xref: /openbmc/linux/tools/perf/util/session.c (revision 6aa7de05)
1 #include <errno.h>
2 #include <inttypes.h>
3 #include <linux/kernel.h>
4 #include <traceevent/event-parse.h>
5 #include <api/fs/fs.h>
6 
7 #include <byteswap.h>
8 #include <unistd.h>
9 #include <sys/types.h>
10 #include <sys/mman.h>
11 
12 #include "evlist.h"
13 #include "evsel.h"
14 #include "memswap.h"
15 #include "session.h"
16 #include "tool.h"
17 #include "sort.h"
18 #include "util.h"
19 #include "cpumap.h"
20 #include "perf_regs.h"
21 #include "asm/bug.h"
22 #include "auxtrace.h"
23 #include "thread.h"
24 #include "thread-stack.h"
25 #include "stat.h"
26 
27 static int perf_session__deliver_event(struct perf_session *session,
28 				       union perf_event *event,
29 				       struct perf_sample *sample,
30 				       struct perf_tool *tool,
31 				       u64 file_offset);
32 
33 static int perf_session__open(struct perf_session *session)
34 {
35 	struct perf_data_file *file = session->file;
36 
37 	if (perf_session__read_header(session) < 0) {
38 		pr_err("incompatible file format (rerun with -v to learn more)\n");
39 		return -1;
40 	}
41 
42 	if (perf_data_file__is_pipe(file))
43 		return 0;
44 
45 	if (perf_header__has_feat(&session->header, HEADER_STAT))
46 		return 0;
47 
48 	if (!perf_evlist__valid_sample_type(session->evlist)) {
49 		pr_err("non matching sample_type\n");
50 		return -1;
51 	}
52 
53 	if (!perf_evlist__valid_sample_id_all(session->evlist)) {
54 		pr_err("non matching sample_id_all\n");
55 		return -1;
56 	}
57 
58 	if (!perf_evlist__valid_read_format(session->evlist)) {
59 		pr_err("non matching read_format\n");
60 		return -1;
61 	}
62 
63 	return 0;
64 }
65 
66 void perf_session__set_id_hdr_size(struct perf_session *session)
67 {
68 	u16 id_hdr_size = perf_evlist__id_hdr_size(session->evlist);
69 
70 	machines__set_id_hdr_size(&session->machines, id_hdr_size);
71 }
72 
73 int perf_session__create_kernel_maps(struct perf_session *session)
74 {
75 	int ret = machine__create_kernel_maps(&session->machines.host);
76 
77 	if (ret >= 0)
78 		ret = machines__create_guest_kernel_maps(&session->machines);
79 	return ret;
80 }
81 
82 static void perf_session__destroy_kernel_maps(struct perf_session *session)
83 {
84 	machines__destroy_kernel_maps(&session->machines);
85 }
86 
87 static bool perf_session__has_comm_exec(struct perf_session *session)
88 {
89 	struct perf_evsel *evsel;
90 
91 	evlist__for_each_entry(session->evlist, evsel) {
92 		if (evsel->attr.comm_exec)
93 			return true;
94 	}
95 
96 	return false;
97 }
98 
99 static void perf_session__set_comm_exec(struct perf_session *session)
100 {
101 	bool comm_exec = perf_session__has_comm_exec(session);
102 
103 	machines__set_comm_exec(&session->machines, comm_exec);
104 }
105 
106 static int ordered_events__deliver_event(struct ordered_events *oe,
107 					 struct ordered_event *event)
108 {
109 	struct perf_sample sample;
110 	struct perf_session *session = container_of(oe, struct perf_session,
111 						    ordered_events);
112 	int ret = perf_evlist__parse_sample(session->evlist, event->event, &sample);
113 
114 	if (ret) {
115 		pr_err("Can't parse sample, err = %d\n", ret);
116 		return ret;
117 	}
118 
119 	return perf_session__deliver_event(session, event->event, &sample,
120 					   session->tool, event->file_offset);
121 }
122 
123 struct perf_session *perf_session__new(struct perf_data_file *file,
124 				       bool repipe, struct perf_tool *tool)
125 {
126 	struct perf_session *session = zalloc(sizeof(*session));
127 
128 	if (!session)
129 		goto out;
130 
131 	session->repipe = repipe;
132 	session->tool   = tool;
133 	INIT_LIST_HEAD(&session->auxtrace_index);
134 	machines__init(&session->machines);
135 	ordered_events__init(&session->ordered_events, ordered_events__deliver_event);
136 
137 	if (file) {
138 		if (perf_data_file__open(file))
139 			goto out_delete;
140 
141 		session->file = file;
142 
143 		if (perf_data_file__is_read(file)) {
144 			if (perf_session__open(session) < 0)
145 				goto out_close;
146 
147 			/*
148 			 * set session attributes that are present in perf.data
149 			 * but not in pipe-mode.
150 			 */
151 			if (!file->is_pipe) {
152 				perf_session__set_id_hdr_size(session);
153 				perf_session__set_comm_exec(session);
154 			}
155 		}
156 	} else  {
157 		session->machines.host.env = &perf_env;
158 	}
159 
160 	if (!file || perf_data_file__is_write(file)) {
161 		/*
162 		 * In O_RDONLY mode this will be performed when reading the
163 		 * kernel MMAP event, in perf_event__process_mmap().
164 		 */
165 		if (perf_session__create_kernel_maps(session) < 0)
166 			pr_warning("Cannot read kernel map\n");
167 	}
168 
169 	/*
170 	 * In pipe-mode, evlist is empty until PERF_RECORD_HEADER_ATTR is
171 	 * processed, so perf_evlist__sample_id_all is not meaningful here.
172 	 */
173 	if ((!file || !file->is_pipe) && tool && tool->ordering_requires_timestamps &&
174 	    tool->ordered_events && !perf_evlist__sample_id_all(session->evlist)) {
175 		dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n");
176 		tool->ordered_events = false;
177 	}
178 
179 	return session;
180 
181  out_close:
182 	perf_data_file__close(file);
183  out_delete:
184 	perf_session__delete(session);
185  out:
186 	return NULL;
187 }
188 
189 static void perf_session__delete_threads(struct perf_session *session)
190 {
191 	machine__delete_threads(&session->machines.host);
192 }
193 
194 void perf_session__delete(struct perf_session *session)
195 {
196 	if (session == NULL)
197 		return;
198 	auxtrace__free(session);
199 	auxtrace_index__free(&session->auxtrace_index);
200 	perf_session__destroy_kernel_maps(session);
201 	perf_session__delete_threads(session);
202 	perf_env__exit(&session->header.env);
203 	machines__exit(&session->machines);
204 	if (session->file)
205 		perf_data_file__close(session->file);
206 	free(session);
207 }
208 
209 static int process_event_synth_tracing_data_stub(struct perf_tool *tool
210 						 __maybe_unused,
211 						 union perf_event *event
212 						 __maybe_unused,
213 						 struct perf_session *session
214 						__maybe_unused)
215 {
216 	dump_printf(": unhandled!\n");
217 	return 0;
218 }
219 
220 static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused,
221 					 union perf_event *event __maybe_unused,
222 					 struct perf_evlist **pevlist
223 					 __maybe_unused)
224 {
225 	dump_printf(": unhandled!\n");
226 	return 0;
227 }
228 
229 static int process_event_synth_event_update_stub(struct perf_tool *tool __maybe_unused,
230 						 union perf_event *event __maybe_unused,
231 						 struct perf_evlist **pevlist
232 						 __maybe_unused)
233 {
234 	if (dump_trace)
235 		perf_event__fprintf_event_update(event, stdout);
236 
237 	dump_printf(": unhandled!\n");
238 	return 0;
239 }
240 
241 static int process_event_sample_stub(struct perf_tool *tool __maybe_unused,
242 				     union perf_event *event __maybe_unused,
243 				     struct perf_sample *sample __maybe_unused,
244 				     struct perf_evsel *evsel __maybe_unused,
245 				     struct machine *machine __maybe_unused)
246 {
247 	dump_printf(": unhandled!\n");
248 	return 0;
249 }
250 
251 static int process_event_stub(struct perf_tool *tool __maybe_unused,
252 			      union perf_event *event __maybe_unused,
253 			      struct perf_sample *sample __maybe_unused,
254 			      struct machine *machine __maybe_unused)
255 {
256 	dump_printf(": unhandled!\n");
257 	return 0;
258 }
259 
260 static int process_finished_round_stub(struct perf_tool *tool __maybe_unused,
261 				       union perf_event *event __maybe_unused,
262 				       struct ordered_events *oe __maybe_unused)
263 {
264 	dump_printf(": unhandled!\n");
265 	return 0;
266 }
267 
268 static int process_finished_round(struct perf_tool *tool,
269 				  union perf_event *event,
270 				  struct ordered_events *oe);
271 
272 static int skipn(int fd, off_t n)
273 {
274 	char buf[4096];
275 	ssize_t ret;
276 
277 	while (n > 0) {
278 		ret = read(fd, buf, min(n, (off_t)sizeof(buf)));
279 		if (ret <= 0)
280 			return ret;
281 		n -= ret;
282 	}
283 
284 	return 0;
285 }
286 
287 static s64 process_event_auxtrace_stub(struct perf_tool *tool __maybe_unused,
288 				       union perf_event *event,
289 				       struct perf_session *session
290 				       __maybe_unused)
291 {
292 	dump_printf(": unhandled!\n");
293 	if (perf_data_file__is_pipe(session->file))
294 		skipn(perf_data_file__fd(session->file), event->auxtrace.size);
295 	return event->auxtrace.size;
296 }
297 
298 static int process_event_op2_stub(struct perf_tool *tool __maybe_unused,
299 				  union perf_event *event __maybe_unused,
300 				  struct perf_session *session __maybe_unused)
301 {
302 	dump_printf(": unhandled!\n");
303 	return 0;
304 }
305 
306 
307 static
308 int process_event_thread_map_stub(struct perf_tool *tool __maybe_unused,
309 				  union perf_event *event __maybe_unused,
310 				  struct perf_session *session __maybe_unused)
311 {
312 	if (dump_trace)
313 		perf_event__fprintf_thread_map(event, stdout);
314 
315 	dump_printf(": unhandled!\n");
316 	return 0;
317 }
318 
319 static
320 int process_event_cpu_map_stub(struct perf_tool *tool __maybe_unused,
321 			       union perf_event *event __maybe_unused,
322 			       struct perf_session *session __maybe_unused)
323 {
324 	if (dump_trace)
325 		perf_event__fprintf_cpu_map(event, stdout);
326 
327 	dump_printf(": unhandled!\n");
328 	return 0;
329 }
330 
331 static
332 int process_event_stat_config_stub(struct perf_tool *tool __maybe_unused,
333 				   union perf_event *event __maybe_unused,
334 				   struct perf_session *session __maybe_unused)
335 {
336 	if (dump_trace)
337 		perf_event__fprintf_stat_config(event, stdout);
338 
339 	dump_printf(": unhandled!\n");
340 	return 0;
341 }
342 
343 static int process_stat_stub(struct perf_tool *tool __maybe_unused,
344 			     union perf_event *event __maybe_unused,
345 			     struct perf_session *perf_session
346 			     __maybe_unused)
347 {
348 	if (dump_trace)
349 		perf_event__fprintf_stat(event, stdout);
350 
351 	dump_printf(": unhandled!\n");
352 	return 0;
353 }
354 
355 static int process_stat_round_stub(struct perf_tool *tool __maybe_unused,
356 				   union perf_event *event __maybe_unused,
357 				   struct perf_session *perf_session
358 				   __maybe_unused)
359 {
360 	if (dump_trace)
361 		perf_event__fprintf_stat_round(event, stdout);
362 
363 	dump_printf(": unhandled!\n");
364 	return 0;
365 }
366 
367 void perf_tool__fill_defaults(struct perf_tool *tool)
368 {
369 	if (tool->sample == NULL)
370 		tool->sample = process_event_sample_stub;
371 	if (tool->mmap == NULL)
372 		tool->mmap = process_event_stub;
373 	if (tool->mmap2 == NULL)
374 		tool->mmap2 = process_event_stub;
375 	if (tool->comm == NULL)
376 		tool->comm = process_event_stub;
377 	if (tool->namespaces == NULL)
378 		tool->namespaces = process_event_stub;
379 	if (tool->fork == NULL)
380 		tool->fork = process_event_stub;
381 	if (tool->exit == NULL)
382 		tool->exit = process_event_stub;
383 	if (tool->lost == NULL)
384 		tool->lost = perf_event__process_lost;
385 	if (tool->lost_samples == NULL)
386 		tool->lost_samples = perf_event__process_lost_samples;
387 	if (tool->aux == NULL)
388 		tool->aux = perf_event__process_aux;
389 	if (tool->itrace_start == NULL)
390 		tool->itrace_start = perf_event__process_itrace_start;
391 	if (tool->context_switch == NULL)
392 		tool->context_switch = perf_event__process_switch;
393 	if (tool->read == NULL)
394 		tool->read = process_event_sample_stub;
395 	if (tool->throttle == NULL)
396 		tool->throttle = process_event_stub;
397 	if (tool->unthrottle == NULL)
398 		tool->unthrottle = process_event_stub;
399 	if (tool->attr == NULL)
400 		tool->attr = process_event_synth_attr_stub;
401 	if (tool->event_update == NULL)
402 		tool->event_update = process_event_synth_event_update_stub;
403 	if (tool->tracing_data == NULL)
404 		tool->tracing_data = process_event_synth_tracing_data_stub;
405 	if (tool->build_id == NULL)
406 		tool->build_id = process_event_op2_stub;
407 	if (tool->finished_round == NULL) {
408 		if (tool->ordered_events)
409 			tool->finished_round = process_finished_round;
410 		else
411 			tool->finished_round = process_finished_round_stub;
412 	}
413 	if (tool->id_index == NULL)
414 		tool->id_index = process_event_op2_stub;
415 	if (tool->auxtrace_info == NULL)
416 		tool->auxtrace_info = process_event_op2_stub;
417 	if (tool->auxtrace == NULL)
418 		tool->auxtrace = process_event_auxtrace_stub;
419 	if (tool->auxtrace_error == NULL)
420 		tool->auxtrace_error = process_event_op2_stub;
421 	if (tool->thread_map == NULL)
422 		tool->thread_map = process_event_thread_map_stub;
423 	if (tool->cpu_map == NULL)
424 		tool->cpu_map = process_event_cpu_map_stub;
425 	if (tool->stat_config == NULL)
426 		tool->stat_config = process_event_stat_config_stub;
427 	if (tool->stat == NULL)
428 		tool->stat = process_stat_stub;
429 	if (tool->stat_round == NULL)
430 		tool->stat_round = process_stat_round_stub;
431 	if (tool->time_conv == NULL)
432 		tool->time_conv = process_event_op2_stub;
433 	if (tool->feature == NULL)
434 		tool->feature = process_event_op2_stub;
435 }
436 
437 static void swap_sample_id_all(union perf_event *event, void *data)
438 {
439 	void *end = (void *) event + event->header.size;
440 	int size = end - data;
441 
442 	BUG_ON(size % sizeof(u64));
443 	mem_bswap_64(data, size);
444 }
445 
446 static void perf_event__all64_swap(union perf_event *event,
447 				   bool sample_id_all __maybe_unused)
448 {
449 	struct perf_event_header *hdr = &event->header;
450 	mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr));
451 }
452 
453 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all)
454 {
455 	event->comm.pid = bswap_32(event->comm.pid);
456 	event->comm.tid = bswap_32(event->comm.tid);
457 
458 	if (sample_id_all) {
459 		void *data = &event->comm.comm;
460 
461 		data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
462 		swap_sample_id_all(event, data);
463 	}
464 }
465 
466 static void perf_event__mmap_swap(union perf_event *event,
467 				  bool sample_id_all)
468 {
469 	event->mmap.pid	  = bswap_32(event->mmap.pid);
470 	event->mmap.tid	  = bswap_32(event->mmap.tid);
471 	event->mmap.start = bswap_64(event->mmap.start);
472 	event->mmap.len	  = bswap_64(event->mmap.len);
473 	event->mmap.pgoff = bswap_64(event->mmap.pgoff);
474 
475 	if (sample_id_all) {
476 		void *data = &event->mmap.filename;
477 
478 		data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
479 		swap_sample_id_all(event, data);
480 	}
481 }
482 
483 static void perf_event__mmap2_swap(union perf_event *event,
484 				  bool sample_id_all)
485 {
486 	event->mmap2.pid   = bswap_32(event->mmap2.pid);
487 	event->mmap2.tid   = bswap_32(event->mmap2.tid);
488 	event->mmap2.start = bswap_64(event->mmap2.start);
489 	event->mmap2.len   = bswap_64(event->mmap2.len);
490 	event->mmap2.pgoff = bswap_64(event->mmap2.pgoff);
491 	event->mmap2.maj   = bswap_32(event->mmap2.maj);
492 	event->mmap2.min   = bswap_32(event->mmap2.min);
493 	event->mmap2.ino   = bswap_64(event->mmap2.ino);
494 
495 	if (sample_id_all) {
496 		void *data = &event->mmap2.filename;
497 
498 		data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
499 		swap_sample_id_all(event, data);
500 	}
501 }
502 static void perf_event__task_swap(union perf_event *event, bool sample_id_all)
503 {
504 	event->fork.pid	 = bswap_32(event->fork.pid);
505 	event->fork.tid	 = bswap_32(event->fork.tid);
506 	event->fork.ppid = bswap_32(event->fork.ppid);
507 	event->fork.ptid = bswap_32(event->fork.ptid);
508 	event->fork.time = bswap_64(event->fork.time);
509 
510 	if (sample_id_all)
511 		swap_sample_id_all(event, &event->fork + 1);
512 }
513 
514 static void perf_event__read_swap(union perf_event *event, bool sample_id_all)
515 {
516 	event->read.pid		 = bswap_32(event->read.pid);
517 	event->read.tid		 = bswap_32(event->read.tid);
518 	event->read.value	 = bswap_64(event->read.value);
519 	event->read.time_enabled = bswap_64(event->read.time_enabled);
520 	event->read.time_running = bswap_64(event->read.time_running);
521 	event->read.id		 = bswap_64(event->read.id);
522 
523 	if (sample_id_all)
524 		swap_sample_id_all(event, &event->read + 1);
525 }
526 
527 static void perf_event__aux_swap(union perf_event *event, bool sample_id_all)
528 {
529 	event->aux.aux_offset = bswap_64(event->aux.aux_offset);
530 	event->aux.aux_size   = bswap_64(event->aux.aux_size);
531 	event->aux.flags      = bswap_64(event->aux.flags);
532 
533 	if (sample_id_all)
534 		swap_sample_id_all(event, &event->aux + 1);
535 }
536 
537 static void perf_event__itrace_start_swap(union perf_event *event,
538 					  bool sample_id_all)
539 {
540 	event->itrace_start.pid	 = bswap_32(event->itrace_start.pid);
541 	event->itrace_start.tid	 = bswap_32(event->itrace_start.tid);
542 
543 	if (sample_id_all)
544 		swap_sample_id_all(event, &event->itrace_start + 1);
545 }
546 
547 static void perf_event__switch_swap(union perf_event *event, bool sample_id_all)
548 {
549 	if (event->header.type == PERF_RECORD_SWITCH_CPU_WIDE) {
550 		event->context_switch.next_prev_pid =
551 				bswap_32(event->context_switch.next_prev_pid);
552 		event->context_switch.next_prev_tid =
553 				bswap_32(event->context_switch.next_prev_tid);
554 	}
555 
556 	if (sample_id_all)
557 		swap_sample_id_all(event, &event->context_switch + 1);
558 }
559 
560 static void perf_event__throttle_swap(union perf_event *event,
561 				      bool sample_id_all)
562 {
563 	event->throttle.time	  = bswap_64(event->throttle.time);
564 	event->throttle.id	  = bswap_64(event->throttle.id);
565 	event->throttle.stream_id = bswap_64(event->throttle.stream_id);
566 
567 	if (sample_id_all)
568 		swap_sample_id_all(event, &event->throttle + 1);
569 }
570 
571 static u8 revbyte(u8 b)
572 {
573 	int rev = (b >> 4) | ((b & 0xf) << 4);
574 	rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2);
575 	rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1);
576 	return (u8) rev;
577 }
578 
579 /*
580  * XXX this is hack in attempt to carry flags bitfield
581  * through endian village. ABI says:
582  *
583  * Bit-fields are allocated from right to left (least to most significant)
584  * on little-endian implementations and from left to right (most to least
585  * significant) on big-endian implementations.
586  *
587  * The above seems to be byte specific, so we need to reverse each
588  * byte of the bitfield. 'Internet' also says this might be implementation
589  * specific and we probably need proper fix and carry perf_event_attr
590  * bitfield flags in separate data file FEAT_ section. Thought this seems
591  * to work for now.
592  */
593 static void swap_bitfield(u8 *p, unsigned len)
594 {
595 	unsigned i;
596 
597 	for (i = 0; i < len; i++) {
598 		*p = revbyte(*p);
599 		p++;
600 	}
601 }
602 
603 /* exported for swapping attributes in file header */
604 void perf_event__attr_swap(struct perf_event_attr *attr)
605 {
606 	attr->type		= bswap_32(attr->type);
607 	attr->size		= bswap_32(attr->size);
608 
609 #define bswap_safe(f, n) 					\
610 	(attr->size > (offsetof(struct perf_event_attr, f) + 	\
611 		       sizeof(attr->f) * (n)))
612 #define bswap_field(f, sz) 			\
613 do { 						\
614 	if (bswap_safe(f, 0))			\
615 		attr->f = bswap_##sz(attr->f);	\
616 } while(0)
617 #define bswap_field_16(f) bswap_field(f, 16)
618 #define bswap_field_32(f) bswap_field(f, 32)
619 #define bswap_field_64(f) bswap_field(f, 64)
620 
621 	bswap_field_64(config);
622 	bswap_field_64(sample_period);
623 	bswap_field_64(sample_type);
624 	bswap_field_64(read_format);
625 	bswap_field_32(wakeup_events);
626 	bswap_field_32(bp_type);
627 	bswap_field_64(bp_addr);
628 	bswap_field_64(bp_len);
629 	bswap_field_64(branch_sample_type);
630 	bswap_field_64(sample_regs_user);
631 	bswap_field_32(sample_stack_user);
632 	bswap_field_32(aux_watermark);
633 	bswap_field_16(sample_max_stack);
634 
635 	/*
636 	 * After read_format are bitfields. Check read_format because
637 	 * we are unable to use offsetof on bitfield.
638 	 */
639 	if (bswap_safe(read_format, 1))
640 		swap_bitfield((u8 *) (&attr->read_format + 1),
641 			      sizeof(u64));
642 #undef bswap_field_64
643 #undef bswap_field_32
644 #undef bswap_field
645 #undef bswap_safe
646 }
647 
648 static void perf_event__hdr_attr_swap(union perf_event *event,
649 				      bool sample_id_all __maybe_unused)
650 {
651 	size_t size;
652 
653 	perf_event__attr_swap(&event->attr.attr);
654 
655 	size = event->header.size;
656 	size -= (void *)&event->attr.id - (void *)event;
657 	mem_bswap_64(event->attr.id, size);
658 }
659 
660 static void perf_event__event_update_swap(union perf_event *event,
661 					  bool sample_id_all __maybe_unused)
662 {
663 	event->event_update.type = bswap_64(event->event_update.type);
664 	event->event_update.id   = bswap_64(event->event_update.id);
665 }
666 
667 static void perf_event__event_type_swap(union perf_event *event,
668 					bool sample_id_all __maybe_unused)
669 {
670 	event->event_type.event_type.event_id =
671 		bswap_64(event->event_type.event_type.event_id);
672 }
673 
674 static void perf_event__tracing_data_swap(union perf_event *event,
675 					  bool sample_id_all __maybe_unused)
676 {
677 	event->tracing_data.size = bswap_32(event->tracing_data.size);
678 }
679 
680 static void perf_event__auxtrace_info_swap(union perf_event *event,
681 					   bool sample_id_all __maybe_unused)
682 {
683 	size_t size;
684 
685 	event->auxtrace_info.type = bswap_32(event->auxtrace_info.type);
686 
687 	size = event->header.size;
688 	size -= (void *)&event->auxtrace_info.priv - (void *)event;
689 	mem_bswap_64(event->auxtrace_info.priv, size);
690 }
691 
692 static void perf_event__auxtrace_swap(union perf_event *event,
693 				      bool sample_id_all __maybe_unused)
694 {
695 	event->auxtrace.size      = bswap_64(event->auxtrace.size);
696 	event->auxtrace.offset    = bswap_64(event->auxtrace.offset);
697 	event->auxtrace.reference = bswap_64(event->auxtrace.reference);
698 	event->auxtrace.idx       = bswap_32(event->auxtrace.idx);
699 	event->auxtrace.tid       = bswap_32(event->auxtrace.tid);
700 	event->auxtrace.cpu       = bswap_32(event->auxtrace.cpu);
701 }
702 
703 static void perf_event__auxtrace_error_swap(union perf_event *event,
704 					    bool sample_id_all __maybe_unused)
705 {
706 	event->auxtrace_error.type = bswap_32(event->auxtrace_error.type);
707 	event->auxtrace_error.code = bswap_32(event->auxtrace_error.code);
708 	event->auxtrace_error.cpu  = bswap_32(event->auxtrace_error.cpu);
709 	event->auxtrace_error.pid  = bswap_32(event->auxtrace_error.pid);
710 	event->auxtrace_error.tid  = bswap_32(event->auxtrace_error.tid);
711 	event->auxtrace_error.ip   = bswap_64(event->auxtrace_error.ip);
712 }
713 
714 static void perf_event__thread_map_swap(union perf_event *event,
715 					bool sample_id_all __maybe_unused)
716 {
717 	unsigned i;
718 
719 	event->thread_map.nr = bswap_64(event->thread_map.nr);
720 
721 	for (i = 0; i < event->thread_map.nr; i++)
722 		event->thread_map.entries[i].pid = bswap_64(event->thread_map.entries[i].pid);
723 }
724 
725 static void perf_event__cpu_map_swap(union perf_event *event,
726 				     bool sample_id_all __maybe_unused)
727 {
728 	struct cpu_map_data *data = &event->cpu_map.data;
729 	struct cpu_map_entries *cpus;
730 	struct cpu_map_mask *mask;
731 	unsigned i;
732 
733 	data->type = bswap_64(data->type);
734 
735 	switch (data->type) {
736 	case PERF_CPU_MAP__CPUS:
737 		cpus = (struct cpu_map_entries *)data->data;
738 
739 		cpus->nr = bswap_16(cpus->nr);
740 
741 		for (i = 0; i < cpus->nr; i++)
742 			cpus->cpu[i] = bswap_16(cpus->cpu[i]);
743 		break;
744 	case PERF_CPU_MAP__MASK:
745 		mask = (struct cpu_map_mask *) data->data;
746 
747 		mask->nr = bswap_16(mask->nr);
748 		mask->long_size = bswap_16(mask->long_size);
749 
750 		switch (mask->long_size) {
751 		case 4: mem_bswap_32(&mask->mask, mask->nr); break;
752 		case 8: mem_bswap_64(&mask->mask, mask->nr); break;
753 		default:
754 			pr_err("cpu_map swap: unsupported long size\n");
755 		}
756 	default:
757 		break;
758 	}
759 }
760 
761 static void perf_event__stat_config_swap(union perf_event *event,
762 					 bool sample_id_all __maybe_unused)
763 {
764 	u64 size;
765 
766 	size  = event->stat_config.nr * sizeof(event->stat_config.data[0]);
767 	size += 1; /* nr item itself */
768 	mem_bswap_64(&event->stat_config.nr, size);
769 }
770 
771 static void perf_event__stat_swap(union perf_event *event,
772 				  bool sample_id_all __maybe_unused)
773 {
774 	event->stat.id     = bswap_64(event->stat.id);
775 	event->stat.thread = bswap_32(event->stat.thread);
776 	event->stat.cpu    = bswap_32(event->stat.cpu);
777 	event->stat.val    = bswap_64(event->stat.val);
778 	event->stat.ena    = bswap_64(event->stat.ena);
779 	event->stat.run    = bswap_64(event->stat.run);
780 }
781 
782 static void perf_event__stat_round_swap(union perf_event *event,
783 					bool sample_id_all __maybe_unused)
784 {
785 	event->stat_round.type = bswap_64(event->stat_round.type);
786 	event->stat_round.time = bswap_64(event->stat_round.time);
787 }
788 
789 typedef void (*perf_event__swap_op)(union perf_event *event,
790 				    bool sample_id_all);
791 
792 static perf_event__swap_op perf_event__swap_ops[] = {
793 	[PERF_RECORD_MMAP]		  = perf_event__mmap_swap,
794 	[PERF_RECORD_MMAP2]		  = perf_event__mmap2_swap,
795 	[PERF_RECORD_COMM]		  = perf_event__comm_swap,
796 	[PERF_RECORD_FORK]		  = perf_event__task_swap,
797 	[PERF_RECORD_EXIT]		  = perf_event__task_swap,
798 	[PERF_RECORD_LOST]		  = perf_event__all64_swap,
799 	[PERF_RECORD_READ]		  = perf_event__read_swap,
800 	[PERF_RECORD_THROTTLE]		  = perf_event__throttle_swap,
801 	[PERF_RECORD_UNTHROTTLE]	  = perf_event__throttle_swap,
802 	[PERF_RECORD_SAMPLE]		  = perf_event__all64_swap,
803 	[PERF_RECORD_AUX]		  = perf_event__aux_swap,
804 	[PERF_RECORD_ITRACE_START]	  = perf_event__itrace_start_swap,
805 	[PERF_RECORD_LOST_SAMPLES]	  = perf_event__all64_swap,
806 	[PERF_RECORD_SWITCH]		  = perf_event__switch_swap,
807 	[PERF_RECORD_SWITCH_CPU_WIDE]	  = perf_event__switch_swap,
808 	[PERF_RECORD_HEADER_ATTR]	  = perf_event__hdr_attr_swap,
809 	[PERF_RECORD_HEADER_EVENT_TYPE]	  = perf_event__event_type_swap,
810 	[PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
811 	[PERF_RECORD_HEADER_BUILD_ID]	  = NULL,
812 	[PERF_RECORD_ID_INDEX]		  = perf_event__all64_swap,
813 	[PERF_RECORD_AUXTRACE_INFO]	  = perf_event__auxtrace_info_swap,
814 	[PERF_RECORD_AUXTRACE]		  = perf_event__auxtrace_swap,
815 	[PERF_RECORD_AUXTRACE_ERROR]	  = perf_event__auxtrace_error_swap,
816 	[PERF_RECORD_THREAD_MAP]	  = perf_event__thread_map_swap,
817 	[PERF_RECORD_CPU_MAP]		  = perf_event__cpu_map_swap,
818 	[PERF_RECORD_STAT_CONFIG]	  = perf_event__stat_config_swap,
819 	[PERF_RECORD_STAT]		  = perf_event__stat_swap,
820 	[PERF_RECORD_STAT_ROUND]	  = perf_event__stat_round_swap,
821 	[PERF_RECORD_EVENT_UPDATE]	  = perf_event__event_update_swap,
822 	[PERF_RECORD_TIME_CONV]		  = perf_event__all64_swap,
823 	[PERF_RECORD_HEADER_MAX]	  = NULL,
824 };
825 
826 /*
827  * When perf record finishes a pass on every buffers, it records this pseudo
828  * event.
829  * We record the max timestamp t found in the pass n.
830  * Assuming these timestamps are monotonic across cpus, we know that if
831  * a buffer still has events with timestamps below t, they will be all
832  * available and then read in the pass n + 1.
833  * Hence when we start to read the pass n + 2, we can safely flush every
834  * events with timestamps below t.
835  *
836  *    ============ PASS n =================
837  *       CPU 0         |   CPU 1
838  *                     |
839  *    cnt1 timestamps  |   cnt2 timestamps
840  *          1          |         2
841  *          2          |         3
842  *          -          |         4  <--- max recorded
843  *
844  *    ============ PASS n + 1 ==============
845  *       CPU 0         |   CPU 1
846  *                     |
847  *    cnt1 timestamps  |   cnt2 timestamps
848  *          3          |         5
849  *          4          |         6
850  *          5          |         7 <---- max recorded
851  *
852  *      Flush every events below timestamp 4
853  *
854  *    ============ PASS n + 2 ==============
855  *       CPU 0         |   CPU 1
856  *                     |
857  *    cnt1 timestamps  |   cnt2 timestamps
858  *          6          |         8
859  *          7          |         9
860  *          -          |         10
861  *
862  *      Flush every events below timestamp 7
863  *      etc...
864  */
865 static int process_finished_round(struct perf_tool *tool __maybe_unused,
866 				  union perf_event *event __maybe_unused,
867 				  struct ordered_events *oe)
868 {
869 	if (dump_trace)
870 		fprintf(stdout, "\n");
871 	return ordered_events__flush(oe, OE_FLUSH__ROUND);
872 }
873 
874 int perf_session__queue_event(struct perf_session *s, union perf_event *event,
875 			      struct perf_sample *sample, u64 file_offset)
876 {
877 	return ordered_events__queue(&s->ordered_events, event, sample, file_offset);
878 }
879 
880 static void callchain__lbr_callstack_printf(struct perf_sample *sample)
881 {
882 	struct ip_callchain *callchain = sample->callchain;
883 	struct branch_stack *lbr_stack = sample->branch_stack;
884 	u64 kernel_callchain_nr = callchain->nr;
885 	unsigned int i;
886 
887 	for (i = 0; i < kernel_callchain_nr; i++) {
888 		if (callchain->ips[i] == PERF_CONTEXT_USER)
889 			break;
890 	}
891 
892 	if ((i != kernel_callchain_nr) && lbr_stack->nr) {
893 		u64 total_nr;
894 		/*
895 		 * LBR callstack can only get user call chain,
896 		 * i is kernel call chain number,
897 		 * 1 is PERF_CONTEXT_USER.
898 		 *
899 		 * The user call chain is stored in LBR registers.
900 		 * LBR are pair registers. The caller is stored
901 		 * in "from" register, while the callee is stored
902 		 * in "to" register.
903 		 * For example, there is a call stack
904 		 * "A"->"B"->"C"->"D".
905 		 * The LBR registers will recorde like
906 		 * "C"->"D", "B"->"C", "A"->"B".
907 		 * So only the first "to" register and all "from"
908 		 * registers are needed to construct the whole stack.
909 		 */
910 		total_nr = i + 1 + lbr_stack->nr + 1;
911 		kernel_callchain_nr = i + 1;
912 
913 		printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr);
914 
915 		for (i = 0; i < kernel_callchain_nr; i++)
916 			printf("..... %2d: %016" PRIx64 "\n",
917 			       i, callchain->ips[i]);
918 
919 		printf("..... %2d: %016" PRIx64 "\n",
920 		       (int)(kernel_callchain_nr), lbr_stack->entries[0].to);
921 		for (i = 0; i < lbr_stack->nr; i++)
922 			printf("..... %2d: %016" PRIx64 "\n",
923 			       (int)(i + kernel_callchain_nr + 1), lbr_stack->entries[i].from);
924 	}
925 }
926 
927 static void callchain__printf(struct perf_evsel *evsel,
928 			      struct perf_sample *sample)
929 {
930 	unsigned int i;
931 	struct ip_callchain *callchain = sample->callchain;
932 
933 	if (perf_evsel__has_branch_callstack(evsel))
934 		callchain__lbr_callstack_printf(sample);
935 
936 	printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr);
937 
938 	for (i = 0; i < callchain->nr; i++)
939 		printf("..... %2d: %016" PRIx64 "\n",
940 		       i, callchain->ips[i]);
941 }
942 
943 static void branch_stack__printf(struct perf_sample *sample)
944 {
945 	uint64_t i;
946 
947 	printf("... branch stack: nr:%" PRIu64 "\n", sample->branch_stack->nr);
948 
949 	for (i = 0; i < sample->branch_stack->nr; i++) {
950 		struct branch_entry *e = &sample->branch_stack->entries[i];
951 
952 		printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x\n",
953 			i, e->from, e->to,
954 			(unsigned short)e->flags.cycles,
955 			e->flags.mispred ? "M" : " ",
956 			e->flags.predicted ? "P" : " ",
957 			e->flags.abort ? "A" : " ",
958 			e->flags.in_tx ? "T" : " ",
959 			(unsigned)e->flags.reserved);
960 	}
961 }
962 
963 static void regs_dump__printf(u64 mask, u64 *regs)
964 {
965 	unsigned rid, i = 0;
966 
967 	for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) {
968 		u64 val = regs[i++];
969 
970 		printf(".... %-5s 0x%" PRIx64 "\n",
971 		       perf_reg_name(rid), val);
972 	}
973 }
974 
975 static const char *regs_abi[] = {
976 	[PERF_SAMPLE_REGS_ABI_NONE] = "none",
977 	[PERF_SAMPLE_REGS_ABI_32] = "32-bit",
978 	[PERF_SAMPLE_REGS_ABI_64] = "64-bit",
979 };
980 
981 static inline const char *regs_dump_abi(struct regs_dump *d)
982 {
983 	if (d->abi > PERF_SAMPLE_REGS_ABI_64)
984 		return "unknown";
985 
986 	return regs_abi[d->abi];
987 }
988 
989 static void regs__printf(const char *type, struct regs_dump *regs)
990 {
991 	u64 mask = regs->mask;
992 
993 	printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n",
994 	       type,
995 	       mask,
996 	       regs_dump_abi(regs));
997 
998 	regs_dump__printf(mask, regs->regs);
999 }
1000 
1001 static void regs_user__printf(struct perf_sample *sample)
1002 {
1003 	struct regs_dump *user_regs = &sample->user_regs;
1004 
1005 	if (user_regs->regs)
1006 		regs__printf("user", user_regs);
1007 }
1008 
1009 static void regs_intr__printf(struct perf_sample *sample)
1010 {
1011 	struct regs_dump *intr_regs = &sample->intr_regs;
1012 
1013 	if (intr_regs->regs)
1014 		regs__printf("intr", intr_regs);
1015 }
1016 
1017 static void stack_user__printf(struct stack_dump *dump)
1018 {
1019 	printf("... ustack: size %" PRIu64 ", offset 0x%x\n",
1020 	       dump->size, dump->offset);
1021 }
1022 
1023 static void perf_evlist__print_tstamp(struct perf_evlist *evlist,
1024 				       union perf_event *event,
1025 				       struct perf_sample *sample)
1026 {
1027 	u64 sample_type = __perf_evlist__combined_sample_type(evlist);
1028 
1029 	if (event->header.type != PERF_RECORD_SAMPLE &&
1030 	    !perf_evlist__sample_id_all(evlist)) {
1031 		fputs("-1 -1 ", stdout);
1032 		return;
1033 	}
1034 
1035 	if ((sample_type & PERF_SAMPLE_CPU))
1036 		printf("%u ", sample->cpu);
1037 
1038 	if (sample_type & PERF_SAMPLE_TIME)
1039 		printf("%" PRIu64 " ", sample->time);
1040 }
1041 
1042 static void sample_read__printf(struct perf_sample *sample, u64 read_format)
1043 {
1044 	printf("... sample_read:\n");
1045 
1046 	if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1047 		printf("...... time enabled %016" PRIx64 "\n",
1048 		       sample->read.time_enabled);
1049 
1050 	if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1051 		printf("...... time running %016" PRIx64 "\n",
1052 		       sample->read.time_running);
1053 
1054 	if (read_format & PERF_FORMAT_GROUP) {
1055 		u64 i;
1056 
1057 		printf(".... group nr %" PRIu64 "\n", sample->read.group.nr);
1058 
1059 		for (i = 0; i < sample->read.group.nr; i++) {
1060 			struct sample_read_value *value;
1061 
1062 			value = &sample->read.group.values[i];
1063 			printf("..... id %016" PRIx64
1064 			       ", value %016" PRIx64 "\n",
1065 			       value->id, value->value);
1066 		}
1067 	} else
1068 		printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n",
1069 			sample->read.one.id, sample->read.one.value);
1070 }
1071 
1072 static void dump_event(struct perf_evlist *evlist, union perf_event *event,
1073 		       u64 file_offset, struct perf_sample *sample)
1074 {
1075 	if (!dump_trace)
1076 		return;
1077 
1078 	printf("\n%#" PRIx64 " [%#x]: event: %d\n",
1079 	       file_offset, event->header.size, event->header.type);
1080 
1081 	trace_event(event);
1082 
1083 	if (sample)
1084 		perf_evlist__print_tstamp(evlist, event, sample);
1085 
1086 	printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset,
1087 	       event->header.size, perf_event__name(event->header.type));
1088 }
1089 
1090 static void dump_sample(struct perf_evsel *evsel, union perf_event *event,
1091 			struct perf_sample *sample)
1092 {
1093 	u64 sample_type;
1094 
1095 	if (!dump_trace)
1096 		return;
1097 
1098 	printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n",
1099 	       event->header.misc, sample->pid, sample->tid, sample->ip,
1100 	       sample->period, sample->addr);
1101 
1102 	sample_type = evsel->attr.sample_type;
1103 
1104 	if (sample_type & PERF_SAMPLE_CALLCHAIN)
1105 		callchain__printf(evsel, sample);
1106 
1107 	if ((sample_type & PERF_SAMPLE_BRANCH_STACK) && !perf_evsel__has_branch_callstack(evsel))
1108 		branch_stack__printf(sample);
1109 
1110 	if (sample_type & PERF_SAMPLE_REGS_USER)
1111 		regs_user__printf(sample);
1112 
1113 	if (sample_type & PERF_SAMPLE_REGS_INTR)
1114 		regs_intr__printf(sample);
1115 
1116 	if (sample_type & PERF_SAMPLE_STACK_USER)
1117 		stack_user__printf(&sample->user_stack);
1118 
1119 	if (sample_type & PERF_SAMPLE_WEIGHT)
1120 		printf("... weight: %" PRIu64 "\n", sample->weight);
1121 
1122 	if (sample_type & PERF_SAMPLE_DATA_SRC)
1123 		printf(" . data_src: 0x%"PRIx64"\n", sample->data_src);
1124 
1125 	if (sample_type & PERF_SAMPLE_PHYS_ADDR)
1126 		printf(" .. phys_addr: 0x%"PRIx64"\n", sample->phys_addr);
1127 
1128 	if (sample_type & PERF_SAMPLE_TRANSACTION)
1129 		printf("... transaction: %" PRIx64 "\n", sample->transaction);
1130 
1131 	if (sample_type & PERF_SAMPLE_READ)
1132 		sample_read__printf(sample, evsel->attr.read_format);
1133 }
1134 
1135 static void dump_read(struct perf_evsel *evsel, union perf_event *event)
1136 {
1137 	struct read_event *read_event = &event->read;
1138 	u64 read_format;
1139 
1140 	if (!dump_trace)
1141 		return;
1142 
1143 	printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid,
1144 	       evsel ? perf_evsel__name(evsel) : "FAIL",
1145 	       event->read.value);
1146 
1147 	read_format = evsel->attr.read_format;
1148 
1149 	if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1150 		printf("... time enabled : %" PRIu64 "\n", read_event->time_enabled);
1151 
1152 	if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1153 		printf("... time running : %" PRIu64 "\n", read_event->time_running);
1154 
1155 	if (read_format & PERF_FORMAT_ID)
1156 		printf("... id           : %" PRIu64 "\n", read_event->id);
1157 }
1158 
1159 static struct machine *machines__find_for_cpumode(struct machines *machines,
1160 					       union perf_event *event,
1161 					       struct perf_sample *sample)
1162 {
1163 	struct machine *machine;
1164 
1165 	if (perf_guest &&
1166 	    ((sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL) ||
1167 	     (sample->cpumode == PERF_RECORD_MISC_GUEST_USER))) {
1168 		u32 pid;
1169 
1170 		if (event->header.type == PERF_RECORD_MMAP
1171 		    || event->header.type == PERF_RECORD_MMAP2)
1172 			pid = event->mmap.pid;
1173 		else
1174 			pid = sample->pid;
1175 
1176 		machine = machines__find(machines, pid);
1177 		if (!machine)
1178 			machine = machines__findnew(machines, DEFAULT_GUEST_KERNEL_ID);
1179 		return machine;
1180 	}
1181 
1182 	return &machines->host;
1183 }
1184 
1185 static int deliver_sample_value(struct perf_evlist *evlist,
1186 				struct perf_tool *tool,
1187 				union perf_event *event,
1188 				struct perf_sample *sample,
1189 				struct sample_read_value *v,
1190 				struct machine *machine)
1191 {
1192 	struct perf_sample_id *sid = perf_evlist__id2sid(evlist, v->id);
1193 
1194 	if (sid) {
1195 		sample->id     = v->id;
1196 		sample->period = v->value - sid->period;
1197 		sid->period    = v->value;
1198 	}
1199 
1200 	if (!sid || sid->evsel == NULL) {
1201 		++evlist->stats.nr_unknown_id;
1202 		return 0;
1203 	}
1204 
1205 	return tool->sample(tool, event, sample, sid->evsel, machine);
1206 }
1207 
1208 static int deliver_sample_group(struct perf_evlist *evlist,
1209 				struct perf_tool *tool,
1210 				union  perf_event *event,
1211 				struct perf_sample *sample,
1212 				struct machine *machine)
1213 {
1214 	int ret = -EINVAL;
1215 	u64 i;
1216 
1217 	for (i = 0; i < sample->read.group.nr; i++) {
1218 		ret = deliver_sample_value(evlist, tool, event, sample,
1219 					   &sample->read.group.values[i],
1220 					   machine);
1221 		if (ret)
1222 			break;
1223 	}
1224 
1225 	return ret;
1226 }
1227 
1228 static int
1229  perf_evlist__deliver_sample(struct perf_evlist *evlist,
1230 			     struct perf_tool *tool,
1231 			     union  perf_event *event,
1232 			     struct perf_sample *sample,
1233 			     struct perf_evsel *evsel,
1234 			     struct machine *machine)
1235 {
1236 	/* We know evsel != NULL. */
1237 	u64 sample_type = evsel->attr.sample_type;
1238 	u64 read_format = evsel->attr.read_format;
1239 
1240 	/* Standard sample delivery. */
1241 	if (!(sample_type & PERF_SAMPLE_READ))
1242 		return tool->sample(tool, event, sample, evsel, machine);
1243 
1244 	/* For PERF_SAMPLE_READ we have either single or group mode. */
1245 	if (read_format & PERF_FORMAT_GROUP)
1246 		return deliver_sample_group(evlist, tool, event, sample,
1247 					    machine);
1248 	else
1249 		return deliver_sample_value(evlist, tool, event, sample,
1250 					    &sample->read.one, machine);
1251 }
1252 
1253 static int machines__deliver_event(struct machines *machines,
1254 				   struct perf_evlist *evlist,
1255 				   union perf_event *event,
1256 				   struct perf_sample *sample,
1257 				   struct perf_tool *tool, u64 file_offset)
1258 {
1259 	struct perf_evsel *evsel;
1260 	struct machine *machine;
1261 
1262 	dump_event(evlist, event, file_offset, sample);
1263 
1264 	evsel = perf_evlist__id2evsel(evlist, sample->id);
1265 
1266 	machine = machines__find_for_cpumode(machines, event, sample);
1267 
1268 	switch (event->header.type) {
1269 	case PERF_RECORD_SAMPLE:
1270 		if (evsel == NULL) {
1271 			++evlist->stats.nr_unknown_id;
1272 			return 0;
1273 		}
1274 		dump_sample(evsel, event, sample);
1275 		if (machine == NULL) {
1276 			++evlist->stats.nr_unprocessable_samples;
1277 			return 0;
1278 		}
1279 		return perf_evlist__deliver_sample(evlist, tool, event, sample, evsel, machine);
1280 	case PERF_RECORD_MMAP:
1281 		return tool->mmap(tool, event, sample, machine);
1282 	case PERF_RECORD_MMAP2:
1283 		if (event->header.misc & PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT)
1284 			++evlist->stats.nr_proc_map_timeout;
1285 		return tool->mmap2(tool, event, sample, machine);
1286 	case PERF_RECORD_COMM:
1287 		return tool->comm(tool, event, sample, machine);
1288 	case PERF_RECORD_NAMESPACES:
1289 		return tool->namespaces(tool, event, sample, machine);
1290 	case PERF_RECORD_FORK:
1291 		return tool->fork(tool, event, sample, machine);
1292 	case PERF_RECORD_EXIT:
1293 		return tool->exit(tool, event, sample, machine);
1294 	case PERF_RECORD_LOST:
1295 		if (tool->lost == perf_event__process_lost)
1296 			evlist->stats.total_lost += event->lost.lost;
1297 		return tool->lost(tool, event, sample, machine);
1298 	case PERF_RECORD_LOST_SAMPLES:
1299 		if (tool->lost_samples == perf_event__process_lost_samples)
1300 			evlist->stats.total_lost_samples += event->lost_samples.lost;
1301 		return tool->lost_samples(tool, event, sample, machine);
1302 	case PERF_RECORD_READ:
1303 		dump_read(evsel, event);
1304 		return tool->read(tool, event, sample, evsel, machine);
1305 	case PERF_RECORD_THROTTLE:
1306 		return tool->throttle(tool, event, sample, machine);
1307 	case PERF_RECORD_UNTHROTTLE:
1308 		return tool->unthrottle(tool, event, sample, machine);
1309 	case PERF_RECORD_AUX:
1310 		if (tool->aux == perf_event__process_aux) {
1311 			if (event->aux.flags & PERF_AUX_FLAG_TRUNCATED)
1312 				evlist->stats.total_aux_lost += 1;
1313 			if (event->aux.flags & PERF_AUX_FLAG_PARTIAL)
1314 				evlist->stats.total_aux_partial += 1;
1315 		}
1316 		return tool->aux(tool, event, sample, machine);
1317 	case PERF_RECORD_ITRACE_START:
1318 		return tool->itrace_start(tool, event, sample, machine);
1319 	case PERF_RECORD_SWITCH:
1320 	case PERF_RECORD_SWITCH_CPU_WIDE:
1321 		return tool->context_switch(tool, event, sample, machine);
1322 	default:
1323 		++evlist->stats.nr_unknown_events;
1324 		return -1;
1325 	}
1326 }
1327 
1328 static int perf_session__deliver_event(struct perf_session *session,
1329 				       union perf_event *event,
1330 				       struct perf_sample *sample,
1331 				       struct perf_tool *tool,
1332 				       u64 file_offset)
1333 {
1334 	int ret;
1335 
1336 	ret = auxtrace__process_event(session, event, sample, tool);
1337 	if (ret < 0)
1338 		return ret;
1339 	if (ret > 0)
1340 		return 0;
1341 
1342 	return machines__deliver_event(&session->machines, session->evlist,
1343 				       event, sample, tool, file_offset);
1344 }
1345 
1346 static s64 perf_session__process_user_event(struct perf_session *session,
1347 					    union perf_event *event,
1348 					    u64 file_offset)
1349 {
1350 	struct ordered_events *oe = &session->ordered_events;
1351 	struct perf_tool *tool = session->tool;
1352 	int fd = perf_data_file__fd(session->file);
1353 	int err;
1354 
1355 	dump_event(session->evlist, event, file_offset, NULL);
1356 
1357 	/* These events are processed right away */
1358 	switch (event->header.type) {
1359 	case PERF_RECORD_HEADER_ATTR:
1360 		err = tool->attr(tool, event, &session->evlist);
1361 		if (err == 0) {
1362 			perf_session__set_id_hdr_size(session);
1363 			perf_session__set_comm_exec(session);
1364 		}
1365 		return err;
1366 	case PERF_RECORD_EVENT_UPDATE:
1367 		return tool->event_update(tool, event, &session->evlist);
1368 	case PERF_RECORD_HEADER_EVENT_TYPE:
1369 		/*
1370 		 * Depreceated, but we need to handle it for sake
1371 		 * of old data files create in pipe mode.
1372 		 */
1373 		return 0;
1374 	case PERF_RECORD_HEADER_TRACING_DATA:
1375 		/* setup for reading amidst mmap */
1376 		lseek(fd, file_offset, SEEK_SET);
1377 		return tool->tracing_data(tool, event, session);
1378 	case PERF_RECORD_HEADER_BUILD_ID:
1379 		return tool->build_id(tool, event, session);
1380 	case PERF_RECORD_FINISHED_ROUND:
1381 		return tool->finished_round(tool, event, oe);
1382 	case PERF_RECORD_ID_INDEX:
1383 		return tool->id_index(tool, event, session);
1384 	case PERF_RECORD_AUXTRACE_INFO:
1385 		return tool->auxtrace_info(tool, event, session);
1386 	case PERF_RECORD_AUXTRACE:
1387 		/* setup for reading amidst mmap */
1388 		lseek(fd, file_offset + event->header.size, SEEK_SET);
1389 		return tool->auxtrace(tool, event, session);
1390 	case PERF_RECORD_AUXTRACE_ERROR:
1391 		perf_session__auxtrace_error_inc(session, event);
1392 		return tool->auxtrace_error(tool, event, session);
1393 	case PERF_RECORD_THREAD_MAP:
1394 		return tool->thread_map(tool, event, session);
1395 	case PERF_RECORD_CPU_MAP:
1396 		return tool->cpu_map(tool, event, session);
1397 	case PERF_RECORD_STAT_CONFIG:
1398 		return tool->stat_config(tool, event, session);
1399 	case PERF_RECORD_STAT:
1400 		return tool->stat(tool, event, session);
1401 	case PERF_RECORD_STAT_ROUND:
1402 		return tool->stat_round(tool, event, session);
1403 	case PERF_RECORD_TIME_CONV:
1404 		session->time_conv = event->time_conv;
1405 		return tool->time_conv(tool, event, session);
1406 	case PERF_RECORD_HEADER_FEATURE:
1407 		return tool->feature(tool, event, session);
1408 	default:
1409 		return -EINVAL;
1410 	}
1411 }
1412 
1413 int perf_session__deliver_synth_event(struct perf_session *session,
1414 				      union perf_event *event,
1415 				      struct perf_sample *sample)
1416 {
1417 	struct perf_evlist *evlist = session->evlist;
1418 	struct perf_tool *tool = session->tool;
1419 
1420 	events_stats__inc(&evlist->stats, event->header.type);
1421 
1422 	if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1423 		return perf_session__process_user_event(session, event, 0);
1424 
1425 	return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0);
1426 }
1427 
1428 static void event_swap(union perf_event *event, bool sample_id_all)
1429 {
1430 	perf_event__swap_op swap;
1431 
1432 	swap = perf_event__swap_ops[event->header.type];
1433 	if (swap)
1434 		swap(event, sample_id_all);
1435 }
1436 
1437 int perf_session__peek_event(struct perf_session *session, off_t file_offset,
1438 			     void *buf, size_t buf_sz,
1439 			     union perf_event **event_ptr,
1440 			     struct perf_sample *sample)
1441 {
1442 	union perf_event *event;
1443 	size_t hdr_sz, rest;
1444 	int fd;
1445 
1446 	if (session->one_mmap && !session->header.needs_swap) {
1447 		event = file_offset - session->one_mmap_offset +
1448 			session->one_mmap_addr;
1449 		goto out_parse_sample;
1450 	}
1451 
1452 	if (perf_data_file__is_pipe(session->file))
1453 		return -1;
1454 
1455 	fd = perf_data_file__fd(session->file);
1456 	hdr_sz = sizeof(struct perf_event_header);
1457 
1458 	if (buf_sz < hdr_sz)
1459 		return -1;
1460 
1461 	if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 ||
1462 	    readn(fd, buf, hdr_sz) != (ssize_t)hdr_sz)
1463 		return -1;
1464 
1465 	event = (union perf_event *)buf;
1466 
1467 	if (session->header.needs_swap)
1468 		perf_event_header__bswap(&event->header);
1469 
1470 	if (event->header.size < hdr_sz || event->header.size > buf_sz)
1471 		return -1;
1472 
1473 	rest = event->header.size - hdr_sz;
1474 
1475 	if (readn(fd, buf, rest) != (ssize_t)rest)
1476 		return -1;
1477 
1478 	if (session->header.needs_swap)
1479 		event_swap(event, perf_evlist__sample_id_all(session->evlist));
1480 
1481 out_parse_sample:
1482 
1483 	if (sample && event->header.type < PERF_RECORD_USER_TYPE_START &&
1484 	    perf_evlist__parse_sample(session->evlist, event, sample))
1485 		return -1;
1486 
1487 	*event_ptr = event;
1488 
1489 	return 0;
1490 }
1491 
1492 static s64 perf_session__process_event(struct perf_session *session,
1493 				       union perf_event *event, u64 file_offset)
1494 {
1495 	struct perf_evlist *evlist = session->evlist;
1496 	struct perf_tool *tool = session->tool;
1497 	struct perf_sample sample;
1498 	int ret;
1499 
1500 	if (session->header.needs_swap)
1501 		event_swap(event, perf_evlist__sample_id_all(evlist));
1502 
1503 	if (event->header.type >= PERF_RECORD_HEADER_MAX)
1504 		return -EINVAL;
1505 
1506 	events_stats__inc(&evlist->stats, event->header.type);
1507 
1508 	if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1509 		return perf_session__process_user_event(session, event, file_offset);
1510 
1511 	/*
1512 	 * For all kernel events we get the sample data
1513 	 */
1514 	ret = perf_evlist__parse_sample(evlist, event, &sample);
1515 	if (ret)
1516 		return ret;
1517 
1518 	if (tool->ordered_events) {
1519 		ret = perf_session__queue_event(session, event, &sample, file_offset);
1520 		if (ret != -ETIME)
1521 			return ret;
1522 	}
1523 
1524 	return perf_session__deliver_event(session, event, &sample, tool,
1525 					   file_offset);
1526 }
1527 
1528 void perf_event_header__bswap(struct perf_event_header *hdr)
1529 {
1530 	hdr->type = bswap_32(hdr->type);
1531 	hdr->misc = bswap_16(hdr->misc);
1532 	hdr->size = bswap_16(hdr->size);
1533 }
1534 
1535 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid)
1536 {
1537 	return machine__findnew_thread(&session->machines.host, -1, pid);
1538 }
1539 
1540 int perf_session__register_idle_thread(struct perf_session *session)
1541 {
1542 	struct thread *thread;
1543 	int err = 0;
1544 
1545 	thread = machine__findnew_thread(&session->machines.host, 0, 0);
1546 	if (thread == NULL || thread__set_comm(thread, "swapper", 0)) {
1547 		pr_err("problem inserting idle task.\n");
1548 		err = -1;
1549 	}
1550 
1551 	if (thread == NULL || thread__set_namespaces(thread, 0, NULL)) {
1552 		pr_err("problem inserting idle task.\n");
1553 		err = -1;
1554 	}
1555 
1556 	/* machine__findnew_thread() got the thread, so put it */
1557 	thread__put(thread);
1558 	return err;
1559 }
1560 
1561 static void
1562 perf_session__warn_order(const struct perf_session *session)
1563 {
1564 	const struct ordered_events *oe = &session->ordered_events;
1565 	struct perf_evsel *evsel;
1566 	bool should_warn = true;
1567 
1568 	evlist__for_each_entry(session->evlist, evsel) {
1569 		if (evsel->attr.write_backward)
1570 			should_warn = false;
1571 	}
1572 
1573 	if (!should_warn)
1574 		return;
1575 	if (oe->nr_unordered_events != 0)
1576 		ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events);
1577 }
1578 
1579 static void perf_session__warn_about_errors(const struct perf_session *session)
1580 {
1581 	const struct events_stats *stats = &session->evlist->stats;
1582 
1583 	if (session->tool->lost == perf_event__process_lost &&
1584 	    stats->nr_events[PERF_RECORD_LOST] != 0) {
1585 		ui__warning("Processed %d events and lost %d chunks!\n\n"
1586 			    "Check IO/CPU overload!\n\n",
1587 			    stats->nr_events[0],
1588 			    stats->nr_events[PERF_RECORD_LOST]);
1589 	}
1590 
1591 	if (session->tool->lost_samples == perf_event__process_lost_samples) {
1592 		double drop_rate;
1593 
1594 		drop_rate = (double)stats->total_lost_samples /
1595 			    (double) (stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples);
1596 		if (drop_rate > 0.05) {
1597 			ui__warning("Processed %" PRIu64 " samples and lost %3.2f%% samples!\n\n",
1598 				    stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples,
1599 				    drop_rate * 100.0);
1600 		}
1601 	}
1602 
1603 	if (session->tool->aux == perf_event__process_aux &&
1604 	    stats->total_aux_lost != 0) {
1605 		ui__warning("AUX data lost %" PRIu64 " times out of %u!\n\n",
1606 			    stats->total_aux_lost,
1607 			    stats->nr_events[PERF_RECORD_AUX]);
1608 	}
1609 
1610 	if (session->tool->aux == perf_event__process_aux &&
1611 	    stats->total_aux_partial != 0) {
1612 		bool vmm_exclusive = false;
1613 
1614 		(void)sysfs__read_bool("module/kvm_intel/parameters/vmm_exclusive",
1615 		                       &vmm_exclusive);
1616 
1617 		ui__warning("AUX data had gaps in it %" PRIu64 " times out of %u!\n\n"
1618 		            "Are you running a KVM guest in the background?%s\n\n",
1619 			    stats->total_aux_partial,
1620 			    stats->nr_events[PERF_RECORD_AUX],
1621 			    vmm_exclusive ?
1622 			    "\nReloading kvm_intel module with vmm_exclusive=0\n"
1623 			    "will reduce the gaps to only guest's timeslices." :
1624 			    "");
1625 	}
1626 
1627 	if (stats->nr_unknown_events != 0) {
1628 		ui__warning("Found %u unknown events!\n\n"
1629 			    "Is this an older tool processing a perf.data "
1630 			    "file generated by a more recent tool?\n\n"
1631 			    "If that is not the case, consider "
1632 			    "reporting to linux-kernel@vger.kernel.org.\n\n",
1633 			    stats->nr_unknown_events);
1634 	}
1635 
1636 	if (stats->nr_unknown_id != 0) {
1637 		ui__warning("%u samples with id not present in the header\n",
1638 			    stats->nr_unknown_id);
1639 	}
1640 
1641 	if (stats->nr_invalid_chains != 0) {
1642 		ui__warning("Found invalid callchains!\n\n"
1643 			    "%u out of %u events were discarded for this reason.\n\n"
1644 			    "Consider reporting to linux-kernel@vger.kernel.org.\n\n",
1645 			    stats->nr_invalid_chains,
1646 			    stats->nr_events[PERF_RECORD_SAMPLE]);
1647 	}
1648 
1649 	if (stats->nr_unprocessable_samples != 0) {
1650 		ui__warning("%u unprocessable samples recorded.\n"
1651 			    "Do you have a KVM guest running and not using 'perf kvm'?\n",
1652 			    stats->nr_unprocessable_samples);
1653 	}
1654 
1655 	perf_session__warn_order(session);
1656 
1657 	events_stats__auxtrace_error_warn(stats);
1658 
1659 	if (stats->nr_proc_map_timeout != 0) {
1660 		ui__warning("%d map information files for pre-existing threads were\n"
1661 			    "not processed, if there are samples for addresses they\n"
1662 			    "will not be resolved, you may find out which are these\n"
1663 			    "threads by running with -v and redirecting the output\n"
1664 			    "to a file.\n"
1665 			    "The time limit to process proc map is too short?\n"
1666 			    "Increase it by --proc-map-timeout\n",
1667 			    stats->nr_proc_map_timeout);
1668 	}
1669 }
1670 
1671 static int perf_session__flush_thread_stack(struct thread *thread,
1672 					    void *p __maybe_unused)
1673 {
1674 	return thread_stack__flush(thread);
1675 }
1676 
1677 static int perf_session__flush_thread_stacks(struct perf_session *session)
1678 {
1679 	return machines__for_each_thread(&session->machines,
1680 					 perf_session__flush_thread_stack,
1681 					 NULL);
1682 }
1683 
1684 volatile int session_done;
1685 
1686 static int __perf_session__process_pipe_events(struct perf_session *session)
1687 {
1688 	struct ordered_events *oe = &session->ordered_events;
1689 	struct perf_tool *tool = session->tool;
1690 	int fd = perf_data_file__fd(session->file);
1691 	union perf_event *event;
1692 	uint32_t size, cur_size = 0;
1693 	void *buf = NULL;
1694 	s64 skip = 0;
1695 	u64 head;
1696 	ssize_t err;
1697 	void *p;
1698 
1699 	perf_tool__fill_defaults(tool);
1700 
1701 	head = 0;
1702 	cur_size = sizeof(union perf_event);
1703 
1704 	buf = malloc(cur_size);
1705 	if (!buf)
1706 		return -errno;
1707 	ordered_events__set_copy_on_queue(oe, true);
1708 more:
1709 	event = buf;
1710 	err = readn(fd, event, sizeof(struct perf_event_header));
1711 	if (err <= 0) {
1712 		if (err == 0)
1713 			goto done;
1714 
1715 		pr_err("failed to read event header\n");
1716 		goto out_err;
1717 	}
1718 
1719 	if (session->header.needs_swap)
1720 		perf_event_header__bswap(&event->header);
1721 
1722 	size = event->header.size;
1723 	if (size < sizeof(struct perf_event_header)) {
1724 		pr_err("bad event header size\n");
1725 		goto out_err;
1726 	}
1727 
1728 	if (size > cur_size) {
1729 		void *new = realloc(buf, size);
1730 		if (!new) {
1731 			pr_err("failed to allocate memory to read event\n");
1732 			goto out_err;
1733 		}
1734 		buf = new;
1735 		cur_size = size;
1736 		event = buf;
1737 	}
1738 	p = event;
1739 	p += sizeof(struct perf_event_header);
1740 
1741 	if (size - sizeof(struct perf_event_header)) {
1742 		err = readn(fd, p, size - sizeof(struct perf_event_header));
1743 		if (err <= 0) {
1744 			if (err == 0) {
1745 				pr_err("unexpected end of event stream\n");
1746 				goto done;
1747 			}
1748 
1749 			pr_err("failed to read event data\n");
1750 			goto out_err;
1751 		}
1752 	}
1753 
1754 	if ((skip = perf_session__process_event(session, event, head)) < 0) {
1755 		pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1756 		       head, event->header.size, event->header.type);
1757 		err = -EINVAL;
1758 		goto out_err;
1759 	}
1760 
1761 	head += size;
1762 
1763 	if (skip > 0)
1764 		head += skip;
1765 
1766 	if (!session_done())
1767 		goto more;
1768 done:
1769 	/* do the final flush for ordered samples */
1770 	err = ordered_events__flush(oe, OE_FLUSH__FINAL);
1771 	if (err)
1772 		goto out_err;
1773 	err = auxtrace__flush_events(session, tool);
1774 	if (err)
1775 		goto out_err;
1776 	err = perf_session__flush_thread_stacks(session);
1777 out_err:
1778 	free(buf);
1779 	perf_session__warn_about_errors(session);
1780 	ordered_events__free(&session->ordered_events);
1781 	auxtrace__free_events(session);
1782 	return err;
1783 }
1784 
1785 static union perf_event *
1786 fetch_mmaped_event(struct perf_session *session,
1787 		   u64 head, size_t mmap_size, char *buf)
1788 {
1789 	union perf_event *event;
1790 
1791 	/*
1792 	 * Ensure we have enough space remaining to read
1793 	 * the size of the event in the headers.
1794 	 */
1795 	if (head + sizeof(event->header) > mmap_size)
1796 		return NULL;
1797 
1798 	event = (union perf_event *)(buf + head);
1799 
1800 	if (session->header.needs_swap)
1801 		perf_event_header__bswap(&event->header);
1802 
1803 	if (head + event->header.size > mmap_size) {
1804 		/* We're not fetching the event so swap back again */
1805 		if (session->header.needs_swap)
1806 			perf_event_header__bswap(&event->header);
1807 		return NULL;
1808 	}
1809 
1810 	return event;
1811 }
1812 
1813 /*
1814  * On 64bit we can mmap the data file in one go. No need for tiny mmap
1815  * slices. On 32bit we use 32MB.
1816  */
1817 #if BITS_PER_LONG == 64
1818 #define MMAP_SIZE ULLONG_MAX
1819 #define NUM_MMAPS 1
1820 #else
1821 #define MMAP_SIZE (32 * 1024 * 1024ULL)
1822 #define NUM_MMAPS 128
1823 #endif
1824 
1825 static int __perf_session__process_events(struct perf_session *session,
1826 					  u64 data_offset, u64 data_size,
1827 					  u64 file_size)
1828 {
1829 	struct ordered_events *oe = &session->ordered_events;
1830 	struct perf_tool *tool = session->tool;
1831 	int fd = perf_data_file__fd(session->file);
1832 	u64 head, page_offset, file_offset, file_pos, size;
1833 	int err, mmap_prot, mmap_flags, map_idx = 0;
1834 	size_t	mmap_size;
1835 	char *buf, *mmaps[NUM_MMAPS];
1836 	union perf_event *event;
1837 	struct ui_progress prog;
1838 	s64 skip;
1839 
1840 	perf_tool__fill_defaults(tool);
1841 
1842 	page_offset = page_size * (data_offset / page_size);
1843 	file_offset = page_offset;
1844 	head = data_offset - page_offset;
1845 
1846 	if (data_size == 0)
1847 		goto out;
1848 
1849 	if (data_offset + data_size < file_size)
1850 		file_size = data_offset + data_size;
1851 
1852 	ui_progress__init(&prog, file_size, "Processing events...");
1853 
1854 	mmap_size = MMAP_SIZE;
1855 	if (mmap_size > file_size) {
1856 		mmap_size = file_size;
1857 		session->one_mmap = true;
1858 	}
1859 
1860 	memset(mmaps, 0, sizeof(mmaps));
1861 
1862 	mmap_prot  = PROT_READ;
1863 	mmap_flags = MAP_SHARED;
1864 
1865 	if (session->header.needs_swap) {
1866 		mmap_prot  |= PROT_WRITE;
1867 		mmap_flags = MAP_PRIVATE;
1868 	}
1869 remap:
1870 	buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, fd,
1871 		   file_offset);
1872 	if (buf == MAP_FAILED) {
1873 		pr_err("failed to mmap file\n");
1874 		err = -errno;
1875 		goto out_err;
1876 	}
1877 	mmaps[map_idx] = buf;
1878 	map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1);
1879 	file_pos = file_offset + head;
1880 	if (session->one_mmap) {
1881 		session->one_mmap_addr = buf;
1882 		session->one_mmap_offset = file_offset;
1883 	}
1884 
1885 more:
1886 	event = fetch_mmaped_event(session, head, mmap_size, buf);
1887 	if (!event) {
1888 		if (mmaps[map_idx]) {
1889 			munmap(mmaps[map_idx], mmap_size);
1890 			mmaps[map_idx] = NULL;
1891 		}
1892 
1893 		page_offset = page_size * (head / page_size);
1894 		file_offset += page_offset;
1895 		head -= page_offset;
1896 		goto remap;
1897 	}
1898 
1899 	size = event->header.size;
1900 
1901 	if (size < sizeof(struct perf_event_header) ||
1902 	    (skip = perf_session__process_event(session, event, file_pos)) < 0) {
1903 		pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1904 		       file_offset + head, event->header.size,
1905 		       event->header.type);
1906 		err = -EINVAL;
1907 		goto out_err;
1908 	}
1909 
1910 	if (skip)
1911 		size += skip;
1912 
1913 	head += size;
1914 	file_pos += size;
1915 
1916 	ui_progress__update(&prog, size);
1917 
1918 	if (session_done())
1919 		goto out;
1920 
1921 	if (file_pos < file_size)
1922 		goto more;
1923 
1924 out:
1925 	/* do the final flush for ordered samples */
1926 	err = ordered_events__flush(oe, OE_FLUSH__FINAL);
1927 	if (err)
1928 		goto out_err;
1929 	err = auxtrace__flush_events(session, tool);
1930 	if (err)
1931 		goto out_err;
1932 	err = perf_session__flush_thread_stacks(session);
1933 out_err:
1934 	ui_progress__finish();
1935 	perf_session__warn_about_errors(session);
1936 	/*
1937 	 * We may switching perf.data output, make ordered_events
1938 	 * reusable.
1939 	 */
1940 	ordered_events__reinit(&session->ordered_events);
1941 	auxtrace__free_events(session);
1942 	session->one_mmap = false;
1943 	return err;
1944 }
1945 
1946 int perf_session__process_events(struct perf_session *session)
1947 {
1948 	u64 size = perf_data_file__size(session->file);
1949 	int err;
1950 
1951 	if (perf_session__register_idle_thread(session) < 0)
1952 		return -ENOMEM;
1953 
1954 	if (!perf_data_file__is_pipe(session->file))
1955 		err = __perf_session__process_events(session,
1956 						     session->header.data_offset,
1957 						     session->header.data_size, size);
1958 	else
1959 		err = __perf_session__process_pipe_events(session);
1960 
1961 	return err;
1962 }
1963 
1964 bool perf_session__has_traces(struct perf_session *session, const char *msg)
1965 {
1966 	struct perf_evsel *evsel;
1967 
1968 	evlist__for_each_entry(session->evlist, evsel) {
1969 		if (evsel->attr.type == PERF_TYPE_TRACEPOINT)
1970 			return true;
1971 	}
1972 
1973 	pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg);
1974 	return false;
1975 }
1976 
1977 int maps__set_kallsyms_ref_reloc_sym(struct map **maps,
1978 				     const char *symbol_name, u64 addr)
1979 {
1980 	char *bracket;
1981 	int i;
1982 	struct ref_reloc_sym *ref;
1983 
1984 	ref = zalloc(sizeof(struct ref_reloc_sym));
1985 	if (ref == NULL)
1986 		return -ENOMEM;
1987 
1988 	ref->name = strdup(symbol_name);
1989 	if (ref->name == NULL) {
1990 		free(ref);
1991 		return -ENOMEM;
1992 	}
1993 
1994 	bracket = strchr(ref->name, ']');
1995 	if (bracket)
1996 		*bracket = '\0';
1997 
1998 	ref->addr = addr;
1999 
2000 	for (i = 0; i < MAP__NR_TYPES; ++i) {
2001 		struct kmap *kmap = map__kmap(maps[i]);
2002 
2003 		if (!kmap)
2004 			continue;
2005 		kmap->ref_reloc_sym = ref;
2006 	}
2007 
2008 	return 0;
2009 }
2010 
2011 size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp)
2012 {
2013 	return machines__fprintf_dsos(&session->machines, fp);
2014 }
2015 
2016 size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp,
2017 					  bool (skip)(struct dso *dso, int parm), int parm)
2018 {
2019 	return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm);
2020 }
2021 
2022 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp)
2023 {
2024 	size_t ret;
2025 	const char *msg = "";
2026 
2027 	if (perf_header__has_feat(&session->header, HEADER_AUXTRACE))
2028 		msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)";
2029 
2030 	ret = fprintf(fp, "\nAggregated stats:%s\n", msg);
2031 
2032 	ret += events_stats__fprintf(&session->evlist->stats, fp);
2033 	return ret;
2034 }
2035 
2036 size_t perf_session__fprintf(struct perf_session *session, FILE *fp)
2037 {
2038 	/*
2039 	 * FIXME: Here we have to actually print all the machines in this
2040 	 * session, not just the host...
2041 	 */
2042 	return machine__fprintf(&session->machines.host, fp);
2043 }
2044 
2045 struct perf_evsel *perf_session__find_first_evtype(struct perf_session *session,
2046 					      unsigned int type)
2047 {
2048 	struct perf_evsel *pos;
2049 
2050 	evlist__for_each_entry(session->evlist, pos) {
2051 		if (pos->attr.type == type)
2052 			return pos;
2053 	}
2054 	return NULL;
2055 }
2056 
2057 int perf_session__cpu_bitmap(struct perf_session *session,
2058 			     const char *cpu_list, unsigned long *cpu_bitmap)
2059 {
2060 	int i, err = -1;
2061 	struct cpu_map *map;
2062 
2063 	for (i = 0; i < PERF_TYPE_MAX; ++i) {
2064 		struct perf_evsel *evsel;
2065 
2066 		evsel = perf_session__find_first_evtype(session, i);
2067 		if (!evsel)
2068 			continue;
2069 
2070 		if (!(evsel->attr.sample_type & PERF_SAMPLE_CPU)) {
2071 			pr_err("File does not contain CPU events. "
2072 			       "Remove -C option to proceed.\n");
2073 			return -1;
2074 		}
2075 	}
2076 
2077 	map = cpu_map__new(cpu_list);
2078 	if (map == NULL) {
2079 		pr_err("Invalid cpu_list\n");
2080 		return -1;
2081 	}
2082 
2083 	for (i = 0; i < map->nr; i++) {
2084 		int cpu = map->map[i];
2085 
2086 		if (cpu >= MAX_NR_CPUS) {
2087 			pr_err("Requested CPU %d too large. "
2088 			       "Consider raising MAX_NR_CPUS\n", cpu);
2089 			goto out_delete_map;
2090 		}
2091 
2092 		set_bit(cpu, cpu_bitmap);
2093 	}
2094 
2095 	err = 0;
2096 
2097 out_delete_map:
2098 	cpu_map__put(map);
2099 	return err;
2100 }
2101 
2102 void perf_session__fprintf_info(struct perf_session *session, FILE *fp,
2103 				bool full)
2104 {
2105 	if (session == NULL || fp == NULL)
2106 		return;
2107 
2108 	fprintf(fp, "# ========\n");
2109 	perf_header__fprintf_info(session, fp, full);
2110 	fprintf(fp, "# ========\n#\n");
2111 }
2112 
2113 
2114 int __perf_session__set_tracepoints_handlers(struct perf_session *session,
2115 					     const struct perf_evsel_str_handler *assocs,
2116 					     size_t nr_assocs)
2117 {
2118 	struct perf_evsel *evsel;
2119 	size_t i;
2120 	int err;
2121 
2122 	for (i = 0; i < nr_assocs; i++) {
2123 		/*
2124 		 * Adding a handler for an event not in the session,
2125 		 * just ignore it.
2126 		 */
2127 		evsel = perf_evlist__find_tracepoint_by_name(session->evlist, assocs[i].name);
2128 		if (evsel == NULL)
2129 			continue;
2130 
2131 		err = -EEXIST;
2132 		if (evsel->handler != NULL)
2133 			goto out;
2134 		evsel->handler = assocs[i].handler;
2135 	}
2136 
2137 	err = 0;
2138 out:
2139 	return err;
2140 }
2141 
2142 int perf_event__process_id_index(struct perf_tool *tool __maybe_unused,
2143 				 union perf_event *event,
2144 				 struct perf_session *session)
2145 {
2146 	struct perf_evlist *evlist = session->evlist;
2147 	struct id_index_event *ie = &event->id_index;
2148 	size_t i, nr, max_nr;
2149 
2150 	max_nr = (ie->header.size - sizeof(struct id_index_event)) /
2151 		 sizeof(struct id_index_entry);
2152 	nr = ie->nr;
2153 	if (nr > max_nr)
2154 		return -EINVAL;
2155 
2156 	if (dump_trace)
2157 		fprintf(stdout, " nr: %zu\n", nr);
2158 
2159 	for (i = 0; i < nr; i++) {
2160 		struct id_index_entry *e = &ie->entries[i];
2161 		struct perf_sample_id *sid;
2162 
2163 		if (dump_trace) {
2164 			fprintf(stdout,	" ... id: %"PRIu64, e->id);
2165 			fprintf(stdout,	"  idx: %"PRIu64, e->idx);
2166 			fprintf(stdout,	"  cpu: %"PRId64, e->cpu);
2167 			fprintf(stdout,	"  tid: %"PRId64"\n", e->tid);
2168 		}
2169 
2170 		sid = perf_evlist__id2sid(evlist, e->id);
2171 		if (!sid)
2172 			return -ENOENT;
2173 		sid->idx = e->idx;
2174 		sid->cpu = e->cpu;
2175 		sid->tid = e->tid;
2176 	}
2177 	return 0;
2178 }
2179 
2180 int perf_event__synthesize_id_index(struct perf_tool *tool,
2181 				    perf_event__handler_t process,
2182 				    struct perf_evlist *evlist,
2183 				    struct machine *machine)
2184 {
2185 	union perf_event *ev;
2186 	struct perf_evsel *evsel;
2187 	size_t nr = 0, i = 0, sz, max_nr, n;
2188 	int err;
2189 
2190 	pr_debug2("Synthesizing id index\n");
2191 
2192 	max_nr = (UINT16_MAX - sizeof(struct id_index_event)) /
2193 		 sizeof(struct id_index_entry);
2194 
2195 	evlist__for_each_entry(evlist, evsel)
2196 		nr += evsel->ids;
2197 
2198 	n = nr > max_nr ? max_nr : nr;
2199 	sz = sizeof(struct id_index_event) + n * sizeof(struct id_index_entry);
2200 	ev = zalloc(sz);
2201 	if (!ev)
2202 		return -ENOMEM;
2203 
2204 	ev->id_index.header.type = PERF_RECORD_ID_INDEX;
2205 	ev->id_index.header.size = sz;
2206 	ev->id_index.nr = n;
2207 
2208 	evlist__for_each_entry(evlist, evsel) {
2209 		u32 j;
2210 
2211 		for (j = 0; j < evsel->ids; j++) {
2212 			struct id_index_entry *e;
2213 			struct perf_sample_id *sid;
2214 
2215 			if (i >= n) {
2216 				err = process(tool, ev, NULL, machine);
2217 				if (err)
2218 					goto out_err;
2219 				nr -= n;
2220 				i = 0;
2221 			}
2222 
2223 			e = &ev->id_index.entries[i++];
2224 
2225 			e->id = evsel->id[j];
2226 
2227 			sid = perf_evlist__id2sid(evlist, e->id);
2228 			if (!sid) {
2229 				free(ev);
2230 				return -ENOENT;
2231 			}
2232 
2233 			e->idx = sid->idx;
2234 			e->cpu = sid->cpu;
2235 			e->tid = sid->tid;
2236 		}
2237 	}
2238 
2239 	sz = sizeof(struct id_index_event) + nr * sizeof(struct id_index_entry);
2240 	ev->id_index.header.size = sz;
2241 	ev->id_index.nr = nr;
2242 
2243 	err = process(tool, ev, NULL, machine);
2244 out_err:
2245 	free(ev);
2246 
2247 	return err;
2248 }
2249