Merge remote-tracking branch 'origin/GP-4782_PtrsubUndo'

This commit is contained in:
Ryan Kurtz 2024-08-26 12:35:12 -04:00
commit aac3e5ad1c
7 changed files with 349 additions and 74 deletions

View file

@ -5920,30 +5920,30 @@ bool AddTreeState::spanAddTree(PcodeOp *op,uint8 treeCoeff)
void AddTreeState::calcSubtype(void)
{
if (size == 0 || nonmultsum < size)
offset = nonmultsum;
uint8 tmpoff = (multsum + nonmultsum) & ptrmask;
if (size == 0 || tmpoff < size)
offset = tmpoff;
else {
// For a sum that falls completely outside the data-type, there is presumably some
// type of constant term added to an array index either at the current level or lower.
// If we knew here whether an array of the baseType was possible we could make a slightly
// better decision.
intb snonmult = sign_extend(nonmultsum,ptrsize*8-1);
snonmult = snonmult % size;
if (snonmult >= 0)
intb stmpoff = sign_extend(tmpoff,ptrsize*8-1);
stmpoff = stmpoff % size;
if (stmpoff >= 0)
// We assume the sum is big enough it represents an array index at this level
offset = (uint8)snonmult;
offset = (uint8)stmpoff;
else {
// For a negative sum, if the baseType is a structure and there is array hints,
// we assume the sum is an array index at a lower level
if (baseType->getMetatype() == TYPE_STRUCT && biggestNonMultCoeff != 0)
offset = nonmultsum;
if (baseType->getMetatype() == TYPE_STRUCT && biggestNonMultCoeff != 0 && multsum == 0)
offset = tmpoff;
else
offset = (uint8)(snonmult + size);
offset = (uint8)(stmpoff + size);
}
}
correct = nonmultsum - offset;
nonmultsum = offset;
multsum = (multsum + correct) & ptrmask; // Some extra multiples of size
correct = nonmultsum; // Non-multiple constants are double counted, correct in final sum
multsum = (tmpoff - offset) & ptrmask; // Some extra multiples of size
if (nonmult.empty()) {
if ((multsum == 0) && multiple.empty()) { // Is there anything at all
valid = false;
@ -5952,31 +5952,33 @@ void AddTreeState::calcSubtype(void)
isSubtype = false; // There are no offsets INTO the pointer
}
else if (baseType->getMetatype() == TYPE_SPACEBASE) {
int8 nonmultbytes = AddrSpace::addressToByteInt(nonmultsum,ct->getWordSize()); // Convert to bytes
int8 offsetbytes = AddrSpace::addressToByteInt(offset,ct->getWordSize()); // Convert to bytes
int8 extra;
// Get offset into mapped variable
if (!hasMatchingSubType(nonmultbytes, biggestNonMultCoeff, &extra)) {
if (!hasMatchingSubType(offsetbytes, biggestNonMultCoeff, &extra)) {
valid = false; // Cannot find mapped variable but nonmult is non-empty
return;
}
extra = AddrSpace::byteToAddress(extra, ct->getWordSize()); // Convert back to address units
offset = (nonmultsum - extra) & ptrmask;
offset = (offset - extra) & ptrmask;
correct = (correct - extra) & ptrmask;
isSubtype = true;
}
else if (baseType->getMetatype() == TYPE_STRUCT) {
intb snonmult = sign_extend(nonmultsum,ptrsize*8-1);
int8 nonmultbytes = AddrSpace::addressToByteInt(snonmult,ct->getWordSize()); // Convert to bytes
intb soffset = sign_extend(offset,ptrsize*8-1);
int8 offsetbytes = AddrSpace::addressToByteInt(soffset,ct->getWordSize()); // Convert to bytes
int8 extra;
// Get offset into field in structure
if (!hasMatchingSubType(nonmultbytes, biggestNonMultCoeff, &extra)) {
if (nonmultbytes < 0 || nonmultbytes >= baseType->getSize()) { // Compare as bytes! not address units
if (!hasMatchingSubType(offsetbytes, biggestNonMultCoeff, &extra)) {
if (offsetbytes < 0 || offsetbytes >= baseType->getSize()) { // Compare as bytes! not address units
valid = false; // Out of structure's bounds
return;
}
extra = 0; // No field, but pretend there is something there
}
extra = AddrSpace::byteToAddressInt(extra, ct->getWordSize()); // Convert back to address units
offset = (nonmultsum - extra) & ptrmask;
offset = (offset - extra) & ptrmask;
correct = (correct - extra) & ptrmask;
if (pRelType != (TypePointerRel *)0 && offset == pRelType->getPointerOffset()) {
// offset falls within basic ptrto
if (!pRelType->evaluateThruParent(0)) { // If we are not representing offset 0 through parent
@ -5988,12 +5990,31 @@ void AddTreeState::calcSubtype(void)
}
else if (baseType->getMetatype() == TYPE_ARRAY) {
isSubtype = true;
correct = (correct - offset) & ptrmask;
offset = 0;
}
else {
// No struct or array, but nonmult is non-empty
valid = false; // There is substructure we don't know about
}
if (pRelType != (const TypePointerRel *)0) {
int4 ptrOff = ((TypePointerRel *)ct)->getPointerOffset();
offset = (offset - ptrOff) & ptrmask;
correct = (correct - ptrOff) & ptrmask;
}
}
/// The data-type from the pointer input (of either a PTRSUB or PTRADD) is propagated to the
/// output of the PcodeOp.
/// \param op is the given PcodeOp
void AddTreeState::assignPropagatedType(PcodeOp *op)
{
Varnode *vn = op->getIn(0);
Datatype *inType = vn->getTypeReadFacing(op);
Datatype *newType = op->getOpcode()->propagateType(inType, op, vn, op->getOut(), 0, -1);
if (newType != (Datatype *)0)
op->getOut()->updateType(newType, false, false);
}
/// Construct part of the tree that sums to a multiple of the base data-type size.
@ -6037,7 +6058,6 @@ Varnode *AddTreeState::buildMultiples(void)
Varnode *AddTreeState::buildExtra(void)
{
correct = correct+offset; // Total correction that needs to be made
Varnode *resNode = (Varnode *)0;
for(int4 i=0;i<nonmult.size();++i) {
Varnode *vn = nonmult[i];
@ -6138,11 +6158,6 @@ bool AddTreeState::apply(void)
void AddTreeState::buildTree(void)
{
if (pRelType != (const TypePointerRel *)0) {
int4 ptrOff = ((TypePointerRel *)ct)->getPointerOffset();
offset -= ptrOff;
offset &= ptrmask;
}
Varnode *multNode = buildMultiples();
Varnode *extraNode = buildExtra();
PcodeOp *newop = (PcodeOp *)0;
@ -6152,6 +6167,8 @@ void AddTreeState::buildTree(void)
newop = data.newOpBefore(baseOp,CPUI_PTRADD,ptr,multNode,data.newConstant(ptrsize,size));
if (ptr->getType()->needsResolution())
data.inheritResolution(ptr->getType(),newop, 0, baseOp, baseSlot);
if (data.isTypeRecoveryExceeded())
assignPropagatedType(newop);
multNode = newop->getOut();
}
else
@ -6162,6 +6179,8 @@ void AddTreeState::buildTree(void)
newop = data.newOpBefore(baseOp,CPUI_PTRSUB,multNode,data.newConstant(ptrsize,offset));
if (multNode->getType()->needsResolution())
data.inheritResolution(multNode->getType(),newop, 0, baseOp, baseSlot);
if (data.isTypeRecoveryExceeded())
assignPropagatedType(newop);
if (size != 0)
newop->setStopTypePropagation();
multNode = newop->getOut();
@ -6555,6 +6574,8 @@ int4 RulePtraddUndo::applyOp(PcodeOp *op,Funcdata &data)
return 1;
}
const int4 RulePtrsubUndo::DEPTH_LIMIT = 8;
/// \class RulePtrsubUndo
/// \brief Remove PTRSUB operations with mismatched data-type information
///
@ -6567,17 +6588,204 @@ void RulePtrsubUndo::getOpList(vector<uint4> &oplist) const
oplist.push_back(CPUI_PTRSUB);
}
/// \brief Recursively search for additive constants and multiplicative constants
///
/// Walking backward from the given Varnode, search for constants being added in and return
/// the sum of all the constants. Additionally pass back the biggest constant coefficient, for any term
/// formed with INT_MULT.
/// \param vn is the given root Varnode of the additive tree
/// \param multiplier will hold the biggest constant coefficient
/// \param maxLevel is the maximum depth to search in the tree
/// \return the sum of all constants in the additive expression
int8 RulePtrsubUndo::getConstOffsetBack(Varnode *vn,int8 &multiplier,int4 maxLevel)
{
multiplier = 1;
int8 submultiplier;
if (vn->isConstant())
return vn->getOffset();
if (!vn->isWritten())
return 0;
maxLevel -= 1;
if (maxLevel < 0)
return 0;
PcodeOp *op = vn->getDef();
OpCode opc = op->code();
int8 retval = 0;
if (opc == CPUI_INT_ADD) {
retval += getConstOffsetBack(op->getIn(0),submultiplier,maxLevel);
if (submultiplier > multiplier)
multiplier = submultiplier;
retval += getConstOffsetBack(op->getIn(1), submultiplier, maxLevel);
if (submultiplier > multiplier)
multiplier = submultiplier;
}
else if (opc == CPUI_INT_MULT) {
Varnode *cvn = op->getIn(1);
if (!cvn->isConstant()) return 0;
multiplier = cvn->getOffset();
getConstOffsetBack(op->getIn(0), submultiplier, maxLevel);
multiplier *= submultiplier; // Only contribute to the multiplier
}
return retval;
}
/// \brief Collect constants and the biggest multiplier in the given PTRSUB expression.
///
/// Walking the additive expression (INT_ADD, PTRADD, and other PTRSUBs) and calculate any additional
/// constant value being added to the PTRSUB. Additionally pass back the biggest constant coefficient of any
/// multiplicative term in the expression.
/// \param op is the given PTRSUB
/// \param multiplier will hold the biggest multiplicative coefficient
int8 RulePtrsubUndo::getExtraOffset(PcodeOp *op,int8 &multiplier)
{
int8 extra = 0;
multiplier = 1;
int8 submultiplier;
Varnode *outvn = op->getOut();
op = outvn->loneDescend();
while(op != (PcodeOp *)0) {
OpCode opc = op->code();
if (opc == CPUI_INT_ADD) {
int4 slot = op->getSlot(outvn);
extra += getConstOffsetBack(op->getIn(1-slot),submultiplier,DEPTH_LIMIT); // Get any constants from other input
if (submultiplier > multiplier)
multiplier = submultiplier;
}
else if (opc == CPUI_PTRSUB) {
extra += op->getIn(1)->getOffset();
}
else if (opc == CPUI_PTRADD) {
if (op->getIn(0) != outvn) break;
int8 ptraddmult = op->getIn(2)->getOffset();
Varnode *invn = op->getIn(1);
if (invn->isConstant()) // Only contribute to the extra
extra += ptraddmult * (int8)invn->getOffset(); // if the index is constant
getConstOffsetBack(invn,submultiplier,DEPTH_LIMIT); // otherwise just contribute to multiplier
submultiplier *= ptraddmult;
if (submultiplier > multiplier)
multiplier = submultiplier;
}
else {
break;
}
outvn = op->getOut();
op = outvn->loneDescend();
}
extra = sign_extend(extra, 8*outvn->getSize()-1);
extra &= calc_mask(outvn->getSize());
return extra;
}
/// \brief Remove any constants in the additive expression rooted at the given PcodeOp
///
/// Walking recursively through the expression, any INT_ADD with a constant input is converted to
/// a COPY. The INT_ADD must only contribute to the root expression.
/// \param op is the given root PcodeOp
/// \param slot is the input slot to walk back from
/// \param maxLevel is the maximum depth to recurse
/// \param data is the function containing the expression
/// \return the sum of all constants that are removed
int8 RulePtrsubUndo::removeLocalAddRecurse(PcodeOp *op,int4 slot,int4 maxLevel,Funcdata &data)
{
Varnode *vn = op->getIn(slot);
if (!vn->isWritten())
return 0;
if (vn->loneDescend() != op)
return 0; // Varnode must not be used anywhere else
maxLevel -= 1;
if (maxLevel < 0)
return 0;
op = vn->getDef();
int8 retval = 0;
if (op->code() == CPUI_INT_ADD) {
if (op->getIn(1)->isConstant()) {
retval += (int8)op->getIn(1)->getOffset();
data.opRemoveInput(op, 1);
data.opSetOpcode(op, CPUI_COPY);
}
else {
retval += removeLocalAddRecurse(op, 0, maxLevel, data);
retval += removeLocalAddRecurse(op, 1, maxLevel, data);
}
}
return retval;
}
/// \brief Remove constants in the additive expression involving the given Varnode
///
/// Any additional PTRADD, PTRSUB, or INT_ADD that uses the Varnode and adds a constant is converted
/// to a COPY. Additionally any other INT_ADD involved in the expression that adds a constant is
/// also converted to COPY.
/// \param vn is the given Varnode
/// \param data is the function containing the expression
/// \return the sum of all constants that are removed
int8 RulePtrsubUndo::removeLocalAdds(Varnode *vn,Funcdata &data)
{
int8 extra = 0;
PcodeOp *op = vn->loneDescend();
while(op != (PcodeOp *)0) {
OpCode opc = op->code();
if (opc == CPUI_INT_ADD) {
int4 slot = op->getSlot(vn);
if (slot == 0 && op->getIn(1)->isConstant()) {
extra += (int8)op->getIn(1)->getOffset();
data.opRemoveInput(op, 1);
data.opSetOpcode(op, CPUI_COPY);
}
else {
extra += removeLocalAddRecurse(op,1-slot,DEPTH_LIMIT, data); // Get any constants from other input
}
}
else if (opc == CPUI_PTRSUB) {
extra += op->getIn(1)->getOffset();
op->clearStopTypePropagation();
data.opRemoveInput(op, 1);
data.opSetOpcode(op, CPUI_COPY);
}
else if (opc == CPUI_PTRADD) {
if (op->getIn(0) != vn) break;
int8 ptraddmult = op->getIn(2)->getOffset();
Varnode *invn = op->getIn(1);
if (invn->isConstant()) {
extra += ptraddmult * (int8)invn->getOffset();
data.opRemoveInput(op,2);
data.opRemoveInput(op,1);
data.opSetOpcode(op, CPUI_COPY);
}
}
else {
break;
}
vn = op->getOut();
op = vn->loneDescend();
}
return extra;
}
int4 RulePtrsubUndo::applyOp(PcodeOp *op,Funcdata &data)
{
if (!data.hasTypeRecoveryStarted()) return 0;
Varnode *basevn = op->getIn(0);
if (basevn->getTypeReadFacing(op)->isPtrsubMatching(op->getIn(1)->getOffset()))
Varnode *cvn = op->getIn(1);
int8 val = cvn->getOffset();
int8 multiplier;
int8 extra = getExtraOffset(op,multiplier);
if (basevn->getTypeReadFacing(op)->isPtrsubMatching(val,extra,multiplier))
return 0;
data.opSetOpcode(op,CPUI_INT_ADD);
op->clearStopTypePropagation();
extra = removeLocalAdds(op->getOut(),data);
if (extra != 0) {
val = val + extra; // Lump extra into additive offset
data.opSetInput(op,data.newConstant(cvn->getSize(), val & calc_mask(cvn->getSize())),1);
}
return 1;
}