tools/perf/util/header.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <errno.h>
   3 #include <inttypes.h>
   4 #include "string2.h"
   5 #include <sys/param.h>
   6 #include <sys/types.h>
   7 #include <byteswap.h>
   8 #include <unistd.h>
   9 #include <stdio.h>
  10 #include <stdlib.h>
  11 #include <linux/compiler.h>
  12 #include <linux/list.h>
  13 #include <linux/kernel.h>
  14 #include <linux/bitops.h>
  15 #include <linux/string.h>
  16 #include <linux/stringify.h>
  17 #include <linux/zalloc.h>
  18 #include <sys/stat.h>
  19 #include <sys/utsname.h>
  20 #include <linux/time64.h>
  21 #include <dirent.h>
  22 #ifdef HAVE_LIBBPF_SUPPORT
  23 #include <bpf/libbpf.h>
  24 #endif
  25 #include <perf/cpumap.h>
  26
  27 #include "dso.h"
  28 #include "evlist.h"
  29 #include "evsel.h"
  30 #include "util/evsel_fprintf.h"
  31 #include "header.h"
  32 #include "memswap.h"
  33 #include "trace-event.h"
  34 #include "session.h"
  35 #include "symbol.h"
  36 #include "debug.h"
  37 #include "cpumap.h"
  38 #include "pmu.h"
  39 #include "vdso.h"
  40 #include "strbuf.h"
  41 #include "build-id.h"
  42 #include "data.h"
  43 #include <api/fs/fs.h>
  44 #include "asm/bug.h"
  45 #include "tool.h"
  46 #include "time-utils.h"
  47 #include "units.h"
  48 #include "util/util.h" // perf_exe()
  49 #include "cputopo.h"
  50 #include "bpf-event.h"
  51 #include "clockid.h"
  52 #include "pmu-hybrid.h"
  53
  54 #include <linux/ctype.h>
  55 #include <internal/lib.h>
  56
  57 /*
  58  * magic2 = "PERFILE2"
  59  * must be a numerical value to let the endianness
  60  * determine the memory layout. That way we are able
  61  * to detect endianness when reading the perf.data file
  62  * back.
  63  *
  64  * we check for legacy (PERFFILE) format.
  65  */
  66 static const char *__perf_magic1 = "PERFFILE";
  67 static const u64 __perf_magic2    = 0x32454c4946524550ULL;
  68 static const u64 __perf_magic2_sw = 0x50455246494c4532ULL;
  69
  70 #define PERF_MAGIC      __perf_magic2
  71
  72 const char perf_version_string[] = PERF_VERSION;
  73
  74 struct perf_file_attr {
  75         struct perf_event_attr  attr;
  76         struct perf_file_section        ids;
  77 };
  78
  79 void perf_header__set_feat(struct perf_header *header, int feat)
  80 {
  81         set_bit(feat, header->adds_features);
  82 }
  83
  84 void perf_header__clear_feat(struct perf_header *header, int feat)
  85 {
  86         clear_bit(feat, header->adds_features);
  87 }
  88
  89 bool perf_header__has_feat(const struct perf_header *header, int feat)
  90 {
  91         return test_bit(feat, header->adds_features);
  92 }
  93
  94 static int __do_write_fd(struct feat_fd *ff, const void *buf, size_t size)
  95 {
  96         ssize_t ret = writen(ff->fd, buf, size);
  97
  98         if (ret != (ssize_t)size)
  99                 return ret < 0 ? (int)ret : -1;
 100         return 0;
 101 }
 102
 103 static int __do_write_buf(struct feat_fd *ff,  const void *buf, size_t size)
 104 {
 105         /* struct perf_event_header::size is u16 */
 106         const size_t max_size = 0xffff - sizeof(struct perf_event_header);
 107         size_t new_size = ff->size;
 108         void *addr;
 109
 110         if (size + ff->offset > max_size)
 111                 return -E2BIG;
 112
 113         while (size > (new_size - ff->offset))
 114                 new_size <<= 1;
 115         new_size = min(max_size, new_size);
 116
 117         if (ff->size < new_size) {
 118                 addr = realloc(ff->buf, new_size);
 119                 if (!addr)
 120                         return -ENOMEM;
 121                 ff->buf = addr;
 122                 ff->size = new_size;
 123         }
 124
 125         memcpy(ff->buf + ff->offset, buf, size);
 126         ff->offset += size;
 127
 128         return 0;
 129 }
 130
 131 /* Return: 0 if succeeded, -ERR if failed. */
 132 int do_write(struct feat_fd *ff, const void *buf, size_t size)
 133 {
 134         if (!ff->buf)
 135                 return __do_write_fd(ff, buf, size);
 136         return __do_write_buf(ff, buf, size);
 137 }
 138
 139 /* Return: 0 if succeeded, -ERR if failed. */
 140 static int do_write_bitmap(struct feat_fd *ff, unsigned long *set, u64 size)
 141 {
 142         u64 *p = (u64 *) set;
 143         int i, ret;
 144
 145         ret = do_write(ff, &size, sizeof(size));
 146         if (ret < 0)
 147                 return ret;
 148
 149         for (i = 0; (u64) i < BITS_TO_U64(size); i++) {
 150                 ret = do_write(ff, p + i, sizeof(*p));
 151                 if (ret < 0)
 152                         return ret;
 153         }
 154
 155         return 0;
 156 }
 157
 158 /* Return: 0 if succeeded, -ERR if failed. */
 159 int write_padded(struct feat_fd *ff, const void *bf,
 160                  size_t count, size_t count_aligned)
 161 {
 162         static const char zero_buf[NAME_ALIGN];
 163         int err = do_write(ff, bf, count);
 164
 165         if (!err)
 166                 err = do_write(ff, zero_buf, count_aligned - count);
 167
 168         return err;
 169 }
 170
 171 #define string_size(str)                                                \
 172         (PERF_ALIGN((strlen(str) + 1), NAME_ALIGN) + sizeof(u32))
 173
 174 /* Return: 0 if succeeded, -ERR if failed. */
 175 static int do_write_string(struct feat_fd *ff, const char *str)
 176 {
 177         u32 len, olen;
 178         int ret;
 179
 180         olen = strlen(str) + 1;
 181         len = PERF_ALIGN(olen, NAME_ALIGN);
 182
 183         /* write len, incl. \0 */
 184         ret = do_write(ff, &len, sizeof(len));
 185         if (ret < 0)
 186                 return ret;
 187
 188         return write_padded(ff, str, olen, len);
 189 }
 190
 191 static int __do_read_fd(struct feat_fd *ff, void *addr, ssize_t size)
 192 {
 193         ssize_t ret = readn(ff->fd, addr, size);
 194
 195         if (ret != size)
 196                 return ret < 0 ? (int)ret : -1;
 197         return 0;
 198 }
 199
 200 static int __do_read_buf(struct feat_fd *ff, void *addr, ssize_t size)
 201 {
 202         if (size > (ssize_t)ff->size - ff->offset)
 203                 return -1;
 204
 205         memcpy(addr, ff->buf + ff->offset, size);
 206         ff->offset += size;
 207
 208         return 0;
 209
 210 }
 211
 212 static int __do_read(struct feat_fd *ff, void *addr, ssize_t size)
 213 {
 214         if (!ff->buf)
 215                 return __do_read_fd(ff, addr, size);
 216         return __do_read_buf(ff, addr, size);
 217 }
 218
 219 static int do_read_u32(struct feat_fd *ff, u32 *addr)
 220 {
 221         int ret;
 222
 223         ret = __do_read(ff, addr, sizeof(*addr));
 224         if (ret)
 225                 return ret;
 226
 227         if (ff->ph->needs_swap)
 228                 *addr = bswap_32(*addr);
 229         return 0;
 230 }
 231
 232 static int do_read_u64(struct feat_fd *ff, u64 *addr)
 233 {
 234         int ret;
 235
 236         ret = __do_read(ff, addr, sizeof(*addr));
 237         if (ret)
 238                 return ret;
 239
 240         if (ff->ph->needs_swap)
 241                 *addr = bswap_64(*addr);
 242         return 0;
 243 }
 244
 245 static char *do_read_string(struct feat_fd *ff)
 246 {
 247         u32 len;
 248         char *buf;
 249
 250         if (do_read_u32(ff, &len))
 251                 return NULL;
 252
 253         buf = malloc(len);
 254         if (!buf)
 255                 return NULL;
 256
 257         if (!__do_read(ff, buf, len)) {
 258                 /*
 259                  * strings are padded by zeroes
 260                  * thus the actual strlen of buf
 261                  * may be less than len
 262                  */
 263                 return buf;
 264         }
 265
 266         free(buf);
 267         return NULL;
 268 }
 269
 270 /* Return: 0 if succeeded, -ERR if failed. */
 271 static int do_read_bitmap(struct feat_fd *ff, unsigned long **pset, u64 *psize)
 272 {
 273         unsigned long *set;
 274         u64 size, *p;
 275         int i, ret;
 276
 277         ret = do_read_u64(ff, &size);
 278         if (ret)
 279                 return ret;
 280
 281         set = bitmap_alloc(size);
 282         if (!set)
 283                 return -ENOMEM;
 284
 285         p = (u64 *) set;
 286
 287         for (i = 0; (u64) i < BITS_TO_U64(size); i++) {
 288                 ret = do_read_u64(ff, p + i);
 289                 if (ret < 0) {
 290                         free(set);
 291                         return ret;
 292                 }
 293         }
 294
 295         *pset  = set;
 296         *psize = size;
 297         return 0;
 298 }
 299
 300 static int write_tracing_data(struct feat_fd *ff,
 301                               struct evlist *evlist)
 302 {
 303         if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__))
 304                 return -1;
 305
 306         return read_tracing_data(ff->fd, &evlist->core.entries);
 307 }
 308
 309 static int write_build_id(struct feat_fd *ff,
 310                           struct evlist *evlist __maybe_unused)
 311 {
 312         struct perf_session *session;
 313         int err;
 314
 315         session = container_of(ff->ph, struct perf_session, header);
 316
 317         if (!perf_session__read_build_ids(session, true))
 318                 return -1;
 319
 320         if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__))
 321                 return -1;
 322
 323         err = perf_session__write_buildid_table(session, ff);
 324         if (err < 0) {
 325                 pr_debug("failed to write buildid table\n");
 326                 return err;
 327         }
 328         perf_session__cache_build_ids(session);
 329
 330         return 0;
 331 }
 332
 333 static int write_hostname(struct feat_fd *ff,
 334                           struct evlist *evlist __maybe_unused)
 335 {
 336         struct utsname uts;
 337         int ret;
 338
 339         ret = uname(&uts);
 340         if (ret < 0)
 341                 return -1;
 342
 343         return do_write_string(ff, uts.nodename);
 344 }
 345
 346 static int write_osrelease(struct feat_fd *ff,
 347                            struct evlist *evlist __maybe_unused)
 348 {
 349         struct utsname uts;
 350         int ret;
 351
 352         ret = uname(&uts);
 353         if (ret < 0)
 354                 return -1;
 355
 356         return do_write_string(ff, uts.release);
 357 }
 358
 359 static int write_arch(struct feat_fd *ff,
 360                       struct evlist *evlist __maybe_unused)
 361 {
 362         struct utsname uts;
 363         int ret;
 364
 365         ret = uname(&uts);
 366         if (ret < 0)
 367                 return -1;
 368
 369         return do_write_string(ff, uts.machine);
 370 }
 371
 372 static int write_version(struct feat_fd *ff,
 373                          struct evlist *evlist __maybe_unused)
 374 {
 375         return do_write_string(ff, perf_version_string);
 376 }
 377
 378 static int __write_cpudesc(struct feat_fd *ff, const char *cpuinfo_proc)
 379 {
 380         FILE *file;
 381         char *buf = NULL;
 382         char *s, *p;
 383         const char *search = cpuinfo_proc;
 384         size_t len = 0;
 385         int ret = -1;
 386
 387         if (!search)
 388                 return -1;
 389
 390         file = fopen("/proc/cpuinfo", "r");
 391         if (!file)
 392                 return -1;
 393
 394         while (getline(&buf, &len, file) > 0) {
 395                 ret = strncmp(buf, search, strlen(search));
 396                 if (!ret)
 397                         break;
 398         }
 399
 400         if (ret) {
 401                 ret = -1;
 402                 goto done;
 403         }
 404
 405         s = buf;
 406
 407         p = strchr(buf, ':');
 408         if (p && *(p+1) == ' ' && *(p+2))
 409                 s = p + 2;
 410         p = strchr(s, '\n');
 411         if (p)
 412                 *p = '\0';
 413
 414         /* squash extra space characters (branding string) */
 415         p = s;
 416         while (*p) {
 417                 if (isspace(*p)) {
 418                         char *r = p + 1;
 419                         char *q = skip_spaces(r);
 420                         *p = ' ';
 421                         if (q != (p+1))
 422                                 while ((*r++ = *q++));
 423                 }
 424                 p++;
 425         }
 426         ret = do_write_string(ff, s);
 427 done:
 428         free(buf);
 429         fclose(file);
 430         return ret;
 431 }
 432
 433 static int write_cpudesc(struct feat_fd *ff,
 434                        struct evlist *evlist __maybe_unused)
 435 {
 436 #if defined(__powerpc__) || defined(__hppa__) || defined(__sparc__)
 437 #define CPUINFO_PROC    { "cpu", }
 438 #elif defined(__s390__)
 439 #define CPUINFO_PROC    { "vendor_id", }
 440 #elif defined(__sh__)
 441 #define CPUINFO_PROC    { "cpu type", }
 442 #elif defined(__alpha__) || defined(__mips__)
 443 #define CPUINFO_PROC    { "cpu model", }
 444 #elif defined(__arm__)
 445 #define CPUINFO_PROC    { "model name", "Processor", }
 446 #elif defined(__arc__)
 447 #define CPUINFO_PROC    { "Processor", }
 448 #elif defined(__xtensa__)
 449 #define CPUINFO_PROC    { "core ID", }
 450 #else
 451 #define CPUINFO_PROC    { "model name", }
 452 #endif
 453         const char *cpuinfo_procs[] = CPUINFO_PROC;
 454 #undef CPUINFO_PROC
 455         unsigned int i;
 456
 457         for (i = 0; i < ARRAY_SIZE(cpuinfo_procs); i++) {
 458                 int ret;
 459                 ret = __write_cpudesc(ff, cpuinfo_procs[i]);
 460                 if (ret >= 0)
 461                         return ret;
 462         }
 463         return -1;
 464 }
 465
 466
 467 static int write_nrcpus(struct feat_fd *ff,
 468                         struct evlist *evlist __maybe_unused)
 469 {
 470         long nr;
 471         u32 nrc, nra;
 472         int ret;
 473
 474         nrc = cpu__max_present_cpu();
 475
 476         nr = sysconf(_SC_NPROCESSORS_ONLN);
 477         if (nr < 0)
 478                 return -1;
 479
 480         nra = (u32)(nr & UINT_MAX);
 481
 482         ret = do_write(ff, &nrc, sizeof(nrc));
 483         if (ret < 0)
 484                 return ret;
 485
 486         return do_write(ff, &nra, sizeof(nra));
 487 }
 488
 489 static int write_event_desc(struct feat_fd *ff,
 490                             struct evlist *evlist)
 491 {
 492         struct evsel *evsel;
 493         u32 nre, nri, sz;
 494         int ret;
 495
 496         nre = evlist->core.nr_entries;
 497
 498         /*
 499          * write number of events
 500          */
 501         ret = do_write(ff, &nre, sizeof(nre));
 502         if (ret < 0)
 503                 return ret;
 504
 505         /*
 506          * size of perf_event_attr struct
 507          */
 508         sz = (u32)sizeof(evsel->core.attr);
 509         ret = do_write(ff, &sz, sizeof(sz));
 510         if (ret < 0)
 511                 return ret;
 512
 513         evlist__for_each_entry(evlist, evsel) {
 514                 ret = do_write(ff, &evsel->core.attr, sz);
 515                 if (ret < 0)
 516                         return ret;
 517                 /*
 518                  * write number of unique id per event
 519                  * there is one id per instance of an event
 520                  *
 521                  * copy into an nri to be independent of the
 522                  * type of ids,
 523                  */
 524                 nri = evsel->core.ids;
 525                 ret = do_write(ff, &nri, sizeof(nri));
 526                 if (ret < 0)
 527                         return ret;
 528
 529                 /*
 530                  * write event string as passed on cmdline
 531                  */
 532                 ret = do_write_string(ff, evsel__name(evsel));
 533                 if (ret < 0)
 534                         return ret;
 535                 /*
 536                  * write unique ids for this event
 537                  */
 538                 ret = do_write(ff, evsel->core.id, evsel->core.ids * sizeof(u64));
 539                 if (ret < 0)
 540                         return ret;
 541         }
 542         return 0;
 543 }
 544
 545 static int write_cmdline(struct feat_fd *ff,
 546                          struct evlist *evlist __maybe_unused)
 547 {
 548         char pbuf[MAXPATHLEN], *buf;
 549         int i, ret, n;
 550
 551         /* actual path to perf binary */
 552         buf = perf_exe(pbuf, MAXPATHLEN);
 553
 554         /* account for binary path */
 555         n = perf_env.nr_cmdline + 1;
 556
 557         ret = do_write(ff, &n, sizeof(n));
 558         if (ret < 0)
 559                 return ret;
 560
 561         ret = do_write_string(ff, buf);
 562         if (ret < 0)
 563                 return ret;
 564
 565         for (i = 0 ; i < perf_env.nr_cmdline; i++) {
 566                 ret = do_write_string(ff, perf_env.cmdline_argv[i]);
 567                 if (ret < 0)
 568                         return ret;
 569         }
 570         return 0;
 571 }
 572
 573
 574 static int write_cpu_topology(struct feat_fd *ff,
 575                               struct evlist *evlist __maybe_unused)
 576 {
 577         struct cpu_topology *tp;
 578         u32 i;
 579         int ret, j;
 580
 581         tp = cpu_topology__new();
 582         if (!tp)
 583                 return -1;
 584
 585         ret = do_write(ff, &tp->core_sib, sizeof(tp->core_sib));
 586         if (ret < 0)
 587                 goto done;
 588
 589         for (i = 0; i < tp->core_sib; i++) {
 590                 ret = do_write_string(ff, tp->core_siblings[i]);
 591                 if (ret < 0)
 592                         goto done;
 593         }
 594         ret = do_write(ff, &tp->thread_sib, sizeof(tp->thread_sib));
 595         if (ret < 0)
 596                 goto done;
 597
 598         for (i = 0; i < tp->thread_sib; i++) {
 599                 ret = do_write_string(ff, tp->thread_siblings[i]);
 600                 if (ret < 0)
 601                         break;
 602         }
 603
 604         ret = perf_env__read_cpu_topology_map(&perf_env);
 605         if (ret < 0)
 606                 goto done;
 607
 608         for (j = 0; j < perf_env.nr_cpus_avail; j++) {
 609                 ret = do_write(ff, &perf_env.cpu[j].core_id,
 610                                sizeof(perf_env.cpu[j].core_id));
 611                 if (ret < 0)
 612                         return ret;
 613                 ret = do_write(ff, &perf_env.cpu[j].socket_id,
 614                                sizeof(perf_env.cpu[j].socket_id));
 615                 if (ret < 0)
 616                         return ret;
 617         }
 618
 619         if (!tp->die_sib)
 620                 goto done;
 621
 622         ret = do_write(ff, &tp->die_sib, sizeof(tp->die_sib));
 623         if (ret < 0)
 624                 goto done;
 625
 626         for (i = 0; i < tp->die_sib; i++) {
 627                 ret = do_write_string(ff, tp->die_siblings[i]);
 628                 if (ret < 0)
 629                         goto done;
 630         }
 631
 632         for (j = 0; j < perf_env.nr_cpus_avail; j++) {
 633                 ret = do_write(ff, &perf_env.cpu[j].die_id,
 634                                sizeof(perf_env.cpu[j].die_id));
 635                 if (ret < 0)
 636                         return ret;
 637         }
 638
 639 done:
 640         cpu_topology__delete(tp);
 641         return ret;
 642 }
 643
 644
 645
 646 static int write_total_mem(struct feat_fd *ff,
 647                            struct evlist *evlist __maybe_unused)
 648 {
 649         char *buf = NULL;
 650         FILE *fp;
 651         size_t len = 0;
 652         int ret = -1, n;
 653         uint64_t mem;
 654
 655         fp = fopen("/proc/meminfo", "r");
 656         if (!fp)
 657                 return -1;
 658
 659         while (getline(&buf, &len, fp) > 0) {
 660                 ret = strncmp(buf, "MemTotal:", 9);
 661                 if (!ret)
 662                         break;
 663         }
 664         if (!ret) {
 665                 n = sscanf(buf, "%*s %"PRIu64, &mem);
 666                 if (n == 1)
 667                         ret = do_write(ff, &mem, sizeof(mem));
 668         } else
 669                 ret = -1;
 670         free(buf);
 671         fclose(fp);
 672         return ret;
 673 }
 674
 675 static int write_numa_topology(struct feat_fd *ff,
 676                                struct evlist *evlist __maybe_unused)
 677 {
 678         struct numa_topology *tp;
 679         int ret = -1;
 680         u32 i;
 681
 682         tp = numa_topology__new();
 683         if (!tp)
 684                 return -ENOMEM;
 685
 686         ret = do_write(ff, &tp->nr, sizeof(u32));
 687         if (ret < 0)
 688                 goto err;
 689
 690         for (i = 0; i < tp->nr; i++) {
 691                 struct numa_topology_node *n = &tp->nodes[i];
 692
 693                 ret = do_write(ff, &n->node, sizeof(u32));
 694                 if (ret < 0)
 695                         goto err;
 696
 697                 ret = do_write(ff, &n->mem_total, sizeof(u64));
 698                 if (ret)
 699                         goto err;
 700
 701                 ret = do_write(ff, &n->mem_free, sizeof(u64));
 702                 if (ret)
 703                         goto err;
 704
 705                 ret = do_write_string(ff, n->cpus);
 706                 if (ret < 0)
 707                         goto err;
 708         }
 709
 710         ret = 0;
 711
 712 err:
 713         numa_topology__delete(tp);
 714         return ret;
 715 }
 716
 717 /*
 718  * File format:
 719  *
 720  * struct pmu_mappings {
 721  *      u32     pmu_num;
 722  *      struct pmu_map {
 723  *              u32     type;
 724  *              char    name[];
 725  *      }[pmu_num];
 726  * };
 727  */
 728
 729 static int write_pmu_mappings(struct feat_fd *ff,
 730                               struct evlist *evlist __maybe_unused)
 731 {
 732         struct perf_pmu *pmu = NULL;
 733         u32 pmu_num = 0;
 734         int ret;
 735
 736         /*
 737          * Do a first pass to count number of pmu to avoid lseek so this
 738          * works in pipe mode as well.
 739          */
 740         while ((pmu = perf_pmu__scan(pmu))) {
 741                 if (!pmu->name)
 742                         continue;
 743                 pmu_num++;
 744         }
 745
 746         ret = do_write(ff, &pmu_num, sizeof(pmu_num));
 747         if (ret < 0)
 748                 return ret;
 749
 750         while ((pmu = perf_pmu__scan(pmu))) {
 751                 if (!pmu->name)
 752                         continue;
 753
 754                 ret = do_write(ff, &pmu->type, sizeof(pmu->type));
 755                 if (ret < 0)
 756                         return ret;
 757
 758                 ret = do_write_string(ff, pmu->name);
 759                 if (ret < 0)
 760                         return ret;
 761         }
 762
 763         return 0;
 764 }
 765
 766 /*
 767  * File format:
 768  *
 769  * struct group_descs {
 770  *      u32     nr_groups;
 771  *      struct group_desc {
 772  *              char    name[];
 773  *              u32     leader_idx;
 774  *              u32     nr_members;
 775  *      }[nr_groups];
 776  * };
 777  */
 778 static int write_group_desc(struct feat_fd *ff,
 779                             struct evlist *evlist)
 780 {
 781         u32 nr_groups = evlist->nr_groups;
 782         struct evsel *evsel;
 783         int ret;
 784
 785         ret = do_write(ff, &nr_groups, sizeof(nr_groups));
 786         if (ret < 0)
 787                 return ret;
 788
 789         evlist__for_each_entry(evlist, evsel) {
 790                 if (evsel__is_group_leader(evsel) && evsel->core.nr_members > 1) {
 791                         const char *name = evsel->group_name ?: "{anon_group}";
 792                         u32 leader_idx = evsel->idx;
 793                         u32 nr_members = evsel->core.nr_members;
 794
 795                         ret = do_write_string(ff, name);
 796                         if (ret < 0)
 797                                 return ret;
 798
 799                         ret = do_write(ff, &leader_idx, sizeof(leader_idx));
 800                         if (ret < 0)
 801                                 return ret;
 802
 803                         ret = do_write(ff, &nr_members, sizeof(nr_members));
 804                         if (ret < 0)
 805                                 return ret;
 806                 }
 807         }
 808         return 0;
 809 }
 810
 811 /*
 812  * Return the CPU id as a raw string.
 813  *
 814  * Each architecture should provide a more precise id string that
 815  * can be use to match the architecture's "mapfile".
 816  */
 817 char * __weak get_cpuid_str(struct perf_pmu *pmu __maybe_unused)
 818 {
 819         return NULL;
 820 }
 821
 822 /* Return zero when the cpuid from the mapfile.csv matches the
 823  * cpuid string generated on this platform.
 824  * Otherwise return non-zero.
 825  */
 826 int __weak strcmp_cpuid_str(const char *mapcpuid, const char *cpuid)
 827 {
 828         regex_t re;
 829         regmatch_t pmatch[1];
 830         int match;
 831
 832         if (regcomp(&re, mapcpuid, REG_EXTENDED) != 0) {
 833                 /* Warn unable to generate match particular string. */
 834                 pr_info("Invalid regular expression %s\n", mapcpuid);
 835                 return 1;
 836         }
 837
 838         match = !regexec(&re, cpuid, 1, pmatch, 0);
 839         regfree(&re);
 840         if (match) {
 841                 size_t match_len = (pmatch[0].rm_eo - pmatch[0].rm_so);
 842
 843                 /* Verify the entire string matched. */
 844                 if (match_len == strlen(cpuid))
 845                         return 0;
 846         }
 847         return 1;
 848 }
 849
 850 /*
 851  * default get_cpuid(): nothing gets recorded
 852  * actual implementation must be in arch/$(SRCARCH)/util/header.c
 853  */
 854 int __weak get_cpuid(char *buffer __maybe_unused, size_t sz __maybe_unused)
 855 {
 856         return ENOSYS; /* Not implemented */
 857 }
 858
 859 static int write_cpuid(struct feat_fd *ff,
 860                        struct evlist *evlist __maybe_unused)
 861 {
 862         char buffer[64];
 863         int ret;
 864
 865         ret = get_cpuid(buffer, sizeof(buffer));
 866         if (ret)
 867                 return -1;
 868
 869         return do_write_string(ff, buffer);
 870 }
 871
 872 static int write_branch_stack(struct feat_fd *ff __maybe_unused,
 873                               struct evlist *evlist __maybe_unused)
 874 {
 875         return 0;
 876 }
 877
 878 static int write_auxtrace(struct feat_fd *ff,
 879                           struct evlist *evlist __maybe_unused)
 880 {
 881         struct perf_session *session;
 882         int err;
 883
 884         if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__))
 885                 return -1;
 886
 887         session = container_of(ff->ph, struct perf_session, header);
 888
 889         err = auxtrace_index__write(ff->fd, &session->auxtrace_index);
 890         if (err < 0)
 891                 pr_err("Failed to write auxtrace index\n");
 892         return err;
 893 }
 894
 895 static int write_clockid(struct feat_fd *ff,
 896                          struct evlist *evlist __maybe_unused)
 897 {
 898         return do_write(ff, &ff->ph->env.clock.clockid_res_ns,
 899                         sizeof(ff->ph->env.clock.clockid_res_ns));
 900 }
 901
 902 static int write_clock_data(struct feat_fd *ff,
 903                             struct evlist *evlist __maybe_unused)
 904 {
 905         u64 *data64;
 906         u32 data32;
 907         int ret;
 908
 909         /* version */
 910         data32 = 1;
 911
 912         ret = do_write(ff, &data32, sizeof(data32));
 913         if (ret < 0)
 914                 return ret;
 915
 916         /* clockid */
 917         data32 = ff->ph->env.clock.clockid;
 918
 919         ret = do_write(ff, &data32, sizeof(data32));
 920         if (ret < 0)
 921                 return ret;
 922
 923         /* TOD ref time */
 924         data64 = &ff->ph->env.clock.tod_ns;
 925
 926         ret = do_write(ff, data64, sizeof(*data64));
 927         if (ret < 0)
 928                 return ret;
 929
 930         /* clockid ref time */
 931         data64 = &ff->ph->env.clock.clockid_ns;
 932
 933         return do_write(ff, data64, sizeof(*data64));
 934 }
 935
 936 static int write_hybrid_topology(struct feat_fd *ff,
 937                                  struct evlist *evlist __maybe_unused)
 938 {
 939         struct hybrid_topology *tp;
 940         int ret;
 941         u32 i;
 942
 943         tp = hybrid_topology__new();
 944         if (!tp)
 945                 return -ENOENT;
 946
 947         ret = do_write(ff, &tp->nr, sizeof(u32));
 948         if (ret < 0)
 949                 goto err;
 950
 951         for (i = 0; i < tp->nr; i++) {
 952                 struct hybrid_topology_node *n = &tp->nodes[i];
 953
 954                 ret = do_write_string(ff, n->pmu_name);
 955                 if (ret < 0)
 956                         goto err;
 957
 958                 ret = do_write_string(ff, n->cpus);
 959                 if (ret < 0)
 960                         goto err;
 961         }
 962
 963         ret = 0;
 964
 965 err:
 966         hybrid_topology__delete(tp);
 967         return ret;
 968 }
 969
 970 static int write_dir_format(struct feat_fd *ff,
 971                             struct evlist *evlist __maybe_unused)
 972 {
 973         struct perf_session *session;
 974         struct perf_data *data;
 975
 976         session = container_of(ff->ph, struct perf_session, header);
 977         data = session->data;
 978
 979         if (WARN_ON(!perf_data__is_dir(data)))
 980                 return -1;
 981
 982         return do_write(ff, &data->dir.version, sizeof(data->dir.version));
 983 }
 984
 985 #ifdef HAVE_LIBBPF_SUPPORT
 986 static int write_bpf_prog_info(struct feat_fd *ff,
 987                                struct evlist *evlist __maybe_unused)
 988 {
 989         struct perf_env *env = &ff->ph->env;
 990         struct rb_root *root;
 991         struct rb_node *next;
 992         int ret;
 993
 994         down_read(&env->bpf_progs.lock);
 995
 996         ret = do_write(ff, &env->bpf_progs.infos_cnt,
 997                        sizeof(env->bpf_progs.infos_cnt));
 998         if (ret < 0)
 999                 goto out;
1000
1001         root = &env->bpf_progs.infos;
1002         next = rb_first(root);
1003         while (next) {
1004                 struct bpf_prog_info_node *node;
1005                 size_t len;
1006
1007                 node = rb_entry(next, struct bpf_prog_info_node, rb_node);
1008                 next = rb_next(&node->rb_node);
1009                 len = sizeof(struct bpf_prog_info_linear) +
1010                         node->info_linear->data_len;
1011
1012                 /* before writing to file, translate address to offset */
1013                 bpf_program__bpil_addr_to_offs(node->info_linear);
1014                 ret = do_write(ff, node->info_linear, len);
1015                 /*
1016                  * translate back to address even when do_write() fails,
1017                  * so that this function never changes the data.
1018                  */
1019                 bpf_program__bpil_offs_to_addr(node->info_linear);
1020                 if (ret < 0)
1021                         goto out;
1022         }
1023 out:
1024         up_read(&env->bpf_progs.lock);
1025         return ret;
1026 }
1027
1028 static int write_bpf_btf(struct feat_fd *ff,
1029                          struct evlist *evlist __maybe_unused)
1030 {
1031         struct perf_env *env = &ff->ph->env;
1032         struct rb_root *root;
1033         struct rb_node *next;
1034         int ret;
1035
1036         down_read(&env->bpf_progs.lock);
1037
1038         ret = do_write(ff, &env->bpf_progs.btfs_cnt,
1039                        sizeof(env->bpf_progs.btfs_cnt));
1040
1041         if (ret < 0)
1042                 goto out;
1043
1044         root = &env->bpf_progs.btfs;
1045         next = rb_first(root);
1046         while (next) {
1047                 struct btf_node *node;
1048
1049                 node = rb_entry(next, struct btf_node, rb_node);
1050                 next = rb_next(&node->rb_node);
1051                 ret = do_write(ff, &node->id,
1052                                sizeof(u32) * 2 + node->data_size);
1053                 if (ret < 0)
1054                         goto out;
1055         }
1056 out:
1057         up_read(&env->bpf_progs.lock);
1058         return ret;
1059 }
1060 #endif // HAVE_LIBBPF_SUPPORT
1061
1062 static int cpu_cache_level__sort(const void *a, const void *b)
1063 {
1064         struct cpu_cache_level *cache_a = (struct cpu_cache_level *)a;
1065         struct cpu_cache_level *cache_b = (struct cpu_cache_level *)b;
1066
1067         return cache_a->level - cache_b->level;
1068 }
1069
1070 static bool cpu_cache_level__cmp(struct cpu_cache_level *a, struct cpu_cache_level *b)
1071 {
1072         if (a->level != b->level)
1073                 return false;
1074
1075         if (a->line_size != b->line_size)
1076                 return false;
1077
1078         if (a->sets != b->sets)
1079                 return false;
1080
1081         if (a->ways != b->ways)
1082                 return false;
1083
1084         if (strcmp(a->type, b->type))
1085                 return false;
1086
1087         if (strcmp(a->size, b->size))
1088                 return false;
1089
1090         if (strcmp(a->map, b->map))
1091                 return false;
1092
1093         return true;
1094 }
1095
1096 static int cpu_cache_level__read(struct cpu_cache_level *cache, u32 cpu, u16 level)
1097 {
1098         char path[PATH_MAX], file[PATH_MAX];
1099         struct stat st;
1100         size_t len;
1101
1102         scnprintf(path, PATH_MAX, "devices/system/cpu/cpu%d/cache/index%d/", cpu, level);
1103         scnprintf(file, PATH_MAX, "%s/%s", sysfs__mountpoint(), path);
1104
1105         if (stat(file, &st))
1106                 return 1;
1107
1108         scnprintf(file, PATH_MAX, "%s/level", path);
1109         if (sysfs__read_int(file, (int *) &cache->level))
1110                 return -1;
1111
1112         scnprintf(file, PATH_MAX, "%s/coherency_line_size", path);
1113         if (sysfs__read_int(file, (int *) &cache->line_size))
1114                 return -1;
1115
1116         scnprintf(file, PATH_MAX, "%s/number_of_sets", path);
1117         if (sysfs__read_int(file, (int *) &cache->sets))
1118                 return -1;
1119
1120         scnprintf(file, PATH_MAX, "%s/ways_of_associativity", path);
1121         if (sysfs__read_int(file, (int *) &cache->ways))
1122                 return -1;
1123
1124         scnprintf(file, PATH_MAX, "%s/type", path);
1125         if (sysfs__read_str(file, &cache->type, &len))
1126                 return -1;
1127
1128         cache->type[len] = 0;
1129         cache->type = strim(cache->type);
1130
1131         scnprintf(file, PATH_MAX, "%s/size", path);
1132         if (sysfs__read_str(file, &cache->size, &len)) {
1133                 zfree(&cache->type);
1134                 return -1;
1135         }
1136
1137         cache->size[len] = 0;
1138         cache->size = strim(cache->size);
1139
1140         scnprintf(file, PATH_MAX, "%s/shared_cpu_list", path);
1141         if (sysfs__read_str(file, &cache->map, &len)) {
1142                 zfree(&cache->size);
1143                 zfree(&cache->type);
1144                 return -1;
1145         }
1146
1147         cache->map[len] = 0;
1148         cache->map = strim(cache->map);
1149         return 0;
1150 }
1151
1152 static void cpu_cache_level__fprintf(FILE *out, struct cpu_cache_level *c)
1153 {
1154         fprintf(out, "L%d %-15s %8s [%s]\n", c->level, c->type, c->size, c->map);
1155 }
1156
1157 #define MAX_CACHE_LVL 4
1158
1159 static int build_caches(struct cpu_cache_level caches[], u32 *cntp)
1160 {
1161         u32 i, cnt = 0;
1162         u32 nr, cpu;
1163         u16 level;
1164
1165         nr = cpu__max_cpu();
1166
1167         for (cpu = 0; cpu < nr; cpu++) {
1168                 for (level = 0; level < MAX_CACHE_LVL; level++) {
1169                         struct cpu_cache_level c;
1170                         int err;
1171
1172                         err = cpu_cache_level__read(&c, cpu, level);
1173                         if (err < 0)
1174                                 return err;
1175
1176                         if (err == 1)
1177                                 break;
1178
1179                         for (i = 0; i < cnt; i++) {
1180                                 if (cpu_cache_level__cmp(&c, &caches[i]))
1181                                         break;
1182                         }
1183
1184                         if (i == cnt)
1185                                 caches[cnt++] = c;
1186                         else
1187                                 cpu_cache_level__free(&c);
1188                 }
1189         }
1190         *cntp = cnt;
1191         return 0;
1192 }
1193
1194 static int write_cache(struct feat_fd *ff,
1195                        struct evlist *evlist __maybe_unused)
1196 {
1197         u32 max_caches = cpu__max_cpu() * MAX_CACHE_LVL;
1198         struct cpu_cache_level caches[max_caches];
1199         u32 cnt = 0, i, version = 1;
1200         int ret;
1201
1202         ret = build_caches(caches, &cnt);
1203         if (ret)
1204                 goto out;
1205
1206         qsort(&caches, cnt, sizeof(struct cpu_cache_level), cpu_cache_level__sort);
1207
1208         ret = do_write(ff, &version, sizeof(u32));
1209         if (ret < 0)
1210                 goto out;
1211
1212         ret = do_write(ff, &cnt, sizeof(u32));
1213         if (ret < 0)
1214                 goto out;
1215
1216         for (i = 0; i < cnt; i++) {
1217                 struct cpu_cache_level *c = &caches[i];
1218
1219                 #define _W(v)                                   \
1220                         ret = do_write(ff, &c->v, sizeof(u32)); \
1221                         if (ret < 0)                            \
1222                                 goto out;
1223
1224                 _W(level)
1225                 _W(line_size)
1226                 _W(sets)
1227                 _W(ways)
1228                 #undef _W
1229
1230                 #define _W(v)                                           \
1231                         ret = do_write_string(ff, (const char *) c->v); \
1232                         if (ret < 0)                                    \
1233                                 goto out;
1234
1235                 _W(type)
1236                 _W(size)
1237                 _W(map)
1238                 #undef _W
1239         }
1240
1241 out:
1242         for (i = 0; i < cnt; i++)
1243                 cpu_cache_level__free(&caches[i]);
1244         return ret;
1245 }
1246
1247 static int write_stat(struct feat_fd *ff __maybe_unused,
1248                       struct evlist *evlist __maybe_unused)
1249 {
1250         return 0;
1251 }
1252
1253 static int write_sample_time(struct feat_fd *ff,
1254                              struct evlist *evlist)
1255 {
1256         int ret;
1257
1258         ret = do_write(ff, &evlist->first_sample_time,
1259                        sizeof(evlist->first_sample_time));
1260         if (ret < 0)
1261                 return ret;
1262
1263         return do_write(ff, &evlist->last_sample_time,
1264                         sizeof(evlist->last_sample_time));
1265 }
1266
1267
1268 static int memory_node__read(struct memory_node *n, unsigned long idx)
1269 {
1270         unsigned int phys, size = 0;
1271         char path[PATH_MAX];
1272         struct dirent *ent;
1273         DIR *dir;
1274
1275 #define for_each_memory(mem, dir)                                       \
1276         while ((ent = readdir(dir)))                                    \
1277                 if (strcmp(ent->d_name, ".") &&                         \
1278                     strcmp(ent->d_name, "..") &&                        \
1279                     sscanf(ent->d_name, "memory%u", &mem) == 1)
1280
1281         scnprintf(path, PATH_MAX,
1282                   "%s/devices/system/node/node%lu",
1283                   sysfs__mountpoint(), idx);
1284
1285         dir = opendir(path);
1286         if (!dir) {
1287                 pr_warning("failed: cant' open memory sysfs data\n");
1288                 return -1;
1289         }
1290
1291         for_each_memory(phys, dir) {
1292                 size = max(phys, size);
1293         }
1294
1295         size++;
1296
1297         n->set = bitmap_alloc(size);
1298         if (!n->set) {
1299                 closedir(dir);
1300                 return -ENOMEM;
1301         }
1302
1303         n->node = idx;
1304         n->size = size;
1305
1306         rewinddir(dir);
1307
1308         for_each_memory(phys, dir) {
1309                 set_bit(phys, n->set);
1310         }
1311
1312         closedir(dir);
1313         return 0;
1314 }
1315
1316 static int memory_node__sort(const void *a, const void *b)
1317 {
1318         const struct memory_node *na = a;
1319         const struct memory_node *nb = b;
1320
1321         return na->node - nb->node;
1322 }
1323
1324 static int build_mem_topology(struct memory_node *nodes, u64 size, u64 *cntp)
1325 {
1326         char path[PATH_MAX];
1327         struct dirent *ent;
1328         DIR *dir;
1329         u64 cnt = 0;
1330         int ret = 0;
1331
1332         scnprintf(path, PATH_MAX, "%s/devices/system/node/",
1333                   sysfs__mountpoint());
1334
1335         dir = opendir(path);
1336         if (!dir) {
1337                 pr_debug2("%s: could't read %s, does this arch have topology information?\n",
1338                           __func__, path);
1339                 return -1;
1340         }
1341
1342         while (!ret && (ent = readdir(dir))) {
1343                 unsigned int idx;
1344                 int r;
1345
1346                 if (!strcmp(ent->d_name, ".") ||
1347                     !strcmp(ent->d_name, ".."))
1348                         continue;
1349
1350                 r = sscanf(ent->d_name, "node%u", &idx);
1351                 if (r != 1)
1352                         continue;
1353
1354                 if (WARN_ONCE(cnt >= size,
1355                         "failed to write MEM_TOPOLOGY, way too many nodes\n")) {
1356                         closedir(dir);
1357                         return -1;
1358                 }
1359
1360                 ret = memory_node__read(&nodes[cnt++], idx);
1361         }
1362
1363         *cntp = cnt;
1364         closedir(dir);
1365
1366         if (!ret)
1367                 qsort(nodes, cnt, sizeof(nodes[0]), memory_node__sort);
1368
1369         return ret;
1370 }
1371
1372 #define MAX_MEMORY_NODES 2000
1373
1374 /*
1375  * The MEM_TOPOLOGY holds physical memory map for every
1376  * node in system. The format of data is as follows:
1377  *
1378  *  0 - version          | for future changes
1379  *  8 - block_size_bytes | /sys/devices/system/memory/block_size_bytes
1380  * 16 - count            | number of nodes
1381  *
1382  * For each node we store map of physical indexes for
1383  * each node:
1384  *
1385  * 32 - node id          | node index
1386  * 40 - size             | size of bitmap
1387  * 48 - bitmap           | bitmap of memory indexes that belongs to node
1388  */
1389 static int write_mem_topology(struct feat_fd *ff __maybe_unused,
1390                               struct evlist *evlist __maybe_unused)
1391 {
1392         static struct memory_node nodes[MAX_MEMORY_NODES];
1393         u64 bsize, version = 1, i, nr;
1394         int ret;
1395
1396         ret = sysfs__read_xll("devices/system/memory/block_size_bytes",
1397                               (unsigned long long *) &bsize);
1398         if (ret)
1399                 return ret;
1400
1401         ret = build_mem_topology(&nodes[0], MAX_MEMORY_NODES, &nr);
1402         if (ret)
1403                 return ret;
1404
1405         ret = do_write(ff, &version, sizeof(version));
1406         if (ret < 0)
1407                 goto out;
1408
1409         ret = do_write(ff, &bsize, sizeof(bsize));
1410         if (ret < 0)
1411                 goto out;
1412
1413         ret = do_write(ff, &nr, sizeof(nr));
1414         if (ret < 0)
1415                 goto out;
1416
1417         for (i = 0; i < nr; i++) {
1418                 struct memory_node *n = &nodes[i];
1419
1420                 #define _W(v)                                           \
1421                         ret = do_write(ff, &n->v, sizeof(n->v));        \
1422                         if (ret < 0)                                    \
1423                                 goto out;
1424
1425                 _W(node)
1426                 _W(size)
1427
1428                 #undef _W
1429
1430                 ret = do_write_bitmap(ff, n->set, n->size);
1431                 if (ret < 0)
1432                         goto out;
1433         }
1434
1435 out:
1436         return ret;
1437 }
1438
1439 static int write_compressed(struct feat_fd *ff __maybe_unused,
1440                             struct evlist *evlist __maybe_unused)
1441 {
1442         int ret;
1443
1444         ret = do_write(ff, &(ff->ph->env.comp_ver), sizeof(ff->ph->env.comp_ver));
1445         if (ret)
1446                 return ret;
1447
1448         ret = do_write(ff, &(ff->ph->env.comp_type), sizeof(ff->ph->env.comp_type));
1449         if (ret)
1450                 return ret;
1451
1452         ret = do_write(ff, &(ff->ph->env.comp_level), sizeof(ff->ph->env.comp_level));
1453         if (ret)
1454                 return ret;
1455
1456         ret = do_write(ff, &(ff->ph->env.comp_ratio), sizeof(ff->ph->env.comp_ratio));
1457         if (ret)
1458                 return ret;
1459
1460         return do_write(ff, &(ff->ph->env.comp_mmap_len), sizeof(ff->ph->env.comp_mmap_len));
1461 }
1462
1463 static int write_per_cpu_pmu_caps(struct feat_fd *ff, struct perf_pmu *pmu,
1464                                   bool write_pmu)
1465 {
1466         struct perf_pmu_caps *caps = NULL;
1467         int nr_caps;
1468         int ret;
1469
1470         nr_caps = perf_pmu__caps_parse(pmu);
1471         if (nr_caps < 0)
1472                 return nr_caps;
1473
1474         ret = do_write(ff, &nr_caps, sizeof(nr_caps));
1475         if (ret < 0)
1476                 return ret;
1477
1478         list_for_each_entry(caps, &pmu->caps, list) {
1479                 ret = do_write_string(ff, caps->name);
1480                 if (ret < 0)
1481                         return ret;
1482
1483                 ret = do_write_string(ff, caps->value);
1484                 if (ret < 0)
1485                         return ret;
1486         }
1487
1488         if (write_pmu) {
1489                 ret = do_write_string(ff, pmu->name);
1490                 if (ret < 0)
1491                         return ret;
1492         }
1493
1494         return ret;
1495 }
1496
1497 static int write_cpu_pmu_caps(struct feat_fd *ff,
1498                               struct evlist *evlist __maybe_unused)
1499 {
1500         struct perf_pmu *cpu_pmu = perf_pmu__find("cpu");
1501
1502         if (!cpu_pmu)
1503                 return -ENOENT;
1504
1505         return write_per_cpu_pmu_caps(ff, cpu_pmu, false);
1506 }
1507
1508 static int write_hybrid_cpu_pmu_caps(struct feat_fd *ff,
1509                                      struct evlist *evlist __maybe_unused)
1510 {
1511         struct perf_pmu *pmu;
1512         u32 nr_pmu = perf_pmu__hybrid_pmu_num();
1513         int ret;
1514
1515         if (nr_pmu == 0)
1516                 return -ENOENT;
1517
1518         ret = do_write(ff, &nr_pmu, sizeof(nr_pmu));
1519         if (ret < 0)
1520                 return ret;
1521
1522         perf_pmu__for_each_hybrid_pmu(pmu) {
1523                 ret = write_per_cpu_pmu_caps(ff, pmu, true);
1524                 if (ret < 0)
1525                         return ret;
1526         }
1527
1528         return 0;
1529 }
1530
1531 static void print_hostname(struct feat_fd *ff, FILE *fp)
1532 {
1533         fprintf(fp, "# hostname : %s\n", ff->ph->env.hostname);
1534 }
1535
1536 static void print_osrelease(struct feat_fd *ff, FILE *fp)
1537 {
1538         fprintf(fp, "# os release : %s\n", ff->ph->env.os_release);
1539 }
1540
1541 static void print_arch(struct feat_fd *ff, FILE *fp)
1542 {
1543         fprintf(fp, "# arch : %s\n", ff->ph->env.arch);
1544 }
1545
1546 static void print_cpudesc(struct feat_fd *ff, FILE *fp)
1547 {
1548         fprintf(fp, "# cpudesc : %s\n", ff->ph->env.cpu_desc);
1549 }
1550
1551 static void print_nrcpus(struct feat_fd *ff, FILE *fp)
1552 {
1553         fprintf(fp, "# nrcpus online : %u\n", ff->ph->env.nr_cpus_online);
1554         fprintf(fp, "# nrcpus avail : %u\n", ff->ph->env.nr_cpus_avail);
1555 }
1556
1557 static void print_version(struct feat_fd *ff, FILE *fp)
1558 {
1559         fprintf(fp, "# perf version : %s\n", ff->ph->env.version);
1560 }
1561
1562 static void print_cmdline(struct feat_fd *ff, FILE *fp)
1563 {
1564         int nr, i;
1565
1566         nr = ff->ph->env.nr_cmdline;
1567
1568         fprintf(fp, "# cmdline : ");
1569
1570         for (i = 0; i < nr; i++) {
1571                 char *argv_i = strdup(ff->ph->env.cmdline_argv[i]);
1572                 if (!argv_i) {
1573                         fprintf(fp, "%s ", ff->ph->env.cmdline_argv[i]);
1574                 } else {
1575                         char *mem = argv_i;
1576                         do {
1577                                 char *quote = strchr(argv_i, '\'');
1578                                 if (!quote)
1579                                         break;
1580                                 *quote++ = '\0';
1581                                 fprintf(fp, "%s\\\'", argv_i);
1582                                 argv_i = quote;
1583                         } while (1);
1584                         fprintf(fp, "%s ", argv_i);
1585                         free(mem);
1586                 }
1587         }
1588         fputc('\n', fp);
1589 }
1590
1591 static void print_cpu_topology(struct feat_fd *ff, FILE *fp)
1592 {
1593         struct perf_header *ph = ff->ph;
1594         int cpu_nr = ph->env.nr_cpus_avail;
1595         int nr, i;
1596         char *str;
1597
1598         nr = ph->env.nr_sibling_cores;
1599         str = ph->env.sibling_cores;
1600
1601         for (i = 0; i < nr; i++) {
1602                 fprintf(fp, "# sibling sockets : %s\n", str);
1603                 str += strlen(str) + 1;
1604         }
1605
1606         if (ph->env.nr_sibling_dies) {
1607                 nr = ph->env.nr_sibling_dies;
1608                 str = ph->env.sibling_dies;
1609
1610                 for (i = 0; i < nr; i++) {
1611                         fprintf(fp, "# sibling dies    : %s\n", str);
1612                         str += strlen(str) + 1;
1613                 }
1614         }
1615
1616         nr = ph->env.nr_sibling_threads;
1617         str = ph->env.sibling_threads;
1618
1619         for (i = 0; i < nr; i++) {
1620                 fprintf(fp, "# sibling threads : %s\n", str);
1621                 str += strlen(str) + 1;
1622         }
1623
1624         if (ph->env.nr_sibling_dies) {
1625                 if (ph->env.cpu != NULL) {
1626                         for (i = 0; i < cpu_nr; i++)
1627                                 fprintf(fp, "# CPU %d: Core ID %d, "
1628                                             "Die ID %d, Socket ID %d\n",
1629                                             i, ph->env.cpu[i].core_id,
1630                                             ph->env.cpu[i].die_id,
1631                                             ph->env.cpu[i].socket_id);
1632                 } else
1633                         fprintf(fp, "# Core ID, Die ID and Socket ID "
1634                                     "information is not available\n");
1635         } else {
1636                 if (ph->env.cpu != NULL) {
1637                         for (i = 0; i < cpu_nr; i++)
1638                                 fprintf(fp, "# CPU %d: Core ID %d, "
1639                                             "Socket ID %d\n",
1640                                             i, ph->env.cpu[i].core_id,
1641                                             ph->env.cpu[i].socket_id);
1642                 } else
1643                         fprintf(fp, "# Core ID and Socket ID "
1644                                     "information is not available\n");
1645         }
1646 }
1647
1648 static void print_clockid(struct feat_fd *ff, FILE *fp)
1649 {
1650         fprintf(fp, "# clockid frequency: %"PRIu64" MHz\n",
1651                 ff->ph->env.clock.clockid_res_ns * 1000);
1652 }
1653
1654 static void print_clock_data(struct feat_fd *ff, FILE *fp)
1655 {
1656         struct timespec clockid_ns;
1657         char tstr[64], date[64];
1658         struct timeval tod_ns;
1659         clockid_t clockid;
1660         struct tm ltime;
1661         u64 ref;
1662
1663         if (!ff->ph->env.clock.enabled) {
1664                 fprintf(fp, "# reference time disabled\n");
1665                 return;
1666         }
1667
1668         /* Compute TOD time. */
1669         ref = ff->ph->env.clock.tod_ns;
1670         tod_ns.tv_sec = ref / NSEC_PER_SEC;
1671         ref -= tod_ns.tv_sec * NSEC_PER_SEC;
1672         tod_ns.tv_usec = ref / NSEC_PER_USEC;
1673
1674         /* Compute clockid time. */
1675         ref = ff->ph->env.clock.clockid_ns;
1676         clockid_ns.tv_sec = ref / NSEC_PER_SEC;
1677         ref -= clockid_ns.tv_sec * NSEC_PER_SEC;
1678         clockid_ns.tv_nsec = ref;
1679
1680         clockid = ff->ph->env.clock.clockid;
1681
1682         if (localtime_r(&tod_ns.tv_sec, &ltime) == NULL)
1683                 snprintf(tstr, sizeof(tstr), "<error>");
1684         else {
1685                 strftime(date, sizeof(date), "%F %T", &ltime);
1686                 scnprintf(tstr, sizeof(tstr), "%s.%06d",
1687                           date, (int) tod_ns.tv_usec);
1688         }
1689
1690         fprintf(fp, "# clockid: %s (%u)\n", clockid_name(clockid), clockid);
1691         fprintf(fp, "# reference time: %s = %ld.%06d (TOD) = %ld.%09ld (%s)\n",
1692                     tstr, (long) tod_ns.tv_sec, (int) tod_ns.tv_usec,
1693                     (long) clockid_ns.tv_sec, clockid_ns.tv_nsec,
1694                     clockid_name(clockid));
1695 }
1696
1697 static void print_hybrid_topology(struct feat_fd *ff, FILE *fp)
1698 {
1699         int i;
1700         struct hybrid_node *n;
1701
1702         fprintf(fp, "# hybrid cpu system:\n");
1703         for (i = 0; i < ff->ph->env.nr_hybrid_nodes; i++) {
1704                 n = &ff->ph->env.hybrid_nodes[i];
1705                 fprintf(fp, "# %s cpu list : %s\n", n->pmu_name, n->cpus);
1706         }
1707 }
1708
1709 static void print_dir_format(struct feat_fd *ff, FILE *fp)
1710 {
1711         struct perf_session *session;
1712         struct perf_data *data;
1713
1714         session = container_of(ff->ph, struct perf_session, header);
1715         data = session->data;
1716
1717         fprintf(fp, "# directory data version : %"PRIu64"\n", data->dir.version);
1718 }
1719
1720 #ifdef HAVE_LIBBPF_SUPPORT
1721 static void print_bpf_prog_info(struct feat_fd *ff, FILE *fp)
1722 {
1723         struct perf_env *env = &ff->ph->env;
1724         struct rb_root *root;
1725         struct rb_node *next;
1726
1727         down_read(&env->bpf_progs.lock);
1728
1729         root = &env->bpf_progs.infos;
1730         next = rb_first(root);
1731
1732         while (next) {
1733                 struct bpf_prog_info_node *node;
1734
1735                 node = rb_entry(next, struct bpf_prog_info_node, rb_node);
1736                 next = rb_next(&node->rb_node);
1737
1738                 bpf_event__print_bpf_prog_info(&node->info_linear->info,
1739                                                env, fp);
1740         }
1741
1742         up_read(&env->bpf_progs.lock);
1743 }
1744
1745 static void print_bpf_btf(struct feat_fd *ff, FILE *fp)
1746 {
1747         struct perf_env *env = &ff->ph->env;
1748         struct rb_root *root;
1749         struct rb_node *next;
1750
1751         down_read(&env->bpf_progs.lock);
1752
1753         root = &env->bpf_progs.btfs;
1754         next = rb_first(root);
1755
1756         while (next) {
1757                 struct btf_node *node;
1758
1759                 node = rb_entry(next, struct btf_node, rb_node);
1760                 next = rb_next(&node->rb_node);
1761                 fprintf(fp, "# btf info of id %u\n", node->id);
1762         }
1763
1764         up_read(&env->bpf_progs.lock);
1765 }
1766 #endif // HAVE_LIBBPF_SUPPORT
1767
1768 static void free_event_desc(struct evsel *events)
1769 {
1770         struct evsel *evsel;
1771
1772         if (!events)
1773                 return;
1774
1775         for (evsel = events; evsel->core.attr.size; evsel++) {
1776                 zfree(&evsel->name);
1777                 zfree(&evsel->core.id);
1778         }
1779
1780         free(events);
1781 }
1782
1783 static bool perf_attr_check(struct perf_event_attr *attr)
1784 {
1785         if (attr->__reserved_1 || attr->__reserved_2 || attr->__reserved_3) {
1786                 pr_warning("Reserved bits are set unexpectedly. "
1787                            "Please update perf tool.\n");
1788                 return false;
1789         }
1790
1791         if (attr->sample_type & ~(PERF_SAMPLE_MAX-1)) {
1792                 pr_warning("Unknown sample type (0x%llx) is detected. "
1793                            "Please update perf tool.\n",
1794                            attr->sample_type);
1795                 return false;
1796         }
1797
1798         if (attr->read_format & ~(PERF_FORMAT_MAX-1)) {
1799                 pr_warning("Unknown read format (0x%llx) is detected. "
1800                            "Please update perf tool.\n",
1801                            attr->read_format);
1802                 return false;
1803         }
1804
1805         if ((attr->sample_type & PERF_SAMPLE_BRANCH_STACK) &&
1806             (attr->branch_sample_type & ~(PERF_SAMPLE_BRANCH_MAX-1))) {
1807                 pr_warning("Unknown branch sample type (0x%llx) is detected. "
1808                            "Please update perf tool.\n",
1809                            attr->branch_sample_type);
1810
1811                 return false;
1812         }
1813
1814         return true;
1815 }
1816
1817 static struct evsel *read_event_desc(struct feat_fd *ff)
1818 {
1819         struct evsel *evsel, *events = NULL;
1820         u64 *id;
1821         void *buf = NULL;
1822         u32 nre, sz, nr, i, j;
1823         size_t msz;
1824
1825         /* number of events */
1826         if (do_read_u32(ff, &nre))
1827                 goto error;
1828
1829         if (do_read_u32(ff, &sz))
1830                 goto error;
1831
1832         /* buffer to hold on file attr struct */
1833         buf = malloc(sz);
1834         if (!buf)
1835                 goto error;
1836
1837         /* the last event terminates with evsel->core.attr.size == 0: */
1838         events = calloc(nre + 1, sizeof(*events));
1839         if (!events)
1840                 goto error;
1841
1842         msz = sizeof(evsel->core.attr);
1843         if (sz < msz)
1844                 msz = sz;
1845
1846         for (i = 0, evsel = events; i < nre; evsel++, i++) {
1847                 evsel->idx = i;
1848
1849                 /*
1850                  * must read entire on-file attr struct to
1851                  * sync up with layout.
1852                  */
1853                 if (__do_read(ff, buf, sz))
1854                         goto error;
1855
1856                 if (ff->ph->needs_swap)
1857                         perf_event__attr_swap(buf);
1858
1859                 memcpy(&evsel->core.attr, buf, msz);
1860
1861                 if (!perf_attr_check(&evsel->core.attr))
1862                         goto error;
1863
1864                 if (do_read_u32(ff, &nr))
1865                         goto error;
1866
1867                 if (ff->ph->needs_swap)
1868                         evsel->needs_swap = true;
1869
1870                 evsel->name = do_read_string(ff);
1871                 if (!evsel->name)
1872                         goto error;
1873
1874                 if (!nr)
1875                         continue;
1876
1877                 id = calloc(nr, sizeof(*id));
1878                 if (!id)
1879                         goto error;
1880                 evsel->core.ids = nr;
1881                 evsel->core.id = id;
1882
1883                 for (j = 0 ; j < nr; j++) {
1884                         if (do_read_u64(ff, id))
1885                                 goto error;
1886                         id++;
1887                 }
1888         }
1889 out:
1890         free(buf);
1891         return events;
1892 error:
1893         free_event_desc(events);
1894         events = NULL;
1895         goto out;
1896 }
1897
1898 static int __desc_attr__fprintf(FILE *fp, const char *name, const char *val,
1899                                 void *priv __maybe_unused)
1900 {
1901         return fprintf(fp, ", %s = %s", name, val);
1902 }
1903
1904 static void print_event_desc(struct feat_fd *ff, FILE *fp)
1905 {
1906         struct evsel *evsel, *events;
1907         u32 j;
1908         u64 *id;
1909
1910         if (ff->events)
1911                 events = ff->events;
1912         else
1913                 events = read_event_desc(ff);
1914
1915         if (!events) {
1916                 fprintf(fp, "# event desc: not available or unable to read\n");
1917                 return;
1918         }
1919
1920         for (evsel = events; evsel->core.attr.size; evsel++) {
1921                 fprintf(fp, "# event : name = %s, ", evsel->name);
1922
1923                 if (evsel->core.ids) {
1924                         fprintf(fp, ", id = {");
1925                         for (j = 0, id = evsel->core.id; j < evsel->core.ids; j++, id++) {
1926                                 if (j)
1927                                         fputc(',', fp);
1928                                 fprintf(fp, " %"PRIu64, *id);
1929                         }
1930                         fprintf(fp, " }");
1931                 }
1932
1933                 perf_event_attr__fprintf(fp, &evsel->core.attr, __desc_attr__fprintf, NULL);
1934
1935                 fputc('\n', fp);
1936         }
1937
1938         free_event_desc(events);
1939         ff->events = NULL;
1940 }
1941
1942 static void print_total_mem(struct feat_fd *ff, FILE *fp)
1943 {
1944         fprintf(fp, "# total memory : %llu kB\n", ff->ph->env.total_mem);
1945 }
1946
1947 static void print_numa_topology(struct feat_fd *ff, FILE *fp)
1948 {
1949         int i;
1950         struct numa_node *n;
1951
1952         for (i = 0; i < ff->ph->env.nr_numa_nodes; i++) {
1953                 n = &ff->ph->env.numa_nodes[i];
1954
1955                 fprintf(fp, "# node%u meminfo  : total = %"PRIu64" kB,"
1956                             " free = %"PRIu64" kB\n",
1957                         n->node, n->mem_total, n->mem_free);
1958
1959                 fprintf(fp, "# node%u cpu list : ", n->node);
1960                 cpu_map__fprintf(n->map, fp);
1961         }
1962 }
1963
1964 static void print_cpuid(struct feat_fd *ff, FILE *fp)
1965 {
1966         fprintf(fp, "# cpuid : %s\n", ff->ph->env.cpuid);
1967 }
1968
1969 static void print_branch_stack(struct feat_fd *ff __maybe_unused, FILE *fp)
1970 {
1971         fprintf(fp, "# contains samples with branch stack\n");
1972 }
1973
1974 static void print_auxtrace(struct feat_fd *ff __maybe_unused, FILE *fp)
1975 {
1976         fprintf(fp, "# contains AUX area data (e.g. instruction trace)\n");
1977 }
1978
1979 static void print_stat(struct feat_fd *ff __maybe_unused, FILE *fp)
1980 {
1981         fprintf(fp, "# contains stat data\n");
1982 }
1983
1984 static void print_cache(struct feat_fd *ff, FILE *fp __maybe_unused)
1985 {
1986         int i;
1987
1988         fprintf(fp, "# CPU cache info:\n");
1989         for (i = 0; i < ff->ph->env.caches_cnt; i++) {
1990                 fprintf(fp, "#  ");
1991                 cpu_cache_level__fprintf(fp, &ff->ph->env.caches[i]);
1992         }
1993 }
1994
1995 static void print_compressed(struct feat_fd *ff, FILE *fp)
1996 {
1997         fprintf(fp, "# compressed : %s, level = %d, ratio = %d\n",
1998                 ff->ph->env.comp_type == PERF_COMP_ZSTD ? "Zstd" : "Unknown",
1999                 ff->ph->env.comp_level, ff->ph->env.comp_ratio);
2000 }
2001
2002 static void print_per_cpu_pmu_caps(FILE *fp, int nr_caps, char *cpu_pmu_caps,
2003                                    char *pmu_name)
2004 {
2005         const char *delimiter;
2006         char *str, buf[128];
2007
2008         if (!nr_caps) {
2009                 if (!pmu_name)
2010                         fprintf(fp, "# cpu pmu capabilities: not available\n");
2011                 else
2012                         fprintf(fp, "# %s pmu capabilities: not available\n", pmu_name);
2013                 return;
2014         }
2015
2016         if (!pmu_name)
2017                 scnprintf(buf, sizeof(buf), "# cpu pmu capabilities: ");
2018         else
2019                 scnprintf(buf, sizeof(buf), "# %s pmu capabilities: ", pmu_name);
2020
2021         delimiter = buf;
2022
2023         str = cpu_pmu_caps;
2024         while (nr_caps--) {
2025                 fprintf(fp, "%s%s", delimiter, str);
2026                 delimiter = ", ";
2027                 str += strlen(str) + 1;
2028         }
2029
2030         fprintf(fp, "\n");
2031 }
2032
2033 static void print_cpu_pmu_caps(struct feat_fd *ff, FILE *fp)
2034 {
2035         print_per_cpu_pmu_caps(fp, ff->ph->env.nr_cpu_pmu_caps,
2036                                ff->ph->env.cpu_pmu_caps, NULL);
2037 }
2038
2039 static void print_hybrid_cpu_pmu_caps(struct feat_fd *ff, FILE *fp)
2040 {
2041         struct hybrid_cpc_node *n;
2042
2043         for (int i = 0; i < ff->ph->env.nr_hybrid_cpc_nodes; i++) {
2044                 n = &ff->ph->env.hybrid_cpc_nodes[i];
2045                 print_per_cpu_pmu_caps(fp, n->nr_cpu_pmu_caps,
2046                                        n->cpu_pmu_caps,
2047                                        n->pmu_name);
2048         }
2049 }
2050
2051 static void print_pmu_mappings(struct feat_fd *ff, FILE *fp)
2052 {
2053         const char *delimiter = "# pmu mappings: ";
2054         char *str, *tmp;
2055         u32 pmu_num;
2056         u32 type;
2057
2058         pmu_num = ff->ph->env.nr_pmu_mappings;
2059         if (!pmu_num) {
2060                 fprintf(fp, "# pmu mappings: not available\n");
2061                 return;
2062         }
2063
2064         str = ff->ph->env.pmu_mappings;
2065
2066         while (pmu_num) {
2067                 type = strtoul(str, &tmp, 0);
2068                 if (*tmp != ':')
2069                         goto error;
2070
2071                 str = tmp + 1;
2072                 fprintf(fp, "%s%s = %" PRIu32, delimiter, str, type);
2073
2074                 delimiter = ", ";
2075                 str += strlen(str) + 1;
2076                 pmu_num--;
2077         }
2078
2079         fprintf(fp, "\n");
2080
2081         if (!pmu_num)
2082                 return;
2083 error:
2084         fprintf(fp, "# pmu mappings: unable to read\n");
2085 }
2086
2087 static void print_group_desc(struct feat_fd *ff, FILE *fp)
2088 {
2089         struct perf_session *session;
2090         struct evsel *evsel;
2091         u32 nr = 0;
2092
2093         session = container_of(ff->ph, struct perf_session, header);
2094
2095         evlist__for_each_entry(session->evlist, evsel) {
2096                 if (evsel__is_group_leader(evsel) && evsel->core.nr_members > 1) {
2097                         fprintf(fp, "# group: %s{%s", evsel->group_name ?: "", evsel__name(evsel));
2098
2099                         nr = evsel->core.nr_members - 1;
2100                 } else if (nr) {
2101                         fprintf(fp, ",%s", evsel__name(evsel));
2102
2103                         if (--nr == 0)
2104                                 fprintf(fp, "}\n");
2105                 }
2106         }
2107 }
2108
2109 static void print_sample_time(struct feat_fd *ff, FILE *fp)
2110 {
2111         struct perf_session *session;
2112         char time_buf[32];
2113         double d;
2114
2115         session = container_of(ff->ph, struct perf_session, header);
2116
2117         timestamp__scnprintf_usec(session->evlist->first_sample_time,
2118                                   time_buf, sizeof(time_buf));
2119         fprintf(fp, "# time of first sample : %s\n", time_buf);
2120
2121         timestamp__scnprintf_usec(session->evlist->last_sample_time,
2122                                   time_buf, sizeof(time_buf));
2123         fprintf(fp, "# time of last sample : %s\n", time_buf);
2124
2125         d = (double)(session->evlist->last_sample_time -
2126                 session->evlist->first_sample_time) / NSEC_PER_MSEC;
2127
2128         fprintf(fp, "# sample duration : %10.3f ms\n", d);
2129 }
2130
2131 static void memory_node__fprintf(struct memory_node *n,
2132                                  unsigned long long bsize, FILE *fp)
2133 {
2134         char buf_map[100], buf_size[50];
2135         unsigned long long size;
2136
2137         size = bsize * bitmap_weight(n->set, n->size);
2138         unit_number__scnprintf(buf_size, 50, size);
2139
2140         bitmap_scnprintf(n->set, n->size, buf_map, 100);
2141         fprintf(fp, "#  %3" PRIu64 " [%s]: %s\n", n->node, buf_size, buf_map);
2142 }
2143
2144 static void print_mem_topology(struct feat_fd *ff, FILE *fp)
2145 {
2146         struct memory_node *nodes;
2147         int i, nr;
2148
2149         nodes = ff->ph->env.memory_nodes;
2150         nr    = ff->ph->env.nr_memory_nodes;
2151
2152         fprintf(fp, "# memory nodes (nr %d, block size 0x%llx):\n",
2153                 nr, ff->ph->env.memory_bsize);
2154
2155         for (i = 0; i < nr; i++) {
2156                 memory_node__fprintf(&nodes[i], ff->ph->env.memory_bsize, fp);
2157         }
2158 }
2159
2160 static int __event_process_build_id(struct perf_record_header_build_id *bev,
2161                                     char *filename,
2162                                     struct perf_session *session)
2163 {
2164         int err = -1;
2165         struct machine *machine;
2166         u16 cpumode;
2167         struct dso *dso;
2168         enum dso_space_type dso_space;
2169
2170         machine = perf_session__findnew_machine(session, bev->pid);
2171         if (!machine)
2172                 goto out;
2173
2174         cpumode = bev->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
2175
2176         switch (cpumode) {
2177         case PERF_RECORD_MISC_KERNEL:
2178                 dso_space = DSO_SPACE__KERNEL;
2179                 break;
2180         case PERF_RECORD_MISC_GUEST_KERNEL:
2181                 dso_space = DSO_SPACE__KERNEL_GUEST;
2182                 break;
2183         case PERF_RECORD_MISC_USER:
2184         case PERF_RECORD_MISC_GUEST_USER:
2185                 dso_space = DSO_SPACE__USER;
2186                 break;
2187         default:
2188                 goto out;
2189         }
2190
2191         dso = machine__findnew_dso(machine, filename);
2192         if (dso != NULL) {
2193                 char sbuild_id[SBUILD_ID_SIZE];
2194                 struct build_id bid;
2195                 size_t size = BUILD_ID_SIZE;
2196
2197                 if (bev->header.misc & PERF_RECORD_MISC_BUILD_ID_SIZE)
2198                         size = bev->size;
2199
2200                 build_id__init(&bid, bev->data, size);
2201                 dso__set_build_id(dso, &bid);
2202
2203                 if (dso_space != DSO_SPACE__USER) {
2204                         struct kmod_path m = { .name = NULL, };
2205
2206                         if (!kmod_path__parse_name(&m, filename) && m.kmod)
2207                                 dso__set_module_info(dso, &m, machine);
2208
2209                         dso->kernel = dso_space;
2210                         free(m.name);
2211                 }
2212
2213                 build_id__sprintf(&dso->bid, sbuild_id);
2214                 pr_debug("build id event received for %s: %s [%zu]\n",
2215                          dso->long_name, sbuild_id, size);
2216                 dso__put(dso);
2217         }
2218
2219         err = 0;
2220 out:
2221         return err;
2222 }
2223
2224 static int perf_header__read_build_ids_abi_quirk(struct perf_header *header,
2225                                                  int input, u64 offset, u64 size)
2226 {
2227         struct perf_session *session = container_of(header, struct perf_session, header);
2228         struct {
2229                 struct perf_event_header   header;
2230                 u8                         build_id[PERF_ALIGN(BUILD_ID_SIZE, sizeof(u64))];
2231                 char                       filename[0];
2232         } old_bev;
2233         struct perf_record_header_build_id bev;
2234         char filename[PATH_MAX];
2235         u64 limit = offset + size;
2236
2237         while (offset < limit) {
2238                 ssize_t len;
2239
2240                 if (readn(input, &old_bev, sizeof(old_bev)) != sizeof(old_bev))
2241                         return -1;
2242
2243                 if (header->needs_swap)
2244                         perf_event_header__bswap(&old_bev.header);
2245
2246                 len = old_bev.header.size - sizeof(old_bev);
2247                 if (readn(input, filename, len) != len)
2248                         return -1;
2249
2250                 bev.header = old_bev.header;
2251
2252                 /*
2253                  * As the pid is the missing value, we need to fill
2254                  * it properly. The header.misc value give us nice hint.
2255                  */
2256                 bev.pid = HOST_KERNEL_ID;
2257                 if (bev.header.misc == PERF_RECORD_MISC_GUEST_USER ||
2258                     bev.header.misc == PERF_RECORD_MISC_GUEST_KERNEL)
2259                         bev.pid = DEFAULT_GUEST_KERNEL_ID;
2260
2261                 memcpy(bev.build_id, old_bev.build_id, sizeof(bev.build_id));
2262                 __event_process_build_id(&bev, filename, session);
2263
2264                 offset += bev.header.size;
2265         }
2266
2267         return 0;
2268 }
2269
2270 static int perf_header__read_build_ids(struct perf_header *header,
2271                                        int input, u64 offset, u64 size)
2272 {
2273         struct perf_session *session = container_of(header, struct perf_session, header);
2274         struct perf_record_header_build_id bev;
2275         char filename[PATH_MAX];
2276         u64 limit = offset + size, orig_offset = offset;
2277         int err = -1;
2278
2279         while (offset < limit) {
2280                 ssize_t len;
2281
2282                 if (readn(input, &bev, sizeof(bev)) != sizeof(bev))
2283                         goto out;
2284
2285                 if (header->needs_swap)
2286                         perf_event_header__bswap(&bev.header);
2287
2288                 len = bev.header.size - sizeof(bev);
2289                 if (readn(input, filename, len) != len)
2290                         goto out;
2291                 /*
2292                  * The a1645ce1 changeset:
2293                  *
2294                  * "perf: 'perf kvm' tool for monitoring guest performance from host"
2295                  *
2296                  * Added a field to struct perf_record_header_build_id that broke the file
2297                  * format.
2298                  *
2299                  * Since the kernel build-id is the first entry, process the
2300                  * table using the old format if the well known
2301                  * '[kernel.kallsyms]' string for the kernel build-id has the
2302                  * first 4 characters chopped off (where the pid_t sits).
2303                  */
2304                 if (memcmp(filename, "nel.kallsyms]", 13) == 0) {
2305                         if (lseek(input, orig_offset, SEEK_SET) == (off_t)-1)
2306                                 return -1;
2307                         return perf_header__read_build_ids_abi_quirk(header, input, offset, size);
2308                 }
2309
2310                 __event_process_build_id(&bev, filename, session);
2311
2312                 offset += bev.header.size;
2313         }
2314         err = 0;
2315 out:
2316         return err;
2317 }
2318
2319 /* Macro for features that simply need to read and store a string. */
2320 #define FEAT_PROCESS_STR_FUN(__feat, __feat_env) \
2321 static int process_##__feat(struct feat_fd *ff, void *data __maybe_unused) \
2322 {\
2323         ff->ph->env.__feat_env = do_read_string(ff); \
2324         return ff->ph->env.__feat_env ? 0 : -ENOMEM; \
2325 }
2326
2327 FEAT_PROCESS_STR_FUN(hostname, hostname);
2328 FEAT_PROCESS_STR_FUN(osrelease, os_release);
2329 FEAT_PROCESS_STR_FUN(version, version);
2330 FEAT_PROCESS_STR_FUN(arch, arch);
2331 FEAT_PROCESS_STR_FUN(cpudesc, cpu_desc);
2332 FEAT_PROCESS_STR_FUN(cpuid, cpuid);
2333
2334 static int process_tracing_data(struct feat_fd *ff, void *data)
2335 {
2336         ssize_t ret = trace_report(ff->fd, data, false);
2337
2338         return ret < 0 ? -1 : 0;
2339 }
2340
2341 static int process_build_id(struct feat_fd *ff, void *data __maybe_unused)
2342 {
2343         if (perf_header__read_build_ids(ff->ph, ff->fd, ff->offset, ff->size))
2344                 pr_debug("Failed to read buildids, continuing...\n");
2345         return 0;
2346 }
2347
2348 static int process_nrcpus(struct feat_fd *ff, void *data __maybe_unused)
2349 {
2350         int ret;
2351         u32 nr_cpus_avail, nr_cpus_online;
2352
2353         ret = do_read_u32(ff, &nr_cpus_avail);
2354         if (ret)
2355                 return ret;
2356
2357         ret = do_read_u32(ff, &nr_cpus_online);
2358         if (ret)
2359                 return ret;
2360         ff->ph->env.nr_cpus_avail = (int)nr_cpus_avail;
2361         ff->ph->env.nr_cpus_online = (int)nr_cpus_online;
2362         return 0;
2363 }
2364
2365 static int process_total_mem(struct feat_fd *ff, void *data __maybe_unused)
2366 {
2367         u64 total_mem;
2368         int ret;
2369
2370         ret = do_read_u64(ff, &total_mem);
2371         if (ret)
2372                 return -1;
2373         ff->ph->env.total_mem = (unsigned long long)total_mem;
2374         return 0;
2375 }
2376
2377 static struct evsel *evlist__find_by_index(struct evlist *evlist, int idx)
2378 {
2379         struct evsel *evsel;
2380
2381         evlist__for_each_entry(evlist, evsel) {
2382                 if (evsel->idx == idx)
2383                         return evsel;
2384         }
2385
2386         return NULL;
2387 }
2388
2389 static void evlist__set_event_name(struct evlist *evlist, struct evsel *event)
2390 {
2391         struct evsel *evsel;
2392
2393         if (!event->name)
2394                 return;
2395
2396         evsel = evlist__find_by_index(evlist, event->idx);
2397         if (!evsel)
2398                 return;
2399
2400         if (evsel->name)
2401                 return;
2402
2403         evsel->name = strdup(event->name);
2404 }
2405
2406 static int
2407 process_event_desc(struct feat_fd *ff, void *data __maybe_unused)
2408 {
2409         struct perf_session *session;
2410         struct evsel *evsel, *events = read_event_desc(ff);
2411
2412         if (!events)
2413                 return 0;
2414
2415         session = container_of(ff->ph, struct perf_session, header);
2416
2417         if (session->data->is_pipe) {
2418                 /* Save events for reading later by print_event_desc,
2419                  * since they can't be read again in pipe mode. */
2420                 ff->events = events;
2421         }
2422
2423         for (evsel = events; evsel->core.attr.size; evsel++)
2424                 evlist__set_event_name(session->evlist, evsel);
2425
2426         if (!session->data->is_pipe)
2427                 free_event_desc(events);
2428
2429         return 0;
2430 }
2431
2432 static int process_cmdline(struct feat_fd *ff, void *data __maybe_unused)
2433 {
2434         char *str, *cmdline = NULL, **argv = NULL;
2435         u32 nr, i, len = 0;
2436
2437         if (do_read_u32(ff, &nr))
2438                 return -1;
2439
2440         ff->ph->env.nr_cmdline = nr;
2441
2442         cmdline = zalloc(ff->size + nr + 1);
2443         if (!cmdline)
2444                 return -1;
2445
2446         argv = zalloc(sizeof(char *) * (nr + 1));
2447         if (!argv)
2448                 goto error;
2449
2450         for (i = 0; i < nr; i++) {
2451                 str = do_read_string(ff);
2452                 if (!str)
2453                         goto error;
2454
2455                 argv[i] = cmdline + len;
2456                 memcpy(argv[i], str, strlen(str) + 1);
2457                 len += strlen(str) + 1;
2458                 free(str);
2459         }
2460         ff->ph->env.cmdline = cmdline;
2461         ff->ph->env.cmdline_argv = (const char **) argv;
2462         return 0;
2463
2464 error:
2465         free(argv);
2466         free(cmdline);
2467         return -1;
2468 }
2469
2470 static int process_cpu_topology(struct feat_fd *ff, void *data __maybe_unused)
2471 {
2472         u32 nr, i;
2473         char *str;
2474         struct strbuf sb;
2475         int cpu_nr = ff->ph->env.nr_cpus_avail;
2476         u64 size = 0;
2477         struct perf_header *ph = ff->ph;
2478         bool do_core_id_test = true;
2479
2480         ph->env.cpu = calloc(cpu_nr, sizeof(*ph->env.cpu));
2481         if (!ph->env.cpu)
2482                 return -1;
2483
2484         if (do_read_u32(ff, &nr))
2485                 goto free_cpu;
2486
2487         ph->env.nr_sibling_cores = nr;
2488         size += sizeof(u32);
2489         if (strbuf_init(&sb, 128) < 0)
2490                 goto free_cpu;
2491
2492         for (i = 0; i < nr; i++) {
2493                 str = do_read_string(ff);
2494                 if (!str)
2495                         goto error;
2496
2497                 /* include a NULL character at the end */
2498                 if (strbuf_add(&sb, str, strlen(str) + 1) < 0)
2499                         goto error;
2500                 size += string_size(str);
2501                 free(str);
2502         }
2503         ph->env.sibling_cores = strbuf_detach(&sb, NULL);
2504
2505         if (do_read_u32(ff, &nr))
2506                 return -1;
2507
2508         ph->env.nr_sibling_threads = nr;
2509         size += sizeof(u32);
2510
2511         for (i = 0; i < nr; i++) {
2512                 str = do_read_string(ff);
2513                 if (!str)
2514                         goto error;
2515
2516                 /* include a NULL character at the end */
2517                 if (strbuf_add(&sb, str, strlen(str) + 1) < 0)
2518                         goto error;
2519                 size += string_size(str);
2520                 free(str);
2521         }
2522         ph->env.sibling_threads = strbuf_detach(&sb, NULL);
2523
2524         /*
2525          * The header may be from old perf,
2526          * which doesn't include core id and socket id information.
2527          */
2528         if (ff->size <= size) {
2529                 zfree(&ph->env.cpu);
2530                 return 0;
2531         }
2532
2533         /* On s390 the socket_id number is not related to the numbers of cpus.
2534          * The socket_id number might be higher than the numbers of cpus.
2535          * This depends on the configuration.
2536          * AArch64 is the same.
2537          */
2538         if (ph->env.arch && (!strncmp(ph->env.arch, "s390", 4)
2539                           || !strncmp(ph->env.arch, "aarch64", 7)))
2540                 do_core_id_test = false;
2541
2542         for (i = 0; i < (u32)cpu_nr; i++) {
2543                 if (do_read_u32(ff, &nr))
2544                         goto free_cpu;
2545
2546                 ph->env.cpu[i].core_id = nr;
2547                 size += sizeof(u32);
2548
2549                 if (do_read_u32(ff, &nr))
2550                         goto free_cpu;
2551
2552                 if (do_core_id_test && nr != (u32)-1 && nr > (u32)cpu_nr) {
2553                         pr_debug("socket_id number is too big."
2554                                  "You may need to upgrade the perf tool.\n");
2555                         goto free_cpu;
2556                 }
2557
2558                 ph->env.cpu[i].socket_id = nr;
2559                 size += sizeof(u32);
2560         }
2561
2562         /*
2563          * The header may be from old perf,
2564          * which doesn't include die information.
2565          */
2566         if (ff->size <= size)
2567                 return 0;
2568
2569         if (do_read_u32(ff, &nr))
2570                 return -1;
2571
2572         ph->env.nr_sibling_dies = nr;
2573         size += sizeof(u32);
2574
2575         for (i = 0; i < nr; i++) {
2576                 str = do_read_string(ff);
2577                 if (!str)
2578                         goto error;
2579
2580                 /* include a NULL character at the end */
2581                 if (strbuf_add(&sb, str, strlen(str) + 1) < 0)
2582                         goto error;
2583                 size += string_size(str);
2584                 free(str);
2585         }
2586         ph->env.sibling_dies = strbuf_detach(&sb, NULL);
2587
2588         for (i = 0; i < (u32)cpu_nr; i++) {
2589                 if (do_read_u32(ff, &nr))
2590                         goto free_cpu;
2591
2592                 ph->env.cpu[i].die_id = nr;
2593         }
2594
2595         return 0;
2596
2597 error:
2598         strbuf_release(&sb);
2599 free_cpu:
2600         zfree(&ph->env.cpu);
2601         return -1;
2602 }
2603
2604 static int process_numa_topology(struct feat_fd *ff, void *data __maybe_unused)
2605 {
2606         struct numa_node *nodes, *n;
2607         u32 nr, i;
2608         char *str;
2609
2610         /* nr nodes */
2611         if (do_read_u32(ff, &nr))
2612                 return -1;
2613
2614         nodes = zalloc(sizeof(*nodes) * nr);
2615         if (!nodes)
2616                 return -ENOMEM;
2617
2618         for (i = 0; i < nr; i++) {
2619                 n = &nodes[i];
2620
2621                 /* node number */
2622                 if (do_read_u32(ff, &n->node))
2623                         goto error;
2624
2625                 if (do_read_u64(ff, &n->mem_total))
2626                         goto error;
2627
2628                 if (do_read_u64(ff, &n->mem_free))
2629                         goto error;
2630
2631                 str = do_read_string(ff);
2632                 if (!str)
2633                         goto error;
2634
2635                 n->map = perf_cpu_map__new(str);
2636                 if (!n->map)
2637                         goto error;
2638
2639                 free(str);
2640         }
2641         ff->ph->env.nr_numa_nodes = nr;
2642         ff->ph->env.numa_nodes = nodes;
2643         return 0;
2644
2645 error:
2646         free(nodes);
2647         return -1;
2648 }
2649
2650 static int process_pmu_mappings(struct feat_fd *ff, void *data __maybe_unused)
2651 {
2652         char *name;
2653         u32 pmu_num;
2654         u32 type;
2655         struct strbuf sb;
2656
2657         if (do_read_u32(ff, &pmu_num))
2658                 return -1;
2659
2660         if (!pmu_num) {
2661                 pr_debug("pmu mappings not available\n");
2662                 return 0;
2663         }
2664
2665         ff->ph->env.nr_pmu_mappings = pmu_num;
2666         if (strbuf_init(&sb, 128) < 0)
2667                 return -1;
2668
2669         while (pmu_num) {
2670                 if (do_read_u32(ff, &type))
2671                         goto error;
2672
2673                 name = do_read_string(ff);
2674                 if (!name)
2675                         goto error;
2676
2677                 if (strbuf_addf(&sb, "%u:%s", type, name) < 0)
2678                         goto error;
2679                 /* include a NULL character at the end */
2680                 if (strbuf_add(&sb, "", 1) < 0)
2681                         goto error;
2682
2683                 if (!strcmp(name, "msr"))
2684                         ff->ph->env.msr_pmu_type = type;
2685
2686                 free(name);
2687                 pmu_num--;
2688         }
2689         ff->ph->env.pmu_mappings = strbuf_detach(&sb, NULL);
2690         return 0;
2691
2692 error:
2693         strbuf_release(&sb);
2694         return -1;
2695 }
2696
2697 static int process_group_desc(struct feat_fd *ff, void *data __maybe_unused)
2698 {
2699         size_t ret = -1;
2700         u32 i, nr, nr_groups;
2701         struct perf_session *session;
2702         struct evsel *evsel, *leader = NULL;
2703         struct group_desc {
2704                 char *name;
2705                 u32 leader_idx;
2706                 u32 nr_members;
2707         } *desc;
2708
2709         if (do_read_u32(ff, &nr_groups))
2710                 return -1;
2711
2712         ff->ph->env.nr_groups = nr_groups;
2713         if (!nr_groups) {
2714                 pr_debug("group desc not available\n");
2715                 return 0;
2716         }
2717
2718         desc = calloc(nr_groups, sizeof(*desc));
2719         if (!desc)
2720                 return -1;
2721
2722         for (i = 0; i < nr_groups; i++) {
2723                 desc[i].name = do_read_string(ff);
2724                 if (!desc[i].name)
2725                         goto out_free;
2726
2727                 if (do_read_u32(ff, &desc[i].leader_idx))
2728                         goto out_free;
2729
2730                 if (do_read_u32(ff, &desc[i].nr_members))
2731                         goto out_free;
2732         }
2733
2734         /*
2735          * Rebuild group relationship based on the group_desc
2736          */
2737         session = container_of(ff->ph, struct perf_session, header);
2738         session->evlist->nr_groups = nr_groups;
2739
2740         i = nr = 0;
2741         evlist__for_each_entry(session->evlist, evsel) {
2742                 if (evsel->idx == (int) desc[i].leader_idx) {
2743                         evsel->leader = evsel;
2744                         /* {anon_group} is a dummy name */
2745                         if (strcmp(desc[i].name, "{anon_group}")) {
2746                                 evsel->group_name = desc[i].name;
2747                                 desc[i].name = NULL;
2748                         }
2749                         evsel->core.nr_members = desc[i].nr_members;
2750
2751                         if (i >= nr_groups || nr > 0) {
2752                                 pr_debug("invalid group desc\n");
2753                                 goto out_free;
2754                         }
2755
2756                         leader = evsel;
2757                         nr = evsel->core.nr_members - 1;
2758                         i++;
2759                 } else if (nr) {
2760                         /* This is a group member */
2761                         evsel->leader = leader;
2762
2763                         nr--;
2764                 }
2765         }
2766
2767         if (i != nr_groups || nr != 0) {
2768                 pr_debug("invalid group desc\n");
2769                 goto out_free;
2770         }
2771
2772         ret = 0;
2773 out_free:
2774         for (i = 0; i < nr_groups; i++)
2775                 zfree(&desc[i].name);
2776         free(desc);
2777
2778         return ret;
2779 }
2780
2781 static int process_auxtrace(struct feat_fd *ff, void *data __maybe_unused)
2782 {
2783         struct perf_session *session;
2784         int err;
2785
2786         session = container_of(ff->ph, struct perf_session, header);
2787
2788         err = auxtrace_index__process(ff->fd, ff->size, session,
2789                                       ff->ph->needs_swap);
2790         if (err < 0)
2791                 pr_err("Failed to process auxtrace index\n");
2792         return err;
2793 }
2794
2795 static int process_cache(struct feat_fd *ff, void *data __maybe_unused)
2796 {
2797         struct cpu_cache_level *caches;
2798         u32 cnt, i, version;
2799
2800         if (do_read_u32(ff, &version))
2801                 return -1;
2802
2803         if (version != 1)
2804                 return -1;
2805
2806         if (do_read_u32(ff, &cnt))
2807                 return -1;
2808
2809         caches = zalloc(sizeof(*caches) * cnt);
2810         if (!caches)
2811                 return -1;
2812
2813         for (i = 0; i < cnt; i++) {
2814                 struct cpu_cache_level c;
2815
2816                 #define _R(v)                                           \
2817                         if (do_read_u32(ff, &c.v))\
2818                                 goto out_free_caches;                   \
2819
2820                 _R(level)
2821                 _R(line_size)
2822                 _R(sets)
2823                 _R(ways)
2824                 #undef _R
2825
2826                 #define _R(v)                                   \
2827                         c.v = do_read_string(ff);               \
2828                         if (!c.v)                               \
2829                                 goto out_free_caches;
2830
2831                 _R(type)
2832                 _R(size)
2833                 _R(map)
2834                 #undef _R
2835
2836                 caches[i] = c;
2837         }
2838
2839         ff->ph->env.caches = caches;
2840         ff->ph->env.caches_cnt = cnt;
2841         return 0;
2842 out_free_caches:
2843         free(caches);
2844         return -1;
2845 }
2846
2847 static int process_sample_time(struct feat_fd *ff, void *data __maybe_unused)
2848 {
2849         struct perf_session *session;
2850         u64 first_sample_time, last_sample_time;
2851         int ret;
2852
2853         session = container_of(ff->ph, struct perf_session, header);
2854
2855         ret = do_read_u64(ff, &first_sample_time);
2856         if (ret)
2857                 return -1;
2858
2859         ret = do_read_u64(ff, &last_sample_time);
2860         if (ret)
2861                 return -1;
2862
2863         session->evlist->first_sample_time = first_sample_time;
2864         session->evlist->last_sample_time = last_sample_time;
2865         return 0;
2866 }
2867
2868 static int process_mem_topology(struct feat_fd *ff,
2869                                 void *data __maybe_unused)
2870 {
2871         struct memory_node *nodes;
2872         u64 version, i, nr, bsize;
2873         int ret = -1;
2874
2875         if (do_read_u64(ff, &version))
2876                 return -1;
2877
2878         if (version != 1)
2879                 return -1;
2880
2881         if (do_read_u64(ff, &bsize))
2882                 return -1;
2883
2884         if (do_read_u64(ff, &nr))
2885                 return -1;
2886
2887         nodes = zalloc(sizeof(*nodes) * nr);
2888         if (!nodes)
2889                 return -1;
2890
2891         for (i = 0; i < nr; i++) {
2892                 struct memory_node n;
2893
2894                 #define _R(v)                           \
2895                         if (do_read_u64(ff, &n.v))      \
2896                                 goto out;               \
2897
2898                 _R(node)
2899                 _R(size)
2900
2901                 #undef _R
2902
2903                 if (do_read_bitmap(ff, &n.set, &n.size))
2904                         goto out;
2905
2906                 nodes[i] = n;
2907         }
2908
2909         ff->ph->env.memory_bsize    = bsize;
2910         ff->ph->env.memory_nodes    = nodes;
2911         ff->ph->env.nr_memory_nodes = nr;
2912         ret = 0;
2913
2914 out:
2915         if (ret)
2916                 free(nodes);
2917         return ret;
2918 }
2919
2920 static int process_clockid(struct feat_fd *ff,
2921                            void *data __maybe_unused)
2922 {
2923         if (do_read_u64(ff, &ff->ph->env.clock.clockid_res_ns))
2924                 return -1;
2925
2926         return 0;
2927 }
2928
2929 static int process_clock_data(struct feat_fd *ff,
2930                               void *_data __maybe_unused)
2931 {
2932         u32 data32;
2933         u64 data64;
2934
2935         /* version */
2936         if (do_read_u32(ff, &data32))
2937                 return -1;
2938
2939         if (data32 != 1)
2940                 return -1;
2941
2942         /* clockid */
2943         if (do_read_u32(ff, &data32))
2944                 return -1;
2945
2946         ff->ph->env.clock.clockid = data32;
2947
2948         /* TOD ref time */
2949         if (do_read_u64(ff, &data64))
2950                 return -1;
2951
2952         ff->ph->env.clock.tod_ns = data64;
2953
2954         /* clockid ref time */
2955         if (do_read_u64(ff, &data64))
2956                 return -1;
2957
2958         ff->ph->env.clock.clockid_ns = data64;
2959         ff->ph->env.clock.enabled = true;
2960         return 0;
2961 }
2962
2963 static int process_hybrid_topology(struct feat_fd *ff,
2964                                    void *data __maybe_unused)
2965 {
2966         struct hybrid_node *nodes, *n;
2967         u32 nr, i;
2968
2969         /* nr nodes */
2970         if (do_read_u32(ff, &nr))
2971                 return -1;
2972
2973         nodes = zalloc(sizeof(*nodes) * nr);
2974         if (!nodes)
2975                 return -ENOMEM;
2976
2977         for (i = 0; i < nr; i++) {
2978                 n = &nodes[i];
2979
2980                 n->pmu_name = do_read_string(ff);
2981                 if (!n->pmu_name)
2982                         goto error;
2983
2984                 n->cpus = do_read_string(ff);
2985                 if (!n->cpus)
2986                         goto error;
2987         }
2988
2989         ff->ph->env.nr_hybrid_nodes = nr;
2990         ff->ph->env.hybrid_nodes = nodes;
2991         return 0;
2992
2993 error:
2994         for (i = 0; i < nr; i++) {
2995                 free(nodes[i].pmu_name);
2996                 free(nodes[i].cpus);
2997         }
2998
2999         free(nodes);
3000         return -1;
3001 }
3002
3003 static int process_dir_format(struct feat_fd *ff,
3004                               void *_data __maybe_unused)
3005 {
3006         struct perf_session *session;
3007         struct perf_data *data;
3008
3009         session = container_of(ff->ph, struct perf_session, header);
3010         data = session->data;
3011
3012         if (WARN_ON(!perf_data__is_dir(data)))
3013                 return -1;
3014
3015         return do_read_u64(ff, &data->dir.version);
3016 }
3017
3018 #ifdef HAVE_LIBBPF_SUPPORT
3019 static int process_bpf_prog_info(struct feat_fd *ff, void *data __maybe_unused)
3020 {
3021         struct bpf_prog_info_linear *info_linear;
3022         struct bpf_prog_info_node *info_node;
3023         struct perf_env *env = &ff->ph->env;
3024         u32 count, i;
3025         int err = -1;
3026
3027         if (ff->ph->needs_swap) {
3028                 pr_warning("interpreting bpf_prog_info from systems with endianness is not yet supported\n");
3029                 return 0;
3030         }
3031
3032         if (do_read_u32(ff, &count))
3033                 return -1;
3034
3035         down_write(&env->bpf_progs.lock);
3036
3037         for (i = 0; i < count; ++i) {
3038                 u32 info_len, data_len;
3039
3040                 info_linear = NULL;
3041                 info_node = NULL;
3042                 if (do_read_u32(ff, &info_len))
3043                         goto out;
3044                 if (do_read_u32(ff, &data_len))
3045                         goto out;
3046
3047                 if (info_len > sizeof(struct bpf_prog_info)) {
3048                         pr_warning("detected invalid bpf_prog_info\n");
3049                         goto out;
3050                 }
3051
3052                 info_linear = malloc(sizeof(struct bpf_prog_info_linear) +
3053                                      data_len);
3054                 if (!info_linear)
3055                         goto out;
3056                 info_linear->info_len = sizeof(struct bpf_prog_info);
3057                 info_linear->data_len = data_len;
3058                 if (do_read_u64(ff, (u64 *)(&info_linear->arrays)))
3059                         goto out;
3060                 if (__do_read(ff, &info_linear->info, info_len))
3061                         goto out;
3062                 if (info_len < sizeof(struct bpf_prog_info))
3063                         memset(((void *)(&info_linear->info)) + info_len, 0,
3064                                sizeof(struct bpf_prog_info) - info_len);
3065
3066                 if (__do_read(ff, info_linear->data, data_len))
3067                         goto out;
3068
3069                 info_node = malloc(sizeof(struct bpf_prog_info_node));
3070                 if (!info_node)
3071                         goto out;
3072
3073                 /* after reading from file, translate offset to address */
3074                 bpf_program__bpil_offs_to_addr(info_linear);
3075                 info_node->info_linear = info_linear;
3076                 perf_env__insert_bpf_prog_info(env, info_node);
3077         }
3078
3079         up_write(&env->bpf_progs.lock);
3080         return 0;
3081 out:
3082         free(info_linear);
3083         free(info_node);
3084         up_write(&env->bpf_progs.lock);
3085         return err;
3086 }
3087
3088 static int process_bpf_btf(struct feat_fd *ff, void *data __maybe_unused)
3089 {
3090         struct perf_env *env = &ff->ph->env;
3091         struct btf_node *node = NULL;
3092         u32 count, i;
3093         int err = -1;
3094
3095         if (ff->ph->needs_swap) {
3096                 pr_warning("interpreting btf from systems with endianness is not yet supported\n");
3097                 return 0;
3098         }
3099
3100         if (do_read_u32(ff, &count))
3101                 return -1;
3102
3103         down_write(&env->bpf_progs.lock);
3104
3105         for (i = 0; i < count; ++i) {
3106                 u32 id, data_size;
3107
3108                 if (do_read_u32(ff, &id))
3109                         goto out;
3110                 if (do_read_u32(ff, &data_size))
3111                         goto out;
3112
3113                 node = malloc(sizeof(struct btf_node) + data_size);
3114                 if (!node)
3115                         goto out;
3116
3117                 node->id = id;
3118                 node->data_size = data_size;
3119
3120                 if (__do_read(ff, node->data, data_size))
3121                         goto out;
3122
3123                 perf_env__insert_btf(env, node);
3124                 node = NULL;
3125         }
3126
3127         err = 0;
3128 out:
3129         up_write(&env->bpf_progs.lock);
3130         free(node);
3131         return err;
3132 }
3133 #endif // HAVE_LIBBPF_SUPPORT
3134
3135 static int process_compressed(struct feat_fd *ff,
3136                               void *data __maybe_unused)
3137 {
3138         if (do_read_u32(ff, &(ff->ph->env.comp_ver)))
3139                 return -1;
3140
3141         if (do_read_u32(ff, &(ff->ph->env.comp_type)))
3142                 return -1;
3143
3144         if (do_read_u32(ff, &(ff->ph->env.comp_level)))
3145                 return -1;
3146
3147         if (do_read_u32(ff, &(ff->ph->env.comp_ratio)))
3148                 return -1;
3149
3150         if (do_read_u32(ff, &(ff->ph->env.comp_mmap_len)))
3151                 return -1;
3152
3153         return 0;
3154 }
3155
3156 static int process_per_cpu_pmu_caps(struct feat_fd *ff, int *nr_cpu_pmu_caps,
3157                                     char **cpu_pmu_caps,
3158                                     unsigned int *max_branches)
3159 {
3160         char *name, *value;
3161         struct strbuf sb;
3162         u32 nr_caps;
3163
3164         if (do_read_u32(ff, &nr_caps))
3165                 return -1;
3166
3167         if (!nr_caps) {
3168                 pr_debug("cpu pmu capabilities not available\n");
3169                 return 0;
3170         }
3171
3172         *nr_cpu_pmu_caps = nr_caps;
3173
3174         if (strbuf_init(&sb, 128) < 0)
3175                 return -1;
3176
3177         while (nr_caps--) {
3178                 name = do_read_string(ff);
3179                 if (!name)
3180                         goto error;
3181
3182                 value = do_read_string(ff);
3183                 if (!value)
3184                         goto free_name;
3185
3186                 if (strbuf_addf(&sb, "%s=%s", name, value) < 0)
3187                         goto free_value;
3188
3189                 /* include a NULL character at the end */
3190                 if (strbuf_add(&sb, "", 1) < 0)
3191                         goto free_value;
3192
3193                 if (!strcmp(name, "branches"))
3194                         *max_branches = atoi(value);
3195
3196                 free(value);
3197                 free(name);
3198         }
3199         *cpu_pmu_caps = strbuf_detach(&sb, NULL);
3200         return 0;
3201
3202 free_value:
3203         free(value);
3204 free_name:
3205         free(name);
3206 error:
3207         strbuf_release(&sb);
3208         return -1;
3209 }
3210
3211 static int process_cpu_pmu_caps(struct feat_fd *ff,
3212                                 void *data __maybe_unused)
3213 {
3214         return process_per_cpu_pmu_caps(ff, &ff->ph->env.nr_cpu_pmu_caps,
3215                                         &ff->ph->env.cpu_pmu_caps,
3216                                         &ff->ph->env.max_branches);
3217 }
3218
3219 static int process_hybrid_cpu_pmu_caps(struct feat_fd *ff,
3220                                        void *data __maybe_unused)
3221 {
3222         struct hybrid_cpc_node *nodes;
3223         u32 nr_pmu, i;
3224         int ret;
3225
3226         if (do_read_u32(ff, &nr_pmu))
3227                 return -1;
3228
3229         if (!nr_pmu) {
3230                 pr_debug("hybrid cpu pmu capabilities not available\n");
3231                 return 0;
3232         }
3233
3234         nodes = zalloc(sizeof(*nodes) * nr_pmu);
3235         if (!nodes)
3236                 return -ENOMEM;
3237
3238         for (i = 0; i < nr_pmu; i++) {
3239                 struct hybrid_cpc_node *n = &nodes[i];
3240
3241                 ret = process_per_cpu_pmu_caps(ff, &n->nr_cpu_pmu_caps,
3242                                                &n->cpu_pmu_caps,
3243                                                &n->max_branches);
3244                 if (ret)
3245                         goto err;
3246
3247                 n->pmu_name = do_read_string(ff);
3248                 if (!n->pmu_name) {
3249                         ret = -1;
3250                         goto err;
3251                 }
3252         }
3253
3254         ff->ph->env.nr_hybrid_cpc_nodes = nr_pmu;
3255         ff->ph->env.hybrid_cpc_nodes = nodes;
3256         return 0;
3257
3258 err:
3259         for (i = 0; i < nr_pmu; i++) {
3260                 free(nodes[i].cpu_pmu_caps);
3261                 free(nodes[i].pmu_name);
3262         }
3263
3264         free(nodes);
3265         return ret;
3266 }
3267
3268 #define FEAT_OPR(n, func, __full_only) \
3269         [HEADER_##n] = {                                        \
3270                 .name       = __stringify(n),                   \
3271                 .write      = write_##func,                     \
3272                 .print      = print_##func,                     \
3273                 .full_only  = __full_only,                      \
3274                 .process    = process_##func,                   \
3275                 .synthesize = true                              \
3276         }
3277
3278 #define FEAT_OPN(n, func, __full_only) \
3279         [HEADER_##n] = {                                        \
3280                 .name       = __stringify(n),                   \
3281                 .write      = write_##func,                     \
3282                 .print      = print_##func,                     \
3283                 .full_only  = __full_only,                      \
3284                 .process    = process_##func                    \
3285         }
3286
3287 /* feature_ops not implemented: */
3288 #define print_tracing_data      NULL
3289 #define print_build_id          NULL
3290
3291 #define process_branch_stack    NULL
3292 #define process_stat            NULL
3293
3294 // Only used in util/synthetic-events.c
3295 const struct perf_header_feature_ops feat_ops[HEADER_LAST_FEATURE];
3296
3297 const struct perf_header_feature_ops feat_ops[HEADER_LAST_FEATURE] = {
3298         FEAT_OPN(TRACING_DATA,  tracing_data,   false),
3299         FEAT_OPN(BUILD_ID,      build_id,       false),
3300         FEAT_OPR(HOSTNAME,      hostname,       false),
3301         FEAT_OPR(OSRELEASE,     osrelease,      false),
3302         FEAT_OPR(VERSION,       version,        false),
3303         FEAT_OPR(ARCH,          arch,           false),
3304         FEAT_OPR(NRCPUS,        nrcpus,         false),
3305         FEAT_OPR(CPUDESC,       cpudesc,        false),
3306         FEAT_OPR(CPUID,         cpuid,          false),
3307         FEAT_OPR(TOTAL_MEM,     total_mem,      false),
3308         FEAT_OPR(EVENT_DESC,    event_desc,     false),
3309         FEAT_OPR(CMDLINE,       cmdline,        false),
3310         FEAT_OPR(CPU_TOPOLOGY,  cpu_topology,   true),
3311         FEAT_OPR(NUMA_TOPOLOGY, numa_topology,  true),
3312         FEAT_OPN(BRANCH_STACK,  branch_stack,   false),
3313         FEAT_OPR(PMU_MAPPINGS,  pmu_mappings,   false),
3314         FEAT_OPR(GROUP_DESC,    group_desc,     false),
3315         FEAT_OPN(AUXTRACE,      auxtrace,       false),
3316         FEAT_OPN(STAT,          stat,           false),
3317         FEAT_OPN(CACHE,         cache,          true),
3318         FEAT_OPR(SAMPLE_TIME,   sample_time,    false),
3319         FEAT_OPR(MEM_TOPOLOGY,  mem_topology,   true),
3320         FEAT_OPR(CLOCKID,       clockid,        false),
3321         FEAT_OPN(DIR_FORMAT,    dir_format,     false),
3322 #ifdef HAVE_LIBBPF_SUPPORT
3323         FEAT_OPR(BPF_PROG_INFO, bpf_prog_info,  false),
3324         FEAT_OPR(BPF_BTF,       bpf_btf,        false),
3325 #endif
3326         FEAT_OPR(COMPRESSED,    compressed,     false),
3327         FEAT_OPR(CPU_PMU_CAPS,  cpu_pmu_caps,   false),
3328         FEAT_OPR(CLOCK_DATA,    clock_data,     false),
3329         FEAT_OPN(HYBRID_TOPOLOGY,       hybrid_topology,        true),
3330         FEAT_OPR(HYBRID_CPU_PMU_CAPS,   hybrid_cpu_pmu_caps,    false),
3331 };
3332
3333 struct header_print_data {
3334         FILE *fp;
3335         bool full; /* extended list of headers */
3336 };
3337
3338 static int perf_file_section__fprintf_info(struct perf_file_section *section,
3339                                            struct perf_header *ph,
3340                                            int feat, int fd, void *data)
3341 {
3342         struct header_print_data *hd = data;
3343         struct feat_fd ff;
3344
3345         if (lseek(fd, section->offset, SEEK_SET) == (off_t)-1) {
3346                 pr_debug("Failed to lseek to %" PRIu64 " offset for feature "
3347                                 "%d, continuing...\n", section->offset, feat);
3348                 return 0;
3349         }
3350         if (feat >= HEADER_LAST_FEATURE) {
3351                 pr_warning("unknown feature %d\n", feat);
3352                 return 0;
3353         }
3354         if (!feat_ops[feat].print)
3355                 return 0;
3356
3357         ff = (struct  feat_fd) {
3358                 .fd = fd,
3359                 .ph = ph,
3360         };
3361
3362         if (!feat_ops[feat].full_only || hd->full)
3363                 feat_ops[feat].print(&ff, hd->fp);
3364         else
3365                 fprintf(hd->fp, "# %s info available, use -I to display\n",
3366                         feat_ops[feat].name);
3367
3368         return 0;
3369 }
3370
3371 int perf_header__fprintf_info(struct perf_session *session, FILE *fp, bool full)
3372 {
3373         struct header_print_data hd;
3374         struct perf_header *header = &session->header;
3375         int fd = perf_data__fd(session->data);
3376         struct stat st;
3377         time_t stctime;
3378         int ret, bit;
3379
3380         hd.fp = fp;
3381         hd.full = full;
3382
3383         ret = fstat(fd, &st);
3384         if (ret == -1)
3385                 return -1;
3386
3387         stctime = st.st_mtime;
3388         fprintf(fp, "# captured on    : %s", ctime(&stctime));
3389
3390         fprintf(fp, "# header version : %u\n", header->version);
3391         fprintf(fp, "# data offset    : %" PRIu64 "\n", header->data_offset);
3392         fprintf(fp, "# data size      : %" PRIu64 "\n", header->data_size);
3393         fprintf(fp, "# feat offset    : %" PRIu64 "\n", header->feat_offset);
3394
3395         perf_header__process_sections(header, fd, &hd,
3396                                       perf_file_section__fprintf_info);
3397
3398         if (session->data->is_pipe)
3399                 return 0;
3400
3401         fprintf(fp, "# missing features: ");
3402         for_each_clear_bit(bit, header->adds_features, HEADER_LAST_FEATURE) {
3403                 if (bit)
3404                         fprintf(fp, "%s ", feat_ops[bit].name);
3405         }
3406
3407         fprintf(fp, "\n");
3408         return 0;
3409 }
3410
3411 static int do_write_feat(struct feat_fd *ff, int type,
3412                          struct perf_file_section **p,
3413                          struct evlist *evlist)
3414 {
3415         int err;
3416         int ret = 0;
3417
3418         if (perf_header__has_feat(ff->ph, type)) {
3419                 if (!feat_ops[type].write)
3420                         return -1;
3421
3422                 if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__))
3423                         return -1;
3424
3425                 (*p)->offset = lseek(ff->fd, 0, SEEK_CUR);
3426
3427                 err = feat_ops[type].write(ff, evlist);
3428                 if (err < 0) {
3429                         pr_debug("failed to write feature %s\n", feat_ops[type].name);
3430
3431                         /* undo anything written */
3432                         lseek(ff->fd, (*p)->offset, SEEK_SET);
3433
3434                         return -1;
3435                 }
3436                 (*p)->size = lseek(ff->fd, 0, SEEK_CUR) - (*p)->offset;
3437                 (*p)++;
3438         }
3439         return ret;
3440 }
3441
3442 static int perf_header__adds_write(struct perf_header *header,
3443                                    struct evlist *evlist, int fd)
3444 {
3445         int nr_sections;
3446         struct feat_fd ff;
3447         struct perf_file_section *feat_sec, *p;
3448         int sec_size;
3449         u64 sec_start;
3450         int feat;
3451         int err;
3452
3453         ff = (struct feat_fd){
3454                 .fd  = fd,
3455                 .ph = header,
3456         };
3457
3458         nr_sections = bitmap_weight(header->adds_features, HEADER_FEAT_BITS);
3459         if (!nr_sections)
3460                 return 0;
3461
3462         feat_sec = p = calloc(nr_sections, sizeof(*feat_sec));
3463         if (feat_sec == NULL)
3464                 return -ENOMEM;
3465
3466         sec_size = sizeof(*feat_sec) * nr_sections;
3467
3468         sec_start = header->feat_offset;
3469         lseek(fd, sec_start + sec_size, SEEK_SET);
3470
3471         for_each_set_bit(feat, header->adds_features, HEADER_FEAT_BITS) {
3472                 if (do_write_feat(&ff, feat, &p, evlist))
3473                         perf_header__clear_feat(header, feat);
3474         }
3475
3476         lseek(fd, sec_start, SEEK_SET);
3477         /*
3478          * may write more than needed due to dropped feature, but
3479          * this is okay, reader will skip the missing entries
3480          */
3481         err = do_write(&ff, feat_sec, sec_size);
3482         if (err < 0)
3483                 pr_debug("failed to write feature section\n");
3484         free(feat_sec);
3485         return err;
3486 }
3487
3488 int perf_header__write_pipe(int fd)
3489 {
3490         struct perf_pipe_file_header f_header;
3491         struct feat_fd ff;
3492         int err;
3493
3494         ff = (struct feat_fd){ .fd = fd };
3495
3496         f_header = (struct perf_pipe_file_header){
3497                 .magic     = PERF_MAGIC,
3498                 .size      = sizeof(f_header),
3499         };
3500
3501         err = do_write(&ff, &f_header, sizeof(f_header));
3502         if (err < 0) {
3503                 pr_debug("failed to write perf pipe header\n");
3504                 return err;
3505         }
3506
3507         return 0;
3508 }
3509
3510 int perf_session__write_header(struct perf_session *session,
3511                                struct evlist *evlist,
3512                                int fd, bool at_exit)
3513 {
3514         struct perf_file_header f_header;
3515         struct perf_file_attr   f_attr;
3516         struct perf_header *header = &session->header;
3517         struct evsel *evsel;
3518         struct feat_fd ff;
3519         u64 attr_offset;
3520         int err;
3521
3522         ff = (struct feat_fd){ .fd = fd};
3523         lseek(fd, sizeof(f_header), SEEK_SET);
3524
3525         evlist__for_each_entry(session->evlist, evsel) {
3526                 evsel->id_offset = lseek(fd, 0, SEEK_CUR);
3527                 err = do_write(&ff, evsel->core.id, evsel->core.ids * sizeof(u64));
3528                 if (err < 0) {
3529                         pr_debug("failed to write perf header\n");
3530                         return err;
3531                 }
3532         }
3533
3534         attr_offset = lseek(ff.fd, 0, SEEK_CUR);
3535
3536         evlist__for_each_entry(evlist, evsel) {
3537                 if (evsel->core.attr.size < sizeof(evsel->core.attr)) {
3538                         /*
3539                          * We are likely in "perf inject" and have read
3540                          * from an older file. Update attr size so that
3541                          * reader gets the right offset to the ids.
3542                          */
3543                         evsel->core.attr.size = sizeof(evsel->core.attr);
3544                 }
3545                 f_attr = (struct perf_file_attr){
3546                         .attr = evsel->core.attr,
3547                         .ids  = {
3548                                 .offset = evsel->id_offset,
3549                                 .size   = evsel->core.ids * sizeof(u64),
3550                         }
3551                 };
3552                 err = do_write(&ff, &f_attr, sizeof(f_attr));
3553                 if (err < 0) {
3554                         pr_debug("failed to write perf header attribute\n");
3555                         return err;
3556                 }
3557         }
3558
3559         if (!header->data_offset)
3560                 header->data_offset = lseek(fd, 0, SEEK_CUR);
3561         header->feat_offset = header->data_offset + header->data_size;
3562
3563         if (at_exit) {
3564                 err = perf_header__adds_write(header, evlist, fd);
3565                 if (err < 0)
3566                         return err;
3567         }
3568
3569         f_header = (struct perf_file_header){
3570                 .magic     = PERF_MAGIC,
3571                 .size      = sizeof(f_header),
3572                 .attr_size = sizeof(f_attr),
3573                 .attrs = {
3574                         .offset = attr_offset,
3575                         .size   = evlist->core.nr_entries * sizeof(f_attr),
3576                 },
3577                 .data = {
3578                         .offset = header->data_offset,
3579                         .size   = header->data_size,
3580                 },
3581                 /* event_types is ignored, store zeros */
3582         };
3583
3584         memcpy(&f_header.adds_features, &header->adds_features, sizeof(header->adds_features));
3585
3586         lseek(fd, 0, SEEK_SET);
3587         err = do_write(&ff, &f_header, sizeof(f_header));
3588         if (err < 0) {
3589                 pr_debug("failed to write perf header\n");
3590                 return err;
3591         }
3592         lseek(fd, header->data_offset + header->data_size, SEEK_SET);
3593
3594         return 0;
3595 }
3596
3597 static int perf_header__getbuffer64(struct perf_header *header,
3598                                     int fd, void *buf, size_t size)
3599 {
3600         if (readn(fd, buf, size) <= 0)
3601                 return -1;
3602
3603         if (header->needs_swap)
3604                 mem_bswap_64(buf, size);
3605
3606         return 0;
3607 }
3608
3609 int perf_header__process_sections(struct perf_header *header, int fd,
3610                                   void *data,
3611                                   int (*process)(struct perf_file_section *section,
3612                                                  struct perf_header *ph,
3613                                                  int feat, int fd, void *data))
3614 {
3615         struct perf_file_section *feat_sec, *sec;
3616         int nr_sections;
3617         int sec_size;
3618         int feat;
3619         int err;
3620
3621         nr_sections = bitmap_weight(header->adds_features, HEADER_FEAT_BITS);
3622         if (!nr_sections)
3623                 return 0;
3624
3625         feat_sec = sec = calloc(nr_sections, sizeof(*feat_sec));
3626         if (!feat_sec)
3627                 return -1;
3628
3629         sec_size = sizeof(*feat_sec) * nr_sections;
3630
3631         lseek(fd, header->feat_offset, SEEK_SET);
3632
3633         err = perf_header__getbuffer64(header, fd, feat_sec, sec_size);
3634         if (err < 0)
3635                 goto out_free;
3636
3637         for_each_set_bit(feat, header->adds_features, HEADER_LAST_FEATURE) {
3638                 err = process(sec++, header, feat, fd, data);
3639                 if (err < 0)
3640                         goto out_free;
3641         }
3642         err = 0;
3643 out_free:
3644         free(feat_sec);
3645         return err;
3646 }
3647
3648 static const int attr_file_abi_sizes[] = {
3649         [0] = PERF_ATTR_SIZE_VER0,
3650         [1] = PERF_ATTR_SIZE_VER1,
3651         [2] = PERF_ATTR_SIZE_VER2,
3652         [3] = PERF_ATTR_SIZE_VER3,
3653         [4] = PERF_ATTR_SIZE_VER4,
3654         0,
3655 };
3656
3657 /*
3658  * In the legacy file format, the magic number is not used to encode endianness.
3659  * hdr_sz was used to encode endianness. But given that hdr_sz can vary based
3660  * on ABI revisions, we need to try all combinations for all endianness to
3661  * detect the endianness.
3662  */
3663 static int try_all_file_abis(uint64_t hdr_sz, struct perf_header *ph)
3664 {
3665         uint64_t ref_size, attr_size;
3666         int i;
3667
3668         for (i = 0 ; attr_file_abi_sizes[i]; i++) {
3669                 ref_size = attr_file_abi_sizes[i]
3670                          + sizeof(struct perf_file_section);
3671                 if (hdr_sz != ref_size) {
3672                         attr_size = bswap_64(hdr_sz);
3673                         if (attr_size != ref_size)
3674                                 continue;
3675
3676                         ph->needs_swap = true;
3677                 }
3678                 pr_debug("ABI%d perf.data file detected, need_swap=%d\n",
3679                          i,
3680                          ph->needs_swap);
3681                 return 0;
3682         }
3683         /* could not determine endianness */
3684         return -1;
3685 }
3686
3687 #define PERF_PIPE_HDR_VER0      16
3688
3689 static const size_t attr_pipe_abi_sizes[] = {
3690         [0] = PERF_PIPE_HDR_VER0,
3691         0,
3692 };
3693
3694 /*
3695  * In the legacy pipe format, there is an implicit assumption that endianness
3696  * between host recording the samples, and host parsing the samples is the
3697  * same. This is not always the case given that the pipe output may always be
3698  * redirected into a file and analyzed on a different machine with possibly a
3699  * different endianness and perf_event ABI revisions in the perf tool itself.
3700  */
3701 static int try_all_pipe_abis(uint64_t hdr_sz, struct perf_header *ph)
3702 {
3703         u64 attr_size;
3704         int i;
3705
3706         for (i = 0 ; attr_pipe_abi_sizes[i]; i++) {
3707                 if (hdr_sz != attr_pipe_abi_sizes[i]) {
3708                         attr_size = bswap_64(hdr_sz);
3709                         if (attr_size != hdr_sz)
3710                                 continue;
3711
3712                         ph->needs_swap = true;
3713                 }
3714                 pr_debug("Pipe ABI%d perf.data file detected\n", i);
3715                 return 0;
3716         }
3717         return -1;
3718 }
3719
3720 bool is_perf_magic(u64 magic)
3721 {
3722         if (!memcmp(&magic, __perf_magic1, sizeof(magic))
3723                 || magic == __perf_magic2
3724                 || magic == __perf_magic2_sw)
3725                 return true;
3726
3727         return false;
3728 }
3729
3730 static int check_magic_endian(u64 magic, uint64_t hdr_sz,
3731                               bool is_pipe, struct perf_header *ph)
3732 {
3733         int ret;
3734
3735         /* check for legacy format */
3736         ret = memcmp(&magic, __perf_magic1, sizeof(magic));
3737         if (ret == 0) {
3738                 ph->version = PERF_HEADER_VERSION_1;
3739                 pr_debug("legacy perf.data format\n");
3740                 if (is_pipe)
3741                         return try_all_pipe_abis(hdr_sz, ph);
3742
3743                 return try_all_file_abis(hdr_sz, ph);
3744         }
3745         /*
3746          * the new magic number serves two purposes:
3747          * - unique number to identify actual perf.data files
3748          * - encode endianness of file
3749          */
3750         ph->version = PERF_HEADER_VERSION_2;
3751
3752         /* check magic number with one endianness */
3753         if (magic == __perf_magic2)
3754                 return 0;
3755
3756         /* check magic number with opposite endianness */
3757         if (magic != __perf_magic2_sw)
3758                 return -1;
3759
3760         ph->needs_swap = true;
3761
3762         return 0;
3763 }
3764
3765 int perf_file_header__read(struct perf_file_header *header,
3766                            struct perf_header *ph, int fd)
3767 {
3768         ssize_t ret;
3769
3770         lseek(fd, 0, SEEK_SET);
3771
3772         ret = readn(fd, header, sizeof(*header));
3773         if (ret <= 0)
3774                 return -1;
3775
3776         if (check_magic_endian(header->magic,
3777                                header->attr_size, false, ph) < 0) {
3778                 pr_debug("magic/endian check failed\n");
3779                 return -1;
3780         }
3781
3782         if (ph->needs_swap) {
3783                 mem_bswap_64(header, offsetof(struct perf_file_header,
3784                              adds_features));
3785         }
3786
3787         if (header->size != sizeof(*header)) {
3788                 /* Support the previous format */
3789                 if (header->size == offsetof(typeof(*header), adds_features))
3790                         bitmap_zero(header->adds_features, HEADER_FEAT_BITS);
3791                 else
3792                         return -1;
3793         } else if (ph->needs_swap) {
3794                 /*
3795                  * feature bitmap is declared as an array of unsigned longs --
3796                  * not good since its size can differ between the host that
3797                  * generated the data file and the host analyzing the file.
3798                  *
3799                  * We need to handle endianness, but we don't know the size of
3800                  * the unsigned long where the file was generated. Take a best
3801                  * guess at determining it: try 64-bit swap first (ie., file
3802                  * created on a 64-bit host), and check if the hostname feature
3803                  * bit is set (this feature bit is forced on as of fbe96f2).
3804                  * If the bit is not, undo the 64-bit swap and try a 32-bit
3805                  * swap. If the hostname bit is still not set (e.g., older data
3806                  * file), punt and fallback to the original behavior --
3807                  * clearing all feature bits and setting buildid.
3808                  */
3809                 mem_bswap_64(&header->adds_features,
3810                             BITS_TO_U64(HEADER_FEAT_BITS));
3811
3812                 if (!test_bit(HEADER_HOSTNAME, header->adds_features)) {
3813                         /* unswap as u64 */
3814                         mem_bswap_64(&header->adds_features,
3815                                     BITS_TO_U64(HEADER_FEAT_BITS));
3816
3817                         /* unswap as u32 */
3818                         mem_bswap_32(&header->adds_features,
3819                                     BITS_TO_U32(HEADER_FEAT_BITS));
3820                 }
3821
3822                 if (!test_bit(HEADER_HOSTNAME, header->adds_features)) {
3823                         bitmap_zero(header->adds_features, HEADER_FEAT_BITS);
3824                         set_bit(HEADER_BUILD_ID, header->adds_features);
3825                 }
3826         }
3827
3828         memcpy(&ph->adds_features, &header->adds_features,
3829                sizeof(ph->adds_features));
3830
3831         ph->data_offset  = header->data.offset;
3832         ph->data_size    = header->data.size;
3833         ph->feat_offset  = header->data.offset + header->data.size;
3834         return 0;
3835 }
3836
3837 static int perf_file_section__process(struct perf_file_section *section,
3838                                       struct perf_header *ph,
3839                                       int feat, int fd, void *data)
3840 {
3841         struct feat_fd fdd = {
3842                 .fd     = fd,
3843                 .ph     = ph,
3844                 .size   = section->size,
3845                 .offset = section->offset,
3846         };
3847
3848         if (lseek(fd, section->offset, SEEK_SET) == (off_t)-1) {
3849                 pr_debug("Failed to lseek to %" PRIu64 " offset for feature "
3850                           "%d, continuing...\n", section->offset, feat);
3851                 return 0;
3852         }
3853
3854         if (feat >= HEADER_LAST_FEATURE) {
3855                 pr_debug("unknown feature %d, continuing...\n", feat);
3856                 return 0;
3857         }
3858
3859         if (!feat_ops[feat].process)
3860                 return 0;
3861
3862         return feat_ops[feat].process(&fdd, data);
3863 }
3864
3865 static int perf_file_header__read_pipe(struct perf_pipe_file_header *header,
3866                                        struct perf_header *ph,
3867                                        struct perf_data* data,
3868                                        bool repipe)
3869 {
3870         struct feat_fd ff = {
3871                 .fd = STDOUT_FILENO,
3872                 .ph = ph,
3873         };
3874         ssize_t ret;
3875
3876         ret = perf_data__read(data, header, sizeof(*header));
3877         if (ret <= 0)
3878                 return -1;
3879
3880         if (check_magic_endian(header->magic, header->size, true, ph) < 0) {
3881                 pr_debug("endian/magic failed\n");
3882                 return -1;
3883         }
3884
3885         if (ph->needs_swap)
3886                 header->size = bswap_64(header->size);
3887
3888         if (repipe && do_write(&ff, header, sizeof(*header)) < 0)
3889                 return -1;
3890
3891         return 0;
3892 }
3893
3894 static int perf_header__read_pipe(struct perf_session *session)
3895 {
3896         struct perf_header *header = &session->header;
3897         struct perf_pipe_file_header f_header;
3898
3899         if (perf_file_header__read_pipe(&f_header, header, session->data,
3900                                         session->repipe) < 0) {
3901                 pr_debug("incompatible file format\n");
3902                 return -EINVAL;
3903         }
3904
3905         return f_header.size == sizeof(f_header) ? 0 : -1;
3906 }
3907
3908 static int read_attr(int fd, struct perf_header *ph,
3909                      struct perf_file_attr *f_attr)
3910 {
3911         struct perf_event_attr *attr = &f_attr->attr;
3912         size_t sz, left;
3913         size_t our_sz = sizeof(f_attr->attr);
3914         ssize_t ret;
3915
3916         memset(f_attr, 0, sizeof(*f_attr));
3917
3918         /* read minimal guaranteed structure */
3919         ret = readn(fd, attr, PERF_ATTR_SIZE_VER0);
3920         if (ret <= 0) {
3921                 pr_debug("cannot read %d bytes of header attr\n",
3922                          PERF_ATTR_SIZE_VER0);
3923                 return -1;
3924         }
3925
3926         /* on file perf_event_attr size */
3927         sz = attr->size;
3928
3929         if (ph->needs_swap)
3930                 sz = bswap_32(sz);
3931
3932         if (sz == 0) {
3933                 /* assume ABI0 */
3934                 sz =  PERF_ATTR_SIZE_VER0;
3935         } else if (sz > our_sz) {
3936                 pr_debug("file uses a more recent and unsupported ABI"
3937                          " (%zu bytes extra)\n", sz - our_sz);
3938                 return -1;
3939         }
3940         /* what we have not yet read and that we know about */
3941         left = sz - PERF_ATTR_SIZE_VER0;
3942         if (left) {
3943                 void *ptr = attr;
3944                 ptr += PERF_ATTR_SIZE_VER0;
3945
3946                 ret = readn(fd, ptr, left);
3947         }
3948         /* read perf_file_section, ids are read in caller */
3949         ret = readn(fd, &f_attr->ids, sizeof(f_attr->ids));
3950
3951         return ret <= 0 ? -1 : 0;
3952 }
3953
3954 static int evsel__prepare_tracepoint_event(struct evsel *evsel, struct tep_handle *pevent)
3955 {
3956         struct tep_event *event;
3957         char bf[128];
3958
3959         /* already prepared */
3960         if (evsel->tp_format)
3961                 return 0;
3962
3963         if (pevent == NULL) {
3964                 pr_debug("broken or missing trace data\n");
3965                 return -1;
3966         }
3967
3968         event = tep_find_event(pevent, evsel->core.attr.config);
3969         if (event == NULL) {
3970                 pr_debug("cannot find event format for %d\n", (int)evsel->core.attr.config);
3971                 return -1;
3972         }
3973
3974         if (!evsel->name) {
3975                 snprintf(bf, sizeof(bf), "%s:%s", event->system, event->name);
3976                 evsel->name = strdup(bf);
3977                 if (evsel->name == NULL)
3978                         return -1;
3979         }
3980
3981         evsel->tp_format = event;
3982         return 0;
3983 }
3984
3985 static int evlist__prepare_tracepoint_events(struct evlist *evlist, struct tep_handle *pevent)
3986 {
3987         struct evsel *pos;
3988
3989         evlist__for_each_entry(evlist, pos) {
3990                 if (pos->core.attr.type == PERF_TYPE_TRACEPOINT &&
3991                     evsel__prepare_tracepoint_event(pos, pevent))
3992                         return -1;
3993         }
3994
3995         return 0;
3996 }
3997
3998 int perf_session__read_header(struct perf_session *session)
3999 {
4000         struct perf_data *data = session->data;
4001         struct perf_header *header = &session->header;
4002         struct perf_file_header f_header;
4003         struct perf_file_attr   f_attr;
4004         u64                     f_id;
4005         int nr_attrs, nr_ids, i, j, err;
4006         int fd = perf_data__fd(data);
4007
4008         session->evlist = evlist__new();
4009         if (session->evlist == NULL)
4010                 return -ENOMEM;
4011
4012         session->evlist->env = &header->env;
4013         session->machines.host.env = &header->env;
4014
4015         /*
4016          * We can read 'pipe' data event from regular file,
4017          * check for the pipe header regardless of source.
4018          */
4019         err = perf_header__read_pipe(session);
4020         if (!err || perf_data__is_pipe(data)) {
4021                 data->is_pipe = true;
4022                 return err;
4023         }
4024
4025         if (perf_file_header__read(&f_header, header, fd) < 0)
4026                 return -EINVAL;
4027
4028         if (header->needs_swap && data->in_place_update) {
4029                 pr_err("In-place update not supported when byte-swapping is required\n");
4030                 return -EINVAL;
4031         }
4032
4033         /*
4034          * Sanity check that perf.data was written cleanly; data size is
4035          * initialized to 0 and updated only if the on_exit function is run.
4036          * If data size is still 0 then the file contains only partial
4037          * information.  Just warn user and process it as much as it can.
4038          */
4039         if (f_header.data.size == 0) {
4040                 pr_warning("WARNING: The %s file's data size field is 0 which is unexpected.\n"
4041                            "Was the 'perf record' command properly terminated?\n",
4042                            data->file.path);
4043         }
4044
4045         if (f_header.attr_size == 0) {
4046                 pr_err("ERROR: The %s file's attr size field is 0 which is unexpected.\n"
4047                        "Was the 'perf record' command properly terminated?\n",
4048                        data->file.path);
4049                 return -EINVAL;
4050         }
4051
4052         nr_attrs = f_header.attrs.size / f_header.attr_size;
4053         lseek(fd, f_header.attrs.offset, SEEK_SET);
4054
4055         for (i = 0; i < nr_attrs; i++) {
4056                 struct evsel *evsel;
4057                 off_t tmp;
4058
4059                 if (read_attr(fd, header, &f_attr) < 0)
4060                         goto out_errno;
4061
4062                 if (header->needs_swap) {
4063                         f_attr.ids.size   = bswap_64(f_attr.ids.size);
4064                         f_attr.ids.offset = bswap_64(f_attr.ids.offset);
4065                         perf_event__attr_swap(&f_attr.attr);
4066                 }
4067
4068                 tmp = lseek(fd, 0, SEEK_CUR);
4069                 evsel = evsel__new(&f_attr.attr);
4070
4071                 if (evsel == NULL)
4072                         goto out_delete_evlist;
4073
4074                 evsel->needs_swap = header->needs_swap;
4075                 /*
4076                  * Do it before so that if perf_evsel__alloc_id fails, this
4077                  * entry gets purged too at evlist__delete().
4078                  */
4079                 evlist__add(session->evlist, evsel);
4080
4081                 nr_ids = f_attr.ids.size / sizeof(u64);
4082                 /*
4083                  * We don't have the cpu and thread maps on the header, so
4084                  * for allocating the perf_sample_id table we fake 1 cpu and
4085                  * hattr->ids threads.
4086                  */
4087                 if (perf_evsel__alloc_id(&evsel->core, 1, nr_ids))
4088                         goto out_delete_evlist;
4089
4090                 lseek(fd, f_attr.ids.offset, SEEK_SET);
4091
4092                 for (j = 0; j < nr_ids; j++) {
4093                         if (perf_header__getbuffer64(header, fd, &f_id, sizeof(f_id)))
4094                                 goto out_errno;
4095
4096                         perf_evlist__id_add(&session->evlist->core, &evsel->core, 0, j, f_id);
4097                 }
4098
4099                 lseek(fd, tmp, SEEK_SET);
4100         }
4101
4102         perf_header__process_sections(header, fd, &session->tevent,
4103                                       perf_file_section__process);
4104
4105         if (evlist__prepare_tracepoint_events(session->evlist, session->tevent.pevent))
4106                 goto out_delete_evlist;
4107
4108         return 0;
4109 out_errno:
4110         return -errno;
4111
4112 out_delete_evlist:
4113         evlist__delete(session->evlist);
4114         session->evlist = NULL;
4115         return -ENOMEM;
4116 }
4117
4118 int perf_event__process_feature(struct perf_session *session,
4119                                 union perf_event *event)
4120 {
4121         struct perf_tool *tool = session->tool;
4122         struct feat_fd ff = { .fd = 0 };
4123         struct perf_record_header_feature *fe = (struct perf_record_header_feature *)event;
4124         int type = fe->header.type;
4125         u64 feat = fe->feat_id;
4126
4127         if (type < 0 || type >= PERF_RECORD_HEADER_MAX) {
4128                 pr_warning("invalid record type %d in pipe-mode\n", type);
4129                 return 0;
4130         }
4131         if (feat == HEADER_RESERVED || feat >= HEADER_LAST_FEATURE) {
4132                 pr_warning("invalid record type %d in pipe-mode\n", type);
4133                 return -1;
4134         }
4135
4136         if (!feat_ops[feat].process)
4137                 return 0;
4138
4139         ff.buf  = (void *)fe->data;
4140         ff.size = event->header.size - sizeof(*fe);
4141         ff.ph = &session->header;
4142
4143         if (feat_ops[feat].process(&ff, NULL))
4144                 return -1;
4145
4146         if (!feat_ops[feat].print || !tool->show_feat_hdr)
4147                 return 0;
4148
4149         if (!feat_ops[feat].full_only ||
4150             tool->show_feat_hdr >= SHOW_FEAT_HEADER_FULL_INFO) {
4151                 feat_ops[feat].print(&ff, stdout);
4152         } else {
4153                 fprintf(stdout, "# %s info available, use -I to display\n",
4154                         feat_ops[feat].name);
4155         }
4156
4157         return 0;
4158 }
4159
4160 size_t perf_event__fprintf_event_update(union perf_event *event, FILE *fp)
4161 {
4162         struct perf_record_event_update *ev = &event->event_update;
4163         struct perf_record_event_update_scale *ev_scale;
4164         struct perf_record_event_update_cpus *ev_cpus;
4165         struct perf_cpu_map *map;
4166         size_t ret;
4167
4168         ret = fprintf(fp, "\n... id:    %" PRI_lu64 "\n", ev->id);
4169
4170         switch (ev->type) {
4171         case PERF_EVENT_UPDATE__SCALE:
4172                 ev_scale = (struct perf_record_event_update_scale *)ev->data;
4173                 ret += fprintf(fp, "... scale: %f\n", ev_scale->scale);
4174                 break;
4175         case PERF_EVENT_UPDATE__UNIT:
4176                 ret += fprintf(fp, "... unit:  %s\n", ev->data);
4177                 break;
4178         case PERF_EVENT_UPDATE__NAME:
4179                 ret += fprintf(fp, "... name:  %s\n", ev->data);
4180                 break;
4181         case PERF_EVENT_UPDATE__CPUS:
4182                 ev_cpus = (struct perf_record_event_update_cpus *)ev->data;
4183                 ret += fprintf(fp, "... ");
4184
4185                 map = cpu_map__new_data(&ev_cpus->cpus);
4186                 if (map)
4187                         ret += cpu_map__fprintf(map, fp);
4188                 else
4189                         ret += fprintf(fp, "failed to get cpus\n");
4190                 break;
4191         default:
4192                 ret += fprintf(fp, "... unknown type\n");
4193                 break;
4194         }
4195
4196         return ret;
4197 }
4198
4199 int perf_event__process_attr(struct perf_tool *tool __maybe_unused,
4200                              union perf_event *event,
4201                              struct evlist **pevlist)
4202 {
4203         u32 i, ids, n_ids;
4204         struct evsel *evsel;
4205         struct evlist *evlist = *pevlist;
4206
4207         if (evlist == NULL) {
4208                 *pevlist = evlist = evlist__new();
4209                 if (evlist == NULL)
4210                         return -ENOMEM;
4211         }
4212
4213         evsel = evsel__new(&event->attr.attr);
4214         if (evsel == NULL)
4215                 return -ENOMEM;
4216
4217         evlist__add(evlist, evsel);
4218
4219         ids = event->header.size;
4220         ids -= (void *)&event->attr.id - (void *)event;
4221         n_ids = ids / sizeof(u64);
4222         /*
4223          * We don't have the cpu and thread maps on the header, so
4224          * for allocating the perf_sample_id table we fake 1 cpu and
4225          * hattr->ids threads.
4226          */
4227         if (perf_evsel__alloc_id(&evsel->core, 1, n_ids))
4228                 return -ENOMEM;
4229
4230         for (i = 0; i < n_ids; i++) {
4231                 perf_evlist__id_add(&evlist->core, &evsel->core, 0, i, event->attr.id[i]);
4232         }
4233
4234         return 0;
4235 }
4236
4237 int perf_event__process_event_update(struct perf_tool *tool __maybe_unused,
4238                                      union perf_event *event,
4239                                      struct evlist **pevlist)
4240 {
4241         struct perf_record_event_update *ev = &event->event_update;
4242         struct perf_record_event_update_scale *ev_scale;
4243         struct perf_record_event_update_cpus *ev_cpus;
4244         struct evlist *evlist;
4245         struct evsel *evsel;
4246         struct perf_cpu_map *map;
4247
4248         if (!pevlist || *pevlist == NULL)
4249                 return -EINVAL;
4250
4251         evlist = *pevlist;
4252
4253         evsel = evlist__id2evsel(evlist, ev->id);
4254         if (evsel == NULL)
4255                 return -EINVAL;
4256
4257         switch (ev->type) {
4258         case PERF_EVENT_UPDATE__UNIT:
4259                 evsel->unit = strdup(ev->data);
4260                 break;
4261         case PERF_EVENT_UPDATE__NAME:
4262                 evsel->name = strdup(ev->data);
4263                 break;
4264         case PERF_EVENT_UPDATE__SCALE:
4265                 ev_scale = (struct perf_record_event_update_scale *)ev->data;
4266                 evsel->scale = ev_scale->scale;
4267                 break;
4268         case PERF_EVENT_UPDATE__CPUS:
4269                 ev_cpus = (struct perf_record_event_update_cpus *)ev->data;
4270
4271                 map = cpu_map__new_data(&ev_cpus->cpus);
4272                 if (map)
4273                         evsel->core.own_cpus = map;
4274                 else
4275                         pr_err("failed to get event_update cpus\n");
4276         default:
4277                 break;
4278         }
4279
4280         return 0;
4281 }
4282
4283 int perf_event__process_tracing_data(struct perf_session *session,
4284                                      union perf_event *event)
4285 {
4286         ssize_t size_read, padding, size = event->tracing_data.size;
4287         int fd = perf_data__fd(session->data);
4288         char buf[BUFSIZ];
4289
4290         /*
4291          * The pipe fd is already in proper place and in any case
4292          * we can't move it, and we'd screw the case where we read
4293          * 'pipe' data from regular file. The trace_report reads
4294          * data from 'fd' so we need to set it directly behind the
4295          * event, where the tracing data starts.
4296          */
4297         if (!perf_data__is_pipe(session->data)) {
4298                 off_t offset = lseek(fd, 0, SEEK_CUR);
4299
4300                 /* setup for reading amidst mmap */
4301                 lseek(fd, offset + sizeof(struct perf_record_header_tracing_data),
4302                       SEEK_SET);
4303         }
4304
4305         size_read = trace_report(fd, &session->tevent,
4306                                  session->repipe);
4307         padding = PERF_ALIGN(size_read, sizeof(u64)) - size_read;
4308
4309         if (readn(fd, buf, padding) < 0) {
4310                 pr_err("%s: reading input file", __func__);
4311                 return -1;
4312         }
4313         if (session->repipe) {
4314                 int retw = write(STDOUT_FILENO, buf, padding);
4315                 if (retw <= 0 || retw != padding) {
4316                         pr_err("%s: repiping tracing data padding", __func__);
4317                         return -1;
4318                 }
4319         }
4320
4321         if (size_read + padding != size) {
4322                 pr_err("%s: tracing data size mismatch", __func__);
4323                 return -1;
4324         }
4325
4326         evlist__prepare_tracepoint_events(session->evlist, session->tevent.pevent);
4327
4328         return size_read + padding;
4329 }
4330
4331 int perf_event__process_build_id(struct perf_session *session,
4332                                  union perf_event *event)
4333 {
4334         __event_process_build_id(&event->build_id,
4335                                  event->build_id.filename,
4336                                  session);
4337         return 0;
4338 }