Linux Audio

Check our new training course

Loading...
v3.1
   1#define _FILE_OFFSET_BITS 64
   2
 
 
   3#include <linux/kernel.h>
 
 
   4
   5#include <byteswap.h>
   6#include <unistd.h>
   7#include <sys/types.h>
   8#include <sys/mman.h>
 
   9
 
 
 
  10#include "evlist.h"
  11#include "evsel.h"
 
 
 
  12#include "session.h"
  13#include "sort.h"
  14#include "util.h"
  15#include "cpumap.h"
  16
  17static int perf_session__open(struct perf_session *self, bool force)
  18{
  19	struct stat input_stat;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  20
  21	if (!strcmp(self->filename, "-")) {
  22		self->fd_pipe = true;
  23		self->fd = STDIN_FILENO;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
  24
  25		if (perf_session__read_header(self, self->fd) < 0)
  26			pr_err("incompatible file format");
  27
  28		return 0;
 
 
 
 
 
  29	}
  30
  31	self->fd = open(self->filename, O_RDONLY);
  32	if (self->fd < 0) {
  33		int err = errno;
  34
  35		pr_err("failed to open %s: %s", self->filename, strerror(err));
  36		if (err == ENOENT && !strcmp(self->filename, "perf.data"))
  37			pr_err("  (try 'perf record' first)");
  38		pr_err("\n");
  39		return -errno;
  40	}
  41
  42	if (fstat(self->fd, &input_stat) < 0)
  43		goto out_close;
 
 
 
  44
  45	if (!force && input_stat.st_uid && (input_stat.st_uid != geteuid())) {
  46		pr_err("file %s not owned by current user or root\n",
  47		       self->filename);
  48		goto out_close;
  49	}
  50
  51	if (!input_stat.st_size) {
  52		pr_info("zero-sized file (%s), nothing to do!\n",
  53			self->filename);
  54		goto out_close;
 
 
 
  55	}
  56
  57	if (perf_session__read_header(self, self->fd) < 0) {
  58		pr_err("incompatible file format");
  59		goto out_close;
 
 
 
 
 
 
  60	}
  61
  62	if (!perf_evlist__valid_sample_type(self->evlist)) {
  63		pr_err("non matching sample_type");
  64		goto out_close;
  65	}
  66
  67	if (!perf_evlist__valid_sample_id_all(self->evlist)) {
  68		pr_err("non matching sample_id_all");
  69		goto out_close;
  70	}
  71
  72	self->size = input_stat.st_size;
  73	return 0;
  74
  75out_close:
  76	close(self->fd);
  77	self->fd = -1;
  78	return -1;
  79}
  80
  81static void perf_session__id_header_size(struct perf_session *session)
  82{
  83       struct perf_sample *data;
  84       u64 sample_type = session->sample_type;
  85       u16 size = 0;
  86
  87	if (!session->sample_id_all)
  88		goto out;
  89
  90       if (sample_type & PERF_SAMPLE_TID)
  91               size += sizeof(data->tid) * 2;
  92
  93       if (sample_type & PERF_SAMPLE_TIME)
  94               size += sizeof(data->time);
  95
  96       if (sample_type & PERF_SAMPLE_ID)
  97               size += sizeof(data->id);
 
  98
  99       if (sample_type & PERF_SAMPLE_STREAM_ID)
 100               size += sizeof(data->stream_id);
 
 
 101
 102       if (sample_type & PERF_SAMPLE_CPU)
 103               size += sizeof(data->cpu) * 2;
 104out:
 105       session->id_hdr_size = size;
 106}
 107
 108void perf_session__update_sample_type(struct perf_session *self)
 109{
 110	self->sample_type = perf_evlist__sample_type(self->evlist);
 111	self->sample_size = __perf_evsel__sample_size(self->sample_type);
 112	self->sample_id_all = perf_evlist__sample_id_all(self->evlist);
 113	perf_session__id_header_size(self);
 
 
 
 
 114}
 115
 116int perf_session__create_kernel_maps(struct perf_session *self)
 117{
 118	int ret = machine__create_kernel_maps(&self->host_machine);
 119
 120	if (ret >= 0)
 121		ret = machines__create_guest_kernel_maps(&self->machines);
 122	return ret;
 123}
 124
 125static void perf_session__destroy_kernel_maps(struct perf_session *self)
 
 126{
 127	machine__destroy_kernel_maps(&self->host_machine);
 128	machines__destroy_guest_kernel_maps(&self->machines);
 
 
 
 129}
 130
 131struct perf_session *perf_session__new(const char *filename, int mode,
 132				       bool force, bool repipe,
 133				       struct perf_event_ops *ops)
 134{
 135	size_t len = filename ? strlen(filename) + 1 : 0;
 136	struct perf_session *self = zalloc(sizeof(*self) + len);
 137
 138	if (self == NULL)
 139		goto out;
 140
 141	memcpy(self->filename, filename, len);
 142	self->threads = RB_ROOT;
 143	INIT_LIST_HEAD(&self->dead_threads);
 144	self->last_match = NULL;
 145	/*
 146	 * On 64bit we can mmap the data file in one go. No need for tiny mmap
 147	 * slices. On 32bit we use 32MB.
 148	 */
 149#if BITS_PER_LONG == 64
 150	self->mmap_window = ULLONG_MAX;
 151#else
 152	self->mmap_window = 32 * 1024 * 1024ULL;
 153#endif
 154	self->machines = RB_ROOT;
 155	self->repipe = repipe;
 156	INIT_LIST_HEAD(&self->ordered_samples.samples);
 157	INIT_LIST_HEAD(&self->ordered_samples.sample_cache);
 158	INIT_LIST_HEAD(&self->ordered_samples.to_free);
 159	machine__init(&self->host_machine, "", HOST_KERNEL_ID);
 160
 161	if (mode == O_RDONLY) {
 162		if (perf_session__open(self, force) < 0)
 163			goto out_delete;
 164		perf_session__update_sample_type(self);
 165	} else if (mode == O_WRONLY) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 166		/*
 167		 * In O_RDONLY mode this will be performed when reading the
 168		 * kernel MMAP event, in perf_event__process_mmap().
 169		 */
 170		if (perf_session__create_kernel_maps(self) < 0)
 171			goto out_delete;
 172	}
 173
 174	if (ops && ops->ordering_requires_timestamps &&
 175	    ops->ordered_samples && !self->sample_id_all) {
 
 
 
 
 176		dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n");
 177		ops->ordered_samples = false;
 178	}
 179
 180out:
 181	return self;
 182out_delete:
 183	perf_session__delete(self);
 184	return NULL;
 
 185}
 186
 187static void perf_session__delete_dead_threads(struct perf_session *self)
 188{
 189	struct thread *n, *t;
 190
 191	list_for_each_entry_safe(t, n, &self->dead_threads, node) {
 192		list_del(&t->node);
 193		thread__delete(t);
 194	}
 195}
 196
 197static void perf_session__delete_threads(struct perf_session *self)
 198{
 199	struct rb_node *nd = rb_first(&self->threads);
 200
 201	while (nd) {
 202		struct thread *t = rb_entry(nd, struct thread, rb_node);
 
 
 
 
 
 
 
 
 203
 204		rb_erase(&t->rb_node, &self->threads);
 205		nd = rb_next(nd);
 206		thread__delete(t);
 207	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 208}
 209
 210void perf_session__delete(struct perf_session *self)
 
 
 
 211{
 212	perf_session__destroy_kernel_maps(self);
 213	perf_session__delete_dead_threads(self);
 214	perf_session__delete_threads(self);
 215	machine__exit(&self->host_machine);
 216	close(self->fd);
 217	free(self);
 218}
 219
 220void perf_session__remove_thread(struct perf_session *self, struct thread *th)
 
 
 
 221{
 222	self->last_match = NULL;
 223	rb_erase(&th->rb_node, &self->threads);
 224	/*
 225	 * We may have references to this thread, for instance in some hist_entry
 226	 * instances, so just move them to a separate list.
 227	 */
 228	list_add_tail(&th->node, &self->dead_threads);
 229}
 230
 231static bool symbol__match_parent_regex(struct symbol *sym)
 
 
 
 
 232{
 233	if (sym->name && !regexec(&parent_regex, sym->name, 0, NULL, 0))
 234		return 1;
 
 235
 
 
 
 
 
 
 236	return 0;
 237}
 238
 239int perf_session__resolve_callchain(struct perf_session *self,
 240				    struct thread *thread,
 241				    struct ip_callchain *chain,
 242				    struct symbol **parent)
 243{
 244	u8 cpumode = PERF_RECORD_MISC_USER;
 245	unsigned int i;
 246	int err;
 247
 248	callchain_cursor_reset(&self->callchain_cursor);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 249
 250	for (i = 0; i < chain->nr; i++) {
 251		u64 ip;
 252		struct addr_location al;
 253
 254		if (callchain_param.order == ORDER_CALLEE)
 255			ip = chain->ips[i];
 256		else
 257			ip = chain->ips[chain->nr - i - 1];
 
 
 
 
 258
 259		if (ip >= PERF_CONTEXT_MAX) {
 260			switch (ip) {
 261			case PERF_CONTEXT_HV:
 262				cpumode = PERF_RECORD_MISC_HYPERVISOR;	break;
 263			case PERF_CONTEXT_KERNEL:
 264				cpumode = PERF_RECORD_MISC_KERNEL;	break;
 265			case PERF_CONTEXT_USER:
 266				cpumode = PERF_RECORD_MISC_USER;	break;
 267			default:
 268				break;
 269			}
 270			continue;
 271		}
 272
 273		al.filtered = false;
 274		thread__find_addr_location(thread, self, cpumode,
 275				MAP__FUNCTION, thread->pid, ip, &al, NULL);
 276		if (al.sym != NULL) {
 277			if (sort__has_parent && !*parent &&
 278			    symbol__match_parent_regex(al.sym))
 279				*parent = al.sym;
 280			if (!symbol_conf.use_callchain)
 281				break;
 282		}
 283
 284		err = callchain_cursor_append(&self->callchain_cursor,
 285					      ip, al.map, al.sym);
 286		if (err)
 287			return err;
 288	}
 
 289
 
 290	return 0;
 291}
 292
 293static int process_event_synth_stub(union perf_event *event __used,
 294				    struct perf_session *session __used)
 
 295{
 
 
 
 296	dump_printf(": unhandled!\n");
 297	return 0;
 298}
 299
 300static int process_event_sample_stub(union perf_event *event __used,
 301				     struct perf_sample *sample __used,
 302				     struct perf_evsel *evsel __used,
 303				     struct perf_session *session __used)
 304{
 
 
 
 305	dump_printf(": unhandled!\n");
 306	return 0;
 307}
 308
 309static int process_event_stub(union perf_event *event __used,
 310			      struct perf_sample *sample __used,
 311			      struct perf_session *session __used)
 312{
 
 
 
 313	dump_printf(": unhandled!\n");
 314	return 0;
 315}
 316
 317static int process_finished_round_stub(union perf_event *event __used,
 318				       struct perf_session *session __used,
 319				       struct perf_event_ops *ops __used)
 320{
 
 
 
 321	dump_printf(": unhandled!\n");
 322	return 0;
 323}
 324
 325static int process_finished_round(union perf_event *event,
 326				  struct perf_session *session,
 327				  struct perf_event_ops *ops);
 328
 329static void perf_event_ops__fill_defaults(struct perf_event_ops *handler)
 330{
 331	if (handler->sample == NULL)
 332		handler->sample = process_event_sample_stub;
 333	if (handler->mmap == NULL)
 334		handler->mmap = process_event_stub;
 335	if (handler->comm == NULL)
 336		handler->comm = process_event_stub;
 337	if (handler->fork == NULL)
 338		handler->fork = process_event_stub;
 339	if (handler->exit == NULL)
 340		handler->exit = process_event_stub;
 341	if (handler->lost == NULL)
 342		handler->lost = perf_event__process_lost;
 343	if (handler->read == NULL)
 344		handler->read = process_event_stub;
 345	if (handler->throttle == NULL)
 346		handler->throttle = process_event_stub;
 347	if (handler->unthrottle == NULL)
 348		handler->unthrottle = process_event_stub;
 349	if (handler->attr == NULL)
 350		handler->attr = process_event_synth_stub;
 351	if (handler->event_type == NULL)
 352		handler->event_type = process_event_synth_stub;
 353	if (handler->tracing_data == NULL)
 354		handler->tracing_data = process_event_synth_stub;
 355	if (handler->build_id == NULL)
 356		handler->build_id = process_event_synth_stub;
 357	if (handler->finished_round == NULL) {
 358		if (handler->ordered_samples)
 359			handler->finished_round = process_finished_round;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 360		else
 361			handler->finished_round = process_finished_round_stub;
 362	}
 363}
 364
 365void mem_bswap_64(void *src, int byte_size)
 366{
 367	u64 *m = src;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 368
 369	while (byte_size > 0) {
 370		*m = bswap_64(*m);
 371		byte_size -= sizeof(u64);
 372		++m;
 373	}
 374}
 375
 376static void perf_event__all64_swap(union perf_event *event)
 
 377{
 378	struct perf_event_header *hdr = &event->header;
 379	mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr));
 380}
 381
 382static void perf_event__comm_swap(union perf_event *event)
 383{
 384	event->comm.pid = bswap_32(event->comm.pid);
 385	event->comm.tid = bswap_32(event->comm.tid);
 
 
 
 
 
 
 
 386}
 387
 388static void perf_event__mmap_swap(union perf_event *event)
 
 389{
 390	event->mmap.pid	  = bswap_32(event->mmap.pid);
 391	event->mmap.tid	  = bswap_32(event->mmap.tid);
 392	event->mmap.start = bswap_64(event->mmap.start);
 393	event->mmap.len	  = bswap_64(event->mmap.len);
 394	event->mmap.pgoff = bswap_64(event->mmap.pgoff);
 
 
 
 
 
 
 
 395}
 396
 397static void perf_event__task_swap(union perf_event *event)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 398{
 399	event->fork.pid	 = bswap_32(event->fork.pid);
 400	event->fork.tid	 = bswap_32(event->fork.tid);
 401	event->fork.ppid = bswap_32(event->fork.ppid);
 402	event->fork.ptid = bswap_32(event->fork.ptid);
 403	event->fork.time = bswap_64(event->fork.time);
 
 
 
 404}
 405
 406static void perf_event__read_swap(union perf_event *event)
 407{
 408	event->read.pid		 = bswap_32(event->read.pid);
 409	event->read.tid		 = bswap_32(event->read.tid);
 410	event->read.value	 = bswap_64(event->read.value);
 411	event->read.time_enabled = bswap_64(event->read.time_enabled);
 412	event->read.time_running = bswap_64(event->read.time_running);
 413	event->read.id		 = bswap_64(event->read.id);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 414}
 415
 416/* exported for swapping attributes in file header */
 417void perf_event__attr_swap(struct perf_event_attr *attr)
 418{
 419	attr->type		= bswap_32(attr->type);
 420	attr->size		= bswap_32(attr->size);
 421	attr->config		= bswap_64(attr->config);
 422	attr->sample_period	= bswap_64(attr->sample_period);
 423	attr->sample_type	= bswap_64(attr->sample_type);
 424	attr->read_format	= bswap_64(attr->read_format);
 425	attr->wakeup_events	= bswap_32(attr->wakeup_events);
 426	attr->bp_type		= bswap_32(attr->bp_type);
 427	attr->bp_addr		= bswap_64(attr->bp_addr);
 428	attr->bp_len		= bswap_64(attr->bp_len);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 429}
 430
 431static void perf_event__hdr_attr_swap(union perf_event *event)
 
 432{
 433	size_t size;
 434
 435	perf_event__attr_swap(&event->attr.attr);
 436
 437	size = event->header.size;
 438	size -= (void *)&event->attr.id - (void *)event;
 439	mem_bswap_64(event->attr.id, size);
 440}
 441
 442static void perf_event__event_type_swap(union perf_event *event)
 
 
 
 
 
 
 
 
 443{
 444	event->event_type.event_type.event_id =
 445		bswap_64(event->event_type.event_type.event_id);
 446}
 447
 448static void perf_event__tracing_data_swap(union perf_event *event)
 
 449{
 450	event->tracing_data.size = bswap_32(event->tracing_data.size);
 451}
 452
 453typedef void (*perf_event__swap_op)(union perf_event *event);
 
 
 
 454
 455static perf_event__swap_op perf_event__swap_ops[] = {
 456	[PERF_RECORD_MMAP]		  = perf_event__mmap_swap,
 457	[PERF_RECORD_COMM]		  = perf_event__comm_swap,
 458	[PERF_RECORD_FORK]		  = perf_event__task_swap,
 459	[PERF_RECORD_EXIT]		  = perf_event__task_swap,
 460	[PERF_RECORD_LOST]		  = perf_event__all64_swap,
 461	[PERF_RECORD_READ]		  = perf_event__read_swap,
 462	[PERF_RECORD_SAMPLE]		  = perf_event__all64_swap,
 463	[PERF_RECORD_HEADER_ATTR]	  = perf_event__hdr_attr_swap,
 464	[PERF_RECORD_HEADER_EVENT_TYPE]	  = perf_event__event_type_swap,
 465	[PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
 466	[PERF_RECORD_HEADER_BUILD_ID]	  = NULL,
 467	[PERF_RECORD_HEADER_MAX]	  = NULL,
 468};
 469
 470struct sample_queue {
 471	u64			timestamp;
 472	u64			file_offset;
 473	union perf_event	*event;
 474	struct list_head	list;
 475};
 
 
 
 
 
 
 
 
 
 476
 477static void perf_session_free_sample_buffers(struct perf_session *session)
 
 478{
 479	struct ordered_samples *os = &session->ordered_samples;
 
 
 
 
 
 
 
 
 
 480
 481	while (!list_empty(&os->to_free)) {
 482		struct sample_queue *sq;
 
 
 483
 484		sq = list_entry(os->to_free.next, struct sample_queue, list);
 485		list_del(&sq->list);
 486		free(sq);
 487	}
 488}
 489
 490static int perf_session_deliver_event(struct perf_session *session,
 491				      union perf_event *event,
 492				      struct perf_sample *sample,
 493				      struct perf_event_ops *ops,
 494				      u64 file_offset);
 495
 496static void flush_sample_queue(struct perf_session *s,
 497			       struct perf_event_ops *ops)
 498{
 499	struct ordered_samples *os = &s->ordered_samples;
 500	struct list_head *head = &os->samples;
 501	struct sample_queue *tmp, *iter;
 502	struct perf_sample sample;
 503	u64 limit = os->next_flush;
 504	u64 last_ts = os->last_sample ? os->last_sample->timestamp : 0ULL;
 505	int ret;
 506
 507	if (!ops->ordered_samples || !limit)
 508		return;
 509
 510	list_for_each_entry_safe(iter, tmp, head, list) {
 511		if (iter->timestamp > limit)
 512			break;
 513
 514		ret = perf_session__parse_sample(s, iter->event, &sample);
 515		if (ret)
 516			pr_err("Can't parse sample, err = %d\n", ret);
 517		else
 518			perf_session_deliver_event(s, iter->event, &sample, ops,
 519						   iter->file_offset);
 520
 521		os->last_flush = iter->timestamp;
 522		list_del(&iter->list);
 523		list_add(&iter->list, &os->sample_cache);
 524	}
 
 525
 526	if (list_empty(head)) {
 527		os->last_sample = NULL;
 528	} else if (last_ts <= limit) {
 529		os->last_sample =
 530			list_entry(head->prev, struct sample_queue, list);
 
 
 
 
 
 
 531	}
 532}
 533
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 534/*
 535 * When perf record finishes a pass on every buffers, it records this pseudo
 536 * event.
 537 * We record the max timestamp t found in the pass n.
 538 * Assuming these timestamps are monotonic across cpus, we know that if
 539 * a buffer still has events with timestamps below t, they will be all
 540 * available and then read in the pass n + 1.
 541 * Hence when we start to read the pass n + 2, we can safely flush every
 542 * events with timestamps below t.
 543 *
 544 *    ============ PASS n =================
 545 *       CPU 0         |   CPU 1
 546 *                     |
 547 *    cnt1 timestamps  |   cnt2 timestamps
 548 *          1          |         2
 549 *          2          |         3
 550 *          -          |         4  <--- max recorded
 551 *
 552 *    ============ PASS n + 1 ==============
 553 *       CPU 0         |   CPU 1
 554 *                     |
 555 *    cnt1 timestamps  |   cnt2 timestamps
 556 *          3          |         5
 557 *          4          |         6
 558 *          5          |         7 <---- max recorded
 559 *
 560 *      Flush every events below timestamp 4
 561 *
 562 *    ============ PASS n + 2 ==============
 563 *       CPU 0         |   CPU 1
 564 *                     |
 565 *    cnt1 timestamps  |   cnt2 timestamps
 566 *          6          |         8
 567 *          7          |         9
 568 *          -          |         10
 569 *
 570 *      Flush every events below timestamp 7
 571 *      etc...
 572 */
 573static int process_finished_round(union perf_event *event __used,
 574				  struct perf_session *session,
 575				  struct perf_event_ops *ops)
 576{
 577	flush_sample_queue(session, ops);
 578	session->ordered_samples.next_flush = session->ordered_samples.max_timestamp;
 
 
 579
 580	return 0;
 
 
 
 581}
 582
 583/* The queue is ordered by time */
 584static void __queue_event(struct sample_queue *new, struct perf_session *s)
 585{
 586	struct ordered_samples *os = &s->ordered_samples;
 587	struct sample_queue *sample = os->last_sample;
 588	u64 timestamp = new->timestamp;
 589	struct list_head *p;
 590
 591	os->last_sample = new;
 592
 593	if (!sample) {
 594		list_add(&new->list, &os->samples);
 595		os->max_timestamp = timestamp;
 596		return;
 597	}
 598
 599	/*
 600	 * last_sample might point to some random place in the list as it's
 601	 * the last queued event. We expect that the new event is close to
 602	 * this.
 603	 */
 604	if (sample->timestamp <= timestamp) {
 605		while (sample->timestamp <= timestamp) {
 606			p = sample->list.next;
 607			if (p == &os->samples) {
 608				list_add_tail(&new->list, &os->samples);
 609				os->max_timestamp = timestamp;
 610				return;
 611			}
 612			sample = list_entry(p, struct sample_queue, list);
 613		}
 614		list_add_tail(&new->list, &sample->list);
 615	} else {
 616		while (sample->timestamp > timestamp) {
 617			p = sample->list.prev;
 618			if (p == &os->samples) {
 619				list_add(&new->list, &os->samples);
 620				return;
 621			}
 622			sample = list_entry(p, struct sample_queue, list);
 623		}
 624		list_add(&new->list, &sample->list);
 
 
 
 
 
 
 625	}
 626}
 627
 628#define MAX_SAMPLE_BUFFER	(64 * 1024 / sizeof(struct sample_queue))
 629
 630static int perf_session_queue_event(struct perf_session *s, union perf_event *event,
 631				    struct perf_sample *sample, u64 file_offset)
 632{
 633	struct ordered_samples *os = &s->ordered_samples;
 634	struct list_head *sc = &os->sample_cache;
 635	u64 timestamp = sample->time;
 636	struct sample_queue *new;
 637
 638	if (!timestamp || timestamp == ~0ULL)
 639		return -ETIME;
 640
 641	if (timestamp < s->ordered_samples.last_flush) {
 642		printf("Warning: Timestamp below last timeslice flush\n");
 643		return -EINVAL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 644	}
 
 645
 646	if (!list_empty(sc)) {
 647		new = list_entry(sc->next, struct sample_queue, list);
 648		list_del(&new->list);
 649	} else if (os->sample_buffer) {
 650		new = os->sample_buffer + os->sample_buffer_idx;
 651		if (++os->sample_buffer_idx == MAX_SAMPLE_BUFFER)
 652			os->sample_buffer = NULL;
 653	} else {
 654		os->sample_buffer = malloc(MAX_SAMPLE_BUFFER * sizeof(*new));
 655		if (!os->sample_buffer)
 656			return -ENOMEM;
 657		list_add(&os->sample_buffer->list, &os->to_free);
 658		os->sample_buffer_idx = 2;
 659		new = os->sample_buffer + 1;
 660	}
 
 661
 662	new->timestamp = timestamp;
 663	new->file_offset = file_offset;
 664	new->event = event;
 
 
 665
 666	__queue_event(new, s);
 
 
 
 667
 668	return 0;
 669}
 670
 671static void callchain__printf(struct perf_sample *sample)
 672{
 673	unsigned int i;
 674
 675	printf("... chain: nr:%" PRIu64 "\n", sample->callchain->nr);
 
 
 
 676
 677	for (i = 0; i < sample->callchain->nr; i++)
 678		printf("..... %2d: %016" PRIx64 "\n",
 679		       i, sample->callchain->ips[i]);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 680}
 681
 682static void perf_session__print_tstamp(struct perf_session *session,
 
 
 
 
 
 
 683				       union perf_event *event,
 684				       struct perf_sample *sample)
 685{
 
 
 686	if (event->header.type != PERF_RECORD_SAMPLE &&
 687	    !session->sample_id_all) {
 688		fputs("-1 -1 ", stdout);
 689		return;
 690	}
 691
 692	if ((session->sample_type & PERF_SAMPLE_CPU))
 693		printf("%u ", sample->cpu);
 694
 695	if (session->sample_type & PERF_SAMPLE_TIME)
 696		printf("%" PRIu64 " ", sample->time);
 697}
 698
 699static void dump_event(struct perf_session *session, union perf_event *event,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 700		       u64 file_offset, struct perf_sample *sample)
 701{
 702	if (!dump_trace)
 703		return;
 704
 705	printf("\n%#" PRIx64 " [%#x]: event: %d\n",
 706	       file_offset, event->header.size, event->header.type);
 707
 708	trace_event(event);
 
 
 709
 710	if (sample)
 711		perf_session__print_tstamp(session, event, sample);
 712
 713	printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset,
 714	       event->header.size, perf_event__name(event->header.type));
 715}
 716
 717static void dump_sample(struct perf_session *session, union perf_event *event,
 718			struct perf_sample *sample)
 719{
 
 
 720	if (!dump_trace)
 721		return;
 722
 723	printf("(IP, %d): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n",
 724	       event->header.misc, sample->pid, sample->tid, sample->ip,
 725	       sample->period, sample->addr);
 726
 727	if (session->sample_type & PERF_SAMPLE_CALLCHAIN)
 728		callchain__printf(sample);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 729}
 730
 731static int perf_session_deliver_event(struct perf_session *session,
 732				      union perf_event *event,
 733				      struct perf_sample *sample,
 734				      struct perf_event_ops *ops,
 735				      u64 file_offset)
 736{
 737	struct perf_evsel *evsel;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 738
 739	dump_event(session, event, file_offset, sample);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 740
 741	switch (event->header.type) {
 742	case PERF_RECORD_SAMPLE:
 743		dump_sample(session, event, sample);
 744		evsel = perf_evlist__id2evsel(session->evlist, sample->id);
 745		if (evsel == NULL) {
 746			++session->hists.stats.nr_unknown_id;
 747			return -1;
 
 
 
 
 
 748		}
 749		return ops->sample(event, sample, evsel, session);
 750	case PERF_RECORD_MMAP:
 751		return ops->mmap(event, sample, session);
 
 
 
 
 752	case PERF_RECORD_COMM:
 753		return ops->comm(event, sample, session);
 
 
 
 
 754	case PERF_RECORD_FORK:
 755		return ops->fork(event, sample, session);
 756	case PERF_RECORD_EXIT:
 757		return ops->exit(event, sample, session);
 758	case PERF_RECORD_LOST:
 759		return ops->lost(event, sample, session);
 
 
 
 
 
 
 760	case PERF_RECORD_READ:
 761		return ops->read(event, sample, session);
 
 762	case PERF_RECORD_THROTTLE:
 763		return ops->throttle(event, sample, session);
 764	case PERF_RECORD_UNTHROTTLE:
 765		return ops->unthrottle(event, sample, session);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 766	default:
 767		++session->hists.stats.nr_unknown_events;
 768		return -1;
 769	}
 770}
 771
 772static int perf_session__preprocess_sample(struct perf_session *session,
 773					   union perf_event *event, struct perf_sample *sample)
 
 
 774{
 775	if (event->header.type != PERF_RECORD_SAMPLE ||
 776	    !(session->sample_type & PERF_SAMPLE_CALLCHAIN))
 777		return 0;
 778
 779	if (!ip_callchain__valid(sample->callchain, event)) {
 780		pr_debug("call-chain problem with event, skipping it.\n");
 781		++session->hists.stats.nr_invalid_chains;
 782		session->hists.stats.total_invalid_chains += sample->period;
 783		return -EINVAL;
 784	}
 785	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 786}
 787
 788static int perf_session__process_user_event(struct perf_session *session, union perf_event *event,
 789					    struct perf_event_ops *ops, u64 file_offset)
 790{
 791	dump_event(session, event, file_offset, NULL);
 
 
 
 
 
 
 
 
 
 792
 793	/* These events are processed right away */
 794	switch (event->header.type) {
 795	case PERF_RECORD_HEADER_ATTR:
 796		return ops->attr(event, session);
 
 
 
 
 
 
 
 797	case PERF_RECORD_HEADER_EVENT_TYPE:
 798		return ops->event_type(event, session);
 
 
 
 
 799	case PERF_RECORD_HEADER_TRACING_DATA:
 800		/* setup for reading amidst mmap */
 801		lseek(session->fd, file_offset, SEEK_SET);
 802		return ops->tracing_data(event, session);
 
 
 
 
 
 803	case PERF_RECORD_HEADER_BUILD_ID:
 804		return ops->build_id(event, session);
 805	case PERF_RECORD_FINISHED_ROUND:
 806		return ops->finished_round(event, session, ops);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 807	default:
 808		return -EINVAL;
 809	}
 810}
 811
 812static int perf_session__process_event(struct perf_session *session,
 813				       union perf_event *event,
 814				       struct perf_event_ops *ops,
 815				       u64 file_offset)
 816{
 817	struct perf_sample sample;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 818	int ret;
 819
 820	if (session->header.needs_swap &&
 821	    perf_event__swap_ops[event->header.type])
 822		perf_event__swap_ops[event->header.type](event);
 823
 824	if (event->header.type >= PERF_RECORD_HEADER_MAX)
 825		return -EINVAL;
 826
 827	hists__inc_nr_events(&session->hists, event->header.type);
 828
 829	if (event->header.type >= PERF_RECORD_USER_TYPE_START)
 830		return perf_session__process_user_event(session, event, ops, file_offset);
 831
 832	/*
 833	 * For all kernel events we get the sample data
 834	 */
 835	ret = perf_session__parse_sample(session, event, &sample);
 836	if (ret)
 837		return ret;
 838
 839	/* Preprocess sample records - precheck callchains */
 840	if (perf_session__preprocess_sample(session, event, &sample))
 841		return 0;
 842
 843	if (ops->ordered_samples) {
 844		ret = perf_session_queue_event(session, event, &sample,
 845					       file_offset);
 846		if (ret != -ETIME)
 847			return ret;
 848	}
 849
 850	return perf_session_deliver_event(session, event, &sample, ops,
 851					  file_offset);
 
 
 
 
 
 
 852}
 853
 854void perf_event_header__bswap(struct perf_event_header *self)
 855{
 856	self->type = bswap_32(self->type);
 857	self->misc = bswap_16(self->misc);
 858	self->size = bswap_16(self->size);
 859}
 860
 861static struct thread *perf_session__register_idle_thread(struct perf_session *self)
 
 
 
 
 
 
 
 862{
 863	struct thread *thread = perf_session__findnew(self, 0);
 
 
 
 
 
 
 
 864
 865	if (thread == NULL || thread__set_comm(thread, "swapper")) {
 866		pr_err("problem inserting idle task.\n");
 867		thread = NULL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 868	}
 869
 870	return thread;
 
 
 
 871}
 872
 873static void perf_session__warn_about_errors(const struct perf_session *session,
 874					    const struct perf_event_ops *ops)
 875{
 876	if (ops->lost == perf_event__process_lost &&
 877	    session->hists.stats.total_lost != 0) {
 878		ui__warning("Processed %" PRIu64 " events and LOST %" PRIu64
 879			    "!\n\nCheck IO/CPU overload!\n\n",
 880			    session->hists.stats.total_period,
 881			    session->hists.stats.total_lost);
 
 
 882	}
 883
 884	if (session->hists.stats.nr_unknown_events != 0) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 885		ui__warning("Found %u unknown events!\n\n"
 886			    "Is this an older tool processing a perf.data "
 887			    "file generated by a more recent tool?\n\n"
 888			    "If that is not the case, consider "
 889			    "reporting to linux-kernel@vger.kernel.org.\n\n",
 890			    session->hists.stats.nr_unknown_events);
 891	}
 892
 893	if (session->hists.stats.nr_unknown_id != 0) {
 894		ui__warning("%u samples with id not present in the header\n",
 895			    session->hists.stats.nr_unknown_id);
 896	}
 897
 898 	if (session->hists.stats.nr_invalid_chains != 0) {
 899 		ui__warning("Found invalid callchains!\n\n"
 900 			    "%u out of %u events were discarded for this reason.\n\n"
 901 			    "Consider reporting to linux-kernel@vger.kernel.org.\n\n",
 902 			    session->hists.stats.nr_invalid_chains,
 903 			    session->hists.stats.nr_events[PERF_RECORD_SAMPLE]);
 904 	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 905}
 906
 907#define session_done()	(*(volatile int *)(&session_done))
 908volatile int session_done;
 909
 910static int __perf_session__process_pipe_events(struct perf_session *self,
 911					       struct perf_event_ops *ops)
 
 912{
 913	union perf_event event;
 914	uint32_t size;
 915	int skip = 0;
 
 
 
 
 916	u64 head;
 917	int err;
 918	void *p;
 919
 920	perf_event_ops__fill_defaults(ops);
 921
 922	head = 0;
 
 
 
 
 
 
 923more:
 924	err = readn(self->fd, &event, sizeof(struct perf_event_header));
 
 925	if (err <= 0) {
 926		if (err == 0)
 927			goto done;
 928
 929		pr_err("failed to read event header\n");
 930		goto out_err;
 931	}
 932
 933	if (self->header.needs_swap)
 934		perf_event_header__bswap(&event.header);
 935
 936	size = event.header.size;
 937	if (size == 0)
 938		size = 8;
 
 
 939
 940	p = &event;
 
 
 
 
 
 
 
 
 
 
 941	p += sizeof(struct perf_event_header);
 942
 943	if (size - sizeof(struct perf_event_header)) {
 944		err = readn(self->fd, p, size - sizeof(struct perf_event_header));
 945		if (err <= 0) {
 946			if (err == 0) {
 947				pr_err("unexpected end of event stream\n");
 948				goto done;
 949			}
 950
 951			pr_err("failed to read event data\n");
 952			goto out_err;
 953		}
 954	}
 955
 956	if (size == 0 ||
 957	    (skip = perf_session__process_event(self, &event, ops, head)) < 0) {
 958		dump_printf("%#" PRIx64 " [%#x]: skipping unknown header type: %d\n",
 959			    head, event.header.size, event.header.type);
 960		/*
 961		 * assume we lost track of the stream, check alignment, and
 962		 * increment a single u64 in the hope to catch on again 'soon'.
 963		 */
 964		if (unlikely(head & 7))
 965			head &= ~7ULL;
 966
 967		size = 8;
 968	}
 969
 970	head += size;
 971
 972	if (skip > 0)
 973		head += skip;
 974
 
 
 
 
 975	if (!session_done())
 976		goto more;
 977done:
 978	err = 0;
 
 
 
 
 
 
 
 979out_err:
 980	perf_session__warn_about_errors(self, ops);
 981	perf_session_free_sample_buffers(self);
 
 
 
 982	return err;
 983}
 984
 985static union perf_event *
 986fetch_mmaped_event(struct perf_session *session,
 987		   u64 head, size_t mmap_size, char *buf)
 988{
 989	union perf_event *event;
 990
 991	/*
 992	 * Ensure we have enough space remaining to read
 993	 * the size of the event in the headers.
 994	 */
 995	if (head + sizeof(event->header) > mmap_size)
 996		return NULL;
 997
 998	event = (union perf_event *)(buf + head);
 
 
 999
1000	if (session->header.needs_swap)
 
 
 
 
1001		perf_event_header__bswap(&event->header);
1002
1003	if (head + event->header.size > mmap_size)
1004		return NULL;
1005
1006	return event;
1007}
1008
1009int __perf_session__process_events(struct perf_session *session,
1010				   u64 data_offset, u64 data_size,
1011				   u64 file_size, struct perf_event_ops *ops)
1012{
1013	u64 head, page_offset, file_offset, file_pos, progress_next;
1014	int err, mmap_prot, mmap_flags, map_idx = 0;
1015	struct ui_progress *progress;
1016	size_t	page_size, mmap_size;
1017	char *buf, *mmaps[8];
1018	union perf_event *event;
1019	uint32_t size;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1020
1021	perf_event_ops__fill_defaults(ops);
1022
1023	page_size = sysconf(_SC_PAGESIZE);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1024
1025	page_offset = page_size * (data_offset / page_size);
1026	file_offset = page_offset;
1027	head = data_offset - page_offset;
1028
1029	if (data_offset + data_size < file_size)
1030		file_size = data_offset + data_size;
1031
1032	progress_next = file_size / 16;
1033	progress = ui_progress__new("Processing events...", file_size);
1034	if (progress == NULL)
1035		return -1;
1036
1037	mmap_size = session->mmap_window;
1038	if (mmap_size > file_size)
1039		mmap_size = file_size;
 
 
1040
1041	memset(mmaps, 0, sizeof(mmaps));
1042
1043	mmap_prot  = PROT_READ;
1044	mmap_flags = MAP_SHARED;
1045
1046	if (session->header.needs_swap) {
1047		mmap_prot  |= PROT_WRITE;
1048		mmap_flags = MAP_PRIVATE;
1049	}
1050remap:
1051	buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, session->fd,
1052		   file_offset);
1053	if (buf == MAP_FAILED) {
1054		pr_err("failed to mmap file\n");
1055		err = -errno;
1056		goto out_err;
1057	}
1058	mmaps[map_idx] = buf;
1059	map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1);
1060	file_pos = file_offset + head;
 
 
 
 
1061
1062more:
1063	event = fetch_mmaped_event(session, head, mmap_size, buf);
 
 
 
1064	if (!event) {
1065		if (mmaps[map_idx]) {
1066			munmap(mmaps[map_idx], mmap_size);
1067			mmaps[map_idx] = NULL;
1068		}
1069
1070		page_offset = page_size * (head / page_size);
1071		file_offset += page_offset;
1072		head -= page_offset;
1073		goto remap;
1074	}
1075
1076	size = event->header.size;
1077
1078	if (size == 0 ||
1079	    perf_session__process_event(session, event, ops, file_pos) < 0) {
1080		dump_printf("%#" PRIx64 " [%#x]: skipping unknown header type: %d\n",
1081			    file_offset + head, event->header.size,
1082			    event->header.type);
1083		/*
1084		 * assume we lost track of the stream, check alignment, and
1085		 * increment a single u64 in the hope to catch on again 'soon'.
1086		 */
1087		if (unlikely(head & 7))
1088			head &= ~7ULL;
1089
1090		size = 8;
 
 
 
 
 
 
1091	}
1092
 
 
 
1093	head += size;
1094	file_pos += size;
1095
1096	if (file_pos >= progress_next) {
1097		progress_next += file_size / 16;
1098		ui_progress__update(progress, file_pos);
1099	}
1100
1101	if (file_pos < file_size)
 
 
 
 
 
1102		goto more;
1103
1104	err = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1105	/* do the final flush for ordered samples */
1106	session->ordered_samples.next_flush = ULLONG_MAX;
1107	flush_sample_queue(session, ops);
 
 
 
 
 
1108out_err:
1109	ui_progress__delete(progress);
1110	perf_session__warn_about_errors(session, ops);
1111	perf_session_free_sample_buffers(session);
 
 
 
 
 
 
 
1112	return err;
1113}
1114
1115int perf_session__process_events(struct perf_session *self,
1116				 struct perf_event_ops *ops)
1117{
1118	int err;
1119
1120	if (perf_session__register_idle_thread(self) == NULL)
1121		return -ENOMEM;
1122
1123	if (!self->fd_pipe)
1124		err = __perf_session__process_events(self,
1125						     self->header.data_offset,
1126						     self->header.data_size,
1127						     self->size, ops);
1128	else
1129		err = __perf_session__process_pipe_events(self, ops);
1130
1131	return err;
1132}
1133
1134bool perf_session__has_traces(struct perf_session *self, const char *msg)
1135{
1136	if (!(self->sample_type & PERF_SAMPLE_RAW)) {
1137		pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg);
1138		return false;
 
 
1139	}
1140
1141	return true;
 
1142}
1143
1144int perf_session__set_kallsyms_ref_reloc_sym(struct map **maps,
1145					     const char *symbol_name,
1146					     u64 addr)
1147{
1148	char *bracket;
1149	enum map_type i;
1150	struct ref_reloc_sym *ref;
 
1151
1152	ref = zalloc(sizeof(struct ref_reloc_sym));
1153	if (ref == NULL)
1154		return -ENOMEM;
1155
1156	ref->name = strdup(symbol_name);
1157	if (ref->name == NULL) {
1158		free(ref);
1159		return -ENOMEM;
1160	}
1161
1162	bracket = strchr(ref->name, ']');
1163	if (bracket)
1164		*bracket = '\0';
1165
1166	ref->addr = addr;
1167
1168	for (i = 0; i < MAP__NR_TYPES; ++i) {
1169		struct kmap *kmap = map__kmap(maps[i]);
1170		kmap->ref_reloc_sym = ref;
1171	}
1172
1173	return 0;
1174}
1175
1176size_t perf_session__fprintf_dsos(struct perf_session *self, FILE *fp)
1177{
1178	return __dsos__fprintf(&self->host_machine.kernel_dsos, fp) +
1179	       __dsos__fprintf(&self->host_machine.user_dsos, fp) +
1180	       machines__fprintf_dsos(&self->machines, fp);
1181}
1182
1183size_t perf_session__fprintf_dsos_buildid(struct perf_session *self, FILE *fp,
1184					  bool with_hits)
1185{
1186	size_t ret = machine__fprintf_dsos_buildid(&self->host_machine, fp, with_hits);
1187	return ret + machines__fprintf_dsos_buildid(&self->machines, fp, with_hits);
1188}
1189
1190size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp)
1191{
1192	struct perf_evsel *pos;
1193	size_t ret = fprintf(fp, "Aggregated stats:\n");
1194
1195	ret += hists__fprintf_nr_events(&session->hists, fp);
 
1196
1197	list_for_each_entry(pos, &session->evlist->entries, node) {
1198		ret += fprintf(fp, "%s stats:\n", event_name(pos));
1199		ret += hists__fprintf_nr_events(&pos->hists, fp);
1200	}
1201
 
1202	return ret;
1203}
1204
1205struct perf_evsel *perf_session__find_first_evtype(struct perf_session *session,
 
 
 
 
 
 
 
 
 
1206					      unsigned int type)
1207{
1208	struct perf_evsel *pos;
1209
1210	list_for_each_entry(pos, &session->evlist->entries, node) {
1211		if (pos->attr.type == type)
1212			return pos;
1213	}
1214	return NULL;
1215}
1216
1217void perf_session__print_ip(union perf_event *event,
1218			    struct perf_sample *sample,
1219			    struct perf_session *session,
1220			    int print_sym, int print_dso)
1221{
1222	struct addr_location al;
1223	const char *symname, *dsoname;
1224	struct callchain_cursor *cursor = &session->callchain_cursor;
1225	struct callchain_cursor_node *node;
1226
1227	if (perf_event__preprocess_sample(event, session, &al, sample,
1228					  NULL) < 0) {
1229		error("problem processing %d event, skipping it.\n",
1230			event->header.type);
1231		return;
1232	}
1233
1234	if (symbol_conf.use_callchain && sample->callchain) {
1235
1236		if (perf_session__resolve_callchain(session, al.thread,
1237						sample->callchain, NULL) != 0) {
1238			if (verbose)
1239				error("Failed to resolve callchain. Skipping\n");
1240			return;
1241		}
1242		callchain_cursor_commit(cursor);
1243
1244		while (1) {
1245			node = callchain_cursor_current(cursor);
1246			if (!node)
1247				break;
1248
1249			printf("\t%16" PRIx64, node->ip);
1250			if (print_sym) {
1251				if (node->sym && node->sym->name)
1252					symname = node->sym->name;
1253				else
1254					symname = "";
1255
1256				printf(" %s", symname);
1257			}
1258			if (print_dso) {
1259				if (node->map && node->map->dso && node->map->dso->name)
1260					dsoname = node->map->dso->name;
1261				else
1262					dsoname = "";
1263
1264				printf(" (%s)", dsoname);
1265			}
1266			printf("\n");
1267
1268			callchain_cursor_advance(cursor);
1269		}
1270
1271	} else {
1272		printf("%16" PRIx64, sample->ip);
1273		if (print_sym) {
1274			if (al.sym && al.sym->name)
1275				symname = al.sym->name;
1276			else
1277				symname = "";
1278
1279			printf(" %s", symname);
1280		}
1281
1282		if (print_dso) {
1283			if (al.map && al.map->dso && al.map->dso->name)
1284				dsoname = al.map->dso->name;
1285			else
1286				dsoname = "";
1287
1288			printf(" (%s)", dsoname);
1289		}
1290	}
1291}
1292
1293int perf_session__cpu_bitmap(struct perf_session *session,
1294			     const char *cpu_list, unsigned long *cpu_bitmap)
1295{
1296	int i;
1297	struct cpu_map *map;
 
1298
1299	for (i = 0; i < PERF_TYPE_MAX; ++i) {
1300		struct perf_evsel *evsel;
1301
1302		evsel = perf_session__find_first_evtype(session, i);
1303		if (!evsel)
1304			continue;
1305
1306		if (!(evsel->attr.sample_type & PERF_SAMPLE_CPU)) {
1307			pr_err("File does not contain CPU events. "
1308			       "Remove -c option to proceed.\n");
1309			return -1;
1310		}
1311	}
1312
1313	map = cpu_map__new(cpu_list);
 
 
 
 
1314
1315	for (i = 0; i < map->nr; i++) {
1316		int cpu = map->map[i];
1317
1318		if (cpu >= MAX_NR_CPUS) {
1319			pr_err("Requested CPU %d too large. "
1320			       "Consider raising MAX_NR_CPUS\n", cpu);
1321			return -1;
1322		}
1323
1324		set_bit(cpu, cpu_bitmap);
1325	}
1326
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1327	return 0;
1328}
v5.9
   1// SPDX-License-Identifier: GPL-2.0
   2#include <errno.h>
   3#include <inttypes.h>
   4#include <linux/err.h>
   5#include <linux/kernel.h>
   6#include <linux/zalloc.h>
   7#include <api/fs/fs.h>
   8
   9#include <byteswap.h>
  10#include <unistd.h>
  11#include <sys/types.h>
  12#include <sys/mman.h>
  13#include <perf/cpumap.h>
  14
  15#include "map_symbol.h"
  16#include "branch.h"
  17#include "debug.h"
  18#include "evlist.h"
  19#include "evsel.h"
  20#include "memswap.h"
  21#include "map.h"
  22#include "symbol.h"
  23#include "session.h"
  24#include "tool.h"
  25#include "perf_regs.h"
  26#include "asm/bug.h"
  27#include "auxtrace.h"
  28#include "thread.h"
  29#include "thread-stack.h"
  30#include "sample-raw.h"
  31#include "stat.h"
  32#include "ui/progress.h"
  33#include "../perf.h"
  34#include "arch/common.h"
  35#include <internal/lib.h>
  36
  37#ifdef HAVE_ZSTD_SUPPORT
  38static int perf_session__process_compressed_event(struct perf_session *session,
  39						  union perf_event *event, u64 file_offset)
  40{
  41	void *src;
  42	size_t decomp_size, src_size;
  43	u64 decomp_last_rem = 0;
  44	size_t mmap_len, decomp_len = session->header.env.comp_mmap_len;
  45	struct decomp *decomp, *decomp_last = session->decomp_last;
  46
  47	if (decomp_last) {
  48		decomp_last_rem = decomp_last->size - decomp_last->head;
  49		decomp_len += decomp_last_rem;
  50	}
  51
  52	mmap_len = sizeof(struct decomp) + decomp_len;
  53	decomp = mmap(NULL, mmap_len, PROT_READ|PROT_WRITE,
  54		      MAP_ANONYMOUS|MAP_PRIVATE, -1, 0);
  55	if (decomp == MAP_FAILED) {
  56		pr_err("Couldn't allocate memory for decompression\n");
  57		return -1;
  58	}
  59
  60	decomp->file_pos = file_offset;
  61	decomp->mmap_len = mmap_len;
  62	decomp->head = 0;
  63
  64	if (decomp_last_rem) {
  65		memcpy(decomp->data, &(decomp_last->data[decomp_last->head]), decomp_last_rem);
  66		decomp->size = decomp_last_rem;
  67	}
  68
  69	src = (void *)event + sizeof(struct perf_record_compressed);
  70	src_size = event->pack.header.size - sizeof(struct perf_record_compressed);
  71
  72	decomp_size = zstd_decompress_stream(&(session->zstd_data), src, src_size,
  73				&(decomp->data[decomp_last_rem]), decomp_len - decomp_last_rem);
  74	if (!decomp_size) {
  75		munmap(decomp, mmap_len);
  76		pr_err("Couldn't decompress data\n");
  77		return -1;
  78	}
  79
  80	decomp->size += decomp_size;
 
  81
  82	if (session->decomp == NULL) {
  83		session->decomp = decomp;
  84		session->decomp_last = decomp;
  85	} else {
  86		session->decomp_last->next = decomp;
  87		session->decomp_last = decomp;
  88	}
  89
  90	pr_debug("decomp (B): %zd to %zd\n", src_size, decomp_size);
 
 
 
 
 
 
 
 
 
  91
  92	return 0;
  93}
  94#else /* !HAVE_ZSTD_SUPPORT */
  95#define perf_session__process_compressed_event perf_session__process_compressed_event_stub
  96#endif
  97
  98static int perf_session__deliver_event(struct perf_session *session,
  99				       union perf_event *event,
 100				       struct perf_tool *tool,
 101				       u64 file_offset);
 
 102
 103static int perf_session__open(struct perf_session *session)
 104{
 105	struct perf_data *data = session->data;
 106
 107	if (perf_session__read_header(session) < 0) {
 108		pr_err("incompatible file format (rerun with -v to learn more)\n");
 109		return -1;
 110	}
 111
 112	if (perf_data__is_pipe(data))
 113		return 0;
 114
 115	if (perf_header__has_feat(&session->header, HEADER_STAT))
 116		return 0;
 117
 118	if (!evlist__valid_sample_type(session->evlist)) {
 119		pr_err("non matching sample_type\n");
 120		return -1;
 121	}
 122
 123	if (!evlist__valid_sample_id_all(session->evlist)) {
 124		pr_err("non matching sample_id_all\n");
 125		return -1;
 126	}
 127
 128	if (!perf_evlist__valid_read_format(session->evlist)) {
 129		pr_err("non matching read_format\n");
 130		return -1;
 131	}
 132
 
 133	return 0;
 
 
 
 
 
 134}
 135
 136void perf_session__set_id_hdr_size(struct perf_session *session)
 137{
 138	u16 id_hdr_size = perf_evlist__id_hdr_size(session->evlist);
 
 
 
 
 
 139
 140	machines__set_id_hdr_size(&session->machines, id_hdr_size);
 141}
 
 
 
 142
 143int perf_session__create_kernel_maps(struct perf_session *session)
 144{
 145	int ret = machine__create_kernel_maps(&session->machines.host);
 146
 147	if (ret >= 0)
 148		ret = machines__create_guest_kernel_maps(&session->machines);
 149	return ret;
 150}
 151
 152static void perf_session__destroy_kernel_maps(struct perf_session *session)
 153{
 154	machines__destroy_kernel_maps(&session->machines);
 
 155}
 156
 157static bool perf_session__has_comm_exec(struct perf_session *session)
 158{
 159	struct evsel *evsel;
 160
 161	evlist__for_each_entry(session->evlist, evsel) {
 162		if (evsel->core.attr.comm_exec)
 163			return true;
 164	}
 165
 166	return false;
 167}
 168
 169static void perf_session__set_comm_exec(struct perf_session *session)
 170{
 171	bool comm_exec = perf_session__has_comm_exec(session);
 172
 173	machines__set_comm_exec(&session->machines, comm_exec);
 
 
 174}
 175
 176static int ordered_events__deliver_event(struct ordered_events *oe,
 177					 struct ordered_event *event)
 178{
 179	struct perf_session *session = container_of(oe, struct perf_session,
 180						    ordered_events);
 181
 182	return perf_session__deliver_event(session, event->event,
 183					   session->tool, event->file_offset);
 184}
 185
 186struct perf_session *perf_session__new(struct perf_data *data,
 187				       bool repipe, struct perf_tool *tool)
 
 188{
 189	int ret = -ENOMEM;
 190	struct perf_session *session = zalloc(sizeof(*session));
 191
 192	if (!session)
 193		goto out;
 194
 195	session->repipe = repipe;
 196	session->tool   = tool;
 197	INIT_LIST_HEAD(&session->auxtrace_index);
 198	machines__init(&session->machines);
 199	ordered_events__init(&session->ordered_events,
 200			     ordered_events__deliver_event, NULL);
 201
 202	perf_env__init(&session->header.env);
 203	if (data) {
 204		ret = perf_data__open(data);
 205		if (ret < 0)
 
 
 
 
 
 
 
 
 
 
 
 206			goto out_delete;
 207
 208		session->data = data;
 209
 210		if (perf_data__is_read(data)) {
 211			ret = perf_session__open(session);
 212			if (ret < 0)
 213				goto out_delete;
 214
 215			/*
 216			 * set session attributes that are present in perf.data
 217			 * but not in pipe-mode.
 218			 */
 219			if (!data->is_pipe) {
 220				perf_session__set_id_hdr_size(session);
 221				perf_session__set_comm_exec(session);
 222			}
 223
 224			perf_evlist__init_trace_event_sample_raw(session->evlist);
 225
 226			/* Open the directory data. */
 227			if (data->is_dir) {
 228				ret = perf_data__open_dir(data);
 229				if (ret)
 230					goto out_delete;
 231			}
 232
 233			if (!symbol_conf.kallsyms_name &&
 234			    !symbol_conf.vmlinux_name)
 235				symbol_conf.kallsyms_name = perf_data__kallsyms_name(data);
 236		}
 237	} else  {
 238		session->machines.host.env = &perf_env;
 239	}
 240
 241	session->machines.host.single_address_space =
 242		perf_env__single_address_space(session->machines.host.env);
 243
 244	if (!data || perf_data__is_write(data)) {
 245		/*
 246		 * In O_RDONLY mode this will be performed when reading the
 247		 * kernel MMAP event, in perf_event__process_mmap().
 248		 */
 249		if (perf_session__create_kernel_maps(session) < 0)
 250			pr_warning("Cannot read kernel map\n");
 251	}
 252
 253	/*
 254	 * In pipe-mode, evlist is empty until PERF_RECORD_HEADER_ATTR is
 255	 * processed, so evlist__sample_id_all is not meaningful here.
 256	 */
 257	if ((!data || !data->is_pipe) && tool && tool->ordering_requires_timestamps &&
 258	    tool->ordered_events && !evlist__sample_id_all(session->evlist)) {
 259		dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n");
 260		tool->ordered_events = false;
 261	}
 262
 263	return session;
 264
 265 out_delete:
 266	perf_session__delete(session);
 267 out:
 268	return ERR_PTR(ret);
 269}
 270
 271static void perf_session__delete_threads(struct perf_session *session)
 272{
 273	machine__delete_threads(&session->machines.host);
 
 
 
 
 
 274}
 275
 276static void perf_session__release_decomp_events(struct perf_session *session)
 277{
 278	struct decomp *next, *decomp;
 279	size_t mmap_len;
 280	next = session->decomp;
 281	do {
 282		decomp = next;
 283		if (decomp == NULL)
 284			break;
 285		next = decomp->next;
 286		mmap_len = decomp->mmap_len;
 287		munmap(decomp, mmap_len);
 288	} while (1);
 289}
 290
 291void perf_session__delete(struct perf_session *session)
 292{
 293	if (session == NULL)
 294		return;
 295	auxtrace__free(session);
 296	auxtrace_index__free(&session->auxtrace_index);
 297	perf_session__destroy_kernel_maps(session);
 298	perf_session__delete_threads(session);
 299	perf_session__release_decomp_events(session);
 300	perf_env__exit(&session->header.env);
 301	machines__exit(&session->machines);
 302	if (session->data)
 303		perf_data__close(session->data);
 304	free(session);
 305}
 306
 307static int process_event_synth_tracing_data_stub(struct perf_session *session
 308						 __maybe_unused,
 309						 union perf_event *event
 310						 __maybe_unused)
 311{
 312	dump_printf(": unhandled!\n");
 313	return 0;
 314}
 315
 316static int process_event_synth_attr_stub(struct perf_tool *tool __maybe_unused,
 317					 union perf_event *event __maybe_unused,
 318					 struct evlist **pevlist
 319					 __maybe_unused)
 320{
 321	dump_printf(": unhandled!\n");
 322	return 0;
 
 
 
 
 323}
 324
 325static int process_event_synth_event_update_stub(struct perf_tool *tool __maybe_unused,
 326						 union perf_event *event __maybe_unused,
 327						 struct evlist **pevlist
 328						 __maybe_unused)
 329{
 330	if (dump_trace)
 331		perf_event__fprintf_event_update(event, stdout);
 332
 333	dump_printf(": unhandled!\n");
 334	return 0;
 
 
 335}
 336
 337static int process_event_sample_stub(struct perf_tool *tool __maybe_unused,
 338				     union perf_event *event __maybe_unused,
 339				     struct perf_sample *sample __maybe_unused,
 340				     struct evsel *evsel __maybe_unused,
 341				     struct machine *machine __maybe_unused)
 342{
 343	dump_printf(": unhandled!\n");
 344	return 0;
 345}
 346
 347static int process_event_stub(struct perf_tool *tool __maybe_unused,
 348			      union perf_event *event __maybe_unused,
 349			      struct perf_sample *sample __maybe_unused,
 350			      struct machine *machine __maybe_unused)
 351{
 352	dump_printf(": unhandled!\n");
 353	return 0;
 354}
 355
 356static int process_finished_round_stub(struct perf_tool *tool __maybe_unused,
 357				       union perf_event *event __maybe_unused,
 358				       struct ordered_events *oe __maybe_unused)
 
 359{
 360	dump_printf(": unhandled!\n");
 361	return 0;
 362}
 363
 364static int process_finished_round(struct perf_tool *tool,
 365				  union perf_event *event,
 366				  struct ordered_events *oe);
 367
 368static int skipn(int fd, off_t n)
 369{
 370	char buf[4096];
 371	ssize_t ret;
 372
 373	while (n > 0) {
 374		ret = read(fd, buf, min(n, (off_t)sizeof(buf)));
 375		if (ret <= 0)
 376			return ret;
 377		n -= ret;
 378	}
 379
 380	return 0;
 381}
 
 382
 383static s64 process_event_auxtrace_stub(struct perf_session *session __maybe_unused,
 384				       union perf_event *event)
 385{
 386	dump_printf(": unhandled!\n");
 387	if (perf_data__is_pipe(session->data))
 388		skipn(perf_data__fd(session->data), event->auxtrace.size);
 389	return event->auxtrace.size;
 390}
 391
 392static int process_event_op2_stub(struct perf_session *session __maybe_unused,
 393				  union perf_event *event __maybe_unused)
 394{
 395	dump_printf(": unhandled!\n");
 396	return 0;
 397}
 
 
 
 
 
 
 
 398
 
 
 
 
 
 
 
 
 
 
 399
 400static
 401int process_event_thread_map_stub(struct perf_session *session __maybe_unused,
 402				  union perf_event *event __maybe_unused)
 403{
 404	if (dump_trace)
 405		perf_event__fprintf_thread_map(event, stdout);
 406
 407	dump_printf(": unhandled!\n");
 408	return 0;
 409}
 410
 411static
 412int process_event_cpu_map_stub(struct perf_session *session __maybe_unused,
 413			       union perf_event *event __maybe_unused)
 414{
 415	if (dump_trace)
 416		perf_event__fprintf_cpu_map(event, stdout);
 417
 418	dump_printf(": unhandled!\n");
 419	return 0;
 420}
 421
 422static
 423int process_event_stat_config_stub(struct perf_session *session __maybe_unused,
 424				   union perf_event *event __maybe_unused)
 
 425{
 426	if (dump_trace)
 427		perf_event__fprintf_stat_config(event, stdout);
 428
 429	dump_printf(": unhandled!\n");
 430	return 0;
 431}
 432
 433static int process_stat_stub(struct perf_session *perf_session __maybe_unused,
 434			     union perf_event *event)
 
 435{
 436	if (dump_trace)
 437		perf_event__fprintf_stat(event, stdout);
 438
 439	dump_printf(": unhandled!\n");
 440	return 0;
 441}
 442
 443static int process_stat_round_stub(struct perf_session *perf_session __maybe_unused,
 444				   union perf_event *event)
 
 445{
 446	if (dump_trace)
 447		perf_event__fprintf_stat_round(event, stdout);
 448
 449	dump_printf(": unhandled!\n");
 450	return 0;
 451}
 452
 453static int perf_session__process_compressed_event_stub(struct perf_session *session __maybe_unused,
 454						       union perf_event *event __maybe_unused,
 455						       u64 file_offset __maybe_unused)
 456{
 457       dump_printf(": unhandled!\n");
 458       return 0;
 459}
 460
 461void perf_tool__fill_defaults(struct perf_tool *tool)
 462{
 463	if (tool->sample == NULL)
 464		tool->sample = process_event_sample_stub;
 465	if (tool->mmap == NULL)
 466		tool->mmap = process_event_stub;
 467	if (tool->mmap2 == NULL)
 468		tool->mmap2 = process_event_stub;
 469	if (tool->comm == NULL)
 470		tool->comm = process_event_stub;
 471	if (tool->namespaces == NULL)
 472		tool->namespaces = process_event_stub;
 473	if (tool->cgroup == NULL)
 474		tool->cgroup = process_event_stub;
 475	if (tool->fork == NULL)
 476		tool->fork = process_event_stub;
 477	if (tool->exit == NULL)
 478		tool->exit = process_event_stub;
 479	if (tool->lost == NULL)
 480		tool->lost = perf_event__process_lost;
 481	if (tool->lost_samples == NULL)
 482		tool->lost_samples = perf_event__process_lost_samples;
 483	if (tool->aux == NULL)
 484		tool->aux = perf_event__process_aux;
 485	if (tool->itrace_start == NULL)
 486		tool->itrace_start = perf_event__process_itrace_start;
 487	if (tool->context_switch == NULL)
 488		tool->context_switch = perf_event__process_switch;
 489	if (tool->ksymbol == NULL)
 490		tool->ksymbol = perf_event__process_ksymbol;
 491	if (tool->bpf == NULL)
 492		tool->bpf = perf_event__process_bpf;
 493	if (tool->text_poke == NULL)
 494		tool->text_poke = perf_event__process_text_poke;
 495	if (tool->read == NULL)
 496		tool->read = process_event_sample_stub;
 497	if (tool->throttle == NULL)
 498		tool->throttle = process_event_stub;
 499	if (tool->unthrottle == NULL)
 500		tool->unthrottle = process_event_stub;
 501	if (tool->attr == NULL)
 502		tool->attr = process_event_synth_attr_stub;
 503	if (tool->event_update == NULL)
 504		tool->event_update = process_event_synth_event_update_stub;
 505	if (tool->tracing_data == NULL)
 506		tool->tracing_data = process_event_synth_tracing_data_stub;
 507	if (tool->build_id == NULL)
 508		tool->build_id = process_event_op2_stub;
 509	if (tool->finished_round == NULL) {
 510		if (tool->ordered_events)
 511			tool->finished_round = process_finished_round;
 512		else
 513			tool->finished_round = process_finished_round_stub;
 514	}
 515	if (tool->id_index == NULL)
 516		tool->id_index = process_event_op2_stub;
 517	if (tool->auxtrace_info == NULL)
 518		tool->auxtrace_info = process_event_op2_stub;
 519	if (tool->auxtrace == NULL)
 520		tool->auxtrace = process_event_auxtrace_stub;
 521	if (tool->auxtrace_error == NULL)
 522		tool->auxtrace_error = process_event_op2_stub;
 523	if (tool->thread_map == NULL)
 524		tool->thread_map = process_event_thread_map_stub;
 525	if (tool->cpu_map == NULL)
 526		tool->cpu_map = process_event_cpu_map_stub;
 527	if (tool->stat_config == NULL)
 528		tool->stat_config = process_event_stat_config_stub;
 529	if (tool->stat == NULL)
 530		tool->stat = process_stat_stub;
 531	if (tool->stat_round == NULL)
 532		tool->stat_round = process_stat_round_stub;
 533	if (tool->time_conv == NULL)
 534		tool->time_conv = process_event_op2_stub;
 535	if (tool->feature == NULL)
 536		tool->feature = process_event_op2_stub;
 537	if (tool->compressed == NULL)
 538		tool->compressed = perf_session__process_compressed_event;
 539}
 540
 541static void swap_sample_id_all(union perf_event *event, void *data)
 542{
 543	void *end = (void *) event + event->header.size;
 544	int size = end - data;
 545
 546	BUG_ON(size % sizeof(u64));
 547	mem_bswap_64(data, size);
 
 
 
 548}
 549
 550static void perf_event__all64_swap(union perf_event *event,
 551				   bool sample_id_all __maybe_unused)
 552{
 553	struct perf_event_header *hdr = &event->header;
 554	mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr));
 555}
 556
 557static void perf_event__comm_swap(union perf_event *event, bool sample_id_all)
 558{
 559	event->comm.pid = bswap_32(event->comm.pid);
 560	event->comm.tid = bswap_32(event->comm.tid);
 561
 562	if (sample_id_all) {
 563		void *data = &event->comm.comm;
 564
 565		data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 566		swap_sample_id_all(event, data);
 567	}
 568}
 569
 570static void perf_event__mmap_swap(union perf_event *event,
 571				  bool sample_id_all)
 572{
 573	event->mmap.pid	  = bswap_32(event->mmap.pid);
 574	event->mmap.tid	  = bswap_32(event->mmap.tid);
 575	event->mmap.start = bswap_64(event->mmap.start);
 576	event->mmap.len	  = bswap_64(event->mmap.len);
 577	event->mmap.pgoff = bswap_64(event->mmap.pgoff);
 578
 579	if (sample_id_all) {
 580		void *data = &event->mmap.filename;
 581
 582		data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 583		swap_sample_id_all(event, data);
 584	}
 585}
 586
 587static void perf_event__mmap2_swap(union perf_event *event,
 588				  bool sample_id_all)
 589{
 590	event->mmap2.pid   = bswap_32(event->mmap2.pid);
 591	event->mmap2.tid   = bswap_32(event->mmap2.tid);
 592	event->mmap2.start = bswap_64(event->mmap2.start);
 593	event->mmap2.len   = bswap_64(event->mmap2.len);
 594	event->mmap2.pgoff = bswap_64(event->mmap2.pgoff);
 595	event->mmap2.maj   = bswap_32(event->mmap2.maj);
 596	event->mmap2.min   = bswap_32(event->mmap2.min);
 597	event->mmap2.ino   = bswap_64(event->mmap2.ino);
 598
 599	if (sample_id_all) {
 600		void *data = &event->mmap2.filename;
 601
 602		data += PERF_ALIGN(strlen(data) + 1, sizeof(u64));
 603		swap_sample_id_all(event, data);
 604	}
 605}
 606static void perf_event__task_swap(union perf_event *event, bool sample_id_all)
 607{
 608	event->fork.pid	 = bswap_32(event->fork.pid);
 609	event->fork.tid	 = bswap_32(event->fork.tid);
 610	event->fork.ppid = bswap_32(event->fork.ppid);
 611	event->fork.ptid = bswap_32(event->fork.ptid);
 612	event->fork.time = bswap_64(event->fork.time);
 613
 614	if (sample_id_all)
 615		swap_sample_id_all(event, &event->fork + 1);
 616}
 617
 618static void perf_event__read_swap(union perf_event *event, bool sample_id_all)
 619{
 620	event->read.pid		 = bswap_32(event->read.pid);
 621	event->read.tid		 = bswap_32(event->read.tid);
 622	event->read.value	 = bswap_64(event->read.value);
 623	event->read.time_enabled = bswap_64(event->read.time_enabled);
 624	event->read.time_running = bswap_64(event->read.time_running);
 625	event->read.id		 = bswap_64(event->read.id);
 626
 627	if (sample_id_all)
 628		swap_sample_id_all(event, &event->read + 1);
 629}
 630
 631static void perf_event__aux_swap(union perf_event *event, bool sample_id_all)
 632{
 633	event->aux.aux_offset = bswap_64(event->aux.aux_offset);
 634	event->aux.aux_size   = bswap_64(event->aux.aux_size);
 635	event->aux.flags      = bswap_64(event->aux.flags);
 636
 637	if (sample_id_all)
 638		swap_sample_id_all(event, &event->aux + 1);
 639}
 640
 641static void perf_event__itrace_start_swap(union perf_event *event,
 642					  bool sample_id_all)
 643{
 644	event->itrace_start.pid	 = bswap_32(event->itrace_start.pid);
 645	event->itrace_start.tid	 = bswap_32(event->itrace_start.tid);
 646
 647	if (sample_id_all)
 648		swap_sample_id_all(event, &event->itrace_start + 1);
 649}
 650
 651static void perf_event__switch_swap(union perf_event *event, bool sample_id_all)
 652{
 653	if (event->header.type == PERF_RECORD_SWITCH_CPU_WIDE) {
 654		event->context_switch.next_prev_pid =
 655				bswap_32(event->context_switch.next_prev_pid);
 656		event->context_switch.next_prev_tid =
 657				bswap_32(event->context_switch.next_prev_tid);
 658	}
 659
 660	if (sample_id_all)
 661		swap_sample_id_all(event, &event->context_switch + 1);
 662}
 663
 664static void perf_event__text_poke_swap(union perf_event *event, bool sample_id_all)
 665{
 666	event->text_poke.addr    = bswap_64(event->text_poke.addr);
 667	event->text_poke.old_len = bswap_16(event->text_poke.old_len);
 668	event->text_poke.new_len = bswap_16(event->text_poke.new_len);
 669
 670	if (sample_id_all) {
 671		size_t len = sizeof(event->text_poke.old_len) +
 672			     sizeof(event->text_poke.new_len) +
 673			     event->text_poke.old_len +
 674			     event->text_poke.new_len;
 675		void *data = &event->text_poke.old_len;
 676
 677		data += PERF_ALIGN(len, sizeof(u64));
 678		swap_sample_id_all(event, data);
 679	}
 680}
 681
 682static void perf_event__throttle_swap(union perf_event *event,
 683				      bool sample_id_all)
 684{
 685	event->throttle.time	  = bswap_64(event->throttle.time);
 686	event->throttle.id	  = bswap_64(event->throttle.id);
 687	event->throttle.stream_id = bswap_64(event->throttle.stream_id);
 688
 689	if (sample_id_all)
 690		swap_sample_id_all(event, &event->throttle + 1);
 691}
 692
 693static void perf_event__namespaces_swap(union perf_event *event,
 694					bool sample_id_all)
 695{
 696	u64 i;
 697
 698	event->namespaces.pid		= bswap_32(event->namespaces.pid);
 699	event->namespaces.tid		= bswap_32(event->namespaces.tid);
 700	event->namespaces.nr_namespaces	= bswap_64(event->namespaces.nr_namespaces);
 701
 702	for (i = 0; i < event->namespaces.nr_namespaces; i++) {
 703		struct perf_ns_link_info *ns = &event->namespaces.link_info[i];
 704
 705		ns->dev = bswap_64(ns->dev);
 706		ns->ino = bswap_64(ns->ino);
 707	}
 708
 709	if (sample_id_all)
 710		swap_sample_id_all(event, &event->namespaces.link_info[i]);
 711}
 712
 713static u8 revbyte(u8 b)
 714{
 715	int rev = (b >> 4) | ((b & 0xf) << 4);
 716	rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2);
 717	rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1);
 718	return (u8) rev;
 719}
 720
 721/*
 722 * XXX this is hack in attempt to carry flags bitfield
 723 * through endian village. ABI says:
 724 *
 725 * Bit-fields are allocated from right to left (least to most significant)
 726 * on little-endian implementations and from left to right (most to least
 727 * significant) on big-endian implementations.
 728 *
 729 * The above seems to be byte specific, so we need to reverse each
 730 * byte of the bitfield. 'Internet' also says this might be implementation
 731 * specific and we probably need proper fix and carry perf_event_attr
 732 * bitfield flags in separate data file FEAT_ section. Thought this seems
 733 * to work for now.
 734 */
 735static void swap_bitfield(u8 *p, unsigned len)
 736{
 737	unsigned i;
 738
 739	for (i = 0; i < len; i++) {
 740		*p = revbyte(*p);
 741		p++;
 742	}
 743}
 744
 745/* exported for swapping attributes in file header */
 746void perf_event__attr_swap(struct perf_event_attr *attr)
 747{
 748	attr->type		= bswap_32(attr->type);
 749	attr->size		= bswap_32(attr->size);
 750
 751#define bswap_safe(f, n) 					\
 752	(attr->size > (offsetof(struct perf_event_attr, f) + 	\
 753		       sizeof(attr->f) * (n)))
 754#define bswap_field(f, sz) 			\
 755do { 						\
 756	if (bswap_safe(f, 0))			\
 757		attr->f = bswap_##sz(attr->f);	\
 758} while(0)
 759#define bswap_field_16(f) bswap_field(f, 16)
 760#define bswap_field_32(f) bswap_field(f, 32)
 761#define bswap_field_64(f) bswap_field(f, 64)
 762
 763	bswap_field_64(config);
 764	bswap_field_64(sample_period);
 765	bswap_field_64(sample_type);
 766	bswap_field_64(read_format);
 767	bswap_field_32(wakeup_events);
 768	bswap_field_32(bp_type);
 769	bswap_field_64(bp_addr);
 770	bswap_field_64(bp_len);
 771	bswap_field_64(branch_sample_type);
 772	bswap_field_64(sample_regs_user);
 773	bswap_field_32(sample_stack_user);
 774	bswap_field_32(aux_watermark);
 775	bswap_field_16(sample_max_stack);
 776	bswap_field_32(aux_sample_size);
 777
 778	/*
 779	 * After read_format are bitfields. Check read_format because
 780	 * we are unable to use offsetof on bitfield.
 781	 */
 782	if (bswap_safe(read_format, 1))
 783		swap_bitfield((u8 *) (&attr->read_format + 1),
 784			      sizeof(u64));
 785#undef bswap_field_64
 786#undef bswap_field_32
 787#undef bswap_field
 788#undef bswap_safe
 789}
 790
 791static void perf_event__hdr_attr_swap(union perf_event *event,
 792				      bool sample_id_all __maybe_unused)
 793{
 794	size_t size;
 795
 796	perf_event__attr_swap(&event->attr.attr);
 797
 798	size = event->header.size;
 799	size -= (void *)&event->attr.id - (void *)event;
 800	mem_bswap_64(event->attr.id, size);
 801}
 802
 803static void perf_event__event_update_swap(union perf_event *event,
 804					  bool sample_id_all __maybe_unused)
 805{
 806	event->event_update.type = bswap_64(event->event_update.type);
 807	event->event_update.id   = bswap_64(event->event_update.id);
 808}
 809
 810static void perf_event__event_type_swap(union perf_event *event,
 811					bool sample_id_all __maybe_unused)
 812{
 813	event->event_type.event_type.event_id =
 814		bswap_64(event->event_type.event_type.event_id);
 815}
 816
 817static void perf_event__tracing_data_swap(union perf_event *event,
 818					  bool sample_id_all __maybe_unused)
 819{
 820	event->tracing_data.size = bswap_32(event->tracing_data.size);
 821}
 822
 823static void perf_event__auxtrace_info_swap(union perf_event *event,
 824					   bool sample_id_all __maybe_unused)
 825{
 826	size_t size;
 827
 828	event->auxtrace_info.type = bswap_32(event->auxtrace_info.type);
 
 
 
 
 
 
 
 
 
 
 
 
 
 829
 830	size = event->header.size;
 831	size -= (void *)&event->auxtrace_info.priv - (void *)event;
 832	mem_bswap_64(event->auxtrace_info.priv, size);
 833}
 834
 835static void perf_event__auxtrace_swap(union perf_event *event,
 836				      bool sample_id_all __maybe_unused)
 837{
 838	event->auxtrace.size      = bswap_64(event->auxtrace.size);
 839	event->auxtrace.offset    = bswap_64(event->auxtrace.offset);
 840	event->auxtrace.reference = bswap_64(event->auxtrace.reference);
 841	event->auxtrace.idx       = bswap_32(event->auxtrace.idx);
 842	event->auxtrace.tid       = bswap_32(event->auxtrace.tid);
 843	event->auxtrace.cpu       = bswap_32(event->auxtrace.cpu);
 844}
 845
 846static void perf_event__auxtrace_error_swap(union perf_event *event,
 847					    bool sample_id_all __maybe_unused)
 848{
 849	event->auxtrace_error.type = bswap_32(event->auxtrace_error.type);
 850	event->auxtrace_error.code = bswap_32(event->auxtrace_error.code);
 851	event->auxtrace_error.cpu  = bswap_32(event->auxtrace_error.cpu);
 852	event->auxtrace_error.pid  = bswap_32(event->auxtrace_error.pid);
 853	event->auxtrace_error.tid  = bswap_32(event->auxtrace_error.tid);
 854	event->auxtrace_error.fmt  = bswap_32(event->auxtrace_error.fmt);
 855	event->auxtrace_error.ip   = bswap_64(event->auxtrace_error.ip);
 856	if (event->auxtrace_error.fmt)
 857		event->auxtrace_error.time = bswap_64(event->auxtrace_error.time);
 858}
 859
 860static void perf_event__thread_map_swap(union perf_event *event,
 861					bool sample_id_all __maybe_unused)
 862{
 863	unsigned i;
 864
 865	event->thread_map.nr = bswap_64(event->thread_map.nr);
 866
 867	for (i = 0; i < event->thread_map.nr; i++)
 868		event->thread_map.entries[i].pid = bswap_64(event->thread_map.entries[i].pid);
 869}
 870
 871static void perf_event__cpu_map_swap(union perf_event *event,
 872				     bool sample_id_all __maybe_unused)
 873{
 874	struct perf_record_cpu_map_data *data = &event->cpu_map.data;
 875	struct cpu_map_entries *cpus;
 876	struct perf_record_record_cpu_map *mask;
 877	unsigned i;
 
 
 
 
 
 
 
 
 
 878
 879	data->type = bswap_64(data->type);
 
 880
 881	switch (data->type) {
 882	case PERF_CPU_MAP__CPUS:
 883		cpus = (struct cpu_map_entries *)data->data;
 884
 885		cpus->nr = bswap_16(cpus->nr);
 
 
 
 
 
 886
 887		for (i = 0; i < cpus->nr; i++)
 888			cpus->cpu[i] = bswap_16(cpus->cpu[i]);
 889		break;
 890	case PERF_CPU_MAP__MASK:
 891		mask = (struct perf_record_record_cpu_map *)data->data;
 892
 893		mask->nr = bswap_16(mask->nr);
 894		mask->long_size = bswap_16(mask->long_size);
 895
 896		switch (mask->long_size) {
 897		case 4: mem_bswap_32(&mask->mask, mask->nr); break;
 898		case 8: mem_bswap_64(&mask->mask, mask->nr); break;
 899		default:
 900			pr_err("cpu_map swap: unsupported long size\n");
 901		}
 902	default:
 903		break;
 904	}
 905}
 906
 907static void perf_event__stat_config_swap(union perf_event *event,
 908					 bool sample_id_all __maybe_unused)
 909{
 910	u64 size;
 911
 912	size  = event->stat_config.nr * sizeof(event->stat_config.data[0]);
 913	size += 1; /* nr item itself */
 914	mem_bswap_64(&event->stat_config.nr, size);
 915}
 916
 917static void perf_event__stat_swap(union perf_event *event,
 918				  bool sample_id_all __maybe_unused)
 919{
 920	event->stat.id     = bswap_64(event->stat.id);
 921	event->stat.thread = bswap_32(event->stat.thread);
 922	event->stat.cpu    = bswap_32(event->stat.cpu);
 923	event->stat.val    = bswap_64(event->stat.val);
 924	event->stat.ena    = bswap_64(event->stat.ena);
 925	event->stat.run    = bswap_64(event->stat.run);
 926}
 927
 928static void perf_event__stat_round_swap(union perf_event *event,
 929					bool sample_id_all __maybe_unused)
 930{
 931	event->stat_round.type = bswap_64(event->stat_round.type);
 932	event->stat_round.time = bswap_64(event->stat_round.time);
 933}
 934
 935typedef void (*perf_event__swap_op)(union perf_event *event,
 936				    bool sample_id_all);
 937
 938static perf_event__swap_op perf_event__swap_ops[] = {
 939	[PERF_RECORD_MMAP]		  = perf_event__mmap_swap,
 940	[PERF_RECORD_MMAP2]		  = perf_event__mmap2_swap,
 941	[PERF_RECORD_COMM]		  = perf_event__comm_swap,
 942	[PERF_RECORD_FORK]		  = perf_event__task_swap,
 943	[PERF_RECORD_EXIT]		  = perf_event__task_swap,
 944	[PERF_RECORD_LOST]		  = perf_event__all64_swap,
 945	[PERF_RECORD_READ]		  = perf_event__read_swap,
 946	[PERF_RECORD_THROTTLE]		  = perf_event__throttle_swap,
 947	[PERF_RECORD_UNTHROTTLE]	  = perf_event__throttle_swap,
 948	[PERF_RECORD_SAMPLE]		  = perf_event__all64_swap,
 949	[PERF_RECORD_AUX]		  = perf_event__aux_swap,
 950	[PERF_RECORD_ITRACE_START]	  = perf_event__itrace_start_swap,
 951	[PERF_RECORD_LOST_SAMPLES]	  = perf_event__all64_swap,
 952	[PERF_RECORD_SWITCH]		  = perf_event__switch_swap,
 953	[PERF_RECORD_SWITCH_CPU_WIDE]	  = perf_event__switch_swap,
 954	[PERF_RECORD_NAMESPACES]	  = perf_event__namespaces_swap,
 955	[PERF_RECORD_TEXT_POKE]		  = perf_event__text_poke_swap,
 956	[PERF_RECORD_HEADER_ATTR]	  = perf_event__hdr_attr_swap,
 957	[PERF_RECORD_HEADER_EVENT_TYPE]	  = perf_event__event_type_swap,
 958	[PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap,
 959	[PERF_RECORD_HEADER_BUILD_ID]	  = NULL,
 960	[PERF_RECORD_ID_INDEX]		  = perf_event__all64_swap,
 961	[PERF_RECORD_AUXTRACE_INFO]	  = perf_event__auxtrace_info_swap,
 962	[PERF_RECORD_AUXTRACE]		  = perf_event__auxtrace_swap,
 963	[PERF_RECORD_AUXTRACE_ERROR]	  = perf_event__auxtrace_error_swap,
 964	[PERF_RECORD_THREAD_MAP]	  = perf_event__thread_map_swap,
 965	[PERF_RECORD_CPU_MAP]		  = perf_event__cpu_map_swap,
 966	[PERF_RECORD_STAT_CONFIG]	  = perf_event__stat_config_swap,
 967	[PERF_RECORD_STAT]		  = perf_event__stat_swap,
 968	[PERF_RECORD_STAT_ROUND]	  = perf_event__stat_round_swap,
 969	[PERF_RECORD_EVENT_UPDATE]	  = perf_event__event_update_swap,
 970	[PERF_RECORD_TIME_CONV]		  = perf_event__all64_swap,
 971	[PERF_RECORD_HEADER_MAX]	  = NULL,
 972};
 973
 974/*
 975 * When perf record finishes a pass on every buffers, it records this pseudo
 976 * event.
 977 * We record the max timestamp t found in the pass n.
 978 * Assuming these timestamps are monotonic across cpus, we know that if
 979 * a buffer still has events with timestamps below t, they will be all
 980 * available and then read in the pass n + 1.
 981 * Hence when we start to read the pass n + 2, we can safely flush every
 982 * events with timestamps below t.
 983 *
 984 *    ============ PASS n =================
 985 *       CPU 0         |   CPU 1
 986 *                     |
 987 *    cnt1 timestamps  |   cnt2 timestamps
 988 *          1          |         2
 989 *          2          |         3
 990 *          -          |         4  <--- max recorded
 991 *
 992 *    ============ PASS n + 1 ==============
 993 *       CPU 0         |   CPU 1
 994 *                     |
 995 *    cnt1 timestamps  |   cnt2 timestamps
 996 *          3          |         5
 997 *          4          |         6
 998 *          5          |         7 <---- max recorded
 999 *
1000 *      Flush every events below timestamp 4
1001 *
1002 *    ============ PASS n + 2 ==============
1003 *       CPU 0         |   CPU 1
1004 *                     |
1005 *    cnt1 timestamps  |   cnt2 timestamps
1006 *          6          |         8
1007 *          7          |         9
1008 *          -          |         10
1009 *
1010 *      Flush every events below timestamp 7
1011 *      etc...
1012 */
1013static int process_finished_round(struct perf_tool *tool __maybe_unused,
1014				  union perf_event *event __maybe_unused,
1015				  struct ordered_events *oe)
1016{
1017	if (dump_trace)
1018		fprintf(stdout, "\n");
1019	return ordered_events__flush(oe, OE_FLUSH__ROUND);
1020}
1021
1022int perf_session__queue_event(struct perf_session *s, union perf_event *event,
1023			      u64 timestamp, u64 file_offset)
1024{
1025	return ordered_events__queue(&s->ordered_events, event, timestamp, file_offset);
1026}
1027
1028static void callchain__lbr_callstack_printf(struct perf_sample *sample)
 
1029{
1030	struct ip_callchain *callchain = sample->callchain;
1031	struct branch_stack *lbr_stack = sample->branch_stack;
1032	struct branch_entry *entries = perf_sample__branch_entries(sample);
1033	u64 kernel_callchain_nr = callchain->nr;
1034	unsigned int i;
1035
1036	for (i = 0; i < kernel_callchain_nr; i++) {
1037		if (callchain->ips[i] == PERF_CONTEXT_USER)
1038			break;
 
 
1039	}
1040
1041	if ((i != kernel_callchain_nr) && lbr_stack->nr) {
1042		u64 total_nr;
1043		/*
1044		 * LBR callstack can only get user call chain,
1045		 * i is kernel call chain number,
1046		 * 1 is PERF_CONTEXT_USER.
1047		 *
1048		 * The user call chain is stored in LBR registers.
1049		 * LBR are pair registers. The caller is stored
1050		 * in "from" register, while the callee is stored
1051		 * in "to" register.
1052		 * For example, there is a call stack
1053		 * "A"->"B"->"C"->"D".
1054		 * The LBR registers will recorde like
1055		 * "C"->"D", "B"->"C", "A"->"B".
1056		 * So only the first "to" register and all "from"
1057		 * registers are needed to construct the whole stack.
1058		 */
1059		total_nr = i + 1 + lbr_stack->nr + 1;
1060		kernel_callchain_nr = i + 1;
1061
1062		printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr);
1063
1064		for (i = 0; i < kernel_callchain_nr; i++)
1065			printf("..... %2d: %016" PRIx64 "\n",
1066			       i, callchain->ips[i]);
1067
1068		printf("..... %2d: %016" PRIx64 "\n",
1069		       (int)(kernel_callchain_nr), entries[0].to);
1070		for (i = 0; i < lbr_stack->nr; i++)
1071			printf("..... %2d: %016" PRIx64 "\n",
1072			       (int)(i + kernel_callchain_nr + 1), entries[i].from);
1073	}
1074}
1075
1076static void callchain__printf(struct evsel *evsel,
1077			      struct perf_sample *sample)
 
 
1078{
1079	unsigned int i;
1080	struct ip_callchain *callchain = sample->callchain;
 
 
1081
1082	if (evsel__has_branch_callstack(evsel))
1083		callchain__lbr_callstack_printf(sample);
1084
1085	printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr);
1086
1087	for (i = 0; i < callchain->nr; i++)
1088		printf("..... %2d: %016" PRIx64 "\n",
1089		       i, callchain->ips[i]);
1090}
1091
1092static void branch_stack__printf(struct perf_sample *sample, bool callstack)
1093{
1094	struct branch_entry *entries = perf_sample__branch_entries(sample);
1095	uint64_t i;
1096
1097	printf("%s: nr:%" PRIu64 "\n",
1098		!callstack ? "... branch stack" : "... branch callstack",
1099		sample->branch_stack->nr);
1100
1101	for (i = 0; i < sample->branch_stack->nr; i++) {
1102		struct branch_entry *e = &entries[i];
1103
1104		if (!callstack) {
1105			printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x\n",
1106				i, e->from, e->to,
1107				(unsigned short)e->flags.cycles,
1108				e->flags.mispred ? "M" : " ",
1109				e->flags.predicted ? "P" : " ",
1110				e->flags.abort ? "A" : " ",
1111				e->flags.in_tx ? "T" : " ",
1112				(unsigned)e->flags.reserved);
1113		} else {
1114			printf("..... %2"PRIu64": %016" PRIx64 "\n",
1115				i, i > 0 ? e->from : e->to);
1116		}
1117	}
1118}
1119
1120static void regs_dump__printf(u64 mask, u64 *regs)
1121{
1122	unsigned rid, i = 0;
1123
1124	for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) {
1125		u64 val = regs[i++];
1126
1127		printf(".... %-5s 0x%016" PRIx64 "\n",
1128		       perf_reg_name(rid), val);
 
 
 
 
 
1129	}
1130}
1131
1132static const char *regs_abi[] = {
1133	[PERF_SAMPLE_REGS_ABI_NONE] = "none",
1134	[PERF_SAMPLE_REGS_ABI_32] = "32-bit",
1135	[PERF_SAMPLE_REGS_ABI_64] = "64-bit",
1136};
1137
1138static inline const char *regs_dump_abi(struct regs_dump *d)
1139{
1140	if (d->abi > PERF_SAMPLE_REGS_ABI_64)
1141		return "unknown";
1142
1143	return regs_abi[d->abi];
1144}
1145
1146static void regs__printf(const char *type, struct regs_dump *regs)
1147{
1148	u64 mask = regs->mask;
1149
1150	printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n",
1151	       type,
1152	       mask,
1153	       regs_dump_abi(regs));
1154
1155	regs_dump__printf(mask, regs->regs);
1156}
1157
1158static void regs_user__printf(struct perf_sample *sample)
1159{
1160	struct regs_dump *user_regs = &sample->user_regs;
1161
1162	if (user_regs->regs)
1163		regs__printf("user", user_regs);
1164}
1165
1166static void regs_intr__printf(struct perf_sample *sample)
1167{
1168	struct regs_dump *intr_regs = &sample->intr_regs;
1169
1170	if (intr_regs->regs)
1171		regs__printf("intr", intr_regs);
1172}
1173
1174static void stack_user__printf(struct stack_dump *dump)
1175{
1176	printf("... ustack: size %" PRIu64 ", offset 0x%x\n",
1177	       dump->size, dump->offset);
1178}
1179
1180static void perf_evlist__print_tstamp(struct evlist *evlist,
1181				       union perf_event *event,
1182				       struct perf_sample *sample)
1183{
1184	u64 sample_type = __evlist__combined_sample_type(evlist);
1185
1186	if (event->header.type != PERF_RECORD_SAMPLE &&
1187	    !evlist__sample_id_all(evlist)) {
1188		fputs("-1 -1 ", stdout);
1189		return;
1190	}
1191
1192	if ((sample_type & PERF_SAMPLE_CPU))
1193		printf("%u ", sample->cpu);
1194
1195	if (sample_type & PERF_SAMPLE_TIME)
1196		printf("%" PRIu64 " ", sample->time);
1197}
1198
1199static void sample_read__printf(struct perf_sample *sample, u64 read_format)
1200{
1201	printf("... sample_read:\n");
1202
1203	if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1204		printf("...... time enabled %016" PRIx64 "\n",
1205		       sample->read.time_enabled);
1206
1207	if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1208		printf("...... time running %016" PRIx64 "\n",
1209		       sample->read.time_running);
1210
1211	if (read_format & PERF_FORMAT_GROUP) {
1212		u64 i;
1213
1214		printf(".... group nr %" PRIu64 "\n", sample->read.group.nr);
1215
1216		for (i = 0; i < sample->read.group.nr; i++) {
1217			struct sample_read_value *value;
1218
1219			value = &sample->read.group.values[i];
1220			printf("..... id %016" PRIx64
1221			       ", value %016" PRIx64 "\n",
1222			       value->id, value->value);
1223		}
1224	} else
1225		printf("..... id %016" PRIx64 ", value %016" PRIx64 "\n",
1226			sample->read.one.id, sample->read.one.value);
1227}
1228
1229static void dump_event(struct evlist *evlist, union perf_event *event,
1230		       u64 file_offset, struct perf_sample *sample)
1231{
1232	if (!dump_trace)
1233		return;
1234
1235	printf("\n%#" PRIx64 " [%#x]: event: %d\n",
1236	       file_offset, event->header.size, event->header.type);
1237
1238	trace_event(event);
1239	if (event->header.type == PERF_RECORD_SAMPLE && evlist->trace_event_sample_raw)
1240		evlist->trace_event_sample_raw(evlist, event, sample);
1241
1242	if (sample)
1243		perf_evlist__print_tstamp(evlist, event, sample);
1244
1245	printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset,
1246	       event->header.size, perf_event__name(event->header.type));
1247}
1248
1249static void dump_sample(struct evsel *evsel, union perf_event *event,
1250			struct perf_sample *sample)
1251{
1252	u64 sample_type;
1253
1254	if (!dump_trace)
1255		return;
1256
1257	printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n",
1258	       event->header.misc, sample->pid, sample->tid, sample->ip,
1259	       sample->period, sample->addr);
1260
1261	sample_type = evsel->core.attr.sample_type;
1262
1263	if (evsel__has_callchain(evsel))
1264		callchain__printf(evsel, sample);
1265
1266	if (evsel__has_br_stack(evsel))
1267		branch_stack__printf(sample, evsel__has_branch_callstack(evsel));
1268
1269	if (sample_type & PERF_SAMPLE_REGS_USER)
1270		regs_user__printf(sample);
1271
1272	if (sample_type & PERF_SAMPLE_REGS_INTR)
1273		regs_intr__printf(sample);
1274
1275	if (sample_type & PERF_SAMPLE_STACK_USER)
1276		stack_user__printf(&sample->user_stack);
1277
1278	if (sample_type & PERF_SAMPLE_WEIGHT)
1279		printf("... weight: %" PRIu64 "\n", sample->weight);
1280
1281	if (sample_type & PERF_SAMPLE_DATA_SRC)
1282		printf(" . data_src: 0x%"PRIx64"\n", sample->data_src);
1283
1284	if (sample_type & PERF_SAMPLE_PHYS_ADDR)
1285		printf(" .. phys_addr: 0x%"PRIx64"\n", sample->phys_addr);
1286
1287	if (sample_type & PERF_SAMPLE_TRANSACTION)
1288		printf("... transaction: %" PRIx64 "\n", sample->transaction);
1289
1290	if (sample_type & PERF_SAMPLE_READ)
1291		sample_read__printf(sample, evsel->core.attr.read_format);
1292}
1293
1294static void dump_read(struct evsel *evsel, union perf_event *event)
 
 
 
 
1295{
1296	struct perf_record_read *read_event = &event->read;
1297	u64 read_format;
1298
1299	if (!dump_trace)
1300		return;
1301
1302	printf(": %d %d %s %" PRI_lu64 "\n", event->read.pid, event->read.tid,
1303	       evsel__name(evsel), event->read.value);
1304
1305	if (!evsel)
1306		return;
1307
1308	read_format = evsel->core.attr.read_format;
1309
1310	if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
1311		printf("... time enabled : %" PRI_lu64 "\n", read_event->time_enabled);
1312
1313	if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
1314		printf("... time running : %" PRI_lu64 "\n", read_event->time_running);
1315
1316	if (read_format & PERF_FORMAT_ID)
1317		printf("... id           : %" PRI_lu64 "\n", read_event->id);
1318}
1319
1320static struct machine *machines__find_for_cpumode(struct machines *machines,
1321					       union perf_event *event,
1322					       struct perf_sample *sample)
1323{
1324	struct machine *machine;
1325
1326	if (perf_guest &&
1327	    ((sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL) ||
1328	     (sample->cpumode == PERF_RECORD_MISC_GUEST_USER))) {
1329		u32 pid;
1330
1331		if (event->header.type == PERF_RECORD_MMAP
1332		    || event->header.type == PERF_RECORD_MMAP2)
1333			pid = event->mmap.pid;
1334		else
1335			pid = sample->pid;
1336
1337		machine = machines__find(machines, pid);
1338		if (!machine)
1339			machine = machines__findnew(machines, DEFAULT_GUEST_KERNEL_ID);
1340		return machine;
1341	}
1342
1343	return &machines->host;
1344}
1345
1346static int deliver_sample_value(struct evlist *evlist,
1347				struct perf_tool *tool,
1348				union perf_event *event,
1349				struct perf_sample *sample,
1350				struct sample_read_value *v,
1351				struct machine *machine)
1352{
1353	struct perf_sample_id *sid = perf_evlist__id2sid(evlist, v->id);
1354	struct evsel *evsel;
1355
1356	if (sid) {
1357		sample->id     = v->id;
1358		sample->period = v->value - sid->period;
1359		sid->period    = v->value;
1360	}
1361
1362	if (!sid || sid->evsel == NULL) {
1363		++evlist->stats.nr_unknown_id;
1364		return 0;
1365	}
1366
1367	/*
1368	 * There's no reason to deliver sample
1369	 * for zero period, bail out.
1370	 */
1371	if (!sample->period)
1372		return 0;
1373
1374	evsel = container_of(sid->evsel, struct evsel, core);
1375	return tool->sample(tool, event, sample, evsel, machine);
1376}
1377
1378static int deliver_sample_group(struct evlist *evlist,
1379				struct perf_tool *tool,
1380				union  perf_event *event,
1381				struct perf_sample *sample,
1382				struct machine *machine)
1383{
1384	int ret = -EINVAL;
1385	u64 i;
1386
1387	for (i = 0; i < sample->read.group.nr; i++) {
1388		ret = deliver_sample_value(evlist, tool, event, sample,
1389					   &sample->read.group.values[i],
1390					   machine);
1391		if (ret)
1392			break;
1393	}
1394
1395	return ret;
1396}
1397
1398static int
1399 perf_evlist__deliver_sample(struct evlist *evlist,
1400			     struct perf_tool *tool,
1401			     union  perf_event *event,
1402			     struct perf_sample *sample,
1403			     struct evsel *evsel,
1404			     struct machine *machine)
1405{
1406	/* We know evsel != NULL. */
1407	u64 sample_type = evsel->core.attr.sample_type;
1408	u64 read_format = evsel->core.attr.read_format;
1409
1410	/* Standard sample delivery. */
1411	if (!(sample_type & PERF_SAMPLE_READ))
1412		return tool->sample(tool, event, sample, evsel, machine);
1413
1414	/* For PERF_SAMPLE_READ we have either single or group mode. */
1415	if (read_format & PERF_FORMAT_GROUP)
1416		return deliver_sample_group(evlist, tool, event, sample,
1417					    machine);
1418	else
1419		return deliver_sample_value(evlist, tool, event, sample,
1420					    &sample->read.one, machine);
1421}
1422
1423static int machines__deliver_event(struct machines *machines,
1424				   struct evlist *evlist,
1425				   union perf_event *event,
1426				   struct perf_sample *sample,
1427				   struct perf_tool *tool, u64 file_offset)
1428{
1429	struct evsel *evsel;
1430	struct machine *machine;
1431
1432	dump_event(evlist, event, file_offset, sample);
1433
1434	evsel = perf_evlist__id2evsel(evlist, sample->id);
1435
1436	machine = machines__find_for_cpumode(machines, event, sample);
1437
1438	switch (event->header.type) {
1439	case PERF_RECORD_SAMPLE:
 
 
1440		if (evsel == NULL) {
1441			++evlist->stats.nr_unknown_id;
1442			return 0;
1443		}
1444		dump_sample(evsel, event, sample);
1445		if (machine == NULL) {
1446			++evlist->stats.nr_unprocessable_samples;
1447			return 0;
1448		}
1449		return perf_evlist__deliver_sample(evlist, tool, event, sample, evsel, machine);
1450	case PERF_RECORD_MMAP:
1451		return tool->mmap(tool, event, sample, machine);
1452	case PERF_RECORD_MMAP2:
1453		if (event->header.misc & PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT)
1454			++evlist->stats.nr_proc_map_timeout;
1455		return tool->mmap2(tool, event, sample, machine);
1456	case PERF_RECORD_COMM:
1457		return tool->comm(tool, event, sample, machine);
1458	case PERF_RECORD_NAMESPACES:
1459		return tool->namespaces(tool, event, sample, machine);
1460	case PERF_RECORD_CGROUP:
1461		return tool->cgroup(tool, event, sample, machine);
1462	case PERF_RECORD_FORK:
1463		return tool->fork(tool, event, sample, machine);
1464	case PERF_RECORD_EXIT:
1465		return tool->exit(tool, event, sample, machine);
1466	case PERF_RECORD_LOST:
1467		if (tool->lost == perf_event__process_lost)
1468			evlist->stats.total_lost += event->lost.lost;
1469		return tool->lost(tool, event, sample, machine);
1470	case PERF_RECORD_LOST_SAMPLES:
1471		if (tool->lost_samples == perf_event__process_lost_samples)
1472			evlist->stats.total_lost_samples += event->lost_samples.lost;
1473		return tool->lost_samples(tool, event, sample, machine);
1474	case PERF_RECORD_READ:
1475		dump_read(evsel, event);
1476		return tool->read(tool, event, sample, evsel, machine);
1477	case PERF_RECORD_THROTTLE:
1478		return tool->throttle(tool, event, sample, machine);
1479	case PERF_RECORD_UNTHROTTLE:
1480		return tool->unthrottle(tool, event, sample, machine);
1481	case PERF_RECORD_AUX:
1482		if (tool->aux == perf_event__process_aux) {
1483			if (event->aux.flags & PERF_AUX_FLAG_TRUNCATED)
1484				evlist->stats.total_aux_lost += 1;
1485			if (event->aux.flags & PERF_AUX_FLAG_PARTIAL)
1486				evlist->stats.total_aux_partial += 1;
1487		}
1488		return tool->aux(tool, event, sample, machine);
1489	case PERF_RECORD_ITRACE_START:
1490		return tool->itrace_start(tool, event, sample, machine);
1491	case PERF_RECORD_SWITCH:
1492	case PERF_RECORD_SWITCH_CPU_WIDE:
1493		return tool->context_switch(tool, event, sample, machine);
1494	case PERF_RECORD_KSYMBOL:
1495		return tool->ksymbol(tool, event, sample, machine);
1496	case PERF_RECORD_BPF_EVENT:
1497		return tool->bpf(tool, event, sample, machine);
1498	case PERF_RECORD_TEXT_POKE:
1499		return tool->text_poke(tool, event, sample, machine);
1500	default:
1501		++evlist->stats.nr_unknown_events;
1502		return -1;
1503	}
1504}
1505
1506static int perf_session__deliver_event(struct perf_session *session,
1507				       union perf_event *event,
1508				       struct perf_tool *tool,
1509				       u64 file_offset)
1510{
1511	struct perf_sample sample;
1512	int ret;
 
1513
1514	ret = perf_evlist__parse_sample(session->evlist, event, &sample);
1515	if (ret) {
1516		pr_err("Can't parse sample, err = %d\n", ret);
1517		return ret;
 
1518	}
1519
1520	ret = auxtrace__process_event(session, event, &sample, tool);
1521	if (ret < 0)
1522		return ret;
1523	if (ret > 0)
1524		return 0;
1525
1526	ret = machines__deliver_event(&session->machines, session->evlist,
1527				      event, &sample, tool, file_offset);
1528
1529	if (dump_trace && sample.aux_sample.size)
1530		auxtrace__dump_auxtrace_sample(session, &sample);
1531
1532	return ret;
1533}
1534
1535static s64 perf_session__process_user_event(struct perf_session *session,
1536					    union perf_event *event,
1537					    u64 file_offset)
1538{
1539	struct ordered_events *oe = &session->ordered_events;
1540	struct perf_tool *tool = session->tool;
1541	struct perf_sample sample = { .time = 0, };
1542	int fd = perf_data__fd(session->data);
1543	int err;
1544
1545	if (event->header.type != PERF_RECORD_COMPRESSED ||
1546	    tool->compressed == perf_session__process_compressed_event_stub)
1547		dump_event(session->evlist, event, file_offset, &sample);
1548
1549	/* These events are processed right away */
1550	switch (event->header.type) {
1551	case PERF_RECORD_HEADER_ATTR:
1552		err = tool->attr(tool, event, &session->evlist);
1553		if (err == 0) {
1554			perf_session__set_id_hdr_size(session);
1555			perf_session__set_comm_exec(session);
1556		}
1557		return err;
1558	case PERF_RECORD_EVENT_UPDATE:
1559		return tool->event_update(tool, event, &session->evlist);
1560	case PERF_RECORD_HEADER_EVENT_TYPE:
1561		/*
1562		 * Depreceated, but we need to handle it for sake
1563		 * of old data files create in pipe mode.
1564		 */
1565		return 0;
1566	case PERF_RECORD_HEADER_TRACING_DATA:
1567		/*
1568		 * Setup for reading amidst mmap, but only when we
1569		 * are in 'file' mode. The 'pipe' fd is in proper
1570		 * place already.
1571		 */
1572		if (!perf_data__is_pipe(session->data))
1573			lseek(fd, file_offset, SEEK_SET);
1574		return tool->tracing_data(session, event);
1575	case PERF_RECORD_HEADER_BUILD_ID:
1576		return tool->build_id(session, event);
1577	case PERF_RECORD_FINISHED_ROUND:
1578		return tool->finished_round(tool, event, oe);
1579	case PERF_RECORD_ID_INDEX:
1580		return tool->id_index(session, event);
1581	case PERF_RECORD_AUXTRACE_INFO:
1582		return tool->auxtrace_info(session, event);
1583	case PERF_RECORD_AUXTRACE:
1584		/* setup for reading amidst mmap */
1585		lseek(fd, file_offset + event->header.size, SEEK_SET);
1586		return tool->auxtrace(session, event);
1587	case PERF_RECORD_AUXTRACE_ERROR:
1588		perf_session__auxtrace_error_inc(session, event);
1589		return tool->auxtrace_error(session, event);
1590	case PERF_RECORD_THREAD_MAP:
1591		return tool->thread_map(session, event);
1592	case PERF_RECORD_CPU_MAP:
1593		return tool->cpu_map(session, event);
1594	case PERF_RECORD_STAT_CONFIG:
1595		return tool->stat_config(session, event);
1596	case PERF_RECORD_STAT:
1597		return tool->stat(session, event);
1598	case PERF_RECORD_STAT_ROUND:
1599		return tool->stat_round(session, event);
1600	case PERF_RECORD_TIME_CONV:
1601		session->time_conv = event->time_conv;
1602		return tool->time_conv(session, event);
1603	case PERF_RECORD_HEADER_FEATURE:
1604		return tool->feature(session, event);
1605	case PERF_RECORD_COMPRESSED:
1606		err = tool->compressed(session, event, file_offset);
1607		if (err)
1608			dump_event(session->evlist, event, file_offset, &sample);
1609		return err;
1610	default:
1611		return -EINVAL;
1612	}
1613}
1614
1615int perf_session__deliver_synth_event(struct perf_session *session,
1616				      union perf_event *event,
1617				      struct perf_sample *sample)
 
1618{
1619	struct evlist *evlist = session->evlist;
1620	struct perf_tool *tool = session->tool;
1621
1622	events_stats__inc(&evlist->stats, event->header.type);
1623
1624	if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1625		return perf_session__process_user_event(session, event, 0);
1626
1627	return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0);
1628}
1629
1630static void event_swap(union perf_event *event, bool sample_id_all)
1631{
1632	perf_event__swap_op swap;
1633
1634	swap = perf_event__swap_ops[event->header.type];
1635	if (swap)
1636		swap(event, sample_id_all);
1637}
1638
1639int perf_session__peek_event(struct perf_session *session, off_t file_offset,
1640			     void *buf, size_t buf_sz,
1641			     union perf_event **event_ptr,
1642			     struct perf_sample *sample)
1643{
1644	union perf_event *event;
1645	size_t hdr_sz, rest;
1646	int fd;
1647
1648	if (session->one_mmap && !session->header.needs_swap) {
1649		event = file_offset - session->one_mmap_offset +
1650			session->one_mmap_addr;
1651		goto out_parse_sample;
1652	}
1653
1654	if (perf_data__is_pipe(session->data))
1655		return -1;
1656
1657	fd = perf_data__fd(session->data);
1658	hdr_sz = sizeof(struct perf_event_header);
1659
1660	if (buf_sz < hdr_sz)
1661		return -1;
1662
1663	if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 ||
1664	    readn(fd, buf, hdr_sz) != (ssize_t)hdr_sz)
1665		return -1;
1666
1667	event = (union perf_event *)buf;
1668
1669	if (session->header.needs_swap)
1670		perf_event_header__bswap(&event->header);
1671
1672	if (event->header.size < hdr_sz || event->header.size > buf_sz)
1673		return -1;
1674
1675	rest = event->header.size - hdr_sz;
1676
1677	if (readn(fd, buf, rest) != (ssize_t)rest)
1678		return -1;
1679
1680	if (session->header.needs_swap)
1681		event_swap(event, evlist__sample_id_all(session->evlist));
1682
1683out_parse_sample:
1684
1685	if (sample && event->header.type < PERF_RECORD_USER_TYPE_START &&
1686	    perf_evlist__parse_sample(session->evlist, event, sample))
1687		return -1;
1688
1689	*event_ptr = event;
1690
1691	return 0;
1692}
1693
1694int perf_session__peek_events(struct perf_session *session, u64 offset,
1695			      u64 size, peek_events_cb_t cb, void *data)
1696{
1697	u64 max_offset = offset + size;
1698	char buf[PERF_SAMPLE_MAX_SIZE];
1699	union perf_event *event;
1700	int err;
1701
1702	do {
1703		err = perf_session__peek_event(session, offset, buf,
1704					       PERF_SAMPLE_MAX_SIZE, &event,
1705					       NULL);
1706		if (err)
1707			return err;
1708
1709		err = cb(session, event, offset, data);
1710		if (err)
1711			return err;
1712
1713		offset += event->header.size;
1714		if (event->header.type == PERF_RECORD_AUXTRACE)
1715			offset += event->auxtrace.size;
1716
1717	} while (offset < max_offset);
1718
1719	return err;
1720}
1721
1722static s64 perf_session__process_event(struct perf_session *session,
1723				       union perf_event *event, u64 file_offset)
1724{
1725	struct evlist *evlist = session->evlist;
1726	struct perf_tool *tool = session->tool;
1727	int ret;
1728
1729	if (session->header.needs_swap)
1730		event_swap(event, evlist__sample_id_all(evlist));
 
1731
1732	if (event->header.type >= PERF_RECORD_HEADER_MAX)
1733		return -EINVAL;
1734
1735	events_stats__inc(&evlist->stats, event->header.type);
1736
1737	if (event->header.type >= PERF_RECORD_USER_TYPE_START)
1738		return perf_session__process_user_event(session, event, file_offset);
1739
1740	if (tool->ordered_events) {
1741		u64 timestamp = -1ULL;
 
 
 
 
1742
1743		ret = perf_evlist__parse_sample_timestamp(evlist, event, &timestamp);
1744		if (ret && ret != -1)
1745			return ret;
1746
1747		ret = perf_session__queue_event(session, event, timestamp, file_offset);
 
 
1748		if (ret != -ETIME)
1749			return ret;
1750	}
1751
1752	return perf_session__deliver_event(session, event, tool, file_offset);
1753}
1754
1755void perf_event_header__bswap(struct perf_event_header *hdr)
1756{
1757	hdr->type = bswap_32(hdr->type);
1758	hdr->misc = bswap_16(hdr->misc);
1759	hdr->size = bswap_16(hdr->size);
1760}
1761
1762struct thread *perf_session__findnew(struct perf_session *session, pid_t pid)
1763{
1764	return machine__findnew_thread(&session->machines.host, -1, pid);
 
 
1765}
1766
1767/*
1768 * Threads are identified by pid and tid, and the idle task has pid == tid == 0.
1769 * So here a single thread is created for that, but actually there is a separate
1770 * idle task per cpu, so there should be one 'struct thread' per cpu, but there
1771 * is only 1. That causes problems for some tools, requiring workarounds. For
1772 * example get_idle_thread() in builtin-sched.c, or thread_stack__per_cpu().
1773 */
1774int perf_session__register_idle_thread(struct perf_session *session)
1775{
1776	struct thread *thread;
1777	int err = 0;
1778
1779	thread = machine__findnew_thread(&session->machines.host, 0, 0);
1780	if (thread == NULL || thread__set_comm(thread, "swapper", 0)) {
1781		pr_err("problem inserting idle task.\n");
1782		err = -1;
1783	}
1784
1785	if (thread == NULL || thread__set_namespaces(thread, 0, NULL)) {
1786		pr_err("problem inserting idle task.\n");
1787		err = -1;
1788	}
1789
1790	/* machine__findnew_thread() got the thread, so put it */
1791	thread__put(thread);
1792	return err;
1793}
1794
1795static void
1796perf_session__warn_order(const struct perf_session *session)
1797{
1798	const struct ordered_events *oe = &session->ordered_events;
1799	struct evsel *evsel;
1800	bool should_warn = true;
1801
1802	evlist__for_each_entry(session->evlist, evsel) {
1803		if (evsel->core.attr.write_backward)
1804			should_warn = false;
1805	}
1806
1807	if (!should_warn)
1808		return;
1809	if (oe->nr_unordered_events != 0)
1810		ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events);
1811}
1812
1813static void perf_session__warn_about_errors(const struct perf_session *session)
 
1814{
1815	const struct events_stats *stats = &session->evlist->stats;
1816
1817	if (session->tool->lost == perf_event__process_lost &&
1818	    stats->nr_events[PERF_RECORD_LOST] != 0) {
1819		ui__warning("Processed %d events and lost %d chunks!\n\n"
1820			    "Check IO/CPU overload!\n\n",
1821			    stats->nr_events[0],
1822			    stats->nr_events[PERF_RECORD_LOST]);
1823	}
1824
1825	if (session->tool->lost_samples == perf_event__process_lost_samples) {
1826		double drop_rate;
1827
1828		drop_rate = (double)stats->total_lost_samples /
1829			    (double) (stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples);
1830		if (drop_rate > 0.05) {
1831			ui__warning("Processed %" PRIu64 " samples and lost %3.2f%%!\n\n",
1832				    stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples,
1833				    drop_rate * 100.0);
1834		}
1835	}
1836
1837	if (session->tool->aux == perf_event__process_aux &&
1838	    stats->total_aux_lost != 0) {
1839		ui__warning("AUX data lost %" PRIu64 " times out of %u!\n\n",
1840			    stats->total_aux_lost,
1841			    stats->nr_events[PERF_RECORD_AUX]);
1842	}
1843
1844	if (session->tool->aux == perf_event__process_aux &&
1845	    stats->total_aux_partial != 0) {
1846		bool vmm_exclusive = false;
1847
1848		(void)sysfs__read_bool("module/kvm_intel/parameters/vmm_exclusive",
1849		                       &vmm_exclusive);
1850
1851		ui__warning("AUX data had gaps in it %" PRIu64 " times out of %u!\n\n"
1852		            "Are you running a KVM guest in the background?%s\n\n",
1853			    stats->total_aux_partial,
1854			    stats->nr_events[PERF_RECORD_AUX],
1855			    vmm_exclusive ?
1856			    "\nReloading kvm_intel module with vmm_exclusive=0\n"
1857			    "will reduce the gaps to only guest's timeslices." :
1858			    "");
1859	}
1860
1861	if (stats->nr_unknown_events != 0) {
1862		ui__warning("Found %u unknown events!\n\n"
1863			    "Is this an older tool processing a perf.data "
1864			    "file generated by a more recent tool?\n\n"
1865			    "If that is not the case, consider "
1866			    "reporting to linux-kernel@vger.kernel.org.\n\n",
1867			    stats->nr_unknown_events);
1868	}
1869
1870	if (stats->nr_unknown_id != 0) {
1871		ui__warning("%u samples with id not present in the header\n",
1872			    stats->nr_unknown_id);
1873	}
1874
1875	if (stats->nr_invalid_chains != 0) {
1876		ui__warning("Found invalid callchains!\n\n"
1877			    "%u out of %u events were discarded for this reason.\n\n"
1878			    "Consider reporting to linux-kernel@vger.kernel.org.\n\n",
1879			    stats->nr_invalid_chains,
1880			    stats->nr_events[PERF_RECORD_SAMPLE]);
1881	}
1882
1883	if (stats->nr_unprocessable_samples != 0) {
1884		ui__warning("%u unprocessable samples recorded.\n"
1885			    "Do you have a KVM guest running and not using 'perf kvm'?\n",
1886			    stats->nr_unprocessable_samples);
1887	}
1888
1889	perf_session__warn_order(session);
1890
1891	events_stats__auxtrace_error_warn(stats);
1892
1893	if (stats->nr_proc_map_timeout != 0) {
1894		ui__warning("%d map information files for pre-existing threads were\n"
1895			    "not processed, if there are samples for addresses they\n"
1896			    "will not be resolved, you may find out which are these\n"
1897			    "threads by running with -v and redirecting the output\n"
1898			    "to a file.\n"
1899			    "The time limit to process proc map is too short?\n"
1900			    "Increase it by --proc-map-timeout\n",
1901			    stats->nr_proc_map_timeout);
1902	}
1903}
1904
1905static int perf_session__flush_thread_stack(struct thread *thread,
1906					    void *p __maybe_unused)
1907{
1908	return thread_stack__flush(thread);
1909}
1910
1911static int perf_session__flush_thread_stacks(struct perf_session *session)
1912{
1913	return machines__for_each_thread(&session->machines,
1914					 perf_session__flush_thread_stack,
1915					 NULL);
1916}
1917
 
1918volatile int session_done;
1919
1920static int __perf_session__process_decomp_events(struct perf_session *session);
1921
1922static int __perf_session__process_pipe_events(struct perf_session *session)
1923{
1924	struct ordered_events *oe = &session->ordered_events;
1925	struct perf_tool *tool = session->tool;
1926	int fd = perf_data__fd(session->data);
1927	union perf_event *event;
1928	uint32_t size, cur_size = 0;
1929	void *buf = NULL;
1930	s64 skip = 0;
1931	u64 head;
1932	ssize_t err;
1933	void *p;
1934
1935	perf_tool__fill_defaults(tool);
1936
1937	head = 0;
1938	cur_size = sizeof(union perf_event);
1939
1940	buf = malloc(cur_size);
1941	if (!buf)
1942		return -errno;
1943	ordered_events__set_copy_on_queue(oe, true);
1944more:
1945	event = buf;
1946	err = readn(fd, event, sizeof(struct perf_event_header));
1947	if (err <= 0) {
1948		if (err == 0)
1949			goto done;
1950
1951		pr_err("failed to read event header\n");
1952		goto out_err;
1953	}
1954
1955	if (session->header.needs_swap)
1956		perf_event_header__bswap(&event->header);
1957
1958	size = event->header.size;
1959	if (size < sizeof(struct perf_event_header)) {
1960		pr_err("bad event header size\n");
1961		goto out_err;
1962	}
1963
1964	if (size > cur_size) {
1965		void *new = realloc(buf, size);
1966		if (!new) {
1967			pr_err("failed to allocate memory to read event\n");
1968			goto out_err;
1969		}
1970		buf = new;
1971		cur_size = size;
1972		event = buf;
1973	}
1974	p = event;
1975	p += sizeof(struct perf_event_header);
1976
1977	if (size - sizeof(struct perf_event_header)) {
1978		err = readn(fd, p, size - sizeof(struct perf_event_header));
1979		if (err <= 0) {
1980			if (err == 0) {
1981				pr_err("unexpected end of event stream\n");
1982				goto done;
1983			}
1984
1985			pr_err("failed to read event data\n");
1986			goto out_err;
1987		}
1988	}
1989
1990	if ((skip = perf_session__process_event(session, event, head)) < 0) {
1991		pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
1992		       head, event->header.size, event->header.type);
1993		err = -EINVAL;
1994		goto out_err;
 
 
 
 
 
 
 
1995	}
1996
1997	head += size;
1998
1999	if (skip > 0)
2000		head += skip;
2001
2002	err = __perf_session__process_decomp_events(session);
2003	if (err)
2004		goto out_err;
2005
2006	if (!session_done())
2007		goto more;
2008done:
2009	/* do the final flush for ordered samples */
2010	err = ordered_events__flush(oe, OE_FLUSH__FINAL);
2011	if (err)
2012		goto out_err;
2013	err = auxtrace__flush_events(session, tool);
2014	if (err)
2015		goto out_err;
2016	err = perf_session__flush_thread_stacks(session);
2017out_err:
2018	free(buf);
2019	if (!tool->no_warn)
2020		perf_session__warn_about_errors(session);
2021	ordered_events__free(&session->ordered_events);
2022	auxtrace__free_events(session);
2023	return err;
2024}
2025
2026static union perf_event *
2027prefetch_event(char *buf, u64 head, size_t mmap_size,
2028	       bool needs_swap, union perf_event *error)
2029{
2030	union perf_event *event;
2031
2032	/*
2033	 * Ensure we have enough space remaining to read
2034	 * the size of the event in the headers.
2035	 */
2036	if (head + sizeof(event->header) > mmap_size)
2037		return NULL;
2038
2039	event = (union perf_event *)(buf + head);
2040	if (needs_swap)
2041		perf_event_header__bswap(&event->header);
2042
2043	if (head + event->header.size <= mmap_size)
2044		return event;
2045
2046	/* We're not fetching the event so swap back again */
2047	if (needs_swap)
2048		perf_event_header__bswap(&event->header);
2049
2050	pr_debug("%s: head=%#" PRIx64 " event->header_size=%#x, mmap_size=%#zx:"
2051		 " fuzzed or compressed perf.data?\n",__func__, head, event->header.size, mmap_size);
2052
2053	return error;
2054}
2055
2056static union perf_event *
2057fetch_mmaped_event(u64 head, size_t mmap_size, char *buf, bool needs_swap)
2058{
2059	return prefetch_event(buf, head, mmap_size, needs_swap, ERR_PTR(-EINVAL));
2060}
2061
2062static union perf_event *
2063fetch_decomp_event(u64 head, size_t mmap_size, char *buf, bool needs_swap)
2064{
2065	return prefetch_event(buf, head, mmap_size, needs_swap, NULL);
2066}
2067
2068static int __perf_session__process_decomp_events(struct perf_session *session)
2069{
2070	s64 skip;
2071	u64 size, file_pos = 0;
2072	struct decomp *decomp = session->decomp_last;
2073
2074	if (!decomp)
2075		return 0;
2076
2077	while (decomp->head < decomp->size && !session_done()) {
2078		union perf_event *event = fetch_decomp_event(decomp->head, decomp->size, decomp->data,
2079							     session->header.needs_swap);
2080
2081		if (!event)
2082			break;
2083
2084		size = event->header.size;
2085
2086		if (size < sizeof(struct perf_event_header) ||
2087		    (skip = perf_session__process_event(session, event, file_pos)) < 0) {
2088			pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n",
2089				decomp->file_pos + decomp->head, event->header.size, event->header.type);
2090			return -EINVAL;
2091		}
2092
2093		if (skip)
2094			size += skip;
2095
2096		decomp->head += size;
2097	}
2098
2099	return 0;
2100}
2101
2102/*
2103 * On 64bit we can mmap the data file in one go. No need for tiny mmap
2104 * slices. On 32bit we use 32MB.
2105 */
2106#if BITS_PER_LONG == 64
2107#define MMAP_SIZE ULLONG_MAX
2108#define NUM_MMAPS 1
2109#else
2110#define MMAP_SIZE (32 * 1024 * 1024ULL)
2111#define NUM_MMAPS 128
2112#endif
2113
2114struct reader;
2115
2116typedef s64 (*reader_cb_t)(struct perf_session *session,
2117			   union perf_event *event,
2118			   u64 file_offset);
2119
2120struct reader {
2121	int		 fd;
2122	u64		 data_size;
2123	u64		 data_offset;
2124	reader_cb_t	 process;
2125};
2126
2127static int
2128reader__process_events(struct reader *rd, struct perf_session *session,
2129		       struct ui_progress *prog)
2130{
2131	u64 data_size = rd->data_size;
2132	u64 head, page_offset, file_offset, file_pos, size;
2133	int err = 0, mmap_prot, mmap_flags, map_idx = 0;
2134	size_t	mmap_size;
2135	char *buf, *mmaps[NUM_MMAPS];
2136	union perf_event *event;
2137	s64 skip;
2138
2139	page_offset = page_size * (rd->data_offset / page_size);
2140	file_offset = page_offset;
2141	head = rd->data_offset - page_offset;
2142
2143	ui_progress__init_size(prog, data_size, "Processing events...");
 
2144
2145	data_size += rd->data_offset;
 
 
 
2146
2147	mmap_size = MMAP_SIZE;
2148	if (mmap_size > data_size) {
2149		mmap_size = data_size;
2150		session->one_mmap = true;
2151	}
2152
2153	memset(mmaps, 0, sizeof(mmaps));
2154
2155	mmap_prot  = PROT_READ;
2156	mmap_flags = MAP_SHARED;
2157
2158	if (session->header.needs_swap) {
2159		mmap_prot  |= PROT_WRITE;
2160		mmap_flags = MAP_PRIVATE;
2161	}
2162remap:
2163	buf = mmap(NULL, mmap_size, mmap_prot, mmap_flags, rd->fd,
2164		   file_offset);
2165	if (buf == MAP_FAILED) {
2166		pr_err("failed to mmap file\n");
2167		err = -errno;
2168		goto out;
2169	}
2170	mmaps[map_idx] = buf;
2171	map_idx = (map_idx + 1) & (ARRAY_SIZE(mmaps) - 1);
2172	file_pos = file_offset + head;
2173	if (session->one_mmap) {
2174		session->one_mmap_addr = buf;
2175		session->one_mmap_offset = file_offset;
2176	}
2177
2178more:
2179	event = fetch_mmaped_event(head, mmap_size, buf, session->header.needs_swap);
2180	if (IS_ERR(event))
2181		return PTR_ERR(event);
2182
2183	if (!event) {
2184		if (mmaps[map_idx]) {
2185			munmap(mmaps[map_idx], mmap_size);
2186			mmaps[map_idx] = NULL;
2187		}
2188
2189		page_offset = page_size * (head / page_size);
2190		file_offset += page_offset;
2191		head -= page_offset;
2192		goto remap;
2193	}
2194
2195	size = event->header.size;
2196
2197	skip = -EINVAL;
 
 
 
 
 
 
 
 
 
 
2198
2199	if (size < sizeof(struct perf_event_header) ||
2200	    (skip = rd->process(session, event, file_pos)) < 0) {
2201		pr_err("%#" PRIx64 " [%#x]: failed to process type: %d [%s]\n",
2202		       file_offset + head, event->header.size,
2203		       event->header.type, strerror(-skip));
2204		err = skip;
2205		goto out;
2206	}
2207
2208	if (skip)
2209		size += skip;
2210
2211	head += size;
2212	file_pos += size;
2213
2214	err = __perf_session__process_decomp_events(session);
2215	if (err)
2216		goto out;
 
2217
2218	ui_progress__update(prog, size);
2219
2220	if (session_done())
2221		goto out;
2222
2223	if (file_pos < data_size)
2224		goto more;
2225
2226out:
2227	return err;
2228}
2229
2230static s64 process_simple(struct perf_session *session,
2231			  union perf_event *event,
2232			  u64 file_offset)
2233{
2234	return perf_session__process_event(session, event, file_offset);
2235}
2236
2237static int __perf_session__process_events(struct perf_session *session)
2238{
2239	struct reader rd = {
2240		.fd		= perf_data__fd(session->data),
2241		.data_size	= session->header.data_size,
2242		.data_offset	= session->header.data_offset,
2243		.process	= process_simple,
2244	};
2245	struct ordered_events *oe = &session->ordered_events;
2246	struct perf_tool *tool = session->tool;
2247	struct ui_progress prog;
2248	int err;
2249
2250	perf_tool__fill_defaults(tool);
2251
2252	if (rd.data_size == 0)
2253		return -1;
2254
2255	ui_progress__init_size(&prog, rd.data_size, "Processing events...");
2256
2257	err = reader__process_events(&rd, session, &prog);
2258	if (err)
2259		goto out_err;
2260	/* do the final flush for ordered samples */
2261	err = ordered_events__flush(oe, OE_FLUSH__FINAL);
2262	if (err)
2263		goto out_err;
2264	err = auxtrace__flush_events(session, tool);
2265	if (err)
2266		goto out_err;
2267	err = perf_session__flush_thread_stacks(session);
2268out_err:
2269	ui_progress__finish();
2270	if (!tool->no_warn)
2271		perf_session__warn_about_errors(session);
2272	/*
2273	 * We may switching perf.data output, make ordered_events
2274	 * reusable.
2275	 */
2276	ordered_events__reinit(&session->ordered_events);
2277	auxtrace__free_events(session);
2278	session->one_mmap = false;
2279	return err;
2280}
2281
2282int perf_session__process_events(struct perf_session *session)
 
2283{
2284	if (perf_session__register_idle_thread(session) < 0)
 
 
2285		return -ENOMEM;
2286
2287	if (perf_data__is_pipe(session->data))
2288		return __perf_session__process_pipe_events(session);
 
 
 
 
 
2289
2290	return __perf_session__process_events(session);
2291}
2292
2293bool perf_session__has_traces(struct perf_session *session, const char *msg)
2294{
2295	struct evsel *evsel;
2296
2297	evlist__for_each_entry(session->evlist, evsel) {
2298		if (evsel->core.attr.type == PERF_TYPE_TRACEPOINT)
2299			return true;
2300	}
2301
2302	pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg);
2303	return false;
2304}
2305
2306int map__set_kallsyms_ref_reloc_sym(struct map *map, const char *symbol_name, u64 addr)
 
 
2307{
2308	char *bracket;
 
2309	struct ref_reloc_sym *ref;
2310	struct kmap *kmap;
2311
2312	ref = zalloc(sizeof(struct ref_reloc_sym));
2313	if (ref == NULL)
2314		return -ENOMEM;
2315
2316	ref->name = strdup(symbol_name);
2317	if (ref->name == NULL) {
2318		free(ref);
2319		return -ENOMEM;
2320	}
2321
2322	bracket = strchr(ref->name, ']');
2323	if (bracket)
2324		*bracket = '\0';
2325
2326	ref->addr = addr;
2327
2328	kmap = map__kmap(map);
2329	if (kmap)
2330		kmap->ref_reloc_sym = ref;
 
2331
2332	return 0;
2333}
2334
2335size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp)
2336{
2337	return machines__fprintf_dsos(&session->machines, fp);
 
 
2338}
2339
2340size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp,
2341					  bool (skip)(struct dso *dso, int parm), int parm)
2342{
2343	return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm);
 
2344}
2345
2346size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp)
2347{
2348	size_t ret;
2349	const char *msg = "";
2350
2351	if (perf_header__has_feat(&session->header, HEADER_AUXTRACE))
2352		msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)";
2353
2354	ret = fprintf(fp, "\nAggregated stats:%s\n", msg);
 
 
 
2355
2356	ret += events_stats__fprintf(&session->evlist->stats, fp);
2357	return ret;
2358}
2359
2360size_t perf_session__fprintf(struct perf_session *session, FILE *fp)
2361{
2362	/*
2363	 * FIXME: Here we have to actually print all the machines in this
2364	 * session, not just the host...
2365	 */
2366	return machine__fprintf(&session->machines.host, fp);
2367}
2368
2369struct evsel *perf_session__find_first_evtype(struct perf_session *session,
2370					      unsigned int type)
2371{
2372	struct evsel *pos;
2373
2374	evlist__for_each_entry(session->evlist, pos) {
2375		if (pos->core.attr.type == type)
2376			return pos;
2377	}
2378	return NULL;
2379}
2380
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2381int perf_session__cpu_bitmap(struct perf_session *session,
2382			     const char *cpu_list, unsigned long *cpu_bitmap)
2383{
2384	int i, err = -1;
2385	struct perf_cpu_map *map;
2386	int nr_cpus = min(session->header.env.nr_cpus_online, MAX_NR_CPUS);
2387
2388	for (i = 0; i < PERF_TYPE_MAX; ++i) {
2389		struct evsel *evsel;
2390
2391		evsel = perf_session__find_first_evtype(session, i);
2392		if (!evsel)
2393			continue;
2394
2395		if (!(evsel->core.attr.sample_type & PERF_SAMPLE_CPU)) {
2396			pr_err("File does not contain CPU events. "
2397			       "Remove -C option to proceed.\n");
2398			return -1;
2399		}
2400	}
2401
2402	map = perf_cpu_map__new(cpu_list);
2403	if (map == NULL) {
2404		pr_err("Invalid cpu_list\n");
2405		return -1;
2406	}
2407
2408	for (i = 0; i < map->nr; i++) {
2409		int cpu = map->map[i];
2410
2411		if (cpu >= nr_cpus) {
2412			pr_err("Requested CPU %d too large. "
2413			       "Consider raising MAX_NR_CPUS\n", cpu);
2414			goto out_delete_map;
2415		}
2416
2417		set_bit(cpu, cpu_bitmap);
2418	}
2419
2420	err = 0;
2421
2422out_delete_map:
2423	perf_cpu_map__put(map);
2424	return err;
2425}
2426
2427void perf_session__fprintf_info(struct perf_session *session, FILE *fp,
2428				bool full)
2429{
2430	if (session == NULL || fp == NULL)
2431		return;
2432
2433	fprintf(fp, "# ========\n");
2434	perf_header__fprintf_info(session, fp, full);
2435	fprintf(fp, "# ========\n#\n");
2436}
2437
2438int perf_event__process_id_index(struct perf_session *session,
2439				 union perf_event *event)
2440{
2441	struct evlist *evlist = session->evlist;
2442	struct perf_record_id_index *ie = &event->id_index;
2443	size_t i, nr, max_nr;
2444
2445	max_nr = (ie->header.size - sizeof(struct perf_record_id_index)) /
2446		 sizeof(struct id_index_entry);
2447	nr = ie->nr;
2448	if (nr > max_nr)
2449		return -EINVAL;
2450
2451	if (dump_trace)
2452		fprintf(stdout, " nr: %zu\n", nr);
2453
2454	for (i = 0; i < nr; i++) {
2455		struct id_index_entry *e = &ie->entries[i];
2456		struct perf_sample_id *sid;
2457
2458		if (dump_trace) {
2459			fprintf(stdout,	" ... id: %"PRI_lu64, e->id);
2460			fprintf(stdout,	"  idx: %"PRI_lu64, e->idx);
2461			fprintf(stdout,	"  cpu: %"PRI_ld64, e->cpu);
2462			fprintf(stdout,	"  tid: %"PRI_ld64"\n", e->tid);
2463		}
2464
2465		sid = perf_evlist__id2sid(evlist, e->id);
2466		if (!sid)
2467			return -ENOENT;
2468		sid->idx = e->idx;
2469		sid->cpu = e->cpu;
2470		sid->tid = e->tid;
2471	}
2472	return 0;
2473}