?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
ceph: use i_size_{read,write} to get/set i_size
[linux/fpc-iii.git] / tools / perf / util / session.c
blobc35ffdd360fe13e1590559217365e0f78414e91c
1 #include <linux/kernel.h>
2 #include <traceevent/event-parse.h>
3
4 #include <byteswap.h>
5 #include <unistd.h>
6 #include <sys/types.h>
7 #include <sys/mman.h>
8
9 #include "evlist.h"
10 #include "evsel.h"
11 #include "session.h"
12 #include "tool.h"
13 #include "sort.h"
14 #include "util.h"
15 #include "cpumap.h"
16 #include "perf_regs.h"
17 #include "asm/bug.h"
18 #include "auxtrace.h"
19 #include "thread-stack.h"
20
21 static int perf_session__deliver_event(struct perf_session *session,
22 union perf_event *event,
23 struct perf_sample *sample,
24 struct perf_tool *tool,
25 u64 file_offset);
26
27 static int perf_session__open(struct perf_session *session)
28 {
29 struct perf_data_file *file = session->file;
30
31 if (perf_session__read_header(session) < 0) {
32 pr_err("incompatible file format (rerun with -v to learn more)\n");
33 return -1;
34 }
35
36 if (perf_data_file__is_pipe(file))
37 return 0;
38
39 if (!perf_evlist__valid_sample_type(session->evlist)) {
40 pr_err("non matching sample_type\n");
41 return -1;
42 }
43
44 if (!perf_evlist__valid_sample_id_all(session->evlist)) {
45 pr_err("non matching sample_id_all\n");
46 return -1;
47 }
48
49 if (!perf_evlist__valid_read_format(session->evlist)) {
50 pr_err("non matching read_format\n");
51 return -1;
52 }
53
54 return 0;
55 }
56
57 void perf_session__set_id_hdr_size(struct perf_session *session)
58 {
59 u16 id_hdr_size = perf_evlist__id_hdr_size(session->evlist);
60
61 machines__set_id_hdr_size(&session->machines, id_hdr_size);
62 }
63
64 int perf_session__create_kernel_maps(struct perf_session *session)
65 {
66 int ret = machine__create_kernel_maps(&session->machines.host);
67
68 if (ret >= 0)
69 ret = machines__create_guest_kernel_maps(&session->machines);
70 return ret;
71 }
72
73 static void perf_session__destroy_kernel_maps(struct perf_session *session)
74 {
75 machines__destroy_kernel_maps(&session->machines);
76 }
77
78 static bool perf_session__has_comm_exec(struct perf_session *session)
79 {
80 struct perf_evsel *evsel;
81
82 evlist__for_each(session->evlist, evsel) {
83 if (evsel->attr.comm_exec)
84 return true;
85 }
86
87 return false;
88 }
89
90 static void perf_session__set_comm_exec(struct perf_session *session)
91 {
92 bool comm_exec = perf_session__has_comm_exec(session);
93
94 machines__set_comm_exec(&session->machines, comm_exec);
95 }
96
97 static int ordered_events__deliver_event(struct ordered_events *oe,
98 struct ordered_event *event)
99 {
100 struct perf_sample sample;
101 struct perf_session *session = container_of(oe, struct perf_session,
102 ordered_events);
103 int ret = perf_evlist__parse_sample(session->evlist, event->event, &sample);
104
105 if (ret) {
106 pr_err("Can't parse sample, err = %d\n", ret);
107 return ret;
108 }
109
110 return perf_session__deliver_event(session, event->event, &sample,
111 session->tool, event->file_offset);
112 }
113
114 struct perf_session *perf_session__new(struct perf_data_file *file,
115 bool repipe, struct perf_tool *tool)
116 {
117 struct perf_session *session = zalloc(sizeof(*session));
118
119 if (!session)
120 goto out;
121
122 session->repipe = repipe;
123 session->tool = tool;
124 INIT_LIST_HEAD(&session->auxtrace_index);
125 machines__init(&session->machines);
126 ordered_events__init(&session->ordered_events, ordered_events__deliver_event);
127
128 if (file) {
129 if (perf_data_file__open(file))
130 goto out_delete;
131
132 session->file = file;
133
134 if (perf_data_file__is_read(file)) {
135 if (perf_session__open(session) < 0)
136 goto out_close;
137
138 perf_session__set_id_hdr_size(session);
139 perf_session__set_comm_exec(session);
140 }
141 } else {
142 session->machines.host.env = &perf_env;
143 }
144
145 if (!file || perf_data_file__is_write(file)) {
146 /*
147 * In O_RDONLY mode this will be performed when reading the
148 * kernel MMAP event, in perf_event__process_mmap().
149 */
150 if (perf_session__create_kernel_maps(session) < 0)
151 pr_warning("Cannot read kernel map\n");
152 }
153
154 if (tool && tool->ordering_requires_timestamps &&
155 tool->ordered_events && !perf_evlist__sample_id_all(session->evlist)) {
156 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n");
157 tool->ordered_events = false;
158 }
159
160 return session;
161
162 out_close:
163 perf_data_file__close(file);
164 out_delete:
165 perf_session__delete(session);
166 out:
167 return NULL;
168 }
169
170 static void perf_session__delete_threads(struct perf_session *session)
171 {
172 machine__delete_threads(&session->machines.host);
173 }
174
175 void perf_session__delete(struct perf_session *session)
176 {
177 auxtrace__free(session);
178 auxtrace_index__free(&session->auxtrace_index);
179 perf_session__destroy_kernel_maps(session);
180 perf_session__delete_threads(session);
181 perf_env__exit(&session->header.env);
182 machines__exit(&session->machines);
183 if (session->file)
184 perf_data_file__close(session->file);
185 free(session);
186 }
187
188 static int process_event_synth_tracing_data_stub(struct perf_tool *tool
189 __maybe_unused,
190 union perf_event *event
191 __maybe_unused,
192 struct perf_session *session
193 __maybe_unused)
194 {
195 dump_printf(": unhandled!\n");
196 return 0;
197 }
198
199 static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused,
200 union perf_event *event __maybe_unused,
201 struct perf_evlist **pevlist
202 __maybe_unused)
203 {
204 dump_printf(": unhandled!\n");
205 return 0;
206 }
207
208 static int process_event_sample_stub(struct perf_tool *tool __maybe_unused,
209 union perf_event *event __maybe_unused,
210 struct perf_sample *sample __maybe_unused,
211 struct perf_evsel *evsel __maybe_unused,
212 struct machine *machine __maybe_unused)
213 {
214 dump_printf(": unhandled!\n");
215 return 0;
216 }
217
218 static int process_event_stub(struct perf_tool *tool __maybe_unused,
219 union perf_event *event __maybe_unused,
220 struct perf_sample *sample __maybe_unused,
221 struct machine *machine __maybe_unused)
222 {
223 dump_printf(": unhandled!\n");
224 return 0;
225 }
226
227 static int process_build_id_stub(struct perf_tool *tool __maybe_unused,
228 union perf_event *event __maybe_unused,
229 struct perf_session *session __maybe_unused)
230 {
231 dump_printf(": unhandled!\n");
232 return 0;
233 }
234
235 static int process_finished_round_stub(struct perf_tool *tool __maybe_unused,
236 union perf_event *event __maybe_unused,
237 struct ordered_events *oe __maybe_unused)
238 {
239 dump_printf(": unhandled!\n");
240 return 0;
241 }
242
243 static int process_finished_round(struct perf_tool *tool,
244 union perf_event *event,
245 struct ordered_events *oe);
246
247 static int process_id_index_stub(struct perf_tool *tool __maybe_unused,
248 union perf_event *event __maybe_unused,
249 struct perf_session *perf_session
250 __maybe_unused)
251 {
252 dump_printf(": unhandled!\n");
253 return 0;
254 }
255
256 static int process_event_auxtrace_info_stub(struct perf_tool *tool __maybe_unused,
257 union perf_event *event __maybe_unused,
258 struct perf_session *session __maybe_unused)
259 {
260 dump_printf(": unhandled!\n");
261 return 0;
262 }
263
264 static int skipn(int fd, off_t n)
265 {
266 char buf[4096];
267 ssize_t ret;
268
269 while (n > 0) {
270 ret = read(fd, buf, min(n, (off_t)sizeof(buf)));
271 if (ret <= 0)
272 return ret;
273 n -= ret;
274 }
275
276 return 0;
277 }
278
279 static s64 process_event_auxtrace_stub(struct perf_tool *tool __maybe_unused,
280 union perf_event *event,
281 struct perf_session *session
282 __maybe_unused)
283 {
284 dump_printf(": unhandled!\n");
285 if (perf_data_file__is_pipe(session->file))
286 skipn(perf_data_file__fd(session->file), event->auxtrace.size);
287 return event->auxtrace.size;
288 }
289
290 static
291 int process_event_auxtrace_error_stub(struct perf_tool *tool __maybe_unused,
292 union perf_event *event __maybe_unused,
293 struct perf_session *session __maybe_unused)
294 {
295 dump_printf(": unhandled!\n");
296 return 0;
297 }
298
299 void perf_tool__fill_defaults(struct perf_tool *tool)
300 {
301 if (tool->sample == NULL)
302 tool->sample = process_event_sample_stub;
303 if (tool->mmap == NULL)
304 tool->mmap = process_event_stub;
305 if (tool->mmap2 == NULL)
306 tool->mmap2 = process_event_stub;
307 if (tool->comm == NULL)
308 tool->comm = process_event_stub;
309 if (tool->fork == NULL)
310 tool->fork = process_event_stub;
311 if (tool->exit == NULL)
312 tool->exit = process_event_stub;
313 if (tool->lost == NULL)
314 tool->lost = perf_event__process_lost;
315 if (tool->lost_samples == NULL)
316 tool->lost_samples = perf_event__process_lost_samples;
317 if (tool->aux == NULL)
318 tool->aux = perf_event__process_aux;
319 if (tool->itrace_start == NULL)
320 tool->itrace_start = perf_event__process_itrace_start;
321 if (tool->context_switch == NULL)
322 tool->context_switch = perf_event__process_switch;
323 if (tool->read == NULL)
324 tool->read = process_event_sample_stub;
325 if (tool->throttle == NULL)
326 tool->throttle = process_event_stub;
327 if (tool->unthrottle == NULL)
328 tool->unthrottle = process_event_stub;
329 if (tool->attr == NULL)
330 tool->attr = process_event_synth_attr_stub;
331 if (tool->tracing_data == NULL)
332 tool->tracing_data = process_event_synth_tracing_data_stub;
333 if (tool->build_id == NULL)
334 tool->build_id = process_build_id_stub;
335 if (tool->finished_round == NULL) {
336 if (tool->ordered_events)
337 tool->finished_round = process_finished_round;
338 else
339 tool->finished_round = process_finished_round_stub;
340 }
341 if (tool->id_index == NULL)
342 tool->id_index = process_id_index_stub;
343 if (tool->auxtrace_info == NULL)
344 tool->auxtrace_info = process_event_auxtrace_info_stub;
345 if (tool->auxtrace == NULL)
346 tool->auxtrace = process_event_auxtrace_stub;
347 if (tool->auxtrace_error == NULL)
348 tool->auxtrace_error = process_event_auxtrace_error_stub;
349 }
350
351 static void swap_sample_id_all(union perf_event *event, void *data)
352 {
353 void *end = (void *) event + event->header.size;
354 int size = end - data;
355
356 BUG_ON(size % sizeof(u64));
357 mem_bswap_64(data, size);
358 }
359
360 static void perf_event__all64_swap(union perf_event *event,
361 bool sample_id_all __maybe_unused)
362 {
363 struct perf_event_header *hdr = &event->header;
364 mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr));
365 }
366
367 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all)
368 {
369 event->comm.pid = bswap_32(event->comm.pid);
370 event->comm.tid = bswap_32(event->comm.tid);
371
372 if (sample_id_all) {
373 void *data = &event->comm.comm;
374
375 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
376 swap_sample_id_all(event, data);
377 }
378 }
379
380 static void perf_event__mmap_swap(union perf_event *event,
381 bool sample_id_all)
382 {
383 event->mmap.pid = bswap_32(event->mmap.pid);
384 event->mmap.tid = bswap_32(event->mmap.tid);
385 event->mmap.start = bswap_64(event->mmap.start);
386 event->mmap.len = bswap_64(event->mmap.len);
387 event->mmap.pgoff = bswap_64(event->mmap.pgoff);
388
389 if (sample_id_all) {
390 void *data = &event->mmap.filename;
391
392 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
393 swap_sample_id_all(event, data);
394 }
395 }
396
397 static void perf_event__mmap2_swap(union perf_event *event,
398 bool sample_id_all)
399 {
400 event->mmap2.pid = bswap_32(event->mmap2.pid);
401 event->mmap2.tid = bswap_32(event->mmap2.tid);
402 event->mmap2.start = bswap_64(event->mmap2.start);
403 event->mmap2.len = bswap_64(event->mmap2.len);
404 event->mmap2.pgoff = bswap_64(event->mmap2.pgoff);
405 event->mmap2.maj = bswap_32(event->mmap2.maj);
406 event->mmap2.min = bswap_32(event->mmap2.min);
407 event->mmap2.ino = bswap_64(event->mmap2.ino);
408
409 if (sample_id_all) {
410 void *data = &event->mmap2.filename;
411
412 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
413 swap_sample_id_all(event, data);
414 }
415 }
416 static void perf_event__task_swap(union perf_event *event, bool sample_id_all)
417 {
418 event->fork.pid = bswap_32(event->fork.pid);
419 event->fork.tid = bswap_32(event->fork.tid);
420 event->fork.ppid = bswap_32(event->fork.ppid);
421 event->fork.ptid = bswap_32(event->fork.ptid);
422 event->fork.time = bswap_64(event->fork.time);
423
424 if (sample_id_all)
425 swap_sample_id_all(event, &event->fork + 1);
426 }
427
428 static void perf_event__read_swap(union perf_event *event, bool sample_id_all)
429 {
430 event->read.pid = bswap_32(event->read.pid);
431 event->read.tid = bswap_32(event->read.tid);
432 event->read.value = bswap_64(event->read.value);
433 event->read.time_enabled = bswap_64(event->read.time_enabled);
434 event->read.time_running = bswap_64(event->read.time_running);
435 event->read.id = bswap_64(event->read.id);
436
437 if (sample_id_all)
438 swap_sample_id_all(event, &event->read + 1);
439 }
440
441 static void perf_event__aux_swap(union perf_event *event, bool sample_id_all)
442 {
443 event->aux.aux_offset = bswap_64(event->aux.aux_offset);
444 event->aux.aux_size = bswap_64(event->aux.aux_size);
445 event->aux.flags = bswap_64(event->aux.flags);
446
447 if (sample_id_all)
448 swap_sample_id_all(event, &event->aux + 1);
449 }
450
451 static void perf_event__itrace_start_swap(union perf_event *event,
452 bool sample_id_all)
453 {
454 event->itrace_start.pid = bswap_32(event->itrace_start.pid);
455 event->itrace_start.tid = bswap_32(event->itrace_start.tid);
456
457 if (sample_id_all)
458 swap_sample_id_all(event, &event->itrace_start + 1);
459 }
460
461 static void perf_event__switch_swap(union perf_event *event, bool sample_id_all)
462 {
463 if (event->header.type == PERF_RECORD_SWITCH_CPU_WIDE) {
464 event->context_switch.next_prev_pid =
465 bswap_32(event->context_switch.next_prev_pid);
466 event->context_switch.next_prev_tid =
467 bswap_32(event->context_switch.next_prev_tid);
468 }
469
470 if (sample_id_all)
471 swap_sample_id_all(event, &event->context_switch + 1);
472 }
473
474 static void perf_event__throttle_swap(union perf_event *event,
475 bool sample_id_all)
476 {
477 event->throttle.time = bswap_64(event->throttle.time);
478 event->throttle.id = bswap_64(event->throttle.id);
479 event->throttle.stream_id = bswap_64(event->throttle.stream_id);
480
481 if (sample_id_all)
482 swap_sample_id_all(event, &event->throttle + 1);
483 }
484
485 static u8 revbyte(u8 b)
486 {
487 int rev = (b >> 4) | ((b & 0xf) << 4);
488 rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2);
489 rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1);
490 return (u8) rev;
491 }
492
493 /*
494 * XXX this is hack in attempt to carry flags bitfield
495 * throught endian village. ABI says:
496 *
497 * Bit-fields are allocated from right to left (least to most significant)
498 * on little-endian implementations and from left to right (most to least
499 * significant) on big-endian implementations.
500 *
501 * The above seems to be byte specific, so we need to reverse each
502 * byte of the bitfield. 'Internet' also says this might be implementation
503 * specific and we probably need proper fix and carry perf_event_attr
504 * bitfield flags in separate data file FEAT_ section. Thought this seems
505 * to work for now.
506 */
507 static void swap_bitfield(u8 *p, unsigned len)
508 {
509 unsigned i;
510
511 for (i = 0; i < len; i++) {
512 *p = revbyte(*p);
513 p++;
514 }
515 }
516
517 /* exported for swapping attributes in file header */
518 void perf_event__attr_swap(struct perf_event_attr *attr)
519 {
520 attr->type = bswap_32(attr->type);
521 attr->size = bswap_32(attr->size);
522
523 #define bswap_safe(f, n) \
524 (attr->size > (offsetof(struct perf_event_attr, f) + \
525 sizeof(attr->f) * (n)))
526 #define bswap_field(f, sz) \
527 do { \
528 if (bswap_safe(f, 0)) \
529 attr->f = bswap_##sz(attr->f); \
530 } while(0)
531 #define bswap_field_32(f) bswap_field(f, 32)
532 #define bswap_field_64(f) bswap_field(f, 64)
533
534 bswap_field_64(config);
535 bswap_field_64(sample_period);
536 bswap_field_64(sample_type);
537 bswap_field_64(read_format);
538 bswap_field_32(wakeup_events);
539 bswap_field_32(bp_type);
540 bswap_field_64(bp_addr);
541 bswap_field_64(bp_len);
542 bswap_field_64(branch_sample_type);
543 bswap_field_64(sample_regs_user);
544 bswap_field_32(sample_stack_user);
545 bswap_field_32(aux_watermark);
546
547 /*
548 * After read_format are bitfields. Check read_format because
549 * we are unable to use offsetof on bitfield.
550 */
551 if (bswap_safe(read_format, 1))
552 swap_bitfield((u8 *) (&attr->read_format + 1),
553 sizeof(u64));
554 #undef bswap_field_64
555 #undef bswap_field_32
556 #undef bswap_field
557 #undef bswap_safe
558 }
559
560 static void perf_event__hdr_attr_swap(union perf_event *event,
561 bool sample_id_all __maybe_unused)
562 {
563 size_t size;
564
565 perf_event__attr_swap(&event->attr.attr);
566
567 size = event->header.size;
568 size -= (void *)&event->attr.id - (void *)event;
569 mem_bswap_64(event->attr.id, size);
570 }
571
572 static void perf_event__event_type_swap(union perf_event *event,
573 bool sample_id_all __maybe_unused)
574 {
575 event->event_type.event_type.event_id =
576 bswap_64(event->event_type.event_type.event_id);
577 }
578
579 static void perf_event__tracing_data_swap(union perf_event *event,
580 bool sample_id_all __maybe_unused)
581 {
582 event->tracing_data.size = bswap_32(event->tracing_data.size);
583 }
584
585 static void perf_event__auxtrace_info_swap(union perf_event *event,
586 bool sample_id_all __maybe_unused)
587 {
588 size_t size;
589
590 event->auxtrace_info.type = bswap_32(event->auxtrace_info.type);
591
592 size = event->header.size;
593 size -= (void *)&event->auxtrace_info.priv - (void *)event;
594 mem_bswap_64(event->auxtrace_info.priv, size);
595 }
596
597 static void perf_event__auxtrace_swap(union perf_event *event,
598 bool sample_id_all __maybe_unused)
599 {
600 event->auxtrace.size = bswap_64(event->auxtrace.size);
601 event->auxtrace.offset = bswap_64(event->auxtrace.offset);
602 event->auxtrace.reference = bswap_64(event->auxtrace.reference);
603 event->auxtrace.idx = bswap_32(event->auxtrace.idx);
604 event->auxtrace.tid = bswap_32(event->auxtrace.tid);
605 event->auxtrace.cpu = bswap_32(event->auxtrace.cpu);
606 }
607
608 static void perf_event__auxtrace_error_swap(union perf_event *event,
609 bool sample_id_all __maybe_unused)
610 {
611 event->auxtrace_error.type = bswap_32(event->auxtrace_error.type);
612 event->auxtrace_error.code = bswap_32(event->auxtrace_error.code);
613 event->auxtrace_error.cpu = bswap_32(event->auxtrace_error.cpu);
614 event->auxtrace_error.pid = bswap_32(event->auxtrace_error.pid);
615 event->auxtrace_error.tid = bswap_32(event->auxtrace_error.tid);
616 event->auxtrace_error.ip = bswap_64(event->auxtrace_error.ip);
617 }
618
619 typedef void (*perf_event__swap_op)(union perf_event *event,
620 bool sample_id_all);
621
622 static perf_event__swap_op perf_event__swap_ops[] = {
623 [PERF_RECORD_MMAP] = perf_event__mmap_swap,
624 [PERF_RECORD_MMAP2] = perf_event__mmap2_swap,
625 [PERF_RECORD_COMM] = perf_event__comm_swap,
626 [PERF_RECORD_FORK] = perf_event__task_swap,
627 [PERF_RECORD_EXIT] = perf_event__task_swap,
628 [PERF_RECORD_LOST] = perf_event__all64_swap,
629 [PERF_RECORD_READ] = perf_event__read_swap,
630 [PERF_RECORD_THROTTLE] = perf_event__throttle_swap,
631 [PERF_RECORD_UNTHROTTLE] = perf_event__throttle_swap,
632 [PERF_RECORD_SAMPLE] = perf_event__all64_swap,
633 [PERF_RECORD_AUX] = perf_event__aux_swap,
634 [PERF_RECORD_ITRACE_START] = perf_event__itrace_start_swap,
635 [PERF_RECORD_LOST_SAMPLES] = perf_event__all64_swap,
636 [PERF_RECORD_SWITCH] = perf_event__switch_swap,
637 [PERF_RECORD_SWITCH_CPU_WIDE] = perf_event__switch_swap,
638 [PERF_RECORD_HEADER_ATTR] = perf_event__hdr_attr_swap,
639 [PERF_RECORD_HEADER_EVENT_TYPE] = perf_event__event_type_swap,
640 [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
641 [PERF_RECORD_HEADER_BUILD_ID] = NULL,
642 [PERF_RECORD_ID_INDEX] = perf_event__all64_swap,
643 [PERF_RECORD_AUXTRACE_INFO] = perf_event__auxtrace_info_swap,
644 [PERF_RECORD_AUXTRACE] = perf_event__auxtrace_swap,
645 [PERF_RECORD_AUXTRACE_ERROR] = perf_event__auxtrace_error_swap,
646 [PERF_RECORD_HEADER_MAX] = NULL,
647 };
648
649 /*
650 * When perf record finishes a pass on every buffers, it records this pseudo
651 * event.
652 * We record the max timestamp t found in the pass n.
653 * Assuming these timestamps are monotonic across cpus, we know that if
654 * a buffer still has events with timestamps below t, they will be all
655 * available and then read in the pass n + 1.
656 * Hence when we start to read the pass n + 2, we can safely flush every
657 * events with timestamps below t.
658 *
659 * ============ PASS n =================
660 * CPU 0 | CPU 1
661 * |
662 * cnt1 timestamps | cnt2 timestamps
663 * 1 | 2
664 * 2 | 3
665 * - | 4 <--- max recorded
666 *
667 * ============ PASS n + 1 ==============
668 * CPU 0 | CPU 1
669 * |
670 * cnt1 timestamps | cnt2 timestamps
671 * 3 | 5
672 * 4 | 6
673 * 5 | 7 <---- max recorded
674 *
675 * Flush every events below timestamp 4
676 *
677 * ============ PASS n + 2 ==============
678 * CPU 0 | CPU 1
679 * |
680 * cnt1 timestamps | cnt2 timestamps
681 * 6 | 8
682 * 7 | 9
683 * - | 10
684 *
685 * Flush every events below timestamp 7
686 * etc...
687 */
688 static int process_finished_round(struct perf_tool *tool __maybe_unused,
689 union perf_event *event __maybe_unused,
690 struct ordered_events *oe)
691 {
692 if (dump_trace)
693 fprintf(stdout, "\n");
694 return ordered_events__flush(oe, OE_FLUSH__ROUND);
695 }
696
697 int perf_session__queue_event(struct perf_session *s, union perf_event *event,
698 struct perf_sample *sample, u64 file_offset)
699 {
700 return ordered_events__queue(&s->ordered_events, event, sample, file_offset);
701 }
702
703 static void callchain__lbr_callstack_printf(struct perf_sample *sample)
704 {
705 struct ip_callchain *callchain = sample->callchain;
706 struct branch_stack *lbr_stack = sample->branch_stack;
707 u64 kernel_callchain_nr = callchain->nr;
708 unsigned int i;
709
710 for (i = 0; i < kernel_callchain_nr; i++) {
711 if (callchain->ips[i] == PERF_CONTEXT_USER)
712 break;
713 }
714
715 if ((i != kernel_callchain_nr) && lbr_stack->nr) {
716 u64 total_nr;
717 /*
718 * LBR callstack can only get user call chain,
719 * i is kernel call chain number,
720 * 1 is PERF_CONTEXT_USER.
721 *
722 * The user call chain is stored in LBR registers.
723 * LBR are pair registers. The caller is stored
724 * in "from" register, while the callee is stored
725 * in "to" register.
726 * For example, there is a call stack
727 * "A"->"B"->"C"->"D".
728 * The LBR registers will recorde like
729 * "C"->"D", "B"->"C", "A"->"B".
730 * So only the first "to" register and all "from"
731 * registers are needed to construct the whole stack.
732 */
733 total_nr = i + 1 + lbr_stack->nr + 1;
734 kernel_callchain_nr = i + 1;
735
736 printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr);
737
738 for (i = 0; i < kernel_callchain_nr; i++)
739 printf("..... %2d: %016" PRIx64 "\n",
740 i, callchain->ips[i]);
741
742 printf("..... %2d: %016" PRIx64 "\n",
743 (int)(kernel_callchain_nr), lbr_stack->entries[0].to);
744 for (i = 0; i < lbr_stack->nr; i++)
745 printf("..... %2d: %016" PRIx64 "\n",
746 (int)(i + kernel_callchain_nr + 1), lbr_stack->entries[i].from);
747 }
748 }
749
750 static void callchain__printf(struct perf_evsel *evsel,
751 struct perf_sample *sample)
752 {
753 unsigned int i;
754 struct ip_callchain *callchain = sample->callchain;
755
756 if (has_branch_callstack(evsel))
757 callchain__lbr_callstack_printf(sample);
758
759 printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr);
760
761 for (i = 0; i < callchain->nr; i++)
762 printf("..... %2d: %016" PRIx64 "\n",
763 i, callchain->ips[i]);
764 }
765
766 static void branch_stack__printf(struct perf_sample *sample)
767 {
768 uint64_t i;
769
770 printf("... branch stack: nr:%" PRIu64 "\n", sample->branch_stack->nr);
771
772 for (i = 0; i < sample->branch_stack->nr; i++) {
773 struct branch_entry *e = &sample->branch_stack->entries[i];
774
775 printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x\n",
776 i, e->from, e->to,
777 e->flags.cycles,
778 e->flags.mispred ? "M" : " ",
779 e->flags.predicted ? "P" : " ",
780 e->flags.abort ? "A" : " ",
781 e->flags.in_tx ? "T" : " ",
782 (unsigned)e->flags.reserved);
783 }
784 }
785
786 static void regs_dump__printf(u64 mask, u64 *regs)
787 {
788 unsigned rid, i = 0;
789
790 for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) {
791 u64 val = regs[i++];
792
793 printf(".... %-5s 0x%" PRIx64 "\n",
794 perf_reg_name(rid), val);
795 }
796 }
797
798 static const char *regs_abi[] = {
799 [PERF_SAMPLE_REGS_ABI_NONE] = "none",
800 [PERF_SAMPLE_REGS_ABI_32] = "32-bit",
801 [PERF_SAMPLE_REGS_ABI_64] = "64-bit",
802 };
803
804 static inline const char *regs_dump_abi(struct regs_dump *d)
805 {
806 if (d->abi > PERF_SAMPLE_REGS_ABI_64)
807 return "unknown";
808
809 return regs_abi[d->abi];
810 }
811
812 static void regs__printf(const char *type, struct regs_dump *regs)
813 {
814 u64 mask = regs->mask;
815
816 printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n",
817 type,
818 mask,
819 regs_dump_abi(regs));
820
821 regs_dump__printf(mask, regs->regs);
822 }
823
824 static void regs_user__printf(struct perf_sample *sample)
825 {
826 struct regs_dump *user_regs = &sample->user_regs;
827
828 if (user_regs->regs)
829 regs__printf("user", user_regs);
830 }
831
832 static void regs_intr__printf(struct perf_sample *sample)
833 {
834 struct regs_dump *intr_regs = &sample->intr_regs;
835
836 if (intr_regs->regs)
837 regs__printf("intr", intr_regs);
838 }
839
840 static void stack_user__printf(struct stack_dump *dump)
841 {
842 printf("... ustack: size %" PRIu64 ", offset 0x%x\n",
843 dump->size, dump->offset);
844 }
845
846 static void perf_evlist__print_tstamp(struct perf_evlist *evlist,
847 union perf_event *event,
848 struct perf_sample *sample)
849 {
850 u64 sample_type = __perf_evlist__combined_sample_type(evlist);
851
852 if (event->header.type != PERF_RECORD_SAMPLE &&
853 !perf_evlist__sample_id_all(evlist)) {
854 fputs("-1 -1 ", stdout);
855 return;
856 }
857
858 if ((sample_type & PERF_SAMPLE_CPU))
859 printf("%u ", sample->cpu);
860
861 if (sample_type & PERF_SAMPLE_TIME)
862 printf("%" PRIu64 " ", sample->time);
863 }
864
865 static void sample_read__printf(struct perf_sample *sample, u64 read_format)
866 {
867 printf("... sample_read:\n");
868
869 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
870 printf("...... time enabled %016" PRIx64 "\n",
871 sample->read.time_enabled);
872
873 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
874 printf("...... time running %016" PRIx64 "\n",
875 sample->read.time_running);
876
877 if (read_format & PERF_FORMAT_GROUP) {
878 u64 i;
879
880 printf(".... group nr %" PRIu64 "\n", sample->read.group.nr);
881
882 for (i = 0; i < sample->read.group.nr; i++) {
883 struct sample_read_value *value;
884
885 value = &sample->read.group.values[i];
886 printf("..... id %016" PRIx64
887 ", value %016" PRIx64 "\n",
888 value->id, value->value);
889 }
890 } else
891 printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n",
892 sample->read.one.id, sample->read.one.value);
893 }
894
895 static void dump_event(struct perf_evlist *evlist, union perf_event *event,
896 u64 file_offset, struct perf_sample *sample)
897 {
898 if (!dump_trace)
899 return;
900
901 printf("\n%#" PRIx64 " [%#x]: event: %d\n",
902 file_offset, event->header.size, event->header.type);
903
904 trace_event(event);
905
906 if (sample)
907 perf_evlist__print_tstamp(evlist, event, sample);
908
909 printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset,
910 event->header.size, perf_event__name(event->header.type));
911 }
912
913 static void dump_sample(struct perf_evsel *evsel, union perf_event *event,
914 struct perf_sample *sample)
915 {
916 u64 sample_type;
917
918 if (!dump_trace)
919 return;
920
921 printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n",
922 event->header.misc, sample->pid, sample->tid, sample->ip,
923 sample->period, sample->addr);
924
925 sample_type = evsel->attr.sample_type;
926
927 if (sample_type & PERF_SAMPLE_CALLCHAIN)
928 callchain__printf(evsel, sample);
929
930 if ((sample_type & PERF_SAMPLE_BRANCH_STACK) && !has_branch_callstack(evsel))
931 branch_stack__printf(sample);
932
933 if (sample_type & PERF_SAMPLE_REGS_USER)
934 regs_user__printf(sample);
935
936 if (sample_type & PERF_SAMPLE_REGS_INTR)
937 regs_intr__printf(sample);
938
939 if (sample_type & PERF_SAMPLE_STACK_USER)
940 stack_user__printf(&sample->user_stack);
941
942 if (sample_type & PERF_SAMPLE_WEIGHT)
943 printf("... weight: %" PRIu64 "\n", sample->weight);
944
945 if (sample_type & PERF_SAMPLE_DATA_SRC)
946 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src);
947
948 if (sample_type & PERF_SAMPLE_TRANSACTION)
949 printf("... transaction: %" PRIx64 "\n", sample->transaction);
950
951 if (sample_type & PERF_SAMPLE_READ)
952 sample_read__printf(sample, evsel->attr.read_format);
953 }
954
955 static struct machine *machines__find_for_cpumode(struct machines *machines,
956 union perf_event *event,
957 struct perf_sample *sample)
958 {
959 const u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
960 struct machine *machine;
961
962 if (perf_guest &&
963 ((cpumode == PERF_RECORD_MISC_GUEST_KERNEL) ||
964 (cpumode == PERF_RECORD_MISC_GUEST_USER))) {
965 u32 pid;
966
967 if (event->header.type == PERF_RECORD_MMAP
968 || event->header.type == PERF_RECORD_MMAP2)
969 pid = event->mmap.pid;
970 else
971 pid = sample->pid;
972
973 machine = machines__find(machines, pid);
974 if (!machine)
975 machine = machines__find(machines, DEFAULT_GUEST_KERNEL_ID);
976 return machine;
977 }
978
979 return &machines->host;
980 }
981
982 static int deliver_sample_value(struct perf_evlist *evlist,
983 struct perf_tool *tool,
984 union perf_event *event,
985 struct perf_sample *sample,
986 struct sample_read_value *v,
987 struct machine *machine)
988 {
989 struct perf_sample_id *sid = perf_evlist__id2sid(evlist, v->id);
990
991 if (sid) {
992 sample->id = v->id;
993 sample->period = v->value - sid->period;
994 sid->period = v->value;
995 }
996
997 if (!sid || sid->evsel == NULL) {
998 ++evlist->stats.nr_unknown_id;
999 return 0;
1000 }
1001
1002 return tool->sample(tool, event, sample, sid->evsel, machine);
1003 }
1004
1005 static int deliver_sample_group(struct perf_evlist *evlist,
1006 struct perf_tool *tool,
1007 union perf_event *event,
1008 struct perf_sample *sample,
1009 struct machine *machine)
1010 {
1011 int ret = -EINVAL;
1012 u64 i;
1013
1014 for (i = 0; i < sample->read.group.nr; i++) {
1015 ret = deliver_sample_value(evlist, tool, event, sample,
1016 &sample->read.group.values[i],
1017 machine);
1018 if (ret)
1019 break;
1020 }
1021
1022 return ret;
1023 }
1024
1025 static int
1026 perf_evlist__deliver_sample(struct perf_evlist *evlist,
1027 struct perf_tool *tool,
1028 union perf_event *event,
1029 struct perf_sample *sample,
1030 struct perf_evsel *evsel,
1031 struct machine *machine)
1032 {
1033 /* We know evsel != NULL. */
1034 u64 sample_type = evsel->attr.sample_type;
1035 u64 read_format = evsel->attr.read_format;
1036
1037 /* Standard sample delievery. */
1038 if (!(sample_type & PERF_SAMPLE_READ))
1039 return tool->sample(tool, event, sample, evsel, machine);
1040
1041 /* For PERF_SAMPLE_READ we have either single or group mode. */
1042 if (read_format & PERF_FORMAT_GROUP)
1043 return deliver_sample_group(evlist, tool, event, sample,
1044 machine);
1045 else
1046 return deliver_sample_value(evlist, tool, event, sample,
1047 &sample->read.one, machine);
1048 }
1049
1050 static int machines__deliver_event(struct machines *machines,
1051 struct perf_evlist *evlist,
1052 union perf_event *event,
1053 struct perf_sample *sample,
1054 struct perf_tool *tool, u64 file_offset)
1055 {
1056 struct perf_evsel *evsel;
1057 struct machine *machine;
1058
1059 dump_event(evlist, event, file_offset, sample);
1060
1061 evsel = perf_evlist__id2evsel(evlist, sample->id);
1062
1063 machine = machines__find_for_cpumode(machines, event, sample);
1064
1065 switch (event->header.type) {
1066 case PERF_RECORD_SAMPLE:
1067 if (evsel == NULL) {
1068 ++evlist->stats.nr_unknown_id;
1069 return 0;
1070 }
1071 dump_sample(evsel, event, sample);
1072 if (machine == NULL) {
1073 ++evlist->stats.nr_unprocessable_samples;
1074 return 0;
1075 }
1076 return perf_evlist__deliver_sample(evlist, tool, event, sample, evsel, machine);
1077 case PERF_RECORD_MMAP:
1078 return tool->mmap(tool, event, sample, machine);
1079 case PERF_RECORD_MMAP2:
1080 if (event->header.misc & PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT)
1081 ++evlist->stats.nr_proc_map_timeout;
1082 return tool->mmap2(tool, event, sample, machine);
1083 case PERF_RECORD_COMM:
1084 return tool->comm(tool, event, sample, machine);
1085 case PERF_RECORD_FORK:
1086 return tool->fork(tool, event, sample, machine);
1087 case PERF_RECORD_EXIT:
1088 return tool->exit(tool, event, sample, machine);
1089 case PERF_RECORD_LOST:
1090 if (tool->lost == perf_event__process_lost)
1091 evlist->stats.total_lost += event->lost.lost;
1092 return tool->lost(tool, event, sample, machine);
1093 case PERF_RECORD_LOST_SAMPLES:
1094 if (tool->lost_samples == perf_event__process_lost_samples)
1095 evlist->stats.total_lost_samples += event->lost_samples.lost;
1096 return tool->lost_samples(tool, event, sample, machine);
1097 case PERF_RECORD_READ:
1098 return tool->read(tool, event, sample, evsel, machine);
1099 case PERF_RECORD_THROTTLE:
1100 return tool->throttle(tool, event, sample, machine);
1101 case PERF_RECORD_UNTHROTTLE:
1102 return tool->unthrottle(tool, event, sample, machine);
1103 case PERF_RECORD_AUX:
1104 if (tool->aux == perf_event__process_aux &&
1105 (event->aux.flags & PERF_AUX_FLAG_TRUNCATED))
1106 evlist->stats.total_aux_lost += 1;
1107 return tool->aux(tool, event, sample, machine);
1108 case PERF_RECORD_ITRACE_START:
1109 return tool->itrace_start(tool, event, sample, machine);
1110 case PERF_RECORD_SWITCH:
1111 case PERF_RECORD_SWITCH_CPU_WIDE:
1112 return tool->context_switch(tool, event, sample, machine);
1113 default:
1114 ++evlist->stats.nr_unknown_events;
1115 return -1;
1116 }
1117 }
1118
1119 static int perf_session__deliver_event(struct perf_session *session,
1120 union perf_event *event,
1121 struct perf_sample *sample,
1122 struct perf_tool *tool,
1123 u64 file_offset)
1124 {
1125 int ret;
1126
1127 ret = auxtrace__process_event(session, event, sample, tool);
1128 if (ret < 0)
1129 return ret;
1130 if (ret > 0)
1131 return 0;
1132
1133 return machines__deliver_event(&session->machines, session->evlist,
1134 event, sample, tool, file_offset);
1135 }
1136
1137 static s64 perf_session__process_user_event(struct perf_session *session,
1138 union perf_event *event,
1139 u64 file_offset)
1140 {
1141 struct ordered_events *oe = &session->ordered_events;
1142 struct perf_tool *tool = session->tool;
1143 int fd = perf_data_file__fd(session->file);
1144 int err;
1145
1146 dump_event(session->evlist, event, file_offset, NULL);
1147
1148 /* These events are processed right away */
1149 switch (event->header.type) {
1150 case PERF_RECORD_HEADER_ATTR:
1151 err = tool->attr(tool, event, &session->evlist);
1152 if (err == 0) {
1153 perf_session__set_id_hdr_size(session);
1154 perf_session__set_comm_exec(session);
1155 }
1156 return err;
1157 case PERF_RECORD_HEADER_EVENT_TYPE:
1158 /*
1159 * Depreceated, but we need to handle it for sake
1160 * of old data files create in pipe mode.
1161 */
1162 return 0;
1163 case PERF_RECORD_HEADER_TRACING_DATA:
1164 /* setup for reading amidst mmap */
1165 lseek(fd, file_offset, SEEK_SET);
1166 return tool->tracing_data(tool, event, session);
1167 case PERF_RECORD_HEADER_BUILD_ID:
1168 return tool->build_id(tool, event, session);
1169 case PERF_RECORD_FINISHED_ROUND:
1170 return tool->finished_round(tool, event, oe);
1171 case PERF_RECORD_ID_INDEX:
1172 return tool->id_index(tool, event, session);
1173 case PERF_RECORD_AUXTRACE_INFO:
1174 return tool->auxtrace_info(tool, event, session);
1175 case PERF_RECORD_AUXTRACE:
1176 /* setup for reading amidst mmap */
1177 lseek(fd, file_offset + event->header.size, SEEK_SET);
1178 return tool->auxtrace(tool, event, session);
1179 case PERF_RECORD_AUXTRACE_ERROR:
1180 perf_session__auxtrace_error_inc(session, event);
1181 return tool->auxtrace_error(tool, event, session);
1182 default:
1183 return -EINVAL;
1184 }
1185 }
1186
1187 int perf_session__deliver_synth_event(struct perf_session *session,
1188 union perf_event *event,
1189 struct perf_sample *sample)
1190 {
1191 struct perf_evlist *evlist = session->evlist;
1192 struct perf_tool *tool = session->tool;
1193
1194 events_stats__inc(&evlist->stats, event->header.type);
1195
1196 if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1197 return perf_session__process_user_event(session, event, 0);
1198
1199 return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0);
1200 }
1201
1202 static void event_swap(union perf_event *event, bool sample_id_all)
1203 {
1204 perf_event__swap_op swap;
1205
1206 swap = perf_event__swap_ops[event->header.type];
1207 if (swap)
1208 swap(event, sample_id_all);
1209 }
1210
1211 int perf_session__peek_event(struct perf_session *session, off_t file_offset,
1212 void *buf, size_t buf_sz,
1213 union perf_event **event_ptr,
1214 struct perf_sample *sample)
1215 {
1216 union perf_event *event;
1217 size_t hdr_sz, rest;
1218 int fd;
1219
1220 if (session->one_mmap && !session->header.needs_swap) {
1221 event = file_offset - session->one_mmap_offset +
1222 session->one_mmap_addr;
1223 goto out_parse_sample;
1224 }
1225
1226 if (perf_data_file__is_pipe(session->file))
1227 return -1;
1228
1229 fd = perf_data_file__fd(session->file);
1230 hdr_sz = sizeof(struct perf_event_header);
1231
1232 if (buf_sz < hdr_sz)
1233 return -1;
1234
1235 if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 ||
1236 readn(fd, buf, hdr_sz) != (ssize_t)hdr_sz)
1237 return -1;
1238
1239 event = (union perf_event *)buf;
1240
1241 if (session->header.needs_swap)
1242 perf_event_header__bswap(&event->header);
1243
1244 if (event->header.size < hdr_sz || event->header.size > buf_sz)
1245 return -1;
1246
1247 rest = event->header.size - hdr_sz;
1248
1249 if (readn(fd, buf, rest) != (ssize_t)rest)
1250 return -1;
1251
1252 if (session->header.needs_swap)
1253 event_swap(event, perf_evlist__sample_id_all(session->evlist));
1254
1255 out_parse_sample:
1256
1257 if (sample && event->header.type < PERF_RECORD_USER_TYPE_START &&
1258 perf_evlist__parse_sample(session->evlist, event, sample))
1259 return -1;
1260
1261 *event_ptr = event;
1262
1263 return 0;
1264 }
1265
1266 static s64 perf_session__process_event(struct perf_session *session,
1267 union perf_event *event, u64 file_offset)
1268 {
1269 struct perf_evlist *evlist = session->evlist;
1270 struct perf_tool *tool = session->tool;
1271 struct perf_sample sample;
1272 int ret;
1273
1274 if (session->header.needs_swap)
1275 event_swap(event, perf_evlist__sample_id_all(evlist));
1276
1277 if (event->header.type >= PERF_RECORD_HEADER_MAX)
1278 return -EINVAL;
1279
1280 events_stats__inc(&evlist->stats, event->header.type);
1281
1282 if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1283 return perf_session__process_user_event(session, event, file_offset);
1284
1285 /*
1286 * For all kernel events we get the sample data
1287 */
1288 ret = perf_evlist__parse_sample(evlist, event, &sample);
1289 if (ret)
1290 return ret;
1291
1292 if (tool->ordered_events) {
1293 ret = perf_session__queue_event(session, event, &sample, file_offset);
1294 if (ret != -ETIME)
1295 return ret;
1296 }
1297
1298 return perf_session__deliver_event(session, event, &sample, tool,
1299 file_offset);
1300 }
1301
1302 void perf_event_header__bswap(struct perf_event_header *hdr)
1303 {
1304 hdr->type = bswap_32(hdr->type);
1305 hdr->misc = bswap_16(hdr->misc);
1306 hdr->size = bswap_16(hdr->size);
1307 }
1308
1309 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid)
1310 {
1311 return machine__findnew_thread(&session->machines.host, -1, pid);
1312 }
1313
1314 struct thread *perf_session__register_idle_thread(struct perf_session *session)
1315 {
1316 struct thread *thread;
1317
1318 thread = machine__findnew_thread(&session->machines.host, 0, 0);
1319 if (thread == NULL || thread__set_comm(thread, "swapper", 0)) {
1320 pr_err("problem inserting idle task.\n");
1321 thread = NULL;
1322 }
1323
1324 return thread;
1325 }
1326
1327 static void perf_session__warn_about_errors(const struct perf_session *session)
1328 {
1329 const struct events_stats *stats = &session->evlist->stats;
1330 const struct ordered_events *oe = &session->ordered_events;
1331
1332 if (session->tool->lost == perf_event__process_lost &&
1333 stats->nr_events[PERF_RECORD_LOST] != 0) {
1334 ui__warning("Processed %d events and lost %d chunks!\n\n"
1335 "Check IO/CPU overload!\n\n",
1336 stats->nr_events[0],
1337 stats->nr_events[PERF_RECORD_LOST]);
1338 }
1339
1340 if (session->tool->lost_samples == perf_event__process_lost_samples) {
1341 double drop_rate;
1342
1343 drop_rate = (double)stats->total_lost_samples /
1344 (double) (stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples);
1345 if (drop_rate > 0.05) {
1346 ui__warning("Processed %" PRIu64 " samples and lost %3.2f%% samples!\n\n",
1347 stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples,
1348 drop_rate * 100.0);
1349 }
1350 }
1351
1352 if (session->tool->aux == perf_event__process_aux &&
1353 stats->total_aux_lost != 0) {
1354 ui__warning("AUX data lost %" PRIu64 " times out of %u!\n\n",
1355 stats->total_aux_lost,
1356 stats->nr_events[PERF_RECORD_AUX]);
1357 }
1358
1359 if (stats->nr_unknown_events != 0) {
1360 ui__warning("Found %u unknown events!\n\n"
1361 "Is this an older tool processing a perf.data "
1362 "file generated by a more recent tool?\n\n"
1363 "If that is not the case, consider "
1364 "reporting to linux-kernel@vger.kernel.org.\n\n",
1365 stats->nr_unknown_events);
1366 }
1367
1368 if (stats->nr_unknown_id != 0) {
1369 ui__warning("%u samples with id not present in the header\n",
1370 stats->nr_unknown_id);
1371 }
1372
1373 if (stats->nr_invalid_chains != 0) {
1374 ui__warning("Found invalid callchains!\n\n"
1375 "%u out of %u events were discarded for this reason.\n\n"
1376 "Consider reporting to linux-kernel@vger.kernel.org.\n\n",
1377 stats->nr_invalid_chains,
1378 stats->nr_events[PERF_RECORD_SAMPLE]);
1379 }
1380
1381 if (stats->nr_unprocessable_samples != 0) {
1382 ui__warning("%u unprocessable samples recorded.\n"
1383 "Do you have a KVM guest running and not using 'perf kvm'?\n",
1384 stats->nr_unprocessable_samples);
1385 }
1386
1387 if (oe->nr_unordered_events != 0)
1388 ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events);
1389
1390 events_stats__auxtrace_error_warn(stats);
1391
1392 if (stats->nr_proc_map_timeout != 0) {
1393 ui__warning("%d map information files for pre-existing threads were\n"
1394 "not processed, if there are samples for addresses they\n"
1395 "will not be resolved, you may find out which are these\n"
1396 "threads by running with -v and redirecting the output\n"
1397 "to a file.\n"
1398 "The time limit to process proc map is too short?\n"
1399 "Increase it by --proc-map-timeout\n",
1400 stats->nr_proc_map_timeout);
1401 }
1402 }
1403
1404 static int perf_session__flush_thread_stack(struct thread *thread,
1405 void *p __maybe_unused)
1406 {
1407 return thread_stack__flush(thread);
1408 }
1409
1410 static int perf_session__flush_thread_stacks(struct perf_session *session)
1411 {
1412 return machines__for_each_thread(&session->machines,
1413 perf_session__flush_thread_stack,
1414 NULL);
1415 }
1416
1417 volatile int session_done;
1418
1419 static int __perf_session__process_pipe_events(struct perf_session *session)
1420 {
1421 struct ordered_events *oe = &session->ordered_events;
1422 struct perf_tool *tool = session->tool;
1423 int fd = perf_data_file__fd(session->file);
1424 union perf_event *event;
1425 uint32_t size, cur_size = 0;
1426 void *buf = NULL;
1427 s64 skip = 0;
1428 u64 head;
1429 ssize_t err;
1430 void *p;
1431
1432 perf_tool__fill_defaults(tool);
1433
1434 head = 0;
1435 cur_size = sizeof(union perf_event);
1436
1437 buf = malloc(cur_size);
1438 if (!buf)
1439 return -errno;
1440 more:
1441 event = buf;
1442 err = readn(fd, event, sizeof(struct perf_event_header));
1443 if (err <= 0) {
1444 if (err == 0)
1445 goto done;
1446
1447 pr_err("failed to read event header\n");
1448 goto out_err;
1449 }
1450
1451 if (session->header.needs_swap)
1452 perf_event_header__bswap(&event->header);
1453
1454 size = event->header.size;
1455 if (size < sizeof(struct perf_event_header)) {
1456 pr_err("bad event header size\n");
1457 goto out_err;
1458 }
1459
1460 if (size > cur_size) {
1461 void *new = realloc(buf, size);
1462 if (!new) {
1463 pr_err("failed to allocate memory to read event\n");
1464 goto out_err;
1465 }
1466 buf = new;
1467 cur_size = size;
1468 event = buf;
1469 }
1470 p = event;
1471 p += sizeof(struct perf_event_header);
1472
1473 if (size - sizeof(struct perf_event_header)) {
1474 err = readn(fd, p, size - sizeof(struct perf_event_header));
1475 if (err <= 0) {
1476 if (err == 0) {
1477 pr_err("unexpected end of event stream\n");
1478 goto done;
1479 }
1480
1481 pr_err("failed to read event data\n");
1482 goto out_err;
1483 }
1484 }
1485
1486 if ((skip = perf_session__process_event(session, event, head)) < 0) {
1487 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1488 head, event->header.size, event->header.type);
1489 err = -EINVAL;
1490 goto out_err;
1491 }
1492
1493 head += size;
1494
1495 if (skip > 0)
1496 head += skip;
1497
1498 if (!session_done())
1499 goto more;
1500 done:
1501 /* do the final flush for ordered samples */
1502 err = ordered_events__flush(oe, OE_FLUSH__FINAL);
1503 if (err)
1504 goto out_err;
1505 err = auxtrace__flush_events(session, tool);
1506 if (err)
1507 goto out_err;
1508 err = perf_session__flush_thread_stacks(session);
1509 out_err:
1510 free(buf);
1511 perf_session__warn_about_errors(session);
1512 ordered_events__free(&session->ordered_events);
1513 auxtrace__free_events(session);
1514 return err;
1515 }
1516
1517 static union perf_event *
1518 fetch_mmaped_event(struct perf_session *session,
1519 u64 head, size_t mmap_size, char *buf)
1520 {
1521 union perf_event *event;
1522
1523 /*
1524 * Ensure we have enough space remaining to read
1525 * the size of the event in the headers.
1526 */
1527 if (head + sizeof(event->header) > mmap_size)
1528 return NULL;
1529
1530 event = (union perf_event *)(buf + head);
1531
1532 if (session->header.needs_swap)
1533 perf_event_header__bswap(&event->header);
1534
1535 if (head + event->header.size > mmap_size) {
1536 /* We're not fetching the event so swap back again */
1537 if (session->header.needs_swap)
1538 perf_event_header__bswap(&event->header);
1539 return NULL;
1540 }
1541
1542 return event;
1543 }
1544
1545 /*
1546 * On 64bit we can mmap the data file in one go. No need for tiny mmap
1547 * slices. On 32bit we use 32MB.
1548 */
1549 #if BITS_PER_LONG == 64
1550 #define MMAP_SIZE ULLONG_MAX
1551 #define NUM_MMAPS 1
1552 #else
1553 #define MMAP_SIZE (32 * 1024 * 1024ULL)
1554 #define NUM_MMAPS 128
1555 #endif
1556
1557 static int __perf_session__process_events(struct perf_session *session,
1558 u64 data_offset, u64 data_size,
1559 u64 file_size)
1560 {
1561 struct ordered_events *oe = &session->ordered_events;
1562 struct perf_tool *tool = session->tool;
1563 int fd = perf_data_file__fd(session->file);
1564 u64 head, page_offset, file_offset, file_pos, size;
1565 int err, mmap_prot, mmap_flags, map_idx = 0;
1566 size_t mmap_size;
1567 char *buf, *mmaps[NUM_MMAPS];
1568 union perf_event *event;
1569 struct ui_progress prog;
1570 s64 skip;
1571
1572 perf_tool__fill_defaults(tool);
1573
1574 page_offset = page_size * (data_offset / page_size);
1575 file_offset = page_offset;
1576 head = data_offset - page_offset;
1577
1578 if (data_size == 0)
1579 goto out;
1580
1581 if (data_offset + data_size < file_size)
1582 file_size = data_offset + data_size;
1583
1584 ui_progress__init(&prog, file_size, "Processing events...");
1585
1586 mmap_size = MMAP_SIZE;
1587 if (mmap_size > file_size) {
1588 mmap_size = file_size;
1589 session->one_mmap = true;
1590 }
1591
1592 memset(mmaps, 0, sizeof(mmaps));
1593
1594 mmap_prot = PROT_READ;
1595 mmap_flags = MAP_SHARED;
1596
1597 if (session->header.needs_swap) {
1598 mmap_prot |= PROT_WRITE;
1599 mmap_flags = MAP_PRIVATE;
1600 }
1601 remap:
1602 buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, fd,
1603 file_offset);
1604 if (buf == MAP_FAILED) {
1605 pr_err("failed to mmap file\n");
1606 err = -errno;
1607 goto out_err;
1608 }
1609 mmaps[map_idx] = buf;
1610 map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1);
1611 file_pos = file_offset + head;
1612 if (session->one_mmap) {
1613 session->one_mmap_addr = buf;
1614 session->one_mmap_offset = file_offset;
1615 }
1616
1617 more:
1618 event = fetch_mmaped_event(session, head, mmap_size, buf);
1619 if (!event) {
1620 if (mmaps[map_idx]) {
1621 munmap(mmaps[map_idx], mmap_size);
1622 mmaps[map_idx] = NULL;
1623 }
1624
1625 page_offset = page_size * (head / page_size);
1626 file_offset += page_offset;
1627 head -= page_offset;
1628 goto remap;
1629 }
1630
1631 size = event->header.size;
1632
1633 if (size < sizeof(struct perf_event_header) ||
1634 (skip = perf_session__process_event(session, event, file_pos)) < 0) {
1635 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1636 file_offset + head, event->header.size,
1637 event->header.type);
1638 err = -EINVAL;
1639 goto out_err;
1640 }
1641
1642 if (skip)
1643 size += skip;
1644
1645 head += size;
1646 file_pos += size;
1647
1648 ui_progress__update(&prog, size);
1649
1650 if (session_done())
1651 goto out;
1652
1653 if (file_pos < file_size)
1654 goto more;
1655
1656 out:
1657 /* do the final flush for ordered samples */
1658 err = ordered_events__flush(oe, OE_FLUSH__FINAL);
1659 if (err)
1660 goto out_err;
1661 err = auxtrace__flush_events(session, tool);
1662 if (err)
1663 goto out_err;
1664 err = perf_session__flush_thread_stacks(session);
1665 out_err:
1666 ui_progress__finish();
1667 perf_session__warn_about_errors(session);
1668 ordered_events__free(&session->ordered_events);
1669 auxtrace__free_events(session);
1670 session->one_mmap = false;
1671 return err;
1672 }
1673
1674 int perf_session__process_events(struct perf_session *session)
1675 {
1676 u64 size = perf_data_file__size(session->file);
1677 int err;
1678
1679 if (perf_session__register_idle_thread(session) == NULL)
1680 return -ENOMEM;
1681
1682 if (!perf_data_file__is_pipe(session->file))
1683 err = __perf_session__process_events(session,
1684 session->header.data_offset,
1685 session->header.data_size, size);
1686 else
1687 err = __perf_session__process_pipe_events(session);
1688
1689 return err;
1690 }
1691
1692 bool perf_session__has_traces(struct perf_session *session, const char *msg)
1693 {
1694 struct perf_evsel *evsel;
1695
1696 evlist__for_each(session->evlist, evsel) {
1697 if (evsel->attr.type == PERF_TYPE_TRACEPOINT)
1698 return true;
1699 }
1700
1701 pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg);
1702 return false;
1703 }
1704
1705 int maps__set_kallsyms_ref_reloc_sym(struct map **maps,
1706 const char *symbol_name, u64 addr)
1707 {
1708 char *bracket;
1709 enum map_type i;
1710 struct ref_reloc_sym *ref;
1711
1712 ref = zalloc(sizeof(struct ref_reloc_sym));
1713 if (ref == NULL)
1714 return -ENOMEM;
1715
1716 ref->name = strdup(symbol_name);
1717 if (ref->name == NULL) {
1718 free(ref);
1719 return -ENOMEM;
1720 }
1721
1722 bracket = strchr(ref->name, ']');
1723 if (bracket)
1724 *bracket = '\0';
1725
1726 ref->addr = addr;
1727
1728 for (i = 0; i < MAP__NR_TYPES; ++i) {
1729 struct kmap *kmap = map__kmap(maps[i]);
1730
1731 if (!kmap)
1732 continue;
1733 kmap->ref_reloc_sym = ref;
1734 }
1735
1736 return 0;
1737 }
1738
1739 size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp)
1740 {
1741 return machines__fprintf_dsos(&session->machines, fp);
1742 }
1743
1744 size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp,
1745 bool (skip)(struct dso *dso, int parm), int parm)
1746 {
1747 return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm);
1748 }
1749
1750 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp)
1751 {
1752 size_t ret;
1753 const char *msg = "";
1754
1755 if (perf_header__has_feat(&session->header, HEADER_AUXTRACE))
1756 msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)";
1757
1758 ret = fprintf(fp, "\nAggregated stats:%s\n", msg);
1759
1760 ret += events_stats__fprintf(&session->evlist->stats, fp);
1761 return ret;
1762 }
1763
1764 size_t perf_session__fprintf(struct perf_session *session, FILE *fp)
1765 {
1766 /*
1767 * FIXME: Here we have to actually print all the machines in this
1768 * session, not just the host...
1769 */
1770 return machine__fprintf(&session->machines.host, fp);
1771 }
1772
1773 struct perf_evsel *perf_session__find_first_evtype(struct perf_session *session,
1774 unsigned int type)
1775 {
1776 struct perf_evsel *pos;
1777
1778 evlist__for_each(session->evlist, pos) {
1779 if (pos->attr.type == type)
1780 return pos;
1781 }
1782 return NULL;
1783 }
1784
1785 void perf_evsel__print_ip(struct perf_evsel *evsel, struct perf_sample *sample,
1786 struct addr_location *al,
1787 unsigned int print_opts, unsigned int stack_depth)
1788 {
1789 struct callchain_cursor_node *node;
1790 int print_ip = print_opts & PRINT_IP_OPT_IP;
1791 int print_sym = print_opts & PRINT_IP_OPT_SYM;
1792 int print_dso = print_opts & PRINT_IP_OPT_DSO;
1793 int print_symoffset = print_opts & PRINT_IP_OPT_SYMOFFSET;
1794 int print_oneline = print_opts & PRINT_IP_OPT_ONELINE;
1795 int print_srcline = print_opts & PRINT_IP_OPT_SRCLINE;
1796 char s = print_oneline ? ' ' : '\t';
1797
1798 if (symbol_conf.use_callchain && sample->callchain) {
1799 struct addr_location node_al;
1800
1801 if (thread__resolve_callchain(al->thread, evsel,
1802 sample, NULL, NULL,
1803 stack_depth) != 0) {
1804 if (verbose)
1805 error("Failed to resolve callchain. Skipping\n");
1806 return;
1807 }
1808 callchain_cursor_commit(&callchain_cursor);
1809
1810 if (print_symoffset)
1811 node_al = *al;
1812
1813 while (stack_depth) {
1814 u64 addr = 0;
1815
1816 node = callchain_cursor_current(&callchain_cursor);
1817 if (!node)
1818 break;
1819
1820 if (node->sym && node->sym->ignore)
1821 goto next;
1822
1823 if (print_ip)
1824 printf("%c%16" PRIx64, s, node->ip);
1825
1826 if (node->map)
1827 addr = node->map->map_ip(node->map, node->ip);
1828
1829 if (print_sym) {
1830 printf(" ");
1831 if (print_symoffset) {
1832 node_al.addr = addr;
1833 node_al.map = node->map;
1834 symbol__fprintf_symname_offs(node->sym, &node_al, stdout);
1835 } else
1836 symbol__fprintf_symname(node->sym, stdout);
1837 }
1838
1839 if (print_dso) {
1840 printf(" (");
1841 map__fprintf_dsoname(node->map, stdout);
1842 printf(")");
1843 }
1844
1845 if (print_srcline)
1846 map__fprintf_srcline(node->map, addr, "\n ",
1847 stdout);
1848
1849 if (!print_oneline)
1850 printf("\n");
1851
1852 stack_depth--;
1853 next:
1854 callchain_cursor_advance(&callchain_cursor);
1855 }
1856
1857 } else {
1858 if (al->sym && al->sym->ignore)
1859 return;
1860
1861 if (print_ip)
1862 printf("%16" PRIx64, sample->ip);
1863
1864 if (print_sym) {
1865 printf(" ");
1866 if (print_symoffset)
1867 symbol__fprintf_symname_offs(al->sym, al,
1868 stdout);
1869 else
1870 symbol__fprintf_symname(al->sym, stdout);
1871 }
1872
1873 if (print_dso) {
1874 printf(" (");
1875 map__fprintf_dsoname(al->map, stdout);
1876 printf(")");
1877 }
1878
1879 if (print_srcline)
1880 map__fprintf_srcline(al->map, al->addr, "\n ", stdout);
1881 }
1882 }
1883
1884 int perf_session__cpu_bitmap(struct perf_session *session,
1885 const char *cpu_list, unsigned long *cpu_bitmap)
1886 {
1887 int i, err = -1;
1888 struct cpu_map *map;
1889
1890 for (i = 0; i < PERF_TYPE_MAX; ++i) {
1891 struct perf_evsel *evsel;
1892
1893 evsel = perf_session__find_first_evtype(session, i);
1894 if (!evsel)
1895 continue;
1896
1897 if (!(evsel->attr.sample_type & PERF_SAMPLE_CPU)) {
1898 pr_err("File does not contain CPU events. "
1899 "Remove -c option to proceed.\n");
1900 return -1;
1901 }
1902 }
1903
1904 map = cpu_map__new(cpu_list);
1905 if (map == NULL) {
1906 pr_err("Invalid cpu_list\n");
1907 return -1;
1908 }
1909
1910 for (i = 0; i < map->nr; i++) {
1911 int cpu = map->map[i];
1912
1913 if (cpu >= MAX_NR_CPUS) {
1914 pr_err("Requested CPU %d too large. "
1915 "Consider raising MAX_NR_CPUS\n", cpu);
1916 goto out_delete_map;
1917 }
1918
1919 set_bit(cpu, cpu_bitmap);
1920 }
1921
1922 err = 0;
1923
1924 out_delete_map:
1925 cpu_map__put(map);
1926 return err;
1927 }
1928
1929 void perf_session__fprintf_info(struct perf_session *session, FILE *fp,
1930 bool full)
1931 {
1932 struct stat st;
1933 int fd, ret;
1934
1935 if (session == NULL || fp == NULL)
1936 return;
1937
1938 fd = perf_data_file__fd(session->file);
1939
1940 ret = fstat(fd, &st);
1941 if (ret == -1)
1942 return;
1943
1944 fprintf(fp, "# ========\n");
1945 fprintf(fp, "# captured on: %s", ctime(&st.st_ctime));
1946 perf_header__fprintf_info(session, fp, full);
1947 fprintf(fp, "# ========\n#\n");
1948 }
1949
1950
1951 int __perf_session__set_tracepoints_handlers(struct perf_session *session,
1952 const struct perf_evsel_str_handler *assocs,
1953 size_t nr_assocs)
1954 {
1955 struct perf_evsel *evsel;
1956 size_t i;
1957 int err;
1958
1959 for (i = 0; i < nr_assocs; i++) {
1960 /*
1961 * Adding a handler for an event not in the session,
1962 * just ignore it.
1963 */
1964 evsel = perf_evlist__find_tracepoint_by_name(session->evlist, assocs[i].name);
1965 if (evsel == NULL)
1966 continue;
1967
1968 err = -EEXIST;
1969 if (evsel->handler != NULL)
1970 goto out;
1971 evsel->handler = assocs[i].handler;
1972 }
1973
1974 err = 0;
1975 out:
1976 return err;
1977 }
1978
1979 int perf_event__process_id_index(struct perf_tool *tool __maybe_unused,
1980 union perf_event *event,
1981 struct perf_session *session)
1982 {
1983 struct perf_evlist *evlist = session->evlist;
1984 struct id_index_event *ie = &event->id_index;
1985 size_t i, nr, max_nr;
1986
1987 max_nr = (ie->header.size - sizeof(struct id_index_event)) /
1988 sizeof(struct id_index_entry);
1989 nr = ie->nr;
1990 if (nr > max_nr)
1991 return -EINVAL;
1992
1993 if (dump_trace)
1994 fprintf(stdout, " nr: %zu\n", nr);
1995
1996 for (i = 0; i < nr; i++) {
1997 struct id_index_entry *e = &ie->entries[i];
1998 struct perf_sample_id *sid;
1999
2000 if (dump_trace) {
2001 fprintf(stdout, " ... id: %"PRIu64, e->id);
2002 fprintf(stdout, " idx: %"PRIu64, e->idx);
2003 fprintf(stdout, " cpu: %"PRId64, e->cpu);
2004 fprintf(stdout, " tid: %"PRId64"\n", e->tid);
2005 }
2006
2007 sid = perf_evlist__id2sid(evlist, e->id);
2008 if (!sid)
2009 return -ENOENT;
2010 sid->idx = e->idx;
2011 sid->cpu = e->cpu;
2012 sid->tid = e->tid;
2013 }
2014 return 0;
2015 }
2016
2017 int perf_event__synthesize_id_index(struct perf_tool *tool,
2018 perf_event__handler_t process,
2019 struct perf_evlist *evlist,
2020 struct machine *machine)
2021 {
2022 union perf_event *ev;
2023 struct perf_evsel *evsel;
2024 size_t nr = 0, i = 0, sz, max_nr, n;
2025 int err;
2026
2027 pr_debug2("Synthesizing id index\n");
2028
2029 max_nr = (UINT16_MAX - sizeof(struct id_index_event)) /
2030 sizeof(struct id_index_entry);
2031
2032 evlist__for_each(evlist, evsel)
2033 nr += evsel->ids;
2034
2035 n = nr > max_nr ? max_nr : nr;
2036 sz = sizeof(struct id_index_event) + n * sizeof(struct id_index_entry);
2037 ev = zalloc(sz);
2038 if (!ev)
2039 return -ENOMEM;
2040
2041 ev->id_index.header.type = PERF_RECORD_ID_INDEX;
2042 ev->id_index.header.size = sz;
2043 ev->id_index.nr = n;
2044
2045 evlist__for_each(evlist, evsel) {
2046 u32 j;
2047
2048 for (j = 0; j < evsel->ids; j++) {
2049 struct id_index_entry *e;
2050 struct perf_sample_id *sid;
2051
2052 if (i >= n) {
2053 err = process(tool, ev, NULL, machine);
2054 if (err)
2055 goto out_err;
2056 nr -= n;
2057 i = 0;
2058 }
2059
2060 e = &ev->id_index.entries[i++];
2061
2062 e->id = evsel->id[j];
2063
2064 sid = perf_evlist__id2sid(evlist, e->id);
2065 if (!sid) {
2066 free(ev);
2067 return -ENOENT;
2068 }
2069
2070 e->idx = sid->idx;
2071 e->cpu = sid->cpu;
2072 e->tid = sid->tid;
2073 }
2074 }
2075
2076 sz = sizeof(struct id_index_event) + nr * sizeof(struct id_index_entry);
2077 ev->id_index.header.size = sz;
2078 ev->id_index.nr = nr;
2079
2080 err = process(tool, ev, NULL, machine);
2081 out_err:
2082 free(ev);
2083
2084 return err;
2085 }
Linux with added FPC-III support