xref: /openbmc/linux/Documentation/bpf/btf.rst (revision b737eecd)
1=====================
2BPF Type Format (BTF)
3=====================
4
51. Introduction
6***************
7
8BTF (BPF Type Format) is the metadata format which encodes the debug info
9related to BPF program/map. The name BTF was used initially to describe data
10types. The BTF was later extended to include function info for defined
11subroutines, and line info for source/line information.
12
13The debug info is used for map pretty print, function signature, etc. The
14function signature enables better bpf program/function kernel symbol. The line
15info helps generate source annotated translated byte code, jited code and
16verifier log.
17
18The BTF specification contains two parts,
19  * BTF kernel API
20  * BTF ELF file format
21
22The kernel API is the contract between user space and kernel. The kernel
23verifies the BTF info before using it. The ELF file format is a user space
24contract between ELF file and libbpf loader.
25
26The type and string sections are part of the BTF kernel API, describing the
27debug info (mostly types related) referenced by the bpf program. These two
28sections are discussed in details in :ref:`BTF_Type_String`.
29
30.. _BTF_Type_String:
31
322. BTF Type and String Encoding
33*******************************
34
35The file ``include/uapi/linux/btf.h`` provides high-level definition of how
36types/strings are encoded.
37
38The beginning of data blob must be::
39
40    struct btf_header {
41        __u16   magic;
42        __u8    version;
43        __u8    flags;
44        __u32   hdr_len;
45
46        /* All offsets are in bytes relative to the end of this header */
47        __u32   type_off;       /* offset of type section       */
48        __u32   type_len;       /* length of type section       */
49        __u32   str_off;        /* offset of string section     */
50        __u32   str_len;        /* length of string section     */
51    };
52
53The magic is ``0xeB9F``, which has different encoding for big and little
54endian systems, and can be used to test whether BTF is generated for big- or
55little-endian target. The ``btf_header`` is designed to be extensible with
56``hdr_len`` equal to ``sizeof(struct btf_header)`` when a data blob is
57generated.
58
592.1 String Encoding
60===================
61
62The first string in the string section must be a null string. The rest of
63string table is a concatenation of other null-terminated strings.
64
652.2 Type Encoding
66=================
67
68The type id ``0`` is reserved for ``void`` type. The type section is parsed
69sequentially and type id is assigned to each recognized type starting from id
70``1``. Currently, the following types are supported::
71
72    #define BTF_KIND_INT            1       /* Integer      */
73    #define BTF_KIND_PTR            2       /* Pointer      */
74    #define BTF_KIND_ARRAY          3       /* Array        */
75    #define BTF_KIND_STRUCT         4       /* Struct       */
76    #define BTF_KIND_UNION          5       /* Union        */
77    #define BTF_KIND_ENUM           6       /* Enumeration  */
78    #define BTF_KIND_FWD            7       /* Forward      */
79    #define BTF_KIND_TYPEDEF        8       /* Typedef      */
80    #define BTF_KIND_VOLATILE       9       /* Volatile     */
81    #define BTF_KIND_CONST          10      /* Const        */
82    #define BTF_KIND_RESTRICT       11      /* Restrict     */
83    #define BTF_KIND_FUNC           12      /* Function     */
84    #define BTF_KIND_FUNC_PROTO     13      /* Function Proto       */
85    #define BTF_KIND_VAR            14      /* Variable     */
86    #define BTF_KIND_DATASEC        15      /* Section      */
87
88Note that the type section encodes debug info, not just pure types.
89``BTF_KIND_FUNC`` is not a type, and it represents a defined subprogram.
90
91Each type contains the following common data::
92
93    struct btf_type {
94        __u32 name_off;
95        /* "info" bits arrangement
96         * bits  0-15: vlen (e.g. # of struct's members)
97         * bits 16-23: unused
98         * bits 24-27: kind (e.g. int, ptr, array...etc)
99         * bits 28-30: unused
100         * bit     31: kind_flag, currently used by
101         *             struct, union and fwd
102         */
103        __u32 info;
104        /* "size" is used by INT, ENUM, STRUCT and UNION.
105         * "size" tells the size of the type it is describing.
106         *
107         * "type" is used by PTR, TYPEDEF, VOLATILE, CONST, RESTRICT,
108         * FUNC and FUNC_PROTO.
109         * "type" is a type_id referring to another type.
110         */
111        union {
112                __u32 size;
113                __u32 type;
114        };
115    };
116
117For certain kinds, the common data are followed by kind-specific data. The
118``name_off`` in ``struct btf_type`` specifies the offset in the string table.
119The following sections detail encoding of each kind.
120
1212.2.1 BTF_KIND_INT
122~~~~~~~~~~~~~~~~~~
123
124``struct btf_type`` encoding requirement:
125 * ``name_off``: any valid offset
126 * ``info.kind_flag``: 0
127 * ``info.kind``: BTF_KIND_INT
128 * ``info.vlen``: 0
129 * ``size``: the size of the int type in bytes.
130
131``btf_type`` is followed by a ``u32`` with the following bits arrangement::
132
133  #define BTF_INT_ENCODING(VAL)   (((VAL) & 0x0f000000) >> 24)
134  #define BTF_INT_OFFSET(VAL)     (((VAL) & 0x00ff0000) >> 16)
135  #define BTF_INT_BITS(VAL)       ((VAL)  & 0x000000ff)
136
137The ``BTF_INT_ENCODING`` has the following attributes::
138
139  #define BTF_INT_SIGNED  (1 << 0)
140  #define BTF_INT_CHAR    (1 << 1)
141  #define BTF_INT_BOOL    (1 << 2)
142
143The ``BTF_INT_ENCODING()`` provides extra information: signedness, char, or
144bool, for the int type. The char and bool encoding are mostly useful for
145pretty print. At most one encoding can be specified for the int type.
146
147The ``BTF_INT_BITS()`` specifies the number of actual bits held by this int
148type. For example, a 4-bit bitfield encodes ``BTF_INT_BITS()`` equals to 4.
149The ``btf_type.size * 8`` must be equal to or greater than ``BTF_INT_BITS()``
150for the type. The maximum value of ``BTF_INT_BITS()`` is 128.
151
152The ``BTF_INT_OFFSET()`` specifies the starting bit offset to calculate values
153for this int. For example, a bitfield struct member has:
154
155 * btf member bit offset 100 from the start of the structure,
156 * btf member pointing to an int type,
157 * the int type has ``BTF_INT_OFFSET() = 2`` and ``BTF_INT_BITS() = 4``
158
159Then in the struct memory layout, this member will occupy ``4`` bits starting
160from bits ``100 + 2 = 102``.
161
162Alternatively, the bitfield struct member can be the following to access the
163same bits as the above:
164
165 * btf member bit offset 102,
166 * btf member pointing to an int type,
167 * the int type has ``BTF_INT_OFFSET() = 0`` and ``BTF_INT_BITS() = 4``
168
169The original intention of ``BTF_INT_OFFSET()`` is to provide flexibility of
170bitfield encoding. Currently, both llvm and pahole generate
171``BTF_INT_OFFSET() = 0`` for all int types.
172
1732.2.2 BTF_KIND_PTR
174~~~~~~~~~~~~~~~~~~
175
176``struct btf_type`` encoding requirement:
177  * ``name_off``: 0
178  * ``info.kind_flag``: 0
179  * ``info.kind``: BTF_KIND_PTR
180  * ``info.vlen``: 0
181  * ``type``: the pointee type of the pointer
182
183No additional type data follow ``btf_type``.
184
1852.2.3 BTF_KIND_ARRAY
186~~~~~~~~~~~~~~~~~~~~
187
188``struct btf_type`` encoding requirement:
189  * ``name_off``: 0
190  * ``info.kind_flag``: 0
191  * ``info.kind``: BTF_KIND_ARRAY
192  * ``info.vlen``: 0
193  * ``size/type``: 0, not used
194
195``btf_type`` is followed by one ``struct btf_array``::
196
197    struct btf_array {
198        __u32   type;
199        __u32   index_type;
200        __u32   nelems;
201    };
202
203The ``struct btf_array`` encoding:
204  * ``type``: the element type
205  * ``index_type``: the index type
206  * ``nelems``: the number of elements for this array (``0`` is also allowed).
207
208The ``index_type`` can be any regular int type (``u8``, ``u16``, ``u32``,
209``u64``, ``unsigned __int128``). The original design of including
210``index_type`` follows DWARF, which has an ``index_type`` for its array type.
211Currently in BTF, beyond type verification, the ``index_type`` is not used.
212
213The ``struct btf_array`` allows chaining through element type to represent
214multidimensional arrays. For example, for ``int a[5][6]``, the following type
215information illustrates the chaining:
216
217  * [1]: int
218  * [2]: array, ``btf_array.type = [1]``, ``btf_array.nelems = 6``
219  * [3]: array, ``btf_array.type = [2]``, ``btf_array.nelems = 5``
220
221Currently, both pahole and llvm collapse multidimensional array into
222one-dimensional array, e.g., for ``a[5][6]``, the ``btf_array.nelems`` is
223equal to ``30``. This is because the original use case is map pretty print
224where the whole array is dumped out so one-dimensional array is enough. As
225more BTF usage is explored, pahole and llvm can be changed to generate proper
226chained representation for multidimensional arrays.
227
2282.2.4 BTF_KIND_STRUCT
229~~~~~~~~~~~~~~~~~~~~~
2302.2.5 BTF_KIND_UNION
231~~~~~~~~~~~~~~~~~~~~
232
233``struct btf_type`` encoding requirement:
234  * ``name_off``: 0 or offset to a valid C identifier
235  * ``info.kind_flag``: 0 or 1
236  * ``info.kind``: BTF_KIND_STRUCT or BTF_KIND_UNION
237  * ``info.vlen``: the number of struct/union members
238  * ``info.size``: the size of the struct/union in bytes
239
240``btf_type`` is followed by ``info.vlen`` number of ``struct btf_member``.::
241
242    struct btf_member {
243        __u32   name_off;
244        __u32   type;
245        __u32   offset;
246    };
247
248``struct btf_member`` encoding:
249  * ``name_off``: offset to a valid C identifier
250  * ``type``: the member type
251  * ``offset``: <see below>
252
253If the type info ``kind_flag`` is not set, the offset contains only bit offset
254of the member. Note that the base type of the bitfield can only be int or enum
255type. If the bitfield size is 32, the base type can be either int or enum
256type. If the bitfield size is not 32, the base type must be int, and int type
257``BTF_INT_BITS()`` encodes the bitfield size.
258
259If the ``kind_flag`` is set, the ``btf_member.offset`` contains both member
260bitfield size and bit offset. The bitfield size and bit offset are calculated
261as below.::
262
263  #define BTF_MEMBER_BITFIELD_SIZE(val)   ((val) >> 24)
264  #define BTF_MEMBER_BIT_OFFSET(val)      ((val) & 0xffffff)
265
266In this case, if the base type is an int type, it must be a regular int type:
267
268  * ``BTF_INT_OFFSET()`` must be 0.
269  * ``BTF_INT_BITS()`` must be equal to ``{1,2,4,8,16} * 8``.
270
271The following kernel patch introduced ``kind_flag`` and explained why both
272modes exist:
273
274  https://github.com/torvalds/linux/commit/9d5f9f701b1891466fb3dbb1806ad97716f95cc3#diff-fa650a64fdd3968396883d2fe8215ff3
275
2762.2.6 BTF_KIND_ENUM
277~~~~~~~~~~~~~~~~~~~
278
279``struct btf_type`` encoding requirement:
280  * ``name_off``: 0 or offset to a valid C identifier
281  * ``info.kind_flag``: 0
282  * ``info.kind``: BTF_KIND_ENUM
283  * ``info.vlen``: number of enum values
284  * ``size``: 4
285
286``btf_type`` is followed by ``info.vlen`` number of ``struct btf_enum``.::
287
288    struct btf_enum {
289        __u32   name_off;
290        __s32   val;
291    };
292
293The ``btf_enum`` encoding:
294  * ``name_off``: offset to a valid C identifier
295  * ``val``: any value
296
2972.2.7 BTF_KIND_FWD
298~~~~~~~~~~~~~~~~~~
299
300``struct btf_type`` encoding requirement:
301  * ``name_off``: offset to a valid C identifier
302  * ``info.kind_flag``: 0 for struct, 1 for union
303  * ``info.kind``: BTF_KIND_FWD
304  * ``info.vlen``: 0
305  * ``type``: 0
306
307No additional type data follow ``btf_type``.
308
3092.2.8 BTF_KIND_TYPEDEF
310~~~~~~~~~~~~~~~~~~~~~~
311
312``struct btf_type`` encoding requirement:
313  * ``name_off``: offset to a valid C identifier
314  * ``info.kind_flag``: 0
315  * ``info.kind``: BTF_KIND_TYPEDEF
316  * ``info.vlen``: 0
317  * ``type``: the type which can be referred by name at ``name_off``
318
319No additional type data follow ``btf_type``.
320
3212.2.9 BTF_KIND_VOLATILE
322~~~~~~~~~~~~~~~~~~~~~~~
323
324``struct btf_type`` encoding requirement:
325  * ``name_off``: 0
326  * ``info.kind_flag``: 0
327  * ``info.kind``: BTF_KIND_VOLATILE
328  * ``info.vlen``: 0
329  * ``type``: the type with ``volatile`` qualifier
330
331No additional type data follow ``btf_type``.
332
3332.2.10 BTF_KIND_CONST
334~~~~~~~~~~~~~~~~~~~~~
335
336``struct btf_type`` encoding requirement:
337  * ``name_off``: 0
338  * ``info.kind_flag``: 0
339  * ``info.kind``: BTF_KIND_CONST
340  * ``info.vlen``: 0
341  * ``type``: the type with ``const`` qualifier
342
343No additional type data follow ``btf_type``.
344
3452.2.11 BTF_KIND_RESTRICT
346~~~~~~~~~~~~~~~~~~~~~~~~
347
348``struct btf_type`` encoding requirement:
349  * ``name_off``: 0
350  * ``info.kind_flag``: 0
351  * ``info.kind``: BTF_KIND_RESTRICT
352  * ``info.vlen``: 0
353  * ``type``: the type with ``restrict`` qualifier
354
355No additional type data follow ``btf_type``.
356
3572.2.12 BTF_KIND_FUNC
358~~~~~~~~~~~~~~~~~~~~
359
360``struct btf_type`` encoding requirement:
361  * ``name_off``: offset to a valid C identifier
362  * ``info.kind_flag``: 0
363  * ``info.kind``: BTF_KIND_FUNC
364  * ``info.vlen``: 0
365  * ``type``: a BTF_KIND_FUNC_PROTO type
366
367No additional type data follow ``btf_type``.
368
369A BTF_KIND_FUNC defines not a type, but a subprogram (function) whose
370signature is defined by ``type``. The subprogram is thus an instance of that
371type. The BTF_KIND_FUNC may in turn be referenced by a func_info in the
372:ref:`BTF_Ext_Section` (ELF) or in the arguments to :ref:`BPF_Prog_Load`
373(ABI).
374
3752.2.13 BTF_KIND_FUNC_PROTO
376~~~~~~~~~~~~~~~~~~~~~~~~~~
377
378``struct btf_type`` encoding requirement:
379  * ``name_off``: 0
380  * ``info.kind_flag``: 0
381  * ``info.kind``: BTF_KIND_FUNC_PROTO
382  * ``info.vlen``: # of parameters
383  * ``type``: the return type
384
385``btf_type`` is followed by ``info.vlen`` number of ``struct btf_param``.::
386
387    struct btf_param {
388        __u32   name_off;
389        __u32   type;
390    };
391
392If a BTF_KIND_FUNC_PROTO type is referred by a BTF_KIND_FUNC type, then
393``btf_param.name_off`` must point to a valid C identifier except for the
394possible last argument representing the variable argument. The btf_param.type
395refers to parameter type.
396
397If the function has variable arguments, the last parameter is encoded with
398``name_off = 0`` and ``type = 0``.
399
4002.2.14 BTF_KIND_VAR
401~~~~~~~~~~~~~~~~~~~
402
403``struct btf_type`` encoding requirement:
404  * ``name_off``: offset to a valid C identifier
405  * ``info.kind_flag``: 0
406  * ``info.kind``: BTF_KIND_VAR
407  * ``info.vlen``: 0
408  * ``type``: the type of the variable
409
410``btf_type`` is followed by a single ``struct btf_variable`` with the
411following data::
412
413    struct btf_var {
414        __u32   linkage;
415    };
416
417``struct btf_var`` encoding:
418  * ``linkage``: currently only static variable 0, or globally allocated
419                 variable in ELF sections 1
420
421Not all type of global variables are supported by LLVM at this point.
422The following is currently available:
423
424  * static variables with or without section attributes
425  * global variables with section attributes
426
427The latter is for future extraction of map key/value type id's from a
428map definition.
429
4302.2.15 BTF_KIND_DATASEC
431~~~~~~~~~~~~~~~~~~~~~~~
432
433``struct btf_type`` encoding requirement:
434  * ``name_off``: offset to a valid name associated with a variable or
435                  one of .data/.bss/.rodata
436  * ``info.kind_flag``: 0
437  * ``info.kind``: BTF_KIND_DATASEC
438  * ``info.vlen``: # of variables
439  * ``size``: total section size in bytes (0 at compilation time, patched
440              to actual size by BPF loaders such as libbpf)
441
442``btf_type`` is followed by ``info.vlen`` number of ``struct btf_var_secinfo``.::
443
444    struct btf_var_secinfo {
445        __u32   type;
446        __u32   offset;
447        __u32   size;
448    };
449
450``struct btf_var_secinfo`` encoding:
451  * ``type``: the type of the BTF_KIND_VAR variable
452  * ``offset``: the in-section offset of the variable
453  * ``size``: the size of the variable in bytes
454
4553. BTF Kernel API
456*****************
457
458The following bpf syscall command involves BTF:
459   * BPF_BTF_LOAD: load a blob of BTF data into kernel
460   * BPF_MAP_CREATE: map creation with btf key and value type info.
461   * BPF_PROG_LOAD: prog load with btf function and line info.
462   * BPF_BTF_GET_FD_BY_ID: get a btf fd
463   * BPF_OBJ_GET_INFO_BY_FD: btf, func_info, line_info
464     and other btf related info are returned.
465
466The workflow typically looks like:
467::
468
469  Application:
470      BPF_BTF_LOAD
471          |
472          v
473      BPF_MAP_CREATE and BPF_PROG_LOAD
474          |
475          V
476      ......
477
478  Introspection tool:
479      ......
480      BPF_{PROG,MAP}_GET_NEXT_ID (get prog/map id's)
481          |
482          V
483      BPF_{PROG,MAP}_GET_FD_BY_ID (get a prog/map fd)
484          |
485          V
486      BPF_OBJ_GET_INFO_BY_FD (get bpf_prog_info/bpf_map_info with btf_id)
487          |                                     |
488          V                                     |
489      BPF_BTF_GET_FD_BY_ID (get btf_fd)         |
490          |                                     |
491          V                                     |
492      BPF_OBJ_GET_INFO_BY_FD (get btf)          |
493          |                                     |
494          V                                     V
495      pretty print types, dump func signatures and line info, etc.
496
497
4983.1 BPF_BTF_LOAD
499================
500
501Load a blob of BTF data into kernel. A blob of data, described in
502:ref:`BTF_Type_String`, can be directly loaded into the kernel. A ``btf_fd``
503is returned to a userspace.
504
5053.2 BPF_MAP_CREATE
506==================
507
508A map can be created with ``btf_fd`` and specified key/value type id.::
509
510    __u32   btf_fd;         /* fd pointing to a BTF type data */
511    __u32   btf_key_type_id;        /* BTF type_id of the key */
512    __u32   btf_value_type_id;      /* BTF type_id of the value */
513
514In libbpf, the map can be defined with extra annotation like below:
515::
516
517    struct bpf_map_def SEC("maps") btf_map = {
518        .type = BPF_MAP_TYPE_ARRAY,
519        .key_size = sizeof(int),
520        .value_size = sizeof(struct ipv_counts),
521        .max_entries = 4,
522    };
523    BPF_ANNOTATE_KV_PAIR(btf_map, int, struct ipv_counts);
524
525Here, the parameters for macro BPF_ANNOTATE_KV_PAIR are map name, key and
526value types for the map. During ELF parsing, libbpf is able to extract
527key/value type_id's and assign them to BPF_MAP_CREATE attributes
528automatically.
529
530.. _BPF_Prog_Load:
531
5323.3 BPF_PROG_LOAD
533=================
534
535During prog_load, func_info and line_info can be passed to kernel with proper
536values for the following attributes:
537::
538
539    __u32           insn_cnt;
540    __aligned_u64   insns;
541    ......
542    __u32           prog_btf_fd;    /* fd pointing to BTF type data */
543    __u32           func_info_rec_size;     /* userspace bpf_func_info size */
544    __aligned_u64   func_info;      /* func info */
545    __u32           func_info_cnt;  /* number of bpf_func_info records */
546    __u32           line_info_rec_size;     /* userspace bpf_line_info size */
547    __aligned_u64   line_info;      /* line info */
548    __u32           line_info_cnt;  /* number of bpf_line_info records */
549
550The func_info and line_info are an array of below, respectively.::
551
552    struct bpf_func_info {
553        __u32   insn_off; /* [0, insn_cnt - 1] */
554        __u32   type_id;  /* pointing to a BTF_KIND_FUNC type */
555    };
556    struct bpf_line_info {
557        __u32   insn_off; /* [0, insn_cnt - 1] */
558        __u32   file_name_off; /* offset to string table for the filename */
559        __u32   line_off; /* offset to string table for the source line */
560        __u32   line_col; /* line number and column number */
561    };
562
563func_info_rec_size is the size of each func_info record, and
564line_info_rec_size is the size of each line_info record. Passing the record
565size to kernel make it possible to extend the record itself in the future.
566
567Below are requirements for func_info:
568  * func_info[0].insn_off must be 0.
569  * the func_info insn_off is in strictly increasing order and matches
570    bpf func boundaries.
571
572Below are requirements for line_info:
573  * the first insn in each func must have a line_info record pointing to it.
574  * the line_info insn_off is in strictly increasing order.
575
576For line_info, the line number and column number are defined as below:
577::
578
579    #define BPF_LINE_INFO_LINE_NUM(line_col)        ((line_col) >> 10)
580    #define BPF_LINE_INFO_LINE_COL(line_col)        ((line_col) & 0x3ff)
581
5823.4 BPF_{PROG,MAP}_GET_NEXT_ID
583==============================
584
585In kernel, every loaded program, map or btf has a unique id. The id won't
586change during the lifetime of a program, map, or btf.
587
588The bpf syscall command BPF_{PROG,MAP}_GET_NEXT_ID returns all id's, one for
589each command, to user space, for bpf program or maps, respectively, so an
590inspection tool can inspect all programs and maps.
591
5923.5 BPF_{PROG,MAP}_GET_FD_BY_ID
593===============================
594
595An introspection tool cannot use id to get details about program or maps.
596A file descriptor needs to be obtained first for reference-counting purpose.
597
5983.6 BPF_OBJ_GET_INFO_BY_FD
599==========================
600
601Once a program/map fd is acquired, an introspection tool can get the detailed
602information from kernel about this fd, some of which are BTF-related. For
603example, ``bpf_map_info`` returns ``btf_id`` and key/value type ids.
604``bpf_prog_info`` returns ``btf_id``, func_info, and line info for translated
605bpf byte codes, and jited_line_info.
606
6073.7 BPF_BTF_GET_FD_BY_ID
608========================
609
610With ``btf_id`` obtained in ``bpf_map_info`` and ``bpf_prog_info``, bpf
611syscall command BPF_BTF_GET_FD_BY_ID can retrieve a btf fd. Then, with
612command BPF_OBJ_GET_INFO_BY_FD, the btf blob, originally loaded into the
613kernel with BPF_BTF_LOAD, can be retrieved.
614
615With the btf blob, ``bpf_map_info``, and ``bpf_prog_info``, an introspection
616tool has full btf knowledge and is able to pretty print map key/values, dump
617func signatures and line info, along with byte/jit codes.
618
6194. ELF File Format Interface
620****************************
621
6224.1 .BTF section
623================
624
625The .BTF section contains type and string data. The format of this section is
626same as the one describe in :ref:`BTF_Type_String`.
627
628.. _BTF_Ext_Section:
629
6304.2 .BTF.ext section
631====================
632
633The .BTF.ext section encodes func_info and line_info which needs loader
634manipulation before loading into the kernel.
635
636The specification for .BTF.ext section is defined at ``tools/lib/bpf/btf.h``
637and ``tools/lib/bpf/btf.c``.
638
639The current header of .BTF.ext section::
640
641    struct btf_ext_header {
642        __u16   magic;
643        __u8    version;
644        __u8    flags;
645        __u32   hdr_len;
646
647        /* All offsets are in bytes relative to the end of this header */
648        __u32   func_info_off;
649        __u32   func_info_len;
650        __u32   line_info_off;
651        __u32   line_info_len;
652    };
653
654It is very similar to .BTF section. Instead of type/string section, it
655contains func_info and line_info section. See :ref:`BPF_Prog_Load` for details
656about func_info and line_info record format.
657
658The func_info is organized as below.::
659
660     func_info_rec_size
661     btf_ext_info_sec for section #1 /* func_info for section #1 */
662     btf_ext_info_sec for section #2 /* func_info for section #2 */
663     ...
664
665``func_info_rec_size`` specifies the size of ``bpf_func_info`` structure when
666.BTF.ext is generated. ``btf_ext_info_sec``, defined below, is a collection of
667func_info for each specific ELF section.::
668
669     struct btf_ext_info_sec {
670        __u32   sec_name_off; /* offset to section name */
671        __u32   num_info;
672        /* Followed by num_info * record_size number of bytes */
673        __u8    data[0];
674     };
675
676Here, num_info must be greater than 0.
677
678The line_info is organized as below.::
679
680     line_info_rec_size
681     btf_ext_info_sec for section #1 /* line_info for section #1 */
682     btf_ext_info_sec for section #2 /* line_info for section #2 */
683     ...
684
685``line_info_rec_size`` specifies the size of ``bpf_line_info`` structure when
686.BTF.ext is generated.
687
688The interpretation of ``bpf_func_info->insn_off`` and
689``bpf_line_info->insn_off`` is different between kernel API and ELF API. For
690kernel API, the ``insn_off`` is the instruction offset in the unit of ``struct
691bpf_insn``. For ELF API, the ``insn_off`` is the byte offset from the
692beginning of section (``btf_ext_info_sec->sec_name_off``).
693
6944.2 .BTF_ids section
695====================
696
697The .BTF_ids section encodes BTF ID values that are used within the kernel.
698
699This section is created during the kernel compilation with the help of
700macros defined in ``include/linux/btf_ids.h`` header file. Kernel code can
701use them to create lists and sets (sorted lists) of BTF ID values.
702
703The ``BTF_ID_LIST`` and ``BTF_ID`` macros define unsorted list of BTF ID values,
704with following syntax::
705
706  BTF_ID_LIST(list)
707  BTF_ID(type1, name1)
708  BTF_ID(type2, name2)
709
710resulting in following layout in .BTF_ids section::
711
712  __BTF_ID__type1__name1__1:
713  .zero 4
714  __BTF_ID__type2__name2__2:
715  .zero 4
716
717The ``u32 list[];`` variable is defined to access the list.
718
719The ``BTF_ID_UNUSED`` macro defines 4 zero bytes. It's used when we
720want to define unused entry in BTF_ID_LIST, like::
721
722      BTF_ID_LIST(bpf_skb_output_btf_ids)
723      BTF_ID(struct, sk_buff)
724      BTF_ID_UNUSED
725      BTF_ID(struct, task_struct)
726
727The ``BTF_SET_START/END`` macros pair defines sorted list of BTF ID values
728and their count, with following syntax::
729
730  BTF_SET_START(set)
731  BTF_ID(type1, name1)
732  BTF_ID(type2, name2)
733  BTF_SET_END(set)
734
735resulting in following layout in .BTF_ids section::
736
737  __BTF_ID__set__set:
738  .zero 4
739  __BTF_ID__type1__name1__3:
740  .zero 4
741  __BTF_ID__type2__name2__4:
742  .zero 4
743
744The ``struct btf_id_set set;`` variable is defined to access the list.
745
746The ``typeX`` name can be one of following::
747
748   struct, union, typedef, func
749
750and is used as a filter when resolving the BTF ID value.
751
752All the BTF ID lists and sets are compiled in the .BTF_ids section and
753resolved during the linking phase of kernel build by ``resolve_btfids`` tool.
754
7555. Using BTF
756************
757
7585.1 bpftool map pretty print
759============================
760
761With BTF, the map key/value can be printed based on fields rather than simply
762raw bytes. This is especially valuable for large structure or if your data
763structure has bitfields. For example, for the following map,::
764
765      enum A { A1, A2, A3, A4, A5 };
766      typedef enum A ___A;
767      struct tmp_t {
768           char a1:4;
769           int  a2:4;
770           int  :4;
771           __u32 a3:4;
772           int b;
773           ___A b1:4;
774           enum A b2:4;
775      };
776      struct bpf_map_def SEC("maps") tmpmap = {
777           .type = BPF_MAP_TYPE_ARRAY,
778           .key_size = sizeof(__u32),
779           .value_size = sizeof(struct tmp_t),
780           .max_entries = 1,
781      };
782      BPF_ANNOTATE_KV_PAIR(tmpmap, int, struct tmp_t);
783
784bpftool is able to pretty print like below:
785::
786
787      [{
788            "key": 0,
789            "value": {
790                "a1": 0x2,
791                "a2": 0x4,
792                "a3": 0x6,
793                "b": 7,
794                "b1": 0x8,
795                "b2": 0xa
796            }
797        }
798      ]
799
8005.2 bpftool prog dump
801=====================
802
803The following is an example showing how func_info and line_info can help prog
804dump with better kernel symbol names, function prototypes and line
805information.::
806
807    $ bpftool prog dump jited pinned /sys/fs/bpf/test_btf_haskv
808    [...]
809    int test_long_fname_2(struct dummy_tracepoint_args * arg):
810    bpf_prog_44a040bf25481309_test_long_fname_2:
811    ; static int test_long_fname_2(struct dummy_tracepoint_args *arg)
812       0:   push   %rbp
813       1:   mov    %rsp,%rbp
814       4:   sub    $0x30,%rsp
815       b:   sub    $0x28,%rbp
816       f:   mov    %rbx,0x0(%rbp)
817      13:   mov    %r13,0x8(%rbp)
818      17:   mov    %r14,0x10(%rbp)
819      1b:   mov    %r15,0x18(%rbp)
820      1f:   xor    %eax,%eax
821      21:   mov    %rax,0x20(%rbp)
822      25:   xor    %esi,%esi
823    ; int key = 0;
824      27:   mov    %esi,-0x4(%rbp)
825    ; if (!arg->sock)
826      2a:   mov    0x8(%rdi),%rdi
827    ; if (!arg->sock)
828      2e:   cmp    $0x0,%rdi
829      32:   je     0x0000000000000070
830      34:   mov    %rbp,%rsi
831    ; counts = bpf_map_lookup_elem(&btf_map, &key);
832    [...]
833
8345.3 Verifier Log
835================
836
837The following is an example of how line_info can help debugging verification
838failure.::
839
840       /* The code at tools/testing/selftests/bpf/test_xdp_noinline.c
841        * is modified as below.
842        */
843       data = (void *)(long)xdp->data;
844       data_end = (void *)(long)xdp->data_end;
845       /*
846       if (data + 4 > data_end)
847               return XDP_DROP;
848       */
849       *(u32 *)data = dst->dst;
850
851    $ bpftool prog load ./test_xdp_noinline.o /sys/fs/bpf/test_xdp_noinline type xdp
852        ; data = (void *)(long)xdp->data;
853        224: (79) r2 = *(u64 *)(r10 -112)
854        225: (61) r2 = *(u32 *)(r2 +0)
855        ; *(u32 *)data = dst->dst;
856        226: (63) *(u32 *)(r2 +0) = r1
857        invalid access to packet, off=0 size=4, R2(id=0,off=0,r=0)
858        R2 offset is outside of the packet
859
8606. BTF Generation
861*****************
862
863You need latest pahole
864
865  https://git.kernel.org/pub/scm/devel/pahole/pahole.git/
866
867or llvm (8.0 or later). The pahole acts as a dwarf2btf converter. It doesn't
868support .BTF.ext and btf BTF_KIND_FUNC type yet. For example,::
869
870      -bash-4.4$ cat t.c
871      struct t {
872        int a:2;
873        int b:3;
874        int c:2;
875      } g;
876      -bash-4.4$ gcc -c -O2 -g t.c
877      -bash-4.4$ pahole -JV t.o
878      File t.o:
879      [1] STRUCT t kind_flag=1 size=4 vlen=3
880              a type_id=2 bitfield_size=2 bits_offset=0
881              b type_id=2 bitfield_size=3 bits_offset=2
882              c type_id=2 bitfield_size=2 bits_offset=5
883      [2] INT int size=4 bit_offset=0 nr_bits=32 encoding=SIGNED
884
885The llvm is able to generate .BTF and .BTF.ext directly with -g for bpf target
886only. The assembly code (-S) is able to show the BTF encoding in assembly
887format.::
888
889    -bash-4.4$ cat t2.c
890    typedef int __int32;
891    struct t2 {
892      int a2;
893      int (*f2)(char q1, __int32 q2, ...);
894      int (*f3)();
895    } g2;
896    int main() { return 0; }
897    int test() { return 0; }
898    -bash-4.4$ clang -c -g -O2 -target bpf t2.c
899    -bash-4.4$ readelf -S t2.o
900      ......
901      [ 8] .BTF              PROGBITS         0000000000000000  00000247
902           000000000000016e  0000000000000000           0     0     1
903      [ 9] .BTF.ext          PROGBITS         0000000000000000  000003b5
904           0000000000000060  0000000000000000           0     0     1
905      [10] .rel.BTF.ext      REL              0000000000000000  000007e0
906           0000000000000040  0000000000000010          16     9     8
907      ......
908    -bash-4.4$ clang -S -g -O2 -target bpf t2.c
909    -bash-4.4$ cat t2.s
910      ......
911            .section        .BTF,"",@progbits
912            .short  60319                   # 0xeb9f
913            .byte   1
914            .byte   0
915            .long   24
916            .long   0
917            .long   220
918            .long   220
919            .long   122
920            .long   0                       # BTF_KIND_FUNC_PROTO(id = 1)
921            .long   218103808               # 0xd000000
922            .long   2
923            .long   83                      # BTF_KIND_INT(id = 2)
924            .long   16777216                # 0x1000000
925            .long   4
926            .long   16777248                # 0x1000020
927      ......
928            .byte   0                       # string offset=0
929            .ascii  ".text"                 # string offset=1
930            .byte   0
931            .ascii  "/home/yhs/tmp-pahole/t2.c" # string offset=7
932            .byte   0
933            .ascii  "int main() { return 0; }" # string offset=33
934            .byte   0
935            .ascii  "int test() { return 0; }" # string offset=58
936            .byte   0
937            .ascii  "int"                   # string offset=83
938      ......
939            .section        .BTF.ext,"",@progbits
940            .short  60319                   # 0xeb9f
941            .byte   1
942            .byte   0
943            .long   24
944            .long   0
945            .long   28
946            .long   28
947            .long   44
948            .long   8                       # FuncInfo
949            .long   1                       # FuncInfo section string offset=1
950            .long   2
951            .long   .Lfunc_begin0
952            .long   3
953            .long   .Lfunc_begin1
954            .long   5
955            .long   16                      # LineInfo
956            .long   1                       # LineInfo section string offset=1
957            .long   2
958            .long   .Ltmp0
959            .long   7
960            .long   33
961            .long   7182                    # Line 7 Col 14
962            .long   .Ltmp3
963            .long   7
964            .long   58
965            .long   8206                    # Line 8 Col 14
966
9677. Testing
968**********
969
970Kernel bpf selftest `test_btf.c` provides extensive set of BTF-related tests.
971