1pahole(1) dwarves pahole(1)
2
6 pahole - Shows, manipulates data structure layout and pretty prints raw
7 data.
8
10 pahole [options] files
11
13 pahole shows data structure layouts encoded in debugging information
14 formats, DWARF, CTF and BTF being supported.
15 This is useful for, among other things: optimizing important data
17 structures by reducing its size, figuring out what is the field sitting
18 at an offset from the start of a data structure, investigating ABI
19 changes and more generally understanding a new codebase you have to
20 work with.
21 It also uses these structure layouts to pretty print data feed to its
23 standard input, e.g.:
24 $ pahole --header elf64_hdr < /lib/modules/5.8.0-rc6+/build/vmlinux
26 {
27 .e_ident = { 127, 69, 76, 70, 2, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
28 .e_type = 2,
29 .e_machine = 62,
30 .e_version = 1,
31 .e_entry = 16777216,
32 .e_phoff = 64,
33 .e_shoff = 604653784,
34 .e_flags = 0,
35 .e_ehsize = 64,
36 .e_phentsize = 56,
37 .e_phnum = 5,
38 .e_shentsize = 64,
39 .e_shnum = 80,
40 .e_shstrndx = 79,
41 },
42 $
43 See the PRETTY PRINTING section for further examples and documentation.
45 The files must have associated debugging information. This information
47 may be inside the file itself, in ELF sections, or in another file.
48 One way to have this information is to specify the -g option to the
50 compiler when building it. When this is done the information will be
51 stored in an ELF section. For the DWARF debugging information format
52 this, adds, among others, the .debug_info ELF section. For CTF it is
53 found in just one ELF section, .SUNW_ctf. BTF comes in at least the
54 .BTF ELF section, and may come also with the .BTF.ext ELF section.
55 The debuginfo packages available in most Linux distributions are also
57 supported by pahole, where the debugging information is available in a
58 separate file.
59 By default, pahole shows the layout of all named structs in the files
61 specified.
62 If no files are specified, then it will look if the /sys/kernel/btf/vm‐
64 linux is present, using the BTF information present in it about the
65 running kernel, i.e. this works:
66 $ pahole list_head
68 struct list_head {
69 struct list_head * next; /* 0 8 */
70 struct list_head * prev; /* 8 8 */
71 /* size: 16, cachelines: 1, members: 2 */
73 /* last cacheline: 16 bytes */
74 };
75 $
76 If BTF is not present and no file is passed, then a vmlinux that
78 matches the build-id for the running kernel will be looked up in the
79 usual places, including where the kernel debuginfo packages put it,
80 looking for DWARF info instead.
81 See the EXAMPLES section for more usage suggestions.
83 It also pretty prints whatever is fed to its standard input, according
85 to the type specified, see the EXAMPLE session.
86 Use --count to state how many records should be pretty printed.
88
91 pahole supports the following options.
92 -C, --class_name=CLASS_NAMES
95 Show just these classes. This can be a comma separated list of
96 class names or file URLs (e.g.: file://class_list.txt)
97 -c, --cacheline_size=SIZE
100 Set cacheline size to SIZE bytes.
101 --count=COUNT
104 Pretty print the first COUNT records from input.
105 --skip=COUNT
108 Skip COUNT input records.
109 -E, --expand_types
112 Expand class members. Useful to find in what member of inner
113 structs where an offset from the beginning of a struct is.
114 -F, --format_path
117 Allows specifying a list of debugging formats to try, in order.
118 Right now this includes "ctf" and "dwarf". The default format
119 path used is equivalent to "-F dwarf,ctf".
120 --hex Print offsets and sizes in hexadecimal.
123 -r, --rel_offset
126 Show relative offsets of members in inner structs.
127 -p, --expand_pointers
130 Expand class pointer members.
131 -R, --reorganize
134 Reorganize struct, demoting and combining bitfields, moving mem‐
135 bers to remove alignment holes and padding.
136 -S, --show_reorg_steps
139 Show the struct layout at each reorganization step.
140 -i, --contains=CLASS_NAME
143 Show classes that contains CLASS_NAME.
144 -a, --anon_include
147 Include anonymous classes.
148 -A, --nested_anon_include
151 Include nested (inside other structs) anonymous classes.
152 -B, --bit_holes=NR_HOLES
155 Show only structs at least NR_HOLES bit holes.
156 -d, --recursive
159 Recursive mode, affects several other flags.
160 -D, --decl_exclude=PREFIX
163 exclude classes declared in files with PREFIX.
164 -f, --find_pointers_to=CLASS_NAME
167 Find pointers to CLASS_NAME.
168 -H, --holes=NR_HOLES
171 Show only structs with at least NR_HOLES holes.
172 -I, --show_decl_info
175 Show the file and line number where the tags were defined, if
176 available in the debugging information.
177 --skip_encoding_btf_vars
180 Do not encode VARs in BTF.
181 -J, --btf_encode
184 Encode BTF information from DWARF, used in the Linux kernel
185 build process when CONFIG_DEBUG_INFO_BTF=y is present, intro‐
186 duced in Linux v5.2. Used to implement features such as BPF CO-
187 RE (Compile Once - Run Everywhere).
188 See https://nakryiko.com/posts/bpf-portability-and-co-re/.
190 --btf_encode_force
193 Ignore those symbols found invalid when encoding BTF.
194 --btf_base=PATH
197 Path to the base BTF file, for instance: vmlinux when encoding
198 kernel module BTF information. This may be inferred when asking
199 for a /sys/kernel/btf/MODULE, when it will be autoconfigured to
200 "/sys/kernel/btf/vmlinux".
201 --btf_gen_floats
204 Allow producing BTF_KIND_FLOAT entries in systems where the vm‐
205 linux DWARF information has float types.
206 --btf_gen_all
209 Allow using all the BTF features supported by pahole.
210 -l, --show_first_biggest_size_base_type_member
213 Show first biggest size base_type member.
214 -m, --nr_methods
217 Show number of methods.
218 -M, --show_only_data_members
221 Show only the members that use space in the class layout. C++
222 methods will be suppressed.
223 -n, --nr_members
226 Show number of members.
227 -N, --class_name_len
230 Show size of classes.
231 -O, --dwarf_offset=OFFSET
234 Show tag with DWARF OFFSET.
235 -P, --packable
238 Show only structs that has holes that can be packed if members
239 are reorganized, for instance when using the --reorganize op‐
240 tion.
241 -q, --quiet
244 Be quieter.
245 -s, --sizes
248 Show size of classes.
249 -t, --separator=SEP
252 Use SEP as the field separator.
253 -T, --nr_definitions
256 Show how many times struct was defined.
257 -u, --defined_in
260 Show CUs where CLASS_NAME (-C) is defined.
261 --flat_arrays
264 Flatten arrays, so that array[10][2] becomes array[20]. Useful
265 when generating from both CTF/BTF and DWARF encodings for the
266 same binary for testing purposes.
267 --suppress_aligned_attribute
270 Suppress forced alignment markers, so that one can compare BTF
271 or CTF output, that don't have that info, to output from DWARF
272 >= 5.
273 --suppress_force_paddings
276 Suppress bitfield forced padding at the end of structs, as this
278 requires something like DWARF's DW_AT_alignment, so that one can
279 compare BTF or CTF output, that don't have that info.
280 --suppress_packed
283 Suppress the output of the inference of __attri‐
285 bute__((__packed__)), so that one can compare BTF or CTF output,
286 the inference algorithm uses things like DW_AT_alignment, so un‐
287 til it is improved to infer that as well for BTF, allow dis‐
288 abling this output.
289 --fixup_silly_bitfields
292 Converts silly bitfields such as "int foo:32" to plain "int
293 foo".
294 -V, --verbose
297 be verbose
298 -w, --word_size=WORD_SIZE
301 Change the arch word size to WORD_SIZE.
302 -x, --exclude=PREFIX
305 Exclude PREFIXed classes.
306 -X, --cu_exclude=PREFIX
309 Exclude PREFIXed compilation units.
310 -y, --prefix_filter=PREFIX
313 Include PREFIXed classes.
314 -z, --hole_size_ge=HOLE_SIZE
317 Show only structs with at least one hole greater or equal to
318 HOLE_SIZE.
319 --structs
322 Show only structs, all the other filters apply, i.e. to show
323 just the sizes of all structs combine --structs with --sizes,
324 etc.
325 --packed
328 Show only packed structs, all the other filters apply, i.e. to
329 show just the sizes of all packed structs combine --packed with
330 --sizes, etc.
331 --unions
334 Show only unions, all the other filters apply, i.e. to show just
335 the sizes of all unions combine --union with --sizes, etc.
336 --version
339 Show a traditional string version, i.e.: "v1.18".
340 --numeric_version
343 Show a numeric only version, suitable for use in Makefiles and
344 scripts where one wants to know what if the installed version
345 has some feature, i.e.: 118 instead of "v1.18".
346
349 To enable the generation of debugging information in the Linux kernel
350 build process select CONFIG_DEBUG_INFO. This can be done using make
351 menuconfig by this path: "Kernel Hacking" -> "Compile-time checks and
352 compiler options" -> "Compile the kernel with debug info". Consider as
353 well enabling CONFIG_DEBUG_INFO_BTF by going thru the aforementioned
354 menuconfig path and then selecting "Generate BTF typeinfo". Most modern
355 distributions with eBPF support should come with that in all its ker‐
356 nels, greatly facilitating the use of pahole.
357 Many distributions also come with debuginfo packages, so just enable it
359 in your package manager repository configuration and install the ker‐
360 nel-debuginfo, or any other userspace program written in a language
361 that the compiler generates debuginfo (C, C++, for instance).
362
365 All the examples here use either /sys/kernel/btf/vmlinux, if present,
366 or lookup a vmlinux file matching the running kernel, using the build-
367 id info found in /sys/kernel/notes to make sure it matches.
368 Show a type:
370 $ pahole -C __u64
372 typedef long long unsigned int __u64;
373 $
374 Works as well if the only argument is a type name:
377 $ pahole raw_spinlock_t
379 typedef struct raw_spinlock raw_spinlock_t;
380 $
381 Multiple types can be passed, separated by commas:
384 $ pahole raw_spinlock_t,raw_spinlock
386 struct raw_spinlock {
387 arch_spinlock_t raw_lock; /* 0 4 */
388 /* size: 4, cachelines: 1, members: 1 */
390 /* last cacheline: 4 bytes */
391 };
392 typedef struct raw_spinlock raw_spinlock_t;
393 $
394 Types can be expanded:
397 $ pahole -E raw_spinlock
399 struct raw_spinlock {
400 /* typedef arch_spinlock_t */ struct qspinlock {
401 union {
402 /* typedef atomic_t */ struct {
403 int counter; /* 0 4 */
404 } val; /* 0 4 */
405 struct {
406 /* typedef u8 -> __u8 */ unsigned char locked; /* 0 1 */
407 /* typedef u8 -> __u8 */ unsigned char pending; /* 1 1 */
408 }; /* 0 2 */
409 struct {
410 /* typedef u16 -> __u16 */ short unsigned int locked_pending; /* 0 2 */
411 /* typedef u16 -> __u16 */ short unsigned int tail; /* 2 2 */
412 }; /* 0 4 */
413 }; /* 0 4 */
414 } raw_lock; /* 0 4 */
415 /* size: 4, cachelines: 1, members: 1 */
417 /* last cacheline: 4 bytes */
418 };
419 $
420 When decoding OOPSes you may want to see the offsets and sizes in hexa‐
423 decimal:
424 $ pahole --hex thread_struct
426 struct thread_struct {
427 struct desc_struct tls_array[3]; /* 0 0x18 */
428 long unsigned int sp; /* 0x18 0x8 */
429 short unsigned int es; /* 0x20 0x2 */
430 short unsigned int ds; /* 0x22 0x2 */
431 short unsigned int fsindex; /* 0x24 0x2 */
432 short unsigned int gsindex; /* 0x26 0x2 */
433 long unsigned int fsbase; /* 0x28 0x8 */
434 long unsigned int gsbase; /* 0x30 0x8 */
435 struct perf_event * ptrace_bps[4]; /* 0x38 0x20 */
436 /* --- cacheline 1 boundary (64 bytes) was 24 bytes ago --- */
437 long unsigned int debugreg6; /* 0x58 0x8 */
438 long unsigned int ptrace_dr7; /* 0x60 0x8 */
439 long unsigned int cr2; /* 0x68 0x8 */
440 long unsigned int trap_nr; /* 0x70 0x8 */
441 long unsigned int error_code; /* 0x78 0x8 */
442 /* --- cacheline 2 boundary (128 bytes) --- */
443 struct io_bitmap * io_bitmap; /* 0x80 0x8 */
444 long unsigned int iopl_emul; /* 0x88 0x8 */
445 mm_segment_t addr_limit; /* 0x90 0x8 */
446 unsigned int sig_on_uaccess_err:1; /* 0x98: 0 0x4 */
447 unsigned int uaccess_err:1; /* 0x98:0x1 0x4 */
448 /* XXX 30 bits hole, try to pack */
450 /* XXX 36 bytes hole, try to pack */
451 /* --- cacheline 3 boundary (192 bytes) --- */
453 struct fpu fpu; /* 0xc0 0x1040 */
454 /* size: 4352, cachelines: 68, members: 20 */
456 /* sum members: 4312, holes: 1, sum holes: 36 */
457 /* sum bitfield members: 2 bits, bit holes: 1, sum bit holes: 30 bits */
458 };
459 $
460 OK, I know the offset that causes its a 'struct thread_struct' and that
463 the offset is 0x178, so must be in that 'fpu' struct... No problem, ex‐
464 pand 'struct thread_struct' and combine with grep:
465 $ pahole --hex -E thread_struct | egrep '(0x178|struct fpu)' -B4 -A4
467 /* XXX 30 bits hole, try to pack */
468 /* XXX 36 bytes hole, try to pack */
469 /* --- cacheline 3 boundary (192 bytes) --- */
471 struct fpu {
472 unsigned int last_cpu; /* 0xc0 0x4 */
473 /* XXX 4 bytes hole, try to pack */
475 --
477 /* typedef u8 -> __u8 */ unsigned char alimit; /* 0x171 0x1 */
478 /* XXX 6 bytes hole, try to pack */
480 struct math_emu_info * info; /* 0x178 0x8 */
482 /* --- cacheline 6 boundary (384 bytes) --- */
483 /* typedef u32 -> __u32 */ unsigned int entry_eip; /* 0x180 0x4 */
484 } soft; /* 0x100 0x88 */
485 struct xregs_state {
486 $
487 Want to know where 'struct thread_struct' is defined in the kernel
490 sources?
491 $ pahole -I thread_struct | head -2
493 /* Used at: /sys/kernel/btf/vmlinux */
494 /* <0> (null):0 */
495 $
496 Not present in BTF, so use DWARF, takes a little bit longer, and assum‐
499 ing it finds the matching vmlinux file:
500 $ pahole -Fdwarf -I thread_struct | head -2
502 /* Used at: /home/acme/git/linux/arch/x86/kernel/head64.c */
503 /* <3333> /home/acme/git/linux/arch/x86/include/asm/processor.h:485 */
504 $
505 To find the biggest data structures in the Linux kernel:
508 $ pahole -s | sort -k2 -nr | head -5
510 cmp_data 290904 1
511 dec_datas 274520 1
512 cpu_entry_area 217088 0
513 pglist_data 172928 4
514 saved_cmdlines_buffer 131104 1
515 $
516 The second column is the size in bytes and the third is the number of
518 alignment holes in that structure.
519 Show data structures that have a raw spinlock and are related to the
521 RCU mechanism:
522 $ pahole --contains raw_spinlock_t --prefix rcu
524 rcu_node
525 rcu_data
526 rcu_state
527 $
528 To see that in context, combine it with grep:
530 $ pahole rcu_state | grep raw_spinlock_t -B1 -A5
532 /* --- cacheline 52 boundary (3328 bytes) --- */
533 raw_spinlock_t ofl_lock; /* 3328 4 */
534 /* size: 3392, cachelines: 53, members: 35 */
536 /* sum members: 3250, holes: 7, sum holes: 82 */
537 /* padding: 60 */
538 };
539 $
540 It can also pretty print raw data from stdin according to the type
542 specified:
543 $ pahole -C modversion_info drivers/scsi/sg.ko
545 struct modversion_info {
546 long unsigned int crc; /* 0 8 */
547 char name[56]; /* 8 56 */
548 /* size: 64, cachelines: 1, members: 2 */
550 };
551 $
552 $ objcopy -O binary --only-section=__versions drivers/scsi/sg.ko versions
553 $
554 $ ls -la versions
555 -rw-rw-r--. 1 acme acme 7616 Jun 25 11:33 versions
556 $
557 $ pahole --count 3 -C modversion_info drivers/scsi/sg.ko < versions
558 {
559 .crc = 0x8dabd84,
560 .name = "module_layout",
561 },
562 {
563 .crc = 0x45e4617b,
564 .name = "no_llseek",
565 },
566 {
567 .crc = 0xa23fae8c,
568 .name = "param_ops_int",
569 },
570 $
571 $ pahole --skip 1 --count 2 -C modversion_info drivers/scsi/sg.ko < versions
572 {
573 .crc = 0x45e4617b,
574 .name = "no_llseek",
575 },
576 {
577 .crc = 0xa23fae8c,
578 .name = "param_ops_int",
579 },
580 $
581 This is equivalent to:
582 $ pahole --seek_bytes 64 --count 1 -C modversion_info drivers/scsi/sg.ko < versions
584 {
585 .crc = 0x45e4617b,
586 .name = "no_llseek",
587 },
588 $
589
591 pahole can also use the data structure types to pretty print raw data
592 coming from its standard input.
593 -C, --class_name=CLASS_NAME
596 Pretty print according to this class. Arguments may be passed to
597 it to affect how the pretty printing is performed, e.g.:
598 -C 'perf_event_header(sizeof,type,type_enum=perf_event_type,filter=type==PERF_RECORD_EXIT)'
601 This would select the 'struct perf_event_header' as the type to use to
603 pretty print records states that the 'size' field in that struct should
604 be used to figure out the size of the record (variable sized records),
605 that the 'enum perf_event_type' should be used to pretty print the nu‐
606 meric value in perf_event_header->type and furthermore that it should
607 be used to heuristically look for structs with the same name (lower‐
608 case) of the enum entry that is converted from the type field, using it
609 to pretty print instead of the base 'perf_event_header' type. See the
610 PRETTY PRINTING EXAMPLES section below.
611 Furthermore the 'filter=' part can be used, so far with only the '=='
613 operator to filter based on the 'type' field and converting the string
614 'PERF_RECORD_EXIT' to a number according to type_enum.
615 The 'sizeof' arg defaults to the 'size' member name, if the name is
617 different, one can use
618 'sizeof=sz' form, ditto for 'type=other_member_name' field, that de‐
619 faults to 'type'.
620
623 Looking at the ELF header for a vmlinux file, using BTF, first lets
624 discover the ELF header type:
625 $ pahole --sizes | grep -i elf | grep -i _h
627 elf64_hdr 64 0
628 elf32_hdr 52 0
629 $
630 Now we can use this to show the first record from offset zero:
632 $ pahole -C elf64_hdr --count 1 < /lib/modules/5.8.0-rc3+/build/vmlinux
634 {
635 .e_ident = { 127, 69, 76, 70, 2, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0 },
636 .e_type = 2,
637 .e_machine = 62,
638 .e_version = 1,
639 .e_entry = 16777216,
640 .e_phoff = 64,
641 .e_shoff = 775923840,
642 .e_flags = 0,
643 .e_ehsize = 64,
644 .e_phentsize = 56,
645 .e_phnum = 5,
646 .e_shentsize = 64,
647 .e_shnum = 80,
648 .e_shstrndx = 79,
649 },
650 $
651 This is equivalent to:
653 $ pahole --header elf64_hdr < /lib/modules/5.8.0-rc3+/build/vmlinux
655 The --header option also allows reference in other command line options
657 to fields in the header. This is useful when one wants to show multi‐
658 ple records in a file and the range where those fields are located is
659 specified in header fields, such as for perf.data files:
660 $ pahole --hex ~/bin/perf --header perf_file_header < perf.data
662 {
663 .magic = 0x32454c4946524550,
664 .size = 0x68,
665 .attr_size = 0x88,
666 .attrs = {
667 .offset = 0xa8,
668 .size = 0x88,
669 },
670 .data = {
671 .offset = 0x130,
672 .size = 0x588,
673 },
674 .event_types = {
675 .offset = 0,
676 .size = 0,
677 },
678 .adds_features = { 0x16717ffc, 0, 0, 0 },
679 },
680 $
681 So to display the cgroups records in the perf_file_header.data section
683 we can use:
684 $ pahole ~/bin/perf --header=perf_file_header --seek_bytes '$header.data.offset' --size_bytes='$header.data.size' -C 'perf_event_header(sizeof,type,type_enum=perf_event_type,filter=type==PERF_RECORD_CGROUP)' < perf.data
686 {
687 .header = {
688 .type = PERF_RECORD_CGROUP,
689 .misc = 0,
690 .size = 40,
691 },
692 .id = 1,
693 .path = "/",
694 },
695 {
696 .header = {
697 .type = PERF_RECORD_CGROUP,
698 .misc = 0,
699 .size = 48,
700 },
701 .id = 1553,
702 .path = "/system.slice",
703 },
704 {
705 .header = {
706 .type = PERF_RECORD_CGROUP,
707 .misc = 0,
708 .size = 48,
709 },
710 .id = 8,
711 .path = "/machine.slice",
712 },
713 {
714 .header = {
715 .type = PERF_RECORD_CGROUP,
716 .misc = 0,
717 .size = 128,
718 },
719 .id = 7828,
720 .path = "/machine.slice/libpod-42be8e8d4eb9d22405845005f0d04ea398548dccc934a150fbaa3c1f1f9492c2.scope",
721 },
722 {
723 .header = {
724 .type = PERF_RECORD_CGROUP,
725 .misc = 0,
726 .size = 88,
727 },
728 .id = 13,
729 .path = "/machine.slice/machine-qemud1drhel6.sandy.scope",
730 },
731 $
732 For the common case of the header having a member that has the 'offset'
734 and 'size' members, it is possible to use this more compact form:
735 $ pahole ~/bin/perf --header=perf_file_header --range=data -C 'perf_event_header(sizeof,type,type_enum=perf_event_type,filter=type==PERF_RECORD_CGROUP)' < perf.data
737 This uses ~/bin/perf to get the type definitions, the defines 'struct
739 perf_file_header' as the header, then seeks '$header.data.offset' bytes
740 from the start of the file, and considers '$header.data.size' bytes
741 worth of such records. The filter expression may omit a common prefix,
742 in this case it could additonally be equivalently written as both 'fil‐
743 ter=type==CGROUP' or the 'filter=' can also be omitted, getting as com‐
744 pact as 'type==CGROUP':
745 If we look at:
747 $ pahole ~/bin/perf -C perf_event_header
749 struct perf_event_header {
750 __u32 type; /* 0 4 */
751 __u16 misc; /* 4 2 */
752 __u16 size; /* 6 2 */
753 /* size: 8, cachelines: 1, members: 3 */
755 /* last cacheline: 8 bytes */
756 };
757 $
758 And:
760 $ pahole ~/bin/perf -C perf_event_type
762 enum perf_event_type {
763 PERF_RECORD_MMAP = 1,
764 PERF_RECORD_LOST = 2,
765 PERF_RECORD_COMM = 3,
766 PERF_RECORD_EXIT = 4,
767 PERF_RECORD_THROTTLE = 5,
768 PERF_RECORD_UNTHROTTLE = 6,
769 PERF_RECORD_FORK = 7,
770 PERF_RECORD_READ = 8,
771 PERF_RECORD_SAMPLE = 9,
772 PERF_RECORD_MMAP2 = 10,
773 PERF_RECORD_AUX = 11,
774 PERF_RECORD_ITRACE_START = 12,
775 PERF_RECORD_LOST_SAMPLES = 13,
776 PERF_RECORD_SWITCH = 14,
777 PERF_RECORD_SWITCH_CPU_WIDE = 15,
778 PERF_RECORD_NAMESPACES = 16,
779 PERF_RECORD_KSYMBOL = 17,
780 PERF_RECORD_BPF_EVENT = 18,
781 PERF_RECORD_CGROUP = 19,
782 PERF_RECORD_TEXT_POKE = 20,
783 PERF_RECORD_MAX = 21,
784 };
785 $
786 And furthermore:
788 $ pahole ~/bin/perf -C perf_record_cgroup
790 struct perf_record_cgroup {
791 struct perf_event_header header; /* 0 8 */
792 __u64 id; /* 8 8 */
793 char path[4096]; /* 16 4096 */
794 /* size: 4112, cachelines: 65, members: 3 */
796 /* last cacheline: 16 bytes */
797 };
798 $
799 Then we can see how the perf_event_header.type could be converted from
801 a __u32 to a string (PERF_RECORD_CGROUP). If we remove that
802 type_enum=perf_event_type, we will lose the conversion of 'struct
803 perf_event_header' to the more descriptive 'struct perf_record_cgroup',
804 and also the beautification of the header.type field:
805 $ pahole ~/bin/perf --header=perf_file_header --seek_bytes '$header.data.offset' --size_bytes='$header.data.size' -C 'perf_event_header(sizeof,type,filter=type==19)' < perf.data
807 {
808 .type = 19,
809 .misc = 0,
810 .size = 40,
811 },
812 {
813 .type = 19,
814 .misc = 0,
815 .size = 48,
816 },
817 {
818 .type = 19,
819 .misc = 0,
820 .size = 48,
821 },
822 {
823 .type = 19,
824 .misc = 0,
825 .size = 128,
826 },
827 {
828 .type = 19,
829 .misc = 0,
830 .size = 88,
831 },
832 $
833 Some of the records are not found in 'type_enum=perf_event_type' so
835 some of the records don't get converted to a type that fully shows its
836 contents. For perf we know that those are in another enumeration, 'enum
837 perf_user_event_type', so, for these cases, we can create a 'virtual
838 enum', i.e. the sum of two enums and then get all those entries decoded
839 and properly casted, first few records with just 'enum
840 perf_event_type':
841 $ pahole ~/bin/perf --header=perf_file_header --seek_bytes '$header.data.offset' --size_bytes='$header.data.size' -C 'perf_event_header(sizeof,type,type_enum=perf_event_type)' --count 4 < perf.data
843 {
844 .type = 79,
845 .misc = 0,
846 .size = 32,
847 },
848 {
849 .type = 73,
850 .misc = 0,
851 .size = 40,
852 },
853 {
854 .type = 74,
855 .misc = 0,
856 .size = 32,
857 },
858 {
859 .header = {
860 .type = PERF_RECORD_CGROUP,
861 .misc = 0,
862 .size = 40,
863 },
864 .id = 1,
865 .path = "/",
866 },
867 $
868 Now with both enumerations, i.e. with
870 'type_enum=perf_event_type+perf_user_event_type':
871 $ pahole ~/bin/perf --header=perf_file_header --seek_bytes '$header.data.offset' --size_bytes='$header.data.size' -C 'perf_event_header(sizeof,type,type_enum=perf_event_type+perf_user_event_type)' --count 5 < perf.data
873 {
874 .header = {
875 .type = PERF_RECORD_TIME_CONV,
876 .misc = 0,
877 .size = 32,
878 },
879 .time_shift = 31,
880 .time_mult = 1016803377,
881 .time_zero = 435759009518382,
882 },
883 {
884 .header = {
885 .type = PERF_RECORD_THREAD_MAP,
886 .misc = 0,
887 .size = 40,
888 },
889 .nr = 1,
890 .entries = 0x50 0x7e 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00 0x00,
891 },
892 {
893 .header = {
894 .type = PERF_RECORD_CPU_MAP,
895 .misc = 0,
896 .size = 32,
897 },
898 .data = {
899 .type = 1,
900 .data = "",
901 },
902 },
903 {
904 .header = {
905 .type = PERF_RECORD_CGROUP,
906 .misc = 0,
907 .size = 40,
908 },
909 .id = 1,
910 .path = "/",
911 },
912 {
913 .header = {
914 .type = PERF_RECORD_CGROUP,
915 .misc = 0,
916 .size = 48,
917 },
918 .id = 1553,
919 .path = "/system.slice",
920 },
921 $
922 It is possible to pass multiple types, one has only to make sure they
924 appear in the file in sequence, i.e. for the perf.data example, see the
925 perf_file_header dump above, one can print the perf_file_attr structs
926 in the header attrs range, then the perf_event_header in the data range
927 with the following command:
928 pahole ~/bin/perf --header=perf_file_header -C 'perf_file_attr(range=attrs),perf_event_header(range=data,sizeof,type,type_enum=perf_event_type+perf_user_event_type)' < perf.data
930
933 eu-readelf(1), readelf(1), objdump(1).
934 https://www.kernel.org/doc/ols/2007/ols2007v2-pages-35-44.pdf.
936
938 pahole was written and is maintained by Arnaldo Carvalho de Melo
939 <acme@kernel.org>.
940 Thanks to Andrii Nakryiko and Martin KaFai Lau for providing the BTF
942 encoder and improving the codebase while making sure the BTF encoder
943 works as needed to be used in encoding the Linux kernel .BTF section
944 from the DWARF info generated by gcc. For that Andrii wrote a BTF dedu‐
945 plicator in libbpf that is used by pahole.
946 Also thanks to Conectiva, Mandriva and Red Hat for allowing me to work
948 on these tools.
949 Please send bug reports to <dwarves@vger.kernel.org>.
951 No subscription is required.
953dwarves January 16, 2020 pahole(1)