1 // SPDX-License-Identifier: GPL-2.0
5 #include <linux/kernel.h>
6 #include <linux/zalloc.h>
11 #include <sys/types.h>
13 #include <perf/cpumap.h>
15 #include "map_symbol.h"
26 #include "perf_regs.h"
30 #include "thread-stack.h"
31 #include "sample-raw.h"
34 #include "ui/progress.h"
36 #include "arch/common.h"
38 #include <internal/lib.h>
40 #ifdef HAVE_ZSTD_SUPPORT
41 static int perf_session__process_compressed_event(struct perf_session *session,
42 union perf_event *event, u64 file_offset,
43 const char *file_path)
46 size_t decomp_size, src_size;
47 u64 decomp_last_rem = 0;
48 size_t mmap_len, decomp_len = session->header.env.comp_mmap_len;
49 struct decomp *decomp, *decomp_last = session->active_decomp->decomp_last;
52 decomp_last_rem = decomp_last->size - decomp_last->head;
53 decomp_len += decomp_last_rem;
56 mmap_len = sizeof(struct decomp) + decomp_len;
57 decomp = mmap(NULL, mmap_len, PROT_READ|PROT_WRITE,
58 MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);
59 if (decomp == MAP_FAILED) {
60 pr_err("Couldn't allocate memory for decompression\n");
64 decomp->file_pos = file_offset;
65 decomp->file_path = file_path;
66 decomp->mmap_len = mmap_len;
69 if (decomp_last_rem) {
70 memcpy(decomp->data, &(decomp_last->data[decomp_last->head]), decomp_last_rem);
71 decomp->size = decomp_last_rem;
74 src = (void *)event + sizeof(struct perf_record_compressed);
75 src_size = event->pack.header.size - sizeof(struct perf_record_compressed);
77 decomp_size = zstd_decompress_stream(session->active_decomp->zstd_decomp, src, src_size,
78 &(decomp->data[decomp_last_rem]), decomp_len - decomp_last_rem);
80 munmap(decomp, mmap_len);
81 pr_err("Couldn't decompress data\n");
85 decomp->size += decomp_size;
87 if (session->active_decomp->decomp == NULL)
88 session->active_decomp->decomp = decomp;
90 session->active_decomp->decomp_last->next = decomp;
92 session->active_decomp->decomp_last = decomp;
94 pr_debug("decomp (B): %zd to %zd\n", src_size, decomp_size);
98 #else /* !HAVE_ZSTD_SUPPORT */
99 #define perf_session__process_compressed_event perf_session__process_compressed_event_stub
102 static int perf_session__deliver_event(struct perf_session *session,
103 union perf_event *event,
104 struct perf_tool *tool,
106 const char *file_path);
108 static int perf_session__open(struct perf_session *session, int repipe_fd)
110 struct perf_data *data = session->data;
112 if (perf_session__read_header(session, repipe_fd) < 0) {
113 pr_err("incompatible file format (rerun with -v to learn more)\n");
117 if (perf_data__is_pipe(data))
120 if (perf_header__has_feat(&session->header, HEADER_STAT))
123 if (!evlist__valid_sample_type(session->evlist)) {
124 pr_err("non matching sample_type\n");
128 if (!evlist__valid_sample_id_all(session->evlist)) {
129 pr_err("non matching sample_id_all\n");
133 if (!evlist__valid_read_format(session->evlist)) {
134 pr_err("non matching read_format\n");
141 void perf_session__set_id_hdr_size(struct perf_session *session)
143 u16 id_hdr_size = evlist__id_hdr_size(session->evlist);
145 machines__set_id_hdr_size(&session->machines, id_hdr_size);
148 int perf_session__create_kernel_maps(struct perf_session *session)
150 int ret = machine__create_kernel_maps(&session->machines.host);
153 ret = machines__create_guest_kernel_maps(&session->machines);
157 static void perf_session__destroy_kernel_maps(struct perf_session *session)
159 machines__destroy_kernel_maps(&session->machines);
162 static bool perf_session__has_comm_exec(struct perf_session *session)
166 evlist__for_each_entry(session->evlist, evsel) {
167 if (evsel->core.attr.comm_exec)
174 static void perf_session__set_comm_exec(struct perf_session *session)
176 bool comm_exec = perf_session__has_comm_exec(session);
178 machines__set_comm_exec(&session->machines, comm_exec);
181 static int ordered_events__deliver_event(struct ordered_events *oe,
182 struct ordered_event *event)
184 struct perf_session *session = container_of(oe, struct perf_session,
187 return perf_session__deliver_event(session, event->event,
188 session->tool, event->file_offset,
192 struct perf_session *__perf_session__new(struct perf_data *data,
193 bool repipe, int repipe_fd,
194 struct perf_tool *tool)
197 struct perf_session *session = zalloc(sizeof(*session));
202 session->repipe = repipe;
203 session->tool = tool;
204 session->decomp_data.zstd_decomp = &session->zstd_data;
205 session->active_decomp = &session->decomp_data;
206 INIT_LIST_HEAD(&session->auxtrace_index);
207 machines__init(&session->machines);
208 ordered_events__init(&session->ordered_events,
209 ordered_events__deliver_event, NULL);
211 perf_env__init(&session->header.env);
213 ret = perf_data__open(data);
217 session->data = data;
219 if (perf_data__is_read(data)) {
220 ret = perf_session__open(session, repipe_fd);
225 * set session attributes that are present in perf.data
226 * but not in pipe-mode.
228 if (!data->is_pipe) {
229 perf_session__set_id_hdr_size(session);
230 perf_session__set_comm_exec(session);
233 evlist__init_trace_event_sample_raw(session->evlist);
235 /* Open the directory data. */
237 ret = perf_data__open_dir(data);
242 if (!symbol_conf.kallsyms_name &&
243 !symbol_conf.vmlinux_name)
244 symbol_conf.kallsyms_name = perf_data__kallsyms_name(data);
247 session->machines.host.env = &perf_env;
250 session->machines.host.single_address_space =
251 perf_env__single_address_space(session->machines.host.env);
253 if (!data || perf_data__is_write(data)) {
255 * In O_RDONLY mode this will be performed when reading the
256 * kernel MMAP event, in perf_event__process_mmap().
258 if (perf_session__create_kernel_maps(session) < 0)
259 pr_warning("Cannot read kernel map\n");
263 * In pipe-mode, evlist is empty until PERF_RECORD_HEADER_ATTR is
264 * processed, so evlist__sample_id_all is not meaningful here.
266 if ((!data || !data->is_pipe) && tool && tool->ordering_requires_timestamps &&
267 tool->ordered_events && !evlist__sample_id_all(session->evlist)) {
268 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n");
269 tool->ordered_events = false;
275 perf_session__delete(session);
280 static void perf_session__delete_threads(struct perf_session *session)
282 machine__delete_threads(&session->machines.host);
285 static void perf_decomp__release_events(struct decomp *next)
287 struct decomp *decomp;
295 mmap_len = decomp->mmap_len;
296 munmap(decomp, mmap_len);
300 void perf_session__delete(struct perf_session *session)
304 auxtrace__free(session);
305 auxtrace_index__free(&session->auxtrace_index);
306 perf_session__destroy_kernel_maps(session);
307 perf_session__delete_threads(session);
308 perf_decomp__release_events(session->decomp_data.decomp);
309 perf_env__exit(&session->header.env);
310 machines__exit(&session->machines);
312 if (perf_data__is_read(session->data))
313 evlist__delete(session->evlist);
314 perf_data__close(session->data);
316 trace_event__cleanup(&session->tevent);
320 static int process_event_synth_tracing_data_stub(struct perf_session *session
322 union perf_event *event
325 dump_printf(": unhandled!\n");
329 static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused,
330 union perf_event *event __maybe_unused,
331 struct evlist **pevlist
334 dump_printf(": unhandled!\n");
338 static int process_event_synth_event_update_stub(struct perf_tool *tool __maybe_unused,
339 union perf_event *event __maybe_unused,
340 struct evlist **pevlist
344 perf_event__fprintf_event_update(event, stdout);
346 dump_printf(": unhandled!\n");
350 static int process_event_sample_stub(struct perf_tool *tool __maybe_unused,
351 union perf_event *event __maybe_unused,
352 struct perf_sample *sample __maybe_unused,
353 struct evsel *evsel __maybe_unused,
354 struct machine *machine __maybe_unused)
356 dump_printf(": unhandled!\n");
360 static int process_event_stub(struct perf_tool *tool __maybe_unused,
361 union perf_event *event __maybe_unused,
362 struct perf_sample *sample __maybe_unused,
363 struct machine *machine __maybe_unused)
365 dump_printf(": unhandled!\n");
369 static int process_finished_round_stub(struct perf_tool *tool __maybe_unused,
370 union perf_event *event __maybe_unused,
371 struct ordered_events *oe __maybe_unused)
373 dump_printf(": unhandled!\n");
377 static int process_finished_round(struct perf_tool *tool,
378 union perf_event *event,
379 struct ordered_events *oe);
381 static int skipn(int fd, off_t n)
387 ret = read(fd, buf, min(n, (off_t)sizeof(buf)));
396 static s64 process_event_auxtrace_stub(struct perf_session *session __maybe_unused,
397 union perf_event *event)
399 dump_printf(": unhandled!\n");
400 if (perf_data__is_pipe(session->data))
401 skipn(perf_data__fd(session->data), event->auxtrace.size);
402 return event->auxtrace.size;
405 static int process_event_op2_stub(struct perf_session *session __maybe_unused,
406 union perf_event *event __maybe_unused)
408 dump_printf(": unhandled!\n");
414 int process_event_thread_map_stub(struct perf_session *session __maybe_unused,
415 union perf_event *event __maybe_unused)
418 perf_event__fprintf_thread_map(event, stdout);
420 dump_printf(": unhandled!\n");
425 int process_event_cpu_map_stub(struct perf_session *session __maybe_unused,
426 union perf_event *event __maybe_unused)
429 perf_event__fprintf_cpu_map(event, stdout);
431 dump_printf(": unhandled!\n");
436 int process_event_stat_config_stub(struct perf_session *session __maybe_unused,
437 union perf_event *event __maybe_unused)
440 perf_event__fprintf_stat_config(event, stdout);
442 dump_printf(": unhandled!\n");
446 static int process_stat_stub(struct perf_session *perf_session __maybe_unused,
447 union perf_event *event)
450 perf_event__fprintf_stat(event, stdout);
452 dump_printf(": unhandled!\n");
456 static int process_stat_round_stub(struct perf_session *perf_session __maybe_unused,
457 union perf_event *event)
460 perf_event__fprintf_stat_round(event, stdout);
462 dump_printf(": unhandled!\n");
466 static int process_event_time_conv_stub(struct perf_session *perf_session __maybe_unused,
467 union perf_event *event)
470 perf_event__fprintf_time_conv(event, stdout);
472 dump_printf(": unhandled!\n");
476 static int perf_session__process_compressed_event_stub(struct perf_session *session __maybe_unused,
477 union perf_event *event __maybe_unused,
478 u64 file_offset __maybe_unused,
479 const char *file_path __maybe_unused)
481 dump_printf(": unhandled!\n");
485 void perf_tool__fill_defaults(struct perf_tool *tool)
487 if (tool->sample == NULL)
488 tool->sample = process_event_sample_stub;
489 if (tool->mmap == NULL)
490 tool->mmap = process_event_stub;
491 if (tool->mmap2 == NULL)
492 tool->mmap2 = process_event_stub;
493 if (tool->comm == NULL)
494 tool->comm = process_event_stub;
495 if (tool->namespaces == NULL)
496 tool->namespaces = process_event_stub;
497 if (tool->cgroup == NULL)
498 tool->cgroup = process_event_stub;
499 if (tool->fork == NULL)
500 tool->fork = process_event_stub;
501 if (tool->exit == NULL)
502 tool->exit = process_event_stub;
503 if (tool->lost == NULL)
504 tool->lost = perf_event__process_lost;
505 if (tool->lost_samples == NULL)
506 tool->lost_samples = perf_event__process_lost_samples;
507 if (tool->aux == NULL)
508 tool->aux = perf_event__process_aux;
509 if (tool->itrace_start == NULL)
510 tool->itrace_start = perf_event__process_itrace_start;
511 if (tool->context_switch == NULL)
512 tool->context_switch = perf_event__process_switch;
513 if (tool->ksymbol == NULL)
514 tool->ksymbol = perf_event__process_ksymbol;
515 if (tool->bpf == NULL)
516 tool->bpf = perf_event__process_bpf;
517 if (tool->text_poke == NULL)
518 tool->text_poke = perf_event__process_text_poke;
519 if (tool->aux_output_hw_id == NULL)
520 tool->aux_output_hw_id = perf_event__process_aux_output_hw_id;
521 if (tool->read == NULL)
522 tool->read = process_event_sample_stub;
523 if (tool->throttle == NULL)
524 tool->throttle = process_event_stub;
525 if (tool->unthrottle == NULL)
526 tool->unthrottle = process_event_stub;
527 if (tool->attr == NULL)
528 tool->attr = process_event_synth_attr_stub;
529 if (tool->event_update == NULL)
530 tool->event_update = process_event_synth_event_update_stub;
531 if (tool->tracing_data == NULL)
532 tool->tracing_data = process_event_synth_tracing_data_stub;
533 if (tool->build_id == NULL)
534 tool->build_id = process_event_op2_stub;
535 if (tool->finished_round == NULL) {
536 if (tool->ordered_events)
537 tool->finished_round = process_finished_round;
539 tool->finished_round = process_finished_round_stub;
541 if (tool->id_index == NULL)
542 tool->id_index = process_event_op2_stub;
543 if (tool->auxtrace_info == NULL)
544 tool->auxtrace_info = process_event_op2_stub;
545 if (tool->auxtrace == NULL)
546 tool->auxtrace = process_event_auxtrace_stub;
547 if (tool->auxtrace_error == NULL)
548 tool->auxtrace_error = process_event_op2_stub;
549 if (tool->thread_map == NULL)
550 tool->thread_map = process_event_thread_map_stub;
551 if (tool->cpu_map == NULL)
552 tool->cpu_map = process_event_cpu_map_stub;
553 if (tool->stat_config == NULL)
554 tool->stat_config = process_event_stat_config_stub;
555 if (tool->stat == NULL)
556 tool->stat = process_stat_stub;
557 if (tool->stat_round == NULL)
558 tool->stat_round = process_stat_round_stub;
559 if (tool->time_conv == NULL)
560 tool->time_conv = process_event_time_conv_stub;
561 if (tool->feature == NULL)
562 tool->feature = process_event_op2_stub;
563 if (tool->compressed == NULL)
564 tool->compressed = perf_session__process_compressed_event;
567 static void swap_sample_id_all(union perf_event *event, void *data)
569 void *end = (void *) event + event->header.size;
570 int size = end - data;
572 BUG_ON(size % sizeof(u64));
573 mem_bswap_64(data, size);
576 static void perf_event__all64_swap(union perf_event *event,
577 bool sample_id_all __maybe_unused)
579 struct perf_event_header *hdr = &event->header;
580 mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr));
583 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all)
585 event->comm.pid = bswap_32(event->comm.pid);
586 event->comm.tid = bswap_32(event->comm.tid);
589 void *data = &event->comm.comm;
591 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
592 swap_sample_id_all(event, data);
596 static void perf_event__mmap_swap(union perf_event *event,
599 event->mmap.pid = bswap_32(event->mmap.pid);
600 event->mmap.tid = bswap_32(event->mmap.tid);
601 event->mmap.start = bswap_64(event->mmap.start);
602 event->mmap.len = bswap_64(event->mmap.len);
603 event->mmap.pgoff = bswap_64(event->mmap.pgoff);
606 void *data = &event->mmap.filename;
608 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
609 swap_sample_id_all(event, data);
613 static void perf_event__mmap2_swap(union perf_event *event,
616 event->mmap2.pid = bswap_32(event->mmap2.pid);
617 event->mmap2.tid = bswap_32(event->mmap2.tid);
618 event->mmap2.start = bswap_64(event->mmap2.start);
619 event->mmap2.len = bswap_64(event->mmap2.len);
620 event->mmap2.pgoff = bswap_64(event->mmap2.pgoff);
622 if (!(event->header.misc & PERF_RECORD_MISC_MMAP_BUILD_ID)) {
623 event->mmap2.maj = bswap_32(event->mmap2.maj);
624 event->mmap2.min = bswap_32(event->mmap2.min);
625 event->mmap2.ino = bswap_64(event->mmap2.ino);
626 event->mmap2.ino_generation = bswap_64(event->mmap2.ino_generation);
630 void *data = &event->mmap2.filename;
632 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
633 swap_sample_id_all(event, data);
636 static void perf_event__task_swap(union perf_event *event, bool sample_id_all)
638 event->fork.pid = bswap_32(event->fork.pid);
639 event->fork.tid = bswap_32(event->fork.tid);
640 event->fork.ppid = bswap_32(event->fork.ppid);
641 event->fork.ptid = bswap_32(event->fork.ptid);
642 event->fork.time = bswap_64(event->fork.time);
645 swap_sample_id_all(event, &event->fork + 1);
648 static void perf_event__read_swap(union perf_event *event, bool sample_id_all)
650 event->read.pid = bswap_32(event->read.pid);
651 event->read.tid = bswap_32(event->read.tid);
652 event->read.value = bswap_64(event->read.value);
653 event->read.time_enabled = bswap_64(event->read.time_enabled);
654 event->read.time_running = bswap_64(event->read.time_running);
655 event->read.id = bswap_64(event->read.id);
658 swap_sample_id_all(event, &event->read + 1);
661 static void perf_event__aux_swap(union perf_event *event, bool sample_id_all)
663 event->aux.aux_offset = bswap_64(event->aux.aux_offset);
664 event->aux.aux_size = bswap_64(event->aux.aux_size);
665 event->aux.flags = bswap_64(event->aux.flags);
668 swap_sample_id_all(event, &event->aux + 1);
671 static void perf_event__itrace_start_swap(union perf_event *event,
674 event->itrace_start.pid = bswap_32(event->itrace_start.pid);
675 event->itrace_start.tid = bswap_32(event->itrace_start.tid);
678 swap_sample_id_all(event, &event->itrace_start + 1);
681 static void perf_event__switch_swap(union perf_event *event, bool sample_id_all)
683 if (event->header.type == PERF_RECORD_SWITCH_CPU_WIDE) {
684 event->context_switch.next_prev_pid =
685 bswap_32(event->context_switch.next_prev_pid);
686 event->context_switch.next_prev_tid =
687 bswap_32(event->context_switch.next_prev_tid);
691 swap_sample_id_all(event, &event->context_switch + 1);
694 static void perf_event__text_poke_swap(union perf_event *event, bool sample_id_all)
696 event->text_poke.addr = bswap_64(event->text_poke.addr);
697 event->text_poke.old_len = bswap_16(event->text_poke.old_len);
698 event->text_poke.new_len = bswap_16(event->text_poke.new_len);
701 size_t len = sizeof(event->text_poke.old_len) +
702 sizeof(event->text_poke.new_len) +
703 event->text_poke.old_len +
704 event->text_poke.new_len;
705 void *data = &event->text_poke.old_len;
707 data += PERF_ALIGN(len, sizeof(u64));
708 swap_sample_id_all(event, data);
712 static void perf_event__throttle_swap(union perf_event *event,
715 event->throttle.time = bswap_64(event->throttle.time);
716 event->throttle.id = bswap_64(event->throttle.id);
717 event->throttle.stream_id = bswap_64(event->throttle.stream_id);
720 swap_sample_id_all(event, &event->throttle + 1);
723 static void perf_event__namespaces_swap(union perf_event *event,
728 event->namespaces.pid = bswap_32(event->namespaces.pid);
729 event->namespaces.tid = bswap_32(event->namespaces.tid);
730 event->namespaces.nr_namespaces = bswap_64(event->namespaces.nr_namespaces);
732 for (i = 0; i < event->namespaces.nr_namespaces; i++) {
733 struct perf_ns_link_info *ns = &event->namespaces.link_info[i];
735 ns->dev = bswap_64(ns->dev);
736 ns->ino = bswap_64(ns->ino);
740 swap_sample_id_all(event, &event->namespaces.link_info[i]);
743 static void perf_event__cgroup_swap(union perf_event *event, bool sample_id_all)
745 event->cgroup.id = bswap_64(event->cgroup.id);
748 void *data = &event->cgroup.path;
750 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
751 swap_sample_id_all(event, data);
755 static u8 revbyte(u8 b)
757 int rev = (b >> 4) | ((b & 0xf) << 4);
758 rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2);
759 rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1);
764 * XXX this is hack in attempt to carry flags bitfield
765 * through endian village. ABI says:
767 * Bit-fields are allocated from right to left (least to most significant)
768 * on little-endian implementations and from left to right (most to least
769 * significant) on big-endian implementations.
771 * The above seems to be byte specific, so we need to reverse each
772 * byte of the bitfield. 'Internet' also says this might be implementation
773 * specific and we probably need proper fix and carry perf_event_attr
774 * bitfield flags in separate data file FEAT_ section. Thought this seems
777 static void swap_bitfield(u8 *p, unsigned len)
781 for (i = 0; i < len; i++) {
787 /* exported for swapping attributes in file header */
788 void perf_event__attr_swap(struct perf_event_attr *attr)
790 attr->type = bswap_32(attr->type);
791 attr->size = bswap_32(attr->size);
793 #define bswap_safe(f, n) \
794 (attr->size > (offsetof(struct perf_event_attr, f) + \
795 sizeof(attr->f) * (n)))
796 #define bswap_field(f, sz) \
798 if (bswap_safe(f, 0)) \
799 attr->f = bswap_##sz(attr->f); \
801 #define bswap_field_16(f) bswap_field(f, 16)
802 #define bswap_field_32(f) bswap_field(f, 32)
803 #define bswap_field_64(f) bswap_field(f, 64)
805 bswap_field_64(config);
806 bswap_field_64(sample_period);
807 bswap_field_64(sample_type);
808 bswap_field_64(read_format);
809 bswap_field_32(wakeup_events);
810 bswap_field_32(bp_type);
811 bswap_field_64(bp_addr);
812 bswap_field_64(bp_len);
813 bswap_field_64(branch_sample_type);
814 bswap_field_64(sample_regs_user);
815 bswap_field_32(sample_stack_user);
816 bswap_field_32(aux_watermark);
817 bswap_field_16(sample_max_stack);
818 bswap_field_32(aux_sample_size);
821 * After read_format are bitfields. Check read_format because
822 * we are unable to use offsetof on bitfield.
824 if (bswap_safe(read_format, 1))
825 swap_bitfield((u8 *) (&attr->read_format + 1),
827 #undef bswap_field_64
828 #undef bswap_field_32
833 static void perf_event__hdr_attr_swap(union perf_event *event,
834 bool sample_id_all __maybe_unused)
838 perf_event__attr_swap(&event->attr.attr);
840 size = event->header.size;
841 size -= (void *)&event->attr.id - (void *)event;
842 mem_bswap_64(event->attr.id, size);
845 static void perf_event__event_update_swap(union perf_event *event,
846 bool sample_id_all __maybe_unused)
848 event->event_update.type = bswap_64(event->event_update.type);
849 event->event_update.id = bswap_64(event->event_update.id);
852 static void perf_event__event_type_swap(union perf_event *event,
853 bool sample_id_all __maybe_unused)
855 event->event_type.event_type.event_id =
856 bswap_64(event->event_type.event_type.event_id);
859 static void perf_event__tracing_data_swap(union perf_event *event,
860 bool sample_id_all __maybe_unused)
862 event->tracing_data.size = bswap_32(event->tracing_data.size);
865 static void perf_event__auxtrace_info_swap(union perf_event *event,
866 bool sample_id_all __maybe_unused)
870 event->auxtrace_info.type = bswap_32(event->auxtrace_info.type);
872 size = event->header.size;
873 size -= (void *)&event->auxtrace_info.priv - (void *)event;
874 mem_bswap_64(event->auxtrace_info.priv, size);
877 static void perf_event__auxtrace_swap(union perf_event *event,
878 bool sample_id_all __maybe_unused)
880 event->auxtrace.size = bswap_64(event->auxtrace.size);
881 event->auxtrace.offset = bswap_64(event->auxtrace.offset);
882 event->auxtrace.reference = bswap_64(event->auxtrace.reference);
883 event->auxtrace.idx = bswap_32(event->auxtrace.idx);
884 event->auxtrace.tid = bswap_32(event->auxtrace.tid);
885 event->auxtrace.cpu = bswap_32(event->auxtrace.cpu);
888 static void perf_event__auxtrace_error_swap(union perf_event *event,
889 bool sample_id_all __maybe_unused)
891 event->auxtrace_error.type = bswap_32(event->auxtrace_error.type);
892 event->auxtrace_error.code = bswap_32(event->auxtrace_error.code);
893 event->auxtrace_error.cpu = bswap_32(event->auxtrace_error.cpu);
894 event->auxtrace_error.pid = bswap_32(event->auxtrace_error.pid);
895 event->auxtrace_error.tid = bswap_32(event->auxtrace_error.tid);
896 event->auxtrace_error.fmt = bswap_32(event->auxtrace_error.fmt);
897 event->auxtrace_error.ip = bswap_64(event->auxtrace_error.ip);
898 if (event->auxtrace_error.fmt)
899 event->auxtrace_error.time = bswap_64(event->auxtrace_error.time);
902 static void perf_event__thread_map_swap(union perf_event *event,
903 bool sample_id_all __maybe_unused)
907 event->thread_map.nr = bswap_64(event->thread_map.nr);
909 for (i = 0; i < event->thread_map.nr; i++)
910 event->thread_map.entries[i].pid = bswap_64(event->thread_map.entries[i].pid);
913 static void perf_event__cpu_map_swap(union perf_event *event,
914 bool sample_id_all __maybe_unused)
916 struct perf_record_cpu_map_data *data = &event->cpu_map.data;
917 struct cpu_map_entries *cpus;
918 struct perf_record_record_cpu_map *mask;
921 data->type = bswap_16(data->type);
923 switch (data->type) {
924 case PERF_CPU_MAP__CPUS:
925 cpus = (struct cpu_map_entries *)data->data;
927 cpus->nr = bswap_16(cpus->nr);
929 for (i = 0; i < cpus->nr; i++)
930 cpus->cpu[i] = bswap_16(cpus->cpu[i]);
932 case PERF_CPU_MAP__MASK:
933 mask = (struct perf_record_record_cpu_map *)data->data;
935 mask->nr = bswap_16(mask->nr);
936 mask->long_size = bswap_16(mask->long_size);
938 switch (mask->long_size) {
939 case 4: mem_bswap_32(&mask->mask, mask->nr); break;
940 case 8: mem_bswap_64(&mask->mask, mask->nr); break;
942 pr_err("cpu_map swap: unsupported long size\n");
949 static void perf_event__stat_config_swap(union perf_event *event,
950 bool sample_id_all __maybe_unused)
954 size = bswap_64(event->stat_config.nr) * sizeof(event->stat_config.data[0]);
955 size += 1; /* nr item itself */
956 mem_bswap_64(&event->stat_config.nr, size);
959 static void perf_event__stat_swap(union perf_event *event,
960 bool sample_id_all __maybe_unused)
962 event->stat.id = bswap_64(event->stat.id);
963 event->stat.thread = bswap_32(event->stat.thread);
964 event->stat.cpu = bswap_32(event->stat.cpu);
965 event->stat.val = bswap_64(event->stat.val);
966 event->stat.ena = bswap_64(event->stat.ena);
967 event->stat.run = bswap_64(event->stat.run);
970 static void perf_event__stat_round_swap(union perf_event *event,
971 bool sample_id_all __maybe_unused)
973 event->stat_round.type = bswap_64(event->stat_round.type);
974 event->stat_round.time = bswap_64(event->stat_round.time);
977 static void perf_event__time_conv_swap(union perf_event *event,
978 bool sample_id_all __maybe_unused)
980 event->time_conv.time_shift = bswap_64(event->time_conv.time_shift);
981 event->time_conv.time_mult = bswap_64(event->time_conv.time_mult);
982 event->time_conv.time_zero = bswap_64(event->time_conv.time_zero);
984 if (event_contains(event->time_conv, time_cycles)) {
985 event->time_conv.time_cycles = bswap_64(event->time_conv.time_cycles);
986 event->time_conv.time_mask = bswap_64(event->time_conv.time_mask);
990 typedef void (*perf_event__swap_op)(union perf_event *event,
993 static perf_event__swap_op perf_event__swap_ops[] = {
994 [PERF_RECORD_MMAP] = perf_event__mmap_swap,
995 [PERF_RECORD_MMAP2] = perf_event__mmap2_swap,
996 [PERF_RECORD_COMM] = perf_event__comm_swap,
997 [PERF_RECORD_FORK] = perf_event__task_swap,
998 [PERF_RECORD_EXIT] = perf_event__task_swap,
999 [PERF_RECORD_LOST] = perf_event__all64_swap,
1000 [PERF_RECORD_READ] = perf_event__read_swap,
1001 [PERF_RECORD_THROTTLE] = perf_event__throttle_swap,
1002 [PERF_RECORD_UNTHROTTLE] = perf_event__throttle_swap,
1003 [PERF_RECORD_SAMPLE] = perf_event__all64_swap,
1004 [PERF_RECORD_AUX] = perf_event__aux_swap,
1005 [PERF_RECORD_ITRACE_START] = perf_event__itrace_start_swap,
1006 [PERF_RECORD_LOST_SAMPLES] = perf_event__all64_swap,
1007 [PERF_RECORD_SWITCH] = perf_event__switch_swap,
1008 [PERF_RECORD_SWITCH_CPU_WIDE] = perf_event__switch_swap,
1009 [PERF_RECORD_NAMESPACES] = perf_event__namespaces_swap,
1010 [PERF_RECORD_CGROUP] = perf_event__cgroup_swap,
1011 [PERF_RECORD_TEXT_POKE] = perf_event__text_poke_swap,
1012 [PERF_RECORD_AUX_OUTPUT_HW_ID] = perf_event__all64_swap,
1013 [PERF_RECORD_HEADER_ATTR] = perf_event__hdr_attr_swap,
1014 [PERF_RECORD_HEADER_EVENT_TYPE] = perf_event__event_type_swap,
1015 [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
1016 [PERF_RECORD_HEADER_BUILD_ID] = NULL,
1017 [PERF_RECORD_ID_INDEX] = perf_event__all64_swap,
1018 [PERF_RECORD_AUXTRACE_INFO] = perf_event__auxtrace_info_swap,
1019 [PERF_RECORD_AUXTRACE] = perf_event__auxtrace_swap,
1020 [PERF_RECORD_AUXTRACE_ERROR] = perf_event__auxtrace_error_swap,
1021 [PERF_RECORD_THREAD_MAP] = perf_event__thread_map_swap,
1022 [PERF_RECORD_CPU_MAP] = perf_event__cpu_map_swap,
1023 [PERF_RECORD_STAT_CONFIG] = perf_event__stat_config_swap,
1024 [PERF_RECORD_STAT] = perf_event__stat_swap,
1025 [PERF_RECORD_STAT_ROUND] = perf_event__stat_round_swap,
1026 [PERF_RECORD_EVENT_UPDATE] = perf_event__event_update_swap,
1027 [PERF_RECORD_TIME_CONV] = perf_event__time_conv_swap,
1028 [PERF_RECORD_HEADER_MAX] = NULL,
1032 * When perf record finishes a pass on every buffers, it records this pseudo
1034 * We record the max timestamp t found in the pass n.
1035 * Assuming these timestamps are monotonic across cpus, we know that if
1036 * a buffer still has events with timestamps below t, they will be all
1037 * available and then read in the pass n + 1.
1038 * Hence when we start to read the pass n + 2, we can safely flush every
1039 * events with timestamps below t.
1041 * ============ PASS n =================
1044 * cnt1 timestamps | cnt2 timestamps
1047 * - | 4 <--- max recorded
1049 * ============ PASS n + 1 ==============
1052 * cnt1 timestamps | cnt2 timestamps
1055 * 5 | 7 <---- max recorded
1057 * Flush every events below timestamp 4
1059 * ============ PASS n + 2 ==============
1062 * cnt1 timestamps | cnt2 timestamps
1067 * Flush every events below timestamp 7
1070 static int process_finished_round(struct perf_tool *tool __maybe_unused,
1071 union perf_event *event __maybe_unused,
1072 struct ordered_events *oe)
1075 fprintf(stdout, "\n");
1076 return ordered_events__flush(oe, OE_FLUSH__ROUND);
1079 int perf_session__queue_event(struct perf_session *s, union perf_event *event,
1080 u64 timestamp, u64 file_offset, const char *file_path)
1082 return ordered_events__queue(&s->ordered_events, event, timestamp, file_offset, file_path);
1085 static void callchain__lbr_callstack_printf(struct perf_sample *sample)
1087 struct ip_callchain *callchain = sample->callchain;
1088 struct branch_stack *lbr_stack = sample->branch_stack;
1089 struct branch_entry *entries = perf_sample__branch_entries(sample);
1090 u64 kernel_callchain_nr = callchain->nr;
1093 for (i = 0; i < kernel_callchain_nr; i++) {
1094 if (callchain->ips[i] == PERF_CONTEXT_USER)
1098 if ((i != kernel_callchain_nr) && lbr_stack->nr) {
1101 * LBR callstack can only get user call chain,
1102 * i is kernel call chain number,
1103 * 1 is PERF_CONTEXT_USER.
1105 * The user call chain is stored in LBR registers.
1106 * LBR are pair registers. The caller is stored
1107 * in "from" register, while the callee is stored
1109 * For example, there is a call stack
1110 * "A"->"B"->"C"->"D".
1111 * The LBR registers will be recorded like
1112 * "C"->"D", "B"->"C", "A"->"B".
1113 * So only the first "to" register and all "from"
1114 * registers are needed to construct the whole stack.
1116 total_nr = i + 1 + lbr_stack->nr + 1;
1117 kernel_callchain_nr = i + 1;
1119 printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr);
1121 for (i = 0; i < kernel_callchain_nr; i++)
1122 printf("..... %2d: %016" PRIx64 "\n",
1123 i, callchain->ips[i]);
1125 printf("..... %2d: %016" PRIx64 "\n",
1126 (int)(kernel_callchain_nr), entries[0].to);
1127 for (i = 0; i < lbr_stack->nr; i++)
1128 printf("..... %2d: %016" PRIx64 "\n",
1129 (int)(i + kernel_callchain_nr + 1), entries[i].from);
1133 static void callchain__printf(struct evsel *evsel,
1134 struct perf_sample *sample)
1137 struct ip_callchain *callchain = sample->callchain;
1139 if (evsel__has_branch_callstack(evsel))
1140 callchain__lbr_callstack_printf(sample);
1142 printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr);
1144 for (i = 0; i < callchain->nr; i++)
1145 printf("..... %2d: %016" PRIx64 "\n",
1146 i, callchain->ips[i]);
1149 static void branch_stack__printf(struct perf_sample *sample, bool callstack)
1151 struct branch_entry *entries = perf_sample__branch_entries(sample);
1154 printf("%s: nr:%" PRIu64 "\n",
1155 !callstack ? "... branch stack" : "... branch callstack",
1156 sample->branch_stack->nr);
1158 for (i = 0; i < sample->branch_stack->nr; i++) {
1159 struct branch_entry *e = &entries[i];
1162 printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x %s\n",
1164 (unsigned short)e->flags.cycles,
1165 e->flags.mispred ? "M" : " ",
1166 e->flags.predicted ? "P" : " ",
1167 e->flags.abort ? "A" : " ",
1168 e->flags.in_tx ? "T" : " ",
1169 (unsigned)e->flags.reserved,
1170 e->flags.type ? branch_type_name(e->flags.type) : "");
1172 printf("..... %2"PRIu64": %016" PRIx64 "\n",
1173 i, i > 0 ? e->from : e->to);
1178 static void regs_dump__printf(u64 mask, u64 *regs, const char *arch)
1180 unsigned rid, i = 0;
1182 for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) {
1183 u64 val = regs[i++];
1185 printf(".... %-5s 0x%016" PRIx64 "\n",
1186 perf_reg_name(rid, arch), val);
1190 static const char *regs_abi[] = {
1191 [PERF_SAMPLE_REGS_ABI_NONE] = "none",
1192 [PERF_SAMPLE_REGS_ABI_32] = "32-bit",
1193 [PERF_SAMPLE_REGS_ABI_64] = "64-bit",
1196 static inline const char *regs_dump_abi(struct regs_dump *d)
1198 if (d->abi > PERF_SAMPLE_REGS_ABI_64)
1201 return regs_abi[d->abi];
1204 static void regs__printf(const char *type, struct regs_dump *regs, const char *arch)
1206 u64 mask = regs->mask;
1208 printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n",
1211 regs_dump_abi(regs));
1213 regs_dump__printf(mask, regs->regs, arch);
1216 static void regs_user__printf(struct perf_sample *sample, const char *arch)
1218 struct regs_dump *user_regs = &sample->user_regs;
1220 if (user_regs->regs)
1221 regs__printf("user", user_regs, arch);
1224 static void regs_intr__printf(struct perf_sample *sample, const char *arch)
1226 struct regs_dump *intr_regs = &sample->intr_regs;
1228 if (intr_regs->regs)
1229 regs__printf("intr", intr_regs, arch);
1232 static void stack_user__printf(struct stack_dump *dump)
1234 printf("... ustack: size %" PRIu64 ", offset 0x%x\n",
1235 dump->size, dump->offset);
1238 static void evlist__print_tstamp(struct evlist *evlist, union perf_event *event, struct perf_sample *sample)
1240 u64 sample_type = __evlist__combined_sample_type(evlist);
1242 if (event->header.type != PERF_RECORD_SAMPLE &&
1243 !evlist__sample_id_all(evlist)) {
1244 fputs("-1 -1 ", stdout);
1248 if ((sample_type & PERF_SAMPLE_CPU))
1249 printf("%u ", sample->cpu);
1251 if (sample_type & PERF_SAMPLE_TIME)
1252 printf("%" PRIu64 " ", sample->time);
1255 static void sample_read__printf(struct perf_sample *sample, u64 read_format)
1257 printf("... sample_read:\n");
1259 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1260 printf("...... time enabled %016" PRIx64 "\n",
1261 sample->read.time_enabled);
1263 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1264 printf("...... time running %016" PRIx64 "\n",
1265 sample->read.time_running);
1267 if (read_format & PERF_FORMAT_GROUP) {
1270 printf(".... group nr %" PRIu64 "\n", sample->read.group.nr);
1272 for (i = 0; i < sample->read.group.nr; i++) {
1273 struct sample_read_value *value;
1275 value = &sample->read.group.values[i];
1276 printf("..... id %016" PRIx64
1277 ", value %016" PRIx64 "\n",
1278 value->id, value->value);
1281 printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n",
1282 sample->read.one.id, sample->read.one.value);
1285 static void dump_event(struct evlist *evlist, union perf_event *event,
1286 u64 file_offset, struct perf_sample *sample,
1287 const char *file_path)
1292 printf("\n%#" PRIx64 "@%s [%#x]: event: %d\n",
1293 file_offset, file_path, event->header.size, event->header.type);
1296 if (event->header.type == PERF_RECORD_SAMPLE && evlist->trace_event_sample_raw)
1297 evlist->trace_event_sample_raw(evlist, event, sample);
1300 evlist__print_tstamp(evlist, event, sample);
1302 printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset,
1303 event->header.size, perf_event__name(event->header.type));
1306 char *get_page_size_name(u64 size, char *str)
1308 if (!size || !unit_number__scnprintf(str, PAGE_SIZE_NAME_LEN, size))
1309 snprintf(str, PAGE_SIZE_NAME_LEN, "%s", "N/A");
1314 static void dump_sample(struct evsel *evsel, union perf_event *event,
1315 struct perf_sample *sample, const char *arch)
1318 char str[PAGE_SIZE_NAME_LEN];
1323 printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n",
1324 event->header.misc, sample->pid, sample->tid, sample->ip,
1325 sample->period, sample->addr);
1327 sample_type = evsel->core.attr.sample_type;
1329 if (evsel__has_callchain(evsel))
1330 callchain__printf(evsel, sample);
1332 if (evsel__has_br_stack(evsel))
1333 branch_stack__printf(sample, evsel__has_branch_callstack(evsel));
1335 if (sample_type & PERF_SAMPLE_REGS_USER)
1336 regs_user__printf(sample, arch);
1338 if (sample_type & PERF_SAMPLE_REGS_INTR)
1339 regs_intr__printf(sample, arch);
1341 if (sample_type & PERF_SAMPLE_STACK_USER)
1342 stack_user__printf(&sample->user_stack);
1344 if (sample_type & PERF_SAMPLE_WEIGHT_TYPE) {
1345 printf("... weight: %" PRIu64 "", sample->weight);
1346 if (sample_type & PERF_SAMPLE_WEIGHT_STRUCT) {
1347 printf(",0x%"PRIx16"", sample->ins_lat);
1348 printf(",0x%"PRIx16"", sample->p_stage_cyc);
1353 if (sample_type & PERF_SAMPLE_DATA_SRC)
1354 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src);
1356 if (sample_type & PERF_SAMPLE_PHYS_ADDR)
1357 printf(" .. phys_addr: 0x%"PRIx64"\n", sample->phys_addr);
1359 if (sample_type & PERF_SAMPLE_DATA_PAGE_SIZE)
1360 printf(" .. data page size: %s\n", get_page_size_name(sample->data_page_size, str));
1362 if (sample_type & PERF_SAMPLE_CODE_PAGE_SIZE)
1363 printf(" .. code page size: %s\n", get_page_size_name(sample->code_page_size, str));
1365 if (sample_type & PERF_SAMPLE_TRANSACTION)
1366 printf("... transaction: %" PRIx64 "\n", sample->transaction);
1368 if (sample_type & PERF_SAMPLE_READ)
1369 sample_read__printf(sample, evsel->core.attr.read_format);
1372 static void dump_read(struct evsel *evsel, union perf_event *event)
1374 struct perf_record_read *read_event = &event->read;
1380 printf(": %d %d %s %" PRI_lu64 "\n", event->read.pid, event->read.tid,
1381 evsel__name(evsel), event->read.value);
1386 read_format = evsel->core.attr.read_format;
1388 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1389 printf("... time enabled : %" PRI_lu64 "\n", read_event->time_enabled);
1391 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1392 printf("... time running : %" PRI_lu64 "\n", read_event->time_running);
1394 if (read_format & PERF_FORMAT_ID)
1395 printf("... id : %" PRI_lu64 "\n", read_event->id);
1398 static struct machine *machines__find_for_cpumode(struct machines *machines,
1399 union perf_event *event,
1400 struct perf_sample *sample)
1403 ((sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL) ||
1404 (sample->cpumode == PERF_RECORD_MISC_GUEST_USER))) {
1407 if (event->header.type == PERF_RECORD_MMAP
1408 || event->header.type == PERF_RECORD_MMAP2)
1409 pid = event->mmap.pid;
1413 return machines__find_guest(machines, pid);
1416 return &machines->host;
1419 static int deliver_sample_value(struct evlist *evlist,
1420 struct perf_tool *tool,
1421 union perf_event *event,
1422 struct perf_sample *sample,
1423 struct sample_read_value *v,
1424 struct machine *machine)
1426 struct perf_sample_id *sid = evlist__id2sid(evlist, v->id);
1427 struct evsel *evsel;
1431 sample->period = v->value - sid->period;
1432 sid->period = v->value;
1435 if (!sid || sid->evsel == NULL) {
1436 ++evlist->stats.nr_unknown_id;
1441 * There's no reason to deliver sample
1442 * for zero period, bail out.
1444 if (!sample->period)
1447 evsel = container_of(sid->evsel, struct evsel, core);
1448 return tool->sample(tool, event, sample, evsel, machine);
1451 static int deliver_sample_group(struct evlist *evlist,
1452 struct perf_tool *tool,
1453 union perf_event *event,
1454 struct perf_sample *sample,
1455 struct machine *machine)
1460 for (i = 0; i < sample->read.group.nr; i++) {
1461 ret = deliver_sample_value(evlist, tool, event, sample,
1462 &sample->read.group.values[i],
1471 static int evlist__deliver_sample(struct evlist *evlist, struct perf_tool *tool,
1472 union perf_event *event, struct perf_sample *sample,
1473 struct evsel *evsel, struct machine *machine)
1475 /* We know evsel != NULL. */
1476 u64 sample_type = evsel->core.attr.sample_type;
1477 u64 read_format = evsel->core.attr.read_format;
1479 /* Standard sample delivery. */
1480 if (!(sample_type & PERF_SAMPLE_READ))
1481 return tool->sample(tool, event, sample, evsel, machine);
1483 /* For PERF_SAMPLE_READ we have either single or group mode. */
1484 if (read_format & PERF_FORMAT_GROUP)
1485 return deliver_sample_group(evlist, tool, event, sample,
1488 return deliver_sample_value(evlist, tool, event, sample,
1489 &sample->read.one, machine);
1492 static int machines__deliver_event(struct machines *machines,
1493 struct evlist *evlist,
1494 union perf_event *event,
1495 struct perf_sample *sample,
1496 struct perf_tool *tool, u64 file_offset,
1497 const char *file_path)
1499 struct evsel *evsel;
1500 struct machine *machine;
1502 dump_event(evlist, event, file_offset, sample, file_path);
1504 evsel = evlist__id2evsel(evlist, sample->id);
1506 machine = machines__find_for_cpumode(machines, event, sample);
1508 switch (event->header.type) {
1509 case PERF_RECORD_SAMPLE:
1510 if (evsel == NULL) {
1511 ++evlist->stats.nr_unknown_id;
1514 if (machine == NULL) {
1515 ++evlist->stats.nr_unprocessable_samples;
1516 dump_sample(evsel, event, sample, perf_env__arch(NULL));
1519 dump_sample(evsel, event, sample, perf_env__arch(machine->env));
1520 return evlist__deliver_sample(evlist, tool, event, sample, evsel, machine);
1521 case PERF_RECORD_MMAP:
1522 return tool->mmap(tool, event, sample, machine);
1523 case PERF_RECORD_MMAP2:
1524 if (event->header.misc & PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT)
1525 ++evlist->stats.nr_proc_map_timeout;
1526 return tool->mmap2(tool, event, sample, machine);
1527 case PERF_RECORD_COMM:
1528 return tool->comm(tool, event, sample, machine);
1529 case PERF_RECORD_NAMESPACES:
1530 return tool->namespaces(tool, event, sample, machine);
1531 case PERF_RECORD_CGROUP:
1532 return tool->cgroup(tool, event, sample, machine);
1533 case PERF_RECORD_FORK:
1534 return tool->fork(tool, event, sample, machine);
1535 case PERF_RECORD_EXIT:
1536 return tool->exit(tool, event, sample, machine);
1537 case PERF_RECORD_LOST:
1538 if (tool->lost == perf_event__process_lost)
1539 evlist->stats.total_lost += event->lost.lost;
1540 return tool->lost(tool, event, sample, machine);
1541 case PERF_RECORD_LOST_SAMPLES:
1542 if (tool->lost_samples == perf_event__process_lost_samples)
1543 evlist->stats.total_lost_samples += event->lost_samples.lost;
1544 return tool->lost_samples(tool, event, sample, machine);
1545 case PERF_RECORD_READ:
1546 dump_read(evsel, event);
1547 return tool->read(tool, event, sample, evsel, machine);
1548 case PERF_RECORD_THROTTLE:
1549 return tool->throttle(tool, event, sample, machine);
1550 case PERF_RECORD_UNTHROTTLE:
1551 return tool->unthrottle(tool, event, sample, machine);
1552 case PERF_RECORD_AUX:
1553 if (tool->aux == perf_event__process_aux) {
1554 if (event->aux.flags & PERF_AUX_FLAG_TRUNCATED)
1555 evlist->stats.total_aux_lost += 1;
1556 if (event->aux.flags & PERF_AUX_FLAG_PARTIAL)
1557 evlist->stats.total_aux_partial += 1;
1558 if (event->aux.flags & PERF_AUX_FLAG_COLLISION)
1559 evlist->stats.total_aux_collision += 1;
1561 return tool->aux(tool, event, sample, machine);
1562 case PERF_RECORD_ITRACE_START:
1563 return tool->itrace_start(tool, event, sample, machine);
1564 case PERF_RECORD_SWITCH:
1565 case PERF_RECORD_SWITCH_CPU_WIDE:
1566 return tool->context_switch(tool, event, sample, machine);
1567 case PERF_RECORD_KSYMBOL:
1568 return tool->ksymbol(tool, event, sample, machine);
1569 case PERF_RECORD_BPF_EVENT:
1570 return tool->bpf(tool, event, sample, machine);
1571 case PERF_RECORD_TEXT_POKE:
1572 return tool->text_poke(tool, event, sample, machine);
1573 case PERF_RECORD_AUX_OUTPUT_HW_ID:
1574 return tool->aux_output_hw_id(tool, event, sample, machine);
1576 ++evlist->stats.nr_unknown_events;
1581 static int perf_session__deliver_event(struct perf_session *session,
1582 union perf_event *event,
1583 struct perf_tool *tool,
1585 const char *file_path)
1587 struct perf_sample sample;
1588 int ret = evlist__parse_sample(session->evlist, event, &sample);
1591 pr_err("Can't parse sample, err = %d\n", ret);
1595 ret = auxtrace__process_event(session, event, &sample, tool);
1601 ret = machines__deliver_event(&session->machines, session->evlist,
1602 event, &sample, tool, file_offset, file_path);
1604 if (dump_trace && sample.aux_sample.size)
1605 auxtrace__dump_auxtrace_sample(session, &sample);
1610 static s64 perf_session__process_user_event(struct perf_session *session,
1611 union perf_event *event,
1613 const char *file_path)
1615 struct ordered_events *oe = &session->ordered_events;
1616 struct perf_tool *tool = session->tool;
1617 struct perf_sample sample = { .time = 0, };
1618 int fd = perf_data__fd(session->data);
1621 if (event->header.type != PERF_RECORD_COMPRESSED ||
1622 tool->compressed == perf_session__process_compressed_event_stub)
1623 dump_event(session->evlist, event, file_offset, &sample, file_path);
1625 /* These events are processed right away */
1626 switch (event->header.type) {
1627 case PERF_RECORD_HEADER_ATTR:
1628 err = tool->attr(tool, event, &session->evlist);
1630 perf_session__set_id_hdr_size(session);
1631 perf_session__set_comm_exec(session);
1634 case PERF_RECORD_EVENT_UPDATE:
1635 return tool->event_update(tool, event, &session->evlist);
1636 case PERF_RECORD_HEADER_EVENT_TYPE:
1638 * Deprecated, but we need to handle it for sake
1639 * of old data files create in pipe mode.
1642 case PERF_RECORD_HEADER_TRACING_DATA:
1644 * Setup for reading amidst mmap, but only when we
1645 * are in 'file' mode. The 'pipe' fd is in proper
1648 if (!perf_data__is_pipe(session->data))
1649 lseek(fd, file_offset, SEEK_SET);
1650 return tool->tracing_data(session, event);
1651 case PERF_RECORD_HEADER_BUILD_ID:
1652 return tool->build_id(session, event);
1653 case PERF_RECORD_FINISHED_ROUND:
1654 return tool->finished_round(tool, event, oe);
1655 case PERF_RECORD_ID_INDEX:
1656 return tool->id_index(session, event);
1657 case PERF_RECORD_AUXTRACE_INFO:
1658 return tool->auxtrace_info(session, event);
1659 case PERF_RECORD_AUXTRACE:
1660 /* setup for reading amidst mmap */
1661 lseek(fd, file_offset + event->header.size, SEEK_SET);
1662 return tool->auxtrace(session, event);
1663 case PERF_RECORD_AUXTRACE_ERROR:
1664 perf_session__auxtrace_error_inc(session, event);
1665 return tool->auxtrace_error(session, event);
1666 case PERF_RECORD_THREAD_MAP:
1667 return tool->thread_map(session, event);
1668 case PERF_RECORD_CPU_MAP:
1669 return tool->cpu_map(session, event);
1670 case PERF_RECORD_STAT_CONFIG:
1671 return tool->stat_config(session, event);
1672 case PERF_RECORD_STAT:
1673 return tool->stat(session, event);
1674 case PERF_RECORD_STAT_ROUND:
1675 return tool->stat_round(session, event);
1676 case PERF_RECORD_TIME_CONV:
1677 session->time_conv = event->time_conv;
1678 return tool->time_conv(session, event);
1679 case PERF_RECORD_HEADER_FEATURE:
1680 return tool->feature(session, event);
1681 case PERF_RECORD_COMPRESSED:
1682 err = tool->compressed(session, event, file_offset, file_path);
1684 dump_event(session->evlist, event, file_offset, &sample, file_path);
1691 int perf_session__deliver_synth_event(struct perf_session *session,
1692 union perf_event *event,
1693 struct perf_sample *sample)
1695 struct evlist *evlist = session->evlist;
1696 struct perf_tool *tool = session->tool;
1698 events_stats__inc(&evlist->stats, event->header.type);
1700 if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1701 return perf_session__process_user_event(session, event, 0, NULL);
1703 return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0, NULL);
1706 static void event_swap(union perf_event *event, bool sample_id_all)
1708 perf_event__swap_op swap;
1710 swap = perf_event__swap_ops[event->header.type];
1712 swap(event, sample_id_all);
1715 int perf_session__peek_event(struct perf_session *session, off_t file_offset,
1716 void *buf, size_t buf_sz,
1717 union perf_event **event_ptr,
1718 struct perf_sample *sample)
1720 union perf_event *event;
1721 size_t hdr_sz, rest;
1724 if (session->one_mmap && !session->header.needs_swap) {
1725 event = file_offset - session->one_mmap_offset +
1726 session->one_mmap_addr;
1727 goto out_parse_sample;
1730 if (perf_data__is_pipe(session->data))
1733 fd = perf_data__fd(session->data);
1734 hdr_sz = sizeof(struct perf_event_header);
1736 if (buf_sz < hdr_sz)
1739 if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 ||
1740 readn(fd, buf, hdr_sz) != (ssize_t)hdr_sz)
1743 event = (union perf_event *)buf;
1745 if (session->header.needs_swap)
1746 perf_event_header__bswap(&event->header);
1748 if (event->header.size < hdr_sz || event->header.size > buf_sz)
1752 rest = event->header.size - hdr_sz;
1754 if (readn(fd, buf, rest) != (ssize_t)rest)
1757 if (session->header.needs_swap)
1758 event_swap(event, evlist__sample_id_all(session->evlist));
1762 if (sample && event->header.type < PERF_RECORD_USER_TYPE_START &&
1763 evlist__parse_sample(session->evlist, event, sample))
1771 int perf_session__peek_events(struct perf_session *session, u64 offset,
1772 u64 size, peek_events_cb_t cb, void *data)
1774 u64 max_offset = offset + size;
1775 char buf[PERF_SAMPLE_MAX_SIZE];
1776 union perf_event *event;
1780 err = perf_session__peek_event(session, offset, buf,
1781 PERF_SAMPLE_MAX_SIZE, &event,
1786 err = cb(session, event, offset, data);
1790 offset += event->header.size;
1791 if (event->header.type == PERF_RECORD_AUXTRACE)
1792 offset += event->auxtrace.size;
1794 } while (offset < max_offset);
1799 static s64 perf_session__process_event(struct perf_session *session,
1800 union perf_event *event, u64 file_offset,
1801 const char *file_path)
1803 struct evlist *evlist = session->evlist;
1804 struct perf_tool *tool = session->tool;
1807 if (session->header.needs_swap)
1808 event_swap(event, evlist__sample_id_all(evlist));
1810 if (event->header.type >= PERF_RECORD_HEADER_MAX)
1813 events_stats__inc(&evlist->stats, event->header.type);
1815 if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1816 return perf_session__process_user_event(session, event, file_offset, file_path);
1818 if (tool->ordered_events) {
1819 u64 timestamp = -1ULL;
1821 ret = evlist__parse_sample_timestamp(evlist, event, ×tamp);
1822 if (ret && ret != -1)
1825 ret = perf_session__queue_event(session, event, timestamp, file_offset, file_path);
1830 return perf_session__deliver_event(session, event, tool, file_offset, file_path);
1833 void perf_event_header__bswap(struct perf_event_header *hdr)
1835 hdr->type = bswap_32(hdr->type);
1836 hdr->misc = bswap_16(hdr->misc);
1837 hdr->size = bswap_16(hdr->size);
1840 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid)
1842 return machine__findnew_thread(&session->machines.host, -1, pid);
1845 int perf_session__register_idle_thread(struct perf_session *session)
1847 struct thread *thread = machine__idle_thread(&session->machines.host);
1849 /* machine__idle_thread() got the thread, so put it */
1850 thread__put(thread);
1851 return thread ? 0 : -1;
1855 perf_session__warn_order(const struct perf_session *session)
1857 const struct ordered_events *oe = &session->ordered_events;
1858 struct evsel *evsel;
1859 bool should_warn = true;
1861 evlist__for_each_entry(session->evlist, evsel) {
1862 if (evsel->core.attr.write_backward)
1863 should_warn = false;
1868 if (oe->nr_unordered_events != 0)
1869 ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events);
1872 static void perf_session__warn_about_errors(const struct perf_session *session)
1874 const struct events_stats *stats = &session->evlist->stats;
1876 if (session->tool->lost == perf_event__process_lost &&
1877 stats->nr_events[PERF_RECORD_LOST] != 0) {
1878 ui__warning("Processed %d events and lost %d chunks!\n\n"
1879 "Check IO/CPU overload!\n\n",
1880 stats->nr_events[0],
1881 stats->nr_events[PERF_RECORD_LOST]);
1884 if (session->tool->lost_samples == perf_event__process_lost_samples) {
1887 drop_rate = (double)stats->total_lost_samples /
1888 (double) (stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples);
1889 if (drop_rate > 0.05) {
1890 ui__warning("Processed %" PRIu64 " samples and lost %3.2f%%!\n\n",
1891 stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples,
1896 if (session->tool->aux == perf_event__process_aux &&
1897 stats->total_aux_lost != 0) {
1898 ui__warning("AUX data lost %" PRIu64 " times out of %u!\n\n",
1899 stats->total_aux_lost,
1900 stats->nr_events[PERF_RECORD_AUX]);
1903 if (session->tool->aux == perf_event__process_aux &&
1904 stats->total_aux_partial != 0) {
1905 bool vmm_exclusive = false;
1907 (void)sysfs__read_bool("module/kvm_intel/parameters/vmm_exclusive",
1910 ui__warning("AUX data had gaps in it %" PRIu64 " times out of %u!\n\n"
1911 "Are you running a KVM guest in the background?%s\n\n",
1912 stats->total_aux_partial,
1913 stats->nr_events[PERF_RECORD_AUX],
1915 "\nReloading kvm_intel module with vmm_exclusive=0\n"
1916 "will reduce the gaps to only guest's timeslices." :
1920 if (session->tool->aux == perf_event__process_aux &&
1921 stats->total_aux_collision != 0) {
1922 ui__warning("AUX data detected collision %" PRIu64 " times out of %u!\n\n",
1923 stats->total_aux_collision,
1924 stats->nr_events[PERF_RECORD_AUX]);
1927 if (stats->nr_unknown_events != 0) {
1928 ui__warning("Found %u unknown events!\n\n"
1929 "Is this an older tool processing a perf.data "
1930 "file generated by a more recent tool?\n\n"
1931 "If that is not the case, consider "
1932 "reporting to linux-kernel@vger.kernel.org.\n\n",
1933 stats->nr_unknown_events);
1936 if (stats->nr_unknown_id != 0) {
1937 ui__warning("%u samples with id not present in the header\n",
1938 stats->nr_unknown_id);
1941 if (stats->nr_invalid_chains != 0) {
1942 ui__warning("Found invalid callchains!\n\n"
1943 "%u out of %u events were discarded for this reason.\n\n"
1944 "Consider reporting to linux-kernel@vger.kernel.org.\n\n",
1945 stats->nr_invalid_chains,
1946 stats->nr_events[PERF_RECORD_SAMPLE]);
1949 if (stats->nr_unprocessable_samples != 0) {
1950 ui__warning("%u unprocessable samples recorded.\n"
1951 "Do you have a KVM guest running and not using 'perf kvm'?\n",
1952 stats->nr_unprocessable_samples);
1955 perf_session__warn_order(session);
1957 events_stats__auxtrace_error_warn(stats);
1959 if (stats->nr_proc_map_timeout != 0) {
1960 ui__warning("%d map information files for pre-existing threads were\n"
1961 "not processed, if there are samples for addresses they\n"
1962 "will not be resolved, you may find out which are these\n"
1963 "threads by running with -v and redirecting the output\n"
1965 "The time limit to process proc map is too short?\n"
1966 "Increase it by --proc-map-timeout\n",
1967 stats->nr_proc_map_timeout);
1971 static int perf_session__flush_thread_stack(struct thread *thread,
1972 void *p __maybe_unused)
1974 return thread_stack__flush(thread);
1977 static int perf_session__flush_thread_stacks(struct perf_session *session)
1979 return machines__for_each_thread(&session->machines,
1980 perf_session__flush_thread_stack,
1984 volatile int session_done;
1986 static int __perf_session__process_decomp_events(struct perf_session *session);
1988 static int __perf_session__process_pipe_events(struct perf_session *session)
1990 struct ordered_events *oe = &session->ordered_events;
1991 struct perf_tool *tool = session->tool;
1992 union perf_event *event;
1993 uint32_t size, cur_size = 0;
2000 perf_tool__fill_defaults(tool);
2003 cur_size = sizeof(union perf_event);
2005 buf = malloc(cur_size);
2008 ordered_events__set_copy_on_queue(oe, true);
2011 err = perf_data__read(session->data, event,
2012 sizeof(struct perf_event_header));
2017 pr_err("failed to read event header\n");
2021 if (session->header.needs_swap)
2022 perf_event_header__bswap(&event->header);
2024 size = event->header.size;
2025 if (size < sizeof(struct perf_event_header)) {
2026 pr_err("bad event header size\n");
2030 if (size > cur_size) {
2031 void *new = realloc(buf, size);
2033 pr_err("failed to allocate memory to read event\n");
2041 p += sizeof(struct perf_event_header);
2043 if (size - sizeof(struct perf_event_header)) {
2044 err = perf_data__read(session->data, p,
2045 size - sizeof(struct perf_event_header));
2048 pr_err("unexpected end of event stream\n");
2052 pr_err("failed to read event data\n");
2057 if ((skip = perf_session__process_event(session, event, head, "pipe")) < 0) {
2058 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
2059 head, event->header.size, event->header.type);
2069 err = __perf_session__process_decomp_events(session);
2073 if (!session_done())
2076 /* do the final flush for ordered samples */
2077 err = ordered_events__flush(oe, OE_FLUSH__FINAL);
2080 err = auxtrace__flush_events(session, tool);
2083 err = perf_session__flush_thread_stacks(session);
2087 perf_session__warn_about_errors(session);
2088 ordered_events__free(&session->ordered_events);
2089 auxtrace__free_events(session);
2093 static union perf_event *
2094 prefetch_event(char *buf, u64 head, size_t mmap_size,
2095 bool needs_swap, union perf_event *error)
2097 union perf_event *event;
2101 * Ensure we have enough space remaining to read
2102 * the size of the event in the headers.
2104 if (head + sizeof(event->header) > mmap_size)
2107 event = (union perf_event *)(buf + head);
2109 perf_event_header__bswap(&event->header);
2111 event_size = event->header.size;
2112 if (head + event_size <= mmap_size)
2115 /* We're not fetching the event so swap back again */
2117 perf_event_header__bswap(&event->header);
2119 /* Check if the event fits into the next mmapped buf. */
2120 if (event_size <= mmap_size - head % page_size) {
2121 /* Remap buf and fetch again. */
2125 /* Invalid input. Event size should never exceed mmap_size. */
2126 pr_debug("%s: head=%#" PRIx64 " event->header.size=%#x, mmap_size=%#zx:"
2127 " fuzzed or compressed perf.data?\n", __func__, head, event_size, mmap_size);
2132 static union perf_event *
2133 fetch_mmaped_event(u64 head, size_t mmap_size, char *buf, bool needs_swap)
2135 return prefetch_event(buf, head, mmap_size, needs_swap, ERR_PTR(-EINVAL));
2138 static union perf_event *
2139 fetch_decomp_event(u64 head, size_t mmap_size, char *buf, bool needs_swap)
2141 return prefetch_event(buf, head, mmap_size, needs_swap, NULL);
2144 static int __perf_session__process_decomp_events(struct perf_session *session)
2148 struct decomp *decomp = session->active_decomp->decomp_last;
2153 while (decomp->head < decomp->size && !session_done()) {
2154 union perf_event *event = fetch_decomp_event(decomp->head, decomp->size, decomp->data,
2155 session->header.needs_swap);
2160 size = event->header.size;
2162 if (size < sizeof(struct perf_event_header) ||
2163 (skip = perf_session__process_event(session, event, decomp->file_pos,
2164 decomp->file_path)) < 0) {
2165 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
2166 decomp->file_pos + decomp->head, event->header.size, event->header.type);
2173 decomp->head += size;
2180 * On 64bit we can mmap the data file in one go. No need for tiny mmap
2181 * slices. On 32bit we use 32MB.
2183 #if BITS_PER_LONG == 64
2184 #define MMAP_SIZE ULLONG_MAX
2187 #define MMAP_SIZE (32 * 1024 * 1024ULL)
2188 #define NUM_MMAPS 128
2193 typedef s64 (*reader_cb_t)(struct perf_session *session,
2194 union perf_event *event,
2196 const char *file_path);
2203 reader_cb_t process;
2204 bool in_place_update;
2205 char *mmaps[NUM_MMAPS];
2214 struct zstd_data zstd_data;
2215 struct decomp_data decomp_data;
2219 reader__init(struct reader *rd, bool *one_mmap)
2221 u64 data_size = rd->data_size;
2222 char **mmaps = rd->mmaps;
2224 rd->head = rd->data_offset;
2225 data_size += rd->data_offset;
2227 rd->mmap_size = MMAP_SIZE;
2228 if (rd->mmap_size > data_size) {
2229 rd->mmap_size = data_size;
2234 memset(mmaps, 0, sizeof(rd->mmaps));
2236 if (zstd_init(&rd->zstd_data, 0))
2238 rd->decomp_data.zstd_decomp = &rd->zstd_data;
2244 reader__release_decomp(struct reader *rd)
2246 perf_decomp__release_events(rd->decomp_data.decomp);
2247 zstd_fini(&rd->zstd_data);
2251 reader__mmap(struct reader *rd, struct perf_session *session)
2253 int mmap_prot, mmap_flags;
2254 char *buf, **mmaps = rd->mmaps;
2257 mmap_prot = PROT_READ;
2258 mmap_flags = MAP_SHARED;
2260 if (rd->in_place_update) {
2261 mmap_prot |= PROT_WRITE;
2262 } else if (session->header.needs_swap) {
2263 mmap_prot |= PROT_WRITE;
2264 mmap_flags = MAP_PRIVATE;
2267 if (mmaps[rd->mmap_idx]) {
2268 munmap(mmaps[rd->mmap_idx], rd->mmap_size);
2269 mmaps[rd->mmap_idx] = NULL;
2272 page_offset = page_size * (rd->head / page_size);
2273 rd->file_offset += page_offset;
2274 rd->head -= page_offset;
2276 buf = mmap(NULL, rd->mmap_size, mmap_prot, mmap_flags, rd->fd,
2278 if (buf == MAP_FAILED) {
2279 pr_err("failed to mmap file\n");
2282 mmaps[rd->mmap_idx] = rd->mmap_cur = buf;
2283 rd->mmap_idx = (rd->mmap_idx + 1) & (ARRAY_SIZE(rd->mmaps) - 1);
2284 rd->file_pos = rd->file_offset + rd->head;
2285 if (session->one_mmap) {
2286 session->one_mmap_addr = buf;
2287 session->one_mmap_offset = rd->file_offset;
2299 reader__read_event(struct reader *rd, struct perf_session *session,
2300 struct ui_progress *prog)
2303 int err = READER_OK;
2304 union perf_event *event;
2307 event = fetch_mmaped_event(rd->head, rd->mmap_size, rd->mmap_cur,
2308 session->header.needs_swap);
2310 return PTR_ERR(event);
2313 return READER_NODATA;
2315 size = event->header.size;
2319 if (size < sizeof(struct perf_event_header) ||
2320 (skip = rd->process(session, event, rd->file_pos, rd->path)) < 0) {
2321 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d [%s]\n",
2322 rd->file_offset + rd->head, event->header.size,
2323 event->header.type, strerror(-skip));
2333 rd->file_pos += size;
2335 err = __perf_session__process_decomp_events(session);
2339 ui_progress__update(prog, size);
2346 reader__eof(struct reader *rd)
2348 return (rd->file_pos >= rd->data_size + rd->data_offset);
2352 reader__process_events(struct reader *rd, struct perf_session *session,
2353 struct ui_progress *prog)
2357 err = reader__init(rd, &session->one_mmap);
2361 session->active_decomp = &rd->decomp_data;
2364 err = reader__mmap(rd, session);
2369 err = reader__read_event(rd, session, prog);
2372 else if (err == READER_NODATA)
2378 if (!reader__eof(rd))
2382 session->active_decomp = &session->decomp_data;
2386 static s64 process_simple(struct perf_session *session,
2387 union perf_event *event,
2389 const char *file_path)
2391 return perf_session__process_event(session, event, file_offset, file_path);
2394 static int __perf_session__process_events(struct perf_session *session)
2396 struct reader rd = {
2397 .fd = perf_data__fd(session->data),
2398 .path = session->data->file.path,
2399 .data_size = session->header.data_size,
2400 .data_offset = session->header.data_offset,
2401 .process = process_simple,
2402 .in_place_update = session->data->in_place_update,
2404 struct ordered_events *oe = &session->ordered_events;
2405 struct perf_tool *tool = session->tool;
2406 struct ui_progress prog;
2409 perf_tool__fill_defaults(tool);
2411 if (rd.data_size == 0)
2414 ui_progress__init_size(&prog, rd.data_size, "Processing events...");
2416 err = reader__process_events(&rd, session, &prog);
2419 /* do the final flush for ordered samples */
2420 err = ordered_events__flush(oe, OE_FLUSH__FINAL);
2423 err = auxtrace__flush_events(session, tool);
2426 err = perf_session__flush_thread_stacks(session);
2428 ui_progress__finish();
2430 perf_session__warn_about_errors(session);
2432 * We may switching perf.data output, make ordered_events
2435 ordered_events__reinit(&session->ordered_events);
2436 auxtrace__free_events(session);
2437 reader__release_decomp(&rd);
2438 session->one_mmap = false;
2443 * Processing 2 MB of data from each reader in sequence,
2444 * because that's the way the ordered events sorting works
2447 #define READER_MAX_SIZE (2 * 1024 * 1024)
2450 * This function reads, merge and process directory data.
2451 * It assumens the version 1 of directory data, where each
2452 * data file holds per-cpu data, already sorted by kernel.
2454 static int __perf_session__process_dir_events(struct perf_session *session)
2456 struct perf_data *data = session->data;
2457 struct perf_tool *tool = session->tool;
2458 int i, ret, readers, nr_readers;
2459 struct ui_progress prog;
2460 u64 total_size = perf_data__size(session->data);
2463 perf_tool__fill_defaults(tool);
2465 ui_progress__init_size(&prog, total_size, "Sorting events...");
2468 for (i = 0; i < data->dir.nr; i++) {
2469 if (data->dir.files[i].size)
2473 rd = zalloc(nr_readers * sizeof(struct reader));
2477 rd[0] = (struct reader) {
2478 .fd = perf_data__fd(session->data),
2479 .path = session->data->file.path,
2480 .data_size = session->header.data_size,
2481 .data_offset = session->header.data_offset,
2482 .process = process_simple,
2483 .in_place_update = session->data->in_place_update,
2485 ret = reader__init(&rd[0], NULL);
2488 ret = reader__mmap(&rd[0], session);
2493 for (i = 0; i < data->dir.nr; i++) {
2494 if (!data->dir.files[i].size)
2496 rd[readers] = (struct reader) {
2497 .fd = data->dir.files[i].fd,
2498 .path = data->dir.files[i].path,
2499 .data_size = data->dir.files[i].size,
2501 .process = process_simple,
2502 .in_place_update = session->data->in_place_update,
2504 ret = reader__init(&rd[readers], NULL);
2507 ret = reader__mmap(&rd[readers], session);
2519 i = (i + 1) % nr_readers;
2522 if (reader__eof(&rd[i])) {
2528 session->active_decomp = &rd[i].decomp_data;
2529 ret = reader__read_event(&rd[i], session, &prog);
2532 } else if (ret == READER_NODATA) {
2533 ret = reader__mmap(&rd[i], session);
2538 if (rd[i].size >= READER_MAX_SIZE) {
2540 i = (i + 1) % nr_readers;
2544 ret = ordered_events__flush(&session->ordered_events, OE_FLUSH__FINAL);
2548 ret = perf_session__flush_thread_stacks(session);
2550 ui_progress__finish();
2553 perf_session__warn_about_errors(session);
2556 * We may switching perf.data output, make ordered_events
2559 ordered_events__reinit(&session->ordered_events);
2561 session->one_mmap = false;
2563 session->active_decomp = &session->decomp_data;
2564 for (i = 0; i < nr_readers; i++)
2565 reader__release_decomp(&rd[i]);
2571 int perf_session__process_events(struct perf_session *session)
2573 if (perf_session__register_idle_thread(session) < 0)
2576 if (perf_data__is_pipe(session->data))
2577 return __perf_session__process_pipe_events(session);
2579 if (perf_data__is_dir(session->data) && session->data->dir.nr)
2580 return __perf_session__process_dir_events(session);
2582 return __perf_session__process_events(session);
2585 bool perf_session__has_traces(struct perf_session *session, const char *msg)
2587 struct evsel *evsel;
2589 evlist__for_each_entry(session->evlist, evsel) {
2590 if (evsel->core.attr.type == PERF_TYPE_TRACEPOINT)
2594 pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg);
2598 int map__set_kallsyms_ref_reloc_sym(struct map *map, const char *symbol_name, u64 addr)
2601 struct ref_reloc_sym *ref;
2604 ref = zalloc(sizeof(struct ref_reloc_sym));
2608 ref->name = strdup(symbol_name);
2609 if (ref->name == NULL) {
2614 bracket = strchr(ref->name, ']');
2620 kmap = map__kmap(map);
2622 kmap->ref_reloc_sym = ref;
2627 size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp)
2629 return machines__fprintf_dsos(&session->machines, fp);
2632 size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp,
2633 bool (skip)(struct dso *dso, int parm), int parm)
2635 return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm);
2638 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp,
2642 const char *msg = "";
2644 if (perf_header__has_feat(&session->header, HEADER_AUXTRACE))
2645 msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)";
2647 ret = fprintf(fp, "\nAggregated stats:%s\n", msg);
2649 ret += events_stats__fprintf(&session->evlist->stats, fp, skip_empty);
2653 size_t perf_session__fprintf(struct perf_session *session, FILE *fp)
2656 * FIXME: Here we have to actually print all the machines in this
2657 * session, not just the host...
2659 return machine__fprintf(&session->machines.host, fp);
2662 struct evsel *perf_session__find_first_evtype(struct perf_session *session,
2667 evlist__for_each_entry(session->evlist, pos) {
2668 if (pos->core.attr.type == type)
2674 int perf_session__cpu_bitmap(struct perf_session *session,
2675 const char *cpu_list, unsigned long *cpu_bitmap)
2678 struct perf_cpu_map *map;
2679 int nr_cpus = min(session->header.env.nr_cpus_avail, MAX_NR_CPUS);
2681 for (i = 0; i < PERF_TYPE_MAX; ++i) {
2682 struct evsel *evsel;
2684 evsel = perf_session__find_first_evtype(session, i);
2688 if (!(evsel->core.attr.sample_type & PERF_SAMPLE_CPU)) {
2689 pr_err("File does not contain CPU events. "
2690 "Remove -C option to proceed.\n");
2695 map = perf_cpu_map__new(cpu_list);
2697 pr_err("Invalid cpu_list\n");
2701 for (i = 0; i < perf_cpu_map__nr(map); i++) {
2702 struct perf_cpu cpu = perf_cpu_map__cpu(map, i);
2704 if (cpu.cpu >= nr_cpus) {
2705 pr_err("Requested CPU %d too large. "
2706 "Consider raising MAX_NR_CPUS\n", cpu.cpu);
2707 goto out_delete_map;
2710 set_bit(cpu.cpu, cpu_bitmap);
2716 perf_cpu_map__put(map);
2720 void perf_session__fprintf_info(struct perf_session *session, FILE *fp,
2723 if (session == NULL || fp == NULL)
2726 fprintf(fp, "# ========\n");
2727 perf_header__fprintf_info(session, fp, full);
2728 fprintf(fp, "# ========\n#\n");
2731 int perf_event__process_id_index(struct perf_session *session,
2732 union perf_event *event)
2734 struct evlist *evlist = session->evlist;
2735 struct perf_record_id_index *ie = &event->id_index;
2736 size_t i, nr, max_nr;
2738 max_nr = (ie->header.size - sizeof(struct perf_record_id_index)) /
2739 sizeof(struct id_index_entry);
2745 fprintf(stdout, " nr: %zu\n", nr);
2747 for (i = 0; i < nr; i++) {
2748 struct id_index_entry *e = &ie->entries[i];
2749 struct perf_sample_id *sid;
2752 fprintf(stdout, " ... id: %"PRI_lu64, e->id);
2753 fprintf(stdout, " idx: %"PRI_lu64, e->idx);
2754 fprintf(stdout, " cpu: %"PRI_ld64, e->cpu);
2755 fprintf(stdout, " tid: %"PRI_ld64"\n", e->tid);
2758 sid = evlist__id2sid(evlist, e->id);
2762 sid->cpu.cpu = e->cpu;