xref: /openbmc/linux/tools/perf/util/session.c (revision 2d99a7ec)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <errno.h>
3 #include <signal.h>
4 #include <inttypes.h>
5 #include <linux/err.h>
6 #include <linux/kernel.h>
7 #include <linux/zalloc.h>
8 #include <api/fs/fs.h>
9 
10 #include <byteswap.h>
11 #include <unistd.h>
12 #include <sys/types.h>
13 #include <sys/mman.h>
14 #include <perf/cpumap.h>
15 
16 #include "map_symbol.h"
17 #include "branch.h"
18 #include "debug.h"
19 #include "env.h"
20 #include "evlist.h"
21 #include "evsel.h"
22 #include "memswap.h"
23 #include "map.h"
24 #include "symbol.h"
25 #include "session.h"
26 #include "tool.h"
27 #include "perf_regs.h"
28 #include "asm/bug.h"
29 #include "auxtrace.h"
30 #include "thread.h"
31 #include "thread-stack.h"
32 #include "sample-raw.h"
33 #include "stat.h"
34 #include "tsc.h"
35 #include "ui/progress.h"
36 #include "../perf.h"
37 #include "arch/common.h"
38 #include "units.h"
39 #include <internal/lib.h>
40 
41 #ifdef HAVE_ZSTD_SUPPORT
42 static int perf_session__process_compressed_event(struct perf_session *session,
43 						  union perf_event *event, u64 file_offset,
44 						  const char *file_path)
45 {
46 	void *src;
47 	size_t decomp_size, src_size;
48 	u64 decomp_last_rem = 0;
49 	size_t mmap_len, decomp_len = session->header.env.comp_mmap_len;
50 	struct decomp *decomp, *decomp_last = session->active_decomp->decomp_last;
51 
52 	if (decomp_last) {
53 		decomp_last_rem = decomp_last->size - decomp_last->head;
54 		decomp_len += decomp_last_rem;
55 	}
56 
57 	mmap_len = sizeof(struct decomp) + decomp_len;
58 	decomp = mmap(NULL, mmap_len, PROT_READ|PROT_WRITE,
59 		      MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);
60 	if (decomp == MAP_FAILED) {
61 		pr_err("Couldn't allocate memory for decompression\n");
62 		return -1;
63 	}
64 
65 	decomp->file_pos = file_offset;
66 	decomp->file_path = file_path;
67 	decomp->mmap_len = mmap_len;
68 	decomp->head = 0;
69 
70 	if (decomp_last_rem) {
71 		memcpy(decomp->data, &(decomp_last->data[decomp_last->head]), decomp_last_rem);
72 		decomp->size = decomp_last_rem;
73 	}
74 
75 	src = (void *)event + sizeof(struct perf_record_compressed);
76 	src_size = event->pack.header.size - sizeof(struct perf_record_compressed);
77 
78 	decomp_size = zstd_decompress_stream(session->active_decomp->zstd_decomp, src, src_size,
79 				&(decomp->data[decomp_last_rem]), decomp_len - decomp_last_rem);
80 	if (!decomp_size) {
81 		munmap(decomp, mmap_len);
82 		pr_err("Couldn't decompress data\n");
83 		return -1;
84 	}
85 
86 	decomp->size += decomp_size;
87 
88 	if (session->active_decomp->decomp == NULL)
89 		session->active_decomp->decomp = decomp;
90 	else
91 		session->active_decomp->decomp_last->next = decomp;
92 
93 	session->active_decomp->decomp_last = decomp;
94 
95 	pr_debug("decomp (B): %zd to %zd\n", src_size, decomp_size);
96 
97 	return 0;
98 }
99 #else /* !HAVE_ZSTD_SUPPORT */
100 #define perf_session__process_compressed_event perf_session__process_compressed_event_stub
101 #endif
102 
103 static int perf_session__deliver_event(struct perf_session *session,
104 				       union perf_event *event,
105 				       struct perf_tool *tool,
106 				       u64 file_offset,
107 				       const char *file_path);
108 
109 static int perf_session__open(struct perf_session *session, int repipe_fd)
110 {
111 	struct perf_data *data = session->data;
112 
113 	if (perf_session__read_header(session, repipe_fd) < 0) {
114 		pr_err("incompatible file format (rerun with -v to learn more)\n");
115 		return -1;
116 	}
117 
118 	if (perf_data__is_pipe(data))
119 		return 0;
120 
121 	if (perf_header__has_feat(&session->header, HEADER_STAT))
122 		return 0;
123 
124 	if (!evlist__valid_sample_type(session->evlist)) {
125 		pr_err("non matching sample_type\n");
126 		return -1;
127 	}
128 
129 	if (!evlist__valid_sample_id_all(session->evlist)) {
130 		pr_err("non matching sample_id_all\n");
131 		return -1;
132 	}
133 
134 	if (!evlist__valid_read_format(session->evlist)) {
135 		pr_err("non matching read_format\n");
136 		return -1;
137 	}
138 
139 	return 0;
140 }
141 
142 void perf_session__set_id_hdr_size(struct perf_session *session)
143 {
144 	u16 id_hdr_size = evlist__id_hdr_size(session->evlist);
145 
146 	machines__set_id_hdr_size(&session->machines, id_hdr_size);
147 }
148 
149 int perf_session__create_kernel_maps(struct perf_session *session)
150 {
151 	int ret = machine__create_kernel_maps(&session->machines.host);
152 
153 	if (ret >= 0)
154 		ret = machines__create_guest_kernel_maps(&session->machines);
155 	return ret;
156 }
157 
158 static void perf_session__destroy_kernel_maps(struct perf_session *session)
159 {
160 	machines__destroy_kernel_maps(&session->machines);
161 }
162 
163 static bool perf_session__has_comm_exec(struct perf_session *session)
164 {
165 	struct evsel *evsel;
166 
167 	evlist__for_each_entry(session->evlist, evsel) {
168 		if (evsel->core.attr.comm_exec)
169 			return true;
170 	}
171 
172 	return false;
173 }
174 
175 static void perf_session__set_comm_exec(struct perf_session *session)
176 {
177 	bool comm_exec = perf_session__has_comm_exec(session);
178 
179 	machines__set_comm_exec(&session->machines, comm_exec);
180 }
181 
182 static int ordered_events__deliver_event(struct ordered_events *oe,
183 					 struct ordered_event *event)
184 {
185 	struct perf_session *session = container_of(oe, struct perf_session,
186 						    ordered_events);
187 
188 	return perf_session__deliver_event(session, event->event,
189 					   session->tool, event->file_offset,
190 					   event->file_path);
191 }
192 
193 struct perf_session *__perf_session__new(struct perf_data *data,
194 					 bool repipe, int repipe_fd,
195 					 struct perf_tool *tool)
196 {
197 	int ret = -ENOMEM;
198 	struct perf_session *session = zalloc(sizeof(*session));
199 
200 	if (!session)
201 		goto out;
202 
203 	session->repipe = repipe;
204 	session->tool   = tool;
205 	session->decomp_data.zstd_decomp = &session->zstd_data;
206 	session->active_decomp = &session->decomp_data;
207 	INIT_LIST_HEAD(&session->auxtrace_index);
208 	machines__init(&session->machines);
209 	ordered_events__init(&session->ordered_events,
210 			     ordered_events__deliver_event, NULL);
211 
212 	perf_env__init(&session->header.env);
213 	if (data) {
214 		ret = perf_data__open(data);
215 		if (ret < 0)
216 			goto out_delete;
217 
218 		session->data = data;
219 
220 		if (perf_data__is_read(data)) {
221 			ret = perf_session__open(session, repipe_fd);
222 			if (ret < 0)
223 				goto out_delete;
224 
225 			/*
226 			 * set session attributes that are present in perf.data
227 			 * but not in pipe-mode.
228 			 */
229 			if (!data->is_pipe) {
230 				perf_session__set_id_hdr_size(session);
231 				perf_session__set_comm_exec(session);
232 			}
233 
234 			evlist__init_trace_event_sample_raw(session->evlist);
235 
236 			/* Open the directory data. */
237 			if (data->is_dir) {
238 				ret = perf_data__open_dir(data);
239 				if (ret)
240 					goto out_delete;
241 			}
242 
243 			if (!symbol_conf.kallsyms_name &&
244 			    !symbol_conf.vmlinux_name)
245 				symbol_conf.kallsyms_name = perf_data__kallsyms_name(data);
246 		}
247 	} else  {
248 		session->machines.host.env = &perf_env;
249 	}
250 
251 	session->machines.host.single_address_space =
252 		perf_env__single_address_space(session->machines.host.env);
253 
254 	if (!data || perf_data__is_write(data)) {
255 		/*
256 		 * In O_RDONLY mode this will be performed when reading the
257 		 * kernel MMAP event, in perf_event__process_mmap().
258 		 */
259 		if (perf_session__create_kernel_maps(session) < 0)
260 			pr_warning("Cannot read kernel map\n");
261 	}
262 
263 	/*
264 	 * In pipe-mode, evlist is empty until PERF_RECORD_HEADER_ATTR is
265 	 * processed, so evlist__sample_id_all is not meaningful here.
266 	 */
267 	if ((!data || !data->is_pipe) && tool && tool->ordering_requires_timestamps &&
268 	    tool->ordered_events && !evlist__sample_id_all(session->evlist)) {
269 		dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n");
270 		tool->ordered_events = false;
271 	}
272 
273 	return session;
274 
275  out_delete:
276 	perf_session__delete(session);
277  out:
278 	return ERR_PTR(ret);
279 }
280 
281 static void perf_session__delete_threads(struct perf_session *session)
282 {
283 	machine__delete_threads(&session->machines.host);
284 }
285 
286 static void perf_decomp__release_events(struct decomp *next)
287 {
288 	struct decomp *decomp;
289 	size_t mmap_len;
290 
291 	do {
292 		decomp = next;
293 		if (decomp == NULL)
294 			break;
295 		next = decomp->next;
296 		mmap_len = decomp->mmap_len;
297 		munmap(decomp, mmap_len);
298 	} while (1);
299 }
300 
301 void perf_session__delete(struct perf_session *session)
302 {
303 	if (session == NULL)
304 		return;
305 	auxtrace__free(session);
306 	auxtrace_index__free(&session->auxtrace_index);
307 	perf_session__destroy_kernel_maps(session);
308 	perf_session__delete_threads(session);
309 	perf_decomp__release_events(session->decomp_data.decomp);
310 	perf_env__exit(&session->header.env);
311 	machines__exit(&session->machines);
312 	if (session->data) {
313 		if (perf_data__is_read(session->data))
314 			evlist__delete(session->evlist);
315 		perf_data__close(session->data);
316 	}
317 #ifdef HAVE_LIBTRACEEVENT
318 	trace_event__cleanup(&session->tevent);
319 #endif
320 	free(session);
321 }
322 
323 static int process_event_synth_tracing_data_stub(struct perf_session *session
324 						 __maybe_unused,
325 						 union perf_event *event
326 						 __maybe_unused)
327 {
328 	dump_printf(": unhandled!\n");
329 	return 0;
330 }
331 
332 static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused,
333 					 union perf_event *event __maybe_unused,
334 					 struct evlist **pevlist
335 					 __maybe_unused)
336 {
337 	dump_printf(": unhandled!\n");
338 	return 0;
339 }
340 
341 static int process_event_synth_event_update_stub(struct perf_tool *tool __maybe_unused,
342 						 union perf_event *event __maybe_unused,
343 						 struct evlist **pevlist
344 						 __maybe_unused)
345 {
346 	if (dump_trace)
347 		perf_event__fprintf_event_update(event, stdout);
348 
349 	dump_printf(": unhandled!\n");
350 	return 0;
351 }
352 
353 static int process_event_sample_stub(struct perf_tool *tool __maybe_unused,
354 				     union perf_event *event __maybe_unused,
355 				     struct perf_sample *sample __maybe_unused,
356 				     struct evsel *evsel __maybe_unused,
357 				     struct machine *machine __maybe_unused)
358 {
359 	dump_printf(": unhandled!\n");
360 	return 0;
361 }
362 
363 static int process_event_stub(struct perf_tool *tool __maybe_unused,
364 			      union perf_event *event __maybe_unused,
365 			      struct perf_sample *sample __maybe_unused,
366 			      struct machine *machine __maybe_unused)
367 {
368 	dump_printf(": unhandled!\n");
369 	return 0;
370 }
371 
372 static int process_finished_round_stub(struct perf_tool *tool __maybe_unused,
373 				       union perf_event *event __maybe_unused,
374 				       struct ordered_events *oe __maybe_unused)
375 {
376 	dump_printf(": unhandled!\n");
377 	return 0;
378 }
379 
380 static int skipn(int fd, off_t n)
381 {
382 	char buf[4096];
383 	ssize_t ret;
384 
385 	while (n > 0) {
386 		ret = read(fd, buf, min(n, (off_t)sizeof(buf)));
387 		if (ret <= 0)
388 			return ret;
389 		n -= ret;
390 	}
391 
392 	return 0;
393 }
394 
395 static s64 process_event_auxtrace_stub(struct perf_session *session __maybe_unused,
396 				       union perf_event *event)
397 {
398 	dump_printf(": unhandled!\n");
399 	if (perf_data__is_pipe(session->data))
400 		skipn(perf_data__fd(session->data), event->auxtrace.size);
401 	return event->auxtrace.size;
402 }
403 
404 static int process_event_op2_stub(struct perf_session *session __maybe_unused,
405 				  union perf_event *event __maybe_unused)
406 {
407 	dump_printf(": unhandled!\n");
408 	return 0;
409 }
410 
411 
412 static
413 int process_event_thread_map_stub(struct perf_session *session __maybe_unused,
414 				  union perf_event *event __maybe_unused)
415 {
416 	if (dump_trace)
417 		perf_event__fprintf_thread_map(event, stdout);
418 
419 	dump_printf(": unhandled!\n");
420 	return 0;
421 }
422 
423 static
424 int process_event_cpu_map_stub(struct perf_session *session __maybe_unused,
425 			       union perf_event *event __maybe_unused)
426 {
427 	if (dump_trace)
428 		perf_event__fprintf_cpu_map(event, stdout);
429 
430 	dump_printf(": unhandled!\n");
431 	return 0;
432 }
433 
434 static
435 int process_event_stat_config_stub(struct perf_session *session __maybe_unused,
436 				   union perf_event *event __maybe_unused)
437 {
438 	if (dump_trace)
439 		perf_event__fprintf_stat_config(event, stdout);
440 
441 	dump_printf(": unhandled!\n");
442 	return 0;
443 }
444 
445 static int process_stat_stub(struct perf_session *perf_session __maybe_unused,
446 			     union perf_event *event)
447 {
448 	if (dump_trace)
449 		perf_event__fprintf_stat(event, stdout);
450 
451 	dump_printf(": unhandled!\n");
452 	return 0;
453 }
454 
455 static int process_stat_round_stub(struct perf_session *perf_session __maybe_unused,
456 				   union perf_event *event)
457 {
458 	if (dump_trace)
459 		perf_event__fprintf_stat_round(event, stdout);
460 
461 	dump_printf(": unhandled!\n");
462 	return 0;
463 }
464 
465 static int process_event_time_conv_stub(struct perf_session *perf_session __maybe_unused,
466 					union perf_event *event)
467 {
468 	if (dump_trace)
469 		perf_event__fprintf_time_conv(event, stdout);
470 
471 	dump_printf(": unhandled!\n");
472 	return 0;
473 }
474 
475 static int perf_session__process_compressed_event_stub(struct perf_session *session __maybe_unused,
476 						       union perf_event *event __maybe_unused,
477 						       u64 file_offset __maybe_unused,
478 						       const char *file_path __maybe_unused)
479 {
480        dump_printf(": unhandled!\n");
481        return 0;
482 }
483 
484 void perf_tool__fill_defaults(struct perf_tool *tool)
485 {
486 	if (tool->sample == NULL)
487 		tool->sample = process_event_sample_stub;
488 	if (tool->mmap == NULL)
489 		tool->mmap = process_event_stub;
490 	if (tool->mmap2 == NULL)
491 		tool->mmap2 = process_event_stub;
492 	if (tool->comm == NULL)
493 		tool->comm = process_event_stub;
494 	if (tool->namespaces == NULL)
495 		tool->namespaces = process_event_stub;
496 	if (tool->cgroup == NULL)
497 		tool->cgroup = process_event_stub;
498 	if (tool->fork == NULL)
499 		tool->fork = process_event_stub;
500 	if (tool->exit == NULL)
501 		tool->exit = process_event_stub;
502 	if (tool->lost == NULL)
503 		tool->lost = perf_event__process_lost;
504 	if (tool->lost_samples == NULL)
505 		tool->lost_samples = perf_event__process_lost_samples;
506 	if (tool->aux == NULL)
507 		tool->aux = perf_event__process_aux;
508 	if (tool->itrace_start == NULL)
509 		tool->itrace_start = perf_event__process_itrace_start;
510 	if (tool->context_switch == NULL)
511 		tool->context_switch = perf_event__process_switch;
512 	if (tool->ksymbol == NULL)
513 		tool->ksymbol = perf_event__process_ksymbol;
514 	if (tool->bpf == NULL)
515 		tool->bpf = perf_event__process_bpf;
516 	if (tool->text_poke == NULL)
517 		tool->text_poke = perf_event__process_text_poke;
518 	if (tool->aux_output_hw_id == NULL)
519 		tool->aux_output_hw_id = perf_event__process_aux_output_hw_id;
520 	if (tool->read == NULL)
521 		tool->read = process_event_sample_stub;
522 	if (tool->throttle == NULL)
523 		tool->throttle = process_event_stub;
524 	if (tool->unthrottle == NULL)
525 		tool->unthrottle = process_event_stub;
526 	if (tool->attr == NULL)
527 		tool->attr = process_event_synth_attr_stub;
528 	if (tool->event_update == NULL)
529 		tool->event_update = process_event_synth_event_update_stub;
530 	if (tool->tracing_data == NULL)
531 		tool->tracing_data = process_event_synth_tracing_data_stub;
532 	if (tool->build_id == NULL)
533 		tool->build_id = process_event_op2_stub;
534 	if (tool->finished_round == NULL) {
535 		if (tool->ordered_events)
536 			tool->finished_round = perf_event__process_finished_round;
537 		else
538 			tool->finished_round = process_finished_round_stub;
539 	}
540 	if (tool->id_index == NULL)
541 		tool->id_index = process_event_op2_stub;
542 	if (tool->auxtrace_info == NULL)
543 		tool->auxtrace_info = process_event_op2_stub;
544 	if (tool->auxtrace == NULL)
545 		tool->auxtrace = process_event_auxtrace_stub;
546 	if (tool->auxtrace_error == NULL)
547 		tool->auxtrace_error = process_event_op2_stub;
548 	if (tool->thread_map == NULL)
549 		tool->thread_map = process_event_thread_map_stub;
550 	if (tool->cpu_map == NULL)
551 		tool->cpu_map = process_event_cpu_map_stub;
552 	if (tool->stat_config == NULL)
553 		tool->stat_config = process_event_stat_config_stub;
554 	if (tool->stat == NULL)
555 		tool->stat = process_stat_stub;
556 	if (tool->stat_round == NULL)
557 		tool->stat_round = process_stat_round_stub;
558 	if (tool->time_conv == NULL)
559 		tool->time_conv = process_event_time_conv_stub;
560 	if (tool->feature == NULL)
561 		tool->feature = process_event_op2_stub;
562 	if (tool->compressed == NULL)
563 		tool->compressed = perf_session__process_compressed_event;
564 	if (tool->finished_init == NULL)
565 		tool->finished_init = process_event_op2_stub;
566 }
567 
568 static void swap_sample_id_all(union perf_event *event, void *data)
569 {
570 	void *end = (void *) event + event->header.size;
571 	int size = end - data;
572 
573 	BUG_ON(size % sizeof(u64));
574 	mem_bswap_64(data, size);
575 }
576 
577 static void perf_event__all64_swap(union perf_event *event,
578 				   bool sample_id_all __maybe_unused)
579 {
580 	struct perf_event_header *hdr = &event->header;
581 	mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr));
582 }
583 
584 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all)
585 {
586 	event->comm.pid = bswap_32(event->comm.pid);
587 	event->comm.tid = bswap_32(event->comm.tid);
588 
589 	if (sample_id_all) {
590 		void *data = &event->comm.comm;
591 
592 		data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
593 		swap_sample_id_all(event, data);
594 	}
595 }
596 
597 static void perf_event__mmap_swap(union perf_event *event,
598 				  bool sample_id_all)
599 {
600 	event->mmap.pid	  = bswap_32(event->mmap.pid);
601 	event->mmap.tid	  = bswap_32(event->mmap.tid);
602 	event->mmap.start = bswap_64(event->mmap.start);
603 	event->mmap.len	  = bswap_64(event->mmap.len);
604 	event->mmap.pgoff = bswap_64(event->mmap.pgoff);
605 
606 	if (sample_id_all) {
607 		void *data = &event->mmap.filename;
608 
609 		data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
610 		swap_sample_id_all(event, data);
611 	}
612 }
613 
614 static void perf_event__mmap2_swap(union perf_event *event,
615 				  bool sample_id_all)
616 {
617 	event->mmap2.pid   = bswap_32(event->mmap2.pid);
618 	event->mmap2.tid   = bswap_32(event->mmap2.tid);
619 	event->mmap2.start = bswap_64(event->mmap2.start);
620 	event->mmap2.len   = bswap_64(event->mmap2.len);
621 	event->mmap2.pgoff = bswap_64(event->mmap2.pgoff);
622 
623 	if (!(event->header.misc & PERF_RECORD_MISC_MMAP_BUILD_ID)) {
624 		event->mmap2.maj   = bswap_32(event->mmap2.maj);
625 		event->mmap2.min   = bswap_32(event->mmap2.min);
626 		event->mmap2.ino   = bswap_64(event->mmap2.ino);
627 		event->mmap2.ino_generation = bswap_64(event->mmap2.ino_generation);
628 	}
629 
630 	if (sample_id_all) {
631 		void *data = &event->mmap2.filename;
632 
633 		data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
634 		swap_sample_id_all(event, data);
635 	}
636 }
637 static void perf_event__task_swap(union perf_event *event, bool sample_id_all)
638 {
639 	event->fork.pid	 = bswap_32(event->fork.pid);
640 	event->fork.tid	 = bswap_32(event->fork.tid);
641 	event->fork.ppid = bswap_32(event->fork.ppid);
642 	event->fork.ptid = bswap_32(event->fork.ptid);
643 	event->fork.time = bswap_64(event->fork.time);
644 
645 	if (sample_id_all)
646 		swap_sample_id_all(event, &event->fork + 1);
647 }
648 
649 static void perf_event__read_swap(union perf_event *event, bool sample_id_all)
650 {
651 	event->read.pid		 = bswap_32(event->read.pid);
652 	event->read.tid		 = bswap_32(event->read.tid);
653 	event->read.value	 = bswap_64(event->read.value);
654 	event->read.time_enabled = bswap_64(event->read.time_enabled);
655 	event->read.time_running = bswap_64(event->read.time_running);
656 	event->read.id		 = bswap_64(event->read.id);
657 
658 	if (sample_id_all)
659 		swap_sample_id_all(event, &event->read + 1);
660 }
661 
662 static void perf_event__aux_swap(union perf_event *event, bool sample_id_all)
663 {
664 	event->aux.aux_offset = bswap_64(event->aux.aux_offset);
665 	event->aux.aux_size   = bswap_64(event->aux.aux_size);
666 	event->aux.flags      = bswap_64(event->aux.flags);
667 
668 	if (sample_id_all)
669 		swap_sample_id_all(event, &event->aux + 1);
670 }
671 
672 static void perf_event__itrace_start_swap(union perf_event *event,
673 					  bool sample_id_all)
674 {
675 	event->itrace_start.pid	 = bswap_32(event->itrace_start.pid);
676 	event->itrace_start.tid	 = bswap_32(event->itrace_start.tid);
677 
678 	if (sample_id_all)
679 		swap_sample_id_all(event, &event->itrace_start + 1);
680 }
681 
682 static void perf_event__switch_swap(union perf_event *event, bool sample_id_all)
683 {
684 	if (event->header.type == PERF_RECORD_SWITCH_CPU_WIDE) {
685 		event->context_switch.next_prev_pid =
686 				bswap_32(event->context_switch.next_prev_pid);
687 		event->context_switch.next_prev_tid =
688 				bswap_32(event->context_switch.next_prev_tid);
689 	}
690 
691 	if (sample_id_all)
692 		swap_sample_id_all(event, &event->context_switch + 1);
693 }
694 
695 static void perf_event__text_poke_swap(union perf_event *event, bool sample_id_all)
696 {
697 	event->text_poke.addr    = bswap_64(event->text_poke.addr);
698 	event->text_poke.old_len = bswap_16(event->text_poke.old_len);
699 	event->text_poke.new_len = bswap_16(event->text_poke.new_len);
700 
701 	if (sample_id_all) {
702 		size_t len = sizeof(event->text_poke.old_len) +
703 			     sizeof(event->text_poke.new_len) +
704 			     event->text_poke.old_len +
705 			     event->text_poke.new_len;
706 		void *data = &event->text_poke.old_len;
707 
708 		data += PERF_ALIGN(len, sizeof(u64));
709 		swap_sample_id_all(event, data);
710 	}
711 }
712 
713 static void perf_event__throttle_swap(union perf_event *event,
714 				      bool sample_id_all)
715 {
716 	event->throttle.time	  = bswap_64(event->throttle.time);
717 	event->throttle.id	  = bswap_64(event->throttle.id);
718 	event->throttle.stream_id = bswap_64(event->throttle.stream_id);
719 
720 	if (sample_id_all)
721 		swap_sample_id_all(event, &event->throttle + 1);
722 }
723 
724 static void perf_event__namespaces_swap(union perf_event *event,
725 					bool sample_id_all)
726 {
727 	u64 i;
728 
729 	event->namespaces.pid		= bswap_32(event->namespaces.pid);
730 	event->namespaces.tid		= bswap_32(event->namespaces.tid);
731 	event->namespaces.nr_namespaces	= bswap_64(event->namespaces.nr_namespaces);
732 
733 	for (i = 0; i < event->namespaces.nr_namespaces; i++) {
734 		struct perf_ns_link_info *ns = &event->namespaces.link_info[i];
735 
736 		ns->dev = bswap_64(ns->dev);
737 		ns->ino = bswap_64(ns->ino);
738 	}
739 
740 	if (sample_id_all)
741 		swap_sample_id_all(event, &event->namespaces.link_info[i]);
742 }
743 
744 static void perf_event__cgroup_swap(union perf_event *event, bool sample_id_all)
745 {
746 	event->cgroup.id = bswap_64(event->cgroup.id);
747 
748 	if (sample_id_all) {
749 		void *data = &event->cgroup.path;
750 
751 		data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
752 		swap_sample_id_all(event, data);
753 	}
754 }
755 
756 static u8 revbyte(u8 b)
757 {
758 	int rev = (b >> 4) | ((b & 0xf) << 4);
759 	rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2);
760 	rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1);
761 	return (u8) rev;
762 }
763 
764 /*
765  * XXX this is hack in attempt to carry flags bitfield
766  * through endian village. ABI says:
767  *
768  * Bit-fields are allocated from right to left (least to most significant)
769  * on little-endian implementations and from left to right (most to least
770  * significant) on big-endian implementations.
771  *
772  * The above seems to be byte specific, so we need to reverse each
773  * byte of the bitfield. 'Internet' also says this might be implementation
774  * specific and we probably need proper fix and carry perf_event_attr
775  * bitfield flags in separate data file FEAT_ section. Thought this seems
776  * to work for now.
777  */
778 static void swap_bitfield(u8 *p, unsigned len)
779 {
780 	unsigned i;
781 
782 	for (i = 0; i < len; i++) {
783 		*p = revbyte(*p);
784 		p++;
785 	}
786 }
787 
788 /* exported for swapping attributes in file header */
789 void perf_event__attr_swap(struct perf_event_attr *attr)
790 {
791 	attr->type		= bswap_32(attr->type);
792 	attr->size		= bswap_32(attr->size);
793 
794 #define bswap_safe(f, n) 					\
795 	(attr->size > (offsetof(struct perf_event_attr, f) + 	\
796 		       sizeof(attr->f) * (n)))
797 #define bswap_field(f, sz) 			\
798 do { 						\
799 	if (bswap_safe(f, 0))			\
800 		attr->f = bswap_##sz(attr->f);	\
801 } while(0)
802 #define bswap_field_16(f) bswap_field(f, 16)
803 #define bswap_field_32(f) bswap_field(f, 32)
804 #define bswap_field_64(f) bswap_field(f, 64)
805 
806 	bswap_field_64(config);
807 	bswap_field_64(sample_period);
808 	bswap_field_64(sample_type);
809 	bswap_field_64(read_format);
810 	bswap_field_32(wakeup_events);
811 	bswap_field_32(bp_type);
812 	bswap_field_64(bp_addr);
813 	bswap_field_64(bp_len);
814 	bswap_field_64(branch_sample_type);
815 	bswap_field_64(sample_regs_user);
816 	bswap_field_32(sample_stack_user);
817 	bswap_field_32(aux_watermark);
818 	bswap_field_16(sample_max_stack);
819 	bswap_field_32(aux_sample_size);
820 
821 	/*
822 	 * After read_format are bitfields. Check read_format because
823 	 * we are unable to use offsetof on bitfield.
824 	 */
825 	if (bswap_safe(read_format, 1))
826 		swap_bitfield((u8 *) (&attr->read_format + 1),
827 			      sizeof(u64));
828 #undef bswap_field_64
829 #undef bswap_field_32
830 #undef bswap_field
831 #undef bswap_safe
832 }
833 
834 static void perf_event__hdr_attr_swap(union perf_event *event,
835 				      bool sample_id_all __maybe_unused)
836 {
837 	size_t size;
838 
839 	perf_event__attr_swap(&event->attr.attr);
840 
841 	size = event->header.size;
842 	size -= (void *)&event->attr.id - (void *)event;
843 	mem_bswap_64(event->attr.id, size);
844 }
845 
846 static void perf_event__event_update_swap(union perf_event *event,
847 					  bool sample_id_all __maybe_unused)
848 {
849 	event->event_update.type = bswap_64(event->event_update.type);
850 	event->event_update.id   = bswap_64(event->event_update.id);
851 }
852 
853 static void perf_event__event_type_swap(union perf_event *event,
854 					bool sample_id_all __maybe_unused)
855 {
856 	event->event_type.event_type.event_id =
857 		bswap_64(event->event_type.event_type.event_id);
858 }
859 
860 static void perf_event__tracing_data_swap(union perf_event *event,
861 					  bool sample_id_all __maybe_unused)
862 {
863 	event->tracing_data.size = bswap_32(event->tracing_data.size);
864 }
865 
866 static void perf_event__auxtrace_info_swap(union perf_event *event,
867 					   bool sample_id_all __maybe_unused)
868 {
869 	size_t size;
870 
871 	event->auxtrace_info.type = bswap_32(event->auxtrace_info.type);
872 
873 	size = event->header.size;
874 	size -= (void *)&event->auxtrace_info.priv - (void *)event;
875 	mem_bswap_64(event->auxtrace_info.priv, size);
876 }
877 
878 static void perf_event__auxtrace_swap(union perf_event *event,
879 				      bool sample_id_all __maybe_unused)
880 {
881 	event->auxtrace.size      = bswap_64(event->auxtrace.size);
882 	event->auxtrace.offset    = bswap_64(event->auxtrace.offset);
883 	event->auxtrace.reference = bswap_64(event->auxtrace.reference);
884 	event->auxtrace.idx       = bswap_32(event->auxtrace.idx);
885 	event->auxtrace.tid       = bswap_32(event->auxtrace.tid);
886 	event->auxtrace.cpu       = bswap_32(event->auxtrace.cpu);
887 }
888 
889 static void perf_event__auxtrace_error_swap(union perf_event *event,
890 					    bool sample_id_all __maybe_unused)
891 {
892 	event->auxtrace_error.type = bswap_32(event->auxtrace_error.type);
893 	event->auxtrace_error.code = bswap_32(event->auxtrace_error.code);
894 	event->auxtrace_error.cpu  = bswap_32(event->auxtrace_error.cpu);
895 	event->auxtrace_error.pid  = bswap_32(event->auxtrace_error.pid);
896 	event->auxtrace_error.tid  = bswap_32(event->auxtrace_error.tid);
897 	event->auxtrace_error.fmt  = bswap_32(event->auxtrace_error.fmt);
898 	event->auxtrace_error.ip   = bswap_64(event->auxtrace_error.ip);
899 	if (event->auxtrace_error.fmt)
900 		event->auxtrace_error.time = bswap_64(event->auxtrace_error.time);
901 	if (event->auxtrace_error.fmt >= 2) {
902 		event->auxtrace_error.machine_pid = bswap_32(event->auxtrace_error.machine_pid);
903 		event->auxtrace_error.vcpu = bswap_32(event->auxtrace_error.vcpu);
904 	}
905 }
906 
907 static void perf_event__thread_map_swap(union perf_event *event,
908 					bool sample_id_all __maybe_unused)
909 {
910 	unsigned i;
911 
912 	event->thread_map.nr = bswap_64(event->thread_map.nr);
913 
914 	for (i = 0; i < event->thread_map.nr; i++)
915 		event->thread_map.entries[i].pid = bswap_64(event->thread_map.entries[i].pid);
916 }
917 
918 static void perf_event__cpu_map_swap(union perf_event *event,
919 				     bool sample_id_all __maybe_unused)
920 {
921 	struct perf_record_cpu_map_data *data = &event->cpu_map.data;
922 
923 	data->type = bswap_16(data->type);
924 
925 	switch (data->type) {
926 	case PERF_CPU_MAP__CPUS:
927 		data->cpus_data.nr = bswap_16(data->cpus_data.nr);
928 
929 		for (unsigned i = 0; i < data->cpus_data.nr; i++)
930 			data->cpus_data.cpu[i] = bswap_16(data->cpus_data.cpu[i]);
931 		break;
932 	case PERF_CPU_MAP__MASK:
933 		data->mask32_data.long_size = bswap_16(data->mask32_data.long_size);
934 
935 		switch (data->mask32_data.long_size) {
936 		case 4:
937 			data->mask32_data.nr = bswap_16(data->mask32_data.nr);
938 			for (unsigned i = 0; i < data->mask32_data.nr; i++)
939 				data->mask32_data.mask[i] = bswap_32(data->mask32_data.mask[i]);
940 			break;
941 		case 8:
942 			data->mask64_data.nr = bswap_16(data->mask64_data.nr);
943 			for (unsigned i = 0; i < data->mask64_data.nr; i++)
944 				data->mask64_data.mask[i] = bswap_64(data->mask64_data.mask[i]);
945 			break;
946 		default:
947 			pr_err("cpu_map swap: unsupported long size\n");
948 		}
949 		break;
950 	case PERF_CPU_MAP__RANGE_CPUS:
951 		data->range_cpu_data.start_cpu = bswap_16(data->range_cpu_data.start_cpu);
952 		data->range_cpu_data.end_cpu = bswap_16(data->range_cpu_data.end_cpu);
953 		break;
954 	default:
955 		break;
956 	}
957 }
958 
959 static void perf_event__stat_config_swap(union perf_event *event,
960 					 bool sample_id_all __maybe_unused)
961 {
962 	u64 size;
963 
964 	size  = bswap_64(event->stat_config.nr) * sizeof(event->stat_config.data[0]);
965 	size += 1; /* nr item itself */
966 	mem_bswap_64(&event->stat_config.nr, size);
967 }
968 
969 static void perf_event__stat_swap(union perf_event *event,
970 				  bool sample_id_all __maybe_unused)
971 {
972 	event->stat.id     = bswap_64(event->stat.id);
973 	event->stat.thread = bswap_32(event->stat.thread);
974 	event->stat.cpu    = bswap_32(event->stat.cpu);
975 	event->stat.val    = bswap_64(event->stat.val);
976 	event->stat.ena    = bswap_64(event->stat.ena);
977 	event->stat.run    = bswap_64(event->stat.run);
978 }
979 
980 static void perf_event__stat_round_swap(union perf_event *event,
981 					bool sample_id_all __maybe_unused)
982 {
983 	event->stat_round.type = bswap_64(event->stat_round.type);
984 	event->stat_round.time = bswap_64(event->stat_round.time);
985 }
986 
987 static void perf_event__time_conv_swap(union perf_event *event,
988 				       bool sample_id_all __maybe_unused)
989 {
990 	event->time_conv.time_shift = bswap_64(event->time_conv.time_shift);
991 	event->time_conv.time_mult  = bswap_64(event->time_conv.time_mult);
992 	event->time_conv.time_zero  = bswap_64(event->time_conv.time_zero);
993 
994 	if (event_contains(event->time_conv, time_cycles)) {
995 		event->time_conv.time_cycles = bswap_64(event->time_conv.time_cycles);
996 		event->time_conv.time_mask = bswap_64(event->time_conv.time_mask);
997 	}
998 }
999 
1000 typedef void (*perf_event__swap_op)(union perf_event *event,
1001 				    bool sample_id_all);
1002 
1003 static perf_event__swap_op perf_event__swap_ops[] = {
1004 	[PERF_RECORD_MMAP]		  = perf_event__mmap_swap,
1005 	[PERF_RECORD_MMAP2]		  = perf_event__mmap2_swap,
1006 	[PERF_RECORD_COMM]		  = perf_event__comm_swap,
1007 	[PERF_RECORD_FORK]		  = perf_event__task_swap,
1008 	[PERF_RECORD_EXIT]		  = perf_event__task_swap,
1009 	[PERF_RECORD_LOST]		  = perf_event__all64_swap,
1010 	[PERF_RECORD_READ]		  = perf_event__read_swap,
1011 	[PERF_RECORD_THROTTLE]		  = perf_event__throttle_swap,
1012 	[PERF_RECORD_UNTHROTTLE]	  = perf_event__throttle_swap,
1013 	[PERF_RECORD_SAMPLE]		  = perf_event__all64_swap,
1014 	[PERF_RECORD_AUX]		  = perf_event__aux_swap,
1015 	[PERF_RECORD_ITRACE_START]	  = perf_event__itrace_start_swap,
1016 	[PERF_RECORD_LOST_SAMPLES]	  = perf_event__all64_swap,
1017 	[PERF_RECORD_SWITCH]		  = perf_event__switch_swap,
1018 	[PERF_RECORD_SWITCH_CPU_WIDE]	  = perf_event__switch_swap,
1019 	[PERF_RECORD_NAMESPACES]	  = perf_event__namespaces_swap,
1020 	[PERF_RECORD_CGROUP]		  = perf_event__cgroup_swap,
1021 	[PERF_RECORD_TEXT_POKE]		  = perf_event__text_poke_swap,
1022 	[PERF_RECORD_AUX_OUTPUT_HW_ID]	  = perf_event__all64_swap,
1023 	[PERF_RECORD_HEADER_ATTR]	  = perf_event__hdr_attr_swap,
1024 	[PERF_RECORD_HEADER_EVENT_TYPE]	  = perf_event__event_type_swap,
1025 	[PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
1026 	[PERF_RECORD_HEADER_BUILD_ID]	  = NULL,
1027 	[PERF_RECORD_ID_INDEX]		  = perf_event__all64_swap,
1028 	[PERF_RECORD_AUXTRACE_INFO]	  = perf_event__auxtrace_info_swap,
1029 	[PERF_RECORD_AUXTRACE]		  = perf_event__auxtrace_swap,
1030 	[PERF_RECORD_AUXTRACE_ERROR]	  = perf_event__auxtrace_error_swap,
1031 	[PERF_RECORD_THREAD_MAP]	  = perf_event__thread_map_swap,
1032 	[PERF_RECORD_CPU_MAP]		  = perf_event__cpu_map_swap,
1033 	[PERF_RECORD_STAT_CONFIG]	  = perf_event__stat_config_swap,
1034 	[PERF_RECORD_STAT]		  = perf_event__stat_swap,
1035 	[PERF_RECORD_STAT_ROUND]	  = perf_event__stat_round_swap,
1036 	[PERF_RECORD_EVENT_UPDATE]	  = perf_event__event_update_swap,
1037 	[PERF_RECORD_TIME_CONV]		  = perf_event__time_conv_swap,
1038 	[PERF_RECORD_HEADER_MAX]	  = NULL,
1039 };
1040 
1041 /*
1042  * When perf record finishes a pass on every buffers, it records this pseudo
1043  * event.
1044  * We record the max timestamp t found in the pass n.
1045  * Assuming these timestamps are monotonic across cpus, we know that if
1046  * a buffer still has events with timestamps below t, they will be all
1047  * available and then read in the pass n + 1.
1048  * Hence when we start to read the pass n + 2, we can safely flush every
1049  * events with timestamps below t.
1050  *
1051  *    ============ PASS n =================
1052  *       CPU 0         |   CPU 1
1053  *                     |
1054  *    cnt1 timestamps  |   cnt2 timestamps
1055  *          1          |         2
1056  *          2          |         3
1057  *          -          |         4  <--- max recorded
1058  *
1059  *    ============ PASS n + 1 ==============
1060  *       CPU 0         |   CPU 1
1061  *                     |
1062  *    cnt1 timestamps  |   cnt2 timestamps
1063  *          3          |         5
1064  *          4          |         6
1065  *          5          |         7 <---- max recorded
1066  *
1067  *      Flush every events below timestamp 4
1068  *
1069  *    ============ PASS n + 2 ==============
1070  *       CPU 0         |   CPU 1
1071  *                     |
1072  *    cnt1 timestamps  |   cnt2 timestamps
1073  *          6          |         8
1074  *          7          |         9
1075  *          -          |         10
1076  *
1077  *      Flush every events below timestamp 7
1078  *      etc...
1079  */
1080 int perf_event__process_finished_round(struct perf_tool *tool __maybe_unused,
1081 				       union perf_event *event __maybe_unused,
1082 				       struct ordered_events *oe)
1083 {
1084 	if (dump_trace)
1085 		fprintf(stdout, "\n");
1086 	return ordered_events__flush(oe, OE_FLUSH__ROUND);
1087 }
1088 
1089 int perf_session__queue_event(struct perf_session *s, union perf_event *event,
1090 			      u64 timestamp, u64 file_offset, const char *file_path)
1091 {
1092 	return ordered_events__queue(&s->ordered_events, event, timestamp, file_offset, file_path);
1093 }
1094 
1095 static void callchain__lbr_callstack_printf(struct perf_sample *sample)
1096 {
1097 	struct ip_callchain *callchain = sample->callchain;
1098 	struct branch_stack *lbr_stack = sample->branch_stack;
1099 	struct branch_entry *entries = perf_sample__branch_entries(sample);
1100 	u64 kernel_callchain_nr = callchain->nr;
1101 	unsigned int i;
1102 
1103 	for (i = 0; i < kernel_callchain_nr; i++) {
1104 		if (callchain->ips[i] == PERF_CONTEXT_USER)
1105 			break;
1106 	}
1107 
1108 	if ((i != kernel_callchain_nr) && lbr_stack->nr) {
1109 		u64 total_nr;
1110 		/*
1111 		 * LBR callstack can only get user call chain,
1112 		 * i is kernel call chain number,
1113 		 * 1 is PERF_CONTEXT_USER.
1114 		 *
1115 		 * The user call chain is stored in LBR registers.
1116 		 * LBR are pair registers. The caller is stored
1117 		 * in "from" register, while the callee is stored
1118 		 * in "to" register.
1119 		 * For example, there is a call stack
1120 		 * "A"->"B"->"C"->"D".
1121 		 * The LBR registers will be recorded like
1122 		 * "C"->"D", "B"->"C", "A"->"B".
1123 		 * So only the first "to" register and all "from"
1124 		 * registers are needed to construct the whole stack.
1125 		 */
1126 		total_nr = i + 1 + lbr_stack->nr + 1;
1127 		kernel_callchain_nr = i + 1;
1128 
1129 		printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr);
1130 
1131 		for (i = 0; i < kernel_callchain_nr; i++)
1132 			printf("..... %2d: %016" PRIx64 "\n",
1133 			       i, callchain->ips[i]);
1134 
1135 		printf("..... %2d: %016" PRIx64 "\n",
1136 		       (int)(kernel_callchain_nr), entries[0].to);
1137 		for (i = 0; i < lbr_stack->nr; i++)
1138 			printf("..... %2d: %016" PRIx64 "\n",
1139 			       (int)(i + kernel_callchain_nr + 1), entries[i].from);
1140 	}
1141 }
1142 
1143 static void callchain__printf(struct evsel *evsel,
1144 			      struct perf_sample *sample)
1145 {
1146 	unsigned int i;
1147 	struct ip_callchain *callchain = sample->callchain;
1148 
1149 	if (evsel__has_branch_callstack(evsel))
1150 		callchain__lbr_callstack_printf(sample);
1151 
1152 	printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr);
1153 
1154 	for (i = 0; i < callchain->nr; i++)
1155 		printf("..... %2d: %016" PRIx64 "\n",
1156 		       i, callchain->ips[i]);
1157 }
1158 
1159 static void branch_stack__printf(struct perf_sample *sample, bool callstack)
1160 {
1161 	struct branch_entry *entries = perf_sample__branch_entries(sample);
1162 	uint64_t i;
1163 
1164 	if (!callstack) {
1165 		printf("%s: nr:%" PRIu64 "\n", "... branch stack", sample->branch_stack->nr);
1166 	} else {
1167 		/* the reason of adding 1 to nr is because after expanding
1168 		 * branch stack it generates nr + 1 callstack records. e.g.,
1169 		 *         B()->C()
1170 		 *         A()->B()
1171 		 * the final callstack should be:
1172 		 *         C()
1173 		 *         B()
1174 		 *         A()
1175 		 */
1176 		printf("%s: nr:%" PRIu64 "\n", "... branch callstack", sample->branch_stack->nr+1);
1177 	}
1178 
1179 	for (i = 0; i < sample->branch_stack->nr; i++) {
1180 		struct branch_entry *e = &entries[i];
1181 
1182 		if (!callstack) {
1183 			printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x %s %s\n",
1184 				i, e->from, e->to,
1185 				(unsigned short)e->flags.cycles,
1186 				e->flags.mispred ? "M" : " ",
1187 				e->flags.predicted ? "P" : " ",
1188 				e->flags.abort ? "A" : " ",
1189 				e->flags.in_tx ? "T" : " ",
1190 				(unsigned)e->flags.reserved,
1191 				get_branch_type(e),
1192 				e->flags.spec ? branch_spec_desc(e->flags.spec) : "");
1193 		} else {
1194 			if (i == 0) {
1195 				printf("..... %2"PRIu64": %016" PRIx64 "\n"
1196 				       "..... %2"PRIu64": %016" PRIx64 "\n",
1197 						i, e->to, i+1, e->from);
1198 			} else {
1199 				printf("..... %2"PRIu64": %016" PRIx64 "\n", i+1, e->from);
1200 			}
1201 		}
1202 	}
1203 }
1204 
1205 static void regs_dump__printf(u64 mask, u64 *regs, const char *arch)
1206 {
1207 	unsigned rid, i = 0;
1208 
1209 	for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) {
1210 		u64 val = regs[i++];
1211 
1212 		printf(".... %-5s 0x%016" PRIx64 "\n",
1213 		       perf_reg_name(rid, arch), val);
1214 	}
1215 }
1216 
1217 static const char *regs_abi[] = {
1218 	[PERF_SAMPLE_REGS_ABI_NONE] = "none",
1219 	[PERF_SAMPLE_REGS_ABI_32] = "32-bit",
1220 	[PERF_SAMPLE_REGS_ABI_64] = "64-bit",
1221 };
1222 
1223 static inline const char *regs_dump_abi(struct regs_dump *d)
1224 {
1225 	if (d->abi > PERF_SAMPLE_REGS_ABI_64)
1226 		return "unknown";
1227 
1228 	return regs_abi[d->abi];
1229 }
1230 
1231 static void regs__printf(const char *type, struct regs_dump *regs, const char *arch)
1232 {
1233 	u64 mask = regs->mask;
1234 
1235 	printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n",
1236 	       type,
1237 	       mask,
1238 	       regs_dump_abi(regs));
1239 
1240 	regs_dump__printf(mask, regs->regs, arch);
1241 }
1242 
1243 static void regs_user__printf(struct perf_sample *sample, const char *arch)
1244 {
1245 	struct regs_dump *user_regs = &sample->user_regs;
1246 
1247 	if (user_regs->regs)
1248 		regs__printf("user", user_regs, arch);
1249 }
1250 
1251 static void regs_intr__printf(struct perf_sample *sample, const char *arch)
1252 {
1253 	struct regs_dump *intr_regs = &sample->intr_regs;
1254 
1255 	if (intr_regs->regs)
1256 		regs__printf("intr", intr_regs, arch);
1257 }
1258 
1259 static void stack_user__printf(struct stack_dump *dump)
1260 {
1261 	printf("... ustack: size %" PRIu64 ", offset 0x%x\n",
1262 	       dump->size, dump->offset);
1263 }
1264 
1265 static void evlist__print_tstamp(struct evlist *evlist, union perf_event *event, struct perf_sample *sample)
1266 {
1267 	u64 sample_type = __evlist__combined_sample_type(evlist);
1268 
1269 	if (event->header.type != PERF_RECORD_SAMPLE &&
1270 	    !evlist__sample_id_all(evlist)) {
1271 		fputs("-1 -1 ", stdout);
1272 		return;
1273 	}
1274 
1275 	if ((sample_type & PERF_SAMPLE_CPU))
1276 		printf("%u ", sample->cpu);
1277 
1278 	if (sample_type & PERF_SAMPLE_TIME)
1279 		printf("%" PRIu64 " ", sample->time);
1280 }
1281 
1282 static void sample_read__printf(struct perf_sample *sample, u64 read_format)
1283 {
1284 	printf("... sample_read:\n");
1285 
1286 	if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1287 		printf("...... time enabled %016" PRIx64 "\n",
1288 		       sample->read.time_enabled);
1289 
1290 	if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1291 		printf("...... time running %016" PRIx64 "\n",
1292 		       sample->read.time_running);
1293 
1294 	if (read_format & PERF_FORMAT_GROUP) {
1295 		struct sample_read_value *value = sample->read.group.values;
1296 
1297 		printf(".... group nr %" PRIu64 "\n", sample->read.group.nr);
1298 
1299 		sample_read_group__for_each(value, sample->read.group.nr, read_format) {
1300 			printf("..... id %016" PRIx64
1301 			       ", value %016" PRIx64,
1302 			       value->id, value->value);
1303 			if (read_format & PERF_FORMAT_LOST)
1304 				printf(", lost %" PRIu64, value->lost);
1305 			printf("\n");
1306 		}
1307 	} else {
1308 		printf("..... id %016" PRIx64 ", value %016" PRIx64,
1309 			sample->read.one.id, sample->read.one.value);
1310 		if (read_format & PERF_FORMAT_LOST)
1311 			printf(", lost %" PRIu64, sample->read.one.lost);
1312 		printf("\n");
1313 	}
1314 }
1315 
1316 static void dump_event(struct evlist *evlist, union perf_event *event,
1317 		       u64 file_offset, struct perf_sample *sample,
1318 		       const char *file_path)
1319 {
1320 	if (!dump_trace)
1321 		return;
1322 
1323 	printf("\n%#" PRIx64 "@%s [%#x]: event: %d\n",
1324 	       file_offset, file_path, event->header.size, event->header.type);
1325 
1326 	trace_event(event);
1327 	if (event->header.type == PERF_RECORD_SAMPLE && evlist->trace_event_sample_raw)
1328 		evlist->trace_event_sample_raw(evlist, event, sample);
1329 
1330 	if (sample)
1331 		evlist__print_tstamp(evlist, event, sample);
1332 
1333 	printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset,
1334 	       event->header.size, perf_event__name(event->header.type));
1335 }
1336 
1337 char *get_page_size_name(u64 size, char *str)
1338 {
1339 	if (!size || !unit_number__scnprintf(str, PAGE_SIZE_NAME_LEN, size))
1340 		snprintf(str, PAGE_SIZE_NAME_LEN, "%s", "N/A");
1341 
1342 	return str;
1343 }
1344 
1345 static void dump_sample(struct evsel *evsel, union perf_event *event,
1346 			struct perf_sample *sample, const char *arch)
1347 {
1348 	u64 sample_type;
1349 	char str[PAGE_SIZE_NAME_LEN];
1350 
1351 	if (!dump_trace)
1352 		return;
1353 
1354 	printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n",
1355 	       event->header.misc, sample->pid, sample->tid, sample->ip,
1356 	       sample->period, sample->addr);
1357 
1358 	sample_type = evsel->core.attr.sample_type;
1359 
1360 	if (evsel__has_callchain(evsel))
1361 		callchain__printf(evsel, sample);
1362 
1363 	if (evsel__has_br_stack(evsel))
1364 		branch_stack__printf(sample, evsel__has_branch_callstack(evsel));
1365 
1366 	if (sample_type & PERF_SAMPLE_REGS_USER)
1367 		regs_user__printf(sample, arch);
1368 
1369 	if (sample_type & PERF_SAMPLE_REGS_INTR)
1370 		regs_intr__printf(sample, arch);
1371 
1372 	if (sample_type & PERF_SAMPLE_STACK_USER)
1373 		stack_user__printf(&sample->user_stack);
1374 
1375 	if (sample_type & PERF_SAMPLE_WEIGHT_TYPE) {
1376 		printf("... weight: %" PRIu64 "", sample->weight);
1377 			if (sample_type & PERF_SAMPLE_WEIGHT_STRUCT) {
1378 				printf(",0x%"PRIx16"", sample->ins_lat);
1379 				printf(",0x%"PRIx16"", sample->p_stage_cyc);
1380 			}
1381 		printf("\n");
1382 	}
1383 
1384 	if (sample_type & PERF_SAMPLE_DATA_SRC)
1385 		printf(" . data_src: 0x%"PRIx64"\n", sample->data_src);
1386 
1387 	if (sample_type & PERF_SAMPLE_PHYS_ADDR)
1388 		printf(" .. phys_addr: 0x%"PRIx64"\n", sample->phys_addr);
1389 
1390 	if (sample_type & PERF_SAMPLE_DATA_PAGE_SIZE)
1391 		printf(" .. data page size: %s\n", get_page_size_name(sample->data_page_size, str));
1392 
1393 	if (sample_type & PERF_SAMPLE_CODE_PAGE_SIZE)
1394 		printf(" .. code page size: %s\n", get_page_size_name(sample->code_page_size, str));
1395 
1396 	if (sample_type & PERF_SAMPLE_TRANSACTION)
1397 		printf("... transaction: %" PRIx64 "\n", sample->transaction);
1398 
1399 	if (sample_type & PERF_SAMPLE_READ)
1400 		sample_read__printf(sample, evsel->core.attr.read_format);
1401 }
1402 
1403 static void dump_read(struct evsel *evsel, union perf_event *event)
1404 {
1405 	struct perf_record_read *read_event = &event->read;
1406 	u64 read_format;
1407 
1408 	if (!dump_trace)
1409 		return;
1410 
1411 	printf(": %d %d %s %" PRI_lu64 "\n", event->read.pid, event->read.tid,
1412 	       evsel__name(evsel), event->read.value);
1413 
1414 	if (!evsel)
1415 		return;
1416 
1417 	read_format = evsel->core.attr.read_format;
1418 
1419 	if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1420 		printf("... time enabled : %" PRI_lu64 "\n", read_event->time_enabled);
1421 
1422 	if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1423 		printf("... time running : %" PRI_lu64 "\n", read_event->time_running);
1424 
1425 	if (read_format & PERF_FORMAT_ID)
1426 		printf("... id           : %" PRI_lu64 "\n", read_event->id);
1427 
1428 	if (read_format & PERF_FORMAT_LOST)
1429 		printf("... lost         : %" PRI_lu64 "\n", read_event->lost);
1430 }
1431 
1432 static struct machine *machines__find_for_cpumode(struct machines *machines,
1433 					       union perf_event *event,
1434 					       struct perf_sample *sample)
1435 {
1436 	if (perf_guest &&
1437 	    ((sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL) ||
1438 	     (sample->cpumode == PERF_RECORD_MISC_GUEST_USER))) {
1439 		u32 pid;
1440 
1441 		if (sample->machine_pid)
1442 			pid = sample->machine_pid;
1443 		else if (event->header.type == PERF_RECORD_MMAP
1444 		    || event->header.type == PERF_RECORD_MMAP2)
1445 			pid = event->mmap.pid;
1446 		else
1447 			pid = sample->pid;
1448 
1449 		/*
1450 		 * Guest code machine is created as needed and does not use
1451 		 * DEFAULT_GUEST_KERNEL_ID.
1452 		 */
1453 		if (symbol_conf.guest_code)
1454 			return machines__findnew(machines, pid);
1455 
1456 		return machines__find_guest(machines, pid);
1457 	}
1458 
1459 	return &machines->host;
1460 }
1461 
1462 static int deliver_sample_value(struct evlist *evlist,
1463 				struct perf_tool *tool,
1464 				union perf_event *event,
1465 				struct perf_sample *sample,
1466 				struct sample_read_value *v,
1467 				struct machine *machine)
1468 {
1469 	struct perf_sample_id *sid = evlist__id2sid(evlist, v->id);
1470 	struct evsel *evsel;
1471 
1472 	if (sid) {
1473 		sample->id     = v->id;
1474 		sample->period = v->value - sid->period;
1475 		sid->period    = v->value;
1476 	}
1477 
1478 	if (!sid || sid->evsel == NULL) {
1479 		++evlist->stats.nr_unknown_id;
1480 		return 0;
1481 	}
1482 
1483 	/*
1484 	 * There's no reason to deliver sample
1485 	 * for zero period, bail out.
1486 	 */
1487 	if (!sample->period)
1488 		return 0;
1489 
1490 	evsel = container_of(sid->evsel, struct evsel, core);
1491 	return tool->sample(tool, event, sample, evsel, machine);
1492 }
1493 
1494 static int deliver_sample_group(struct evlist *evlist,
1495 				struct perf_tool *tool,
1496 				union  perf_event *event,
1497 				struct perf_sample *sample,
1498 				struct machine *machine,
1499 				u64 read_format)
1500 {
1501 	int ret = -EINVAL;
1502 	struct sample_read_value *v = sample->read.group.values;
1503 
1504 	sample_read_group__for_each(v, sample->read.group.nr, read_format) {
1505 		ret = deliver_sample_value(evlist, tool, event, sample, v,
1506 					   machine);
1507 		if (ret)
1508 			break;
1509 	}
1510 
1511 	return ret;
1512 }
1513 
1514 static int evlist__deliver_sample(struct evlist *evlist, struct perf_tool *tool,
1515 				  union  perf_event *event, struct perf_sample *sample,
1516 				  struct evsel *evsel, struct machine *machine)
1517 {
1518 	/* We know evsel != NULL. */
1519 	u64 sample_type = evsel->core.attr.sample_type;
1520 	u64 read_format = evsel->core.attr.read_format;
1521 
1522 	/* Standard sample delivery. */
1523 	if (!(sample_type & PERF_SAMPLE_READ))
1524 		return tool->sample(tool, event, sample, evsel, machine);
1525 
1526 	/* For PERF_SAMPLE_READ we have either single or group mode. */
1527 	if (read_format & PERF_FORMAT_GROUP)
1528 		return deliver_sample_group(evlist, tool, event, sample,
1529 					    machine, read_format);
1530 	else
1531 		return deliver_sample_value(evlist, tool, event, sample,
1532 					    &sample->read.one, machine);
1533 }
1534 
1535 static int machines__deliver_event(struct machines *machines,
1536 				   struct evlist *evlist,
1537 				   union perf_event *event,
1538 				   struct perf_sample *sample,
1539 				   struct perf_tool *tool, u64 file_offset,
1540 				   const char *file_path)
1541 {
1542 	struct evsel *evsel;
1543 	struct machine *machine;
1544 
1545 	dump_event(evlist, event, file_offset, sample, file_path);
1546 
1547 	evsel = evlist__id2evsel(evlist, sample->id);
1548 
1549 	machine = machines__find_for_cpumode(machines, event, sample);
1550 
1551 	switch (event->header.type) {
1552 	case PERF_RECORD_SAMPLE:
1553 		if (evsel == NULL) {
1554 			++evlist->stats.nr_unknown_id;
1555 			return 0;
1556 		}
1557 		if (machine == NULL) {
1558 			++evlist->stats.nr_unprocessable_samples;
1559 			dump_sample(evsel, event, sample, perf_env__arch(NULL));
1560 			return 0;
1561 		}
1562 		dump_sample(evsel, event, sample, perf_env__arch(machine->env));
1563 		return evlist__deliver_sample(evlist, tool, event, sample, evsel, machine);
1564 	case PERF_RECORD_MMAP:
1565 		return tool->mmap(tool, event, sample, machine);
1566 	case PERF_RECORD_MMAP2:
1567 		if (event->header.misc & PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT)
1568 			++evlist->stats.nr_proc_map_timeout;
1569 		return tool->mmap2(tool, event, sample, machine);
1570 	case PERF_RECORD_COMM:
1571 		return tool->comm(tool, event, sample, machine);
1572 	case PERF_RECORD_NAMESPACES:
1573 		return tool->namespaces(tool, event, sample, machine);
1574 	case PERF_RECORD_CGROUP:
1575 		return tool->cgroup(tool, event, sample, machine);
1576 	case PERF_RECORD_FORK:
1577 		return tool->fork(tool, event, sample, machine);
1578 	case PERF_RECORD_EXIT:
1579 		return tool->exit(tool, event, sample, machine);
1580 	case PERF_RECORD_LOST:
1581 		if (tool->lost == perf_event__process_lost)
1582 			evlist->stats.total_lost += event->lost.lost;
1583 		return tool->lost(tool, event, sample, machine);
1584 	case PERF_RECORD_LOST_SAMPLES:
1585 		if (tool->lost_samples == perf_event__process_lost_samples)
1586 			evlist->stats.total_lost_samples += event->lost_samples.lost;
1587 		return tool->lost_samples(tool, event, sample, machine);
1588 	case PERF_RECORD_READ:
1589 		dump_read(evsel, event);
1590 		return tool->read(tool, event, sample, evsel, machine);
1591 	case PERF_RECORD_THROTTLE:
1592 		return tool->throttle(tool, event, sample, machine);
1593 	case PERF_RECORD_UNTHROTTLE:
1594 		return tool->unthrottle(tool, event, sample, machine);
1595 	case PERF_RECORD_AUX:
1596 		if (tool->aux == perf_event__process_aux) {
1597 			if (event->aux.flags & PERF_AUX_FLAG_TRUNCATED)
1598 				evlist->stats.total_aux_lost += 1;
1599 			if (event->aux.flags & PERF_AUX_FLAG_PARTIAL)
1600 				evlist->stats.total_aux_partial += 1;
1601 			if (event->aux.flags & PERF_AUX_FLAG_COLLISION)
1602 				evlist->stats.total_aux_collision += 1;
1603 		}
1604 		return tool->aux(tool, event, sample, machine);
1605 	case PERF_RECORD_ITRACE_START:
1606 		return tool->itrace_start(tool, event, sample, machine);
1607 	case PERF_RECORD_SWITCH:
1608 	case PERF_RECORD_SWITCH_CPU_WIDE:
1609 		return tool->context_switch(tool, event, sample, machine);
1610 	case PERF_RECORD_KSYMBOL:
1611 		return tool->ksymbol(tool, event, sample, machine);
1612 	case PERF_RECORD_BPF_EVENT:
1613 		return tool->bpf(tool, event, sample, machine);
1614 	case PERF_RECORD_TEXT_POKE:
1615 		return tool->text_poke(tool, event, sample, machine);
1616 	case PERF_RECORD_AUX_OUTPUT_HW_ID:
1617 		return tool->aux_output_hw_id(tool, event, sample, machine);
1618 	default:
1619 		++evlist->stats.nr_unknown_events;
1620 		return -1;
1621 	}
1622 }
1623 
1624 static int perf_session__deliver_event(struct perf_session *session,
1625 				       union perf_event *event,
1626 				       struct perf_tool *tool,
1627 				       u64 file_offset,
1628 				       const char *file_path)
1629 {
1630 	struct perf_sample sample;
1631 	int ret = evlist__parse_sample(session->evlist, event, &sample);
1632 
1633 	if (ret) {
1634 		pr_err("Can't parse sample, err = %d\n", ret);
1635 		return ret;
1636 	}
1637 
1638 	ret = auxtrace__process_event(session, event, &sample, tool);
1639 	if (ret < 0)
1640 		return ret;
1641 	if (ret > 0)
1642 		return 0;
1643 
1644 	ret = machines__deliver_event(&session->machines, session->evlist,
1645 				      event, &sample, tool, file_offset, file_path);
1646 
1647 	if (dump_trace && sample.aux_sample.size)
1648 		auxtrace__dump_auxtrace_sample(session, &sample);
1649 
1650 	return ret;
1651 }
1652 
1653 static s64 perf_session__process_user_event(struct perf_session *session,
1654 					    union perf_event *event,
1655 					    u64 file_offset,
1656 					    const char *file_path)
1657 {
1658 	struct ordered_events *oe = &session->ordered_events;
1659 	struct perf_tool *tool = session->tool;
1660 	struct perf_sample sample = { .time = 0, };
1661 	int fd = perf_data__fd(session->data);
1662 	int err;
1663 
1664 	if (event->header.type != PERF_RECORD_COMPRESSED ||
1665 	    tool->compressed == perf_session__process_compressed_event_stub)
1666 		dump_event(session->evlist, event, file_offset, &sample, file_path);
1667 
1668 	/* These events are processed right away */
1669 	switch (event->header.type) {
1670 	case PERF_RECORD_HEADER_ATTR:
1671 		err = tool->attr(tool, event, &session->evlist);
1672 		if (err == 0) {
1673 			perf_session__set_id_hdr_size(session);
1674 			perf_session__set_comm_exec(session);
1675 		}
1676 		return err;
1677 	case PERF_RECORD_EVENT_UPDATE:
1678 		return tool->event_update(tool, event, &session->evlist);
1679 	case PERF_RECORD_HEADER_EVENT_TYPE:
1680 		/*
1681 		 * Deprecated, but we need to handle it for sake
1682 		 * of old data files create in pipe mode.
1683 		 */
1684 		return 0;
1685 	case PERF_RECORD_HEADER_TRACING_DATA:
1686 		/*
1687 		 * Setup for reading amidst mmap, but only when we
1688 		 * are in 'file' mode. The 'pipe' fd is in proper
1689 		 * place already.
1690 		 */
1691 		if (!perf_data__is_pipe(session->data))
1692 			lseek(fd, file_offset, SEEK_SET);
1693 		return tool->tracing_data(session, event);
1694 	case PERF_RECORD_HEADER_BUILD_ID:
1695 		return tool->build_id(session, event);
1696 	case PERF_RECORD_FINISHED_ROUND:
1697 		return tool->finished_round(tool, event, oe);
1698 	case PERF_RECORD_ID_INDEX:
1699 		return tool->id_index(session, event);
1700 	case PERF_RECORD_AUXTRACE_INFO:
1701 		return tool->auxtrace_info(session, event);
1702 	case PERF_RECORD_AUXTRACE:
1703 		/*
1704 		 * Setup for reading amidst mmap, but only when we
1705 		 * are in 'file' mode.  The 'pipe' fd is in proper
1706 		 * place already.
1707 		 */
1708 		if (!perf_data__is_pipe(session->data))
1709 			lseek(fd, file_offset + event->header.size, SEEK_SET);
1710 		return tool->auxtrace(session, event);
1711 	case PERF_RECORD_AUXTRACE_ERROR:
1712 		perf_session__auxtrace_error_inc(session, event);
1713 		return tool->auxtrace_error(session, event);
1714 	case PERF_RECORD_THREAD_MAP:
1715 		return tool->thread_map(session, event);
1716 	case PERF_RECORD_CPU_MAP:
1717 		return tool->cpu_map(session, event);
1718 	case PERF_RECORD_STAT_CONFIG:
1719 		return tool->stat_config(session, event);
1720 	case PERF_RECORD_STAT:
1721 		return tool->stat(session, event);
1722 	case PERF_RECORD_STAT_ROUND:
1723 		return tool->stat_round(session, event);
1724 	case PERF_RECORD_TIME_CONV:
1725 		session->time_conv = event->time_conv;
1726 		return tool->time_conv(session, event);
1727 	case PERF_RECORD_HEADER_FEATURE:
1728 		return tool->feature(session, event);
1729 	case PERF_RECORD_COMPRESSED:
1730 		err = tool->compressed(session, event, file_offset, file_path);
1731 		if (err)
1732 			dump_event(session->evlist, event, file_offset, &sample, file_path);
1733 		return err;
1734 	case PERF_RECORD_FINISHED_INIT:
1735 		return tool->finished_init(session, event);
1736 	default:
1737 		return -EINVAL;
1738 	}
1739 }
1740 
1741 int perf_session__deliver_synth_event(struct perf_session *session,
1742 				      union perf_event *event,
1743 				      struct perf_sample *sample)
1744 {
1745 	struct evlist *evlist = session->evlist;
1746 	struct perf_tool *tool = session->tool;
1747 
1748 	events_stats__inc(&evlist->stats, event->header.type);
1749 
1750 	if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1751 		return perf_session__process_user_event(session, event, 0, NULL);
1752 
1753 	return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0, NULL);
1754 }
1755 
1756 static void event_swap(union perf_event *event, bool sample_id_all)
1757 {
1758 	perf_event__swap_op swap;
1759 
1760 	swap = perf_event__swap_ops[event->header.type];
1761 	if (swap)
1762 		swap(event, sample_id_all);
1763 }
1764 
1765 int perf_session__peek_event(struct perf_session *session, off_t file_offset,
1766 			     void *buf, size_t buf_sz,
1767 			     union perf_event **event_ptr,
1768 			     struct perf_sample *sample)
1769 {
1770 	union perf_event *event;
1771 	size_t hdr_sz, rest;
1772 	int fd;
1773 
1774 	if (session->one_mmap && !session->header.needs_swap) {
1775 		event = file_offset - session->one_mmap_offset +
1776 			session->one_mmap_addr;
1777 		goto out_parse_sample;
1778 	}
1779 
1780 	if (perf_data__is_pipe(session->data))
1781 		return -1;
1782 
1783 	fd = perf_data__fd(session->data);
1784 	hdr_sz = sizeof(struct perf_event_header);
1785 
1786 	if (buf_sz < hdr_sz)
1787 		return -1;
1788 
1789 	if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 ||
1790 	    readn(fd, buf, hdr_sz) != (ssize_t)hdr_sz)
1791 		return -1;
1792 
1793 	event = (union perf_event *)buf;
1794 
1795 	if (session->header.needs_swap)
1796 		perf_event_header__bswap(&event->header);
1797 
1798 	if (event->header.size < hdr_sz || event->header.size > buf_sz)
1799 		return -1;
1800 
1801 	buf += hdr_sz;
1802 	rest = event->header.size - hdr_sz;
1803 
1804 	if (readn(fd, buf, rest) != (ssize_t)rest)
1805 		return -1;
1806 
1807 	if (session->header.needs_swap)
1808 		event_swap(event, evlist__sample_id_all(session->evlist));
1809 
1810 out_parse_sample:
1811 
1812 	if (sample && event->header.type < PERF_RECORD_USER_TYPE_START &&
1813 	    evlist__parse_sample(session->evlist, event, sample))
1814 		return -1;
1815 
1816 	*event_ptr = event;
1817 
1818 	return 0;
1819 }
1820 
1821 int perf_session__peek_events(struct perf_session *session, u64 offset,
1822 			      u64 size, peek_events_cb_t cb, void *data)
1823 {
1824 	u64 max_offset = offset + size;
1825 	char buf[PERF_SAMPLE_MAX_SIZE];
1826 	union perf_event *event;
1827 	int err;
1828 
1829 	do {
1830 		err = perf_session__peek_event(session, offset, buf,
1831 					       PERF_SAMPLE_MAX_SIZE, &event,
1832 					       NULL);
1833 		if (err)
1834 			return err;
1835 
1836 		err = cb(session, event, offset, data);
1837 		if (err)
1838 			return err;
1839 
1840 		offset += event->header.size;
1841 		if (event->header.type == PERF_RECORD_AUXTRACE)
1842 			offset += event->auxtrace.size;
1843 
1844 	} while (offset < max_offset);
1845 
1846 	return err;
1847 }
1848 
1849 static s64 perf_session__process_event(struct perf_session *session,
1850 				       union perf_event *event, u64 file_offset,
1851 				       const char *file_path)
1852 {
1853 	struct evlist *evlist = session->evlist;
1854 	struct perf_tool *tool = session->tool;
1855 	int ret;
1856 
1857 	if (session->header.needs_swap)
1858 		event_swap(event, evlist__sample_id_all(evlist));
1859 
1860 	if (event->header.type >= PERF_RECORD_HEADER_MAX)
1861 		return -EINVAL;
1862 
1863 	events_stats__inc(&evlist->stats, event->header.type);
1864 
1865 	if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1866 		return perf_session__process_user_event(session, event, file_offset, file_path);
1867 
1868 	if (tool->ordered_events) {
1869 		u64 timestamp = -1ULL;
1870 
1871 		ret = evlist__parse_sample_timestamp(evlist, event, &timestamp);
1872 		if (ret && ret != -1)
1873 			return ret;
1874 
1875 		ret = perf_session__queue_event(session, event, timestamp, file_offset, file_path);
1876 		if (ret != -ETIME)
1877 			return ret;
1878 	}
1879 
1880 	return perf_session__deliver_event(session, event, tool, file_offset, file_path);
1881 }
1882 
1883 void perf_event_header__bswap(struct perf_event_header *hdr)
1884 {
1885 	hdr->type = bswap_32(hdr->type);
1886 	hdr->misc = bswap_16(hdr->misc);
1887 	hdr->size = bswap_16(hdr->size);
1888 }
1889 
1890 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid)
1891 {
1892 	return machine__findnew_thread(&session->machines.host, -1, pid);
1893 }
1894 
1895 int perf_session__register_idle_thread(struct perf_session *session)
1896 {
1897 	struct thread *thread = machine__idle_thread(&session->machines.host);
1898 
1899 	/* machine__idle_thread() got the thread, so put it */
1900 	thread__put(thread);
1901 	return thread ? 0 : -1;
1902 }
1903 
1904 static void
1905 perf_session__warn_order(const struct perf_session *session)
1906 {
1907 	const struct ordered_events *oe = &session->ordered_events;
1908 	struct evsel *evsel;
1909 	bool should_warn = true;
1910 
1911 	evlist__for_each_entry(session->evlist, evsel) {
1912 		if (evsel->core.attr.write_backward)
1913 			should_warn = false;
1914 	}
1915 
1916 	if (!should_warn)
1917 		return;
1918 	if (oe->nr_unordered_events != 0)
1919 		ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events);
1920 }
1921 
1922 static void perf_session__warn_about_errors(const struct perf_session *session)
1923 {
1924 	const struct events_stats *stats = &session->evlist->stats;
1925 
1926 	if (session->tool->lost == perf_event__process_lost &&
1927 	    stats->nr_events[PERF_RECORD_LOST] != 0) {
1928 		ui__warning("Processed %d events and lost %d chunks!\n\n"
1929 			    "Check IO/CPU overload!\n\n",
1930 			    stats->nr_events[0],
1931 			    stats->nr_events[PERF_RECORD_LOST]);
1932 	}
1933 
1934 	if (session->tool->lost_samples == perf_event__process_lost_samples) {
1935 		double drop_rate;
1936 
1937 		drop_rate = (double)stats->total_lost_samples /
1938 			    (double) (stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples);
1939 		if (drop_rate > 0.05) {
1940 			ui__warning("Processed %" PRIu64 " samples and lost %3.2f%%!\n\n",
1941 				    stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples,
1942 				    drop_rate * 100.0);
1943 		}
1944 	}
1945 
1946 	if (session->tool->aux == perf_event__process_aux &&
1947 	    stats->total_aux_lost != 0) {
1948 		ui__warning("AUX data lost %" PRIu64 " times out of %u!\n\n",
1949 			    stats->total_aux_lost,
1950 			    stats->nr_events[PERF_RECORD_AUX]);
1951 	}
1952 
1953 	if (session->tool->aux == perf_event__process_aux &&
1954 	    stats->total_aux_partial != 0) {
1955 		bool vmm_exclusive = false;
1956 
1957 		(void)sysfs__read_bool("module/kvm_intel/parameters/vmm_exclusive",
1958 		                       &vmm_exclusive);
1959 
1960 		ui__warning("AUX data had gaps in it %" PRIu64 " times out of %u!\n\n"
1961 		            "Are you running a KVM guest in the background?%s\n\n",
1962 			    stats->total_aux_partial,
1963 			    stats->nr_events[PERF_RECORD_AUX],
1964 			    vmm_exclusive ?
1965 			    "\nReloading kvm_intel module with vmm_exclusive=0\n"
1966 			    "will reduce the gaps to only guest's timeslices." :
1967 			    "");
1968 	}
1969 
1970 	if (session->tool->aux == perf_event__process_aux &&
1971 	    stats->total_aux_collision != 0) {
1972 		ui__warning("AUX data detected collision  %" PRIu64 " times out of %u!\n\n",
1973 			    stats->total_aux_collision,
1974 			    stats->nr_events[PERF_RECORD_AUX]);
1975 	}
1976 
1977 	if (stats->nr_unknown_events != 0) {
1978 		ui__warning("Found %u unknown events!\n\n"
1979 			    "Is this an older tool processing a perf.data "
1980 			    "file generated by a more recent tool?\n\n"
1981 			    "If that is not the case, consider "
1982 			    "reporting to linux-kernel@vger.kernel.org.\n\n",
1983 			    stats->nr_unknown_events);
1984 	}
1985 
1986 	if (stats->nr_unknown_id != 0) {
1987 		ui__warning("%u samples with id not present in the header\n",
1988 			    stats->nr_unknown_id);
1989 	}
1990 
1991 	if (stats->nr_invalid_chains != 0) {
1992 		ui__warning("Found invalid callchains!\n\n"
1993 			    "%u out of %u events were discarded for this reason.\n\n"
1994 			    "Consider reporting to linux-kernel@vger.kernel.org.\n\n",
1995 			    stats->nr_invalid_chains,
1996 			    stats->nr_events[PERF_RECORD_SAMPLE]);
1997 	}
1998 
1999 	if (stats->nr_unprocessable_samples != 0) {
2000 		ui__warning("%u unprocessable samples recorded.\n"
2001 			    "Do you have a KVM guest running and not using 'perf kvm'?\n",
2002 			    stats->nr_unprocessable_samples);
2003 	}
2004 
2005 	perf_session__warn_order(session);
2006 
2007 	events_stats__auxtrace_error_warn(stats);
2008 
2009 	if (stats->nr_proc_map_timeout != 0) {
2010 		ui__warning("%d map information files for pre-existing threads were\n"
2011 			    "not processed, if there are samples for addresses they\n"
2012 			    "will not be resolved, you may find out which are these\n"
2013 			    "threads by running with -v and redirecting the output\n"
2014 			    "to a file.\n"
2015 			    "The time limit to process proc map is too short?\n"
2016 			    "Increase it by --proc-map-timeout\n",
2017 			    stats->nr_proc_map_timeout);
2018 	}
2019 }
2020 
2021 static int perf_session__flush_thread_stack(struct thread *thread,
2022 					    void *p __maybe_unused)
2023 {
2024 	return thread_stack__flush(thread);
2025 }
2026 
2027 static int perf_session__flush_thread_stacks(struct perf_session *session)
2028 {
2029 	return machines__for_each_thread(&session->machines,
2030 					 perf_session__flush_thread_stack,
2031 					 NULL);
2032 }
2033 
2034 volatile sig_atomic_t session_done;
2035 
2036 static int __perf_session__process_decomp_events(struct perf_session *session);
2037 
2038 static int __perf_session__process_pipe_events(struct perf_session *session)
2039 {
2040 	struct ordered_events *oe = &session->ordered_events;
2041 	struct perf_tool *tool = session->tool;
2042 	union perf_event *event;
2043 	uint32_t size, cur_size = 0;
2044 	void *buf = NULL;
2045 	s64 skip = 0;
2046 	u64 head;
2047 	ssize_t err;
2048 	void *p;
2049 
2050 	perf_tool__fill_defaults(tool);
2051 
2052 	head = 0;
2053 	cur_size = sizeof(union perf_event);
2054 
2055 	buf = malloc(cur_size);
2056 	if (!buf)
2057 		return -errno;
2058 	ordered_events__set_copy_on_queue(oe, true);
2059 more:
2060 	event = buf;
2061 	err = perf_data__read(session->data, event,
2062 			      sizeof(struct perf_event_header));
2063 	if (err <= 0) {
2064 		if (err == 0)
2065 			goto done;
2066 
2067 		pr_err("failed to read event header\n");
2068 		goto out_err;
2069 	}
2070 
2071 	if (session->header.needs_swap)
2072 		perf_event_header__bswap(&event->header);
2073 
2074 	size = event->header.size;
2075 	if (size < sizeof(struct perf_event_header)) {
2076 		pr_err("bad event header size\n");
2077 		goto out_err;
2078 	}
2079 
2080 	if (size > cur_size) {
2081 		void *new = realloc(buf, size);
2082 		if (!new) {
2083 			pr_err("failed to allocate memory to read event\n");
2084 			goto out_err;
2085 		}
2086 		buf = new;
2087 		cur_size = size;
2088 		event = buf;
2089 	}
2090 	p = event;
2091 	p += sizeof(struct perf_event_header);
2092 
2093 	if (size - sizeof(struct perf_event_header)) {
2094 		err = perf_data__read(session->data, p,
2095 				      size - sizeof(struct perf_event_header));
2096 		if (err <= 0) {
2097 			if (err == 0) {
2098 				pr_err("unexpected end of event stream\n");
2099 				goto done;
2100 			}
2101 
2102 			pr_err("failed to read event data\n");
2103 			goto out_err;
2104 		}
2105 	}
2106 
2107 	if ((skip = perf_session__process_event(session, event, head, "pipe")) < 0) {
2108 		pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
2109 		       head, event->header.size, event->header.type);
2110 		err = -EINVAL;
2111 		goto out_err;
2112 	}
2113 
2114 	head += size;
2115 
2116 	if (skip > 0)
2117 		head += skip;
2118 
2119 	err = __perf_session__process_decomp_events(session);
2120 	if (err)
2121 		goto out_err;
2122 
2123 	if (!session_done())
2124 		goto more;
2125 done:
2126 	/* do the final flush for ordered samples */
2127 	err = ordered_events__flush(oe, OE_FLUSH__FINAL);
2128 	if (err)
2129 		goto out_err;
2130 	err = auxtrace__flush_events(session, tool);
2131 	if (err)
2132 		goto out_err;
2133 	err = perf_session__flush_thread_stacks(session);
2134 out_err:
2135 	free(buf);
2136 	if (!tool->no_warn)
2137 		perf_session__warn_about_errors(session);
2138 	ordered_events__free(&session->ordered_events);
2139 	auxtrace__free_events(session);
2140 	return err;
2141 }
2142 
2143 static union perf_event *
2144 prefetch_event(char *buf, u64 head, size_t mmap_size,
2145 	       bool needs_swap, union perf_event *error)
2146 {
2147 	union perf_event *event;
2148 	u16 event_size;
2149 
2150 	/*
2151 	 * Ensure we have enough space remaining to read
2152 	 * the size of the event in the headers.
2153 	 */
2154 	if (head + sizeof(event->header) > mmap_size)
2155 		return NULL;
2156 
2157 	event = (union perf_event *)(buf + head);
2158 	if (needs_swap)
2159 		perf_event_header__bswap(&event->header);
2160 
2161 	event_size = event->header.size;
2162 	if (head + event_size <= mmap_size)
2163 		return event;
2164 
2165 	/* We're not fetching the event so swap back again */
2166 	if (needs_swap)
2167 		perf_event_header__bswap(&event->header);
2168 
2169 	/* Check if the event fits into the next mmapped buf. */
2170 	if (event_size <= mmap_size - head % page_size) {
2171 		/* Remap buf and fetch again. */
2172 		return NULL;
2173 	}
2174 
2175 	/* Invalid input. Event size should never exceed mmap_size. */
2176 	pr_debug("%s: head=%#" PRIx64 " event->header.size=%#x, mmap_size=%#zx:"
2177 		 " fuzzed or compressed perf.data?\n", __func__, head, event_size, mmap_size);
2178 
2179 	return error;
2180 }
2181 
2182 static union perf_event *
2183 fetch_mmaped_event(u64 head, size_t mmap_size, char *buf, bool needs_swap)
2184 {
2185 	return prefetch_event(buf, head, mmap_size, needs_swap, ERR_PTR(-EINVAL));
2186 }
2187 
2188 static union perf_event *
2189 fetch_decomp_event(u64 head, size_t mmap_size, char *buf, bool needs_swap)
2190 {
2191 	return prefetch_event(buf, head, mmap_size, needs_swap, NULL);
2192 }
2193 
2194 static int __perf_session__process_decomp_events(struct perf_session *session)
2195 {
2196 	s64 skip;
2197 	u64 size;
2198 	struct decomp *decomp = session->active_decomp->decomp_last;
2199 
2200 	if (!decomp)
2201 		return 0;
2202 
2203 	while (decomp->head < decomp->size && !session_done()) {
2204 		union perf_event *event = fetch_decomp_event(decomp->head, decomp->size, decomp->data,
2205 							     session->header.needs_swap);
2206 
2207 		if (!event)
2208 			break;
2209 
2210 		size = event->header.size;
2211 
2212 		if (size < sizeof(struct perf_event_header) ||
2213 		    (skip = perf_session__process_event(session, event, decomp->file_pos,
2214 							decomp->file_path)) < 0) {
2215 			pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
2216 				decomp->file_pos + decomp->head, event->header.size, event->header.type);
2217 			return -EINVAL;
2218 		}
2219 
2220 		if (skip)
2221 			size += skip;
2222 
2223 		decomp->head += size;
2224 	}
2225 
2226 	return 0;
2227 }
2228 
2229 /*
2230  * On 64bit we can mmap the data file in one go. No need for tiny mmap
2231  * slices. On 32bit we use 32MB.
2232  */
2233 #if BITS_PER_LONG == 64
2234 #define MMAP_SIZE ULLONG_MAX
2235 #define NUM_MMAPS 1
2236 #else
2237 #define MMAP_SIZE (32 * 1024 * 1024ULL)
2238 #define NUM_MMAPS 128
2239 #endif
2240 
2241 struct reader;
2242 
2243 typedef s64 (*reader_cb_t)(struct perf_session *session,
2244 			   union perf_event *event,
2245 			   u64 file_offset,
2246 			   const char *file_path);
2247 
2248 struct reader {
2249 	int		 fd;
2250 	const char	 *path;
2251 	u64		 data_size;
2252 	u64		 data_offset;
2253 	reader_cb_t	 process;
2254 	bool		 in_place_update;
2255 	char		 *mmaps[NUM_MMAPS];
2256 	size_t		 mmap_size;
2257 	int		 mmap_idx;
2258 	char		 *mmap_cur;
2259 	u64		 file_pos;
2260 	u64		 file_offset;
2261 	u64		 head;
2262 	u64		 size;
2263 	bool		 done;
2264 	struct zstd_data   zstd_data;
2265 	struct decomp_data decomp_data;
2266 };
2267 
2268 static int
2269 reader__init(struct reader *rd, bool *one_mmap)
2270 {
2271 	u64 data_size = rd->data_size;
2272 	char **mmaps = rd->mmaps;
2273 
2274 	rd->head = rd->data_offset;
2275 	data_size += rd->data_offset;
2276 
2277 	rd->mmap_size = MMAP_SIZE;
2278 	if (rd->mmap_size > data_size) {
2279 		rd->mmap_size = data_size;
2280 		if (one_mmap)
2281 			*one_mmap = true;
2282 	}
2283 
2284 	memset(mmaps, 0, sizeof(rd->mmaps));
2285 
2286 	if (zstd_init(&rd->zstd_data, 0))
2287 		return -1;
2288 	rd->decomp_data.zstd_decomp = &rd->zstd_data;
2289 
2290 	return 0;
2291 }
2292 
2293 static void
2294 reader__release_decomp(struct reader *rd)
2295 {
2296 	perf_decomp__release_events(rd->decomp_data.decomp);
2297 	zstd_fini(&rd->zstd_data);
2298 }
2299 
2300 static int
2301 reader__mmap(struct reader *rd, struct perf_session *session)
2302 {
2303 	int mmap_prot, mmap_flags;
2304 	char *buf, **mmaps = rd->mmaps;
2305 	u64 page_offset;
2306 
2307 	mmap_prot  = PROT_READ;
2308 	mmap_flags = MAP_SHARED;
2309 
2310 	if (rd->in_place_update) {
2311 		mmap_prot  |= PROT_WRITE;
2312 	} else if (session->header.needs_swap) {
2313 		mmap_prot  |= PROT_WRITE;
2314 		mmap_flags = MAP_PRIVATE;
2315 	}
2316 
2317 	if (mmaps[rd->mmap_idx]) {
2318 		munmap(mmaps[rd->mmap_idx], rd->mmap_size);
2319 		mmaps[rd->mmap_idx] = NULL;
2320 	}
2321 
2322 	page_offset = page_size * (rd->head / page_size);
2323 	rd->file_offset += page_offset;
2324 	rd->head -= page_offset;
2325 
2326 	buf = mmap(NULL, rd->mmap_size, mmap_prot, mmap_flags, rd->fd,
2327 		   rd->file_offset);
2328 	if (buf == MAP_FAILED) {
2329 		pr_err("failed to mmap file\n");
2330 		return -errno;
2331 	}
2332 	mmaps[rd->mmap_idx] = rd->mmap_cur = buf;
2333 	rd->mmap_idx = (rd->mmap_idx + 1) & (ARRAY_SIZE(rd->mmaps) - 1);
2334 	rd->file_pos = rd->file_offset + rd->head;
2335 	if (session->one_mmap) {
2336 		session->one_mmap_addr = buf;
2337 		session->one_mmap_offset = rd->file_offset;
2338 	}
2339 
2340 	return 0;
2341 }
2342 
2343 enum {
2344 	READER_OK,
2345 	READER_NODATA,
2346 };
2347 
2348 static int
2349 reader__read_event(struct reader *rd, struct perf_session *session,
2350 		   struct ui_progress *prog)
2351 {
2352 	u64 size;
2353 	int err = READER_OK;
2354 	union perf_event *event;
2355 	s64 skip;
2356 
2357 	event = fetch_mmaped_event(rd->head, rd->mmap_size, rd->mmap_cur,
2358 				   session->header.needs_swap);
2359 	if (IS_ERR(event))
2360 		return PTR_ERR(event);
2361 
2362 	if (!event)
2363 		return READER_NODATA;
2364 
2365 	size = event->header.size;
2366 
2367 	skip = -EINVAL;
2368 
2369 	if (size < sizeof(struct perf_event_header) ||
2370 	    (skip = rd->process(session, event, rd->file_pos, rd->path)) < 0) {
2371 		pr_err("%#" PRIx64 " [%#x]: failed to process type: %d [%s]\n",
2372 		       rd->file_offset + rd->head, event->header.size,
2373 		       event->header.type, strerror(-skip));
2374 		err = skip;
2375 		goto out;
2376 	}
2377 
2378 	if (skip)
2379 		size += skip;
2380 
2381 	rd->size += size;
2382 	rd->head += size;
2383 	rd->file_pos += size;
2384 
2385 	err = __perf_session__process_decomp_events(session);
2386 	if (err)
2387 		goto out;
2388 
2389 	ui_progress__update(prog, size);
2390 
2391 out:
2392 	return err;
2393 }
2394 
2395 static inline bool
2396 reader__eof(struct reader *rd)
2397 {
2398 	return (rd->file_pos >= rd->data_size + rd->data_offset);
2399 }
2400 
2401 static int
2402 reader__process_events(struct reader *rd, struct perf_session *session,
2403 		       struct ui_progress *prog)
2404 {
2405 	int err;
2406 
2407 	err = reader__init(rd, &session->one_mmap);
2408 	if (err)
2409 		goto out;
2410 
2411 	session->active_decomp = &rd->decomp_data;
2412 
2413 remap:
2414 	err = reader__mmap(rd, session);
2415 	if (err)
2416 		goto out;
2417 
2418 more:
2419 	err = reader__read_event(rd, session, prog);
2420 	if (err < 0)
2421 		goto out;
2422 	else if (err == READER_NODATA)
2423 		goto remap;
2424 
2425 	if (session_done())
2426 		goto out;
2427 
2428 	if (!reader__eof(rd))
2429 		goto more;
2430 
2431 out:
2432 	session->active_decomp = &session->decomp_data;
2433 	return err;
2434 }
2435 
2436 static s64 process_simple(struct perf_session *session,
2437 			  union perf_event *event,
2438 			  u64 file_offset,
2439 			  const char *file_path)
2440 {
2441 	return perf_session__process_event(session, event, file_offset, file_path);
2442 }
2443 
2444 static int __perf_session__process_events(struct perf_session *session)
2445 {
2446 	struct reader rd = {
2447 		.fd		= perf_data__fd(session->data),
2448 		.path		= session->data->file.path,
2449 		.data_size	= session->header.data_size,
2450 		.data_offset	= session->header.data_offset,
2451 		.process	= process_simple,
2452 		.in_place_update = session->data->in_place_update,
2453 	};
2454 	struct ordered_events *oe = &session->ordered_events;
2455 	struct perf_tool *tool = session->tool;
2456 	struct ui_progress prog;
2457 	int err;
2458 
2459 	perf_tool__fill_defaults(tool);
2460 
2461 	if (rd.data_size == 0)
2462 		return -1;
2463 
2464 	ui_progress__init_size(&prog, rd.data_size, "Processing events...");
2465 
2466 	err = reader__process_events(&rd, session, &prog);
2467 	if (err)
2468 		goto out_err;
2469 	/* do the final flush for ordered samples */
2470 	err = ordered_events__flush(oe, OE_FLUSH__FINAL);
2471 	if (err)
2472 		goto out_err;
2473 	err = auxtrace__flush_events(session, tool);
2474 	if (err)
2475 		goto out_err;
2476 	err = perf_session__flush_thread_stacks(session);
2477 out_err:
2478 	ui_progress__finish();
2479 	if (!tool->no_warn)
2480 		perf_session__warn_about_errors(session);
2481 	/*
2482 	 * We may switching perf.data output, make ordered_events
2483 	 * reusable.
2484 	 */
2485 	ordered_events__reinit(&session->ordered_events);
2486 	auxtrace__free_events(session);
2487 	reader__release_decomp(&rd);
2488 	session->one_mmap = false;
2489 	return err;
2490 }
2491 
2492 /*
2493  * Processing 2 MB of data from each reader in sequence,
2494  * because that's the way the ordered events sorting works
2495  * most efficiently.
2496  */
2497 #define READER_MAX_SIZE (2 * 1024 * 1024)
2498 
2499 /*
2500  * This function reads, merge and process directory data.
2501  * It assumens the version 1 of directory data, where each
2502  * data file holds per-cpu data, already sorted by kernel.
2503  */
2504 static int __perf_session__process_dir_events(struct perf_session *session)
2505 {
2506 	struct perf_data *data = session->data;
2507 	struct perf_tool *tool = session->tool;
2508 	int i, ret, readers, nr_readers;
2509 	struct ui_progress prog;
2510 	u64 total_size = perf_data__size(session->data);
2511 	struct reader *rd;
2512 
2513 	perf_tool__fill_defaults(tool);
2514 
2515 	ui_progress__init_size(&prog, total_size, "Sorting events...");
2516 
2517 	nr_readers = 1;
2518 	for (i = 0; i < data->dir.nr; i++) {
2519 		if (data->dir.files[i].size)
2520 			nr_readers++;
2521 	}
2522 
2523 	rd = zalloc(nr_readers * sizeof(struct reader));
2524 	if (!rd)
2525 		return -ENOMEM;
2526 
2527 	rd[0] = (struct reader) {
2528 		.fd		 = perf_data__fd(session->data),
2529 		.path		 = session->data->file.path,
2530 		.data_size	 = session->header.data_size,
2531 		.data_offset	 = session->header.data_offset,
2532 		.process	 = process_simple,
2533 		.in_place_update = session->data->in_place_update,
2534 	};
2535 	ret = reader__init(&rd[0], NULL);
2536 	if (ret)
2537 		goto out_err;
2538 	ret = reader__mmap(&rd[0], session);
2539 	if (ret)
2540 		goto out_err;
2541 	readers = 1;
2542 
2543 	for (i = 0; i < data->dir.nr; i++) {
2544 		if (!data->dir.files[i].size)
2545 			continue;
2546 		rd[readers] = (struct reader) {
2547 			.fd		 = data->dir.files[i].fd,
2548 			.path		 = data->dir.files[i].path,
2549 			.data_size	 = data->dir.files[i].size,
2550 			.data_offset	 = 0,
2551 			.process	 = process_simple,
2552 			.in_place_update = session->data->in_place_update,
2553 		};
2554 		ret = reader__init(&rd[readers], NULL);
2555 		if (ret)
2556 			goto out_err;
2557 		ret = reader__mmap(&rd[readers], session);
2558 		if (ret)
2559 			goto out_err;
2560 		readers++;
2561 	}
2562 
2563 	i = 0;
2564 	while (readers) {
2565 		if (session_done())
2566 			break;
2567 
2568 		if (rd[i].done) {
2569 			i = (i + 1) % nr_readers;
2570 			continue;
2571 		}
2572 		if (reader__eof(&rd[i])) {
2573 			rd[i].done = true;
2574 			readers--;
2575 			continue;
2576 		}
2577 
2578 		session->active_decomp = &rd[i].decomp_data;
2579 		ret = reader__read_event(&rd[i], session, &prog);
2580 		if (ret < 0) {
2581 			goto out_err;
2582 		} else if (ret == READER_NODATA) {
2583 			ret = reader__mmap(&rd[i], session);
2584 			if (ret)
2585 				goto out_err;
2586 		}
2587 
2588 		if (rd[i].size >= READER_MAX_SIZE) {
2589 			rd[i].size = 0;
2590 			i = (i + 1) % nr_readers;
2591 		}
2592 	}
2593 
2594 	ret = ordered_events__flush(&session->ordered_events, OE_FLUSH__FINAL);
2595 	if (ret)
2596 		goto out_err;
2597 
2598 	ret = perf_session__flush_thread_stacks(session);
2599 out_err:
2600 	ui_progress__finish();
2601 
2602 	if (!tool->no_warn)
2603 		perf_session__warn_about_errors(session);
2604 
2605 	/*
2606 	 * We may switching perf.data output, make ordered_events
2607 	 * reusable.
2608 	 */
2609 	ordered_events__reinit(&session->ordered_events);
2610 
2611 	session->one_mmap = false;
2612 
2613 	session->active_decomp = &session->decomp_data;
2614 	for (i = 0; i < nr_readers; i++)
2615 		reader__release_decomp(&rd[i]);
2616 	zfree(&rd);
2617 
2618 	return ret;
2619 }
2620 
2621 int perf_session__process_events(struct perf_session *session)
2622 {
2623 	if (perf_session__register_idle_thread(session) < 0)
2624 		return -ENOMEM;
2625 
2626 	if (perf_data__is_pipe(session->data))
2627 		return __perf_session__process_pipe_events(session);
2628 
2629 	if (perf_data__is_dir(session->data) && session->data->dir.nr)
2630 		return __perf_session__process_dir_events(session);
2631 
2632 	return __perf_session__process_events(session);
2633 }
2634 
2635 bool perf_session__has_traces(struct perf_session *session, const char *msg)
2636 {
2637 	struct evsel *evsel;
2638 
2639 	evlist__for_each_entry(session->evlist, evsel) {
2640 		if (evsel->core.attr.type == PERF_TYPE_TRACEPOINT)
2641 			return true;
2642 	}
2643 
2644 	pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg);
2645 	return false;
2646 }
2647 
2648 int map__set_kallsyms_ref_reloc_sym(struct map *map, const char *symbol_name, u64 addr)
2649 {
2650 	char *bracket;
2651 	struct ref_reloc_sym *ref;
2652 	struct kmap *kmap;
2653 
2654 	ref = zalloc(sizeof(struct ref_reloc_sym));
2655 	if (ref == NULL)
2656 		return -ENOMEM;
2657 
2658 	ref->name = strdup(symbol_name);
2659 	if (ref->name == NULL) {
2660 		free(ref);
2661 		return -ENOMEM;
2662 	}
2663 
2664 	bracket = strchr(ref->name, ']');
2665 	if (bracket)
2666 		*bracket = '\0';
2667 
2668 	ref->addr = addr;
2669 
2670 	kmap = map__kmap(map);
2671 	if (kmap)
2672 		kmap->ref_reloc_sym = ref;
2673 
2674 	return 0;
2675 }
2676 
2677 size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp)
2678 {
2679 	return machines__fprintf_dsos(&session->machines, fp);
2680 }
2681 
2682 size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp,
2683 					  bool (skip)(struct dso *dso, int parm), int parm)
2684 {
2685 	return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm);
2686 }
2687 
2688 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp,
2689 				       bool skip_empty)
2690 {
2691 	size_t ret;
2692 	const char *msg = "";
2693 
2694 	if (perf_header__has_feat(&session->header, HEADER_AUXTRACE))
2695 		msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)";
2696 
2697 	ret = fprintf(fp, "\nAggregated stats:%s\n", msg);
2698 
2699 	ret += events_stats__fprintf(&session->evlist->stats, fp, skip_empty);
2700 	return ret;
2701 }
2702 
2703 size_t perf_session__fprintf(struct perf_session *session, FILE *fp)
2704 {
2705 	/*
2706 	 * FIXME: Here we have to actually print all the machines in this
2707 	 * session, not just the host...
2708 	 */
2709 	return machine__fprintf(&session->machines.host, fp);
2710 }
2711 
2712 struct evsel *perf_session__find_first_evtype(struct perf_session *session,
2713 					      unsigned int type)
2714 {
2715 	struct evsel *pos;
2716 
2717 	evlist__for_each_entry(session->evlist, pos) {
2718 		if (pos->core.attr.type == type)
2719 			return pos;
2720 	}
2721 	return NULL;
2722 }
2723 
2724 int perf_session__cpu_bitmap(struct perf_session *session,
2725 			     const char *cpu_list, unsigned long *cpu_bitmap)
2726 {
2727 	int i, err = -1;
2728 	struct perf_cpu_map *map;
2729 	int nr_cpus = min(session->header.env.nr_cpus_avail, MAX_NR_CPUS);
2730 
2731 	for (i = 0; i < PERF_TYPE_MAX; ++i) {
2732 		struct evsel *evsel;
2733 
2734 		evsel = perf_session__find_first_evtype(session, i);
2735 		if (!evsel)
2736 			continue;
2737 
2738 		if (!(evsel->core.attr.sample_type & PERF_SAMPLE_CPU)) {
2739 			pr_err("File does not contain CPU events. "
2740 			       "Remove -C option to proceed.\n");
2741 			return -1;
2742 		}
2743 	}
2744 
2745 	map = perf_cpu_map__new(cpu_list);
2746 	if (map == NULL) {
2747 		pr_err("Invalid cpu_list\n");
2748 		return -1;
2749 	}
2750 
2751 	for (i = 0; i < perf_cpu_map__nr(map); i++) {
2752 		struct perf_cpu cpu = perf_cpu_map__cpu(map, i);
2753 
2754 		if (cpu.cpu >= nr_cpus) {
2755 			pr_err("Requested CPU %d too large. "
2756 			       "Consider raising MAX_NR_CPUS\n", cpu.cpu);
2757 			goto out_delete_map;
2758 		}
2759 
2760 		__set_bit(cpu.cpu, cpu_bitmap);
2761 	}
2762 
2763 	err = 0;
2764 
2765 out_delete_map:
2766 	perf_cpu_map__put(map);
2767 	return err;
2768 }
2769 
2770 void perf_session__fprintf_info(struct perf_session *session, FILE *fp,
2771 				bool full)
2772 {
2773 	if (session == NULL || fp == NULL)
2774 		return;
2775 
2776 	fprintf(fp, "# ========\n");
2777 	perf_header__fprintf_info(session, fp, full);
2778 	fprintf(fp, "# ========\n#\n");
2779 }
2780 
2781 static int perf_session__register_guest(struct perf_session *session, pid_t machine_pid)
2782 {
2783 	struct machine *machine = machines__findnew(&session->machines, machine_pid);
2784 	struct thread *thread;
2785 
2786 	if (!machine)
2787 		return -ENOMEM;
2788 
2789 	machine->single_address_space = session->machines.host.single_address_space;
2790 
2791 	thread = machine__idle_thread(machine);
2792 	if (!thread)
2793 		return -ENOMEM;
2794 	thread__put(thread);
2795 
2796 	machine->kallsyms_filename = perf_data__guest_kallsyms_name(session->data, machine_pid);
2797 
2798 	return 0;
2799 }
2800 
2801 static int perf_session__set_guest_cpu(struct perf_session *session, pid_t pid,
2802 				       pid_t tid, int guest_cpu)
2803 {
2804 	struct machine *machine = &session->machines.host;
2805 	struct thread *thread = machine__findnew_thread(machine, pid, tid);
2806 
2807 	if (!thread)
2808 		return -ENOMEM;
2809 	thread->guest_cpu = guest_cpu;
2810 	thread__put(thread);
2811 
2812 	return 0;
2813 }
2814 
2815 int perf_event__process_id_index(struct perf_session *session,
2816 				 union perf_event *event)
2817 {
2818 	struct evlist *evlist = session->evlist;
2819 	struct perf_record_id_index *ie = &event->id_index;
2820 	size_t sz = ie->header.size - sizeof(*ie);
2821 	size_t i, nr, max_nr;
2822 	size_t e1_sz = sizeof(struct id_index_entry);
2823 	size_t e2_sz = sizeof(struct id_index_entry_2);
2824 	size_t etot_sz = e1_sz + e2_sz;
2825 	struct id_index_entry_2 *e2;
2826 	pid_t last_pid = 0;
2827 
2828 	max_nr = sz / e1_sz;
2829 	nr = ie->nr;
2830 	if (nr > max_nr) {
2831 		printf("Too big: nr %zu max_nr %zu\n", nr, max_nr);
2832 		return -EINVAL;
2833 	}
2834 
2835 	if (sz >= nr * etot_sz) {
2836 		max_nr = sz / etot_sz;
2837 		if (nr > max_nr) {
2838 			printf("Too big2: nr %zu max_nr %zu\n", nr, max_nr);
2839 			return -EINVAL;
2840 		}
2841 		e2 = (void *)ie + sizeof(*ie) + nr * e1_sz;
2842 	} else {
2843 		e2 = NULL;
2844 	}
2845 
2846 	if (dump_trace)
2847 		fprintf(stdout, " nr: %zu\n", nr);
2848 
2849 	for (i = 0; i < nr; i++, (e2 ? e2++ : 0)) {
2850 		struct id_index_entry *e = &ie->entries[i];
2851 		struct perf_sample_id *sid;
2852 		int ret;
2853 
2854 		if (dump_trace) {
2855 			fprintf(stdout,	" ... id: %"PRI_lu64, e->id);
2856 			fprintf(stdout,	"  idx: %"PRI_lu64, e->idx);
2857 			fprintf(stdout,	"  cpu: %"PRI_ld64, e->cpu);
2858 			fprintf(stdout, "  tid: %"PRI_ld64, e->tid);
2859 			if (e2) {
2860 				fprintf(stdout, "  machine_pid: %"PRI_ld64, e2->machine_pid);
2861 				fprintf(stdout, "  vcpu: %"PRI_lu64"\n", e2->vcpu);
2862 			} else {
2863 				fprintf(stdout, "\n");
2864 			}
2865 		}
2866 
2867 		sid = evlist__id2sid(evlist, e->id);
2868 		if (!sid)
2869 			return -ENOENT;
2870 
2871 		sid->idx = e->idx;
2872 		sid->cpu.cpu = e->cpu;
2873 		sid->tid = e->tid;
2874 
2875 		if (!e2)
2876 			continue;
2877 
2878 		sid->machine_pid = e2->machine_pid;
2879 		sid->vcpu.cpu = e2->vcpu;
2880 
2881 		if (!sid->machine_pid)
2882 			continue;
2883 
2884 		if (sid->machine_pid != last_pid) {
2885 			ret = perf_session__register_guest(session, sid->machine_pid);
2886 			if (ret)
2887 				return ret;
2888 			last_pid = sid->machine_pid;
2889 			perf_guest = true;
2890 		}
2891 
2892 		ret = perf_session__set_guest_cpu(session, sid->machine_pid, e->tid, e2->vcpu);
2893 		if (ret)
2894 			return ret;
2895 	}
2896 	return 0;
2897 }
2898