this->UseFP && ((BaseReg == MD_FRAME_POINTER_REG) || (IndexReg == MD_FRAME_POINTER_REG))) {
		SignedOffset -= this->FuncInfo.frregs; // base offsets from entry ESP value
		SignedOffset -= BaseValue; // convert to StackFrameMap index
		offset = (ea_t) SignedOffset;
		DataSize = GetOpDataSize(TempOp);
		FP = true;
		Indexed = ((BaseReg != R_none) && (IndexReg != R_none)); // two regs used
		assert(Indexed || (!this->StackPtrAnalysisSucceeded()) || !this->HasSTARSStackPtrAnalysisCompleted()); // Else we should never get here with unnormalized stack operands
#else
		if (!(Indexed || (!this->StackPtrAnalysisSucceeded()) || !this->HasSTARSStackPtrAnalysisCompleted())) {
			SMP_msg("WARNING: Unnormalized FP-relative stack offset at %lx after stack analysis succeeded.\n",
				(unsigned long) Instr->GetAddr());
		unsigned short opcode = Instr->GetCmd().itype;
		Unsigned = (opcode == NN_movzx);
		Signed = (opcode == NN_movsx);
		if ((0 > SignedOffset) && (!Indexed) && (BaseValue == this->MinStackAccessOffset)) {
			// Consider asserting here.
			SMP_msg("ERROR: Negative offset %d in MDGetStackOffsetAndSize: frregs: %d MinStackDelta: %ld Inst dump: \n",
				SignedOffset, this->FuncInfo.frregs, (long) this->MinStackDelta);
		return true;
	}
	else {
		return false;
	}
} // end of SMPFunction::MDGetStackOffsetAndSize()
		
// Return fine grained stack entry for stack op TempOp from instruction at InstAddr
bool SMPFunction::MDGetFGStackLocInfo(ea_t InstAddr, op_t TempOp, struct FineGrainedInfo &FGEntry) {
	int BaseReg;
	int IndexReg;
	ushort ScaleFactor;
	ea_t offset;
	int SignedOffset;

	assert((o_displ == TempOp.type) || (o_phrase == TempOp.type));
	MDExtractAddressFields(TempOp, BaseReg, IndexReg, ScaleFactor, offset);

	SignedOffset = (int) offset;

	if (TempOp.type == o_phrase) {
		assert(SignedOffset == 0);  // implicit zero, as in [esp] ==> [esp+0]
	if ((BaseReg == MD_STACK_POINTER_REG) || (IndexReg == MD_STACK_POINTER_REG)) {
		// ESP-relative constant offset
		SignedOffset -= this->MinStackAccessOffset; // convert to StackFrameMap index
	else if (this->UseFP && ((BaseReg == MD_FRAME_POINTER_REG) || (IndexReg == MD_FRAME_POINTER_REG))) {
		assert(false); // should never get here with unnormalized stack operand
		SignedOffset -= this->FuncInfo.frregs; // base offsets from entry ESP value
		SignedOffset -= this->MinStackAccessOffset; // convert to StackFrameMap index
	}
	else {
		return false;
	}
	// We did not return false, so we should have a good offset. Use it to
	//  pass back the fine grained stack table entry for that offset.
	if ((0 > SignedOffset) || (SignedOffset >= (int) this->FineGrainedStackTable.size())) {
		if (this->OutgoingArgsComputed) {
			SMP_msg("ERROR: FG stack table index out of range in MDGetFGStackLocInfo at %lx\n",
				(unsigned long) InstAddr);
		FGEntry.SignMiscInfo = 0; // We cannot figure out signedness info without an FG info stack table.
		FGEntry.SizeInfo = ComputeOperandBitWidthMask(TempOp, 0); // IDA can figure out width, anyway.
		FGEntry = this->FineGrainedStackTable.at((size_t) SignedOffset);
	}
	return true;
} // end of SMPFunction::MDGetFGStackLocInfo()

// Return true if we update fine grained stack entry for stack op TempOp from instruction at InstAddr
bool SMPFunction::MDUpdateFGStackLocInfo(ea_t InstAddr, op_t TempOp, struct FineGrainedInfo NewFG) {
	int BaseReg;
	int IndexReg;
	ushort ScaleFactor;
	ea_t offset;
	int SignedOffset;
	struct FineGrainedInfo OldFG, UnionFG;

	assert((o_displ == TempOp.type) || (o_phrase == TempOp.type));
	MDExtractAddressFields(TempOp, BaseReg, IndexReg, ScaleFactor, offset);

	SignedOffset = (int) offset;

	if (TempOp.type == o_phrase) {
		assert(SignedOffset == 0);  // implicit zero, as in [esp] ==> [esp+0]
	}
	if ((BaseReg == MD_STACK_POINTER_REG) || (IndexReg == MD_STACK_POINTER_REG)) {
		// ESP-relative constant offset
		SignedOffset -= this->MinStackAccessOffset; // convert to StackFrameMap index
	else if (this->UseFP && ((BaseReg == MD_FRAME_POINTER_REG) || (IndexReg == MD_FRAME_POINTER_REG))) {
		assert(false); // should never get here with unnormalized stack operands
		SignedOffset -= this->FuncInfo.frregs; // base offsets from entry ESP value
		SignedOffset -= this->MinStackAccessOffset; // convert to StackFrameMap index
	}
	else {
		return false;
	}
	// We did not return false, so we should have a good offset. Use it to
	//  retrieve the fine grained stack table entry for that offset.
	if ((0 > SignedOffset) || (SignedOffset >= (int) this->FineGrainedStackTable.size())) {
		if (this->OutgoingArgsComputed) {
			SMP_msg("ERROR: FG stack table index out of range in MDGetFGStackLocInfo at %lx\n",
				(unsigned long) InstAddr);
		}
		return false;
	}
	else if (this->OutgoingArgsComputed && (((size_t)SignedOffset) < this->OutgoingArgsSize)) {
		// We don't want to update the outgoing args region, as it will not be consistent
		//  over multiple function calls. NOTE: We could fine tune this by seeing if we
		//  call mutliple target functions or not; if only one, then outgoing args region
		//  would be consistent in the absence of varargs targets.
		return false;
	}
	else {
		OldFG = this->FineGrainedStackTable.at((size_t) SignedOffset);
		UnionFG.SignMiscInfo = OldFG.SignMiscInfo | NewFG.SignMiscInfo;
		UnionFG.SizeInfo = OldFG.SizeInfo | NewFG.SizeInfo;
		if ((OldFG.SignMiscInfo != UnionFG.SignMiscInfo) || (OldFG.SizeInfo != UnionFG.SizeInfo)) {
			// The signs they are a-changin'. Or maybe the sizes.
			this->FineGrainedStackTable.at(SignedOffset).SignMiscInfo |= NewFG.SignMiscInfo;
			this->FineGrainedStackTable.at(SignedOffset).SizeInfo |= NewFG.SizeInfo;
		}
	}
	return true;
} // end of SMPFunction::MDUpdateFGStackLocInfo()

// retrieve DEF addr from GlobalDefAddrBySSA or return BADADDR
ea_t SMPFunction::GetGlobalDefAddr(op_t DefOp, int SSANum) {
	map<int, ea_t>::iterator MapResult;
	ea_t DefAddr = BADADDR; // BADADDR means we did not find it
	bool RegDef = (o_reg == DefOp.type);
	if (RegDef) {
		int HashedName = HashGlobalNameAndSSA(DefOp, SSANum);
		MapResult = this->GlobalDefAddrBySSA.find(HashedName);
		if (MapResult != this->GlobalDefAddrBySSA.end()) { // Found it.
			DefAddr = (ea_t) MapResult->second;
		}
	}
	else if (MDIsStackAccessOpnd(DefOp, this->UsesFramePointer())) {
		// Until we get stack operands into the GlobalDefAddrBySSA map,
		//  do a linear search.
		list<SMPInstr *>::iterator InstIter = this->Instrs.begin();
		set<DefOrUse, LessDefUse>::iterator DefIter;
		SMPInstr *CurrInst = (*InstIter);
		if (CurrInst->IsFloatNop()) { // marker inst
			if (0 == SSANum) { // Live-in-to-func stack locations get DEF at marker inst.
				DefIter = CurrInst->FindDef(DefOp);
				if (DefIter != CurrInst->GetLastDef()) {
					// Found it. Must be SSA 0.
					assert(0 == DefIter->GetSSANum());
					DefAddr = CurrInst->GetAddr();
					return DefAddr;
				}
			}
			++InstIter;
		}
		for ( ; InstIter != this->Instrs.end(); ++InstIter) {
			CurrInst = (*InstIter);
			if (CurrInst->HasDestMemoryOperand()) {
				op_t MemDefOp = CurrInst->GetMemDef();
				if (IsEqOp(DefOp, MemDefOp)) {
					DefIter = CurrInst->FindDef(MemDefOp);
					assert(DefIter != CurrInst->GetLastDef());
					int DefSSANum = DefIter->GetSSANum();
					if (DefSSANum == SSANum) { // found it
						DefAddr = CurrInst->GetAddr();
						break;
					}
				}
			}
		}
	}
	return DefAddr;
} // end of SMPFunction::GetGlobalDefAddr()

int SMPFunction::GetBlockNumForPhiDef(op_t DefOp, int SSANum) {
	size_t BlockIndex;
	for (BlockIndex = 0; BlockIndex < this->RPOBlocks.size(); ++BlockIndex) {
		SMPBasicBlock *CurrBlock = this->RPOBlocks.at(BlockIndex);
		set<SMPPhiFunction, LessPhi>::iterator PhiIter = CurrBlock->FindPhi(DefOp);
		if (PhiIter != CurrBlock->GetLastPhi()) {
			if (PhiIter->GetDefSSANum() == SSANum) {
				return CurrBlock->GetNumber();
			}
		}
	}
	return (int) BADADDR;
} // end of SMPFunction::GetBlockNumForPhiDef()

// Retrieve block iterator for InstAddr from InstBlockMap; assert if failure
SMPBasicBlock *SMPFunction::GetBlockFromInstAddr(ea_t InstAddr) {
	map<ea_t, SMPBasicBlock *>::iterator MapEntry;
	MapEntry = this->InstBlockMap.find(InstAddr);
	assert(MapEntry != this->InstBlockMap.end());
	return MapEntry->second;
}

// Retrieve inst pointer for InstAddr; assert if failure on block find.
SMPInstr *SMPFunction::GetInstFromAddr(ea_t InstAddr) {
	SMPBasicBlock *CurrBlock = this->GetBlockFromInstAddr(InstAddr);
	SMPInstr *CurrInst = CurrBlock->FindInstr(InstAddr);
	return CurrInst;
// Given block # and PhiDef op_t and SSANum, return the Phi iterator or assert.
set<SMPPhiFunction, LessPhi>::iterator SMPFunction::GetPhiIterForPhiDef(size_t BlockNumber, op_t DefOp, int SSANum) {
	SMPBasicBlock *DefBlock = this->RPOBlocks.at(BlockNumber);
	set<SMPPhiFunction, LessPhi>::iterator PhiIter = DefBlock->FindPhi(DefOp);
	assert(PhiIter != DefBlock->GetLastPhi());
	return PhiIter;
}
// Is DestOp within the outgoing args area? Assume it must be an ESP-relative
//  DEF operand in order to be a write to the outgoing args area.
// NOTE: DestOp should be already normalized to the entry stack delta.
bool SMPFunction::IsInOutgoingArgsRegion(op_t DestOp) {
	bool OutArgWrite = false;
	int BaseReg, IndexReg;
	ushort ScaleFactor;
	ea_t offset;

	if (this->IsLeaf())
		return false;

	MDExtractAddressFields(DestOp, BaseReg, IndexReg, ScaleFactor, offset);
	if ((BaseReg != R_sp) && (IndexReg != R_sp))
		return false;
	if (((BaseReg == R_sp) && (IndexReg != R_none))
		|| ((IndexReg == R_sp) && (BaseReg != R_none))
		|| (0 < ScaleFactor)) {

		SMP_msg("WARNING: WritesToOutgoingArgs called with indexed write.");
		return false;
	}

	if (!this->OutgoingArgsComputed) {
		OutArgWrite = true; // be conservative
	}
	else {
		int SignedOffset = (int) offset;
		SignedOffset -= this->MinStackDelta; // convert to zero-based from bottom of stack frame
		OutArgWrite = (((size_t) SignedOffset) < this->OutgoingArgsSize);
} // end of SMPFunction::IsInOutgoingArgsRegion()

// Is DestOp a direct memory access above the local vars frame?
bool SMPFunction::WritesAboveLocalFrame(op_t DestOp, bool OpNormalized) {
	bool InArgWrite = false;
	int BaseReg, IndexReg;
	ushort ScaleFactor;
	ea_t offset;
	long SignedOffset;

	MDExtractAddressFields(DestOp, BaseReg, IndexReg, ScaleFactor, offset);
	SignedOffset = (long) offset;
	bool ESPrelative = (BaseReg == MD_STACK_POINTER_REG) || (IndexReg == MD_STACK_POINTER_REG);
	bool EBPrelative = this->UseFP && ((BaseReg == MD_FRAME_POINTER_REG) || (IndexReg == MD_FRAME_POINTER_REG));
	assert(!EBPrelative || !OpNormalized); // stack operands should be normalized by now
	if (!(ESPrelative || EBPrelative))
		return false;
	if (((IndexReg != R_none) && (BaseReg != R_none))
		|| (0 < ScaleFactor)) {

		SMP_msg("WARNING: WritesAboveLocalFrame called with indexed write.");
	// The next statement omits a complication: The possibility that OpNormalized is false,
	//  and an ESPRelative access is above the stack frame. For the purposes of determining
	//  whether a function is safe, this is irrelevant, because !OpNormalized would indicate
	//  that AnalyzedSP is false, which will make the function unsafe anyway. Future uses for
	//  other purposes need to fix this.
	InArgWrite = (ESPrelative && OpNormalized && (SignedOffset >= 0))
		|| (EBPrelative && (SignedOffset > 0));
	if (InArgWrite && OpNormalized && (0 == SignedOffset)) {
		SMP_msg("DANGER: Write to saved return address detected in function that begins at %lx\n",
			(unsigned long) this->FirstEA);
}// end of SMPFunction::WritesAboveLocalFrame()

// Is DestOp an indexed write above the local vars frame?
bool SMPFunction::IndexedWritesAboveLocalFrame(op_t DestOp) {
	bool InArgWrite = false;
	int BaseReg, IndexReg;
	ushort ScaleFactor;
	ea_t offset;

	MDExtractAddressFields(DestOp, BaseReg, IndexReg, ScaleFactor, offset);
	bool ESPrelative = (BaseReg == MD_STACK_POINTER_REG) || (IndexReg == MD_STACK_POINTER_REG);
	bool EBPrelative = this->UseFP && ((BaseReg == MD_FRAME_POINTER_REG) || (IndexReg == MD_FRAME_POINTER_REG));
	assert(!EBPrelative || !this->StackPtrAnalysisSucceeded() || !this->HasSTARSStackPtrAnalysisCompleted()); // stack operands should be normalized by now
	if (!(ESPrelative || EBPrelative))
		return false;

	InArgWrite = (ESPrelative && (SignedOffset > 0))
		|| (EBPrelative && (SignedOffset > 0));
} // end of SMPFunction::IndexedWritesAboveLocalFrame()
// Is CurrOp found anywhere in the StackPtrCopySet, regardless of which address and stack delta
//  values are associated with it?
bool SMPFunction::IsInStackPtrCopySet(op_t CurrOp) {
	bool found = false;
	// Set is composed of triples, so we have to iterate through it and compare operands.
	set<pair<op_t, pair<ea_t, sval_t> >, LessStackDeltaCopy>::iterator CopyIter;
	for (CopyIter = this->StackPtrCopySet.begin(); CopyIter != this->StackPtrCopySet.end(); ++CopyIter) {
		pair<op_t, pair<ea_t, sval_t> > CurrCopy = *CopyIter;
		op_t CopyOp = CurrCopy.first;
		if (IsEqOp(CopyOp, CurrOp)) {
			// Found it.
			found = true;
			break;
		}
		else if (CopyOp.type > CurrOp.type) {
			// already moved past its spot; not found
			break;
		}
	}

	return found;
} // end of SMPFunction::IsInStackPtrCopySet()

// Find evidence of calls to alloca(), which appear as stack space allocations (i.e.
//  subtractions [of unknown values(?)] from the stack pointer) AFTER the local frame allocation instruction
//  for this function.
// Return true if such an allocation is found and false otherwise.
bool SMPFunction::FindAlloca(void) {
	bool FoundAlloca = false;
	list<SMPInstr *>::iterator InstIter = this->Instrs.begin();
	SMPInstr *CurrInst;
	ea_t InstAddr;
#if SMP_USE_SSA_FNOP_MARKER
	++InstIter;  // skip marker instruction
	for ( ; InstIter != this->Instrs.end(); ++InstIter) {
		CurrInst = (*InstIter);
		InstAddr = CurrInst->GetAddr();
		if (InstAddr > this->LocalVarsAllocInstr) {
			if (CurrInst->MDIsFrameAllocInstr()) {
				FoundAlloca = true;
				if (CurrInst->HasAllocaRTL()) {
					CurrInst->SetAllocaCall();
				}
			}
			else if (CurrInst->MDIsPushInstr()) {
				this->PushAfterLocalVarAlloc = true;
			}
} // end of SMPFunction::FindAlloca()

// Emit the annotations describing the regions of the stack frame.
void SMPFunction::EmitStackFrameAnnotations(FILE *AnnotFile, SMPInstr *Instr) {
	ea_t addr = Instr->GetAddr();

#if 0
	if (0 < IncomingArgsSize) {
		SMP_fprintf(AnnotFile, "%10lx %6d INARGS STACK esp + %ld %s \n",
				(unsigned long) addr, IncomingArgsSize,
				(long) (LocalVarsSize + CalleeSavedRegsSize + RetAddrSize),
				Instr->GetDisasm());
		SMP_fprintf(AnnotFile, "%10lx %6d MEMORYHOLE STACK esp + %lu ReturnAddress \n",
				(unsigned long) addr, RetAddrSize, (unsigned long) (this->LocalVarsSize + this->CalleeSavedRegsSize));
	if (0 < this->CalleeSavedRegsSize) {
		SMP_fprintf(AnnotFile, "%10lx %6u MEMORYHOLE STACK esp + %lu CalleeSavedRegs \n",
				(unsigned long) addr, this->CalleeSavedRegsSize, (unsigned long) this->LocalVarsSize);
	if ((0 < this->LocalVarsSize) && this->GoodLocalVarTable) {
		unsigned long ParentReferentID = DataReferentID++;
		SMP_fprintf(AnnotFile, "%10lx %6lu DATAREF STACK %lu esp + %d PARENT LocalFrame LOCALFRAME\n",
				(unsigned long) addr, (unsigned long) this->LocalVarsSize, ParentReferentID, 0);
#if SMP_COMPUTE_STACK_GRANULARITY
		if (this->AnalyzedSP && !this->CallsAlloca && (BADADDR != this->LocalVarsAllocInstr)) {
			// We can only fine-grain the stack frame if we were able to analyze the stack
			if (this->OutgoingArgsSize > 0) {
				SMP_fprintf(AnnotFile, "%10lx %6zu DATAREF STACK %lu esp + %d CHILDOF %lu OFFSET %d OutArgsRegion OUTARGS\n",
					(unsigned long) addr, this->OutgoingArgsSize, DataReferentID, 0, ParentReferentID, 0);
#if SMP_DEBUG_STACK_GRANULARITY
			SMP_msg("LocalVarTable of size %d for function %s\n", this->LocalVarTable.size(),
				this->GetFuncName());
			for (size_t i = 0; i < this->LocalVarTable.size(); ++i) {
#if SMP_DEBUG_STACK_GRANULARITY
				SMP_msg("Entry %d offset %ld size %d name %s\n", i, this->LocalVarTable[i].offset,
					this->LocalVarTable[i].size, this->LocalVarTable[i].VarName);
				// Don't emit annotations for incoming or outgoing args or anything else
				//  above or below the current local frame.
				if ((this->LocalVarTable[i].offset >= (long) this->FuncInfo.frsize)
					|| (this->LocalVarTable[i].offset < (long) this->OutgoingArgsSize))
					continue;
				SMP_fprintf(AnnotFile, "%10lx %6zu DATAREF STACK %lu esp + %ld CHILDOF %lu OFFSET %ld LOCALVAR %s \n",
					(unsigned long) addr, this->LocalVarTable[i].size, DataReferentID,
					this->LocalVarTable[i].offset, ParentReferentID,
					this->LocalVarTable[i].offset, this->LocalVarTable[i].VarName);
				++DataReferentID;
		} // end if (this->AnalyzedSP and not Alloca .... )
	} // end if (0 < LocalVarsSize)
	return;
} // end of SMPFunction::EmitStackFrameAnnotations() 

// Audit and fix the IDA Pro code cross references for jumps and jump targets.
void SMPFunction::MDAuditJumpXrefs(void) {
	func_tail_iterator_t FuncTail(this->GetFuncInfo());
	enum cref_t NearJump = (cref_t)(fl_JN | XREF_USER);
	enum cref_t FarJump = (cref_t)(fl_JF | XREF_USER);

 	for (bool ChunkOK = FuncTail.main(); ChunkOK; ChunkOK = FuncTail.next()) {
		const area_t &CurrChunk = FuncTail.chunk();

		// Find the instructions for each chunk, audit the xrefs.
		for (ea_t addr = CurrChunk.startEA; addr < CurrChunk.endEA; addr = get_item_end(addr)) {
			flags_t InstrFlags = getFlags(addr);
			if (isHead(InstrFlags) && isCode(InstrFlags)) {
				// Fill cmd structure with disassembly of instr
				insn_t LocalCmd;
				ulong LocalFeatures;
				if (!SMPGetCmd(addr, LocalCmd, LocalFeatures)) {
					SMP_msg("ERROR: SMPGetCmd failed from MDAuditJumpXrefs at %lx\n",
						(unsigned long) addr);
				}
				else {
					// Determine whether the instruction is a jump target by looking
					//  at its cross references and seeing if it has "TO" code xrefs.
					SMP_xref_t xrefs, Distant_xrefs;
					for (bool ok = xrefs.SMP_first_to(addr, XREF_FAR); ok; ok = xrefs.SMP_next_to()) {
						ea_t DistantAddr = xrefs.GetFrom();
						if ((DistantAddr != 0) && (xrefs.GetIscode())) {
							// Now we see if the distant instruction has an xref to this instruction.
							bool FoundDistantXref = false;
							for (bool ok2 = Distant_xrefs.SMP_first_from(DistantAddr, XREF_FAR); ok2; ok2 = Distant_xrefs.SMP_next_from()) {
								ea_t TargetAddr = Distant_xrefs.GetTo();
								if (TargetAddr == addr) {
									FoundDistantXref = true;
									break;
								}
							}
							if (!FoundDistantXref) {
								SMP_msg("WARNING: Missing code Xref from %lx to %lx\n",
									(unsigned long) DistantAddr, (unsigned long) addr);
								long SignedAddrDiff = (long) (DistantAddr - addr);
								if ((SignedAddrDiff < -128) || (SignedAddrDiff > 127)) {
									add_cref(DistantAddr, addr, FarJump);
								}
								else {
									add_cref(DistantAddr, addr, NearJump);
								}
							}
						}
					} // end for all "to" xrefs
					// Now check the "from" xrefs to see if the target inst has the corresponding "to" xref.
					for (bool ok = xrefs.SMP_first_from(addr, XREF_FAR); ok; ok = xrefs.SMP_next_from()) {
						ea_t DistantAddr = xrefs.GetTo();
						if ((DistantAddr != 0) && (xrefs.GetIscode())) {
							// Now we see if the distant instruction has an xref to this instruction.
							bool FoundDistantXref = false;
							for (bool ok2 = Distant_xrefs.SMP_first_to(DistantAddr, XREF_FAR); ok2; ok2 = Distant_xrefs.SMP_next_to()) {
								ea_t SourceAddr = Distant_xrefs.GetFrom();
								if (SourceAddr == addr) {
									FoundDistantXref = true;
									break;
								}
							}
							if (!FoundDistantXref) {
								SMP_msg("WARNING: Missing code Xref to %lx from %lx\n",
									(unsigned long) DistantAddr, (unsigned long) addr);
								long SignedAddrDiff = (long) (DistantAddr - addr);
								if ((SignedAddrDiff < -128) || (SignedAddrDiff > 127)) {
									add_cref(DistantAddr, addr, FarJump);
								}
								else {
									add_cref(DistantAddr, addr, NearJump);
								}
							}
						}
					} // end for all "from" xrefs
				} // end if (!SMPGetCmd() ... else ...
			} // end if (IsHead() and IsCode())
		} // end for all addrs in chunk
	} // end for all chunks

	return;
} // end of SMPFunction::MDAuditJumpXrefs()

// Main data flow analysis driver. Goes through the function and
//  fills all objects for instructions, basic blocks, and the function
//  itself.
void SMPFunction::Analyze(void) {
	list<SMPInstr *>::iterator FirstInBlock = this->Instrs.end();
	   // For starting a basic block
	list<SMPInstr *>::iterator LastInBlock = this->Instrs.end();
	   // Terminating a basic block
	sval_t CurrStackPointerOffset = 0;
	set<ea_t> FragmentWorkList;  // Distant code fragments that belong to this function and need processing
	ea_t InstAddr; // grab address to help in debugging, conditional breakpoints, etc.
	ea_t PreviousIndirJumpAddr = BADADDR;
	enum cref_t NearJump = (cref_t)(fl_JN | XREF_USER);
	enum cref_t FarJump = (cref_t)(fl_JF | XREF_USER);

#if SMP_DEBUG_CONTROLFLOW
	SMP_msg("Entering SMPFunction::Analyze.\n");
#endif

	// Get some basic info from the FuncInfo structure.
	this->Size = this->FuncInfo.endEA - this->FuncInfo.startEA;
	this->UseFP = (0 != (this->FuncInfo.flags & (FUNC_FRAME | FUNC_BOTTOMBP)));
	this->StaticFunc = (0 != (this->FuncInfo.flags & FUNC_STATIC));
	this->LibFunc = (0 != (this->FuncInfo.flags & FUNC_LIB));
	this->AnalyzedSP = this->FuncInfo.analyzed_sp();

#if SMP_DEBUG_CONTROLFLOW
	SMP_msg("SMPFunction::Analyze: got basic info.\n");
	// Determine if we are dealing with shared chunks.
	size_t ChunkCounter = 0;
	func_tail_iterator_t FuncTail(this->GetFuncInfo());
	ea_t FuncHeadLastAddr = 0;
	for (bool ChunkOK = FuncTail.main(); ChunkOK; ChunkOK = FuncTail.next()) {
		const area_t &CurrChunk = FuncTail.chunk();
		if (1 == ChunkCounter) { // head chunk
			FuncHeadLastAddr = CurrChunk.endEA;
		}
		else { // a tail chunk
#if STARS_FIND_UNSHARED_CHUNKS
			if (this->GetProg()->IsChunkUnshared(CurrChunk.startEA, this->FirstEA, FuncHeadLastAddr)) {
				this->UnsharedChunks = true;
#if SMP_DEBUG_CHUNKS
				SMP_msg("INFO: Found unshared tail chunk for %s at %lx\n",
					this->GetFuncName(), (unsigned long) CurrChunk.startEA);
#endif
			}
			else {
#endif // STARS_FIND_UNSHARED_CHUNKS
				this->SharedChunks = true;
#if SMP_DEBUG_CHUNKS
				SMP_msg("INFO: Found tail chunk for %s at %lx\n",
					this->GetFuncName(), (unsigned long) CurrChunk.startEA);
#if STARS_FIND_UNSHARED_CHUNKS 
			}
#endif // STARS_FIND_UNSHARED_CHUNKS
	// Cycle through all chunks that belong to the function.
	ChunkCounter = 0;
	bool GoodRTL;
	this->BuiltRTLs = true;
 	for (bool ChunkOK = FuncTail.main(); ChunkOK; ChunkOK = FuncTail.next()) {
		const area_t &CurrChunk = FuncTail.chunk();
		++ChunkCounter;
clc5q's avatar
clc5q committed
#if 0
		if (CurrChunk.startEA < this->FirstEA) {
			this->FirstEA = CurrChunk.startEA;
		}
#endif
#if STARS_DEBUG_MEMORY_CORRUPTION
	bool DebugFlag = (0 == strcmp("sub_8063BE0", this->GetFuncName()));
clc5q's avatar
clc5q committed
#endif
		// Build the instruction and block lists for the function.
		for (ea_t addr = CurrChunk.startEA; addr < CurrChunk.endEA;
			addr = get_item_end(addr)) {
			flags_t InstrFlags = getFlags(addr);
			if (!isHead(InstrFlags)) {
				continue;
			}
			if (!isCode(InstrFlags)) { //data
				// Check for code xrefs from the data.
				SMP_xref_t xrefs;
				for (bool ok = xrefs.SMP_first_from(addr, XREF_ALL); ok; ok = xrefs.SMP_next_from()) {
					if ((xrefs.GetTo() != 0) && (xrefs.GetIscode())) {
						// Found a code target, with its address in xrefs.to
						PrintDataToCodeXref(addr, xrefs.GetTo(), 0);
					}
				}
			}
			else { // code
				SMPInstr *CurrInst = new SMPInstr(addr);
				// Fill in the instruction data members.
#if SMP_DEBUG_CONTROLFLOW
				SMP_msg("SMPFunction::Analyze: calling CurrInst::Analyze.\n");
				if (SMPBinaryDebug) {
					SMP_msg("Disasm:  %s \n", CurrInst->GetDisasm());
#if SMP_COUNT_MEMORY_ALLOCATIONS
				SMPInstBytes += sizeof(*CurrInst);
#if SMP_USE_SSA_FNOP_MARKER
				if (this->Instrs.empty()) {
					// First instruction in function. We want to create a pseudo-instruction
					//  at the top of the function that can hold SSA DEFs for LiveIn names
					//  to the function. We use a floating point no-op as the pseudo-inst.
					//  The code address is one less than the start address of the function.
					SMPInstr *MarkerInst = new SMPInstr(addr - 1);
					MarkerInst->AnalyzeMarker();
					GoodRTL = MarkerInst->BuildRTL();
					this->BuiltRTLs = (this->BuiltRTLs && GoodRTL);
					if (GoodRTL) {
						MarkerInst->SetGoodRTL();
					}
					assert(FirstInBlock == this->Instrs.end());
					this->Instrs.push_back(MarkerInst);
#if SMP_COUNT_MEMORY_ALLOCATIONS
					SMPInstBytes += sizeof(*MarkerInst);
				// Find all functions that call the current function.
					for (bool ok = CurrXrefs.SMP_first_to(addr, XREF_ALL); ok; ok = CurrXrefs.SMP_next_to()) {
						ea_t FromAddr = CurrXrefs.GetFrom();
						if ((FromAddr != 0) && (CurrXrefs.GetIscode())) {
							// Make sure it is not a fall-through. Must be a
							//  control-flow instruction of some sort, including
							//  direct or indirect calls or tail calls.
							SMPInstr CallInst(FromAddr);
							CallInst.Analyze();
							SMPitype CallType = CallInst.GetDataFlowType();
							if ((COND_BRANCH <= CallType) && (RETURN >= CallType)) {
								// Found a caller, with its call address in CurrXrefs.from
								this->AddCallSource(FromAddr);
				SMPitype DataFlowType = CurrInst->GetDataFlowType();
				if ((DataFlowType == INDIR_CALL) || (DataFlowType == CALL)) {
					// See if IDA has determined the target of the call.
#if 0
					CurrInst->AnalyzeCallInst(this->FirstEA, this->FuncInfo.endEA);
#endif
					ea_t TargetAddr = CurrInst->GetCallTarget();
					bool LinkedToTarget = (BADADDR != TargetAddr);
					if (LinkedToTarget) {
							SMP_msg("WARNING: Ignoring NULL call target (unreachable) at %lx\n", 
								(unsigned long) CurrInst->GetAddr());
							pair<set<ea_t>::iterator, bool> InsertResult;
							if (INDIR_CALL == DataFlowType) {
								InsertResult = this->IndirectCallTargets.insert(TargetAddr);
								InsertResult = this->DirectCallTargets.insert(TargetAddr);
							}
							if (InsertResult.second) {
								this->AllCallTargets.push_back(TargetAddr);
					if (DataFlowType == INDIR_CALL) {
						this->IndirectCalls = true;
						this->UnresolvedIndirectCalls = (!LinkedToTarget);
					}
				} // end if INDIR_CALL or CALL
				else if (DataFlowType == INDIR_JUMP) {
#if STARS_AUDIT_INDIR_JUMP_XREFS
					PreviousIndirJumpAddr = addr;
#endif
				}
				else if (DataFlowType == RETURN) {
					this->HasReturnInst = true;
				}
				// Add call targets for tail call jumps.
				else if (CurrInst->IsBranchToFarChunk()) {
					ea_t FarTargetAddr = CurrInst->GetFarBranchTarget();
					if (BADADDR != FarTargetAddr) {
						assert((RETURN == DataFlowType) || (JUMP == DataFlowType) || (COND_BRANCH == DataFlowType));
						// Optimized tail calls, where the stack frame is down to zero at the call point,
						//  get RETURN as their DataFlowType. Might have to revisit that idea at some point. !!!!****!!!!
						if (this->FindDistantCodeFragment(FarTargetAddr)) {
							if (this->GetProg()->InsertUnsharedFragment(FarTargetAddr)) {
								// Fragment address was inserted in SMPProgram set, was not already there.
								pair<set<ea_t>::iterator, bool> InsertResult;
								InsertResult = FragmentWorkList.insert(FarTargetAddr);
								if (InsertResult.second) {
									SMP_msg("INFO: Found distant code fragment at %lx that can be added to func, reached from %lx\n",
										(unsigned long) FarTargetAddr, (unsigned long) addr);
clc5q's avatar
clc5q committed
#if 0
									if (FarTargetAddr < this->FirstEA) {
										this->FirstEA = FarTargetAddr;
									}
#endif
								}
								else {
									// These kind of fragments are generally only jumped to from one place,
									//  and jump back into the function that jumped into them. Very suspicious
									//  to encounter such a fragment more than once, and even if it happens,
									//  the insertion into the SMPProgram set should have failed due to already
									//  being present. This message and assertion should never be reached.
									SMP_msg("FATAL ERROR: Distant fragment at %lx reached from %lx already reached from same function.\n",
										(unsigned long) FarTargetAddr, (unsigned long) addr);
									assert(InsertResult.second); // sanity lost; shut down
								}
							}
							else { // Fragment address was already in SMPProgram set
								; // Probably added in loop at beginning that found unshared fragments.
#if 0
								// These kind of fragments are generally only jumped to from one place,
								//  and jump back into the function that jumped into them. Very suspicious
								//  to encounter such a fragment more than once.
								SMP_msg("WARNING: Distant fragment at %x reached from %x has already been processed.\n",
									FarTargetAddr, addr);
#endif
							}
						}
						else if (!this->GetProg()->IsUnsharedFragment(FarTargetAddr)) {
							pair<set<ea_t>::iterator, bool> InsertResult;
							InsertResult = this->DirectCallTargets.insert(FarTargetAddr);
							if (InsertResult.second) {
								this->AllCallTargets.push_back(FarTargetAddr);
							}
#if STARS_AUDIT_INDIR_JUMP_XREFS
				if (FirstInBlock == this->Instrs.end()) { // CurrInst will start a block
					if (CurrInst->HasNoCodeXrefs() && (BADADDR != PreviousIndirJumpAddr) && (addr != PreviousIndirJumpAddr)) {
						// This block appears unreachable, but it can probably be reached by
						//  the most recent indirect jump. IDA Pro sometimes thinks it has
						//  resolved an indirect jump completely but has only done so partially.
						SMP_msg("WARNING: Adding possible missing indirect jump code Xref to %lx from %lx\n",
							(unsigned long) addr, (unsigned long) PreviousIndirJumpAddr);
						long SignedAddrDiff = (long) (addr - PreviousIndirJumpAddr);
						if ((SignedAddrDiff < -128) || (SignedAddrDiff > 127)) {
							add_cref(PreviousIndirJumpAddr, addr, FarJump);
						}
						else {
							add_cref(PreviousIndirJumpAddr, addr, NearJump);
						}
						CurrInst->SetJumpTarget();
					}
				}
#endif

				// Before we insert the instruction into the instruction
				//  list, determine if it is a jump target that does not
				//  follow a basic block terminator. This is the special case
				//  of a CASE in a SWITCH that falls through into another
				//  CASE, for example. The first sequence of statements
				//  was not terminated by a C "break;" statement, so it
				//  looks like straight line code, but there is an entry
				//  point at the beginning of the second CASE sequence and
				//  we have to split basic blocks at the entry point.
				if ((FirstInBlock != this->Instrs.end()) && CurrInst->IsJumpTarget()) {
#if SMP_DEBUG_CONTROLFLOW
					SMP_msg("SMPFunction::Analyze: hit special jump target case.\n");
#endif
					LastInBlock = --(this->Instrs.end());
					SMPBasicBlock *NewBlock = new SMPBasicBlock(this, FirstInBlock,	LastInBlock);
					// If not the first chunk in the function, it is a shared
					//  tail chunk.
					if (ChunkCounter > 1) {
						NewBlock->SetShared();
					}
					FirstInBlock = this->Instrs.end();
					LastInBlock = this->Instrs.end();
					this->Blocks.push_back(NewBlock);
				// Build tree RTLs for the instruction.
				GoodRTL = CurrInst->BuildRTL();
				this->BuiltRTLs = (this->BuiltRTLs && GoodRTL);
				if (GoodRTL) {
					CurrInst->SetGoodRTL();
				}
					SMP_msg("ERROR: Cannot build RTL at %lx for %s\n", 
						(unsigned long) CurrInst->GetAddr(), CurrInst->GetDisasm());
#endif
#if SMP_DEBUG_CONTROLFLOW
		SMP_msg("SMPFunction::Analyze: putting CurrInst on list.\n");
#endif
				// Insert instruction at end of list.
				this->Instrs.push_back(CurrInst);

				// Find basic block leaders and terminators.
				if (FirstInBlock == this->Instrs.end()) {
#if SMP_DEBUG_CONTROLFLOW
					SMP_msg("SMPFunction::Analyze: setting FirstInBlock.\n");
#if SMP_USE_SSA_FNOP_MARKER
					if (2 == this->Instrs.size()) {
						// Just pushed first real instruction, after the fnop marker.
						FirstInBlock = this->Instrs.begin();
					}
					else {
						FirstInBlock = --(this->Instrs.end());
					}
#else
					FirstInBlock = --(this->Instrs.end());
				if (CurrInst->IsBasicBlockTerminator()) {
#if SMP_DEBUG_CONTROLFLOW
		SMP_msg("SMPFunction::Analyze: found block terminator.\n");
#endif
					LastInBlock = --(this->Instrs.end());
					SMPBasicBlock *NewBlock = new SMPBasicBlock(this, FirstInBlock, LastInBlock);
					// If not the first chunk in the function, it is a shared
					//  tail chunk.
					if (ChunkCounter > 1) {
						NewBlock->SetShared();
					}
					FirstInBlock = this->Instrs.end();
					LastInBlock = this->Instrs.end();
					this->Blocks.push_back(NewBlock);
			} // end if (isCode(InstrFlags))
		} // end for (ea_t addr = CurrChunk.startEA; ... )

		// Handle the special case in which a function does not terminate
		//  with a return instruction or any other basic block terminator.
		//  Sometimes IDA Pro sees a call to a NORET function and decides
		//  to not include the dead code after it in the function. That
		//  dead code includes the return instruction, so the function no
		//  longer includes a return instruction and terminates with a CALL.
		if (FirstInBlock != this->Instrs.end()) {
			LastInBlock = --(this->Instrs.end());
			SMPBasicBlock *NewBlock = new SMPBasicBlock(this, FirstInBlock, LastInBlock);
			// If not the first chunk in the function, it is a shared
			//  tail chunk.
			if (ChunkCounter > 1) {
				NewBlock->SetShared();
			}
			FirstInBlock = this->Instrs.end();