1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * auxtrace.c: AUX area trace support
   4 * Copyright (c) 2013-2015, Intel Corporation.
   5 */
   6
   7#include <inttypes.h>
   8#include <sys/types.h>
   9#include <sys/mman.h>
  10#include <stdbool.h>
  11#include <string.h>
  12#include <limits.h>
  13#include <errno.h>
  14
  15#include <linux/kernel.h>
  16#include <linux/perf_event.h>
  17#include <linux/types.h>
  18#include <linux/bitops.h>
  19#include <linux/log2.h>
  20#include <linux/string.h>
  21#include <linux/time64.h>
  22
  23#include <sys/param.h>
  24#include <stdlib.h>
  25#include <stdio.h>
  26#include <linux/list.h>
  27#include <linux/zalloc.h>
  28
  29#include "config.h"
  30#include "evlist.h"
  31#include "dso.h"
  32#include "map.h"
  33#include "pmu.h"
  34#include "evsel.h"
  35#include "evsel_config.h"
  36#include "symbol.h"
  37#include "util/perf_api_probe.h"
  38#include "util/synthetic-events.h"
  39#include "thread_map.h"
  40#include "asm/bug.h"
  41#include "auxtrace.h"
  42
  43#include <linux/hash.h>
  44
  45#include "event.h"
  46#include "record.h"
  47#include "session.h"
  48#include "debug.h"
  49#include <subcmd/parse-options.h>
  50
  51#include "cs-etm.h"
  52#include "intel-pt.h"
  53#include "intel-bts.h"
  54#include "arm-spe.h"
  55#include "hisi-ptt.h"
  56#include "s390-cpumsf.h"
  57#include "util/mmap.h"
  58
  59#include <linux/ctype.h>
  60#include "symbol/kallsyms.h"
  61#include <internal/lib.h>
  62#include "util/sample.h"
  63
  64/*
  65 * Make a group from 'leader' to 'last', requiring that the events were not
  66 * already grouped to a different leader.
  67 */
  68static int evlist__regroup(struct evlist *evlist, struct evsel *leader, struct evsel *last)
  69{
  70	struct evsel *evsel;
  71	bool grp;
  72
  73	if (!evsel__is_group_leader(leader))
  74		return -EINVAL;
  75
  76	grp = false;
  77	evlist__for_each_entry(evlist, evsel) {
  78		if (grp) {
  79			if (!(evsel__leader(evsel) == leader ||
  80			     (evsel__leader(evsel) == evsel &&
  81			      evsel->core.nr_members <= 1)))
  82				return -EINVAL;
  83		} else if (evsel == leader) {
  84			grp = true;
  85		}
  86		if (evsel == last)
  87			break;
  88	}
  89
  90	grp = false;
  91	evlist__for_each_entry(evlist, evsel) {
  92		if (grp) {
  93			if (!evsel__has_leader(evsel, leader)) {
  94				evsel__set_leader(evsel, leader);
  95				if (leader->core.nr_members < 1)
  96					leader->core.nr_members = 1;
  97				leader->core.nr_members += 1;
  98			}
  99		} else if (evsel == leader) {
 100			grp = true;
 101		}
 102		if (evsel == last)
 103			break;
 104	}
 105
 106	return 0;
 107}
 108
 109static bool auxtrace__dont_decode(struct perf_session *session)
 110{
 111	return !session->itrace_synth_opts ||
 112	       session->itrace_synth_opts->dont_decode;
 113}
 114
 115int auxtrace_mmap__mmap(struct auxtrace_mmap *mm,
 116			struct auxtrace_mmap_params *mp,
 117			void *userpg, int fd)
 118{
 119	struct perf_event_mmap_page *pc = userpg;
 120
 121	WARN_ONCE(mm->base, "Uninitialized auxtrace_mmap\n");
 122
 123	mm->userpg = userpg;
 124	mm->mask = mp->mask;
 125	mm->len = mp->len;
 126	mm->prev = 0;
 127	mm->idx = mp->idx;
 128	mm->tid = mp->tid;
 129	mm->cpu = mp->cpu.cpu;
 130
 131	if (!mp->len || !mp->mmap_needed) {
 132		mm->base = NULL;
 133		return 0;
 134	}
 135
 136	pc->aux_offset = mp->offset;
 137	pc->aux_size = mp->len;
 138
 139	mm->base = mmap(NULL, mp->len, mp->prot, MAP_SHARED, fd, mp->offset);
 140	if (mm->base == MAP_FAILED) {
 141		pr_debug2("failed to mmap AUX area\n");
 142		mm->base = NULL;
 143		return -1;
 144	}
 145
 146	return 0;
 147}
 148
 149void auxtrace_mmap__munmap(struct auxtrace_mmap *mm)
 150{
 151	if (mm->base) {
 152		munmap(mm->base, mm->len);
 153		mm->base = NULL;
 154	}
 155}
 156
 157void auxtrace_mmap_params__init(struct auxtrace_mmap_params *mp,
 158				off_t auxtrace_offset,
 159				unsigned int auxtrace_pages,
 160				bool auxtrace_overwrite)
 161{
 162	if (auxtrace_pages) {
 163		mp->offset = auxtrace_offset;
 164		mp->len = auxtrace_pages * (size_t)page_size;
 165		mp->mask = is_power_of_2(mp->len) ? mp->len - 1 : 0;
 166		mp->prot = PROT_READ | (auxtrace_overwrite ? 0 : PROT_WRITE);
 167		pr_debug2("AUX area mmap length %zu\n", mp->len);
 168	} else {
 169		mp->len = 0;
 170	}
 171}
 172
 173void auxtrace_mmap_params__set_idx(struct auxtrace_mmap_params *mp,
 174				   struct evlist *evlist,
 175				   struct evsel *evsel, int idx)
 176{
 177	bool per_cpu = !perf_cpu_map__has_any_cpu_or_is_empty(evlist->core.user_requested_cpus);
 178
 179	mp->mmap_needed = evsel->needs_auxtrace_mmap;
 180
 181	if (!mp->mmap_needed)
 182		return;
 183
 184	mp->idx = idx;
 185
 186	if (per_cpu) {
 187		mp->cpu = perf_cpu_map__cpu(evlist->core.all_cpus, idx);
 188		if (evlist->core.threads)
 189			mp->tid = perf_thread_map__pid(evlist->core.threads, 0);
 190		else
 191			mp->tid = -1;
 192	} else {
 193		mp->cpu.cpu = -1;
 194		mp->tid = perf_thread_map__pid(evlist->core.threads, idx);
 195	}
 196}
 197
 198#define AUXTRACE_INIT_NR_QUEUES	32
 199
 200static struct auxtrace_queue *auxtrace_alloc_queue_array(unsigned int nr_queues)
 201{
 202	struct auxtrace_queue *queue_array;
 203	unsigned int max_nr_queues, i;
 204
 205	max_nr_queues = UINT_MAX / sizeof(struct auxtrace_queue);
 206	if (nr_queues > max_nr_queues)
 207		return NULL;
 208
 209	queue_array = calloc(nr_queues, sizeof(struct auxtrace_queue));
 210	if (!queue_array)
 211		return NULL;
 212
 213	for (i = 0; i < nr_queues; i++) {
 214		INIT_LIST_HEAD(&queue_array[i].head);
 215		queue_array[i].priv = NULL;
 216	}
 217
 218	return queue_array;
 219}
 220
 221int auxtrace_queues__init(struct auxtrace_queues *queues)
 222{
 223	queues->nr_queues = AUXTRACE_INIT_NR_QUEUES;
 224	queues->queue_array = auxtrace_alloc_queue_array(queues->nr_queues);
 225	if (!queues->queue_array)
 226		return -ENOMEM;
 227	return 0;
 228}
 229
 230static int auxtrace_queues__grow(struct auxtrace_queues *queues,
 231				 unsigned int new_nr_queues)
 232{
 233	unsigned int nr_queues = queues->nr_queues;
 234	struct auxtrace_queue *queue_array;
 235	unsigned int i;
 236
 237	if (!nr_queues)
 238		nr_queues = AUXTRACE_INIT_NR_QUEUES;
 239
 240	while (nr_queues && nr_queues < new_nr_queues)
 241		nr_queues <<= 1;
 242
 243	if (nr_queues < queues->nr_queues || nr_queues < new_nr_queues)
 244		return -EINVAL;
 245
 246	queue_array = auxtrace_alloc_queue_array(nr_queues);
 247	if (!queue_array)
 248		return -ENOMEM;
 249
 250	for (i = 0; i < queues->nr_queues; i++) {
 251		list_splice_tail(&queues->queue_array[i].head,
 252				 &queue_array[i].head);
 253		queue_array[i].tid = queues->queue_array[i].tid;
 254		queue_array[i].cpu = queues->queue_array[i].cpu;
 255		queue_array[i].set = queues->queue_array[i].set;
 256		queue_array[i].priv = queues->queue_array[i].priv;
 257	}
 258
 259	queues->nr_queues = nr_queues;
 260	queues->queue_array = queue_array;
 261
 262	return 0;
 263}
 264
 265static void *auxtrace_copy_data(u64 size, struct perf_session *session)
 266{
 267	int fd = perf_data__fd(session->data);
 268	void *p;
 269	ssize_t ret;
 270
 271	if (size > SSIZE_MAX)
 272		return NULL;
 273
 274	p = malloc(size);
 275	if (!p)
 276		return NULL;
 277
 278	ret = readn(fd, p, size);
 279	if (ret != (ssize_t)size) {
 280		free(p);
 281		return NULL;
 282	}
 283
 284	return p;
 285}
 286
 287static int auxtrace_queues__queue_buffer(struct auxtrace_queues *queues,
 288					 unsigned int idx,
 289					 struct auxtrace_buffer *buffer)
 290{
 291	struct auxtrace_queue *queue;
 292	int err;
 293
 294	if (idx >= queues->nr_queues) {
 295		err = auxtrace_queues__grow(queues, idx + 1);
 296		if (err)
 297			return err;
 298	}
 299
 300	queue = &queues->queue_array[idx];
 301
 302	if (!queue->set) {
 303		queue->set = true;
 304		queue->tid = buffer->tid;
 305		queue->cpu = buffer->cpu.cpu;
 306	}
 307
 308	buffer->buffer_nr = queues->next_buffer_nr++;
 309
 310	list_add_tail(&buffer->list, &queue->head);
 311
 312	queues->new_data = true;
 313	queues->populated = true;
 314
 315	return 0;
 316}
 317
 318/* Limit buffers to 32MiB on 32-bit */
 319#define BUFFER_LIMIT_FOR_32_BIT (32 * 1024 * 1024)
 320
 321static int auxtrace_queues__split_buffer(struct auxtrace_queues *queues,
 322					 unsigned int idx,
 323					 struct auxtrace_buffer *buffer)
 324{
 325	u64 sz = buffer->size;
 326	bool consecutive = false;
 327	struct auxtrace_buffer *b;
 328	int err;
 329
 330	while (sz > BUFFER_LIMIT_FOR_32_BIT) {
 331		b = memdup(buffer, sizeof(struct auxtrace_buffer));
 332		if (!b)
 333			return -ENOMEM;
 334		b->size = BUFFER_LIMIT_FOR_32_BIT;
 335		b->consecutive = consecutive;
 336		err = auxtrace_queues__queue_buffer(queues, idx, b);
 337		if (err) {
 338			auxtrace_buffer__free(b);
 339			return err;
 340		}
 341		buffer->data_offset += BUFFER_LIMIT_FOR_32_BIT;
 342		sz -= BUFFER_LIMIT_FOR_32_BIT;
 343		consecutive = true;
 344	}
 345
 346	buffer->size = sz;
 347	buffer->consecutive = consecutive;
 348
 349	return 0;
 350}
 351
 352static bool filter_cpu(struct perf_session *session, struct perf_cpu cpu)
 353{
 354	unsigned long *cpu_bitmap = session->itrace_synth_opts->cpu_bitmap;
 355
 356	return cpu_bitmap && cpu.cpu != -1 && !test_bit(cpu.cpu, cpu_bitmap);
 357}
 358
 359static int auxtrace_queues__add_buffer(struct auxtrace_queues *queues,
 360				       struct perf_session *session,
 361				       unsigned int idx,
 362				       struct auxtrace_buffer *buffer,
 363				       struct auxtrace_buffer **buffer_ptr)
 364{
 365	int err = -ENOMEM;
 366
 367	if (filter_cpu(session, buffer->cpu))
 368		return 0;
 369
 370	buffer = memdup(buffer, sizeof(*buffer));
 371	if (!buffer)
 372		return -ENOMEM;
 373
 374	if (session->one_mmap) {
 375		buffer->data = buffer->data_offset - session->one_mmap_offset +
 376			       session->one_mmap_addr;
 377	} else if (perf_data__is_pipe(session->data)) {
 378		buffer->data = auxtrace_copy_data(buffer->size, session);
 379		if (!buffer->data)
 380			goto out_free;
 381		buffer->data_needs_freeing = true;
 382	} else if (BITS_PER_LONG == 32 &&
 383		   buffer->size > BUFFER_LIMIT_FOR_32_BIT) {
 384		err = auxtrace_queues__split_buffer(queues, idx, buffer);
 385		if (err)
 386			goto out_free;
 387	}
 388
 389	err = auxtrace_queues__queue_buffer(queues, idx, buffer);
 390	if (err)
 391		goto out_free;
 392
 393	/* FIXME: Doesn't work for split buffer */
 394	if (buffer_ptr)
 395		*buffer_ptr = buffer;
 396
 397	return 0;
 398
 399out_free:
 400	auxtrace_buffer__free(buffer);
 401	return err;
 402}
 403
 404int auxtrace_queues__add_event(struct auxtrace_queues *queues,
 405			       struct perf_session *session,
 406			       union perf_event *event, off_t data_offset,
 407			       struct auxtrace_buffer **buffer_ptr)
 408{
 409	struct auxtrace_buffer buffer = {
 410		.pid = -1,
 411		.tid = event->auxtrace.tid,
 412		.cpu = { event->auxtrace.cpu },
 413		.data_offset = data_offset,
 414		.offset = event->auxtrace.offset,
 415		.reference = event->auxtrace.reference,
 416		.size = event->auxtrace.size,
 417	};
 418	unsigned int idx = event->auxtrace.idx;
 419
 420	return auxtrace_queues__add_buffer(queues, session, idx, &buffer,
 421					   buffer_ptr);
 422}
 423
 424static int auxtrace_queues__add_indexed_event(struct auxtrace_queues *queues,
 425					      struct perf_session *session,
 426					      off_t file_offset, size_t sz)
 427{
 428	union perf_event *event;
 429	int err;
 430	char buf[PERF_SAMPLE_MAX_SIZE];
 431
 432	err = perf_session__peek_event(session, file_offset, buf,
 433				       PERF_SAMPLE_MAX_SIZE, &event, NULL);
 434	if (err)
 435		return err;
 436
 437	if (event->header.type == PERF_RECORD_AUXTRACE) {
 438		if (event->header.size < sizeof(struct perf_record_auxtrace) ||
 439		    event->header.size != sz) {
 440			err = -EINVAL;
 441			goto out;
 442		}
 443		file_offset += event->header.size;
 444		err = auxtrace_queues__add_event(queues, session, event,
 445						 file_offset, NULL);
 446	}
 447out:
 448	return err;
 449}
 450
 451void auxtrace_queues__free(struct auxtrace_queues *queues)
 452{
 453	unsigned int i;
 454
 455	for (i = 0; i < queues->nr_queues; i++) {
 456		while (!list_empty(&queues->queue_array[i].head)) {
 457			struct auxtrace_buffer *buffer;
 458
 459			buffer = list_entry(queues->queue_array[i].head.next,
 460					    struct auxtrace_buffer, list);
 461			list_del_init(&buffer->list);
 462			auxtrace_buffer__free(buffer);
 463		}
 464	}
 465
 466	zfree(&queues->queue_array);
 467	queues->nr_queues = 0;
 468}
 469
 470static void auxtrace_heapify(struct auxtrace_heap_item *heap_array,
 471			     unsigned int pos, unsigned int queue_nr,
 472			     u64 ordinal)
 473{
 474	unsigned int parent;
 475
 476	while (pos) {
 477		parent = (pos - 1) >> 1;
 478		if (heap_array[parent].ordinal <= ordinal)
 479			break;
 480		heap_array[pos] = heap_array[parent];
 481		pos = parent;
 482	}
 483	heap_array[pos].queue_nr = queue_nr;
 484	heap_array[pos].ordinal = ordinal;
 485}
 486
 487int auxtrace_heap__add(struct auxtrace_heap *heap, unsigned int queue_nr,
 488		       u64 ordinal)
 489{
 490	struct auxtrace_heap_item *heap_array;
 491
 492	if (queue_nr >= heap->heap_sz) {
 493		unsigned int heap_sz = AUXTRACE_INIT_NR_QUEUES;
 494
 495		while (heap_sz <= queue_nr)
 496			heap_sz <<= 1;
 497		heap_array = realloc(heap->heap_array,
 498				     heap_sz * sizeof(struct auxtrace_heap_item));
 499		if (!heap_array)
 500			return -ENOMEM;
 501		heap->heap_array = heap_array;
 502		heap->heap_sz = heap_sz;
 503	}
 504
 505	auxtrace_heapify(heap->heap_array, heap->heap_cnt++, queue_nr, ordinal);
 506
 507	return 0;
 508}
 509
 510void auxtrace_heap__free(struct auxtrace_heap *heap)
 511{
 512	zfree(&heap->heap_array);
 513	heap->heap_cnt = 0;
 514	heap->heap_sz = 0;
 515}
 516
 517void auxtrace_heap__pop(struct auxtrace_heap *heap)
 518{
 519	unsigned int pos, last, heap_cnt = heap->heap_cnt;
 520	struct auxtrace_heap_item *heap_array;
 521
 522	if (!heap_cnt)
 523		return;
 524
 525	heap->heap_cnt -= 1;
 526
 527	heap_array = heap->heap_array;
 528
 529	pos = 0;
 530	while (1) {
 531		unsigned int left, right;
 532
 533		left = (pos << 1) + 1;
 534		if (left >= heap_cnt)
 535			break;
 536		right = left + 1;
 537		if (right >= heap_cnt) {
 538			heap_array[pos] = heap_array[left];
 539			return;
 540		}
 541		if (heap_array[left].ordinal < heap_array[right].ordinal) {
 542			heap_array[pos] = heap_array[left];
 543			pos = left;
 544		} else {
 545			heap_array[pos] = heap_array[right];
 546			pos = right;
 547		}
 548	}
 549
 550	last = heap_cnt - 1;
 551	auxtrace_heapify(heap_array, pos, heap_array[last].queue_nr,
 552			 heap_array[last].ordinal);
 553}
 554
 555size_t auxtrace_record__info_priv_size(struct auxtrace_record *itr,
 556				       struct evlist *evlist)
 557{
 558	if (itr)
 559		return itr->info_priv_size(itr, evlist);
 560	return 0;
 561}
 562
 563static int auxtrace_not_supported(void)
 564{
 565	pr_err("AUX area tracing is not supported on this architecture\n");
 566	return -EINVAL;
 567}
 568
 569int auxtrace_record__info_fill(struct auxtrace_record *itr,
 570			       struct perf_session *session,
 571			       struct perf_record_auxtrace_info *auxtrace_info,
 572			       size_t priv_size)
 573{
 574	if (itr)
 575		return itr->info_fill(itr, session, auxtrace_info, priv_size);
 576	return auxtrace_not_supported();
 577}
 578
 579void auxtrace_record__free(struct auxtrace_record *itr)
 580{
 581	if (itr)
 582		itr->free(itr);
 583}
 584
 585int auxtrace_record__snapshot_start(struct auxtrace_record *itr)
 586{
 587	if (itr && itr->snapshot_start)
 588		return itr->snapshot_start(itr);
 589	return 0;
 590}
 591
 592int auxtrace_record__snapshot_finish(struct auxtrace_record *itr, bool on_exit)
 593{
 594	if (!on_exit && itr && itr->snapshot_finish)
 595		return itr->snapshot_finish(itr);
 596	return 0;
 597}
 598
 599int auxtrace_record__find_snapshot(struct auxtrace_record *itr, int idx,
 600				   struct auxtrace_mmap *mm,
 601				   unsigned char *data, u64 *head, u64 *old)
 602{
 603	if (itr && itr->find_snapshot)
 604		return itr->find_snapshot(itr, idx, mm, data, head, old);
 605	return 0;
 606}
 607
 608int auxtrace_record__options(struct auxtrace_record *itr,
 609			     struct evlist *evlist,
 610			     struct record_opts *opts)
 611{
 612	if (itr) {
 613		itr->evlist = evlist;
 614		return itr->recording_options(itr, evlist, opts);
 615	}
 616	return 0;
 617}
 618
 619u64 auxtrace_record__reference(struct auxtrace_record *itr)
 620{
 621	if (itr)
 622		return itr->reference(itr);
 623	return 0;
 624}
 625
 626int auxtrace_parse_snapshot_options(struct auxtrace_record *itr,
 627				    struct record_opts *opts, const char *str)
 628{
 629	if (!str)
 630		return 0;
 631
 632	/* PMU-agnostic options */
 633	switch (*str) {
 634	case 'e':
 635		opts->auxtrace_snapshot_on_exit = true;
 636		str++;
 637		break;
 638	default:
 639		break;
 640	}
 641
 642	if (itr && itr->parse_snapshot_options)
 643		return itr->parse_snapshot_options(itr, opts, str);
 644
 645	pr_err("No AUX area tracing to snapshot\n");
 646	return -EINVAL;
 647}
 648
 649static int evlist__enable_event_idx(struct evlist *evlist, struct evsel *evsel, int idx)
 650{
 651	bool per_cpu_mmaps = !perf_cpu_map__has_any_cpu_or_is_empty(evlist->core.user_requested_cpus);
 652
 653	if (per_cpu_mmaps) {
 654		struct perf_cpu evlist_cpu = perf_cpu_map__cpu(evlist->core.all_cpus, idx);
 655		int cpu_map_idx = perf_cpu_map__idx(evsel->core.cpus, evlist_cpu);
 656
 657		if (cpu_map_idx == -1)
 658			return -EINVAL;
 659		return perf_evsel__enable_cpu(&evsel->core, cpu_map_idx);
 660	}
 661
 662	return perf_evsel__enable_thread(&evsel->core, idx);
 663}
 664
 665int auxtrace_record__read_finish(struct auxtrace_record *itr, int idx)
 666{
 667	struct evsel *evsel;
 668
 669	if (!itr->evlist || !itr->pmu)
 670		return -EINVAL;
 671
 672	evlist__for_each_entry(itr->evlist, evsel) {
 673		if (evsel->core.attr.type == itr->pmu->type) {
 674			if (evsel->disabled)
 675				return 0;
 676			return evlist__enable_event_idx(itr->evlist, evsel, idx);
 677		}
 678	}
 679	return -EINVAL;
 680}
 681
 682/*
 683 * Event record size is 16-bit which results in a maximum size of about 64KiB.
 684 * Allow about 4KiB for the rest of the sample record, to give a maximum
 685 * AUX area sample size of 60KiB.
 686 */
 687#define MAX_AUX_SAMPLE_SIZE (60 * 1024)
 688
 689/* Arbitrary default size if no other default provided */
 690#define DEFAULT_AUX_SAMPLE_SIZE (4 * 1024)
 691
 692static int auxtrace_validate_aux_sample_size(struct evlist *evlist,
 693					     struct record_opts *opts)
 694{
 695	struct evsel *evsel;
 696	bool has_aux_leader = false;
 697	u32 sz;
 698
 699	evlist__for_each_entry(evlist, evsel) {
 700		sz = evsel->core.attr.aux_sample_size;
 701		if (evsel__is_group_leader(evsel)) {
 702			has_aux_leader = evsel__is_aux_event(evsel);
 703			if (sz) {
 704				if (has_aux_leader)
 705					pr_err("Cannot add AUX area sampling to an AUX area event\n");
 706				else
 707					pr_err("Cannot add AUX area sampling to a group leader\n");
 708				return -EINVAL;
 709			}
 710		}
 711		if (sz > MAX_AUX_SAMPLE_SIZE) {
 712			pr_err("AUX area sample size %u too big, max. %d\n",
 713			       sz, MAX_AUX_SAMPLE_SIZE);
 714			return -EINVAL;
 715		}
 716		if (sz) {
 717			if (!has_aux_leader) {
 718				pr_err("Cannot add AUX area sampling because group leader is not an AUX area event\n");
 719				return -EINVAL;
 720			}
 721			evsel__set_sample_bit(evsel, AUX);
 722			opts->auxtrace_sample_mode = true;
 723		} else {
 724			evsel__reset_sample_bit(evsel, AUX);
 725		}
 726	}
 727
 728	if (!opts->auxtrace_sample_mode) {
 729		pr_err("AUX area sampling requires an AUX area event group leader plus other events to which to add samples\n");
 730		return -EINVAL;
 731	}
 732
 733	if (!perf_can_aux_sample()) {
 734		pr_err("AUX area sampling is not supported by kernel\n");
 735		return -EINVAL;
 736	}
 737
 738	return 0;
 739}
 740
 741int auxtrace_parse_sample_options(struct auxtrace_record *itr,
 742				  struct evlist *evlist,
 743				  struct record_opts *opts, const char *str)
 744{
 745	struct evsel_config_term *term;
 746	struct evsel *aux_evsel;
 747	bool has_aux_sample_size = false;
 748	bool has_aux_leader = false;
 749	struct evsel *evsel;
 750	char *endptr;
 751	unsigned long sz;
 752
 753	if (!str)
 754		goto no_opt;
 755
 756	if (!itr) {
 757		pr_err("No AUX area event to sample\n");
 758		return -EINVAL;
 759	}
 760
 761	sz = strtoul(str, &endptr, 0);
 762	if (*endptr || sz > UINT_MAX) {
 763		pr_err("Bad AUX area sampling option: '%s'\n", str);
 764		return -EINVAL;
 765	}
 766
 767	if (!sz)
 768		sz = itr->default_aux_sample_size;
 769
 770	if (!sz)
 771		sz = DEFAULT_AUX_SAMPLE_SIZE;
 772
 773	/* Set aux_sample_size based on --aux-sample option */
 774	evlist__for_each_entry(evlist, evsel) {
 775		if (evsel__is_group_leader(evsel)) {
 776			has_aux_leader = evsel__is_aux_event(evsel);
 777		} else if (has_aux_leader) {
 778			evsel->core.attr.aux_sample_size = sz;
 779		}
 780	}
 781no_opt:
 782	aux_evsel = NULL;
 783	/* Override with aux_sample_size from config term */
 784	evlist__for_each_entry(evlist, evsel) {
 785		if (evsel__is_aux_event(evsel))
 786			aux_evsel = evsel;
 787		term = evsel__get_config_term(evsel, AUX_SAMPLE_SIZE);
 788		if (term) {
 789			has_aux_sample_size = true;
 790			evsel->core.attr.aux_sample_size = term->val.aux_sample_size;
 791			/* If possible, group with the AUX event */
 792			if (aux_evsel && evsel->core.attr.aux_sample_size)
 793				evlist__regroup(evlist, aux_evsel, evsel);
 794		}
 795	}
 796
 797	if (!str && !has_aux_sample_size)
 798		return 0;
 799
 800	if (!itr) {
 801		pr_err("No AUX area event to sample\n");
 802		return -EINVAL;
 803	}
 804
 805	return auxtrace_validate_aux_sample_size(evlist, opts);
 806}
 807
 808void auxtrace_regroup_aux_output(struct evlist *evlist)
 809{
 810	struct evsel *evsel, *aux_evsel = NULL;
 811	struct evsel_config_term *term;
 812
 813	evlist__for_each_entry(evlist, evsel) {
 814		if (evsel__is_aux_event(evsel))
 815			aux_evsel = evsel;
 816		term = evsel__get_config_term(evsel, AUX_OUTPUT);
 817		/* If possible, group with the AUX event */
 818		if (term && aux_evsel)
 819			evlist__regroup(evlist, aux_evsel, evsel);
 820	}
 821}
 822
 823struct auxtrace_record *__weak
 824auxtrace_record__init(struct evlist *evlist __maybe_unused, int *err)
 825{
 826	*err = 0;
 827	return NULL;
 828}
 829
 830static int auxtrace_index__alloc(struct list_head *head)
 831{
 832	struct auxtrace_index *auxtrace_index;
 833
 834	auxtrace_index = malloc(sizeof(struct auxtrace_index));
 835	if (!auxtrace_index)
 836		return -ENOMEM;
 837
 838	auxtrace_index->nr = 0;
 839	INIT_LIST_HEAD(&auxtrace_index->list);
 840
 841	list_add_tail(&auxtrace_index->list, head);
 842
 843	return 0;
 844}
 845
 846void auxtrace_index__free(struct list_head *head)
 847{
 848	struct auxtrace_index *auxtrace_index, *n;
 849
 850	list_for_each_entry_safe(auxtrace_index, n, head, list) {
 851		list_del_init(&auxtrace_index->list);
 852		free(auxtrace_index);
 853	}
 854}
 855
 856static struct auxtrace_index *auxtrace_index__last(struct list_head *head)
 857{
 858	struct auxtrace_index *auxtrace_index;
 859	int err;
 860
 861	if (list_empty(head)) {
 862		err = auxtrace_index__alloc(head);
 863		if (err)
 864			return NULL;
 865	}
 866
 867	auxtrace_index = list_entry(head->prev, struct auxtrace_index, list);
 868
 869	if (auxtrace_index->nr >= PERF_AUXTRACE_INDEX_ENTRY_COUNT) {
 870		err = auxtrace_index__alloc(head);
 871		if (err)
 872			return NULL;
 873		auxtrace_index = list_entry(head->prev, struct auxtrace_index,
 874					    list);
 875	}
 876
 877	return auxtrace_index;
 878}
 879
 880int auxtrace_index__auxtrace_event(struct list_head *head,
 881				   union perf_event *event, off_t file_offset)
 882{
 883	struct auxtrace_index *auxtrace_index;
 884	size_t nr;
 885
 886	auxtrace_index = auxtrace_index__last(head);
 887	if (!auxtrace_index)
 888		return -ENOMEM;
 889
 890	nr = auxtrace_index->nr;
 891	auxtrace_index->entries[nr].file_offset = file_offset;
 892	auxtrace_index->entries[nr].sz = event->header.size;
 893	auxtrace_index->nr += 1;
 894
 895	return 0;
 896}
 897
 898static int auxtrace_index__do_write(int fd,
 899				    struct auxtrace_index *auxtrace_index)
 900{
 901	struct auxtrace_index_entry ent;
 902	size_t i;
 903
 904	for (i = 0; i < auxtrace_index->nr; i++) {
 905		ent.file_offset = auxtrace_index->entries[i].file_offset;
 906		ent.sz = auxtrace_index->entries[i].sz;
 907		if (writen(fd, &ent, sizeof(ent)) != sizeof(ent))
 908			return -errno;
 909	}
 910	return 0;
 911}
 912
 913int auxtrace_index__write(int fd, struct list_head *head)
 914{
 915	struct auxtrace_index *auxtrace_index;
 916	u64 total = 0;
 917	int err;
 918
 919	list_for_each_entry(auxtrace_index, head, list)
 920		total += auxtrace_index->nr;
 921
 922	if (writen(fd, &total, sizeof(total)) != sizeof(total))
 923		return -errno;
 924
 925	list_for_each_entry(auxtrace_index, head, list) {
 926		err = auxtrace_index__do_write(fd, auxtrace_index);
 927		if (err)
 928			return err;
 929	}
 930
 931	return 0;
 932}
 933
 934static int auxtrace_index__process_entry(int fd, struct list_head *head,
 935					 bool needs_swap)
 936{
 937	struct auxtrace_index *auxtrace_index;
 938	struct auxtrace_index_entry ent;
 939	size_t nr;
 940
 941	if (readn(fd, &ent, sizeof(ent)) != sizeof(ent))
 942		return -1;
 943
 944	auxtrace_index = auxtrace_index__last(head);
 945	if (!auxtrace_index)
 946		return -1;
 947
 948	nr = auxtrace_index->nr;
 949	if (needs_swap) {
 950		auxtrace_index->entries[nr].file_offset =
 951						bswap_64(ent.file_offset);
 952		auxtrace_index->entries[nr].sz = bswap_64(ent.sz);
 953	} else {
 954		auxtrace_index->entries[nr].file_offset = ent.file_offset;
 955		auxtrace_index->entries[nr].sz = ent.sz;
 956	}
 957
 958	auxtrace_index->nr = nr + 1;
 959
 960	return 0;
 961}
 962
 963int auxtrace_index__process(int fd, u64 size, struct perf_session *session,
 964			    bool needs_swap)
 965{
 966	struct list_head *head = &session->auxtrace_index;
 967	u64 nr;
 968
 969	if (readn(fd, &nr, sizeof(u64)) != sizeof(u64))
 970		return -1;
 971
 972	if (needs_swap)
 973		nr = bswap_64(nr);
 974
 975	if (sizeof(u64) + nr * sizeof(struct auxtrace_index_entry) > size)
 976		return -1;
 977
 978	while (nr--) {
 979		int err;
 980
 981		err = auxtrace_index__process_entry(fd, head, needs_swap);
 982		if (err)
 983			return -1;
 984	}
 985
 986	return 0;
 987}
 988
 989static int auxtrace_queues__process_index_entry(struct auxtrace_queues *queues,
 990						struct perf_session *session,
 991						struct auxtrace_index_entry *ent)
 992{
 993	return auxtrace_queues__add_indexed_event(queues, session,
 994						  ent->file_offset, ent->sz);
 995}
 996
 997int auxtrace_queues__process_index(struct auxtrace_queues *queues,
 998				   struct perf_session *session)
 999{
1000	struct auxtrace_index *auxtrace_index;
1001	struct auxtrace_index_entry *ent;
1002	size_t i;
1003	int err;
1004
1005	if (auxtrace__dont_decode(session))
1006		return 0;
1007
1008	list_for_each_entry(auxtrace_index, &session->auxtrace_index, list) {
1009		for (i = 0; i < auxtrace_index->nr; i++) {
1010			ent = &auxtrace_index->entries[i];
1011			err = auxtrace_queues__process_index_entry(queues,
1012								   session,
1013								   ent);
1014			if (err)
1015				return err;
1016		}
1017	}
1018	return 0;
1019}
1020
1021struct auxtrace_buffer *auxtrace_buffer__next(struct auxtrace_queue *queue,
1022					      struct auxtrace_buffer *buffer)
1023{
1024	if (buffer) {
1025		if (list_is_last(&buffer->list, &queue->head))
1026			return NULL;
1027		return list_entry(buffer->list.next, struct auxtrace_buffer,
1028				  list);
1029	} else {
1030		if (list_empty(&queue->head))
1031			return NULL;
1032		return list_entry(queue->head.next, struct auxtrace_buffer,
1033				  list);
1034	}
1035}
1036
1037struct auxtrace_queue *auxtrace_queues__sample_queue(struct auxtrace_queues *queues,
1038						     struct perf_sample *sample,
1039						     struct perf_session *session)
1040{
1041	struct perf_sample_id *sid;
1042	unsigned int idx;
1043	u64 id;
1044
1045	id = sample->id;
1046	if (!id)
1047		return NULL;
1048
1049	sid = evlist__id2sid(session->evlist, id);
1050	if (!sid)
1051		return NULL;
1052
1053	idx = sid->idx;
1054
1055	if (idx >= queues->nr_queues)
1056		return NULL;
1057
1058	return &queues->queue_array[idx];
1059}
1060
1061int auxtrace_queues__add_sample(struct auxtrace_queues *queues,
1062				struct perf_session *session,
1063				struct perf_sample *sample, u64 data_offset,
1064				u64 reference)
1065{
1066	struct auxtrace_buffer buffer = {
1067		.pid = -1,
1068		.data_offset = data_offset,
1069		.reference = reference,
1070		.size = sample->aux_sample.size,
1071	};
1072	struct perf_sample_id *sid;
1073	u64 id = sample->id;
1074	unsigned int idx;
1075
1076	if (!id)
1077		return -EINVAL;
1078
1079	sid = evlist__id2sid(session->evlist, id);
1080	if (!sid)
1081		return -ENOENT;
1082
1083	idx = sid->idx;
1084	buffer.tid = sid->tid;
1085	buffer.cpu = sid->cpu;
1086
1087	return auxtrace_queues__add_buffer(queues, session, idx, &buffer, NULL);
1088}
1089
1090struct queue_data {
1091	bool samples;
1092	bool events;
1093};
1094
1095static int auxtrace_queue_data_cb(struct perf_session *session,
1096				  union perf_event *event, u64 offset,
1097				  void *data)
1098{
1099	struct queue_data *qd = data;
1100	struct perf_sample sample;
1101	int err;
1102
1103	if (qd->events && event->header.type == PERF_RECORD_AUXTRACE) {
1104		if (event->header.size < sizeof(struct perf_record_auxtrace))
1105			return -EINVAL;
1106		offset += event->header.size;
1107		return session->auxtrace->queue_data(session, NULL, event,
1108						     offset);
1109	}
1110
1111	if (!qd->samples || event->header.type != PERF_RECORD_SAMPLE)
1112		return 0;
1113
1114	err = evlist__parse_sample(session->evlist, event, &sample);
1115	if (err)
1116		return err;
1117
1118	if (!sample.aux_sample.size)
1119		return 0;
1120
1121	offset += sample.aux_sample.data - (void *)event;
1122
1123	return session->auxtrace->queue_data(session, &sample, NULL, offset);
1124}
1125
1126int auxtrace_queue_data(struct perf_session *session, bool samples, bool events)
1127{
1128	struct queue_data qd = {
1129		.samples = samples,
1130		.events = events,
1131	};
1132
1133	if (auxtrace__dont_decode(session))
1134		return 0;
1135
1136	if (perf_data__is_pipe(session->data))
1137		return 0;
1138
1139	if (!session->auxtrace || !session->auxtrace->queue_data)
1140		return -EINVAL;
1141
1142	return perf_session__peek_events(session, session->header.data_offset,
1143					 session->header.data_size,
1144					 auxtrace_queue_data_cb, &qd);
1145}
1146
1147void *auxtrace_buffer__get_data_rw(struct auxtrace_buffer *buffer, int fd, bool rw)
1148{
1149	int prot = rw ? PROT_READ | PROT_WRITE : PROT_READ;
1150	size_t adj = buffer->data_offset & (page_size - 1);
1151	size_t size = buffer->size + adj;
1152	off_t file_offset = buffer->data_offset - adj;
1153	void *addr;
1154
1155	if (buffer->data)
1156		return buffer->data;
1157
1158	addr = mmap(NULL, size, prot, MAP_SHARED, fd, file_offset);
1159	if (addr == MAP_FAILED)
1160		return NULL;
1161
1162	buffer->mmap_addr = addr;
1163	buffer->mmap_size = size;
1164
1165	buffer->data = addr + adj;
1166
1167	return buffer->data;
1168}
1169
1170void auxtrace_buffer__put_data(struct auxtrace_buffer *buffer)
1171{
1172	if (!buffer->data || !buffer->mmap_addr)
1173		return;
1174	munmap(buffer->mmap_addr, buffer->mmap_size);
1175	buffer->mmap_addr = NULL;
1176	buffer->mmap_size = 0;
1177	buffer->data = NULL;
1178	buffer->use_data = NULL;
1179}
1180
1181void auxtrace_buffer__drop_data(struct auxtrace_buffer *buffer)
1182{
1183	auxtrace_buffer__put_data(buffer);
1184	if (buffer->data_needs_freeing) {
1185		buffer->data_needs_freeing = false;
1186		zfree(&buffer->data);
1187		buffer->use_data = NULL;
1188		buffer->size = 0;
1189	}
1190}
1191
1192void auxtrace_buffer__free(struct auxtrace_buffer *buffer)
1193{
1194	auxtrace_buffer__drop_data(buffer);
1195	free(buffer);
1196}
1197
1198void auxtrace_synth_guest_error(struct perf_record_auxtrace_error *auxtrace_error, int type,
1199				int code, int cpu, pid_t pid, pid_t tid, u64 ip,
1200				const char *msg, u64 timestamp,
1201				pid_t machine_pid, int vcpu)
1202{
1203	size_t size;
1204
1205	memset(auxtrace_error, 0, sizeof(struct perf_record_auxtrace_error));
1206
1207	auxtrace_error->header.type = PERF_RECORD_AUXTRACE_ERROR;
1208	auxtrace_error->type = type;
1209	auxtrace_error->code = code;
1210	auxtrace_error->cpu = cpu;
1211	auxtrace_error->pid = pid;
1212	auxtrace_error->tid = tid;
1213	auxtrace_error->fmt = 1;
1214	auxtrace_error->ip = ip;
1215	auxtrace_error->time = timestamp;
1216	strlcpy(auxtrace_error->msg, msg, MAX_AUXTRACE_ERROR_MSG);
1217	if (machine_pid) {
1218		auxtrace_error->fmt = 2;
1219		auxtrace_error->machine_pid = machine_pid;
1220		auxtrace_error->vcpu = vcpu;
1221		size = sizeof(*auxtrace_error);
1222	} else {
1223		size = (void *)auxtrace_error->msg - (void *)auxtrace_error +
1224		       strlen(auxtrace_error->msg) + 1;
1225	}
1226	auxtrace_error->header.size = PERF_ALIGN(size, sizeof(u64));
1227}
1228
1229void auxtrace_synth_error(struct perf_record_auxtrace_error *auxtrace_error, int type,
1230			  int code, int cpu, pid_t pid, pid_t tid, u64 ip,
1231			  const char *msg, u64 timestamp)
1232{
1233	auxtrace_synth_guest_error(auxtrace_error, type, code, cpu, pid, tid,
1234				   ip, msg, timestamp, 0, -1);
1235}
1236
1237int perf_event__synthesize_auxtrace_info(struct auxtrace_record *itr,
1238					 struct perf_tool *tool,
1239					 struct perf_session *session,
1240					 perf_event__handler_t process)
1241{
1242	union perf_event *ev;
1243	size_t priv_size;
1244	int err;
1245
1246	pr_debug2("Synthesizing auxtrace information\n");
1247	priv_size = auxtrace_record__info_priv_size(itr, session->evlist);
1248	ev = zalloc(sizeof(struct perf_record_auxtrace_info) + priv_size);
1249	if (!ev)
1250		return -ENOMEM;
1251
1252	ev->auxtrace_info.header.type = PERF_RECORD_AUXTRACE_INFO;
1253	ev->auxtrace_info.header.size = sizeof(struct perf_record_auxtrace_info) +
1254					priv_size;
1255	err = auxtrace_record__info_fill(itr, session, &ev->auxtrace_info,
1256					 priv_size);
1257	if (err)
1258		goto out_free;
1259
1260	err = process(tool, ev, NULL, NULL);
1261out_free:
1262	free(ev);
1263	return err;
1264}
1265
1266static void unleader_evsel(struct evlist *evlist, struct evsel *leader)
1267{
1268	struct evsel *new_leader = NULL;
1269	struct evsel *evsel;
1270
1271	/* Find new leader for the group */
1272	evlist__for_each_entry(evlist, evsel) {
1273		if (!evsel__has_leader(evsel, leader) || evsel == leader)
1274			continue;
1275		if (!new_leader)
1276			new_leader = evsel;
1277		evsel__set_leader(evsel, new_leader);
1278	}
1279
1280	/* Update group information */
1281	if (new_leader) {
1282		zfree(&new_leader->group_name);
1283		new_leader->group_name = leader->group_name;
1284		leader->group_name = NULL;
1285
1286		new_leader->core.nr_members = leader->core.nr_members - 1;
1287		leader->core.nr_members = 1;
1288	}
1289}
1290
1291static void unleader_auxtrace(struct perf_session *session)
1292{
1293	struct evsel *evsel;
1294
1295	evlist__for_each_entry(session->evlist, evsel) {
1296		if (auxtrace__evsel_is_auxtrace(session, evsel) &&
1297		    evsel__is_group_leader(evsel)) {
1298			unleader_evsel(session->evlist, evsel);
1299		}
1300	}
1301}
1302
1303int perf_event__process_auxtrace_info(struct perf_session *session,
1304				      union perf_event *event)
1305{
1306	enum auxtrace_type type = event->auxtrace_info.type;
1307	int err;
1308
1309	if (dump_trace)
1310		fprintf(stdout, " type: %u\n", type);
1311
1312	switch (type) {
1313	case PERF_AUXTRACE_INTEL_PT:
1314		err = intel_pt_process_auxtrace_info(event, session);
1315		break;
1316	case PERF_AUXTRACE_INTEL_BTS:
1317		err = intel_bts_process_auxtrace_info(event, session);
1318		break;
1319	case PERF_AUXTRACE_ARM_SPE:
1320		err = arm_spe_process_auxtrace_info(event, session);
1321		break;
1322	case PERF_AUXTRACE_CS_ETM:
1323		err = cs_etm__process_auxtrace_info(event, session);
1324		break;
1325	case PERF_AUXTRACE_S390_CPUMSF:
1326		err = s390_cpumsf_process_auxtrace_info(event, session);
1327		break;
1328	case PERF_AUXTRACE_HISI_PTT:
1329		err = hisi_ptt_process_auxtrace_info(event, session);
1330		break;
1331	case PERF_AUXTRACE_UNKNOWN:
1332	default:
1333		return -EINVAL;
1334	}
1335
1336	if (err)
1337		return err;
1338
1339	unleader_auxtrace(session);
1340
1341	return 0;
1342}
1343
1344s64 perf_event__process_auxtrace(struct perf_session *session,
1345				 union perf_event *event)
1346{
1347	s64 err;
1348
1349	if (dump_trace)
1350		fprintf(stdout, " size: %#"PRI_lx64"  offset: %#"PRI_lx64"  ref: %#"PRI_lx64"  idx: %u  tid: %d  cpu: %d\n",
1351			event->auxtrace.size, event->auxtrace.offset,
1352			event->auxtrace.reference, event->auxtrace.idx,
1353			event->auxtrace.tid, event->auxtrace.cpu);
1354
1355	if (auxtrace__dont_decode(session))
1356		return event->auxtrace.size;
1357
1358	if (!session->auxtrace || event->header.type != PERF_RECORD_AUXTRACE)
1359		return -EINVAL;
1360
1361	err = session->auxtrace->process_auxtrace_event(session, event, session->tool);
1362	if (err < 0)
1363		return err;
1364
1365	return event->auxtrace.size;
1366}
1367
1368#define PERF_ITRACE_DEFAULT_PERIOD_TYPE		PERF_ITRACE_PERIOD_NANOSECS
1369#define PERF_ITRACE_DEFAULT_PERIOD		100000
1370#define PERF_ITRACE_DEFAULT_CALLCHAIN_SZ	16
1371#define PERF_ITRACE_MAX_CALLCHAIN_SZ		1024
1372#define PERF_ITRACE_DEFAULT_LAST_BRANCH_SZ	64
1373#define PERF_ITRACE_MAX_LAST_BRANCH_SZ		1024
1374
1375void itrace_synth_opts__set_default(struct itrace_synth_opts *synth_opts,
1376				    bool no_sample)
1377{
1378	synth_opts->branches = true;
1379	synth_opts->transactions = true;
1380	synth_opts->ptwrites = true;
1381	synth_opts->pwr_events = true;
1382	synth_opts->other_events = true;
1383	synth_opts->intr_events = true;
1384	synth_opts->errors = true;
1385	synth_opts->flc = true;
1386	synth_opts->llc = true;
1387	synth_opts->tlb = true;
1388	synth_opts->mem = true;
1389	synth_opts->remote_access = true;
1390
1391	if (no_sample) {
1392		synth_opts->period_type = PERF_ITRACE_PERIOD_INSTRUCTIONS;
1393		synth_opts->period = 1;
1394		synth_opts->calls = true;
1395	} else {
1396		synth_opts->instructions = true;
1397		synth_opts->cycles = true;
1398		synth_opts->period_type = PERF_ITRACE_DEFAULT_PERIOD_TYPE;
1399		synth_opts->period = PERF_ITRACE_DEFAULT_PERIOD;
1400	}
1401	synth_opts->callchain_sz = PERF_ITRACE_DEFAULT_CALLCHAIN_SZ;
1402	synth_opts->last_branch_sz = PERF_ITRACE_DEFAULT_LAST_BRANCH_SZ;
1403	synth_opts->initial_skip = 0;
1404}
1405
1406static int get_flag(const char **ptr, unsigned int *flags)
1407{
1408	while (1) {
1409		char c = **ptr;
1410
1411		if (c >= 'a' && c <= 'z') {
1412			*flags |= 1 << (c - 'a');
1413			++*ptr;
1414			return 0;
1415		} else if (c == ' ') {
1416			++*ptr;
1417			continue;
1418		} else {
1419			return -1;
1420		}
1421	}
1422}
1423
1424static int get_flags(const char **ptr, unsigned int *plus_flags, unsigned int *minus_flags)
1425{
1426	while (1) {
1427		switch (**ptr) {
1428		case '+':
1429			++*ptr;
1430			if (get_flag(ptr, plus_flags))
1431				return -1;
1432			break;
1433		case '-':
1434			++*ptr;
1435			if (get_flag(ptr, minus_flags))
1436				return -1;
1437			break;
1438		case ' ':
1439			++*ptr;
1440			break;
1441		default:
1442			return 0;
1443		}
1444	}
1445}
1446
1447#define ITRACE_DFLT_LOG_ON_ERROR_SZ 16384
1448
1449static unsigned int itrace_log_on_error_size(void)
1450{
1451	unsigned int sz = 0;
1452
1453	perf_config_scan("itrace.debug-log-buffer-size", "%u", &sz);
1454	return sz ?: ITRACE_DFLT_LOG_ON_ERROR_SZ;
1455}
1456
1457/*
1458 * Please check tools/perf/Documentation/perf-script.txt for information
1459 * about the options parsed here, which is introduced after this cset,
1460 * when support in 'perf script' for these options is introduced.
1461 */
1462int itrace_do_parse_synth_opts(struct itrace_synth_opts *synth_opts,
1463			       const char *str, int unset)
1464{
1465	const char *p;
1466	char *endptr;
1467	bool period_type_set = false;
1468	bool period_set = false;
1469
1470	synth_opts->set = true;
1471
1472	if (unset) {
1473		synth_opts->dont_decode = true;
1474		return 0;
1475	}
1476
1477	if (!str) {
1478		itrace_synth_opts__set_default(synth_opts,
1479					       synth_opts->default_no_sample);
1480		return 0;
1481	}
1482
1483	for (p = str; *p;) {
1484		switch (*p++) {
1485		case 'i':
1486		case 'y':
1487			if (p[-1] == 'y')
1488				synth_opts->cycles = true;
1489			else
1490				synth_opts->instructions = true;
1491			while (*p == ' ' || *p == ',')
1492				p += 1;
1493			if (isdigit(*p)) {
1494				synth_opts->period = strtoull(p, &endptr, 10);
1495				period_set = true;
1496				p = endptr;
1497				while (*p == ' ' || *p == ',')
1498					p += 1;
1499				switch (*p++) {
1500				case 'i':
1501					synth_opts->period_type =
1502						PERF_ITRACE_PERIOD_INSTRUCTIONS;
1503					period_type_set = true;
1504					break;
1505				case 't':
1506					synth_opts->period_type =
1507						PERF_ITRACE_PERIOD_TICKS;
1508					period_type_set = true;
1509					break;
1510				case 'm':
1511					synth_opts->period *= 1000;
1512					/* Fall through */
1513				case 'u':
1514					synth_opts->period *= 1000;
1515					/* Fall through */
1516				case 'n':
1517					if (*p++ != 's')
1518						goto out_err;
1519					synth_opts->period_type =
1520						PERF_ITRACE_PERIOD_NANOSECS;
1521					period_type_set = true;
1522					break;
1523				case '\0':
1524					goto out;
1525				default:
1526					goto out_err;
1527				}
1528			}
1529			break;
1530		case 'b':
1531			synth_opts->branches = true;
1532			break;
1533		case 'x':
1534			synth_opts->transactions = true;
1535			break;
1536		case 'w':
1537			synth_opts->ptwrites = true;
1538			break;
1539		case 'p':
1540			synth_opts->pwr_events = true;
1541			break;
1542		case 'o':
1543			synth_opts->other_events = true;
1544			break;
1545		case 'I':
1546			synth_opts->intr_events = true;
1547			break;
1548		case 'e':
1549			synth_opts->errors = true;
1550			if (get_flags(&p, &synth_opts->error_plus_flags,
1551				      &synth_opts->error_minus_flags))
1552				goto out_err;
1553			break;
1554		case 'd':
1555			synth_opts->log = true;
1556			if (get_flags(&p, &synth_opts->log_plus_flags,
1557				      &synth_opts->log_minus_flags))
1558				goto out_err;
1559			if (synth_opts->log_plus_flags & AUXTRACE_LOG_FLG_ON_ERROR)
1560				synth_opts->log_on_error_size = itrace_log_on_error_size();
1561			break;
1562		case 'c':
1563			synth_opts->branches = true;
1564			synth_opts->calls = true;
1565			break;
1566		case 'r':
1567			synth_opts->branches = true;
1568			synth_opts->returns = true;
1569			break;
1570		case 'G':
1571		case 'g':
1572			if (p[-1] == 'G')
1573				synth_opts->add_callchain = true;
1574			else
1575				synth_opts->callchain = true;
1576			synth_opts->callchain_sz =
1577					PERF_ITRACE_DEFAULT_CALLCHAIN_SZ;
1578			while (*p == ' ' || *p == ',')
1579				p += 1;
1580			if (isdigit(*p)) {
1581				unsigned int val;
1582
1583				val = strtoul(p, &endptr, 10);
1584				p = endptr;
1585				if (!val || val > PERF_ITRACE_MAX_CALLCHAIN_SZ)
1586					goto out_err;
1587				synth_opts->callchain_sz = val;
1588			}
1589			break;
1590		case 'L':
1591		case 'l':
1592			if (p[-1] == 'L')
1593				synth_opts->add_last_branch = true;
1594			else
1595				synth_opts->last_branch = true;
1596			synth_opts->last_branch_sz =
1597					PERF_ITRACE_DEFAULT_LAST_BRANCH_SZ;
1598			while (*p == ' ' || *p == ',')
1599				p += 1;
1600			if (isdigit(*p)) {
1601				unsigned int val;
1602
1603				val = strtoul(p, &endptr, 10);
1604				p = endptr;
1605				if (!val ||
1606				    val > PERF_ITRACE_MAX_LAST_BRANCH_SZ)
1607					goto out_err;
1608				synth_opts->last_branch_sz = val;
1609			}
1610			break;
1611		case 's':
1612			synth_opts->initial_skip = strtoul(p, &endptr, 10);
1613			if (p == endptr)
1614				goto out_err;
1615			p = endptr;
1616			break;
1617		case 'f':
1618			synth_opts->flc = true;
1619			break;
1620		case 'm':
1621			synth_opts->llc = true;
1622			break;
1623		case 't':
1624			synth_opts->tlb = true;
1625			break;
1626		case 'a':
1627			synth_opts->remote_access = true;
1628			break;
1629		case 'M':
1630			synth_opts->mem = true;
1631			break;
1632		case 'q':
1633			synth_opts->quick += 1;
1634			break;
1635		case 'A':
1636			synth_opts->approx_ipc = true;
1637			break;
1638		case 'Z':
1639			synth_opts->timeless_decoding = true;
1640			break;
1641		case 'T':
1642			synth_opts->use_timestamp = true;
1643			break;
1644		case ' ':
1645		case ',':
1646			break;
1647		default:
1648			goto out_err;
1649		}
1650	}
1651out:
1652	if (synth_opts->instructions || synth_opts->cycles) {
1653		if (!period_type_set)
1654			synth_opts->period_type =
1655					PERF_ITRACE_DEFAULT_PERIOD_TYPE;
1656		if (!period_set)
1657			synth_opts->period = PERF_ITRACE_DEFAULT_PERIOD;
1658	}
1659
1660	return 0;
1661
1662out_err:
1663	pr_err("Bad Instruction Tracing options '%s'\n", str);
1664	return -EINVAL;
1665}
1666
1667int itrace_parse_synth_opts(const struct option *opt, const char *str, int unset)
1668{
1669	return itrace_do_parse_synth_opts(opt->value, str, unset);
1670}
1671
1672static const char * const auxtrace_error_type_name[] = {
1673	[PERF_AUXTRACE_ERROR_ITRACE] = "instruction trace",
1674};
1675
1676static const char *auxtrace_error_name(int type)
1677{
1678	const char *error_type_name = NULL;
1679
1680	if (type < PERF_AUXTRACE_ERROR_MAX)
1681		error_type_name = auxtrace_error_type_name[type];
1682	if (!error_type_name)
1683		error_type_name = "unknown AUX";
1684	return error_type_name;
1685}
1686
1687size_t perf_event__fprintf_auxtrace_error(union perf_event *event, FILE *fp)
1688{
1689	struct perf_record_auxtrace_error *e = &event->auxtrace_error;
1690	unsigned long long nsecs = e->time;
1691	const char *msg = e->msg;
1692	int ret;
1693
1694	ret = fprintf(fp, " %s error type %u",
1695		      auxtrace_error_name(e->type), e->type);
1696
1697	if (e->fmt && nsecs) {
1698		unsigned long secs = nsecs / NSEC_PER_SEC;
1699
1700		nsecs -= secs * NSEC_PER_SEC;
1701		ret += fprintf(fp, " time %lu.%09llu", secs, nsecs);
1702	} else {
1703		ret += fprintf(fp, " time 0");
1704	}
1705
1706	if (!e->fmt)
1707		msg = (const char *)&e->time;
1708
1709	if (e->fmt >= 2 && e->machine_pid)
1710		ret += fprintf(fp, " machine_pid %d vcpu %d", e->machine_pid, e->vcpu);
1711
1712	ret += fprintf(fp, " cpu %d pid %d tid %d ip %#"PRI_lx64" code %u: %s\n",
1713		       e->cpu, e->pid, e->tid, e->ip, e->code, msg);
1714	return ret;
1715}
1716
1717void perf_session__auxtrace_error_inc(struct perf_session *session,
1718				      union perf_event *event)
1719{
1720	struct perf_record_auxtrace_error *e = &event->auxtrace_error;
1721
1722	if (e->type < PERF_AUXTRACE_ERROR_MAX)
1723		session->evlist->stats.nr_auxtrace_errors[e->type] += 1;
1724}
1725
1726void events_stats__auxtrace_error_warn(const struct events_stats *stats)
1727{
1728	int i;
1729
1730	for (i = 0; i < PERF_AUXTRACE_ERROR_MAX; i++) {
1731		if (!stats->nr_auxtrace_errors[i])
1732			continue;
1733		ui__warning("%u %s errors\n",
1734			    stats->nr_auxtrace_errors[i],
1735			    auxtrace_error_name(i));
1736	}
1737}
1738
1739int perf_event__process_auxtrace_error(struct perf_session *session,
1740				       union perf_event *event)
1741{
1742	if (auxtrace__dont_decode(session))
1743		return 0;
1744
1745	perf_event__fprintf_auxtrace_error(event, stdout);
1746	return 0;
1747}
1748
1749/*
1750 * In the compat mode kernel runs in 64-bit and perf tool runs in 32-bit mode,
1751 * 32-bit perf tool cannot access 64-bit value atomically, which might lead to
1752 * the issues caused by the below sequence on multiple CPUs: when perf tool
1753 * accesses either the load operation or the store operation for 64-bit value,
1754 * on some architectures the operation is divided into two instructions, one
1755 * is for accessing the low 32-bit value and another is for the high 32-bit;
1756 * thus these two user operations can give the kernel chances to access the
1757 * 64-bit value, and thus leads to the unexpected load values.
1758 *
1759 *   kernel (64-bit)                        user (32-bit)
1760 *
1761 *   if (LOAD ->aux_tail) { --,             LOAD ->aux_head_lo
1762 *       STORE $aux_data      |       ,--->
1763 *       FLUSH $aux_data      |       |     LOAD ->aux_head_hi
1764 *       STORE ->aux_head   --|-------`     smp_rmb()
1765 *   }                        |             LOAD $data
1766 *                            |             smp_mb()
1767 *                            |             STORE ->aux_tail_lo
1768 *                            `----------->
1769 *                                          STORE ->aux_tail_hi
1770 *
1771 * For this reason, it's impossible for the perf tool to work correctly when
1772 * the AUX head or tail is bigger than 4GB (more than 32 bits length); and we
1773 * can not simply limit the AUX ring buffer to less than 4GB, the reason is
1774 * the pointers can be increased monotonically, whatever the buffer size it is,
1775 * at the end the head and tail can be bigger than 4GB and carry out to the
1776 * high 32-bit.
1777 *
1778 * To mitigate the issues and improve the user experience, we can allow the
1779 * perf tool working in certain conditions and bail out with error if detect
1780 * any overflow cannot be handled.
1781 *
1782 * For reading the AUX head, it reads out the values for three times, and
1783 * compares the high 4 bytes of the values between the first time and the last
1784 * time, if there has no change for high 4 bytes injected by the kernel during
1785 * the user reading sequence, it's safe for use the second value.
1786 *
1787 * When compat_auxtrace_mmap__write_tail() detects any carrying in the high
1788 * 32 bits, it means there have two store operations in user space and it cannot
1789 * promise the atomicity for 64-bit write, so return '-1' in this case to tell
1790 * the caller an overflow error has happened.
1791 */
1792u64 __weak compat_auxtrace_mmap__read_head(struct auxtrace_mmap *mm)
1793{
1794	struct perf_event_mmap_page *pc = mm->userpg;
1795	u64 first, second, last;
1796	u64 mask = (u64)(UINT32_MAX) << 32;
1797
1798	do {
1799		first = READ_ONCE(pc->aux_head);
1800		/* Ensure all reads are done after we read the head */
1801		smp_rmb();
1802		second = READ_ONCE(pc->aux_head);
1803		/* Ensure all reads are done after we read the head */
1804		smp_rmb();
1805		last = READ_ONCE(pc->aux_head);
1806	} while ((first & mask) != (last & mask));
1807
1808	return second;
1809}
1810
1811int __weak compat_auxtrace_mmap__write_tail(struct auxtrace_mmap *mm, u64 tail)
1812{
1813	struct perf_event_mmap_page *pc = mm->userpg;
1814	u64 mask = (u64)(UINT32_MAX) << 32;
1815
1816	if (tail & mask)
1817		return -1;
1818
1819	/* Ensure all reads are done before we write the tail out */
1820	smp_mb();
1821	WRITE_ONCE(pc->aux_tail, tail);
1822	return 0;
1823}
1824
1825static int __auxtrace_mmap__read(struct mmap *map,
1826				 struct auxtrace_record *itr,
1827				 struct perf_tool *tool, process_auxtrace_t fn,
1828				 bool snapshot, size_t snapshot_size)
1829{
1830	struct auxtrace_mmap *mm = &map->auxtrace_mmap;
1831	u64 head, old = mm->prev, offset, ref;
1832	unsigned char *data = mm->base;
1833	size_t size, head_off, old_off, len1, len2, padding;
1834	union perf_event ev;
1835	void *data1, *data2;
1836	int kernel_is_64_bit = perf_env__kernel_is_64_bit(evsel__env(NULL));
1837
1838	head = auxtrace_mmap__read_head(mm, kernel_is_64_bit);
1839
1840	if (snapshot &&
1841	    auxtrace_record__find_snapshot(itr, mm->idx, mm, data, &head, &old))
1842		return -1;
1843
1844	if (old == head)
1845		return 0;
1846
1847	pr_debug3("auxtrace idx %d old %#"PRIx64" head %#"PRIx64" diff %#"PRIx64"\n",
1848		  mm->idx, old, head, head - old);
1849
1850	if (mm->mask) {
1851		head_off = head & mm->mask;
1852		old_off = old & mm->mask;
1853	} else {
1854		head_off = head % mm->len;
1855		old_off = old % mm->len;
1856	}
1857
1858	if (head_off > old_off)
1859		size = head_off - old_off;
1860	else
1861		size = mm->len - (old_off - head_off);
1862
1863	if (snapshot && size > snapshot_size)
1864		size = snapshot_size;
1865
1866	ref = auxtrace_record__reference(itr);
1867
1868	if (head > old || size <= head || mm->mask) {
1869		offset = head - size;
1870	} else {
1871		/*
1872		 * When the buffer size is not a power of 2, 'head' wraps at the
1873		 * highest multiple of the buffer size, so we have to subtract
1874		 * the remainder here.
1875		 */
1876		u64 rem = (0ULL - mm->len) % mm->len;
1877
1878		offset = head - size - rem;
1879	}
1880
1881	if (size > head_off) {
1882		len1 = size - head_off;
1883		data1 = &data[mm->len - len1];
1884		len2 = head_off;
1885		data2 = &data[0];
1886	} else {
1887		len1 = size;
1888		data1 = &data[head_off - len1];
1889		len2 = 0;
1890		data2 = NULL;
1891	}
1892
1893	if (itr->alignment) {
1894		unsigned int unwanted = len1 % itr->alignment;
1895
1896		len1 -= unwanted;
1897		size -= unwanted;
1898	}
1899
1900	/* padding must be written by fn() e.g. record__process_auxtrace() */
1901	padding = size & (PERF_AUXTRACE_RECORD_ALIGNMENT - 1);
1902	if (padding)
1903		padding = PERF_AUXTRACE_RECORD_ALIGNMENT - padding;
1904
1905	memset(&ev, 0, sizeof(ev));
1906	ev.auxtrace.header.type = PERF_RECORD_AUXTRACE;
1907	ev.auxtrace.header.size = sizeof(ev.auxtrace);
1908	ev.auxtrace.size = size + padding;
1909	ev.auxtrace.offset = offset;
1910	ev.auxtrace.reference = ref;
1911	ev.auxtrace.idx = mm->idx;
1912	ev.auxtrace.tid = mm->tid;
1913	ev.auxtrace.cpu = mm->cpu;
1914
1915	if (fn(tool, map, &ev, data1, len1, data2, len2))
1916		return -1;
1917
1918	mm->prev = head;
1919
1920	if (!snapshot) {
1921		int err;
1922
1923		err = auxtrace_mmap__write_tail(mm, head, kernel_is_64_bit);
1924		if (err < 0)
1925			return err;
1926
1927		if (itr->read_finish) {
1928			err = itr->read_finish(itr, mm->idx);
1929			if (err < 0)
1930				return err;
1931		}
1932	}
1933
1934	return 1;
1935}
1936
1937int auxtrace_mmap__read(struct mmap *map, struct auxtrace_record *itr,
1938			struct perf_tool *tool, process_auxtrace_t fn)
1939{
1940	return __auxtrace_mmap__read(map, itr, tool, fn, false, 0);
1941}
1942
1943int auxtrace_mmap__read_snapshot(struct mmap *map,
1944				 struct auxtrace_record *itr,
1945				 struct perf_tool *tool, process_auxtrace_t fn,
1946				 size_t snapshot_size)
1947{
1948	return __auxtrace_mmap__read(map, itr, tool, fn, true, snapshot_size);
1949}
1950
1951/**
1952 * struct auxtrace_cache - hash table to implement a cache
1953 * @hashtable: the hashtable
1954 * @sz: hashtable size (number of hlists)
1955 * @entry_size: size of an entry
1956 * @limit: limit the number of entries to this maximum, when reached the cache
1957 *         is dropped and caching begins again with an empty cache
1958 * @cnt: current number of entries
1959 * @bits: hashtable size (@sz = 2^@bits)
1960 */
1961struct auxtrace_cache {
1962	struct hlist_head *hashtable;
1963	size_t sz;
1964	size_t entry_size;
1965	size_t limit;
1966	size_t cnt;
1967	unsigned int bits;
1968};
1969
1970struct auxtrace_cache *auxtrace_cache__new(unsigned int bits, size_t entry_size,
1971					   unsigned int limit_percent)
1972{
1973	struct auxtrace_cache *c;
1974	struct hlist_head *ht;
1975	size_t sz, i;
1976
1977	c = zalloc(sizeof(struct auxtrace_cache));
1978	if (!c)
1979		return NULL;
1980
1981	sz = 1UL << bits;
1982
1983	ht = calloc(sz, sizeof(struct hlist_head));
1984	if (!ht)
1985		goto out_free;
1986
1987	for (i = 0; i < sz; i++)
1988		INIT_HLIST_HEAD(&ht[i]);
1989
1990	c->hashtable = ht;
1991	c->sz = sz;
1992	c->entry_size = entry_size;
1993	c->limit = (c->sz * limit_percent) / 100;
1994	c->bits = bits;
1995
1996	return c;
1997
1998out_free:
1999	free(c);
2000	return NULL;
2001}
2002
2003static void auxtrace_cache__drop(struct auxtrace_cache *c)
2004{
2005	struct auxtrace_cache_entry *entry;
2006	struct hlist_node *tmp;
2007	size_t i;
2008
2009	if (!c)
2010		return;
2011
2012	for (i = 0; i < c->sz; i++) {
2013		hlist_for_each_entry_safe(entry, tmp, &c->hashtable[i], hash) {
2014			hlist_del(&entry->hash);
2015			auxtrace_cache__free_entry(c, entry);
2016		}
2017	}
2018
2019	c->cnt = 0;
2020}
2021
2022void auxtrace_cache__free(struct auxtrace_cache *c)
2023{
2024	if (!c)
2025		return;
2026
2027	auxtrace_cache__drop(c);
2028	zfree(&c->hashtable);
2029	free(c);
2030}
2031
2032void *auxtrace_cache__alloc_entry(struct auxtrace_cache *c)
2033{
2034	return malloc(c->entry_size);
2035}
2036
2037void auxtrace_cache__free_entry(struct auxtrace_cache *c __maybe_unused,
2038				void *entry)
2039{
2040	free(entry);
2041}
2042
2043int auxtrace_cache__add(struct auxtrace_cache *c, u32 key,
2044			struct auxtrace_cache_entry *entry)
2045{
2046	if (c->limit && ++c->cnt > c->limit)
2047		auxtrace_cache__drop(c);
2048
2049	entry->key = key;
2050	hlist_add_head(&entry->hash, &c->hashtable[hash_32(key, c->bits)]);
2051
2052	return 0;
2053}
2054
2055static struct auxtrace_cache_entry *auxtrace_cache__rm(struct auxtrace_cache *c,
2056						       u32 key)
2057{
2058	struct auxtrace_cache_entry *entry;
2059	struct hlist_head *hlist;
2060	struct hlist_node *n;
2061
2062	if (!c)
2063		return NULL;
2064
2065	hlist = &c->hashtable[hash_32(key, c->bits)];
2066	hlist_for_each_entry_safe(entry, n, hlist, hash) {
2067		if (entry->key == key) {
2068			hlist_del(&entry->hash);
2069			return entry;
2070		}
2071	}
2072
2073	return NULL;
2074}
2075
2076void auxtrace_cache__remove(struct auxtrace_cache *c, u32 key)
2077{
2078	struct auxtrace_cache_entry *entry = auxtrace_cache__rm(c, key);
2079
2080	auxtrace_cache__free_entry(c, entry);
2081}
2082
2083void *auxtrace_cache__lookup(struct auxtrace_cache *c, u32 key)
2084{
2085	struct auxtrace_cache_entry *entry;
2086	struct hlist_head *hlist;
2087
2088	if (!c)
2089		return NULL;
2090
2091	hlist = &c->hashtable[hash_32(key, c->bits)];
2092	hlist_for_each_entry(entry, hlist, hash) {
2093		if (entry->key == key)
2094			return entry;
2095	}
2096
2097	return NULL;
2098}
2099
2100static void addr_filter__free_str(struct addr_filter *filt)
2101{
2102	zfree(&filt->str);
2103	filt->action   = NULL;
2104	filt->sym_from = NULL;
2105	filt->sym_to   = NULL;
2106	filt->filename = NULL;
2107}
2108
2109static struct addr_filter *addr_filter__new(void)
2110{
2111	struct addr_filter *filt = zalloc(sizeof(*filt));
2112
2113	if (filt)
2114		INIT_LIST_HEAD(&filt->list);
2115
2116	return filt;
2117}
2118
2119static void addr_filter__free(struct addr_filter *filt)
2120{
2121	if (filt)
2122		addr_filter__free_str(filt);
2123	free(filt);
2124}
2125
2126static void addr_filters__add(struct addr_filters *filts,
2127			      struct addr_filter *filt)
2128{
2129	list_add_tail(&filt->list, &filts->head);
2130	filts->cnt += 1;
2131}
2132
2133static void addr_filters__del(struct addr_filters *filts,
2134			      struct addr_filter *filt)
2135{
2136	list_del_init(&filt->list);
2137	filts->cnt -= 1;
2138}
2139
2140void addr_filters__init(struct addr_filters *filts)
2141{
2142	INIT_LIST_HEAD(&filts->head);
2143	filts->cnt = 0;
2144}
2145
2146void addr_filters__exit(struct addr_filters *filts)
2147{
2148	struct addr_filter *filt, *n;
2149
2150	list_for_each_entry_safe(filt, n, &filts->head, list) {
2151		addr_filters__del(filts, filt);
2152		addr_filter__free(filt);
2153	}
2154}
2155
2156static int parse_num_or_str(char **inp, u64 *num, const char **str,
2157			    const char *str_delim)
2158{
2159	*inp += strspn(*inp, " ");
2160
2161	if (isdigit(**inp)) {
2162		char *endptr;
2163
2164		if (!num)
2165			return -EINVAL;
2166		errno = 0;
2167		*num = strtoull(*inp, &endptr, 0);
2168		if (errno)
2169			return -errno;
2170		if (endptr == *inp)
2171			return -EINVAL;
2172		*inp = endptr;
2173	} else {
2174		size_t n;
2175
2176		if (!str)
2177			return -EINVAL;
2178		*inp += strspn(*inp, " ");
2179		*str = *inp;
2180		n = strcspn(*inp, str_delim);
2181		if (!n)
2182			return -EINVAL;
2183		*inp += n;
2184		if (**inp) {
2185			**inp = '\0';
2186			*inp += 1;
2187		}
2188	}
2189	return 0;
2190}
2191
2192static int parse_action(struct addr_filter *filt)
2193{
2194	if (!strcmp(filt->action, "filter")) {
2195		filt->start = true;
2196		filt->range = true;
2197	} else if (!strcmp(filt->action, "start")) {
2198		filt->start = true;
2199	} else if (!strcmp(filt->action, "stop")) {
2200		filt->start = false;
2201	} else if (!strcmp(filt->action, "tracestop")) {
2202		filt->start = false;
2203		filt->range = true;
2204		filt->action += 5; /* Change 'tracestop' to 'stop' */
2205	} else {
2206		return -EINVAL;
2207	}
2208	return 0;
2209}
2210
2211static int parse_sym_idx(char **inp, int *idx)
2212{
2213	*idx = -1;
2214
2215	*inp += strspn(*inp, " ");
2216
2217	if (**inp != '#')
2218		return 0;
2219
2220	*inp += 1;
2221
2222	if (**inp == 'g' || **inp == 'G') {
2223		*inp += 1;
2224		*idx = 0;
2225	} else {
2226		unsigned long num;
2227		char *endptr;
2228
2229		errno = 0;
2230		num = strtoul(*inp, &endptr, 0);
2231		if (errno)
2232			return -errno;
2233		if (endptr == *inp || num > INT_MAX)
2234			return -EINVAL;
2235		*inp = endptr;
2236		*idx = num;
2237	}
2238
2239	return 0;
2240}
2241
2242static int parse_addr_size(char **inp, u64 *num, const char **str, int *idx)
2243{
2244	int err = parse_num_or_str(inp, num, str, " ");
2245
2246	if (!err && *str)
2247		err = parse_sym_idx(inp, idx);
2248
2249	return err;
2250}
2251
2252static int parse_one_filter(struct addr_filter *filt, const char **filter_inp)
2253{
2254	char *fstr;
2255	int err;
2256
2257	filt->str = fstr = strdup(*filter_inp);
2258	if (!fstr)
2259		return -ENOMEM;
2260
2261	err = parse_num_or_str(&fstr, NULL, &filt->action, " ");
2262	if (err)
2263		goto out_err;
2264
2265	err = parse_action(filt);
2266	if (err)
2267		goto out_err;
2268
2269	err = parse_addr_size(&fstr, &filt->addr, &filt->sym_from,
2270			      &filt->sym_from_idx);
2271	if (err)
2272		goto out_err;
2273
2274	fstr += strspn(fstr, " ");
2275
2276	if (*fstr == '/') {
2277		fstr += 1;
2278		err = parse_addr_size(&fstr, &filt->size, &filt->sym_to,
2279				      &filt->sym_to_idx);
2280		if (err)
2281			goto out_err;
2282		filt->range = true;
2283	}
2284
2285	fstr += strspn(fstr, " ");
2286
2287	if (*fstr == '@') {
2288		fstr += 1;
2289		err = parse_num_or_str(&fstr, NULL, &filt->filename, " ,");
2290		if (err)
2291			goto out_err;
2292	}
2293
2294	fstr += strspn(fstr, " ,");
2295
2296	*filter_inp += fstr - filt->str;
2297
2298	return 0;
2299
2300out_err:
2301	addr_filter__free_str(filt);
2302
2303	return err;
2304}
2305
2306int addr_filters__parse_bare_filter(struct addr_filters *filts,
2307				    const char *filter)
2308{
2309	struct addr_filter *filt;
2310	const char *fstr = filter;
2311	int err;
2312
2313	while (*fstr) {
2314		filt = addr_filter__new();
2315		err = parse_one_filter(filt, &fstr);
2316		if (err) {
2317			addr_filter__free(filt);
2318			addr_filters__exit(filts);
2319			return err;
2320		}
2321		addr_filters__add(filts, filt);
2322	}
2323
2324	return 0;
2325}
2326
2327struct sym_args {
2328	const char	*name;
2329	u64		start;
2330	u64		size;
2331	int		idx;
2332	int		cnt;
2333	bool		started;
2334	bool		global;
2335	bool		selected;
2336	bool		duplicate;
2337	bool		near;
2338};
2339
2340static bool kern_sym_name_match(const char *kname, const char *name)
2341{
2342	size_t n = strlen(name);
2343
2344	return !strcmp(kname, name) ||
2345	       (!strncmp(kname, name, n) && kname[n] == '\t');
2346}
2347
2348static bool kern_sym_match(struct sym_args *args, const char *name, char type)
2349{
2350	/* A function with the same name, and global or the n'th found or any */
2351	return kallsyms__is_function(type) &&
2352	       kern_sym_name_match(name, args->name) &&
2353	       ((args->global && isupper(type)) ||
2354		(args->selected && ++(args->cnt) == args->idx) ||
2355		(!args->global && !args->selected));
2356}
2357
2358static int find_kern_sym_cb(void *arg, const char *name, char type, u64 start)
2359{
2360	struct sym_args *args = arg;
2361
2362	if (args->started) {
2363		if (!args->size)
2364			args->size = start - args->start;
2365		if (args->selected) {
2366			if (args->size)
2367				return 1;
2368		} else if (kern_sym_match(args, name, type)) {
2369			args->duplicate = true;
2370			return 1;
2371		}
2372	} else if (kern_sym_match(args, name, type)) {
2373		args->started = true;
2374		args->start = start;
2375	}
2376
2377	return 0;
2378}
2379
2380static int print_kern_sym_cb(void *arg, const char *name, char type, u64 start)
2381{
2382	struct sym_args *args = arg;
2383
2384	if (kern_sym_match(args, name, type)) {
2385		pr_err("#%d\t0x%"PRIx64"\t%c\t%s\n",
2386		       ++args->cnt, start, type, name);
2387		args->near = true;
2388	} else if (args->near) {
2389		args->near = false;
2390		pr_err("\t\twhich is near\t\t%s\n", name);
2391	}
2392
2393	return 0;
2394}
2395
2396static int sym_not_found_error(const char *sym_name, int idx)
2397{
2398	if (idx > 0) {
2399		pr_err("N'th occurrence (N=%d) of symbol '%s' not found.\n",
2400		       idx, sym_name);
2401	} else if (!idx) {
2402		pr_err("Global symbol '%s' not found.\n", sym_name);
2403	} else {
2404		pr_err("Symbol '%s' not found.\n", sym_name);
2405	}
2406	pr_err("Note that symbols must be functions.\n");
2407
2408	return -EINVAL;
2409}
2410
2411static int find_kern_sym(const char *sym_name, u64 *start, u64 *size, int idx)
2412{
2413	struct sym_args args = {
2414		.name = sym_name,
2415		.idx = idx,
2416		.global = !idx,
2417		.selected = idx > 0,
2418	};
2419	int err;
2420
2421	*start = 0;
2422	*size = 0;
2423
2424	err = kallsyms__parse("/proc/kallsyms", &args, find_kern_sym_cb);
2425	if (err < 0) {
2426		pr_err("Failed to parse /proc/kallsyms\n");
2427		return err;
2428	}
2429
2430	if (args.duplicate) {
2431		pr_err("Multiple kernel symbols with name '%s'\n", sym_name);
2432		args.cnt = 0;
2433		kallsyms__parse("/proc/kallsyms", &args, print_kern_sym_cb);
2434		pr_err("Disambiguate symbol name by inserting #n after the name e.g. %s #2\n",
2435		       sym_name);
2436		pr_err("Or select a global symbol by inserting #0 or #g or #G\n");
2437		return -EINVAL;
2438	}
2439
2440	if (!args.started) {
2441		pr_err("Kernel symbol lookup: ");
2442		return sym_not_found_error(sym_name, idx);
2443	}
2444
2445	*start = args.start;
2446	*size = args.size;
2447
2448	return 0;
2449}
2450
2451static int find_entire_kern_cb(void *arg, const char *name __maybe_unused,
2452			       char type, u64 start)
2453{
2454	struct sym_args *args = arg;
2455	u64 size;
2456
2457	if (!kallsyms__is_function(type))
2458		return 0;
2459
2460	if (!args->started) {
2461		args->started = true;
2462		args->start = start;
2463	}
2464	/* Don't know exactly where the kernel ends, so we add a page */
2465	size = round_up(start, page_size) + page_size - args->start;
2466	if (size > args->size)
2467		args->size = size;
2468
2469	return 0;
2470}
2471
2472static int addr_filter__entire_kernel(struct addr_filter *filt)
2473{
2474	struct sym_args args = { .started = false };
2475	int err;
2476
2477	err = kallsyms__parse("/proc/kallsyms", &args, find_entire_kern_cb);
2478	if (err < 0 || !args.started) {
2479		pr_err("Failed to parse /proc/kallsyms\n");
2480		return err;
2481	}
2482
2483	filt->addr = args.start;
2484	filt->size = args.size;
2485
2486	return 0;
2487}
2488
2489static int check_end_after_start(struct addr_filter *filt, u64 start, u64 size)
2490{
2491	if (start + size >= filt->addr)
2492		return 0;
2493
2494	if (filt->sym_from) {
2495		pr_err("Symbol '%s' (0x%"PRIx64") comes before '%s' (0x%"PRIx64")\n",
2496		       filt->sym_to, start, filt->sym_from, filt->addr);
2497	} else {
2498		pr_err("Symbol '%s' (0x%"PRIx64") comes before address 0x%"PRIx64")\n",
2499		       filt->sym_to, start, filt->addr);
2500	}
2501
2502	return -EINVAL;
2503}
2504
2505static int addr_filter__resolve_kernel_syms(struct addr_filter *filt)
2506{
2507	bool no_size = false;
2508	u64 start, size;
2509	int err;
2510
2511	if (symbol_conf.kptr_restrict) {
2512		pr_err("Kernel addresses are restricted. Unable to resolve kernel symbols.\n");
2513		return -EINVAL;
2514	}
2515
2516	if (filt->sym_from && !strcmp(filt->sym_from, "*"))
2517		return addr_filter__entire_kernel(filt);
2518
2519	if (filt->sym_from) {
2520		err = find_kern_sym(filt->sym_from, &start, &size,
2521				    filt->sym_from_idx);
2522		if (err)
2523			return err;
2524		filt->addr = start;
2525		if (filt->range && !filt->size && !filt->sym_to) {
2526			filt->size = size;
2527			no_size = !size;
2528		}
2529	}
2530
2531	if (filt->sym_to) {
2532		err = find_kern_sym(filt->sym_to, &start, &size,
2533				    filt->sym_to_idx);
2534		if (err)
2535			return err;
2536
2537		err = check_end_after_start(filt, start, size);
2538		if (err)
2539			return err;
2540		filt->size = start + size - filt->addr;
2541		no_size = !size;
2542	}
2543
2544	/* The very last symbol in kallsyms does not imply a particular size */
2545	if (no_size) {
2546		pr_err("Cannot determine size of symbol '%s'\n",
2547		       filt->sym_to ? filt->sym_to : filt->sym_from);
2548		return -EINVAL;
2549	}
2550
2551	return 0;
2552}
2553
2554static struct dso *load_dso(const char *name)
2555{
2556	struct map *map;
2557	struct dso *dso;
2558
2559	map = dso__new_map(name);
2560	if (!map)
2561		return NULL;
2562
2563	if (map__load(map) < 0)
2564		pr_err("File '%s' not found or has no symbols.\n", name);
2565
2566	dso = dso__get(map__dso(map));
2567
2568	map__put(map);
2569
2570	return dso;
2571}
2572
2573static bool dso_sym_match(struct symbol *sym, const char *name, int *cnt,
2574			  int idx)
2575{
2576	/* Same name, and global or the n'th found or any */
2577	return !arch__compare_symbol_names(name, sym->name) &&
2578	       ((!idx && sym->binding == STB_GLOBAL) ||
2579		(idx > 0 && ++*cnt == idx) ||
2580		idx < 0);
2581}
2582
2583static void print_duplicate_syms(struct dso *dso, const char *sym_name)
2584{
2585	struct symbol *sym;
2586	bool near = false;
2587	int cnt = 0;
2588
2589	pr_err("Multiple symbols with name '%s'\n", sym_name);
2590
2591	sym = dso__first_symbol(dso);
2592	while (sym) {
2593		if (dso_sym_match(sym, sym_name, &cnt, -1)) {
2594			pr_err("#%d\t0x%"PRIx64"\t%c\t%s\n",
2595			       ++cnt, sym->start,
2596			       sym->binding == STB_GLOBAL ? 'g' :
2597			       sym->binding == STB_LOCAL  ? 'l' : 'w',
2598			       sym->name);
2599			near = true;
2600		} else if (near) {
2601			near = false;
2602			pr_err("\t\twhich is near\t\t%s\n", sym->name);
2603		}
2604		sym = dso__next_symbol(sym);
2605	}
2606
2607	pr_err("Disambiguate symbol name by inserting #n after the name e.g. %s #2\n",
2608	       sym_name);
2609	pr_err("Or select a global symbol by inserting #0 or #g or #G\n");
2610}
2611
2612static int find_dso_sym(struct dso *dso, const char *sym_name, u64 *start,
2613			u64 *size, int idx)
2614{
2615	struct symbol *sym;
2616	int cnt = 0;
2617
2618	*start = 0;
2619	*size = 0;
2620
2621	sym = dso__first_symbol(dso);
2622	while (sym) {
2623		if (*start) {
2624			if (!*size)
2625				*size = sym->start - *start;
2626			if (idx > 0) {
2627				if (*size)
2628					return 0;
2629			} else if (dso_sym_match(sym, sym_name, &cnt, idx)) {
2630				print_duplicate_syms(dso, sym_name);
2631				return -EINVAL;
2632			}
2633		} else if (dso_sym_match(sym, sym_name, &cnt, idx)) {
2634			*start = sym->start;
2635			*size = sym->end - sym->start;
2636		}
2637		sym = dso__next_symbol(sym);
2638	}
2639
2640	if (!*start)
2641		return sym_not_found_error(sym_name, idx);
2642
2643	return 0;
2644}
2645
2646static int addr_filter__entire_dso(struct addr_filter *filt, struct dso *dso)
2647{
2648	if (dso__data_file_size(dso, NULL)) {
2649		pr_err("Failed to determine filter for %s\nCannot determine file size.\n",
2650		       filt->filename);
2651		return -EINVAL;
2652	}
2653
2654	filt->addr = 0;
2655	filt->size = dso->data.file_size;
2656
2657	return 0;
2658}
2659
2660static int addr_filter__resolve_syms(struct addr_filter *filt)
2661{
2662	u64 start, size;
2663	struct dso *dso;
2664	int err = 0;
2665
2666	if (!filt->sym_from && !filt->sym_to)
2667		return 0;
2668
2669	if (!filt->filename)
2670		return addr_filter__resolve_kernel_syms(filt);
2671
2672	dso = load_dso(filt->filename);
2673	if (!dso) {
2674		pr_err("Failed to load symbols from: %s\n", filt->filename);
2675		return -EINVAL;
2676	}
2677
2678	if (filt->sym_from && !strcmp(filt->sym_from, "*")) {
2679		err = addr_filter__entire_dso(filt, dso);
2680		goto put_dso;
2681	}
2682
2683	if (filt->sym_from) {
2684		err = find_dso_sym(dso, filt->sym_from, &start, &size,
2685				   filt->sym_from_idx);
2686		if (err)
2687			goto put_dso;
2688		filt->addr = start;
2689		if (filt->range && !filt->size && !filt->sym_to)
2690			filt->size = size;
2691	}
2692
2693	if (filt->sym_to) {
2694		err = find_dso_sym(dso, filt->sym_to, &start, &size,
2695				   filt->sym_to_idx);
2696		if (err)
2697			goto put_dso;
2698
2699		err = check_end_after_start(filt, start, size);
2700		if (err)
2701			return err;
2702
2703		filt->size = start + size - filt->addr;
2704	}
2705
2706put_dso:
2707	dso__put(dso);
2708
2709	return err;
2710}
2711
2712static char *addr_filter__to_str(struct addr_filter *filt)
2713{
2714	char filename_buf[PATH_MAX];
2715	const char *at = "";
2716	const char *fn = "";
2717	char *filter;
2718	int err;
2719
2720	if (filt->filename) {
2721		at = "@";
2722		fn = realpath(filt->filename, filename_buf);
2723		if (!fn)
2724			return NULL;
2725	}
2726
2727	if (filt->range) {
2728		err = asprintf(&filter, "%s 0x%"PRIx64"/0x%"PRIx64"%s%s",
2729			       filt->action, filt->addr, filt->size, at, fn);
2730	} else {
2731		err = asprintf(&filter, "%s 0x%"PRIx64"%s%s",
2732			       filt->action, filt->addr, at, fn);
2733	}
2734
2735	return err < 0 ? NULL : filter;
2736}
2737
2738static int parse_addr_filter(struct evsel *evsel, const char *filter,
2739			     int max_nr)
2740{
2741	struct addr_filters filts;
2742	struct addr_filter *filt;
2743	int err;
2744
2745	addr_filters__init(&filts);
2746
2747	err = addr_filters__parse_bare_filter(&filts, filter);
2748	if (err)
2749		goto out_exit;
2750
2751	if (filts.cnt > max_nr) {
2752		pr_err("Error: number of address filters (%d) exceeds maximum (%d)\n",
2753		       filts.cnt, max_nr);
2754		err = -EINVAL;
2755		goto out_exit;
2756	}
2757
2758	list_for_each_entry(filt, &filts.head, list) {
2759		char *new_filter;
2760
2761		err = addr_filter__resolve_syms(filt);
2762		if (err)
2763			goto out_exit;
2764
2765		new_filter = addr_filter__to_str(filt);
2766		if (!new_filter) {
2767			err = -ENOMEM;
2768			goto out_exit;
2769		}
2770
2771		if (evsel__append_addr_filter(evsel, new_filter)) {
2772			err = -ENOMEM;
2773			goto out_exit;
2774		}
2775	}
2776
2777out_exit:
2778	addr_filters__exit(&filts);
2779
2780	if (err) {
2781		pr_err("Failed to parse address filter: '%s'\n", filter);
2782		pr_err("Filter format is: filter|start|stop|tracestop <start symbol or address> [/ <end symbol or size>] [@<file name>]\n");
2783		pr_err("Where multiple filters are separated by space or comma.\n");
2784	}
2785
2786	return err;
2787}
2788
2789static int evsel__nr_addr_filter(struct evsel *evsel)
2790{
2791	struct perf_pmu *pmu = evsel__find_pmu(evsel);
2792	int nr_addr_filters = 0;
2793
2794	if (!pmu)
2795		return 0;
2796
2797	perf_pmu__scan_file(pmu, "nr_addr_filters", "%d", &nr_addr_filters);
2798
2799	return nr_addr_filters;
2800}
2801
2802int auxtrace_parse_filters(struct evlist *evlist)
2803{
2804	struct evsel *evsel;
2805	char *filter;
2806	int err, max_nr;
2807
2808	evlist__for_each_entry(evlist, evsel) {
2809		filter = evsel->filter;
2810		max_nr = evsel__nr_addr_filter(evsel);
2811		if (!filter || !max_nr)
2812			continue;
2813		evsel->filter = NULL;
2814		err = parse_addr_filter(evsel, filter, max_nr);
2815		free(filter);
2816		if (err)
2817			return err;
2818		pr_debug("Address filter: %s\n", evsel->filter);
2819	}
2820
2821	return 0;
2822}
2823
2824int auxtrace__process_event(struct perf_session *session, union perf_event *event,
2825			    struct perf_sample *sample, struct perf_tool *tool)
2826{
2827	if (!session->auxtrace)
2828		return 0;
2829
2830	return session->auxtrace->process_event(session, event, sample, tool);
2831}
2832
2833void auxtrace__dump_auxtrace_sample(struct perf_session *session,
2834				    struct perf_sample *sample)
2835{
2836	if (!session->auxtrace || !session->auxtrace->dump_auxtrace_sample ||
2837	    auxtrace__dont_decode(session))
2838		return;
2839
2840	session->auxtrace->dump_auxtrace_sample(session, sample);
2841}
2842
2843int auxtrace__flush_events(struct perf_session *session, struct perf_tool *tool)
2844{
2845	if (!session->auxtrace)
2846		return 0;
2847
2848	return session->auxtrace->flush_events(session, tool);
2849}
2850
2851void auxtrace__free_events(struct perf_session *session)
2852{
2853	if (!session->auxtrace)
2854		return;
2855
2856	return session->auxtrace->free_events(session);
2857}
2858
2859void auxtrace__free(struct perf_session *session)
2860{
2861	if (!session->auxtrace)
2862		return;
2863
2864	return session->auxtrace->free(session);
2865}
2866
2867bool auxtrace__evsel_is_auxtrace(struct perf_session *session,
2868				 struct evsel *evsel)
2869{
2870	if (!session->auxtrace || !session->auxtrace->evsel_is_auxtrace)
2871		return false;
2872
2873	return session->auxtrace->evsel_is_auxtrace(session, evsel);
2874}