1 // SPDX-License-Identifier: GPL-2.0
4 #include <linux/kernel.h>
5 #include <traceevent/event-parse.h>
10 #include <sys/types.h>
21 #include "perf_regs.h"
25 #include "thread-stack.h"
28 static int perf_session__deliver_event(struct perf_session *session,
29 union perf_event *event,
30 struct perf_sample *sample,
31 struct perf_tool *tool,
34 static int perf_session__open(struct perf_session *session)
36 struct perf_data *data = session->data;
38 if (perf_session__read_header(session) < 0) {
39 pr_err("incompatible file format (rerun with -v to learn more)\n");
43 if (perf_data__is_pipe(data))
46 if (perf_header__has_feat(&session->header, HEADER_STAT))
49 if (!perf_evlist__valid_sample_type(session->evlist)) {
50 pr_err("non matching sample_type\n");
54 if (!perf_evlist__valid_sample_id_all(session->evlist)) {
55 pr_err("non matching sample_id_all\n");
59 if (!perf_evlist__valid_read_format(session->evlist)) {
60 pr_err("non matching read_format\n");
67 void perf_session__set_id_hdr_size(struct perf_session *session)
69 u16 id_hdr_size = perf_evlist__id_hdr_size(session->evlist);
71 machines__set_id_hdr_size(&session->machines, id_hdr_size);
74 int perf_session__create_kernel_maps(struct perf_session *session)
76 int ret = machine__create_kernel_maps(&session->machines.host);
79 ret = machines__create_guest_kernel_maps(&session->machines);
83 static void perf_session__destroy_kernel_maps(struct perf_session *session)
85 machines__destroy_kernel_maps(&session->machines);
88 static bool perf_session__has_comm_exec(struct perf_session *session)
90 struct perf_evsel *evsel;
92 evlist__for_each_entry(session->evlist, evsel) {
93 if (evsel->attr.comm_exec)
100 static void perf_session__set_comm_exec(struct perf_session *session)
102 bool comm_exec = perf_session__has_comm_exec(session);
104 machines__set_comm_exec(&session->machines, comm_exec);
107 static int ordered_events__deliver_event(struct ordered_events *oe,
108 struct ordered_event *event)
110 struct perf_sample sample;
111 struct perf_session *session = container_of(oe, struct perf_session,
113 int ret = perf_evlist__parse_sample(session->evlist, event->event, &sample);
116 pr_err("Can't parse sample, err = %d\n", ret);
120 return perf_session__deliver_event(session, event->event, &sample,
121 session->tool, event->file_offset);
124 struct perf_session *perf_session__new(struct perf_data *data,
125 bool repipe, struct perf_tool *tool)
127 struct perf_session *session = zalloc(sizeof(*session));
132 session->repipe = repipe;
133 session->tool = tool;
134 INIT_LIST_HEAD(&session->auxtrace_index);
135 machines__init(&session->machines);
136 ordered_events__init(&session->ordered_events, ordered_events__deliver_event);
139 if (perf_data__open(data))
142 session->data = data;
144 if (perf_data__is_read(data)) {
145 if (perf_session__open(session) < 0)
149 * set session attributes that are present in perf.data
150 * but not in pipe-mode.
152 if (!data->is_pipe) {
153 perf_session__set_id_hdr_size(session);
154 perf_session__set_comm_exec(session);
158 session->machines.host.env = &perf_env;
161 if (!data || perf_data__is_write(data)) {
163 * In O_RDONLY mode this will be performed when reading the
164 * kernel MMAP event, in perf_event__process_mmap().
166 if (perf_session__create_kernel_maps(session) < 0)
167 pr_warning("Cannot read kernel map\n");
171 * In pipe-mode, evlist is empty until PERF_RECORD_HEADER_ATTR is
172 * processed, so perf_evlist__sample_id_all is not meaningful here.
174 if ((!data || !data->is_pipe) && tool && tool->ordering_requires_timestamps &&
175 tool->ordered_events && !perf_evlist__sample_id_all(session->evlist)) {
176 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n");
177 tool->ordered_events = false;
183 perf_data__close(data);
185 perf_session__delete(session);
190 static void perf_session__delete_threads(struct perf_session *session)
192 machine__delete_threads(&session->machines.host);
195 void perf_session__delete(struct perf_session *session)
199 auxtrace__free(session);
200 auxtrace_index__free(&session->auxtrace_index);
201 perf_session__destroy_kernel_maps(session);
202 perf_session__delete_threads(session);
203 perf_env__exit(&session->header.env);
204 machines__exit(&session->machines);
206 perf_data__close(session->data);
210 static int process_event_synth_tracing_data_stub(struct perf_tool *tool
212 union perf_event *event
214 struct perf_session *session
217 dump_printf(": unhandled!\n");
221 static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused,
222 union perf_event *event __maybe_unused,
223 struct perf_evlist **pevlist
226 dump_printf(": unhandled!\n");
230 static int process_event_synth_event_update_stub(struct perf_tool *tool __maybe_unused,
231 union perf_event *event __maybe_unused,
232 struct perf_evlist **pevlist
236 perf_event__fprintf_event_update(event, stdout);
238 dump_printf(": unhandled!\n");
242 static int process_event_sample_stub(struct perf_tool *tool __maybe_unused,
243 union perf_event *event __maybe_unused,
244 struct perf_sample *sample __maybe_unused,
245 struct perf_evsel *evsel __maybe_unused,
246 struct machine *machine __maybe_unused)
248 dump_printf(": unhandled!\n");
252 static int process_event_stub(struct perf_tool *tool __maybe_unused,
253 union perf_event *event __maybe_unused,
254 struct perf_sample *sample __maybe_unused,
255 struct machine *machine __maybe_unused)
257 dump_printf(": unhandled!\n");
261 static int process_finished_round_stub(struct perf_tool *tool __maybe_unused,
262 union perf_event *event __maybe_unused,
263 struct ordered_events *oe __maybe_unused)
265 dump_printf(": unhandled!\n");
269 static int process_finished_round(struct perf_tool *tool,
270 union perf_event *event,
271 struct ordered_events *oe);
273 static int skipn(int fd, off_t n)
279 ret = read(fd, buf, min(n, (off_t)sizeof(buf)));
288 static s64 process_event_auxtrace_stub(struct perf_tool *tool __maybe_unused,
289 union perf_event *event,
290 struct perf_session *session
293 dump_printf(": unhandled!\n");
294 if (perf_data__is_pipe(session->data))
295 skipn(perf_data__fd(session->data), event->auxtrace.size);
296 return event->auxtrace.size;
299 static int process_event_op2_stub(struct perf_tool *tool __maybe_unused,
300 union perf_event *event __maybe_unused,
301 struct perf_session *session __maybe_unused)
303 dump_printf(": unhandled!\n");
309 int process_event_thread_map_stub(struct perf_tool *tool __maybe_unused,
310 union perf_event *event __maybe_unused,
311 struct perf_session *session __maybe_unused)
314 perf_event__fprintf_thread_map(event, stdout);
316 dump_printf(": unhandled!\n");
321 int process_event_cpu_map_stub(struct perf_tool *tool __maybe_unused,
322 union perf_event *event __maybe_unused,
323 struct perf_session *session __maybe_unused)
326 perf_event__fprintf_cpu_map(event, stdout);
328 dump_printf(": unhandled!\n");
333 int process_event_stat_config_stub(struct perf_tool *tool __maybe_unused,
334 union perf_event *event __maybe_unused,
335 struct perf_session *session __maybe_unused)
338 perf_event__fprintf_stat_config(event, stdout);
340 dump_printf(": unhandled!\n");
344 static int process_stat_stub(struct perf_tool *tool __maybe_unused,
345 union perf_event *event __maybe_unused,
346 struct perf_session *perf_session
350 perf_event__fprintf_stat(event, stdout);
352 dump_printf(": unhandled!\n");
356 static int process_stat_round_stub(struct perf_tool *tool __maybe_unused,
357 union perf_event *event __maybe_unused,
358 struct perf_session *perf_session
362 perf_event__fprintf_stat_round(event, stdout);
364 dump_printf(": unhandled!\n");
368 void perf_tool__fill_defaults(struct perf_tool *tool)
370 if (tool->sample == NULL)
371 tool->sample = process_event_sample_stub;
372 if (tool->mmap == NULL)
373 tool->mmap = process_event_stub;
374 if (tool->mmap2 == NULL)
375 tool->mmap2 = process_event_stub;
376 if (tool->comm == NULL)
377 tool->comm = process_event_stub;
378 if (tool->namespaces == NULL)
379 tool->namespaces = process_event_stub;
380 if (tool->fork == NULL)
381 tool->fork = process_event_stub;
382 if (tool->exit == NULL)
383 tool->exit = process_event_stub;
384 if (tool->lost == NULL)
385 tool->lost = perf_event__process_lost;
386 if (tool->lost_samples == NULL)
387 tool->lost_samples = perf_event__process_lost_samples;
388 if (tool->aux == NULL)
389 tool->aux = perf_event__process_aux;
390 if (tool->itrace_start == NULL)
391 tool->itrace_start = perf_event__process_itrace_start;
392 if (tool->context_switch == NULL)
393 tool->context_switch = perf_event__process_switch;
394 if (tool->read == NULL)
395 tool->read = process_event_sample_stub;
396 if (tool->throttle == NULL)
397 tool->throttle = process_event_stub;
398 if (tool->unthrottle == NULL)
399 tool->unthrottle = process_event_stub;
400 if (tool->attr == NULL)
401 tool->attr = process_event_synth_attr_stub;
402 if (tool->event_update == NULL)
403 tool->event_update = process_event_synth_event_update_stub;
404 if (tool->tracing_data == NULL)
405 tool->tracing_data = process_event_synth_tracing_data_stub;
406 if (tool->build_id == NULL)
407 tool->build_id = process_event_op2_stub;
408 if (tool->finished_round == NULL) {
409 if (tool->ordered_events)
410 tool->finished_round = process_finished_round;
412 tool->finished_round = process_finished_round_stub;
414 if (tool->id_index == NULL)
415 tool->id_index = process_event_op2_stub;
416 if (tool->auxtrace_info == NULL)
417 tool->auxtrace_info = process_event_op2_stub;
418 if (tool->auxtrace == NULL)
419 tool->auxtrace = process_event_auxtrace_stub;
420 if (tool->auxtrace_error == NULL)
421 tool->auxtrace_error = process_event_op2_stub;
422 if (tool->thread_map == NULL)
423 tool->thread_map = process_event_thread_map_stub;
424 if (tool->cpu_map == NULL)
425 tool->cpu_map = process_event_cpu_map_stub;
426 if (tool->stat_config == NULL)
427 tool->stat_config = process_event_stat_config_stub;
428 if (tool->stat == NULL)
429 tool->stat = process_stat_stub;
430 if (tool->stat_round == NULL)
431 tool->stat_round = process_stat_round_stub;
432 if (tool->time_conv == NULL)
433 tool->time_conv = process_event_op2_stub;
434 if (tool->feature == NULL)
435 tool->feature = process_event_op2_stub;
438 static void swap_sample_id_all(union perf_event *event, void *data)
440 void *end = (void *) event + event->header.size;
441 int size = end - data;
443 BUG_ON(size % sizeof(u64));
444 mem_bswap_64(data, size);
447 static void perf_event__all64_swap(union perf_event *event,
448 bool sample_id_all __maybe_unused)
450 struct perf_event_header *hdr = &event->header;
451 mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr));
454 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all)
456 event->comm.pid = bswap_32(event->comm.pid);
457 event->comm.tid = bswap_32(event->comm.tid);
460 void *data = &event->comm.comm;
462 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
463 swap_sample_id_all(event, data);
467 static void perf_event__mmap_swap(union perf_event *event,
470 event->mmap.pid = bswap_32(event->mmap.pid);
471 event->mmap.tid = bswap_32(event->mmap.tid);
472 event->mmap.start = bswap_64(event->mmap.start);
473 event->mmap.len = bswap_64(event->mmap.len);
474 event->mmap.pgoff = bswap_64(event->mmap.pgoff);
477 void *data = &event->mmap.filename;
479 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
480 swap_sample_id_all(event, data);
484 static void perf_event__mmap2_swap(union perf_event *event,
487 event->mmap2.pid = bswap_32(event->mmap2.pid);
488 event->mmap2.tid = bswap_32(event->mmap2.tid);
489 event->mmap2.start = bswap_64(event->mmap2.start);
490 event->mmap2.len = bswap_64(event->mmap2.len);
491 event->mmap2.pgoff = bswap_64(event->mmap2.pgoff);
492 event->mmap2.maj = bswap_32(event->mmap2.maj);
493 event->mmap2.min = bswap_32(event->mmap2.min);
494 event->mmap2.ino = bswap_64(event->mmap2.ino);
497 void *data = &event->mmap2.filename;
499 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
500 swap_sample_id_all(event, data);
503 static void perf_event__task_swap(union perf_event *event, bool sample_id_all)
505 event->fork.pid = bswap_32(event->fork.pid);
506 event->fork.tid = bswap_32(event->fork.tid);
507 event->fork.ppid = bswap_32(event->fork.ppid);
508 event->fork.ptid = bswap_32(event->fork.ptid);
509 event->fork.time = bswap_64(event->fork.time);
512 swap_sample_id_all(event, &event->fork + 1);
515 static void perf_event__read_swap(union perf_event *event, bool sample_id_all)
517 event->read.pid = bswap_32(event->read.pid);
518 event->read.tid = bswap_32(event->read.tid);
519 event->read.value = bswap_64(event->read.value);
520 event->read.time_enabled = bswap_64(event->read.time_enabled);
521 event->read.time_running = bswap_64(event->read.time_running);
522 event->read.id = bswap_64(event->read.id);
525 swap_sample_id_all(event, &event->read + 1);
528 static void perf_event__aux_swap(union perf_event *event, bool sample_id_all)
530 event->aux.aux_offset = bswap_64(event->aux.aux_offset);
531 event->aux.aux_size = bswap_64(event->aux.aux_size);
532 event->aux.flags = bswap_64(event->aux.flags);
535 swap_sample_id_all(event, &event->aux + 1);
538 static void perf_event__itrace_start_swap(union perf_event *event,
541 event->itrace_start.pid = bswap_32(event->itrace_start.pid);
542 event->itrace_start.tid = bswap_32(event->itrace_start.tid);
545 swap_sample_id_all(event, &event->itrace_start + 1);
548 static void perf_event__switch_swap(union perf_event *event, bool sample_id_all)
550 if (event->header.type == PERF_RECORD_SWITCH_CPU_WIDE) {
551 event->context_switch.next_prev_pid =
552 bswap_32(event->context_switch.next_prev_pid);
553 event->context_switch.next_prev_tid =
554 bswap_32(event->context_switch.next_prev_tid);
558 swap_sample_id_all(event, &event->context_switch + 1);
561 static void perf_event__throttle_swap(union perf_event *event,
564 event->throttle.time = bswap_64(event->throttle.time);
565 event->throttle.id = bswap_64(event->throttle.id);
566 event->throttle.stream_id = bswap_64(event->throttle.stream_id);
569 swap_sample_id_all(event, &event->throttle + 1);
572 static u8 revbyte(u8 b)
574 int rev = (b >> 4) | ((b & 0xf) << 4);
575 rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2);
576 rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1);
581 * XXX this is hack in attempt to carry flags bitfield
582 * through endian village. ABI says:
584 * Bit-fields are allocated from right to left (least to most significant)
585 * on little-endian implementations and from left to right (most to least
586 * significant) on big-endian implementations.
588 * The above seems to be byte specific, so we need to reverse each
589 * byte of the bitfield. 'Internet' also says this might be implementation
590 * specific and we probably need proper fix and carry perf_event_attr
591 * bitfield flags in separate data file FEAT_ section. Thought this seems
594 static void swap_bitfield(u8 *p, unsigned len)
598 for (i = 0; i < len; i++) {
604 /* exported for swapping attributes in file header */
605 void perf_event__attr_swap(struct perf_event_attr *attr)
607 attr->type = bswap_32(attr->type);
608 attr->size = bswap_32(attr->size);
610 #define bswap_safe(f, n) \
611 (attr->size > (offsetof(struct perf_event_attr, f) + \
612 sizeof(attr->f) * (n)))
613 #define bswap_field(f, sz) \
615 if (bswap_safe(f, 0)) \
616 attr->f = bswap_##sz(attr->f); \
618 #define bswap_field_16(f) bswap_field(f, 16)
619 #define bswap_field_32(f) bswap_field(f, 32)
620 #define bswap_field_64(f) bswap_field(f, 64)
622 bswap_field_64(config);
623 bswap_field_64(sample_period);
624 bswap_field_64(sample_type);
625 bswap_field_64(read_format);
626 bswap_field_32(wakeup_events);
627 bswap_field_32(bp_type);
628 bswap_field_64(bp_addr);
629 bswap_field_64(bp_len);
630 bswap_field_64(branch_sample_type);
631 bswap_field_64(sample_regs_user);
632 bswap_field_32(sample_stack_user);
633 bswap_field_32(aux_watermark);
634 bswap_field_16(sample_max_stack);
637 * After read_format are bitfields. Check read_format because
638 * we are unable to use offsetof on bitfield.
640 if (bswap_safe(read_format, 1))
641 swap_bitfield((u8 *) (&attr->read_format + 1),
643 #undef bswap_field_64
644 #undef bswap_field_32
649 static void perf_event__hdr_attr_swap(union perf_event *event,
650 bool sample_id_all __maybe_unused)
654 perf_event__attr_swap(&event->attr.attr);
656 size = event->header.size;
657 size -= (void *)&event->attr.id - (void *)event;
658 mem_bswap_64(event->attr.id, size);
661 static void perf_event__event_update_swap(union perf_event *event,
662 bool sample_id_all __maybe_unused)
664 event->event_update.type = bswap_64(event->event_update.type);
665 event->event_update.id = bswap_64(event->event_update.id);
668 static void perf_event__event_type_swap(union perf_event *event,
669 bool sample_id_all __maybe_unused)
671 event->event_type.event_type.event_id =
672 bswap_64(event->event_type.event_type.event_id);
675 static void perf_event__tracing_data_swap(union perf_event *event,
676 bool sample_id_all __maybe_unused)
678 event->tracing_data.size = bswap_32(event->tracing_data.size);
681 static void perf_event__auxtrace_info_swap(union perf_event *event,
682 bool sample_id_all __maybe_unused)
686 event->auxtrace_info.type = bswap_32(event->auxtrace_info.type);
688 size = event->header.size;
689 size -= (void *)&event->auxtrace_info.priv - (void *)event;
690 mem_bswap_64(event->auxtrace_info.priv, size);
693 static void perf_event__auxtrace_swap(union perf_event *event,
694 bool sample_id_all __maybe_unused)
696 event->auxtrace.size = bswap_64(event->auxtrace.size);
697 event->auxtrace.offset = bswap_64(event->auxtrace.offset);
698 event->auxtrace.reference = bswap_64(event->auxtrace.reference);
699 event->auxtrace.idx = bswap_32(event->auxtrace.idx);
700 event->auxtrace.tid = bswap_32(event->auxtrace.tid);
701 event->auxtrace.cpu = bswap_32(event->auxtrace.cpu);
704 static void perf_event__auxtrace_error_swap(union perf_event *event,
705 bool sample_id_all __maybe_unused)
707 event->auxtrace_error.type = bswap_32(event->auxtrace_error.type);
708 event->auxtrace_error.code = bswap_32(event->auxtrace_error.code);
709 event->auxtrace_error.cpu = bswap_32(event->auxtrace_error.cpu);
710 event->auxtrace_error.pid = bswap_32(event->auxtrace_error.pid);
711 event->auxtrace_error.tid = bswap_32(event->auxtrace_error.tid);
712 event->auxtrace_error.ip = bswap_64(event->auxtrace_error.ip);
715 static void perf_event__thread_map_swap(union perf_event *event,
716 bool sample_id_all __maybe_unused)
720 event->thread_map.nr = bswap_64(event->thread_map.nr);
722 for (i = 0; i < event->thread_map.nr; i++)
723 event->thread_map.entries[i].pid = bswap_64(event->thread_map.entries[i].pid);
726 static void perf_event__cpu_map_swap(union perf_event *event,
727 bool sample_id_all __maybe_unused)
729 struct cpu_map_data *data = &event->cpu_map.data;
730 struct cpu_map_entries *cpus;
731 struct cpu_map_mask *mask;
734 data->type = bswap_64(data->type);
736 switch (data->type) {
737 case PERF_CPU_MAP__CPUS:
738 cpus = (struct cpu_map_entries *)data->data;
740 cpus->nr = bswap_16(cpus->nr);
742 for (i = 0; i < cpus->nr; i++)
743 cpus->cpu[i] = bswap_16(cpus->cpu[i]);
745 case PERF_CPU_MAP__MASK:
746 mask = (struct cpu_map_mask *) data->data;
748 mask->nr = bswap_16(mask->nr);
749 mask->long_size = bswap_16(mask->long_size);
751 switch (mask->long_size) {
752 case 4: mem_bswap_32(&mask->mask, mask->nr); break;
753 case 8: mem_bswap_64(&mask->mask, mask->nr); break;
755 pr_err("cpu_map swap: unsupported long size\n");
762 static void perf_event__stat_config_swap(union perf_event *event,
763 bool sample_id_all __maybe_unused)
767 size = event->stat_config.nr * sizeof(event->stat_config.data[0]);
768 size += 1; /* nr item itself */
769 mem_bswap_64(&event->stat_config.nr, size);
772 static void perf_event__stat_swap(union perf_event *event,
773 bool sample_id_all __maybe_unused)
775 event->stat.id = bswap_64(event->stat.id);
776 event->stat.thread = bswap_32(event->stat.thread);
777 event->stat.cpu = bswap_32(event->stat.cpu);
778 event->stat.val = bswap_64(event->stat.val);
779 event->stat.ena = bswap_64(event->stat.ena);
780 event->stat.run = bswap_64(event->stat.run);
783 static void perf_event__stat_round_swap(union perf_event *event,
784 bool sample_id_all __maybe_unused)
786 event->stat_round.type = bswap_64(event->stat_round.type);
787 event->stat_round.time = bswap_64(event->stat_round.time);
790 typedef void (*perf_event__swap_op)(union perf_event *event,
793 static perf_event__swap_op perf_event__swap_ops[] = {
794 [PERF_RECORD_MMAP] = perf_event__mmap_swap,
795 [PERF_RECORD_MMAP2] = perf_event__mmap2_swap,
796 [PERF_RECORD_COMM] = perf_event__comm_swap,
797 [PERF_RECORD_FORK] = perf_event__task_swap,
798 [PERF_RECORD_EXIT] = perf_event__task_swap,
799 [PERF_RECORD_LOST] = perf_event__all64_swap,
800 [PERF_RECORD_READ] = perf_event__read_swap,
801 [PERF_RECORD_THROTTLE] = perf_event__throttle_swap,
802 [PERF_RECORD_UNTHROTTLE] = perf_event__throttle_swap,
803 [PERF_RECORD_SAMPLE] = perf_event__all64_swap,
804 [PERF_RECORD_AUX] = perf_event__aux_swap,
805 [PERF_RECORD_ITRACE_START] = perf_event__itrace_start_swap,
806 [PERF_RECORD_LOST_SAMPLES] = perf_event__all64_swap,
807 [PERF_RECORD_SWITCH] = perf_event__switch_swap,
808 [PERF_RECORD_SWITCH_CPU_WIDE] = perf_event__switch_swap,
809 [PERF_RECORD_HEADER_ATTR] = perf_event__hdr_attr_swap,
810 [PERF_RECORD_HEADER_EVENT_TYPE] = perf_event__event_type_swap,
811 [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
812 [PERF_RECORD_HEADER_BUILD_ID] = NULL,
813 [PERF_RECORD_ID_INDEX] = perf_event__all64_swap,
814 [PERF_RECORD_AUXTRACE_INFO] = perf_event__auxtrace_info_swap,
815 [PERF_RECORD_AUXTRACE] = perf_event__auxtrace_swap,
816 [PERF_RECORD_AUXTRACE_ERROR] = perf_event__auxtrace_error_swap,
817 [PERF_RECORD_THREAD_MAP] = perf_event__thread_map_swap,
818 [PERF_RECORD_CPU_MAP] = perf_event__cpu_map_swap,
819 [PERF_RECORD_STAT_CONFIG] = perf_event__stat_config_swap,
820 [PERF_RECORD_STAT] = perf_event__stat_swap,
821 [PERF_RECORD_STAT_ROUND] = perf_event__stat_round_swap,
822 [PERF_RECORD_EVENT_UPDATE] = perf_event__event_update_swap,
823 [PERF_RECORD_TIME_CONV] = perf_event__all64_swap,
824 [PERF_RECORD_HEADER_MAX] = NULL,
828 * When perf record finishes a pass on every buffers, it records this pseudo
830 * We record the max timestamp t found in the pass n.
831 * Assuming these timestamps are monotonic across cpus, we know that if
832 * a buffer still has events with timestamps below t, they will be all
833 * available and then read in the pass n + 1.
834 * Hence when we start to read the pass n + 2, we can safely flush every
835 * events with timestamps below t.
837 * ============ PASS n =================
840 * cnt1 timestamps | cnt2 timestamps
843 * - | 4 <--- max recorded
845 * ============ PASS n + 1 ==============
848 * cnt1 timestamps | cnt2 timestamps
851 * 5 | 7 <---- max recorded
853 * Flush every events below timestamp 4
855 * ============ PASS n + 2 ==============
858 * cnt1 timestamps | cnt2 timestamps
863 * Flush every events below timestamp 7
866 static int process_finished_round(struct perf_tool *tool __maybe_unused,
867 union perf_event *event __maybe_unused,
868 struct ordered_events *oe)
871 fprintf(stdout, "\n");
872 return ordered_events__flush(oe, OE_FLUSH__ROUND);
875 int perf_session__queue_event(struct perf_session *s, union perf_event *event,
876 struct perf_sample *sample, u64 file_offset)
878 return ordered_events__queue(&s->ordered_events, event, sample, file_offset);
881 static void callchain__lbr_callstack_printf(struct perf_sample *sample)
883 struct ip_callchain *callchain = sample->callchain;
884 struct branch_stack *lbr_stack = sample->branch_stack;
885 u64 kernel_callchain_nr = callchain->nr;
888 for (i = 0; i < kernel_callchain_nr; i++) {
889 if (callchain->ips[i] == PERF_CONTEXT_USER)
893 if ((i != kernel_callchain_nr) && lbr_stack->nr) {
896 * LBR callstack can only get user call chain,
897 * i is kernel call chain number,
898 * 1 is PERF_CONTEXT_USER.
900 * The user call chain is stored in LBR registers.
901 * LBR are pair registers. The caller is stored
902 * in "from" register, while the callee is stored
904 * For example, there is a call stack
905 * "A"->"B"->"C"->"D".
906 * The LBR registers will recorde like
907 * "C"->"D", "B"->"C", "A"->"B".
908 * So only the first "to" register and all "from"
909 * registers are needed to construct the whole stack.
911 total_nr = i + 1 + lbr_stack->nr + 1;
912 kernel_callchain_nr = i + 1;
914 printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr);
916 for (i = 0; i < kernel_callchain_nr; i++)
917 printf("..... %2d: %016" PRIx64 "\n",
918 i, callchain->ips[i]);
920 printf("..... %2d: %016" PRIx64 "\n",
921 (int)(kernel_callchain_nr), lbr_stack->entries[0].to);
922 for (i = 0; i < lbr_stack->nr; i++)
923 printf("..... %2d: %016" PRIx64 "\n",
924 (int)(i + kernel_callchain_nr + 1), lbr_stack->entries[i].from);
928 static void callchain__printf(struct perf_evsel *evsel,
929 struct perf_sample *sample)
932 struct ip_callchain *callchain = sample->callchain;
934 if (perf_evsel__has_branch_callstack(evsel))
935 callchain__lbr_callstack_printf(sample);
937 printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr);
939 for (i = 0; i < callchain->nr; i++)
940 printf("..... %2d: %016" PRIx64 "\n",
941 i, callchain->ips[i]);
944 static void branch_stack__printf(struct perf_sample *sample)
948 printf("... branch stack: nr:%" PRIu64 "\n", sample->branch_stack->nr);
950 for (i = 0; i < sample->branch_stack->nr; i++) {
951 struct branch_entry *e = &sample->branch_stack->entries[i];
953 printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x\n",
955 (unsigned short)e->flags.cycles,
956 e->flags.mispred ? "M" : " ",
957 e->flags.predicted ? "P" : " ",
958 e->flags.abort ? "A" : " ",
959 e->flags.in_tx ? "T" : " ",
960 (unsigned)e->flags.reserved);
964 static void regs_dump__printf(u64 mask, u64 *regs)
968 for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) {
971 printf(".... %-5s 0x%" PRIx64 "\n",
972 perf_reg_name(rid), val);
976 static const char *regs_abi[] = {
977 [PERF_SAMPLE_REGS_ABI_NONE] = "none",
978 [PERF_SAMPLE_REGS_ABI_32] = "32-bit",
979 [PERF_SAMPLE_REGS_ABI_64] = "64-bit",
982 static inline const char *regs_dump_abi(struct regs_dump *d)
984 if (d->abi > PERF_SAMPLE_REGS_ABI_64)
987 return regs_abi[d->abi];
990 static void regs__printf(const char *type, struct regs_dump *regs)
992 u64 mask = regs->mask;
994 printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n",
997 regs_dump_abi(regs));
999 regs_dump__printf(mask, regs->regs);
1002 static void regs_user__printf(struct perf_sample *sample)
1004 struct regs_dump *user_regs = &sample->user_regs;
1006 if (user_regs->regs)
1007 regs__printf("user", user_regs);
1010 static void regs_intr__printf(struct perf_sample *sample)
1012 struct regs_dump *intr_regs = &sample->intr_regs;
1014 if (intr_regs->regs)
1015 regs__printf("intr", intr_regs);
1018 static void stack_user__printf(struct stack_dump *dump)
1020 printf("... ustack: size %" PRIu64 ", offset 0x%x\n",
1021 dump->size, dump->offset);
1024 static void perf_evlist__print_tstamp(struct perf_evlist *evlist,
1025 union perf_event *event,
1026 struct perf_sample *sample)
1028 u64 sample_type = __perf_evlist__combined_sample_type(evlist);
1030 if (event->header.type != PERF_RECORD_SAMPLE &&
1031 !perf_evlist__sample_id_all(evlist)) {
1032 fputs("-1 -1 ", stdout);
1036 if ((sample_type & PERF_SAMPLE_CPU))
1037 printf("%u ", sample->cpu);
1039 if (sample_type & PERF_SAMPLE_TIME)
1040 printf("%" PRIu64 " ", sample->time);
1043 static void sample_read__printf(struct perf_sample *sample, u64 read_format)
1045 printf("... sample_read:\n");
1047 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1048 printf("...... time enabled %016" PRIx64 "\n",
1049 sample->read.time_enabled);
1051 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1052 printf("...... time running %016" PRIx64 "\n",
1053 sample->read.time_running);
1055 if (read_format & PERF_FORMAT_GROUP) {
1058 printf(".... group nr %" PRIu64 "\n", sample->read.group.nr);
1060 for (i = 0; i < sample->read.group.nr; i++) {
1061 struct sample_read_value *value;
1063 value = &sample->read.group.values[i];
1064 printf("..... id %016" PRIx64
1065 ", value %016" PRIx64 "\n",
1066 value->id, value->value);
1069 printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n",
1070 sample->read.one.id, sample->read.one.value);
1073 static void dump_event(struct perf_evlist *evlist, union perf_event *event,
1074 u64 file_offset, struct perf_sample *sample)
1079 printf("\n%#" PRIx64 " [%#x]: event: %d\n",
1080 file_offset, event->header.size, event->header.type);
1085 perf_evlist__print_tstamp(evlist, event, sample);
1087 printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset,
1088 event->header.size, perf_event__name(event->header.type));
1091 static void dump_sample(struct perf_evsel *evsel, union perf_event *event,
1092 struct perf_sample *sample)
1099 printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n",
1100 event->header.misc, sample->pid, sample->tid, sample->ip,
1101 sample->period, sample->addr);
1103 sample_type = evsel->attr.sample_type;
1105 if (sample_type & PERF_SAMPLE_CALLCHAIN)
1106 callchain__printf(evsel, sample);
1108 if ((sample_type & PERF_SAMPLE_BRANCH_STACK) && !perf_evsel__has_branch_callstack(evsel))
1109 branch_stack__printf(sample);
1111 if (sample_type & PERF_SAMPLE_REGS_USER)
1112 regs_user__printf(sample);
1114 if (sample_type & PERF_SAMPLE_REGS_INTR)
1115 regs_intr__printf(sample);
1117 if (sample_type & PERF_SAMPLE_STACK_USER)
1118 stack_user__printf(&sample->user_stack);
1120 if (sample_type & PERF_SAMPLE_WEIGHT)
1121 printf("... weight: %" PRIu64 "\n", sample->weight);
1123 if (sample_type & PERF_SAMPLE_DATA_SRC)
1124 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src);
1126 if (sample_type & PERF_SAMPLE_PHYS_ADDR)
1127 printf(" .. phys_addr: 0x%"PRIx64"\n", sample->phys_addr);
1129 if (sample_type & PERF_SAMPLE_TRANSACTION)
1130 printf("... transaction: %" PRIx64 "\n", sample->transaction);
1132 if (sample_type & PERF_SAMPLE_READ)
1133 sample_read__printf(sample, evsel->attr.read_format);
1136 static void dump_read(struct perf_evsel *evsel, union perf_event *event)
1138 struct read_event *read_event = &event->read;
1144 printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid,
1145 evsel ? perf_evsel__name(evsel) : "FAIL",
1148 read_format = evsel->attr.read_format;
1150 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1151 printf("... time enabled : %" PRIu64 "\n", read_event->time_enabled);
1153 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1154 printf("... time running : %" PRIu64 "\n", read_event->time_running);
1156 if (read_format & PERF_FORMAT_ID)
1157 printf("... id : %" PRIu64 "\n", read_event->id);
1160 static struct machine *machines__find_for_cpumode(struct machines *machines,
1161 union perf_event *event,
1162 struct perf_sample *sample)
1164 struct machine *machine;
1167 ((sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL) ||
1168 (sample->cpumode == PERF_RECORD_MISC_GUEST_USER))) {
1171 if (event->header.type == PERF_RECORD_MMAP
1172 || event->header.type == PERF_RECORD_MMAP2)
1173 pid = event->mmap.pid;
1177 machine = machines__find(machines, pid);
1179 machine = machines__findnew(machines, DEFAULT_GUEST_KERNEL_ID);
1183 return &machines->host;
1186 static int deliver_sample_value(struct perf_evlist *evlist,
1187 struct perf_tool *tool,
1188 union perf_event *event,
1189 struct perf_sample *sample,
1190 struct sample_read_value *v,
1191 struct machine *machine)
1193 struct perf_sample_id *sid = perf_evlist__id2sid(evlist, v->id);
1197 sample->period = v->value - sid->period;
1198 sid->period = v->value;
1201 if (!sid || sid->evsel == NULL) {
1202 ++evlist->stats.nr_unknown_id;
1206 return tool->sample(tool, event, sample, sid->evsel, machine);
1209 static int deliver_sample_group(struct perf_evlist *evlist,
1210 struct perf_tool *tool,
1211 union perf_event *event,
1212 struct perf_sample *sample,
1213 struct machine *machine)
1218 for (i = 0; i < sample->read.group.nr; i++) {
1219 ret = deliver_sample_value(evlist, tool, event, sample,
1220 &sample->read.group.values[i],
1230 perf_evlist__deliver_sample(struct perf_evlist *evlist,
1231 struct perf_tool *tool,
1232 union perf_event *event,
1233 struct perf_sample *sample,
1234 struct perf_evsel *evsel,
1235 struct machine *machine)
1237 /* We know evsel != NULL. */
1238 u64 sample_type = evsel->attr.sample_type;
1239 u64 read_format = evsel->attr.read_format;
1241 /* Standard sample delivery. */
1242 if (!(sample_type & PERF_SAMPLE_READ))
1243 return tool->sample(tool, event, sample, evsel, machine);
1245 /* For PERF_SAMPLE_READ we have either single or group mode. */
1246 if (read_format & PERF_FORMAT_GROUP)
1247 return deliver_sample_group(evlist, tool, event, sample,
1250 return deliver_sample_value(evlist, tool, event, sample,
1251 &sample->read.one, machine);
1254 static int machines__deliver_event(struct machines *machines,
1255 struct perf_evlist *evlist,
1256 union perf_event *event,
1257 struct perf_sample *sample,
1258 struct perf_tool *tool, u64 file_offset)
1260 struct perf_evsel *evsel;
1261 struct machine *machine;
1263 dump_event(evlist, event, file_offset, sample);
1265 evsel = perf_evlist__id2evsel(evlist, sample->id);
1267 machine = machines__find_for_cpumode(machines, event, sample);
1269 switch (event->header.type) {
1270 case PERF_RECORD_SAMPLE:
1271 if (evsel == NULL) {
1272 ++evlist->stats.nr_unknown_id;
1275 dump_sample(evsel, event, sample);
1276 if (machine == NULL) {
1277 ++evlist->stats.nr_unprocessable_samples;
1280 return perf_evlist__deliver_sample(evlist, tool, event, sample, evsel, machine);
1281 case PERF_RECORD_MMAP:
1282 return tool->mmap(tool, event, sample, machine);
1283 case PERF_RECORD_MMAP2:
1284 if (event->header.misc & PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT)
1285 ++evlist->stats.nr_proc_map_timeout;
1286 return tool->mmap2(tool, event, sample, machine);
1287 case PERF_RECORD_COMM:
1288 return tool->comm(tool, event, sample, machine);
1289 case PERF_RECORD_NAMESPACES:
1290 return tool->namespaces(tool, event, sample, machine);
1291 case PERF_RECORD_FORK:
1292 return tool->fork(tool, event, sample, machine);
1293 case PERF_RECORD_EXIT:
1294 return tool->exit(tool, event, sample, machine);
1295 case PERF_RECORD_LOST:
1296 if (tool->lost == perf_event__process_lost)
1297 evlist->stats.total_lost += event->lost.lost;
1298 return tool->lost(tool, event, sample, machine);
1299 case PERF_RECORD_LOST_SAMPLES:
1300 if (tool->lost_samples == perf_event__process_lost_samples)
1301 evlist->stats.total_lost_samples += event->lost_samples.lost;
1302 return tool->lost_samples(tool, event, sample, machine);
1303 case PERF_RECORD_READ:
1304 dump_read(evsel, event);
1305 return tool->read(tool, event, sample, evsel, machine);
1306 case PERF_RECORD_THROTTLE:
1307 return tool->throttle(tool, event, sample, machine);
1308 case PERF_RECORD_UNTHROTTLE:
1309 return tool->unthrottle(tool, event, sample, machine);
1310 case PERF_RECORD_AUX:
1311 if (tool->aux == perf_event__process_aux) {
1312 if (event->aux.flags & PERF_AUX_FLAG_TRUNCATED)
1313 evlist->stats.total_aux_lost += 1;
1314 if (event->aux.flags & PERF_AUX_FLAG_PARTIAL)
1315 evlist->stats.total_aux_partial += 1;
1317 return tool->aux(tool, event, sample, machine);
1318 case PERF_RECORD_ITRACE_START:
1319 return tool->itrace_start(tool, event, sample, machine);
1320 case PERF_RECORD_SWITCH:
1321 case PERF_RECORD_SWITCH_CPU_WIDE:
1322 return tool->context_switch(tool, event, sample, machine);
1324 ++evlist->stats.nr_unknown_events;
1329 static int perf_session__deliver_event(struct perf_session *session,
1330 union perf_event *event,
1331 struct perf_sample *sample,
1332 struct perf_tool *tool,
1337 ret = auxtrace__process_event(session, event, sample, tool);
1343 return machines__deliver_event(&session->machines, session->evlist,
1344 event, sample, tool, file_offset);
1347 static s64 perf_session__process_user_event(struct perf_session *session,
1348 union perf_event *event,
1351 struct ordered_events *oe = &session->ordered_events;
1352 struct perf_tool *tool = session->tool;
1353 int fd = perf_data__fd(session->data);
1356 dump_event(session->evlist, event, file_offset, NULL);
1358 /* These events are processed right away */
1359 switch (event->header.type) {
1360 case PERF_RECORD_HEADER_ATTR:
1361 err = tool->attr(tool, event, &session->evlist);
1363 perf_session__set_id_hdr_size(session);
1364 perf_session__set_comm_exec(session);
1367 case PERF_RECORD_EVENT_UPDATE:
1368 return tool->event_update(tool, event, &session->evlist);
1369 case PERF_RECORD_HEADER_EVENT_TYPE:
1371 * Depreceated, but we need to handle it for sake
1372 * of old data files create in pipe mode.
1375 case PERF_RECORD_HEADER_TRACING_DATA:
1376 /* setup for reading amidst mmap */
1377 lseek(fd, file_offset, SEEK_SET);
1378 return tool->tracing_data(tool, event, session);
1379 case PERF_RECORD_HEADER_BUILD_ID:
1380 return tool->build_id(tool, event, session);
1381 case PERF_RECORD_FINISHED_ROUND:
1382 return tool->finished_round(tool, event, oe);
1383 case PERF_RECORD_ID_INDEX:
1384 return tool->id_index(tool, event, session);
1385 case PERF_RECORD_AUXTRACE_INFO:
1386 return tool->auxtrace_info(tool, event, session);
1387 case PERF_RECORD_AUXTRACE:
1388 /* setup for reading amidst mmap */
1389 lseek(fd, file_offset + event->header.size, SEEK_SET);
1390 return tool->auxtrace(tool, event, session);
1391 case PERF_RECORD_AUXTRACE_ERROR:
1392 perf_session__auxtrace_error_inc(session, event);
1393 return tool->auxtrace_error(tool, event, session);
1394 case PERF_RECORD_THREAD_MAP:
1395 return tool->thread_map(tool, event, session);
1396 case PERF_RECORD_CPU_MAP:
1397 return tool->cpu_map(tool, event, session);
1398 case PERF_RECORD_STAT_CONFIG:
1399 return tool->stat_config(tool, event, session);
1400 case PERF_RECORD_STAT:
1401 return tool->stat(tool, event, session);
1402 case PERF_RECORD_STAT_ROUND:
1403 return tool->stat_round(tool, event, session);
1404 case PERF_RECORD_TIME_CONV:
1405 session->time_conv = event->time_conv;
1406 return tool->time_conv(tool, event, session);
1407 case PERF_RECORD_HEADER_FEATURE:
1408 return tool->feature(tool, event, session);
1414 int perf_session__deliver_synth_event(struct perf_session *session,
1415 union perf_event *event,
1416 struct perf_sample *sample)
1418 struct perf_evlist *evlist = session->evlist;
1419 struct perf_tool *tool = session->tool;
1421 events_stats__inc(&evlist->stats, event->header.type);
1423 if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1424 return perf_session__process_user_event(session, event, 0);
1426 return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0);
1429 static void event_swap(union perf_event *event, bool sample_id_all)
1431 perf_event__swap_op swap;
1433 swap = perf_event__swap_ops[event->header.type];
1435 swap(event, sample_id_all);
1438 int perf_session__peek_event(struct perf_session *session, off_t file_offset,
1439 void *buf, size_t buf_sz,
1440 union perf_event **event_ptr,
1441 struct perf_sample *sample)
1443 union perf_event *event;
1444 size_t hdr_sz, rest;
1447 if (session->one_mmap && !session->header.needs_swap) {
1448 event = file_offset - session->one_mmap_offset +
1449 session->one_mmap_addr;
1450 goto out_parse_sample;
1453 if (perf_data__is_pipe(session->data))
1456 fd = perf_data__fd(session->data);
1457 hdr_sz = sizeof(struct perf_event_header);
1459 if (buf_sz < hdr_sz)
1462 if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 ||
1463 readn(fd, buf, hdr_sz) != (ssize_t)hdr_sz)
1466 event = (union perf_event *)buf;
1468 if (session->header.needs_swap)
1469 perf_event_header__bswap(&event->header);
1471 if (event->header.size < hdr_sz || event->header.size > buf_sz)
1474 rest = event->header.size - hdr_sz;
1476 if (readn(fd, buf, rest) != (ssize_t)rest)
1479 if (session->header.needs_swap)
1480 event_swap(event, perf_evlist__sample_id_all(session->evlist));
1484 if (sample && event->header.type < PERF_RECORD_USER_TYPE_START &&
1485 perf_evlist__parse_sample(session->evlist, event, sample))
1493 static s64 perf_session__process_event(struct perf_session *session,
1494 union perf_event *event, u64 file_offset)
1496 struct perf_evlist *evlist = session->evlist;
1497 struct perf_tool *tool = session->tool;
1498 struct perf_sample sample;
1501 if (session->header.needs_swap)
1502 event_swap(event, perf_evlist__sample_id_all(evlist));
1504 if (event->header.type >= PERF_RECORD_HEADER_MAX)
1507 events_stats__inc(&evlist->stats, event->header.type);
1509 if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1510 return perf_session__process_user_event(session, event, file_offset);
1513 * For all kernel events we get the sample data
1515 ret = perf_evlist__parse_sample(evlist, event, &sample);
1519 if (tool->ordered_events) {
1520 ret = perf_session__queue_event(session, event, &sample, file_offset);
1525 return perf_session__deliver_event(session, event, &sample, tool,
1529 void perf_event_header__bswap(struct perf_event_header *hdr)
1531 hdr->type = bswap_32(hdr->type);
1532 hdr->misc = bswap_16(hdr->misc);
1533 hdr->size = bswap_16(hdr->size);
1536 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid)
1538 return machine__findnew_thread(&session->machines.host, -1, pid);
1541 int perf_session__register_idle_thread(struct perf_session *session)
1543 struct thread *thread;
1546 thread = machine__findnew_thread(&session->machines.host, 0, 0);
1547 if (thread == NULL || thread__set_comm(thread, "swapper", 0)) {
1548 pr_err("problem inserting idle task.\n");
1552 if (thread == NULL || thread__set_namespaces(thread, 0, NULL)) {
1553 pr_err("problem inserting idle task.\n");
1557 /* machine__findnew_thread() got the thread, so put it */
1558 thread__put(thread);
1563 perf_session__warn_order(const struct perf_session *session)
1565 const struct ordered_events *oe = &session->ordered_events;
1566 struct perf_evsel *evsel;
1567 bool should_warn = true;
1569 evlist__for_each_entry(session->evlist, evsel) {
1570 if (evsel->attr.write_backward)
1571 should_warn = false;
1576 if (oe->nr_unordered_events != 0)
1577 ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events);
1580 static void perf_session__warn_about_errors(const struct perf_session *session)
1582 const struct events_stats *stats = &session->evlist->stats;
1584 if (session->tool->lost == perf_event__process_lost &&
1585 stats->nr_events[PERF_RECORD_LOST] != 0) {
1586 ui__warning("Processed %d events and lost %d chunks!\n\n"
1587 "Check IO/CPU overload!\n\n",
1588 stats->nr_events[0],
1589 stats->nr_events[PERF_RECORD_LOST]);
1592 if (session->tool->lost_samples == perf_event__process_lost_samples) {
1595 drop_rate = (double)stats->total_lost_samples /
1596 (double) (stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples);
1597 if (drop_rate > 0.05) {
1598 ui__warning("Processed %" PRIu64 " samples and lost %3.2f%% samples!\n\n",
1599 stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples,
1604 if (session->tool->aux == perf_event__process_aux &&
1605 stats->total_aux_lost != 0) {
1606 ui__warning("AUX data lost %" PRIu64 " times out of %u!\n\n",
1607 stats->total_aux_lost,
1608 stats->nr_events[PERF_RECORD_AUX]);
1611 if (session->tool->aux == perf_event__process_aux &&
1612 stats->total_aux_partial != 0) {
1613 bool vmm_exclusive = false;
1615 (void)sysfs__read_bool("module/kvm_intel/parameters/vmm_exclusive",
1618 ui__warning("AUX data had gaps in it %" PRIu64 " times out of %u!\n\n"
1619 "Are you running a KVM guest in the background?%s\n\n",
1620 stats->total_aux_partial,
1621 stats->nr_events[PERF_RECORD_AUX],
1623 "\nReloading kvm_intel module with vmm_exclusive=0\n"
1624 "will reduce the gaps to only guest's timeslices." :
1628 if (stats->nr_unknown_events != 0) {
1629 ui__warning("Found %u unknown events!\n\n"
1630 "Is this an older tool processing a perf.data "
1631 "file generated by a more recent tool?\n\n"
1632 "If that is not the case, consider "
1633 "reporting to linux-kernel@vger.kernel.org.\n\n",
1634 stats->nr_unknown_events);
1637 if (stats->nr_unknown_id != 0) {
1638 ui__warning("%u samples with id not present in the header\n",
1639 stats->nr_unknown_id);
1642 if (stats->nr_invalid_chains != 0) {
1643 ui__warning("Found invalid callchains!\n\n"
1644 "%u out of %u events were discarded for this reason.\n\n"
1645 "Consider reporting to linux-kernel@vger.kernel.org.\n\n",
1646 stats->nr_invalid_chains,
1647 stats->nr_events[PERF_RECORD_SAMPLE]);
1650 if (stats->nr_unprocessable_samples != 0) {
1651 ui__warning("%u unprocessable samples recorded.\n"
1652 "Do you have a KVM guest running and not using 'perf kvm'?\n",
1653 stats->nr_unprocessable_samples);
1656 perf_session__warn_order(session);
1658 events_stats__auxtrace_error_warn(stats);
1660 if (stats->nr_proc_map_timeout != 0) {
1661 ui__warning("%d map information files for pre-existing threads were\n"
1662 "not processed, if there are samples for addresses they\n"
1663 "will not be resolved, you may find out which are these\n"
1664 "threads by running with -v and redirecting the output\n"
1666 "The time limit to process proc map is too short?\n"
1667 "Increase it by --proc-map-timeout\n",
1668 stats->nr_proc_map_timeout);
1672 static int perf_session__flush_thread_stack(struct thread *thread,
1673 void *p __maybe_unused)
1675 return thread_stack__flush(thread);
1678 static int perf_session__flush_thread_stacks(struct perf_session *session)
1680 return machines__for_each_thread(&session->machines,
1681 perf_session__flush_thread_stack,
1685 volatile int session_done;
1687 static int __perf_session__process_pipe_events(struct perf_session *session)
1689 struct ordered_events *oe = &session->ordered_events;
1690 struct perf_tool *tool = session->tool;
1691 int fd = perf_data__fd(session->data);
1692 union perf_event *event;
1693 uint32_t size, cur_size = 0;
1700 perf_tool__fill_defaults(tool);
1703 cur_size = sizeof(union perf_event);
1705 buf = malloc(cur_size);
1708 ordered_events__set_copy_on_queue(oe, true);
1711 err = readn(fd, event, sizeof(struct perf_event_header));
1716 pr_err("failed to read event header\n");
1720 if (session->header.needs_swap)
1721 perf_event_header__bswap(&event->header);
1723 size = event->header.size;
1724 if (size < sizeof(struct perf_event_header)) {
1725 pr_err("bad event header size\n");
1729 if (size > cur_size) {
1730 void *new = realloc(buf, size);
1732 pr_err("failed to allocate memory to read event\n");
1740 p += sizeof(struct perf_event_header);
1742 if (size - sizeof(struct perf_event_header)) {
1743 err = readn(fd, p, size - sizeof(struct perf_event_header));
1746 pr_err("unexpected end of event stream\n");
1750 pr_err("failed to read event data\n");
1755 if ((skip = perf_session__process_event(session, event, head)) < 0) {
1756 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1757 head, event->header.size, event->header.type);
1767 if (!session_done())
1770 /* do the final flush for ordered samples */
1771 err = ordered_events__flush(oe, OE_FLUSH__FINAL);
1774 err = auxtrace__flush_events(session, tool);
1777 err = perf_session__flush_thread_stacks(session);
1780 perf_session__warn_about_errors(session);
1781 ordered_events__free(&session->ordered_events);
1782 auxtrace__free_events(session);
1786 static union perf_event *
1787 fetch_mmaped_event(struct perf_session *session,
1788 u64 head, size_t mmap_size, char *buf)
1790 union perf_event *event;
1793 * Ensure we have enough space remaining to read
1794 * the size of the event in the headers.
1796 if (head + sizeof(event->header) > mmap_size)
1799 event = (union perf_event *)(buf + head);
1801 if (session->header.needs_swap)
1802 perf_event_header__bswap(&event->header);
1804 if (head + event->header.size > mmap_size) {
1805 /* We're not fetching the event so swap back again */
1806 if (session->header.needs_swap)
1807 perf_event_header__bswap(&event->header);
1815 * On 64bit we can mmap the data file in one go. No need for tiny mmap
1816 * slices. On 32bit we use 32MB.
1818 #if BITS_PER_LONG == 64
1819 #define MMAP_SIZE ULLONG_MAX
1822 #define MMAP_SIZE (32 * 1024 * 1024ULL)
1823 #define NUM_MMAPS 128
1826 static int __perf_session__process_events(struct perf_session *session,
1827 u64 data_offset, u64 data_size,
1830 struct ordered_events *oe = &session->ordered_events;
1831 struct perf_tool *tool = session->tool;
1832 int fd = perf_data__fd(session->data);
1833 u64 head, page_offset, file_offset, file_pos, size;
1834 int err, mmap_prot, mmap_flags, map_idx = 0;
1836 char *buf, *mmaps[NUM_MMAPS];
1837 union perf_event *event;
1838 struct ui_progress prog;
1841 perf_tool__fill_defaults(tool);
1843 page_offset = page_size * (data_offset / page_size);
1844 file_offset = page_offset;
1845 head = data_offset - page_offset;
1850 if (data_offset + data_size < file_size)
1851 file_size = data_offset + data_size;
1853 ui_progress__init_size(&prog, file_size, "Processing events...");
1855 mmap_size = MMAP_SIZE;
1856 if (mmap_size > file_size) {
1857 mmap_size = file_size;
1858 session->one_mmap = true;
1861 memset(mmaps, 0, sizeof(mmaps));
1863 mmap_prot = PROT_READ;
1864 mmap_flags = MAP_SHARED;
1866 if (session->header.needs_swap) {
1867 mmap_prot |= PROT_WRITE;
1868 mmap_flags = MAP_PRIVATE;
1871 buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, fd,
1873 if (buf == MAP_FAILED) {
1874 pr_err("failed to mmap file\n");
1878 mmaps[map_idx] = buf;
1879 map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1);
1880 file_pos = file_offset + head;
1881 if (session->one_mmap) {
1882 session->one_mmap_addr = buf;
1883 session->one_mmap_offset = file_offset;
1887 event = fetch_mmaped_event(session, head, mmap_size, buf);
1889 if (mmaps[map_idx]) {
1890 munmap(mmaps[map_idx], mmap_size);
1891 mmaps[map_idx] = NULL;
1894 page_offset = page_size * (head / page_size);
1895 file_offset += page_offset;
1896 head -= page_offset;
1900 size = event->header.size;
1902 if (size < sizeof(struct perf_event_header) ||
1903 (skip = perf_session__process_event(session, event, file_pos)) < 0) {
1904 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1905 file_offset + head, event->header.size,
1906 event->header.type);
1917 ui_progress__update(&prog, size);
1922 if (file_pos < file_size)
1926 /* do the final flush for ordered samples */
1927 err = ordered_events__flush(oe, OE_FLUSH__FINAL);
1930 err = auxtrace__flush_events(session, tool);
1933 err = perf_session__flush_thread_stacks(session);
1935 ui_progress__finish();
1936 perf_session__warn_about_errors(session);
1938 * We may switching perf.data output, make ordered_events
1941 ordered_events__reinit(&session->ordered_events);
1942 auxtrace__free_events(session);
1943 session->one_mmap = false;
1947 int perf_session__process_events(struct perf_session *session)
1949 u64 size = perf_data__size(session->data);
1952 if (perf_session__register_idle_thread(session) < 0)
1955 if (!perf_data__is_pipe(session->data))
1956 err = __perf_session__process_events(session,
1957 session->header.data_offset,
1958 session->header.data_size, size);
1960 err = __perf_session__process_pipe_events(session);
1965 bool perf_session__has_traces(struct perf_session *session, const char *msg)
1967 struct perf_evsel *evsel;
1969 evlist__for_each_entry(session->evlist, evsel) {
1970 if (evsel->attr.type == PERF_TYPE_TRACEPOINT)
1974 pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg);
1978 int maps__set_kallsyms_ref_reloc_sym(struct map **maps,
1979 const char *symbol_name, u64 addr)
1983 struct ref_reloc_sym *ref;
1985 ref = zalloc(sizeof(struct ref_reloc_sym));
1989 ref->name = strdup(symbol_name);
1990 if (ref->name == NULL) {
1995 bracket = strchr(ref->name, ']');
2001 for (i = 0; i < MAP__NR_TYPES; ++i) {
2002 struct kmap *kmap = map__kmap(maps[i]);
2006 kmap->ref_reloc_sym = ref;
2012 size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp)
2014 return machines__fprintf_dsos(&session->machines, fp);
2017 size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp,
2018 bool (skip)(struct dso *dso, int parm), int parm)
2020 return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm);
2023 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp)
2026 const char *msg = "";
2028 if (perf_header__has_feat(&session->header, HEADER_AUXTRACE))
2029 msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)";
2031 ret = fprintf(fp, "\nAggregated stats:%s\n", msg);
2033 ret += events_stats__fprintf(&session->evlist->stats, fp);
2037 size_t perf_session__fprintf(struct perf_session *session, FILE *fp)
2040 * FIXME: Here we have to actually print all the machines in this
2041 * session, not just the host...
2043 return machine__fprintf(&session->machines.host, fp);
2046 struct perf_evsel *perf_session__find_first_evtype(struct perf_session *session,
2049 struct perf_evsel *pos;
2051 evlist__for_each_entry(session->evlist, pos) {
2052 if (pos->attr.type == type)
2058 int perf_session__cpu_bitmap(struct perf_session *session,
2059 const char *cpu_list, unsigned long *cpu_bitmap)
2062 struct cpu_map *map;
2064 for (i = 0; i < PERF_TYPE_MAX; ++i) {
2065 struct perf_evsel *evsel;
2067 evsel = perf_session__find_first_evtype(session, i);
2071 if (!(evsel->attr.sample_type & PERF_SAMPLE_CPU)) {
2072 pr_err("File does not contain CPU events. "
2073 "Remove -C option to proceed.\n");
2078 map = cpu_map__new(cpu_list);
2080 pr_err("Invalid cpu_list\n");
2084 for (i = 0; i < map->nr; i++) {
2085 int cpu = map->map[i];
2087 if (cpu >= MAX_NR_CPUS) {
2088 pr_err("Requested CPU %d too large. "
2089 "Consider raising MAX_NR_CPUS\n", cpu);
2090 goto out_delete_map;
2093 set_bit(cpu, cpu_bitmap);
2103 void perf_session__fprintf_info(struct perf_session *session, FILE *fp,
2106 if (session == NULL || fp == NULL)
2109 fprintf(fp, "# ========\n");
2110 perf_header__fprintf_info(session, fp, full);
2111 fprintf(fp, "# ========\n#\n");
2115 int __perf_session__set_tracepoints_handlers(struct perf_session *session,
2116 const struct perf_evsel_str_handler *assocs,
2119 struct perf_evsel *evsel;
2123 for (i = 0; i < nr_assocs; i++) {
2125 * Adding a handler for an event not in the session,
2128 evsel = perf_evlist__find_tracepoint_by_name(session->evlist, assocs[i].name);
2133 if (evsel->handler != NULL)
2135 evsel->handler = assocs[i].handler;
2143 int perf_event__process_id_index(struct perf_tool *tool __maybe_unused,
2144 union perf_event *event,
2145 struct perf_session *session)
2147 struct perf_evlist *evlist = session->evlist;
2148 struct id_index_event *ie = &event->id_index;
2149 size_t i, nr, max_nr;
2151 max_nr = (ie->header.size - sizeof(struct id_index_event)) /
2152 sizeof(struct id_index_entry);
2158 fprintf(stdout, " nr: %zu\n", nr);
2160 for (i = 0; i < nr; i++) {
2161 struct id_index_entry *e = &ie->entries[i];
2162 struct perf_sample_id *sid;
2165 fprintf(stdout, " ... id: %"PRIu64, e->id);
2166 fprintf(stdout, " idx: %"PRIu64, e->idx);
2167 fprintf(stdout, " cpu: %"PRId64, e->cpu);
2168 fprintf(stdout, " tid: %"PRId64"\n", e->tid);
2171 sid = perf_evlist__id2sid(evlist, e->id);
2181 int perf_event__synthesize_id_index(struct perf_tool *tool,
2182 perf_event__handler_t process,
2183 struct perf_evlist *evlist,
2184 struct machine *machine)
2186 union perf_event *ev;
2187 struct perf_evsel *evsel;
2188 size_t nr = 0, i = 0, sz, max_nr, n;
2191 pr_debug2("Synthesizing id index\n");
2193 max_nr = (UINT16_MAX - sizeof(struct id_index_event)) /
2194 sizeof(struct id_index_entry);
2196 evlist__for_each_entry(evlist, evsel)
2199 n = nr > max_nr ? max_nr : nr;
2200 sz = sizeof(struct id_index_event) + n * sizeof(struct id_index_entry);
2205 ev->id_index.header.type = PERF_RECORD_ID_INDEX;
2206 ev->id_index.header.size = sz;
2207 ev->id_index.nr = n;
2209 evlist__for_each_entry(evlist, evsel) {
2212 for (j = 0; j < evsel->ids; j++) {
2213 struct id_index_entry *e;
2214 struct perf_sample_id *sid;
2217 err = process(tool, ev, NULL, machine);
2224 e = &ev->id_index.entries[i++];
2226 e->id = evsel->id[j];
2228 sid = perf_evlist__id2sid(evlist, e->id);
2240 sz = sizeof(struct id_index_event) + nr * sizeof(struct id_index_entry);
2241 ev->id_index.header.size = sz;
2242 ev->id_index.nr = nr;
2244 err = process(tool, ev, NULL, machine);