SMPFunction.cpp

//  we return BADADDR.
ea_t SMPFunction::FindAllocPoint(asize_t OriginalLocSize) {
	sval_t TargetSize = - ((sval_t) OriginalLocSize);  // negate; stack grows down 

#if SMP_DEBUG_FRAMEFIXUP
	bool DebugFlag = (0 == strcmp("_dl_runtime_resolve", this->GetFuncName()));
	if (DebugFlag)
		SMP_msg("%s OriginalLocSize: %d\n", this->GetFuncName(), OriginalLocSize);
#endif

	if (this->AnalyzedSP) {
		// Limit our analysis to the first basic block in the function.
		list<SMPInstr *>::iterator InstIter = this->Instrs.begin();
#if SMP_USE_SSA_FNOP_MARKER
		++InstIter;  // skip marker instruction
#endif
		for ( ; InstIter != this->Instrs.end(); ++InstIter) {
			SMPInstr *CurrInst = (*InstIter);
			ea_t addr = CurrInst->GetAddr();
			// get_spd() returns a cumulative delta of ESP
			sval_t sp_delta = get_spd(this->GetFuncInfo(), addr);
#if SMP_DEBUG_FRAMEFIXUP
			if (DebugFlag)
				SMP_msg("%s delta: %d at %x\n", this->GetFuncName(), sp_delta, addr);
#endif
			if (sp_delta == TargetSize) { // <= instead of == here?  **!!**
				// Previous instruction hit the frame size.
				if (InstIter == this->Instrs.begin()) {
					return BADADDR;  // cannot back up from first instruction
				}
				else {
					ea_t PrevAddr = (*(--InstIter))->GetAddr();
#if SMP_USE_SSA_FNOP_MARKER
					if ((*(this->Instrs.begin()))->GetAddr() == PrevAddr)
						return BADADDR;  // don't return marker instruction
					else
						return PrevAddr;
#else
					return PrevAddr;
#endif
				}
			}
			if (CurrInst->IsLastInBlock()) {
				// It could be that the current instruction will cause the stack pointer
				//  delta to reach the TargetSize. sp_delta is not updated until after the
				//  current instruction, so we need to look ahead one instruction if the
				//  current block falls through. On the other hand, if the current block
				//  ends with a jump or return, we cannot hit TargetSize.
				if (CurrInst->IsBasicBlockTerminator())
					return BADADDR;
				list<SMPInstr *>::iterator NextInstIter = InstIter;
				++NextInstIter;
				if (NextInstIter == this->Instrs.end())
					return BADADDR;
				sp_delta = get_spd(this->GetFuncInfo(), (*NextInstIter)->GetAddr());
				if (sp_delta == TargetSize) {
					// CurrInst will cause stack pointer delta to hit TargetSize.
					return addr;
				}
				else {
					return BADADDR;
				}
			} // end if LastInBlock
		} // end for all instructions
	} // end if (this->AnalyzedSP)
#if SMP_DEBUG_FRAMEFIXUP
	else {
		SMP_msg("AnalyzedSP is false for %s\n", this->GetFuncName());
	}
#endif
	return BADADDR;
} // end of SMPFunction::FindAllocPoint()

// IDA Pro is sometimes confused by a function that uses the frame pointer
//  register for other purposes. For the x86, a function that uses EBP
//  as a frame pointer would begin with: push ebp; mov ebp,esp to save
//  the old value of EBP and give it a new value as a frame pointer. The
//  allocation of local variable space would have to come AFTER the move
//  instruction. A function that begins: push ebp; push esi; sub esp,24
//  is obviously not using EBP as a frame pointer. IDA is apparently
//  confused by the push ebp instruction being the first instruction
//  in the function. We will reset UseFP to false in this case.
// The inverse problem happens with a function that begins with instructions
//  other than push ebp; mov ebp,esp; ... etc. but eventually has those
//  instructions in the first basic block. For example, a C compiler generates
//  for the first block of main():
//    lea ecx,[esp+arg0]
//    and esp, 0xfffffff0
//    push dword ptr [ecx-4]
//    push ebp
//    mov ebp,esp
//    push ecx
//    sub esp,<framesize>
//
//  This function is obviously using EBP as a frame pointer, but IDA Pro marks
//  the function as not using a frame pointer. We will reset UseFP to true in
//  this case.
// NOTE: This logic should work for both Linux and Windows x86 prologues.
bool SMPFunction::MDFixUseFP(void) {
	list<SMPInstr *>::iterator InstIter = this->Instrs.begin();
	ea_t addr = (*InstIter)->GetAddr();

#if SMP_USE_SSA_FNOP_MARKER
	++InstIter;  // skip marker instruction
#endif
	SMPInstr *CurrInst = (*InstIter);

	if (!(this->UseFP)) {
		// See if we can detect the instruction "push ebp" followed by the instruction
		//  "mov ebp,esp" in the first basic block. The instructions do not have to be
		//  consecutive. If we find them, we will reset UseFP to true.
		bool FirstBlockProcessed = false;
		bool EBPSaved = false;
		bool ESPintoEBP = false;
		do {
			FirstBlockProcessed = CurrInst->IsLastInBlock();
			if (!EBPSaved) { // still looking for "push ebp"
				if (CurrInst->MDIsPushInstr() && CurrInst->GetCmd().Operands[0].is_reg(R_bp)) {
					EBPSaved = true;
				}
			}
			else if (!ESPintoEBP) { // found "push ebp", looking for "mov ebp,esp"
				insn_t CurrCmd = CurrInst->GetCmd();
				if ((CurrCmd.itype == NN_mov) 
					&& (CurrInst->GetFirstDef()->GetOp().is_reg(R_bp))
					&& (CurrInst->GetFirstUse()->GetOp().is_reg(R_sp))) {
					ESPintoEBP = true;
					FirstBlockProcessed = true; // exit loop
				}
			}
			++InstIter;
			CurrInst = (*InstIter);
			addr = CurrInst->GetAddr();
			// We must get EBP set to its frame pointer value before we reach the
			//  local frame allocation instruction (i.e. the subtraction of locals space
			//   from the stack pointer).
			FirstBlockProcessed |= (addr >= this->LocalVarsAllocInstr);
		} while (!FirstBlockProcessed);
		// If we found ESPintoEBP, we also found EBPSaved first, and we need to change
		//  this->UseFP to true and return true. Otherwise, return false.
		this->UseFP = ESPintoEBP;
		if (ESPintoEBP)
			SMP_msg("MDFixUseFP reset UseFP to true for %s\n", this->GetFuncName());
		return ESPintoEBP;
	} // end if (!(this->UseFP))

	// At this point, this->UseFP must have been true on entry to this method and we will
	//  check whether it should be reset to false.
	while (addr < this->LocalVarsAllocInstr) {
		set<DefOrUse, LessDefUse>::iterator CurrDef = CurrInst->GetFirstDef();
		while (CurrDef != CurrInst->GetLastDef()) {
			if (CurrDef->GetOp().is_reg(R_bp))
				return false; // EBP got set before locals were allocated
			++CurrDef;
		}
		++InstIter;
		CurrInst = (*InstIter);
		addr = CurrInst->GetAddr();
	}
	// If we found no defs of the frame pointer before the local vars
	//  allocation, then the frame pointer register is not being used
	//  as a frame pointer, just as a general callee-saved register.
	this->UseFP = false;
	SMP_msg("MDFixUseFP reset UseFP to false for %s\n", this->GetFuncName());
	return true;
} // end of SMPFunction::MDFixUseFP()

// Find the callee-saved reg offsets (negative offset from return address)
//  for all registers pushed onto the stack before the stack frame allocation
//  instruction.
void SMPFunction::MDFindSavedRegs(void) {
	list<SMPInstr *>::iterator InstIter;
	int RegIndex;
	func_t *CurrFunc = SMP_get_func(this->GetStartAddr());
	assert(NULL != CurrFunc);

	for (InstIter = this->Instrs.begin(); InstIter != this->Instrs.end(); ++InstIter) {
		SMPInstr *CurrInst = (*InstIter);
		if (CurrInst->GetAddr() > this->LocalVarsAllocInstr)
			break;
		if (!(CurrInst->MDIsPushInstr()))
			continue;

		sval_t CurrOffset = get_spd(CurrFunc, CurrInst->GetAddr());
		if (CurrInst->GetCmd().itype == NN_push) {
			op_t PushedReg = CurrInst->GetPushedOpnd();
			if (o_reg == PushedReg.type) {
				RegIndex = (int) PushedReg.reg;
				if (RegIndex > R_di) {
					SMP_msg("WARNING: Skipping save of register %d\n", RegIndex);
					continue;
				}
				if (this->SavedRegLoc.at((size_t) RegIndex) == 0) {
					this->SavedRegLoc[(size_t) RegIndex] = CurrOffset - 4;
				}
				else {
					SMP_msg("WARNING: Multiple saves of register %d\n", RegIndex);
				}
			} // end if register push operand
		} // end if PUSH instruction
		else if (NN_pusha == CurrInst->GetCmd().itype) {
			// **!!** Handle pushes of all regs.
			this->SavedRegLoc[(size_t) R_ax] = CurrOffset - 4;
			this->SavedRegLoc[(size_t) R_cx] = CurrOffset - 8;
			this->SavedRegLoc[(size_t) R_dx] = CurrOffset - 12;
			this->SavedRegLoc[(size_t) R_bx] = CurrOffset - 16;
			this->SavedRegLoc[(size_t) R_sp] = CurrOffset - 20;
			this->SavedRegLoc[(size_t) R_bp] = CurrOffset - 24;
			this->SavedRegLoc[(size_t) R_si] = CurrOffset - 28;
			this->SavedRegLoc[(size_t) R_di] = CurrOffset - 32;
			break; // all regs accounted for
		}
		else if (CurrInst->MDIsEnterInstr()) {
			this->SavedRegLoc[(size_t) R_bp] = CurrOffset - 4;
		}
	} // end for all instructions

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

// Compute the ReturnRegTypes[] as the meet over all register types
//  at all return instructions.
void SMPFunction::MDFindReturnTypes(void) {
	list<SMPBasicBlock *>::iterator BlockIter;
	SMPBasicBlock *CurrBlock;
	list<SMPInstr *>::iterator InstIter;
	vector<SMPOperandType> RegTypes;
	SMPInstr *CurrInst;

	for (BlockIter = this->Blocks.begin(); BlockIter != this->Blocks.end(); ++BlockIter) {
		CurrBlock = (*BlockIter);
		if (CurrBlock->HasReturn()) {
			// Get the types of all registers at the RETURN point.
			//  Calculate the meet function over them.
			InstIter = CurrBlock->GetLastInstr();
			--InstIter;
			CurrInst = (*InstIter);
			assert(RETURN == CurrInst->GetDataFlowType());
			set<DefOrUse, LessDefUse>::iterator CurrUse;
			for (CurrUse = CurrInst->GetFirstUse();
				CurrUse != CurrInst->GetLastUse();
				++CurrUse) {

				op_t UseOp = CurrUse->GetOp();
				if ((o_reg != UseOp.type) || (R_di < UseOp.reg))
					continue;
				this->ReturnRegTypes[UseOp.reg]
					= SMPTypeMeet(this->ReturnRegTypes.at(UseOp.reg),
						CurrUse->GetType());
			} // for all USEs in the RETURN instruction
		} // end if current block has a RETURN
	} // end for all blocks
	return;
} // end of SMPFunction::MDFindReturnTypes()

// Determine local variable boundaries in the stack frame.
void SMPFunction::BuildLocalVarTable(void) {
	// Currently we just use the info that IDA Pro has inferred from the direct
	//  addressing of stack locations.
	this->SemiNaiveLocalVarID();
	return;
} // end of SMPFunction::BuildLocalVarTable()

// Limit damage from garbage stack offset values produced by IDA Pro.
#define IDAPRO_KLUDGE_STACK_FRAME_SIZE_LIMIT 5000000

// Use the local variable offset list from IDA's stack frame structure to compute
//  the table of local variable boundaries.
void SMPFunction::SemiNaiveLocalVarID(void) {
	// NOTE: We use IDA Pro's offsets from this->FuncInfo (e.g. frsize) and NOT
	//  our own corrected values in our private data members. The offsets we
	//  read from the stack frame structure returned by get_frame() are consistent
	//  with other IDA Pro values, not with our corrected values.
	list<SMPInstr *>::iterator InstIter;
	bool DebugFlag = false;
	this->LocalVarOffsetLimit = -20000;
#if SMP_DEBUG_STACK_GRANULARITY
	DebugFlag |= (0 == strcmp("qSort3", this->GetFuncName()));
#endif
	func_t *FuncPtr = SMP_get_func(this->FuncInfo.startEA);
	if (NULL == FuncPtr) {
		SMP_msg("ERROR in SMPFunction::SemiNaiveLocalVarID; no func ptr\n");
	}
	assert(NULL != FuncPtr);
	struc_t *StackFrame = get_frame(FuncPtr);

	if (NULL == StackFrame) {
		SMP_msg("WARNING: No stack frame info from get_frame for %s\n", this->GetFuncName());
		return;
	}

	member_t *Member = StackFrame->members;
	for (size_t i = 0; i < StackFrame->memqty; ++i, ++Member) {
		long offset;
		char MemberName[MAXSMPVARSTR] = {'\0'};
		if (NULL == Member) {
			SMP_msg("NULL stack frame member pointer in %s\n", this->GetFuncName());
			break;
		}
		get_member_name(Member->id, MemberName, MAXSMPVARSTR - 1);
		if (MemberName == NULL) {
#if SMP_DEBUG_STACK_GRANULARITY
			SMP_msg("NULL stack frame member in %s\n", this->GetFuncName());
#endif
			continue;
		}
		if (Member->unimem()) {
			// Not a separate variable; name for member of a union.
			// The union itself should have a separate entry, so we skip this.
			SMP_msg("STACK INFO: Skipping union member %s frame member %zu in stack frame for %s\n",
				MemberName, i, this->GetFuncName());
			continue;
		}
		offset = (long) Member->get_soff(); // Would be 0 for union member, so we skipped them above.
		if (DebugFlag) {
			SMP_msg("%s local var %s at offset %ld\n", this->GetFuncName(), MemberName, offset);
		}
		if (offset > IDAPRO_KLUDGE_STACK_FRAME_SIZE_LIMIT) {
			SMP_msg("ERROR: Rejected enormous stack offset %ld for var %s in func %s\n", offset, MemberName, this->GetFuncName());
			continue;
		}
		struct LocalVar TempLocal;
		TempLocal.offset = offset;
		TempLocal.size = Member->eoff - Member->soff; // audit later
		SMP_strncpy(TempLocal.VarName, MemberName, sizeof(TempLocal.VarName) - 1);
		this->LocalVarTable.push_back(TempLocal);
		if ((offset + (long) TempLocal.size) >= this->LocalVarOffsetLimit) {
			this->LocalVarOffsetLimit = (long) (TempLocal.offset + TempLocal.size);
		}
	} // end for all stack frame members

	// If AnalyzedSP is false, that is all we can do.
	if (!this->AnalyzedSP) {
		// No allocations; sometimes happens in library functions.
		this->OutgoingArgsSize = 0;
		this->MinStackDelta = 0;
		this->AllocPointDelta = 0;
		return;
	}

	// Calculate min and max stack point deltas.
	this->MinStackDelta = 20000; // Final value should be negative or zero
	this->MaxStackDelta = -1000; // Final value should be zero.
	InstIter = this->Instrs.begin();
#if SMP_USE_SSA_FNOP_MARKER
	if ((*InstIter)->IsFloatNop())
		++InstIter;  // skip marker instruction
#endif
	for ( ; InstIter != this->Instrs.end(); ++InstIter) {
		SMPInstr *CurrInst = (*InstIter);
		ea_t addr = CurrInst->GetAddr();
		sval_t sp_delta = get_spd(this->GetFuncInfo(), addr);
		if (sp_delta < this->MinStackDelta)
			this->MinStackDelta = sp_delta;
		if (sp_delta > this->MaxStackDelta)
			this->MaxStackDelta = sp_delta;
		if (addr == this->LocalVarsAllocInstr) {
			// Total stack pointer delta is sp_delta for the next instruction,
			//  because IDA updates the sp delta AFTER each instruction.
			list<SMPInstr *>::iterator NextInstIter = InstIter;
			++NextInstIter;
			sp_delta = get_spd(this->GetFuncInfo(), (*NextInstIter)->GetAddr());
			this->AllocPointDelta = sp_delta;
		}
	}

	// IDA Pro sometimes fails to add stack frame members for all incoming args, etc.
	//  Find and correct these omissions by examining stack accesses in instructions
	//  and extend the LocalVarTable to cover whatever is out of range.
	if (!this->AuditLocalVarTable()) {
		// Catastrophic error must have occurred, probably due to errors in IDA's
		//  stack pointer analysis, despite AnalyzedSP being true.
		return;
	}

	if (!(this->LocalVarTable.empty())) {
		this->GoodLocalVarTable = true;

		// Sort the LocalVarTable so that we do not depend on IDA Pro
		//  presenting the stack frame members in order.
		std::sort(this->LocalVarTable.begin(), this->LocalVarTable.end(), LocalVarCompare);
	}

#if SMP_DEBUG_STACK_GRANULARITY
	SMP_msg("Computing %d local var sizes\n", this->LocalVarTable.size());
#endif
	// Now we want to audit the size field for each local
	if (this->GoodLocalVarTable) {
		size_t VarLimit = this->LocalVarTable.size() - 1;
		assert(this->LocalVarTable.size() > 0);
		for (size_t VarIndex = 0; VarIndex < VarLimit; ++VarIndex) {
			struct LocalVar TempLocEntry = this->LocalVarTable[VarIndex];
			bool AboveLocalsRegion = (TempLocEntry.offset >= this->LocalVarsSize);
			size_t TempSize = this->LocalVarTable[VarIndex + 1].offset
				- TempLocEntry.offset;
			int DiffSize = ((int) TempSize) - ((int) TempLocEntry.size);
			// We don't have IDA Pro stack frame members for callee saved registers. This
			//  omission can make it seem that there is a gap between the uppermost local
			//  variable and the return address or saved frame pointer. Avoid expanding the
			//  last local variable into the callee saved registers region.
			if (DiffSize > 0) { // We are expanding the size.
				if (!AboveLocalsRegion && ((TempLocEntry.offset + TempLocEntry.size + DiffSize) > this->LocalVarsSize)) {
					// Current local does not start above the locals region, but its new size will
					//  carry it into the locals region.
					if ((TempLocEntry.offset + TempLocEntry.size) > this->LocalVarsSize) {
						// Weird. It already overlapped the callee saved regs region.
						SMP_msg("WARNING: Local var at offset %ld size %zu in %s extends above local vars region.\n",
							TempLocEntry.offset, TempLocEntry.size, this->GetFuncName());
					}
					// Limit DiffSize to avoid overlapping callee saved regs.
					DiffSize = this->LocalVarsSize - (TempLocEntry.offset + TempLocEntry.size);
					if (DiffSize < 0)
						DiffSize = 0; // started out positive, cap it at zero.
				}
			}
			if (DiffSize < 0)
				DiffSize = 0; // should not happen with sorted LocalVarTable unless duplicate entries.
			if (DiffSize != 0) {
#if SMP_DEBUG_STACK_GRANULARITY
				SMP_msg("STACK INFO: Adjusted size for stack frame member at %ld in %s\n",
					TempLocEntry.offset, this->GetFuncName());
#endif
				this->LocalVarTable[VarIndex].size += DiffSize;
			}
		}
#if 0 // Using Member->eoff seems to be working for all members, including the last one.
#if SMP_DEBUG_STACK_GRANULARITY
		SMP_msg("Computing last local var size for frsize %d\n", this->FuncInfo.frsize);
#endif
		// Size of last local is total frsize minus savedregs in frame minus offset of last local
		size_t SavedRegsSpace = 0; // portion of frsize that is saved regs, not locals.
		if (this->CalleeSavedRegsSize > this->FuncInfo.frregs) {
			// IDA Pro counts the save of EBP in frregs, but then EBP gets its new
			//  value and callee saved regs other than the old EBP push get counted
			//  in frsize rather than frregs. CalleeSavedRegsSize includes all saved
			//  regs on the stack, both above and below the current EBP offset.
			// NOTE: For windows, this has to be done differently, as callee saved regs
			//  happen at the bottom of the local frame, not the top.
#if 0
			SavedRegsSpace = this->CalleeSavedRegsSize - this->FuncInfo.frregs;
#else
			SavedRegsSpace = this->FuncInfo.frsize - this->LocalVarsSize;
#endif
		}

		this->LocalVarTable.back().size = this->FuncInfo.frsize
			- SavedRegsSpace - this->LocalVarTable.back().offset;
		this->LocalVarOffsetLimit = this->LocalVarTable.back().offset 
			+ (adiff_t) this->LocalVarTable.back().size;
#endif
	}

#if 0 // AboveLocalsSize is not a reliable number.
	// IDA Pro can have difficulty with some irregular functions such as are found
	//  in the C startup code. The frsize value might be bogus. Just punt on the
	//  local variable ID if that is the case.
	if ((this->LocalVarOffsetLimit - AboveLocalsSize) > (adiff_t) this->FuncInfo.frsize) {
		this->LocalVarTable.clear();
		this->GoodLocalVarTable = false;
		SMP_msg("WARNING: Bad frsize %d for %s OffsetLimit: %d AboveLocalsSize: %d LocalVarsSize: %d ; abandoning SemiNaiveLocalVarID.\n",
			this->FuncInfo.frsize, this->GetFuncName(), this->LocalVarOffsetLimit, AboveLocalsSize, this->LocalVarsSize);
		return;
	}
	assert((this->LocalVarOffsetLimit - AboveLocalsSize) <= (adiff_t) this->FuncInfo.frsize);
#endif

	// Find out how many of the locals are really outgoing args.
	if (this->AnalyzedSP && !this->CallsAlloca && (BADADDR != this->LocalVarsAllocInstr)) {
		this->FindOutgoingArgsSize();
	}
	else {
		SMP_msg("FindOutgoingArgsSize not called for %s ", this->GetFuncName());
		SMP_msg("AnalyzedSP: %d CallsAlloca: %d LocalVarsAllocInstr: %x \n",
			this->AnalyzedSP, this->CallsAlloca, this->LocalVarsAllocInstr);
	}
	return;
} // end of SMPFunction::SemiNaiveLocalVarID()

// Check and correct the LocalVarTable derived from IDA Pro stack frame members.
//  Examine each instruction and see if any stack accesses are beyond the LocalVarTable
//  and create new entries in the LocalVarTable if so.
bool SMPFunction::AuditLocalVarTable(void) {
	list<SMPInstr *>::iterator InstIter;
	int SignedOffset;

	// We cannot depend on IDA Pro making Member
	//  entries for everything that is accessed on the stack.
	//  When an incoming arg is accessed but no Member is
	//  created, then LocalVarOffsetLimit will be too small
	//  and we will get ERROR messages. Just loop through the
	//  instructions, find offsets higher than the LocalVarTable
	//  currently holds, and add new entries to LocalVarTable to
	//  handle them.
	// Iterate through all instructions and record stack frame accesses in the StackFrameMap.
	InstIter = this->Instrs.begin();
#if SMP_USE_SSA_FNOP_MARKER
	if ((*InstIter)->IsFloatNop())
		++InstIter;  // skip marker instruction
#endif
	for ( ; InstIter != this->Instrs.end(); ++InstIter) {
		SMPInstr *CurrInst = (*InstIter);
		ea_t InstAddr = CurrInst->GetAddr();
		sval_t sp_delta = get_spd(this->GetFuncInfo(), InstAddr);
		if (0 < sp_delta) {
			// Stack underflow; about to assert
			SMP_msg("Stack underflow at %x %s sp_delta: %d\n", CurrInst->GetAddr(),
				CurrInst->GetDisasm(), sp_delta);
			return false;
		}
		assert(0 >= sp_delta);
		ea_t offset;
		size_t DataSize;
		bool UsedFramePointer;
		bool IndexedAccess;
		bool SignedMove;
		bool UnsignedMove;
		if (CurrInst->HasDestMemoryOperand()) {
			// NOTE: We need to catch stack pushes here also (callee-saved regs). !!!!!*******!!!!!!!!
			set<DefOrUse, LessDefUse>::iterator CurrDef;
			for (CurrDef = CurrInst->GetFirstDef(); CurrDef != CurrInst->GetLastDef(); ++CurrDef) {
				op_t TempOp = CurrDef->GetOp();
				if (TempOp.type != o_phrase && TempOp.type != o_displ)
					continue;
				if (this->MDGetStackOffsetAndSize(CurrInst, TempOp, sp_delta, offset, DataSize, UsedFramePointer,
					IndexedAccess, SignedMove, UnsignedMove)) {
					SignedOffset = (int) offset;
					if (IndexedAccess && ((0 > SignedOffset) || ((offset + DataSize) > this->StackFrameMap.size()))) {
						continue; // Indexed expressions can be within frame but offset is outside frame
					}
#if 0 // ls_O3.exe has IDA trouble on chunked function get_funky_string().
					assert(0 <= SignedOffset);
#else
					if (0 > SignedOffset) { // negative offset but not Indexed; very bad
						SMP_msg("ERROR: Negative stack offset at %x in %s. Abandoning LocalVar ID.\n", CurrInst->GetAddr(), this->GetFuncName());
						return false;
					}
#endif
					if ((SignedOffset + (long) DataSize) > this->LocalVarOffsetLimit) {
						// Going out of range. Extend LocalVarTable.
						struct LocalVar TempLocal;
						char TempStr[20];
						TempLocal.offset = (long) SignedOffset;
						TempLocal.size = DataSize;
						SMP_strncpy(TempLocal.VarName, "SMP_InArg", sizeof(TempLocal.VarName) - 1);
						(void) SMP_snprintf(TempStr, 18, "%d", offset);
						SMP_strncat(TempLocal.VarName, TempStr, sizeof(TempLocal.VarName) - 1);
						this->LocalVarTable.push_back(TempLocal);
						this->LocalVarOffsetLimit = (long) (SignedOffset + (long) DataSize);
					}
				}
			}
		}
		if (CurrInst->HasSourceMemoryOperand()) {
			set<DefOrUse, LessDefUse>::iterator CurrUse;
			for (CurrUse = CurrInst->GetFirstUse(); CurrUse != CurrInst->GetLastUse(); ++CurrUse) {
				op_t TempOp = CurrUse->GetOp();
				if ((TempOp.type != o_phrase) && (TempOp.type != o_displ))
					continue;
				if (this->MDGetStackOffsetAndSize(CurrInst, TempOp, sp_delta, offset, DataSize, UsedFramePointer,
					IndexedAccess, SignedMove, UnsignedMove)) {
					SignedOffset = (int) offset;
					if (IndexedAccess && ((0 > SignedOffset) || ((offset + DataSize) > this->StackFrameMap.size()))) {
						continue; // Indexed expressions can be within frame but offset is outside frame
					}
					assert(0 <= SignedOffset);
					if ((SignedOffset + (long) DataSize) > this->LocalVarOffsetLimit) {
						// Going out of range. Extend LocalVarTable.
						struct LocalVar TempLocal;
						char TempStr[20];
						TempLocal.offset = (long) SignedOffset;
						TempLocal.size = DataSize;
						SMP_strncpy(TempLocal.VarName, "SMP_InArg", sizeof(TempLocal.VarName) - 1);
						(void) SMP_snprintf(TempStr, 18, "%d", offset);
						SMP_strncat(TempLocal.VarName, TempStr, sizeof(TempLocal.VarName) - 1);
						this->LocalVarTable.push_back(TempLocal);
						this->LocalVarOffsetLimit = (long) (SignedOffset + (long) DataSize);
					}
				}
			}
		}
	} // end for all instructions

	// Fill in the gaps with new variables as well. SHOULD WE? WHY?

	return true;
} // end of SMPFunction::AuditLocalVarTable()


// Determine how many bytes at the bottom of the stack frame (i.e. at bottom of
//  this->LocalVarsSize) are used for outgoing args. This is the case when the cdecl
//  calling convention is used, e.g. gcc/linux allocates local var space + out args space
//  in a single allocation and then writes outarg values directly to ESP+0, ESP+4, etc.
void SMPFunction::FindOutgoingArgsSize(void) {
	// Compute the lowest value reached by the stack pointer.
	list<SMPInstr *>::iterator InstIter;
	unsigned short BitWidthMask;
	bool DebugFlag = false;
	int SignedOffset;
#if SMP_DEBUG_STACK_GRANULARITY
	DebugFlag = (0 == strcmp("BZ2_blockSort", this->GetFuncName()));
#endif

	this->OutgoingArgsComputed = true;

	if (DebugFlag) {
		SMP_msg("DEBUG: Entered FindOutgoingArgsSize for %s\n", this->GetFuncName());
#if SMP_IDAPRO52_WORKAROUND
		this->OutgoingArgsSize = 16;
		return;
#endif
	}

#if SMP_DEBUG_STACK_GRANULARITY
	SMP_msg("AllocPointDelta: %d MinStackDelta: %d\n", this->AllocPointDelta, this->MinStackDelta);
#endif
	if ((0 <= this->MinStackDelta) || (0 <= this->AllocPointDelta)) {
		// No allocations; sometimes happens in library functions.
		this->OutgoingArgsSize = 0;
		this->MinStackDelta = 0;
		this->AllocPointDelta = 0;
		return;
	}
	assert(0 > this->MinStackDelta);

	// Allocate a vector of stack frame entries, one for each byte of the stack frame.
	//  This will be our memory map for analyzing stack usage.
	int limit = 0;
#if 1
	if (this->LocalVarOffsetLimit > 0) {
		if (limit < (this->LocalVarOffsetLimit + this->MinStackDelta)) {
			// Make room for incoming args, other stuff above local vars.
			limit = this->LocalVarOffsetLimit + this->MinStackDelta;
			if (this->MinStackDelta < this->AllocPointDelta) {
				// Also have stuff below alloc point to make room for.
				limit += (this->AllocPointDelta - this->MinStackDelta);
			}
		}
	}
#endif
	for (int i = this->MinStackDelta; i < limit; ++i) {
		struct StackFrameEntry TempEntry;
		TempEntry.VarPtr = NULL;
		TempEntry.offset = (long) i;
		TempEntry.Read = false;
		TempEntry.Written = false;
		TempEntry.AddressTaken = false;
		TempEntry.ESPRelativeAccess = false;
		TempEntry.EBPRelativeAccess = false;
		TempEntry.IndexedAccess = false;
		this->StackFrameMap.push_back(TempEntry);
		struct FineGrainedInfo TempFineGrained;
		TempFineGrained.SignMiscInfo = 0;
		TempFineGrained.SizeInfo = 0;
		this->FineGrainedStackTable.push_back(TempFineGrained);
	}
#if 0
	for (int i = 0; i < this->LocalVarOffsetLimit; ++i) {
		struct FineGrainedInfo TempFineGrained;
		TempFineGrained.SignMiscInfo = 0;
		TempFineGrained.SizeInfo = 0;
		this->FineGrainedStackTable.push_back(TempFineGrained);
	}
#endif

	// Fill in the VarPtr fields for each StackFrameMap entry.
	if (0 <= this->AllocPointDelta) {
		SMP_msg("FATAL ERROR: AllocPointDelta = %d in %s\n", this->AllocPointDelta, this->GetFuncName());
	}
	assert(0 > this->AllocPointDelta);
	for (size_t i = 0; i < this->LocalVarTable.size(); ++i) {
		assert(this->LocalVarTable.at(i).offset >= 0);
		// Picture that AllocPointDelta is -200, MinStackDelta is -210, and
		//  the LocalVarTable[i].offset is +8 (i.e. 8 bytes above alloc point).
		//  Then base = 8 + (-200 - -210) = 8 + 10 = 18, the proper offset into
		//  the StackFrameMap.
		size_t base = (size_t) (this->LocalVarTable.at(i).offset
			+ (this->AllocPointDelta - this->MinStackDelta));
		size_t limit = base + this->LocalVarTable.at(i).size;
		if (limit > this->StackFrameMap.size()) {
			SMP_msg("ERROR: FindOutArgsSize: base = %zu limit = %zu StackFrameMap size = %zu in %s\n",
				base, limit, this->StackFrameMap.size(), this->GetFuncName());
			this->OutgoingArgsComputed = false;
			this->OutgoingArgsSize = 0;
			return;
		}
		assert(limit <= this->StackFrameMap.size());
		for (size_t MapIndex = base; MapIndex < limit; ++MapIndex) {
			this->StackFrameMap[MapIndex].VarPtr = &(this->LocalVarTable.at(i));
		}
	}

	// Iterate through all instructions and record stack frame accesses in the StackFrameMap.
	InstIter = this->Instrs.begin();
#if SMP_USE_SSA_FNOP_MARKER
	if ((*InstIter)->IsFloatNop())
		++InstIter;  // skip marker instruction
#endif
	for ( ; InstIter != this->Instrs.end(); ++InstIter) {
		SMPInstr *CurrInst = (*InstIter);
		ea_t InstAddr = CurrInst->GetAddr();
		sval_t sp_delta = get_spd(this->GetFuncInfo(), InstAddr);
		if (0 < sp_delta) {
			// Stack underflow; about to assert
			SMP_msg("FATAL ERROR: Stack underflow at %x %s sp_delta: %d\n", CurrInst->GetAddr(),
				CurrInst->GetDisasm(), sp_delta);
		}
		assert(0 >= sp_delta);
		ea_t offset;
		size_t DataSize;
		bool UsedFramePointer;
		bool IndexedAccess;
		bool SignedMove;
		bool UnsignedMove;
		if (CurrInst->HasDestMemoryOperand()) {
			set<DefOrUse, LessDefUse>::iterator CurrDef;
			for (CurrDef = CurrInst->GetFirstDef(); CurrDef != CurrInst->GetLastDef(); ++CurrDef) {
				op_t TempOp = CurrDef->GetOp();
				if (TempOp.type != o_phrase && TempOp.type != o_displ)
					continue;
				if (this->MDGetStackOffsetAndSize(CurrInst, TempOp, sp_delta, offset, DataSize, UsedFramePointer,
					IndexedAccess, SignedMove, UnsignedMove)) {
					SignedOffset = (int) offset;
					if (IndexedAccess && ((0 > SignedOffset) || ((offset + DataSize) > this->StackFrameMap.size()))) {
						continue; // Indexed expressions can be within frame even when offset is outside frame
					}
					assert(0 <= SignedOffset);
#if 0
					if (offset >= this->FuncInfo.frsize)
						continue;  // limit processing to outgoing args and locals
#endif
					if ((offset + DataSize) > this->StackFrameMap.size()) {
						SMP_msg("ERROR: offset = %u DataSize = %zu FrameMapSize = %zu\n",
							offset, DataSize, this->StackFrameMap.size());
						continue;
					}
					assert((offset + DataSize) <= this->StackFrameMap.size());
					for (int j = 0; j < (int) DataSize; ++j) {
						this->StackFrameMap[offset + j].Written = true;
						this->StackFrameMap[offset + j].IndexedAccess = IndexedAccess;
						if (!UsedFramePointer) {
							this->StackFrameMap[offset + j].ESPRelativeAccess = true;
						}
						else {
							this->StackFrameMap[offset + j].EBPRelativeAccess = true;
						}
					}
					BitWidthMask = ComputeOperandBitWidthMask(TempOp, DataSize);
					this->FineGrainedStackTable.at(offset).SizeInfo |= BitWidthMask;
					this->FineGrainedStackTable.at(offset).SignMiscInfo |= FG_MASK_WRITTEN;
					if (IndexedAccess) {
						this->FineGrainedStackTable.at(offset).SignMiscInfo |= FG_MASK_INDEXED_ACCESS;
					}
					if (!UsedFramePointer) {
						this->FineGrainedStackTable.at(offset).SignMiscInfo |= FG_MASK_SP_RELATIVE;
					}
					else {
						this->FineGrainedStackTable.at(offset).SignMiscInfo |= FG_MASK_FP_RELATIVE;
					}
					// We will process the signedness of stores later, so that loads can take precedence
					//  over stores in determining signedness.
				} // end if MDGetStackOffsetAndSize()
			} // end for all DEFs
		} // end if DestMemoryOperand

		if (CurrInst->HasSourceMemoryOperand()) {
			set<DefOrUse, LessDefUse>::iterator CurrUse;
			for (CurrUse = CurrInst->GetFirstUse(); CurrUse != CurrInst->GetLastUse(); ++CurrUse) {
				op_t TempOp = CurrUse->GetOp();
				if (TempOp.type != o_phrase && TempOp.type != o_displ)
					continue;
				if (this->MDGetStackOffsetAndSize(CurrInst, TempOp, sp_delta, offset, DataSize, UsedFramePointer,
					IndexedAccess, SignedMove, UnsignedMove)) {
					SignedOffset = (int) offset;
					if (IndexedAccess && ((0 > SignedOffset) || ((SignedOffset + DataSize) > this->StackFrameMap.size()))) {
						continue; // Indexed expressions can be within frame but offset is outside frame
					}
					assert(0 <= SignedOffset);
#if 0
					if (offset >= this->FuncInfo.frsize)
						continue;  // limit processing to outgoing args and locals
#endif
					if ((SignedOffset + DataSize) > this->StackFrameMap.size()) {
						SMP_msg("ERROR: offset = %u DataSize = %zu FrameMapSize = %zu\n",
							offset, DataSize, this->StackFrameMap.size());
						continue;
					}
					assert((SignedOffset + DataSize) <= this->StackFrameMap.size());
					for (int j = 0; j < (int) DataSize; ++j) {
						this->StackFrameMap[offset + j].Read = true;
						this->StackFrameMap[offset + j].IndexedAccess |= IndexedAccess;
						if (!UsedFramePointer)
							this->StackFrameMap[offset + j].ESPRelativeAccess = true;
						else
							this->StackFrameMap[offset + j].EBPRelativeAccess = true;
					}
					BitWidthMask = ComputeOperandBitWidthMask(TempOp, DataSize);
					this->FineGrainedStackTable.at(offset).SizeInfo |= BitWidthMask;
					this->FineGrainedStackTable.at(offset).SignMiscInfo |= FG_MASK_READ;
					if (IndexedAccess) {
						this->FineGrainedStackTable.at(offset).SignMiscInfo |= FG_MASK_INDEXED_ACCESS;
					}
					if (!UsedFramePointer) {
						this->FineGrainedStackTable.at(offset).SignMiscInfo |= FG_MASK_SP_RELATIVE;
					}
					else {
						this->FineGrainedStackTable.at(offset).SignMiscInfo |= FG_MASK_FP_RELATIVE;
					}
					if (SignedMove) {
						this->FineGrainedStackTable.at(offset).SignMiscInfo |= FG_MASK_SIGNED;
					}
					else if (UnsignedMove) {
						this->FineGrainedStackTable.at(offset).SignMiscInfo |= FG_MASK_UNSIGNED;
					}
				} // end if MDGetStackOffsetAndSize()
			} // end for all USEs
		} // end if SourceMemoryOperand
		// NOTE: Detect taking the address of stack locations. **!!**
	} // end for all instructions

	// If function is a leaf function, set OutgoingArgsSize to zero and return.
	if (this->IsLeaf()) {
		this->OutgoingArgsSize = 0;
		return;
	}

	// For non-leaf functions, set the OutgoingArgsSize to the write-only, ESP-relative
	//  region of the bottom of the StackFrameMap.
	bool OutgoingArgsRegionFinished = false;
	bool IndexedOutgoingArgs = false; // Any indexed accesses to outgoing args?
	size_t FramePadSize = 0;
	for (size_t MapIndex = 0; MapIndex < this->StackFrameMap.size(); ++MapIndex) {
		// Some of the bottom of the stack frame might be below the local frame allocation.
		//  These are pushes that happened after allocation, etc. We skip over these
		//  locations and define the outgoing args region to start strictly at the bottom
		//  of the local frame allocation.
		struct StackFrameEntry TempEntry = this->StackFrameMap.at(MapIndex);
		if (DebugFlag) {
			SMP_msg("StackFrameMap entry %zu: offset: %ld Read: %d Written: %d ESP: %d EBP: %d\n",
				MapIndex, TempEntry.offset, TempEntry.Read, TempEntry.Written,
				TempEntry.ESPRelativeAccess, TempEntry.EBPRelativeAccess);
		}
		if (TempEntry.offset < this->AllocPointDelta)
			continue;
		if (OutgoingArgsRegionFinished) {
			// We are just processing the stack frame padding.
			if (!TempEntry.Read && !TempEntry.Written) {
				// Could be stack frame padding.
				++FramePadSize;
			}
			else {
				break; // No more padding region
			}
		}
		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;
			}
		}
	}

	// 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;
	}

#if 0
	// 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)) {
#if SMP_DEBUG_FRAMEFIXUP
			SMP_msg("OutGoingArgsSize = %d", this->OutgoingArgsSize);
#endif
			this->OutgoingArgsSize = TempEntry.VarPtr->offset + TempEntry.VarPtr->size;
#if SMP_DEBUG_FRAMEFIXUP
			SMP_msg(" adjusted to %d\n", this->OutgoingArgsSize);
#endif
		}
	}
#endif

	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: TempOp must be of type o_displ or o_phrase, as no other operand type could be a
//  stack memory access.
// sp_delta is the stack pointer delta of the current instruction, relative to the initial
//  stack pointer value for the function.
// Return true if a stack memory access was found in TempOp, false otherwise.
bool SMPFunction::MDGetStackOffsetAndSize(SMPInstr *Instr, op_t TempOp, sval_t sp_delta, ea_t &offset, size_t &DataSize, bool &FP,
										  bool &Indexed, bool &Signed, bool &Unsigned) {
	int BaseReg;
	int IndexReg;
	ushort ScaleFactor;
	int SignedOffset;

	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
		SignedOffset += sp_delta; // base offsets from entry ESP value
		SignedOffset -= this->MinStackDelta; // convert to StackFrameMap index
		offset = (ea_t) SignedOffset;
		// 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)) {
			// Consider asserting here.
			SMP_msg("ERROR: Negative offset in MDGetStackOffsetAndSize for inst dump: \n");
			Instr->Dump();
		}
		return true;
	}
	else if (this->UseFP && ((BaseReg == R_bp) || (IndexReg == R_bp))) {
		SignedOffset -= this->FuncInfo.frregs; // base offsets from entry ESP value
		SignedOffset -= this->MinStackDelta; // convert to StackFrameMap index
		offset = (ea_t) SignedOffset;
		DataSize = GetOpDataSize(TempOp);
		FP = true;
		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)) {
			// Consider asserting here.
			SMP_msg("ERROR: Negative offset in MDGetStackOffsetAndSize for inst dump: \n");
			Instr->Dump();
		}
		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);
	sval_t sp_delta = get_spd(this->GetFuncInfo(), InstAddr);

	SignedOffset = (int) offset;

	if (TempOp.type == o_phrase) {
		assert(SignedOffset == 0);  // implicit zero, as in [esp] ==> [esp+0]
	}
	if ((BaseReg == R_sp) || (IndexReg == R_sp)) {
		// ESP-relative constant offset
		SignedOffset += sp_delta; // base offsets from entry ESP value
		SignedOffset -= this->MinStackDelta; // convert to StackFrameMap index