2 * auxtrace.c: AUX area trace support
3 * Copyright (c) 2013-2015, Intel Corporation.
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
17 #include <sys/types.h>
24 #include <linux/kernel.h>
25 #include <linux/perf_event.h>
26 #include <linux/types.h>
27 #include <linux/bitops.h>
28 #include <linux/log2.h>
29 #include <linux/string.h>
31 #include <sys/param.h>
34 #include <linux/list.h>
44 #include "thread_map.h"
48 #include <linux/hash.h>
53 #include <subcmd/parse-options.h>
57 #include "intel-bts.h"
59 #include "s390-cpumsf.h"
61 #include "sane_ctype.h"
62 #include "symbol/kallsyms.h"
64 static bool auxtrace__dont_decode(struct perf_session *session)
66 return !session->itrace_synth_opts ||
67 session->itrace_synth_opts->dont_decode;
70 int auxtrace_mmap__mmap(struct auxtrace_mmap *mm,
71 struct auxtrace_mmap_params *mp,
74 struct perf_event_mmap_page *pc = userpg;
76 WARN_ONCE(mm->base, "Uninitialized auxtrace_mmap\n");
91 #if BITS_PER_LONG != 64 && !defined(HAVE_SYNC_COMPARE_AND_SWAP_SUPPORT)
92 pr_err("Cannot use AUX area tracing mmaps\n");
96 pc->aux_offset = mp->offset;
97 pc->aux_size = mp->len;
99 mm->base = mmap(NULL, mp->len, mp->prot, MAP_SHARED, fd, mp->offset);
100 if (mm->base == MAP_FAILED) {
101 pr_debug2("failed to mmap AUX area\n");
109 void auxtrace_mmap__munmap(struct auxtrace_mmap *mm)
112 munmap(mm->base, mm->len);
117 void auxtrace_mmap_params__init(struct auxtrace_mmap_params *mp,
118 off_t auxtrace_offset,
119 unsigned int auxtrace_pages,
120 bool auxtrace_overwrite)
122 if (auxtrace_pages) {
123 mp->offset = auxtrace_offset;
124 mp->len = auxtrace_pages * (size_t)page_size;
125 mp->mask = is_power_of_2(mp->len) ? mp->len - 1 : 0;
126 mp->prot = PROT_READ | (auxtrace_overwrite ? 0 : PROT_WRITE);
127 pr_debug2("AUX area mmap length %zu\n", mp->len);
133 void auxtrace_mmap_params__set_idx(struct auxtrace_mmap_params *mp,
134 struct perf_evlist *evlist, int idx,
140 mp->cpu = evlist->cpus->map[idx];
142 mp->tid = thread_map__pid(evlist->threads, 0);
147 mp->tid = thread_map__pid(evlist->threads, idx);
151 #define AUXTRACE_INIT_NR_QUEUES 32
153 static struct auxtrace_queue *auxtrace_alloc_queue_array(unsigned int nr_queues)
155 struct auxtrace_queue *queue_array;
156 unsigned int max_nr_queues, i;
158 max_nr_queues = UINT_MAX / sizeof(struct auxtrace_queue);
159 if (nr_queues > max_nr_queues)
162 queue_array = calloc(nr_queues, sizeof(struct auxtrace_queue));
166 for (i = 0; i < nr_queues; i++) {
167 INIT_LIST_HEAD(&queue_array[i].head);
168 queue_array[i].priv = NULL;
174 int auxtrace_queues__init(struct auxtrace_queues *queues)
176 queues->nr_queues = AUXTRACE_INIT_NR_QUEUES;
177 queues->queue_array = auxtrace_alloc_queue_array(queues->nr_queues);
178 if (!queues->queue_array)
183 static int auxtrace_queues__grow(struct auxtrace_queues *queues,
184 unsigned int new_nr_queues)
186 unsigned int nr_queues = queues->nr_queues;
187 struct auxtrace_queue *queue_array;
191 nr_queues = AUXTRACE_INIT_NR_QUEUES;
193 while (nr_queues && nr_queues < new_nr_queues)
196 if (nr_queues < queues->nr_queues || nr_queues < new_nr_queues)
199 queue_array = auxtrace_alloc_queue_array(nr_queues);
203 for (i = 0; i < queues->nr_queues; i++) {
204 list_splice_tail(&queues->queue_array[i].head,
205 &queue_array[i].head);
206 queue_array[i].tid = queues->queue_array[i].tid;
207 queue_array[i].cpu = queues->queue_array[i].cpu;
208 queue_array[i].set = queues->queue_array[i].set;
209 queue_array[i].priv = queues->queue_array[i].priv;
212 queues->nr_queues = nr_queues;
213 queues->queue_array = queue_array;
218 static void *auxtrace_copy_data(u64 size, struct perf_session *session)
220 int fd = perf_data__fd(session->data);
224 if (size > SSIZE_MAX)
231 ret = readn(fd, p, size);
232 if (ret != (ssize_t)size) {
240 static int auxtrace_queues__queue_buffer(struct auxtrace_queues *queues,
242 struct auxtrace_buffer *buffer)
244 struct auxtrace_queue *queue;
247 if (idx >= queues->nr_queues) {
248 err = auxtrace_queues__grow(queues, idx + 1);
253 queue = &queues->queue_array[idx];
257 queue->tid = buffer->tid;
258 queue->cpu = buffer->cpu;
261 buffer->buffer_nr = queues->next_buffer_nr++;
263 list_add_tail(&buffer->list, &queue->head);
265 queues->new_data = true;
266 queues->populated = true;
271 /* Limit buffers to 32MiB on 32-bit */
272 #define BUFFER_LIMIT_FOR_32_BIT (32 * 1024 * 1024)
274 static int auxtrace_queues__split_buffer(struct auxtrace_queues *queues,
276 struct auxtrace_buffer *buffer)
278 u64 sz = buffer->size;
279 bool consecutive = false;
280 struct auxtrace_buffer *b;
283 while (sz > BUFFER_LIMIT_FOR_32_BIT) {
284 b = memdup(buffer, sizeof(struct auxtrace_buffer));
287 b->size = BUFFER_LIMIT_FOR_32_BIT;
288 b->consecutive = consecutive;
289 err = auxtrace_queues__queue_buffer(queues, idx, b);
291 auxtrace_buffer__free(b);
294 buffer->data_offset += BUFFER_LIMIT_FOR_32_BIT;
295 sz -= BUFFER_LIMIT_FOR_32_BIT;
300 buffer->consecutive = consecutive;
305 static bool filter_cpu(struct perf_session *session, int cpu)
307 unsigned long *cpu_bitmap = session->itrace_synth_opts->cpu_bitmap;
309 return cpu_bitmap && cpu != -1 && !test_bit(cpu, cpu_bitmap);
312 static int auxtrace_queues__add_buffer(struct auxtrace_queues *queues,
313 struct perf_session *session,
315 struct auxtrace_buffer *buffer,
316 struct auxtrace_buffer **buffer_ptr)
320 if (filter_cpu(session, buffer->cpu))
323 buffer = memdup(buffer, sizeof(*buffer));
327 if (session->one_mmap) {
328 buffer->data = buffer->data_offset - session->one_mmap_offset +
329 session->one_mmap_addr;
330 } else if (perf_data__is_pipe(session->data)) {
331 buffer->data = auxtrace_copy_data(buffer->size, session);
334 buffer->data_needs_freeing = true;
335 } else if (BITS_PER_LONG == 32 &&
336 buffer->size > BUFFER_LIMIT_FOR_32_BIT) {
337 err = auxtrace_queues__split_buffer(queues, idx, buffer);
342 err = auxtrace_queues__queue_buffer(queues, idx, buffer);
346 /* FIXME: Doesn't work for split buffer */
348 *buffer_ptr = buffer;
353 auxtrace_buffer__free(buffer);
357 int auxtrace_queues__add_event(struct auxtrace_queues *queues,
358 struct perf_session *session,
359 union perf_event *event, off_t data_offset,
360 struct auxtrace_buffer **buffer_ptr)
362 struct auxtrace_buffer buffer = {
364 .tid = event->auxtrace.tid,
365 .cpu = event->auxtrace.cpu,
366 .data_offset = data_offset,
367 .offset = event->auxtrace.offset,
368 .reference = event->auxtrace.reference,
369 .size = event->auxtrace.size,
371 unsigned int idx = event->auxtrace.idx;
373 return auxtrace_queues__add_buffer(queues, session, idx, &buffer,
377 static int auxtrace_queues__add_indexed_event(struct auxtrace_queues *queues,
378 struct perf_session *session,
379 off_t file_offset, size_t sz)
381 union perf_event *event;
383 char buf[PERF_SAMPLE_MAX_SIZE];
385 err = perf_session__peek_event(session, file_offset, buf,
386 PERF_SAMPLE_MAX_SIZE, &event, NULL);
390 if (event->header.type == PERF_RECORD_AUXTRACE) {
391 if (event->header.size < sizeof(struct auxtrace_event) ||
392 event->header.size != sz) {
396 file_offset += event->header.size;
397 err = auxtrace_queues__add_event(queues, session, event,
404 void auxtrace_queues__free(struct auxtrace_queues *queues)
408 for (i = 0; i < queues->nr_queues; i++) {
409 while (!list_empty(&queues->queue_array[i].head)) {
410 struct auxtrace_buffer *buffer;
412 buffer = list_entry(queues->queue_array[i].head.next,
413 struct auxtrace_buffer, list);
414 list_del(&buffer->list);
415 auxtrace_buffer__free(buffer);
419 zfree(&queues->queue_array);
420 queues->nr_queues = 0;
423 static void auxtrace_heapify(struct auxtrace_heap_item *heap_array,
424 unsigned int pos, unsigned int queue_nr,
430 parent = (pos - 1) >> 1;
431 if (heap_array[parent].ordinal <= ordinal)
433 heap_array[pos] = heap_array[parent];
436 heap_array[pos].queue_nr = queue_nr;
437 heap_array[pos].ordinal = ordinal;
440 int auxtrace_heap__add(struct auxtrace_heap *heap, unsigned int queue_nr,
443 struct auxtrace_heap_item *heap_array;
445 if (queue_nr >= heap->heap_sz) {
446 unsigned int heap_sz = AUXTRACE_INIT_NR_QUEUES;
448 while (heap_sz <= queue_nr)
450 heap_array = realloc(heap->heap_array,
451 heap_sz * sizeof(struct auxtrace_heap_item));
454 heap->heap_array = heap_array;
455 heap->heap_sz = heap_sz;
458 auxtrace_heapify(heap->heap_array, heap->heap_cnt++, queue_nr, ordinal);
463 void auxtrace_heap__free(struct auxtrace_heap *heap)
465 zfree(&heap->heap_array);
470 void auxtrace_heap__pop(struct auxtrace_heap *heap)
472 unsigned int pos, last, heap_cnt = heap->heap_cnt;
473 struct auxtrace_heap_item *heap_array;
480 heap_array = heap->heap_array;
484 unsigned int left, right;
486 left = (pos << 1) + 1;
487 if (left >= heap_cnt)
490 if (right >= heap_cnt) {
491 heap_array[pos] = heap_array[left];
494 if (heap_array[left].ordinal < heap_array[right].ordinal) {
495 heap_array[pos] = heap_array[left];
498 heap_array[pos] = heap_array[right];
504 auxtrace_heapify(heap_array, pos, heap_array[last].queue_nr,
505 heap_array[last].ordinal);
508 size_t auxtrace_record__info_priv_size(struct auxtrace_record *itr,
509 struct perf_evlist *evlist)
512 return itr->info_priv_size(itr, evlist);
516 static int auxtrace_not_supported(void)
518 pr_err("AUX area tracing is not supported on this architecture\n");
522 int auxtrace_record__info_fill(struct auxtrace_record *itr,
523 struct perf_session *session,
524 struct auxtrace_info_event *auxtrace_info,
528 return itr->info_fill(itr, session, auxtrace_info, priv_size);
529 return auxtrace_not_supported();
532 void auxtrace_record__free(struct auxtrace_record *itr)
538 int auxtrace_record__snapshot_start(struct auxtrace_record *itr)
540 if (itr && itr->snapshot_start)
541 return itr->snapshot_start(itr);
545 int auxtrace_record__snapshot_finish(struct auxtrace_record *itr)
547 if (itr && itr->snapshot_finish)
548 return itr->snapshot_finish(itr);
552 int auxtrace_record__find_snapshot(struct auxtrace_record *itr, int idx,
553 struct auxtrace_mmap *mm,
554 unsigned char *data, u64 *head, u64 *old)
556 if (itr && itr->find_snapshot)
557 return itr->find_snapshot(itr, idx, mm, data, head, old);
561 int auxtrace_record__options(struct auxtrace_record *itr,
562 struct perf_evlist *evlist,
563 struct record_opts *opts)
566 return itr->recording_options(itr, evlist, opts);
570 u64 auxtrace_record__reference(struct auxtrace_record *itr)
573 return itr->reference(itr);
577 int auxtrace_parse_snapshot_options(struct auxtrace_record *itr,
578 struct record_opts *opts, const char *str)
584 return itr->parse_snapshot_options(itr, opts, str);
586 pr_err("No AUX area tracing to snapshot\n");
590 struct auxtrace_record *__weak
591 auxtrace_record__init(struct perf_evlist *evlist __maybe_unused, int *err)
597 static int auxtrace_index__alloc(struct list_head *head)
599 struct auxtrace_index *auxtrace_index;
601 auxtrace_index = malloc(sizeof(struct auxtrace_index));
605 auxtrace_index->nr = 0;
606 INIT_LIST_HEAD(&auxtrace_index->list);
608 list_add_tail(&auxtrace_index->list, head);
613 void auxtrace_index__free(struct list_head *head)
615 struct auxtrace_index *auxtrace_index, *n;
617 list_for_each_entry_safe(auxtrace_index, n, head, list) {
618 list_del(&auxtrace_index->list);
619 free(auxtrace_index);
623 static struct auxtrace_index *auxtrace_index__last(struct list_head *head)
625 struct auxtrace_index *auxtrace_index;
628 if (list_empty(head)) {
629 err = auxtrace_index__alloc(head);
634 auxtrace_index = list_entry(head->prev, struct auxtrace_index, list);
636 if (auxtrace_index->nr >= PERF_AUXTRACE_INDEX_ENTRY_COUNT) {
637 err = auxtrace_index__alloc(head);
640 auxtrace_index = list_entry(head->prev, struct auxtrace_index,
644 return auxtrace_index;
647 int auxtrace_index__auxtrace_event(struct list_head *head,
648 union perf_event *event, off_t file_offset)
650 struct auxtrace_index *auxtrace_index;
653 auxtrace_index = auxtrace_index__last(head);
657 nr = auxtrace_index->nr;
658 auxtrace_index->entries[nr].file_offset = file_offset;
659 auxtrace_index->entries[nr].sz = event->header.size;
660 auxtrace_index->nr += 1;
665 static int auxtrace_index__do_write(int fd,
666 struct auxtrace_index *auxtrace_index)
668 struct auxtrace_index_entry ent;
671 for (i = 0; i < auxtrace_index->nr; i++) {
672 ent.file_offset = auxtrace_index->entries[i].file_offset;
673 ent.sz = auxtrace_index->entries[i].sz;
674 if (writen(fd, &ent, sizeof(ent)) != sizeof(ent))
680 int auxtrace_index__write(int fd, struct list_head *head)
682 struct auxtrace_index *auxtrace_index;
686 list_for_each_entry(auxtrace_index, head, list)
687 total += auxtrace_index->nr;
689 if (writen(fd, &total, sizeof(total)) != sizeof(total))
692 list_for_each_entry(auxtrace_index, head, list) {
693 err = auxtrace_index__do_write(fd, auxtrace_index);
701 static int auxtrace_index__process_entry(int fd, struct list_head *head,
704 struct auxtrace_index *auxtrace_index;
705 struct auxtrace_index_entry ent;
708 if (readn(fd, &ent, sizeof(ent)) != sizeof(ent))
711 auxtrace_index = auxtrace_index__last(head);
715 nr = auxtrace_index->nr;
717 auxtrace_index->entries[nr].file_offset =
718 bswap_64(ent.file_offset);
719 auxtrace_index->entries[nr].sz = bswap_64(ent.sz);
721 auxtrace_index->entries[nr].file_offset = ent.file_offset;
722 auxtrace_index->entries[nr].sz = ent.sz;
725 auxtrace_index->nr = nr + 1;
730 int auxtrace_index__process(int fd, u64 size, struct perf_session *session,
733 struct list_head *head = &session->auxtrace_index;
736 if (readn(fd, &nr, sizeof(u64)) != sizeof(u64))
742 if (sizeof(u64) + nr * sizeof(struct auxtrace_index_entry) > size)
748 err = auxtrace_index__process_entry(fd, head, needs_swap);
756 static int auxtrace_queues__process_index_entry(struct auxtrace_queues *queues,
757 struct perf_session *session,
758 struct auxtrace_index_entry *ent)
760 return auxtrace_queues__add_indexed_event(queues, session,
761 ent->file_offset, ent->sz);
764 int auxtrace_queues__process_index(struct auxtrace_queues *queues,
765 struct perf_session *session)
767 struct auxtrace_index *auxtrace_index;
768 struct auxtrace_index_entry *ent;
772 if (auxtrace__dont_decode(session))
775 list_for_each_entry(auxtrace_index, &session->auxtrace_index, list) {
776 for (i = 0; i < auxtrace_index->nr; i++) {
777 ent = &auxtrace_index->entries[i];
778 err = auxtrace_queues__process_index_entry(queues,
788 struct auxtrace_buffer *auxtrace_buffer__next(struct auxtrace_queue *queue,
789 struct auxtrace_buffer *buffer)
792 if (list_is_last(&buffer->list, &queue->head))
794 return list_entry(buffer->list.next, struct auxtrace_buffer,
797 if (list_empty(&queue->head))
799 return list_entry(queue->head.next, struct auxtrace_buffer,
804 void *auxtrace_buffer__get_data(struct auxtrace_buffer *buffer, int fd)
806 size_t adj = buffer->data_offset & (page_size - 1);
807 size_t size = buffer->size + adj;
808 off_t file_offset = buffer->data_offset - adj;
814 addr = mmap(NULL, size, PROT_READ, MAP_SHARED, fd, file_offset);
815 if (addr == MAP_FAILED)
818 buffer->mmap_addr = addr;
819 buffer->mmap_size = size;
821 buffer->data = addr + adj;
826 void auxtrace_buffer__put_data(struct auxtrace_buffer *buffer)
828 if (!buffer->data || !buffer->mmap_addr)
830 munmap(buffer->mmap_addr, buffer->mmap_size);
831 buffer->mmap_addr = NULL;
832 buffer->mmap_size = 0;
834 buffer->use_data = NULL;
837 void auxtrace_buffer__drop_data(struct auxtrace_buffer *buffer)
839 auxtrace_buffer__put_data(buffer);
840 if (buffer->data_needs_freeing) {
841 buffer->data_needs_freeing = false;
842 zfree(&buffer->data);
843 buffer->use_data = NULL;
848 void auxtrace_buffer__free(struct auxtrace_buffer *buffer)
850 auxtrace_buffer__drop_data(buffer);
854 void auxtrace_synth_error(struct auxtrace_error_event *auxtrace_error, int type,
855 int code, int cpu, pid_t pid, pid_t tid, u64 ip,
860 memset(auxtrace_error, 0, sizeof(struct auxtrace_error_event));
862 auxtrace_error->header.type = PERF_RECORD_AUXTRACE_ERROR;
863 auxtrace_error->type = type;
864 auxtrace_error->code = code;
865 auxtrace_error->cpu = cpu;
866 auxtrace_error->pid = pid;
867 auxtrace_error->tid = tid;
868 auxtrace_error->ip = ip;
869 strlcpy(auxtrace_error->msg, msg, MAX_AUXTRACE_ERROR_MSG);
871 size = (void *)auxtrace_error->msg - (void *)auxtrace_error +
872 strlen(auxtrace_error->msg) + 1;
873 auxtrace_error->header.size = PERF_ALIGN(size, sizeof(u64));
876 int perf_event__synthesize_auxtrace_info(struct auxtrace_record *itr,
877 struct perf_tool *tool,
878 struct perf_session *session,
879 perf_event__handler_t process)
881 union perf_event *ev;
885 pr_debug2("Synthesizing auxtrace information\n");
886 priv_size = auxtrace_record__info_priv_size(itr, session->evlist);
887 ev = zalloc(sizeof(struct auxtrace_info_event) + priv_size);
891 ev->auxtrace_info.header.type = PERF_RECORD_AUXTRACE_INFO;
892 ev->auxtrace_info.header.size = sizeof(struct auxtrace_info_event) +
894 err = auxtrace_record__info_fill(itr, session, &ev->auxtrace_info,
899 err = process(tool, ev, NULL, NULL);
905 int perf_event__process_auxtrace_info(struct perf_tool *tool __maybe_unused,
906 union perf_event *event,
907 struct perf_session *session)
909 enum auxtrace_type type = event->auxtrace_info.type;
912 fprintf(stdout, " type: %u\n", type);
915 case PERF_AUXTRACE_INTEL_PT:
916 return intel_pt_process_auxtrace_info(event, session);
917 case PERF_AUXTRACE_INTEL_BTS:
918 return intel_bts_process_auxtrace_info(event, session);
919 case PERF_AUXTRACE_ARM_SPE:
920 return arm_spe_process_auxtrace_info(event, session);
921 case PERF_AUXTRACE_CS_ETM:
922 return cs_etm__process_auxtrace_info(event, session);
923 case PERF_AUXTRACE_S390_CPUMSF:
924 return s390_cpumsf_process_auxtrace_info(event, session);
925 case PERF_AUXTRACE_UNKNOWN:
931 s64 perf_event__process_auxtrace(struct perf_tool *tool,
932 union perf_event *event,
933 struct perf_session *session)
938 fprintf(stdout, " size: %#"PRIx64" offset: %#"PRIx64" ref: %#"PRIx64" idx: %u tid: %d cpu: %d\n",
939 event->auxtrace.size, event->auxtrace.offset,
940 event->auxtrace.reference, event->auxtrace.idx,
941 event->auxtrace.tid, event->auxtrace.cpu);
943 if (auxtrace__dont_decode(session))
944 return event->auxtrace.size;
946 if (!session->auxtrace || event->header.type != PERF_RECORD_AUXTRACE)
949 err = session->auxtrace->process_auxtrace_event(session, event, tool);
953 return event->auxtrace.size;
956 #define PERF_ITRACE_DEFAULT_PERIOD_TYPE PERF_ITRACE_PERIOD_NANOSECS
957 #define PERF_ITRACE_DEFAULT_PERIOD 100000
958 #define PERF_ITRACE_DEFAULT_CALLCHAIN_SZ 16
959 #define PERF_ITRACE_MAX_CALLCHAIN_SZ 1024
960 #define PERF_ITRACE_DEFAULT_LAST_BRANCH_SZ 64
961 #define PERF_ITRACE_MAX_LAST_BRANCH_SZ 1024
963 void itrace_synth_opts__set_default(struct itrace_synth_opts *synth_opts)
965 synth_opts->instructions = true;
966 synth_opts->branches = true;
967 synth_opts->transactions = true;
968 synth_opts->ptwrites = true;
969 synth_opts->pwr_events = true;
970 synth_opts->errors = true;
971 synth_opts->period_type = PERF_ITRACE_DEFAULT_PERIOD_TYPE;
972 synth_opts->period = PERF_ITRACE_DEFAULT_PERIOD;
973 synth_opts->callchain_sz = PERF_ITRACE_DEFAULT_CALLCHAIN_SZ;
974 synth_opts->last_branch_sz = PERF_ITRACE_DEFAULT_LAST_BRANCH_SZ;
975 synth_opts->initial_skip = 0;
979 * Please check tools/perf/Documentation/perf-script.txt for information
980 * about the options parsed here, which is introduced after this cset,
981 * when support in 'perf script' for these options is introduced.
983 int itrace_parse_synth_opts(const struct option *opt, const char *str,
986 struct itrace_synth_opts *synth_opts = opt->value;
989 bool period_type_set = false;
990 bool period_set = false;
992 synth_opts->set = true;
995 synth_opts->dont_decode = true;
1000 itrace_synth_opts__set_default(synth_opts);
1004 for (p = str; *p;) {
1007 synth_opts->instructions = true;
1008 while (*p == ' ' || *p == ',')
1011 synth_opts->period = strtoull(p, &endptr, 10);
1014 while (*p == ' ' || *p == ',')
1018 synth_opts->period_type =
1019 PERF_ITRACE_PERIOD_INSTRUCTIONS;
1020 period_type_set = true;
1023 synth_opts->period_type =
1024 PERF_ITRACE_PERIOD_TICKS;
1025 period_type_set = true;
1028 synth_opts->period *= 1000;
1031 synth_opts->period *= 1000;
1036 synth_opts->period_type =
1037 PERF_ITRACE_PERIOD_NANOSECS;
1038 period_type_set = true;
1048 synth_opts->branches = true;
1051 synth_opts->transactions = true;
1054 synth_opts->ptwrites = true;
1057 synth_opts->pwr_events = true;
1060 synth_opts->errors = true;
1063 synth_opts->log = true;
1066 synth_opts->branches = true;
1067 synth_opts->calls = true;
1070 synth_opts->branches = true;
1071 synth_opts->returns = true;
1074 synth_opts->callchain = true;
1075 synth_opts->callchain_sz =
1076 PERF_ITRACE_DEFAULT_CALLCHAIN_SZ;
1077 while (*p == ' ' || *p == ',')
1082 val = strtoul(p, &endptr, 10);
1084 if (!val || val > PERF_ITRACE_MAX_CALLCHAIN_SZ)
1086 synth_opts->callchain_sz = val;
1090 synth_opts->last_branch = true;
1091 synth_opts->last_branch_sz =
1092 PERF_ITRACE_DEFAULT_LAST_BRANCH_SZ;
1093 while (*p == ' ' || *p == ',')
1098 val = strtoul(p, &endptr, 10);
1101 val > PERF_ITRACE_MAX_LAST_BRANCH_SZ)
1103 synth_opts->last_branch_sz = val;
1107 synth_opts->initial_skip = strtoul(p, &endptr, 10);
1120 if (synth_opts->instructions) {
1121 if (!period_type_set)
1122 synth_opts->period_type =
1123 PERF_ITRACE_DEFAULT_PERIOD_TYPE;
1125 synth_opts->period = PERF_ITRACE_DEFAULT_PERIOD;
1131 pr_err("Bad Instruction Tracing options '%s'\n", str);
1135 static const char * const auxtrace_error_type_name[] = {
1136 [PERF_AUXTRACE_ERROR_ITRACE] = "instruction trace",
1139 static const char *auxtrace_error_name(int type)
1141 const char *error_type_name = NULL;
1143 if (type < PERF_AUXTRACE_ERROR_MAX)
1144 error_type_name = auxtrace_error_type_name[type];
1145 if (!error_type_name)
1146 error_type_name = "unknown AUX";
1147 return error_type_name;
1150 size_t perf_event__fprintf_auxtrace_error(union perf_event *event, FILE *fp)
1152 struct auxtrace_error_event *e = &event->auxtrace_error;
1155 ret = fprintf(fp, " %s error type %u",
1156 auxtrace_error_name(e->type), e->type);
1157 ret += fprintf(fp, " cpu %d pid %d tid %d ip %#"PRIx64" code %u: %s\n",
1158 e->cpu, e->pid, e->tid, e->ip, e->code, e->msg);
1162 void perf_session__auxtrace_error_inc(struct perf_session *session,
1163 union perf_event *event)
1165 struct auxtrace_error_event *e = &event->auxtrace_error;
1167 if (e->type < PERF_AUXTRACE_ERROR_MAX)
1168 session->evlist->stats.nr_auxtrace_errors[e->type] += 1;
1171 void events_stats__auxtrace_error_warn(const struct events_stats *stats)
1175 for (i = 0; i < PERF_AUXTRACE_ERROR_MAX; i++) {
1176 if (!stats->nr_auxtrace_errors[i])
1178 ui__warning("%u %s errors\n",
1179 stats->nr_auxtrace_errors[i],
1180 auxtrace_error_name(i));
1184 int perf_event__process_auxtrace_error(struct perf_tool *tool __maybe_unused,
1185 union perf_event *event,
1186 struct perf_session *session)
1188 if (auxtrace__dont_decode(session))
1191 perf_event__fprintf_auxtrace_error(event, stdout);
1195 static int __auxtrace_mmap__read(struct auxtrace_mmap *mm,
1196 struct auxtrace_record *itr,
1197 struct perf_tool *tool, process_auxtrace_t fn,
1198 bool snapshot, size_t snapshot_size)
1200 u64 head, old = mm->prev, offset, ref;
1201 unsigned char *data = mm->base;
1202 size_t size, head_off, old_off, len1, len2, padding;
1203 union perf_event ev;
1204 void *data1, *data2;
1207 head = auxtrace_mmap__read_snapshot_head(mm);
1208 if (auxtrace_record__find_snapshot(itr, mm->idx, mm, data,
1212 head = auxtrace_mmap__read_head(mm);
1218 pr_debug3("auxtrace idx %d old %#"PRIx64" head %#"PRIx64" diff %#"PRIx64"\n",
1219 mm->idx, old, head, head - old);
1222 head_off = head & mm->mask;
1223 old_off = old & mm->mask;
1225 head_off = head % mm->len;
1226 old_off = old % mm->len;
1229 if (head_off > old_off)
1230 size = head_off - old_off;
1232 size = mm->len - (old_off - head_off);
1234 if (snapshot && size > snapshot_size)
1235 size = snapshot_size;
1237 ref = auxtrace_record__reference(itr);
1239 if (head > old || size <= head || mm->mask) {
1240 offset = head - size;
1243 * When the buffer size is not a power of 2, 'head' wraps at the
1244 * highest multiple of the buffer size, so we have to subtract
1245 * the remainder here.
1247 u64 rem = (0ULL - mm->len) % mm->len;
1249 offset = head - size - rem;
1252 if (size > head_off) {
1253 len1 = size - head_off;
1254 data1 = &data[mm->len - len1];
1259 data1 = &data[head_off - len1];
1264 if (itr->alignment) {
1265 unsigned int unwanted = len1 % itr->alignment;
1271 /* padding must be written by fn() e.g. record__process_auxtrace() */
1272 padding = size & (PERF_AUXTRACE_RECORD_ALIGNMENT - 1);
1274 padding = PERF_AUXTRACE_RECORD_ALIGNMENT - padding;
1276 memset(&ev, 0, sizeof(ev));
1277 ev.auxtrace.header.type = PERF_RECORD_AUXTRACE;
1278 ev.auxtrace.header.size = sizeof(ev.auxtrace);
1279 ev.auxtrace.size = size + padding;
1280 ev.auxtrace.offset = offset;
1281 ev.auxtrace.reference = ref;
1282 ev.auxtrace.idx = mm->idx;
1283 ev.auxtrace.tid = mm->tid;
1284 ev.auxtrace.cpu = mm->cpu;
1286 if (fn(tool, &ev, data1, len1, data2, len2))
1292 auxtrace_mmap__write_tail(mm, head);
1293 if (itr->read_finish) {
1296 err = itr->read_finish(itr, mm->idx);
1305 int auxtrace_mmap__read(struct auxtrace_mmap *mm, struct auxtrace_record *itr,
1306 struct perf_tool *tool, process_auxtrace_t fn)
1308 return __auxtrace_mmap__read(mm, itr, tool, fn, false, 0);
1311 int auxtrace_mmap__read_snapshot(struct auxtrace_mmap *mm,
1312 struct auxtrace_record *itr,
1313 struct perf_tool *tool, process_auxtrace_t fn,
1314 size_t snapshot_size)
1316 return __auxtrace_mmap__read(mm, itr, tool, fn, true, snapshot_size);
1320 * struct auxtrace_cache - hash table to implement a cache
1321 * @hashtable: the hashtable
1322 * @sz: hashtable size (number of hlists)
1323 * @entry_size: size of an entry
1324 * @limit: limit the number of entries to this maximum, when reached the cache
1325 * is dropped and caching begins again with an empty cache
1326 * @cnt: current number of entries
1327 * @bits: hashtable size (@sz = 2^@bits)
1329 struct auxtrace_cache {
1330 struct hlist_head *hashtable;
1338 struct auxtrace_cache *auxtrace_cache__new(unsigned int bits, size_t entry_size,
1339 unsigned int limit_percent)
1341 struct auxtrace_cache *c;
1342 struct hlist_head *ht;
1345 c = zalloc(sizeof(struct auxtrace_cache));
1351 ht = calloc(sz, sizeof(struct hlist_head));
1355 for (i = 0; i < sz; i++)
1356 INIT_HLIST_HEAD(&ht[i]);
1360 c->entry_size = entry_size;
1361 c->limit = (c->sz * limit_percent) / 100;
1371 static void auxtrace_cache__drop(struct auxtrace_cache *c)
1373 struct auxtrace_cache_entry *entry;
1374 struct hlist_node *tmp;
1380 for (i = 0; i < c->sz; i++) {
1381 hlist_for_each_entry_safe(entry, tmp, &c->hashtable[i], hash) {
1382 hlist_del(&entry->hash);
1383 auxtrace_cache__free_entry(c, entry);
1390 void auxtrace_cache__free(struct auxtrace_cache *c)
1395 auxtrace_cache__drop(c);
1400 void *auxtrace_cache__alloc_entry(struct auxtrace_cache *c)
1402 return malloc(c->entry_size);
1405 void auxtrace_cache__free_entry(struct auxtrace_cache *c __maybe_unused,
1411 int auxtrace_cache__add(struct auxtrace_cache *c, u32 key,
1412 struct auxtrace_cache_entry *entry)
1414 if (c->limit && ++c->cnt > c->limit)
1415 auxtrace_cache__drop(c);
1418 hlist_add_head(&entry->hash, &c->hashtable[hash_32(key, c->bits)]);
1423 void *auxtrace_cache__lookup(struct auxtrace_cache *c, u32 key)
1425 struct auxtrace_cache_entry *entry;
1426 struct hlist_head *hlist;
1431 hlist = &c->hashtable[hash_32(key, c->bits)];
1432 hlist_for_each_entry(entry, hlist, hash) {
1433 if (entry->key == key)
1440 static void addr_filter__free_str(struct addr_filter *filt)
1443 filt->action = NULL;
1444 filt->sym_from = NULL;
1445 filt->sym_to = NULL;
1446 filt->filename = NULL;
1450 static struct addr_filter *addr_filter__new(void)
1452 struct addr_filter *filt = zalloc(sizeof(*filt));
1455 INIT_LIST_HEAD(&filt->list);
1460 static void addr_filter__free(struct addr_filter *filt)
1463 addr_filter__free_str(filt);
1467 static void addr_filters__add(struct addr_filters *filts,
1468 struct addr_filter *filt)
1470 list_add_tail(&filt->list, &filts->head);
1474 static void addr_filters__del(struct addr_filters *filts,
1475 struct addr_filter *filt)
1477 list_del_init(&filt->list);
1481 void addr_filters__init(struct addr_filters *filts)
1483 INIT_LIST_HEAD(&filts->head);
1487 void addr_filters__exit(struct addr_filters *filts)
1489 struct addr_filter *filt, *n;
1491 list_for_each_entry_safe(filt, n, &filts->head, list) {
1492 addr_filters__del(filts, filt);
1493 addr_filter__free(filt);
1497 static int parse_num_or_str(char **inp, u64 *num, const char **str,
1498 const char *str_delim)
1500 *inp += strspn(*inp, " ");
1502 if (isdigit(**inp)) {
1508 *num = strtoull(*inp, &endptr, 0);
1519 *inp += strspn(*inp, " ");
1521 n = strcspn(*inp, str_delim);
1533 static int parse_action(struct addr_filter *filt)
1535 if (!strcmp(filt->action, "filter")) {
1538 } else if (!strcmp(filt->action, "start")) {
1540 } else if (!strcmp(filt->action, "stop")) {
1541 filt->start = false;
1542 } else if (!strcmp(filt->action, "tracestop")) {
1543 filt->start = false;
1545 filt->action += 5; /* Change 'tracestop' to 'stop' */
1552 static int parse_sym_idx(char **inp, int *idx)
1556 *inp += strspn(*inp, " ");
1563 if (**inp == 'g' || **inp == 'G') {
1571 num = strtoul(*inp, &endptr, 0);
1574 if (endptr == *inp || num > INT_MAX)
1583 static int parse_addr_size(char **inp, u64 *num, const char **str, int *idx)
1585 int err = parse_num_or_str(inp, num, str, " ");
1588 err = parse_sym_idx(inp, idx);
1593 static int parse_one_filter(struct addr_filter *filt, const char **filter_inp)
1598 filt->str = fstr = strdup(*filter_inp);
1602 err = parse_num_or_str(&fstr, NULL, &filt->action, " ");
1606 err = parse_action(filt);
1610 err = parse_addr_size(&fstr, &filt->addr, &filt->sym_from,
1611 &filt->sym_from_idx);
1615 fstr += strspn(fstr, " ");
1619 err = parse_addr_size(&fstr, &filt->size, &filt->sym_to,
1626 fstr += strspn(fstr, " ");
1630 err = parse_num_or_str(&fstr, NULL, &filt->filename, " ,");
1635 fstr += strspn(fstr, " ,");
1637 *filter_inp += fstr - filt->str;
1642 addr_filter__free_str(filt);
1647 int addr_filters__parse_bare_filter(struct addr_filters *filts,
1650 struct addr_filter *filt;
1651 const char *fstr = filter;
1655 filt = addr_filter__new();
1656 err = parse_one_filter(filt, &fstr);
1658 addr_filter__free(filt);
1659 addr_filters__exit(filts);
1662 addr_filters__add(filts, filt);
1681 static bool kern_sym_match(struct sym_args *args, const char *name, char type)
1683 /* A function with the same name, and global or the n'th found or any */
1684 return kallsyms__is_function(type) &&
1685 !strcmp(name, args->name) &&
1686 ((args->global && isupper(type)) ||
1687 (args->selected && ++(args->cnt) == args->idx) ||
1688 (!args->global && !args->selected));
1691 static int find_kern_sym_cb(void *arg, const char *name, char type, u64 start)
1693 struct sym_args *args = arg;
1695 if (args->started) {
1697 args->size = start - args->start;
1698 if (args->selected) {
1701 } else if (kern_sym_match(args, name, type)) {
1702 args->duplicate = true;
1705 } else if (kern_sym_match(args, name, type)) {
1706 args->started = true;
1707 args->start = start;
1713 static int print_kern_sym_cb(void *arg, const char *name, char type, u64 start)
1715 struct sym_args *args = arg;
1717 if (kern_sym_match(args, name, type)) {
1718 pr_err("#%d\t0x%"PRIx64"\t%c\t%s\n",
1719 ++args->cnt, start, type, name);
1721 } else if (args->near) {
1723 pr_err("\t\twhich is near\t\t%s\n", name);
1729 static int sym_not_found_error(const char *sym_name, int idx)
1732 pr_err("N'th occurrence (N=%d) of symbol '%s' not found.\n",
1735 pr_err("Global symbol '%s' not found.\n", sym_name);
1737 pr_err("Symbol '%s' not found.\n", sym_name);
1739 pr_err("Note that symbols must be functions.\n");
1744 static int find_kern_sym(const char *sym_name, u64 *start, u64 *size, int idx)
1746 struct sym_args args = {
1750 .selected = idx > 0,
1757 err = kallsyms__parse("/proc/kallsyms", &args, find_kern_sym_cb);
1759 pr_err("Failed to parse /proc/kallsyms\n");
1763 if (args.duplicate) {
1764 pr_err("Multiple kernel symbols with name '%s'\n", sym_name);
1766 kallsyms__parse("/proc/kallsyms", &args, print_kern_sym_cb);
1767 pr_err("Disambiguate symbol name by inserting #n after the name e.g. %s #2\n",
1769 pr_err("Or select a global symbol by inserting #0 or #g or #G\n");
1773 if (!args.started) {
1774 pr_err("Kernel symbol lookup: ");
1775 return sym_not_found_error(sym_name, idx);
1778 *start = args.start;
1784 static int find_entire_kern_cb(void *arg, const char *name __maybe_unused,
1785 char type, u64 start)
1787 struct sym_args *args = arg;
1789 if (!kallsyms__is_function(type))
1792 if (!args->started) {
1793 args->started = true;
1794 args->start = start;
1796 /* Don't know exactly where the kernel ends, so we add a page */
1797 args->size = round_up(start, page_size) + page_size - args->start;
1802 static int addr_filter__entire_kernel(struct addr_filter *filt)
1804 struct sym_args args = { .started = false };
1807 err = kallsyms__parse("/proc/kallsyms", &args, find_entire_kern_cb);
1808 if (err < 0 || !args.started) {
1809 pr_err("Failed to parse /proc/kallsyms\n");
1813 filt->addr = args.start;
1814 filt->size = args.size;
1819 static int check_end_after_start(struct addr_filter *filt, u64 start, u64 size)
1821 if (start + size >= filt->addr)
1824 if (filt->sym_from) {
1825 pr_err("Symbol '%s' (0x%"PRIx64") comes before '%s' (0x%"PRIx64")\n",
1826 filt->sym_to, start, filt->sym_from, filt->addr);
1828 pr_err("Symbol '%s' (0x%"PRIx64") comes before address 0x%"PRIx64")\n",
1829 filt->sym_to, start, filt->addr);
1835 static int addr_filter__resolve_kernel_syms(struct addr_filter *filt)
1837 bool no_size = false;
1841 if (symbol_conf.kptr_restrict) {
1842 pr_err("Kernel addresses are restricted. Unable to resolve kernel symbols.\n");
1846 if (filt->sym_from && !strcmp(filt->sym_from, "*"))
1847 return addr_filter__entire_kernel(filt);
1849 if (filt->sym_from) {
1850 err = find_kern_sym(filt->sym_from, &start, &size,
1851 filt->sym_from_idx);
1855 if (filt->range && !filt->size && !filt->sym_to) {
1862 err = find_kern_sym(filt->sym_to, &start, &size,
1867 err = check_end_after_start(filt, start, size);
1870 filt->size = start + size - filt->addr;
1874 /* The very last symbol in kallsyms does not imply a particular size */
1876 pr_err("Cannot determine size of symbol '%s'\n",
1877 filt->sym_to ? filt->sym_to : filt->sym_from);
1884 static struct dso *load_dso(const char *name)
1889 map = dso__new_map(name);
1895 dso = dso__get(map->dso);
1902 static bool dso_sym_match(struct symbol *sym, const char *name, int *cnt,
1905 /* Same name, and global or the n'th found or any */
1906 return !arch__compare_symbol_names(name, sym->name) &&
1907 ((!idx && sym->binding == STB_GLOBAL) ||
1908 (idx > 0 && ++*cnt == idx) ||
1912 static void print_duplicate_syms(struct dso *dso, const char *sym_name)
1918 pr_err("Multiple symbols with name '%s'\n", sym_name);
1920 sym = dso__first_symbol(dso);
1922 if (dso_sym_match(sym, sym_name, &cnt, -1)) {
1923 pr_err("#%d\t0x%"PRIx64"\t%c\t%s\n",
1925 sym->binding == STB_GLOBAL ? 'g' :
1926 sym->binding == STB_LOCAL ? 'l' : 'w',
1931 pr_err("\t\twhich is near\t\t%s\n", sym->name);
1933 sym = dso__next_symbol(sym);
1936 pr_err("Disambiguate symbol name by inserting #n after the name e.g. %s #2\n",
1938 pr_err("Or select a global symbol by inserting #0 or #g or #G\n");
1941 static int find_dso_sym(struct dso *dso, const char *sym_name, u64 *start,
1950 sym = dso__first_symbol(dso);
1954 *size = sym->start - *start;
1958 } else if (dso_sym_match(sym, sym_name, &cnt, idx)) {
1959 print_duplicate_syms(dso, sym_name);
1962 } else if (dso_sym_match(sym, sym_name, &cnt, idx)) {
1963 *start = sym->start;
1964 *size = sym->end - sym->start;
1966 sym = dso__next_symbol(sym);
1970 return sym_not_found_error(sym_name, idx);
1975 static int addr_filter__entire_dso(struct addr_filter *filt, struct dso *dso)
1977 struct symbol *first_sym = dso__first_symbol(dso);
1978 struct symbol *last_sym = dso__last_symbol(dso);
1980 if (!first_sym || !last_sym) {
1981 pr_err("Failed to determine filter for %s\nNo symbols found.\n",
1986 filt->addr = first_sym->start;
1987 filt->size = last_sym->end - first_sym->start;
1992 static int addr_filter__resolve_syms(struct addr_filter *filt)
1998 if (!filt->sym_from && !filt->sym_to)
2001 if (!filt->filename)
2002 return addr_filter__resolve_kernel_syms(filt);
2004 dso = load_dso(filt->filename);
2006 pr_err("Failed to load symbols from: %s\n", filt->filename);
2010 if (filt->sym_from && !strcmp(filt->sym_from, "*")) {
2011 err = addr_filter__entire_dso(filt, dso);
2015 if (filt->sym_from) {
2016 err = find_dso_sym(dso, filt->sym_from, &start, &size,
2017 filt->sym_from_idx);
2021 if (filt->range && !filt->size && !filt->sym_to)
2026 err = find_dso_sym(dso, filt->sym_to, &start, &size,
2031 err = check_end_after_start(filt, start, size);
2035 filt->size = start + size - filt->addr;
2044 static char *addr_filter__to_str(struct addr_filter *filt)
2046 char filename_buf[PATH_MAX];
2047 const char *at = "";
2048 const char *fn = "";
2052 if (filt->filename) {
2054 fn = realpath(filt->filename, filename_buf);
2060 err = asprintf(&filter, "%s 0x%"PRIx64"/0x%"PRIx64"%s%s",
2061 filt->action, filt->addr, filt->size, at, fn);
2063 err = asprintf(&filter, "%s 0x%"PRIx64"%s%s",
2064 filt->action, filt->addr, at, fn);
2067 return err < 0 ? NULL : filter;
2070 static int parse_addr_filter(struct perf_evsel *evsel, const char *filter,
2073 struct addr_filters filts;
2074 struct addr_filter *filt;
2077 addr_filters__init(&filts);
2079 err = addr_filters__parse_bare_filter(&filts, filter);
2083 if (filts.cnt > max_nr) {
2084 pr_err("Error: number of address filters (%d) exceeds maximum (%d)\n",
2090 list_for_each_entry(filt, &filts.head, list) {
2093 err = addr_filter__resolve_syms(filt);
2097 new_filter = addr_filter__to_str(filt);
2103 if (perf_evsel__append_addr_filter(evsel, new_filter)) {
2110 addr_filters__exit(&filts);
2113 pr_err("Failed to parse address filter: '%s'\n", filter);
2114 pr_err("Filter format is: filter|start|stop|tracestop <start symbol or address> [/ <end symbol or size>] [@<file name>]\n");
2115 pr_err("Where multiple filters are separated by space or comma.\n");
2121 static struct perf_pmu *perf_evsel__find_pmu(struct perf_evsel *evsel)
2123 struct perf_pmu *pmu = NULL;
2125 while ((pmu = perf_pmu__scan(pmu)) != NULL) {
2126 if (pmu->type == evsel->attr.type)
2133 static int perf_evsel__nr_addr_filter(struct perf_evsel *evsel)
2135 struct perf_pmu *pmu = perf_evsel__find_pmu(evsel);
2136 int nr_addr_filters = 0;
2141 perf_pmu__scan_file(pmu, "nr_addr_filters", "%d", &nr_addr_filters);
2143 return nr_addr_filters;
2146 int auxtrace_parse_filters(struct perf_evlist *evlist)
2148 struct perf_evsel *evsel;
2152 evlist__for_each_entry(evlist, evsel) {
2153 filter = evsel->filter;
2154 max_nr = perf_evsel__nr_addr_filter(evsel);
2155 if (!filter || !max_nr)
2157 evsel->filter = NULL;
2158 err = parse_addr_filter(evsel, filter, max_nr);
2162 pr_debug("Address filter: %s\n", evsel->filter);