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