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>
35 #include "evsel_config.h"
37 #include "util/perf_api_probe.h"
38 #include "util/synthetic-events.h"
39 #include "thread_map.h"
43 #include <linux/hash.h>
49 #include <subcmd/parse-options.h>
53 #include "intel-bts.h"
56 #include "s390-cpumsf.h"
57 #include "util/mmap.h"
59 #include <linux/ctype.h>
60 #include "symbol/kallsyms.h"
61 #include <internal/lib.h>
62 #include "util/sample.h"
65 * Make a group from 'leader' to 'last', requiring that the events were not
66 * already grouped to a different leader.
68 static int evlist__regroup(struct evlist *evlist, struct evsel *leader, struct evsel *last)
73 if (!evsel__is_group_leader(leader))
77 evlist__for_each_entry(evlist, evsel) {
79 if (!(evsel__leader(evsel) == leader ||
80 (evsel__leader(evsel) == evsel &&
81 evsel->core.nr_members <= 1)))
83 } else if (evsel == leader) {
91 evlist__for_each_entry(evlist, evsel) {
93 if (!evsel__has_leader(evsel, leader)) {
94 evsel__set_leader(evsel, leader);
95 if (leader->core.nr_members < 1)
96 leader->core.nr_members = 1;
97 leader->core.nr_members += 1;
99 } else if (evsel == leader) {
109 static bool auxtrace__dont_decode(struct perf_session *session)
111 return !session->itrace_synth_opts ||
112 session->itrace_synth_opts->dont_decode;
115 int auxtrace_mmap__mmap(struct auxtrace_mmap *mm,
116 struct auxtrace_mmap_params *mp,
117 void *userpg, int fd)
119 struct perf_event_mmap_page *pc = userpg;
121 WARN_ONCE(mm->base, "Uninitialized auxtrace_mmap\n");
129 mm->cpu = mp->cpu.cpu;
131 if (!mp->len || !mp->mmap_needed) {
136 pc->aux_offset = mp->offset;
137 pc->aux_size = mp->len;
139 mm->base = mmap(NULL, mp->len, mp->prot, MAP_SHARED, fd, mp->offset);
140 if (mm->base == MAP_FAILED) {
141 pr_debug2("failed to mmap AUX area\n");
149 void auxtrace_mmap__munmap(struct auxtrace_mmap *mm)
152 munmap(mm->base, mm->len);
157 void auxtrace_mmap_params__init(struct auxtrace_mmap_params *mp,
158 off_t auxtrace_offset,
159 unsigned int auxtrace_pages,
160 bool auxtrace_overwrite)
162 if (auxtrace_pages) {
163 mp->offset = auxtrace_offset;
164 mp->len = auxtrace_pages * (size_t)page_size;
165 mp->mask = is_power_of_2(mp->len) ? mp->len - 1 : 0;
166 mp->prot = PROT_READ | (auxtrace_overwrite ? 0 : PROT_WRITE);
167 pr_debug2("AUX area mmap length %zu\n", mp->len);
173 void auxtrace_mmap_params__set_idx(struct auxtrace_mmap_params *mp,
174 struct evlist *evlist,
175 struct evsel *evsel, int idx)
177 bool per_cpu = !perf_cpu_map__has_any_cpu(evlist->core.user_requested_cpus);
179 mp->mmap_needed = evsel->needs_auxtrace_mmap;
181 if (!mp->mmap_needed)
187 mp->cpu = perf_cpu_map__cpu(evlist->core.all_cpus, idx);
188 if (evlist->core.threads)
189 mp->tid = perf_thread_map__pid(evlist->core.threads, 0);
194 mp->tid = perf_thread_map__pid(evlist->core.threads, idx);
198 #define AUXTRACE_INIT_NR_QUEUES 32
200 static struct auxtrace_queue *auxtrace_alloc_queue_array(unsigned int nr_queues)
202 struct auxtrace_queue *queue_array;
203 unsigned int max_nr_queues, i;
205 max_nr_queues = UINT_MAX / sizeof(struct auxtrace_queue);
206 if (nr_queues > max_nr_queues)
209 queue_array = calloc(nr_queues, sizeof(struct auxtrace_queue));
213 for (i = 0; i < nr_queues; i++) {
214 INIT_LIST_HEAD(&queue_array[i].head);
215 queue_array[i].priv = NULL;
221 int auxtrace_queues__init_nr(struct auxtrace_queues *queues, int nr_queues)
223 queues->nr_queues = nr_queues;
224 queues->queue_array = auxtrace_alloc_queue_array(queues->nr_queues);
225 if (!queues->queue_array)
230 int auxtrace_queues__init(struct auxtrace_queues *queues)
232 return auxtrace_queues__init_nr(queues, AUXTRACE_INIT_NR_QUEUES);
235 static int auxtrace_queues__grow(struct auxtrace_queues *queues,
236 unsigned int new_nr_queues)
238 unsigned int nr_queues = queues->nr_queues;
239 struct auxtrace_queue *queue_array;
243 nr_queues = AUXTRACE_INIT_NR_QUEUES;
245 while (nr_queues && nr_queues < new_nr_queues)
248 if (nr_queues < queues->nr_queues || nr_queues < new_nr_queues)
251 queue_array = auxtrace_alloc_queue_array(nr_queues);
255 for (i = 0; i < queues->nr_queues; i++) {
256 list_splice_tail(&queues->queue_array[i].head,
257 &queue_array[i].head);
258 queue_array[i].tid = queues->queue_array[i].tid;
259 queue_array[i].cpu = queues->queue_array[i].cpu;
260 queue_array[i].set = queues->queue_array[i].set;
261 queue_array[i].priv = queues->queue_array[i].priv;
264 queues->nr_queues = nr_queues;
265 queues->queue_array = queue_array;
270 static void *auxtrace_copy_data(u64 size, struct perf_session *session)
272 int fd = perf_data__fd(session->data);
276 if (size > SSIZE_MAX)
283 ret = readn(fd, p, size);
284 if (ret != (ssize_t)size) {
292 static int auxtrace_queues__queue_buffer(struct auxtrace_queues *queues,
294 struct auxtrace_buffer *buffer)
296 struct auxtrace_queue *queue;
299 if (idx >= queues->nr_queues) {
300 err = auxtrace_queues__grow(queues, idx + 1);
305 queue = &queues->queue_array[idx];
309 queue->tid = buffer->tid;
310 queue->cpu = buffer->cpu.cpu;
313 buffer->buffer_nr = queues->next_buffer_nr++;
315 list_add_tail(&buffer->list, &queue->head);
317 queues->new_data = true;
318 queues->populated = true;
323 /* Limit buffers to 32MiB on 32-bit */
324 #define BUFFER_LIMIT_FOR_32_BIT (32 * 1024 * 1024)
326 static int auxtrace_queues__split_buffer(struct auxtrace_queues *queues,
328 struct auxtrace_buffer *buffer)
330 u64 sz = buffer->size;
331 bool consecutive = false;
332 struct auxtrace_buffer *b;
335 while (sz > BUFFER_LIMIT_FOR_32_BIT) {
336 b = memdup(buffer, sizeof(struct auxtrace_buffer));
339 b->size = BUFFER_LIMIT_FOR_32_BIT;
340 b->consecutive = consecutive;
341 err = auxtrace_queues__queue_buffer(queues, idx, b);
343 auxtrace_buffer__free(b);
346 buffer->data_offset += BUFFER_LIMIT_FOR_32_BIT;
347 sz -= BUFFER_LIMIT_FOR_32_BIT;
352 buffer->consecutive = consecutive;
357 static bool filter_cpu(struct perf_session *session, struct perf_cpu cpu)
359 unsigned long *cpu_bitmap = session->itrace_synth_opts->cpu_bitmap;
361 return cpu_bitmap && cpu.cpu != -1 && !test_bit(cpu.cpu, cpu_bitmap);
364 static int auxtrace_queues__add_buffer(struct auxtrace_queues *queues,
365 struct perf_session *session,
367 struct auxtrace_buffer *buffer,
368 struct auxtrace_buffer **buffer_ptr)
372 if (filter_cpu(session, buffer->cpu))
375 buffer = memdup(buffer, sizeof(*buffer));
379 if (session->one_mmap) {
380 buffer->data = buffer->data_offset - session->one_mmap_offset +
381 session->one_mmap_addr;
382 } else if (perf_data__is_pipe(session->data)) {
383 buffer->data = auxtrace_copy_data(buffer->size, session);
386 buffer->data_needs_freeing = true;
387 } else if (BITS_PER_LONG == 32 &&
388 buffer->size > BUFFER_LIMIT_FOR_32_BIT) {
389 err = auxtrace_queues__split_buffer(queues, idx, buffer);
394 err = auxtrace_queues__queue_buffer(queues, idx, buffer);
398 /* FIXME: Doesn't work for split buffer */
400 *buffer_ptr = buffer;
405 auxtrace_buffer__free(buffer);
409 int auxtrace_queues__add_event(struct auxtrace_queues *queues,
410 struct perf_session *session,
411 union perf_event *event, off_t data_offset,
412 struct auxtrace_buffer **buffer_ptr)
414 struct auxtrace_buffer buffer = {
416 .tid = event->auxtrace.tid,
417 .cpu = { event->auxtrace.cpu },
418 .data_offset = data_offset,
419 .offset = event->auxtrace.offset,
420 .reference = event->auxtrace.reference,
421 .size = event->auxtrace.size,
423 unsigned int idx = event->auxtrace.idx;
425 return auxtrace_queues__add_buffer(queues, session, idx, &buffer,
429 static int auxtrace_queues__add_indexed_event(struct auxtrace_queues *queues,
430 struct perf_session *session,
431 off_t file_offset, size_t sz)
433 union perf_event *event;
435 char buf[PERF_SAMPLE_MAX_SIZE];
437 err = perf_session__peek_event(session, file_offset, buf,
438 PERF_SAMPLE_MAX_SIZE, &event, NULL);
442 if (event->header.type == PERF_RECORD_AUXTRACE) {
443 if (event->header.size < sizeof(struct perf_record_auxtrace) ||
444 event->header.size != sz) {
448 file_offset += event->header.size;
449 err = auxtrace_queues__add_event(queues, session, event,
456 void auxtrace_queues__free(struct auxtrace_queues *queues)
460 for (i = 0; i < queues->nr_queues; i++) {
461 while (!list_empty(&queues->queue_array[i].head)) {
462 struct auxtrace_buffer *buffer;
464 buffer = list_entry(queues->queue_array[i].head.next,
465 struct auxtrace_buffer, list);
466 list_del_init(&buffer->list);
467 auxtrace_buffer__free(buffer);
471 zfree(&queues->queue_array);
472 queues->nr_queues = 0;
475 static void auxtrace_heapify(struct auxtrace_heap_item *heap_array,
476 unsigned int pos, unsigned int queue_nr,
482 parent = (pos - 1) >> 1;
483 if (heap_array[parent].ordinal <= ordinal)
485 heap_array[pos] = heap_array[parent];
488 heap_array[pos].queue_nr = queue_nr;
489 heap_array[pos].ordinal = ordinal;
492 int auxtrace_heap__add(struct auxtrace_heap *heap, unsigned int queue_nr,
495 struct auxtrace_heap_item *heap_array;
497 if (queue_nr >= heap->heap_sz) {
498 unsigned int heap_sz = AUXTRACE_INIT_NR_QUEUES;
500 while (heap_sz <= queue_nr)
502 heap_array = realloc(heap->heap_array,
503 heap_sz * sizeof(struct auxtrace_heap_item));
506 heap->heap_array = heap_array;
507 heap->heap_sz = heap_sz;
510 auxtrace_heapify(heap->heap_array, heap->heap_cnt++, queue_nr, ordinal);
515 void auxtrace_heap__free(struct auxtrace_heap *heap)
517 zfree(&heap->heap_array);
522 void auxtrace_heap__pop(struct auxtrace_heap *heap)
524 unsigned int pos, last, heap_cnt = heap->heap_cnt;
525 struct auxtrace_heap_item *heap_array;
532 heap_array = heap->heap_array;
536 unsigned int left, right;
538 left = (pos << 1) + 1;
539 if (left >= heap_cnt)
542 if (right >= heap_cnt) {
543 heap_array[pos] = heap_array[left];
546 if (heap_array[left].ordinal < heap_array[right].ordinal) {
547 heap_array[pos] = heap_array[left];
550 heap_array[pos] = heap_array[right];
556 auxtrace_heapify(heap_array, pos, heap_array[last].queue_nr,
557 heap_array[last].ordinal);
560 size_t auxtrace_record__info_priv_size(struct auxtrace_record *itr,
561 struct evlist *evlist)
564 return itr->info_priv_size(itr, evlist);
568 static int auxtrace_not_supported(void)
570 pr_err("AUX area tracing is not supported on this architecture\n");
574 int auxtrace_record__info_fill(struct auxtrace_record *itr,
575 struct perf_session *session,
576 struct perf_record_auxtrace_info *auxtrace_info,
580 return itr->info_fill(itr, session, auxtrace_info, priv_size);
581 return auxtrace_not_supported();
584 void auxtrace_record__free(struct auxtrace_record *itr)
590 int auxtrace_record__snapshot_start(struct auxtrace_record *itr)
592 if (itr && itr->snapshot_start)
593 return itr->snapshot_start(itr);
597 int auxtrace_record__snapshot_finish(struct auxtrace_record *itr, bool on_exit)
599 if (!on_exit && itr && itr->snapshot_finish)
600 return itr->snapshot_finish(itr);
604 int auxtrace_record__find_snapshot(struct auxtrace_record *itr, int idx,
605 struct auxtrace_mmap *mm,
606 unsigned char *data, u64 *head, u64 *old)
608 if (itr && itr->find_snapshot)
609 return itr->find_snapshot(itr, idx, mm, data, head, old);
613 int auxtrace_record__options(struct auxtrace_record *itr,
614 struct evlist *evlist,
615 struct record_opts *opts)
618 itr->evlist = evlist;
619 return itr->recording_options(itr, evlist, opts);
624 u64 auxtrace_record__reference(struct auxtrace_record *itr)
627 return itr->reference(itr);
631 int auxtrace_parse_snapshot_options(struct auxtrace_record *itr,
632 struct record_opts *opts, const char *str)
637 /* PMU-agnostic options */
640 opts->auxtrace_snapshot_on_exit = true;
647 if (itr && itr->parse_snapshot_options)
648 return itr->parse_snapshot_options(itr, opts, str);
650 pr_err("No AUX area tracing to snapshot\n");
654 static int evlist__enable_event_idx(struct evlist *evlist, struct evsel *evsel, int idx)
656 bool per_cpu_mmaps = !perf_cpu_map__has_any_cpu(evlist->core.user_requested_cpus);
659 struct perf_cpu evlist_cpu = perf_cpu_map__cpu(evlist->core.all_cpus, idx);
660 int cpu_map_idx = perf_cpu_map__idx(evsel->core.cpus, evlist_cpu);
662 if (cpu_map_idx == -1)
664 return perf_evsel__enable_cpu(&evsel->core, cpu_map_idx);
667 return perf_evsel__enable_thread(&evsel->core, idx);
670 int auxtrace_record__read_finish(struct auxtrace_record *itr, int idx)
674 if (!itr->evlist || !itr->pmu)
677 evlist__for_each_entry(itr->evlist, evsel) {
678 if (evsel->core.attr.type == itr->pmu->type) {
681 return evlist__enable_event_idx(itr->evlist, evsel, idx);
688 * Event record size is 16-bit which results in a maximum size of about 64KiB.
689 * Allow about 4KiB for the rest of the sample record, to give a maximum
690 * AUX area sample size of 60KiB.
692 #define MAX_AUX_SAMPLE_SIZE (60 * 1024)
694 /* Arbitrary default size if no other default provided */
695 #define DEFAULT_AUX_SAMPLE_SIZE (4 * 1024)
697 static int auxtrace_validate_aux_sample_size(struct evlist *evlist,
698 struct record_opts *opts)
701 bool has_aux_leader = false;
704 evlist__for_each_entry(evlist, evsel) {
705 sz = evsel->core.attr.aux_sample_size;
706 if (evsel__is_group_leader(evsel)) {
707 has_aux_leader = evsel__is_aux_event(evsel);
710 pr_err("Cannot add AUX area sampling to an AUX area event\n");
712 pr_err("Cannot add AUX area sampling to a group leader\n");
716 if (sz > MAX_AUX_SAMPLE_SIZE) {
717 pr_err("AUX area sample size %u too big, max. %d\n",
718 sz, MAX_AUX_SAMPLE_SIZE);
722 if (!has_aux_leader) {
723 pr_err("Cannot add AUX area sampling because group leader is not an AUX area event\n");
726 evsel__set_sample_bit(evsel, AUX);
727 opts->auxtrace_sample_mode = true;
729 evsel__reset_sample_bit(evsel, AUX);
733 if (!opts->auxtrace_sample_mode) {
734 pr_err("AUX area sampling requires an AUX area event group leader plus other events to which to add samples\n");
738 if (!perf_can_aux_sample()) {
739 pr_err("AUX area sampling is not supported by kernel\n");
746 int auxtrace_parse_sample_options(struct auxtrace_record *itr,
747 struct evlist *evlist,
748 struct record_opts *opts, const char *str)
750 struct evsel_config_term *term;
751 struct evsel *aux_evsel;
752 bool has_aux_sample_size = false;
753 bool has_aux_leader = false;
762 pr_err("No AUX area event to sample\n");
766 sz = strtoul(str, &endptr, 0);
767 if (*endptr || sz > UINT_MAX) {
768 pr_err("Bad AUX area sampling option: '%s'\n", str);
773 sz = itr->default_aux_sample_size;
776 sz = DEFAULT_AUX_SAMPLE_SIZE;
778 /* Set aux_sample_size based on --aux-sample option */
779 evlist__for_each_entry(evlist, evsel) {
780 if (evsel__is_group_leader(evsel)) {
781 has_aux_leader = evsel__is_aux_event(evsel);
782 } else if (has_aux_leader) {
783 evsel->core.attr.aux_sample_size = sz;
788 /* Override with aux_sample_size from config term */
789 evlist__for_each_entry(evlist, evsel) {
790 if (evsel__is_aux_event(evsel))
792 term = evsel__get_config_term(evsel, AUX_SAMPLE_SIZE);
794 has_aux_sample_size = true;
795 evsel->core.attr.aux_sample_size = term->val.aux_sample_size;
796 /* If possible, group with the AUX event */
797 if (aux_evsel && evsel->core.attr.aux_sample_size)
798 evlist__regroup(evlist, aux_evsel, evsel);
802 if (!str && !has_aux_sample_size)
806 pr_err("No AUX area event to sample\n");
810 return auxtrace_validate_aux_sample_size(evlist, opts);
813 void auxtrace_regroup_aux_output(struct evlist *evlist)
815 struct evsel *evsel, *aux_evsel = NULL;
816 struct evsel_config_term *term;
818 evlist__for_each_entry(evlist, evsel) {
819 if (evsel__is_aux_event(evsel))
821 term = evsel__get_config_term(evsel, AUX_OUTPUT);
822 /* If possible, group with the AUX event */
823 if (term && aux_evsel)
824 evlist__regroup(evlist, aux_evsel, evsel);
828 struct auxtrace_record *__weak
829 auxtrace_record__init(struct evlist *evlist __maybe_unused, int *err)
835 static int auxtrace_index__alloc(struct list_head *head)
837 struct auxtrace_index *auxtrace_index;
839 auxtrace_index = malloc(sizeof(struct auxtrace_index));
843 auxtrace_index->nr = 0;
844 INIT_LIST_HEAD(&auxtrace_index->list);
846 list_add_tail(&auxtrace_index->list, head);
851 void auxtrace_index__free(struct list_head *head)
853 struct auxtrace_index *auxtrace_index, *n;
855 list_for_each_entry_safe(auxtrace_index, n, head, list) {
856 list_del_init(&auxtrace_index->list);
857 free(auxtrace_index);
861 static struct auxtrace_index *auxtrace_index__last(struct list_head *head)
863 struct auxtrace_index *auxtrace_index;
866 if (list_empty(head)) {
867 err = auxtrace_index__alloc(head);
872 auxtrace_index = list_entry(head->prev, struct auxtrace_index, list);
874 if (auxtrace_index->nr >= PERF_AUXTRACE_INDEX_ENTRY_COUNT) {
875 err = auxtrace_index__alloc(head);
878 auxtrace_index = list_entry(head->prev, struct auxtrace_index,
882 return auxtrace_index;
885 int auxtrace_index__auxtrace_event(struct list_head *head,
886 union perf_event *event, off_t file_offset)
888 struct auxtrace_index *auxtrace_index;
891 auxtrace_index = auxtrace_index__last(head);
895 nr = auxtrace_index->nr;
896 auxtrace_index->entries[nr].file_offset = file_offset;
897 auxtrace_index->entries[nr].sz = event->header.size;
898 auxtrace_index->nr += 1;
903 static int auxtrace_index__do_write(int fd,
904 struct auxtrace_index *auxtrace_index)
906 struct auxtrace_index_entry ent;
909 for (i = 0; i < auxtrace_index->nr; i++) {
910 ent.file_offset = auxtrace_index->entries[i].file_offset;
911 ent.sz = auxtrace_index->entries[i].sz;
912 if (writen(fd, &ent, sizeof(ent)) != sizeof(ent))
918 int auxtrace_index__write(int fd, struct list_head *head)
920 struct auxtrace_index *auxtrace_index;
924 list_for_each_entry(auxtrace_index, head, list)
925 total += auxtrace_index->nr;
927 if (writen(fd, &total, sizeof(total)) != sizeof(total))
930 list_for_each_entry(auxtrace_index, head, list) {
931 err = auxtrace_index__do_write(fd, auxtrace_index);
939 static int auxtrace_index__process_entry(int fd, struct list_head *head,
942 struct auxtrace_index *auxtrace_index;
943 struct auxtrace_index_entry ent;
946 if (readn(fd, &ent, sizeof(ent)) != sizeof(ent))
949 auxtrace_index = auxtrace_index__last(head);
953 nr = auxtrace_index->nr;
955 auxtrace_index->entries[nr].file_offset =
956 bswap_64(ent.file_offset);
957 auxtrace_index->entries[nr].sz = bswap_64(ent.sz);
959 auxtrace_index->entries[nr].file_offset = ent.file_offset;
960 auxtrace_index->entries[nr].sz = ent.sz;
963 auxtrace_index->nr = nr + 1;
968 int auxtrace_index__process(int fd, u64 size, struct perf_session *session,
971 struct list_head *head = &session->auxtrace_index;
974 if (readn(fd, &nr, sizeof(u64)) != sizeof(u64))
980 if (sizeof(u64) + nr * sizeof(struct auxtrace_index_entry) > size)
986 err = auxtrace_index__process_entry(fd, head, needs_swap);
994 static int auxtrace_queues__process_index_entry(struct auxtrace_queues *queues,
995 struct perf_session *session,
996 struct auxtrace_index_entry *ent)
998 return auxtrace_queues__add_indexed_event(queues, session,
999 ent->file_offset, ent->sz);
1002 int auxtrace_queues__process_index(struct auxtrace_queues *queues,
1003 struct perf_session *session)
1005 struct auxtrace_index *auxtrace_index;
1006 struct auxtrace_index_entry *ent;
1010 if (auxtrace__dont_decode(session))
1013 list_for_each_entry(auxtrace_index, &session->auxtrace_index, list) {
1014 for (i = 0; i < auxtrace_index->nr; i++) {
1015 ent = &auxtrace_index->entries[i];
1016 err = auxtrace_queues__process_index_entry(queues,
1026 struct auxtrace_buffer *auxtrace_buffer__next(struct auxtrace_queue *queue,
1027 struct auxtrace_buffer *buffer)
1030 if (list_is_last(&buffer->list, &queue->head))
1032 return list_entry(buffer->list.next, struct auxtrace_buffer,
1035 if (list_empty(&queue->head))
1037 return list_entry(queue->head.next, struct auxtrace_buffer,
1042 struct auxtrace_queue *auxtrace_queues__sample_queue(struct auxtrace_queues *queues,
1043 struct perf_sample *sample,
1044 struct perf_session *session)
1046 struct perf_sample_id *sid;
1054 sid = evlist__id2sid(session->evlist, id);
1060 if (idx >= queues->nr_queues)
1063 return &queues->queue_array[idx];
1066 int auxtrace_queues__add_sample(struct auxtrace_queues *queues,
1067 struct perf_session *session,
1068 struct perf_sample *sample, u64 data_offset,
1071 struct auxtrace_buffer buffer = {
1073 .data_offset = data_offset,
1074 .reference = reference,
1075 .size = sample->aux_sample.size,
1077 struct perf_sample_id *sid;
1078 u64 id = sample->id;
1084 sid = evlist__id2sid(session->evlist, id);
1089 buffer.tid = sid->tid;
1090 buffer.cpu = sid->cpu;
1092 return auxtrace_queues__add_buffer(queues, session, idx, &buffer, NULL);
1100 static int auxtrace_queue_data_cb(struct perf_session *session,
1101 union perf_event *event, u64 offset,
1104 struct queue_data *qd = data;
1105 struct perf_sample sample;
1108 if (qd->events && event->header.type == PERF_RECORD_AUXTRACE) {
1109 if (event->header.size < sizeof(struct perf_record_auxtrace))
1111 offset += event->header.size;
1112 return session->auxtrace->queue_data(session, NULL, event,
1116 if (!qd->samples || event->header.type != PERF_RECORD_SAMPLE)
1119 err = evlist__parse_sample(session->evlist, event, &sample);
1123 if (!sample.aux_sample.size)
1126 offset += sample.aux_sample.data - (void *)event;
1128 return session->auxtrace->queue_data(session, &sample, NULL, offset);
1131 int auxtrace_queue_data(struct perf_session *session, bool samples, bool events)
1133 struct queue_data qd = {
1138 if (auxtrace__dont_decode(session))
1141 if (perf_data__is_pipe(session->data))
1144 if (!session->auxtrace || !session->auxtrace->queue_data)
1147 return perf_session__peek_events(session, session->header.data_offset,
1148 session->header.data_size,
1149 auxtrace_queue_data_cb, &qd);
1152 void *auxtrace_buffer__get_data_rw(struct auxtrace_buffer *buffer, int fd, bool rw)
1154 int prot = rw ? PROT_READ | PROT_WRITE : PROT_READ;
1155 size_t adj = buffer->data_offset & (page_size - 1);
1156 size_t size = buffer->size + adj;
1157 off_t file_offset = buffer->data_offset - adj;
1161 return buffer->data;
1163 addr = mmap(NULL, size, prot, MAP_SHARED, fd, file_offset);
1164 if (addr == MAP_FAILED)
1167 buffer->mmap_addr = addr;
1168 buffer->mmap_size = size;
1170 buffer->data = addr + adj;
1172 return buffer->data;
1175 void auxtrace_buffer__put_data(struct auxtrace_buffer *buffer)
1177 if (!buffer->data || !buffer->mmap_addr)
1179 munmap(buffer->mmap_addr, buffer->mmap_size);
1180 buffer->mmap_addr = NULL;
1181 buffer->mmap_size = 0;
1182 buffer->data = NULL;
1183 buffer->use_data = NULL;
1186 void auxtrace_buffer__drop_data(struct auxtrace_buffer *buffer)
1188 auxtrace_buffer__put_data(buffer);
1189 if (buffer->data_needs_freeing) {
1190 buffer->data_needs_freeing = false;
1191 zfree(&buffer->data);
1192 buffer->use_data = NULL;
1197 void auxtrace_buffer__free(struct auxtrace_buffer *buffer)
1199 auxtrace_buffer__drop_data(buffer);
1203 void auxtrace_synth_guest_error(struct perf_record_auxtrace_error *auxtrace_error, int type,
1204 int code, int cpu, pid_t pid, pid_t tid, u64 ip,
1205 const char *msg, u64 timestamp,
1206 pid_t machine_pid, int vcpu)
1210 memset(auxtrace_error, 0, sizeof(struct perf_record_auxtrace_error));
1212 auxtrace_error->header.type = PERF_RECORD_AUXTRACE_ERROR;
1213 auxtrace_error->type = type;
1214 auxtrace_error->code = code;
1215 auxtrace_error->cpu = cpu;
1216 auxtrace_error->pid = pid;
1217 auxtrace_error->tid = tid;
1218 auxtrace_error->fmt = 1;
1219 auxtrace_error->ip = ip;
1220 auxtrace_error->time = timestamp;
1221 strlcpy(auxtrace_error->msg, msg, MAX_AUXTRACE_ERROR_MSG);
1223 auxtrace_error->fmt = 2;
1224 auxtrace_error->machine_pid = machine_pid;
1225 auxtrace_error->vcpu = vcpu;
1226 size = sizeof(*auxtrace_error);
1228 size = (void *)auxtrace_error->msg - (void *)auxtrace_error +
1229 strlen(auxtrace_error->msg) + 1;
1231 auxtrace_error->header.size = PERF_ALIGN(size, sizeof(u64));
1234 void auxtrace_synth_error(struct perf_record_auxtrace_error *auxtrace_error, int type,
1235 int code, int cpu, pid_t pid, pid_t tid, u64 ip,
1236 const char *msg, u64 timestamp)
1238 auxtrace_synth_guest_error(auxtrace_error, type, code, cpu, pid, tid,
1239 ip, msg, timestamp, 0, -1);
1242 int perf_event__synthesize_auxtrace_info(struct auxtrace_record *itr,
1243 struct perf_tool *tool,
1244 struct perf_session *session,
1245 perf_event__handler_t process)
1247 union perf_event *ev;
1251 pr_debug2("Synthesizing auxtrace information\n");
1252 priv_size = auxtrace_record__info_priv_size(itr, session->evlist);
1253 ev = zalloc(sizeof(struct perf_record_auxtrace_info) + priv_size);
1257 ev->auxtrace_info.header.type = PERF_RECORD_AUXTRACE_INFO;
1258 ev->auxtrace_info.header.size = sizeof(struct perf_record_auxtrace_info) +
1260 err = auxtrace_record__info_fill(itr, session, &ev->auxtrace_info,
1265 err = process(tool, ev, NULL, NULL);
1271 static void unleader_evsel(struct evlist *evlist, struct evsel *leader)
1273 struct evsel *new_leader = NULL;
1274 struct evsel *evsel;
1276 /* Find new leader for the group */
1277 evlist__for_each_entry(evlist, evsel) {
1278 if (!evsel__has_leader(evsel, leader) || evsel == leader)
1282 evsel__set_leader(evsel, new_leader);
1285 /* Update group information */
1287 zfree(&new_leader->group_name);
1288 new_leader->group_name = leader->group_name;
1289 leader->group_name = NULL;
1291 new_leader->core.nr_members = leader->core.nr_members - 1;
1292 leader->core.nr_members = 1;
1296 static void unleader_auxtrace(struct perf_session *session)
1298 struct evsel *evsel;
1300 evlist__for_each_entry(session->evlist, evsel) {
1301 if (auxtrace__evsel_is_auxtrace(session, evsel) &&
1302 evsel__is_group_leader(evsel)) {
1303 unleader_evsel(session->evlist, evsel);
1308 int perf_event__process_auxtrace_info(struct perf_session *session,
1309 union perf_event *event)
1311 enum auxtrace_type type = event->auxtrace_info.type;
1315 fprintf(stdout, " type: %u\n", type);
1318 case PERF_AUXTRACE_INTEL_PT:
1319 err = intel_pt_process_auxtrace_info(event, session);
1321 case PERF_AUXTRACE_INTEL_BTS:
1322 err = intel_bts_process_auxtrace_info(event, session);
1324 case PERF_AUXTRACE_ARM_SPE:
1325 err = arm_spe_process_auxtrace_info(event, session);
1327 case PERF_AUXTRACE_CS_ETM:
1328 err = cs_etm__process_auxtrace_info(event, session);
1330 case PERF_AUXTRACE_S390_CPUMSF:
1331 err = s390_cpumsf_process_auxtrace_info(event, session);
1333 case PERF_AUXTRACE_HISI_PTT:
1334 err = hisi_ptt_process_auxtrace_info(event, session);
1336 case PERF_AUXTRACE_UNKNOWN:
1344 unleader_auxtrace(session);
1349 s64 perf_event__process_auxtrace(struct perf_session *session,
1350 union perf_event *event)
1355 fprintf(stdout, " size: %#"PRI_lx64" offset: %#"PRI_lx64" ref: %#"PRI_lx64" idx: %u tid: %d cpu: %d\n",
1356 event->auxtrace.size, event->auxtrace.offset,
1357 event->auxtrace.reference, event->auxtrace.idx,
1358 event->auxtrace.tid, event->auxtrace.cpu);
1360 if (auxtrace__dont_decode(session))
1361 return event->auxtrace.size;
1363 if (!session->auxtrace || event->header.type != PERF_RECORD_AUXTRACE)
1366 err = session->auxtrace->process_auxtrace_event(session, event, session->tool);
1370 return event->auxtrace.size;
1373 #define PERF_ITRACE_DEFAULT_PERIOD_TYPE PERF_ITRACE_PERIOD_NANOSECS
1374 #define PERF_ITRACE_DEFAULT_PERIOD 100000
1375 #define PERF_ITRACE_DEFAULT_CALLCHAIN_SZ 16
1376 #define PERF_ITRACE_MAX_CALLCHAIN_SZ 1024
1377 #define PERF_ITRACE_DEFAULT_LAST_BRANCH_SZ 64
1378 #define PERF_ITRACE_MAX_LAST_BRANCH_SZ 1024
1380 void itrace_synth_opts__set_default(struct itrace_synth_opts *synth_opts,
1383 synth_opts->branches = true;
1384 synth_opts->transactions = true;
1385 synth_opts->ptwrites = true;
1386 synth_opts->pwr_events = true;
1387 synth_opts->other_events = true;
1388 synth_opts->intr_events = true;
1389 synth_opts->errors = true;
1390 synth_opts->flc = true;
1391 synth_opts->llc = true;
1392 synth_opts->tlb = true;
1393 synth_opts->mem = true;
1394 synth_opts->remote_access = true;
1397 synth_opts->period_type = PERF_ITRACE_PERIOD_INSTRUCTIONS;
1398 synth_opts->period = 1;
1399 synth_opts->calls = true;
1401 synth_opts->instructions = true;
1402 synth_opts->cycles = true;
1403 synth_opts->period_type = PERF_ITRACE_DEFAULT_PERIOD_TYPE;
1404 synth_opts->period = PERF_ITRACE_DEFAULT_PERIOD;
1406 synth_opts->callchain_sz = PERF_ITRACE_DEFAULT_CALLCHAIN_SZ;
1407 synth_opts->last_branch_sz = PERF_ITRACE_DEFAULT_LAST_BRANCH_SZ;
1408 synth_opts->initial_skip = 0;
1411 static int get_flag(const char **ptr, unsigned int *flags)
1416 if (c >= 'a' && c <= 'z') {
1417 *flags |= 1 << (c - 'a');
1420 } else if (c == ' ') {
1429 static int get_flags(const char **ptr, unsigned int *plus_flags, unsigned int *minus_flags)
1435 if (get_flag(ptr, plus_flags))
1440 if (get_flag(ptr, minus_flags))
1452 #define ITRACE_DFLT_LOG_ON_ERROR_SZ 16384
1454 static unsigned int itrace_log_on_error_size(void)
1456 unsigned int sz = 0;
1458 perf_config_scan("itrace.debug-log-buffer-size", "%u", &sz);
1459 return sz ?: ITRACE_DFLT_LOG_ON_ERROR_SZ;
1463 * Please check tools/perf/Documentation/perf-script.txt for information
1464 * about the options parsed here, which is introduced after this cset,
1465 * when support in 'perf script' for these options is introduced.
1467 int itrace_do_parse_synth_opts(struct itrace_synth_opts *synth_opts,
1468 const char *str, int unset)
1472 bool period_type_set = false;
1473 bool period_set = false;
1476 synth_opts->set = true;
1479 synth_opts->dont_decode = true;
1484 itrace_synth_opts__set_default(synth_opts,
1485 synth_opts->default_no_sample);
1489 for (p = str; *p;) {
1495 synth_opts->cycles = true;
1497 synth_opts->instructions = true;
1498 while (*p == ' ' || *p == ',')
1501 synth_opts->period = strtoull(p, &endptr, 10);
1504 while (*p == ' ' || *p == ',')
1508 synth_opts->period_type =
1509 PERF_ITRACE_PERIOD_INSTRUCTIONS;
1510 period_type_set = true;
1513 synth_opts->period_type =
1514 PERF_ITRACE_PERIOD_TICKS;
1515 period_type_set = true;
1518 synth_opts->period *= 1000;
1521 synth_opts->period *= 1000;
1526 synth_opts->period_type =
1527 PERF_ITRACE_PERIOD_NANOSECS;
1528 period_type_set = true;
1538 synth_opts->branches = true;
1541 synth_opts->transactions = true;
1544 synth_opts->ptwrites = true;
1547 synth_opts->pwr_events = true;
1550 synth_opts->other_events = true;
1553 synth_opts->intr_events = true;
1556 synth_opts->errors = true;
1557 if (get_flags(&p, &synth_opts->error_plus_flags,
1558 &synth_opts->error_minus_flags))
1562 synth_opts->log = true;
1563 if (get_flags(&p, &synth_opts->log_plus_flags,
1564 &synth_opts->log_minus_flags))
1566 if (synth_opts->log_plus_flags & AUXTRACE_LOG_FLG_ON_ERROR)
1567 synth_opts->log_on_error_size = itrace_log_on_error_size();
1570 synth_opts->branches = true;
1571 synth_opts->calls = true;
1574 synth_opts->branches = true;
1575 synth_opts->returns = true;
1580 synth_opts->add_callchain = true;
1582 synth_opts->callchain = true;
1583 synth_opts->callchain_sz =
1584 PERF_ITRACE_DEFAULT_CALLCHAIN_SZ;
1585 while (*p == ' ' || *p == ',')
1590 val = strtoul(p, &endptr, 10);
1592 if (!val || val > PERF_ITRACE_MAX_CALLCHAIN_SZ)
1594 synth_opts->callchain_sz = val;
1600 synth_opts->add_last_branch = true;
1602 synth_opts->last_branch = true;
1603 synth_opts->last_branch_sz =
1604 PERF_ITRACE_DEFAULT_LAST_BRANCH_SZ;
1605 while (*p == ' ' || *p == ',')
1610 val = strtoul(p, &endptr, 10);
1613 val > PERF_ITRACE_MAX_LAST_BRANCH_SZ)
1615 synth_opts->last_branch_sz = val;
1619 synth_opts->initial_skip = strtoul(p, &endptr, 10);
1625 synth_opts->flc = true;
1628 synth_opts->llc = true;
1631 synth_opts->tlb = true;
1634 synth_opts->remote_access = true;
1637 synth_opts->mem = true;
1640 synth_opts->quick += 1;
1643 synth_opts->approx_ipc = true;
1646 synth_opts->timeless_decoding = true;
1649 synth_opts->use_timestamp = true;
1660 if (!period_type_set)
1661 synth_opts->period_type =
1662 PERF_ITRACE_DEFAULT_PERIOD_TYPE;
1664 synth_opts->period = PERF_ITRACE_DEFAULT_PERIOD;
1670 pr_err("Bad Instruction Tracing options '%s'\n", str);
1674 int itrace_parse_synth_opts(const struct option *opt, const char *str, int unset)
1676 return itrace_do_parse_synth_opts(opt->value, str, unset);
1679 static const char * const auxtrace_error_type_name[] = {
1680 [PERF_AUXTRACE_ERROR_ITRACE] = "instruction trace",
1683 static const char *auxtrace_error_name(int type)
1685 const char *error_type_name = NULL;
1687 if (type < PERF_AUXTRACE_ERROR_MAX)
1688 error_type_name = auxtrace_error_type_name[type];
1689 if (!error_type_name)
1690 error_type_name = "unknown AUX";
1691 return error_type_name;
1694 size_t perf_event__fprintf_auxtrace_error(union perf_event *event, FILE *fp)
1696 struct perf_record_auxtrace_error *e = &event->auxtrace_error;
1697 unsigned long long nsecs = e->time;
1698 const char *msg = e->msg;
1701 ret = fprintf(fp, " %s error type %u",
1702 auxtrace_error_name(e->type), e->type);
1704 if (e->fmt && nsecs) {
1705 unsigned long secs = nsecs / NSEC_PER_SEC;
1707 nsecs -= secs * NSEC_PER_SEC;
1708 ret += fprintf(fp, " time %lu.%09llu", secs, nsecs);
1710 ret += fprintf(fp, " time 0");
1714 msg = (const char *)&e->time;
1716 if (e->fmt >= 2 && e->machine_pid)
1717 ret += fprintf(fp, " machine_pid %d vcpu %d", e->machine_pid, e->vcpu);
1719 ret += fprintf(fp, " cpu %d pid %d tid %d ip %#"PRI_lx64" code %u: %s\n",
1720 e->cpu, e->pid, e->tid, e->ip, e->code, msg);
1724 void perf_session__auxtrace_error_inc(struct perf_session *session,
1725 union perf_event *event)
1727 struct perf_record_auxtrace_error *e = &event->auxtrace_error;
1729 if (e->type < PERF_AUXTRACE_ERROR_MAX)
1730 session->evlist->stats.nr_auxtrace_errors[e->type] += 1;
1733 void events_stats__auxtrace_error_warn(const struct events_stats *stats)
1737 for (i = 0; i < PERF_AUXTRACE_ERROR_MAX; i++) {
1738 if (!stats->nr_auxtrace_errors[i])
1740 ui__warning("%u %s errors\n",
1741 stats->nr_auxtrace_errors[i],
1742 auxtrace_error_name(i));
1746 int perf_event__process_auxtrace_error(struct perf_session *session,
1747 union perf_event *event)
1749 if (auxtrace__dont_decode(session))
1752 perf_event__fprintf_auxtrace_error(event, stdout);
1757 * In the compat mode kernel runs in 64-bit and perf tool runs in 32-bit mode,
1758 * 32-bit perf tool cannot access 64-bit value atomically, which might lead to
1759 * the issues caused by the below sequence on multiple CPUs: when perf tool
1760 * accesses either the load operation or the store operation for 64-bit value,
1761 * on some architectures the operation is divided into two instructions, one
1762 * is for accessing the low 32-bit value and another is for the high 32-bit;
1763 * thus these two user operations can give the kernel chances to access the
1764 * 64-bit value, and thus leads to the unexpected load values.
1766 * kernel (64-bit) user (32-bit)
1768 * if (LOAD ->aux_tail) { --, LOAD ->aux_head_lo
1769 * STORE $aux_data | ,--->
1770 * FLUSH $aux_data | | LOAD ->aux_head_hi
1771 * STORE ->aux_head --|-------` smp_rmb()
1774 * | STORE ->aux_tail_lo
1776 * STORE ->aux_tail_hi
1778 * For this reason, it's impossible for the perf tool to work correctly when
1779 * the AUX head or tail is bigger than 4GB (more than 32 bits length); and we
1780 * can not simply limit the AUX ring buffer to less than 4GB, the reason is
1781 * the pointers can be increased monotonically, whatever the buffer size it is,
1782 * at the end the head and tail can be bigger than 4GB and carry out to the
1785 * To mitigate the issues and improve the user experience, we can allow the
1786 * perf tool working in certain conditions and bail out with error if detect
1787 * any overflow cannot be handled.
1789 * For reading the AUX head, it reads out the values for three times, and
1790 * compares the high 4 bytes of the values between the first time and the last
1791 * time, if there has no change for high 4 bytes injected by the kernel during
1792 * the user reading sequence, it's safe for use the second value.
1794 * When compat_auxtrace_mmap__write_tail() detects any carrying in the high
1795 * 32 bits, it means there have two store operations in user space and it cannot
1796 * promise the atomicity for 64-bit write, so return '-1' in this case to tell
1797 * the caller an overflow error has happened.
1799 u64 __weak compat_auxtrace_mmap__read_head(struct auxtrace_mmap *mm)
1801 struct perf_event_mmap_page *pc = mm->userpg;
1802 u64 first, second, last;
1803 u64 mask = (u64)(UINT32_MAX) << 32;
1806 first = READ_ONCE(pc->aux_head);
1807 /* Ensure all reads are done after we read the head */
1809 second = READ_ONCE(pc->aux_head);
1810 /* Ensure all reads are done after we read the head */
1812 last = READ_ONCE(pc->aux_head);
1813 } while ((first & mask) != (last & mask));
1818 int __weak compat_auxtrace_mmap__write_tail(struct auxtrace_mmap *mm, u64 tail)
1820 struct perf_event_mmap_page *pc = mm->userpg;
1821 u64 mask = (u64)(UINT32_MAX) << 32;
1826 /* Ensure all reads are done before we write the tail out */
1828 WRITE_ONCE(pc->aux_tail, tail);
1832 static int __auxtrace_mmap__read(struct mmap *map,
1833 struct auxtrace_record *itr,
1834 struct perf_tool *tool, process_auxtrace_t fn,
1835 bool snapshot, size_t snapshot_size)
1837 struct auxtrace_mmap *mm = &map->auxtrace_mmap;
1838 u64 head, old = mm->prev, offset, ref;
1839 unsigned char *data = mm->base;
1840 size_t size, head_off, old_off, len1, len2, padding;
1841 union perf_event ev;
1842 void *data1, *data2;
1843 int kernel_is_64_bit = perf_env__kernel_is_64_bit(evsel__env(NULL));
1845 head = auxtrace_mmap__read_head(mm, kernel_is_64_bit);
1848 auxtrace_record__find_snapshot(itr, mm->idx, mm, data, &head, &old))
1854 pr_debug3("auxtrace idx %d old %#"PRIx64" head %#"PRIx64" diff %#"PRIx64"\n",
1855 mm->idx, old, head, head - old);
1858 head_off = head & mm->mask;
1859 old_off = old & mm->mask;
1861 head_off = head % mm->len;
1862 old_off = old % mm->len;
1865 if (head_off > old_off)
1866 size = head_off - old_off;
1868 size = mm->len - (old_off - head_off);
1870 if (snapshot && size > snapshot_size)
1871 size = snapshot_size;
1873 ref = auxtrace_record__reference(itr);
1875 if (head > old || size <= head || mm->mask) {
1876 offset = head - size;
1879 * When the buffer size is not a power of 2, 'head' wraps at the
1880 * highest multiple of the buffer size, so we have to subtract
1881 * the remainder here.
1883 u64 rem = (0ULL - mm->len) % mm->len;
1885 offset = head - size - rem;
1888 if (size > head_off) {
1889 len1 = size - head_off;
1890 data1 = &data[mm->len - len1];
1895 data1 = &data[head_off - len1];
1900 if (itr->alignment) {
1901 unsigned int unwanted = len1 % itr->alignment;
1907 /* padding must be written by fn() e.g. record__process_auxtrace() */
1908 padding = size & (PERF_AUXTRACE_RECORD_ALIGNMENT - 1);
1910 padding = PERF_AUXTRACE_RECORD_ALIGNMENT - padding;
1912 memset(&ev, 0, sizeof(ev));
1913 ev.auxtrace.header.type = PERF_RECORD_AUXTRACE;
1914 ev.auxtrace.header.size = sizeof(ev.auxtrace);
1915 ev.auxtrace.size = size + padding;
1916 ev.auxtrace.offset = offset;
1917 ev.auxtrace.reference = ref;
1918 ev.auxtrace.idx = mm->idx;
1919 ev.auxtrace.tid = mm->tid;
1920 ev.auxtrace.cpu = mm->cpu;
1922 if (fn(tool, map, &ev, data1, len1, data2, len2))
1930 err = auxtrace_mmap__write_tail(mm, head, kernel_is_64_bit);
1934 if (itr->read_finish) {
1935 err = itr->read_finish(itr, mm->idx);
1944 int auxtrace_mmap__read(struct mmap *map, struct auxtrace_record *itr,
1945 struct perf_tool *tool, process_auxtrace_t fn)
1947 return __auxtrace_mmap__read(map, itr, tool, fn, false, 0);
1950 int auxtrace_mmap__read_snapshot(struct mmap *map,
1951 struct auxtrace_record *itr,
1952 struct perf_tool *tool, process_auxtrace_t fn,
1953 size_t snapshot_size)
1955 return __auxtrace_mmap__read(map, itr, tool, fn, true, snapshot_size);
1959 * struct auxtrace_cache - hash table to implement a cache
1960 * @hashtable: the hashtable
1961 * @sz: hashtable size (number of hlists)
1962 * @entry_size: size of an entry
1963 * @limit: limit the number of entries to this maximum, when reached the cache
1964 * is dropped and caching begins again with an empty cache
1965 * @cnt: current number of entries
1966 * @bits: hashtable size (@sz = 2^@bits)
1968 struct auxtrace_cache {
1969 struct hlist_head *hashtable;
1977 struct auxtrace_cache *auxtrace_cache__new(unsigned int bits, size_t entry_size,
1978 unsigned int limit_percent)
1980 struct auxtrace_cache *c;
1981 struct hlist_head *ht;
1984 c = zalloc(sizeof(struct auxtrace_cache));
1990 ht = calloc(sz, sizeof(struct hlist_head));
1994 for (i = 0; i < sz; i++)
1995 INIT_HLIST_HEAD(&ht[i]);
1999 c->entry_size = entry_size;
2000 c->limit = (c->sz * limit_percent) / 100;
2010 static void auxtrace_cache__drop(struct auxtrace_cache *c)
2012 struct auxtrace_cache_entry *entry;
2013 struct hlist_node *tmp;
2019 for (i = 0; i < c->sz; i++) {
2020 hlist_for_each_entry_safe(entry, tmp, &c->hashtable[i], hash) {
2021 hlist_del(&entry->hash);
2022 auxtrace_cache__free_entry(c, entry);
2029 void auxtrace_cache__free(struct auxtrace_cache *c)
2034 auxtrace_cache__drop(c);
2035 zfree(&c->hashtable);
2039 void *auxtrace_cache__alloc_entry(struct auxtrace_cache *c)
2041 return malloc(c->entry_size);
2044 void auxtrace_cache__free_entry(struct auxtrace_cache *c __maybe_unused,
2050 int auxtrace_cache__add(struct auxtrace_cache *c, u32 key,
2051 struct auxtrace_cache_entry *entry)
2053 if (c->limit && ++c->cnt > c->limit)
2054 auxtrace_cache__drop(c);
2057 hlist_add_head(&entry->hash, &c->hashtable[hash_32(key, c->bits)]);
2062 static struct auxtrace_cache_entry *auxtrace_cache__rm(struct auxtrace_cache *c,
2065 struct auxtrace_cache_entry *entry;
2066 struct hlist_head *hlist;
2067 struct hlist_node *n;
2072 hlist = &c->hashtable[hash_32(key, c->bits)];
2073 hlist_for_each_entry_safe(entry, n, hlist, hash) {
2074 if (entry->key == key) {
2075 hlist_del(&entry->hash);
2083 void auxtrace_cache__remove(struct auxtrace_cache *c, u32 key)
2085 struct auxtrace_cache_entry *entry = auxtrace_cache__rm(c, key);
2087 auxtrace_cache__free_entry(c, entry);
2090 void *auxtrace_cache__lookup(struct auxtrace_cache *c, u32 key)
2092 struct auxtrace_cache_entry *entry;
2093 struct hlist_head *hlist;
2098 hlist = &c->hashtable[hash_32(key, c->bits)];
2099 hlist_for_each_entry(entry, hlist, hash) {
2100 if (entry->key == key)
2107 static void addr_filter__free_str(struct addr_filter *filt)
2110 filt->action = NULL;
2111 filt->sym_from = NULL;
2112 filt->sym_to = NULL;
2113 filt->filename = NULL;
2116 static struct addr_filter *addr_filter__new(void)
2118 struct addr_filter *filt = zalloc(sizeof(*filt));
2121 INIT_LIST_HEAD(&filt->list);
2126 static void addr_filter__free(struct addr_filter *filt)
2129 addr_filter__free_str(filt);
2133 static void addr_filters__add(struct addr_filters *filts,
2134 struct addr_filter *filt)
2136 list_add_tail(&filt->list, &filts->head);
2140 static void addr_filters__del(struct addr_filters *filts,
2141 struct addr_filter *filt)
2143 list_del_init(&filt->list);
2147 void addr_filters__init(struct addr_filters *filts)
2149 INIT_LIST_HEAD(&filts->head);
2153 void addr_filters__exit(struct addr_filters *filts)
2155 struct addr_filter *filt, *n;
2157 list_for_each_entry_safe(filt, n, &filts->head, list) {
2158 addr_filters__del(filts, filt);
2159 addr_filter__free(filt);
2163 static int parse_num_or_str(char **inp, u64 *num, const char **str,
2164 const char *str_delim)
2166 *inp += strspn(*inp, " ");
2168 if (isdigit(**inp)) {
2174 *num = strtoull(*inp, &endptr, 0);
2185 *inp += strspn(*inp, " ");
2187 n = strcspn(*inp, str_delim);
2199 static int parse_action(struct addr_filter *filt)
2201 if (!strcmp(filt->action, "filter")) {
2204 } else if (!strcmp(filt->action, "start")) {
2206 } else if (!strcmp(filt->action, "stop")) {
2207 filt->start = false;
2208 } else if (!strcmp(filt->action, "tracestop")) {
2209 filt->start = false;
2211 filt->action += 5; /* Change 'tracestop' to 'stop' */
2218 static int parse_sym_idx(char **inp, int *idx)
2222 *inp += strspn(*inp, " ");
2229 if (**inp == 'g' || **inp == 'G') {
2237 num = strtoul(*inp, &endptr, 0);
2240 if (endptr == *inp || num > INT_MAX)
2249 static int parse_addr_size(char **inp, u64 *num, const char **str, int *idx)
2251 int err = parse_num_or_str(inp, num, str, " ");
2254 err = parse_sym_idx(inp, idx);
2259 static int parse_one_filter(struct addr_filter *filt, const char **filter_inp)
2264 filt->str = fstr = strdup(*filter_inp);
2268 err = parse_num_or_str(&fstr, NULL, &filt->action, " ");
2272 err = parse_action(filt);
2276 err = parse_addr_size(&fstr, &filt->addr, &filt->sym_from,
2277 &filt->sym_from_idx);
2281 fstr += strspn(fstr, " ");
2285 err = parse_addr_size(&fstr, &filt->size, &filt->sym_to,
2292 fstr += strspn(fstr, " ");
2296 err = parse_num_or_str(&fstr, NULL, &filt->filename, " ,");
2301 fstr += strspn(fstr, " ,");
2303 *filter_inp += fstr - filt->str;
2308 addr_filter__free_str(filt);
2313 int addr_filters__parse_bare_filter(struct addr_filters *filts,
2316 struct addr_filter *filt;
2317 const char *fstr = filter;
2321 filt = addr_filter__new();
2322 err = parse_one_filter(filt, &fstr);
2324 addr_filter__free(filt);
2325 addr_filters__exit(filts);
2328 addr_filters__add(filts, filt);
2347 static bool kern_sym_name_match(const char *kname, const char *name)
2349 size_t n = strlen(name);
2351 return !strcmp(kname, name) ||
2352 (!strncmp(kname, name, n) && kname[n] == '\t');
2355 static bool kern_sym_match(struct sym_args *args, const char *name, char type)
2357 /* A function with the same name, and global or the n'th found or any */
2358 return kallsyms__is_function(type) &&
2359 kern_sym_name_match(name, args->name) &&
2360 ((args->global && isupper(type)) ||
2361 (args->selected && ++(args->cnt) == args->idx) ||
2362 (!args->global && !args->selected));
2365 static int find_kern_sym_cb(void *arg, const char *name, char type, u64 start)
2367 struct sym_args *args = arg;
2369 if (args->started) {
2371 args->size = start - args->start;
2372 if (args->selected) {
2375 } else if (kern_sym_match(args, name, type)) {
2376 args->duplicate = true;
2379 } else if (kern_sym_match(args, name, type)) {
2380 args->started = true;
2381 args->start = start;
2387 static int print_kern_sym_cb(void *arg, const char *name, char type, u64 start)
2389 struct sym_args *args = arg;
2391 if (kern_sym_match(args, name, type)) {
2392 pr_err("#%d\t0x%"PRIx64"\t%c\t%s\n",
2393 ++args->cnt, start, type, name);
2395 } else if (args->near) {
2397 pr_err("\t\twhich is near\t\t%s\n", name);
2403 static int sym_not_found_error(const char *sym_name, int idx)
2406 pr_err("N'th occurrence (N=%d) of symbol '%s' not found.\n",
2409 pr_err("Global symbol '%s' not found.\n", sym_name);
2411 pr_err("Symbol '%s' not found.\n", sym_name);
2413 pr_err("Note that symbols must be functions.\n");
2418 static int find_kern_sym(const char *sym_name, u64 *start, u64 *size, int idx)
2420 struct sym_args args = {
2424 .selected = idx > 0,
2431 err = kallsyms__parse("/proc/kallsyms", &args, find_kern_sym_cb);
2433 pr_err("Failed to parse /proc/kallsyms\n");
2437 if (args.duplicate) {
2438 pr_err("Multiple kernel symbols with name '%s'\n", sym_name);
2440 kallsyms__parse("/proc/kallsyms", &args, print_kern_sym_cb);
2441 pr_err("Disambiguate symbol name by inserting #n after the name e.g. %s #2\n",
2443 pr_err("Or select a global symbol by inserting #0 or #g or #G\n");
2447 if (!args.started) {
2448 pr_err("Kernel symbol lookup: ");
2449 return sym_not_found_error(sym_name, idx);
2452 *start = args.start;
2458 static int find_entire_kern_cb(void *arg, const char *name __maybe_unused,
2459 char type, u64 start)
2461 struct sym_args *args = arg;
2464 if (!kallsyms__is_function(type))
2467 if (!args->started) {
2468 args->started = true;
2469 args->start = start;
2471 /* Don't know exactly where the kernel ends, so we add a page */
2472 size = round_up(start, page_size) + page_size - args->start;
2473 if (size > args->size)
2479 static int addr_filter__entire_kernel(struct addr_filter *filt)
2481 struct sym_args args = { .started = false };
2484 err = kallsyms__parse("/proc/kallsyms", &args, find_entire_kern_cb);
2485 if (err < 0 || !args.started) {
2486 pr_err("Failed to parse /proc/kallsyms\n");
2490 filt->addr = args.start;
2491 filt->size = args.size;
2496 static int check_end_after_start(struct addr_filter *filt, u64 start, u64 size)
2498 if (start + size >= filt->addr)
2501 if (filt->sym_from) {
2502 pr_err("Symbol '%s' (0x%"PRIx64") comes before '%s' (0x%"PRIx64")\n",
2503 filt->sym_to, start, filt->sym_from, filt->addr);
2505 pr_err("Symbol '%s' (0x%"PRIx64") comes before address 0x%"PRIx64")\n",
2506 filt->sym_to, start, filt->addr);
2512 static int addr_filter__resolve_kernel_syms(struct addr_filter *filt)
2514 bool no_size = false;
2518 if (symbol_conf.kptr_restrict) {
2519 pr_err("Kernel addresses are restricted. Unable to resolve kernel symbols.\n");
2523 if (filt->sym_from && !strcmp(filt->sym_from, "*"))
2524 return addr_filter__entire_kernel(filt);
2526 if (filt->sym_from) {
2527 err = find_kern_sym(filt->sym_from, &start, &size,
2528 filt->sym_from_idx);
2532 if (filt->range && !filt->size && !filt->sym_to) {
2539 err = find_kern_sym(filt->sym_to, &start, &size,
2544 err = check_end_after_start(filt, start, size);
2547 filt->size = start + size - filt->addr;
2551 /* The very last symbol in kallsyms does not imply a particular size */
2553 pr_err("Cannot determine size of symbol '%s'\n",
2554 filt->sym_to ? filt->sym_to : filt->sym_from);
2561 static struct dso *load_dso(const char *name)
2566 map = dso__new_map(name);
2570 if (map__load(map) < 0)
2571 pr_err("File '%s' not found or has no symbols.\n", name);
2573 dso = dso__get(map__dso(map));
2580 static bool dso_sym_match(struct symbol *sym, const char *name, int *cnt,
2583 /* Same name, and global or the n'th found or any */
2584 return !arch__compare_symbol_names(name, sym->name) &&
2585 ((!idx && sym->binding == STB_GLOBAL) ||
2586 (idx > 0 && ++*cnt == idx) ||
2590 static void print_duplicate_syms(struct dso *dso, const char *sym_name)
2596 pr_err("Multiple symbols with name '%s'\n", sym_name);
2598 sym = dso__first_symbol(dso);
2600 if (dso_sym_match(sym, sym_name, &cnt, -1)) {
2601 pr_err("#%d\t0x%"PRIx64"\t%c\t%s\n",
2603 sym->binding == STB_GLOBAL ? 'g' :
2604 sym->binding == STB_LOCAL ? 'l' : 'w',
2609 pr_err("\t\twhich is near\t\t%s\n", sym->name);
2611 sym = dso__next_symbol(sym);
2614 pr_err("Disambiguate symbol name by inserting #n after the name e.g. %s #2\n",
2616 pr_err("Or select a global symbol by inserting #0 or #g or #G\n");
2619 static int find_dso_sym(struct dso *dso, const char *sym_name, u64 *start,
2628 sym = dso__first_symbol(dso);
2632 *size = sym->start - *start;
2636 } else if (dso_sym_match(sym, sym_name, &cnt, idx)) {
2637 print_duplicate_syms(dso, sym_name);
2640 } else if (dso_sym_match(sym, sym_name, &cnt, idx)) {
2641 *start = sym->start;
2642 *size = sym->end - sym->start;
2644 sym = dso__next_symbol(sym);
2648 return sym_not_found_error(sym_name, idx);
2653 static int addr_filter__entire_dso(struct addr_filter *filt, struct dso *dso)
2655 if (dso__data_file_size(dso, NULL)) {
2656 pr_err("Failed to determine filter for %s\nCannot determine file size.\n",
2662 filt->size = dso__data(dso)->file_size;
2667 static int addr_filter__resolve_syms(struct addr_filter *filt)
2673 if (!filt->sym_from && !filt->sym_to)
2676 if (!filt->filename)
2677 return addr_filter__resolve_kernel_syms(filt);
2679 dso = load_dso(filt->filename);
2681 pr_err("Failed to load symbols from: %s\n", filt->filename);
2685 if (filt->sym_from && !strcmp(filt->sym_from, "*")) {
2686 err = addr_filter__entire_dso(filt, dso);
2690 if (filt->sym_from) {
2691 err = find_dso_sym(dso, filt->sym_from, &start, &size,
2692 filt->sym_from_idx);
2696 if (filt->range && !filt->size && !filt->sym_to)
2701 err = find_dso_sym(dso, filt->sym_to, &start, &size,
2706 err = check_end_after_start(filt, start, size);
2710 filt->size = start + size - filt->addr;
2719 static char *addr_filter__to_str(struct addr_filter *filt)
2721 char filename_buf[PATH_MAX];
2722 const char *at = "";
2723 const char *fn = "";
2727 if (filt->filename) {
2729 fn = realpath(filt->filename, filename_buf);
2735 err = asprintf(&filter, "%s 0x%"PRIx64"/0x%"PRIx64"%s%s",
2736 filt->action, filt->addr, filt->size, at, fn);
2738 err = asprintf(&filter, "%s 0x%"PRIx64"%s%s",
2739 filt->action, filt->addr, at, fn);
2742 return err < 0 ? NULL : filter;
2745 static int parse_addr_filter(struct evsel *evsel, const char *filter,
2748 struct addr_filters filts;
2749 struct addr_filter *filt;
2752 addr_filters__init(&filts);
2754 err = addr_filters__parse_bare_filter(&filts, filter);
2758 if (filts.cnt > max_nr) {
2759 pr_err("Error: number of address filters (%d) exceeds maximum (%d)\n",
2765 list_for_each_entry(filt, &filts.head, list) {
2768 err = addr_filter__resolve_syms(filt);
2772 new_filter = addr_filter__to_str(filt);
2778 if (evsel__append_addr_filter(evsel, new_filter)) {
2785 addr_filters__exit(&filts);
2788 pr_err("Failed to parse address filter: '%s'\n", filter);
2789 pr_err("Filter format is: filter|start|stop|tracestop <start symbol or address> [/ <end symbol or size>] [@<file name>]\n");
2790 pr_err("Where multiple filters are separated by space or comma.\n");
2796 static int evsel__nr_addr_filter(struct evsel *evsel)
2798 struct perf_pmu *pmu = evsel__find_pmu(evsel);
2799 int nr_addr_filters = 0;
2804 perf_pmu__scan_file(pmu, "nr_addr_filters", "%d", &nr_addr_filters);
2806 return nr_addr_filters;
2809 int auxtrace_parse_filters(struct evlist *evlist)
2811 struct evsel *evsel;
2815 evlist__for_each_entry(evlist, evsel) {
2816 filter = evsel->filter;
2817 max_nr = evsel__nr_addr_filter(evsel);
2818 if (!filter || !max_nr)
2820 evsel->filter = NULL;
2821 err = parse_addr_filter(evsel, filter, max_nr);
2825 pr_debug("Address filter: %s\n", evsel->filter);
2831 int auxtrace__process_event(struct perf_session *session, union perf_event *event,
2832 struct perf_sample *sample, struct perf_tool *tool)
2834 if (!session->auxtrace)
2837 return session->auxtrace->process_event(session, event, sample, tool);
2840 void auxtrace__dump_auxtrace_sample(struct perf_session *session,
2841 struct perf_sample *sample)
2843 if (!session->auxtrace || !session->auxtrace->dump_auxtrace_sample ||
2844 auxtrace__dont_decode(session))
2847 session->auxtrace->dump_auxtrace_sample(session, sample);
2850 int auxtrace__flush_events(struct perf_session *session, struct perf_tool *tool)
2852 if (!session->auxtrace)
2855 return session->auxtrace->flush_events(session, tool);
2858 void auxtrace__free_events(struct perf_session *session)
2860 if (!session->auxtrace)
2863 return session->auxtrace->free_events(session);
2866 void auxtrace__free(struct perf_session *session)
2868 if (!session->auxtrace)
2871 return session->auxtrace->free(session);
2874 bool auxtrace__evsel_is_auxtrace(struct perf_session *session,
2875 struct evsel *evsel)
2877 if (!session->auxtrace || !session->auxtrace->evsel_is_auxtrace)
2880 return session->auxtrace->evsel_is_auxtrace(session, evsel);