projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branch 'drm-fixes-4.3' of git://people.freedesktop.org/~agd5f/linux into drm...
[deliverable/linux.git] / tools / perf / util / session.c
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)");
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");
41 return -1;
42 }
43
44 if (!perf_evlist__valid_sample_id_all(session->evlist)) {
45 pr_err("non matching sample_id_all");
46 return -1;
47 }
48
49 if (!perf_evlist__valid_read_format(session->evlist)) {
50 pr_err("non matching read_format");
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 }
142
143 if (!file || perf_data_file__is_write(file)) {
144 /*
145 * In O_RDONLY mode this will be performed when reading the
146 * kernel MMAP event, in perf_event__process_mmap().
147 */
148 if (perf_session__create_kernel_maps(session) < 0)
149 pr_warning("Cannot read kernel map\n");
150 }
151
152 if (tool && tool->ordering_requires_timestamps &&
153 tool->ordered_events && !perf_evlist__sample_id_all(session->evlist)) {
154 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n");
155 tool->ordered_events = false;
156 }
157
158 return session;
159
160 out_close:
161 perf_data_file__close(file);
162 out_delete:
163 perf_session__delete(session);
164 out:
165 return NULL;
166 }
167
168 static void perf_session__delete_threads(struct perf_session *session)
169 {
170 machine__delete_threads(&session->machines.host);
171 }
172
173 static void perf_session_env__exit(struct perf_env *env)
174 {
175 zfree(&env->hostname);
176 zfree(&env->os_release);
177 zfree(&env->version);
178 zfree(&env->arch);
179 zfree(&env->cpu_desc);
180 zfree(&env->cpuid);
181
182 zfree(&env->cmdline);
183 zfree(&env->cmdline_argv);
184 zfree(&env->sibling_cores);
185 zfree(&env->sibling_threads);
186 zfree(&env->numa_nodes);
187 zfree(&env->pmu_mappings);
188 }
189
190 void perf_session__delete(struct perf_session *session)
191 {
192 auxtrace__free(session);
193 auxtrace_index__free(&session->auxtrace_index);
194 perf_session__destroy_kernel_maps(session);
195 perf_session__delete_threads(session);
196 perf_session_env__exit(&session->header.env);
197 machines__exit(&session->machines);
198 if (session->file)
199 perf_data_file__close(session->file);
200 free(session);
201 }
202
203 static int process_event_synth_tracing_data_stub(struct perf_tool *tool
204 __maybe_unused,
205 union perf_event *event
206 __maybe_unused,
207 struct perf_session *session
208 __maybe_unused)
209 {
210 dump_printf(": unhandled!\n");
211 return 0;
212 }
213
214 static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused,
215 union perf_event *event __maybe_unused,
216 struct perf_evlist **pevlist
217 __maybe_unused)
218 {
219 dump_printf(": unhandled!\n");
220 return 0;
221 }
222
223 static int process_event_sample_stub(struct perf_tool *tool __maybe_unused,
224 union perf_event *event __maybe_unused,
225 struct perf_sample *sample __maybe_unused,
226 struct perf_evsel *evsel __maybe_unused,
227 struct machine *machine __maybe_unused)
228 {
229 dump_printf(": unhandled!\n");
230 return 0;
231 }
232
233 static int process_event_stub(struct perf_tool *tool __maybe_unused,
234 union perf_event *event __maybe_unused,
235 struct perf_sample *sample __maybe_unused,
236 struct machine *machine __maybe_unused)
237 {
238 dump_printf(": unhandled!\n");
239 return 0;
240 }
241
242 static int process_build_id_stub(struct perf_tool *tool __maybe_unused,
243 union perf_event *event __maybe_unused,
244 struct perf_session *session __maybe_unused)
245 {
246 dump_printf(": unhandled!\n");
247 return 0;
248 }
249
250 static int process_finished_round_stub(struct perf_tool *tool __maybe_unused,
251 union perf_event *event __maybe_unused,
252 struct ordered_events *oe __maybe_unused)
253 {
254 dump_printf(": unhandled!\n");
255 return 0;
256 }
257
258 static int process_finished_round(struct perf_tool *tool,
259 union perf_event *event,
260 struct ordered_events *oe);
261
262 static int process_id_index_stub(struct perf_tool *tool __maybe_unused,
263 union perf_event *event __maybe_unused,
264 struct perf_session *perf_session
265 __maybe_unused)
266 {
267 dump_printf(": unhandled!\n");
268 return 0;
269 }
270
271 static int process_event_auxtrace_info_stub(struct perf_tool *tool __maybe_unused,
272 union perf_event *event __maybe_unused,
273 struct perf_session *session __maybe_unused)
274 {
275 dump_printf(": unhandled!\n");
276 return 0;
277 }
278
279 static int skipn(int fd, off_t n)
280 {
281 char buf[4096];
282 ssize_t ret;
283
284 while (n > 0) {
285 ret = read(fd, buf, min(n, (off_t)sizeof(buf)));
286 if (ret <= 0)
287 return ret;
288 n -= ret;
289 }
290
291 return 0;
292 }
293
294 static s64 process_event_auxtrace_stub(struct perf_tool *tool __maybe_unused,
295 union perf_event *event,
296 struct perf_session *session
297 __maybe_unused)
298 {
299 dump_printf(": unhandled!\n");
300 if (perf_data_file__is_pipe(session->file))
301 skipn(perf_data_file__fd(session->file), event->auxtrace.size);
302 return event->auxtrace.size;
303 }
304
305 static
306 int process_event_auxtrace_error_stub(struct perf_tool *tool __maybe_unused,
307 union perf_event *event __maybe_unused,
308 struct perf_session *session __maybe_unused)
309 {
310 dump_printf(": unhandled!\n");
311 return 0;
312 }
313
314 void perf_tool__fill_defaults(struct perf_tool *tool)
315 {
316 if (tool->sample == NULL)
317 tool->sample = process_event_sample_stub;
318 if (tool->mmap == NULL)
319 tool->mmap = process_event_stub;
320 if (tool->mmap2 == NULL)
321 tool->mmap2 = process_event_stub;
322 if (tool->comm == NULL)
323 tool->comm = process_event_stub;
324 if (tool->fork == NULL)
325 tool->fork = process_event_stub;
326 if (tool->exit == NULL)
327 tool->exit = process_event_stub;
328 if (tool->lost == NULL)
329 tool->lost = perf_event__process_lost;
330 if (tool->lost_samples == NULL)
331 tool->lost_samples = perf_event__process_lost_samples;
332 if (tool->aux == NULL)
333 tool->aux = perf_event__process_aux;
334 if (tool->itrace_start == NULL)
335 tool->itrace_start = perf_event__process_itrace_start;
336 if (tool->context_switch == NULL)
337 tool->context_switch = perf_event__process_switch;
338 if (tool->read == NULL)
339 tool->read = process_event_sample_stub;
340 if (tool->throttle == NULL)
341 tool->throttle = process_event_stub;
342 if (tool->unthrottle == NULL)
343 tool->unthrottle = process_event_stub;
344 if (tool->attr == NULL)
345 tool->attr = process_event_synth_attr_stub;
346 if (tool->tracing_data == NULL)
347 tool->tracing_data = process_event_synth_tracing_data_stub;
348 if (tool->build_id == NULL)
349 tool->build_id = process_build_id_stub;
350 if (tool->finished_round == NULL) {
351 if (tool->ordered_events)
352 tool->finished_round = process_finished_round;
353 else
354 tool->finished_round = process_finished_round_stub;
355 }
356 if (tool->id_index == NULL)
357 tool->id_index = process_id_index_stub;
358 if (tool->auxtrace_info == NULL)
359 tool->auxtrace_info = process_event_auxtrace_info_stub;
360 if (tool->auxtrace == NULL)
361 tool->auxtrace = process_event_auxtrace_stub;
362 if (tool->auxtrace_error == NULL)
363 tool->auxtrace_error = process_event_auxtrace_error_stub;
364 }
365
366 static void swap_sample_id_all(union perf_event *event, void *data)
367 {
368 void *end = (void *) event + event->header.size;
369 int size = end - data;
370
371 BUG_ON(size % sizeof(u64));
372 mem_bswap_64(data, size);
373 }
374
375 static void perf_event__all64_swap(union perf_event *event,
376 bool sample_id_all __maybe_unused)
377 {
378 struct perf_event_header *hdr = &event->header;
379 mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr));
380 }
381
382 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all)
383 {
384 event->comm.pid = bswap_32(event->comm.pid);
385 event->comm.tid = bswap_32(event->comm.tid);
386
387 if (sample_id_all) {
388 void *data = &event->comm.comm;
389
390 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
391 swap_sample_id_all(event, data);
392 }
393 }
394
395 static void perf_event__mmap_swap(union perf_event *event,
396 bool sample_id_all)
397 {
398 event->mmap.pid = bswap_32(event->mmap.pid);
399 event->mmap.tid = bswap_32(event->mmap.tid);
400 event->mmap.start = bswap_64(event->mmap.start);
401 event->mmap.len = bswap_64(event->mmap.len);
402 event->mmap.pgoff = bswap_64(event->mmap.pgoff);
403
404 if (sample_id_all) {
405 void *data = &event->mmap.filename;
406
407 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
408 swap_sample_id_all(event, data);
409 }
410 }
411
412 static void perf_event__mmap2_swap(union perf_event *event,
413 bool sample_id_all)
414 {
415 event->mmap2.pid = bswap_32(event->mmap2.pid);
416 event->mmap2.tid = bswap_32(event->mmap2.tid);
417 event->mmap2.start = bswap_64(event->mmap2.start);
418 event->mmap2.len = bswap_64(event->mmap2.len);
419 event->mmap2.pgoff = bswap_64(event->mmap2.pgoff);
420 event->mmap2.maj = bswap_32(event->mmap2.maj);
421 event->mmap2.min = bswap_32(event->mmap2.min);
422 event->mmap2.ino = bswap_64(event->mmap2.ino);
423
424 if (sample_id_all) {
425 void *data = &event->mmap2.filename;
426
427 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
428 swap_sample_id_all(event, data);
429 }
430 }
431 static void perf_event__task_swap(union perf_event *event, bool sample_id_all)
432 {
433 event->fork.pid = bswap_32(event->fork.pid);
434 event->fork.tid = bswap_32(event->fork.tid);
435 event->fork.ppid = bswap_32(event->fork.ppid);
436 event->fork.ptid = bswap_32(event->fork.ptid);
437 event->fork.time = bswap_64(event->fork.time);
438
439 if (sample_id_all)
440 swap_sample_id_all(event, &event->fork + 1);
441 }
442
443 static void perf_event__read_swap(union perf_event *event, bool sample_id_all)
444 {
445 event->read.pid = bswap_32(event->read.pid);
446 event->read.tid = bswap_32(event->read.tid);
447 event->read.value = bswap_64(event->read.value);
448 event->read.time_enabled = bswap_64(event->read.time_enabled);
449 event->read.time_running = bswap_64(event->read.time_running);
450 event->read.id = bswap_64(event->read.id);
451
452 if (sample_id_all)
453 swap_sample_id_all(event, &event->read + 1);
454 }
455
456 static void perf_event__aux_swap(union perf_event *event, bool sample_id_all)
457 {
458 event->aux.aux_offset = bswap_64(event->aux.aux_offset);
459 event->aux.aux_size = bswap_64(event->aux.aux_size);
460 event->aux.flags = bswap_64(event->aux.flags);
461
462 if (sample_id_all)
463 swap_sample_id_all(event, &event->aux + 1);
464 }
465
466 static void perf_event__itrace_start_swap(union perf_event *event,
467 bool sample_id_all)
468 {
469 event->itrace_start.pid = bswap_32(event->itrace_start.pid);
470 event->itrace_start.tid = bswap_32(event->itrace_start.tid);
471
472 if (sample_id_all)
473 swap_sample_id_all(event, &event->itrace_start + 1);
474 }
475
476 static void perf_event__switch_swap(union perf_event *event, bool sample_id_all)
477 {
478 if (event->header.type == PERF_RECORD_SWITCH_CPU_WIDE) {
479 event->context_switch.next_prev_pid =
480 bswap_32(event->context_switch.next_prev_pid);
481 event->context_switch.next_prev_tid =
482 bswap_32(event->context_switch.next_prev_tid);
483 }
484
485 if (sample_id_all)
486 swap_sample_id_all(event, &event->context_switch + 1);
487 }
488
489 static void perf_event__throttle_swap(union perf_event *event,
490 bool sample_id_all)
491 {
492 event->throttle.time = bswap_64(event->throttle.time);
493 event->throttle.id = bswap_64(event->throttle.id);
494 event->throttle.stream_id = bswap_64(event->throttle.stream_id);
495
496 if (sample_id_all)
497 swap_sample_id_all(event, &event->throttle + 1);
498 }
499
500 static u8 revbyte(u8 b)
501 {
502 int rev = (b >> 4) | ((b & 0xf) << 4);
503 rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2);
504 rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1);
505 return (u8) rev;
506 }
507
508 /*
509 * XXX this is hack in attempt to carry flags bitfield
510 * throught endian village. ABI says:
511 *
512 * Bit-fields are allocated from right to left (least to most significant)
513 * on little-endian implementations and from left to right (most to least
514 * significant) on big-endian implementations.
515 *
516 * The above seems to be byte specific, so we need to reverse each
517 * byte of the bitfield. 'Internet' also says this might be implementation
518 * specific and we probably need proper fix and carry perf_event_attr
519 * bitfield flags in separate data file FEAT_ section. Thought this seems
520 * to work for now.
521 */
522 static void swap_bitfield(u8 *p, unsigned len)
523 {
524 unsigned i;
525
526 for (i = 0; i < len; i++) {
527 *p = revbyte(*p);
528 p++;
529 }
530 }
531
532 /* exported for swapping attributes in file header */
533 void perf_event__attr_swap(struct perf_event_attr *attr)
534 {
535 attr->type = bswap_32(attr->type);
536 attr->size = bswap_32(attr->size);
537
538 #define bswap_safe(f, n) \
539 (attr->size > (offsetof(struct perf_event_attr, f) + \
540 sizeof(attr->f) * (n)))
541 #define bswap_field(f, sz) \
542 do { \
543 if (bswap_safe(f, 0)) \
544 attr->f = bswap_##sz(attr->f); \
545 } while(0)
546 #define bswap_field_32(f) bswap_field(f, 32)
547 #define bswap_field_64(f) bswap_field(f, 64)
548
549 bswap_field_64(config);
550 bswap_field_64(sample_period);
551 bswap_field_64(sample_type);
552 bswap_field_64(read_format);
553 bswap_field_32(wakeup_events);
554 bswap_field_32(bp_type);
555 bswap_field_64(bp_addr);
556 bswap_field_64(bp_len);
557 bswap_field_64(branch_sample_type);
558 bswap_field_64(sample_regs_user);
559 bswap_field_32(sample_stack_user);
560 bswap_field_32(aux_watermark);
561
562 /*
563 * After read_format are bitfields. Check read_format because
564 * we are unable to use offsetof on bitfield.
565 */
566 if (bswap_safe(read_format, 1))
567 swap_bitfield((u8 *) (&attr->read_format + 1),
568 sizeof(u64));
569 #undef bswap_field_64
570 #undef bswap_field_32
571 #undef bswap_field
572 #undef bswap_safe
573 }
574
575 static void perf_event__hdr_attr_swap(union perf_event *event,
576 bool sample_id_all __maybe_unused)
577 {
578 size_t size;
579
580 perf_event__attr_swap(&event->attr.attr);
581
582 size = event->header.size;
583 size -= (void *)&event->attr.id - (void *)event;
584 mem_bswap_64(event->attr.id, size);
585 }
586
587 static void perf_event__event_type_swap(union perf_event *event,
588 bool sample_id_all __maybe_unused)
589 {
590 event->event_type.event_type.event_id =
591 bswap_64(event->event_type.event_type.event_id);
592 }
593
594 static void perf_event__tracing_data_swap(union perf_event *event,
595 bool sample_id_all __maybe_unused)
596 {
597 event->tracing_data.size = bswap_32(event->tracing_data.size);
598 }
599
600 static void perf_event__auxtrace_info_swap(union perf_event *event,
601 bool sample_id_all __maybe_unused)
602 {
603 size_t size;
604
605 event->auxtrace_info.type = bswap_32(event->auxtrace_info.type);
606
607 size = event->header.size;
608 size -= (void *)&event->auxtrace_info.priv - (void *)event;
609 mem_bswap_64(event->auxtrace_info.priv, size);
610 }
611
612 static void perf_event__auxtrace_swap(union perf_event *event,
613 bool sample_id_all __maybe_unused)
614 {
615 event->auxtrace.size = bswap_64(event->auxtrace.size);
616 event->auxtrace.offset = bswap_64(event->auxtrace.offset);
617 event->auxtrace.reference = bswap_64(event->auxtrace.reference);
618 event->auxtrace.idx = bswap_32(event->auxtrace.idx);
619 event->auxtrace.tid = bswap_32(event->auxtrace.tid);
620 event->auxtrace.cpu = bswap_32(event->auxtrace.cpu);
621 }
622
623 static void perf_event__auxtrace_error_swap(union perf_event *event,
624 bool sample_id_all __maybe_unused)
625 {
626 event->auxtrace_error.type = bswap_32(event->auxtrace_error.type);
627 event->auxtrace_error.code = bswap_32(event->auxtrace_error.code);
628 event->auxtrace_error.cpu = bswap_32(event->auxtrace_error.cpu);
629 event->auxtrace_error.pid = bswap_32(event->auxtrace_error.pid);
630 event->auxtrace_error.tid = bswap_32(event->auxtrace_error.tid);
631 event->auxtrace_error.ip = bswap_64(event->auxtrace_error.ip);
632 }
633
634 typedef void (*perf_event__swap_op)(union perf_event *event,
635 bool sample_id_all);
636
637 static perf_event__swap_op perf_event__swap_ops[] = {
638 [PERF_RECORD_MMAP] = perf_event__mmap_swap,
639 [PERF_RECORD_MMAP2] = perf_event__mmap2_swap,
640 [PERF_RECORD_COMM] = perf_event__comm_swap,
641 [PERF_RECORD_FORK] = perf_event__task_swap,
642 [PERF_RECORD_EXIT] = perf_event__task_swap,
643 [PERF_RECORD_LOST] = perf_event__all64_swap,
644 [PERF_RECORD_READ] = perf_event__read_swap,
645 [PERF_RECORD_THROTTLE] = perf_event__throttle_swap,
646 [PERF_RECORD_UNTHROTTLE] = perf_event__throttle_swap,
647 [PERF_RECORD_SAMPLE] = perf_event__all64_swap,
648 [PERF_RECORD_AUX] = perf_event__aux_swap,
649 [PERF_RECORD_ITRACE_START] = perf_event__itrace_start_swap,
650 [PERF_RECORD_LOST_SAMPLES] = perf_event__all64_swap,
651 [PERF_RECORD_SWITCH] = perf_event__switch_swap,
652 [PERF_RECORD_SWITCH_CPU_WIDE] = perf_event__switch_swap,
653 [PERF_RECORD_HEADER_ATTR] = perf_event__hdr_attr_swap,
654 [PERF_RECORD_HEADER_EVENT_TYPE] = perf_event__event_type_swap,
655 [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
656 [PERF_RECORD_HEADER_BUILD_ID] = NULL,
657 [PERF_RECORD_ID_INDEX] = perf_event__all64_swap,
658 [PERF_RECORD_AUXTRACE_INFO] = perf_event__auxtrace_info_swap,
659 [PERF_RECORD_AUXTRACE] = perf_event__auxtrace_swap,
660 [PERF_RECORD_AUXTRACE_ERROR] = perf_event__auxtrace_error_swap,
661 [PERF_RECORD_HEADER_MAX] = NULL,
662 };
663
664 /*
665 * When perf record finishes a pass on every buffers, it records this pseudo
666 * event.
667 * We record the max timestamp t found in the pass n.
668 * Assuming these timestamps are monotonic across cpus, we know that if
669 * a buffer still has events with timestamps below t, they will be all
670 * available and then read in the pass n + 1.
671 * Hence when we start to read the pass n + 2, we can safely flush every
672 * events with timestamps below t.
673 *
674 * ============ PASS n =================
675 * CPU 0 | CPU 1
676 * |
677 * cnt1 timestamps | cnt2 timestamps
678 * 1 | 2
679 * 2 | 3
680 * - | 4 <--- max recorded
681 *
682 * ============ PASS n + 1 ==============
683 * CPU 0 | CPU 1
684 * |
685 * cnt1 timestamps | cnt2 timestamps
686 * 3 | 5
687 * 4 | 6
688 * 5 | 7 <---- max recorded
689 *
690 * Flush every events below timestamp 4
691 *
692 * ============ PASS n + 2 ==============
693 * CPU 0 | CPU 1
694 * |
695 * cnt1 timestamps | cnt2 timestamps
696 * 6 | 8
697 * 7 | 9
698 * - | 10
699 *
700 * Flush every events below timestamp 7
701 * etc...
702 */
703 static int process_finished_round(struct perf_tool *tool __maybe_unused,
704 union perf_event *event __maybe_unused,
705 struct ordered_events *oe)
706 {
707 if (dump_trace)
708 fprintf(stdout, "\n");
709 return ordered_events__flush(oe, OE_FLUSH__ROUND);
710 }
711
712 int perf_session__queue_event(struct perf_session *s, union perf_event *event,
713 struct perf_sample *sample, u64 file_offset)
714 {
715 return ordered_events__queue(&s->ordered_events, event, sample, file_offset);
716 }
717
718 static void callchain__lbr_callstack_printf(struct perf_sample *sample)
719 {
720 struct ip_callchain *callchain = sample->callchain;
721 struct branch_stack *lbr_stack = sample->branch_stack;
722 u64 kernel_callchain_nr = callchain->nr;
723 unsigned int i;
724
725 for (i = 0; i < kernel_callchain_nr; i++) {
726 if (callchain->ips[i] == PERF_CONTEXT_USER)
727 break;
728 }
729
730 if ((i != kernel_callchain_nr) && lbr_stack->nr) {
731 u64 total_nr;
732 /*
733 * LBR callstack can only get user call chain,
734 * i is kernel call chain number,
735 * 1 is PERF_CONTEXT_USER.
736 *
737 * The user call chain is stored in LBR registers.
738 * LBR are pair registers. The caller is stored
739 * in "from" register, while the callee is stored
740 * in "to" register.
741 * For example, there is a call stack
742 * "A"->"B"->"C"->"D".
743 * The LBR registers will recorde like
744 * "C"->"D", "B"->"C", "A"->"B".
745 * So only the first "to" register and all "from"
746 * registers are needed to construct the whole stack.
747 */
748 total_nr = i + 1 + lbr_stack->nr + 1;
749 kernel_callchain_nr = i + 1;
750
751 printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr);
752
753 for (i = 0; i < kernel_callchain_nr; i++)
754 printf("..... %2d: %016" PRIx64 "\n",
755 i, callchain->ips[i]);
756
757 printf("..... %2d: %016" PRIx64 "\n",
758 (int)(kernel_callchain_nr), lbr_stack->entries[0].to);
759 for (i = 0; i < lbr_stack->nr; i++)
760 printf("..... %2d: %016" PRIx64 "\n",
761 (int)(i + kernel_callchain_nr + 1), lbr_stack->entries[i].from);
762 }
763 }
764
765 static void callchain__printf(struct perf_evsel *evsel,
766 struct perf_sample *sample)
767 {
768 unsigned int i;
769 struct ip_callchain *callchain = sample->callchain;
770
771 if (has_branch_callstack(evsel))
772 callchain__lbr_callstack_printf(sample);
773
774 printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr);
775
776 for (i = 0; i < callchain->nr; i++)
777 printf("..... %2d: %016" PRIx64 "\n",
778 i, callchain->ips[i]);
779 }
780
781 static void branch_stack__printf(struct perf_sample *sample)
782 {
783 uint64_t i;
784
785 printf("... branch stack: nr:%" PRIu64 "\n", sample->branch_stack->nr);
786
787 for (i = 0; i < sample->branch_stack->nr; i++) {
788 struct branch_entry *e = &sample->branch_stack->entries[i];
789
790 printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x\n",
791 i, e->from, e->to,
792 e->flags.cycles,
793 e->flags.mispred ? "M" : " ",
794 e->flags.predicted ? "P" : " ",
795 e->flags.abort ? "A" : " ",
796 e->flags.in_tx ? "T" : " ",
797 (unsigned)e->flags.reserved);
798 }
799 }
800
801 static void regs_dump__printf(u64 mask, u64 *regs)
802 {
803 unsigned rid, i = 0;
804
805 for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) {
806 u64 val = regs[i++];
807
808 printf(".... %-5s 0x%" PRIx64 "\n",
809 perf_reg_name(rid), val);
810 }
811 }
812
813 static const char *regs_abi[] = {
814 [PERF_SAMPLE_REGS_ABI_NONE] = "none",
815 [PERF_SAMPLE_REGS_ABI_32] = "32-bit",
816 [PERF_SAMPLE_REGS_ABI_64] = "64-bit",
817 };
818
819 static inline const char *regs_dump_abi(struct regs_dump *d)
820 {
821 if (d->abi > PERF_SAMPLE_REGS_ABI_64)
822 return "unknown";
823
824 return regs_abi[d->abi];
825 }
826
827 static void regs__printf(const char *type, struct regs_dump *regs)
828 {
829 u64 mask = regs->mask;
830
831 printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n",
832 type,
833 mask,
834 regs_dump_abi(regs));
835
836 regs_dump__printf(mask, regs->regs);
837 }
838
839 static void regs_user__printf(struct perf_sample *sample)
840 {
841 struct regs_dump *user_regs = &sample->user_regs;
842
843 if (user_regs->regs)
844 regs__printf("user", user_regs);
845 }
846
847 static void regs_intr__printf(struct perf_sample *sample)
848 {
849 struct regs_dump *intr_regs = &sample->intr_regs;
850
851 if (intr_regs->regs)
852 regs__printf("intr", intr_regs);
853 }
854
855 static void stack_user__printf(struct stack_dump *dump)
856 {
857 printf("... ustack: size %" PRIu64 ", offset 0x%x\n",
858 dump->size, dump->offset);
859 }
860
861 static void perf_evlist__print_tstamp(struct perf_evlist *evlist,
862 union perf_event *event,
863 struct perf_sample *sample)
864 {
865 u64 sample_type = __perf_evlist__combined_sample_type(evlist);
866
867 if (event->header.type != PERF_RECORD_SAMPLE &&
868 !perf_evlist__sample_id_all(evlist)) {
869 fputs("-1 -1 ", stdout);
870 return;
871 }
872
873 if ((sample_type & PERF_SAMPLE_CPU))
874 printf("%u ", sample->cpu);
875
876 if (sample_type & PERF_SAMPLE_TIME)
877 printf("%" PRIu64 " ", sample->time);
878 }
879
880 static void sample_read__printf(struct perf_sample *sample, u64 read_format)
881 {
882 printf("... sample_read:\n");
883
884 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
885 printf("...... time enabled %016" PRIx64 "\n",
886 sample->read.time_enabled);
887
888 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
889 printf("...... time running %016" PRIx64 "\n",
890 sample->read.time_running);
891
892 if (read_format & PERF_FORMAT_GROUP) {
893 u64 i;
894
895 printf(".... group nr %" PRIu64 "\n", sample->read.group.nr);
896
897 for (i = 0; i < sample->read.group.nr; i++) {
898 struct sample_read_value *value;
899
900 value = &sample->read.group.values[i];
901 printf("..... id %016" PRIx64
902 ", value %016" PRIx64 "\n",
903 value->id, value->value);
904 }
905 } else
906 printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n",
907 sample->read.one.id, sample->read.one.value);
908 }
909
910 static void dump_event(struct perf_evlist *evlist, union perf_event *event,
911 u64 file_offset, struct perf_sample *sample)
912 {
913 if (!dump_trace)
914 return;
915
916 printf("\n%#" PRIx64 " [%#x]: event: %d\n",
917 file_offset, event->header.size, event->header.type);
918
919 trace_event(event);
920
921 if (sample)
922 perf_evlist__print_tstamp(evlist, event, sample);
923
924 printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset,
925 event->header.size, perf_event__name(event->header.type));
926 }
927
928 static void dump_sample(struct perf_evsel *evsel, union perf_event *event,
929 struct perf_sample *sample)
930 {
931 u64 sample_type;
932
933 if (!dump_trace)
934 return;
935
936 printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n",
937 event->header.misc, sample->pid, sample->tid, sample->ip,
938 sample->period, sample->addr);
939
940 sample_type = evsel->attr.sample_type;
941
942 if (sample_type & PERF_SAMPLE_CALLCHAIN)
943 callchain__printf(evsel, sample);
944
945 if ((sample_type & PERF_SAMPLE_BRANCH_STACK) && !has_branch_callstack(evsel))
946 branch_stack__printf(sample);
947
948 if (sample_type & PERF_SAMPLE_REGS_USER)
949 regs_user__printf(sample);
950
951 if (sample_type & PERF_SAMPLE_REGS_INTR)
952 regs_intr__printf(sample);
953
954 if (sample_type & PERF_SAMPLE_STACK_USER)
955 stack_user__printf(&sample->user_stack);
956
957 if (sample_type & PERF_SAMPLE_WEIGHT)
958 printf("... weight: %" PRIu64 "\n", sample->weight);
959
960 if (sample_type & PERF_SAMPLE_DATA_SRC)
961 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src);
962
963 if (sample_type & PERF_SAMPLE_TRANSACTION)
964 printf("... transaction: %" PRIx64 "\n", sample->transaction);
965
966 if (sample_type & PERF_SAMPLE_READ)
967 sample_read__printf(sample, evsel->attr.read_format);
968 }
969
970 static struct machine *machines__find_for_cpumode(struct machines *machines,
971 union perf_event *event,
972 struct perf_sample *sample)
973 {
974 const u8 cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
975 struct machine *machine;
976
977 if (perf_guest &&
978 ((cpumode == PERF_RECORD_MISC_GUEST_KERNEL) ||
979 (cpumode == PERF_RECORD_MISC_GUEST_USER))) {
980 u32 pid;
981
982 if (event->header.type == PERF_RECORD_MMAP
983 || event->header.type == PERF_RECORD_MMAP2)
984 pid = event->mmap.pid;
985 else
986 pid = sample->pid;
987
988 machine = machines__find(machines, pid);
989 if (!machine)
990 machine = machines__find(machines, DEFAULT_GUEST_KERNEL_ID);
991 return machine;
992 }
993
994 return &machines->host;
995 }
996
997 static int deliver_sample_value(struct perf_evlist *evlist,
998 struct perf_tool *tool,
999 union perf_event *event,
1000 struct perf_sample *sample,
1001 struct sample_read_value *v,
1002 struct machine *machine)
1003 {
1004 struct perf_sample_id *sid = perf_evlist__id2sid(evlist, v->id);
1005
1006 if (sid) {
1007 sample->id = v->id;
1008 sample->period = v->value - sid->period;
1009 sid->period = v->value;
1010 }
1011
1012 if (!sid || sid->evsel == NULL) {
1013 ++evlist->stats.nr_unknown_id;
1014 return 0;
1015 }
1016
1017 return tool->sample(tool, event, sample, sid->evsel, machine);
1018 }
1019
1020 static int deliver_sample_group(struct perf_evlist *evlist,
1021 struct perf_tool *tool,
1022 union perf_event *event,
1023 struct perf_sample *sample,
1024 struct machine *machine)
1025 {
1026 int ret = -EINVAL;
1027 u64 i;
1028
1029 for (i = 0; i < sample->read.group.nr; i++) {
1030 ret = deliver_sample_value(evlist, tool, event, sample,
1031 &sample->read.group.values[i],
1032 machine);
1033 if (ret)
1034 break;
1035 }
1036
1037 return ret;
1038 }
1039
1040 static int
1041 perf_evlist__deliver_sample(struct perf_evlist *evlist,
1042 struct perf_tool *tool,
1043 union perf_event *event,
1044 struct perf_sample *sample,
1045 struct perf_evsel *evsel,
1046 struct machine *machine)
1047 {
1048 /* We know evsel != NULL. */
1049 u64 sample_type = evsel->attr.sample_type;
1050 u64 read_format = evsel->attr.read_format;
1051
1052 /* Standard sample delievery. */
1053 if (!(sample_type & PERF_SAMPLE_READ))
1054 return tool->sample(tool, event, sample, evsel, machine);
1055
1056 /* For PERF_SAMPLE_READ we have either single or group mode. */
1057 if (read_format & PERF_FORMAT_GROUP)
1058 return deliver_sample_group(evlist, tool, event, sample,
1059 machine);
1060 else
1061 return deliver_sample_value(evlist, tool, event, sample,
1062 &sample->read.one, machine);
1063 }
1064
1065 static int machines__deliver_event(struct machines *machines,
1066 struct perf_evlist *evlist,
1067 union perf_event *event,
1068 struct perf_sample *sample,
1069 struct perf_tool *tool, u64 file_offset)
1070 {
1071 struct perf_evsel *evsel;
1072 struct machine *machine;
1073
1074 dump_event(evlist, event, file_offset, sample);
1075
1076 evsel = perf_evlist__id2evsel(evlist, sample->id);
1077
1078 machine = machines__find_for_cpumode(machines, event, sample);
1079
1080 switch (event->header.type) {
1081 case PERF_RECORD_SAMPLE:
1082 dump_sample(evsel, event, sample);
1083 if (evsel == NULL) {
1084 ++evlist->stats.nr_unknown_id;
1085 return 0;
1086 }
1087 if (machine == NULL) {
1088 ++evlist->stats.nr_unprocessable_samples;
1089 return 0;
1090 }
1091 return perf_evlist__deliver_sample(evlist, tool, event, sample, evsel, machine);
1092 case PERF_RECORD_MMAP:
1093 return tool->mmap(tool, event, sample, machine);
1094 case PERF_RECORD_MMAP2:
1095 if (event->header.misc & PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT)
1096 ++evlist->stats.nr_proc_map_timeout;
1097 return tool->mmap2(tool, event, sample, machine);
1098 case PERF_RECORD_COMM:
1099 return tool->comm(tool, event, sample, machine);
1100 case PERF_RECORD_FORK:
1101 return tool->fork(tool, event, sample, machine);
1102 case PERF_RECORD_EXIT:
1103 return tool->exit(tool, event, sample, machine);
1104 case PERF_RECORD_LOST:
1105 if (tool->lost == perf_event__process_lost)
1106 evlist->stats.total_lost += event->lost.lost;
1107 return tool->lost(tool, event, sample, machine);
1108 case PERF_RECORD_LOST_SAMPLES:
1109 if (tool->lost_samples == perf_event__process_lost_samples)
1110 evlist->stats.total_lost_samples += event->lost_samples.lost;
1111 return tool->lost_samples(tool, event, sample, machine);
1112 case PERF_RECORD_READ:
1113 return tool->read(tool, event, sample, evsel, machine);
1114 case PERF_RECORD_THROTTLE:
1115 return tool->throttle(tool, event, sample, machine);
1116 case PERF_RECORD_UNTHROTTLE:
1117 return tool->unthrottle(tool, event, sample, machine);
1118 case PERF_RECORD_AUX:
1119 return tool->aux(tool, event, sample, machine);
1120 case PERF_RECORD_ITRACE_START:
1121 return tool->itrace_start(tool, event, sample, machine);
1122 case PERF_RECORD_SWITCH:
1123 case PERF_RECORD_SWITCH_CPU_WIDE:
1124 return tool->context_switch(tool, event, sample, machine);
1125 default:
1126 ++evlist->stats.nr_unknown_events;
1127 return -1;
1128 }
1129 }
1130
1131 static int perf_session__deliver_event(struct perf_session *session,
1132 union perf_event *event,
1133 struct perf_sample *sample,
1134 struct perf_tool *tool,
1135 u64 file_offset)
1136 {
1137 int ret;
1138
1139 ret = auxtrace__process_event(session, event, sample, tool);
1140 if (ret < 0)
1141 return ret;
1142 if (ret > 0)
1143 return 0;
1144
1145 return machines__deliver_event(&session->machines, session->evlist,
1146 event, sample, tool, file_offset);
1147 }
1148
1149 static s64 perf_session__process_user_event(struct perf_session *session,
1150 union perf_event *event,
1151 u64 file_offset)
1152 {
1153 struct ordered_events *oe = &session->ordered_events;
1154 struct perf_tool *tool = session->tool;
1155 int fd = perf_data_file__fd(session->file);
1156 int err;
1157
1158 dump_event(session->evlist, event, file_offset, NULL);
1159
1160 /* These events are processed right away */
1161 switch (event->header.type) {
1162 case PERF_RECORD_HEADER_ATTR:
1163 err = tool->attr(tool, event, &session->evlist);
1164 if (err == 0) {
1165 perf_session__set_id_hdr_size(session);
1166 perf_session__set_comm_exec(session);
1167 }
1168 return err;
1169 case PERF_RECORD_HEADER_EVENT_TYPE:
1170 /*
1171 * Depreceated, but we need to handle it for sake
1172 * of old data files create in pipe mode.
1173 */
1174 return 0;
1175 case PERF_RECORD_HEADER_TRACING_DATA:
1176 /* setup for reading amidst mmap */
1177 lseek(fd, file_offset, SEEK_SET);
1178 return tool->tracing_data(tool, event, session);
1179 case PERF_RECORD_HEADER_BUILD_ID:
1180 return tool->build_id(tool, event, session);
1181 case PERF_RECORD_FINISHED_ROUND:
1182 return tool->finished_round(tool, event, oe);
1183 case PERF_RECORD_ID_INDEX:
1184 return tool->id_index(tool, event, session);
1185 case PERF_RECORD_AUXTRACE_INFO:
1186 return tool->auxtrace_info(tool, event, session);
1187 case PERF_RECORD_AUXTRACE:
1188 /* setup for reading amidst mmap */
1189 lseek(fd, file_offset + event->header.size, SEEK_SET);
1190 return tool->auxtrace(tool, event, session);
1191 case PERF_RECORD_AUXTRACE_ERROR:
1192 perf_session__auxtrace_error_inc(session, event);
1193 return tool->auxtrace_error(tool, event, session);
1194 default:
1195 return -EINVAL;
1196 }
1197 }
1198
1199 int perf_session__deliver_synth_event(struct perf_session *session,
1200 union perf_event *event,
1201 struct perf_sample *sample)
1202 {
1203 struct perf_evlist *evlist = session->evlist;
1204 struct perf_tool *tool = session->tool;
1205
1206 events_stats__inc(&evlist->stats, event->header.type);
1207
1208 if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1209 return perf_session__process_user_event(session, event, 0);
1210
1211 return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0);
1212 }
1213
1214 static void event_swap(union perf_event *event, bool sample_id_all)
1215 {
1216 perf_event__swap_op swap;
1217
1218 swap = perf_event__swap_ops[event->header.type];
1219 if (swap)
1220 swap(event, sample_id_all);
1221 }
1222
1223 int perf_session__peek_event(struct perf_session *session, off_t file_offset,
1224 void *buf, size_t buf_sz,
1225 union perf_event **event_ptr,
1226 struct perf_sample *sample)
1227 {
1228 union perf_event *event;
1229 size_t hdr_sz, rest;
1230 int fd;
1231
1232 if (session->one_mmap && !session->header.needs_swap) {
1233 event = file_offset - session->one_mmap_offset +
1234 session->one_mmap_addr;
1235 goto out_parse_sample;
1236 }
1237
1238 if (perf_data_file__is_pipe(session->file))
1239 return -1;
1240
1241 fd = perf_data_file__fd(session->file);
1242 hdr_sz = sizeof(struct perf_event_header);
1243
1244 if (buf_sz < hdr_sz)
1245 return -1;
1246
1247 if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 ||
1248 readn(fd, buf, hdr_sz) != (ssize_t)hdr_sz)
1249 return -1;
1250
1251 event = (union perf_event *)buf;
1252
1253 if (session->header.needs_swap)
1254 perf_event_header__bswap(&event->header);
1255
1256 if (event->header.size < hdr_sz || event->header.size > buf_sz)
1257 return -1;
1258
1259 rest = event->header.size - hdr_sz;
1260
1261 if (readn(fd, buf, rest) != (ssize_t)rest)
1262 return -1;
1263
1264 if (session->header.needs_swap)
1265 event_swap(event, perf_evlist__sample_id_all(session->evlist));
1266
1267 out_parse_sample:
1268
1269 if (sample && event->header.type < PERF_RECORD_USER_TYPE_START &&
1270 perf_evlist__parse_sample(session->evlist, event, sample))
1271 return -1;
1272
1273 *event_ptr = event;
1274
1275 return 0;
1276 }
1277
1278 static s64 perf_session__process_event(struct perf_session *session,
1279 union perf_event *event, u64 file_offset)
1280 {
1281 struct perf_evlist *evlist = session->evlist;
1282 struct perf_tool *tool = session->tool;
1283 struct perf_sample sample;
1284 int ret;
1285
1286 if (session->header.needs_swap)
1287 event_swap(event, perf_evlist__sample_id_all(evlist));
1288
1289 if (event->header.type >= PERF_RECORD_HEADER_MAX)
1290 return -EINVAL;
1291
1292 events_stats__inc(&evlist->stats, event->header.type);
1293
1294 if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1295 return perf_session__process_user_event(session, event, file_offset);
1296
1297 /*
1298 * For all kernel events we get the sample data
1299 */
1300 ret = perf_evlist__parse_sample(evlist, event, &sample);
1301 if (ret)
1302 return ret;
1303
1304 if (tool->ordered_events) {
1305 ret = perf_session__queue_event(session, event, &sample, file_offset);
1306 if (ret != -ETIME)
1307 return ret;
1308 }
1309
1310 return perf_session__deliver_event(session, event, &sample, tool,
1311 file_offset);
1312 }
1313
1314 void perf_event_header__bswap(struct perf_event_header *hdr)
1315 {
1316 hdr->type = bswap_32(hdr->type);
1317 hdr->misc = bswap_16(hdr->misc);
1318 hdr->size = bswap_16(hdr->size);
1319 }
1320
1321 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid)
1322 {
1323 return machine__findnew_thread(&session->machines.host, -1, pid);
1324 }
1325
1326 static struct thread *perf_session__register_idle_thread(struct perf_session *session)
1327 {
1328 struct thread *thread;
1329
1330 thread = machine__findnew_thread(&session->machines.host, 0, 0);
1331 if (thread == NULL || thread__set_comm(thread, "swapper", 0)) {
1332 pr_err("problem inserting idle task.\n");
1333 thread = NULL;
1334 }
1335
1336 return thread;
1337 }
1338
1339 static void perf_session__warn_about_errors(const struct perf_session *session)
1340 {
1341 const struct events_stats *stats = &session->evlist->stats;
1342 const struct ordered_events *oe = &session->ordered_events;
1343
1344 if (session->tool->lost == perf_event__process_lost &&
1345 stats->nr_events[PERF_RECORD_LOST] != 0) {
1346 ui__warning("Processed %d events and lost %d chunks!\n\n"
1347 "Check IO/CPU overload!\n\n",
1348 stats->nr_events[0],
1349 stats->nr_events[PERF_RECORD_LOST]);
1350 }
1351
1352 if (session->tool->lost_samples == perf_event__process_lost_samples) {
1353 double drop_rate;
1354
1355 drop_rate = (double)stats->total_lost_samples /
1356 (double) (stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples);
1357 if (drop_rate > 0.05) {
1358 ui__warning("Processed %" PRIu64 " samples and lost %3.2f%% samples!\n\n",
1359 stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples,
1360 drop_rate * 100.0);
1361 }
1362 }
1363
1364 if (stats->nr_unknown_events != 0) {
1365 ui__warning("Found %u unknown events!\n\n"
1366 "Is this an older tool processing a perf.data "
1367 "file generated by a more recent tool?\n\n"
1368 "If that is not the case, consider "
1369 "reporting to linux-kernel@vger.kernel.org.\n\n",
1370 stats->nr_unknown_events);
1371 }
1372
1373 if (stats->nr_unknown_id != 0) {
1374 ui__warning("%u samples with id not present in the header\n",
1375 stats->nr_unknown_id);
1376 }
1377
1378 if (stats->nr_invalid_chains != 0) {
1379 ui__warning("Found invalid callchains!\n\n"
1380 "%u out of %u events were discarded for this reason.\n\n"
1381 "Consider reporting to linux-kernel@vger.kernel.org.\n\n",
1382 stats->nr_invalid_chains,
1383 stats->nr_events[PERF_RECORD_SAMPLE]);
1384 }
1385
1386 if (stats->nr_unprocessable_samples != 0) {
1387 ui__warning("%u unprocessable samples recorded.\n"
1388 "Do you have a KVM guest running and not using 'perf kvm'?\n",
1389 stats->nr_unprocessable_samples);
1390 }
1391
1392 if (oe->nr_unordered_events != 0)
1393 ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events);
1394
1395 events_stats__auxtrace_error_warn(stats);
1396
1397 if (stats->nr_proc_map_timeout != 0) {
1398 ui__warning("%d map information files for pre-existing threads were\n"
1399 "not processed, if there are samples for addresses they\n"
1400 "will not be resolved, you may find out which are these\n"
1401 "threads by running with -v and redirecting the output\n"
1402 "to a file.\n"
1403 "The time limit to process proc map is too short?\n"
1404 "Increase it by --proc-map-timeout\n",
1405 stats->nr_proc_map_timeout);
1406 }
1407 }
1408
1409 static int perf_session__flush_thread_stack(struct thread *thread,
1410 void *p __maybe_unused)
1411 {
1412 return thread_stack__flush(thread);
1413 }
1414
1415 static int perf_session__flush_thread_stacks(struct perf_session *session)
1416 {
1417 return machines__for_each_thread(&session->machines,
1418 perf_session__flush_thread_stack,
1419 NULL);
1420 }
1421
1422 volatile int session_done;
1423
1424 static int __perf_session__process_pipe_events(struct perf_session *session)
1425 {
1426 struct ordered_events *oe = &session->ordered_events;
1427 struct perf_tool *tool = session->tool;
1428 int fd = perf_data_file__fd(session->file);
1429 union perf_event *event;
1430 uint32_t size, cur_size = 0;
1431 void *buf = NULL;
1432 s64 skip = 0;
1433 u64 head;
1434 ssize_t err;
1435 void *p;
1436
1437 perf_tool__fill_defaults(tool);
1438
1439 head = 0;
1440 cur_size = sizeof(union perf_event);
1441
1442 buf = malloc(cur_size);
1443 if (!buf)
1444 return -errno;
1445 more:
1446 event = buf;
1447 err = readn(fd, event, sizeof(struct perf_event_header));
1448 if (err <= 0) {
1449 if (err == 0)
1450 goto done;
1451
1452 pr_err("failed to read event header\n");
1453 goto out_err;
1454 }
1455
1456 if (session->header.needs_swap)
1457 perf_event_header__bswap(&event->header);
1458
1459 size = event->header.size;
1460 if (size < sizeof(struct perf_event_header)) {
1461 pr_err("bad event header size\n");
1462 goto out_err;
1463 }
1464
1465 if (size > cur_size) {
1466 void *new = realloc(buf, size);
1467 if (!new) {
1468 pr_err("failed to allocate memory to read event\n");
1469 goto out_err;
1470 }
1471 buf = new;
1472 cur_size = size;
1473 event = buf;
1474 }
1475 p = event;
1476 p += sizeof(struct perf_event_header);
1477
1478 if (size - sizeof(struct perf_event_header)) {
1479 err = readn(fd, p, size - sizeof(struct perf_event_header));
1480 if (err <= 0) {
1481 if (err == 0) {
1482 pr_err("unexpected end of event stream\n");
1483 goto done;
1484 }
1485
1486 pr_err("failed to read event data\n");
1487 goto out_err;
1488 }
1489 }
1490
1491 if ((skip = perf_session__process_event(session, event, head)) < 0) {
1492 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1493 head, event->header.size, event->header.type);
1494 err = -EINVAL;
1495 goto out_err;
1496 }
1497
1498 head += size;
1499
1500 if (skip > 0)
1501 head += skip;
1502
1503 if (!session_done())
1504 goto more;
1505 done:
1506 /* do the final flush for ordered samples */
1507 err = ordered_events__flush(oe, OE_FLUSH__FINAL);
1508 if (err)
1509 goto out_err;
1510 err = auxtrace__flush_events(session, tool);
1511 if (err)
1512 goto out_err;
1513 err = perf_session__flush_thread_stacks(session);
1514 out_err:
1515 free(buf);
1516 perf_session__warn_about_errors(session);
1517 ordered_events__free(&session->ordered_events);
1518 auxtrace__free_events(session);
1519 return err;
1520 }
1521
1522 static union perf_event *
1523 fetch_mmaped_event(struct perf_session *session,
1524 u64 head, size_t mmap_size, char *buf)
1525 {
1526 union perf_event *event;
1527
1528 /*
1529 * Ensure we have enough space remaining to read
1530 * the size of the event in the headers.
1531 */
1532 if (head + sizeof(event->header) > mmap_size)
1533 return NULL;
1534
1535 event = (union perf_event *)(buf + head);
1536
1537 if (session->header.needs_swap)
1538 perf_event_header__bswap(&event->header);
1539
1540 if (head + event->header.size > mmap_size) {
1541 /* We're not fetching the event so swap back again */
1542 if (session->header.needs_swap)
1543 perf_event_header__bswap(&event->header);
1544 return NULL;
1545 }
1546
1547 return event;
1548 }
1549
1550 /*
1551 * On 64bit we can mmap the data file in one go. No need for tiny mmap
1552 * slices. On 32bit we use 32MB.
1553 */
1554 #if BITS_PER_LONG == 64
1555 #define MMAP_SIZE ULLONG_MAX
1556 #define NUM_MMAPS 1
1557 #else
1558 #define MMAP_SIZE (32 * 1024 * 1024ULL)
1559 #define NUM_MMAPS 128
1560 #endif
1561
1562 static int __perf_session__process_events(struct perf_session *session,
1563 u64 data_offset, u64 data_size,
1564 u64 file_size)
1565 {
1566 struct ordered_events *oe = &session->ordered_events;
1567 struct perf_tool *tool = session->tool;
1568 int fd = perf_data_file__fd(session->file);
1569 u64 head, page_offset, file_offset, file_pos, size;
1570 int err, mmap_prot, mmap_flags, map_idx = 0;
1571 size_t mmap_size;
1572 char *buf, *mmaps[NUM_MMAPS];
1573 union perf_event *event;
1574 struct ui_progress prog;
1575 s64 skip;
1576
1577 perf_tool__fill_defaults(tool);
1578
1579 page_offset = page_size * (data_offset / page_size);
1580 file_offset = page_offset;
1581 head = data_offset - page_offset;
1582
1583 if (data_size && (data_offset + data_size < file_size))
1584 file_size = data_offset + data_size;
1585
1586 ui_progress__init(&prog, file_size, "Processing events...");
1587
1588 mmap_size = MMAP_SIZE;
1589 if (mmap_size > file_size) {
1590 mmap_size = file_size;
1591 session->one_mmap = true;
1592 }
1593
1594 memset(mmaps, 0, sizeof(mmaps));
1595
1596 mmap_prot = PROT_READ;
1597 mmap_flags = MAP_SHARED;
1598
1599 if (session->header.needs_swap) {
1600 mmap_prot |= PROT_WRITE;
1601 mmap_flags = MAP_PRIVATE;
1602 }
1603 remap:
1604 buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, fd,
1605 file_offset);
1606 if (buf == MAP_FAILED) {
1607 pr_err("failed to mmap file\n");
1608 err = -errno;
1609 goto out_err;
1610 }
1611 mmaps[map_idx] = buf;
1612 map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1);
1613 file_pos = file_offset + head;
1614 if (session->one_mmap) {
1615 session->one_mmap_addr = buf;
1616 session->one_mmap_offset = file_offset;
1617 }
1618
1619 more:
1620 event = fetch_mmaped_event(session, head, mmap_size, buf);
1621 if (!event) {
1622 if (mmaps[map_idx]) {
1623 munmap(mmaps[map_idx], mmap_size);
1624 mmaps[map_idx] = NULL;
1625 }
1626
1627 page_offset = page_size * (head / page_size);
1628 file_offset += page_offset;
1629 head -= page_offset;
1630 goto remap;
1631 }
1632
1633 size = event->header.size;
1634
1635 if (size < sizeof(struct perf_event_header) ||
1636 (skip = perf_session__process_event(session, event, file_pos)) < 0) {
1637 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1638 file_offset + head, event->header.size,
1639 event->header.type);
1640 err = -EINVAL;
1641 goto out_err;
1642 }
1643
1644 if (skip)
1645 size += skip;
1646
1647 head += size;
1648 file_pos += size;
1649
1650 ui_progress__update(&prog, size);
1651
1652 if (session_done())
1653 goto out;
1654
1655 if (file_pos < file_size)
1656 goto more;
1657
1658 out:
1659 /* do the final flush for ordered samples */
1660 err = ordered_events__flush(oe, OE_FLUSH__FINAL);
1661 if (err)
1662 goto out_err;
1663 err = auxtrace__flush_events(session, tool);
1664 if (err)
1665 goto out_err;
1666 err = perf_session__flush_thread_stacks(session);
1667 out_err:
1668 ui_progress__finish();
1669 perf_session__warn_about_errors(session);
1670 ordered_events__free(&session->ordered_events);
1671 auxtrace__free_events(session);
1672 session->one_mmap = false;
1673 return err;
1674 }
1675
1676 int perf_session__process_events(struct perf_session *session)
1677 {
1678 u64 size = perf_data_file__size(session->file);
1679 int err;
1680
1681 if (perf_session__register_idle_thread(session) == NULL)
1682 return -ENOMEM;
1683
1684 if (!perf_data_file__is_pipe(session->file))
1685 err = __perf_session__process_events(session,
1686 session->header.data_offset,
1687 session->header.data_size, size);
1688 else
1689 err = __perf_session__process_pipe_events(session);
1690
1691 return err;
1692 }
1693
1694 bool perf_session__has_traces(struct perf_session *session, const char *msg)
1695 {
1696 struct perf_evsel *evsel;
1697
1698 evlist__for_each(session->evlist, evsel) {
1699 if (evsel->attr.type == PERF_TYPE_TRACEPOINT)
1700 return true;
1701 }
1702
1703 pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg);
1704 return false;
1705 }
1706
1707 int maps__set_kallsyms_ref_reloc_sym(struct map **maps,
1708 const char *symbol_name, u64 addr)
1709 {
1710 char *bracket;
1711 enum map_type i;
1712 struct ref_reloc_sym *ref;
1713
1714 ref = zalloc(sizeof(struct ref_reloc_sym));
1715 if (ref == NULL)
1716 return -ENOMEM;
1717
1718 ref->name = strdup(symbol_name);
1719 if (ref->name == NULL) {
1720 free(ref);
1721 return -ENOMEM;
1722 }
1723
1724 bracket = strchr(ref->name, ']');
1725 if (bracket)
1726 *bracket = '\0';
1727
1728 ref->addr = addr;
1729
1730 for (i = 0; i < MAP__NR_TYPES; ++i) {
1731 struct kmap *kmap = map__kmap(maps[i]);
1732
1733 if (!kmap)
1734 continue;
1735 kmap->ref_reloc_sym = ref;
1736 }
1737
1738 return 0;
1739 }
1740
1741 size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp)
1742 {
1743 return machines__fprintf_dsos(&session->machines, fp);
1744 }
1745
1746 size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp,
1747 bool (skip)(struct dso *dso, int parm), int parm)
1748 {
1749 return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm);
1750 }
1751
1752 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp)
1753 {
1754 size_t ret;
1755 const char *msg = "";
1756
1757 if (perf_header__has_feat(&session->header, HEADER_AUXTRACE))
1758 msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)";
1759
1760 ret = fprintf(fp, "\nAggregated stats:%s\n", msg);
1761
1762 ret += events_stats__fprintf(&session->evlist->stats, fp);
1763 return ret;
1764 }
1765
1766 size_t perf_session__fprintf(struct perf_session *session, FILE *fp)
1767 {
1768 /*
1769 * FIXME: Here we have to actually print all the machines in this
1770 * session, not just the host...
1771 */
1772 return machine__fprintf(&session->machines.host, fp);
1773 }
1774
1775 struct perf_evsel *perf_session__find_first_evtype(struct perf_session *session,
1776 unsigned int type)
1777 {
1778 struct perf_evsel *pos;
1779
1780 evlist__for_each(session->evlist, pos) {
1781 if (pos->attr.type == type)
1782 return pos;
1783 }
1784 return NULL;
1785 }
1786
1787 void perf_evsel__print_ip(struct perf_evsel *evsel, struct perf_sample *sample,
1788 struct addr_location *al,
1789 unsigned int print_opts, unsigned int stack_depth)
1790 {
1791 struct callchain_cursor_node *node;
1792 int print_ip = print_opts & PRINT_IP_OPT_IP;
1793 int print_sym = print_opts & PRINT_IP_OPT_SYM;
1794 int print_dso = print_opts & PRINT_IP_OPT_DSO;
1795 int print_symoffset = print_opts & PRINT_IP_OPT_SYMOFFSET;
1796 int print_oneline = print_opts & PRINT_IP_OPT_ONELINE;
1797 int print_srcline = print_opts & PRINT_IP_OPT_SRCLINE;
1798 char s = print_oneline ? ' ' : '\t';
1799
1800 if (symbol_conf.use_callchain && sample->callchain) {
1801 struct addr_location node_al;
1802
1803 if (thread__resolve_callchain(al->thread, evsel,
1804 sample, NULL, NULL,
1805 PERF_MAX_STACK_DEPTH) != 0) {
1806 if (verbose)
1807 error("Failed to resolve callchain. Skipping\n");
1808 return;
1809 }
1810 callchain_cursor_commit(&callchain_cursor);
1811
1812 if (print_symoffset)
1813 node_al = *al;
1814
1815 while (stack_depth) {
1816 u64 addr = 0;
1817
1818 node = callchain_cursor_current(&callchain_cursor);
1819 if (!node)
1820 break;
1821
1822 if (node->sym && node->sym->ignore)
1823 goto next;
1824
1825 if (print_ip)
1826 printf("%c%16" PRIx64, s, node->ip);
1827
1828 if (node->map)
1829 addr = node->map->map_ip(node->map, node->ip);
1830
1831 if (print_sym) {
1832 printf(" ");
1833 if (print_symoffset) {
1834 node_al.addr = addr;
1835 node_al.map = node->map;
1836 symbol__fprintf_symname_offs(node->sym, &node_al, stdout);
1837 } else
1838 symbol__fprintf_symname(node->sym, stdout);
1839 }
1840
1841 if (print_dso) {
1842 printf(" (");
1843 map__fprintf_dsoname(node->map, stdout);
1844 printf(")");
1845 }
1846
1847 if (print_srcline)
1848 map__fprintf_srcline(node->map, addr, "\n ",
1849 stdout);
1850
1851 if (!print_oneline)
1852 printf("\n");
1853
1854 stack_depth--;
1855 next:
1856 callchain_cursor_advance(&callchain_cursor);
1857 }
1858
1859 } else {
1860 if (al->sym && al->sym->ignore)
1861 return;
1862
1863 if (print_ip)
1864 printf("%16" PRIx64, sample->ip);
1865
1866 if (print_sym) {
1867 printf(" ");
1868 if (print_symoffset)
1869 symbol__fprintf_symname_offs(al->sym, al,
1870 stdout);
1871 else
1872 symbol__fprintf_symname(al->sym, stdout);
1873 }
1874
1875 if (print_dso) {
1876 printf(" (");
1877 map__fprintf_dsoname(al->map, stdout);
1878 printf(")");
1879 }
1880
1881 if (print_srcline)
1882 map__fprintf_srcline(al->map, al->addr, "\n ", stdout);
1883 }
1884 }
1885
1886 int perf_session__cpu_bitmap(struct perf_session *session,
1887 const char *cpu_list, unsigned long *cpu_bitmap)
1888 {
1889 int i, err = -1;
1890 struct cpu_map *map;
1891
1892 for (i = 0; i < PERF_TYPE_MAX; ++i) {
1893 struct perf_evsel *evsel;
1894
1895 evsel = perf_session__find_first_evtype(session, i);
1896 if (!evsel)
1897 continue;
1898
1899 if (!(evsel->attr.sample_type & PERF_SAMPLE_CPU)) {
1900 pr_err("File does not contain CPU events. "
1901 "Remove -c option to proceed.\n");
1902 return -1;
1903 }
1904 }
1905
1906 map = cpu_map__new(cpu_list);
1907 if (map == NULL) {
1908 pr_err("Invalid cpu_list\n");
1909 return -1;
1910 }
1911
1912 for (i = 0; i < map->nr; i++) {
1913 int cpu = map->map[i];
1914
1915 if (cpu >= MAX_NR_CPUS) {
1916 pr_err("Requested CPU %d too large. "
1917 "Consider raising MAX_NR_CPUS\n", cpu);
1918 goto out_delete_map;
1919 }
1920
1921 set_bit(cpu, cpu_bitmap);
1922 }
1923
1924 err = 0;
1925
1926 out_delete_map:
1927 cpu_map__put(map);
1928 return err;
1929 }
1930
1931 void perf_session__fprintf_info(struct perf_session *session, FILE *fp,
1932 bool full)
1933 {
1934 struct stat st;
1935 int fd, ret;
1936
1937 if (session == NULL || fp == NULL)
1938 return;
1939
1940 fd = perf_data_file__fd(session->file);
1941
1942 ret = fstat(fd, &st);
1943 if (ret == -1)
1944 return;
1945
1946 fprintf(fp, "# ========\n");
1947 fprintf(fp, "# captured on: %s", ctime(&st.st_ctime));
1948 perf_header__fprintf_info(session, fp, full);
1949 fprintf(fp, "# ========\n#\n");
1950 }
1951
1952
1953 int __perf_session__set_tracepoints_handlers(struct perf_session *session,
1954 const struct perf_evsel_str_handler *assocs,
1955 size_t nr_assocs)
1956 {
1957 struct perf_evsel *evsel;
1958 size_t i;
1959 int err;
1960
1961 for (i = 0; i < nr_assocs; i++) {
1962 /*
1963 * Adding a handler for an event not in the session,
1964 * just ignore it.
1965 */
1966 evsel = perf_evlist__find_tracepoint_by_name(session->evlist, assocs[i].name);
1967 if (evsel == NULL)
1968 continue;
1969
1970 err = -EEXIST;
1971 if (evsel->handler != NULL)
1972 goto out;
1973 evsel->handler = assocs[i].handler;
1974 }
1975
1976 err = 0;
1977 out:
1978 return err;
1979 }
1980
1981 int perf_event__process_id_index(struct perf_tool *tool __maybe_unused,
1982 union perf_event *event,
1983 struct perf_session *session)
1984 {
1985 struct perf_evlist *evlist = session->evlist;
1986 struct id_index_event *ie = &event->id_index;
1987 size_t i, nr, max_nr;
1988
1989 max_nr = (ie->header.size - sizeof(struct id_index_event)) /
1990 sizeof(struct id_index_entry);
1991 nr = ie->nr;
1992 if (nr > max_nr)
1993 return -EINVAL;
1994
1995 if (dump_trace)
1996 fprintf(stdout, " nr: %zu\n", nr);
1997
1998 for (i = 0; i < nr; i++) {
1999 struct id_index_entry *e = &ie->entries[i];
2000 struct perf_sample_id *sid;
2001
2002 if (dump_trace) {
2003 fprintf(stdout, " ... id: %"PRIu64, e->id);
2004 fprintf(stdout, " idx: %"PRIu64, e->idx);
2005 fprintf(stdout, " cpu: %"PRId64, e->cpu);
2006 fprintf(stdout, " tid: %"PRId64"\n", e->tid);
2007 }
2008
2009 sid = perf_evlist__id2sid(evlist, e->id);
2010 if (!sid)
2011 return -ENOENT;
2012 sid->idx = e->idx;
2013 sid->cpu = e->cpu;
2014 sid->tid = e->tid;
2015 }
2016 return 0;
2017 }
2018
2019 int perf_event__synthesize_id_index(struct perf_tool *tool,
2020 perf_event__handler_t process,
2021 struct perf_evlist *evlist,
2022 struct machine *machine)
2023 {
2024 union perf_event *ev;
2025 struct perf_evsel *evsel;
2026 size_t nr = 0, i = 0, sz, max_nr, n;
2027 int err;
2028
2029 pr_debug2("Synthesizing id index\n");
2030
2031 max_nr = (UINT16_MAX - sizeof(struct id_index_event)) /
2032 sizeof(struct id_index_entry);
2033
2034 evlist__for_each(evlist, evsel)
2035 nr += evsel->ids;
2036
2037 n = nr > max_nr ? max_nr : nr;
2038 sz = sizeof(struct id_index_event) + n * sizeof(struct id_index_entry);
2039 ev = zalloc(sz);
2040 if (!ev)
2041 return -ENOMEM;
2042
2043 ev->id_index.header.type = PERF_RECORD_ID_INDEX;
2044 ev->id_index.header.size = sz;
2045 ev->id_index.nr = n;
2046
2047 evlist__for_each(evlist, evsel) {
2048 u32 j;
2049
2050 for (j = 0; j < evsel->ids; j++) {
2051 struct id_index_entry *e;
2052 struct perf_sample_id *sid;
2053
2054 if (i >= n) {
2055 err = process(tool, ev, NULL, machine);
2056 if (err)
2057 goto out_err;
2058 nr -= n;
2059 i = 0;
2060 }
2061
2062 e = &ev->id_index.entries[i++];
2063
2064 e->id = evsel->id[j];
2065
2066 sid = perf_evlist__id2sid(evlist, e->id);
2067 if (!sid) {
2068 free(ev);
2069 return -ENOENT;
2070 }
2071
2072 e->idx = sid->idx;
2073 e->cpu = sid->cpu;
2074 e->tid = sid->tid;
2075 }
2076 }
2077
2078 sz = sizeof(struct id_index_event) + nr * sizeof(struct id_index_entry);
2079 ev->id_index.header.size = sz;
2080 ev->id_index.nr = nr;
2081
2082 err = process(tool, ev, NULL, machine);
2083 out_err:
2084 free(ev);
2085
2086 return err;
2087 }
Linux kernel - Deliverables
This page took 0.309688 seconds and 5 git commands to generate.