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__has_any_cpu(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_nr(struct auxtrace_queues *queues, int nr_queues)
 222 {
 223         queues->nr_queues = 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 int auxtrace_queues__init(struct auxtrace_queues *queues)
 231 {
 232         return auxtrace_queues__init_nr(queues, AUXTRACE_INIT_NR_QUEUES);
 233 }
 234
 235 static int auxtrace_queues__grow(struct auxtrace_queues *queues,
 236                                  unsigned int new_nr_queues)
 237 {
 238         unsigned int nr_queues = queues->nr_queues;
 239         struct auxtrace_queue *queue_array;
 240         unsigned int i;
 241
 242         if (!nr_queues)
 243                 nr_queues = AUXTRACE_INIT_NR_QUEUES;
 244
 245         while (nr_queues && nr_queues < new_nr_queues)
 246                 nr_queues <<= 1;
 247
 248         if (nr_queues < queues->nr_queues || nr_queues < new_nr_queues)
 249                 return -EINVAL;
 250
 251         queue_array = auxtrace_alloc_queue_array(nr_queues);
 252         if (!queue_array)
 253                 return -ENOMEM;
 254
 255         for (i = 0; i < queues->nr_queues; i++) {
 256                 list_splice_tail(&queues->queue_array[i].head,
 257                                  &queue_array[i].head);
 258                 queue_array[i].tid = queues->queue_array[i].tid;
 259                 queue_array[i].cpu = queues->queue_array[i].cpu;
 260                 queue_array[i].set = queues->queue_array[i].set;
 261                 queue_array[i].priv = queues->queue_array[i].priv;
 262         }
 263
 264         queues->nr_queues = nr_queues;
 265         queues->queue_array = queue_array;
 266
 267         return 0;
 268 }
 269
 270 static void *auxtrace_copy_data(u64 size, struct perf_session *session)
 271 {
 272         int fd = perf_data__fd(session->data);
 273         void *p;
 274         ssize_t ret;
 275
 276         if (size > SSIZE_MAX)
 277                 return NULL;
 278
 279         p = malloc(size);
 280         if (!p)
 281                 return NULL;
 282
 283         ret = readn(fd, p, size);
 284         if (ret != (ssize_t)size) {
 285                 free(p);
 286                 return NULL;
 287         }
 288
 289         return p;
 290 }
 291
 292 static int auxtrace_queues__queue_buffer(struct auxtrace_queues *queues,
 293                                          unsigned int idx,
 294                                          struct auxtrace_buffer *buffer)
 295 {
 296         struct auxtrace_queue *queue;
 297         int err;
 298
 299         if (idx >= queues->nr_queues) {
 300                 err = auxtrace_queues__grow(queues, idx + 1);
 301                 if (err)
 302                         return err;
 303         }
 304
 305         queue = &queues->queue_array[idx];
 306
 307         if (!queue->set) {
 308                 queue->set = true;
 309                 queue->tid = buffer->tid;
 310                 queue->cpu = buffer->cpu.cpu;
 311         }
 312
 313         buffer->buffer_nr = queues->next_buffer_nr++;
 314
 315         list_add_tail(&buffer->list, &queue->head);
 316
 317         queues->new_data = true;
 318         queues->populated = true;
 319
 320         return 0;
 321 }
 322
 323 /* Limit buffers to 32MiB on 32-bit */
 324 #define BUFFER_LIMIT_FOR_32_BIT (32 * 1024 * 1024)
 325
 326 static int auxtrace_queues__split_buffer(struct auxtrace_queues *queues,
 327                                          unsigned int idx,
 328                                          struct auxtrace_buffer *buffer)
 329 {
 330         u64 sz = buffer->size;
 331         bool consecutive = false;
 332         struct auxtrace_buffer *b;
 333         int err;
 334
 335         while (sz > BUFFER_LIMIT_FOR_32_BIT) {
 336                 b = memdup(buffer, sizeof(struct auxtrace_buffer));
 337                 if (!b)
 338                         return -ENOMEM;
 339                 b->size = BUFFER_LIMIT_FOR_32_BIT;
 340                 b->consecutive = consecutive;
 341                 err = auxtrace_queues__queue_buffer(queues, idx, b);
 342                 if (err) {
 343                         auxtrace_buffer__free(b);
 344                         return err;
 345                 }
 346                 buffer->data_offset += BUFFER_LIMIT_FOR_32_BIT;
 347                 sz -= BUFFER_LIMIT_FOR_32_BIT;
 348                 consecutive = true;
 349         }
 350
 351         buffer->size = sz;
 352         buffer->consecutive = consecutive;
 353
 354         return 0;
 355 }
 356
 357 static bool filter_cpu(struct perf_session *session, struct perf_cpu cpu)
 358 {
 359         unsigned long *cpu_bitmap = session->itrace_synth_opts->cpu_bitmap;
 360
 361         return cpu_bitmap && cpu.cpu != -1 && !test_bit(cpu.cpu, cpu_bitmap);
 362 }
 363
 364 static int auxtrace_queues__add_buffer(struct auxtrace_queues *queues,
 365                                        struct perf_session *session,
 366                                        unsigned int idx,
 367                                        struct auxtrace_buffer *buffer,
 368                                        struct auxtrace_buffer **buffer_ptr)
 369 {
 370         int err = -ENOMEM;
 371
 372         if (filter_cpu(session, buffer->cpu))
 373                 return 0;
 374
 375         buffer = memdup(buffer, sizeof(*buffer));
 376         if (!buffer)
 377                 return -ENOMEM;
 378
 379         if (session->one_mmap) {
 380                 buffer->data = buffer->data_offset - session->one_mmap_offset +
 381                                session->one_mmap_addr;
 382         } else if (perf_data__is_pipe(session->data)) {
 383                 buffer->data = auxtrace_copy_data(buffer->size, session);
 384                 if (!buffer->data)
 385                         goto out_free;
 386                 buffer->data_needs_freeing = true;
 387         } else if (BITS_PER_LONG == 32 &&
 388                    buffer->size > BUFFER_LIMIT_FOR_32_BIT) {
 389                 err = auxtrace_queues__split_buffer(queues, idx, buffer);
 390                 if (err)
 391                         goto out_free;
 392         }
 393
 394         err = auxtrace_queues__queue_buffer(queues, idx, buffer);
 395         if (err)
 396                 goto out_free;
 397
 398         /* FIXME: Doesn't work for split buffer */
 399         if (buffer_ptr)
 400                 *buffer_ptr = buffer;
 401
 402         return 0;
 403
 404 out_free:
 405         auxtrace_buffer__free(buffer);
 406         return err;
 407 }
 408
 409 int auxtrace_queues__add_event(struct auxtrace_queues *queues,
 410                                struct perf_session *session,
 411                                union perf_event *event, off_t data_offset,
 412                                struct auxtrace_buffer **buffer_ptr)
 413 {
 414         struct auxtrace_buffer buffer = {
 415                 .pid = -1,
 416                 .tid = event->auxtrace.tid,
 417                 .cpu = { event->auxtrace.cpu },
 418                 .data_offset = data_offset,
 419                 .offset = event->auxtrace.offset,
 420                 .reference = event->auxtrace.reference,
 421                 .size = event->auxtrace.size,
 422         };
 423         unsigned int idx = event->auxtrace.idx;
 424
 425         return auxtrace_queues__add_buffer(queues, session, idx, &buffer,
 426                                            buffer_ptr);
 427 }
 428
 429 static int auxtrace_queues__add_indexed_event(struct auxtrace_queues *queues,
 430                                               struct perf_session *session,
 431                                               off_t file_offset, size_t sz)
 432 {
 433         union perf_event *event;
 434         int err;
 435         char buf[PERF_SAMPLE_MAX_SIZE];
 436
 437         err = perf_session__peek_event(session, file_offset, buf,
 438                                        PERF_SAMPLE_MAX_SIZE, &event, NULL);
 439         if (err)
 440                 return err;
 441
 442         if (event->header.type == PERF_RECORD_AUXTRACE) {
 443                 if (event->header.size < sizeof(struct perf_record_auxtrace) ||
 444                     event->header.size != sz) {
 445                         err = -EINVAL;
 446                         goto out;
 447                 }
 448                 file_offset += event->header.size;
 449                 err = auxtrace_queues__add_event(queues, session, event,
 450                                                  file_offset, NULL);
 451         }
 452 out:
 453         return err;
 454 }
 455
 456 void auxtrace_queues__free(struct auxtrace_queues *queues)
 457 {
 458         unsigned int i;
 459
 460         for (i = 0; i < queues->nr_queues; i++) {
 461                 while (!list_empty(&queues->queue_array[i].head)) {
 462                         struct auxtrace_buffer *buffer;
 463
 464                         buffer = list_entry(queues->queue_array[i].head.next,
 465                                             struct auxtrace_buffer, list);
 466                         list_del_init(&buffer->list);
 467                         auxtrace_buffer__free(buffer);
 468                 }
 469         }
 470
 471         zfree(&queues->queue_array);
 472         queues->nr_queues = 0;
 473 }
 474
 475 static void auxtrace_heapify(struct auxtrace_heap_item *heap_array,
 476                              unsigned int pos, unsigned int queue_nr,
 477                              u64 ordinal)
 478 {
 479         unsigned int parent;
 480
 481         while (pos) {
 482                 parent = (pos - 1) >> 1;
 483                 if (heap_array[parent].ordinal <= ordinal)
 484                         break;
 485                 heap_array[pos] = heap_array[parent];
 486                 pos = parent;
 487         }
 488         heap_array[pos].queue_nr = queue_nr;
 489         heap_array[pos].ordinal = ordinal;
 490 }
 491
 492 int auxtrace_heap__add(struct auxtrace_heap *heap, unsigned int queue_nr,
 493                        u64 ordinal)
 494 {
 495         struct auxtrace_heap_item *heap_array;
 496
 497         if (queue_nr >= heap->heap_sz) {
 498                 unsigned int heap_sz = AUXTRACE_INIT_NR_QUEUES;
 499
 500                 while (heap_sz <= queue_nr)
 501                         heap_sz <<= 1;
 502                 heap_array = realloc(heap->heap_array,
 503                                      heap_sz * sizeof(struct auxtrace_heap_item));
 504                 if (!heap_array)
 505                         return -ENOMEM;
 506                 heap->heap_array = heap_array;
 507                 heap->heap_sz = heap_sz;
 508         }
 509
 510         auxtrace_heapify(heap->heap_array, heap->heap_cnt++, queue_nr, ordinal);
 511
 512         return 0;
 513 }
 514
 515 void auxtrace_heap__free(struct auxtrace_heap *heap)
 516 {
 517         zfree(&heap->heap_array);
 518         heap->heap_cnt = 0;
 519         heap->heap_sz = 0;
 520 }
 521
 522 void auxtrace_heap__pop(struct auxtrace_heap *heap)
 523 {
 524         unsigned int pos, last, heap_cnt = heap->heap_cnt;
 525         struct auxtrace_heap_item *heap_array;
 526
 527         if (!heap_cnt)
 528                 return;
 529
 530         heap->heap_cnt -= 1;
 531
 532         heap_array = heap->heap_array;
 533
 534         pos = 0;
 535         while (1) {
 536                 unsigned int left, right;
 537
 538                 left = (pos << 1) + 1;
 539                 if (left >= heap_cnt)
 540                         break;
 541                 right = left + 1;
 542                 if (right >= heap_cnt) {
 543                         heap_array[pos] = heap_array[left];
 544                         return;
 545                 }
 546                 if (heap_array[left].ordinal < heap_array[right].ordinal) {
 547                         heap_array[pos] = heap_array[left];
 548                         pos = left;
 549                 } else {
 550                         heap_array[pos] = heap_array[right];
 551                         pos = right;
 552                 }
 553         }
 554
 555         last = heap_cnt - 1;
 556         auxtrace_heapify(heap_array, pos, heap_array[last].queue_nr,
 557                          heap_array[last].ordinal);
 558 }
 559
 560 size_t auxtrace_record__info_priv_size(struct auxtrace_record *itr,
 561                                        struct evlist *evlist)
 562 {
 563         if (itr)
 564                 return itr->info_priv_size(itr, evlist);
 565         return 0;
 566 }
 567
 568 static int auxtrace_not_supported(void)
 569 {
 570         pr_err("AUX area tracing is not supported on this architecture\n");
 571         return -EINVAL;
 572 }
 573
 574 int auxtrace_record__info_fill(struct auxtrace_record *itr,
 575                                struct perf_session *session,
 576                                struct perf_record_auxtrace_info *auxtrace_info,
 577                                size_t priv_size)
 578 {
 579         if (itr)
 580                 return itr->info_fill(itr, session, auxtrace_info, priv_size);
 581         return auxtrace_not_supported();
 582 }
 583
 584 void auxtrace_record__free(struct auxtrace_record *itr)
 585 {
 586         if (itr)
 587                 itr->free(itr);
 588 }
 589
 590 int auxtrace_record__snapshot_start(struct auxtrace_record *itr)
 591 {
 592         if (itr && itr->snapshot_start)
 593                 return itr->snapshot_start(itr);
 594         return 0;
 595 }
 596
 597 int auxtrace_record__snapshot_finish(struct auxtrace_record *itr, bool on_exit)
 598 {
 599         if (!on_exit && itr && itr->snapshot_finish)
 600                 return itr->snapshot_finish(itr);
 601         return 0;
 602 }
 603
 604 int auxtrace_record__find_snapshot(struct auxtrace_record *itr, int idx,
 605                                    struct auxtrace_mmap *mm,
 606                                    unsigned char *data, u64 *head, u64 *old)
 607 {
 608         if (itr && itr->find_snapshot)
 609                 return itr->find_snapshot(itr, idx, mm, data, head, old);
 610         return 0;
 611 }
 612
 613 int auxtrace_record__options(struct auxtrace_record *itr,
 614                              struct evlist *evlist,
 615                              struct record_opts *opts)
 616 {
 617         if (itr) {
 618                 itr->evlist = evlist;
 619                 return itr->recording_options(itr, evlist, opts);
 620         }
 621         return 0;
 622 }
 623
 624 u64 auxtrace_record__reference(struct auxtrace_record *itr)
 625 {
 626         if (itr)
 627                 return itr->reference(itr);
 628         return 0;
 629 }
 630
 631 int auxtrace_parse_snapshot_options(struct auxtrace_record *itr,
 632                                     struct record_opts *opts, const char *str)
 633 {
 634         if (!str)
 635                 return 0;
 636
 637         /* PMU-agnostic options */
 638         switch (*str) {
 639         case 'e':
 640                 opts->auxtrace_snapshot_on_exit = true;
 641                 str++;
 642                 break;
 643         default:
 644                 break;
 645         }
 646
 647         if (itr && itr->parse_snapshot_options)
 648                 return itr->parse_snapshot_options(itr, opts, str);
 649
 650         pr_err("No AUX area tracing to snapshot\n");
 651         return -EINVAL;
 652 }
 653
 654 static int evlist__enable_event_idx(struct evlist *evlist, struct evsel *evsel, int idx)
 655 {
 656         bool per_cpu_mmaps = !perf_cpu_map__has_any_cpu(evlist->core.user_requested_cpus);
 657
 658         if (per_cpu_mmaps) {
 659                 struct perf_cpu evlist_cpu = perf_cpu_map__cpu(evlist->core.all_cpus, idx);
 660                 int cpu_map_idx = perf_cpu_map__idx(evsel->core.cpus, evlist_cpu);
 661
 662                 if (cpu_map_idx == -1)
 663                         return -EINVAL;
 664                 return perf_evsel__enable_cpu(&evsel->core, cpu_map_idx);
 665         }
 666
 667         return perf_evsel__enable_thread(&evsel->core, idx);
 668 }
 669
 670 int auxtrace_record__read_finish(struct auxtrace_record *itr, int idx)
 671 {
 672         struct evsel *evsel;
 673
 674         if (!itr->evlist || !itr->pmu)
 675                 return -EINVAL;
 676
 677         evlist__for_each_entry(itr->evlist, evsel) {
 678                 if (evsel->core.attr.type == itr->pmu->type) {
 679                         if (evsel->disabled)
 680                                 return 0;
 681                         return evlist__enable_event_idx(itr->evlist, evsel, idx);
 682                 }
 683         }
 684         return -EINVAL;
 685 }
 686
 687 /*
 688  * Event record size is 16-bit which results in a maximum size of about 64KiB.
 689  * Allow about 4KiB for the rest of the sample record, to give a maximum
 690  * AUX area sample size of 60KiB.
 691  */
 692 #define MAX_AUX_SAMPLE_SIZE (60 * 1024)
 693
 694 /* Arbitrary default size if no other default provided */
 695 #define DEFAULT_AUX_SAMPLE_SIZE (4 * 1024)
 696
 697 static int auxtrace_validate_aux_sample_size(struct evlist *evlist,
 698                                              struct record_opts *opts)
 699 {
 700         struct evsel *evsel;
 701         bool has_aux_leader = false;
 702         u32 sz;
 703
 704         evlist__for_each_entry(evlist, evsel) {
 705                 sz = evsel->core.attr.aux_sample_size;
 706                 if (evsel__is_group_leader(evsel)) {
 707                         has_aux_leader = evsel__is_aux_event(evsel);
 708                         if (sz) {
 709                                 if (has_aux_leader)
 710                                         pr_err("Cannot add AUX area sampling to an AUX area event\n");
 711                                 else
 712                                         pr_err("Cannot add AUX area sampling to a group leader\n");
 713                                 return -EINVAL;
 714                         }
 715                 }
 716                 if (sz > MAX_AUX_SAMPLE_SIZE) {
 717                         pr_err("AUX area sample size %u too big, max. %d\n",
 718                                sz, MAX_AUX_SAMPLE_SIZE);
 719                         return -EINVAL;
 720                 }
 721                 if (sz) {
 722                         if (!has_aux_leader) {
 723                                 pr_err("Cannot add AUX area sampling because group leader is not an AUX area event\n");
 724                                 return -EINVAL;
 725                         }
 726                         evsel__set_sample_bit(evsel, AUX);
 727                         opts->auxtrace_sample_mode = true;
 728                 } else {
 729                         evsel__reset_sample_bit(evsel, AUX);
 730                 }
 731         }
 732
 733         if (!opts->auxtrace_sample_mode) {
 734                 pr_err("AUX area sampling requires an AUX area event group leader plus other events to which to add samples\n");
 735                 return -EINVAL;
 736         }
 737
 738         if (!perf_can_aux_sample()) {
 739                 pr_err("AUX area sampling is not supported by kernel\n");
 740                 return -EINVAL;
 741         }
 742
 743         return 0;
 744 }
 745
 746 int auxtrace_parse_sample_options(struct auxtrace_record *itr,
 747                                   struct evlist *evlist,
 748                                   struct record_opts *opts, const char *str)
 749 {
 750         struct evsel_config_term *term;
 751         struct evsel *aux_evsel;
 752         bool has_aux_sample_size = false;
 753         bool has_aux_leader = false;
 754         struct evsel *evsel;
 755         char *endptr;
 756         unsigned long sz;
 757
 758         if (!str)
 759                 goto no_opt;
 760
 761         if (!itr) {
 762                 pr_err("No AUX area event to sample\n");
 763                 return -EINVAL;
 764         }
 765
 766         sz = strtoul(str, &endptr, 0);
 767         if (*endptr || sz > UINT_MAX) {
 768                 pr_err("Bad AUX area sampling option: '%s'\n", str);
 769                 return -EINVAL;
 770         }
 771
 772         if (!sz)
 773                 sz = itr->default_aux_sample_size;
 774
 775         if (!sz)
 776                 sz = DEFAULT_AUX_SAMPLE_SIZE;
 777
 778         /* Set aux_sample_size based on --aux-sample option */
 779         evlist__for_each_entry(evlist, evsel) {
 780                 if (evsel__is_group_leader(evsel)) {
 781                         has_aux_leader = evsel__is_aux_event(evsel);
 782                 } else if (has_aux_leader) {
 783                         evsel->core.attr.aux_sample_size = sz;
 784                 }
 785         }
 786 no_opt:
 787         aux_evsel = NULL;
 788         /* Override with aux_sample_size from config term */
 789         evlist__for_each_entry(evlist, evsel) {
 790                 if (evsel__is_aux_event(evsel))
 791                         aux_evsel = evsel;
 792                 term = evsel__get_config_term(evsel, AUX_SAMPLE_SIZE);
 793                 if (term) {
 794                         has_aux_sample_size = true;
 795                         evsel->core.attr.aux_sample_size = term->val.aux_sample_size;
 796                         /* If possible, group with the AUX event */
 797                         if (aux_evsel && evsel->core.attr.aux_sample_size)
 798                                 evlist__regroup(evlist, aux_evsel, evsel);
 799                 }
 800         }
 801
 802         if (!str && !has_aux_sample_size)
 803                 return 0;
 804
 805         if (!itr) {
 806                 pr_err("No AUX area event to sample\n");
 807                 return -EINVAL;
 808         }
 809
 810         return auxtrace_validate_aux_sample_size(evlist, opts);
 811 }
 812
 813 void auxtrace_regroup_aux_output(struct evlist *evlist)
 814 {
 815         struct evsel *evsel, *aux_evsel = NULL;
 816         struct evsel_config_term *term;
 817
 818         evlist__for_each_entry(evlist, evsel) {
 819                 if (evsel__is_aux_event(evsel))
 820                         aux_evsel = evsel;
 821                 term = evsel__get_config_term(evsel, AUX_OUTPUT);
 822                 /* If possible, group with the AUX event */
 823                 if (term && aux_evsel)
 824                         evlist__regroup(evlist, aux_evsel, evsel);
 825         }
 826 }
 827
 828 struct auxtrace_record *__weak
 829 auxtrace_record__init(struct evlist *evlist __maybe_unused, int *err)
 830 {
 831         *err = 0;
 832         return NULL;
 833 }
 834
 835 static int auxtrace_index__alloc(struct list_head *head)
 836 {
 837         struct auxtrace_index *auxtrace_index;
 838
 839         auxtrace_index = malloc(sizeof(struct auxtrace_index));
 840         if (!auxtrace_index)
 841                 return -ENOMEM;
 842
 843         auxtrace_index->nr = 0;
 844         INIT_LIST_HEAD(&auxtrace_index->list);
 845
 846         list_add_tail(&auxtrace_index->list, head);
 847
 848         return 0;
 849 }
 850
 851 void auxtrace_index__free(struct list_head *head)
 852 {
 853         struct auxtrace_index *auxtrace_index, *n;
 854
 855         list_for_each_entry_safe(auxtrace_index, n, head, list) {
 856                 list_del_init(&auxtrace_index->list);
 857                 free(auxtrace_index);
 858         }
 859 }
 860
 861 static struct auxtrace_index *auxtrace_index__last(struct list_head *head)
 862 {
 863         struct auxtrace_index *auxtrace_index;
 864         int err;
 865
 866         if (list_empty(head)) {
 867                 err = auxtrace_index__alloc(head);
 868                 if (err)
 869                         return NULL;
 870         }
 871
 872         auxtrace_index = list_entry(head->prev, struct auxtrace_index, list);
 873
 874         if (auxtrace_index->nr >= PERF_AUXTRACE_INDEX_ENTRY_COUNT) {
 875                 err = auxtrace_index__alloc(head);
 876                 if (err)
 877                         return NULL;
 878                 auxtrace_index = list_entry(head->prev, struct auxtrace_index,
 879                                             list);
 880         }
 881
 882         return auxtrace_index;
 883 }
 884
 885 int auxtrace_index__auxtrace_event(struct list_head *head,
 886                                    union perf_event *event, off_t file_offset)
 887 {
 888         struct auxtrace_index *auxtrace_index;
 889         size_t nr;
 890
 891         auxtrace_index = auxtrace_index__last(head);
 892         if (!auxtrace_index)
 893                 return -ENOMEM;
 894
 895         nr = auxtrace_index->nr;
 896         auxtrace_index->entries[nr].file_offset = file_offset;
 897         auxtrace_index->entries[nr].sz = event->header.size;
 898         auxtrace_index->nr += 1;
 899
 900         return 0;
 901 }
 902
 903 static int auxtrace_index__do_write(int fd,
 904                                     struct auxtrace_index *auxtrace_index)
 905 {
 906         struct auxtrace_index_entry ent;
 907         size_t i;
 908
 909         for (i = 0; i < auxtrace_index->nr; i++) {
 910                 ent.file_offset = auxtrace_index->entries[i].file_offset;
 911                 ent.sz = auxtrace_index->entries[i].sz;
 912                 if (writen(fd, &ent, sizeof(ent)) != sizeof(ent))
 913                         return -errno;
 914         }
 915         return 0;
 916 }
 917
 918 int auxtrace_index__write(int fd, struct list_head *head)
 919 {
 920         struct auxtrace_index *auxtrace_index;
 921         u64 total = 0;
 922         int err;
 923
 924         list_for_each_entry(auxtrace_index, head, list)
 925                 total += auxtrace_index->nr;
 926
 927         if (writen(fd, &total, sizeof(total)) != sizeof(total))
 928                 return -errno;
 929
 930         list_for_each_entry(auxtrace_index, head, list) {
 931                 err = auxtrace_index__do_write(fd, auxtrace_index);
 932                 if (err)
 933                         return err;
 934         }
 935
 936         return 0;
 937 }
 938
 939 static int auxtrace_index__process_entry(int fd, struct list_head *head,
 940                                          bool needs_swap)
 941 {
 942         struct auxtrace_index *auxtrace_index;
 943         struct auxtrace_index_entry ent;
 944         size_t nr;
 945
 946         if (readn(fd, &ent, sizeof(ent)) != sizeof(ent))
 947                 return -1;
 948
 949         auxtrace_index = auxtrace_index__last(head);
 950         if (!auxtrace_index)
 951                 return -1;
 952
 953         nr = auxtrace_index->nr;
 954         if (needs_swap) {
 955                 auxtrace_index->entries[nr].file_offset =
 956                                                 bswap_64(ent.file_offset);
 957                 auxtrace_index->entries[nr].sz = bswap_64(ent.sz);
 958         } else {
 959                 auxtrace_index->entries[nr].file_offset = ent.file_offset;
 960                 auxtrace_index->entries[nr].sz = ent.sz;
 961         }
 962
 963         auxtrace_index->nr = nr + 1;
 964
 965         return 0;
 966 }
 967
 968 int auxtrace_index__process(int fd, u64 size, struct perf_session *session,
 969                             bool needs_swap)
 970 {
 971         struct list_head *head = &session->auxtrace_index;
 972         u64 nr;
 973
 974         if (readn(fd, &nr, sizeof(u64)) != sizeof(u64))
 975                 return -1;
 976
 977         if (needs_swap)
 978                 nr = bswap_64(nr);
 979
 980         if (sizeof(u64) + nr * sizeof(struct auxtrace_index_entry) > size)
 981                 return -1;
 982
 983         while (nr--) {
 984                 int err;
 985
 986                 err = auxtrace_index__process_entry(fd, head, needs_swap);
 987                 if (err)
 988                         return -1;
 989         }
 990
 991         return 0;
 992 }
 993
 994 static int auxtrace_queues__process_index_entry(struct auxtrace_queues *queues,
 995                                                 struct perf_session *session,
 996                                                 struct auxtrace_index_entry *ent)
 997 {
 998         return auxtrace_queues__add_indexed_event(queues, session,
 999                                                   ent->file_offset, ent->sz);
1000 }
1001
1002 int auxtrace_queues__process_index(struct auxtrace_queues *queues,
1003                                    struct perf_session *session)
1004 {
1005         struct auxtrace_index *auxtrace_index;
1006         struct auxtrace_index_entry *ent;
1007         size_t i;
1008         int err;
1009
1010         if (auxtrace__dont_decode(session))
1011                 return 0;
1012
1013         list_for_each_entry(auxtrace_index, &session->auxtrace_index, list) {
1014                 for (i = 0; i < auxtrace_index->nr; i++) {
1015                         ent = &auxtrace_index->entries[i];
1016                         err = auxtrace_queues__process_index_entry(queues,
1017                                                                    session,
1018                                                                    ent);
1019                         if (err)
1020                                 return err;
1021                 }
1022         }
1023         return 0;
1024 }
1025
1026 struct auxtrace_buffer *auxtrace_buffer__next(struct auxtrace_queue *queue,
1027                                               struct auxtrace_buffer *buffer)
1028 {
1029         if (buffer) {
1030                 if (list_is_last(&buffer->list, &queue->head))
1031                         return NULL;
1032                 return list_entry(buffer->list.next, struct auxtrace_buffer,
1033                                   list);
1034         } else {
1035                 if (list_empty(&queue->head))
1036                         return NULL;
1037                 return list_entry(queue->head.next, struct auxtrace_buffer,
1038                                   list);
1039         }
1040 }
1041
1042 struct auxtrace_queue *auxtrace_queues__sample_queue(struct auxtrace_queues *queues,
1043                                                      struct perf_sample *sample,
1044                                                      struct perf_session *session)
1045 {
1046         struct perf_sample_id *sid;
1047         unsigned int idx;
1048         u64 id;
1049
1050         id = sample->id;
1051         if (!id)
1052                 return NULL;
1053
1054         sid = evlist__id2sid(session->evlist, id);
1055         if (!sid)
1056                 return NULL;
1057
1058         idx = sid->idx;
1059
1060         if (idx >= queues->nr_queues)
1061                 return NULL;
1062
1063         return &queues->queue_array[idx];
1064 }
1065
1066 int auxtrace_queues__add_sample(struct auxtrace_queues *queues,
1067                                 struct perf_session *session,
1068                                 struct perf_sample *sample, u64 data_offset,
1069                                 u64 reference)
1070 {
1071         struct auxtrace_buffer buffer = {
1072                 .pid = -1,
1073                 .data_offset = data_offset,
1074                 .reference = reference,
1075                 .size = sample->aux_sample.size,
1076         };
1077         struct perf_sample_id *sid;
1078         u64 id = sample->id;
1079         unsigned int idx;
1080
1081         if (!id)
1082                 return -EINVAL;
1083
1084         sid = evlist__id2sid(session->evlist, id);
1085         if (!sid)
1086                 return -ENOENT;
1087
1088         idx = sid->idx;
1089         buffer.tid = sid->tid;
1090         buffer.cpu = sid->cpu;
1091
1092         return auxtrace_queues__add_buffer(queues, session, idx, &buffer, NULL);
1093 }
1094
1095 struct queue_data {
1096         bool samples;
1097         bool events;
1098 };
1099
1100 static int auxtrace_queue_data_cb(struct perf_session *session,
1101                                   union perf_event *event, u64 offset,
1102                                   void *data)
1103 {
1104         struct queue_data *qd = data;
1105         struct perf_sample sample;
1106         int err;
1107
1108         if (qd->events && event->header.type == PERF_RECORD_AUXTRACE) {
1109                 if (event->header.size < sizeof(struct perf_record_auxtrace))
1110                         return -EINVAL;
1111                 offset += event->header.size;
1112                 return session->auxtrace->queue_data(session, NULL, event,
1113                                                      offset);
1114         }
1115
1116         if (!qd->samples || event->header.type != PERF_RECORD_SAMPLE)
1117                 return 0;
1118
1119         err = evlist__parse_sample(session->evlist, event, &sample);
1120         if (err)
1121                 return err;
1122
1123         if (!sample.aux_sample.size)
1124                 return 0;
1125
1126         offset += sample.aux_sample.data - (void *)event;
1127
1128         return session->auxtrace->queue_data(session, &sample, NULL, offset);
1129 }
1130
1131 int auxtrace_queue_data(struct perf_session *session, bool samples, bool events)
1132 {
1133         struct queue_data qd = {
1134                 .samples = samples,
1135                 .events = events,
1136         };
1137
1138         if (auxtrace__dont_decode(session))
1139                 return 0;
1140
1141         if (perf_data__is_pipe(session->data))
1142                 return 0;
1143
1144         if (!session->auxtrace || !session->auxtrace->queue_data)
1145                 return -EINVAL;
1146
1147         return perf_session__peek_events(session, session->header.data_offset,
1148                                          session->header.data_size,
1149                                          auxtrace_queue_data_cb, &qd);
1150 }
1151
1152 void *auxtrace_buffer__get_data_rw(struct auxtrace_buffer *buffer, int fd, bool rw)
1153 {
1154         int prot = rw ? PROT_READ | PROT_WRITE : PROT_READ;
1155         size_t adj = buffer->data_offset & (page_size - 1);
1156         size_t size = buffer->size + adj;
1157         off_t file_offset = buffer->data_offset - adj;
1158         void *addr;
1159
1160         if (buffer->data)
1161                 return buffer->data;
1162
1163         addr = mmap(NULL, size, prot, MAP_SHARED, fd, file_offset);
1164         if (addr == MAP_FAILED)
1165                 return NULL;
1166
1167         buffer->mmap_addr = addr;
1168         buffer->mmap_size = size;
1169
1170         buffer->data = addr + adj;
1171
1172         return buffer->data;
1173 }
1174
1175 void auxtrace_buffer__put_data(struct auxtrace_buffer *buffer)
1176 {
1177         if (!buffer->data || !buffer->mmap_addr)
1178                 return;
1179         munmap(buffer->mmap_addr, buffer->mmap_size);
1180         buffer->mmap_addr = NULL;
1181         buffer->mmap_size = 0;
1182         buffer->data = NULL;
1183         buffer->use_data = NULL;
1184 }
1185
1186 void auxtrace_buffer__drop_data(struct auxtrace_buffer *buffer)
1187 {
1188         auxtrace_buffer__put_data(buffer);
1189         if (buffer->data_needs_freeing) {
1190                 buffer->data_needs_freeing = false;
1191                 zfree(&buffer->data);
1192                 buffer->use_data = NULL;
1193                 buffer->size = 0;
1194         }
1195 }
1196
1197 void auxtrace_buffer__free(struct auxtrace_buffer *buffer)
1198 {
1199         auxtrace_buffer__drop_data(buffer);
1200         free(buffer);
1201 }
1202
1203 void auxtrace_synth_guest_error(struct perf_record_auxtrace_error *auxtrace_error, int type,
1204                                 int code, int cpu, pid_t pid, pid_t tid, u64 ip,
1205                                 const char *msg, u64 timestamp,
1206                                 pid_t machine_pid, int vcpu)
1207 {
1208         size_t size;
1209
1210         memset(auxtrace_error, 0, sizeof(struct perf_record_auxtrace_error));
1211
1212         auxtrace_error->header.type = PERF_RECORD_AUXTRACE_ERROR;
1213         auxtrace_error->type = type;
1214         auxtrace_error->code = code;
1215         auxtrace_error->cpu = cpu;
1216         auxtrace_error->pid = pid;
1217         auxtrace_error->tid = tid;
1218         auxtrace_error->fmt = 1;
1219         auxtrace_error->ip = ip;
1220         auxtrace_error->time = timestamp;
1221         strlcpy(auxtrace_error->msg, msg, MAX_AUXTRACE_ERROR_MSG);
1222         if (machine_pid) {
1223                 auxtrace_error->fmt = 2;
1224                 auxtrace_error->machine_pid = machine_pid;
1225                 auxtrace_error->vcpu = vcpu;
1226                 size = sizeof(*auxtrace_error);
1227         } else {
1228                 size = (void *)auxtrace_error->msg - (void *)auxtrace_error +
1229                        strlen(auxtrace_error->msg) + 1;
1230         }
1231         auxtrace_error->header.size = PERF_ALIGN(size, sizeof(u64));
1232 }
1233
1234 void auxtrace_synth_error(struct perf_record_auxtrace_error *auxtrace_error, int type,
1235                           int code, int cpu, pid_t pid, pid_t tid, u64 ip,
1236                           const char *msg, u64 timestamp)
1237 {
1238         auxtrace_synth_guest_error(auxtrace_error, type, code, cpu, pid, tid,
1239                                    ip, msg, timestamp, 0, -1);
1240 }
1241
1242 int perf_event__synthesize_auxtrace_info(struct auxtrace_record *itr,
1243                                          struct perf_tool *tool,
1244                                          struct perf_session *session,
1245                                          perf_event__handler_t process)
1246 {
1247         union perf_event *ev;
1248         size_t priv_size;
1249         int err;
1250
1251         pr_debug2("Synthesizing auxtrace information\n");
1252         priv_size = auxtrace_record__info_priv_size(itr, session->evlist);
1253         ev = zalloc(sizeof(struct perf_record_auxtrace_info) + priv_size);
1254         if (!ev)
1255                 return -ENOMEM;
1256
1257         ev->auxtrace_info.header.type = PERF_RECORD_AUXTRACE_INFO;
1258         ev->auxtrace_info.header.size = sizeof(struct perf_record_auxtrace_info) +
1259                                         priv_size;
1260         err = auxtrace_record__info_fill(itr, session, &ev->auxtrace_info,
1261                                          priv_size);
1262         if (err)
1263                 goto out_free;
1264
1265         err = process(tool, ev, NULL, NULL);
1266 out_free:
1267         free(ev);
1268         return err;
1269 }
1270
1271 static void unleader_evsel(struct evlist *evlist, struct evsel *leader)
1272 {
1273         struct evsel *new_leader = NULL;
1274         struct evsel *evsel;
1275
1276         /* Find new leader for the group */
1277         evlist__for_each_entry(evlist, evsel) {
1278                 if (!evsel__has_leader(evsel, leader) || evsel == leader)
1279                         continue;
1280                 if (!new_leader)
1281                         new_leader = evsel;
1282                 evsel__set_leader(evsel, new_leader);
1283         }
1284
1285         /* Update group information */
1286         if (new_leader) {
1287                 zfree(&new_leader->group_name);
1288                 new_leader->group_name = leader->group_name;
1289                 leader->group_name = NULL;
1290
1291                 new_leader->core.nr_members = leader->core.nr_members - 1;
1292                 leader->core.nr_members = 1;
1293         }
1294 }
1295
1296 static void unleader_auxtrace(struct perf_session *session)
1297 {
1298         struct evsel *evsel;
1299
1300         evlist__for_each_entry(session->evlist, evsel) {
1301                 if (auxtrace__evsel_is_auxtrace(session, evsel) &&
1302                     evsel__is_group_leader(evsel)) {
1303                         unleader_evsel(session->evlist, evsel);
1304                 }
1305         }
1306 }
1307
1308 int perf_event__process_auxtrace_info(struct perf_session *session,
1309                                       union perf_event *event)
1310 {
1311         enum auxtrace_type type = event->auxtrace_info.type;
1312         int err;
1313
1314         if (dump_trace)
1315                 fprintf(stdout, " type: %u\n", type);
1316
1317         switch (type) {
1318         case PERF_AUXTRACE_INTEL_PT:
1319                 err = intel_pt_process_auxtrace_info(event, session);
1320                 break;
1321         case PERF_AUXTRACE_INTEL_BTS:
1322                 err = intel_bts_process_auxtrace_info(event, session);
1323                 break;
1324         case PERF_AUXTRACE_ARM_SPE:
1325                 err = arm_spe_process_auxtrace_info(event, session);
1326                 break;
1327         case PERF_AUXTRACE_CS_ETM:
1328                 err = cs_etm__process_auxtrace_info(event, session);
1329                 break;
1330         case PERF_AUXTRACE_S390_CPUMSF:
1331                 err = s390_cpumsf_process_auxtrace_info(event, session);
1332                 break;
1333         case PERF_AUXTRACE_HISI_PTT:
1334                 err = hisi_ptt_process_auxtrace_info(event, session);
1335                 break;
1336         case PERF_AUXTRACE_UNKNOWN:
1337         default:
1338                 return -EINVAL;
1339         }
1340
1341         if (err)
1342                 return err;
1343
1344         unleader_auxtrace(session);
1345
1346         return 0;
1347 }
1348
1349 s64 perf_event__process_auxtrace(struct perf_session *session,
1350                                  union perf_event *event)
1351 {
1352         s64 err;
1353
1354         if (dump_trace)
1355                 fprintf(stdout, " size: %#"PRI_lx64"  offset: %#"PRI_lx64"  ref: %#"PRI_lx64"  idx: %u  tid: %d  cpu: %d\n",
1356                         event->auxtrace.size, event->auxtrace.offset,
1357                         event->auxtrace.reference, event->auxtrace.idx,
1358                         event->auxtrace.tid, event->auxtrace.cpu);
1359
1360         if (auxtrace__dont_decode(session))
1361                 return event->auxtrace.size;
1362
1363         if (!session->auxtrace || event->header.type != PERF_RECORD_AUXTRACE)
1364                 return -EINVAL;
1365
1366         err = session->auxtrace->process_auxtrace_event(session, event, session->tool);
1367         if (err < 0)
1368                 return err;
1369
1370         return event->auxtrace.size;
1371 }
1372
1373 #define PERF_ITRACE_DEFAULT_PERIOD_TYPE         PERF_ITRACE_PERIOD_NANOSECS
1374 #define PERF_ITRACE_DEFAULT_PERIOD              100000
1375 #define PERF_ITRACE_DEFAULT_CALLCHAIN_SZ        16
1376 #define PERF_ITRACE_MAX_CALLCHAIN_SZ            1024
1377 #define PERF_ITRACE_DEFAULT_LAST_BRANCH_SZ      64
1378 #define PERF_ITRACE_MAX_LAST_BRANCH_SZ          1024
1379
1380 void itrace_synth_opts__set_default(struct itrace_synth_opts *synth_opts,
1381                                     bool no_sample)
1382 {
1383         synth_opts->branches = true;
1384         synth_opts->transactions = true;
1385         synth_opts->ptwrites = true;
1386         synth_opts->pwr_events = true;
1387         synth_opts->other_events = true;
1388         synth_opts->intr_events = true;
1389         synth_opts->errors = true;
1390         synth_opts->flc = true;
1391         synth_opts->llc = true;
1392         synth_opts->tlb = true;
1393         synth_opts->mem = true;
1394         synth_opts->remote_access = true;
1395
1396         if (no_sample) {
1397                 synth_opts->period_type = PERF_ITRACE_PERIOD_INSTRUCTIONS;
1398                 synth_opts->period = 1;
1399                 synth_opts->calls = true;
1400         } else {
1401                 synth_opts->instructions = true;
1402                 synth_opts->cycles = true;
1403                 synth_opts->period_type = PERF_ITRACE_DEFAULT_PERIOD_TYPE;
1404                 synth_opts->period = PERF_ITRACE_DEFAULT_PERIOD;
1405         }
1406         synth_opts->callchain_sz = PERF_ITRACE_DEFAULT_CALLCHAIN_SZ;
1407         synth_opts->last_branch_sz = PERF_ITRACE_DEFAULT_LAST_BRANCH_SZ;
1408         synth_opts->initial_skip = 0;
1409 }
1410
1411 static int get_flag(const char **ptr, unsigned int *flags)
1412 {
1413         while (1) {
1414                 char c = **ptr;
1415
1416                 if (c >= 'a' && c <= 'z') {
1417                         *flags |= 1 << (c - 'a');
1418                         ++*ptr;
1419                         return 0;
1420                 } else if (c == ' ') {
1421                         ++*ptr;
1422                         continue;
1423                 } else {
1424                         return -1;
1425                 }
1426         }
1427 }
1428
1429 static int get_flags(const char **ptr, unsigned int *plus_flags, unsigned int *minus_flags)
1430 {
1431         while (1) {
1432                 switch (**ptr) {
1433                 case '+':
1434                         ++*ptr;
1435                         if (get_flag(ptr, plus_flags))
1436                                 return -1;
1437                         break;
1438                 case '-':
1439                         ++*ptr;
1440                         if (get_flag(ptr, minus_flags))
1441                                 return -1;
1442                         break;
1443                 case ' ':
1444                         ++*ptr;
1445                         break;
1446                 default:
1447                         return 0;
1448                 }
1449         }
1450 }
1451
1452 #define ITRACE_DFLT_LOG_ON_ERROR_SZ 16384
1453
1454 static unsigned int itrace_log_on_error_size(void)
1455 {
1456         unsigned int sz = 0;
1457
1458         perf_config_scan("itrace.debug-log-buffer-size", "%u", &sz);
1459         return sz ?: ITRACE_DFLT_LOG_ON_ERROR_SZ;
1460 }
1461
1462 /*
1463  * Please check tools/perf/Documentation/perf-script.txt for information
1464  * about the options parsed here, which is introduced after this cset,
1465  * when support in 'perf script' for these options is introduced.
1466  */
1467 int itrace_do_parse_synth_opts(struct itrace_synth_opts *synth_opts,
1468                                const char *str, int unset)
1469 {
1470         const char *p;
1471         char *endptr;
1472         bool period_type_set = false;
1473         bool period_set = false;
1474         bool iy = false;
1475
1476         synth_opts->set = true;
1477
1478         if (unset) {
1479                 synth_opts->dont_decode = true;
1480                 return 0;
1481         }
1482
1483         if (!str) {
1484                 itrace_synth_opts__set_default(synth_opts,
1485                                                synth_opts->default_no_sample);
1486                 return 0;
1487         }
1488
1489         for (p = str; *p;) {
1490                 switch (*p++) {
1491                 case 'i':
1492                 case 'y':
1493                         iy = true;
1494                         if (p[-1] == 'y')
1495                                 synth_opts->cycles = true;
1496                         else
1497                                 synth_opts->instructions = true;
1498                         while (*p == ' ' || *p == ',')
1499                                 p += 1;
1500                         if (isdigit(*p)) {
1501                                 synth_opts->period = strtoull(p, &endptr, 10);
1502                                 period_set = true;
1503                                 p = endptr;
1504                                 while (*p == ' ' || *p == ',')
1505                                         p += 1;
1506                                 switch (*p++) {
1507                                 case 'i':
1508                                         synth_opts->period_type =
1509                                                 PERF_ITRACE_PERIOD_INSTRUCTIONS;
1510                                         period_type_set = true;
1511                                         break;
1512                                 case 't':
1513                                         synth_opts->period_type =
1514                                                 PERF_ITRACE_PERIOD_TICKS;
1515                                         period_type_set = true;
1516                                         break;
1517                                 case 'm':
1518                                         synth_opts->period *= 1000;
1519                                         /* Fall through */
1520                                 case 'u':
1521                                         synth_opts->period *= 1000;
1522                                         /* Fall through */
1523                                 case 'n':
1524                                         if (*p++ != 's')
1525                                                 goto out_err;
1526                                         synth_opts->period_type =
1527                                                 PERF_ITRACE_PERIOD_NANOSECS;
1528                                         period_type_set = true;
1529                                         break;
1530                                 case '\0':
1531                                         goto out;
1532                                 default:
1533                                         goto out_err;
1534                                 }
1535                         }
1536                         break;
1537                 case 'b':
1538                         synth_opts->branches = true;
1539                         break;
1540                 case 'x':
1541                         synth_opts->transactions = true;
1542                         break;
1543                 case 'w':
1544                         synth_opts->ptwrites = true;
1545                         break;
1546                 case 'p':
1547                         synth_opts->pwr_events = true;
1548                         break;
1549                 case 'o':
1550                         synth_opts->other_events = true;
1551                         break;
1552                 case 'I':
1553                         synth_opts->intr_events = true;
1554                         break;
1555                 case 'e':
1556                         synth_opts->errors = true;
1557                         if (get_flags(&p, &synth_opts->error_plus_flags,
1558                                       &synth_opts->error_minus_flags))
1559                                 goto out_err;
1560                         break;
1561                 case 'd':
1562                         synth_opts->log = true;
1563                         if (get_flags(&p, &synth_opts->log_plus_flags,
1564                                       &synth_opts->log_minus_flags))
1565                                 goto out_err;
1566                         if (synth_opts->log_plus_flags & AUXTRACE_LOG_FLG_ON_ERROR)
1567                                 synth_opts->log_on_error_size = itrace_log_on_error_size();
1568                         break;
1569                 case 'c':
1570                         synth_opts->branches = true;
1571                         synth_opts->calls = true;
1572                         break;
1573                 case 'r':
1574                         synth_opts->branches = true;
1575                         synth_opts->returns = true;
1576                         break;
1577                 case 'G':
1578                 case 'g':
1579                         if (p[-1] == 'G')
1580                                 synth_opts->add_callchain = true;
1581                         else
1582                                 synth_opts->callchain = true;
1583                         synth_opts->callchain_sz =
1584                                         PERF_ITRACE_DEFAULT_CALLCHAIN_SZ;
1585                         while (*p == ' ' || *p == ',')
1586                                 p += 1;
1587                         if (isdigit(*p)) {
1588                                 unsigned int val;
1589
1590                                 val = strtoul(p, &endptr, 10);
1591                                 p = endptr;
1592                                 if (!val || val > PERF_ITRACE_MAX_CALLCHAIN_SZ)
1593                                         goto out_err;
1594                                 synth_opts->callchain_sz = val;
1595                         }
1596                         break;
1597                 case 'L':
1598                 case 'l':
1599                         if (p[-1] == 'L')
1600                                 synth_opts->add_last_branch = true;
1601                         else
1602                                 synth_opts->last_branch = true;
1603                         synth_opts->last_branch_sz =
1604                                         PERF_ITRACE_DEFAULT_LAST_BRANCH_SZ;
1605                         while (*p == ' ' || *p == ',')
1606                                 p += 1;
1607                         if (isdigit(*p)) {
1608                                 unsigned int val;
1609
1610                                 val = strtoul(p, &endptr, 10);
1611                                 p = endptr;
1612                                 if (!val ||
1613                                     val > PERF_ITRACE_MAX_LAST_BRANCH_SZ)
1614                                         goto out_err;
1615                                 synth_opts->last_branch_sz = val;
1616                         }
1617                         break;
1618                 case 's':
1619                         synth_opts->initial_skip = strtoul(p, &endptr, 10);
1620                         if (p == endptr)
1621                                 goto out_err;
1622                         p = endptr;
1623                         break;
1624                 case 'f':
1625                         synth_opts->flc = true;
1626                         break;
1627                 case 'm':
1628                         synth_opts->llc = true;
1629                         break;
1630                 case 't':
1631                         synth_opts->tlb = true;
1632                         break;
1633                 case 'a':
1634                         synth_opts->remote_access = true;
1635                         break;
1636                 case 'M':
1637                         synth_opts->mem = true;
1638                         break;
1639                 case 'q':
1640                         synth_opts->quick += 1;
1641                         break;
1642                 case 'A':
1643                         synth_opts->approx_ipc = true;
1644                         break;
1645                 case 'Z':
1646                         synth_opts->timeless_decoding = true;
1647                         break;
1648                 case 'T':
1649                         synth_opts->use_timestamp = true;
1650                         break;
1651                 case ' ':
1652                 case ',':
1653                         break;
1654                 default:
1655                         goto out_err;
1656                 }
1657         }
1658 out:
1659         if (iy) {
1660                 if (!period_type_set)
1661                         synth_opts->period_type =
1662                                         PERF_ITRACE_DEFAULT_PERIOD_TYPE;
1663                 if (!period_set)
1664                         synth_opts->period = PERF_ITRACE_DEFAULT_PERIOD;
1665         }
1666
1667         return 0;
1668
1669 out_err:
1670         pr_err("Bad Instruction Tracing options '%s'\n", str);
1671         return -EINVAL;
1672 }
1673
1674 int itrace_parse_synth_opts(const struct option *opt, const char *str, int unset)
1675 {
1676         return itrace_do_parse_synth_opts(opt->value, str, unset);
1677 }
1678
1679 static const char * const auxtrace_error_type_name[] = {
1680         [PERF_AUXTRACE_ERROR_ITRACE] = "instruction trace",
1681 };
1682
1683 static const char *auxtrace_error_name(int type)
1684 {
1685         const char *error_type_name = NULL;
1686
1687         if (type < PERF_AUXTRACE_ERROR_MAX)
1688                 error_type_name = auxtrace_error_type_name[type];
1689         if (!error_type_name)
1690                 error_type_name = "unknown AUX";
1691         return error_type_name;
1692 }
1693
1694 size_t perf_event__fprintf_auxtrace_error(union perf_event *event, FILE *fp)
1695 {
1696         struct perf_record_auxtrace_error *e = &event->auxtrace_error;
1697         unsigned long long nsecs = e->time;
1698         const char *msg = e->msg;
1699         int ret;
1700
1701         ret = fprintf(fp, " %s error type %u",
1702                       auxtrace_error_name(e->type), e->type);
1703
1704         if (e->fmt && nsecs) {
1705                 unsigned long secs = nsecs / NSEC_PER_SEC;
1706
1707                 nsecs -= secs * NSEC_PER_SEC;
1708                 ret += fprintf(fp, " time %lu.%09llu", secs, nsecs);
1709         } else {
1710                 ret += fprintf(fp, " time 0");
1711         }
1712
1713         if (!e->fmt)
1714                 msg = (const char *)&e->time;
1715
1716         if (e->fmt >= 2 && e->machine_pid)
1717                 ret += fprintf(fp, " machine_pid %d vcpu %d", e->machine_pid, e->vcpu);
1718
1719         ret += fprintf(fp, " cpu %d pid %d tid %d ip %#"PRI_lx64" code %u: %s\n",
1720                        e->cpu, e->pid, e->tid, e->ip, e->code, msg);
1721         return ret;
1722 }
1723
1724 void perf_session__auxtrace_error_inc(struct perf_session *session,
1725                                       union perf_event *event)
1726 {
1727         struct perf_record_auxtrace_error *e = &event->auxtrace_error;
1728
1729         if (e->type < PERF_AUXTRACE_ERROR_MAX)
1730                 session->evlist->stats.nr_auxtrace_errors[e->type] += 1;
1731 }
1732
1733 void events_stats__auxtrace_error_warn(const struct events_stats *stats)
1734 {
1735         int i;
1736
1737         for (i = 0; i < PERF_AUXTRACE_ERROR_MAX; i++) {
1738                 if (!stats->nr_auxtrace_errors[i])
1739                         continue;
1740                 ui__warning("%u %s errors\n",
1741                             stats->nr_auxtrace_errors[i],
1742                             auxtrace_error_name(i));
1743         }
1744 }
1745
1746 int perf_event__process_auxtrace_error(struct perf_session *session,
1747                                        union perf_event *event)
1748 {
1749         if (auxtrace__dont_decode(session))
1750                 return 0;
1751
1752         perf_event__fprintf_auxtrace_error(event, stdout);
1753         return 0;
1754 }
1755
1756 /*
1757  * In the compat mode kernel runs in 64-bit and perf tool runs in 32-bit mode,
1758  * 32-bit perf tool cannot access 64-bit value atomically, which might lead to
1759  * the issues caused by the below sequence on multiple CPUs: when perf tool
1760  * accesses either the load operation or the store operation for 64-bit value,
1761  * on some architectures the operation is divided into two instructions, one
1762  * is for accessing the low 32-bit value and another is for the high 32-bit;
1763  * thus these two user operations can give the kernel chances to access the
1764  * 64-bit value, and thus leads to the unexpected load values.
1765  *
1766  *   kernel (64-bit)                        user (32-bit)
1767  *
1768  *   if (LOAD ->aux_tail) { --,             LOAD ->aux_head_lo
1769  *       STORE $aux_data      |       ,--->
1770  *       FLUSH $aux_data      |       |     LOAD ->aux_head_hi
1771  *       STORE ->aux_head   --|-------`     smp_rmb()
1772  *   }                        |             LOAD $data
1773  *                            |             smp_mb()
1774  *                            |             STORE ->aux_tail_lo
1775  *                            `----------->
1776  *                                          STORE ->aux_tail_hi
1777  *
1778  * For this reason, it's impossible for the perf tool to work correctly when
1779  * the AUX head or tail is bigger than 4GB (more than 32 bits length); and we
1780  * can not simply limit the AUX ring buffer to less than 4GB, the reason is
1781  * the pointers can be increased monotonically, whatever the buffer size it is,
1782  * at the end the head and tail can be bigger than 4GB and carry out to the
1783  * high 32-bit.
1784  *
1785  * To mitigate the issues and improve the user experience, we can allow the
1786  * perf tool working in certain conditions and bail out with error if detect
1787  * any overflow cannot be handled.
1788  *
1789  * For reading the AUX head, it reads out the values for three times, and
1790  * compares the high 4 bytes of the values between the first time and the last
1791  * time, if there has no change for high 4 bytes injected by the kernel during
1792  * the user reading sequence, it's safe for use the second value.
1793  *
1794  * When compat_auxtrace_mmap__write_tail() detects any carrying in the high
1795  * 32 bits, it means there have two store operations in user space and it cannot
1796  * promise the atomicity for 64-bit write, so return '-1' in this case to tell
1797  * the caller an overflow error has happened.
1798  */
1799 u64 __weak compat_auxtrace_mmap__read_head(struct auxtrace_mmap *mm)
1800 {
1801         struct perf_event_mmap_page *pc = mm->userpg;
1802         u64 first, second, last;
1803         u64 mask = (u64)(UINT32_MAX) << 32;
1804
1805         do {
1806                 first = READ_ONCE(pc->aux_head);
1807                 /* Ensure all reads are done after we read the head */
1808                 smp_rmb();
1809                 second = READ_ONCE(pc->aux_head);
1810                 /* Ensure all reads are done after we read the head */
1811                 smp_rmb();
1812                 last = READ_ONCE(pc->aux_head);
1813         } while ((first & mask) != (last & mask));
1814
1815         return second;
1816 }
1817
1818 int __weak compat_auxtrace_mmap__write_tail(struct auxtrace_mmap *mm, u64 tail)
1819 {
1820         struct perf_event_mmap_page *pc = mm->userpg;
1821         u64 mask = (u64)(UINT32_MAX) << 32;
1822
1823         if (tail & mask)
1824                 return -1;
1825
1826         /* Ensure all reads are done before we write the tail out */
1827         smp_mb();
1828         WRITE_ONCE(pc->aux_tail, tail);
1829         return 0;
1830 }
1831
1832 static int __auxtrace_mmap__read(struct mmap *map,
1833                                  struct auxtrace_record *itr,
1834                                  struct perf_tool *tool, process_auxtrace_t fn,
1835                                  bool snapshot, size_t snapshot_size)
1836 {
1837         struct auxtrace_mmap *mm = &map->auxtrace_mmap;
1838         u64 head, old = mm->prev, offset, ref;
1839         unsigned char *data = mm->base;
1840         size_t size, head_off, old_off, len1, len2, padding;
1841         union perf_event ev;
1842         void *data1, *data2;
1843         int kernel_is_64_bit = perf_env__kernel_is_64_bit(evsel__env(NULL));
1844
1845         head = auxtrace_mmap__read_head(mm, kernel_is_64_bit);
1846
1847         if (snapshot &&
1848             auxtrace_record__find_snapshot(itr, mm->idx, mm, data, &head, &old))
1849                 return -1;
1850
1851         if (old == head)
1852                 return 0;
1853
1854         pr_debug3("auxtrace idx %d old %#"PRIx64" head %#"PRIx64" diff %#"PRIx64"\n",
1855                   mm->idx, old, head, head - old);
1856
1857         if (mm->mask) {
1858                 head_off = head & mm->mask;
1859                 old_off = old & mm->mask;
1860         } else {
1861                 head_off = head % mm->len;
1862                 old_off = old % mm->len;
1863         }
1864
1865         if (head_off > old_off)
1866                 size = head_off - old_off;
1867         else
1868                 size = mm->len - (old_off - head_off);
1869
1870         if (snapshot && size > snapshot_size)
1871                 size = snapshot_size;
1872
1873         ref = auxtrace_record__reference(itr);
1874
1875         if (head > old || size <= head || mm->mask) {
1876                 offset = head - size;
1877         } else {
1878                 /*
1879                  * When the buffer size is not a power of 2, 'head' wraps at the
1880                  * highest multiple of the buffer size, so we have to subtract
1881                  * the remainder here.
1882                  */
1883                 u64 rem = (0ULL - mm->len) % mm->len;
1884
1885                 offset = head - size - rem;
1886         }
1887
1888         if (size > head_off) {
1889                 len1 = size - head_off;
1890                 data1 = &data[mm->len - len1];
1891                 len2 = head_off;
1892                 data2 = &data[0];
1893         } else {
1894                 len1 = size;
1895                 data1 = &data[head_off - len1];
1896                 len2 = 0;
1897                 data2 = NULL;
1898         }
1899
1900         if (itr->alignment) {
1901                 unsigned int unwanted = len1 % itr->alignment;
1902
1903                 len1 -= unwanted;
1904                 size -= unwanted;
1905         }
1906
1907         /* padding must be written by fn() e.g. record__process_auxtrace() */
1908         padding = size & (PERF_AUXTRACE_RECORD_ALIGNMENT - 1);
1909         if (padding)
1910                 padding = PERF_AUXTRACE_RECORD_ALIGNMENT - padding;
1911
1912         memset(&ev, 0, sizeof(ev));
1913         ev.auxtrace.header.type = PERF_RECORD_AUXTRACE;
1914         ev.auxtrace.header.size = sizeof(ev.auxtrace);
1915         ev.auxtrace.size = size + padding;
1916         ev.auxtrace.offset = offset;
1917         ev.auxtrace.reference = ref;
1918         ev.auxtrace.idx = mm->idx;
1919         ev.auxtrace.tid = mm->tid;
1920         ev.auxtrace.cpu = mm->cpu;
1921
1922         if (fn(tool, map, &ev, data1, len1, data2, len2))
1923                 return -1;
1924
1925         mm->prev = head;
1926
1927         if (!snapshot) {
1928                 int err;
1929
1930                 err = auxtrace_mmap__write_tail(mm, head, kernel_is_64_bit);
1931                 if (err < 0)
1932                         return err;
1933
1934                 if (itr->read_finish) {
1935                         err = itr->read_finish(itr, mm->idx);
1936                         if (err < 0)
1937                                 return err;
1938                 }
1939         }
1940
1941         return 1;
1942 }
1943
1944 int auxtrace_mmap__read(struct mmap *map, struct auxtrace_record *itr,
1945                         struct perf_tool *tool, process_auxtrace_t fn)
1946 {
1947         return __auxtrace_mmap__read(map, itr, tool, fn, false, 0);
1948 }
1949
1950 int auxtrace_mmap__read_snapshot(struct mmap *map,
1951                                  struct auxtrace_record *itr,
1952                                  struct perf_tool *tool, process_auxtrace_t fn,
1953                                  size_t snapshot_size)
1954 {
1955         return __auxtrace_mmap__read(map, itr, tool, fn, true, snapshot_size);
1956 }
1957
1958 /**
1959  * struct auxtrace_cache - hash table to implement a cache
1960  * @hashtable: the hashtable
1961  * @sz: hashtable size (number of hlists)
1962  * @entry_size: size of an entry
1963  * @limit: limit the number of entries to this maximum, when reached the cache
1964  *         is dropped and caching begins again with an empty cache
1965  * @cnt: current number of entries
1966  * @bits: hashtable size (@sz = 2^@bits)
1967  */
1968 struct auxtrace_cache {
1969         struct hlist_head *hashtable;
1970         size_t sz;
1971         size_t entry_size;
1972         size_t limit;
1973         size_t cnt;
1974         unsigned int bits;
1975 };
1976
1977 struct auxtrace_cache *auxtrace_cache__new(unsigned int bits, size_t entry_size,
1978                                            unsigned int limit_percent)
1979 {
1980         struct auxtrace_cache *c;
1981         struct hlist_head *ht;
1982         size_t sz, i;
1983
1984         c = zalloc(sizeof(struct auxtrace_cache));
1985         if (!c)
1986                 return NULL;
1987
1988         sz = 1UL << bits;
1989
1990         ht = calloc(sz, sizeof(struct hlist_head));
1991         if (!ht)
1992                 goto out_free;
1993
1994         for (i = 0; i < sz; i++)
1995                 INIT_HLIST_HEAD(&ht[i]);
1996
1997         c->hashtable = ht;
1998         c->sz = sz;
1999         c->entry_size = entry_size;
2000         c->limit = (c->sz * limit_percent) / 100;
2001         c->bits = bits;
2002
2003         return c;
2004
2005 out_free:
2006         free(c);
2007         return NULL;
2008 }
2009
2010 static void auxtrace_cache__drop(struct auxtrace_cache *c)
2011 {
2012         struct auxtrace_cache_entry *entry;
2013         struct hlist_node *tmp;
2014         size_t i;
2015
2016         if (!c)
2017                 return;
2018
2019         for (i = 0; i < c->sz; i++) {
2020                 hlist_for_each_entry_safe(entry, tmp, &c->hashtable[i], hash) {
2021                         hlist_del(&entry->hash);
2022                         auxtrace_cache__free_entry(c, entry);
2023                 }
2024         }
2025
2026         c->cnt = 0;
2027 }
2028
2029 void auxtrace_cache__free(struct auxtrace_cache *c)
2030 {
2031         if (!c)
2032                 return;
2033
2034         auxtrace_cache__drop(c);
2035         zfree(&c->hashtable);
2036         free(c);
2037 }
2038
2039 void *auxtrace_cache__alloc_entry(struct auxtrace_cache *c)
2040 {
2041         return malloc(c->entry_size);
2042 }
2043
2044 void auxtrace_cache__free_entry(struct auxtrace_cache *c __maybe_unused,
2045                                 void *entry)
2046 {
2047         free(entry);
2048 }
2049
2050 int auxtrace_cache__add(struct auxtrace_cache *c, u32 key,
2051                         struct auxtrace_cache_entry *entry)
2052 {
2053         if (c->limit && ++c->cnt > c->limit)
2054                 auxtrace_cache__drop(c);
2055
2056         entry->key = key;
2057         hlist_add_head(&entry->hash, &c->hashtable[hash_32(key, c->bits)]);
2058
2059         return 0;
2060 }
2061
2062 static struct auxtrace_cache_entry *auxtrace_cache__rm(struct auxtrace_cache *c,
2063                                                        u32 key)
2064 {
2065         struct auxtrace_cache_entry *entry;
2066         struct hlist_head *hlist;
2067         struct hlist_node *n;
2068
2069         if (!c)
2070                 return NULL;
2071
2072         hlist = &c->hashtable[hash_32(key, c->bits)];
2073         hlist_for_each_entry_safe(entry, n, hlist, hash) {
2074                 if (entry->key == key) {
2075                         hlist_del(&entry->hash);
2076                         return entry;
2077                 }
2078         }
2079
2080         return NULL;
2081 }
2082
2083 void auxtrace_cache__remove(struct auxtrace_cache *c, u32 key)
2084 {
2085         struct auxtrace_cache_entry *entry = auxtrace_cache__rm(c, key);
2086
2087         auxtrace_cache__free_entry(c, entry);
2088 }
2089
2090 void *auxtrace_cache__lookup(struct auxtrace_cache *c, u32 key)
2091 {
2092         struct auxtrace_cache_entry *entry;
2093         struct hlist_head *hlist;
2094
2095         if (!c)
2096                 return NULL;
2097
2098         hlist = &c->hashtable[hash_32(key, c->bits)];
2099         hlist_for_each_entry(entry, hlist, hash) {
2100                 if (entry->key == key)
2101                         return entry;
2102         }
2103
2104         return NULL;
2105 }
2106
2107 static void addr_filter__free_str(struct addr_filter *filt)
2108 {
2109         zfree(&filt->str);
2110         filt->action   = NULL;
2111         filt->sym_from = NULL;
2112         filt->sym_to   = NULL;
2113         filt->filename = NULL;
2114 }
2115
2116 static struct addr_filter *addr_filter__new(void)
2117 {
2118         struct addr_filter *filt = zalloc(sizeof(*filt));
2119
2120         if (filt)
2121                 INIT_LIST_HEAD(&filt->list);
2122
2123         return filt;
2124 }
2125
2126 static void addr_filter__free(struct addr_filter *filt)
2127 {
2128         if (filt)
2129                 addr_filter__free_str(filt);
2130         free(filt);
2131 }
2132
2133 static void addr_filters__add(struct addr_filters *filts,
2134                               struct addr_filter *filt)
2135 {
2136         list_add_tail(&filt->list, &filts->head);
2137         filts->cnt += 1;
2138 }
2139
2140 static void addr_filters__del(struct addr_filters *filts,
2141                               struct addr_filter *filt)
2142 {
2143         list_del_init(&filt->list);
2144         filts->cnt -= 1;
2145 }
2146
2147 void addr_filters__init(struct addr_filters *filts)
2148 {
2149         INIT_LIST_HEAD(&filts->head);
2150         filts->cnt = 0;
2151 }
2152
2153 void addr_filters__exit(struct addr_filters *filts)
2154 {
2155         struct addr_filter *filt, *n;
2156
2157         list_for_each_entry_safe(filt, n, &filts->head, list) {
2158                 addr_filters__del(filts, filt);
2159                 addr_filter__free(filt);
2160         }
2161 }
2162
2163 static int parse_num_or_str(char **inp, u64 *num, const char **str,
2164                             const char *str_delim)
2165 {
2166         *inp += strspn(*inp, " ");
2167
2168         if (isdigit(**inp)) {
2169                 char *endptr;
2170
2171                 if (!num)
2172                         return -EINVAL;
2173                 errno = 0;
2174                 *num = strtoull(*inp, &endptr, 0);
2175                 if (errno)
2176                         return -errno;
2177                 if (endptr == *inp)
2178                         return -EINVAL;
2179                 *inp = endptr;
2180         } else {
2181                 size_t n;
2182
2183                 if (!str)
2184                         return -EINVAL;
2185                 *inp += strspn(*inp, " ");
2186                 *str = *inp;
2187                 n = strcspn(*inp, str_delim);
2188                 if (!n)
2189                         return -EINVAL;
2190                 *inp += n;
2191                 if (**inp) {
2192                         **inp = '\0';
2193                         *inp += 1;
2194                 }
2195         }
2196         return 0;
2197 }
2198
2199 static int parse_action(struct addr_filter *filt)
2200 {
2201         if (!strcmp(filt->action, "filter")) {
2202                 filt->start = true;
2203                 filt->range = true;
2204         } else if (!strcmp(filt->action, "start")) {
2205                 filt->start = true;
2206         } else if (!strcmp(filt->action, "stop")) {
2207                 filt->start = false;
2208         } else if (!strcmp(filt->action, "tracestop")) {
2209                 filt->start = false;
2210                 filt->range = true;
2211                 filt->action += 5; /* Change 'tracestop' to 'stop' */
2212         } else {
2213                 return -EINVAL;
2214         }
2215         return 0;
2216 }
2217
2218 static int parse_sym_idx(char **inp, int *idx)
2219 {
2220         *idx = -1;
2221
2222         *inp += strspn(*inp, " ");
2223
2224         if (**inp != '#')
2225                 return 0;
2226
2227         *inp += 1;
2228
2229         if (**inp == 'g' || **inp == 'G') {
2230                 *inp += 1;
2231                 *idx = 0;
2232         } else {
2233                 unsigned long num;
2234                 char *endptr;
2235
2236                 errno = 0;
2237                 num = strtoul(*inp, &endptr, 0);
2238                 if (errno)
2239                         return -errno;
2240                 if (endptr == *inp || num > INT_MAX)
2241                         return -EINVAL;
2242                 *inp = endptr;
2243                 *idx = num;
2244         }
2245
2246         return 0;
2247 }
2248
2249 static int parse_addr_size(char **inp, u64 *num, const char **str, int *idx)
2250 {
2251         int err = parse_num_or_str(inp, num, str, " ");
2252
2253         if (!err && *str)
2254                 err = parse_sym_idx(inp, idx);
2255
2256         return err;
2257 }
2258
2259 static int parse_one_filter(struct addr_filter *filt, const char **filter_inp)
2260 {
2261         char *fstr;
2262         int err;
2263
2264         filt->str = fstr = strdup(*filter_inp);
2265         if (!fstr)
2266                 return -ENOMEM;
2267
2268         err = parse_num_or_str(&fstr, NULL, &filt->action, " ");
2269         if (err)
2270                 goto out_err;
2271
2272         err = parse_action(filt);
2273         if (err)
2274                 goto out_err;
2275
2276         err = parse_addr_size(&fstr, &filt->addr, &filt->sym_from,
2277                               &filt->sym_from_idx);
2278         if (err)
2279                 goto out_err;
2280
2281         fstr += strspn(fstr, " ");
2282
2283         if (*fstr == '/') {
2284                 fstr += 1;
2285                 err = parse_addr_size(&fstr, &filt->size, &filt->sym_to,
2286                                       &filt->sym_to_idx);
2287                 if (err)
2288                         goto out_err;
2289                 filt->range = true;
2290         }
2291
2292         fstr += strspn(fstr, " ");
2293
2294         if (*fstr == '@') {
2295                 fstr += 1;
2296                 err = parse_num_or_str(&fstr, NULL, &filt->filename, " ,");
2297                 if (err)
2298                         goto out_err;
2299         }
2300
2301         fstr += strspn(fstr, " ,");
2302
2303         *filter_inp += fstr - filt->str;
2304
2305         return 0;
2306
2307 out_err:
2308         addr_filter__free_str(filt);
2309
2310         return err;
2311 }
2312
2313 int addr_filters__parse_bare_filter(struct addr_filters *filts,
2314                                     const char *filter)
2315 {
2316         struct addr_filter *filt;
2317         const char *fstr = filter;
2318         int err;
2319
2320         while (*fstr) {
2321                 filt = addr_filter__new();
2322                 err = parse_one_filter(filt, &fstr);
2323                 if (err) {
2324                         addr_filter__free(filt);
2325                         addr_filters__exit(filts);
2326                         return err;
2327                 }
2328                 addr_filters__add(filts, filt);
2329         }
2330
2331         return 0;
2332 }
2333
2334 struct sym_args {
2335         const char      *name;
2336         u64             start;
2337         u64             size;
2338         int             idx;
2339         int             cnt;
2340         bool            started;
2341         bool            global;
2342         bool            selected;
2343         bool            duplicate;
2344         bool            near;
2345 };
2346
2347 static bool kern_sym_name_match(const char *kname, const char *name)
2348 {
2349         size_t n = strlen(name);
2350
2351         return !strcmp(kname, name) ||
2352                (!strncmp(kname, name, n) && kname[n] == '\t');
2353 }
2354
2355 static bool kern_sym_match(struct sym_args *args, const char *name, char type)
2356 {
2357         /* A function with the same name, and global or the n'th found or any */
2358         return kallsyms__is_function(type) &&
2359                kern_sym_name_match(name, args->name) &&
2360                ((args->global && isupper(type)) ||
2361                 (args->selected && ++(args->cnt) == args->idx) ||
2362                 (!args->global && !args->selected));
2363 }
2364
2365 static int find_kern_sym_cb(void *arg, const char *name, char type, u64 start)
2366 {
2367         struct sym_args *args = arg;
2368
2369         if (args->started) {
2370                 if (!args->size)
2371                         args->size = start - args->start;
2372                 if (args->selected) {
2373                         if (args->size)
2374                                 return 1;
2375                 } else if (kern_sym_match(args, name, type)) {
2376                         args->duplicate = true;
2377                         return 1;
2378                 }
2379         } else if (kern_sym_match(args, name, type)) {
2380                 args->started = true;
2381                 args->start = start;
2382         }
2383
2384         return 0;
2385 }
2386
2387 static int print_kern_sym_cb(void *arg, const char *name, char type, u64 start)
2388 {
2389         struct sym_args *args = arg;
2390
2391         if (kern_sym_match(args, name, type)) {
2392                 pr_err("#%d\t0x%"PRIx64"\t%c\t%s\n",
2393                        ++args->cnt, start, type, name);
2394                 args->near = true;
2395         } else if (args->near) {
2396                 args->near = false;
2397                 pr_err("\t\twhich is near\t\t%s\n", name);
2398         }
2399
2400         return 0;
2401 }
2402
2403 static int sym_not_found_error(const char *sym_name, int idx)
2404 {
2405         if (idx > 0) {
2406                 pr_err("N'th occurrence (N=%d) of symbol '%s' not found.\n",
2407                        idx, sym_name);
2408         } else if (!idx) {
2409                 pr_err("Global symbol '%s' not found.\n", sym_name);
2410         } else {
2411                 pr_err("Symbol '%s' not found.\n", sym_name);
2412         }
2413         pr_err("Note that symbols must be functions.\n");
2414
2415         return -EINVAL;
2416 }
2417
2418 static int find_kern_sym(const char *sym_name, u64 *start, u64 *size, int idx)
2419 {
2420         struct sym_args args = {
2421                 .name = sym_name,
2422                 .idx = idx,
2423                 .global = !idx,
2424                 .selected = idx > 0,
2425         };
2426         int err;
2427
2428         *start = 0;
2429         *size = 0;
2430
2431         err = kallsyms__parse("/proc/kallsyms", &args, find_kern_sym_cb);
2432         if (err < 0) {
2433                 pr_err("Failed to parse /proc/kallsyms\n");
2434                 return err;
2435         }
2436
2437         if (args.duplicate) {
2438                 pr_err("Multiple kernel symbols with name '%s'\n", sym_name);
2439                 args.cnt = 0;
2440                 kallsyms__parse("/proc/kallsyms", &args, print_kern_sym_cb);
2441                 pr_err("Disambiguate symbol name by inserting #n after the name e.g. %s #2\n",
2442                        sym_name);
2443                 pr_err("Or select a global symbol by inserting #0 or #g or #G\n");
2444                 return -EINVAL;
2445         }
2446
2447         if (!args.started) {
2448                 pr_err("Kernel symbol lookup: ");
2449                 return sym_not_found_error(sym_name, idx);
2450         }
2451
2452         *start = args.start;
2453         *size = args.size;
2454
2455         return 0;
2456 }
2457
2458 static int find_entire_kern_cb(void *arg, const char *name __maybe_unused,
2459                                char type, u64 start)
2460 {
2461         struct sym_args *args = arg;
2462         u64 size;
2463
2464         if (!kallsyms__is_function(type))
2465                 return 0;
2466
2467         if (!args->started) {
2468                 args->started = true;
2469                 args->start = start;
2470         }
2471         /* Don't know exactly where the kernel ends, so we add a page */
2472         size = round_up(start, page_size) + page_size - args->start;
2473         if (size > args->size)
2474                 args->size = size;
2475
2476         return 0;
2477 }
2478
2479 static int addr_filter__entire_kernel(struct addr_filter *filt)
2480 {
2481         struct sym_args args = { .started = false };
2482         int err;
2483
2484         err = kallsyms__parse("/proc/kallsyms", &args, find_entire_kern_cb);
2485         if (err < 0 || !args.started) {
2486                 pr_err("Failed to parse /proc/kallsyms\n");
2487                 return err;
2488         }
2489
2490         filt->addr = args.start;
2491         filt->size = args.size;
2492
2493         return 0;
2494 }
2495
2496 static int check_end_after_start(struct addr_filter *filt, u64 start, u64 size)
2497 {
2498         if (start + size >= filt->addr)
2499                 return 0;
2500
2501         if (filt->sym_from) {
2502                 pr_err("Symbol '%s' (0x%"PRIx64") comes before '%s' (0x%"PRIx64")\n",
2503                        filt->sym_to, start, filt->sym_from, filt->addr);
2504         } else {
2505                 pr_err("Symbol '%s' (0x%"PRIx64") comes before address 0x%"PRIx64")\n",
2506                        filt->sym_to, start, filt->addr);
2507         }
2508
2509         return -EINVAL;
2510 }
2511
2512 static int addr_filter__resolve_kernel_syms(struct addr_filter *filt)
2513 {
2514         bool no_size = false;
2515         u64 start, size;
2516         int err;
2517
2518         if (symbol_conf.kptr_restrict) {
2519                 pr_err("Kernel addresses are restricted. Unable to resolve kernel symbols.\n");
2520                 return -EINVAL;
2521         }
2522
2523         if (filt->sym_from && !strcmp(filt->sym_from, "*"))
2524                 return addr_filter__entire_kernel(filt);
2525
2526         if (filt->sym_from) {
2527                 err = find_kern_sym(filt->sym_from, &start, &size,
2528                                     filt->sym_from_idx);
2529                 if (err)
2530                         return err;
2531                 filt->addr = start;
2532                 if (filt->range && !filt->size && !filt->sym_to) {
2533                         filt->size = size;
2534                         no_size = !size;
2535                 }
2536         }
2537
2538         if (filt->sym_to) {
2539                 err = find_kern_sym(filt->sym_to, &start, &size,
2540                                     filt->sym_to_idx);
2541                 if (err)
2542                         return err;
2543
2544                 err = check_end_after_start(filt, start, size);
2545                 if (err)
2546                         return err;
2547                 filt->size = start + size - filt->addr;
2548                 no_size = !size;
2549         }
2550
2551         /* The very last symbol in kallsyms does not imply a particular size */
2552         if (no_size) {
2553                 pr_err("Cannot determine size of symbol '%s'\n",
2554                        filt->sym_to ? filt->sym_to : filt->sym_from);
2555                 return -EINVAL;
2556         }
2557
2558         return 0;
2559 }
2560
2561 static struct dso *load_dso(const char *name)
2562 {
2563         struct map *map;
2564         struct dso *dso;
2565
2566         map = dso__new_map(name);
2567         if (!map)
2568                 return NULL;
2569
2570         if (map__load(map) < 0)
2571                 pr_err("File '%s' not found or has no symbols.\n", name);
2572
2573         dso = dso__get(map__dso(map));
2574
2575         map__put(map);
2576
2577         return dso;
2578 }
2579
2580 static bool dso_sym_match(struct symbol *sym, const char *name, int *cnt,
2581                           int idx)
2582 {
2583         /* Same name, and global or the n'th found or any */
2584         return !arch__compare_symbol_names(name, sym->name) &&
2585                ((!idx && sym->binding == STB_GLOBAL) ||
2586                 (idx > 0 && ++*cnt == idx) ||
2587                 idx < 0);
2588 }
2589
2590 static void print_duplicate_syms(struct dso *dso, const char *sym_name)
2591 {
2592         struct symbol *sym;
2593         bool near = false;
2594         int cnt = 0;
2595
2596         pr_err("Multiple symbols with name '%s'\n", sym_name);
2597
2598         sym = dso__first_symbol(dso);
2599         while (sym) {
2600                 if (dso_sym_match(sym, sym_name, &cnt, -1)) {
2601                         pr_err("#%d\t0x%"PRIx64"\t%c\t%s\n",
2602                                ++cnt, sym->start,
2603                                sym->binding == STB_GLOBAL ? 'g' :
2604                                sym->binding == STB_LOCAL  ? 'l' : 'w',
2605                                sym->name);
2606                         near = true;
2607                 } else if (near) {
2608                         near = false;
2609                         pr_err("\t\twhich is near\t\t%s\n", sym->name);
2610                 }
2611                 sym = dso__next_symbol(sym);
2612         }
2613
2614         pr_err("Disambiguate symbol name by inserting #n after the name e.g. %s #2\n",
2615                sym_name);
2616         pr_err("Or select a global symbol by inserting #0 or #g or #G\n");
2617 }
2618
2619 static int find_dso_sym(struct dso *dso, const char *sym_name, u64 *start,
2620                         u64 *size, int idx)
2621 {
2622         struct symbol *sym;
2623         int cnt = 0;
2624
2625         *start = 0;
2626         *size = 0;
2627
2628         sym = dso__first_symbol(dso);
2629         while (sym) {
2630                 if (*start) {
2631                         if (!*size)
2632                                 *size = sym->start - *start;
2633                         if (idx > 0) {
2634                                 if (*size)
2635                                         return 0;
2636                         } else if (dso_sym_match(sym, sym_name, &cnt, idx)) {
2637                                 print_duplicate_syms(dso, sym_name);
2638                                 return -EINVAL;
2639                         }
2640                 } else if (dso_sym_match(sym, sym_name, &cnt, idx)) {
2641                         *start = sym->start;
2642                         *size = sym->end - sym->start;
2643                 }
2644                 sym = dso__next_symbol(sym);
2645         }
2646
2647         if (!*start)
2648                 return sym_not_found_error(sym_name, idx);
2649
2650         return 0;
2651 }
2652
2653 static int addr_filter__entire_dso(struct addr_filter *filt, struct dso *dso)
2654 {
2655         if (dso__data_file_size(dso, NULL)) {
2656                 pr_err("Failed to determine filter for %s\nCannot determine file size.\n",
2657                        filt->filename);
2658                 return -EINVAL;
2659         }
2660
2661         filt->addr = 0;
2662         filt->size = dso__data(dso)->file_size;
2663
2664         return 0;
2665 }
2666
2667 static int addr_filter__resolve_syms(struct addr_filter *filt)
2668 {
2669         u64 start, size;
2670         struct dso *dso;
2671         int err = 0;
2672
2673         if (!filt->sym_from && !filt->sym_to)
2674                 return 0;
2675
2676         if (!filt->filename)
2677                 return addr_filter__resolve_kernel_syms(filt);
2678
2679         dso = load_dso(filt->filename);
2680         if (!dso) {
2681                 pr_err("Failed to load symbols from: %s\n", filt->filename);
2682                 return -EINVAL;
2683         }
2684
2685         if (filt->sym_from && !strcmp(filt->sym_from, "*")) {
2686                 err = addr_filter__entire_dso(filt, dso);
2687                 goto put_dso;
2688         }
2689
2690         if (filt->sym_from) {
2691                 err = find_dso_sym(dso, filt->sym_from, &start, &size,
2692                                    filt->sym_from_idx);
2693                 if (err)
2694                         goto put_dso;
2695                 filt->addr = start;
2696                 if (filt->range && !filt->size && !filt->sym_to)
2697                         filt->size = size;
2698         }
2699
2700         if (filt->sym_to) {
2701                 err = find_dso_sym(dso, filt->sym_to, &start, &size,
2702                                    filt->sym_to_idx);
2703                 if (err)
2704                         goto put_dso;
2705
2706                 err = check_end_after_start(filt, start, size);
2707                 if (err)
2708                         return err;
2709
2710                 filt->size = start + size - filt->addr;
2711         }
2712
2713 put_dso:
2714         dso__put(dso);
2715
2716         return err;
2717 }
2718
2719 static char *addr_filter__to_str(struct addr_filter *filt)
2720 {
2721         char filename_buf[PATH_MAX];
2722         const char *at = "";
2723         const char *fn = "";
2724         char *filter;
2725         int err;
2726
2727         if (filt->filename) {
2728                 at = "@";
2729                 fn = realpath(filt->filename, filename_buf);
2730                 if (!fn)
2731                         return NULL;
2732         }
2733
2734         if (filt->range) {
2735                 err = asprintf(&filter, "%s 0x%"PRIx64"/0x%"PRIx64"%s%s",
2736                                filt->action, filt->addr, filt->size, at, fn);
2737         } else {
2738                 err = asprintf(&filter, "%s 0x%"PRIx64"%s%s",
2739                                filt->action, filt->addr, at, fn);
2740         }
2741
2742         return err < 0 ? NULL : filter;
2743 }
2744
2745 static int parse_addr_filter(struct evsel *evsel, const char *filter,
2746                              int max_nr)
2747 {
2748         struct addr_filters filts;
2749         struct addr_filter *filt;
2750         int err;
2751
2752         addr_filters__init(&filts);
2753
2754         err = addr_filters__parse_bare_filter(&filts, filter);
2755         if (err)
2756                 goto out_exit;
2757
2758         if (filts.cnt > max_nr) {
2759                 pr_err("Error: number of address filters (%d) exceeds maximum (%d)\n",
2760                        filts.cnt, max_nr);
2761                 err = -EINVAL;
2762                 goto out_exit;
2763         }
2764
2765         list_for_each_entry(filt, &filts.head, list) {
2766                 char *new_filter;
2767
2768                 err = addr_filter__resolve_syms(filt);
2769                 if (err)
2770                         goto out_exit;
2771
2772                 new_filter = addr_filter__to_str(filt);
2773                 if (!new_filter) {
2774                         err = -ENOMEM;
2775                         goto out_exit;
2776                 }
2777
2778                 if (evsel__append_addr_filter(evsel, new_filter)) {
2779                         err = -ENOMEM;
2780                         goto out_exit;
2781                 }
2782         }
2783
2784 out_exit:
2785         addr_filters__exit(&filts);
2786
2787         if (err) {
2788                 pr_err("Failed to parse address filter: '%s'\n", filter);
2789                 pr_err("Filter format is: filter|start|stop|tracestop <start symbol or address> [/ <end symbol or size>] [@<file name>]\n");
2790                 pr_err("Where multiple filters are separated by space or comma.\n");
2791         }
2792
2793         return err;
2794 }
2795
2796 static int evsel__nr_addr_filter(struct evsel *evsel)
2797 {
2798         struct perf_pmu *pmu = evsel__find_pmu(evsel);
2799         int nr_addr_filters = 0;
2800
2801         if (!pmu)
2802                 return 0;
2803
2804         perf_pmu__scan_file(pmu, "nr_addr_filters", "%d", &nr_addr_filters);
2805
2806         return nr_addr_filters;
2807 }
2808
2809 int auxtrace_parse_filters(struct evlist *evlist)
2810 {
2811         struct evsel *evsel;
2812         char *filter;
2813         int err, max_nr;
2814
2815         evlist__for_each_entry(evlist, evsel) {
2816                 filter = evsel->filter;
2817                 max_nr = evsel__nr_addr_filter(evsel);
2818                 if (!filter || !max_nr)
2819                         continue;
2820                 evsel->filter = NULL;
2821                 err = parse_addr_filter(evsel, filter, max_nr);
2822                 free(filter);
2823                 if (err)
2824                         return err;
2825                 pr_debug("Address filter: %s\n", evsel->filter);
2826         }
2827
2828         return 0;
2829 }
2830
2831 int auxtrace__process_event(struct perf_session *session, union perf_event *event,
2832                             struct perf_sample *sample, struct perf_tool *tool)
2833 {
2834         if (!session->auxtrace)
2835                 return 0;
2836
2837         return session->auxtrace->process_event(session, event, sample, tool);
2838 }
2839
2840 void auxtrace__dump_auxtrace_sample(struct perf_session *session,
2841                                     struct perf_sample *sample)
2842 {
2843         if (!session->auxtrace || !session->auxtrace->dump_auxtrace_sample ||
2844             auxtrace__dont_decode(session))
2845                 return;
2846
2847         session->auxtrace->dump_auxtrace_sample(session, sample);
2848 }
2849
2850 int auxtrace__flush_events(struct perf_session *session, struct perf_tool *tool)
2851 {
2852         if (!session->auxtrace)
2853                 return 0;
2854
2855         return session->auxtrace->flush_events(session, tool);
2856 }
2857
2858 void auxtrace__free_events(struct perf_session *session)
2859 {
2860         if (!session->auxtrace)
2861                 return;
2862
2863         return session->auxtrace->free_events(session);
2864 }
2865
2866 void auxtrace__free(struct perf_session *session)
2867 {
2868         if (!session->auxtrace)
2869                 return;
2870
2871         return session->auxtrace->free(session);
2872 }
2873
2874 bool auxtrace__evsel_is_auxtrace(struct perf_session *session,
2875                                  struct evsel *evsel)
2876 {
2877         if (!session->auxtrace || !session->auxtrace->evsel_is_auxtrace)
2878                 return false;
2879
2880         return session->auxtrace->evsel_is_auxtrace(session, evsel);
2881 }