Documentation/bpf/standardization/instruction-set.rst

.. contents::
.. sectnum::

======================================
BPF Instruction Set Architecture (ISA)
======================================

eBPF, also commonly
referred to as BPF, is a technology with origins in the Linux kernel
that can run untrusted programs in a privileged context such as an
operating system kernel. This document specifies the BPF instruction
set architecture (ISA).

As a historical note, BPF originally stood for Berkeley Packet Filter,
but now that it can do so much more than packet filtering, the acronym
no longer makes sense. BPF is now considered a standalone term that
does not stand for anything.  The original BPF is sometimes referred to
as cBPF (classic BPF) to distinguish it from the now widely deployed
eBPF (extended BPF).

Documentation conventions
=========================

The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT",
"SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and
"OPTIONAL" in this document are to be interpreted as described in
BCP 14 `<https://www.rfc-editor.org/info/rfc2119>`_
`<https://www.rfc-editor.org/info/rfc8174>`_
when, and only when, they appear in all capitals, as shown here.

For brevity and consistency, this document refers to families
of types using a shorthand syntax and refers to several expository,
mnemonic functions when describing the semantics of instructions.
The range of valid values for those types and the semantics of those
functions are defined in the following subsections.

Types
-----
This document refers to integer types with the notation `SN` to specify
a type's signedness (`S`) and bit width (`N`), respectively.

.. table:: Meaning of signedness notation

  ==== =========
  S    Meaning
  ==== =========
  u    unsigned
  s    signed
  ==== =========

.. table:: Meaning of bit-width notation

  ===== =========
  N     Bit width
  ===== =========
  8     8 bits
  16    16 bits
  32    32 bits
  64    64 bits
  128   128 bits
  ===== =========

For example, `u32` is a type whose valid values are all the 32-bit unsigned
numbers and `s16` is a type whose valid values are all the 16-bit signed
numbers.

Functions
---------

The following byteswap functions are direction-agnostic.  That is,
the same function is used for conversion in either direction discussed
below.

* be16: Takes an unsigned 16-bit number and converts it between
  host byte order and big-endian
  (`IEN137 <https://www.rfc-editor.org/ien/ien137.txt>`_) byte order.
* be32: Takes an unsigned 32-bit number and converts it between
  host byte order and big-endian byte order.
* be64: Takes an unsigned 64-bit number and converts it between
  host byte order and big-endian byte order.
* bswap16: Takes an unsigned 16-bit number in either big- or little-endian
  format and returns the equivalent number with the same bit width but
  opposite endianness.
* bswap32: Takes an unsigned 32-bit number in either big- or little-endian
  format and returns the equivalent number with the same bit width but
  opposite endianness.
* bswap64: Takes an unsigned 64-bit number in either big- or little-endian
  format and returns the equivalent number with the same bit width but
  opposite endianness.
* le16: Takes an unsigned 16-bit number and converts it between
  host byte order and little-endian byte order.
* le32: Takes an unsigned 32-bit number and converts it between
  host byte order and little-endian byte order.
* le64: Takes an unsigned 64-bit number and converts it between
  host byte order and little-endian byte order.

Definitions
-----------

.. glossary::

  Sign Extend
    To `sign extend an` ``X`` `-bit number, A, to a` ``Y`` `-bit number, B  ,` means to

    #. Copy all ``X`` bits from `A` to the lower ``X`` bits of `B`.
    #. Set the value of the remaining ``Y`` - ``X`` bits of `B` to the value of
       the  most-significant bit of `A`.

.. admonition:: Example

  Sign extend an 8-bit number ``A`` to a 16-bit number ``B`` on a big-endian platform:
  ::

    A:          10000110
    B: 11111111 10000110

Conformance groups
------------------

An implementation does not need to support all instructions specified in this
document (e.g., deprecated instructions).  Instead, a number of conformance
groups are specified.  An implementation MUST support the base32 conformance
group and MAY support additional conformance groups, where supporting a
conformance group means it MUST support all instructions in that conformance
group.

The use of named conformance groups enables interoperability between a runtime
that executes instructions, and tools such as compilers that generate
instructions for the runtime.  Thus, capability discovery in terms of
conformance groups might be done manually by users or automatically by tools.

Each conformance group has a short ASCII label (e.g., "base32") that
corresponds to a set of instructions that are mandatory.  That is, each
instruction has one or more conformance groups of which it is a member.

This document defines the following conformance groups:

* base32: includes all instructions defined in this
  specification unless otherwise noted.
* base64: includes base32, plus instructions explicitly noted
  as being in the base64 conformance group.
* atomic32: includes 32-bit atomic operation instructions (see `Atomic operations`_).
* atomic64: includes atomic32, plus 64-bit atomic operation instructions.
* divmul32: includes 32-bit division, multiplication, and modulo instructions.
* divmul64: includes divmul32, plus 64-bit division, multiplication,
  and modulo instructions.
* packet: deprecated packet access instructions.

Instruction encoding
====================

BPF has two instruction encodings:

* the basic instruction encoding, which uses 64 bits to encode an instruction
* the wide instruction encoding, which appends a second 64 bits
  after the basic instruction for a total of 128 bits.

Basic instruction encoding
--------------------------

A basic instruction is encoded as follows::

  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  |    opcode     |     regs      |            offset             |
  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  |                              imm                              |
  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

**opcode**
  operation to perform, encoded as follows::

    +-+-+-+-+-+-+-+-+
    |specific |class|
    +-+-+-+-+-+-+-+-+

  **specific**
    The format of these bits varies by instruction class

  **class**
    The instruction class (see `Instruction classes`_)

**regs**
  The source and destination register numbers, encoded as follows
  on a little-endian host::

    +-+-+-+-+-+-+-+-+
    |src_reg|dst_reg|
    +-+-+-+-+-+-+-+-+

  and as follows on a big-endian host::

    +-+-+-+-+-+-+-+-+
    |dst_reg|src_reg|
    +-+-+-+-+-+-+-+-+

  **src_reg**
    the source register number (0-10), except where otherwise specified
    (`64-bit immediate instructions`_ reuse this field for other purposes)

  **dst_reg**
    destination register number (0-10), unless otherwise specified
    (future instructions might reuse this field for other purposes)

**offset**
  signed integer offset used with pointer arithmetic, except where
  otherwise specified (some arithmetic instructions reuse this field
  for other purposes)

**imm**
  signed integer immediate value

Note that the contents of multi-byte fields ('offset' and 'imm') are
stored using big-endian byte ordering on big-endian hosts and
little-endian byte ordering on little-endian hosts.

For example::

  opcode                  offset imm          assembly
         src_reg dst_reg
  07     0       1        00 00  44 33 22 11  r1 += 0x11223344 // little
         dst_reg src_reg
  07     1       0        00 00  11 22 33 44  r1 += 0x11223344 // big

Note that most instructions do not use all of the fields.
Unused fields SHALL be cleared to zero.

Wide instruction encoding
--------------------------

Some instructions are defined to use the wide instruction encoding,
which uses two 32-bit immediate values.  The 64 bits following
the basic instruction format contain a pseudo instruction
with 'opcode', 'dst_reg', 'src_reg', and 'offset' all set to zero.

This is depicted in the following figure::

  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  |    opcode     |     regs      |            offset             |
  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  |                              imm                              |
  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  |                           reserved                            |
  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
  |                           next_imm                            |
  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

**opcode**
  operation to perform, encoded as explained above

**regs**
  The source and destination register numbers (unless otherwise
  specified), encoded as explained above

**offset**
  signed integer offset used with pointer arithmetic, unless
  otherwise specified

**imm**
  signed integer immediate value

**reserved**
  unused, set to zero

**next_imm**
  second signed integer immediate value

Instruction classes
-------------------

The three least significant bits of the 'opcode' field store the instruction class:

.. table:: Instruction class

  =====  =====  ===============================  ===================================
  class  value  description                      reference
  =====  =====  ===============================  ===================================
  LD     0x0    non-standard load operations     `Load and store instructions`_
  LDX    0x1    load into register operations    `Load and store instructions`_
  ST     0x2    store from immediate operations  `Load and store instructions`_
  STX    0x3    store from register operations   `Load and store instructions`_
  ALU    0x4    32-bit arithmetic operations     `Arithmetic and jump instructions`_
  JMP    0x5    64-bit jump operations           `Arithmetic and jump instructions`_
  JMP32  0x6    32-bit jump operations           `Arithmetic and jump instructions`_
  ALU64  0x7    64-bit arithmetic operations     `Arithmetic and jump instructions`_
  =====  =====  ===============================  ===================================

Arithmetic and jump instructions
================================

For arithmetic and jump instructions (``ALU``, ``ALU64``, ``JMP`` and
``JMP32``), the 8-bit 'opcode' field is divided into three parts::

  +-+-+-+-+-+-+-+-+
  |  code |s|class|
  +-+-+-+-+-+-+-+-+

**code**
  the operation code, whose meaning varies by instruction class

**s (source)**
  the source operand location, which unless otherwise specified is one of:

  .. table:: Source operand location

    ======  =====  ==============================================
    source  value  description
    ======  =====  ==============================================
    K       0      use 32-bit 'imm' value as source operand
    X       1      use 'src_reg' register value as source operand
    ======  =====  ==============================================

**instruction class**
  the instruction class (see `Instruction classes`_)

Arithmetic instructions
-----------------------

``ALU`` uses 32-bit wide operands while ``ALU64`` uses 64-bit wide operands for
otherwise identical operations. ``ALU64`` instructions belong to the
base64 conformance group unless noted otherwise.
The 'code' field encodes the operation as below, where 'src' refers to the
the source operand and 'dst' refers to the value of the destination
register.

.. table:: Arithmetic instructions

  =====  =====  =======  ===================================================================================
  name   code   offset   description
  =====  =====  =======  ===================================================================================
  ADD    0x0    0        dst += src
  SUB    0x1    0        dst -= src
  MUL    0x2    0        dst \*= src
  DIV    0x3    0        dst = (src != 0) ? (dst / src) : 0
  SDIV   0x3    1        dst = (src == 0) ? 0 : ((src == -1 && dst == LLONG_MIN) ? LLONG_MIN : (dst s/ src))
  OR     0x4    0        dst \|= src
  AND    0x5    0        dst &= src
  LSH    0x6    0        dst <<= (src & mask)
  RSH    0x7    0        dst >>= (src & mask)
  NEG    0x8    0        dst = -dst
  MOD    0x9    0        dst = (src != 0) ? (dst % src) : dst
  SMOD   0x9    1        dst = (src == 0) ? dst : ((src == -1 && dst == LLONG_MIN) ? 0: (dst s% src))
  XOR    0xa    0        dst ^= src
  MOV    0xb    0        dst = src
  MOVSX  0xb    8/16/32  dst = (s8,s16,s32)src
  ARSH   0xc    0        :term:`sign extending<Sign Extend>` dst >>= (src & mask)
  END    0xd    0        byte swap operations (see `Byte swap instructions`_ below)
  =====  =====  =======  ===================================================================================

Underflow and overflow are allowed during arithmetic operations, meaning
the 64-bit or 32-bit value will wrap. If BPF program execution would
result in division by zero, the destination register is instead set to zero.
Otherwise, for ``ALU64``, if execution would result in ``LLONG_MIN``
divided by -1, the destination register is instead set to ``LLONG_MIN``. For
``ALU``, if execution would result in ``INT_MIN`` divided by -1, the
destination register is instead set to ``INT_MIN``.

If execution would result in modulo by zero, for ``ALU64`` the value of
the destination register is unchanged whereas for ``ALU`` the upper
32 bits of the destination register are zeroed. Otherwise, for ``ALU64``,
if execution would resuslt in ``LLONG_MIN`` modulo -1, the destination
register is instead set to 0. For ``ALU``, if execution would result in
``INT_MIN`` modulo -1, the destination register is instead set to 0.

``{ADD, X, ALU}``, where 'code' = ``ADD``, 'source' = ``X``, and 'class' = ``ALU``, means::

  dst = (u32) ((u32) dst + (u32) src)

where '(u32)' indicates that the upper 32 bits are zeroed.

``{ADD, X, ALU64}`` means::

  dst = dst + src

``{XOR, K, ALU}`` means::

  dst = (u32) dst ^ (u32) imm

``{XOR, K, ALU64}`` means::

  dst = dst ^ imm

Note that most arithmetic instructions have 'offset' set to 0. Only three instructions
(``SDIV``, ``SMOD``, ``MOVSX``) have a non-zero 'offset'.

Division, multiplication, and modulo operations for ``ALU`` are part
of the "divmul32" conformance group, and division, multiplication, and
modulo operations for ``ALU64`` are part of the "divmul64" conformance
group.
The division and modulo operations support both unsigned and signed flavors.

For unsigned operations (``DIV`` and ``MOD``), for ``ALU``,
'imm' is interpreted as a 32-bit unsigned value. For ``ALU64``,
'imm' is first :term:`sign extended<Sign Extend>` from 32 to 64 bits, and then
interpreted as a 64-bit unsigned value.

For signed operations (``SDIV`` and ``SMOD``), for ``ALU``,
'imm' is interpreted as a 32-bit signed value. For ``ALU64``, 'imm'
is first :term:`sign extended<Sign Extend>` from 32 to 64 bits, and then
interpreted as a 64-bit signed value.

Note that there are varying definitions of the signed modulo operation
when the dividend or divisor are negative, where implementations often
vary by language such that Python, Ruby, etc.  differ from C, Go, Java,
etc. This specification requires that signed modulo MUST use truncated division
(where -13 % 3 == -1) as implemented in C, Go, etc.::

   a % n = a - n * trunc(a / n)

The ``MOVSX`` instruction does a move operation with sign extension.
``{MOVSX, X, ALU}`` :term:`sign extends<Sign Extend>` 8-bit and 16-bit operands into
32-bit operands, and zeroes the remaining upper 32 bits.
``{MOVSX, X, ALU64}`` :term:`sign extends<Sign Extend>` 8-bit, 16-bit, and 32-bit
operands into 64-bit operands.  Unlike other arithmetic instructions,
``MOVSX`` is only defined for register source operands (``X``).

``{MOV, K, ALU64}`` means::

  dst = (s64)imm

``{MOV, X, ALU}`` means::

  dst = (u32)src

``{MOVSX, X, ALU}`` with 'offset' 8 means::

  dst = (u32)(s32)(s8)src


The ``NEG`` instruction is only defined when the source bit is clear
(``K``).

Shift operations use a mask of 0x3F (63) for 64-bit operations and 0x1F (31)
for 32-bit operations.

Byte swap instructions
----------------------

The byte swap instructions use instruction classes of ``ALU`` and ``ALU64``
and a 4-bit 'code' field of ``END``.

The byte swap instructions operate on the destination register
only and do not use a separate source register or immediate value.

For ``ALU``, the 1-bit source operand field in the opcode is used to
select what byte order the operation converts from or to. For
``ALU64``, the 1-bit source operand field in the opcode is reserved
and MUST be set to 0.

.. table:: Byte swap instructions

  =====  ========  =====  =================================================
  class  source    value  description
  =====  ========  =====  =================================================
  ALU    LE        0      convert between host byte order and little endian
  ALU    BE        1      convert between host byte order and big endian
  ALU64  Reserved  0      do byte swap unconditionally
  =====  ========  =====  =================================================

The 'imm' field encodes the width of the swap operations.  The following widths
are supported: 16, 32 and 64.  Width 64 operations belong to the base64
conformance group and other swap operations belong to the base32
conformance group.

Examples:

``{END, LE, ALU}`` with 'imm' = 16/32/64 means::

  dst = le16(dst)
  dst = le32(dst)
  dst = le64(dst)

``{END, BE, ALU}`` with 'imm' = 16/32/64 means::

  dst = be16(dst)
  dst = be32(dst)
  dst = be64(dst)

``{END, TO, ALU64}`` with 'imm' = 16/32/64 means::

  dst = bswap16(dst)
  dst = bswap32(dst)
  dst = bswap64(dst)

Jump instructions
-----------------

``JMP32`` uses 32-bit wide operands and indicates the base32
conformance group, while ``JMP`` uses 64-bit wide operands for
otherwise identical operations, and indicates the base64 conformance
group unless otherwise specified.
The 'code' field encodes the operation as below:

.. table:: Jump instructions

  ========  =====  =======  =================================  ===================================================
  code      value  src_reg  description                        notes
  ========  =====  =======  =================================  ===================================================
  JA        0x0    0x0      PC += offset                       {JA, K, JMP} only
  JA        0x0    0x0      PC += imm                          {JA, K, JMP32} only
  JEQ       0x1    any      PC += offset if dst == src
  JGT       0x2    any      PC += offset if dst > src          unsigned
  JGE       0x3    any      PC += offset if dst >= src         unsigned
  JSET      0x4    any      PC += offset if dst & src
  JNE       0x5    any      PC += offset if dst != src
  JSGT      0x6    any      PC += offset if dst > src          signed
  JSGE      0x7    any      PC += offset if dst >= src         signed
  CALL      0x8    0x0      call helper function by static ID  {CALL, K, JMP} only, see `Helper functions`_
  CALL      0x8    0x1      call PC += imm                     {CALL, K, JMP} only, see `Program-local functions`_
  CALL      0x8    0x2      call helper function by BTF ID     {CALL, K, JMP} only, see `Helper functions`_
  EXIT      0x9    0x0      return                             {CALL, K, JMP} only
  JLT       0xa    any      PC += offset if dst < src          unsigned
  JLE       0xb    any      PC += offset if dst <= src         unsigned
  JSLT      0xc    any      PC += offset if dst < src          signed
  JSLE      0xd    any      PC += offset if dst <= src         signed
  ========  =====  =======  =================================  ===================================================

where 'PC' denotes the program counter, and the offset to increment by
is in units of 64-bit instructions relative to the instruction following
the jump instruction.  Thus 'PC += 1' skips execution of the next
instruction if it's a basic instruction or results in undefined behavior
if the next instruction is a 128-bit wide instruction.

Example:

``{JSGE, X, JMP32}`` means::

  if (s32)dst s>= (s32)src goto +offset

where 's>=' indicates a signed '>=' comparison.

``{JLE, K, JMP}`` means::

  if dst <= (u64)(s64)imm goto +offset

``{JA, K, JMP32}`` means::

  gotol +imm

where 'imm' means the branch offset comes from the 'imm' field.

Note that there are two flavors of ``JA`` instructions. The
``JMP`` class permits a 16-bit jump offset specified by the 'offset'
field, whereas the ``JMP32`` class permits a 32-bit jump offset
specified by the 'imm' field. A > 16-bit conditional jump may be
converted to a < 16-bit conditional jump plus a 32-bit unconditional
jump.

All ``CALL`` and ``JA`` instructions belong to the
base32 conformance group.

Helper functions
~~~~~~~~~~~~~~~~

Helper functions are a concept whereby BPF programs can call into a
set of function calls exposed by the underlying platform.

Historically, each helper function was identified by a static ID
encoded in the 'imm' field.  Further documentation of helper functions
is outside the scope of this document and standardization is left for
future work, but use is widely deployed and more information can be
found in platform-specific documentation (e.g., Linux kernel documentation).

Platforms that support the BPF Type Format (BTF) support identifying
a helper function by a BTF ID encoded in the 'imm' field, where the BTF ID
identifies the helper name and type.  Further documentation of BTF
is outside the scope of this document and standardization is left for
future work, but use is widely deployed and more information can be
found in platform-specific documentation (e.g., Linux kernel documentation).

Program-local functions
~~~~~~~~~~~~~~~~~~~~~~~
Program-local functions are functions exposed by the same BPF program as the
caller, and are referenced by offset from the instruction following the call
instruction, similar to ``JA``.  The offset is encoded in the 'imm' field of
the call instruction. An ``EXIT`` within the program-local function will
return to the caller.

Load and store instructions
===========================

For load and store instructions (``LD``, ``LDX``, ``ST``, and ``STX``), the
8-bit 'opcode' field is divided as follows::

  +-+-+-+-+-+-+-+-+
  |mode |sz |class|
  +-+-+-+-+-+-+-+-+

**mode**
  The mode modifier is one of:

  .. table:: Mode modifier

    =============  =====  ====================================  =============
    mode modifier  value  description                           reference
    =============  =====  ====================================  =============
    IMM            0      64-bit immediate instructions         `64-bit immediate instructions`_
    ABS            1      legacy BPF packet access (absolute)   `Legacy BPF Packet access instructions`_
    IND            2      legacy BPF packet access (indirect)   `Legacy BPF Packet access instructions`_
    MEM            3      regular load and store operations     `Regular load and store operations`_
    MEMSX          4      sign-extension load operations        `Sign-extension load operations`_
    ATOMIC         6      atomic operations                     `Atomic operations`_
    =============  =====  ====================================  =============

**sz (size)**
  The size modifier is one of:

  .. table:: Size modifier

    ====  =====  =====================
    size  value  description
    ====  =====  =====================
    W     0      word        (4 bytes)
    H     1      half word   (2 bytes)
    B     2      byte
    DW    3      double word (8 bytes)
    ====  =====  =====================

  Instructions using ``DW`` belong to the base64 conformance group.

**class**
  The instruction class (see `Instruction classes`_)

Regular load and store operations
---------------------------------

The ``MEM`` mode modifier is used to encode regular load and store
instructions that transfer data between a register and memory.

``{MEM, <size>, STX}`` means::

  *(size *) (dst + offset) = src

``{MEM, <size>, ST}`` means::

  *(size *) (dst + offset) = imm

``{MEM, <size>, LDX}`` means::

  dst = *(unsigned size *) (src + offset)

Where '<size>' is one of: ``B``, ``H``, ``W``, or ``DW``, and
'unsigned size' is one of: u8, u16, u32, or u64.

Sign-extension load operations
------------------------------

The ``MEMSX`` mode modifier is used to encode :term:`sign-extension<Sign Extend>` load
instructions that transfer data between a register and memory.

``{MEMSX, <size>, LDX}`` means::

  dst = *(signed size *) (src + offset)

Where '<size>' is one of: ``B``, ``H``, or ``W``, and
'signed size' is one of: s8, s16, or s32.

Atomic operations
-----------------

Atomic operations are operations that operate on memory and can not be
interrupted or corrupted by other access to the same memory region
by other BPF programs or means outside of this specification.

All atomic operations supported by BPF are encoded as store operations
that use the ``ATOMIC`` mode modifier as follows:

* ``{ATOMIC, W, STX}`` for 32-bit operations, which are
  part of the "atomic32" conformance group.
* ``{ATOMIC, DW, STX}`` for 64-bit operations, which are
  part of the "atomic64" conformance group.
* 8-bit and 16-bit wide atomic operations are not supported.

The 'imm' field is used to encode the actual atomic operation.
Simple atomic operation use a subset of the values defined to encode
arithmetic operations in the 'imm' field to encode the atomic operation:

.. table:: Simple atomic operations

  ========  =====  ===========
  imm       value  description
  ========  =====  ===========
  ADD       0x00   atomic add
  OR        0x40   atomic or
  AND       0x50   atomic and
  XOR       0xa0   atomic xor
  ========  =====  ===========


``{ATOMIC, W, STX}`` with 'imm' = ADD means::

  *(u32 *)(dst + offset) += src

``{ATOMIC, DW, STX}`` with 'imm' = ADD means::

  *(u64 *)(dst + offset) += src

In addition to the simple atomic operations, there also is a modifier and
two complex atomic operations:

.. table:: Complex atomic operations

  ===========  ================  ===========================
  imm          value             description
  ===========  ================  ===========================
  FETCH        0x01              modifier: return old value
  XCHG         0xe0 | FETCH      atomic exchange
  CMPXCHG      0xf0 | FETCH      atomic compare and exchange
  ===========  ================  ===========================

The ``FETCH`` modifier is optional for simple atomic operations, and
always set for the complex atomic operations.  If the ``FETCH`` flag
is set, then the operation also overwrites ``src`` with the value that
was in memory before it was modified.

The ``XCHG`` operation atomically exchanges ``src`` with the value
addressed by ``dst + offset``.

The ``CMPXCHG`` operation atomically compares the value addressed by
``dst + offset`` with ``R0``. If they match, the value addressed by
``dst + offset`` is replaced with ``src``. In either case, the
value that was at ``dst + offset`` before the operation is zero-extended
and loaded back to ``R0``.

64-bit immediate instructions
-----------------------------

Instructions with the ``IMM`` 'mode' modifier use the wide instruction
encoding defined in `Instruction encoding`_, and use the 'src_reg' field of the
basic instruction to hold an opcode subtype.

The following table defines a set of ``{IMM, DW, LD}`` instructions
with opcode subtypes in the 'src_reg' field, using new terms such as "map"
defined further below:

.. table:: 64-bit immediate instructions

  =======  =========================================  ===========  ==============
  src_reg  pseudocode                                 imm type     dst type
  =======  =========================================  ===========  ==============
  0x0      dst = (next_imm << 32) | imm               integer      integer
  0x1      dst = map_by_fd(imm)                       map fd       map
  0x2      dst = map_val(map_by_fd(imm)) + next_imm   map fd       data address
  0x3      dst = var_addr(imm)                        variable id  data address
  0x4      dst = code_addr(imm)                       integer      code address
  0x5      dst = map_by_idx(imm)                      map index    map
  0x6      dst = map_val(map_by_idx(imm)) + next_imm  map index    data address
  =======  =========================================  ===========  ==============

where

* map_by_fd(imm) means to convert a 32-bit file descriptor into an address of a map (see `Maps`_)
* map_by_idx(imm) means to convert a 32-bit index into an address of a map
* map_val(map) gets the address of the first value in a given map
* var_addr(imm) gets the address of a platform variable (see `Platform Variables`_) with a given id
* code_addr(imm) gets the address of the instruction at a specified relative offset in number of (64-bit) instructions
* the 'imm type' can be used by disassemblers for display
* the 'dst type' can be used for verification and JIT compilation purposes

Maps
~~~~

Maps are shared memory regions accessible by BPF programs on some platforms.
A map can have various semantics as defined in a separate document, and may or
may not have a single contiguous memory region, but the 'map_val(map)' is
currently only defined for maps that do have a single contiguous memory region.

Each map can have a file descriptor (fd) if supported by the platform, where
'map_by_fd(imm)' means to get the map with the specified file descriptor. Each
BPF program can also be defined to use a set of maps associated with the
program at load time, and 'map_by_idx(imm)' means to get the map with the given
index in the set associated with the BPF program containing the instruction.

Platform Variables
~~~~~~~~~~~~~~~~~~

Platform variables are memory regions, identified by integer ids, exposed by
the runtime and accessible by BPF programs on some platforms.  The
'var_addr(imm)' operation means to get the address of the memory region
identified by the given id.

Legacy BPF Packet access instructions
-------------------------------------

BPF previously introduced special instructions for access to packet data that were
carried over from classic BPF. These instructions used an instruction
class of ``LD``, a size modifier of ``W``, ``H``, or ``B``, and a
mode modifier of ``ABS`` or ``IND``.  The 'dst_reg' and 'offset' fields were
set to zero, and 'src_reg' was set to zero for ``ABS``.  However, these
instructions are deprecated and SHOULD no longer be used.  All legacy packet
access instructions belong to the "packet" conformance group.