xref: /linux/tools/perf/util/session.c (revision be59dba332e1e8edd3e88d991ba0e4795ae2bcb2)
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 #include <perf/event.h>
16 
17 #include "map_symbol.h"
18 #include "branch.h"
19 #include "debug.h"
20 #include "env.h"
21 #include "evlist.h"
22 #include "evsel.h"
23 #include "memswap.h"
24 #include "map.h"
25 #include "symbol.h"
26 #include "session.h"
27 #include "tool.h"
28 #include "perf_regs.h"
29 #include "asm/bug.h"
30 #include "auxtrace.h"
31 #include "thread.h"
32 #include "thread-stack.h"
33 #include "sample-raw.h"
34 #include "stat.h"
35 #include "tsc.h"
36 #include "ui/progress.h"
37 #include "util.h"
38 #include "arch/common.h"
39 #include "units.h"
40 #include "annotate.h"
41 #include "perf.h"
42 #include <internal/lib.h>
43 
44 static int perf_session__deliver_event(struct perf_session *session,
45 				       union perf_event *event,
46 				       const struct perf_tool *tool,
47 				       u64 file_offset,
48 				       const char *file_path);
49 
50 static int perf_session__open(struct perf_session *session)
51 {
52 	struct perf_data *data = session->data;
53 
54 	if (perf_session__read_header(session) < 0) {
55 		pr_err("incompatible file format (rerun with -v to learn more)\n");
56 		return -1;
57 	}
58 
59 	if (perf_header__has_feat(&session->header, HEADER_AUXTRACE)) {
60 		/* Auxiliary events may reference exited threads, hold onto dead ones. */
61 		symbol_conf.keep_exited_threads = true;
62 	}
63 
64 	if (perf_data__is_pipe(data))
65 		return 0;
66 
67 	if (perf_header__has_feat(&session->header, HEADER_STAT))
68 		return 0;
69 
70 	if (!evlist__valid_sample_type(session->evlist)) {
71 		pr_err("non matching sample_type\n");
72 		return -1;
73 	}
74 
75 	if (!evlist__valid_sample_id_all(session->evlist)) {
76 		pr_err("non matching sample_id_all\n");
77 		return -1;
78 	}
79 
80 	if (!evlist__valid_read_format(session->evlist)) {
81 		pr_err("non matching read_format\n");
82 		return -1;
83 	}
84 
85 	return 0;
86 }
87 
88 void perf_session__set_id_hdr_size(struct perf_session *session)
89 {
90 	u16 id_hdr_size = evlist__id_hdr_size(session->evlist);
91 
92 	machines__set_id_hdr_size(&session->machines, id_hdr_size);
93 }
94 
95 int perf_session__create_kernel_maps(struct perf_session *session)
96 {
97 	int ret = machine__create_kernel_maps(&session->machines.host);
98 
99 	if (ret >= 0)
100 		ret = machines__create_guest_kernel_maps(&session->machines);
101 	return ret;
102 }
103 
104 static void perf_session__destroy_kernel_maps(struct perf_session *session)
105 {
106 	machines__destroy_kernel_maps(&session->machines);
107 }
108 
109 static bool perf_session__has_comm_exec(struct perf_session *session)
110 {
111 	struct evsel *evsel;
112 
113 	evlist__for_each_entry(session->evlist, evsel) {
114 		if (evsel->core.attr.comm_exec)
115 			return true;
116 	}
117 
118 	return false;
119 }
120 
121 static void perf_session__set_comm_exec(struct perf_session *session)
122 {
123 	bool comm_exec = perf_session__has_comm_exec(session);
124 
125 	machines__set_comm_exec(&session->machines, comm_exec);
126 }
127 
128 static int ordered_events__deliver_event(struct ordered_events *oe,
129 					 struct ordered_event *event)
130 {
131 	struct perf_session *session = container_of(oe, struct perf_session,
132 						    ordered_events);
133 
134 	return perf_session__deliver_event(session, event->event,
135 					   session->tool, event->file_offset,
136 					   event->file_path);
137 }
138 
139 struct perf_session *__perf_session__new(struct perf_data *data,
140 					 struct perf_tool *tool,
141 					 bool trace_event_repipe)
142 {
143 	int ret = -ENOMEM;
144 	struct perf_session *session = zalloc(sizeof(*session));
145 
146 	if (!session)
147 		goto out;
148 
149 	session->trace_event_repipe = trace_event_repipe;
150 	session->tool   = tool;
151 	session->decomp_data.zstd_decomp = &session->zstd_data;
152 	session->active_decomp = &session->decomp_data;
153 	INIT_LIST_HEAD(&session->auxtrace_index);
154 	machines__init(&session->machines);
155 	ordered_events__init(&session->ordered_events,
156 			     ordered_events__deliver_event, NULL);
157 
158 	perf_env__init(&session->header.env);
159 	if (data) {
160 		ret = perf_data__open(data);
161 		if (ret < 0)
162 			goto out_delete;
163 
164 		session->data = data;
165 
166 		if (perf_data__is_read(data)) {
167 			ret = perf_session__open(session);
168 			if (ret < 0)
169 				goto out_delete;
170 
171 			/*
172 			 * set session attributes that are present in perf.data
173 			 * but not in pipe-mode.
174 			 */
175 			if (!data->is_pipe) {
176 				perf_session__set_id_hdr_size(session);
177 				perf_session__set_comm_exec(session);
178 			}
179 
180 			evlist__init_trace_event_sample_raw(session->evlist);
181 
182 			/* Open the directory data. */
183 			if (data->is_dir) {
184 				ret = perf_data__open_dir(data);
185 				if (ret)
186 					goto out_delete;
187 			}
188 
189 			if (!symbol_conf.kallsyms_name &&
190 			    !symbol_conf.vmlinux_name)
191 				symbol_conf.kallsyms_name = perf_data__kallsyms_name(data);
192 		}
193 	} else  {
194 		session->machines.host.env = &perf_env;
195 	}
196 
197 	session->machines.host.single_address_space =
198 		perf_env__single_address_space(session->machines.host.env);
199 
200 	if (!data || perf_data__is_write(data)) {
201 		/*
202 		 * In O_RDONLY mode this will be performed when reading the
203 		 * kernel MMAP event, in perf_event__process_mmap().
204 		 */
205 		if (perf_session__create_kernel_maps(session) < 0)
206 			pr_warning("Cannot read kernel map\n");
207 	}
208 
209 	/*
210 	 * In pipe-mode, evlist is empty until PERF_RECORD_HEADER_ATTR is
211 	 * processed, so evlist__sample_id_all is not meaningful here.
212 	 */
213 	if ((!data || !data->is_pipe) && tool && tool->ordering_requires_timestamps &&
214 	    tool->ordered_events && !evlist__sample_id_all(session->evlist)) {
215 		dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n");
216 		tool->ordered_events = false;
217 	}
218 
219 	return session;
220 
221  out_delete:
222 	perf_session__delete(session);
223  out:
224 	return ERR_PTR(ret);
225 }
226 
227 static void perf_decomp__release_events(struct decomp *next)
228 {
229 	struct decomp *decomp;
230 	size_t mmap_len;
231 
232 	do {
233 		decomp = next;
234 		if (decomp == NULL)
235 			break;
236 		next = decomp->next;
237 		mmap_len = decomp->mmap_len;
238 		munmap(decomp, mmap_len);
239 	} while (1);
240 }
241 
242 void perf_session__delete(struct perf_session *session)
243 {
244 	if (session == NULL)
245 		return;
246 	auxtrace__free(session);
247 	auxtrace_index__free(&session->auxtrace_index);
248 	debuginfo_cache__delete();
249 	perf_session__destroy_kernel_maps(session);
250 	perf_decomp__release_events(session->decomp_data.decomp);
251 	perf_env__exit(&session->header.env);
252 	machines__exit(&session->machines);
253 	if (session->data) {
254 		if (perf_data__is_read(session->data))
255 			evlist__delete(session->evlist);
256 		perf_data__close(session->data);
257 	}
258 #ifdef HAVE_LIBTRACEEVENT
259 	trace_event__cleanup(&session->tevent);
260 #endif
261 	free(session);
262 }
263 
264 static void swap_sample_id_all(union perf_event *event, void *data)
265 {
266 	void *end = (void *) event + event->header.size;
267 	int size = end - data;
268 
269 	BUG_ON(size % sizeof(u64));
270 	mem_bswap_64(data, size);
271 }
272 
273 static void perf_event__all64_swap(union perf_event *event,
274 				   bool sample_id_all __maybe_unused)
275 {
276 	struct perf_event_header *hdr = &event->header;
277 	mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr));
278 }
279 
280 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all)
281 {
282 	event->comm.pid = bswap_32(event->comm.pid);
283 	event->comm.tid = bswap_32(event->comm.tid);
284 
285 	if (sample_id_all) {
286 		void *data = &event->comm.comm;
287 
288 		data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
289 		swap_sample_id_all(event, data);
290 	}
291 }
292 
293 static void perf_event__mmap_swap(union perf_event *event,
294 				  bool sample_id_all)
295 {
296 	event->mmap.pid	  = bswap_32(event->mmap.pid);
297 	event->mmap.tid	  = bswap_32(event->mmap.tid);
298 	event->mmap.start = bswap_64(event->mmap.start);
299 	event->mmap.len	  = bswap_64(event->mmap.len);
300 	event->mmap.pgoff = bswap_64(event->mmap.pgoff);
301 
302 	if (sample_id_all) {
303 		void *data = &event->mmap.filename;
304 
305 		data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
306 		swap_sample_id_all(event, data);
307 	}
308 }
309 
310 static void perf_event__mmap2_swap(union perf_event *event,
311 				  bool sample_id_all)
312 {
313 	event->mmap2.pid   = bswap_32(event->mmap2.pid);
314 	event->mmap2.tid   = bswap_32(event->mmap2.tid);
315 	event->mmap2.start = bswap_64(event->mmap2.start);
316 	event->mmap2.len   = bswap_64(event->mmap2.len);
317 	event->mmap2.pgoff = bswap_64(event->mmap2.pgoff);
318 
319 	if (!(event->header.misc & PERF_RECORD_MISC_MMAP_BUILD_ID)) {
320 		event->mmap2.maj   = bswap_32(event->mmap2.maj);
321 		event->mmap2.min   = bswap_32(event->mmap2.min);
322 		event->mmap2.ino   = bswap_64(event->mmap2.ino);
323 		event->mmap2.ino_generation = bswap_64(event->mmap2.ino_generation);
324 	}
325 
326 	if (sample_id_all) {
327 		void *data = &event->mmap2.filename;
328 
329 		data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
330 		swap_sample_id_all(event, data);
331 	}
332 }
333 static void perf_event__task_swap(union perf_event *event, bool sample_id_all)
334 {
335 	event->fork.pid	 = bswap_32(event->fork.pid);
336 	event->fork.tid	 = bswap_32(event->fork.tid);
337 	event->fork.ppid = bswap_32(event->fork.ppid);
338 	event->fork.ptid = bswap_32(event->fork.ptid);
339 	event->fork.time = bswap_64(event->fork.time);
340 
341 	if (sample_id_all)
342 		swap_sample_id_all(event, &event->fork + 1);
343 }
344 
345 static void perf_event__read_swap(union perf_event *event, bool sample_id_all)
346 {
347 	event->read.pid		 = bswap_32(event->read.pid);
348 	event->read.tid		 = bswap_32(event->read.tid);
349 	event->read.value	 = bswap_64(event->read.value);
350 	event->read.time_enabled = bswap_64(event->read.time_enabled);
351 	event->read.time_running = bswap_64(event->read.time_running);
352 	event->read.id		 = bswap_64(event->read.id);
353 
354 	if (sample_id_all)
355 		swap_sample_id_all(event, &event->read + 1);
356 }
357 
358 static void perf_event__aux_swap(union perf_event *event, bool sample_id_all)
359 {
360 	event->aux.aux_offset = bswap_64(event->aux.aux_offset);
361 	event->aux.aux_size   = bswap_64(event->aux.aux_size);
362 	event->aux.flags      = bswap_64(event->aux.flags);
363 
364 	if (sample_id_all)
365 		swap_sample_id_all(event, &event->aux + 1);
366 }
367 
368 static void perf_event__itrace_start_swap(union perf_event *event,
369 					  bool sample_id_all)
370 {
371 	event->itrace_start.pid	 = bswap_32(event->itrace_start.pid);
372 	event->itrace_start.tid	 = bswap_32(event->itrace_start.tid);
373 
374 	if (sample_id_all)
375 		swap_sample_id_all(event, &event->itrace_start + 1);
376 }
377 
378 static void perf_event__switch_swap(union perf_event *event, bool sample_id_all)
379 {
380 	if (event->header.type == PERF_RECORD_SWITCH_CPU_WIDE) {
381 		event->context_switch.next_prev_pid =
382 				bswap_32(event->context_switch.next_prev_pid);
383 		event->context_switch.next_prev_tid =
384 				bswap_32(event->context_switch.next_prev_tid);
385 	}
386 
387 	if (sample_id_all)
388 		swap_sample_id_all(event, &event->context_switch + 1);
389 }
390 
391 static void perf_event__text_poke_swap(union perf_event *event, bool sample_id_all)
392 {
393 	event->text_poke.addr    = bswap_64(event->text_poke.addr);
394 	event->text_poke.old_len = bswap_16(event->text_poke.old_len);
395 	event->text_poke.new_len = bswap_16(event->text_poke.new_len);
396 
397 	if (sample_id_all) {
398 		size_t len = sizeof(event->text_poke.old_len) +
399 			     sizeof(event->text_poke.new_len) +
400 			     event->text_poke.old_len +
401 			     event->text_poke.new_len;
402 		void *data = &event->text_poke.old_len;
403 
404 		data += PERF_ALIGN(len, sizeof(u64));
405 		swap_sample_id_all(event, data);
406 	}
407 }
408 
409 static void perf_event__throttle_swap(union perf_event *event,
410 				      bool sample_id_all)
411 {
412 	event->throttle.time	  = bswap_64(event->throttle.time);
413 	event->throttle.id	  = bswap_64(event->throttle.id);
414 	event->throttle.stream_id = bswap_64(event->throttle.stream_id);
415 
416 	if (sample_id_all)
417 		swap_sample_id_all(event, &event->throttle + 1);
418 }
419 
420 static void perf_event__namespaces_swap(union perf_event *event,
421 					bool sample_id_all)
422 {
423 	u64 i;
424 
425 	event->namespaces.pid		= bswap_32(event->namespaces.pid);
426 	event->namespaces.tid		= bswap_32(event->namespaces.tid);
427 	event->namespaces.nr_namespaces	= bswap_64(event->namespaces.nr_namespaces);
428 
429 	for (i = 0; i < event->namespaces.nr_namespaces; i++) {
430 		struct perf_ns_link_info *ns = &event->namespaces.link_info[i];
431 
432 		ns->dev = bswap_64(ns->dev);
433 		ns->ino = bswap_64(ns->ino);
434 	}
435 
436 	if (sample_id_all)
437 		swap_sample_id_all(event, &event->namespaces.link_info[i]);
438 }
439 
440 static void perf_event__cgroup_swap(union perf_event *event, bool sample_id_all)
441 {
442 	event->cgroup.id = bswap_64(event->cgroup.id);
443 
444 	if (sample_id_all) {
445 		void *data = &event->cgroup.path;
446 
447 		data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
448 		swap_sample_id_all(event, data);
449 	}
450 }
451 
452 static u8 revbyte(u8 b)
453 {
454 	int rev = (b >> 4) | ((b & 0xf) << 4);
455 	rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2);
456 	rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1);
457 	return (u8) rev;
458 }
459 
460 /*
461  * XXX this is hack in attempt to carry flags bitfield
462  * through endian village. ABI says:
463  *
464  * Bit-fields are allocated from right to left (least to most significant)
465  * on little-endian implementations and from left to right (most to least
466  * significant) on big-endian implementations.
467  *
468  * The above seems to be byte specific, so we need to reverse each
469  * byte of the bitfield. 'Internet' also says this might be implementation
470  * specific and we probably need proper fix and carry perf_event_attr
471  * bitfield flags in separate data file FEAT_ section. Thought this seems
472  * to work for now.
473  */
474 static void swap_bitfield(u8 *p, unsigned len)
475 {
476 	unsigned i;
477 
478 	for (i = 0; i < len; i++) {
479 		*p = revbyte(*p);
480 		p++;
481 	}
482 }
483 
484 /* exported for swapping attributes in file header */
485 void perf_event__attr_swap(struct perf_event_attr *attr)
486 {
487 	attr->type		= bswap_32(attr->type);
488 	attr->size		= bswap_32(attr->size);
489 
490 #define bswap_safe(f, n) 					\
491 	(attr->size > (offsetof(struct perf_event_attr, f) + 	\
492 		       sizeof(attr->f) * (n)))
493 #define bswap_field(f, sz) 			\
494 do { 						\
495 	if (bswap_safe(f, 0))			\
496 		attr->f = bswap_##sz(attr->f);	\
497 } while(0)
498 #define bswap_field_16(f) bswap_field(f, 16)
499 #define bswap_field_32(f) bswap_field(f, 32)
500 #define bswap_field_64(f) bswap_field(f, 64)
501 
502 	bswap_field_64(config);
503 	bswap_field_64(sample_period);
504 	bswap_field_64(sample_type);
505 	bswap_field_64(read_format);
506 	bswap_field_32(wakeup_events);
507 	bswap_field_32(bp_type);
508 	bswap_field_64(bp_addr);
509 	bswap_field_64(bp_len);
510 	bswap_field_64(branch_sample_type);
511 	bswap_field_64(sample_regs_user);
512 	bswap_field_32(sample_stack_user);
513 	bswap_field_32(aux_watermark);
514 	bswap_field_16(sample_max_stack);
515 	bswap_field_32(aux_sample_size);
516 
517 	/*
518 	 * After read_format are bitfields. Check read_format because
519 	 * we are unable to use offsetof on bitfield.
520 	 */
521 	if (bswap_safe(read_format, 1))
522 		swap_bitfield((u8 *) (&attr->read_format + 1),
523 			      sizeof(u64));
524 #undef bswap_field_64
525 #undef bswap_field_32
526 #undef bswap_field
527 #undef bswap_safe
528 }
529 
530 static void perf_event__hdr_attr_swap(union perf_event *event,
531 				      bool sample_id_all __maybe_unused)
532 {
533 	size_t size;
534 
535 	perf_event__attr_swap(&event->attr.attr);
536 
537 	size = event->header.size;
538 	size -= perf_record_header_attr_id(event) - (void *)event;
539 	mem_bswap_64(perf_record_header_attr_id(event), size);
540 }
541 
542 static void perf_event__event_update_swap(union perf_event *event,
543 					  bool sample_id_all __maybe_unused)
544 {
545 	event->event_update.type = bswap_64(event->event_update.type);
546 	event->event_update.id   = bswap_64(event->event_update.id);
547 }
548 
549 static void perf_event__event_type_swap(union perf_event *event,
550 					bool sample_id_all __maybe_unused)
551 {
552 	event->event_type.event_type.event_id =
553 		bswap_64(event->event_type.event_type.event_id);
554 }
555 
556 static void perf_event__tracing_data_swap(union perf_event *event,
557 					  bool sample_id_all __maybe_unused)
558 {
559 	event->tracing_data.size = bswap_32(event->tracing_data.size);
560 }
561 
562 static void perf_event__auxtrace_info_swap(union perf_event *event,
563 					   bool sample_id_all __maybe_unused)
564 {
565 	size_t size;
566 
567 	event->auxtrace_info.type = bswap_32(event->auxtrace_info.type);
568 
569 	size = event->header.size;
570 	size -= (void *)&event->auxtrace_info.priv - (void *)event;
571 	mem_bswap_64(event->auxtrace_info.priv, size);
572 }
573 
574 static void perf_event__auxtrace_swap(union perf_event *event,
575 				      bool sample_id_all __maybe_unused)
576 {
577 	event->auxtrace.size      = bswap_64(event->auxtrace.size);
578 	event->auxtrace.offset    = bswap_64(event->auxtrace.offset);
579 	event->auxtrace.reference = bswap_64(event->auxtrace.reference);
580 	event->auxtrace.idx       = bswap_32(event->auxtrace.idx);
581 	event->auxtrace.tid       = bswap_32(event->auxtrace.tid);
582 	event->auxtrace.cpu       = bswap_32(event->auxtrace.cpu);
583 }
584 
585 static void perf_event__auxtrace_error_swap(union perf_event *event,
586 					    bool sample_id_all __maybe_unused)
587 {
588 	event->auxtrace_error.type = bswap_32(event->auxtrace_error.type);
589 	event->auxtrace_error.code = bswap_32(event->auxtrace_error.code);
590 	event->auxtrace_error.cpu  = bswap_32(event->auxtrace_error.cpu);
591 	event->auxtrace_error.pid  = bswap_32(event->auxtrace_error.pid);
592 	event->auxtrace_error.tid  = bswap_32(event->auxtrace_error.tid);
593 	event->auxtrace_error.fmt  = bswap_32(event->auxtrace_error.fmt);
594 	event->auxtrace_error.ip   = bswap_64(event->auxtrace_error.ip);
595 	if (event->auxtrace_error.fmt)
596 		event->auxtrace_error.time = bswap_64(event->auxtrace_error.time);
597 	if (event->auxtrace_error.fmt >= 2) {
598 		event->auxtrace_error.machine_pid = bswap_32(event->auxtrace_error.machine_pid);
599 		event->auxtrace_error.vcpu = bswap_32(event->auxtrace_error.vcpu);
600 	}
601 }
602 
603 static void perf_event__thread_map_swap(union perf_event *event,
604 					bool sample_id_all __maybe_unused)
605 {
606 	unsigned i;
607 
608 	event->thread_map.nr = bswap_64(event->thread_map.nr);
609 
610 	for (i = 0; i < event->thread_map.nr; i++)
611 		event->thread_map.entries[i].pid = bswap_64(event->thread_map.entries[i].pid);
612 }
613 
614 static void perf_event__cpu_map_swap(union perf_event *event,
615 				     bool sample_id_all __maybe_unused)
616 {
617 	struct perf_record_cpu_map_data *data = &event->cpu_map.data;
618 
619 	data->type = bswap_16(data->type);
620 
621 	switch (data->type) {
622 	case PERF_CPU_MAP__CPUS:
623 		data->cpus_data.nr = bswap_16(data->cpus_data.nr);
624 
625 		for (unsigned i = 0; i < data->cpus_data.nr; i++)
626 			data->cpus_data.cpu[i] = bswap_16(data->cpus_data.cpu[i]);
627 		break;
628 	case PERF_CPU_MAP__MASK:
629 		data->mask32_data.long_size = bswap_16(data->mask32_data.long_size);
630 
631 		switch (data->mask32_data.long_size) {
632 		case 4:
633 			data->mask32_data.nr = bswap_16(data->mask32_data.nr);
634 			for (unsigned i = 0; i < data->mask32_data.nr; i++)
635 				data->mask32_data.mask[i] = bswap_32(data->mask32_data.mask[i]);
636 			break;
637 		case 8:
638 			data->mask64_data.nr = bswap_16(data->mask64_data.nr);
639 			for (unsigned i = 0; i < data->mask64_data.nr; i++)
640 				data->mask64_data.mask[i] = bswap_64(data->mask64_data.mask[i]);
641 			break;
642 		default:
643 			pr_err("cpu_map swap: unsupported long size\n");
644 		}
645 		break;
646 	case PERF_CPU_MAP__RANGE_CPUS:
647 		data->range_cpu_data.start_cpu = bswap_16(data->range_cpu_data.start_cpu);
648 		data->range_cpu_data.end_cpu = bswap_16(data->range_cpu_data.end_cpu);
649 		break;
650 	default:
651 		break;
652 	}
653 }
654 
655 static void perf_event__stat_config_swap(union perf_event *event,
656 					 bool sample_id_all __maybe_unused)
657 {
658 	u64 size;
659 
660 	size  = bswap_64(event->stat_config.nr) * sizeof(event->stat_config.data[0]);
661 	size += 1; /* nr item itself */
662 	mem_bswap_64(&event->stat_config.nr, size);
663 }
664 
665 static void perf_event__stat_swap(union perf_event *event,
666 				  bool sample_id_all __maybe_unused)
667 {
668 	event->stat.id     = bswap_64(event->stat.id);
669 	event->stat.thread = bswap_32(event->stat.thread);
670 	event->stat.cpu    = bswap_32(event->stat.cpu);
671 	event->stat.val    = bswap_64(event->stat.val);
672 	event->stat.ena    = bswap_64(event->stat.ena);
673 	event->stat.run    = bswap_64(event->stat.run);
674 }
675 
676 static void perf_event__stat_round_swap(union perf_event *event,
677 					bool sample_id_all __maybe_unused)
678 {
679 	event->stat_round.type = bswap_64(event->stat_round.type);
680 	event->stat_round.time = bswap_64(event->stat_round.time);
681 }
682 
683 static void perf_event__time_conv_swap(union perf_event *event,
684 				       bool sample_id_all __maybe_unused)
685 {
686 	event->time_conv.time_shift = bswap_64(event->time_conv.time_shift);
687 	event->time_conv.time_mult  = bswap_64(event->time_conv.time_mult);
688 	event->time_conv.time_zero  = bswap_64(event->time_conv.time_zero);
689 
690 	if (event_contains(event->time_conv, time_cycles)) {
691 		event->time_conv.time_cycles = bswap_64(event->time_conv.time_cycles);
692 		event->time_conv.time_mask = bswap_64(event->time_conv.time_mask);
693 	}
694 }
695 
696 typedef void (*perf_event__swap_op)(union perf_event *event,
697 				    bool sample_id_all);
698 
699 static perf_event__swap_op perf_event__swap_ops[] = {
700 	[PERF_RECORD_MMAP]		  = perf_event__mmap_swap,
701 	[PERF_RECORD_MMAP2]		  = perf_event__mmap2_swap,
702 	[PERF_RECORD_COMM]		  = perf_event__comm_swap,
703 	[PERF_RECORD_FORK]		  = perf_event__task_swap,
704 	[PERF_RECORD_EXIT]		  = perf_event__task_swap,
705 	[PERF_RECORD_LOST]		  = perf_event__all64_swap,
706 	[PERF_RECORD_READ]		  = perf_event__read_swap,
707 	[PERF_RECORD_THROTTLE]		  = perf_event__throttle_swap,
708 	[PERF_RECORD_UNTHROTTLE]	  = perf_event__throttle_swap,
709 	[PERF_RECORD_SAMPLE]		  = perf_event__all64_swap,
710 	[PERF_RECORD_AUX]		  = perf_event__aux_swap,
711 	[PERF_RECORD_ITRACE_START]	  = perf_event__itrace_start_swap,
712 	[PERF_RECORD_LOST_SAMPLES]	  = perf_event__all64_swap,
713 	[PERF_RECORD_SWITCH]		  = perf_event__switch_swap,
714 	[PERF_RECORD_SWITCH_CPU_WIDE]	  = perf_event__switch_swap,
715 	[PERF_RECORD_NAMESPACES]	  = perf_event__namespaces_swap,
716 	[PERF_RECORD_CGROUP]		  = perf_event__cgroup_swap,
717 	[PERF_RECORD_TEXT_POKE]		  = perf_event__text_poke_swap,
718 	[PERF_RECORD_AUX_OUTPUT_HW_ID]	  = perf_event__all64_swap,
719 	[PERF_RECORD_HEADER_ATTR]	  = perf_event__hdr_attr_swap,
720 	[PERF_RECORD_HEADER_EVENT_TYPE]	  = perf_event__event_type_swap,
721 	[PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
722 	[PERF_RECORD_HEADER_BUILD_ID]	  = NULL,
723 	[PERF_RECORD_ID_INDEX]		  = perf_event__all64_swap,
724 	[PERF_RECORD_AUXTRACE_INFO]	  = perf_event__auxtrace_info_swap,
725 	[PERF_RECORD_AUXTRACE]		  = perf_event__auxtrace_swap,
726 	[PERF_RECORD_AUXTRACE_ERROR]	  = perf_event__auxtrace_error_swap,
727 	[PERF_RECORD_THREAD_MAP]	  = perf_event__thread_map_swap,
728 	[PERF_RECORD_CPU_MAP]		  = perf_event__cpu_map_swap,
729 	[PERF_RECORD_STAT_CONFIG]	  = perf_event__stat_config_swap,
730 	[PERF_RECORD_STAT]		  = perf_event__stat_swap,
731 	[PERF_RECORD_STAT_ROUND]	  = perf_event__stat_round_swap,
732 	[PERF_RECORD_EVENT_UPDATE]	  = perf_event__event_update_swap,
733 	[PERF_RECORD_TIME_CONV]		  = perf_event__time_conv_swap,
734 	[PERF_RECORD_HEADER_MAX]	  = NULL,
735 };
736 
737 /*
738  * When perf record finishes a pass on every buffers, it records this pseudo
739  * event.
740  * We record the max timestamp t found in the pass n.
741  * Assuming these timestamps are monotonic across cpus, we know that if
742  * a buffer still has events with timestamps below t, they will be all
743  * available and then read in the pass n + 1.
744  * Hence when we start to read the pass n + 2, we can safely flush every
745  * events with timestamps below t.
746  *
747  *    ============ PASS n =================
748  *       CPU 0         |   CPU 1
749  *                     |
750  *    cnt1 timestamps  |   cnt2 timestamps
751  *          1          |         2
752  *          2          |         3
753  *          -          |         4  <--- max recorded
754  *
755  *    ============ PASS n + 1 ==============
756  *       CPU 0         |   CPU 1
757  *                     |
758  *    cnt1 timestamps  |   cnt2 timestamps
759  *          3          |         5
760  *          4          |         6
761  *          5          |         7 <---- max recorded
762  *
763  *      Flush every events below timestamp 4
764  *
765  *    ============ PASS n + 2 ==============
766  *       CPU 0         |   CPU 1
767  *                     |
768  *    cnt1 timestamps  |   cnt2 timestamps
769  *          6          |         8
770  *          7          |         9
771  *          -          |         10
772  *
773  *      Flush every events below timestamp 7
774  *      etc...
775  */
776 int perf_event__process_finished_round(const struct perf_tool *tool __maybe_unused,
777 				       union perf_event *event __maybe_unused,
778 				       struct ordered_events *oe)
779 {
780 	if (dump_trace)
781 		fprintf(stdout, "\n");
782 	return ordered_events__flush(oe, OE_FLUSH__ROUND);
783 }
784 
785 int perf_session__queue_event(struct perf_session *s, union perf_event *event,
786 			      u64 timestamp, u64 file_offset, const char *file_path)
787 {
788 	return ordered_events__queue(&s->ordered_events, event, timestamp, file_offset, file_path);
789 }
790 
791 static void callchain__lbr_callstack_printf(struct perf_sample *sample)
792 {
793 	struct ip_callchain *callchain = sample->callchain;
794 	struct branch_stack *lbr_stack = sample->branch_stack;
795 	struct branch_entry *entries = perf_sample__branch_entries(sample);
796 	u64 kernel_callchain_nr = callchain->nr;
797 	unsigned int i;
798 
799 	for (i = 0; i < kernel_callchain_nr; i++) {
800 		if (callchain->ips[i] == PERF_CONTEXT_USER)
801 			break;
802 	}
803 
804 	if ((i != kernel_callchain_nr) && lbr_stack->nr) {
805 		u64 total_nr;
806 		/*
807 		 * LBR callstack can only get user call chain,
808 		 * i is kernel call chain number,
809 		 * 1 is PERF_CONTEXT_USER.
810 		 *
811 		 * The user call chain is stored in LBR registers.
812 		 * LBR are pair registers. The caller is stored
813 		 * in "from" register, while the callee is stored
814 		 * in "to" register.
815 		 * For example, there is a call stack
816 		 * "A"->"B"->"C"->"D".
817 		 * The LBR registers will be recorded like
818 		 * "C"->"D", "B"->"C", "A"->"B".
819 		 * So only the first "to" register and all "from"
820 		 * registers are needed to construct the whole stack.
821 		 */
822 		total_nr = i + 1 + lbr_stack->nr + 1;
823 		kernel_callchain_nr = i + 1;
824 
825 		printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr);
826 
827 		for (i = 0; i < kernel_callchain_nr; i++)
828 			printf("..... %2d: %016" PRIx64 "\n",
829 			       i, callchain->ips[i]);
830 
831 		printf("..... %2d: %016" PRIx64 "\n",
832 		       (int)(kernel_callchain_nr), entries[0].to);
833 		for (i = 0; i < lbr_stack->nr; i++)
834 			printf("..... %2d: %016" PRIx64 "\n",
835 			       (int)(i + kernel_callchain_nr + 1), entries[i].from);
836 	}
837 }
838 
839 static void callchain__printf(struct evsel *evsel,
840 			      struct perf_sample *sample)
841 {
842 	unsigned int i;
843 	struct ip_callchain *callchain = sample->callchain;
844 
845 	if (evsel__has_branch_callstack(evsel))
846 		callchain__lbr_callstack_printf(sample);
847 
848 	printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr);
849 
850 	for (i = 0; i < callchain->nr; i++)
851 		printf("..... %2d: %016" PRIx64 "\n",
852 		       i, callchain->ips[i]);
853 }
854 
855 static void branch_stack__printf(struct perf_sample *sample,
856 				 struct evsel *evsel)
857 {
858 	struct branch_entry *entries = perf_sample__branch_entries(sample);
859 	bool callstack = evsel__has_branch_callstack(evsel);
860 	u64 *branch_stack_cntr = sample->branch_stack_cntr;
861 	uint64_t i;
862 
863 	if (!callstack) {
864 		printf("%s: nr:%" PRIu64 "\n", "... branch stack", sample->branch_stack->nr);
865 	} else {
866 		/* the reason of adding 1 to nr is because after expanding
867 		 * branch stack it generates nr + 1 callstack records. e.g.,
868 		 *         B()->C()
869 		 *         A()->B()
870 		 * the final callstack should be:
871 		 *         C()
872 		 *         B()
873 		 *         A()
874 		 */
875 		printf("%s: nr:%" PRIu64 "\n", "... branch callstack", sample->branch_stack->nr+1);
876 	}
877 
878 	for (i = 0; i < sample->branch_stack->nr; i++) {
879 		struct branch_entry *e = &entries[i];
880 
881 		if (!callstack) {
882 			printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x %s %s\n",
883 				i, e->from, e->to,
884 				(unsigned short)e->flags.cycles,
885 				e->flags.mispred ? "M" : " ",
886 				e->flags.predicted ? "P" : " ",
887 				e->flags.abort ? "A" : " ",
888 				e->flags.in_tx ? "T" : " ",
889 				(unsigned)e->flags.reserved,
890 				get_branch_type(e),
891 				e->flags.spec ? branch_spec_desc(e->flags.spec) : "");
892 		} else {
893 			if (i == 0) {
894 				printf("..... %2"PRIu64": %016" PRIx64 "\n"
895 				       "..... %2"PRIu64": %016" PRIx64 "\n",
896 						i, e->to, i+1, e->from);
897 			} else {
898 				printf("..... %2"PRIu64": %016" PRIx64 "\n", i+1, e->from);
899 			}
900 		}
901 	}
902 
903 	if (branch_stack_cntr) {
904 		unsigned int br_cntr_width, br_cntr_nr;
905 
906 		perf_env__find_br_cntr_info(evsel__env(evsel), &br_cntr_nr, &br_cntr_width);
907 		printf("... branch stack counters: nr:%" PRIu64 " (counter width: %u max counter nr:%u)\n",
908 			sample->branch_stack->nr, br_cntr_width, br_cntr_nr);
909 		for (i = 0; i < sample->branch_stack->nr; i++)
910 			printf("..... %2"PRIu64": %016" PRIx64 "\n", i, branch_stack_cntr[i]);
911 	}
912 }
913 
914 static void regs_dump__printf(u64 mask, u64 *regs, const char *arch)
915 {
916 	unsigned rid, i = 0;
917 
918 	for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) {
919 		u64 val = regs[i++];
920 
921 		printf(".... %-5s 0x%016" PRIx64 "\n",
922 		       perf_reg_name(rid, arch), val);
923 	}
924 }
925 
926 static const char *regs_abi[] = {
927 	[PERF_SAMPLE_REGS_ABI_NONE] = "none",
928 	[PERF_SAMPLE_REGS_ABI_32] = "32-bit",
929 	[PERF_SAMPLE_REGS_ABI_64] = "64-bit",
930 };
931 
932 static inline const char *regs_dump_abi(struct regs_dump *d)
933 {
934 	if (d->abi > PERF_SAMPLE_REGS_ABI_64)
935 		return "unknown";
936 
937 	return regs_abi[d->abi];
938 }
939 
940 static void regs__printf(const char *type, struct regs_dump *regs, const char *arch)
941 {
942 	u64 mask = regs->mask;
943 
944 	printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n",
945 	       type,
946 	       mask,
947 	       regs_dump_abi(regs));
948 
949 	regs_dump__printf(mask, regs->regs, arch);
950 }
951 
952 static void regs_user__printf(struct perf_sample *sample, const char *arch)
953 {
954 	struct regs_dump *user_regs;
955 
956 	if (!sample->user_regs)
957 		return;
958 
959 	user_regs = perf_sample__user_regs(sample);
960 
961 	if (user_regs->regs)
962 		regs__printf("user", user_regs, arch);
963 }
964 
965 static void regs_intr__printf(struct perf_sample *sample, const char *arch)
966 {
967 	struct regs_dump *intr_regs;
968 
969 	if (!sample->intr_regs)
970 		return;
971 
972 	intr_regs = perf_sample__intr_regs(sample);
973 
974 	if (intr_regs->regs)
975 		regs__printf("intr", intr_regs, arch);
976 }
977 
978 static void stack_user__printf(struct stack_dump *dump)
979 {
980 	printf("... ustack: size %" PRIu64 ", offset 0x%x\n",
981 	       dump->size, dump->offset);
982 }
983 
984 static void evlist__print_tstamp(struct evlist *evlist, union perf_event *event, struct perf_sample *sample)
985 {
986 	u64 sample_type = __evlist__combined_sample_type(evlist);
987 
988 	if (event->header.type != PERF_RECORD_SAMPLE &&
989 	    !evlist__sample_id_all(evlist)) {
990 		fputs("-1 -1 ", stdout);
991 		return;
992 	}
993 
994 	if ((sample_type & PERF_SAMPLE_CPU))
995 		printf("%u ", sample->cpu);
996 
997 	if (sample_type & PERF_SAMPLE_TIME)
998 		printf("%" PRIu64 " ", sample->time);
999 }
1000 
1001 static void sample_read__printf(struct perf_sample *sample, u64 read_format)
1002 {
1003 	printf("... sample_read:\n");
1004 
1005 	if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1006 		printf("...... time enabled %016" PRIx64 "\n",
1007 		       sample->read.time_enabled);
1008 
1009 	if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1010 		printf("...... time running %016" PRIx64 "\n",
1011 		       sample->read.time_running);
1012 
1013 	if (read_format & PERF_FORMAT_GROUP) {
1014 		struct sample_read_value *value = sample->read.group.values;
1015 
1016 		printf(".... group nr %" PRIu64 "\n", sample->read.group.nr);
1017 
1018 		sample_read_group__for_each(value, sample->read.group.nr, read_format) {
1019 			printf("..... id %016" PRIx64
1020 			       ", value %016" PRIx64,
1021 			       value->id, value->value);
1022 			if (read_format & PERF_FORMAT_LOST)
1023 				printf(", lost %" PRIu64, value->lost);
1024 			printf("\n");
1025 		}
1026 	} else {
1027 		printf("..... id %016" PRIx64 ", value %016" PRIx64,
1028 			sample->read.one.id, sample->read.one.value);
1029 		if (read_format & PERF_FORMAT_LOST)
1030 			printf(", lost %" PRIu64, sample->read.one.lost);
1031 		printf("\n");
1032 	}
1033 }
1034 
1035 static void dump_event(struct evlist *evlist, union perf_event *event,
1036 		       u64 file_offset, struct perf_sample *sample,
1037 		       const char *file_path)
1038 {
1039 	if (!dump_trace)
1040 		return;
1041 
1042 	printf("\n%#" PRIx64 "@%s [%#x]: event: %d\n",
1043 	       file_offset, file_path, event->header.size, event->header.type);
1044 
1045 	trace_event(event);
1046 	if (event->header.type == PERF_RECORD_SAMPLE && evlist->trace_event_sample_raw)
1047 		evlist->trace_event_sample_raw(evlist, event, sample);
1048 
1049 	if (sample)
1050 		evlist__print_tstamp(evlist, event, sample);
1051 
1052 	printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset,
1053 	       event->header.size, perf_event__name(event->header.type));
1054 }
1055 
1056 char *get_page_size_name(u64 size, char *str)
1057 {
1058 	if (!size || !unit_number__scnprintf(str, PAGE_SIZE_NAME_LEN, size))
1059 		snprintf(str, PAGE_SIZE_NAME_LEN, "%s", "N/A");
1060 
1061 	return str;
1062 }
1063 
1064 static void dump_sample(struct evsel *evsel, union perf_event *event,
1065 			struct perf_sample *sample, const char *arch)
1066 {
1067 	u64 sample_type;
1068 	char str[PAGE_SIZE_NAME_LEN];
1069 
1070 	if (!dump_trace)
1071 		return;
1072 
1073 	printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n",
1074 	       event->header.misc, sample->pid, sample->tid, sample->ip,
1075 	       sample->period, sample->addr);
1076 
1077 	sample_type = evsel->core.attr.sample_type;
1078 
1079 	if (evsel__has_callchain(evsel))
1080 		callchain__printf(evsel, sample);
1081 
1082 	if (evsel__has_br_stack(evsel))
1083 		branch_stack__printf(sample, evsel);
1084 
1085 	if (sample_type & PERF_SAMPLE_REGS_USER)
1086 		regs_user__printf(sample, arch);
1087 
1088 	if (sample_type & PERF_SAMPLE_REGS_INTR)
1089 		regs_intr__printf(sample, arch);
1090 
1091 	if (sample_type & PERF_SAMPLE_STACK_USER)
1092 		stack_user__printf(&sample->user_stack);
1093 
1094 	if (sample_type & PERF_SAMPLE_WEIGHT_TYPE) {
1095 		printf("... weight: %" PRIu64 "", sample->weight);
1096 			if (sample_type & PERF_SAMPLE_WEIGHT_STRUCT) {
1097 				printf(",0x%"PRIx16"", sample->ins_lat);
1098 				printf(",0x%"PRIx16"", sample->p_stage_cyc);
1099 			}
1100 		printf("\n");
1101 	}
1102 
1103 	if (sample_type & PERF_SAMPLE_DATA_SRC)
1104 		printf(" . data_src: 0x%"PRIx64"\n", sample->data_src);
1105 
1106 	if (sample_type & PERF_SAMPLE_PHYS_ADDR)
1107 		printf(" .. phys_addr: 0x%"PRIx64"\n", sample->phys_addr);
1108 
1109 	if (sample_type & PERF_SAMPLE_DATA_PAGE_SIZE)
1110 		printf(" .. data page size: %s\n", get_page_size_name(sample->data_page_size, str));
1111 
1112 	if (sample_type & PERF_SAMPLE_CODE_PAGE_SIZE)
1113 		printf(" .. code page size: %s\n", get_page_size_name(sample->code_page_size, str));
1114 
1115 	if (sample_type & PERF_SAMPLE_TRANSACTION)
1116 		printf("... transaction: %" PRIx64 "\n", sample->transaction);
1117 
1118 	if (sample_type & PERF_SAMPLE_READ)
1119 		sample_read__printf(sample, evsel->core.attr.read_format);
1120 }
1121 
1122 static void dump_read(struct evsel *evsel, union perf_event *event)
1123 {
1124 	struct perf_record_read *read_event = &event->read;
1125 	u64 read_format;
1126 
1127 	if (!dump_trace)
1128 		return;
1129 
1130 	printf(": %d %d %s %" PRI_lu64 "\n", event->read.pid, event->read.tid,
1131 	       evsel__name(evsel), event->read.value);
1132 
1133 	if (!evsel)
1134 		return;
1135 
1136 	read_format = evsel->core.attr.read_format;
1137 
1138 	if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1139 		printf("... time enabled : %" PRI_lu64 "\n", read_event->time_enabled);
1140 
1141 	if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1142 		printf("... time running : %" PRI_lu64 "\n", read_event->time_running);
1143 
1144 	if (read_format & PERF_FORMAT_ID)
1145 		printf("... id           : %" PRI_lu64 "\n", read_event->id);
1146 
1147 	if (read_format & PERF_FORMAT_LOST)
1148 		printf("... lost         : %" PRI_lu64 "\n", read_event->lost);
1149 }
1150 
1151 static struct machine *machines__find_for_cpumode(struct machines *machines,
1152 					       union perf_event *event,
1153 					       struct perf_sample *sample)
1154 {
1155 	if (perf_guest &&
1156 	    ((sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL) ||
1157 	     (sample->cpumode == PERF_RECORD_MISC_GUEST_USER))) {
1158 		u32 pid;
1159 
1160 		if (sample->machine_pid)
1161 			pid = sample->machine_pid;
1162 		else if (event->header.type == PERF_RECORD_MMAP
1163 		    || event->header.type == PERF_RECORD_MMAP2)
1164 			pid = event->mmap.pid;
1165 		else
1166 			pid = sample->pid;
1167 
1168 		/*
1169 		 * Guest code machine is created as needed and does not use
1170 		 * DEFAULT_GUEST_KERNEL_ID.
1171 		 */
1172 		if (symbol_conf.guest_code)
1173 			return machines__findnew(machines, pid);
1174 
1175 		return machines__find_guest(machines, pid);
1176 	}
1177 
1178 	return &machines->host;
1179 }
1180 
1181 static int deliver_sample_value(struct evlist *evlist,
1182 				const struct perf_tool *tool,
1183 				union perf_event *event,
1184 				struct perf_sample *sample,
1185 				struct sample_read_value *v,
1186 				struct machine *machine,
1187 				bool per_thread)
1188 {
1189 	struct perf_sample_id *sid = evlist__id2sid(evlist, v->id);
1190 	struct evsel *evsel;
1191 	u64 *storage = NULL;
1192 
1193 	if (sid) {
1194 		storage = perf_sample_id__get_period_storage(sid, sample->tid, per_thread);
1195 	}
1196 
1197 	if (storage) {
1198 		sample->id     = v->id;
1199 		sample->period = v->value - *storage;
1200 		*storage       = v->value;
1201 	}
1202 
1203 	if (!storage || sid->evsel == NULL) {
1204 		++evlist->stats.nr_unknown_id;
1205 		return 0;
1206 	}
1207 
1208 	/*
1209 	 * There's no reason to deliver sample
1210 	 * for zero period, bail out.
1211 	 */
1212 	if (!sample->period)
1213 		return 0;
1214 
1215 	evsel = container_of(sid->evsel, struct evsel, core);
1216 	return tool->sample(tool, event, sample, evsel, machine);
1217 }
1218 
1219 static int deliver_sample_group(struct evlist *evlist,
1220 				const struct perf_tool *tool,
1221 				union  perf_event *event,
1222 				struct perf_sample *sample,
1223 				struct machine *machine,
1224 				u64 read_format,
1225 				bool per_thread)
1226 {
1227 	int ret = -EINVAL;
1228 	struct sample_read_value *v = sample->read.group.values;
1229 
1230 	if (tool->dont_split_sample_group)
1231 		return deliver_sample_value(evlist, tool, event, sample, v, machine,
1232 					    per_thread);
1233 
1234 	sample_read_group__for_each(v, sample->read.group.nr, read_format) {
1235 		ret = deliver_sample_value(evlist, tool, event, sample, v,
1236 					   machine, per_thread);
1237 		if (ret)
1238 			break;
1239 	}
1240 
1241 	return ret;
1242 }
1243 
1244 static int evlist__deliver_sample(struct evlist *evlist, const struct perf_tool *tool,
1245 				  union  perf_event *event, struct perf_sample *sample,
1246 				  struct evsel *evsel, struct machine *machine)
1247 {
1248 	/* We know evsel != NULL. */
1249 	u64 sample_type = evsel->core.attr.sample_type;
1250 	u64 read_format = evsel->core.attr.read_format;
1251 	bool per_thread = perf_evsel__attr_has_per_thread_sample_period(&evsel->core);
1252 
1253 	/* Standard sample delivery. */
1254 	if (!(sample_type & PERF_SAMPLE_READ))
1255 		return tool->sample(tool, event, sample, evsel, machine);
1256 
1257 	/* For PERF_SAMPLE_READ we have either single or group mode. */
1258 	if (read_format & PERF_FORMAT_GROUP)
1259 		return deliver_sample_group(evlist, tool, event, sample,
1260 					    machine, read_format, per_thread);
1261 	else
1262 		return deliver_sample_value(evlist, tool, event, sample,
1263 					    &sample->read.one, machine,
1264 					    per_thread);
1265 }
1266 
1267 static int machines__deliver_event(struct machines *machines,
1268 				   struct evlist *evlist,
1269 				   union perf_event *event,
1270 				   struct perf_sample *sample,
1271 				   const struct perf_tool *tool, u64 file_offset,
1272 				   const char *file_path)
1273 {
1274 	struct evsel *evsel;
1275 	struct machine *machine;
1276 
1277 	dump_event(evlist, event, file_offset, sample, file_path);
1278 
1279 	evsel = evlist__id2evsel(evlist, sample->id);
1280 
1281 	machine = machines__find_for_cpumode(machines, event, sample);
1282 
1283 	switch (event->header.type) {
1284 	case PERF_RECORD_SAMPLE:
1285 		if (evsel == NULL) {
1286 			++evlist->stats.nr_unknown_id;
1287 			return 0;
1288 		}
1289 		if (machine == NULL) {
1290 			++evlist->stats.nr_unprocessable_samples;
1291 			dump_sample(evsel, event, sample, perf_env__arch(NULL));
1292 			return 0;
1293 		}
1294 		dump_sample(evsel, event, sample, perf_env__arch(machine->env));
1295 		return evlist__deliver_sample(evlist, tool, event, sample, evsel, machine);
1296 	case PERF_RECORD_MMAP:
1297 		return tool->mmap(tool, event, sample, machine);
1298 	case PERF_RECORD_MMAP2:
1299 		if (event->header.misc & PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT)
1300 			++evlist->stats.nr_proc_map_timeout;
1301 		return tool->mmap2(tool, event, sample, machine);
1302 	case PERF_RECORD_COMM:
1303 		return tool->comm(tool, event, sample, machine);
1304 	case PERF_RECORD_NAMESPACES:
1305 		return tool->namespaces(tool, event, sample, machine);
1306 	case PERF_RECORD_CGROUP:
1307 		return tool->cgroup(tool, event, sample, machine);
1308 	case PERF_RECORD_FORK:
1309 		return tool->fork(tool, event, sample, machine);
1310 	case PERF_RECORD_EXIT:
1311 		return tool->exit(tool, event, sample, machine);
1312 	case PERF_RECORD_LOST:
1313 		if (tool->lost == perf_event__process_lost)
1314 			evlist->stats.total_lost += event->lost.lost;
1315 		return tool->lost(tool, event, sample, machine);
1316 	case PERF_RECORD_LOST_SAMPLES:
1317 		if (event->header.misc & PERF_RECORD_MISC_LOST_SAMPLES_BPF)
1318 			evlist->stats.total_dropped_samples += event->lost_samples.lost;
1319 		else if (tool->lost_samples == perf_event__process_lost_samples)
1320 			evlist->stats.total_lost_samples += event->lost_samples.lost;
1321 		return tool->lost_samples(tool, event, sample, machine);
1322 	case PERF_RECORD_READ:
1323 		dump_read(evsel, event);
1324 		return tool->read(tool, event, sample, evsel, machine);
1325 	case PERF_RECORD_THROTTLE:
1326 		return tool->throttle(tool, event, sample, machine);
1327 	case PERF_RECORD_UNTHROTTLE:
1328 		return tool->unthrottle(tool, event, sample, machine);
1329 	case PERF_RECORD_AUX:
1330 		if (tool->aux == perf_event__process_aux) {
1331 			if (event->aux.flags & PERF_AUX_FLAG_TRUNCATED)
1332 				evlist->stats.total_aux_lost += 1;
1333 			if (event->aux.flags & PERF_AUX_FLAG_PARTIAL)
1334 				evlist->stats.total_aux_partial += 1;
1335 			if (event->aux.flags & PERF_AUX_FLAG_COLLISION)
1336 				evlist->stats.total_aux_collision += 1;
1337 		}
1338 		return tool->aux(tool, event, sample, machine);
1339 	case PERF_RECORD_ITRACE_START:
1340 		return tool->itrace_start(tool, event, sample, machine);
1341 	case PERF_RECORD_SWITCH:
1342 	case PERF_RECORD_SWITCH_CPU_WIDE:
1343 		return tool->context_switch(tool, event, sample, machine);
1344 	case PERF_RECORD_KSYMBOL:
1345 		return tool->ksymbol(tool, event, sample, machine);
1346 	case PERF_RECORD_BPF_EVENT:
1347 		return tool->bpf(tool, event, sample, machine);
1348 	case PERF_RECORD_TEXT_POKE:
1349 		return tool->text_poke(tool, event, sample, machine);
1350 	case PERF_RECORD_AUX_OUTPUT_HW_ID:
1351 		return tool->aux_output_hw_id(tool, event, sample, machine);
1352 	default:
1353 		++evlist->stats.nr_unknown_events;
1354 		return -1;
1355 	}
1356 }
1357 
1358 static int perf_session__deliver_event(struct perf_session *session,
1359 				       union perf_event *event,
1360 				       const struct perf_tool *tool,
1361 				       u64 file_offset,
1362 				       const char *file_path)
1363 {
1364 	struct perf_sample sample;
1365 	int ret;
1366 
1367 	perf_sample__init(&sample, /*all=*/false);
1368 	ret = evlist__parse_sample(session->evlist, event, &sample);
1369 	if (ret) {
1370 		pr_err("Can't parse sample, err = %d\n", ret);
1371 		goto out;
1372 	}
1373 
1374 	ret = auxtrace__process_event(session, event, &sample, tool);
1375 	if (ret < 0)
1376 		goto out;
1377 	if (ret > 0) {
1378 		ret = 0;
1379 		goto out;
1380 	}
1381 
1382 	ret = machines__deliver_event(&session->machines, session->evlist,
1383 				      event, &sample, tool, file_offset, file_path);
1384 
1385 	if (dump_trace && sample.aux_sample.size)
1386 		auxtrace__dump_auxtrace_sample(session, &sample);
1387 out:
1388 	perf_sample__exit(&sample);
1389 	return ret;
1390 }
1391 
1392 static s64 perf_session__process_user_event(struct perf_session *session,
1393 					    union perf_event *event,
1394 					    u64 file_offset,
1395 					    const char *file_path)
1396 {
1397 	struct ordered_events *oe = &session->ordered_events;
1398 	const struct perf_tool *tool = session->tool;
1399 	struct perf_sample sample;
1400 	int fd = perf_data__fd(session->data);
1401 	int err;
1402 
1403 	perf_sample__init(&sample, /*all=*/true);
1404 	if ((event->header.type != PERF_RECORD_COMPRESSED &&
1405 	     event->header.type != PERF_RECORD_COMPRESSED2) ||
1406 	    perf_tool__compressed_is_stub(tool))
1407 		dump_event(session->evlist, event, file_offset, &sample, file_path);
1408 
1409 	/* These events are processed right away */
1410 	switch (event->header.type) {
1411 	case PERF_RECORD_HEADER_ATTR:
1412 		err = tool->attr(tool, event, &session->evlist);
1413 		if (err == 0) {
1414 			perf_session__set_id_hdr_size(session);
1415 			perf_session__set_comm_exec(session);
1416 		}
1417 		break;
1418 	case PERF_RECORD_EVENT_UPDATE:
1419 		err = tool->event_update(tool, event, &session->evlist);
1420 		break;
1421 	case PERF_RECORD_HEADER_EVENT_TYPE:
1422 		/*
1423 		 * Deprecated, but we need to handle it for sake
1424 		 * of old data files create in pipe mode.
1425 		 */
1426 		err = 0;
1427 		break;
1428 	case PERF_RECORD_HEADER_TRACING_DATA:
1429 		/*
1430 		 * Setup for reading amidst mmap, but only when we
1431 		 * are in 'file' mode. The 'pipe' fd is in proper
1432 		 * place already.
1433 		 */
1434 		if (!perf_data__is_pipe(session->data))
1435 			lseek(fd, file_offset, SEEK_SET);
1436 		err = tool->tracing_data(session, event);
1437 		break;
1438 	case PERF_RECORD_HEADER_BUILD_ID:
1439 		err = tool->build_id(session, event);
1440 		break;
1441 	case PERF_RECORD_FINISHED_ROUND:
1442 		err = tool->finished_round(tool, event, oe);
1443 		break;
1444 	case PERF_RECORD_ID_INDEX:
1445 		err = tool->id_index(session, event);
1446 		break;
1447 	case PERF_RECORD_AUXTRACE_INFO:
1448 		err = tool->auxtrace_info(session, event);
1449 		break;
1450 	case PERF_RECORD_AUXTRACE:
1451 		/*
1452 		 * Setup for reading amidst mmap, but only when we
1453 		 * are in 'file' mode.  The 'pipe' fd is in proper
1454 		 * place already.
1455 		 */
1456 		if (!perf_data__is_pipe(session->data))
1457 			lseek(fd, file_offset + event->header.size, SEEK_SET);
1458 		err = tool->auxtrace(session, event);
1459 		break;
1460 	case PERF_RECORD_AUXTRACE_ERROR:
1461 		perf_session__auxtrace_error_inc(session, event);
1462 		err = tool->auxtrace_error(session, event);
1463 		break;
1464 	case PERF_RECORD_THREAD_MAP:
1465 		err = tool->thread_map(session, event);
1466 		break;
1467 	case PERF_RECORD_CPU_MAP:
1468 		err = tool->cpu_map(session, event);
1469 		break;
1470 	case PERF_RECORD_STAT_CONFIG:
1471 		err = tool->stat_config(session, event);
1472 		break;
1473 	case PERF_RECORD_STAT:
1474 		err = tool->stat(session, event);
1475 		break;
1476 	case PERF_RECORD_STAT_ROUND:
1477 		err = tool->stat_round(session, event);
1478 		break;
1479 	case PERF_RECORD_TIME_CONV:
1480 		session->time_conv = event->time_conv;
1481 		err = tool->time_conv(session, event);
1482 		break;
1483 	case PERF_RECORD_HEADER_FEATURE:
1484 		err = tool->feature(session, event);
1485 		break;
1486 	case PERF_RECORD_COMPRESSED:
1487 	case PERF_RECORD_COMPRESSED2:
1488 		err = tool->compressed(session, event, file_offset, file_path);
1489 		if (err)
1490 			dump_event(session->evlist, event, file_offset, &sample, file_path);
1491 		break;
1492 	case PERF_RECORD_FINISHED_INIT:
1493 		err = tool->finished_init(session, event);
1494 		break;
1495 	case PERF_RECORD_BPF_METADATA:
1496 		err = tool->bpf_metadata(session, event);
1497 		break;
1498 	default:
1499 		err = -EINVAL;
1500 		break;
1501 	}
1502 	perf_sample__exit(&sample);
1503 	return err;
1504 }
1505 
1506 int perf_session__deliver_synth_event(struct perf_session *session,
1507 				      union perf_event *event,
1508 				      struct perf_sample *sample)
1509 {
1510 	struct evlist *evlist = session->evlist;
1511 	const struct perf_tool *tool = session->tool;
1512 
1513 	events_stats__inc(&evlist->stats, event->header.type);
1514 
1515 	if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1516 		return perf_session__process_user_event(session, event, 0, NULL);
1517 
1518 	return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0, NULL);
1519 }
1520 
1521 int perf_session__deliver_synth_attr_event(struct perf_session *session,
1522 					   const struct perf_event_attr *attr,
1523 					   u64 id)
1524 {
1525 	union {
1526 		struct {
1527 			struct perf_record_header_attr attr;
1528 			u64 ids[1];
1529 		} attr_id;
1530 		union perf_event ev;
1531 	} ev = {
1532 		.attr_id.attr.header.type = PERF_RECORD_HEADER_ATTR,
1533 		.attr_id.attr.header.size = sizeof(ev.attr_id),
1534 		.attr_id.ids[0] = id,
1535 	};
1536 
1537 	if (attr->size != sizeof(ev.attr_id.attr.attr)) {
1538 		pr_debug("Unexpected perf_event_attr size\n");
1539 		return -EINVAL;
1540 	}
1541 	ev.attr_id.attr.attr = *attr;
1542 	return perf_session__deliver_synth_event(session, &ev.ev, NULL);
1543 }
1544 
1545 static void event_swap(union perf_event *event, bool sample_id_all)
1546 {
1547 	perf_event__swap_op swap;
1548 
1549 	swap = perf_event__swap_ops[event->header.type];
1550 	if (swap)
1551 		swap(event, sample_id_all);
1552 }
1553 
1554 int perf_session__peek_event(struct perf_session *session, off_t file_offset,
1555 			     void *buf, size_t buf_sz,
1556 			     union perf_event **event_ptr,
1557 			     struct perf_sample *sample)
1558 {
1559 	union perf_event *event;
1560 	size_t hdr_sz, rest;
1561 	int fd;
1562 
1563 	if (session->one_mmap && !session->header.needs_swap) {
1564 		event = file_offset - session->one_mmap_offset +
1565 			session->one_mmap_addr;
1566 		goto out_parse_sample;
1567 	}
1568 
1569 	if (perf_data__is_pipe(session->data))
1570 		return -1;
1571 
1572 	fd = perf_data__fd(session->data);
1573 	hdr_sz = sizeof(struct perf_event_header);
1574 
1575 	if (buf_sz < hdr_sz)
1576 		return -1;
1577 
1578 	if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 ||
1579 	    readn(fd, buf, hdr_sz) != (ssize_t)hdr_sz)
1580 		return -1;
1581 
1582 	event = (union perf_event *)buf;
1583 
1584 	if (session->header.needs_swap)
1585 		perf_event_header__bswap(&event->header);
1586 
1587 	if (event->header.size < hdr_sz || event->header.size > buf_sz)
1588 		return -1;
1589 
1590 	buf += hdr_sz;
1591 	rest = event->header.size - hdr_sz;
1592 
1593 	if (readn(fd, buf, rest) != (ssize_t)rest)
1594 		return -1;
1595 
1596 	if (session->header.needs_swap)
1597 		event_swap(event, evlist__sample_id_all(session->evlist));
1598 
1599 out_parse_sample:
1600 
1601 	if (sample && event->header.type < PERF_RECORD_USER_TYPE_START &&
1602 	    evlist__parse_sample(session->evlist, event, sample))
1603 		return -1;
1604 
1605 	*event_ptr = event;
1606 
1607 	return 0;
1608 }
1609 
1610 int perf_session__peek_events(struct perf_session *session, u64 offset,
1611 			      u64 size, peek_events_cb_t cb, void *data)
1612 {
1613 	u64 max_offset = offset + size;
1614 	char buf[PERF_SAMPLE_MAX_SIZE];
1615 	union perf_event *event;
1616 	int err;
1617 
1618 	do {
1619 		err = perf_session__peek_event(session, offset, buf,
1620 					       PERF_SAMPLE_MAX_SIZE, &event,
1621 					       NULL);
1622 		if (err)
1623 			return err;
1624 
1625 		err = cb(session, event, offset, data);
1626 		if (err)
1627 			return err;
1628 
1629 		offset += event->header.size;
1630 		if (event->header.type == PERF_RECORD_AUXTRACE)
1631 			offset += event->auxtrace.size;
1632 
1633 	} while (offset < max_offset);
1634 
1635 	return err;
1636 }
1637 
1638 static s64 perf_session__process_event(struct perf_session *session,
1639 				       union perf_event *event, u64 file_offset,
1640 				       const char *file_path)
1641 {
1642 	struct evlist *evlist = session->evlist;
1643 	const struct perf_tool *tool = session->tool;
1644 	int ret;
1645 
1646 	if (session->header.needs_swap)
1647 		event_swap(event, evlist__sample_id_all(evlist));
1648 
1649 	if (event->header.type >= PERF_RECORD_HEADER_MAX) {
1650 		/* perf should not support unaligned event, stop here. */
1651 		if (event->header.size % sizeof(u64))
1652 			return -EINVAL;
1653 
1654 		/* This perf is outdated and does not support the latest event type. */
1655 		ui__warning("Unsupported header type %u, please consider updating perf.\n",
1656 			    event->header.type);
1657 		/* Skip unsupported event by returning its size. */
1658 		return event->header.size;
1659 	}
1660 
1661 	events_stats__inc(&evlist->stats, event->header.type);
1662 
1663 	if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1664 		return perf_session__process_user_event(session, event, file_offset, file_path);
1665 
1666 	if (tool->ordered_events) {
1667 		u64 timestamp = -1ULL;
1668 
1669 		ret = evlist__parse_sample_timestamp(evlist, event, &timestamp);
1670 		if (ret && ret != -1)
1671 			return ret;
1672 
1673 		ret = perf_session__queue_event(session, event, timestamp, file_offset, file_path);
1674 		if (ret != -ETIME)
1675 			return ret;
1676 	}
1677 
1678 	return perf_session__deliver_event(session, event, tool, file_offset, file_path);
1679 }
1680 
1681 void perf_event_header__bswap(struct perf_event_header *hdr)
1682 {
1683 	hdr->type = bswap_32(hdr->type);
1684 	hdr->misc = bswap_16(hdr->misc);
1685 	hdr->size = bswap_16(hdr->size);
1686 }
1687 
1688 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid)
1689 {
1690 	return machine__findnew_thread(&session->machines.host, -1, pid);
1691 }
1692 
1693 int perf_session__register_idle_thread(struct perf_session *session)
1694 {
1695 	struct thread *thread = machine__idle_thread(&session->machines.host);
1696 
1697 	/* machine__idle_thread() got the thread, so put it */
1698 	thread__put(thread);
1699 	return thread ? 0 : -1;
1700 }
1701 
1702 static void
1703 perf_session__warn_order(const struct perf_session *session)
1704 {
1705 	const struct ordered_events *oe = &session->ordered_events;
1706 	struct evsel *evsel;
1707 	bool should_warn = true;
1708 
1709 	evlist__for_each_entry(session->evlist, evsel) {
1710 		if (evsel->core.attr.write_backward)
1711 			should_warn = false;
1712 	}
1713 
1714 	if (!should_warn)
1715 		return;
1716 	if (oe->nr_unordered_events != 0)
1717 		ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events);
1718 }
1719 
1720 static void perf_session__warn_about_errors(const struct perf_session *session)
1721 {
1722 	const struct events_stats *stats = &session->evlist->stats;
1723 
1724 	if (session->tool->lost == perf_event__process_lost &&
1725 	    stats->nr_events[PERF_RECORD_LOST] != 0) {
1726 		ui__warning("Processed %d events and lost %d chunks!\n\n"
1727 			    "Check IO/CPU overload!\n\n",
1728 			    stats->nr_events[0],
1729 			    stats->nr_events[PERF_RECORD_LOST]);
1730 	}
1731 
1732 	if (session->tool->lost_samples == perf_event__process_lost_samples) {
1733 		double drop_rate;
1734 
1735 		drop_rate = (double)stats->total_lost_samples /
1736 			    (double) (stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples);
1737 		if (drop_rate > 0.05) {
1738 			ui__warning("Processed %" PRIu64 " samples and lost %3.2f%%!\n\n",
1739 				    stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples,
1740 				    drop_rate * 100.0);
1741 		}
1742 	}
1743 
1744 	if (session->tool->aux == perf_event__process_aux &&
1745 	    stats->total_aux_lost != 0) {
1746 		ui__warning("AUX data lost %" PRIu64 " times out of %u!\n\n",
1747 			    stats->total_aux_lost,
1748 			    stats->nr_events[PERF_RECORD_AUX]);
1749 	}
1750 
1751 	if (session->tool->aux == perf_event__process_aux &&
1752 	    stats->total_aux_partial != 0) {
1753 		bool vmm_exclusive = false;
1754 
1755 		(void)sysfs__read_bool("module/kvm_intel/parameters/vmm_exclusive",
1756 		                       &vmm_exclusive);
1757 
1758 		ui__warning("AUX data had gaps in it %" PRIu64 " times out of %u!\n\n"
1759 		            "Are you running a KVM guest in the background?%s\n\n",
1760 			    stats->total_aux_partial,
1761 			    stats->nr_events[PERF_RECORD_AUX],
1762 			    vmm_exclusive ?
1763 			    "\nReloading kvm_intel module with vmm_exclusive=0\n"
1764 			    "will reduce the gaps to only guest's timeslices." :
1765 			    "");
1766 	}
1767 
1768 	if (session->tool->aux == perf_event__process_aux &&
1769 	    stats->total_aux_collision != 0) {
1770 		ui__warning("AUX data detected collision  %" PRIu64 " times out of %u!\n\n",
1771 			    stats->total_aux_collision,
1772 			    stats->nr_events[PERF_RECORD_AUX]);
1773 	}
1774 
1775 	if (stats->nr_unknown_events != 0) {
1776 		ui__warning("Found %u unknown events!\n\n"
1777 			    "Is this an older tool processing a perf.data "
1778 			    "file generated by a more recent tool?\n\n"
1779 			    "If that is not the case, consider "
1780 			    "reporting to linux-kernel@vger.kernel.org.\n\n",
1781 			    stats->nr_unknown_events);
1782 	}
1783 
1784 	if (stats->nr_unknown_id != 0) {
1785 		ui__warning("%u samples with id not present in the header\n",
1786 			    stats->nr_unknown_id);
1787 	}
1788 
1789 	if (stats->nr_invalid_chains != 0) {
1790 		ui__warning("Found invalid callchains!\n\n"
1791 			    "%u out of %u events were discarded for this reason.\n\n"
1792 			    "Consider reporting to linux-kernel@vger.kernel.org.\n\n",
1793 			    stats->nr_invalid_chains,
1794 			    stats->nr_events[PERF_RECORD_SAMPLE]);
1795 	}
1796 
1797 	if (stats->nr_unprocessable_samples != 0) {
1798 		ui__warning("%u unprocessable samples recorded.\n"
1799 			    "Do you have a KVM guest running and not using 'perf kvm'?\n",
1800 			    stats->nr_unprocessable_samples);
1801 	}
1802 
1803 	perf_session__warn_order(session);
1804 
1805 	events_stats__auxtrace_error_warn(stats);
1806 
1807 	if (stats->nr_proc_map_timeout != 0) {
1808 		ui__warning("%d map information files for pre-existing threads were\n"
1809 			    "not processed, if there are samples for addresses they\n"
1810 			    "will not be resolved, you may find out which are these\n"
1811 			    "threads by running with -v and redirecting the output\n"
1812 			    "to a file.\n"
1813 			    "The time limit to process proc map is too short?\n"
1814 			    "Increase it by --proc-map-timeout\n",
1815 			    stats->nr_proc_map_timeout);
1816 	}
1817 }
1818 
1819 static int perf_session__flush_thread_stack(struct thread *thread,
1820 					    void *p __maybe_unused)
1821 {
1822 	return thread_stack__flush(thread);
1823 }
1824 
1825 static int perf_session__flush_thread_stacks(struct perf_session *session)
1826 {
1827 	return machines__for_each_thread(&session->machines,
1828 					 perf_session__flush_thread_stack,
1829 					 NULL);
1830 }
1831 
1832 volatile sig_atomic_t session_done;
1833 
1834 static int __perf_session__process_decomp_events(struct perf_session *session);
1835 
1836 static int __perf_session__process_pipe_events(struct perf_session *session)
1837 {
1838 	struct ordered_events *oe = &session->ordered_events;
1839 	const struct perf_tool *tool = session->tool;
1840 	struct ui_progress prog;
1841 	union perf_event *event;
1842 	uint32_t size, cur_size = 0;
1843 	void *buf = NULL;
1844 	s64 skip = 0;
1845 	u64 head;
1846 	ssize_t err;
1847 	void *p;
1848 	bool update_prog = false;
1849 
1850 	/*
1851 	 * If it's from a file saving pipe data (by redirection), it would have
1852 	 * a file name other than "-".  Then we can get the total size and show
1853 	 * the progress.
1854 	 */
1855 	if (strcmp(session->data->path, "-") && session->data->file.size) {
1856 		ui_progress__init_size(&prog, session->data->file.size,
1857 				       "Processing events...");
1858 		update_prog = true;
1859 	}
1860 
1861 	head = 0;
1862 	cur_size = sizeof(union perf_event);
1863 
1864 	buf = malloc(cur_size);
1865 	if (!buf)
1866 		return -errno;
1867 	ordered_events__set_copy_on_queue(oe, true);
1868 more:
1869 	event = buf;
1870 	err = perf_data__read(session->data, event,
1871 			      sizeof(struct perf_event_header));
1872 	if (err <= 0) {
1873 		if (err == 0)
1874 			goto done;
1875 
1876 		pr_err("failed to read event header\n");
1877 		goto out_err;
1878 	}
1879 
1880 	if (session->header.needs_swap)
1881 		perf_event_header__bswap(&event->header);
1882 
1883 	size = event->header.size;
1884 	if (size < sizeof(struct perf_event_header)) {
1885 		pr_err("bad event header size\n");
1886 		goto out_err;
1887 	}
1888 
1889 	if (size > cur_size) {
1890 		void *new = realloc(buf, size);
1891 		if (!new) {
1892 			pr_err("failed to allocate memory to read event\n");
1893 			goto out_err;
1894 		}
1895 		buf = new;
1896 		cur_size = size;
1897 		event = buf;
1898 	}
1899 	p = event;
1900 	p += sizeof(struct perf_event_header);
1901 
1902 	if (size - sizeof(struct perf_event_header)) {
1903 		err = perf_data__read(session->data, p,
1904 				      size - sizeof(struct perf_event_header));
1905 		if (err <= 0) {
1906 			if (err == 0) {
1907 				pr_err("unexpected end of event stream\n");
1908 				goto done;
1909 			}
1910 
1911 			pr_err("failed to read event data\n");
1912 			goto out_err;
1913 		}
1914 	}
1915 
1916 	if ((skip = perf_session__process_event(session, event, head, "pipe")) < 0) {
1917 		pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1918 		       head, event->header.size, event->header.type);
1919 		err = -EINVAL;
1920 		goto out_err;
1921 	}
1922 
1923 	head += size;
1924 
1925 	if (skip > 0)
1926 		head += skip;
1927 
1928 	err = __perf_session__process_decomp_events(session);
1929 	if (err)
1930 		goto out_err;
1931 
1932 	if (update_prog)
1933 		ui_progress__update(&prog, size);
1934 
1935 	if (!session_done())
1936 		goto more;
1937 done:
1938 	/* do the final flush for ordered samples */
1939 	err = ordered_events__flush(oe, OE_FLUSH__FINAL);
1940 	if (err)
1941 		goto out_err;
1942 	err = auxtrace__flush_events(session, tool);
1943 	if (err)
1944 		goto out_err;
1945 	err = perf_session__flush_thread_stacks(session);
1946 out_err:
1947 	free(buf);
1948 	if (update_prog)
1949 		ui_progress__finish();
1950 	if (!tool->no_warn)
1951 		perf_session__warn_about_errors(session);
1952 	ordered_events__free(&session->ordered_events);
1953 	auxtrace__free_events(session);
1954 	return err;
1955 }
1956 
1957 static union perf_event *
1958 prefetch_event(char *buf, u64 head, size_t mmap_size,
1959 	       bool needs_swap, union perf_event *error)
1960 {
1961 	union perf_event *event;
1962 	u16 event_size;
1963 
1964 	/*
1965 	 * Ensure we have enough space remaining to read
1966 	 * the size of the event in the headers.
1967 	 */
1968 	if (head + sizeof(event->header) > mmap_size)
1969 		return NULL;
1970 
1971 	event = (union perf_event *)(buf + head);
1972 	if (needs_swap)
1973 		perf_event_header__bswap(&event->header);
1974 
1975 	event_size = event->header.size;
1976 	if (head + event_size <= mmap_size)
1977 		return event;
1978 
1979 	/* We're not fetching the event so swap back again */
1980 	if (needs_swap)
1981 		perf_event_header__bswap(&event->header);
1982 
1983 	/* Check if the event fits into the next mmapped buf. */
1984 	if (event_size <= mmap_size - head % page_size) {
1985 		/* Remap buf and fetch again. */
1986 		return NULL;
1987 	}
1988 
1989 	/* Invalid input. Event size should never exceed mmap_size. */
1990 	pr_debug("%s: head=%#" PRIx64 " event->header.size=%#x, mmap_size=%#zx:"
1991 		 " fuzzed or compressed perf.data?\n", __func__, head, event_size, mmap_size);
1992 
1993 	return error;
1994 }
1995 
1996 static union perf_event *
1997 fetch_mmaped_event(u64 head, size_t mmap_size, char *buf, bool needs_swap)
1998 {
1999 	return prefetch_event(buf, head, mmap_size, needs_swap, ERR_PTR(-EINVAL));
2000 }
2001 
2002 static union perf_event *
2003 fetch_decomp_event(u64 head, size_t mmap_size, char *buf, bool needs_swap)
2004 {
2005 	return prefetch_event(buf, head, mmap_size, needs_swap, NULL);
2006 }
2007 
2008 static int __perf_session__process_decomp_events(struct perf_session *session)
2009 {
2010 	s64 skip;
2011 	u64 size;
2012 	struct decomp *decomp = session->active_decomp->decomp_last;
2013 
2014 	if (!decomp)
2015 		return 0;
2016 
2017 	while (decomp->head < decomp->size && !session_done()) {
2018 		union perf_event *event = fetch_decomp_event(decomp->head, decomp->size, decomp->data,
2019 							     session->header.needs_swap);
2020 
2021 		if (!event)
2022 			break;
2023 
2024 		size = event->header.size;
2025 
2026 		if (size < sizeof(struct perf_event_header) ||
2027 		    (skip = perf_session__process_event(session, event, decomp->file_pos,
2028 							decomp->file_path)) < 0) {
2029 			pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
2030 				decomp->file_pos + decomp->head, event->header.size, event->header.type);
2031 			return -EINVAL;
2032 		}
2033 
2034 		if (skip)
2035 			size += skip;
2036 
2037 		decomp->head += size;
2038 	}
2039 
2040 	return 0;
2041 }
2042 
2043 /*
2044  * On 64bit we can mmap the data file in one go. No need for tiny mmap
2045  * slices. On 32bit we use 32MB.
2046  */
2047 #if BITS_PER_LONG == 64
2048 #define MMAP_SIZE ULLONG_MAX
2049 #define NUM_MMAPS 1
2050 #else
2051 #define MMAP_SIZE (32 * 1024 * 1024ULL)
2052 #define NUM_MMAPS 128
2053 #endif
2054 
2055 struct reader;
2056 
2057 typedef s64 (*reader_cb_t)(struct perf_session *session,
2058 			   union perf_event *event,
2059 			   u64 file_offset,
2060 			   const char *file_path);
2061 
2062 struct reader {
2063 	int		 fd;
2064 	const char	 *path;
2065 	u64		 data_size;
2066 	u64		 data_offset;
2067 	reader_cb_t	 process;
2068 	bool		 in_place_update;
2069 	char		 *mmaps[NUM_MMAPS];
2070 	size_t		 mmap_size;
2071 	int		 mmap_idx;
2072 	char		 *mmap_cur;
2073 	u64		 file_pos;
2074 	u64		 file_offset;
2075 	u64		 head;
2076 	u64		 size;
2077 	bool		 done;
2078 	struct zstd_data   zstd_data;
2079 	struct decomp_data decomp_data;
2080 };
2081 
2082 static int
2083 reader__init(struct reader *rd, bool *one_mmap)
2084 {
2085 	u64 data_size = rd->data_size;
2086 	char **mmaps = rd->mmaps;
2087 
2088 	rd->head = rd->data_offset;
2089 	data_size += rd->data_offset;
2090 
2091 	rd->mmap_size = MMAP_SIZE;
2092 	if (rd->mmap_size > data_size) {
2093 		rd->mmap_size = data_size;
2094 		if (one_mmap)
2095 			*one_mmap = true;
2096 	}
2097 
2098 	memset(mmaps, 0, sizeof(rd->mmaps));
2099 
2100 	if (zstd_init(&rd->zstd_data, 0))
2101 		return -1;
2102 	rd->decomp_data.zstd_decomp = &rd->zstd_data;
2103 
2104 	return 0;
2105 }
2106 
2107 static void
2108 reader__release_decomp(struct reader *rd)
2109 {
2110 	perf_decomp__release_events(rd->decomp_data.decomp);
2111 	zstd_fini(&rd->zstd_data);
2112 }
2113 
2114 static int
2115 reader__mmap(struct reader *rd, struct perf_session *session)
2116 {
2117 	int mmap_prot, mmap_flags;
2118 	char *buf, **mmaps = rd->mmaps;
2119 	u64 page_offset;
2120 
2121 	mmap_prot  = PROT_READ;
2122 	mmap_flags = MAP_SHARED;
2123 
2124 	if (rd->in_place_update) {
2125 		mmap_prot  |= PROT_WRITE;
2126 	} else if (session->header.needs_swap) {
2127 		mmap_prot  |= PROT_WRITE;
2128 		mmap_flags = MAP_PRIVATE;
2129 	}
2130 
2131 	if (mmaps[rd->mmap_idx]) {
2132 		munmap(mmaps[rd->mmap_idx], rd->mmap_size);
2133 		mmaps[rd->mmap_idx] = NULL;
2134 	}
2135 
2136 	page_offset = page_size * (rd->head / page_size);
2137 	rd->file_offset += page_offset;
2138 	rd->head -= page_offset;
2139 
2140 	buf = mmap(NULL, rd->mmap_size, mmap_prot, mmap_flags, rd->fd,
2141 		   rd->file_offset);
2142 	if (buf == MAP_FAILED) {
2143 		pr_err("failed to mmap file\n");
2144 		return -errno;
2145 	}
2146 	mmaps[rd->mmap_idx] = rd->mmap_cur = buf;
2147 	rd->mmap_idx = (rd->mmap_idx + 1) & (ARRAY_SIZE(rd->mmaps) - 1);
2148 	rd->file_pos = rd->file_offset + rd->head;
2149 	if (session->one_mmap) {
2150 		session->one_mmap_addr = buf;
2151 		session->one_mmap_offset = rd->file_offset;
2152 	}
2153 
2154 	return 0;
2155 }
2156 
2157 enum {
2158 	READER_OK,
2159 	READER_NODATA,
2160 };
2161 
2162 static int
2163 reader__read_event(struct reader *rd, struct perf_session *session,
2164 		   struct ui_progress *prog)
2165 {
2166 	u64 size;
2167 	int err = READER_OK;
2168 	union perf_event *event;
2169 	s64 skip;
2170 
2171 	event = fetch_mmaped_event(rd->head, rd->mmap_size, rd->mmap_cur,
2172 				   session->header.needs_swap);
2173 	if (IS_ERR(event))
2174 		return PTR_ERR(event);
2175 
2176 	if (!event)
2177 		return READER_NODATA;
2178 
2179 	size = event->header.size;
2180 
2181 	skip = -EINVAL;
2182 
2183 	if (size < sizeof(struct perf_event_header) ||
2184 	    (skip = rd->process(session, event, rd->file_pos, rd->path)) < 0) {
2185 		pr_err("%#" PRIx64 " [%#x]: failed to process type: %d [%s]\n",
2186 		       rd->file_offset + rd->head, event->header.size,
2187 		       event->header.type, strerror(-skip));
2188 		err = skip;
2189 		goto out;
2190 	}
2191 
2192 	if (skip)
2193 		size += skip;
2194 
2195 	rd->size += size;
2196 	rd->head += size;
2197 	rd->file_pos += size;
2198 
2199 	err = __perf_session__process_decomp_events(session);
2200 	if (err)
2201 		goto out;
2202 
2203 	ui_progress__update(prog, size);
2204 
2205 out:
2206 	return err;
2207 }
2208 
2209 static inline bool
2210 reader__eof(struct reader *rd)
2211 {
2212 	return (rd->file_pos >= rd->data_size + rd->data_offset);
2213 }
2214 
2215 static int
2216 reader__process_events(struct reader *rd, struct perf_session *session,
2217 		       struct ui_progress *prog)
2218 {
2219 	int err;
2220 
2221 	err = reader__init(rd, &session->one_mmap);
2222 	if (err)
2223 		goto out;
2224 
2225 	session->active_decomp = &rd->decomp_data;
2226 
2227 remap:
2228 	err = reader__mmap(rd, session);
2229 	if (err)
2230 		goto out;
2231 
2232 more:
2233 	err = reader__read_event(rd, session, prog);
2234 	if (err < 0)
2235 		goto out;
2236 	else if (err == READER_NODATA)
2237 		goto remap;
2238 
2239 	if (session_done())
2240 		goto out;
2241 
2242 	if (!reader__eof(rd))
2243 		goto more;
2244 
2245 out:
2246 	session->active_decomp = &session->decomp_data;
2247 	return err;
2248 }
2249 
2250 static s64 process_simple(struct perf_session *session,
2251 			  union perf_event *event,
2252 			  u64 file_offset,
2253 			  const char *file_path)
2254 {
2255 	return perf_session__process_event(session, event, file_offset, file_path);
2256 }
2257 
2258 static int __perf_session__process_events(struct perf_session *session)
2259 {
2260 	struct reader rd = {
2261 		.fd		= perf_data__fd(session->data),
2262 		.path		= session->data->file.path,
2263 		.data_size	= session->header.data_size,
2264 		.data_offset	= session->header.data_offset,
2265 		.process	= process_simple,
2266 		.in_place_update = session->data->in_place_update,
2267 	};
2268 	struct ordered_events *oe = &session->ordered_events;
2269 	const struct perf_tool *tool = session->tool;
2270 	struct ui_progress prog;
2271 	int err;
2272 
2273 	if (rd.data_size == 0)
2274 		return -1;
2275 
2276 	ui_progress__init_size(&prog, rd.data_size, "Processing events...");
2277 
2278 	err = reader__process_events(&rd, session, &prog);
2279 	if (err)
2280 		goto out_err;
2281 	/* do the final flush for ordered samples */
2282 	err = ordered_events__flush(oe, OE_FLUSH__FINAL);
2283 	if (err)
2284 		goto out_err;
2285 	err = auxtrace__flush_events(session, tool);
2286 	if (err)
2287 		goto out_err;
2288 	err = perf_session__flush_thread_stacks(session);
2289 out_err:
2290 	ui_progress__finish();
2291 	if (!tool->no_warn)
2292 		perf_session__warn_about_errors(session);
2293 	/*
2294 	 * We may switching perf.data output, make ordered_events
2295 	 * reusable.
2296 	 */
2297 	ordered_events__reinit(&session->ordered_events);
2298 	auxtrace__free_events(session);
2299 	reader__release_decomp(&rd);
2300 	session->one_mmap = false;
2301 	return err;
2302 }
2303 
2304 /*
2305  * Processing 2 MB of data from each reader in sequence,
2306  * because that's the way the ordered events sorting works
2307  * most efficiently.
2308  */
2309 #define READER_MAX_SIZE (2 * 1024 * 1024)
2310 
2311 /*
2312  * This function reads, merge and process directory data.
2313  * It assumens the version 1 of directory data, where each
2314  * data file holds per-cpu data, already sorted by kernel.
2315  */
2316 static int __perf_session__process_dir_events(struct perf_session *session)
2317 {
2318 	struct perf_data *data = session->data;
2319 	const struct perf_tool *tool = session->tool;
2320 	int i, ret, readers, nr_readers;
2321 	struct ui_progress prog;
2322 	u64 total_size = perf_data__size(session->data);
2323 	struct reader *rd;
2324 
2325 	ui_progress__init_size(&prog, total_size, "Processing events...");
2326 
2327 	nr_readers = 1;
2328 	for (i = 0; i < data->dir.nr; i++) {
2329 		if (data->dir.files[i].size)
2330 			nr_readers++;
2331 	}
2332 
2333 	rd = zalloc(nr_readers * sizeof(struct reader));
2334 	if (!rd)
2335 		return -ENOMEM;
2336 
2337 	rd[0] = (struct reader) {
2338 		.fd		 = perf_data__fd(session->data),
2339 		.path		 = session->data->file.path,
2340 		.data_size	 = session->header.data_size,
2341 		.data_offset	 = session->header.data_offset,
2342 		.process	 = process_simple,
2343 		.in_place_update = session->data->in_place_update,
2344 	};
2345 	ret = reader__init(&rd[0], NULL);
2346 	if (ret)
2347 		goto out_err;
2348 	ret = reader__mmap(&rd[0], session);
2349 	if (ret)
2350 		goto out_err;
2351 	readers = 1;
2352 
2353 	for (i = 0; i < data->dir.nr; i++) {
2354 		if (!data->dir.files[i].size)
2355 			continue;
2356 		rd[readers] = (struct reader) {
2357 			.fd		 = data->dir.files[i].fd,
2358 			.path		 = data->dir.files[i].path,
2359 			.data_size	 = data->dir.files[i].size,
2360 			.data_offset	 = 0,
2361 			.process	 = process_simple,
2362 			.in_place_update = session->data->in_place_update,
2363 		};
2364 		ret = reader__init(&rd[readers], NULL);
2365 		if (ret)
2366 			goto out_err;
2367 		ret = reader__mmap(&rd[readers], session);
2368 		if (ret)
2369 			goto out_err;
2370 		readers++;
2371 	}
2372 
2373 	i = 0;
2374 	while (readers) {
2375 		if (session_done())
2376 			break;
2377 
2378 		if (rd[i].done) {
2379 			i = (i + 1) % nr_readers;
2380 			continue;
2381 		}
2382 		if (reader__eof(&rd[i])) {
2383 			rd[i].done = true;
2384 			readers--;
2385 			continue;
2386 		}
2387 
2388 		session->active_decomp = &rd[i].decomp_data;
2389 		ret = reader__read_event(&rd[i], session, &prog);
2390 		if (ret < 0) {
2391 			goto out_err;
2392 		} else if (ret == READER_NODATA) {
2393 			ret = reader__mmap(&rd[i], session);
2394 			if (ret)
2395 				goto out_err;
2396 		}
2397 
2398 		if (rd[i].size >= READER_MAX_SIZE) {
2399 			rd[i].size = 0;
2400 			i = (i + 1) % nr_readers;
2401 		}
2402 	}
2403 
2404 	ret = ordered_events__flush(&session->ordered_events, OE_FLUSH__FINAL);
2405 	if (ret)
2406 		goto out_err;
2407 
2408 	ret = perf_session__flush_thread_stacks(session);
2409 out_err:
2410 	ui_progress__finish();
2411 
2412 	if (!tool->no_warn)
2413 		perf_session__warn_about_errors(session);
2414 
2415 	/*
2416 	 * We may switching perf.data output, make ordered_events
2417 	 * reusable.
2418 	 */
2419 	ordered_events__reinit(&session->ordered_events);
2420 
2421 	session->one_mmap = false;
2422 
2423 	session->active_decomp = &session->decomp_data;
2424 	for (i = 0; i < nr_readers; i++)
2425 		reader__release_decomp(&rd[i]);
2426 	zfree(&rd);
2427 
2428 	return ret;
2429 }
2430 
2431 int perf_session__process_events(struct perf_session *session)
2432 {
2433 	if (perf_session__register_idle_thread(session) < 0)
2434 		return -ENOMEM;
2435 
2436 	if (perf_data__is_pipe(session->data))
2437 		return __perf_session__process_pipe_events(session);
2438 
2439 	if (perf_data__is_dir(session->data) && session->data->dir.nr)
2440 		return __perf_session__process_dir_events(session);
2441 
2442 	return __perf_session__process_events(session);
2443 }
2444 
2445 bool perf_session__has_traces(struct perf_session *session, const char *msg)
2446 {
2447 	struct evsel *evsel;
2448 
2449 	evlist__for_each_entry(session->evlist, evsel) {
2450 		if (evsel->core.attr.type == PERF_TYPE_TRACEPOINT)
2451 			return true;
2452 	}
2453 
2454 	pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg);
2455 	return false;
2456 }
2457 
2458 bool perf_session__has_switch_events(struct perf_session *session)
2459 {
2460 	struct evsel *evsel;
2461 
2462 	evlist__for_each_entry(session->evlist, evsel) {
2463 		if (evsel->core.attr.context_switch)
2464 			return true;
2465 	}
2466 
2467 	return false;
2468 }
2469 
2470 int map__set_kallsyms_ref_reloc_sym(struct map *map, const char *symbol_name, u64 addr)
2471 {
2472 	char *bracket;
2473 	struct ref_reloc_sym *ref;
2474 	struct kmap *kmap;
2475 
2476 	ref = zalloc(sizeof(struct ref_reloc_sym));
2477 	if (ref == NULL)
2478 		return -ENOMEM;
2479 
2480 	ref->name = strdup(symbol_name);
2481 	if (ref->name == NULL) {
2482 		free(ref);
2483 		return -ENOMEM;
2484 	}
2485 
2486 	bracket = strchr(ref->name, ']');
2487 	if (bracket)
2488 		*bracket = '\0';
2489 
2490 	ref->addr = addr;
2491 
2492 	kmap = map__kmap(map);
2493 	if (kmap)
2494 		kmap->ref_reloc_sym = ref;
2495 
2496 	return 0;
2497 }
2498 
2499 size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp)
2500 {
2501 	return machines__fprintf_dsos(&session->machines, fp);
2502 }
2503 
2504 size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp,
2505 					  bool (skip)(struct dso *dso, int parm), int parm)
2506 {
2507 	return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm);
2508 }
2509 
2510 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp)
2511 {
2512 	size_t ret;
2513 	const char *msg = "";
2514 
2515 	if (perf_header__has_feat(&session->header, HEADER_AUXTRACE))
2516 		msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)";
2517 
2518 	ret = fprintf(fp, "\nAggregated stats:%s\n", msg);
2519 
2520 	ret += events_stats__fprintf(&session->evlist->stats, fp);
2521 	return ret;
2522 }
2523 
2524 size_t perf_session__fprintf(struct perf_session *session, FILE *fp)
2525 {
2526 	/*
2527 	 * FIXME: Here we have to actually print all the machines in this
2528 	 * session, not just the host...
2529 	 */
2530 	return machine__fprintf(&session->machines.host, fp);
2531 }
2532 
2533 void perf_session__dump_kmaps(struct perf_session *session)
2534 {
2535 	int save_verbose = verbose;
2536 
2537 	fflush(stdout);
2538 	fprintf(stderr, "Kernel and module maps:\n");
2539 	verbose = 0; /* Suppress verbose to print a summary only */
2540 	maps__fprintf(machine__kernel_maps(&session->machines.host), stderr);
2541 	verbose = save_verbose;
2542 }
2543 
2544 struct evsel *perf_session__find_first_evtype(struct perf_session *session,
2545 					      unsigned int type)
2546 {
2547 	struct evsel *pos;
2548 
2549 	evlist__for_each_entry(session->evlist, pos) {
2550 		if (pos->core.attr.type == type)
2551 			return pos;
2552 	}
2553 	return NULL;
2554 }
2555 
2556 int perf_session__cpu_bitmap(struct perf_session *session,
2557 			     const char *cpu_list, unsigned long *cpu_bitmap)
2558 {
2559 	int i, err = -1;
2560 	struct perf_cpu_map *map;
2561 	int nr_cpus = min(session->header.env.nr_cpus_avail, MAX_NR_CPUS);
2562 	struct perf_cpu cpu;
2563 
2564 	for (i = 0; i < PERF_TYPE_MAX; ++i) {
2565 		struct evsel *evsel;
2566 
2567 		evsel = perf_session__find_first_evtype(session, i);
2568 		if (!evsel)
2569 			continue;
2570 
2571 		if (!(evsel->core.attr.sample_type & PERF_SAMPLE_CPU)) {
2572 			pr_err("File does not contain CPU events. "
2573 			       "Remove -C option to proceed.\n");
2574 			return -1;
2575 		}
2576 	}
2577 
2578 	map = perf_cpu_map__new(cpu_list);
2579 	if (map == NULL) {
2580 		pr_err("Invalid cpu_list\n");
2581 		return -1;
2582 	}
2583 
2584 	perf_cpu_map__for_each_cpu(cpu, i, map) {
2585 		if (cpu.cpu >= nr_cpus) {
2586 			pr_err("Requested CPU %d too large. "
2587 			       "Consider raising MAX_NR_CPUS\n", cpu.cpu);
2588 			goto out_delete_map;
2589 		}
2590 
2591 		__set_bit(cpu.cpu, cpu_bitmap);
2592 	}
2593 
2594 	err = 0;
2595 
2596 out_delete_map:
2597 	perf_cpu_map__put(map);
2598 	return err;
2599 }
2600 
2601 void perf_session__fprintf_info(struct perf_session *session, FILE *fp,
2602 				bool full)
2603 {
2604 	if (session == NULL || fp == NULL)
2605 		return;
2606 
2607 	fprintf(fp, "# ========\n");
2608 	perf_header__fprintf_info(session, fp, full);
2609 	fprintf(fp, "# ========\n#\n");
2610 }
2611 
2612 static int perf_session__register_guest(struct perf_session *session, pid_t machine_pid)
2613 {
2614 	struct machine *machine = machines__findnew(&session->machines, machine_pid);
2615 	struct thread *thread;
2616 
2617 	if (!machine)
2618 		return -ENOMEM;
2619 
2620 	machine->single_address_space = session->machines.host.single_address_space;
2621 
2622 	thread = machine__idle_thread(machine);
2623 	if (!thread)
2624 		return -ENOMEM;
2625 	thread__put(thread);
2626 
2627 	machine->kallsyms_filename = perf_data__guest_kallsyms_name(session->data, machine_pid);
2628 
2629 	return 0;
2630 }
2631 
2632 static int perf_session__set_guest_cpu(struct perf_session *session, pid_t pid,
2633 				       pid_t tid, int guest_cpu)
2634 {
2635 	struct machine *machine = &session->machines.host;
2636 	struct thread *thread = machine__findnew_thread(machine, pid, tid);
2637 
2638 	if (!thread)
2639 		return -ENOMEM;
2640 	thread__set_guest_cpu(thread, guest_cpu);
2641 	thread__put(thread);
2642 
2643 	return 0;
2644 }
2645 
2646 int perf_event__process_id_index(struct perf_session *session,
2647 				 union perf_event *event)
2648 {
2649 	struct evlist *evlist = session->evlist;
2650 	struct perf_record_id_index *ie = &event->id_index;
2651 	size_t sz = ie->header.size - sizeof(*ie);
2652 	size_t i, nr, max_nr;
2653 	size_t e1_sz = sizeof(struct id_index_entry);
2654 	size_t e2_sz = sizeof(struct id_index_entry_2);
2655 	size_t etot_sz = e1_sz + e2_sz;
2656 	struct id_index_entry_2 *e2;
2657 	pid_t last_pid = 0;
2658 
2659 	max_nr = sz / e1_sz;
2660 	nr = ie->nr;
2661 	if (nr > max_nr) {
2662 		printf("Too big: nr %zu max_nr %zu\n", nr, max_nr);
2663 		return -EINVAL;
2664 	}
2665 
2666 	if (sz >= nr * etot_sz) {
2667 		max_nr = sz / etot_sz;
2668 		if (nr > max_nr) {
2669 			printf("Too big2: nr %zu max_nr %zu\n", nr, max_nr);
2670 			return -EINVAL;
2671 		}
2672 		e2 = (void *)ie + sizeof(*ie) + nr * e1_sz;
2673 	} else {
2674 		e2 = NULL;
2675 	}
2676 
2677 	if (dump_trace)
2678 		fprintf(stdout, " nr: %zu\n", nr);
2679 
2680 	for (i = 0; i < nr; i++, (e2 ? e2++ : 0)) {
2681 		struct id_index_entry *e = &ie->entries[i];
2682 		struct perf_sample_id *sid;
2683 		int ret;
2684 
2685 		if (dump_trace) {
2686 			fprintf(stdout,	" ... id: %"PRI_lu64, e->id);
2687 			fprintf(stdout,	"  idx: %"PRI_lu64, e->idx);
2688 			fprintf(stdout,	"  cpu: %"PRI_ld64, e->cpu);
2689 			fprintf(stdout, "  tid: %"PRI_ld64, e->tid);
2690 			if (e2) {
2691 				fprintf(stdout, "  machine_pid: %"PRI_ld64, e2->machine_pid);
2692 				fprintf(stdout, "  vcpu: %"PRI_lu64"\n", e2->vcpu);
2693 			} else {
2694 				fprintf(stdout, "\n");
2695 			}
2696 		}
2697 
2698 		sid = evlist__id2sid(evlist, e->id);
2699 		if (!sid)
2700 			return -ENOENT;
2701 
2702 		sid->idx = e->idx;
2703 		sid->cpu.cpu = e->cpu;
2704 		sid->tid = e->tid;
2705 
2706 		if (!e2)
2707 			continue;
2708 
2709 		sid->machine_pid = e2->machine_pid;
2710 		sid->vcpu.cpu = e2->vcpu;
2711 
2712 		if (!sid->machine_pid)
2713 			continue;
2714 
2715 		if (sid->machine_pid != last_pid) {
2716 			ret = perf_session__register_guest(session, sid->machine_pid);
2717 			if (ret)
2718 				return ret;
2719 			last_pid = sid->machine_pid;
2720 			perf_guest = true;
2721 		}
2722 
2723 		ret = perf_session__set_guest_cpu(session, sid->machine_pid, e->tid, e2->vcpu);
2724 		if (ret)
2725 			return ret;
2726 	}
2727 	return 0;
2728 }
2729 
2730 int perf_session__dsos_hit_all(struct perf_session *session)
2731 {
2732 	struct rb_node *nd;
2733 	int err;
2734 
2735 	err = machine__hit_all_dsos(&session->machines.host);
2736 	if (err)
2737 		return err;
2738 
2739 	for (nd = rb_first_cached(&session->machines.guests); nd;
2740 	     nd = rb_next(nd)) {
2741 		struct machine *pos = rb_entry(nd, struct machine, rb_node);
2742 
2743 		err = machine__hit_all_dsos(pos);
2744 		if (err)
2745 			return err;
2746 	}
2747 
2748 	return 0;
2749 }
2750