1 #include <linux/kernel.h>
2 #include <traceevent/event-parse.h>
16 #include "perf_regs.h"
19 #include "thread-stack.h"
22 static int perf_session__deliver_event(struct perf_session *session,
23 union perf_event *event,
24 struct perf_sample *sample,
25 struct perf_tool *tool,
28 static int perf_session__open(struct perf_session *session)
30 struct perf_data_file *file = session->file;
32 if (perf_session__read_header(session) < 0) {
33 pr_err("incompatible file format (rerun with -v to learn more)\n");
37 if (perf_data_file__is_pipe(file))
40 if (perf_header__has_feat(&session->header, HEADER_STAT))
43 if (!perf_evlist__valid_sample_type(session->evlist)) {
44 pr_err("non matching sample_type\n");
48 if (!perf_evlist__valid_sample_id_all(session->evlist)) {
49 pr_err("non matching sample_id_all\n");
53 if (!perf_evlist__valid_read_format(session->evlist)) {
54 pr_err("non matching read_format\n");
61 void perf_session__set_id_hdr_size(struct perf_session *session)
63 u16 id_hdr_size = perf_evlist__id_hdr_size(session->evlist);
65 machines__set_id_hdr_size(&session->machines, id_hdr_size);
68 int perf_session__create_kernel_maps(struct perf_session *session)
70 int ret = machine__create_kernel_maps(&session->machines.host);
73 ret = machines__create_guest_kernel_maps(&session->machines);
77 static void perf_session__destroy_kernel_maps(struct perf_session *session)
79 machines__destroy_kernel_maps(&session->machines);
82 static bool perf_session__has_comm_exec(struct perf_session *session)
84 struct perf_evsel *evsel;
86 evlist__for_each(session->evlist, evsel) {
87 if (evsel->attr.comm_exec)
94 static void perf_session__set_comm_exec(struct perf_session *session)
96 bool comm_exec = perf_session__has_comm_exec(session);
98 machines__set_comm_exec(&session->machines, comm_exec);
101 static int ordered_events__deliver_event(struct ordered_events *oe,
102 struct ordered_event *event)
104 struct perf_sample sample;
105 struct perf_session *session = container_of(oe, struct perf_session,
107 int ret = perf_evlist__parse_sample(session->evlist, event->event, &sample);
110 pr_err("Can't parse sample, err = %d\n", ret);
114 return perf_session__deliver_event(session, event->event, &sample,
115 session->tool, event->file_offset);
118 struct perf_session *perf_session__new(struct perf_data_file *file,
119 bool repipe, struct perf_tool *tool)
121 struct perf_session *session = zalloc(sizeof(*session));
126 session->repipe = repipe;
127 session->tool = tool;
128 INIT_LIST_HEAD(&session->auxtrace_index);
129 machines__init(&session->machines);
130 ordered_events__init(&session->ordered_events, ordered_events__deliver_event);
133 if (perf_data_file__open(file))
136 session->file = file;
138 if (perf_data_file__is_read(file)) {
139 if (perf_session__open(session) < 0)
142 perf_session__set_id_hdr_size(session);
143 perf_session__set_comm_exec(session);
146 session->machines.host.env = &perf_env;
149 if (!file || perf_data_file__is_write(file)) {
151 * In O_RDONLY mode this will be performed when reading the
152 * kernel MMAP event, in perf_event__process_mmap().
154 if (perf_session__create_kernel_maps(session) < 0)
155 pr_warning("Cannot read kernel map\n");
158 if (tool && tool->ordering_requires_timestamps &&
159 tool->ordered_events && !perf_evlist__sample_id_all(session->evlist)) {
160 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n");
161 tool->ordered_events = false;
167 perf_data_file__close(file);
169 perf_session__delete(session);
174 static void perf_session__delete_threads(struct perf_session *session)
176 machine__delete_threads(&session->machines.host);
179 void perf_session__delete(struct perf_session *session)
181 auxtrace__free(session);
182 auxtrace_index__free(&session->auxtrace_index);
183 perf_session__destroy_kernel_maps(session);
184 perf_session__delete_threads(session);
185 perf_env__exit(&session->header.env);
186 machines__exit(&session->machines);
188 perf_data_file__close(session->file);
192 static int process_event_synth_tracing_data_stub(struct perf_tool *tool
194 union perf_event *event
196 struct perf_session *session
199 dump_printf(": unhandled!\n");
203 static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused,
204 union perf_event *event __maybe_unused,
205 struct perf_evlist **pevlist
208 dump_printf(": unhandled!\n");
212 static int process_event_synth_event_update_stub(struct perf_tool *tool __maybe_unused,
213 union perf_event *event __maybe_unused,
214 struct perf_evlist **pevlist
218 perf_event__fprintf_event_update(event, stdout);
220 dump_printf(": unhandled!\n");
224 static int process_event_sample_stub(struct perf_tool *tool __maybe_unused,
225 union perf_event *event __maybe_unused,
226 struct perf_sample *sample __maybe_unused,
227 struct perf_evsel *evsel __maybe_unused,
228 struct machine *machine __maybe_unused)
230 dump_printf(": unhandled!\n");
234 static int process_event_stub(struct perf_tool *tool __maybe_unused,
235 union perf_event *event __maybe_unused,
236 struct perf_sample *sample __maybe_unused,
237 struct machine *machine __maybe_unused)
239 dump_printf(": unhandled!\n");
243 static int process_finished_round_stub(struct perf_tool *tool __maybe_unused,
244 union perf_event *event __maybe_unused,
245 struct ordered_events *oe __maybe_unused)
247 dump_printf(": unhandled!\n");
251 static int process_finished_round(struct perf_tool *tool,
252 union perf_event *event,
253 struct ordered_events *oe);
255 static int skipn(int fd, off_t n)
261 ret = read(fd, buf, min(n, (off_t)sizeof(buf)));
270 static s64 process_event_auxtrace_stub(struct perf_tool *tool __maybe_unused,
271 union perf_event *event,
272 struct perf_session *session
275 dump_printf(": unhandled!\n");
276 if (perf_data_file__is_pipe(session->file))
277 skipn(perf_data_file__fd(session->file), event->auxtrace.size);
278 return event->auxtrace.size;
281 static int process_event_op2_stub(struct perf_tool *tool __maybe_unused,
282 union perf_event *event __maybe_unused,
283 struct perf_session *session __maybe_unused)
285 dump_printf(": unhandled!\n");
291 int process_event_thread_map_stub(struct perf_tool *tool __maybe_unused,
292 union perf_event *event __maybe_unused,
293 struct perf_session *session __maybe_unused)
296 perf_event__fprintf_thread_map(event, stdout);
298 dump_printf(": unhandled!\n");
303 int process_event_cpu_map_stub(struct perf_tool *tool __maybe_unused,
304 union perf_event *event __maybe_unused,
305 struct perf_session *session __maybe_unused)
308 perf_event__fprintf_cpu_map(event, stdout);
310 dump_printf(": unhandled!\n");
315 int process_event_stat_config_stub(struct perf_tool *tool __maybe_unused,
316 union perf_event *event __maybe_unused,
317 struct perf_session *session __maybe_unused)
320 perf_event__fprintf_stat_config(event, stdout);
322 dump_printf(": unhandled!\n");
326 static int process_stat_stub(struct perf_tool *tool __maybe_unused,
327 union perf_event *event __maybe_unused,
328 struct perf_session *perf_session
332 perf_event__fprintf_stat(event, stdout);
334 dump_printf(": unhandled!\n");
338 static int process_stat_round_stub(struct perf_tool *tool __maybe_unused,
339 union perf_event *event __maybe_unused,
340 struct perf_session *perf_session
344 perf_event__fprintf_stat_round(event, stdout);
346 dump_printf(": unhandled!\n");
350 void perf_tool__fill_defaults(struct perf_tool *tool)
352 if (tool->sample == NULL)
353 tool->sample = process_event_sample_stub;
354 if (tool->mmap == NULL)
355 tool->mmap = process_event_stub;
356 if (tool->mmap2 == NULL)
357 tool->mmap2 = process_event_stub;
358 if (tool->comm == NULL)
359 tool->comm = process_event_stub;
360 if (tool->fork == NULL)
361 tool->fork = process_event_stub;
362 if (tool->exit == NULL)
363 tool->exit = process_event_stub;
364 if (tool->lost == NULL)
365 tool->lost = perf_event__process_lost;
366 if (tool->lost_samples == NULL)
367 tool->lost_samples = perf_event__process_lost_samples;
368 if (tool->aux == NULL)
369 tool->aux = perf_event__process_aux;
370 if (tool->itrace_start == NULL)
371 tool->itrace_start = perf_event__process_itrace_start;
372 if (tool->context_switch == NULL)
373 tool->context_switch = perf_event__process_switch;
374 if (tool->read == NULL)
375 tool->read = process_event_sample_stub;
376 if (tool->throttle == NULL)
377 tool->throttle = process_event_stub;
378 if (tool->unthrottle == NULL)
379 tool->unthrottle = process_event_stub;
380 if (tool->attr == NULL)
381 tool->attr = process_event_synth_attr_stub;
382 if (tool->event_update == NULL)
383 tool->event_update = process_event_synth_event_update_stub;
384 if (tool->tracing_data == NULL)
385 tool->tracing_data = process_event_synth_tracing_data_stub;
386 if (tool->build_id == NULL)
387 tool->build_id = process_event_op2_stub;
388 if (tool->finished_round == NULL) {
389 if (tool->ordered_events)
390 tool->finished_round = process_finished_round;
392 tool->finished_round = process_finished_round_stub;
394 if (tool->id_index == NULL)
395 tool->id_index = process_event_op2_stub;
396 if (tool->auxtrace_info == NULL)
397 tool->auxtrace_info = process_event_op2_stub;
398 if (tool->auxtrace == NULL)
399 tool->auxtrace = process_event_auxtrace_stub;
400 if (tool->auxtrace_error == NULL)
401 tool->auxtrace_error = process_event_op2_stub;
402 if (tool->thread_map == NULL)
403 tool->thread_map = process_event_thread_map_stub;
404 if (tool->cpu_map == NULL)
405 tool->cpu_map = process_event_cpu_map_stub;
406 if (tool->stat_config == NULL)
407 tool->stat_config = process_event_stat_config_stub;
408 if (tool->stat == NULL)
409 tool->stat = process_stat_stub;
410 if (tool->stat_round == NULL)
411 tool->stat_round = process_stat_round_stub;
412 if (tool->time_conv == NULL)
413 tool->time_conv = process_event_op2_stub;
416 static void swap_sample_id_all(union perf_event *event, void *data)
418 void *end = (void *) event + event->header.size;
419 int size = end - data;
421 BUG_ON(size % sizeof(u64));
422 mem_bswap_64(data, size);
425 static void perf_event__all64_swap(union perf_event *event,
426 bool sample_id_all __maybe_unused)
428 struct perf_event_header *hdr = &event->header;
429 mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr));
432 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all)
434 event->comm.pid = bswap_32(event->comm.pid);
435 event->comm.tid = bswap_32(event->comm.tid);
438 void *data = &event->comm.comm;
440 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
441 swap_sample_id_all(event, data);
445 static void perf_event__mmap_swap(union perf_event *event,
448 event->mmap.pid = bswap_32(event->mmap.pid);
449 event->mmap.tid = bswap_32(event->mmap.tid);
450 event->mmap.start = bswap_64(event->mmap.start);
451 event->mmap.len = bswap_64(event->mmap.len);
452 event->mmap.pgoff = bswap_64(event->mmap.pgoff);
455 void *data = &event->mmap.filename;
457 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
458 swap_sample_id_all(event, data);
462 static void perf_event__mmap2_swap(union perf_event *event,
465 event->mmap2.pid = bswap_32(event->mmap2.pid);
466 event->mmap2.tid = bswap_32(event->mmap2.tid);
467 event->mmap2.start = bswap_64(event->mmap2.start);
468 event->mmap2.len = bswap_64(event->mmap2.len);
469 event->mmap2.pgoff = bswap_64(event->mmap2.pgoff);
470 event->mmap2.maj = bswap_32(event->mmap2.maj);
471 event->mmap2.min = bswap_32(event->mmap2.min);
472 event->mmap2.ino = bswap_64(event->mmap2.ino);
475 void *data = &event->mmap2.filename;
477 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
478 swap_sample_id_all(event, data);
481 static void perf_event__task_swap(union perf_event *event, bool sample_id_all)
483 event->fork.pid = bswap_32(event->fork.pid);
484 event->fork.tid = bswap_32(event->fork.tid);
485 event->fork.ppid = bswap_32(event->fork.ppid);
486 event->fork.ptid = bswap_32(event->fork.ptid);
487 event->fork.time = bswap_64(event->fork.time);
490 swap_sample_id_all(event, &event->fork + 1);
493 static void perf_event__read_swap(union perf_event *event, bool sample_id_all)
495 event->read.pid = bswap_32(event->read.pid);
496 event->read.tid = bswap_32(event->read.tid);
497 event->read.value = bswap_64(event->read.value);
498 event->read.time_enabled = bswap_64(event->read.time_enabled);
499 event->read.time_running = bswap_64(event->read.time_running);
500 event->read.id = bswap_64(event->read.id);
503 swap_sample_id_all(event, &event->read + 1);
506 static void perf_event__aux_swap(union perf_event *event, bool sample_id_all)
508 event->aux.aux_offset = bswap_64(event->aux.aux_offset);
509 event->aux.aux_size = bswap_64(event->aux.aux_size);
510 event->aux.flags = bswap_64(event->aux.flags);
513 swap_sample_id_all(event, &event->aux + 1);
516 static void perf_event__itrace_start_swap(union perf_event *event,
519 event->itrace_start.pid = bswap_32(event->itrace_start.pid);
520 event->itrace_start.tid = bswap_32(event->itrace_start.tid);
523 swap_sample_id_all(event, &event->itrace_start + 1);
526 static void perf_event__switch_swap(union perf_event *event, bool sample_id_all)
528 if (event->header.type == PERF_RECORD_SWITCH_CPU_WIDE) {
529 event->context_switch.next_prev_pid =
530 bswap_32(event->context_switch.next_prev_pid);
531 event->context_switch.next_prev_tid =
532 bswap_32(event->context_switch.next_prev_tid);
536 swap_sample_id_all(event, &event->context_switch + 1);
539 static void perf_event__throttle_swap(union perf_event *event,
542 event->throttle.time = bswap_64(event->throttle.time);
543 event->throttle.id = bswap_64(event->throttle.id);
544 event->throttle.stream_id = bswap_64(event->throttle.stream_id);
547 swap_sample_id_all(event, &event->throttle + 1);
550 static u8 revbyte(u8 b)
552 int rev = (b >> 4) | ((b & 0xf) << 4);
553 rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2);
554 rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1);
559 * XXX this is hack in attempt to carry flags bitfield
560 * through endian village. ABI says:
562 * Bit-fields are allocated from right to left (least to most significant)
563 * on little-endian implementations and from left to right (most to least
564 * significant) on big-endian implementations.
566 * The above seems to be byte specific, so we need to reverse each
567 * byte of the bitfield. 'Internet' also says this might be implementation
568 * specific and we probably need proper fix and carry perf_event_attr
569 * bitfield flags in separate data file FEAT_ section. Thought this seems
572 static void swap_bitfield(u8 *p, unsigned len)
576 for (i = 0; i < len; i++) {
582 /* exported for swapping attributes in file header */
583 void perf_event__attr_swap(struct perf_event_attr *attr)
585 attr->type = bswap_32(attr->type);
586 attr->size = bswap_32(attr->size);
588 #define bswap_safe(f, n) \
589 (attr->size > (offsetof(struct perf_event_attr, f) + \
590 sizeof(attr->f) * (n)))
591 #define bswap_field(f, sz) \
593 if (bswap_safe(f, 0)) \
594 attr->f = bswap_##sz(attr->f); \
596 #define bswap_field_32(f) bswap_field(f, 32)
597 #define bswap_field_64(f) bswap_field(f, 64)
599 bswap_field_64(config);
600 bswap_field_64(sample_period);
601 bswap_field_64(sample_type);
602 bswap_field_64(read_format);
603 bswap_field_32(wakeup_events);
604 bswap_field_32(bp_type);
605 bswap_field_64(bp_addr);
606 bswap_field_64(bp_len);
607 bswap_field_64(branch_sample_type);
608 bswap_field_64(sample_regs_user);
609 bswap_field_32(sample_stack_user);
610 bswap_field_32(aux_watermark);
613 * After read_format are bitfields. Check read_format because
614 * we are unable to use offsetof on bitfield.
616 if (bswap_safe(read_format, 1))
617 swap_bitfield((u8 *) (&attr->read_format + 1),
619 #undef bswap_field_64
620 #undef bswap_field_32
625 static void perf_event__hdr_attr_swap(union perf_event *event,
626 bool sample_id_all __maybe_unused)
630 perf_event__attr_swap(&event->attr.attr);
632 size = event->header.size;
633 size -= (void *)&event->attr.id - (void *)event;
634 mem_bswap_64(event->attr.id, size);
637 static void perf_event__event_update_swap(union perf_event *event,
638 bool sample_id_all __maybe_unused)
640 event->event_update.type = bswap_64(event->event_update.type);
641 event->event_update.id = bswap_64(event->event_update.id);
644 static void perf_event__event_type_swap(union perf_event *event,
645 bool sample_id_all __maybe_unused)
647 event->event_type.event_type.event_id =
648 bswap_64(event->event_type.event_type.event_id);
651 static void perf_event__tracing_data_swap(union perf_event *event,
652 bool sample_id_all __maybe_unused)
654 event->tracing_data.size = bswap_32(event->tracing_data.size);
657 static void perf_event__auxtrace_info_swap(union perf_event *event,
658 bool sample_id_all __maybe_unused)
662 event->auxtrace_info.type = bswap_32(event->auxtrace_info.type);
664 size = event->header.size;
665 size -= (void *)&event->auxtrace_info.priv - (void *)event;
666 mem_bswap_64(event->auxtrace_info.priv, size);
669 static void perf_event__auxtrace_swap(union perf_event *event,
670 bool sample_id_all __maybe_unused)
672 event->auxtrace.size = bswap_64(event->auxtrace.size);
673 event->auxtrace.offset = bswap_64(event->auxtrace.offset);
674 event->auxtrace.reference = bswap_64(event->auxtrace.reference);
675 event->auxtrace.idx = bswap_32(event->auxtrace.idx);
676 event->auxtrace.tid = bswap_32(event->auxtrace.tid);
677 event->auxtrace.cpu = bswap_32(event->auxtrace.cpu);
680 static void perf_event__auxtrace_error_swap(union perf_event *event,
681 bool sample_id_all __maybe_unused)
683 event->auxtrace_error.type = bswap_32(event->auxtrace_error.type);
684 event->auxtrace_error.code = bswap_32(event->auxtrace_error.code);
685 event->auxtrace_error.cpu = bswap_32(event->auxtrace_error.cpu);
686 event->auxtrace_error.pid = bswap_32(event->auxtrace_error.pid);
687 event->auxtrace_error.tid = bswap_32(event->auxtrace_error.tid);
688 event->auxtrace_error.ip = bswap_64(event->auxtrace_error.ip);
691 static void perf_event__thread_map_swap(union perf_event *event,
692 bool sample_id_all __maybe_unused)
696 event->thread_map.nr = bswap_64(event->thread_map.nr);
698 for (i = 0; i < event->thread_map.nr; i++)
699 event->thread_map.entries[i].pid = bswap_64(event->thread_map.entries[i].pid);
702 static void perf_event__cpu_map_swap(union perf_event *event,
703 bool sample_id_all __maybe_unused)
705 struct cpu_map_data *data = &event->cpu_map.data;
706 struct cpu_map_entries *cpus;
707 struct cpu_map_mask *mask;
710 data->type = bswap_64(data->type);
712 switch (data->type) {
713 case PERF_CPU_MAP__CPUS:
714 cpus = (struct cpu_map_entries *)data->data;
716 cpus->nr = bswap_16(cpus->nr);
718 for (i = 0; i < cpus->nr; i++)
719 cpus->cpu[i] = bswap_16(cpus->cpu[i]);
721 case PERF_CPU_MAP__MASK:
722 mask = (struct cpu_map_mask *) data->data;
724 mask->nr = bswap_16(mask->nr);
725 mask->long_size = bswap_16(mask->long_size);
727 switch (mask->long_size) {
728 case 4: mem_bswap_32(&mask->mask, mask->nr); break;
729 case 8: mem_bswap_64(&mask->mask, mask->nr); break;
731 pr_err("cpu_map swap: unsupported long size\n");
738 static void perf_event__stat_config_swap(union perf_event *event,
739 bool sample_id_all __maybe_unused)
743 size = event->stat_config.nr * sizeof(event->stat_config.data[0]);
744 size += 1; /* nr item itself */
745 mem_bswap_64(&event->stat_config.nr, size);
748 static void perf_event__stat_swap(union perf_event *event,
749 bool sample_id_all __maybe_unused)
751 event->stat.id = bswap_64(event->stat.id);
752 event->stat.thread = bswap_32(event->stat.thread);
753 event->stat.cpu = bswap_32(event->stat.cpu);
754 event->stat.val = bswap_64(event->stat.val);
755 event->stat.ena = bswap_64(event->stat.ena);
756 event->stat.run = bswap_64(event->stat.run);
759 static void perf_event__stat_round_swap(union perf_event *event,
760 bool sample_id_all __maybe_unused)
762 event->stat_round.type = bswap_64(event->stat_round.type);
763 event->stat_round.time = bswap_64(event->stat_round.time);
766 typedef void (*perf_event__swap_op)(union perf_event *event,
769 static perf_event__swap_op perf_event__swap_ops[] = {
770 [PERF_RECORD_MMAP] = perf_event__mmap_swap,
771 [PERF_RECORD_MMAP2] = perf_event__mmap2_swap,
772 [PERF_RECORD_COMM] = perf_event__comm_swap,
773 [PERF_RECORD_FORK] = perf_event__task_swap,
774 [PERF_RECORD_EXIT] = perf_event__task_swap,
775 [PERF_RECORD_LOST] = perf_event__all64_swap,
776 [PERF_RECORD_READ] = perf_event__read_swap,
777 [PERF_RECORD_THROTTLE] = perf_event__throttle_swap,
778 [PERF_RECORD_UNTHROTTLE] = perf_event__throttle_swap,
779 [PERF_RECORD_SAMPLE] = perf_event__all64_swap,
780 [PERF_RECORD_AUX] = perf_event__aux_swap,
781 [PERF_RECORD_ITRACE_START] = perf_event__itrace_start_swap,
782 [PERF_RECORD_LOST_SAMPLES] = perf_event__all64_swap,
783 [PERF_RECORD_SWITCH] = perf_event__switch_swap,
784 [PERF_RECORD_SWITCH_CPU_WIDE] = perf_event__switch_swap,
785 [PERF_RECORD_HEADER_ATTR] = perf_event__hdr_attr_swap,
786 [PERF_RECORD_HEADER_EVENT_TYPE] = perf_event__event_type_swap,
787 [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
788 [PERF_RECORD_HEADER_BUILD_ID] = NULL,
789 [PERF_RECORD_ID_INDEX] = perf_event__all64_swap,
790 [PERF_RECORD_AUXTRACE_INFO] = perf_event__auxtrace_info_swap,
791 [PERF_RECORD_AUXTRACE] = perf_event__auxtrace_swap,
792 [PERF_RECORD_AUXTRACE_ERROR] = perf_event__auxtrace_error_swap,
793 [PERF_RECORD_THREAD_MAP] = perf_event__thread_map_swap,
794 [PERF_RECORD_CPU_MAP] = perf_event__cpu_map_swap,
795 [PERF_RECORD_STAT_CONFIG] = perf_event__stat_config_swap,
796 [PERF_RECORD_STAT] = perf_event__stat_swap,
797 [PERF_RECORD_STAT_ROUND] = perf_event__stat_round_swap,
798 [PERF_RECORD_EVENT_UPDATE] = perf_event__event_update_swap,
799 [PERF_RECORD_TIME_CONV] = perf_event__all64_swap,
800 [PERF_RECORD_HEADER_MAX] = NULL,
804 * When perf record finishes a pass on every buffers, it records this pseudo
806 * We record the max timestamp t found in the pass n.
807 * Assuming these timestamps are monotonic across cpus, we know that if
808 * a buffer still has events with timestamps below t, they will be all
809 * available and then read in the pass n + 1.
810 * Hence when we start to read the pass n + 2, we can safely flush every
811 * events with timestamps below t.
813 * ============ PASS n =================
816 * cnt1 timestamps | cnt2 timestamps
819 * - | 4 <--- max recorded
821 * ============ PASS n + 1 ==============
824 * cnt1 timestamps | cnt2 timestamps
827 * 5 | 7 <---- max recorded
829 * Flush every events below timestamp 4
831 * ============ PASS n + 2 ==============
834 * cnt1 timestamps | cnt2 timestamps
839 * Flush every events below timestamp 7
842 static int process_finished_round(struct perf_tool *tool __maybe_unused,
843 union perf_event *event __maybe_unused,
844 struct ordered_events *oe)
847 fprintf(stdout, "\n");
848 return ordered_events__flush(oe, OE_FLUSH__ROUND);
851 int perf_session__queue_event(struct perf_session *s, union perf_event *event,
852 struct perf_sample *sample, u64 file_offset)
854 return ordered_events__queue(&s->ordered_events, event, sample, file_offset);
857 static void callchain__lbr_callstack_printf(struct perf_sample *sample)
859 struct ip_callchain *callchain = sample->callchain;
860 struct branch_stack *lbr_stack = sample->branch_stack;
861 u64 kernel_callchain_nr = callchain->nr;
864 for (i = 0; i < kernel_callchain_nr; i++) {
865 if (callchain->ips[i] == PERF_CONTEXT_USER)
869 if ((i != kernel_callchain_nr) && lbr_stack->nr) {
872 * LBR callstack can only get user call chain,
873 * i is kernel call chain number,
874 * 1 is PERF_CONTEXT_USER.
876 * The user call chain is stored in LBR registers.
877 * LBR are pair registers. The caller is stored
878 * in "from" register, while the callee is stored
880 * For example, there is a call stack
881 * "A"->"B"->"C"->"D".
882 * The LBR registers will recorde like
883 * "C"->"D", "B"->"C", "A"->"B".
884 * So only the first "to" register and all "from"
885 * registers are needed to construct the whole stack.
887 total_nr = i + 1 + lbr_stack->nr + 1;
888 kernel_callchain_nr = i + 1;
890 printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr);
892 for (i = 0; i < kernel_callchain_nr; i++)
893 printf("..... %2d: %016" PRIx64 "\n",
894 i, callchain->ips[i]);
896 printf("..... %2d: %016" PRIx64 "\n",
897 (int)(kernel_callchain_nr), lbr_stack->entries[0].to);
898 for (i = 0; i < lbr_stack->nr; i++)
899 printf("..... %2d: %016" PRIx64 "\n",
900 (int)(i + kernel_callchain_nr + 1), lbr_stack->entries[i].from);
904 static void callchain__printf(struct perf_evsel *evsel,
905 struct perf_sample *sample)
908 struct ip_callchain *callchain = sample->callchain;
910 if (perf_evsel__has_branch_callstack(evsel))
911 callchain__lbr_callstack_printf(sample);
913 printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr);
915 for (i = 0; i < callchain->nr; i++)
916 printf("..... %2d: %016" PRIx64 "\n",
917 i, callchain->ips[i]);
920 static void branch_stack__printf(struct perf_sample *sample)
924 printf("... branch stack: nr:%" PRIu64 "\n", sample->branch_stack->nr);
926 for (i = 0; i < sample->branch_stack->nr; i++) {
927 struct branch_entry *e = &sample->branch_stack->entries[i];
929 printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x\n",
932 e->flags.mispred ? "M" : " ",
933 e->flags.predicted ? "P" : " ",
934 e->flags.abort ? "A" : " ",
935 e->flags.in_tx ? "T" : " ",
936 (unsigned)e->flags.reserved);
940 static void regs_dump__printf(u64 mask, u64 *regs)
944 for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) {
947 printf(".... %-5s 0x%" PRIx64 "\n",
948 perf_reg_name(rid), val);
952 static const char *regs_abi[] = {
953 [PERF_SAMPLE_REGS_ABI_NONE] = "none",
954 [PERF_SAMPLE_REGS_ABI_32] = "32-bit",
955 [PERF_SAMPLE_REGS_ABI_64] = "64-bit",
958 static inline const char *regs_dump_abi(struct regs_dump *d)
960 if (d->abi > PERF_SAMPLE_REGS_ABI_64)
963 return regs_abi[d->abi];
966 static void regs__printf(const char *type, struct regs_dump *regs)
968 u64 mask = regs->mask;
970 printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n",
973 regs_dump_abi(regs));
975 regs_dump__printf(mask, regs->regs);
978 static void regs_user__printf(struct perf_sample *sample)
980 struct regs_dump *user_regs = &sample->user_regs;
983 regs__printf("user", user_regs);
986 static void regs_intr__printf(struct perf_sample *sample)
988 struct regs_dump *intr_regs = &sample->intr_regs;
991 regs__printf("intr", intr_regs);
994 static void stack_user__printf(struct stack_dump *dump)
996 printf("... ustack: size %" PRIu64 ", offset 0x%x\n",
997 dump->size, dump->offset);
1000 static void perf_evlist__print_tstamp(struct perf_evlist *evlist,
1001 union perf_event *event,
1002 struct perf_sample *sample)
1004 u64 sample_type = __perf_evlist__combined_sample_type(evlist);
1006 if (event->header.type != PERF_RECORD_SAMPLE &&
1007 !perf_evlist__sample_id_all(evlist)) {
1008 fputs("-1 -1 ", stdout);
1012 if ((sample_type & PERF_SAMPLE_CPU))
1013 printf("%u ", sample->cpu);
1015 if (sample_type & PERF_SAMPLE_TIME)
1016 printf("%" PRIu64 " ", sample->time);
1019 static void sample_read__printf(struct perf_sample *sample, u64 read_format)
1021 printf("... sample_read:\n");
1023 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1024 printf("...... time enabled %016" PRIx64 "\n",
1025 sample->read.time_enabled);
1027 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1028 printf("...... time running %016" PRIx64 "\n",
1029 sample->read.time_running);
1031 if (read_format & PERF_FORMAT_GROUP) {
1034 printf(".... group nr %" PRIu64 "\n", sample->read.group.nr);
1036 for (i = 0; i < sample->read.group.nr; i++) {
1037 struct sample_read_value *value;
1039 value = &sample->read.group.values[i];
1040 printf("..... id %016" PRIx64
1041 ", value %016" PRIx64 "\n",
1042 value->id, value->value);
1045 printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n",
1046 sample->read.one.id, sample->read.one.value);
1049 static void dump_event(struct perf_evlist *evlist, union perf_event *event,
1050 u64 file_offset, struct perf_sample *sample)
1055 printf("\n%#" PRIx64 " [%#x]: event: %d\n",
1056 file_offset, event->header.size, event->header.type);
1061 perf_evlist__print_tstamp(evlist, event, sample);
1063 printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset,
1064 event->header.size, perf_event__name(event->header.type));
1067 static void dump_sample(struct perf_evsel *evsel, union perf_event *event,
1068 struct perf_sample *sample)
1075 printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n",
1076 event->header.misc, sample->pid, sample->tid, sample->ip,
1077 sample->period, sample->addr);
1079 sample_type = evsel->attr.sample_type;
1081 if (sample_type & PERF_SAMPLE_CALLCHAIN)
1082 callchain__printf(evsel, sample);
1084 if ((sample_type & PERF_SAMPLE_BRANCH_STACK) && !perf_evsel__has_branch_callstack(evsel))
1085 branch_stack__printf(sample);
1087 if (sample_type & PERF_SAMPLE_REGS_USER)
1088 regs_user__printf(sample);
1090 if (sample_type & PERF_SAMPLE_REGS_INTR)
1091 regs_intr__printf(sample);
1093 if (sample_type & PERF_SAMPLE_STACK_USER)
1094 stack_user__printf(&sample->user_stack);
1096 if (sample_type & PERF_SAMPLE_WEIGHT)
1097 printf("... weight: %" PRIu64 "\n", sample->weight);
1099 if (sample_type & PERF_SAMPLE_DATA_SRC)
1100 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src);
1102 if (sample_type & PERF_SAMPLE_TRANSACTION)
1103 printf("... transaction: %" PRIx64 "\n", sample->transaction);
1105 if (sample_type & PERF_SAMPLE_READ)
1106 sample_read__printf(sample, evsel->attr.read_format);
1109 static struct machine *machines__find_for_cpumode(struct machines *machines,
1110 union perf_event *event,
1111 struct perf_sample *sample)
1113 struct machine *machine;
1116 ((sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL) ||
1117 (sample->cpumode == PERF_RECORD_MISC_GUEST_USER))) {
1120 if (event->header.type == PERF_RECORD_MMAP
1121 || event->header.type == PERF_RECORD_MMAP2)
1122 pid = event->mmap.pid;
1126 machine = machines__find(machines, pid);
1128 machine = machines__findnew(machines, DEFAULT_GUEST_KERNEL_ID);
1132 return &machines->host;
1135 static int deliver_sample_value(struct perf_evlist *evlist,
1136 struct perf_tool *tool,
1137 union perf_event *event,
1138 struct perf_sample *sample,
1139 struct sample_read_value *v,
1140 struct machine *machine)
1142 struct perf_sample_id *sid = perf_evlist__id2sid(evlist, v->id);
1146 sample->period = v->value - sid->period;
1147 sid->period = v->value;
1150 if (!sid || sid->evsel == NULL) {
1151 ++evlist->stats.nr_unknown_id;
1155 return tool->sample(tool, event, sample, sid->evsel, machine);
1158 static int deliver_sample_group(struct perf_evlist *evlist,
1159 struct perf_tool *tool,
1160 union perf_event *event,
1161 struct perf_sample *sample,
1162 struct machine *machine)
1167 for (i = 0; i < sample->read.group.nr; i++) {
1168 ret = deliver_sample_value(evlist, tool, event, sample,
1169 &sample->read.group.values[i],
1179 perf_evlist__deliver_sample(struct perf_evlist *evlist,
1180 struct perf_tool *tool,
1181 union perf_event *event,
1182 struct perf_sample *sample,
1183 struct perf_evsel *evsel,
1184 struct machine *machine)
1186 /* We know evsel != NULL. */
1187 u64 sample_type = evsel->attr.sample_type;
1188 u64 read_format = evsel->attr.read_format;
1190 /* Standard sample delievery. */
1191 if (!(sample_type & PERF_SAMPLE_READ))
1192 return tool->sample(tool, event, sample, evsel, machine);
1194 /* For PERF_SAMPLE_READ we have either single or group mode. */
1195 if (read_format & PERF_FORMAT_GROUP)
1196 return deliver_sample_group(evlist, tool, event, sample,
1199 return deliver_sample_value(evlist, tool, event, sample,
1200 &sample->read.one, machine);
1203 static int machines__deliver_event(struct machines *machines,
1204 struct perf_evlist *evlist,
1205 union perf_event *event,
1206 struct perf_sample *sample,
1207 struct perf_tool *tool, u64 file_offset)
1209 struct perf_evsel *evsel;
1210 struct machine *machine;
1212 dump_event(evlist, event, file_offset, sample);
1214 evsel = perf_evlist__id2evsel(evlist, sample->id);
1216 machine = machines__find_for_cpumode(machines, event, sample);
1218 switch (event->header.type) {
1219 case PERF_RECORD_SAMPLE:
1220 if (evsel == NULL) {
1221 ++evlist->stats.nr_unknown_id;
1224 dump_sample(evsel, event, sample);
1225 if (machine == NULL) {
1226 ++evlist->stats.nr_unprocessable_samples;
1229 return perf_evlist__deliver_sample(evlist, tool, event, sample, evsel, machine);
1230 case PERF_RECORD_MMAP:
1231 return tool->mmap(tool, event, sample, machine);
1232 case PERF_RECORD_MMAP2:
1233 if (event->header.misc & PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT)
1234 ++evlist->stats.nr_proc_map_timeout;
1235 return tool->mmap2(tool, event, sample, machine);
1236 case PERF_RECORD_COMM:
1237 return tool->comm(tool, event, sample, machine);
1238 case PERF_RECORD_FORK:
1239 return tool->fork(tool, event, sample, machine);
1240 case PERF_RECORD_EXIT:
1241 return tool->exit(tool, event, sample, machine);
1242 case PERF_RECORD_LOST:
1243 if (tool->lost == perf_event__process_lost)
1244 evlist->stats.total_lost += event->lost.lost;
1245 return tool->lost(tool, event, sample, machine);
1246 case PERF_RECORD_LOST_SAMPLES:
1247 if (tool->lost_samples == perf_event__process_lost_samples)
1248 evlist->stats.total_lost_samples += event->lost_samples.lost;
1249 return tool->lost_samples(tool, event, sample, machine);
1250 case PERF_RECORD_READ:
1251 return tool->read(tool, event, sample, evsel, machine);
1252 case PERF_RECORD_THROTTLE:
1253 return tool->throttle(tool, event, sample, machine);
1254 case PERF_RECORD_UNTHROTTLE:
1255 return tool->unthrottle(tool, event, sample, machine);
1256 case PERF_RECORD_AUX:
1257 if (tool->aux == perf_event__process_aux &&
1258 (event->aux.flags & PERF_AUX_FLAG_TRUNCATED))
1259 evlist->stats.total_aux_lost += 1;
1260 return tool->aux(tool, event, sample, machine);
1261 case PERF_RECORD_ITRACE_START:
1262 return tool->itrace_start(tool, event, sample, machine);
1263 case PERF_RECORD_SWITCH:
1264 case PERF_RECORD_SWITCH_CPU_WIDE:
1265 return tool->context_switch(tool, event, sample, machine);
1267 ++evlist->stats.nr_unknown_events;
1272 static int perf_session__deliver_event(struct perf_session *session,
1273 union perf_event *event,
1274 struct perf_sample *sample,
1275 struct perf_tool *tool,
1280 ret = auxtrace__process_event(session, event, sample, tool);
1286 return machines__deliver_event(&session->machines, session->evlist,
1287 event, sample, tool, file_offset);
1290 static s64 perf_session__process_user_event(struct perf_session *session,
1291 union perf_event *event,
1294 struct ordered_events *oe = &session->ordered_events;
1295 struct perf_tool *tool = session->tool;
1296 int fd = perf_data_file__fd(session->file);
1299 dump_event(session->evlist, event, file_offset, NULL);
1301 /* These events are processed right away */
1302 switch (event->header.type) {
1303 case PERF_RECORD_HEADER_ATTR:
1304 err = tool->attr(tool, event, &session->evlist);
1306 perf_session__set_id_hdr_size(session);
1307 perf_session__set_comm_exec(session);
1310 case PERF_RECORD_EVENT_UPDATE:
1311 return tool->event_update(tool, event, &session->evlist);
1312 case PERF_RECORD_HEADER_EVENT_TYPE:
1314 * Depreceated, but we need to handle it for sake
1315 * of old data files create in pipe mode.
1318 case PERF_RECORD_HEADER_TRACING_DATA:
1319 /* setup for reading amidst mmap */
1320 lseek(fd, file_offset, SEEK_SET);
1321 return tool->tracing_data(tool, event, session);
1322 case PERF_RECORD_HEADER_BUILD_ID:
1323 return tool->build_id(tool, event, session);
1324 case PERF_RECORD_FINISHED_ROUND:
1325 return tool->finished_round(tool, event, oe);
1326 case PERF_RECORD_ID_INDEX:
1327 return tool->id_index(tool, event, session);
1328 case PERF_RECORD_AUXTRACE_INFO:
1329 return tool->auxtrace_info(tool, event, session);
1330 case PERF_RECORD_AUXTRACE:
1331 /* setup for reading amidst mmap */
1332 lseek(fd, file_offset + event->header.size, SEEK_SET);
1333 return tool->auxtrace(tool, event, session);
1334 case PERF_RECORD_AUXTRACE_ERROR:
1335 perf_session__auxtrace_error_inc(session, event);
1336 return tool->auxtrace_error(tool, event, session);
1337 case PERF_RECORD_THREAD_MAP:
1338 return tool->thread_map(tool, event, session);
1339 case PERF_RECORD_CPU_MAP:
1340 return tool->cpu_map(tool, event, session);
1341 case PERF_RECORD_STAT_CONFIG:
1342 return tool->stat_config(tool, event, session);
1343 case PERF_RECORD_STAT:
1344 return tool->stat(tool, event, session);
1345 case PERF_RECORD_STAT_ROUND:
1346 return tool->stat_round(tool, event, session);
1347 case PERF_RECORD_TIME_CONV:
1348 session->time_conv = event->time_conv;
1349 return tool->time_conv(tool, event, session);
1355 int perf_session__deliver_synth_event(struct perf_session *session,
1356 union perf_event *event,
1357 struct perf_sample *sample)
1359 struct perf_evlist *evlist = session->evlist;
1360 struct perf_tool *tool = session->tool;
1362 events_stats__inc(&evlist->stats, event->header.type);
1364 if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1365 return perf_session__process_user_event(session, event, 0);
1367 return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0);
1370 static void event_swap(union perf_event *event, bool sample_id_all)
1372 perf_event__swap_op swap;
1374 swap = perf_event__swap_ops[event->header.type];
1376 swap(event, sample_id_all);
1379 int perf_session__peek_event(struct perf_session *session, off_t file_offset,
1380 void *buf, size_t buf_sz,
1381 union perf_event **event_ptr,
1382 struct perf_sample *sample)
1384 union perf_event *event;
1385 size_t hdr_sz, rest;
1388 if (session->one_mmap && !session->header.needs_swap) {
1389 event = file_offset - session->one_mmap_offset +
1390 session->one_mmap_addr;
1391 goto out_parse_sample;
1394 if (perf_data_file__is_pipe(session->file))
1397 fd = perf_data_file__fd(session->file);
1398 hdr_sz = sizeof(struct perf_event_header);
1400 if (buf_sz < hdr_sz)
1403 if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 ||
1404 readn(fd, buf, hdr_sz) != (ssize_t)hdr_sz)
1407 event = (union perf_event *)buf;
1409 if (session->header.needs_swap)
1410 perf_event_header__bswap(&event->header);
1412 if (event->header.size < hdr_sz || event->header.size > buf_sz)
1415 rest = event->header.size - hdr_sz;
1417 if (readn(fd, buf, rest) != (ssize_t)rest)
1420 if (session->header.needs_swap)
1421 event_swap(event, perf_evlist__sample_id_all(session->evlist));
1425 if (sample && event->header.type < PERF_RECORD_USER_TYPE_START &&
1426 perf_evlist__parse_sample(session->evlist, event, sample))
1434 static s64 perf_session__process_event(struct perf_session *session,
1435 union perf_event *event, u64 file_offset)
1437 struct perf_evlist *evlist = session->evlist;
1438 struct perf_tool *tool = session->tool;
1439 struct perf_sample sample;
1442 if (session->header.needs_swap)
1443 event_swap(event, perf_evlist__sample_id_all(evlist));
1445 if (event->header.type >= PERF_RECORD_HEADER_MAX)
1448 events_stats__inc(&evlist->stats, event->header.type);
1450 if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1451 return perf_session__process_user_event(session, event, file_offset);
1454 * For all kernel events we get the sample data
1456 ret = perf_evlist__parse_sample(evlist, event, &sample);
1460 if (tool->ordered_events) {
1461 ret = perf_session__queue_event(session, event, &sample, file_offset);
1466 return perf_session__deliver_event(session, event, &sample, tool,
1470 void perf_event_header__bswap(struct perf_event_header *hdr)
1472 hdr->type = bswap_32(hdr->type);
1473 hdr->misc = bswap_16(hdr->misc);
1474 hdr->size = bswap_16(hdr->size);
1477 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid)
1479 return machine__findnew_thread(&session->machines.host, -1, pid);
1482 int perf_session__register_idle_thread(struct perf_session *session)
1484 struct thread *thread;
1487 thread = machine__findnew_thread(&session->machines.host, 0, 0);
1488 if (thread == NULL || thread__set_comm(thread, "swapper", 0)) {
1489 pr_err("problem inserting idle task.\n");
1493 /* machine__findnew_thread() got the thread, so put it */
1494 thread__put(thread);
1498 static void perf_session__warn_about_errors(const struct perf_session *session)
1500 const struct events_stats *stats = &session->evlist->stats;
1501 const struct ordered_events *oe = &session->ordered_events;
1503 if (session->tool->lost == perf_event__process_lost &&
1504 stats->nr_events[PERF_RECORD_LOST] != 0) {
1505 ui__warning("Processed %d events and lost %d chunks!\n\n"
1506 "Check IO/CPU overload!\n\n",
1507 stats->nr_events[0],
1508 stats->nr_events[PERF_RECORD_LOST]);
1511 if (session->tool->lost_samples == perf_event__process_lost_samples) {
1514 drop_rate = (double)stats->total_lost_samples /
1515 (double) (stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples);
1516 if (drop_rate > 0.05) {
1517 ui__warning("Processed %" PRIu64 " samples and lost %3.2f%% samples!\n\n",
1518 stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples,
1523 if (session->tool->aux == perf_event__process_aux &&
1524 stats->total_aux_lost != 0) {
1525 ui__warning("AUX data lost %" PRIu64 " times out of %u!\n\n",
1526 stats->total_aux_lost,
1527 stats->nr_events[PERF_RECORD_AUX]);
1530 if (stats->nr_unknown_events != 0) {
1531 ui__warning("Found %u unknown events!\n\n"
1532 "Is this an older tool processing a perf.data "
1533 "file generated by a more recent tool?\n\n"
1534 "If that is not the case, consider "
1535 "reporting to linux-kernel@vger.kernel.org.\n\n",
1536 stats->nr_unknown_events);
1539 if (stats->nr_unknown_id != 0) {
1540 ui__warning("%u samples with id not present in the header\n",
1541 stats->nr_unknown_id);
1544 if (stats->nr_invalid_chains != 0) {
1545 ui__warning("Found invalid callchains!\n\n"
1546 "%u out of %u events were discarded for this reason.\n\n"
1547 "Consider reporting to linux-kernel@vger.kernel.org.\n\n",
1548 stats->nr_invalid_chains,
1549 stats->nr_events[PERF_RECORD_SAMPLE]);
1552 if (stats->nr_unprocessable_samples != 0) {
1553 ui__warning("%u unprocessable samples recorded.\n"
1554 "Do you have a KVM guest running and not using 'perf kvm'?\n",
1555 stats->nr_unprocessable_samples);
1558 if (oe->nr_unordered_events != 0)
1559 ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events);
1561 events_stats__auxtrace_error_warn(stats);
1563 if (stats->nr_proc_map_timeout != 0) {
1564 ui__warning("%d map information files for pre-existing threads were\n"
1565 "not processed, if there are samples for addresses they\n"
1566 "will not be resolved, you may find out which are these\n"
1567 "threads by running with -v and redirecting the output\n"
1569 "The time limit to process proc map is too short?\n"
1570 "Increase it by --proc-map-timeout\n",
1571 stats->nr_proc_map_timeout);
1575 static int perf_session__flush_thread_stack(struct thread *thread,
1576 void *p __maybe_unused)
1578 return thread_stack__flush(thread);
1581 static int perf_session__flush_thread_stacks(struct perf_session *session)
1583 return machines__for_each_thread(&session->machines,
1584 perf_session__flush_thread_stack,
1588 volatile int session_done;
1590 static int __perf_session__process_pipe_events(struct perf_session *session)
1592 struct ordered_events *oe = &session->ordered_events;
1593 struct perf_tool *tool = session->tool;
1594 int fd = perf_data_file__fd(session->file);
1595 union perf_event *event;
1596 uint32_t size, cur_size = 0;
1603 perf_tool__fill_defaults(tool);
1606 cur_size = sizeof(union perf_event);
1608 buf = malloc(cur_size);
1613 err = readn(fd, event, sizeof(struct perf_event_header));
1618 pr_err("failed to read event header\n");
1622 if (session->header.needs_swap)
1623 perf_event_header__bswap(&event->header);
1625 size = event->header.size;
1626 if (size < sizeof(struct perf_event_header)) {
1627 pr_err("bad event header size\n");
1631 if (size > cur_size) {
1632 void *new = realloc(buf, size);
1634 pr_err("failed to allocate memory to read event\n");
1642 p += sizeof(struct perf_event_header);
1644 if (size - sizeof(struct perf_event_header)) {
1645 err = readn(fd, p, size - sizeof(struct perf_event_header));
1648 pr_err("unexpected end of event stream\n");
1652 pr_err("failed to read event data\n");
1657 if ((skip = perf_session__process_event(session, event, head)) < 0) {
1658 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1659 head, event->header.size, event->header.type);
1669 if (!session_done())
1672 /* do the final flush for ordered samples */
1673 err = ordered_events__flush(oe, OE_FLUSH__FINAL);
1676 err = auxtrace__flush_events(session, tool);
1679 err = perf_session__flush_thread_stacks(session);
1682 perf_session__warn_about_errors(session);
1683 ordered_events__free(&session->ordered_events);
1684 auxtrace__free_events(session);
1688 static union perf_event *
1689 fetch_mmaped_event(struct perf_session *session,
1690 u64 head, size_t mmap_size, char *buf)
1692 union perf_event *event;
1695 * Ensure we have enough space remaining to read
1696 * the size of the event in the headers.
1698 if (head + sizeof(event->header) > mmap_size)
1701 event = (union perf_event *)(buf + head);
1703 if (session->header.needs_swap)
1704 perf_event_header__bswap(&event->header);
1706 if (head + event->header.size > mmap_size) {
1707 /* We're not fetching the event so swap back again */
1708 if (session->header.needs_swap)
1709 perf_event_header__bswap(&event->header);
1717 * On 64bit we can mmap the data file in one go. No need for tiny mmap
1718 * slices. On 32bit we use 32MB.
1720 #if BITS_PER_LONG == 64
1721 #define MMAP_SIZE ULLONG_MAX
1724 #define MMAP_SIZE (32 * 1024 * 1024ULL)
1725 #define NUM_MMAPS 128
1728 static int __perf_session__process_events(struct perf_session *session,
1729 u64 data_offset, u64 data_size,
1732 struct ordered_events *oe = &session->ordered_events;
1733 struct perf_tool *tool = session->tool;
1734 int fd = perf_data_file__fd(session->file);
1735 u64 head, page_offset, file_offset, file_pos, size;
1736 int err, mmap_prot, mmap_flags, map_idx = 0;
1738 char *buf, *mmaps[NUM_MMAPS];
1739 union perf_event *event;
1740 struct ui_progress prog;
1743 perf_tool__fill_defaults(tool);
1745 page_offset = page_size * (data_offset / page_size);
1746 file_offset = page_offset;
1747 head = data_offset - page_offset;
1752 if (data_offset + data_size < file_size)
1753 file_size = data_offset + data_size;
1755 ui_progress__init(&prog, file_size, "Processing events...");
1757 mmap_size = MMAP_SIZE;
1758 if (mmap_size > file_size) {
1759 mmap_size = file_size;
1760 session->one_mmap = true;
1763 memset(mmaps, 0, sizeof(mmaps));
1765 mmap_prot = PROT_READ;
1766 mmap_flags = MAP_SHARED;
1768 if (session->header.needs_swap) {
1769 mmap_prot |= PROT_WRITE;
1770 mmap_flags = MAP_PRIVATE;
1773 buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, fd,
1775 if (buf == MAP_FAILED) {
1776 pr_err("failed to mmap file\n");
1780 mmaps[map_idx] = buf;
1781 map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1);
1782 file_pos = file_offset + head;
1783 if (session->one_mmap) {
1784 session->one_mmap_addr = buf;
1785 session->one_mmap_offset = file_offset;
1789 event = fetch_mmaped_event(session, head, mmap_size, buf);
1791 if (mmaps[map_idx]) {
1792 munmap(mmaps[map_idx], mmap_size);
1793 mmaps[map_idx] = NULL;
1796 page_offset = page_size * (head / page_size);
1797 file_offset += page_offset;
1798 head -= page_offset;
1802 size = event->header.size;
1804 if (size < sizeof(struct perf_event_header) ||
1805 (skip = perf_session__process_event(session, event, file_pos)) < 0) {
1806 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1807 file_offset + head, event->header.size,
1808 event->header.type);
1819 ui_progress__update(&prog, size);
1824 if (file_pos < file_size)
1828 /* do the final flush for ordered samples */
1829 err = ordered_events__flush(oe, OE_FLUSH__FINAL);
1832 err = auxtrace__flush_events(session, tool);
1835 err = perf_session__flush_thread_stacks(session);
1837 ui_progress__finish();
1838 perf_session__warn_about_errors(session);
1840 * We may switching perf.data output, make ordered_events
1843 ordered_events__reinit(&session->ordered_events);
1844 auxtrace__free_events(session);
1845 session->one_mmap = false;
1849 int perf_session__process_events(struct perf_session *session)
1851 u64 size = perf_data_file__size(session->file);
1854 if (perf_session__register_idle_thread(session) < 0)
1857 if (!perf_data_file__is_pipe(session->file))
1858 err = __perf_session__process_events(session,
1859 session->header.data_offset,
1860 session->header.data_size, size);
1862 err = __perf_session__process_pipe_events(session);
1867 bool perf_session__has_traces(struct perf_session *session, const char *msg)
1869 struct perf_evsel *evsel;
1871 evlist__for_each(session->evlist, evsel) {
1872 if (evsel->attr.type == PERF_TYPE_TRACEPOINT)
1876 pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg);
1880 int maps__set_kallsyms_ref_reloc_sym(struct map **maps,
1881 const char *symbol_name, u64 addr)
1885 struct ref_reloc_sym *ref;
1887 ref = zalloc(sizeof(struct ref_reloc_sym));
1891 ref->name = strdup(symbol_name);
1892 if (ref->name == NULL) {
1897 bracket = strchr(ref->name, ']');
1903 for (i = 0; i < MAP__NR_TYPES; ++i) {
1904 struct kmap *kmap = map__kmap(maps[i]);
1908 kmap->ref_reloc_sym = ref;
1914 size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp)
1916 return machines__fprintf_dsos(&session->machines, fp);
1919 size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp,
1920 bool (skip)(struct dso *dso, int parm), int parm)
1922 return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm);
1925 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp)
1928 const char *msg = "";
1930 if (perf_header__has_feat(&session->header, HEADER_AUXTRACE))
1931 msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)";
1933 ret = fprintf(fp, "\nAggregated stats:%s\n", msg);
1935 ret += events_stats__fprintf(&session->evlist->stats, fp);
1939 size_t perf_session__fprintf(struct perf_session *session, FILE *fp)
1942 * FIXME: Here we have to actually print all the machines in this
1943 * session, not just the host...
1945 return machine__fprintf(&session->machines.host, fp);
1948 struct perf_evsel *perf_session__find_first_evtype(struct perf_session *session,
1951 struct perf_evsel *pos;
1953 evlist__for_each(session->evlist, pos) {
1954 if (pos->attr.type == type)
1960 int perf_session__cpu_bitmap(struct perf_session *session,
1961 const char *cpu_list, unsigned long *cpu_bitmap)
1964 struct cpu_map *map;
1966 for (i = 0; i < PERF_TYPE_MAX; ++i) {
1967 struct perf_evsel *evsel;
1969 evsel = perf_session__find_first_evtype(session, i);
1973 if (!(evsel->attr.sample_type & PERF_SAMPLE_CPU)) {
1974 pr_err("File does not contain CPU events. "
1975 "Remove -c option to proceed.\n");
1980 map = cpu_map__new(cpu_list);
1982 pr_err("Invalid cpu_list\n");
1986 for (i = 0; i < map->nr; i++) {
1987 int cpu = map->map[i];
1989 if (cpu >= MAX_NR_CPUS) {
1990 pr_err("Requested CPU %d too large. "
1991 "Consider raising MAX_NR_CPUS\n", cpu);
1992 goto out_delete_map;
1995 set_bit(cpu, cpu_bitmap);
2005 void perf_session__fprintf_info(struct perf_session *session, FILE *fp,
2011 if (session == NULL || fp == NULL)
2014 fd = perf_data_file__fd(session->file);
2016 ret = fstat(fd, &st);
2020 fprintf(fp, "# ========\n");
2021 fprintf(fp, "# captured on: %s", ctime(&st.st_ctime));
2022 perf_header__fprintf_info(session, fp, full);
2023 fprintf(fp, "# ========\n#\n");
2027 int __perf_session__set_tracepoints_handlers(struct perf_session *session,
2028 const struct perf_evsel_str_handler *assocs,
2031 struct perf_evsel *evsel;
2035 for (i = 0; i < nr_assocs; i++) {
2037 * Adding a handler for an event not in the session,
2040 evsel = perf_evlist__find_tracepoint_by_name(session->evlist, assocs[i].name);
2045 if (evsel->handler != NULL)
2047 evsel->handler = assocs[i].handler;
2055 int perf_event__process_id_index(struct perf_tool *tool __maybe_unused,
2056 union perf_event *event,
2057 struct perf_session *session)
2059 struct perf_evlist *evlist = session->evlist;
2060 struct id_index_event *ie = &event->id_index;
2061 size_t i, nr, max_nr;
2063 max_nr = (ie->header.size - sizeof(struct id_index_event)) /
2064 sizeof(struct id_index_entry);
2070 fprintf(stdout, " nr: %zu\n", nr);
2072 for (i = 0; i < nr; i++) {
2073 struct id_index_entry *e = &ie->entries[i];
2074 struct perf_sample_id *sid;
2077 fprintf(stdout, " ... id: %"PRIu64, e->id);
2078 fprintf(stdout, " idx: %"PRIu64, e->idx);
2079 fprintf(stdout, " cpu: %"PRId64, e->cpu);
2080 fprintf(stdout, " tid: %"PRId64"\n", e->tid);
2083 sid = perf_evlist__id2sid(evlist, e->id);
2093 int perf_event__synthesize_id_index(struct perf_tool *tool,
2094 perf_event__handler_t process,
2095 struct perf_evlist *evlist,
2096 struct machine *machine)
2098 union perf_event *ev;
2099 struct perf_evsel *evsel;
2100 size_t nr = 0, i = 0, sz, max_nr, n;
2103 pr_debug2("Synthesizing id index\n");
2105 max_nr = (UINT16_MAX - sizeof(struct id_index_event)) /
2106 sizeof(struct id_index_entry);
2108 evlist__for_each(evlist, evsel)
2111 n = nr > max_nr ? max_nr : nr;
2112 sz = sizeof(struct id_index_event) + n * sizeof(struct id_index_entry);
2117 ev->id_index.header.type = PERF_RECORD_ID_INDEX;
2118 ev->id_index.header.size = sz;
2119 ev->id_index.nr = n;
2121 evlist__for_each(evlist, evsel) {
2124 for (j = 0; j < evsel->ids; j++) {
2125 struct id_index_entry *e;
2126 struct perf_sample_id *sid;
2129 err = process(tool, ev, NULL, machine);
2136 e = &ev->id_index.entries[i++];
2138 e->id = evsel->id[j];
2140 sid = perf_evlist__id2sid(evlist, e->id);
2152 sz = sizeof(struct id_index_event) + nr * sizeof(struct id_index_entry);
2153 ev->id_index.header.size = sz;
2154 ev->id_index.nr = nr;
2156 err = process(tool, ev, NULL, machine);