P-Code Reference Manual

Last updated September 5, 2019


COPY INT_ADD BOOL_OR
LOAD INT_SUB FLOAT_EQUAL
STORE INT_CARRY FLOAT_NOTEQUAL
BRANCH INT_SCARRY FLOAT_LESS
CBRANCH INT_SBORROW FLOAT_LESSEQUAL
BRANCHIND INT_2COMP FLOAT_ADD
CALL INT_NEGATE FLOAT_SUB
CALLIND INT_XOR FLOAT_MULT
USERDEFINED INT_AND FLOAT_DIV
RETURN INT_OR FLOAT_NEG
PIECE INT_LEFT FLOAT_ABS
SUBPIECE INT_RIGHT FLOAT_SQRT
POPCOUNT INT_SRIGHT FLOAT_CEIL
INT_EQUAL INT_MULT FLOAT_FLOOR
INT_NOTEQUAL INT_DIV FLOAT_ROUND
INT_LESS INT_REM FLOAT_NAN
INT_SLESS INT_SDIV INT2FLOAT
INT_LESSEQUAL INT_SREM FLOAT2FLOAT
INT_SLESSEQUAL BOOL_NEGATE TRUNC
INT_ZEXT BOOL_XOR CPOOLREF
INT_SEXT BOOL_AND NEW

A Brief Introduction to P-Code

P-code is a register transfer language designed for reverse engineering applications. The language is general enough to model the behavior of many different processors. By modeling in this way, the analysis of different processors is put into a common framework, facilitating the development of retargetable analysis algorithms and applications.

Fundamentally, p-code works by translating individual processor instructions into a sequence of p-code operations that take parts of the processor state as input and output variables (varnodes). The set of unique p-code operations (distinguished by opcode) comprise a fairly tight set of the arithmetic and logical actions performed by general purpose processors. The direct translation of instructions into these operations is referred to as raw p-code. Raw p-code can be used to directly emulate instruction execution and generally follows the same control-flow, although it may add some of its own internal control-flow. The subset of opcodes that can occur in raw p-code is described in the section called “P-Code Operation Reference” and in the section called “Pseudo P-CODE Operations”, making up the bulk of this document.

P-code is designed specifically to facilitate the construction of data-flow graphs for follow-on analysis of disassembled instructions. Varnodes and p-code operators can be thought of explicitly as nodes in these graphs. Generation of raw p-code is a necessary first step in graph construction, but additional steps are required, which introduces some new opcodes. Two of these, MULTIEQUAL and INDIRECT, are specific to the graph construction process, but other opcodes can be introduced during subsequent analysis and transformation of a graph and help hold recovered data-type relationships. All of the new opcodes are described in the section called “Additional P-CODE Operations”, none of which can occur in the original raw p-code translation. Finally, a few of the p-code operators, CALL, CALLIND, and RETURN, may have their input and output varnodes changed during analysis so that they no longer match their raw p-code form.

The core concepts of p-code are:

Address Space

The address space for p-code is a generalization of RAM. It is defined simply as an indexed sequence of bytes that can be read and written by the p-code operations. For a specific byte, the unique index that labels it is the byte's address. An address space has a name to identify it, a size that indicates the number of distinct indices into the space, and an endianess associated with it that indicates how integers and other multi-byte values are encoded into the space. A typical processor will have a ram space, to model memory accessible via its main data bus, and a register space for modeling the processor's general purpose registers. Any data that a processor manipulates must be in some address space. The specification for a processor is free to define as many address spaces as it needs. There is always a special address space, called a constant address space, which is used to encode any constant values needed for p-code operations. Systems generating p-code also generally use a dedicated temporary space, which can be viewed as a bottomless source of temporary registers. These are used to hold intermediate values when modeling instruction behavior.

P-code specifications allow the addressable unit of an address space to be bigger than just a byte. Each address space has a wordsize attribute that can be set to indicate the number of bytes in a unit. A wordsize which is bigger than one makes little difference to the representation of p-code. All the offsets into an address space are still represented internally as a byte offset. The only exceptions are the LOAD and STORE p-code operations. These operations read a pointer offset that must be scaled properly to get the right byte offset when dereferencing the pointer. The wordsize attribute has no effect on any of the other p-code operations.

Varnode

A varnode is a generalization of either a register or a memory location. It is represented by the formal triple: an address space, an offset into the space, and a size. Intuitively, a varnode is a contiguous sequence of bytes in some address space that can be treated as a single value. All manipulation of data by p-code operations occurs on varnodes.

Varnodes by themselves are just a contiguous chunk of bytes, identified by their address and size, and they have no type. The p-code operations however can force one of three type interpretations on the varnodes: integer, boolean, and floating-point.

  • Operations that manipulate integers always interpret a varnode as a twos-complement encoding using the endianess associated with the address space containing the varnode.
  • A varnode being used as a boolean value is assumed to be a single byte that can only take the value 0, for false, and 1, for true.
  • Floating-point operations use the encoding expected by the processor being modeled, which varies depending on the size of the varnode. For most processors, these encodings are described by the IEEE 754 standard, but other encodings are possible in principle.

If a varnode is specified as an offset into the constant address space, that offset is interpreted as a constant, or immediate value, in any p-code operation that uses that varnode. The size of the varnode, in this case, can be treated as the size or precision available for the encoding of the constant. As with other varnodes, constants only have a type forced on them by the p-code operations that use them.

P-code Operation

A p-code operation is the analog of a machine instruction. All p-code operations have the same basic format internally. They all take one or more varnodes as input and optionally produce a single output varnode. The action of the operation is determined by its opcode. For almost all p-code operations, only the output varnode can have its value modified; there are no indirect effects of the operation. The only possible exceptions are pseudo operations, see the section called “Pseudo P-CODE Operations”, which are sometimes necessary when there is incomplete knowledge of an instruction's behavior.

All p-code operations are associated with the address of the original processor instruction they were translated from. For a single instruction, a 1-up counter, starting at zero, is used to enumerate the multiple p-code operations involved in its translation. The address and counter as a pair are referred to as the p-code op's unique sequence number. Control-flow of p-code operations generally follows sequence number order. When execution of all p-code for one instruction is completed, if the instruction has fall-through semantics, p-code control-flow picks up with the first p-code operation in sequence corresponding to the instruction at the fall-through address. Similarly, if a p-code operation results in a control-flow branch, the first p-code operation in sequence executes at the destination address.

The list of possible opcodes are similar to many RISC based instruction sets. The effect of each opcode is described in detail in the following sections, and a reference table is given in the section called “Syntax Reference”. In general, the size or precision of a particular p-code operation is determined by the size of the varnode inputs or output, not by the opcode.