SMPStaticAnalyzer.cpp

		index = (min + max) / 2;
	}

	// IDAProLocs[index] contains addr.
	vector<ea_t>::iterator RemovalIterator = IDAProLocs.begin();
	RemovalIterator += index;
	RemovalIterator = IDAProLocs.erase(RemovalIterator);
	return;
} // end of RemoveIDACodeAddr()

// Driver for all other fixing functions. Upon its return, the IDA
//  database (IDB file) should be fixed up as much as we can fix it.
void FixupIDB(void) {
	FindCodeAddresses();
#if SMP_DEBUG_FIXUP_IDB
	SpecialDebugOutput();
#endif
	AuditCodeTargets();
	FindDataInCode();
	AuditTailChunkOwnership();
	if (DisasmLocs.size() > 0) {
		FixCodeIdentification();
		int fixes = FixupNewCodeChunks();
#if SMP_DEBUG_FIXUP_IDB
		SpecialDebugOutput();
#endif
	}
	DisasmLocs.clear();
	IDAProLocs.clear();
} // end of FixupIDB()

// Find and print all data head addresses in code segments. 
// If an isolated code instruction is found in the midst of a run
//  of data bytes and has no code xrefs jumping to it, it is not
//  reachable as code and is undoubtedly a mixup by IDA. Possibly
//  the whole data region will be converted to code later, in which
//  case the isolated code is not necessarily properly aligned and
//  parsed at its present address, so we are glad to convert it into
//  data anyway so that FindDataToConvert() will succeed on it later.
// Data to code conversion, and isolated code detection, are inhibited
//  by IDA identifying several consecutive instructions in the midst
//  of a data region, with the code addresses not agreeing with the
//  external disassembler's code addresses. We will convert these
//  misaligned instructions to data as we detect them. We will also
//  convert unexplored bytes (isUnknown(flags) == true) into data if
//  they are in the midst of a data sequence.
#define MIN_DATARUN_LEN 24  // #bytes on either side of "isolated" code
void FindDataInCode(void) {
	size_t DataRunLen = 0; // How many data bytes in a row have we seen?
	bool IsolatedCodeTrigger = false; // Have seen data, then isolated code
									// Now looking for data
	ea_t IsolatedCodeAddr;
	int IsolatedCodeLen;
	int InstrLen;
	bool InstOK;
	insn_t LocalCmd;
	ulong LocalFeatures;

	ea_t RecentAddr = BADADDR;
#if IDA_SDK_VERSION < 600
	for (int SegIndex = 0; SegIndex < SMP_get_segm_qty(); ++SegIndex) {
		segment_t *seg = SMP_getnseg(SegIndex);
#else
	for (segment_t *seg = SMP_get_first_seg(); NULL != seg; seg = SMP_get_next_seg(RecentAddr)) {
		RecentAddr = seg->startEA;
#endif
		if (SEG_CODE != seg->type)
			continue;
#if SMP_DEBUG_FIXUP_IDB
		char SegName[MAXSTR];
		ssize_t SegNameSize = SMP_get_segm_name(seg, SegName, sizeof(SegName) - 1);
		SMP_msg("Non-code addresses for code segment %s from %x to %x\n",
			SegName, seg->startEA, seg->endEA);
#endif
		for (ea_t addr = seg->startEA; addr < seg->endEA; addr = get_item_end(addr)) {
			flags_t AddrFlags = getFlags(addr);
			if (isHead(AddrFlags)) {
				if (isData(AddrFlags)) {
					DataRunLen += get_item_size(addr);
#if SMP_DEBUG_FIXUP_IDB
					SMP_msg("Data: %x\n", addr);
#endif
					if (MIN_DATARUN_LEN <= DataRunLen) {
						if (IsolatedCodeTrigger) {
							// Saw data, then one isolated code, then data
							do_unknown_range(IsolatedCodeAddr, IsolatedCodeLen, DOUNK_SIMPLE);
							RemoveIDACodeAddr(IsolatedCodeAddr);
							if (do_data_ex(IsolatedCodeAddr, byteflag(),
								IsolatedCodeLen, BADNODE)) {
									SMP_msg("Converted isolated code to data: %lx\n",
										(unsigned long) IsolatedCodeAddr);
							}
							else {
								SMP_msg("Failed to convert isolated code to data: %lx len: %x\n",
									(unsigned long) IsolatedCodeAddr, IsolatedCodeLen);
							}
							IsolatedCodeTrigger = false;
						} // end if (IsolatedCodeTrigger)
					} // end if (MIN_DATARUN_LEN <= DataRunLen)
				} // end if (isData(AddrFlags)
				else if (isUnknown(AddrFlags)) {
					// Just in case; unknown usually means not head or tail
					// If in a data run, convert to data.
					InstrLen = get_item_size(addr);
#if SMP_DEBUG_FIXUP_IDB
					SMP_msg("Unknown: %x len: %x\n", addr, InstrLen);
#endif
					if (0 < DataRunLen) {
						if (do_data_ex(addr, byteflag(), InstrLen, BADNODE)) {
#if SMP_DEBUG_FIXUP_IDB
							SMP_msg("Converted unknown to data at %x len: %x\n", addr, InstrLen);
#endif
							DataRunLen += InstrLen;
						}
						else {
#if SMP_DEBUG_FIXUP_IDB
							SMP_msg("Failed to convert unknown to data at %x len: %x\n", addr, InstrLen);
#endif
							DataRunLen = 0;
							IsolatedCodeTrigger = false;
						}
					}
				}
				else if (isCode(AddrFlags)) {  // must be true
					if (MIN_DATARUN_LEN <= DataRunLen) {
#if SMP_DEBUG_FIXUP_IDB
						SMP_msg("DataRunLen: %d at %x\n", DataRunLen, addr);
#endif
						InstOK = SMPGetCmd(addr, LocalCmd, LocalFeatures);
						assert(InstOK);

						InstrLen = (int) LocalCmd.size;

						// We don't check the returned InstrLen for validity because IsCodeMisaligned()
						//  will check for validity immediately below.
#if SMP_DEBUG_FIXUP_IDB
						SMP_msg("Calling IsCodeMisaligned: len %d\n", InstrLen);
#endif
						if (IsCodeMisaligned(addr)) {
#if SMP_DEBUG_FIXUP_IDB
							SMP_msg("Code was misaligned.\n");
#endif
							do_unknown_range(addr, InstrLen, DOUNK_SIMPLE);
							RemoveIDACodeAddr(addr);
							if (do_data_ex(addr, byteflag(), InstrLen, BADNODE)) {
#if SMP_DEBUG_FIXUP_IDB
								SMP_msg("Converted misaligned code to data at %x : len: %x\n",
									addr, InstrLen);
#endif
								// Step back so data gets processed.
								DataRunLen += get_item_size(addr);
								continue; // skip reset of DataRunLen
							}
							else {
#if SMP_DEBUG_FIXUP_IDB
								SMP_msg("Misaligned code left as unknown at %x : len: %x\n",
									addr, InstrLen);
#endif
								IsolatedCodeTrigger = false;
							}
						} // end if (IsCodeMisaligned() ...)
						else if (!hasRef(AddrFlags)) {
							// No references at all --> isolated code.
							IsolatedCodeTrigger = true;
							IsolatedCodeAddr = addr;
							IsolatedCodeLen = InstrLen;
						}
						else {
							SMP_xref_t xb;
							bool ok = xb.SMP_first_to(addr, XREF_ALL);
							if (!ok) {
								// No code xrefs to this target addr.
								IsolatedCodeTrigger = true;
								IsolatedCodeAddr = addr;
								IsolatedCodeLen = InstrLen;
							}
						}
					} // end if (MIN_DATARUN_LEN <= DataRunLen)
					else if (IsolatedCodeTrigger) {
						// Two instructions in a row does not fit the pattern.
						IsolatedCodeTrigger = false;
					}
					DataRunLen = 0;
				} // end if (isData) ... else if (isUnknown) ... else isCode
			} // end if (isHead)
			else if (isUnknown(AddrFlags)) {
				// If in a data run, convert to data.
				InstrLen = get_item_size(addr);
#if SMP_DEBUG_FIXUP_IDB
				SMP_msg("Unknown: %x len: %x\n", addr, InstrLen);
#endif
				if (0 < DataRunLen) {
					if (do_data_ex(addr, byteflag(), InstrLen, BADNODE)) {
#if SMP_DEBUG_FIXUP_IDB
						SMP_msg("Converted unknown to data at %x len: %x\n", addr, InstrLen);
#endif
						DataRunLen += InstrLen;
					}
					else {
#if SMP_DEBUG_FIXUP_IDB
						SMP_msg("Failed to convert unknown to data at %x len: %x\n", addr, InstrLen);
#endif
						DataRunLen = 0;
						IsolatedCodeTrigger = false;
					}
				}
			}
		} // end for (ea_t addr =  seg->startEA; ...)
	} // end for all segments
	return;
} // end of FindDataInCode()


// The choices that IDA makes for deciding which parent function of a
//  TAIL chunk is the primary owner of the tail can be counterintuitive.
//  A function entry can both fall into and jump to a tail chunk that
//  is contiguous with it, yet the "owner" might be a function that is
//  far below it in the executable address space. This function will
//  change the ownership to a more sensible arrangement.
void AuditTailChunkOwnership(void) {
	char FuncName[MAXSTR];
	// Iterate through all chunks in the program.
	size_t NumChunks = get_fchunk_qty();
	for (size_t ChunkIndex = 0; ChunkIndex < NumChunks; ++ChunkIndex) {
		func_t *ChunkInfo = getn_fchunk((int) ChunkIndex);
		if (is_func_tail(ChunkInfo)) {
			// For each TAIL chunk, find all the parent chunks. Find the last
			//  parent chunk with an address less than the TAIL chunk address.
			ea_t BestCandidate = 0;
			func_parent_iterator_t FuncParent(ChunkInfo);
#if SMP_DEBUG_CHUNKS
			SMP_msg("Tail chunk: %x ", ChunkInfo->startEA);
#endif
			for (bool ok = FuncParent.first(); ok; ok = FuncParent.next()) {
				ea_t parent = FuncParent.parent();
#if SMP_DEBUG_CHUNKS
				SMP_msg(" parent: %x ", parent);
#endif
				if ((parent > BestCandidate) && (parent < ChunkInfo->startEA))
					BestCandidate = parent;
			}
#if SMP_DEBUG_CHUNKS
			SMP_msg("\n");
#endif
			//  Make the best parent chunk the owner of the TAIL chunk if it is
			//  not already the owner.
			if (ChunkInfo->owner != BestCandidate) {
				if (0 < BestCandidate) {
					if (set_tail_owner(ChunkInfo, BestCandidate)) {
						func_t *FuncInfo = get_func(BestCandidate);
						SMP_msg("Set %lx as new owner of tail %lx\n",
							(unsigned long) BestCandidate, (unsigned long) ChunkInfo->startEA);
						// Reanalyze the parent function (and all its
						//  tail chunks) now that the structure has changed.
						reanalyze_function(FuncInfo);
					}
					else {
						SMP_msg("set_tail_owner failed for tail %lx and parent %lx\n",
							(unsigned long) ChunkInfo->startEA, (unsigned long) BestCandidate);
					}
				}
				else {
					func_t *FuncInfo = get_func(ChunkInfo->owner);
					get_func_name(FuncInfo->startEA, FuncName, sizeof(FuncName) - 1);
#if SMP_DEBUG_CHUNKS
					SMP_msg("No good parent candidate before tail at %x\n",
						ChunkInfo->startEA);
					SMP_msg("Current parent is %x: %s\n", FuncInfo->startEA, FuncName);
#endif
					// Find out if a function entry chunk that comes before the
					//  tail is a better candidate for the owner (i.e. it falls
					//  through to the tail, or jumps to it).
					BestCandidate = 0;
#if SMP_DEBUG_CHUNKS
					SMP_msg("Finding parent func candidates for %x:", ChunkInfo->startEA);
#endif
					SMP_bounds_t CurrFunc;
					for (size_t FuncIndex = 0; FuncIndex < FuncBounds.size(); ++FuncIndex) {
						CurrFunc = FuncBounds[FuncIndex];
						if ((CurrFunc.startEA < ChunkInfo->startEA)
							&& (CurrFunc.startEA > BestCandidate)) {
							BestCandidate = CurrFunc.startEA;
#if SMP_DEBUG_CHUNKS
							SMP_msg(" candidate: %x tail: %x", BestCandidate,
								ChunkInfo->startEA);
#endif
						}
						else {
#if SMP_DEBUG_CHUNKS
							SMP_msg(" not a candidate: %x tail: %x best: %x\n",
								CurrFunc.startEA, ChunkInfo->startEA, BestCandidate);
#endif
							break;
						}
					} // end for (size_t FuncIndex = 0; ...)
					if (0 >= BestCandidate) { // highly unlikely
						SMP_msg("No good func entry parent candidate.\n");
					}
					else {
						FuncInfo = get_func(BestCandidate);
						get_func_name(FuncInfo->startEA, FuncName, sizeof(FuncName) - 1);
#if SMP_DEBUG_CHUNKS
						SMP_msg("Best func entry parent candidate: %s at %x",
							FuncName, BestCandidate);
						if (FuncInfo->endEA == ChunkInfo->startEA)
							SMP_msg(" Function endEA == tail chunk startEA");
						SMP_msg("\n");
#endif
					}
				}
			} // end if (ChunkInfo->owner != BestCandidate)
#if SMP_DEBUG_CHUNKS
			else {
				SMP_msg("Already best parent for %x is %x\n", ChunkInfo->startEA,
					ChunkInfo->owner);
			}
#endif
		} // end if (is_func_tail(ChunkInfo))
	} // end for (size_t ChunkIndex = 0; ...)

	return;
} // end of AuditTailChunkOwnership()

// If the addresses signified from DisasmIndex to IDAProIndex are
//  all considered data and do NOT follow a return instruction,
//  return false and update AreaSize to reflect the area to be
//  converted.
// Return value: true -> skip to IDAProIndex; false -> convert AreaSize bytes.
bool FindDataToConvert(size_t IDAProIndex, size_t DisasmIndex, int &AreaSize) {
	ea_t PrevIDAAddr;
	ea_t NextIDAAddr;
	size_t ShadowDisasmIndex = DisasmIndex - 1;
	ea_t DisasmAddr = DisasmLocs[ShadowDisasmIndex];
	bool CannotConvert = false;  // return value
	bool DebugAddress = false;
#if SMP_DEBUG_FIXUP_IDB
	DebugAddress = (DisasmAddr == 0x806c19a);
#endif

	if (DebugAddress) {
		SMP_msg("IDAProIndex: %zu DisasmIndex: %zu\n", IDAProIndex, DisasmIndex);
		SMP_msg("IDA locs size %zu Disasm locs size %zu\n", IDAProLocs.size(),
			DisasmLocs.size());
	}
	if (IDAProIndex >= IDAProLocs.size()) {
		// Have already processed the last IDA address.
		if (DebugAddress) SMP_msg(" Already done with IDAProLocs.\n");
		return true;
	}
	else if (DisasmIndex >= DisasmLocs.size()) {
		// Strange. Last Disasm address is only one to convert, and
		//  IDA still has addresses after that?
		if (DebugAddress) SMP_msg(" Already done with DisasmLocs.\n");
		return true;
	}
	else if (IDAProIndex < 2) {
		// We have Disasm addrs before the very first IDA addr. We
		//  don't trust this boundary case.
		if (DebugAddress) SMP_msg(" Boundary case with IDAProLocs.\n");
		return true;
	}
	NextIDAAddr = IDAProLocs[IDAProIndex - 1];
	PrevIDAAddr = IDAProLocs[IDAProIndex - 2];
	if (DebugAddress) SMP_msg(" PrevIDAAddr: %lx NextIDAAddr: %lx\n", (unsigned long) PrevIDAAddr, (unsigned long) NextIDAAddr);

	// See if previous IDA address was a return.
	flags_t PrevFlags = getFlags(PrevIDAAddr);
	if (!isCode(PrevFlags) || !isHead(PrevFlags)) {
		SMP_msg("PrevIDAAddr %lx not isCode or not isHead.\n", (unsigned long) PrevIDAAddr);
		return true;
	}
	SMPInstr PrevInstr(PrevIDAAddr);
	PrevInstr.Analyze();
	if (DebugAddress) SMP_msg("Finished PrevInstr.Analyze()\n");
	if (PrevInstr.MDIsReturnInstr()) {
		// Right after a return come no-ops and 2-byte no-ops
		//  that are just for alignment. IDA does not seem to be
		//  happy when we convert all those to code.
		if (DebugAddress) SMP_msg(" Data followed a return instruction.\n");
		return true;
	}
	// Now, see if the area from DisasmAddr to NextIDAAddr is all data
	//  according to IDA.
	while (DisasmAddr < NextIDAAddr) {
		flags_t DataFlags = getFlags(DisasmAddr);
		if (isTail(DataFlags)) {
			if (DebugAddress) SMP_msg(" tail byte: %lx\n", (unsigned long) DisasmAddr);
			DisasmAddr = get_item_end(DisasmAddr);
		}
		else if (isData(DataFlags)) {
			if (DebugAddress) SMP_msg(" data byte: %lx\n", (unsigned long) DisasmAddr);
			DisasmAddr = get_item_end(DisasmAddr);
		}
		else if (isCode(DataFlags)) {
			// How could this ever happen?
			if (DebugAddress) SMP_msg(" isCode: %lx\n", (unsigned long) DisasmAddr);
			return true;
		}
		else { // must be isUnknown()
			// Very conservative here; only want to convert when the whole
			//  region is data, because that is a symptom of IDA missing
			//  a piece of code within a function (usually a piece of code
			//  that is only reachable via an indirect jump).
			if (DebugAddress) SMP_msg(" Not isData: %lx\n", (unsigned long) DisasmAddr);
			return true;
		}
		if (DebugAddress) SMP_msg(" new DisasmAddr: %lx\n", (unsigned long) DisasmAddr);
	} // end while (DisasmAddr < NextIDAAddr)
	if (DebugAddress) SMP_msg(" loop exit CannotConvert: %d\n", CannotConvert);
	if (!CannotConvert) {
		// Success.
		DisasmAddr = DisasmLocs[ShadowDisasmIndex];
		AreaSize = NextIDAAddr - DisasmAddr;
		if (DebugAddress) { 
			SMP_msg(" Success! AreaSize: %x Old index: %zu new index: %zu\n",
				AreaSize, ShadowDisasmIndex, DisasmIndex);
			SMP_msg(" exiting FindDataToConvert()\n");
			SMP_msg("\n");
		}
	} // end if (!CannotConvert)
	return CannotConvert;
} // end of FindDataToConvert()

// Does a converted code region look like a function prologue? If so,
//  we should not include it in the previous function.
bool IsFunctionPrologue(ea_t StartAddr, ea_t EndAddr) {
	return false;  // **!!** TODO 
} // end of IsFunctionPrologue()

// Patch program bytes that could not be converted from
//  data to code, if it can be determined that the bytes represent code
//  that IDA has a hard time with.
// Currently limited to finding "call near ptr 0" instructions, which
//  often are found in optimized glibc code because gcc was able to
//  determine that a function pointer was zero and did constant propagation,
//  but unfortunately was not able to determine that the code was unreachable.
//  IDA will not succeed in ua_code() for "call 0", but there is no danger
//  of a working program ever executing this code. Replacing the call with
//  no-ops permits us to continue converting a contiguous range of data to
//  code, and permits IDA to reanalyze the function later.
// Returns true if program bytes were patched.
bool MDPatchUnconvertedBytes(ea_t CurrDisasmAddr) {
	flags_t AddrFlags = getFlags(CurrDisasmAddr);
	if (isData(AddrFlags) || isTail(AddrFlags)) {
		// Bytes should have been converted to unknown already.
#if SMP_DEBUG_FIXUP_IDB
		SMP_msg("Cannot patch data bytes or tail bytes at %x\n", CurrDisasmAddr);
#endif
		return false;
	}
	SMPInstr PatchInstr(CurrDisasmAddr);
	PatchInstr.Analyze();
	int InstrLen = PatchInstr.GetCmd().size;
	if (0 >= InstrLen) {
#if SMP_DEBUG_FIXUP_IDB
		SMP_msg("decode_insn() failed on patch location %x\n", CurrDisasmAddr);
#endif
		return false;
	}
	else {
		if (PatchInstr.GetCmd().itype != NN_call) {
#if SMP_DEBUG_FIXUP_IDB
			SMP_msg("Cannot patch non-call instruction at %x\n", CurrDisasmAddr);
#endif
			return false;
		}
		PatchInstr.PrintOperands();
		op_t CallDest = PatchInstr.GetFirstUse()->GetOp();
		if ((o_near != CallDest.type) || (0 != CallDest.addr)) {
#if SMP_DEBUG_FIXUP_IDB
			SMP_msg("Cannot patch call unless it is call near ptr 0 at %x",
				CurrDisasmAddr);
#endif
			return false;
		}
		ea_t PatchAddr = CurrDisasmAddr;
		for (int i = 0; i < InstrLen; ++i) {
			bool ok = patch_byte(PatchAddr, 0x90);  // x86 no-op
			if (!ok) {
#if SMP_DEBUG_FIXUP_IDB
				SMP_msg("patch_byte() failed at %x\n", PatchAddr);
#endif
				return false;
			}
			++PatchAddr;
		}
#if SMP_DEBUG_FIXUP_IDB
		SMP_msg("Patched %d bytes successfully at %x\n", InstrLen, CurrDisasmAddr);
#endif
#if IDA_SDK_VERSION < 600
		InstrLen = ua_code(CurrDisasmAddr);
#else
		InstrLen = create_insn(CurrDisasmAddr);
#endif
		if (0 >= InstrLen) {
#if SMP_DEBUG_FIXUP_IDB
			SMP_msg(" ... but ua_code() still failed!\n");
#endif
			return false;
		}
	} // end if (0 >= InstrLen) ... else ...
	return true;
} // end of MDPatchUnconvertedBytes()

// Use the lists of code addresses identified by IDA Pro (in IDAProLocs)
//  and an external disassembler (in DisasmLocs). Compare the lists and
//  try to convert addresses to code that are found in DisasmLocs but
//  not in IDAProLocs. Emit warnings when IDAProLocs has a code address
//  not found in DisasmLocs.
void FixCodeIdentification(void) {
	size_t DisasmIndex = 0;
	ea_t CurrDisasmAddr = DisasmLocs[DisasmIndex++];
	size_t IDAProIndex = 0;
	ea_t CurrAddr = IDAProLocs[IDAProIndex++];

	while (DisasmIndex <= DisasmLocs.size()) {
		// If the current address is less than the current
		//  external disasm address, we have the rare case in
		//  which IDA Pro has identified an address as code
		//  but the external disasm has not. Emit a warning
		//  message and go on to the next IDA address.
		if (CurrAddr < CurrDisasmAddr) {
			SMPInstr TempInstr(CurrAddr);
			TempInstr.Analyze();
			SMP_msg("AUDIT: Address %lx is code in IDB but not in external disassembler: %s\n",
				(unsigned long) CurrAddr, TempInstr.GetDisasm());
			if (IDAProIndex < IDAProLocs.size())
				CurrAddr = IDAProLocs[IDAProIndex++];
			else {
				// Last IDA addr; might still process Disasm addrs
				//  after loop exit.
				break;
			}
		}
		else if (CurrAddr == CurrDisasmAddr) {
			// If equal, no problem, we are moving through the
			//  code addresses in lockstep. Grab the next address
			//  from each source.
			if (DisasmIndex < DisasmLocs.size()) {
				CurrDisasmAddr = DisasmLocs[DisasmIndex++];
			}
			else {
				++DisasmIndex;  // cause loop exit; skip cleanup loop
			}
			if (IDAProIndex < IDAProLocs.size())
				CurrAddr = IDAProLocs[IDAProIndex++];
			else {
				// Last IDA addr; might still process Disasm addrs
				//  after loop exit in cleanup loop.
				break;
			}
		}
		else {
			// We must have CurrAddr > CurrDisasmAddr. That means
			//  IDA has jumped over some code addresses in
			//  DisasmLocs. We need to try to convert addresses
			//  to code until we can reach the current addr.
			// For now, we will address only the case in which IDA
			//  has identified addresses as data bytes, and the
			//  external disassembler(e.g. objdump) has identified
			//  the same addresses as code. We only want to deal with
			//  contiguous areas of data-to-code conversion that do NOT
			//  follow a return statement.
			int AreaSize = 0;
			ea_t AreaStart = CurrDisasmAddr;
			ea_t AreaEnd;
#if SMP_DEBUG_FIXUP_IDB
			SMP_msg("CurrDisasmAddr: %x  CurrAddr: %x\n", CurrDisasmAddr, CurrAddr);
#endif
			bool SkipArea = FindDataToConvert(IDAProIndex, DisasmIndex, AreaSize);
			if (SkipArea) {
				// Skip over the extra external disasm addresses.
				while (CurrDisasmAddr < CurrAddr)
					CurrDisasmAddr = DisasmLocs[DisasmIndex++];
			}
			else { 
				// Convert the overlooked code region to unexplored.
				AreaEnd = CurrDisasmAddr + AreaSize;
#if SMP_DEBUG_FIXUP_IDB
				SMP_msg("Found data to convert: %x to %x\n", AreaStart, AreaEnd);
#endif
				do_unknown_range(AreaStart, AreaSize, DOUNK_SIMPLE);
				SMP_bounds_t ConvertRegion;
				ConvertRegion.startEA = AreaStart;
				ConvertRegion.endEA = AreaEnd;
				FixupRegion CurrRegion(ConvertRegion);
				CodeReanalyzeList.push_back(CurrRegion);
				bool AllConverted = true;
				bool AllNops = true;
				do {
					flags_t InstrFlags = getFlags(CurrDisasmAddr);
					if (!isUnknown(InstrFlags)) {
						SMP_msg("Sync problem in FixCodeID: %lx\n", (unsigned long) CurrDisasmAddr);
					}
					else {
#if IDA_SDK_VERSION < 600
						int InstrLen = ua_code(CurrDisasmAddr);
#else
						int InstrLen = create_insn(CurrDisasmAddr);
#endif
						if (InstrLen > 0) { // Successfully converted to code
							SMPInstr NewInstr(CurrDisasmAddr);
							NewInstr.Analyze();
							if (!NewInstr.IsNop())
								AllNops = false;
#if SMP_DEBUG_FIXUP_IDB
#if 0
							SMP_msg("FixCodeID success at %x: len: %d %s\n", CurrDisasmAddr,
									InstrLen, NewInstr.GetDisasm());
#endif
#endif
						}
						else {
							if (MDPatchUnconvertedBytes(CurrDisasmAddr)) {
								;
#if SMP_DEBUG_FIXUP_IDB
								SMP_msg(" Patched bytes at %x\n", CurrDisasmAddr);
#endif
							}
							else {
								CurrRegion.FixupInstrs.push_back(CurrDisasmAddr);
								AllConverted = false;
#if SMP_DEBUG_FIXUP_IDB
								SMP_msg("FixCodeID failure at %x\n", CurrDisasmAddr);
#endif
							}
						}
					} // end if (isCode(InstrFlags) ... else ...
					if (DisasmIndex < DisasmLocs.size()) {
						CurrDisasmAddr = DisasmLocs[DisasmIndex++];
					}
					else {
						// cause loops to exit
						CurrDisasmAddr = CurrAddr;
						++DisasmIndex; // skip cleanup loop
					}
				} while (CurrDisasmAddr < CurrAddr);
				if (AllConverted && AllNops) {
					// We want to convert the region back to unexplored bytes
					//  and take it off the work list. Regions that are all nops
					//  create data flow analysis problems sometimes. The region
					//  is often unreachable code and produces a basic block with
					//  no predecessors within a function. This often happens when
					//  an optimizing compiler uses nops as padding to align jump
					//  targets on cache line bounaries. With no fall through into
					//  the nops, they are unreachable and should be left as unknown.
#if SMP_DEBUG_FIXUP_IDB
					SMP_msg("FixCodeID nops region from %x to %x\n", CurrRegion.GetStart(),
						CurrRegion.GetEnd());
#endif
					do_unknown_range(CurrRegion.GetStart(),
						CurrRegion.GetEnd() - CurrRegion.GetStart(), DOUNK_SIMPLE);
					CodeReanalyzeList.pop_back();
				}
			} // end if (SkipArea) ... else ...
		} // end if (addr < CurrDisasmAddr) .. else if ... else ...
	} // end while (DisasmIndex <= DisasmLocs.size()

#if 0  // Make this code use FindDataToConvert()  **!!**
	// Cleanup loop:
	// If there are still Disasm addrs to process, try to turn them
	//  into code in the IDB.
	while (DisasmIndex <= DisasmLocs.size()) {
		flags_t InstrFlags = getFlags(CurrDisasmAddr);
		if (isCode(InstrFlags)) {
			SMP_msg("Sync problem in FixCodeID: %x\n", CurrDisasmAddr);
		}
		else {
			// Clear bytes to unexplored.
			segment_t *seg = SMP_getseg(CurrDisasmAddr);
			if (SEG_CODE == seg->type) {
				do_unknown_range(CurrDisasmAddr, seg->endEA - CurrDisasmAddr, DOUNK_SIMPLE);
			}
			else {
				// Might be safest to just discontinue processing
				//  if we wander into a non-code segment.
				//  DisasmLocs should not have an entire code segment
				//  that IDA Pro missed.
				break;
			}
			int InstrLen = ua_code(CurrDisasmAddr);
			if (InstrLen > 0) { // Successfully converted to code
				SMPInstr NewInstr(CurrDisasmAddr);
				NewInstr.Analyze();
				SMP_msg("FixCodeID success at %x: %s\n", CurrDisasmAddr,
						NewInstr.GetDisasm());
			}
			else {
				SMP_msg("FixCodeID failure at %x\n", CurrDisasmAddr);
			}
		} // end if (isCode(InstrFlags) ... else ...
		if (DisasmIndex < DisasmLocs.size()) {
			CurrDisasmAddr = DisasmLocs[DisasmIndex++];
		}
		else {
			++DisasmIndex; // cause loop to exit
		}
	} // end while (DisasmIndex <= DisasmLocs.size()
#endif

	return;
} // end of FixCodeIdentification()

// Analyze instructions that could not be analyzed earlier and were placed on the CodeReanalyzeList.
//  Earlier failures are usually because the instruction branches to an address that has not
//  yet been converted from data to code, so ua_code() failed. Now that all data to code
//  conversions have completed, ua_code() should succeed.
// Return the number of instructions successfully analyzed.
int FixupNewCodeChunks(void) {
	list<FixupRegion>::iterator CurrRegion;
	int changes = 0;
	for (CurrRegion = CodeReanalyzeList.begin(); CurrRegion != CodeReanalyzeList.end(); ++CurrRegion) {
		bool AllConverted = true;
		bool AllNops = true;
		bool NoFixups = (0 == CurrRegion->FixupInstrs.size());
		if (NoFixups) {
			CurrRegion->SetStart(BADADDR);  // mark for removal
			continue;  // skip to next region
		}
		list<ea_t>::iterator CurrInstr;
		for (CurrInstr = CurrRegion->FixupInstrs.begin(); CurrInstr != CurrRegion->FixupInstrs.end(); ++CurrInstr)  {
#if IDA_SDK_VERSION < 600
			int InstrLen = ua_code(*CurrInstr);
#else
			int InstrLen = create_insn(*CurrInstr);
#endif
			if (InstrLen > 0) { // Successfully converted to code
				SMPInstr NewInstr(*CurrInstr);
				NewInstr.Analyze();
#if SMP_DEBUG_FIXUP_IDB
				SMP_msg("FixupNewCodeChunks success at %x: len: %d\n", *CurrInstr, InstrLen);
#endif
				if (!NewInstr.IsNop()) {
					AllNops = false;
					*CurrInstr = BADADDR; // mark for removal
				}
			}
			else {
				AllConverted = false;
#if SMP_DEBUG_FIXUP_IDB
				SMP_msg("FixupNewCodeChunks failure at %x\n", *CurrInstr);
#endif
			}
		} // end for all instrs in CurrRegion
		if (AllConverted && !AllNops) {
#if SMP_DEBUG_FIXUP_IDB
			SMP_msg("FixupNewCodeChunks success for region from %x to %x\n",
				CurrRegion->GetStart(), CurrRegion->GetEnd());
#endif
			CurrRegion->SetStart(BADADDR); // mark for removal
		}
		else if (AllConverted && AllNops) {
#if SMP_DEBUG_FIXUP_IDB
			SMP_msg("FixupNewCodeChunks re-converting nops region from %x to %x\n",
				CurrRegion->GetStart(), CurrRegion->GetEnd());
#endif
			do_unknown_range(CurrRegion->GetStart(),
				CurrRegion->GetEnd() - CurrRegion->GetStart(), DOUNK_SIMPLE);
			CurrRegion->SetStart(BADADDR); // mark for removal
		}
		else {
			// Remove only the instructions that were fixed up.
			CurrInstr = CurrRegion->FixupInstrs.begin(); 
			while (CurrInstr != CurrRegion->FixupInstrs.end()) {
				if (BADADDR == *CurrInstr) {
					CurrInstr = CurrRegion->FixupInstrs.erase(CurrInstr);
				}
				else {
					++CurrInstr;
				}
			}
		}
	} // end for all regions in the CodeReanalyzeList

	// Remove completed regions from the CodeReanalyzeList
	CurrRegion = CodeReanalyzeList.begin();
	while (CurrRegion != CodeReanalyzeList.end()) {
		if (BADADDR == CurrRegion->GetStart())
			CurrRegion = CodeReanalyzeList.erase(CurrRegion);
		else
			++CurrRegion;
	}

#if 0
	if (AllConverted) {
					if (IsFunctionPrologue(AreaStart, AreaEnd)) {
						// Create a new function entry chunk here.
						//  **!!** TODO
						;
					}
					else {
						// Extend the previous chunk to include the
						//  converted code.
						ea_t PrevIDAAddr = IDAProLocs[IDAProIndex - 2];
						func_t *PrevChunk = get_fchunk(PrevIDAAddr);
#if SMP_DEBUG_FIXUP_IDB
						SMP_msg(" addr in chunk to extend: %x\n", PrevIDAAddr);
						SMP_msg(" func_t pointer for chunk: %x\n", PrevChunk);
#endif
#if 0  // temporary for debugging
						if (is_func_entry(PrevChunk)) {
							// Extend the func entry to contain the new code.
							if (func_setend(PrevIDAAddr, AreaEnd)) {
								SMP_msg("Func extended to include code from %x to %x\n",
									AreaStart, AreaEnd);
								FuncReanalyzeList.push_back(PrevIDAAddr);
							}
							else {
								SMP_msg("Failed to extend func from %x to %x\n",
									AreaStart, AreaEnd);
							}
						}
						else { // tail
							// See if this works for function tails, also.
							// Extend the func entry to contain the new code.
							if (func_setend(PrevIDAAddr, AreaEnd)) {
								SMP_msg("Tail extended to include code from %x to %x\n",
									AreaStart, AreaEnd);
								func_t *TailOwner = get_func(PrevChunk->owner);
								FuncReanalyzeList.push_back(PrevIDAAddr);
							}
							else {
								SMP_msg("Failed to extend tail from %x to %x\n",
									AreaStart, AreaEnd);
							}
						} // end if (is_func_entry()) ... else ...
#endif
					} // end if (IsFunctionPrologue()) ... else ...
				} // end if (AllConverted)
				else {
					SMP_msg("not AllConverted; cannot include new code in previous chunk.\n");
				}
#endif

	return changes;
} // end of FixupNewCodeChunnks()

// Audit the IDA code database by looking at all instructions in the
//  code segment and printing all those that are not contained in a
//  function. Emit the context-free annotations that we are able to
//  emit on a per-instruction basis.
void FindOrphanedCode(segment_t *CurrSeg, FILE *AnnotFile, FILE *InfoAnnotFile) {
	char disasm[MAXSTR];
	for (ea_t addr = CurrSeg->startEA; addr < CurrSeg->endEA;
		addr = get_item_end(addr)) {
		
		flags_t InstrFlags = getFlags(addr);
		if (isTail(InstrFlags))
			continue;
		if (isHead(InstrFlags) && isCode(InstrFlags)) {
			ea_t FirstFuncAddr;
			if (!(CurrProg->IsInstAddrStillInFunction(addr, FirstFuncAddr))) {
				SMPInstr CurrInst(addr);
				CurrInst.Analyze();
#if SMP_DEBUG_FIXUP_IDB
				SMP_msg("Orphan code at %x : %s\n", addr, CurrInst.GetDisasm());
#endif
				// TODO: If there are code xrefs to the orphan code,
				//  see what kind. If a CALL, and orphan code looks
				//  like a prologue, make a function. If a JUMP of
				//  some kind, then make a function chunk and make
				//  it a tail of all functions that jump to it. **!!**

				// Do machine-dependent fixes for DEF and USE lists.
				//  The fixes can help produce better annotations.
				CurrInst.MDFixupDefUseLists();

				// If instruction is still not included in a code chunk,
				//  emit annotations for it in isolation.
				CurrInst.EmitAnnotations(true, false, true, AnnotFile, InfoAnnotFile, CurrProg);

				// If instruction is an indirect branch, emit an XREF
				//  annotation for each of its targets.
				SMPitype CurrDataFlow = CurrInst.GetDataFlowType();
				if ((CurrDataFlow == INDIR_JUMP) || (CurrDataFlow == INDIR_CALL)) {
					SMP_xref_t xrefs;
					for (bool ok = xrefs.SMP_first_from(addr, XREF_ALL); ok; ok = xrefs.SMP_next_from()) {
						if (xrefs.GetTo() != 0) {
							if (xrefs.GetIscode() && (xrefs.GetType() != fl_F)) {
								// Found a code target, with its address in xrefs.to
								PrintCodeToCodeXref(addr, xrefs.GetTo(), CurrInst.GetSize());
							}
						}
					}
				}

			}
		}
		else if (isUnknown(InstrFlags)) {
#if SMP_DEBUG_FIXUP_IDB
			SMP_msg("Unanalyzed byte at %x\n", addr);
#endif
			// Can IDA analyze this to be code?
			int InstrLen;
#if IDA_SDK_VERSION < 600
			InstrLen = ua_code(addr);
#else
			InstrLen = create_insn(addr);
#endif
			if (InstrLen > 0) {
				bool IDAsuccess = generate_disasm_line(addr, disasm, sizeof(disasm) - 1);
				if (IDAsuccess) {
					// Remove interactive color-coding tags.
					ssize_t StringLen = tag_remove(disasm, disasm, 0);
					if (-1 >= StringLen) {
						SMP_msg("ERROR: tag_remove failed at addr %lx \n", (unsigned long) addr);
					}
					else {
#if SMP_DEBUG_FIXUP_IDB
						SMP_msg("Successfully analyzed!  %s\n", disasm);
#endif
						SMPInstr UnknownInstr(addr);
						UnknownInstr.Analyze();
						// TODO: Get new code into a chunk.  **!!**

						// If instruction is still not included in a code chunk,
						//  emit annotations for it in isolation.
						UnknownInstr.EmitAnnotations(true, false, true, AnnotFile, InfoAnnotFile, CurrProg);
					}
				}
				else {
					SMP_msg("ERROR: generate_disasm_line failed at addr %lx \n", (unsigned long) addr);
				}
			}
		}
	} // end for (ea_t addr = CurrSeg->startEA; ...)
} // end of FindOrphanedCode()

// Version of FindOrphanedCode that does not emit annotations but can be used
//  to determine at what point in time code becomes orphaned.
void Debug_FindOrphanedCode(segment_t *CurrSeg, bool FirstRun) {
	ea_t DebugAddr = 0x8050db0;
	for (ea_t addr = CurrSeg->startEA; addr < CurrSeg->endEA;
		addr = get_item_end(addr)) {

		flags_t InstrFlags = getFlags(addr);
		if (isHead(InstrFlags) && isCode(InstrFlags)) {
			func_t *CurrFunc = get_func(addr);
			if (NULL == CurrFunc) {  // Code not in a func; orphaned
				pair<set<ea_t>::iterator, bool> pairib;
				pairib = CodeOrphans.insert(addr);
				if (DebugAddr == addr) {
					SMP_msg("DEBUG: Orphaned code addr %lx found.\n", (unsigned long) addr);
				}
				if ((!FirstRun) && (pairib.second)) {
					SMP_msg("SERIOUS WARNING: Newly orphaned code at %lx \n", (unsigned long) addr);
				}
			}
		}
	} // end for (ea_t addr = CurrSeg->startEA; ...)
} // end of Debug_FindOrphanedCode()

// Audit the IDA database with respect to branches and calls. They should
//  each have valid code targets (not data or unknown bytes) and the code
//  cross references should reflect the linkage.
void AuditCodeTargets(void) {
	// Cover all the code that IDA has grouped into functions by iterating
	//  through all function chunks in the program.
	size_t NumChunks = get_fchunk_qty();
	for (size_t ChunkIndex = 0; ChunkIndex < NumChunks; ++ChunkIndex) {
		func_t *ChunkInfo = getn_fchunk((int) ChunkIndex);
		char FuncName[MAXSTR];
		get_func_name(ChunkInfo->startEA, FuncName, sizeof(FuncName) - 1);

		// First, see if any calls to this function (if this chunk is
		//  an entry point) are not coming from within functions.
		if (is_func_entry(ChunkInfo)) {
			SMP_xref_t xb;
			ea_t addr = ChunkInfo->startEA;
			for (bool ok = xb.SMP_first_to(addr, XREF_ALL);	ok; ok = xb.SMP_next_to()) {
				uchar XrefType = xb.GetType() & XREF_MASK;
				if (xb.GetIscode()) {
					if ((XrefType == fl_U) || (XrefType == fl_USobsolete)) {
						SMP_msg("AUDIT: Bad xref type: %lx %s\n", (unsigned long) addr, FuncName);
					}
#if SMP_DEBUG_FIXUP_IDB
					else if ((XrefType == fl_JF) || (XrefType == fl_JN)) {
						SMP_msg("Jump to func: %x %s from: %x\n",
							addr, FuncName, xb.GetFrom());
					}
#endif
					else if (XrefType == fl_F) {
						SMP_msg("AUDIT: Fall through to func: %lx %s from: %lx\n",
							(unsigned long) addr, FuncName, (unsigned long) xb.GetFrom());
					}
					else if ((XrefType == fl_CF) || (XrefType == fl_CN)) {
						// Far call or Near call
						func_t *CallingFunc = get_func(xb.GetFrom());
						if (NULL == CallingFunc) {
							;
#if SMP_DEBUG_FIXUP_IDB
							SMP_msg("Call to %x Func %s from %x not in function.\n",
								addr, FuncName, xb.GetFrom());
#endif
						}
					}
				} // end if (xb.GetIscode())
				else { // DATA xref
					if (XrefType == dr_O) {
#if SMP_DEBUG_FIXUP_IDB