1// SPDX-License-Identifier: GPL-2.0-only 2/* 3 * Copyright (C) 2011, Red Hat Inc, Arnaldo Carvalho de Melo <acme@redhat.com> 4 * 5 * Parts came from builtin-{top,stat,record}.c, see those files for further 6 * copyright notes. 7 */ 8 9#include <byteswap.h> 10#include <errno.h> 11#include <inttypes.h> 12#include <linux/bitops.h> 13#include <api/fs/fs.h> 14#include <api/fs/tracing_path.h> 15#include <linux/hw_breakpoint.h> 16#include <linux/perf_event.h> 17#include <linux/compiler.h> 18#include <linux/err.h> 19#include <linux/zalloc.h> 20#include <sys/ioctl.h> 21#include <sys/resource.h> 22#include <sys/types.h> 23#include <dirent.h> 24#include <stdlib.h> 25#include <perf/evsel.h> 26#include "asm/bug.h" 27#include "bpf_counter.h" 28#include "callchain.h" 29#include "cgroup.h" 30#include "counts.h" 31#include "event.h" 32#include "evsel.h" 33#include "util/env.h" 34#include "util/evsel_config.h" 35#include "util/evsel_fprintf.h" 36#include "evlist.h" 37#include <perf/cpumap.h> 38#include "thread_map.h" 39#include "target.h" 40#include "perf_regs.h" 41#include "record.h" 42#include "debug.h" 43#include "trace-event.h" 44#include "stat.h" 45#include "string2.h" 46#include "memswap.h" 47#include "util.h" 48#include "util/hashmap.h" 49#include "off_cpu.h" 50#include "pmu.h" 51#include "pmus.h" 52#include "../perf-sys.h" 53#include "util/parse-branch-options.h" 54#include "util/bpf-filter.h" 55#include <internal/xyarray.h> 56#include <internal/lib.h> 57#include <internal/threadmap.h> 58 59#include <linux/ctype.h> 60 61#ifdef HAVE_LIBTRACEEVENT 62#include <traceevent/event-parse.h> 63#endif 64 65struct perf_missing_features perf_missing_features; 66 67static clockid_t clockid; 68 69static const char *const perf_tool_event__tool_names[PERF_TOOL_MAX] = { 70 NULL, 71 "duration_time", 72 "user_time", 73 "system_time", 74}; 75 76const char *perf_tool_event__to_str(enum perf_tool_event ev) 77{ 78 if (ev > PERF_TOOL_NONE && ev < PERF_TOOL_MAX) 79 return perf_tool_event__tool_names[ev]; 80 81 return NULL; 82} 83 84enum perf_tool_event perf_tool_event__from_str(const char *str) 85{ 86 int i; 87 88 perf_tool_event__for_each_event(i) { 89 if (!strcmp(str, perf_tool_event__tool_names[i])) 90 return i; 91 } 92 return PERF_TOOL_NONE; 93} 94 95 96static int evsel__no_extra_init(struct evsel *evsel __maybe_unused) 97{ 98 return 0; 99} 100 101void __weak test_attr__ready(void) { } 102 103static void evsel__no_extra_fini(struct evsel *evsel __maybe_unused) 104{ 105} 106 107static struct { 108 size_t size; 109 int (*init)(struct evsel *evsel); 110 void (*fini)(struct evsel *evsel); 111} perf_evsel__object = { 112 .size = sizeof(struct evsel), 113 .init = evsel__no_extra_init, 114 .fini = evsel__no_extra_fini, 115}; 116 117int evsel__object_config(size_t object_size, int (*init)(struct evsel *evsel), 118 void (*fini)(struct evsel *evsel)) 119{ 120 121 if (object_size == 0) 122 goto set_methods; 123 124 if (perf_evsel__object.size > object_size) 125 return -EINVAL; 126 127 perf_evsel__object.size = object_size; 128 129set_methods: 130 if (init != NULL) 131 perf_evsel__object.init = init; 132 133 if (fini != NULL) 134 perf_evsel__object.fini = fini; 135 136 return 0; 137} 138 139#define FD(e, x, y) (*(int *)xyarray__entry(e->core.fd, x, y)) 140 141int __evsel__sample_size(u64 sample_type) 142{ 143 u64 mask = sample_type & PERF_SAMPLE_MASK; 144 int size = 0; 145 int i; 146 147 for (i = 0; i < 64; i++) { 148 if (mask & (1ULL << i)) 149 size++; 150 } 151 152 size *= sizeof(u64); 153 154 return size; 155} 156 157/** 158 * __perf_evsel__calc_id_pos - calculate id_pos. 159 * @sample_type: sample type 160 * 161 * This function returns the position of the event id (PERF_SAMPLE_ID or 162 * PERF_SAMPLE_IDENTIFIER) in a sample event i.e. in the array of struct 163 * perf_record_sample. 164 */ 165static int __perf_evsel__calc_id_pos(u64 sample_type) 166{ 167 int idx = 0; 168 169 if (sample_type & PERF_SAMPLE_IDENTIFIER) 170 return 0; 171 172 if (!(sample_type & PERF_SAMPLE_ID)) 173 return -1; 174 175 if (sample_type & PERF_SAMPLE_IP) 176 idx += 1; 177 178 if (sample_type & PERF_SAMPLE_TID) 179 idx += 1; 180 181 if (sample_type & PERF_SAMPLE_TIME) 182 idx += 1; 183 184 if (sample_type & PERF_SAMPLE_ADDR) 185 idx += 1; 186 187 return idx; 188} 189 190/** 191 * __perf_evsel__calc_is_pos - calculate is_pos. 192 * @sample_type: sample type 193 * 194 * This function returns the position (counting backwards) of the event id 195 * (PERF_SAMPLE_ID or PERF_SAMPLE_IDENTIFIER) in a non-sample event i.e. if 196 * sample_id_all is used there is an id sample appended to non-sample events. 197 */ 198static int __perf_evsel__calc_is_pos(u64 sample_type) 199{ 200 int idx = 1; 201 202 if (sample_type & PERF_SAMPLE_IDENTIFIER) 203 return 1; 204 205 if (!(sample_type & PERF_SAMPLE_ID)) 206 return -1; 207 208 if (sample_type & PERF_SAMPLE_CPU) 209 idx += 1; 210 211 if (sample_type & PERF_SAMPLE_STREAM_ID) 212 idx += 1; 213 214 return idx; 215} 216 217void evsel__calc_id_pos(struct evsel *evsel) 218{ 219 evsel->id_pos = __perf_evsel__calc_id_pos(evsel->core.attr.sample_type); 220 evsel->is_pos = __perf_evsel__calc_is_pos(evsel->core.attr.sample_type); 221} 222 223void __evsel__set_sample_bit(struct evsel *evsel, 224 enum perf_event_sample_format bit) 225{ 226 if (!(evsel->core.attr.sample_type & bit)) { 227 evsel->core.attr.sample_type |= bit; 228 evsel->sample_size += sizeof(u64); 229 evsel__calc_id_pos(evsel); 230 } 231} 232 233void __evsel__reset_sample_bit(struct evsel *evsel, 234 enum perf_event_sample_format bit) 235{ 236 if (evsel->core.attr.sample_type & bit) { 237 evsel->core.attr.sample_type &= ~bit; 238 evsel->sample_size -= sizeof(u64); 239 evsel__calc_id_pos(evsel); 240 } 241} 242 243void evsel__set_sample_id(struct evsel *evsel, 244 bool can_sample_identifier) 245{ 246 if (can_sample_identifier) { 247 evsel__reset_sample_bit(evsel, ID); 248 evsel__set_sample_bit(evsel, IDENTIFIER); 249 } else { 250 evsel__set_sample_bit(evsel, ID); 251 } 252 evsel->core.attr.read_format |= PERF_FORMAT_ID; 253} 254 255/** 256 * evsel__is_function_event - Return whether given evsel is a function 257 * trace event 258 * 259 * @evsel - evsel selector to be tested 260 * 261 * Return %true if event is function trace event 262 */ 263bool evsel__is_function_event(struct evsel *evsel) 264{ 265#define FUNCTION_EVENT "ftrace:function" 266 267 return evsel->name && 268 !strncmp(FUNCTION_EVENT, evsel->name, sizeof(FUNCTION_EVENT)); 269 270#undef FUNCTION_EVENT 271} 272 273void evsel__init(struct evsel *evsel, 274 struct perf_event_attr *attr, int idx) 275{ 276 perf_evsel__init(&evsel->core, attr, idx); 277 evsel->tracking = !idx; 278 evsel->unit = strdup(""); 279 evsel->scale = 1.0; 280 evsel->max_events = ULONG_MAX; 281 evsel->evlist = NULL; 282 evsel->bpf_obj = NULL; 283 evsel->bpf_fd = -1; 284 INIT_LIST_HEAD(&evsel->config_terms); 285 INIT_LIST_HEAD(&evsel->bpf_counter_list); 286 INIT_LIST_HEAD(&evsel->bpf_filters); 287 perf_evsel__object.init(evsel); 288 evsel->sample_size = __evsel__sample_size(attr->sample_type); 289 evsel__calc_id_pos(evsel); 290 evsel->cmdline_group_boundary = false; 291 evsel->metric_events = NULL; 292 evsel->per_pkg_mask = NULL; 293 evsel->collect_stat = false; 294 evsel->pmu_name = NULL; 295 evsel->group_pmu_name = NULL; 296 evsel->skippable = false; 297} 298 299struct evsel *evsel__new_idx(struct perf_event_attr *attr, int idx) 300{ 301 struct evsel *evsel = zalloc(perf_evsel__object.size); 302 303 if (!evsel) 304 return NULL; 305 evsel__init(evsel, attr, idx); 306 307 if (evsel__is_bpf_output(evsel) && !attr->sample_type) { 308 evsel->core.attr.sample_type = (PERF_SAMPLE_RAW | PERF_SAMPLE_TIME | 309 PERF_SAMPLE_CPU | PERF_SAMPLE_PERIOD), 310 evsel->core.attr.sample_period = 1; 311 } 312 313 if (evsel__is_clock(evsel)) { 314 free((char *)evsel->unit); 315 evsel->unit = strdup("msec"); 316 evsel->scale = 1e-6; 317 } 318 319 return evsel; 320} 321 322int copy_config_terms(struct list_head *dst, struct list_head *src) 323{ 324 struct evsel_config_term *pos, *tmp; 325 326 list_for_each_entry(pos, src, list) { 327 tmp = malloc(sizeof(*tmp)); 328 if (tmp == NULL) 329 return -ENOMEM; 330 331 *tmp = *pos; 332 if (tmp->free_str) { 333 tmp->val.str = strdup(pos->val.str); 334 if (tmp->val.str == NULL) { 335 free(tmp); 336 return -ENOMEM; 337 } 338 } 339 list_add_tail(&tmp->list, dst); 340 } 341 return 0; 342} 343 344static int evsel__copy_config_terms(struct evsel *dst, struct evsel *src) 345{ 346 return copy_config_terms(&dst->config_terms, &src->config_terms); 347} 348 349/** 350 * evsel__clone - create a new evsel copied from @orig 351 * @orig: original evsel 352 * 353 * The assumption is that @orig is not configured nor opened yet. 354 * So we only care about the attributes that can be set while it's parsed. 355 */ 356struct evsel *evsel__clone(struct evsel *orig) 357{ 358 struct evsel *evsel; 359 360 BUG_ON(orig->core.fd); 361 BUG_ON(orig->counts); 362 BUG_ON(orig->priv); 363 BUG_ON(orig->per_pkg_mask); 364 365 /* cannot handle BPF objects for now */ 366 if (orig->bpf_obj) 367 return NULL; 368 369 evsel = evsel__new(&orig->core.attr); 370 if (evsel == NULL) 371 return NULL; 372 373 evsel->core.cpus = perf_cpu_map__get(orig->core.cpus); 374 evsel->core.own_cpus = perf_cpu_map__get(orig->core.own_cpus); 375 evsel->core.threads = perf_thread_map__get(orig->core.threads); 376 evsel->core.nr_members = orig->core.nr_members; 377 evsel->core.system_wide = orig->core.system_wide; 378 evsel->core.requires_cpu = orig->core.requires_cpu; 379 evsel->core.is_pmu_core = orig->core.is_pmu_core; 380 381 if (orig->name) { 382 evsel->name = strdup(orig->name); 383 if (evsel->name == NULL) 384 goto out_err; 385 } 386 if (orig->group_name) { 387 evsel->group_name = strdup(orig->group_name); 388 if (evsel->group_name == NULL) 389 goto out_err; 390 } 391 if (orig->pmu_name) { 392 evsel->pmu_name = strdup(orig->pmu_name); 393 if (evsel->pmu_name == NULL) 394 goto out_err; 395 } 396 if (orig->group_pmu_name) { 397 evsel->group_pmu_name = strdup(orig->group_pmu_name); 398 if (evsel->group_pmu_name == NULL) 399 goto out_err; 400 } 401 if (orig->filter) { 402 evsel->filter = strdup(orig->filter); 403 if (evsel->filter == NULL) 404 goto out_err; 405 } 406 if (orig->metric_id) { 407 evsel->metric_id = strdup(orig->metric_id); 408 if (evsel->metric_id == NULL) 409 goto out_err; 410 } 411 evsel->cgrp = cgroup__get(orig->cgrp); 412#ifdef HAVE_LIBTRACEEVENT 413 evsel->tp_format = orig->tp_format; 414#endif 415 evsel->handler = orig->handler; 416 evsel->core.leader = orig->core.leader; 417 418 evsel->max_events = orig->max_events; 419 evsel->tool_event = orig->tool_event; 420 free((char *)evsel->unit); 421 evsel->unit = strdup(orig->unit); 422 if (evsel->unit == NULL) 423 goto out_err; 424 425 evsel->scale = orig->scale; 426 evsel->snapshot = orig->snapshot; 427 evsel->per_pkg = orig->per_pkg; 428 evsel->percore = orig->percore; 429 evsel->precise_max = orig->precise_max; 430 evsel->is_libpfm_event = orig->is_libpfm_event; 431 432 evsel->exclude_GH = orig->exclude_GH; 433 evsel->sample_read = orig->sample_read; 434 evsel->auto_merge_stats = orig->auto_merge_stats; 435 evsel->collect_stat = orig->collect_stat; 436 evsel->weak_group = orig->weak_group; 437 evsel->use_config_name = orig->use_config_name; 438 evsel->pmu = orig->pmu; 439 440 if (evsel__copy_config_terms(evsel, orig) < 0) 441 goto out_err; 442 443 return evsel; 444 445out_err: 446 evsel__delete(evsel); 447 return NULL; 448} 449 450/* 451 * Returns pointer with encoded error via <linux/err.h> interface. 452 */ 453#ifdef HAVE_LIBTRACEEVENT 454struct evsel *evsel__newtp_idx(const char *sys, const char *name, int idx) 455{ 456 struct evsel *evsel = zalloc(perf_evsel__object.size); 457 int err = -ENOMEM; 458 459 if (evsel == NULL) { 460 goto out_err; 461 } else { 462 struct perf_event_attr attr = { 463 .type = PERF_TYPE_TRACEPOINT, 464 .sample_type = (PERF_SAMPLE_RAW | PERF_SAMPLE_TIME | 465 PERF_SAMPLE_CPU | PERF_SAMPLE_PERIOD), 466 }; 467 468 if (asprintf(&evsel->name, "%s:%s", sys, name) < 0) 469 goto out_free; 470 471 evsel->tp_format = trace_event__tp_format(sys, name); 472 if (IS_ERR(evsel->tp_format)) { 473 err = PTR_ERR(evsel->tp_format); 474 goto out_free; 475 } 476 477 event_attr_init(&attr); 478 attr.config = evsel->tp_format->id; 479 attr.sample_period = 1; 480 evsel__init(evsel, &attr, idx); 481 } 482 483 return evsel; 484 485out_free: 486 zfree(&evsel->name); 487 free(evsel); 488out_err: 489 return ERR_PTR(err); 490} 491#endif 492 493const char *const evsel__hw_names[PERF_COUNT_HW_MAX] = { 494 "cycles", 495 "instructions", 496 "cache-references", 497 "cache-misses", 498 "branches", 499 "branch-misses", 500 "bus-cycles", 501 "stalled-cycles-frontend", 502 "stalled-cycles-backend", 503 "ref-cycles", 504}; 505 506char *evsel__bpf_counter_events; 507 508bool evsel__match_bpf_counter_events(const char *name) 509{ 510 int name_len; 511 bool match; 512 char *ptr; 513 514 if (!evsel__bpf_counter_events) 515 return false; 516 517 ptr = strstr(evsel__bpf_counter_events, name); 518 name_len = strlen(name); 519 520 /* check name matches a full token in evsel__bpf_counter_events */ 521 match = (ptr != NULL) && 522 ((ptr == evsel__bpf_counter_events) || (*(ptr - 1) == ',')) && 523 ((*(ptr + name_len) == ',') || (*(ptr + name_len) == '\0')); 524 525 return match; 526} 527 528static const char *__evsel__hw_name(u64 config) 529{ 530 if (config < PERF_COUNT_HW_MAX && evsel__hw_names[config]) 531 return evsel__hw_names[config]; 532 533 return "unknown-hardware"; 534} 535 536static int evsel__add_modifiers(struct evsel *evsel, char *bf, size_t size) 537{ 538 int colon = 0, r = 0; 539 struct perf_event_attr *attr = &evsel->core.attr; 540 bool exclude_guest_default = false; 541 542#define MOD_PRINT(context, mod) do { \ 543 if (!attr->exclude_##context) { \ 544 if (!colon) colon = ++r; \ 545 r += scnprintf(bf + r, size - r, "%c", mod); \ 546 } } while(0) 547 548 if (attr->exclude_kernel || attr->exclude_user || attr->exclude_hv) { 549 MOD_PRINT(kernel, 'k'); 550 MOD_PRINT(user, 'u'); 551 MOD_PRINT(hv, 'h'); 552 exclude_guest_default = true; 553 } 554 555 if (attr->precise_ip) { 556 if (!colon) 557 colon = ++r; 558 r += scnprintf(bf + r, size - r, "%.*s", attr->precise_ip, "ppp"); 559 exclude_guest_default = true; 560 } 561 562 if (attr->exclude_host || attr->exclude_guest == exclude_guest_default) { 563 MOD_PRINT(host, 'H'); 564 MOD_PRINT(guest, 'G'); 565 } 566#undef MOD_PRINT 567 if (colon) 568 bf[colon - 1] = ':'; 569 return r; 570} 571 572int __weak arch_evsel__hw_name(struct evsel *evsel, char *bf, size_t size) 573{ 574 return scnprintf(bf, size, "%s", __evsel__hw_name(evsel->core.attr.config)); 575} 576 577static int evsel__hw_name(struct evsel *evsel, char *bf, size_t size) 578{ 579 int r = arch_evsel__hw_name(evsel, bf, size); 580 return r + evsel__add_modifiers(evsel, bf + r, size - r); 581} 582 583const char *const evsel__sw_names[PERF_COUNT_SW_MAX] = { 584 "cpu-clock", 585 "task-clock", 586 "page-faults", 587 "context-switches", 588 "cpu-migrations", 589 "minor-faults", 590 "major-faults", 591 "alignment-faults", 592 "emulation-faults", 593 "dummy", 594}; 595 596static const char *__evsel__sw_name(u64 config) 597{ 598 if (config < PERF_COUNT_SW_MAX && evsel__sw_names[config]) 599 return evsel__sw_names[config]; 600 return "unknown-software"; 601} 602 603static int evsel__sw_name(struct evsel *evsel, char *bf, size_t size) 604{ 605 int r = scnprintf(bf, size, "%s", __evsel__sw_name(evsel->core.attr.config)); 606 return r + evsel__add_modifiers(evsel, bf + r, size - r); 607} 608 609static int evsel__tool_name(enum perf_tool_event ev, char *bf, size_t size) 610{ 611 return scnprintf(bf, size, "%s", perf_tool_event__to_str(ev)); 612} 613 614static int __evsel__bp_name(char *bf, size_t size, u64 addr, u64 type) 615{ 616 int r; 617 618 r = scnprintf(bf, size, "mem:0x%" PRIx64 ":", addr); 619 620 if (type & HW_BREAKPOINT_R) 621 r += scnprintf(bf + r, size - r, "r"); 622 623 if (type & HW_BREAKPOINT_W) 624 r += scnprintf(bf + r, size - r, "w"); 625 626 if (type & HW_BREAKPOINT_X) 627 r += scnprintf(bf + r, size - r, "x"); 628 629 return r; 630} 631 632static int evsel__bp_name(struct evsel *evsel, char *bf, size_t size) 633{ 634 struct perf_event_attr *attr = &evsel->core.attr; 635 int r = __evsel__bp_name(bf, size, attr->bp_addr, attr->bp_type); 636 return r + evsel__add_modifiers(evsel, bf + r, size - r); 637} 638 639const char *const evsel__hw_cache[PERF_COUNT_HW_CACHE_MAX][EVSEL__MAX_ALIASES] = { 640 { "L1-dcache", "l1-d", "l1d", "L1-data", }, 641 { "L1-icache", "l1-i", "l1i", "L1-instruction", }, 642 { "LLC", "L2", }, 643 { "dTLB", "d-tlb", "Data-TLB", }, 644 { "iTLB", "i-tlb", "Instruction-TLB", }, 645 { "branch", "branches", "bpu", "btb", "bpc", }, 646 { "node", }, 647}; 648 649const char *const evsel__hw_cache_op[PERF_COUNT_HW_CACHE_OP_MAX][EVSEL__MAX_ALIASES] = { 650 { "load", "loads", "read", }, 651 { "store", "stores", "write", }, 652 { "prefetch", "prefetches", "speculative-read", "speculative-load", }, 653}; 654 655const char *const evsel__hw_cache_result[PERF_COUNT_HW_CACHE_RESULT_MAX][EVSEL__MAX_ALIASES] = { 656 { "refs", "Reference", "ops", "access", }, 657 { "misses", "miss", }, 658}; 659 660#define C(x) PERF_COUNT_HW_CACHE_##x 661#define CACHE_READ (1 << C(OP_READ)) 662#define CACHE_WRITE (1 << C(OP_WRITE)) 663#define CACHE_PREFETCH (1 << C(OP_PREFETCH)) 664#define COP(x) (1 << x) 665 666/* 667 * cache operation stat 668 * L1I : Read and prefetch only 669 * ITLB and BPU : Read-only 670 */ 671static const unsigned long evsel__hw_cache_stat[C(MAX)] = { 672 [C(L1D)] = (CACHE_READ | CACHE_WRITE | CACHE_PREFETCH), 673 [C(L1I)] = (CACHE_READ | CACHE_PREFETCH), 674 [C(LL)] = (CACHE_READ | CACHE_WRITE | CACHE_PREFETCH), 675 [C(DTLB)] = (CACHE_READ | CACHE_WRITE | CACHE_PREFETCH), 676 [C(ITLB)] = (CACHE_READ), 677 [C(BPU)] = (CACHE_READ), 678 [C(NODE)] = (CACHE_READ | CACHE_WRITE | CACHE_PREFETCH), 679}; 680 681bool evsel__is_cache_op_valid(u8 type, u8 op) 682{ 683 if (evsel__hw_cache_stat[type] & COP(op)) 684 return true; /* valid */ 685 else 686 return false; /* invalid */ 687} 688 689int __evsel__hw_cache_type_op_res_name(u8 type, u8 op, u8 result, char *bf, size_t size) 690{ 691 if (result) { 692 return scnprintf(bf, size, "%s-%s-%s", evsel__hw_cache[type][0], 693 evsel__hw_cache_op[op][0], 694 evsel__hw_cache_result[result][0]); 695 } 696 697 return scnprintf(bf, size, "%s-%s", evsel__hw_cache[type][0], 698 evsel__hw_cache_op[op][1]); 699} 700 701static int __evsel__hw_cache_name(u64 config, char *bf, size_t size) 702{ 703 u8 op, result, type = (config >> 0) & 0xff; 704 const char *err = "unknown-ext-hardware-cache-type"; 705 706 if (type >= PERF_COUNT_HW_CACHE_MAX) 707 goto out_err; 708 709 op = (config >> 8) & 0xff; 710 err = "unknown-ext-hardware-cache-op"; 711 if (op >= PERF_COUNT_HW_CACHE_OP_MAX) 712 goto out_err; 713 714 result = (config >> 16) & 0xff; 715 err = "unknown-ext-hardware-cache-result"; 716 if (result >= PERF_COUNT_HW_CACHE_RESULT_MAX) 717 goto out_err; 718 719 err = "invalid-cache"; 720 if (!evsel__is_cache_op_valid(type, op)) 721 goto out_err; 722 723 return __evsel__hw_cache_type_op_res_name(type, op, result, bf, size); 724out_err: 725 return scnprintf(bf, size, "%s", err); 726} 727 728static int evsel__hw_cache_name(struct evsel *evsel, char *bf, size_t size) 729{ 730 int ret = __evsel__hw_cache_name(evsel->core.attr.config, bf, size); 731 return ret + evsel__add_modifiers(evsel, bf + ret, size - ret); 732} 733 734static int evsel__raw_name(struct evsel *evsel, char *bf, size_t size) 735{ 736 int ret = scnprintf(bf, size, "raw 0x%" PRIx64, evsel->core.attr.config); 737 return ret + evsel__add_modifiers(evsel, bf + ret, size - ret); 738} 739 740const char *evsel__name(struct evsel *evsel) 741{ 742 char bf[128]; 743 744 if (!evsel) 745 goto out_unknown; 746 747 if (evsel->name) 748 return evsel->name; 749 750 switch (evsel->core.attr.type) { 751 case PERF_TYPE_RAW: 752 evsel__raw_name(evsel, bf, sizeof(bf)); 753 break; 754 755 case PERF_TYPE_HARDWARE: 756 evsel__hw_name(evsel, bf, sizeof(bf)); 757 break; 758 759 case PERF_TYPE_HW_CACHE: 760 evsel__hw_cache_name(evsel, bf, sizeof(bf)); 761 break; 762 763 case PERF_TYPE_SOFTWARE: 764 if (evsel__is_tool(evsel)) 765 evsel__tool_name(evsel->tool_event, bf, sizeof(bf)); 766 else 767 evsel__sw_name(evsel, bf, sizeof(bf)); 768 break; 769 770 case PERF_TYPE_TRACEPOINT: 771 scnprintf(bf, sizeof(bf), "%s", "unknown tracepoint"); 772 break; 773 774 case PERF_TYPE_BREAKPOINT: 775 evsel__bp_name(evsel, bf, sizeof(bf)); 776 break; 777 778 default: 779 scnprintf(bf, sizeof(bf), "unknown attr type: %d", 780 evsel->core.attr.type); 781 break; 782 } 783 784 evsel->name = strdup(bf); 785 786 if (evsel->name) 787 return evsel->name; 788out_unknown: 789 return "unknown"; 790} 791 792bool evsel__name_is(struct evsel *evsel, const char *name) 793{ 794 return !strcmp(evsel__name(evsel), name); 795} 796 797const char *evsel__metric_id(const struct evsel *evsel) 798{ 799 if (evsel->metric_id) 800 return evsel->metric_id; 801 802 if (evsel__is_tool(evsel)) 803 return perf_tool_event__to_str(evsel->tool_event); 804 805 return "unknown"; 806} 807 808const char *evsel__group_name(struct evsel *evsel) 809{ 810 return evsel->group_name ?: "anon group"; 811} 812 813/* 814 * Returns the group details for the specified leader, 815 * with following rules. 816 * 817 * For record -e '{cycles,instructions}' 818 * 'anon group { cycles:u, instructions:u }' 819 * 820 * For record -e 'cycles,instructions' and report --group 821 * 'cycles:u, instructions:u' 822 */ 823int evsel__group_desc(struct evsel *evsel, char *buf, size_t size) 824{ 825 int ret = 0; 826 struct evsel *pos; 827 const char *group_name = evsel__group_name(evsel); 828 829 if (!evsel->forced_leader) 830 ret = scnprintf(buf, size, "%s { ", group_name); 831 832 ret += scnprintf(buf + ret, size - ret, "%s", evsel__name(evsel)); 833 834 for_each_group_member(pos, evsel) 835 ret += scnprintf(buf + ret, size - ret, ", %s", evsel__name(pos)); 836 837 if (!evsel->forced_leader) 838 ret += scnprintf(buf + ret, size - ret, " }"); 839 840 return ret; 841} 842 843static void __evsel__config_callchain(struct evsel *evsel, struct record_opts *opts, 844 struct callchain_param *param) 845{ 846 bool function = evsel__is_function_event(evsel); 847 struct perf_event_attr *attr = &evsel->core.attr; 848 const char *arch = perf_env__arch(evsel__env(evsel)); 849 850 evsel__set_sample_bit(evsel, CALLCHAIN); 851 852 attr->sample_max_stack = param->max_stack; 853 854 if (opts->kernel_callchains) 855 attr->exclude_callchain_user = 1; 856 if (opts->user_callchains) 857 attr->exclude_callchain_kernel = 1; 858 if (param->record_mode == CALLCHAIN_LBR) { 859 if (!opts->branch_stack) { 860 if (attr->exclude_user) { 861 pr_warning("LBR callstack option is only available " 862 "to get user callchain information. " 863 "Falling back to framepointers.\n"); 864 } else { 865 evsel__set_sample_bit(evsel, BRANCH_STACK); 866 attr->branch_sample_type = PERF_SAMPLE_BRANCH_USER | 867 PERF_SAMPLE_BRANCH_CALL_STACK | 868 PERF_SAMPLE_BRANCH_NO_CYCLES | 869 PERF_SAMPLE_BRANCH_NO_FLAGS | 870 PERF_SAMPLE_BRANCH_HW_INDEX; 871 } 872 } else 873 pr_warning("Cannot use LBR callstack with branch stack. " 874 "Falling back to framepointers.\n"); 875 } 876 877 if (param->record_mode == CALLCHAIN_DWARF) { 878 if (!function) { 879 evsel__set_sample_bit(evsel, REGS_USER); 880 evsel__set_sample_bit(evsel, STACK_USER); 881 if (opts->sample_user_regs && 882 DWARF_MINIMAL_REGS(arch) != arch__user_reg_mask()) { 883 attr->sample_regs_user |= DWARF_MINIMAL_REGS(arch); 884 pr_warning("WARNING: The use of --call-graph=dwarf may require all the user registers, " 885 "specifying a subset with --user-regs may render DWARF unwinding unreliable, " 886 "so the minimal registers set (IP, SP) is explicitly forced.\n"); 887 } else { 888 attr->sample_regs_user |= arch__user_reg_mask(); 889 } 890 attr->sample_stack_user = param->dump_size; 891 attr->exclude_callchain_user = 1; 892 } else { 893 pr_info("Cannot use DWARF unwind for function trace event," 894 " falling back to framepointers.\n"); 895 } 896 } 897 898 if (function) { 899 pr_info("Disabling user space callchains for function trace event.\n"); 900 attr->exclude_callchain_user = 1; 901 } 902} 903 904void evsel__config_callchain(struct evsel *evsel, struct record_opts *opts, 905 struct callchain_param *param) 906{ 907 if (param->enabled) 908 return __evsel__config_callchain(evsel, opts, param); 909} 910 911static void evsel__reset_callgraph(struct evsel *evsel, struct callchain_param *param) 912{ 913 struct perf_event_attr *attr = &evsel->core.attr; 914 915 evsel__reset_sample_bit(evsel, CALLCHAIN); 916 if (param->record_mode == CALLCHAIN_LBR) { 917 evsel__reset_sample_bit(evsel, BRANCH_STACK); 918 attr->branch_sample_type &= ~(PERF_SAMPLE_BRANCH_USER | 919 PERF_SAMPLE_BRANCH_CALL_STACK | 920 PERF_SAMPLE_BRANCH_HW_INDEX); 921 } 922 if (param->record_mode == CALLCHAIN_DWARF) { 923 evsel__reset_sample_bit(evsel, REGS_USER); 924 evsel__reset_sample_bit(evsel, STACK_USER); 925 } 926} 927 928static void evsel__apply_config_terms(struct evsel *evsel, 929 struct record_opts *opts, bool track) 930{ 931 struct evsel_config_term *term; 932 struct list_head *config_terms = &evsel->config_terms; 933 struct perf_event_attr *attr = &evsel->core.attr; 934 /* callgraph default */ 935 struct callchain_param param = { 936 .record_mode = callchain_param.record_mode, 937 }; 938 u32 dump_size = 0; 939 int max_stack = 0; 940 const char *callgraph_buf = NULL; 941 942 list_for_each_entry(term, config_terms, list) { 943 switch (term->type) { 944 case EVSEL__CONFIG_TERM_PERIOD: 945 if (!(term->weak && opts->user_interval != ULLONG_MAX)) { 946 attr->sample_period = term->val.period; 947 attr->freq = 0; 948 evsel__reset_sample_bit(evsel, PERIOD); 949 } 950 break; 951 case EVSEL__CONFIG_TERM_FREQ: 952 if (!(term->weak && opts->user_freq != UINT_MAX)) { 953 attr->sample_freq = term->val.freq; 954 attr->freq = 1; 955 evsel__set_sample_bit(evsel, PERIOD); 956 } 957 break; 958 case EVSEL__CONFIG_TERM_TIME: 959 if (term->val.time) 960 evsel__set_sample_bit(evsel, TIME); 961 else 962 evsel__reset_sample_bit(evsel, TIME); 963 break; 964 case EVSEL__CONFIG_TERM_CALLGRAPH: 965 callgraph_buf = term->val.str; 966 break; 967 case EVSEL__CONFIG_TERM_BRANCH: 968 if (term->val.str && strcmp(term->val.str, "no")) { 969 evsel__set_sample_bit(evsel, BRANCH_STACK); 970 parse_branch_str(term->val.str, 971 &attr->branch_sample_type); 972 } else 973 evsel__reset_sample_bit(evsel, BRANCH_STACK); 974 break; 975 case EVSEL__CONFIG_TERM_STACK_USER: 976 dump_size = term->val.stack_user; 977 break; 978 case EVSEL__CONFIG_TERM_MAX_STACK: 979 max_stack = term->val.max_stack; 980 break; 981 case EVSEL__CONFIG_TERM_MAX_EVENTS: 982 evsel->max_events = term->val.max_events; 983 break; 984 case EVSEL__CONFIG_TERM_INHERIT: 985 /* 986 * attr->inherit should has already been set by 987 * evsel__config. If user explicitly set 988 * inherit using config terms, override global 989 * opt->no_inherit setting. 990 */ 991 attr->inherit = term->val.inherit ? 1 : 0; 992 break; 993 case EVSEL__CONFIG_TERM_OVERWRITE: 994 attr->write_backward = term->val.overwrite ? 1 : 0; 995 break; 996 case EVSEL__CONFIG_TERM_DRV_CFG: 997 break; 998 case EVSEL__CONFIG_TERM_PERCORE: 999 break; 1000 case EVSEL__CONFIG_TERM_AUX_OUTPUT: 1001 attr->aux_output = term->val.aux_output ? 1 : 0; 1002 break; 1003 case EVSEL__CONFIG_TERM_AUX_SAMPLE_SIZE: 1004 /* Already applied by auxtrace */ 1005 break; 1006 case EVSEL__CONFIG_TERM_CFG_CHG: 1007 break; 1008 default: 1009 break; 1010 } 1011 } 1012 1013 /* User explicitly set per-event callgraph, clear the old setting and reset. */ 1014 if ((callgraph_buf != NULL) || (dump_size > 0) || max_stack) { 1015 bool sample_address = false; 1016 1017 if (max_stack) { 1018 param.max_stack = max_stack; 1019 if (callgraph_buf == NULL) 1020 callgraph_buf = "fp"; 1021 } 1022 1023 /* parse callgraph parameters */ 1024 if (callgraph_buf != NULL) { 1025 if (!strcmp(callgraph_buf, "no")) { 1026 param.enabled = false; 1027 param.record_mode = CALLCHAIN_NONE; 1028 } else { 1029 param.enabled = true; 1030 if (parse_callchain_record(callgraph_buf, ¶m)) { 1031 pr_err("per-event callgraph setting for %s failed. " 1032 "Apply callgraph global setting for it\n", 1033 evsel->name); 1034 return; 1035 } 1036 if (param.record_mode == CALLCHAIN_DWARF) 1037 sample_address = true; 1038 } 1039 } 1040 if (dump_size > 0) { 1041 dump_size = round_up(dump_size, sizeof(u64)); 1042 param.dump_size = dump_size; 1043 } 1044 1045 /* If global callgraph set, clear it */ 1046 if (callchain_param.enabled) 1047 evsel__reset_callgraph(evsel, &callchain_param); 1048 1049 /* set perf-event callgraph */ 1050 if (param.enabled) { 1051 if (sample_address) { 1052 evsel__set_sample_bit(evsel, ADDR); 1053 evsel__set_sample_bit(evsel, DATA_SRC); 1054 evsel->core.attr.mmap_data = track; 1055 } 1056 evsel__config_callchain(evsel, opts, ¶m); 1057 } 1058 } 1059} 1060 1061struct evsel_config_term *__evsel__get_config_term(struct evsel *evsel, enum evsel_term_type type) 1062{ 1063 struct evsel_config_term *term, *found_term = NULL; 1064 1065 list_for_each_entry(term, &evsel->config_terms, list) { 1066 if (term->type == type) 1067 found_term = term; 1068 } 1069 1070 return found_term; 1071} 1072 1073void __weak arch_evsel__set_sample_weight(struct evsel *evsel) 1074{ 1075 evsel__set_sample_bit(evsel, WEIGHT); 1076} 1077 1078void __weak arch__post_evsel_config(struct evsel *evsel __maybe_unused, 1079 struct perf_event_attr *attr __maybe_unused) 1080{ 1081} 1082 1083static void evsel__set_default_freq_period(struct record_opts *opts, 1084 struct perf_event_attr *attr) 1085{ 1086 if (opts->freq) { 1087 attr->freq = 1; 1088 attr->sample_freq = opts->freq; 1089 } else { 1090 attr->sample_period = opts->default_interval; 1091 } 1092} 1093 1094static bool evsel__is_offcpu_event(struct evsel *evsel) 1095{ 1096 return evsel__is_bpf_output(evsel) && evsel__name_is(evsel, OFFCPU_EVENT); 1097} 1098 1099/* 1100 * The enable_on_exec/disabled value strategy: 1101 * 1102 * 1) For any type of traced program: 1103 * - all independent events and group leaders are disabled 1104 * - all group members are enabled 1105 * 1106 * Group members are ruled by group leaders. They need to 1107 * be enabled, because the group scheduling relies on that. 1108 * 1109 * 2) For traced programs executed by perf: 1110 * - all independent events and group leaders have 1111 * enable_on_exec set 1112 * - we don't specifically enable or disable any event during 1113 * the record command 1114 * 1115 * Independent events and group leaders are initially disabled 1116 * and get enabled by exec. Group members are ruled by group 1117 * leaders as stated in 1). 1118 * 1119 * 3) For traced programs attached by perf (pid/tid): 1120 * - we specifically enable or disable all events during 1121 * the record command 1122 * 1123 * When attaching events to already running traced we 1124 * enable/disable events specifically, as there's no 1125 * initial traced exec call. 1126 */ 1127void evsel__config(struct evsel *evsel, struct record_opts *opts, 1128 struct callchain_param *callchain) 1129{ 1130 struct evsel *leader = evsel__leader(evsel); 1131 struct perf_event_attr *attr = &evsel->core.attr; 1132 int track = evsel->tracking; 1133 bool per_cpu = opts->target.default_per_cpu && !opts->target.per_thread; 1134 1135 attr->sample_id_all = perf_missing_features.sample_id_all ? 0 : 1; 1136 attr->inherit = !opts->no_inherit; 1137 attr->write_backward = opts->overwrite ? 1 : 0; 1138 attr->read_format = PERF_FORMAT_LOST; 1139 1140 evsel__set_sample_bit(evsel, IP); 1141 evsel__set_sample_bit(evsel, TID); 1142 1143 if (evsel->sample_read) { 1144 evsel__set_sample_bit(evsel, READ); 1145 1146 /* 1147 * We need ID even in case of single event, because 1148 * PERF_SAMPLE_READ process ID specific data. 1149 */ 1150 evsel__set_sample_id(evsel, false); 1151 1152 /* 1153 * Apply group format only if we belong to group 1154 * with more than one members. 1155 */ 1156 if (leader->core.nr_members > 1) { 1157 attr->read_format |= PERF_FORMAT_GROUP; 1158 attr->inherit = 0; 1159 } 1160 } 1161 1162 /* 1163 * We default some events to have a default interval. But keep 1164 * it a weak assumption overridable by the user. 1165 */ 1166 if ((evsel->is_libpfm_event && !attr->sample_period) || 1167 (!evsel->is_libpfm_event && (!attr->sample_period || 1168 opts->user_freq != UINT_MAX || 1169 opts->user_interval != ULLONG_MAX))) 1170 evsel__set_default_freq_period(opts, attr); 1171 1172 /* 1173 * If attr->freq was set (here or earlier), ask for period 1174 * to be sampled. 1175 */ 1176 if (attr->freq) 1177 evsel__set_sample_bit(evsel, PERIOD); 1178 1179 if (opts->no_samples) 1180 attr->sample_freq = 0; 1181 1182 if (opts->inherit_stat) { 1183 evsel->core.attr.read_format |= 1184 PERF_FORMAT_TOTAL_TIME_ENABLED | 1185 PERF_FORMAT_TOTAL_TIME_RUNNING | 1186 PERF_FORMAT_ID; 1187 attr->inherit_stat = 1; 1188 } 1189 1190 if (opts->sample_address) { 1191 evsel__set_sample_bit(evsel, ADDR); 1192 attr->mmap_data = track; 1193 } 1194 1195 /* 1196 * We don't allow user space callchains for function trace 1197 * event, due to issues with page faults while tracing page 1198 * fault handler and its overall trickiness nature. 1199 */ 1200 if (evsel__is_function_event(evsel)) 1201 evsel->core.attr.exclude_callchain_user = 1; 1202 1203 if (callchain && callchain->enabled && !evsel->no_aux_samples) 1204 evsel__config_callchain(evsel, opts, callchain); 1205 1206 if (opts->sample_intr_regs && !evsel->no_aux_samples && 1207 !evsel__is_dummy_event(evsel)) { 1208 attr->sample_regs_intr = opts->sample_intr_regs; 1209 evsel__set_sample_bit(evsel, REGS_INTR); 1210 } 1211 1212 if (opts->sample_user_regs && !evsel->no_aux_samples && 1213 !evsel__is_dummy_event(evsel)) { 1214 attr->sample_regs_user |= opts->sample_user_regs; 1215 evsel__set_sample_bit(evsel, REGS_USER); 1216 } 1217 1218 if (target__has_cpu(&opts->target) || opts->sample_cpu) 1219 evsel__set_sample_bit(evsel, CPU); 1220 1221 /* 1222 * When the user explicitly disabled time don't force it here. 1223 */ 1224 if (opts->sample_time && 1225 (!perf_missing_features.sample_id_all && 1226 (!opts->no_inherit || target__has_cpu(&opts->target) || per_cpu || 1227 opts->sample_time_set))) 1228 evsel__set_sample_bit(evsel, TIME); 1229 1230 if (opts->raw_samples && !evsel->no_aux_samples) { 1231 evsel__set_sample_bit(evsel, TIME); 1232 evsel__set_sample_bit(evsel, RAW); 1233 evsel__set_sample_bit(evsel, CPU); 1234 } 1235 1236 if (opts->sample_address) 1237 evsel__set_sample_bit(evsel, DATA_SRC); 1238 1239 if (opts->sample_phys_addr) 1240 evsel__set_sample_bit(evsel, PHYS_ADDR); 1241 1242 if (opts->no_buffering) { 1243 attr->watermark = 0; 1244 attr->wakeup_events = 1; 1245 } 1246 if (opts->branch_stack && !evsel->no_aux_samples) { 1247 evsel__set_sample_bit(evsel, BRANCH_STACK); 1248 attr->branch_sample_type = opts->branch_stack; 1249 } 1250 1251 if (opts->sample_weight) 1252 arch_evsel__set_sample_weight(evsel); 1253 1254 attr->task = track; 1255 attr->mmap = track; 1256 attr->mmap2 = track && !perf_missing_features.mmap2; 1257 attr->comm = track; 1258 attr->build_id = track && opts->build_id; 1259 1260 /* 1261 * ksymbol is tracked separately with text poke because it needs to be 1262 * system wide and enabled immediately. 1263 */ 1264 if (!opts->text_poke) 1265 attr->ksymbol = track && !perf_missing_features.ksymbol; 1266 attr->bpf_event = track && !opts->no_bpf_event && !perf_missing_features.bpf; 1267 1268 if (opts->record_namespaces) 1269 attr->namespaces = track; 1270 1271 if (opts->record_cgroup) { 1272 attr->cgroup = track && !perf_missing_features.cgroup; 1273 evsel__set_sample_bit(evsel, CGROUP); 1274 } 1275 1276 if (opts->sample_data_page_size) 1277 evsel__set_sample_bit(evsel, DATA_PAGE_SIZE); 1278 1279 if (opts->sample_code_page_size) 1280 evsel__set_sample_bit(evsel, CODE_PAGE_SIZE); 1281 1282 if (opts->record_switch_events) 1283 attr->context_switch = track; 1284 1285 if (opts->sample_transaction) 1286 evsel__set_sample_bit(evsel, TRANSACTION); 1287 1288 if (opts->running_time) { 1289 evsel->core.attr.read_format |= 1290 PERF_FORMAT_TOTAL_TIME_ENABLED | 1291 PERF_FORMAT_TOTAL_TIME_RUNNING; 1292 } 1293 1294 /* 1295 * XXX see the function comment above 1296 * 1297 * Disabling only independent events or group leaders, 1298 * keeping group members enabled. 1299 */ 1300 if (evsel__is_group_leader(evsel)) 1301 attr->disabled = 1; 1302 1303 /* 1304 * Setting enable_on_exec for independent events and 1305 * group leaders for traced executed by perf. 1306 */ 1307 if (target__none(&opts->target) && evsel__is_group_leader(evsel) && 1308 !opts->target.initial_delay) 1309 attr->enable_on_exec = 1; 1310 1311 if (evsel->immediate) { 1312 attr->disabled = 0; 1313 attr->enable_on_exec = 0; 1314 } 1315 1316 clockid = opts->clockid; 1317 if (opts->use_clockid) { 1318 attr->use_clockid = 1; 1319 attr->clockid = opts->clockid; 1320 } 1321 1322 if (evsel->precise_max) 1323 attr->precise_ip = 3; 1324 1325 if (opts->all_user) { 1326 attr->exclude_kernel = 1; 1327 attr->exclude_user = 0; 1328 } 1329 1330 if (opts->all_kernel) { 1331 attr->exclude_kernel = 0; 1332 attr->exclude_user = 1; 1333 } 1334 1335 if (evsel->core.own_cpus || evsel->unit) 1336 evsel->core.attr.read_format |= PERF_FORMAT_ID; 1337 1338 /* 1339 * Apply event specific term settings, 1340 * it overloads any global configuration. 1341 */ 1342 evsel__apply_config_terms(evsel, opts, track); 1343 1344 evsel->ignore_missing_thread = opts->ignore_missing_thread; 1345 1346 /* The --period option takes the precedence. */ 1347 if (opts->period_set) { 1348 if (opts->period) 1349 evsel__set_sample_bit(evsel, PERIOD); 1350 else 1351 evsel__reset_sample_bit(evsel, PERIOD); 1352 } 1353 1354 /* 1355 * A dummy event never triggers any actual counter and therefore 1356 * cannot be used with branch_stack. 1357 * 1358 * For initial_delay, a dummy event is added implicitly. 1359 * The software event will trigger -EOPNOTSUPP error out, 1360 * if BRANCH_STACK bit is set. 1361 */ 1362 if (evsel__is_dummy_event(evsel)) 1363 evsel__reset_sample_bit(evsel, BRANCH_STACK); 1364 1365 if (evsel__is_offcpu_event(evsel)) 1366 evsel->core.attr.sample_type &= OFFCPU_SAMPLE_TYPES; 1367 1368 arch__post_evsel_config(evsel, attr); 1369} 1370 1371int evsel__set_filter(struct evsel *evsel, const char *filter) 1372{ 1373 char *new_filter = strdup(filter); 1374 1375 if (new_filter != NULL) { 1376 free(evsel->filter); 1377 evsel->filter = new_filter; 1378 return 0; 1379 } 1380 1381 return -1; 1382} 1383 1384static int evsel__append_filter(struct evsel *evsel, const char *fmt, const char *filter) 1385{ 1386 char *new_filter; 1387 1388 if (evsel->filter == NULL) 1389 return evsel__set_filter(evsel, filter); 1390 1391 if (asprintf(&new_filter, fmt, evsel->filter, filter) > 0) { 1392 free(evsel->filter); 1393 evsel->filter = new_filter; 1394 return 0; 1395 } 1396 1397 return -1; 1398} 1399 1400int evsel__append_tp_filter(struct evsel *evsel, const char *filter) 1401{ 1402 return evsel__append_filter(evsel, "(%s) && (%s)", filter); 1403} 1404 1405int evsel__append_addr_filter(struct evsel *evsel, const char *filter) 1406{ 1407 return evsel__append_filter(evsel, "%s,%s", filter); 1408} 1409 1410/* Caller has to clear disabled after going through all CPUs. */ 1411int evsel__enable_cpu(struct evsel *evsel, int cpu_map_idx) 1412{ 1413 return perf_evsel__enable_cpu(&evsel->core, cpu_map_idx); 1414} 1415 1416int evsel__enable(struct evsel *evsel) 1417{ 1418 int err = perf_evsel__enable(&evsel->core); 1419 1420 if (!err) 1421 evsel->disabled = false; 1422 return err; 1423} 1424 1425/* Caller has to set disabled after going through all CPUs. */ 1426int evsel__disable_cpu(struct evsel *evsel, int cpu_map_idx) 1427{ 1428 return perf_evsel__disable_cpu(&evsel->core, cpu_map_idx); 1429} 1430 1431int evsel__disable(struct evsel *evsel) 1432{ 1433 int err = perf_evsel__disable(&evsel->core); 1434 /* 1435 * We mark it disabled here so that tools that disable a event can 1436 * ignore events after they disable it. I.e. the ring buffer may have 1437 * already a few more events queued up before the kernel got the stop 1438 * request. 1439 */ 1440 if (!err) 1441 evsel->disabled = true; 1442 1443 return err; 1444} 1445 1446void free_config_terms(struct list_head *config_terms) 1447{ 1448 struct evsel_config_term *term, *h; 1449 1450 list_for_each_entry_safe(term, h, config_terms, list) { 1451 list_del_init(&term->list); 1452 if (term->free_str) 1453 zfree(&term->val.str); 1454 free(term); 1455 } 1456} 1457 1458static void evsel__free_config_terms(struct evsel *evsel) 1459{ 1460 free_config_terms(&evsel->config_terms); 1461} 1462 1463void evsel__exit(struct evsel *evsel) 1464{ 1465 assert(list_empty(&evsel->core.node)); 1466 assert(evsel->evlist == NULL); 1467 bpf_counter__destroy(evsel); 1468 perf_bpf_filter__destroy(evsel); 1469 evsel__free_counts(evsel); 1470 perf_evsel__free_fd(&evsel->core); 1471 perf_evsel__free_id(&evsel->core); 1472 evsel__free_config_terms(evsel); 1473 cgroup__put(evsel->cgrp); 1474 perf_cpu_map__put(evsel->core.cpus); 1475 perf_cpu_map__put(evsel->core.own_cpus); 1476 perf_thread_map__put(evsel->core.threads); 1477 zfree(&evsel->group_name); 1478 zfree(&evsel->name); 1479 zfree(&evsel->filter); 1480 zfree(&evsel->pmu_name); 1481 zfree(&evsel->group_pmu_name); 1482 zfree(&evsel->unit); 1483 zfree(&evsel->metric_id); 1484 evsel__zero_per_pkg(evsel); 1485 hashmap__free(evsel->per_pkg_mask); 1486 evsel->per_pkg_mask = NULL; 1487 zfree(&evsel->metric_events); 1488 perf_evsel__object.fini(evsel); 1489} 1490 1491void evsel__delete(struct evsel *evsel) 1492{ 1493 if (!evsel) 1494 return; 1495 1496 evsel__exit(evsel); 1497 free(evsel); 1498} 1499 1500void evsel__compute_deltas(struct evsel *evsel, int cpu_map_idx, int thread, 1501 struct perf_counts_values *count) 1502{ 1503 struct perf_counts_values tmp; 1504 1505 if (!evsel->prev_raw_counts) 1506 return; 1507 1508 tmp = *perf_counts(evsel->prev_raw_counts, cpu_map_idx, thread); 1509 *perf_counts(evsel->prev_raw_counts, cpu_map_idx, thread) = *count; 1510 1511 count->val = count->val - tmp.val; 1512 count->ena = count->ena - tmp.ena; 1513 count->run = count->run - tmp.run; 1514} 1515 1516static int evsel__read_one(struct evsel *evsel, int cpu_map_idx, int thread) 1517{ 1518 struct perf_counts_values *count = perf_counts(evsel->counts, cpu_map_idx, thread); 1519 1520 return perf_evsel__read(&evsel->core, cpu_map_idx, thread, count); 1521} 1522 1523static void evsel__set_count(struct evsel *counter, int cpu_map_idx, int thread, 1524 u64 val, u64 ena, u64 run, u64 lost) 1525{ 1526 struct perf_counts_values *count; 1527 1528 count = perf_counts(counter->counts, cpu_map_idx, thread); 1529 1530 count->val = val; 1531 count->ena = ena; 1532 count->run = run; 1533 count->lost = lost; 1534 1535 perf_counts__set_loaded(counter->counts, cpu_map_idx, thread, true); 1536} 1537 1538static int evsel__process_group_data(struct evsel *leader, int cpu_map_idx, int thread, u64 *data) 1539{ 1540 u64 read_format = leader->core.attr.read_format; 1541 struct sample_read_value *v; 1542 u64 nr, ena = 0, run = 0, lost = 0; 1543 1544 nr = *data++; 1545 1546 if (nr != (u64) leader->core.nr_members) 1547 return -EINVAL; 1548 1549 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) 1550 ena = *data++; 1551 1552 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) 1553 run = *data++; 1554 1555 v = (void *)data; 1556 sample_read_group__for_each(v, nr, read_format) { 1557 struct evsel *counter; 1558 1559 counter = evlist__id2evsel(leader->evlist, v->id); 1560 if (!counter) 1561 return -EINVAL; 1562 1563 if (read_format & PERF_FORMAT_LOST) 1564 lost = v->lost; 1565 1566 evsel__set_count(counter, cpu_map_idx, thread, v->value, ena, run, lost); 1567 } 1568 1569 return 0; 1570} 1571 1572static int evsel__read_group(struct evsel *leader, int cpu_map_idx, int thread) 1573{ 1574 struct perf_stat_evsel *ps = leader->stats; 1575 u64 read_format = leader->core.attr.read_format; 1576 int size = perf_evsel__read_size(&leader->core); 1577 u64 *data = ps->group_data; 1578 1579 if (!(read_format & PERF_FORMAT_ID)) 1580 return -EINVAL; 1581 1582 if (!evsel__is_group_leader(leader)) 1583 return -EINVAL; 1584 1585 if (!data) { 1586 data = zalloc(size); 1587 if (!data) 1588 return -ENOMEM; 1589 1590 ps->group_data = data; 1591 } 1592 1593 if (FD(leader, cpu_map_idx, thread) < 0) 1594 return -EINVAL; 1595 1596 if (readn(FD(leader, cpu_map_idx, thread), data, size) <= 0) 1597 return -errno; 1598 1599 return evsel__process_group_data(leader, cpu_map_idx, thread, data); 1600} 1601 1602int evsel__read_counter(struct evsel *evsel, int cpu_map_idx, int thread) 1603{ 1604 u64 read_format = evsel->core.attr.read_format; 1605 1606 if (read_format & PERF_FORMAT_GROUP) 1607 return evsel__read_group(evsel, cpu_map_idx, thread); 1608 1609 return evsel__read_one(evsel, cpu_map_idx, thread); 1610} 1611 1612int __evsel__read_on_cpu(struct evsel *evsel, int cpu_map_idx, int thread, bool scale) 1613{ 1614 struct perf_counts_values count; 1615 size_t nv = scale ? 3 : 1; 1616 1617 if (FD(evsel, cpu_map_idx, thread) < 0) 1618 return -EINVAL; 1619 1620 if (evsel->counts == NULL && evsel__alloc_counts(evsel) < 0) 1621 return -ENOMEM; 1622 1623 if (readn(FD(evsel, cpu_map_idx, thread), &count, nv * sizeof(u64)) <= 0) 1624 return -errno; 1625 1626 evsel__compute_deltas(evsel, cpu_map_idx, thread, &count); 1627 perf_counts_values__scale(&count, scale, NULL); 1628 *perf_counts(evsel->counts, cpu_map_idx, thread) = count; 1629 return 0; 1630} 1631 1632static int evsel__match_other_cpu(struct evsel *evsel, struct evsel *other, 1633 int cpu_map_idx) 1634{ 1635 struct perf_cpu cpu; 1636 1637 cpu = perf_cpu_map__cpu(evsel->core.cpus, cpu_map_idx); 1638 return perf_cpu_map__idx(other->core.cpus, cpu); 1639} 1640 1641static int evsel__hybrid_group_cpu_map_idx(struct evsel *evsel, int cpu_map_idx) 1642{ 1643 struct evsel *leader = evsel__leader(evsel); 1644 1645 if ((evsel__is_hybrid(evsel) && !evsel__is_hybrid(leader)) || 1646 (!evsel__is_hybrid(evsel) && evsel__is_hybrid(leader))) { 1647 return evsel__match_other_cpu(evsel, leader, cpu_map_idx); 1648 } 1649 1650 return cpu_map_idx; 1651} 1652 1653static int get_group_fd(struct evsel *evsel, int cpu_map_idx, int thread) 1654{ 1655 struct evsel *leader = evsel__leader(evsel); 1656 int fd; 1657 1658 if (evsel__is_group_leader(evsel)) 1659 return -1; 1660 1661 /* 1662 * Leader must be already processed/open, 1663 * if not it's a bug. 1664 */ 1665 BUG_ON(!leader->core.fd); 1666 1667 cpu_map_idx = evsel__hybrid_group_cpu_map_idx(evsel, cpu_map_idx); 1668 if (cpu_map_idx == -1) 1669 return -1; 1670 1671 fd = FD(leader, cpu_map_idx, thread); 1672 BUG_ON(fd == -1 && !leader->skippable); 1673 1674 /* 1675 * When the leader has been skipped, return -2 to distinguish from no 1676 * group leader case. 1677 */ 1678 return fd == -1 ? -2 : fd; 1679} 1680 1681static void evsel__remove_fd(struct evsel *pos, int nr_cpus, int nr_threads, int thread_idx) 1682{ 1683 for (int cpu = 0; cpu < nr_cpus; cpu++) 1684 for (int thread = thread_idx; thread < nr_threads - 1; thread++) 1685 FD(pos, cpu, thread) = FD(pos, cpu, thread + 1); 1686} 1687 1688static int update_fds(struct evsel *evsel, 1689 int nr_cpus, int cpu_map_idx, 1690 int nr_threads, int thread_idx) 1691{ 1692 struct evsel *pos; 1693 1694 if (cpu_map_idx >= nr_cpus || thread_idx >= nr_threads) 1695 return -EINVAL; 1696 1697 evlist__for_each_entry(evsel->evlist, pos) { 1698 nr_cpus = pos != evsel ? nr_cpus : cpu_map_idx; 1699 1700 evsel__remove_fd(pos, nr_cpus, nr_threads, thread_idx); 1701 1702 /* 1703 * Since fds for next evsel has not been created, 1704 * there is no need to iterate whole event list. 1705 */ 1706 if (pos == evsel) 1707 break; 1708 } 1709 return 0; 1710} 1711 1712static bool evsel__ignore_missing_thread(struct evsel *evsel, 1713 int nr_cpus, int cpu_map_idx, 1714 struct perf_thread_map *threads, 1715 int thread, int err) 1716{ 1717 pid_t ignore_pid = perf_thread_map__pid(threads, thread); 1718 1719 if (!evsel->ignore_missing_thread) 1720 return false; 1721 1722 /* The system wide setup does not work with threads. */ 1723 if (evsel->core.system_wide) 1724 return false; 1725 1726 /* The -ESRCH is perf event syscall errno for pid's not found. */ 1727 if (err != -ESRCH) 1728 return false; 1729 1730 /* If there's only one thread, let it fail. */ 1731 if (threads->nr == 1) 1732 return false; 1733 1734 /* 1735 * We should remove fd for missing_thread first 1736 * because thread_map__remove() will decrease threads->nr. 1737 */ 1738 if (update_fds(evsel, nr_cpus, cpu_map_idx, threads->nr, thread)) 1739 return false; 1740 1741 if (thread_map__remove(threads, thread)) 1742 return false; 1743 1744 pr_warning("WARNING: Ignored open failure for pid %d\n", 1745 ignore_pid); 1746 return true; 1747} 1748 1749static int __open_attr__fprintf(FILE *fp, const char *name, const char *val, 1750 void *priv __maybe_unused) 1751{ 1752 return fprintf(fp, " %-32s %s\n", name, val); 1753} 1754 1755static void display_attr(struct perf_event_attr *attr) 1756{ 1757 if (verbose >= 2 || debug_peo_args) { 1758 fprintf(stderr, "%.60s\n", graph_dotted_line); 1759 fprintf(stderr, "perf_event_attr:\n"); 1760 perf_event_attr__fprintf(stderr, attr, __open_attr__fprintf, NULL); 1761 fprintf(stderr, "%.60s\n", graph_dotted_line); 1762 } 1763} 1764 1765bool evsel__precise_ip_fallback(struct evsel *evsel) 1766{ 1767 /* Do not try less precise if not requested. */ 1768 if (!evsel->precise_max) 1769 return false; 1770 1771 /* 1772 * We tried all the precise_ip values, and it's 1773 * still failing, so leave it to standard fallback. 1774 */ 1775 if (!evsel->core.attr.precise_ip) { 1776 evsel->core.attr.precise_ip = evsel->precise_ip_original; 1777 return false; 1778 } 1779 1780 if (!evsel->precise_ip_original) 1781 evsel->precise_ip_original = evsel->core.attr.precise_ip; 1782 1783 evsel->core.attr.precise_ip--; 1784 pr_debug2_peo("decreasing precise_ip by one (%d)\n", evsel->core.attr.precise_ip); 1785 display_attr(&evsel->core.attr); 1786 return true; 1787} 1788 1789static struct perf_cpu_map *empty_cpu_map; 1790static struct perf_thread_map *empty_thread_map; 1791 1792static int __evsel__prepare_open(struct evsel *evsel, struct perf_cpu_map *cpus, 1793 struct perf_thread_map *threads) 1794{ 1795 int nthreads = perf_thread_map__nr(threads); 1796 1797 if ((perf_missing_features.write_backward && evsel->core.attr.write_backward) || 1798 (perf_missing_features.aux_output && evsel->core.attr.aux_output)) 1799 return -EINVAL; 1800 1801 if (cpus == NULL) { 1802 if (empty_cpu_map == NULL) { 1803 empty_cpu_map = perf_cpu_map__dummy_new(); 1804 if (empty_cpu_map == NULL) 1805 return -ENOMEM; 1806 } 1807 1808 cpus = empty_cpu_map; 1809 } 1810 1811 if (threads == NULL) { 1812 if (empty_thread_map == NULL) { 1813 empty_thread_map = thread_map__new_by_tid(-1); 1814 if (empty_thread_map == NULL) 1815 return -ENOMEM; 1816 } 1817 1818 threads = empty_thread_map; 1819 } 1820 1821 if (evsel->core.fd == NULL && 1822 perf_evsel__alloc_fd(&evsel->core, perf_cpu_map__nr(cpus), nthreads) < 0) 1823 return -ENOMEM; 1824 1825 evsel->open_flags = PERF_FLAG_FD_CLOEXEC; 1826 if (evsel->cgrp) 1827 evsel->open_flags |= PERF_FLAG_PID_CGROUP; 1828 1829 return 0; 1830} 1831 1832static void evsel__disable_missing_features(struct evsel *evsel) 1833{ 1834 if (perf_missing_features.read_lost) 1835 evsel->core.attr.read_format &= ~PERF_FORMAT_LOST; 1836 if (perf_missing_features.weight_struct) { 1837 evsel__set_sample_bit(evsel, WEIGHT); 1838 evsel__reset_sample_bit(evsel, WEIGHT_STRUCT); 1839 } 1840 if (perf_missing_features.clockid_wrong) 1841 evsel->core.attr.clockid = CLOCK_MONOTONIC; /* should always work */ 1842 if (perf_missing_features.clockid) { 1843 evsel->core.attr.use_clockid = 0; 1844 evsel->core.attr.clockid = 0; 1845 } 1846 if (perf_missing_features.cloexec) 1847 evsel->open_flags &= ~(unsigned long)PERF_FLAG_FD_CLOEXEC; 1848 if (perf_missing_features.mmap2) 1849 evsel->core.attr.mmap2 = 0; 1850 if (evsel->pmu && evsel->pmu->missing_features.exclude_guest) 1851 evsel->core.attr.exclude_guest = evsel->core.attr.exclude_host = 0; 1852 if (perf_missing_features.lbr_flags) 1853 evsel->core.attr.branch_sample_type &= ~(PERF_SAMPLE_BRANCH_NO_FLAGS | 1854 PERF_SAMPLE_BRANCH_NO_CYCLES); 1855 if (perf_missing_features.group_read && evsel->core.attr.inherit) 1856 evsel->core.attr.read_format &= ~(PERF_FORMAT_GROUP|PERF_FORMAT_ID); 1857 if (perf_missing_features.ksymbol) 1858 evsel->core.attr.ksymbol = 0; 1859 if (perf_missing_features.bpf) 1860 evsel->core.attr.bpf_event = 0; 1861 if (perf_missing_features.branch_hw_idx) 1862 evsel->core.attr.branch_sample_type &= ~PERF_SAMPLE_BRANCH_HW_INDEX; 1863 if (perf_missing_features.sample_id_all) 1864 evsel->core.attr.sample_id_all = 0; 1865} 1866 1867int evsel__prepare_open(struct evsel *evsel, struct perf_cpu_map *cpus, 1868 struct perf_thread_map *threads) 1869{ 1870 int err; 1871 1872 err = __evsel__prepare_open(evsel, cpus, threads); 1873 if (err) 1874 return err; 1875 1876 evsel__disable_missing_features(evsel); 1877 1878 return err; 1879} 1880 1881bool evsel__detect_missing_features(struct evsel *evsel) 1882{ 1883 /* 1884 * Must probe features in the order they were added to the 1885 * perf_event_attr interface. 1886 */ 1887 if (!perf_missing_features.read_lost && 1888 (evsel->core.attr.read_format & PERF_FORMAT_LOST)) { 1889 perf_missing_features.read_lost = true; 1890 pr_debug2("switching off PERF_FORMAT_LOST support\n"); 1891 return true; 1892 } else if (!perf_missing_features.weight_struct && 1893 (evsel->core.attr.sample_type & PERF_SAMPLE_WEIGHT_STRUCT)) { 1894 perf_missing_features.weight_struct = true; 1895 pr_debug2("switching off weight struct support\n"); 1896 return true; 1897 } else if (!perf_missing_features.code_page_size && 1898 (evsel->core.attr.sample_type & PERF_SAMPLE_CODE_PAGE_SIZE)) { 1899 perf_missing_features.code_page_size = true; 1900 pr_debug2_peo("Kernel has no PERF_SAMPLE_CODE_PAGE_SIZE support, bailing out\n"); 1901 return false; 1902 } else if (!perf_missing_features.data_page_size && 1903 (evsel->core.attr.sample_type & PERF_SAMPLE_DATA_PAGE_SIZE)) { 1904 perf_missing_features.data_page_size = true; 1905 pr_debug2_peo("Kernel has no PERF_SAMPLE_DATA_PAGE_SIZE support, bailing out\n"); 1906 return false; 1907 } else if (!perf_missing_features.cgroup && evsel->core.attr.cgroup) { 1908 perf_missing_features.cgroup = true; 1909 pr_debug2_peo("Kernel has no cgroup sampling support, bailing out\n"); 1910 return false; 1911 } else if (!perf_missing_features.branch_hw_idx && 1912 (evsel->core.attr.branch_sample_type & PERF_SAMPLE_BRANCH_HW_INDEX)) { 1913 perf_missing_features.branch_hw_idx = true; 1914 pr_debug2("switching off branch HW index support\n"); 1915 return true; 1916 } else if (!perf_missing_features.aux_output && evsel->core.attr.aux_output) { 1917 perf_missing_features.aux_output = true; 1918 pr_debug2_peo("Kernel has no attr.aux_output support, bailing out\n"); 1919 return false; 1920 } else if (!perf_missing_features.bpf && evsel->core.attr.bpf_event) { 1921 perf_missing_features.bpf = true; 1922 pr_debug2_peo("switching off bpf_event\n"); 1923 return true; 1924 } else if (!perf_missing_features.ksymbol && evsel->core.attr.ksymbol) { 1925 perf_missing_features.ksymbol = true; 1926 pr_debug2_peo("switching off ksymbol\n"); 1927 return true; 1928 } else if (!perf_missing_features.write_backward && evsel->core.attr.write_backward) { 1929 perf_missing_features.write_backward = true; 1930 pr_debug2_peo("switching off write_backward\n"); 1931 return false; 1932 } else if (!perf_missing_features.clockid_wrong && evsel->core.attr.use_clockid) { 1933 perf_missing_features.clockid_wrong = true; 1934 pr_debug2_peo("switching off clockid\n"); 1935 return true; 1936 } else if (!perf_missing_features.clockid && evsel->core.attr.use_clockid) { 1937 perf_missing_features.clockid = true; 1938 pr_debug2_peo("switching off use_clockid\n"); 1939 return true; 1940 } else if (!perf_missing_features.cloexec && (evsel->open_flags & PERF_FLAG_FD_CLOEXEC)) { 1941 perf_missing_features.cloexec = true; 1942 pr_debug2_peo("switching off cloexec flag\n"); 1943 return true; 1944 } else if (!perf_missing_features.mmap2 && evsel->core.attr.mmap2) { 1945 perf_missing_features.mmap2 = true; 1946 pr_debug2_peo("switching off mmap2\n"); 1947 return true; 1948 } else if (evsel->core.attr.exclude_guest || evsel->core.attr.exclude_host) { 1949 if (evsel->pmu == NULL) 1950 evsel->pmu = evsel__find_pmu(evsel); 1951 1952 if (evsel->pmu) 1953 evsel->pmu->missing_features.exclude_guest = true; 1954 else { 1955 /* we cannot find PMU, disable attrs now */ 1956 evsel->core.attr.exclude_host = false; 1957 evsel->core.attr.exclude_guest = false; 1958 } 1959 1960 if (evsel->exclude_GH) { 1961 pr_debug2_peo("PMU has no exclude_host/guest support, bailing out\n"); 1962 return false; 1963 } 1964 if (!perf_missing_features.exclude_guest) { 1965 perf_missing_features.exclude_guest = true; 1966 pr_debug2_peo("switching off exclude_guest, exclude_host\n"); 1967 } 1968 return true; 1969 } else if (!perf_missing_features.sample_id_all) { 1970 perf_missing_features.sample_id_all = true; 1971 pr_debug2_peo("switching off sample_id_all\n"); 1972 return true; 1973 } else if (!perf_missing_features.lbr_flags && 1974 (evsel->core.attr.branch_sample_type & 1975 (PERF_SAMPLE_BRANCH_NO_CYCLES | 1976 PERF_SAMPLE_BRANCH_NO_FLAGS))) { 1977 perf_missing_features.lbr_flags = true; 1978 pr_debug2_peo("switching off branch sample type no (cycles/flags)\n"); 1979 return true; 1980 } else if (!perf_missing_features.group_read && 1981 evsel->core.attr.inherit && 1982 (evsel->core.attr.read_format & PERF_FORMAT_GROUP) && 1983 evsel__is_group_leader(evsel)) { 1984 perf_missing_features.group_read = true; 1985 pr_debug2_peo("switching off group read\n"); 1986 return true; 1987 } else { 1988 return false; 1989 } 1990} 1991 1992bool evsel__increase_rlimit(enum rlimit_action *set_rlimit) 1993{ 1994 int old_errno; 1995 struct rlimit l; 1996 1997 if (*set_rlimit < INCREASED_MAX) { 1998 old_errno = errno; 1999 2000 if (getrlimit(RLIMIT_NOFILE, &l) == 0) { 2001 if (*set_rlimit == NO_CHANGE) { 2002 l.rlim_cur = l.rlim_max; 2003 } else { 2004 l.rlim_cur = l.rlim_max + 1000; 2005 l.rlim_max = l.rlim_cur; 2006 } 2007 if (setrlimit(RLIMIT_NOFILE, &l) == 0) { 2008 (*set_rlimit) += 1; 2009 errno = old_errno; 2010 return true; 2011 } 2012 } 2013 errno = old_errno; 2014 } 2015 2016 return false; 2017} 2018 2019static int evsel__open_cpu(struct evsel *evsel, struct perf_cpu_map *cpus, 2020 struct perf_thread_map *threads, 2021 int start_cpu_map_idx, int end_cpu_map_idx) 2022{ 2023 int idx, thread, nthreads; 2024 int pid = -1, err, old_errno; 2025 enum rlimit_action set_rlimit = NO_CHANGE; 2026 2027 err = __evsel__prepare_open(evsel, cpus, threads); 2028 if (err) 2029 return err; 2030 2031 if (cpus == NULL) 2032 cpus = empty_cpu_map; 2033 2034 if (threads == NULL) 2035 threads = empty_thread_map; 2036 2037 nthreads = perf_thread_map__nr(threads); 2038 2039 if (evsel->cgrp) 2040 pid = evsel->cgrp->fd; 2041 2042fallback_missing_features: 2043 evsel__disable_missing_features(evsel); 2044 2045 pr_debug3("Opening: %s\n", evsel__name(evsel)); 2046 display_attr(&evsel->core.attr); 2047 2048 for (idx = start_cpu_map_idx; idx < end_cpu_map_idx; idx++) { 2049 2050 for (thread = 0; thread < nthreads; thread++) { 2051 int fd, group_fd; 2052retry_open: 2053 if (thread >= nthreads) 2054 break; 2055 2056 if (!evsel->cgrp && !evsel->core.system_wide) 2057 pid = perf_thread_map__pid(threads, thread); 2058 2059 group_fd = get_group_fd(evsel, idx, thread); 2060 2061 if (group_fd == -2) { 2062 pr_debug("broken group leader for %s\n", evsel->name); 2063 err = -EINVAL; 2064 goto out_close; 2065 } 2066 2067 test_attr__ready(); 2068 2069 /* Debug message used by test scripts */ 2070 pr_debug2_peo("sys_perf_event_open: pid %d cpu %d group_fd %d flags %#lx", 2071 pid, perf_cpu_map__cpu(cpus, idx).cpu, group_fd, evsel->open_flags); 2072 2073 fd = sys_perf_event_open(&evsel->core.attr, pid, 2074 perf_cpu_map__cpu(cpus, idx).cpu, 2075 group_fd, evsel->open_flags); 2076 2077 FD(evsel, idx, thread) = fd; 2078 2079 if (fd < 0) { 2080 err = -errno; 2081 2082 pr_debug2_peo("\nsys_perf_event_open failed, error %d\n", 2083 err); 2084 goto try_fallback; 2085 } 2086 2087 bpf_counter__install_pe(evsel, idx, fd); 2088 2089 if (unlikely(test_attr__enabled)) { 2090 test_attr__open(&evsel->core.attr, pid, 2091 perf_cpu_map__cpu(cpus, idx), 2092 fd, group_fd, evsel->open_flags); 2093 } 2094 2095 /* Debug message used by test scripts */ 2096 pr_debug2_peo(" = %d\n", fd); 2097 2098 if (evsel->bpf_fd >= 0) { 2099 int evt_fd = fd; 2100 int bpf_fd = evsel->bpf_fd; 2101 2102 err = ioctl(evt_fd, 2103 PERF_EVENT_IOC_SET_BPF, 2104 bpf_fd); 2105 if (err && errno != EEXIST) { 2106 pr_err("failed to attach bpf fd %d: %s\n", 2107 bpf_fd, strerror(errno)); 2108 err = -EINVAL; 2109 goto out_close; 2110 } 2111 } 2112 2113 set_rlimit = NO_CHANGE; 2114 2115 /* 2116 * If we succeeded but had to kill clockid, fail and 2117 * have evsel__open_strerror() print us a nice error. 2118 */ 2119 if (perf_missing_features.clockid || 2120 perf_missing_features.clockid_wrong) { 2121 err = -EINVAL; 2122 goto out_close; 2123 } 2124 } 2125 } 2126 2127 return 0; 2128 2129try_fallback: 2130 if (evsel__precise_ip_fallback(evsel)) 2131 goto retry_open; 2132 2133 if (evsel__ignore_missing_thread(evsel, perf_cpu_map__nr(cpus), 2134 idx, threads, thread, err)) { 2135 /* We just removed 1 thread, so lower the upper nthreads limit. */ 2136 nthreads--; 2137 2138 /* ... and pretend like nothing have happened. */ 2139 err = 0; 2140 goto retry_open; 2141 } 2142 /* 2143 * perf stat needs between 5 and 22 fds per CPU. When we run out 2144 * of them try to increase the limits. 2145 */ 2146 if (err == -EMFILE && evsel__increase_rlimit(&set_rlimit)) 2147 goto retry_open; 2148 2149 if (err != -EINVAL || idx > 0 || thread > 0) 2150 goto out_close; 2151 2152 if (evsel__detect_missing_features(evsel)) 2153 goto fallback_missing_features; 2154out_close: 2155 if (err) 2156 threads->err_thread = thread; 2157 2158 old_errno = errno; 2159 do { 2160 while (--thread >= 0) { 2161 if (FD(evsel, idx, thread) >= 0) 2162 close(FD(evsel, idx, thread)); 2163 FD(evsel, idx, thread) = -1; 2164 } 2165 thread = nthreads; 2166 } while (--idx >= 0); 2167 errno = old_errno; 2168 return err; 2169} 2170 2171int evsel__open(struct evsel *evsel, struct perf_cpu_map *cpus, 2172 struct perf_thread_map *threads) 2173{ 2174 return evsel__open_cpu(evsel, cpus, threads, 0, perf_cpu_map__nr(cpus)); 2175} 2176 2177void evsel__close(struct evsel *evsel) 2178{ 2179 perf_evsel__close(&evsel->core); 2180 perf_evsel__free_id(&evsel->core); 2181} 2182 2183int evsel__open_per_cpu(struct evsel *evsel, struct perf_cpu_map *cpus, int cpu_map_idx) 2184{ 2185 if (cpu_map_idx == -1) 2186 return evsel__open_cpu(evsel, cpus, NULL, 0, perf_cpu_map__nr(cpus)); 2187 2188 return evsel__open_cpu(evsel, cpus, NULL, cpu_map_idx, cpu_map_idx + 1); 2189} 2190 2191int evsel__open_per_thread(struct evsel *evsel, struct perf_thread_map *threads) 2192{ 2193 return evsel__open(evsel, NULL, threads); 2194} 2195 2196static int perf_evsel__parse_id_sample(const struct evsel *evsel, 2197 const union perf_event *event, 2198 struct perf_sample *sample) 2199{ 2200 u64 type = evsel->core.attr.sample_type; 2201 const __u64 *array = event->sample.array; 2202 bool swapped = evsel->needs_swap; 2203 union u64_swap u; 2204 2205 array += ((event->header.size - 2206 sizeof(event->header)) / sizeof(u64)) - 1; 2207 2208 if (type & PERF_SAMPLE_IDENTIFIER) { 2209 sample->id = *array; 2210 array--; 2211 } 2212 2213 if (type & PERF_SAMPLE_CPU) { 2214 u.val64 = *array; 2215 if (swapped) { 2216 /* undo swap of u64, then swap on individual u32s */ 2217 u.val64 = bswap_64(u.val64); 2218 u.val32[0] = bswap_32(u.val32[0]); 2219 } 2220 2221 sample->cpu = u.val32[0]; 2222 array--; 2223 } 2224 2225 if (type & PERF_SAMPLE_STREAM_ID) { 2226 sample->stream_id = *array; 2227 array--; 2228 } 2229 2230 if (type & PERF_SAMPLE_ID) { 2231 sample->id = *array; 2232 array--; 2233 } 2234 2235 if (type & PERF_SAMPLE_TIME) { 2236 sample->time = *array; 2237 array--; 2238 } 2239 2240 if (type & PERF_SAMPLE_TID) { 2241 u.val64 = *array; 2242 if (swapped) { 2243 /* undo swap of u64, then swap on individual u32s */ 2244 u.val64 = bswap_64(u.val64); 2245 u.val32[0] = bswap_32(u.val32[0]); 2246 u.val32[1] = bswap_32(u.val32[1]); 2247 } 2248 2249 sample->pid = u.val32[0]; 2250 sample->tid = u.val32[1]; 2251 array--; 2252 } 2253 2254 return 0; 2255} 2256 2257static inline bool overflow(const void *endp, u16 max_size, const void *offset, 2258 u64 size) 2259{ 2260 return size > max_size || offset + size > endp; 2261} 2262 2263#define OVERFLOW_CHECK(offset, size, max_size) \ 2264 do { \ 2265 if (overflow(endp, (max_size), (offset), (size))) \ 2266 return -EFAULT; \ 2267 } while (0) 2268 2269#define OVERFLOW_CHECK_u64(offset) \ 2270 OVERFLOW_CHECK(offset, sizeof(u64), sizeof(u64)) 2271 2272static int 2273perf_event__check_size(union perf_event *event, unsigned int sample_size) 2274{ 2275 /* 2276 * The evsel's sample_size is based on PERF_SAMPLE_MASK which includes 2277 * up to PERF_SAMPLE_PERIOD. After that overflow() must be used to 2278 * check the format does not go past the end of the event. 2279 */ 2280 if (sample_size + sizeof(event->header) > event->header.size) 2281 return -EFAULT; 2282 2283 return 0; 2284} 2285 2286void __weak arch_perf_parse_sample_weight(struct perf_sample *data, 2287 const __u64 *array, 2288 u64 type __maybe_unused) 2289{ 2290 data->weight = *array; 2291} 2292 2293u64 evsel__bitfield_swap_branch_flags(u64 value) 2294{ 2295 u64 new_val = 0; 2296 2297 /* 2298 * branch_flags 2299 * union { 2300 * u64 values; 2301 * struct { 2302 * mispred:1 //target mispredicted 2303 * predicted:1 //target predicted 2304 * in_tx:1 //in transaction 2305 * abort:1 //transaction abort 2306 * cycles:16 //cycle count to last branch 2307 * type:4 //branch type 2308 * spec:2 //branch speculation info 2309 * new_type:4 //additional branch type 2310 * priv:3 //privilege level 2311 * reserved:31 2312 * } 2313 * } 2314 * 2315 * Avoid bswap64() the entire branch_flag.value, 2316 * as it has variable bit-field sizes. Instead the 2317 * macro takes the bit-field position/size, 2318 * swaps it based on the host endianness. 2319 */ 2320 if (host_is_bigendian()) { 2321 new_val = bitfield_swap(value, 0, 1); 2322 new_val |= bitfield_swap(value, 1, 1); 2323 new_val |= bitfield_swap(value, 2, 1); 2324 new_val |= bitfield_swap(value, 3, 1); 2325 new_val |= bitfield_swap(value, 4, 16); 2326 new_val |= bitfield_swap(value, 20, 4); 2327 new_val |= bitfield_swap(value, 24, 2); 2328 new_val |= bitfield_swap(value, 26, 4); 2329 new_val |= bitfield_swap(value, 30, 3); 2330 new_val |= bitfield_swap(value, 33, 31); 2331 } else { 2332 new_val = bitfield_swap(value, 63, 1); 2333 new_val |= bitfield_swap(value, 62, 1); 2334 new_val |= bitfield_swap(value, 61, 1); 2335 new_val |= bitfield_swap(value, 60, 1); 2336 new_val |= bitfield_swap(value, 44, 16); 2337 new_val |= bitfield_swap(value, 40, 4); 2338 new_val |= bitfield_swap(value, 38, 2); 2339 new_val |= bitfield_swap(value, 34, 4); 2340 new_val |= bitfield_swap(value, 31, 3); 2341 new_val |= bitfield_swap(value, 0, 31); 2342 } 2343 2344 return new_val; 2345} 2346 2347int evsel__parse_sample(struct evsel *evsel, union perf_event *event, 2348 struct perf_sample *data) 2349{ 2350 u64 type = evsel->core.attr.sample_type; 2351 bool swapped = evsel->needs_swap; 2352 const __u64 *array; 2353 u16 max_size = event->header.size; 2354 const void *endp = (void *)event + max_size; 2355 u64 sz; 2356 2357 /* 2358 * used for cross-endian analysis. See git commit 65014ab3 2359 * for why this goofiness is needed. 2360 */ 2361 union u64_swap u; 2362 2363 memset(data, 0, sizeof(*data)); 2364 data->cpu = data->pid = data->tid = -1; 2365 data->stream_id = data->id = data->time = -1ULL; 2366 data->period = evsel->core.attr.sample_period; 2367 data->cpumode = event->header.misc & PERF_RECORD_MISC_CPUMODE_MASK; 2368 data->misc = event->header.misc; 2369 data->data_src = PERF_MEM_DATA_SRC_NONE; 2370 data->vcpu = -1; 2371 2372 if (event->header.type != PERF_RECORD_SAMPLE) { 2373 if (!evsel->core.attr.sample_id_all) 2374 return 0; 2375 return perf_evsel__parse_id_sample(evsel, event, data); 2376 } 2377 2378 array = event->sample.array; 2379 2380 if (perf_event__check_size(event, evsel->sample_size)) 2381 return -EFAULT; 2382 2383 if (type & PERF_SAMPLE_IDENTIFIER) { 2384 data->id = *array; 2385 array++; 2386 } 2387 2388 if (type & PERF_SAMPLE_IP) { 2389 data->ip = *array; 2390 array++; 2391 } 2392 2393 if (type & PERF_SAMPLE_TID) { 2394 u.val64 = *array; 2395 if (swapped) { 2396 /* undo swap of u64, then swap on individual u32s */ 2397 u.val64 = bswap_64(u.val64); 2398 u.val32[0] = bswap_32(u.val32[0]); 2399 u.val32[1] = bswap_32(u.val32[1]); 2400 } 2401 2402 data->pid = u.val32[0]; 2403 data->tid = u.val32[1]; 2404 array++; 2405 } 2406 2407 if (type & PERF_SAMPLE_TIME) { 2408 data->time = *array; 2409 array++; 2410 } 2411 2412 if (type & PERF_SAMPLE_ADDR) { 2413 data->addr = *array; 2414 array++; 2415 } 2416 2417 if (type & PERF_SAMPLE_ID) { 2418 data->id = *array; 2419 array++; 2420 } 2421 2422 if (type & PERF_SAMPLE_STREAM_ID) { 2423 data->stream_id = *array; 2424 array++; 2425 } 2426 2427 if (type & PERF_SAMPLE_CPU) { 2428 2429 u.val64 = *array; 2430 if (swapped) { 2431 /* undo swap of u64, then swap on individual u32s */ 2432 u.val64 = bswap_64(u.val64); 2433 u.val32[0] = bswap_32(u.val32[0]); 2434 } 2435 2436 data->cpu = u.val32[0]; 2437 array++; 2438 } 2439 2440 if (type & PERF_SAMPLE_PERIOD) { 2441 data->period = *array; 2442 array++; 2443 } 2444 2445 if (type & PERF_SAMPLE_READ) { 2446 u64 read_format = evsel->core.attr.read_format; 2447 2448 OVERFLOW_CHECK_u64(array); 2449 if (read_format & PERF_FORMAT_GROUP) 2450 data->read.group.nr = *array; 2451 else 2452 data->read.one.value = *array; 2453 2454 array++; 2455 2456 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) { 2457 OVERFLOW_CHECK_u64(array); 2458 data->read.time_enabled = *array; 2459 array++; 2460 } 2461 2462 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) { 2463 OVERFLOW_CHECK_u64(array); 2464 data->read.time_running = *array; 2465 array++; 2466 } 2467 2468 /* PERF_FORMAT_ID is forced for PERF_SAMPLE_READ */ 2469 if (read_format & PERF_FORMAT_GROUP) { 2470 const u64 max_group_nr = UINT64_MAX / 2471 sizeof(struct sample_read_value); 2472 2473 if (data->read.group.nr > max_group_nr) 2474 return -EFAULT; 2475 2476 sz = data->read.group.nr * sample_read_value_size(read_format); 2477 OVERFLOW_CHECK(array, sz, max_size); 2478 data->read.group.values = 2479 (struct sample_read_value *)array; 2480 array = (void *)array + sz; 2481 } else { 2482 OVERFLOW_CHECK_u64(array); 2483 data->read.one.id = *array; 2484 array++; 2485 2486 if (read_format & PERF_FORMAT_LOST) { 2487 OVERFLOW_CHECK_u64(array); 2488 data->read.one.lost = *array; 2489 array++; 2490 } 2491 } 2492 } 2493 2494 if (type & PERF_SAMPLE_CALLCHAIN) { 2495 const u64 max_callchain_nr = UINT64_MAX / sizeof(u64); 2496 2497 OVERFLOW_CHECK_u64(array); 2498 data->callchain = (struct ip_callchain *)array++; 2499 if (data->callchain->nr > max_callchain_nr) 2500 return -EFAULT; 2501 sz = data->callchain->nr * sizeof(u64); 2502 OVERFLOW_CHECK(array, sz, max_size); 2503 array = (void *)array + sz; 2504 } 2505 2506 if (type & PERF_SAMPLE_RAW) { 2507 OVERFLOW_CHECK_u64(array); 2508 u.val64 = *array; 2509 2510 /* 2511 * Undo swap of u64, then swap on individual u32s, 2512 * get the size of the raw area and undo all of the 2513 * swap. The pevent interface handles endianness by 2514 * itself. 2515 */ 2516 if (swapped) { 2517 u.val64 = bswap_64(u.val64); 2518 u.val32[0] = bswap_32(u.val32[0]); 2519 u.val32[1] = bswap_32(u.val32[1]); 2520 } 2521 data->raw_size = u.val32[0]; 2522 2523 /* 2524 * The raw data is aligned on 64bits including the 2525 * u32 size, so it's safe to use mem_bswap_64. 2526 */ 2527 if (swapped) 2528 mem_bswap_64((void *) array, data->raw_size); 2529 2530 array = (void *)array + sizeof(u32); 2531 2532 OVERFLOW_CHECK(array, data->raw_size, max_size); 2533 data->raw_data = (void *)array; 2534 array = (void *)array + data->raw_size; 2535 } 2536 2537 if (type & PERF_SAMPLE_BRANCH_STACK) { 2538 const u64 max_branch_nr = UINT64_MAX / 2539 sizeof(struct branch_entry); 2540 struct branch_entry *e; 2541 unsigned int i; 2542 2543 OVERFLOW_CHECK_u64(array); 2544 data->branch_stack = (struct branch_stack *)array++; 2545 2546 if (data->branch_stack->nr > max_branch_nr) 2547 return -EFAULT; 2548 2549 sz = data->branch_stack->nr * sizeof(struct branch_entry); 2550 if (evsel__has_branch_hw_idx(evsel)) { 2551 sz += sizeof(u64); 2552 e = &data->branch_stack->entries[0]; 2553 } else { 2554 data->no_hw_idx = true; 2555 /* 2556 * if the PERF_SAMPLE_BRANCH_HW_INDEX is not applied, 2557 * only nr and entries[] will be output by kernel. 2558 */ 2559 e = (struct branch_entry *)&data->branch_stack->hw_idx; 2560 } 2561 2562 if (swapped) { 2563 /* 2564 * struct branch_flag does not have endian 2565 * specific bit field definition. And bswap 2566 * will not resolve the issue, since these 2567 * are bit fields. 2568 * 2569 * evsel__bitfield_swap_branch_flags() uses a 2570 * bitfield_swap macro to swap the bit position 2571 * based on the host endians. 2572 */ 2573 for (i = 0; i < data->branch_stack->nr; i++, e++) 2574 e->flags.value = evsel__bitfield_swap_branch_flags(e->flags.value); 2575 } 2576 2577 OVERFLOW_CHECK(array, sz, max_size); 2578 array = (void *)array + sz; 2579 } 2580 2581 if (type & PERF_SAMPLE_REGS_USER) { 2582 OVERFLOW_CHECK_u64(array); 2583 data->user_regs.abi = *array; 2584 array++; 2585 2586 if (data->user_regs.abi) { 2587 u64 mask = evsel->core.attr.sample_regs_user; 2588 2589 sz = hweight64(mask) * sizeof(u64); 2590 OVERFLOW_CHECK(array, sz, max_size); 2591 data->user_regs.mask = mask; 2592 data->user_regs.regs = (u64 *)array; 2593 array = (void *)array + sz; 2594 } 2595 } 2596 2597 if (type & PERF_SAMPLE_STACK_USER) { 2598 OVERFLOW_CHECK_u64(array); 2599 sz = *array++; 2600 2601 data->user_stack.offset = ((char *)(array - 1) 2602 - (char *) event); 2603 2604 if (!sz) { 2605 data->user_stack.size = 0; 2606 } else { 2607 OVERFLOW_CHECK(array, sz, max_size); 2608 data->user_stack.data = (char *)array; 2609 array = (void *)array + sz; 2610 OVERFLOW_CHECK_u64(array); 2611 data->user_stack.size = *array++; 2612 if (WARN_ONCE(data->user_stack.size > sz, 2613 "user stack dump failure\n")) 2614 return -EFAULT; 2615 } 2616 } 2617 2618 if (type & PERF_SAMPLE_WEIGHT_TYPE) { 2619 OVERFLOW_CHECK_u64(array); 2620 arch_perf_parse_sample_weight(data, array, type); 2621 array++; 2622 } 2623 2624 if (type & PERF_SAMPLE_DATA_SRC) { 2625 OVERFLOW_CHECK_u64(array); 2626 data->data_src = *array; 2627 array++; 2628 } 2629 2630 if (type & PERF_SAMPLE_TRANSACTION) { 2631 OVERFLOW_CHECK_u64(array); 2632 data->transaction = *array; 2633 array++; 2634 } 2635 2636 data->intr_regs.abi = PERF_SAMPLE_REGS_ABI_NONE; 2637 if (type & PERF_SAMPLE_REGS_INTR) { 2638 OVERFLOW_CHECK_u64(array); 2639 data->intr_regs.abi = *array; 2640 array++; 2641 2642 if (data->intr_regs.abi != PERF_SAMPLE_REGS_ABI_NONE) { 2643 u64 mask = evsel->core.attr.sample_regs_intr; 2644 2645 sz = hweight64(mask) * sizeof(u64); 2646 OVERFLOW_CHECK(array, sz, max_size); 2647 data->intr_regs.mask = mask; 2648 data->intr_regs.regs = (u64 *)array; 2649 array = (void *)array + sz; 2650 } 2651 } 2652 2653 data->phys_addr = 0; 2654 if (type & PERF_SAMPLE_PHYS_ADDR) { 2655 data->phys_addr = *array; 2656 array++; 2657 } 2658 2659 data->cgroup = 0; 2660 if (type & PERF_SAMPLE_CGROUP) { 2661 data->cgroup = *array; 2662 array++; 2663 } 2664 2665 data->data_page_size = 0; 2666 if (type & PERF_SAMPLE_DATA_PAGE_SIZE) { 2667 data->data_page_size = *array; 2668 array++; 2669 } 2670 2671 data->code_page_size = 0; 2672 if (type & PERF_SAMPLE_CODE_PAGE_SIZE) { 2673 data->code_page_size = *array; 2674 array++; 2675 } 2676 2677 if (type & PERF_SAMPLE_AUX) { 2678 OVERFLOW_CHECK_u64(array); 2679 sz = *array++; 2680 2681 OVERFLOW_CHECK(array, sz, max_size); 2682 /* Undo swap of data */ 2683 if (swapped) 2684 mem_bswap_64((char *)array, sz); 2685 data->aux_sample.size = sz; 2686 data->aux_sample.data = (char *)array; 2687 array = (void *)array + sz; 2688 } 2689 2690 return 0; 2691} 2692 2693int evsel__parse_sample_timestamp(struct evsel *evsel, union perf_event *event, 2694 u64 *timestamp) 2695{ 2696 u64 type = evsel->core.attr.sample_type; 2697 const __u64 *array; 2698 2699 if (!(type & PERF_SAMPLE_TIME)) 2700 return -1; 2701 2702 if (event->header.type != PERF_RECORD_SAMPLE) { 2703 struct perf_sample data = { 2704 .time = -1ULL, 2705 }; 2706 2707 if (!evsel->core.attr.sample_id_all) 2708 return -1; 2709 if (perf_evsel__parse_id_sample(evsel, event, &data)) 2710 return -1; 2711 2712 *timestamp = data.time; 2713 return 0; 2714 } 2715 2716 array = event->sample.array; 2717 2718 if (perf_event__check_size(event, evsel->sample_size)) 2719 return -EFAULT; 2720 2721 if (type & PERF_SAMPLE_IDENTIFIER) 2722 array++; 2723 2724 if (type & PERF_SAMPLE_IP) 2725 array++; 2726 2727 if (type & PERF_SAMPLE_TID) 2728 array++; 2729 2730 if (type & PERF_SAMPLE_TIME) 2731 *timestamp = *array; 2732 2733 return 0; 2734} 2735 2736u16 evsel__id_hdr_size(struct evsel *evsel) 2737{ 2738 u64 sample_type = evsel->core.attr.sample_type; 2739 u16 size = 0; 2740 2741 if (sample_type & PERF_SAMPLE_TID) 2742 size += sizeof(u64); 2743 2744 if (sample_type & PERF_SAMPLE_TIME) 2745 size += sizeof(u64); 2746 2747 if (sample_type & PERF_SAMPLE_ID) 2748 size += sizeof(u64); 2749 2750 if (sample_type & PERF_SAMPLE_STREAM_ID) 2751 size += sizeof(u64); 2752 2753 if (sample_type & PERF_SAMPLE_CPU) 2754 size += sizeof(u64); 2755 2756 if (sample_type & PERF_SAMPLE_IDENTIFIER) 2757 size += sizeof(u64); 2758 2759 return size; 2760} 2761 2762#ifdef HAVE_LIBTRACEEVENT 2763struct tep_format_field *evsel__field(struct evsel *evsel, const char *name) 2764{ 2765 return tep_find_field(evsel->tp_format, name); 2766} 2767 2768void *evsel__rawptr(struct evsel *evsel, struct perf_sample *sample, const char *name) 2769{ 2770 struct tep_format_field *field = evsel__field(evsel, name); 2771 int offset; 2772 2773 if (!field) 2774 return NULL; 2775 2776 offset = field->offset; 2777 2778 if (field->flags & TEP_FIELD_IS_DYNAMIC) { 2779 offset = *(int *)(sample->raw_data + field->offset); 2780 offset &= 0xffff; 2781 if (tep_field_is_relative(field->flags)) 2782 offset += field->offset + field->size; 2783 } 2784 2785 return sample->raw_data + offset; 2786} 2787 2788u64 format_field__intval(struct tep_format_field *field, struct perf_sample *sample, 2789 bool needs_swap) 2790{ 2791 u64 value; 2792 void *ptr = sample->raw_data + field->offset; 2793 2794 switch (field->size) { 2795 case 1: 2796 return *(u8 *)ptr; 2797 case 2: 2798 value = *(u16 *)ptr; 2799 break; 2800 case 4: 2801 value = *(u32 *)ptr; 2802 break; 2803 case 8: 2804 memcpy(&value, ptr, sizeof(u64)); 2805 break; 2806 default: 2807 return 0; 2808 } 2809 2810 if (!needs_swap) 2811 return value; 2812 2813 switch (field->size) { 2814 case 2: 2815 return bswap_16(value); 2816 case 4: 2817 return bswap_32(value); 2818 case 8: 2819 return bswap_64(value); 2820 default: 2821 return 0; 2822 } 2823 2824 return 0; 2825} 2826 2827u64 evsel__intval(struct evsel *evsel, struct perf_sample *sample, const char *name) 2828{ 2829 struct tep_format_field *field = evsel__field(evsel, name); 2830 2831 return field ? format_field__intval(field, sample, evsel->needs_swap) : 0; 2832} 2833#endif 2834 2835bool evsel__fallback(struct evsel *evsel, int err, char *msg, size_t msgsize) 2836{ 2837 int paranoid; 2838 2839 if ((err == ENOENT || err == ENXIO || err == ENODEV) && 2840 evsel->core.attr.type == PERF_TYPE_HARDWARE && 2841 evsel->core.attr.config == PERF_COUNT_HW_CPU_CYCLES) { 2842 /* 2843 * If it's cycles then fall back to hrtimer based 2844 * cpu-clock-tick sw counter, which is always available even if 2845 * no PMU support. 2846 * 2847 * PPC returns ENXIO until 2.6.37 (behavior changed with commit 2848 * b0a873e). 2849 */ 2850 scnprintf(msg, msgsize, "%s", 2851"The cycles event is not supported, trying to fall back to cpu-clock-ticks"); 2852 2853 evsel->core.attr.type = PERF_TYPE_SOFTWARE; 2854 evsel->core.attr.config = PERF_COUNT_SW_CPU_CLOCK; 2855 2856 zfree(&evsel->name); 2857 return true; 2858 } else if (err == EACCES && !evsel->core.attr.exclude_kernel && 2859 (paranoid = perf_event_paranoid()) > 1) { 2860 const char *name = evsel__name(evsel); 2861 char *new_name; 2862 const char *sep = ":"; 2863 2864 /* If event has exclude user then don't exclude kernel. */ 2865 if (evsel->core.attr.exclude_user) 2866 return false; 2867 2868 /* Is there already the separator in the name. */ 2869 if (strchr(name, '/') || 2870 (strchr(name, ':') && !evsel->is_libpfm_event)) 2871 sep = ""; 2872 2873 if (asprintf(&new_name, "%s%su", name, sep) < 0) 2874 return false; 2875 2876 free(evsel->name); 2877 evsel->name = new_name; 2878 scnprintf(msg, msgsize, "kernel.perf_event_paranoid=%d, trying " 2879 "to fall back to excluding kernel and hypervisor " 2880 " samples", paranoid); 2881 evsel->core.attr.exclude_kernel = 1; 2882 evsel->core.attr.exclude_hv = 1; 2883 2884 return true; 2885 } 2886 2887 return false; 2888} 2889 2890static bool find_process(const char *name) 2891{ 2892 size_t len = strlen(name); 2893 DIR *dir; 2894 struct dirent *d; 2895 int ret = -1; 2896 2897 dir = opendir(procfs__mountpoint()); 2898 if (!dir) 2899 return false; 2900 2901 /* Walk through the directory. */ 2902 while (ret && (d = readdir(dir)) != NULL) { 2903 char path[PATH_MAX]; 2904 char *data; 2905 size_t size; 2906 2907 if ((d->d_type != DT_DIR) || 2908 !strcmp(".", d->d_name) || 2909 !strcmp("..", d->d_name)) 2910 continue; 2911 2912 scnprintf(path, sizeof(path), "%s/%s/comm", 2913 procfs__mountpoint(), d->d_name); 2914 2915 if (filename__read_str(path, &data, &size)) 2916 continue; 2917 2918 ret = strncmp(name, data, len); 2919 free(data); 2920 } 2921 2922 closedir(dir); 2923 return ret ? false : true; 2924} 2925 2926int __weak arch_evsel__open_strerror(struct evsel *evsel __maybe_unused, 2927 char *msg __maybe_unused, 2928 size_t size __maybe_unused) 2929{ 2930 return 0; 2931} 2932 2933int evsel__open_strerror(struct evsel *evsel, struct target *target, 2934 int err, char *msg, size_t size) 2935{ 2936 char sbuf[STRERR_BUFSIZE]; 2937 int printed = 0, enforced = 0; 2938 int ret; 2939 2940 switch (err) { 2941 case EPERM: 2942 case EACCES: 2943 printed += scnprintf(msg + printed, size - printed, 2944 "Access to performance monitoring and observability operations is limited.\n"); 2945 2946 if (!sysfs__read_int("fs/selinux/enforce", &enforced)) { 2947 if (enforced) { 2948 printed += scnprintf(msg + printed, size - printed, 2949 "Enforced MAC policy settings (SELinux) can limit access to performance\n" 2950 "monitoring and observability operations. Inspect system audit records for\n" 2951 "more perf_event access control information and adjusting the policy.\n"); 2952 } 2953 } 2954 2955 if (err == EPERM) 2956 printed += scnprintf(msg, size, 2957 "No permission to enable %s event.\n\n", evsel__name(evsel)); 2958 2959 return scnprintf(msg + printed, size - printed, 2960 "Consider adjusting /proc/sys/kernel/perf_event_paranoid setting to open\n" 2961 "access to performance monitoring and observability operations for processes\n" 2962 "without CAP_PERFMON, CAP_SYS_PTRACE or CAP_SYS_ADMIN Linux capability.\n" 2963 "More information can be found at 'Perf events and tool security' document:\n" 2964 "https://www.kernel.org/doc/html/latest/admin-guide/perf-security.html\n" 2965 "perf_event_paranoid setting is %d:\n" 2966 " -1: Allow use of (almost) all events by all users\n" 2967 " Ignore mlock limit after perf_event_mlock_kb without CAP_IPC_LOCK\n" 2968 ">= 0: Disallow raw and ftrace function tracepoint access\n" 2969 ">= 1: Disallow CPU event access\n" 2970 ">= 2: Disallow kernel profiling\n" 2971 "To make the adjusted perf_event_paranoid setting permanent preserve it\n" 2972 "in /etc/sysctl.conf (e.g. kernel.perf_event_paranoid = <setting>)", 2973 perf_event_paranoid()); 2974 case ENOENT: 2975 return scnprintf(msg, size, "The %s event is not supported.", evsel__name(evsel)); 2976 case EMFILE: 2977 return scnprintf(msg, size, "%s", 2978 "Too many events are opened.\n" 2979 "Probably the maximum number of open file descriptors has been reached.\n" 2980 "Hint: Try again after reducing the number of events.\n" 2981 "Hint: Try increasing the limit with 'ulimit -n <limit>'"); 2982 case ENOMEM: 2983 if (evsel__has_callchain(evsel) && 2984 access("/proc/sys/kernel/perf_event_max_stack", F_OK) == 0) 2985 return scnprintf(msg, size, 2986 "Not enough memory to setup event with callchain.\n" 2987 "Hint: Try tweaking /proc/sys/kernel/perf_event_max_stack\n" 2988 "Hint: Current value: %d", sysctl__max_stack()); 2989 break; 2990 case ENODEV: 2991 if (target->cpu_list) 2992 return scnprintf(msg, size, "%s", 2993 "No such device - did you specify an out-of-range profile CPU?"); 2994 break; 2995 case EOPNOTSUPP: 2996 if (evsel->core.attr.sample_type & PERF_SAMPLE_BRANCH_STACK) 2997 return scnprintf(msg, size, 2998 "%s: PMU Hardware or event type doesn't support branch stack sampling.", 2999 evsel__name(evsel)); 3000 if (evsel->core.attr.aux_output) 3001 return scnprintf(msg, size, 3002 "%s: PMU Hardware doesn't support 'aux_output' feature", 3003 evsel__name(evsel)); 3004 if (evsel->core.attr.sample_period != 0) 3005 return scnprintf(msg, size, 3006 "%s: PMU Hardware doesn't support sampling/overflow-interrupts. Try 'perf stat'", 3007 evsel__name(evsel)); 3008 if (evsel->core.attr.precise_ip) 3009 return scnprintf(msg, size, "%s", 3010 "\'precise\' request may not be supported. Try removing 'p' modifier."); 3011#if defined(__i386__) || defined(__x86_64__) 3012 if (evsel->core.attr.type == PERF_TYPE_HARDWARE) 3013 return scnprintf(msg, size, "%s", 3014 "No hardware sampling interrupt available.\n"); 3015#endif 3016 break; 3017 case EBUSY: 3018 if (find_process("oprofiled")) 3019 return scnprintf(msg, size, 3020 "The PMU counters are busy/taken by another profiler.\n" 3021 "We found oprofile daemon running, please stop it and try again."); 3022 break; 3023 case EINVAL: 3024 if (evsel->core.attr.sample_type & PERF_SAMPLE_CODE_PAGE_SIZE && perf_missing_features.code_page_size) 3025 return scnprintf(msg, size, "Asking for the code page size isn't supported by this kernel."); 3026 if (evsel->core.attr.sample_type & PERF_SAMPLE_DATA_PAGE_SIZE && perf_missing_features.data_page_size) 3027 return scnprintf(msg, size, "Asking for the data page size isn't supported by this kernel."); 3028 if (evsel->core.attr.write_backward && perf_missing_features.write_backward) 3029 return scnprintf(msg, size, "Reading from overwrite event is not supported by this kernel."); 3030 if (perf_missing_features.clockid) 3031 return scnprintf(msg, size, "clockid feature not supported."); 3032 if (perf_missing_features.clockid_wrong) 3033 return scnprintf(msg, size, "wrong clockid (%d).", clockid); 3034 if (perf_missing_features.aux_output) 3035 return scnprintf(msg, size, "The 'aux_output' feature is not supported, update the kernel."); 3036 if (!target__has_cpu(target)) 3037 return scnprintf(msg, size, 3038 "Invalid event (%s) in per-thread mode, enable system wide with '-a'.", 3039 evsel__name(evsel)); 3040 3041 break; 3042 case ENODATA: 3043 return scnprintf(msg, size, "Cannot collect data source with the load latency event alone. " 3044 "Please add an auxiliary event in front of the load latency event."); 3045 default: 3046 break; 3047 } 3048 3049 ret = arch_evsel__open_strerror(evsel, msg, size); 3050 if (ret) 3051 return ret; 3052 3053 return scnprintf(msg, size, 3054 "The sys_perf_event_open() syscall returned with %d (%s) for event (%s).\n" 3055 "/bin/dmesg | grep -i perf may provide additional information.\n", 3056 err, str_error_r(err, sbuf, sizeof(sbuf)), evsel__name(evsel)); 3057} 3058 3059struct perf_env *evsel__env(struct evsel *evsel) 3060{ 3061 if (evsel && evsel->evlist && evsel->evlist->env) 3062 return evsel->evlist->env; 3063 return &perf_env; 3064} 3065 3066static int store_evsel_ids(struct evsel *evsel, struct evlist *evlist) 3067{ 3068 int cpu_map_idx, thread; 3069 3070 for (cpu_map_idx = 0; cpu_map_idx < xyarray__max_x(evsel->core.fd); cpu_map_idx++) { 3071 for (thread = 0; thread < xyarray__max_y(evsel->core.fd); 3072 thread++) { 3073 int fd = FD(evsel, cpu_map_idx, thread); 3074 3075 if (perf_evlist__id_add_fd(&evlist->core, &evsel->core, 3076 cpu_map_idx, thread, fd) < 0) 3077 return -1; 3078 } 3079 } 3080 3081 return 0; 3082} 3083 3084int evsel__store_ids(struct evsel *evsel, struct evlist *evlist) 3085{ 3086 struct perf_cpu_map *cpus = evsel->core.cpus; 3087 struct perf_thread_map *threads = evsel->core.threads; 3088 3089 if (perf_evsel__alloc_id(&evsel->core, perf_cpu_map__nr(cpus), threads->nr)) 3090 return -ENOMEM; 3091 3092 return store_evsel_ids(evsel, evlist); 3093} 3094 3095void evsel__zero_per_pkg(struct evsel *evsel) 3096{ 3097 struct hashmap_entry *cur; 3098 size_t bkt; 3099 3100 if (evsel->per_pkg_mask) { 3101 hashmap__for_each_entry(evsel->per_pkg_mask, cur, bkt) 3102 zfree(&cur->pkey); 3103 3104 hashmap__clear(evsel->per_pkg_mask); 3105 } 3106} 3107 3108/** 3109 * evsel__is_hybrid - does the evsel have a known PMU that is hybrid. Note, this 3110 * will be false on hybrid systems for hardware and legacy 3111 * cache events. 3112 */ 3113bool evsel__is_hybrid(const struct evsel *evsel) 3114{ 3115 if (perf_pmus__num_core_pmus() == 1) 3116 return false; 3117 3118 return evsel->core.is_pmu_core; 3119} 3120 3121struct evsel *evsel__leader(const struct evsel *evsel) 3122{ 3123 return container_of(evsel->core.leader, struct evsel, core); 3124} 3125 3126bool evsel__has_leader(struct evsel *evsel, struct evsel *leader) 3127{ 3128 return evsel->core.leader == &leader->core; 3129} 3130 3131bool evsel__is_leader(struct evsel *evsel) 3132{ 3133 return evsel__has_leader(evsel, evsel); 3134} 3135 3136void evsel__set_leader(struct evsel *evsel, struct evsel *leader) 3137{ 3138 evsel->core.leader = &leader->core; 3139} 3140 3141int evsel__source_count(const struct evsel *evsel) 3142{ 3143 struct evsel *pos; 3144 int count = 0; 3145 3146 evlist__for_each_entry(evsel->evlist, pos) { 3147 if (pos->metric_leader == evsel) 3148 count++; 3149 } 3150 return count; 3151} 3152 3153bool __weak arch_evsel__must_be_in_group(const struct evsel *evsel __maybe_unused) 3154{ 3155 return false; 3156} 3157 3158/* 3159 * Remove an event from a given group (leader). 3160 * Some events, e.g., perf metrics Topdown events, 3161 * must always be grouped. Ignore the events. 3162 */ 3163void evsel__remove_from_group(struct evsel *evsel, struct evsel *leader) 3164{ 3165 if (!arch_evsel__must_be_in_group(evsel) && evsel != leader) { 3166 evsel__set_leader(evsel, evsel); 3167 evsel->core.nr_members = 0; 3168 leader->core.nr_members--; 3169 } 3170} 3171