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_entry(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)
183 auxtrace__free(session);
184 auxtrace_index__free(&session->auxtrace_index);
185 perf_session__destroy_kernel_maps(session);
186 perf_session__delete_threads(session);
187 perf_env__exit(&session->header.env);
188 machines__exit(&session->machines);
190 perf_data_file__close(session->file);
194 static int process_event_synth_tracing_data_stub(struct perf_tool *tool
196 union perf_event *event
198 struct perf_session *session
201 dump_printf(": unhandled!\n");
205 static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused,
206 union perf_event *event __maybe_unused,
207 struct perf_evlist **pevlist
210 dump_printf(": unhandled!\n");
214 static int process_event_synth_event_update_stub(struct perf_tool *tool __maybe_unused,
215 union perf_event *event __maybe_unused,
216 struct perf_evlist **pevlist
220 perf_event__fprintf_event_update(event, stdout);
222 dump_printf(": unhandled!\n");
226 static int process_event_sample_stub(struct perf_tool *tool __maybe_unused,
227 union perf_event *event __maybe_unused,
228 struct perf_sample *sample __maybe_unused,
229 struct perf_evsel *evsel __maybe_unused,
230 struct machine *machine __maybe_unused)
232 dump_printf(": unhandled!\n");
236 static int process_event_stub(struct perf_tool *tool __maybe_unused,
237 union perf_event *event __maybe_unused,
238 struct perf_sample *sample __maybe_unused,
239 struct machine *machine __maybe_unused)
241 dump_printf(": unhandled!\n");
245 static int process_finished_round_stub(struct perf_tool *tool __maybe_unused,
246 union perf_event *event __maybe_unused,
247 struct ordered_events *oe __maybe_unused)
249 dump_printf(": unhandled!\n");
253 static int process_finished_round(struct perf_tool *tool,
254 union perf_event *event,
255 struct ordered_events *oe);
257 static int skipn(int fd, off_t n)
263 ret = read(fd, buf, min(n, (off_t)sizeof(buf)));
272 static s64 process_event_auxtrace_stub(struct perf_tool *tool __maybe_unused,
273 union perf_event *event,
274 struct perf_session *session
277 dump_printf(": unhandled!\n");
278 if (perf_data_file__is_pipe(session->file))
279 skipn(perf_data_file__fd(session->file), event->auxtrace.size);
280 return event->auxtrace.size;
283 static int process_event_op2_stub(struct perf_tool *tool __maybe_unused,
284 union perf_event *event __maybe_unused,
285 struct perf_session *session __maybe_unused)
287 dump_printf(": unhandled!\n");
293 int process_event_thread_map_stub(struct perf_tool *tool __maybe_unused,
294 union perf_event *event __maybe_unused,
295 struct perf_session *session __maybe_unused)
298 perf_event__fprintf_thread_map(event, stdout);
300 dump_printf(": unhandled!\n");
305 int process_event_cpu_map_stub(struct perf_tool *tool __maybe_unused,
306 union perf_event *event __maybe_unused,
307 struct perf_session *session __maybe_unused)
310 perf_event__fprintf_cpu_map(event, stdout);
312 dump_printf(": unhandled!\n");
317 int process_event_stat_config_stub(struct perf_tool *tool __maybe_unused,
318 union perf_event *event __maybe_unused,
319 struct perf_session *session __maybe_unused)
322 perf_event__fprintf_stat_config(event, stdout);
324 dump_printf(": unhandled!\n");
328 static int process_stat_stub(struct perf_tool *tool __maybe_unused,
329 union perf_event *event __maybe_unused,
330 struct perf_session *perf_session
334 perf_event__fprintf_stat(event, stdout);
336 dump_printf(": unhandled!\n");
340 static int process_stat_round_stub(struct perf_tool *tool __maybe_unused,
341 union perf_event *event __maybe_unused,
342 struct perf_session *perf_session
346 perf_event__fprintf_stat_round(event, stdout);
348 dump_printf(": unhandled!\n");
352 void perf_tool__fill_defaults(struct perf_tool *tool)
354 if (tool->sample == NULL)
355 tool->sample = process_event_sample_stub;
356 if (tool->mmap == NULL)
357 tool->mmap = process_event_stub;
358 if (tool->mmap2 == NULL)
359 tool->mmap2 = process_event_stub;
360 if (tool->comm == NULL)
361 tool->comm = process_event_stub;
362 if (tool->fork == NULL)
363 tool->fork = process_event_stub;
364 if (tool->exit == NULL)
365 tool->exit = process_event_stub;
366 if (tool->lost == NULL)
367 tool->lost = perf_event__process_lost;
368 if (tool->lost_samples == NULL)
369 tool->lost_samples = perf_event__process_lost_samples;
370 if (tool->aux == NULL)
371 tool->aux = perf_event__process_aux;
372 if (tool->itrace_start == NULL)
373 tool->itrace_start = perf_event__process_itrace_start;
374 if (tool->context_switch == NULL)
375 tool->context_switch = perf_event__process_switch;
376 if (tool->read == NULL)
377 tool->read = process_event_sample_stub;
378 if (tool->throttle == NULL)
379 tool->throttle = process_event_stub;
380 if (tool->unthrottle == NULL)
381 tool->unthrottle = process_event_stub;
382 if (tool->attr == NULL)
383 tool->attr = process_event_synth_attr_stub;
384 if (tool->event_update == NULL)
385 tool->event_update = process_event_synth_event_update_stub;
386 if (tool->tracing_data == NULL)
387 tool->tracing_data = process_event_synth_tracing_data_stub;
388 if (tool->build_id == NULL)
389 tool->build_id = process_event_op2_stub;
390 if (tool->finished_round == NULL) {
391 if (tool->ordered_events)
392 tool->finished_round = process_finished_round;
394 tool->finished_round = process_finished_round_stub;
396 if (tool->id_index == NULL)
397 tool->id_index = process_event_op2_stub;
398 if (tool->auxtrace_info == NULL)
399 tool->auxtrace_info = process_event_op2_stub;
400 if (tool->auxtrace == NULL)
401 tool->auxtrace = process_event_auxtrace_stub;
402 if (tool->auxtrace_error == NULL)
403 tool->auxtrace_error = process_event_op2_stub;
404 if (tool->thread_map == NULL)
405 tool->thread_map = process_event_thread_map_stub;
406 if (tool->cpu_map == NULL)
407 tool->cpu_map = process_event_cpu_map_stub;
408 if (tool->stat_config == NULL)
409 tool->stat_config = process_event_stat_config_stub;
410 if (tool->stat == NULL)
411 tool->stat = process_stat_stub;
412 if (tool->stat_round == NULL)
413 tool->stat_round = process_stat_round_stub;
414 if (tool->time_conv == NULL)
415 tool->time_conv = process_event_op2_stub;
418 static void swap_sample_id_all(union perf_event *event, void *data)
420 void *end = (void *) event + event->header.size;
421 int size = end - data;
423 BUG_ON(size % sizeof(u64));
424 mem_bswap_64(data, size);
427 static void perf_event__all64_swap(union perf_event *event,
428 bool sample_id_all __maybe_unused)
430 struct perf_event_header *hdr = &event->header;
431 mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr));
434 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all)
436 event->comm.pid = bswap_32(event->comm.pid);
437 event->comm.tid = bswap_32(event->comm.tid);
440 void *data = &event->comm.comm;
442 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
443 swap_sample_id_all(event, data);
447 static void perf_event__mmap_swap(union perf_event *event,
450 event->mmap.pid = bswap_32(event->mmap.pid);
451 event->mmap.tid = bswap_32(event->mmap.tid);
452 event->mmap.start = bswap_64(event->mmap.start);
453 event->mmap.len = bswap_64(event->mmap.len);
454 event->mmap.pgoff = bswap_64(event->mmap.pgoff);
457 void *data = &event->mmap.filename;
459 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
460 swap_sample_id_all(event, data);
464 static void perf_event__mmap2_swap(union perf_event *event,
467 event->mmap2.pid = bswap_32(event->mmap2.pid);
468 event->mmap2.tid = bswap_32(event->mmap2.tid);
469 event->mmap2.start = bswap_64(event->mmap2.start);
470 event->mmap2.len = bswap_64(event->mmap2.len);
471 event->mmap2.pgoff = bswap_64(event->mmap2.pgoff);
472 event->mmap2.maj = bswap_32(event->mmap2.maj);
473 event->mmap2.min = bswap_32(event->mmap2.min);
474 event->mmap2.ino = bswap_64(event->mmap2.ino);
477 void *data = &event->mmap2.filename;
479 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
480 swap_sample_id_all(event, data);
483 static void perf_event__task_swap(union perf_event *event, bool sample_id_all)
485 event->fork.pid = bswap_32(event->fork.pid);
486 event->fork.tid = bswap_32(event->fork.tid);
487 event->fork.ppid = bswap_32(event->fork.ppid);
488 event->fork.ptid = bswap_32(event->fork.ptid);
489 event->fork.time = bswap_64(event->fork.time);
492 swap_sample_id_all(event, &event->fork + 1);
495 static void perf_event__read_swap(union perf_event *event, bool sample_id_all)
497 event->read.pid = bswap_32(event->read.pid);
498 event->read.tid = bswap_32(event->read.tid);
499 event->read.value = bswap_64(event->read.value);
500 event->read.time_enabled = bswap_64(event->read.time_enabled);
501 event->read.time_running = bswap_64(event->read.time_running);
502 event->read.id = bswap_64(event->read.id);
505 swap_sample_id_all(event, &event->read + 1);
508 static void perf_event__aux_swap(union perf_event *event, bool sample_id_all)
510 event->aux.aux_offset = bswap_64(event->aux.aux_offset);
511 event->aux.aux_size = bswap_64(event->aux.aux_size);
512 event->aux.flags = bswap_64(event->aux.flags);
515 swap_sample_id_all(event, &event->aux + 1);
518 static void perf_event__itrace_start_swap(union perf_event *event,
521 event->itrace_start.pid = bswap_32(event->itrace_start.pid);
522 event->itrace_start.tid = bswap_32(event->itrace_start.tid);
525 swap_sample_id_all(event, &event->itrace_start + 1);
528 static void perf_event__switch_swap(union perf_event *event, bool sample_id_all)
530 if (event->header.type == PERF_RECORD_SWITCH_CPU_WIDE) {
531 event->context_switch.next_prev_pid =
532 bswap_32(event->context_switch.next_prev_pid);
533 event->context_switch.next_prev_tid =
534 bswap_32(event->context_switch.next_prev_tid);
538 swap_sample_id_all(event, &event->context_switch + 1);
541 static void perf_event__throttle_swap(union perf_event *event,
544 event->throttle.time = bswap_64(event->throttle.time);
545 event->throttle.id = bswap_64(event->throttle.id);
546 event->throttle.stream_id = bswap_64(event->throttle.stream_id);
549 swap_sample_id_all(event, &event->throttle + 1);
552 static u8 revbyte(u8 b)
554 int rev = (b >> 4) | ((b & 0xf) << 4);
555 rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2);
556 rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1);
561 * XXX this is hack in attempt to carry flags bitfield
562 * through endian village. ABI says:
564 * Bit-fields are allocated from right to left (least to most significant)
565 * on little-endian implementations and from left to right (most to least
566 * significant) on big-endian implementations.
568 * The above seems to be byte specific, so we need to reverse each
569 * byte of the bitfield. 'Internet' also says this might be implementation
570 * specific and we probably need proper fix and carry perf_event_attr
571 * bitfield flags in separate data file FEAT_ section. Thought this seems
574 static void swap_bitfield(u8 *p, unsigned len)
578 for (i = 0; i < len; i++) {
584 /* exported for swapping attributes in file header */
585 void perf_event__attr_swap(struct perf_event_attr *attr)
587 attr->type = bswap_32(attr->type);
588 attr->size = bswap_32(attr->size);
590 #define bswap_safe(f, n) \
591 (attr->size > (offsetof(struct perf_event_attr, f) + \
592 sizeof(attr->f) * (n)))
593 #define bswap_field(f, sz) \
595 if (bswap_safe(f, 0)) \
596 attr->f = bswap_##sz(attr->f); \
598 #define bswap_field_16(f) bswap_field(f, 16)
599 #define bswap_field_32(f) bswap_field(f, 32)
600 #define bswap_field_64(f) bswap_field(f, 64)
602 bswap_field_64(config);
603 bswap_field_64(sample_period);
604 bswap_field_64(sample_type);
605 bswap_field_64(read_format);
606 bswap_field_32(wakeup_events);
607 bswap_field_32(bp_type);
608 bswap_field_64(bp_addr);
609 bswap_field_64(bp_len);
610 bswap_field_64(branch_sample_type);
611 bswap_field_64(sample_regs_user);
612 bswap_field_32(sample_stack_user);
613 bswap_field_32(aux_watermark);
614 bswap_field_16(sample_max_stack);
617 * After read_format are bitfields. Check read_format because
618 * we are unable to use offsetof on bitfield.
620 if (bswap_safe(read_format, 1))
621 swap_bitfield((u8 *) (&attr->read_format + 1),
623 #undef bswap_field_64
624 #undef bswap_field_32
629 static void perf_event__hdr_attr_swap(union perf_event *event,
630 bool sample_id_all __maybe_unused)
634 perf_event__attr_swap(&event->attr.attr);
636 size = event->header.size;
637 size -= (void *)&event->attr.id - (void *)event;
638 mem_bswap_64(event->attr.id, size);
641 static void perf_event__event_update_swap(union perf_event *event,
642 bool sample_id_all __maybe_unused)
644 event->event_update.type = bswap_64(event->event_update.type);
645 event->event_update.id = bswap_64(event->event_update.id);
648 static void perf_event__event_type_swap(union perf_event *event,
649 bool sample_id_all __maybe_unused)
651 event->event_type.event_type.event_id =
652 bswap_64(event->event_type.event_type.event_id);
655 static void perf_event__tracing_data_swap(union perf_event *event,
656 bool sample_id_all __maybe_unused)
658 event->tracing_data.size = bswap_32(event->tracing_data.size);
661 static void perf_event__auxtrace_info_swap(union perf_event *event,
662 bool sample_id_all __maybe_unused)
666 event->auxtrace_info.type = bswap_32(event->auxtrace_info.type);
668 size = event->header.size;
669 size -= (void *)&event->auxtrace_info.priv - (void *)event;
670 mem_bswap_64(event->auxtrace_info.priv, size);
673 static void perf_event__auxtrace_swap(union perf_event *event,
674 bool sample_id_all __maybe_unused)
676 event->auxtrace.size = bswap_64(event->auxtrace.size);
677 event->auxtrace.offset = bswap_64(event->auxtrace.offset);
678 event->auxtrace.reference = bswap_64(event->auxtrace.reference);
679 event->auxtrace.idx = bswap_32(event->auxtrace.idx);
680 event->auxtrace.tid = bswap_32(event->auxtrace.tid);
681 event->auxtrace.cpu = bswap_32(event->auxtrace.cpu);
684 static void perf_event__auxtrace_error_swap(union perf_event *event,
685 bool sample_id_all __maybe_unused)
687 event->auxtrace_error.type = bswap_32(event->auxtrace_error.type);
688 event->auxtrace_error.code = bswap_32(event->auxtrace_error.code);
689 event->auxtrace_error.cpu = bswap_32(event->auxtrace_error.cpu);
690 event->auxtrace_error.pid = bswap_32(event->auxtrace_error.pid);
691 event->auxtrace_error.tid = bswap_32(event->auxtrace_error.tid);
692 event->auxtrace_error.ip = bswap_64(event->auxtrace_error.ip);
695 static void perf_event__thread_map_swap(union perf_event *event,
696 bool sample_id_all __maybe_unused)
700 event->thread_map.nr = bswap_64(event->thread_map.nr);
702 for (i = 0; i < event->thread_map.nr; i++)
703 event->thread_map.entries[i].pid = bswap_64(event->thread_map.entries[i].pid);
706 static void perf_event__cpu_map_swap(union perf_event *event,
707 bool sample_id_all __maybe_unused)
709 struct cpu_map_data *data = &event->cpu_map.data;
710 struct cpu_map_entries *cpus;
711 struct cpu_map_mask *mask;
714 data->type = bswap_64(data->type);
716 switch (data->type) {
717 case PERF_CPU_MAP__CPUS:
718 cpus = (struct cpu_map_entries *)data->data;
720 cpus->nr = bswap_16(cpus->nr);
722 for (i = 0; i < cpus->nr; i++)
723 cpus->cpu[i] = bswap_16(cpus->cpu[i]);
725 case PERF_CPU_MAP__MASK:
726 mask = (struct cpu_map_mask *) data->data;
728 mask->nr = bswap_16(mask->nr);
729 mask->long_size = bswap_16(mask->long_size);
731 switch (mask->long_size) {
732 case 4: mem_bswap_32(&mask->mask, mask->nr); break;
733 case 8: mem_bswap_64(&mask->mask, mask->nr); break;
735 pr_err("cpu_map swap: unsupported long size\n");
742 static void perf_event__stat_config_swap(union perf_event *event,
743 bool sample_id_all __maybe_unused)
747 size = event->stat_config.nr * sizeof(event->stat_config.data[0]);
748 size += 1; /* nr item itself */
749 mem_bswap_64(&event->stat_config.nr, size);
752 static void perf_event__stat_swap(union perf_event *event,
753 bool sample_id_all __maybe_unused)
755 event->stat.id = bswap_64(event->stat.id);
756 event->stat.thread = bswap_32(event->stat.thread);
757 event->stat.cpu = bswap_32(event->stat.cpu);
758 event->stat.val = bswap_64(event->stat.val);
759 event->stat.ena = bswap_64(event->stat.ena);
760 event->stat.run = bswap_64(event->stat.run);
763 static void perf_event__stat_round_swap(union perf_event *event,
764 bool sample_id_all __maybe_unused)
766 event->stat_round.type = bswap_64(event->stat_round.type);
767 event->stat_round.time = bswap_64(event->stat_round.time);
770 typedef void (*perf_event__swap_op)(union perf_event *event,
773 static perf_event__swap_op perf_event__swap_ops[] = {
774 [PERF_RECORD_MMAP] = perf_event__mmap_swap,
775 [PERF_RECORD_MMAP2] = perf_event__mmap2_swap,
776 [PERF_RECORD_COMM] = perf_event__comm_swap,
777 [PERF_RECORD_FORK] = perf_event__task_swap,
778 [PERF_RECORD_EXIT] = perf_event__task_swap,
779 [PERF_RECORD_LOST] = perf_event__all64_swap,
780 [PERF_RECORD_READ] = perf_event__read_swap,
781 [PERF_RECORD_THROTTLE] = perf_event__throttle_swap,
782 [PERF_RECORD_UNTHROTTLE] = perf_event__throttle_swap,
783 [PERF_RECORD_SAMPLE] = perf_event__all64_swap,
784 [PERF_RECORD_AUX] = perf_event__aux_swap,
785 [PERF_RECORD_ITRACE_START] = perf_event__itrace_start_swap,
786 [PERF_RECORD_LOST_SAMPLES] = perf_event__all64_swap,
787 [PERF_RECORD_SWITCH] = perf_event__switch_swap,
788 [PERF_RECORD_SWITCH_CPU_WIDE] = perf_event__switch_swap,
789 [PERF_RECORD_HEADER_ATTR] = perf_event__hdr_attr_swap,
790 [PERF_RECORD_HEADER_EVENT_TYPE] = perf_event__event_type_swap,
791 [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
792 [PERF_RECORD_HEADER_BUILD_ID] = NULL,
793 [PERF_RECORD_ID_INDEX] = perf_event__all64_swap,
794 [PERF_RECORD_AUXTRACE_INFO] = perf_event__auxtrace_info_swap,
795 [PERF_RECORD_AUXTRACE] = perf_event__auxtrace_swap,
796 [PERF_RECORD_AUXTRACE_ERROR] = perf_event__auxtrace_error_swap,
797 [PERF_RECORD_THREAD_MAP] = perf_event__thread_map_swap,
798 [PERF_RECORD_CPU_MAP] = perf_event__cpu_map_swap,
799 [PERF_RECORD_STAT_CONFIG] = perf_event__stat_config_swap,
800 [PERF_RECORD_STAT] = perf_event__stat_swap,
801 [PERF_RECORD_STAT_ROUND] = perf_event__stat_round_swap,
802 [PERF_RECORD_EVENT_UPDATE] = perf_event__event_update_swap,
803 [PERF_RECORD_TIME_CONV] = perf_event__all64_swap,
804 [PERF_RECORD_HEADER_MAX] = NULL,
808 * When perf record finishes a pass on every buffers, it records this pseudo
810 * We record the max timestamp t found in the pass n.
811 * Assuming these timestamps are monotonic across cpus, we know that if
812 * a buffer still has events with timestamps below t, they will be all
813 * available and then read in the pass n + 1.
814 * Hence when we start to read the pass n + 2, we can safely flush every
815 * events with timestamps below t.
817 * ============ PASS n =================
820 * cnt1 timestamps | cnt2 timestamps
823 * - | 4 <--- max recorded
825 * ============ PASS n + 1 ==============
828 * cnt1 timestamps | cnt2 timestamps
831 * 5 | 7 <---- max recorded
833 * Flush every events below timestamp 4
835 * ============ PASS n + 2 ==============
838 * cnt1 timestamps | cnt2 timestamps
843 * Flush every events below timestamp 7
846 static int process_finished_round(struct perf_tool *tool __maybe_unused,
847 union perf_event *event __maybe_unused,
848 struct ordered_events *oe)
851 fprintf(stdout, "\n");
852 return ordered_events__flush(oe, OE_FLUSH__ROUND);
855 int perf_session__queue_event(struct perf_session *s, union perf_event *event,
856 struct perf_sample *sample, u64 file_offset)
858 return ordered_events__queue(&s->ordered_events, event, sample, file_offset);
861 static void callchain__lbr_callstack_printf(struct perf_sample *sample)
863 struct ip_callchain *callchain = sample->callchain;
864 struct branch_stack *lbr_stack = sample->branch_stack;
865 u64 kernel_callchain_nr = callchain->nr;
868 for (i = 0; i < kernel_callchain_nr; i++) {
869 if (callchain->ips[i] == PERF_CONTEXT_USER)
873 if ((i != kernel_callchain_nr) && lbr_stack->nr) {
876 * LBR callstack can only get user call chain,
877 * i is kernel call chain number,
878 * 1 is PERF_CONTEXT_USER.
880 * The user call chain is stored in LBR registers.
881 * LBR are pair registers. The caller is stored
882 * in "from" register, while the callee is stored
884 * For example, there is a call stack
885 * "A"->"B"->"C"->"D".
886 * The LBR registers will recorde like
887 * "C"->"D", "B"->"C", "A"->"B".
888 * So only the first "to" register and all "from"
889 * registers are needed to construct the whole stack.
891 total_nr = i + 1 + lbr_stack->nr + 1;
892 kernel_callchain_nr = i + 1;
894 printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr);
896 for (i = 0; i < kernel_callchain_nr; i++)
897 printf("..... %2d: %016" PRIx64 "\n",
898 i, callchain->ips[i]);
900 printf("..... %2d: %016" PRIx64 "\n",
901 (int)(kernel_callchain_nr), lbr_stack->entries[0].to);
902 for (i = 0; i < lbr_stack->nr; i++)
903 printf("..... %2d: %016" PRIx64 "\n",
904 (int)(i + kernel_callchain_nr + 1), lbr_stack->entries[i].from);
908 static void callchain__printf(struct perf_evsel *evsel,
909 struct perf_sample *sample)
912 struct ip_callchain *callchain = sample->callchain;
914 if (perf_evsel__has_branch_callstack(evsel))
915 callchain__lbr_callstack_printf(sample);
917 printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr);
919 for (i = 0; i < callchain->nr; i++)
920 printf("..... %2d: %016" PRIx64 "\n",
921 i, callchain->ips[i]);
924 static void branch_stack__printf(struct perf_sample *sample)
928 printf("... branch stack: nr:%" PRIu64 "\n", sample->branch_stack->nr);
930 for (i = 0; i < sample->branch_stack->nr; i++) {
931 struct branch_entry *e = &sample->branch_stack->entries[i];
933 printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x\n",
936 e->flags.mispred ? "M" : " ",
937 e->flags.predicted ? "P" : " ",
938 e->flags.abort ? "A" : " ",
939 e->flags.in_tx ? "T" : " ",
940 (unsigned)e->flags.reserved);
944 static void regs_dump__printf(u64 mask, u64 *regs)
948 for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) {
951 printf(".... %-5s 0x%" PRIx64 "\n",
952 perf_reg_name(rid), val);
956 static const char *regs_abi[] = {
957 [PERF_SAMPLE_REGS_ABI_NONE] = "none",
958 [PERF_SAMPLE_REGS_ABI_32] = "32-bit",
959 [PERF_SAMPLE_REGS_ABI_64] = "64-bit",
962 static inline const char *regs_dump_abi(struct regs_dump *d)
964 if (d->abi > PERF_SAMPLE_REGS_ABI_64)
967 return regs_abi[d->abi];
970 static void regs__printf(const char *type, struct regs_dump *regs)
972 u64 mask = regs->mask;
974 printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n",
977 regs_dump_abi(regs));
979 regs_dump__printf(mask, regs->regs);
982 static void regs_user__printf(struct perf_sample *sample)
984 struct regs_dump *user_regs = &sample->user_regs;
987 regs__printf("user", user_regs);
990 static void regs_intr__printf(struct perf_sample *sample)
992 struct regs_dump *intr_regs = &sample->intr_regs;
995 regs__printf("intr", intr_regs);
998 static void stack_user__printf(struct stack_dump *dump)
1000 printf("... ustack: size %" PRIu64 ", offset 0x%x\n",
1001 dump->size, dump->offset);
1004 static void perf_evlist__print_tstamp(struct perf_evlist *evlist,
1005 union perf_event *event,
1006 struct perf_sample *sample)
1008 u64 sample_type = __perf_evlist__combined_sample_type(evlist);
1010 if (event->header.type != PERF_RECORD_SAMPLE &&
1011 !perf_evlist__sample_id_all(evlist)) {
1012 fputs("-1 -1 ", stdout);
1016 if ((sample_type & PERF_SAMPLE_CPU))
1017 printf("%u ", sample->cpu);
1019 if (sample_type & PERF_SAMPLE_TIME)
1020 printf("%" PRIu64 " ", sample->time);
1023 static void sample_read__printf(struct perf_sample *sample, u64 read_format)
1025 printf("... sample_read:\n");
1027 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1028 printf("...... time enabled %016" PRIx64 "\n",
1029 sample->read.time_enabled);
1031 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1032 printf("...... time running %016" PRIx64 "\n",
1033 sample->read.time_running);
1035 if (read_format & PERF_FORMAT_GROUP) {
1038 printf(".... group nr %" PRIu64 "\n", sample->read.group.nr);
1040 for (i = 0; i < sample->read.group.nr; i++) {
1041 struct sample_read_value *value;
1043 value = &sample->read.group.values[i];
1044 printf("..... id %016" PRIx64
1045 ", value %016" PRIx64 "\n",
1046 value->id, value->value);
1049 printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n",
1050 sample->read.one.id, sample->read.one.value);
1053 static void dump_event(struct perf_evlist *evlist, union perf_event *event,
1054 u64 file_offset, struct perf_sample *sample)
1059 printf("\n%#" PRIx64 " [%#x]: event: %d\n",
1060 file_offset, event->header.size, event->header.type);
1065 perf_evlist__print_tstamp(evlist, event, sample);
1067 printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset,
1068 event->header.size, perf_event__name(event->header.type));
1071 static void dump_sample(struct perf_evsel *evsel, union perf_event *event,
1072 struct perf_sample *sample)
1079 printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n",
1080 event->header.misc, sample->pid, sample->tid, sample->ip,
1081 sample->period, sample->addr);
1083 sample_type = evsel->attr.sample_type;
1085 if (sample_type & PERF_SAMPLE_CALLCHAIN)
1086 callchain__printf(evsel, sample);
1088 if ((sample_type & PERF_SAMPLE_BRANCH_STACK) && !perf_evsel__has_branch_callstack(evsel))
1089 branch_stack__printf(sample);
1091 if (sample_type & PERF_SAMPLE_REGS_USER)
1092 regs_user__printf(sample);
1094 if (sample_type & PERF_SAMPLE_REGS_INTR)
1095 regs_intr__printf(sample);
1097 if (sample_type & PERF_SAMPLE_STACK_USER)
1098 stack_user__printf(&sample->user_stack);
1100 if (sample_type & PERF_SAMPLE_WEIGHT)
1101 printf("... weight: %" PRIu64 "\n", sample->weight);
1103 if (sample_type & PERF_SAMPLE_DATA_SRC)
1104 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src);
1106 if (sample_type & PERF_SAMPLE_TRANSACTION)
1107 printf("... transaction: %" PRIx64 "\n", sample->transaction);
1109 if (sample_type & PERF_SAMPLE_READ)
1110 sample_read__printf(sample, evsel->attr.read_format);
1113 static struct machine *machines__find_for_cpumode(struct machines *machines,
1114 union perf_event *event,
1115 struct perf_sample *sample)
1117 struct machine *machine;
1120 ((sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL) ||
1121 (sample->cpumode == PERF_RECORD_MISC_GUEST_USER))) {
1124 if (event->header.type == PERF_RECORD_MMAP
1125 || event->header.type == PERF_RECORD_MMAP2)
1126 pid = event->mmap.pid;
1130 machine = machines__find(machines, pid);
1132 machine = machines__findnew(machines, DEFAULT_GUEST_KERNEL_ID);
1136 return &machines->host;
1139 static int deliver_sample_value(struct perf_evlist *evlist,
1140 struct perf_tool *tool,
1141 union perf_event *event,
1142 struct perf_sample *sample,
1143 struct sample_read_value *v,
1144 struct machine *machine)
1146 struct perf_sample_id *sid = perf_evlist__id2sid(evlist, v->id);
1150 sample->period = v->value - sid->period;
1151 sid->period = v->value;
1154 if (!sid || sid->evsel == NULL) {
1155 ++evlist->stats.nr_unknown_id;
1159 return tool->sample(tool, event, sample, sid->evsel, machine);
1162 static int deliver_sample_group(struct perf_evlist *evlist,
1163 struct perf_tool *tool,
1164 union perf_event *event,
1165 struct perf_sample *sample,
1166 struct machine *machine)
1171 for (i = 0; i < sample->read.group.nr; i++) {
1172 ret = deliver_sample_value(evlist, tool, event, sample,
1173 &sample->read.group.values[i],
1183 perf_evlist__deliver_sample(struct perf_evlist *evlist,
1184 struct perf_tool *tool,
1185 union perf_event *event,
1186 struct perf_sample *sample,
1187 struct perf_evsel *evsel,
1188 struct machine *machine)
1190 /* We know evsel != NULL. */
1191 u64 sample_type = evsel->attr.sample_type;
1192 u64 read_format = evsel->attr.read_format;
1194 /* Standard sample delivery. */
1195 if (!(sample_type & PERF_SAMPLE_READ))
1196 return tool->sample(tool, event, sample, evsel, machine);
1198 /* For PERF_SAMPLE_READ we have either single or group mode. */
1199 if (read_format & PERF_FORMAT_GROUP)
1200 return deliver_sample_group(evlist, tool, event, sample,
1203 return deliver_sample_value(evlist, tool, event, sample,
1204 &sample->read.one, machine);
1207 static int machines__deliver_event(struct machines *machines,
1208 struct perf_evlist *evlist,
1209 union perf_event *event,
1210 struct perf_sample *sample,
1211 struct perf_tool *tool, u64 file_offset)
1213 struct perf_evsel *evsel;
1214 struct machine *machine;
1216 dump_event(evlist, event, file_offset, sample);
1218 evsel = perf_evlist__id2evsel(evlist, sample->id);
1220 machine = machines__find_for_cpumode(machines, event, sample);
1222 switch (event->header.type) {
1223 case PERF_RECORD_SAMPLE:
1224 if (evsel == NULL) {
1225 ++evlist->stats.nr_unknown_id;
1228 dump_sample(evsel, event, sample);
1229 if (machine == NULL) {
1230 ++evlist->stats.nr_unprocessable_samples;
1233 return perf_evlist__deliver_sample(evlist, tool, event, sample, evsel, machine);
1234 case PERF_RECORD_MMAP:
1235 return tool->mmap(tool, event, sample, machine);
1236 case PERF_RECORD_MMAP2:
1237 if (event->header.misc & PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT)
1238 ++evlist->stats.nr_proc_map_timeout;
1239 return tool->mmap2(tool, event, sample, machine);
1240 case PERF_RECORD_COMM:
1241 return tool->comm(tool, event, sample, machine);
1242 case PERF_RECORD_FORK:
1243 return tool->fork(tool, event, sample, machine);
1244 case PERF_RECORD_EXIT:
1245 return tool->exit(tool, event, sample, machine);
1246 case PERF_RECORD_LOST:
1247 if (tool->lost == perf_event__process_lost)
1248 evlist->stats.total_lost += event->lost.lost;
1249 return tool->lost(tool, event, sample, machine);
1250 case PERF_RECORD_LOST_SAMPLES:
1251 if (tool->lost_samples == perf_event__process_lost_samples)
1252 evlist->stats.total_lost_samples += event->lost_samples.lost;
1253 return tool->lost_samples(tool, event, sample, machine);
1254 case PERF_RECORD_READ:
1255 return tool->read(tool, event, sample, evsel, machine);
1256 case PERF_RECORD_THROTTLE:
1257 return tool->throttle(tool, event, sample, machine);
1258 case PERF_RECORD_UNTHROTTLE:
1259 return tool->unthrottle(tool, event, sample, machine);
1260 case PERF_RECORD_AUX:
1261 if (tool->aux == perf_event__process_aux &&
1262 (event->aux.flags & PERF_AUX_FLAG_TRUNCATED))
1263 evlist->stats.total_aux_lost += 1;
1264 return tool->aux(tool, event, sample, machine);
1265 case PERF_RECORD_ITRACE_START:
1266 return tool->itrace_start(tool, event, sample, machine);
1267 case PERF_RECORD_SWITCH:
1268 case PERF_RECORD_SWITCH_CPU_WIDE:
1269 return tool->context_switch(tool, event, sample, machine);
1271 ++evlist->stats.nr_unknown_events;
1276 static int perf_session__deliver_event(struct perf_session *session,
1277 union perf_event *event,
1278 struct perf_sample *sample,
1279 struct perf_tool *tool,
1284 ret = auxtrace__process_event(session, event, sample, tool);
1290 return machines__deliver_event(&session->machines, session->evlist,
1291 event, sample, tool, file_offset);
1294 static s64 perf_session__process_user_event(struct perf_session *session,
1295 union perf_event *event,
1298 struct ordered_events *oe = &session->ordered_events;
1299 struct perf_tool *tool = session->tool;
1300 int fd = perf_data_file__fd(session->file);
1303 dump_event(session->evlist, event, file_offset, NULL);
1305 /* These events are processed right away */
1306 switch (event->header.type) {
1307 case PERF_RECORD_HEADER_ATTR:
1308 err = tool->attr(tool, event, &session->evlist);
1310 perf_session__set_id_hdr_size(session);
1311 perf_session__set_comm_exec(session);
1314 case PERF_RECORD_EVENT_UPDATE:
1315 return tool->event_update(tool, event, &session->evlist);
1316 case PERF_RECORD_HEADER_EVENT_TYPE:
1318 * Depreceated, but we need to handle it for sake
1319 * of old data files create in pipe mode.
1322 case PERF_RECORD_HEADER_TRACING_DATA:
1323 /* setup for reading amidst mmap */
1324 lseek(fd, file_offset, SEEK_SET);
1325 return tool->tracing_data(tool, event, session);
1326 case PERF_RECORD_HEADER_BUILD_ID:
1327 return tool->build_id(tool, event, session);
1328 case PERF_RECORD_FINISHED_ROUND:
1329 return tool->finished_round(tool, event, oe);
1330 case PERF_RECORD_ID_INDEX:
1331 return tool->id_index(tool, event, session);
1332 case PERF_RECORD_AUXTRACE_INFO:
1333 return tool->auxtrace_info(tool, event, session);
1334 case PERF_RECORD_AUXTRACE:
1335 /* setup for reading amidst mmap */
1336 lseek(fd, file_offset + event->header.size, SEEK_SET);
1337 return tool->auxtrace(tool, event, session);
1338 case PERF_RECORD_AUXTRACE_ERROR:
1339 perf_session__auxtrace_error_inc(session, event);
1340 return tool->auxtrace_error(tool, event, session);
1341 case PERF_RECORD_THREAD_MAP:
1342 return tool->thread_map(tool, event, session);
1343 case PERF_RECORD_CPU_MAP:
1344 return tool->cpu_map(tool, event, session);
1345 case PERF_RECORD_STAT_CONFIG:
1346 return tool->stat_config(tool, event, session);
1347 case PERF_RECORD_STAT:
1348 return tool->stat(tool, event, session);
1349 case PERF_RECORD_STAT_ROUND:
1350 return tool->stat_round(tool, event, session);
1351 case PERF_RECORD_TIME_CONV:
1352 session->time_conv = event->time_conv;
1353 return tool->time_conv(tool, event, session);
1359 int perf_session__deliver_synth_event(struct perf_session *session,
1360 union perf_event *event,
1361 struct perf_sample *sample)
1363 struct perf_evlist *evlist = session->evlist;
1364 struct perf_tool *tool = session->tool;
1366 events_stats__inc(&evlist->stats, event->header.type);
1368 if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1369 return perf_session__process_user_event(session, event, 0);
1371 return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0);
1374 static void event_swap(union perf_event *event, bool sample_id_all)
1376 perf_event__swap_op swap;
1378 swap = perf_event__swap_ops[event->header.type];
1380 swap(event, sample_id_all);
1383 int perf_session__peek_event(struct perf_session *session, off_t file_offset,
1384 void *buf, size_t buf_sz,
1385 union perf_event **event_ptr,
1386 struct perf_sample *sample)
1388 union perf_event *event;
1389 size_t hdr_sz, rest;
1392 if (session->one_mmap && !session->header.needs_swap) {
1393 event = file_offset - session->one_mmap_offset +
1394 session->one_mmap_addr;
1395 goto out_parse_sample;
1398 if (perf_data_file__is_pipe(session->file))
1401 fd = perf_data_file__fd(session->file);
1402 hdr_sz = sizeof(struct perf_event_header);
1404 if (buf_sz < hdr_sz)
1407 if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 ||
1408 readn(fd, buf, hdr_sz) != (ssize_t)hdr_sz)
1411 event = (union perf_event *)buf;
1413 if (session->header.needs_swap)
1414 perf_event_header__bswap(&event->header);
1416 if (event->header.size < hdr_sz || event->header.size > buf_sz)
1419 rest = event->header.size - hdr_sz;
1421 if (readn(fd, buf, rest) != (ssize_t)rest)
1424 if (session->header.needs_swap)
1425 event_swap(event, perf_evlist__sample_id_all(session->evlist));
1429 if (sample && event->header.type < PERF_RECORD_USER_TYPE_START &&
1430 perf_evlist__parse_sample(session->evlist, event, sample))
1438 static s64 perf_session__process_event(struct perf_session *session,
1439 union perf_event *event, u64 file_offset)
1441 struct perf_evlist *evlist = session->evlist;
1442 struct perf_tool *tool = session->tool;
1443 struct perf_sample sample;
1446 if (session->header.needs_swap)
1447 event_swap(event, perf_evlist__sample_id_all(evlist));
1449 if (event->header.type >= PERF_RECORD_HEADER_MAX)
1452 events_stats__inc(&evlist->stats, event->header.type);
1454 if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1455 return perf_session__process_user_event(session, event, file_offset);
1458 * For all kernel events we get the sample data
1460 ret = perf_evlist__parse_sample(evlist, event, &sample);
1464 if (tool->ordered_events) {
1465 ret = perf_session__queue_event(session, event, &sample, file_offset);
1470 return perf_session__deliver_event(session, event, &sample, tool,
1474 void perf_event_header__bswap(struct perf_event_header *hdr)
1476 hdr->type = bswap_32(hdr->type);
1477 hdr->misc = bswap_16(hdr->misc);
1478 hdr->size = bswap_16(hdr->size);
1481 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid)
1483 return machine__findnew_thread(&session->machines.host, -1, pid);
1486 int perf_session__register_idle_thread(struct perf_session *session)
1488 struct thread *thread;
1491 thread = machine__findnew_thread(&session->machines.host, 0, 0);
1492 if (thread == NULL || thread__set_comm(thread, "swapper", 0)) {
1493 pr_err("problem inserting idle task.\n");
1497 /* machine__findnew_thread() got the thread, so put it */
1498 thread__put(thread);
1503 perf_session__warn_order(const struct perf_session *session)
1505 const struct ordered_events *oe = &session->ordered_events;
1506 struct perf_evsel *evsel;
1507 bool should_warn = true;
1509 evlist__for_each_entry(session->evlist, evsel) {
1510 if (evsel->attr.write_backward)
1511 should_warn = false;
1516 if (oe->nr_unordered_events != 0)
1517 ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events);
1520 static void perf_session__warn_about_errors(const struct perf_session *session)
1522 const struct events_stats *stats = &session->evlist->stats;
1524 if (session->tool->lost == perf_event__process_lost &&
1525 stats->nr_events[PERF_RECORD_LOST] != 0) {
1526 ui__warning("Processed %d events and lost %d chunks!\n\n"
1527 "Check IO/CPU overload!\n\n",
1528 stats->nr_events[0],
1529 stats->nr_events[PERF_RECORD_LOST]);
1532 if (session->tool->lost_samples == perf_event__process_lost_samples) {
1535 drop_rate = (double)stats->total_lost_samples /
1536 (double) (stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples);
1537 if (drop_rate > 0.05) {
1538 ui__warning("Processed %" PRIu64 " samples and lost %3.2f%% samples!\n\n",
1539 stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples,
1544 if (session->tool->aux == perf_event__process_aux &&
1545 stats->total_aux_lost != 0) {
1546 ui__warning("AUX data lost %" PRIu64 " times out of %u!\n\n",
1547 stats->total_aux_lost,
1548 stats->nr_events[PERF_RECORD_AUX]);
1551 if (stats->nr_unknown_events != 0) {
1552 ui__warning("Found %u unknown events!\n\n"
1553 "Is this an older tool processing a perf.data "
1554 "file generated by a more recent tool?\n\n"
1555 "If that is not the case, consider "
1556 "reporting to linux-kernel@vger.kernel.org.\n\n",
1557 stats->nr_unknown_events);
1560 if (stats->nr_unknown_id != 0) {
1561 ui__warning("%u samples with id not present in the header\n",
1562 stats->nr_unknown_id);
1565 if (stats->nr_invalid_chains != 0) {
1566 ui__warning("Found invalid callchains!\n\n"
1567 "%u out of %u events were discarded for this reason.\n\n"
1568 "Consider reporting to linux-kernel@vger.kernel.org.\n\n",
1569 stats->nr_invalid_chains,
1570 stats->nr_events[PERF_RECORD_SAMPLE]);
1573 if (stats->nr_unprocessable_samples != 0) {
1574 ui__warning("%u unprocessable samples recorded.\n"
1575 "Do you have a KVM guest running and not using 'perf kvm'?\n",
1576 stats->nr_unprocessable_samples);
1579 perf_session__warn_order(session);
1581 events_stats__auxtrace_error_warn(stats);
1583 if (stats->nr_proc_map_timeout != 0) {
1584 ui__warning("%d map information files for pre-existing threads were\n"
1585 "not processed, if there are samples for addresses they\n"
1586 "will not be resolved, you may find out which are these\n"
1587 "threads by running with -v and redirecting the output\n"
1589 "The time limit to process proc map is too short?\n"
1590 "Increase it by --proc-map-timeout\n",
1591 stats->nr_proc_map_timeout);
1595 static int perf_session__flush_thread_stack(struct thread *thread,
1596 void *p __maybe_unused)
1598 return thread_stack__flush(thread);
1601 static int perf_session__flush_thread_stacks(struct perf_session *session)
1603 return machines__for_each_thread(&session->machines,
1604 perf_session__flush_thread_stack,
1608 volatile int session_done;
1610 static int __perf_session__process_pipe_events(struct perf_session *session)
1612 struct ordered_events *oe = &session->ordered_events;
1613 struct perf_tool *tool = session->tool;
1614 int fd = perf_data_file__fd(session->file);
1615 union perf_event *event;
1616 uint32_t size, cur_size = 0;
1623 perf_tool__fill_defaults(tool);
1626 cur_size = sizeof(union perf_event);
1628 buf = malloc(cur_size);
1633 err = readn(fd, event, sizeof(struct perf_event_header));
1638 pr_err("failed to read event header\n");
1642 if (session->header.needs_swap)
1643 perf_event_header__bswap(&event->header);
1645 size = event->header.size;
1646 if (size < sizeof(struct perf_event_header)) {
1647 pr_err("bad event header size\n");
1651 if (size > cur_size) {
1652 void *new = realloc(buf, size);
1654 pr_err("failed to allocate memory to read event\n");
1662 p += sizeof(struct perf_event_header);
1664 if (size - sizeof(struct perf_event_header)) {
1665 err = readn(fd, p, size - sizeof(struct perf_event_header));
1668 pr_err("unexpected end of event stream\n");
1672 pr_err("failed to read event data\n");
1677 if ((skip = perf_session__process_event(session, event, head)) < 0) {
1678 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1679 head, event->header.size, event->header.type);
1689 if (!session_done())
1692 /* do the final flush for ordered samples */
1693 err = ordered_events__flush(oe, OE_FLUSH__FINAL);
1696 err = auxtrace__flush_events(session, tool);
1699 err = perf_session__flush_thread_stacks(session);
1702 perf_session__warn_about_errors(session);
1703 ordered_events__free(&session->ordered_events);
1704 auxtrace__free_events(session);
1708 static union perf_event *
1709 fetch_mmaped_event(struct perf_session *session,
1710 u64 head, size_t mmap_size, char *buf)
1712 union perf_event *event;
1715 * Ensure we have enough space remaining to read
1716 * the size of the event in the headers.
1718 if (head + sizeof(event->header) > mmap_size)
1721 event = (union perf_event *)(buf + head);
1723 if (session->header.needs_swap)
1724 perf_event_header__bswap(&event->header);
1726 if (head + event->header.size > mmap_size) {
1727 /* We're not fetching the event so swap back again */
1728 if (session->header.needs_swap)
1729 perf_event_header__bswap(&event->header);
1737 * On 64bit we can mmap the data file in one go. No need for tiny mmap
1738 * slices. On 32bit we use 32MB.
1740 #if BITS_PER_LONG == 64
1741 #define MMAP_SIZE ULLONG_MAX
1744 #define MMAP_SIZE (32 * 1024 * 1024ULL)
1745 #define NUM_MMAPS 128
1748 static int __perf_session__process_events(struct perf_session *session,
1749 u64 data_offset, u64 data_size,
1752 struct ordered_events *oe = &session->ordered_events;
1753 struct perf_tool *tool = session->tool;
1754 int fd = perf_data_file__fd(session->file);
1755 u64 head, page_offset, file_offset, file_pos, size;
1756 int err, mmap_prot, mmap_flags, map_idx = 0;
1758 char *buf, *mmaps[NUM_MMAPS];
1759 union perf_event *event;
1760 struct ui_progress prog;
1763 perf_tool__fill_defaults(tool);
1765 page_offset = page_size * (data_offset / page_size);
1766 file_offset = page_offset;
1767 head = data_offset - page_offset;
1772 if (data_offset + data_size < file_size)
1773 file_size = data_offset + data_size;
1775 ui_progress__init(&prog, file_size, "Processing events...");
1777 mmap_size = MMAP_SIZE;
1778 if (mmap_size > file_size) {
1779 mmap_size = file_size;
1780 session->one_mmap = true;
1783 memset(mmaps, 0, sizeof(mmaps));
1785 mmap_prot = PROT_READ;
1786 mmap_flags = MAP_SHARED;
1788 if (session->header.needs_swap) {
1789 mmap_prot |= PROT_WRITE;
1790 mmap_flags = MAP_PRIVATE;
1793 buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, fd,
1795 if (buf == MAP_FAILED) {
1796 pr_err("failed to mmap file\n");
1800 mmaps[map_idx] = buf;
1801 map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1);
1802 file_pos = file_offset + head;
1803 if (session->one_mmap) {
1804 session->one_mmap_addr = buf;
1805 session->one_mmap_offset = file_offset;
1809 event = fetch_mmaped_event(session, head, mmap_size, buf);
1811 if (mmaps[map_idx]) {
1812 munmap(mmaps[map_idx], mmap_size);
1813 mmaps[map_idx] = NULL;
1816 page_offset = page_size * (head / page_size);
1817 file_offset += page_offset;
1818 head -= page_offset;
1822 size = event->header.size;
1824 if (size < sizeof(struct perf_event_header) ||
1825 (skip = perf_session__process_event(session, event, file_pos)) < 0) {
1826 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1827 file_offset + head, event->header.size,
1828 event->header.type);
1839 ui_progress__update(&prog, size);
1844 if (file_pos < file_size)
1848 /* do the final flush for ordered samples */
1849 err = ordered_events__flush(oe, OE_FLUSH__FINAL);
1852 err = auxtrace__flush_events(session, tool);
1855 err = perf_session__flush_thread_stacks(session);
1857 ui_progress__finish();
1858 perf_session__warn_about_errors(session);
1860 * We may switching perf.data output, make ordered_events
1863 ordered_events__reinit(&session->ordered_events);
1864 auxtrace__free_events(session);
1865 session->one_mmap = false;
1869 int perf_session__process_events(struct perf_session *session)
1871 u64 size = perf_data_file__size(session->file);
1874 if (perf_session__register_idle_thread(session) < 0)
1877 if (!perf_data_file__is_pipe(session->file))
1878 err = __perf_session__process_events(session,
1879 session->header.data_offset,
1880 session->header.data_size, size);
1882 err = __perf_session__process_pipe_events(session);
1887 bool perf_session__has_traces(struct perf_session *session, const char *msg)
1889 struct perf_evsel *evsel;
1891 evlist__for_each_entry(session->evlist, evsel) {
1892 if (evsel->attr.type == PERF_TYPE_TRACEPOINT)
1896 pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg);
1900 int maps__set_kallsyms_ref_reloc_sym(struct map **maps,
1901 const char *symbol_name, u64 addr)
1905 struct ref_reloc_sym *ref;
1907 ref = zalloc(sizeof(struct ref_reloc_sym));
1911 ref->name = strdup(symbol_name);
1912 if (ref->name == NULL) {
1917 bracket = strchr(ref->name, ']');
1923 for (i = 0; i < MAP__NR_TYPES; ++i) {
1924 struct kmap *kmap = map__kmap(maps[i]);
1928 kmap->ref_reloc_sym = ref;
1934 size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp)
1936 return machines__fprintf_dsos(&session->machines, fp);
1939 size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp,
1940 bool (skip)(struct dso *dso, int parm), int parm)
1942 return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm);
1945 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp)
1948 const char *msg = "";
1950 if (perf_header__has_feat(&session->header, HEADER_AUXTRACE))
1951 msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)";
1953 ret = fprintf(fp, "\nAggregated stats:%s\n", msg);
1955 ret += events_stats__fprintf(&session->evlist->stats, fp);
1959 size_t perf_session__fprintf(struct perf_session *session, FILE *fp)
1962 * FIXME: Here we have to actually print all the machines in this
1963 * session, not just the host...
1965 return machine__fprintf(&session->machines.host, fp);
1968 struct perf_evsel *perf_session__find_first_evtype(struct perf_session *session,
1971 struct perf_evsel *pos;
1973 evlist__for_each_entry(session->evlist, pos) {
1974 if (pos->attr.type == type)
1980 int perf_session__cpu_bitmap(struct perf_session *session,
1981 const char *cpu_list, unsigned long *cpu_bitmap)
1984 struct cpu_map *map;
1986 for (i = 0; i < PERF_TYPE_MAX; ++i) {
1987 struct perf_evsel *evsel;
1989 evsel = perf_session__find_first_evtype(session, i);
1993 if (!(evsel->attr.sample_type & PERF_SAMPLE_CPU)) {
1994 pr_err("File does not contain CPU events. "
1995 "Remove -c option to proceed.\n");
2000 map = cpu_map__new(cpu_list);
2002 pr_err("Invalid cpu_list\n");
2006 for (i = 0; i < map->nr; i++) {
2007 int cpu = map->map[i];
2009 if (cpu >= MAX_NR_CPUS) {
2010 pr_err("Requested CPU %d too large. "
2011 "Consider raising MAX_NR_CPUS\n", cpu);
2012 goto out_delete_map;
2015 set_bit(cpu, cpu_bitmap);
2025 void perf_session__fprintf_info(struct perf_session *session, FILE *fp,
2028 if (session == NULL || fp == NULL)
2031 fprintf(fp, "# ========\n");
2032 perf_header__fprintf_info(session, fp, full);
2033 fprintf(fp, "# ========\n#\n");
2037 int __perf_session__set_tracepoints_handlers(struct perf_session *session,
2038 const struct perf_evsel_str_handler *assocs,
2041 struct perf_evsel *evsel;
2045 for (i = 0; i < nr_assocs; i++) {
2047 * Adding a handler for an event not in the session,
2050 evsel = perf_evlist__find_tracepoint_by_name(session->evlist, assocs[i].name);
2055 if (evsel->handler != NULL)
2057 evsel->handler = assocs[i].handler;
2065 int perf_event__process_id_index(struct perf_tool *tool __maybe_unused,
2066 union perf_event *event,
2067 struct perf_session *session)
2069 struct perf_evlist *evlist = session->evlist;
2070 struct id_index_event *ie = &event->id_index;
2071 size_t i, nr, max_nr;
2073 max_nr = (ie->header.size - sizeof(struct id_index_event)) /
2074 sizeof(struct id_index_entry);
2080 fprintf(stdout, " nr: %zu\n", nr);
2082 for (i = 0; i < nr; i++) {
2083 struct id_index_entry *e = &ie->entries[i];
2084 struct perf_sample_id *sid;
2087 fprintf(stdout, " ... id: %"PRIu64, e->id);
2088 fprintf(stdout, " idx: %"PRIu64, e->idx);
2089 fprintf(stdout, " cpu: %"PRId64, e->cpu);
2090 fprintf(stdout, " tid: %"PRId64"\n", e->tid);
2093 sid = perf_evlist__id2sid(evlist, e->id);
2103 int perf_event__synthesize_id_index(struct perf_tool *tool,
2104 perf_event__handler_t process,
2105 struct perf_evlist *evlist,
2106 struct machine *machine)
2108 union perf_event *ev;
2109 struct perf_evsel *evsel;
2110 size_t nr = 0, i = 0, sz, max_nr, n;
2113 pr_debug2("Synthesizing id index\n");
2115 max_nr = (UINT16_MAX - sizeof(struct id_index_event)) /
2116 sizeof(struct id_index_entry);
2118 evlist__for_each_entry(evlist, evsel)
2121 n = nr > max_nr ? max_nr : nr;
2122 sz = sizeof(struct id_index_event) + n * sizeof(struct id_index_entry);
2127 ev->id_index.header.type = PERF_RECORD_ID_INDEX;
2128 ev->id_index.header.size = sz;
2129 ev->id_index.nr = n;
2131 evlist__for_each_entry(evlist, evsel) {
2134 for (j = 0; j < evsel->ids; j++) {
2135 struct id_index_entry *e;
2136 struct perf_sample_id *sid;
2139 err = process(tool, ev, NULL, machine);
2146 e = &ev->id_index.entries[i++];
2148 e->id = evsel->id[j];
2150 sid = perf_evlist__id2sid(evlist, e->id);
2162 sz = sizeof(struct id_index_event) + nr * sizeof(struct id_index_entry);
2163 ev->id_index.header.size = sz;
2164 ev->id_index.nr = nr;
2166 err = process(tool, ev, NULL, machine);