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 <regex.h> 10 #include <stdio.h> 11 #include <stdlib.h> 12 #include <linux/compiler.h> 13 #include <linux/list.h> 14 #include <linux/kernel.h> 15 #include <linux/bitops.h> 16 #include <linux/string.h> 17 #include <linux/stringify.h> 18 #include <linux/zalloc.h> 19 #include <sys/stat.h> 20 #include <sys/utsname.h> 21 #include <linux/time64.h> 22 #include <dirent.h> 23 #ifdef HAVE_LIBBPF_SUPPORT 24 #include <bpf/libbpf.h> 25 #endif 26 #include <perf/cpumap.h> 27 #include <tools/libc_compat.h> // reallocarray 28 29 #include "dso.h" 30 #include "evlist.h" 31 #include "evsel.h" 32 #include "util/evsel_fprintf.h" 33 #include "header.h" 34 #include "memswap.h" 35 #include "trace-event.h" 36 #include "session.h" 37 #include "symbol.h" 38 #include "debug.h" 39 #include "cpumap.h" 40 #include "pmu.h" 41 #include "pmus.h" 42 #include "vdso.h" 43 #include "strbuf.h" 44 #include "build-id.h" 45 #include "data.h" 46 #include <api/fs/fs.h> 47 #include "asm/bug.h" 48 #include "tool.h" 49 #include "time-utils.h" 50 #include "units.h" 51 #include "util/util.h" // perf_exe() 52 #include "cputopo.h" 53 #include "bpf-event.h" 54 #include "bpf-utils.h" 55 #include "clockid.h" 56 57 #include <linux/ctype.h> 58 #include <internal/lib.h> 59 60 #ifdef HAVE_LIBTRACEEVENT 61 #include <traceevent/event-parse.h> 62 #endif 63 64 /* 65 * magic2 = "PERFILE2" 66 * must be a numerical value to let the endianness 67 * determine the memory layout. That way we are able 68 * to detect endianness when reading the perf.data file 69 * back. 70 * 71 * we check for legacy (PERFFILE) format. 72 */ 73 static const char *__perf_magic1 = "PERFFILE"; 74 static const u64 __perf_magic2 = 0x32454c4946524550ULL; 75 static const u64 __perf_magic2_sw = 0x50455246494c4532ULL; 76 77 #define PERF_MAGIC __perf_magic2 78 79 const char perf_version_string[] = PERF_VERSION; 80 81 struct perf_file_attr { 82 struct perf_event_attr attr; 83 struct perf_file_section ids; 84 }; 85 86 void perf_header__set_feat(struct perf_header *header, int feat) 87 { 88 __set_bit(feat, header->adds_features); 89 } 90 91 void perf_header__clear_feat(struct perf_header *header, int feat) 92 { 93 __clear_bit(feat, header->adds_features); 94 } 95 96 bool perf_header__has_feat(const struct perf_header *header, int feat) 97 { 98 return test_bit(feat, header->adds_features); 99 } 100 101 static int __do_write_fd(struct feat_fd *ff, const void *buf, size_t size) 102 { 103 ssize_t ret = writen(ff->fd, buf, size); 104 105 if (ret != (ssize_t)size) 106 return ret < 0 ? (int)ret : -1; 107 return 0; 108 } 109 110 static int __do_write_buf(struct feat_fd *ff, const void *buf, size_t size) 111 { 112 /* struct perf_event_header::size is u16 */ 113 const size_t max_size = 0xffff - sizeof(struct perf_event_header); 114 size_t new_size = ff->size; 115 void *addr; 116 117 if (size + ff->offset > max_size) 118 return -E2BIG; 119 120 while (size > (new_size - ff->offset)) 121 new_size <<= 1; 122 new_size = min(max_size, new_size); 123 124 if (ff->size < new_size) { 125 addr = realloc(ff->buf, new_size); 126 if (!addr) 127 return -ENOMEM; 128 ff->buf = addr; 129 ff->size = new_size; 130 } 131 132 memcpy(ff->buf + ff->offset, buf, size); 133 ff->offset += size; 134 135 return 0; 136 } 137 138 /* Return: 0 if succeeded, -ERR if failed. */ 139 int do_write(struct feat_fd *ff, const void *buf, size_t size) 140 { 141 if (!ff->buf) 142 return __do_write_fd(ff, buf, size); 143 return __do_write_buf(ff, buf, size); 144 } 145 146 /* Return: 0 if succeeded, -ERR if failed. */ 147 static int do_write_bitmap(struct feat_fd *ff, unsigned long *set, u64 size) 148 { 149 u64 *p = (u64 *) set; 150 int i, ret; 151 152 ret = do_write(ff, &size, sizeof(size)); 153 if (ret < 0) 154 return ret; 155 156 for (i = 0; (u64) i < BITS_TO_U64(size); i++) { 157 ret = do_write(ff, p + i, sizeof(*p)); 158 if (ret < 0) 159 return ret; 160 } 161 162 return 0; 163 } 164 165 /* Return: 0 if succeeded, -ERR if failed. */ 166 int write_padded(struct feat_fd *ff, const void *bf, 167 size_t count, size_t count_aligned) 168 { 169 static const char zero_buf[NAME_ALIGN]; 170 int err = do_write(ff, bf, count); 171 172 if (!err) 173 err = do_write(ff, zero_buf, count_aligned - count); 174 175 return err; 176 } 177 178 #define string_size(str) \ 179 (PERF_ALIGN((strlen(str) + 1), NAME_ALIGN) + sizeof(u32)) 180 181 /* Return: 0 if succeeded, -ERR if failed. */ 182 static int do_write_string(struct feat_fd *ff, const char *str) 183 { 184 u32 len, olen; 185 int ret; 186 187 olen = strlen(str) + 1; 188 len = PERF_ALIGN(olen, NAME_ALIGN); 189 190 /* write len, incl. \0 */ 191 ret = do_write(ff, &len, sizeof(len)); 192 if (ret < 0) 193 return ret; 194 195 return write_padded(ff, str, olen, len); 196 } 197 198 static int __do_read_fd(struct feat_fd *ff, void *addr, ssize_t size) 199 { 200 ssize_t ret = readn(ff->fd, addr, size); 201 202 if (ret != size) 203 return ret < 0 ? (int)ret : -1; 204 return 0; 205 } 206 207 static int __do_read_buf(struct feat_fd *ff, void *addr, ssize_t size) 208 { 209 if (size > (ssize_t)ff->size - ff->offset) 210 return -1; 211 212 memcpy(addr, ff->buf + ff->offset, size); 213 ff->offset += size; 214 215 return 0; 216 217 } 218 219 static int __do_read(struct feat_fd *ff, void *addr, ssize_t size) 220 { 221 if (!ff->buf) 222 return __do_read_fd(ff, addr, size); 223 return __do_read_buf(ff, addr, size); 224 } 225 226 static int do_read_u32(struct feat_fd *ff, u32 *addr) 227 { 228 int ret; 229 230 ret = __do_read(ff, addr, sizeof(*addr)); 231 if (ret) 232 return ret; 233 234 if (ff->ph->needs_swap) 235 *addr = bswap_32(*addr); 236 return 0; 237 } 238 239 static int do_read_u64(struct feat_fd *ff, u64 *addr) 240 { 241 int ret; 242 243 ret = __do_read(ff, addr, sizeof(*addr)); 244 if (ret) 245 return ret; 246 247 if (ff->ph->needs_swap) 248 *addr = bswap_64(*addr); 249 return 0; 250 } 251 252 static char *do_read_string(struct feat_fd *ff) 253 { 254 u32 len; 255 char *buf; 256 257 if (do_read_u32(ff, &len)) 258 return NULL; 259 260 buf = malloc(len); 261 if (!buf) 262 return NULL; 263 264 if (!__do_read(ff, buf, len)) { 265 /* 266 * strings are padded by zeroes 267 * thus the actual strlen of buf 268 * may be less than len 269 */ 270 return buf; 271 } 272 273 free(buf); 274 return NULL; 275 } 276 277 /* Return: 0 if succeeded, -ERR if failed. */ 278 static int do_read_bitmap(struct feat_fd *ff, unsigned long **pset, u64 *psize) 279 { 280 unsigned long *set; 281 u64 size, *p; 282 int i, ret; 283 284 ret = do_read_u64(ff, &size); 285 if (ret) 286 return ret; 287 288 set = bitmap_zalloc(size); 289 if (!set) 290 return -ENOMEM; 291 292 p = (u64 *) set; 293 294 for (i = 0; (u64) i < BITS_TO_U64(size); i++) { 295 ret = do_read_u64(ff, p + i); 296 if (ret < 0) { 297 free(set); 298 return ret; 299 } 300 } 301 302 *pset = set; 303 *psize = size; 304 return 0; 305 } 306 307 #ifdef HAVE_LIBTRACEEVENT 308 static int write_tracing_data(struct feat_fd *ff, 309 struct evlist *evlist) 310 { 311 if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__)) 312 return -1; 313 314 return read_tracing_data(ff->fd, &evlist->core.entries); 315 } 316 #endif 317 318 static int write_build_id(struct feat_fd *ff, 319 struct evlist *evlist __maybe_unused) 320 { 321 struct perf_session *session; 322 int err; 323 324 session = container_of(ff->ph, struct perf_session, header); 325 326 if (!perf_session__read_build_ids(session, true)) 327 return -1; 328 329 if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__)) 330 return -1; 331 332 err = perf_session__write_buildid_table(session, ff); 333 if (err < 0) { 334 pr_debug("failed to write buildid table\n"); 335 return err; 336 } 337 perf_session__cache_build_ids(session); 338 339 return 0; 340 } 341 342 static int write_hostname(struct feat_fd *ff, 343 struct evlist *evlist __maybe_unused) 344 { 345 struct utsname uts; 346 int ret; 347 348 ret = uname(&uts); 349 if (ret < 0) 350 return -1; 351 352 return do_write_string(ff, uts.nodename); 353 } 354 355 static int write_osrelease(struct feat_fd *ff, 356 struct evlist *evlist __maybe_unused) 357 { 358 struct utsname uts; 359 int ret; 360 361 ret = uname(&uts); 362 if (ret < 0) 363 return -1; 364 365 return do_write_string(ff, uts.release); 366 } 367 368 static int write_arch(struct feat_fd *ff, 369 struct evlist *evlist __maybe_unused) 370 { 371 struct utsname uts; 372 int ret; 373 374 ret = uname(&uts); 375 if (ret < 0) 376 return -1; 377 378 return do_write_string(ff, uts.machine); 379 } 380 381 static int write_version(struct feat_fd *ff, 382 struct evlist *evlist __maybe_unused) 383 { 384 return do_write_string(ff, perf_version_string); 385 } 386 387 static int __write_cpudesc(struct feat_fd *ff, const char *cpuinfo_proc) 388 { 389 FILE *file; 390 char *buf = NULL; 391 char *s, *p; 392 const char *search = cpuinfo_proc; 393 size_t len = 0; 394 int ret = -1; 395 396 if (!search) 397 return -1; 398 399 file = fopen("/proc/cpuinfo", "r"); 400 if (!file) 401 return -1; 402 403 while (getline(&buf, &len, file) > 0) { 404 ret = strncmp(buf, search, strlen(search)); 405 if (!ret) 406 break; 407 } 408 409 if (ret) { 410 ret = -1; 411 goto done; 412 } 413 414 s = buf; 415 416 p = strchr(buf, ':'); 417 if (p && *(p+1) == ' ' && *(p+2)) 418 s = p + 2; 419 p = strchr(s, '\n'); 420 if (p) 421 *p = '\0'; 422 423 /* squash extra space characters (branding string) */ 424 p = s; 425 while (*p) { 426 if (isspace(*p)) { 427 char *r = p + 1; 428 char *q = skip_spaces(r); 429 *p = ' '; 430 if (q != (p+1)) 431 while ((*r++ = *q++)); 432 } 433 p++; 434 } 435 ret = do_write_string(ff, s); 436 done: 437 free(buf); 438 fclose(file); 439 return ret; 440 } 441 442 static int write_cpudesc(struct feat_fd *ff, 443 struct evlist *evlist __maybe_unused) 444 { 445 #if defined(__powerpc__) || defined(__hppa__) || defined(__sparc__) 446 #define CPUINFO_PROC { "cpu", } 447 #elif defined(__s390__) 448 #define CPUINFO_PROC { "vendor_id", } 449 #elif defined(__sh__) 450 #define CPUINFO_PROC { "cpu type", } 451 #elif defined(__alpha__) || defined(__mips__) 452 #define CPUINFO_PROC { "cpu model", } 453 #elif defined(__arm__) 454 #define CPUINFO_PROC { "model name", "Processor", } 455 #elif defined(__arc__) 456 #define CPUINFO_PROC { "Processor", } 457 #elif defined(__xtensa__) 458 #define CPUINFO_PROC { "core ID", } 459 #else 460 #define CPUINFO_PROC { "model name", } 461 #endif 462 const char *cpuinfo_procs[] = CPUINFO_PROC; 463 #undef CPUINFO_PROC 464 unsigned int i; 465 466 for (i = 0; i < ARRAY_SIZE(cpuinfo_procs); i++) { 467 int ret; 468 ret = __write_cpudesc(ff, cpuinfo_procs[i]); 469 if (ret >= 0) 470 return ret; 471 } 472 return -1; 473 } 474 475 476 static int write_nrcpus(struct feat_fd *ff, 477 struct evlist *evlist __maybe_unused) 478 { 479 long nr; 480 u32 nrc, nra; 481 int ret; 482 483 nrc = cpu__max_present_cpu().cpu; 484 485 nr = sysconf(_SC_NPROCESSORS_ONLN); 486 if (nr < 0) 487 return -1; 488 489 nra = (u32)(nr & UINT_MAX); 490 491 ret = do_write(ff, &nrc, sizeof(nrc)); 492 if (ret < 0) 493 return ret; 494 495 return do_write(ff, &nra, sizeof(nra)); 496 } 497 498 static int write_event_desc(struct feat_fd *ff, 499 struct evlist *evlist) 500 { 501 struct evsel *evsel; 502 u32 nre, nri, sz; 503 int ret; 504 505 nre = evlist->core.nr_entries; 506 507 /* 508 * write number of events 509 */ 510 ret = do_write(ff, &nre, sizeof(nre)); 511 if (ret < 0) 512 return ret; 513 514 /* 515 * size of perf_event_attr struct 516 */ 517 sz = (u32)sizeof(evsel->core.attr); 518 ret = do_write(ff, &sz, sizeof(sz)); 519 if (ret < 0) 520 return ret; 521 522 evlist__for_each_entry(evlist, evsel) { 523 ret = do_write(ff, &evsel->core.attr, sz); 524 if (ret < 0) 525 return ret; 526 /* 527 * write number of unique id per event 528 * there is one id per instance of an event 529 * 530 * copy into an nri to be independent of the 531 * type of ids, 532 */ 533 nri = evsel->core.ids; 534 ret = do_write(ff, &nri, sizeof(nri)); 535 if (ret < 0) 536 return ret; 537 538 /* 539 * write event string as passed on cmdline 540 */ 541 ret = do_write_string(ff, evsel__name(evsel)); 542 if (ret < 0) 543 return ret; 544 /* 545 * write unique ids for this event 546 */ 547 ret = do_write(ff, evsel->core.id, evsel->core.ids * sizeof(u64)); 548 if (ret < 0) 549 return ret; 550 } 551 return 0; 552 } 553 554 static int write_cmdline(struct feat_fd *ff, 555 struct evlist *evlist __maybe_unused) 556 { 557 char pbuf[MAXPATHLEN], *buf; 558 int i, ret, n; 559 560 /* actual path to perf binary */ 561 buf = perf_exe(pbuf, MAXPATHLEN); 562 563 /* account for binary path */ 564 n = perf_env.nr_cmdline + 1; 565 566 ret = do_write(ff, &n, sizeof(n)); 567 if (ret < 0) 568 return ret; 569 570 ret = do_write_string(ff, buf); 571 if (ret < 0) 572 return ret; 573 574 for (i = 0 ; i < perf_env.nr_cmdline; i++) { 575 ret = do_write_string(ff, perf_env.cmdline_argv[i]); 576 if (ret < 0) 577 return ret; 578 } 579 return 0; 580 } 581 582 583 static int write_cpu_topology(struct feat_fd *ff, 584 struct evlist *evlist __maybe_unused) 585 { 586 struct cpu_topology *tp; 587 u32 i; 588 int ret, j; 589 590 tp = cpu_topology__new(); 591 if (!tp) 592 return -1; 593 594 ret = do_write(ff, &tp->package_cpus_lists, sizeof(tp->package_cpus_lists)); 595 if (ret < 0) 596 goto done; 597 598 for (i = 0; i < tp->package_cpus_lists; i++) { 599 ret = do_write_string(ff, tp->package_cpus_list[i]); 600 if (ret < 0) 601 goto done; 602 } 603 ret = do_write(ff, &tp->core_cpus_lists, sizeof(tp->core_cpus_lists)); 604 if (ret < 0) 605 goto done; 606 607 for (i = 0; i < tp->core_cpus_lists; i++) { 608 ret = do_write_string(ff, tp->core_cpus_list[i]); 609 if (ret < 0) 610 break; 611 } 612 613 ret = perf_env__read_cpu_topology_map(&perf_env); 614 if (ret < 0) 615 goto done; 616 617 for (j = 0; j < perf_env.nr_cpus_avail; j++) { 618 ret = do_write(ff, &perf_env.cpu[j].core_id, 619 sizeof(perf_env.cpu[j].core_id)); 620 if (ret < 0) 621 return ret; 622 ret = do_write(ff, &perf_env.cpu[j].socket_id, 623 sizeof(perf_env.cpu[j].socket_id)); 624 if (ret < 0) 625 return ret; 626 } 627 628 if (!tp->die_cpus_lists) 629 goto done; 630 631 ret = do_write(ff, &tp->die_cpus_lists, sizeof(tp->die_cpus_lists)); 632 if (ret < 0) 633 goto done; 634 635 for (i = 0; i < tp->die_cpus_lists; i++) { 636 ret = do_write_string(ff, tp->die_cpus_list[i]); 637 if (ret < 0) 638 goto done; 639 } 640 641 for (j = 0; j < perf_env.nr_cpus_avail; j++) { 642 ret = do_write(ff, &perf_env.cpu[j].die_id, 643 sizeof(perf_env.cpu[j].die_id)); 644 if (ret < 0) 645 return ret; 646 } 647 648 done: 649 cpu_topology__delete(tp); 650 return ret; 651 } 652 653 654 655 static int write_total_mem(struct feat_fd *ff, 656 struct evlist *evlist __maybe_unused) 657 { 658 char *buf = NULL; 659 FILE *fp; 660 size_t len = 0; 661 int ret = -1, n; 662 uint64_t mem; 663 664 fp = fopen("/proc/meminfo", "r"); 665 if (!fp) 666 return -1; 667 668 while (getline(&buf, &len, fp) > 0) { 669 ret = strncmp(buf, "MemTotal:", 9); 670 if (!ret) 671 break; 672 } 673 if (!ret) { 674 n = sscanf(buf, "%*s %"PRIu64, &mem); 675 if (n == 1) 676 ret = do_write(ff, &mem, sizeof(mem)); 677 } else 678 ret = -1; 679 free(buf); 680 fclose(fp); 681 return ret; 682 } 683 684 static int write_numa_topology(struct feat_fd *ff, 685 struct evlist *evlist __maybe_unused) 686 { 687 struct numa_topology *tp; 688 int ret = -1; 689 u32 i; 690 691 tp = numa_topology__new(); 692 if (!tp) 693 return -ENOMEM; 694 695 ret = do_write(ff, &tp->nr, sizeof(u32)); 696 if (ret < 0) 697 goto err; 698 699 for (i = 0; i < tp->nr; i++) { 700 struct numa_topology_node *n = &tp->nodes[i]; 701 702 ret = do_write(ff, &n->node, sizeof(u32)); 703 if (ret < 0) 704 goto err; 705 706 ret = do_write(ff, &n->mem_total, sizeof(u64)); 707 if (ret) 708 goto err; 709 710 ret = do_write(ff, &n->mem_free, sizeof(u64)); 711 if (ret) 712 goto err; 713 714 ret = do_write_string(ff, n->cpus); 715 if (ret < 0) 716 goto err; 717 } 718 719 ret = 0; 720 721 err: 722 numa_topology__delete(tp); 723 return ret; 724 } 725 726 /* 727 * File format: 728 * 729 * struct pmu_mappings { 730 * u32 pmu_num; 731 * struct pmu_map { 732 * u32 type; 733 * char name[]; 734 * }[pmu_num]; 735 * }; 736 */ 737 738 static int write_pmu_mappings(struct feat_fd *ff, 739 struct evlist *evlist __maybe_unused) 740 { 741 struct perf_pmu *pmu = NULL; 742 u32 pmu_num = 0; 743 int ret; 744 745 /* 746 * Do a first pass to count number of pmu to avoid lseek so this 747 * works in pipe mode as well. 748 */ 749 while ((pmu = perf_pmus__scan(pmu))) 750 pmu_num++; 751 752 ret = do_write(ff, &pmu_num, sizeof(pmu_num)); 753 if (ret < 0) 754 return ret; 755 756 while ((pmu = perf_pmus__scan(pmu))) { 757 ret = do_write(ff, &pmu->type, sizeof(pmu->type)); 758 if (ret < 0) 759 return ret; 760 761 ret = do_write_string(ff, pmu->name); 762 if (ret < 0) 763 return ret; 764 } 765 766 return 0; 767 } 768 769 /* 770 * File format: 771 * 772 * struct group_descs { 773 * u32 nr_groups; 774 * struct group_desc { 775 * char name[]; 776 * u32 leader_idx; 777 * u32 nr_members; 778 * }[nr_groups]; 779 * }; 780 */ 781 static int write_group_desc(struct feat_fd *ff, 782 struct evlist *evlist) 783 { 784 u32 nr_groups = evlist__nr_groups(evlist); 785 struct evsel *evsel; 786 int ret; 787 788 ret = do_write(ff, &nr_groups, sizeof(nr_groups)); 789 if (ret < 0) 790 return ret; 791 792 evlist__for_each_entry(evlist, evsel) { 793 if (evsel__is_group_leader(evsel) && evsel->core.nr_members > 1) { 794 const char *name = evsel->group_name ?: "{anon_group}"; 795 u32 leader_idx = evsel->core.idx; 796 u32 nr_members = evsel->core.nr_members; 797 798 ret = do_write_string(ff, name); 799 if (ret < 0) 800 return ret; 801 802 ret = do_write(ff, &leader_idx, sizeof(leader_idx)); 803 if (ret < 0) 804 return ret; 805 806 ret = do_write(ff, &nr_members, sizeof(nr_members)); 807 if (ret < 0) 808 return ret; 809 } 810 } 811 return 0; 812 } 813 814 /* 815 * Return the CPU id as a raw string. 816 * 817 * Each architecture should provide a more precise id string that 818 * can be use to match the architecture's "mapfile". 819 */ 820 char * __weak get_cpuid_str(struct perf_pmu *pmu __maybe_unused) 821 { 822 return NULL; 823 } 824 825 /* Return zero when the cpuid from the mapfile.csv matches the 826 * cpuid string generated on this platform. 827 * Otherwise return non-zero. 828 */ 829 int __weak strcmp_cpuid_str(const char *mapcpuid, const char *cpuid) 830 { 831 regex_t re; 832 regmatch_t pmatch[1]; 833 int match; 834 835 if (regcomp(&re, mapcpuid, REG_EXTENDED) != 0) { 836 /* Warn unable to generate match particular string. */ 837 pr_info("Invalid regular expression %s\n", mapcpuid); 838 return 1; 839 } 840 841 match = !regexec(&re, cpuid, 1, pmatch, 0); 842 regfree(&re); 843 if (match) { 844 size_t match_len = (pmatch[0].rm_eo - pmatch[0].rm_so); 845 846 /* Verify the entire string matched. */ 847 if (match_len == strlen(cpuid)) 848 return 0; 849 } 850 return 1; 851 } 852 853 /* 854 * default get_cpuid(): nothing gets recorded 855 * actual implementation must be in arch/$(SRCARCH)/util/header.c 856 */ 857 int __weak get_cpuid(char *buffer __maybe_unused, size_t sz __maybe_unused) 858 { 859 return ENOSYS; /* Not implemented */ 860 } 861 862 static int write_cpuid(struct feat_fd *ff, 863 struct evlist *evlist __maybe_unused) 864 { 865 char buffer[64]; 866 int ret; 867 868 ret = get_cpuid(buffer, sizeof(buffer)); 869 if (ret) 870 return -1; 871 872 return do_write_string(ff, buffer); 873 } 874 875 static int write_branch_stack(struct feat_fd *ff __maybe_unused, 876 struct evlist *evlist __maybe_unused) 877 { 878 return 0; 879 } 880 881 static int write_auxtrace(struct feat_fd *ff, 882 struct evlist *evlist __maybe_unused) 883 { 884 struct perf_session *session; 885 int err; 886 887 if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__)) 888 return -1; 889 890 session = container_of(ff->ph, struct perf_session, header); 891 892 err = auxtrace_index__write(ff->fd, &session->auxtrace_index); 893 if (err < 0) 894 pr_err("Failed to write auxtrace index\n"); 895 return err; 896 } 897 898 static int write_clockid(struct feat_fd *ff, 899 struct evlist *evlist __maybe_unused) 900 { 901 return do_write(ff, &ff->ph->env.clock.clockid_res_ns, 902 sizeof(ff->ph->env.clock.clockid_res_ns)); 903 } 904 905 static int write_clock_data(struct feat_fd *ff, 906 struct evlist *evlist __maybe_unused) 907 { 908 u64 *data64; 909 u32 data32; 910 int ret; 911 912 /* version */ 913 data32 = 1; 914 915 ret = do_write(ff, &data32, sizeof(data32)); 916 if (ret < 0) 917 return ret; 918 919 /* clockid */ 920 data32 = ff->ph->env.clock.clockid; 921 922 ret = do_write(ff, &data32, sizeof(data32)); 923 if (ret < 0) 924 return ret; 925 926 /* TOD ref time */ 927 data64 = &ff->ph->env.clock.tod_ns; 928 929 ret = do_write(ff, data64, sizeof(*data64)); 930 if (ret < 0) 931 return ret; 932 933 /* clockid ref time */ 934 data64 = &ff->ph->env.clock.clockid_ns; 935 936 return do_write(ff, data64, sizeof(*data64)); 937 } 938 939 static int write_hybrid_topology(struct feat_fd *ff, 940 struct evlist *evlist __maybe_unused) 941 { 942 struct hybrid_topology *tp; 943 int ret; 944 u32 i; 945 946 tp = hybrid_topology__new(); 947 if (!tp) 948 return -ENOENT; 949 950 ret = do_write(ff, &tp->nr, sizeof(u32)); 951 if (ret < 0) 952 goto err; 953 954 for (i = 0; i < tp->nr; i++) { 955 struct hybrid_topology_node *n = &tp->nodes[i]; 956 957 ret = do_write_string(ff, n->pmu_name); 958 if (ret < 0) 959 goto err; 960 961 ret = do_write_string(ff, n->cpus); 962 if (ret < 0) 963 goto err; 964 } 965 966 ret = 0; 967 968 err: 969 hybrid_topology__delete(tp); 970 return ret; 971 } 972 973 static int write_dir_format(struct feat_fd *ff, 974 struct evlist *evlist __maybe_unused) 975 { 976 struct perf_session *session; 977 struct perf_data *data; 978 979 session = container_of(ff->ph, struct perf_session, header); 980 data = session->data; 981 982 if (WARN_ON(!perf_data__is_dir(data))) 983 return -1; 984 985 return do_write(ff, &data->dir.version, sizeof(data->dir.version)); 986 } 987 988 /* 989 * Check whether a CPU is online 990 * 991 * Returns: 992 * 1 -> if CPU is online 993 * 0 -> if CPU is offline 994 * -1 -> error case 995 */ 996 int is_cpu_online(unsigned int cpu) 997 { 998 char *str; 999 size_t strlen; 1000 char buf[256]; 1001 int status = -1; 1002 struct stat statbuf; 1003 1004 snprintf(buf, sizeof(buf), 1005 "/sys/devices/system/cpu/cpu%d", cpu); 1006 if (stat(buf, &statbuf) != 0) 1007 return 0; 1008 1009 /* 1010 * Check if /sys/devices/system/cpu/cpux/online file 1011 * exists. Some cases cpu0 won't have online file since 1012 * it is not expected to be turned off generally. 1013 * In kernels without CONFIG_HOTPLUG_CPU, this 1014 * file won't exist 1015 */ 1016 snprintf(buf, sizeof(buf), 1017 "/sys/devices/system/cpu/cpu%d/online", cpu); 1018 if (stat(buf, &statbuf) != 0) 1019 return 1; 1020 1021 /* 1022 * Read online file using sysfs__read_str. 1023 * If read or open fails, return -1. 1024 * If read succeeds, return value from file 1025 * which gets stored in "str" 1026 */ 1027 snprintf(buf, sizeof(buf), 1028 "devices/system/cpu/cpu%d/online", cpu); 1029 1030 if (sysfs__read_str(buf, &str, &strlen) < 0) 1031 return status; 1032 1033 status = atoi(str); 1034 1035 free(str); 1036 return status; 1037 } 1038 1039 #ifdef HAVE_LIBBPF_SUPPORT 1040 static int write_bpf_prog_info(struct feat_fd *ff, 1041 struct evlist *evlist __maybe_unused) 1042 { 1043 struct perf_env *env = &ff->ph->env; 1044 struct rb_root *root; 1045 struct rb_node *next; 1046 int ret; 1047 1048 down_read(&env->bpf_progs.lock); 1049 1050 ret = do_write(ff, &env->bpf_progs.infos_cnt, 1051 sizeof(env->bpf_progs.infos_cnt)); 1052 if (ret < 0) 1053 goto out; 1054 1055 root = &env->bpf_progs.infos; 1056 next = rb_first(root); 1057 while (next) { 1058 struct bpf_prog_info_node *node; 1059 size_t len; 1060 1061 node = rb_entry(next, struct bpf_prog_info_node, rb_node); 1062 next = rb_next(&node->rb_node); 1063 len = sizeof(struct perf_bpil) + 1064 node->info_linear->data_len; 1065 1066 /* before writing to file, translate address to offset */ 1067 bpil_addr_to_offs(node->info_linear); 1068 ret = do_write(ff, node->info_linear, len); 1069 /* 1070 * translate back to address even when do_write() fails, 1071 * so that this function never changes the data. 1072 */ 1073 bpil_offs_to_addr(node->info_linear); 1074 if (ret < 0) 1075 goto out; 1076 } 1077 out: 1078 up_read(&env->bpf_progs.lock); 1079 return ret; 1080 } 1081 1082 static int write_bpf_btf(struct feat_fd *ff, 1083 struct evlist *evlist __maybe_unused) 1084 { 1085 struct perf_env *env = &ff->ph->env; 1086 struct rb_root *root; 1087 struct rb_node *next; 1088 int ret; 1089 1090 down_read(&env->bpf_progs.lock); 1091 1092 ret = do_write(ff, &env->bpf_progs.btfs_cnt, 1093 sizeof(env->bpf_progs.btfs_cnt)); 1094 1095 if (ret < 0) 1096 goto out; 1097 1098 root = &env->bpf_progs.btfs; 1099 next = rb_first(root); 1100 while (next) { 1101 struct btf_node *node; 1102 1103 node = rb_entry(next, struct btf_node, rb_node); 1104 next = rb_next(&node->rb_node); 1105 ret = do_write(ff, &node->id, 1106 sizeof(u32) * 2 + node->data_size); 1107 if (ret < 0) 1108 goto out; 1109 } 1110 out: 1111 up_read(&env->bpf_progs.lock); 1112 return ret; 1113 } 1114 #endif // HAVE_LIBBPF_SUPPORT 1115 1116 static int cpu_cache_level__sort(const void *a, const void *b) 1117 { 1118 struct cpu_cache_level *cache_a = (struct cpu_cache_level *)a; 1119 struct cpu_cache_level *cache_b = (struct cpu_cache_level *)b; 1120 1121 return cache_a->level - cache_b->level; 1122 } 1123 1124 static bool cpu_cache_level__cmp(struct cpu_cache_level *a, struct cpu_cache_level *b) 1125 { 1126 if (a->level != b->level) 1127 return false; 1128 1129 if (a->line_size != b->line_size) 1130 return false; 1131 1132 if (a->sets != b->sets) 1133 return false; 1134 1135 if (a->ways != b->ways) 1136 return false; 1137 1138 if (strcmp(a->type, b->type)) 1139 return false; 1140 1141 if (strcmp(a->size, b->size)) 1142 return false; 1143 1144 if (strcmp(a->map, b->map)) 1145 return false; 1146 1147 return true; 1148 } 1149 1150 static int cpu_cache_level__read(struct cpu_cache_level *cache, u32 cpu, u16 level) 1151 { 1152 char path[PATH_MAX], file[PATH_MAX]; 1153 struct stat st; 1154 size_t len; 1155 1156 scnprintf(path, PATH_MAX, "devices/system/cpu/cpu%d/cache/index%d/", cpu, level); 1157 scnprintf(file, PATH_MAX, "%s/%s", sysfs__mountpoint(), path); 1158 1159 if (stat(file, &st)) 1160 return 1; 1161 1162 scnprintf(file, PATH_MAX, "%s/level", path); 1163 if (sysfs__read_int(file, (int *) &cache->level)) 1164 return -1; 1165 1166 scnprintf(file, PATH_MAX, "%s/coherency_line_size", path); 1167 if (sysfs__read_int(file, (int *) &cache->line_size)) 1168 return -1; 1169 1170 scnprintf(file, PATH_MAX, "%s/number_of_sets", path); 1171 if (sysfs__read_int(file, (int *) &cache->sets)) 1172 return -1; 1173 1174 scnprintf(file, PATH_MAX, "%s/ways_of_associativity", path); 1175 if (sysfs__read_int(file, (int *) &cache->ways)) 1176 return -1; 1177 1178 scnprintf(file, PATH_MAX, "%s/type", path); 1179 if (sysfs__read_str(file, &cache->type, &len)) 1180 return -1; 1181 1182 cache->type[len] = 0; 1183 cache->type = strim(cache->type); 1184 1185 scnprintf(file, PATH_MAX, "%s/size", path); 1186 if (sysfs__read_str(file, &cache->size, &len)) { 1187 zfree(&cache->type); 1188 return -1; 1189 } 1190 1191 cache->size[len] = 0; 1192 cache->size = strim(cache->size); 1193 1194 scnprintf(file, PATH_MAX, "%s/shared_cpu_list", path); 1195 if (sysfs__read_str(file, &cache->map, &len)) { 1196 zfree(&cache->size); 1197 zfree(&cache->type); 1198 return -1; 1199 } 1200 1201 cache->map[len] = 0; 1202 cache->map = strim(cache->map); 1203 return 0; 1204 } 1205 1206 static void cpu_cache_level__fprintf(FILE *out, struct cpu_cache_level *c) 1207 { 1208 fprintf(out, "L%d %-15s %8s [%s]\n", c->level, c->type, c->size, c->map); 1209 } 1210 1211 /* 1212 * Build caches levels for a particular CPU from the data in 1213 * /sys/devices/system/cpu/cpu<cpu>/cache/ 1214 * The cache level data is stored in caches[] from index at 1215 * *cntp. 1216 */ 1217 int build_caches_for_cpu(u32 cpu, struct cpu_cache_level caches[], u32 *cntp) 1218 { 1219 u16 level; 1220 1221 for (level = 0; level < MAX_CACHE_LVL; level++) { 1222 struct cpu_cache_level c; 1223 int err; 1224 u32 i; 1225 1226 err = cpu_cache_level__read(&c, cpu, level); 1227 if (err < 0) 1228 return err; 1229 1230 if (err == 1) 1231 break; 1232 1233 for (i = 0; i < *cntp; i++) { 1234 if (cpu_cache_level__cmp(&c, &caches[i])) 1235 break; 1236 } 1237 1238 if (i == *cntp) { 1239 caches[*cntp] = c; 1240 *cntp = *cntp + 1; 1241 } else 1242 cpu_cache_level__free(&c); 1243 } 1244 1245 return 0; 1246 } 1247 1248 static int build_caches(struct cpu_cache_level caches[], u32 *cntp) 1249 { 1250 u32 nr, cpu, cnt = 0; 1251 1252 nr = cpu__max_cpu().cpu; 1253 1254 for (cpu = 0; cpu < nr; cpu++) { 1255 int ret = build_caches_for_cpu(cpu, caches, &cnt); 1256 1257 if (ret) 1258 return ret; 1259 } 1260 *cntp = cnt; 1261 return 0; 1262 } 1263 1264 static int write_cache(struct feat_fd *ff, 1265 struct evlist *evlist __maybe_unused) 1266 { 1267 u32 max_caches = cpu__max_cpu().cpu * MAX_CACHE_LVL; 1268 struct cpu_cache_level caches[max_caches]; 1269 u32 cnt = 0, i, version = 1; 1270 int ret; 1271 1272 ret = build_caches(caches, &cnt); 1273 if (ret) 1274 goto out; 1275 1276 qsort(&caches, cnt, sizeof(struct cpu_cache_level), cpu_cache_level__sort); 1277 1278 ret = do_write(ff, &version, sizeof(u32)); 1279 if (ret < 0) 1280 goto out; 1281 1282 ret = do_write(ff, &cnt, sizeof(u32)); 1283 if (ret < 0) 1284 goto out; 1285 1286 for (i = 0; i < cnt; i++) { 1287 struct cpu_cache_level *c = &caches[i]; 1288 1289 #define _W(v) \ 1290 ret = do_write(ff, &c->v, sizeof(u32)); \ 1291 if (ret < 0) \ 1292 goto out; 1293 1294 _W(level) 1295 _W(line_size) 1296 _W(sets) 1297 _W(ways) 1298 #undef _W 1299 1300 #define _W(v) \ 1301 ret = do_write_string(ff, (const char *) c->v); \ 1302 if (ret < 0) \ 1303 goto out; 1304 1305 _W(type) 1306 _W(size) 1307 _W(map) 1308 #undef _W 1309 } 1310 1311 out: 1312 for (i = 0; i < cnt; i++) 1313 cpu_cache_level__free(&caches[i]); 1314 return ret; 1315 } 1316 1317 static int write_stat(struct feat_fd *ff __maybe_unused, 1318 struct evlist *evlist __maybe_unused) 1319 { 1320 return 0; 1321 } 1322 1323 static int write_sample_time(struct feat_fd *ff, 1324 struct evlist *evlist) 1325 { 1326 int ret; 1327 1328 ret = do_write(ff, &evlist->first_sample_time, 1329 sizeof(evlist->first_sample_time)); 1330 if (ret < 0) 1331 return ret; 1332 1333 return do_write(ff, &evlist->last_sample_time, 1334 sizeof(evlist->last_sample_time)); 1335 } 1336 1337 1338 static int memory_node__read(struct memory_node *n, unsigned long idx) 1339 { 1340 unsigned int phys, size = 0; 1341 char path[PATH_MAX]; 1342 struct dirent *ent; 1343 DIR *dir; 1344 1345 #define for_each_memory(mem, dir) \ 1346 while ((ent = readdir(dir))) \ 1347 if (strcmp(ent->d_name, ".") && \ 1348 strcmp(ent->d_name, "..") && \ 1349 sscanf(ent->d_name, "memory%u", &mem) == 1) 1350 1351 scnprintf(path, PATH_MAX, 1352 "%s/devices/system/node/node%lu", 1353 sysfs__mountpoint(), idx); 1354 1355 dir = opendir(path); 1356 if (!dir) { 1357 pr_warning("failed: can't open memory sysfs data\n"); 1358 return -1; 1359 } 1360 1361 for_each_memory(phys, dir) { 1362 size = max(phys, size); 1363 } 1364 1365 size++; 1366 1367 n->set = bitmap_zalloc(size); 1368 if (!n->set) { 1369 closedir(dir); 1370 return -ENOMEM; 1371 } 1372 1373 n->node = idx; 1374 n->size = size; 1375 1376 rewinddir(dir); 1377 1378 for_each_memory(phys, dir) { 1379 __set_bit(phys, n->set); 1380 } 1381 1382 closedir(dir); 1383 return 0; 1384 } 1385 1386 static void memory_node__delete_nodes(struct memory_node *nodesp, u64 cnt) 1387 { 1388 for (u64 i = 0; i < cnt; i++) 1389 bitmap_free(nodesp[i].set); 1390 1391 free(nodesp); 1392 } 1393 1394 static int memory_node__sort(const void *a, const void *b) 1395 { 1396 const struct memory_node *na = a; 1397 const struct memory_node *nb = b; 1398 1399 return na->node - nb->node; 1400 } 1401 1402 static int build_mem_topology(struct memory_node **nodesp, u64 *cntp) 1403 { 1404 char path[PATH_MAX]; 1405 struct dirent *ent; 1406 DIR *dir; 1407 int ret = 0; 1408 size_t cnt = 0, size = 0; 1409 struct memory_node *nodes = NULL; 1410 1411 scnprintf(path, PATH_MAX, "%s/devices/system/node/", 1412 sysfs__mountpoint()); 1413 1414 dir = opendir(path); 1415 if (!dir) { 1416 pr_debug2("%s: couldn't read %s, does this arch have topology information?\n", 1417 __func__, path); 1418 return -1; 1419 } 1420 1421 while (!ret && (ent = readdir(dir))) { 1422 unsigned int idx; 1423 int r; 1424 1425 if (!strcmp(ent->d_name, ".") || 1426 !strcmp(ent->d_name, "..")) 1427 continue; 1428 1429 r = sscanf(ent->d_name, "node%u", &idx); 1430 if (r != 1) 1431 continue; 1432 1433 if (cnt >= size) { 1434 struct memory_node *new_nodes = 1435 reallocarray(nodes, cnt + 4, sizeof(*nodes)); 1436 1437 if (!new_nodes) { 1438 pr_err("Failed to write MEM_TOPOLOGY, size %zd nodes\n", size); 1439 ret = -ENOMEM; 1440 goto out; 1441 } 1442 nodes = new_nodes; 1443 size += 4; 1444 } 1445 ret = memory_node__read(&nodes[cnt++], idx); 1446 } 1447 out: 1448 closedir(dir); 1449 if (!ret) { 1450 *cntp = cnt; 1451 *nodesp = nodes; 1452 qsort(nodes, cnt, sizeof(nodes[0]), memory_node__sort); 1453 } else 1454 memory_node__delete_nodes(nodes, cnt); 1455 1456 return ret; 1457 } 1458 1459 /* 1460 * The MEM_TOPOLOGY holds physical memory map for every 1461 * node in system. The format of data is as follows: 1462 * 1463 * 0 - version | for future changes 1464 * 8 - block_size_bytes | /sys/devices/system/memory/block_size_bytes 1465 * 16 - count | number of nodes 1466 * 1467 * For each node we store map of physical indexes for 1468 * each node: 1469 * 1470 * 32 - node id | node index 1471 * 40 - size | size of bitmap 1472 * 48 - bitmap | bitmap of memory indexes that belongs to node 1473 */ 1474 static int write_mem_topology(struct feat_fd *ff __maybe_unused, 1475 struct evlist *evlist __maybe_unused) 1476 { 1477 struct memory_node *nodes = NULL; 1478 u64 bsize, version = 1, i, nr = 0; 1479 int ret; 1480 1481 ret = sysfs__read_xll("devices/system/memory/block_size_bytes", 1482 (unsigned long long *) &bsize); 1483 if (ret) 1484 return ret; 1485 1486 ret = build_mem_topology(&nodes, &nr); 1487 if (ret) 1488 return ret; 1489 1490 ret = do_write(ff, &version, sizeof(version)); 1491 if (ret < 0) 1492 goto out; 1493 1494 ret = do_write(ff, &bsize, sizeof(bsize)); 1495 if (ret < 0) 1496 goto out; 1497 1498 ret = do_write(ff, &nr, sizeof(nr)); 1499 if (ret < 0) 1500 goto out; 1501 1502 for (i = 0; i < nr; i++) { 1503 struct memory_node *n = &nodes[i]; 1504 1505 #define _W(v) \ 1506 ret = do_write(ff, &n->v, sizeof(n->v)); \ 1507 if (ret < 0) \ 1508 goto out; 1509 1510 _W(node) 1511 _W(size) 1512 1513 #undef _W 1514 1515 ret = do_write_bitmap(ff, n->set, n->size); 1516 if (ret < 0) 1517 goto out; 1518 } 1519 1520 out: 1521 memory_node__delete_nodes(nodes, nr); 1522 return ret; 1523 } 1524 1525 static int write_compressed(struct feat_fd *ff __maybe_unused, 1526 struct evlist *evlist __maybe_unused) 1527 { 1528 int ret; 1529 1530 ret = do_write(ff, &(ff->ph->env.comp_ver), sizeof(ff->ph->env.comp_ver)); 1531 if (ret) 1532 return ret; 1533 1534 ret = do_write(ff, &(ff->ph->env.comp_type), sizeof(ff->ph->env.comp_type)); 1535 if (ret) 1536 return ret; 1537 1538 ret = do_write(ff, &(ff->ph->env.comp_level), sizeof(ff->ph->env.comp_level)); 1539 if (ret) 1540 return ret; 1541 1542 ret = do_write(ff, &(ff->ph->env.comp_ratio), sizeof(ff->ph->env.comp_ratio)); 1543 if (ret) 1544 return ret; 1545 1546 return do_write(ff, &(ff->ph->env.comp_mmap_len), sizeof(ff->ph->env.comp_mmap_len)); 1547 } 1548 1549 static int __write_pmu_caps(struct feat_fd *ff, struct perf_pmu *pmu, 1550 bool write_pmu) 1551 { 1552 struct perf_pmu_caps *caps = NULL; 1553 int ret; 1554 1555 ret = do_write(ff, &pmu->nr_caps, sizeof(pmu->nr_caps)); 1556 if (ret < 0) 1557 return ret; 1558 1559 list_for_each_entry(caps, &pmu->caps, list) { 1560 ret = do_write_string(ff, caps->name); 1561 if (ret < 0) 1562 return ret; 1563 1564 ret = do_write_string(ff, caps->value); 1565 if (ret < 0) 1566 return ret; 1567 } 1568 1569 if (write_pmu) { 1570 ret = do_write_string(ff, pmu->name); 1571 if (ret < 0) 1572 return ret; 1573 } 1574 1575 return ret; 1576 } 1577 1578 static int write_cpu_pmu_caps(struct feat_fd *ff, 1579 struct evlist *evlist __maybe_unused) 1580 { 1581 struct perf_pmu *cpu_pmu = perf_pmus__find("cpu"); 1582 int ret; 1583 1584 if (!cpu_pmu) 1585 return -ENOENT; 1586 1587 ret = perf_pmu__caps_parse(cpu_pmu); 1588 if (ret < 0) 1589 return ret; 1590 1591 return __write_pmu_caps(ff, cpu_pmu, false); 1592 } 1593 1594 static int write_pmu_caps(struct feat_fd *ff, 1595 struct evlist *evlist __maybe_unused) 1596 { 1597 struct perf_pmu *pmu = NULL; 1598 int nr_pmu = 0; 1599 int ret; 1600 1601 while ((pmu = perf_pmus__scan(pmu))) { 1602 if (!strcmp(pmu->name, "cpu")) { 1603 /* 1604 * The "cpu" PMU is special and covered by 1605 * HEADER_CPU_PMU_CAPS. Note, core PMUs are 1606 * counted/written here for ARM, s390 and Intel hybrid. 1607 */ 1608 continue; 1609 } 1610 if (perf_pmu__caps_parse(pmu) <= 0) 1611 continue; 1612 nr_pmu++; 1613 } 1614 1615 ret = do_write(ff, &nr_pmu, sizeof(nr_pmu)); 1616 if (ret < 0) 1617 return ret; 1618 1619 if (!nr_pmu) 1620 return 0; 1621 1622 /* 1623 * Note older perf tools assume core PMUs come first, this is a property 1624 * of perf_pmus__scan. 1625 */ 1626 pmu = NULL; 1627 while ((pmu = perf_pmus__scan(pmu))) { 1628 if (!strcmp(pmu->name, "cpu")) { 1629 /* Skip as above. */ 1630 continue; 1631 } 1632 if (perf_pmu__caps_parse(pmu) <= 0) 1633 continue; 1634 ret = __write_pmu_caps(ff, pmu, true); 1635 if (ret < 0) 1636 return ret; 1637 } 1638 return 0; 1639 } 1640 1641 static void print_hostname(struct feat_fd *ff, FILE *fp) 1642 { 1643 fprintf(fp, "# hostname : %s\n", ff->ph->env.hostname); 1644 } 1645 1646 static void print_osrelease(struct feat_fd *ff, FILE *fp) 1647 { 1648 fprintf(fp, "# os release : %s\n", ff->ph->env.os_release); 1649 } 1650 1651 static void print_arch(struct feat_fd *ff, FILE *fp) 1652 { 1653 fprintf(fp, "# arch : %s\n", ff->ph->env.arch); 1654 } 1655 1656 static void print_cpudesc(struct feat_fd *ff, FILE *fp) 1657 { 1658 fprintf(fp, "# cpudesc : %s\n", ff->ph->env.cpu_desc); 1659 } 1660 1661 static void print_nrcpus(struct feat_fd *ff, FILE *fp) 1662 { 1663 fprintf(fp, "# nrcpus online : %u\n", ff->ph->env.nr_cpus_online); 1664 fprintf(fp, "# nrcpus avail : %u\n", ff->ph->env.nr_cpus_avail); 1665 } 1666 1667 static void print_version(struct feat_fd *ff, FILE *fp) 1668 { 1669 fprintf(fp, "# perf version : %s\n", ff->ph->env.version); 1670 } 1671 1672 static void print_cmdline(struct feat_fd *ff, FILE *fp) 1673 { 1674 int nr, i; 1675 1676 nr = ff->ph->env.nr_cmdline; 1677 1678 fprintf(fp, "# cmdline : "); 1679 1680 for (i = 0; i < nr; i++) { 1681 char *argv_i = strdup(ff->ph->env.cmdline_argv[i]); 1682 if (!argv_i) { 1683 fprintf(fp, "%s ", ff->ph->env.cmdline_argv[i]); 1684 } else { 1685 char *mem = argv_i; 1686 do { 1687 char *quote = strchr(argv_i, '\''); 1688 if (!quote) 1689 break; 1690 *quote++ = '\0'; 1691 fprintf(fp, "%s\\\'", argv_i); 1692 argv_i = quote; 1693 } while (1); 1694 fprintf(fp, "%s ", argv_i); 1695 free(mem); 1696 } 1697 } 1698 fputc('\n', fp); 1699 } 1700 1701 static void print_cpu_topology(struct feat_fd *ff, FILE *fp) 1702 { 1703 struct perf_header *ph = ff->ph; 1704 int cpu_nr = ph->env.nr_cpus_avail; 1705 int nr, i; 1706 char *str; 1707 1708 nr = ph->env.nr_sibling_cores; 1709 str = ph->env.sibling_cores; 1710 1711 for (i = 0; i < nr; i++) { 1712 fprintf(fp, "# sibling sockets : %s\n", str); 1713 str += strlen(str) + 1; 1714 } 1715 1716 if (ph->env.nr_sibling_dies) { 1717 nr = ph->env.nr_sibling_dies; 1718 str = ph->env.sibling_dies; 1719 1720 for (i = 0; i < nr; i++) { 1721 fprintf(fp, "# sibling dies : %s\n", str); 1722 str += strlen(str) + 1; 1723 } 1724 } 1725 1726 nr = ph->env.nr_sibling_threads; 1727 str = ph->env.sibling_threads; 1728 1729 for (i = 0; i < nr; i++) { 1730 fprintf(fp, "# sibling threads : %s\n", str); 1731 str += strlen(str) + 1; 1732 } 1733 1734 if (ph->env.nr_sibling_dies) { 1735 if (ph->env.cpu != NULL) { 1736 for (i = 0; i < cpu_nr; i++) 1737 fprintf(fp, "# CPU %d: Core ID %d, " 1738 "Die ID %d, Socket ID %d\n", 1739 i, ph->env.cpu[i].core_id, 1740 ph->env.cpu[i].die_id, 1741 ph->env.cpu[i].socket_id); 1742 } else 1743 fprintf(fp, "# Core ID, Die ID and Socket ID " 1744 "information is not available\n"); 1745 } else { 1746 if (ph->env.cpu != NULL) { 1747 for (i = 0; i < cpu_nr; i++) 1748 fprintf(fp, "# CPU %d: Core ID %d, " 1749 "Socket ID %d\n", 1750 i, ph->env.cpu[i].core_id, 1751 ph->env.cpu[i].socket_id); 1752 } else 1753 fprintf(fp, "# Core ID and Socket ID " 1754 "information is not available\n"); 1755 } 1756 } 1757 1758 static void print_clockid(struct feat_fd *ff, FILE *fp) 1759 { 1760 fprintf(fp, "# clockid frequency: %"PRIu64" MHz\n", 1761 ff->ph->env.clock.clockid_res_ns * 1000); 1762 } 1763 1764 static void print_clock_data(struct feat_fd *ff, FILE *fp) 1765 { 1766 struct timespec clockid_ns; 1767 char tstr[64], date[64]; 1768 struct timeval tod_ns; 1769 clockid_t clockid; 1770 struct tm ltime; 1771 u64 ref; 1772 1773 if (!ff->ph->env.clock.enabled) { 1774 fprintf(fp, "# reference time disabled\n"); 1775 return; 1776 } 1777 1778 /* Compute TOD time. */ 1779 ref = ff->ph->env.clock.tod_ns; 1780 tod_ns.tv_sec = ref / NSEC_PER_SEC; 1781 ref -= tod_ns.tv_sec * NSEC_PER_SEC; 1782 tod_ns.tv_usec = ref / NSEC_PER_USEC; 1783 1784 /* Compute clockid time. */ 1785 ref = ff->ph->env.clock.clockid_ns; 1786 clockid_ns.tv_sec = ref / NSEC_PER_SEC; 1787 ref -= clockid_ns.tv_sec * NSEC_PER_SEC; 1788 clockid_ns.tv_nsec = ref; 1789 1790 clockid = ff->ph->env.clock.clockid; 1791 1792 if (localtime_r(&tod_ns.tv_sec, <ime) == NULL) 1793 snprintf(tstr, sizeof(tstr), "<error>"); 1794 else { 1795 strftime(date, sizeof(date), "%F %T", <ime); 1796 scnprintf(tstr, sizeof(tstr), "%s.%06d", 1797 date, (int) tod_ns.tv_usec); 1798 } 1799 1800 fprintf(fp, "# clockid: %s (%u)\n", clockid_name(clockid), clockid); 1801 fprintf(fp, "# reference time: %s = %ld.%06d (TOD) = %ld.%09ld (%s)\n", 1802 tstr, (long) tod_ns.tv_sec, (int) tod_ns.tv_usec, 1803 (long) clockid_ns.tv_sec, clockid_ns.tv_nsec, 1804 clockid_name(clockid)); 1805 } 1806 1807 static void print_hybrid_topology(struct feat_fd *ff, FILE *fp) 1808 { 1809 int i; 1810 struct hybrid_node *n; 1811 1812 fprintf(fp, "# hybrid cpu system:\n"); 1813 for (i = 0; i < ff->ph->env.nr_hybrid_nodes; i++) { 1814 n = &ff->ph->env.hybrid_nodes[i]; 1815 fprintf(fp, "# %s cpu list : %s\n", n->pmu_name, n->cpus); 1816 } 1817 } 1818 1819 static void print_dir_format(struct feat_fd *ff, FILE *fp) 1820 { 1821 struct perf_session *session; 1822 struct perf_data *data; 1823 1824 session = container_of(ff->ph, struct perf_session, header); 1825 data = session->data; 1826 1827 fprintf(fp, "# directory data version : %"PRIu64"\n", data->dir.version); 1828 } 1829 1830 #ifdef HAVE_LIBBPF_SUPPORT 1831 static void print_bpf_prog_info(struct feat_fd *ff, FILE *fp) 1832 { 1833 struct perf_env *env = &ff->ph->env; 1834 struct rb_root *root; 1835 struct rb_node *next; 1836 1837 down_read(&env->bpf_progs.lock); 1838 1839 root = &env->bpf_progs.infos; 1840 next = rb_first(root); 1841 1842 while (next) { 1843 struct bpf_prog_info_node *node; 1844 1845 node = rb_entry(next, struct bpf_prog_info_node, rb_node); 1846 next = rb_next(&node->rb_node); 1847 1848 bpf_event__print_bpf_prog_info(&node->info_linear->info, 1849 env, fp); 1850 } 1851 1852 up_read(&env->bpf_progs.lock); 1853 } 1854 1855 static void print_bpf_btf(struct feat_fd *ff, FILE *fp) 1856 { 1857 struct perf_env *env = &ff->ph->env; 1858 struct rb_root *root; 1859 struct rb_node *next; 1860 1861 down_read(&env->bpf_progs.lock); 1862 1863 root = &env->bpf_progs.btfs; 1864 next = rb_first(root); 1865 1866 while (next) { 1867 struct btf_node *node; 1868 1869 node = rb_entry(next, struct btf_node, rb_node); 1870 next = rb_next(&node->rb_node); 1871 fprintf(fp, "# btf info of id %u\n", node->id); 1872 } 1873 1874 up_read(&env->bpf_progs.lock); 1875 } 1876 #endif // HAVE_LIBBPF_SUPPORT 1877 1878 static void free_event_desc(struct evsel *events) 1879 { 1880 struct evsel *evsel; 1881 1882 if (!events) 1883 return; 1884 1885 for (evsel = events; evsel->core.attr.size; evsel++) { 1886 zfree(&evsel->name); 1887 zfree(&evsel->core.id); 1888 } 1889 1890 free(events); 1891 } 1892 1893 static bool perf_attr_check(struct perf_event_attr *attr) 1894 { 1895 if (attr->__reserved_1 || attr->__reserved_2 || attr->__reserved_3) { 1896 pr_warning("Reserved bits are set unexpectedly. " 1897 "Please update perf tool.\n"); 1898 return false; 1899 } 1900 1901 if (attr->sample_type & ~(PERF_SAMPLE_MAX-1)) { 1902 pr_warning("Unknown sample type (0x%llx) is detected. " 1903 "Please update perf tool.\n", 1904 attr->sample_type); 1905 return false; 1906 } 1907 1908 if (attr->read_format & ~(PERF_FORMAT_MAX-1)) { 1909 pr_warning("Unknown read format (0x%llx) is detected. " 1910 "Please update perf tool.\n", 1911 attr->read_format); 1912 return false; 1913 } 1914 1915 if ((attr->sample_type & PERF_SAMPLE_BRANCH_STACK) && 1916 (attr->branch_sample_type & ~(PERF_SAMPLE_BRANCH_MAX-1))) { 1917 pr_warning("Unknown branch sample type (0x%llx) is detected. " 1918 "Please update perf tool.\n", 1919 attr->branch_sample_type); 1920 1921 return false; 1922 } 1923 1924 return true; 1925 } 1926 1927 static struct evsel *read_event_desc(struct feat_fd *ff) 1928 { 1929 struct evsel *evsel, *events = NULL; 1930 u64 *id; 1931 void *buf = NULL; 1932 u32 nre, sz, nr, i, j; 1933 size_t msz; 1934 1935 /* number of events */ 1936 if (do_read_u32(ff, &nre)) 1937 goto error; 1938 1939 if (do_read_u32(ff, &sz)) 1940 goto error; 1941 1942 /* buffer to hold on file attr struct */ 1943 buf = malloc(sz); 1944 if (!buf) 1945 goto error; 1946 1947 /* the last event terminates with evsel->core.attr.size == 0: */ 1948 events = calloc(nre + 1, sizeof(*events)); 1949 if (!events) 1950 goto error; 1951 1952 msz = sizeof(evsel->core.attr); 1953 if (sz < msz) 1954 msz = sz; 1955 1956 for (i = 0, evsel = events; i < nre; evsel++, i++) { 1957 evsel->core.idx = i; 1958 1959 /* 1960 * must read entire on-file attr struct to 1961 * sync up with layout. 1962 */ 1963 if (__do_read(ff, buf, sz)) 1964 goto error; 1965 1966 if (ff->ph->needs_swap) 1967 perf_event__attr_swap(buf); 1968 1969 memcpy(&evsel->core.attr, buf, msz); 1970 1971 if (!perf_attr_check(&evsel->core.attr)) 1972 goto error; 1973 1974 if (do_read_u32(ff, &nr)) 1975 goto error; 1976 1977 if (ff->ph->needs_swap) 1978 evsel->needs_swap = true; 1979 1980 evsel->name = do_read_string(ff); 1981 if (!evsel->name) 1982 goto error; 1983 1984 if (!nr) 1985 continue; 1986 1987 id = calloc(nr, sizeof(*id)); 1988 if (!id) 1989 goto error; 1990 evsel->core.ids = nr; 1991 evsel->core.id = id; 1992 1993 for (j = 0 ; j < nr; j++) { 1994 if (do_read_u64(ff, id)) 1995 goto error; 1996 id++; 1997 } 1998 } 1999 out: 2000 free(buf); 2001 return events; 2002 error: 2003 free_event_desc(events); 2004 events = NULL; 2005 goto out; 2006 } 2007 2008 static int __desc_attr__fprintf(FILE *fp, const char *name, const char *val, 2009 void *priv __maybe_unused) 2010 { 2011 return fprintf(fp, ", %s = %s", name, val); 2012 } 2013 2014 static void print_event_desc(struct feat_fd *ff, FILE *fp) 2015 { 2016 struct evsel *evsel, *events; 2017 u32 j; 2018 u64 *id; 2019 2020 if (ff->events) 2021 events = ff->events; 2022 else 2023 events = read_event_desc(ff); 2024 2025 if (!events) { 2026 fprintf(fp, "# event desc: not available or unable to read\n"); 2027 return; 2028 } 2029 2030 for (evsel = events; evsel->core.attr.size; evsel++) { 2031 fprintf(fp, "# event : name = %s, ", evsel->name); 2032 2033 if (evsel->core.ids) { 2034 fprintf(fp, ", id = {"); 2035 for (j = 0, id = evsel->core.id; j < evsel->core.ids; j++, id++) { 2036 if (j) 2037 fputc(',', fp); 2038 fprintf(fp, " %"PRIu64, *id); 2039 } 2040 fprintf(fp, " }"); 2041 } 2042 2043 perf_event_attr__fprintf(fp, &evsel->core.attr, __desc_attr__fprintf, NULL); 2044 2045 fputc('\n', fp); 2046 } 2047 2048 free_event_desc(events); 2049 ff->events = NULL; 2050 } 2051 2052 static void print_total_mem(struct feat_fd *ff, FILE *fp) 2053 { 2054 fprintf(fp, "# total memory : %llu kB\n", ff->ph->env.total_mem); 2055 } 2056 2057 static void print_numa_topology(struct feat_fd *ff, FILE *fp) 2058 { 2059 int i; 2060 struct numa_node *n; 2061 2062 for (i = 0; i < ff->ph->env.nr_numa_nodes; i++) { 2063 n = &ff->ph->env.numa_nodes[i]; 2064 2065 fprintf(fp, "# node%u meminfo : total = %"PRIu64" kB," 2066 " free = %"PRIu64" kB\n", 2067 n->node, n->mem_total, n->mem_free); 2068 2069 fprintf(fp, "# node%u cpu list : ", n->node); 2070 cpu_map__fprintf(n->map, fp); 2071 } 2072 } 2073 2074 static void print_cpuid(struct feat_fd *ff, FILE *fp) 2075 { 2076 fprintf(fp, "# cpuid : %s\n", ff->ph->env.cpuid); 2077 } 2078 2079 static void print_branch_stack(struct feat_fd *ff __maybe_unused, FILE *fp) 2080 { 2081 fprintf(fp, "# contains samples with branch stack\n"); 2082 } 2083 2084 static void print_auxtrace(struct feat_fd *ff __maybe_unused, FILE *fp) 2085 { 2086 fprintf(fp, "# contains AUX area data (e.g. instruction trace)\n"); 2087 } 2088 2089 static void print_stat(struct feat_fd *ff __maybe_unused, FILE *fp) 2090 { 2091 fprintf(fp, "# contains stat data\n"); 2092 } 2093 2094 static void print_cache(struct feat_fd *ff, FILE *fp __maybe_unused) 2095 { 2096 int i; 2097 2098 fprintf(fp, "# CPU cache info:\n"); 2099 for (i = 0; i < ff->ph->env.caches_cnt; i++) { 2100 fprintf(fp, "# "); 2101 cpu_cache_level__fprintf(fp, &ff->ph->env.caches[i]); 2102 } 2103 } 2104 2105 static void print_compressed(struct feat_fd *ff, FILE *fp) 2106 { 2107 fprintf(fp, "# compressed : %s, level = %d, ratio = %d\n", 2108 ff->ph->env.comp_type == PERF_COMP_ZSTD ? "Zstd" : "Unknown", 2109 ff->ph->env.comp_level, ff->ph->env.comp_ratio); 2110 } 2111 2112 static void __print_pmu_caps(FILE *fp, int nr_caps, char **caps, char *pmu_name) 2113 { 2114 const char *delimiter = ""; 2115 int i; 2116 2117 if (!nr_caps) { 2118 fprintf(fp, "# %s pmu capabilities: not available\n", pmu_name); 2119 return; 2120 } 2121 2122 fprintf(fp, "# %s pmu capabilities: ", pmu_name); 2123 for (i = 0; i < nr_caps; i++) { 2124 fprintf(fp, "%s%s", delimiter, caps[i]); 2125 delimiter = ", "; 2126 } 2127 2128 fprintf(fp, "\n"); 2129 } 2130 2131 static void print_cpu_pmu_caps(struct feat_fd *ff, FILE *fp) 2132 { 2133 __print_pmu_caps(fp, ff->ph->env.nr_cpu_pmu_caps, 2134 ff->ph->env.cpu_pmu_caps, (char *)"cpu"); 2135 } 2136 2137 static void print_pmu_caps(struct feat_fd *ff, FILE *fp) 2138 { 2139 struct pmu_caps *pmu_caps; 2140 2141 for (int i = 0; i < ff->ph->env.nr_pmus_with_caps; i++) { 2142 pmu_caps = &ff->ph->env.pmu_caps[i]; 2143 __print_pmu_caps(fp, pmu_caps->nr_caps, pmu_caps->caps, 2144 pmu_caps->pmu_name); 2145 } 2146 } 2147 2148 static void print_pmu_mappings(struct feat_fd *ff, FILE *fp) 2149 { 2150 const char *delimiter = "# pmu mappings: "; 2151 char *str, *tmp; 2152 u32 pmu_num; 2153 u32 type; 2154 2155 pmu_num = ff->ph->env.nr_pmu_mappings; 2156 if (!pmu_num) { 2157 fprintf(fp, "# pmu mappings: not available\n"); 2158 return; 2159 } 2160 2161 str = ff->ph->env.pmu_mappings; 2162 2163 while (pmu_num) { 2164 type = strtoul(str, &tmp, 0); 2165 if (*tmp != ':') 2166 goto error; 2167 2168 str = tmp + 1; 2169 fprintf(fp, "%s%s = %" PRIu32, delimiter, str, type); 2170 2171 delimiter = ", "; 2172 str += strlen(str) + 1; 2173 pmu_num--; 2174 } 2175 2176 fprintf(fp, "\n"); 2177 2178 if (!pmu_num) 2179 return; 2180 error: 2181 fprintf(fp, "# pmu mappings: unable to read\n"); 2182 } 2183 2184 static void print_group_desc(struct feat_fd *ff, FILE *fp) 2185 { 2186 struct perf_session *session; 2187 struct evsel *evsel; 2188 u32 nr = 0; 2189 2190 session = container_of(ff->ph, struct perf_session, header); 2191 2192 evlist__for_each_entry(session->evlist, evsel) { 2193 if (evsel__is_group_leader(evsel) && evsel->core.nr_members > 1) { 2194 fprintf(fp, "# group: %s{%s", evsel->group_name ?: "", evsel__name(evsel)); 2195 2196 nr = evsel->core.nr_members - 1; 2197 } else if (nr) { 2198 fprintf(fp, ",%s", evsel__name(evsel)); 2199 2200 if (--nr == 0) 2201 fprintf(fp, "}\n"); 2202 } 2203 } 2204 } 2205 2206 static void print_sample_time(struct feat_fd *ff, FILE *fp) 2207 { 2208 struct perf_session *session; 2209 char time_buf[32]; 2210 double d; 2211 2212 session = container_of(ff->ph, struct perf_session, header); 2213 2214 timestamp__scnprintf_usec(session->evlist->first_sample_time, 2215 time_buf, sizeof(time_buf)); 2216 fprintf(fp, "# time of first sample : %s\n", time_buf); 2217 2218 timestamp__scnprintf_usec(session->evlist->last_sample_time, 2219 time_buf, sizeof(time_buf)); 2220 fprintf(fp, "# time of last sample : %s\n", time_buf); 2221 2222 d = (double)(session->evlist->last_sample_time - 2223 session->evlist->first_sample_time) / NSEC_PER_MSEC; 2224 2225 fprintf(fp, "# sample duration : %10.3f ms\n", d); 2226 } 2227 2228 static void memory_node__fprintf(struct memory_node *n, 2229 unsigned long long bsize, FILE *fp) 2230 { 2231 char buf_map[100], buf_size[50]; 2232 unsigned long long size; 2233 2234 size = bsize * bitmap_weight(n->set, n->size); 2235 unit_number__scnprintf(buf_size, 50, size); 2236 2237 bitmap_scnprintf(n->set, n->size, buf_map, 100); 2238 fprintf(fp, "# %3" PRIu64 " [%s]: %s\n", n->node, buf_size, buf_map); 2239 } 2240 2241 static void print_mem_topology(struct feat_fd *ff, FILE *fp) 2242 { 2243 struct memory_node *nodes; 2244 int i, nr; 2245 2246 nodes = ff->ph->env.memory_nodes; 2247 nr = ff->ph->env.nr_memory_nodes; 2248 2249 fprintf(fp, "# memory nodes (nr %d, block size 0x%llx):\n", 2250 nr, ff->ph->env.memory_bsize); 2251 2252 for (i = 0; i < nr; i++) { 2253 memory_node__fprintf(&nodes[i], ff->ph->env.memory_bsize, fp); 2254 } 2255 } 2256 2257 static int __event_process_build_id(struct perf_record_header_build_id *bev, 2258 char *filename, 2259 struct perf_session *session) 2260 { 2261 int err = -1; 2262 struct machine *machine; 2263 u16 cpumode; 2264 struct dso *dso; 2265 enum dso_space_type dso_space; 2266 2267 machine = perf_session__findnew_machine(session, bev->pid); 2268 if (!machine) 2269 goto out; 2270 2271 cpumode = bev->header.misc & PERF_RECORD_MISC_CPUMODE_MASK; 2272 2273 switch (cpumode) { 2274 case PERF_RECORD_MISC_KERNEL: 2275 dso_space = DSO_SPACE__KERNEL; 2276 break; 2277 case PERF_RECORD_MISC_GUEST_KERNEL: 2278 dso_space = DSO_SPACE__KERNEL_GUEST; 2279 break; 2280 case PERF_RECORD_MISC_USER: 2281 case PERF_RECORD_MISC_GUEST_USER: 2282 dso_space = DSO_SPACE__USER; 2283 break; 2284 default: 2285 goto out; 2286 } 2287 2288 dso = machine__findnew_dso(machine, filename); 2289 if (dso != NULL) { 2290 char sbuild_id[SBUILD_ID_SIZE]; 2291 struct build_id bid; 2292 size_t size = BUILD_ID_SIZE; 2293 2294 if (bev->header.misc & PERF_RECORD_MISC_BUILD_ID_SIZE) 2295 size = bev->size; 2296 2297 build_id__init(&bid, bev->data, size); 2298 dso__set_build_id(dso, &bid); 2299 dso->header_build_id = 1; 2300 2301 if (dso_space != DSO_SPACE__USER) { 2302 struct kmod_path m = { .name = NULL, }; 2303 2304 if (!kmod_path__parse_name(&m, filename) && m.kmod) 2305 dso__set_module_info(dso, &m, machine); 2306 2307 dso->kernel = dso_space; 2308 free(m.name); 2309 } 2310 2311 build_id__sprintf(&dso->bid, sbuild_id); 2312 pr_debug("build id event received for %s: %s [%zu]\n", 2313 dso->long_name, sbuild_id, size); 2314 dso__put(dso); 2315 } 2316 2317 err = 0; 2318 out: 2319 return err; 2320 } 2321 2322 static int perf_header__read_build_ids_abi_quirk(struct perf_header *header, 2323 int input, u64 offset, u64 size) 2324 { 2325 struct perf_session *session = container_of(header, struct perf_session, header); 2326 struct { 2327 struct perf_event_header header; 2328 u8 build_id[PERF_ALIGN(BUILD_ID_SIZE, sizeof(u64))]; 2329 char filename[0]; 2330 } old_bev; 2331 struct perf_record_header_build_id bev; 2332 char filename[PATH_MAX]; 2333 u64 limit = offset + size; 2334 2335 while (offset < limit) { 2336 ssize_t len; 2337 2338 if (readn(input, &old_bev, sizeof(old_bev)) != sizeof(old_bev)) 2339 return -1; 2340 2341 if (header->needs_swap) 2342 perf_event_header__bswap(&old_bev.header); 2343 2344 len = old_bev.header.size - sizeof(old_bev); 2345 if (readn(input, filename, len) != len) 2346 return -1; 2347 2348 bev.header = old_bev.header; 2349 2350 /* 2351 * As the pid is the missing value, we need to fill 2352 * it properly. The header.misc value give us nice hint. 2353 */ 2354 bev.pid = HOST_KERNEL_ID; 2355 if (bev.header.misc == PERF_RECORD_MISC_GUEST_USER || 2356 bev.header.misc == PERF_RECORD_MISC_GUEST_KERNEL) 2357 bev.pid = DEFAULT_GUEST_KERNEL_ID; 2358 2359 memcpy(bev.build_id, old_bev.build_id, sizeof(bev.build_id)); 2360 __event_process_build_id(&bev, filename, session); 2361 2362 offset += bev.header.size; 2363 } 2364 2365 return 0; 2366 } 2367 2368 static int perf_header__read_build_ids(struct perf_header *header, 2369 int input, u64 offset, u64 size) 2370 { 2371 struct perf_session *session = container_of(header, struct perf_session, header); 2372 struct perf_record_header_build_id bev; 2373 char filename[PATH_MAX]; 2374 u64 limit = offset + size, orig_offset = offset; 2375 int err = -1; 2376 2377 while (offset < limit) { 2378 ssize_t len; 2379 2380 if (readn(input, &bev, sizeof(bev)) != sizeof(bev)) 2381 goto out; 2382 2383 if (header->needs_swap) 2384 perf_event_header__bswap(&bev.header); 2385 2386 len = bev.header.size - sizeof(bev); 2387 if (readn(input, filename, len) != len) 2388 goto out; 2389 /* 2390 * The a1645ce1 changeset: 2391 * 2392 * "perf: 'perf kvm' tool for monitoring guest performance from host" 2393 * 2394 * Added a field to struct perf_record_header_build_id that broke the file 2395 * format. 2396 * 2397 * Since the kernel build-id is the first entry, process the 2398 * table using the old format if the well known 2399 * '[kernel.kallsyms]' string for the kernel build-id has the 2400 * first 4 characters chopped off (where the pid_t sits). 2401 */ 2402 if (memcmp(filename, "nel.kallsyms]", 13) == 0) { 2403 if (lseek(input, orig_offset, SEEK_SET) == (off_t)-1) 2404 return -1; 2405 return perf_header__read_build_ids_abi_quirk(header, input, offset, size); 2406 } 2407 2408 __event_process_build_id(&bev, filename, session); 2409 2410 offset += bev.header.size; 2411 } 2412 err = 0; 2413 out: 2414 return err; 2415 } 2416 2417 /* Macro for features that simply need to read and store a string. */ 2418 #define FEAT_PROCESS_STR_FUN(__feat, __feat_env) \ 2419 static int process_##__feat(struct feat_fd *ff, void *data __maybe_unused) \ 2420 {\ 2421 free(ff->ph->env.__feat_env); \ 2422 ff->ph->env.__feat_env = do_read_string(ff); \ 2423 return ff->ph->env.__feat_env ? 0 : -ENOMEM; \ 2424 } 2425 2426 FEAT_PROCESS_STR_FUN(hostname, hostname); 2427 FEAT_PROCESS_STR_FUN(osrelease, os_release); 2428 FEAT_PROCESS_STR_FUN(version, version); 2429 FEAT_PROCESS_STR_FUN(arch, arch); 2430 FEAT_PROCESS_STR_FUN(cpudesc, cpu_desc); 2431 FEAT_PROCESS_STR_FUN(cpuid, cpuid); 2432 2433 #ifdef HAVE_LIBTRACEEVENT 2434 static int process_tracing_data(struct feat_fd *ff, void *data) 2435 { 2436 ssize_t ret = trace_report(ff->fd, data, false); 2437 2438 return ret < 0 ? -1 : 0; 2439 } 2440 #endif 2441 2442 static int process_build_id(struct feat_fd *ff, void *data __maybe_unused) 2443 { 2444 if (perf_header__read_build_ids(ff->ph, ff->fd, ff->offset, ff->size)) 2445 pr_debug("Failed to read buildids, continuing...\n"); 2446 return 0; 2447 } 2448 2449 static int process_nrcpus(struct feat_fd *ff, void *data __maybe_unused) 2450 { 2451 int ret; 2452 u32 nr_cpus_avail, nr_cpus_online; 2453 2454 ret = do_read_u32(ff, &nr_cpus_avail); 2455 if (ret) 2456 return ret; 2457 2458 ret = do_read_u32(ff, &nr_cpus_online); 2459 if (ret) 2460 return ret; 2461 ff->ph->env.nr_cpus_avail = (int)nr_cpus_avail; 2462 ff->ph->env.nr_cpus_online = (int)nr_cpus_online; 2463 return 0; 2464 } 2465 2466 static int process_total_mem(struct feat_fd *ff, void *data __maybe_unused) 2467 { 2468 u64 total_mem; 2469 int ret; 2470 2471 ret = do_read_u64(ff, &total_mem); 2472 if (ret) 2473 return -1; 2474 ff->ph->env.total_mem = (unsigned long long)total_mem; 2475 return 0; 2476 } 2477 2478 static struct evsel *evlist__find_by_index(struct evlist *evlist, int idx) 2479 { 2480 struct evsel *evsel; 2481 2482 evlist__for_each_entry(evlist, evsel) { 2483 if (evsel->core.idx == idx) 2484 return evsel; 2485 } 2486 2487 return NULL; 2488 } 2489 2490 static void evlist__set_event_name(struct evlist *evlist, struct evsel *event) 2491 { 2492 struct evsel *evsel; 2493 2494 if (!event->name) 2495 return; 2496 2497 evsel = evlist__find_by_index(evlist, event->core.idx); 2498 if (!evsel) 2499 return; 2500 2501 if (evsel->name) 2502 return; 2503 2504 evsel->name = strdup(event->name); 2505 } 2506 2507 static int 2508 process_event_desc(struct feat_fd *ff, void *data __maybe_unused) 2509 { 2510 struct perf_session *session; 2511 struct evsel *evsel, *events = read_event_desc(ff); 2512 2513 if (!events) 2514 return 0; 2515 2516 session = container_of(ff->ph, struct perf_session, header); 2517 2518 if (session->data->is_pipe) { 2519 /* Save events for reading later by print_event_desc, 2520 * since they can't be read again in pipe mode. */ 2521 ff->events = events; 2522 } 2523 2524 for (evsel = events; evsel->core.attr.size; evsel++) 2525 evlist__set_event_name(session->evlist, evsel); 2526 2527 if (!session->data->is_pipe) 2528 free_event_desc(events); 2529 2530 return 0; 2531 } 2532 2533 static int process_cmdline(struct feat_fd *ff, void *data __maybe_unused) 2534 { 2535 char *str, *cmdline = NULL, **argv = NULL; 2536 u32 nr, i, len = 0; 2537 2538 if (do_read_u32(ff, &nr)) 2539 return -1; 2540 2541 ff->ph->env.nr_cmdline = nr; 2542 2543 cmdline = zalloc(ff->size + nr + 1); 2544 if (!cmdline) 2545 return -1; 2546 2547 argv = zalloc(sizeof(char *) * (nr + 1)); 2548 if (!argv) 2549 goto error; 2550 2551 for (i = 0; i < nr; i++) { 2552 str = do_read_string(ff); 2553 if (!str) 2554 goto error; 2555 2556 argv[i] = cmdline + len; 2557 memcpy(argv[i], str, strlen(str) + 1); 2558 len += strlen(str) + 1; 2559 free(str); 2560 } 2561 ff->ph->env.cmdline = cmdline; 2562 ff->ph->env.cmdline_argv = (const char **) argv; 2563 return 0; 2564 2565 error: 2566 free(argv); 2567 free(cmdline); 2568 return -1; 2569 } 2570 2571 static int process_cpu_topology(struct feat_fd *ff, void *data __maybe_unused) 2572 { 2573 u32 nr, i; 2574 char *str; 2575 struct strbuf sb; 2576 int cpu_nr = ff->ph->env.nr_cpus_avail; 2577 u64 size = 0; 2578 struct perf_header *ph = ff->ph; 2579 bool do_core_id_test = true; 2580 2581 ph->env.cpu = calloc(cpu_nr, sizeof(*ph->env.cpu)); 2582 if (!ph->env.cpu) 2583 return -1; 2584 2585 if (do_read_u32(ff, &nr)) 2586 goto free_cpu; 2587 2588 ph->env.nr_sibling_cores = nr; 2589 size += sizeof(u32); 2590 if (strbuf_init(&sb, 128) < 0) 2591 goto free_cpu; 2592 2593 for (i = 0; i < nr; i++) { 2594 str = do_read_string(ff); 2595 if (!str) 2596 goto error; 2597 2598 /* include a NULL character at the end */ 2599 if (strbuf_add(&sb, str, strlen(str) + 1) < 0) 2600 goto error; 2601 size += string_size(str); 2602 free(str); 2603 } 2604 ph->env.sibling_cores = strbuf_detach(&sb, NULL); 2605 2606 if (do_read_u32(ff, &nr)) 2607 return -1; 2608 2609 ph->env.nr_sibling_threads = nr; 2610 size += sizeof(u32); 2611 2612 for (i = 0; i < nr; i++) { 2613 str = do_read_string(ff); 2614 if (!str) 2615 goto error; 2616 2617 /* include a NULL character at the end */ 2618 if (strbuf_add(&sb, str, strlen(str) + 1) < 0) 2619 goto error; 2620 size += string_size(str); 2621 free(str); 2622 } 2623 ph->env.sibling_threads = strbuf_detach(&sb, NULL); 2624 2625 /* 2626 * The header may be from old perf, 2627 * which doesn't include core id and socket id information. 2628 */ 2629 if (ff->size <= size) { 2630 zfree(&ph->env.cpu); 2631 return 0; 2632 } 2633 2634 /* On s390 the socket_id number is not related to the numbers of cpus. 2635 * The socket_id number might be higher than the numbers of cpus. 2636 * This depends on the configuration. 2637 * AArch64 is the same. 2638 */ 2639 if (ph->env.arch && (!strncmp(ph->env.arch, "s390", 4) 2640 || !strncmp(ph->env.arch, "aarch64", 7))) 2641 do_core_id_test = false; 2642 2643 for (i = 0; i < (u32)cpu_nr; i++) { 2644 if (do_read_u32(ff, &nr)) 2645 goto free_cpu; 2646 2647 ph->env.cpu[i].core_id = nr; 2648 size += sizeof(u32); 2649 2650 if (do_read_u32(ff, &nr)) 2651 goto free_cpu; 2652 2653 if (do_core_id_test && nr != (u32)-1 && nr > (u32)cpu_nr) { 2654 pr_debug("socket_id number is too big." 2655 "You may need to upgrade the perf tool.\n"); 2656 goto free_cpu; 2657 } 2658 2659 ph->env.cpu[i].socket_id = nr; 2660 size += sizeof(u32); 2661 } 2662 2663 /* 2664 * The header may be from old perf, 2665 * which doesn't include die information. 2666 */ 2667 if (ff->size <= size) 2668 return 0; 2669 2670 if (do_read_u32(ff, &nr)) 2671 return -1; 2672 2673 ph->env.nr_sibling_dies = nr; 2674 size += sizeof(u32); 2675 2676 for (i = 0; i < nr; i++) { 2677 str = do_read_string(ff); 2678 if (!str) 2679 goto error; 2680 2681 /* include a NULL character at the end */ 2682 if (strbuf_add(&sb, str, strlen(str) + 1) < 0) 2683 goto error; 2684 size += string_size(str); 2685 free(str); 2686 } 2687 ph->env.sibling_dies = strbuf_detach(&sb, NULL); 2688 2689 for (i = 0; i < (u32)cpu_nr; i++) { 2690 if (do_read_u32(ff, &nr)) 2691 goto free_cpu; 2692 2693 ph->env.cpu[i].die_id = nr; 2694 } 2695 2696 return 0; 2697 2698 error: 2699 strbuf_release(&sb); 2700 free_cpu: 2701 zfree(&ph->env.cpu); 2702 return -1; 2703 } 2704 2705 static int process_numa_topology(struct feat_fd *ff, void *data __maybe_unused) 2706 { 2707 struct numa_node *nodes, *n; 2708 u32 nr, i; 2709 char *str; 2710 2711 /* nr nodes */ 2712 if (do_read_u32(ff, &nr)) 2713 return -1; 2714 2715 nodes = zalloc(sizeof(*nodes) * nr); 2716 if (!nodes) 2717 return -ENOMEM; 2718 2719 for (i = 0; i < nr; i++) { 2720 n = &nodes[i]; 2721 2722 /* node number */ 2723 if (do_read_u32(ff, &n->node)) 2724 goto error; 2725 2726 if (do_read_u64(ff, &n->mem_total)) 2727 goto error; 2728 2729 if (do_read_u64(ff, &n->mem_free)) 2730 goto error; 2731 2732 str = do_read_string(ff); 2733 if (!str) 2734 goto error; 2735 2736 n->map = perf_cpu_map__new(str); 2737 if (!n->map) 2738 goto error; 2739 2740 free(str); 2741 } 2742 ff->ph->env.nr_numa_nodes = nr; 2743 ff->ph->env.numa_nodes = nodes; 2744 return 0; 2745 2746 error: 2747 free(nodes); 2748 return -1; 2749 } 2750 2751 static int process_pmu_mappings(struct feat_fd *ff, void *data __maybe_unused) 2752 { 2753 char *name; 2754 u32 pmu_num; 2755 u32 type; 2756 struct strbuf sb; 2757 2758 if (do_read_u32(ff, &pmu_num)) 2759 return -1; 2760 2761 if (!pmu_num) { 2762 pr_debug("pmu mappings not available\n"); 2763 return 0; 2764 } 2765 2766 ff->ph->env.nr_pmu_mappings = pmu_num; 2767 if (strbuf_init(&sb, 128) < 0) 2768 return -1; 2769 2770 while (pmu_num) { 2771 if (do_read_u32(ff, &type)) 2772 goto error; 2773 2774 name = do_read_string(ff); 2775 if (!name) 2776 goto error; 2777 2778 if (strbuf_addf(&sb, "%u:%s", type, name) < 0) 2779 goto error; 2780 /* include a NULL character at the end */ 2781 if (strbuf_add(&sb, "", 1) < 0) 2782 goto error; 2783 2784 if (!strcmp(name, "msr")) 2785 ff->ph->env.msr_pmu_type = type; 2786 2787 free(name); 2788 pmu_num--; 2789 } 2790 ff->ph->env.pmu_mappings = strbuf_detach(&sb, NULL); 2791 return 0; 2792 2793 error: 2794 strbuf_release(&sb); 2795 return -1; 2796 } 2797 2798 static int process_group_desc(struct feat_fd *ff, void *data __maybe_unused) 2799 { 2800 size_t ret = -1; 2801 u32 i, nr, nr_groups; 2802 struct perf_session *session; 2803 struct evsel *evsel, *leader = NULL; 2804 struct group_desc { 2805 char *name; 2806 u32 leader_idx; 2807 u32 nr_members; 2808 } *desc; 2809 2810 if (do_read_u32(ff, &nr_groups)) 2811 return -1; 2812 2813 ff->ph->env.nr_groups = nr_groups; 2814 if (!nr_groups) { 2815 pr_debug("group desc not available\n"); 2816 return 0; 2817 } 2818 2819 desc = calloc(nr_groups, sizeof(*desc)); 2820 if (!desc) 2821 return -1; 2822 2823 for (i = 0; i < nr_groups; i++) { 2824 desc[i].name = do_read_string(ff); 2825 if (!desc[i].name) 2826 goto out_free; 2827 2828 if (do_read_u32(ff, &desc[i].leader_idx)) 2829 goto out_free; 2830 2831 if (do_read_u32(ff, &desc[i].nr_members)) 2832 goto out_free; 2833 } 2834 2835 /* 2836 * Rebuild group relationship based on the group_desc 2837 */ 2838 session = container_of(ff->ph, struct perf_session, header); 2839 2840 i = nr = 0; 2841 evlist__for_each_entry(session->evlist, evsel) { 2842 if (i < nr_groups && evsel->core.idx == (int) desc[i].leader_idx) { 2843 evsel__set_leader(evsel, evsel); 2844 /* {anon_group} is a dummy name */ 2845 if (strcmp(desc[i].name, "{anon_group}")) { 2846 evsel->group_name = desc[i].name; 2847 desc[i].name = NULL; 2848 } 2849 evsel->core.nr_members = desc[i].nr_members; 2850 2851 if (i >= nr_groups || nr > 0) { 2852 pr_debug("invalid group desc\n"); 2853 goto out_free; 2854 } 2855 2856 leader = evsel; 2857 nr = evsel->core.nr_members - 1; 2858 i++; 2859 } else if (nr) { 2860 /* This is a group member */ 2861 evsel__set_leader(evsel, leader); 2862 2863 nr--; 2864 } 2865 } 2866 2867 if (i != nr_groups || nr != 0) { 2868 pr_debug("invalid group desc\n"); 2869 goto out_free; 2870 } 2871 2872 ret = 0; 2873 out_free: 2874 for (i = 0; i < nr_groups; i++) 2875 zfree(&desc[i].name); 2876 free(desc); 2877 2878 return ret; 2879 } 2880 2881 static int process_auxtrace(struct feat_fd *ff, void *data __maybe_unused) 2882 { 2883 struct perf_session *session; 2884 int err; 2885 2886 session = container_of(ff->ph, struct perf_session, header); 2887 2888 err = auxtrace_index__process(ff->fd, ff->size, session, 2889 ff->ph->needs_swap); 2890 if (err < 0) 2891 pr_err("Failed to process auxtrace index\n"); 2892 return err; 2893 } 2894 2895 static int process_cache(struct feat_fd *ff, void *data __maybe_unused) 2896 { 2897 struct cpu_cache_level *caches; 2898 u32 cnt, i, version; 2899 2900 if (do_read_u32(ff, &version)) 2901 return -1; 2902 2903 if (version != 1) 2904 return -1; 2905 2906 if (do_read_u32(ff, &cnt)) 2907 return -1; 2908 2909 caches = zalloc(sizeof(*caches) * cnt); 2910 if (!caches) 2911 return -1; 2912 2913 for (i = 0; i < cnt; i++) { 2914 struct cpu_cache_level c; 2915 2916 #define _R(v) \ 2917 if (do_read_u32(ff, &c.v))\ 2918 goto out_free_caches; \ 2919 2920 _R(level) 2921 _R(line_size) 2922 _R(sets) 2923 _R(ways) 2924 #undef _R 2925 2926 #define _R(v) \ 2927 c.v = do_read_string(ff); \ 2928 if (!c.v) \ 2929 goto out_free_caches; 2930 2931 _R(type) 2932 _R(size) 2933 _R(map) 2934 #undef _R 2935 2936 caches[i] = c; 2937 } 2938 2939 ff->ph->env.caches = caches; 2940 ff->ph->env.caches_cnt = cnt; 2941 return 0; 2942 out_free_caches: 2943 free(caches); 2944 return -1; 2945 } 2946 2947 static int process_sample_time(struct feat_fd *ff, void *data __maybe_unused) 2948 { 2949 struct perf_session *session; 2950 u64 first_sample_time, last_sample_time; 2951 int ret; 2952 2953 session = container_of(ff->ph, struct perf_session, header); 2954 2955 ret = do_read_u64(ff, &first_sample_time); 2956 if (ret) 2957 return -1; 2958 2959 ret = do_read_u64(ff, &last_sample_time); 2960 if (ret) 2961 return -1; 2962 2963 session->evlist->first_sample_time = first_sample_time; 2964 session->evlist->last_sample_time = last_sample_time; 2965 return 0; 2966 } 2967 2968 static int process_mem_topology(struct feat_fd *ff, 2969 void *data __maybe_unused) 2970 { 2971 struct memory_node *nodes; 2972 u64 version, i, nr, bsize; 2973 int ret = -1; 2974 2975 if (do_read_u64(ff, &version)) 2976 return -1; 2977 2978 if (version != 1) 2979 return -1; 2980 2981 if (do_read_u64(ff, &bsize)) 2982 return -1; 2983 2984 if (do_read_u64(ff, &nr)) 2985 return -1; 2986 2987 nodes = zalloc(sizeof(*nodes) * nr); 2988 if (!nodes) 2989 return -1; 2990 2991 for (i = 0; i < nr; i++) { 2992 struct memory_node n; 2993 2994 #define _R(v) \ 2995 if (do_read_u64(ff, &n.v)) \ 2996 goto out; \ 2997 2998 _R(node) 2999 _R(size) 3000 3001 #undef _R 3002 3003 if (do_read_bitmap(ff, &n.set, &n.size)) 3004 goto out; 3005 3006 nodes[i] = n; 3007 } 3008 3009 ff->ph->env.memory_bsize = bsize; 3010 ff->ph->env.memory_nodes = nodes; 3011 ff->ph->env.nr_memory_nodes = nr; 3012 ret = 0; 3013 3014 out: 3015 if (ret) 3016 free(nodes); 3017 return ret; 3018 } 3019 3020 static int process_clockid(struct feat_fd *ff, 3021 void *data __maybe_unused) 3022 { 3023 if (do_read_u64(ff, &ff->ph->env.clock.clockid_res_ns)) 3024 return -1; 3025 3026 return 0; 3027 } 3028 3029 static int process_clock_data(struct feat_fd *ff, 3030 void *_data __maybe_unused) 3031 { 3032 u32 data32; 3033 u64 data64; 3034 3035 /* version */ 3036 if (do_read_u32(ff, &data32)) 3037 return -1; 3038 3039 if (data32 != 1) 3040 return -1; 3041 3042 /* clockid */ 3043 if (do_read_u32(ff, &data32)) 3044 return -1; 3045 3046 ff->ph->env.clock.clockid = data32; 3047 3048 /* TOD ref time */ 3049 if (do_read_u64(ff, &data64)) 3050 return -1; 3051 3052 ff->ph->env.clock.tod_ns = data64; 3053 3054 /* clockid ref time */ 3055 if (do_read_u64(ff, &data64)) 3056 return -1; 3057 3058 ff->ph->env.clock.clockid_ns = data64; 3059 ff->ph->env.clock.enabled = true; 3060 return 0; 3061 } 3062 3063 static int process_hybrid_topology(struct feat_fd *ff, 3064 void *data __maybe_unused) 3065 { 3066 struct hybrid_node *nodes, *n; 3067 u32 nr, i; 3068 3069 /* nr nodes */ 3070 if (do_read_u32(ff, &nr)) 3071 return -1; 3072 3073 nodes = zalloc(sizeof(*nodes) * nr); 3074 if (!nodes) 3075 return -ENOMEM; 3076 3077 for (i = 0; i < nr; i++) { 3078 n = &nodes[i]; 3079 3080 n->pmu_name = do_read_string(ff); 3081 if (!n->pmu_name) 3082 goto error; 3083 3084 n->cpus = do_read_string(ff); 3085 if (!n->cpus) 3086 goto error; 3087 } 3088 3089 ff->ph->env.nr_hybrid_nodes = nr; 3090 ff->ph->env.hybrid_nodes = nodes; 3091 return 0; 3092 3093 error: 3094 for (i = 0; i < nr; i++) { 3095 free(nodes[i].pmu_name); 3096 free(nodes[i].cpus); 3097 } 3098 3099 free(nodes); 3100 return -1; 3101 } 3102 3103 static int process_dir_format(struct feat_fd *ff, 3104 void *_data __maybe_unused) 3105 { 3106 struct perf_session *session; 3107 struct perf_data *data; 3108 3109 session = container_of(ff->ph, struct perf_session, header); 3110 data = session->data; 3111 3112 if (WARN_ON(!perf_data__is_dir(data))) 3113 return -1; 3114 3115 return do_read_u64(ff, &data->dir.version); 3116 } 3117 3118 #ifdef HAVE_LIBBPF_SUPPORT 3119 static int process_bpf_prog_info(struct feat_fd *ff, void *data __maybe_unused) 3120 { 3121 struct bpf_prog_info_node *info_node; 3122 struct perf_env *env = &ff->ph->env; 3123 struct perf_bpil *info_linear; 3124 u32 count, i; 3125 int err = -1; 3126 3127 if (ff->ph->needs_swap) { 3128 pr_warning("interpreting bpf_prog_info from systems with endianness is not yet supported\n"); 3129 return 0; 3130 } 3131 3132 if (do_read_u32(ff, &count)) 3133 return -1; 3134 3135 down_write(&env->bpf_progs.lock); 3136 3137 for (i = 0; i < count; ++i) { 3138 u32 info_len, data_len; 3139 3140 info_linear = NULL; 3141 info_node = NULL; 3142 if (do_read_u32(ff, &info_len)) 3143 goto out; 3144 if (do_read_u32(ff, &data_len)) 3145 goto out; 3146 3147 if (info_len > sizeof(struct bpf_prog_info)) { 3148 pr_warning("detected invalid bpf_prog_info\n"); 3149 goto out; 3150 } 3151 3152 info_linear = malloc(sizeof(struct perf_bpil) + 3153 data_len); 3154 if (!info_linear) 3155 goto out; 3156 info_linear->info_len = sizeof(struct bpf_prog_info); 3157 info_linear->data_len = data_len; 3158 if (do_read_u64(ff, (u64 *)(&info_linear->arrays))) 3159 goto out; 3160 if (__do_read(ff, &info_linear->info, info_len)) 3161 goto out; 3162 if (info_len < sizeof(struct bpf_prog_info)) 3163 memset(((void *)(&info_linear->info)) + info_len, 0, 3164 sizeof(struct bpf_prog_info) - info_len); 3165 3166 if (__do_read(ff, info_linear->data, data_len)) 3167 goto out; 3168 3169 info_node = malloc(sizeof(struct bpf_prog_info_node)); 3170 if (!info_node) 3171 goto out; 3172 3173 /* after reading from file, translate offset to address */ 3174 bpil_offs_to_addr(info_linear); 3175 info_node->info_linear = info_linear; 3176 perf_env__insert_bpf_prog_info(env, info_node); 3177 } 3178 3179 up_write(&env->bpf_progs.lock); 3180 return 0; 3181 out: 3182 free(info_linear); 3183 free(info_node); 3184 up_write(&env->bpf_progs.lock); 3185 return err; 3186 } 3187 3188 static int process_bpf_btf(struct feat_fd *ff, void *data __maybe_unused) 3189 { 3190 struct perf_env *env = &ff->ph->env; 3191 struct btf_node *node = NULL; 3192 u32 count, i; 3193 int err = -1; 3194 3195 if (ff->ph->needs_swap) { 3196 pr_warning("interpreting btf from systems with endianness is not yet supported\n"); 3197 return 0; 3198 } 3199 3200 if (do_read_u32(ff, &count)) 3201 return -1; 3202 3203 down_write(&env->bpf_progs.lock); 3204 3205 for (i = 0; i < count; ++i) { 3206 u32 id, data_size; 3207 3208 if (do_read_u32(ff, &id)) 3209 goto out; 3210 if (do_read_u32(ff, &data_size)) 3211 goto out; 3212 3213 node = malloc(sizeof(struct btf_node) + data_size); 3214 if (!node) 3215 goto out; 3216 3217 node->id = id; 3218 node->data_size = data_size; 3219 3220 if (__do_read(ff, node->data, data_size)) 3221 goto out; 3222 3223 perf_env__insert_btf(env, node); 3224 node = NULL; 3225 } 3226 3227 err = 0; 3228 out: 3229 up_write(&env->bpf_progs.lock); 3230 free(node); 3231 return err; 3232 } 3233 #endif // HAVE_LIBBPF_SUPPORT 3234 3235 static int process_compressed(struct feat_fd *ff, 3236 void *data __maybe_unused) 3237 { 3238 if (do_read_u32(ff, &(ff->ph->env.comp_ver))) 3239 return -1; 3240 3241 if (do_read_u32(ff, &(ff->ph->env.comp_type))) 3242 return -1; 3243 3244 if (do_read_u32(ff, &(ff->ph->env.comp_level))) 3245 return -1; 3246 3247 if (do_read_u32(ff, &(ff->ph->env.comp_ratio))) 3248 return -1; 3249 3250 if (do_read_u32(ff, &(ff->ph->env.comp_mmap_len))) 3251 return -1; 3252 3253 return 0; 3254 } 3255 3256 static int __process_pmu_caps(struct feat_fd *ff, int *nr_caps, 3257 char ***caps, unsigned int *max_branches) 3258 { 3259 char *name, *value, *ptr; 3260 u32 nr_pmu_caps, i; 3261 3262 *nr_caps = 0; 3263 *caps = NULL; 3264 3265 if (do_read_u32(ff, &nr_pmu_caps)) 3266 return -1; 3267 3268 if (!nr_pmu_caps) 3269 return 0; 3270 3271 *caps = zalloc(sizeof(char *) * nr_pmu_caps); 3272 if (!*caps) 3273 return -1; 3274 3275 for (i = 0; i < nr_pmu_caps; i++) { 3276 name = do_read_string(ff); 3277 if (!name) 3278 goto error; 3279 3280 value = do_read_string(ff); 3281 if (!value) 3282 goto free_name; 3283 3284 if (asprintf(&ptr, "%s=%s", name, value) < 0) 3285 goto free_value; 3286 3287 (*caps)[i] = ptr; 3288 3289 if (!strcmp(name, "branches")) 3290 *max_branches = atoi(value); 3291 3292 free(value); 3293 free(name); 3294 } 3295 *nr_caps = nr_pmu_caps; 3296 return 0; 3297 3298 free_value: 3299 free(value); 3300 free_name: 3301 free(name); 3302 error: 3303 for (; i > 0; i--) 3304 free((*caps)[i - 1]); 3305 free(*caps); 3306 *caps = NULL; 3307 *nr_caps = 0; 3308 return -1; 3309 } 3310 3311 static int process_cpu_pmu_caps(struct feat_fd *ff, 3312 void *data __maybe_unused) 3313 { 3314 int ret = __process_pmu_caps(ff, &ff->ph->env.nr_cpu_pmu_caps, 3315 &ff->ph->env.cpu_pmu_caps, 3316 &ff->ph->env.max_branches); 3317 3318 if (!ret && !ff->ph->env.cpu_pmu_caps) 3319 pr_debug("cpu pmu capabilities not available\n"); 3320 return ret; 3321 } 3322 3323 static int process_pmu_caps(struct feat_fd *ff, void *data __maybe_unused) 3324 { 3325 struct pmu_caps *pmu_caps; 3326 u32 nr_pmu, i; 3327 int ret; 3328 int j; 3329 3330 if (do_read_u32(ff, &nr_pmu)) 3331 return -1; 3332 3333 if (!nr_pmu) { 3334 pr_debug("pmu capabilities not available\n"); 3335 return 0; 3336 } 3337 3338 pmu_caps = zalloc(sizeof(*pmu_caps) * nr_pmu); 3339 if (!pmu_caps) 3340 return -ENOMEM; 3341 3342 for (i = 0; i < nr_pmu; i++) { 3343 ret = __process_pmu_caps(ff, &pmu_caps[i].nr_caps, 3344 &pmu_caps[i].caps, 3345 &pmu_caps[i].max_branches); 3346 if (ret) 3347 goto err; 3348 3349 pmu_caps[i].pmu_name = do_read_string(ff); 3350 if (!pmu_caps[i].pmu_name) { 3351 ret = -1; 3352 goto err; 3353 } 3354 if (!pmu_caps[i].nr_caps) { 3355 pr_debug("%s pmu capabilities not available\n", 3356 pmu_caps[i].pmu_name); 3357 } 3358 } 3359 3360 ff->ph->env.nr_pmus_with_caps = nr_pmu; 3361 ff->ph->env.pmu_caps = pmu_caps; 3362 return 0; 3363 3364 err: 3365 for (i = 0; i < nr_pmu; i++) { 3366 for (j = 0; j < pmu_caps[i].nr_caps; j++) 3367 free(pmu_caps[i].caps[j]); 3368 free(pmu_caps[i].caps); 3369 free(pmu_caps[i].pmu_name); 3370 } 3371 3372 free(pmu_caps); 3373 return ret; 3374 } 3375 3376 #define FEAT_OPR(n, func, __full_only) \ 3377 [HEADER_##n] = { \ 3378 .name = __stringify(n), \ 3379 .write = write_##func, \ 3380 .print = print_##func, \ 3381 .full_only = __full_only, \ 3382 .process = process_##func, \ 3383 .synthesize = true \ 3384 } 3385 3386 #define FEAT_OPN(n, func, __full_only) \ 3387 [HEADER_##n] = { \ 3388 .name = __stringify(n), \ 3389 .write = write_##func, \ 3390 .print = print_##func, \ 3391 .full_only = __full_only, \ 3392 .process = process_##func \ 3393 } 3394 3395 /* feature_ops not implemented: */ 3396 #define print_tracing_data NULL 3397 #define print_build_id NULL 3398 3399 #define process_branch_stack NULL 3400 #define process_stat NULL 3401 3402 // Only used in util/synthetic-events.c 3403 const struct perf_header_feature_ops feat_ops[HEADER_LAST_FEATURE]; 3404 3405 const struct perf_header_feature_ops feat_ops[HEADER_LAST_FEATURE] = { 3406 #ifdef HAVE_LIBTRACEEVENT 3407 FEAT_OPN(TRACING_DATA, tracing_data, false), 3408 #endif 3409 FEAT_OPN(BUILD_ID, build_id, false), 3410 FEAT_OPR(HOSTNAME, hostname, false), 3411 FEAT_OPR(OSRELEASE, osrelease, false), 3412 FEAT_OPR(VERSION, version, false), 3413 FEAT_OPR(ARCH, arch, false), 3414 FEAT_OPR(NRCPUS, nrcpus, false), 3415 FEAT_OPR(CPUDESC, cpudesc, false), 3416 FEAT_OPR(CPUID, cpuid, false), 3417 FEAT_OPR(TOTAL_MEM, total_mem, false), 3418 FEAT_OPR(EVENT_DESC, event_desc, false), 3419 FEAT_OPR(CMDLINE, cmdline, false), 3420 FEAT_OPR(CPU_TOPOLOGY, cpu_topology, true), 3421 FEAT_OPR(NUMA_TOPOLOGY, numa_topology, true), 3422 FEAT_OPN(BRANCH_STACK, branch_stack, false), 3423 FEAT_OPR(PMU_MAPPINGS, pmu_mappings, false), 3424 FEAT_OPR(GROUP_DESC, group_desc, false), 3425 FEAT_OPN(AUXTRACE, auxtrace, false), 3426 FEAT_OPN(STAT, stat, false), 3427 FEAT_OPN(CACHE, cache, true), 3428 FEAT_OPR(SAMPLE_TIME, sample_time, false), 3429 FEAT_OPR(MEM_TOPOLOGY, mem_topology, true), 3430 FEAT_OPR(CLOCKID, clockid, false), 3431 FEAT_OPN(DIR_FORMAT, dir_format, false), 3432 #ifdef HAVE_LIBBPF_SUPPORT 3433 FEAT_OPR(BPF_PROG_INFO, bpf_prog_info, false), 3434 FEAT_OPR(BPF_BTF, bpf_btf, false), 3435 #endif 3436 FEAT_OPR(COMPRESSED, compressed, false), 3437 FEAT_OPR(CPU_PMU_CAPS, cpu_pmu_caps, false), 3438 FEAT_OPR(CLOCK_DATA, clock_data, false), 3439 FEAT_OPN(HYBRID_TOPOLOGY, hybrid_topology, true), 3440 FEAT_OPR(PMU_CAPS, pmu_caps, false), 3441 }; 3442 3443 struct header_print_data { 3444 FILE *fp; 3445 bool full; /* extended list of headers */ 3446 }; 3447 3448 static int perf_file_section__fprintf_info(struct perf_file_section *section, 3449 struct perf_header *ph, 3450 int feat, int fd, void *data) 3451 { 3452 struct header_print_data *hd = data; 3453 struct feat_fd ff; 3454 3455 if (lseek(fd, section->offset, SEEK_SET) == (off_t)-1) { 3456 pr_debug("Failed to lseek to %" PRIu64 " offset for feature " 3457 "%d, continuing...\n", section->offset, feat); 3458 return 0; 3459 } 3460 if (feat >= HEADER_LAST_FEATURE) { 3461 pr_warning("unknown feature %d\n", feat); 3462 return 0; 3463 } 3464 if (!feat_ops[feat].print) 3465 return 0; 3466 3467 ff = (struct feat_fd) { 3468 .fd = fd, 3469 .ph = ph, 3470 }; 3471 3472 if (!feat_ops[feat].full_only || hd->full) 3473 feat_ops[feat].print(&ff, hd->fp); 3474 else 3475 fprintf(hd->fp, "# %s info available, use -I to display\n", 3476 feat_ops[feat].name); 3477 3478 return 0; 3479 } 3480 3481 int perf_header__fprintf_info(struct perf_session *session, FILE *fp, bool full) 3482 { 3483 struct header_print_data hd; 3484 struct perf_header *header = &session->header; 3485 int fd = perf_data__fd(session->data); 3486 struct stat st; 3487 time_t stctime; 3488 int ret, bit; 3489 3490 hd.fp = fp; 3491 hd.full = full; 3492 3493 ret = fstat(fd, &st); 3494 if (ret == -1) 3495 return -1; 3496 3497 stctime = st.st_mtime; 3498 fprintf(fp, "# captured on : %s", ctime(&stctime)); 3499 3500 fprintf(fp, "# header version : %u\n", header->version); 3501 fprintf(fp, "# data offset : %" PRIu64 "\n", header->data_offset); 3502 fprintf(fp, "# data size : %" PRIu64 "\n", header->data_size); 3503 fprintf(fp, "# feat offset : %" PRIu64 "\n", header->feat_offset); 3504 3505 perf_header__process_sections(header, fd, &hd, 3506 perf_file_section__fprintf_info); 3507 3508 if (session->data->is_pipe) 3509 return 0; 3510 3511 fprintf(fp, "# missing features: "); 3512 for_each_clear_bit(bit, header->adds_features, HEADER_LAST_FEATURE) { 3513 if (bit) 3514 fprintf(fp, "%s ", feat_ops[bit].name); 3515 } 3516 3517 fprintf(fp, "\n"); 3518 return 0; 3519 } 3520 3521 struct header_fw { 3522 struct feat_writer fw; 3523 struct feat_fd *ff; 3524 }; 3525 3526 static int feat_writer_cb(struct feat_writer *fw, void *buf, size_t sz) 3527 { 3528 struct header_fw *h = container_of(fw, struct header_fw, fw); 3529 3530 return do_write(h->ff, buf, sz); 3531 } 3532 3533 static int do_write_feat(struct feat_fd *ff, int type, 3534 struct perf_file_section **p, 3535 struct evlist *evlist, 3536 struct feat_copier *fc) 3537 { 3538 int err; 3539 int ret = 0; 3540 3541 if (perf_header__has_feat(ff->ph, type)) { 3542 if (!feat_ops[type].write) 3543 return -1; 3544 3545 if (WARN(ff->buf, "Error: calling %s in pipe-mode.\n", __func__)) 3546 return -1; 3547 3548 (*p)->offset = lseek(ff->fd, 0, SEEK_CUR); 3549 3550 /* 3551 * Hook to let perf inject copy features sections from the input 3552 * file. 3553 */ 3554 if (fc && fc->copy) { 3555 struct header_fw h = { 3556 .fw.write = feat_writer_cb, 3557 .ff = ff, 3558 }; 3559 3560 /* ->copy() returns 0 if the feature was not copied */ 3561 err = fc->copy(fc, type, &h.fw); 3562 } else { 3563 err = 0; 3564 } 3565 if (!err) 3566 err = feat_ops[type].write(ff, evlist); 3567 if (err < 0) { 3568 pr_debug("failed to write feature %s\n", feat_ops[type].name); 3569 3570 /* undo anything written */ 3571 lseek(ff->fd, (*p)->offset, SEEK_SET); 3572 3573 return -1; 3574 } 3575 (*p)->size = lseek(ff->fd, 0, SEEK_CUR) - (*p)->offset; 3576 (*p)++; 3577 } 3578 return ret; 3579 } 3580 3581 static int perf_header__adds_write(struct perf_header *header, 3582 struct evlist *evlist, int fd, 3583 struct feat_copier *fc) 3584 { 3585 int nr_sections; 3586 struct feat_fd ff; 3587 struct perf_file_section *feat_sec, *p; 3588 int sec_size; 3589 u64 sec_start; 3590 int feat; 3591 int err; 3592 3593 ff = (struct feat_fd){ 3594 .fd = fd, 3595 .ph = header, 3596 }; 3597 3598 nr_sections = bitmap_weight(header->adds_features, HEADER_FEAT_BITS); 3599 if (!nr_sections) 3600 return 0; 3601 3602 feat_sec = p = calloc(nr_sections, sizeof(*feat_sec)); 3603 if (feat_sec == NULL) 3604 return -ENOMEM; 3605 3606 sec_size = sizeof(*feat_sec) * nr_sections; 3607 3608 sec_start = header->feat_offset; 3609 lseek(fd, sec_start + sec_size, SEEK_SET); 3610 3611 for_each_set_bit(feat, header->adds_features, HEADER_FEAT_BITS) { 3612 if (do_write_feat(&ff, feat, &p, evlist, fc)) 3613 perf_header__clear_feat(header, feat); 3614 } 3615 3616 lseek(fd, sec_start, SEEK_SET); 3617 /* 3618 * may write more than needed due to dropped feature, but 3619 * this is okay, reader will skip the missing entries 3620 */ 3621 err = do_write(&ff, feat_sec, sec_size); 3622 if (err < 0) 3623 pr_debug("failed to write feature section\n"); 3624 free(feat_sec); 3625 return err; 3626 } 3627 3628 int perf_header__write_pipe(int fd) 3629 { 3630 struct perf_pipe_file_header f_header; 3631 struct feat_fd ff; 3632 int err; 3633 3634 ff = (struct feat_fd){ .fd = fd }; 3635 3636 f_header = (struct perf_pipe_file_header){ 3637 .magic = PERF_MAGIC, 3638 .size = sizeof(f_header), 3639 }; 3640 3641 err = do_write(&ff, &f_header, sizeof(f_header)); 3642 if (err < 0) { 3643 pr_debug("failed to write perf pipe header\n"); 3644 return err; 3645 } 3646 3647 return 0; 3648 } 3649 3650 static int perf_session__do_write_header(struct perf_session *session, 3651 struct evlist *evlist, 3652 int fd, bool at_exit, 3653 struct feat_copier *fc) 3654 { 3655 struct perf_file_header f_header; 3656 struct perf_file_attr f_attr; 3657 struct perf_header *header = &session->header; 3658 struct evsel *evsel; 3659 struct feat_fd ff; 3660 u64 attr_offset; 3661 int err; 3662 3663 ff = (struct feat_fd){ .fd = fd}; 3664 lseek(fd, sizeof(f_header), SEEK_SET); 3665 3666 evlist__for_each_entry(session->evlist, evsel) { 3667 evsel->id_offset = lseek(fd, 0, SEEK_CUR); 3668 err = do_write(&ff, evsel->core.id, evsel->core.ids * sizeof(u64)); 3669 if (err < 0) { 3670 pr_debug("failed to write perf header\n"); 3671 return err; 3672 } 3673 } 3674 3675 attr_offset = lseek(ff.fd, 0, SEEK_CUR); 3676 3677 evlist__for_each_entry(evlist, evsel) { 3678 if (evsel->core.attr.size < sizeof(evsel->core.attr)) { 3679 /* 3680 * We are likely in "perf inject" and have read 3681 * from an older file. Update attr size so that 3682 * reader gets the right offset to the ids. 3683 */ 3684 evsel->core.attr.size = sizeof(evsel->core.attr); 3685 } 3686 f_attr = (struct perf_file_attr){ 3687 .attr = evsel->core.attr, 3688 .ids = { 3689 .offset = evsel->id_offset, 3690 .size = evsel->core.ids * sizeof(u64), 3691 } 3692 }; 3693 err = do_write(&ff, &f_attr, sizeof(f_attr)); 3694 if (err < 0) { 3695 pr_debug("failed to write perf header attribute\n"); 3696 return err; 3697 } 3698 } 3699 3700 if (!header->data_offset) 3701 header->data_offset = lseek(fd, 0, SEEK_CUR); 3702 header->feat_offset = header->data_offset + header->data_size; 3703 3704 if (at_exit) { 3705 err = perf_header__adds_write(header, evlist, fd, fc); 3706 if (err < 0) 3707 return err; 3708 } 3709 3710 f_header = (struct perf_file_header){ 3711 .magic = PERF_MAGIC, 3712 .size = sizeof(f_header), 3713 .attr_size = sizeof(f_attr), 3714 .attrs = { 3715 .offset = attr_offset, 3716 .size = evlist->core.nr_entries * sizeof(f_attr), 3717 }, 3718 .data = { 3719 .offset = header->data_offset, 3720 .size = header->data_size, 3721 }, 3722 /* event_types is ignored, store zeros */ 3723 }; 3724 3725 memcpy(&f_header.adds_features, &header->adds_features, sizeof(header->adds_features)); 3726 3727 lseek(fd, 0, SEEK_SET); 3728 err = do_write(&ff, &f_header, sizeof(f_header)); 3729 if (err < 0) { 3730 pr_debug("failed to write perf header\n"); 3731 return err; 3732 } 3733 lseek(fd, header->data_offset + header->data_size, SEEK_SET); 3734 3735 return 0; 3736 } 3737 3738 int perf_session__write_header(struct perf_session *session, 3739 struct evlist *evlist, 3740 int fd, bool at_exit) 3741 { 3742 return perf_session__do_write_header(session, evlist, fd, at_exit, NULL); 3743 } 3744 3745 size_t perf_session__data_offset(const struct evlist *evlist) 3746 { 3747 struct evsel *evsel; 3748 size_t data_offset; 3749 3750 data_offset = sizeof(struct perf_file_header); 3751 evlist__for_each_entry(evlist, evsel) { 3752 data_offset += evsel->core.ids * sizeof(u64); 3753 } 3754 data_offset += evlist->core.nr_entries * sizeof(struct perf_file_attr); 3755 3756 return data_offset; 3757 } 3758 3759 int perf_session__inject_header(struct perf_session *session, 3760 struct evlist *evlist, 3761 int fd, 3762 struct feat_copier *fc) 3763 { 3764 return perf_session__do_write_header(session, evlist, fd, true, fc); 3765 } 3766 3767 static int perf_header__getbuffer64(struct perf_header *header, 3768 int fd, void *buf, size_t size) 3769 { 3770 if (readn(fd, buf, size) <= 0) 3771 return -1; 3772 3773 if (header->needs_swap) 3774 mem_bswap_64(buf, size); 3775 3776 return 0; 3777 } 3778 3779 int perf_header__process_sections(struct perf_header *header, int fd, 3780 void *data, 3781 int (*process)(struct perf_file_section *section, 3782 struct perf_header *ph, 3783 int feat, int fd, void *data)) 3784 { 3785 struct perf_file_section *feat_sec, *sec; 3786 int nr_sections; 3787 int sec_size; 3788 int feat; 3789 int err; 3790 3791 nr_sections = bitmap_weight(header->adds_features, HEADER_FEAT_BITS); 3792 if (!nr_sections) 3793 return 0; 3794 3795 feat_sec = sec = calloc(nr_sections, sizeof(*feat_sec)); 3796 if (!feat_sec) 3797 return -1; 3798 3799 sec_size = sizeof(*feat_sec) * nr_sections; 3800 3801 lseek(fd, header->feat_offset, SEEK_SET); 3802 3803 err = perf_header__getbuffer64(header, fd, feat_sec, sec_size); 3804 if (err < 0) 3805 goto out_free; 3806 3807 for_each_set_bit(feat, header->adds_features, HEADER_LAST_FEATURE) { 3808 err = process(sec++, header, feat, fd, data); 3809 if (err < 0) 3810 goto out_free; 3811 } 3812 err = 0; 3813 out_free: 3814 free(feat_sec); 3815 return err; 3816 } 3817 3818 static const int attr_file_abi_sizes[] = { 3819 [0] = PERF_ATTR_SIZE_VER0, 3820 [1] = PERF_ATTR_SIZE_VER1, 3821 [2] = PERF_ATTR_SIZE_VER2, 3822 [3] = PERF_ATTR_SIZE_VER3, 3823 [4] = PERF_ATTR_SIZE_VER4, 3824 0, 3825 }; 3826 3827 /* 3828 * In the legacy file format, the magic number is not used to encode endianness. 3829 * hdr_sz was used to encode endianness. But given that hdr_sz can vary based 3830 * on ABI revisions, we need to try all combinations for all endianness to 3831 * detect the endianness. 3832 */ 3833 static int try_all_file_abis(uint64_t hdr_sz, struct perf_header *ph) 3834 { 3835 uint64_t ref_size, attr_size; 3836 int i; 3837 3838 for (i = 0 ; attr_file_abi_sizes[i]; i++) { 3839 ref_size = attr_file_abi_sizes[i] 3840 + sizeof(struct perf_file_section); 3841 if (hdr_sz != ref_size) { 3842 attr_size = bswap_64(hdr_sz); 3843 if (attr_size != ref_size) 3844 continue; 3845 3846 ph->needs_swap = true; 3847 } 3848 pr_debug("ABI%d perf.data file detected, need_swap=%d\n", 3849 i, 3850 ph->needs_swap); 3851 return 0; 3852 } 3853 /* could not determine endianness */ 3854 return -1; 3855 } 3856 3857 #define PERF_PIPE_HDR_VER0 16 3858 3859 static const size_t attr_pipe_abi_sizes[] = { 3860 [0] = PERF_PIPE_HDR_VER0, 3861 0, 3862 }; 3863 3864 /* 3865 * In the legacy pipe format, there is an implicit assumption that endianness 3866 * between host recording the samples, and host parsing the samples is the 3867 * same. This is not always the case given that the pipe output may always be 3868 * redirected into a file and analyzed on a different machine with possibly a 3869 * different endianness and perf_event ABI revisions in the perf tool itself. 3870 */ 3871 static int try_all_pipe_abis(uint64_t hdr_sz, struct perf_header *ph) 3872 { 3873 u64 attr_size; 3874 int i; 3875 3876 for (i = 0 ; attr_pipe_abi_sizes[i]; i++) { 3877 if (hdr_sz != attr_pipe_abi_sizes[i]) { 3878 attr_size = bswap_64(hdr_sz); 3879 if (attr_size != hdr_sz) 3880 continue; 3881 3882 ph->needs_swap = true; 3883 } 3884 pr_debug("Pipe ABI%d perf.data file detected\n", i); 3885 return 0; 3886 } 3887 return -1; 3888 } 3889 3890 bool is_perf_magic(u64 magic) 3891 { 3892 if (!memcmp(&magic, __perf_magic1, sizeof(magic)) 3893 || magic == __perf_magic2 3894 || magic == __perf_magic2_sw) 3895 return true; 3896 3897 return false; 3898 } 3899 3900 static int check_magic_endian(u64 magic, uint64_t hdr_sz, 3901 bool is_pipe, struct perf_header *ph) 3902 { 3903 int ret; 3904 3905 /* check for legacy format */ 3906 ret = memcmp(&magic, __perf_magic1, sizeof(magic)); 3907 if (ret == 0) { 3908 ph->version = PERF_HEADER_VERSION_1; 3909 pr_debug("legacy perf.data format\n"); 3910 if (is_pipe) 3911 return try_all_pipe_abis(hdr_sz, ph); 3912 3913 return try_all_file_abis(hdr_sz, ph); 3914 } 3915 /* 3916 * the new magic number serves two purposes: 3917 * - unique number to identify actual perf.data files 3918 * - encode endianness of file 3919 */ 3920 ph->version = PERF_HEADER_VERSION_2; 3921 3922 /* check magic number with one endianness */ 3923 if (magic == __perf_magic2) 3924 return 0; 3925 3926 /* check magic number with opposite endianness */ 3927 if (magic != __perf_magic2_sw) 3928 return -1; 3929 3930 ph->needs_swap = true; 3931 3932 return 0; 3933 } 3934 3935 int perf_file_header__read(struct perf_file_header *header, 3936 struct perf_header *ph, int fd) 3937 { 3938 ssize_t ret; 3939 3940 lseek(fd, 0, SEEK_SET); 3941 3942 ret = readn(fd, header, sizeof(*header)); 3943 if (ret <= 0) 3944 return -1; 3945 3946 if (check_magic_endian(header->magic, 3947 header->attr_size, false, ph) < 0) { 3948 pr_debug("magic/endian check failed\n"); 3949 return -1; 3950 } 3951 3952 if (ph->needs_swap) { 3953 mem_bswap_64(header, offsetof(struct perf_file_header, 3954 adds_features)); 3955 } 3956 3957 if (header->size != sizeof(*header)) { 3958 /* Support the previous format */ 3959 if (header->size == offsetof(typeof(*header), adds_features)) 3960 bitmap_zero(header->adds_features, HEADER_FEAT_BITS); 3961 else 3962 return -1; 3963 } else if (ph->needs_swap) { 3964 /* 3965 * feature bitmap is declared as an array of unsigned longs -- 3966 * not good since its size can differ between the host that 3967 * generated the data file and the host analyzing the file. 3968 * 3969 * We need to handle endianness, but we don't know the size of 3970 * the unsigned long where the file was generated. Take a best 3971 * guess at determining it: try 64-bit swap first (ie., file 3972 * created on a 64-bit host), and check if the hostname feature 3973 * bit is set (this feature bit is forced on as of fbe96f2). 3974 * If the bit is not, undo the 64-bit swap and try a 32-bit 3975 * swap. If the hostname bit is still not set (e.g., older data 3976 * file), punt and fallback to the original behavior -- 3977 * clearing all feature bits and setting buildid. 3978 */ 3979 mem_bswap_64(&header->adds_features, 3980 BITS_TO_U64(HEADER_FEAT_BITS)); 3981 3982 if (!test_bit(HEADER_HOSTNAME, header->adds_features)) { 3983 /* unswap as u64 */ 3984 mem_bswap_64(&header->adds_features, 3985 BITS_TO_U64(HEADER_FEAT_BITS)); 3986 3987 /* unswap as u32 */ 3988 mem_bswap_32(&header->adds_features, 3989 BITS_TO_U32(HEADER_FEAT_BITS)); 3990 } 3991 3992 if (!test_bit(HEADER_HOSTNAME, header->adds_features)) { 3993 bitmap_zero(header->adds_features, HEADER_FEAT_BITS); 3994 __set_bit(HEADER_BUILD_ID, header->adds_features); 3995 } 3996 } 3997 3998 memcpy(&ph->adds_features, &header->adds_features, 3999 sizeof(ph->adds_features)); 4000 4001 ph->data_offset = header->data.offset; 4002 ph->data_size = header->data.size; 4003 ph->feat_offset = header->data.offset + header->data.size; 4004 return 0; 4005 } 4006 4007 static int perf_file_section__process(struct perf_file_section *section, 4008 struct perf_header *ph, 4009 int feat, int fd, void *data) 4010 { 4011 struct feat_fd fdd = { 4012 .fd = fd, 4013 .ph = ph, 4014 .size = section->size, 4015 .offset = section->offset, 4016 }; 4017 4018 if (lseek(fd, section->offset, SEEK_SET) == (off_t)-1) { 4019 pr_debug("Failed to lseek to %" PRIu64 " offset for feature " 4020 "%d, continuing...\n", section->offset, feat); 4021 return 0; 4022 } 4023 4024 if (feat >= HEADER_LAST_FEATURE) { 4025 pr_debug("unknown feature %d, continuing...\n", feat); 4026 return 0; 4027 } 4028 4029 if (!feat_ops[feat].process) 4030 return 0; 4031 4032 return feat_ops[feat].process(&fdd, data); 4033 } 4034 4035 static int perf_file_header__read_pipe(struct perf_pipe_file_header *header, 4036 struct perf_header *ph, 4037 struct perf_data* data, 4038 bool repipe, int repipe_fd) 4039 { 4040 struct feat_fd ff = { 4041 .fd = repipe_fd, 4042 .ph = ph, 4043 }; 4044 ssize_t ret; 4045 4046 ret = perf_data__read(data, header, sizeof(*header)); 4047 if (ret <= 0) 4048 return -1; 4049 4050 if (check_magic_endian(header->magic, header->size, true, ph) < 0) { 4051 pr_debug("endian/magic failed\n"); 4052 return -1; 4053 } 4054 4055 if (ph->needs_swap) 4056 header->size = bswap_64(header->size); 4057 4058 if (repipe && do_write(&ff, header, sizeof(*header)) < 0) 4059 return -1; 4060 4061 return 0; 4062 } 4063 4064 static int perf_header__read_pipe(struct perf_session *session, int repipe_fd) 4065 { 4066 struct perf_header *header = &session->header; 4067 struct perf_pipe_file_header f_header; 4068 4069 if (perf_file_header__read_pipe(&f_header, header, session->data, 4070 session->repipe, repipe_fd) < 0) { 4071 pr_debug("incompatible file format\n"); 4072 return -EINVAL; 4073 } 4074 4075 return f_header.size == sizeof(f_header) ? 0 : -1; 4076 } 4077 4078 static int read_attr(int fd, struct perf_header *ph, 4079 struct perf_file_attr *f_attr) 4080 { 4081 struct perf_event_attr *attr = &f_attr->attr; 4082 size_t sz, left; 4083 size_t our_sz = sizeof(f_attr->attr); 4084 ssize_t ret; 4085 4086 memset(f_attr, 0, sizeof(*f_attr)); 4087 4088 /* read minimal guaranteed structure */ 4089 ret = readn(fd, attr, PERF_ATTR_SIZE_VER0); 4090 if (ret <= 0) { 4091 pr_debug("cannot read %d bytes of header attr\n", 4092 PERF_ATTR_SIZE_VER0); 4093 return -1; 4094 } 4095 4096 /* on file perf_event_attr size */ 4097 sz = attr->size; 4098 4099 if (ph->needs_swap) 4100 sz = bswap_32(sz); 4101 4102 if (sz == 0) { 4103 /* assume ABI0 */ 4104 sz = PERF_ATTR_SIZE_VER0; 4105 } else if (sz > our_sz) { 4106 pr_debug("file uses a more recent and unsupported ABI" 4107 " (%zu bytes extra)\n", sz - our_sz); 4108 return -1; 4109 } 4110 /* what we have not yet read and that we know about */ 4111 left = sz - PERF_ATTR_SIZE_VER0; 4112 if (left) { 4113 void *ptr = attr; 4114 ptr += PERF_ATTR_SIZE_VER0; 4115 4116 ret = readn(fd, ptr, left); 4117 } 4118 /* read perf_file_section, ids are read in caller */ 4119 ret = readn(fd, &f_attr->ids, sizeof(f_attr->ids)); 4120 4121 return ret <= 0 ? -1 : 0; 4122 } 4123 4124 #ifdef HAVE_LIBTRACEEVENT 4125 static int evsel__prepare_tracepoint_event(struct evsel *evsel, struct tep_handle *pevent) 4126 { 4127 struct tep_event *event; 4128 char bf[128]; 4129 4130 /* already prepared */ 4131 if (evsel->tp_format) 4132 return 0; 4133 4134 if (pevent == NULL) { 4135 pr_debug("broken or missing trace data\n"); 4136 return -1; 4137 } 4138 4139 event = tep_find_event(pevent, evsel->core.attr.config); 4140 if (event == NULL) { 4141 pr_debug("cannot find event format for %d\n", (int)evsel->core.attr.config); 4142 return -1; 4143 } 4144 4145 if (!evsel->name) { 4146 snprintf(bf, sizeof(bf), "%s:%s", event->system, event->name); 4147 evsel->name = strdup(bf); 4148 if (evsel->name == NULL) 4149 return -1; 4150 } 4151 4152 evsel->tp_format = event; 4153 return 0; 4154 } 4155 4156 static int evlist__prepare_tracepoint_events(struct evlist *evlist, struct tep_handle *pevent) 4157 { 4158 struct evsel *pos; 4159 4160 evlist__for_each_entry(evlist, pos) { 4161 if (pos->core.attr.type == PERF_TYPE_TRACEPOINT && 4162 evsel__prepare_tracepoint_event(pos, pevent)) 4163 return -1; 4164 } 4165 4166 return 0; 4167 } 4168 #endif 4169 4170 int perf_session__read_header(struct perf_session *session, int repipe_fd) 4171 { 4172 struct perf_data *data = session->data; 4173 struct perf_header *header = &session->header; 4174 struct perf_file_header f_header; 4175 struct perf_file_attr f_attr; 4176 u64 f_id; 4177 int nr_attrs, nr_ids, i, j, err; 4178 int fd = perf_data__fd(data); 4179 4180 session->evlist = evlist__new(); 4181 if (session->evlist == NULL) 4182 return -ENOMEM; 4183 4184 session->evlist->env = &header->env; 4185 session->machines.host.env = &header->env; 4186 4187 /* 4188 * We can read 'pipe' data event from regular file, 4189 * check for the pipe header regardless of source. 4190 */ 4191 err = perf_header__read_pipe(session, repipe_fd); 4192 if (!err || perf_data__is_pipe(data)) { 4193 data->is_pipe = true; 4194 return err; 4195 } 4196 4197 if (perf_file_header__read(&f_header, header, fd) < 0) 4198 return -EINVAL; 4199 4200 if (header->needs_swap && data->in_place_update) { 4201 pr_err("In-place update not supported when byte-swapping is required\n"); 4202 return -EINVAL; 4203 } 4204 4205 /* 4206 * Sanity check that perf.data was written cleanly; data size is 4207 * initialized to 0 and updated only if the on_exit function is run. 4208 * If data size is still 0 then the file contains only partial 4209 * information. Just warn user and process it as much as it can. 4210 */ 4211 if (f_header.data.size == 0) { 4212 pr_warning("WARNING: The %s file's data size field is 0 which is unexpected.\n" 4213 "Was the 'perf record' command properly terminated?\n", 4214 data->file.path); 4215 } 4216 4217 if (f_header.attr_size == 0) { 4218 pr_err("ERROR: The %s file's attr size field is 0 which is unexpected.\n" 4219 "Was the 'perf record' command properly terminated?\n", 4220 data->file.path); 4221 return -EINVAL; 4222 } 4223 4224 nr_attrs = f_header.attrs.size / f_header.attr_size; 4225 lseek(fd, f_header.attrs.offset, SEEK_SET); 4226 4227 for (i = 0; i < nr_attrs; i++) { 4228 struct evsel *evsel; 4229 off_t tmp; 4230 4231 if (read_attr(fd, header, &f_attr) < 0) 4232 goto out_errno; 4233 4234 if (header->needs_swap) { 4235 f_attr.ids.size = bswap_64(f_attr.ids.size); 4236 f_attr.ids.offset = bswap_64(f_attr.ids.offset); 4237 perf_event__attr_swap(&f_attr.attr); 4238 } 4239 4240 tmp = lseek(fd, 0, SEEK_CUR); 4241 evsel = evsel__new(&f_attr.attr); 4242 4243 if (evsel == NULL) 4244 goto out_delete_evlist; 4245 4246 evsel->needs_swap = header->needs_swap; 4247 /* 4248 * Do it before so that if perf_evsel__alloc_id fails, this 4249 * entry gets purged too at evlist__delete(). 4250 */ 4251 evlist__add(session->evlist, evsel); 4252 4253 nr_ids = f_attr.ids.size / sizeof(u64); 4254 /* 4255 * We don't have the cpu and thread maps on the header, so 4256 * for allocating the perf_sample_id table we fake 1 cpu and 4257 * hattr->ids threads. 4258 */ 4259 if (perf_evsel__alloc_id(&evsel->core, 1, nr_ids)) 4260 goto out_delete_evlist; 4261 4262 lseek(fd, f_attr.ids.offset, SEEK_SET); 4263 4264 for (j = 0; j < nr_ids; j++) { 4265 if (perf_header__getbuffer64(header, fd, &f_id, sizeof(f_id))) 4266 goto out_errno; 4267 4268 perf_evlist__id_add(&session->evlist->core, &evsel->core, 0, j, f_id); 4269 } 4270 4271 lseek(fd, tmp, SEEK_SET); 4272 } 4273 4274 #ifdef HAVE_LIBTRACEEVENT 4275 perf_header__process_sections(header, fd, &session->tevent, 4276 perf_file_section__process); 4277 4278 if (evlist__prepare_tracepoint_events(session->evlist, session->tevent.pevent)) 4279 goto out_delete_evlist; 4280 #else 4281 perf_header__process_sections(header, fd, NULL, perf_file_section__process); 4282 #endif 4283 4284 return 0; 4285 out_errno: 4286 return -errno; 4287 4288 out_delete_evlist: 4289 evlist__delete(session->evlist); 4290 session->evlist = NULL; 4291 return -ENOMEM; 4292 } 4293 4294 int perf_event__process_feature(struct perf_session *session, 4295 union perf_event *event) 4296 { 4297 struct perf_tool *tool = session->tool; 4298 struct feat_fd ff = { .fd = 0 }; 4299 struct perf_record_header_feature *fe = (struct perf_record_header_feature *)event; 4300 int type = fe->header.type; 4301 u64 feat = fe->feat_id; 4302 int ret = 0; 4303 4304 if (type < 0 || type >= PERF_RECORD_HEADER_MAX) { 4305 pr_warning("invalid record type %d in pipe-mode\n", type); 4306 return 0; 4307 } 4308 if (feat == HEADER_RESERVED || feat >= HEADER_LAST_FEATURE) { 4309 pr_warning("invalid record type %d in pipe-mode\n", type); 4310 return -1; 4311 } 4312 4313 if (!feat_ops[feat].process) 4314 return 0; 4315 4316 ff.buf = (void *)fe->data; 4317 ff.size = event->header.size - sizeof(*fe); 4318 ff.ph = &session->header; 4319 4320 if (feat_ops[feat].process(&ff, NULL)) { 4321 ret = -1; 4322 goto out; 4323 } 4324 4325 if (!feat_ops[feat].print || !tool->show_feat_hdr) 4326 goto out; 4327 4328 if (!feat_ops[feat].full_only || 4329 tool->show_feat_hdr >= SHOW_FEAT_HEADER_FULL_INFO) { 4330 feat_ops[feat].print(&ff, stdout); 4331 } else { 4332 fprintf(stdout, "# %s info available, use -I to display\n", 4333 feat_ops[feat].name); 4334 } 4335 out: 4336 free_event_desc(ff.events); 4337 return ret; 4338 } 4339 4340 size_t perf_event__fprintf_event_update(union perf_event *event, FILE *fp) 4341 { 4342 struct perf_record_event_update *ev = &event->event_update; 4343 struct perf_cpu_map *map; 4344 size_t ret; 4345 4346 ret = fprintf(fp, "\n... id: %" PRI_lu64 "\n", ev->id); 4347 4348 switch (ev->type) { 4349 case PERF_EVENT_UPDATE__SCALE: 4350 ret += fprintf(fp, "... scale: %f\n", ev->scale.scale); 4351 break; 4352 case PERF_EVENT_UPDATE__UNIT: 4353 ret += fprintf(fp, "... unit: %s\n", ev->unit); 4354 break; 4355 case PERF_EVENT_UPDATE__NAME: 4356 ret += fprintf(fp, "... name: %s\n", ev->name); 4357 break; 4358 case PERF_EVENT_UPDATE__CPUS: 4359 ret += fprintf(fp, "... "); 4360 4361 map = cpu_map__new_data(&ev->cpus.cpus); 4362 if (map) 4363 ret += cpu_map__fprintf(map, fp); 4364 else 4365 ret += fprintf(fp, "failed to get cpus\n"); 4366 break; 4367 default: 4368 ret += fprintf(fp, "... unknown type\n"); 4369 break; 4370 } 4371 4372 return ret; 4373 } 4374 4375 int perf_event__process_attr(struct perf_tool *tool __maybe_unused, 4376 union perf_event *event, 4377 struct evlist **pevlist) 4378 { 4379 u32 i, ids, n_ids; 4380 struct evsel *evsel; 4381 struct evlist *evlist = *pevlist; 4382 4383 if (evlist == NULL) { 4384 *pevlist = evlist = evlist__new(); 4385 if (evlist == NULL) 4386 return -ENOMEM; 4387 } 4388 4389 evsel = evsel__new(&event->attr.attr); 4390 if (evsel == NULL) 4391 return -ENOMEM; 4392 4393 evlist__add(evlist, evsel); 4394 4395 ids = event->header.size; 4396 ids -= (void *)&event->attr.id - (void *)event; 4397 n_ids = ids / sizeof(u64); 4398 /* 4399 * We don't have the cpu and thread maps on the header, so 4400 * for allocating the perf_sample_id table we fake 1 cpu and 4401 * hattr->ids threads. 4402 */ 4403 if (perf_evsel__alloc_id(&evsel->core, 1, n_ids)) 4404 return -ENOMEM; 4405 4406 for (i = 0; i < n_ids; i++) { 4407 perf_evlist__id_add(&evlist->core, &evsel->core, 0, i, event->attr.id[i]); 4408 } 4409 4410 return 0; 4411 } 4412 4413 int perf_event__process_event_update(struct perf_tool *tool __maybe_unused, 4414 union perf_event *event, 4415 struct evlist **pevlist) 4416 { 4417 struct perf_record_event_update *ev = &event->event_update; 4418 struct evlist *evlist; 4419 struct evsel *evsel; 4420 struct perf_cpu_map *map; 4421 4422 if (dump_trace) 4423 perf_event__fprintf_event_update(event, stdout); 4424 4425 if (!pevlist || *pevlist == NULL) 4426 return -EINVAL; 4427 4428 evlist = *pevlist; 4429 4430 evsel = evlist__id2evsel(evlist, ev->id); 4431 if (evsel == NULL) 4432 return -EINVAL; 4433 4434 switch (ev->type) { 4435 case PERF_EVENT_UPDATE__UNIT: 4436 free((char *)evsel->unit); 4437 evsel->unit = strdup(ev->unit); 4438 break; 4439 case PERF_EVENT_UPDATE__NAME: 4440 free(evsel->name); 4441 evsel->name = strdup(ev->name); 4442 break; 4443 case PERF_EVENT_UPDATE__SCALE: 4444 evsel->scale = ev->scale.scale; 4445 break; 4446 case PERF_EVENT_UPDATE__CPUS: 4447 map = cpu_map__new_data(&ev->cpus.cpus); 4448 if (map) { 4449 perf_cpu_map__put(evsel->core.own_cpus); 4450 evsel->core.own_cpus = map; 4451 } else 4452 pr_err("failed to get event_update cpus\n"); 4453 default: 4454 break; 4455 } 4456 4457 return 0; 4458 } 4459 4460 #ifdef HAVE_LIBTRACEEVENT 4461 int perf_event__process_tracing_data(struct perf_session *session, 4462 union perf_event *event) 4463 { 4464 ssize_t size_read, padding, size = event->tracing_data.size; 4465 int fd = perf_data__fd(session->data); 4466 char buf[BUFSIZ]; 4467 4468 /* 4469 * The pipe fd is already in proper place and in any case 4470 * we can't move it, and we'd screw the case where we read 4471 * 'pipe' data from regular file. The trace_report reads 4472 * data from 'fd' so we need to set it directly behind the 4473 * event, where the tracing data starts. 4474 */ 4475 if (!perf_data__is_pipe(session->data)) { 4476 off_t offset = lseek(fd, 0, SEEK_CUR); 4477 4478 /* setup for reading amidst mmap */ 4479 lseek(fd, offset + sizeof(struct perf_record_header_tracing_data), 4480 SEEK_SET); 4481 } 4482 4483 size_read = trace_report(fd, &session->tevent, 4484 session->repipe); 4485 padding = PERF_ALIGN(size_read, sizeof(u64)) - size_read; 4486 4487 if (readn(fd, buf, padding) < 0) { 4488 pr_err("%s: reading input file", __func__); 4489 return -1; 4490 } 4491 if (session->repipe) { 4492 int retw = write(STDOUT_FILENO, buf, padding); 4493 if (retw <= 0 || retw != padding) { 4494 pr_err("%s: repiping tracing data padding", __func__); 4495 return -1; 4496 } 4497 } 4498 4499 if (size_read + padding != size) { 4500 pr_err("%s: tracing data size mismatch", __func__); 4501 return -1; 4502 } 4503 4504 evlist__prepare_tracepoint_events(session->evlist, session->tevent.pevent); 4505 4506 return size_read + padding; 4507 } 4508 #endif 4509 4510 int perf_event__process_build_id(struct perf_session *session, 4511 union perf_event *event) 4512 { 4513 __event_process_build_id(&event->build_id, 4514 event->build_id.filename, 4515 session); 4516 return 0; 4517 } 4518