summaryrefslogtreecommitdiff
path: root/Documentation/bpf/standardization/instruction-set.rst
blob: 45cffe94c7520519df20ca9302d970721bff093a (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
.. contents::
.. sectnum::

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

This document specifies the BPF instruction set architecture (ISA).

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

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 types whose valid values are all the 16-bit signed
numbers.

Functions
---------
* `htobe16`: Takes an unsigned 16-bit number in host-endian format and
  returns the equivalent number as an unsigned 16-bit number in big-endian
  format.
* `htobe32`: Takes an unsigned 32-bit number in host-endian format and
  returns the equivalent number as an unsigned 32-bit number in big-endian
  format.
* `htobe64`: Takes an unsigned 64-bit number in host-endian format and
  returns the equivalent number as an unsigned 64-bit number in big-endian
  format.
* `htole16`: Takes an unsigned 16-bit number in host-endian format and
  returns the equivalent number as an unsigned 16-bit number in little-endian
  format.
* `htole32`: Takes an unsigned 32-bit number in host-endian format and
  returns the equivalent number as an unsigned 32-bit number in little-endian
  format.
* `htole64`: Takes an unsigned 64-bit number in host-endian format and
  returns the equivalent number as an unsigned 64-bit number in little-endian
  format.
* `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.


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 as such 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.
* legacy: 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-bit immediate (i.e.,
  constant) value after the basic instruction for a total of 128 bits.

The fields conforming an encoded basic instruction are stored in the
following order::

  opcode:8 src_reg:4 dst_reg:4 offset:16 imm:32 // In little-endian BPF.
  opcode:8 dst_reg:4 src_reg:4 offset:16 imm:32 // In big-endian BPF.

**imm**
  signed integer immediate value

**offset**
  signed integer offset used with pointer arithmetic

**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)

**opcode**
  operation to perform

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

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.

As discussed below in `64-bit immediate instructions`_, a 64-bit immediate
instruction uses two 32-bit immediate values that are constructed as follows.
The 64 bits following the basic instruction contain a pseudo instruction
using the same format but with 'opcode', 'dst_reg', 'src_reg', and 'offset' all
set to zero, and imm containing the high 32 bits of the immediate value.

This is depicted in the following figure::

        basic_instruction
  .------------------------------.
  |                              |
  opcode:8 regs:8 offset:16 imm:32 unused:32 imm:32
                                   |              |
                                   '--------------'
                                  pseudo instruction

Here, the imm value of the pseudo instruction is called 'next_imm'. The unused
bytes in the pseudo instruction are reserved and shall be cleared to zero.

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

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

=========  =====  ===============================  ===================================
class      value  description                      reference
=========  =====  ===============================  ===================================
BPF_LD     0x00   non-standard load operations     `Load and store instructions`_
BPF_LDX    0x01   load into register operations    `Load and store instructions`_
BPF_ST     0x02   store from immediate operations  `Load and store instructions`_
BPF_STX    0x03   store from register operations   `Load and store instructions`_
BPF_ALU    0x04   32-bit arithmetic operations     `Arithmetic and jump instructions`_
BPF_JMP    0x05   64-bit jump operations           `Arithmetic and jump instructions`_
BPF_JMP32  0x06   32-bit jump operations           `Arithmetic and jump instructions`_
BPF_ALU64  0x07   64-bit arithmetic operations     `Arithmetic and jump instructions`_
=========  =====  ===============================  ===================================

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

For arithmetic and jump instructions (``BPF_ALU``, ``BPF_ALU64``, ``BPF_JMP`` and
``BPF_JMP32``), the 8-bit 'opcode' field is divided into three parts:

==============  ======  =================
4 bits (MSB)    1 bit   3 bits (LSB)
==============  ======  =================
code            source  instruction class
==============  ======  =================

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

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

  ======  =====  ==============================================
  source  value  description
  ======  =====  ==============================================
  BPF_K   0x00   use 32-bit 'imm' value as source operand
  BPF_X   0x08   use 'src_reg' register value as source operand
  ======  =====  ==============================================

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

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

``BPF_ALU`` uses 32-bit wide operands while ``BPF_ALU64`` uses 64-bit wide operands for
otherwise identical operations. ``BPF_ALU64`` instructions belong to the
base64 conformance group unless noted otherwise.
The 'code' field encodes the operation as below, where 'src' and 'dst' refer
to the values of the source and destination registers, respectively.

=========  =====  =======  ==========================================================
code       value  offset   description
=========  =====  =======  ==========================================================
BPF_ADD    0x00   0        dst += src
BPF_SUB    0x10   0        dst -= src
BPF_MUL    0x20   0        dst \*= src
BPF_DIV    0x30   0        dst = (src != 0) ? (dst / src) : 0
BPF_SDIV   0x30   1        dst = (src != 0) ? (dst s/ src) : 0
BPF_OR     0x40   0        dst \|= src
BPF_AND    0x50   0        dst &= src
BPF_LSH    0x60   0        dst <<= (src & mask)
BPF_RSH    0x70   0        dst >>= (src & mask)
BPF_NEG    0x80   0        dst = -dst
BPF_MOD    0x90   0        dst = (src != 0) ? (dst % src) : dst
BPF_SMOD   0x90   1        dst = (src != 0) ? (dst s% src) : dst
BPF_XOR    0xa0   0        dst ^= src
BPF_MOV    0xb0   0        dst = src
BPF_MOVSX  0xb0   8/16/32  dst = (s8,s16,s32)src
BPF_ARSH   0xc0   0        :term:`sign extending<Sign Extend>` dst >>= (src & mask)
BPF_END    0xd0   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.
If execution would result in modulo by zero, for ``BPF_ALU64`` the value of
the destination register is unchanged whereas for ``BPF_ALU`` the upper
32 bits of the destination register are zeroed.

``BPF_ADD | BPF_X | BPF_ALU`` means::

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

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

``BPF_ADD | BPF_X | BPF_ALU64`` means::

  dst = dst + src

``BPF_XOR | BPF_K | BPF_ALU`` means::

  dst = (u32) dst ^ (u32) imm

``BPF_XOR | BPF_K | BPF_ALU64`` means::

  dst = dst ^ imm

Note that most instructions have instruction offset of 0. Only three instructions
(``BPF_SDIV``, ``BPF_SMOD``, ``BPF_MOVSX``) have a non-zero offset.

Division, multiplication, and modulo operations for ``BPF_ALU`` are part
of the "divmul32" conformance group, and division, multiplication, and
modulo operations for ``BPF_ALU64`` are part of the "divmul64" conformance
group.
The division and modulo operations support both unsigned and signed flavors.

For unsigned operations (``BPF_DIV`` and ``BPF_MOD``), for ``BPF_ALU``,
'imm' is interpreted as a 32-bit unsigned value. For ``BPF_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 (``BPF_SDIV`` and ``BPF_SMOD``), for ``BPF_ALU``,
'imm' is interpreted as a 32-bit signed value. For ``BPF_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 use truncated division
(where -13 % 3 == -1) as implemented in C, Go, etc.:

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

The ``BPF_MOVSX`` instruction does a move operation with sign extension.
``BPF_ALU | BPF_MOVSX`` :term:`sign extends<Sign Extend>` 8-bit and 16-bit operands into 32
bit operands, and zeroes the remaining upper 32 bits.
``BPF_ALU64 | BPF_MOVSX`` :term:`sign extends<Sign Extend>` 8-bit, 16-bit, and 32-bit
operands into 64 bit operands.  Unlike other arithmetic instructions,
``BPF_MOVSX`` is only defined for register source operands (``BPF_X``).

The ``BPF_NEG`` instruction is only defined when the source bit is clear
(``BPF_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 ``BPF_ALU`` and ``BPF_ALU64``
and a 4-bit 'code' field of ``BPF_END``.

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

For ``BPF_ALU``, the 1-bit source operand field in the opcode is used to
select what byte order the operation converts from or to. For
``BPF_ALU64``, the 1-bit source operand field in the opcode is reserved
and must be set to 0.

=========  =========  =====  =================================================
class      source     value  description
=========  =========  =====  =================================================
BPF_ALU    BPF_TO_LE  0x00   convert between host byte order and little endian
BPF_ALU    BPF_TO_BE  0x08   convert between host byte order and big endian
BPF_ALU64  Reserved   0x00   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:

``BPF_ALU | BPF_TO_LE | BPF_END`` with imm = 16/32/64 means::

  dst = htole16(dst)
  dst = htole32(dst)
  dst = htole64(dst)

``BPF_ALU | BPF_TO_BE | BPF_END`` with imm = 16/32/64 means::

  dst = htobe16(dst)
  dst = htobe32(dst)
  dst = htobe64(dst)

``BPF_ALU64 | BPF_TO_LE | BPF_END`` with imm = 16/32/64 means::

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

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

``BPF_JMP32`` uses 32-bit wide operands and indicates the base32
conformance group, while ``BPF_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:

========  =====  =======  ===============================  =============================================
code      value  src_reg  description                      notes
========  =====  =======  ===============================  =============================================
BPF_JA    0x0    0x0      PC += offset                     BPF_JMP | BPF_K only
BPF_JA    0x0    0x0      PC += imm                        BPF_JMP32 | BPF_K only
BPF_JEQ   0x1    any      PC += offset if dst == src
BPF_JGT   0x2    any      PC += offset if dst > src        unsigned
BPF_JGE   0x3    any      PC += offset if dst >= src       unsigned
BPF_JSET  0x4    any      PC += offset if dst & src
BPF_JNE   0x5    any      PC += offset if dst != src
BPF_JSGT  0x6    any      PC += offset if dst > src        signed
BPF_JSGE  0x7    any      PC += offset if dst >= src       signed
BPF_CALL  0x8    0x0      call helper function by address  BPF_JMP | BPF_K only, see `Helper functions`_
BPF_CALL  0x8    0x1      call PC += imm                   BPF_JMP | BPF_K only, see `Program-local functions`_
BPF_CALL  0x8    0x2      call helper function by BTF ID   BPF_JMP | BPF_K only, see `Helper functions`_
BPF_EXIT  0x9    0x0      return                           BPF_JMP | BPF_K only
BPF_JLT   0xa    any      PC += offset if dst < src        unsigned
BPF_JLE   0xb    any      PC += offset if dst <= src       unsigned
BPF_JSLT  0xc    any      PC += offset if dst < src        signed
BPF_JSLE  0xd    any      PC += offset if dst <= src       signed
========  =====  =======  ===============================  =============================================

The BPF program needs to store the return value into register R0 before doing a
``BPF_EXIT``.

Example:

``BPF_JSGE | BPF_X | BPF_JMP32`` (0x7e) means::

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

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

``BPF_JA | BPF_K | BPF_JMP32`` (0x06) means::

  gotol +imm

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

Note that there are two flavors of ``BPF_JA`` instructions. The
``BPF_JMP`` class permits a 16-bit jump offset specified by the 'offset'
field, whereas the ``BPF_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 ``BPF_CALL`` and ``BPF_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 an address
encoded in the imm field.  The available helper functions may differ
for each program type, but address values are unique across all program types.

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.

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

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

For load and store instructions (``BPF_LD``, ``BPF_LDX``, ``BPF_ST``, and ``BPF_STX``), the
8-bit 'opcode' field is divided as:

============  ======  =================
3 bits (MSB)  2 bits  3 bits (LSB)
============  ======  =================
mode          size    instruction class
============  ======  =================

The mode modifier is one of:

  =============  =====  ====================================  =============
  mode modifier  value  description                           reference
  =============  =====  ====================================  =============
  BPF_IMM        0x00   64-bit immediate instructions         `64-bit immediate instructions`_
  BPF_ABS        0x20   legacy BPF packet access (absolute)   `Legacy BPF Packet access instructions`_
  BPF_IND        0x40   legacy BPF packet access (indirect)   `Legacy BPF Packet access instructions`_
  BPF_MEM        0x60   regular load and store operations     `Regular load and store operations`_
  BPF_MEMSX      0x80   sign-extension load operations        `Sign-extension load operations`_
  BPF_ATOMIC     0xc0   atomic operations                     `Atomic operations`_
  =============  =====  ====================================  =============

The size modifier is one of:

  =============  =====  =====================
  size modifier  value  description
  =============  =====  =====================
  BPF_W          0x00   word        (4 bytes)
  BPF_H          0x08   half word   (2 bytes)
  BPF_B          0x10   byte
  BPF_DW         0x18   double word (8 bytes)
  =============  =====  =====================

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

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

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

``BPF_MEM | <size> | BPF_STX`` means::

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

``BPF_MEM | <size> | BPF_ST`` means::

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

``BPF_MEM | <size> | BPF_LDX`` means::

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

Where size is one of: ``BPF_B``, ``BPF_H``, ``BPF_W``, or ``BPF_DW`` and
'unsigned size' is one of u8, u16, u32 or u64.

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

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

``BPF_MEMSX | <size> | BPF_LDX`` means::

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

Where size is one of: ``BPF_B``, ``BPF_H`` or ``BPF_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 ``BPF_ATOMIC`` mode modifier as follows:

* ``BPF_ATOMIC | BPF_W | BPF_STX`` for 32-bit operations, which are
  part of the "atomic32" conformance group.
* ``BPF_ATOMIC | BPF_DW | BPF_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:

========  =====  ===========
imm       value  description
========  =====  ===========
BPF_ADD   0x00   atomic add
BPF_OR    0x40   atomic or
BPF_AND   0x50   atomic and
BPF_XOR   0xa0   atomic xor
========  =====  ===========


``BPF_ATOMIC | BPF_W  | BPF_STX`` with 'imm' = BPF_ADD means::

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

``BPF_ATOMIC | BPF_DW | BPF_STX`` with 'imm' = BPF_ADD means::

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

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

===========  ================  ===========================
imm          value             description
===========  ================  ===========================
BPF_FETCH    0x01              modifier: return old value
BPF_XCHG     0xe0 | BPF_FETCH  atomic exchange
BPF_CMPXCHG  0xf0 | BPF_FETCH  atomic compare and exchange
===========  ================  ===========================

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

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

The ``BPF_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 ``BPF_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 ``BPF_IMM | BPF_DW | BPF_LD`` instructions
with opcode subtypes in the 'src_reg' field, using new terms such as "map"
defined further below:

=========================  ======  =======  =========================================  ===========  ==============
opcode construction        opcode  src_reg  pseudocode                                 imm type     dst type
=========================  ======  =======  =========================================  ===========  ==============
BPF_IMM | BPF_DW | BPF_LD  0x18    0x0      dst = (next_imm << 32) | imm               integer      integer
BPF_IMM | BPF_DW | BPF_LD  0x18    0x1      dst = map_by_fd(imm)                       map fd       map
BPF_IMM | BPF_DW | BPF_LD  0x18    0x2      dst = map_val(map_by_fd(imm)) + next_imm   map fd       data pointer
BPF_IMM | BPF_DW | BPF_LD  0x18    0x3      dst = var_addr(imm)                        variable id  data pointer
BPF_IMM | BPF_DW | BPF_LD  0x18    0x4      dst = code_addr(imm)                       integer      code pointer
BPF_IMM | BPF_DW | BPF_LD  0x18    0x5      dst = map_by_idx(imm)                      map index    map
BPF_IMM | BPF_DW | BPF_LD  0x18    0x6      dst = map_val(map_by_idx(imm)) + next_imm  map index    data pointer
=========================  ======  =======  =========================================  ===========  ==============

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 BPF_LD, a size modifier of BPF_W, BPF_H, or BPF_B, and a
mode modifier of BPF_ABS or BPF_IND.  However, these instructions are
deprecated and should no longer be used.  All legacy packet access
instructions belong to the "legacy" conformance group.