Loading...
1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef __PERF_RECORD_H
3#define __PERF_RECORD_H
4
5#include <limits.h>
6#include <stdio.h>
7#include <linux/kernel.h>
8
9#include "../perf.h"
10#include "build-id.h"
11#include "perf_regs.h"
12
13struct mmap_event {
14 struct perf_event_header header;
15 u32 pid, tid;
16 u64 start;
17 u64 len;
18 u64 pgoff;
19 char filename[PATH_MAX];
20};
21
22struct mmap2_event {
23 struct perf_event_header header;
24 u32 pid, tid;
25 u64 start;
26 u64 len;
27 u64 pgoff;
28 u32 maj;
29 u32 min;
30 u64 ino;
31 u64 ino_generation;
32 u32 prot;
33 u32 flags;
34 char filename[PATH_MAX];
35};
36
37struct comm_event {
38 struct perf_event_header header;
39 u32 pid, tid;
40 char comm[16];
41};
42
43struct namespaces_event {
44 struct perf_event_header header;
45 u32 pid, tid;
46 u64 nr_namespaces;
47 struct perf_ns_link_info link_info[];
48};
49
50struct fork_event {
51 struct perf_event_header header;
52 u32 pid, ppid;
53 u32 tid, ptid;
54 u64 time;
55};
56
57struct lost_event {
58 struct perf_event_header header;
59 u64 id;
60 u64 lost;
61};
62
63struct lost_samples_event {
64 struct perf_event_header header;
65 u64 lost;
66};
67
68/*
69 * PERF_FORMAT_ENABLED | PERF_FORMAT_RUNNING | PERF_FORMAT_ID
70 */
71struct read_event {
72 struct perf_event_header header;
73 u32 pid, tid;
74 u64 value;
75 u64 time_enabled;
76 u64 time_running;
77 u64 id;
78};
79
80struct throttle_event {
81 struct perf_event_header header;
82 u64 time;
83 u64 id;
84 u64 stream_id;
85};
86
87#define PERF_SAMPLE_MASK \
88 (PERF_SAMPLE_IP | PERF_SAMPLE_TID | \
89 PERF_SAMPLE_TIME | PERF_SAMPLE_ADDR | \
90 PERF_SAMPLE_ID | PERF_SAMPLE_STREAM_ID | \
91 PERF_SAMPLE_CPU | PERF_SAMPLE_PERIOD | \
92 PERF_SAMPLE_IDENTIFIER)
93
94/* perf sample has 16 bits size limit */
95#define PERF_SAMPLE_MAX_SIZE (1 << 16)
96
97struct sample_event {
98 struct perf_event_header header;
99 u64 array[];
100};
101
102struct regs_dump {
103 u64 abi;
104 u64 mask;
105 u64 *regs;
106
107 /* Cached values/mask filled by first register access. */
108 u64 cache_regs[PERF_REGS_MAX];
109 u64 cache_mask;
110};
111
112struct stack_dump {
113 u16 offset;
114 u64 size;
115 char *data;
116};
117
118struct sample_read_value {
119 u64 value;
120 u64 id;
121};
122
123struct sample_read {
124 u64 time_enabled;
125 u64 time_running;
126 union {
127 struct {
128 u64 nr;
129 struct sample_read_value *values;
130 } group;
131 struct sample_read_value one;
132 };
133};
134
135struct ip_callchain {
136 u64 nr;
137 u64 ips[0];
138};
139
140struct branch_flags {
141 u64 mispred:1;
142 u64 predicted:1;
143 u64 in_tx:1;
144 u64 abort:1;
145 u64 cycles:16;
146 u64 type:4;
147 u64 reserved:40;
148};
149
150struct branch_entry {
151 u64 from;
152 u64 to;
153 struct branch_flags flags;
154};
155
156struct branch_stack {
157 u64 nr;
158 struct branch_entry entries[0];
159};
160
161enum {
162 PERF_IP_FLAG_BRANCH = 1ULL << 0,
163 PERF_IP_FLAG_CALL = 1ULL << 1,
164 PERF_IP_FLAG_RETURN = 1ULL << 2,
165 PERF_IP_FLAG_CONDITIONAL = 1ULL << 3,
166 PERF_IP_FLAG_SYSCALLRET = 1ULL << 4,
167 PERF_IP_FLAG_ASYNC = 1ULL << 5,
168 PERF_IP_FLAG_INTERRUPT = 1ULL << 6,
169 PERF_IP_FLAG_TX_ABORT = 1ULL << 7,
170 PERF_IP_FLAG_TRACE_BEGIN = 1ULL << 8,
171 PERF_IP_FLAG_TRACE_END = 1ULL << 9,
172 PERF_IP_FLAG_IN_TX = 1ULL << 10,
173};
174
175#define PERF_IP_FLAG_CHARS "bcrosyiABEx"
176
177#define PERF_BRANCH_MASK (\
178 PERF_IP_FLAG_BRANCH |\
179 PERF_IP_FLAG_CALL |\
180 PERF_IP_FLAG_RETURN |\
181 PERF_IP_FLAG_CONDITIONAL |\
182 PERF_IP_FLAG_SYSCALLRET |\
183 PERF_IP_FLAG_ASYNC |\
184 PERF_IP_FLAG_INTERRUPT |\
185 PERF_IP_FLAG_TX_ABORT |\
186 PERF_IP_FLAG_TRACE_BEGIN |\
187 PERF_IP_FLAG_TRACE_END)
188
189#define MAX_INSN 16
190
191struct perf_sample {
192 u64 ip;
193 u32 pid, tid;
194 u64 time;
195 u64 addr;
196 u64 id;
197 u64 stream_id;
198 u64 period;
199 u64 weight;
200 u64 transaction;
201 u32 cpu;
202 u32 raw_size;
203 u64 data_src;
204 u64 phys_addr;
205 u32 flags;
206 u16 insn_len;
207 u8 cpumode;
208 u16 misc;
209 char insn[MAX_INSN];
210 void *raw_data;
211 struct ip_callchain *callchain;
212 struct branch_stack *branch_stack;
213 struct regs_dump user_regs;
214 struct regs_dump intr_regs;
215 struct stack_dump user_stack;
216 struct sample_read read;
217};
218
219#define PERF_MEM_DATA_SRC_NONE \
220 (PERF_MEM_S(OP, NA) |\
221 PERF_MEM_S(LVL, NA) |\
222 PERF_MEM_S(SNOOP, NA) |\
223 PERF_MEM_S(LOCK, NA) |\
224 PERF_MEM_S(TLB, NA))
225
226struct build_id_event {
227 struct perf_event_header header;
228 pid_t pid;
229 u8 build_id[PERF_ALIGN(BUILD_ID_SIZE, sizeof(u64))];
230 char filename[];
231};
232
233enum perf_user_event_type { /* above any possible kernel type */
234 PERF_RECORD_USER_TYPE_START = 64,
235 PERF_RECORD_HEADER_ATTR = 64,
236 PERF_RECORD_HEADER_EVENT_TYPE = 65, /* deprecated */
237 PERF_RECORD_HEADER_TRACING_DATA = 66,
238 PERF_RECORD_HEADER_BUILD_ID = 67,
239 PERF_RECORD_FINISHED_ROUND = 68,
240 PERF_RECORD_ID_INDEX = 69,
241 PERF_RECORD_AUXTRACE_INFO = 70,
242 PERF_RECORD_AUXTRACE = 71,
243 PERF_RECORD_AUXTRACE_ERROR = 72,
244 PERF_RECORD_THREAD_MAP = 73,
245 PERF_RECORD_CPU_MAP = 74,
246 PERF_RECORD_STAT_CONFIG = 75,
247 PERF_RECORD_STAT = 76,
248 PERF_RECORD_STAT_ROUND = 77,
249 PERF_RECORD_EVENT_UPDATE = 78,
250 PERF_RECORD_TIME_CONV = 79,
251 PERF_RECORD_HEADER_FEATURE = 80,
252 PERF_RECORD_HEADER_MAX
253};
254
255enum auxtrace_error_type {
256 PERF_AUXTRACE_ERROR_ITRACE = 1,
257 PERF_AUXTRACE_ERROR_MAX
258};
259
260/* Attribute type for custom synthesized events */
261#define PERF_TYPE_SYNTH (INT_MAX + 1U)
262
263/* Attribute config for custom synthesized events */
264enum perf_synth_id {
265 PERF_SYNTH_INTEL_PTWRITE,
266 PERF_SYNTH_INTEL_MWAIT,
267 PERF_SYNTH_INTEL_PWRE,
268 PERF_SYNTH_INTEL_EXSTOP,
269 PERF_SYNTH_INTEL_PWRX,
270 PERF_SYNTH_INTEL_CBR,
271};
272
273/*
274 * Raw data formats for synthesized events. Note that 4 bytes of padding are
275 * present to match the 'size' member of PERF_SAMPLE_RAW data which is always
276 * 8-byte aligned. That means we must dereference raw_data with an offset of 4.
277 * Refer perf_sample__synth_ptr() and perf_synth__raw_data(). It also means the
278 * structure sizes are 4 bytes bigger than the raw_size, refer
279 * perf_synth__raw_size().
280 */
281
282struct perf_synth_intel_ptwrite {
283 u32 padding;
284 union {
285 struct {
286 u32 ip : 1,
287 reserved : 31;
288 };
289 u32 flags;
290 };
291 u64 payload;
292};
293
294struct perf_synth_intel_mwait {
295 u32 padding;
296 u32 reserved;
297 union {
298 struct {
299 u64 hints : 8,
300 reserved1 : 24,
301 extensions : 2,
302 reserved2 : 30;
303 };
304 u64 payload;
305 };
306};
307
308struct perf_synth_intel_pwre {
309 u32 padding;
310 u32 reserved;
311 union {
312 struct {
313 u64 reserved1 : 7,
314 hw : 1,
315 subcstate : 4,
316 cstate : 4,
317 reserved2 : 48;
318 };
319 u64 payload;
320 };
321};
322
323struct perf_synth_intel_exstop {
324 u32 padding;
325 union {
326 struct {
327 u32 ip : 1,
328 reserved : 31;
329 };
330 u32 flags;
331 };
332};
333
334struct perf_synth_intel_pwrx {
335 u32 padding;
336 u32 reserved;
337 union {
338 struct {
339 u64 deepest_cstate : 4,
340 last_cstate : 4,
341 wake_reason : 4,
342 reserved1 : 52;
343 };
344 u64 payload;
345 };
346};
347
348struct perf_synth_intel_cbr {
349 u32 padding;
350 union {
351 struct {
352 u32 cbr : 8,
353 reserved1 : 8,
354 max_nonturbo : 8,
355 reserved2 : 8;
356 };
357 u32 flags;
358 };
359 u32 freq;
360 u32 reserved3;
361};
362
363/*
364 * raw_data is always 4 bytes from an 8-byte boundary, so subtract 4 to get
365 * 8-byte alignment.
366 */
367static inline void *perf_sample__synth_ptr(struct perf_sample *sample)
368{
369 return sample->raw_data - 4;
370}
371
372static inline void *perf_synth__raw_data(void *p)
373{
374 return p + 4;
375}
376
377#define perf_synth__raw_size(d) (sizeof(d) - 4)
378
379#define perf_sample__bad_synth_size(s, d) ((s)->raw_size < sizeof(d) - 4)
380
381/*
382 * The kernel collects the number of events it couldn't send in a stretch and
383 * when possible sends this number in a PERF_RECORD_LOST event. The number of
384 * such "chunks" of lost events is stored in .nr_events[PERF_EVENT_LOST] while
385 * total_lost tells exactly how many events the kernel in fact lost, i.e. it is
386 * the sum of all struct lost_event.lost fields reported.
387 *
388 * The kernel discards mixed up samples and sends the number in a
389 * PERF_RECORD_LOST_SAMPLES event. The number of lost-samples events is stored
390 * in .nr_events[PERF_RECORD_LOST_SAMPLES] while total_lost_samples tells
391 * exactly how many samples the kernel in fact dropped, i.e. it is the sum of
392 * all struct lost_samples_event.lost fields reported.
393 *
394 * The total_period is needed because by default auto-freq is used, so
395 * multipling nr_events[PERF_EVENT_SAMPLE] by a frequency isn't possible to get
396 * the total number of low level events, it is necessary to to sum all struct
397 * sample_event.period and stash the result in total_period.
398 */
399struct events_stats {
400 u64 total_period;
401 u64 total_non_filtered_period;
402 u64 total_lost;
403 u64 total_lost_samples;
404 u64 total_aux_lost;
405 u64 total_aux_partial;
406 u64 total_invalid_chains;
407 u32 nr_events[PERF_RECORD_HEADER_MAX];
408 u32 nr_non_filtered_samples;
409 u32 nr_lost_warned;
410 u32 nr_unknown_events;
411 u32 nr_invalid_chains;
412 u32 nr_unknown_id;
413 u32 nr_unprocessable_samples;
414 u32 nr_auxtrace_errors[PERF_AUXTRACE_ERROR_MAX];
415 u32 nr_proc_map_timeout;
416};
417
418enum {
419 PERF_CPU_MAP__CPUS = 0,
420 PERF_CPU_MAP__MASK = 1,
421};
422
423struct cpu_map_entries {
424 u16 nr;
425 u16 cpu[];
426};
427
428struct cpu_map_mask {
429 u16 nr;
430 u16 long_size;
431 unsigned long mask[];
432};
433
434struct cpu_map_data {
435 u16 type;
436 char data[];
437};
438
439struct cpu_map_event {
440 struct perf_event_header header;
441 struct cpu_map_data data;
442};
443
444struct attr_event {
445 struct perf_event_header header;
446 struct perf_event_attr attr;
447 u64 id[];
448};
449
450enum {
451 PERF_EVENT_UPDATE__UNIT = 0,
452 PERF_EVENT_UPDATE__SCALE = 1,
453 PERF_EVENT_UPDATE__NAME = 2,
454 PERF_EVENT_UPDATE__CPUS = 3,
455};
456
457struct event_update_event_cpus {
458 struct cpu_map_data cpus;
459};
460
461struct event_update_event_scale {
462 double scale;
463};
464
465struct event_update_event {
466 struct perf_event_header header;
467 u64 type;
468 u64 id;
469
470 char data[];
471};
472
473#define MAX_EVENT_NAME 64
474
475struct perf_trace_event_type {
476 u64 event_id;
477 char name[MAX_EVENT_NAME];
478};
479
480struct event_type_event {
481 struct perf_event_header header;
482 struct perf_trace_event_type event_type;
483};
484
485struct tracing_data_event {
486 struct perf_event_header header;
487 u32 size;
488};
489
490struct id_index_entry {
491 u64 id;
492 u64 idx;
493 u64 cpu;
494 u64 tid;
495};
496
497struct id_index_event {
498 struct perf_event_header header;
499 u64 nr;
500 struct id_index_entry entries[0];
501};
502
503struct auxtrace_info_event {
504 struct perf_event_header header;
505 u32 type;
506 u32 reserved__; /* For alignment */
507 u64 priv[];
508};
509
510struct auxtrace_event {
511 struct perf_event_header header;
512 u64 size;
513 u64 offset;
514 u64 reference;
515 u32 idx;
516 u32 tid;
517 u32 cpu;
518 u32 reserved__; /* For alignment */
519};
520
521#define MAX_AUXTRACE_ERROR_MSG 64
522
523struct auxtrace_error_event {
524 struct perf_event_header header;
525 u32 type;
526 u32 code;
527 u32 cpu;
528 u32 pid;
529 u32 tid;
530 u32 reserved__; /* For alignment */
531 u64 ip;
532 char msg[MAX_AUXTRACE_ERROR_MSG];
533};
534
535struct aux_event {
536 struct perf_event_header header;
537 u64 aux_offset;
538 u64 aux_size;
539 u64 flags;
540};
541
542struct itrace_start_event {
543 struct perf_event_header header;
544 u32 pid, tid;
545};
546
547struct context_switch_event {
548 struct perf_event_header header;
549 u32 next_prev_pid;
550 u32 next_prev_tid;
551};
552
553struct thread_map_event_entry {
554 u64 pid;
555 char comm[16];
556};
557
558struct thread_map_event {
559 struct perf_event_header header;
560 u64 nr;
561 struct thread_map_event_entry entries[];
562};
563
564enum {
565 PERF_STAT_CONFIG_TERM__AGGR_MODE = 0,
566 PERF_STAT_CONFIG_TERM__INTERVAL = 1,
567 PERF_STAT_CONFIG_TERM__SCALE = 2,
568 PERF_STAT_CONFIG_TERM__MAX = 3,
569};
570
571struct stat_config_event_entry {
572 u64 tag;
573 u64 val;
574};
575
576struct stat_config_event {
577 struct perf_event_header header;
578 u64 nr;
579 struct stat_config_event_entry data[];
580};
581
582struct stat_event {
583 struct perf_event_header header;
584
585 u64 id;
586 u32 cpu;
587 u32 thread;
588
589 union {
590 struct {
591 u64 val;
592 u64 ena;
593 u64 run;
594 };
595 u64 values[3];
596 };
597};
598
599enum {
600 PERF_STAT_ROUND_TYPE__INTERVAL = 0,
601 PERF_STAT_ROUND_TYPE__FINAL = 1,
602};
603
604struct stat_round_event {
605 struct perf_event_header header;
606 u64 type;
607 u64 time;
608};
609
610struct time_conv_event {
611 struct perf_event_header header;
612 u64 time_shift;
613 u64 time_mult;
614 u64 time_zero;
615};
616
617struct feature_event {
618 struct perf_event_header header;
619 u64 feat_id;
620 char data[];
621};
622
623union perf_event {
624 struct perf_event_header header;
625 struct mmap_event mmap;
626 struct mmap2_event mmap2;
627 struct comm_event comm;
628 struct namespaces_event namespaces;
629 struct fork_event fork;
630 struct lost_event lost;
631 struct lost_samples_event lost_samples;
632 struct read_event read;
633 struct throttle_event throttle;
634 struct sample_event sample;
635 struct attr_event attr;
636 struct event_update_event event_update;
637 struct event_type_event event_type;
638 struct tracing_data_event tracing_data;
639 struct build_id_event build_id;
640 struct id_index_event id_index;
641 struct auxtrace_info_event auxtrace_info;
642 struct auxtrace_event auxtrace;
643 struct auxtrace_error_event auxtrace_error;
644 struct aux_event aux;
645 struct itrace_start_event itrace_start;
646 struct context_switch_event context_switch;
647 struct thread_map_event thread_map;
648 struct cpu_map_event cpu_map;
649 struct stat_config_event stat_config;
650 struct stat_event stat;
651 struct stat_round_event stat_round;
652 struct time_conv_event time_conv;
653 struct feature_event feat;
654};
655
656void perf_event__print_totals(void);
657
658struct perf_tool;
659struct thread_map;
660struct cpu_map;
661struct perf_stat_config;
662struct perf_counts_values;
663
664typedef int (*perf_event__handler_t)(struct perf_tool *tool,
665 union perf_event *event,
666 struct perf_sample *sample,
667 struct machine *machine);
668
669int perf_event__synthesize_thread_map(struct perf_tool *tool,
670 struct thread_map *threads,
671 perf_event__handler_t process,
672 struct machine *machine, bool mmap_data,
673 unsigned int proc_map_timeout);
674int perf_event__synthesize_thread_map2(struct perf_tool *tool,
675 struct thread_map *threads,
676 perf_event__handler_t process,
677 struct machine *machine);
678int perf_event__synthesize_cpu_map(struct perf_tool *tool,
679 struct cpu_map *cpus,
680 perf_event__handler_t process,
681 struct machine *machine);
682int perf_event__synthesize_threads(struct perf_tool *tool,
683 perf_event__handler_t process,
684 struct machine *machine, bool mmap_data,
685 unsigned int proc_map_timeout,
686 unsigned int nr_threads_synthesize);
687int perf_event__synthesize_kernel_mmap(struct perf_tool *tool,
688 perf_event__handler_t process,
689 struct machine *machine);
690int perf_event__synthesize_stat_config(struct perf_tool *tool,
691 struct perf_stat_config *config,
692 perf_event__handler_t process,
693 struct machine *machine);
694void perf_event__read_stat_config(struct perf_stat_config *config,
695 struct stat_config_event *event);
696int perf_event__synthesize_stat(struct perf_tool *tool,
697 u32 cpu, u32 thread, u64 id,
698 struct perf_counts_values *count,
699 perf_event__handler_t process,
700 struct machine *machine);
701int perf_event__synthesize_stat_round(struct perf_tool *tool,
702 u64 time, u64 type,
703 perf_event__handler_t process,
704 struct machine *machine);
705int perf_event__synthesize_modules(struct perf_tool *tool,
706 perf_event__handler_t process,
707 struct machine *machine);
708
709int perf_event__process_comm(struct perf_tool *tool,
710 union perf_event *event,
711 struct perf_sample *sample,
712 struct machine *machine);
713int perf_event__process_lost(struct perf_tool *tool,
714 union perf_event *event,
715 struct perf_sample *sample,
716 struct machine *machine);
717int perf_event__process_lost_samples(struct perf_tool *tool,
718 union perf_event *event,
719 struct perf_sample *sample,
720 struct machine *machine);
721int perf_event__process_aux(struct perf_tool *tool,
722 union perf_event *event,
723 struct perf_sample *sample,
724 struct machine *machine);
725int perf_event__process_itrace_start(struct perf_tool *tool,
726 union perf_event *event,
727 struct perf_sample *sample,
728 struct machine *machine);
729int perf_event__process_switch(struct perf_tool *tool,
730 union perf_event *event,
731 struct perf_sample *sample,
732 struct machine *machine);
733int perf_event__process_namespaces(struct perf_tool *tool,
734 union perf_event *event,
735 struct perf_sample *sample,
736 struct machine *machine);
737int perf_event__process_mmap(struct perf_tool *tool,
738 union perf_event *event,
739 struct perf_sample *sample,
740 struct machine *machine);
741int perf_event__process_mmap2(struct perf_tool *tool,
742 union perf_event *event,
743 struct perf_sample *sample,
744 struct machine *machine);
745int perf_event__process_fork(struct perf_tool *tool,
746 union perf_event *event,
747 struct perf_sample *sample,
748 struct machine *machine);
749int perf_event__process_exit(struct perf_tool *tool,
750 union perf_event *event,
751 struct perf_sample *sample,
752 struct machine *machine);
753int perf_event__process(struct perf_tool *tool,
754 union perf_event *event,
755 struct perf_sample *sample,
756 struct machine *machine);
757
758struct addr_location;
759
760int machine__resolve(struct machine *machine, struct addr_location *al,
761 struct perf_sample *sample);
762
763void addr_location__put(struct addr_location *al);
764
765struct thread;
766
767bool is_bts_event(struct perf_event_attr *attr);
768bool sample_addr_correlates_sym(struct perf_event_attr *attr);
769void thread__resolve(struct thread *thread, struct addr_location *al,
770 struct perf_sample *sample);
771
772const char *perf_event__name(unsigned int id);
773
774size_t perf_event__sample_event_size(const struct perf_sample *sample, u64 type,
775 u64 read_format);
776int perf_event__synthesize_sample(union perf_event *event, u64 type,
777 u64 read_format,
778 const struct perf_sample *sample);
779
780pid_t perf_event__synthesize_comm(struct perf_tool *tool,
781 union perf_event *event, pid_t pid,
782 perf_event__handler_t process,
783 struct machine *machine);
784
785int perf_event__synthesize_namespaces(struct perf_tool *tool,
786 union perf_event *event,
787 pid_t pid, pid_t tgid,
788 perf_event__handler_t process,
789 struct machine *machine);
790
791int perf_event__synthesize_mmap_events(struct perf_tool *tool,
792 union perf_event *event,
793 pid_t pid, pid_t tgid,
794 perf_event__handler_t process,
795 struct machine *machine,
796 bool mmap_data,
797 unsigned int proc_map_timeout);
798
799size_t perf_event__fprintf_comm(union perf_event *event, FILE *fp);
800size_t perf_event__fprintf_mmap(union perf_event *event, FILE *fp);
801size_t perf_event__fprintf_mmap2(union perf_event *event, FILE *fp);
802size_t perf_event__fprintf_task(union perf_event *event, FILE *fp);
803size_t perf_event__fprintf_aux(union perf_event *event, FILE *fp);
804size_t perf_event__fprintf_itrace_start(union perf_event *event, FILE *fp);
805size_t perf_event__fprintf_switch(union perf_event *event, FILE *fp);
806size_t perf_event__fprintf_thread_map(union perf_event *event, FILE *fp);
807size_t perf_event__fprintf_cpu_map(union perf_event *event, FILE *fp);
808size_t perf_event__fprintf_namespaces(union perf_event *event, FILE *fp);
809size_t perf_event__fprintf(union perf_event *event, FILE *fp);
810
811int kallsyms__get_function_start(const char *kallsyms_filename,
812 const char *symbol_name, u64 *addr);
813
814void *cpu_map_data__alloc(struct cpu_map *map, size_t *size, u16 *type, int *max);
815void cpu_map_data__synthesize(struct cpu_map_data *data, struct cpu_map *map,
816 u16 type, int max);
817
818void event_attr_init(struct perf_event_attr *attr);
819
820int perf_event_paranoid(void);
821
822extern int sysctl_perf_event_max_stack;
823extern int sysctl_perf_event_max_contexts_per_stack;
824
825#endif /* __PERF_RECORD_H */
1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef __PERF_RECORD_H
3#define __PERF_RECORD_H
4/*
5 * The linux/stddef.h isn't need here, but is needed for __always_inline used
6 * in files included from uapi/linux/perf_event.h such as
7 * /usr/include/linux/swab.h and /usr/include/linux/byteorder/little_endian.h,
8 * detected in at least musl libc, used in Alpine Linux. -acme
9 */
10#include <stdio.h>
11#include <linux/stddef.h>
12#include <perf/event.h>
13#include <linux/types.h>
14
15#include "perf_regs.h"
16
17struct dso;
18struct machine;
19struct perf_event_attr;
20
21#ifdef __LP64__
22/*
23 * /usr/include/inttypes.h uses just 'lu' for PRIu64, but we end up defining
24 * __u64 as long long unsigned int, and then -Werror=format= kicks in and
25 * complains of the mismatched types, so use these two special extra PRI
26 * macros to overcome that.
27 */
28#define PRI_lu64 "l" PRIu64
29#define PRI_lx64 "l" PRIx64
30#define PRI_ld64 "l" PRId64
31#else
32#define PRI_lu64 PRIu64
33#define PRI_lx64 PRIx64
34#define PRI_ld64 PRId64
35#endif
36
37#define PERF_SAMPLE_MASK \
38 (PERF_SAMPLE_IP | PERF_SAMPLE_TID | \
39 PERF_SAMPLE_TIME | PERF_SAMPLE_ADDR | \
40 PERF_SAMPLE_ID | PERF_SAMPLE_STREAM_ID | \
41 PERF_SAMPLE_CPU | PERF_SAMPLE_PERIOD | \
42 PERF_SAMPLE_IDENTIFIER)
43
44/* perf sample has 16 bits size limit */
45#define PERF_SAMPLE_MAX_SIZE (1 << 16)
46
47struct regs_dump {
48 u64 abi;
49 u64 mask;
50 u64 *regs;
51
52 /* Cached values/mask filled by first register access. */
53 u64 cache_regs[PERF_REGS_MAX];
54 u64 cache_mask;
55};
56
57struct stack_dump {
58 u16 offset;
59 u64 size;
60 char *data;
61};
62
63struct sample_read_value {
64 u64 value;
65 u64 id;
66};
67
68struct sample_read {
69 u64 time_enabled;
70 u64 time_running;
71 union {
72 struct {
73 u64 nr;
74 struct sample_read_value *values;
75 } group;
76 struct sample_read_value one;
77 };
78};
79
80struct ip_callchain {
81 u64 nr;
82 u64 ips[];
83};
84
85struct branch_stack;
86
87enum {
88 PERF_IP_FLAG_BRANCH = 1ULL << 0,
89 PERF_IP_FLAG_CALL = 1ULL << 1,
90 PERF_IP_FLAG_RETURN = 1ULL << 2,
91 PERF_IP_FLAG_CONDITIONAL = 1ULL << 3,
92 PERF_IP_FLAG_SYSCALLRET = 1ULL << 4,
93 PERF_IP_FLAG_ASYNC = 1ULL << 5,
94 PERF_IP_FLAG_INTERRUPT = 1ULL << 6,
95 PERF_IP_FLAG_TX_ABORT = 1ULL << 7,
96 PERF_IP_FLAG_TRACE_BEGIN = 1ULL << 8,
97 PERF_IP_FLAG_TRACE_END = 1ULL << 9,
98 PERF_IP_FLAG_IN_TX = 1ULL << 10,
99};
100
101#define PERF_IP_FLAG_CHARS "bcrosyiABEx"
102
103#define PERF_BRANCH_MASK (\
104 PERF_IP_FLAG_BRANCH |\
105 PERF_IP_FLAG_CALL |\
106 PERF_IP_FLAG_RETURN |\
107 PERF_IP_FLAG_CONDITIONAL |\
108 PERF_IP_FLAG_SYSCALLRET |\
109 PERF_IP_FLAG_ASYNC |\
110 PERF_IP_FLAG_INTERRUPT |\
111 PERF_IP_FLAG_TX_ABORT |\
112 PERF_IP_FLAG_TRACE_BEGIN |\
113 PERF_IP_FLAG_TRACE_END)
114
115#define MAX_INSN 16
116
117struct aux_sample {
118 u64 size;
119 void *data;
120};
121
122struct perf_sample {
123 u64 ip;
124 u32 pid, tid;
125 u64 time;
126 u64 addr;
127 u64 id;
128 u64 stream_id;
129 u64 period;
130 u64 weight;
131 u64 transaction;
132 u64 insn_cnt;
133 u64 cyc_cnt;
134 u32 cpu;
135 u32 raw_size;
136 u64 data_src;
137 u64 phys_addr;
138 u64 cgroup;
139 u32 flags;
140 u16 insn_len;
141 u8 cpumode;
142 u16 misc;
143 bool no_hw_idx; /* No hw_idx collected in branch_stack */
144 char insn[MAX_INSN];
145 void *raw_data;
146 struct ip_callchain *callchain;
147 struct branch_stack *branch_stack;
148 struct regs_dump user_regs;
149 struct regs_dump intr_regs;
150 struct stack_dump user_stack;
151 struct sample_read read;
152 struct aux_sample aux_sample;
153};
154
155#define PERF_MEM_DATA_SRC_NONE \
156 (PERF_MEM_S(OP, NA) |\
157 PERF_MEM_S(LVL, NA) |\
158 PERF_MEM_S(SNOOP, NA) |\
159 PERF_MEM_S(LOCK, NA) |\
160 PERF_MEM_S(TLB, NA))
161
162/* Attribute type for custom synthesized events */
163#define PERF_TYPE_SYNTH (INT_MAX + 1U)
164
165/* Attribute config for custom synthesized events */
166enum perf_synth_id {
167 PERF_SYNTH_INTEL_PTWRITE,
168 PERF_SYNTH_INTEL_MWAIT,
169 PERF_SYNTH_INTEL_PWRE,
170 PERF_SYNTH_INTEL_EXSTOP,
171 PERF_SYNTH_INTEL_PWRX,
172 PERF_SYNTH_INTEL_CBR,
173};
174
175/*
176 * Raw data formats for synthesized events. Note that 4 bytes of padding are
177 * present to match the 'size' member of PERF_SAMPLE_RAW data which is always
178 * 8-byte aligned. That means we must dereference raw_data with an offset of 4.
179 * Refer perf_sample__synth_ptr() and perf_synth__raw_data(). It also means the
180 * structure sizes are 4 bytes bigger than the raw_size, refer
181 * perf_synth__raw_size().
182 */
183
184struct perf_synth_intel_ptwrite {
185 u32 padding;
186 union {
187 struct {
188 u32 ip : 1,
189 reserved : 31;
190 };
191 u32 flags;
192 };
193 u64 payload;
194};
195
196struct perf_synth_intel_mwait {
197 u32 padding;
198 u32 reserved;
199 union {
200 struct {
201 u64 hints : 8,
202 reserved1 : 24,
203 extensions : 2,
204 reserved2 : 30;
205 };
206 u64 payload;
207 };
208};
209
210struct perf_synth_intel_pwre {
211 u32 padding;
212 u32 reserved;
213 union {
214 struct {
215 u64 reserved1 : 7,
216 hw : 1,
217 subcstate : 4,
218 cstate : 4,
219 reserved2 : 48;
220 };
221 u64 payload;
222 };
223};
224
225struct perf_synth_intel_exstop {
226 u32 padding;
227 union {
228 struct {
229 u32 ip : 1,
230 reserved : 31;
231 };
232 u32 flags;
233 };
234};
235
236struct perf_synth_intel_pwrx {
237 u32 padding;
238 u32 reserved;
239 union {
240 struct {
241 u64 deepest_cstate : 4,
242 last_cstate : 4,
243 wake_reason : 4,
244 reserved1 : 52;
245 };
246 u64 payload;
247 };
248};
249
250struct perf_synth_intel_cbr {
251 u32 padding;
252 union {
253 struct {
254 u32 cbr : 8,
255 reserved1 : 8,
256 max_nonturbo : 8,
257 reserved2 : 8;
258 };
259 u32 flags;
260 };
261 u32 freq;
262 u32 reserved3;
263};
264
265/*
266 * raw_data is always 4 bytes from an 8-byte boundary, so subtract 4 to get
267 * 8-byte alignment.
268 */
269static inline void *perf_sample__synth_ptr(struct perf_sample *sample)
270{
271 return sample->raw_data - 4;
272}
273
274static inline void *perf_synth__raw_data(void *p)
275{
276 return p + 4;
277}
278
279#define perf_synth__raw_size(d) (sizeof(d) - 4)
280
281#define perf_sample__bad_synth_size(s, d) ((s)->raw_size < sizeof(d) - 4)
282
283enum {
284 PERF_STAT_ROUND_TYPE__INTERVAL = 0,
285 PERF_STAT_ROUND_TYPE__FINAL = 1,
286};
287
288void perf_event__print_totals(void);
289
290struct perf_cpu_map;
291struct perf_record_stat_config;
292struct perf_stat_config;
293struct perf_tool;
294
295void perf_event__read_stat_config(struct perf_stat_config *config,
296 struct perf_record_stat_config *event);
297
298int perf_event__process_comm(struct perf_tool *tool,
299 union perf_event *event,
300 struct perf_sample *sample,
301 struct machine *machine);
302int perf_event__process_lost(struct perf_tool *tool,
303 union perf_event *event,
304 struct perf_sample *sample,
305 struct machine *machine);
306int perf_event__process_lost_samples(struct perf_tool *tool,
307 union perf_event *event,
308 struct perf_sample *sample,
309 struct machine *machine);
310int perf_event__process_aux(struct perf_tool *tool,
311 union perf_event *event,
312 struct perf_sample *sample,
313 struct machine *machine);
314int perf_event__process_itrace_start(struct perf_tool *tool,
315 union perf_event *event,
316 struct perf_sample *sample,
317 struct machine *machine);
318int perf_event__process_switch(struct perf_tool *tool,
319 union perf_event *event,
320 struct perf_sample *sample,
321 struct machine *machine);
322int perf_event__process_namespaces(struct perf_tool *tool,
323 union perf_event *event,
324 struct perf_sample *sample,
325 struct machine *machine);
326int perf_event__process_cgroup(struct perf_tool *tool,
327 union perf_event *event,
328 struct perf_sample *sample,
329 struct machine *machine);
330int perf_event__process_mmap(struct perf_tool *tool,
331 union perf_event *event,
332 struct perf_sample *sample,
333 struct machine *machine);
334int perf_event__process_mmap2(struct perf_tool *tool,
335 union perf_event *event,
336 struct perf_sample *sample,
337 struct machine *machine);
338int perf_event__process_fork(struct perf_tool *tool,
339 union perf_event *event,
340 struct perf_sample *sample,
341 struct machine *machine);
342int perf_event__process_exit(struct perf_tool *tool,
343 union perf_event *event,
344 struct perf_sample *sample,
345 struct machine *machine);
346int perf_event__process_ksymbol(struct perf_tool *tool,
347 union perf_event *event,
348 struct perf_sample *sample,
349 struct machine *machine);
350int perf_event__process_bpf(struct perf_tool *tool,
351 union perf_event *event,
352 struct perf_sample *sample,
353 struct machine *machine);
354int perf_event__process_text_poke(struct perf_tool *tool,
355 union perf_event *event,
356 struct perf_sample *sample,
357 struct machine *machine);
358int perf_event__process(struct perf_tool *tool,
359 union perf_event *event,
360 struct perf_sample *sample,
361 struct machine *machine);
362
363struct addr_location;
364
365int machine__resolve(struct machine *machine, struct addr_location *al,
366 struct perf_sample *sample);
367
368void addr_location__put(struct addr_location *al);
369
370struct thread;
371
372bool is_bts_event(struct perf_event_attr *attr);
373bool sample_addr_correlates_sym(struct perf_event_attr *attr);
374void thread__resolve(struct thread *thread, struct addr_location *al,
375 struct perf_sample *sample);
376
377const char *perf_event__name(unsigned int id);
378
379size_t perf_event__fprintf_comm(union perf_event *event, FILE *fp);
380size_t perf_event__fprintf_mmap(union perf_event *event, FILE *fp);
381size_t perf_event__fprintf_mmap2(union perf_event *event, FILE *fp);
382size_t perf_event__fprintf_task(union perf_event *event, FILE *fp);
383size_t perf_event__fprintf_aux(union perf_event *event, FILE *fp);
384size_t perf_event__fprintf_itrace_start(union perf_event *event, FILE *fp);
385size_t perf_event__fprintf_switch(union perf_event *event, FILE *fp);
386size_t perf_event__fprintf_thread_map(union perf_event *event, FILE *fp);
387size_t perf_event__fprintf_cpu_map(union perf_event *event, FILE *fp);
388size_t perf_event__fprintf_namespaces(union perf_event *event, FILE *fp);
389size_t perf_event__fprintf_cgroup(union perf_event *event, FILE *fp);
390size_t perf_event__fprintf_ksymbol(union perf_event *event, FILE *fp);
391size_t perf_event__fprintf_bpf(union perf_event *event, FILE *fp);
392size_t perf_event__fprintf_text_poke(union perf_event *event, struct machine *machine,FILE *fp);
393size_t perf_event__fprintf(union perf_event *event, struct machine *machine, FILE *fp);
394
395int kallsyms__get_function_start(const char *kallsyms_filename,
396 const char *symbol_name, u64 *addr);
397
398void *cpu_map_data__alloc(struct perf_cpu_map *map, size_t *size, u16 *type, int *max);
399void cpu_map_data__synthesize(struct perf_record_cpu_map_data *data, struct perf_cpu_map *map,
400 u16 type, int max);
401
402void event_attr_init(struct perf_event_attr *attr);
403
404int perf_event_paranoid(void);
405bool perf_event_paranoid_check(int max_level);
406
407extern int sysctl_perf_event_max_stack;
408extern int sysctl_perf_event_max_contexts_per_stack;
409extern unsigned int proc_map_timeout;
410
411#endif /* __PERF_RECORD_H */