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__empty(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(struct auxtrace_queues *queues)
223 queues->nr_queues = AUXTRACE_INIT_NR_QUEUES;
224 queues->queue_array = auxtrace_alloc_queue_array(queues->nr_queues);
225 if (!queues->queue_array)
230 static int auxtrace_queues__grow(struct auxtrace_queues *queues,
231 unsigned int new_nr_queues)
233 unsigned int nr_queues = queues->nr_queues;
234 struct auxtrace_queue *queue_array;
238 nr_queues = AUXTRACE_INIT_NR_QUEUES;
240 while (nr_queues && nr_queues < new_nr_queues)
243 if (nr_queues < queues->nr_queues || nr_queues < new_nr_queues)
246 queue_array = auxtrace_alloc_queue_array(nr_queues);
250 for (i = 0; i < queues->nr_queues; i++) {
251 list_splice_tail(&queues->queue_array[i].head,
252 &queue_array[i].head);
253 queue_array[i].tid = queues->queue_array[i].tid;
254 queue_array[i].cpu = queues->queue_array[i].cpu;
255 queue_array[i].set = queues->queue_array[i].set;
256 queue_array[i].priv = queues->queue_array[i].priv;
259 queues->nr_queues = nr_queues;
260 queues->queue_array = queue_array;
265 static void *auxtrace_copy_data(u64 size, struct perf_session *session)
267 int fd = perf_data__fd(session->data);
271 if (size > SSIZE_MAX)
278 ret = readn(fd, p, size);
279 if (ret != (ssize_t)size) {
287 static int auxtrace_queues__queue_buffer(struct auxtrace_queues *queues,
289 struct auxtrace_buffer *buffer)
291 struct auxtrace_queue *queue;
294 if (idx >= queues->nr_queues) {
295 err = auxtrace_queues__grow(queues, idx + 1);
300 queue = &queues->queue_array[idx];
304 queue->tid = buffer->tid;
305 queue->cpu = buffer->cpu.cpu;
308 buffer->buffer_nr = queues->next_buffer_nr++;
310 list_add_tail(&buffer->list, &queue->head);
312 queues->new_data = true;
313 queues->populated = true;
318 /* Limit buffers to 32MiB on 32-bit */
319 #define BUFFER_LIMIT_FOR_32_BIT (32 * 1024 * 1024)
321 static int auxtrace_queues__split_buffer(struct auxtrace_queues *queues,
323 struct auxtrace_buffer *buffer)
325 u64 sz = buffer->size;
326 bool consecutive = false;
327 struct auxtrace_buffer *b;
330 while (sz > BUFFER_LIMIT_FOR_32_BIT) {
331 b = memdup(buffer, sizeof(struct auxtrace_buffer));
334 b->size = BUFFER_LIMIT_FOR_32_BIT;
335 b->consecutive = consecutive;
336 err = auxtrace_queues__queue_buffer(queues, idx, b);
338 auxtrace_buffer__free(b);
341 buffer->data_offset += BUFFER_LIMIT_FOR_32_BIT;
342 sz -= BUFFER_LIMIT_FOR_32_BIT;
347 buffer->consecutive = consecutive;
352 static bool filter_cpu(struct perf_session *session, struct perf_cpu cpu)
354 unsigned long *cpu_bitmap = session->itrace_synth_opts->cpu_bitmap;
356 return cpu_bitmap && cpu.cpu != -1 && !test_bit(cpu.cpu, cpu_bitmap);
359 static int auxtrace_queues__add_buffer(struct auxtrace_queues *queues,
360 struct perf_session *session,
362 struct auxtrace_buffer *buffer,
363 struct auxtrace_buffer **buffer_ptr)
367 if (filter_cpu(session, buffer->cpu))
370 buffer = memdup(buffer, sizeof(*buffer));
374 if (session->one_mmap) {
375 buffer->data = buffer->data_offset - session->one_mmap_offset +
376 session->one_mmap_addr;
377 } else if (perf_data__is_pipe(session->data)) {
378 buffer->data = auxtrace_copy_data(buffer->size, session);
381 buffer->data_needs_freeing = true;
382 } else if (BITS_PER_LONG == 32 &&
383 buffer->size > BUFFER_LIMIT_FOR_32_BIT) {
384 err = auxtrace_queues__split_buffer(queues, idx, buffer);
389 err = auxtrace_queues__queue_buffer(queues, idx, buffer);
393 /* FIXME: Doesn't work for split buffer */
395 *buffer_ptr = buffer;
400 auxtrace_buffer__free(buffer);
404 int auxtrace_queues__add_event(struct auxtrace_queues *queues,
405 struct perf_session *session,
406 union perf_event *event, off_t data_offset,
407 struct auxtrace_buffer **buffer_ptr)
409 struct auxtrace_buffer buffer = {
411 .tid = event->auxtrace.tid,
412 .cpu = { event->auxtrace.cpu },
413 .data_offset = data_offset,
414 .offset = event->auxtrace.offset,
415 .reference = event->auxtrace.reference,
416 .size = event->auxtrace.size,
418 unsigned int idx = event->auxtrace.idx;
420 return auxtrace_queues__add_buffer(queues, session, idx, &buffer,
424 static int auxtrace_queues__add_indexed_event(struct auxtrace_queues *queues,
425 struct perf_session *session,
426 off_t file_offset, size_t sz)
428 union perf_event *event;
430 char buf[PERF_SAMPLE_MAX_SIZE];
432 err = perf_session__peek_event(session, file_offset, buf,
433 PERF_SAMPLE_MAX_SIZE, &event, NULL);
437 if (event->header.type == PERF_RECORD_AUXTRACE) {
438 if (event->header.size < sizeof(struct perf_record_auxtrace) ||
439 event->header.size != sz) {
443 file_offset += event->header.size;
444 err = auxtrace_queues__add_event(queues, session, event,
451 void auxtrace_queues__free(struct auxtrace_queues *queues)
455 for (i = 0; i < queues->nr_queues; i++) {
456 while (!list_empty(&queues->queue_array[i].head)) {
457 struct auxtrace_buffer *buffer;
459 buffer = list_entry(queues->queue_array[i].head.next,
460 struct auxtrace_buffer, list);
461 list_del_init(&buffer->list);
462 auxtrace_buffer__free(buffer);
466 zfree(&queues->queue_array);
467 queues->nr_queues = 0;
470 static void auxtrace_heapify(struct auxtrace_heap_item *heap_array,
471 unsigned int pos, unsigned int queue_nr,
477 parent = (pos - 1) >> 1;
478 if (heap_array[parent].ordinal <= ordinal)
480 heap_array[pos] = heap_array[parent];
483 heap_array[pos].queue_nr = queue_nr;
484 heap_array[pos].ordinal = ordinal;
487 int auxtrace_heap__add(struct auxtrace_heap *heap, unsigned int queue_nr,
490 struct auxtrace_heap_item *heap_array;
492 if (queue_nr >= heap->heap_sz) {
493 unsigned int heap_sz = AUXTRACE_INIT_NR_QUEUES;
495 while (heap_sz <= queue_nr)
497 heap_array = realloc(heap->heap_array,
498 heap_sz * sizeof(struct auxtrace_heap_item));
501 heap->heap_array = heap_array;
502 heap->heap_sz = heap_sz;
505 auxtrace_heapify(heap->heap_array, heap->heap_cnt++, queue_nr, ordinal);
510 void auxtrace_heap__free(struct auxtrace_heap *heap)
512 zfree(&heap->heap_array);
517 void auxtrace_heap__pop(struct auxtrace_heap *heap)
519 unsigned int pos, last, heap_cnt = heap->heap_cnt;
520 struct auxtrace_heap_item *heap_array;
527 heap_array = heap->heap_array;
531 unsigned int left, right;
533 left = (pos << 1) + 1;
534 if (left >= heap_cnt)
537 if (right >= heap_cnt) {
538 heap_array[pos] = heap_array[left];
541 if (heap_array[left].ordinal < heap_array[right].ordinal) {
542 heap_array[pos] = heap_array[left];
545 heap_array[pos] = heap_array[right];
551 auxtrace_heapify(heap_array, pos, heap_array[last].queue_nr,
552 heap_array[last].ordinal);
555 size_t auxtrace_record__info_priv_size(struct auxtrace_record *itr,
556 struct evlist *evlist)
559 return itr->info_priv_size(itr, evlist);
563 static int auxtrace_not_supported(void)
565 pr_err("AUX area tracing is not supported on this architecture\n");
569 int auxtrace_record__info_fill(struct auxtrace_record *itr,
570 struct perf_session *session,
571 struct perf_record_auxtrace_info *auxtrace_info,
575 return itr->info_fill(itr, session, auxtrace_info, priv_size);
576 return auxtrace_not_supported();
579 void auxtrace_record__free(struct auxtrace_record *itr)
585 int auxtrace_record__snapshot_start(struct auxtrace_record *itr)
587 if (itr && itr->snapshot_start)
588 return itr->snapshot_start(itr);
592 int auxtrace_record__snapshot_finish(struct auxtrace_record *itr, bool on_exit)
594 if (!on_exit && itr && itr->snapshot_finish)
595 return itr->snapshot_finish(itr);
599 int auxtrace_record__find_snapshot(struct auxtrace_record *itr, int idx,
600 struct auxtrace_mmap *mm,
601 unsigned char *data, u64 *head, u64 *old)
603 if (itr && itr->find_snapshot)
604 return itr->find_snapshot(itr, idx, mm, data, head, old);
608 int auxtrace_record__options(struct auxtrace_record *itr,
609 struct evlist *evlist,
610 struct record_opts *opts)
613 itr->evlist = evlist;
614 return itr->recording_options(itr, evlist, opts);
619 u64 auxtrace_record__reference(struct auxtrace_record *itr)
622 return itr->reference(itr);
626 int auxtrace_parse_snapshot_options(struct auxtrace_record *itr,
627 struct record_opts *opts, const char *str)
632 /* PMU-agnostic options */
635 opts->auxtrace_snapshot_on_exit = true;
642 if (itr && itr->parse_snapshot_options)
643 return itr->parse_snapshot_options(itr, opts, str);
645 pr_err("No AUX area tracing to snapshot\n");
649 static int evlist__enable_event_idx(struct evlist *evlist, struct evsel *evsel, int idx)
651 bool per_cpu_mmaps = !perf_cpu_map__empty(evlist->core.user_requested_cpus);
654 struct perf_cpu evlist_cpu = perf_cpu_map__cpu(evlist->core.all_cpus, idx);
655 int cpu_map_idx = perf_cpu_map__idx(evsel->core.cpus, evlist_cpu);
657 if (cpu_map_idx == -1)
659 return perf_evsel__enable_cpu(&evsel->core, cpu_map_idx);
662 return perf_evsel__enable_thread(&evsel->core, idx);
665 int auxtrace_record__read_finish(struct auxtrace_record *itr, int idx)
669 if (!itr->evlist || !itr->pmu)
672 evlist__for_each_entry(itr->evlist, evsel) {
673 if (evsel->core.attr.type == itr->pmu->type) {
676 return evlist__enable_event_idx(itr->evlist, evsel, idx);
683 * Event record size is 16-bit which results in a maximum size of about 64KiB.
684 * Allow about 4KiB for the rest of the sample record, to give a maximum
685 * AUX area sample size of 60KiB.
687 #define MAX_AUX_SAMPLE_SIZE (60 * 1024)
689 /* Arbitrary default size if no other default provided */
690 #define DEFAULT_AUX_SAMPLE_SIZE (4 * 1024)
692 static int auxtrace_validate_aux_sample_size(struct evlist *evlist,
693 struct record_opts *opts)
696 bool has_aux_leader = false;
699 evlist__for_each_entry(evlist, evsel) {
700 sz = evsel->core.attr.aux_sample_size;
701 if (evsel__is_group_leader(evsel)) {
702 has_aux_leader = evsel__is_aux_event(evsel);
705 pr_err("Cannot add AUX area sampling to an AUX area event\n");
707 pr_err("Cannot add AUX area sampling to a group leader\n");
711 if (sz > MAX_AUX_SAMPLE_SIZE) {
712 pr_err("AUX area sample size %u too big, max. %d\n",
713 sz, MAX_AUX_SAMPLE_SIZE);
717 if (!has_aux_leader) {
718 pr_err("Cannot add AUX area sampling because group leader is not an AUX area event\n");
721 evsel__set_sample_bit(evsel, AUX);
722 opts->auxtrace_sample_mode = true;
724 evsel__reset_sample_bit(evsel, AUX);
728 if (!opts->auxtrace_sample_mode) {
729 pr_err("AUX area sampling requires an AUX area event group leader plus other events to which to add samples\n");
733 if (!perf_can_aux_sample()) {
734 pr_err("AUX area sampling is not supported by kernel\n");
741 int auxtrace_parse_sample_options(struct auxtrace_record *itr,
742 struct evlist *evlist,
743 struct record_opts *opts, const char *str)
745 struct evsel_config_term *term;
746 struct evsel *aux_evsel;
747 bool has_aux_sample_size = false;
748 bool has_aux_leader = false;
757 pr_err("No AUX area event to sample\n");
761 sz = strtoul(str, &endptr, 0);
762 if (*endptr || sz > UINT_MAX) {
763 pr_err("Bad AUX area sampling option: '%s'\n", str);
768 sz = itr->default_aux_sample_size;
771 sz = DEFAULT_AUX_SAMPLE_SIZE;
773 /* Set aux_sample_size based on --aux-sample option */
774 evlist__for_each_entry(evlist, evsel) {
775 if (evsel__is_group_leader(evsel)) {
776 has_aux_leader = evsel__is_aux_event(evsel);
777 } else if (has_aux_leader) {
778 evsel->core.attr.aux_sample_size = sz;
783 /* Override with aux_sample_size from config term */
784 evlist__for_each_entry(evlist, evsel) {
785 if (evsel__is_aux_event(evsel))
787 term = evsel__get_config_term(evsel, AUX_SAMPLE_SIZE);
789 has_aux_sample_size = true;
790 evsel->core.attr.aux_sample_size = term->val.aux_sample_size;
791 /* If possible, group with the AUX event */
792 if (aux_evsel && evsel->core.attr.aux_sample_size)
793 evlist__regroup(evlist, aux_evsel, evsel);
797 if (!str && !has_aux_sample_size)
801 pr_err("No AUX area event to sample\n");
805 return auxtrace_validate_aux_sample_size(evlist, opts);
808 void auxtrace_regroup_aux_output(struct evlist *evlist)
810 struct evsel *evsel, *aux_evsel = NULL;
811 struct evsel_config_term *term;
813 evlist__for_each_entry(evlist, evsel) {
814 if (evsel__is_aux_event(evsel))
816 term = evsel__get_config_term(evsel, AUX_OUTPUT);
817 /* If possible, group with the AUX event */
818 if (term && aux_evsel)
819 evlist__regroup(evlist, aux_evsel, evsel);
823 struct auxtrace_record *__weak
824 auxtrace_record__init(struct evlist *evlist __maybe_unused, int *err)
830 static int auxtrace_index__alloc(struct list_head *head)
832 struct auxtrace_index *auxtrace_index;
834 auxtrace_index = malloc(sizeof(struct auxtrace_index));
838 auxtrace_index->nr = 0;
839 INIT_LIST_HEAD(&auxtrace_index->list);
841 list_add_tail(&auxtrace_index->list, head);
846 void auxtrace_index__free(struct list_head *head)
848 struct auxtrace_index *auxtrace_index, *n;
850 list_for_each_entry_safe(auxtrace_index, n, head, list) {
851 list_del_init(&auxtrace_index->list);
852 free(auxtrace_index);
856 static struct auxtrace_index *auxtrace_index__last(struct list_head *head)
858 struct auxtrace_index *auxtrace_index;
861 if (list_empty(head)) {
862 err = auxtrace_index__alloc(head);
867 auxtrace_index = list_entry(head->prev, struct auxtrace_index, list);
869 if (auxtrace_index->nr >= PERF_AUXTRACE_INDEX_ENTRY_COUNT) {
870 err = auxtrace_index__alloc(head);
873 auxtrace_index = list_entry(head->prev, struct auxtrace_index,
877 return auxtrace_index;
880 int auxtrace_index__auxtrace_event(struct list_head *head,
881 union perf_event *event, off_t file_offset)
883 struct auxtrace_index *auxtrace_index;
886 auxtrace_index = auxtrace_index__last(head);
890 nr = auxtrace_index->nr;
891 auxtrace_index->entries[nr].file_offset = file_offset;
892 auxtrace_index->entries[nr].sz = event->header.size;
893 auxtrace_index->nr += 1;
898 static int auxtrace_index__do_write(int fd,
899 struct auxtrace_index *auxtrace_index)
901 struct auxtrace_index_entry ent;
904 for (i = 0; i < auxtrace_index->nr; i++) {
905 ent.file_offset = auxtrace_index->entries[i].file_offset;
906 ent.sz = auxtrace_index->entries[i].sz;
907 if (writen(fd, &ent, sizeof(ent)) != sizeof(ent))
913 int auxtrace_index__write(int fd, struct list_head *head)
915 struct auxtrace_index *auxtrace_index;
919 list_for_each_entry(auxtrace_index, head, list)
920 total += auxtrace_index->nr;
922 if (writen(fd, &total, sizeof(total)) != sizeof(total))
925 list_for_each_entry(auxtrace_index, head, list) {
926 err = auxtrace_index__do_write(fd, auxtrace_index);
934 static int auxtrace_index__process_entry(int fd, struct list_head *head,
937 struct auxtrace_index *auxtrace_index;
938 struct auxtrace_index_entry ent;
941 if (readn(fd, &ent, sizeof(ent)) != sizeof(ent))
944 auxtrace_index = auxtrace_index__last(head);
948 nr = auxtrace_index->nr;
950 auxtrace_index->entries[nr].file_offset =
951 bswap_64(ent.file_offset);
952 auxtrace_index->entries[nr].sz = bswap_64(ent.sz);
954 auxtrace_index->entries[nr].file_offset = ent.file_offset;
955 auxtrace_index->entries[nr].sz = ent.sz;
958 auxtrace_index->nr = nr + 1;
963 int auxtrace_index__process(int fd, u64 size, struct perf_session *session,
966 struct list_head *head = &session->auxtrace_index;
969 if (readn(fd, &nr, sizeof(u64)) != sizeof(u64))
975 if (sizeof(u64) + nr * sizeof(struct auxtrace_index_entry) > size)
981 err = auxtrace_index__process_entry(fd, head, needs_swap);
989 static int auxtrace_queues__process_index_entry(struct auxtrace_queues *queues,
990 struct perf_session *session,
991 struct auxtrace_index_entry *ent)
993 return auxtrace_queues__add_indexed_event(queues, session,
994 ent->file_offset, ent->sz);
997 int auxtrace_queues__process_index(struct auxtrace_queues *queues,
998 struct perf_session *session)
1000 struct auxtrace_index *auxtrace_index;
1001 struct auxtrace_index_entry *ent;
1005 if (auxtrace__dont_decode(session))
1008 list_for_each_entry(auxtrace_index, &session->auxtrace_index, list) {
1009 for (i = 0; i < auxtrace_index->nr; i++) {
1010 ent = &auxtrace_index->entries[i];
1011 err = auxtrace_queues__process_index_entry(queues,
1021 struct auxtrace_buffer *auxtrace_buffer__next(struct auxtrace_queue *queue,
1022 struct auxtrace_buffer *buffer)
1025 if (list_is_last(&buffer->list, &queue->head))
1027 return list_entry(buffer->list.next, struct auxtrace_buffer,
1030 if (list_empty(&queue->head))
1032 return list_entry(queue->head.next, struct auxtrace_buffer,
1037 struct auxtrace_queue *auxtrace_queues__sample_queue(struct auxtrace_queues *queues,
1038 struct perf_sample *sample,
1039 struct perf_session *session)
1041 struct perf_sample_id *sid;
1049 sid = evlist__id2sid(session->evlist, id);
1055 if (idx >= queues->nr_queues)
1058 return &queues->queue_array[idx];
1061 int auxtrace_queues__add_sample(struct auxtrace_queues *queues,
1062 struct perf_session *session,
1063 struct perf_sample *sample, u64 data_offset,
1066 struct auxtrace_buffer buffer = {
1068 .data_offset = data_offset,
1069 .reference = reference,
1070 .size = sample->aux_sample.size,
1072 struct perf_sample_id *sid;
1073 u64 id = sample->id;
1079 sid = evlist__id2sid(session->evlist, id);
1084 buffer.tid = sid->tid;
1085 buffer.cpu = sid->cpu;
1087 return auxtrace_queues__add_buffer(queues, session, idx, &buffer, NULL);
1095 static int auxtrace_queue_data_cb(struct perf_session *session,
1096 union perf_event *event, u64 offset,
1099 struct queue_data *qd = data;
1100 struct perf_sample sample;
1103 if (qd->events && event->header.type == PERF_RECORD_AUXTRACE) {
1104 if (event->header.size < sizeof(struct perf_record_auxtrace))
1106 offset += event->header.size;
1107 return session->auxtrace->queue_data(session, NULL, event,
1111 if (!qd->samples || event->header.type != PERF_RECORD_SAMPLE)
1114 err = evlist__parse_sample(session->evlist, event, &sample);
1118 if (!sample.aux_sample.size)
1121 offset += sample.aux_sample.data - (void *)event;
1123 return session->auxtrace->queue_data(session, &sample, NULL, offset);
1126 int auxtrace_queue_data(struct perf_session *session, bool samples, bool events)
1128 struct queue_data qd = {
1133 if (auxtrace__dont_decode(session))
1136 if (!session->auxtrace || !session->auxtrace->queue_data)
1139 return perf_session__peek_events(session, session->header.data_offset,
1140 session->header.data_size,
1141 auxtrace_queue_data_cb, &qd);
1144 void *auxtrace_buffer__get_data_rw(struct auxtrace_buffer *buffer, int fd, bool rw)
1146 int prot = rw ? PROT_READ | PROT_WRITE : PROT_READ;
1147 size_t adj = buffer->data_offset & (page_size - 1);
1148 size_t size = buffer->size + adj;
1149 off_t file_offset = buffer->data_offset - adj;
1153 return buffer->data;
1155 addr = mmap(NULL, size, prot, MAP_SHARED, fd, file_offset);
1156 if (addr == MAP_FAILED)
1159 buffer->mmap_addr = addr;
1160 buffer->mmap_size = size;
1162 buffer->data = addr + adj;
1164 return buffer->data;
1167 void auxtrace_buffer__put_data(struct auxtrace_buffer *buffer)
1169 if (!buffer->data || !buffer->mmap_addr)
1171 munmap(buffer->mmap_addr, buffer->mmap_size);
1172 buffer->mmap_addr = NULL;
1173 buffer->mmap_size = 0;
1174 buffer->data = NULL;
1175 buffer->use_data = NULL;
1178 void auxtrace_buffer__drop_data(struct auxtrace_buffer *buffer)
1180 auxtrace_buffer__put_data(buffer);
1181 if (buffer->data_needs_freeing) {
1182 buffer->data_needs_freeing = false;
1183 zfree(&buffer->data);
1184 buffer->use_data = NULL;
1189 void auxtrace_buffer__free(struct auxtrace_buffer *buffer)
1191 auxtrace_buffer__drop_data(buffer);
1195 void auxtrace_synth_guest_error(struct perf_record_auxtrace_error *auxtrace_error, int type,
1196 int code, int cpu, pid_t pid, pid_t tid, u64 ip,
1197 const char *msg, u64 timestamp,
1198 pid_t machine_pid, int vcpu)
1202 memset(auxtrace_error, 0, sizeof(struct perf_record_auxtrace_error));
1204 auxtrace_error->header.type = PERF_RECORD_AUXTRACE_ERROR;
1205 auxtrace_error->type = type;
1206 auxtrace_error->code = code;
1207 auxtrace_error->cpu = cpu;
1208 auxtrace_error->pid = pid;
1209 auxtrace_error->tid = tid;
1210 auxtrace_error->fmt = 1;
1211 auxtrace_error->ip = ip;
1212 auxtrace_error->time = timestamp;
1213 strlcpy(auxtrace_error->msg, msg, MAX_AUXTRACE_ERROR_MSG);
1215 auxtrace_error->fmt = 2;
1216 auxtrace_error->machine_pid = machine_pid;
1217 auxtrace_error->vcpu = vcpu;
1218 size = sizeof(*auxtrace_error);
1220 size = (void *)auxtrace_error->msg - (void *)auxtrace_error +
1221 strlen(auxtrace_error->msg) + 1;
1223 auxtrace_error->header.size = PERF_ALIGN(size, sizeof(u64));
1226 void auxtrace_synth_error(struct perf_record_auxtrace_error *auxtrace_error, int type,
1227 int code, int cpu, pid_t pid, pid_t tid, u64 ip,
1228 const char *msg, u64 timestamp)
1230 auxtrace_synth_guest_error(auxtrace_error, type, code, cpu, pid, tid,
1231 ip, msg, timestamp, 0, -1);
1234 int perf_event__synthesize_auxtrace_info(struct auxtrace_record *itr,
1235 struct perf_tool *tool,
1236 struct perf_session *session,
1237 perf_event__handler_t process)
1239 union perf_event *ev;
1243 pr_debug2("Synthesizing auxtrace information\n");
1244 priv_size = auxtrace_record__info_priv_size(itr, session->evlist);
1245 ev = zalloc(sizeof(struct perf_record_auxtrace_info) + priv_size);
1249 ev->auxtrace_info.header.type = PERF_RECORD_AUXTRACE_INFO;
1250 ev->auxtrace_info.header.size = sizeof(struct perf_record_auxtrace_info) +
1252 err = auxtrace_record__info_fill(itr, session, &ev->auxtrace_info,
1257 err = process(tool, ev, NULL, NULL);
1263 static void unleader_evsel(struct evlist *evlist, struct evsel *leader)
1265 struct evsel *new_leader = NULL;
1266 struct evsel *evsel;
1268 /* Find new leader for the group */
1269 evlist__for_each_entry(evlist, evsel) {
1270 if (!evsel__has_leader(evsel, leader) || evsel == leader)
1274 evsel__set_leader(evsel, new_leader);
1277 /* Update group information */
1279 zfree(&new_leader->group_name);
1280 new_leader->group_name = leader->group_name;
1281 leader->group_name = NULL;
1283 new_leader->core.nr_members = leader->core.nr_members - 1;
1284 leader->core.nr_members = 1;
1288 static void unleader_auxtrace(struct perf_session *session)
1290 struct evsel *evsel;
1292 evlist__for_each_entry(session->evlist, evsel) {
1293 if (auxtrace__evsel_is_auxtrace(session, evsel) &&
1294 evsel__is_group_leader(evsel)) {
1295 unleader_evsel(session->evlist, evsel);
1300 int perf_event__process_auxtrace_info(struct perf_session *session,
1301 union perf_event *event)
1303 enum auxtrace_type type = event->auxtrace_info.type;
1307 fprintf(stdout, " type: %u\n", type);
1310 case PERF_AUXTRACE_INTEL_PT:
1311 err = intel_pt_process_auxtrace_info(event, session);
1313 case PERF_AUXTRACE_INTEL_BTS:
1314 err = intel_bts_process_auxtrace_info(event, session);
1316 case PERF_AUXTRACE_ARM_SPE:
1317 err = arm_spe_process_auxtrace_info(event, session);
1319 case PERF_AUXTRACE_CS_ETM:
1320 err = cs_etm__process_auxtrace_info(event, session);
1322 case PERF_AUXTRACE_S390_CPUMSF:
1323 err = s390_cpumsf_process_auxtrace_info(event, session);
1325 case PERF_AUXTRACE_HISI_PTT:
1326 err = hisi_ptt_process_auxtrace_info(event, session);
1328 case PERF_AUXTRACE_UNKNOWN:
1336 unleader_auxtrace(session);
1341 s64 perf_event__process_auxtrace(struct perf_session *session,
1342 union perf_event *event)
1347 fprintf(stdout, " size: %#"PRI_lx64" offset: %#"PRI_lx64" ref: %#"PRI_lx64" idx: %u tid: %d cpu: %d\n",
1348 event->auxtrace.size, event->auxtrace.offset,
1349 event->auxtrace.reference, event->auxtrace.idx,
1350 event->auxtrace.tid, event->auxtrace.cpu);
1352 if (auxtrace__dont_decode(session))
1353 return event->auxtrace.size;
1355 if (!session->auxtrace || event->header.type != PERF_RECORD_AUXTRACE)
1358 err = session->auxtrace->process_auxtrace_event(session, event, session->tool);
1362 return event->auxtrace.size;
1365 #define PERF_ITRACE_DEFAULT_PERIOD_TYPE PERF_ITRACE_PERIOD_NANOSECS
1366 #define PERF_ITRACE_DEFAULT_PERIOD 100000
1367 #define PERF_ITRACE_DEFAULT_CALLCHAIN_SZ 16
1368 #define PERF_ITRACE_MAX_CALLCHAIN_SZ 1024
1369 #define PERF_ITRACE_DEFAULT_LAST_BRANCH_SZ 64
1370 #define PERF_ITRACE_MAX_LAST_BRANCH_SZ 1024
1372 void itrace_synth_opts__set_default(struct itrace_synth_opts *synth_opts,
1375 synth_opts->branches = true;
1376 synth_opts->transactions = true;
1377 synth_opts->ptwrites = true;
1378 synth_opts->pwr_events = true;
1379 synth_opts->other_events = true;
1380 synth_opts->intr_events = true;
1381 synth_opts->errors = true;
1382 synth_opts->flc = true;
1383 synth_opts->llc = true;
1384 synth_opts->tlb = true;
1385 synth_opts->mem = true;
1386 synth_opts->remote_access = true;
1389 synth_opts->period_type = PERF_ITRACE_PERIOD_INSTRUCTIONS;
1390 synth_opts->period = 1;
1391 synth_opts->calls = true;
1393 synth_opts->instructions = true;
1394 synth_opts->period_type = PERF_ITRACE_DEFAULT_PERIOD_TYPE;
1395 synth_opts->period = PERF_ITRACE_DEFAULT_PERIOD;
1397 synth_opts->callchain_sz = PERF_ITRACE_DEFAULT_CALLCHAIN_SZ;
1398 synth_opts->last_branch_sz = PERF_ITRACE_DEFAULT_LAST_BRANCH_SZ;
1399 synth_opts->initial_skip = 0;
1402 static int get_flag(const char **ptr, unsigned int *flags)
1407 if (c >= 'a' && c <= 'z') {
1408 *flags |= 1 << (c - 'a');
1411 } else if (c == ' ') {
1420 static int get_flags(const char **ptr, unsigned int *plus_flags, unsigned int *minus_flags)
1426 if (get_flag(ptr, plus_flags))
1431 if (get_flag(ptr, minus_flags))
1443 #define ITRACE_DFLT_LOG_ON_ERROR_SZ 16384
1445 static unsigned int itrace_log_on_error_size(void)
1447 unsigned int sz = 0;
1449 perf_config_scan("itrace.debug-log-buffer-size", "%u", &sz);
1450 return sz ?: ITRACE_DFLT_LOG_ON_ERROR_SZ;
1454 * Please check tools/perf/Documentation/perf-script.txt for information
1455 * about the options parsed here, which is introduced after this cset,
1456 * when support in 'perf script' for these options is introduced.
1458 int itrace_do_parse_synth_opts(struct itrace_synth_opts *synth_opts,
1459 const char *str, int unset)
1463 bool period_type_set = false;
1464 bool period_set = false;
1466 synth_opts->set = true;
1469 synth_opts->dont_decode = true;
1474 itrace_synth_opts__set_default(synth_opts,
1475 synth_opts->default_no_sample);
1479 for (p = str; *p;) {
1482 synth_opts->instructions = true;
1483 while (*p == ' ' || *p == ',')
1486 synth_opts->period = strtoull(p, &endptr, 10);
1489 while (*p == ' ' || *p == ',')
1493 synth_opts->period_type =
1494 PERF_ITRACE_PERIOD_INSTRUCTIONS;
1495 period_type_set = true;
1498 synth_opts->period_type =
1499 PERF_ITRACE_PERIOD_TICKS;
1500 period_type_set = true;
1503 synth_opts->period *= 1000;
1506 synth_opts->period *= 1000;
1511 synth_opts->period_type =
1512 PERF_ITRACE_PERIOD_NANOSECS;
1513 period_type_set = true;
1523 synth_opts->branches = true;
1526 synth_opts->transactions = true;
1529 synth_opts->ptwrites = true;
1532 synth_opts->pwr_events = true;
1535 synth_opts->other_events = true;
1538 synth_opts->intr_events = true;
1541 synth_opts->errors = true;
1542 if (get_flags(&p, &synth_opts->error_plus_flags,
1543 &synth_opts->error_minus_flags))
1547 synth_opts->log = true;
1548 if (get_flags(&p, &synth_opts->log_plus_flags,
1549 &synth_opts->log_minus_flags))
1551 if (synth_opts->log_plus_flags & AUXTRACE_LOG_FLG_ON_ERROR)
1552 synth_opts->log_on_error_size = itrace_log_on_error_size();
1555 synth_opts->branches = true;
1556 synth_opts->calls = true;
1559 synth_opts->branches = true;
1560 synth_opts->returns = true;
1565 synth_opts->add_callchain = true;
1567 synth_opts->callchain = true;
1568 synth_opts->callchain_sz =
1569 PERF_ITRACE_DEFAULT_CALLCHAIN_SZ;
1570 while (*p == ' ' || *p == ',')
1575 val = strtoul(p, &endptr, 10);
1577 if (!val || val > PERF_ITRACE_MAX_CALLCHAIN_SZ)
1579 synth_opts->callchain_sz = val;
1585 synth_opts->add_last_branch = true;
1587 synth_opts->last_branch = true;
1588 synth_opts->last_branch_sz =
1589 PERF_ITRACE_DEFAULT_LAST_BRANCH_SZ;
1590 while (*p == ' ' || *p == ',')
1595 val = strtoul(p, &endptr, 10);
1598 val > PERF_ITRACE_MAX_LAST_BRANCH_SZ)
1600 synth_opts->last_branch_sz = val;
1604 synth_opts->initial_skip = strtoul(p, &endptr, 10);
1610 synth_opts->flc = true;
1613 synth_opts->llc = true;
1616 synth_opts->tlb = true;
1619 synth_opts->remote_access = true;
1622 synth_opts->mem = true;
1625 synth_opts->quick += 1;
1628 synth_opts->approx_ipc = true;
1631 synth_opts->timeless_decoding = true;
1641 if (synth_opts->instructions) {
1642 if (!period_type_set)
1643 synth_opts->period_type =
1644 PERF_ITRACE_DEFAULT_PERIOD_TYPE;
1646 synth_opts->period = PERF_ITRACE_DEFAULT_PERIOD;
1652 pr_err("Bad Instruction Tracing options '%s'\n", str);
1656 int itrace_parse_synth_opts(const struct option *opt, const char *str, int unset)
1658 return itrace_do_parse_synth_opts(opt->value, str, unset);
1661 static const char * const auxtrace_error_type_name[] = {
1662 [PERF_AUXTRACE_ERROR_ITRACE] = "instruction trace",
1665 static const char *auxtrace_error_name(int type)
1667 const char *error_type_name = NULL;
1669 if (type < PERF_AUXTRACE_ERROR_MAX)
1670 error_type_name = auxtrace_error_type_name[type];
1671 if (!error_type_name)
1672 error_type_name = "unknown AUX";
1673 return error_type_name;
1676 size_t perf_event__fprintf_auxtrace_error(union perf_event *event, FILE *fp)
1678 struct perf_record_auxtrace_error *e = &event->auxtrace_error;
1679 unsigned long long nsecs = e->time;
1680 const char *msg = e->msg;
1683 ret = fprintf(fp, " %s error type %u",
1684 auxtrace_error_name(e->type), e->type);
1686 if (e->fmt && nsecs) {
1687 unsigned long secs = nsecs / NSEC_PER_SEC;
1689 nsecs -= secs * NSEC_PER_SEC;
1690 ret += fprintf(fp, " time %lu.%09llu", secs, nsecs);
1692 ret += fprintf(fp, " time 0");
1696 msg = (const char *)&e->time;
1698 if (e->fmt >= 2 && e->machine_pid)
1699 ret += fprintf(fp, " machine_pid %d vcpu %d", e->machine_pid, e->vcpu);
1701 ret += fprintf(fp, " cpu %d pid %d tid %d ip %#"PRI_lx64" code %u: %s\n",
1702 e->cpu, e->pid, e->tid, e->ip, e->code, msg);
1706 void perf_session__auxtrace_error_inc(struct perf_session *session,
1707 union perf_event *event)
1709 struct perf_record_auxtrace_error *e = &event->auxtrace_error;
1711 if (e->type < PERF_AUXTRACE_ERROR_MAX)
1712 session->evlist->stats.nr_auxtrace_errors[e->type] += 1;
1715 void events_stats__auxtrace_error_warn(const struct events_stats *stats)
1719 for (i = 0; i < PERF_AUXTRACE_ERROR_MAX; i++) {
1720 if (!stats->nr_auxtrace_errors[i])
1722 ui__warning("%u %s errors\n",
1723 stats->nr_auxtrace_errors[i],
1724 auxtrace_error_name(i));
1728 int perf_event__process_auxtrace_error(struct perf_session *session,
1729 union perf_event *event)
1731 if (auxtrace__dont_decode(session))
1734 perf_event__fprintf_auxtrace_error(event, stdout);
1739 * In the compat mode kernel runs in 64-bit and perf tool runs in 32-bit mode,
1740 * 32-bit perf tool cannot access 64-bit value atomically, which might lead to
1741 * the issues caused by the below sequence on multiple CPUs: when perf tool
1742 * accesses either the load operation or the store operation for 64-bit value,
1743 * on some architectures the operation is divided into two instructions, one
1744 * is for accessing the low 32-bit value and another is for the high 32-bit;
1745 * thus these two user operations can give the kernel chances to access the
1746 * 64-bit value, and thus leads to the unexpected load values.
1748 * kernel (64-bit) user (32-bit)
1750 * if (LOAD ->aux_tail) { --, LOAD ->aux_head_lo
1751 * STORE $aux_data | ,--->
1752 * FLUSH $aux_data | | LOAD ->aux_head_hi
1753 * STORE ->aux_head --|-------` smp_rmb()
1756 * | STORE ->aux_tail_lo
1758 * STORE ->aux_tail_hi
1760 * For this reason, it's impossible for the perf tool to work correctly when
1761 * the AUX head or tail is bigger than 4GB (more than 32 bits length); and we
1762 * can not simply limit the AUX ring buffer to less than 4GB, the reason is
1763 * the pointers can be increased monotonically, whatever the buffer size it is,
1764 * at the end the head and tail can be bigger than 4GB and carry out to the
1767 * To mitigate the issues and improve the user experience, we can allow the
1768 * perf tool working in certain conditions and bail out with error if detect
1769 * any overflow cannot be handled.
1771 * For reading the AUX head, it reads out the values for three times, and
1772 * compares the high 4 bytes of the values between the first time and the last
1773 * time, if there has no change for high 4 bytes injected by the kernel during
1774 * the user reading sequence, it's safe for use the second value.
1776 * When compat_auxtrace_mmap__write_tail() detects any carrying in the high
1777 * 32 bits, it means there have two store operations in user space and it cannot
1778 * promise the atomicity for 64-bit write, so return '-1' in this case to tell
1779 * the caller an overflow error has happened.
1781 u64 __weak compat_auxtrace_mmap__read_head(struct auxtrace_mmap *mm)
1783 struct perf_event_mmap_page *pc = mm->userpg;
1784 u64 first, second, last;
1785 u64 mask = (u64)(UINT32_MAX) << 32;
1788 first = READ_ONCE(pc->aux_head);
1789 /* Ensure all reads are done after we read the head */
1791 second = READ_ONCE(pc->aux_head);
1792 /* Ensure all reads are done after we read the head */
1794 last = READ_ONCE(pc->aux_head);
1795 } while ((first & mask) != (last & mask));
1800 int __weak compat_auxtrace_mmap__write_tail(struct auxtrace_mmap *mm, u64 tail)
1802 struct perf_event_mmap_page *pc = mm->userpg;
1803 u64 mask = (u64)(UINT32_MAX) << 32;
1808 /* Ensure all reads are done before we write the tail out */
1810 WRITE_ONCE(pc->aux_tail, tail);
1814 static int __auxtrace_mmap__read(struct mmap *map,
1815 struct auxtrace_record *itr,
1816 struct perf_tool *tool, process_auxtrace_t fn,
1817 bool snapshot, size_t snapshot_size)
1819 struct auxtrace_mmap *mm = &map->auxtrace_mmap;
1820 u64 head, old = mm->prev, offset, ref;
1821 unsigned char *data = mm->base;
1822 size_t size, head_off, old_off, len1, len2, padding;
1823 union perf_event ev;
1824 void *data1, *data2;
1825 int kernel_is_64_bit = perf_env__kernel_is_64_bit(evsel__env(NULL));
1827 head = auxtrace_mmap__read_head(mm, kernel_is_64_bit);
1830 auxtrace_record__find_snapshot(itr, mm->idx, mm, data, &head, &old))
1836 pr_debug3("auxtrace idx %d old %#"PRIx64" head %#"PRIx64" diff %#"PRIx64"\n",
1837 mm->idx, old, head, head - old);
1840 head_off = head & mm->mask;
1841 old_off = old & mm->mask;
1843 head_off = head % mm->len;
1844 old_off = old % mm->len;
1847 if (head_off > old_off)
1848 size = head_off - old_off;
1850 size = mm->len - (old_off - head_off);
1852 if (snapshot && size > snapshot_size)
1853 size = snapshot_size;
1855 ref = auxtrace_record__reference(itr);
1857 if (head > old || size <= head || mm->mask) {
1858 offset = head - size;
1861 * When the buffer size is not a power of 2, 'head' wraps at the
1862 * highest multiple of the buffer size, so we have to subtract
1863 * the remainder here.
1865 u64 rem = (0ULL - mm->len) % mm->len;
1867 offset = head - size - rem;
1870 if (size > head_off) {
1871 len1 = size - head_off;
1872 data1 = &data[mm->len - len1];
1877 data1 = &data[head_off - len1];
1882 if (itr->alignment) {
1883 unsigned int unwanted = len1 % itr->alignment;
1889 /* padding must be written by fn() e.g. record__process_auxtrace() */
1890 padding = size & (PERF_AUXTRACE_RECORD_ALIGNMENT - 1);
1892 padding = PERF_AUXTRACE_RECORD_ALIGNMENT - padding;
1894 memset(&ev, 0, sizeof(ev));
1895 ev.auxtrace.header.type = PERF_RECORD_AUXTRACE;
1896 ev.auxtrace.header.size = sizeof(ev.auxtrace);
1897 ev.auxtrace.size = size + padding;
1898 ev.auxtrace.offset = offset;
1899 ev.auxtrace.reference = ref;
1900 ev.auxtrace.idx = mm->idx;
1901 ev.auxtrace.tid = mm->tid;
1902 ev.auxtrace.cpu = mm->cpu;
1904 if (fn(tool, map, &ev, data1, len1, data2, len2))
1912 err = auxtrace_mmap__write_tail(mm, head, kernel_is_64_bit);
1916 if (itr->read_finish) {
1917 err = itr->read_finish(itr, mm->idx);
1926 int auxtrace_mmap__read(struct mmap *map, struct auxtrace_record *itr,
1927 struct perf_tool *tool, process_auxtrace_t fn)
1929 return __auxtrace_mmap__read(map, itr, tool, fn, false, 0);
1932 int auxtrace_mmap__read_snapshot(struct mmap *map,
1933 struct auxtrace_record *itr,
1934 struct perf_tool *tool, process_auxtrace_t fn,
1935 size_t snapshot_size)
1937 return __auxtrace_mmap__read(map, itr, tool, fn, true, snapshot_size);
1941 * struct auxtrace_cache - hash table to implement a cache
1942 * @hashtable: the hashtable
1943 * @sz: hashtable size (number of hlists)
1944 * @entry_size: size of an entry
1945 * @limit: limit the number of entries to this maximum, when reached the cache
1946 * is dropped and caching begins again with an empty cache
1947 * @cnt: current number of entries
1948 * @bits: hashtable size (@sz = 2^@bits)
1950 struct auxtrace_cache {
1951 struct hlist_head *hashtable;
1959 struct auxtrace_cache *auxtrace_cache__new(unsigned int bits, size_t entry_size,
1960 unsigned int limit_percent)
1962 struct auxtrace_cache *c;
1963 struct hlist_head *ht;
1966 c = zalloc(sizeof(struct auxtrace_cache));
1972 ht = calloc(sz, sizeof(struct hlist_head));
1976 for (i = 0; i < sz; i++)
1977 INIT_HLIST_HEAD(&ht[i]);
1981 c->entry_size = entry_size;
1982 c->limit = (c->sz * limit_percent) / 100;
1992 static void auxtrace_cache__drop(struct auxtrace_cache *c)
1994 struct auxtrace_cache_entry *entry;
1995 struct hlist_node *tmp;
2001 for (i = 0; i < c->sz; i++) {
2002 hlist_for_each_entry_safe(entry, tmp, &c->hashtable[i], hash) {
2003 hlist_del(&entry->hash);
2004 auxtrace_cache__free_entry(c, entry);
2011 void auxtrace_cache__free(struct auxtrace_cache *c)
2016 auxtrace_cache__drop(c);
2017 zfree(&c->hashtable);
2021 void *auxtrace_cache__alloc_entry(struct auxtrace_cache *c)
2023 return malloc(c->entry_size);
2026 void auxtrace_cache__free_entry(struct auxtrace_cache *c __maybe_unused,
2032 int auxtrace_cache__add(struct auxtrace_cache *c, u32 key,
2033 struct auxtrace_cache_entry *entry)
2035 if (c->limit && ++c->cnt > c->limit)
2036 auxtrace_cache__drop(c);
2039 hlist_add_head(&entry->hash, &c->hashtable[hash_32(key, c->bits)]);
2044 static struct auxtrace_cache_entry *auxtrace_cache__rm(struct auxtrace_cache *c,
2047 struct auxtrace_cache_entry *entry;
2048 struct hlist_head *hlist;
2049 struct hlist_node *n;
2054 hlist = &c->hashtable[hash_32(key, c->bits)];
2055 hlist_for_each_entry_safe(entry, n, hlist, hash) {
2056 if (entry->key == key) {
2057 hlist_del(&entry->hash);
2065 void auxtrace_cache__remove(struct auxtrace_cache *c, u32 key)
2067 struct auxtrace_cache_entry *entry = auxtrace_cache__rm(c, key);
2069 auxtrace_cache__free_entry(c, entry);
2072 void *auxtrace_cache__lookup(struct auxtrace_cache *c, u32 key)
2074 struct auxtrace_cache_entry *entry;
2075 struct hlist_head *hlist;
2080 hlist = &c->hashtable[hash_32(key, c->bits)];
2081 hlist_for_each_entry(entry, hlist, hash) {
2082 if (entry->key == key)
2089 static void addr_filter__free_str(struct addr_filter *filt)
2092 filt->action = NULL;
2093 filt->sym_from = NULL;
2094 filt->sym_to = NULL;
2095 filt->filename = NULL;
2098 static struct addr_filter *addr_filter__new(void)
2100 struct addr_filter *filt = zalloc(sizeof(*filt));
2103 INIT_LIST_HEAD(&filt->list);
2108 static void addr_filter__free(struct addr_filter *filt)
2111 addr_filter__free_str(filt);
2115 static void addr_filters__add(struct addr_filters *filts,
2116 struct addr_filter *filt)
2118 list_add_tail(&filt->list, &filts->head);
2122 static void addr_filters__del(struct addr_filters *filts,
2123 struct addr_filter *filt)
2125 list_del_init(&filt->list);
2129 void addr_filters__init(struct addr_filters *filts)
2131 INIT_LIST_HEAD(&filts->head);
2135 void addr_filters__exit(struct addr_filters *filts)
2137 struct addr_filter *filt, *n;
2139 list_for_each_entry_safe(filt, n, &filts->head, list) {
2140 addr_filters__del(filts, filt);
2141 addr_filter__free(filt);
2145 static int parse_num_or_str(char **inp, u64 *num, const char **str,
2146 const char *str_delim)
2148 *inp += strspn(*inp, " ");
2150 if (isdigit(**inp)) {
2156 *num = strtoull(*inp, &endptr, 0);
2167 *inp += strspn(*inp, " ");
2169 n = strcspn(*inp, str_delim);
2181 static int parse_action(struct addr_filter *filt)
2183 if (!strcmp(filt->action, "filter")) {
2186 } else if (!strcmp(filt->action, "start")) {
2188 } else if (!strcmp(filt->action, "stop")) {
2189 filt->start = false;
2190 } else if (!strcmp(filt->action, "tracestop")) {
2191 filt->start = false;
2193 filt->action += 5; /* Change 'tracestop' to 'stop' */
2200 static int parse_sym_idx(char **inp, int *idx)
2204 *inp += strspn(*inp, " ");
2211 if (**inp == 'g' || **inp == 'G') {
2219 num = strtoul(*inp, &endptr, 0);
2222 if (endptr == *inp || num > INT_MAX)
2231 static int parse_addr_size(char **inp, u64 *num, const char **str, int *idx)
2233 int err = parse_num_or_str(inp, num, str, " ");
2236 err = parse_sym_idx(inp, idx);
2241 static int parse_one_filter(struct addr_filter *filt, const char **filter_inp)
2246 filt->str = fstr = strdup(*filter_inp);
2250 err = parse_num_or_str(&fstr, NULL, &filt->action, " ");
2254 err = parse_action(filt);
2258 err = parse_addr_size(&fstr, &filt->addr, &filt->sym_from,
2259 &filt->sym_from_idx);
2263 fstr += strspn(fstr, " ");
2267 err = parse_addr_size(&fstr, &filt->size, &filt->sym_to,
2274 fstr += strspn(fstr, " ");
2278 err = parse_num_or_str(&fstr, NULL, &filt->filename, " ,");
2283 fstr += strspn(fstr, " ,");
2285 *filter_inp += fstr - filt->str;
2290 addr_filter__free_str(filt);
2295 int addr_filters__parse_bare_filter(struct addr_filters *filts,
2298 struct addr_filter *filt;
2299 const char *fstr = filter;
2303 filt = addr_filter__new();
2304 err = parse_one_filter(filt, &fstr);
2306 addr_filter__free(filt);
2307 addr_filters__exit(filts);
2310 addr_filters__add(filts, filt);
2329 static bool kern_sym_name_match(const char *kname, const char *name)
2331 size_t n = strlen(name);
2333 return !strcmp(kname, name) ||
2334 (!strncmp(kname, name, n) && kname[n] == '\t');
2337 static bool kern_sym_match(struct sym_args *args, const char *name, char type)
2339 /* A function with the same name, and global or the n'th found or any */
2340 return kallsyms__is_function(type) &&
2341 kern_sym_name_match(name, args->name) &&
2342 ((args->global && isupper(type)) ||
2343 (args->selected && ++(args->cnt) == args->idx) ||
2344 (!args->global && !args->selected));
2347 static int find_kern_sym_cb(void *arg, const char *name, char type, u64 start)
2349 struct sym_args *args = arg;
2351 if (args->started) {
2353 args->size = start - args->start;
2354 if (args->selected) {
2357 } else if (kern_sym_match(args, name, type)) {
2358 args->duplicate = true;
2361 } else if (kern_sym_match(args, name, type)) {
2362 args->started = true;
2363 args->start = start;
2369 static int print_kern_sym_cb(void *arg, const char *name, char type, u64 start)
2371 struct sym_args *args = arg;
2373 if (kern_sym_match(args, name, type)) {
2374 pr_err("#%d\t0x%"PRIx64"\t%c\t%s\n",
2375 ++args->cnt, start, type, name);
2377 } else if (args->near) {
2379 pr_err("\t\twhich is near\t\t%s\n", name);
2385 static int sym_not_found_error(const char *sym_name, int idx)
2388 pr_err("N'th occurrence (N=%d) of symbol '%s' not found.\n",
2391 pr_err("Global symbol '%s' not found.\n", sym_name);
2393 pr_err("Symbol '%s' not found.\n", sym_name);
2395 pr_err("Note that symbols must be functions.\n");
2400 static int find_kern_sym(const char *sym_name, u64 *start, u64 *size, int idx)
2402 struct sym_args args = {
2406 .selected = idx > 0,
2413 err = kallsyms__parse("/proc/kallsyms", &args, find_kern_sym_cb);
2415 pr_err("Failed to parse /proc/kallsyms\n");
2419 if (args.duplicate) {
2420 pr_err("Multiple kernel symbols with name '%s'\n", sym_name);
2422 kallsyms__parse("/proc/kallsyms", &args, print_kern_sym_cb);
2423 pr_err("Disambiguate symbol name by inserting #n after the name e.g. %s #2\n",
2425 pr_err("Or select a global symbol by inserting #0 or #g or #G\n");
2429 if (!args.started) {
2430 pr_err("Kernel symbol lookup: ");
2431 return sym_not_found_error(sym_name, idx);
2434 *start = args.start;
2440 static int find_entire_kern_cb(void *arg, const char *name __maybe_unused,
2441 char type, u64 start)
2443 struct sym_args *args = arg;
2445 if (!kallsyms__is_function(type))
2448 if (!args->started) {
2449 args->started = true;
2450 args->start = start;
2452 /* Don't know exactly where the kernel ends, so we add a page */
2453 args->size = round_up(start, page_size) + page_size - args->start;
2458 static int addr_filter__entire_kernel(struct addr_filter *filt)
2460 struct sym_args args = { .started = false };
2463 err = kallsyms__parse("/proc/kallsyms", &args, find_entire_kern_cb);
2464 if (err < 0 || !args.started) {
2465 pr_err("Failed to parse /proc/kallsyms\n");
2469 filt->addr = args.start;
2470 filt->size = args.size;
2475 static int check_end_after_start(struct addr_filter *filt, u64 start, u64 size)
2477 if (start + size >= filt->addr)
2480 if (filt->sym_from) {
2481 pr_err("Symbol '%s' (0x%"PRIx64") comes before '%s' (0x%"PRIx64")\n",
2482 filt->sym_to, start, filt->sym_from, filt->addr);
2484 pr_err("Symbol '%s' (0x%"PRIx64") comes before address 0x%"PRIx64")\n",
2485 filt->sym_to, start, filt->addr);
2491 static int addr_filter__resolve_kernel_syms(struct addr_filter *filt)
2493 bool no_size = false;
2497 if (symbol_conf.kptr_restrict) {
2498 pr_err("Kernel addresses are restricted. Unable to resolve kernel symbols.\n");
2502 if (filt->sym_from && !strcmp(filt->sym_from, "*"))
2503 return addr_filter__entire_kernel(filt);
2505 if (filt->sym_from) {
2506 err = find_kern_sym(filt->sym_from, &start, &size,
2507 filt->sym_from_idx);
2511 if (filt->range && !filt->size && !filt->sym_to) {
2518 err = find_kern_sym(filt->sym_to, &start, &size,
2523 err = check_end_after_start(filt, start, size);
2526 filt->size = start + size - filt->addr;
2530 /* The very last symbol in kallsyms does not imply a particular size */
2532 pr_err("Cannot determine size of symbol '%s'\n",
2533 filt->sym_to ? filt->sym_to : filt->sym_from);
2540 static struct dso *load_dso(const char *name)
2545 map = dso__new_map(name);
2549 if (map__load(map) < 0)
2550 pr_err("File '%s' not found or has no symbols.\n", name);
2552 dso = dso__get(map->dso);
2559 static bool dso_sym_match(struct symbol *sym, const char *name, int *cnt,
2562 /* Same name, and global or the n'th found or any */
2563 return !arch__compare_symbol_names(name, sym->name) &&
2564 ((!idx && sym->binding == STB_GLOBAL) ||
2565 (idx > 0 && ++*cnt == idx) ||
2569 static void print_duplicate_syms(struct dso *dso, const char *sym_name)
2575 pr_err("Multiple symbols with name '%s'\n", sym_name);
2577 sym = dso__first_symbol(dso);
2579 if (dso_sym_match(sym, sym_name, &cnt, -1)) {
2580 pr_err("#%d\t0x%"PRIx64"\t%c\t%s\n",
2582 sym->binding == STB_GLOBAL ? 'g' :
2583 sym->binding == STB_LOCAL ? 'l' : 'w',
2588 pr_err("\t\twhich is near\t\t%s\n", sym->name);
2590 sym = dso__next_symbol(sym);
2593 pr_err("Disambiguate symbol name by inserting #n after the name e.g. %s #2\n",
2595 pr_err("Or select a global symbol by inserting #0 or #g or #G\n");
2598 static int find_dso_sym(struct dso *dso, const char *sym_name, u64 *start,
2607 sym = dso__first_symbol(dso);
2611 *size = sym->start - *start;
2615 } else if (dso_sym_match(sym, sym_name, &cnt, idx)) {
2616 print_duplicate_syms(dso, sym_name);
2619 } else if (dso_sym_match(sym, sym_name, &cnt, idx)) {
2620 *start = sym->start;
2621 *size = sym->end - sym->start;
2623 sym = dso__next_symbol(sym);
2627 return sym_not_found_error(sym_name, idx);
2632 static int addr_filter__entire_dso(struct addr_filter *filt, struct dso *dso)
2634 if (dso__data_file_size(dso, NULL)) {
2635 pr_err("Failed to determine filter for %s\nCannot determine file size.\n",
2641 filt->size = dso->data.file_size;
2646 static int addr_filter__resolve_syms(struct addr_filter *filt)
2652 if (!filt->sym_from && !filt->sym_to)
2655 if (!filt->filename)
2656 return addr_filter__resolve_kernel_syms(filt);
2658 dso = load_dso(filt->filename);
2660 pr_err("Failed to load symbols from: %s\n", filt->filename);
2664 if (filt->sym_from && !strcmp(filt->sym_from, "*")) {
2665 err = addr_filter__entire_dso(filt, dso);
2669 if (filt->sym_from) {
2670 err = find_dso_sym(dso, filt->sym_from, &start, &size,
2671 filt->sym_from_idx);
2675 if (filt->range && !filt->size && !filt->sym_to)
2680 err = find_dso_sym(dso, filt->sym_to, &start, &size,
2685 err = check_end_after_start(filt, start, size);
2689 filt->size = start + size - filt->addr;
2698 static char *addr_filter__to_str(struct addr_filter *filt)
2700 char filename_buf[PATH_MAX];
2701 const char *at = "";
2702 const char *fn = "";
2706 if (filt->filename) {
2708 fn = realpath(filt->filename, filename_buf);
2714 err = asprintf(&filter, "%s 0x%"PRIx64"/0x%"PRIx64"%s%s",
2715 filt->action, filt->addr, filt->size, at, fn);
2717 err = asprintf(&filter, "%s 0x%"PRIx64"%s%s",
2718 filt->action, filt->addr, at, fn);
2721 return err < 0 ? NULL : filter;
2724 static int parse_addr_filter(struct evsel *evsel, const char *filter,
2727 struct addr_filters filts;
2728 struct addr_filter *filt;
2731 addr_filters__init(&filts);
2733 err = addr_filters__parse_bare_filter(&filts, filter);
2737 if (filts.cnt > max_nr) {
2738 pr_err("Error: number of address filters (%d) exceeds maximum (%d)\n",
2744 list_for_each_entry(filt, &filts.head, list) {
2747 err = addr_filter__resolve_syms(filt);
2751 new_filter = addr_filter__to_str(filt);
2757 if (evsel__append_addr_filter(evsel, new_filter)) {
2764 addr_filters__exit(&filts);
2767 pr_err("Failed to parse address filter: '%s'\n", filter);
2768 pr_err("Filter format is: filter|start|stop|tracestop <start symbol or address> [/ <end symbol or size>] [@<file name>]\n");
2769 pr_err("Where multiple filters are separated by space or comma.\n");
2775 static int evsel__nr_addr_filter(struct evsel *evsel)
2777 struct perf_pmu *pmu = evsel__find_pmu(evsel);
2778 int nr_addr_filters = 0;
2783 perf_pmu__scan_file(pmu, "nr_addr_filters", "%d", &nr_addr_filters);
2785 return nr_addr_filters;
2788 int auxtrace_parse_filters(struct evlist *evlist)
2790 struct evsel *evsel;
2794 evlist__for_each_entry(evlist, evsel) {
2795 filter = evsel->filter;
2796 max_nr = evsel__nr_addr_filter(evsel);
2797 if (!filter || !max_nr)
2799 evsel->filter = NULL;
2800 err = parse_addr_filter(evsel, filter, max_nr);
2804 pr_debug("Address filter: %s\n", evsel->filter);
2810 int auxtrace__process_event(struct perf_session *session, union perf_event *event,
2811 struct perf_sample *sample, struct perf_tool *tool)
2813 if (!session->auxtrace)
2816 return session->auxtrace->process_event(session, event, sample, tool);
2819 void auxtrace__dump_auxtrace_sample(struct perf_session *session,
2820 struct perf_sample *sample)
2822 if (!session->auxtrace || !session->auxtrace->dump_auxtrace_sample ||
2823 auxtrace__dont_decode(session))
2826 session->auxtrace->dump_auxtrace_sample(session, sample);
2829 int auxtrace__flush_events(struct perf_session *session, struct perf_tool *tool)
2831 if (!session->auxtrace)
2834 return session->auxtrace->flush_events(session, tool);
2837 void auxtrace__free_events(struct perf_session *session)
2839 if (!session->auxtrace)
2842 return session->auxtrace->free_events(session);
2845 void auxtrace__free(struct perf_session *session)
2847 if (!session->auxtrace)
2850 return session->auxtrace->free(session);
2853 bool auxtrace__evsel_is_auxtrace(struct perf_session *session,
2854 struct evsel *evsel)
2856 if (!session->auxtrace || !session->auxtrace->evsel_is_auxtrace)
2859 return session->auxtrace->evsel_is_auxtrace(session, evsel);