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 <signal.h>
4#include <inttypes.h>
5#include <linux/err.h>
6#include <linux/kernel.h>
7#include <linux/zalloc.h>
8#include <api/fs/fs.h>
9
10#include <byteswap.h>
11#include <unistd.h>
12#include <sys/types.h>
13#include <sys/mman.h>
14#include <perf/cpumap.h>
15
16#include "map_symbol.h"
17#include "branch.h"
18#include "debug.h"
19#include "env.h"
20#include "evlist.h"
21#include "evsel.h"
22#include "memswap.h"
23#include "map.h"
24#include "symbol.h"
25#include "session.h"
26#include "tool.h"
27#include "perf_regs.h"
28#include "asm/bug.h"
29#include "auxtrace.h"
30#include "thread.h"
31#include "thread-stack.h"
32#include "sample-raw.h"
33#include "stat.h"
34#include "tsc.h"
35#include "ui/progress.h"
36#include "util.h"
37#include "arch/common.h"
38#include "units.h"
39#include <internal/lib.h>
40
41#ifdef HAVE_ZSTD_SUPPORT
42static int perf_session__process_compressed_event(struct perf_session *session,
43 union perf_event *event, u64 file_offset,
44 const char *file_path)
45{
46 void *src;
47 size_t decomp_size, src_size;
48 u64 decomp_last_rem = 0;
49 size_t mmap_len, decomp_len = session->header.env.comp_mmap_len;
50 struct decomp *decomp, *decomp_last = session->active_decomp->decomp_last;
51
52 if (decomp_last) {
53 decomp_last_rem = decomp_last->size - decomp_last->head;
54 decomp_len += decomp_last_rem;
55 }
56
57 mmap_len = sizeof(struct decomp) + decomp_len;
58 decomp = mmap(NULL, mmap_len, PROT_READ|PROT_WRITE,
59 MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);
60 if (decomp == MAP_FAILED) {
61 pr_err("Couldn't allocate memory for decompression\n");
62 return -1;
63 }
64
65 decomp->file_pos = file_offset;
66 decomp->file_path = file_path;
67 decomp->mmap_len = mmap_len;
68 decomp->head = 0;
69
70 if (decomp_last_rem) {
71 memcpy(decomp->data, &(decomp_last->data[decomp_last->head]), decomp_last_rem);
72 decomp->size = decomp_last_rem;
73 }
74
75 src = (void *)event + sizeof(struct perf_record_compressed);
76 src_size = event->pack.header.size - sizeof(struct perf_record_compressed);
77
78 decomp_size = zstd_decompress_stream(session->active_decomp->zstd_decomp, src, src_size,
79 &(decomp->data[decomp_last_rem]), decomp_len - decomp_last_rem);
80 if (!decomp_size) {
81 munmap(decomp, mmap_len);
82 pr_err("Couldn't decompress data\n");
83 return -1;
84 }
85
86 decomp->size += decomp_size;
87
88 if (session->active_decomp->decomp == NULL)
89 session->active_decomp->decomp = decomp;
90 else
91 session->active_decomp->decomp_last->next = decomp;
92
93 session->active_decomp->decomp_last = decomp;
94
95 pr_debug("decomp (B): %zd to %zd\n", src_size, decomp_size);
96
97 return 0;
98}
99#else /* !HAVE_ZSTD_SUPPORT */
100#define perf_session__process_compressed_event perf_session__process_compressed_event_stub
101#endif
102
103static int perf_session__deliver_event(struct perf_session *session,
104 union perf_event *event,
105 struct perf_tool *tool,
106 u64 file_offset,
107 const char *file_path);
108
109static int perf_session__open(struct perf_session *session, int repipe_fd)
110{
111 struct perf_data *data = session->data;
112
113 if (perf_session__read_header(session, repipe_fd) < 0) {
114 pr_err("incompatible file format (rerun with -v to learn more)\n");
115 return -1;
116 }
117
118 if (perf_header__has_feat(&session->header, HEADER_AUXTRACE)) {
119 /* Auxiliary events may reference exited threads, hold onto dead ones. */
120 symbol_conf.keep_exited_threads = true;
121 }
122
123 if (perf_data__is_pipe(data))
124 return 0;
125
126 if (perf_header__has_feat(&session->header, HEADER_STAT))
127 return 0;
128
129 if (!evlist__valid_sample_type(session->evlist)) {
130 pr_err("non matching sample_type\n");
131 return -1;
132 }
133
134 if (!evlist__valid_sample_id_all(session->evlist)) {
135 pr_err("non matching sample_id_all\n");
136 return -1;
137 }
138
139 if (!evlist__valid_read_format(session->evlist)) {
140 pr_err("non matching read_format\n");
141 return -1;
142 }
143
144 return 0;
145}
146
147void perf_session__set_id_hdr_size(struct perf_session *session)
148{
149 u16 id_hdr_size = evlist__id_hdr_size(session->evlist);
150
151 machines__set_id_hdr_size(&session->machines, id_hdr_size);
152}
153
154int perf_session__create_kernel_maps(struct perf_session *session)
155{
156 int ret = machine__create_kernel_maps(&session->machines.host);
157
158 if (ret >= 0)
159 ret = machines__create_guest_kernel_maps(&session->machines);
160 return ret;
161}
162
163static void perf_session__destroy_kernel_maps(struct perf_session *session)
164{
165 machines__destroy_kernel_maps(&session->machines);
166}
167
168static bool perf_session__has_comm_exec(struct perf_session *session)
169{
170 struct evsel *evsel;
171
172 evlist__for_each_entry(session->evlist, evsel) {
173 if (evsel->core.attr.comm_exec)
174 return true;
175 }
176
177 return false;
178}
179
180static void perf_session__set_comm_exec(struct perf_session *session)
181{
182 bool comm_exec = perf_session__has_comm_exec(session);
183
184 machines__set_comm_exec(&session->machines, comm_exec);
185}
186
187static int ordered_events__deliver_event(struct ordered_events *oe,
188 struct ordered_event *event)
189{
190 struct perf_session *session = container_of(oe, struct perf_session,
191 ordered_events);
192
193 return perf_session__deliver_event(session, event->event,
194 session->tool, event->file_offset,
195 event->file_path);
196}
197
198struct perf_session *__perf_session__new(struct perf_data *data,
199 bool repipe, int repipe_fd,
200 struct perf_tool *tool)
201{
202 int ret = -ENOMEM;
203 struct perf_session *session = zalloc(sizeof(*session));
204
205 if (!session)
206 goto out;
207
208 session->repipe = repipe;
209 session->tool = tool;
210 session->decomp_data.zstd_decomp = &session->zstd_data;
211 session->active_decomp = &session->decomp_data;
212 INIT_LIST_HEAD(&session->auxtrace_index);
213 machines__init(&session->machines);
214 ordered_events__init(&session->ordered_events,
215 ordered_events__deliver_event, NULL);
216
217 perf_env__init(&session->header.env);
218 if (data) {
219 ret = perf_data__open(data);
220 if (ret < 0)
221 goto out_delete;
222
223 session->data = data;
224
225 if (perf_data__is_read(data)) {
226 ret = perf_session__open(session, repipe_fd);
227 if (ret < 0)
228 goto out_delete;
229
230 /*
231 * set session attributes that are present in perf.data
232 * but not in pipe-mode.
233 */
234 if (!data->is_pipe) {
235 perf_session__set_id_hdr_size(session);
236 perf_session__set_comm_exec(session);
237 }
238
239 evlist__init_trace_event_sample_raw(session->evlist);
240
241 /* Open the directory data. */
242 if (data->is_dir) {
243 ret = perf_data__open_dir(data);
244 if (ret)
245 goto out_delete;
246 }
247
248 if (!symbol_conf.kallsyms_name &&
249 !symbol_conf.vmlinux_name)
250 symbol_conf.kallsyms_name = perf_data__kallsyms_name(data);
251 }
252 } else {
253 session->machines.host.env = &perf_env;
254 }
255
256 session->machines.host.single_address_space =
257 perf_env__single_address_space(session->machines.host.env);
258
259 if (!data || perf_data__is_write(data)) {
260 /*
261 * In O_RDONLY mode this will be performed when reading the
262 * kernel MMAP event, in perf_event__process_mmap().
263 */
264 if (perf_session__create_kernel_maps(session) < 0)
265 pr_warning("Cannot read kernel map\n");
266 }
267
268 /*
269 * In pipe-mode, evlist is empty until PERF_RECORD_HEADER_ATTR is
270 * processed, so evlist__sample_id_all is not meaningful here.
271 */
272 if ((!data || !data->is_pipe) && tool && tool->ordering_requires_timestamps &&
273 tool->ordered_events && !evlist__sample_id_all(session->evlist)) {
274 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n");
275 tool->ordered_events = false;
276 }
277
278 return session;
279
280 out_delete:
281 perf_session__delete(session);
282 out:
283 return ERR_PTR(ret);
284}
285
286static void perf_decomp__release_events(struct decomp *next)
287{
288 struct decomp *decomp;
289 size_t mmap_len;
290
291 do {
292 decomp = next;
293 if (decomp == NULL)
294 break;
295 next = decomp->next;
296 mmap_len = decomp->mmap_len;
297 munmap(decomp, mmap_len);
298 } while (1);
299}
300
301void perf_session__delete(struct perf_session *session)
302{
303 if (session == NULL)
304 return;
305 auxtrace__free(session);
306 auxtrace_index__free(&session->auxtrace_index);
307 perf_session__destroy_kernel_maps(session);
308 perf_decomp__release_events(session->decomp_data.decomp);
309 perf_env__exit(&session->header.env);
310 machines__exit(&session->machines);
311 if (session->data) {
312 if (perf_data__is_read(session->data))
313 evlist__delete(session->evlist);
314 perf_data__close(session->data);
315 }
316#ifdef HAVE_LIBTRACEEVENT
317 trace_event__cleanup(&session->tevent);
318#endif
319 free(session);
320}
321
322static int process_event_synth_tracing_data_stub(struct perf_session *session
323 __maybe_unused,
324 union perf_event *event
325 __maybe_unused)
326{
327 dump_printf(": unhandled!\n");
328 return 0;
329}
330
331static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused,
332 union perf_event *event __maybe_unused,
333 struct evlist **pevlist
334 __maybe_unused)
335{
336 dump_printf(": unhandled!\n");
337 return 0;
338}
339
340static int process_event_synth_event_update_stub(struct perf_tool *tool __maybe_unused,
341 union perf_event *event __maybe_unused,
342 struct evlist **pevlist
343 __maybe_unused)
344{
345 if (dump_trace)
346 perf_event__fprintf_event_update(event, stdout);
347
348 dump_printf(": unhandled!\n");
349 return 0;
350}
351
352static int process_event_sample_stub(struct perf_tool *tool __maybe_unused,
353 union perf_event *event __maybe_unused,
354 struct perf_sample *sample __maybe_unused,
355 struct evsel *evsel __maybe_unused,
356 struct machine *machine __maybe_unused)
357{
358 dump_printf(": unhandled!\n");
359 return 0;
360}
361
362static int process_event_stub(struct perf_tool *tool __maybe_unused,
363 union perf_event *event __maybe_unused,
364 struct perf_sample *sample __maybe_unused,
365 struct machine *machine __maybe_unused)
366{
367 dump_printf(": unhandled!\n");
368 return 0;
369}
370
371static int process_finished_round_stub(struct perf_tool *tool __maybe_unused,
372 union perf_event *event __maybe_unused,
373 struct ordered_events *oe __maybe_unused)
374{
375 dump_printf(": unhandled!\n");
376 return 0;
377}
378
379static int skipn(int fd, off_t n)
380{
381 char buf[4096];
382 ssize_t ret;
383
384 while (n > 0) {
385 ret = read(fd, buf, min(n, (off_t)sizeof(buf)));
386 if (ret <= 0)
387 return ret;
388 n -= ret;
389 }
390
391 return 0;
392}
393
394static s64 process_event_auxtrace_stub(struct perf_session *session __maybe_unused,
395 union perf_event *event)
396{
397 dump_printf(": unhandled!\n");
398 if (perf_data__is_pipe(session->data))
399 skipn(perf_data__fd(session->data), event->auxtrace.size);
400 return event->auxtrace.size;
401}
402
403static int process_event_op2_stub(struct perf_session *session __maybe_unused,
404 union perf_event *event __maybe_unused)
405{
406 dump_printf(": unhandled!\n");
407 return 0;
408}
409
410
411static
412int process_event_thread_map_stub(struct perf_session *session __maybe_unused,
413 union perf_event *event __maybe_unused)
414{
415 if (dump_trace)
416 perf_event__fprintf_thread_map(event, stdout);
417
418 dump_printf(": unhandled!\n");
419 return 0;
420}
421
422static
423int process_event_cpu_map_stub(struct perf_session *session __maybe_unused,
424 union perf_event *event __maybe_unused)
425{
426 if (dump_trace)
427 perf_event__fprintf_cpu_map(event, stdout);
428
429 dump_printf(": unhandled!\n");
430 return 0;
431}
432
433static
434int process_event_stat_config_stub(struct perf_session *session __maybe_unused,
435 union perf_event *event __maybe_unused)
436{
437 if (dump_trace)
438 perf_event__fprintf_stat_config(event, stdout);
439
440 dump_printf(": unhandled!\n");
441 return 0;
442}
443
444static int process_stat_stub(struct perf_session *perf_session __maybe_unused,
445 union perf_event *event)
446{
447 if (dump_trace)
448 perf_event__fprintf_stat(event, stdout);
449
450 dump_printf(": unhandled!\n");
451 return 0;
452}
453
454static int process_stat_round_stub(struct perf_session *perf_session __maybe_unused,
455 union perf_event *event)
456{
457 if (dump_trace)
458 perf_event__fprintf_stat_round(event, stdout);
459
460 dump_printf(": unhandled!\n");
461 return 0;
462}
463
464static int process_event_time_conv_stub(struct perf_session *perf_session __maybe_unused,
465 union perf_event *event)
466{
467 if (dump_trace)
468 perf_event__fprintf_time_conv(event, stdout);
469
470 dump_printf(": unhandled!\n");
471 return 0;
472}
473
474static int perf_session__process_compressed_event_stub(struct perf_session *session __maybe_unused,
475 union perf_event *event __maybe_unused,
476 u64 file_offset __maybe_unused,
477 const char *file_path __maybe_unused)
478{
479 dump_printf(": unhandled!\n");
480 return 0;
481}
482
483void perf_tool__fill_defaults(struct perf_tool *tool)
484{
485 if (tool->sample == NULL)
486 tool->sample = process_event_sample_stub;
487 if (tool->mmap == NULL)
488 tool->mmap = process_event_stub;
489 if (tool->mmap2 == NULL)
490 tool->mmap2 = process_event_stub;
491 if (tool->comm == NULL)
492 tool->comm = process_event_stub;
493 if (tool->namespaces == NULL)
494 tool->namespaces = process_event_stub;
495 if (tool->cgroup == NULL)
496 tool->cgroup = process_event_stub;
497 if (tool->fork == NULL)
498 tool->fork = process_event_stub;
499 if (tool->exit == NULL)
500 tool->exit = process_event_stub;
501 if (tool->lost == NULL)
502 tool->lost = perf_event__process_lost;
503 if (tool->lost_samples == NULL)
504 tool->lost_samples = perf_event__process_lost_samples;
505 if (tool->aux == NULL)
506 tool->aux = perf_event__process_aux;
507 if (tool->itrace_start == NULL)
508 tool->itrace_start = perf_event__process_itrace_start;
509 if (tool->context_switch == NULL)
510 tool->context_switch = perf_event__process_switch;
511 if (tool->ksymbol == NULL)
512 tool->ksymbol = perf_event__process_ksymbol;
513 if (tool->bpf == NULL)
514 tool->bpf = perf_event__process_bpf;
515 if (tool->text_poke == NULL)
516 tool->text_poke = perf_event__process_text_poke;
517 if (tool->aux_output_hw_id == NULL)
518 tool->aux_output_hw_id = perf_event__process_aux_output_hw_id;
519 if (tool->read == NULL)
520 tool->read = process_event_sample_stub;
521 if (tool->throttle == NULL)
522 tool->throttle = process_event_stub;
523 if (tool->unthrottle == NULL)
524 tool->unthrottle = process_event_stub;
525 if (tool->attr == NULL)
526 tool->attr = process_event_synth_attr_stub;
527 if (tool->event_update == NULL)
528 tool->event_update = process_event_synth_event_update_stub;
529 if (tool->tracing_data == NULL)
530 tool->tracing_data = process_event_synth_tracing_data_stub;
531 if (tool->build_id == NULL)
532 tool->build_id = process_event_op2_stub;
533 if (tool->finished_round == NULL) {
534 if (tool->ordered_events)
535 tool->finished_round = perf_event__process_finished_round;
536 else
537 tool->finished_round = process_finished_round_stub;
538 }
539 if (tool->id_index == NULL)
540 tool->id_index = process_event_op2_stub;
541 if (tool->auxtrace_info == NULL)
542 tool->auxtrace_info = process_event_op2_stub;
543 if (tool->auxtrace == NULL)
544 tool->auxtrace = process_event_auxtrace_stub;
545 if (tool->auxtrace_error == NULL)
546 tool->auxtrace_error = process_event_op2_stub;
547 if (tool->thread_map == NULL)
548 tool->thread_map = process_event_thread_map_stub;
549 if (tool->cpu_map == NULL)
550 tool->cpu_map = process_event_cpu_map_stub;
551 if (tool->stat_config == NULL)
552 tool->stat_config = process_event_stat_config_stub;
553 if (tool->stat == NULL)
554 tool->stat = process_stat_stub;
555 if (tool->stat_round == NULL)
556 tool->stat_round = process_stat_round_stub;
557 if (tool->time_conv == NULL)
558 tool->time_conv = process_event_time_conv_stub;
559 if (tool->feature == NULL)
560 tool->feature = process_event_op2_stub;
561 if (tool->compressed == NULL)
562 tool->compressed = perf_session__process_compressed_event;
563 if (tool->finished_init == NULL)
564 tool->finished_init = process_event_op2_stub;
565}
566
567static void swap_sample_id_all(union perf_event *event, void *data)
568{
569 void *end = (void *) event + event->header.size;
570 int size = end - data;
571
572 BUG_ON(size % sizeof(u64));
573 mem_bswap_64(data, size);
574}
575
576static void perf_event__all64_swap(union perf_event *event,
577 bool sample_id_all __maybe_unused)
578{
579 struct perf_event_header *hdr = &event->header;
580 mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr));
581}
582
583static void perf_event__comm_swap(union perf_event *event, bool sample_id_all)
584{
585 event->comm.pid = bswap_32(event->comm.pid);
586 event->comm.tid = bswap_32(event->comm.tid);
587
588 if (sample_id_all) {
589 void *data = &event->comm.comm;
590
591 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
592 swap_sample_id_all(event, data);
593 }
594}
595
596static void perf_event__mmap_swap(union perf_event *event,
597 bool sample_id_all)
598{
599 event->mmap.pid = bswap_32(event->mmap.pid);
600 event->mmap.tid = bswap_32(event->mmap.tid);
601 event->mmap.start = bswap_64(event->mmap.start);
602 event->mmap.len = bswap_64(event->mmap.len);
603 event->mmap.pgoff = bswap_64(event->mmap.pgoff);
604
605 if (sample_id_all) {
606 void *data = &event->mmap.filename;
607
608 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
609 swap_sample_id_all(event, data);
610 }
611}
612
613static void perf_event__mmap2_swap(union perf_event *event,
614 bool sample_id_all)
615{
616 event->mmap2.pid = bswap_32(event->mmap2.pid);
617 event->mmap2.tid = bswap_32(event->mmap2.tid);
618 event->mmap2.start = bswap_64(event->mmap2.start);
619 event->mmap2.len = bswap_64(event->mmap2.len);
620 event->mmap2.pgoff = bswap_64(event->mmap2.pgoff);
621
622 if (!(event->header.misc & PERF_RECORD_MISC_MMAP_BUILD_ID)) {
623 event->mmap2.maj = bswap_32(event->mmap2.maj);
624 event->mmap2.min = bswap_32(event->mmap2.min);
625 event->mmap2.ino = bswap_64(event->mmap2.ino);
626 event->mmap2.ino_generation = bswap_64(event->mmap2.ino_generation);
627 }
628
629 if (sample_id_all) {
630 void *data = &event->mmap2.filename;
631
632 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
633 swap_sample_id_all(event, data);
634 }
635}
636static void perf_event__task_swap(union perf_event *event, bool sample_id_all)
637{
638 event->fork.pid = bswap_32(event->fork.pid);
639 event->fork.tid = bswap_32(event->fork.tid);
640 event->fork.ppid = bswap_32(event->fork.ppid);
641 event->fork.ptid = bswap_32(event->fork.ptid);
642 event->fork.time = bswap_64(event->fork.time);
643
644 if (sample_id_all)
645 swap_sample_id_all(event, &event->fork + 1);
646}
647
648static void perf_event__read_swap(union perf_event *event, bool sample_id_all)
649{
650 event->read.pid = bswap_32(event->read.pid);
651 event->read.tid = bswap_32(event->read.tid);
652 event->read.value = bswap_64(event->read.value);
653 event->read.time_enabled = bswap_64(event->read.time_enabled);
654 event->read.time_running = bswap_64(event->read.time_running);
655 event->read.id = bswap_64(event->read.id);
656
657 if (sample_id_all)
658 swap_sample_id_all(event, &event->read + 1);
659}
660
661static void perf_event__aux_swap(union perf_event *event, bool sample_id_all)
662{
663 event->aux.aux_offset = bswap_64(event->aux.aux_offset);
664 event->aux.aux_size = bswap_64(event->aux.aux_size);
665 event->aux.flags = bswap_64(event->aux.flags);
666
667 if (sample_id_all)
668 swap_sample_id_all(event, &event->aux + 1);
669}
670
671static void perf_event__itrace_start_swap(union perf_event *event,
672 bool sample_id_all)
673{
674 event->itrace_start.pid = bswap_32(event->itrace_start.pid);
675 event->itrace_start.tid = bswap_32(event->itrace_start.tid);
676
677 if (sample_id_all)
678 swap_sample_id_all(event, &event->itrace_start + 1);
679}
680
681static void perf_event__switch_swap(union perf_event *event, bool sample_id_all)
682{
683 if (event->header.type == PERF_RECORD_SWITCH_CPU_WIDE) {
684 event->context_switch.next_prev_pid =
685 bswap_32(event->context_switch.next_prev_pid);
686 event->context_switch.next_prev_tid =
687 bswap_32(event->context_switch.next_prev_tid);
688 }
689
690 if (sample_id_all)
691 swap_sample_id_all(event, &event->context_switch + 1);
692}
693
694static void perf_event__text_poke_swap(union perf_event *event, bool sample_id_all)
695{
696 event->text_poke.addr = bswap_64(event->text_poke.addr);
697 event->text_poke.old_len = bswap_16(event->text_poke.old_len);
698 event->text_poke.new_len = bswap_16(event->text_poke.new_len);
699
700 if (sample_id_all) {
701 size_t len = sizeof(event->text_poke.old_len) +
702 sizeof(event->text_poke.new_len) +
703 event->text_poke.old_len +
704 event->text_poke.new_len;
705 void *data = &event->text_poke.old_len;
706
707 data += PERF_ALIGN(len, sizeof(u64));
708 swap_sample_id_all(event, data);
709 }
710}
711
712static void perf_event__throttle_swap(union perf_event *event,
713 bool sample_id_all)
714{
715 event->throttle.time = bswap_64(event->throttle.time);
716 event->throttle.id = bswap_64(event->throttle.id);
717 event->throttle.stream_id = bswap_64(event->throttle.stream_id);
718
719 if (sample_id_all)
720 swap_sample_id_all(event, &event->throttle + 1);
721}
722
723static void perf_event__namespaces_swap(union perf_event *event,
724 bool sample_id_all)
725{
726 u64 i;
727
728 event->namespaces.pid = bswap_32(event->namespaces.pid);
729 event->namespaces.tid = bswap_32(event->namespaces.tid);
730 event->namespaces.nr_namespaces = bswap_64(event->namespaces.nr_namespaces);
731
732 for (i = 0; i < event->namespaces.nr_namespaces; i++) {
733 struct perf_ns_link_info *ns = &event->namespaces.link_info[i];
734
735 ns->dev = bswap_64(ns->dev);
736 ns->ino = bswap_64(ns->ino);
737 }
738
739 if (sample_id_all)
740 swap_sample_id_all(event, &event->namespaces.link_info[i]);
741}
742
743static void perf_event__cgroup_swap(union perf_event *event, bool sample_id_all)
744{
745 event->cgroup.id = bswap_64(event->cgroup.id);
746
747 if (sample_id_all) {
748 void *data = &event->cgroup.path;
749
750 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
751 swap_sample_id_all(event, data);
752 }
753}
754
755static u8 revbyte(u8 b)
756{
757 int rev = (b >> 4) | ((b & 0xf) << 4);
758 rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2);
759 rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1);
760 return (u8) rev;
761}
762
763/*
764 * XXX this is hack in attempt to carry flags bitfield
765 * through endian village. ABI says:
766 *
767 * Bit-fields are allocated from right to left (least to most significant)
768 * on little-endian implementations and from left to right (most to least
769 * significant) on big-endian implementations.
770 *
771 * The above seems to be byte specific, so we need to reverse each
772 * byte of the bitfield. 'Internet' also says this might be implementation
773 * specific and we probably need proper fix and carry perf_event_attr
774 * bitfield flags in separate data file FEAT_ section. Thought this seems
775 * to work for now.
776 */
777static void swap_bitfield(u8 *p, unsigned len)
778{
779 unsigned i;
780
781 for (i = 0; i < len; i++) {
782 *p = revbyte(*p);
783 p++;
784 }
785}
786
787/* exported for swapping attributes in file header */
788void perf_event__attr_swap(struct perf_event_attr *attr)
789{
790 attr->type = bswap_32(attr->type);
791 attr->size = bswap_32(attr->size);
792
793#define bswap_safe(f, n) \
794 (attr->size > (offsetof(struct perf_event_attr, f) + \
795 sizeof(attr->f) * (n)))
796#define bswap_field(f, sz) \
797do { \
798 if (bswap_safe(f, 0)) \
799 attr->f = bswap_##sz(attr->f); \
800} while(0)
801#define bswap_field_16(f) bswap_field(f, 16)
802#define bswap_field_32(f) bswap_field(f, 32)
803#define bswap_field_64(f) bswap_field(f, 64)
804
805 bswap_field_64(config);
806 bswap_field_64(sample_period);
807 bswap_field_64(sample_type);
808 bswap_field_64(read_format);
809 bswap_field_32(wakeup_events);
810 bswap_field_32(bp_type);
811 bswap_field_64(bp_addr);
812 bswap_field_64(bp_len);
813 bswap_field_64(branch_sample_type);
814 bswap_field_64(sample_regs_user);
815 bswap_field_32(sample_stack_user);
816 bswap_field_32(aux_watermark);
817 bswap_field_16(sample_max_stack);
818 bswap_field_32(aux_sample_size);
819
820 /*
821 * After read_format are bitfields. Check read_format because
822 * we are unable to use offsetof on bitfield.
823 */
824 if (bswap_safe(read_format, 1))
825 swap_bitfield((u8 *) (&attr->read_format + 1),
826 sizeof(u64));
827#undef bswap_field_64
828#undef bswap_field_32
829#undef bswap_field
830#undef bswap_safe
831}
832
833static void perf_event__hdr_attr_swap(union perf_event *event,
834 bool sample_id_all __maybe_unused)
835{
836 size_t size;
837
838 perf_event__attr_swap(&event->attr.attr);
839
840 size = event->header.size;
841 size -= perf_record_header_attr_id(event) - (void *)event;
842 mem_bswap_64(perf_record_header_attr_id(event), size);
843}
844
845static void perf_event__event_update_swap(union perf_event *event,
846 bool sample_id_all __maybe_unused)
847{
848 event->event_update.type = bswap_64(event->event_update.type);
849 event->event_update.id = bswap_64(event->event_update.id);
850}
851
852static void perf_event__event_type_swap(union perf_event *event,
853 bool sample_id_all __maybe_unused)
854{
855 event->event_type.event_type.event_id =
856 bswap_64(event->event_type.event_type.event_id);
857}
858
859static void perf_event__tracing_data_swap(union perf_event *event,
860 bool sample_id_all __maybe_unused)
861{
862 event->tracing_data.size = bswap_32(event->tracing_data.size);
863}
864
865static void perf_event__auxtrace_info_swap(union perf_event *event,
866 bool sample_id_all __maybe_unused)
867{
868 size_t size;
869
870 event->auxtrace_info.type = bswap_32(event->auxtrace_info.type);
871
872 size = event->header.size;
873 size -= (void *)&event->auxtrace_info.priv - (void *)event;
874 mem_bswap_64(event->auxtrace_info.priv, size);
875}
876
877static void perf_event__auxtrace_swap(union perf_event *event,
878 bool sample_id_all __maybe_unused)
879{
880 event->auxtrace.size = bswap_64(event->auxtrace.size);
881 event->auxtrace.offset = bswap_64(event->auxtrace.offset);
882 event->auxtrace.reference = bswap_64(event->auxtrace.reference);
883 event->auxtrace.idx = bswap_32(event->auxtrace.idx);
884 event->auxtrace.tid = bswap_32(event->auxtrace.tid);
885 event->auxtrace.cpu = bswap_32(event->auxtrace.cpu);
886}
887
888static void perf_event__auxtrace_error_swap(union perf_event *event,
889 bool sample_id_all __maybe_unused)
890{
891 event->auxtrace_error.type = bswap_32(event->auxtrace_error.type);
892 event->auxtrace_error.code = bswap_32(event->auxtrace_error.code);
893 event->auxtrace_error.cpu = bswap_32(event->auxtrace_error.cpu);
894 event->auxtrace_error.pid = bswap_32(event->auxtrace_error.pid);
895 event->auxtrace_error.tid = bswap_32(event->auxtrace_error.tid);
896 event->auxtrace_error.fmt = bswap_32(event->auxtrace_error.fmt);
897 event->auxtrace_error.ip = bswap_64(event->auxtrace_error.ip);
898 if (event->auxtrace_error.fmt)
899 event->auxtrace_error.time = bswap_64(event->auxtrace_error.time);
900 if (event->auxtrace_error.fmt >= 2) {
901 event->auxtrace_error.machine_pid = bswap_32(event->auxtrace_error.machine_pid);
902 event->auxtrace_error.vcpu = bswap_32(event->auxtrace_error.vcpu);
903 }
904}
905
906static void perf_event__thread_map_swap(union perf_event *event,
907 bool sample_id_all __maybe_unused)
908{
909 unsigned i;
910
911 event->thread_map.nr = bswap_64(event->thread_map.nr);
912
913 for (i = 0; i < event->thread_map.nr; i++)
914 event->thread_map.entries[i].pid = bswap_64(event->thread_map.entries[i].pid);
915}
916
917static void perf_event__cpu_map_swap(union perf_event *event,
918 bool sample_id_all __maybe_unused)
919{
920 struct perf_record_cpu_map_data *data = &event->cpu_map.data;
921
922 data->type = bswap_16(data->type);
923
924 switch (data->type) {
925 case PERF_CPU_MAP__CPUS:
926 data->cpus_data.nr = bswap_16(data->cpus_data.nr);
927
928 for (unsigned i = 0; i < data->cpus_data.nr; i++)
929 data->cpus_data.cpu[i] = bswap_16(data->cpus_data.cpu[i]);
930 break;
931 case PERF_CPU_MAP__MASK:
932 data->mask32_data.long_size = bswap_16(data->mask32_data.long_size);
933
934 switch (data->mask32_data.long_size) {
935 case 4:
936 data->mask32_data.nr = bswap_16(data->mask32_data.nr);
937 for (unsigned i = 0; i < data->mask32_data.nr; i++)
938 data->mask32_data.mask[i] = bswap_32(data->mask32_data.mask[i]);
939 break;
940 case 8:
941 data->mask64_data.nr = bswap_16(data->mask64_data.nr);
942 for (unsigned i = 0; i < data->mask64_data.nr; i++)
943 data->mask64_data.mask[i] = bswap_64(data->mask64_data.mask[i]);
944 break;
945 default:
946 pr_err("cpu_map swap: unsupported long size\n");
947 }
948 break;
949 case PERF_CPU_MAP__RANGE_CPUS:
950 data->range_cpu_data.start_cpu = bswap_16(data->range_cpu_data.start_cpu);
951 data->range_cpu_data.end_cpu = bswap_16(data->range_cpu_data.end_cpu);
952 break;
953 default:
954 break;
955 }
956}
957
958static void perf_event__stat_config_swap(union perf_event *event,
959 bool sample_id_all __maybe_unused)
960{
961 u64 size;
962
963 size = bswap_64(event->stat_config.nr) * sizeof(event->stat_config.data[0]);
964 size += 1; /* nr item itself */
965 mem_bswap_64(&event->stat_config.nr, size);
966}
967
968static void perf_event__stat_swap(union perf_event *event,
969 bool sample_id_all __maybe_unused)
970{
971 event->stat.id = bswap_64(event->stat.id);
972 event->stat.thread = bswap_32(event->stat.thread);
973 event->stat.cpu = bswap_32(event->stat.cpu);
974 event->stat.val = bswap_64(event->stat.val);
975 event->stat.ena = bswap_64(event->stat.ena);
976 event->stat.run = bswap_64(event->stat.run);
977}
978
979static void perf_event__stat_round_swap(union perf_event *event,
980 bool sample_id_all __maybe_unused)
981{
982 event->stat_round.type = bswap_64(event->stat_round.type);
983 event->stat_round.time = bswap_64(event->stat_round.time);
984}
985
986static void perf_event__time_conv_swap(union perf_event *event,
987 bool sample_id_all __maybe_unused)
988{
989 event->time_conv.time_shift = bswap_64(event->time_conv.time_shift);
990 event->time_conv.time_mult = bswap_64(event->time_conv.time_mult);
991 event->time_conv.time_zero = bswap_64(event->time_conv.time_zero);
992
993 if (event_contains(event->time_conv, time_cycles)) {
994 event->time_conv.time_cycles = bswap_64(event->time_conv.time_cycles);
995 event->time_conv.time_mask = bswap_64(event->time_conv.time_mask);
996 }
997}
998
999typedef void (*perf_event__swap_op)(union perf_event *event,
1000 bool sample_id_all);
1001
1002static perf_event__swap_op perf_event__swap_ops[] = {
1003 [PERF_RECORD_MMAP] = perf_event__mmap_swap,
1004 [PERF_RECORD_MMAP2] = perf_event__mmap2_swap,
1005 [PERF_RECORD_COMM] = perf_event__comm_swap,
1006 [PERF_RECORD_FORK] = perf_event__task_swap,
1007 [PERF_RECORD_EXIT] = perf_event__task_swap,
1008 [PERF_RECORD_LOST] = perf_event__all64_swap,
1009 [PERF_RECORD_READ] = perf_event__read_swap,
1010 [PERF_RECORD_THROTTLE] = perf_event__throttle_swap,
1011 [PERF_RECORD_UNTHROTTLE] = perf_event__throttle_swap,
1012 [PERF_RECORD_SAMPLE] = perf_event__all64_swap,
1013 [PERF_RECORD_AUX] = perf_event__aux_swap,
1014 [PERF_RECORD_ITRACE_START] = perf_event__itrace_start_swap,
1015 [PERF_RECORD_LOST_SAMPLES] = perf_event__all64_swap,
1016 [PERF_RECORD_SWITCH] = perf_event__switch_swap,
1017 [PERF_RECORD_SWITCH_CPU_WIDE] = perf_event__switch_swap,
1018 [PERF_RECORD_NAMESPACES] = perf_event__namespaces_swap,
1019 [PERF_RECORD_CGROUP] = perf_event__cgroup_swap,
1020 [PERF_RECORD_TEXT_POKE] = perf_event__text_poke_swap,
1021 [PERF_RECORD_AUX_OUTPUT_HW_ID] = perf_event__all64_swap,
1022 [PERF_RECORD_HEADER_ATTR] = perf_event__hdr_attr_swap,
1023 [PERF_RECORD_HEADER_EVENT_TYPE] = perf_event__event_type_swap,
1024 [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
1025 [PERF_RECORD_HEADER_BUILD_ID] = NULL,
1026 [PERF_RECORD_ID_INDEX] = perf_event__all64_swap,
1027 [PERF_RECORD_AUXTRACE_INFO] = perf_event__auxtrace_info_swap,
1028 [PERF_RECORD_AUXTRACE] = perf_event__auxtrace_swap,
1029 [PERF_RECORD_AUXTRACE_ERROR] = perf_event__auxtrace_error_swap,
1030 [PERF_RECORD_THREAD_MAP] = perf_event__thread_map_swap,
1031 [PERF_RECORD_CPU_MAP] = perf_event__cpu_map_swap,
1032 [PERF_RECORD_STAT_CONFIG] = perf_event__stat_config_swap,
1033 [PERF_RECORD_STAT] = perf_event__stat_swap,
1034 [PERF_RECORD_STAT_ROUND] = perf_event__stat_round_swap,
1035 [PERF_RECORD_EVENT_UPDATE] = perf_event__event_update_swap,
1036 [PERF_RECORD_TIME_CONV] = perf_event__time_conv_swap,
1037 [PERF_RECORD_HEADER_MAX] = NULL,
1038};
1039
1040/*
1041 * When perf record finishes a pass on every buffers, it records this pseudo
1042 * event.
1043 * We record the max timestamp t found in the pass n.
1044 * Assuming these timestamps are monotonic across cpus, we know that if
1045 * a buffer still has events with timestamps below t, they will be all
1046 * available and then read in the pass n + 1.
1047 * Hence when we start to read the pass n + 2, we can safely flush every
1048 * events with timestamps below t.
1049 *
1050 * ============ PASS n =================
1051 * CPU 0 | CPU 1
1052 * |
1053 * cnt1 timestamps | cnt2 timestamps
1054 * 1 | 2
1055 * 2 | 3
1056 * - | 4 <--- max recorded
1057 *
1058 * ============ PASS n + 1 ==============
1059 * CPU 0 | CPU 1
1060 * |
1061 * cnt1 timestamps | cnt2 timestamps
1062 * 3 | 5
1063 * 4 | 6
1064 * 5 | 7 <---- max recorded
1065 *
1066 * Flush every events below timestamp 4
1067 *
1068 * ============ PASS n + 2 ==============
1069 * CPU 0 | CPU 1
1070 * |
1071 * cnt1 timestamps | cnt2 timestamps
1072 * 6 | 8
1073 * 7 | 9
1074 * - | 10
1075 *
1076 * Flush every events below timestamp 7
1077 * etc...
1078 */
1079int perf_event__process_finished_round(struct perf_tool *tool __maybe_unused,
1080 union perf_event *event __maybe_unused,
1081 struct ordered_events *oe)
1082{
1083 if (dump_trace)
1084 fprintf(stdout, "\n");
1085 return ordered_events__flush(oe, OE_FLUSH__ROUND);
1086}
1087
1088int perf_session__queue_event(struct perf_session *s, union perf_event *event,
1089 u64 timestamp, u64 file_offset, const char *file_path)
1090{
1091 return ordered_events__queue(&s->ordered_events, event, timestamp, file_offset, file_path);
1092}
1093
1094static void callchain__lbr_callstack_printf(struct perf_sample *sample)
1095{
1096 struct ip_callchain *callchain = sample->callchain;
1097 struct branch_stack *lbr_stack = sample->branch_stack;
1098 struct branch_entry *entries = perf_sample__branch_entries(sample);
1099 u64 kernel_callchain_nr = callchain->nr;
1100 unsigned int i;
1101
1102 for (i = 0; i < kernel_callchain_nr; i++) {
1103 if (callchain->ips[i] == PERF_CONTEXT_USER)
1104 break;
1105 }
1106
1107 if ((i != kernel_callchain_nr) && lbr_stack->nr) {
1108 u64 total_nr;
1109 /*
1110 * LBR callstack can only get user call chain,
1111 * i is kernel call chain number,
1112 * 1 is PERF_CONTEXT_USER.
1113 *
1114 * The user call chain is stored in LBR registers.
1115 * LBR are pair registers. The caller is stored
1116 * in "from" register, while the callee is stored
1117 * in "to" register.
1118 * For example, there is a call stack
1119 * "A"->"B"->"C"->"D".
1120 * The LBR registers will be recorded like
1121 * "C"->"D", "B"->"C", "A"->"B".
1122 * So only the first "to" register and all "from"
1123 * registers are needed to construct the whole stack.
1124 */
1125 total_nr = i + 1 + lbr_stack->nr + 1;
1126 kernel_callchain_nr = i + 1;
1127
1128 printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr);
1129
1130 for (i = 0; i < kernel_callchain_nr; i++)
1131 printf("..... %2d: %016" PRIx64 "\n",
1132 i, callchain->ips[i]);
1133
1134 printf("..... %2d: %016" PRIx64 "\n",
1135 (int)(kernel_callchain_nr), entries[0].to);
1136 for (i = 0; i < lbr_stack->nr; i++)
1137 printf("..... %2d: %016" PRIx64 "\n",
1138 (int)(i + kernel_callchain_nr + 1), entries[i].from);
1139 }
1140}
1141
1142static void callchain__printf(struct evsel *evsel,
1143 struct perf_sample *sample)
1144{
1145 unsigned int i;
1146 struct ip_callchain *callchain = sample->callchain;
1147
1148 if (evsel__has_branch_callstack(evsel))
1149 callchain__lbr_callstack_printf(sample);
1150
1151 printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr);
1152
1153 for (i = 0; i < callchain->nr; i++)
1154 printf("..... %2d: %016" PRIx64 "\n",
1155 i, callchain->ips[i]);
1156}
1157
1158static void branch_stack__printf(struct perf_sample *sample,
1159 struct evsel *evsel)
1160{
1161 struct branch_entry *entries = perf_sample__branch_entries(sample);
1162 bool callstack = evsel__has_branch_callstack(evsel);
1163 u64 *branch_stack_cntr = sample->branch_stack_cntr;
1164 struct perf_env *env = evsel__env(evsel);
1165 uint64_t i;
1166
1167 if (!callstack) {
1168 printf("%s: nr:%" PRIu64 "\n", "... branch stack", sample->branch_stack->nr);
1169 } else {
1170 /* the reason of adding 1 to nr is because after expanding
1171 * branch stack it generates nr + 1 callstack records. e.g.,
1172 * B()->C()
1173 * A()->B()
1174 * the final callstack should be:
1175 * C()
1176 * B()
1177 * A()
1178 */
1179 printf("%s: nr:%" PRIu64 "\n", "... branch callstack", sample->branch_stack->nr+1);
1180 }
1181
1182 for (i = 0; i < sample->branch_stack->nr; i++) {
1183 struct branch_entry *e = &entries[i];
1184
1185 if (!callstack) {
1186 printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x %s %s\n",
1187 i, e->from, e->to,
1188 (unsigned short)e->flags.cycles,
1189 e->flags.mispred ? "M" : " ",
1190 e->flags.predicted ? "P" : " ",
1191 e->flags.abort ? "A" : " ",
1192 e->flags.in_tx ? "T" : " ",
1193 (unsigned)e->flags.reserved,
1194 get_branch_type(e),
1195 e->flags.spec ? branch_spec_desc(e->flags.spec) : "");
1196 } else {
1197 if (i == 0) {
1198 printf("..... %2"PRIu64": %016" PRIx64 "\n"
1199 "..... %2"PRIu64": %016" PRIx64 "\n",
1200 i, e->to, i+1, e->from);
1201 } else {
1202 printf("..... %2"PRIu64": %016" PRIx64 "\n", i+1, e->from);
1203 }
1204 }
1205 }
1206
1207 if (branch_stack_cntr) {
1208 printf("... branch stack counters: nr:%" PRIu64 " (counter width: %u max counter nr:%u)\n",
1209 sample->branch_stack->nr, env->br_cntr_width, env->br_cntr_nr);
1210 for (i = 0; i < sample->branch_stack->nr; i++)
1211 printf("..... %2"PRIu64": %016" PRIx64 "\n", i, branch_stack_cntr[i]);
1212 }
1213}
1214
1215static void regs_dump__printf(u64 mask, u64 *regs, const char *arch)
1216{
1217 unsigned rid, i = 0;
1218
1219 for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) {
1220 u64 val = regs[i++];
1221
1222 printf(".... %-5s 0x%016" PRIx64 "\n",
1223 perf_reg_name(rid, arch), val);
1224 }
1225}
1226
1227static const char *regs_abi[] = {
1228 [PERF_SAMPLE_REGS_ABI_NONE] = "none",
1229 [PERF_SAMPLE_REGS_ABI_32] = "32-bit",
1230 [PERF_SAMPLE_REGS_ABI_64] = "64-bit",
1231};
1232
1233static inline const char *regs_dump_abi(struct regs_dump *d)
1234{
1235 if (d->abi > PERF_SAMPLE_REGS_ABI_64)
1236 return "unknown";
1237
1238 return regs_abi[d->abi];
1239}
1240
1241static void regs__printf(const char *type, struct regs_dump *regs, const char *arch)
1242{
1243 u64 mask = regs->mask;
1244
1245 printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n",
1246 type,
1247 mask,
1248 regs_dump_abi(regs));
1249
1250 regs_dump__printf(mask, regs->regs, arch);
1251}
1252
1253static void regs_user__printf(struct perf_sample *sample, const char *arch)
1254{
1255 struct regs_dump *user_regs = &sample->user_regs;
1256
1257 if (user_regs->regs)
1258 regs__printf("user", user_regs, arch);
1259}
1260
1261static void regs_intr__printf(struct perf_sample *sample, const char *arch)
1262{
1263 struct regs_dump *intr_regs = &sample->intr_regs;
1264
1265 if (intr_regs->regs)
1266 regs__printf("intr", intr_regs, arch);
1267}
1268
1269static void stack_user__printf(struct stack_dump *dump)
1270{
1271 printf("... ustack: size %" PRIu64 ", offset 0x%x\n",
1272 dump->size, dump->offset);
1273}
1274
1275static void evlist__print_tstamp(struct evlist *evlist, union perf_event *event, struct perf_sample *sample)
1276{
1277 u64 sample_type = __evlist__combined_sample_type(evlist);
1278
1279 if (event->header.type != PERF_RECORD_SAMPLE &&
1280 !evlist__sample_id_all(evlist)) {
1281 fputs("-1 -1 ", stdout);
1282 return;
1283 }
1284
1285 if ((sample_type & PERF_SAMPLE_CPU))
1286 printf("%u ", sample->cpu);
1287
1288 if (sample_type & PERF_SAMPLE_TIME)
1289 printf("%" PRIu64 " ", sample->time);
1290}
1291
1292static void sample_read__printf(struct perf_sample *sample, u64 read_format)
1293{
1294 printf("... sample_read:\n");
1295
1296 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1297 printf("...... time enabled %016" PRIx64 "\n",
1298 sample->read.time_enabled);
1299
1300 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1301 printf("...... time running %016" PRIx64 "\n",
1302 sample->read.time_running);
1303
1304 if (read_format & PERF_FORMAT_GROUP) {
1305 struct sample_read_value *value = sample->read.group.values;
1306
1307 printf(".... group nr %" PRIu64 "\n", sample->read.group.nr);
1308
1309 sample_read_group__for_each(value, sample->read.group.nr, read_format) {
1310 printf("..... id %016" PRIx64
1311 ", value %016" PRIx64,
1312 value->id, value->value);
1313 if (read_format & PERF_FORMAT_LOST)
1314 printf(", lost %" PRIu64, value->lost);
1315 printf("\n");
1316 }
1317 } else {
1318 printf("..... id %016" PRIx64 ", value %016" PRIx64,
1319 sample->read.one.id, sample->read.one.value);
1320 if (read_format & PERF_FORMAT_LOST)
1321 printf(", lost %" PRIu64, sample->read.one.lost);
1322 printf("\n");
1323 }
1324}
1325
1326static void dump_event(struct evlist *evlist, union perf_event *event,
1327 u64 file_offset, struct perf_sample *sample,
1328 const char *file_path)
1329{
1330 if (!dump_trace)
1331 return;
1332
1333 printf("\n%#" PRIx64 "@%s [%#x]: event: %d\n",
1334 file_offset, file_path, event->header.size, event->header.type);
1335
1336 trace_event(event);
1337 if (event->header.type == PERF_RECORD_SAMPLE && evlist->trace_event_sample_raw)
1338 evlist->trace_event_sample_raw(evlist, event, sample);
1339
1340 if (sample)
1341 evlist__print_tstamp(evlist, event, sample);
1342
1343 printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset,
1344 event->header.size, perf_event__name(event->header.type));
1345}
1346
1347char *get_page_size_name(u64 size, char *str)
1348{
1349 if (!size || !unit_number__scnprintf(str, PAGE_SIZE_NAME_LEN, size))
1350 snprintf(str, PAGE_SIZE_NAME_LEN, "%s", "N/A");
1351
1352 return str;
1353}
1354
1355static void dump_sample(struct evsel *evsel, union perf_event *event,
1356 struct perf_sample *sample, const char *arch)
1357{
1358 u64 sample_type;
1359 char str[PAGE_SIZE_NAME_LEN];
1360
1361 if (!dump_trace)
1362 return;
1363
1364 printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n",
1365 event->header.misc, sample->pid, sample->tid, sample->ip,
1366 sample->period, sample->addr);
1367
1368 sample_type = evsel->core.attr.sample_type;
1369
1370 if (evsel__has_callchain(evsel))
1371 callchain__printf(evsel, sample);
1372
1373 if (evsel__has_br_stack(evsel))
1374 branch_stack__printf(sample, evsel);
1375
1376 if (sample_type & PERF_SAMPLE_REGS_USER)
1377 regs_user__printf(sample, arch);
1378
1379 if (sample_type & PERF_SAMPLE_REGS_INTR)
1380 regs_intr__printf(sample, arch);
1381
1382 if (sample_type & PERF_SAMPLE_STACK_USER)
1383 stack_user__printf(&sample->user_stack);
1384
1385 if (sample_type & PERF_SAMPLE_WEIGHT_TYPE) {
1386 printf("... weight: %" PRIu64 "", sample->weight);
1387 if (sample_type & PERF_SAMPLE_WEIGHT_STRUCT) {
1388 printf(",0x%"PRIx16"", sample->ins_lat);
1389 printf(",0x%"PRIx16"", sample->p_stage_cyc);
1390 }
1391 printf("\n");
1392 }
1393
1394 if (sample_type & PERF_SAMPLE_DATA_SRC)
1395 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src);
1396
1397 if (sample_type & PERF_SAMPLE_PHYS_ADDR)
1398 printf(" .. phys_addr: 0x%"PRIx64"\n", sample->phys_addr);
1399
1400 if (sample_type & PERF_SAMPLE_DATA_PAGE_SIZE)
1401 printf(" .. data page size: %s\n", get_page_size_name(sample->data_page_size, str));
1402
1403 if (sample_type & PERF_SAMPLE_CODE_PAGE_SIZE)
1404 printf(" .. code page size: %s\n", get_page_size_name(sample->code_page_size, str));
1405
1406 if (sample_type & PERF_SAMPLE_TRANSACTION)
1407 printf("... transaction: %" PRIx64 "\n", sample->transaction);
1408
1409 if (sample_type & PERF_SAMPLE_READ)
1410 sample_read__printf(sample, evsel->core.attr.read_format);
1411}
1412
1413static void dump_read(struct evsel *evsel, union perf_event *event)
1414{
1415 struct perf_record_read *read_event = &event->read;
1416 u64 read_format;
1417
1418 if (!dump_trace)
1419 return;
1420
1421 printf(": %d %d %s %" PRI_lu64 "\n", event->read.pid, event->read.tid,
1422 evsel__name(evsel), event->read.value);
1423
1424 if (!evsel)
1425 return;
1426
1427 read_format = evsel->core.attr.read_format;
1428
1429 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1430 printf("... time enabled : %" PRI_lu64 "\n", read_event->time_enabled);
1431
1432 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1433 printf("... time running : %" PRI_lu64 "\n", read_event->time_running);
1434
1435 if (read_format & PERF_FORMAT_ID)
1436 printf("... id : %" PRI_lu64 "\n", read_event->id);
1437
1438 if (read_format & PERF_FORMAT_LOST)
1439 printf("... lost : %" PRI_lu64 "\n", read_event->lost);
1440}
1441
1442static struct machine *machines__find_for_cpumode(struct machines *machines,
1443 union perf_event *event,
1444 struct perf_sample *sample)
1445{
1446 if (perf_guest &&
1447 ((sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL) ||
1448 (sample->cpumode == PERF_RECORD_MISC_GUEST_USER))) {
1449 u32 pid;
1450
1451 if (sample->machine_pid)
1452 pid = sample->machine_pid;
1453 else if (event->header.type == PERF_RECORD_MMAP
1454 || event->header.type == PERF_RECORD_MMAP2)
1455 pid = event->mmap.pid;
1456 else
1457 pid = sample->pid;
1458
1459 /*
1460 * Guest code machine is created as needed and does not use
1461 * DEFAULT_GUEST_KERNEL_ID.
1462 */
1463 if (symbol_conf.guest_code)
1464 return machines__findnew(machines, pid);
1465
1466 return machines__find_guest(machines, pid);
1467 }
1468
1469 return &machines->host;
1470}
1471
1472static int deliver_sample_value(struct evlist *evlist,
1473 struct perf_tool *tool,
1474 union perf_event *event,
1475 struct perf_sample *sample,
1476 struct sample_read_value *v,
1477 struct machine *machine)
1478{
1479 struct perf_sample_id *sid = evlist__id2sid(evlist, v->id);
1480 struct evsel *evsel;
1481
1482 if (sid) {
1483 sample->id = v->id;
1484 sample->period = v->value - sid->period;
1485 sid->period = v->value;
1486 }
1487
1488 if (!sid || sid->evsel == NULL) {
1489 ++evlist->stats.nr_unknown_id;
1490 return 0;
1491 }
1492
1493 /*
1494 * There's no reason to deliver sample
1495 * for zero period, bail out.
1496 */
1497 if (!sample->period)
1498 return 0;
1499
1500 evsel = container_of(sid->evsel, struct evsel, core);
1501 return tool->sample(tool, event, sample, evsel, machine);
1502}
1503
1504static int deliver_sample_group(struct evlist *evlist,
1505 struct perf_tool *tool,
1506 union perf_event *event,
1507 struct perf_sample *sample,
1508 struct machine *machine,
1509 u64 read_format)
1510{
1511 int ret = -EINVAL;
1512 struct sample_read_value *v = sample->read.group.values;
1513
1514 sample_read_group__for_each(v, sample->read.group.nr, read_format) {
1515 ret = deliver_sample_value(evlist, tool, event, sample, v,
1516 machine);
1517 if (ret)
1518 break;
1519 }
1520
1521 return ret;
1522}
1523
1524static int evlist__deliver_sample(struct evlist *evlist, struct perf_tool *tool,
1525 union perf_event *event, struct perf_sample *sample,
1526 struct evsel *evsel, struct machine *machine)
1527{
1528 /* We know evsel != NULL. */
1529 u64 sample_type = evsel->core.attr.sample_type;
1530 u64 read_format = evsel->core.attr.read_format;
1531
1532 /* Standard sample delivery. */
1533 if (!(sample_type & PERF_SAMPLE_READ))
1534 return tool->sample(tool, event, sample, evsel, machine);
1535
1536 /* For PERF_SAMPLE_READ we have either single or group mode. */
1537 if (read_format & PERF_FORMAT_GROUP)
1538 return deliver_sample_group(evlist, tool, event, sample,
1539 machine, read_format);
1540 else
1541 return deliver_sample_value(evlist, tool, event, sample,
1542 &sample->read.one, machine);
1543}
1544
1545static int machines__deliver_event(struct machines *machines,
1546 struct evlist *evlist,
1547 union perf_event *event,
1548 struct perf_sample *sample,
1549 struct perf_tool *tool, u64 file_offset,
1550 const char *file_path)
1551{
1552 struct evsel *evsel;
1553 struct machine *machine;
1554
1555 dump_event(evlist, event, file_offset, sample, file_path);
1556
1557 evsel = evlist__id2evsel(evlist, sample->id);
1558
1559 machine = machines__find_for_cpumode(machines, event, sample);
1560
1561 switch (event->header.type) {
1562 case PERF_RECORD_SAMPLE:
1563 if (evsel == NULL) {
1564 ++evlist->stats.nr_unknown_id;
1565 return 0;
1566 }
1567 if (machine == NULL) {
1568 ++evlist->stats.nr_unprocessable_samples;
1569 dump_sample(evsel, event, sample, perf_env__arch(NULL));
1570 return 0;
1571 }
1572 dump_sample(evsel, event, sample, perf_env__arch(machine->env));
1573 return evlist__deliver_sample(evlist, tool, event, sample, evsel, machine);
1574 case PERF_RECORD_MMAP:
1575 return tool->mmap(tool, event, sample, machine);
1576 case PERF_RECORD_MMAP2:
1577 if (event->header.misc & PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT)
1578 ++evlist->stats.nr_proc_map_timeout;
1579 return tool->mmap2(tool, event, sample, machine);
1580 case PERF_RECORD_COMM:
1581 return tool->comm(tool, event, sample, machine);
1582 case PERF_RECORD_NAMESPACES:
1583 return tool->namespaces(tool, event, sample, machine);
1584 case PERF_RECORD_CGROUP:
1585 return tool->cgroup(tool, event, sample, machine);
1586 case PERF_RECORD_FORK:
1587 return tool->fork(tool, event, sample, machine);
1588 case PERF_RECORD_EXIT:
1589 return tool->exit(tool, event, sample, machine);
1590 case PERF_RECORD_LOST:
1591 if (tool->lost == perf_event__process_lost)
1592 evlist->stats.total_lost += event->lost.lost;
1593 return tool->lost(tool, event, sample, machine);
1594 case PERF_RECORD_LOST_SAMPLES:
1595 if (tool->lost_samples == perf_event__process_lost_samples &&
1596 !(event->header.misc & PERF_RECORD_MISC_LOST_SAMPLES_BPF))
1597 evlist->stats.total_lost_samples += event->lost_samples.lost;
1598 return tool->lost_samples(tool, event, sample, machine);
1599 case PERF_RECORD_READ:
1600 dump_read(evsel, event);
1601 return tool->read(tool, event, sample, evsel, machine);
1602 case PERF_RECORD_THROTTLE:
1603 return tool->throttle(tool, event, sample, machine);
1604 case PERF_RECORD_UNTHROTTLE:
1605 return tool->unthrottle(tool, event, sample, machine);
1606 case PERF_RECORD_AUX:
1607 if (tool->aux == perf_event__process_aux) {
1608 if (event->aux.flags & PERF_AUX_FLAG_TRUNCATED)
1609 evlist->stats.total_aux_lost += 1;
1610 if (event->aux.flags & PERF_AUX_FLAG_PARTIAL)
1611 evlist->stats.total_aux_partial += 1;
1612 if (event->aux.flags & PERF_AUX_FLAG_COLLISION)
1613 evlist->stats.total_aux_collision += 1;
1614 }
1615 return tool->aux(tool, event, sample, machine);
1616 case PERF_RECORD_ITRACE_START:
1617 return tool->itrace_start(tool, event, sample, machine);
1618 case PERF_RECORD_SWITCH:
1619 case PERF_RECORD_SWITCH_CPU_WIDE:
1620 return tool->context_switch(tool, event, sample, machine);
1621 case PERF_RECORD_KSYMBOL:
1622 return tool->ksymbol(tool, event, sample, machine);
1623 case PERF_RECORD_BPF_EVENT:
1624 return tool->bpf(tool, event, sample, machine);
1625 case PERF_RECORD_TEXT_POKE:
1626 return tool->text_poke(tool, event, sample, machine);
1627 case PERF_RECORD_AUX_OUTPUT_HW_ID:
1628 return tool->aux_output_hw_id(tool, event, sample, machine);
1629 default:
1630 ++evlist->stats.nr_unknown_events;
1631 return -1;
1632 }
1633}
1634
1635static int perf_session__deliver_event(struct perf_session *session,
1636 union perf_event *event,
1637 struct perf_tool *tool,
1638 u64 file_offset,
1639 const char *file_path)
1640{
1641 struct perf_sample sample;
1642 int ret = evlist__parse_sample(session->evlist, event, &sample);
1643
1644 if (ret) {
1645 pr_err("Can't parse sample, err = %d\n", ret);
1646 return ret;
1647 }
1648
1649 ret = auxtrace__process_event(session, event, &sample, tool);
1650 if (ret < 0)
1651 return ret;
1652 if (ret > 0)
1653 return 0;
1654
1655 ret = machines__deliver_event(&session->machines, session->evlist,
1656 event, &sample, tool, file_offset, file_path);
1657
1658 if (dump_trace && sample.aux_sample.size)
1659 auxtrace__dump_auxtrace_sample(session, &sample);
1660
1661 return ret;
1662}
1663
1664static s64 perf_session__process_user_event(struct perf_session *session,
1665 union perf_event *event,
1666 u64 file_offset,
1667 const char *file_path)
1668{
1669 struct ordered_events *oe = &session->ordered_events;
1670 struct perf_tool *tool = session->tool;
1671 struct perf_sample sample = { .time = 0, };
1672 int fd = perf_data__fd(session->data);
1673 int err;
1674
1675 if (event->header.type != PERF_RECORD_COMPRESSED ||
1676 tool->compressed == perf_session__process_compressed_event_stub)
1677 dump_event(session->evlist, event, file_offset, &sample, file_path);
1678
1679 /* These events are processed right away */
1680 switch (event->header.type) {
1681 case PERF_RECORD_HEADER_ATTR:
1682 err = tool->attr(tool, event, &session->evlist);
1683 if (err == 0) {
1684 perf_session__set_id_hdr_size(session);
1685 perf_session__set_comm_exec(session);
1686 }
1687 return err;
1688 case PERF_RECORD_EVENT_UPDATE:
1689 return tool->event_update(tool, event, &session->evlist);
1690 case PERF_RECORD_HEADER_EVENT_TYPE:
1691 /*
1692 * Deprecated, but we need to handle it for sake
1693 * of old data files create in pipe mode.
1694 */
1695 return 0;
1696 case PERF_RECORD_HEADER_TRACING_DATA:
1697 /*
1698 * Setup for reading amidst mmap, but only when we
1699 * are in 'file' mode. The 'pipe' fd is in proper
1700 * place already.
1701 */
1702 if (!perf_data__is_pipe(session->data))
1703 lseek(fd, file_offset, SEEK_SET);
1704 return tool->tracing_data(session, event);
1705 case PERF_RECORD_HEADER_BUILD_ID:
1706 return tool->build_id(session, event);
1707 case PERF_RECORD_FINISHED_ROUND:
1708 return tool->finished_round(tool, event, oe);
1709 case PERF_RECORD_ID_INDEX:
1710 return tool->id_index(session, event);
1711 case PERF_RECORD_AUXTRACE_INFO:
1712 return tool->auxtrace_info(session, event);
1713 case PERF_RECORD_AUXTRACE:
1714 /*
1715 * Setup for reading amidst mmap, but only when we
1716 * are in 'file' mode. The 'pipe' fd is in proper
1717 * place already.
1718 */
1719 if (!perf_data__is_pipe(session->data))
1720 lseek(fd, file_offset + event->header.size, SEEK_SET);
1721 return tool->auxtrace(session, event);
1722 case PERF_RECORD_AUXTRACE_ERROR:
1723 perf_session__auxtrace_error_inc(session, event);
1724 return tool->auxtrace_error(session, event);
1725 case PERF_RECORD_THREAD_MAP:
1726 return tool->thread_map(session, event);
1727 case PERF_RECORD_CPU_MAP:
1728 return tool->cpu_map(session, event);
1729 case PERF_RECORD_STAT_CONFIG:
1730 return tool->stat_config(session, event);
1731 case PERF_RECORD_STAT:
1732 return tool->stat(session, event);
1733 case PERF_RECORD_STAT_ROUND:
1734 return tool->stat_round(session, event);
1735 case PERF_RECORD_TIME_CONV:
1736 session->time_conv = event->time_conv;
1737 return tool->time_conv(session, event);
1738 case PERF_RECORD_HEADER_FEATURE:
1739 return tool->feature(session, event);
1740 case PERF_RECORD_COMPRESSED:
1741 err = tool->compressed(session, event, file_offset, file_path);
1742 if (err)
1743 dump_event(session->evlist, event, file_offset, &sample, file_path);
1744 return err;
1745 case PERF_RECORD_FINISHED_INIT:
1746 return tool->finished_init(session, event);
1747 default:
1748 return -EINVAL;
1749 }
1750}
1751
1752int perf_session__deliver_synth_event(struct perf_session *session,
1753 union perf_event *event,
1754 struct perf_sample *sample)
1755{
1756 struct evlist *evlist = session->evlist;
1757 struct perf_tool *tool = session->tool;
1758
1759 events_stats__inc(&evlist->stats, event->header.type);
1760
1761 if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1762 return perf_session__process_user_event(session, event, 0, NULL);
1763
1764 return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0, NULL);
1765}
1766
1767static void event_swap(union perf_event *event, bool sample_id_all)
1768{
1769 perf_event__swap_op swap;
1770
1771 swap = perf_event__swap_ops[event->header.type];
1772 if (swap)
1773 swap(event, sample_id_all);
1774}
1775
1776int perf_session__peek_event(struct perf_session *session, off_t file_offset,
1777 void *buf, size_t buf_sz,
1778 union perf_event **event_ptr,
1779 struct perf_sample *sample)
1780{
1781 union perf_event *event;
1782 size_t hdr_sz, rest;
1783 int fd;
1784
1785 if (session->one_mmap && !session->header.needs_swap) {
1786 event = file_offset - session->one_mmap_offset +
1787 session->one_mmap_addr;
1788 goto out_parse_sample;
1789 }
1790
1791 if (perf_data__is_pipe(session->data))
1792 return -1;
1793
1794 fd = perf_data__fd(session->data);
1795 hdr_sz = sizeof(struct perf_event_header);
1796
1797 if (buf_sz < hdr_sz)
1798 return -1;
1799
1800 if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 ||
1801 readn(fd, buf, hdr_sz) != (ssize_t)hdr_sz)
1802 return -1;
1803
1804 event = (union perf_event *)buf;
1805
1806 if (session->header.needs_swap)
1807 perf_event_header__bswap(&event->header);
1808
1809 if (event->header.size < hdr_sz || event->header.size > buf_sz)
1810 return -1;
1811
1812 buf += hdr_sz;
1813 rest = event->header.size - hdr_sz;
1814
1815 if (readn(fd, buf, rest) != (ssize_t)rest)
1816 return -1;
1817
1818 if (session->header.needs_swap)
1819 event_swap(event, evlist__sample_id_all(session->evlist));
1820
1821out_parse_sample:
1822
1823 if (sample && event->header.type < PERF_RECORD_USER_TYPE_START &&
1824 evlist__parse_sample(session->evlist, event, sample))
1825 return -1;
1826
1827 *event_ptr = event;
1828
1829 return 0;
1830}
1831
1832int perf_session__peek_events(struct perf_session *session, u64 offset,
1833 u64 size, peek_events_cb_t cb, void *data)
1834{
1835 u64 max_offset = offset + size;
1836 char buf[PERF_SAMPLE_MAX_SIZE];
1837 union perf_event *event;
1838 int err;
1839
1840 do {
1841 err = perf_session__peek_event(session, offset, buf,
1842 PERF_SAMPLE_MAX_SIZE, &event,
1843 NULL);
1844 if (err)
1845 return err;
1846
1847 err = cb(session, event, offset, data);
1848 if (err)
1849 return err;
1850
1851 offset += event->header.size;
1852 if (event->header.type == PERF_RECORD_AUXTRACE)
1853 offset += event->auxtrace.size;
1854
1855 } while (offset < max_offset);
1856
1857 return err;
1858}
1859
1860static s64 perf_session__process_event(struct perf_session *session,
1861 union perf_event *event, u64 file_offset,
1862 const char *file_path)
1863{
1864 struct evlist *evlist = session->evlist;
1865 struct perf_tool *tool = session->tool;
1866 int ret;
1867
1868 if (session->header.needs_swap)
1869 event_swap(event, evlist__sample_id_all(evlist));
1870
1871 if (event->header.type >= PERF_RECORD_HEADER_MAX)
1872 return -EINVAL;
1873
1874 events_stats__inc(&evlist->stats, event->header.type);
1875
1876 if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1877 return perf_session__process_user_event(session, event, file_offset, file_path);
1878
1879 if (tool->ordered_events) {
1880 u64 timestamp = -1ULL;
1881
1882 ret = evlist__parse_sample_timestamp(evlist, event, ×tamp);
1883 if (ret && ret != -1)
1884 return ret;
1885
1886 ret = perf_session__queue_event(session, event, timestamp, file_offset, file_path);
1887 if (ret != -ETIME)
1888 return ret;
1889 }
1890
1891 return perf_session__deliver_event(session, event, tool, file_offset, file_path);
1892}
1893
1894void perf_event_header__bswap(struct perf_event_header *hdr)
1895{
1896 hdr->type = bswap_32(hdr->type);
1897 hdr->misc = bswap_16(hdr->misc);
1898 hdr->size = bswap_16(hdr->size);
1899}
1900
1901struct thread *perf_session__findnew(struct perf_session *session, pid_t pid)
1902{
1903 return machine__findnew_thread(&session->machines.host, -1, pid);
1904}
1905
1906int perf_session__register_idle_thread(struct perf_session *session)
1907{
1908 struct thread *thread = machine__idle_thread(&session->machines.host);
1909
1910 /* machine__idle_thread() got the thread, so put it */
1911 thread__put(thread);
1912 return thread ? 0 : -1;
1913}
1914
1915static void
1916perf_session__warn_order(const struct perf_session *session)
1917{
1918 const struct ordered_events *oe = &session->ordered_events;
1919 struct evsel *evsel;
1920 bool should_warn = true;
1921
1922 evlist__for_each_entry(session->evlist, evsel) {
1923 if (evsel->core.attr.write_backward)
1924 should_warn = false;
1925 }
1926
1927 if (!should_warn)
1928 return;
1929 if (oe->nr_unordered_events != 0)
1930 ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events);
1931}
1932
1933static void perf_session__warn_about_errors(const struct perf_session *session)
1934{
1935 const struct events_stats *stats = &session->evlist->stats;
1936
1937 if (session->tool->lost == perf_event__process_lost &&
1938 stats->nr_events[PERF_RECORD_LOST] != 0) {
1939 ui__warning("Processed %d events and lost %d chunks!\n\n"
1940 "Check IO/CPU overload!\n\n",
1941 stats->nr_events[0],
1942 stats->nr_events[PERF_RECORD_LOST]);
1943 }
1944
1945 if (session->tool->lost_samples == perf_event__process_lost_samples) {
1946 double drop_rate;
1947
1948 drop_rate = (double)stats->total_lost_samples /
1949 (double) (stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples);
1950 if (drop_rate > 0.05) {
1951 ui__warning("Processed %" PRIu64 " samples and lost %3.2f%%!\n\n",
1952 stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples,
1953 drop_rate * 100.0);
1954 }
1955 }
1956
1957 if (session->tool->aux == perf_event__process_aux &&
1958 stats->total_aux_lost != 0) {
1959 ui__warning("AUX data lost %" PRIu64 " times out of %u!\n\n",
1960 stats->total_aux_lost,
1961 stats->nr_events[PERF_RECORD_AUX]);
1962 }
1963
1964 if (session->tool->aux == perf_event__process_aux &&
1965 stats->total_aux_partial != 0) {
1966 bool vmm_exclusive = false;
1967
1968 (void)sysfs__read_bool("module/kvm_intel/parameters/vmm_exclusive",
1969 &vmm_exclusive);
1970
1971 ui__warning("AUX data had gaps in it %" PRIu64 " times out of %u!\n\n"
1972 "Are you running a KVM guest in the background?%s\n\n",
1973 stats->total_aux_partial,
1974 stats->nr_events[PERF_RECORD_AUX],
1975 vmm_exclusive ?
1976 "\nReloading kvm_intel module with vmm_exclusive=0\n"
1977 "will reduce the gaps to only guest's timeslices." :
1978 "");
1979 }
1980
1981 if (session->tool->aux == perf_event__process_aux &&
1982 stats->total_aux_collision != 0) {
1983 ui__warning("AUX data detected collision %" PRIu64 " times out of %u!\n\n",
1984 stats->total_aux_collision,
1985 stats->nr_events[PERF_RECORD_AUX]);
1986 }
1987
1988 if (stats->nr_unknown_events != 0) {
1989 ui__warning("Found %u unknown events!\n\n"
1990 "Is this an older tool processing a perf.data "
1991 "file generated by a more recent tool?\n\n"
1992 "If that is not the case, consider "
1993 "reporting to linux-kernel@vger.kernel.org.\n\n",
1994 stats->nr_unknown_events);
1995 }
1996
1997 if (stats->nr_unknown_id != 0) {
1998 ui__warning("%u samples with id not present in the header\n",
1999 stats->nr_unknown_id);
2000 }
2001
2002 if (stats->nr_invalid_chains != 0) {
2003 ui__warning("Found invalid callchains!\n\n"
2004 "%u out of %u events were discarded for this reason.\n\n"
2005 "Consider reporting to linux-kernel@vger.kernel.org.\n\n",
2006 stats->nr_invalid_chains,
2007 stats->nr_events[PERF_RECORD_SAMPLE]);
2008 }
2009
2010 if (stats->nr_unprocessable_samples != 0) {
2011 ui__warning("%u unprocessable samples recorded.\n"
2012 "Do you have a KVM guest running and not using 'perf kvm'?\n",
2013 stats->nr_unprocessable_samples);
2014 }
2015
2016 perf_session__warn_order(session);
2017
2018 events_stats__auxtrace_error_warn(stats);
2019
2020 if (stats->nr_proc_map_timeout != 0) {
2021 ui__warning("%d map information files for pre-existing threads were\n"
2022 "not processed, if there are samples for addresses they\n"
2023 "will not be resolved, you may find out which are these\n"
2024 "threads by running with -v and redirecting the output\n"
2025 "to a file.\n"
2026 "The time limit to process proc map is too short?\n"
2027 "Increase it by --proc-map-timeout\n",
2028 stats->nr_proc_map_timeout);
2029 }
2030}
2031
2032static int perf_session__flush_thread_stack(struct thread *thread,
2033 void *p __maybe_unused)
2034{
2035 return thread_stack__flush(thread);
2036}
2037
2038static int perf_session__flush_thread_stacks(struct perf_session *session)
2039{
2040 return machines__for_each_thread(&session->machines,
2041 perf_session__flush_thread_stack,
2042 NULL);
2043}
2044
2045volatile sig_atomic_t session_done;
2046
2047static int __perf_session__process_decomp_events(struct perf_session *session);
2048
2049static int __perf_session__process_pipe_events(struct perf_session *session)
2050{
2051 struct ordered_events *oe = &session->ordered_events;
2052 struct perf_tool *tool = session->tool;
2053 union perf_event *event;
2054 uint32_t size, cur_size = 0;
2055 void *buf = NULL;
2056 s64 skip = 0;
2057 u64 head;
2058 ssize_t err;
2059 void *p;
2060
2061 perf_tool__fill_defaults(tool);
2062
2063 head = 0;
2064 cur_size = sizeof(union perf_event);
2065
2066 buf = malloc(cur_size);
2067 if (!buf)
2068 return -errno;
2069 ordered_events__set_copy_on_queue(oe, true);
2070more:
2071 event = buf;
2072 err = perf_data__read(session->data, event,
2073 sizeof(struct perf_event_header));
2074 if (err <= 0) {
2075 if (err == 0)
2076 goto done;
2077
2078 pr_err("failed to read event header\n");
2079 goto out_err;
2080 }
2081
2082 if (session->header.needs_swap)
2083 perf_event_header__bswap(&event->header);
2084
2085 size = event->header.size;
2086 if (size < sizeof(struct perf_event_header)) {
2087 pr_err("bad event header size\n");
2088 goto out_err;
2089 }
2090
2091 if (size > cur_size) {
2092 void *new = realloc(buf, size);
2093 if (!new) {
2094 pr_err("failed to allocate memory to read event\n");
2095 goto out_err;
2096 }
2097 buf = new;
2098 cur_size = size;
2099 event = buf;
2100 }
2101 p = event;
2102 p += sizeof(struct perf_event_header);
2103
2104 if (size - sizeof(struct perf_event_header)) {
2105 err = perf_data__read(session->data, p,
2106 size - sizeof(struct perf_event_header));
2107 if (err <= 0) {
2108 if (err == 0) {
2109 pr_err("unexpected end of event stream\n");
2110 goto done;
2111 }
2112
2113 pr_err("failed to read event data\n");
2114 goto out_err;
2115 }
2116 }
2117
2118 if ((skip = perf_session__process_event(session, event, head, "pipe")) < 0) {
2119 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
2120 head, event->header.size, event->header.type);
2121 err = -EINVAL;
2122 goto out_err;
2123 }
2124
2125 head += size;
2126
2127 if (skip > 0)
2128 head += skip;
2129
2130 err = __perf_session__process_decomp_events(session);
2131 if (err)
2132 goto out_err;
2133
2134 if (!session_done())
2135 goto more;
2136done:
2137 /* do the final flush for ordered samples */
2138 err = ordered_events__flush(oe, OE_FLUSH__FINAL);
2139 if (err)
2140 goto out_err;
2141 err = auxtrace__flush_events(session, tool);
2142 if (err)
2143 goto out_err;
2144 err = perf_session__flush_thread_stacks(session);
2145out_err:
2146 free(buf);
2147 if (!tool->no_warn)
2148 perf_session__warn_about_errors(session);
2149 ordered_events__free(&session->ordered_events);
2150 auxtrace__free_events(session);
2151 return err;
2152}
2153
2154static union perf_event *
2155prefetch_event(char *buf, u64 head, size_t mmap_size,
2156 bool needs_swap, union perf_event *error)
2157{
2158 union perf_event *event;
2159 u16 event_size;
2160
2161 /*
2162 * Ensure we have enough space remaining to read
2163 * the size of the event in the headers.
2164 */
2165 if (head + sizeof(event->header) > mmap_size)
2166 return NULL;
2167
2168 event = (union perf_event *)(buf + head);
2169 if (needs_swap)
2170 perf_event_header__bswap(&event->header);
2171
2172 event_size = event->header.size;
2173 if (head + event_size <= mmap_size)
2174 return event;
2175
2176 /* We're not fetching the event so swap back again */
2177 if (needs_swap)
2178 perf_event_header__bswap(&event->header);
2179
2180 /* Check if the event fits into the next mmapped buf. */
2181 if (event_size <= mmap_size - head % page_size) {
2182 /* Remap buf and fetch again. */
2183 return NULL;
2184 }
2185
2186 /* Invalid input. Event size should never exceed mmap_size. */
2187 pr_debug("%s: head=%#" PRIx64 " event->header.size=%#x, mmap_size=%#zx:"
2188 " fuzzed or compressed perf.data?\n", __func__, head, event_size, mmap_size);
2189
2190 return error;
2191}
2192
2193static union perf_event *
2194fetch_mmaped_event(u64 head, size_t mmap_size, char *buf, bool needs_swap)
2195{
2196 return prefetch_event(buf, head, mmap_size, needs_swap, ERR_PTR(-EINVAL));
2197}
2198
2199static union perf_event *
2200fetch_decomp_event(u64 head, size_t mmap_size, char *buf, bool needs_swap)
2201{
2202 return prefetch_event(buf, head, mmap_size, needs_swap, NULL);
2203}
2204
2205static int __perf_session__process_decomp_events(struct perf_session *session)
2206{
2207 s64 skip;
2208 u64 size;
2209 struct decomp *decomp = session->active_decomp->decomp_last;
2210
2211 if (!decomp)
2212 return 0;
2213
2214 while (decomp->head < decomp->size && !session_done()) {
2215 union perf_event *event = fetch_decomp_event(decomp->head, decomp->size, decomp->data,
2216 session->header.needs_swap);
2217
2218 if (!event)
2219 break;
2220
2221 size = event->header.size;
2222
2223 if (size < sizeof(struct perf_event_header) ||
2224 (skip = perf_session__process_event(session, event, decomp->file_pos,
2225 decomp->file_path)) < 0) {
2226 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
2227 decomp->file_pos + decomp->head, event->header.size, event->header.type);
2228 return -EINVAL;
2229 }
2230
2231 if (skip)
2232 size += skip;
2233
2234 decomp->head += size;
2235 }
2236
2237 return 0;
2238}
2239
2240/*
2241 * On 64bit we can mmap the data file in one go. No need for tiny mmap
2242 * slices. On 32bit we use 32MB.
2243 */
2244#if BITS_PER_LONG == 64
2245#define MMAP_SIZE ULLONG_MAX
2246#define NUM_MMAPS 1
2247#else
2248#define MMAP_SIZE (32 * 1024 * 1024ULL)
2249#define NUM_MMAPS 128
2250#endif
2251
2252struct reader;
2253
2254typedef s64 (*reader_cb_t)(struct perf_session *session,
2255 union perf_event *event,
2256 u64 file_offset,
2257 const char *file_path);
2258
2259struct reader {
2260 int fd;
2261 const char *path;
2262 u64 data_size;
2263 u64 data_offset;
2264 reader_cb_t process;
2265 bool in_place_update;
2266 char *mmaps[NUM_MMAPS];
2267 size_t mmap_size;
2268 int mmap_idx;
2269 char *mmap_cur;
2270 u64 file_pos;
2271 u64 file_offset;
2272 u64 head;
2273 u64 size;
2274 bool done;
2275 struct zstd_data zstd_data;
2276 struct decomp_data decomp_data;
2277};
2278
2279static int
2280reader__init(struct reader *rd, bool *one_mmap)
2281{
2282 u64 data_size = rd->data_size;
2283 char **mmaps = rd->mmaps;
2284
2285 rd->head = rd->data_offset;
2286 data_size += rd->data_offset;
2287
2288 rd->mmap_size = MMAP_SIZE;
2289 if (rd->mmap_size > data_size) {
2290 rd->mmap_size = data_size;
2291 if (one_mmap)
2292 *one_mmap = true;
2293 }
2294
2295 memset(mmaps, 0, sizeof(rd->mmaps));
2296
2297 if (zstd_init(&rd->zstd_data, 0))
2298 return -1;
2299 rd->decomp_data.zstd_decomp = &rd->zstd_data;
2300
2301 return 0;
2302}
2303
2304static void
2305reader__release_decomp(struct reader *rd)
2306{
2307 perf_decomp__release_events(rd->decomp_data.decomp);
2308 zstd_fini(&rd->zstd_data);
2309}
2310
2311static int
2312reader__mmap(struct reader *rd, struct perf_session *session)
2313{
2314 int mmap_prot, mmap_flags;
2315 char *buf, **mmaps = rd->mmaps;
2316 u64 page_offset;
2317
2318 mmap_prot = PROT_READ;
2319 mmap_flags = MAP_SHARED;
2320
2321 if (rd->in_place_update) {
2322 mmap_prot |= PROT_WRITE;
2323 } else if (session->header.needs_swap) {
2324 mmap_prot |= PROT_WRITE;
2325 mmap_flags = MAP_PRIVATE;
2326 }
2327
2328 if (mmaps[rd->mmap_idx]) {
2329 munmap(mmaps[rd->mmap_idx], rd->mmap_size);
2330 mmaps[rd->mmap_idx] = NULL;
2331 }
2332
2333 page_offset = page_size * (rd->head / page_size);
2334 rd->file_offset += page_offset;
2335 rd->head -= page_offset;
2336
2337 buf = mmap(NULL, rd->mmap_size, mmap_prot, mmap_flags, rd->fd,
2338 rd->file_offset);
2339 if (buf == MAP_FAILED) {
2340 pr_err("failed to mmap file\n");
2341 return -errno;
2342 }
2343 mmaps[rd->mmap_idx] = rd->mmap_cur = buf;
2344 rd->mmap_idx = (rd->mmap_idx + 1) & (ARRAY_SIZE(rd->mmaps) - 1);
2345 rd->file_pos = rd->file_offset + rd->head;
2346 if (session->one_mmap) {
2347 session->one_mmap_addr = buf;
2348 session->one_mmap_offset = rd->file_offset;
2349 }
2350
2351 return 0;
2352}
2353
2354enum {
2355 READER_OK,
2356 READER_NODATA,
2357};
2358
2359static int
2360reader__read_event(struct reader *rd, struct perf_session *session,
2361 struct ui_progress *prog)
2362{
2363 u64 size;
2364 int err = READER_OK;
2365 union perf_event *event;
2366 s64 skip;
2367
2368 event = fetch_mmaped_event(rd->head, rd->mmap_size, rd->mmap_cur,
2369 session->header.needs_swap);
2370 if (IS_ERR(event))
2371 return PTR_ERR(event);
2372
2373 if (!event)
2374 return READER_NODATA;
2375
2376 size = event->header.size;
2377
2378 skip = -EINVAL;
2379
2380 if (size < sizeof(struct perf_event_header) ||
2381 (skip = rd->process(session, event, rd->file_pos, rd->path)) < 0) {
2382 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d [%s]\n",
2383 rd->file_offset + rd->head, event->header.size,
2384 event->header.type, strerror(-skip));
2385 err = skip;
2386 goto out;
2387 }
2388
2389 if (skip)
2390 size += skip;
2391
2392 rd->size += size;
2393 rd->head += size;
2394 rd->file_pos += size;
2395
2396 err = __perf_session__process_decomp_events(session);
2397 if (err)
2398 goto out;
2399
2400 ui_progress__update(prog, size);
2401
2402out:
2403 return err;
2404}
2405
2406static inline bool
2407reader__eof(struct reader *rd)
2408{
2409 return (rd->file_pos >= rd->data_size + rd->data_offset);
2410}
2411
2412static int
2413reader__process_events(struct reader *rd, struct perf_session *session,
2414 struct ui_progress *prog)
2415{
2416 int err;
2417
2418 err = reader__init(rd, &session->one_mmap);
2419 if (err)
2420 goto out;
2421
2422 session->active_decomp = &rd->decomp_data;
2423
2424remap:
2425 err = reader__mmap(rd, session);
2426 if (err)
2427 goto out;
2428
2429more:
2430 err = reader__read_event(rd, session, prog);
2431 if (err < 0)
2432 goto out;
2433 else if (err == READER_NODATA)
2434 goto remap;
2435
2436 if (session_done())
2437 goto out;
2438
2439 if (!reader__eof(rd))
2440 goto more;
2441
2442out:
2443 session->active_decomp = &session->decomp_data;
2444 return err;
2445}
2446
2447static s64 process_simple(struct perf_session *session,
2448 union perf_event *event,
2449 u64 file_offset,
2450 const char *file_path)
2451{
2452 return perf_session__process_event(session, event, file_offset, file_path);
2453}
2454
2455static int __perf_session__process_events(struct perf_session *session)
2456{
2457 struct reader rd = {
2458 .fd = perf_data__fd(session->data),
2459 .path = session->data->file.path,
2460 .data_size = session->header.data_size,
2461 .data_offset = session->header.data_offset,
2462 .process = process_simple,
2463 .in_place_update = session->data->in_place_update,
2464 };
2465 struct ordered_events *oe = &session->ordered_events;
2466 struct perf_tool *tool = session->tool;
2467 struct ui_progress prog;
2468 int err;
2469
2470 perf_tool__fill_defaults(tool);
2471
2472 if (rd.data_size == 0)
2473 return -1;
2474
2475 ui_progress__init_size(&prog, rd.data_size, "Processing events...");
2476
2477 err = reader__process_events(&rd, session, &prog);
2478 if (err)
2479 goto out_err;
2480 /* do the final flush for ordered samples */
2481 err = ordered_events__flush(oe, OE_FLUSH__FINAL);
2482 if (err)
2483 goto out_err;
2484 err = auxtrace__flush_events(session, tool);
2485 if (err)
2486 goto out_err;
2487 err = perf_session__flush_thread_stacks(session);
2488out_err:
2489 ui_progress__finish();
2490 if (!tool->no_warn)
2491 perf_session__warn_about_errors(session);
2492 /*
2493 * We may switching perf.data output, make ordered_events
2494 * reusable.
2495 */
2496 ordered_events__reinit(&session->ordered_events);
2497 auxtrace__free_events(session);
2498 reader__release_decomp(&rd);
2499 session->one_mmap = false;
2500 return err;
2501}
2502
2503/*
2504 * Processing 2 MB of data from each reader in sequence,
2505 * because that's the way the ordered events sorting works
2506 * most efficiently.
2507 */
2508#define READER_MAX_SIZE (2 * 1024 * 1024)
2509
2510/*
2511 * This function reads, merge and process directory data.
2512 * It assumens the version 1 of directory data, where each
2513 * data file holds per-cpu data, already sorted by kernel.
2514 */
2515static int __perf_session__process_dir_events(struct perf_session *session)
2516{
2517 struct perf_data *data = session->data;
2518 struct perf_tool *tool = session->tool;
2519 int i, ret, readers, nr_readers;
2520 struct ui_progress prog;
2521 u64 total_size = perf_data__size(session->data);
2522 struct reader *rd;
2523
2524 perf_tool__fill_defaults(tool);
2525
2526 ui_progress__init_size(&prog, total_size, "Sorting events...");
2527
2528 nr_readers = 1;
2529 for (i = 0; i < data->dir.nr; i++) {
2530 if (data->dir.files[i].size)
2531 nr_readers++;
2532 }
2533
2534 rd = zalloc(nr_readers * sizeof(struct reader));
2535 if (!rd)
2536 return -ENOMEM;
2537
2538 rd[0] = (struct reader) {
2539 .fd = perf_data__fd(session->data),
2540 .path = session->data->file.path,
2541 .data_size = session->header.data_size,
2542 .data_offset = session->header.data_offset,
2543 .process = process_simple,
2544 .in_place_update = session->data->in_place_update,
2545 };
2546 ret = reader__init(&rd[0], NULL);
2547 if (ret)
2548 goto out_err;
2549 ret = reader__mmap(&rd[0], session);
2550 if (ret)
2551 goto out_err;
2552 readers = 1;
2553
2554 for (i = 0; i < data->dir.nr; i++) {
2555 if (!data->dir.files[i].size)
2556 continue;
2557 rd[readers] = (struct reader) {
2558 .fd = data->dir.files[i].fd,
2559 .path = data->dir.files[i].path,
2560 .data_size = data->dir.files[i].size,
2561 .data_offset = 0,
2562 .process = process_simple,
2563 .in_place_update = session->data->in_place_update,
2564 };
2565 ret = reader__init(&rd[readers], NULL);
2566 if (ret)
2567 goto out_err;
2568 ret = reader__mmap(&rd[readers], session);
2569 if (ret)
2570 goto out_err;
2571 readers++;
2572 }
2573
2574 i = 0;
2575 while (readers) {
2576 if (session_done())
2577 break;
2578
2579 if (rd[i].done) {
2580 i = (i + 1) % nr_readers;
2581 continue;
2582 }
2583 if (reader__eof(&rd[i])) {
2584 rd[i].done = true;
2585 readers--;
2586 continue;
2587 }
2588
2589 session->active_decomp = &rd[i].decomp_data;
2590 ret = reader__read_event(&rd[i], session, &prog);
2591 if (ret < 0) {
2592 goto out_err;
2593 } else if (ret == READER_NODATA) {
2594 ret = reader__mmap(&rd[i], session);
2595 if (ret)
2596 goto out_err;
2597 }
2598
2599 if (rd[i].size >= READER_MAX_SIZE) {
2600 rd[i].size = 0;
2601 i = (i + 1) % nr_readers;
2602 }
2603 }
2604
2605 ret = ordered_events__flush(&session->ordered_events, OE_FLUSH__FINAL);
2606 if (ret)
2607 goto out_err;
2608
2609 ret = perf_session__flush_thread_stacks(session);
2610out_err:
2611 ui_progress__finish();
2612
2613 if (!tool->no_warn)
2614 perf_session__warn_about_errors(session);
2615
2616 /*
2617 * We may switching perf.data output, make ordered_events
2618 * reusable.
2619 */
2620 ordered_events__reinit(&session->ordered_events);
2621
2622 session->one_mmap = false;
2623
2624 session->active_decomp = &session->decomp_data;
2625 for (i = 0; i < nr_readers; i++)
2626 reader__release_decomp(&rd[i]);
2627 zfree(&rd);
2628
2629 return ret;
2630}
2631
2632int perf_session__process_events(struct perf_session *session)
2633{
2634 if (perf_session__register_idle_thread(session) < 0)
2635 return -ENOMEM;
2636
2637 if (perf_data__is_pipe(session->data))
2638 return __perf_session__process_pipe_events(session);
2639
2640 if (perf_data__is_dir(session->data) && session->data->dir.nr)
2641 return __perf_session__process_dir_events(session);
2642
2643 return __perf_session__process_events(session);
2644}
2645
2646bool perf_session__has_traces(struct perf_session *session, const char *msg)
2647{
2648 struct evsel *evsel;
2649
2650 evlist__for_each_entry(session->evlist, evsel) {
2651 if (evsel->core.attr.type == PERF_TYPE_TRACEPOINT)
2652 return true;
2653 }
2654
2655 pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg);
2656 return false;
2657}
2658
2659int map__set_kallsyms_ref_reloc_sym(struct map *map, const char *symbol_name, u64 addr)
2660{
2661 char *bracket;
2662 struct ref_reloc_sym *ref;
2663 struct kmap *kmap;
2664
2665 ref = zalloc(sizeof(struct ref_reloc_sym));
2666 if (ref == NULL)
2667 return -ENOMEM;
2668
2669 ref->name = strdup(symbol_name);
2670 if (ref->name == NULL) {
2671 free(ref);
2672 return -ENOMEM;
2673 }
2674
2675 bracket = strchr(ref->name, ']');
2676 if (bracket)
2677 *bracket = '\0';
2678
2679 ref->addr = addr;
2680
2681 kmap = map__kmap(map);
2682 if (kmap)
2683 kmap->ref_reloc_sym = ref;
2684
2685 return 0;
2686}
2687
2688size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp)
2689{
2690 return machines__fprintf_dsos(&session->machines, fp);
2691}
2692
2693size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp,
2694 bool (skip)(struct dso *dso, int parm), int parm)
2695{
2696 return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm);
2697}
2698
2699size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp,
2700 bool skip_empty)
2701{
2702 size_t ret;
2703 const char *msg = "";
2704
2705 if (perf_header__has_feat(&session->header, HEADER_AUXTRACE))
2706 msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)";
2707
2708 ret = fprintf(fp, "\nAggregated stats:%s\n", msg);
2709
2710 ret += events_stats__fprintf(&session->evlist->stats, fp, skip_empty);
2711 return ret;
2712}
2713
2714size_t perf_session__fprintf(struct perf_session *session, FILE *fp)
2715{
2716 /*
2717 * FIXME: Here we have to actually print all the machines in this
2718 * session, not just the host...
2719 */
2720 return machine__fprintf(&session->machines.host, fp);
2721}
2722
2723void perf_session__dump_kmaps(struct perf_session *session)
2724{
2725 int save_verbose = verbose;
2726
2727 fflush(stdout);
2728 fprintf(stderr, "Kernel and module maps:\n");
2729 verbose = 0; /* Suppress verbose to print a summary only */
2730 maps__fprintf(machine__kernel_maps(&session->machines.host), stderr);
2731 verbose = save_verbose;
2732}
2733
2734struct evsel *perf_session__find_first_evtype(struct perf_session *session,
2735 unsigned int type)
2736{
2737 struct evsel *pos;
2738
2739 evlist__for_each_entry(session->evlist, pos) {
2740 if (pos->core.attr.type == type)
2741 return pos;
2742 }
2743 return NULL;
2744}
2745
2746int perf_session__cpu_bitmap(struct perf_session *session,
2747 const char *cpu_list, unsigned long *cpu_bitmap)
2748{
2749 int i, err = -1;
2750 struct perf_cpu_map *map;
2751 int nr_cpus = min(session->header.env.nr_cpus_avail, MAX_NR_CPUS);
2752
2753 for (i = 0; i < PERF_TYPE_MAX; ++i) {
2754 struct evsel *evsel;
2755
2756 evsel = perf_session__find_first_evtype(session, i);
2757 if (!evsel)
2758 continue;
2759
2760 if (!(evsel->core.attr.sample_type & PERF_SAMPLE_CPU)) {
2761 pr_err("File does not contain CPU events. "
2762 "Remove -C option to proceed.\n");
2763 return -1;
2764 }
2765 }
2766
2767 map = perf_cpu_map__new(cpu_list);
2768 if (map == NULL) {
2769 pr_err("Invalid cpu_list\n");
2770 return -1;
2771 }
2772
2773 for (i = 0; i < perf_cpu_map__nr(map); i++) {
2774 struct perf_cpu cpu = perf_cpu_map__cpu(map, i);
2775
2776 if (cpu.cpu >= nr_cpus) {
2777 pr_err("Requested CPU %d too large. "
2778 "Consider raising MAX_NR_CPUS\n", cpu.cpu);
2779 goto out_delete_map;
2780 }
2781
2782 __set_bit(cpu.cpu, cpu_bitmap);
2783 }
2784
2785 err = 0;
2786
2787out_delete_map:
2788 perf_cpu_map__put(map);
2789 return err;
2790}
2791
2792void perf_session__fprintf_info(struct perf_session *session, FILE *fp,
2793 bool full)
2794{
2795 if (session == NULL || fp == NULL)
2796 return;
2797
2798 fprintf(fp, "# ========\n");
2799 perf_header__fprintf_info(session, fp, full);
2800 fprintf(fp, "# ========\n#\n");
2801}
2802
2803static int perf_session__register_guest(struct perf_session *session, pid_t machine_pid)
2804{
2805 struct machine *machine = machines__findnew(&session->machines, machine_pid);
2806 struct thread *thread;
2807
2808 if (!machine)
2809 return -ENOMEM;
2810
2811 machine->single_address_space = session->machines.host.single_address_space;
2812
2813 thread = machine__idle_thread(machine);
2814 if (!thread)
2815 return -ENOMEM;
2816 thread__put(thread);
2817
2818 machine->kallsyms_filename = perf_data__guest_kallsyms_name(session->data, machine_pid);
2819
2820 return 0;
2821}
2822
2823static int perf_session__set_guest_cpu(struct perf_session *session, pid_t pid,
2824 pid_t tid, int guest_cpu)
2825{
2826 struct machine *machine = &session->machines.host;
2827 struct thread *thread = machine__findnew_thread(machine, pid, tid);
2828
2829 if (!thread)
2830 return -ENOMEM;
2831 thread__set_guest_cpu(thread, guest_cpu);
2832 thread__put(thread);
2833
2834 return 0;
2835}
2836
2837int perf_event__process_id_index(struct perf_session *session,
2838 union perf_event *event)
2839{
2840 struct evlist *evlist = session->evlist;
2841 struct perf_record_id_index *ie = &event->id_index;
2842 size_t sz = ie->header.size - sizeof(*ie);
2843 size_t i, nr, max_nr;
2844 size_t e1_sz = sizeof(struct id_index_entry);
2845 size_t e2_sz = sizeof(struct id_index_entry_2);
2846 size_t etot_sz = e1_sz + e2_sz;
2847 struct id_index_entry_2 *e2;
2848 pid_t last_pid = 0;
2849
2850 max_nr = sz / e1_sz;
2851 nr = ie->nr;
2852 if (nr > max_nr) {
2853 printf("Too big: nr %zu max_nr %zu\n", nr, max_nr);
2854 return -EINVAL;
2855 }
2856
2857 if (sz >= nr * etot_sz) {
2858 max_nr = sz / etot_sz;
2859 if (nr > max_nr) {
2860 printf("Too big2: nr %zu max_nr %zu\n", nr, max_nr);
2861 return -EINVAL;
2862 }
2863 e2 = (void *)ie + sizeof(*ie) + nr * e1_sz;
2864 } else {
2865 e2 = NULL;
2866 }
2867
2868 if (dump_trace)
2869 fprintf(stdout, " nr: %zu\n", nr);
2870
2871 for (i = 0; i < nr; i++, (e2 ? e2++ : 0)) {
2872 struct id_index_entry *e = &ie->entries[i];
2873 struct perf_sample_id *sid;
2874 int ret;
2875
2876 if (dump_trace) {
2877 fprintf(stdout, " ... id: %"PRI_lu64, e->id);
2878 fprintf(stdout, " idx: %"PRI_lu64, e->idx);
2879 fprintf(stdout, " cpu: %"PRI_ld64, e->cpu);
2880 fprintf(stdout, " tid: %"PRI_ld64, e->tid);
2881 if (e2) {
2882 fprintf(stdout, " machine_pid: %"PRI_ld64, e2->machine_pid);
2883 fprintf(stdout, " vcpu: %"PRI_lu64"\n", e2->vcpu);
2884 } else {
2885 fprintf(stdout, "\n");
2886 }
2887 }
2888
2889 sid = evlist__id2sid(evlist, e->id);
2890 if (!sid)
2891 return -ENOENT;
2892
2893 sid->idx = e->idx;
2894 sid->cpu.cpu = e->cpu;
2895 sid->tid = e->tid;
2896
2897 if (!e2)
2898 continue;
2899
2900 sid->machine_pid = e2->machine_pid;
2901 sid->vcpu.cpu = e2->vcpu;
2902
2903 if (!sid->machine_pid)
2904 continue;
2905
2906 if (sid->machine_pid != last_pid) {
2907 ret = perf_session__register_guest(session, sid->machine_pid);
2908 if (ret)
2909 return ret;
2910 last_pid = sid->machine_pid;
2911 perf_guest = true;
2912 }
2913
2914 ret = perf_session__set_guest_cpu(session, sid->machine_pid, e->tid, e2->vcpu);
2915 if (ret)
2916 return ret;
2917 }
2918 return 0;
2919}