tools/perf/util/auxtrace.c

   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  */
  68 static 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
 109 static bool auxtrace__dont_decode(struct perf_session *session)
 110 {
 111         return !session->itrace_synth_opts ||
 112                session->itrace_synth_opts->dont_decode;
 113 }
 114
 115 int 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
 149 void 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
 157 void 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
 173 void 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__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
 200 static 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
 221 int 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
 230 static 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
 265 static 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
 287 static 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
 321 static 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
 352 static 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
 359 static 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
 399 out_free:
 400         auxtrace_buffer__free(buffer);
 401         return err;
 402 }
 403
 404 int 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
 424 static 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         }
 447 out:
 448         return err;
 449 }
 450
 451 void 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
 470 static 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
 487 int 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
 510 void 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
 517 void 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
 555 size_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
 563 static int auxtrace_not_supported(void)
 564 {
 565         pr_err("AUX area tracing is not supported on this architecture\n");
 566         return -EINVAL;
 567 }
 568
 569 int 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
 579 void auxtrace_record__free(struct auxtrace_record *itr)
 580 {
 581         if (itr)
 582                 itr->free(itr);
 583 }
 584
 585 int 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
 592 int 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
 599 int 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
 608 int 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
 619 u64 auxtrace_record__reference(struct auxtrace_record *itr)
 620 {
 621         if (itr)
 622                 return itr->reference(itr);
 623         return 0;
 624 }
 625
 626 int 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
 649 static int evlist__enable_event_idx(struct evlist *evlist, struct evsel *evsel, int idx)
 650 {
 651         bool per_cpu_mmaps = !perf_cpu_map__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
 665 int 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
 692 static 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
 741 int 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         }
 781 no_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
 808 void 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
 823 struct auxtrace_record *__weak
 824 auxtrace_record__init(struct evlist *evlist __maybe_unused, int *err)
 825 {
 826         *err = 0;
 827         return NULL;
 828 }
 829
 830 static 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
 846 void 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
 856 static 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
 880 int 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
 898 static 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
 913 int 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
 934 static 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
 963 int 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
 989 static 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
 997 int 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
1021 struct 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
1037 struct 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
1061 int 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
1090 struct queue_data {
1091         bool samples;
1092         bool events;
1093 };
1094
1095 static 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
1126 int 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
1147 void *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
1170 void 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
1181 void 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
1192 void auxtrace_buffer__free(struct auxtrace_buffer *buffer)
1193 {
1194         auxtrace_buffer__drop_data(buffer);
1195         free(buffer);
1196 }
1197
1198 void 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
1229 void 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
1237 int 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);
1261 out_free:
1262         free(ev);
1263         return err;
1264 }
1265
1266 static 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
1291 static 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
1303 int 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
1344 s64 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
1375 void 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
1406 static 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
1424 static 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
1449 static 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  */
1462 int 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 ' ':
1642                 case ',':
1643                         break;
1644                 default:
1645                         goto out_err;
1646                 }
1647         }
1648 out:
1649         if (synth_opts->instructions || synth_opts->cycles) {
1650                 if (!period_type_set)
1651                         synth_opts->period_type =
1652                                         PERF_ITRACE_DEFAULT_PERIOD_TYPE;
1653                 if (!period_set)
1654                         synth_opts->period = PERF_ITRACE_DEFAULT_PERIOD;
1655         }
1656
1657         return 0;
1658
1659 out_err:
1660         pr_err("Bad Instruction Tracing options '%s'\n", str);
1661         return -EINVAL;
1662 }
1663
1664 int itrace_parse_synth_opts(const struct option *opt, const char *str, int unset)
1665 {
1666         return itrace_do_parse_synth_opts(opt->value, str, unset);
1667 }
1668
1669 static const char * const auxtrace_error_type_name[] = {
1670         [PERF_AUXTRACE_ERROR_ITRACE] = "instruction trace",
1671 };
1672
1673 static const char *auxtrace_error_name(int type)
1674 {
1675         const char *error_type_name = NULL;
1676
1677         if (type < PERF_AUXTRACE_ERROR_MAX)
1678                 error_type_name = auxtrace_error_type_name[type];
1679         if (!error_type_name)
1680                 error_type_name = "unknown AUX";
1681         return error_type_name;
1682 }
1683
1684 size_t perf_event__fprintf_auxtrace_error(union perf_event *event, FILE *fp)
1685 {
1686         struct perf_record_auxtrace_error *e = &event->auxtrace_error;
1687         unsigned long long nsecs = e->time;
1688         const char *msg = e->msg;
1689         int ret;
1690
1691         ret = fprintf(fp, " %s error type %u",
1692                       auxtrace_error_name(e->type), e->type);
1693
1694         if (e->fmt && nsecs) {
1695                 unsigned long secs = nsecs / NSEC_PER_SEC;
1696
1697                 nsecs -= secs * NSEC_PER_SEC;
1698                 ret += fprintf(fp, " time %lu.%09llu", secs, nsecs);
1699         } else {
1700                 ret += fprintf(fp, " time 0");
1701         }
1702
1703         if (!e->fmt)
1704                 msg = (const char *)&e->time;
1705
1706         if (e->fmt >= 2 && e->machine_pid)
1707                 ret += fprintf(fp, " machine_pid %d vcpu %d", e->machine_pid, e->vcpu);
1708
1709         ret += fprintf(fp, " cpu %d pid %d tid %d ip %#"PRI_lx64" code %u: %s\n",
1710                        e->cpu, e->pid, e->tid, e->ip, e->code, msg);
1711         return ret;
1712 }
1713
1714 void perf_session__auxtrace_error_inc(struct perf_session *session,
1715                                       union perf_event *event)
1716 {
1717         struct perf_record_auxtrace_error *e = &event->auxtrace_error;
1718
1719         if (e->type < PERF_AUXTRACE_ERROR_MAX)
1720                 session->evlist->stats.nr_auxtrace_errors[e->type] += 1;
1721 }
1722
1723 void events_stats__auxtrace_error_warn(const struct events_stats *stats)
1724 {
1725         int i;
1726
1727         for (i = 0; i < PERF_AUXTRACE_ERROR_MAX; i++) {
1728                 if (!stats->nr_auxtrace_errors[i])
1729                         continue;
1730                 ui__warning("%u %s errors\n",
1731                             stats->nr_auxtrace_errors[i],
1732                             auxtrace_error_name(i));
1733         }
1734 }
1735
1736 int perf_event__process_auxtrace_error(struct perf_session *session,
1737                                        union perf_event *event)
1738 {
1739         if (auxtrace__dont_decode(session))
1740                 return 0;
1741
1742         perf_event__fprintf_auxtrace_error(event, stdout);
1743         return 0;
1744 }
1745
1746 /*
1747  * In the compat mode kernel runs in 64-bit and perf tool runs in 32-bit mode,
1748  * 32-bit perf tool cannot access 64-bit value atomically, which might lead to
1749  * the issues caused by the below sequence on multiple CPUs: when perf tool
1750  * accesses either the load operation or the store operation for 64-bit value,
1751  * on some architectures the operation is divided into two instructions, one
1752  * is for accessing the low 32-bit value and another is for the high 32-bit;
1753  * thus these two user operations can give the kernel chances to access the
1754  * 64-bit value, and thus leads to the unexpected load values.
1755  *
1756  *   kernel (64-bit)                        user (32-bit)
1757  *
1758  *   if (LOAD ->aux_tail) { --,             LOAD ->aux_head_lo
1759  *       STORE $aux_data      |       ,--->
1760  *       FLUSH $aux_data      |       |     LOAD ->aux_head_hi
1761  *       STORE ->aux_head   --|-------`     smp_rmb()
1762  *   }                        |             LOAD $data
1763  *                            |             smp_mb()
1764  *                            |             STORE ->aux_tail_lo
1765  *                            `----------->
1766  *                                          STORE ->aux_tail_hi
1767  *
1768  * For this reason, it's impossible for the perf tool to work correctly when
1769  * the AUX head or tail is bigger than 4GB (more than 32 bits length); and we
1770  * can not simply limit the AUX ring buffer to less than 4GB, the reason is
1771  * the pointers can be increased monotonically, whatever the buffer size it is,
1772  * at the end the head and tail can be bigger than 4GB and carry out to the
1773  * high 32-bit.
1774  *
1775  * To mitigate the issues and improve the user experience, we can allow the
1776  * perf tool working in certain conditions and bail out with error if detect
1777  * any overflow cannot be handled.
1778  *
1779  * For reading the AUX head, it reads out the values for three times, and
1780  * compares the high 4 bytes of the values between the first time and the last
1781  * time, if there has no change for high 4 bytes injected by the kernel during
1782  * the user reading sequence, it's safe for use the second value.
1783  *
1784  * When compat_auxtrace_mmap__write_tail() detects any carrying in the high
1785  * 32 bits, it means there have two store operations in user space and it cannot
1786  * promise the atomicity for 64-bit write, so return '-1' in this case to tell
1787  * the caller an overflow error has happened.
1788  */
1789 u64 __weak compat_auxtrace_mmap__read_head(struct auxtrace_mmap *mm)
1790 {
1791         struct perf_event_mmap_page *pc = mm->userpg;
1792         u64 first, second, last;
1793         u64 mask = (u64)(UINT32_MAX) << 32;
1794
1795         do {
1796                 first = READ_ONCE(pc->aux_head);
1797                 /* Ensure all reads are done after we read the head */
1798                 smp_rmb();
1799                 second = READ_ONCE(pc->aux_head);
1800                 /* Ensure all reads are done after we read the head */
1801                 smp_rmb();
1802                 last = READ_ONCE(pc->aux_head);
1803         } while ((first & mask) != (last & mask));
1804
1805         return second;
1806 }
1807
1808 int __weak compat_auxtrace_mmap__write_tail(struct auxtrace_mmap *mm, u64 tail)
1809 {
1810         struct perf_event_mmap_page *pc = mm->userpg;
1811         u64 mask = (u64)(UINT32_MAX) << 32;
1812
1813         if (tail & mask)
1814                 return -1;
1815
1816         /* Ensure all reads are done before we write the tail out */
1817         smp_mb();
1818         WRITE_ONCE(pc->aux_tail, tail);
1819         return 0;
1820 }
1821
1822 static int __auxtrace_mmap__read(struct mmap *map,
1823                                  struct auxtrace_record *itr,
1824                                  struct perf_tool *tool, process_auxtrace_t fn,
1825                                  bool snapshot, size_t snapshot_size)
1826 {
1827         struct auxtrace_mmap *mm = &map->auxtrace_mmap;
1828         u64 head, old = mm->prev, offset, ref;
1829         unsigned char *data = mm->base;
1830         size_t size, head_off, old_off, len1, len2, padding;
1831         union perf_event ev;
1832         void *data1, *data2;
1833         int kernel_is_64_bit = perf_env__kernel_is_64_bit(evsel__env(NULL));
1834
1835         head = auxtrace_mmap__read_head(mm, kernel_is_64_bit);
1836
1837         if (snapshot &&
1838             auxtrace_record__find_snapshot(itr, mm->idx, mm, data, &head, &old))
1839                 return -1;
1840
1841         if (old == head)
1842                 return 0;
1843
1844         pr_debug3("auxtrace idx %d old %#"PRIx64" head %#"PRIx64" diff %#"PRIx64"\n",
1845                   mm->idx, old, head, head - old);
1846
1847         if (mm->mask) {
1848                 head_off = head & mm->mask;
1849                 old_off = old & mm->mask;
1850         } else {
1851                 head_off = head % mm->len;
1852                 old_off = old % mm->len;
1853         }
1854
1855         if (head_off > old_off)
1856                 size = head_off - old_off;
1857         else
1858                 size = mm->len - (old_off - head_off);
1859
1860         if (snapshot && size > snapshot_size)
1861                 size = snapshot_size;
1862
1863         ref = auxtrace_record__reference(itr);
1864
1865         if (head > old || size <= head || mm->mask) {
1866                 offset = head - size;
1867         } else {
1868                 /*
1869                  * When the buffer size is not a power of 2, 'head' wraps at the
1870                  * highest multiple of the buffer size, so we have to subtract
1871                  * the remainder here.
1872                  */
1873                 u64 rem = (0ULL - mm->len) % mm->len;
1874
1875                 offset = head - size - rem;
1876         }
1877
1878         if (size > head_off) {
1879                 len1 = size - head_off;
1880                 data1 = &data[mm->len - len1];
1881                 len2 = head_off;
1882                 data2 = &data[0];
1883         } else {
1884                 len1 = size;
1885                 data1 = &data[head_off - len1];
1886                 len2 = 0;
1887                 data2 = NULL;
1888         }
1889
1890         if (itr->alignment) {
1891                 unsigned int unwanted = len1 % itr->alignment;
1892
1893                 len1 -= unwanted;
1894                 size -= unwanted;
1895         }
1896
1897         /* padding must be written by fn() e.g. record__process_auxtrace() */
1898         padding = size & (PERF_AUXTRACE_RECORD_ALIGNMENT - 1);
1899         if (padding)
1900                 padding = PERF_AUXTRACE_RECORD_ALIGNMENT - padding;
1901
1902         memset(&ev, 0, sizeof(ev));
1903         ev.auxtrace.header.type = PERF_RECORD_AUXTRACE;
1904         ev.auxtrace.header.size = sizeof(ev.auxtrace);
1905         ev.auxtrace.size = size + padding;
1906         ev.auxtrace.offset = offset;
1907         ev.auxtrace.reference = ref;
1908         ev.auxtrace.idx = mm->idx;
1909         ev.auxtrace.tid = mm->tid;
1910         ev.auxtrace.cpu = mm->cpu;
1911
1912         if (fn(tool, map, &ev, data1, len1, data2, len2))
1913                 return -1;
1914
1915         mm->prev = head;
1916
1917         if (!snapshot) {
1918                 int err;
1919
1920                 err = auxtrace_mmap__write_tail(mm, head, kernel_is_64_bit);
1921                 if (err < 0)
1922                         return err;
1923
1924                 if (itr->read_finish) {
1925                         err = itr->read_finish(itr, mm->idx);
1926                         if (err < 0)
1927                                 return err;
1928                 }
1929         }
1930
1931         return 1;
1932 }
1933
1934 int auxtrace_mmap__read(struct mmap *map, struct auxtrace_record *itr,
1935                         struct perf_tool *tool, process_auxtrace_t fn)
1936 {
1937         return __auxtrace_mmap__read(map, itr, tool, fn, false, 0);
1938 }
1939
1940 int auxtrace_mmap__read_snapshot(struct mmap *map,
1941                                  struct auxtrace_record *itr,
1942                                  struct perf_tool *tool, process_auxtrace_t fn,
1943                                  size_t snapshot_size)
1944 {
1945         return __auxtrace_mmap__read(map, itr, tool, fn, true, snapshot_size);
1946 }
1947
1948 /**
1949  * struct auxtrace_cache - hash table to implement a cache
1950  * @hashtable: the hashtable
1951  * @sz: hashtable size (number of hlists)
1952  * @entry_size: size of an entry
1953  * @limit: limit the number of entries to this maximum, when reached the cache
1954  *         is dropped and caching begins again with an empty cache
1955  * @cnt: current number of entries
1956  * @bits: hashtable size (@sz = 2^@bits)
1957  */
1958 struct auxtrace_cache {
1959         struct hlist_head *hashtable;
1960         size_t sz;
1961         size_t entry_size;
1962         size_t limit;
1963         size_t cnt;
1964         unsigned int bits;
1965 };
1966
1967 struct auxtrace_cache *auxtrace_cache__new(unsigned int bits, size_t entry_size,
1968                                            unsigned int limit_percent)
1969 {
1970         struct auxtrace_cache *c;
1971         struct hlist_head *ht;
1972         size_t sz, i;
1973
1974         c = zalloc(sizeof(struct auxtrace_cache));
1975         if (!c)
1976                 return NULL;
1977
1978         sz = 1UL << bits;
1979
1980         ht = calloc(sz, sizeof(struct hlist_head));
1981         if (!ht)
1982                 goto out_free;
1983
1984         for (i = 0; i < sz; i++)
1985                 INIT_HLIST_HEAD(&ht[i]);
1986
1987         c->hashtable = ht;
1988         c->sz = sz;
1989         c->entry_size = entry_size;
1990         c->limit = (c->sz * limit_percent) / 100;
1991         c->bits = bits;
1992
1993         return c;
1994
1995 out_free:
1996         free(c);
1997         return NULL;
1998 }
1999
2000 static void auxtrace_cache__drop(struct auxtrace_cache *c)
2001 {
2002         struct auxtrace_cache_entry *entry;
2003         struct hlist_node *tmp;
2004         size_t i;
2005
2006         if (!c)
2007                 return;
2008
2009         for (i = 0; i < c->sz; i++) {
2010                 hlist_for_each_entry_safe(entry, tmp, &c->hashtable[i], hash) {
2011                         hlist_del(&entry->hash);
2012                         auxtrace_cache__free_entry(c, entry);
2013                 }
2014         }
2015
2016         c->cnt = 0;
2017 }
2018
2019 void auxtrace_cache__free(struct auxtrace_cache *c)
2020 {
2021         if (!c)
2022                 return;
2023
2024         auxtrace_cache__drop(c);
2025         zfree(&c->hashtable);
2026         free(c);
2027 }
2028
2029 void *auxtrace_cache__alloc_entry(struct auxtrace_cache *c)
2030 {
2031         return malloc(c->entry_size);
2032 }
2033
2034 void auxtrace_cache__free_entry(struct auxtrace_cache *c __maybe_unused,
2035                                 void *entry)
2036 {
2037         free(entry);
2038 }
2039
2040 int auxtrace_cache__add(struct auxtrace_cache *c, u32 key,
2041                         struct auxtrace_cache_entry *entry)
2042 {
2043         if (c->limit && ++c->cnt > c->limit)
2044                 auxtrace_cache__drop(c);
2045
2046         entry->key = key;
2047         hlist_add_head(&entry->hash, &c->hashtable[hash_32(key, c->bits)]);
2048
2049         return 0;
2050 }
2051
2052 static struct auxtrace_cache_entry *auxtrace_cache__rm(struct auxtrace_cache *c,
2053                                                        u32 key)
2054 {
2055         struct auxtrace_cache_entry *entry;
2056         struct hlist_head *hlist;
2057         struct hlist_node *n;
2058
2059         if (!c)
2060                 return NULL;
2061
2062         hlist = &c->hashtable[hash_32(key, c->bits)];
2063         hlist_for_each_entry_safe(entry, n, hlist, hash) {
2064                 if (entry->key == key) {
2065                         hlist_del(&entry->hash);
2066                         return entry;
2067                 }
2068         }
2069
2070         return NULL;
2071 }
2072
2073 void auxtrace_cache__remove(struct auxtrace_cache *c, u32 key)
2074 {
2075         struct auxtrace_cache_entry *entry = auxtrace_cache__rm(c, key);
2076
2077         auxtrace_cache__free_entry(c, entry);
2078 }
2079
2080 void *auxtrace_cache__lookup(struct auxtrace_cache *c, u32 key)
2081 {
2082         struct auxtrace_cache_entry *entry;
2083         struct hlist_head *hlist;
2084
2085         if (!c)
2086                 return NULL;
2087
2088         hlist = &c->hashtable[hash_32(key, c->bits)];
2089         hlist_for_each_entry(entry, hlist, hash) {
2090                 if (entry->key == key)
2091                         return entry;
2092         }
2093
2094         return NULL;
2095 }
2096
2097 static void addr_filter__free_str(struct addr_filter *filt)
2098 {
2099         zfree(&filt->str);
2100         filt->action   = NULL;
2101         filt->sym_from = NULL;
2102         filt->sym_to   = NULL;
2103         filt->filename = NULL;
2104 }
2105
2106 static struct addr_filter *addr_filter__new(void)
2107 {
2108         struct addr_filter *filt = zalloc(sizeof(*filt));
2109
2110         if (filt)
2111                 INIT_LIST_HEAD(&filt->list);
2112
2113         return filt;
2114 }
2115
2116 static void addr_filter__free(struct addr_filter *filt)
2117 {
2118         if (filt)
2119                 addr_filter__free_str(filt);
2120         free(filt);
2121 }
2122
2123 static void addr_filters__add(struct addr_filters *filts,
2124                               struct addr_filter *filt)
2125 {
2126         list_add_tail(&filt->list, &filts->head);
2127         filts->cnt += 1;
2128 }
2129
2130 static void addr_filters__del(struct addr_filters *filts,
2131                               struct addr_filter *filt)
2132 {
2133         list_del_init(&filt->list);
2134         filts->cnt -= 1;
2135 }
2136
2137 void addr_filters__init(struct addr_filters *filts)
2138 {
2139         INIT_LIST_HEAD(&filts->head);
2140         filts->cnt = 0;
2141 }
2142
2143 void addr_filters__exit(struct addr_filters *filts)
2144 {
2145         struct addr_filter *filt, *n;
2146
2147         list_for_each_entry_safe(filt, n, &filts->head, list) {
2148                 addr_filters__del(filts, filt);
2149                 addr_filter__free(filt);
2150         }
2151 }
2152
2153 static int parse_num_or_str(char **inp, u64 *num, const char **str,
2154                             const char *str_delim)
2155 {
2156         *inp += strspn(*inp, " ");
2157
2158         if (isdigit(**inp)) {
2159                 char *endptr;
2160
2161                 if (!num)
2162                         return -EINVAL;
2163                 errno = 0;
2164                 *num = strtoull(*inp, &endptr, 0);
2165                 if (errno)
2166                         return -errno;
2167                 if (endptr == *inp)
2168                         return -EINVAL;
2169                 *inp = endptr;
2170         } else {
2171                 size_t n;
2172
2173                 if (!str)
2174                         return -EINVAL;
2175                 *inp += strspn(*inp, " ");
2176                 *str = *inp;
2177                 n = strcspn(*inp, str_delim);
2178                 if (!n)
2179                         return -EINVAL;
2180                 *inp += n;
2181                 if (**inp) {
2182                         **inp = '\0';
2183                         *inp += 1;
2184                 }
2185         }
2186         return 0;
2187 }
2188
2189 static int parse_action(struct addr_filter *filt)
2190 {
2191         if (!strcmp(filt->action, "filter")) {
2192                 filt->start = true;
2193                 filt->range = true;
2194         } else if (!strcmp(filt->action, "start")) {
2195                 filt->start = true;
2196         } else if (!strcmp(filt->action, "stop")) {
2197                 filt->start = false;
2198         } else if (!strcmp(filt->action, "tracestop")) {
2199                 filt->start = false;
2200                 filt->range = true;
2201                 filt->action += 5; /* Change 'tracestop' to 'stop' */
2202         } else {
2203                 return -EINVAL;
2204         }
2205         return 0;
2206 }
2207
2208 static int parse_sym_idx(char **inp, int *idx)
2209 {
2210         *idx = -1;
2211
2212         *inp += strspn(*inp, " ");
2213
2214         if (**inp != '#')
2215                 return 0;
2216
2217         *inp += 1;
2218
2219         if (**inp == 'g' || **inp == 'G') {
2220                 *inp += 1;
2221                 *idx = 0;
2222         } else {
2223                 unsigned long num;
2224                 char *endptr;
2225
2226                 errno = 0;
2227                 num = strtoul(*inp, &endptr, 0);
2228                 if (errno)
2229                         return -errno;
2230                 if (endptr == *inp || num > INT_MAX)
2231                         return -EINVAL;
2232                 *inp = endptr;
2233                 *idx = num;
2234         }
2235
2236         return 0;
2237 }
2238
2239 static int parse_addr_size(char **inp, u64 *num, const char **str, int *idx)
2240 {
2241         int err = parse_num_or_str(inp, num, str, " ");
2242
2243         if (!err && *str)
2244                 err = parse_sym_idx(inp, idx);
2245
2246         return err;
2247 }
2248
2249 static int parse_one_filter(struct addr_filter *filt, const char **filter_inp)
2250 {
2251         char *fstr;
2252         int err;
2253
2254         filt->str = fstr = strdup(*filter_inp);
2255         if (!fstr)
2256                 return -ENOMEM;
2257
2258         err = parse_num_or_str(&fstr, NULL, &filt->action, " ");
2259         if (err)
2260                 goto out_err;
2261
2262         err = parse_action(filt);
2263         if (err)
2264                 goto out_err;
2265
2266         err = parse_addr_size(&fstr, &filt->addr, &filt->sym_from,
2267                               &filt->sym_from_idx);
2268         if (err)
2269                 goto out_err;
2270
2271         fstr += strspn(fstr, " ");
2272
2273         if (*fstr == '/') {
2274                 fstr += 1;
2275                 err = parse_addr_size(&fstr, &filt->size, &filt->sym_to,
2276                                       &filt->sym_to_idx);
2277                 if (err)
2278                         goto out_err;
2279                 filt->range = true;
2280         }
2281
2282         fstr += strspn(fstr, " ");
2283
2284         if (*fstr == '@') {
2285                 fstr += 1;
2286                 err = parse_num_or_str(&fstr, NULL, &filt->filename, " ,");
2287                 if (err)
2288                         goto out_err;
2289         }
2290
2291         fstr += strspn(fstr, " ,");
2292
2293         *filter_inp += fstr - filt->str;
2294
2295         return 0;
2296
2297 out_err:
2298         addr_filter__free_str(filt);
2299
2300         return err;
2301 }
2302
2303 int addr_filters__parse_bare_filter(struct addr_filters *filts,
2304                                     const char *filter)
2305 {
2306         struct addr_filter *filt;
2307         const char *fstr = filter;
2308         int err;
2309
2310         while (*fstr) {
2311                 filt = addr_filter__new();
2312                 err = parse_one_filter(filt, &fstr);
2313                 if (err) {
2314                         addr_filter__free(filt);
2315                         addr_filters__exit(filts);
2316                         return err;
2317                 }
2318                 addr_filters__add(filts, filt);
2319         }
2320
2321         return 0;
2322 }
2323
2324 struct sym_args {
2325         const char      *name;
2326         u64             start;
2327         u64             size;
2328         int             idx;
2329         int             cnt;
2330         bool            started;
2331         bool            global;
2332         bool            selected;
2333         bool            duplicate;
2334         bool            near;
2335 };
2336
2337 static bool kern_sym_name_match(const char *kname, const char *name)
2338 {
2339         size_t n = strlen(name);
2340
2341         return !strcmp(kname, name) ||
2342                (!strncmp(kname, name, n) && kname[n] == '\t');
2343 }
2344
2345 static bool kern_sym_match(struct sym_args *args, const char *name, char type)
2346 {
2347         /* A function with the same name, and global or the n'th found or any */
2348         return kallsyms__is_function(type) &&
2349                kern_sym_name_match(name, args->name) &&
2350                ((args->global && isupper(type)) ||
2351                 (args->selected && ++(args->cnt) == args->idx) ||
2352                 (!args->global && !args->selected));
2353 }
2354
2355 static int find_kern_sym_cb(void *arg, const char *name, char type, u64 start)
2356 {
2357         struct sym_args *args = arg;
2358
2359         if (args->started) {
2360                 if (!args->size)
2361                         args->size = start - args->start;
2362                 if (args->selected) {
2363                         if (args->size)
2364                                 return 1;
2365                 } else if (kern_sym_match(args, name, type)) {
2366                         args->duplicate = true;
2367                         return 1;
2368                 }
2369         } else if (kern_sym_match(args, name, type)) {
2370                 args->started = true;
2371                 args->start = start;
2372         }
2373
2374         return 0;
2375 }
2376
2377 static int print_kern_sym_cb(void *arg, const char *name, char type, u64 start)
2378 {
2379         struct sym_args *args = arg;
2380
2381         if (kern_sym_match(args, name, type)) {
2382                 pr_err("#%d\t0x%"PRIx64"\t%c\t%s\n",
2383                        ++args->cnt, start, type, name);
2384                 args->near = true;
2385         } else if (args->near) {
2386                 args->near = false;
2387                 pr_err("\t\twhich is near\t\t%s\n", name);
2388         }
2389
2390         return 0;
2391 }
2392
2393 static int sym_not_found_error(const char *sym_name, int idx)
2394 {
2395         if (idx > 0) {
2396                 pr_err("N'th occurrence (N=%d) of symbol '%s' not found.\n",
2397                        idx, sym_name);
2398         } else if (!idx) {
2399                 pr_err("Global symbol '%s' not found.\n", sym_name);
2400         } else {
2401                 pr_err("Symbol '%s' not found.\n", sym_name);
2402         }
2403         pr_err("Note that symbols must be functions.\n");
2404
2405         return -EINVAL;
2406 }
2407
2408 static int find_kern_sym(const char *sym_name, u64 *start, u64 *size, int idx)
2409 {
2410         struct sym_args args = {
2411                 .name = sym_name,
2412                 .idx = idx,
2413                 .global = !idx,
2414                 .selected = idx > 0,
2415         };
2416         int err;
2417
2418         *start = 0;
2419         *size = 0;
2420
2421         err = kallsyms__parse("/proc/kallsyms", &args, find_kern_sym_cb);
2422         if (err < 0) {
2423                 pr_err("Failed to parse /proc/kallsyms\n");
2424                 return err;
2425         }
2426
2427         if (args.duplicate) {
2428                 pr_err("Multiple kernel symbols with name '%s'\n", sym_name);
2429                 args.cnt = 0;
2430                 kallsyms__parse("/proc/kallsyms", &args, print_kern_sym_cb);
2431                 pr_err("Disambiguate symbol name by inserting #n after the name e.g. %s #2\n",
2432                        sym_name);
2433                 pr_err("Or select a global symbol by inserting #0 or #g or #G\n");
2434                 return -EINVAL;
2435         }
2436
2437         if (!args.started) {
2438                 pr_err("Kernel symbol lookup: ");
2439                 return sym_not_found_error(sym_name, idx);
2440         }
2441
2442         *start = args.start;
2443         *size = args.size;
2444
2445         return 0;
2446 }
2447
2448 static int find_entire_kern_cb(void *arg, const char *name __maybe_unused,
2449                                char type, u64 start)
2450 {
2451         struct sym_args *args = arg;
2452         u64 size;
2453
2454         if (!kallsyms__is_function(type))
2455                 return 0;
2456
2457         if (!args->started) {
2458                 args->started = true;
2459                 args->start = start;
2460         }
2461         /* Don't know exactly where the kernel ends, so we add a page */
2462         size = round_up(start, page_size) + page_size - args->start;
2463         if (size > args->size)
2464                 args->size = size;
2465
2466         return 0;
2467 }
2468
2469 static int addr_filter__entire_kernel(struct addr_filter *filt)
2470 {
2471         struct sym_args args = { .started = false };
2472         int err;
2473
2474         err = kallsyms__parse("/proc/kallsyms", &args, find_entire_kern_cb);
2475         if (err < 0 || !args.started) {
2476                 pr_err("Failed to parse /proc/kallsyms\n");
2477                 return err;
2478         }
2479
2480         filt->addr = args.start;
2481         filt->size = args.size;
2482
2483         return 0;
2484 }
2485
2486 static int check_end_after_start(struct addr_filter *filt, u64 start, u64 size)
2487 {
2488         if (start + size >= filt->addr)
2489                 return 0;
2490
2491         if (filt->sym_from) {
2492                 pr_err("Symbol '%s' (0x%"PRIx64") comes before '%s' (0x%"PRIx64")\n",
2493                        filt->sym_to, start, filt->sym_from, filt->addr);
2494         } else {
2495                 pr_err("Symbol '%s' (0x%"PRIx64") comes before address 0x%"PRIx64")\n",
2496                        filt->sym_to, start, filt->addr);
2497         }
2498
2499         return -EINVAL;
2500 }
2501
2502 static int addr_filter__resolve_kernel_syms(struct addr_filter *filt)
2503 {
2504         bool no_size = false;
2505         u64 start, size;
2506         int err;
2507
2508         if (symbol_conf.kptr_restrict) {
2509                 pr_err("Kernel addresses are restricted. Unable to resolve kernel symbols.\n");
2510                 return -EINVAL;
2511         }
2512
2513         if (filt->sym_from && !strcmp(filt->sym_from, "*"))
2514                 return addr_filter__entire_kernel(filt);
2515
2516         if (filt->sym_from) {
2517                 err = find_kern_sym(filt->sym_from, &start, &size,
2518                                     filt->sym_from_idx);
2519                 if (err)
2520                         return err;
2521                 filt->addr = start;
2522                 if (filt->range && !filt->size && !filt->sym_to) {
2523                         filt->size = size;
2524                         no_size = !size;
2525                 }
2526         }
2527
2528         if (filt->sym_to) {
2529                 err = find_kern_sym(filt->sym_to, &start, &size,
2530                                     filt->sym_to_idx);
2531                 if (err)
2532                         return err;
2533
2534                 err = check_end_after_start(filt, start, size);
2535                 if (err)
2536                         return err;
2537                 filt->size = start + size - filt->addr;
2538                 no_size = !size;
2539         }
2540
2541         /* The very last symbol in kallsyms does not imply a particular size */
2542         if (no_size) {
2543                 pr_err("Cannot determine size of symbol '%s'\n",
2544                        filt->sym_to ? filt->sym_to : filt->sym_from);
2545                 return -EINVAL;
2546         }
2547
2548         return 0;
2549 }
2550
2551 static struct dso *load_dso(const char *name)
2552 {
2553         struct map *map;
2554         struct dso *dso;
2555
2556         map = dso__new_map(name);
2557         if (!map)
2558                 return NULL;
2559
2560         if (map__load(map) < 0)
2561                 pr_err("File '%s' not found or has no symbols.\n", name);
2562
2563         dso = dso__get(map__dso(map));
2564
2565         map__put(map);
2566
2567         return dso;
2568 }
2569
2570 static bool dso_sym_match(struct symbol *sym, const char *name, int *cnt,
2571                           int idx)
2572 {
2573         /* Same name, and global or the n'th found or any */
2574         return !arch__compare_symbol_names(name, sym->name) &&
2575                ((!idx && sym->binding == STB_GLOBAL) ||
2576                 (idx > 0 && ++*cnt == idx) ||
2577                 idx < 0);
2578 }
2579
2580 static void print_duplicate_syms(struct dso *dso, const char *sym_name)
2581 {
2582         struct symbol *sym;
2583         bool near = false;
2584         int cnt = 0;
2585
2586         pr_err("Multiple symbols with name '%s'\n", sym_name);
2587
2588         sym = dso__first_symbol(dso);
2589         while (sym) {
2590                 if (dso_sym_match(sym, sym_name, &cnt, -1)) {
2591                         pr_err("#%d\t0x%"PRIx64"\t%c\t%s\n",
2592                                ++cnt, sym->start,
2593                                sym->binding == STB_GLOBAL ? 'g' :
2594                                sym->binding == STB_LOCAL  ? 'l' : 'w',
2595                                sym->name);
2596                         near = true;
2597                 } else if (near) {
2598                         near = false;
2599                         pr_err("\t\twhich is near\t\t%s\n", sym->name);
2600                 }
2601                 sym = dso__next_symbol(sym);
2602         }
2603
2604         pr_err("Disambiguate symbol name by inserting #n after the name e.g. %s #2\n",
2605                sym_name);
2606         pr_err("Or select a global symbol by inserting #0 or #g or #G\n");
2607 }
2608
2609 static int find_dso_sym(struct dso *dso, const char *sym_name, u64 *start,
2610                         u64 *size, int idx)
2611 {
2612         struct symbol *sym;
2613         int cnt = 0;
2614
2615         *start = 0;
2616         *size = 0;
2617
2618         sym = dso__first_symbol(dso);
2619         while (sym) {
2620                 if (*start) {
2621                         if (!*size)
2622                                 *size = sym->start - *start;
2623                         if (idx > 0) {
2624                                 if (*size)
2625                                         return 0;
2626                         } else if (dso_sym_match(sym, sym_name, &cnt, idx)) {
2627                                 print_duplicate_syms(dso, sym_name);
2628                                 return -EINVAL;
2629                         }
2630                 } else if (dso_sym_match(sym, sym_name, &cnt, idx)) {
2631                         *start = sym->start;
2632                         *size = sym->end - sym->start;
2633                 }
2634                 sym = dso__next_symbol(sym);
2635         }
2636
2637         if (!*start)
2638                 return sym_not_found_error(sym_name, idx);
2639
2640         return 0;
2641 }
2642
2643 static int addr_filter__entire_dso(struct addr_filter *filt, struct dso *dso)
2644 {
2645         if (dso__data_file_size(dso, NULL)) {
2646                 pr_err("Failed to determine filter for %s\nCannot determine file size.\n",
2647                        filt->filename);
2648                 return -EINVAL;
2649         }
2650
2651         filt->addr = 0;
2652         filt->size = dso->data.file_size;
2653
2654         return 0;
2655 }
2656
2657 static int addr_filter__resolve_syms(struct addr_filter *filt)
2658 {
2659         u64 start, size;
2660         struct dso *dso;
2661         int err = 0;
2662
2663         if (!filt->sym_from && !filt->sym_to)
2664                 return 0;
2665
2666         if (!filt->filename)
2667                 return addr_filter__resolve_kernel_syms(filt);
2668
2669         dso = load_dso(filt->filename);
2670         if (!dso) {
2671                 pr_err("Failed to load symbols from: %s\n", filt->filename);
2672                 return -EINVAL;
2673         }
2674
2675         if (filt->sym_from && !strcmp(filt->sym_from, "*")) {
2676                 err = addr_filter__entire_dso(filt, dso);
2677                 goto put_dso;
2678         }
2679
2680         if (filt->sym_from) {
2681                 err = find_dso_sym(dso, filt->sym_from, &start, &size,
2682                                    filt->sym_from_idx);
2683                 if (err)
2684                         goto put_dso;
2685                 filt->addr = start;
2686                 if (filt->range && !filt->size && !filt->sym_to)
2687                         filt->size = size;
2688         }
2689
2690         if (filt->sym_to) {
2691                 err = find_dso_sym(dso, filt->sym_to, &start, &size,
2692                                    filt->sym_to_idx);
2693                 if (err)
2694                         goto put_dso;
2695
2696                 err = check_end_after_start(filt, start, size);
2697                 if (err)
2698                         return err;
2699
2700                 filt->size = start + size - filt->addr;
2701         }
2702
2703 put_dso:
2704         dso__put(dso);
2705
2706         return err;
2707 }
2708
2709 static char *addr_filter__to_str(struct addr_filter *filt)
2710 {
2711         char filename_buf[PATH_MAX];
2712         const char *at = "";
2713         const char *fn = "";
2714         char *filter;
2715         int err;
2716
2717         if (filt->filename) {
2718                 at = "@";
2719                 fn = realpath(filt->filename, filename_buf);
2720                 if (!fn)
2721                         return NULL;
2722         }
2723
2724         if (filt->range) {
2725                 err = asprintf(&filter, "%s 0x%"PRIx64"/0x%"PRIx64"%s%s",
2726                                filt->action, filt->addr, filt->size, at, fn);
2727         } else {
2728                 err = asprintf(&filter, "%s 0x%"PRIx64"%s%s",
2729                                filt->action, filt->addr, at, fn);
2730         }
2731
2732         return err < 0 ? NULL : filter;
2733 }
2734
2735 static int parse_addr_filter(struct evsel *evsel, const char *filter,
2736                              int max_nr)
2737 {
2738         struct addr_filters filts;
2739         struct addr_filter *filt;
2740         int err;
2741
2742         addr_filters__init(&filts);
2743
2744         err = addr_filters__parse_bare_filter(&filts, filter);
2745         if (err)
2746                 goto out_exit;
2747
2748         if (filts.cnt > max_nr) {
2749                 pr_err("Error: number of address filters (%d) exceeds maximum (%d)\n",
2750                        filts.cnt, max_nr);
2751                 err = -EINVAL;
2752                 goto out_exit;
2753         }
2754
2755         list_for_each_entry(filt, &filts.head, list) {
2756                 char *new_filter;
2757
2758                 err = addr_filter__resolve_syms(filt);
2759                 if (err)
2760                         goto out_exit;
2761
2762                 new_filter = addr_filter__to_str(filt);
2763                 if (!new_filter) {
2764                         err = -ENOMEM;
2765                         goto out_exit;
2766                 }
2767
2768                 if (evsel__append_addr_filter(evsel, new_filter)) {
2769                         err = -ENOMEM;
2770                         goto out_exit;
2771                 }
2772         }
2773
2774 out_exit:
2775         addr_filters__exit(&filts);
2776
2777         if (err) {
2778                 pr_err("Failed to parse address filter: '%s'\n", filter);
2779                 pr_err("Filter format is: filter|start|stop|tracestop <start symbol or address> [/ <end symbol or size>] [@<file name>]\n");
2780                 pr_err("Where multiple filters are separated by space or comma.\n");
2781         }
2782
2783         return err;
2784 }
2785
2786 static int evsel__nr_addr_filter(struct evsel *evsel)
2787 {
2788         struct perf_pmu *pmu = evsel__find_pmu(evsel);
2789         int nr_addr_filters = 0;
2790
2791         if (!pmu)
2792                 return 0;
2793
2794         perf_pmu__scan_file(pmu, "nr_addr_filters", "%d", &nr_addr_filters);
2795
2796         return nr_addr_filters;
2797 }
2798
2799 int auxtrace_parse_filters(struct evlist *evlist)
2800 {
2801         struct evsel *evsel;
2802         char *filter;
2803         int err, max_nr;
2804
2805         evlist__for_each_entry(evlist, evsel) {
2806                 filter = evsel->filter;
2807                 max_nr = evsel__nr_addr_filter(evsel);
2808                 if (!filter || !max_nr)
2809                         continue;
2810                 evsel->filter = NULL;
2811                 err = parse_addr_filter(evsel, filter, max_nr);
2812                 free(filter);
2813                 if (err)
2814                         return err;
2815                 pr_debug("Address filter: %s\n", evsel->filter);
2816         }
2817
2818         return 0;
2819 }
2820
2821 int auxtrace__process_event(struct perf_session *session, union perf_event *event,
2822                             struct perf_sample *sample, struct perf_tool *tool)
2823 {
2824         if (!session->auxtrace)
2825                 return 0;
2826
2827         return session->auxtrace->process_event(session, event, sample, tool);
2828 }
2829
2830 void auxtrace__dump_auxtrace_sample(struct perf_session *session,
2831                                     struct perf_sample *sample)
2832 {
2833         if (!session->auxtrace || !session->auxtrace->dump_auxtrace_sample ||
2834             auxtrace__dont_decode(session))
2835                 return;
2836
2837         session->auxtrace->dump_auxtrace_sample(session, sample);
2838 }
2839
2840 int auxtrace__flush_events(struct perf_session *session, struct perf_tool *tool)
2841 {
2842         if (!session->auxtrace)
2843                 return 0;
2844
2845         return session->auxtrace->flush_events(session, tool);
2846 }
2847
2848 void auxtrace__free_events(struct perf_session *session)
2849 {
2850         if (!session->auxtrace)
2851                 return;
2852
2853         return session->auxtrace->free_events(session);
2854 }
2855
2856 void auxtrace__free(struct perf_session *session)
2857 {
2858         if (!session->auxtrace)
2859                 return;
2860
2861         return session->auxtrace->free(session);
2862 }
2863
2864 bool auxtrace__evsel_is_auxtrace(struct perf_session *session,
2865                                  struct evsel *evsel)
2866 {
2867         if (!session->auxtrace || !session->auxtrace->evsel_is_auxtrace)
2868                 return false;
2869
2870         return session->auxtrace->evsel_is_auxtrace(session, evsel);
2871 }