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 "tool.h"
14#include "sort.h"
15#include "util.h"
16#include "cpumap.h"
17
18static int perf_session__open(struct perf_session *self, bool force)
19{
20 struct stat input_stat;
21
22 if (!strcmp(self->filename, "-")) {
23 self->fd_pipe = true;
24 self->fd = STDIN_FILENO;
25
26 if (perf_session__read_header(self, self->fd) < 0)
27 pr_err("incompatible file format (rerun with -v to learn more)");
28
29 return 0;
30 }
31
32 self->fd = open(self->filename, O_RDONLY);
33 if (self->fd < 0) {
34 int err = errno;
35
36 pr_err("failed to open %s: %s", self->filename, strerror(err));
37 if (err == ENOENT && !strcmp(self->filename, "perf.data"))
38 pr_err(" (try 'perf record' first)");
39 pr_err("\n");
40 return -errno;
41 }
42
43 if (fstat(self->fd, &input_stat) < 0)
44 goto out_close;
45
46 if (!force && input_stat.st_uid && (input_stat.st_uid != geteuid())) {
47 pr_err("file %s not owned by current user or root\n",
48 self->filename);
49 goto out_close;
50 }
51
52 if (!input_stat.st_size) {
53 pr_info("zero-sized file (%s), nothing to do!\n",
54 self->filename);
55 goto out_close;
56 }
57
58 if (perf_session__read_header(self, self->fd) < 0) {
59 pr_err("incompatible file format (rerun with -v to learn more)");
60 goto out_close;
61 }
62
63 if (!perf_evlist__valid_sample_type(self->evlist)) {
64 pr_err("non matching sample_type");
65 goto out_close;
66 }
67
68 if (!perf_evlist__valid_sample_id_all(self->evlist)) {
69 pr_err("non matching sample_id_all");
70 goto out_close;
71 }
72
73 self->size = input_stat.st_size;
74 return 0;
75
76out_close:
77 close(self->fd);
78 self->fd = -1;
79 return -1;
80}
81
82void perf_session__update_sample_type(struct perf_session *self)
83{
84 self->sample_type = perf_evlist__sample_type(self->evlist);
85 self->sample_size = __perf_evsel__sample_size(self->sample_type);
86 self->sample_id_all = perf_evlist__sample_id_all(self->evlist);
87 self->id_hdr_size = perf_evlist__id_hdr_size(self->evlist);
88 self->host_machine.id_hdr_size = self->id_hdr_size;
89}
90
91int perf_session__create_kernel_maps(struct perf_session *self)
92{
93 int ret = machine__create_kernel_maps(&self->host_machine);
94
95 if (ret >= 0)
96 ret = machines__create_guest_kernel_maps(&self->machines);
97 return ret;
98}
99
100static void perf_session__destroy_kernel_maps(struct perf_session *self)
101{
102 machine__destroy_kernel_maps(&self->host_machine);
103 machines__destroy_guest_kernel_maps(&self->machines);
104}
105
106struct perf_session *perf_session__new(const char *filename, int mode,
107 bool force, bool repipe,
108 struct perf_tool *tool)
109{
110 struct perf_session *self;
111 struct stat st;
112 size_t len;
113
114 if (!filename || !strlen(filename)) {
115 if (!fstat(STDIN_FILENO, &st) && S_ISFIFO(st.st_mode))
116 filename = "-";
117 else
118 filename = "perf.data";
119 }
120
121 len = strlen(filename);
122 self = zalloc(sizeof(*self) + len);
123
124 if (self == NULL)
125 goto out;
126
127 memcpy(self->filename, filename, len);
128 /*
129 * On 64bit we can mmap the data file in one go. No need for tiny mmap
130 * slices. On 32bit we use 32MB.
131 */
132#if BITS_PER_LONG == 64
133 self->mmap_window = ULLONG_MAX;
134#else
135 self->mmap_window = 32 * 1024 * 1024ULL;
136#endif
137 self->machines = RB_ROOT;
138 self->repipe = repipe;
139 INIT_LIST_HEAD(&self->ordered_samples.samples);
140 INIT_LIST_HEAD(&self->ordered_samples.sample_cache);
141 INIT_LIST_HEAD(&self->ordered_samples.to_free);
142 machine__init(&self->host_machine, "", HOST_KERNEL_ID);
143 hists__init(&self->hists);
144
145 if (mode == O_RDONLY) {
146 if (perf_session__open(self, force) < 0)
147 goto out_delete;
148 perf_session__update_sample_type(self);
149 } else if (mode == O_WRONLY) {
150 /*
151 * In O_RDONLY mode this will be performed when reading the
152 * kernel MMAP event, in perf_event__process_mmap().
153 */
154 if (perf_session__create_kernel_maps(self) < 0)
155 goto out_delete;
156 }
157
158 if (tool && tool->ordering_requires_timestamps &&
159 tool->ordered_samples && !self->sample_id_all) {
160 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n");
161 tool->ordered_samples = false;
162 }
163
164out:
165 return self;
166out_delete:
167 perf_session__delete(self);
168 return NULL;
169}
170
171static void machine__delete_dead_threads(struct machine *machine)
172{
173 struct thread *n, *t;
174
175 list_for_each_entry_safe(t, n, &machine->dead_threads, node) {
176 list_del(&t->node);
177 thread__delete(t);
178 }
179}
180
181static void perf_session__delete_dead_threads(struct perf_session *session)
182{
183 machine__delete_dead_threads(&session->host_machine);
184}
185
186static void machine__delete_threads(struct machine *self)
187{
188 struct rb_node *nd = rb_first(&self->threads);
189
190 while (nd) {
191 struct thread *t = rb_entry(nd, struct thread, rb_node);
192
193 rb_erase(&t->rb_node, &self->threads);
194 nd = rb_next(nd);
195 thread__delete(t);
196 }
197}
198
199static void perf_session__delete_threads(struct perf_session *session)
200{
201 machine__delete_threads(&session->host_machine);
202}
203
204void perf_session__delete(struct perf_session *self)
205{
206 perf_session__destroy_kernel_maps(self);
207 perf_session__delete_dead_threads(self);
208 perf_session__delete_threads(self);
209 machine__exit(&self->host_machine);
210 close(self->fd);
211 free(self);
212}
213
214void machine__remove_thread(struct machine *self, struct thread *th)
215{
216 self->last_match = NULL;
217 rb_erase(&th->rb_node, &self->threads);
218 /*
219 * We may have references to this thread, for instance in some hist_entry
220 * instances, so just move them to a separate list.
221 */
222 list_add_tail(&th->node, &self->dead_threads);
223}
224
225static bool symbol__match_parent_regex(struct symbol *sym)
226{
227 if (sym->name && !regexec(&parent_regex, sym->name, 0, NULL, 0))
228 return 1;
229
230 return 0;
231}
232
233static const u8 cpumodes[] = {
234 PERF_RECORD_MISC_USER,
235 PERF_RECORD_MISC_KERNEL,
236 PERF_RECORD_MISC_GUEST_USER,
237 PERF_RECORD_MISC_GUEST_KERNEL
238};
239#define NCPUMODES (sizeof(cpumodes)/sizeof(u8))
240
241static void ip__resolve_ams(struct machine *self, struct thread *thread,
242 struct addr_map_symbol *ams,
243 u64 ip)
244{
245 struct addr_location al;
246 size_t i;
247 u8 m;
248
249 memset(&al, 0, sizeof(al));
250
251 for (i = 0; i < NCPUMODES; i++) {
252 m = cpumodes[i];
253 /*
254 * We cannot use the header.misc hint to determine whether a
255 * branch stack address is user, kernel, guest, hypervisor.
256 * Branches may straddle the kernel/user/hypervisor boundaries.
257 * Thus, we have to try consecutively until we find a match
258 * or else, the symbol is unknown
259 */
260 thread__find_addr_location(thread, self, m, MAP__FUNCTION,
261 ip, &al, NULL);
262 if (al.sym)
263 goto found;
264 }
265found:
266 ams->addr = ip;
267 ams->al_addr = al.addr;
268 ams->sym = al.sym;
269 ams->map = al.map;
270}
271
272struct branch_info *machine__resolve_bstack(struct machine *self,
273 struct thread *thr,
274 struct branch_stack *bs)
275{
276 struct branch_info *bi;
277 unsigned int i;
278
279 bi = calloc(bs->nr, sizeof(struct branch_info));
280 if (!bi)
281 return NULL;
282
283 for (i = 0; i < bs->nr; i++) {
284 ip__resolve_ams(self, thr, &bi[i].to, bs->entries[i].to);
285 ip__resolve_ams(self, thr, &bi[i].from, bs->entries[i].from);
286 bi[i].flags = bs->entries[i].flags;
287 }
288 return bi;
289}
290
291int machine__resolve_callchain(struct machine *self,
292 struct perf_evsel *evsel __used,
293 struct thread *thread,
294 struct ip_callchain *chain,
295 struct symbol **parent)
296{
297 u8 cpumode = PERF_RECORD_MISC_USER;
298 unsigned int i;
299 int err;
300
301 callchain_cursor_reset(&callchain_cursor);
302
303 if (chain->nr > PERF_MAX_STACK_DEPTH) {
304 pr_warning("corrupted callchain. skipping...\n");
305 return 0;
306 }
307
308 for (i = 0; i < chain->nr; i++) {
309 u64 ip;
310 struct addr_location al;
311
312 if (callchain_param.order == ORDER_CALLEE)
313 ip = chain->ips[i];
314 else
315 ip = chain->ips[chain->nr - i - 1];
316
317 if (ip >= PERF_CONTEXT_MAX) {
318 switch (ip) {
319 case PERF_CONTEXT_HV:
320 cpumode = PERF_RECORD_MISC_HYPERVISOR; break;
321 case PERF_CONTEXT_KERNEL:
322 cpumode = PERF_RECORD_MISC_KERNEL; break;
323 case PERF_CONTEXT_USER:
324 cpumode = PERF_RECORD_MISC_USER; break;
325 default:
326 pr_debug("invalid callchain context: "
327 "%"PRId64"\n", (s64) ip);
328 /*
329 * It seems the callchain is corrupted.
330 * Discard all.
331 */
332 callchain_cursor_reset(&callchain_cursor);
333 return 0;
334 }
335 continue;
336 }
337
338 al.filtered = false;
339 thread__find_addr_location(thread, self, cpumode,
340 MAP__FUNCTION, ip, &al, NULL);
341 if (al.sym != NULL) {
342 if (sort__has_parent && !*parent &&
343 symbol__match_parent_regex(al.sym))
344 *parent = al.sym;
345 if (!symbol_conf.use_callchain)
346 break;
347 }
348
349 err = callchain_cursor_append(&callchain_cursor,
350 ip, al.map, al.sym);
351 if (err)
352 return err;
353 }
354
355 return 0;
356}
357
358static int process_event_synth_tracing_data_stub(union perf_event *event __used,
359 struct perf_session *session __used)
360{
361 dump_printf(": unhandled!\n");
362 return 0;
363}
364
365static int process_event_synth_attr_stub(union perf_event *event __used,
366 struct perf_evlist **pevlist __used)
367{
368 dump_printf(": unhandled!\n");
369 return 0;
370}
371
372static int process_event_sample_stub(struct perf_tool *tool __used,
373 union perf_event *event __used,
374 struct perf_sample *sample __used,
375 struct perf_evsel *evsel __used,
376 struct machine *machine __used)
377{
378 dump_printf(": unhandled!\n");
379 return 0;
380}
381
382static int process_event_stub(struct perf_tool *tool __used,
383 union perf_event *event __used,
384 struct perf_sample *sample __used,
385 struct machine *machine __used)
386{
387 dump_printf(": unhandled!\n");
388 return 0;
389}
390
391static int process_finished_round_stub(struct perf_tool *tool __used,
392 union perf_event *event __used,
393 struct perf_session *perf_session __used)
394{
395 dump_printf(": unhandled!\n");
396 return 0;
397}
398
399static int process_event_type_stub(struct perf_tool *tool __used,
400 union perf_event *event __used)
401{
402 dump_printf(": unhandled!\n");
403 return 0;
404}
405
406static int process_finished_round(struct perf_tool *tool,
407 union perf_event *event,
408 struct perf_session *session);
409
410static void perf_tool__fill_defaults(struct perf_tool *tool)
411{
412 if (tool->sample == NULL)
413 tool->sample = process_event_sample_stub;
414 if (tool->mmap == NULL)
415 tool->mmap = process_event_stub;
416 if (tool->comm == NULL)
417 tool->comm = process_event_stub;
418 if (tool->fork == NULL)
419 tool->fork = process_event_stub;
420 if (tool->exit == NULL)
421 tool->exit = process_event_stub;
422 if (tool->lost == NULL)
423 tool->lost = perf_event__process_lost;
424 if (tool->read == NULL)
425 tool->read = process_event_sample_stub;
426 if (tool->throttle == NULL)
427 tool->throttle = process_event_stub;
428 if (tool->unthrottle == NULL)
429 tool->unthrottle = process_event_stub;
430 if (tool->attr == NULL)
431 tool->attr = process_event_synth_attr_stub;
432 if (tool->event_type == NULL)
433 tool->event_type = process_event_type_stub;
434 if (tool->tracing_data == NULL)
435 tool->tracing_data = process_event_synth_tracing_data_stub;
436 if (tool->build_id == NULL)
437 tool->build_id = process_finished_round_stub;
438 if (tool->finished_round == NULL) {
439 if (tool->ordered_samples)
440 tool->finished_round = process_finished_round;
441 else
442 tool->finished_round = process_finished_round_stub;
443 }
444}
445
446void mem_bswap_32(void *src, int byte_size)
447{
448 u32 *m = src;
449 while (byte_size > 0) {
450 *m = bswap_32(*m);
451 byte_size -= sizeof(u32);
452 ++m;
453 }
454}
455
456void mem_bswap_64(void *src, int byte_size)
457{
458 u64 *m = src;
459
460 while (byte_size > 0) {
461 *m = bswap_64(*m);
462 byte_size -= sizeof(u64);
463 ++m;
464 }
465}
466
467static void swap_sample_id_all(union perf_event *event, void *data)
468{
469 void *end = (void *) event + event->header.size;
470 int size = end - data;
471
472 BUG_ON(size % sizeof(u64));
473 mem_bswap_64(data, size);
474}
475
476static void perf_event__all64_swap(union perf_event *event,
477 bool sample_id_all __used)
478{
479 struct perf_event_header *hdr = &event->header;
480 mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr));
481}
482
483static void perf_event__comm_swap(union perf_event *event, bool sample_id_all)
484{
485 event->comm.pid = bswap_32(event->comm.pid);
486 event->comm.tid = bswap_32(event->comm.tid);
487
488 if (sample_id_all) {
489 void *data = &event->comm.comm;
490
491 data += ALIGN(strlen(data) + 1, sizeof(u64));
492 swap_sample_id_all(event, data);
493 }
494}
495
496static void perf_event__mmap_swap(union perf_event *event,
497 bool sample_id_all)
498{
499 event->mmap.pid = bswap_32(event->mmap.pid);
500 event->mmap.tid = bswap_32(event->mmap.tid);
501 event->mmap.start = bswap_64(event->mmap.start);
502 event->mmap.len = bswap_64(event->mmap.len);
503 event->mmap.pgoff = bswap_64(event->mmap.pgoff);
504
505 if (sample_id_all) {
506 void *data = &event->mmap.filename;
507
508 data += ALIGN(strlen(data) + 1, sizeof(u64));
509 swap_sample_id_all(event, data);
510 }
511}
512
513static void perf_event__task_swap(union perf_event *event, bool sample_id_all)
514{
515 event->fork.pid = bswap_32(event->fork.pid);
516 event->fork.tid = bswap_32(event->fork.tid);
517 event->fork.ppid = bswap_32(event->fork.ppid);
518 event->fork.ptid = bswap_32(event->fork.ptid);
519 event->fork.time = bswap_64(event->fork.time);
520
521 if (sample_id_all)
522 swap_sample_id_all(event, &event->fork + 1);
523}
524
525static void perf_event__read_swap(union perf_event *event, bool sample_id_all)
526{
527 event->read.pid = bswap_32(event->read.pid);
528 event->read.tid = bswap_32(event->read.tid);
529 event->read.value = bswap_64(event->read.value);
530 event->read.time_enabled = bswap_64(event->read.time_enabled);
531 event->read.time_running = bswap_64(event->read.time_running);
532 event->read.id = bswap_64(event->read.id);
533
534 if (sample_id_all)
535 swap_sample_id_all(event, &event->read + 1);
536}
537
538static u8 revbyte(u8 b)
539{
540 int rev = (b >> 4) | ((b & 0xf) << 4);
541 rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2);
542 rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1);
543 return (u8) rev;
544}
545
546/*
547 * XXX this is hack in attempt to carry flags bitfield
548 * throught endian village. ABI says:
549 *
550 * Bit-fields are allocated from right to left (least to most significant)
551 * on little-endian implementations and from left to right (most to least
552 * significant) on big-endian implementations.
553 *
554 * The above seems to be byte specific, so we need to reverse each
555 * byte of the bitfield. 'Internet' also says this might be implementation
556 * specific and we probably need proper fix and carry perf_event_attr
557 * bitfield flags in separate data file FEAT_ section. Thought this seems
558 * to work for now.
559 */
560static void swap_bitfield(u8 *p, unsigned len)
561{
562 unsigned i;
563
564 for (i = 0; i < len; i++) {
565 *p = revbyte(*p);
566 p++;
567 }
568}
569
570/* exported for swapping attributes in file header */
571void perf_event__attr_swap(struct perf_event_attr *attr)
572{
573 attr->type = bswap_32(attr->type);
574 attr->size = bswap_32(attr->size);
575 attr->config = bswap_64(attr->config);
576 attr->sample_period = bswap_64(attr->sample_period);
577 attr->sample_type = bswap_64(attr->sample_type);
578 attr->read_format = bswap_64(attr->read_format);
579 attr->wakeup_events = bswap_32(attr->wakeup_events);
580 attr->bp_type = bswap_32(attr->bp_type);
581 attr->bp_addr = bswap_64(attr->bp_addr);
582 attr->bp_len = bswap_64(attr->bp_len);
583
584 swap_bitfield((u8 *) (&attr->read_format + 1), sizeof(u64));
585}
586
587static void perf_event__hdr_attr_swap(union perf_event *event,
588 bool sample_id_all __used)
589{
590 size_t size;
591
592 perf_event__attr_swap(&event->attr.attr);
593
594 size = event->header.size;
595 size -= (void *)&event->attr.id - (void *)event;
596 mem_bswap_64(event->attr.id, size);
597}
598
599static void perf_event__event_type_swap(union perf_event *event,
600 bool sample_id_all __used)
601{
602 event->event_type.event_type.event_id =
603 bswap_64(event->event_type.event_type.event_id);
604}
605
606static void perf_event__tracing_data_swap(union perf_event *event,
607 bool sample_id_all __used)
608{
609 event->tracing_data.size = bswap_32(event->tracing_data.size);
610}
611
612typedef void (*perf_event__swap_op)(union perf_event *event,
613 bool sample_id_all);
614
615static perf_event__swap_op perf_event__swap_ops[] = {
616 [PERF_RECORD_MMAP] = perf_event__mmap_swap,
617 [PERF_RECORD_COMM] = perf_event__comm_swap,
618 [PERF_RECORD_FORK] = perf_event__task_swap,
619 [PERF_RECORD_EXIT] = perf_event__task_swap,
620 [PERF_RECORD_LOST] = perf_event__all64_swap,
621 [PERF_RECORD_READ] = perf_event__read_swap,
622 [PERF_RECORD_SAMPLE] = perf_event__all64_swap,
623 [PERF_RECORD_HEADER_ATTR] = perf_event__hdr_attr_swap,
624 [PERF_RECORD_HEADER_EVENT_TYPE] = perf_event__event_type_swap,
625 [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
626 [PERF_RECORD_HEADER_BUILD_ID] = NULL,
627 [PERF_RECORD_HEADER_MAX] = NULL,
628};
629
630struct sample_queue {
631 u64 timestamp;
632 u64 file_offset;
633 union perf_event *event;
634 struct list_head list;
635};
636
637static void perf_session_free_sample_buffers(struct perf_session *session)
638{
639 struct ordered_samples *os = &session->ordered_samples;
640
641 while (!list_empty(&os->to_free)) {
642 struct sample_queue *sq;
643
644 sq = list_entry(os->to_free.next, struct sample_queue, list);
645 list_del(&sq->list);
646 free(sq);
647 }
648}
649
650static int perf_session_deliver_event(struct perf_session *session,
651 union perf_event *event,
652 struct perf_sample *sample,
653 struct perf_tool *tool,
654 u64 file_offset);
655
656static void flush_sample_queue(struct perf_session *s,
657 struct perf_tool *tool)
658{
659 struct ordered_samples *os = &s->ordered_samples;
660 struct list_head *head = &os->samples;
661 struct sample_queue *tmp, *iter;
662 struct perf_sample sample;
663 u64 limit = os->next_flush;
664 u64 last_ts = os->last_sample ? os->last_sample->timestamp : 0ULL;
665 unsigned idx = 0, progress_next = os->nr_samples / 16;
666 int ret;
667
668 if (!tool->ordered_samples || !limit)
669 return;
670
671 list_for_each_entry_safe(iter, tmp, head, list) {
672 if (iter->timestamp > limit)
673 break;
674
675 ret = perf_session__parse_sample(s, iter->event, &sample);
676 if (ret)
677 pr_err("Can't parse sample, err = %d\n", ret);
678 else
679 perf_session_deliver_event(s, iter->event, &sample, tool,
680 iter->file_offset);
681
682 os->last_flush = iter->timestamp;
683 list_del(&iter->list);
684 list_add(&iter->list, &os->sample_cache);
685 if (++idx >= progress_next) {
686 progress_next += os->nr_samples / 16;
687 ui_progress__update(idx, os->nr_samples,
688 "Processing time ordered events...");
689 }
690 }
691
692 if (list_empty(head)) {
693 os->last_sample = NULL;
694 } else if (last_ts <= limit) {
695 os->last_sample =
696 list_entry(head->prev, struct sample_queue, list);
697 }
698
699 os->nr_samples = 0;
700}
701
702/*
703 * When perf record finishes a pass on every buffers, it records this pseudo
704 * event.
705 * We record the max timestamp t found in the pass n.
706 * Assuming these timestamps are monotonic across cpus, we know that if
707 * a buffer still has events with timestamps below t, they will be all
708 * available and then read in the pass n + 1.
709 * Hence when we start to read the pass n + 2, we can safely flush every
710 * events with timestamps below t.
711 *
712 * ============ PASS n =================
713 * CPU 0 | CPU 1
714 * |
715 * cnt1 timestamps | cnt2 timestamps
716 * 1 | 2
717 * 2 | 3
718 * - | 4 <--- max recorded
719 *
720 * ============ PASS n + 1 ==============
721 * CPU 0 | CPU 1
722 * |
723 * cnt1 timestamps | cnt2 timestamps
724 * 3 | 5
725 * 4 | 6
726 * 5 | 7 <---- max recorded
727 *
728 * Flush every events below timestamp 4
729 *
730 * ============ PASS n + 2 ==============
731 * CPU 0 | CPU 1
732 * |
733 * cnt1 timestamps | cnt2 timestamps
734 * 6 | 8
735 * 7 | 9
736 * - | 10
737 *
738 * Flush every events below timestamp 7
739 * etc...
740 */
741static int process_finished_round(struct perf_tool *tool,
742 union perf_event *event __used,
743 struct perf_session *session)
744{
745 flush_sample_queue(session, tool);
746 session->ordered_samples.next_flush = session->ordered_samples.max_timestamp;
747
748 return 0;
749}
750
751/* The queue is ordered by time */
752static void __queue_event(struct sample_queue *new, struct perf_session *s)
753{
754 struct ordered_samples *os = &s->ordered_samples;
755 struct sample_queue *sample = os->last_sample;
756 u64 timestamp = new->timestamp;
757 struct list_head *p;
758
759 ++os->nr_samples;
760 os->last_sample = new;
761
762 if (!sample) {
763 list_add(&new->list, &os->samples);
764 os->max_timestamp = timestamp;
765 return;
766 }
767
768 /*
769 * last_sample might point to some random place in the list as it's
770 * the last queued event. We expect that the new event is close to
771 * this.
772 */
773 if (sample->timestamp <= timestamp) {
774 while (sample->timestamp <= timestamp) {
775 p = sample->list.next;
776 if (p == &os->samples) {
777 list_add_tail(&new->list, &os->samples);
778 os->max_timestamp = timestamp;
779 return;
780 }
781 sample = list_entry(p, struct sample_queue, list);
782 }
783 list_add_tail(&new->list, &sample->list);
784 } else {
785 while (sample->timestamp > timestamp) {
786 p = sample->list.prev;
787 if (p == &os->samples) {
788 list_add(&new->list, &os->samples);
789 return;
790 }
791 sample = list_entry(p, struct sample_queue, list);
792 }
793 list_add(&new->list, &sample->list);
794 }
795}
796
797#define MAX_SAMPLE_BUFFER (64 * 1024 / sizeof(struct sample_queue))
798
799static int perf_session_queue_event(struct perf_session *s, union perf_event *event,
800 struct perf_sample *sample, u64 file_offset)
801{
802 struct ordered_samples *os = &s->ordered_samples;
803 struct list_head *sc = &os->sample_cache;
804 u64 timestamp = sample->time;
805 struct sample_queue *new;
806
807 if (!timestamp || timestamp == ~0ULL)
808 return -ETIME;
809
810 if (timestamp < s->ordered_samples.last_flush) {
811 printf("Warning: Timestamp below last timeslice flush\n");
812 return -EINVAL;
813 }
814
815 if (!list_empty(sc)) {
816 new = list_entry(sc->next, struct sample_queue, list);
817 list_del(&new->list);
818 } else if (os->sample_buffer) {
819 new = os->sample_buffer + os->sample_buffer_idx;
820 if (++os->sample_buffer_idx == MAX_SAMPLE_BUFFER)
821 os->sample_buffer = NULL;
822 } else {
823 os->sample_buffer = malloc(MAX_SAMPLE_BUFFER * sizeof(*new));
824 if (!os->sample_buffer)
825 return -ENOMEM;
826 list_add(&os->sample_buffer->list, &os->to_free);
827 os->sample_buffer_idx = 2;
828 new = os->sample_buffer + 1;
829 }
830
831 new->timestamp = timestamp;
832 new->file_offset = file_offset;
833 new->event = event;
834
835 __queue_event(new, s);
836
837 return 0;
838}
839
840static void callchain__printf(struct perf_sample *sample)
841{
842 unsigned int i;
843
844 printf("... chain: nr:%" PRIu64 "\n", sample->callchain->nr);
845
846 for (i = 0; i < sample->callchain->nr; i++)
847 printf("..... %2d: %016" PRIx64 "\n",
848 i, sample->callchain->ips[i]);
849}
850
851static void branch_stack__printf(struct perf_sample *sample)
852{
853 uint64_t i;
854
855 printf("... branch stack: nr:%" PRIu64 "\n", sample->branch_stack->nr);
856
857 for (i = 0; i < sample->branch_stack->nr; i++)
858 printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 "\n",
859 i, sample->branch_stack->entries[i].from,
860 sample->branch_stack->entries[i].to);
861}
862
863static void perf_session__print_tstamp(struct perf_session *session,
864 union perf_event *event,
865 struct perf_sample *sample)
866{
867 if (event->header.type != PERF_RECORD_SAMPLE &&
868 !session->sample_id_all) {
869 fputs("-1 -1 ", stdout);
870 return;
871 }
872
873 if ((session->sample_type & PERF_SAMPLE_CPU))
874 printf("%u ", sample->cpu);
875
876 if (session->sample_type & PERF_SAMPLE_TIME)
877 printf("%" PRIu64 " ", sample->time);
878}
879
880static void dump_event(struct perf_session *session, union perf_event *event,
881 u64 file_offset, struct perf_sample *sample)
882{
883 if (!dump_trace)
884 return;
885
886 printf("\n%#" PRIx64 " [%#x]: event: %d\n",
887 file_offset, event->header.size, event->header.type);
888
889 trace_event(event);
890
891 if (sample)
892 perf_session__print_tstamp(session, event, sample);
893
894 printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset,
895 event->header.size, perf_event__name(event->header.type));
896}
897
898static void dump_sample(struct perf_session *session, union perf_event *event,
899 struct perf_sample *sample)
900{
901 if (!dump_trace)
902 return;
903
904 printf("(IP, %d): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n",
905 event->header.misc, sample->pid, sample->tid, sample->ip,
906 sample->period, sample->addr);
907
908 if (session->sample_type & PERF_SAMPLE_CALLCHAIN)
909 callchain__printf(sample);
910
911 if (session->sample_type & PERF_SAMPLE_BRANCH_STACK)
912 branch_stack__printf(sample);
913}
914
915static struct machine *
916 perf_session__find_machine_for_cpumode(struct perf_session *session,
917 union perf_event *event)
918{
919 const u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
920
921 if (cpumode == PERF_RECORD_MISC_GUEST_KERNEL && perf_guest) {
922 u32 pid;
923
924 if (event->header.type == PERF_RECORD_MMAP)
925 pid = event->mmap.pid;
926 else
927 pid = event->ip.pid;
928
929 return perf_session__findnew_machine(session, pid);
930 }
931
932 return perf_session__find_host_machine(session);
933}
934
935static int perf_session_deliver_event(struct perf_session *session,
936 union perf_event *event,
937 struct perf_sample *sample,
938 struct perf_tool *tool,
939 u64 file_offset)
940{
941 struct perf_evsel *evsel;
942 struct machine *machine;
943
944 dump_event(session, event, file_offset, sample);
945
946 evsel = perf_evlist__id2evsel(session->evlist, sample->id);
947 if (evsel != NULL && event->header.type != PERF_RECORD_SAMPLE) {
948 /*
949 * XXX We're leaving PERF_RECORD_SAMPLE unnacounted here
950 * because the tools right now may apply filters, discarding
951 * some of the samples. For consistency, in the future we
952 * should have something like nr_filtered_samples and remove
953 * the sample->period from total_sample_period, etc, KISS for
954 * now tho.
955 *
956 * Also testing against NULL allows us to handle files without
957 * attr.sample_id_all and/or without PERF_SAMPLE_ID. In the
958 * future probably it'll be a good idea to restrict event
959 * processing via perf_session to files with both set.
960 */
961 hists__inc_nr_events(&evsel->hists, event->header.type);
962 }
963
964 machine = perf_session__find_machine_for_cpumode(session, event);
965
966 switch (event->header.type) {
967 case PERF_RECORD_SAMPLE:
968 dump_sample(session, event, sample);
969 if (evsel == NULL) {
970 ++session->hists.stats.nr_unknown_id;
971 return 0;
972 }
973 if (machine == NULL) {
974 ++session->hists.stats.nr_unprocessable_samples;
975 return 0;
976 }
977 return tool->sample(tool, event, sample, evsel, machine);
978 case PERF_RECORD_MMAP:
979 return tool->mmap(tool, event, sample, machine);
980 case PERF_RECORD_COMM:
981 return tool->comm(tool, event, sample, machine);
982 case PERF_RECORD_FORK:
983 return tool->fork(tool, event, sample, machine);
984 case PERF_RECORD_EXIT:
985 return tool->exit(tool, event, sample, machine);
986 case PERF_RECORD_LOST:
987 if (tool->lost == perf_event__process_lost)
988 session->hists.stats.total_lost += event->lost.lost;
989 return tool->lost(tool, event, sample, machine);
990 case PERF_RECORD_READ:
991 return tool->read(tool, event, sample, evsel, machine);
992 case PERF_RECORD_THROTTLE:
993 return tool->throttle(tool, event, sample, machine);
994 case PERF_RECORD_UNTHROTTLE:
995 return tool->unthrottle(tool, event, sample, machine);
996 default:
997 ++session->hists.stats.nr_unknown_events;
998 return -1;
999 }
1000}
1001
1002static int perf_session__preprocess_sample(struct perf_session *session,
1003 union perf_event *event, struct perf_sample *sample)
1004{
1005 if (event->header.type != PERF_RECORD_SAMPLE ||
1006 !(session->sample_type & PERF_SAMPLE_CALLCHAIN))
1007 return 0;
1008
1009 if (!ip_callchain__valid(sample->callchain, event)) {
1010 pr_debug("call-chain problem with event, skipping it.\n");
1011 ++session->hists.stats.nr_invalid_chains;
1012 session->hists.stats.total_invalid_chains += sample->period;
1013 return -EINVAL;
1014 }
1015 return 0;
1016}
1017
1018static int perf_session__process_user_event(struct perf_session *session, union perf_event *event,
1019 struct perf_tool *tool, u64 file_offset)
1020{
1021 int err;
1022
1023 dump_event(session, event, file_offset, NULL);
1024
1025 /* These events are processed right away */
1026 switch (event->header.type) {
1027 case PERF_RECORD_HEADER_ATTR:
1028 err = tool->attr(event, &session->evlist);
1029 if (err == 0)
1030 perf_session__update_sample_type(session);
1031 return err;
1032 case PERF_RECORD_HEADER_EVENT_TYPE:
1033 return tool->event_type(tool, event);
1034 case PERF_RECORD_HEADER_TRACING_DATA:
1035 /* setup for reading amidst mmap */
1036 lseek(session->fd, file_offset, SEEK_SET);
1037 return tool->tracing_data(event, session);
1038 case PERF_RECORD_HEADER_BUILD_ID:
1039 return tool->build_id(tool, event, session);
1040 case PERF_RECORD_FINISHED_ROUND:
1041 return tool->finished_round(tool, event, session);
1042 default:
1043 return -EINVAL;
1044 }
1045}
1046
1047static void event_swap(union perf_event *event, bool sample_id_all)
1048{
1049 perf_event__swap_op swap;
1050
1051 swap = perf_event__swap_ops[event->header.type];
1052 if (swap)
1053 swap(event, sample_id_all);
1054}
1055
1056static int perf_session__process_event(struct perf_session *session,
1057 union perf_event *event,
1058 struct perf_tool *tool,
1059 u64 file_offset)
1060{
1061 struct perf_sample sample;
1062 int ret;
1063
1064 if (session->header.needs_swap)
1065 event_swap(event, session->sample_id_all);
1066
1067 if (event->header.type >= PERF_RECORD_HEADER_MAX)
1068 return -EINVAL;
1069
1070 hists__inc_nr_events(&session->hists, event->header.type);
1071
1072 if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1073 return perf_session__process_user_event(session, event, tool, file_offset);
1074
1075 /*
1076 * For all kernel events we get the sample data
1077 */
1078 ret = perf_session__parse_sample(session, event, &sample);
1079 if (ret)
1080 return ret;
1081
1082 /* Preprocess sample records - precheck callchains */
1083 if (perf_session__preprocess_sample(session, event, &sample))
1084 return 0;
1085
1086 if (tool->ordered_samples) {
1087 ret = perf_session_queue_event(session, event, &sample,
1088 file_offset);
1089 if (ret != -ETIME)
1090 return ret;
1091 }
1092
1093 return perf_session_deliver_event(session, event, &sample, tool,
1094 file_offset);
1095}
1096
1097void perf_event_header__bswap(struct perf_event_header *self)
1098{
1099 self->type = bswap_32(self->type);
1100 self->misc = bswap_16(self->misc);
1101 self->size = bswap_16(self->size);
1102}
1103
1104struct thread *perf_session__findnew(struct perf_session *session, pid_t pid)
1105{
1106 return machine__findnew_thread(&session->host_machine, pid);
1107}
1108
1109static struct thread *perf_session__register_idle_thread(struct perf_session *self)
1110{
1111 struct thread *thread = perf_session__findnew(self, 0);
1112
1113 if (thread == NULL || thread__set_comm(thread, "swapper")) {
1114 pr_err("problem inserting idle task.\n");
1115 thread = NULL;
1116 }
1117
1118 return thread;
1119}
1120
1121static void perf_session__warn_about_errors(const struct perf_session *session,
1122 const struct perf_tool *tool)
1123{
1124 if (tool->lost == perf_event__process_lost &&
1125 session->hists.stats.nr_events[PERF_RECORD_LOST] != 0) {
1126 ui__warning("Processed %d events and lost %d chunks!\n\n"
1127 "Check IO/CPU overload!\n\n",
1128 session->hists.stats.nr_events[0],
1129 session->hists.stats.nr_events[PERF_RECORD_LOST]);
1130 }
1131
1132 if (session->hists.stats.nr_unknown_events != 0) {
1133 ui__warning("Found %u unknown events!\n\n"
1134 "Is this an older tool processing a perf.data "
1135 "file generated by a more recent tool?\n\n"
1136 "If that is not the case, consider "
1137 "reporting to linux-kernel@vger.kernel.org.\n\n",
1138 session->hists.stats.nr_unknown_events);
1139 }
1140
1141 if (session->hists.stats.nr_unknown_id != 0) {
1142 ui__warning("%u samples with id not present in the header\n",
1143 session->hists.stats.nr_unknown_id);
1144 }
1145
1146 if (session->hists.stats.nr_invalid_chains != 0) {
1147 ui__warning("Found invalid callchains!\n\n"
1148 "%u out of %u events were discarded for this reason.\n\n"
1149 "Consider reporting to linux-kernel@vger.kernel.org.\n\n",
1150 session->hists.stats.nr_invalid_chains,
1151 session->hists.stats.nr_events[PERF_RECORD_SAMPLE]);
1152 }
1153
1154 if (session->hists.stats.nr_unprocessable_samples != 0) {
1155 ui__warning("%u unprocessable samples recorded.\n"
1156 "Do you have a KVM guest running and not using 'perf kvm'?\n",
1157 session->hists.stats.nr_unprocessable_samples);
1158 }
1159}
1160
1161#define session_done() (*(volatile int *)(&session_done))
1162volatile int session_done;
1163
1164static int __perf_session__process_pipe_events(struct perf_session *self,
1165 struct perf_tool *tool)
1166{
1167 union perf_event *event;
1168 uint32_t size, cur_size = 0;
1169 void *buf = NULL;
1170 int skip = 0;
1171 u64 head;
1172 int err;
1173 void *p;
1174
1175 perf_tool__fill_defaults(tool);
1176
1177 head = 0;
1178 cur_size = sizeof(union perf_event);
1179
1180 buf = malloc(cur_size);
1181 if (!buf)
1182 return -errno;
1183more:
1184 event = buf;
1185 err = readn(self->fd, event, sizeof(struct perf_event_header));
1186 if (err <= 0) {
1187 if (err == 0)
1188 goto done;
1189
1190 pr_err("failed to read event header\n");
1191 goto out_err;
1192 }
1193
1194 if (self->header.needs_swap)
1195 perf_event_header__bswap(&event->header);
1196
1197 size = event->header.size;
1198 if (size == 0)
1199 size = 8;
1200
1201 if (size > cur_size) {
1202 void *new = realloc(buf, size);
1203 if (!new) {
1204 pr_err("failed to allocate memory to read event\n");
1205 goto out_err;
1206 }
1207 buf = new;
1208 cur_size = size;
1209 event = buf;
1210 }
1211 p = event;
1212 p += sizeof(struct perf_event_header);
1213
1214 if (size - sizeof(struct perf_event_header)) {
1215 err = readn(self->fd, p, size - sizeof(struct perf_event_header));
1216 if (err <= 0) {
1217 if (err == 0) {
1218 pr_err("unexpected end of event stream\n");
1219 goto done;
1220 }
1221
1222 pr_err("failed to read event data\n");
1223 goto out_err;
1224 }
1225 }
1226
1227 if ((skip = perf_session__process_event(self, event, tool, head)) < 0) {
1228 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1229 head, event->header.size, event->header.type);
1230 err = -EINVAL;
1231 goto out_err;
1232 }
1233
1234 head += size;
1235
1236 if (skip > 0)
1237 head += skip;
1238
1239 if (!session_done())
1240 goto more;
1241done:
1242 err = 0;
1243out_err:
1244 free(buf);
1245 perf_session__warn_about_errors(self, tool);
1246 perf_session_free_sample_buffers(self);
1247 return err;
1248}
1249
1250static union perf_event *
1251fetch_mmaped_event(struct perf_session *session,
1252 u64 head, size_t mmap_size, char *buf)
1253{
1254 union perf_event *event;
1255
1256 /*
1257 * Ensure we have enough space remaining to read
1258 * the size of the event in the headers.
1259 */
1260 if (head + sizeof(event->header) > mmap_size)
1261 return NULL;
1262
1263 event = (union perf_event *)(buf + head);
1264
1265 if (session->header.needs_swap)
1266 perf_event_header__bswap(&event->header);
1267
1268 if (head + event->header.size > mmap_size)
1269 return NULL;
1270
1271 return event;
1272}
1273
1274int __perf_session__process_events(struct perf_session *session,
1275 u64 data_offset, u64 data_size,
1276 u64 file_size, struct perf_tool *tool)
1277{
1278 u64 head, page_offset, file_offset, file_pos, progress_next;
1279 int err, mmap_prot, mmap_flags, map_idx = 0;
1280 size_t page_size, mmap_size;
1281 char *buf, *mmaps[8];
1282 union perf_event *event;
1283 uint32_t size;
1284
1285 perf_tool__fill_defaults(tool);
1286
1287 page_size = sysconf(_SC_PAGESIZE);
1288
1289 page_offset = page_size * (data_offset / page_size);
1290 file_offset = page_offset;
1291 head = data_offset - page_offset;
1292
1293 if (data_offset + data_size < file_size)
1294 file_size = data_offset + data_size;
1295
1296 progress_next = file_size / 16;
1297
1298 mmap_size = session->mmap_window;
1299 if (mmap_size > file_size)
1300 mmap_size = file_size;
1301
1302 memset(mmaps, 0, sizeof(mmaps));
1303
1304 mmap_prot = PROT_READ;
1305 mmap_flags = MAP_SHARED;
1306
1307 if (session->header.needs_swap) {
1308 mmap_prot |= PROT_WRITE;
1309 mmap_flags = MAP_PRIVATE;
1310 }
1311remap:
1312 buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, session->fd,
1313 file_offset);
1314 if (buf == MAP_FAILED) {
1315 pr_err("failed to mmap file\n");
1316 err = -errno;
1317 goto out_err;
1318 }
1319 mmaps[map_idx] = buf;
1320 map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1);
1321 file_pos = file_offset + head;
1322
1323more:
1324 event = fetch_mmaped_event(session, head, mmap_size, buf);
1325 if (!event) {
1326 if (mmaps[map_idx]) {
1327 munmap(mmaps[map_idx], mmap_size);
1328 mmaps[map_idx] = NULL;
1329 }
1330
1331 page_offset = page_size * (head / page_size);
1332 file_offset += page_offset;
1333 head -= page_offset;
1334 goto remap;
1335 }
1336
1337 size = event->header.size;
1338
1339 if (size == 0 ||
1340 perf_session__process_event(session, event, tool, file_pos) < 0) {
1341 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1342 file_offset + head, event->header.size,
1343 event->header.type);
1344 err = -EINVAL;
1345 goto out_err;
1346 }
1347
1348 head += size;
1349 file_pos += size;
1350
1351 if (file_pos >= progress_next) {
1352 progress_next += file_size / 16;
1353 ui_progress__update(file_pos, file_size,
1354 "Processing events...");
1355 }
1356
1357 if (file_pos < file_size)
1358 goto more;
1359
1360 err = 0;
1361 /* do the final flush for ordered samples */
1362 session->ordered_samples.next_flush = ULLONG_MAX;
1363 flush_sample_queue(session, tool);
1364out_err:
1365 perf_session__warn_about_errors(session, tool);
1366 perf_session_free_sample_buffers(session);
1367 return err;
1368}
1369
1370int perf_session__process_events(struct perf_session *self,
1371 struct perf_tool *tool)
1372{
1373 int err;
1374
1375 if (perf_session__register_idle_thread(self) == NULL)
1376 return -ENOMEM;
1377
1378 if (!self->fd_pipe)
1379 err = __perf_session__process_events(self,
1380 self->header.data_offset,
1381 self->header.data_size,
1382 self->size, tool);
1383 else
1384 err = __perf_session__process_pipe_events(self, tool);
1385
1386 return err;
1387}
1388
1389bool perf_session__has_traces(struct perf_session *self, const char *msg)
1390{
1391 if (!(self->sample_type & PERF_SAMPLE_RAW)) {
1392 pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg);
1393 return false;
1394 }
1395
1396 return true;
1397}
1398
1399int maps__set_kallsyms_ref_reloc_sym(struct map **maps,
1400 const char *symbol_name, u64 addr)
1401{
1402 char *bracket;
1403 enum map_type i;
1404 struct ref_reloc_sym *ref;
1405
1406 ref = zalloc(sizeof(struct ref_reloc_sym));
1407 if (ref == NULL)
1408 return -ENOMEM;
1409
1410 ref->name = strdup(symbol_name);
1411 if (ref->name == NULL) {
1412 free(ref);
1413 return -ENOMEM;
1414 }
1415
1416 bracket = strchr(ref->name, ']');
1417 if (bracket)
1418 *bracket = '\0';
1419
1420 ref->addr = addr;
1421
1422 for (i = 0; i < MAP__NR_TYPES; ++i) {
1423 struct kmap *kmap = map__kmap(maps[i]);
1424 kmap->ref_reloc_sym = ref;
1425 }
1426
1427 return 0;
1428}
1429
1430size_t perf_session__fprintf_dsos(struct perf_session *self, FILE *fp)
1431{
1432 return __dsos__fprintf(&self->host_machine.kernel_dsos, fp) +
1433 __dsos__fprintf(&self->host_machine.user_dsos, fp) +
1434 machines__fprintf_dsos(&self->machines, fp);
1435}
1436
1437size_t perf_session__fprintf_dsos_buildid(struct perf_session *self, FILE *fp,
1438 bool with_hits)
1439{
1440 size_t ret = machine__fprintf_dsos_buildid(&self->host_machine, fp, with_hits);
1441 return ret + machines__fprintf_dsos_buildid(&self->machines, fp, with_hits);
1442}
1443
1444size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp)
1445{
1446 struct perf_evsel *pos;
1447 size_t ret = fprintf(fp, "Aggregated stats:\n");
1448
1449 ret += hists__fprintf_nr_events(&session->hists, fp);
1450
1451 list_for_each_entry(pos, &session->evlist->entries, node) {
1452 ret += fprintf(fp, "%s stats:\n", event_name(pos));
1453 ret += hists__fprintf_nr_events(&pos->hists, fp);
1454 }
1455
1456 return ret;
1457}
1458
1459size_t perf_session__fprintf(struct perf_session *session, FILE *fp)
1460{
1461 /*
1462 * FIXME: Here we have to actually print all the machines in this
1463 * session, not just the host...
1464 */
1465 return machine__fprintf(&session->host_machine, fp);
1466}
1467
1468void perf_session__remove_thread(struct perf_session *session,
1469 struct thread *th)
1470{
1471 /*
1472 * FIXME: This one makes no sense, we need to remove the thread from
1473 * the machine it belongs to, perf_session can have many machines, so
1474 * doing it always on ->host_machine is wrong. Fix when auditing all
1475 * the 'perf kvm' code.
1476 */
1477 machine__remove_thread(&session->host_machine, th);
1478}
1479
1480struct perf_evsel *perf_session__find_first_evtype(struct perf_session *session,
1481 unsigned int type)
1482{
1483 struct perf_evsel *pos;
1484
1485 list_for_each_entry(pos, &session->evlist->entries, node) {
1486 if (pos->attr.type == type)
1487 return pos;
1488 }
1489 return NULL;
1490}
1491
1492void perf_event__print_ip(union perf_event *event, struct perf_sample *sample,
1493 struct machine *machine, struct perf_evsel *evsel,
1494 int print_sym, int print_dso, int print_symoffset)
1495{
1496 struct addr_location al;
1497 struct callchain_cursor_node *node;
1498
1499 if (perf_event__preprocess_sample(event, machine, &al, sample,
1500 NULL) < 0) {
1501 error("problem processing %d event, skipping it.\n",
1502 event->header.type);
1503 return;
1504 }
1505
1506 if (symbol_conf.use_callchain && sample->callchain) {
1507
1508 if (machine__resolve_callchain(machine, evsel, al.thread,
1509 sample->callchain, NULL) != 0) {
1510 if (verbose)
1511 error("Failed to resolve callchain. Skipping\n");
1512 return;
1513 }
1514 callchain_cursor_commit(&callchain_cursor);
1515
1516 while (1) {
1517 node = callchain_cursor_current(&callchain_cursor);
1518 if (!node)
1519 break;
1520
1521 printf("\t%16" PRIx64, node->ip);
1522 if (print_sym) {
1523 printf(" ");
1524 symbol__fprintf_symname(node->sym, stdout);
1525 }
1526 if (print_dso) {
1527 printf(" (");
1528 map__fprintf_dsoname(node->map, stdout);
1529 printf(")");
1530 }
1531 printf("\n");
1532
1533 callchain_cursor_advance(&callchain_cursor);
1534 }
1535
1536 } else {
1537 printf("%16" PRIx64, sample->ip);
1538 if (print_sym) {
1539 printf(" ");
1540 if (print_symoffset)
1541 symbol__fprintf_symname_offs(al.sym, &al,
1542 stdout);
1543 else
1544 symbol__fprintf_symname(al.sym, stdout);
1545 }
1546
1547 if (print_dso) {
1548 printf(" (");
1549 map__fprintf_dsoname(al.map, stdout);
1550 printf(")");
1551 }
1552 }
1553}
1554
1555int perf_session__cpu_bitmap(struct perf_session *session,
1556 const char *cpu_list, unsigned long *cpu_bitmap)
1557{
1558 int i;
1559 struct cpu_map *map;
1560
1561 for (i = 0; i < PERF_TYPE_MAX; ++i) {
1562 struct perf_evsel *evsel;
1563
1564 evsel = perf_session__find_first_evtype(session, i);
1565 if (!evsel)
1566 continue;
1567
1568 if (!(evsel->attr.sample_type & PERF_SAMPLE_CPU)) {
1569 pr_err("File does not contain CPU events. "
1570 "Remove -c option to proceed.\n");
1571 return -1;
1572 }
1573 }
1574
1575 map = cpu_map__new(cpu_list);
1576 if (map == NULL) {
1577 pr_err("Invalid cpu_list\n");
1578 return -1;
1579 }
1580
1581 for (i = 0; i < map->nr; i++) {
1582 int cpu = map->map[i];
1583
1584 if (cpu >= MAX_NR_CPUS) {
1585 pr_err("Requested CPU %d too large. "
1586 "Consider raising MAX_NR_CPUS\n", cpu);
1587 return -1;
1588 }
1589
1590 set_bit(cpu, cpu_bitmap);
1591 }
1592
1593 return 0;
1594}
1595
1596void perf_session__fprintf_info(struct perf_session *session, FILE *fp,
1597 bool full)
1598{
1599 struct stat st;
1600 int ret;
1601
1602 if (session == NULL || fp == NULL)
1603 return;
1604
1605 ret = fstat(session->fd, &st);
1606 if (ret == -1)
1607 return;
1608
1609 fprintf(fp, "# ========\n");
1610 fprintf(fp, "# captured on: %s", ctime(&st.st_ctime));
1611 perf_header__fprintf_info(session, fp, full);
1612 fprintf(fp, "# ========\n#\n");
1613}
1// SPDX-License-Identifier: GPL-2.0
2#include <errno.h>
3#include <inttypes.h>
4#include <linux/err.h>
5#include <linux/kernel.h>
6#include <linux/zalloc.h>
7#include <api/fs/fs.h>
8
9#include <byteswap.h>
10#include <unistd.h>
11#include <sys/types.h>
12#include <sys/mman.h>
13#include <perf/cpumap.h>
14
15#include "map_symbol.h"
16#include "branch.h"
17#include "debug.h"
18#include "evlist.h"
19#include "evsel.h"
20#include "memswap.h"
21#include "map.h"
22#include "symbol.h"
23#include "session.h"
24#include "tool.h"
25#include "perf_regs.h"
26#include "asm/bug.h"
27#include "auxtrace.h"
28#include "thread.h"
29#include "thread-stack.h"
30#include "sample-raw.h"
31#include "stat.h"
32#include "tsc.h"
33#include "ui/progress.h"
34#include "../perf.h"
35#include "arch/common.h"
36#include "units.h"
37#include <internal/lib.h>
38
39#ifdef HAVE_ZSTD_SUPPORT
40static int perf_session__process_compressed_event(struct perf_session *session,
41 union perf_event *event, u64 file_offset)
42{
43 void *src;
44 size_t decomp_size, src_size;
45 u64 decomp_last_rem = 0;
46 size_t mmap_len, decomp_len = session->header.env.comp_mmap_len;
47 struct decomp *decomp, *decomp_last = session->decomp_last;
48
49 if (decomp_last) {
50 decomp_last_rem = decomp_last->size - decomp_last->head;
51 decomp_len += decomp_last_rem;
52 }
53
54 mmap_len = sizeof(struct decomp) + decomp_len;
55 decomp = mmap(NULL, mmap_len, PROT_READ|PROT_WRITE,
56 MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);
57 if (decomp == MAP_FAILED) {
58 pr_err("Couldn't allocate memory for decompression\n");
59 return -1;
60 }
61
62 decomp->file_pos = file_offset;
63 decomp->mmap_len = mmap_len;
64 decomp->head = 0;
65
66 if (decomp_last_rem) {
67 memcpy(decomp->data, &(decomp_last->data[decomp_last->head]), decomp_last_rem);
68 decomp->size = decomp_last_rem;
69 }
70
71 src = (void *)event + sizeof(struct perf_record_compressed);
72 src_size = event->pack.header.size - sizeof(struct perf_record_compressed);
73
74 decomp_size = zstd_decompress_stream(&(session->zstd_data), src, src_size,
75 &(decomp->data[decomp_last_rem]), decomp_len - decomp_last_rem);
76 if (!decomp_size) {
77 munmap(decomp, mmap_len);
78 pr_err("Couldn't decompress data\n");
79 return -1;
80 }
81
82 decomp->size += decomp_size;
83
84 if (session->decomp == NULL) {
85 session->decomp = decomp;
86 session->decomp_last = decomp;
87 } else {
88 session->decomp_last->next = decomp;
89 session->decomp_last = decomp;
90 }
91
92 pr_debug("decomp (B): %zd to %zd\n", src_size, decomp_size);
93
94 return 0;
95}
96#else /* !HAVE_ZSTD_SUPPORT */
97#define perf_session__process_compressed_event perf_session__process_compressed_event_stub
98#endif
99
100static int perf_session__deliver_event(struct perf_session *session,
101 union perf_event *event,
102 struct perf_tool *tool,
103 u64 file_offset);
104
105static int perf_session__open(struct perf_session *session)
106{
107 struct perf_data *data = session->data;
108
109 if (perf_session__read_header(session) < 0) {
110 pr_err("incompatible file format (rerun with -v to learn more)\n");
111 return -1;
112 }
113
114 if (perf_data__is_pipe(data))
115 return 0;
116
117 if (perf_header__has_feat(&session->header, HEADER_STAT))
118 return 0;
119
120 if (!evlist__valid_sample_type(session->evlist)) {
121 pr_err("non matching sample_type\n");
122 return -1;
123 }
124
125 if (!evlist__valid_sample_id_all(session->evlist)) {
126 pr_err("non matching sample_id_all\n");
127 return -1;
128 }
129
130 if (!evlist__valid_read_format(session->evlist)) {
131 pr_err("non matching read_format\n");
132 return -1;
133 }
134
135 return 0;
136}
137
138void perf_session__set_id_hdr_size(struct perf_session *session)
139{
140 u16 id_hdr_size = evlist__id_hdr_size(session->evlist);
141
142 machines__set_id_hdr_size(&session->machines, id_hdr_size);
143}
144
145int perf_session__create_kernel_maps(struct perf_session *session)
146{
147 int ret = machine__create_kernel_maps(&session->machines.host);
148
149 if (ret >= 0)
150 ret = machines__create_guest_kernel_maps(&session->machines);
151 return ret;
152}
153
154static void perf_session__destroy_kernel_maps(struct perf_session *session)
155{
156 machines__destroy_kernel_maps(&session->machines);
157}
158
159static bool perf_session__has_comm_exec(struct perf_session *session)
160{
161 struct evsel *evsel;
162
163 evlist__for_each_entry(session->evlist, evsel) {
164 if (evsel->core.attr.comm_exec)
165 return true;
166 }
167
168 return false;
169}
170
171static void perf_session__set_comm_exec(struct perf_session *session)
172{
173 bool comm_exec = perf_session__has_comm_exec(session);
174
175 machines__set_comm_exec(&session->machines, comm_exec);
176}
177
178static int ordered_events__deliver_event(struct ordered_events *oe,
179 struct ordered_event *event)
180{
181 struct perf_session *session = container_of(oe, struct perf_session,
182 ordered_events);
183
184 return perf_session__deliver_event(session, event->event,
185 session->tool, event->file_offset);
186}
187
188struct perf_session *perf_session__new(struct perf_data *data,
189 bool repipe, struct perf_tool *tool)
190{
191 int ret = -ENOMEM;
192 struct perf_session *session = zalloc(sizeof(*session));
193
194 if (!session)
195 goto out;
196
197 session->repipe = repipe;
198 session->tool = tool;
199 INIT_LIST_HEAD(&session->auxtrace_index);
200 machines__init(&session->machines);
201 ordered_events__init(&session->ordered_events,
202 ordered_events__deliver_event, NULL);
203
204 perf_env__init(&session->header.env);
205 if (data) {
206 ret = perf_data__open(data);
207 if (ret < 0)
208 goto out_delete;
209
210 session->data = data;
211
212 if (perf_data__is_read(data)) {
213 ret = perf_session__open(session);
214 if (ret < 0)
215 goto out_delete;
216
217 /*
218 * set session attributes that are present in perf.data
219 * but not in pipe-mode.
220 */
221 if (!data->is_pipe) {
222 perf_session__set_id_hdr_size(session);
223 perf_session__set_comm_exec(session);
224 }
225
226 evlist__init_trace_event_sample_raw(session->evlist);
227
228 /* Open the directory data. */
229 if (data->is_dir) {
230 ret = perf_data__open_dir(data);
231 if (ret)
232 goto out_delete;
233 }
234
235 if (!symbol_conf.kallsyms_name &&
236 !symbol_conf.vmlinux_name)
237 symbol_conf.kallsyms_name = perf_data__kallsyms_name(data);
238 }
239 } else {
240 session->machines.host.env = &perf_env;
241 }
242
243 session->machines.host.single_address_space =
244 perf_env__single_address_space(session->machines.host.env);
245
246 if (!data || perf_data__is_write(data)) {
247 /*
248 * In O_RDONLY mode this will be performed when reading the
249 * kernel MMAP event, in perf_event__process_mmap().
250 */
251 if (perf_session__create_kernel_maps(session) < 0)
252 pr_warning("Cannot read kernel map\n");
253 }
254
255 /*
256 * In pipe-mode, evlist is empty until PERF_RECORD_HEADER_ATTR is
257 * processed, so evlist__sample_id_all is not meaningful here.
258 */
259 if ((!data || !data->is_pipe) && tool && tool->ordering_requires_timestamps &&
260 tool->ordered_events && !evlist__sample_id_all(session->evlist)) {
261 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n");
262 tool->ordered_events = false;
263 }
264
265 return session;
266
267 out_delete:
268 perf_session__delete(session);
269 out:
270 return ERR_PTR(ret);
271}
272
273static void perf_session__delete_threads(struct perf_session *session)
274{
275 machine__delete_threads(&session->machines.host);
276}
277
278static void perf_session__release_decomp_events(struct perf_session *session)
279{
280 struct decomp *next, *decomp;
281 size_t mmap_len;
282 next = session->decomp;
283 do {
284 decomp = next;
285 if (decomp == NULL)
286 break;
287 next = decomp->next;
288 mmap_len = decomp->mmap_len;
289 munmap(decomp, mmap_len);
290 } while (1);
291}
292
293void perf_session__delete(struct perf_session *session)
294{
295 if (session == NULL)
296 return;
297 auxtrace__free(session);
298 auxtrace_index__free(&session->auxtrace_index);
299 perf_session__destroy_kernel_maps(session);
300 perf_session__delete_threads(session);
301 perf_session__release_decomp_events(session);
302 perf_env__exit(&session->header.env);
303 machines__exit(&session->machines);
304 if (session->data) {
305 if (perf_data__is_read(session->data))
306 evlist__delete(session->evlist);
307 perf_data__close(session->data);
308 }
309 trace_event__cleanup(&session->tevent);
310 free(session);
311}
312
313static int process_event_synth_tracing_data_stub(struct perf_session *session
314 __maybe_unused,
315 union perf_event *event
316 __maybe_unused)
317{
318 dump_printf(": unhandled!\n");
319 return 0;
320}
321
322static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused,
323 union perf_event *event __maybe_unused,
324 struct evlist **pevlist
325 __maybe_unused)
326{
327 dump_printf(": unhandled!\n");
328 return 0;
329}
330
331static int process_event_synth_event_update_stub(struct perf_tool *tool __maybe_unused,
332 union perf_event *event __maybe_unused,
333 struct evlist **pevlist
334 __maybe_unused)
335{
336 if (dump_trace)
337 perf_event__fprintf_event_update(event, stdout);
338
339 dump_printf(": unhandled!\n");
340 return 0;
341}
342
343static int process_event_sample_stub(struct perf_tool *tool __maybe_unused,
344 union perf_event *event __maybe_unused,
345 struct perf_sample *sample __maybe_unused,
346 struct evsel *evsel __maybe_unused,
347 struct machine *machine __maybe_unused)
348{
349 dump_printf(": unhandled!\n");
350 return 0;
351}
352
353static int process_event_stub(struct perf_tool *tool __maybe_unused,
354 union perf_event *event __maybe_unused,
355 struct perf_sample *sample __maybe_unused,
356 struct machine *machine __maybe_unused)
357{
358 dump_printf(": unhandled!\n");
359 return 0;
360}
361
362static int process_finished_round_stub(struct perf_tool *tool __maybe_unused,
363 union perf_event *event __maybe_unused,
364 struct ordered_events *oe __maybe_unused)
365{
366 dump_printf(": unhandled!\n");
367 return 0;
368}
369
370static int process_finished_round(struct perf_tool *tool,
371 union perf_event *event,
372 struct ordered_events *oe);
373
374static int skipn(int fd, off_t n)
375{
376 char buf[4096];
377 ssize_t ret;
378
379 while (n > 0) {
380 ret = read(fd, buf, min(n, (off_t)sizeof(buf)));
381 if (ret <= 0)
382 return ret;
383 n -= ret;
384 }
385
386 return 0;
387}
388
389static s64 process_event_auxtrace_stub(struct perf_session *session __maybe_unused,
390 union perf_event *event)
391{
392 dump_printf(": unhandled!\n");
393 if (perf_data__is_pipe(session->data))
394 skipn(perf_data__fd(session->data), event->auxtrace.size);
395 return event->auxtrace.size;
396}
397
398static int process_event_op2_stub(struct perf_session *session __maybe_unused,
399 union perf_event *event __maybe_unused)
400{
401 dump_printf(": unhandled!\n");
402 return 0;
403}
404
405
406static
407int process_event_thread_map_stub(struct perf_session *session __maybe_unused,
408 union perf_event *event __maybe_unused)
409{
410 if (dump_trace)
411 perf_event__fprintf_thread_map(event, stdout);
412
413 dump_printf(": unhandled!\n");
414 return 0;
415}
416
417static
418int process_event_cpu_map_stub(struct perf_session *session __maybe_unused,
419 union perf_event *event __maybe_unused)
420{
421 if (dump_trace)
422 perf_event__fprintf_cpu_map(event, stdout);
423
424 dump_printf(": unhandled!\n");
425 return 0;
426}
427
428static
429int process_event_stat_config_stub(struct perf_session *session __maybe_unused,
430 union perf_event *event __maybe_unused)
431{
432 if (dump_trace)
433 perf_event__fprintf_stat_config(event, stdout);
434
435 dump_printf(": unhandled!\n");
436 return 0;
437}
438
439static int process_stat_stub(struct perf_session *perf_session __maybe_unused,
440 union perf_event *event)
441{
442 if (dump_trace)
443 perf_event__fprintf_stat(event, stdout);
444
445 dump_printf(": unhandled!\n");
446 return 0;
447}
448
449static int process_stat_round_stub(struct perf_session *perf_session __maybe_unused,
450 union perf_event *event)
451{
452 if (dump_trace)
453 perf_event__fprintf_stat_round(event, stdout);
454
455 dump_printf(": unhandled!\n");
456 return 0;
457}
458
459static int process_event_time_conv_stub(struct perf_session *perf_session __maybe_unused,
460 union perf_event *event)
461{
462 if (dump_trace)
463 perf_event__fprintf_time_conv(event, stdout);
464
465 dump_printf(": unhandled!\n");
466 return 0;
467}
468
469static int perf_session__process_compressed_event_stub(struct perf_session *session __maybe_unused,
470 union perf_event *event __maybe_unused,
471 u64 file_offset __maybe_unused)
472{
473 dump_printf(": unhandled!\n");
474 return 0;
475}
476
477void perf_tool__fill_defaults(struct perf_tool *tool)
478{
479 if (tool->sample == NULL)
480 tool->sample = process_event_sample_stub;
481 if (tool->mmap == NULL)
482 tool->mmap = process_event_stub;
483 if (tool->mmap2 == NULL)
484 tool->mmap2 = process_event_stub;
485 if (tool->comm == NULL)
486 tool->comm = process_event_stub;
487 if (tool->namespaces == NULL)
488 tool->namespaces = process_event_stub;
489 if (tool->cgroup == NULL)
490 tool->cgroup = process_event_stub;
491 if (tool->fork == NULL)
492 tool->fork = process_event_stub;
493 if (tool->exit == NULL)
494 tool->exit = process_event_stub;
495 if (tool->lost == NULL)
496 tool->lost = perf_event__process_lost;
497 if (tool->lost_samples == NULL)
498 tool->lost_samples = perf_event__process_lost_samples;
499 if (tool->aux == NULL)
500 tool->aux = perf_event__process_aux;
501 if (tool->itrace_start == NULL)
502 tool->itrace_start = perf_event__process_itrace_start;
503 if (tool->context_switch == NULL)
504 tool->context_switch = perf_event__process_switch;
505 if (tool->ksymbol == NULL)
506 tool->ksymbol = perf_event__process_ksymbol;
507 if (tool->bpf == NULL)
508 tool->bpf = perf_event__process_bpf;
509 if (tool->text_poke == NULL)
510 tool->text_poke = perf_event__process_text_poke;
511 if (tool->read == NULL)
512 tool->read = process_event_sample_stub;
513 if (tool->throttle == NULL)
514 tool->throttle = process_event_stub;
515 if (tool->unthrottle == NULL)
516 tool->unthrottle = process_event_stub;
517 if (tool->attr == NULL)
518 tool->attr = process_event_synth_attr_stub;
519 if (tool->event_update == NULL)
520 tool->event_update = process_event_synth_event_update_stub;
521 if (tool->tracing_data == NULL)
522 tool->tracing_data = process_event_synth_tracing_data_stub;
523 if (tool->build_id == NULL)
524 tool->build_id = process_event_op2_stub;
525 if (tool->finished_round == NULL) {
526 if (tool->ordered_events)
527 tool->finished_round = process_finished_round;
528 else
529 tool->finished_round = process_finished_round_stub;
530 }
531 if (tool->id_index == NULL)
532 tool->id_index = process_event_op2_stub;
533 if (tool->auxtrace_info == NULL)
534 tool->auxtrace_info = process_event_op2_stub;
535 if (tool->auxtrace == NULL)
536 tool->auxtrace = process_event_auxtrace_stub;
537 if (tool->auxtrace_error == NULL)
538 tool->auxtrace_error = process_event_op2_stub;
539 if (tool->thread_map == NULL)
540 tool->thread_map = process_event_thread_map_stub;
541 if (tool->cpu_map == NULL)
542 tool->cpu_map = process_event_cpu_map_stub;
543 if (tool->stat_config == NULL)
544 tool->stat_config = process_event_stat_config_stub;
545 if (tool->stat == NULL)
546 tool->stat = process_stat_stub;
547 if (tool->stat_round == NULL)
548 tool->stat_round = process_stat_round_stub;
549 if (tool->time_conv == NULL)
550 tool->time_conv = process_event_time_conv_stub;
551 if (tool->feature == NULL)
552 tool->feature = process_event_op2_stub;
553 if (tool->compressed == NULL)
554 tool->compressed = perf_session__process_compressed_event;
555}
556
557static void swap_sample_id_all(union perf_event *event, void *data)
558{
559 void *end = (void *) event + event->header.size;
560 int size = end - data;
561
562 BUG_ON(size % sizeof(u64));
563 mem_bswap_64(data, size);
564}
565
566static void perf_event__all64_swap(union perf_event *event,
567 bool sample_id_all __maybe_unused)
568{
569 struct perf_event_header *hdr = &event->header;
570 mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr));
571}
572
573static void perf_event__comm_swap(union perf_event *event, bool sample_id_all)
574{
575 event->comm.pid = bswap_32(event->comm.pid);
576 event->comm.tid = bswap_32(event->comm.tid);
577
578 if (sample_id_all) {
579 void *data = &event->comm.comm;
580
581 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
582 swap_sample_id_all(event, data);
583 }
584}
585
586static void perf_event__mmap_swap(union perf_event *event,
587 bool sample_id_all)
588{
589 event->mmap.pid = bswap_32(event->mmap.pid);
590 event->mmap.tid = bswap_32(event->mmap.tid);
591 event->mmap.start = bswap_64(event->mmap.start);
592 event->mmap.len = bswap_64(event->mmap.len);
593 event->mmap.pgoff = bswap_64(event->mmap.pgoff);
594
595 if (sample_id_all) {
596 void *data = &event->mmap.filename;
597
598 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
599 swap_sample_id_all(event, data);
600 }
601}
602
603static void perf_event__mmap2_swap(union perf_event *event,
604 bool sample_id_all)
605{
606 event->mmap2.pid = bswap_32(event->mmap2.pid);
607 event->mmap2.tid = bswap_32(event->mmap2.tid);
608 event->mmap2.start = bswap_64(event->mmap2.start);
609 event->mmap2.len = bswap_64(event->mmap2.len);
610 event->mmap2.pgoff = bswap_64(event->mmap2.pgoff);
611
612 if (!(event->header.misc & PERF_RECORD_MISC_MMAP_BUILD_ID)) {
613 event->mmap2.maj = bswap_32(event->mmap2.maj);
614 event->mmap2.min = bswap_32(event->mmap2.min);
615 event->mmap2.ino = bswap_64(event->mmap2.ino);
616 event->mmap2.ino_generation = bswap_64(event->mmap2.ino_generation);
617 }
618
619 if (sample_id_all) {
620 void *data = &event->mmap2.filename;
621
622 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
623 swap_sample_id_all(event, data);
624 }
625}
626static void perf_event__task_swap(union perf_event *event, bool sample_id_all)
627{
628 event->fork.pid = bswap_32(event->fork.pid);
629 event->fork.tid = bswap_32(event->fork.tid);
630 event->fork.ppid = bswap_32(event->fork.ppid);
631 event->fork.ptid = bswap_32(event->fork.ptid);
632 event->fork.time = bswap_64(event->fork.time);
633
634 if (sample_id_all)
635 swap_sample_id_all(event, &event->fork + 1);
636}
637
638static void perf_event__read_swap(union perf_event *event, bool sample_id_all)
639{
640 event->read.pid = bswap_32(event->read.pid);
641 event->read.tid = bswap_32(event->read.tid);
642 event->read.value = bswap_64(event->read.value);
643 event->read.time_enabled = bswap_64(event->read.time_enabled);
644 event->read.time_running = bswap_64(event->read.time_running);
645 event->read.id = bswap_64(event->read.id);
646
647 if (sample_id_all)
648 swap_sample_id_all(event, &event->read + 1);
649}
650
651static void perf_event__aux_swap(union perf_event *event, bool sample_id_all)
652{
653 event->aux.aux_offset = bswap_64(event->aux.aux_offset);
654 event->aux.aux_size = bswap_64(event->aux.aux_size);
655 event->aux.flags = bswap_64(event->aux.flags);
656
657 if (sample_id_all)
658 swap_sample_id_all(event, &event->aux + 1);
659}
660
661static void perf_event__itrace_start_swap(union perf_event *event,
662 bool sample_id_all)
663{
664 event->itrace_start.pid = bswap_32(event->itrace_start.pid);
665 event->itrace_start.tid = bswap_32(event->itrace_start.tid);
666
667 if (sample_id_all)
668 swap_sample_id_all(event, &event->itrace_start + 1);
669}
670
671static void perf_event__switch_swap(union perf_event *event, bool sample_id_all)
672{
673 if (event->header.type == PERF_RECORD_SWITCH_CPU_WIDE) {
674 event->context_switch.next_prev_pid =
675 bswap_32(event->context_switch.next_prev_pid);
676 event->context_switch.next_prev_tid =
677 bswap_32(event->context_switch.next_prev_tid);
678 }
679
680 if (sample_id_all)
681 swap_sample_id_all(event, &event->context_switch + 1);
682}
683
684static void perf_event__text_poke_swap(union perf_event *event, bool sample_id_all)
685{
686 event->text_poke.addr = bswap_64(event->text_poke.addr);
687 event->text_poke.old_len = bswap_16(event->text_poke.old_len);
688 event->text_poke.new_len = bswap_16(event->text_poke.new_len);
689
690 if (sample_id_all) {
691 size_t len = sizeof(event->text_poke.old_len) +
692 sizeof(event->text_poke.new_len) +
693 event->text_poke.old_len +
694 event->text_poke.new_len;
695 void *data = &event->text_poke.old_len;
696
697 data += PERF_ALIGN(len, sizeof(u64));
698 swap_sample_id_all(event, data);
699 }
700}
701
702static void perf_event__throttle_swap(union perf_event *event,
703 bool sample_id_all)
704{
705 event->throttle.time = bswap_64(event->throttle.time);
706 event->throttle.id = bswap_64(event->throttle.id);
707 event->throttle.stream_id = bswap_64(event->throttle.stream_id);
708
709 if (sample_id_all)
710 swap_sample_id_all(event, &event->throttle + 1);
711}
712
713static void perf_event__namespaces_swap(union perf_event *event,
714 bool sample_id_all)
715{
716 u64 i;
717
718 event->namespaces.pid = bswap_32(event->namespaces.pid);
719 event->namespaces.tid = bswap_32(event->namespaces.tid);
720 event->namespaces.nr_namespaces = bswap_64(event->namespaces.nr_namespaces);
721
722 for (i = 0; i < event->namespaces.nr_namespaces; i++) {
723 struct perf_ns_link_info *ns = &event->namespaces.link_info[i];
724
725 ns->dev = bswap_64(ns->dev);
726 ns->ino = bswap_64(ns->ino);
727 }
728
729 if (sample_id_all)
730 swap_sample_id_all(event, &event->namespaces.link_info[i]);
731}
732
733static void perf_event__cgroup_swap(union perf_event *event, bool sample_id_all)
734{
735 event->cgroup.id = bswap_64(event->cgroup.id);
736
737 if (sample_id_all) {
738 void *data = &event->cgroup.path;
739
740 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
741 swap_sample_id_all(event, data);
742 }
743}
744
745static u8 revbyte(u8 b)
746{
747 int rev = (b >> 4) | ((b & 0xf) << 4);
748 rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2);
749 rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1);
750 return (u8) rev;
751}
752
753/*
754 * XXX this is hack in attempt to carry flags bitfield
755 * through endian village. ABI says:
756 *
757 * Bit-fields are allocated from right to left (least to most significant)
758 * on little-endian implementations and from left to right (most to least
759 * significant) on big-endian implementations.
760 *
761 * The above seems to be byte specific, so we need to reverse each
762 * byte of the bitfield. 'Internet' also says this might be implementation
763 * specific and we probably need proper fix and carry perf_event_attr
764 * bitfield flags in separate data file FEAT_ section. Thought this seems
765 * to work for now.
766 */
767static void swap_bitfield(u8 *p, unsigned len)
768{
769 unsigned i;
770
771 for (i = 0; i < len; i++) {
772 *p = revbyte(*p);
773 p++;
774 }
775}
776
777/* exported for swapping attributes in file header */
778void perf_event__attr_swap(struct perf_event_attr *attr)
779{
780 attr->type = bswap_32(attr->type);
781 attr->size = bswap_32(attr->size);
782
783#define bswap_safe(f, n) \
784 (attr->size > (offsetof(struct perf_event_attr, f) + \
785 sizeof(attr->f) * (n)))
786#define bswap_field(f, sz) \
787do { \
788 if (bswap_safe(f, 0)) \
789 attr->f = bswap_##sz(attr->f); \
790} while(0)
791#define bswap_field_16(f) bswap_field(f, 16)
792#define bswap_field_32(f) bswap_field(f, 32)
793#define bswap_field_64(f) bswap_field(f, 64)
794
795 bswap_field_64(config);
796 bswap_field_64(sample_period);
797 bswap_field_64(sample_type);
798 bswap_field_64(read_format);
799 bswap_field_32(wakeup_events);
800 bswap_field_32(bp_type);
801 bswap_field_64(bp_addr);
802 bswap_field_64(bp_len);
803 bswap_field_64(branch_sample_type);
804 bswap_field_64(sample_regs_user);
805 bswap_field_32(sample_stack_user);
806 bswap_field_32(aux_watermark);
807 bswap_field_16(sample_max_stack);
808 bswap_field_32(aux_sample_size);
809
810 /*
811 * After read_format are bitfields. Check read_format because
812 * we are unable to use offsetof on bitfield.
813 */
814 if (bswap_safe(read_format, 1))
815 swap_bitfield((u8 *) (&attr->read_format + 1),
816 sizeof(u64));
817#undef bswap_field_64
818#undef bswap_field_32
819#undef bswap_field
820#undef bswap_safe
821}
822
823static void perf_event__hdr_attr_swap(union perf_event *event,
824 bool sample_id_all __maybe_unused)
825{
826 size_t size;
827
828 perf_event__attr_swap(&event->attr.attr);
829
830 size = event->header.size;
831 size -= (void *)&event->attr.id - (void *)event;
832 mem_bswap_64(event->attr.id, size);
833}
834
835static void perf_event__event_update_swap(union perf_event *event,
836 bool sample_id_all __maybe_unused)
837{
838 event->event_update.type = bswap_64(event->event_update.type);
839 event->event_update.id = bswap_64(event->event_update.id);
840}
841
842static void perf_event__event_type_swap(union perf_event *event,
843 bool sample_id_all __maybe_unused)
844{
845 event->event_type.event_type.event_id =
846 bswap_64(event->event_type.event_type.event_id);
847}
848
849static void perf_event__tracing_data_swap(union perf_event *event,
850 bool sample_id_all __maybe_unused)
851{
852 event->tracing_data.size = bswap_32(event->tracing_data.size);
853}
854
855static void perf_event__auxtrace_info_swap(union perf_event *event,
856 bool sample_id_all __maybe_unused)
857{
858 size_t size;
859
860 event->auxtrace_info.type = bswap_32(event->auxtrace_info.type);
861
862 size = event->header.size;
863 size -= (void *)&event->auxtrace_info.priv - (void *)event;
864 mem_bswap_64(event->auxtrace_info.priv, size);
865}
866
867static void perf_event__auxtrace_swap(union perf_event *event,
868 bool sample_id_all __maybe_unused)
869{
870 event->auxtrace.size = bswap_64(event->auxtrace.size);
871 event->auxtrace.offset = bswap_64(event->auxtrace.offset);
872 event->auxtrace.reference = bswap_64(event->auxtrace.reference);
873 event->auxtrace.idx = bswap_32(event->auxtrace.idx);
874 event->auxtrace.tid = bswap_32(event->auxtrace.tid);
875 event->auxtrace.cpu = bswap_32(event->auxtrace.cpu);
876}
877
878static void perf_event__auxtrace_error_swap(union perf_event *event,
879 bool sample_id_all __maybe_unused)
880{
881 event->auxtrace_error.type = bswap_32(event->auxtrace_error.type);
882 event->auxtrace_error.code = bswap_32(event->auxtrace_error.code);
883 event->auxtrace_error.cpu = bswap_32(event->auxtrace_error.cpu);
884 event->auxtrace_error.pid = bswap_32(event->auxtrace_error.pid);
885 event->auxtrace_error.tid = bswap_32(event->auxtrace_error.tid);
886 event->auxtrace_error.fmt = bswap_32(event->auxtrace_error.fmt);
887 event->auxtrace_error.ip = bswap_64(event->auxtrace_error.ip);
888 if (event->auxtrace_error.fmt)
889 event->auxtrace_error.time = bswap_64(event->auxtrace_error.time);
890}
891
892static void perf_event__thread_map_swap(union perf_event *event,
893 bool sample_id_all __maybe_unused)
894{
895 unsigned i;
896
897 event->thread_map.nr = bswap_64(event->thread_map.nr);
898
899 for (i = 0; i < event->thread_map.nr; i++)
900 event->thread_map.entries[i].pid = bswap_64(event->thread_map.entries[i].pid);
901}
902
903static void perf_event__cpu_map_swap(union perf_event *event,
904 bool sample_id_all __maybe_unused)
905{
906 struct perf_record_cpu_map_data *data = &event->cpu_map.data;
907 struct cpu_map_entries *cpus;
908 struct perf_record_record_cpu_map *mask;
909 unsigned i;
910
911 data->type = bswap_16(data->type);
912
913 switch (data->type) {
914 case PERF_CPU_MAP__CPUS:
915 cpus = (struct cpu_map_entries *)data->data;
916
917 cpus->nr = bswap_16(cpus->nr);
918
919 for (i = 0; i < cpus->nr; i++)
920 cpus->cpu[i] = bswap_16(cpus->cpu[i]);
921 break;
922 case PERF_CPU_MAP__MASK:
923 mask = (struct perf_record_record_cpu_map *)data->data;
924
925 mask->nr = bswap_16(mask->nr);
926 mask->long_size = bswap_16(mask->long_size);
927
928 switch (mask->long_size) {
929 case 4: mem_bswap_32(&mask->mask, mask->nr); break;
930 case 8: mem_bswap_64(&mask->mask, mask->nr); break;
931 default:
932 pr_err("cpu_map swap: unsupported long size\n");
933 }
934 default:
935 break;
936 }
937}
938
939static void perf_event__stat_config_swap(union perf_event *event,
940 bool sample_id_all __maybe_unused)
941{
942 u64 size;
943
944 size = bswap_64(event->stat_config.nr) * sizeof(event->stat_config.data[0]);
945 size += 1; /* nr item itself */
946 mem_bswap_64(&event->stat_config.nr, size);
947}
948
949static void perf_event__stat_swap(union perf_event *event,
950 bool sample_id_all __maybe_unused)
951{
952 event->stat.id = bswap_64(event->stat.id);
953 event->stat.thread = bswap_32(event->stat.thread);
954 event->stat.cpu = bswap_32(event->stat.cpu);
955 event->stat.val = bswap_64(event->stat.val);
956 event->stat.ena = bswap_64(event->stat.ena);
957 event->stat.run = bswap_64(event->stat.run);
958}
959
960static void perf_event__stat_round_swap(union perf_event *event,
961 bool sample_id_all __maybe_unused)
962{
963 event->stat_round.type = bswap_64(event->stat_round.type);
964 event->stat_round.time = bswap_64(event->stat_round.time);
965}
966
967static void perf_event__time_conv_swap(union perf_event *event,
968 bool sample_id_all __maybe_unused)
969{
970 event->time_conv.time_shift = bswap_64(event->time_conv.time_shift);
971 event->time_conv.time_mult = bswap_64(event->time_conv.time_mult);
972 event->time_conv.time_zero = bswap_64(event->time_conv.time_zero);
973
974 if (event_contains(event->time_conv, time_cycles)) {
975 event->time_conv.time_cycles = bswap_64(event->time_conv.time_cycles);
976 event->time_conv.time_mask = bswap_64(event->time_conv.time_mask);
977 }
978}
979
980typedef void (*perf_event__swap_op)(union perf_event *event,
981 bool sample_id_all);
982
983static perf_event__swap_op perf_event__swap_ops[] = {
984 [PERF_RECORD_MMAP] = perf_event__mmap_swap,
985 [PERF_RECORD_MMAP2] = perf_event__mmap2_swap,
986 [PERF_RECORD_COMM] = perf_event__comm_swap,
987 [PERF_RECORD_FORK] = perf_event__task_swap,
988 [PERF_RECORD_EXIT] = perf_event__task_swap,
989 [PERF_RECORD_LOST] = perf_event__all64_swap,
990 [PERF_RECORD_READ] = perf_event__read_swap,
991 [PERF_RECORD_THROTTLE] = perf_event__throttle_swap,
992 [PERF_RECORD_UNTHROTTLE] = perf_event__throttle_swap,
993 [PERF_RECORD_SAMPLE] = perf_event__all64_swap,
994 [PERF_RECORD_AUX] = perf_event__aux_swap,
995 [PERF_RECORD_ITRACE_START] = perf_event__itrace_start_swap,
996 [PERF_RECORD_LOST_SAMPLES] = perf_event__all64_swap,
997 [PERF_RECORD_SWITCH] = perf_event__switch_swap,
998 [PERF_RECORD_SWITCH_CPU_WIDE] = perf_event__switch_swap,
999 [PERF_RECORD_NAMESPACES] = perf_event__namespaces_swap,
1000 [PERF_RECORD_CGROUP] = perf_event__cgroup_swap,
1001 [PERF_RECORD_TEXT_POKE] = perf_event__text_poke_swap,
1002 [PERF_RECORD_HEADER_ATTR] = perf_event__hdr_attr_swap,
1003 [PERF_RECORD_HEADER_EVENT_TYPE] = perf_event__event_type_swap,
1004 [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
1005 [PERF_RECORD_HEADER_BUILD_ID] = NULL,
1006 [PERF_RECORD_ID_INDEX] = perf_event__all64_swap,
1007 [PERF_RECORD_AUXTRACE_INFO] = perf_event__auxtrace_info_swap,
1008 [PERF_RECORD_AUXTRACE] = perf_event__auxtrace_swap,
1009 [PERF_RECORD_AUXTRACE_ERROR] = perf_event__auxtrace_error_swap,
1010 [PERF_RECORD_THREAD_MAP] = perf_event__thread_map_swap,
1011 [PERF_RECORD_CPU_MAP] = perf_event__cpu_map_swap,
1012 [PERF_RECORD_STAT_CONFIG] = perf_event__stat_config_swap,
1013 [PERF_RECORD_STAT] = perf_event__stat_swap,
1014 [PERF_RECORD_STAT_ROUND] = perf_event__stat_round_swap,
1015 [PERF_RECORD_EVENT_UPDATE] = perf_event__event_update_swap,
1016 [PERF_RECORD_TIME_CONV] = perf_event__time_conv_swap,
1017 [PERF_RECORD_HEADER_MAX] = NULL,
1018};
1019
1020/*
1021 * When perf record finishes a pass on every buffers, it records this pseudo
1022 * event.
1023 * We record the max timestamp t found in the pass n.
1024 * Assuming these timestamps are monotonic across cpus, we know that if
1025 * a buffer still has events with timestamps below t, they will be all
1026 * available and then read in the pass n + 1.
1027 * Hence when we start to read the pass n + 2, we can safely flush every
1028 * events with timestamps below t.
1029 *
1030 * ============ PASS n =================
1031 * CPU 0 | CPU 1
1032 * |
1033 * cnt1 timestamps | cnt2 timestamps
1034 * 1 | 2
1035 * 2 | 3
1036 * - | 4 <--- max recorded
1037 *
1038 * ============ PASS n + 1 ==============
1039 * CPU 0 | CPU 1
1040 * |
1041 * cnt1 timestamps | cnt2 timestamps
1042 * 3 | 5
1043 * 4 | 6
1044 * 5 | 7 <---- max recorded
1045 *
1046 * Flush every events below timestamp 4
1047 *
1048 * ============ PASS n + 2 ==============
1049 * CPU 0 | CPU 1
1050 * |
1051 * cnt1 timestamps | cnt2 timestamps
1052 * 6 | 8
1053 * 7 | 9
1054 * - | 10
1055 *
1056 * Flush every events below timestamp 7
1057 * etc...
1058 */
1059static int process_finished_round(struct perf_tool *tool __maybe_unused,
1060 union perf_event *event __maybe_unused,
1061 struct ordered_events *oe)
1062{
1063 if (dump_trace)
1064 fprintf(stdout, "\n");
1065 return ordered_events__flush(oe, OE_FLUSH__ROUND);
1066}
1067
1068int perf_session__queue_event(struct perf_session *s, union perf_event *event,
1069 u64 timestamp, u64 file_offset)
1070{
1071 return ordered_events__queue(&s->ordered_events, event, timestamp, file_offset);
1072}
1073
1074static void callchain__lbr_callstack_printf(struct perf_sample *sample)
1075{
1076 struct ip_callchain *callchain = sample->callchain;
1077 struct branch_stack *lbr_stack = sample->branch_stack;
1078 struct branch_entry *entries = perf_sample__branch_entries(sample);
1079 u64 kernel_callchain_nr = callchain->nr;
1080 unsigned int i;
1081
1082 for (i = 0; i < kernel_callchain_nr; i++) {
1083 if (callchain->ips[i] == PERF_CONTEXT_USER)
1084 break;
1085 }
1086
1087 if ((i != kernel_callchain_nr) && lbr_stack->nr) {
1088 u64 total_nr;
1089 /*
1090 * LBR callstack can only get user call chain,
1091 * i is kernel call chain number,
1092 * 1 is PERF_CONTEXT_USER.
1093 *
1094 * The user call chain is stored in LBR registers.
1095 * LBR are pair registers. The caller is stored
1096 * in "from" register, while the callee is stored
1097 * in "to" register.
1098 * For example, there is a call stack
1099 * "A"->"B"->"C"->"D".
1100 * The LBR registers will be recorded like
1101 * "C"->"D", "B"->"C", "A"->"B".
1102 * So only the first "to" register and all "from"
1103 * registers are needed to construct the whole stack.
1104 */
1105 total_nr = i + 1 + lbr_stack->nr + 1;
1106 kernel_callchain_nr = i + 1;
1107
1108 printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr);
1109
1110 for (i = 0; i < kernel_callchain_nr; i++)
1111 printf("..... %2d: %016" PRIx64 "\n",
1112 i, callchain->ips[i]);
1113
1114 printf("..... %2d: %016" PRIx64 "\n",
1115 (int)(kernel_callchain_nr), entries[0].to);
1116 for (i = 0; i < lbr_stack->nr; i++)
1117 printf("..... %2d: %016" PRIx64 "\n",
1118 (int)(i + kernel_callchain_nr + 1), entries[i].from);
1119 }
1120}
1121
1122static void callchain__printf(struct evsel *evsel,
1123 struct perf_sample *sample)
1124{
1125 unsigned int i;
1126 struct ip_callchain *callchain = sample->callchain;
1127
1128 if (evsel__has_branch_callstack(evsel))
1129 callchain__lbr_callstack_printf(sample);
1130
1131 printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr);
1132
1133 for (i = 0; i < callchain->nr; i++)
1134 printf("..... %2d: %016" PRIx64 "\n",
1135 i, callchain->ips[i]);
1136}
1137
1138static void branch_stack__printf(struct perf_sample *sample, bool callstack)
1139{
1140 struct branch_entry *entries = perf_sample__branch_entries(sample);
1141 uint64_t i;
1142
1143 printf("%s: nr:%" PRIu64 "\n",
1144 !callstack ? "... branch stack" : "... branch callstack",
1145 sample->branch_stack->nr);
1146
1147 for (i = 0; i < sample->branch_stack->nr; i++) {
1148 struct branch_entry *e = &entries[i];
1149
1150 if (!callstack) {
1151 printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x\n",
1152 i, e->from, e->to,
1153 (unsigned short)e->flags.cycles,
1154 e->flags.mispred ? "M" : " ",
1155 e->flags.predicted ? "P" : " ",
1156 e->flags.abort ? "A" : " ",
1157 e->flags.in_tx ? "T" : " ",
1158 (unsigned)e->flags.reserved);
1159 } else {
1160 printf("..... %2"PRIu64": %016" PRIx64 "\n",
1161 i, i > 0 ? e->from : e->to);
1162 }
1163 }
1164}
1165
1166static void regs_dump__printf(u64 mask, u64 *regs)
1167{
1168 unsigned rid, i = 0;
1169
1170 for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) {
1171 u64 val = regs[i++];
1172
1173 printf(".... %-5s 0x%016" PRIx64 "\n",
1174 perf_reg_name(rid), val);
1175 }
1176}
1177
1178static const char *regs_abi[] = {
1179 [PERF_SAMPLE_REGS_ABI_NONE] = "none",
1180 [PERF_SAMPLE_REGS_ABI_32] = "32-bit",
1181 [PERF_SAMPLE_REGS_ABI_64] = "64-bit",
1182};
1183
1184static inline const char *regs_dump_abi(struct regs_dump *d)
1185{
1186 if (d->abi > PERF_SAMPLE_REGS_ABI_64)
1187 return "unknown";
1188
1189 return regs_abi[d->abi];
1190}
1191
1192static void regs__printf(const char *type, struct regs_dump *regs)
1193{
1194 u64 mask = regs->mask;
1195
1196 printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n",
1197 type,
1198 mask,
1199 regs_dump_abi(regs));
1200
1201 regs_dump__printf(mask, regs->regs);
1202}
1203
1204static void regs_user__printf(struct perf_sample *sample)
1205{
1206 struct regs_dump *user_regs = &sample->user_regs;
1207
1208 if (user_regs->regs)
1209 regs__printf("user", user_regs);
1210}
1211
1212static void regs_intr__printf(struct perf_sample *sample)
1213{
1214 struct regs_dump *intr_regs = &sample->intr_regs;
1215
1216 if (intr_regs->regs)
1217 regs__printf("intr", intr_regs);
1218}
1219
1220static void stack_user__printf(struct stack_dump *dump)
1221{
1222 printf("... ustack: size %" PRIu64 ", offset 0x%x\n",
1223 dump->size, dump->offset);
1224}
1225
1226static void evlist__print_tstamp(struct evlist *evlist, union perf_event *event, struct perf_sample *sample)
1227{
1228 u64 sample_type = __evlist__combined_sample_type(evlist);
1229
1230 if (event->header.type != PERF_RECORD_SAMPLE &&
1231 !evlist__sample_id_all(evlist)) {
1232 fputs("-1 -1 ", stdout);
1233 return;
1234 }
1235
1236 if ((sample_type & PERF_SAMPLE_CPU))
1237 printf("%u ", sample->cpu);
1238
1239 if (sample_type & PERF_SAMPLE_TIME)
1240 printf("%" PRIu64 " ", sample->time);
1241}
1242
1243static void sample_read__printf(struct perf_sample *sample, u64 read_format)
1244{
1245 printf("... sample_read:\n");
1246
1247 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1248 printf("...... time enabled %016" PRIx64 "\n",
1249 sample->read.time_enabled);
1250
1251 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1252 printf("...... time running %016" PRIx64 "\n",
1253 sample->read.time_running);
1254
1255 if (read_format & PERF_FORMAT_GROUP) {
1256 u64 i;
1257
1258 printf(".... group nr %" PRIu64 "\n", sample->read.group.nr);
1259
1260 for (i = 0; i < sample->read.group.nr; i++) {
1261 struct sample_read_value *value;
1262
1263 value = &sample->read.group.values[i];
1264 printf("..... id %016" PRIx64
1265 ", value %016" PRIx64 "\n",
1266 value->id, value->value);
1267 }
1268 } else
1269 printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n",
1270 sample->read.one.id, sample->read.one.value);
1271}
1272
1273static void dump_event(struct evlist *evlist, union perf_event *event,
1274 u64 file_offset, struct perf_sample *sample)
1275{
1276 if (!dump_trace)
1277 return;
1278
1279 printf("\n%#" PRIx64 " [%#x]: event: %d\n",
1280 file_offset, event->header.size, event->header.type);
1281
1282 trace_event(event);
1283 if (event->header.type == PERF_RECORD_SAMPLE && evlist->trace_event_sample_raw)
1284 evlist->trace_event_sample_raw(evlist, event, sample);
1285
1286 if (sample)
1287 evlist__print_tstamp(evlist, event, sample);
1288
1289 printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset,
1290 event->header.size, perf_event__name(event->header.type));
1291}
1292
1293char *get_page_size_name(u64 size, char *str)
1294{
1295 if (!size || !unit_number__scnprintf(str, PAGE_SIZE_NAME_LEN, size))
1296 snprintf(str, PAGE_SIZE_NAME_LEN, "%s", "N/A");
1297
1298 return str;
1299}
1300
1301static void dump_sample(struct evsel *evsel, union perf_event *event,
1302 struct perf_sample *sample)
1303{
1304 u64 sample_type;
1305 char str[PAGE_SIZE_NAME_LEN];
1306
1307 if (!dump_trace)
1308 return;
1309
1310 printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n",
1311 event->header.misc, sample->pid, sample->tid, sample->ip,
1312 sample->period, sample->addr);
1313
1314 sample_type = evsel->core.attr.sample_type;
1315
1316 if (evsel__has_callchain(evsel))
1317 callchain__printf(evsel, sample);
1318
1319 if (evsel__has_br_stack(evsel))
1320 branch_stack__printf(sample, evsel__has_branch_callstack(evsel));
1321
1322 if (sample_type & PERF_SAMPLE_REGS_USER)
1323 regs_user__printf(sample);
1324
1325 if (sample_type & PERF_SAMPLE_REGS_INTR)
1326 regs_intr__printf(sample);
1327
1328 if (sample_type & PERF_SAMPLE_STACK_USER)
1329 stack_user__printf(&sample->user_stack);
1330
1331 if (sample_type & PERF_SAMPLE_WEIGHT_TYPE) {
1332 printf("... weight: %" PRIu64 "", sample->weight);
1333 if (sample_type & PERF_SAMPLE_WEIGHT_STRUCT) {
1334 printf(",0x%"PRIx16"", sample->ins_lat);
1335 printf(",0x%"PRIx16"", sample->p_stage_cyc);
1336 }
1337 printf("\n");
1338 }
1339
1340 if (sample_type & PERF_SAMPLE_DATA_SRC)
1341 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src);
1342
1343 if (sample_type & PERF_SAMPLE_PHYS_ADDR)
1344 printf(" .. phys_addr: 0x%"PRIx64"\n", sample->phys_addr);
1345
1346 if (sample_type & PERF_SAMPLE_DATA_PAGE_SIZE)
1347 printf(" .. data page size: %s\n", get_page_size_name(sample->data_page_size, str));
1348
1349 if (sample_type & PERF_SAMPLE_CODE_PAGE_SIZE)
1350 printf(" .. code page size: %s\n", get_page_size_name(sample->code_page_size, str));
1351
1352 if (sample_type & PERF_SAMPLE_TRANSACTION)
1353 printf("... transaction: %" PRIx64 "\n", sample->transaction);
1354
1355 if (sample_type & PERF_SAMPLE_READ)
1356 sample_read__printf(sample, evsel->core.attr.read_format);
1357}
1358
1359static void dump_read(struct evsel *evsel, union perf_event *event)
1360{
1361 struct perf_record_read *read_event = &event->read;
1362 u64 read_format;
1363
1364 if (!dump_trace)
1365 return;
1366
1367 printf(": %d %d %s %" PRI_lu64 "\n", event->read.pid, event->read.tid,
1368 evsel__name(evsel), event->read.value);
1369
1370 if (!evsel)
1371 return;
1372
1373 read_format = evsel->core.attr.read_format;
1374
1375 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1376 printf("... time enabled : %" PRI_lu64 "\n", read_event->time_enabled);
1377
1378 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1379 printf("... time running : %" PRI_lu64 "\n", read_event->time_running);
1380
1381 if (read_format & PERF_FORMAT_ID)
1382 printf("... id : %" PRI_lu64 "\n", read_event->id);
1383}
1384
1385static struct machine *machines__find_for_cpumode(struct machines *machines,
1386 union perf_event *event,
1387 struct perf_sample *sample)
1388{
1389 if (perf_guest &&
1390 ((sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL) ||
1391 (sample->cpumode == PERF_RECORD_MISC_GUEST_USER))) {
1392 u32 pid;
1393
1394 if (event->header.type == PERF_RECORD_MMAP
1395 || event->header.type == PERF_RECORD_MMAP2)
1396 pid = event->mmap.pid;
1397 else
1398 pid = sample->pid;
1399
1400 return machines__find_guest(machines, pid);
1401 }
1402
1403 return &machines->host;
1404}
1405
1406static int deliver_sample_value(struct evlist *evlist,
1407 struct perf_tool *tool,
1408 union perf_event *event,
1409 struct perf_sample *sample,
1410 struct sample_read_value *v,
1411 struct machine *machine)
1412{
1413 struct perf_sample_id *sid = evlist__id2sid(evlist, v->id);
1414 struct evsel *evsel;
1415
1416 if (sid) {
1417 sample->id = v->id;
1418 sample->period = v->value - sid->period;
1419 sid->period = v->value;
1420 }
1421
1422 if (!sid || sid->evsel == NULL) {
1423 ++evlist->stats.nr_unknown_id;
1424 return 0;
1425 }
1426
1427 /*
1428 * There's no reason to deliver sample
1429 * for zero period, bail out.
1430 */
1431 if (!sample->period)
1432 return 0;
1433
1434 evsel = container_of(sid->evsel, struct evsel, core);
1435 return tool->sample(tool, event, sample, evsel, machine);
1436}
1437
1438static int deliver_sample_group(struct evlist *evlist,
1439 struct perf_tool *tool,
1440 union perf_event *event,
1441 struct perf_sample *sample,
1442 struct machine *machine)
1443{
1444 int ret = -EINVAL;
1445 u64 i;
1446
1447 for (i = 0; i < sample->read.group.nr; i++) {
1448 ret = deliver_sample_value(evlist, tool, event, sample,
1449 &sample->read.group.values[i],
1450 machine);
1451 if (ret)
1452 break;
1453 }
1454
1455 return ret;
1456}
1457
1458static int evlist__deliver_sample(struct evlist *evlist, struct perf_tool *tool,
1459 union perf_event *event, struct perf_sample *sample,
1460 struct evsel *evsel, struct machine *machine)
1461{
1462 /* We know evsel != NULL. */
1463 u64 sample_type = evsel->core.attr.sample_type;
1464 u64 read_format = evsel->core.attr.read_format;
1465
1466 /* Standard sample delivery. */
1467 if (!(sample_type & PERF_SAMPLE_READ))
1468 return tool->sample(tool, event, sample, evsel, machine);
1469
1470 /* For PERF_SAMPLE_READ we have either single or group mode. */
1471 if (read_format & PERF_FORMAT_GROUP)
1472 return deliver_sample_group(evlist, tool, event, sample,
1473 machine);
1474 else
1475 return deliver_sample_value(evlist, tool, event, sample,
1476 &sample->read.one, machine);
1477}
1478
1479static int machines__deliver_event(struct machines *machines,
1480 struct evlist *evlist,
1481 union perf_event *event,
1482 struct perf_sample *sample,
1483 struct perf_tool *tool, u64 file_offset)
1484{
1485 struct evsel *evsel;
1486 struct machine *machine;
1487
1488 dump_event(evlist, event, file_offset, sample);
1489
1490 evsel = evlist__id2evsel(evlist, sample->id);
1491
1492 machine = machines__find_for_cpumode(machines, event, sample);
1493
1494 switch (event->header.type) {
1495 case PERF_RECORD_SAMPLE:
1496 if (evsel == NULL) {
1497 ++evlist->stats.nr_unknown_id;
1498 return 0;
1499 }
1500 dump_sample(evsel, event, sample);
1501 if (machine == NULL) {
1502 ++evlist->stats.nr_unprocessable_samples;
1503 return 0;
1504 }
1505 return evlist__deliver_sample(evlist, tool, event, sample, evsel, machine);
1506 case PERF_RECORD_MMAP:
1507 return tool->mmap(tool, event, sample, machine);
1508 case PERF_RECORD_MMAP2:
1509 if (event->header.misc & PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT)
1510 ++evlist->stats.nr_proc_map_timeout;
1511 return tool->mmap2(tool, event, sample, machine);
1512 case PERF_RECORD_COMM:
1513 return tool->comm(tool, event, sample, machine);
1514 case PERF_RECORD_NAMESPACES:
1515 return tool->namespaces(tool, event, sample, machine);
1516 case PERF_RECORD_CGROUP:
1517 return tool->cgroup(tool, event, sample, machine);
1518 case PERF_RECORD_FORK:
1519 return tool->fork(tool, event, sample, machine);
1520 case PERF_RECORD_EXIT:
1521 return tool->exit(tool, event, sample, machine);
1522 case PERF_RECORD_LOST:
1523 if (tool->lost == perf_event__process_lost)
1524 evlist->stats.total_lost += event->lost.lost;
1525 return tool->lost(tool, event, sample, machine);
1526 case PERF_RECORD_LOST_SAMPLES:
1527 if (tool->lost_samples == perf_event__process_lost_samples)
1528 evlist->stats.total_lost_samples += event->lost_samples.lost;
1529 return tool->lost_samples(tool, event, sample, machine);
1530 case PERF_RECORD_READ:
1531 dump_read(evsel, event);
1532 return tool->read(tool, event, sample, evsel, machine);
1533 case PERF_RECORD_THROTTLE:
1534 return tool->throttle(tool, event, sample, machine);
1535 case PERF_RECORD_UNTHROTTLE:
1536 return tool->unthrottle(tool, event, sample, machine);
1537 case PERF_RECORD_AUX:
1538 if (tool->aux == perf_event__process_aux) {
1539 if (event->aux.flags & PERF_AUX_FLAG_TRUNCATED)
1540 evlist->stats.total_aux_lost += 1;
1541 if (event->aux.flags & PERF_AUX_FLAG_PARTIAL)
1542 evlist->stats.total_aux_partial += 1;
1543 }
1544 return tool->aux(tool, event, sample, machine);
1545 case PERF_RECORD_ITRACE_START:
1546 return tool->itrace_start(tool, event, sample, machine);
1547 case PERF_RECORD_SWITCH:
1548 case PERF_RECORD_SWITCH_CPU_WIDE:
1549 return tool->context_switch(tool, event, sample, machine);
1550 case PERF_RECORD_KSYMBOL:
1551 return tool->ksymbol(tool, event, sample, machine);
1552 case PERF_RECORD_BPF_EVENT:
1553 return tool->bpf(tool, event, sample, machine);
1554 case PERF_RECORD_TEXT_POKE:
1555 return tool->text_poke(tool, event, sample, machine);
1556 default:
1557 ++evlist->stats.nr_unknown_events;
1558 return -1;
1559 }
1560}
1561
1562static int perf_session__deliver_event(struct perf_session *session,
1563 union perf_event *event,
1564 struct perf_tool *tool,
1565 u64 file_offset)
1566{
1567 struct perf_sample sample;
1568 int ret = evlist__parse_sample(session->evlist, event, &sample);
1569
1570 if (ret) {
1571 pr_err("Can't parse sample, err = %d\n", ret);
1572 return ret;
1573 }
1574
1575 ret = auxtrace__process_event(session, event, &sample, tool);
1576 if (ret < 0)
1577 return ret;
1578 if (ret > 0)
1579 return 0;
1580
1581 ret = machines__deliver_event(&session->machines, session->evlist,
1582 event, &sample, tool, file_offset);
1583
1584 if (dump_trace && sample.aux_sample.size)
1585 auxtrace__dump_auxtrace_sample(session, &sample);
1586
1587 return ret;
1588}
1589
1590static s64 perf_session__process_user_event(struct perf_session *session,
1591 union perf_event *event,
1592 u64 file_offset)
1593{
1594 struct ordered_events *oe = &session->ordered_events;
1595 struct perf_tool *tool = session->tool;
1596 struct perf_sample sample = { .time = 0, };
1597 int fd = perf_data__fd(session->data);
1598 int err;
1599
1600 if (event->header.type != PERF_RECORD_COMPRESSED ||
1601 tool->compressed == perf_session__process_compressed_event_stub)
1602 dump_event(session->evlist, event, file_offset, &sample);
1603
1604 /* These events are processed right away */
1605 switch (event->header.type) {
1606 case PERF_RECORD_HEADER_ATTR:
1607 err = tool->attr(tool, event, &session->evlist);
1608 if (err == 0) {
1609 perf_session__set_id_hdr_size(session);
1610 perf_session__set_comm_exec(session);
1611 }
1612 return err;
1613 case PERF_RECORD_EVENT_UPDATE:
1614 return tool->event_update(tool, event, &session->evlist);
1615 case PERF_RECORD_HEADER_EVENT_TYPE:
1616 /*
1617 * Deprecated, but we need to handle it for sake
1618 * of old data files create in pipe mode.
1619 */
1620 return 0;
1621 case PERF_RECORD_HEADER_TRACING_DATA:
1622 /*
1623 * Setup for reading amidst mmap, but only when we
1624 * are in 'file' mode. The 'pipe' fd is in proper
1625 * place already.
1626 */
1627 if (!perf_data__is_pipe(session->data))
1628 lseek(fd, file_offset, SEEK_SET);
1629 return tool->tracing_data(session, event);
1630 case PERF_RECORD_HEADER_BUILD_ID:
1631 return tool->build_id(session, event);
1632 case PERF_RECORD_FINISHED_ROUND:
1633 return tool->finished_round(tool, event, oe);
1634 case PERF_RECORD_ID_INDEX:
1635 return tool->id_index(session, event);
1636 case PERF_RECORD_AUXTRACE_INFO:
1637 return tool->auxtrace_info(session, event);
1638 case PERF_RECORD_AUXTRACE:
1639 /* setup for reading amidst mmap */
1640 lseek(fd, file_offset + event->header.size, SEEK_SET);
1641 return tool->auxtrace(session, event);
1642 case PERF_RECORD_AUXTRACE_ERROR:
1643 perf_session__auxtrace_error_inc(session, event);
1644 return tool->auxtrace_error(session, event);
1645 case PERF_RECORD_THREAD_MAP:
1646 return tool->thread_map(session, event);
1647 case PERF_RECORD_CPU_MAP:
1648 return tool->cpu_map(session, event);
1649 case PERF_RECORD_STAT_CONFIG:
1650 return tool->stat_config(session, event);
1651 case PERF_RECORD_STAT:
1652 return tool->stat(session, event);
1653 case PERF_RECORD_STAT_ROUND:
1654 return tool->stat_round(session, event);
1655 case PERF_RECORD_TIME_CONV:
1656 session->time_conv = event->time_conv;
1657 return tool->time_conv(session, event);
1658 case PERF_RECORD_HEADER_FEATURE:
1659 return tool->feature(session, event);
1660 case PERF_RECORD_COMPRESSED:
1661 err = tool->compressed(session, event, file_offset);
1662 if (err)
1663 dump_event(session->evlist, event, file_offset, &sample);
1664 return err;
1665 default:
1666 return -EINVAL;
1667 }
1668}
1669
1670int perf_session__deliver_synth_event(struct perf_session *session,
1671 union perf_event *event,
1672 struct perf_sample *sample)
1673{
1674 struct evlist *evlist = session->evlist;
1675 struct perf_tool *tool = session->tool;
1676
1677 events_stats__inc(&evlist->stats, event->header.type);
1678
1679 if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1680 return perf_session__process_user_event(session, event, 0);
1681
1682 return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0);
1683}
1684
1685static void event_swap(union perf_event *event, bool sample_id_all)
1686{
1687 perf_event__swap_op swap;
1688
1689 swap = perf_event__swap_ops[event->header.type];
1690 if (swap)
1691 swap(event, sample_id_all);
1692}
1693
1694int perf_session__peek_event(struct perf_session *session, off_t file_offset,
1695 void *buf, size_t buf_sz,
1696 union perf_event **event_ptr,
1697 struct perf_sample *sample)
1698{
1699 union perf_event *event;
1700 size_t hdr_sz, rest;
1701 int fd;
1702
1703 if (session->one_mmap && !session->header.needs_swap) {
1704 event = file_offset - session->one_mmap_offset +
1705 session->one_mmap_addr;
1706 goto out_parse_sample;
1707 }
1708
1709 if (perf_data__is_pipe(session->data))
1710 return -1;
1711
1712 fd = perf_data__fd(session->data);
1713 hdr_sz = sizeof(struct perf_event_header);
1714
1715 if (buf_sz < hdr_sz)
1716 return -1;
1717
1718 if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 ||
1719 readn(fd, buf, hdr_sz) != (ssize_t)hdr_sz)
1720 return -1;
1721
1722 event = (union perf_event *)buf;
1723
1724 if (session->header.needs_swap)
1725 perf_event_header__bswap(&event->header);
1726
1727 if (event->header.size < hdr_sz || event->header.size > buf_sz)
1728 return -1;
1729
1730 buf += hdr_sz;
1731 rest = event->header.size - hdr_sz;
1732
1733 if (readn(fd, buf, rest) != (ssize_t)rest)
1734 return -1;
1735
1736 if (session->header.needs_swap)
1737 event_swap(event, evlist__sample_id_all(session->evlist));
1738
1739out_parse_sample:
1740
1741 if (sample && event->header.type < PERF_RECORD_USER_TYPE_START &&
1742 evlist__parse_sample(session->evlist, event, sample))
1743 return -1;
1744
1745 *event_ptr = event;
1746
1747 return 0;
1748}
1749
1750int perf_session__peek_events(struct perf_session *session, u64 offset,
1751 u64 size, peek_events_cb_t cb, void *data)
1752{
1753 u64 max_offset = offset + size;
1754 char buf[PERF_SAMPLE_MAX_SIZE];
1755 union perf_event *event;
1756 int err;
1757
1758 do {
1759 err = perf_session__peek_event(session, offset, buf,
1760 PERF_SAMPLE_MAX_SIZE, &event,
1761 NULL);
1762 if (err)
1763 return err;
1764
1765 err = cb(session, event, offset, data);
1766 if (err)
1767 return err;
1768
1769 offset += event->header.size;
1770 if (event->header.type == PERF_RECORD_AUXTRACE)
1771 offset += event->auxtrace.size;
1772
1773 } while (offset < max_offset);
1774
1775 return err;
1776}
1777
1778static s64 perf_session__process_event(struct perf_session *session,
1779 union perf_event *event, u64 file_offset)
1780{
1781 struct evlist *evlist = session->evlist;
1782 struct perf_tool *tool = session->tool;
1783 int ret;
1784
1785 if (session->header.needs_swap)
1786 event_swap(event, evlist__sample_id_all(evlist));
1787
1788 if (event->header.type >= PERF_RECORD_HEADER_MAX)
1789 return -EINVAL;
1790
1791 events_stats__inc(&evlist->stats, event->header.type);
1792
1793 if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1794 return perf_session__process_user_event(session, event, file_offset);
1795
1796 if (tool->ordered_events) {
1797 u64 timestamp = -1ULL;
1798
1799 ret = evlist__parse_sample_timestamp(evlist, event, ×tamp);
1800 if (ret && ret != -1)
1801 return ret;
1802
1803 ret = perf_session__queue_event(session, event, timestamp, file_offset);
1804 if (ret != -ETIME)
1805 return ret;
1806 }
1807
1808 return perf_session__deliver_event(session, event, tool, file_offset);
1809}
1810
1811void perf_event_header__bswap(struct perf_event_header *hdr)
1812{
1813 hdr->type = bswap_32(hdr->type);
1814 hdr->misc = bswap_16(hdr->misc);
1815 hdr->size = bswap_16(hdr->size);
1816}
1817
1818struct thread *perf_session__findnew(struct perf_session *session, pid_t pid)
1819{
1820 return machine__findnew_thread(&session->machines.host, -1, pid);
1821}
1822
1823int perf_session__register_idle_thread(struct perf_session *session)
1824{
1825 struct thread *thread = machine__idle_thread(&session->machines.host);
1826
1827 /* machine__idle_thread() got the thread, so put it */
1828 thread__put(thread);
1829 return thread ? 0 : -1;
1830}
1831
1832static void
1833perf_session__warn_order(const struct perf_session *session)
1834{
1835 const struct ordered_events *oe = &session->ordered_events;
1836 struct evsel *evsel;
1837 bool should_warn = true;
1838
1839 evlist__for_each_entry(session->evlist, evsel) {
1840 if (evsel->core.attr.write_backward)
1841 should_warn = false;
1842 }
1843
1844 if (!should_warn)
1845 return;
1846 if (oe->nr_unordered_events != 0)
1847 ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events);
1848}
1849
1850static void perf_session__warn_about_errors(const struct perf_session *session)
1851{
1852 const struct events_stats *stats = &session->evlist->stats;
1853
1854 if (session->tool->lost == perf_event__process_lost &&
1855 stats->nr_events[PERF_RECORD_LOST] != 0) {
1856 ui__warning("Processed %d events and lost %d chunks!\n\n"
1857 "Check IO/CPU overload!\n\n",
1858 stats->nr_events[0],
1859 stats->nr_events[PERF_RECORD_LOST]);
1860 }
1861
1862 if (session->tool->lost_samples == perf_event__process_lost_samples) {
1863 double drop_rate;
1864
1865 drop_rate = (double)stats->total_lost_samples /
1866 (double) (stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples);
1867 if (drop_rate > 0.05) {
1868 ui__warning("Processed %" PRIu64 " samples and lost %3.2f%%!\n\n",
1869 stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples,
1870 drop_rate * 100.0);
1871 }
1872 }
1873
1874 if (session->tool->aux == perf_event__process_aux &&
1875 stats->total_aux_lost != 0) {
1876 ui__warning("AUX data lost %" PRIu64 " times out of %u!\n\n",
1877 stats->total_aux_lost,
1878 stats->nr_events[PERF_RECORD_AUX]);
1879 }
1880
1881 if (session->tool->aux == perf_event__process_aux &&
1882 stats->total_aux_partial != 0) {
1883 bool vmm_exclusive = false;
1884
1885 (void)sysfs__read_bool("module/kvm_intel/parameters/vmm_exclusive",
1886 &vmm_exclusive);
1887
1888 ui__warning("AUX data had gaps in it %" PRIu64 " times out of %u!\n\n"
1889 "Are you running a KVM guest in the background?%s\n\n",
1890 stats->total_aux_partial,
1891 stats->nr_events[PERF_RECORD_AUX],
1892 vmm_exclusive ?
1893 "\nReloading kvm_intel module with vmm_exclusive=0\n"
1894 "will reduce the gaps to only guest's timeslices." :
1895 "");
1896 }
1897
1898 if (stats->nr_unknown_events != 0) {
1899 ui__warning("Found %u unknown events!\n\n"
1900 "Is this an older tool processing a perf.data "
1901 "file generated by a more recent tool?\n\n"
1902 "If that is not the case, consider "
1903 "reporting to linux-kernel@vger.kernel.org.\n\n",
1904 stats->nr_unknown_events);
1905 }
1906
1907 if (stats->nr_unknown_id != 0) {
1908 ui__warning("%u samples with id not present in the header\n",
1909 stats->nr_unknown_id);
1910 }
1911
1912 if (stats->nr_invalid_chains != 0) {
1913 ui__warning("Found invalid callchains!\n\n"
1914 "%u out of %u events were discarded for this reason.\n\n"
1915 "Consider reporting to linux-kernel@vger.kernel.org.\n\n",
1916 stats->nr_invalid_chains,
1917 stats->nr_events[PERF_RECORD_SAMPLE]);
1918 }
1919
1920 if (stats->nr_unprocessable_samples != 0) {
1921 ui__warning("%u unprocessable samples recorded.\n"
1922 "Do you have a KVM guest running and not using 'perf kvm'?\n",
1923 stats->nr_unprocessable_samples);
1924 }
1925
1926 perf_session__warn_order(session);
1927
1928 events_stats__auxtrace_error_warn(stats);
1929
1930 if (stats->nr_proc_map_timeout != 0) {
1931 ui__warning("%d map information files for pre-existing threads were\n"
1932 "not processed, if there are samples for addresses they\n"
1933 "will not be resolved, you may find out which are these\n"
1934 "threads by running with -v and redirecting the output\n"
1935 "to a file.\n"
1936 "The time limit to process proc map is too short?\n"
1937 "Increase it by --proc-map-timeout\n",
1938 stats->nr_proc_map_timeout);
1939 }
1940}
1941
1942static int perf_session__flush_thread_stack(struct thread *thread,
1943 void *p __maybe_unused)
1944{
1945 return thread_stack__flush(thread);
1946}
1947
1948static int perf_session__flush_thread_stacks(struct perf_session *session)
1949{
1950 return machines__for_each_thread(&session->machines,
1951 perf_session__flush_thread_stack,
1952 NULL);
1953}
1954
1955volatile int session_done;
1956
1957static int __perf_session__process_decomp_events(struct perf_session *session);
1958
1959static int __perf_session__process_pipe_events(struct perf_session *session)
1960{
1961 struct ordered_events *oe = &session->ordered_events;
1962 struct perf_tool *tool = session->tool;
1963 union perf_event *event;
1964 uint32_t size, cur_size = 0;
1965 void *buf = NULL;
1966 s64 skip = 0;
1967 u64 head;
1968 ssize_t err;
1969 void *p;
1970
1971 perf_tool__fill_defaults(tool);
1972
1973 head = 0;
1974 cur_size = sizeof(union perf_event);
1975
1976 buf = malloc(cur_size);
1977 if (!buf)
1978 return -errno;
1979 ordered_events__set_copy_on_queue(oe, true);
1980more:
1981 event = buf;
1982 err = perf_data__read(session->data, event,
1983 sizeof(struct perf_event_header));
1984 if (err <= 0) {
1985 if (err == 0)
1986 goto done;
1987
1988 pr_err("failed to read event header\n");
1989 goto out_err;
1990 }
1991
1992 if (session->header.needs_swap)
1993 perf_event_header__bswap(&event->header);
1994
1995 size = event->header.size;
1996 if (size < sizeof(struct perf_event_header)) {
1997 pr_err("bad event header size\n");
1998 goto out_err;
1999 }
2000
2001 if (size > cur_size) {
2002 void *new = realloc(buf, size);
2003 if (!new) {
2004 pr_err("failed to allocate memory to read event\n");
2005 goto out_err;
2006 }
2007 buf = new;
2008 cur_size = size;
2009 event = buf;
2010 }
2011 p = event;
2012 p += sizeof(struct perf_event_header);
2013
2014 if (size - sizeof(struct perf_event_header)) {
2015 err = perf_data__read(session->data, p,
2016 size - sizeof(struct perf_event_header));
2017 if (err <= 0) {
2018 if (err == 0) {
2019 pr_err("unexpected end of event stream\n");
2020 goto done;
2021 }
2022
2023 pr_err("failed to read event data\n");
2024 goto out_err;
2025 }
2026 }
2027
2028 if ((skip = perf_session__process_event(session, event, head)) < 0) {
2029 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
2030 head, event->header.size, event->header.type);
2031 err = -EINVAL;
2032 goto out_err;
2033 }
2034
2035 head += size;
2036
2037 if (skip > 0)
2038 head += skip;
2039
2040 err = __perf_session__process_decomp_events(session);
2041 if (err)
2042 goto out_err;
2043
2044 if (!session_done())
2045 goto more;
2046done:
2047 /* do the final flush for ordered samples */
2048 err = ordered_events__flush(oe, OE_FLUSH__FINAL);
2049 if (err)
2050 goto out_err;
2051 err = auxtrace__flush_events(session, tool);
2052 if (err)
2053 goto out_err;
2054 err = perf_session__flush_thread_stacks(session);
2055out_err:
2056 free(buf);
2057 if (!tool->no_warn)
2058 perf_session__warn_about_errors(session);
2059 ordered_events__free(&session->ordered_events);
2060 auxtrace__free_events(session);
2061 return err;
2062}
2063
2064static union perf_event *
2065prefetch_event(char *buf, u64 head, size_t mmap_size,
2066 bool needs_swap, union perf_event *error)
2067{
2068 union perf_event *event;
2069
2070 /*
2071 * Ensure we have enough space remaining to read
2072 * the size of the event in the headers.
2073 */
2074 if (head + sizeof(event->header) > mmap_size)
2075 return NULL;
2076
2077 event = (union perf_event *)(buf + head);
2078 if (needs_swap)
2079 perf_event_header__bswap(&event->header);
2080
2081 if (head + event->header.size <= mmap_size)
2082 return event;
2083
2084 /* We're not fetching the event so swap back again */
2085 if (needs_swap)
2086 perf_event_header__bswap(&event->header);
2087
2088 pr_debug("%s: head=%#" PRIx64 " event->header_size=%#x, mmap_size=%#zx:"
2089 " fuzzed or compressed perf.data?\n",__func__, head, event->header.size, mmap_size);
2090
2091 return error;
2092}
2093
2094static union perf_event *
2095fetch_mmaped_event(u64 head, size_t mmap_size, char *buf, bool needs_swap)
2096{
2097 return prefetch_event(buf, head, mmap_size, needs_swap, ERR_PTR(-EINVAL));
2098}
2099
2100static union perf_event *
2101fetch_decomp_event(u64 head, size_t mmap_size, char *buf, bool needs_swap)
2102{
2103 return prefetch_event(buf, head, mmap_size, needs_swap, NULL);
2104}
2105
2106static int __perf_session__process_decomp_events(struct perf_session *session)
2107{
2108 s64 skip;
2109 u64 size, file_pos = 0;
2110 struct decomp *decomp = session->decomp_last;
2111
2112 if (!decomp)
2113 return 0;
2114
2115 while (decomp->head < decomp->size && !session_done()) {
2116 union perf_event *event = fetch_decomp_event(decomp->head, decomp->size, decomp->data,
2117 session->header.needs_swap);
2118
2119 if (!event)
2120 break;
2121
2122 size = event->header.size;
2123
2124 if (size < sizeof(struct perf_event_header) ||
2125 (skip = perf_session__process_event(session, event, file_pos)) < 0) {
2126 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
2127 decomp->file_pos + decomp->head, event->header.size, event->header.type);
2128 return -EINVAL;
2129 }
2130
2131 if (skip)
2132 size += skip;
2133
2134 decomp->head += size;
2135 }
2136
2137 return 0;
2138}
2139
2140/*
2141 * On 64bit we can mmap the data file in one go. No need for tiny mmap
2142 * slices. On 32bit we use 32MB.
2143 */
2144#if BITS_PER_LONG == 64
2145#define MMAP_SIZE ULLONG_MAX
2146#define NUM_MMAPS 1
2147#else
2148#define MMAP_SIZE (32 * 1024 * 1024ULL)
2149#define NUM_MMAPS 128
2150#endif
2151
2152struct reader;
2153
2154typedef s64 (*reader_cb_t)(struct perf_session *session,
2155 union perf_event *event,
2156 u64 file_offset);
2157
2158struct reader {
2159 int fd;
2160 u64 data_size;
2161 u64 data_offset;
2162 reader_cb_t process;
2163 bool in_place_update;
2164};
2165
2166static int
2167reader__process_events(struct reader *rd, struct perf_session *session,
2168 struct ui_progress *prog)
2169{
2170 u64 data_size = rd->data_size;
2171 u64 head, page_offset, file_offset, file_pos, size;
2172 int err = 0, mmap_prot, mmap_flags, map_idx = 0;
2173 size_t mmap_size;
2174 char *buf, *mmaps[NUM_MMAPS];
2175 union perf_event *event;
2176 s64 skip;
2177
2178 page_offset = page_size * (rd->data_offset / page_size);
2179 file_offset = page_offset;
2180 head = rd->data_offset - page_offset;
2181
2182 ui_progress__init_size(prog, data_size, "Processing events...");
2183
2184 data_size += rd->data_offset;
2185
2186 mmap_size = MMAP_SIZE;
2187 if (mmap_size > data_size) {
2188 mmap_size = data_size;
2189 session->one_mmap = true;
2190 }
2191
2192 memset(mmaps, 0, sizeof(mmaps));
2193
2194 mmap_prot = PROT_READ;
2195 mmap_flags = MAP_SHARED;
2196
2197 if (rd->in_place_update) {
2198 mmap_prot |= PROT_WRITE;
2199 } else if (session->header.needs_swap) {
2200 mmap_prot |= PROT_WRITE;
2201 mmap_flags = MAP_PRIVATE;
2202 }
2203remap:
2204 buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, rd->fd,
2205 file_offset);
2206 if (buf == MAP_FAILED) {
2207 pr_err("failed to mmap file\n");
2208 err = -errno;
2209 goto out;
2210 }
2211 mmaps[map_idx] = buf;
2212 map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1);
2213 file_pos = file_offset + head;
2214 if (session->one_mmap) {
2215 session->one_mmap_addr = buf;
2216 session->one_mmap_offset = file_offset;
2217 }
2218
2219more:
2220 event = fetch_mmaped_event(head, mmap_size, buf, session->header.needs_swap);
2221 if (IS_ERR(event))
2222 return PTR_ERR(event);
2223
2224 if (!event) {
2225 if (mmaps[map_idx]) {
2226 munmap(mmaps[map_idx], mmap_size);
2227 mmaps[map_idx] = NULL;
2228 }
2229
2230 page_offset = page_size * (head / page_size);
2231 file_offset += page_offset;
2232 head -= page_offset;
2233 goto remap;
2234 }
2235
2236 size = event->header.size;
2237
2238 skip = -EINVAL;
2239
2240 if (size < sizeof(struct perf_event_header) ||
2241 (skip = rd->process(session, event, file_pos)) < 0) {
2242 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d [%s]\n",
2243 file_offset + head, event->header.size,
2244 event->header.type, strerror(-skip));
2245 err = skip;
2246 goto out;
2247 }
2248
2249 if (skip)
2250 size += skip;
2251
2252 head += size;
2253 file_pos += size;
2254
2255 err = __perf_session__process_decomp_events(session);
2256 if (err)
2257 goto out;
2258
2259 ui_progress__update(prog, size);
2260
2261 if (session_done())
2262 goto out;
2263
2264 if (file_pos < data_size)
2265 goto more;
2266
2267out:
2268 return err;
2269}
2270
2271static s64 process_simple(struct perf_session *session,
2272 union perf_event *event,
2273 u64 file_offset)
2274{
2275 return perf_session__process_event(session, event, file_offset);
2276}
2277
2278static int __perf_session__process_events(struct perf_session *session)
2279{
2280 struct reader rd = {
2281 .fd = perf_data__fd(session->data),
2282 .data_size = session->header.data_size,
2283 .data_offset = session->header.data_offset,
2284 .process = process_simple,
2285 .in_place_update = session->data->in_place_update,
2286 };
2287 struct ordered_events *oe = &session->ordered_events;
2288 struct perf_tool *tool = session->tool;
2289 struct ui_progress prog;
2290 int err;
2291
2292 perf_tool__fill_defaults(tool);
2293
2294 if (rd.data_size == 0)
2295 return -1;
2296
2297 ui_progress__init_size(&prog, rd.data_size, "Processing events...");
2298
2299 err = reader__process_events(&rd, session, &prog);
2300 if (err)
2301 goto out_err;
2302 /* do the final flush for ordered samples */
2303 err = ordered_events__flush(oe, OE_FLUSH__FINAL);
2304 if (err)
2305 goto out_err;
2306 err = auxtrace__flush_events(session, tool);
2307 if (err)
2308 goto out_err;
2309 err = perf_session__flush_thread_stacks(session);
2310out_err:
2311 ui_progress__finish();
2312 if (!tool->no_warn)
2313 perf_session__warn_about_errors(session);
2314 /*
2315 * We may switching perf.data output, make ordered_events
2316 * reusable.
2317 */
2318 ordered_events__reinit(&session->ordered_events);
2319 auxtrace__free_events(session);
2320 session->one_mmap = false;
2321 return err;
2322}
2323
2324int perf_session__process_events(struct perf_session *session)
2325{
2326 if (perf_session__register_idle_thread(session) < 0)
2327 return -ENOMEM;
2328
2329 if (perf_data__is_pipe(session->data))
2330 return __perf_session__process_pipe_events(session);
2331
2332 return __perf_session__process_events(session);
2333}
2334
2335bool perf_session__has_traces(struct perf_session *session, const char *msg)
2336{
2337 struct evsel *evsel;
2338
2339 evlist__for_each_entry(session->evlist, evsel) {
2340 if (evsel->core.attr.type == PERF_TYPE_TRACEPOINT)
2341 return true;
2342 }
2343
2344 pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg);
2345 return false;
2346}
2347
2348int map__set_kallsyms_ref_reloc_sym(struct map *map, const char *symbol_name, u64 addr)
2349{
2350 char *bracket;
2351 struct ref_reloc_sym *ref;
2352 struct kmap *kmap;
2353
2354 ref = zalloc(sizeof(struct ref_reloc_sym));
2355 if (ref == NULL)
2356 return -ENOMEM;
2357
2358 ref->name = strdup(symbol_name);
2359 if (ref->name == NULL) {
2360 free(ref);
2361 return -ENOMEM;
2362 }
2363
2364 bracket = strchr(ref->name, ']');
2365 if (bracket)
2366 *bracket = '\0';
2367
2368 ref->addr = addr;
2369
2370 kmap = map__kmap(map);
2371 if (kmap)
2372 kmap->ref_reloc_sym = ref;
2373
2374 return 0;
2375}
2376
2377size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp)
2378{
2379 return machines__fprintf_dsos(&session->machines, fp);
2380}
2381
2382size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp,
2383 bool (skip)(struct dso *dso, int parm), int parm)
2384{
2385 return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm);
2386}
2387
2388size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp,
2389 bool skip_empty)
2390{
2391 size_t ret;
2392 const char *msg = "";
2393
2394 if (perf_header__has_feat(&session->header, HEADER_AUXTRACE))
2395 msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)";
2396
2397 ret = fprintf(fp, "\nAggregated stats:%s\n", msg);
2398
2399 ret += events_stats__fprintf(&session->evlist->stats, fp, skip_empty);
2400 return ret;
2401}
2402
2403size_t perf_session__fprintf(struct perf_session *session, FILE *fp)
2404{
2405 /*
2406 * FIXME: Here we have to actually print all the machines in this
2407 * session, not just the host...
2408 */
2409 return machine__fprintf(&session->machines.host, fp);
2410}
2411
2412struct evsel *perf_session__find_first_evtype(struct perf_session *session,
2413 unsigned int type)
2414{
2415 struct evsel *pos;
2416
2417 evlist__for_each_entry(session->evlist, pos) {
2418 if (pos->core.attr.type == type)
2419 return pos;
2420 }
2421 return NULL;
2422}
2423
2424int perf_session__cpu_bitmap(struct perf_session *session,
2425 const char *cpu_list, unsigned long *cpu_bitmap)
2426{
2427 int i, err = -1;
2428 struct perf_cpu_map *map;
2429 int nr_cpus = min(session->header.env.nr_cpus_avail, MAX_NR_CPUS);
2430
2431 for (i = 0; i < PERF_TYPE_MAX; ++i) {
2432 struct evsel *evsel;
2433
2434 evsel = perf_session__find_first_evtype(session, i);
2435 if (!evsel)
2436 continue;
2437
2438 if (!(evsel->core.attr.sample_type & PERF_SAMPLE_CPU)) {
2439 pr_err("File does not contain CPU events. "
2440 "Remove -C option to proceed.\n");
2441 return -1;
2442 }
2443 }
2444
2445 map = perf_cpu_map__new(cpu_list);
2446 if (map == NULL) {
2447 pr_err("Invalid cpu_list\n");
2448 return -1;
2449 }
2450
2451 for (i = 0; i < map->nr; i++) {
2452 int cpu = map->map[i];
2453
2454 if (cpu >= nr_cpus) {
2455 pr_err("Requested CPU %d too large. "
2456 "Consider raising MAX_NR_CPUS\n", cpu);
2457 goto out_delete_map;
2458 }
2459
2460 set_bit(cpu, cpu_bitmap);
2461 }
2462
2463 err = 0;
2464
2465out_delete_map:
2466 perf_cpu_map__put(map);
2467 return err;
2468}
2469
2470void perf_session__fprintf_info(struct perf_session *session, FILE *fp,
2471 bool full)
2472{
2473 if (session == NULL || fp == NULL)
2474 return;
2475
2476 fprintf(fp, "# ========\n");
2477 perf_header__fprintf_info(session, fp, full);
2478 fprintf(fp, "# ========\n#\n");
2479}
2480
2481int perf_event__process_id_index(struct perf_session *session,
2482 union perf_event *event)
2483{
2484 struct evlist *evlist = session->evlist;
2485 struct perf_record_id_index *ie = &event->id_index;
2486 size_t i, nr, max_nr;
2487
2488 max_nr = (ie->header.size - sizeof(struct perf_record_id_index)) /
2489 sizeof(struct id_index_entry);
2490 nr = ie->nr;
2491 if (nr > max_nr)
2492 return -EINVAL;
2493
2494 if (dump_trace)
2495 fprintf(stdout, " nr: %zu\n", nr);
2496
2497 for (i = 0; i < nr; i++) {
2498 struct id_index_entry *e = &ie->entries[i];
2499 struct perf_sample_id *sid;
2500
2501 if (dump_trace) {
2502 fprintf(stdout, " ... id: %"PRI_lu64, e->id);
2503 fprintf(stdout, " idx: %"PRI_lu64, e->idx);
2504 fprintf(stdout, " cpu: %"PRI_ld64, e->cpu);
2505 fprintf(stdout, " tid: %"PRI_ld64"\n", e->tid);
2506 }
2507
2508 sid = evlist__id2sid(evlist, e->id);
2509 if (!sid)
2510 return -ENOENT;
2511 sid->idx = e->idx;
2512 sid->cpu = e->cpu;
2513 sid->tid = e->tid;
2514 }
2515 return 0;
2516}