xref: /openbmc/linux/tools/perf/util/session.c (revision 29c37341)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <errno.h>
3 #include <inttypes.h>
4 #include <linux/err.h>
5 #include <linux/kernel.h>
6 #include <linux/zalloc.h>
7 #include <api/fs/fs.h>
8 
9 #include <byteswap.h>
10 #include <unistd.h>
11 #include <sys/types.h>
12 #include <sys/mman.h>
13 #include <perf/cpumap.h>
14 
15 #include "map_symbol.h"
16 #include "branch.h"
17 #include "debug.h"
18 #include "evlist.h"
19 #include "evsel.h"
20 #include "memswap.h"
21 #include "map.h"
22 #include "symbol.h"
23 #include "session.h"
24 #include "tool.h"
25 #include "perf_regs.h"
26 #include "asm/bug.h"
27 #include "auxtrace.h"
28 #include "thread.h"
29 #include "thread-stack.h"
30 #include "sample-raw.h"
31 #include "stat.h"
32 #include "ui/progress.h"
33 #include "../perf.h"
34 #include "arch/common.h"
35 #include <internal/lib.h>
36 
37 #ifdef HAVE_ZSTD_SUPPORT
38 static int perf_session__process_compressed_event(struct perf_session *session,
39 						  union perf_event *event, u64 file_offset)
40 {
41 	void *src;
42 	size_t decomp_size, src_size;
43 	u64 decomp_last_rem = 0;
44 	size_t mmap_len, decomp_len = session->header.env.comp_mmap_len;
45 	struct decomp *decomp, *decomp_last = session->decomp_last;
46 
47 	if (decomp_last) {
48 		decomp_last_rem = decomp_last->size - decomp_last->head;
49 		decomp_len += decomp_last_rem;
50 	}
51 
52 	mmap_len = sizeof(struct decomp) + decomp_len;
53 	decomp = mmap(NULL, mmap_len, PROT_READ|PROT_WRITE,
54 		      MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);
55 	if (decomp == MAP_FAILED) {
56 		pr_err("Couldn't allocate memory for decompression\n");
57 		return -1;
58 	}
59 
60 	decomp->file_pos = file_offset;
61 	decomp->mmap_len = mmap_len;
62 	decomp->head = 0;
63 
64 	if (decomp_last_rem) {
65 		memcpy(decomp->data, &(decomp_last->data[decomp_last->head]), decomp_last_rem);
66 		decomp->size = decomp_last_rem;
67 	}
68 
69 	src = (void *)event + sizeof(struct perf_record_compressed);
70 	src_size = event->pack.header.size - sizeof(struct perf_record_compressed);
71 
72 	decomp_size = zstd_decompress_stream(&(session->zstd_data), src, src_size,
73 				&(decomp->data[decomp_last_rem]), decomp_len - decomp_last_rem);
74 	if (!decomp_size) {
75 		munmap(decomp, mmap_len);
76 		pr_err("Couldn't decompress data\n");
77 		return -1;
78 	}
79 
80 	decomp->size += decomp_size;
81 
82 	if (session->decomp == NULL) {
83 		session->decomp = decomp;
84 		session->decomp_last = decomp;
85 	} else {
86 		session->decomp_last->next = decomp;
87 		session->decomp_last = decomp;
88 	}
89 
90 	pr_debug("decomp (B): %ld to %ld\n", src_size, decomp_size);
91 
92 	return 0;
93 }
94 #else /* !HAVE_ZSTD_SUPPORT */
95 #define perf_session__process_compressed_event perf_session__process_compressed_event_stub
96 #endif
97 
98 static int perf_session__deliver_event(struct perf_session *session,
99 				       union perf_event *event,
100 				       struct perf_tool *tool,
101 				       u64 file_offset);
102 
103 static int perf_session__open(struct perf_session *session)
104 {
105 	struct perf_data *data = session->data;
106 
107 	if (perf_session__read_header(session) < 0) {
108 		pr_err("incompatible file format (rerun with -v to learn more)\n");
109 		return -1;
110 	}
111 
112 	if (perf_data__is_pipe(data))
113 		return 0;
114 
115 	if (perf_header__has_feat(&session->header, HEADER_STAT))
116 		return 0;
117 
118 	if (!evlist__valid_sample_type(session->evlist)) {
119 		pr_err("non matching sample_type\n");
120 		return -1;
121 	}
122 
123 	if (!evlist__valid_sample_id_all(session->evlist)) {
124 		pr_err("non matching sample_id_all\n");
125 		return -1;
126 	}
127 
128 	if (!perf_evlist__valid_read_format(session->evlist)) {
129 		pr_err("non matching read_format\n");
130 		return -1;
131 	}
132 
133 	return 0;
134 }
135 
136 void perf_session__set_id_hdr_size(struct perf_session *session)
137 {
138 	u16 id_hdr_size = perf_evlist__id_hdr_size(session->evlist);
139 
140 	machines__set_id_hdr_size(&session->machines, id_hdr_size);
141 }
142 
143 int perf_session__create_kernel_maps(struct perf_session *session)
144 {
145 	int ret = machine__create_kernel_maps(&session->machines.host);
146 
147 	if (ret >= 0)
148 		ret = machines__create_guest_kernel_maps(&session->machines);
149 	return ret;
150 }
151 
152 static void perf_session__destroy_kernel_maps(struct perf_session *session)
153 {
154 	machines__destroy_kernel_maps(&session->machines);
155 }
156 
157 static bool perf_session__has_comm_exec(struct perf_session *session)
158 {
159 	struct evsel *evsel;
160 
161 	evlist__for_each_entry(session->evlist, evsel) {
162 		if (evsel->core.attr.comm_exec)
163 			return true;
164 	}
165 
166 	return false;
167 }
168 
169 static void perf_session__set_comm_exec(struct perf_session *session)
170 {
171 	bool comm_exec = perf_session__has_comm_exec(session);
172 
173 	machines__set_comm_exec(&session->machines, comm_exec);
174 }
175 
176 static int ordered_events__deliver_event(struct ordered_events *oe,
177 					 struct ordered_event *event)
178 {
179 	struct perf_session *session = container_of(oe, struct perf_session,
180 						    ordered_events);
181 
182 	return perf_session__deliver_event(session, event->event,
183 					   session->tool, event->file_offset);
184 }
185 
186 struct perf_session *perf_session__new(struct perf_data *data,
187 				       bool repipe, struct perf_tool *tool)
188 {
189 	int ret = -ENOMEM;
190 	struct perf_session *session = zalloc(sizeof(*session));
191 
192 	if (!session)
193 		goto out;
194 
195 	session->repipe = repipe;
196 	session->tool   = tool;
197 	INIT_LIST_HEAD(&session->auxtrace_index);
198 	machines__init(&session->machines);
199 	ordered_events__init(&session->ordered_events,
200 			     ordered_events__deliver_event, NULL);
201 
202 	perf_env__init(&session->header.env);
203 	if (data) {
204 		ret = perf_data__open(data);
205 		if (ret < 0)
206 			goto out_delete;
207 
208 		session->data = data;
209 
210 		if (perf_data__is_read(data)) {
211 			ret = perf_session__open(session);
212 			if (ret < 0)
213 				goto out_delete;
214 
215 			/*
216 			 * set session attributes that are present in perf.data
217 			 * but not in pipe-mode.
218 			 */
219 			if (!data->is_pipe) {
220 				perf_session__set_id_hdr_size(session);
221 				perf_session__set_comm_exec(session);
222 			}
223 
224 			perf_evlist__init_trace_event_sample_raw(session->evlist);
225 
226 			/* Open the directory data. */
227 			if (data->is_dir) {
228 				ret = perf_data__open_dir(data);
229 				if (ret)
230 					goto out_delete;
231 			}
232 
233 			if (!symbol_conf.kallsyms_name &&
234 			    !symbol_conf.vmlinux_name)
235 				symbol_conf.kallsyms_name = perf_data__kallsyms_name(data);
236 		}
237 	} else  {
238 		session->machines.host.env = &perf_env;
239 	}
240 
241 	session->machines.host.single_address_space =
242 		perf_env__single_address_space(session->machines.host.env);
243 
244 	if (!data || perf_data__is_write(data)) {
245 		/*
246 		 * In O_RDONLY mode this will be performed when reading the
247 		 * kernel MMAP event, in perf_event__process_mmap().
248 		 */
249 		if (perf_session__create_kernel_maps(session) < 0)
250 			pr_warning("Cannot read kernel map\n");
251 	}
252 
253 	/*
254 	 * In pipe-mode, evlist is empty until PERF_RECORD_HEADER_ATTR is
255 	 * processed, so evlist__sample_id_all is not meaningful here.
256 	 */
257 	if ((!data || !data->is_pipe) && tool && tool->ordering_requires_timestamps &&
258 	    tool->ordered_events && !evlist__sample_id_all(session->evlist)) {
259 		dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n");
260 		tool->ordered_events = false;
261 	}
262 
263 	return session;
264 
265  out_delete:
266 	perf_session__delete(session);
267  out:
268 	return ERR_PTR(ret);
269 }
270 
271 static void perf_session__delete_threads(struct perf_session *session)
272 {
273 	machine__delete_threads(&session->machines.host);
274 }
275 
276 static void perf_session__release_decomp_events(struct perf_session *session)
277 {
278 	struct decomp *next, *decomp;
279 	size_t mmap_len;
280 	next = session->decomp;
281 	do {
282 		decomp = next;
283 		if (decomp == NULL)
284 			break;
285 		next = decomp->next;
286 		mmap_len = decomp->mmap_len;
287 		munmap(decomp, mmap_len);
288 	} while (1);
289 }
290 
291 void perf_session__delete(struct perf_session *session)
292 {
293 	if (session == NULL)
294 		return;
295 	auxtrace__free(session);
296 	auxtrace_index__free(&session->auxtrace_index);
297 	perf_session__destroy_kernel_maps(session);
298 	perf_session__delete_threads(session);
299 	perf_session__release_decomp_events(session);
300 	perf_env__exit(&session->header.env);
301 	machines__exit(&session->machines);
302 	if (session->data)
303 		perf_data__close(session->data);
304 	free(session);
305 }
306 
307 static int process_event_synth_tracing_data_stub(struct perf_session *session
308 						 __maybe_unused,
309 						 union perf_event *event
310 						 __maybe_unused)
311 {
312 	dump_printf(": unhandled!\n");
313 	return 0;
314 }
315 
316 static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused,
317 					 union perf_event *event __maybe_unused,
318 					 struct evlist **pevlist
319 					 __maybe_unused)
320 {
321 	dump_printf(": unhandled!\n");
322 	return 0;
323 }
324 
325 static int process_event_synth_event_update_stub(struct perf_tool *tool __maybe_unused,
326 						 union perf_event *event __maybe_unused,
327 						 struct evlist **pevlist
328 						 __maybe_unused)
329 {
330 	if (dump_trace)
331 		perf_event__fprintf_event_update(event, stdout);
332 
333 	dump_printf(": unhandled!\n");
334 	return 0;
335 }
336 
337 static int process_event_sample_stub(struct perf_tool *tool __maybe_unused,
338 				     union perf_event *event __maybe_unused,
339 				     struct perf_sample *sample __maybe_unused,
340 				     struct evsel *evsel __maybe_unused,
341 				     struct machine *machine __maybe_unused)
342 {
343 	dump_printf(": unhandled!\n");
344 	return 0;
345 }
346 
347 static int process_event_stub(struct perf_tool *tool __maybe_unused,
348 			      union perf_event *event __maybe_unused,
349 			      struct perf_sample *sample __maybe_unused,
350 			      struct machine *machine __maybe_unused)
351 {
352 	dump_printf(": unhandled!\n");
353 	return 0;
354 }
355 
356 static int process_finished_round_stub(struct perf_tool *tool __maybe_unused,
357 				       union perf_event *event __maybe_unused,
358 				       struct ordered_events *oe __maybe_unused)
359 {
360 	dump_printf(": unhandled!\n");
361 	return 0;
362 }
363 
364 static int process_finished_round(struct perf_tool *tool,
365 				  union perf_event *event,
366 				  struct ordered_events *oe);
367 
368 static int skipn(int fd, off_t n)
369 {
370 	char buf[4096];
371 	ssize_t ret;
372 
373 	while (n > 0) {
374 		ret = read(fd, buf, min(n, (off_t)sizeof(buf)));
375 		if (ret <= 0)
376 			return ret;
377 		n -= ret;
378 	}
379 
380 	return 0;
381 }
382 
383 static s64 process_event_auxtrace_stub(struct perf_session *session __maybe_unused,
384 				       union perf_event *event)
385 {
386 	dump_printf(": unhandled!\n");
387 	if (perf_data__is_pipe(session->data))
388 		skipn(perf_data__fd(session->data), event->auxtrace.size);
389 	return event->auxtrace.size;
390 }
391 
392 static int process_event_op2_stub(struct perf_session *session __maybe_unused,
393 				  union perf_event *event __maybe_unused)
394 {
395 	dump_printf(": unhandled!\n");
396 	return 0;
397 }
398 
399 
400 static
401 int process_event_thread_map_stub(struct perf_session *session __maybe_unused,
402 				  union perf_event *event __maybe_unused)
403 {
404 	if (dump_trace)
405 		perf_event__fprintf_thread_map(event, stdout);
406 
407 	dump_printf(": unhandled!\n");
408 	return 0;
409 }
410 
411 static
412 int process_event_cpu_map_stub(struct perf_session *session __maybe_unused,
413 			       union perf_event *event __maybe_unused)
414 {
415 	if (dump_trace)
416 		perf_event__fprintf_cpu_map(event, stdout);
417 
418 	dump_printf(": unhandled!\n");
419 	return 0;
420 }
421 
422 static
423 int process_event_stat_config_stub(struct perf_session *session __maybe_unused,
424 				   union perf_event *event __maybe_unused)
425 {
426 	if (dump_trace)
427 		perf_event__fprintf_stat_config(event, stdout);
428 
429 	dump_printf(": unhandled!\n");
430 	return 0;
431 }
432 
433 static int process_stat_stub(struct perf_session *perf_session __maybe_unused,
434 			     union perf_event *event)
435 {
436 	if (dump_trace)
437 		perf_event__fprintf_stat(event, stdout);
438 
439 	dump_printf(": unhandled!\n");
440 	return 0;
441 }
442 
443 static int process_stat_round_stub(struct perf_session *perf_session __maybe_unused,
444 				   union perf_event *event)
445 {
446 	if (dump_trace)
447 		perf_event__fprintf_stat_round(event, stdout);
448 
449 	dump_printf(": unhandled!\n");
450 	return 0;
451 }
452 
453 static int perf_session__process_compressed_event_stub(struct perf_session *session __maybe_unused,
454 						       union perf_event *event __maybe_unused,
455 						       u64 file_offset __maybe_unused)
456 {
457        dump_printf(": unhandled!\n");
458        return 0;
459 }
460 
461 void perf_tool__fill_defaults(struct perf_tool *tool)
462 {
463 	if (tool->sample == NULL)
464 		tool->sample = process_event_sample_stub;
465 	if (tool->mmap == NULL)
466 		tool->mmap = process_event_stub;
467 	if (tool->mmap2 == NULL)
468 		tool->mmap2 = process_event_stub;
469 	if (tool->comm == NULL)
470 		tool->comm = process_event_stub;
471 	if (tool->namespaces == NULL)
472 		tool->namespaces = process_event_stub;
473 	if (tool->cgroup == NULL)
474 		tool->cgroup = process_event_stub;
475 	if (tool->fork == NULL)
476 		tool->fork = process_event_stub;
477 	if (tool->exit == NULL)
478 		tool->exit = process_event_stub;
479 	if (tool->lost == NULL)
480 		tool->lost = perf_event__process_lost;
481 	if (tool->lost_samples == NULL)
482 		tool->lost_samples = perf_event__process_lost_samples;
483 	if (tool->aux == NULL)
484 		tool->aux = perf_event__process_aux;
485 	if (tool->itrace_start == NULL)
486 		tool->itrace_start = perf_event__process_itrace_start;
487 	if (tool->context_switch == NULL)
488 		tool->context_switch = perf_event__process_switch;
489 	if (tool->ksymbol == NULL)
490 		tool->ksymbol = perf_event__process_ksymbol;
491 	if (tool->bpf == NULL)
492 		tool->bpf = perf_event__process_bpf;
493 	if (tool->text_poke == NULL)
494 		tool->text_poke = perf_event__process_text_poke;
495 	if (tool->read == NULL)
496 		tool->read = process_event_sample_stub;
497 	if (tool->throttle == NULL)
498 		tool->throttle = process_event_stub;
499 	if (tool->unthrottle == NULL)
500 		tool->unthrottle = process_event_stub;
501 	if (tool->attr == NULL)
502 		tool->attr = process_event_synth_attr_stub;
503 	if (tool->event_update == NULL)
504 		tool->event_update = process_event_synth_event_update_stub;
505 	if (tool->tracing_data == NULL)
506 		tool->tracing_data = process_event_synth_tracing_data_stub;
507 	if (tool->build_id == NULL)
508 		tool->build_id = process_event_op2_stub;
509 	if (tool->finished_round == NULL) {
510 		if (tool->ordered_events)
511 			tool->finished_round = process_finished_round;
512 		else
513 			tool->finished_round = process_finished_round_stub;
514 	}
515 	if (tool->id_index == NULL)
516 		tool->id_index = process_event_op2_stub;
517 	if (tool->auxtrace_info == NULL)
518 		tool->auxtrace_info = process_event_op2_stub;
519 	if (tool->auxtrace == NULL)
520 		tool->auxtrace = process_event_auxtrace_stub;
521 	if (tool->auxtrace_error == NULL)
522 		tool->auxtrace_error = process_event_op2_stub;
523 	if (tool->thread_map == NULL)
524 		tool->thread_map = process_event_thread_map_stub;
525 	if (tool->cpu_map == NULL)
526 		tool->cpu_map = process_event_cpu_map_stub;
527 	if (tool->stat_config == NULL)
528 		tool->stat_config = process_event_stat_config_stub;
529 	if (tool->stat == NULL)
530 		tool->stat = process_stat_stub;
531 	if (tool->stat_round == NULL)
532 		tool->stat_round = process_stat_round_stub;
533 	if (tool->time_conv == NULL)
534 		tool->time_conv = process_event_op2_stub;
535 	if (tool->feature == NULL)
536 		tool->feature = process_event_op2_stub;
537 	if (tool->compressed == NULL)
538 		tool->compressed = perf_session__process_compressed_event;
539 }
540 
541 static void swap_sample_id_all(union perf_event *event, void *data)
542 {
543 	void *end = (void *) event + event->header.size;
544 	int size = end - data;
545 
546 	BUG_ON(size % sizeof(u64));
547 	mem_bswap_64(data, size);
548 }
549 
550 static void perf_event__all64_swap(union perf_event *event,
551 				   bool sample_id_all __maybe_unused)
552 {
553 	struct perf_event_header *hdr = &event->header;
554 	mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr));
555 }
556 
557 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all)
558 {
559 	event->comm.pid = bswap_32(event->comm.pid);
560 	event->comm.tid = bswap_32(event->comm.tid);
561 
562 	if (sample_id_all) {
563 		void *data = &event->comm.comm;
564 
565 		data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
566 		swap_sample_id_all(event, data);
567 	}
568 }
569 
570 static void perf_event__mmap_swap(union perf_event *event,
571 				  bool sample_id_all)
572 {
573 	event->mmap.pid	  = bswap_32(event->mmap.pid);
574 	event->mmap.tid	  = bswap_32(event->mmap.tid);
575 	event->mmap.start = bswap_64(event->mmap.start);
576 	event->mmap.len	  = bswap_64(event->mmap.len);
577 	event->mmap.pgoff = bswap_64(event->mmap.pgoff);
578 
579 	if (sample_id_all) {
580 		void *data = &event->mmap.filename;
581 
582 		data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
583 		swap_sample_id_all(event, data);
584 	}
585 }
586 
587 static void perf_event__mmap2_swap(union perf_event *event,
588 				  bool sample_id_all)
589 {
590 	event->mmap2.pid   = bswap_32(event->mmap2.pid);
591 	event->mmap2.tid   = bswap_32(event->mmap2.tid);
592 	event->mmap2.start = bswap_64(event->mmap2.start);
593 	event->mmap2.len   = bswap_64(event->mmap2.len);
594 	event->mmap2.pgoff = bswap_64(event->mmap2.pgoff);
595 	event->mmap2.maj   = bswap_32(event->mmap2.maj);
596 	event->mmap2.min   = bswap_32(event->mmap2.min);
597 	event->mmap2.ino   = bswap_64(event->mmap2.ino);
598 
599 	if (sample_id_all) {
600 		void *data = &event->mmap2.filename;
601 
602 		data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
603 		swap_sample_id_all(event, data);
604 	}
605 }
606 static void perf_event__task_swap(union perf_event *event, bool sample_id_all)
607 {
608 	event->fork.pid	 = bswap_32(event->fork.pid);
609 	event->fork.tid	 = bswap_32(event->fork.tid);
610 	event->fork.ppid = bswap_32(event->fork.ppid);
611 	event->fork.ptid = bswap_32(event->fork.ptid);
612 	event->fork.time = bswap_64(event->fork.time);
613 
614 	if (sample_id_all)
615 		swap_sample_id_all(event, &event->fork + 1);
616 }
617 
618 static void perf_event__read_swap(union perf_event *event, bool sample_id_all)
619 {
620 	event->read.pid		 = bswap_32(event->read.pid);
621 	event->read.tid		 = bswap_32(event->read.tid);
622 	event->read.value	 = bswap_64(event->read.value);
623 	event->read.time_enabled = bswap_64(event->read.time_enabled);
624 	event->read.time_running = bswap_64(event->read.time_running);
625 	event->read.id		 = bswap_64(event->read.id);
626 
627 	if (sample_id_all)
628 		swap_sample_id_all(event, &event->read + 1);
629 }
630 
631 static void perf_event__aux_swap(union perf_event *event, bool sample_id_all)
632 {
633 	event->aux.aux_offset = bswap_64(event->aux.aux_offset);
634 	event->aux.aux_size   = bswap_64(event->aux.aux_size);
635 	event->aux.flags      = bswap_64(event->aux.flags);
636 
637 	if (sample_id_all)
638 		swap_sample_id_all(event, &event->aux + 1);
639 }
640 
641 static void perf_event__itrace_start_swap(union perf_event *event,
642 					  bool sample_id_all)
643 {
644 	event->itrace_start.pid	 = bswap_32(event->itrace_start.pid);
645 	event->itrace_start.tid	 = bswap_32(event->itrace_start.tid);
646 
647 	if (sample_id_all)
648 		swap_sample_id_all(event, &event->itrace_start + 1);
649 }
650 
651 static void perf_event__switch_swap(union perf_event *event, bool sample_id_all)
652 {
653 	if (event->header.type == PERF_RECORD_SWITCH_CPU_WIDE) {
654 		event->context_switch.next_prev_pid =
655 				bswap_32(event->context_switch.next_prev_pid);
656 		event->context_switch.next_prev_tid =
657 				bswap_32(event->context_switch.next_prev_tid);
658 	}
659 
660 	if (sample_id_all)
661 		swap_sample_id_all(event, &event->context_switch + 1);
662 }
663 
664 static void perf_event__text_poke_swap(union perf_event *event, bool sample_id_all)
665 {
666 	event->text_poke.addr    = bswap_64(event->text_poke.addr);
667 	event->text_poke.old_len = bswap_16(event->text_poke.old_len);
668 	event->text_poke.new_len = bswap_16(event->text_poke.new_len);
669 
670 	if (sample_id_all) {
671 		size_t len = sizeof(event->text_poke.old_len) +
672 			     sizeof(event->text_poke.new_len) +
673 			     event->text_poke.old_len +
674 			     event->text_poke.new_len;
675 		void *data = &event->text_poke.old_len;
676 
677 		data += PERF_ALIGN(len, sizeof(u64));
678 		swap_sample_id_all(event, data);
679 	}
680 }
681 
682 static void perf_event__throttle_swap(union perf_event *event,
683 				      bool sample_id_all)
684 {
685 	event->throttle.time	  = bswap_64(event->throttle.time);
686 	event->throttle.id	  = bswap_64(event->throttle.id);
687 	event->throttle.stream_id = bswap_64(event->throttle.stream_id);
688 
689 	if (sample_id_all)
690 		swap_sample_id_all(event, &event->throttle + 1);
691 }
692 
693 static void perf_event__namespaces_swap(union perf_event *event,
694 					bool sample_id_all)
695 {
696 	u64 i;
697 
698 	event->namespaces.pid		= bswap_32(event->namespaces.pid);
699 	event->namespaces.tid		= bswap_32(event->namespaces.tid);
700 	event->namespaces.nr_namespaces	= bswap_64(event->namespaces.nr_namespaces);
701 
702 	for (i = 0; i < event->namespaces.nr_namespaces; i++) {
703 		struct perf_ns_link_info *ns = &event->namespaces.link_info[i];
704 
705 		ns->dev = bswap_64(ns->dev);
706 		ns->ino = bswap_64(ns->ino);
707 	}
708 
709 	if (sample_id_all)
710 		swap_sample_id_all(event, &event->namespaces.link_info[i]);
711 }
712 
713 static u8 revbyte(u8 b)
714 {
715 	int rev = (b >> 4) | ((b & 0xf) << 4);
716 	rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2);
717 	rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1);
718 	return (u8) rev;
719 }
720 
721 /*
722  * XXX this is hack in attempt to carry flags bitfield
723  * through endian village. ABI says:
724  *
725  * Bit-fields are allocated from right to left (least to most significant)
726  * on little-endian implementations and from left to right (most to least
727  * significant) on big-endian implementations.
728  *
729  * The above seems to be byte specific, so we need to reverse each
730  * byte of the bitfield. 'Internet' also says this might be implementation
731  * specific and we probably need proper fix and carry perf_event_attr
732  * bitfield flags in separate data file FEAT_ section. Thought this seems
733  * to work for now.
734  */
735 static void swap_bitfield(u8 *p, unsigned len)
736 {
737 	unsigned i;
738 
739 	for (i = 0; i < len; i++) {
740 		*p = revbyte(*p);
741 		p++;
742 	}
743 }
744 
745 /* exported for swapping attributes in file header */
746 void perf_event__attr_swap(struct perf_event_attr *attr)
747 {
748 	attr->type		= bswap_32(attr->type);
749 	attr->size		= bswap_32(attr->size);
750 
751 #define bswap_safe(f, n) 					\
752 	(attr->size > (offsetof(struct perf_event_attr, f) + 	\
753 		       sizeof(attr->f) * (n)))
754 #define bswap_field(f, sz) 			\
755 do { 						\
756 	if (bswap_safe(f, 0))			\
757 		attr->f = bswap_##sz(attr->f);	\
758 } while(0)
759 #define bswap_field_16(f) bswap_field(f, 16)
760 #define bswap_field_32(f) bswap_field(f, 32)
761 #define bswap_field_64(f) bswap_field(f, 64)
762 
763 	bswap_field_64(config);
764 	bswap_field_64(sample_period);
765 	bswap_field_64(sample_type);
766 	bswap_field_64(read_format);
767 	bswap_field_32(wakeup_events);
768 	bswap_field_32(bp_type);
769 	bswap_field_64(bp_addr);
770 	bswap_field_64(bp_len);
771 	bswap_field_64(branch_sample_type);
772 	bswap_field_64(sample_regs_user);
773 	bswap_field_32(sample_stack_user);
774 	bswap_field_32(aux_watermark);
775 	bswap_field_16(sample_max_stack);
776 	bswap_field_32(aux_sample_size);
777 
778 	/*
779 	 * After read_format are bitfields. Check read_format because
780 	 * we are unable to use offsetof on bitfield.
781 	 */
782 	if (bswap_safe(read_format, 1))
783 		swap_bitfield((u8 *) (&attr->read_format + 1),
784 			      sizeof(u64));
785 #undef bswap_field_64
786 #undef bswap_field_32
787 #undef bswap_field
788 #undef bswap_safe
789 }
790 
791 static void perf_event__hdr_attr_swap(union perf_event *event,
792 				      bool sample_id_all __maybe_unused)
793 {
794 	size_t size;
795 
796 	perf_event__attr_swap(&event->attr.attr);
797 
798 	size = event->header.size;
799 	size -= (void *)&event->attr.id - (void *)event;
800 	mem_bswap_64(event->attr.id, size);
801 }
802 
803 static void perf_event__event_update_swap(union perf_event *event,
804 					  bool sample_id_all __maybe_unused)
805 {
806 	event->event_update.type = bswap_64(event->event_update.type);
807 	event->event_update.id   = bswap_64(event->event_update.id);
808 }
809 
810 static void perf_event__event_type_swap(union perf_event *event,
811 					bool sample_id_all __maybe_unused)
812 {
813 	event->event_type.event_type.event_id =
814 		bswap_64(event->event_type.event_type.event_id);
815 }
816 
817 static void perf_event__tracing_data_swap(union perf_event *event,
818 					  bool sample_id_all __maybe_unused)
819 {
820 	event->tracing_data.size = bswap_32(event->tracing_data.size);
821 }
822 
823 static void perf_event__auxtrace_info_swap(union perf_event *event,
824 					   bool sample_id_all __maybe_unused)
825 {
826 	size_t size;
827 
828 	event->auxtrace_info.type = bswap_32(event->auxtrace_info.type);
829 
830 	size = event->header.size;
831 	size -= (void *)&event->auxtrace_info.priv - (void *)event;
832 	mem_bswap_64(event->auxtrace_info.priv, size);
833 }
834 
835 static void perf_event__auxtrace_swap(union perf_event *event,
836 				      bool sample_id_all __maybe_unused)
837 {
838 	event->auxtrace.size      = bswap_64(event->auxtrace.size);
839 	event->auxtrace.offset    = bswap_64(event->auxtrace.offset);
840 	event->auxtrace.reference = bswap_64(event->auxtrace.reference);
841 	event->auxtrace.idx       = bswap_32(event->auxtrace.idx);
842 	event->auxtrace.tid       = bswap_32(event->auxtrace.tid);
843 	event->auxtrace.cpu       = bswap_32(event->auxtrace.cpu);
844 }
845 
846 static void perf_event__auxtrace_error_swap(union perf_event *event,
847 					    bool sample_id_all __maybe_unused)
848 {
849 	event->auxtrace_error.type = bswap_32(event->auxtrace_error.type);
850 	event->auxtrace_error.code = bswap_32(event->auxtrace_error.code);
851 	event->auxtrace_error.cpu  = bswap_32(event->auxtrace_error.cpu);
852 	event->auxtrace_error.pid  = bswap_32(event->auxtrace_error.pid);
853 	event->auxtrace_error.tid  = bswap_32(event->auxtrace_error.tid);
854 	event->auxtrace_error.fmt  = bswap_32(event->auxtrace_error.fmt);
855 	event->auxtrace_error.ip   = bswap_64(event->auxtrace_error.ip);
856 	if (event->auxtrace_error.fmt)
857 		event->auxtrace_error.time = bswap_64(event->auxtrace_error.time);
858 }
859 
860 static void perf_event__thread_map_swap(union perf_event *event,
861 					bool sample_id_all __maybe_unused)
862 {
863 	unsigned i;
864 
865 	event->thread_map.nr = bswap_64(event->thread_map.nr);
866 
867 	for (i = 0; i < event->thread_map.nr; i++)
868 		event->thread_map.entries[i].pid = bswap_64(event->thread_map.entries[i].pid);
869 }
870 
871 static void perf_event__cpu_map_swap(union perf_event *event,
872 				     bool sample_id_all __maybe_unused)
873 {
874 	struct perf_record_cpu_map_data *data = &event->cpu_map.data;
875 	struct cpu_map_entries *cpus;
876 	struct perf_record_record_cpu_map *mask;
877 	unsigned i;
878 
879 	data->type = bswap_64(data->type);
880 
881 	switch (data->type) {
882 	case PERF_CPU_MAP__CPUS:
883 		cpus = (struct cpu_map_entries *)data->data;
884 
885 		cpus->nr = bswap_16(cpus->nr);
886 
887 		for (i = 0; i < cpus->nr; i++)
888 			cpus->cpu[i] = bswap_16(cpus->cpu[i]);
889 		break;
890 	case PERF_CPU_MAP__MASK:
891 		mask = (struct perf_record_record_cpu_map *)data->data;
892 
893 		mask->nr = bswap_16(mask->nr);
894 		mask->long_size = bswap_16(mask->long_size);
895 
896 		switch (mask->long_size) {
897 		case 4: mem_bswap_32(&mask->mask, mask->nr); break;
898 		case 8: mem_bswap_64(&mask->mask, mask->nr); break;
899 		default:
900 			pr_err("cpu_map swap: unsupported long size\n");
901 		}
902 	default:
903 		break;
904 	}
905 }
906 
907 static void perf_event__stat_config_swap(union perf_event *event,
908 					 bool sample_id_all __maybe_unused)
909 {
910 	u64 size;
911 
912 	size  = event->stat_config.nr * sizeof(event->stat_config.data[0]);
913 	size += 1; /* nr item itself */
914 	mem_bswap_64(&event->stat_config.nr, size);
915 }
916 
917 static void perf_event__stat_swap(union perf_event *event,
918 				  bool sample_id_all __maybe_unused)
919 {
920 	event->stat.id     = bswap_64(event->stat.id);
921 	event->stat.thread = bswap_32(event->stat.thread);
922 	event->stat.cpu    = bswap_32(event->stat.cpu);
923 	event->stat.val    = bswap_64(event->stat.val);
924 	event->stat.ena    = bswap_64(event->stat.ena);
925 	event->stat.run    = bswap_64(event->stat.run);
926 }
927 
928 static void perf_event__stat_round_swap(union perf_event *event,
929 					bool sample_id_all __maybe_unused)
930 {
931 	event->stat_round.type = bswap_64(event->stat_round.type);
932 	event->stat_round.time = bswap_64(event->stat_round.time);
933 }
934 
935 typedef void (*perf_event__swap_op)(union perf_event *event,
936 				    bool sample_id_all);
937 
938 static perf_event__swap_op perf_event__swap_ops[] = {
939 	[PERF_RECORD_MMAP]		  = perf_event__mmap_swap,
940 	[PERF_RECORD_MMAP2]		  = perf_event__mmap2_swap,
941 	[PERF_RECORD_COMM]		  = perf_event__comm_swap,
942 	[PERF_RECORD_FORK]		  = perf_event__task_swap,
943 	[PERF_RECORD_EXIT]		  = perf_event__task_swap,
944 	[PERF_RECORD_LOST]		  = perf_event__all64_swap,
945 	[PERF_RECORD_READ]		  = perf_event__read_swap,
946 	[PERF_RECORD_THROTTLE]		  = perf_event__throttle_swap,
947 	[PERF_RECORD_UNTHROTTLE]	  = perf_event__throttle_swap,
948 	[PERF_RECORD_SAMPLE]		  = perf_event__all64_swap,
949 	[PERF_RECORD_AUX]		  = perf_event__aux_swap,
950 	[PERF_RECORD_ITRACE_START]	  = perf_event__itrace_start_swap,
951 	[PERF_RECORD_LOST_SAMPLES]	  = perf_event__all64_swap,
952 	[PERF_RECORD_SWITCH]		  = perf_event__switch_swap,
953 	[PERF_RECORD_SWITCH_CPU_WIDE]	  = perf_event__switch_swap,
954 	[PERF_RECORD_NAMESPACES]	  = perf_event__namespaces_swap,
955 	[PERF_RECORD_TEXT_POKE]		  = perf_event__text_poke_swap,
956 	[PERF_RECORD_HEADER_ATTR]	  = perf_event__hdr_attr_swap,
957 	[PERF_RECORD_HEADER_EVENT_TYPE]	  = perf_event__event_type_swap,
958 	[PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
959 	[PERF_RECORD_HEADER_BUILD_ID]	  = NULL,
960 	[PERF_RECORD_ID_INDEX]		  = perf_event__all64_swap,
961 	[PERF_RECORD_AUXTRACE_INFO]	  = perf_event__auxtrace_info_swap,
962 	[PERF_RECORD_AUXTRACE]		  = perf_event__auxtrace_swap,
963 	[PERF_RECORD_AUXTRACE_ERROR]	  = perf_event__auxtrace_error_swap,
964 	[PERF_RECORD_THREAD_MAP]	  = perf_event__thread_map_swap,
965 	[PERF_RECORD_CPU_MAP]		  = perf_event__cpu_map_swap,
966 	[PERF_RECORD_STAT_CONFIG]	  = perf_event__stat_config_swap,
967 	[PERF_RECORD_STAT]		  = perf_event__stat_swap,
968 	[PERF_RECORD_STAT_ROUND]	  = perf_event__stat_round_swap,
969 	[PERF_RECORD_EVENT_UPDATE]	  = perf_event__event_update_swap,
970 	[PERF_RECORD_TIME_CONV]		  = perf_event__all64_swap,
971 	[PERF_RECORD_HEADER_MAX]	  = NULL,
972 };
973 
974 /*
975  * When perf record finishes a pass on every buffers, it records this pseudo
976  * event.
977  * We record the max timestamp t found in the pass n.
978  * Assuming these timestamps are monotonic across cpus, we know that if
979  * a buffer still has events with timestamps below t, they will be all
980  * available and then read in the pass n + 1.
981  * Hence when we start to read the pass n + 2, we can safely flush every
982  * events with timestamps below t.
983  *
984  *    ============ PASS n =================
985  *       CPU 0         |   CPU 1
986  *                     |
987  *    cnt1 timestamps  |   cnt2 timestamps
988  *          1          |         2
989  *          2          |         3
990  *          -          |         4  <--- max recorded
991  *
992  *    ============ PASS n + 1 ==============
993  *       CPU 0         |   CPU 1
994  *                     |
995  *    cnt1 timestamps  |   cnt2 timestamps
996  *          3          |         5
997  *          4          |         6
998  *          5          |         7 <---- max recorded
999  *
1000  *      Flush every events below timestamp 4
1001  *
1002  *    ============ PASS n + 2 ==============
1003  *       CPU 0         |   CPU 1
1004  *                     |
1005  *    cnt1 timestamps  |   cnt2 timestamps
1006  *          6          |         8
1007  *          7          |         9
1008  *          -          |         10
1009  *
1010  *      Flush every events below timestamp 7
1011  *      etc...
1012  */
1013 static int process_finished_round(struct perf_tool *tool __maybe_unused,
1014 				  union perf_event *event __maybe_unused,
1015 				  struct ordered_events *oe)
1016 {
1017 	if (dump_trace)
1018 		fprintf(stdout, "\n");
1019 	return ordered_events__flush(oe, OE_FLUSH__ROUND);
1020 }
1021 
1022 int perf_session__queue_event(struct perf_session *s, union perf_event *event,
1023 			      u64 timestamp, u64 file_offset)
1024 {
1025 	return ordered_events__queue(&s->ordered_events, event, timestamp, file_offset);
1026 }
1027 
1028 static void callchain__lbr_callstack_printf(struct perf_sample *sample)
1029 {
1030 	struct ip_callchain *callchain = sample->callchain;
1031 	struct branch_stack *lbr_stack = sample->branch_stack;
1032 	struct branch_entry *entries = perf_sample__branch_entries(sample);
1033 	u64 kernel_callchain_nr = callchain->nr;
1034 	unsigned int i;
1035 
1036 	for (i = 0; i < kernel_callchain_nr; i++) {
1037 		if (callchain->ips[i] == PERF_CONTEXT_USER)
1038 			break;
1039 	}
1040 
1041 	if ((i != kernel_callchain_nr) && lbr_stack->nr) {
1042 		u64 total_nr;
1043 		/*
1044 		 * LBR callstack can only get user call chain,
1045 		 * i is kernel call chain number,
1046 		 * 1 is PERF_CONTEXT_USER.
1047 		 *
1048 		 * The user call chain is stored in LBR registers.
1049 		 * LBR are pair registers. The caller is stored
1050 		 * in "from" register, while the callee is stored
1051 		 * in "to" register.
1052 		 * For example, there is a call stack
1053 		 * "A"->"B"->"C"->"D".
1054 		 * The LBR registers will recorde like
1055 		 * "C"->"D", "B"->"C", "A"->"B".
1056 		 * So only the first "to" register and all "from"
1057 		 * registers are needed to construct the whole stack.
1058 		 */
1059 		total_nr = i + 1 + lbr_stack->nr + 1;
1060 		kernel_callchain_nr = i + 1;
1061 
1062 		printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr);
1063 
1064 		for (i = 0; i < kernel_callchain_nr; i++)
1065 			printf("..... %2d: %016" PRIx64 "\n",
1066 			       i, callchain->ips[i]);
1067 
1068 		printf("..... %2d: %016" PRIx64 "\n",
1069 		       (int)(kernel_callchain_nr), entries[0].to);
1070 		for (i = 0; i < lbr_stack->nr; i++)
1071 			printf("..... %2d: %016" PRIx64 "\n",
1072 			       (int)(i + kernel_callchain_nr + 1), entries[i].from);
1073 	}
1074 }
1075 
1076 static void callchain__printf(struct evsel *evsel,
1077 			      struct perf_sample *sample)
1078 {
1079 	unsigned int i;
1080 	struct ip_callchain *callchain = sample->callchain;
1081 
1082 	if (evsel__has_branch_callstack(evsel))
1083 		callchain__lbr_callstack_printf(sample);
1084 
1085 	printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr);
1086 
1087 	for (i = 0; i < callchain->nr; i++)
1088 		printf("..... %2d: %016" PRIx64 "\n",
1089 		       i, callchain->ips[i]);
1090 }
1091 
1092 static void branch_stack__printf(struct perf_sample *sample, bool callstack)
1093 {
1094 	struct branch_entry *entries = perf_sample__branch_entries(sample);
1095 	uint64_t i;
1096 
1097 	printf("%s: nr:%" PRIu64 "\n",
1098 		!callstack ? "... branch stack" : "... branch callstack",
1099 		sample->branch_stack->nr);
1100 
1101 	for (i = 0; i < sample->branch_stack->nr; i++) {
1102 		struct branch_entry *e = &entries[i];
1103 
1104 		if (!callstack) {
1105 			printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x\n",
1106 				i, e->from, e->to,
1107 				(unsigned short)e->flags.cycles,
1108 				e->flags.mispred ? "M" : " ",
1109 				e->flags.predicted ? "P" : " ",
1110 				e->flags.abort ? "A" : " ",
1111 				e->flags.in_tx ? "T" : " ",
1112 				(unsigned)e->flags.reserved);
1113 		} else {
1114 			printf("..... %2"PRIu64": %016" PRIx64 "\n",
1115 				i, i > 0 ? e->from : e->to);
1116 		}
1117 	}
1118 }
1119 
1120 static void regs_dump__printf(u64 mask, u64 *regs)
1121 {
1122 	unsigned rid, i = 0;
1123 
1124 	for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) {
1125 		u64 val = regs[i++];
1126 
1127 		printf(".... %-5s 0x%016" PRIx64 "\n",
1128 		       perf_reg_name(rid), val);
1129 	}
1130 }
1131 
1132 static const char *regs_abi[] = {
1133 	[PERF_SAMPLE_REGS_ABI_NONE] = "none",
1134 	[PERF_SAMPLE_REGS_ABI_32] = "32-bit",
1135 	[PERF_SAMPLE_REGS_ABI_64] = "64-bit",
1136 };
1137 
1138 static inline const char *regs_dump_abi(struct regs_dump *d)
1139 {
1140 	if (d->abi > PERF_SAMPLE_REGS_ABI_64)
1141 		return "unknown";
1142 
1143 	return regs_abi[d->abi];
1144 }
1145 
1146 static void regs__printf(const char *type, struct regs_dump *regs)
1147 {
1148 	u64 mask = regs->mask;
1149 
1150 	printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n",
1151 	       type,
1152 	       mask,
1153 	       regs_dump_abi(regs));
1154 
1155 	regs_dump__printf(mask, regs->regs);
1156 }
1157 
1158 static void regs_user__printf(struct perf_sample *sample)
1159 {
1160 	struct regs_dump *user_regs = &sample->user_regs;
1161 
1162 	if (user_regs->regs)
1163 		regs__printf("user", user_regs);
1164 }
1165 
1166 static void regs_intr__printf(struct perf_sample *sample)
1167 {
1168 	struct regs_dump *intr_regs = &sample->intr_regs;
1169 
1170 	if (intr_regs->regs)
1171 		regs__printf("intr", intr_regs);
1172 }
1173 
1174 static void stack_user__printf(struct stack_dump *dump)
1175 {
1176 	printf("... ustack: size %" PRIu64 ", offset 0x%x\n",
1177 	       dump->size, dump->offset);
1178 }
1179 
1180 static void perf_evlist__print_tstamp(struct evlist *evlist,
1181 				       union perf_event *event,
1182 				       struct perf_sample *sample)
1183 {
1184 	u64 sample_type = __evlist__combined_sample_type(evlist);
1185 
1186 	if (event->header.type != PERF_RECORD_SAMPLE &&
1187 	    !evlist__sample_id_all(evlist)) {
1188 		fputs("-1 -1 ", stdout);
1189 		return;
1190 	}
1191 
1192 	if ((sample_type & PERF_SAMPLE_CPU))
1193 		printf("%u ", sample->cpu);
1194 
1195 	if (sample_type & PERF_SAMPLE_TIME)
1196 		printf("%" PRIu64 " ", sample->time);
1197 }
1198 
1199 static void sample_read__printf(struct perf_sample *sample, u64 read_format)
1200 {
1201 	printf("... sample_read:\n");
1202 
1203 	if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1204 		printf("...... time enabled %016" PRIx64 "\n",
1205 		       sample->read.time_enabled);
1206 
1207 	if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1208 		printf("...... time running %016" PRIx64 "\n",
1209 		       sample->read.time_running);
1210 
1211 	if (read_format & PERF_FORMAT_GROUP) {
1212 		u64 i;
1213 
1214 		printf(".... group nr %" PRIu64 "\n", sample->read.group.nr);
1215 
1216 		for (i = 0; i < sample->read.group.nr; i++) {
1217 			struct sample_read_value *value;
1218 
1219 			value = &sample->read.group.values[i];
1220 			printf("..... id %016" PRIx64
1221 			       ", value %016" PRIx64 "\n",
1222 			       value->id, value->value);
1223 		}
1224 	} else
1225 		printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n",
1226 			sample->read.one.id, sample->read.one.value);
1227 }
1228 
1229 static void dump_event(struct evlist *evlist, union perf_event *event,
1230 		       u64 file_offset, struct perf_sample *sample)
1231 {
1232 	if (!dump_trace)
1233 		return;
1234 
1235 	printf("\n%#" PRIx64 " [%#x]: event: %d\n",
1236 	       file_offset, event->header.size, event->header.type);
1237 
1238 	trace_event(event);
1239 	if (event->header.type == PERF_RECORD_SAMPLE && evlist->trace_event_sample_raw)
1240 		evlist->trace_event_sample_raw(evlist, event, sample);
1241 
1242 	if (sample)
1243 		perf_evlist__print_tstamp(evlist, event, sample);
1244 
1245 	printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset,
1246 	       event->header.size, perf_event__name(event->header.type));
1247 }
1248 
1249 static void dump_sample(struct evsel *evsel, union perf_event *event,
1250 			struct perf_sample *sample)
1251 {
1252 	u64 sample_type;
1253 
1254 	if (!dump_trace)
1255 		return;
1256 
1257 	printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n",
1258 	       event->header.misc, sample->pid, sample->tid, sample->ip,
1259 	       sample->period, sample->addr);
1260 
1261 	sample_type = evsel->core.attr.sample_type;
1262 
1263 	if (evsel__has_callchain(evsel))
1264 		callchain__printf(evsel, sample);
1265 
1266 	if (evsel__has_br_stack(evsel))
1267 		branch_stack__printf(sample, evsel__has_branch_callstack(evsel));
1268 
1269 	if (sample_type & PERF_SAMPLE_REGS_USER)
1270 		regs_user__printf(sample);
1271 
1272 	if (sample_type & PERF_SAMPLE_REGS_INTR)
1273 		regs_intr__printf(sample);
1274 
1275 	if (sample_type & PERF_SAMPLE_STACK_USER)
1276 		stack_user__printf(&sample->user_stack);
1277 
1278 	if (sample_type & PERF_SAMPLE_WEIGHT)
1279 		printf("... weight: %" PRIu64 "\n", sample->weight);
1280 
1281 	if (sample_type & PERF_SAMPLE_DATA_SRC)
1282 		printf(" . data_src: 0x%"PRIx64"\n", sample->data_src);
1283 
1284 	if (sample_type & PERF_SAMPLE_PHYS_ADDR)
1285 		printf(" .. phys_addr: 0x%"PRIx64"\n", sample->phys_addr);
1286 
1287 	if (sample_type & PERF_SAMPLE_TRANSACTION)
1288 		printf("... transaction: %" PRIx64 "\n", sample->transaction);
1289 
1290 	if (sample_type & PERF_SAMPLE_READ)
1291 		sample_read__printf(sample, evsel->core.attr.read_format);
1292 }
1293 
1294 static void dump_read(struct evsel *evsel, union perf_event *event)
1295 {
1296 	struct perf_record_read *read_event = &event->read;
1297 	u64 read_format;
1298 
1299 	if (!dump_trace)
1300 		return;
1301 
1302 	printf(": %d %d %s %" PRI_lu64 "\n", event->read.pid, event->read.tid,
1303 	       evsel__name(evsel), event->read.value);
1304 
1305 	if (!evsel)
1306 		return;
1307 
1308 	read_format = evsel->core.attr.read_format;
1309 
1310 	if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1311 		printf("... time enabled : %" PRI_lu64 "\n", read_event->time_enabled);
1312 
1313 	if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1314 		printf("... time running : %" PRI_lu64 "\n", read_event->time_running);
1315 
1316 	if (read_format & PERF_FORMAT_ID)
1317 		printf("... id           : %" PRI_lu64 "\n", read_event->id);
1318 }
1319 
1320 static struct machine *machines__find_for_cpumode(struct machines *machines,
1321 					       union perf_event *event,
1322 					       struct perf_sample *sample)
1323 {
1324 	struct machine *machine;
1325 
1326 	if (perf_guest &&
1327 	    ((sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL) ||
1328 	     (sample->cpumode == PERF_RECORD_MISC_GUEST_USER))) {
1329 		u32 pid;
1330 
1331 		if (event->header.type == PERF_RECORD_MMAP
1332 		    || event->header.type == PERF_RECORD_MMAP2)
1333 			pid = event->mmap.pid;
1334 		else
1335 			pid = sample->pid;
1336 
1337 		machine = machines__find(machines, pid);
1338 		if (!machine)
1339 			machine = machines__findnew(machines, DEFAULT_GUEST_KERNEL_ID);
1340 		return machine;
1341 	}
1342 
1343 	return &machines->host;
1344 }
1345 
1346 static int deliver_sample_value(struct evlist *evlist,
1347 				struct perf_tool *tool,
1348 				union perf_event *event,
1349 				struct perf_sample *sample,
1350 				struct sample_read_value *v,
1351 				struct machine *machine)
1352 {
1353 	struct perf_sample_id *sid = perf_evlist__id2sid(evlist, v->id);
1354 	struct evsel *evsel;
1355 
1356 	if (sid) {
1357 		sample->id     = v->id;
1358 		sample->period = v->value - sid->period;
1359 		sid->period    = v->value;
1360 	}
1361 
1362 	if (!sid || sid->evsel == NULL) {
1363 		++evlist->stats.nr_unknown_id;
1364 		return 0;
1365 	}
1366 
1367 	/*
1368 	 * There's no reason to deliver sample
1369 	 * for zero period, bail out.
1370 	 */
1371 	if (!sample->period)
1372 		return 0;
1373 
1374 	evsel = container_of(sid->evsel, struct evsel, core);
1375 	return tool->sample(tool, event, sample, evsel, machine);
1376 }
1377 
1378 static int deliver_sample_group(struct evlist *evlist,
1379 				struct perf_tool *tool,
1380 				union  perf_event *event,
1381 				struct perf_sample *sample,
1382 				struct machine *machine)
1383 {
1384 	int ret = -EINVAL;
1385 	u64 i;
1386 
1387 	for (i = 0; i < sample->read.group.nr; i++) {
1388 		ret = deliver_sample_value(evlist, tool, event, sample,
1389 					   &sample->read.group.values[i],
1390 					   machine);
1391 		if (ret)
1392 			break;
1393 	}
1394 
1395 	return ret;
1396 }
1397 
1398 static int
1399  perf_evlist__deliver_sample(struct evlist *evlist,
1400 			     struct perf_tool *tool,
1401 			     union  perf_event *event,
1402 			     struct perf_sample *sample,
1403 			     struct evsel *evsel,
1404 			     struct machine *machine)
1405 {
1406 	/* We know evsel != NULL. */
1407 	u64 sample_type = evsel->core.attr.sample_type;
1408 	u64 read_format = evsel->core.attr.read_format;
1409 
1410 	/* Standard sample delivery. */
1411 	if (!(sample_type & PERF_SAMPLE_READ))
1412 		return tool->sample(tool, event, sample, evsel, machine);
1413 
1414 	/* For PERF_SAMPLE_READ we have either single or group mode. */
1415 	if (read_format & PERF_FORMAT_GROUP)
1416 		return deliver_sample_group(evlist, tool, event, sample,
1417 					    machine);
1418 	else
1419 		return deliver_sample_value(evlist, tool, event, sample,
1420 					    &sample->read.one, machine);
1421 }
1422 
1423 static int machines__deliver_event(struct machines *machines,
1424 				   struct evlist *evlist,
1425 				   union perf_event *event,
1426 				   struct perf_sample *sample,
1427 				   struct perf_tool *tool, u64 file_offset)
1428 {
1429 	struct evsel *evsel;
1430 	struct machine *machine;
1431 
1432 	dump_event(evlist, event, file_offset, sample);
1433 
1434 	evsel = perf_evlist__id2evsel(evlist, sample->id);
1435 
1436 	machine = machines__find_for_cpumode(machines, event, sample);
1437 
1438 	switch (event->header.type) {
1439 	case PERF_RECORD_SAMPLE:
1440 		if (evsel == NULL) {
1441 			++evlist->stats.nr_unknown_id;
1442 			return 0;
1443 		}
1444 		dump_sample(evsel, event, sample);
1445 		if (machine == NULL) {
1446 			++evlist->stats.nr_unprocessable_samples;
1447 			return 0;
1448 		}
1449 		return perf_evlist__deliver_sample(evlist, tool, event, sample, evsel, machine);
1450 	case PERF_RECORD_MMAP:
1451 		return tool->mmap(tool, event, sample, machine);
1452 	case PERF_RECORD_MMAP2:
1453 		if (event->header.misc & PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT)
1454 			++evlist->stats.nr_proc_map_timeout;
1455 		return tool->mmap2(tool, event, sample, machine);
1456 	case PERF_RECORD_COMM:
1457 		return tool->comm(tool, event, sample, machine);
1458 	case PERF_RECORD_NAMESPACES:
1459 		return tool->namespaces(tool, event, sample, machine);
1460 	case PERF_RECORD_CGROUP:
1461 		return tool->cgroup(tool, event, sample, machine);
1462 	case PERF_RECORD_FORK:
1463 		return tool->fork(tool, event, sample, machine);
1464 	case PERF_RECORD_EXIT:
1465 		return tool->exit(tool, event, sample, machine);
1466 	case PERF_RECORD_LOST:
1467 		if (tool->lost == perf_event__process_lost)
1468 			evlist->stats.total_lost += event->lost.lost;
1469 		return tool->lost(tool, event, sample, machine);
1470 	case PERF_RECORD_LOST_SAMPLES:
1471 		if (tool->lost_samples == perf_event__process_lost_samples)
1472 			evlist->stats.total_lost_samples += event->lost_samples.lost;
1473 		return tool->lost_samples(tool, event, sample, machine);
1474 	case PERF_RECORD_READ:
1475 		dump_read(evsel, event);
1476 		return tool->read(tool, event, sample, evsel, machine);
1477 	case PERF_RECORD_THROTTLE:
1478 		return tool->throttle(tool, event, sample, machine);
1479 	case PERF_RECORD_UNTHROTTLE:
1480 		return tool->unthrottle(tool, event, sample, machine);
1481 	case PERF_RECORD_AUX:
1482 		if (tool->aux == perf_event__process_aux) {
1483 			if (event->aux.flags & PERF_AUX_FLAG_TRUNCATED)
1484 				evlist->stats.total_aux_lost += 1;
1485 			if (event->aux.flags & PERF_AUX_FLAG_PARTIAL)
1486 				evlist->stats.total_aux_partial += 1;
1487 		}
1488 		return tool->aux(tool, event, sample, machine);
1489 	case PERF_RECORD_ITRACE_START:
1490 		return tool->itrace_start(tool, event, sample, machine);
1491 	case PERF_RECORD_SWITCH:
1492 	case PERF_RECORD_SWITCH_CPU_WIDE:
1493 		return tool->context_switch(tool, event, sample, machine);
1494 	case PERF_RECORD_KSYMBOL:
1495 		return tool->ksymbol(tool, event, sample, machine);
1496 	case PERF_RECORD_BPF_EVENT:
1497 		return tool->bpf(tool, event, sample, machine);
1498 	case PERF_RECORD_TEXT_POKE:
1499 		return tool->text_poke(tool, event, sample, machine);
1500 	default:
1501 		++evlist->stats.nr_unknown_events;
1502 		return -1;
1503 	}
1504 }
1505 
1506 static int perf_session__deliver_event(struct perf_session *session,
1507 				       union perf_event *event,
1508 				       struct perf_tool *tool,
1509 				       u64 file_offset)
1510 {
1511 	struct perf_sample sample;
1512 	int ret;
1513 
1514 	ret = perf_evlist__parse_sample(session->evlist, event, &sample);
1515 	if (ret) {
1516 		pr_err("Can't parse sample, err = %d\n", ret);
1517 		return ret;
1518 	}
1519 
1520 	ret = auxtrace__process_event(session, event, &sample, tool);
1521 	if (ret < 0)
1522 		return ret;
1523 	if (ret > 0)
1524 		return 0;
1525 
1526 	ret = machines__deliver_event(&session->machines, session->evlist,
1527 				      event, &sample, tool, file_offset);
1528 
1529 	if (dump_trace && sample.aux_sample.size)
1530 		auxtrace__dump_auxtrace_sample(session, &sample);
1531 
1532 	return ret;
1533 }
1534 
1535 static s64 perf_session__process_user_event(struct perf_session *session,
1536 					    union perf_event *event,
1537 					    u64 file_offset)
1538 {
1539 	struct ordered_events *oe = &session->ordered_events;
1540 	struct perf_tool *tool = session->tool;
1541 	struct perf_sample sample = { .time = 0, };
1542 	int fd = perf_data__fd(session->data);
1543 	int err;
1544 
1545 	if (event->header.type != PERF_RECORD_COMPRESSED ||
1546 	    tool->compressed == perf_session__process_compressed_event_stub)
1547 		dump_event(session->evlist, event, file_offset, &sample);
1548 
1549 	/* These events are processed right away */
1550 	switch (event->header.type) {
1551 	case PERF_RECORD_HEADER_ATTR:
1552 		err = tool->attr(tool, event, &session->evlist);
1553 		if (err == 0) {
1554 			perf_session__set_id_hdr_size(session);
1555 			perf_session__set_comm_exec(session);
1556 		}
1557 		return err;
1558 	case PERF_RECORD_EVENT_UPDATE:
1559 		return tool->event_update(tool, event, &session->evlist);
1560 	case PERF_RECORD_HEADER_EVENT_TYPE:
1561 		/*
1562 		 * Depreceated, but we need to handle it for sake
1563 		 * of old data files create in pipe mode.
1564 		 */
1565 		return 0;
1566 	case PERF_RECORD_HEADER_TRACING_DATA:
1567 		/*
1568 		 * Setup for reading amidst mmap, but only when we
1569 		 * are in 'file' mode. The 'pipe' fd is in proper
1570 		 * place already.
1571 		 */
1572 		if (!perf_data__is_pipe(session->data))
1573 			lseek(fd, file_offset, SEEK_SET);
1574 		return tool->tracing_data(session, event);
1575 	case PERF_RECORD_HEADER_BUILD_ID:
1576 		return tool->build_id(session, event);
1577 	case PERF_RECORD_FINISHED_ROUND:
1578 		return tool->finished_round(tool, event, oe);
1579 	case PERF_RECORD_ID_INDEX:
1580 		return tool->id_index(session, event);
1581 	case PERF_RECORD_AUXTRACE_INFO:
1582 		return tool->auxtrace_info(session, event);
1583 	case PERF_RECORD_AUXTRACE:
1584 		/* setup for reading amidst mmap */
1585 		lseek(fd, file_offset + event->header.size, SEEK_SET);
1586 		return tool->auxtrace(session, event);
1587 	case PERF_RECORD_AUXTRACE_ERROR:
1588 		perf_session__auxtrace_error_inc(session, event);
1589 		return tool->auxtrace_error(session, event);
1590 	case PERF_RECORD_THREAD_MAP:
1591 		return tool->thread_map(session, event);
1592 	case PERF_RECORD_CPU_MAP:
1593 		return tool->cpu_map(session, event);
1594 	case PERF_RECORD_STAT_CONFIG:
1595 		return tool->stat_config(session, event);
1596 	case PERF_RECORD_STAT:
1597 		return tool->stat(session, event);
1598 	case PERF_RECORD_STAT_ROUND:
1599 		return tool->stat_round(session, event);
1600 	case PERF_RECORD_TIME_CONV:
1601 		session->time_conv = event->time_conv;
1602 		return tool->time_conv(session, event);
1603 	case PERF_RECORD_HEADER_FEATURE:
1604 		return tool->feature(session, event);
1605 	case PERF_RECORD_COMPRESSED:
1606 		err = tool->compressed(session, event, file_offset);
1607 		if (err)
1608 			dump_event(session->evlist, event, file_offset, &sample);
1609 		return err;
1610 	default:
1611 		return -EINVAL;
1612 	}
1613 }
1614 
1615 int perf_session__deliver_synth_event(struct perf_session *session,
1616 				      union perf_event *event,
1617 				      struct perf_sample *sample)
1618 {
1619 	struct evlist *evlist = session->evlist;
1620 	struct perf_tool *tool = session->tool;
1621 
1622 	events_stats__inc(&evlist->stats, event->header.type);
1623 
1624 	if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1625 		return perf_session__process_user_event(session, event, 0);
1626 
1627 	return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0);
1628 }
1629 
1630 static void event_swap(union perf_event *event, bool sample_id_all)
1631 {
1632 	perf_event__swap_op swap;
1633 
1634 	swap = perf_event__swap_ops[event->header.type];
1635 	if (swap)
1636 		swap(event, sample_id_all);
1637 }
1638 
1639 int perf_session__peek_event(struct perf_session *session, off_t file_offset,
1640 			     void *buf, size_t buf_sz,
1641 			     union perf_event **event_ptr,
1642 			     struct perf_sample *sample)
1643 {
1644 	union perf_event *event;
1645 	size_t hdr_sz, rest;
1646 	int fd;
1647 
1648 	if (session->one_mmap && !session->header.needs_swap) {
1649 		event = file_offset - session->one_mmap_offset +
1650 			session->one_mmap_addr;
1651 		goto out_parse_sample;
1652 	}
1653 
1654 	if (perf_data__is_pipe(session->data))
1655 		return -1;
1656 
1657 	fd = perf_data__fd(session->data);
1658 	hdr_sz = sizeof(struct perf_event_header);
1659 
1660 	if (buf_sz < hdr_sz)
1661 		return -1;
1662 
1663 	if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 ||
1664 	    readn(fd, buf, hdr_sz) != (ssize_t)hdr_sz)
1665 		return -1;
1666 
1667 	event = (union perf_event *)buf;
1668 
1669 	if (session->header.needs_swap)
1670 		perf_event_header__bswap(&event->header);
1671 
1672 	if (event->header.size < hdr_sz || event->header.size > buf_sz)
1673 		return -1;
1674 
1675 	rest = event->header.size - hdr_sz;
1676 
1677 	if (readn(fd, buf, rest) != (ssize_t)rest)
1678 		return -1;
1679 
1680 	if (session->header.needs_swap)
1681 		event_swap(event, evlist__sample_id_all(session->evlist));
1682 
1683 out_parse_sample:
1684 
1685 	if (sample && event->header.type < PERF_RECORD_USER_TYPE_START &&
1686 	    perf_evlist__parse_sample(session->evlist, event, sample))
1687 		return -1;
1688 
1689 	*event_ptr = event;
1690 
1691 	return 0;
1692 }
1693 
1694 int perf_session__peek_events(struct perf_session *session, u64 offset,
1695 			      u64 size, peek_events_cb_t cb, void *data)
1696 {
1697 	u64 max_offset = offset + size;
1698 	char buf[PERF_SAMPLE_MAX_SIZE];
1699 	union perf_event *event;
1700 	int err;
1701 
1702 	do {
1703 		err = perf_session__peek_event(session, offset, buf,
1704 					       PERF_SAMPLE_MAX_SIZE, &event,
1705 					       NULL);
1706 		if (err)
1707 			return err;
1708 
1709 		err = cb(session, event, offset, data);
1710 		if (err)
1711 			return err;
1712 
1713 		offset += event->header.size;
1714 		if (event->header.type == PERF_RECORD_AUXTRACE)
1715 			offset += event->auxtrace.size;
1716 
1717 	} while (offset < max_offset);
1718 
1719 	return err;
1720 }
1721 
1722 static s64 perf_session__process_event(struct perf_session *session,
1723 				       union perf_event *event, u64 file_offset)
1724 {
1725 	struct evlist *evlist = session->evlist;
1726 	struct perf_tool *tool = session->tool;
1727 	int ret;
1728 
1729 	if (session->header.needs_swap)
1730 		event_swap(event, evlist__sample_id_all(evlist));
1731 
1732 	if (event->header.type >= PERF_RECORD_HEADER_MAX)
1733 		return -EINVAL;
1734 
1735 	events_stats__inc(&evlist->stats, event->header.type);
1736 
1737 	if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1738 		return perf_session__process_user_event(session, event, file_offset);
1739 
1740 	if (tool->ordered_events) {
1741 		u64 timestamp = -1ULL;
1742 
1743 		ret = perf_evlist__parse_sample_timestamp(evlist, event, &timestamp);
1744 		if (ret && ret != -1)
1745 			return ret;
1746 
1747 		ret = perf_session__queue_event(session, event, timestamp, file_offset);
1748 		if (ret != -ETIME)
1749 			return ret;
1750 	}
1751 
1752 	return perf_session__deliver_event(session, event, tool, file_offset);
1753 }
1754 
1755 void perf_event_header__bswap(struct perf_event_header *hdr)
1756 {
1757 	hdr->type = bswap_32(hdr->type);
1758 	hdr->misc = bswap_16(hdr->misc);
1759 	hdr->size = bswap_16(hdr->size);
1760 }
1761 
1762 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid)
1763 {
1764 	return machine__findnew_thread(&session->machines.host, -1, pid);
1765 }
1766 
1767 /*
1768  * Threads are identified by pid and tid, and the idle task has pid == tid == 0.
1769  * So here a single thread is created for that, but actually there is a separate
1770  * idle task per cpu, so there should be one 'struct thread' per cpu, but there
1771  * is only 1. That causes problems for some tools, requiring workarounds. For
1772  * example get_idle_thread() in builtin-sched.c, or thread_stack__per_cpu().
1773  */
1774 int perf_session__register_idle_thread(struct perf_session *session)
1775 {
1776 	struct thread *thread;
1777 	int err = 0;
1778 
1779 	thread = machine__findnew_thread(&session->machines.host, 0, 0);
1780 	if (thread == NULL || thread__set_comm(thread, "swapper", 0)) {
1781 		pr_err("problem inserting idle task.\n");
1782 		err = -1;
1783 	}
1784 
1785 	if (thread == NULL || thread__set_namespaces(thread, 0, NULL)) {
1786 		pr_err("problem inserting idle task.\n");
1787 		err = -1;
1788 	}
1789 
1790 	/* machine__findnew_thread() got the thread, so put it */
1791 	thread__put(thread);
1792 	return err;
1793 }
1794 
1795 static void
1796 perf_session__warn_order(const struct perf_session *session)
1797 {
1798 	const struct ordered_events *oe = &session->ordered_events;
1799 	struct evsel *evsel;
1800 	bool should_warn = true;
1801 
1802 	evlist__for_each_entry(session->evlist, evsel) {
1803 		if (evsel->core.attr.write_backward)
1804 			should_warn = false;
1805 	}
1806 
1807 	if (!should_warn)
1808 		return;
1809 	if (oe->nr_unordered_events != 0)
1810 		ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events);
1811 }
1812 
1813 static void perf_session__warn_about_errors(const struct perf_session *session)
1814 {
1815 	const struct events_stats *stats = &session->evlist->stats;
1816 
1817 	if (session->tool->lost == perf_event__process_lost &&
1818 	    stats->nr_events[PERF_RECORD_LOST] != 0) {
1819 		ui__warning("Processed %d events and lost %d chunks!\n\n"
1820 			    "Check IO/CPU overload!\n\n",
1821 			    stats->nr_events[0],
1822 			    stats->nr_events[PERF_RECORD_LOST]);
1823 	}
1824 
1825 	if (session->tool->lost_samples == perf_event__process_lost_samples) {
1826 		double drop_rate;
1827 
1828 		drop_rate = (double)stats->total_lost_samples /
1829 			    (double) (stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples);
1830 		if (drop_rate > 0.05) {
1831 			ui__warning("Processed %" PRIu64 " samples and lost %3.2f%%!\n\n",
1832 				    stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples,
1833 				    drop_rate * 100.0);
1834 		}
1835 	}
1836 
1837 	if (session->tool->aux == perf_event__process_aux &&
1838 	    stats->total_aux_lost != 0) {
1839 		ui__warning("AUX data lost %" PRIu64 " times out of %u!\n\n",
1840 			    stats->total_aux_lost,
1841 			    stats->nr_events[PERF_RECORD_AUX]);
1842 	}
1843 
1844 	if (session->tool->aux == perf_event__process_aux &&
1845 	    stats->total_aux_partial != 0) {
1846 		bool vmm_exclusive = false;
1847 
1848 		(void)sysfs__read_bool("module/kvm_intel/parameters/vmm_exclusive",
1849 		                       &vmm_exclusive);
1850 
1851 		ui__warning("AUX data had gaps in it %" PRIu64 " times out of %u!\n\n"
1852 		            "Are you running a KVM guest in the background?%s\n\n",
1853 			    stats->total_aux_partial,
1854 			    stats->nr_events[PERF_RECORD_AUX],
1855 			    vmm_exclusive ?
1856 			    "\nReloading kvm_intel module with vmm_exclusive=0\n"
1857 			    "will reduce the gaps to only guest's timeslices." :
1858 			    "");
1859 	}
1860 
1861 	if (stats->nr_unknown_events != 0) {
1862 		ui__warning("Found %u unknown events!\n\n"
1863 			    "Is this an older tool processing a perf.data "
1864 			    "file generated by a more recent tool?\n\n"
1865 			    "If that is not the case, consider "
1866 			    "reporting to linux-kernel@vger.kernel.org.\n\n",
1867 			    stats->nr_unknown_events);
1868 	}
1869 
1870 	if (stats->nr_unknown_id != 0) {
1871 		ui__warning("%u samples with id not present in the header\n",
1872 			    stats->nr_unknown_id);
1873 	}
1874 
1875 	if (stats->nr_invalid_chains != 0) {
1876 		ui__warning("Found invalid callchains!\n\n"
1877 			    "%u out of %u events were discarded for this reason.\n\n"
1878 			    "Consider reporting to linux-kernel@vger.kernel.org.\n\n",
1879 			    stats->nr_invalid_chains,
1880 			    stats->nr_events[PERF_RECORD_SAMPLE]);
1881 	}
1882 
1883 	if (stats->nr_unprocessable_samples != 0) {
1884 		ui__warning("%u unprocessable samples recorded.\n"
1885 			    "Do you have a KVM guest running and not using 'perf kvm'?\n",
1886 			    stats->nr_unprocessable_samples);
1887 	}
1888 
1889 	perf_session__warn_order(session);
1890 
1891 	events_stats__auxtrace_error_warn(stats);
1892 
1893 	if (stats->nr_proc_map_timeout != 0) {
1894 		ui__warning("%d map information files for pre-existing threads were\n"
1895 			    "not processed, if there are samples for addresses they\n"
1896 			    "will not be resolved, you may find out which are these\n"
1897 			    "threads by running with -v and redirecting the output\n"
1898 			    "to a file.\n"
1899 			    "The time limit to process proc map is too short?\n"
1900 			    "Increase it by --proc-map-timeout\n",
1901 			    stats->nr_proc_map_timeout);
1902 	}
1903 }
1904 
1905 static int perf_session__flush_thread_stack(struct thread *thread,
1906 					    void *p __maybe_unused)
1907 {
1908 	return thread_stack__flush(thread);
1909 }
1910 
1911 static int perf_session__flush_thread_stacks(struct perf_session *session)
1912 {
1913 	return machines__for_each_thread(&session->machines,
1914 					 perf_session__flush_thread_stack,
1915 					 NULL);
1916 }
1917 
1918 volatile int session_done;
1919 
1920 static int __perf_session__process_decomp_events(struct perf_session *session);
1921 
1922 static int __perf_session__process_pipe_events(struct perf_session *session)
1923 {
1924 	struct ordered_events *oe = &session->ordered_events;
1925 	struct perf_tool *tool = session->tool;
1926 	int fd = perf_data__fd(session->data);
1927 	union perf_event *event;
1928 	uint32_t size, cur_size = 0;
1929 	void *buf = NULL;
1930 	s64 skip = 0;
1931 	u64 head;
1932 	ssize_t err;
1933 	void *p;
1934 
1935 	perf_tool__fill_defaults(tool);
1936 
1937 	head = 0;
1938 	cur_size = sizeof(union perf_event);
1939 
1940 	buf = malloc(cur_size);
1941 	if (!buf)
1942 		return -errno;
1943 	ordered_events__set_copy_on_queue(oe, true);
1944 more:
1945 	event = buf;
1946 	err = readn(fd, event, sizeof(struct perf_event_header));
1947 	if (err <= 0) {
1948 		if (err == 0)
1949 			goto done;
1950 
1951 		pr_err("failed to read event header\n");
1952 		goto out_err;
1953 	}
1954 
1955 	if (session->header.needs_swap)
1956 		perf_event_header__bswap(&event->header);
1957 
1958 	size = event->header.size;
1959 	if (size < sizeof(struct perf_event_header)) {
1960 		pr_err("bad event header size\n");
1961 		goto out_err;
1962 	}
1963 
1964 	if (size > cur_size) {
1965 		void *new = realloc(buf, size);
1966 		if (!new) {
1967 			pr_err("failed to allocate memory to read event\n");
1968 			goto out_err;
1969 		}
1970 		buf = new;
1971 		cur_size = size;
1972 		event = buf;
1973 	}
1974 	p = event;
1975 	p += sizeof(struct perf_event_header);
1976 
1977 	if (size - sizeof(struct perf_event_header)) {
1978 		err = readn(fd, p, size - sizeof(struct perf_event_header));
1979 		if (err <= 0) {
1980 			if (err == 0) {
1981 				pr_err("unexpected end of event stream\n");
1982 				goto done;
1983 			}
1984 
1985 			pr_err("failed to read event data\n");
1986 			goto out_err;
1987 		}
1988 	}
1989 
1990 	if ((skip = perf_session__process_event(session, event, head)) < 0) {
1991 		pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1992 		       head, event->header.size, event->header.type);
1993 		err = -EINVAL;
1994 		goto out_err;
1995 	}
1996 
1997 	head += size;
1998 
1999 	if (skip > 0)
2000 		head += skip;
2001 
2002 	err = __perf_session__process_decomp_events(session);
2003 	if (err)
2004 		goto out_err;
2005 
2006 	if (!session_done())
2007 		goto more;
2008 done:
2009 	/* do the final flush for ordered samples */
2010 	err = ordered_events__flush(oe, OE_FLUSH__FINAL);
2011 	if (err)
2012 		goto out_err;
2013 	err = auxtrace__flush_events(session, tool);
2014 	if (err)
2015 		goto out_err;
2016 	err = perf_session__flush_thread_stacks(session);
2017 out_err:
2018 	free(buf);
2019 	if (!tool->no_warn)
2020 		perf_session__warn_about_errors(session);
2021 	ordered_events__free(&session->ordered_events);
2022 	auxtrace__free_events(session);
2023 	return err;
2024 }
2025 
2026 static union perf_event *
2027 prefetch_event(char *buf, u64 head, size_t mmap_size,
2028 	       bool needs_swap, union perf_event *error)
2029 {
2030 	union perf_event *event;
2031 
2032 	/*
2033 	 * Ensure we have enough space remaining to read
2034 	 * the size of the event in the headers.
2035 	 */
2036 	if (head + sizeof(event->header) > mmap_size)
2037 		return NULL;
2038 
2039 	event = (union perf_event *)(buf + head);
2040 	if (needs_swap)
2041 		perf_event_header__bswap(&event->header);
2042 
2043 	if (head + event->header.size <= mmap_size)
2044 		return event;
2045 
2046 	/* We're not fetching the event so swap back again */
2047 	if (needs_swap)
2048 		perf_event_header__bswap(&event->header);
2049 
2050 	pr_debug("%s: head=%#" PRIx64 " event->header_size=%#x, mmap_size=%#zx:"
2051 		 " fuzzed or compressed perf.data?\n",__func__, head, event->header.size, mmap_size);
2052 
2053 	return error;
2054 }
2055 
2056 static union perf_event *
2057 fetch_mmaped_event(u64 head, size_t mmap_size, char *buf, bool needs_swap)
2058 {
2059 	return prefetch_event(buf, head, mmap_size, needs_swap, ERR_PTR(-EINVAL));
2060 }
2061 
2062 static union perf_event *
2063 fetch_decomp_event(u64 head, size_t mmap_size, char *buf, bool needs_swap)
2064 {
2065 	return prefetch_event(buf, head, mmap_size, needs_swap, NULL);
2066 }
2067 
2068 static int __perf_session__process_decomp_events(struct perf_session *session)
2069 {
2070 	s64 skip;
2071 	u64 size, file_pos = 0;
2072 	struct decomp *decomp = session->decomp_last;
2073 
2074 	if (!decomp)
2075 		return 0;
2076 
2077 	while (decomp->head < decomp->size && !session_done()) {
2078 		union perf_event *event = fetch_decomp_event(decomp->head, decomp->size, decomp->data,
2079 							     session->header.needs_swap);
2080 
2081 		if (!event)
2082 			break;
2083 
2084 		size = event->header.size;
2085 
2086 		if (size < sizeof(struct perf_event_header) ||
2087 		    (skip = perf_session__process_event(session, event, file_pos)) < 0) {
2088 			pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
2089 				decomp->file_pos + decomp->head, event->header.size, event->header.type);
2090 			return -EINVAL;
2091 		}
2092 
2093 		if (skip)
2094 			size += skip;
2095 
2096 		decomp->head += size;
2097 	}
2098 
2099 	return 0;
2100 }
2101 
2102 /*
2103  * On 64bit we can mmap the data file in one go. No need for tiny mmap
2104  * slices. On 32bit we use 32MB.
2105  */
2106 #if BITS_PER_LONG == 64
2107 #define MMAP_SIZE ULLONG_MAX
2108 #define NUM_MMAPS 1
2109 #else
2110 #define MMAP_SIZE (32 * 1024 * 1024ULL)
2111 #define NUM_MMAPS 128
2112 #endif
2113 
2114 struct reader;
2115 
2116 typedef s64 (*reader_cb_t)(struct perf_session *session,
2117 			   union perf_event *event,
2118 			   u64 file_offset);
2119 
2120 struct reader {
2121 	int		 fd;
2122 	u64		 data_size;
2123 	u64		 data_offset;
2124 	reader_cb_t	 process;
2125 };
2126 
2127 static int
2128 reader__process_events(struct reader *rd, struct perf_session *session,
2129 		       struct ui_progress *prog)
2130 {
2131 	u64 data_size = rd->data_size;
2132 	u64 head, page_offset, file_offset, file_pos, size;
2133 	int err = 0, mmap_prot, mmap_flags, map_idx = 0;
2134 	size_t	mmap_size;
2135 	char *buf, *mmaps[NUM_MMAPS];
2136 	union perf_event *event;
2137 	s64 skip;
2138 
2139 	page_offset = page_size * (rd->data_offset / page_size);
2140 	file_offset = page_offset;
2141 	head = rd->data_offset - page_offset;
2142 
2143 	ui_progress__init_size(prog, data_size, "Processing events...");
2144 
2145 	data_size += rd->data_offset;
2146 
2147 	mmap_size = MMAP_SIZE;
2148 	if (mmap_size > data_size) {
2149 		mmap_size = data_size;
2150 		session->one_mmap = true;
2151 	}
2152 
2153 	memset(mmaps, 0, sizeof(mmaps));
2154 
2155 	mmap_prot  = PROT_READ;
2156 	mmap_flags = MAP_SHARED;
2157 
2158 	if (session->header.needs_swap) {
2159 		mmap_prot  |= PROT_WRITE;
2160 		mmap_flags = MAP_PRIVATE;
2161 	}
2162 remap:
2163 	buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, rd->fd,
2164 		   file_offset);
2165 	if (buf == MAP_FAILED) {
2166 		pr_err("failed to mmap file\n");
2167 		err = -errno;
2168 		goto out;
2169 	}
2170 	mmaps[map_idx] = buf;
2171 	map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1);
2172 	file_pos = file_offset + head;
2173 	if (session->one_mmap) {
2174 		session->one_mmap_addr = buf;
2175 		session->one_mmap_offset = file_offset;
2176 	}
2177 
2178 more:
2179 	event = fetch_mmaped_event(head, mmap_size, buf, session->header.needs_swap);
2180 	if (IS_ERR(event))
2181 		return PTR_ERR(event);
2182 
2183 	if (!event) {
2184 		if (mmaps[map_idx]) {
2185 			munmap(mmaps[map_idx], mmap_size);
2186 			mmaps[map_idx] = NULL;
2187 		}
2188 
2189 		page_offset = page_size * (head / page_size);
2190 		file_offset += page_offset;
2191 		head -= page_offset;
2192 		goto remap;
2193 	}
2194 
2195 	size = event->header.size;
2196 
2197 	skip = -EINVAL;
2198 
2199 	if (size < sizeof(struct perf_event_header) ||
2200 	    (skip = rd->process(session, event, file_pos)) < 0) {
2201 		pr_err("%#" PRIx64 " [%#x]: failed to process type: %d [%s]\n",
2202 		       file_offset + head, event->header.size,
2203 		       event->header.type, strerror(-skip));
2204 		err = skip;
2205 		goto out;
2206 	}
2207 
2208 	if (skip)
2209 		size += skip;
2210 
2211 	head += size;
2212 	file_pos += size;
2213 
2214 	err = __perf_session__process_decomp_events(session);
2215 	if (err)
2216 		goto out;
2217 
2218 	ui_progress__update(prog, size);
2219 
2220 	if (session_done())
2221 		goto out;
2222 
2223 	if (file_pos < data_size)
2224 		goto more;
2225 
2226 out:
2227 	return err;
2228 }
2229 
2230 static s64 process_simple(struct perf_session *session,
2231 			  union perf_event *event,
2232 			  u64 file_offset)
2233 {
2234 	return perf_session__process_event(session, event, file_offset);
2235 }
2236 
2237 static int __perf_session__process_events(struct perf_session *session)
2238 {
2239 	struct reader rd = {
2240 		.fd		= perf_data__fd(session->data),
2241 		.data_size	= session->header.data_size,
2242 		.data_offset	= session->header.data_offset,
2243 		.process	= process_simple,
2244 	};
2245 	struct ordered_events *oe = &session->ordered_events;
2246 	struct perf_tool *tool = session->tool;
2247 	struct ui_progress prog;
2248 	int err;
2249 
2250 	perf_tool__fill_defaults(tool);
2251 
2252 	if (rd.data_size == 0)
2253 		return -1;
2254 
2255 	ui_progress__init_size(&prog, rd.data_size, "Processing events...");
2256 
2257 	err = reader__process_events(&rd, session, &prog);
2258 	if (err)
2259 		goto out_err;
2260 	/* do the final flush for ordered samples */
2261 	err = ordered_events__flush(oe, OE_FLUSH__FINAL);
2262 	if (err)
2263 		goto out_err;
2264 	err = auxtrace__flush_events(session, tool);
2265 	if (err)
2266 		goto out_err;
2267 	err = perf_session__flush_thread_stacks(session);
2268 out_err:
2269 	ui_progress__finish();
2270 	if (!tool->no_warn)
2271 		perf_session__warn_about_errors(session);
2272 	/*
2273 	 * We may switching perf.data output, make ordered_events
2274 	 * reusable.
2275 	 */
2276 	ordered_events__reinit(&session->ordered_events);
2277 	auxtrace__free_events(session);
2278 	session->one_mmap = false;
2279 	return err;
2280 }
2281 
2282 int perf_session__process_events(struct perf_session *session)
2283 {
2284 	if (perf_session__register_idle_thread(session) < 0)
2285 		return -ENOMEM;
2286 
2287 	if (perf_data__is_pipe(session->data))
2288 		return __perf_session__process_pipe_events(session);
2289 
2290 	return __perf_session__process_events(session);
2291 }
2292 
2293 bool perf_session__has_traces(struct perf_session *session, const char *msg)
2294 {
2295 	struct evsel *evsel;
2296 
2297 	evlist__for_each_entry(session->evlist, evsel) {
2298 		if (evsel->core.attr.type == PERF_TYPE_TRACEPOINT)
2299 			return true;
2300 	}
2301 
2302 	pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg);
2303 	return false;
2304 }
2305 
2306 int map__set_kallsyms_ref_reloc_sym(struct map *map, const char *symbol_name, u64 addr)
2307 {
2308 	char *bracket;
2309 	struct ref_reloc_sym *ref;
2310 	struct kmap *kmap;
2311 
2312 	ref = zalloc(sizeof(struct ref_reloc_sym));
2313 	if (ref == NULL)
2314 		return -ENOMEM;
2315 
2316 	ref->name = strdup(symbol_name);
2317 	if (ref->name == NULL) {
2318 		free(ref);
2319 		return -ENOMEM;
2320 	}
2321 
2322 	bracket = strchr(ref->name, ']');
2323 	if (bracket)
2324 		*bracket = '\0';
2325 
2326 	ref->addr = addr;
2327 
2328 	kmap = map__kmap(map);
2329 	if (kmap)
2330 		kmap->ref_reloc_sym = ref;
2331 
2332 	return 0;
2333 }
2334 
2335 size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp)
2336 {
2337 	return machines__fprintf_dsos(&session->machines, fp);
2338 }
2339 
2340 size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp,
2341 					  bool (skip)(struct dso *dso, int parm), int parm)
2342 {
2343 	return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm);
2344 }
2345 
2346 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp)
2347 {
2348 	size_t ret;
2349 	const char *msg = "";
2350 
2351 	if (perf_header__has_feat(&session->header, HEADER_AUXTRACE))
2352 		msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)";
2353 
2354 	ret = fprintf(fp, "\nAggregated stats:%s\n", msg);
2355 
2356 	ret += events_stats__fprintf(&session->evlist->stats, fp);
2357 	return ret;
2358 }
2359 
2360 size_t perf_session__fprintf(struct perf_session *session, FILE *fp)
2361 {
2362 	/*
2363 	 * FIXME: Here we have to actually print all the machines in this
2364 	 * session, not just the host...
2365 	 */
2366 	return machine__fprintf(&session->machines.host, fp);
2367 }
2368 
2369 struct evsel *perf_session__find_first_evtype(struct perf_session *session,
2370 					      unsigned int type)
2371 {
2372 	struct evsel *pos;
2373 
2374 	evlist__for_each_entry(session->evlist, pos) {
2375 		if (pos->core.attr.type == type)
2376 			return pos;
2377 	}
2378 	return NULL;
2379 }
2380 
2381 int perf_session__cpu_bitmap(struct perf_session *session,
2382 			     const char *cpu_list, unsigned long *cpu_bitmap)
2383 {
2384 	int i, err = -1;
2385 	struct perf_cpu_map *map;
2386 	int nr_cpus = min(session->header.env.nr_cpus_online, MAX_NR_CPUS);
2387 
2388 	for (i = 0; i < PERF_TYPE_MAX; ++i) {
2389 		struct evsel *evsel;
2390 
2391 		evsel = perf_session__find_first_evtype(session, i);
2392 		if (!evsel)
2393 			continue;
2394 
2395 		if (!(evsel->core.attr.sample_type & PERF_SAMPLE_CPU)) {
2396 			pr_err("File does not contain CPU events. "
2397 			       "Remove -C option to proceed.\n");
2398 			return -1;
2399 		}
2400 	}
2401 
2402 	map = perf_cpu_map__new(cpu_list);
2403 	if (map == NULL) {
2404 		pr_err("Invalid cpu_list\n");
2405 		return -1;
2406 	}
2407 
2408 	for (i = 0; i < map->nr; i++) {
2409 		int cpu = map->map[i];
2410 
2411 		if (cpu >= nr_cpus) {
2412 			pr_err("Requested CPU %d too large. "
2413 			       "Consider raising MAX_NR_CPUS\n", cpu);
2414 			goto out_delete_map;
2415 		}
2416 
2417 		set_bit(cpu, cpu_bitmap);
2418 	}
2419 
2420 	err = 0;
2421 
2422 out_delete_map:
2423 	perf_cpu_map__put(map);
2424 	return err;
2425 }
2426 
2427 void perf_session__fprintf_info(struct perf_session *session, FILE *fp,
2428 				bool full)
2429 {
2430 	if (session == NULL || fp == NULL)
2431 		return;
2432 
2433 	fprintf(fp, "# ========\n");
2434 	perf_header__fprintf_info(session, fp, full);
2435 	fprintf(fp, "# ========\n#\n");
2436 }
2437 
2438 int perf_event__process_id_index(struct perf_session *session,
2439 				 union perf_event *event)
2440 {
2441 	struct evlist *evlist = session->evlist;
2442 	struct perf_record_id_index *ie = &event->id_index;
2443 	size_t i, nr, max_nr;
2444 
2445 	max_nr = (ie->header.size - sizeof(struct perf_record_id_index)) /
2446 		 sizeof(struct id_index_entry);
2447 	nr = ie->nr;
2448 	if (nr > max_nr)
2449 		return -EINVAL;
2450 
2451 	if (dump_trace)
2452 		fprintf(stdout, " nr: %zu\n", nr);
2453 
2454 	for (i = 0; i < nr; i++) {
2455 		struct id_index_entry *e = &ie->entries[i];
2456 		struct perf_sample_id *sid;
2457 
2458 		if (dump_trace) {
2459 			fprintf(stdout,	" ... id: %"PRI_lu64, e->id);
2460 			fprintf(stdout,	"  idx: %"PRI_lu64, e->idx);
2461 			fprintf(stdout,	"  cpu: %"PRI_ld64, e->cpu);
2462 			fprintf(stdout,	"  tid: %"PRI_ld64"\n", e->tid);
2463 		}
2464 
2465 		sid = perf_evlist__id2sid(evlist, e->id);
2466 		if (!sid)
2467 			return -ENOENT;
2468 		sid->idx = e->idx;
2469 		sid->cpu = e->cpu;
2470 		sid->tid = e->tid;
2471 	}
2472 	return 0;
2473 }
2474