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 #include <bpf/libbpf.h>
  23 #include <perf/cpumap.h>
  24
  25 #include "dso.h"
  26 #include "evlist.h"
  27 #include "evsel.h"
  28 #include "util/evsel_fprintf.h"
  29 #include "header.h"
  30 #include "memswap.h"
  31 #include "trace-event.h"
  32 #include "session.h"
  33 #include "symbol.h"
  34 #include "debug.h"
  35 #include "cpumap.h"
  36 #include "pmu.h"
  37 #include "vdso.h"
  38 #include "strbuf.h"
  39 #include "build-id.h"
  40 #include "data.h"
  41 #include <api/fs/fs.h>
  42 #include "asm/bug.h"
  43 #include "tool.h"
  44 #include "time-utils.h"
  45 #include "units.h"
  46 #include "util/util.h" // perf_exe()
  47 #include "cputopo.h"
  48 #include "bpf-event.h"
  49 #include "clockid.h"
  50
  51 #include <linux/ctype.h>
  52 #include <internal/lib.h>
  53
  54 /*
  55  * magic2 = "PERFILE2"
  56  * must be a numerical value to let the endianness
  57  * determine the memory layout. That way we are able
  58  * to detect endianness when reading the perf.data file
  59  * back.
  60  *
  61  * we check for legacy (PERFFILE) format.
  62  */
  63 static const char *__perf_magic1 = "PERFFILE";
  64 static const u64 __perf_magic2    = 0x32454c4946524550ULL;
  65 static const u64 __perf_magic2_sw = 0x50455246494c4532ULL;
  66
  67 #define PERF_MAGIC      __perf_magic2
  68
  69 const char perf_version_string[] = PERF_VERSION;
  70
  71 struct perf_file_attr {
  72         struct perf_event_attr  attr;
  73         struct perf_file_section        ids;
  74 };
  75
  76 void perf_header__set_feat(struct perf_header *header, int feat)
  77 {
  78         set_bit(feat, header->adds_features);
  79 }
  80
  81 void perf_header__clear_feat(struct perf_header *header, int feat)
  82 {
  83         clear_bit(feat, header->adds_features);
  84 }
  85
  86 bool perf_header__has_feat(const struct perf_header *header, int feat)
  87 {
  88         return test_bit(feat, header->adds_features);
  89 }
  90
  91 static int __do_write_fd(struct feat_fd *ff, const void *buf, size_t size)
  92 {
  93         ssize_t ret = writen(ff->fd, buf, size);
  94
  95         if (ret != (ssize_t)size)
  96                 return ret < 0 ? (int)ret : -1;
  97         return 0;
  98 }
  99
 100 static int __do_write_buf(struct feat_fd *ff,  const void *buf, size_t size)
 101 {
 102         /* struct perf_event_header::size is u16 */
 103         const size_t max_size = 0xffff - sizeof(struct perf_event_header);
 104         size_t new_size = ff->size;
 105         void *addr;
 106
 107         if (size + ff->offset > max_size)
 108                 return -E2BIG;
 109
 110         while (size > (new_size - ff->offset))
 111                 new_size <<= 1;
 112         new_size = min(max_size, new_size);
 113
 114         if (ff->size < new_size) {
 115                 addr = realloc(ff->buf, new_size);
 116                 if (!addr)
 117                         return -ENOMEM;
 118                 ff->buf = addr;
 119                 ff->size = new_size;
 120         }
 121
 122         memcpy(ff->buf + ff->offset, buf, size);
 123         ff->offset += size;
 124
 125         return 0;
 126 }
 127
 128 /* Return: 0 if succeded, -ERR if failed. */
 129 int do_write(struct feat_fd *ff, const void *buf, size_t size)
 130 {
 131         if (!ff->buf)
 132                 return __do_write_fd(ff, buf, size);
 133         return __do_write_buf(ff, buf, size);
 134 }
 135
 136 /* Return: 0 if succeded, -ERR if failed. */
 137 static int do_write_bitmap(struct feat_fd *ff, unsigned long *set, u64 size)
 138 {
 139         u64 *p = (u64 *) set;
 140         int i, ret;
 141
 142         ret = do_write(ff, &size, sizeof(size));
 143         if (ret < 0)
 144                 return ret;
 145
 146         for (i = 0; (u64) i < BITS_TO_U64(size); i++) {
 147                 ret = do_write(ff, p + i, sizeof(*p));
 148                 if (ret < 0)
 149                         return ret;
 150         }
 151
 152         return 0;
 153 }
 154
 155 /* Return: 0 if succeded, -ERR if failed. */
 156 int write_padded(struct feat_fd *ff, const void *bf,
 157                  size_t count, size_t count_aligned)
 158 {
 159         static const char zero_buf[NAME_ALIGN];
 160         int err = do_write(ff, bf, count);
 161
 162         if (!err)
 163                 err = do_write(ff, zero_buf, count_aligned - count);
 164
 165         return err;
 166 }
 167
 168 #define string_size(str)                                                \
 169         (PERF_ALIGN((strlen(str) + 1), NAME_ALIGN) + sizeof(u32))
 170
 171 /* Return: 0 if succeded, -ERR if failed. */
 172 static int do_write_string(struct feat_fd *ff, const char *str)
 173 {
 174         u32 len, olen;
 175         int ret;
 176
 177         olen = strlen(str) + 1;
 178         len = PERF_ALIGN(olen, NAME_ALIGN);
 179
 180         /* write len, incl. \0 */
 181         ret = do_write(ff, &len, sizeof(len));
 182         if (ret < 0)
 183                 return ret;
 184
 185         return write_padded(ff, str, olen, len);
 186 }
 187
 188 static int __do_read_fd(struct feat_fd *ff, void *addr, ssize_t size)
 189 {
 190         ssize_t ret = readn(ff->fd, addr, size);
 191
 192         if (ret != size)
 193                 return ret < 0 ? (int)ret : -1;
 194         return 0;
 195 }
 196
 197 static int __do_read_buf(struct feat_fd *ff, void *addr, ssize_t size)
 198 {
 199         if (size > (ssize_t)ff->size - ff->offset)
 200                 return -1;
 201
 202         memcpy(addr, ff->buf + ff->offset, size);
 203         ff->offset += size;
 204
 205         return 0;
 206
 207 }
 208
 209 static int __do_read(struct feat_fd *ff, void *addr, ssize_t size)
 210 {
 211         if (!ff->buf)
 212                 return __do_read_fd(ff, addr, size);
 213         return __do_read_buf(ff, addr, size);
 214 }
 215
 216 static int do_read_u32(struct feat_fd *ff, u32 *addr)
 217 {
 218         int ret;
 219
 220         ret = __do_read(ff, addr, sizeof(*addr));
 221         if (ret)
 222                 return ret;
 223
 224         if (ff->ph->needs_swap)
 225                 *addr = bswap_32(*addr);
 226         return 0;
 227 }
 228
 229 static int do_read_u64(struct feat_fd *ff, u64 *addr)
 230 {
 231         int ret;
 232
 233         ret = __do_read(ff, addr, sizeof(*addr));
 234         if (ret)
 235                 return ret;
 236
 237         if (ff->ph->needs_swap)
 238                 *addr = bswap_64(*addr);
 239         return 0;
 240 }
 241
 242 static char *do_read_string(struct feat_fd *ff)
 243 {
 244         u32 len;
 245         char *buf;
 246
 247         if (do_read_u32(ff, &len))
 248                 return NULL;
 249
 250         buf = malloc(len);
 251         if (!buf)
 252                 return NULL;
 253
 254         if (!__do_read(ff, buf, len)) {
 255                 /*
 256                  * strings are padded by zeroes
 257                  * thus the actual strlen of buf
 258                  * may be less than len
 259                  */
 260                 return buf;
 261         }
 262
 263         free(buf);
 264         return NULL;
 265 }
 266
 267 /* Return: 0 if succeded, -ERR if failed. */
 268 static int do_read_bitmap(struct feat_fd *ff, unsigned long **pset, u64 *psize)
 269 {
 270         unsigned long *set;
 271         u64 size, *p;
 272         int i, ret;
 273
 274         ret = do_read_u64(ff, &size);
 275         if (ret)
 276                 return ret;
 277
 278         set = bitmap_alloc(size);
 279         if (!set)
 280                 return -ENOMEM;
 281
 282         p = (u64 *) set;
 283
 284         for (i = 0; (u64) i < BITS_TO_U64(size); i++) {
 285                 ret = do_read_u64(ff, p + i);
 286                 if (ret < 0) {
 287                         free(set);
 288                         return ret;
 289                 }
 290         }
 291
 292         *pset  = set;
 293         *psize = size;
 294         return 0;
 295 }
 296
 297 static int write_tracing_data(struct feat_fd *ff,
 298                               struct evlist *evlist)
 299 {
 300         if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__))
 301                 return -1;
 302
 303         return read_tracing_data(ff->fd, &evlist->core.entries);
 304 }
 305
 306 static int write_build_id(struct feat_fd *ff,
 307                           struct evlist *evlist __maybe_unused)
 308 {
 309         struct perf_session *session;
 310         int err;
 311
 312         session = container_of(ff->ph, struct perf_session, header);
 313
 314         if (!perf_session__read_build_ids(session, true))
 315                 return -1;
 316
 317         if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__))
 318                 return -1;
 319
 320         err = perf_session__write_buildid_table(session, ff);
 321         if (err < 0) {
 322                 pr_debug("failed to write buildid table\n");
 323                 return err;
 324         }
 325         perf_session__cache_build_ids(session);
 326
 327         return 0;
 328 }
 329
 330 static int write_hostname(struct feat_fd *ff,
 331                           struct evlist *evlist __maybe_unused)
 332 {
 333         struct utsname uts;
 334         int ret;
 335
 336         ret = uname(&uts);
 337         if (ret < 0)
 338                 return -1;
 339
 340         return do_write_string(ff, uts.nodename);
 341 }
 342
 343 static int write_osrelease(struct feat_fd *ff,
 344                            struct evlist *evlist __maybe_unused)
 345 {
 346         struct utsname uts;
 347         int ret;
 348
 349         ret = uname(&uts);
 350         if (ret < 0)
 351                 return -1;
 352
 353         return do_write_string(ff, uts.release);
 354 }
 355
 356 static int write_arch(struct feat_fd *ff,
 357                       struct evlist *evlist __maybe_unused)
 358 {
 359         struct utsname uts;
 360         int ret;
 361
 362         ret = uname(&uts);
 363         if (ret < 0)
 364                 return -1;
 365
 366         return do_write_string(ff, uts.machine);
 367 }
 368
 369 static int write_version(struct feat_fd *ff,
 370                          struct evlist *evlist __maybe_unused)
 371 {
 372         return do_write_string(ff, perf_version_string);
 373 }
 374
 375 static int __write_cpudesc(struct feat_fd *ff, const char *cpuinfo_proc)
 376 {
 377         FILE *file;
 378         char *buf = NULL;
 379         char *s, *p;
 380         const char *search = cpuinfo_proc;
 381         size_t len = 0;
 382         int ret = -1;
 383
 384         if (!search)
 385                 return -1;
 386
 387         file = fopen("/proc/cpuinfo", "r");
 388         if (!file)
 389                 return -1;
 390
 391         while (getline(&buf, &len, file) > 0) {
 392                 ret = strncmp(buf, search, strlen(search));
 393                 if (!ret)
 394                         break;
 395         }
 396
 397         if (ret) {
 398                 ret = -1;
 399                 goto done;
 400         }
 401
 402         s = buf;
 403
 404         p = strchr(buf, ':');
 405         if (p && *(p+1) == ' ' && *(p+2))
 406                 s = p + 2;
 407         p = strchr(s, '\n');
 408         if (p)
 409                 *p = '\0';
 410
 411         /* squash extra space characters (branding string) */
 412         p = s;
 413         while (*p) {
 414                 if (isspace(*p)) {
 415                         char *r = p + 1;
 416                         char *q = skip_spaces(r);
 417                         *p = ' ';
 418                         if (q != (p+1))
 419                                 while ((*r++ = *q++));
 420                 }
 421                 p++;
 422         }
 423         ret = do_write_string(ff, s);
 424 done:
 425         free(buf);
 426         fclose(file);
 427         return ret;
 428 }
 429
 430 static int write_cpudesc(struct feat_fd *ff,
 431                        struct evlist *evlist __maybe_unused)
 432 {
 433 #if defined(__powerpc__) || defined(__hppa__) || defined(__sparc__)
 434 #define CPUINFO_PROC    { "cpu", }
 435 #elif defined(__s390__)
 436 #define CPUINFO_PROC    { "vendor_id", }
 437 #elif defined(__sh__)
 438 #define CPUINFO_PROC    { "cpu type", }
 439 #elif defined(__alpha__) || defined(__mips__)
 440 #define CPUINFO_PROC    { "cpu model", }
 441 #elif defined(__arm__)
 442 #define CPUINFO_PROC    { "model name", "Processor", }
 443 #elif defined(__arc__)
 444 #define CPUINFO_PROC    { "Processor", }
 445 #elif defined(__xtensa__)
 446 #define CPUINFO_PROC    { "core ID", }
 447 #else
 448 #define CPUINFO_PROC    { "model name", }
 449 #endif
 450         const char *cpuinfo_procs[] = CPUINFO_PROC;
 451 #undef CPUINFO_PROC
 452         unsigned int i;
 453
 454         for (i = 0; i < ARRAY_SIZE(cpuinfo_procs); i++) {
 455                 int ret;
 456                 ret = __write_cpudesc(ff, cpuinfo_procs[i]);
 457                 if (ret >= 0)
 458                         return ret;
 459         }
 460         return -1;
 461 }
 462
 463
 464 static int write_nrcpus(struct feat_fd *ff,
 465                         struct evlist *evlist __maybe_unused)
 466 {
 467         long nr;
 468         u32 nrc, nra;
 469         int ret;
 470
 471         nrc = cpu__max_present_cpu();
 472
 473         nr = sysconf(_SC_NPROCESSORS_ONLN);
 474         if (nr < 0)
 475                 return -1;
 476
 477         nra = (u32)(nr & UINT_MAX);
 478
 479         ret = do_write(ff, &nrc, sizeof(nrc));
 480         if (ret < 0)
 481                 return ret;
 482
 483         return do_write(ff, &nra, sizeof(nra));
 484 }
 485
 486 static int write_event_desc(struct feat_fd *ff,
 487                             struct evlist *evlist)
 488 {
 489         struct evsel *evsel;
 490         u32 nre, nri, sz;
 491         int ret;
 492
 493         nre = evlist->core.nr_entries;
 494
 495         /*
 496          * write number of events
 497          */
 498         ret = do_write(ff, &nre, sizeof(nre));
 499         if (ret < 0)
 500                 return ret;
 501
 502         /*
 503          * size of perf_event_attr struct
 504          */
 505         sz = (u32)sizeof(evsel->core.attr);
 506         ret = do_write(ff, &sz, sizeof(sz));
 507         if (ret < 0)
 508                 return ret;
 509
 510         evlist__for_each_entry(evlist, evsel) {
 511                 ret = do_write(ff, &evsel->core.attr, sz);
 512                 if (ret < 0)
 513                         return ret;
 514                 /*
 515                  * write number of unique id per event
 516                  * there is one id per instance of an event
 517                  *
 518                  * copy into an nri to be independent of the
 519                  * type of ids,
 520                  */
 521                 nri = evsel->core.ids;
 522                 ret = do_write(ff, &nri, sizeof(nri));
 523                 if (ret < 0)
 524                         return ret;
 525
 526                 /*
 527                  * write event string as passed on cmdline
 528                  */
 529                 ret = do_write_string(ff, evsel__name(evsel));
 530                 if (ret < 0)
 531                         return ret;
 532                 /*
 533                  * write unique ids for this event
 534                  */
 535                 ret = do_write(ff, evsel->core.id, evsel->core.ids * sizeof(u64));
 536                 if (ret < 0)
 537                         return ret;
 538         }
 539         return 0;
 540 }
 541
 542 static int write_cmdline(struct feat_fd *ff,
 543                          struct evlist *evlist __maybe_unused)
 544 {
 545         char pbuf[MAXPATHLEN], *buf;
 546         int i, ret, n;
 547
 548         /* actual path to perf binary */
 549         buf = perf_exe(pbuf, MAXPATHLEN);
 550
 551         /* account for binary path */
 552         n = perf_env.nr_cmdline + 1;
 553
 554         ret = do_write(ff, &n, sizeof(n));
 555         if (ret < 0)
 556                 return ret;
 557
 558         ret = do_write_string(ff, buf);
 559         if (ret < 0)
 560                 return ret;
 561
 562         for (i = 0 ; i < perf_env.nr_cmdline; i++) {
 563                 ret = do_write_string(ff, perf_env.cmdline_argv[i]);
 564                 if (ret < 0)
 565                         return ret;
 566         }
 567         return 0;
 568 }
 569
 570
 571 static int write_cpu_topology(struct feat_fd *ff,
 572                               struct evlist *evlist __maybe_unused)
 573 {
 574         struct cpu_topology *tp;
 575         u32 i;
 576         int ret, j;
 577
 578         tp = cpu_topology__new();
 579         if (!tp)
 580                 return -1;
 581
 582         ret = do_write(ff, &tp->core_sib, sizeof(tp->core_sib));
 583         if (ret < 0)
 584                 goto done;
 585
 586         for (i = 0; i < tp->core_sib; i++) {
 587                 ret = do_write_string(ff, tp->core_siblings[i]);
 588                 if (ret < 0)
 589                         goto done;
 590         }
 591         ret = do_write(ff, &tp->thread_sib, sizeof(tp->thread_sib));
 592         if (ret < 0)
 593                 goto done;
 594
 595         for (i = 0; i < tp->thread_sib; i++) {
 596                 ret = do_write_string(ff, tp->thread_siblings[i]);
 597                 if (ret < 0)
 598                         break;
 599         }
 600
 601         ret = perf_env__read_cpu_topology_map(&perf_env);
 602         if (ret < 0)
 603                 goto done;
 604
 605         for (j = 0; j < perf_env.nr_cpus_avail; j++) {
 606                 ret = do_write(ff, &perf_env.cpu[j].core_id,
 607                                sizeof(perf_env.cpu[j].core_id));
 608                 if (ret < 0)
 609                         return ret;
 610                 ret = do_write(ff, &perf_env.cpu[j].socket_id,
 611                                sizeof(perf_env.cpu[j].socket_id));
 612                 if (ret < 0)
 613                         return ret;
 614         }
 615
 616         if (!tp->die_sib)
 617                 goto done;
 618
 619         ret = do_write(ff, &tp->die_sib, sizeof(tp->die_sib));
 620         if (ret < 0)
 621                 goto done;
 622
 623         for (i = 0; i < tp->die_sib; i++) {
 624                 ret = do_write_string(ff, tp->die_siblings[i]);
 625                 if (ret < 0)
 626                         goto done;
 627         }
 628
 629         for (j = 0; j < perf_env.nr_cpus_avail; j++) {
 630                 ret = do_write(ff, &perf_env.cpu[j].die_id,
 631                                sizeof(perf_env.cpu[j].die_id));
 632                 if (ret < 0)
 633                         return ret;
 634         }
 635
 636 done:
 637         cpu_topology__delete(tp);
 638         return ret;
 639 }
 640
 641
 642
 643 static int write_total_mem(struct feat_fd *ff,
 644                            struct evlist *evlist __maybe_unused)
 645 {
 646         char *buf = NULL;
 647         FILE *fp;
 648         size_t len = 0;
 649         int ret = -1, n;
 650         uint64_t mem;
 651
 652         fp = fopen("/proc/meminfo", "r");
 653         if (!fp)
 654                 return -1;
 655
 656         while (getline(&buf, &len, fp) > 0) {
 657                 ret = strncmp(buf, "MemTotal:", 9);
 658                 if (!ret)
 659                         break;
 660         }
 661         if (!ret) {
 662                 n = sscanf(buf, "%*s %"PRIu64, &mem);
 663                 if (n == 1)
 664                         ret = do_write(ff, &mem, sizeof(mem));
 665         } else
 666                 ret = -1;
 667         free(buf);
 668         fclose(fp);
 669         return ret;
 670 }
 671
 672 static int write_numa_topology(struct feat_fd *ff,
 673                                struct evlist *evlist __maybe_unused)
 674 {
 675         struct numa_topology *tp;
 676         int ret = -1;
 677         u32 i;
 678
 679         tp = numa_topology__new();
 680         if (!tp)
 681                 return -ENOMEM;
 682
 683         ret = do_write(ff, &tp->nr, sizeof(u32));
 684         if (ret < 0)
 685                 goto err;
 686
 687         for (i = 0; i < tp->nr; i++) {
 688                 struct numa_topology_node *n = &tp->nodes[i];
 689
 690                 ret = do_write(ff, &n->node, sizeof(u32));
 691                 if (ret < 0)
 692                         goto err;
 693
 694                 ret = do_write(ff, &n->mem_total, sizeof(u64));
 695                 if (ret)
 696                         goto err;
 697
 698                 ret = do_write(ff, &n->mem_free, sizeof(u64));
 699                 if (ret)
 700                         goto err;
 701
 702                 ret = do_write_string(ff, n->cpus);
 703                 if (ret < 0)
 704                         goto err;
 705         }
 706
 707         ret = 0;
 708
 709 err:
 710         numa_topology__delete(tp);
 711         return ret;
 712 }
 713
 714 /*
 715  * File format:
 716  *
 717  * struct pmu_mappings {
 718  *      u32     pmu_num;
 719  *      struct pmu_map {
 720  *              u32     type;
 721  *              char    name[];
 722  *      }[pmu_num];
 723  * };
 724  */
 725
 726 static int write_pmu_mappings(struct feat_fd *ff,
 727                               struct evlist *evlist __maybe_unused)
 728 {
 729         struct perf_pmu *pmu = NULL;
 730         u32 pmu_num = 0;
 731         int ret;
 732
 733         /*
 734          * Do a first pass to count number of pmu to avoid lseek so this
 735          * works in pipe mode as well.
 736          */
 737         while ((pmu = perf_pmu__scan(pmu))) {
 738                 if (!pmu->name)
 739                         continue;
 740                 pmu_num++;
 741         }
 742
 743         ret = do_write(ff, &pmu_num, sizeof(pmu_num));
 744         if (ret < 0)
 745                 return ret;
 746
 747         while ((pmu = perf_pmu__scan(pmu))) {
 748                 if (!pmu->name)
 749                         continue;
 750
 751                 ret = do_write(ff, &pmu->type, sizeof(pmu->type));
 752                 if (ret < 0)
 753                         return ret;
 754
 755                 ret = do_write_string(ff, pmu->name);
 756                 if (ret < 0)
 757                         return ret;
 758         }
 759
 760         return 0;
 761 }
 762
 763 /*
 764  * File format:
 765  *
 766  * struct group_descs {
 767  *      u32     nr_groups;
 768  *      struct group_desc {
 769  *              char    name[];
 770  *              u32     leader_idx;
 771  *              u32     nr_members;
 772  *      }[nr_groups];
 773  * };
 774  */
 775 static int write_group_desc(struct feat_fd *ff,
 776                             struct evlist *evlist)
 777 {
 778         u32 nr_groups = evlist->nr_groups;
 779         struct evsel *evsel;
 780         int ret;
 781
 782         ret = do_write(ff, &nr_groups, sizeof(nr_groups));
 783         if (ret < 0)
 784                 return ret;
 785
 786         evlist__for_each_entry(evlist, evsel) {
 787                 if (evsel__is_group_leader(evsel) && evsel->core.nr_members > 1) {
 788                         const char *name = evsel->group_name ?: "{anon_group}";
 789                         u32 leader_idx = evsel->idx;
 790                         u32 nr_members = evsel->core.nr_members;
 791
 792                         ret = do_write_string(ff, name);
 793                         if (ret < 0)
 794                                 return ret;
 795
 796                         ret = do_write(ff, &leader_idx, sizeof(leader_idx));
 797                         if (ret < 0)
 798                                 return ret;
 799
 800                         ret = do_write(ff, &nr_members, sizeof(nr_members));
 801                         if (ret < 0)
 802                                 return ret;
 803                 }
 804         }
 805         return 0;
 806 }
 807
 808 /*
 809  * Return the CPU id as a raw string.
 810  *
 811  * Each architecture should provide a more precise id string that
 812  * can be use to match the architecture's "mapfile".
 813  */
 814 char * __weak get_cpuid_str(struct perf_pmu *pmu __maybe_unused)
 815 {
 816         return NULL;
 817 }
 818
 819 /* Return zero when the cpuid from the mapfile.csv matches the
 820  * cpuid string generated on this platform.
 821  * Otherwise return non-zero.
 822  */
 823 int __weak strcmp_cpuid_str(const char *mapcpuid, const char *cpuid)
 824 {
 825         regex_t re;
 826         regmatch_t pmatch[1];
 827         int match;
 828
 829         if (regcomp(&re, mapcpuid, REG_EXTENDED) != 0) {
 830                 /* Warn unable to generate match particular string. */
 831                 pr_info("Invalid regular expression %s\n", mapcpuid);
 832                 return 1;
 833         }
 834
 835         match = !regexec(&re, cpuid, 1, pmatch, 0);
 836         regfree(&re);
 837         if (match) {
 838                 size_t match_len = (pmatch[0].rm_eo - pmatch[0].rm_so);
 839
 840                 /* Verify the entire string matched. */
 841                 if (match_len == strlen(cpuid))
 842                         return 0;
 843         }
 844         return 1;
 845 }
 846
 847 /*
 848  * default get_cpuid(): nothing gets recorded
 849  * actual implementation must be in arch/$(SRCARCH)/util/header.c
 850  */
 851 int __weak get_cpuid(char *buffer __maybe_unused, size_t sz __maybe_unused)
 852 {
 853         return ENOSYS; /* Not implemented */
 854 }
 855
 856 static int write_cpuid(struct feat_fd *ff,
 857                        struct evlist *evlist __maybe_unused)
 858 {
 859         char buffer[64];
 860         int ret;
 861
 862         ret = get_cpuid(buffer, sizeof(buffer));
 863         if (ret)
 864                 return -1;
 865
 866         return do_write_string(ff, buffer);
 867 }
 868
 869 static int write_branch_stack(struct feat_fd *ff __maybe_unused,
 870                               struct evlist *evlist __maybe_unused)
 871 {
 872         return 0;
 873 }
 874
 875 static int write_auxtrace(struct feat_fd *ff,
 876                           struct evlist *evlist __maybe_unused)
 877 {
 878         struct perf_session *session;
 879         int err;
 880
 881         if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__))
 882                 return -1;
 883
 884         session = container_of(ff->ph, struct perf_session, header);
 885
 886         err = auxtrace_index__write(ff->fd, &session->auxtrace_index);
 887         if (err < 0)
 888                 pr_err("Failed to write auxtrace index\n");
 889         return err;
 890 }
 891
 892 static int write_clockid(struct feat_fd *ff,
 893                          struct evlist *evlist __maybe_unused)
 894 {
 895         return do_write(ff, &ff->ph->env.clock.clockid_res_ns,
 896                         sizeof(ff->ph->env.clock.clockid_res_ns));
 897 }
 898
 899 static int write_clock_data(struct feat_fd *ff,
 900                             struct evlist *evlist __maybe_unused)
 901 {
 902         u64 *data64;
 903         u32 data32;
 904         int ret;
 905
 906         /* version */
 907         data32 = 1;
 908
 909         ret = do_write(ff, &data32, sizeof(data32));
 910         if (ret < 0)
 911                 return ret;
 912
 913         /* clockid */
 914         data32 = ff->ph->env.clock.clockid;
 915
 916         ret = do_write(ff, &data32, sizeof(data32));
 917         if (ret < 0)
 918                 return ret;
 919
 920         /* TOD ref time */
 921         data64 = &ff->ph->env.clock.tod_ns;
 922
 923         ret = do_write(ff, data64, sizeof(*data64));
 924         if (ret < 0)
 925                 return ret;
 926
 927         /* clockid ref time */
 928         data64 = &ff->ph->env.clock.clockid_ns;
 929
 930         return do_write(ff, data64, sizeof(*data64));
 931 }
 932
 933 static int write_dir_format(struct feat_fd *ff,
 934                             struct evlist *evlist __maybe_unused)
 935 {
 936         struct perf_session *session;
 937         struct perf_data *data;
 938
 939         session = container_of(ff->ph, struct perf_session, header);
 940         data = session->data;
 941
 942         if (WARN_ON(!perf_data__is_dir(data)))
 943                 return -1;
 944
 945         return do_write(ff, &data->dir.version, sizeof(data->dir.version));
 946 }
 947
 948 #ifdef HAVE_LIBBPF_SUPPORT
 949 static int write_bpf_prog_info(struct feat_fd *ff,
 950                                struct evlist *evlist __maybe_unused)
 951 {
 952         struct perf_env *env = &ff->ph->env;
 953         struct rb_root *root;
 954         struct rb_node *next;
 955         int ret;
 956
 957         down_read(&env->bpf_progs.lock);
 958
 959         ret = do_write(ff, &env->bpf_progs.infos_cnt,
 960                        sizeof(env->bpf_progs.infos_cnt));
 961         if (ret < 0)
 962                 goto out;
 963
 964         root = &env->bpf_progs.infos;
 965         next = rb_first(root);
 966         while (next) {
 967                 struct bpf_prog_info_node *node;
 968                 size_t len;
 969
 970                 node = rb_entry(next, struct bpf_prog_info_node, rb_node);
 971                 next = rb_next(&node->rb_node);
 972                 len = sizeof(struct bpf_prog_info_linear) +
 973                         node->info_linear->data_len;
 974
 975                 /* before writing to file, translate address to offset */
 976                 bpf_program__bpil_addr_to_offs(node->info_linear);
 977                 ret = do_write(ff, node->info_linear, len);
 978                 /*
 979                  * translate back to address even when do_write() fails,
 980                  * so that this function never changes the data.
 981                  */
 982                 bpf_program__bpil_offs_to_addr(node->info_linear);
 983                 if (ret < 0)
 984                         goto out;
 985         }
 986 out:
 987         up_read(&env->bpf_progs.lock);
 988         return ret;
 989 }
 990 #else // HAVE_LIBBPF_SUPPORT
 991 static int write_bpf_prog_info(struct feat_fd *ff __maybe_unused,
 992                                struct evlist *evlist __maybe_unused)
 993 {
 994         return 0;
 995 }
 996 #endif // HAVE_LIBBPF_SUPPORT
 997
 998 static int write_bpf_btf(struct feat_fd *ff,
 999                          struct evlist *evlist __maybe_unused)
1000 {
1001         struct perf_env *env = &ff->ph->env;
1002         struct rb_root *root;
1003         struct rb_node *next;
1004         int ret;
1005
1006         down_read(&env->bpf_progs.lock);
1007
1008         ret = do_write(ff, &env->bpf_progs.btfs_cnt,
1009                        sizeof(env->bpf_progs.btfs_cnt));
1010
1011         if (ret < 0)
1012                 goto out;
1013
1014         root = &env->bpf_progs.btfs;
1015         next = rb_first(root);
1016         while (next) {
1017                 struct btf_node *node;
1018
1019                 node = rb_entry(next, struct btf_node, rb_node);
1020                 next = rb_next(&node->rb_node);
1021                 ret = do_write(ff, &node->id,
1022                                sizeof(u32) * 2 + node->data_size);
1023                 if (ret < 0)
1024                         goto out;
1025         }
1026 out:
1027         up_read(&env->bpf_progs.lock);
1028         return ret;
1029 }
1030
1031 static int cpu_cache_level__sort(const void *a, const void *b)
1032 {
1033         struct cpu_cache_level *cache_a = (struct cpu_cache_level *)a;
1034         struct cpu_cache_level *cache_b = (struct cpu_cache_level *)b;
1035
1036         return cache_a->level - cache_b->level;
1037 }
1038
1039 static bool cpu_cache_level__cmp(struct cpu_cache_level *a, struct cpu_cache_level *b)
1040 {
1041         if (a->level != b->level)
1042                 return false;
1043
1044         if (a->line_size != b->line_size)
1045                 return false;
1046
1047         if (a->sets != b->sets)
1048                 return false;
1049
1050         if (a->ways != b->ways)
1051                 return false;
1052
1053         if (strcmp(a->type, b->type))
1054                 return false;
1055
1056         if (strcmp(a->size, b->size))
1057                 return false;
1058
1059         if (strcmp(a->map, b->map))
1060                 return false;
1061
1062         return true;
1063 }
1064
1065 static int cpu_cache_level__read(struct cpu_cache_level *cache, u32 cpu, u16 level)
1066 {
1067         char path[PATH_MAX], file[PATH_MAX];
1068         struct stat st;
1069         size_t len;
1070
1071         scnprintf(path, PATH_MAX, "devices/system/cpu/cpu%d/cache/index%d/", cpu, level);
1072         scnprintf(file, PATH_MAX, "%s/%s", sysfs__mountpoint(), path);
1073
1074         if (stat(file, &st))
1075                 return 1;
1076
1077         scnprintf(file, PATH_MAX, "%s/level", path);
1078         if (sysfs__read_int(file, (int *) &cache->level))
1079                 return -1;
1080
1081         scnprintf(file, PATH_MAX, "%s/coherency_line_size", path);
1082         if (sysfs__read_int(file, (int *) &cache->line_size))
1083                 return -1;
1084
1085         scnprintf(file, PATH_MAX, "%s/number_of_sets", path);
1086         if (sysfs__read_int(file, (int *) &cache->sets))
1087                 return -1;
1088
1089         scnprintf(file, PATH_MAX, "%s/ways_of_associativity", path);
1090         if (sysfs__read_int(file, (int *) &cache->ways))
1091                 return -1;
1092
1093         scnprintf(file, PATH_MAX, "%s/type", path);
1094         if (sysfs__read_str(file, &cache->type, &len))
1095                 return -1;
1096
1097         cache->type[len] = 0;
1098         cache->type = strim(cache->type);
1099
1100         scnprintf(file, PATH_MAX, "%s/size", path);
1101         if (sysfs__read_str(file, &cache->size, &len)) {
1102                 zfree(&cache->type);
1103                 return -1;
1104         }
1105
1106         cache->size[len] = 0;
1107         cache->size = strim(cache->size);
1108
1109         scnprintf(file, PATH_MAX, "%s/shared_cpu_list", path);
1110         if (sysfs__read_str(file, &cache->map, &len)) {
1111                 zfree(&cache->size);
1112                 zfree(&cache->type);
1113                 return -1;
1114         }
1115
1116         cache->map[len] = 0;
1117         cache->map = strim(cache->map);
1118         return 0;
1119 }
1120
1121 static void cpu_cache_level__fprintf(FILE *out, struct cpu_cache_level *c)
1122 {
1123         fprintf(out, "L%d %-15s %8s [%s]\n", c->level, c->type, c->size, c->map);
1124 }
1125
1126 #define MAX_CACHE_LVL 4
1127
1128 static int build_caches(struct cpu_cache_level caches[], u32 *cntp)
1129 {
1130         u32 i, cnt = 0;
1131         u32 nr, cpu;
1132         u16 level;
1133
1134         nr = cpu__max_cpu();
1135
1136         for (cpu = 0; cpu < nr; cpu++) {
1137                 for (level = 0; level < MAX_CACHE_LVL; level++) {
1138                         struct cpu_cache_level c;
1139                         int err;
1140
1141                         err = cpu_cache_level__read(&c, cpu, level);
1142                         if (err < 0)
1143                                 return err;
1144
1145                         if (err == 1)
1146                                 break;
1147
1148                         for (i = 0; i < cnt; i++) {
1149                                 if (cpu_cache_level__cmp(&c, &caches[i]))
1150                                         break;
1151                         }
1152
1153                         if (i == cnt)
1154                                 caches[cnt++] = c;
1155                         else
1156                                 cpu_cache_level__free(&c);
1157                 }
1158         }
1159         *cntp = cnt;
1160         return 0;
1161 }
1162
1163 static int write_cache(struct feat_fd *ff,
1164                        struct evlist *evlist __maybe_unused)
1165 {
1166         u32 max_caches = cpu__max_cpu() * MAX_CACHE_LVL;
1167         struct cpu_cache_level caches[max_caches];
1168         u32 cnt = 0, i, version = 1;
1169         int ret;
1170
1171         ret = build_caches(caches, &cnt);
1172         if (ret)
1173                 goto out;
1174
1175         qsort(&caches, cnt, sizeof(struct cpu_cache_level), cpu_cache_level__sort);
1176
1177         ret = do_write(ff, &version, sizeof(u32));
1178         if (ret < 0)
1179                 goto out;
1180
1181         ret = do_write(ff, &cnt, sizeof(u32));
1182         if (ret < 0)
1183                 goto out;
1184
1185         for (i = 0; i < cnt; i++) {
1186                 struct cpu_cache_level *c = &caches[i];
1187
1188                 #define _W(v)                                   \
1189                         ret = do_write(ff, &c->v, sizeof(u32)); \
1190                         if (ret < 0)                            \
1191                                 goto out;
1192
1193                 _W(level)
1194                 _W(line_size)
1195                 _W(sets)
1196                 _W(ways)
1197                 #undef _W
1198
1199                 #define _W(v)                                           \
1200                         ret = do_write_string(ff, (const char *) c->v); \
1201                         if (ret < 0)                                    \
1202                                 goto out;
1203
1204                 _W(type)
1205                 _W(size)
1206                 _W(map)
1207                 #undef _W
1208         }
1209
1210 out:
1211         for (i = 0; i < cnt; i++)
1212                 cpu_cache_level__free(&caches[i]);
1213         return ret;
1214 }
1215
1216 static int write_stat(struct feat_fd *ff __maybe_unused,
1217                       struct evlist *evlist __maybe_unused)
1218 {
1219         return 0;
1220 }
1221
1222 static int write_sample_time(struct feat_fd *ff,
1223                              struct evlist *evlist)
1224 {
1225         int ret;
1226
1227         ret = do_write(ff, &evlist->first_sample_time,
1228                        sizeof(evlist->first_sample_time));
1229         if (ret < 0)
1230                 return ret;
1231
1232         return do_write(ff, &evlist->last_sample_time,
1233                         sizeof(evlist->last_sample_time));
1234 }
1235
1236
1237 static int memory_node__read(struct memory_node *n, unsigned long idx)
1238 {
1239         unsigned int phys, size = 0;
1240         char path[PATH_MAX];
1241         struct dirent *ent;
1242         DIR *dir;
1243
1244 #define for_each_memory(mem, dir)                                       \
1245         while ((ent = readdir(dir)))                                    \
1246                 if (strcmp(ent->d_name, ".") &&                         \
1247                     strcmp(ent->d_name, "..") &&                        \
1248                     sscanf(ent->d_name, "memory%u", &mem) == 1)
1249
1250         scnprintf(path, PATH_MAX,
1251                   "%s/devices/system/node/node%lu",
1252                   sysfs__mountpoint(), idx);
1253
1254         dir = opendir(path);
1255         if (!dir) {
1256                 pr_warning("failed: cant' open memory sysfs data\n");
1257                 return -1;
1258         }
1259
1260         for_each_memory(phys, dir) {
1261                 size = max(phys, size);
1262         }
1263
1264         size++;
1265
1266         n->set = bitmap_alloc(size);
1267         if (!n->set) {
1268                 closedir(dir);
1269                 return -ENOMEM;
1270         }
1271
1272         n->node = idx;
1273         n->size = size;
1274
1275         rewinddir(dir);
1276
1277         for_each_memory(phys, dir) {
1278                 set_bit(phys, n->set);
1279         }
1280
1281         closedir(dir);
1282         return 0;
1283 }
1284
1285 static int memory_node__sort(const void *a, const void *b)
1286 {
1287         const struct memory_node *na = a;
1288         const struct memory_node *nb = b;
1289
1290         return na->node - nb->node;
1291 }
1292
1293 static int build_mem_topology(struct memory_node *nodes, u64 size, u64 *cntp)
1294 {
1295         char path[PATH_MAX];
1296         struct dirent *ent;
1297         DIR *dir;
1298         u64 cnt = 0;
1299         int ret = 0;
1300
1301         scnprintf(path, PATH_MAX, "%s/devices/system/node/",
1302                   sysfs__mountpoint());
1303
1304         dir = opendir(path);
1305         if (!dir) {
1306                 pr_debug2("%s: could't read %s, does this arch have topology information?\n",
1307                           __func__, path);
1308                 return -1;
1309         }
1310
1311         while (!ret && (ent = readdir(dir))) {
1312                 unsigned int idx;
1313                 int r;
1314
1315                 if (!strcmp(ent->d_name, ".") ||
1316                     !strcmp(ent->d_name, ".."))
1317                         continue;
1318
1319                 r = sscanf(ent->d_name, "node%u", &idx);
1320                 if (r != 1)
1321                         continue;
1322
1323                 if (WARN_ONCE(cnt >= size,
1324                         "failed to write MEM_TOPOLOGY, way too many nodes\n")) {
1325                         closedir(dir);
1326                         return -1;
1327                 }
1328
1329                 ret = memory_node__read(&nodes[cnt++], idx);
1330         }
1331
1332         *cntp = cnt;
1333         closedir(dir);
1334
1335         if (!ret)
1336                 qsort(nodes, cnt, sizeof(nodes[0]), memory_node__sort);
1337
1338         return ret;
1339 }
1340
1341 #define MAX_MEMORY_NODES 2000
1342
1343 /*
1344  * The MEM_TOPOLOGY holds physical memory map for every
1345  * node in system. The format of data is as follows:
1346  *
1347  *  0 - version          | for future changes
1348  *  8 - block_size_bytes | /sys/devices/system/memory/block_size_bytes
1349  * 16 - count            | number of nodes
1350  *
1351  * For each node we store map of physical indexes for
1352  * each node:
1353  *
1354  * 32 - node id          | node index
1355  * 40 - size             | size of bitmap
1356  * 48 - bitmap           | bitmap of memory indexes that belongs to node
1357  */
1358 static int write_mem_topology(struct feat_fd *ff __maybe_unused,
1359                               struct evlist *evlist __maybe_unused)
1360 {
1361         static struct memory_node nodes[MAX_MEMORY_NODES];
1362         u64 bsize, version = 1, i, nr;
1363         int ret;
1364
1365         ret = sysfs__read_xll("devices/system/memory/block_size_bytes",
1366                               (unsigned long long *) &bsize);
1367         if (ret)
1368                 return ret;
1369
1370         ret = build_mem_topology(&nodes[0], MAX_MEMORY_NODES, &nr);
1371         if (ret)
1372                 return ret;
1373
1374         ret = do_write(ff, &version, sizeof(version));
1375         if (ret < 0)
1376                 goto out;
1377
1378         ret = do_write(ff, &bsize, sizeof(bsize));
1379         if (ret < 0)
1380                 goto out;
1381
1382         ret = do_write(ff, &nr, sizeof(nr));
1383         if (ret < 0)
1384                 goto out;
1385
1386         for (i = 0; i < nr; i++) {
1387                 struct memory_node *n = &nodes[i];
1388
1389                 #define _W(v)                                           \
1390                         ret = do_write(ff, &n->v, sizeof(n->v));        \
1391                         if (ret < 0)                                    \
1392                                 goto out;
1393
1394                 _W(node)
1395                 _W(size)
1396
1397                 #undef _W
1398
1399                 ret = do_write_bitmap(ff, n->set, n->size);
1400                 if (ret < 0)
1401                         goto out;
1402         }
1403
1404 out:
1405         return ret;
1406 }
1407
1408 static int write_compressed(struct feat_fd *ff __maybe_unused,
1409                             struct evlist *evlist __maybe_unused)
1410 {
1411         int ret;
1412
1413         ret = do_write(ff, &(ff->ph->env.comp_ver), sizeof(ff->ph->env.comp_ver));
1414         if (ret)
1415                 return ret;
1416
1417         ret = do_write(ff, &(ff->ph->env.comp_type), sizeof(ff->ph->env.comp_type));
1418         if (ret)
1419                 return ret;
1420
1421         ret = do_write(ff, &(ff->ph->env.comp_level), sizeof(ff->ph->env.comp_level));
1422         if (ret)
1423                 return ret;
1424
1425         ret = do_write(ff, &(ff->ph->env.comp_ratio), sizeof(ff->ph->env.comp_ratio));
1426         if (ret)
1427                 return ret;
1428
1429         return do_write(ff, &(ff->ph->env.comp_mmap_len), sizeof(ff->ph->env.comp_mmap_len));
1430 }
1431
1432 static int write_cpu_pmu_caps(struct feat_fd *ff,
1433                               struct evlist *evlist __maybe_unused)
1434 {
1435         struct perf_pmu *cpu_pmu = perf_pmu__find("cpu");
1436         struct perf_pmu_caps *caps = NULL;
1437         int nr_caps;
1438         int ret;
1439
1440         if (!cpu_pmu)
1441                 return -ENOENT;
1442
1443         nr_caps = perf_pmu__caps_parse(cpu_pmu);
1444         if (nr_caps < 0)
1445                 return nr_caps;
1446
1447         ret = do_write(ff, &nr_caps, sizeof(nr_caps));
1448         if (ret < 0)
1449                 return ret;
1450
1451         list_for_each_entry(caps, &cpu_pmu->caps, list) {
1452                 ret = do_write_string(ff, caps->name);
1453                 if (ret < 0)
1454                         return ret;
1455
1456                 ret = do_write_string(ff, caps->value);
1457                 if (ret < 0)
1458                         return ret;
1459         }
1460
1461         return ret;
1462 }
1463
1464 static void print_hostname(struct feat_fd *ff, FILE *fp)
1465 {
1466         fprintf(fp, "# hostname : %s\n", ff->ph->env.hostname);
1467 }
1468
1469 static void print_osrelease(struct feat_fd *ff, FILE *fp)
1470 {
1471         fprintf(fp, "# os release : %s\n", ff->ph->env.os_release);
1472 }
1473
1474 static void print_arch(struct feat_fd *ff, FILE *fp)
1475 {
1476         fprintf(fp, "# arch : %s\n", ff->ph->env.arch);
1477 }
1478
1479 static void print_cpudesc(struct feat_fd *ff, FILE *fp)
1480 {
1481         fprintf(fp, "# cpudesc : %s\n", ff->ph->env.cpu_desc);
1482 }
1483
1484 static void print_nrcpus(struct feat_fd *ff, FILE *fp)
1485 {
1486         fprintf(fp, "# nrcpus online : %u\n", ff->ph->env.nr_cpus_online);
1487         fprintf(fp, "# nrcpus avail : %u\n", ff->ph->env.nr_cpus_avail);
1488 }
1489
1490 static void print_version(struct feat_fd *ff, FILE *fp)
1491 {
1492         fprintf(fp, "# perf version : %s\n", ff->ph->env.version);
1493 }
1494
1495 static void print_cmdline(struct feat_fd *ff, FILE *fp)
1496 {
1497         int nr, i;
1498
1499         nr = ff->ph->env.nr_cmdline;
1500
1501         fprintf(fp, "# cmdline : ");
1502
1503         for (i = 0; i < nr; i++) {
1504                 char *argv_i = strdup(ff->ph->env.cmdline_argv[i]);
1505                 if (!argv_i) {
1506                         fprintf(fp, "%s ", ff->ph->env.cmdline_argv[i]);
1507                 } else {
1508                         char *mem = argv_i;
1509                         do {
1510                                 char *quote = strchr(argv_i, '\'');
1511                                 if (!quote)
1512                                         break;
1513                                 *quote++ = '\0';
1514                                 fprintf(fp, "%s\\\'", argv_i);
1515                                 argv_i = quote;
1516                         } while (1);
1517                         fprintf(fp, "%s ", argv_i);
1518                         free(mem);
1519                 }
1520         }
1521         fputc('\n', fp);
1522 }
1523
1524 static void print_cpu_topology(struct feat_fd *ff, FILE *fp)
1525 {
1526         struct perf_header *ph = ff->ph;
1527         int cpu_nr = ph->env.nr_cpus_avail;
1528         int nr, i;
1529         char *str;
1530
1531         nr = ph->env.nr_sibling_cores;
1532         str = ph->env.sibling_cores;
1533
1534         for (i = 0; i < nr; i++) {
1535                 fprintf(fp, "# sibling sockets : %s\n", str);
1536                 str += strlen(str) + 1;
1537         }
1538
1539         if (ph->env.nr_sibling_dies) {
1540                 nr = ph->env.nr_sibling_dies;
1541                 str = ph->env.sibling_dies;
1542
1543                 for (i = 0; i < nr; i++) {
1544                         fprintf(fp, "# sibling dies    : %s\n", str);
1545                         str += strlen(str) + 1;
1546                 }
1547         }
1548
1549         nr = ph->env.nr_sibling_threads;
1550         str = ph->env.sibling_threads;
1551
1552         for (i = 0; i < nr; i++) {
1553                 fprintf(fp, "# sibling threads : %s\n", str);
1554                 str += strlen(str) + 1;
1555         }
1556
1557         if (ph->env.nr_sibling_dies) {
1558                 if (ph->env.cpu != NULL) {
1559                         for (i = 0; i < cpu_nr; i++)
1560                                 fprintf(fp, "# CPU %d: Core ID %d, "
1561                                             "Die ID %d, Socket ID %d\n",
1562                                             i, ph->env.cpu[i].core_id,
1563                                             ph->env.cpu[i].die_id,
1564                                             ph->env.cpu[i].socket_id);
1565                 } else
1566                         fprintf(fp, "# Core ID, Die ID and Socket ID "
1567                                     "information is not available\n");
1568         } else {
1569                 if (ph->env.cpu != NULL) {
1570                         for (i = 0; i < cpu_nr; i++)
1571                                 fprintf(fp, "# CPU %d: Core ID %d, "
1572                                             "Socket ID %d\n",
1573                                             i, ph->env.cpu[i].core_id,
1574                                             ph->env.cpu[i].socket_id);
1575                 } else
1576                         fprintf(fp, "# Core ID and Socket ID "
1577                                     "information is not available\n");
1578         }
1579 }
1580
1581 static void print_clockid(struct feat_fd *ff, FILE *fp)
1582 {
1583         fprintf(fp, "# clockid frequency: %"PRIu64" MHz\n",
1584                 ff->ph->env.clock.clockid_res_ns * 1000);
1585 }
1586
1587 static void print_clock_data(struct feat_fd *ff, FILE *fp)
1588 {
1589         struct timespec clockid_ns;
1590         char tstr[64], date[64];
1591         struct timeval tod_ns;
1592         clockid_t clockid;
1593         struct tm ltime;
1594         u64 ref;
1595
1596         if (!ff->ph->env.clock.enabled) {
1597                 fprintf(fp, "# reference time disabled\n");
1598                 return;
1599         }
1600
1601         /* Compute TOD time. */
1602         ref = ff->ph->env.clock.tod_ns;
1603         tod_ns.tv_sec = ref / NSEC_PER_SEC;
1604         ref -= tod_ns.tv_sec * NSEC_PER_SEC;
1605         tod_ns.tv_usec = ref / NSEC_PER_USEC;
1606
1607         /* Compute clockid time. */
1608         ref = ff->ph->env.clock.clockid_ns;
1609         clockid_ns.tv_sec = ref / NSEC_PER_SEC;
1610         ref -= clockid_ns.tv_sec * NSEC_PER_SEC;
1611         clockid_ns.tv_nsec = ref;
1612
1613         clockid = ff->ph->env.clock.clockid;
1614
1615         if (localtime_r(&tod_ns.tv_sec, &ltime) == NULL)
1616                 snprintf(tstr, sizeof(tstr), "<error>");
1617         else {
1618                 strftime(date, sizeof(date), "%F %T", &ltime);
1619                 scnprintf(tstr, sizeof(tstr), "%s.%06d",
1620                           date, (int) tod_ns.tv_usec);
1621         }
1622
1623         fprintf(fp, "# clockid: %s (%u)\n", clockid_name(clockid), clockid);
1624         fprintf(fp, "# reference time: %s = %ld.%06d (TOD) = %ld.%09ld (%s)\n",
1625                     tstr, tod_ns.tv_sec, (int) tod_ns.tv_usec,
1626                     clockid_ns.tv_sec, clockid_ns.tv_nsec,
1627                     clockid_name(clockid));
1628 }
1629
1630 static void print_dir_format(struct feat_fd *ff, FILE *fp)
1631 {
1632         struct perf_session *session;
1633         struct perf_data *data;
1634
1635         session = container_of(ff->ph, struct perf_session, header);
1636         data = session->data;
1637
1638         fprintf(fp, "# directory data version : %"PRIu64"\n", data->dir.version);
1639 }
1640
1641 static void print_bpf_prog_info(struct feat_fd *ff, FILE *fp)
1642 {
1643         struct perf_env *env = &ff->ph->env;
1644         struct rb_root *root;
1645         struct rb_node *next;
1646
1647         down_read(&env->bpf_progs.lock);
1648
1649         root = &env->bpf_progs.infos;
1650         next = rb_first(root);
1651
1652         while (next) {
1653                 struct bpf_prog_info_node *node;
1654
1655                 node = rb_entry(next, struct bpf_prog_info_node, rb_node);
1656                 next = rb_next(&node->rb_node);
1657
1658                 __bpf_event__print_bpf_prog_info(&node->info_linear->info,
1659                                                  env, fp);
1660         }
1661
1662         up_read(&env->bpf_progs.lock);
1663 }
1664
1665 static void print_bpf_btf(struct feat_fd *ff, FILE *fp)
1666 {
1667         struct perf_env *env = &ff->ph->env;
1668         struct rb_root *root;
1669         struct rb_node *next;
1670
1671         down_read(&env->bpf_progs.lock);
1672
1673         root = &env->bpf_progs.btfs;
1674         next = rb_first(root);
1675
1676         while (next) {
1677                 struct btf_node *node;
1678
1679                 node = rb_entry(next, struct btf_node, rb_node);
1680                 next = rb_next(&node->rb_node);
1681                 fprintf(fp, "# btf info of id %u\n", node->id);
1682         }
1683
1684         up_read(&env->bpf_progs.lock);
1685 }
1686
1687 static void free_event_desc(struct evsel *events)
1688 {
1689         struct evsel *evsel;
1690
1691         if (!events)
1692                 return;
1693
1694         for (evsel = events; evsel->core.attr.size; evsel++) {
1695                 zfree(&evsel->name);
1696                 zfree(&evsel->core.id);
1697         }
1698
1699         free(events);
1700 }
1701
1702 static bool perf_attr_check(struct perf_event_attr *attr)
1703 {
1704         if (attr->__reserved_1 || attr->__reserved_2 || attr->__reserved_3) {
1705                 pr_warning("Reserved bits are set unexpectedly. "
1706                            "Please update perf tool.\n");
1707                 return false;
1708         }
1709
1710         if (attr->sample_type & ~(PERF_SAMPLE_MAX-1)) {
1711                 pr_warning("Unknown sample type (0x%llx) is detected. "
1712                            "Please update perf tool.\n",
1713                            attr->sample_type);
1714                 return false;
1715         }
1716
1717         if (attr->read_format & ~(PERF_FORMAT_MAX-1)) {
1718                 pr_warning("Unknown read format (0x%llx) is detected. "
1719                            "Please update perf tool.\n",
1720                            attr->read_format);
1721                 return false;
1722         }
1723
1724         if ((attr->sample_type & PERF_SAMPLE_BRANCH_STACK) &&
1725             (attr->branch_sample_type & ~(PERF_SAMPLE_BRANCH_MAX-1))) {
1726                 pr_warning("Unknown branch sample type (0x%llx) is detected. "
1727                            "Please update perf tool.\n",
1728                            attr->branch_sample_type);
1729
1730                 return false;
1731         }
1732
1733         return true;
1734 }
1735
1736 static struct evsel *read_event_desc(struct feat_fd *ff)
1737 {
1738         struct evsel *evsel, *events = NULL;
1739         u64 *id;
1740         void *buf = NULL;
1741         u32 nre, sz, nr, i, j;
1742         size_t msz;
1743
1744         /* number of events */
1745         if (do_read_u32(ff, &nre))
1746                 goto error;
1747
1748         if (do_read_u32(ff, &sz))
1749                 goto error;
1750
1751         /* buffer to hold on file attr struct */
1752         buf = malloc(sz);
1753         if (!buf)
1754                 goto error;
1755
1756         /* the last event terminates with evsel->core.attr.size == 0: */
1757         events = calloc(nre + 1, sizeof(*events));
1758         if (!events)
1759                 goto error;
1760
1761         msz = sizeof(evsel->core.attr);
1762         if (sz < msz)
1763                 msz = sz;
1764
1765         for (i = 0, evsel = events; i < nre; evsel++, i++) {
1766                 evsel->idx = i;
1767
1768                 /*
1769                  * must read entire on-file attr struct to
1770                  * sync up with layout.
1771                  */
1772                 if (__do_read(ff, buf, sz))
1773                         goto error;
1774
1775                 if (ff->ph->needs_swap)
1776                         perf_event__attr_swap(buf);
1777
1778                 memcpy(&evsel->core.attr, buf, msz);
1779
1780                 if (!perf_attr_check(&evsel->core.attr))
1781                         goto error;
1782
1783                 if (do_read_u32(ff, &nr))
1784                         goto error;
1785
1786                 if (ff->ph->needs_swap)
1787                         evsel->needs_swap = true;
1788
1789                 evsel->name = do_read_string(ff);
1790                 if (!evsel->name)
1791                         goto error;
1792
1793                 if (!nr)
1794                         continue;
1795
1796                 id = calloc(nr, sizeof(*id));
1797                 if (!id)
1798                         goto error;
1799                 evsel->core.ids = nr;
1800                 evsel->core.id = id;
1801
1802                 for (j = 0 ; j < nr; j++) {
1803                         if (do_read_u64(ff, id))
1804                                 goto error;
1805                         id++;
1806                 }
1807         }
1808 out:
1809         free(buf);
1810         return events;
1811 error:
1812         free_event_desc(events);
1813         events = NULL;
1814         goto out;
1815 }
1816
1817 static int __desc_attr__fprintf(FILE *fp, const char *name, const char *val,
1818                                 void *priv __maybe_unused)
1819 {
1820         return fprintf(fp, ", %s = %s", name, val);
1821 }
1822
1823 static void print_event_desc(struct feat_fd *ff, FILE *fp)
1824 {
1825         struct evsel *evsel, *events;
1826         u32 j;
1827         u64 *id;
1828
1829         if (ff->events)
1830                 events = ff->events;
1831         else
1832                 events = read_event_desc(ff);
1833
1834         if (!events) {
1835                 fprintf(fp, "# event desc: not available or unable to read\n");
1836                 return;
1837         }
1838
1839         for (evsel = events; evsel->core.attr.size; evsel++) {
1840                 fprintf(fp, "# event : name = %s, ", evsel->name);
1841
1842                 if (evsel->core.ids) {
1843                         fprintf(fp, ", id = {");
1844                         for (j = 0, id = evsel->core.id; j < evsel->core.ids; j++, id++) {
1845                                 if (j)
1846                                         fputc(',', fp);
1847                                 fprintf(fp, " %"PRIu64, *id);
1848                         }
1849                         fprintf(fp, " }");
1850                 }
1851
1852                 perf_event_attr__fprintf(fp, &evsel->core.attr, __desc_attr__fprintf, NULL);
1853
1854                 fputc('\n', fp);
1855         }
1856
1857         free_event_desc(events);
1858         ff->events = NULL;
1859 }
1860
1861 static void print_total_mem(struct feat_fd *ff, FILE *fp)
1862 {
1863         fprintf(fp, "# total memory : %llu kB\n", ff->ph->env.total_mem);
1864 }
1865
1866 static void print_numa_topology(struct feat_fd *ff, FILE *fp)
1867 {
1868         int i;
1869         struct numa_node *n;
1870
1871         for (i = 0; i < ff->ph->env.nr_numa_nodes; i++) {
1872                 n = &ff->ph->env.numa_nodes[i];
1873
1874                 fprintf(fp, "# node%u meminfo  : total = %"PRIu64" kB,"
1875                             " free = %"PRIu64" kB\n",
1876                         n->node, n->mem_total, n->mem_free);
1877
1878                 fprintf(fp, "# node%u cpu list : ", n->node);
1879                 cpu_map__fprintf(n->map, fp);
1880         }
1881 }
1882
1883 static void print_cpuid(struct feat_fd *ff, FILE *fp)
1884 {
1885         fprintf(fp, "# cpuid : %s\n", ff->ph->env.cpuid);
1886 }
1887
1888 static void print_branch_stack(struct feat_fd *ff __maybe_unused, FILE *fp)
1889 {
1890         fprintf(fp, "# contains samples with branch stack\n");
1891 }
1892
1893 static void print_auxtrace(struct feat_fd *ff __maybe_unused, FILE *fp)
1894 {
1895         fprintf(fp, "# contains AUX area data (e.g. instruction trace)\n");
1896 }
1897
1898 static void print_stat(struct feat_fd *ff __maybe_unused, FILE *fp)
1899 {
1900         fprintf(fp, "# contains stat data\n");
1901 }
1902
1903 static void print_cache(struct feat_fd *ff, FILE *fp __maybe_unused)
1904 {
1905         int i;
1906
1907         fprintf(fp, "# CPU cache info:\n");
1908         for (i = 0; i < ff->ph->env.caches_cnt; i++) {
1909                 fprintf(fp, "#  ");
1910                 cpu_cache_level__fprintf(fp, &ff->ph->env.caches[i]);
1911         }
1912 }
1913
1914 static void print_compressed(struct feat_fd *ff, FILE *fp)
1915 {
1916         fprintf(fp, "# compressed : %s, level = %d, ratio = %d\n",
1917                 ff->ph->env.comp_type == PERF_COMP_ZSTD ? "Zstd" : "Unknown",
1918                 ff->ph->env.comp_level, ff->ph->env.comp_ratio);
1919 }
1920
1921 static void print_cpu_pmu_caps(struct feat_fd *ff, FILE *fp)
1922 {
1923         const char *delimiter = "# cpu pmu capabilities: ";
1924         u32 nr_caps = ff->ph->env.nr_cpu_pmu_caps;
1925         char *str;
1926
1927         if (!nr_caps) {
1928                 fprintf(fp, "# cpu pmu capabilities: not available\n");
1929                 return;
1930         }
1931
1932         str = ff->ph->env.cpu_pmu_caps;
1933         while (nr_caps--) {
1934                 fprintf(fp, "%s%s", delimiter, str);
1935                 delimiter = ", ";
1936                 str += strlen(str) + 1;
1937         }
1938
1939         fprintf(fp, "\n");
1940 }
1941
1942 static void print_pmu_mappings(struct feat_fd *ff, FILE *fp)
1943 {
1944         const char *delimiter = "# pmu mappings: ";
1945         char *str, *tmp;
1946         u32 pmu_num;
1947         u32 type;
1948
1949         pmu_num = ff->ph->env.nr_pmu_mappings;
1950         if (!pmu_num) {
1951                 fprintf(fp, "# pmu mappings: not available\n");
1952                 return;
1953         }
1954
1955         str = ff->ph->env.pmu_mappings;
1956
1957         while (pmu_num) {
1958                 type = strtoul(str, &tmp, 0);
1959                 if (*tmp != ':')
1960                         goto error;
1961
1962                 str = tmp + 1;
1963                 fprintf(fp, "%s%s = %" PRIu32, delimiter, str, type);
1964
1965                 delimiter = ", ";
1966                 str += strlen(str) + 1;
1967                 pmu_num--;
1968         }
1969
1970         fprintf(fp, "\n");
1971
1972         if (!pmu_num)
1973                 return;
1974 error:
1975         fprintf(fp, "# pmu mappings: unable to read\n");
1976 }
1977
1978 static void print_group_desc(struct feat_fd *ff, FILE *fp)
1979 {
1980         struct perf_session *session;
1981         struct evsel *evsel;
1982         u32 nr = 0;
1983
1984         session = container_of(ff->ph, struct perf_session, header);
1985
1986         evlist__for_each_entry(session->evlist, evsel) {
1987                 if (evsel__is_group_leader(evsel) && evsel->core.nr_members > 1) {
1988                         fprintf(fp, "# group: %s{%s", evsel->group_name ?: "", evsel__name(evsel));
1989
1990                         nr = evsel->core.nr_members - 1;
1991                 } else if (nr) {
1992                         fprintf(fp, ",%s", evsel__name(evsel));
1993
1994                         if (--nr == 0)
1995                                 fprintf(fp, "}\n");
1996                 }
1997         }
1998 }
1999
2000 static void print_sample_time(struct feat_fd *ff, FILE *fp)
2001 {
2002         struct perf_session *session;
2003         char time_buf[32];
2004         double d;
2005
2006         session = container_of(ff->ph, struct perf_session, header);
2007
2008         timestamp__scnprintf_usec(session->evlist->first_sample_time,
2009                                   time_buf, sizeof(time_buf));
2010         fprintf(fp, "# time of first sample : %s\n", time_buf);
2011
2012         timestamp__scnprintf_usec(session->evlist->last_sample_time,
2013                                   time_buf, sizeof(time_buf));
2014         fprintf(fp, "# time of last sample : %s\n", time_buf);
2015
2016         d = (double)(session->evlist->last_sample_time -
2017                 session->evlist->first_sample_time) / NSEC_PER_MSEC;
2018
2019         fprintf(fp, "# sample duration : %10.3f ms\n", d);
2020 }
2021
2022 static void memory_node__fprintf(struct memory_node *n,
2023                                  unsigned long long bsize, FILE *fp)
2024 {
2025         char buf_map[100], buf_size[50];
2026         unsigned long long size;
2027
2028         size = bsize * bitmap_weight(n->set, n->size);
2029         unit_number__scnprintf(buf_size, 50, size);
2030
2031         bitmap_scnprintf(n->set, n->size, buf_map, 100);
2032         fprintf(fp, "#  %3" PRIu64 " [%s]: %s\n", n->node, buf_size, buf_map);
2033 }
2034
2035 static void print_mem_topology(struct feat_fd *ff, FILE *fp)
2036 {
2037         struct memory_node *nodes;
2038         int i, nr;
2039
2040         nodes = ff->ph->env.memory_nodes;
2041         nr    = ff->ph->env.nr_memory_nodes;
2042
2043         fprintf(fp, "# memory nodes (nr %d, block size 0x%llx):\n",
2044                 nr, ff->ph->env.memory_bsize);
2045
2046         for (i = 0; i < nr; i++) {
2047                 memory_node__fprintf(&nodes[i], ff->ph->env.memory_bsize, fp);
2048         }
2049 }
2050
2051 static int __event_process_build_id(struct perf_record_header_build_id *bev,
2052                                     char *filename,
2053                                     struct perf_session *session)
2054 {
2055         int err = -1;
2056         struct machine *machine;
2057         u16 cpumode;
2058         struct dso *dso;
2059         enum dso_space_type dso_space;
2060
2061         machine = perf_session__findnew_machine(session, bev->pid);
2062         if (!machine)
2063                 goto out;
2064
2065         cpumode = bev->header.misc & PERF_RECORD_MISC_CPUMODE_MASK;
2066
2067         switch (cpumode) {
2068         case PERF_RECORD_MISC_KERNEL:
2069                 dso_space = DSO_SPACE__KERNEL;
2070                 break;
2071         case PERF_RECORD_MISC_GUEST_KERNEL:
2072                 dso_space = DSO_SPACE__KERNEL_GUEST;
2073                 break;
2074         case PERF_RECORD_MISC_USER:
2075         case PERF_RECORD_MISC_GUEST_USER:
2076                 dso_space = DSO_SPACE__USER;
2077                 break;
2078         default:
2079                 goto out;
2080         }
2081
2082         dso = machine__findnew_dso(machine, filename);
2083         if (dso != NULL) {
2084                 char sbuild_id[SBUILD_ID_SIZE];
2085                 struct build_id bid;
2086                 size_t size = BUILD_ID_SIZE;
2087
2088                 if (bev->header.misc & PERF_RECORD_MISC_BUILD_ID_SIZE)
2089                         size = bev->size;
2090
2091                 build_id__init(&bid, bev->data, size);
2092                 dso__set_build_id(dso, &bid);
2093
2094                 if (dso_space != DSO_SPACE__USER) {
2095                         struct kmod_path m = { .name = NULL, };
2096
2097                         if (!kmod_path__parse_name(&m, filename) && m.kmod)
2098                                 dso__set_module_info(dso, &m, machine);
2099
2100                         dso->kernel = dso_space;
2101                         free(m.name);
2102                 }
2103
2104                 build_id__sprintf(&dso->bid, sbuild_id);
2105                 pr_debug("build id event received for %s: %s [%zu]\n",
2106                          dso->long_name, sbuild_id, size);
2107                 dso__put(dso);
2108         }
2109
2110         err = 0;
2111 out:
2112         return err;
2113 }
2114
2115 static int perf_header__read_build_ids_abi_quirk(struct perf_header *header,
2116                                                  int input, u64 offset, u64 size)
2117 {
2118         struct perf_session *session = container_of(header, struct perf_session, header);
2119         struct {
2120                 struct perf_event_header   header;
2121                 u8                         build_id[PERF_ALIGN(BUILD_ID_SIZE, sizeof(u64))];
2122                 char                       filename[0];
2123         } old_bev;
2124         struct perf_record_header_build_id bev;
2125         char filename[PATH_MAX];
2126         u64 limit = offset + size;
2127
2128         while (offset < limit) {
2129                 ssize_t len;
2130
2131                 if (readn(input, &old_bev, sizeof(old_bev)) != sizeof(old_bev))
2132                         return -1;
2133
2134                 if (header->needs_swap)
2135                         perf_event_header__bswap(&old_bev.header);
2136
2137                 len = old_bev.header.size - sizeof(old_bev);
2138                 if (readn(input, filename, len) != len)
2139                         return -1;
2140
2141                 bev.header = old_bev.header;
2142
2143                 /*
2144                  * As the pid is the missing value, we need to fill
2145                  * it properly. The header.misc value give us nice hint.
2146                  */
2147                 bev.pid = HOST_KERNEL_ID;
2148                 if (bev.header.misc == PERF_RECORD_MISC_GUEST_USER ||
2149                     bev.header.misc == PERF_RECORD_MISC_GUEST_KERNEL)
2150                         bev.pid = DEFAULT_GUEST_KERNEL_ID;
2151
2152                 memcpy(bev.build_id, old_bev.build_id, sizeof(bev.build_id));
2153                 __event_process_build_id(&bev, filename, session);
2154
2155                 offset += bev.header.size;
2156         }
2157
2158         return 0;
2159 }
2160
2161 static int perf_header__read_build_ids(struct perf_header *header,
2162                                        int input, u64 offset, u64 size)
2163 {
2164         struct perf_session *session = container_of(header, struct perf_session, header);
2165         struct perf_record_header_build_id bev;
2166         char filename[PATH_MAX];
2167         u64 limit = offset + size, orig_offset = offset;
2168         int err = -1;
2169
2170         while (offset < limit) {
2171                 ssize_t len;
2172
2173                 if (readn(input, &bev, sizeof(bev)) != sizeof(bev))
2174                         goto out;
2175
2176                 if (header->needs_swap)
2177                         perf_event_header__bswap(&bev.header);
2178
2179                 len = bev.header.size - sizeof(bev);
2180                 if (readn(input, filename, len) != len)
2181                         goto out;
2182                 /*
2183                  * The a1645ce1 changeset:
2184                  *
2185                  * "perf: 'perf kvm' tool for monitoring guest performance from host"
2186                  *
2187                  * Added a field to struct perf_record_header_build_id that broke the file
2188                  * format.
2189                  *
2190                  * Since the kernel build-id is the first entry, process the
2191                  * table using the old format if the well known
2192                  * '[kernel.kallsyms]' string for the kernel build-id has the
2193                  * first 4 characters chopped off (where the pid_t sits).
2194                  */
2195                 if (memcmp(filename, "nel.kallsyms]", 13) == 0) {
2196                         if (lseek(input, orig_offset, SEEK_SET) == (off_t)-1)
2197                                 return -1;
2198                         return perf_header__read_build_ids_abi_quirk(header, input, offset, size);
2199                 }
2200
2201                 __event_process_build_id(&bev, filename, session);
2202
2203                 offset += bev.header.size;
2204         }
2205         err = 0;
2206 out:
2207         return err;
2208 }
2209
2210 /* Macro for features that simply need to read and store a string. */
2211 #define FEAT_PROCESS_STR_FUN(__feat, __feat_env) \
2212 static int process_##__feat(struct feat_fd *ff, void *data __maybe_unused) \
2213 {\
2214         ff->ph->env.__feat_env = do_read_string(ff); \
2215         return ff->ph->env.__feat_env ? 0 : -ENOMEM; \
2216 }
2217
2218 FEAT_PROCESS_STR_FUN(hostname, hostname);
2219 FEAT_PROCESS_STR_FUN(osrelease, os_release);
2220 FEAT_PROCESS_STR_FUN(version, version);
2221 FEAT_PROCESS_STR_FUN(arch, arch);
2222 FEAT_PROCESS_STR_FUN(cpudesc, cpu_desc);
2223 FEAT_PROCESS_STR_FUN(cpuid, cpuid);
2224
2225 static int process_tracing_data(struct feat_fd *ff, void *data)
2226 {
2227         ssize_t ret = trace_report(ff->fd, data, false);
2228
2229         return ret < 0 ? -1 : 0;
2230 }
2231
2232 static int process_build_id(struct feat_fd *ff, void *data __maybe_unused)
2233 {
2234         if (perf_header__read_build_ids(ff->ph, ff->fd, ff->offset, ff->size))
2235                 pr_debug("Failed to read buildids, continuing...\n");
2236         return 0;
2237 }
2238
2239 static int process_nrcpus(struct feat_fd *ff, void *data __maybe_unused)
2240 {
2241         int ret;
2242         u32 nr_cpus_avail, nr_cpus_online;
2243
2244         ret = do_read_u32(ff, &nr_cpus_avail);
2245         if (ret)
2246                 return ret;
2247
2248         ret = do_read_u32(ff, &nr_cpus_online);
2249         if (ret)
2250                 return ret;
2251         ff->ph->env.nr_cpus_avail = (int)nr_cpus_avail;
2252         ff->ph->env.nr_cpus_online = (int)nr_cpus_online;
2253         return 0;
2254 }
2255
2256 static int process_total_mem(struct feat_fd *ff, void *data __maybe_unused)
2257 {
2258         u64 total_mem;
2259         int ret;
2260
2261         ret = do_read_u64(ff, &total_mem);
2262         if (ret)
2263                 return -1;
2264         ff->ph->env.total_mem = (unsigned long long)total_mem;
2265         return 0;
2266 }
2267
2268 static struct evsel *
2269 perf_evlist__find_by_index(struct evlist *evlist, int idx)
2270 {
2271         struct evsel *evsel;
2272
2273         evlist__for_each_entry(evlist, evsel) {
2274                 if (evsel->idx == idx)
2275                         return evsel;
2276         }
2277
2278         return NULL;
2279 }
2280
2281 static void
2282 perf_evlist__set_event_name(struct evlist *evlist,
2283                             struct evsel *event)
2284 {
2285         struct evsel *evsel;
2286
2287         if (!event->name)
2288                 return;
2289
2290         evsel = perf_evlist__find_by_index(evlist, event->idx);
2291         if (!evsel)
2292                 return;
2293
2294         if (evsel->name)
2295                 return;
2296
2297         evsel->name = strdup(event->name);
2298 }
2299
2300 static int
2301 process_event_desc(struct feat_fd *ff, void *data __maybe_unused)
2302 {
2303         struct perf_session *session;
2304         struct evsel *evsel, *events = read_event_desc(ff);
2305
2306         if (!events)
2307                 return 0;
2308
2309         session = container_of(ff->ph, struct perf_session, header);
2310
2311         if (session->data->is_pipe) {
2312                 /* Save events for reading later by print_event_desc,
2313                  * since they can't be read again in pipe mode. */
2314                 ff->events = events;
2315         }
2316
2317         for (evsel = events; evsel->core.attr.size; evsel++)
2318                 perf_evlist__set_event_name(session->evlist, evsel);
2319
2320         if (!session->data->is_pipe)
2321                 free_event_desc(events);
2322
2323         return 0;
2324 }
2325
2326 static int process_cmdline(struct feat_fd *ff, void *data __maybe_unused)
2327 {
2328         char *str, *cmdline = NULL, **argv = NULL;
2329         u32 nr, i, len = 0;
2330
2331         if (do_read_u32(ff, &nr))
2332                 return -1;
2333
2334         ff->ph->env.nr_cmdline = nr;
2335
2336         cmdline = zalloc(ff->size + nr + 1);
2337         if (!cmdline)
2338                 return -1;
2339
2340         argv = zalloc(sizeof(char *) * (nr + 1));
2341         if (!argv)
2342                 goto error;
2343
2344         for (i = 0; i < nr; i++) {
2345                 str = do_read_string(ff);
2346                 if (!str)
2347                         goto error;
2348
2349                 argv[i] = cmdline + len;
2350                 memcpy(argv[i], str, strlen(str) + 1);
2351                 len += strlen(str) + 1;
2352                 free(str);
2353         }
2354         ff->ph->env.cmdline = cmdline;
2355         ff->ph->env.cmdline_argv = (const char **) argv;
2356         return 0;
2357
2358 error:
2359         free(argv);
2360         free(cmdline);
2361         return -1;
2362 }
2363
2364 static int process_cpu_topology(struct feat_fd *ff, void *data __maybe_unused)
2365 {
2366         u32 nr, i;
2367         char *str;
2368         struct strbuf sb;
2369         int cpu_nr = ff->ph->env.nr_cpus_avail;
2370         u64 size = 0;
2371         struct perf_header *ph = ff->ph;
2372         bool do_core_id_test = true;
2373
2374         ph->env.cpu = calloc(cpu_nr, sizeof(*ph->env.cpu));
2375         if (!ph->env.cpu)
2376                 return -1;
2377
2378         if (do_read_u32(ff, &nr))
2379                 goto free_cpu;
2380
2381         ph->env.nr_sibling_cores = nr;
2382         size += sizeof(u32);
2383         if (strbuf_init(&sb, 128) < 0)
2384                 goto free_cpu;
2385
2386         for (i = 0; i < nr; i++) {
2387                 str = do_read_string(ff);
2388                 if (!str)
2389                         goto error;
2390
2391                 /* include a NULL character at the end */
2392                 if (strbuf_add(&sb, str, strlen(str) + 1) < 0)
2393                         goto error;
2394                 size += string_size(str);
2395                 free(str);
2396         }
2397         ph->env.sibling_cores = strbuf_detach(&sb, NULL);
2398
2399         if (do_read_u32(ff, &nr))
2400                 return -1;
2401
2402         ph->env.nr_sibling_threads = nr;
2403         size += sizeof(u32);
2404
2405         for (i = 0; i < nr; i++) {
2406                 str = do_read_string(ff);
2407                 if (!str)
2408                         goto error;
2409
2410                 /* include a NULL character at the end */
2411                 if (strbuf_add(&sb, str, strlen(str) + 1) < 0)
2412                         goto error;
2413                 size += string_size(str);
2414                 free(str);
2415         }
2416         ph->env.sibling_threads = strbuf_detach(&sb, NULL);
2417
2418         /*
2419          * The header may be from old perf,
2420          * which doesn't include core id and socket id information.
2421          */
2422         if (ff->size <= size) {
2423                 zfree(&ph->env.cpu);
2424                 return 0;
2425         }
2426
2427         /* On s390 the socket_id number is not related to the numbers of cpus.
2428          * The socket_id number might be higher than the numbers of cpus.
2429          * This depends on the configuration.
2430          * AArch64 is the same.
2431          */
2432         if (ph->env.arch && (!strncmp(ph->env.arch, "s390", 4)
2433                           || !strncmp(ph->env.arch, "aarch64", 7)))
2434                 do_core_id_test = false;
2435
2436         for (i = 0; i < (u32)cpu_nr; i++) {
2437                 if (do_read_u32(ff, &nr))
2438                         goto free_cpu;
2439
2440                 ph->env.cpu[i].core_id = nr;
2441                 size += sizeof(u32);
2442
2443                 if (do_read_u32(ff, &nr))
2444                         goto free_cpu;
2445
2446                 if (do_core_id_test && nr != (u32)-1 && nr > (u32)cpu_nr) {
2447                         pr_debug("socket_id number is too big."
2448                                  "You may need to upgrade the perf tool.\n");
2449                         goto free_cpu;
2450                 }
2451
2452                 ph->env.cpu[i].socket_id = nr;
2453                 size += sizeof(u32);
2454         }
2455
2456         /*
2457          * The header may be from old perf,
2458          * which doesn't include die information.
2459          */
2460         if (ff->size <= size)
2461                 return 0;
2462
2463         if (do_read_u32(ff, &nr))
2464                 return -1;
2465
2466         ph->env.nr_sibling_dies = nr;
2467         size += sizeof(u32);
2468
2469         for (i = 0; i < nr; i++) {
2470                 str = do_read_string(ff);
2471                 if (!str)
2472                         goto error;
2473
2474                 /* include a NULL character at the end */
2475                 if (strbuf_add(&sb, str, strlen(str) + 1) < 0)
2476                         goto error;
2477                 size += string_size(str);
2478                 free(str);
2479         }
2480         ph->env.sibling_dies = strbuf_detach(&sb, NULL);
2481
2482         for (i = 0; i < (u32)cpu_nr; i++) {
2483                 if (do_read_u32(ff, &nr))
2484                         goto free_cpu;
2485
2486                 ph->env.cpu[i].die_id = nr;
2487         }
2488
2489         return 0;
2490
2491 error:
2492         strbuf_release(&sb);
2493 free_cpu:
2494         zfree(&ph->env.cpu);
2495         return -1;
2496 }
2497
2498 static int process_numa_topology(struct feat_fd *ff, void *data __maybe_unused)
2499 {
2500         struct numa_node *nodes, *n;
2501         u32 nr, i;
2502         char *str;
2503
2504         /* nr nodes */
2505         if (do_read_u32(ff, &nr))
2506                 return -1;
2507
2508         nodes = zalloc(sizeof(*nodes) * nr);
2509         if (!nodes)
2510                 return -ENOMEM;
2511
2512         for (i = 0; i < nr; i++) {
2513                 n = &nodes[i];
2514
2515                 /* node number */
2516                 if (do_read_u32(ff, &n->node))
2517                         goto error;
2518
2519                 if (do_read_u64(ff, &n->mem_total))
2520                         goto error;
2521
2522                 if (do_read_u64(ff, &n->mem_free))
2523                         goto error;
2524
2525                 str = do_read_string(ff);
2526                 if (!str)
2527                         goto error;
2528
2529                 n->map = perf_cpu_map__new(str);
2530                 if (!n->map)
2531                         goto error;
2532
2533                 free(str);
2534         }
2535         ff->ph->env.nr_numa_nodes = nr;
2536         ff->ph->env.numa_nodes = nodes;
2537         return 0;
2538
2539 error:
2540         free(nodes);
2541         return -1;
2542 }
2543
2544 static int process_pmu_mappings(struct feat_fd *ff, void *data __maybe_unused)
2545 {
2546         char *name;
2547         u32 pmu_num;
2548         u32 type;
2549         struct strbuf sb;
2550
2551         if (do_read_u32(ff, &pmu_num))
2552                 return -1;
2553
2554         if (!pmu_num) {
2555                 pr_debug("pmu mappings not available\n");
2556                 return 0;
2557         }
2558
2559         ff->ph->env.nr_pmu_mappings = pmu_num;
2560         if (strbuf_init(&sb, 128) < 0)
2561                 return -1;
2562
2563         while (pmu_num) {
2564                 if (do_read_u32(ff, &type))
2565                         goto error;
2566
2567                 name = do_read_string(ff);
2568                 if (!name)
2569                         goto error;
2570
2571                 if (strbuf_addf(&sb, "%u:%s", type, name) < 0)
2572                         goto error;
2573                 /* include a NULL character at the end */
2574                 if (strbuf_add(&sb, "", 1) < 0)
2575                         goto error;
2576
2577                 if (!strcmp(name, "msr"))
2578                         ff->ph->env.msr_pmu_type = type;
2579
2580                 free(name);
2581                 pmu_num--;
2582         }
2583         ff->ph->env.pmu_mappings = strbuf_detach(&sb, NULL);
2584         return 0;
2585
2586 error:
2587         strbuf_release(&sb);
2588         return -1;
2589 }
2590
2591 static int process_group_desc(struct feat_fd *ff, void *data __maybe_unused)
2592 {
2593         size_t ret = -1;
2594         u32 i, nr, nr_groups;
2595         struct perf_session *session;
2596         struct evsel *evsel, *leader = NULL;
2597         struct group_desc {
2598                 char *name;
2599                 u32 leader_idx;
2600                 u32 nr_members;
2601         } *desc;
2602
2603         if (do_read_u32(ff, &nr_groups))
2604                 return -1;
2605
2606         ff->ph->env.nr_groups = nr_groups;
2607         if (!nr_groups) {
2608                 pr_debug("group desc not available\n");
2609                 return 0;
2610         }
2611
2612         desc = calloc(nr_groups, sizeof(*desc));
2613         if (!desc)
2614                 return -1;
2615
2616         for (i = 0; i < nr_groups; i++) {
2617                 desc[i].name = do_read_string(ff);
2618                 if (!desc[i].name)
2619                         goto out_free;
2620
2621                 if (do_read_u32(ff, &desc[i].leader_idx))
2622                         goto out_free;
2623
2624                 if (do_read_u32(ff, &desc[i].nr_members))
2625                         goto out_free;
2626         }
2627
2628         /*
2629          * Rebuild group relationship based on the group_desc
2630          */
2631         session = container_of(ff->ph, struct perf_session, header);
2632         session->evlist->nr_groups = nr_groups;
2633
2634         i = nr = 0;
2635         evlist__for_each_entry(session->evlist, evsel) {
2636                 if (evsel->idx == (int) desc[i].leader_idx) {
2637                         evsel->leader = evsel;
2638                         /* {anon_group} is a dummy name */
2639                         if (strcmp(desc[i].name, "{anon_group}")) {
2640                                 evsel->group_name = desc[i].name;
2641                                 desc[i].name = NULL;
2642                         }
2643                         evsel->core.nr_members = desc[i].nr_members;
2644
2645                         if (i >= nr_groups || nr > 0) {
2646                                 pr_debug("invalid group desc\n");
2647                                 goto out_free;
2648                         }
2649
2650                         leader = evsel;
2651                         nr = evsel->core.nr_members - 1;
2652                         i++;
2653                 } else if (nr) {
2654                         /* This is a group member */
2655                         evsel->leader = leader;
2656
2657                         nr--;
2658                 }
2659         }
2660
2661         if (i != nr_groups || nr != 0) {
2662                 pr_debug("invalid group desc\n");
2663                 goto out_free;
2664         }
2665
2666         ret = 0;
2667 out_free:
2668         for (i = 0; i < nr_groups; i++)
2669                 zfree(&desc[i].name);
2670         free(desc);
2671
2672         return ret;
2673 }
2674
2675 static int process_auxtrace(struct feat_fd *ff, void *data __maybe_unused)
2676 {
2677         struct perf_session *session;
2678         int err;
2679
2680         session = container_of(ff->ph, struct perf_session, header);
2681
2682         err = auxtrace_index__process(ff->fd, ff->size, session,
2683                                       ff->ph->needs_swap);
2684         if (err < 0)
2685                 pr_err("Failed to process auxtrace index\n");
2686         return err;
2687 }
2688
2689 static int process_cache(struct feat_fd *ff, void *data __maybe_unused)
2690 {
2691         struct cpu_cache_level *caches;
2692         u32 cnt, i, version;
2693
2694         if (do_read_u32(ff, &version))
2695                 return -1;
2696
2697         if (version != 1)
2698                 return -1;
2699
2700         if (do_read_u32(ff, &cnt))
2701                 return -1;
2702
2703         caches = zalloc(sizeof(*caches) * cnt);
2704         if (!caches)
2705                 return -1;
2706
2707         for (i = 0; i < cnt; i++) {
2708                 struct cpu_cache_level c;
2709
2710                 #define _R(v)                                           \
2711                         if (do_read_u32(ff, &c.v))\
2712                                 goto out_free_caches;                   \
2713
2714                 _R(level)
2715                 _R(line_size)
2716                 _R(sets)
2717                 _R(ways)
2718                 #undef _R
2719
2720                 #define _R(v)                                   \
2721                         c.v = do_read_string(ff);               \
2722                         if (!c.v)                               \
2723                                 goto out_free_caches;
2724
2725                 _R(type)
2726                 _R(size)
2727                 _R(map)
2728                 #undef _R
2729
2730                 caches[i] = c;
2731         }
2732
2733         ff->ph->env.caches = caches;
2734         ff->ph->env.caches_cnt = cnt;
2735         return 0;
2736 out_free_caches:
2737         free(caches);
2738         return -1;
2739 }
2740
2741 static int process_sample_time(struct feat_fd *ff, void *data __maybe_unused)
2742 {
2743         struct perf_session *session;
2744         u64 first_sample_time, last_sample_time;
2745         int ret;
2746
2747         session = container_of(ff->ph, struct perf_session, header);
2748
2749         ret = do_read_u64(ff, &first_sample_time);
2750         if (ret)
2751                 return -1;
2752
2753         ret = do_read_u64(ff, &last_sample_time);
2754         if (ret)
2755                 return -1;
2756
2757         session->evlist->first_sample_time = first_sample_time;
2758         session->evlist->last_sample_time = last_sample_time;
2759         return 0;
2760 }
2761
2762 static int process_mem_topology(struct feat_fd *ff,
2763                                 void *data __maybe_unused)
2764 {
2765         struct memory_node *nodes;
2766         u64 version, i, nr, bsize;
2767         int ret = -1;
2768
2769         if (do_read_u64(ff, &version))
2770                 return -1;
2771
2772         if (version != 1)
2773                 return -1;
2774
2775         if (do_read_u64(ff, &bsize))
2776                 return -1;
2777
2778         if (do_read_u64(ff, &nr))
2779                 return -1;
2780
2781         nodes = zalloc(sizeof(*nodes) * nr);
2782         if (!nodes)
2783                 return -1;
2784
2785         for (i = 0; i < nr; i++) {
2786                 struct memory_node n;
2787
2788                 #define _R(v)                           \
2789                         if (do_read_u64(ff, &n.v))      \
2790                                 goto out;               \
2791
2792                 _R(node)
2793                 _R(size)
2794
2795                 #undef _R
2796
2797                 if (do_read_bitmap(ff, &n.set, &n.size))
2798                         goto out;
2799
2800                 nodes[i] = n;
2801         }
2802
2803         ff->ph->env.memory_bsize    = bsize;
2804         ff->ph->env.memory_nodes    = nodes;
2805         ff->ph->env.nr_memory_nodes = nr;
2806         ret = 0;
2807
2808 out:
2809         if (ret)
2810                 free(nodes);
2811         return ret;
2812 }
2813
2814 static int process_clockid(struct feat_fd *ff,
2815                            void *data __maybe_unused)
2816 {
2817         if (do_read_u64(ff, &ff->ph->env.clock.clockid_res_ns))
2818                 return -1;
2819
2820         return 0;
2821 }
2822
2823 static int process_clock_data(struct feat_fd *ff,
2824                               void *_data __maybe_unused)
2825 {
2826         u32 data32;
2827         u64 data64;
2828
2829         /* version */
2830         if (do_read_u32(ff, &data32))
2831                 return -1;
2832
2833         if (data32 != 1)
2834                 return -1;
2835
2836         /* clockid */
2837         if (do_read_u32(ff, &data32))
2838                 return -1;
2839
2840         ff->ph->env.clock.clockid = data32;
2841
2842         /* TOD ref time */
2843         if (do_read_u64(ff, &data64))
2844                 return -1;
2845
2846         ff->ph->env.clock.tod_ns = data64;
2847
2848         /* clockid ref time */
2849         if (do_read_u64(ff, &data64))
2850                 return -1;
2851
2852         ff->ph->env.clock.clockid_ns = data64;
2853         ff->ph->env.clock.enabled = true;
2854         return 0;
2855 }
2856
2857 static int process_dir_format(struct feat_fd *ff,
2858                               void *_data __maybe_unused)
2859 {
2860         struct perf_session *session;
2861         struct perf_data *data;
2862
2863         session = container_of(ff->ph, struct perf_session, header);
2864         data = session->data;
2865
2866         if (WARN_ON(!perf_data__is_dir(data)))
2867                 return -1;
2868
2869         return do_read_u64(ff, &data->dir.version);
2870 }
2871
2872 #ifdef HAVE_LIBBPF_SUPPORT
2873 static int process_bpf_prog_info(struct feat_fd *ff, void *data __maybe_unused)
2874 {
2875         struct bpf_prog_info_linear *info_linear;
2876         struct bpf_prog_info_node *info_node;
2877         struct perf_env *env = &ff->ph->env;
2878         u32 count, i;
2879         int err = -1;
2880
2881         if (ff->ph->needs_swap) {
2882                 pr_warning("interpreting bpf_prog_info from systems with endianity is not yet supported\n");
2883                 return 0;
2884         }
2885
2886         if (do_read_u32(ff, &count))
2887                 return -1;
2888
2889         down_write(&env->bpf_progs.lock);
2890
2891         for (i = 0; i < count; ++i) {
2892                 u32 info_len, data_len;
2893
2894                 info_linear = NULL;
2895                 info_node = NULL;
2896                 if (do_read_u32(ff, &info_len))
2897                         goto out;
2898                 if (do_read_u32(ff, &data_len))
2899                         goto out;
2900
2901                 if (info_len > sizeof(struct bpf_prog_info)) {
2902                         pr_warning("detected invalid bpf_prog_info\n");
2903                         goto out;
2904                 }
2905
2906                 info_linear = malloc(sizeof(struct bpf_prog_info_linear) +
2907                                      data_len);
2908                 if (!info_linear)
2909                         goto out;
2910                 info_linear->info_len = sizeof(struct bpf_prog_info);
2911                 info_linear->data_len = data_len;
2912                 if (do_read_u64(ff, (u64 *)(&info_linear->arrays)))
2913                         goto out;
2914                 if (__do_read(ff, &info_linear->info, info_len))
2915                         goto out;
2916                 if (info_len < sizeof(struct bpf_prog_info))
2917                         memset(((void *)(&info_linear->info)) + info_len, 0,
2918                                sizeof(struct bpf_prog_info) - info_len);
2919
2920                 if (__do_read(ff, info_linear->data, data_len))
2921                         goto out;
2922
2923                 info_node = malloc(sizeof(struct bpf_prog_info_node));
2924                 if (!info_node)
2925                         goto out;
2926
2927                 /* after reading from file, translate offset to address */
2928                 bpf_program__bpil_offs_to_addr(info_linear);
2929                 info_node->info_linear = info_linear;
2930                 __perf_env__insert_bpf_prog_info(env, info_node);
2931         }
2932
2933         up_write(&env->bpf_progs.lock);
2934         return 0;
2935 out:
2936         free(info_linear);
2937         free(info_node);
2938         up_write(&env->bpf_progs.lock);
2939         return err;
2940 }
2941 #else // HAVE_LIBBPF_SUPPORT
2942 static int process_bpf_prog_info(struct feat_fd *ff __maybe_unused, void *data __maybe_unused)
2943 {
2944         return 0;
2945 }
2946 #endif // HAVE_LIBBPF_SUPPORT
2947
2948 static int process_bpf_btf(struct feat_fd *ff, void *data __maybe_unused)
2949 {
2950         struct perf_env *env = &ff->ph->env;
2951         struct btf_node *node = NULL;
2952         u32 count, i;
2953         int err = -1;
2954
2955         if (ff->ph->needs_swap) {
2956                 pr_warning("interpreting btf from systems with endianity is not yet supported\n");
2957                 return 0;
2958         }
2959
2960         if (do_read_u32(ff, &count))
2961                 return -1;
2962
2963         down_write(&env->bpf_progs.lock);
2964
2965         for (i = 0; i < count; ++i) {
2966                 u32 id, data_size;
2967
2968                 if (do_read_u32(ff, &id))
2969                         goto out;
2970                 if (do_read_u32(ff, &data_size))
2971                         goto out;
2972
2973                 node = malloc(sizeof(struct btf_node) + data_size);
2974                 if (!node)
2975                         goto out;
2976
2977                 node->id = id;
2978                 node->data_size = data_size;
2979
2980                 if (__do_read(ff, node->data, data_size))
2981                         goto out;
2982
2983                 __perf_env__insert_btf(env, node);
2984                 node = NULL;
2985         }
2986
2987         err = 0;
2988 out:
2989         up_write(&env->bpf_progs.lock);
2990         free(node);
2991         return err;
2992 }
2993
2994 static int process_compressed(struct feat_fd *ff,
2995                               void *data __maybe_unused)
2996 {
2997         if (do_read_u32(ff, &(ff->ph->env.comp_ver)))
2998                 return -1;
2999
3000         if (do_read_u32(ff, &(ff->ph->env.comp_type)))
3001                 return -1;
3002
3003         if (do_read_u32(ff, &(ff->ph->env.comp_level)))
3004                 return -1;
3005
3006         if (do_read_u32(ff, &(ff->ph->env.comp_ratio)))
3007                 return -1;
3008
3009         if (do_read_u32(ff, &(ff->ph->env.comp_mmap_len)))
3010                 return -1;
3011
3012         return 0;
3013 }
3014
3015 static int process_cpu_pmu_caps(struct feat_fd *ff,
3016                                 void *data __maybe_unused)
3017 {
3018         char *name, *value;
3019         struct strbuf sb;
3020         u32 nr_caps;
3021
3022         if (do_read_u32(ff, &nr_caps))
3023                 return -1;
3024
3025         if (!nr_caps) {
3026                 pr_debug("cpu pmu capabilities not available\n");
3027                 return 0;
3028         }
3029
3030         ff->ph->env.nr_cpu_pmu_caps = nr_caps;
3031
3032         if (strbuf_init(&sb, 128) < 0)
3033                 return -1;
3034
3035         while (nr_caps--) {
3036                 name = do_read_string(ff);
3037                 if (!name)
3038                         goto error;
3039
3040                 value = do_read_string(ff);
3041                 if (!value)
3042                         goto free_name;
3043
3044                 if (strbuf_addf(&sb, "%s=%s", name, value) < 0)
3045                         goto free_value;
3046
3047                 /* include a NULL character at the end */
3048                 if (strbuf_add(&sb, "", 1) < 0)
3049                         goto free_value;
3050
3051                 if (!strcmp(name, "branches"))
3052                         ff->ph->env.max_branches = atoi(value);
3053
3054                 free(value);
3055                 free(name);
3056         }
3057         ff->ph->env.cpu_pmu_caps = strbuf_detach(&sb, NULL);
3058         return 0;
3059
3060 free_value:
3061         free(value);
3062 free_name:
3063         free(name);
3064 error:
3065         strbuf_release(&sb);
3066         return -1;
3067 }
3068
3069 #define FEAT_OPR(n, func, __full_only) \
3070         [HEADER_##n] = {                                        \
3071                 .name       = __stringify(n),                   \
3072                 .write      = write_##func,                     \
3073                 .print      = print_##func,                     \
3074                 .full_only  = __full_only,                      \
3075                 .process    = process_##func,                   \
3076                 .synthesize = true                              \
3077         }
3078
3079 #define FEAT_OPN(n, func, __full_only) \
3080         [HEADER_##n] = {                                        \
3081                 .name       = __stringify(n),                   \
3082                 .write      = write_##func,                     \
3083                 .print      = print_##func,                     \
3084                 .full_only  = __full_only,                      \
3085                 .process    = process_##func                    \
3086         }
3087
3088 /* feature_ops not implemented: */
3089 #define print_tracing_data      NULL
3090 #define print_build_id          NULL
3091
3092 #define process_branch_stack    NULL
3093 #define process_stat            NULL
3094
3095 // Only used in util/synthetic-events.c
3096 const struct perf_header_feature_ops feat_ops[HEADER_LAST_FEATURE];
3097
3098 const struct perf_header_feature_ops feat_ops[HEADER_LAST_FEATURE] = {
3099         FEAT_OPN(TRACING_DATA,  tracing_data,   false),
3100         FEAT_OPN(BUILD_ID,      build_id,       false),
3101         FEAT_OPR(HOSTNAME,      hostname,       false),
3102         FEAT_OPR(OSRELEASE,     osrelease,      false),
3103         FEAT_OPR(VERSION,       version,        false),
3104         FEAT_OPR(ARCH,          arch,           false),
3105         FEAT_OPR(NRCPUS,        nrcpus,         false),
3106         FEAT_OPR(CPUDESC,       cpudesc,        false),
3107         FEAT_OPR(CPUID,         cpuid,          false),
3108         FEAT_OPR(TOTAL_MEM,     total_mem,      false),
3109         FEAT_OPR(EVENT_DESC,    event_desc,     false),
3110         FEAT_OPR(CMDLINE,       cmdline,        false),
3111         FEAT_OPR(CPU_TOPOLOGY,  cpu_topology,   true),
3112         FEAT_OPR(NUMA_TOPOLOGY, numa_topology,  true),
3113         FEAT_OPN(BRANCH_STACK,  branch_stack,   false),
3114         FEAT_OPR(PMU_MAPPINGS,  pmu_mappings,   false),
3115         FEAT_OPR(GROUP_DESC,    group_desc,     false),
3116         FEAT_OPN(AUXTRACE,      auxtrace,       false),
3117         FEAT_OPN(STAT,          stat,           false),
3118         FEAT_OPN(CACHE,         cache,          true),
3119         FEAT_OPR(SAMPLE_TIME,   sample_time,    false),
3120         FEAT_OPR(MEM_TOPOLOGY,  mem_topology,   true),
3121         FEAT_OPR(CLOCKID,       clockid,        false),
3122         FEAT_OPN(DIR_FORMAT,    dir_format,     false),
3123         FEAT_OPR(BPF_PROG_INFO, bpf_prog_info,  false),
3124         FEAT_OPR(BPF_BTF,       bpf_btf,        false),
3125         FEAT_OPR(COMPRESSED,    compressed,     false),
3126         FEAT_OPR(CPU_PMU_CAPS,  cpu_pmu_caps,   false),
3127         FEAT_OPR(CLOCK_DATA,    clock_data,     false),
3128 };
3129
3130 struct header_print_data {
3131         FILE *fp;
3132         bool full; /* extended list of headers */
3133 };
3134
3135 static int perf_file_section__fprintf_info(struct perf_file_section *section,
3136                                            struct perf_header *ph,
3137                                            int feat, int fd, void *data)
3138 {
3139         struct header_print_data *hd = data;
3140         struct feat_fd ff;
3141
3142         if (lseek(fd, section->offset, SEEK_SET) == (off_t)-1) {
3143                 pr_debug("Failed to lseek to %" PRIu64 " offset for feature "
3144                                 "%d, continuing...\n", section->offset, feat);
3145                 return 0;
3146         }
3147         if (feat >= HEADER_LAST_FEATURE) {
3148                 pr_warning("unknown feature %d\n", feat);
3149                 return 0;
3150         }
3151         if (!feat_ops[feat].print)
3152                 return 0;
3153
3154         ff = (struct  feat_fd) {
3155                 .fd = fd,
3156                 .ph = ph,
3157         };
3158
3159         if (!feat_ops[feat].full_only || hd->full)
3160                 feat_ops[feat].print(&ff, hd->fp);
3161         else
3162                 fprintf(hd->fp, "# %s info available, use -I to display\n",
3163                         feat_ops[feat].name);
3164
3165         return 0;
3166 }
3167
3168 int perf_header__fprintf_info(struct perf_session *session, FILE *fp, bool full)
3169 {
3170         struct header_print_data hd;
3171         struct perf_header *header = &session->header;
3172         int fd = perf_data__fd(session->data);
3173         struct stat st;
3174         time_t stctime;
3175         int ret, bit;
3176
3177         hd.fp = fp;
3178         hd.full = full;
3179
3180         ret = fstat(fd, &st);
3181         if (ret == -1)
3182                 return -1;
3183
3184         stctime = st.st_mtime;
3185         fprintf(fp, "# captured on    : %s", ctime(&stctime));
3186
3187         fprintf(fp, "# header version : %u\n", header->version);
3188         fprintf(fp, "# data offset    : %" PRIu64 "\n", header->data_offset);
3189         fprintf(fp, "# data size      : %" PRIu64 "\n", header->data_size);
3190         fprintf(fp, "# feat offset    : %" PRIu64 "\n", header->feat_offset);
3191
3192         perf_header__process_sections(header, fd, &hd,
3193                                       perf_file_section__fprintf_info);
3194
3195         if (session->data->is_pipe)
3196                 return 0;
3197
3198         fprintf(fp, "# missing features: ");
3199         for_each_clear_bit(bit, header->adds_features, HEADER_LAST_FEATURE) {
3200                 if (bit)
3201                         fprintf(fp, "%s ", feat_ops[bit].name);
3202         }
3203
3204         fprintf(fp, "\n");
3205         return 0;
3206 }
3207
3208 static int do_write_feat(struct feat_fd *ff, int type,
3209                          struct perf_file_section **p,
3210                          struct evlist *evlist)
3211 {
3212         int err;
3213         int ret = 0;
3214
3215         if (perf_header__has_feat(ff->ph, type)) {
3216                 if (!feat_ops[type].write)
3217                         return -1;
3218
3219                 if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__))
3220                         return -1;
3221
3222                 (*p)->offset = lseek(ff->fd, 0, SEEK_CUR);
3223
3224                 err = feat_ops[type].write(ff, evlist);
3225                 if (err < 0) {
3226                         pr_debug("failed to write feature %s\n", feat_ops[type].name);
3227
3228                         /* undo anything written */
3229                         lseek(ff->fd, (*p)->offset, SEEK_SET);
3230
3231                         return -1;
3232                 }
3233                 (*p)->size = lseek(ff->fd, 0, SEEK_CUR) - (*p)->offset;
3234                 (*p)++;
3235         }
3236         return ret;
3237 }
3238
3239 static int perf_header__adds_write(struct perf_header *header,
3240                                    struct evlist *evlist, int fd)
3241 {
3242         int nr_sections;
3243         struct feat_fd ff;
3244         struct perf_file_section *feat_sec, *p;
3245         int sec_size;
3246         u64 sec_start;
3247         int feat;
3248         int err;
3249
3250         ff = (struct feat_fd){
3251                 .fd  = fd,
3252                 .ph = header,
3253         };
3254
3255         nr_sections = bitmap_weight(header->adds_features, HEADER_FEAT_BITS);
3256         if (!nr_sections)
3257                 return 0;
3258
3259         feat_sec = p = calloc(nr_sections, sizeof(*feat_sec));
3260         if (feat_sec == NULL)
3261                 return -ENOMEM;
3262
3263         sec_size = sizeof(*feat_sec) * nr_sections;
3264
3265         sec_start = header->feat_offset;
3266         lseek(fd, sec_start + sec_size, SEEK_SET);
3267
3268         for_each_set_bit(feat, header->adds_features, HEADER_FEAT_BITS) {
3269                 if (do_write_feat(&ff, feat, &p, evlist))
3270                         perf_header__clear_feat(header, feat);
3271         }
3272
3273         lseek(fd, sec_start, SEEK_SET);
3274         /*
3275          * may write more than needed due to dropped feature, but
3276          * this is okay, reader will skip the missing entries
3277          */
3278         err = do_write(&ff, feat_sec, sec_size);
3279         if (err < 0)
3280                 pr_debug("failed to write feature section\n");
3281         free(feat_sec);
3282         return err;
3283 }
3284
3285 int perf_header__write_pipe(int fd)
3286 {
3287         struct perf_pipe_file_header f_header;
3288         struct feat_fd ff;
3289         int err;
3290
3291         ff = (struct feat_fd){ .fd = fd };
3292
3293         f_header = (struct perf_pipe_file_header){
3294                 .magic     = PERF_MAGIC,
3295                 .size      = sizeof(f_header),
3296         };
3297
3298         err = do_write(&ff, &f_header, sizeof(f_header));
3299         if (err < 0) {
3300                 pr_debug("failed to write perf pipe header\n");
3301                 return err;
3302         }
3303
3304         return 0;
3305 }
3306
3307 int perf_session__write_header(struct perf_session *session,
3308                                struct evlist *evlist,
3309                                int fd, bool at_exit)
3310 {
3311         struct perf_file_header f_header;
3312         struct perf_file_attr   f_attr;
3313         struct perf_header *header = &session->header;
3314         struct evsel *evsel;
3315         struct feat_fd ff;
3316         u64 attr_offset;
3317         int err;
3318
3319         ff = (struct feat_fd){ .fd = fd};
3320         lseek(fd, sizeof(f_header), SEEK_SET);
3321
3322         evlist__for_each_entry(session->evlist, evsel) {
3323                 evsel->id_offset = lseek(fd, 0, SEEK_CUR);
3324                 err = do_write(&ff, evsel->core.id, evsel->core.ids * sizeof(u64));
3325                 if (err < 0) {
3326                         pr_debug("failed to write perf header\n");
3327                         return err;
3328                 }
3329         }
3330
3331         attr_offset = lseek(ff.fd, 0, SEEK_CUR);
3332
3333         evlist__for_each_entry(evlist, evsel) {
3334                 f_attr = (struct perf_file_attr){
3335                         .attr = evsel->core.attr,
3336                         .ids  = {
3337                                 .offset = evsel->id_offset,
3338                                 .size   = evsel->core.ids * sizeof(u64),
3339                         }
3340                 };
3341                 err = do_write(&ff, &f_attr, sizeof(f_attr));
3342                 if (err < 0) {
3343                         pr_debug("failed to write perf header attribute\n");
3344                         return err;
3345                 }
3346         }
3347
3348         if (!header->data_offset)
3349                 header->data_offset = lseek(fd, 0, SEEK_CUR);
3350         header->feat_offset = header->data_offset + header->data_size;
3351
3352         if (at_exit) {
3353                 err = perf_header__adds_write(header, evlist, fd);
3354                 if (err < 0)
3355                         return err;
3356         }
3357
3358         f_header = (struct perf_file_header){
3359                 .magic     = PERF_MAGIC,
3360                 .size      = sizeof(f_header),
3361                 .attr_size = sizeof(f_attr),
3362                 .attrs = {
3363                         .offset = attr_offset,
3364                         .size   = evlist->core.nr_entries * sizeof(f_attr),
3365                 },
3366                 .data = {
3367                         .offset = header->data_offset,
3368                         .size   = header->data_size,
3369                 },
3370                 /* event_types is ignored, store zeros */
3371         };
3372
3373         memcpy(&f_header.adds_features, &header->adds_features, sizeof(header->adds_features));
3374
3375         lseek(fd, 0, SEEK_SET);
3376         err = do_write(&ff, &f_header, sizeof(f_header));
3377         if (err < 0) {
3378                 pr_debug("failed to write perf header\n");
3379                 return err;
3380         }
3381         lseek(fd, header->data_offset + header->data_size, SEEK_SET);
3382
3383         return 0;
3384 }
3385
3386 static int perf_header__getbuffer64(struct perf_header *header,
3387                                     int fd, void *buf, size_t size)
3388 {
3389         if (readn(fd, buf, size) <= 0)
3390                 return -1;
3391
3392         if (header->needs_swap)
3393                 mem_bswap_64(buf, size);
3394
3395         return 0;
3396 }
3397
3398 int perf_header__process_sections(struct perf_header *header, int fd,
3399                                   void *data,
3400                                   int (*process)(struct perf_file_section *section,
3401                                                  struct perf_header *ph,
3402                                                  int feat, int fd, void *data))
3403 {
3404         struct perf_file_section *feat_sec, *sec;
3405         int nr_sections;
3406         int sec_size;
3407         int feat;
3408         int err;
3409
3410         nr_sections = bitmap_weight(header->adds_features, HEADER_FEAT_BITS);
3411         if (!nr_sections)
3412                 return 0;
3413
3414         feat_sec = sec = calloc(nr_sections, sizeof(*feat_sec));
3415         if (!feat_sec)
3416                 return -1;
3417
3418         sec_size = sizeof(*feat_sec) * nr_sections;
3419
3420         lseek(fd, header->feat_offset, SEEK_SET);
3421
3422         err = perf_header__getbuffer64(header, fd, feat_sec, sec_size);
3423         if (err < 0)
3424                 goto out_free;
3425
3426         for_each_set_bit(feat, header->adds_features, HEADER_LAST_FEATURE) {
3427                 err = process(sec++, header, feat, fd, data);
3428                 if (err < 0)
3429                         goto out_free;
3430         }
3431         err = 0;
3432 out_free:
3433         free(feat_sec);
3434         return err;
3435 }
3436
3437 static const int attr_file_abi_sizes[] = {
3438         [0] = PERF_ATTR_SIZE_VER0,
3439         [1] = PERF_ATTR_SIZE_VER1,
3440         [2] = PERF_ATTR_SIZE_VER2,
3441         [3] = PERF_ATTR_SIZE_VER3,
3442         [4] = PERF_ATTR_SIZE_VER4,
3443         0,
3444 };
3445
3446 /*
3447  * In the legacy file format, the magic number is not used to encode endianness.
3448  * hdr_sz was used to encode endianness. But given that hdr_sz can vary based
3449  * on ABI revisions, we need to try all combinations for all endianness to
3450  * detect the endianness.
3451  */
3452 static int try_all_file_abis(uint64_t hdr_sz, struct perf_header *ph)
3453 {
3454         uint64_t ref_size, attr_size;
3455         int i;
3456
3457         for (i = 0 ; attr_file_abi_sizes[i]; i++) {
3458                 ref_size = attr_file_abi_sizes[i]
3459                          + sizeof(struct perf_file_section);
3460                 if (hdr_sz != ref_size) {
3461                         attr_size = bswap_64(hdr_sz);
3462                         if (attr_size != ref_size)
3463                                 continue;
3464
3465                         ph->needs_swap = true;
3466                 }
3467                 pr_debug("ABI%d perf.data file detected, need_swap=%d\n",
3468                          i,
3469                          ph->needs_swap);
3470                 return 0;
3471         }
3472         /* could not determine endianness */
3473         return -1;
3474 }
3475
3476 #define PERF_PIPE_HDR_VER0      16
3477
3478 static const size_t attr_pipe_abi_sizes[] = {
3479         [0] = PERF_PIPE_HDR_VER0,
3480         0,
3481 };
3482
3483 /*
3484  * In the legacy pipe format, there is an implicit assumption that endiannesss
3485  * between host recording the samples, and host parsing the samples is the
3486  * same. This is not always the case given that the pipe output may always be
3487  * redirected into a file and analyzed on a different machine with possibly a
3488  * different endianness and perf_event ABI revsions in the perf tool itself.
3489  */
3490 static int try_all_pipe_abis(uint64_t hdr_sz, struct perf_header *ph)
3491 {
3492         u64 attr_size;
3493         int i;
3494
3495         for (i = 0 ; attr_pipe_abi_sizes[i]; i++) {
3496                 if (hdr_sz != attr_pipe_abi_sizes[i]) {
3497                         attr_size = bswap_64(hdr_sz);
3498                         if (attr_size != hdr_sz)
3499                                 continue;
3500
3501                         ph->needs_swap = true;
3502                 }
3503                 pr_debug("Pipe ABI%d perf.data file detected\n", i);
3504                 return 0;
3505         }
3506         return -1;
3507 }
3508
3509 bool is_perf_magic(u64 magic)
3510 {
3511         if (!memcmp(&magic, __perf_magic1, sizeof(magic))
3512                 || magic == __perf_magic2
3513                 || magic == __perf_magic2_sw)
3514                 return true;
3515
3516         return false;
3517 }
3518
3519 static int check_magic_endian(u64 magic, uint64_t hdr_sz,
3520                               bool is_pipe, struct perf_header *ph)
3521 {
3522         int ret;
3523
3524         /* check for legacy format */
3525         ret = memcmp(&magic, __perf_magic1, sizeof(magic));
3526         if (ret == 0) {
3527                 ph->version = PERF_HEADER_VERSION_1;
3528                 pr_debug("legacy perf.data format\n");
3529                 if (is_pipe)
3530                         return try_all_pipe_abis(hdr_sz, ph);
3531
3532                 return try_all_file_abis(hdr_sz, ph);
3533         }
3534         /*
3535          * the new magic number serves two purposes:
3536          * - unique number to identify actual perf.data files
3537          * - encode endianness of file
3538          */
3539         ph->version = PERF_HEADER_VERSION_2;
3540
3541         /* check magic number with one endianness */
3542         if (magic == __perf_magic2)
3543                 return 0;
3544
3545         /* check magic number with opposite endianness */
3546         if (magic != __perf_magic2_sw)
3547                 return -1;
3548
3549         ph->needs_swap = true;
3550
3551         return 0;
3552 }
3553
3554 int perf_file_header__read(struct perf_file_header *header,
3555                            struct perf_header *ph, int fd)
3556 {
3557         ssize_t ret;
3558
3559         lseek(fd, 0, SEEK_SET);
3560
3561         ret = readn(fd, header, sizeof(*header));
3562         if (ret <= 0)
3563                 return -1;
3564
3565         if (check_magic_endian(header->magic,
3566                                header->attr_size, false, ph) < 0) {
3567                 pr_debug("magic/endian check failed\n");
3568                 return -1;
3569         }
3570
3571         if (ph->needs_swap) {
3572                 mem_bswap_64(header, offsetof(struct perf_file_header,
3573                              adds_features));
3574         }
3575
3576         if (header->size != sizeof(*header)) {
3577                 /* Support the previous format */
3578                 if (header->size == offsetof(typeof(*header), adds_features))
3579                         bitmap_zero(header->adds_features, HEADER_FEAT_BITS);
3580                 else
3581                         return -1;
3582         } else if (ph->needs_swap) {
3583                 /*
3584                  * feature bitmap is declared as an array of unsigned longs --
3585                  * not good since its size can differ between the host that
3586                  * generated the data file and the host analyzing the file.
3587                  *
3588                  * We need to handle endianness, but we don't know the size of
3589                  * the unsigned long where the file was generated. Take a best
3590                  * guess at determining it: try 64-bit swap first (ie., file
3591                  * created on a 64-bit host), and check if the hostname feature
3592                  * bit is set (this feature bit is forced on as of fbe96f2).
3593                  * If the bit is not, undo the 64-bit swap and try a 32-bit
3594                  * swap. If the hostname bit is still not set (e.g., older data
3595                  * file), punt and fallback to the original behavior --
3596                  * clearing all feature bits and setting buildid.
3597                  */
3598                 mem_bswap_64(&header->adds_features,
3599                             BITS_TO_U64(HEADER_FEAT_BITS));
3600
3601                 if (!test_bit(HEADER_HOSTNAME, header->adds_features)) {
3602                         /* unswap as u64 */
3603                         mem_bswap_64(&header->adds_features,
3604                                     BITS_TO_U64(HEADER_FEAT_BITS));
3605
3606                         /* unswap as u32 */
3607                         mem_bswap_32(&header->adds_features,
3608                                     BITS_TO_U32(HEADER_FEAT_BITS));
3609                 }
3610
3611                 if (!test_bit(HEADER_HOSTNAME, header->adds_features)) {
3612                         bitmap_zero(header->adds_features, HEADER_FEAT_BITS);
3613                         set_bit(HEADER_BUILD_ID, header->adds_features);
3614                 }
3615         }
3616
3617         memcpy(&ph->adds_features, &header->adds_features,
3618                sizeof(ph->adds_features));
3619
3620         ph->data_offset  = header->data.offset;
3621         ph->data_size    = header->data.size;
3622         ph->feat_offset  = header->data.offset + header->data.size;
3623         return 0;
3624 }
3625
3626 static int perf_file_section__process(struct perf_file_section *section,
3627                                       struct perf_header *ph,
3628                                       int feat, int fd, void *data)
3629 {
3630         struct feat_fd fdd = {
3631                 .fd     = fd,
3632                 .ph     = ph,
3633                 .size   = section->size,
3634                 .offset = section->offset,
3635         };
3636
3637         if (lseek(fd, section->offset, SEEK_SET) == (off_t)-1) {
3638                 pr_debug("Failed to lseek to %" PRIu64 " offset for feature "
3639                           "%d, continuing...\n", section->offset, feat);
3640                 return 0;
3641         }
3642
3643         if (feat >= HEADER_LAST_FEATURE) {
3644                 pr_debug("unknown feature %d, continuing...\n", feat);
3645                 return 0;
3646         }
3647
3648         if (!feat_ops[feat].process)
3649                 return 0;
3650
3651         return feat_ops[feat].process(&fdd, data);
3652 }
3653
3654 static int perf_file_header__read_pipe(struct perf_pipe_file_header *header,
3655                                        struct perf_header *ph, int fd,
3656                                        bool repipe)
3657 {
3658         struct feat_fd ff = {
3659                 .fd = STDOUT_FILENO,
3660                 .ph = ph,
3661         };
3662         ssize_t ret;
3663
3664         ret = readn(fd, header, sizeof(*header));
3665         if (ret <= 0)
3666                 return -1;
3667
3668         if (check_magic_endian(header->magic, header->size, true, ph) < 0) {
3669                 pr_debug("endian/magic failed\n");
3670                 return -1;
3671         }
3672
3673         if (ph->needs_swap)
3674                 header->size = bswap_64(header->size);
3675
3676         if (repipe && do_write(&ff, header, sizeof(*header)) < 0)
3677                 return -1;
3678
3679         return 0;
3680 }
3681
3682 static int perf_header__read_pipe(struct perf_session *session)
3683 {
3684         struct perf_header *header = &session->header;
3685         struct perf_pipe_file_header f_header;
3686
3687         if (perf_file_header__read_pipe(&f_header, header,
3688                                         perf_data__fd(session->data),
3689                                         session->repipe) < 0) {
3690                 pr_debug("incompatible file format\n");
3691                 return -EINVAL;
3692         }
3693
3694         return f_header.size == sizeof(f_header) ? 0 : -1;
3695 }
3696
3697 static int read_attr(int fd, struct perf_header *ph,
3698                      struct perf_file_attr *f_attr)
3699 {
3700         struct perf_event_attr *attr = &f_attr->attr;
3701         size_t sz, left;
3702         size_t our_sz = sizeof(f_attr->attr);
3703         ssize_t ret;
3704
3705         memset(f_attr, 0, sizeof(*f_attr));
3706
3707         /* read minimal guaranteed structure */
3708         ret = readn(fd, attr, PERF_ATTR_SIZE_VER0);
3709         if (ret <= 0) {
3710                 pr_debug("cannot read %d bytes of header attr\n",
3711                          PERF_ATTR_SIZE_VER0);
3712                 return -1;
3713         }
3714
3715         /* on file perf_event_attr size */
3716         sz = attr->size;
3717
3718         if (ph->needs_swap)
3719                 sz = bswap_32(sz);
3720
3721         if (sz == 0) {
3722                 /* assume ABI0 */
3723                 sz =  PERF_ATTR_SIZE_VER0;
3724         } else if (sz > our_sz) {
3725                 pr_debug("file uses a more recent and unsupported ABI"
3726                          " (%zu bytes extra)\n", sz - our_sz);
3727                 return -1;
3728         }
3729         /* what we have not yet read and that we know about */
3730         left = sz - PERF_ATTR_SIZE_VER0;
3731         if (left) {
3732                 void *ptr = attr;
3733                 ptr += PERF_ATTR_SIZE_VER0;
3734
3735                 ret = readn(fd, ptr, left);
3736         }
3737         /* read perf_file_section, ids are read in caller */
3738         ret = readn(fd, &f_attr->ids, sizeof(f_attr->ids));
3739
3740         return ret <= 0 ? -1 : 0;
3741 }
3742
3743 static int perf_evsel__prepare_tracepoint_event(struct evsel *evsel,
3744                                                 struct tep_handle *pevent)
3745 {
3746         struct tep_event *event;
3747         char bf[128];
3748
3749         /* already prepared */
3750         if (evsel->tp_format)
3751                 return 0;
3752
3753         if (pevent == NULL) {
3754                 pr_debug("broken or missing trace data\n");
3755                 return -1;
3756         }
3757
3758         event = tep_find_event(pevent, evsel->core.attr.config);
3759         if (event == NULL) {
3760                 pr_debug("cannot find event format for %d\n", (int)evsel->core.attr.config);
3761                 return -1;
3762         }
3763
3764         if (!evsel->name) {
3765                 snprintf(bf, sizeof(bf), "%s:%s", event->system, event->name);
3766                 evsel->name = strdup(bf);
3767                 if (evsel->name == NULL)
3768                         return -1;
3769         }
3770
3771         evsel->tp_format = event;
3772         return 0;
3773 }
3774
3775 static int perf_evlist__prepare_tracepoint_events(struct evlist *evlist,
3776                                                   struct tep_handle *pevent)
3777 {
3778         struct evsel *pos;
3779
3780         evlist__for_each_entry(evlist, pos) {
3781                 if (pos->core.attr.type == PERF_TYPE_TRACEPOINT &&
3782                     perf_evsel__prepare_tracepoint_event(pos, pevent))
3783                         return -1;
3784         }
3785
3786         return 0;
3787 }
3788
3789 int perf_session__read_header(struct perf_session *session)
3790 {
3791         struct perf_data *data = session->data;
3792         struct perf_header *header = &session->header;
3793         struct perf_file_header f_header;
3794         struct perf_file_attr   f_attr;
3795         u64                     f_id;
3796         int nr_attrs, nr_ids, i, j, err;
3797         int fd = perf_data__fd(data);
3798
3799         session->evlist = evlist__new();
3800         if (session->evlist == NULL)
3801                 return -ENOMEM;
3802
3803         session->evlist->env = &header->env;
3804         session->machines.host.env = &header->env;
3805
3806         /*
3807          * We can read 'pipe' data event from regular file,
3808          * check for the pipe header regardless of source.
3809          */
3810         err = perf_header__read_pipe(session);
3811         if (!err || (err && perf_data__is_pipe(data))) {
3812                 data->is_pipe = true;
3813                 return err;
3814         }
3815
3816         if (perf_file_header__read(&f_header, header, fd) < 0)
3817                 return -EINVAL;
3818
3819         /*
3820          * Sanity check that perf.data was written cleanly; data size is
3821          * initialized to 0 and updated only if the on_exit function is run.
3822          * If data size is still 0 then the file contains only partial
3823          * information.  Just warn user and process it as much as it can.
3824          */
3825         if (f_header.data.size == 0) {
3826                 pr_warning("WARNING: The %s file's data size field is 0 which is unexpected.\n"
3827                            "Was the 'perf record' command properly terminated?\n",
3828                            data->file.path);
3829         }
3830
3831         if (f_header.attr_size == 0) {
3832                 pr_err("ERROR: The %s file's attr size field is 0 which is unexpected.\n"
3833                        "Was the 'perf record' command properly terminated?\n",
3834                        data->file.path);
3835                 return -EINVAL;
3836         }
3837
3838         nr_attrs = f_header.attrs.size / f_header.attr_size;
3839         lseek(fd, f_header.attrs.offset, SEEK_SET);
3840
3841         for (i = 0; i < nr_attrs; i++) {
3842                 struct evsel *evsel;
3843                 off_t tmp;
3844
3845                 if (read_attr(fd, header, &f_attr) < 0)
3846                         goto out_errno;
3847
3848                 if (header->needs_swap) {
3849                         f_attr.ids.size   = bswap_64(f_attr.ids.size);
3850                         f_attr.ids.offset = bswap_64(f_attr.ids.offset);
3851                         perf_event__attr_swap(&f_attr.attr);
3852                 }
3853
3854                 tmp = lseek(fd, 0, SEEK_CUR);
3855                 evsel = evsel__new(&f_attr.attr);
3856
3857                 if (evsel == NULL)
3858                         goto out_delete_evlist;
3859
3860                 evsel->needs_swap = header->needs_swap;
3861                 /*
3862                  * Do it before so that if perf_evsel__alloc_id fails, this
3863                  * entry gets purged too at evlist__delete().
3864                  */
3865                 evlist__add(session->evlist, evsel);
3866
3867                 nr_ids = f_attr.ids.size / sizeof(u64);
3868                 /*
3869                  * We don't have the cpu and thread maps on the header, so
3870                  * for allocating the perf_sample_id table we fake 1 cpu and
3871                  * hattr->ids threads.
3872                  */
3873                 if (perf_evsel__alloc_id(&evsel->core, 1, nr_ids))
3874                         goto out_delete_evlist;
3875
3876                 lseek(fd, f_attr.ids.offset, SEEK_SET);
3877
3878                 for (j = 0; j < nr_ids; j++) {
3879                         if (perf_header__getbuffer64(header, fd, &f_id, sizeof(f_id)))
3880                                 goto out_errno;
3881
3882                         perf_evlist__id_add(&session->evlist->core, &evsel->core, 0, j, f_id);
3883                 }
3884
3885                 lseek(fd, tmp, SEEK_SET);
3886         }
3887
3888         perf_header__process_sections(header, fd, &session->tevent,
3889                                       perf_file_section__process);
3890
3891         if (perf_evlist__prepare_tracepoint_events(session->evlist,
3892                                                    session->tevent.pevent))
3893                 goto out_delete_evlist;
3894
3895         return 0;
3896 out_errno:
3897         return -errno;
3898
3899 out_delete_evlist:
3900         evlist__delete(session->evlist);
3901         session->evlist = NULL;
3902         return -ENOMEM;
3903 }
3904
3905 int perf_event__process_feature(struct perf_session *session,
3906                                 union perf_event *event)
3907 {
3908         struct perf_tool *tool = session->tool;
3909         struct feat_fd ff = { .fd = 0 };
3910         struct perf_record_header_feature *fe = (struct perf_record_header_feature *)event;
3911         int type = fe->header.type;
3912         u64 feat = fe->feat_id;
3913
3914         if (type < 0 || type >= PERF_RECORD_HEADER_MAX) {
3915                 pr_warning("invalid record type %d in pipe-mode\n", type);
3916                 return 0;
3917         }
3918         if (feat == HEADER_RESERVED || feat >= HEADER_LAST_FEATURE) {
3919                 pr_warning("invalid record type %d in pipe-mode\n", type);
3920                 return -1;
3921         }
3922
3923         if (!feat_ops[feat].process)
3924                 return 0;
3925
3926         ff.buf  = (void *)fe->data;
3927         ff.size = event->header.size - sizeof(*fe);
3928         ff.ph = &session->header;
3929
3930         if (feat_ops[feat].process(&ff, NULL))
3931                 return -1;
3932
3933         if (!feat_ops[feat].print || !tool->show_feat_hdr)
3934                 return 0;
3935
3936         if (!feat_ops[feat].full_only ||
3937             tool->show_feat_hdr >= SHOW_FEAT_HEADER_FULL_INFO) {
3938                 feat_ops[feat].print(&ff, stdout);
3939         } else {
3940                 fprintf(stdout, "# %s info available, use -I to display\n",
3941                         feat_ops[feat].name);
3942         }
3943
3944         return 0;
3945 }
3946
3947 size_t perf_event__fprintf_event_update(union perf_event *event, FILE *fp)
3948 {
3949         struct perf_record_event_update *ev = &event->event_update;
3950         struct perf_record_event_update_scale *ev_scale;
3951         struct perf_record_event_update_cpus *ev_cpus;
3952         struct perf_cpu_map *map;
3953         size_t ret;
3954
3955         ret = fprintf(fp, "\n... id:    %" PRI_lu64 "\n", ev->id);
3956
3957         switch (ev->type) {
3958         case PERF_EVENT_UPDATE__SCALE:
3959                 ev_scale = (struct perf_record_event_update_scale *)ev->data;
3960                 ret += fprintf(fp, "... scale: %f\n", ev_scale->scale);
3961                 break;
3962         case PERF_EVENT_UPDATE__UNIT:
3963                 ret += fprintf(fp, "... unit:  %s\n", ev->data);
3964                 break;
3965         case PERF_EVENT_UPDATE__NAME:
3966                 ret += fprintf(fp, "... name:  %s\n", ev->data);
3967                 break;
3968         case PERF_EVENT_UPDATE__CPUS:
3969                 ev_cpus = (struct perf_record_event_update_cpus *)ev->data;
3970                 ret += fprintf(fp, "... ");
3971
3972                 map = cpu_map__new_data(&ev_cpus->cpus);
3973                 if (map)
3974                         ret += cpu_map__fprintf(map, fp);
3975                 else
3976                         ret += fprintf(fp, "failed to get cpus\n");
3977                 break;
3978         default:
3979                 ret += fprintf(fp, "... unknown type\n");
3980                 break;
3981         }
3982
3983         return ret;
3984 }
3985
3986 int perf_event__process_attr(struct perf_tool *tool __maybe_unused,
3987                              union perf_event *event,
3988                              struct evlist **pevlist)
3989 {
3990         u32 i, n_ids;
3991         u64 *ids;
3992         struct evsel *evsel;
3993         struct evlist *evlist = *pevlist;
3994
3995         if (evlist == NULL) {
3996                 *pevlist = evlist = evlist__new();
3997                 if (evlist == NULL)
3998                         return -ENOMEM;
3999         }
4000
4001         evsel = evsel__new(&event->attr.attr);
4002         if (evsel == NULL)
4003                 return -ENOMEM;
4004
4005         evlist__add(evlist, evsel);
4006
4007         n_ids = event->header.size - sizeof(event->header) - event->attr.attr.size;
4008         n_ids = n_ids / sizeof(u64);
4009         /*
4010          * We don't have the cpu and thread maps on the header, so
4011          * for allocating the perf_sample_id table we fake 1 cpu and
4012          * hattr->ids threads.
4013          */
4014         if (perf_evsel__alloc_id(&evsel->core, 1, n_ids))
4015                 return -ENOMEM;
4016
4017         ids = (void *)&event->attr.attr + event->attr.attr.size;
4018         for (i = 0; i < n_ids; i++) {
4019                 perf_evlist__id_add(&evlist->core, &evsel->core, 0, i, ids[i]);
4020         }
4021
4022         return 0;
4023 }
4024
4025 int perf_event__process_event_update(struct perf_tool *tool __maybe_unused,
4026                                      union perf_event *event,
4027                                      struct evlist **pevlist)
4028 {
4029         struct perf_record_event_update *ev = &event->event_update;
4030         struct perf_record_event_update_scale *ev_scale;
4031         struct perf_record_event_update_cpus *ev_cpus;
4032         struct evlist *evlist;
4033         struct evsel *evsel;
4034         struct perf_cpu_map *map;
4035
4036         if (!pevlist || *pevlist == NULL)
4037                 return -EINVAL;
4038
4039         evlist = *pevlist;
4040
4041         evsel = perf_evlist__id2evsel(evlist, ev->id);
4042         if (evsel == NULL)
4043                 return -EINVAL;
4044
4045         switch (ev->type) {
4046         case PERF_EVENT_UPDATE__UNIT:
4047                 evsel->unit = strdup(ev->data);
4048                 break;
4049         case PERF_EVENT_UPDATE__NAME:
4050                 evsel->name = strdup(ev->data);
4051                 break;
4052         case PERF_EVENT_UPDATE__SCALE:
4053                 ev_scale = (struct perf_record_event_update_scale *)ev->data;
4054                 evsel->scale = ev_scale->scale;
4055                 break;
4056         case PERF_EVENT_UPDATE__CPUS:
4057                 ev_cpus = (struct perf_record_event_update_cpus *)ev->data;
4058
4059                 map = cpu_map__new_data(&ev_cpus->cpus);
4060                 if (map)
4061                         evsel->core.own_cpus = map;
4062                 else
4063                         pr_err("failed to get event_update cpus\n");
4064         default:
4065                 break;
4066         }
4067
4068         return 0;
4069 }
4070
4071 int perf_event__process_tracing_data(struct perf_session *session,
4072                                      union perf_event *event)
4073 {
4074         ssize_t size_read, padding, size = event->tracing_data.size;
4075         int fd = perf_data__fd(session->data);
4076         char buf[BUFSIZ];
4077
4078         /*
4079          * The pipe fd is already in proper place and in any case
4080          * we can't move it, and we'd screw the case where we read
4081          * 'pipe' data from regular file. The trace_report reads
4082          * data from 'fd' so we need to set it directly behind the
4083          * event, where the tracing data starts.
4084          */
4085         if (!perf_data__is_pipe(session->data)) {
4086                 off_t offset = lseek(fd, 0, SEEK_CUR);
4087
4088                 /* setup for reading amidst mmap */
4089                 lseek(fd, offset + sizeof(struct perf_record_header_tracing_data),
4090                       SEEK_SET);
4091         }
4092
4093         size_read = trace_report(fd, &session->tevent,
4094                                  session->repipe);
4095         padding = PERF_ALIGN(size_read, sizeof(u64)) - size_read;
4096
4097         if (readn(fd, buf, padding) < 0) {
4098                 pr_err("%s: reading input file", __func__);
4099                 return -1;
4100         }
4101         if (session->repipe) {
4102                 int retw = write(STDOUT_FILENO, buf, padding);
4103                 if (retw <= 0 || retw != padding) {
4104                         pr_err("%s: repiping tracing data padding", __func__);
4105                         return -1;
4106                 }
4107         }
4108
4109         if (size_read + padding != size) {
4110                 pr_err("%s: tracing data size mismatch", __func__);
4111                 return -1;
4112         }
4113
4114         perf_evlist__prepare_tracepoint_events(session->evlist,
4115                                                session->tevent.pevent);
4116
4117         return size_read + padding;
4118 }
4119
4120 int perf_event__process_build_id(struct perf_session *session,
4121                                  union perf_event *event)
4122 {
4123         __event_process_build_id(&event->build_id,
4124                                  event->build_id.filename,
4125                                  session);
4126         return 0;
4127 }