Loading...
1#define _FILE_OFFSET_BITS 64
2
3#include <linux/kernel.h>
4
5#include <byteswap.h>
6#include <unistd.h>
7#include <sys/types.h>
8#include <sys/mman.h>
9
10#include "evlist.h"
11#include "evsel.h"
12#include "session.h"
13#include "sort.h"
14#include "util.h"
15#include "cpumap.h"
16
17static int perf_session__open(struct perf_session *self, bool force)
18{
19 struct stat input_stat;
20
21 if (!strcmp(self->filename, "-")) {
22 self->fd_pipe = true;
23 self->fd = STDIN_FILENO;
24
25 if (perf_session__read_header(self, self->fd) < 0)
26 pr_err("incompatible file format");
27
28 return 0;
29 }
30
31 self->fd = open(self->filename, O_RDONLY);
32 if (self->fd < 0) {
33 int err = errno;
34
35 pr_err("failed to open %s: %s", self->filename, strerror(err));
36 if (err == ENOENT && !strcmp(self->filename, "perf.data"))
37 pr_err(" (try 'perf record' first)");
38 pr_err("\n");
39 return -errno;
40 }
41
42 if (fstat(self->fd, &input_stat) < 0)
43 goto out_close;
44
45 if (!force && input_stat.st_uid && (input_stat.st_uid != geteuid())) {
46 pr_err("file %s not owned by current user or root\n",
47 self->filename);
48 goto out_close;
49 }
50
51 if (!input_stat.st_size) {
52 pr_info("zero-sized file (%s), nothing to do!\n",
53 self->filename);
54 goto out_close;
55 }
56
57 if (perf_session__read_header(self, self->fd) < 0) {
58 pr_err("incompatible file format");
59 goto out_close;
60 }
61
62 if (!perf_evlist__valid_sample_type(self->evlist)) {
63 pr_err("non matching sample_type");
64 goto out_close;
65 }
66
67 if (!perf_evlist__valid_sample_id_all(self->evlist)) {
68 pr_err("non matching sample_id_all");
69 goto out_close;
70 }
71
72 self->size = input_stat.st_size;
73 return 0;
74
75out_close:
76 close(self->fd);
77 self->fd = -1;
78 return -1;
79}
80
81static void perf_session__id_header_size(struct perf_session *session)
82{
83 struct perf_sample *data;
84 u64 sample_type = session->sample_type;
85 u16 size = 0;
86
87 if (!session->sample_id_all)
88 goto out;
89
90 if (sample_type & PERF_SAMPLE_TID)
91 size += sizeof(data->tid) * 2;
92
93 if (sample_type & PERF_SAMPLE_TIME)
94 size += sizeof(data->time);
95
96 if (sample_type & PERF_SAMPLE_ID)
97 size += sizeof(data->id);
98
99 if (sample_type & PERF_SAMPLE_STREAM_ID)
100 size += sizeof(data->stream_id);
101
102 if (sample_type & PERF_SAMPLE_CPU)
103 size += sizeof(data->cpu) * 2;
104out:
105 session->id_hdr_size = size;
106}
107
108void perf_session__update_sample_type(struct perf_session *self)
109{
110 self->sample_type = perf_evlist__sample_type(self->evlist);
111 self->sample_size = __perf_evsel__sample_size(self->sample_type);
112 self->sample_id_all = perf_evlist__sample_id_all(self->evlist);
113 perf_session__id_header_size(self);
114}
115
116int perf_session__create_kernel_maps(struct perf_session *self)
117{
118 int ret = machine__create_kernel_maps(&self->host_machine);
119
120 if (ret >= 0)
121 ret = machines__create_guest_kernel_maps(&self->machines);
122 return ret;
123}
124
125static void perf_session__destroy_kernel_maps(struct perf_session *self)
126{
127 machine__destroy_kernel_maps(&self->host_machine);
128 machines__destroy_guest_kernel_maps(&self->machines);
129}
130
131struct perf_session *perf_session__new(const char *filename, int mode,
132 bool force, bool repipe,
133 struct perf_event_ops *ops)
134{
135 size_t len = filename ? strlen(filename) + 1 : 0;
136 struct perf_session *self = zalloc(sizeof(*self) + len);
137
138 if (self == NULL)
139 goto out;
140
141 memcpy(self->filename, filename, len);
142 self->threads = RB_ROOT;
143 INIT_LIST_HEAD(&self->dead_threads);
144 self->last_match = NULL;
145 /*
146 * On 64bit we can mmap the data file in one go. No need for tiny mmap
147 * slices. On 32bit we use 32MB.
148 */
149#if BITS_PER_LONG == 64
150 self->mmap_window = ULLONG_MAX;
151#else
152 self->mmap_window = 32 * 1024 * 1024ULL;
153#endif
154 self->machines = RB_ROOT;
155 self->repipe = repipe;
156 INIT_LIST_HEAD(&self->ordered_samples.samples);
157 INIT_LIST_HEAD(&self->ordered_samples.sample_cache);
158 INIT_LIST_HEAD(&self->ordered_samples.to_free);
159 machine__init(&self->host_machine, "", HOST_KERNEL_ID);
160
161 if (mode == O_RDONLY) {
162 if (perf_session__open(self, force) < 0)
163 goto out_delete;
164 perf_session__update_sample_type(self);
165 } else if (mode == O_WRONLY) {
166 /*
167 * In O_RDONLY mode this will be performed when reading the
168 * kernel MMAP event, in perf_event__process_mmap().
169 */
170 if (perf_session__create_kernel_maps(self) < 0)
171 goto out_delete;
172 }
173
174 if (ops && ops->ordering_requires_timestamps &&
175 ops->ordered_samples && !self->sample_id_all) {
176 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n");
177 ops->ordered_samples = false;
178 }
179
180out:
181 return self;
182out_delete:
183 perf_session__delete(self);
184 return NULL;
185}
186
187static void perf_session__delete_dead_threads(struct perf_session *self)
188{
189 struct thread *n, *t;
190
191 list_for_each_entry_safe(t, n, &self->dead_threads, node) {
192 list_del(&t->node);
193 thread__delete(t);
194 }
195}
196
197static void perf_session__delete_threads(struct perf_session *self)
198{
199 struct rb_node *nd = rb_first(&self->threads);
200
201 while (nd) {
202 struct thread *t = rb_entry(nd, struct thread, rb_node);
203
204 rb_erase(&t->rb_node, &self->threads);
205 nd = rb_next(nd);
206 thread__delete(t);
207 }
208}
209
210void perf_session__delete(struct perf_session *self)
211{
212 perf_session__destroy_kernel_maps(self);
213 perf_session__delete_dead_threads(self);
214 perf_session__delete_threads(self);
215 machine__exit(&self->host_machine);
216 close(self->fd);
217 free(self);
218}
219
220void perf_session__remove_thread(struct perf_session *self, struct thread *th)
221{
222 self->last_match = NULL;
223 rb_erase(&th->rb_node, &self->threads);
224 /*
225 * We may have references to this thread, for instance in some hist_entry
226 * instances, so just move them to a separate list.
227 */
228 list_add_tail(&th->node, &self->dead_threads);
229}
230
231static bool symbol__match_parent_regex(struct symbol *sym)
232{
233 if (sym->name && !regexec(&parent_regex, sym->name, 0, NULL, 0))
234 return 1;
235
236 return 0;
237}
238
239int perf_session__resolve_callchain(struct perf_session *self,
240 struct thread *thread,
241 struct ip_callchain *chain,
242 struct symbol **parent)
243{
244 u8 cpumode = PERF_RECORD_MISC_USER;
245 unsigned int i;
246 int err;
247
248 callchain_cursor_reset(&self->callchain_cursor);
249
250 for (i = 0; i < chain->nr; i++) {
251 u64 ip;
252 struct addr_location al;
253
254 if (callchain_param.order == ORDER_CALLEE)
255 ip = chain->ips[i];
256 else
257 ip = chain->ips[chain->nr - i - 1];
258
259 if (ip >= PERF_CONTEXT_MAX) {
260 switch (ip) {
261 case PERF_CONTEXT_HV:
262 cpumode = PERF_RECORD_MISC_HYPERVISOR; break;
263 case PERF_CONTEXT_KERNEL:
264 cpumode = PERF_RECORD_MISC_KERNEL; break;
265 case PERF_CONTEXT_USER:
266 cpumode = PERF_RECORD_MISC_USER; break;
267 default:
268 break;
269 }
270 continue;
271 }
272
273 al.filtered = false;
274 thread__find_addr_location(thread, self, cpumode,
275 MAP__FUNCTION, thread->pid, ip, &al, NULL);
276 if (al.sym != NULL) {
277 if (sort__has_parent && !*parent &&
278 symbol__match_parent_regex(al.sym))
279 *parent = al.sym;
280 if (!symbol_conf.use_callchain)
281 break;
282 }
283
284 err = callchain_cursor_append(&self->callchain_cursor,
285 ip, al.map, al.sym);
286 if (err)
287 return err;
288 }
289
290 return 0;
291}
292
293static int process_event_synth_stub(union perf_event *event __used,
294 struct perf_session *session __used)
295{
296 dump_printf(": unhandled!\n");
297 return 0;
298}
299
300static int process_event_sample_stub(union perf_event *event __used,
301 struct perf_sample *sample __used,
302 struct perf_evsel *evsel __used,
303 struct perf_session *session __used)
304{
305 dump_printf(": unhandled!\n");
306 return 0;
307}
308
309static int process_event_stub(union perf_event *event __used,
310 struct perf_sample *sample __used,
311 struct perf_session *session __used)
312{
313 dump_printf(": unhandled!\n");
314 return 0;
315}
316
317static int process_finished_round_stub(union perf_event *event __used,
318 struct perf_session *session __used,
319 struct perf_event_ops *ops __used)
320{
321 dump_printf(": unhandled!\n");
322 return 0;
323}
324
325static int process_finished_round(union perf_event *event,
326 struct perf_session *session,
327 struct perf_event_ops *ops);
328
329static void perf_event_ops__fill_defaults(struct perf_event_ops *handler)
330{
331 if (handler->sample == NULL)
332 handler->sample = process_event_sample_stub;
333 if (handler->mmap == NULL)
334 handler->mmap = process_event_stub;
335 if (handler->comm == NULL)
336 handler->comm = process_event_stub;
337 if (handler->fork == NULL)
338 handler->fork = process_event_stub;
339 if (handler->exit == NULL)
340 handler->exit = process_event_stub;
341 if (handler->lost == NULL)
342 handler->lost = perf_event__process_lost;
343 if (handler->read == NULL)
344 handler->read = process_event_stub;
345 if (handler->throttle == NULL)
346 handler->throttle = process_event_stub;
347 if (handler->unthrottle == NULL)
348 handler->unthrottle = process_event_stub;
349 if (handler->attr == NULL)
350 handler->attr = process_event_synth_stub;
351 if (handler->event_type == NULL)
352 handler->event_type = process_event_synth_stub;
353 if (handler->tracing_data == NULL)
354 handler->tracing_data = process_event_synth_stub;
355 if (handler->build_id == NULL)
356 handler->build_id = process_event_synth_stub;
357 if (handler->finished_round == NULL) {
358 if (handler->ordered_samples)
359 handler->finished_round = process_finished_round;
360 else
361 handler->finished_round = process_finished_round_stub;
362 }
363}
364
365void mem_bswap_64(void *src, int byte_size)
366{
367 u64 *m = src;
368
369 while (byte_size > 0) {
370 *m = bswap_64(*m);
371 byte_size -= sizeof(u64);
372 ++m;
373 }
374}
375
376static void perf_event__all64_swap(union perf_event *event)
377{
378 struct perf_event_header *hdr = &event->header;
379 mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr));
380}
381
382static void perf_event__comm_swap(union perf_event *event)
383{
384 event->comm.pid = bswap_32(event->comm.pid);
385 event->comm.tid = bswap_32(event->comm.tid);
386}
387
388static void perf_event__mmap_swap(union perf_event *event)
389{
390 event->mmap.pid = bswap_32(event->mmap.pid);
391 event->mmap.tid = bswap_32(event->mmap.tid);
392 event->mmap.start = bswap_64(event->mmap.start);
393 event->mmap.len = bswap_64(event->mmap.len);
394 event->mmap.pgoff = bswap_64(event->mmap.pgoff);
395}
396
397static void perf_event__task_swap(union perf_event *event)
398{
399 event->fork.pid = bswap_32(event->fork.pid);
400 event->fork.tid = bswap_32(event->fork.tid);
401 event->fork.ppid = bswap_32(event->fork.ppid);
402 event->fork.ptid = bswap_32(event->fork.ptid);
403 event->fork.time = bswap_64(event->fork.time);
404}
405
406static void perf_event__read_swap(union perf_event *event)
407{
408 event->read.pid = bswap_32(event->read.pid);
409 event->read.tid = bswap_32(event->read.tid);
410 event->read.value = bswap_64(event->read.value);
411 event->read.time_enabled = bswap_64(event->read.time_enabled);
412 event->read.time_running = bswap_64(event->read.time_running);
413 event->read.id = bswap_64(event->read.id);
414}
415
416/* exported for swapping attributes in file header */
417void perf_event__attr_swap(struct perf_event_attr *attr)
418{
419 attr->type = bswap_32(attr->type);
420 attr->size = bswap_32(attr->size);
421 attr->config = bswap_64(attr->config);
422 attr->sample_period = bswap_64(attr->sample_period);
423 attr->sample_type = bswap_64(attr->sample_type);
424 attr->read_format = bswap_64(attr->read_format);
425 attr->wakeup_events = bswap_32(attr->wakeup_events);
426 attr->bp_type = bswap_32(attr->bp_type);
427 attr->bp_addr = bswap_64(attr->bp_addr);
428 attr->bp_len = bswap_64(attr->bp_len);
429}
430
431static void perf_event__hdr_attr_swap(union perf_event *event)
432{
433 size_t size;
434
435 perf_event__attr_swap(&event->attr.attr);
436
437 size = event->header.size;
438 size -= (void *)&event->attr.id - (void *)event;
439 mem_bswap_64(event->attr.id, size);
440}
441
442static void perf_event__event_type_swap(union perf_event *event)
443{
444 event->event_type.event_type.event_id =
445 bswap_64(event->event_type.event_type.event_id);
446}
447
448static void perf_event__tracing_data_swap(union perf_event *event)
449{
450 event->tracing_data.size = bswap_32(event->tracing_data.size);
451}
452
453typedef void (*perf_event__swap_op)(union perf_event *event);
454
455static perf_event__swap_op perf_event__swap_ops[] = {
456 [PERF_RECORD_MMAP] = perf_event__mmap_swap,
457 [PERF_RECORD_COMM] = perf_event__comm_swap,
458 [PERF_RECORD_FORK] = perf_event__task_swap,
459 [PERF_RECORD_EXIT] = perf_event__task_swap,
460 [PERF_RECORD_LOST] = perf_event__all64_swap,
461 [PERF_RECORD_READ] = perf_event__read_swap,
462 [PERF_RECORD_SAMPLE] = perf_event__all64_swap,
463 [PERF_RECORD_HEADER_ATTR] = perf_event__hdr_attr_swap,
464 [PERF_RECORD_HEADER_EVENT_TYPE] = perf_event__event_type_swap,
465 [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
466 [PERF_RECORD_HEADER_BUILD_ID] = NULL,
467 [PERF_RECORD_HEADER_MAX] = NULL,
468};
469
470struct sample_queue {
471 u64 timestamp;
472 u64 file_offset;
473 union perf_event *event;
474 struct list_head list;
475};
476
477static void perf_session_free_sample_buffers(struct perf_session *session)
478{
479 struct ordered_samples *os = &session->ordered_samples;
480
481 while (!list_empty(&os->to_free)) {
482 struct sample_queue *sq;
483
484 sq = list_entry(os->to_free.next, struct sample_queue, list);
485 list_del(&sq->list);
486 free(sq);
487 }
488}
489
490static int perf_session_deliver_event(struct perf_session *session,
491 union perf_event *event,
492 struct perf_sample *sample,
493 struct perf_event_ops *ops,
494 u64 file_offset);
495
496static void flush_sample_queue(struct perf_session *s,
497 struct perf_event_ops *ops)
498{
499 struct ordered_samples *os = &s->ordered_samples;
500 struct list_head *head = &os->samples;
501 struct sample_queue *tmp, *iter;
502 struct perf_sample sample;
503 u64 limit = os->next_flush;
504 u64 last_ts = os->last_sample ? os->last_sample->timestamp : 0ULL;
505 int ret;
506
507 if (!ops->ordered_samples || !limit)
508 return;
509
510 list_for_each_entry_safe(iter, tmp, head, list) {
511 if (iter->timestamp > limit)
512 break;
513
514 ret = perf_session__parse_sample(s, iter->event, &sample);
515 if (ret)
516 pr_err("Can't parse sample, err = %d\n", ret);
517 else
518 perf_session_deliver_event(s, iter->event, &sample, ops,
519 iter->file_offset);
520
521 os->last_flush = iter->timestamp;
522 list_del(&iter->list);
523 list_add(&iter->list, &os->sample_cache);
524 }
525
526 if (list_empty(head)) {
527 os->last_sample = NULL;
528 } else if (last_ts <= limit) {
529 os->last_sample =
530 list_entry(head->prev, struct sample_queue, list);
531 }
532}
533
534/*
535 * When perf record finishes a pass on every buffers, it records this pseudo
536 * event.
537 * We record the max timestamp t found in the pass n.
538 * Assuming these timestamps are monotonic across cpus, we know that if
539 * a buffer still has events with timestamps below t, they will be all
540 * available and then read in the pass n + 1.
541 * Hence when we start to read the pass n + 2, we can safely flush every
542 * events with timestamps below t.
543 *
544 * ============ PASS n =================
545 * CPU 0 | CPU 1
546 * |
547 * cnt1 timestamps | cnt2 timestamps
548 * 1 | 2
549 * 2 | 3
550 * - | 4 <--- max recorded
551 *
552 * ============ PASS n + 1 ==============
553 * CPU 0 | CPU 1
554 * |
555 * cnt1 timestamps | cnt2 timestamps
556 * 3 | 5
557 * 4 | 6
558 * 5 | 7 <---- max recorded
559 *
560 * Flush every events below timestamp 4
561 *
562 * ============ PASS n + 2 ==============
563 * CPU 0 | CPU 1
564 * |
565 * cnt1 timestamps | cnt2 timestamps
566 * 6 | 8
567 * 7 | 9
568 * - | 10
569 *
570 * Flush every events below timestamp 7
571 * etc...
572 */
573static int process_finished_round(union perf_event *event __used,
574 struct perf_session *session,
575 struct perf_event_ops *ops)
576{
577 flush_sample_queue(session, ops);
578 session->ordered_samples.next_flush = session->ordered_samples.max_timestamp;
579
580 return 0;
581}
582
583/* The queue is ordered by time */
584static void __queue_event(struct sample_queue *new, struct perf_session *s)
585{
586 struct ordered_samples *os = &s->ordered_samples;
587 struct sample_queue *sample = os->last_sample;
588 u64 timestamp = new->timestamp;
589 struct list_head *p;
590
591 os->last_sample = new;
592
593 if (!sample) {
594 list_add(&new->list, &os->samples);
595 os->max_timestamp = timestamp;
596 return;
597 }
598
599 /*
600 * last_sample might point to some random place in the list as it's
601 * the last queued event. We expect that the new event is close to
602 * this.
603 */
604 if (sample->timestamp <= timestamp) {
605 while (sample->timestamp <= timestamp) {
606 p = sample->list.next;
607 if (p == &os->samples) {
608 list_add_tail(&new->list, &os->samples);
609 os->max_timestamp = timestamp;
610 return;
611 }
612 sample = list_entry(p, struct sample_queue, list);
613 }
614 list_add_tail(&new->list, &sample->list);
615 } else {
616 while (sample->timestamp > timestamp) {
617 p = sample->list.prev;
618 if (p == &os->samples) {
619 list_add(&new->list, &os->samples);
620 return;
621 }
622 sample = list_entry(p, struct sample_queue, list);
623 }
624 list_add(&new->list, &sample->list);
625 }
626}
627
628#define MAX_SAMPLE_BUFFER (64 * 1024 / sizeof(struct sample_queue))
629
630static int perf_session_queue_event(struct perf_session *s, union perf_event *event,
631 struct perf_sample *sample, u64 file_offset)
632{
633 struct ordered_samples *os = &s->ordered_samples;
634 struct list_head *sc = &os->sample_cache;
635 u64 timestamp = sample->time;
636 struct sample_queue *new;
637
638 if (!timestamp || timestamp == ~0ULL)
639 return -ETIME;
640
641 if (timestamp < s->ordered_samples.last_flush) {
642 printf("Warning: Timestamp below last timeslice flush\n");
643 return -EINVAL;
644 }
645
646 if (!list_empty(sc)) {
647 new = list_entry(sc->next, struct sample_queue, list);
648 list_del(&new->list);
649 } else if (os->sample_buffer) {
650 new = os->sample_buffer + os->sample_buffer_idx;
651 if (++os->sample_buffer_idx == MAX_SAMPLE_BUFFER)
652 os->sample_buffer = NULL;
653 } else {
654 os->sample_buffer = malloc(MAX_SAMPLE_BUFFER * sizeof(*new));
655 if (!os->sample_buffer)
656 return -ENOMEM;
657 list_add(&os->sample_buffer->list, &os->to_free);
658 os->sample_buffer_idx = 2;
659 new = os->sample_buffer + 1;
660 }
661
662 new->timestamp = timestamp;
663 new->file_offset = file_offset;
664 new->event = event;
665
666 __queue_event(new, s);
667
668 return 0;
669}
670
671static void callchain__printf(struct perf_sample *sample)
672{
673 unsigned int i;
674
675 printf("... chain: nr:%" PRIu64 "\n", sample->callchain->nr);
676
677 for (i = 0; i < sample->callchain->nr; i++)
678 printf("..... %2d: %016" PRIx64 "\n",
679 i, sample->callchain->ips[i]);
680}
681
682static void perf_session__print_tstamp(struct perf_session *session,
683 union perf_event *event,
684 struct perf_sample *sample)
685{
686 if (event->header.type != PERF_RECORD_SAMPLE &&
687 !session->sample_id_all) {
688 fputs("-1 -1 ", stdout);
689 return;
690 }
691
692 if ((session->sample_type & PERF_SAMPLE_CPU))
693 printf("%u ", sample->cpu);
694
695 if (session->sample_type & PERF_SAMPLE_TIME)
696 printf("%" PRIu64 " ", sample->time);
697}
698
699static void dump_event(struct perf_session *session, union perf_event *event,
700 u64 file_offset, struct perf_sample *sample)
701{
702 if (!dump_trace)
703 return;
704
705 printf("\n%#" PRIx64 " [%#x]: event: %d\n",
706 file_offset, event->header.size, event->header.type);
707
708 trace_event(event);
709
710 if (sample)
711 perf_session__print_tstamp(session, event, sample);
712
713 printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset,
714 event->header.size, perf_event__name(event->header.type));
715}
716
717static void dump_sample(struct perf_session *session, union perf_event *event,
718 struct perf_sample *sample)
719{
720 if (!dump_trace)
721 return;
722
723 printf("(IP, %d): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n",
724 event->header.misc, sample->pid, sample->tid, sample->ip,
725 sample->period, sample->addr);
726
727 if (session->sample_type & PERF_SAMPLE_CALLCHAIN)
728 callchain__printf(sample);
729}
730
731static int perf_session_deliver_event(struct perf_session *session,
732 union perf_event *event,
733 struct perf_sample *sample,
734 struct perf_event_ops *ops,
735 u64 file_offset)
736{
737 struct perf_evsel *evsel;
738
739 dump_event(session, event, file_offset, sample);
740
741 switch (event->header.type) {
742 case PERF_RECORD_SAMPLE:
743 dump_sample(session, event, sample);
744 evsel = perf_evlist__id2evsel(session->evlist, sample->id);
745 if (evsel == NULL) {
746 ++session->hists.stats.nr_unknown_id;
747 return -1;
748 }
749 return ops->sample(event, sample, evsel, session);
750 case PERF_RECORD_MMAP:
751 return ops->mmap(event, sample, session);
752 case PERF_RECORD_COMM:
753 return ops->comm(event, sample, session);
754 case PERF_RECORD_FORK:
755 return ops->fork(event, sample, session);
756 case PERF_RECORD_EXIT:
757 return ops->exit(event, sample, session);
758 case PERF_RECORD_LOST:
759 return ops->lost(event, sample, session);
760 case PERF_RECORD_READ:
761 return ops->read(event, sample, session);
762 case PERF_RECORD_THROTTLE:
763 return ops->throttle(event, sample, session);
764 case PERF_RECORD_UNTHROTTLE:
765 return ops->unthrottle(event, sample, session);
766 default:
767 ++session->hists.stats.nr_unknown_events;
768 return -1;
769 }
770}
771
772static int perf_session__preprocess_sample(struct perf_session *session,
773 union perf_event *event, struct perf_sample *sample)
774{
775 if (event->header.type != PERF_RECORD_SAMPLE ||
776 !(session->sample_type & PERF_SAMPLE_CALLCHAIN))
777 return 0;
778
779 if (!ip_callchain__valid(sample->callchain, event)) {
780 pr_debug("call-chain problem with event, skipping it.\n");
781 ++session->hists.stats.nr_invalid_chains;
782 session->hists.stats.total_invalid_chains += sample->period;
783 return -EINVAL;
784 }
785 return 0;
786}
787
788static int perf_session__process_user_event(struct perf_session *session, union perf_event *event,
789 struct perf_event_ops *ops, u64 file_offset)
790{
791 dump_event(session, event, file_offset, NULL);
792
793 /* These events are processed right away */
794 switch (event->header.type) {
795 case PERF_RECORD_HEADER_ATTR:
796 return ops->attr(event, session);
797 case PERF_RECORD_HEADER_EVENT_TYPE:
798 return ops->event_type(event, session);
799 case PERF_RECORD_HEADER_TRACING_DATA:
800 /* setup for reading amidst mmap */
801 lseek(session->fd, file_offset, SEEK_SET);
802 return ops->tracing_data(event, session);
803 case PERF_RECORD_HEADER_BUILD_ID:
804 return ops->build_id(event, session);
805 case PERF_RECORD_FINISHED_ROUND:
806 return ops->finished_round(event, session, ops);
807 default:
808 return -EINVAL;
809 }
810}
811
812static int perf_session__process_event(struct perf_session *session,
813 union perf_event *event,
814 struct perf_event_ops *ops,
815 u64 file_offset)
816{
817 struct perf_sample sample;
818 int ret;
819
820 if (session->header.needs_swap &&
821 perf_event__swap_ops[event->header.type])
822 perf_event__swap_ops[event->header.type](event);
823
824 if (event->header.type >= PERF_RECORD_HEADER_MAX)
825 return -EINVAL;
826
827 hists__inc_nr_events(&session->hists, event->header.type);
828
829 if (event->header.type >= PERF_RECORD_USER_TYPE_START)
830 return perf_session__process_user_event(session, event, ops, file_offset);
831
832 /*
833 * For all kernel events we get the sample data
834 */
835 ret = perf_session__parse_sample(session, event, &sample);
836 if (ret)
837 return ret;
838
839 /* Preprocess sample records - precheck callchains */
840 if (perf_session__preprocess_sample(session, event, &sample))
841 return 0;
842
843 if (ops->ordered_samples) {
844 ret = perf_session_queue_event(session, event, &sample,
845 file_offset);
846 if (ret != -ETIME)
847 return ret;
848 }
849
850 return perf_session_deliver_event(session, event, &sample, ops,
851 file_offset);
852}
853
854void perf_event_header__bswap(struct perf_event_header *self)
855{
856 self->type = bswap_32(self->type);
857 self->misc = bswap_16(self->misc);
858 self->size = bswap_16(self->size);
859}
860
861static struct thread *perf_session__register_idle_thread(struct perf_session *self)
862{
863 struct thread *thread = perf_session__findnew(self, 0);
864
865 if (thread == NULL || thread__set_comm(thread, "swapper")) {
866 pr_err("problem inserting idle task.\n");
867 thread = NULL;
868 }
869
870 return thread;
871}
872
873static void perf_session__warn_about_errors(const struct perf_session *session,
874 const struct perf_event_ops *ops)
875{
876 if (ops->lost == perf_event__process_lost &&
877 session->hists.stats.total_lost != 0) {
878 ui__warning("Processed %" PRIu64 " events and LOST %" PRIu64
879 "!\n\nCheck IO/CPU overload!\n\n",
880 session->hists.stats.total_period,
881 session->hists.stats.total_lost);
882 }
883
884 if (session->hists.stats.nr_unknown_events != 0) {
885 ui__warning("Found %u unknown events!\n\n"
886 "Is this an older tool processing a perf.data "
887 "file generated by a more recent tool?\n\n"
888 "If that is not the case, consider "
889 "reporting to linux-kernel@vger.kernel.org.\n\n",
890 session->hists.stats.nr_unknown_events);
891 }
892
893 if (session->hists.stats.nr_unknown_id != 0) {
894 ui__warning("%u samples with id not present in the header\n",
895 session->hists.stats.nr_unknown_id);
896 }
897
898 if (session->hists.stats.nr_invalid_chains != 0) {
899 ui__warning("Found invalid callchains!\n\n"
900 "%u out of %u events were discarded for this reason.\n\n"
901 "Consider reporting to linux-kernel@vger.kernel.org.\n\n",
902 session->hists.stats.nr_invalid_chains,
903 session->hists.stats.nr_events[PERF_RECORD_SAMPLE]);
904 }
905}
906
907#define session_done() (*(volatile int *)(&session_done))
908volatile int session_done;
909
910static int __perf_session__process_pipe_events(struct perf_session *self,
911 struct perf_event_ops *ops)
912{
913 union perf_event event;
914 uint32_t size;
915 int skip = 0;
916 u64 head;
917 int err;
918 void *p;
919
920 perf_event_ops__fill_defaults(ops);
921
922 head = 0;
923more:
924 err = readn(self->fd, &event, sizeof(struct perf_event_header));
925 if (err <= 0) {
926 if (err == 0)
927 goto done;
928
929 pr_err("failed to read event header\n");
930 goto out_err;
931 }
932
933 if (self->header.needs_swap)
934 perf_event_header__bswap(&event.header);
935
936 size = event.header.size;
937 if (size == 0)
938 size = 8;
939
940 p = &event;
941 p += sizeof(struct perf_event_header);
942
943 if (size - sizeof(struct perf_event_header)) {
944 err = readn(self->fd, p, size - sizeof(struct perf_event_header));
945 if (err <= 0) {
946 if (err == 0) {
947 pr_err("unexpected end of event stream\n");
948 goto done;
949 }
950
951 pr_err("failed to read event data\n");
952 goto out_err;
953 }
954 }
955
956 if (size == 0 ||
957 (skip = perf_session__process_event(self, &event, ops, head)) < 0) {
958 dump_printf("%#" PRIx64 " [%#x]: skipping unknown header type: %d\n",
959 head, event.header.size, event.header.type);
960 /*
961 * assume we lost track of the stream, check alignment, and
962 * increment a single u64 in the hope to catch on again 'soon'.
963 */
964 if (unlikely(head & 7))
965 head &= ~7ULL;
966
967 size = 8;
968 }
969
970 head += size;
971
972 if (skip > 0)
973 head += skip;
974
975 if (!session_done())
976 goto more;
977done:
978 err = 0;
979out_err:
980 perf_session__warn_about_errors(self, ops);
981 perf_session_free_sample_buffers(self);
982 return err;
983}
984
985static union perf_event *
986fetch_mmaped_event(struct perf_session *session,
987 u64 head, size_t mmap_size, char *buf)
988{
989 union perf_event *event;
990
991 /*
992 * Ensure we have enough space remaining to read
993 * the size of the event in the headers.
994 */
995 if (head + sizeof(event->header) > mmap_size)
996 return NULL;
997
998 event = (union perf_event *)(buf + head);
999
1000 if (session->header.needs_swap)
1001 perf_event_header__bswap(&event->header);
1002
1003 if (head + event->header.size > mmap_size)
1004 return NULL;
1005
1006 return event;
1007}
1008
1009int __perf_session__process_events(struct perf_session *session,
1010 u64 data_offset, u64 data_size,
1011 u64 file_size, struct perf_event_ops *ops)
1012{
1013 u64 head, page_offset, file_offset, file_pos, progress_next;
1014 int err, mmap_prot, mmap_flags, map_idx = 0;
1015 struct ui_progress *progress;
1016 size_t page_size, mmap_size;
1017 char *buf, *mmaps[8];
1018 union perf_event *event;
1019 uint32_t size;
1020
1021 perf_event_ops__fill_defaults(ops);
1022
1023 page_size = sysconf(_SC_PAGESIZE);
1024
1025 page_offset = page_size * (data_offset / page_size);
1026 file_offset = page_offset;
1027 head = data_offset - page_offset;
1028
1029 if (data_offset + data_size < file_size)
1030 file_size = data_offset + data_size;
1031
1032 progress_next = file_size / 16;
1033 progress = ui_progress__new("Processing events...", file_size);
1034 if (progress == NULL)
1035 return -1;
1036
1037 mmap_size = session->mmap_window;
1038 if (mmap_size > file_size)
1039 mmap_size = file_size;
1040
1041 memset(mmaps, 0, sizeof(mmaps));
1042
1043 mmap_prot = PROT_READ;
1044 mmap_flags = MAP_SHARED;
1045
1046 if (session->header.needs_swap) {
1047 mmap_prot |= PROT_WRITE;
1048 mmap_flags = MAP_PRIVATE;
1049 }
1050remap:
1051 buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, session->fd,
1052 file_offset);
1053 if (buf == MAP_FAILED) {
1054 pr_err("failed to mmap file\n");
1055 err = -errno;
1056 goto out_err;
1057 }
1058 mmaps[map_idx] = buf;
1059 map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1);
1060 file_pos = file_offset + head;
1061
1062more:
1063 event = fetch_mmaped_event(session, head, mmap_size, buf);
1064 if (!event) {
1065 if (mmaps[map_idx]) {
1066 munmap(mmaps[map_idx], mmap_size);
1067 mmaps[map_idx] = NULL;
1068 }
1069
1070 page_offset = page_size * (head / page_size);
1071 file_offset += page_offset;
1072 head -= page_offset;
1073 goto remap;
1074 }
1075
1076 size = event->header.size;
1077
1078 if (size == 0 ||
1079 perf_session__process_event(session, event, ops, file_pos) < 0) {
1080 dump_printf("%#" PRIx64 " [%#x]: skipping unknown header type: %d\n",
1081 file_offset + head, event->header.size,
1082 event->header.type);
1083 /*
1084 * assume we lost track of the stream, check alignment, and
1085 * increment a single u64 in the hope to catch on again 'soon'.
1086 */
1087 if (unlikely(head & 7))
1088 head &= ~7ULL;
1089
1090 size = 8;
1091 }
1092
1093 head += size;
1094 file_pos += size;
1095
1096 if (file_pos >= progress_next) {
1097 progress_next += file_size / 16;
1098 ui_progress__update(progress, file_pos);
1099 }
1100
1101 if (file_pos < file_size)
1102 goto more;
1103
1104 err = 0;
1105 /* do the final flush for ordered samples */
1106 session->ordered_samples.next_flush = ULLONG_MAX;
1107 flush_sample_queue(session, ops);
1108out_err:
1109 ui_progress__delete(progress);
1110 perf_session__warn_about_errors(session, ops);
1111 perf_session_free_sample_buffers(session);
1112 return err;
1113}
1114
1115int perf_session__process_events(struct perf_session *self,
1116 struct perf_event_ops *ops)
1117{
1118 int err;
1119
1120 if (perf_session__register_idle_thread(self) == NULL)
1121 return -ENOMEM;
1122
1123 if (!self->fd_pipe)
1124 err = __perf_session__process_events(self,
1125 self->header.data_offset,
1126 self->header.data_size,
1127 self->size, ops);
1128 else
1129 err = __perf_session__process_pipe_events(self, ops);
1130
1131 return err;
1132}
1133
1134bool perf_session__has_traces(struct perf_session *self, const char *msg)
1135{
1136 if (!(self->sample_type & PERF_SAMPLE_RAW)) {
1137 pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg);
1138 return false;
1139 }
1140
1141 return true;
1142}
1143
1144int perf_session__set_kallsyms_ref_reloc_sym(struct map **maps,
1145 const char *symbol_name,
1146 u64 addr)
1147{
1148 char *bracket;
1149 enum map_type i;
1150 struct ref_reloc_sym *ref;
1151
1152 ref = zalloc(sizeof(struct ref_reloc_sym));
1153 if (ref == NULL)
1154 return -ENOMEM;
1155
1156 ref->name = strdup(symbol_name);
1157 if (ref->name == NULL) {
1158 free(ref);
1159 return -ENOMEM;
1160 }
1161
1162 bracket = strchr(ref->name, ']');
1163 if (bracket)
1164 *bracket = '\0';
1165
1166 ref->addr = addr;
1167
1168 for (i = 0; i < MAP__NR_TYPES; ++i) {
1169 struct kmap *kmap = map__kmap(maps[i]);
1170 kmap->ref_reloc_sym = ref;
1171 }
1172
1173 return 0;
1174}
1175
1176size_t perf_session__fprintf_dsos(struct perf_session *self, FILE *fp)
1177{
1178 return __dsos__fprintf(&self->host_machine.kernel_dsos, fp) +
1179 __dsos__fprintf(&self->host_machine.user_dsos, fp) +
1180 machines__fprintf_dsos(&self->machines, fp);
1181}
1182
1183size_t perf_session__fprintf_dsos_buildid(struct perf_session *self, FILE *fp,
1184 bool with_hits)
1185{
1186 size_t ret = machine__fprintf_dsos_buildid(&self->host_machine, fp, with_hits);
1187 return ret + machines__fprintf_dsos_buildid(&self->machines, fp, with_hits);
1188}
1189
1190size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp)
1191{
1192 struct perf_evsel *pos;
1193 size_t ret = fprintf(fp, "Aggregated stats:\n");
1194
1195 ret += hists__fprintf_nr_events(&session->hists, fp);
1196
1197 list_for_each_entry(pos, &session->evlist->entries, node) {
1198 ret += fprintf(fp, "%s stats:\n", event_name(pos));
1199 ret += hists__fprintf_nr_events(&pos->hists, fp);
1200 }
1201
1202 return ret;
1203}
1204
1205struct perf_evsel *perf_session__find_first_evtype(struct perf_session *session,
1206 unsigned int type)
1207{
1208 struct perf_evsel *pos;
1209
1210 list_for_each_entry(pos, &session->evlist->entries, node) {
1211 if (pos->attr.type == type)
1212 return pos;
1213 }
1214 return NULL;
1215}
1216
1217void perf_session__print_ip(union perf_event *event,
1218 struct perf_sample *sample,
1219 struct perf_session *session,
1220 int print_sym, int print_dso)
1221{
1222 struct addr_location al;
1223 const char *symname, *dsoname;
1224 struct callchain_cursor *cursor = &session->callchain_cursor;
1225 struct callchain_cursor_node *node;
1226
1227 if (perf_event__preprocess_sample(event, session, &al, sample,
1228 NULL) < 0) {
1229 error("problem processing %d event, skipping it.\n",
1230 event->header.type);
1231 return;
1232 }
1233
1234 if (symbol_conf.use_callchain && sample->callchain) {
1235
1236 if (perf_session__resolve_callchain(session, al.thread,
1237 sample->callchain, NULL) != 0) {
1238 if (verbose)
1239 error("Failed to resolve callchain. Skipping\n");
1240 return;
1241 }
1242 callchain_cursor_commit(cursor);
1243
1244 while (1) {
1245 node = callchain_cursor_current(cursor);
1246 if (!node)
1247 break;
1248
1249 printf("\t%16" PRIx64, node->ip);
1250 if (print_sym) {
1251 if (node->sym && node->sym->name)
1252 symname = node->sym->name;
1253 else
1254 symname = "";
1255
1256 printf(" %s", symname);
1257 }
1258 if (print_dso) {
1259 if (node->map && node->map->dso && node->map->dso->name)
1260 dsoname = node->map->dso->name;
1261 else
1262 dsoname = "";
1263
1264 printf(" (%s)", dsoname);
1265 }
1266 printf("\n");
1267
1268 callchain_cursor_advance(cursor);
1269 }
1270
1271 } else {
1272 printf("%16" PRIx64, sample->ip);
1273 if (print_sym) {
1274 if (al.sym && al.sym->name)
1275 symname = al.sym->name;
1276 else
1277 symname = "";
1278
1279 printf(" %s", symname);
1280 }
1281
1282 if (print_dso) {
1283 if (al.map && al.map->dso && al.map->dso->name)
1284 dsoname = al.map->dso->name;
1285 else
1286 dsoname = "";
1287
1288 printf(" (%s)", dsoname);
1289 }
1290 }
1291}
1292
1293int perf_session__cpu_bitmap(struct perf_session *session,
1294 const char *cpu_list, unsigned long *cpu_bitmap)
1295{
1296 int i;
1297 struct cpu_map *map;
1298
1299 for (i = 0; i < PERF_TYPE_MAX; ++i) {
1300 struct perf_evsel *evsel;
1301
1302 evsel = perf_session__find_first_evtype(session, i);
1303 if (!evsel)
1304 continue;
1305
1306 if (!(evsel->attr.sample_type & PERF_SAMPLE_CPU)) {
1307 pr_err("File does not contain CPU events. "
1308 "Remove -c option to proceed.\n");
1309 return -1;
1310 }
1311 }
1312
1313 map = cpu_map__new(cpu_list);
1314
1315 for (i = 0; i < map->nr; i++) {
1316 int cpu = map->map[i];
1317
1318 if (cpu >= MAX_NR_CPUS) {
1319 pr_err("Requested CPU %d too large. "
1320 "Consider raising MAX_NR_CPUS\n", cpu);
1321 return -1;
1322 }
1323
1324 set_bit(cpu, cpu_bitmap);
1325 }
1326
1327 return 0;
1328}
1// SPDX-License-Identifier: GPL-2.0
2#include <errno.h>
3#include <inttypes.h>
4#include <linux/kernel.h>
5#include <traceevent/event-parse.h>
6#include <api/fs/fs.h>
7
8#include <byteswap.h>
9#include <unistd.h>
10#include <sys/types.h>
11#include <sys/mman.h>
12
13#include "evlist.h"
14#include "evsel.h"
15#include "memswap.h"
16#include "session.h"
17#include "tool.h"
18#include "sort.h"
19#include "util.h"
20#include "cpumap.h"
21#include "perf_regs.h"
22#include "asm/bug.h"
23#include "auxtrace.h"
24#include "thread.h"
25#include "thread-stack.h"
26#include "stat.h"
27
28static int perf_session__deliver_event(struct perf_session *session,
29 union perf_event *event,
30 struct perf_tool *tool,
31 u64 file_offset);
32
33static int perf_session__open(struct perf_session *session)
34{
35 struct perf_data *data = session->data;
36
37 if (perf_session__read_header(session) < 0) {
38 pr_err("incompatible file format (rerun with -v to learn more)\n");
39 return -1;
40 }
41
42 if (perf_data__is_pipe(data))
43 return 0;
44
45 if (perf_header__has_feat(&session->header, HEADER_STAT))
46 return 0;
47
48 if (!perf_evlist__valid_sample_type(session->evlist)) {
49 pr_err("non matching sample_type\n");
50 return -1;
51 }
52
53 if (!perf_evlist__valid_sample_id_all(session->evlist)) {
54 pr_err("non matching sample_id_all\n");
55 return -1;
56 }
57
58 if (!perf_evlist__valid_read_format(session->evlist)) {
59 pr_err("non matching read_format\n");
60 return -1;
61 }
62
63 return 0;
64}
65
66void perf_session__set_id_hdr_size(struct perf_session *session)
67{
68 u16 id_hdr_size = perf_evlist__id_hdr_size(session->evlist);
69
70 machines__set_id_hdr_size(&session->machines, id_hdr_size);
71}
72
73int perf_session__create_kernel_maps(struct perf_session *session)
74{
75 int ret = machine__create_kernel_maps(&session->machines.host);
76
77 if (ret >= 0)
78 ret = machines__create_guest_kernel_maps(&session->machines);
79 return ret;
80}
81
82static void perf_session__destroy_kernel_maps(struct perf_session *session)
83{
84 machines__destroy_kernel_maps(&session->machines);
85}
86
87static bool perf_session__has_comm_exec(struct perf_session *session)
88{
89 struct perf_evsel *evsel;
90
91 evlist__for_each_entry(session->evlist, evsel) {
92 if (evsel->attr.comm_exec)
93 return true;
94 }
95
96 return false;
97}
98
99static void perf_session__set_comm_exec(struct perf_session *session)
100{
101 bool comm_exec = perf_session__has_comm_exec(session);
102
103 machines__set_comm_exec(&session->machines, comm_exec);
104}
105
106static int ordered_events__deliver_event(struct ordered_events *oe,
107 struct ordered_event *event)
108{
109 struct perf_session *session = container_of(oe, struct perf_session,
110 ordered_events);
111
112 return perf_session__deliver_event(session, event->event,
113 session->tool, event->file_offset);
114}
115
116struct perf_session *perf_session__new(struct perf_data *data,
117 bool repipe, struct perf_tool *tool)
118{
119 struct perf_session *session = zalloc(sizeof(*session));
120
121 if (!session)
122 goto out;
123
124 session->repipe = repipe;
125 session->tool = tool;
126 INIT_LIST_HEAD(&session->auxtrace_index);
127 machines__init(&session->machines);
128 ordered_events__init(&session->ordered_events, ordered_events__deliver_event);
129
130 if (data) {
131 if (perf_data__open(data))
132 goto out_delete;
133
134 session->data = data;
135
136 if (perf_data__is_read(data)) {
137 if (perf_session__open(session) < 0)
138 goto out_close;
139
140 /*
141 * set session attributes that are present in perf.data
142 * but not in pipe-mode.
143 */
144 if (!data->is_pipe) {
145 perf_session__set_id_hdr_size(session);
146 perf_session__set_comm_exec(session);
147 }
148 }
149 } else {
150 session->machines.host.env = &perf_env;
151 }
152
153 if (!data || perf_data__is_write(data)) {
154 /*
155 * In O_RDONLY mode this will be performed when reading the
156 * kernel MMAP event, in perf_event__process_mmap().
157 */
158 if (perf_session__create_kernel_maps(session) < 0)
159 pr_warning("Cannot read kernel map\n");
160 }
161
162 /*
163 * In pipe-mode, evlist is empty until PERF_RECORD_HEADER_ATTR is
164 * processed, so perf_evlist__sample_id_all is not meaningful here.
165 */
166 if ((!data || !data->is_pipe) && tool && tool->ordering_requires_timestamps &&
167 tool->ordered_events && !perf_evlist__sample_id_all(session->evlist)) {
168 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n");
169 tool->ordered_events = false;
170 }
171
172 return session;
173
174 out_close:
175 perf_data__close(data);
176 out_delete:
177 perf_session__delete(session);
178 out:
179 return NULL;
180}
181
182static void perf_session__delete_threads(struct perf_session *session)
183{
184 machine__delete_threads(&session->machines.host);
185}
186
187void perf_session__delete(struct perf_session *session)
188{
189 if (session == NULL)
190 return;
191 auxtrace__free(session);
192 auxtrace_index__free(&session->auxtrace_index);
193 perf_session__destroy_kernel_maps(session);
194 perf_session__delete_threads(session);
195 perf_env__exit(&session->header.env);
196 machines__exit(&session->machines);
197 if (session->data)
198 perf_data__close(session->data);
199 free(session);
200}
201
202static int process_event_synth_tracing_data_stub(struct perf_tool *tool
203 __maybe_unused,
204 union perf_event *event
205 __maybe_unused,
206 struct perf_session *session
207 __maybe_unused)
208{
209 dump_printf(": unhandled!\n");
210 return 0;
211}
212
213static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused,
214 union perf_event *event __maybe_unused,
215 struct perf_evlist **pevlist
216 __maybe_unused)
217{
218 dump_printf(": unhandled!\n");
219 return 0;
220}
221
222static int process_event_synth_event_update_stub(struct perf_tool *tool __maybe_unused,
223 union perf_event *event __maybe_unused,
224 struct perf_evlist **pevlist
225 __maybe_unused)
226{
227 if (dump_trace)
228 perf_event__fprintf_event_update(event, stdout);
229
230 dump_printf(": unhandled!\n");
231 return 0;
232}
233
234static int process_event_sample_stub(struct perf_tool *tool __maybe_unused,
235 union perf_event *event __maybe_unused,
236 struct perf_sample *sample __maybe_unused,
237 struct perf_evsel *evsel __maybe_unused,
238 struct machine *machine __maybe_unused)
239{
240 dump_printf(": unhandled!\n");
241 return 0;
242}
243
244static int process_event_stub(struct perf_tool *tool __maybe_unused,
245 union perf_event *event __maybe_unused,
246 struct perf_sample *sample __maybe_unused,
247 struct machine *machine __maybe_unused)
248{
249 dump_printf(": unhandled!\n");
250 return 0;
251}
252
253static int process_finished_round_stub(struct perf_tool *tool __maybe_unused,
254 union perf_event *event __maybe_unused,
255 struct ordered_events *oe __maybe_unused)
256{
257 dump_printf(": unhandled!\n");
258 return 0;
259}
260
261static int process_finished_round(struct perf_tool *tool,
262 union perf_event *event,
263 struct ordered_events *oe);
264
265static int skipn(int fd, off_t n)
266{
267 char buf[4096];
268 ssize_t ret;
269
270 while (n > 0) {
271 ret = read(fd, buf, min(n, (off_t)sizeof(buf)));
272 if (ret <= 0)
273 return ret;
274 n -= ret;
275 }
276
277 return 0;
278}
279
280static s64 process_event_auxtrace_stub(struct perf_tool *tool __maybe_unused,
281 union perf_event *event,
282 struct perf_session *session
283 __maybe_unused)
284{
285 dump_printf(": unhandled!\n");
286 if (perf_data__is_pipe(session->data))
287 skipn(perf_data__fd(session->data), event->auxtrace.size);
288 return event->auxtrace.size;
289}
290
291static int process_event_op2_stub(struct perf_tool *tool __maybe_unused,
292 union perf_event *event __maybe_unused,
293 struct perf_session *session __maybe_unused)
294{
295 dump_printf(": unhandled!\n");
296 return 0;
297}
298
299
300static
301int process_event_thread_map_stub(struct perf_tool *tool __maybe_unused,
302 union perf_event *event __maybe_unused,
303 struct perf_session *session __maybe_unused)
304{
305 if (dump_trace)
306 perf_event__fprintf_thread_map(event, stdout);
307
308 dump_printf(": unhandled!\n");
309 return 0;
310}
311
312static
313int process_event_cpu_map_stub(struct perf_tool *tool __maybe_unused,
314 union perf_event *event __maybe_unused,
315 struct perf_session *session __maybe_unused)
316{
317 if (dump_trace)
318 perf_event__fprintf_cpu_map(event, stdout);
319
320 dump_printf(": unhandled!\n");
321 return 0;
322}
323
324static
325int process_event_stat_config_stub(struct perf_tool *tool __maybe_unused,
326 union perf_event *event __maybe_unused,
327 struct perf_session *session __maybe_unused)
328{
329 if (dump_trace)
330 perf_event__fprintf_stat_config(event, stdout);
331
332 dump_printf(": unhandled!\n");
333 return 0;
334}
335
336static int process_stat_stub(struct perf_tool *tool __maybe_unused,
337 union perf_event *event __maybe_unused,
338 struct perf_session *perf_session
339 __maybe_unused)
340{
341 if (dump_trace)
342 perf_event__fprintf_stat(event, stdout);
343
344 dump_printf(": unhandled!\n");
345 return 0;
346}
347
348static int process_stat_round_stub(struct perf_tool *tool __maybe_unused,
349 union perf_event *event __maybe_unused,
350 struct perf_session *perf_session
351 __maybe_unused)
352{
353 if (dump_trace)
354 perf_event__fprintf_stat_round(event, stdout);
355
356 dump_printf(": unhandled!\n");
357 return 0;
358}
359
360void perf_tool__fill_defaults(struct perf_tool *tool)
361{
362 if (tool->sample == NULL)
363 tool->sample = process_event_sample_stub;
364 if (tool->mmap == NULL)
365 tool->mmap = process_event_stub;
366 if (tool->mmap2 == NULL)
367 tool->mmap2 = process_event_stub;
368 if (tool->comm == NULL)
369 tool->comm = process_event_stub;
370 if (tool->namespaces == NULL)
371 tool->namespaces = process_event_stub;
372 if (tool->fork == NULL)
373 tool->fork = process_event_stub;
374 if (tool->exit == NULL)
375 tool->exit = process_event_stub;
376 if (tool->lost == NULL)
377 tool->lost = perf_event__process_lost;
378 if (tool->lost_samples == NULL)
379 tool->lost_samples = perf_event__process_lost_samples;
380 if (tool->aux == NULL)
381 tool->aux = perf_event__process_aux;
382 if (tool->itrace_start == NULL)
383 tool->itrace_start = perf_event__process_itrace_start;
384 if (tool->context_switch == NULL)
385 tool->context_switch = perf_event__process_switch;
386 if (tool->read == NULL)
387 tool->read = process_event_sample_stub;
388 if (tool->throttle == NULL)
389 tool->throttle = process_event_stub;
390 if (tool->unthrottle == NULL)
391 tool->unthrottle = process_event_stub;
392 if (tool->attr == NULL)
393 tool->attr = process_event_synth_attr_stub;
394 if (tool->event_update == NULL)
395 tool->event_update = process_event_synth_event_update_stub;
396 if (tool->tracing_data == NULL)
397 tool->tracing_data = process_event_synth_tracing_data_stub;
398 if (tool->build_id == NULL)
399 tool->build_id = process_event_op2_stub;
400 if (tool->finished_round == NULL) {
401 if (tool->ordered_events)
402 tool->finished_round = process_finished_round;
403 else
404 tool->finished_round = process_finished_round_stub;
405 }
406 if (tool->id_index == NULL)
407 tool->id_index = process_event_op2_stub;
408 if (tool->auxtrace_info == NULL)
409 tool->auxtrace_info = process_event_op2_stub;
410 if (tool->auxtrace == NULL)
411 tool->auxtrace = process_event_auxtrace_stub;
412 if (tool->auxtrace_error == NULL)
413 tool->auxtrace_error = process_event_op2_stub;
414 if (tool->thread_map == NULL)
415 tool->thread_map = process_event_thread_map_stub;
416 if (tool->cpu_map == NULL)
417 tool->cpu_map = process_event_cpu_map_stub;
418 if (tool->stat_config == NULL)
419 tool->stat_config = process_event_stat_config_stub;
420 if (tool->stat == NULL)
421 tool->stat = process_stat_stub;
422 if (tool->stat_round == NULL)
423 tool->stat_round = process_stat_round_stub;
424 if (tool->time_conv == NULL)
425 tool->time_conv = process_event_op2_stub;
426 if (tool->feature == NULL)
427 tool->feature = process_event_op2_stub;
428}
429
430static void swap_sample_id_all(union perf_event *event, void *data)
431{
432 void *end = (void *) event + event->header.size;
433 int size = end - data;
434
435 BUG_ON(size % sizeof(u64));
436 mem_bswap_64(data, size);
437}
438
439static void perf_event__all64_swap(union perf_event *event,
440 bool sample_id_all __maybe_unused)
441{
442 struct perf_event_header *hdr = &event->header;
443 mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr));
444}
445
446static void perf_event__comm_swap(union perf_event *event, bool sample_id_all)
447{
448 event->comm.pid = bswap_32(event->comm.pid);
449 event->comm.tid = bswap_32(event->comm.tid);
450
451 if (sample_id_all) {
452 void *data = &event->comm.comm;
453
454 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
455 swap_sample_id_all(event, data);
456 }
457}
458
459static void perf_event__mmap_swap(union perf_event *event,
460 bool sample_id_all)
461{
462 event->mmap.pid = bswap_32(event->mmap.pid);
463 event->mmap.tid = bswap_32(event->mmap.tid);
464 event->mmap.start = bswap_64(event->mmap.start);
465 event->mmap.len = bswap_64(event->mmap.len);
466 event->mmap.pgoff = bswap_64(event->mmap.pgoff);
467
468 if (sample_id_all) {
469 void *data = &event->mmap.filename;
470
471 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
472 swap_sample_id_all(event, data);
473 }
474}
475
476static void perf_event__mmap2_swap(union perf_event *event,
477 bool sample_id_all)
478{
479 event->mmap2.pid = bswap_32(event->mmap2.pid);
480 event->mmap2.tid = bswap_32(event->mmap2.tid);
481 event->mmap2.start = bswap_64(event->mmap2.start);
482 event->mmap2.len = bswap_64(event->mmap2.len);
483 event->mmap2.pgoff = bswap_64(event->mmap2.pgoff);
484 event->mmap2.maj = bswap_32(event->mmap2.maj);
485 event->mmap2.min = bswap_32(event->mmap2.min);
486 event->mmap2.ino = bswap_64(event->mmap2.ino);
487
488 if (sample_id_all) {
489 void *data = &event->mmap2.filename;
490
491 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
492 swap_sample_id_all(event, data);
493 }
494}
495static void perf_event__task_swap(union perf_event *event, bool sample_id_all)
496{
497 event->fork.pid = bswap_32(event->fork.pid);
498 event->fork.tid = bswap_32(event->fork.tid);
499 event->fork.ppid = bswap_32(event->fork.ppid);
500 event->fork.ptid = bswap_32(event->fork.ptid);
501 event->fork.time = bswap_64(event->fork.time);
502
503 if (sample_id_all)
504 swap_sample_id_all(event, &event->fork + 1);
505}
506
507static void perf_event__read_swap(union perf_event *event, bool sample_id_all)
508{
509 event->read.pid = bswap_32(event->read.pid);
510 event->read.tid = bswap_32(event->read.tid);
511 event->read.value = bswap_64(event->read.value);
512 event->read.time_enabled = bswap_64(event->read.time_enabled);
513 event->read.time_running = bswap_64(event->read.time_running);
514 event->read.id = bswap_64(event->read.id);
515
516 if (sample_id_all)
517 swap_sample_id_all(event, &event->read + 1);
518}
519
520static void perf_event__aux_swap(union perf_event *event, bool sample_id_all)
521{
522 event->aux.aux_offset = bswap_64(event->aux.aux_offset);
523 event->aux.aux_size = bswap_64(event->aux.aux_size);
524 event->aux.flags = bswap_64(event->aux.flags);
525
526 if (sample_id_all)
527 swap_sample_id_all(event, &event->aux + 1);
528}
529
530static void perf_event__itrace_start_swap(union perf_event *event,
531 bool sample_id_all)
532{
533 event->itrace_start.pid = bswap_32(event->itrace_start.pid);
534 event->itrace_start.tid = bswap_32(event->itrace_start.tid);
535
536 if (sample_id_all)
537 swap_sample_id_all(event, &event->itrace_start + 1);
538}
539
540static void perf_event__switch_swap(union perf_event *event, bool sample_id_all)
541{
542 if (event->header.type == PERF_RECORD_SWITCH_CPU_WIDE) {
543 event->context_switch.next_prev_pid =
544 bswap_32(event->context_switch.next_prev_pid);
545 event->context_switch.next_prev_tid =
546 bswap_32(event->context_switch.next_prev_tid);
547 }
548
549 if (sample_id_all)
550 swap_sample_id_all(event, &event->context_switch + 1);
551}
552
553static void perf_event__throttle_swap(union perf_event *event,
554 bool sample_id_all)
555{
556 event->throttle.time = bswap_64(event->throttle.time);
557 event->throttle.id = bswap_64(event->throttle.id);
558 event->throttle.stream_id = bswap_64(event->throttle.stream_id);
559
560 if (sample_id_all)
561 swap_sample_id_all(event, &event->throttle + 1);
562}
563
564static u8 revbyte(u8 b)
565{
566 int rev = (b >> 4) | ((b & 0xf) << 4);
567 rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2);
568 rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1);
569 return (u8) rev;
570}
571
572/*
573 * XXX this is hack in attempt to carry flags bitfield
574 * through endian village. ABI says:
575 *
576 * Bit-fields are allocated from right to left (least to most significant)
577 * on little-endian implementations and from left to right (most to least
578 * significant) on big-endian implementations.
579 *
580 * The above seems to be byte specific, so we need to reverse each
581 * byte of the bitfield. 'Internet' also says this might be implementation
582 * specific and we probably need proper fix and carry perf_event_attr
583 * bitfield flags in separate data file FEAT_ section. Thought this seems
584 * to work for now.
585 */
586static void swap_bitfield(u8 *p, unsigned len)
587{
588 unsigned i;
589
590 for (i = 0; i < len; i++) {
591 *p = revbyte(*p);
592 p++;
593 }
594}
595
596/* exported for swapping attributes in file header */
597void perf_event__attr_swap(struct perf_event_attr *attr)
598{
599 attr->type = bswap_32(attr->type);
600 attr->size = bswap_32(attr->size);
601
602#define bswap_safe(f, n) \
603 (attr->size > (offsetof(struct perf_event_attr, f) + \
604 sizeof(attr->f) * (n)))
605#define bswap_field(f, sz) \
606do { \
607 if (bswap_safe(f, 0)) \
608 attr->f = bswap_##sz(attr->f); \
609} while(0)
610#define bswap_field_16(f) bswap_field(f, 16)
611#define bswap_field_32(f) bswap_field(f, 32)
612#define bswap_field_64(f) bswap_field(f, 64)
613
614 bswap_field_64(config);
615 bswap_field_64(sample_period);
616 bswap_field_64(sample_type);
617 bswap_field_64(read_format);
618 bswap_field_32(wakeup_events);
619 bswap_field_32(bp_type);
620 bswap_field_64(bp_addr);
621 bswap_field_64(bp_len);
622 bswap_field_64(branch_sample_type);
623 bswap_field_64(sample_regs_user);
624 bswap_field_32(sample_stack_user);
625 bswap_field_32(aux_watermark);
626 bswap_field_16(sample_max_stack);
627
628 /*
629 * After read_format are bitfields. Check read_format because
630 * we are unable to use offsetof on bitfield.
631 */
632 if (bswap_safe(read_format, 1))
633 swap_bitfield((u8 *) (&attr->read_format + 1),
634 sizeof(u64));
635#undef bswap_field_64
636#undef bswap_field_32
637#undef bswap_field
638#undef bswap_safe
639}
640
641static void perf_event__hdr_attr_swap(union perf_event *event,
642 bool sample_id_all __maybe_unused)
643{
644 size_t size;
645
646 perf_event__attr_swap(&event->attr.attr);
647
648 size = event->header.size;
649 size -= (void *)&event->attr.id - (void *)event;
650 mem_bswap_64(event->attr.id, size);
651}
652
653static void perf_event__event_update_swap(union perf_event *event,
654 bool sample_id_all __maybe_unused)
655{
656 event->event_update.type = bswap_64(event->event_update.type);
657 event->event_update.id = bswap_64(event->event_update.id);
658}
659
660static void perf_event__event_type_swap(union perf_event *event,
661 bool sample_id_all __maybe_unused)
662{
663 event->event_type.event_type.event_id =
664 bswap_64(event->event_type.event_type.event_id);
665}
666
667static void perf_event__tracing_data_swap(union perf_event *event,
668 bool sample_id_all __maybe_unused)
669{
670 event->tracing_data.size = bswap_32(event->tracing_data.size);
671}
672
673static void perf_event__auxtrace_info_swap(union perf_event *event,
674 bool sample_id_all __maybe_unused)
675{
676 size_t size;
677
678 event->auxtrace_info.type = bswap_32(event->auxtrace_info.type);
679
680 size = event->header.size;
681 size -= (void *)&event->auxtrace_info.priv - (void *)event;
682 mem_bswap_64(event->auxtrace_info.priv, size);
683}
684
685static void perf_event__auxtrace_swap(union perf_event *event,
686 bool sample_id_all __maybe_unused)
687{
688 event->auxtrace.size = bswap_64(event->auxtrace.size);
689 event->auxtrace.offset = bswap_64(event->auxtrace.offset);
690 event->auxtrace.reference = bswap_64(event->auxtrace.reference);
691 event->auxtrace.idx = bswap_32(event->auxtrace.idx);
692 event->auxtrace.tid = bswap_32(event->auxtrace.tid);
693 event->auxtrace.cpu = bswap_32(event->auxtrace.cpu);
694}
695
696static void perf_event__auxtrace_error_swap(union perf_event *event,
697 bool sample_id_all __maybe_unused)
698{
699 event->auxtrace_error.type = bswap_32(event->auxtrace_error.type);
700 event->auxtrace_error.code = bswap_32(event->auxtrace_error.code);
701 event->auxtrace_error.cpu = bswap_32(event->auxtrace_error.cpu);
702 event->auxtrace_error.pid = bswap_32(event->auxtrace_error.pid);
703 event->auxtrace_error.tid = bswap_32(event->auxtrace_error.tid);
704 event->auxtrace_error.ip = bswap_64(event->auxtrace_error.ip);
705}
706
707static void perf_event__thread_map_swap(union perf_event *event,
708 bool sample_id_all __maybe_unused)
709{
710 unsigned i;
711
712 event->thread_map.nr = bswap_64(event->thread_map.nr);
713
714 for (i = 0; i < event->thread_map.nr; i++)
715 event->thread_map.entries[i].pid = bswap_64(event->thread_map.entries[i].pid);
716}
717
718static void perf_event__cpu_map_swap(union perf_event *event,
719 bool sample_id_all __maybe_unused)
720{
721 struct cpu_map_data *data = &event->cpu_map.data;
722 struct cpu_map_entries *cpus;
723 struct cpu_map_mask *mask;
724 unsigned i;
725
726 data->type = bswap_64(data->type);
727
728 switch (data->type) {
729 case PERF_CPU_MAP__CPUS:
730 cpus = (struct cpu_map_entries *)data->data;
731
732 cpus->nr = bswap_16(cpus->nr);
733
734 for (i = 0; i < cpus->nr; i++)
735 cpus->cpu[i] = bswap_16(cpus->cpu[i]);
736 break;
737 case PERF_CPU_MAP__MASK:
738 mask = (struct cpu_map_mask *) data->data;
739
740 mask->nr = bswap_16(mask->nr);
741 mask->long_size = bswap_16(mask->long_size);
742
743 switch (mask->long_size) {
744 case 4: mem_bswap_32(&mask->mask, mask->nr); break;
745 case 8: mem_bswap_64(&mask->mask, mask->nr); break;
746 default:
747 pr_err("cpu_map swap: unsupported long size\n");
748 }
749 default:
750 break;
751 }
752}
753
754static void perf_event__stat_config_swap(union perf_event *event,
755 bool sample_id_all __maybe_unused)
756{
757 u64 size;
758
759 size = event->stat_config.nr * sizeof(event->stat_config.data[0]);
760 size += 1; /* nr item itself */
761 mem_bswap_64(&event->stat_config.nr, size);
762}
763
764static void perf_event__stat_swap(union perf_event *event,
765 bool sample_id_all __maybe_unused)
766{
767 event->stat.id = bswap_64(event->stat.id);
768 event->stat.thread = bswap_32(event->stat.thread);
769 event->stat.cpu = bswap_32(event->stat.cpu);
770 event->stat.val = bswap_64(event->stat.val);
771 event->stat.ena = bswap_64(event->stat.ena);
772 event->stat.run = bswap_64(event->stat.run);
773}
774
775static void perf_event__stat_round_swap(union perf_event *event,
776 bool sample_id_all __maybe_unused)
777{
778 event->stat_round.type = bswap_64(event->stat_round.type);
779 event->stat_round.time = bswap_64(event->stat_round.time);
780}
781
782typedef void (*perf_event__swap_op)(union perf_event *event,
783 bool sample_id_all);
784
785static perf_event__swap_op perf_event__swap_ops[] = {
786 [PERF_RECORD_MMAP] = perf_event__mmap_swap,
787 [PERF_RECORD_MMAP2] = perf_event__mmap2_swap,
788 [PERF_RECORD_COMM] = perf_event__comm_swap,
789 [PERF_RECORD_FORK] = perf_event__task_swap,
790 [PERF_RECORD_EXIT] = perf_event__task_swap,
791 [PERF_RECORD_LOST] = perf_event__all64_swap,
792 [PERF_RECORD_READ] = perf_event__read_swap,
793 [PERF_RECORD_THROTTLE] = perf_event__throttle_swap,
794 [PERF_RECORD_UNTHROTTLE] = perf_event__throttle_swap,
795 [PERF_RECORD_SAMPLE] = perf_event__all64_swap,
796 [PERF_RECORD_AUX] = perf_event__aux_swap,
797 [PERF_RECORD_ITRACE_START] = perf_event__itrace_start_swap,
798 [PERF_RECORD_LOST_SAMPLES] = perf_event__all64_swap,
799 [PERF_RECORD_SWITCH] = perf_event__switch_swap,
800 [PERF_RECORD_SWITCH_CPU_WIDE] = perf_event__switch_swap,
801 [PERF_RECORD_HEADER_ATTR] = perf_event__hdr_attr_swap,
802 [PERF_RECORD_HEADER_EVENT_TYPE] = perf_event__event_type_swap,
803 [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
804 [PERF_RECORD_HEADER_BUILD_ID] = NULL,
805 [PERF_RECORD_ID_INDEX] = perf_event__all64_swap,
806 [PERF_RECORD_AUXTRACE_INFO] = perf_event__auxtrace_info_swap,
807 [PERF_RECORD_AUXTRACE] = perf_event__auxtrace_swap,
808 [PERF_RECORD_AUXTRACE_ERROR] = perf_event__auxtrace_error_swap,
809 [PERF_RECORD_THREAD_MAP] = perf_event__thread_map_swap,
810 [PERF_RECORD_CPU_MAP] = perf_event__cpu_map_swap,
811 [PERF_RECORD_STAT_CONFIG] = perf_event__stat_config_swap,
812 [PERF_RECORD_STAT] = perf_event__stat_swap,
813 [PERF_RECORD_STAT_ROUND] = perf_event__stat_round_swap,
814 [PERF_RECORD_EVENT_UPDATE] = perf_event__event_update_swap,
815 [PERF_RECORD_TIME_CONV] = perf_event__all64_swap,
816 [PERF_RECORD_HEADER_MAX] = NULL,
817};
818
819/*
820 * When perf record finishes a pass on every buffers, it records this pseudo
821 * event.
822 * We record the max timestamp t found in the pass n.
823 * Assuming these timestamps are monotonic across cpus, we know that if
824 * a buffer still has events with timestamps below t, they will be all
825 * available and then read in the pass n + 1.
826 * Hence when we start to read the pass n + 2, we can safely flush every
827 * events with timestamps below t.
828 *
829 * ============ PASS n =================
830 * CPU 0 | CPU 1
831 * |
832 * cnt1 timestamps | cnt2 timestamps
833 * 1 | 2
834 * 2 | 3
835 * - | 4 <--- max recorded
836 *
837 * ============ PASS n + 1 ==============
838 * CPU 0 | CPU 1
839 * |
840 * cnt1 timestamps | cnt2 timestamps
841 * 3 | 5
842 * 4 | 6
843 * 5 | 7 <---- max recorded
844 *
845 * Flush every events below timestamp 4
846 *
847 * ============ PASS n + 2 ==============
848 * CPU 0 | CPU 1
849 * |
850 * cnt1 timestamps | cnt2 timestamps
851 * 6 | 8
852 * 7 | 9
853 * - | 10
854 *
855 * Flush every events below timestamp 7
856 * etc...
857 */
858static int process_finished_round(struct perf_tool *tool __maybe_unused,
859 union perf_event *event __maybe_unused,
860 struct ordered_events *oe)
861{
862 if (dump_trace)
863 fprintf(stdout, "\n");
864 return ordered_events__flush(oe, OE_FLUSH__ROUND);
865}
866
867int perf_session__queue_event(struct perf_session *s, union perf_event *event,
868 u64 timestamp, u64 file_offset)
869{
870 return ordered_events__queue(&s->ordered_events, event, timestamp, file_offset);
871}
872
873static void callchain__lbr_callstack_printf(struct perf_sample *sample)
874{
875 struct ip_callchain *callchain = sample->callchain;
876 struct branch_stack *lbr_stack = sample->branch_stack;
877 u64 kernel_callchain_nr = callchain->nr;
878 unsigned int i;
879
880 for (i = 0; i < kernel_callchain_nr; i++) {
881 if (callchain->ips[i] == PERF_CONTEXT_USER)
882 break;
883 }
884
885 if ((i != kernel_callchain_nr) && lbr_stack->nr) {
886 u64 total_nr;
887 /*
888 * LBR callstack can only get user call chain,
889 * i is kernel call chain number,
890 * 1 is PERF_CONTEXT_USER.
891 *
892 * The user call chain is stored in LBR registers.
893 * LBR are pair registers. The caller is stored
894 * in "from" register, while the callee is stored
895 * in "to" register.
896 * For example, there is a call stack
897 * "A"->"B"->"C"->"D".
898 * The LBR registers will recorde like
899 * "C"->"D", "B"->"C", "A"->"B".
900 * So only the first "to" register and all "from"
901 * registers are needed to construct the whole stack.
902 */
903 total_nr = i + 1 + lbr_stack->nr + 1;
904 kernel_callchain_nr = i + 1;
905
906 printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr);
907
908 for (i = 0; i < kernel_callchain_nr; i++)
909 printf("..... %2d: %016" PRIx64 "\n",
910 i, callchain->ips[i]);
911
912 printf("..... %2d: %016" PRIx64 "\n",
913 (int)(kernel_callchain_nr), lbr_stack->entries[0].to);
914 for (i = 0; i < lbr_stack->nr; i++)
915 printf("..... %2d: %016" PRIx64 "\n",
916 (int)(i + kernel_callchain_nr + 1), lbr_stack->entries[i].from);
917 }
918}
919
920static void callchain__printf(struct perf_evsel *evsel,
921 struct perf_sample *sample)
922{
923 unsigned int i;
924 struct ip_callchain *callchain = sample->callchain;
925
926 if (perf_evsel__has_branch_callstack(evsel))
927 callchain__lbr_callstack_printf(sample);
928
929 printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr);
930
931 for (i = 0; i < callchain->nr; i++)
932 printf("..... %2d: %016" PRIx64 "\n",
933 i, callchain->ips[i]);
934}
935
936static void branch_stack__printf(struct perf_sample *sample)
937{
938 uint64_t i;
939
940 printf("... branch stack: nr:%" PRIu64 "\n", sample->branch_stack->nr);
941
942 for (i = 0; i < sample->branch_stack->nr; i++) {
943 struct branch_entry *e = &sample->branch_stack->entries[i];
944
945 printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x\n",
946 i, e->from, e->to,
947 (unsigned short)e->flags.cycles,
948 e->flags.mispred ? "M" : " ",
949 e->flags.predicted ? "P" : " ",
950 e->flags.abort ? "A" : " ",
951 e->flags.in_tx ? "T" : " ",
952 (unsigned)e->flags.reserved);
953 }
954}
955
956static void regs_dump__printf(u64 mask, u64 *regs)
957{
958 unsigned rid, i = 0;
959
960 for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) {
961 u64 val = regs[i++];
962
963 printf(".... %-5s 0x%" PRIx64 "\n",
964 perf_reg_name(rid), val);
965 }
966}
967
968static const char *regs_abi[] = {
969 [PERF_SAMPLE_REGS_ABI_NONE] = "none",
970 [PERF_SAMPLE_REGS_ABI_32] = "32-bit",
971 [PERF_SAMPLE_REGS_ABI_64] = "64-bit",
972};
973
974static inline const char *regs_dump_abi(struct regs_dump *d)
975{
976 if (d->abi > PERF_SAMPLE_REGS_ABI_64)
977 return "unknown";
978
979 return regs_abi[d->abi];
980}
981
982static void regs__printf(const char *type, struct regs_dump *regs)
983{
984 u64 mask = regs->mask;
985
986 printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n",
987 type,
988 mask,
989 regs_dump_abi(regs));
990
991 regs_dump__printf(mask, regs->regs);
992}
993
994static void regs_user__printf(struct perf_sample *sample)
995{
996 struct regs_dump *user_regs = &sample->user_regs;
997
998 if (user_regs->regs)
999 regs__printf("user", user_regs);
1000}
1001
1002static void regs_intr__printf(struct perf_sample *sample)
1003{
1004 struct regs_dump *intr_regs = &sample->intr_regs;
1005
1006 if (intr_regs->regs)
1007 regs__printf("intr", intr_regs);
1008}
1009
1010static void stack_user__printf(struct stack_dump *dump)
1011{
1012 printf("... ustack: size %" PRIu64 ", offset 0x%x\n",
1013 dump->size, dump->offset);
1014}
1015
1016static void perf_evlist__print_tstamp(struct perf_evlist *evlist,
1017 union perf_event *event,
1018 struct perf_sample *sample)
1019{
1020 u64 sample_type = __perf_evlist__combined_sample_type(evlist);
1021
1022 if (event->header.type != PERF_RECORD_SAMPLE &&
1023 !perf_evlist__sample_id_all(evlist)) {
1024 fputs("-1 -1 ", stdout);
1025 return;
1026 }
1027
1028 if ((sample_type & PERF_SAMPLE_CPU))
1029 printf("%u ", sample->cpu);
1030
1031 if (sample_type & PERF_SAMPLE_TIME)
1032 printf("%" PRIu64 " ", sample->time);
1033}
1034
1035static void sample_read__printf(struct perf_sample *sample, u64 read_format)
1036{
1037 printf("... sample_read:\n");
1038
1039 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1040 printf("...... time enabled %016" PRIx64 "\n",
1041 sample->read.time_enabled);
1042
1043 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1044 printf("...... time running %016" PRIx64 "\n",
1045 sample->read.time_running);
1046
1047 if (read_format & PERF_FORMAT_GROUP) {
1048 u64 i;
1049
1050 printf(".... group nr %" PRIu64 "\n", sample->read.group.nr);
1051
1052 for (i = 0; i < sample->read.group.nr; i++) {
1053 struct sample_read_value *value;
1054
1055 value = &sample->read.group.values[i];
1056 printf("..... id %016" PRIx64
1057 ", value %016" PRIx64 "\n",
1058 value->id, value->value);
1059 }
1060 } else
1061 printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n",
1062 sample->read.one.id, sample->read.one.value);
1063}
1064
1065static void dump_event(struct perf_evlist *evlist, union perf_event *event,
1066 u64 file_offset, struct perf_sample *sample)
1067{
1068 if (!dump_trace)
1069 return;
1070
1071 printf("\n%#" PRIx64 " [%#x]: event: %d\n",
1072 file_offset, event->header.size, event->header.type);
1073
1074 trace_event(event);
1075
1076 if (sample)
1077 perf_evlist__print_tstamp(evlist, event, sample);
1078
1079 printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset,
1080 event->header.size, perf_event__name(event->header.type));
1081}
1082
1083static void dump_sample(struct perf_evsel *evsel, union perf_event *event,
1084 struct perf_sample *sample)
1085{
1086 u64 sample_type;
1087
1088 if (!dump_trace)
1089 return;
1090
1091 printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n",
1092 event->header.misc, sample->pid, sample->tid, sample->ip,
1093 sample->period, sample->addr);
1094
1095 sample_type = evsel->attr.sample_type;
1096
1097 if (sample_type & PERF_SAMPLE_CALLCHAIN)
1098 callchain__printf(evsel, sample);
1099
1100 if ((sample_type & PERF_SAMPLE_BRANCH_STACK) && !perf_evsel__has_branch_callstack(evsel))
1101 branch_stack__printf(sample);
1102
1103 if (sample_type & PERF_SAMPLE_REGS_USER)
1104 regs_user__printf(sample);
1105
1106 if (sample_type & PERF_SAMPLE_REGS_INTR)
1107 regs_intr__printf(sample);
1108
1109 if (sample_type & PERF_SAMPLE_STACK_USER)
1110 stack_user__printf(&sample->user_stack);
1111
1112 if (sample_type & PERF_SAMPLE_WEIGHT)
1113 printf("... weight: %" PRIu64 "\n", sample->weight);
1114
1115 if (sample_type & PERF_SAMPLE_DATA_SRC)
1116 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src);
1117
1118 if (sample_type & PERF_SAMPLE_PHYS_ADDR)
1119 printf(" .. phys_addr: 0x%"PRIx64"\n", sample->phys_addr);
1120
1121 if (sample_type & PERF_SAMPLE_TRANSACTION)
1122 printf("... transaction: %" PRIx64 "\n", sample->transaction);
1123
1124 if (sample_type & PERF_SAMPLE_READ)
1125 sample_read__printf(sample, evsel->attr.read_format);
1126}
1127
1128static void dump_read(struct perf_evsel *evsel, union perf_event *event)
1129{
1130 struct read_event *read_event = &event->read;
1131 u64 read_format;
1132
1133 if (!dump_trace)
1134 return;
1135
1136 printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid,
1137 evsel ? perf_evsel__name(evsel) : "FAIL",
1138 event->read.value);
1139
1140 read_format = evsel->attr.read_format;
1141
1142 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1143 printf("... time enabled : %" PRIu64 "\n", read_event->time_enabled);
1144
1145 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1146 printf("... time running : %" PRIu64 "\n", read_event->time_running);
1147
1148 if (read_format & PERF_FORMAT_ID)
1149 printf("... id : %" PRIu64 "\n", read_event->id);
1150}
1151
1152static struct machine *machines__find_for_cpumode(struct machines *machines,
1153 union perf_event *event,
1154 struct perf_sample *sample)
1155{
1156 struct machine *machine;
1157
1158 if (perf_guest &&
1159 ((sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL) ||
1160 (sample->cpumode == PERF_RECORD_MISC_GUEST_USER))) {
1161 u32 pid;
1162
1163 if (event->header.type == PERF_RECORD_MMAP
1164 || event->header.type == PERF_RECORD_MMAP2)
1165 pid = event->mmap.pid;
1166 else
1167 pid = sample->pid;
1168
1169 machine = machines__find(machines, pid);
1170 if (!machine)
1171 machine = machines__findnew(machines, DEFAULT_GUEST_KERNEL_ID);
1172 return machine;
1173 }
1174
1175 return &machines->host;
1176}
1177
1178static int deliver_sample_value(struct perf_evlist *evlist,
1179 struct perf_tool *tool,
1180 union perf_event *event,
1181 struct perf_sample *sample,
1182 struct sample_read_value *v,
1183 struct machine *machine)
1184{
1185 struct perf_sample_id *sid = perf_evlist__id2sid(evlist, v->id);
1186
1187 if (sid) {
1188 sample->id = v->id;
1189 sample->period = v->value - sid->period;
1190 sid->period = v->value;
1191 }
1192
1193 if (!sid || sid->evsel == NULL) {
1194 ++evlist->stats.nr_unknown_id;
1195 return 0;
1196 }
1197
1198 return tool->sample(tool, event, sample, sid->evsel, machine);
1199}
1200
1201static int deliver_sample_group(struct perf_evlist *evlist,
1202 struct perf_tool *tool,
1203 union perf_event *event,
1204 struct perf_sample *sample,
1205 struct machine *machine)
1206{
1207 int ret = -EINVAL;
1208 u64 i;
1209
1210 for (i = 0; i < sample->read.group.nr; i++) {
1211 ret = deliver_sample_value(evlist, tool, event, sample,
1212 &sample->read.group.values[i],
1213 machine);
1214 if (ret)
1215 break;
1216 }
1217
1218 return ret;
1219}
1220
1221static int
1222 perf_evlist__deliver_sample(struct perf_evlist *evlist,
1223 struct perf_tool *tool,
1224 union perf_event *event,
1225 struct perf_sample *sample,
1226 struct perf_evsel *evsel,
1227 struct machine *machine)
1228{
1229 /* We know evsel != NULL. */
1230 u64 sample_type = evsel->attr.sample_type;
1231 u64 read_format = evsel->attr.read_format;
1232
1233 /* Standard sample delivery. */
1234 if (!(sample_type & PERF_SAMPLE_READ))
1235 return tool->sample(tool, event, sample, evsel, machine);
1236
1237 /* For PERF_SAMPLE_READ we have either single or group mode. */
1238 if (read_format & PERF_FORMAT_GROUP)
1239 return deliver_sample_group(evlist, tool, event, sample,
1240 machine);
1241 else
1242 return deliver_sample_value(evlist, tool, event, sample,
1243 &sample->read.one, machine);
1244}
1245
1246static int machines__deliver_event(struct machines *machines,
1247 struct perf_evlist *evlist,
1248 union perf_event *event,
1249 struct perf_sample *sample,
1250 struct perf_tool *tool, u64 file_offset)
1251{
1252 struct perf_evsel *evsel;
1253 struct machine *machine;
1254
1255 dump_event(evlist, event, file_offset, sample);
1256
1257 evsel = perf_evlist__id2evsel(evlist, sample->id);
1258
1259 machine = machines__find_for_cpumode(machines, event, sample);
1260
1261 switch (event->header.type) {
1262 case PERF_RECORD_SAMPLE:
1263 if (evsel == NULL) {
1264 ++evlist->stats.nr_unknown_id;
1265 return 0;
1266 }
1267 dump_sample(evsel, event, sample);
1268 if (machine == NULL) {
1269 ++evlist->stats.nr_unprocessable_samples;
1270 return 0;
1271 }
1272 return perf_evlist__deliver_sample(evlist, tool, event, sample, evsel, machine);
1273 case PERF_RECORD_MMAP:
1274 return tool->mmap(tool, event, sample, machine);
1275 case PERF_RECORD_MMAP2:
1276 if (event->header.misc & PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT)
1277 ++evlist->stats.nr_proc_map_timeout;
1278 return tool->mmap2(tool, event, sample, machine);
1279 case PERF_RECORD_COMM:
1280 return tool->comm(tool, event, sample, machine);
1281 case PERF_RECORD_NAMESPACES:
1282 return tool->namespaces(tool, event, sample, machine);
1283 case PERF_RECORD_FORK:
1284 return tool->fork(tool, event, sample, machine);
1285 case PERF_RECORD_EXIT:
1286 return tool->exit(tool, event, sample, machine);
1287 case PERF_RECORD_LOST:
1288 if (tool->lost == perf_event__process_lost)
1289 evlist->stats.total_lost += event->lost.lost;
1290 return tool->lost(tool, event, sample, machine);
1291 case PERF_RECORD_LOST_SAMPLES:
1292 if (tool->lost_samples == perf_event__process_lost_samples)
1293 evlist->stats.total_lost_samples += event->lost_samples.lost;
1294 return tool->lost_samples(tool, event, sample, machine);
1295 case PERF_RECORD_READ:
1296 dump_read(evsel, event);
1297 return tool->read(tool, event, sample, evsel, machine);
1298 case PERF_RECORD_THROTTLE:
1299 return tool->throttle(tool, event, sample, machine);
1300 case PERF_RECORD_UNTHROTTLE:
1301 return tool->unthrottle(tool, event, sample, machine);
1302 case PERF_RECORD_AUX:
1303 if (tool->aux == perf_event__process_aux) {
1304 if (event->aux.flags & PERF_AUX_FLAG_TRUNCATED)
1305 evlist->stats.total_aux_lost += 1;
1306 if (event->aux.flags & PERF_AUX_FLAG_PARTIAL)
1307 evlist->stats.total_aux_partial += 1;
1308 }
1309 return tool->aux(tool, event, sample, machine);
1310 case PERF_RECORD_ITRACE_START:
1311 return tool->itrace_start(tool, event, sample, machine);
1312 case PERF_RECORD_SWITCH:
1313 case PERF_RECORD_SWITCH_CPU_WIDE:
1314 return tool->context_switch(tool, event, sample, machine);
1315 default:
1316 ++evlist->stats.nr_unknown_events;
1317 return -1;
1318 }
1319}
1320
1321static int perf_session__deliver_event(struct perf_session *session,
1322 union perf_event *event,
1323 struct perf_tool *tool,
1324 u64 file_offset)
1325{
1326 struct perf_sample sample;
1327 int ret;
1328
1329 ret = perf_evlist__parse_sample(session->evlist, event, &sample);
1330 if (ret) {
1331 pr_err("Can't parse sample, err = %d\n", ret);
1332 return ret;
1333 }
1334
1335 ret = auxtrace__process_event(session, event, &sample, tool);
1336 if (ret < 0)
1337 return ret;
1338 if (ret > 0)
1339 return 0;
1340
1341 return machines__deliver_event(&session->machines, session->evlist,
1342 event, &sample, tool, file_offset);
1343}
1344
1345static s64 perf_session__process_user_event(struct perf_session *session,
1346 union perf_event *event,
1347 u64 file_offset)
1348{
1349 struct ordered_events *oe = &session->ordered_events;
1350 struct perf_tool *tool = session->tool;
1351 struct perf_sample sample = { .time = 0, };
1352 int fd = perf_data__fd(session->data);
1353 int err;
1354
1355 dump_event(session->evlist, event, file_offset, &sample);
1356
1357 /* These events are processed right away */
1358 switch (event->header.type) {
1359 case PERF_RECORD_HEADER_ATTR:
1360 err = tool->attr(tool, event, &session->evlist);
1361 if (err == 0) {
1362 perf_session__set_id_hdr_size(session);
1363 perf_session__set_comm_exec(session);
1364 }
1365 return err;
1366 case PERF_RECORD_EVENT_UPDATE:
1367 return tool->event_update(tool, event, &session->evlist);
1368 case PERF_RECORD_HEADER_EVENT_TYPE:
1369 /*
1370 * Depreceated, but we need to handle it for sake
1371 * of old data files create in pipe mode.
1372 */
1373 return 0;
1374 case PERF_RECORD_HEADER_TRACING_DATA:
1375 /* setup for reading amidst mmap */
1376 lseek(fd, file_offset, SEEK_SET);
1377 return tool->tracing_data(tool, event, session);
1378 case PERF_RECORD_HEADER_BUILD_ID:
1379 return tool->build_id(tool, event, session);
1380 case PERF_RECORD_FINISHED_ROUND:
1381 return tool->finished_round(tool, event, oe);
1382 case PERF_RECORD_ID_INDEX:
1383 return tool->id_index(tool, event, session);
1384 case PERF_RECORD_AUXTRACE_INFO:
1385 return tool->auxtrace_info(tool, event, session);
1386 case PERF_RECORD_AUXTRACE:
1387 /* setup for reading amidst mmap */
1388 lseek(fd, file_offset + event->header.size, SEEK_SET);
1389 return tool->auxtrace(tool, event, session);
1390 case PERF_RECORD_AUXTRACE_ERROR:
1391 perf_session__auxtrace_error_inc(session, event);
1392 return tool->auxtrace_error(tool, event, session);
1393 case PERF_RECORD_THREAD_MAP:
1394 return tool->thread_map(tool, event, session);
1395 case PERF_RECORD_CPU_MAP:
1396 return tool->cpu_map(tool, event, session);
1397 case PERF_RECORD_STAT_CONFIG:
1398 return tool->stat_config(tool, event, session);
1399 case PERF_RECORD_STAT:
1400 return tool->stat(tool, event, session);
1401 case PERF_RECORD_STAT_ROUND:
1402 return tool->stat_round(tool, event, session);
1403 case PERF_RECORD_TIME_CONV:
1404 session->time_conv = event->time_conv;
1405 return tool->time_conv(tool, event, session);
1406 case PERF_RECORD_HEADER_FEATURE:
1407 return tool->feature(tool, event, session);
1408 default:
1409 return -EINVAL;
1410 }
1411}
1412
1413int perf_session__deliver_synth_event(struct perf_session *session,
1414 union perf_event *event,
1415 struct perf_sample *sample)
1416{
1417 struct perf_evlist *evlist = session->evlist;
1418 struct perf_tool *tool = session->tool;
1419
1420 events_stats__inc(&evlist->stats, event->header.type);
1421
1422 if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1423 return perf_session__process_user_event(session, event, 0);
1424
1425 return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0);
1426}
1427
1428static void event_swap(union perf_event *event, bool sample_id_all)
1429{
1430 perf_event__swap_op swap;
1431
1432 swap = perf_event__swap_ops[event->header.type];
1433 if (swap)
1434 swap(event, sample_id_all);
1435}
1436
1437int perf_session__peek_event(struct perf_session *session, off_t file_offset,
1438 void *buf, size_t buf_sz,
1439 union perf_event **event_ptr,
1440 struct perf_sample *sample)
1441{
1442 union perf_event *event;
1443 size_t hdr_sz, rest;
1444 int fd;
1445
1446 if (session->one_mmap && !session->header.needs_swap) {
1447 event = file_offset - session->one_mmap_offset +
1448 session->one_mmap_addr;
1449 goto out_parse_sample;
1450 }
1451
1452 if (perf_data__is_pipe(session->data))
1453 return -1;
1454
1455 fd = perf_data__fd(session->data);
1456 hdr_sz = sizeof(struct perf_event_header);
1457
1458 if (buf_sz < hdr_sz)
1459 return -1;
1460
1461 if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 ||
1462 readn(fd, buf, hdr_sz) != (ssize_t)hdr_sz)
1463 return -1;
1464
1465 event = (union perf_event *)buf;
1466
1467 if (session->header.needs_swap)
1468 perf_event_header__bswap(&event->header);
1469
1470 if (event->header.size < hdr_sz || event->header.size > buf_sz)
1471 return -1;
1472
1473 rest = event->header.size - hdr_sz;
1474
1475 if (readn(fd, buf, rest) != (ssize_t)rest)
1476 return -1;
1477
1478 if (session->header.needs_swap)
1479 event_swap(event, perf_evlist__sample_id_all(session->evlist));
1480
1481out_parse_sample:
1482
1483 if (sample && event->header.type < PERF_RECORD_USER_TYPE_START &&
1484 perf_evlist__parse_sample(session->evlist, event, sample))
1485 return -1;
1486
1487 *event_ptr = event;
1488
1489 return 0;
1490}
1491
1492static s64 perf_session__process_event(struct perf_session *session,
1493 union perf_event *event, u64 file_offset)
1494{
1495 struct perf_evlist *evlist = session->evlist;
1496 struct perf_tool *tool = session->tool;
1497 int ret;
1498
1499 if (session->header.needs_swap)
1500 event_swap(event, perf_evlist__sample_id_all(evlist));
1501
1502 if (event->header.type >= PERF_RECORD_HEADER_MAX)
1503 return -EINVAL;
1504
1505 events_stats__inc(&evlist->stats, event->header.type);
1506
1507 if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1508 return perf_session__process_user_event(session, event, file_offset);
1509
1510 if (tool->ordered_events) {
1511 u64 timestamp = -1ULL;
1512
1513 ret = perf_evlist__parse_sample_timestamp(evlist, event, ×tamp);
1514 if (ret && ret != -1)
1515 return ret;
1516
1517 ret = perf_session__queue_event(session, event, timestamp, file_offset);
1518 if (ret != -ETIME)
1519 return ret;
1520 }
1521
1522 return perf_session__deliver_event(session, event, tool, file_offset);
1523}
1524
1525void perf_event_header__bswap(struct perf_event_header *hdr)
1526{
1527 hdr->type = bswap_32(hdr->type);
1528 hdr->misc = bswap_16(hdr->misc);
1529 hdr->size = bswap_16(hdr->size);
1530}
1531
1532struct thread *perf_session__findnew(struct perf_session *session, pid_t pid)
1533{
1534 return machine__findnew_thread(&session->machines.host, -1, pid);
1535}
1536
1537int perf_session__register_idle_thread(struct perf_session *session)
1538{
1539 struct thread *thread;
1540 int err = 0;
1541
1542 thread = machine__findnew_thread(&session->machines.host, 0, 0);
1543 if (thread == NULL || thread__set_comm(thread, "swapper", 0)) {
1544 pr_err("problem inserting idle task.\n");
1545 err = -1;
1546 }
1547
1548 if (thread == NULL || thread__set_namespaces(thread, 0, NULL)) {
1549 pr_err("problem inserting idle task.\n");
1550 err = -1;
1551 }
1552
1553 /* machine__findnew_thread() got the thread, so put it */
1554 thread__put(thread);
1555 return err;
1556}
1557
1558static void
1559perf_session__warn_order(const struct perf_session *session)
1560{
1561 const struct ordered_events *oe = &session->ordered_events;
1562 struct perf_evsel *evsel;
1563 bool should_warn = true;
1564
1565 evlist__for_each_entry(session->evlist, evsel) {
1566 if (evsel->attr.write_backward)
1567 should_warn = false;
1568 }
1569
1570 if (!should_warn)
1571 return;
1572 if (oe->nr_unordered_events != 0)
1573 ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events);
1574}
1575
1576static void perf_session__warn_about_errors(const struct perf_session *session)
1577{
1578 const struct events_stats *stats = &session->evlist->stats;
1579
1580 if (session->tool->lost == perf_event__process_lost &&
1581 stats->nr_events[PERF_RECORD_LOST] != 0) {
1582 ui__warning("Processed %d events and lost %d chunks!\n\n"
1583 "Check IO/CPU overload!\n\n",
1584 stats->nr_events[0],
1585 stats->nr_events[PERF_RECORD_LOST]);
1586 }
1587
1588 if (session->tool->lost_samples == perf_event__process_lost_samples) {
1589 double drop_rate;
1590
1591 drop_rate = (double)stats->total_lost_samples /
1592 (double) (stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples);
1593 if (drop_rate > 0.05) {
1594 ui__warning("Processed %" PRIu64 " samples and lost %3.2f%%!\n\n",
1595 stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples,
1596 drop_rate * 100.0);
1597 }
1598 }
1599
1600 if (session->tool->aux == perf_event__process_aux &&
1601 stats->total_aux_lost != 0) {
1602 ui__warning("AUX data lost %" PRIu64 " times out of %u!\n\n",
1603 stats->total_aux_lost,
1604 stats->nr_events[PERF_RECORD_AUX]);
1605 }
1606
1607 if (session->tool->aux == perf_event__process_aux &&
1608 stats->total_aux_partial != 0) {
1609 bool vmm_exclusive = false;
1610
1611 (void)sysfs__read_bool("module/kvm_intel/parameters/vmm_exclusive",
1612 &vmm_exclusive);
1613
1614 ui__warning("AUX data had gaps in it %" PRIu64 " times out of %u!\n\n"
1615 "Are you running a KVM guest in the background?%s\n\n",
1616 stats->total_aux_partial,
1617 stats->nr_events[PERF_RECORD_AUX],
1618 vmm_exclusive ?
1619 "\nReloading kvm_intel module with vmm_exclusive=0\n"
1620 "will reduce the gaps to only guest's timeslices." :
1621 "");
1622 }
1623
1624 if (stats->nr_unknown_events != 0) {
1625 ui__warning("Found %u unknown events!\n\n"
1626 "Is this an older tool processing a perf.data "
1627 "file generated by a more recent tool?\n\n"
1628 "If that is not the case, consider "
1629 "reporting to linux-kernel@vger.kernel.org.\n\n",
1630 stats->nr_unknown_events);
1631 }
1632
1633 if (stats->nr_unknown_id != 0) {
1634 ui__warning("%u samples with id not present in the header\n",
1635 stats->nr_unknown_id);
1636 }
1637
1638 if (stats->nr_invalid_chains != 0) {
1639 ui__warning("Found invalid callchains!\n\n"
1640 "%u out of %u events were discarded for this reason.\n\n"
1641 "Consider reporting to linux-kernel@vger.kernel.org.\n\n",
1642 stats->nr_invalid_chains,
1643 stats->nr_events[PERF_RECORD_SAMPLE]);
1644 }
1645
1646 if (stats->nr_unprocessable_samples != 0) {
1647 ui__warning("%u unprocessable samples recorded.\n"
1648 "Do you have a KVM guest running and not using 'perf kvm'?\n",
1649 stats->nr_unprocessable_samples);
1650 }
1651
1652 perf_session__warn_order(session);
1653
1654 events_stats__auxtrace_error_warn(stats);
1655
1656 if (stats->nr_proc_map_timeout != 0) {
1657 ui__warning("%d map information files for pre-existing threads were\n"
1658 "not processed, if there are samples for addresses they\n"
1659 "will not be resolved, you may find out which are these\n"
1660 "threads by running with -v and redirecting the output\n"
1661 "to a file.\n"
1662 "The time limit to process proc map is too short?\n"
1663 "Increase it by --proc-map-timeout\n",
1664 stats->nr_proc_map_timeout);
1665 }
1666}
1667
1668static int perf_session__flush_thread_stack(struct thread *thread,
1669 void *p __maybe_unused)
1670{
1671 return thread_stack__flush(thread);
1672}
1673
1674static int perf_session__flush_thread_stacks(struct perf_session *session)
1675{
1676 return machines__for_each_thread(&session->machines,
1677 perf_session__flush_thread_stack,
1678 NULL);
1679}
1680
1681volatile int session_done;
1682
1683static int __perf_session__process_pipe_events(struct perf_session *session)
1684{
1685 struct ordered_events *oe = &session->ordered_events;
1686 struct perf_tool *tool = session->tool;
1687 int fd = perf_data__fd(session->data);
1688 union perf_event *event;
1689 uint32_t size, cur_size = 0;
1690 void *buf = NULL;
1691 s64 skip = 0;
1692 u64 head;
1693 ssize_t err;
1694 void *p;
1695
1696 perf_tool__fill_defaults(tool);
1697
1698 head = 0;
1699 cur_size = sizeof(union perf_event);
1700
1701 buf = malloc(cur_size);
1702 if (!buf)
1703 return -errno;
1704 ordered_events__set_copy_on_queue(oe, true);
1705more:
1706 event = buf;
1707 err = readn(fd, event, sizeof(struct perf_event_header));
1708 if (err <= 0) {
1709 if (err == 0)
1710 goto done;
1711
1712 pr_err("failed to read event header\n");
1713 goto out_err;
1714 }
1715
1716 if (session->header.needs_swap)
1717 perf_event_header__bswap(&event->header);
1718
1719 size = event->header.size;
1720 if (size < sizeof(struct perf_event_header)) {
1721 pr_err("bad event header size\n");
1722 goto out_err;
1723 }
1724
1725 if (size > cur_size) {
1726 void *new = realloc(buf, size);
1727 if (!new) {
1728 pr_err("failed to allocate memory to read event\n");
1729 goto out_err;
1730 }
1731 buf = new;
1732 cur_size = size;
1733 event = buf;
1734 }
1735 p = event;
1736 p += sizeof(struct perf_event_header);
1737
1738 if (size - sizeof(struct perf_event_header)) {
1739 err = readn(fd, p, size - sizeof(struct perf_event_header));
1740 if (err <= 0) {
1741 if (err == 0) {
1742 pr_err("unexpected end of event stream\n");
1743 goto done;
1744 }
1745
1746 pr_err("failed to read event data\n");
1747 goto out_err;
1748 }
1749 }
1750
1751 if ((skip = perf_session__process_event(session, event, head)) < 0) {
1752 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1753 head, event->header.size, event->header.type);
1754 err = -EINVAL;
1755 goto out_err;
1756 }
1757
1758 head += size;
1759
1760 if (skip > 0)
1761 head += skip;
1762
1763 if (!session_done())
1764 goto more;
1765done:
1766 /* do the final flush for ordered samples */
1767 err = ordered_events__flush(oe, OE_FLUSH__FINAL);
1768 if (err)
1769 goto out_err;
1770 err = auxtrace__flush_events(session, tool);
1771 if (err)
1772 goto out_err;
1773 err = perf_session__flush_thread_stacks(session);
1774out_err:
1775 free(buf);
1776 if (!tool->no_warn)
1777 perf_session__warn_about_errors(session);
1778 ordered_events__free(&session->ordered_events);
1779 auxtrace__free_events(session);
1780 return err;
1781}
1782
1783static union perf_event *
1784fetch_mmaped_event(struct perf_session *session,
1785 u64 head, size_t mmap_size, char *buf)
1786{
1787 union perf_event *event;
1788
1789 /*
1790 * Ensure we have enough space remaining to read
1791 * the size of the event in the headers.
1792 */
1793 if (head + sizeof(event->header) > mmap_size)
1794 return NULL;
1795
1796 event = (union perf_event *)(buf + head);
1797
1798 if (session->header.needs_swap)
1799 perf_event_header__bswap(&event->header);
1800
1801 if (head + event->header.size > mmap_size) {
1802 /* We're not fetching the event so swap back again */
1803 if (session->header.needs_swap)
1804 perf_event_header__bswap(&event->header);
1805 return NULL;
1806 }
1807
1808 return event;
1809}
1810
1811/*
1812 * On 64bit we can mmap the data file in one go. No need for tiny mmap
1813 * slices. On 32bit we use 32MB.
1814 */
1815#if BITS_PER_LONG == 64
1816#define MMAP_SIZE ULLONG_MAX
1817#define NUM_MMAPS 1
1818#else
1819#define MMAP_SIZE (32 * 1024 * 1024ULL)
1820#define NUM_MMAPS 128
1821#endif
1822
1823static int __perf_session__process_events(struct perf_session *session,
1824 u64 data_offset, u64 data_size,
1825 u64 file_size)
1826{
1827 struct ordered_events *oe = &session->ordered_events;
1828 struct perf_tool *tool = session->tool;
1829 int fd = perf_data__fd(session->data);
1830 u64 head, page_offset, file_offset, file_pos, size;
1831 int err, mmap_prot, mmap_flags, map_idx = 0;
1832 size_t mmap_size;
1833 char *buf, *mmaps[NUM_MMAPS];
1834 union perf_event *event;
1835 struct ui_progress prog;
1836 s64 skip;
1837
1838 perf_tool__fill_defaults(tool);
1839
1840 page_offset = page_size * (data_offset / page_size);
1841 file_offset = page_offset;
1842 head = data_offset - page_offset;
1843
1844 if (data_size == 0)
1845 goto out;
1846
1847 if (data_offset + data_size < file_size)
1848 file_size = data_offset + data_size;
1849
1850 ui_progress__init_size(&prog, file_size, "Processing events...");
1851
1852 mmap_size = MMAP_SIZE;
1853 if (mmap_size > file_size) {
1854 mmap_size = file_size;
1855 session->one_mmap = true;
1856 }
1857
1858 memset(mmaps, 0, sizeof(mmaps));
1859
1860 mmap_prot = PROT_READ;
1861 mmap_flags = MAP_SHARED;
1862
1863 if (session->header.needs_swap) {
1864 mmap_prot |= PROT_WRITE;
1865 mmap_flags = MAP_PRIVATE;
1866 }
1867remap:
1868 buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, fd,
1869 file_offset);
1870 if (buf == MAP_FAILED) {
1871 pr_err("failed to mmap file\n");
1872 err = -errno;
1873 goto out_err;
1874 }
1875 mmaps[map_idx] = buf;
1876 map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1);
1877 file_pos = file_offset + head;
1878 if (session->one_mmap) {
1879 session->one_mmap_addr = buf;
1880 session->one_mmap_offset = file_offset;
1881 }
1882
1883more:
1884 event = fetch_mmaped_event(session, head, mmap_size, buf);
1885 if (!event) {
1886 if (mmaps[map_idx]) {
1887 munmap(mmaps[map_idx], mmap_size);
1888 mmaps[map_idx] = NULL;
1889 }
1890
1891 page_offset = page_size * (head / page_size);
1892 file_offset += page_offset;
1893 head -= page_offset;
1894 goto remap;
1895 }
1896
1897 size = event->header.size;
1898
1899 if (size < sizeof(struct perf_event_header) ||
1900 (skip = perf_session__process_event(session, event, file_pos)) < 0) {
1901 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1902 file_offset + head, event->header.size,
1903 event->header.type);
1904 err = -EINVAL;
1905 goto out_err;
1906 }
1907
1908 if (skip)
1909 size += skip;
1910
1911 head += size;
1912 file_pos += size;
1913
1914 ui_progress__update(&prog, size);
1915
1916 if (session_done())
1917 goto out;
1918
1919 if (file_pos < file_size)
1920 goto more;
1921
1922out:
1923 /* do the final flush for ordered samples */
1924 err = ordered_events__flush(oe, OE_FLUSH__FINAL);
1925 if (err)
1926 goto out_err;
1927 err = auxtrace__flush_events(session, tool);
1928 if (err)
1929 goto out_err;
1930 err = perf_session__flush_thread_stacks(session);
1931out_err:
1932 ui_progress__finish();
1933 if (!tool->no_warn)
1934 perf_session__warn_about_errors(session);
1935 /*
1936 * We may switching perf.data output, make ordered_events
1937 * reusable.
1938 */
1939 ordered_events__reinit(&session->ordered_events);
1940 auxtrace__free_events(session);
1941 session->one_mmap = false;
1942 return err;
1943}
1944
1945int perf_session__process_events(struct perf_session *session)
1946{
1947 u64 size = perf_data__size(session->data);
1948 int err;
1949
1950 if (perf_session__register_idle_thread(session) < 0)
1951 return -ENOMEM;
1952
1953 if (!perf_data__is_pipe(session->data))
1954 err = __perf_session__process_events(session,
1955 session->header.data_offset,
1956 session->header.data_size, size);
1957 else
1958 err = __perf_session__process_pipe_events(session);
1959
1960 return err;
1961}
1962
1963bool perf_session__has_traces(struct perf_session *session, const char *msg)
1964{
1965 struct perf_evsel *evsel;
1966
1967 evlist__for_each_entry(session->evlist, evsel) {
1968 if (evsel->attr.type == PERF_TYPE_TRACEPOINT)
1969 return true;
1970 }
1971
1972 pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg);
1973 return false;
1974}
1975
1976int maps__set_kallsyms_ref_reloc_sym(struct map **maps,
1977 const char *symbol_name, u64 addr)
1978{
1979 char *bracket;
1980 int i;
1981 struct ref_reloc_sym *ref;
1982
1983 ref = zalloc(sizeof(struct ref_reloc_sym));
1984 if (ref == NULL)
1985 return -ENOMEM;
1986
1987 ref->name = strdup(symbol_name);
1988 if (ref->name == NULL) {
1989 free(ref);
1990 return -ENOMEM;
1991 }
1992
1993 bracket = strchr(ref->name, ']');
1994 if (bracket)
1995 *bracket = '\0';
1996
1997 ref->addr = addr;
1998
1999 for (i = 0; i < MAP__NR_TYPES; ++i) {
2000 struct kmap *kmap = map__kmap(maps[i]);
2001
2002 if (!kmap)
2003 continue;
2004 kmap->ref_reloc_sym = ref;
2005 }
2006
2007 return 0;
2008}
2009
2010size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp)
2011{
2012 return machines__fprintf_dsos(&session->machines, fp);
2013}
2014
2015size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp,
2016 bool (skip)(struct dso *dso, int parm), int parm)
2017{
2018 return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm);
2019}
2020
2021size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp)
2022{
2023 size_t ret;
2024 const char *msg = "";
2025
2026 if (perf_header__has_feat(&session->header, HEADER_AUXTRACE))
2027 msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)";
2028
2029 ret = fprintf(fp, "\nAggregated stats:%s\n", msg);
2030
2031 ret += events_stats__fprintf(&session->evlist->stats, fp);
2032 return ret;
2033}
2034
2035size_t perf_session__fprintf(struct perf_session *session, FILE *fp)
2036{
2037 /*
2038 * FIXME: Here we have to actually print all the machines in this
2039 * session, not just the host...
2040 */
2041 return machine__fprintf(&session->machines.host, fp);
2042}
2043
2044struct perf_evsel *perf_session__find_first_evtype(struct perf_session *session,
2045 unsigned int type)
2046{
2047 struct perf_evsel *pos;
2048
2049 evlist__for_each_entry(session->evlist, pos) {
2050 if (pos->attr.type == type)
2051 return pos;
2052 }
2053 return NULL;
2054}
2055
2056int perf_session__cpu_bitmap(struct perf_session *session,
2057 const char *cpu_list, unsigned long *cpu_bitmap)
2058{
2059 int i, err = -1;
2060 struct cpu_map *map;
2061
2062 for (i = 0; i < PERF_TYPE_MAX; ++i) {
2063 struct perf_evsel *evsel;
2064
2065 evsel = perf_session__find_first_evtype(session, i);
2066 if (!evsel)
2067 continue;
2068
2069 if (!(evsel->attr.sample_type & PERF_SAMPLE_CPU)) {
2070 pr_err("File does not contain CPU events. "
2071 "Remove -C option to proceed.\n");
2072 return -1;
2073 }
2074 }
2075
2076 map = cpu_map__new(cpu_list);
2077 if (map == NULL) {
2078 pr_err("Invalid cpu_list\n");
2079 return -1;
2080 }
2081
2082 for (i = 0; i < map->nr; i++) {
2083 int cpu = map->map[i];
2084
2085 if (cpu >= MAX_NR_CPUS) {
2086 pr_err("Requested CPU %d too large. "
2087 "Consider raising MAX_NR_CPUS\n", cpu);
2088 goto out_delete_map;
2089 }
2090
2091 set_bit(cpu, cpu_bitmap);
2092 }
2093
2094 err = 0;
2095
2096out_delete_map:
2097 cpu_map__put(map);
2098 return err;
2099}
2100
2101void perf_session__fprintf_info(struct perf_session *session, FILE *fp,
2102 bool full)
2103{
2104 if (session == NULL || fp == NULL)
2105 return;
2106
2107 fprintf(fp, "# ========\n");
2108 perf_header__fprintf_info(session, fp, full);
2109 fprintf(fp, "# ========\n#\n");
2110}
2111
2112
2113int __perf_session__set_tracepoints_handlers(struct perf_session *session,
2114 const struct perf_evsel_str_handler *assocs,
2115 size_t nr_assocs)
2116{
2117 struct perf_evsel *evsel;
2118 size_t i;
2119 int err;
2120
2121 for (i = 0; i < nr_assocs; i++) {
2122 /*
2123 * Adding a handler for an event not in the session,
2124 * just ignore it.
2125 */
2126 evsel = perf_evlist__find_tracepoint_by_name(session->evlist, assocs[i].name);
2127 if (evsel == NULL)
2128 continue;
2129
2130 err = -EEXIST;
2131 if (evsel->handler != NULL)
2132 goto out;
2133 evsel->handler = assocs[i].handler;
2134 }
2135
2136 err = 0;
2137out:
2138 return err;
2139}
2140
2141int perf_event__process_id_index(struct perf_tool *tool __maybe_unused,
2142 union perf_event *event,
2143 struct perf_session *session)
2144{
2145 struct perf_evlist *evlist = session->evlist;
2146 struct id_index_event *ie = &event->id_index;
2147 size_t i, nr, max_nr;
2148
2149 max_nr = (ie->header.size - sizeof(struct id_index_event)) /
2150 sizeof(struct id_index_entry);
2151 nr = ie->nr;
2152 if (nr > max_nr)
2153 return -EINVAL;
2154
2155 if (dump_trace)
2156 fprintf(stdout, " nr: %zu\n", nr);
2157
2158 for (i = 0; i < nr; i++) {
2159 struct id_index_entry *e = &ie->entries[i];
2160 struct perf_sample_id *sid;
2161
2162 if (dump_trace) {
2163 fprintf(stdout, " ... id: %"PRIu64, e->id);
2164 fprintf(stdout, " idx: %"PRIu64, e->idx);
2165 fprintf(stdout, " cpu: %"PRId64, e->cpu);
2166 fprintf(stdout, " tid: %"PRId64"\n", e->tid);
2167 }
2168
2169 sid = perf_evlist__id2sid(evlist, e->id);
2170 if (!sid)
2171 return -ENOENT;
2172 sid->idx = e->idx;
2173 sid->cpu = e->cpu;
2174 sid->tid = e->tid;
2175 }
2176 return 0;
2177}
2178
2179int perf_event__synthesize_id_index(struct perf_tool *tool,
2180 perf_event__handler_t process,
2181 struct perf_evlist *evlist,
2182 struct machine *machine)
2183{
2184 union perf_event *ev;
2185 struct perf_evsel *evsel;
2186 size_t nr = 0, i = 0, sz, max_nr, n;
2187 int err;
2188
2189 pr_debug2("Synthesizing id index\n");
2190
2191 max_nr = (UINT16_MAX - sizeof(struct id_index_event)) /
2192 sizeof(struct id_index_entry);
2193
2194 evlist__for_each_entry(evlist, evsel)
2195 nr += evsel->ids;
2196
2197 n = nr > max_nr ? max_nr : nr;
2198 sz = sizeof(struct id_index_event) + n * sizeof(struct id_index_entry);
2199 ev = zalloc(sz);
2200 if (!ev)
2201 return -ENOMEM;
2202
2203 ev->id_index.header.type = PERF_RECORD_ID_INDEX;
2204 ev->id_index.header.size = sz;
2205 ev->id_index.nr = n;
2206
2207 evlist__for_each_entry(evlist, evsel) {
2208 u32 j;
2209
2210 for (j = 0; j < evsel->ids; j++) {
2211 struct id_index_entry *e;
2212 struct perf_sample_id *sid;
2213
2214 if (i >= n) {
2215 err = process(tool, ev, NULL, machine);
2216 if (err)
2217 goto out_err;
2218 nr -= n;
2219 i = 0;
2220 }
2221
2222 e = &ev->id_index.entries[i++];
2223
2224 e->id = evsel->id[j];
2225
2226 sid = perf_evlist__id2sid(evlist, e->id);
2227 if (!sid) {
2228 free(ev);
2229 return -ENOENT;
2230 }
2231
2232 e->idx = sid->idx;
2233 e->cpu = sid->cpu;
2234 e->tid = sid->tid;
2235 }
2236 }
2237
2238 sz = sizeof(struct id_index_event) + nr * sizeof(struct id_index_entry);
2239 ev->id_index.header.size = sz;
2240 ev->id_index.nr = nr;
2241
2242 err = process(tool, ev, NULL, machine);
2243out_err:
2244 free(ev);
2245
2246 return err;
2247}