xref: /linux/Documentation/bpf/standardization/instruction-set.rst (revision 4b660dbd9ee2059850fd30e0df420ca7a38a1856)
1.. contents::
2.. sectnum::
3
4======================================
5BPF Instruction Set Architecture (ISA)
6======================================
7
8This document specifies the BPF instruction set architecture (ISA).
9
10Documentation conventions
11=========================
12
13For brevity and consistency, this document refers to families
14of types using a shorthand syntax and refers to several expository,
15mnemonic functions when describing the semantics of instructions.
16The range of valid values for those types and the semantics of those
17functions are defined in the following subsections.
18
19Types
20-----
21This document refers to integer types with the notation `SN` to specify
22a type's signedness (`S`) and bit width (`N`), respectively.
23
24.. table:: Meaning of signedness notation.
25
26  ==== =========
27  S    Meaning
28  ==== =========
29  u    unsigned
30  s    signed
31  ==== =========
32
33.. table:: Meaning of bit-width notation.
34
35  ===== =========
36  N     Bit width
37  ===== =========
38  8     8 bits
39  16    16 bits
40  32    32 bits
41  64    64 bits
42  128   128 bits
43  ===== =========
44
45For example, `u32` is a type whose valid values are all the 32-bit unsigned
46numbers and `s16` is a types whose valid values are all the 16-bit signed
47numbers.
48
49Functions
50---------
51* htobe16: Takes an unsigned 16-bit number in host-endian format and
52  returns the equivalent number as an unsigned 16-bit number in big-endian
53  format.
54* htobe32: Takes an unsigned 32-bit number in host-endian format and
55  returns the equivalent number as an unsigned 32-bit number in big-endian
56  format.
57* htobe64: Takes an unsigned 64-bit number in host-endian format and
58  returns the equivalent number as an unsigned 64-bit number in big-endian
59  format.
60* htole16: Takes an unsigned 16-bit number in host-endian format and
61  returns the equivalent number as an unsigned 16-bit number in little-endian
62  format.
63* htole32: Takes an unsigned 32-bit number in host-endian format and
64  returns the equivalent number as an unsigned 32-bit number in little-endian
65  format.
66* htole64: Takes an unsigned 64-bit number in host-endian format and
67  returns the equivalent number as an unsigned 64-bit number in little-endian
68  format.
69* bswap16: Takes an unsigned 16-bit number in either big- or little-endian
70  format and returns the equivalent number with the same bit width but
71  opposite endianness.
72* bswap32: Takes an unsigned 32-bit number in either big- or little-endian
73  format and returns the equivalent number with the same bit width but
74  opposite endianness.
75* bswap64: Takes an unsigned 64-bit number in either big- or little-endian
76  format and returns the equivalent number with the same bit width but
77  opposite endianness.
78
79
80Definitions
81-----------
82
83.. glossary::
84
85  Sign Extend
86    To `sign extend an` ``X`` `-bit number, A, to a` ``Y`` `-bit number, B  ,` means to
87
88    #. Copy all ``X`` bits from `A` to the lower ``X`` bits of `B`.
89    #. Set the value of the remaining ``Y`` - ``X`` bits of `B` to the value of
90       the  most-significant bit of `A`.
91
92.. admonition:: Example
93
94  Sign extend an 8-bit number ``A`` to a 16-bit number ``B`` on a big-endian platform:
95  ::
96
97    A:          10000110
98    B: 11111111 10000110
99
100Conformance groups
101------------------
102
103An implementation does not need to support all instructions specified in this
104document (e.g., deprecated instructions).  Instead, a number of conformance
105groups are specified.  An implementation must support the base32 conformance
106group and may support additional conformance groups, where supporting a
107conformance group means it must support all instructions in that conformance
108group.
109
110The use of named conformance groups enables interoperability between a runtime
111that executes instructions, and tools as such compilers that generate
112instructions for the runtime.  Thus, capability discovery in terms of
113conformance groups might be done manually by users or automatically by tools.
114
115Each conformance group has a short ASCII label (e.g., "base32") that
116corresponds to a set of instructions that are mandatory.  That is, each
117instruction has one or more conformance groups of which it is a member.
118
119This document defines the following conformance groups:
120
121* base32: includes all instructions defined in this
122  specification unless otherwise noted.
123* base64: includes base32, plus instructions explicitly noted
124  as being in the base64 conformance group.
125* atomic32: includes 32-bit atomic operation instructions (see `Atomic operations`_).
126* atomic64: includes atomic32, plus 64-bit atomic operation instructions.
127* divmul32: includes 32-bit division, multiplication, and modulo instructions.
128* divmul64: includes divmul32, plus 64-bit division, multiplication,
129  and modulo instructions.
130* packet: deprecated packet access instructions.
131
132Instruction encoding
133====================
134
135BPF has two instruction encodings:
136
137* the basic instruction encoding, which uses 64 bits to encode an instruction
138* the wide instruction encoding, which appends a second 64 bits
139  after the basic instruction for a total of 128 bits.
140
141Basic instruction encoding
142--------------------------
143
144A basic instruction is encoded as follows::
145
146  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
147  |    opcode     |     regs      |            offset             |
148  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
149  |                              imm                              |
150  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
151
152**opcode**
153  operation to perform, encoded as follows::
154
155    +-+-+-+-+-+-+-+-+
156    |specific |class|
157    +-+-+-+-+-+-+-+-+
158
159  **specific**
160    The format of these bits varies by instruction class
161
162  **class**
163    The instruction class (see `Instruction classes`_)
164
165**regs**
166  The source and destination register numbers, encoded as follows
167  on a little-endian host::
168
169    +-+-+-+-+-+-+-+-+
170    |src_reg|dst_reg|
171    +-+-+-+-+-+-+-+-+
172
173  and as follows on a big-endian host::
174
175    +-+-+-+-+-+-+-+-+
176    |dst_reg|src_reg|
177    +-+-+-+-+-+-+-+-+
178
179  **src_reg**
180    the source register number (0-10), except where otherwise specified
181    (`64-bit immediate instructions`_ reuse this field for other purposes)
182
183  **dst_reg**
184    destination register number (0-10)
185
186**offset**
187  signed integer offset used with pointer arithmetic
188
189**imm**
190  signed integer immediate value
191
192Note that the contents of multi-byte fields ('offset' and 'imm') are
193stored using big-endian byte ordering on big-endian hosts and
194little-endian byte ordering on little-endian hosts.
195
196For example::
197
198  opcode                  offset imm          assembly
199         src_reg dst_reg
200  07     0       1        00 00  44 33 22 11  r1 += 0x11223344 // little
201         dst_reg src_reg
202  07     1       0        00 00  11 22 33 44  r1 += 0x11223344 // big
203
204Note that most instructions do not use all of the fields.
205Unused fields shall be cleared to zero.
206
207Wide instruction encoding
208--------------------------
209
210Some instructions are defined to use the wide instruction encoding,
211which uses two 32-bit immediate values.  The 64 bits following
212the basic instruction format contain a pseudo instruction
213with 'opcode', 'dst_reg', 'src_reg', and 'offset' all set to zero.
214
215This is depicted in the following figure::
216
217  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
218  |    opcode     |     regs      |            offset             |
219  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
220  |                              imm                              |
221  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
222  |                           reserved                            |
223  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
224  |                           next_imm                            |
225  +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
226
227**opcode**
228  operation to perform, encoded as explained above
229
230**regs**
231  The source and destination register numbers, encoded as explained above
232
233**offset**
234  signed integer offset used with pointer arithmetic
235
236**imm**
237  signed integer immediate value
238
239**reserved**
240  unused, set to zero
241
242**next_imm**
243  second signed integer immediate value
244
245Instruction classes
246-------------------
247
248The three least significant bits of the 'opcode' field store the instruction class:
249
250=====  =====  ===============================  ===================================
251class  value  description                      reference
252=====  =====  ===============================  ===================================
253LD     0x0    non-standard load operations     `Load and store instructions`_
254LDX    0x1    load into register operations    `Load and store instructions`_
255ST     0x2    store from immediate operations  `Load and store instructions`_
256STX    0x3    store from register operations   `Load and store instructions`_
257ALU    0x4    32-bit arithmetic operations     `Arithmetic and jump instructions`_
258JMP    0x5    64-bit jump operations           `Arithmetic and jump instructions`_
259JMP32  0x6    32-bit jump operations           `Arithmetic and jump instructions`_
260ALU64  0x7    64-bit arithmetic operations     `Arithmetic and jump instructions`_
261=====  =====  ===============================  ===================================
262
263Arithmetic and jump instructions
264================================
265
266For arithmetic and jump instructions (``ALU``, ``ALU64``, ``JMP`` and
267``JMP32``), the 8-bit 'opcode' field is divided into three parts::
268
269  +-+-+-+-+-+-+-+-+
270  |  code |s|class|
271  +-+-+-+-+-+-+-+-+
272
273**code**
274  the operation code, whose meaning varies by instruction class
275
276**s (source)**
277  the source operand location, which unless otherwise specified is one of:
278
279  ======  =====  ==============================================
280  source  value  description
281  ======  =====  ==============================================
282  K       0      use 32-bit 'imm' value as source operand
283  X       1      use 'src_reg' register value as source operand
284  ======  =====  ==============================================
285
286**instruction class**
287  the instruction class (see `Instruction classes`_)
288
289Arithmetic instructions
290-----------------------
291
292``ALU`` uses 32-bit wide operands while ``ALU64`` uses 64-bit wide operands for
293otherwise identical operations. ``ALU64`` instructions belong to the
294base64 conformance group unless noted otherwise.
295The 'code' field encodes the operation as below, where 'src' and 'dst' refer
296to the values of the source and destination registers, respectively.
297
298=====  =====  =======  ==========================================================
299name   code   offset   description
300=====  =====  =======  ==========================================================
301ADD    0x0    0        dst += src
302SUB    0x1    0        dst -= src
303MUL    0x2    0        dst \*= src
304DIV    0x3    0        dst = (src != 0) ? (dst / src) : 0
305SDIV   0x3    1        dst = (src != 0) ? (dst s/ src) : 0
306OR     0x4    0        dst \|= src
307AND    0x5    0        dst &= src
308LSH    0x6    0        dst <<= (src & mask)
309RSH    0x7    0        dst >>= (src & mask)
310NEG    0x8    0        dst = -dst
311MOD    0x9    0        dst = (src != 0) ? (dst % src) : dst
312SMOD   0x9    1        dst = (src != 0) ? (dst s% src) : dst
313XOR    0xa    0        dst ^= src
314MOV    0xb    0        dst = src
315MOVSX  0xb    8/16/32  dst = (s8,s16,s32)src
316ARSH   0xc    0        :term:`sign extending<Sign Extend>` dst >>= (src & mask)
317END    0xd    0        byte swap operations (see `Byte swap instructions`_ below)
318=====  =====  =======  ==========================================================
319
320Underflow and overflow are allowed during arithmetic operations, meaning
321the 64-bit or 32-bit value will wrap. If BPF program execution would
322result in division by zero, the destination register is instead set to zero.
323If execution would result in modulo by zero, for ``ALU64`` the value of
324the destination register is unchanged whereas for ``ALU`` the upper
32532 bits of the destination register are zeroed.
326
327``{ADD, X, ALU}``, where 'code' = ``ADD``, 'source' = ``X``, and 'class' = ``ALU``, means::
328
329  dst = (u32) ((u32) dst + (u32) src)
330
331where '(u32)' indicates that the upper 32 bits are zeroed.
332
333``{ADD, X, ALU64}`` means::
334
335  dst = dst + src
336
337``{XOR, K, ALU}`` means::
338
339  dst = (u32) dst ^ (u32) imm
340
341``{XOR, K, ALU64}`` means::
342
343  dst = dst ^ imm
344
345Note that most instructions have instruction offset of 0. Only three instructions
346(``SDIV``, ``SMOD``, ``MOVSX``) have a non-zero offset.
347
348Division, multiplication, and modulo operations for ``ALU`` are part
349of the "divmul32" conformance group, and division, multiplication, and
350modulo operations for ``ALU64`` are part of the "divmul64" conformance
351group.
352The division and modulo operations support both unsigned and signed flavors.
353
354For unsigned operations (``DIV`` and ``MOD``), for ``ALU``,
355'imm' is interpreted as a 32-bit unsigned value. For ``ALU64``,
356'imm' is first :term:`sign extended<Sign Extend>` from 32 to 64 bits, and then
357interpreted as a 64-bit unsigned value.
358
359For signed operations (``SDIV`` and ``SMOD``), for ``ALU``,
360'imm' is interpreted as a 32-bit signed value. For ``ALU64``, 'imm'
361is first :term:`sign extended<Sign Extend>` from 32 to 64 bits, and then
362interpreted as a 64-bit signed value.
363
364Note that there are varying definitions of the signed modulo operation
365when the dividend or divisor are negative, where implementations often
366vary by language such that Python, Ruby, etc.  differ from C, Go, Java,
367etc. This specification requires that signed modulo use truncated division
368(where -13 % 3 == -1) as implemented in C, Go, etc.:
369
370   a % n = a - n * trunc(a / n)
371
372The ``MOVSX`` instruction does a move operation with sign extension.
373``{MOVSX, X, ALU}`` :term:`sign extends<Sign Extend>` 8-bit and 16-bit operands into 32
374bit operands, and zeroes the remaining upper 32 bits.
375``{MOVSX, X, ALU64}`` :term:`sign extends<Sign Extend>` 8-bit, 16-bit, and 32-bit
376operands into 64 bit operands.  Unlike other arithmetic instructions,
377``MOVSX`` is only defined for register source operands (``X``).
378
379The ``NEG`` instruction is only defined when the source bit is clear
380(``K``).
381
382Shift operations use a mask of 0x3F (63) for 64-bit operations and 0x1F (31)
383for 32-bit operations.
384
385Byte swap instructions
386----------------------
387
388The byte swap instructions use instruction classes of ``ALU`` and ``ALU64``
389and a 4-bit 'code' field of ``END``.
390
391The byte swap instructions operate on the destination register
392only and do not use a separate source register or immediate value.
393
394For ``ALU``, the 1-bit source operand field in the opcode is used to
395select what byte order the operation converts from or to. For
396``ALU64``, the 1-bit source operand field in the opcode is reserved
397and must be set to 0.
398
399=====  ========  =====  =================================================
400class  source    value  description
401=====  ========  =====  =================================================
402ALU    TO_LE     0      convert between host byte order and little endian
403ALU    TO_BE     1      convert between host byte order and big endian
404ALU64  Reserved  0      do byte swap unconditionally
405=====  ========  =====  =================================================
406
407The 'imm' field encodes the width of the swap operations.  The following widths
408are supported: 16, 32 and 64.  Width 64 operations belong to the base64
409conformance group and other swap operations belong to the base32
410conformance group.
411
412Examples:
413
414``{END, TO_LE, ALU}`` with imm = 16/32/64 means::
415
416  dst = htole16(dst)
417  dst = htole32(dst)
418  dst = htole64(dst)
419
420``{END, TO_BE, ALU}`` with imm = 16/32/64 means::
421
422  dst = htobe16(dst)
423  dst = htobe32(dst)
424  dst = htobe64(dst)
425
426``{END, TO_LE, ALU64}`` with imm = 16/32/64 means::
427
428  dst = bswap16(dst)
429  dst = bswap32(dst)
430  dst = bswap64(dst)
431
432Jump instructions
433-----------------
434
435``JMP32`` uses 32-bit wide operands and indicates the base32
436conformance group, while ``JMP`` uses 64-bit wide operands for
437otherwise identical operations, and indicates the base64 conformance
438group unless otherwise specified.
439The 'code' field encodes the operation as below:
440
441========  =====  =======  ===============================  ===================================================
442code      value  src_reg  description                      notes
443========  =====  =======  ===============================  ===================================================
444JA        0x0    0x0      PC += offset                     {JA, K, JMP} only
445JA        0x0    0x0      PC += imm                        {JA, K, JMP32} only
446JEQ       0x1    any      PC += offset if dst == src
447JGT       0x2    any      PC += offset if dst > src        unsigned
448JGE       0x3    any      PC += offset if dst >= src       unsigned
449JSET      0x4    any      PC += offset if dst & src
450JNE       0x5    any      PC += offset if dst != src
451JSGT      0x6    any      PC += offset if dst > src        signed
452JSGE      0x7    any      PC += offset if dst >= src       signed
453CALL      0x8    0x0      call helper function by address  {CALL, K, JMP} only, see `Helper functions`_
454CALL      0x8    0x1      call PC += imm                   {CALL, K, JMP} only, see `Program-local functions`_
455CALL      0x8    0x2      call helper function by BTF ID   {CALL, K, JMP} only, see `Helper functions`_
456EXIT      0x9    0x0      return                           {CALL, K, JMP} only
457JLT       0xa    any      PC += offset if dst < src        unsigned
458JLE       0xb    any      PC += offset if dst <= src       unsigned
459JSLT      0xc    any      PC += offset if dst < src        signed
460JSLE      0xd    any      PC += offset if dst <= src       signed
461========  =====  =======  ===============================  ===================================================
462
463The BPF program needs to store the return value into register R0 before doing an
464``EXIT``.
465
466Example:
467
468``{JSGE, X, JMP32}`` means::
469
470  if (s32)dst s>= (s32)src goto +offset
471
472where 's>=' indicates a signed '>=' comparison.
473
474``{JA, K, JMP32}`` means::
475
476  gotol +imm
477
478where 'imm' means the branch offset comes from insn 'imm' field.
479
480Note that there are two flavors of ``JA`` instructions. The
481``JMP`` class permits a 16-bit jump offset specified by the 'offset'
482field, whereas the ``JMP32`` class permits a 32-bit jump offset
483specified by the 'imm' field. A > 16-bit conditional jump may be
484converted to a < 16-bit conditional jump plus a 32-bit unconditional
485jump.
486
487All ``CALL`` and ``JA`` instructions belong to the
488base32 conformance group.
489
490Helper functions
491~~~~~~~~~~~~~~~~
492
493Helper functions are a concept whereby BPF programs can call into a
494set of function calls exposed by the underlying platform.
495
496Historically, each helper function was identified by an address
497encoded in the imm field.  The available helper functions may differ
498for each program type, but address values are unique across all program types.
499
500Platforms that support the BPF Type Format (BTF) support identifying
501a helper function by a BTF ID encoded in the imm field, where the BTF ID
502identifies the helper name and type.
503
504Program-local functions
505~~~~~~~~~~~~~~~~~~~~~~~
506Program-local functions are functions exposed by the same BPF program as the
507caller, and are referenced by offset from the call instruction, similar to
508``JA``.  The offset is encoded in the imm field of the call instruction.
509A ``EXIT`` within the program-local function will return to the caller.
510
511Load and store instructions
512===========================
513
514For load and store instructions (``LD``, ``LDX``, ``ST``, and ``STX``), the
5158-bit 'opcode' field is divided as::
516
517  +-+-+-+-+-+-+-+-+
518  |mode |sz |class|
519  +-+-+-+-+-+-+-+-+
520
521**mode**
522  The mode modifier is one of:
523
524    =============  =====  ====================================  =============
525    mode modifier  value  description                           reference
526    =============  =====  ====================================  =============
527    IMM            0      64-bit immediate instructions         `64-bit immediate instructions`_
528    ABS            1      legacy BPF packet access (absolute)   `Legacy BPF Packet access instructions`_
529    IND            2      legacy BPF packet access (indirect)   `Legacy BPF Packet access instructions`_
530    MEM            3      regular load and store operations     `Regular load and store operations`_
531    MEMSX          4      sign-extension load operations        `Sign-extension load operations`_
532    ATOMIC         6      atomic operations                     `Atomic operations`_
533    =============  =====  ====================================  =============
534
535**sz (size)**
536  The size modifier is one of:
537
538    ====  =====  =====================
539    size  value  description
540    ====  =====  =====================
541    W     0      word        (4 bytes)
542    H     1      half word   (2 bytes)
543    B     2      byte
544    DW    3      double word (8 bytes)
545    ====  =====  =====================
546
547  Instructions using ``DW`` belong to the base64 conformance group.
548
549**class**
550  The instruction class (see `Instruction classes`_)
551
552Regular load and store operations
553---------------------------------
554
555The ``MEM`` mode modifier is used to encode regular load and store
556instructions that transfer data between a register and memory.
557
558``{MEM, <size>, STX}`` means::
559
560  *(size *) (dst + offset) = src
561
562``{MEM, <size>, ST}`` means::
563
564  *(size *) (dst + offset) = imm
565
566``{MEM, <size>, LDX}`` means::
567
568  dst = *(unsigned size *) (src + offset)
569
570Where '<size>' is one of: ``B``, ``H``, ``W``, or ``DW``, and
571'unsigned size' is one of: u8, u16, u32, or u64.
572
573Sign-extension load operations
574------------------------------
575
576The ``MEMSX`` mode modifier is used to encode :term:`sign-extension<Sign Extend>` load
577instructions that transfer data between a register and memory.
578
579``{MEMSX, <size>, LDX}`` means::
580
581  dst = *(signed size *) (src + offset)
582
583Where size is one of: ``B``, ``H``, or ``W``, and
584'signed size' is one of: s8, s16, or s32.
585
586Atomic operations
587-----------------
588
589Atomic operations are operations that operate on memory and can not be
590interrupted or corrupted by other access to the same memory region
591by other BPF programs or means outside of this specification.
592
593All atomic operations supported by BPF are encoded as store operations
594that use the ``ATOMIC`` mode modifier as follows:
595
596* ``{ATOMIC, W, STX}`` for 32-bit operations, which are
597  part of the "atomic32" conformance group.
598* ``{ATOMIC, DW, STX}`` for 64-bit operations, which are
599  part of the "atomic64" conformance group.
600* 8-bit and 16-bit wide atomic operations are not supported.
601
602The 'imm' field is used to encode the actual atomic operation.
603Simple atomic operation use a subset of the values defined to encode
604arithmetic operations in the 'imm' field to encode the atomic operation:
605
606========  =====  ===========
607imm       value  description
608========  =====  ===========
609ADD       0x00   atomic add
610OR        0x40   atomic or
611AND       0x50   atomic and
612XOR       0xa0   atomic xor
613========  =====  ===========
614
615
616``{ATOMIC, W, STX}`` with 'imm' = ADD means::
617
618  *(u32 *)(dst + offset) += src
619
620``{ATOMIC, DW, STX}`` with 'imm' = ADD means::
621
622  *(u64 *)(dst + offset) += src
623
624In addition to the simple atomic operations, there also is a modifier and
625two complex atomic operations:
626
627===========  ================  ===========================
628imm          value             description
629===========  ================  ===========================
630FETCH        0x01              modifier: return old value
631XCHG         0xe0 | FETCH      atomic exchange
632CMPXCHG      0xf0 | FETCH      atomic compare and exchange
633===========  ================  ===========================
634
635The ``FETCH`` modifier is optional for simple atomic operations, and
636always set for the complex atomic operations.  If the ``FETCH`` flag
637is set, then the operation also overwrites ``src`` with the value that
638was in memory before it was modified.
639
640The ``XCHG`` operation atomically exchanges ``src`` with the value
641addressed by ``dst + offset``.
642
643The ``CMPXCHG`` operation atomically compares the value addressed by
644``dst + offset`` with ``R0``. If they match, the value addressed by
645``dst + offset`` is replaced with ``src``. In either case, the
646value that was at ``dst + offset`` before the operation is zero-extended
647and loaded back to ``R0``.
648
64964-bit immediate instructions
650-----------------------------
651
652Instructions with the ``IMM`` 'mode' modifier use the wide instruction
653encoding defined in `Instruction encoding`_, and use the 'src_reg' field of the
654basic instruction to hold an opcode subtype.
655
656The following table defines a set of ``{IMM, DW, LD}`` instructions
657with opcode subtypes in the 'src_reg' field, using new terms such as "map"
658defined further below:
659
660=======  =========================================  ===========  ==============
661src_reg  pseudocode                                 imm type     dst type
662=======  =========================================  ===========  ==============
6630x0      dst = (next_imm << 32) | imm               integer      integer
6640x1      dst = map_by_fd(imm)                       map fd       map
6650x2      dst = map_val(map_by_fd(imm)) + next_imm   map fd       data pointer
6660x3      dst = var_addr(imm)                        variable id  data pointer
6670x4      dst = code_addr(imm)                       integer      code pointer
6680x5      dst = map_by_idx(imm)                      map index    map
6690x6      dst = map_val(map_by_idx(imm)) + next_imm  map index    data pointer
670=======  =========================================  ===========  ==============
671
672where
673
674* map_by_fd(imm) means to convert a 32-bit file descriptor into an address of a map (see `Maps`_)
675* map_by_idx(imm) means to convert a 32-bit index into an address of a map
676* map_val(map) gets the address of the first value in a given map
677* var_addr(imm) gets the address of a platform variable (see `Platform Variables`_) with a given id
678* code_addr(imm) gets the address of the instruction at a specified relative offset in number of (64-bit) instructions
679* the 'imm type' can be used by disassemblers for display
680* the 'dst type' can be used for verification and JIT compilation purposes
681
682Maps
683~~~~
684
685Maps are shared memory regions accessible by BPF programs on some platforms.
686A map can have various semantics as defined in a separate document, and may or
687may not have a single contiguous memory region, but the 'map_val(map)' is
688currently only defined for maps that do have a single contiguous memory region.
689
690Each map can have a file descriptor (fd) if supported by the platform, where
691'map_by_fd(imm)' means to get the map with the specified file descriptor. Each
692BPF program can also be defined to use a set of maps associated with the
693program at load time, and 'map_by_idx(imm)' means to get the map with the given
694index in the set associated with the BPF program containing the instruction.
695
696Platform Variables
697~~~~~~~~~~~~~~~~~~
698
699Platform variables are memory regions, identified by integer ids, exposed by
700the runtime and accessible by BPF programs on some platforms.  The
701'var_addr(imm)' operation means to get the address of the memory region
702identified by the given id.
703
704Legacy BPF Packet access instructions
705-------------------------------------
706
707BPF previously introduced special instructions for access to packet data that were
708carried over from classic BPF. These instructions used an instruction
709class of ``LD``, a size modifier of ``W``, ``H``, or ``B``, and a
710mode modifier of ``ABS`` or ``IND``.  The 'dst_reg' and 'offset' fields were
711set to zero, and 'src_reg' was set to zero for ``ABS``.  However, these
712instructions are deprecated and should no longer be used.  All legacy packet
713access instructions belong to the "packet" conformance group.
714