]> git.proxmox.com Git - mirror_ubuntu-hirsute-kernel.git/blob - tools/perf/util/session.c
projects / mirror_ubuntu-hirsute-kernel.git / blob
? search:


052181f9c1cbad78090b7d84b38abf8edd6ab3e4
[mirror_ubuntu-hirsute-kernel.git] / tools / perf / util / session.c
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 "ui/progress.h"
33 #include "../perf.h"
34 #include "arch/common.h"
35 #include "units.h"
36 #include <internal/lib.h>
37
38 #ifdef HAVE_ZSTD_SUPPORT
39 static int perf_session__process_compressed_event(struct perf_session *session,
40 union perf_event *event, u64 file_offset)
41 {
42 void *src;
43 size_t decomp_size, src_size;
44 u64 decomp_last_rem = 0;
45 size_t mmap_len, decomp_len = session->header.env.comp_mmap_len;
46 struct decomp *decomp, *decomp_last = session->decomp_last;
47
48 if (decomp_last) {
49 decomp_last_rem = decomp_last->size - decomp_last->head;
50 decomp_len += decomp_last_rem;
51 }
52
53 mmap_len = sizeof(struct decomp) + decomp_len;
54 decomp = mmap(NULL, mmap_len, PROT_READ|PROT_WRITE,
55 MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);
56 if (decomp == MAP_FAILED) {
57 pr_err("Couldn't allocate memory for decompression\n");
58 return -1;
59 }
60
61 decomp->file_pos = file_offset;
62 decomp->mmap_len = mmap_len;
63 decomp->head = 0;
64
65 if (decomp_last_rem) {
66 memcpy(decomp->data, &(decomp_last->data[decomp_last->head]), decomp_last_rem);
67 decomp->size = decomp_last_rem;
68 }
69
70 src = (void *)event + sizeof(struct perf_record_compressed);
71 src_size = event->pack.header.size - sizeof(struct perf_record_compressed);
72
73 decomp_size = zstd_decompress_stream(&(session->zstd_data), src, src_size,
74 &(decomp->data[decomp_last_rem]), decomp_len - decomp_last_rem);
75 if (!decomp_size) {
76 munmap(decomp, mmap_len);
77 pr_err("Couldn't decompress data\n");
78 return -1;
79 }
80
81 decomp->size += decomp_size;
82
83 if (session->decomp == NULL) {
84 session->decomp = decomp;
85 session->decomp_last = decomp;
86 } else {
87 session->decomp_last->next = decomp;
88 session->decomp_last = decomp;
89 }
90
91 pr_debug("decomp (B): %zd to %zd\n", src_size, decomp_size);
92
93 return 0;
94 }
95 #else /* !HAVE_ZSTD_SUPPORT */
96 #define perf_session__process_compressed_event perf_session__process_compressed_event_stub
97 #endif
98
99 static int perf_session__deliver_event(struct perf_session *session,
100 union perf_event *event,
101 struct perf_tool *tool,
102 u64 file_offset);
103
104 static int perf_session__open(struct perf_session *session)
105 {
106 struct perf_data *data = session->data;
107
108 if (perf_session__read_header(session) < 0) {
109 pr_err("incompatible file format (rerun with -v to learn more)\n");
110 return -1;
111 }
112
113 if (perf_data__is_pipe(data))
114 return 0;
115
116 if (perf_header__has_feat(&session->header, HEADER_STAT))
117 return 0;
118
119 if (!evlist__valid_sample_type(session->evlist)) {
120 pr_err("non matching sample_type\n");
121 return -1;
122 }
123
124 if (!evlist__valid_sample_id_all(session->evlist)) {
125 pr_err("non matching sample_id_all\n");
126 return -1;
127 }
128
129 if (!evlist__valid_read_format(session->evlist)) {
130 pr_err("non matching read_format\n");
131 return -1;
132 }
133
134 return 0;
135 }
136
137 void perf_session__set_id_hdr_size(struct perf_session *session)
138 {
139 u16 id_hdr_size = evlist__id_hdr_size(session->evlist);
140
141 machines__set_id_hdr_size(&session->machines, id_hdr_size);
142 }
143
144 int perf_session__create_kernel_maps(struct perf_session *session)
145 {
146 int ret = machine__create_kernel_maps(&session->machines.host);
147
148 if (ret >= 0)
149 ret = machines__create_guest_kernel_maps(&session->machines);
150 return ret;
151 }
152
153 static void perf_session__destroy_kernel_maps(struct perf_session *session)
154 {
155 machines__destroy_kernel_maps(&session->machines);
156 }
157
158 static bool perf_session__has_comm_exec(struct perf_session *session)
159 {
160 struct evsel *evsel;
161
162 evlist__for_each_entry(session->evlist, evsel) {
163 if (evsel->core.attr.comm_exec)
164 return true;
165 }
166
167 return false;
168 }
169
170 static void perf_session__set_comm_exec(struct perf_session *session)
171 {
172 bool comm_exec = perf_session__has_comm_exec(session);
173
174 machines__set_comm_exec(&session->machines, comm_exec);
175 }
176
177 static int ordered_events__deliver_event(struct ordered_events *oe,
178 struct ordered_event *event)
179 {
180 struct perf_session *session = container_of(oe, struct perf_session,
181 ordered_events);
182
183 return perf_session__deliver_event(session, event->event,
184 session->tool, event->file_offset);
185 }
186
187 struct perf_session *perf_session__new(struct perf_data *data,
188 bool repipe, struct perf_tool *tool)
189 {
190 int ret = -ENOMEM;
191 struct perf_session *session = zalloc(sizeof(*session));
192
193 if (!session)
194 goto out;
195
196 session->repipe = repipe;
197 session->tool = tool;
198 INIT_LIST_HEAD(&session->auxtrace_index);
199 machines__init(&session->machines);
200 ordered_events__init(&session->ordered_events,
201 ordered_events__deliver_event, NULL);
202
203 perf_env__init(&session->header.env);
204 if (data) {
205 ret = perf_data__open(data);
206 if (ret < 0)
207 goto out_delete;
208
209 session->data = data;
210
211 if (perf_data__is_read(data)) {
212 ret = perf_session__open(session);
213 if (ret < 0)
214 goto out_delete;
215
216 /*
217 * set session attributes that are present in perf.data
218 * but not in pipe-mode.
219 */
220 if (!data->is_pipe) {
221 perf_session__set_id_hdr_size(session);
222 perf_session__set_comm_exec(session);
223 }
224
225 evlist__init_trace_event_sample_raw(session->evlist);
226
227 /* Open the directory data. */
228 if (data->is_dir) {
229 ret = perf_data__open_dir(data);
230 if (ret)
231 goto out_delete;
232 }
233
234 if (!symbol_conf.kallsyms_name &&
235 !symbol_conf.vmlinux_name)
236 symbol_conf.kallsyms_name = perf_data__kallsyms_name(data);
237 }
238 } else {
239 session->machines.host.env = &perf_env;
240 }
241
242 session->machines.host.single_address_space =
243 perf_env__single_address_space(session->machines.host.env);
244
245 if (!data || perf_data__is_write(data)) {
246 /*
247 * In O_RDONLY mode this will be performed when reading the
248 * kernel MMAP event, in perf_event__process_mmap().
249 */
250 if (perf_session__create_kernel_maps(session) < 0)
251 pr_warning("Cannot read kernel map\n");
252 }
253
254 /*
255 * In pipe-mode, evlist is empty until PERF_RECORD_HEADER_ATTR is
256 * processed, so evlist__sample_id_all is not meaningful here.
257 */
258 if ((!data || !data->is_pipe) && tool && tool->ordering_requires_timestamps &&
259 tool->ordered_events && !evlist__sample_id_all(session->evlist)) {
260 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n");
261 tool->ordered_events = false;
262 }
263
264 return session;
265
266 out_delete:
267 perf_session__delete(session);
268 out:
269 return ERR_PTR(ret);
270 }
271
272 static void perf_session__delete_threads(struct perf_session *session)
273 {
274 machine__delete_threads(&session->machines.host);
275 }
276
277 static void perf_session__release_decomp_events(struct perf_session *session)
278 {
279 struct decomp *next, *decomp;
280 size_t mmap_len;
281 next = session->decomp;
282 do {
283 decomp = next;
284 if (decomp == NULL)
285 break;
286 next = decomp->next;
287 mmap_len = decomp->mmap_len;
288 munmap(decomp, mmap_len);
289 } while (1);
290 }
291
292 void perf_session__delete(struct perf_session *session)
293 {
294 if (session == NULL)
295 return;
296 auxtrace__free(session);
297 auxtrace_index__free(&session->auxtrace_index);
298 perf_session__destroy_kernel_maps(session);
299 perf_session__delete_threads(session);
300 perf_session__release_decomp_events(session);
301 perf_env__exit(&session->header.env);
302 machines__exit(&session->machines);
303 if (session->data)
304 perf_data__close(session->data);
305 free(session);
306 }
307
308 static int process_event_synth_tracing_data_stub(struct perf_session *session
309 __maybe_unused,
310 union perf_event *event
311 __maybe_unused)
312 {
313 dump_printf(": unhandled!\n");
314 return 0;
315 }
316
317 static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused,
318 union perf_event *event __maybe_unused,
319 struct evlist **pevlist
320 __maybe_unused)
321 {
322 dump_printf(": unhandled!\n");
323 return 0;
324 }
325
326 static int process_event_synth_event_update_stub(struct perf_tool *tool __maybe_unused,
327 union perf_event *event __maybe_unused,
328 struct evlist **pevlist
329 __maybe_unused)
330 {
331 if (dump_trace)
332 perf_event__fprintf_event_update(event, stdout);
333
334 dump_printf(": unhandled!\n");
335 return 0;
336 }
337
338 static int process_event_sample_stub(struct perf_tool *tool __maybe_unused,
339 union perf_event *event __maybe_unused,
340 struct perf_sample *sample __maybe_unused,
341 struct evsel *evsel __maybe_unused,
342 struct machine *machine __maybe_unused)
343 {
344 dump_printf(": unhandled!\n");
345 return 0;
346 }
347
348 static int process_event_stub(struct perf_tool *tool __maybe_unused,
349 union perf_event *event __maybe_unused,
350 struct perf_sample *sample __maybe_unused,
351 struct machine *machine __maybe_unused)
352 {
353 dump_printf(": unhandled!\n");
354 return 0;
355 }
356
357 static int process_finished_round_stub(struct perf_tool *tool __maybe_unused,
358 union perf_event *event __maybe_unused,
359 struct ordered_events *oe __maybe_unused)
360 {
361 dump_printf(": unhandled!\n");
362 return 0;
363 }
364
365 static int process_finished_round(struct perf_tool *tool,
366 union perf_event *event,
367 struct ordered_events *oe);
368
369 static int skipn(int fd, off_t n)
370 {
371 char buf[4096];
372 ssize_t ret;
373
374 while (n > 0) {
375 ret = read(fd, buf, min(n, (off_t)sizeof(buf)));
376 if (ret <= 0)
377 return ret;
378 n -= ret;
379 }
380
381 return 0;
382 }
383
384 static s64 process_event_auxtrace_stub(struct perf_session *session __maybe_unused,
385 union perf_event *event)
386 {
387 dump_printf(": unhandled!\n");
388 if (perf_data__is_pipe(session->data))
389 skipn(perf_data__fd(session->data), event->auxtrace.size);
390 return event->auxtrace.size;
391 }
392
393 static int process_event_op2_stub(struct perf_session *session __maybe_unused,
394 union perf_event *event __maybe_unused)
395 {
396 dump_printf(": unhandled!\n");
397 return 0;
398 }
399
400
401 static
402 int process_event_thread_map_stub(struct perf_session *session __maybe_unused,
403 union perf_event *event __maybe_unused)
404 {
405 if (dump_trace)
406 perf_event__fprintf_thread_map(event, stdout);
407
408 dump_printf(": unhandled!\n");
409 return 0;
410 }
411
412 static
413 int process_event_cpu_map_stub(struct perf_session *session __maybe_unused,
414 union perf_event *event __maybe_unused)
415 {
416 if (dump_trace)
417 perf_event__fprintf_cpu_map(event, stdout);
418
419 dump_printf(": unhandled!\n");
420 return 0;
421 }
422
423 static
424 int process_event_stat_config_stub(struct perf_session *session __maybe_unused,
425 union perf_event *event __maybe_unused)
426 {
427 if (dump_trace)
428 perf_event__fprintf_stat_config(event, stdout);
429
430 dump_printf(": unhandled!\n");
431 return 0;
432 }
433
434 static int process_stat_stub(struct perf_session *perf_session __maybe_unused,
435 union perf_event *event)
436 {
437 if (dump_trace)
438 perf_event__fprintf_stat(event, stdout);
439
440 dump_printf(": unhandled!\n");
441 return 0;
442 }
443
444 static int process_stat_round_stub(struct perf_session *perf_session __maybe_unused,
445 union perf_event *event)
446 {
447 if (dump_trace)
448 perf_event__fprintf_stat_round(event, stdout);
449
450 dump_printf(": unhandled!\n");
451 return 0;
452 }
453
454 static int perf_session__process_compressed_event_stub(struct perf_session *session __maybe_unused,
455 union perf_event *event __maybe_unused,
456 u64 file_offset __maybe_unused)
457 {
458 dump_printf(": unhandled!\n");
459 return 0;
460 }
461
462 void perf_tool__fill_defaults(struct perf_tool *tool)
463 {
464 if (tool->sample == NULL)
465 tool->sample = process_event_sample_stub;
466 if (tool->mmap == NULL)
467 tool->mmap = process_event_stub;
468 if (tool->mmap2 == NULL)
469 tool->mmap2 = process_event_stub;
470 if (tool->comm == NULL)
471 tool->comm = process_event_stub;
472 if (tool->namespaces == NULL)
473 tool->namespaces = process_event_stub;
474 if (tool->cgroup == NULL)
475 tool->cgroup = process_event_stub;
476 if (tool->fork == NULL)
477 tool->fork = process_event_stub;
478 if (tool->exit == NULL)
479 tool->exit = process_event_stub;
480 if (tool->lost == NULL)
481 tool->lost = perf_event__process_lost;
482 if (tool->lost_samples == NULL)
483 tool->lost_samples = perf_event__process_lost_samples;
484 if (tool->aux == NULL)
485 tool->aux = perf_event__process_aux;
486 if (tool->itrace_start == NULL)
487 tool->itrace_start = perf_event__process_itrace_start;
488 if (tool->context_switch == NULL)
489 tool->context_switch = perf_event__process_switch;
490 if (tool->ksymbol == NULL)
491 tool->ksymbol = perf_event__process_ksymbol;
492 if (tool->bpf == NULL)
493 tool->bpf = perf_event__process_bpf;
494 if (tool->text_poke == NULL)
495 tool->text_poke = perf_event__process_text_poke;
496 if (tool->read == NULL)
497 tool->read = process_event_sample_stub;
498 if (tool->throttle == NULL)
499 tool->throttle = process_event_stub;
500 if (tool->unthrottle == NULL)
501 tool->unthrottle = process_event_stub;
502 if (tool->attr == NULL)
503 tool->attr = process_event_synth_attr_stub;
504 if (tool->event_update == NULL)
505 tool->event_update = process_event_synth_event_update_stub;
506 if (tool->tracing_data == NULL)
507 tool->tracing_data = process_event_synth_tracing_data_stub;
508 if (tool->build_id == NULL)
509 tool->build_id = process_event_op2_stub;
510 if (tool->finished_round == NULL) {
511 if (tool->ordered_events)
512 tool->finished_round = process_finished_round;
513 else
514 tool->finished_round = process_finished_round_stub;
515 }
516 if (tool->id_index == NULL)
517 tool->id_index = process_event_op2_stub;
518 if (tool->auxtrace_info == NULL)
519 tool->auxtrace_info = process_event_op2_stub;
520 if (tool->auxtrace == NULL)
521 tool->auxtrace = process_event_auxtrace_stub;
522 if (tool->auxtrace_error == NULL)
523 tool->auxtrace_error = process_event_op2_stub;
524 if (tool->thread_map == NULL)
525 tool->thread_map = process_event_thread_map_stub;
526 if (tool->cpu_map == NULL)
527 tool->cpu_map = process_event_cpu_map_stub;
528 if (tool->stat_config == NULL)
529 tool->stat_config = process_event_stat_config_stub;
530 if (tool->stat == NULL)
531 tool->stat = process_stat_stub;
532 if (tool->stat_round == NULL)
533 tool->stat_round = process_stat_round_stub;
534 if (tool->time_conv == NULL)
535 tool->time_conv = process_event_op2_stub;
536 if (tool->feature == NULL)
537 tool->feature = process_event_op2_stub;
538 if (tool->compressed == NULL)
539 tool->compressed = perf_session__process_compressed_event;
540 }
541
542 static void swap_sample_id_all(union perf_event *event, void *data)
543 {
544 void *end = (void *) event + event->header.size;
545 int size = end - data;
546
547 BUG_ON(size % sizeof(u64));
548 mem_bswap_64(data, size);
549 }
550
551 static void perf_event__all64_swap(union perf_event *event,
552 bool sample_id_all __maybe_unused)
553 {
554 struct perf_event_header *hdr = &event->header;
555 mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr));
556 }
557
558 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all)
559 {
560 event->comm.pid = bswap_32(event->comm.pid);
561 event->comm.tid = bswap_32(event->comm.tid);
562
563 if (sample_id_all) {
564 void *data = &event->comm.comm;
565
566 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
567 swap_sample_id_all(event, data);
568 }
569 }
570
571 static void perf_event__mmap_swap(union perf_event *event,
572 bool sample_id_all)
573 {
574 event->mmap.pid = bswap_32(event->mmap.pid);
575 event->mmap.tid = bswap_32(event->mmap.tid);
576 event->mmap.start = bswap_64(event->mmap.start);
577 event->mmap.len = bswap_64(event->mmap.len);
578 event->mmap.pgoff = bswap_64(event->mmap.pgoff);
579
580 if (sample_id_all) {
581 void *data = &event->mmap.filename;
582
583 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
584 swap_sample_id_all(event, data);
585 }
586 }
587
588 static void perf_event__mmap2_swap(union perf_event *event,
589 bool sample_id_all)
590 {
591 event->mmap2.pid = bswap_32(event->mmap2.pid);
592 event->mmap2.tid = bswap_32(event->mmap2.tid);
593 event->mmap2.start = bswap_64(event->mmap2.start);
594 event->mmap2.len = bswap_64(event->mmap2.len);
595 event->mmap2.pgoff = bswap_64(event->mmap2.pgoff);
596 event->mmap2.maj = bswap_32(event->mmap2.maj);
597 event->mmap2.min = bswap_32(event->mmap2.min);
598 event->mmap2.ino = bswap_64(event->mmap2.ino);
599 event->mmap2.ino_generation = bswap_64(event->mmap2.ino_generation);
600
601 if (sample_id_all) {
602 void *data = &event->mmap2.filename;
603
604 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
605 swap_sample_id_all(event, data);
606 }
607 }
608 static void perf_event__task_swap(union perf_event *event, bool sample_id_all)
609 {
610 event->fork.pid = bswap_32(event->fork.pid);
611 event->fork.tid = bswap_32(event->fork.tid);
612 event->fork.ppid = bswap_32(event->fork.ppid);
613 event->fork.ptid = bswap_32(event->fork.ptid);
614 event->fork.time = bswap_64(event->fork.time);
615
616 if (sample_id_all)
617 swap_sample_id_all(event, &event->fork + 1);
618 }
619
620 static void perf_event__read_swap(union perf_event *event, bool sample_id_all)
621 {
622 event->read.pid = bswap_32(event->read.pid);
623 event->read.tid = bswap_32(event->read.tid);
624 event->read.value = bswap_64(event->read.value);
625 event->read.time_enabled = bswap_64(event->read.time_enabled);
626 event->read.time_running = bswap_64(event->read.time_running);
627 event->read.id = bswap_64(event->read.id);
628
629 if (sample_id_all)
630 swap_sample_id_all(event, &event->read + 1);
631 }
632
633 static void perf_event__aux_swap(union perf_event *event, bool sample_id_all)
634 {
635 event->aux.aux_offset = bswap_64(event->aux.aux_offset);
636 event->aux.aux_size = bswap_64(event->aux.aux_size);
637 event->aux.flags = bswap_64(event->aux.flags);
638
639 if (sample_id_all)
640 swap_sample_id_all(event, &event->aux + 1);
641 }
642
643 static void perf_event__itrace_start_swap(union perf_event *event,
644 bool sample_id_all)
645 {
646 event->itrace_start.pid = bswap_32(event->itrace_start.pid);
647 event->itrace_start.tid = bswap_32(event->itrace_start.tid);
648
649 if (sample_id_all)
650 swap_sample_id_all(event, &event->itrace_start + 1);
651 }
652
653 static void perf_event__switch_swap(union perf_event *event, bool sample_id_all)
654 {
655 if (event->header.type == PERF_RECORD_SWITCH_CPU_WIDE) {
656 event->context_switch.next_prev_pid =
657 bswap_32(event->context_switch.next_prev_pid);
658 event->context_switch.next_prev_tid =
659 bswap_32(event->context_switch.next_prev_tid);
660 }
661
662 if (sample_id_all)
663 swap_sample_id_all(event, &event->context_switch + 1);
664 }
665
666 static void perf_event__text_poke_swap(union perf_event *event, bool sample_id_all)
667 {
668 event->text_poke.addr = bswap_64(event->text_poke.addr);
669 event->text_poke.old_len = bswap_16(event->text_poke.old_len);
670 event->text_poke.new_len = bswap_16(event->text_poke.new_len);
671
672 if (sample_id_all) {
673 size_t len = sizeof(event->text_poke.old_len) +
674 sizeof(event->text_poke.new_len) +
675 event->text_poke.old_len +
676 event->text_poke.new_len;
677 void *data = &event->text_poke.old_len;
678
679 data += PERF_ALIGN(len, sizeof(u64));
680 swap_sample_id_all(event, data);
681 }
682 }
683
684 static void perf_event__throttle_swap(union perf_event *event,
685 bool sample_id_all)
686 {
687 event->throttle.time = bswap_64(event->throttle.time);
688 event->throttle.id = bswap_64(event->throttle.id);
689 event->throttle.stream_id = bswap_64(event->throttle.stream_id);
690
691 if (sample_id_all)
692 swap_sample_id_all(event, &event->throttle + 1);
693 }
694
695 static void perf_event__namespaces_swap(union perf_event *event,
696 bool sample_id_all)
697 {
698 u64 i;
699
700 event->namespaces.pid = bswap_32(event->namespaces.pid);
701 event->namespaces.tid = bswap_32(event->namespaces.tid);
702 event->namespaces.nr_namespaces = bswap_64(event->namespaces.nr_namespaces);
703
704 for (i = 0; i < event->namespaces.nr_namespaces; i++) {
705 struct perf_ns_link_info *ns = &event->namespaces.link_info[i];
706
707 ns->dev = bswap_64(ns->dev);
708 ns->ino = bswap_64(ns->ino);
709 }
710
711 if (sample_id_all)
712 swap_sample_id_all(event, &event->namespaces.link_info[i]);
713 }
714
715 static void perf_event__cgroup_swap(union perf_event *event, bool sample_id_all)
716 {
717 event->cgroup.id = bswap_64(event->cgroup.id);
718
719 if (sample_id_all) {
720 void *data = &event->cgroup.path;
721
722 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
723 swap_sample_id_all(event, data);
724 }
725 }
726
727 static u8 revbyte(u8 b)
728 {
729 int rev = (b >> 4) | ((b & 0xf) << 4);
730 rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2);
731 rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1);
732 return (u8) rev;
733 }
734
735 /*
736 * XXX this is hack in attempt to carry flags bitfield
737 * through endian village. ABI says:
738 *
739 * Bit-fields are allocated from right to left (least to most significant)
740 * on little-endian implementations and from left to right (most to least
741 * significant) on big-endian implementations.
742 *
743 * The above seems to be byte specific, so we need to reverse each
744 * byte of the bitfield. 'Internet' also says this might be implementation
745 * specific and we probably need proper fix and carry perf_event_attr
746 * bitfield flags in separate data file FEAT_ section. Thought this seems
747 * to work for now.
748 */
749 static void swap_bitfield(u8 *p, unsigned len)
750 {
751 unsigned i;
752
753 for (i = 0; i < len; i++) {
754 *p = revbyte(*p);
755 p++;
756 }
757 }
758
759 /* exported for swapping attributes in file header */
760 void perf_event__attr_swap(struct perf_event_attr *attr)
761 {
762 attr->type = bswap_32(attr->type);
763 attr->size = bswap_32(attr->size);
764
765 #define bswap_safe(f, n) \
766 (attr->size > (offsetof(struct perf_event_attr, f) + \
767 sizeof(attr->f) * (n)))
768 #define bswap_field(f, sz) \
769 do { \
770 if (bswap_safe(f, 0)) \
771 attr->f = bswap_##sz(attr->f); \
772 } while(0)
773 #define bswap_field_16(f) bswap_field(f, 16)
774 #define bswap_field_32(f) bswap_field(f, 32)
775 #define bswap_field_64(f) bswap_field(f, 64)
776
777 bswap_field_64(config);
778 bswap_field_64(sample_period);
779 bswap_field_64(sample_type);
780 bswap_field_64(read_format);
781 bswap_field_32(wakeup_events);
782 bswap_field_32(bp_type);
783 bswap_field_64(bp_addr);
784 bswap_field_64(bp_len);
785 bswap_field_64(branch_sample_type);
786 bswap_field_64(sample_regs_user);
787 bswap_field_32(sample_stack_user);
788 bswap_field_32(aux_watermark);
789 bswap_field_16(sample_max_stack);
790 bswap_field_32(aux_sample_size);
791
792 /*
793 * After read_format are bitfields. Check read_format because
794 * we are unable to use offsetof on bitfield.
795 */
796 if (bswap_safe(read_format, 1))
797 swap_bitfield((u8 *) (&attr->read_format + 1),
798 sizeof(u64));
799 #undef bswap_field_64
800 #undef bswap_field_32
801 #undef bswap_field
802 #undef bswap_safe
803 }
804
805 static void perf_event__hdr_attr_swap(union perf_event *event,
806 bool sample_id_all __maybe_unused)
807 {
808 size_t size;
809
810 perf_event__attr_swap(&event->attr.attr);
811
812 size = event->header.size;
813 size -= (void *)&event->attr.id - (void *)event;
814 mem_bswap_64(event->attr.id, size);
815 }
816
817 static void perf_event__event_update_swap(union perf_event *event,
818 bool sample_id_all __maybe_unused)
819 {
820 event->event_update.type = bswap_64(event->event_update.type);
821 event->event_update.id = bswap_64(event->event_update.id);
822 }
823
824 static void perf_event__event_type_swap(union perf_event *event,
825 bool sample_id_all __maybe_unused)
826 {
827 event->event_type.event_type.event_id =
828 bswap_64(event->event_type.event_type.event_id);
829 }
830
831 static void perf_event__tracing_data_swap(union perf_event *event,
832 bool sample_id_all __maybe_unused)
833 {
834 event->tracing_data.size = bswap_32(event->tracing_data.size);
835 }
836
837 static void perf_event__auxtrace_info_swap(union perf_event *event,
838 bool sample_id_all __maybe_unused)
839 {
840 size_t size;
841
842 event->auxtrace_info.type = bswap_32(event->auxtrace_info.type);
843
844 size = event->header.size;
845 size -= (void *)&event->auxtrace_info.priv - (void *)event;
846 mem_bswap_64(event->auxtrace_info.priv, size);
847 }
848
849 static void perf_event__auxtrace_swap(union perf_event *event,
850 bool sample_id_all __maybe_unused)
851 {
852 event->auxtrace.size = bswap_64(event->auxtrace.size);
853 event->auxtrace.offset = bswap_64(event->auxtrace.offset);
854 event->auxtrace.reference = bswap_64(event->auxtrace.reference);
855 event->auxtrace.idx = bswap_32(event->auxtrace.idx);
856 event->auxtrace.tid = bswap_32(event->auxtrace.tid);
857 event->auxtrace.cpu = bswap_32(event->auxtrace.cpu);
858 }
859
860 static void perf_event__auxtrace_error_swap(union perf_event *event,
861 bool sample_id_all __maybe_unused)
862 {
863 event->auxtrace_error.type = bswap_32(event->auxtrace_error.type);
864 event->auxtrace_error.code = bswap_32(event->auxtrace_error.code);
865 event->auxtrace_error.cpu = bswap_32(event->auxtrace_error.cpu);
866 event->auxtrace_error.pid = bswap_32(event->auxtrace_error.pid);
867 event->auxtrace_error.tid = bswap_32(event->auxtrace_error.tid);
868 event->auxtrace_error.fmt = bswap_32(event->auxtrace_error.fmt);
869 event->auxtrace_error.ip = bswap_64(event->auxtrace_error.ip);
870 if (event->auxtrace_error.fmt)
871 event->auxtrace_error.time = bswap_64(event->auxtrace_error.time);
872 }
873
874 static void perf_event__thread_map_swap(union perf_event *event,
875 bool sample_id_all __maybe_unused)
876 {
877 unsigned i;
878
879 event->thread_map.nr = bswap_64(event->thread_map.nr);
880
881 for (i = 0; i < event->thread_map.nr; i++)
882 event->thread_map.entries[i].pid = bswap_64(event->thread_map.entries[i].pid);
883 }
884
885 static void perf_event__cpu_map_swap(union perf_event *event,
886 bool sample_id_all __maybe_unused)
887 {
888 struct perf_record_cpu_map_data *data = &event->cpu_map.data;
889 struct cpu_map_entries *cpus;
890 struct perf_record_record_cpu_map *mask;
891 unsigned i;
892
893 data->type = bswap_64(data->type);
894
895 switch (data->type) {
896 case PERF_CPU_MAP__CPUS:
897 cpus = (struct cpu_map_entries *)data->data;
898
899 cpus->nr = bswap_16(cpus->nr);
900
901 for (i = 0; i < cpus->nr; i++)
902 cpus->cpu[i] = bswap_16(cpus->cpu[i]);
903 break;
904 case PERF_CPU_MAP__MASK:
905 mask = (struct perf_record_record_cpu_map *)data->data;
906
907 mask->nr = bswap_16(mask->nr);
908 mask->long_size = bswap_16(mask->long_size);
909
910 switch (mask->long_size) {
911 case 4: mem_bswap_32(&mask->mask, mask->nr); break;
912 case 8: mem_bswap_64(&mask->mask, mask->nr); break;
913 default:
914 pr_err("cpu_map swap: unsupported long size\n");
915 }
916 default:
917 break;
918 }
919 }
920
921 static void perf_event__stat_config_swap(union perf_event *event,
922 bool sample_id_all __maybe_unused)
923 {
924 u64 size;
925
926 size = event->stat_config.nr * sizeof(event->stat_config.data[0]);
927 size += 1; /* nr item itself */
928 mem_bswap_64(&event->stat_config.nr, size);
929 }
930
931 static void perf_event__stat_swap(union perf_event *event,
932 bool sample_id_all __maybe_unused)
933 {
934 event->stat.id = bswap_64(event->stat.id);
935 event->stat.thread = bswap_32(event->stat.thread);
936 event->stat.cpu = bswap_32(event->stat.cpu);
937 event->stat.val = bswap_64(event->stat.val);
938 event->stat.ena = bswap_64(event->stat.ena);
939 event->stat.run = bswap_64(event->stat.run);
940 }
941
942 static void perf_event__stat_round_swap(union perf_event *event,
943 bool sample_id_all __maybe_unused)
944 {
945 event->stat_round.type = bswap_64(event->stat_round.type);
946 event->stat_round.time = bswap_64(event->stat_round.time);
947 }
948
949 static void perf_event__time_conv_swap(union perf_event *event,
950 bool sample_id_all __maybe_unused)
951 {
952 event->time_conv.time_shift = bswap_64(event->time_conv.time_shift);
953 event->time_conv.time_mult = bswap_64(event->time_conv.time_mult);
954 event->time_conv.time_zero = bswap_64(event->time_conv.time_zero);
955
956 if (event_contains(event->time_conv, time_cycles)) {
957 event->time_conv.time_cycles = bswap_64(event->time_conv.time_cycles);
958 event->time_conv.time_mask = bswap_64(event->time_conv.time_mask);
959 }
960 }
961
962 typedef void (*perf_event__swap_op)(union perf_event *event,
963 bool sample_id_all);
964
965 static perf_event__swap_op perf_event__swap_ops[] = {
966 [PERF_RECORD_MMAP] = perf_event__mmap_swap,
967 [PERF_RECORD_MMAP2] = perf_event__mmap2_swap,
968 [PERF_RECORD_COMM] = perf_event__comm_swap,
969 [PERF_RECORD_FORK] = perf_event__task_swap,
970 [PERF_RECORD_EXIT] = perf_event__task_swap,
971 [PERF_RECORD_LOST] = perf_event__all64_swap,
972 [PERF_RECORD_READ] = perf_event__read_swap,
973 [PERF_RECORD_THROTTLE] = perf_event__throttle_swap,
974 [PERF_RECORD_UNTHROTTLE] = perf_event__throttle_swap,
975 [PERF_RECORD_SAMPLE] = perf_event__all64_swap,
976 [PERF_RECORD_AUX] = perf_event__aux_swap,
977 [PERF_RECORD_ITRACE_START] = perf_event__itrace_start_swap,
978 [PERF_RECORD_LOST_SAMPLES] = perf_event__all64_swap,
979 [PERF_RECORD_SWITCH] = perf_event__switch_swap,
980 [PERF_RECORD_SWITCH_CPU_WIDE] = perf_event__switch_swap,
981 [PERF_RECORD_NAMESPACES] = perf_event__namespaces_swap,
982 [PERF_RECORD_CGROUP] = perf_event__cgroup_swap,
983 [PERF_RECORD_TEXT_POKE] = perf_event__text_poke_swap,
984 [PERF_RECORD_HEADER_ATTR] = perf_event__hdr_attr_swap,
985 [PERF_RECORD_HEADER_EVENT_TYPE] = perf_event__event_type_swap,
986 [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
987 [PERF_RECORD_HEADER_BUILD_ID] = NULL,
988 [PERF_RECORD_ID_INDEX] = perf_event__all64_swap,
989 [PERF_RECORD_AUXTRACE_INFO] = perf_event__auxtrace_info_swap,
990 [PERF_RECORD_AUXTRACE] = perf_event__auxtrace_swap,
991 [PERF_RECORD_AUXTRACE_ERROR] = perf_event__auxtrace_error_swap,
992 [PERF_RECORD_THREAD_MAP] = perf_event__thread_map_swap,
993 [PERF_RECORD_CPU_MAP] = perf_event__cpu_map_swap,
994 [PERF_RECORD_STAT_CONFIG] = perf_event__stat_config_swap,
995 [PERF_RECORD_STAT] = perf_event__stat_swap,
996 [PERF_RECORD_STAT_ROUND] = perf_event__stat_round_swap,
997 [PERF_RECORD_EVENT_UPDATE] = perf_event__event_update_swap,
998 [PERF_RECORD_TIME_CONV] = perf_event__time_conv_swap,
999 [PERF_RECORD_HEADER_MAX] = NULL,
1000 };
1001
1002 /*
1003 * When perf record finishes a pass on every buffers, it records this pseudo
1004 * event.
1005 * We record the max timestamp t found in the pass n.
1006 * Assuming these timestamps are monotonic across cpus, we know that if
1007 * a buffer still has events with timestamps below t, they will be all
1008 * available and then read in the pass n + 1.
1009 * Hence when we start to read the pass n + 2, we can safely flush every
1010 * events with timestamps below t.
1011 *
1012 * ============ PASS n =================
1013 * CPU 0 | CPU 1
1014 * |
1015 * cnt1 timestamps | cnt2 timestamps
1016 * 1 | 2
1017 * 2 | 3
1018 * - | 4 <--- max recorded
1019 *
1020 * ============ PASS n + 1 ==============
1021 * CPU 0 | CPU 1
1022 * |
1023 * cnt1 timestamps | cnt2 timestamps
1024 * 3 | 5
1025 * 4 | 6
1026 * 5 | 7 <---- max recorded
1027 *
1028 * Flush every events below timestamp 4
1029 *
1030 * ============ PASS n + 2 ==============
1031 * CPU 0 | CPU 1
1032 * |
1033 * cnt1 timestamps | cnt2 timestamps
1034 * 6 | 8
1035 * 7 | 9
1036 * - | 10
1037 *
1038 * Flush every events below timestamp 7
1039 * etc...
1040 */
1041 static int process_finished_round(struct perf_tool *tool __maybe_unused,
1042 union perf_event *event __maybe_unused,
1043 struct ordered_events *oe)
1044 {
1045 if (dump_trace)
1046 fprintf(stdout, "\n");
1047 return ordered_events__flush(oe, OE_FLUSH__ROUND);
1048 }
1049
1050 int perf_session__queue_event(struct perf_session *s, union perf_event *event,
1051 u64 timestamp, u64 file_offset)
1052 {
1053 return ordered_events__queue(&s->ordered_events, event, timestamp, file_offset);
1054 }
1055
1056 static void callchain__lbr_callstack_printf(struct perf_sample *sample)
1057 {
1058 struct ip_callchain *callchain = sample->callchain;
1059 struct branch_stack *lbr_stack = sample->branch_stack;
1060 struct branch_entry *entries = perf_sample__branch_entries(sample);
1061 u64 kernel_callchain_nr = callchain->nr;
1062 unsigned int i;
1063
1064 for (i = 0; i < kernel_callchain_nr; i++) {
1065 if (callchain->ips[i] == PERF_CONTEXT_USER)
1066 break;
1067 }
1068
1069 if ((i != kernel_callchain_nr) && lbr_stack->nr) {
1070 u64 total_nr;
1071 /*
1072 * LBR callstack can only get user call chain,
1073 * i is kernel call chain number,
1074 * 1 is PERF_CONTEXT_USER.
1075 *
1076 * The user call chain is stored in LBR registers.
1077 * LBR are pair registers. The caller is stored
1078 * in "from" register, while the callee is stored
1079 * in "to" register.
1080 * For example, there is a call stack
1081 * "A"->"B"->"C"->"D".
1082 * The LBR registers will recorde like
1083 * "C"->"D", "B"->"C", "A"->"B".
1084 * So only the first "to" register and all "from"
1085 * registers are needed to construct the whole stack.
1086 */
1087 total_nr = i + 1 + lbr_stack->nr + 1;
1088 kernel_callchain_nr = i + 1;
1089
1090 printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr);
1091
1092 for (i = 0; i < kernel_callchain_nr; i++)
1093 printf("..... %2d: %016" PRIx64 "\n",
1094 i, callchain->ips[i]);
1095
1096 printf("..... %2d: %016" PRIx64 "\n",
1097 (int)(kernel_callchain_nr), entries[0].to);
1098 for (i = 0; i < lbr_stack->nr; i++)
1099 printf("..... %2d: %016" PRIx64 "\n",
1100 (int)(i + kernel_callchain_nr + 1), entries[i].from);
1101 }
1102 }
1103
1104 static void callchain__printf(struct evsel *evsel,
1105 struct perf_sample *sample)
1106 {
1107 unsigned int i;
1108 struct ip_callchain *callchain = sample->callchain;
1109
1110 if (evsel__has_branch_callstack(evsel))
1111 callchain__lbr_callstack_printf(sample);
1112
1113 printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr);
1114
1115 for (i = 0; i < callchain->nr; i++)
1116 printf("..... %2d: %016" PRIx64 "\n",
1117 i, callchain->ips[i]);
1118 }
1119
1120 static void branch_stack__printf(struct perf_sample *sample, bool callstack)
1121 {
1122 struct branch_entry *entries = perf_sample__branch_entries(sample);
1123 uint64_t i;
1124
1125 printf("%s: nr:%" PRIu64 "\n",
1126 !callstack ? "... branch stack" : "... branch callstack",
1127 sample->branch_stack->nr);
1128
1129 for (i = 0; i < sample->branch_stack->nr; i++) {
1130 struct branch_entry *e = &entries[i];
1131
1132 if (!callstack) {
1133 printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x\n",
1134 i, e->from, e->to,
1135 (unsigned short)e->flags.cycles,
1136 e->flags.mispred ? "M" : " ",
1137 e->flags.predicted ? "P" : " ",
1138 e->flags.abort ? "A" : " ",
1139 e->flags.in_tx ? "T" : " ",
1140 (unsigned)e->flags.reserved);
1141 } else {
1142 printf("..... %2"PRIu64": %016" PRIx64 "\n",
1143 i, i > 0 ? e->from : e->to);
1144 }
1145 }
1146 }
1147
1148 static void regs_dump__printf(u64 mask, u64 *regs)
1149 {
1150 unsigned rid, i = 0;
1151
1152 for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) {
1153 u64 val = regs[i++];
1154
1155 printf(".... %-5s 0x%016" PRIx64 "\n",
1156 perf_reg_name(rid), val);
1157 }
1158 }
1159
1160 static const char *regs_abi[] = {
1161 [PERF_SAMPLE_REGS_ABI_NONE] = "none",
1162 [PERF_SAMPLE_REGS_ABI_32] = "32-bit",
1163 [PERF_SAMPLE_REGS_ABI_64] = "64-bit",
1164 };
1165
1166 static inline const char *regs_dump_abi(struct regs_dump *d)
1167 {
1168 if (d->abi > PERF_SAMPLE_REGS_ABI_64)
1169 return "unknown";
1170
1171 return regs_abi[d->abi];
1172 }
1173
1174 static void regs__printf(const char *type, struct regs_dump *regs)
1175 {
1176 u64 mask = regs->mask;
1177
1178 printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n",
1179 type,
1180 mask,
1181 regs_dump_abi(regs));
1182
1183 regs_dump__printf(mask, regs->regs);
1184 }
1185
1186 static void regs_user__printf(struct perf_sample *sample)
1187 {
1188 struct regs_dump *user_regs = &sample->user_regs;
1189
1190 if (user_regs->regs)
1191 regs__printf("user", user_regs);
1192 }
1193
1194 static void regs_intr__printf(struct perf_sample *sample)
1195 {
1196 struct regs_dump *intr_regs = &sample->intr_regs;
1197
1198 if (intr_regs->regs)
1199 regs__printf("intr", intr_regs);
1200 }
1201
1202 static void stack_user__printf(struct stack_dump *dump)
1203 {
1204 printf("... ustack: size %" PRIu64 ", offset 0x%x\n",
1205 dump->size, dump->offset);
1206 }
1207
1208 static void evlist__print_tstamp(struct evlist *evlist, union perf_event *event, struct perf_sample *sample)
1209 {
1210 u64 sample_type = __evlist__combined_sample_type(evlist);
1211
1212 if (event->header.type != PERF_RECORD_SAMPLE &&
1213 !evlist__sample_id_all(evlist)) {
1214 fputs("-1 -1 ", stdout);
1215 return;
1216 }
1217
1218 if ((sample_type & PERF_SAMPLE_CPU))
1219 printf("%u ", sample->cpu);
1220
1221 if (sample_type & PERF_SAMPLE_TIME)
1222 printf("%" PRIu64 " ", sample->time);
1223 }
1224
1225 static void sample_read__printf(struct perf_sample *sample, u64 read_format)
1226 {
1227 printf("... sample_read:\n");
1228
1229 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1230 printf("...... time enabled %016" PRIx64 "\n",
1231 sample->read.time_enabled);
1232
1233 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1234 printf("...... time running %016" PRIx64 "\n",
1235 sample->read.time_running);
1236
1237 if (read_format & PERF_FORMAT_GROUP) {
1238 u64 i;
1239
1240 printf(".... group nr %" PRIu64 "\n", sample->read.group.nr);
1241
1242 for (i = 0; i < sample->read.group.nr; i++) {
1243 struct sample_read_value *value;
1244
1245 value = &sample->read.group.values[i];
1246 printf("..... id %016" PRIx64
1247 ", value %016" PRIx64 "\n",
1248 value->id, value->value);
1249 }
1250 } else
1251 printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n",
1252 sample->read.one.id, sample->read.one.value);
1253 }
1254
1255 static void dump_event(struct evlist *evlist, union perf_event *event,
1256 u64 file_offset, struct perf_sample *sample)
1257 {
1258 if (!dump_trace)
1259 return;
1260
1261 printf("\n%#" PRIx64 " [%#x]: event: %d\n",
1262 file_offset, event->header.size, event->header.type);
1263
1264 trace_event(event);
1265 if (event->header.type == PERF_RECORD_SAMPLE && evlist->trace_event_sample_raw)
1266 evlist->trace_event_sample_raw(evlist, event, sample);
1267
1268 if (sample)
1269 evlist__print_tstamp(evlist, event, sample);
1270
1271 printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset,
1272 event->header.size, perf_event__name(event->header.type));
1273 }
1274
1275 char *get_page_size_name(u64 size, char *str)
1276 {
1277 if (!size || !unit_number__scnprintf(str, PAGE_SIZE_NAME_LEN, size))
1278 snprintf(str, PAGE_SIZE_NAME_LEN, "%s", "N/A");
1279
1280 return str;
1281 }
1282
1283 static void dump_sample(struct evsel *evsel, union perf_event *event,
1284 struct perf_sample *sample)
1285 {
1286 u64 sample_type;
1287 char str[PAGE_SIZE_NAME_LEN];
1288
1289 if (!dump_trace)
1290 return;
1291
1292 printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n",
1293 event->header.misc, sample->pid, sample->tid, sample->ip,
1294 sample->period, sample->addr);
1295
1296 sample_type = evsel->core.attr.sample_type;
1297
1298 if (evsel__has_callchain(evsel))
1299 callchain__printf(evsel, sample);
1300
1301 if (evsel__has_br_stack(evsel))
1302 branch_stack__printf(sample, evsel__has_branch_callstack(evsel));
1303
1304 if (sample_type & PERF_SAMPLE_REGS_USER)
1305 regs_user__printf(sample);
1306
1307 if (sample_type & PERF_SAMPLE_REGS_INTR)
1308 regs_intr__printf(sample);
1309
1310 if (sample_type & PERF_SAMPLE_STACK_USER)
1311 stack_user__printf(&sample->user_stack);
1312
1313 if (sample_type & PERF_SAMPLE_WEIGHT)
1314 printf("... weight: %" PRIu64 "\n", sample->weight);
1315
1316 if (sample_type & PERF_SAMPLE_DATA_SRC)
1317 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src);
1318
1319 if (sample_type & PERF_SAMPLE_PHYS_ADDR)
1320 printf(" .. phys_addr: 0x%"PRIx64"\n", sample->phys_addr);
1321
1322 if (sample_type & PERF_SAMPLE_DATA_PAGE_SIZE)
1323 printf(" .. data page size: %s\n", get_page_size_name(sample->data_page_size, str));
1324
1325 if (sample_type & PERF_SAMPLE_TRANSACTION)
1326 printf("... transaction: %" PRIx64 "\n", sample->transaction);
1327
1328 if (sample_type & PERF_SAMPLE_READ)
1329 sample_read__printf(sample, evsel->core.attr.read_format);
1330 }
1331
1332 static void dump_read(struct evsel *evsel, union perf_event *event)
1333 {
1334 struct perf_record_read *read_event = &event->read;
1335 u64 read_format;
1336
1337 if (!dump_trace)
1338 return;
1339
1340 printf(": %d %d %s %" PRI_lu64 "\n", event->read.pid, event->read.tid,
1341 evsel__name(evsel), event->read.value);
1342
1343 if (!evsel)
1344 return;
1345
1346 read_format = evsel->core.attr.read_format;
1347
1348 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1349 printf("... time enabled : %" PRI_lu64 "\n", read_event->time_enabled);
1350
1351 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1352 printf("... time running : %" PRI_lu64 "\n", read_event->time_running);
1353
1354 if (read_format & PERF_FORMAT_ID)
1355 printf("... id : %" PRI_lu64 "\n", read_event->id);
1356 }
1357
1358 static struct machine *machines__find_for_cpumode(struct machines *machines,
1359 union perf_event *event,
1360 struct perf_sample *sample)
1361 {
1362 struct machine *machine;
1363
1364 if (perf_guest &&
1365 ((sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL) ||
1366 (sample->cpumode == PERF_RECORD_MISC_GUEST_USER))) {
1367 u32 pid;
1368
1369 if (event->header.type == PERF_RECORD_MMAP
1370 || event->header.type == PERF_RECORD_MMAP2)
1371 pid = event->mmap.pid;
1372 else
1373 pid = sample->pid;
1374
1375 machine = machines__find(machines, pid);
1376 if (!machine)
1377 machine = machines__findnew(machines, DEFAULT_GUEST_KERNEL_ID);
1378 return machine;
1379 }
1380
1381 return &machines->host;
1382 }
1383
1384 static int deliver_sample_value(struct evlist *evlist,
1385 struct perf_tool *tool,
1386 union perf_event *event,
1387 struct perf_sample *sample,
1388 struct sample_read_value *v,
1389 struct machine *machine)
1390 {
1391 struct perf_sample_id *sid = evlist__id2sid(evlist, v->id);
1392 struct evsel *evsel;
1393
1394 if (sid) {
1395 sample->id = v->id;
1396 sample->period = v->value - sid->period;
1397 sid->period = v->value;
1398 }
1399
1400 if (!sid || sid->evsel == NULL) {
1401 ++evlist->stats.nr_unknown_id;
1402 return 0;
1403 }
1404
1405 /*
1406 * There's no reason to deliver sample
1407 * for zero period, bail out.
1408 */
1409 if (!sample->period)
1410 return 0;
1411
1412 evsel = container_of(sid->evsel, struct evsel, core);
1413 return tool->sample(tool, event, sample, evsel, machine);
1414 }
1415
1416 static int deliver_sample_group(struct evlist *evlist,
1417 struct perf_tool *tool,
1418 union perf_event *event,
1419 struct perf_sample *sample,
1420 struct machine *machine)
1421 {
1422 int ret = -EINVAL;
1423 u64 i;
1424
1425 for (i = 0; i < sample->read.group.nr; i++) {
1426 ret = deliver_sample_value(evlist, tool, event, sample,
1427 &sample->read.group.values[i],
1428 machine);
1429 if (ret)
1430 break;
1431 }
1432
1433 return ret;
1434 }
1435
1436 static int evlist__deliver_sample(struct evlist *evlist, struct perf_tool *tool,
1437 union perf_event *event, struct perf_sample *sample,
1438 struct evsel *evsel, struct machine *machine)
1439 {
1440 /* We know evsel != NULL. */
1441 u64 sample_type = evsel->core.attr.sample_type;
1442 u64 read_format = evsel->core.attr.read_format;
1443
1444 /* Standard sample delivery. */
1445 if (!(sample_type & PERF_SAMPLE_READ))
1446 return tool->sample(tool, event, sample, evsel, machine);
1447
1448 /* For PERF_SAMPLE_READ we have either single or group mode. */
1449 if (read_format & PERF_FORMAT_GROUP)
1450 return deliver_sample_group(evlist, tool, event, sample,
1451 machine);
1452 else
1453 return deliver_sample_value(evlist, tool, event, sample,
1454 &sample->read.one, machine);
1455 }
1456
1457 static int machines__deliver_event(struct machines *machines,
1458 struct evlist *evlist,
1459 union perf_event *event,
1460 struct perf_sample *sample,
1461 struct perf_tool *tool, u64 file_offset)
1462 {
1463 struct evsel *evsel;
1464 struct machine *machine;
1465
1466 dump_event(evlist, event, file_offset, sample);
1467
1468 evsel = evlist__id2evsel(evlist, sample->id);
1469
1470 machine = machines__find_for_cpumode(machines, event, sample);
1471
1472 switch (event->header.type) {
1473 case PERF_RECORD_SAMPLE:
1474 if (evsel == NULL) {
1475 ++evlist->stats.nr_unknown_id;
1476 return 0;
1477 }
1478 dump_sample(evsel, event, sample);
1479 if (machine == NULL) {
1480 ++evlist->stats.nr_unprocessable_samples;
1481 return 0;
1482 }
1483 return evlist__deliver_sample(evlist, tool, event, sample, evsel, machine);
1484 case PERF_RECORD_MMAP:
1485 return tool->mmap(tool, event, sample, machine);
1486 case PERF_RECORD_MMAP2:
1487 if (event->header.misc & PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT)
1488 ++evlist->stats.nr_proc_map_timeout;
1489 return tool->mmap2(tool, event, sample, machine);
1490 case PERF_RECORD_COMM:
1491 return tool->comm(tool, event, sample, machine);
1492 case PERF_RECORD_NAMESPACES:
1493 return tool->namespaces(tool, event, sample, machine);
1494 case PERF_RECORD_CGROUP:
1495 return tool->cgroup(tool, event, sample, machine);
1496 case PERF_RECORD_FORK:
1497 return tool->fork(tool, event, sample, machine);
1498 case PERF_RECORD_EXIT:
1499 return tool->exit(tool, event, sample, machine);
1500 case PERF_RECORD_LOST:
1501 if (tool->lost == perf_event__process_lost)
1502 evlist->stats.total_lost += event->lost.lost;
1503 return tool->lost(tool, event, sample, machine);
1504 case PERF_RECORD_LOST_SAMPLES:
1505 if (tool->lost_samples == perf_event__process_lost_samples)
1506 evlist->stats.total_lost_samples += event->lost_samples.lost;
1507 return tool->lost_samples(tool, event, sample, machine);
1508 case PERF_RECORD_READ:
1509 dump_read(evsel, event);
1510 return tool->read(tool, event, sample, evsel, machine);
1511 case PERF_RECORD_THROTTLE:
1512 return tool->throttle(tool, event, sample, machine);
1513 case PERF_RECORD_UNTHROTTLE:
1514 return tool->unthrottle(tool, event, sample, machine);
1515 case PERF_RECORD_AUX:
1516 if (tool->aux == perf_event__process_aux) {
1517 if (event->aux.flags & PERF_AUX_FLAG_TRUNCATED)
1518 evlist->stats.total_aux_lost += 1;
1519 if (event->aux.flags & PERF_AUX_FLAG_PARTIAL)
1520 evlist->stats.total_aux_partial += 1;
1521 }
1522 return tool->aux(tool, event, sample, machine);
1523 case PERF_RECORD_ITRACE_START:
1524 return tool->itrace_start(tool, event, sample, machine);
1525 case PERF_RECORD_SWITCH:
1526 case PERF_RECORD_SWITCH_CPU_WIDE:
1527 return tool->context_switch(tool, event, sample, machine);
1528 case PERF_RECORD_KSYMBOL:
1529 return tool->ksymbol(tool, event, sample, machine);
1530 case PERF_RECORD_BPF_EVENT:
1531 return tool->bpf(tool, event, sample, machine);
1532 case PERF_RECORD_TEXT_POKE:
1533 return tool->text_poke(tool, event, sample, machine);
1534 default:
1535 ++evlist->stats.nr_unknown_events;
1536 return -1;
1537 }
1538 }
1539
1540 static int perf_session__deliver_event(struct perf_session *session,
1541 union perf_event *event,
1542 struct perf_tool *tool,
1543 u64 file_offset)
1544 {
1545 struct perf_sample sample;
1546 int ret = evlist__parse_sample(session->evlist, event, &sample);
1547
1548 if (ret) {
1549 pr_err("Can't parse sample, err = %d\n", ret);
1550 return ret;
1551 }
1552
1553 ret = auxtrace__process_event(session, event, &sample, tool);
1554 if (ret < 0)
1555 return ret;
1556 if (ret > 0)
1557 return 0;
1558
1559 ret = machines__deliver_event(&session->machines, session->evlist,
1560 event, &sample, tool, file_offset);
1561
1562 if (dump_trace && sample.aux_sample.size)
1563 auxtrace__dump_auxtrace_sample(session, &sample);
1564
1565 return ret;
1566 }
1567
1568 static s64 perf_session__process_user_event(struct perf_session *session,
1569 union perf_event *event,
1570 u64 file_offset)
1571 {
1572 struct ordered_events *oe = &session->ordered_events;
1573 struct perf_tool *tool = session->tool;
1574 struct perf_sample sample = { .time = 0, };
1575 int fd = perf_data__fd(session->data);
1576 int err;
1577
1578 if (event->header.type != PERF_RECORD_COMPRESSED ||
1579 tool->compressed == perf_session__process_compressed_event_stub)
1580 dump_event(session->evlist, event, file_offset, &sample);
1581
1582 /* These events are processed right away */
1583 switch (event->header.type) {
1584 case PERF_RECORD_HEADER_ATTR:
1585 err = tool->attr(tool, event, &session->evlist);
1586 if (err == 0) {
1587 perf_session__set_id_hdr_size(session);
1588 perf_session__set_comm_exec(session);
1589 }
1590 return err;
1591 case PERF_RECORD_EVENT_UPDATE:
1592 return tool->event_update(tool, event, &session->evlist);
1593 case PERF_RECORD_HEADER_EVENT_TYPE:
1594 /*
1595 * Depreceated, but we need to handle it for sake
1596 * of old data files create in pipe mode.
1597 */
1598 return 0;
1599 case PERF_RECORD_HEADER_TRACING_DATA:
1600 /*
1601 * Setup for reading amidst mmap, but only when we
1602 * are in 'file' mode. The 'pipe' fd is in proper
1603 * place already.
1604 */
1605 if (!perf_data__is_pipe(session->data))
1606 lseek(fd, file_offset, SEEK_SET);
1607 return tool->tracing_data(session, event);
1608 case PERF_RECORD_HEADER_BUILD_ID:
1609 return tool->build_id(session, event);
1610 case PERF_RECORD_FINISHED_ROUND:
1611 return tool->finished_round(tool, event, oe);
1612 case PERF_RECORD_ID_INDEX:
1613 return tool->id_index(session, event);
1614 case PERF_RECORD_AUXTRACE_INFO:
1615 return tool->auxtrace_info(session, event);
1616 case PERF_RECORD_AUXTRACE:
1617 /* setup for reading amidst mmap */
1618 lseek(fd, file_offset + event->header.size, SEEK_SET);
1619 return tool->auxtrace(session, event);
1620 case PERF_RECORD_AUXTRACE_ERROR:
1621 perf_session__auxtrace_error_inc(session, event);
1622 return tool->auxtrace_error(session, event);
1623 case PERF_RECORD_THREAD_MAP:
1624 return tool->thread_map(session, event);
1625 case PERF_RECORD_CPU_MAP:
1626 return tool->cpu_map(session, event);
1627 case PERF_RECORD_STAT_CONFIG:
1628 return tool->stat_config(session, event);
1629 case PERF_RECORD_STAT:
1630 return tool->stat(session, event);
1631 case PERF_RECORD_STAT_ROUND:
1632 return tool->stat_round(session, event);
1633 case PERF_RECORD_TIME_CONV:
1634 session->time_conv = event->time_conv;
1635 return tool->time_conv(session, event);
1636 case PERF_RECORD_HEADER_FEATURE:
1637 return tool->feature(session, event);
1638 case PERF_RECORD_COMPRESSED:
1639 err = tool->compressed(session, event, file_offset);
1640 if (err)
1641 dump_event(session->evlist, event, file_offset, &sample);
1642 return err;
1643 default:
1644 return -EINVAL;
1645 }
1646 }
1647
1648 int perf_session__deliver_synth_event(struct perf_session *session,
1649 union perf_event *event,
1650 struct perf_sample *sample)
1651 {
1652 struct evlist *evlist = session->evlist;
1653 struct perf_tool *tool = session->tool;
1654
1655 events_stats__inc(&evlist->stats, event->header.type);
1656
1657 if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1658 return perf_session__process_user_event(session, event, 0);
1659
1660 return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0);
1661 }
1662
1663 static void event_swap(union perf_event *event, bool sample_id_all)
1664 {
1665 perf_event__swap_op swap;
1666
1667 swap = perf_event__swap_ops[event->header.type];
1668 if (swap)
1669 swap(event, sample_id_all);
1670 }
1671
1672 int perf_session__peek_event(struct perf_session *session, off_t file_offset,
1673 void *buf, size_t buf_sz,
1674 union perf_event **event_ptr,
1675 struct perf_sample *sample)
1676 {
1677 union perf_event *event;
1678 size_t hdr_sz, rest;
1679 int fd;
1680
1681 if (session->one_mmap && !session->header.needs_swap) {
1682 event = file_offset - session->one_mmap_offset +
1683 session->one_mmap_addr;
1684 goto out_parse_sample;
1685 }
1686
1687 if (perf_data__is_pipe(session->data))
1688 return -1;
1689
1690 fd = perf_data__fd(session->data);
1691 hdr_sz = sizeof(struct perf_event_header);
1692
1693 if (buf_sz < hdr_sz)
1694 return -1;
1695
1696 if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 ||
1697 readn(fd, buf, hdr_sz) != (ssize_t)hdr_sz)
1698 return -1;
1699
1700 event = (union perf_event *)buf;
1701
1702 if (session->header.needs_swap)
1703 perf_event_header__bswap(&event->header);
1704
1705 if (event->header.size < hdr_sz || event->header.size > buf_sz)
1706 return -1;
1707
1708 rest = event->header.size - hdr_sz;
1709
1710 if (readn(fd, buf, rest) != (ssize_t)rest)
1711 return -1;
1712
1713 if (session->header.needs_swap)
1714 event_swap(event, evlist__sample_id_all(session->evlist));
1715
1716 out_parse_sample:
1717
1718 if (sample && event->header.type < PERF_RECORD_USER_TYPE_START &&
1719 evlist__parse_sample(session->evlist, event, sample))
1720 return -1;
1721
1722 *event_ptr = event;
1723
1724 return 0;
1725 }
1726
1727 int perf_session__peek_events(struct perf_session *session, u64 offset,
1728 u64 size, peek_events_cb_t cb, void *data)
1729 {
1730 u64 max_offset = offset + size;
1731 char buf[PERF_SAMPLE_MAX_SIZE];
1732 union perf_event *event;
1733 int err;
1734
1735 do {
1736 err = perf_session__peek_event(session, offset, buf,
1737 PERF_SAMPLE_MAX_SIZE, &event,
1738 NULL);
1739 if (err)
1740 return err;
1741
1742 err = cb(session, event, offset, data);
1743 if (err)
1744 return err;
1745
1746 offset += event->header.size;
1747 if (event->header.type == PERF_RECORD_AUXTRACE)
1748 offset += event->auxtrace.size;
1749
1750 } while (offset < max_offset);
1751
1752 return err;
1753 }
1754
1755 static s64 perf_session__process_event(struct perf_session *session,
1756 union perf_event *event, u64 file_offset)
1757 {
1758 struct evlist *evlist = session->evlist;
1759 struct perf_tool *tool = session->tool;
1760 int ret;
1761
1762 if (session->header.needs_swap)
1763 event_swap(event, evlist__sample_id_all(evlist));
1764
1765 if (event->header.type >= PERF_RECORD_HEADER_MAX)
1766 return -EINVAL;
1767
1768 events_stats__inc(&evlist->stats, event->header.type);
1769
1770 if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1771 return perf_session__process_user_event(session, event, file_offset);
1772
1773 if (tool->ordered_events) {
1774 u64 timestamp = -1ULL;
1775
1776 ret = evlist__parse_sample_timestamp(evlist, event, &timestamp);
1777 if (ret && ret != -1)
1778 return ret;
1779
1780 ret = perf_session__queue_event(session, event, timestamp, file_offset);
1781 if (ret != -ETIME)
1782 return ret;
1783 }
1784
1785 return perf_session__deliver_event(session, event, tool, file_offset);
1786 }
1787
1788 void perf_event_header__bswap(struct perf_event_header *hdr)
1789 {
1790 hdr->type = bswap_32(hdr->type);
1791 hdr->misc = bswap_16(hdr->misc);
1792 hdr->size = bswap_16(hdr->size);
1793 }
1794
1795 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid)
1796 {
1797 return machine__findnew_thread(&session->machines.host, -1, pid);
1798 }
1799
1800 /*
1801 * Threads are identified by pid and tid, and the idle task has pid == tid == 0.
1802 * So here a single thread is created for that, but actually there is a separate
1803 * idle task per cpu, so there should be one 'struct thread' per cpu, but there
1804 * is only 1. That causes problems for some tools, requiring workarounds. For
1805 * example get_idle_thread() in builtin-sched.c, or thread_stack__per_cpu().
1806 */
1807 int perf_session__register_idle_thread(struct perf_session *session)
1808 {
1809 struct thread *thread;
1810 int err = 0;
1811
1812 thread = machine__findnew_thread(&session->machines.host, 0, 0);
1813 if (thread == NULL || thread__set_comm(thread, "swapper", 0)) {
1814 pr_err("problem inserting idle task.\n");
1815 err = -1;
1816 }
1817
1818 if (thread == NULL || thread__set_namespaces(thread, 0, NULL)) {
1819 pr_err("problem inserting idle task.\n");
1820 err = -1;
1821 }
1822
1823 /* machine__findnew_thread() got the thread, so put it */
1824 thread__put(thread);
1825 return err;
1826 }
1827
1828 static void
1829 perf_session__warn_order(const struct perf_session *session)
1830 {
1831 const struct ordered_events *oe = &session->ordered_events;
1832 struct evsel *evsel;
1833 bool should_warn = true;
1834
1835 evlist__for_each_entry(session->evlist, evsel) {
1836 if (evsel->core.attr.write_backward)
1837 should_warn = false;
1838 }
1839
1840 if (!should_warn)
1841 return;
1842 if (oe->nr_unordered_events != 0)
1843 ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events);
1844 }
1845
1846 static void perf_session__warn_about_errors(const struct perf_session *session)
1847 {
1848 const struct events_stats *stats = &session->evlist->stats;
1849
1850 if (session->tool->lost == perf_event__process_lost &&
1851 stats->nr_events[PERF_RECORD_LOST] != 0) {
1852 ui__warning("Processed %d events and lost %d chunks!\n\n"
1853 "Check IO/CPU overload!\n\n",
1854 stats->nr_events[0],
1855 stats->nr_events[PERF_RECORD_LOST]);
1856 }
1857
1858 if (session->tool->lost_samples == perf_event__process_lost_samples) {
1859 double drop_rate;
1860
1861 drop_rate = (double)stats->total_lost_samples /
1862 (double) (stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples);
1863 if (drop_rate > 0.05) {
1864 ui__warning("Processed %" PRIu64 " samples and lost %3.2f%%!\n\n",
1865 stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples,
1866 drop_rate * 100.0);
1867 }
1868 }
1869
1870 if (session->tool->aux == perf_event__process_aux &&
1871 stats->total_aux_lost != 0) {
1872 ui__warning("AUX data lost %" PRIu64 " times out of %u!\n\n",
1873 stats->total_aux_lost,
1874 stats->nr_events[PERF_RECORD_AUX]);
1875 }
1876
1877 if (session->tool->aux == perf_event__process_aux &&
1878 stats->total_aux_partial != 0) {
1879 bool vmm_exclusive = false;
1880
1881 (void)sysfs__read_bool("module/kvm_intel/parameters/vmm_exclusive",
1882 &vmm_exclusive);
1883
1884 ui__warning("AUX data had gaps in it %" PRIu64 " times out of %u!\n\n"
1885 "Are you running a KVM guest in the background?%s\n\n",
1886 stats->total_aux_partial,
1887 stats->nr_events[PERF_RECORD_AUX],
1888 vmm_exclusive ?
1889 "\nReloading kvm_intel module with vmm_exclusive=0\n"
1890 "will reduce the gaps to only guest's timeslices." :
1891 "");
1892 }
1893
1894 if (stats->nr_unknown_events != 0) {
1895 ui__warning("Found %u unknown events!\n\n"
1896 "Is this an older tool processing a perf.data "
1897 "file generated by a more recent tool?\n\n"
1898 "If that is not the case, consider "
1899 "reporting to linux-kernel@vger.kernel.org.\n\n",
1900 stats->nr_unknown_events);
1901 }
1902
1903 if (stats->nr_unknown_id != 0) {
1904 ui__warning("%u samples with id not present in the header\n",
1905 stats->nr_unknown_id);
1906 }
1907
1908 if (stats->nr_invalid_chains != 0) {
1909 ui__warning("Found invalid callchains!\n\n"
1910 "%u out of %u events were discarded for this reason.\n\n"
1911 "Consider reporting to linux-kernel@vger.kernel.org.\n\n",
1912 stats->nr_invalid_chains,
1913 stats->nr_events[PERF_RECORD_SAMPLE]);
1914 }
1915
1916 if (stats->nr_unprocessable_samples != 0) {
1917 ui__warning("%u unprocessable samples recorded.\n"
1918 "Do you have a KVM guest running and not using 'perf kvm'?\n",
1919 stats->nr_unprocessable_samples);
1920 }
1921
1922 perf_session__warn_order(session);
1923
1924 events_stats__auxtrace_error_warn(stats);
1925
1926 if (stats->nr_proc_map_timeout != 0) {
1927 ui__warning("%d map information files for pre-existing threads were\n"
1928 "not processed, if there are samples for addresses they\n"
1929 "will not be resolved, you may find out which are these\n"
1930 "threads by running with -v and redirecting the output\n"
1931 "to a file.\n"
1932 "The time limit to process proc map is too short?\n"
1933 "Increase it by --proc-map-timeout\n",
1934 stats->nr_proc_map_timeout);
1935 }
1936 }
1937
1938 static int perf_session__flush_thread_stack(struct thread *thread,
1939 void *p __maybe_unused)
1940 {
1941 return thread_stack__flush(thread);
1942 }
1943
1944 static int perf_session__flush_thread_stacks(struct perf_session *session)
1945 {
1946 return machines__for_each_thread(&session->machines,
1947 perf_session__flush_thread_stack,
1948 NULL);
1949 }
1950
1951 volatile int session_done;
1952
1953 static int __perf_session__process_decomp_events(struct perf_session *session);
1954
1955 static int __perf_session__process_pipe_events(struct perf_session *session)
1956 {
1957 struct ordered_events *oe = &session->ordered_events;
1958 struct perf_tool *tool = session->tool;
1959 union perf_event *event;
1960 uint32_t size, cur_size = 0;
1961 void *buf = NULL;
1962 s64 skip = 0;
1963 u64 head;
1964 ssize_t err;
1965 void *p;
1966
1967 perf_tool__fill_defaults(tool);
1968
1969 head = 0;
1970 cur_size = sizeof(union perf_event);
1971
1972 buf = malloc(cur_size);
1973 if (!buf)
1974 return -errno;
1975 ordered_events__set_copy_on_queue(oe, true);
1976 more:
1977 event = buf;
1978 err = perf_data__read(session->data, event,
1979 sizeof(struct perf_event_header));
1980 if (err <= 0) {
1981 if (err == 0)
1982 goto done;
1983
1984 pr_err("failed to read event header\n");
1985 goto out_err;
1986 }
1987
1988 if (session->header.needs_swap)
1989 perf_event_header__bswap(&event->header);
1990
1991 size = event->header.size;
1992 if (size < sizeof(struct perf_event_header)) {
1993 pr_err("bad event header size\n");
1994 goto out_err;
1995 }
1996
1997 if (size > cur_size) {
1998 void *new = realloc(buf, size);
1999 if (!new) {
2000 pr_err("failed to allocate memory to read event\n");
2001 goto out_err;
2002 }
2003 buf = new;
2004 cur_size = size;
2005 event = buf;
2006 }
2007 p = event;
2008 p += sizeof(struct perf_event_header);
2009
2010 if (size - sizeof(struct perf_event_header)) {
2011 err = perf_data__read(session->data, p,
2012 size - sizeof(struct perf_event_header));
2013 if (err <= 0) {
2014 if (err == 0) {
2015 pr_err("unexpected end of event stream\n");
2016 goto done;
2017 }
2018
2019 pr_err("failed to read event data\n");
2020 goto out_err;
2021 }
2022 }
2023
2024 if ((skip = perf_session__process_event(session, event, head)) < 0) {
2025 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
2026 head, event->header.size, event->header.type);
2027 err = -EINVAL;
2028 goto out_err;
2029 }
2030
2031 head += size;
2032
2033 if (skip > 0)
2034 head += skip;
2035
2036 err = __perf_session__process_decomp_events(session);
2037 if (err)
2038 goto out_err;
2039
2040 if (!session_done())
2041 goto more;
2042 done:
2043 /* do the final flush for ordered samples */
2044 err = ordered_events__flush(oe, OE_FLUSH__FINAL);
2045 if (err)
2046 goto out_err;
2047 err = auxtrace__flush_events(session, tool);
2048 if (err)
2049 goto out_err;
2050 err = perf_session__flush_thread_stacks(session);
2051 out_err:
2052 free(buf);
2053 if (!tool->no_warn)
2054 perf_session__warn_about_errors(session);
2055 ordered_events__free(&session->ordered_events);
2056 auxtrace__free_events(session);
2057 return err;
2058 }
2059
2060 static union perf_event *
2061 prefetch_event(char *buf, u64 head, size_t mmap_size,
2062 bool needs_swap, union perf_event *error)
2063 {
2064 union perf_event *event;
2065
2066 /*
2067 * Ensure we have enough space remaining to read
2068 * the size of the event in the headers.
2069 */
2070 if (head + sizeof(event->header) > mmap_size)
2071 return NULL;
2072
2073 event = (union perf_event *)(buf + head);
2074 if (needs_swap)
2075 perf_event_header__bswap(&event->header);
2076
2077 if (head + event->header.size <= mmap_size)
2078 return event;
2079
2080 /* We're not fetching the event so swap back again */
2081 if (needs_swap)
2082 perf_event_header__bswap(&event->header);
2083
2084 pr_debug("%s: head=%#" PRIx64 " event->header_size=%#x, mmap_size=%#zx:"
2085 " fuzzed or compressed perf.data?\n",__func__, head, event->header.size, mmap_size);
2086
2087 return error;
2088 }
2089
2090 static union perf_event *
2091 fetch_mmaped_event(u64 head, size_t mmap_size, char *buf, bool needs_swap)
2092 {
2093 return prefetch_event(buf, head, mmap_size, needs_swap, ERR_PTR(-EINVAL));
2094 }
2095
2096 static union perf_event *
2097 fetch_decomp_event(u64 head, size_t mmap_size, char *buf, bool needs_swap)
2098 {
2099 return prefetch_event(buf, head, mmap_size, needs_swap, NULL);
2100 }
2101
2102 static int __perf_session__process_decomp_events(struct perf_session *session)
2103 {
2104 s64 skip;
2105 u64 size, file_pos = 0;
2106 struct decomp *decomp = session->decomp_last;
2107
2108 if (!decomp)
2109 return 0;
2110
2111 while (decomp->head < decomp->size && !session_done()) {
2112 union perf_event *event = fetch_decomp_event(decomp->head, decomp->size, decomp->data,
2113 session->header.needs_swap);
2114
2115 if (!event)
2116 break;
2117
2118 size = event->header.size;
2119
2120 if (size < sizeof(struct perf_event_header) ||
2121 (skip = perf_session__process_event(session, event, file_pos)) < 0) {
2122 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
2123 decomp->file_pos + decomp->head, event->header.size, event->header.type);
2124 return -EINVAL;
2125 }
2126
2127 if (skip)
2128 size += skip;
2129
2130 decomp->head += size;
2131 }
2132
2133 return 0;
2134 }
2135
2136 /*
2137 * On 64bit we can mmap the data file in one go. No need for tiny mmap
2138 * slices. On 32bit we use 32MB.
2139 */
2140 #if BITS_PER_LONG == 64
2141 #define MMAP_SIZE ULLONG_MAX
2142 #define NUM_MMAPS 1
2143 #else
2144 #define MMAP_SIZE (32 * 1024 * 1024ULL)
2145 #define NUM_MMAPS 128
2146 #endif
2147
2148 struct reader;
2149
2150 typedef s64 (*reader_cb_t)(struct perf_session *session,
2151 union perf_event *event,
2152 u64 file_offset);
2153
2154 struct reader {
2155 int fd;
2156 u64 data_size;
2157 u64 data_offset;
2158 reader_cb_t process;
2159 };
2160
2161 static int
2162 reader__process_events(struct reader *rd, struct perf_session *session,
2163 struct ui_progress *prog)
2164 {
2165 u64 data_size = rd->data_size;
2166 u64 head, page_offset, file_offset, file_pos, size;
2167 int err = 0, mmap_prot, mmap_flags, map_idx = 0;
2168 size_t mmap_size;
2169 char *buf, *mmaps[NUM_MMAPS];
2170 union perf_event *event;
2171 s64 skip;
2172
2173 page_offset = page_size * (rd->data_offset / page_size);
2174 file_offset = page_offset;
2175 head = rd->data_offset - page_offset;
2176
2177 ui_progress__init_size(prog, data_size, "Processing events...");
2178
2179 data_size += rd->data_offset;
2180
2181 mmap_size = MMAP_SIZE;
2182 if (mmap_size > data_size) {
2183 mmap_size = data_size;
2184 session->one_mmap = true;
2185 }
2186
2187 memset(mmaps, 0, sizeof(mmaps));
2188
2189 mmap_prot = PROT_READ;
2190 mmap_flags = MAP_SHARED;
2191
2192 if (session->header.needs_swap) {
2193 mmap_prot |= PROT_WRITE;
2194 mmap_flags = MAP_PRIVATE;
2195 }
2196 remap:
2197 buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, rd->fd,
2198 file_offset);
2199 if (buf == MAP_FAILED) {
2200 pr_err("failed to mmap file\n");
2201 err = -errno;
2202 goto out;
2203 }
2204 mmaps[map_idx] = buf;
2205 map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1);
2206 file_pos = file_offset + head;
2207 if (session->one_mmap) {
2208 session->one_mmap_addr = buf;
2209 session->one_mmap_offset = file_offset;
2210 }
2211
2212 more:
2213 event = fetch_mmaped_event(head, mmap_size, buf, session->header.needs_swap);
2214 if (IS_ERR(event))
2215 return PTR_ERR(event);
2216
2217 if (!event) {
2218 if (mmaps[map_idx]) {
2219 munmap(mmaps[map_idx], mmap_size);
2220 mmaps[map_idx] = NULL;
2221 }
2222
2223 page_offset = page_size * (head / page_size);
2224 file_offset += page_offset;
2225 head -= page_offset;
2226 goto remap;
2227 }
2228
2229 size = event->header.size;
2230
2231 skip = -EINVAL;
2232
2233 if (size < sizeof(struct perf_event_header) ||
2234 (skip = rd->process(session, event, file_pos)) < 0) {
2235 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d [%s]\n",
2236 file_offset + head, event->header.size,
2237 event->header.type, strerror(-skip));
2238 err = skip;
2239 goto out;
2240 }
2241
2242 if (skip)
2243 size += skip;
2244
2245 head += size;
2246 file_pos += size;
2247
2248 err = __perf_session__process_decomp_events(session);
2249 if (err)
2250 goto out;
2251
2252 ui_progress__update(prog, size);
2253
2254 if (session_done())
2255 goto out;
2256
2257 if (file_pos < data_size)
2258 goto more;
2259
2260 out:
2261 return err;
2262 }
2263
2264 static s64 process_simple(struct perf_session *session,
2265 union perf_event *event,
2266 u64 file_offset)
2267 {
2268 return perf_session__process_event(session, event, file_offset);
2269 }
2270
2271 static int __perf_session__process_events(struct perf_session *session)
2272 {
2273 struct reader rd = {
2274 .fd = perf_data__fd(session->data),
2275 .data_size = session->header.data_size,
2276 .data_offset = session->header.data_offset,
2277 .process = process_simple,
2278 };
2279 struct ordered_events *oe = &session->ordered_events;
2280 struct perf_tool *tool = session->tool;
2281 struct ui_progress prog;
2282 int err;
2283
2284 perf_tool__fill_defaults(tool);
2285
2286 if (rd.data_size == 0)
2287 return -1;
2288
2289 ui_progress__init_size(&prog, rd.data_size, "Processing events...");
2290
2291 err = reader__process_events(&rd, session, &prog);
2292 if (err)
2293 goto out_err;
2294 /* do the final flush for ordered samples */
2295 err = ordered_events__flush(oe, OE_FLUSH__FINAL);
2296 if (err)
2297 goto out_err;
2298 err = auxtrace__flush_events(session, tool);
2299 if (err)
2300 goto out_err;
2301 err = perf_session__flush_thread_stacks(session);
2302 out_err:
2303 ui_progress__finish();
2304 if (!tool->no_warn)
2305 perf_session__warn_about_errors(session);
2306 /*
2307 * We may switching perf.data output, make ordered_events
2308 * reusable.
2309 */
2310 ordered_events__reinit(&session->ordered_events);
2311 auxtrace__free_events(session);
2312 session->one_mmap = false;
2313 return err;
2314 }
2315
2316 int perf_session__process_events(struct perf_session *session)
2317 {
2318 if (perf_session__register_idle_thread(session) < 0)
2319 return -ENOMEM;
2320
2321 if (perf_data__is_pipe(session->data))
2322 return __perf_session__process_pipe_events(session);
2323
2324 return __perf_session__process_events(session);
2325 }
2326
2327 bool perf_session__has_traces(struct perf_session *session, const char *msg)
2328 {
2329 struct evsel *evsel;
2330
2331 evlist__for_each_entry(session->evlist, evsel) {
2332 if (evsel->core.attr.type == PERF_TYPE_TRACEPOINT)
2333 return true;
2334 }
2335
2336 pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg);
2337 return false;
2338 }
2339
2340 int map__set_kallsyms_ref_reloc_sym(struct map *map, const char *symbol_name, u64 addr)
2341 {
2342 char *bracket;
2343 struct ref_reloc_sym *ref;
2344 struct kmap *kmap;
2345
2346 ref = zalloc(sizeof(struct ref_reloc_sym));
2347 if (ref == NULL)
2348 return -ENOMEM;
2349
2350 ref->name = strdup(symbol_name);
2351 if (ref->name == NULL) {
2352 free(ref);
2353 return -ENOMEM;
2354 }
2355
2356 bracket = strchr(ref->name, ']');
2357 if (bracket)
2358 *bracket = '\0';
2359
2360 ref->addr = addr;
2361
2362 kmap = map__kmap(map);
2363 if (kmap)
2364 kmap->ref_reloc_sym = ref;
2365
2366 return 0;
2367 }
2368
2369 size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp)
2370 {
2371 return machines__fprintf_dsos(&session->machines, fp);
2372 }
2373
2374 size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp,
2375 bool (skip)(struct dso *dso, int parm), int parm)
2376 {
2377 return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm);
2378 }
2379
2380 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp)
2381 {
2382 size_t ret;
2383 const char *msg = "";
2384
2385 if (perf_header__has_feat(&session->header, HEADER_AUXTRACE))
2386 msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)";
2387
2388 ret = fprintf(fp, "\nAggregated stats:%s\n", msg);
2389
2390 ret += events_stats__fprintf(&session->evlist->stats, fp);
2391 return ret;
2392 }
2393
2394 size_t perf_session__fprintf(struct perf_session *session, FILE *fp)
2395 {
2396 /*
2397 * FIXME: Here we have to actually print all the machines in this
2398 * session, not just the host...
2399 */
2400 return machine__fprintf(&session->machines.host, fp);
2401 }
2402
2403 struct evsel *perf_session__find_first_evtype(struct perf_session *session,
2404 unsigned int type)
2405 {
2406 struct evsel *pos;
2407
2408 evlist__for_each_entry(session->evlist, pos) {
2409 if (pos->core.attr.type == type)
2410 return pos;
2411 }
2412 return NULL;
2413 }
2414
2415 int perf_session__cpu_bitmap(struct perf_session *session,
2416 const char *cpu_list, unsigned long *cpu_bitmap)
2417 {
2418 int i, err = -1;
2419 struct perf_cpu_map *map;
2420 int nr_cpus = min(session->header.env.nr_cpus_avail, MAX_NR_CPUS);
2421
2422 for (i = 0; i < PERF_TYPE_MAX; ++i) {
2423 struct evsel *evsel;
2424
2425 evsel = perf_session__find_first_evtype(session, i);
2426 if (!evsel)
2427 continue;
2428
2429 if (!(evsel->core.attr.sample_type & PERF_SAMPLE_CPU)) {
2430 pr_err("File does not contain CPU events. "
2431 "Remove -C option to proceed.\n");
2432 return -1;
2433 }
2434 }
2435
2436 map = perf_cpu_map__new(cpu_list);
2437 if (map == NULL) {
2438 pr_err("Invalid cpu_list\n");
2439 return -1;
2440 }
2441
2442 for (i = 0; i < map->nr; i++) {
2443 int cpu = map->map[i];
2444
2445 if (cpu >= nr_cpus) {
2446 pr_err("Requested CPU %d too large. "
2447 "Consider raising MAX_NR_CPUS\n", cpu);
2448 goto out_delete_map;
2449 }
2450
2451 set_bit(cpu, cpu_bitmap);
2452 }
2453
2454 err = 0;
2455
2456 out_delete_map:
2457 perf_cpu_map__put(map);
2458 return err;
2459 }
2460
2461 void perf_session__fprintf_info(struct perf_session *session, FILE *fp,
2462 bool full)
2463 {
2464 if (session == NULL || fp == NULL)
2465 return;
2466
2467 fprintf(fp, "# ========\n");
2468 perf_header__fprintf_info(session, fp, full);
2469 fprintf(fp, "# ========\n#\n");
2470 }
2471
2472 int perf_event__process_id_index(struct perf_session *session,
2473 union perf_event *event)
2474 {
2475 struct evlist *evlist = session->evlist;
2476 struct perf_record_id_index *ie = &event->id_index;
2477 size_t i, nr, max_nr;
2478
2479 max_nr = (ie->header.size - sizeof(struct perf_record_id_index)) /
2480 sizeof(struct id_index_entry);
2481 nr = ie->nr;
2482 if (nr > max_nr)
2483 return -EINVAL;
2484
2485 if (dump_trace)
2486 fprintf(stdout, " nr: %zu\n", nr);
2487
2488 for (i = 0; i < nr; i++) {
2489 struct id_index_entry *e = &ie->entries[i];
2490 struct perf_sample_id *sid;
2491
2492 if (dump_trace) {
2493 fprintf(stdout, " ... id: %"PRI_lu64, e->id);
2494 fprintf(stdout, " idx: %"PRI_lu64, e->idx);
2495 fprintf(stdout, " cpu: %"PRI_ld64, e->cpu);
2496 fprintf(stdout, " tid: %"PRI_ld64"\n", e->tid);
2497 }
2498
2499 sid = evlist__id2sid(evlist, e->id);
2500 if (!sid)
2501 return -ENOENT;
2502 sid->idx = e->idx;
2503 sid->cpu = e->cpu;
2504 sid->tid = e->tid;
2505 }
2506 return 0;
2507 }
Ubuntu Hirsute Linux kernel mirror