1 #include <linux/kernel.h>
2 #include <traceevent/event-parse.h>
16 #include "perf_regs.h"
19 #include "thread-stack.h"
21 static int perf_session__deliver_event(struct perf_session *session,
22 union perf_event *event,
23 struct perf_sample *sample,
24 struct perf_tool *tool,
27 static int perf_session__open(struct perf_session *session)
29 struct perf_data_file *file = session->file;
31 if (perf_session__read_header(session) < 0) {
32 pr_err("incompatible file format (rerun with -v to learn more)\n");
36 if (perf_data_file__is_pipe(file))
39 if (!perf_evlist__valid_sample_type(session->evlist)) {
40 pr_err("non matching sample_type\n");
44 if (!perf_evlist__valid_sample_id_all(session->evlist)) {
45 pr_err("non matching sample_id_all\n");
49 if (!perf_evlist__valid_read_format(session->evlist)) {
50 pr_err("non matching read_format\n");
57 void perf_session__set_id_hdr_size(struct perf_session *session)
59 u16 id_hdr_size = perf_evlist__id_hdr_size(session->evlist);
61 machines__set_id_hdr_size(&session->machines, id_hdr_size);
64 int perf_session__create_kernel_maps(struct perf_session *session)
66 int ret = machine__create_kernel_maps(&session->machines.host);
69 ret = machines__create_guest_kernel_maps(&session->machines);
73 static void perf_session__destroy_kernel_maps(struct perf_session *session)
75 machines__destroy_kernel_maps(&session->machines);
78 static bool perf_session__has_comm_exec(struct perf_session *session)
80 struct perf_evsel *evsel;
82 evlist__for_each(session->evlist, evsel) {
83 if (evsel->attr.comm_exec)
90 static void perf_session__set_comm_exec(struct perf_session *session)
92 bool comm_exec = perf_session__has_comm_exec(session);
94 machines__set_comm_exec(&session->machines, comm_exec);
97 static int ordered_events__deliver_event(struct ordered_events *oe,
98 struct ordered_event *event)
100 struct perf_sample sample;
101 struct perf_session *session = container_of(oe, struct perf_session,
103 int ret = perf_evlist__parse_sample(session->evlist, event->event, &sample);
106 pr_err("Can't parse sample, err = %d\n", ret);
110 return perf_session__deliver_event(session, event->event, &sample,
111 session->tool, event->file_offset);
114 struct perf_session *perf_session__new(struct perf_data_file *file,
115 bool repipe, struct perf_tool *tool)
117 struct perf_session *session = zalloc(sizeof(*session));
122 session->repipe = repipe;
123 session->tool = tool;
124 INIT_LIST_HEAD(&session->auxtrace_index);
125 machines__init(&session->machines);
126 ordered_events__init(&session->ordered_events, ordered_events__deliver_event);
129 if (perf_data_file__open(file))
132 session->file = file;
134 if (perf_data_file__is_read(file)) {
135 if (perf_session__open(session) < 0)
139 * set session attributes that are present in perf.data
140 * but not in pipe-mode.
142 if (!file->is_pipe) {
143 perf_session__set_id_hdr_size(session);
144 perf_session__set_comm_exec(session);
148 session->machines.host.env = &perf_env;
151 if (!file || perf_data_file__is_write(file)) {
153 * In O_RDONLY mode this will be performed when reading the
154 * kernel MMAP event, in perf_event__process_mmap().
156 if (perf_session__create_kernel_maps(session) < 0)
157 pr_warning("Cannot read kernel map\n");
161 * In pipe-mode, evlist is empty until PERF_RECORD_HEADER_ATTR is
162 * processed, so perf_evlist__sample_id_all is not meaningful here.
164 if ((!file || !file->is_pipe) && tool && tool->ordering_requires_timestamps &&
165 tool->ordered_events && !perf_evlist__sample_id_all(session->evlist)) {
166 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n");
167 tool->ordered_events = false;
173 perf_data_file__close(file);
175 perf_session__delete(session);
180 static void perf_session__delete_threads(struct perf_session *session)
182 machine__delete_threads(&session->machines.host);
185 void perf_session__delete(struct perf_session *session)
187 auxtrace__free(session);
188 auxtrace_index__free(&session->auxtrace_index);
189 perf_session__destroy_kernel_maps(session);
190 perf_session__delete_threads(session);
191 perf_env__exit(&session->header.env);
192 machines__exit(&session->machines);
194 perf_data_file__close(session->file);
198 static int process_event_synth_tracing_data_stub(struct perf_tool *tool
200 union perf_event *event
202 struct perf_session *session
205 dump_printf(": unhandled!\n");
209 static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused,
210 union perf_event *event __maybe_unused,
211 struct perf_evlist **pevlist
214 dump_printf(": unhandled!\n");
218 static int process_event_sample_stub(struct perf_tool *tool __maybe_unused,
219 union perf_event *event __maybe_unused,
220 struct perf_sample *sample __maybe_unused,
221 struct perf_evsel *evsel __maybe_unused,
222 struct machine *machine __maybe_unused)
224 dump_printf(": unhandled!\n");
228 static int process_event_stub(struct perf_tool *tool __maybe_unused,
229 union perf_event *event __maybe_unused,
230 struct perf_sample *sample __maybe_unused,
231 struct machine *machine __maybe_unused)
233 dump_printf(": unhandled!\n");
237 static int process_build_id_stub(struct perf_tool *tool __maybe_unused,
238 union perf_event *event __maybe_unused,
239 struct perf_session *session __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 process_id_index_stub(struct perf_tool *tool __maybe_unused,
258 union perf_event *event __maybe_unused,
259 struct perf_session *perf_session
262 dump_printf(": unhandled!\n");
266 static int process_event_auxtrace_info_stub(struct perf_tool *tool __maybe_unused,
267 union perf_event *event __maybe_unused,
268 struct perf_session *session __maybe_unused)
270 dump_printf(": unhandled!\n");
274 static int skipn(int fd, off_t n)
280 ret = read(fd, buf, min(n, (off_t)sizeof(buf)));
289 static s64 process_event_auxtrace_stub(struct perf_tool *tool __maybe_unused,
290 union perf_event *event,
291 struct perf_session *session
294 dump_printf(": unhandled!\n");
295 if (perf_data_file__is_pipe(session->file))
296 skipn(perf_data_file__fd(session->file), event->auxtrace.size);
297 return event->auxtrace.size;
301 int process_event_auxtrace_error_stub(struct perf_tool *tool __maybe_unused,
302 union perf_event *event __maybe_unused,
303 struct perf_session *session __maybe_unused)
305 dump_printf(": unhandled!\n");
309 void perf_tool__fill_defaults(struct perf_tool *tool)
311 if (tool->sample == NULL)
312 tool->sample = process_event_sample_stub;
313 if (tool->mmap == NULL)
314 tool->mmap = process_event_stub;
315 if (tool->mmap2 == NULL)
316 tool->mmap2 = process_event_stub;
317 if (tool->comm == NULL)
318 tool->comm = process_event_stub;
319 if (tool->fork == NULL)
320 tool->fork = process_event_stub;
321 if (tool->exit == NULL)
322 tool->exit = process_event_stub;
323 if (tool->lost == NULL)
324 tool->lost = perf_event__process_lost;
325 if (tool->lost_samples == NULL)
326 tool->lost_samples = perf_event__process_lost_samples;
327 if (tool->aux == NULL)
328 tool->aux = perf_event__process_aux;
329 if (tool->itrace_start == NULL)
330 tool->itrace_start = perf_event__process_itrace_start;
331 if (tool->context_switch == NULL)
332 tool->context_switch = perf_event__process_switch;
333 if (tool->read == NULL)
334 tool->read = process_event_sample_stub;
335 if (tool->throttle == NULL)
336 tool->throttle = process_event_stub;
337 if (tool->unthrottle == NULL)
338 tool->unthrottle = process_event_stub;
339 if (tool->attr == NULL)
340 tool->attr = process_event_synth_attr_stub;
341 if (tool->tracing_data == NULL)
342 tool->tracing_data = process_event_synth_tracing_data_stub;
343 if (tool->build_id == NULL)
344 tool->build_id = process_build_id_stub;
345 if (tool->finished_round == NULL) {
346 if (tool->ordered_events)
347 tool->finished_round = process_finished_round;
349 tool->finished_round = process_finished_round_stub;
351 if (tool->id_index == NULL)
352 tool->id_index = process_id_index_stub;
353 if (tool->auxtrace_info == NULL)
354 tool->auxtrace_info = process_event_auxtrace_info_stub;
355 if (tool->auxtrace == NULL)
356 tool->auxtrace = process_event_auxtrace_stub;
357 if (tool->auxtrace_error == NULL)
358 tool->auxtrace_error = process_event_auxtrace_error_stub;
361 static void swap_sample_id_all(union perf_event *event, void *data)
363 void *end = (void *) event + event->header.size;
364 int size = end - data;
366 BUG_ON(size % sizeof(u64));
367 mem_bswap_64(data, size);
370 static void perf_event__all64_swap(union perf_event *event,
371 bool sample_id_all __maybe_unused)
373 struct perf_event_header *hdr = &event->header;
374 mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr));
377 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all)
379 event->comm.pid = bswap_32(event->comm.pid);
380 event->comm.tid = bswap_32(event->comm.tid);
383 void *data = &event->comm.comm;
385 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
386 swap_sample_id_all(event, data);
390 static void perf_event__mmap_swap(union perf_event *event,
393 event->mmap.pid = bswap_32(event->mmap.pid);
394 event->mmap.tid = bswap_32(event->mmap.tid);
395 event->mmap.start = bswap_64(event->mmap.start);
396 event->mmap.len = bswap_64(event->mmap.len);
397 event->mmap.pgoff = bswap_64(event->mmap.pgoff);
400 void *data = &event->mmap.filename;
402 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
403 swap_sample_id_all(event, data);
407 static void perf_event__mmap2_swap(union perf_event *event,
410 event->mmap2.pid = bswap_32(event->mmap2.pid);
411 event->mmap2.tid = bswap_32(event->mmap2.tid);
412 event->mmap2.start = bswap_64(event->mmap2.start);
413 event->mmap2.len = bswap_64(event->mmap2.len);
414 event->mmap2.pgoff = bswap_64(event->mmap2.pgoff);
415 event->mmap2.maj = bswap_32(event->mmap2.maj);
416 event->mmap2.min = bswap_32(event->mmap2.min);
417 event->mmap2.ino = bswap_64(event->mmap2.ino);
418 event->mmap2.ino_generation = bswap_64(event->mmap2.ino_generation);
421 void *data = &event->mmap2.filename;
423 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
424 swap_sample_id_all(event, data);
427 static void perf_event__task_swap(union perf_event *event, bool sample_id_all)
429 event->fork.pid = bswap_32(event->fork.pid);
430 event->fork.tid = bswap_32(event->fork.tid);
431 event->fork.ppid = bswap_32(event->fork.ppid);
432 event->fork.ptid = bswap_32(event->fork.ptid);
433 event->fork.time = bswap_64(event->fork.time);
436 swap_sample_id_all(event, &event->fork + 1);
439 static void perf_event__read_swap(union perf_event *event, bool sample_id_all)
441 event->read.pid = bswap_32(event->read.pid);
442 event->read.tid = bswap_32(event->read.tid);
443 event->read.value = bswap_64(event->read.value);
444 event->read.time_enabled = bswap_64(event->read.time_enabled);
445 event->read.time_running = bswap_64(event->read.time_running);
446 event->read.id = bswap_64(event->read.id);
449 swap_sample_id_all(event, &event->read + 1);
452 static void perf_event__aux_swap(union perf_event *event, bool sample_id_all)
454 event->aux.aux_offset = bswap_64(event->aux.aux_offset);
455 event->aux.aux_size = bswap_64(event->aux.aux_size);
456 event->aux.flags = bswap_64(event->aux.flags);
459 swap_sample_id_all(event, &event->aux + 1);
462 static void perf_event__itrace_start_swap(union perf_event *event,
465 event->itrace_start.pid = bswap_32(event->itrace_start.pid);
466 event->itrace_start.tid = bswap_32(event->itrace_start.tid);
469 swap_sample_id_all(event, &event->itrace_start + 1);
472 static void perf_event__switch_swap(union perf_event *event, bool sample_id_all)
474 if (event->header.type == PERF_RECORD_SWITCH_CPU_WIDE) {
475 event->context_switch.next_prev_pid =
476 bswap_32(event->context_switch.next_prev_pid);
477 event->context_switch.next_prev_tid =
478 bswap_32(event->context_switch.next_prev_tid);
482 swap_sample_id_all(event, &event->context_switch + 1);
485 static void perf_event__throttle_swap(union perf_event *event,
488 event->throttle.time = bswap_64(event->throttle.time);
489 event->throttle.id = bswap_64(event->throttle.id);
490 event->throttle.stream_id = bswap_64(event->throttle.stream_id);
493 swap_sample_id_all(event, &event->throttle + 1);
496 static u8 revbyte(u8 b)
498 int rev = (b >> 4) | ((b & 0xf) << 4);
499 rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2);
500 rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1);
505 * XXX this is hack in attempt to carry flags bitfield
506 * throught endian village. ABI says:
508 * Bit-fields are allocated from right to left (least to most significant)
509 * on little-endian implementations and from left to right (most to least
510 * significant) on big-endian implementations.
512 * The above seems to be byte specific, so we need to reverse each
513 * byte of the bitfield. 'Internet' also says this might be implementation
514 * specific and we probably need proper fix and carry perf_event_attr
515 * bitfield flags in separate data file FEAT_ section. Thought this seems
518 static void swap_bitfield(u8 *p, unsigned len)
522 for (i = 0; i < len; i++) {
528 /* exported for swapping attributes in file header */
529 void perf_event__attr_swap(struct perf_event_attr *attr)
531 attr->type = bswap_32(attr->type);
532 attr->size = bswap_32(attr->size);
534 #define bswap_safe(f, n) \
535 (attr->size > (offsetof(struct perf_event_attr, f) + \
536 sizeof(attr->f) * (n)))
537 #define bswap_field(f, sz) \
539 if (bswap_safe(f, 0)) \
540 attr->f = bswap_##sz(attr->f); \
542 #define bswap_field_32(f) bswap_field(f, 32)
543 #define bswap_field_64(f) bswap_field(f, 64)
545 bswap_field_64(config);
546 bswap_field_64(sample_period);
547 bswap_field_64(sample_type);
548 bswap_field_64(read_format);
549 bswap_field_32(wakeup_events);
550 bswap_field_32(bp_type);
551 bswap_field_64(bp_addr);
552 bswap_field_64(bp_len);
553 bswap_field_64(branch_sample_type);
554 bswap_field_64(sample_regs_user);
555 bswap_field_32(sample_stack_user);
556 bswap_field_32(aux_watermark);
559 * After read_format are bitfields. Check read_format because
560 * we are unable to use offsetof on bitfield.
562 if (bswap_safe(read_format, 1))
563 swap_bitfield((u8 *) (&attr->read_format + 1),
565 #undef bswap_field_64
566 #undef bswap_field_32
571 static void perf_event__hdr_attr_swap(union perf_event *event,
572 bool sample_id_all __maybe_unused)
576 perf_event__attr_swap(&event->attr.attr);
578 size = event->header.size;
579 size -= (void *)&event->attr.id - (void *)event;
580 mem_bswap_64(event->attr.id, size);
583 static void perf_event__event_type_swap(union perf_event *event,
584 bool sample_id_all __maybe_unused)
586 event->event_type.event_type.event_id =
587 bswap_64(event->event_type.event_type.event_id);
590 static void perf_event__tracing_data_swap(union perf_event *event,
591 bool sample_id_all __maybe_unused)
593 event->tracing_data.size = bswap_32(event->tracing_data.size);
596 static void perf_event__auxtrace_info_swap(union perf_event *event,
597 bool sample_id_all __maybe_unused)
601 event->auxtrace_info.type = bswap_32(event->auxtrace_info.type);
603 size = event->header.size;
604 size -= (void *)&event->auxtrace_info.priv - (void *)event;
605 mem_bswap_64(event->auxtrace_info.priv, size);
608 static void perf_event__auxtrace_swap(union perf_event *event,
609 bool sample_id_all __maybe_unused)
611 event->auxtrace.size = bswap_64(event->auxtrace.size);
612 event->auxtrace.offset = bswap_64(event->auxtrace.offset);
613 event->auxtrace.reference = bswap_64(event->auxtrace.reference);
614 event->auxtrace.idx = bswap_32(event->auxtrace.idx);
615 event->auxtrace.tid = bswap_32(event->auxtrace.tid);
616 event->auxtrace.cpu = bswap_32(event->auxtrace.cpu);
619 static void perf_event__auxtrace_error_swap(union perf_event *event,
620 bool sample_id_all __maybe_unused)
622 event->auxtrace_error.type = bswap_32(event->auxtrace_error.type);
623 event->auxtrace_error.code = bswap_32(event->auxtrace_error.code);
624 event->auxtrace_error.cpu = bswap_32(event->auxtrace_error.cpu);
625 event->auxtrace_error.pid = bswap_32(event->auxtrace_error.pid);
626 event->auxtrace_error.tid = bswap_32(event->auxtrace_error.tid);
627 event->auxtrace_error.ip = bswap_64(event->auxtrace_error.ip);
630 typedef void (*perf_event__swap_op)(union perf_event *event,
633 static perf_event__swap_op perf_event__swap_ops[] = {
634 [PERF_RECORD_MMAP] = perf_event__mmap_swap,
635 [PERF_RECORD_MMAP2] = perf_event__mmap2_swap,
636 [PERF_RECORD_COMM] = perf_event__comm_swap,
637 [PERF_RECORD_FORK] = perf_event__task_swap,
638 [PERF_RECORD_EXIT] = perf_event__task_swap,
639 [PERF_RECORD_LOST] = perf_event__all64_swap,
640 [PERF_RECORD_READ] = perf_event__read_swap,
641 [PERF_RECORD_THROTTLE] = perf_event__throttle_swap,
642 [PERF_RECORD_UNTHROTTLE] = perf_event__throttle_swap,
643 [PERF_RECORD_SAMPLE] = perf_event__all64_swap,
644 [PERF_RECORD_AUX] = perf_event__aux_swap,
645 [PERF_RECORD_ITRACE_START] = perf_event__itrace_start_swap,
646 [PERF_RECORD_LOST_SAMPLES] = perf_event__all64_swap,
647 [PERF_RECORD_SWITCH] = perf_event__switch_swap,
648 [PERF_RECORD_SWITCH_CPU_WIDE] = perf_event__switch_swap,
649 [PERF_RECORD_HEADER_ATTR] = perf_event__hdr_attr_swap,
650 [PERF_RECORD_HEADER_EVENT_TYPE] = perf_event__event_type_swap,
651 [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
652 [PERF_RECORD_HEADER_BUILD_ID] = NULL,
653 [PERF_RECORD_ID_INDEX] = perf_event__all64_swap,
654 [PERF_RECORD_AUXTRACE_INFO] = perf_event__auxtrace_info_swap,
655 [PERF_RECORD_AUXTRACE] = perf_event__auxtrace_swap,
656 [PERF_RECORD_AUXTRACE_ERROR] = perf_event__auxtrace_error_swap,
657 [PERF_RECORD_HEADER_MAX] = NULL,
661 * When perf record finishes a pass on every buffers, it records this pseudo
663 * We record the max timestamp t found in the pass n.
664 * Assuming these timestamps are monotonic across cpus, we know that if
665 * a buffer still has events with timestamps below t, they will be all
666 * available and then read in the pass n + 1.
667 * Hence when we start to read the pass n + 2, we can safely flush every
668 * events with timestamps below t.
670 * ============ PASS n =================
673 * cnt1 timestamps | cnt2 timestamps
676 * - | 4 <--- max recorded
678 * ============ PASS n + 1 ==============
681 * cnt1 timestamps | cnt2 timestamps
684 * 5 | 7 <---- max recorded
686 * Flush every events below timestamp 4
688 * ============ PASS n + 2 ==============
691 * cnt1 timestamps | cnt2 timestamps
696 * Flush every events below timestamp 7
699 static int process_finished_round(struct perf_tool *tool __maybe_unused,
700 union perf_event *event __maybe_unused,
701 struct ordered_events *oe)
704 fprintf(stdout, "\n");
705 return ordered_events__flush(oe, OE_FLUSH__ROUND);
708 int perf_session__queue_event(struct perf_session *s, union perf_event *event,
709 struct perf_sample *sample, u64 file_offset)
711 return ordered_events__queue(&s->ordered_events, event, sample, file_offset);
714 static void callchain__lbr_callstack_printf(struct perf_sample *sample)
716 struct ip_callchain *callchain = sample->callchain;
717 struct branch_stack *lbr_stack = sample->branch_stack;
718 u64 kernel_callchain_nr = callchain->nr;
721 for (i = 0; i < kernel_callchain_nr; i++) {
722 if (callchain->ips[i] == PERF_CONTEXT_USER)
726 if ((i != kernel_callchain_nr) && lbr_stack->nr) {
729 * LBR callstack can only get user call chain,
730 * i is kernel call chain number,
731 * 1 is PERF_CONTEXT_USER.
733 * The user call chain is stored in LBR registers.
734 * LBR are pair registers. The caller is stored
735 * in "from" register, while the callee is stored
737 * For example, there is a call stack
738 * "A"->"B"->"C"->"D".
739 * The LBR registers will recorde like
740 * "C"->"D", "B"->"C", "A"->"B".
741 * So only the first "to" register and all "from"
742 * registers are needed to construct the whole stack.
744 total_nr = i + 1 + lbr_stack->nr + 1;
745 kernel_callchain_nr = i + 1;
747 printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr);
749 for (i = 0; i < kernel_callchain_nr; i++)
750 printf("..... %2d: %016" PRIx64 "\n",
751 i, callchain->ips[i]);
753 printf("..... %2d: %016" PRIx64 "\n",
754 (int)(kernel_callchain_nr), lbr_stack->entries[0].to);
755 for (i = 0; i < lbr_stack->nr; i++)
756 printf("..... %2d: %016" PRIx64 "\n",
757 (int)(i + kernel_callchain_nr + 1), lbr_stack->entries[i].from);
761 static void callchain__printf(struct perf_evsel *evsel,
762 struct perf_sample *sample)
765 struct ip_callchain *callchain = sample->callchain;
767 if (has_branch_callstack(evsel))
768 callchain__lbr_callstack_printf(sample);
770 printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr);
772 for (i = 0; i < callchain->nr; i++)
773 printf("..... %2d: %016" PRIx64 "\n",
774 i, callchain->ips[i]);
777 static void branch_stack__printf(struct perf_sample *sample)
781 printf("... branch stack: nr:%" PRIu64 "\n", sample->branch_stack->nr);
783 for (i = 0; i < sample->branch_stack->nr; i++) {
784 struct branch_entry *e = &sample->branch_stack->entries[i];
786 printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x\n",
789 e->flags.mispred ? "M" : " ",
790 e->flags.predicted ? "P" : " ",
791 e->flags.abort ? "A" : " ",
792 e->flags.in_tx ? "T" : " ",
793 (unsigned)e->flags.reserved);
797 static void regs_dump__printf(u64 mask, u64 *regs)
801 for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) {
804 printf(".... %-5s 0x%" PRIx64 "\n",
805 perf_reg_name(rid), val);
809 static const char *regs_abi[] = {
810 [PERF_SAMPLE_REGS_ABI_NONE] = "none",
811 [PERF_SAMPLE_REGS_ABI_32] = "32-bit",
812 [PERF_SAMPLE_REGS_ABI_64] = "64-bit",
815 static inline const char *regs_dump_abi(struct regs_dump *d)
817 if (d->abi > PERF_SAMPLE_REGS_ABI_64)
820 return regs_abi[d->abi];
823 static void regs__printf(const char *type, struct regs_dump *regs)
825 u64 mask = regs->mask;
827 printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n",
830 regs_dump_abi(regs));
832 regs_dump__printf(mask, regs->regs);
835 static void regs_user__printf(struct perf_sample *sample)
837 struct regs_dump *user_regs = &sample->user_regs;
840 regs__printf("user", user_regs);
843 static void regs_intr__printf(struct perf_sample *sample)
845 struct regs_dump *intr_regs = &sample->intr_regs;
848 regs__printf("intr", intr_regs);
851 static void stack_user__printf(struct stack_dump *dump)
853 printf("... ustack: size %" PRIu64 ", offset 0x%x\n",
854 dump->size, dump->offset);
857 static void perf_evlist__print_tstamp(struct perf_evlist *evlist,
858 union perf_event *event,
859 struct perf_sample *sample)
861 u64 sample_type = __perf_evlist__combined_sample_type(evlist);
863 if (event->header.type != PERF_RECORD_SAMPLE &&
864 !perf_evlist__sample_id_all(evlist)) {
865 fputs("-1 -1 ", stdout);
869 if ((sample_type & PERF_SAMPLE_CPU))
870 printf("%u ", sample->cpu);
872 if (sample_type & PERF_SAMPLE_TIME)
873 printf("%" PRIu64 " ", sample->time);
876 static void sample_read__printf(struct perf_sample *sample, u64 read_format)
878 printf("... sample_read:\n");
880 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
881 printf("...... time enabled %016" PRIx64 "\n",
882 sample->read.time_enabled);
884 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
885 printf("...... time running %016" PRIx64 "\n",
886 sample->read.time_running);
888 if (read_format & PERF_FORMAT_GROUP) {
891 printf(".... group nr %" PRIu64 "\n", sample->read.group.nr);
893 for (i = 0; i < sample->read.group.nr; i++) {
894 struct sample_read_value *value;
896 value = &sample->read.group.values[i];
897 printf("..... id %016" PRIx64
898 ", value %016" PRIx64 "\n",
899 value->id, value->value);
902 printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n",
903 sample->read.one.id, sample->read.one.value);
906 static void dump_event(struct perf_evlist *evlist, union perf_event *event,
907 u64 file_offset, struct perf_sample *sample)
912 printf("\n%#" PRIx64 " [%#x]: event: %d\n",
913 file_offset, event->header.size, event->header.type);
918 perf_evlist__print_tstamp(evlist, event, sample);
920 printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset,
921 event->header.size, perf_event__name(event->header.type));
924 static void dump_sample(struct perf_evsel *evsel, union perf_event *event,
925 struct perf_sample *sample)
932 printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n",
933 event->header.misc, sample->pid, sample->tid, sample->ip,
934 sample->period, sample->addr);
936 sample_type = evsel->attr.sample_type;
938 if (sample_type & PERF_SAMPLE_CALLCHAIN)
939 callchain__printf(evsel, sample);
941 if ((sample_type & PERF_SAMPLE_BRANCH_STACK) && !has_branch_callstack(evsel))
942 branch_stack__printf(sample);
944 if (sample_type & PERF_SAMPLE_REGS_USER)
945 regs_user__printf(sample);
947 if (sample_type & PERF_SAMPLE_REGS_INTR)
948 regs_intr__printf(sample);
950 if (sample_type & PERF_SAMPLE_STACK_USER)
951 stack_user__printf(&sample->user_stack);
953 if (sample_type & PERF_SAMPLE_WEIGHT)
954 printf("... weight: %" PRIu64 "\n", sample->weight);
956 if (sample_type & PERF_SAMPLE_DATA_SRC)
957 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src);
959 if (sample_type & PERF_SAMPLE_TRANSACTION)
960 printf("... transaction: %" PRIx64 "\n", sample->transaction);
962 if (sample_type & PERF_SAMPLE_READ)
963 sample_read__printf(sample, evsel->attr.read_format);
966 static struct machine *machines__find_for_cpumode(struct machines *machines,
967 union perf_event *event,
968 struct perf_sample *sample)
970 const u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
971 struct machine *machine;
974 ((cpumode == PERF_RECORD_MISC_GUEST_KERNEL) ||
975 (cpumode == PERF_RECORD_MISC_GUEST_USER))) {
978 if (event->header.type == PERF_RECORD_MMAP
979 || event->header.type == PERF_RECORD_MMAP2)
980 pid = event->mmap.pid;
984 machine = machines__find(machines, pid);
986 machine = machines__findnew(machines, DEFAULT_GUEST_KERNEL_ID);
990 return &machines->host;
993 static int deliver_sample_value(struct perf_evlist *evlist,
994 struct perf_tool *tool,
995 union perf_event *event,
996 struct perf_sample *sample,
997 struct sample_read_value *v,
998 struct machine *machine)
1000 struct perf_sample_id *sid = perf_evlist__id2sid(evlist, v->id);
1004 sample->period = v->value - sid->period;
1005 sid->period = v->value;
1008 if (!sid || sid->evsel == NULL) {
1009 ++evlist->stats.nr_unknown_id;
1013 return tool->sample(tool, event, sample, sid->evsel, machine);
1016 static int deliver_sample_group(struct perf_evlist *evlist,
1017 struct perf_tool *tool,
1018 union perf_event *event,
1019 struct perf_sample *sample,
1020 struct machine *machine)
1025 for (i = 0; i < sample->read.group.nr; i++) {
1026 ret = deliver_sample_value(evlist, tool, event, sample,
1027 &sample->read.group.values[i],
1037 perf_evlist__deliver_sample(struct perf_evlist *evlist,
1038 struct perf_tool *tool,
1039 union perf_event *event,
1040 struct perf_sample *sample,
1041 struct perf_evsel *evsel,
1042 struct machine *machine)
1044 /* We know evsel != NULL. */
1045 u64 sample_type = evsel->attr.sample_type;
1046 u64 read_format = evsel->attr.read_format;
1048 /* Standard sample delievery. */
1049 if (!(sample_type & PERF_SAMPLE_READ))
1050 return tool->sample(tool, event, sample, evsel, machine);
1052 /* For PERF_SAMPLE_READ we have either single or group mode. */
1053 if (read_format & PERF_FORMAT_GROUP)
1054 return deliver_sample_group(evlist, tool, event, sample,
1057 return deliver_sample_value(evlist, tool, event, sample,
1058 &sample->read.one, machine);
1061 static int machines__deliver_event(struct machines *machines,
1062 struct perf_evlist *evlist,
1063 union perf_event *event,
1064 struct perf_sample *sample,
1065 struct perf_tool *tool, u64 file_offset)
1067 struct perf_evsel *evsel;
1068 struct machine *machine;
1070 dump_event(evlist, event, file_offset, sample);
1072 evsel = perf_evlist__id2evsel(evlist, sample->id);
1074 machine = machines__find_for_cpumode(machines, event, sample);
1076 switch (event->header.type) {
1077 case PERF_RECORD_SAMPLE:
1078 if (evsel == NULL) {
1079 ++evlist->stats.nr_unknown_id;
1082 dump_sample(evsel, event, sample);
1083 if (machine == NULL) {
1084 ++evlist->stats.nr_unprocessable_samples;
1087 return perf_evlist__deliver_sample(evlist, tool, event, sample, evsel, machine);
1088 case PERF_RECORD_MMAP:
1089 return tool->mmap(tool, event, sample, machine);
1090 case PERF_RECORD_MMAP2:
1091 if (event->header.misc & PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT)
1092 ++evlist->stats.nr_proc_map_timeout;
1093 return tool->mmap2(tool, event, sample, machine);
1094 case PERF_RECORD_COMM:
1095 return tool->comm(tool, event, sample, machine);
1096 case PERF_RECORD_FORK:
1097 return tool->fork(tool, event, sample, machine);
1098 case PERF_RECORD_EXIT:
1099 return tool->exit(tool, event, sample, machine);
1100 case PERF_RECORD_LOST:
1101 if (tool->lost == perf_event__process_lost)
1102 evlist->stats.total_lost += event->lost.lost;
1103 return tool->lost(tool, event, sample, machine);
1104 case PERF_RECORD_LOST_SAMPLES:
1105 if (tool->lost_samples == perf_event__process_lost_samples)
1106 evlist->stats.total_lost_samples += event->lost_samples.lost;
1107 return tool->lost_samples(tool, event, sample, machine);
1108 case PERF_RECORD_READ:
1109 return tool->read(tool, event, sample, evsel, machine);
1110 case PERF_RECORD_THROTTLE:
1111 return tool->throttle(tool, event, sample, machine);
1112 case PERF_RECORD_UNTHROTTLE:
1113 return tool->unthrottle(tool, event, sample, machine);
1114 case PERF_RECORD_AUX:
1115 if (tool->aux == perf_event__process_aux &&
1116 (event->aux.flags & PERF_AUX_FLAG_TRUNCATED))
1117 evlist->stats.total_aux_lost += 1;
1118 return tool->aux(tool, event, sample, machine);
1119 case PERF_RECORD_ITRACE_START:
1120 return tool->itrace_start(tool, event, sample, machine);
1121 case PERF_RECORD_SWITCH:
1122 case PERF_RECORD_SWITCH_CPU_WIDE:
1123 return tool->context_switch(tool, event, sample, machine);
1125 ++evlist->stats.nr_unknown_events;
1130 static int perf_session__deliver_event(struct perf_session *session,
1131 union perf_event *event,
1132 struct perf_sample *sample,
1133 struct perf_tool *tool,
1138 ret = auxtrace__process_event(session, event, sample, tool);
1144 return machines__deliver_event(&session->machines, session->evlist,
1145 event, sample, tool, file_offset);
1148 static s64 perf_session__process_user_event(struct perf_session *session,
1149 union perf_event *event,
1152 struct ordered_events *oe = &session->ordered_events;
1153 struct perf_tool *tool = session->tool;
1154 int fd = perf_data_file__fd(session->file);
1157 dump_event(session->evlist, event, file_offset, NULL);
1159 /* These events are processed right away */
1160 switch (event->header.type) {
1161 case PERF_RECORD_HEADER_ATTR:
1162 err = tool->attr(tool, event, &session->evlist);
1164 perf_session__set_id_hdr_size(session);
1165 perf_session__set_comm_exec(session);
1168 case PERF_RECORD_HEADER_EVENT_TYPE:
1170 * Depreceated, but we need to handle it for sake
1171 * of old data files create in pipe mode.
1174 case PERF_RECORD_HEADER_TRACING_DATA:
1175 /* setup for reading amidst mmap */
1176 lseek(fd, file_offset, SEEK_SET);
1177 return tool->tracing_data(tool, event, session);
1178 case PERF_RECORD_HEADER_BUILD_ID:
1179 return tool->build_id(tool, event, session);
1180 case PERF_RECORD_FINISHED_ROUND:
1181 return tool->finished_round(tool, event, oe);
1182 case PERF_RECORD_ID_INDEX:
1183 return tool->id_index(tool, event, session);
1184 case PERF_RECORD_AUXTRACE_INFO:
1185 return tool->auxtrace_info(tool, event, session);
1186 case PERF_RECORD_AUXTRACE:
1187 /* setup for reading amidst mmap */
1188 lseek(fd, file_offset + event->header.size, SEEK_SET);
1189 return tool->auxtrace(tool, event, session);
1190 case PERF_RECORD_AUXTRACE_ERROR:
1191 perf_session__auxtrace_error_inc(session, event);
1192 return tool->auxtrace_error(tool, event, session);
1198 int perf_session__deliver_synth_event(struct perf_session *session,
1199 union perf_event *event,
1200 struct perf_sample *sample)
1202 struct perf_evlist *evlist = session->evlist;
1203 struct perf_tool *tool = session->tool;
1205 events_stats__inc(&evlist->stats, event->header.type);
1207 if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1208 return perf_session__process_user_event(session, event, 0);
1210 return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0);
1213 static void event_swap(union perf_event *event, bool sample_id_all)
1215 perf_event__swap_op swap;
1217 swap = perf_event__swap_ops[event->header.type];
1219 swap(event, sample_id_all);
1222 int perf_session__peek_event(struct perf_session *session, off_t file_offset,
1223 void *buf, size_t buf_sz,
1224 union perf_event **event_ptr,
1225 struct perf_sample *sample)
1227 union perf_event *event;
1228 size_t hdr_sz, rest;
1231 if (session->one_mmap && !session->header.needs_swap) {
1232 event = file_offset - session->one_mmap_offset +
1233 session->one_mmap_addr;
1234 goto out_parse_sample;
1237 if (perf_data_file__is_pipe(session->file))
1240 fd = perf_data_file__fd(session->file);
1241 hdr_sz = sizeof(struct perf_event_header);
1243 if (buf_sz < hdr_sz)
1246 if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 ||
1247 readn(fd, buf, hdr_sz) != (ssize_t)hdr_sz)
1250 event = (union perf_event *)buf;
1252 if (session->header.needs_swap)
1253 perf_event_header__bswap(&event->header);
1255 if (event->header.size < hdr_sz || event->header.size > buf_sz)
1259 rest = event->header.size - hdr_sz;
1261 if (readn(fd, buf, rest) != (ssize_t)rest)
1264 if (session->header.needs_swap)
1265 event_swap(event, perf_evlist__sample_id_all(session->evlist));
1269 if (sample && event->header.type < PERF_RECORD_USER_TYPE_START &&
1270 perf_evlist__parse_sample(session->evlist, event, sample))
1278 static s64 perf_session__process_event(struct perf_session *session,
1279 union perf_event *event, u64 file_offset)
1281 struct perf_evlist *evlist = session->evlist;
1282 struct perf_tool *tool = session->tool;
1283 struct perf_sample sample;
1286 if (session->header.needs_swap)
1287 event_swap(event, perf_evlist__sample_id_all(evlist));
1289 if (event->header.type >= PERF_RECORD_HEADER_MAX)
1292 events_stats__inc(&evlist->stats, event->header.type);
1294 if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1295 return perf_session__process_user_event(session, event, file_offset);
1298 * For all kernel events we get the sample data
1300 ret = perf_evlist__parse_sample(evlist, event, &sample);
1304 if (tool->ordered_events) {
1305 ret = perf_session__queue_event(session, event, &sample, file_offset);
1310 return perf_session__deliver_event(session, event, &sample, tool,
1314 void perf_event_header__bswap(struct perf_event_header *hdr)
1316 hdr->type = bswap_32(hdr->type);
1317 hdr->misc = bswap_16(hdr->misc);
1318 hdr->size = bswap_16(hdr->size);
1321 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid)
1323 return machine__findnew_thread(&session->machines.host, -1, pid);
1326 struct thread *perf_session__register_idle_thread(struct perf_session *session)
1328 struct thread *thread;
1330 thread = machine__findnew_thread(&session->machines.host, 0, 0);
1331 if (thread == NULL || thread__set_comm(thread, "swapper", 0)) {
1332 pr_err("problem inserting idle task.\n");
1339 static void perf_session__warn_about_errors(const struct perf_session *session)
1341 const struct events_stats *stats = &session->evlist->stats;
1342 const struct ordered_events *oe = &session->ordered_events;
1344 if (session->tool->lost == perf_event__process_lost &&
1345 stats->nr_events[PERF_RECORD_LOST] != 0) {
1346 ui__warning("Processed %d events and lost %d chunks!\n\n"
1347 "Check IO/CPU overload!\n\n",
1348 stats->nr_events[0],
1349 stats->nr_events[PERF_RECORD_LOST]);
1352 if (session->tool->lost_samples == perf_event__process_lost_samples) {
1355 drop_rate = (double)stats->total_lost_samples /
1356 (double) (stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples);
1357 if (drop_rate > 0.05) {
1358 ui__warning("Processed %" PRIu64 " samples and lost %3.2f%% samples!\n\n",
1359 stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples,
1364 if (session->tool->aux == perf_event__process_aux &&
1365 stats->total_aux_lost != 0) {
1366 ui__warning("AUX data lost %" PRIu64 " times out of %u!\n\n",
1367 stats->total_aux_lost,
1368 stats->nr_events[PERF_RECORD_AUX]);
1371 if (stats->nr_unknown_events != 0) {
1372 ui__warning("Found %u unknown events!\n\n"
1373 "Is this an older tool processing a perf.data "
1374 "file generated by a more recent tool?\n\n"
1375 "If that is not the case, consider "
1376 "reporting to linux-kernel@vger.kernel.org.\n\n",
1377 stats->nr_unknown_events);
1380 if (stats->nr_unknown_id != 0) {
1381 ui__warning("%u samples with id not present in the header\n",
1382 stats->nr_unknown_id);
1385 if (stats->nr_invalid_chains != 0) {
1386 ui__warning("Found invalid callchains!\n\n"
1387 "%u out of %u events were discarded for this reason.\n\n"
1388 "Consider reporting to linux-kernel@vger.kernel.org.\n\n",
1389 stats->nr_invalid_chains,
1390 stats->nr_events[PERF_RECORD_SAMPLE]);
1393 if (stats->nr_unprocessable_samples != 0) {
1394 ui__warning("%u unprocessable samples recorded.\n"
1395 "Do you have a KVM guest running and not using 'perf kvm'?\n",
1396 stats->nr_unprocessable_samples);
1399 if (oe->nr_unordered_events != 0)
1400 ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events);
1402 events_stats__auxtrace_error_warn(stats);
1404 if (stats->nr_proc_map_timeout != 0) {
1405 ui__warning("%d map information files for pre-existing threads were\n"
1406 "not processed, if there are samples for addresses they\n"
1407 "will not be resolved, you may find out which are these\n"
1408 "threads by running with -v and redirecting the output\n"
1410 "The time limit to process proc map is too short?\n"
1411 "Increase it by --proc-map-timeout\n",
1412 stats->nr_proc_map_timeout);
1416 static int perf_session__flush_thread_stack(struct thread *thread,
1417 void *p __maybe_unused)
1419 return thread_stack__flush(thread);
1422 static int perf_session__flush_thread_stacks(struct perf_session *session)
1424 return machines__for_each_thread(&session->machines,
1425 perf_session__flush_thread_stack,
1429 volatile int session_done;
1431 static int __perf_session__process_pipe_events(struct perf_session *session)
1433 struct ordered_events *oe = &session->ordered_events;
1434 struct perf_tool *tool = session->tool;
1435 int fd = perf_data_file__fd(session->file);
1436 union perf_event *event;
1437 uint32_t size, cur_size = 0;
1444 perf_tool__fill_defaults(tool);
1447 cur_size = sizeof(union perf_event);
1449 buf = malloc(cur_size);
1452 ordered_events__set_copy_on_queue(oe, true);
1455 err = readn(fd, event, sizeof(struct perf_event_header));
1460 pr_err("failed to read event header\n");
1464 if (session->header.needs_swap)
1465 perf_event_header__bswap(&event->header);
1467 size = event->header.size;
1468 if (size < sizeof(struct perf_event_header)) {
1469 pr_err("bad event header size\n");
1473 if (size > cur_size) {
1474 void *new = realloc(buf, size);
1476 pr_err("failed to allocate memory to read event\n");
1484 p += sizeof(struct perf_event_header);
1486 if (size - sizeof(struct perf_event_header)) {
1487 err = readn(fd, p, size - sizeof(struct perf_event_header));
1490 pr_err("unexpected end of event stream\n");
1494 pr_err("failed to read event data\n");
1499 if ((skip = perf_session__process_event(session, event, head)) < 0) {
1500 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1501 head, event->header.size, event->header.type);
1511 if (!session_done())
1514 /* do the final flush for ordered samples */
1515 err = ordered_events__flush(oe, OE_FLUSH__FINAL);
1518 err = auxtrace__flush_events(session, tool);
1521 err = perf_session__flush_thread_stacks(session);
1524 perf_session__warn_about_errors(session);
1525 ordered_events__free(&session->ordered_events);
1526 auxtrace__free_events(session);
1530 static union perf_event *
1531 fetch_mmaped_event(struct perf_session *session,
1532 u64 head, size_t mmap_size, char *buf)
1534 union perf_event *event;
1537 * Ensure we have enough space remaining to read
1538 * the size of the event in the headers.
1540 if (head + sizeof(event->header) > mmap_size)
1543 event = (union perf_event *)(buf + head);
1545 if (session->header.needs_swap)
1546 perf_event_header__bswap(&event->header);
1548 if (head + event->header.size > mmap_size) {
1549 /* We're not fetching the event so swap back again */
1550 if (session->header.needs_swap)
1551 perf_event_header__bswap(&event->header);
1559 * On 64bit we can mmap the data file in one go. No need for tiny mmap
1560 * slices. On 32bit we use 32MB.
1562 #if BITS_PER_LONG == 64
1563 #define MMAP_SIZE ULLONG_MAX
1566 #define MMAP_SIZE (32 * 1024 * 1024ULL)
1567 #define NUM_MMAPS 128
1570 static int __perf_session__process_events(struct perf_session *session,
1571 u64 data_offset, u64 data_size,
1574 struct ordered_events *oe = &session->ordered_events;
1575 struct perf_tool *tool = session->tool;
1576 int fd = perf_data_file__fd(session->file);
1577 u64 head, page_offset, file_offset, file_pos, size;
1578 int err, mmap_prot, mmap_flags, map_idx = 0;
1580 char *buf, *mmaps[NUM_MMAPS];
1581 union perf_event *event;
1582 struct ui_progress prog;
1585 perf_tool__fill_defaults(tool);
1587 page_offset = page_size * (data_offset / page_size);
1588 file_offset = page_offset;
1589 head = data_offset - page_offset;
1594 if (data_offset + data_size < file_size)
1595 file_size = data_offset + data_size;
1597 ui_progress__init(&prog, file_size, "Processing events...");
1599 mmap_size = MMAP_SIZE;
1600 if (mmap_size > file_size) {
1601 mmap_size = file_size;
1602 session->one_mmap = true;
1605 memset(mmaps, 0, sizeof(mmaps));
1607 mmap_prot = PROT_READ;
1608 mmap_flags = MAP_SHARED;
1610 if (session->header.needs_swap) {
1611 mmap_prot |= PROT_WRITE;
1612 mmap_flags = MAP_PRIVATE;
1615 buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, fd,
1617 if (buf == MAP_FAILED) {
1618 pr_err("failed to mmap file\n");
1622 mmaps[map_idx] = buf;
1623 map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1);
1624 file_pos = file_offset + head;
1625 if (session->one_mmap) {
1626 session->one_mmap_addr = buf;
1627 session->one_mmap_offset = file_offset;
1631 event = fetch_mmaped_event(session, head, mmap_size, buf);
1633 if (mmaps[map_idx]) {
1634 munmap(mmaps[map_idx], mmap_size);
1635 mmaps[map_idx] = NULL;
1638 page_offset = page_size * (head / page_size);
1639 file_offset += page_offset;
1640 head -= page_offset;
1644 size = event->header.size;
1646 if (size < sizeof(struct perf_event_header) ||
1647 (skip = perf_session__process_event(session, event, file_pos)) < 0) {
1648 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1649 file_offset + head, event->header.size,
1650 event->header.type);
1661 ui_progress__update(&prog, size);
1666 if (file_pos < file_size)
1670 /* do the final flush for ordered samples */
1671 err = ordered_events__flush(oe, OE_FLUSH__FINAL);
1674 err = auxtrace__flush_events(session, tool);
1677 err = perf_session__flush_thread_stacks(session);
1679 ui_progress__finish();
1680 perf_session__warn_about_errors(session);
1681 ordered_events__free(&session->ordered_events);
1682 auxtrace__free_events(session);
1683 session->one_mmap = false;
1687 int perf_session__process_events(struct perf_session *session)
1689 u64 size = perf_data_file__size(session->file);
1692 if (perf_session__register_idle_thread(session) == NULL)
1695 if (!perf_data_file__is_pipe(session->file))
1696 err = __perf_session__process_events(session,
1697 session->header.data_offset,
1698 session->header.data_size, size);
1700 err = __perf_session__process_pipe_events(session);
1705 bool perf_session__has_traces(struct perf_session *session, const char *msg)
1707 struct perf_evsel *evsel;
1709 evlist__for_each(session->evlist, evsel) {
1710 if (evsel->attr.type == PERF_TYPE_TRACEPOINT)
1714 pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg);
1718 int maps__set_kallsyms_ref_reloc_sym(struct map **maps,
1719 const char *symbol_name, u64 addr)
1723 struct ref_reloc_sym *ref;
1725 ref = zalloc(sizeof(struct ref_reloc_sym));
1729 ref->name = strdup(symbol_name);
1730 if (ref->name == NULL) {
1735 bracket = strchr(ref->name, ']');
1741 for (i = 0; i < MAP__NR_TYPES; ++i) {
1742 struct kmap *kmap = map__kmap(maps[i]);
1746 kmap->ref_reloc_sym = ref;
1752 size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp)
1754 return machines__fprintf_dsos(&session->machines, fp);
1757 size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp,
1758 bool (skip)(struct dso *dso, int parm), int parm)
1760 return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm);
1763 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp)
1766 const char *msg = "";
1768 if (perf_header__has_feat(&session->header, HEADER_AUXTRACE))
1769 msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)";
1771 ret = fprintf(fp, "\nAggregated stats:%s\n", msg);
1773 ret += events_stats__fprintf(&session->evlist->stats, fp);
1777 size_t perf_session__fprintf(struct perf_session *session, FILE *fp)
1780 * FIXME: Here we have to actually print all the machines in this
1781 * session, not just the host...
1783 return machine__fprintf(&session->machines.host, fp);
1786 struct perf_evsel *perf_session__find_first_evtype(struct perf_session *session,
1789 struct perf_evsel *pos;
1791 evlist__for_each(session->evlist, pos) {
1792 if (pos->attr.type == type)
1798 void perf_evsel__print_ip(struct perf_evsel *evsel, struct perf_sample *sample,
1799 struct addr_location *al,
1800 unsigned int print_opts, unsigned int stack_depth)
1802 struct callchain_cursor_node *node;
1803 int print_ip = print_opts & PRINT_IP_OPT_IP;
1804 int print_sym = print_opts & PRINT_IP_OPT_SYM;
1805 int print_dso = print_opts & PRINT_IP_OPT_DSO;
1806 int print_symoffset = print_opts & PRINT_IP_OPT_SYMOFFSET;
1807 int print_oneline = print_opts & PRINT_IP_OPT_ONELINE;
1808 int print_srcline = print_opts & PRINT_IP_OPT_SRCLINE;
1809 char s = print_oneline ? ' ' : '\t';
1811 if (symbol_conf.use_callchain && sample->callchain) {
1812 struct addr_location node_al;
1814 if (thread__resolve_callchain(al->thread, evsel,
1816 stack_depth) != 0) {
1818 error("Failed to resolve callchain. Skipping\n");
1821 callchain_cursor_commit(&callchain_cursor);
1823 if (print_symoffset)
1826 while (stack_depth) {
1829 node = callchain_cursor_current(&callchain_cursor);
1833 if (node->sym && node->sym->ignore)
1837 printf("%c%16" PRIx64, s, node->ip);
1840 addr = node->map->map_ip(node->map, node->ip);
1844 if (print_symoffset) {
1845 node_al.addr = addr;
1846 node_al.map = node->map;
1847 symbol__fprintf_symname_offs(node->sym, &node_al, stdout);
1849 symbol__fprintf_symname(node->sym, stdout);
1854 map__fprintf_dsoname(node->map, stdout);
1859 map__fprintf_srcline(node->map, addr, "\n ",
1867 callchain_cursor_advance(&callchain_cursor);
1871 if (al->sym && al->sym->ignore)
1875 printf("%16" PRIx64, sample->ip);
1879 if (print_symoffset)
1880 symbol__fprintf_symname_offs(al->sym, al,
1883 symbol__fprintf_symname(al->sym, stdout);
1888 map__fprintf_dsoname(al->map, stdout);
1893 map__fprintf_srcline(al->map, al->addr, "\n ", stdout);
1897 int perf_session__cpu_bitmap(struct perf_session *session,
1898 const char *cpu_list, unsigned long *cpu_bitmap)
1901 struct cpu_map *map;
1903 for (i = 0; i < PERF_TYPE_MAX; ++i) {
1904 struct perf_evsel *evsel;
1906 evsel = perf_session__find_first_evtype(session, i);
1910 if (!(evsel->attr.sample_type & PERF_SAMPLE_CPU)) {
1911 pr_err("File does not contain CPU events. "
1912 "Remove -c option to proceed.\n");
1917 map = cpu_map__new(cpu_list);
1919 pr_err("Invalid cpu_list\n");
1923 for (i = 0; i < map->nr; i++) {
1924 int cpu = map->map[i];
1926 if (cpu >= MAX_NR_CPUS) {
1927 pr_err("Requested CPU %d too large. "
1928 "Consider raising MAX_NR_CPUS\n", cpu);
1929 goto out_delete_map;
1932 set_bit(cpu, cpu_bitmap);
1942 void perf_session__fprintf_info(struct perf_session *session, FILE *fp,
1948 if (session == NULL || fp == NULL)
1951 fd = perf_data_file__fd(session->file);
1953 ret = fstat(fd, &st);
1957 fprintf(fp, "# ========\n");
1958 fprintf(fp, "# captured on: %s", ctime(&st.st_ctime));
1959 perf_header__fprintf_info(session, fp, full);
1960 fprintf(fp, "# ========\n#\n");
1964 int __perf_session__set_tracepoints_handlers(struct perf_session *session,
1965 const struct perf_evsel_str_handler *assocs,
1968 struct perf_evsel *evsel;
1972 for (i = 0; i < nr_assocs; i++) {
1974 * Adding a handler for an event not in the session,
1977 evsel = perf_evlist__find_tracepoint_by_name(session->evlist, assocs[i].name);
1982 if (evsel->handler != NULL)
1984 evsel->handler = assocs[i].handler;
1992 int perf_event__process_id_index(struct perf_tool *tool __maybe_unused,
1993 union perf_event *event,
1994 struct perf_session *session)
1996 struct perf_evlist *evlist = session->evlist;
1997 struct id_index_event *ie = &event->id_index;
1998 size_t i, nr, max_nr;
2000 max_nr = (ie->header.size - sizeof(struct id_index_event)) /
2001 sizeof(struct id_index_entry);
2007 fprintf(stdout, " nr: %zu\n", nr);
2009 for (i = 0; i < nr; i++) {
2010 struct id_index_entry *e = &ie->entries[i];
2011 struct perf_sample_id *sid;
2014 fprintf(stdout, " ... id: %"PRIu64, e->id);
2015 fprintf(stdout, " idx: %"PRIu64, e->idx);
2016 fprintf(stdout, " cpu: %"PRId64, e->cpu);
2017 fprintf(stdout, " tid: %"PRId64"\n", e->tid);
2020 sid = perf_evlist__id2sid(evlist, e->id);
2030 int perf_event__synthesize_id_index(struct perf_tool *tool,
2031 perf_event__handler_t process,
2032 struct perf_evlist *evlist,
2033 struct machine *machine)
2035 union perf_event *ev;
2036 struct perf_evsel *evsel;
2037 size_t nr = 0, i = 0, sz, max_nr, n;
2040 pr_debug2("Synthesizing id index\n");
2042 max_nr = (UINT16_MAX - sizeof(struct id_index_event)) /
2043 sizeof(struct id_index_entry);
2045 evlist__for_each(evlist, evsel)
2048 n = nr > max_nr ? max_nr : nr;
2049 sz = sizeof(struct id_index_event) + n * sizeof(struct id_index_entry);
2054 ev->id_index.header.type = PERF_RECORD_ID_INDEX;
2055 ev->id_index.header.size = sz;
2056 ev->id_index.nr = n;
2058 evlist__for_each(evlist, evsel) {
2061 for (j = 0; j < evsel->ids; j++) {
2062 struct id_index_entry *e;
2063 struct perf_sample_id *sid;
2066 err = process(tool, ev, NULL, machine);
2073 e = &ev->id_index.entries[i++];
2075 e->id = evsel->id[j];
2077 sid = perf_evlist__id2sid(evlist, e->id);
2089 sz = sizeof(struct id_index_event) + nr * sizeof(struct id_index_entry);
2090 ev->id_index.header.size = sz;
2091 ev->id_index.nr = nr;
2093 err = process(tool, ev, NULL, machine);