xref: /linux/tools/perf/util/session.c (revision b7019ac550eb3916f34d79db583e9b7ea2524afa)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <errno.h>
3 #include <inttypes.h>
4 #include <linux/kernel.h>
5 #include <traceevent/event-parse.h>
6 #include <api/fs/fs.h>
7 
8 #include <byteswap.h>
9 #include <unistd.h>
10 #include <sys/types.h>
11 #include <sys/mman.h>
12 
13 #include "evlist.h"
14 #include "evsel.h"
15 #include "memswap.h"
16 #include "map.h"
17 #include "symbol.h"
18 #include "session.h"
19 #include "tool.h"
20 #include "sort.h"
21 #include "util.h"
22 #include "cpumap.h"
23 #include "perf_regs.h"
24 #include "asm/bug.h"
25 #include "auxtrace.h"
26 #include "thread.h"
27 #include "thread-stack.h"
28 #include "sample-raw.h"
29 #include "stat.h"
30 #include "arch/common.h"
31 
32 #ifdef HAVE_ZSTD_SUPPORT
33 static int perf_session__process_compressed_event(struct perf_session *session,
34 						  union perf_event *event, u64 file_offset)
35 {
36 	void *src;
37 	size_t decomp_size, src_size;
38 	u64 decomp_last_rem = 0;
39 	size_t decomp_len = session->header.env.comp_mmap_len;
40 	struct decomp *decomp, *decomp_last = session->decomp_last;
41 
42 	decomp = mmap(NULL, sizeof(struct decomp) + decomp_len, PROT_READ|PROT_WRITE,
43 		      MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);
44 	if (decomp == MAP_FAILED) {
45 		pr_err("Couldn't allocate memory for decompression\n");
46 		return -1;
47 	}
48 
49 	decomp->file_pos = file_offset;
50 	decomp->head = 0;
51 
52 	if (decomp_last) {
53 		decomp_last_rem = decomp_last->size - decomp_last->head;
54 		memcpy(decomp->data, &(decomp_last->data[decomp_last->head]), decomp_last_rem);
55 		decomp->size = decomp_last_rem;
56 	}
57 
58 	src = (void *)event + sizeof(struct compressed_event);
59 	src_size = event->pack.header.size - sizeof(struct compressed_event);
60 
61 	decomp_size = zstd_decompress_stream(&(session->zstd_data), src, src_size,
62 				&(decomp->data[decomp_last_rem]), decomp_len - decomp_last_rem);
63 	if (!decomp_size) {
64 		munmap(decomp, sizeof(struct decomp) + decomp_len);
65 		pr_err("Couldn't decompress data\n");
66 		return -1;
67 	}
68 
69 	decomp->size += decomp_size;
70 
71 	if (session->decomp == NULL) {
72 		session->decomp = decomp;
73 		session->decomp_last = decomp;
74 	} else {
75 		session->decomp_last->next = decomp;
76 		session->decomp_last = decomp;
77 	}
78 
79 	pr_debug("decomp (B): %ld to %ld\n", src_size, decomp_size);
80 
81 	return 0;
82 }
83 #else /* !HAVE_ZSTD_SUPPORT */
84 #define perf_session__process_compressed_event perf_session__process_compressed_event_stub
85 #endif
86 
87 static int perf_session__deliver_event(struct perf_session *session,
88 				       union perf_event *event,
89 				       struct perf_tool *tool,
90 				       u64 file_offset);
91 
92 static int perf_session__open(struct perf_session *session)
93 {
94 	struct perf_data *data = session->data;
95 
96 	if (perf_session__read_header(session) < 0) {
97 		pr_err("incompatible file format (rerun with -v to learn more)\n");
98 		return -1;
99 	}
100 
101 	if (perf_data__is_pipe(data))
102 		return 0;
103 
104 	if (perf_header__has_feat(&session->header, HEADER_STAT))
105 		return 0;
106 
107 	if (!perf_evlist__valid_sample_type(session->evlist)) {
108 		pr_err("non matching sample_type\n");
109 		return -1;
110 	}
111 
112 	if (!perf_evlist__valid_sample_id_all(session->evlist)) {
113 		pr_err("non matching sample_id_all\n");
114 		return -1;
115 	}
116 
117 	if (!perf_evlist__valid_read_format(session->evlist)) {
118 		pr_err("non matching read_format\n");
119 		return -1;
120 	}
121 
122 	return 0;
123 }
124 
125 void perf_session__set_id_hdr_size(struct perf_session *session)
126 {
127 	u16 id_hdr_size = perf_evlist__id_hdr_size(session->evlist);
128 
129 	machines__set_id_hdr_size(&session->machines, id_hdr_size);
130 }
131 
132 int perf_session__create_kernel_maps(struct perf_session *session)
133 {
134 	int ret = machine__create_kernel_maps(&session->machines.host);
135 
136 	if (ret >= 0)
137 		ret = machines__create_guest_kernel_maps(&session->machines);
138 	return ret;
139 }
140 
141 static void perf_session__destroy_kernel_maps(struct perf_session *session)
142 {
143 	machines__destroy_kernel_maps(&session->machines);
144 }
145 
146 static bool perf_session__has_comm_exec(struct perf_session *session)
147 {
148 	struct perf_evsel *evsel;
149 
150 	evlist__for_each_entry(session->evlist, evsel) {
151 		if (evsel->attr.comm_exec)
152 			return true;
153 	}
154 
155 	return false;
156 }
157 
158 static void perf_session__set_comm_exec(struct perf_session *session)
159 {
160 	bool comm_exec = perf_session__has_comm_exec(session);
161 
162 	machines__set_comm_exec(&session->machines, comm_exec);
163 }
164 
165 static int ordered_events__deliver_event(struct ordered_events *oe,
166 					 struct ordered_event *event)
167 {
168 	struct perf_session *session = container_of(oe, struct perf_session,
169 						    ordered_events);
170 
171 	return perf_session__deliver_event(session, event->event,
172 					   session->tool, event->file_offset);
173 }
174 
175 struct perf_session *perf_session__new(struct perf_data *data,
176 				       bool repipe, struct perf_tool *tool)
177 {
178 	struct perf_session *session = zalloc(sizeof(*session));
179 
180 	if (!session)
181 		goto out;
182 
183 	session->repipe = repipe;
184 	session->tool   = tool;
185 	INIT_LIST_HEAD(&session->auxtrace_index);
186 	machines__init(&session->machines);
187 	ordered_events__init(&session->ordered_events,
188 			     ordered_events__deliver_event, NULL);
189 
190 	perf_env__init(&session->header.env);
191 	if (data) {
192 		if (perf_data__open(data))
193 			goto out_delete;
194 
195 		session->data = data;
196 
197 		if (perf_data__is_read(data)) {
198 			if (perf_session__open(session) < 0)
199 				goto out_delete;
200 
201 			/*
202 			 * set session attributes that are present in perf.data
203 			 * but not in pipe-mode.
204 			 */
205 			if (!data->is_pipe) {
206 				perf_session__set_id_hdr_size(session);
207 				perf_session__set_comm_exec(session);
208 			}
209 
210 			perf_evlist__init_trace_event_sample_raw(session->evlist);
211 
212 			/* Open the directory data. */
213 			if (data->is_dir && perf_data__open_dir(data))
214 				goto out_delete;
215 		}
216 	} else  {
217 		session->machines.host.env = &perf_env;
218 	}
219 
220 	session->machines.host.single_address_space =
221 		perf_env__single_address_space(session->machines.host.env);
222 
223 	if (!data || perf_data__is_write(data)) {
224 		/*
225 		 * In O_RDONLY mode this will be performed when reading the
226 		 * kernel MMAP event, in perf_event__process_mmap().
227 		 */
228 		if (perf_session__create_kernel_maps(session) < 0)
229 			pr_warning("Cannot read kernel map\n");
230 	}
231 
232 	/*
233 	 * In pipe-mode, evlist is empty until PERF_RECORD_HEADER_ATTR is
234 	 * processed, so perf_evlist__sample_id_all is not meaningful here.
235 	 */
236 	if ((!data || !data->is_pipe) && tool && tool->ordering_requires_timestamps &&
237 	    tool->ordered_events && !perf_evlist__sample_id_all(session->evlist)) {
238 		dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n");
239 		tool->ordered_events = false;
240 	}
241 
242 	return session;
243 
244  out_delete:
245 	perf_session__delete(session);
246  out:
247 	return NULL;
248 }
249 
250 static void perf_session__delete_threads(struct perf_session *session)
251 {
252 	machine__delete_threads(&session->machines.host);
253 }
254 
255 static void perf_session__release_decomp_events(struct perf_session *session)
256 {
257 	struct decomp *next, *decomp;
258 	size_t decomp_len;
259 	next = session->decomp;
260 	decomp_len = session->header.env.comp_mmap_len;
261 	do {
262 		decomp = next;
263 		if (decomp == NULL)
264 			break;
265 		next = decomp->next;
266 		munmap(decomp, decomp_len + sizeof(struct decomp));
267 	} while (1);
268 }
269 
270 void perf_session__delete(struct perf_session *session)
271 {
272 	if (session == NULL)
273 		return;
274 	auxtrace__free(session);
275 	auxtrace_index__free(&session->auxtrace_index);
276 	perf_session__destroy_kernel_maps(session);
277 	perf_session__delete_threads(session);
278 	perf_session__release_decomp_events(session);
279 	perf_env__exit(&session->header.env);
280 	machines__exit(&session->machines);
281 	if (session->data)
282 		perf_data__close(session->data);
283 	free(session);
284 }
285 
286 static int process_event_synth_tracing_data_stub(struct perf_session *session
287 						 __maybe_unused,
288 						 union perf_event *event
289 						 __maybe_unused)
290 {
291 	dump_printf(": unhandled!\n");
292 	return 0;
293 }
294 
295 static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused,
296 					 union perf_event *event __maybe_unused,
297 					 struct perf_evlist **pevlist
298 					 __maybe_unused)
299 {
300 	dump_printf(": unhandled!\n");
301 	return 0;
302 }
303 
304 static int process_event_synth_event_update_stub(struct perf_tool *tool __maybe_unused,
305 						 union perf_event *event __maybe_unused,
306 						 struct perf_evlist **pevlist
307 						 __maybe_unused)
308 {
309 	if (dump_trace)
310 		perf_event__fprintf_event_update(event, stdout);
311 
312 	dump_printf(": unhandled!\n");
313 	return 0;
314 }
315 
316 static int process_event_sample_stub(struct perf_tool *tool __maybe_unused,
317 				     union perf_event *event __maybe_unused,
318 				     struct perf_sample *sample __maybe_unused,
319 				     struct perf_evsel *evsel __maybe_unused,
320 				     struct machine *machine __maybe_unused)
321 {
322 	dump_printf(": unhandled!\n");
323 	return 0;
324 }
325 
326 static int process_event_stub(struct perf_tool *tool __maybe_unused,
327 			      union perf_event *event __maybe_unused,
328 			      struct perf_sample *sample __maybe_unused,
329 			      struct machine *machine __maybe_unused)
330 {
331 	dump_printf(": unhandled!\n");
332 	return 0;
333 }
334 
335 static int process_finished_round_stub(struct perf_tool *tool __maybe_unused,
336 				       union perf_event *event __maybe_unused,
337 				       struct ordered_events *oe __maybe_unused)
338 {
339 	dump_printf(": unhandled!\n");
340 	return 0;
341 }
342 
343 static int process_finished_round(struct perf_tool *tool,
344 				  union perf_event *event,
345 				  struct ordered_events *oe);
346 
347 static int skipn(int fd, off_t n)
348 {
349 	char buf[4096];
350 	ssize_t ret;
351 
352 	while (n > 0) {
353 		ret = read(fd, buf, min(n, (off_t)sizeof(buf)));
354 		if (ret <= 0)
355 			return ret;
356 		n -= ret;
357 	}
358 
359 	return 0;
360 }
361 
362 static s64 process_event_auxtrace_stub(struct perf_session *session __maybe_unused,
363 				       union perf_event *event)
364 {
365 	dump_printf(": unhandled!\n");
366 	if (perf_data__is_pipe(session->data))
367 		skipn(perf_data__fd(session->data), event->auxtrace.size);
368 	return event->auxtrace.size;
369 }
370 
371 static int process_event_op2_stub(struct perf_session *session __maybe_unused,
372 				  union perf_event *event __maybe_unused)
373 {
374 	dump_printf(": unhandled!\n");
375 	return 0;
376 }
377 
378 
379 static
380 int process_event_thread_map_stub(struct perf_session *session __maybe_unused,
381 				  union perf_event *event __maybe_unused)
382 {
383 	if (dump_trace)
384 		perf_event__fprintf_thread_map(event, stdout);
385 
386 	dump_printf(": unhandled!\n");
387 	return 0;
388 }
389 
390 static
391 int process_event_cpu_map_stub(struct perf_session *session __maybe_unused,
392 			       union perf_event *event __maybe_unused)
393 {
394 	if (dump_trace)
395 		perf_event__fprintf_cpu_map(event, stdout);
396 
397 	dump_printf(": unhandled!\n");
398 	return 0;
399 }
400 
401 static
402 int process_event_stat_config_stub(struct perf_session *session __maybe_unused,
403 				   union perf_event *event __maybe_unused)
404 {
405 	if (dump_trace)
406 		perf_event__fprintf_stat_config(event, stdout);
407 
408 	dump_printf(": unhandled!\n");
409 	return 0;
410 }
411 
412 static int process_stat_stub(struct perf_session *perf_session __maybe_unused,
413 			     union perf_event *event)
414 {
415 	if (dump_trace)
416 		perf_event__fprintf_stat(event, stdout);
417 
418 	dump_printf(": unhandled!\n");
419 	return 0;
420 }
421 
422 static int process_stat_round_stub(struct perf_session *perf_session __maybe_unused,
423 				   union perf_event *event)
424 {
425 	if (dump_trace)
426 		perf_event__fprintf_stat_round(event, stdout);
427 
428 	dump_printf(": unhandled!\n");
429 	return 0;
430 }
431 
432 static int perf_session__process_compressed_event_stub(struct perf_session *session __maybe_unused,
433 						       union perf_event *event __maybe_unused,
434 						       u64 file_offset __maybe_unused)
435 {
436        dump_printf(": unhandled!\n");
437        return 0;
438 }
439 
440 void perf_tool__fill_defaults(struct perf_tool *tool)
441 {
442 	if (tool->sample == NULL)
443 		tool->sample = process_event_sample_stub;
444 	if (tool->mmap == NULL)
445 		tool->mmap = process_event_stub;
446 	if (tool->mmap2 == NULL)
447 		tool->mmap2 = process_event_stub;
448 	if (tool->comm == NULL)
449 		tool->comm = process_event_stub;
450 	if (tool->namespaces == NULL)
451 		tool->namespaces = process_event_stub;
452 	if (tool->fork == NULL)
453 		tool->fork = process_event_stub;
454 	if (tool->exit == NULL)
455 		tool->exit = process_event_stub;
456 	if (tool->lost == NULL)
457 		tool->lost = perf_event__process_lost;
458 	if (tool->lost_samples == NULL)
459 		tool->lost_samples = perf_event__process_lost_samples;
460 	if (tool->aux == NULL)
461 		tool->aux = perf_event__process_aux;
462 	if (tool->itrace_start == NULL)
463 		tool->itrace_start = perf_event__process_itrace_start;
464 	if (tool->context_switch == NULL)
465 		tool->context_switch = perf_event__process_switch;
466 	if (tool->ksymbol == NULL)
467 		tool->ksymbol = perf_event__process_ksymbol;
468 	if (tool->bpf_event == NULL)
469 		tool->bpf_event = perf_event__process_bpf_event;
470 	if (tool->read == NULL)
471 		tool->read = process_event_sample_stub;
472 	if (tool->throttle == NULL)
473 		tool->throttle = process_event_stub;
474 	if (tool->unthrottle == NULL)
475 		tool->unthrottle = process_event_stub;
476 	if (tool->attr == NULL)
477 		tool->attr = process_event_synth_attr_stub;
478 	if (tool->event_update == NULL)
479 		tool->event_update = process_event_synth_event_update_stub;
480 	if (tool->tracing_data == NULL)
481 		tool->tracing_data = process_event_synth_tracing_data_stub;
482 	if (tool->build_id == NULL)
483 		tool->build_id = process_event_op2_stub;
484 	if (tool->finished_round == NULL) {
485 		if (tool->ordered_events)
486 			tool->finished_round = process_finished_round;
487 		else
488 			tool->finished_round = process_finished_round_stub;
489 	}
490 	if (tool->id_index == NULL)
491 		tool->id_index = process_event_op2_stub;
492 	if (tool->auxtrace_info == NULL)
493 		tool->auxtrace_info = process_event_op2_stub;
494 	if (tool->auxtrace == NULL)
495 		tool->auxtrace = process_event_auxtrace_stub;
496 	if (tool->auxtrace_error == NULL)
497 		tool->auxtrace_error = process_event_op2_stub;
498 	if (tool->thread_map == NULL)
499 		tool->thread_map = process_event_thread_map_stub;
500 	if (tool->cpu_map == NULL)
501 		tool->cpu_map = process_event_cpu_map_stub;
502 	if (tool->stat_config == NULL)
503 		tool->stat_config = process_event_stat_config_stub;
504 	if (tool->stat == NULL)
505 		tool->stat = process_stat_stub;
506 	if (tool->stat_round == NULL)
507 		tool->stat_round = process_stat_round_stub;
508 	if (tool->time_conv == NULL)
509 		tool->time_conv = process_event_op2_stub;
510 	if (tool->feature == NULL)
511 		tool->feature = process_event_op2_stub;
512 	if (tool->compressed == NULL)
513 		tool->compressed = perf_session__process_compressed_event;
514 }
515 
516 static void swap_sample_id_all(union perf_event *event, void *data)
517 {
518 	void *end = (void *) event + event->header.size;
519 	int size = end - data;
520 
521 	BUG_ON(size % sizeof(u64));
522 	mem_bswap_64(data, size);
523 }
524 
525 static void perf_event__all64_swap(union perf_event *event,
526 				   bool sample_id_all __maybe_unused)
527 {
528 	struct perf_event_header *hdr = &event->header;
529 	mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr));
530 }
531 
532 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all)
533 {
534 	event->comm.pid = bswap_32(event->comm.pid);
535 	event->comm.tid = bswap_32(event->comm.tid);
536 
537 	if (sample_id_all) {
538 		void *data = &event->comm.comm;
539 
540 		data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
541 		swap_sample_id_all(event, data);
542 	}
543 }
544 
545 static void perf_event__mmap_swap(union perf_event *event,
546 				  bool sample_id_all)
547 {
548 	event->mmap.pid	  = bswap_32(event->mmap.pid);
549 	event->mmap.tid	  = bswap_32(event->mmap.tid);
550 	event->mmap.start = bswap_64(event->mmap.start);
551 	event->mmap.len	  = bswap_64(event->mmap.len);
552 	event->mmap.pgoff = bswap_64(event->mmap.pgoff);
553 
554 	if (sample_id_all) {
555 		void *data = &event->mmap.filename;
556 
557 		data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
558 		swap_sample_id_all(event, data);
559 	}
560 }
561 
562 static void perf_event__mmap2_swap(union perf_event *event,
563 				  bool sample_id_all)
564 {
565 	event->mmap2.pid   = bswap_32(event->mmap2.pid);
566 	event->mmap2.tid   = bswap_32(event->mmap2.tid);
567 	event->mmap2.start = bswap_64(event->mmap2.start);
568 	event->mmap2.len   = bswap_64(event->mmap2.len);
569 	event->mmap2.pgoff = bswap_64(event->mmap2.pgoff);
570 	event->mmap2.maj   = bswap_32(event->mmap2.maj);
571 	event->mmap2.min   = bswap_32(event->mmap2.min);
572 	event->mmap2.ino   = bswap_64(event->mmap2.ino);
573 
574 	if (sample_id_all) {
575 		void *data = &event->mmap2.filename;
576 
577 		data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
578 		swap_sample_id_all(event, data);
579 	}
580 }
581 static void perf_event__task_swap(union perf_event *event, bool sample_id_all)
582 {
583 	event->fork.pid	 = bswap_32(event->fork.pid);
584 	event->fork.tid	 = bswap_32(event->fork.tid);
585 	event->fork.ppid = bswap_32(event->fork.ppid);
586 	event->fork.ptid = bswap_32(event->fork.ptid);
587 	event->fork.time = bswap_64(event->fork.time);
588 
589 	if (sample_id_all)
590 		swap_sample_id_all(event, &event->fork + 1);
591 }
592 
593 static void perf_event__read_swap(union perf_event *event, bool sample_id_all)
594 {
595 	event->read.pid		 = bswap_32(event->read.pid);
596 	event->read.tid		 = bswap_32(event->read.tid);
597 	event->read.value	 = bswap_64(event->read.value);
598 	event->read.time_enabled = bswap_64(event->read.time_enabled);
599 	event->read.time_running = bswap_64(event->read.time_running);
600 	event->read.id		 = bswap_64(event->read.id);
601 
602 	if (sample_id_all)
603 		swap_sample_id_all(event, &event->read + 1);
604 }
605 
606 static void perf_event__aux_swap(union perf_event *event, bool sample_id_all)
607 {
608 	event->aux.aux_offset = bswap_64(event->aux.aux_offset);
609 	event->aux.aux_size   = bswap_64(event->aux.aux_size);
610 	event->aux.flags      = bswap_64(event->aux.flags);
611 
612 	if (sample_id_all)
613 		swap_sample_id_all(event, &event->aux + 1);
614 }
615 
616 static void perf_event__itrace_start_swap(union perf_event *event,
617 					  bool sample_id_all)
618 {
619 	event->itrace_start.pid	 = bswap_32(event->itrace_start.pid);
620 	event->itrace_start.tid	 = bswap_32(event->itrace_start.tid);
621 
622 	if (sample_id_all)
623 		swap_sample_id_all(event, &event->itrace_start + 1);
624 }
625 
626 static void perf_event__switch_swap(union perf_event *event, bool sample_id_all)
627 {
628 	if (event->header.type == PERF_RECORD_SWITCH_CPU_WIDE) {
629 		event->context_switch.next_prev_pid =
630 				bswap_32(event->context_switch.next_prev_pid);
631 		event->context_switch.next_prev_tid =
632 				bswap_32(event->context_switch.next_prev_tid);
633 	}
634 
635 	if (sample_id_all)
636 		swap_sample_id_all(event, &event->context_switch + 1);
637 }
638 
639 static void perf_event__throttle_swap(union perf_event *event,
640 				      bool sample_id_all)
641 {
642 	event->throttle.time	  = bswap_64(event->throttle.time);
643 	event->throttle.id	  = bswap_64(event->throttle.id);
644 	event->throttle.stream_id = bswap_64(event->throttle.stream_id);
645 
646 	if (sample_id_all)
647 		swap_sample_id_all(event, &event->throttle + 1);
648 }
649 
650 static void perf_event__namespaces_swap(union perf_event *event,
651 					bool sample_id_all)
652 {
653 	u64 i;
654 
655 	event->namespaces.pid		= bswap_32(event->namespaces.pid);
656 	event->namespaces.tid		= bswap_32(event->namespaces.tid);
657 	event->namespaces.nr_namespaces	= bswap_64(event->namespaces.nr_namespaces);
658 
659 	for (i = 0; i < event->namespaces.nr_namespaces; i++) {
660 		struct perf_ns_link_info *ns = &event->namespaces.link_info[i];
661 
662 		ns->dev = bswap_64(ns->dev);
663 		ns->ino = bswap_64(ns->ino);
664 	}
665 
666 	if (sample_id_all)
667 		swap_sample_id_all(event, &event->namespaces.link_info[i]);
668 }
669 
670 static u8 revbyte(u8 b)
671 {
672 	int rev = (b >> 4) | ((b & 0xf) << 4);
673 	rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2);
674 	rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1);
675 	return (u8) rev;
676 }
677 
678 /*
679  * XXX this is hack in attempt to carry flags bitfield
680  * through endian village. ABI says:
681  *
682  * Bit-fields are allocated from right to left (least to most significant)
683  * on little-endian implementations and from left to right (most to least
684  * significant) on big-endian implementations.
685  *
686  * The above seems to be byte specific, so we need to reverse each
687  * byte of the bitfield. 'Internet' also says this might be implementation
688  * specific and we probably need proper fix and carry perf_event_attr
689  * bitfield flags in separate data file FEAT_ section. Thought this seems
690  * to work for now.
691  */
692 static void swap_bitfield(u8 *p, unsigned len)
693 {
694 	unsigned i;
695 
696 	for (i = 0; i < len; i++) {
697 		*p = revbyte(*p);
698 		p++;
699 	}
700 }
701 
702 /* exported for swapping attributes in file header */
703 void perf_event__attr_swap(struct perf_event_attr *attr)
704 {
705 	attr->type		= bswap_32(attr->type);
706 	attr->size		= bswap_32(attr->size);
707 
708 #define bswap_safe(f, n) 					\
709 	(attr->size > (offsetof(struct perf_event_attr, f) + 	\
710 		       sizeof(attr->f) * (n)))
711 #define bswap_field(f, sz) 			\
712 do { 						\
713 	if (bswap_safe(f, 0))			\
714 		attr->f = bswap_##sz(attr->f);	\
715 } while(0)
716 #define bswap_field_16(f) bswap_field(f, 16)
717 #define bswap_field_32(f) bswap_field(f, 32)
718 #define bswap_field_64(f) bswap_field(f, 64)
719 
720 	bswap_field_64(config);
721 	bswap_field_64(sample_period);
722 	bswap_field_64(sample_type);
723 	bswap_field_64(read_format);
724 	bswap_field_32(wakeup_events);
725 	bswap_field_32(bp_type);
726 	bswap_field_64(bp_addr);
727 	bswap_field_64(bp_len);
728 	bswap_field_64(branch_sample_type);
729 	bswap_field_64(sample_regs_user);
730 	bswap_field_32(sample_stack_user);
731 	bswap_field_32(aux_watermark);
732 	bswap_field_16(sample_max_stack);
733 
734 	/*
735 	 * After read_format are bitfields. Check read_format because
736 	 * we are unable to use offsetof on bitfield.
737 	 */
738 	if (bswap_safe(read_format, 1))
739 		swap_bitfield((u8 *) (&attr->read_format + 1),
740 			      sizeof(u64));
741 #undef bswap_field_64
742 #undef bswap_field_32
743 #undef bswap_field
744 #undef bswap_safe
745 }
746 
747 static void perf_event__hdr_attr_swap(union perf_event *event,
748 				      bool sample_id_all __maybe_unused)
749 {
750 	size_t size;
751 
752 	perf_event__attr_swap(&event->attr.attr);
753 
754 	size = event->header.size;
755 	size -= (void *)&event->attr.id - (void *)event;
756 	mem_bswap_64(event->attr.id, size);
757 }
758 
759 static void perf_event__event_update_swap(union perf_event *event,
760 					  bool sample_id_all __maybe_unused)
761 {
762 	event->event_update.type = bswap_64(event->event_update.type);
763 	event->event_update.id   = bswap_64(event->event_update.id);
764 }
765 
766 static void perf_event__event_type_swap(union perf_event *event,
767 					bool sample_id_all __maybe_unused)
768 {
769 	event->event_type.event_type.event_id =
770 		bswap_64(event->event_type.event_type.event_id);
771 }
772 
773 static void perf_event__tracing_data_swap(union perf_event *event,
774 					  bool sample_id_all __maybe_unused)
775 {
776 	event->tracing_data.size = bswap_32(event->tracing_data.size);
777 }
778 
779 static void perf_event__auxtrace_info_swap(union perf_event *event,
780 					   bool sample_id_all __maybe_unused)
781 {
782 	size_t size;
783 
784 	event->auxtrace_info.type = bswap_32(event->auxtrace_info.type);
785 
786 	size = event->header.size;
787 	size -= (void *)&event->auxtrace_info.priv - (void *)event;
788 	mem_bswap_64(event->auxtrace_info.priv, size);
789 }
790 
791 static void perf_event__auxtrace_swap(union perf_event *event,
792 				      bool sample_id_all __maybe_unused)
793 {
794 	event->auxtrace.size      = bswap_64(event->auxtrace.size);
795 	event->auxtrace.offset    = bswap_64(event->auxtrace.offset);
796 	event->auxtrace.reference = bswap_64(event->auxtrace.reference);
797 	event->auxtrace.idx       = bswap_32(event->auxtrace.idx);
798 	event->auxtrace.tid       = bswap_32(event->auxtrace.tid);
799 	event->auxtrace.cpu       = bswap_32(event->auxtrace.cpu);
800 }
801 
802 static void perf_event__auxtrace_error_swap(union perf_event *event,
803 					    bool sample_id_all __maybe_unused)
804 {
805 	event->auxtrace_error.type = bswap_32(event->auxtrace_error.type);
806 	event->auxtrace_error.code = bswap_32(event->auxtrace_error.code);
807 	event->auxtrace_error.cpu  = bswap_32(event->auxtrace_error.cpu);
808 	event->auxtrace_error.pid  = bswap_32(event->auxtrace_error.pid);
809 	event->auxtrace_error.tid  = bswap_32(event->auxtrace_error.tid);
810 	event->auxtrace_error.fmt  = bswap_32(event->auxtrace_error.fmt);
811 	event->auxtrace_error.ip   = bswap_64(event->auxtrace_error.ip);
812 	if (event->auxtrace_error.fmt)
813 		event->auxtrace_error.time = bswap_64(event->auxtrace_error.time);
814 }
815 
816 static void perf_event__thread_map_swap(union perf_event *event,
817 					bool sample_id_all __maybe_unused)
818 {
819 	unsigned i;
820 
821 	event->thread_map.nr = bswap_64(event->thread_map.nr);
822 
823 	for (i = 0; i < event->thread_map.nr; i++)
824 		event->thread_map.entries[i].pid = bswap_64(event->thread_map.entries[i].pid);
825 }
826 
827 static void perf_event__cpu_map_swap(union perf_event *event,
828 				     bool sample_id_all __maybe_unused)
829 {
830 	struct cpu_map_data *data = &event->cpu_map.data;
831 	struct cpu_map_entries *cpus;
832 	struct cpu_map_mask *mask;
833 	unsigned i;
834 
835 	data->type = bswap_64(data->type);
836 
837 	switch (data->type) {
838 	case PERF_CPU_MAP__CPUS:
839 		cpus = (struct cpu_map_entries *)data->data;
840 
841 		cpus->nr = bswap_16(cpus->nr);
842 
843 		for (i = 0; i < cpus->nr; i++)
844 			cpus->cpu[i] = bswap_16(cpus->cpu[i]);
845 		break;
846 	case PERF_CPU_MAP__MASK:
847 		mask = (struct cpu_map_mask *) data->data;
848 
849 		mask->nr = bswap_16(mask->nr);
850 		mask->long_size = bswap_16(mask->long_size);
851 
852 		switch (mask->long_size) {
853 		case 4: mem_bswap_32(&mask->mask, mask->nr); break;
854 		case 8: mem_bswap_64(&mask->mask, mask->nr); break;
855 		default:
856 			pr_err("cpu_map swap: unsupported long size\n");
857 		}
858 	default:
859 		break;
860 	}
861 }
862 
863 static void perf_event__stat_config_swap(union perf_event *event,
864 					 bool sample_id_all __maybe_unused)
865 {
866 	u64 size;
867 
868 	size  = event->stat_config.nr * sizeof(event->stat_config.data[0]);
869 	size += 1; /* nr item itself */
870 	mem_bswap_64(&event->stat_config.nr, size);
871 }
872 
873 static void perf_event__stat_swap(union perf_event *event,
874 				  bool sample_id_all __maybe_unused)
875 {
876 	event->stat.id     = bswap_64(event->stat.id);
877 	event->stat.thread = bswap_32(event->stat.thread);
878 	event->stat.cpu    = bswap_32(event->stat.cpu);
879 	event->stat.val    = bswap_64(event->stat.val);
880 	event->stat.ena    = bswap_64(event->stat.ena);
881 	event->stat.run    = bswap_64(event->stat.run);
882 }
883 
884 static void perf_event__stat_round_swap(union perf_event *event,
885 					bool sample_id_all __maybe_unused)
886 {
887 	event->stat_round.type = bswap_64(event->stat_round.type);
888 	event->stat_round.time = bswap_64(event->stat_round.time);
889 }
890 
891 typedef void (*perf_event__swap_op)(union perf_event *event,
892 				    bool sample_id_all);
893 
894 static perf_event__swap_op perf_event__swap_ops[] = {
895 	[PERF_RECORD_MMAP]		  = perf_event__mmap_swap,
896 	[PERF_RECORD_MMAP2]		  = perf_event__mmap2_swap,
897 	[PERF_RECORD_COMM]		  = perf_event__comm_swap,
898 	[PERF_RECORD_FORK]		  = perf_event__task_swap,
899 	[PERF_RECORD_EXIT]		  = perf_event__task_swap,
900 	[PERF_RECORD_LOST]		  = perf_event__all64_swap,
901 	[PERF_RECORD_READ]		  = perf_event__read_swap,
902 	[PERF_RECORD_THROTTLE]		  = perf_event__throttle_swap,
903 	[PERF_RECORD_UNTHROTTLE]	  = perf_event__throttle_swap,
904 	[PERF_RECORD_SAMPLE]		  = perf_event__all64_swap,
905 	[PERF_RECORD_AUX]		  = perf_event__aux_swap,
906 	[PERF_RECORD_ITRACE_START]	  = perf_event__itrace_start_swap,
907 	[PERF_RECORD_LOST_SAMPLES]	  = perf_event__all64_swap,
908 	[PERF_RECORD_SWITCH]		  = perf_event__switch_swap,
909 	[PERF_RECORD_SWITCH_CPU_WIDE]	  = perf_event__switch_swap,
910 	[PERF_RECORD_NAMESPACES]	  = perf_event__namespaces_swap,
911 	[PERF_RECORD_HEADER_ATTR]	  = perf_event__hdr_attr_swap,
912 	[PERF_RECORD_HEADER_EVENT_TYPE]	  = perf_event__event_type_swap,
913 	[PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
914 	[PERF_RECORD_HEADER_BUILD_ID]	  = NULL,
915 	[PERF_RECORD_ID_INDEX]		  = perf_event__all64_swap,
916 	[PERF_RECORD_AUXTRACE_INFO]	  = perf_event__auxtrace_info_swap,
917 	[PERF_RECORD_AUXTRACE]		  = perf_event__auxtrace_swap,
918 	[PERF_RECORD_AUXTRACE_ERROR]	  = perf_event__auxtrace_error_swap,
919 	[PERF_RECORD_THREAD_MAP]	  = perf_event__thread_map_swap,
920 	[PERF_RECORD_CPU_MAP]		  = perf_event__cpu_map_swap,
921 	[PERF_RECORD_STAT_CONFIG]	  = perf_event__stat_config_swap,
922 	[PERF_RECORD_STAT]		  = perf_event__stat_swap,
923 	[PERF_RECORD_STAT_ROUND]	  = perf_event__stat_round_swap,
924 	[PERF_RECORD_EVENT_UPDATE]	  = perf_event__event_update_swap,
925 	[PERF_RECORD_TIME_CONV]		  = perf_event__all64_swap,
926 	[PERF_RECORD_HEADER_MAX]	  = NULL,
927 };
928 
929 /*
930  * When perf record finishes a pass on every buffers, it records this pseudo
931  * event.
932  * We record the max timestamp t found in the pass n.
933  * Assuming these timestamps are monotonic across cpus, we know that if
934  * a buffer still has events with timestamps below t, they will be all
935  * available and then read in the pass n + 1.
936  * Hence when we start to read the pass n + 2, we can safely flush every
937  * events with timestamps below t.
938  *
939  *    ============ PASS n =================
940  *       CPU 0         |   CPU 1
941  *                     |
942  *    cnt1 timestamps  |   cnt2 timestamps
943  *          1          |         2
944  *          2          |         3
945  *          -          |         4  <--- max recorded
946  *
947  *    ============ PASS n + 1 ==============
948  *       CPU 0         |   CPU 1
949  *                     |
950  *    cnt1 timestamps  |   cnt2 timestamps
951  *          3          |         5
952  *          4          |         6
953  *          5          |         7 <---- max recorded
954  *
955  *      Flush every events below timestamp 4
956  *
957  *    ============ PASS n + 2 ==============
958  *       CPU 0         |   CPU 1
959  *                     |
960  *    cnt1 timestamps  |   cnt2 timestamps
961  *          6          |         8
962  *          7          |         9
963  *          -          |         10
964  *
965  *      Flush every events below timestamp 7
966  *      etc...
967  */
968 static int process_finished_round(struct perf_tool *tool __maybe_unused,
969 				  union perf_event *event __maybe_unused,
970 				  struct ordered_events *oe)
971 {
972 	if (dump_trace)
973 		fprintf(stdout, "\n");
974 	return ordered_events__flush(oe, OE_FLUSH__ROUND);
975 }
976 
977 int perf_session__queue_event(struct perf_session *s, union perf_event *event,
978 			      u64 timestamp, u64 file_offset)
979 {
980 	return ordered_events__queue(&s->ordered_events, event, timestamp, file_offset);
981 }
982 
983 static void callchain__lbr_callstack_printf(struct perf_sample *sample)
984 {
985 	struct ip_callchain *callchain = sample->callchain;
986 	struct branch_stack *lbr_stack = sample->branch_stack;
987 	u64 kernel_callchain_nr = callchain->nr;
988 	unsigned int i;
989 
990 	for (i = 0; i < kernel_callchain_nr; i++) {
991 		if (callchain->ips[i] == PERF_CONTEXT_USER)
992 			break;
993 	}
994 
995 	if ((i != kernel_callchain_nr) && lbr_stack->nr) {
996 		u64 total_nr;
997 		/*
998 		 * LBR callstack can only get user call chain,
999 		 * i is kernel call chain number,
1000 		 * 1 is PERF_CONTEXT_USER.
1001 		 *
1002 		 * The user call chain is stored in LBR registers.
1003 		 * LBR are pair registers. The caller is stored
1004 		 * in "from" register, while the callee is stored
1005 		 * in "to" register.
1006 		 * For example, there is a call stack
1007 		 * "A"->"B"->"C"->"D".
1008 		 * The LBR registers will recorde like
1009 		 * "C"->"D", "B"->"C", "A"->"B".
1010 		 * So only the first "to" register and all "from"
1011 		 * registers are needed to construct the whole stack.
1012 		 */
1013 		total_nr = i + 1 + lbr_stack->nr + 1;
1014 		kernel_callchain_nr = i + 1;
1015 
1016 		printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr);
1017 
1018 		for (i = 0; i < kernel_callchain_nr; i++)
1019 			printf("..... %2d: %016" PRIx64 "\n",
1020 			       i, callchain->ips[i]);
1021 
1022 		printf("..... %2d: %016" PRIx64 "\n",
1023 		       (int)(kernel_callchain_nr), lbr_stack->entries[0].to);
1024 		for (i = 0; i < lbr_stack->nr; i++)
1025 			printf("..... %2d: %016" PRIx64 "\n",
1026 			       (int)(i + kernel_callchain_nr + 1), lbr_stack->entries[i].from);
1027 	}
1028 }
1029 
1030 static void callchain__printf(struct perf_evsel *evsel,
1031 			      struct perf_sample *sample)
1032 {
1033 	unsigned int i;
1034 	struct ip_callchain *callchain = sample->callchain;
1035 
1036 	if (perf_evsel__has_branch_callstack(evsel))
1037 		callchain__lbr_callstack_printf(sample);
1038 
1039 	printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr);
1040 
1041 	for (i = 0; i < callchain->nr; i++)
1042 		printf("..... %2d: %016" PRIx64 "\n",
1043 		       i, callchain->ips[i]);
1044 }
1045 
1046 static void branch_stack__printf(struct perf_sample *sample)
1047 {
1048 	uint64_t i;
1049 
1050 	printf("... branch stack: nr:%" PRIu64 "\n", sample->branch_stack->nr);
1051 
1052 	for (i = 0; i < sample->branch_stack->nr; i++) {
1053 		struct branch_entry *e = &sample->branch_stack->entries[i];
1054 
1055 		printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x\n",
1056 			i, e->from, e->to,
1057 			(unsigned short)e->flags.cycles,
1058 			e->flags.mispred ? "M" : " ",
1059 			e->flags.predicted ? "P" : " ",
1060 			e->flags.abort ? "A" : " ",
1061 			e->flags.in_tx ? "T" : " ",
1062 			(unsigned)e->flags.reserved);
1063 	}
1064 }
1065 
1066 static void regs_dump__printf(u64 mask, u64 *regs)
1067 {
1068 	unsigned rid, i = 0;
1069 
1070 	for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) {
1071 		u64 val = regs[i++];
1072 
1073 		printf(".... %-5s 0x%" PRIx64 "\n",
1074 		       perf_reg_name(rid), val);
1075 	}
1076 }
1077 
1078 static const char *regs_abi[] = {
1079 	[PERF_SAMPLE_REGS_ABI_NONE] = "none",
1080 	[PERF_SAMPLE_REGS_ABI_32] = "32-bit",
1081 	[PERF_SAMPLE_REGS_ABI_64] = "64-bit",
1082 };
1083 
1084 static inline const char *regs_dump_abi(struct regs_dump *d)
1085 {
1086 	if (d->abi > PERF_SAMPLE_REGS_ABI_64)
1087 		return "unknown";
1088 
1089 	return regs_abi[d->abi];
1090 }
1091 
1092 static void regs__printf(const char *type, struct regs_dump *regs)
1093 {
1094 	u64 mask = regs->mask;
1095 
1096 	printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n",
1097 	       type,
1098 	       mask,
1099 	       regs_dump_abi(regs));
1100 
1101 	regs_dump__printf(mask, regs->regs);
1102 }
1103 
1104 static void regs_user__printf(struct perf_sample *sample)
1105 {
1106 	struct regs_dump *user_regs = &sample->user_regs;
1107 
1108 	if (user_regs->regs)
1109 		regs__printf("user", user_regs);
1110 }
1111 
1112 static void regs_intr__printf(struct perf_sample *sample)
1113 {
1114 	struct regs_dump *intr_regs = &sample->intr_regs;
1115 
1116 	if (intr_regs->regs)
1117 		regs__printf("intr", intr_regs);
1118 }
1119 
1120 static void stack_user__printf(struct stack_dump *dump)
1121 {
1122 	printf("... ustack: size %" PRIu64 ", offset 0x%x\n",
1123 	       dump->size, dump->offset);
1124 }
1125 
1126 static void perf_evlist__print_tstamp(struct perf_evlist *evlist,
1127 				       union perf_event *event,
1128 				       struct perf_sample *sample)
1129 {
1130 	u64 sample_type = __perf_evlist__combined_sample_type(evlist);
1131 
1132 	if (event->header.type != PERF_RECORD_SAMPLE &&
1133 	    !perf_evlist__sample_id_all(evlist)) {
1134 		fputs("-1 -1 ", stdout);
1135 		return;
1136 	}
1137 
1138 	if ((sample_type & PERF_SAMPLE_CPU))
1139 		printf("%u ", sample->cpu);
1140 
1141 	if (sample_type & PERF_SAMPLE_TIME)
1142 		printf("%" PRIu64 " ", sample->time);
1143 }
1144 
1145 static void sample_read__printf(struct perf_sample *sample, u64 read_format)
1146 {
1147 	printf("... sample_read:\n");
1148 
1149 	if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1150 		printf("...... time enabled %016" PRIx64 "\n",
1151 		       sample->read.time_enabled);
1152 
1153 	if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1154 		printf("...... time running %016" PRIx64 "\n",
1155 		       sample->read.time_running);
1156 
1157 	if (read_format & PERF_FORMAT_GROUP) {
1158 		u64 i;
1159 
1160 		printf(".... group nr %" PRIu64 "\n", sample->read.group.nr);
1161 
1162 		for (i = 0; i < sample->read.group.nr; i++) {
1163 			struct sample_read_value *value;
1164 
1165 			value = &sample->read.group.values[i];
1166 			printf("..... id %016" PRIx64
1167 			       ", value %016" PRIx64 "\n",
1168 			       value->id, value->value);
1169 		}
1170 	} else
1171 		printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n",
1172 			sample->read.one.id, sample->read.one.value);
1173 }
1174 
1175 static void dump_event(struct perf_evlist *evlist, union perf_event *event,
1176 		       u64 file_offset, struct perf_sample *sample)
1177 {
1178 	if (!dump_trace)
1179 		return;
1180 
1181 	printf("\n%#" PRIx64 " [%#x]: event: %d\n",
1182 	       file_offset, event->header.size, event->header.type);
1183 
1184 	trace_event(event);
1185 	if (event->header.type == PERF_RECORD_SAMPLE && evlist->trace_event_sample_raw)
1186 		evlist->trace_event_sample_raw(evlist, event, sample);
1187 
1188 	if (sample)
1189 		perf_evlist__print_tstamp(evlist, event, sample);
1190 
1191 	printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset,
1192 	       event->header.size, perf_event__name(event->header.type));
1193 }
1194 
1195 static void dump_sample(struct perf_evsel *evsel, union perf_event *event,
1196 			struct perf_sample *sample)
1197 {
1198 	u64 sample_type;
1199 
1200 	if (!dump_trace)
1201 		return;
1202 
1203 	printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n",
1204 	       event->header.misc, sample->pid, sample->tid, sample->ip,
1205 	       sample->period, sample->addr);
1206 
1207 	sample_type = evsel->attr.sample_type;
1208 
1209 	if (evsel__has_callchain(evsel))
1210 		callchain__printf(evsel, sample);
1211 
1212 	if ((sample_type & PERF_SAMPLE_BRANCH_STACK) && !perf_evsel__has_branch_callstack(evsel))
1213 		branch_stack__printf(sample);
1214 
1215 	if (sample_type & PERF_SAMPLE_REGS_USER)
1216 		regs_user__printf(sample);
1217 
1218 	if (sample_type & PERF_SAMPLE_REGS_INTR)
1219 		regs_intr__printf(sample);
1220 
1221 	if (sample_type & PERF_SAMPLE_STACK_USER)
1222 		stack_user__printf(&sample->user_stack);
1223 
1224 	if (sample_type & PERF_SAMPLE_WEIGHT)
1225 		printf("... weight: %" PRIu64 "\n", sample->weight);
1226 
1227 	if (sample_type & PERF_SAMPLE_DATA_SRC)
1228 		printf(" . data_src: 0x%"PRIx64"\n", sample->data_src);
1229 
1230 	if (sample_type & PERF_SAMPLE_PHYS_ADDR)
1231 		printf(" .. phys_addr: 0x%"PRIx64"\n", sample->phys_addr);
1232 
1233 	if (sample_type & PERF_SAMPLE_TRANSACTION)
1234 		printf("... transaction: %" PRIx64 "\n", sample->transaction);
1235 
1236 	if (sample_type & PERF_SAMPLE_READ)
1237 		sample_read__printf(sample, evsel->attr.read_format);
1238 }
1239 
1240 static void dump_read(struct perf_evsel *evsel, union perf_event *event)
1241 {
1242 	struct read_event *read_event = &event->read;
1243 	u64 read_format;
1244 
1245 	if (!dump_trace)
1246 		return;
1247 
1248 	printf(": %d %d %s %" PRIu64 "\n", event->read.pid, event->read.tid,
1249 	       evsel ? perf_evsel__name(evsel) : "FAIL",
1250 	       event->read.value);
1251 
1252 	read_format = evsel->attr.read_format;
1253 
1254 	if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1255 		printf("... time enabled : %" PRIu64 "\n", read_event->time_enabled);
1256 
1257 	if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1258 		printf("... time running : %" PRIu64 "\n", read_event->time_running);
1259 
1260 	if (read_format & PERF_FORMAT_ID)
1261 		printf("... id           : %" PRIu64 "\n", read_event->id);
1262 }
1263 
1264 static struct machine *machines__find_for_cpumode(struct machines *machines,
1265 					       union perf_event *event,
1266 					       struct perf_sample *sample)
1267 {
1268 	struct machine *machine;
1269 
1270 	if (perf_guest &&
1271 	    ((sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL) ||
1272 	     (sample->cpumode == PERF_RECORD_MISC_GUEST_USER))) {
1273 		u32 pid;
1274 
1275 		if (event->header.type == PERF_RECORD_MMAP
1276 		    || event->header.type == PERF_RECORD_MMAP2)
1277 			pid = event->mmap.pid;
1278 		else
1279 			pid = sample->pid;
1280 
1281 		machine = machines__find(machines, pid);
1282 		if (!machine)
1283 			machine = machines__findnew(machines, DEFAULT_GUEST_KERNEL_ID);
1284 		return machine;
1285 	}
1286 
1287 	return &machines->host;
1288 }
1289 
1290 static int deliver_sample_value(struct perf_evlist *evlist,
1291 				struct perf_tool *tool,
1292 				union perf_event *event,
1293 				struct perf_sample *sample,
1294 				struct sample_read_value *v,
1295 				struct machine *machine)
1296 {
1297 	struct perf_sample_id *sid = perf_evlist__id2sid(evlist, v->id);
1298 
1299 	if (sid) {
1300 		sample->id     = v->id;
1301 		sample->period = v->value - sid->period;
1302 		sid->period    = v->value;
1303 	}
1304 
1305 	if (!sid || sid->evsel == NULL) {
1306 		++evlist->stats.nr_unknown_id;
1307 		return 0;
1308 	}
1309 
1310 	/*
1311 	 * There's no reason to deliver sample
1312 	 * for zero period, bail out.
1313 	 */
1314 	if (!sample->period)
1315 		return 0;
1316 
1317 	return tool->sample(tool, event, sample, sid->evsel, machine);
1318 }
1319 
1320 static int deliver_sample_group(struct perf_evlist *evlist,
1321 				struct perf_tool *tool,
1322 				union  perf_event *event,
1323 				struct perf_sample *sample,
1324 				struct machine *machine)
1325 {
1326 	int ret = -EINVAL;
1327 	u64 i;
1328 
1329 	for (i = 0; i < sample->read.group.nr; i++) {
1330 		ret = deliver_sample_value(evlist, tool, event, sample,
1331 					   &sample->read.group.values[i],
1332 					   machine);
1333 		if (ret)
1334 			break;
1335 	}
1336 
1337 	return ret;
1338 }
1339 
1340 static int
1341  perf_evlist__deliver_sample(struct perf_evlist *evlist,
1342 			     struct perf_tool *tool,
1343 			     union  perf_event *event,
1344 			     struct perf_sample *sample,
1345 			     struct perf_evsel *evsel,
1346 			     struct machine *machine)
1347 {
1348 	/* We know evsel != NULL. */
1349 	u64 sample_type = evsel->attr.sample_type;
1350 	u64 read_format = evsel->attr.read_format;
1351 
1352 	/* Standard sample delivery. */
1353 	if (!(sample_type & PERF_SAMPLE_READ))
1354 		return tool->sample(tool, event, sample, evsel, machine);
1355 
1356 	/* For PERF_SAMPLE_READ we have either single or group mode. */
1357 	if (read_format & PERF_FORMAT_GROUP)
1358 		return deliver_sample_group(evlist, tool, event, sample,
1359 					    machine);
1360 	else
1361 		return deliver_sample_value(evlist, tool, event, sample,
1362 					    &sample->read.one, machine);
1363 }
1364 
1365 static int machines__deliver_event(struct machines *machines,
1366 				   struct perf_evlist *evlist,
1367 				   union perf_event *event,
1368 				   struct perf_sample *sample,
1369 				   struct perf_tool *tool, u64 file_offset)
1370 {
1371 	struct perf_evsel *evsel;
1372 	struct machine *machine;
1373 
1374 	dump_event(evlist, event, file_offset, sample);
1375 
1376 	evsel = perf_evlist__id2evsel(evlist, sample->id);
1377 
1378 	machine = machines__find_for_cpumode(machines, event, sample);
1379 
1380 	switch (event->header.type) {
1381 	case PERF_RECORD_SAMPLE:
1382 		if (evsel == NULL) {
1383 			++evlist->stats.nr_unknown_id;
1384 			return 0;
1385 		}
1386 		dump_sample(evsel, event, sample);
1387 		if (machine == NULL) {
1388 			++evlist->stats.nr_unprocessable_samples;
1389 			return 0;
1390 		}
1391 		return perf_evlist__deliver_sample(evlist, tool, event, sample, evsel, machine);
1392 	case PERF_RECORD_MMAP:
1393 		return tool->mmap(tool, event, sample, machine);
1394 	case PERF_RECORD_MMAP2:
1395 		if (event->header.misc & PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT)
1396 			++evlist->stats.nr_proc_map_timeout;
1397 		return tool->mmap2(tool, event, sample, machine);
1398 	case PERF_RECORD_COMM:
1399 		return tool->comm(tool, event, sample, machine);
1400 	case PERF_RECORD_NAMESPACES:
1401 		return tool->namespaces(tool, event, sample, machine);
1402 	case PERF_RECORD_FORK:
1403 		return tool->fork(tool, event, sample, machine);
1404 	case PERF_RECORD_EXIT:
1405 		return tool->exit(tool, event, sample, machine);
1406 	case PERF_RECORD_LOST:
1407 		if (tool->lost == perf_event__process_lost)
1408 			evlist->stats.total_lost += event->lost.lost;
1409 		return tool->lost(tool, event, sample, machine);
1410 	case PERF_RECORD_LOST_SAMPLES:
1411 		if (tool->lost_samples == perf_event__process_lost_samples)
1412 			evlist->stats.total_lost_samples += event->lost_samples.lost;
1413 		return tool->lost_samples(tool, event, sample, machine);
1414 	case PERF_RECORD_READ:
1415 		dump_read(evsel, event);
1416 		return tool->read(tool, event, sample, evsel, machine);
1417 	case PERF_RECORD_THROTTLE:
1418 		return tool->throttle(tool, event, sample, machine);
1419 	case PERF_RECORD_UNTHROTTLE:
1420 		return tool->unthrottle(tool, event, sample, machine);
1421 	case PERF_RECORD_AUX:
1422 		if (tool->aux == perf_event__process_aux) {
1423 			if (event->aux.flags & PERF_AUX_FLAG_TRUNCATED)
1424 				evlist->stats.total_aux_lost += 1;
1425 			if (event->aux.flags & PERF_AUX_FLAG_PARTIAL)
1426 				evlist->stats.total_aux_partial += 1;
1427 		}
1428 		return tool->aux(tool, event, sample, machine);
1429 	case PERF_RECORD_ITRACE_START:
1430 		return tool->itrace_start(tool, event, sample, machine);
1431 	case PERF_RECORD_SWITCH:
1432 	case PERF_RECORD_SWITCH_CPU_WIDE:
1433 		return tool->context_switch(tool, event, sample, machine);
1434 	case PERF_RECORD_KSYMBOL:
1435 		return tool->ksymbol(tool, event, sample, machine);
1436 	case PERF_RECORD_BPF_EVENT:
1437 		return tool->bpf_event(tool, event, sample, machine);
1438 	default:
1439 		++evlist->stats.nr_unknown_events;
1440 		return -1;
1441 	}
1442 }
1443 
1444 static int perf_session__deliver_event(struct perf_session *session,
1445 				       union perf_event *event,
1446 				       struct perf_tool *tool,
1447 				       u64 file_offset)
1448 {
1449 	struct perf_sample sample;
1450 	int ret;
1451 
1452 	ret = perf_evlist__parse_sample(session->evlist, event, &sample);
1453 	if (ret) {
1454 		pr_err("Can't parse sample, err = %d\n", ret);
1455 		return ret;
1456 	}
1457 
1458 	ret = auxtrace__process_event(session, event, &sample, tool);
1459 	if (ret < 0)
1460 		return ret;
1461 	if (ret > 0)
1462 		return 0;
1463 
1464 	return machines__deliver_event(&session->machines, session->evlist,
1465 				       event, &sample, tool, file_offset);
1466 }
1467 
1468 static s64 perf_session__process_user_event(struct perf_session *session,
1469 					    union perf_event *event,
1470 					    u64 file_offset)
1471 {
1472 	struct ordered_events *oe = &session->ordered_events;
1473 	struct perf_tool *tool = session->tool;
1474 	struct perf_sample sample = { .time = 0, };
1475 	int fd = perf_data__fd(session->data);
1476 	int err;
1477 
1478 	if (event->header.type != PERF_RECORD_COMPRESSED ||
1479 	    tool->compressed == perf_session__process_compressed_event_stub)
1480 		dump_event(session->evlist, event, file_offset, &sample);
1481 
1482 	/* These events are processed right away */
1483 	switch (event->header.type) {
1484 	case PERF_RECORD_HEADER_ATTR:
1485 		err = tool->attr(tool, event, &session->evlist);
1486 		if (err == 0) {
1487 			perf_session__set_id_hdr_size(session);
1488 			perf_session__set_comm_exec(session);
1489 		}
1490 		return err;
1491 	case PERF_RECORD_EVENT_UPDATE:
1492 		return tool->event_update(tool, event, &session->evlist);
1493 	case PERF_RECORD_HEADER_EVENT_TYPE:
1494 		/*
1495 		 * Depreceated, but we need to handle it for sake
1496 		 * of old data files create in pipe mode.
1497 		 */
1498 		return 0;
1499 	case PERF_RECORD_HEADER_TRACING_DATA:
1500 		/* setup for reading amidst mmap */
1501 		lseek(fd, file_offset, SEEK_SET);
1502 		return tool->tracing_data(session, event);
1503 	case PERF_RECORD_HEADER_BUILD_ID:
1504 		return tool->build_id(session, event);
1505 	case PERF_RECORD_FINISHED_ROUND:
1506 		return tool->finished_round(tool, event, oe);
1507 	case PERF_RECORD_ID_INDEX:
1508 		return tool->id_index(session, event);
1509 	case PERF_RECORD_AUXTRACE_INFO:
1510 		return tool->auxtrace_info(session, event);
1511 	case PERF_RECORD_AUXTRACE:
1512 		/* setup for reading amidst mmap */
1513 		lseek(fd, file_offset + event->header.size, SEEK_SET);
1514 		return tool->auxtrace(session, event);
1515 	case PERF_RECORD_AUXTRACE_ERROR:
1516 		perf_session__auxtrace_error_inc(session, event);
1517 		return tool->auxtrace_error(session, event);
1518 	case PERF_RECORD_THREAD_MAP:
1519 		return tool->thread_map(session, event);
1520 	case PERF_RECORD_CPU_MAP:
1521 		return tool->cpu_map(session, event);
1522 	case PERF_RECORD_STAT_CONFIG:
1523 		return tool->stat_config(session, event);
1524 	case PERF_RECORD_STAT:
1525 		return tool->stat(session, event);
1526 	case PERF_RECORD_STAT_ROUND:
1527 		return tool->stat_round(session, event);
1528 	case PERF_RECORD_TIME_CONV:
1529 		session->time_conv = event->time_conv;
1530 		return tool->time_conv(session, event);
1531 	case PERF_RECORD_HEADER_FEATURE:
1532 		return tool->feature(session, event);
1533 	case PERF_RECORD_COMPRESSED:
1534 		err = tool->compressed(session, event, file_offset);
1535 		if (err)
1536 			dump_event(session->evlist, event, file_offset, &sample);
1537 		return err;
1538 	default:
1539 		return -EINVAL;
1540 	}
1541 }
1542 
1543 int perf_session__deliver_synth_event(struct perf_session *session,
1544 				      union perf_event *event,
1545 				      struct perf_sample *sample)
1546 {
1547 	struct perf_evlist *evlist = session->evlist;
1548 	struct perf_tool *tool = session->tool;
1549 
1550 	events_stats__inc(&evlist->stats, event->header.type);
1551 
1552 	if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1553 		return perf_session__process_user_event(session, event, 0);
1554 
1555 	return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0);
1556 }
1557 
1558 static void event_swap(union perf_event *event, bool sample_id_all)
1559 {
1560 	perf_event__swap_op swap;
1561 
1562 	swap = perf_event__swap_ops[event->header.type];
1563 	if (swap)
1564 		swap(event, sample_id_all);
1565 }
1566 
1567 int perf_session__peek_event(struct perf_session *session, off_t file_offset,
1568 			     void *buf, size_t buf_sz,
1569 			     union perf_event **event_ptr,
1570 			     struct perf_sample *sample)
1571 {
1572 	union perf_event *event;
1573 	size_t hdr_sz, rest;
1574 	int fd;
1575 
1576 	if (session->one_mmap && !session->header.needs_swap) {
1577 		event = file_offset - session->one_mmap_offset +
1578 			session->one_mmap_addr;
1579 		goto out_parse_sample;
1580 	}
1581 
1582 	if (perf_data__is_pipe(session->data))
1583 		return -1;
1584 
1585 	fd = perf_data__fd(session->data);
1586 	hdr_sz = sizeof(struct perf_event_header);
1587 
1588 	if (buf_sz < hdr_sz)
1589 		return -1;
1590 
1591 	if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 ||
1592 	    readn(fd, buf, hdr_sz) != (ssize_t)hdr_sz)
1593 		return -1;
1594 
1595 	event = (union perf_event *)buf;
1596 
1597 	if (session->header.needs_swap)
1598 		perf_event_header__bswap(&event->header);
1599 
1600 	if (event->header.size < hdr_sz || event->header.size > buf_sz)
1601 		return -1;
1602 
1603 	rest = event->header.size - hdr_sz;
1604 
1605 	if (readn(fd, buf, rest) != (ssize_t)rest)
1606 		return -1;
1607 
1608 	if (session->header.needs_swap)
1609 		event_swap(event, perf_evlist__sample_id_all(session->evlist));
1610 
1611 out_parse_sample:
1612 
1613 	if (sample && event->header.type < PERF_RECORD_USER_TYPE_START &&
1614 	    perf_evlist__parse_sample(session->evlist, event, sample))
1615 		return -1;
1616 
1617 	*event_ptr = event;
1618 
1619 	return 0;
1620 }
1621 
1622 static s64 perf_session__process_event(struct perf_session *session,
1623 				       union perf_event *event, u64 file_offset)
1624 {
1625 	struct perf_evlist *evlist = session->evlist;
1626 	struct perf_tool *tool = session->tool;
1627 	int ret;
1628 
1629 	if (session->header.needs_swap)
1630 		event_swap(event, perf_evlist__sample_id_all(evlist));
1631 
1632 	if (event->header.type >= PERF_RECORD_HEADER_MAX)
1633 		return -EINVAL;
1634 
1635 	events_stats__inc(&evlist->stats, event->header.type);
1636 
1637 	if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1638 		return perf_session__process_user_event(session, event, file_offset);
1639 
1640 	if (tool->ordered_events) {
1641 		u64 timestamp = -1ULL;
1642 
1643 		ret = perf_evlist__parse_sample_timestamp(evlist, event, &timestamp);
1644 		if (ret && ret != -1)
1645 			return ret;
1646 
1647 		ret = perf_session__queue_event(session, event, timestamp, file_offset);
1648 		if (ret != -ETIME)
1649 			return ret;
1650 	}
1651 
1652 	return perf_session__deliver_event(session, event, tool, file_offset);
1653 }
1654 
1655 void perf_event_header__bswap(struct perf_event_header *hdr)
1656 {
1657 	hdr->type = bswap_32(hdr->type);
1658 	hdr->misc = bswap_16(hdr->misc);
1659 	hdr->size = bswap_16(hdr->size);
1660 }
1661 
1662 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid)
1663 {
1664 	return machine__findnew_thread(&session->machines.host, -1, pid);
1665 }
1666 
1667 /*
1668  * Threads are identified by pid and tid, and the idle task has pid == tid == 0.
1669  * So here a single thread is created for that, but actually there is a separate
1670  * idle task per cpu, so there should be one 'struct thread' per cpu, but there
1671  * is only 1. That causes problems for some tools, requiring workarounds. For
1672  * example get_idle_thread() in builtin-sched.c, or thread_stack__per_cpu().
1673  */
1674 int perf_session__register_idle_thread(struct perf_session *session)
1675 {
1676 	struct thread *thread;
1677 	int err = 0;
1678 
1679 	thread = machine__findnew_thread(&session->machines.host, 0, 0);
1680 	if (thread == NULL || thread__set_comm(thread, "swapper", 0)) {
1681 		pr_err("problem inserting idle task.\n");
1682 		err = -1;
1683 	}
1684 
1685 	if (thread == NULL || thread__set_namespaces(thread, 0, NULL)) {
1686 		pr_err("problem inserting idle task.\n");
1687 		err = -1;
1688 	}
1689 
1690 	/* machine__findnew_thread() got the thread, so put it */
1691 	thread__put(thread);
1692 	return err;
1693 }
1694 
1695 static void
1696 perf_session__warn_order(const struct perf_session *session)
1697 {
1698 	const struct ordered_events *oe = &session->ordered_events;
1699 	struct perf_evsel *evsel;
1700 	bool should_warn = true;
1701 
1702 	evlist__for_each_entry(session->evlist, evsel) {
1703 		if (evsel->attr.write_backward)
1704 			should_warn = false;
1705 	}
1706 
1707 	if (!should_warn)
1708 		return;
1709 	if (oe->nr_unordered_events != 0)
1710 		ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events);
1711 }
1712 
1713 static void perf_session__warn_about_errors(const struct perf_session *session)
1714 {
1715 	const struct events_stats *stats = &session->evlist->stats;
1716 
1717 	if (session->tool->lost == perf_event__process_lost &&
1718 	    stats->nr_events[PERF_RECORD_LOST] != 0) {
1719 		ui__warning("Processed %d events and lost %d chunks!\n\n"
1720 			    "Check IO/CPU overload!\n\n",
1721 			    stats->nr_events[0],
1722 			    stats->nr_events[PERF_RECORD_LOST]);
1723 	}
1724 
1725 	if (session->tool->lost_samples == perf_event__process_lost_samples) {
1726 		double drop_rate;
1727 
1728 		drop_rate = (double)stats->total_lost_samples /
1729 			    (double) (stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples);
1730 		if (drop_rate > 0.05) {
1731 			ui__warning("Processed %" PRIu64 " samples and lost %3.2f%%!\n\n",
1732 				    stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples,
1733 				    drop_rate * 100.0);
1734 		}
1735 	}
1736 
1737 	if (session->tool->aux == perf_event__process_aux &&
1738 	    stats->total_aux_lost != 0) {
1739 		ui__warning("AUX data lost %" PRIu64 " times out of %u!\n\n",
1740 			    stats->total_aux_lost,
1741 			    stats->nr_events[PERF_RECORD_AUX]);
1742 	}
1743 
1744 	if (session->tool->aux == perf_event__process_aux &&
1745 	    stats->total_aux_partial != 0) {
1746 		bool vmm_exclusive = false;
1747 
1748 		(void)sysfs__read_bool("module/kvm_intel/parameters/vmm_exclusive",
1749 		                       &vmm_exclusive);
1750 
1751 		ui__warning("AUX data had gaps in it %" PRIu64 " times out of %u!\n\n"
1752 		            "Are you running a KVM guest in the background?%s\n\n",
1753 			    stats->total_aux_partial,
1754 			    stats->nr_events[PERF_RECORD_AUX],
1755 			    vmm_exclusive ?
1756 			    "\nReloading kvm_intel module with vmm_exclusive=0\n"
1757 			    "will reduce the gaps to only guest's timeslices." :
1758 			    "");
1759 	}
1760 
1761 	if (stats->nr_unknown_events != 0) {
1762 		ui__warning("Found %u unknown events!\n\n"
1763 			    "Is this an older tool processing a perf.data "
1764 			    "file generated by a more recent tool?\n\n"
1765 			    "If that is not the case, consider "
1766 			    "reporting to linux-kernel@vger.kernel.org.\n\n",
1767 			    stats->nr_unknown_events);
1768 	}
1769 
1770 	if (stats->nr_unknown_id != 0) {
1771 		ui__warning("%u samples with id not present in the header\n",
1772 			    stats->nr_unknown_id);
1773 	}
1774 
1775 	if (stats->nr_invalid_chains != 0) {
1776 		ui__warning("Found invalid callchains!\n\n"
1777 			    "%u out of %u events were discarded for this reason.\n\n"
1778 			    "Consider reporting to linux-kernel@vger.kernel.org.\n\n",
1779 			    stats->nr_invalid_chains,
1780 			    stats->nr_events[PERF_RECORD_SAMPLE]);
1781 	}
1782 
1783 	if (stats->nr_unprocessable_samples != 0) {
1784 		ui__warning("%u unprocessable samples recorded.\n"
1785 			    "Do you have a KVM guest running and not using 'perf kvm'?\n",
1786 			    stats->nr_unprocessable_samples);
1787 	}
1788 
1789 	perf_session__warn_order(session);
1790 
1791 	events_stats__auxtrace_error_warn(stats);
1792 
1793 	if (stats->nr_proc_map_timeout != 0) {
1794 		ui__warning("%d map information files for pre-existing threads were\n"
1795 			    "not processed, if there are samples for addresses they\n"
1796 			    "will not be resolved, you may find out which are these\n"
1797 			    "threads by running with -v and redirecting the output\n"
1798 			    "to a file.\n"
1799 			    "The time limit to process proc map is too short?\n"
1800 			    "Increase it by --proc-map-timeout\n",
1801 			    stats->nr_proc_map_timeout);
1802 	}
1803 }
1804 
1805 static int perf_session__flush_thread_stack(struct thread *thread,
1806 					    void *p __maybe_unused)
1807 {
1808 	return thread_stack__flush(thread);
1809 }
1810 
1811 static int perf_session__flush_thread_stacks(struct perf_session *session)
1812 {
1813 	return machines__for_each_thread(&session->machines,
1814 					 perf_session__flush_thread_stack,
1815 					 NULL);
1816 }
1817 
1818 volatile int session_done;
1819 
1820 static int __perf_session__process_decomp_events(struct perf_session *session);
1821 
1822 static int __perf_session__process_pipe_events(struct perf_session *session)
1823 {
1824 	struct ordered_events *oe = &session->ordered_events;
1825 	struct perf_tool *tool = session->tool;
1826 	int fd = perf_data__fd(session->data);
1827 	union perf_event *event;
1828 	uint32_t size, cur_size = 0;
1829 	void *buf = NULL;
1830 	s64 skip = 0;
1831 	u64 head;
1832 	ssize_t err;
1833 	void *p;
1834 
1835 	perf_tool__fill_defaults(tool);
1836 
1837 	head = 0;
1838 	cur_size = sizeof(union perf_event);
1839 
1840 	buf = malloc(cur_size);
1841 	if (!buf)
1842 		return -errno;
1843 	ordered_events__set_copy_on_queue(oe, true);
1844 more:
1845 	event = buf;
1846 	err = readn(fd, event, sizeof(struct perf_event_header));
1847 	if (err <= 0) {
1848 		if (err == 0)
1849 			goto done;
1850 
1851 		pr_err("failed to read event header\n");
1852 		goto out_err;
1853 	}
1854 
1855 	if (session->header.needs_swap)
1856 		perf_event_header__bswap(&event->header);
1857 
1858 	size = event->header.size;
1859 	if (size < sizeof(struct perf_event_header)) {
1860 		pr_err("bad event header size\n");
1861 		goto out_err;
1862 	}
1863 
1864 	if (size > cur_size) {
1865 		void *new = realloc(buf, size);
1866 		if (!new) {
1867 			pr_err("failed to allocate memory to read event\n");
1868 			goto out_err;
1869 		}
1870 		buf = new;
1871 		cur_size = size;
1872 		event = buf;
1873 	}
1874 	p = event;
1875 	p += sizeof(struct perf_event_header);
1876 
1877 	if (size - sizeof(struct perf_event_header)) {
1878 		err = readn(fd, p, size - sizeof(struct perf_event_header));
1879 		if (err <= 0) {
1880 			if (err == 0) {
1881 				pr_err("unexpected end of event stream\n");
1882 				goto done;
1883 			}
1884 
1885 			pr_err("failed to read event data\n");
1886 			goto out_err;
1887 		}
1888 	}
1889 
1890 	if ((skip = perf_session__process_event(session, event, head)) < 0) {
1891 		pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1892 		       head, event->header.size, event->header.type);
1893 		err = -EINVAL;
1894 		goto out_err;
1895 	}
1896 
1897 	head += size;
1898 
1899 	if (skip > 0)
1900 		head += skip;
1901 
1902 	err = __perf_session__process_decomp_events(session);
1903 	if (err)
1904 		goto out_err;
1905 
1906 	if (!session_done())
1907 		goto more;
1908 done:
1909 	/* do the final flush for ordered samples */
1910 	err = ordered_events__flush(oe, OE_FLUSH__FINAL);
1911 	if (err)
1912 		goto out_err;
1913 	err = auxtrace__flush_events(session, tool);
1914 	if (err)
1915 		goto out_err;
1916 	err = perf_session__flush_thread_stacks(session);
1917 out_err:
1918 	free(buf);
1919 	if (!tool->no_warn)
1920 		perf_session__warn_about_errors(session);
1921 	ordered_events__free(&session->ordered_events);
1922 	auxtrace__free_events(session);
1923 	return err;
1924 }
1925 
1926 static union perf_event *
1927 fetch_mmaped_event(struct perf_session *session,
1928 		   u64 head, size_t mmap_size, char *buf)
1929 {
1930 	union perf_event *event;
1931 
1932 	/*
1933 	 * Ensure we have enough space remaining to read
1934 	 * the size of the event in the headers.
1935 	 */
1936 	if (head + sizeof(event->header) > mmap_size)
1937 		return NULL;
1938 
1939 	event = (union perf_event *)(buf + head);
1940 
1941 	if (session->header.needs_swap)
1942 		perf_event_header__bswap(&event->header);
1943 
1944 	if (head + event->header.size > mmap_size) {
1945 		/* We're not fetching the event so swap back again */
1946 		if (session->header.needs_swap)
1947 			perf_event_header__bswap(&event->header);
1948 		return NULL;
1949 	}
1950 
1951 	return event;
1952 }
1953 
1954 static int __perf_session__process_decomp_events(struct perf_session *session)
1955 {
1956 	s64 skip;
1957 	u64 size, file_pos = 0;
1958 	struct decomp *decomp = session->decomp_last;
1959 
1960 	if (!decomp)
1961 		return 0;
1962 
1963 	while (decomp->head < decomp->size && !session_done()) {
1964 		union perf_event *event = fetch_mmaped_event(session, decomp->head, decomp->size, decomp->data);
1965 
1966 		if (!event)
1967 			break;
1968 
1969 		size = event->header.size;
1970 
1971 		if (size < sizeof(struct perf_event_header) ||
1972 		    (skip = perf_session__process_event(session, event, file_pos)) < 0) {
1973 			pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1974 				decomp->file_pos + decomp->head, event->header.size, event->header.type);
1975 			return -EINVAL;
1976 		}
1977 
1978 		if (skip)
1979 			size += skip;
1980 
1981 		decomp->head += size;
1982 	}
1983 
1984 	return 0;
1985 }
1986 
1987 /*
1988  * On 64bit we can mmap the data file in one go. No need for tiny mmap
1989  * slices. On 32bit we use 32MB.
1990  */
1991 #if BITS_PER_LONG == 64
1992 #define MMAP_SIZE ULLONG_MAX
1993 #define NUM_MMAPS 1
1994 #else
1995 #define MMAP_SIZE (32 * 1024 * 1024ULL)
1996 #define NUM_MMAPS 128
1997 #endif
1998 
1999 struct reader;
2000 
2001 typedef s64 (*reader_cb_t)(struct perf_session *session,
2002 			   union perf_event *event,
2003 			   u64 file_offset);
2004 
2005 struct reader {
2006 	int		 fd;
2007 	u64		 data_size;
2008 	u64		 data_offset;
2009 	reader_cb_t	 process;
2010 };
2011 
2012 static int
2013 reader__process_events(struct reader *rd, struct perf_session *session,
2014 		       struct ui_progress *prog)
2015 {
2016 	u64 data_size = rd->data_size;
2017 	u64 head, page_offset, file_offset, file_pos, size;
2018 	int err = 0, mmap_prot, mmap_flags, map_idx = 0;
2019 	size_t	mmap_size;
2020 	char *buf, *mmaps[NUM_MMAPS];
2021 	union perf_event *event;
2022 	s64 skip;
2023 
2024 	page_offset = page_size * (rd->data_offset / page_size);
2025 	file_offset = page_offset;
2026 	head = rd->data_offset - page_offset;
2027 
2028 	ui_progress__init_size(prog, data_size, "Processing events...");
2029 
2030 	data_size += rd->data_offset;
2031 
2032 	mmap_size = MMAP_SIZE;
2033 	if (mmap_size > data_size) {
2034 		mmap_size = data_size;
2035 		session->one_mmap = true;
2036 	}
2037 
2038 	memset(mmaps, 0, sizeof(mmaps));
2039 
2040 	mmap_prot  = PROT_READ;
2041 	mmap_flags = MAP_SHARED;
2042 
2043 	if (session->header.needs_swap) {
2044 		mmap_prot  |= PROT_WRITE;
2045 		mmap_flags = MAP_PRIVATE;
2046 	}
2047 remap:
2048 	buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, rd->fd,
2049 		   file_offset);
2050 	if (buf == MAP_FAILED) {
2051 		pr_err("failed to mmap file\n");
2052 		err = -errno;
2053 		goto out;
2054 	}
2055 	mmaps[map_idx] = buf;
2056 	map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1);
2057 	file_pos = file_offset + head;
2058 	if (session->one_mmap) {
2059 		session->one_mmap_addr = buf;
2060 		session->one_mmap_offset = file_offset;
2061 	}
2062 
2063 more:
2064 	event = fetch_mmaped_event(session, head, mmap_size, buf);
2065 	if (!event) {
2066 		if (mmaps[map_idx]) {
2067 			munmap(mmaps[map_idx], mmap_size);
2068 			mmaps[map_idx] = NULL;
2069 		}
2070 
2071 		page_offset = page_size * (head / page_size);
2072 		file_offset += page_offset;
2073 		head -= page_offset;
2074 		goto remap;
2075 	}
2076 
2077 	size = event->header.size;
2078 
2079 	skip = -EINVAL;
2080 
2081 	if (size < sizeof(struct perf_event_header) ||
2082 	    (skip = rd->process(session, event, file_pos)) < 0) {
2083 		pr_err("%#" PRIx64 " [%#x]: failed to process type: %d [%s]\n",
2084 		       file_offset + head, event->header.size,
2085 		       event->header.type, strerror(-skip));
2086 		err = skip;
2087 		goto out;
2088 	}
2089 
2090 	if (skip)
2091 		size += skip;
2092 
2093 	head += size;
2094 	file_pos += size;
2095 
2096 	err = __perf_session__process_decomp_events(session);
2097 	if (err)
2098 		goto out;
2099 
2100 	ui_progress__update(prog, size);
2101 
2102 	if (session_done())
2103 		goto out;
2104 
2105 	if (file_pos < data_size)
2106 		goto more;
2107 
2108 out:
2109 	return err;
2110 }
2111 
2112 static s64 process_simple(struct perf_session *session,
2113 			  union perf_event *event,
2114 			  u64 file_offset)
2115 {
2116 	return perf_session__process_event(session, event, file_offset);
2117 }
2118 
2119 static int __perf_session__process_events(struct perf_session *session)
2120 {
2121 	struct reader rd = {
2122 		.fd		= perf_data__fd(session->data),
2123 		.data_size	= session->header.data_size,
2124 		.data_offset	= session->header.data_offset,
2125 		.process	= process_simple,
2126 	};
2127 	struct ordered_events *oe = &session->ordered_events;
2128 	struct perf_tool *tool = session->tool;
2129 	struct ui_progress prog;
2130 	int err;
2131 
2132 	perf_tool__fill_defaults(tool);
2133 
2134 	if (rd.data_size == 0)
2135 		return -1;
2136 
2137 	ui_progress__init_size(&prog, rd.data_size, "Processing events...");
2138 
2139 	err = reader__process_events(&rd, session, &prog);
2140 	if (err)
2141 		goto out_err;
2142 	/* do the final flush for ordered samples */
2143 	err = ordered_events__flush(oe, OE_FLUSH__FINAL);
2144 	if (err)
2145 		goto out_err;
2146 	err = auxtrace__flush_events(session, tool);
2147 	if (err)
2148 		goto out_err;
2149 	err = perf_session__flush_thread_stacks(session);
2150 out_err:
2151 	ui_progress__finish();
2152 	if (!tool->no_warn)
2153 		perf_session__warn_about_errors(session);
2154 	/*
2155 	 * We may switching perf.data output, make ordered_events
2156 	 * reusable.
2157 	 */
2158 	ordered_events__reinit(&session->ordered_events);
2159 	auxtrace__free_events(session);
2160 	session->one_mmap = false;
2161 	return err;
2162 }
2163 
2164 int perf_session__process_events(struct perf_session *session)
2165 {
2166 	if (perf_session__register_idle_thread(session) < 0)
2167 		return -ENOMEM;
2168 
2169 	if (perf_data__is_pipe(session->data))
2170 		return __perf_session__process_pipe_events(session);
2171 
2172 	return __perf_session__process_events(session);
2173 }
2174 
2175 bool perf_session__has_traces(struct perf_session *session, const char *msg)
2176 {
2177 	struct perf_evsel *evsel;
2178 
2179 	evlist__for_each_entry(session->evlist, evsel) {
2180 		if (evsel->attr.type == PERF_TYPE_TRACEPOINT)
2181 			return true;
2182 	}
2183 
2184 	pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg);
2185 	return false;
2186 }
2187 
2188 int map__set_kallsyms_ref_reloc_sym(struct map *map, const char *symbol_name, u64 addr)
2189 {
2190 	char *bracket;
2191 	struct ref_reloc_sym *ref;
2192 	struct kmap *kmap;
2193 
2194 	ref = zalloc(sizeof(struct ref_reloc_sym));
2195 	if (ref == NULL)
2196 		return -ENOMEM;
2197 
2198 	ref->name = strdup(symbol_name);
2199 	if (ref->name == NULL) {
2200 		free(ref);
2201 		return -ENOMEM;
2202 	}
2203 
2204 	bracket = strchr(ref->name, ']');
2205 	if (bracket)
2206 		*bracket = '\0';
2207 
2208 	ref->addr = addr;
2209 
2210 	kmap = map__kmap(map);
2211 	if (kmap)
2212 		kmap->ref_reloc_sym = ref;
2213 
2214 	return 0;
2215 }
2216 
2217 size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp)
2218 {
2219 	return machines__fprintf_dsos(&session->machines, fp);
2220 }
2221 
2222 size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp,
2223 					  bool (skip)(struct dso *dso, int parm), int parm)
2224 {
2225 	return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm);
2226 }
2227 
2228 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp)
2229 {
2230 	size_t ret;
2231 	const char *msg = "";
2232 
2233 	if (perf_header__has_feat(&session->header, HEADER_AUXTRACE))
2234 		msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)";
2235 
2236 	ret = fprintf(fp, "\nAggregated stats:%s\n", msg);
2237 
2238 	ret += events_stats__fprintf(&session->evlist->stats, fp);
2239 	return ret;
2240 }
2241 
2242 size_t perf_session__fprintf(struct perf_session *session, FILE *fp)
2243 {
2244 	/*
2245 	 * FIXME: Here we have to actually print all the machines in this
2246 	 * session, not just the host...
2247 	 */
2248 	return machine__fprintf(&session->machines.host, fp);
2249 }
2250 
2251 struct perf_evsel *perf_session__find_first_evtype(struct perf_session *session,
2252 					      unsigned int type)
2253 {
2254 	struct perf_evsel *pos;
2255 
2256 	evlist__for_each_entry(session->evlist, pos) {
2257 		if (pos->attr.type == type)
2258 			return pos;
2259 	}
2260 	return NULL;
2261 }
2262 
2263 int perf_session__cpu_bitmap(struct perf_session *session,
2264 			     const char *cpu_list, unsigned long *cpu_bitmap)
2265 {
2266 	int i, err = -1;
2267 	struct cpu_map *map;
2268 
2269 	for (i = 0; i < PERF_TYPE_MAX; ++i) {
2270 		struct perf_evsel *evsel;
2271 
2272 		evsel = perf_session__find_first_evtype(session, i);
2273 		if (!evsel)
2274 			continue;
2275 
2276 		if (!(evsel->attr.sample_type & PERF_SAMPLE_CPU)) {
2277 			pr_err("File does not contain CPU events. "
2278 			       "Remove -C option to proceed.\n");
2279 			return -1;
2280 		}
2281 	}
2282 
2283 	map = cpu_map__new(cpu_list);
2284 	if (map == NULL) {
2285 		pr_err("Invalid cpu_list\n");
2286 		return -1;
2287 	}
2288 
2289 	for (i = 0; i < map->nr; i++) {
2290 		int cpu = map->map[i];
2291 
2292 		if (cpu >= MAX_NR_CPUS) {
2293 			pr_err("Requested CPU %d too large. "
2294 			       "Consider raising MAX_NR_CPUS\n", cpu);
2295 			goto out_delete_map;
2296 		}
2297 
2298 		set_bit(cpu, cpu_bitmap);
2299 	}
2300 
2301 	err = 0;
2302 
2303 out_delete_map:
2304 	cpu_map__put(map);
2305 	return err;
2306 }
2307 
2308 void perf_session__fprintf_info(struct perf_session *session, FILE *fp,
2309 				bool full)
2310 {
2311 	if (session == NULL || fp == NULL)
2312 		return;
2313 
2314 	fprintf(fp, "# ========\n");
2315 	perf_header__fprintf_info(session, fp, full);
2316 	fprintf(fp, "# ========\n#\n");
2317 }
2318 
2319 
2320 int __perf_session__set_tracepoints_handlers(struct perf_session *session,
2321 					     const struct perf_evsel_str_handler *assocs,
2322 					     size_t nr_assocs)
2323 {
2324 	struct perf_evsel *evsel;
2325 	size_t i;
2326 	int err;
2327 
2328 	for (i = 0; i < nr_assocs; i++) {
2329 		/*
2330 		 * Adding a handler for an event not in the session,
2331 		 * just ignore it.
2332 		 */
2333 		evsel = perf_evlist__find_tracepoint_by_name(session->evlist, assocs[i].name);
2334 		if (evsel == NULL)
2335 			continue;
2336 
2337 		err = -EEXIST;
2338 		if (evsel->handler != NULL)
2339 			goto out;
2340 		evsel->handler = assocs[i].handler;
2341 	}
2342 
2343 	err = 0;
2344 out:
2345 	return err;
2346 }
2347 
2348 int perf_event__process_id_index(struct perf_session *session,
2349 				 union perf_event *event)
2350 {
2351 	struct perf_evlist *evlist = session->evlist;
2352 	struct id_index_event *ie = &event->id_index;
2353 	size_t i, nr, max_nr;
2354 
2355 	max_nr = (ie->header.size - sizeof(struct id_index_event)) /
2356 		 sizeof(struct id_index_entry);
2357 	nr = ie->nr;
2358 	if (nr > max_nr)
2359 		return -EINVAL;
2360 
2361 	if (dump_trace)
2362 		fprintf(stdout, " nr: %zu\n", nr);
2363 
2364 	for (i = 0; i < nr; i++) {
2365 		struct id_index_entry *e = &ie->entries[i];
2366 		struct perf_sample_id *sid;
2367 
2368 		if (dump_trace) {
2369 			fprintf(stdout,	" ... id: %"PRIu64, e->id);
2370 			fprintf(stdout,	"  idx: %"PRIu64, e->idx);
2371 			fprintf(stdout,	"  cpu: %"PRId64, e->cpu);
2372 			fprintf(stdout,	"  tid: %"PRId64"\n", e->tid);
2373 		}
2374 
2375 		sid = perf_evlist__id2sid(evlist, e->id);
2376 		if (!sid)
2377 			return -ENOENT;
2378 		sid->idx = e->idx;
2379 		sid->cpu = e->cpu;
2380 		sid->tid = e->tid;
2381 	}
2382 	return 0;
2383 }
2384 
2385 int perf_event__synthesize_id_index(struct perf_tool *tool,
2386 				    perf_event__handler_t process,
2387 				    struct perf_evlist *evlist,
2388 				    struct machine *machine)
2389 {
2390 	union perf_event *ev;
2391 	struct perf_evsel *evsel;
2392 	size_t nr = 0, i = 0, sz, max_nr, n;
2393 	int err;
2394 
2395 	pr_debug2("Synthesizing id index\n");
2396 
2397 	max_nr = (UINT16_MAX - sizeof(struct id_index_event)) /
2398 		 sizeof(struct id_index_entry);
2399 
2400 	evlist__for_each_entry(evlist, evsel)
2401 		nr += evsel->ids;
2402 
2403 	n = nr > max_nr ? max_nr : nr;
2404 	sz = sizeof(struct id_index_event) + n * sizeof(struct id_index_entry);
2405 	ev = zalloc(sz);
2406 	if (!ev)
2407 		return -ENOMEM;
2408 
2409 	ev->id_index.header.type = PERF_RECORD_ID_INDEX;
2410 	ev->id_index.header.size = sz;
2411 	ev->id_index.nr = n;
2412 
2413 	evlist__for_each_entry(evlist, evsel) {
2414 		u32 j;
2415 
2416 		for (j = 0; j < evsel->ids; j++) {
2417 			struct id_index_entry *e;
2418 			struct perf_sample_id *sid;
2419 
2420 			if (i >= n) {
2421 				err = process(tool, ev, NULL, machine);
2422 				if (err)
2423 					goto out_err;
2424 				nr -= n;
2425 				i = 0;
2426 			}
2427 
2428 			e = &ev->id_index.entries[i++];
2429 
2430 			e->id = evsel->id[j];
2431 
2432 			sid = perf_evlist__id2sid(evlist, e->id);
2433 			if (!sid) {
2434 				free(ev);
2435 				return -ENOENT;
2436 			}
2437 
2438 			e->idx = sid->idx;
2439 			e->cpu = sid->cpu;
2440 			e->tid = sid->tid;
2441 		}
2442 	}
2443 
2444 	sz = sizeof(struct id_index_event) + nr * sizeof(struct id_index_entry);
2445 	ev->id_index.header.size = sz;
2446 	ev->id_index.nr = nr;
2447 
2448 	err = process(tool, ev, NULL, machine);
2449 out_err:
2450 	free(ev);
2451 
2452 	return err;
2453 }
2454