void CountTraversal::visit ( SgNode* n )
{
SgAsmInstruction* asmInstruction = isSgAsmInstruction(n);
if (asmInstruction != NULL)
{
// Use the new interface support for this (this detects all multi-byte nop instructions).
if (SageInterface::isNOP(asmInstruction) == true)
{
if (previousInstructionWasNop == true)
{
// Increment the length of the identified NOP sequence
count++;
}
else
{
count = 1;
// Record the starting address of the NOP sequence
nopSequenceStart = asmInstruction;
}
previousInstructionWasNop = true;
}
else
{
if (count > 0)
{
// Report the sequence when we have detected the end of the sequence.
SgAsmFunction* functionDeclaration = getAsmFunction(asmInstruction);
printf ("Reporting NOP sequence of length %3d at address %zu in function %s (reason for this being a function = %u = %s) \n",
count,nopSequenceStart->get_address(),functionDeclaration->get_name().c_str(),
functionDeclaration->get_reason(),
stringifySgAsmFunctionFunctionReason(functionDeclaration->get_reason()).c_str());
nopSequences.push_back(pair<SgAsmInstruction*,int>(nopSequenceStart,count));
SgAsmBlock* block = isSgAsmBlock(nopSequenceStart->get_parent());
ROSE_ASSERT(block != NULL);
SgAsmStatementPtrList & l = block->get_statementList();
// Now iterate over the nop instructions in the sequence and report the lenght of each (can be multi-byte nop instructions).
SgAsmStatementPtrList::iterator i = find(l.begin(),l.end(),nopSequenceStart);
ROSE_ASSERT(i != l.end());
int counter = 0;
while ( (*i != asmInstruction) && (i != l.end()) )
{
printf ("--- NOP #%2d is length = %2d \n",counter++,(int)isSgAsmInstruction(*i)->get_raw_bytes().size());
i++;
}
}
count = 0;
previousInstructionWasNop = false;
}
}
}
void visit(SgNode *node) {
if (SgAsmInstruction *insn = isSgAsmInstruction(node)) {
SgAsmFunction *func = SageInterface::getEnclosingNode<SgAsmFunction>(insn);
if (func && 0==(func->get_reason() & SgAsmFunction::FUNC_LEFTOVERS))
insns.insert(std::make_pair(insn->get_address(), insn));
}
}
bool operator()(ControlFlow *analyzer, SgAsmNode *node) {
SgAsmFunction *func = SageInterface::getEnclosingNode<SgAsmFunction>(node, true);
return func && 0==(func->get_reason() & SgAsmFunction::FUNC_LEFTOVERS);
}
int main(int argc, char** argv) {
std::string binaryFilename = (argc >= 1 ? argv[argc-1] : "" );
std::vector<std::string> newArgv(argv,argv+argc);
newArgv.push_back("-rose:output");
newArgv.push_back(binaryFilename+"-binarySemantics.C");
SgProject* proj = frontend(newArgv);
ROSE_ASSERT (proj);
SgSourceFile* newFile = isSgSourceFile(proj->get_fileList().front());
ROSE_ASSERT(newFile != NULL);
SgGlobal* g = newFile->get_globalScope();
ROSE_ASSERT (g);
//I am doing some experimental work to enable functions in the C representation
//Set this flag to true in order to enable that work
bool enable_functions = true;
//Jeremiah did some work to enable a simplification and normalization of the
//C representation. Enable this work by setting this flag to true.
bool enable_normalizations = false;
vector<SgNode*> asmFiles = NodeQuery::querySubTree(proj, V_SgAsmGenericFile);
ROSE_ASSERT (asmFiles.size() == 1);
if( enable_functions == false)
{
//Representation of C normalizations withotu functions
SgFunctionDeclaration* decl = buildDefiningFunctionDeclaration("run", SgTypeVoid::createType(), buildFunctionParameterList(), g);
appendStatement(decl, g);
SgBasicBlock* body = decl->get_definition()->get_body();
// ROSE_ASSERT(isSgAsmFile(asmFiles[0]));
// X86CTranslationPolicy policy(newFile, isSgAsmFile(asmFiles[0]));
X86CTranslationPolicy policy(newFile, isSgAsmGenericFile(asmFiles[0]));
ROSE_ASSERT( isSgAsmGenericFile(asmFiles[0]) != NULL);
policy.switchBody = buildBasicBlock();
removeDeadStores(policy.switchBody,policy);
SgSwitchStatement* sw = buildSwitchStatement(buildVarRefExp(policy.ipSym), policy.switchBody);
ROSE_ASSERT(isSgBasicBlock(sw->get_body()));
SgWhileStmt* whileStmt = buildWhileStmt(buildBoolValExp(true), sw);
appendStatement(whileStmt, body);
policy.whileBody = sw;
X86InstructionSemantics<X86CTranslationPolicy, WordWithExpression> t(policy);
//AS FIXME: This query gets noting in the form in the repository. Doing this hack since we only
//have one binary file anyways.
//vector<SgNode*> instructions = NodeQuery::querySubTree(asmFiles[0], V_SgAsmX86Instruction);
vector<SgNode*> instructions = NodeQuery::querySubTree(proj, V_SgAsmX86Instruction);
std::cout << "Instruction\n";
for (size_t i = 0; i < instructions.size(); ++i) {
SgAsmX86Instruction* insn = isSgAsmX86Instruction(instructions[i]);
ROSE_ASSERT (insn);
try {
t.processInstruction(insn);
} catch (const X86InstructionSemantics<X86CTranslationPolicy, WordWithExpression>::Exception &e) {
std::cout <<e.mesg <<": " <<unparseInstructionWithAddress(e.insn) <<"\n";
}
}
if ( enable_normalizations == true )
{
//Enable normalizations of C representation
//This is done heuristically where some steps
//are repeated. It is not clear which order is
//the best
{
plugInAllConstVarDefs(policy.switchBody,policy) ;
simplifyAllExpressions(policy.switchBody);
removeIfConstants(policy.switchBody);
removeDeadStores(policy.switchBody,policy);
removeUnusedVariables(policy.switchBody);
}
{
plugInAllConstVarDefs(policy.switchBody,policy) ;
simplifyAllExpressions(policy.switchBody);
removeIfConstants(policy.switchBody);
removeDeadStores(policy.switchBody,policy);
}
removeUnusedVariables(policy.switchBody);
}
}else{ //Experimental changes to introduce functions into the C representation
//When trying to add function I get that symbols are not defined
//Iterate over the functions separately
vector<SgNode*> asmFunctions = NodeQuery::querySubTree(proj, V_SgAsmFunction);
for(size_t j = 0; j < asmFunctions.size(); j++ )
{
SgAsmFunction* binFunc = isSgAsmFunction( asmFunctions[j] );
// Some functions (probably just one) are generated to hold basic blocks that could not
// be assigned to a particular function. This happens when the Disassembler is overzealous
// and the Partitioner cannot statically determine where the block belongs. The name of
// one such function is "***uncategorized blocks***". [matzke 2010-06-29]
if ((binFunc->get_reason() & SgAsmFunction::FUNC_LEFTOVERS))
continue;
//Some functions may be unnamed so we need to generate a name for those
std::string funcName;
if (binFunc->get_name().size()==0) {
char addr_str[64];
sprintf(addr_str, "0x%"PRIx64, binFunc->get_statementList()[0]->get_address());
funcName = std::string("my_") + addr_str;;
} else {
funcName = "my" + binFunc->get_name();
}
//Functions can have illegal characters in their name. Need to replace those characters
for ( int i = 0 ; i < funcName.size(); i++ )
{
char& currentCharacter = funcName.at(i);
if ( currentCharacter == '.' )
currentCharacter = '_';
}
SgFunctionDeclaration* decl = buildDefiningFunctionDeclaration(funcName, SgTypeVoid::createType(), buildFunctionParameterList(), g);
appendStatement(decl, g);
SgBasicBlock* body = decl->get_definition()->get_body();
X86CTranslationPolicy policy(newFile, isSgAsmGenericFile(asmFiles[0]));
ROSE_ASSERT( isSgAsmGenericFile(asmFiles[0]) != NULL);
policy.switchBody = buildBasicBlock();
SgSwitchStatement* sw = buildSwitchStatement(buildVarRefExp(policy.ipSym), policy.switchBody);
SgWhileStmt* whileStmt = buildWhileStmt(buildBoolValExp(true), sw);
appendStatement(whileStmt, body);
policy.whileBody = sw;
X86InstructionSemantics<X86CTranslationPolicy, WordWithExpression> t(policy);
vector<SgNode*> instructions = NodeQuery::querySubTree(binFunc, V_SgAsmX86Instruction);
for (size_t i = 0; i < instructions.size(); ++i) {
SgAsmX86Instruction* insn = isSgAsmX86Instruction(instructions[i]);
if( insn->get_kind() == x86_nop )
continue;
ROSE_ASSERT (insn);
try {
t.processInstruction(insn);
} catch (const X86InstructionSemantics<X86CTranslationPolicy, WordWithExpression>::Exception &e) {
std::cout <<e.mesg <<": " <<unparseInstructionWithAddress(e.insn) <<"\n";
}
}
}
//addDirectJumpsToSwitchCases(policy);
}
proj->get_fileList().erase(proj->get_fileList().end() - 1); // Remove binary file before calling backend
// AstTests::runAllTests(proj);
//Compile the resulting project
return backend(proj);
}