BinaryAnalysis::Disassembler::AddressSet
SgAsmM68kInstruction::getSuccessors(const std::vector<SgAsmInstruction*>& insns, bool *complete,
                                    const BinaryAnalysis::MemoryMap::Ptr &initial_memory)
{
    using namespace Rose::BinaryAnalysis::InstructionSemantics2;
    Stream debug(mlog[DEBUG]);

    if (debug) {
        debug <<"SgAsmM68kInstruction::getSuccessors(" <<StringUtility::addrToString(insns.front()->get_address())
              <<" for " <<insns.size() <<" instruction" <<(1==insns.size()?"":"s") <<"):" <<"\n";
    }

    BinaryAnalysis::Disassembler::AddressSet successors = SgAsmInstruction::getSuccessors(insns, complete);

    // If we couldn't determine all the successors, or a cursory analysis couldn't narrow it down to a single successor then
    // we'll do a more thorough analysis now. In the case where the cursory analysis returned a complete set containing two
    // successors, a thorough analysis might be able to narrow it down to a single successor. We should not make special
    // assumptions about function call instructions -- their only successor is the specified address operand. */
    if (!*complete || successors.size()>1) {
        using namespace Rose::BinaryAnalysis::InstructionSemantics2::PartialSymbolicSemantics;

        const RegisterDictionary *regdict = RegisterDictionary::dictionary_coldfire_emac();
        RiscOperatorsPtr ops = RiscOperators::instance(regdict);
        ops->set_memory_map(initial_memory);
        DispatcherM68kPtr dispatcher = DispatcherM68k::instance(ops, 32);
        
        try {
            for (size_t i=0; i<insns.size(); ++i) {
                dispatcher->processInstruction(insns[i]);
                if (debug)
                    debug << "  state after " <<insns[i]->toString() <<"\n" <<*ops;
            }
            SValuePtr ip = SValue::promote(ops->readRegister(dispatcher->REG_PC));
            if (ip->is_number()) {
                successors.clear();
                successors.insert(ip->get_number());
                *complete = true; /*this is the complete set of successors*/
            }
        } catch(const BaseSemantics::Exception& e) {
            /* Abandon entire basic block if we hit an instruction that's not implemented. */
            debug <<e <<"\n";
        }
    }

    if (debug) {
        debug <<"  successors:";
        BOOST_FOREACH (rose_addr_t va, successors)
            debug <<" " <<StringUtility::addrToString(va);
        debug <<(*complete?"":"...") <<"\n";
    }

    return successors;
}
Esempio n. 2
0
// see base class
bool
SgAsmX86Instruction::isFunctionCallSlow(const std::vector<SgAsmInstruction*>& insns, rose_addr_t *target, rose_addr_t *return_va)
{
    if (isFunctionCallFast(insns, target, return_va))
        return true;

    // The following stuff works only if we have a relatively complete AST.
    static const size_t EXECUTION_LIMIT = 10; // max size of basic blocks for expensive analyses
    if (insns.empty())
        return false;
    SgAsmX86Instruction *last = isSgAsmX86Instruction(insns.back());
    if (!last)
        return false;
    SgAsmFunction *func = SageInterface::getEnclosingNode<SgAsmFunction>(last);
    SgAsmInterpretation *interp = SageInterface::getEnclosingNode<SgAsmInterpretation>(func);

    // Slow method: Emulate the instructions and then look at the EIP and stack.  If the EIP points outside the current
    // function and the top of the stack holds an address of an instruction within the current function, then this must be a
    // function call.
    if (interp && insns.size()<=EXECUTION_LIMIT) {
        using namespace Rose::BinaryAnalysis;
        using namespace Rose::BinaryAnalysis::InstructionSemantics2;
        using namespace Rose::BinaryAnalysis::InstructionSemantics2::SymbolicSemantics;
        const InstructionMap &imap = interp->get_instruction_map();
        const RegisterDictionary *regdict = RegisterDictionary::dictionary_for_isa(interp);
        SmtSolverPtr solver = SmtSolver::instance(Rose::CommandLine::genericSwitchArgs.smtSolver);
        BaseSemantics::RiscOperatorsPtr ops = RiscOperators::instance(regdict, solver);
        ASSERT_not_null(ops);
        const RegisterDescriptor SP = regdict->findLargestRegister(x86_regclass_gpr, x86_gpr_sp);
        DispatcherX86Ptr dispatcher = DispatcherX86::instance(ops, SP.get_nbits());
        SValuePtr orig_esp = SValue::promote(ops->readRegister(dispatcher->REG_anySP));
        try {
            for (size_t i=0; i<insns.size(); ++i)
                dispatcher->processInstruction(insns[i]);
        } catch (const BaseSemantics::Exception &e) {
            return false;
        }

        // If the next instruction address is concrete but does not point to a function entry point, then this is not a call.
        SValuePtr eip = SValue::promote(ops->readRegister(dispatcher->REG_anyIP));
        if (eip->is_number()) {
            rose_addr_t target_va = eip->get_number();
            SgAsmFunction *target_func = SageInterface::getEnclosingNode<SgAsmFunction>(imap.get_value_or(target_va, NULL));
            if (!target_func || target_va!=target_func->get_entry_va())
                return false;
        }

        // If nothing was pushed onto the stack, then this isn't a function call.
        const size_t spWidth = dispatcher->REG_anySP.get_nbits();
        SValuePtr esp = SValue::promote(ops->readRegister(dispatcher->REG_anySP));
        SValuePtr stack_delta = SValue::promote(ops->add(esp, ops->negate(orig_esp)));
        SValuePtr stack_delta_sign = SValue::promote(ops->extract(stack_delta, spWidth-1, spWidth));
        if (stack_delta_sign->is_number() && 0==stack_delta_sign->get_number())
            return false;

        // If the top of the stack does not contain a concrete value or the top of the stack does not point to an instruction
        // in this basic block's function, then this is not a function call.
        const size_t ipWidth = dispatcher->REG_anyIP.get_nbits();
        SValuePtr top = SValue::promote(ops->readMemory(dispatcher->REG_SS, esp, esp->undefined_(ipWidth), esp->boolean_(true)));
        if (top->is_number()) {
            rose_addr_t va = top->get_number();
            SgAsmFunction *return_func = SageInterface::getEnclosingNode<SgAsmFunction>(imap.get_value_or(va, NULL));
            if (!return_func || return_func!=func) {
                return false;
            }
        } else {
            return false;
        }

        // Since EIP might point to a function entry address and since the top of the stack contains a pointer to an
        // instruction in this function, we assume that this is a function call.
        if (target && eip->is_number())
            *target = eip->get_number();
        if (return_va && top->is_number())
            *return_va = top->get_number();
        return true;
    }

    // Similar to the above method, but works when all we have is the basic block (e.g., this case gets hit quite a bit from
    // the Partitioner).  Returns true if, after executing the basic block, the top of the stack contains the fall-through
    // address of the basic block. We depend on our caller to figure out if EIP is reasonably a function entry address.
    if (!interp && insns.size()<=EXECUTION_LIMIT) {
        using namespace Rose::BinaryAnalysis;
        using namespace Rose::BinaryAnalysis::InstructionSemantics2;
        using namespace Rose::BinaryAnalysis::InstructionSemantics2::SymbolicSemantics;
        SmtSolverPtr solver = SmtSolver::instance(Rose::CommandLine::genericSwitchArgs.smtSolver);
        SgAsmX86Instruction *x86insn = isSgAsmX86Instruction(insns.front());
        ASSERT_not_null(x86insn);
#if 1 // [Robb P. Matzke 2015-03-03]: FIXME[Robb P. Matzke 2015-03-03]: not ready yet; x86-64 semantics still under construction
        if (x86insn->get_addressSize() != x86_insnsize_32)
            return false;
#endif
        const RegisterDictionary *regdict = registersForInstructionSize(x86insn->get_addressSize());
        const RegisterDescriptor SP = regdict->findLargestRegister(x86_regclass_gpr, x86_gpr_sp);
        BaseSemantics::RiscOperatorsPtr ops = RiscOperators::instance(regdict, solver);
        DispatcherX86Ptr dispatcher = DispatcherX86::instance(ops, SP.get_nbits());
        try {
            for (size_t i=0; i<insns.size(); ++i)
                dispatcher->processInstruction(insns[i]);
        } catch (const BaseSemantics::Exception &e) {
            return false;
        }

        // Look at the top of the stack
        const size_t ipWidth = dispatcher->REG_anyIP.get_nbits();
        SValuePtr top = SValue::promote(ops->readMemory(dispatcher->REG_SS, ops->readRegister(SP),
                                                        ops->protoval()->undefined_(ipWidth),
                                                        ops->protoval()->boolean_(true)));
        if (top->is_number() && top->get_number() == last->get_address()+last->get_size()) {
            if (target) {
                SValuePtr eip = SValue::promote(ops->readRegister(dispatcher->REG_anyIP));
                if (eip->is_number())
                    *target = eip->get_number();
            }
            if (return_va)
                *return_va = top->get_number();
            return true;
        }
    }

    return false;
}
// see base class; don't modify target_va or return_va if they are not known
bool
SgAsmM68kInstruction::isFunctionCallSlow(const std::vector<SgAsmInstruction*>& insns, rose_addr_t *target_va,
                                         rose_addr_t *return_va)
{
    if (isFunctionCallFast(insns, target_va, return_va))
        return true;

    static const size_t EXECUTION_LIMIT = 25; // max size of basic blocks for expensive analyses
    if (insns.empty())
        return false;
    SgAsmM68kInstruction *last = isSgAsmM68kInstruction(insns.back());
    if (!last)
        return false;
    SgAsmFunction *func = SageInterface::getEnclosingNode<SgAsmFunction>(last);
    SgAsmInterpretation *interp = SageInterface::getEnclosingNode<SgAsmInterpretation>(func);

    // Slow method: Emulate the instructions and then look at the program counter (PC) and stack (A7).  If the PC points
    // outside the current function and the top of the stack holds an address of an instruction within the current function,
    // then this must be a function call.
    if (interp && insns.size()<=EXECUTION_LIMIT) {
        using namespace Rose::BinaryAnalysis;
        using namespace Rose::BinaryAnalysis::InstructionSemantics2;
        using namespace Rose::BinaryAnalysis::InstructionSemantics2::SymbolicSemantics;
        const InstructionMap &imap = interp->get_instruction_map();
        const RegisterDictionary *regdict = RegisterDictionary::dictionary_for_isa(interp);
        SmtSolverPtr solver = SmtSolver::instance(Rose::CommandLine::genericSwitchArgs.smtSolver);
        BaseSemantics::RiscOperatorsPtr ops = RiscOperators::instance(regdict, solver);
        DispatcherM68kPtr dispatcher = DispatcherM68k::instance(ops, 32);
        SValuePtr orig_sp = SValue::promote(ops->readRegister(dispatcher->REG_A[7]));
        try {
            for (size_t i=0; i<insns.size(); ++i)
                dispatcher->processInstruction(insns[i]);
        } catch (const BaseSemantics::Exception &e) {
            return false;
        }

        // If the next instruction address is concrete but does not point to a function entry point, then this is not a call.
        SValuePtr ip = SValue::promote(ops->readRegister(dispatcher->REG_PC));
        if (ip->is_number()) {
            rose_addr_t target_va = ip->get_number();
            SgAsmFunction *target_func = SageInterface::getEnclosingNode<SgAsmFunction>(imap.get_value_or(target_va, NULL));
            if (!target_func || target_va!=target_func->get_entry_va())
                return false;
        }

        // If nothing was pushed onto the stack, then this isn't a function call.
        SValuePtr sp = SValue::promote(ops->readRegister(dispatcher->REG_A[7]));
        SValuePtr stack_delta = SValue::promote(ops->add(sp, ops->negate(orig_sp)));
        SValuePtr stack_delta_sign = SValue::promote(ops->extract(stack_delta, 31, 32));
        if (stack_delta_sign->is_number() && 0==stack_delta_sign->get_number())
            return false;

        // If the top of the stack does not contain a concrete value or the top of the stack does not point to an instruction
        // in this basic block's function, then this is not a function call.
        SValuePtr top = SValue::promote(ops->readMemory(RegisterDescriptor(), sp, sp->undefined_(32), sp->boolean_(true)));
        if (top->is_number()) {
            rose_addr_t va = top->get_number();
            SgAsmFunction *return_func = SageInterface::getEnclosingNode<SgAsmFunction>(imap.get_value_or(va, NULL));
            if (!return_func || return_func!=func) {
                return false;
            }
        } else {
            return false;
        }

        // Since the instruction pointer might point to a function entry address and since the top of the stack contains a
        // pointer to an instruction in this function, we assume that this is a function call.
        if (target_va && ip->is_number())
            *target_va = ip->get_number();
        if (return_va && top->is_number())
            *return_va = top->get_number();
        return true;
    }

    // Similar to the above method, but works when all we have is the basic block (e.g., this case gets hit quite a bit from
    // the Partitioner).  Returns true if, after executing the basic block, the top of the stack contains the fall-through
    // address of the basic block. We depend on our caller to figure out if the instruction pointer is reasonably a function
    // entry address.
    if (!interp && insns.size()<=EXECUTION_LIMIT) {
        using namespace Rose::BinaryAnalysis;
        using namespace Rose::BinaryAnalysis::InstructionSemantics2;
        using namespace Rose::BinaryAnalysis::InstructionSemantics2::SymbolicSemantics;
        const RegisterDictionary *regdict = RegisterDictionary::dictionary_coldfire_emac();
        SmtSolverPtr solver = SmtSolver::instance(Rose::CommandLine::genericSwitchArgs.smtSolver);
        BaseSemantics::RiscOperatorsPtr ops = RiscOperators::instance(regdict, solver);
        DispatcherM68kPtr dispatcher = DispatcherM68k::instance(ops, 32);
        try {
            for (size_t i=0; i<insns.size(); ++i)
                dispatcher->processInstruction(insns[i]);
        } catch (const BaseSemantics::Exception &e) {
            return false;
        }

        // Look at the top of the stack
        SValuePtr top = SValue::promote(ops->readMemory(RegisterDescriptor(), ops->readRegister(dispatcher->REG_A[7]),
                                                        ops->protoval()->undefined_(32),
                                                        ops->protoval()->boolean_(true)));
        if (top->is_number() && top->get_number() == last->get_address()+last->get_size()) {
            if (target_va) {
                SValuePtr ip = SValue::promote(ops->readRegister(dispatcher->REG_PC));
                if (ip->is_number())
                    *target_va = ip->get_number();
            }
            if (return_va)
                *return_va = top->get_number();
            return true;
        }
    }

    return false;
}