1 // SPDX-License-Identifier: GPL-2.0-only
3 * auxtrace.c: AUX area trace support
4 * Copyright (c) 2013-2015, Intel Corporation.
15 #include <linux/kernel.h>
16 #include <linux/perf_event.h>
17 #include <linux/types.h>
18 #include <linux/bitops.h>
19 #include <linux/log2.h>
20 #include <linux/string.h>
21 #include <linux/time64.h>
23 #include <sys/param.h>
26 #include <linux/list.h>
27 #include <linux/zalloc.h>
34 #include "evsel_config.h"
36 #include "util/perf_api_probe.h"
37 #include "util/synthetic-events.h"
38 #include "thread_map.h"
42 #include <linux/hash.h>
48 #include <subcmd/parse-options.h>
52 #include "intel-bts.h"
54 #include "s390-cpumsf.h"
55 #include "util/mmap.h"
57 #include <linux/ctype.h>
58 #include "symbol/kallsyms.h"
59 #include <internal/lib.h>
62 * Make a group from 'leader' to 'last', requiring that the events were not
63 * already grouped to a different leader.
65 static int evlist__regroup(struct evlist *evlist, struct evsel *leader, struct evsel *last)
70 if (!evsel__is_group_leader(leader))
74 evlist__for_each_entry(evlist, evsel) {
76 if (!(evsel__leader(evsel) == leader ||
77 (evsel__leader(evsel) == evsel &&
78 evsel->core.nr_members <= 1)))
80 } else if (evsel == leader) {
88 evlist__for_each_entry(evlist, evsel) {
90 if (!evsel__has_leader(evsel, leader)) {
91 evsel__set_leader(evsel, leader);
92 if (leader->core.nr_members < 1)
93 leader->core.nr_members = 1;
94 leader->core.nr_members += 1;
96 } else if (evsel == leader) {
106 static bool auxtrace__dont_decode(struct perf_session *session)
108 return !session->itrace_synth_opts ||
109 session->itrace_synth_opts->dont_decode;
112 int auxtrace_mmap__mmap(struct auxtrace_mmap *mm,
113 struct auxtrace_mmap_params *mp,
114 void *userpg, int fd)
116 struct perf_event_mmap_page *pc = userpg;
118 WARN_ONCE(mm->base, "Uninitialized auxtrace_mmap\n");
126 mm->cpu = mp->cpu.cpu;
133 pc->aux_offset = mp->offset;
134 pc->aux_size = mp->len;
136 mm->base = mmap(NULL, mp->len, mp->prot, MAP_SHARED, fd, mp->offset);
137 if (mm->base == MAP_FAILED) {
138 pr_debug2("failed to mmap AUX area\n");
146 void auxtrace_mmap__munmap(struct auxtrace_mmap *mm)
149 munmap(mm->base, mm->len);
154 void auxtrace_mmap_params__init(struct auxtrace_mmap_params *mp,
155 off_t auxtrace_offset,
156 unsigned int auxtrace_pages,
157 bool auxtrace_overwrite)
159 if (auxtrace_pages) {
160 mp->offset = auxtrace_offset;
161 mp->len = auxtrace_pages * (size_t)page_size;
162 mp->mask = is_power_of_2(mp->len) ? mp->len - 1 : 0;
163 mp->prot = PROT_READ | (auxtrace_overwrite ? 0 : PROT_WRITE);
164 pr_debug2("AUX area mmap length %zu\n", mp->len);
170 void auxtrace_mmap_params__set_idx(struct auxtrace_mmap_params *mp,
171 struct evlist *evlist, int idx,
177 mp->cpu = perf_cpu_map__cpu(evlist->core.cpus, idx);
178 if (evlist->core.threads)
179 mp->tid = perf_thread_map__pid(evlist->core.threads, 0);
184 mp->tid = perf_thread_map__pid(evlist->core.threads, idx);
188 #define AUXTRACE_INIT_NR_QUEUES 32
190 static struct auxtrace_queue *auxtrace_alloc_queue_array(unsigned int nr_queues)
192 struct auxtrace_queue *queue_array;
193 unsigned int max_nr_queues, i;
195 max_nr_queues = UINT_MAX / sizeof(struct auxtrace_queue);
196 if (nr_queues > max_nr_queues)
199 queue_array = calloc(nr_queues, sizeof(struct auxtrace_queue));
203 for (i = 0; i < nr_queues; i++) {
204 INIT_LIST_HEAD(&queue_array[i].head);
205 queue_array[i].priv = NULL;
211 int auxtrace_queues__init(struct auxtrace_queues *queues)
213 queues->nr_queues = AUXTRACE_INIT_NR_QUEUES;
214 queues->queue_array = auxtrace_alloc_queue_array(queues->nr_queues);
215 if (!queues->queue_array)
220 static int auxtrace_queues__grow(struct auxtrace_queues *queues,
221 unsigned int new_nr_queues)
223 unsigned int nr_queues = queues->nr_queues;
224 struct auxtrace_queue *queue_array;
228 nr_queues = AUXTRACE_INIT_NR_QUEUES;
230 while (nr_queues && nr_queues < new_nr_queues)
233 if (nr_queues < queues->nr_queues || nr_queues < new_nr_queues)
236 queue_array = auxtrace_alloc_queue_array(nr_queues);
240 for (i = 0; i < queues->nr_queues; i++) {
241 list_splice_tail(&queues->queue_array[i].head,
242 &queue_array[i].head);
243 queue_array[i].tid = queues->queue_array[i].tid;
244 queue_array[i].cpu = queues->queue_array[i].cpu;
245 queue_array[i].set = queues->queue_array[i].set;
246 queue_array[i].priv = queues->queue_array[i].priv;
249 queues->nr_queues = nr_queues;
250 queues->queue_array = queue_array;
255 static void *auxtrace_copy_data(u64 size, struct perf_session *session)
257 int fd = perf_data__fd(session->data);
261 if (size > SSIZE_MAX)
268 ret = readn(fd, p, size);
269 if (ret != (ssize_t)size) {
277 static int auxtrace_queues__queue_buffer(struct auxtrace_queues *queues,
279 struct auxtrace_buffer *buffer)
281 struct auxtrace_queue *queue;
284 if (idx >= queues->nr_queues) {
285 err = auxtrace_queues__grow(queues, idx + 1);
290 queue = &queues->queue_array[idx];
294 queue->tid = buffer->tid;
295 queue->cpu = buffer->cpu.cpu;
298 buffer->buffer_nr = queues->next_buffer_nr++;
300 list_add_tail(&buffer->list, &queue->head);
302 queues->new_data = true;
303 queues->populated = true;
308 /* Limit buffers to 32MiB on 32-bit */
309 #define BUFFER_LIMIT_FOR_32_BIT (32 * 1024 * 1024)
311 static int auxtrace_queues__split_buffer(struct auxtrace_queues *queues,
313 struct auxtrace_buffer *buffer)
315 u64 sz = buffer->size;
316 bool consecutive = false;
317 struct auxtrace_buffer *b;
320 while (sz > BUFFER_LIMIT_FOR_32_BIT) {
321 b = memdup(buffer, sizeof(struct auxtrace_buffer));
324 b->size = BUFFER_LIMIT_FOR_32_BIT;
325 b->consecutive = consecutive;
326 err = auxtrace_queues__queue_buffer(queues, idx, b);
328 auxtrace_buffer__free(b);
331 buffer->data_offset += BUFFER_LIMIT_FOR_32_BIT;
332 sz -= BUFFER_LIMIT_FOR_32_BIT;
337 buffer->consecutive = consecutive;
342 static bool filter_cpu(struct perf_session *session, struct perf_cpu cpu)
344 unsigned long *cpu_bitmap = session->itrace_synth_opts->cpu_bitmap;
346 return cpu_bitmap && cpu.cpu != -1 && !test_bit(cpu.cpu, cpu_bitmap);
349 static int auxtrace_queues__add_buffer(struct auxtrace_queues *queues,
350 struct perf_session *session,
352 struct auxtrace_buffer *buffer,
353 struct auxtrace_buffer **buffer_ptr)
357 if (filter_cpu(session, buffer->cpu))
360 buffer = memdup(buffer, sizeof(*buffer));
364 if (session->one_mmap) {
365 buffer->data = buffer->data_offset - session->one_mmap_offset +
366 session->one_mmap_addr;
367 } else if (perf_data__is_pipe(session->data)) {
368 buffer->data = auxtrace_copy_data(buffer->size, session);
371 buffer->data_needs_freeing = true;
372 } else if (BITS_PER_LONG == 32 &&
373 buffer->size > BUFFER_LIMIT_FOR_32_BIT) {
374 err = auxtrace_queues__split_buffer(queues, idx, buffer);
379 err = auxtrace_queues__queue_buffer(queues, idx, buffer);
383 /* FIXME: Doesn't work for split buffer */
385 *buffer_ptr = buffer;
390 auxtrace_buffer__free(buffer);
394 int auxtrace_queues__add_event(struct auxtrace_queues *queues,
395 struct perf_session *session,
396 union perf_event *event, off_t data_offset,
397 struct auxtrace_buffer **buffer_ptr)
399 struct auxtrace_buffer buffer = {
401 .tid = event->auxtrace.tid,
402 .cpu = { event->auxtrace.cpu },
403 .data_offset = data_offset,
404 .offset = event->auxtrace.offset,
405 .reference = event->auxtrace.reference,
406 .size = event->auxtrace.size,
408 unsigned int idx = event->auxtrace.idx;
410 return auxtrace_queues__add_buffer(queues, session, idx, &buffer,
414 static int auxtrace_queues__add_indexed_event(struct auxtrace_queues *queues,
415 struct perf_session *session,
416 off_t file_offset, size_t sz)
418 union perf_event *event;
420 char buf[PERF_SAMPLE_MAX_SIZE];
422 err = perf_session__peek_event(session, file_offset, buf,
423 PERF_SAMPLE_MAX_SIZE, &event, NULL);
427 if (event->header.type == PERF_RECORD_AUXTRACE) {
428 if (event->header.size < sizeof(struct perf_record_auxtrace) ||
429 event->header.size != sz) {
433 file_offset += event->header.size;
434 err = auxtrace_queues__add_event(queues, session, event,
441 void auxtrace_queues__free(struct auxtrace_queues *queues)
445 for (i = 0; i < queues->nr_queues; i++) {
446 while (!list_empty(&queues->queue_array[i].head)) {
447 struct auxtrace_buffer *buffer;
449 buffer = list_entry(queues->queue_array[i].head.next,
450 struct auxtrace_buffer, list);
451 list_del_init(&buffer->list);
452 auxtrace_buffer__free(buffer);
456 zfree(&queues->queue_array);
457 queues->nr_queues = 0;
460 static void auxtrace_heapify(struct auxtrace_heap_item *heap_array,
461 unsigned int pos, unsigned int queue_nr,
467 parent = (pos - 1) >> 1;
468 if (heap_array[parent].ordinal <= ordinal)
470 heap_array[pos] = heap_array[parent];
473 heap_array[pos].queue_nr = queue_nr;
474 heap_array[pos].ordinal = ordinal;
477 int auxtrace_heap__add(struct auxtrace_heap *heap, unsigned int queue_nr,
480 struct auxtrace_heap_item *heap_array;
482 if (queue_nr >= heap->heap_sz) {
483 unsigned int heap_sz = AUXTRACE_INIT_NR_QUEUES;
485 while (heap_sz <= queue_nr)
487 heap_array = realloc(heap->heap_array,
488 heap_sz * sizeof(struct auxtrace_heap_item));
491 heap->heap_array = heap_array;
492 heap->heap_sz = heap_sz;
495 auxtrace_heapify(heap->heap_array, heap->heap_cnt++, queue_nr, ordinal);
500 void auxtrace_heap__free(struct auxtrace_heap *heap)
502 zfree(&heap->heap_array);
507 void auxtrace_heap__pop(struct auxtrace_heap *heap)
509 unsigned int pos, last, heap_cnt = heap->heap_cnt;
510 struct auxtrace_heap_item *heap_array;
517 heap_array = heap->heap_array;
521 unsigned int left, right;
523 left = (pos << 1) + 1;
524 if (left >= heap_cnt)
527 if (right >= heap_cnt) {
528 heap_array[pos] = heap_array[left];
531 if (heap_array[left].ordinal < heap_array[right].ordinal) {
532 heap_array[pos] = heap_array[left];
535 heap_array[pos] = heap_array[right];
541 auxtrace_heapify(heap_array, pos, heap_array[last].queue_nr,
542 heap_array[last].ordinal);
545 size_t auxtrace_record__info_priv_size(struct auxtrace_record *itr,
546 struct evlist *evlist)
549 return itr->info_priv_size(itr, evlist);
553 static int auxtrace_not_supported(void)
555 pr_err("AUX area tracing is not supported on this architecture\n");
559 int auxtrace_record__info_fill(struct auxtrace_record *itr,
560 struct perf_session *session,
561 struct perf_record_auxtrace_info *auxtrace_info,
565 return itr->info_fill(itr, session, auxtrace_info, priv_size);
566 return auxtrace_not_supported();
569 void auxtrace_record__free(struct auxtrace_record *itr)
575 int auxtrace_record__snapshot_start(struct auxtrace_record *itr)
577 if (itr && itr->snapshot_start)
578 return itr->snapshot_start(itr);
582 int auxtrace_record__snapshot_finish(struct auxtrace_record *itr, bool on_exit)
584 if (!on_exit && itr && itr->snapshot_finish)
585 return itr->snapshot_finish(itr);
589 int auxtrace_record__find_snapshot(struct auxtrace_record *itr, int idx,
590 struct auxtrace_mmap *mm,
591 unsigned char *data, u64 *head, u64 *old)
593 if (itr && itr->find_snapshot)
594 return itr->find_snapshot(itr, idx, mm, data, head, old);
598 int auxtrace_record__options(struct auxtrace_record *itr,
599 struct evlist *evlist,
600 struct record_opts *opts)
603 itr->evlist = evlist;
604 return itr->recording_options(itr, evlist, opts);
609 u64 auxtrace_record__reference(struct auxtrace_record *itr)
612 return itr->reference(itr);
616 int auxtrace_parse_snapshot_options(struct auxtrace_record *itr,
617 struct record_opts *opts, const char *str)
622 /* PMU-agnostic options */
625 opts->auxtrace_snapshot_on_exit = true;
632 if (itr && itr->parse_snapshot_options)
633 return itr->parse_snapshot_options(itr, opts, str);
635 pr_err("No AUX area tracing to snapshot\n");
639 int auxtrace_record__read_finish(struct auxtrace_record *itr, int idx)
643 if (!itr->evlist || !itr->pmu)
646 evlist__for_each_entry(itr->evlist, evsel) {
647 if (evsel->core.attr.type == itr->pmu->type) {
650 return evlist__enable_event_idx(itr->evlist, evsel, idx);
657 * Event record size is 16-bit which results in a maximum size of about 64KiB.
658 * Allow about 4KiB for the rest of the sample record, to give a maximum
659 * AUX area sample size of 60KiB.
661 #define MAX_AUX_SAMPLE_SIZE (60 * 1024)
663 /* Arbitrary default size if no other default provided */
664 #define DEFAULT_AUX_SAMPLE_SIZE (4 * 1024)
666 static int auxtrace_validate_aux_sample_size(struct evlist *evlist,
667 struct record_opts *opts)
670 bool has_aux_leader = false;
673 evlist__for_each_entry(evlist, evsel) {
674 sz = evsel->core.attr.aux_sample_size;
675 if (evsel__is_group_leader(evsel)) {
676 has_aux_leader = evsel__is_aux_event(evsel);
679 pr_err("Cannot add AUX area sampling to an AUX area event\n");
681 pr_err("Cannot add AUX area sampling to a group leader\n");
685 if (sz > MAX_AUX_SAMPLE_SIZE) {
686 pr_err("AUX area sample size %u too big, max. %d\n",
687 sz, MAX_AUX_SAMPLE_SIZE);
691 if (!has_aux_leader) {
692 pr_err("Cannot add AUX area sampling because group leader is not an AUX area event\n");
695 evsel__set_sample_bit(evsel, AUX);
696 opts->auxtrace_sample_mode = true;
698 evsel__reset_sample_bit(evsel, AUX);
702 if (!opts->auxtrace_sample_mode) {
703 pr_err("AUX area sampling requires an AUX area event group leader plus other events to which to add samples\n");
707 if (!perf_can_aux_sample()) {
708 pr_err("AUX area sampling is not supported by kernel\n");
715 int auxtrace_parse_sample_options(struct auxtrace_record *itr,
716 struct evlist *evlist,
717 struct record_opts *opts, const char *str)
719 struct evsel_config_term *term;
720 struct evsel *aux_evsel;
721 bool has_aux_sample_size = false;
722 bool has_aux_leader = false;
731 pr_err("No AUX area event to sample\n");
735 sz = strtoul(str, &endptr, 0);
736 if (*endptr || sz > UINT_MAX) {
737 pr_err("Bad AUX area sampling option: '%s'\n", str);
742 sz = itr->default_aux_sample_size;
745 sz = DEFAULT_AUX_SAMPLE_SIZE;
747 /* Set aux_sample_size based on --aux-sample option */
748 evlist__for_each_entry(evlist, evsel) {
749 if (evsel__is_group_leader(evsel)) {
750 has_aux_leader = evsel__is_aux_event(evsel);
751 } else if (has_aux_leader) {
752 evsel->core.attr.aux_sample_size = sz;
757 /* Override with aux_sample_size from config term */
758 evlist__for_each_entry(evlist, evsel) {
759 if (evsel__is_aux_event(evsel))
761 term = evsel__get_config_term(evsel, AUX_SAMPLE_SIZE);
763 has_aux_sample_size = true;
764 evsel->core.attr.aux_sample_size = term->val.aux_sample_size;
765 /* If possible, group with the AUX event */
766 if (aux_evsel && evsel->core.attr.aux_sample_size)
767 evlist__regroup(evlist, aux_evsel, evsel);
771 if (!str && !has_aux_sample_size)
775 pr_err("No AUX area event to sample\n");
779 return auxtrace_validate_aux_sample_size(evlist, opts);
782 void auxtrace_regroup_aux_output(struct evlist *evlist)
784 struct evsel *evsel, *aux_evsel = NULL;
785 struct evsel_config_term *term;
787 evlist__for_each_entry(evlist, evsel) {
788 if (evsel__is_aux_event(evsel))
790 term = evsel__get_config_term(evsel, AUX_OUTPUT);
791 /* If possible, group with the AUX event */
792 if (term && aux_evsel)
793 evlist__regroup(evlist, aux_evsel, evsel);
797 struct auxtrace_record *__weak
798 auxtrace_record__init(struct evlist *evlist __maybe_unused, int *err)
804 static int auxtrace_index__alloc(struct list_head *head)
806 struct auxtrace_index *auxtrace_index;
808 auxtrace_index = malloc(sizeof(struct auxtrace_index));
812 auxtrace_index->nr = 0;
813 INIT_LIST_HEAD(&auxtrace_index->list);
815 list_add_tail(&auxtrace_index->list, head);
820 void auxtrace_index__free(struct list_head *head)
822 struct auxtrace_index *auxtrace_index, *n;
824 list_for_each_entry_safe(auxtrace_index, n, head, list) {
825 list_del_init(&auxtrace_index->list);
826 free(auxtrace_index);
830 static struct auxtrace_index *auxtrace_index__last(struct list_head *head)
832 struct auxtrace_index *auxtrace_index;
835 if (list_empty(head)) {
836 err = auxtrace_index__alloc(head);
841 auxtrace_index = list_entry(head->prev, struct auxtrace_index, list);
843 if (auxtrace_index->nr >= PERF_AUXTRACE_INDEX_ENTRY_COUNT) {
844 err = auxtrace_index__alloc(head);
847 auxtrace_index = list_entry(head->prev, struct auxtrace_index,
851 return auxtrace_index;
854 int auxtrace_index__auxtrace_event(struct list_head *head,
855 union perf_event *event, off_t file_offset)
857 struct auxtrace_index *auxtrace_index;
860 auxtrace_index = auxtrace_index__last(head);
864 nr = auxtrace_index->nr;
865 auxtrace_index->entries[nr].file_offset = file_offset;
866 auxtrace_index->entries[nr].sz = event->header.size;
867 auxtrace_index->nr += 1;
872 static int auxtrace_index__do_write(int fd,
873 struct auxtrace_index *auxtrace_index)
875 struct auxtrace_index_entry ent;
878 for (i = 0; i < auxtrace_index->nr; i++) {
879 ent.file_offset = auxtrace_index->entries[i].file_offset;
880 ent.sz = auxtrace_index->entries[i].sz;
881 if (writen(fd, &ent, sizeof(ent)) != sizeof(ent))
887 int auxtrace_index__write(int fd, struct list_head *head)
889 struct auxtrace_index *auxtrace_index;
893 list_for_each_entry(auxtrace_index, head, list)
894 total += auxtrace_index->nr;
896 if (writen(fd, &total, sizeof(total)) != sizeof(total))
899 list_for_each_entry(auxtrace_index, head, list) {
900 err = auxtrace_index__do_write(fd, auxtrace_index);
908 static int auxtrace_index__process_entry(int fd, struct list_head *head,
911 struct auxtrace_index *auxtrace_index;
912 struct auxtrace_index_entry ent;
915 if (readn(fd, &ent, sizeof(ent)) != sizeof(ent))
918 auxtrace_index = auxtrace_index__last(head);
922 nr = auxtrace_index->nr;
924 auxtrace_index->entries[nr].file_offset =
925 bswap_64(ent.file_offset);
926 auxtrace_index->entries[nr].sz = bswap_64(ent.sz);
928 auxtrace_index->entries[nr].file_offset = ent.file_offset;
929 auxtrace_index->entries[nr].sz = ent.sz;
932 auxtrace_index->nr = nr + 1;
937 int auxtrace_index__process(int fd, u64 size, struct perf_session *session,
940 struct list_head *head = &session->auxtrace_index;
943 if (readn(fd, &nr, sizeof(u64)) != sizeof(u64))
949 if (sizeof(u64) + nr * sizeof(struct auxtrace_index_entry) > size)
955 err = auxtrace_index__process_entry(fd, head, needs_swap);
963 static int auxtrace_queues__process_index_entry(struct auxtrace_queues *queues,
964 struct perf_session *session,
965 struct auxtrace_index_entry *ent)
967 return auxtrace_queues__add_indexed_event(queues, session,
968 ent->file_offset, ent->sz);
971 int auxtrace_queues__process_index(struct auxtrace_queues *queues,
972 struct perf_session *session)
974 struct auxtrace_index *auxtrace_index;
975 struct auxtrace_index_entry *ent;
979 if (auxtrace__dont_decode(session))
982 list_for_each_entry(auxtrace_index, &session->auxtrace_index, list) {
983 for (i = 0; i < auxtrace_index->nr; i++) {
984 ent = &auxtrace_index->entries[i];
985 err = auxtrace_queues__process_index_entry(queues,
995 struct auxtrace_buffer *auxtrace_buffer__next(struct auxtrace_queue *queue,
996 struct auxtrace_buffer *buffer)
999 if (list_is_last(&buffer->list, &queue->head))
1001 return list_entry(buffer->list.next, struct auxtrace_buffer,
1004 if (list_empty(&queue->head))
1006 return list_entry(queue->head.next, struct auxtrace_buffer,
1011 struct auxtrace_queue *auxtrace_queues__sample_queue(struct auxtrace_queues *queues,
1012 struct perf_sample *sample,
1013 struct perf_session *session)
1015 struct perf_sample_id *sid;
1023 sid = evlist__id2sid(session->evlist, id);
1029 if (idx >= queues->nr_queues)
1032 return &queues->queue_array[idx];
1035 int auxtrace_queues__add_sample(struct auxtrace_queues *queues,
1036 struct perf_session *session,
1037 struct perf_sample *sample, u64 data_offset,
1040 struct auxtrace_buffer buffer = {
1042 .data_offset = data_offset,
1043 .reference = reference,
1044 .size = sample->aux_sample.size,
1046 struct perf_sample_id *sid;
1047 u64 id = sample->id;
1053 sid = evlist__id2sid(session->evlist, id);
1058 buffer.tid = sid->tid;
1059 buffer.cpu = sid->cpu;
1061 return auxtrace_queues__add_buffer(queues, session, idx, &buffer, NULL);
1069 static int auxtrace_queue_data_cb(struct perf_session *session,
1070 union perf_event *event, u64 offset,
1073 struct queue_data *qd = data;
1074 struct perf_sample sample;
1077 if (qd->events && event->header.type == PERF_RECORD_AUXTRACE) {
1078 if (event->header.size < sizeof(struct perf_record_auxtrace))
1080 offset += event->header.size;
1081 return session->auxtrace->queue_data(session, NULL, event,
1085 if (!qd->samples || event->header.type != PERF_RECORD_SAMPLE)
1088 err = evlist__parse_sample(session->evlist, event, &sample);
1092 if (!sample.aux_sample.size)
1095 offset += sample.aux_sample.data - (void *)event;
1097 return session->auxtrace->queue_data(session, &sample, NULL, offset);
1100 int auxtrace_queue_data(struct perf_session *session, bool samples, bool events)
1102 struct queue_data qd = {
1107 if (auxtrace__dont_decode(session))
1110 if (!session->auxtrace || !session->auxtrace->queue_data)
1113 return perf_session__peek_events(session, session->header.data_offset,
1114 session->header.data_size,
1115 auxtrace_queue_data_cb, &qd);
1118 void *auxtrace_buffer__get_data_rw(struct auxtrace_buffer *buffer, int fd, bool rw)
1120 int prot = rw ? PROT_READ | PROT_WRITE : PROT_READ;
1121 size_t adj = buffer->data_offset & (page_size - 1);
1122 size_t size = buffer->size + adj;
1123 off_t file_offset = buffer->data_offset - adj;
1127 return buffer->data;
1129 addr = mmap(NULL, size, prot, MAP_SHARED, fd, file_offset);
1130 if (addr == MAP_FAILED)
1133 buffer->mmap_addr = addr;
1134 buffer->mmap_size = size;
1136 buffer->data = addr + adj;
1138 return buffer->data;
1141 void auxtrace_buffer__put_data(struct auxtrace_buffer *buffer)
1143 if (!buffer->data || !buffer->mmap_addr)
1145 munmap(buffer->mmap_addr, buffer->mmap_size);
1146 buffer->mmap_addr = NULL;
1147 buffer->mmap_size = 0;
1148 buffer->data = NULL;
1149 buffer->use_data = NULL;
1152 void auxtrace_buffer__drop_data(struct auxtrace_buffer *buffer)
1154 auxtrace_buffer__put_data(buffer);
1155 if (buffer->data_needs_freeing) {
1156 buffer->data_needs_freeing = false;
1157 zfree(&buffer->data);
1158 buffer->use_data = NULL;
1163 void auxtrace_buffer__free(struct auxtrace_buffer *buffer)
1165 auxtrace_buffer__drop_data(buffer);
1169 void auxtrace_synth_error(struct perf_record_auxtrace_error *auxtrace_error, int type,
1170 int code, int cpu, pid_t pid, pid_t tid, u64 ip,
1171 const char *msg, u64 timestamp)
1175 memset(auxtrace_error, 0, sizeof(struct perf_record_auxtrace_error));
1177 auxtrace_error->header.type = PERF_RECORD_AUXTRACE_ERROR;
1178 auxtrace_error->type = type;
1179 auxtrace_error->code = code;
1180 auxtrace_error->cpu = cpu;
1181 auxtrace_error->pid = pid;
1182 auxtrace_error->tid = tid;
1183 auxtrace_error->fmt = 1;
1184 auxtrace_error->ip = ip;
1185 auxtrace_error->time = timestamp;
1186 strlcpy(auxtrace_error->msg, msg, MAX_AUXTRACE_ERROR_MSG);
1188 size = (void *)auxtrace_error->msg - (void *)auxtrace_error +
1189 strlen(auxtrace_error->msg) + 1;
1190 auxtrace_error->header.size = PERF_ALIGN(size, sizeof(u64));
1193 int perf_event__synthesize_auxtrace_info(struct auxtrace_record *itr,
1194 struct perf_tool *tool,
1195 struct perf_session *session,
1196 perf_event__handler_t process)
1198 union perf_event *ev;
1202 pr_debug2("Synthesizing auxtrace information\n");
1203 priv_size = auxtrace_record__info_priv_size(itr, session->evlist);
1204 ev = zalloc(sizeof(struct perf_record_auxtrace_info) + priv_size);
1208 ev->auxtrace_info.header.type = PERF_RECORD_AUXTRACE_INFO;
1209 ev->auxtrace_info.header.size = sizeof(struct perf_record_auxtrace_info) +
1211 err = auxtrace_record__info_fill(itr, session, &ev->auxtrace_info,
1216 err = process(tool, ev, NULL, NULL);
1222 static void unleader_evsel(struct evlist *evlist, struct evsel *leader)
1224 struct evsel *new_leader = NULL;
1225 struct evsel *evsel;
1227 /* Find new leader for the group */
1228 evlist__for_each_entry(evlist, evsel) {
1229 if (!evsel__has_leader(evsel, leader) || evsel == leader)
1233 evsel__set_leader(evsel, new_leader);
1236 /* Update group information */
1238 zfree(&new_leader->group_name);
1239 new_leader->group_name = leader->group_name;
1240 leader->group_name = NULL;
1242 new_leader->core.nr_members = leader->core.nr_members - 1;
1243 leader->core.nr_members = 1;
1247 static void unleader_auxtrace(struct perf_session *session)
1249 struct evsel *evsel;
1251 evlist__for_each_entry(session->evlist, evsel) {
1252 if (auxtrace__evsel_is_auxtrace(session, evsel) &&
1253 evsel__is_group_leader(evsel)) {
1254 unleader_evsel(session->evlist, evsel);
1259 int perf_event__process_auxtrace_info(struct perf_session *session,
1260 union perf_event *event)
1262 enum auxtrace_type type = event->auxtrace_info.type;
1266 fprintf(stdout, " type: %u\n", type);
1269 case PERF_AUXTRACE_INTEL_PT:
1270 err = intel_pt_process_auxtrace_info(event, session);
1272 case PERF_AUXTRACE_INTEL_BTS:
1273 err = intel_bts_process_auxtrace_info(event, session);
1275 case PERF_AUXTRACE_ARM_SPE:
1276 err = arm_spe_process_auxtrace_info(event, session);
1278 case PERF_AUXTRACE_CS_ETM:
1279 err = cs_etm__process_auxtrace_info(event, session);
1281 case PERF_AUXTRACE_S390_CPUMSF:
1282 err = s390_cpumsf_process_auxtrace_info(event, session);
1284 case PERF_AUXTRACE_UNKNOWN:
1292 unleader_auxtrace(session);
1297 s64 perf_event__process_auxtrace(struct perf_session *session,
1298 union perf_event *event)
1303 fprintf(stdout, " size: %#"PRI_lx64" offset: %#"PRI_lx64" ref: %#"PRI_lx64" idx: %u tid: %d cpu: %d\n",
1304 event->auxtrace.size, event->auxtrace.offset,
1305 event->auxtrace.reference, event->auxtrace.idx,
1306 event->auxtrace.tid, event->auxtrace.cpu);
1308 if (auxtrace__dont_decode(session))
1309 return event->auxtrace.size;
1311 if (!session->auxtrace || event->header.type != PERF_RECORD_AUXTRACE)
1314 err = session->auxtrace->process_auxtrace_event(session, event, session->tool);
1318 return event->auxtrace.size;
1321 #define PERF_ITRACE_DEFAULT_PERIOD_TYPE PERF_ITRACE_PERIOD_NANOSECS
1322 #define PERF_ITRACE_DEFAULT_PERIOD 100000
1323 #define PERF_ITRACE_DEFAULT_CALLCHAIN_SZ 16
1324 #define PERF_ITRACE_MAX_CALLCHAIN_SZ 1024
1325 #define PERF_ITRACE_DEFAULT_LAST_BRANCH_SZ 64
1326 #define PERF_ITRACE_MAX_LAST_BRANCH_SZ 1024
1328 void itrace_synth_opts__set_default(struct itrace_synth_opts *synth_opts,
1331 synth_opts->branches = true;
1332 synth_opts->transactions = true;
1333 synth_opts->ptwrites = true;
1334 synth_opts->pwr_events = true;
1335 synth_opts->other_events = true;
1336 synth_opts->errors = true;
1337 synth_opts->flc = true;
1338 synth_opts->llc = true;
1339 synth_opts->tlb = true;
1340 synth_opts->mem = true;
1341 synth_opts->remote_access = true;
1344 synth_opts->period_type = PERF_ITRACE_PERIOD_INSTRUCTIONS;
1345 synth_opts->period = 1;
1346 synth_opts->calls = true;
1348 synth_opts->instructions = true;
1349 synth_opts->period_type = PERF_ITRACE_DEFAULT_PERIOD_TYPE;
1350 synth_opts->period = PERF_ITRACE_DEFAULT_PERIOD;
1352 synth_opts->callchain_sz = PERF_ITRACE_DEFAULT_CALLCHAIN_SZ;
1353 synth_opts->last_branch_sz = PERF_ITRACE_DEFAULT_LAST_BRANCH_SZ;
1354 synth_opts->initial_skip = 0;
1357 static int get_flag(const char **ptr, unsigned int *flags)
1362 if (c >= 'a' && c <= 'z') {
1363 *flags |= 1 << (c - 'a');
1366 } else if (c == ' ') {
1375 static int get_flags(const char **ptr, unsigned int *plus_flags, unsigned int *minus_flags)
1381 if (get_flag(ptr, plus_flags))
1386 if (get_flag(ptr, minus_flags))
1399 * Please check tools/perf/Documentation/perf-script.txt for information
1400 * about the options parsed here, which is introduced after this cset,
1401 * when support in 'perf script' for these options is introduced.
1403 int itrace_do_parse_synth_opts(struct itrace_synth_opts *synth_opts,
1404 const char *str, int unset)
1408 bool period_type_set = false;
1409 bool period_set = false;
1411 synth_opts->set = true;
1414 synth_opts->dont_decode = true;
1419 itrace_synth_opts__set_default(synth_opts,
1420 synth_opts->default_no_sample);
1424 for (p = str; *p;) {
1427 synth_opts->instructions = true;
1428 while (*p == ' ' || *p == ',')
1431 synth_opts->period = strtoull(p, &endptr, 10);
1434 while (*p == ' ' || *p == ',')
1438 synth_opts->period_type =
1439 PERF_ITRACE_PERIOD_INSTRUCTIONS;
1440 period_type_set = true;
1443 synth_opts->period_type =
1444 PERF_ITRACE_PERIOD_TICKS;
1445 period_type_set = true;
1448 synth_opts->period *= 1000;
1451 synth_opts->period *= 1000;
1456 synth_opts->period_type =
1457 PERF_ITRACE_PERIOD_NANOSECS;
1458 period_type_set = true;
1468 synth_opts->branches = true;
1471 synth_opts->transactions = true;
1474 synth_opts->ptwrites = true;
1477 synth_opts->pwr_events = true;
1480 synth_opts->other_events = true;
1483 synth_opts->errors = true;
1484 if (get_flags(&p, &synth_opts->error_plus_flags,
1485 &synth_opts->error_minus_flags))
1489 synth_opts->log = true;
1490 if (get_flags(&p, &synth_opts->log_plus_flags,
1491 &synth_opts->log_minus_flags))
1495 synth_opts->branches = true;
1496 synth_opts->calls = true;
1499 synth_opts->branches = true;
1500 synth_opts->returns = true;
1505 synth_opts->add_callchain = true;
1507 synth_opts->callchain = true;
1508 synth_opts->callchain_sz =
1509 PERF_ITRACE_DEFAULT_CALLCHAIN_SZ;
1510 while (*p == ' ' || *p == ',')
1515 val = strtoul(p, &endptr, 10);
1517 if (!val || val > PERF_ITRACE_MAX_CALLCHAIN_SZ)
1519 synth_opts->callchain_sz = val;
1525 synth_opts->add_last_branch = true;
1527 synth_opts->last_branch = true;
1528 synth_opts->last_branch_sz =
1529 PERF_ITRACE_DEFAULT_LAST_BRANCH_SZ;
1530 while (*p == ' ' || *p == ',')
1535 val = strtoul(p, &endptr, 10);
1538 val > PERF_ITRACE_MAX_LAST_BRANCH_SZ)
1540 synth_opts->last_branch_sz = val;
1544 synth_opts->initial_skip = strtoul(p, &endptr, 10);
1550 synth_opts->flc = true;
1553 synth_opts->llc = true;
1556 synth_opts->tlb = true;
1559 synth_opts->remote_access = true;
1562 synth_opts->mem = true;
1565 synth_opts->quick += 1;
1568 synth_opts->approx_ipc = true;
1571 synth_opts->timeless_decoding = true;
1581 if (synth_opts->instructions) {
1582 if (!period_type_set)
1583 synth_opts->period_type =
1584 PERF_ITRACE_DEFAULT_PERIOD_TYPE;
1586 synth_opts->period = PERF_ITRACE_DEFAULT_PERIOD;
1592 pr_err("Bad Instruction Tracing options '%s'\n", str);
1596 int itrace_parse_synth_opts(const struct option *opt, const char *str, int unset)
1598 return itrace_do_parse_synth_opts(opt->value, str, unset);
1601 static const char * const auxtrace_error_type_name[] = {
1602 [PERF_AUXTRACE_ERROR_ITRACE] = "instruction trace",
1605 static const char *auxtrace_error_name(int type)
1607 const char *error_type_name = NULL;
1609 if (type < PERF_AUXTRACE_ERROR_MAX)
1610 error_type_name = auxtrace_error_type_name[type];
1611 if (!error_type_name)
1612 error_type_name = "unknown AUX";
1613 return error_type_name;
1616 size_t perf_event__fprintf_auxtrace_error(union perf_event *event, FILE *fp)
1618 struct perf_record_auxtrace_error *e = &event->auxtrace_error;
1619 unsigned long long nsecs = e->time;
1620 const char *msg = e->msg;
1623 ret = fprintf(fp, " %s error type %u",
1624 auxtrace_error_name(e->type), e->type);
1626 if (e->fmt && nsecs) {
1627 unsigned long secs = nsecs / NSEC_PER_SEC;
1629 nsecs -= secs * NSEC_PER_SEC;
1630 ret += fprintf(fp, " time %lu.%09llu", secs, nsecs);
1632 ret += fprintf(fp, " time 0");
1636 msg = (const char *)&e->time;
1638 ret += fprintf(fp, " cpu %d pid %d tid %d ip %#"PRI_lx64" code %u: %s\n",
1639 e->cpu, e->pid, e->tid, e->ip, e->code, msg);
1643 void perf_session__auxtrace_error_inc(struct perf_session *session,
1644 union perf_event *event)
1646 struct perf_record_auxtrace_error *e = &event->auxtrace_error;
1648 if (e->type < PERF_AUXTRACE_ERROR_MAX)
1649 session->evlist->stats.nr_auxtrace_errors[e->type] += 1;
1652 void events_stats__auxtrace_error_warn(const struct events_stats *stats)
1656 for (i = 0; i < PERF_AUXTRACE_ERROR_MAX; i++) {
1657 if (!stats->nr_auxtrace_errors[i])
1659 ui__warning("%u %s errors\n",
1660 stats->nr_auxtrace_errors[i],
1661 auxtrace_error_name(i));
1665 int perf_event__process_auxtrace_error(struct perf_session *session,
1666 union perf_event *event)
1668 if (auxtrace__dont_decode(session))
1671 perf_event__fprintf_auxtrace_error(event, stdout);
1676 * In the compat mode kernel runs in 64-bit and perf tool runs in 32-bit mode,
1677 * 32-bit perf tool cannot access 64-bit value atomically, which might lead to
1678 * the issues caused by the below sequence on multiple CPUs: when perf tool
1679 * accesses either the load operation or the store operation for 64-bit value,
1680 * on some architectures the operation is divided into two instructions, one
1681 * is for accessing the low 32-bit value and another is for the high 32-bit;
1682 * thus these two user operations can give the kernel chances to access the
1683 * 64-bit value, and thus leads to the unexpected load values.
1685 * kernel (64-bit) user (32-bit)
1687 * if (LOAD ->aux_tail) { --, LOAD ->aux_head_lo
1688 * STORE $aux_data | ,--->
1689 * FLUSH $aux_data | | LOAD ->aux_head_hi
1690 * STORE ->aux_head --|-------` smp_rmb()
1693 * | STORE ->aux_tail_lo
1695 * STORE ->aux_tail_hi
1697 * For this reason, it's impossible for the perf tool to work correctly when
1698 * the AUX head or tail is bigger than 4GB (more than 32 bits length); and we
1699 * can not simply limit the AUX ring buffer to less than 4GB, the reason is
1700 * the pointers can be increased monotonically, whatever the buffer size it is,
1701 * at the end the head and tail can be bigger than 4GB and carry out to the
1704 * To mitigate the issues and improve the user experience, we can allow the
1705 * perf tool working in certain conditions and bail out with error if detect
1706 * any overflow cannot be handled.
1708 * For reading the AUX head, it reads out the values for three times, and
1709 * compares the high 4 bytes of the values between the first time and the last
1710 * time, if there has no change for high 4 bytes injected by the kernel during
1711 * the user reading sequence, it's safe for use the second value.
1713 * When compat_auxtrace_mmap__write_tail() detects any carrying in the high
1714 * 32 bits, it means there have two store operations in user space and it cannot
1715 * promise the atomicity for 64-bit write, so return '-1' in this case to tell
1716 * the caller an overflow error has happened.
1718 u64 __weak compat_auxtrace_mmap__read_head(struct auxtrace_mmap *mm)
1720 struct perf_event_mmap_page *pc = mm->userpg;
1721 u64 first, second, last;
1722 u64 mask = (u64)(UINT32_MAX) << 32;
1725 first = READ_ONCE(pc->aux_head);
1726 /* Ensure all reads are done after we read the head */
1728 second = READ_ONCE(pc->aux_head);
1729 /* Ensure all reads are done after we read the head */
1731 last = READ_ONCE(pc->aux_head);
1732 } while ((first & mask) != (last & mask));
1737 int __weak compat_auxtrace_mmap__write_tail(struct auxtrace_mmap *mm, u64 tail)
1739 struct perf_event_mmap_page *pc = mm->userpg;
1740 u64 mask = (u64)(UINT32_MAX) << 32;
1745 /* Ensure all reads are done before we write the tail out */
1747 WRITE_ONCE(pc->aux_tail, tail);
1751 static int __auxtrace_mmap__read(struct mmap *map,
1752 struct auxtrace_record *itr,
1753 struct perf_tool *tool, process_auxtrace_t fn,
1754 bool snapshot, size_t snapshot_size)
1756 struct auxtrace_mmap *mm = &map->auxtrace_mmap;
1757 u64 head, old = mm->prev, offset, ref;
1758 unsigned char *data = mm->base;
1759 size_t size, head_off, old_off, len1, len2, padding;
1760 union perf_event ev;
1761 void *data1, *data2;
1762 int kernel_is_64_bit = perf_env__kernel_is_64_bit(evsel__env(NULL));
1764 head = auxtrace_mmap__read_head(mm, kernel_is_64_bit);
1767 auxtrace_record__find_snapshot(itr, mm->idx, mm, data, &head, &old))
1773 pr_debug3("auxtrace idx %d old %#"PRIx64" head %#"PRIx64" diff %#"PRIx64"\n",
1774 mm->idx, old, head, head - old);
1777 head_off = head & mm->mask;
1778 old_off = old & mm->mask;
1780 head_off = head % mm->len;
1781 old_off = old % mm->len;
1784 if (head_off > old_off)
1785 size = head_off - old_off;
1787 size = mm->len - (old_off - head_off);
1789 if (snapshot && size > snapshot_size)
1790 size = snapshot_size;
1792 ref = auxtrace_record__reference(itr);
1794 if (head > old || size <= head || mm->mask) {
1795 offset = head - size;
1798 * When the buffer size is not a power of 2, 'head' wraps at the
1799 * highest multiple of the buffer size, so we have to subtract
1800 * the remainder here.
1802 u64 rem = (0ULL - mm->len) % mm->len;
1804 offset = head - size - rem;
1807 if (size > head_off) {
1808 len1 = size - head_off;
1809 data1 = &data[mm->len - len1];
1814 data1 = &data[head_off - len1];
1819 if (itr->alignment) {
1820 unsigned int unwanted = len1 % itr->alignment;
1826 /* padding must be written by fn() e.g. record__process_auxtrace() */
1827 padding = size & (PERF_AUXTRACE_RECORD_ALIGNMENT - 1);
1829 padding = PERF_AUXTRACE_RECORD_ALIGNMENT - padding;
1831 memset(&ev, 0, sizeof(ev));
1832 ev.auxtrace.header.type = PERF_RECORD_AUXTRACE;
1833 ev.auxtrace.header.size = sizeof(ev.auxtrace);
1834 ev.auxtrace.size = size + padding;
1835 ev.auxtrace.offset = offset;
1836 ev.auxtrace.reference = ref;
1837 ev.auxtrace.idx = mm->idx;
1838 ev.auxtrace.tid = mm->tid;
1839 ev.auxtrace.cpu = mm->cpu;
1841 if (fn(tool, map, &ev, data1, len1, data2, len2))
1849 err = auxtrace_mmap__write_tail(mm, head, kernel_is_64_bit);
1853 if (itr->read_finish) {
1854 err = itr->read_finish(itr, mm->idx);
1863 int auxtrace_mmap__read(struct mmap *map, struct auxtrace_record *itr,
1864 struct perf_tool *tool, process_auxtrace_t fn)
1866 return __auxtrace_mmap__read(map, itr, tool, fn, false, 0);
1869 int auxtrace_mmap__read_snapshot(struct mmap *map,
1870 struct auxtrace_record *itr,
1871 struct perf_tool *tool, process_auxtrace_t fn,
1872 size_t snapshot_size)
1874 return __auxtrace_mmap__read(map, itr, tool, fn, true, snapshot_size);
1878 * struct auxtrace_cache - hash table to implement a cache
1879 * @hashtable: the hashtable
1880 * @sz: hashtable size (number of hlists)
1881 * @entry_size: size of an entry
1882 * @limit: limit the number of entries to this maximum, when reached the cache
1883 * is dropped and caching begins again with an empty cache
1884 * @cnt: current number of entries
1885 * @bits: hashtable size (@sz = 2^@bits)
1887 struct auxtrace_cache {
1888 struct hlist_head *hashtable;
1896 struct auxtrace_cache *auxtrace_cache__new(unsigned int bits, size_t entry_size,
1897 unsigned int limit_percent)
1899 struct auxtrace_cache *c;
1900 struct hlist_head *ht;
1903 c = zalloc(sizeof(struct auxtrace_cache));
1909 ht = calloc(sz, sizeof(struct hlist_head));
1913 for (i = 0; i < sz; i++)
1914 INIT_HLIST_HEAD(&ht[i]);
1918 c->entry_size = entry_size;
1919 c->limit = (c->sz * limit_percent) / 100;
1929 static void auxtrace_cache__drop(struct auxtrace_cache *c)
1931 struct auxtrace_cache_entry *entry;
1932 struct hlist_node *tmp;
1938 for (i = 0; i < c->sz; i++) {
1939 hlist_for_each_entry_safe(entry, tmp, &c->hashtable[i], hash) {
1940 hlist_del(&entry->hash);
1941 auxtrace_cache__free_entry(c, entry);
1948 void auxtrace_cache__free(struct auxtrace_cache *c)
1953 auxtrace_cache__drop(c);
1954 zfree(&c->hashtable);
1958 void *auxtrace_cache__alloc_entry(struct auxtrace_cache *c)
1960 return malloc(c->entry_size);
1963 void auxtrace_cache__free_entry(struct auxtrace_cache *c __maybe_unused,
1969 int auxtrace_cache__add(struct auxtrace_cache *c, u32 key,
1970 struct auxtrace_cache_entry *entry)
1972 if (c->limit && ++c->cnt > c->limit)
1973 auxtrace_cache__drop(c);
1976 hlist_add_head(&entry->hash, &c->hashtable[hash_32(key, c->bits)]);
1981 static struct auxtrace_cache_entry *auxtrace_cache__rm(struct auxtrace_cache *c,
1984 struct auxtrace_cache_entry *entry;
1985 struct hlist_head *hlist;
1986 struct hlist_node *n;
1991 hlist = &c->hashtable[hash_32(key, c->bits)];
1992 hlist_for_each_entry_safe(entry, n, hlist, hash) {
1993 if (entry->key == key) {
1994 hlist_del(&entry->hash);
2002 void auxtrace_cache__remove(struct auxtrace_cache *c, u32 key)
2004 struct auxtrace_cache_entry *entry = auxtrace_cache__rm(c, key);
2006 auxtrace_cache__free_entry(c, entry);
2009 void *auxtrace_cache__lookup(struct auxtrace_cache *c, u32 key)
2011 struct auxtrace_cache_entry *entry;
2012 struct hlist_head *hlist;
2017 hlist = &c->hashtable[hash_32(key, c->bits)];
2018 hlist_for_each_entry(entry, hlist, hash) {
2019 if (entry->key == key)
2026 static void addr_filter__free_str(struct addr_filter *filt)
2029 filt->action = NULL;
2030 filt->sym_from = NULL;
2031 filt->sym_to = NULL;
2032 filt->filename = NULL;
2035 static struct addr_filter *addr_filter__new(void)
2037 struct addr_filter *filt = zalloc(sizeof(*filt));
2040 INIT_LIST_HEAD(&filt->list);
2045 static void addr_filter__free(struct addr_filter *filt)
2048 addr_filter__free_str(filt);
2052 static void addr_filters__add(struct addr_filters *filts,
2053 struct addr_filter *filt)
2055 list_add_tail(&filt->list, &filts->head);
2059 static void addr_filters__del(struct addr_filters *filts,
2060 struct addr_filter *filt)
2062 list_del_init(&filt->list);
2066 void addr_filters__init(struct addr_filters *filts)
2068 INIT_LIST_HEAD(&filts->head);
2072 void addr_filters__exit(struct addr_filters *filts)
2074 struct addr_filter *filt, *n;
2076 list_for_each_entry_safe(filt, n, &filts->head, list) {
2077 addr_filters__del(filts, filt);
2078 addr_filter__free(filt);
2082 static int parse_num_or_str(char **inp, u64 *num, const char **str,
2083 const char *str_delim)
2085 *inp += strspn(*inp, " ");
2087 if (isdigit(**inp)) {
2093 *num = strtoull(*inp, &endptr, 0);
2104 *inp += strspn(*inp, " ");
2106 n = strcspn(*inp, str_delim);
2118 static int parse_action(struct addr_filter *filt)
2120 if (!strcmp(filt->action, "filter")) {
2123 } else if (!strcmp(filt->action, "start")) {
2125 } else if (!strcmp(filt->action, "stop")) {
2126 filt->start = false;
2127 } else if (!strcmp(filt->action, "tracestop")) {
2128 filt->start = false;
2130 filt->action += 5; /* Change 'tracestop' to 'stop' */
2137 static int parse_sym_idx(char **inp, int *idx)
2141 *inp += strspn(*inp, " ");
2148 if (**inp == 'g' || **inp == 'G') {
2156 num = strtoul(*inp, &endptr, 0);
2159 if (endptr == *inp || num > INT_MAX)
2168 static int parse_addr_size(char **inp, u64 *num, const char **str, int *idx)
2170 int err = parse_num_or_str(inp, num, str, " ");
2173 err = parse_sym_idx(inp, idx);
2178 static int parse_one_filter(struct addr_filter *filt, const char **filter_inp)
2183 filt->str = fstr = strdup(*filter_inp);
2187 err = parse_num_or_str(&fstr, NULL, &filt->action, " ");
2191 err = parse_action(filt);
2195 err = parse_addr_size(&fstr, &filt->addr, &filt->sym_from,
2196 &filt->sym_from_idx);
2200 fstr += strspn(fstr, " ");
2204 err = parse_addr_size(&fstr, &filt->size, &filt->sym_to,
2211 fstr += strspn(fstr, " ");
2215 err = parse_num_or_str(&fstr, NULL, &filt->filename, " ,");
2220 fstr += strspn(fstr, " ,");
2222 *filter_inp += fstr - filt->str;
2227 addr_filter__free_str(filt);
2232 int addr_filters__parse_bare_filter(struct addr_filters *filts,
2235 struct addr_filter *filt;
2236 const char *fstr = filter;
2240 filt = addr_filter__new();
2241 err = parse_one_filter(filt, &fstr);
2243 addr_filter__free(filt);
2244 addr_filters__exit(filts);
2247 addr_filters__add(filts, filt);
2266 static bool kern_sym_match(struct sym_args *args, const char *name, char type)
2268 /* A function with the same name, and global or the n'th found or any */
2269 return kallsyms__is_function(type) &&
2270 !strcmp(name, args->name) &&
2271 ((args->global && isupper(type)) ||
2272 (args->selected && ++(args->cnt) == args->idx) ||
2273 (!args->global && !args->selected));
2276 static int find_kern_sym_cb(void *arg, const char *name, char type, u64 start)
2278 struct sym_args *args = arg;
2280 if (args->started) {
2282 args->size = start - args->start;
2283 if (args->selected) {
2286 } else if (kern_sym_match(args, name, type)) {
2287 args->duplicate = true;
2290 } else if (kern_sym_match(args, name, type)) {
2291 args->started = true;
2292 args->start = start;
2298 static int print_kern_sym_cb(void *arg, const char *name, char type, u64 start)
2300 struct sym_args *args = arg;
2302 if (kern_sym_match(args, name, type)) {
2303 pr_err("#%d\t0x%"PRIx64"\t%c\t%s\n",
2304 ++args->cnt, start, type, name);
2306 } else if (args->near) {
2308 pr_err("\t\twhich is near\t\t%s\n", name);
2314 static int sym_not_found_error(const char *sym_name, int idx)
2317 pr_err("N'th occurrence (N=%d) of symbol '%s' not found.\n",
2320 pr_err("Global symbol '%s' not found.\n", sym_name);
2322 pr_err("Symbol '%s' not found.\n", sym_name);
2324 pr_err("Note that symbols must be functions.\n");
2329 static int find_kern_sym(const char *sym_name, u64 *start, u64 *size, int idx)
2331 struct sym_args args = {
2335 .selected = idx > 0,
2342 err = kallsyms__parse("/proc/kallsyms", &args, find_kern_sym_cb);
2344 pr_err("Failed to parse /proc/kallsyms\n");
2348 if (args.duplicate) {
2349 pr_err("Multiple kernel symbols with name '%s'\n", sym_name);
2351 kallsyms__parse("/proc/kallsyms", &args, print_kern_sym_cb);
2352 pr_err("Disambiguate symbol name by inserting #n after the name e.g. %s #2\n",
2354 pr_err("Or select a global symbol by inserting #0 or #g or #G\n");
2358 if (!args.started) {
2359 pr_err("Kernel symbol lookup: ");
2360 return sym_not_found_error(sym_name, idx);
2363 *start = args.start;
2369 static int find_entire_kern_cb(void *arg, const char *name __maybe_unused,
2370 char type, u64 start)
2372 struct sym_args *args = arg;
2374 if (!kallsyms__is_function(type))
2377 if (!args->started) {
2378 args->started = true;
2379 args->start = start;
2381 /* Don't know exactly where the kernel ends, so we add a page */
2382 args->size = round_up(start, page_size) + page_size - args->start;
2387 static int addr_filter__entire_kernel(struct addr_filter *filt)
2389 struct sym_args args = { .started = false };
2392 err = kallsyms__parse("/proc/kallsyms", &args, find_entire_kern_cb);
2393 if (err < 0 || !args.started) {
2394 pr_err("Failed to parse /proc/kallsyms\n");
2398 filt->addr = args.start;
2399 filt->size = args.size;
2404 static int check_end_after_start(struct addr_filter *filt, u64 start, u64 size)
2406 if (start + size >= filt->addr)
2409 if (filt->sym_from) {
2410 pr_err("Symbol '%s' (0x%"PRIx64") comes before '%s' (0x%"PRIx64")\n",
2411 filt->sym_to, start, filt->sym_from, filt->addr);
2413 pr_err("Symbol '%s' (0x%"PRIx64") comes before address 0x%"PRIx64")\n",
2414 filt->sym_to, start, filt->addr);
2420 static int addr_filter__resolve_kernel_syms(struct addr_filter *filt)
2422 bool no_size = false;
2426 if (symbol_conf.kptr_restrict) {
2427 pr_err("Kernel addresses are restricted. Unable to resolve kernel symbols.\n");
2431 if (filt->sym_from && !strcmp(filt->sym_from, "*"))
2432 return addr_filter__entire_kernel(filt);
2434 if (filt->sym_from) {
2435 err = find_kern_sym(filt->sym_from, &start, &size,
2436 filt->sym_from_idx);
2440 if (filt->range && !filt->size && !filt->sym_to) {
2447 err = find_kern_sym(filt->sym_to, &start, &size,
2452 err = check_end_after_start(filt, start, size);
2455 filt->size = start + size - filt->addr;
2459 /* The very last symbol in kallsyms does not imply a particular size */
2461 pr_err("Cannot determine size of symbol '%s'\n",
2462 filt->sym_to ? filt->sym_to : filt->sym_from);
2469 static struct dso *load_dso(const char *name)
2474 map = dso__new_map(name);
2478 if (map__load(map) < 0)
2479 pr_err("File '%s' not found or has no symbols.\n", name);
2481 dso = dso__get(map->dso);
2488 static bool dso_sym_match(struct symbol *sym, const char *name, int *cnt,
2491 /* Same name, and global or the n'th found or any */
2492 return !arch__compare_symbol_names(name, sym->name) &&
2493 ((!idx && sym->binding == STB_GLOBAL) ||
2494 (idx > 0 && ++*cnt == idx) ||
2498 static void print_duplicate_syms(struct dso *dso, const char *sym_name)
2504 pr_err("Multiple symbols with name '%s'\n", sym_name);
2506 sym = dso__first_symbol(dso);
2508 if (dso_sym_match(sym, sym_name, &cnt, -1)) {
2509 pr_err("#%d\t0x%"PRIx64"\t%c\t%s\n",
2511 sym->binding == STB_GLOBAL ? 'g' :
2512 sym->binding == STB_LOCAL ? 'l' : 'w',
2517 pr_err("\t\twhich is near\t\t%s\n", sym->name);
2519 sym = dso__next_symbol(sym);
2522 pr_err("Disambiguate symbol name by inserting #n after the name e.g. %s #2\n",
2524 pr_err("Or select a global symbol by inserting #0 or #g or #G\n");
2527 static int find_dso_sym(struct dso *dso, const char *sym_name, u64 *start,
2536 sym = dso__first_symbol(dso);
2540 *size = sym->start - *start;
2544 } else if (dso_sym_match(sym, sym_name, &cnt, idx)) {
2545 print_duplicate_syms(dso, sym_name);
2548 } else if (dso_sym_match(sym, sym_name, &cnt, idx)) {
2549 *start = sym->start;
2550 *size = sym->end - sym->start;
2552 sym = dso__next_symbol(sym);
2556 return sym_not_found_error(sym_name, idx);
2561 static int addr_filter__entire_dso(struct addr_filter *filt, struct dso *dso)
2563 if (dso__data_file_size(dso, NULL)) {
2564 pr_err("Failed to determine filter for %s\nCannot determine file size.\n",
2570 filt->size = dso->data.file_size;
2575 static int addr_filter__resolve_syms(struct addr_filter *filt)
2581 if (!filt->sym_from && !filt->sym_to)
2584 if (!filt->filename)
2585 return addr_filter__resolve_kernel_syms(filt);
2587 dso = load_dso(filt->filename);
2589 pr_err("Failed to load symbols from: %s\n", filt->filename);
2593 if (filt->sym_from && !strcmp(filt->sym_from, "*")) {
2594 err = addr_filter__entire_dso(filt, dso);
2598 if (filt->sym_from) {
2599 err = find_dso_sym(dso, filt->sym_from, &start, &size,
2600 filt->sym_from_idx);
2604 if (filt->range && !filt->size && !filt->sym_to)
2609 err = find_dso_sym(dso, filt->sym_to, &start, &size,
2614 err = check_end_after_start(filt, start, size);
2618 filt->size = start + size - filt->addr;
2627 static char *addr_filter__to_str(struct addr_filter *filt)
2629 char filename_buf[PATH_MAX];
2630 const char *at = "";
2631 const char *fn = "";
2635 if (filt->filename) {
2637 fn = realpath(filt->filename, filename_buf);
2643 err = asprintf(&filter, "%s 0x%"PRIx64"/0x%"PRIx64"%s%s",
2644 filt->action, filt->addr, filt->size, at, fn);
2646 err = asprintf(&filter, "%s 0x%"PRIx64"%s%s",
2647 filt->action, filt->addr, at, fn);
2650 return err < 0 ? NULL : filter;
2653 static int parse_addr_filter(struct evsel *evsel, const char *filter,
2656 struct addr_filters filts;
2657 struct addr_filter *filt;
2660 addr_filters__init(&filts);
2662 err = addr_filters__parse_bare_filter(&filts, filter);
2666 if (filts.cnt > max_nr) {
2667 pr_err("Error: number of address filters (%d) exceeds maximum (%d)\n",
2673 list_for_each_entry(filt, &filts.head, list) {
2676 err = addr_filter__resolve_syms(filt);
2680 new_filter = addr_filter__to_str(filt);
2686 if (evsel__append_addr_filter(evsel, new_filter)) {
2693 addr_filters__exit(&filts);
2696 pr_err("Failed to parse address filter: '%s'\n", filter);
2697 pr_err("Filter format is: filter|start|stop|tracestop <start symbol or address> [/ <end symbol or size>] [@<file name>]\n");
2698 pr_err("Where multiple filters are separated by space or comma.\n");
2704 static int evsel__nr_addr_filter(struct evsel *evsel)
2706 struct perf_pmu *pmu = evsel__find_pmu(evsel);
2707 int nr_addr_filters = 0;
2712 perf_pmu__scan_file(pmu, "nr_addr_filters", "%d", &nr_addr_filters);
2714 return nr_addr_filters;
2717 int auxtrace_parse_filters(struct evlist *evlist)
2719 struct evsel *evsel;
2723 evlist__for_each_entry(evlist, evsel) {
2724 filter = evsel->filter;
2725 max_nr = evsel__nr_addr_filter(evsel);
2726 if (!filter || !max_nr)
2728 evsel->filter = NULL;
2729 err = parse_addr_filter(evsel, filter, max_nr);
2733 pr_debug("Address filter: %s\n", evsel->filter);
2739 int auxtrace__process_event(struct perf_session *session, union perf_event *event,
2740 struct perf_sample *sample, struct perf_tool *tool)
2742 if (!session->auxtrace)
2745 return session->auxtrace->process_event(session, event, sample, tool);
2748 void auxtrace__dump_auxtrace_sample(struct perf_session *session,
2749 struct perf_sample *sample)
2751 if (!session->auxtrace || !session->auxtrace->dump_auxtrace_sample ||
2752 auxtrace__dont_decode(session))
2755 session->auxtrace->dump_auxtrace_sample(session, sample);
2758 int auxtrace__flush_events(struct perf_session *session, struct perf_tool *tool)
2760 if (!session->auxtrace)
2763 return session->auxtrace->flush_events(session, tool);
2766 void auxtrace__free_events(struct perf_session *session)
2768 if (!session->auxtrace)
2771 return session->auxtrace->free_events(session);
2774 void auxtrace__free(struct perf_session *session)
2776 if (!session->auxtrace)
2779 return session->auxtrace->free(session);
2782 bool auxtrace__evsel_is_auxtrace(struct perf_session *session,
2783 struct evsel *evsel)
2785 if (!session->auxtrace || !session->auxtrace->evsel_is_auxtrace)
2788 return session->auxtrace->evsel_is_auxtrace(session, evsel);