Split a p-code COPY, LOAD, or STORE op based on underlying composite data-type. More...

#include <subflow.hh>

Classes
class	Component
	A helper class describing a pair of matching data-types for the split. More...

class	RootPointer
	A helper class describing the pointer being passed to a LOAD or STORE. More...

Public Member Functions
	SplitDatatype (Funcdata &func)
	Constructor.

bool	splitCopy (PcodeOp copyOp, Datatype inType, Datatype *outType)
	Split a COPY operation. More...

bool	splitLoad (PcodeOp loadOp, Datatype inType)
	Split a LOAD operation. More...

bool	splitStore (PcodeOp storeOp, Datatype outType)
	Split a STORE operation. More...

Static Public Member Functions
static Datatype *	getValueDatatype (PcodeOp loadStore, int4 size, TypeFactory tlst)
	Get a data-type description of the value being pointed at by the given LOAD or STORE. More...

Private Member Functions
Datatype *	getComponent (Datatype *ct, int4 offset, bool &isHole)
	Obtain the component of the given data-type at the specified offset. More...

int4	categorizeDatatype (Datatype *ct)
	Categorize if and how data-type should be split. More...

bool	testDatatypeCompatibility (Datatype inBase, Datatype outBase, bool inConstant)
	Can the two given data-types be mutually split into matching logical components. More...

bool	testCopyConstraints (PcodeOp *copyOp)
	Test specific constraints for splitting the given COPY operation into pieces. More...

bool	generateConstants (Varnode vn, vector< Varnode > &inVarnodes)
	If the given Varnode is an extended precision constant, create split constants. More...

void	buildInConstants (Varnode rootVn, vector< Varnode > &inVarnodes)
	Assuming the input is a constant, build split constants. More...

void	buildInSubpieces (Varnode rootVn, PcodeOp followOp, vector< Varnode *> &inVarnodes)
	Build input Varnodes by extracting SUBPIECEs from the root. More...

void	buildOutVarnodes (Varnode rootVn, vector< Varnode > &outVarnodes)
	Build output Varnodes with storage based on the given root. More...

void	buildOutConcats (Varnode rootVn, PcodeOp previousOp, vector< Varnode *> &outVarnodes)
	Concatenate output Varnodes into given root Varnode. More...

void	buildPointers (Varnode rootVn, TypePointer ptrType, int4 baseOffset, PcodeOp followOp, vector< Varnode > &ptrVarnodes, bool isInput)
	Build a a series of PTRSUB ops at different offsets, given a root pointer. More...

Static Private Member Functions
static bool	isArithmeticInput (Varnode *vn)
	Is this the input to an arithmetic operation. More...

static bool	isArithmeticOutput (Varnode *vn)
	Is this defined by an arithmetic operation. More...

Private Attributes
Funcdata &	data
	The containing function.

TypeFactory *	types
	The data-type container.

vector< Component >	dataTypePieces
	Sequence of all data-type pairs being copied.

bool	splitStructures
	Whether or not structures should be split.

bool	splitArrays
	Whether or not arrays should be split.

Detailed Description

Split a p-code COPY, LOAD, or STORE op based on underlying composite data-type.

During the cleanup phase, if a COPY, LOAD, or STORE occurs on a partial structure or array (TypePartialStruct), try to break it up into multiple operations that each act on logical component of the structure or array.

Member Function Documentation

◆ buildInConstants()

void ghidra::SplitDatatype::buildInConstants	(	Varnode *	rootVn,
		vector< Varnode *> &	inVarnodes
	)

private

Assuming the input is a constant, build split constants.

Build constant input Varnodes, extracting the constant value from the given root constant based on the input offsets in dataTypePieces.

Parameters

rootVn	is the given root constant
inVarnodes	is the container for the new Varnodes

References ghidra::calc_mask(), ghidra::Varnode::getOffset(), ghidra::Varnode::getSize(), ghidra::Datatype::getSize(), ghidra::Varnode::getSpace(), and ghidra::AddrSpace::isBigEndian().

Referenced by splitCopy(), and splitStore().

◆ buildInSubpieces()

void ghidra::SplitDatatype::buildInSubpieces	(	Varnode *	rootVn,
		PcodeOp *	followOp,
		vector< Varnode *> &	inVarnodes
	)

private

Build input Varnodes by extracting SUBPIECEs from the root.

Extract different pieces from the given root based on the offsets and input data-types in dataTypePieces.

Parameters

rootVn	is the given root Varnode
followOp	is the point at which the SUBPIECEs should be inserted (before)
inVarnodes	is the container for the new Varnodes

References ghidra::CPUI_SUBPIECE, ghidra::PcodeOp::getAddr(), ghidra::Varnode::getAddr(), ghidra::Varnode::getSize(), ghidra::Datatype::getSize(), ghidra::Address::isBigEndian(), ghidra::Address::renormalize(), and ghidra::Varnode::updateType().

Referenced by splitCopy(), and splitStore().

◆ buildOutConcats()

void ghidra::SplitDatatype::buildOutConcats	(	Varnode *	rootVn,
		PcodeOp *	previousOp,
		vector< Varnode *> &	outVarnodes
	)

private

Concatenate output Varnodes into given root Varnode.

Insert PIECE operators concatenating all output Varnodes from most significant to least significant producing the root Varnode as the final result.

Parameters

rootVn	is the given root Varnode
previousOp	is the point at which to insert (after)
outVarnodes	is the list of output Varnodes

References ghidra::CPUI_PIECE, ghidra::PcodeOp::getAddr(), ghidra::Varnode::getAddr(), ghidra::Varnode::getSize(), ghidra::Varnode::hasNoDescend(), ghidra::Varnode::isAddrTied(), ghidra::Address::isBigEndian(), ghidra::Address::renormalize(), ghidra::PcodeOp::setPartialRoot(), and ghidra::Varnode::setProtoPartial().

Referenced by splitCopy(), and splitLoad().

◆ buildOutVarnodes()

void ghidra::SplitDatatype::buildOutVarnodes	(	Varnode *	rootVn,
		vector< Varnode *> &	outVarnodes
	)

private

Build output Varnodes with storage based on the given root.

Extract different pieces from the given root based on the offsets and output data-types in dataTypePieces.

Parameters

rootVn	is the given root Varnode
inVarnodes	is the container for the new Varnodes

References ghidra::Varnode::getAddr(), ghidra::Datatype::getSize(), and ghidra::Address::renormalize().

Referenced by splitCopy(), and splitLoad().

◆ buildPointers()

void ghidra::SplitDatatype::buildPointers	(	Varnode *	rootVn,
		TypePointer *	ptrType,
		int4	baseOffset,
		PcodeOp *	followOp,
		vector< Varnode *> &	ptrVarnodes,
		bool	isInput
	)

private

Build a a series of PTRSUB ops at different offsets, given a root pointer.

Offsets and data-types are based on dataTypePieces, taking input data-types if isInput is true, output data-types otherwise. The data-types, relative to the root pointer, are assumed to start at the given base offset.

Parameters

rootVn	is the root pointer
ptrType	is the pointer data-type associated with the root
baseOffset	is the given base offset
followOp	is the point at which the new PTRSUB ops are inserted (before)
ptrVarnodes	is the container for the new pointer Varnodes
isInput	specifies either input (true) or output (false) data-types

References ghidra::AddrSpace::byteToAddressInt(), ghidra::CPUI_PTRADD, ghidra::CPUI_PTRSUB, ghidra::PcodeOp::getAddr(), ghidra::Datatype::getMetatype(), ghidra::TypePointer::getPtrTo(), ghidra::Varnode::getSize(), ghidra::Datatype::getSize(), ghidra::Datatype::getSubType(), ghidra::TypePointer::getWordSize(), ghidra::TransformManager::newOp(), ghidra::TYPE_ARRAY, ghidra::TYPE_INT, and ghidra::Varnode::updateType().

Referenced by splitLoad(), and splitStore().

◆ categorizeDatatype()

int4 ghidra::SplitDatatype::categorizeDatatype ( Datatype * ct )

private

Categorize if and how data-type should be split.

For the given data-type, taking into account configuration options, return:

-1 for not splittable
0 for data-type that needs to be split
1 for data-type that can be split multiple ways
Parameters

ct is the given data-type

Returns
the categorization

References ghidra::Datatype::getMetatype(), ghidra::Datatype::getSize(), ghidra::Datatype::numDepend(), ghidra::TYPE_ARRAY, ghidra::TYPE_INT, ghidra::TYPE_PARTIALSTRUCT, ghidra::TYPE_STRUCT, ghidra::TYPE_UINT, and ghidra::TYPE_UNKNOWN.

◆ generateConstants()

bool ghidra::SplitDatatype::generateConstants	(	Varnode *	vn,
		vector< Varnode *> &	inVarnodes
	)

private

If the given Varnode is an extended precision constant, create split constants.

Look for ZEXT(#c) and CONCAT(#c1,#c2) forms. Try to split into single precision Varnodes.

Parameters

vn	is the given Varnode
inVarnodes	will contain the split constant Varnodes

Returns: true if the Varnode is an extended precision constant and the split is successful

References ghidra::calc_mask(), ghidra::PcodeOp::code(), ghidra::CPUI_INT_ZEXT, ghidra::CPUI_PIECE, ghidra::Varnode::getDef(), ghidra::PcodeOp::getIn(), ghidra::Varnode::getOffset(), ghidra::Varnode::getSize(), ghidra::Datatype::getSize(), ghidra::Varnode::getSpace(), ghidra::AddrSpace::isBigEndian(), ghidra::Varnode::isConstant(), ghidra::Varnode::isWritten(), and ghidra::Varnode::loneDescend().

◆ getComponent()

Datatype * ghidra::SplitDatatype::getComponent	(	Datatype *	ct,
		int4	offset,
		bool &	isHole
	)

private

Obtain the component of the given data-type at the specified offset.

The data-type must be a composite of some form. This method finds a component data-type starting exactly at the offset, if it exists. The component may be nested more than 1 level deep. If the given data-type is of composite form and has no component defined at the specified offset, an undefined data-type matching the size of the hole is returned and isHole is set to true.

Parameters

ct	is the given data-type
offset	is the specified offset
isHole	passes back whether a hole in the composite was encountered

Returns: the component data-type at the offset or null, if no such component exists

References ghidra::Datatype::getHoleSize(), ghidra::Datatype::getMetatype(), ghidra::Datatype::getSubType(), ghidra::TYPE_ARRAY, and ghidra::TYPE_UNKNOWN.

◆ getValueDatatype()

Datatype * ghidra::SplitDatatype::getValueDatatype	(	PcodeOp *	loadStore,
		int4	size,
		TypeFactory *	tlst
	)

static

Get a data-type description of the value being pointed at by the given LOAD or STORE.

Take the data-type of the pointer and construct the data-type of the thing being pointed at so that it matches a specific size. This takes into account TypePointerRel and can produce TypePartialStruct in order to match the size. If no interpretation of the value as a splittable data-type is possible, null is returned.

Parameters

loadStore	is the given LOAD or STORE
size	is the number of bytes in the value being pointed at
tlst	is the TypeFactory for constructing partial data-types if necessary

Returns: the data-type description of the value or null

References ghidra::AddrSpace::addressToByteInt(), ghidra::TypeFactory::getExactPiece(), ghidra::PcodeOp::getIn(), ghidra::Datatype::getMetatype(), ghidra::TypePointerRel::getParent(), ghidra::TypePointerRel::getPointerOffset(), ghidra::Varnode::getTypeReadFacing(), ghidra::TypePointer::getWordSize(), ghidra::Datatype::isPointerRel(), ghidra::TYPE_ARRAY, ghidra::TYPE_PTR, and ghidra::TYPE_STRUCT.

Referenced by ghidra::RuleSplitLoad::applyOp(), ghidra::RuleSplitStore::applyOp(), and splitStore().

◆ isArithmeticInput()

bool ghidra::SplitDatatype::isArithmeticInput ( Varnode * vn )

staticprivate

Is this the input to an arithmetic operation.

Iterate through descendants of the given Varnode, looking for arithmetic ops.

Parameters

vn	is the given Varnode

Returns: true if the Varnode has an arithmetic op as a descendant

References ghidra::Varnode::beginDescend(), ghidra::Varnode::endDescend(), ghidra::PcodeOp::getOpcode(), and ghidra::TypeOp::isArithmeticOp().

Referenced by splitCopy(), and splitLoad().

◆ isArithmeticOutput()

bool ghidra::SplitDatatype::isArithmeticOutput ( Varnode * vn )

staticprivate

Is this defined by an arithmetic operation.

Check if the defining PcodeOp is arithmetic.

Parameters

vn	is the given Varnode

Returns: true if the defining op is arithemetic

References ghidra::Varnode::getDef(), ghidra::PcodeOp::getOpcode(), ghidra::TypeOp::isArithmeticOp(), and ghidra::Varnode::isWritten().

Referenced by splitCopy(), and splitStore().

◆ splitCopy()

bool ghidra::SplitDatatype::splitCopy	(	PcodeOp *	copyOp,
		Datatype *	inType,
		Datatype *	outType
	)

Split a COPY operation.

Based on the input and output data-types, determine if and how the given COPY operation should be split into pieces. Then if possible, perform the split.

Parameters

copyOp	is the given COPY
inType	is the data-type of the COPY input
outType	is the data-type of the COPY output

Returns: true if the split was performed

References buildInConstants(), buildInSubpieces(), buildOutConcats(), buildOutVarnodes(), ghidra::CPUI_COPY, data, ghidra::PcodeOp::getAddr(), ghidra::PcodeOp::getIn(), ghidra::PcodeOp::getOut(), isArithmeticInput(), isArithmeticOutput(), ghidra::Varnode::isConstant(), ghidra::Funcdata::newOp(), ghidra::Funcdata::opDestroy(), ghidra::Funcdata::opInsertBefore(), ghidra::Funcdata::opSetInput(), ghidra::Funcdata::opSetOpcode(), ghidra::Funcdata::opSetOutput(), testCopyConstraints(), and testDatatypeCompatibility().

Referenced by ghidra::RuleSplitCopy::applyOp().

◆ splitLoad()

bool ghidra::SplitDatatype::splitLoad	(	PcodeOp *	loadOp,
		Datatype *	inType
	)

Split a LOAD operation.

Based on the LOAD data-type, determine if the given LOAD can be split into smaller LOADs. Then, if possible, perform the split. The input data-type describes the size and composition of the value being loaded. Check for the special case where, the LOAD output is a lone input to a COPY, and split the outputs of the COPY as well.

Parameters

loadOp	is the given LOAD to split
inType	is the data-type associated with the value being loaded

Returns: true if the split was performed

References ghidra::SplitDatatype::RootPointer::baseOffset, buildOutConcats(), buildOutVarnodes(), buildPointers(), ghidra::PcodeOp::code(), ghidra::CPUI_COPY, ghidra::CPUI_LOAD, ghidra::CPUI_STORE, data, ghidra::SplitDatatype::RootPointer::find(), ghidra::SplitDatatype::RootPointer::freePointerChain(), ghidra::PcodeOp::getAddr(), ghidra::PcodeOp::getIn(), ghidra::PcodeOp::getOut(), ghidra::Varnode::getSpaceFromConst(), ghidra::Varnode::getTypeDefFacing(), ghidra::Varnode::isAddrTied(), isArithmeticInput(), ghidra::Varnode::loneDescend(), ghidra::Funcdata::newOp(), ghidra::Funcdata::newVarnodeSpace(), ghidra::Funcdata::opDestroy(), ghidra::Funcdata::opInsertBefore(), ghidra::Funcdata::opSetInput(), ghidra::Funcdata::opSetOpcode(), ghidra::Funcdata::opSetOutput(), ghidra::SplitDatatype::RootPointer::pointer, ghidra::SplitDatatype::RootPointer::ptrType, and testDatatypeCompatibility().

Referenced by ghidra::RuleSplitLoad::applyOp().

◆ splitStore()

bool ghidra::SplitDatatype::splitStore	(	PcodeOp *	storeOp,
		Datatype *	outType
	)

Split a STORE operation.

Based on the STORE data-type, determine if the given STORE can be split into smaller STOREs. Then, if possible, perform the split. The output data-type describes the size and composition of the value being stored.

Parameters

storeOp	is the given STORE to split
outType	is the data-type associated with the value being stored

Returns: true if the split was performed

References ghidra::SplitDatatype::RootPointer::baseOffset, buildInConstants(), buildInSubpieces(), buildPointers(), ghidra::PcodeOp::code(), ghidra::CPUI_LOAD, ghidra::CPUI_STORE, data, dataTypePieces, ghidra::SplitDatatype::RootPointer::find(), ghidra::SplitDatatype::RootPointer::freePointerChain(), ghidra::PcodeOp::getAddr(), ghidra::Funcdata::getArch(), ghidra::Varnode::getDef(), ghidra::PcodeOp::getIn(), ghidra::Varnode::getSize(), ghidra::Datatype::getSize(), ghidra::Varnode::getSpaceFromConst(), ghidra::Varnode::getTypeReadFacing(), getValueDatatype(), isArithmeticOutput(), ghidra::Varnode::isConstant(), ghidra::Varnode::isWritten(), ghidra::Varnode::loneDescend(), ghidra::Funcdata::newOp(), ghidra::Funcdata::newUniqueOut(), ghidra::Funcdata::newVarnodeSpace(), ghidra::Funcdata::opDestroy(), ghidra::Funcdata::opInsertAfter(), ghidra::Funcdata::opInsertBefore(), ghidra::Funcdata::opSetInput(), ghidra::Funcdata::opSetOpcode(), ghidra::SplitDatatype::RootPointer::pointer, ghidra::SplitDatatype::RootPointer::ptrType, testDatatypeCompatibility(), ghidra::Architecture::types, and ghidra::Varnode::updateType().

Referenced by ghidra::RuleSplitStore::applyOp().

◆ testCopyConstraints()

bool ghidra::SplitDatatype::testCopyConstraints ( PcodeOp * copyOp )

private

Test specific constraints for splitting the given COPY operation into pieces.

Don't split function inputs. Don't split hidden COPYs.

Returns: true if the split can proceed

References ghidra::PcodeOp::code(), ghidra::CPUI_LOAD, ghidra::Varnode::getAddr(), ghidra::Varnode::getDef(), ghidra::PcodeOp::getIn(), ghidra::PcodeOp::getOut(), ghidra::Varnode::isAddrTied(), ghidra::Varnode::isInput(), ghidra::Varnode::isWritten(), and ghidra::Varnode::loneDescend().

Referenced by splitCopy().

◆ testDatatypeCompatibility()

bool ghidra::SplitDatatype::testDatatypeCompatibility	(	Datatype *	inBase,
		Datatype *	outBase,
		bool	inConstant
	)

private

Can the two given data-types be mutually split into matching logical components.

Test if the data-types have components with matching size and offset. If so, the component data-types and offsets are saved to the pieces array and true is returned. At least one of the data-types must be a partial data-type, but the other may be a TYPE_UNKNOWN, which this method assumes can be split into components of arbitrary size.

Parameters

inBase	is the data-type coming into the operation
outBase	is the data-type coming out of the operation
inConstant	is true if the incoming data-type labels a constant

Returns: true if the data-types have compatible components, false otherwise

References ghidra::Datatype::getMetatype(), ghidra::Datatype::getSize(), ghidra::TYPE_STRUCT, and ghidra::TYPE_UNKNOWN.

Referenced by splitCopy(), splitLoad(), and splitStore().

The documentation for this class was generated from the following files:

subflow.hh
subflow.cc

Classes

Public Member Functions

Static Public Member Functions

Private Member Functions

Static Private Member Functions

Private Attributes

Detailed Description

Member Function Documentation

◆ buildInConstants()

◆ buildInSubpieces()

◆ buildOutConcats()

◆ buildOutVarnodes()

◆ buildPointers()

◆ categorizeDatatype()

◆ generateConstants()

◆ getComponent()

◆ getValueDatatype()

◆ isArithmeticInput()

◆ isArithmeticOutput()

◆ splitCopy()

◆ splitLoad()

◆ splitStore()

◆ testCopyConstraints()

◆ testDatatypeCompatibility()