1 // SPDX-License-Identifier: GPL-2.0 2 #include <errno.h> 3 #include <signal.h> 4 #include <inttypes.h> 5 #include <linux/err.h> 6 #include <linux/kernel.h> 7 #include <linux/zalloc.h> 8 #include <api/fs/fs.h> 9 10 #include <byteswap.h> 11 #include <unistd.h> 12 #include <sys/types.h> 13 #include <sys/mman.h> 14 #include <perf/cpumap.h> 15 #include <perf/event.h> 16 17 #include "map_symbol.h" 18 #include "branch.h" 19 #include "debug.h" 20 #include "env.h" 21 #include "evlist.h" 22 #include "evsel.h" 23 #include "memswap.h" 24 #include "map.h" 25 #include "symbol.h" 26 #include "session.h" 27 #include "tool.h" 28 #include "perf_regs.h" 29 #include "asm/bug.h" 30 #include "auxtrace.h" 31 #include "thread.h" 32 #include "thread-stack.h" 33 #include "sample-raw.h" 34 #include "stat.h" 35 #include "tsc.h" 36 #include "ui/progress.h" 37 #include "util.h" 38 #include "arch/common.h" 39 #include "units.h" 40 #include "annotate.h" 41 #include "perf.h" 42 #include <internal/lib.h> 43 44 static int perf_session__deliver_event(struct perf_session *session, 45 union perf_event *event, 46 const struct perf_tool *tool, 47 u64 file_offset, 48 const char *file_path); 49 50 static int perf_session__open(struct perf_session *session) 51 { 52 struct perf_data *data = session->data; 53 54 if (perf_session__read_header(session) < 0) { 55 pr_err("incompatible file format (rerun with -v to learn more)\n"); 56 return -1; 57 } 58 59 if (perf_header__has_feat(&session->header, HEADER_AUXTRACE)) { 60 /* Auxiliary events may reference exited threads, hold onto dead ones. */ 61 symbol_conf.keep_exited_threads = true; 62 } 63 64 if (perf_data__is_pipe(data)) 65 return 0; 66 67 if (perf_header__has_feat(&session->header, HEADER_STAT)) 68 return 0; 69 70 if (!evlist__valid_sample_type(session->evlist)) { 71 pr_err("non matching sample_type\n"); 72 return -1; 73 } 74 75 if (!evlist__valid_sample_id_all(session->evlist)) { 76 pr_err("non matching sample_id_all\n"); 77 return -1; 78 } 79 80 if (!evlist__valid_read_format(session->evlist)) { 81 pr_err("non matching read_format\n"); 82 return -1; 83 } 84 85 return 0; 86 } 87 88 void perf_session__set_id_hdr_size(struct perf_session *session) 89 { 90 u16 id_hdr_size = evlist__id_hdr_size(session->evlist); 91 92 machines__set_id_hdr_size(&session->machines, id_hdr_size); 93 } 94 95 int perf_session__create_kernel_maps(struct perf_session *session) 96 { 97 int ret = machine__create_kernel_maps(&session->machines.host); 98 99 if (ret >= 0) 100 ret = machines__create_guest_kernel_maps(&session->machines); 101 return ret; 102 } 103 104 static void perf_session__destroy_kernel_maps(struct perf_session *session) 105 { 106 machines__destroy_kernel_maps(&session->machines); 107 } 108 109 static bool perf_session__has_comm_exec(struct perf_session *session) 110 { 111 struct evsel *evsel; 112 113 evlist__for_each_entry(session->evlist, evsel) { 114 if (evsel->core.attr.comm_exec) 115 return true; 116 } 117 118 return false; 119 } 120 121 static void perf_session__set_comm_exec(struct perf_session *session) 122 { 123 bool comm_exec = perf_session__has_comm_exec(session); 124 125 machines__set_comm_exec(&session->machines, comm_exec); 126 } 127 128 static int ordered_events__deliver_event(struct ordered_events *oe, 129 struct ordered_event *event) 130 { 131 struct perf_session *session = container_of(oe, struct perf_session, 132 ordered_events); 133 134 return perf_session__deliver_event(session, event->event, 135 session->tool, event->file_offset, 136 event->file_path); 137 } 138 139 struct perf_session *__perf_session__new(struct perf_data *data, 140 struct perf_tool *tool, 141 bool trace_event_repipe, 142 struct perf_env *host_env) 143 { 144 int ret = -ENOMEM; 145 struct perf_session *session = zalloc(sizeof(*session)); 146 147 if (!session) 148 goto out; 149 150 session->trace_event_repipe = trace_event_repipe; 151 session->tool = tool; 152 session->decomp_data.zstd_decomp = &session->zstd_data; 153 session->active_decomp = &session->decomp_data; 154 INIT_LIST_HEAD(&session->auxtrace_index); 155 machines__init(&session->machines); 156 ordered_events__init(&session->ordered_events, 157 ordered_events__deliver_event, NULL); 158 159 perf_env__init(&session->header.env); 160 if (data) { 161 ret = perf_data__open(data); 162 if (ret < 0) 163 goto out_delete; 164 165 session->data = data; 166 167 if (perf_data__is_read(data)) { 168 ret = perf_session__open(session); 169 if (ret < 0) 170 goto out_delete; 171 172 /* 173 * set session attributes that are present in perf.data 174 * but not in pipe-mode. 175 */ 176 if (!data->is_pipe) { 177 perf_session__set_id_hdr_size(session); 178 perf_session__set_comm_exec(session); 179 } 180 181 evlist__init_trace_event_sample_raw(session->evlist, &session->header.env); 182 183 /* Open the directory data. */ 184 if (data->is_dir) { 185 ret = perf_data__open_dir(data); 186 if (ret) 187 goto out_delete; 188 } 189 190 if (!symbol_conf.kallsyms_name && 191 !symbol_conf.vmlinux_name) 192 symbol_conf.kallsyms_name = perf_data__kallsyms_name(data); 193 } 194 } else { 195 assert(host_env != NULL); 196 session->machines.host.env = host_env; 197 } 198 if (session->evlist) 199 session->evlist->session = session; 200 201 session->machines.host.single_address_space = 202 perf_env__single_address_space(session->machines.host.env); 203 204 if (!data || perf_data__is_write(data)) { 205 /* 206 * In O_RDONLY mode this will be performed when reading the 207 * kernel MMAP event, in perf_event__process_mmap(). 208 */ 209 if (perf_session__create_kernel_maps(session) < 0) 210 pr_warning("Cannot read kernel map\n"); 211 } 212 213 /* 214 * In pipe-mode, evlist is empty until PERF_RECORD_HEADER_ATTR is 215 * processed, so evlist__sample_id_all is not meaningful here. 216 */ 217 if ((!data || !data->is_pipe) && tool && tool->ordering_requires_timestamps && 218 tool->ordered_events && !evlist__sample_id_all(session->evlist)) { 219 dump_printf("WARNING: No sample_id_all support, falling back to unordered processing\n"); 220 tool->ordered_events = false; 221 } 222 223 return session; 224 225 out_delete: 226 perf_session__delete(session); 227 out: 228 return ERR_PTR(ret); 229 } 230 231 static void perf_decomp__release_events(struct decomp *next) 232 { 233 struct decomp *decomp; 234 size_t mmap_len; 235 236 do { 237 decomp = next; 238 if (decomp == NULL) 239 break; 240 next = decomp->next; 241 mmap_len = decomp->mmap_len; 242 munmap(decomp, mmap_len); 243 } while (1); 244 } 245 246 void perf_session__delete(struct perf_session *session) 247 { 248 if (session == NULL) 249 return; 250 auxtrace__free(session); 251 auxtrace_index__free(&session->auxtrace_index); 252 debuginfo_cache__delete(); 253 perf_session__destroy_kernel_maps(session); 254 perf_decomp__release_events(session->decomp_data.decomp); 255 perf_env__exit(&session->header.env); 256 machines__exit(&session->machines); 257 if (session->data) { 258 if (perf_data__is_read(session->data)) 259 evlist__delete(session->evlist); 260 perf_data__close(session->data); 261 } 262 #ifdef HAVE_LIBTRACEEVENT 263 trace_event__cleanup(&session->tevent); 264 #endif 265 free(session); 266 } 267 268 static void swap_sample_id_all(union perf_event *event, void *data) 269 { 270 void *end = (void *) event + event->header.size; 271 int size = end - data; 272 273 BUG_ON(size % sizeof(u64)); 274 mem_bswap_64(data, size); 275 } 276 277 static void perf_event__all64_swap(union perf_event *event, 278 bool sample_id_all __maybe_unused) 279 { 280 struct perf_event_header *hdr = &event->header; 281 mem_bswap_64(hdr + 1, event->header.size - sizeof(*hdr)); 282 } 283 284 static void perf_event__comm_swap(union perf_event *event, bool sample_id_all) 285 { 286 event->comm.pid = bswap_32(event->comm.pid); 287 event->comm.tid = bswap_32(event->comm.tid); 288 289 if (sample_id_all) { 290 void *data = &event->comm.comm; 291 292 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64)); 293 swap_sample_id_all(event, data); 294 } 295 } 296 297 static void perf_event__mmap_swap(union perf_event *event, 298 bool sample_id_all) 299 { 300 event->mmap.pid = bswap_32(event->mmap.pid); 301 event->mmap.tid = bswap_32(event->mmap.tid); 302 event->mmap.start = bswap_64(event->mmap.start); 303 event->mmap.len = bswap_64(event->mmap.len); 304 event->mmap.pgoff = bswap_64(event->mmap.pgoff); 305 306 if (sample_id_all) { 307 void *data = &event->mmap.filename; 308 309 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64)); 310 swap_sample_id_all(event, data); 311 } 312 } 313 314 static void perf_event__mmap2_swap(union perf_event *event, 315 bool sample_id_all) 316 { 317 event->mmap2.pid = bswap_32(event->mmap2.pid); 318 event->mmap2.tid = bswap_32(event->mmap2.tid); 319 event->mmap2.start = bswap_64(event->mmap2.start); 320 event->mmap2.len = bswap_64(event->mmap2.len); 321 event->mmap2.pgoff = bswap_64(event->mmap2.pgoff); 322 323 if (!(event->header.misc & PERF_RECORD_MISC_MMAP_BUILD_ID)) { 324 event->mmap2.maj = bswap_32(event->mmap2.maj); 325 event->mmap2.min = bswap_32(event->mmap2.min); 326 event->mmap2.ino = bswap_64(event->mmap2.ino); 327 event->mmap2.ino_generation = bswap_64(event->mmap2.ino_generation); 328 } 329 330 if (sample_id_all) { 331 void *data = &event->mmap2.filename; 332 333 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64)); 334 swap_sample_id_all(event, data); 335 } 336 } 337 static void perf_event__task_swap(union perf_event *event, bool sample_id_all) 338 { 339 event->fork.pid = bswap_32(event->fork.pid); 340 event->fork.tid = bswap_32(event->fork.tid); 341 event->fork.ppid = bswap_32(event->fork.ppid); 342 event->fork.ptid = bswap_32(event->fork.ptid); 343 event->fork.time = bswap_64(event->fork.time); 344 345 if (sample_id_all) 346 swap_sample_id_all(event, &event->fork + 1); 347 } 348 349 static void perf_event__read_swap(union perf_event *event, bool sample_id_all) 350 { 351 event->read.pid = bswap_32(event->read.pid); 352 event->read.tid = bswap_32(event->read.tid); 353 event->read.value = bswap_64(event->read.value); 354 event->read.time_enabled = bswap_64(event->read.time_enabled); 355 event->read.time_running = bswap_64(event->read.time_running); 356 event->read.id = bswap_64(event->read.id); 357 358 if (sample_id_all) 359 swap_sample_id_all(event, &event->read + 1); 360 } 361 362 static void perf_event__aux_swap(union perf_event *event, bool sample_id_all) 363 { 364 event->aux.aux_offset = bswap_64(event->aux.aux_offset); 365 event->aux.aux_size = bswap_64(event->aux.aux_size); 366 event->aux.flags = bswap_64(event->aux.flags); 367 368 if (sample_id_all) 369 swap_sample_id_all(event, &event->aux + 1); 370 } 371 372 static void perf_event__itrace_start_swap(union perf_event *event, 373 bool sample_id_all) 374 { 375 event->itrace_start.pid = bswap_32(event->itrace_start.pid); 376 event->itrace_start.tid = bswap_32(event->itrace_start.tid); 377 378 if (sample_id_all) 379 swap_sample_id_all(event, &event->itrace_start + 1); 380 } 381 382 static void perf_event__switch_swap(union perf_event *event, bool sample_id_all) 383 { 384 if (event->header.type == PERF_RECORD_SWITCH_CPU_WIDE) { 385 event->context_switch.next_prev_pid = 386 bswap_32(event->context_switch.next_prev_pid); 387 event->context_switch.next_prev_tid = 388 bswap_32(event->context_switch.next_prev_tid); 389 } 390 391 if (sample_id_all) 392 swap_sample_id_all(event, &event->context_switch + 1); 393 } 394 395 static void perf_event__text_poke_swap(union perf_event *event, bool sample_id_all) 396 { 397 event->text_poke.addr = bswap_64(event->text_poke.addr); 398 event->text_poke.old_len = bswap_16(event->text_poke.old_len); 399 event->text_poke.new_len = bswap_16(event->text_poke.new_len); 400 401 if (sample_id_all) { 402 size_t len = sizeof(event->text_poke.old_len) + 403 sizeof(event->text_poke.new_len) + 404 event->text_poke.old_len + 405 event->text_poke.new_len; 406 void *data = &event->text_poke.old_len; 407 408 data += PERF_ALIGN(len, sizeof(u64)); 409 swap_sample_id_all(event, data); 410 } 411 } 412 413 static void perf_event__throttle_swap(union perf_event *event, 414 bool sample_id_all) 415 { 416 event->throttle.time = bswap_64(event->throttle.time); 417 event->throttle.id = bswap_64(event->throttle.id); 418 event->throttle.stream_id = bswap_64(event->throttle.stream_id); 419 420 if (sample_id_all) 421 swap_sample_id_all(event, &event->throttle + 1); 422 } 423 424 static void perf_event__namespaces_swap(union perf_event *event, 425 bool sample_id_all) 426 { 427 u64 i; 428 429 event->namespaces.pid = bswap_32(event->namespaces.pid); 430 event->namespaces.tid = bswap_32(event->namespaces.tid); 431 event->namespaces.nr_namespaces = bswap_64(event->namespaces.nr_namespaces); 432 433 for (i = 0; i < event->namespaces.nr_namespaces; i++) { 434 struct perf_ns_link_info *ns = &event->namespaces.link_info[i]; 435 436 ns->dev = bswap_64(ns->dev); 437 ns->ino = bswap_64(ns->ino); 438 } 439 440 if (sample_id_all) 441 swap_sample_id_all(event, &event->namespaces.link_info[i]); 442 } 443 444 static void perf_event__cgroup_swap(union perf_event *event, bool sample_id_all) 445 { 446 event->cgroup.id = bswap_64(event->cgroup.id); 447 448 if (sample_id_all) { 449 void *data = &event->cgroup.path; 450 451 data += PERF_ALIGN(strlen(data) + 1, sizeof(u64)); 452 swap_sample_id_all(event, data); 453 } 454 } 455 456 static u8 revbyte(u8 b) 457 { 458 int rev = (b >> 4) | ((b & 0xf) << 4); 459 rev = ((rev & 0xcc) >> 2) | ((rev & 0x33) << 2); 460 rev = ((rev & 0xaa) >> 1) | ((rev & 0x55) << 1); 461 return (u8) rev; 462 } 463 464 /* 465 * XXX this is hack in attempt to carry flags bitfield 466 * through endian village. ABI says: 467 * 468 * Bit-fields are allocated from right to left (least to most significant) 469 * on little-endian implementations and from left to right (most to least 470 * significant) on big-endian implementations. 471 * 472 * The above seems to be byte specific, so we need to reverse each 473 * byte of the bitfield. 'Internet' also says this might be implementation 474 * specific and we probably need proper fix and carry perf_event_attr 475 * bitfield flags in separate data file FEAT_ section. Thought this seems 476 * to work for now. 477 */ 478 static void swap_bitfield(u8 *p, unsigned len) 479 { 480 unsigned i; 481 482 for (i = 0; i < len; i++) { 483 *p = revbyte(*p); 484 p++; 485 } 486 } 487 488 /* exported for swapping attributes in file header */ 489 void perf_event__attr_swap(struct perf_event_attr *attr) 490 { 491 attr->type = bswap_32(attr->type); 492 attr->size = bswap_32(attr->size); 493 494 #define bswap_safe(f, n) \ 495 (attr->size > (offsetof(struct perf_event_attr, f) + \ 496 sizeof(attr->f) * (n))) 497 #define bswap_field(f, sz) \ 498 do { \ 499 if (bswap_safe(f, 0)) \ 500 attr->f = bswap_##sz(attr->f); \ 501 } while(0) 502 #define bswap_field_16(f) bswap_field(f, 16) 503 #define bswap_field_32(f) bswap_field(f, 32) 504 #define bswap_field_64(f) bswap_field(f, 64) 505 506 bswap_field_64(config); 507 bswap_field_64(sample_period); 508 bswap_field_64(sample_type); 509 bswap_field_64(read_format); 510 bswap_field_32(wakeup_events); 511 bswap_field_32(bp_type); 512 bswap_field_64(bp_addr); 513 bswap_field_64(bp_len); 514 bswap_field_64(branch_sample_type); 515 bswap_field_64(sample_regs_user); 516 bswap_field_32(sample_stack_user); 517 bswap_field_32(aux_watermark); 518 bswap_field_16(sample_max_stack); 519 bswap_field_32(aux_sample_size); 520 521 /* 522 * After read_format are bitfields. Check read_format because 523 * we are unable to use offsetof on bitfield. 524 */ 525 if (bswap_safe(read_format, 1)) 526 swap_bitfield((u8 *) (&attr->read_format + 1), 527 sizeof(u64)); 528 #undef bswap_field_64 529 #undef bswap_field_32 530 #undef bswap_field 531 #undef bswap_safe 532 } 533 534 static void perf_event__hdr_attr_swap(union perf_event *event, 535 bool sample_id_all __maybe_unused) 536 { 537 size_t size; 538 539 perf_event__attr_swap(&event->attr.attr); 540 541 size = event->header.size; 542 size -= perf_record_header_attr_id(event) - (void *)event; 543 mem_bswap_64(perf_record_header_attr_id(event), size); 544 } 545 546 static void perf_event__event_update_swap(union perf_event *event, 547 bool sample_id_all __maybe_unused) 548 { 549 event->event_update.type = bswap_64(event->event_update.type); 550 event->event_update.id = bswap_64(event->event_update.id); 551 } 552 553 static void perf_event__event_type_swap(union perf_event *event, 554 bool sample_id_all __maybe_unused) 555 { 556 event->event_type.event_type.event_id = 557 bswap_64(event->event_type.event_type.event_id); 558 } 559 560 static void perf_event__tracing_data_swap(union perf_event *event, 561 bool sample_id_all __maybe_unused) 562 { 563 event->tracing_data.size = bswap_32(event->tracing_data.size); 564 } 565 566 static void perf_event__auxtrace_info_swap(union perf_event *event, 567 bool sample_id_all __maybe_unused) 568 { 569 size_t size; 570 571 event->auxtrace_info.type = bswap_32(event->auxtrace_info.type); 572 573 size = event->header.size; 574 size -= (void *)&event->auxtrace_info.priv - (void *)event; 575 mem_bswap_64(event->auxtrace_info.priv, size); 576 } 577 578 static void perf_event__auxtrace_swap(union perf_event *event, 579 bool sample_id_all __maybe_unused) 580 { 581 event->auxtrace.size = bswap_64(event->auxtrace.size); 582 event->auxtrace.offset = bswap_64(event->auxtrace.offset); 583 event->auxtrace.reference = bswap_64(event->auxtrace.reference); 584 event->auxtrace.idx = bswap_32(event->auxtrace.idx); 585 event->auxtrace.tid = bswap_32(event->auxtrace.tid); 586 event->auxtrace.cpu = bswap_32(event->auxtrace.cpu); 587 } 588 589 static void perf_event__auxtrace_error_swap(union perf_event *event, 590 bool sample_id_all __maybe_unused) 591 { 592 event->auxtrace_error.type = bswap_32(event->auxtrace_error.type); 593 event->auxtrace_error.code = bswap_32(event->auxtrace_error.code); 594 event->auxtrace_error.cpu = bswap_32(event->auxtrace_error.cpu); 595 event->auxtrace_error.pid = bswap_32(event->auxtrace_error.pid); 596 event->auxtrace_error.tid = bswap_32(event->auxtrace_error.tid); 597 event->auxtrace_error.fmt = bswap_32(event->auxtrace_error.fmt); 598 event->auxtrace_error.ip = bswap_64(event->auxtrace_error.ip); 599 if (event->auxtrace_error.fmt) 600 event->auxtrace_error.time = bswap_64(event->auxtrace_error.time); 601 if (event->auxtrace_error.fmt >= 2) { 602 event->auxtrace_error.machine_pid = bswap_32(event->auxtrace_error.machine_pid); 603 event->auxtrace_error.vcpu = bswap_32(event->auxtrace_error.vcpu); 604 } 605 } 606 607 static void perf_event__thread_map_swap(union perf_event *event, 608 bool sample_id_all __maybe_unused) 609 { 610 unsigned i; 611 612 event->thread_map.nr = bswap_64(event->thread_map.nr); 613 614 for (i = 0; i < event->thread_map.nr; i++) 615 event->thread_map.entries[i].pid = bswap_64(event->thread_map.entries[i].pid); 616 } 617 618 static void perf_event__cpu_map_swap(union perf_event *event, 619 bool sample_id_all __maybe_unused) 620 { 621 struct perf_record_cpu_map_data *data = &event->cpu_map.data; 622 623 data->type = bswap_16(data->type); 624 625 switch (data->type) { 626 case PERF_CPU_MAP__CPUS: 627 data->cpus_data.nr = bswap_16(data->cpus_data.nr); 628 629 for (unsigned i = 0; i < data->cpus_data.nr; i++) 630 data->cpus_data.cpu[i] = bswap_16(data->cpus_data.cpu[i]); 631 break; 632 case PERF_CPU_MAP__MASK: 633 data->mask32_data.long_size = bswap_16(data->mask32_data.long_size); 634 635 switch (data->mask32_data.long_size) { 636 case 4: 637 data->mask32_data.nr = bswap_16(data->mask32_data.nr); 638 for (unsigned i = 0; i < data->mask32_data.nr; i++) 639 data->mask32_data.mask[i] = bswap_32(data->mask32_data.mask[i]); 640 break; 641 case 8: 642 data->mask64_data.nr = bswap_16(data->mask64_data.nr); 643 for (unsigned i = 0; i < data->mask64_data.nr; i++) 644 data->mask64_data.mask[i] = bswap_64(data->mask64_data.mask[i]); 645 break; 646 default: 647 pr_err("cpu_map swap: unsupported long size\n"); 648 } 649 break; 650 case PERF_CPU_MAP__RANGE_CPUS: 651 data->range_cpu_data.start_cpu = bswap_16(data->range_cpu_data.start_cpu); 652 data->range_cpu_data.end_cpu = bswap_16(data->range_cpu_data.end_cpu); 653 break; 654 default: 655 break; 656 } 657 } 658 659 static void perf_event__stat_config_swap(union perf_event *event, 660 bool sample_id_all __maybe_unused) 661 { 662 u64 size; 663 664 size = bswap_64(event->stat_config.nr) * sizeof(event->stat_config.data[0]); 665 size += 1; /* nr item itself */ 666 mem_bswap_64(&event->stat_config.nr, size); 667 } 668 669 static void perf_event__stat_swap(union perf_event *event, 670 bool sample_id_all __maybe_unused) 671 { 672 event->stat.id = bswap_64(event->stat.id); 673 event->stat.thread = bswap_32(event->stat.thread); 674 event->stat.cpu = bswap_32(event->stat.cpu); 675 event->stat.val = bswap_64(event->stat.val); 676 event->stat.ena = bswap_64(event->stat.ena); 677 event->stat.run = bswap_64(event->stat.run); 678 } 679 680 static void perf_event__stat_round_swap(union perf_event *event, 681 bool sample_id_all __maybe_unused) 682 { 683 event->stat_round.type = bswap_64(event->stat_round.type); 684 event->stat_round.time = bswap_64(event->stat_round.time); 685 } 686 687 static void perf_event__time_conv_swap(union perf_event *event, 688 bool sample_id_all __maybe_unused) 689 { 690 event->time_conv.time_shift = bswap_64(event->time_conv.time_shift); 691 event->time_conv.time_mult = bswap_64(event->time_conv.time_mult); 692 event->time_conv.time_zero = bswap_64(event->time_conv.time_zero); 693 694 if (event_contains(event->time_conv, time_cycles)) { 695 event->time_conv.time_cycles = bswap_64(event->time_conv.time_cycles); 696 event->time_conv.time_mask = bswap_64(event->time_conv.time_mask); 697 } 698 } 699 700 typedef void (*perf_event__swap_op)(union perf_event *event, 701 bool sample_id_all); 702 703 static perf_event__swap_op perf_event__swap_ops[] = { 704 [PERF_RECORD_MMAP] = perf_event__mmap_swap, 705 [PERF_RECORD_MMAP2] = perf_event__mmap2_swap, 706 [PERF_RECORD_COMM] = perf_event__comm_swap, 707 [PERF_RECORD_FORK] = perf_event__task_swap, 708 [PERF_RECORD_EXIT] = perf_event__task_swap, 709 [PERF_RECORD_LOST] = perf_event__all64_swap, 710 [PERF_RECORD_READ] = perf_event__read_swap, 711 [PERF_RECORD_THROTTLE] = perf_event__throttle_swap, 712 [PERF_RECORD_UNTHROTTLE] = perf_event__throttle_swap, 713 [PERF_RECORD_SAMPLE] = perf_event__all64_swap, 714 [PERF_RECORD_AUX] = perf_event__aux_swap, 715 [PERF_RECORD_ITRACE_START] = perf_event__itrace_start_swap, 716 [PERF_RECORD_LOST_SAMPLES] = perf_event__all64_swap, 717 [PERF_RECORD_SWITCH] = perf_event__switch_swap, 718 [PERF_RECORD_SWITCH_CPU_WIDE] = perf_event__switch_swap, 719 [PERF_RECORD_NAMESPACES] = perf_event__namespaces_swap, 720 [PERF_RECORD_CGROUP] = perf_event__cgroup_swap, 721 [PERF_RECORD_TEXT_POKE] = perf_event__text_poke_swap, 722 [PERF_RECORD_AUX_OUTPUT_HW_ID] = perf_event__all64_swap, 723 [PERF_RECORD_CALLCHAIN_DEFERRED] = perf_event__all64_swap, 724 [PERF_RECORD_HEADER_ATTR] = perf_event__hdr_attr_swap, 725 [PERF_RECORD_HEADER_EVENT_TYPE] = perf_event__event_type_swap, 726 [PERF_RECORD_HEADER_TRACING_DATA] = perf_event__tracing_data_swap, 727 [PERF_RECORD_HEADER_BUILD_ID] = NULL, 728 [PERF_RECORD_ID_INDEX] = perf_event__all64_swap, 729 [PERF_RECORD_AUXTRACE_INFO] = perf_event__auxtrace_info_swap, 730 [PERF_RECORD_AUXTRACE] = perf_event__auxtrace_swap, 731 [PERF_RECORD_AUXTRACE_ERROR] = perf_event__auxtrace_error_swap, 732 [PERF_RECORD_THREAD_MAP] = perf_event__thread_map_swap, 733 [PERF_RECORD_CPU_MAP] = perf_event__cpu_map_swap, 734 [PERF_RECORD_STAT_CONFIG] = perf_event__stat_config_swap, 735 [PERF_RECORD_STAT] = perf_event__stat_swap, 736 [PERF_RECORD_STAT_ROUND] = perf_event__stat_round_swap, 737 [PERF_RECORD_EVENT_UPDATE] = perf_event__event_update_swap, 738 [PERF_RECORD_TIME_CONV] = perf_event__time_conv_swap, 739 [PERF_RECORD_HEADER_MAX] = NULL, 740 }; 741 742 /* 743 * When perf record finishes a pass on every buffers, it records this pseudo 744 * event. 745 * We record the max timestamp t found in the pass n. 746 * Assuming these timestamps are monotonic across cpus, we know that if 747 * a buffer still has events with timestamps below t, they will be all 748 * available and then read in the pass n + 1. 749 * Hence when we start to read the pass n + 2, we can safely flush every 750 * events with timestamps below t. 751 * 752 * ============ PASS n ================= 753 * CPU 0 | CPU 1 754 * | 755 * cnt1 timestamps | cnt2 timestamps 756 * 1 | 2 757 * 2 | 3 758 * - | 4 <--- max recorded 759 * 760 * ============ PASS n + 1 ============== 761 * CPU 0 | CPU 1 762 * | 763 * cnt1 timestamps | cnt2 timestamps 764 * 3 | 5 765 * 4 | 6 766 * 5 | 7 <---- max recorded 767 * 768 * Flush every events below timestamp 4 769 * 770 * ============ PASS n + 2 ============== 771 * CPU 0 | CPU 1 772 * | 773 * cnt1 timestamps | cnt2 timestamps 774 * 6 | 8 775 * 7 | 9 776 * - | 10 777 * 778 * Flush every events below timestamp 7 779 * etc... 780 */ 781 int perf_event__process_finished_round(const struct perf_tool *tool __maybe_unused, 782 union perf_event *event __maybe_unused, 783 struct ordered_events *oe) 784 { 785 if (dump_trace) 786 fprintf(stdout, "\n"); 787 return ordered_events__flush(oe, OE_FLUSH__ROUND); 788 } 789 790 int perf_session__queue_event(struct perf_session *s, union perf_event *event, 791 u64 timestamp, u64 file_offset, const char *file_path) 792 { 793 return ordered_events__queue(&s->ordered_events, event, timestamp, file_offset, file_path); 794 } 795 796 static void callchain__lbr_callstack_printf(struct perf_sample *sample) 797 { 798 struct ip_callchain *callchain = sample->callchain; 799 struct branch_stack *lbr_stack = sample->branch_stack; 800 struct branch_entry *entries = perf_sample__branch_entries(sample); 801 u64 kernel_callchain_nr = callchain->nr; 802 unsigned int i; 803 804 for (i = 0; i < kernel_callchain_nr; i++) { 805 if (callchain->ips[i] == PERF_CONTEXT_USER) 806 break; 807 } 808 809 if ((i != kernel_callchain_nr) && lbr_stack->nr) { 810 u64 total_nr; 811 /* 812 * LBR callstack can only get user call chain, 813 * i is kernel call chain number, 814 * 1 is PERF_CONTEXT_USER. 815 * 816 * The user call chain is stored in LBR registers. 817 * LBR are pair registers. The caller is stored 818 * in "from" register, while the callee is stored 819 * in "to" register. 820 * For example, there is a call stack 821 * "A"->"B"->"C"->"D". 822 * The LBR registers will be recorded like 823 * "C"->"D", "B"->"C", "A"->"B". 824 * So only the first "to" register and all "from" 825 * registers are needed to construct the whole stack. 826 */ 827 total_nr = i + 1 + lbr_stack->nr + 1; 828 kernel_callchain_nr = i + 1; 829 830 printf("... LBR call chain: nr:%" PRIu64 "\n", total_nr); 831 832 for (i = 0; i < kernel_callchain_nr; i++) 833 printf("..... %2d: %016" PRIx64 "\n", 834 i, callchain->ips[i]); 835 836 printf("..... %2d: %016" PRIx64 "\n", 837 (int)(kernel_callchain_nr), entries[0].to); 838 for (i = 0; i < lbr_stack->nr; i++) 839 printf("..... %2d: %016" PRIx64 "\n", 840 (int)(i + kernel_callchain_nr + 1), entries[i].from); 841 } 842 } 843 844 static void callchain__printf(struct evsel *evsel, 845 struct perf_sample *sample) 846 { 847 unsigned int i; 848 struct ip_callchain *callchain = sample->callchain; 849 850 if (evsel__has_branch_callstack(evsel)) 851 callchain__lbr_callstack_printf(sample); 852 853 printf("... FP chain: nr:%" PRIu64 "\n", callchain->nr); 854 855 for (i = 0; i < callchain->nr; i++) 856 printf("..... %2d: %016" PRIx64 "\n", 857 i, callchain->ips[i]); 858 859 if (sample->deferred_callchain) 860 printf("...... (deferred)\n"); 861 } 862 863 static void branch_stack__printf(struct perf_sample *sample, 864 struct evsel *evsel) 865 { 866 struct branch_entry *entries = perf_sample__branch_entries(sample); 867 bool callstack = evsel__has_branch_callstack(evsel); 868 u64 *branch_stack_cntr = sample->branch_stack_cntr; 869 uint64_t i; 870 871 if (!callstack) { 872 printf("%s: nr:%" PRIu64 "\n", "... branch stack", sample->branch_stack->nr); 873 } else { 874 /* the reason of adding 1 to nr is because after expanding 875 * branch stack it generates nr + 1 callstack records. e.g., 876 * B()->C() 877 * A()->B() 878 * the final callstack should be: 879 * C() 880 * B() 881 * A() 882 */ 883 printf("%s: nr:%" PRIu64 "\n", "... branch callstack", sample->branch_stack->nr+1); 884 } 885 886 for (i = 0; i < sample->branch_stack->nr; i++) { 887 struct branch_entry *e = &entries[i]; 888 889 if (!callstack) { 890 printf("..... %2"PRIu64": %016" PRIx64 " -> %016" PRIx64 " %hu cycles %s%s%s%s %x %s %s\n", 891 i, e->from, e->to, 892 (unsigned short)e->flags.cycles, 893 e->flags.mispred ? "M" : " ", 894 e->flags.predicted ? "P" : " ", 895 e->flags.abort ? "A" : " ", 896 e->flags.in_tx ? "T" : " ", 897 (unsigned)e->flags.reserved, 898 get_branch_type(e), 899 e->flags.spec ? branch_spec_desc(e->flags.spec) : ""); 900 } else { 901 if (i == 0) { 902 printf("..... %2"PRIu64": %016" PRIx64 "\n" 903 "..... %2"PRIu64": %016" PRIx64 "\n", 904 i, e->to, i+1, e->from); 905 } else { 906 printf("..... %2"PRIu64": %016" PRIx64 "\n", i+1, e->from); 907 } 908 } 909 } 910 911 if (branch_stack_cntr) { 912 unsigned int br_cntr_width, br_cntr_nr; 913 914 perf_env__find_br_cntr_info(evsel__env(evsel), &br_cntr_nr, &br_cntr_width); 915 printf("... branch stack counters: nr:%" PRIu64 " (counter width: %u max counter nr:%u)\n", 916 sample->branch_stack->nr, br_cntr_width, br_cntr_nr); 917 for (i = 0; i < sample->branch_stack->nr; i++) 918 printf("..... %2"PRIu64": %016" PRIx64 "\n", i, branch_stack_cntr[i]); 919 } 920 } 921 922 static void regs_dump__printf(u64 mask, u64 *regs, const char *arch) 923 { 924 unsigned rid, i = 0; 925 926 for_each_set_bit(rid, (unsigned long *) &mask, sizeof(mask) * 8) { 927 u64 val = regs[i++]; 928 929 printf(".... %-5s 0x%016" PRIx64 "\n", 930 perf_reg_name(rid, arch), val); 931 } 932 } 933 934 static const char *regs_abi[] = { 935 [PERF_SAMPLE_REGS_ABI_NONE] = "none", 936 [PERF_SAMPLE_REGS_ABI_32] = "32-bit", 937 [PERF_SAMPLE_REGS_ABI_64] = "64-bit", 938 }; 939 940 static inline const char *regs_dump_abi(struct regs_dump *d) 941 { 942 if (d->abi > PERF_SAMPLE_REGS_ABI_64) 943 return "unknown"; 944 945 return regs_abi[d->abi]; 946 } 947 948 static void regs__printf(const char *type, struct regs_dump *regs, const char *arch) 949 { 950 u64 mask = regs->mask; 951 952 printf("... %s regs: mask 0x%" PRIx64 " ABI %s\n", 953 type, 954 mask, 955 regs_dump_abi(regs)); 956 957 regs_dump__printf(mask, regs->regs, arch); 958 } 959 960 static void regs_user__printf(struct perf_sample *sample, const char *arch) 961 { 962 struct regs_dump *user_regs; 963 964 if (!sample->user_regs) 965 return; 966 967 user_regs = perf_sample__user_regs(sample); 968 969 if (user_regs->regs) 970 regs__printf("user", user_regs, arch); 971 } 972 973 static void regs_intr__printf(struct perf_sample *sample, const char *arch) 974 { 975 struct regs_dump *intr_regs; 976 977 if (!sample->intr_regs) 978 return; 979 980 intr_regs = perf_sample__intr_regs(sample); 981 982 if (intr_regs->regs) 983 regs__printf("intr", intr_regs, arch); 984 } 985 986 static void stack_user__printf(struct stack_dump *dump) 987 { 988 printf("... ustack: size %" PRIu64 ", offset 0x%x\n", 989 dump->size, dump->offset); 990 } 991 992 static void evlist__print_tstamp(struct evlist *evlist, union perf_event *event, struct perf_sample *sample) 993 { 994 u64 sample_type = __evlist__combined_sample_type(evlist); 995 996 if (event->header.type != PERF_RECORD_SAMPLE && 997 !evlist__sample_id_all(evlist)) { 998 fputs("-1 -1 ", stdout); 999 return; 1000 } 1001 1002 if ((sample_type & PERF_SAMPLE_CPU)) 1003 printf("%u ", sample->cpu); 1004 1005 if (sample_type & PERF_SAMPLE_TIME) 1006 printf("%" PRIu64 " ", sample->time); 1007 } 1008 1009 static void sample_read__printf(struct perf_sample *sample, u64 read_format) 1010 { 1011 printf("... sample_read:\n"); 1012 1013 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) 1014 printf("...... time enabled %016" PRIx64 "\n", 1015 sample->read.time_enabled); 1016 1017 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) 1018 printf("...... time running %016" PRIx64 "\n", 1019 sample->read.time_running); 1020 1021 if (read_format & PERF_FORMAT_GROUP) { 1022 struct sample_read_value *value = sample->read.group.values; 1023 1024 printf(".... group nr %" PRIu64 "\n", sample->read.group.nr); 1025 1026 sample_read_group__for_each(value, sample->read.group.nr, read_format) { 1027 printf("..... id %016" PRIx64 1028 ", value %016" PRIx64, 1029 value->id, value->value); 1030 if (read_format & PERF_FORMAT_LOST) 1031 printf(", lost %" PRIu64, value->lost); 1032 printf("\n"); 1033 } 1034 } else { 1035 printf("..... id %016" PRIx64 ", value %016" PRIx64, 1036 sample->read.one.id, sample->read.one.value); 1037 if (read_format & PERF_FORMAT_LOST) 1038 printf(", lost %" PRIu64, sample->read.one.lost); 1039 printf("\n"); 1040 } 1041 } 1042 1043 static void dump_event(struct evlist *evlist, union perf_event *event, 1044 u64 file_offset, struct perf_sample *sample, 1045 const char *file_path) 1046 { 1047 if (!dump_trace) 1048 return; 1049 1050 printf("\n%#" PRIx64 "@%s [%#x]: event: %d\n", 1051 file_offset, file_path, event->header.size, event->header.type); 1052 1053 trace_event(event); 1054 if (event->header.type == PERF_RECORD_SAMPLE && evlist->trace_event_sample_raw) 1055 evlist->trace_event_sample_raw(evlist, event, sample); 1056 1057 if (sample) 1058 evlist__print_tstamp(evlist, event, sample); 1059 1060 printf("%#" PRIx64 " [%#x]: PERF_RECORD_%s", file_offset, 1061 event->header.size, perf_event__name(event->header.type)); 1062 } 1063 1064 char *get_page_size_name(u64 size, char *str) 1065 { 1066 if (!size || !unit_number__scnprintf(str, PAGE_SIZE_NAME_LEN, size)) 1067 snprintf(str, PAGE_SIZE_NAME_LEN, "%s", "N/A"); 1068 1069 return str; 1070 } 1071 1072 static void dump_sample(struct evsel *evsel, union perf_event *event, 1073 struct perf_sample *sample, const char *arch) 1074 { 1075 u64 sample_type; 1076 char str[PAGE_SIZE_NAME_LEN]; 1077 1078 if (!dump_trace) 1079 return; 1080 1081 printf("(IP, 0x%x): %d/%d: %#" PRIx64 " period: %" PRIu64 " addr: %#" PRIx64 "\n", 1082 event->header.misc, sample->pid, sample->tid, sample->ip, 1083 sample->period, sample->addr); 1084 1085 sample_type = evsel->core.attr.sample_type; 1086 1087 if (evsel__has_callchain(evsel)) 1088 callchain__printf(evsel, sample); 1089 1090 if (evsel__has_br_stack(evsel)) 1091 branch_stack__printf(sample, evsel); 1092 1093 if (sample_type & PERF_SAMPLE_REGS_USER) 1094 regs_user__printf(sample, arch); 1095 1096 if (sample_type & PERF_SAMPLE_REGS_INTR) 1097 regs_intr__printf(sample, arch); 1098 1099 if (sample_type & PERF_SAMPLE_STACK_USER) 1100 stack_user__printf(&sample->user_stack); 1101 1102 if (sample_type & PERF_SAMPLE_WEIGHT_TYPE) { 1103 printf("... weight: %" PRIu64 "", sample->weight); 1104 if (sample_type & PERF_SAMPLE_WEIGHT_STRUCT) { 1105 printf(",0x%"PRIx16"", sample->ins_lat); 1106 printf(",0x%"PRIx16"", sample->weight3); 1107 } 1108 printf("\n"); 1109 } 1110 1111 if (sample_type & PERF_SAMPLE_DATA_SRC) 1112 printf(" . data_src: 0x%"PRIx64"\n", sample->data_src); 1113 1114 if (sample_type & PERF_SAMPLE_PHYS_ADDR) 1115 printf(" .. phys_addr: 0x%"PRIx64"\n", sample->phys_addr); 1116 1117 if (sample_type & PERF_SAMPLE_DATA_PAGE_SIZE) 1118 printf(" .. data page size: %s\n", get_page_size_name(sample->data_page_size, str)); 1119 1120 if (sample_type & PERF_SAMPLE_CODE_PAGE_SIZE) 1121 printf(" .. code page size: %s\n", get_page_size_name(sample->code_page_size, str)); 1122 1123 if (sample_type & PERF_SAMPLE_TRANSACTION) 1124 printf("... transaction: %" PRIx64 "\n", sample->transaction); 1125 1126 if (sample_type & PERF_SAMPLE_READ) 1127 sample_read__printf(sample, evsel->core.attr.read_format); 1128 } 1129 1130 static void dump_deferred_callchain(struct evsel *evsel, union perf_event *event, 1131 struct perf_sample *sample) 1132 { 1133 if (!dump_trace) 1134 return; 1135 1136 printf("(IP, 0x%x): %d/%d: %#" PRIx64 "\n", 1137 event->header.misc, sample->pid, sample->tid, sample->deferred_cookie); 1138 1139 if (evsel__has_callchain(evsel)) 1140 callchain__printf(evsel, sample); 1141 } 1142 1143 static void dump_read(struct evsel *evsel, union perf_event *event) 1144 { 1145 struct perf_record_read *read_event = &event->read; 1146 u64 read_format; 1147 1148 if (!dump_trace) 1149 return; 1150 1151 printf(": %d %d %s %" PRI_lu64 "\n", event->read.pid, event->read.tid, 1152 evsel__name(evsel), event->read.value); 1153 1154 if (!evsel) 1155 return; 1156 1157 read_format = evsel->core.attr.read_format; 1158 1159 if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) 1160 printf("... time enabled : %" PRI_lu64 "\n", read_event->time_enabled); 1161 1162 if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) 1163 printf("... time running : %" PRI_lu64 "\n", read_event->time_running); 1164 1165 if (read_format & PERF_FORMAT_ID) 1166 printf("... id : %" PRI_lu64 "\n", read_event->id); 1167 1168 if (read_format & PERF_FORMAT_LOST) 1169 printf("... lost : %" PRI_lu64 "\n", read_event->lost); 1170 } 1171 1172 static struct machine *machines__find_for_cpumode(struct machines *machines, 1173 union perf_event *event, 1174 struct perf_sample *sample) 1175 { 1176 if (perf_guest && 1177 ((sample->cpumode == PERF_RECORD_MISC_GUEST_KERNEL) || 1178 (sample->cpumode == PERF_RECORD_MISC_GUEST_USER))) { 1179 u32 pid; 1180 1181 if (sample->machine_pid) 1182 pid = sample->machine_pid; 1183 else if (event->header.type == PERF_RECORD_MMAP 1184 || event->header.type == PERF_RECORD_MMAP2) 1185 pid = event->mmap.pid; 1186 else 1187 pid = sample->pid; 1188 1189 /* 1190 * Guest code machine is created as needed and does not use 1191 * DEFAULT_GUEST_KERNEL_ID. 1192 */ 1193 if (symbol_conf.guest_code) 1194 return machines__findnew(machines, pid); 1195 1196 return machines__find_guest(machines, pid); 1197 } 1198 1199 return &machines->host; 1200 } 1201 1202 static int deliver_sample_value(struct evlist *evlist, 1203 const struct perf_tool *tool, 1204 union perf_event *event, 1205 struct perf_sample *sample, 1206 struct sample_read_value *v, 1207 struct machine *machine, 1208 bool per_thread) 1209 { 1210 struct perf_sample_id *sid = evlist__id2sid(evlist, v->id); 1211 struct evsel *evsel; 1212 u64 *storage = NULL; 1213 1214 if (sid) { 1215 storage = perf_sample_id__get_period_storage(sid, sample->tid, per_thread); 1216 } 1217 1218 if (storage) { 1219 sample->id = v->id; 1220 sample->period = v->value - *storage; 1221 *storage = v->value; 1222 } 1223 1224 if (!storage || sid->evsel == NULL) { 1225 ++evlist->stats.nr_unknown_id; 1226 return 0; 1227 } 1228 1229 /* 1230 * There's no reason to deliver sample 1231 * for zero period, bail out. 1232 */ 1233 if (!sample->period) 1234 return 0; 1235 1236 evsel = container_of(sid->evsel, struct evsel, core); 1237 return tool->sample(tool, event, sample, evsel, machine); 1238 } 1239 1240 static int deliver_sample_group(struct evlist *evlist, 1241 const struct perf_tool *tool, 1242 union perf_event *event, 1243 struct perf_sample *sample, 1244 struct machine *machine, 1245 u64 read_format, 1246 bool per_thread) 1247 { 1248 int ret = -EINVAL; 1249 struct sample_read_value *v = sample->read.group.values; 1250 1251 if (tool->dont_split_sample_group) 1252 return deliver_sample_value(evlist, tool, event, sample, v, machine, 1253 per_thread); 1254 1255 sample_read_group__for_each(v, sample->read.group.nr, read_format) { 1256 ret = deliver_sample_value(evlist, tool, event, sample, v, 1257 machine, per_thread); 1258 if (ret) 1259 break; 1260 } 1261 1262 return ret; 1263 } 1264 1265 static int evlist__deliver_sample(struct evlist *evlist, const struct perf_tool *tool, 1266 union perf_event *event, struct perf_sample *sample, 1267 struct evsel *evsel, struct machine *machine) 1268 { 1269 /* We know evsel != NULL. */ 1270 u64 sample_type = evsel->core.attr.sample_type; 1271 u64 read_format = evsel->core.attr.read_format; 1272 bool per_thread = perf_evsel__attr_has_per_thread_sample_period(&evsel->core); 1273 1274 /* Standard sample delivery. */ 1275 if (!(sample_type & PERF_SAMPLE_READ)) 1276 return tool->sample(tool, event, sample, evsel, machine); 1277 1278 /* For PERF_SAMPLE_READ we have either single or group mode. */ 1279 if (read_format & PERF_FORMAT_GROUP) 1280 return deliver_sample_group(evlist, tool, event, sample, 1281 machine, read_format, per_thread); 1282 else 1283 return deliver_sample_value(evlist, tool, event, sample, 1284 &sample->read.one, machine, 1285 per_thread); 1286 } 1287 1288 /* 1289 * Samples with deferred callchains should wait for the next matching 1290 * PERF_RECORD_CALLCHAIN_RECORD entries. Keep the events in a list and 1291 * deliver them once it finds the callchains. 1292 */ 1293 struct deferred_event { 1294 struct list_head list; 1295 union perf_event *event; 1296 }; 1297 1298 /* 1299 * This is called when a deferred callchain record comes up. Find all matching 1300 * samples, merge the callchains and process them. 1301 */ 1302 static int evlist__deliver_deferred_callchain(struct evlist *evlist, 1303 const struct perf_tool *tool, 1304 union perf_event *event, 1305 struct perf_sample *sample, 1306 struct machine *machine) 1307 { 1308 struct deferred_event *de, *tmp; 1309 struct evsel *evsel; 1310 int ret = 0; 1311 1312 if (!tool->merge_deferred_callchains) { 1313 evsel = evlist__id2evsel(evlist, sample->id); 1314 return tool->callchain_deferred(tool, event, sample, 1315 evsel, machine); 1316 } 1317 1318 list_for_each_entry_safe(de, tmp, &evlist->deferred_samples, list) { 1319 struct perf_sample orig_sample; 1320 1321 ret = evlist__parse_sample(evlist, de->event, &orig_sample); 1322 if (ret < 0) { 1323 pr_err("failed to parse original sample\n"); 1324 break; 1325 } 1326 1327 if (sample->tid != orig_sample.tid) 1328 continue; 1329 1330 if (event->callchain_deferred.cookie == orig_sample.deferred_cookie) 1331 sample__merge_deferred_callchain(&orig_sample, sample); 1332 else 1333 orig_sample.deferred_callchain = false; 1334 1335 evsel = evlist__id2evsel(evlist, orig_sample.id); 1336 ret = evlist__deliver_sample(evlist, tool, de->event, 1337 &orig_sample, evsel, machine); 1338 1339 if (orig_sample.deferred_callchain) 1340 free(orig_sample.callchain); 1341 1342 list_del(&de->list); 1343 free(de->event); 1344 free(de); 1345 1346 if (ret) 1347 break; 1348 } 1349 return ret; 1350 } 1351 1352 /* 1353 * This is called at the end of the data processing for the session. Flush the 1354 * remaining samples as there's no hope for matching deferred callchains. 1355 */ 1356 static int session__flush_deferred_samples(struct perf_session *session, 1357 const struct perf_tool *tool) 1358 { 1359 struct evlist *evlist = session->evlist; 1360 struct machine *machine = &session->machines.host; 1361 struct deferred_event *de, *tmp; 1362 struct evsel *evsel; 1363 int ret = 0; 1364 1365 list_for_each_entry_safe(de, tmp, &evlist->deferred_samples, list) { 1366 struct perf_sample sample; 1367 1368 ret = evlist__parse_sample(evlist, de->event, &sample); 1369 if (ret < 0) { 1370 pr_err("failed to parse original sample\n"); 1371 break; 1372 } 1373 1374 evsel = evlist__id2evsel(evlist, sample.id); 1375 ret = evlist__deliver_sample(evlist, tool, de->event, 1376 &sample, evsel, machine); 1377 1378 list_del(&de->list); 1379 free(de->event); 1380 free(de); 1381 1382 if (ret) 1383 break; 1384 } 1385 return ret; 1386 } 1387 1388 static int machines__deliver_event(struct machines *machines, 1389 struct evlist *evlist, 1390 union perf_event *event, 1391 struct perf_sample *sample, 1392 const struct perf_tool *tool, u64 file_offset, 1393 const char *file_path) 1394 { 1395 struct evsel *evsel; 1396 struct machine *machine; 1397 1398 dump_event(evlist, event, file_offset, sample, file_path); 1399 1400 evsel = evlist__id2evsel(evlist, sample->id); 1401 1402 machine = machines__find_for_cpumode(machines, event, sample); 1403 1404 switch (event->header.type) { 1405 case PERF_RECORD_SAMPLE: 1406 if (evsel == NULL) { 1407 ++evlist->stats.nr_unknown_id; 1408 return 0; 1409 } 1410 if (machine == NULL) { 1411 ++evlist->stats.nr_unprocessable_samples; 1412 dump_sample(evsel, event, sample, perf_env__arch(NULL)); 1413 return 0; 1414 } 1415 dump_sample(evsel, event, sample, perf_env__arch(machine->env)); 1416 if (sample->deferred_callchain && tool->merge_deferred_callchains) { 1417 struct deferred_event *de = malloc(sizeof(*de)); 1418 size_t sz = event->header.size; 1419 1420 if (de == NULL) 1421 return -ENOMEM; 1422 1423 de->event = malloc(sz); 1424 if (de->event == NULL) { 1425 free(de); 1426 return -ENOMEM; 1427 } 1428 memcpy(de->event, event, sz); 1429 list_add_tail(&de->list, &evlist->deferred_samples); 1430 return 0; 1431 } 1432 return evlist__deliver_sample(evlist, tool, event, sample, evsel, machine); 1433 case PERF_RECORD_MMAP: 1434 return tool->mmap(tool, event, sample, machine); 1435 case PERF_RECORD_MMAP2: 1436 if (event->header.misc & PERF_RECORD_MISC_PROC_MAP_PARSE_TIMEOUT) 1437 ++evlist->stats.nr_proc_map_timeout; 1438 return tool->mmap2(tool, event, sample, machine); 1439 case PERF_RECORD_COMM: 1440 return tool->comm(tool, event, sample, machine); 1441 case PERF_RECORD_NAMESPACES: 1442 return tool->namespaces(tool, event, sample, machine); 1443 case PERF_RECORD_CGROUP: 1444 return tool->cgroup(tool, event, sample, machine); 1445 case PERF_RECORD_FORK: 1446 return tool->fork(tool, event, sample, machine); 1447 case PERF_RECORD_EXIT: 1448 return tool->exit(tool, event, sample, machine); 1449 case PERF_RECORD_LOST: 1450 if (tool->lost == perf_event__process_lost) 1451 evlist->stats.total_lost += event->lost.lost; 1452 return tool->lost(tool, event, sample, machine); 1453 case PERF_RECORD_LOST_SAMPLES: 1454 if (event->header.misc & PERF_RECORD_MISC_LOST_SAMPLES_BPF) 1455 evlist->stats.total_dropped_samples += event->lost_samples.lost; 1456 else if (tool->lost_samples == perf_event__process_lost_samples) 1457 evlist->stats.total_lost_samples += event->lost_samples.lost; 1458 return tool->lost_samples(tool, event, sample, machine); 1459 case PERF_RECORD_READ: 1460 dump_read(evsel, event); 1461 return tool->read(tool, event, sample, evsel, machine); 1462 case PERF_RECORD_THROTTLE: 1463 return tool->throttle(tool, event, sample, machine); 1464 case PERF_RECORD_UNTHROTTLE: 1465 return tool->unthrottle(tool, event, sample, machine); 1466 case PERF_RECORD_AUX: 1467 if (tool->aux == perf_event__process_aux) { 1468 if (event->aux.flags & PERF_AUX_FLAG_TRUNCATED) 1469 evlist->stats.total_aux_lost += 1; 1470 if (event->aux.flags & PERF_AUX_FLAG_PARTIAL) 1471 evlist->stats.total_aux_partial += 1; 1472 if (event->aux.flags & PERF_AUX_FLAG_COLLISION) 1473 evlist->stats.total_aux_collision += 1; 1474 } 1475 return tool->aux(tool, event, sample, machine); 1476 case PERF_RECORD_ITRACE_START: 1477 return tool->itrace_start(tool, event, sample, machine); 1478 case PERF_RECORD_SWITCH: 1479 case PERF_RECORD_SWITCH_CPU_WIDE: 1480 return tool->context_switch(tool, event, sample, machine); 1481 case PERF_RECORD_KSYMBOL: 1482 return tool->ksymbol(tool, event, sample, machine); 1483 case PERF_RECORD_BPF_EVENT: 1484 return tool->bpf(tool, event, sample, machine); 1485 case PERF_RECORD_TEXT_POKE: 1486 return tool->text_poke(tool, event, sample, machine); 1487 case PERF_RECORD_AUX_OUTPUT_HW_ID: 1488 return tool->aux_output_hw_id(tool, event, sample, machine); 1489 case PERF_RECORD_CALLCHAIN_DEFERRED: 1490 dump_deferred_callchain(evsel, event, sample); 1491 return evlist__deliver_deferred_callchain(evlist, tool, event, 1492 sample, machine); 1493 default: 1494 ++evlist->stats.nr_unknown_events; 1495 return -1; 1496 } 1497 } 1498 1499 static int perf_session__deliver_event(struct perf_session *session, 1500 union perf_event *event, 1501 const struct perf_tool *tool, 1502 u64 file_offset, 1503 const char *file_path) 1504 { 1505 struct perf_sample sample; 1506 int ret; 1507 1508 perf_sample__init(&sample, /*all=*/false); 1509 ret = evlist__parse_sample(session->evlist, event, &sample); 1510 if (ret) { 1511 pr_err("Can't parse sample, err = %d\n", ret); 1512 goto out; 1513 } 1514 1515 ret = auxtrace__process_event(session, event, &sample, tool); 1516 if (ret < 0) 1517 goto out; 1518 if (ret > 0) { 1519 ret = 0; 1520 goto out; 1521 } 1522 1523 ret = machines__deliver_event(&session->machines, session->evlist, 1524 event, &sample, tool, file_offset, file_path); 1525 1526 if (dump_trace && sample.aux_sample.size) 1527 auxtrace__dump_auxtrace_sample(session, &sample); 1528 out: 1529 perf_sample__exit(&sample); 1530 return ret; 1531 } 1532 1533 static s64 perf_session__process_user_event(struct perf_session *session, 1534 union perf_event *event, 1535 u64 file_offset, 1536 const char *file_path) 1537 { 1538 struct ordered_events *oe = &session->ordered_events; 1539 const struct perf_tool *tool = session->tool; 1540 struct perf_sample sample; 1541 int fd = perf_data__fd(session->data); 1542 s64 err; 1543 1544 perf_sample__init(&sample, /*all=*/true); 1545 if ((event->header.type != PERF_RECORD_COMPRESSED && 1546 event->header.type != PERF_RECORD_COMPRESSED2) || 1547 perf_tool__compressed_is_stub(tool)) 1548 dump_event(session->evlist, event, file_offset, &sample, file_path); 1549 1550 /* These events are processed right away */ 1551 switch (event->header.type) { 1552 case PERF_RECORD_HEADER_ATTR: 1553 err = tool->attr(tool, event, &session->evlist); 1554 if (err == 0) { 1555 perf_session__set_id_hdr_size(session); 1556 perf_session__set_comm_exec(session); 1557 } 1558 break; 1559 case PERF_RECORD_EVENT_UPDATE: 1560 err = tool->event_update(tool, event, &session->evlist); 1561 break; 1562 case PERF_RECORD_HEADER_EVENT_TYPE: 1563 /* 1564 * Deprecated, but we need to handle it for sake 1565 * of old data files create in pipe mode. 1566 */ 1567 err = 0; 1568 break; 1569 case PERF_RECORD_HEADER_TRACING_DATA: 1570 /* 1571 * Setup for reading amidst mmap, but only when we 1572 * are in 'file' mode. The 'pipe' fd is in proper 1573 * place already. 1574 */ 1575 if (!perf_data__is_pipe(session->data)) 1576 lseek(fd, file_offset, SEEK_SET); 1577 err = tool->tracing_data(tool, session, event); 1578 break; 1579 case PERF_RECORD_HEADER_BUILD_ID: 1580 err = tool->build_id(tool, session, event); 1581 break; 1582 case PERF_RECORD_FINISHED_ROUND: 1583 err = tool->finished_round(tool, event, oe); 1584 break; 1585 case PERF_RECORD_ID_INDEX: 1586 err = tool->id_index(tool, session, event); 1587 break; 1588 case PERF_RECORD_AUXTRACE_INFO: 1589 err = tool->auxtrace_info(tool, session, event); 1590 break; 1591 case PERF_RECORD_AUXTRACE: 1592 /* 1593 * Setup for reading amidst mmap, but only when we 1594 * are in 'file' mode. The 'pipe' fd is in proper 1595 * place already. 1596 */ 1597 if (!perf_data__is_pipe(session->data)) 1598 lseek(fd, file_offset + event->header.size, SEEK_SET); 1599 err = tool->auxtrace(tool, session, event); 1600 break; 1601 case PERF_RECORD_AUXTRACE_ERROR: 1602 perf_session__auxtrace_error_inc(session, event); 1603 err = tool->auxtrace_error(tool, session, event); 1604 break; 1605 case PERF_RECORD_THREAD_MAP: 1606 err = tool->thread_map(tool, session, event); 1607 break; 1608 case PERF_RECORD_CPU_MAP: 1609 err = tool->cpu_map(tool, session, event); 1610 break; 1611 case PERF_RECORD_STAT_CONFIG: 1612 err = tool->stat_config(tool, session, event); 1613 break; 1614 case PERF_RECORD_STAT: 1615 err = tool->stat(tool, session, event); 1616 break; 1617 case PERF_RECORD_STAT_ROUND: 1618 err = tool->stat_round(tool, session, event); 1619 break; 1620 case PERF_RECORD_TIME_CONV: 1621 session->time_conv = event->time_conv; 1622 err = tool->time_conv(tool, session, event); 1623 break; 1624 case PERF_RECORD_HEADER_FEATURE: 1625 err = tool->feature(tool, session, event); 1626 break; 1627 case PERF_RECORD_COMPRESSED: 1628 case PERF_RECORD_COMPRESSED2: 1629 err = tool->compressed(tool, session, event, file_offset, file_path); 1630 if (err) 1631 dump_event(session->evlist, event, file_offset, &sample, file_path); 1632 break; 1633 case PERF_RECORD_FINISHED_INIT: 1634 err = tool->finished_init(tool, session, event); 1635 break; 1636 case PERF_RECORD_BPF_METADATA: 1637 err = tool->bpf_metadata(tool, session, event); 1638 break; 1639 default: 1640 err = -EINVAL; 1641 break; 1642 } 1643 perf_sample__exit(&sample); 1644 return err; 1645 } 1646 1647 int perf_session__deliver_synth_event(struct perf_session *session, 1648 union perf_event *event, 1649 struct perf_sample *sample) 1650 { 1651 struct evlist *evlist = session->evlist; 1652 const struct perf_tool *tool = session->tool; 1653 1654 events_stats__inc(&evlist->stats, event->header.type); 1655 1656 if (event->header.type >= PERF_RECORD_USER_TYPE_START) 1657 return perf_session__process_user_event(session, event, 0, NULL); 1658 1659 return machines__deliver_event(&session->machines, evlist, event, sample, tool, 0, NULL); 1660 } 1661 1662 int perf_session__deliver_synth_attr_event(struct perf_session *session, 1663 const struct perf_event_attr *attr, 1664 u64 id) 1665 { 1666 union { 1667 struct { 1668 struct perf_record_header_attr attr; 1669 u64 ids[1]; 1670 } attr_id; 1671 union perf_event ev; 1672 } ev = { 1673 .attr_id.attr.header.type = PERF_RECORD_HEADER_ATTR, 1674 .attr_id.attr.header.size = sizeof(ev.attr_id), 1675 .attr_id.ids[0] = id, 1676 }; 1677 1678 if (attr->size != sizeof(ev.attr_id.attr.attr)) { 1679 pr_debug("Unexpected perf_event_attr size\n"); 1680 return -EINVAL; 1681 } 1682 ev.attr_id.attr.attr = *attr; 1683 return perf_session__deliver_synth_event(session, &ev.ev, NULL); 1684 } 1685 1686 static void event_swap(union perf_event *event, bool sample_id_all) 1687 { 1688 perf_event__swap_op swap; 1689 1690 swap = perf_event__swap_ops[event->header.type]; 1691 if (swap) 1692 swap(event, sample_id_all); 1693 } 1694 1695 int perf_session__peek_event(struct perf_session *session, off_t file_offset, 1696 void *buf, size_t buf_sz, 1697 union perf_event **event_ptr, 1698 struct perf_sample *sample) 1699 { 1700 union perf_event *event; 1701 size_t hdr_sz, rest; 1702 int fd; 1703 1704 if (session->one_mmap && !session->header.needs_swap) { 1705 event = file_offset - session->one_mmap_offset + 1706 session->one_mmap_addr; 1707 goto out_parse_sample; 1708 } 1709 1710 if (perf_data__is_pipe(session->data)) 1711 return -1; 1712 1713 fd = perf_data__fd(session->data); 1714 hdr_sz = sizeof(struct perf_event_header); 1715 1716 if (buf_sz < hdr_sz) 1717 return -1; 1718 1719 if (lseek(fd, file_offset, SEEK_SET) == (off_t)-1 || 1720 readn(fd, buf, hdr_sz) != (ssize_t)hdr_sz) 1721 return -1; 1722 1723 event = (union perf_event *)buf; 1724 1725 if (session->header.needs_swap) 1726 perf_event_header__bswap(&event->header); 1727 1728 if (event->header.size < hdr_sz || event->header.size > buf_sz) 1729 return -1; 1730 1731 buf += hdr_sz; 1732 rest = event->header.size - hdr_sz; 1733 1734 if (readn(fd, buf, rest) != (ssize_t)rest) 1735 return -1; 1736 1737 if (session->header.needs_swap) 1738 event_swap(event, evlist__sample_id_all(session->evlist)); 1739 1740 out_parse_sample: 1741 1742 if (sample && event->header.type < PERF_RECORD_USER_TYPE_START && 1743 evlist__parse_sample(session->evlist, event, sample)) 1744 return -1; 1745 1746 *event_ptr = event; 1747 1748 return 0; 1749 } 1750 1751 int perf_session__peek_events(struct perf_session *session, u64 offset, 1752 u64 size, peek_events_cb_t cb, void *data) 1753 { 1754 u64 max_offset = offset + size; 1755 char buf[PERF_SAMPLE_MAX_SIZE]; 1756 union perf_event *event; 1757 int err; 1758 1759 do { 1760 err = perf_session__peek_event(session, offset, buf, 1761 PERF_SAMPLE_MAX_SIZE, &event, 1762 NULL); 1763 if (err) 1764 return err; 1765 1766 err = cb(session, event, offset, data); 1767 if (err) 1768 return err; 1769 1770 offset += event->header.size; 1771 if (event->header.type == PERF_RECORD_AUXTRACE) 1772 offset += event->auxtrace.size; 1773 1774 } while (offset < max_offset); 1775 1776 return err; 1777 } 1778 1779 static s64 perf_session__process_event(struct perf_session *session, 1780 union perf_event *event, u64 file_offset, 1781 const char *file_path) 1782 { 1783 struct evlist *evlist = session->evlist; 1784 const struct perf_tool *tool = session->tool; 1785 int ret; 1786 1787 if (session->header.needs_swap) 1788 event_swap(event, evlist__sample_id_all(evlist)); 1789 1790 if (event->header.type >= PERF_RECORD_HEADER_MAX) { 1791 /* perf should not support unaligned event, stop here. */ 1792 if (event->header.size % sizeof(u64)) 1793 return -EINVAL; 1794 1795 /* This perf is outdated and does not support the latest event type. */ 1796 ui__warning("Unsupported header type %u, please consider updating perf.\n", 1797 event->header.type); 1798 /* Skip unsupported event by returning its size. */ 1799 return event->header.size; 1800 } 1801 1802 events_stats__inc(&evlist->stats, event->header.type); 1803 1804 if (event->header.type >= PERF_RECORD_USER_TYPE_START) 1805 return perf_session__process_user_event(session, event, file_offset, file_path); 1806 1807 if (tool->ordered_events) { 1808 u64 timestamp = -1ULL; 1809 1810 ret = evlist__parse_sample_timestamp(evlist, event, ×tamp); 1811 if (ret && ret != -1) 1812 return ret; 1813 1814 ret = perf_session__queue_event(session, event, timestamp, file_offset, file_path); 1815 if (ret != -ETIME) 1816 return ret; 1817 } 1818 1819 return perf_session__deliver_event(session, event, tool, file_offset, file_path); 1820 } 1821 1822 void perf_event_header__bswap(struct perf_event_header *hdr) 1823 { 1824 hdr->type = bswap_32(hdr->type); 1825 hdr->misc = bswap_16(hdr->misc); 1826 hdr->size = bswap_16(hdr->size); 1827 } 1828 1829 struct thread *perf_session__findnew(struct perf_session *session, pid_t pid) 1830 { 1831 return machine__findnew_thread(&session->machines.host, -1, pid); 1832 } 1833 1834 int perf_session__register_idle_thread(struct perf_session *session) 1835 { 1836 struct thread *thread = machine__idle_thread(&session->machines.host); 1837 1838 /* machine__idle_thread() got the thread, so put it */ 1839 thread__put(thread); 1840 return thread ? 0 : -1; 1841 } 1842 1843 static void 1844 perf_session__warn_order(const struct perf_session *session) 1845 { 1846 const struct ordered_events *oe = &session->ordered_events; 1847 struct evsel *evsel; 1848 bool should_warn = true; 1849 1850 evlist__for_each_entry(session->evlist, evsel) { 1851 if (evsel->core.attr.write_backward) 1852 should_warn = false; 1853 } 1854 1855 if (!should_warn) 1856 return; 1857 if (oe->nr_unordered_events != 0) 1858 ui__warning("%u out of order events recorded.\n", oe->nr_unordered_events); 1859 } 1860 1861 static void perf_session__warn_about_errors(const struct perf_session *session) 1862 { 1863 const struct events_stats *stats = &session->evlist->stats; 1864 1865 if (session->tool->lost == perf_event__process_lost && 1866 stats->nr_events[PERF_RECORD_LOST] != 0) { 1867 ui__warning("Processed %d events and lost %d chunks!\n\n" 1868 "Check IO/CPU overload!\n\n", 1869 stats->nr_events[0], 1870 stats->nr_events[PERF_RECORD_LOST]); 1871 } 1872 1873 if (session->tool->lost_samples == perf_event__process_lost_samples) { 1874 double drop_rate; 1875 1876 drop_rate = (double)stats->total_lost_samples / 1877 (double) (stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples); 1878 if (drop_rate > 0.05) { 1879 ui__warning("Processed %" PRIu64 " samples and lost %3.2f%%!\n\n", 1880 stats->nr_events[PERF_RECORD_SAMPLE] + stats->total_lost_samples, 1881 drop_rate * 100.0); 1882 } 1883 } 1884 1885 if (session->tool->aux == perf_event__process_aux && 1886 stats->total_aux_lost != 0) { 1887 ui__warning("AUX data lost %" PRIu64 " times out of %u!\n\n", 1888 stats->total_aux_lost, 1889 stats->nr_events[PERF_RECORD_AUX]); 1890 } 1891 1892 if (session->tool->aux == perf_event__process_aux && 1893 stats->total_aux_partial != 0) { 1894 bool vmm_exclusive = false; 1895 1896 (void)sysfs__read_bool("module/kvm_intel/parameters/vmm_exclusive", 1897 &vmm_exclusive); 1898 1899 ui__warning("AUX data had gaps in it %" PRIu64 " times out of %u!\n\n" 1900 "Are you running a KVM guest in the background?%s\n\n", 1901 stats->total_aux_partial, 1902 stats->nr_events[PERF_RECORD_AUX], 1903 vmm_exclusive ? 1904 "\nReloading kvm_intel module with vmm_exclusive=0\n" 1905 "will reduce the gaps to only guest's timeslices." : 1906 ""); 1907 } 1908 1909 if (session->tool->aux == perf_event__process_aux && 1910 stats->total_aux_collision != 0) { 1911 ui__warning("AUX data detected collision %" PRIu64 " times out of %u!\n\n", 1912 stats->total_aux_collision, 1913 stats->nr_events[PERF_RECORD_AUX]); 1914 } 1915 1916 if (stats->nr_unknown_events != 0) { 1917 ui__warning("Found %u unknown events!\n\n" 1918 "Is this an older tool processing a perf.data " 1919 "file generated by a more recent tool?\n\n" 1920 "If that is not the case, consider " 1921 "reporting to linux-kernel@vger.kernel.org.\n\n", 1922 stats->nr_unknown_events); 1923 } 1924 1925 if (stats->nr_unknown_id != 0) { 1926 ui__warning("%u samples with id not present in the header\n", 1927 stats->nr_unknown_id); 1928 } 1929 1930 if (stats->nr_invalid_chains != 0) { 1931 ui__warning("Found invalid callchains!\n\n" 1932 "%u out of %u events were discarded for this reason.\n\n" 1933 "Consider reporting to linux-kernel@vger.kernel.org.\n\n", 1934 stats->nr_invalid_chains, 1935 stats->nr_events[PERF_RECORD_SAMPLE]); 1936 } 1937 1938 if (stats->nr_unprocessable_samples != 0) { 1939 ui__warning("%u unprocessable samples recorded.\n" 1940 "Do you have a KVM guest running and not using 'perf kvm'?\n", 1941 stats->nr_unprocessable_samples); 1942 } 1943 1944 perf_session__warn_order(session); 1945 1946 events_stats__auxtrace_error_warn(stats); 1947 1948 if (stats->nr_proc_map_timeout != 0) { 1949 ui__warning("%d map information files for pre-existing threads were\n" 1950 "not processed, if there are samples for addresses they\n" 1951 "will not be resolved, you may find out which are these\n" 1952 "threads by running with -v and redirecting the output\n" 1953 "to a file.\n" 1954 "The time limit to process proc map is too short?\n" 1955 "Increase it by --proc-map-timeout\n", 1956 stats->nr_proc_map_timeout); 1957 } 1958 } 1959 1960 static int perf_session__flush_thread_stack(struct thread *thread, 1961 void *p __maybe_unused) 1962 { 1963 return thread_stack__flush(thread); 1964 } 1965 1966 static int perf_session__flush_thread_stacks(struct perf_session *session) 1967 { 1968 return machines__for_each_thread(&session->machines, 1969 perf_session__flush_thread_stack, 1970 NULL); 1971 } 1972 1973 volatile sig_atomic_t session_done; 1974 1975 static int __perf_session__process_decomp_events(struct perf_session *session); 1976 1977 static int __perf_session__process_pipe_events(struct perf_session *session) 1978 { 1979 struct ordered_events *oe = &session->ordered_events; 1980 const struct perf_tool *tool = session->tool; 1981 struct ui_progress prog; 1982 union perf_event *event; 1983 uint32_t size, cur_size = 0; 1984 void *buf = NULL; 1985 s64 skip = 0; 1986 u64 head; 1987 ssize_t err; 1988 void *p; 1989 bool update_prog = false; 1990 1991 /* 1992 * If it's from a file saving pipe data (by redirection), it would have 1993 * a file name other than "-". Then we can get the total size and show 1994 * the progress. 1995 */ 1996 if (strcmp(session->data->path, "-") && session->data->file.size) { 1997 ui_progress__init_size(&prog, session->data->file.size, 1998 "Processing events..."); 1999 update_prog = true; 2000 } 2001 2002 head = 0; 2003 cur_size = sizeof(union perf_event); 2004 2005 buf = malloc(cur_size); 2006 if (!buf) 2007 return -errno; 2008 ordered_events__set_copy_on_queue(oe, true); 2009 more: 2010 event = buf; 2011 err = perf_data__read(session->data, event, 2012 sizeof(struct perf_event_header)); 2013 if (err <= 0) { 2014 if (err == 0) 2015 goto done; 2016 2017 pr_err("failed to read event header\n"); 2018 goto out_err; 2019 } 2020 2021 if (session->header.needs_swap) 2022 perf_event_header__bswap(&event->header); 2023 2024 size = event->header.size; 2025 if (size < sizeof(struct perf_event_header)) { 2026 pr_err("bad event header size\n"); 2027 goto out_err; 2028 } 2029 2030 if (size > cur_size) { 2031 void *new = realloc(buf, size); 2032 if (!new) { 2033 pr_err("failed to allocate memory to read event\n"); 2034 goto out_err; 2035 } 2036 buf = new; 2037 cur_size = size; 2038 event = buf; 2039 } 2040 p = event; 2041 p += sizeof(struct perf_event_header); 2042 2043 if (size - sizeof(struct perf_event_header)) { 2044 err = perf_data__read(session->data, p, 2045 size - sizeof(struct perf_event_header)); 2046 if (err <= 0) { 2047 if (err == 0) { 2048 pr_err("unexpected end of event stream\n"); 2049 goto done; 2050 } 2051 2052 pr_err("failed to read event data\n"); 2053 goto out_err; 2054 } 2055 } 2056 2057 if ((skip = perf_session__process_event(session, event, head, "pipe")) < 0) { 2058 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n", 2059 head, event->header.size, event->header.type); 2060 err = -EINVAL; 2061 goto out_err; 2062 } 2063 2064 head += size; 2065 2066 if (skip > 0) 2067 head += skip; 2068 2069 err = __perf_session__process_decomp_events(session); 2070 if (err) 2071 goto out_err; 2072 2073 if (update_prog) 2074 ui_progress__update(&prog, size); 2075 2076 if (!session_done()) 2077 goto more; 2078 done: 2079 /* do the final flush for ordered samples */ 2080 err = ordered_events__flush(oe, OE_FLUSH__FINAL); 2081 if (err) 2082 goto out_err; 2083 err = session__flush_deferred_samples(session, tool); 2084 if (err) 2085 goto out_err; 2086 err = auxtrace__flush_events(session, tool); 2087 if (err) 2088 goto out_err; 2089 err = perf_session__flush_thread_stacks(session); 2090 out_err: 2091 free(buf); 2092 if (update_prog) 2093 ui_progress__finish(); 2094 if (!tool->no_warn) 2095 perf_session__warn_about_errors(session); 2096 ordered_events__free(&session->ordered_events); 2097 auxtrace__free_events(session); 2098 return err; 2099 } 2100 2101 static union perf_event * 2102 prefetch_event(char *buf, u64 head, size_t mmap_size, 2103 bool needs_swap, union perf_event *error) 2104 { 2105 union perf_event *event; 2106 u16 event_size; 2107 2108 /* 2109 * Ensure we have enough space remaining to read 2110 * the size of the event in the headers. 2111 */ 2112 if (head + sizeof(event->header) > mmap_size) 2113 return NULL; 2114 2115 event = (union perf_event *)(buf + head); 2116 if (needs_swap) 2117 perf_event_header__bswap(&event->header); 2118 2119 event_size = event->header.size; 2120 if (head + event_size <= mmap_size) 2121 return event; 2122 2123 /* We're not fetching the event so swap back again */ 2124 if (needs_swap) 2125 perf_event_header__bswap(&event->header); 2126 2127 /* Check if the event fits into the next mmapped buf. */ 2128 if (event_size <= mmap_size - head % page_size) { 2129 /* Remap buf and fetch again. */ 2130 return NULL; 2131 } 2132 2133 /* Invalid input. Event size should never exceed mmap_size. */ 2134 pr_debug("%s: head=%#" PRIx64 " event->header.size=%#x, mmap_size=%#zx:" 2135 " fuzzed or compressed perf.data?\n", __func__, head, event_size, mmap_size); 2136 2137 return error; 2138 } 2139 2140 static union perf_event * 2141 fetch_mmaped_event(u64 head, size_t mmap_size, char *buf, bool needs_swap) 2142 { 2143 return prefetch_event(buf, head, mmap_size, needs_swap, ERR_PTR(-EINVAL)); 2144 } 2145 2146 static union perf_event * 2147 fetch_decomp_event(u64 head, size_t mmap_size, char *buf, bool needs_swap) 2148 { 2149 return prefetch_event(buf, head, mmap_size, needs_swap, NULL); 2150 } 2151 2152 static int __perf_session__process_decomp_events(struct perf_session *session) 2153 { 2154 s64 skip; 2155 u64 size; 2156 struct decomp *decomp = session->active_decomp->decomp_last; 2157 2158 if (!decomp) 2159 return 0; 2160 2161 while (decomp->head < decomp->size && !session_done()) { 2162 union perf_event *event = fetch_decomp_event(decomp->head, decomp->size, decomp->data, 2163 session->header.needs_swap); 2164 2165 if (!event) 2166 break; 2167 2168 size = event->header.size; 2169 2170 if (size < sizeof(struct perf_event_header) || 2171 (skip = perf_session__process_event(session, event, decomp->file_pos, 2172 decomp->file_path)) < 0) { 2173 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d\n", 2174 decomp->file_pos + decomp->head, event->header.size, event->header.type); 2175 return -EINVAL; 2176 } 2177 2178 if (skip) 2179 size += skip; 2180 2181 decomp->head += size; 2182 } 2183 2184 return 0; 2185 } 2186 2187 /* 2188 * On 64bit we can mmap the data file in one go. No need for tiny mmap 2189 * slices. On 32bit we use 32MB. 2190 */ 2191 #if BITS_PER_LONG == 64 2192 #define MMAP_SIZE ULLONG_MAX 2193 #define NUM_MMAPS 1 2194 #else 2195 #define MMAP_SIZE (32 * 1024 * 1024ULL) 2196 #define NUM_MMAPS 128 2197 #endif 2198 2199 struct reader; 2200 2201 typedef s64 (*reader_cb_t)(struct perf_session *session, 2202 union perf_event *event, 2203 u64 file_offset, 2204 const char *file_path); 2205 2206 struct reader { 2207 int fd; 2208 const char *path; 2209 u64 data_size; 2210 u64 data_offset; 2211 reader_cb_t process; 2212 bool in_place_update; 2213 char *mmaps[NUM_MMAPS]; 2214 size_t mmap_size; 2215 int mmap_idx; 2216 char *mmap_cur; 2217 u64 file_pos; 2218 u64 file_offset; 2219 u64 head; 2220 u64 size; 2221 bool done; 2222 struct zstd_data zstd_data; 2223 struct decomp_data decomp_data; 2224 }; 2225 2226 static int 2227 reader__init(struct reader *rd, bool *one_mmap) 2228 { 2229 u64 data_size = rd->data_size; 2230 char **mmaps = rd->mmaps; 2231 2232 rd->head = rd->data_offset; 2233 data_size += rd->data_offset; 2234 2235 rd->mmap_size = MMAP_SIZE; 2236 if (rd->mmap_size > data_size) { 2237 rd->mmap_size = data_size; 2238 if (one_mmap) 2239 *one_mmap = true; 2240 } 2241 2242 memset(mmaps, 0, sizeof(rd->mmaps)); 2243 2244 if (zstd_init(&rd->zstd_data, 0)) 2245 return -1; 2246 rd->decomp_data.zstd_decomp = &rd->zstd_data; 2247 2248 return 0; 2249 } 2250 2251 static void 2252 reader__release_decomp(struct reader *rd) 2253 { 2254 perf_decomp__release_events(rd->decomp_data.decomp); 2255 zstd_fini(&rd->zstd_data); 2256 } 2257 2258 static int 2259 reader__mmap(struct reader *rd, struct perf_session *session) 2260 { 2261 int mmap_prot, mmap_flags; 2262 char *buf, **mmaps = rd->mmaps; 2263 u64 page_offset; 2264 2265 mmap_prot = PROT_READ; 2266 mmap_flags = MAP_SHARED; 2267 2268 if (rd->in_place_update) { 2269 mmap_prot |= PROT_WRITE; 2270 } else if (session->header.needs_swap) { 2271 mmap_prot |= PROT_WRITE; 2272 mmap_flags = MAP_PRIVATE; 2273 } 2274 2275 if (mmaps[rd->mmap_idx]) { 2276 munmap(mmaps[rd->mmap_idx], rd->mmap_size); 2277 mmaps[rd->mmap_idx] = NULL; 2278 } 2279 2280 page_offset = page_size * (rd->head / page_size); 2281 rd->file_offset += page_offset; 2282 rd->head -= page_offset; 2283 2284 buf = mmap(NULL, rd->mmap_size, mmap_prot, mmap_flags, rd->fd, 2285 rd->file_offset); 2286 if (buf == MAP_FAILED) { 2287 pr_err("failed to mmap file\n"); 2288 return -errno; 2289 } 2290 mmaps[rd->mmap_idx] = rd->mmap_cur = buf; 2291 rd->mmap_idx = (rd->mmap_idx + 1) & (ARRAY_SIZE(rd->mmaps) - 1); 2292 rd->file_pos = rd->file_offset + rd->head; 2293 if (session->one_mmap) { 2294 session->one_mmap_addr = buf; 2295 session->one_mmap_offset = rd->file_offset; 2296 } 2297 2298 return 0; 2299 } 2300 2301 enum { 2302 READER_OK, 2303 READER_NODATA, 2304 }; 2305 2306 static int 2307 reader__read_event(struct reader *rd, struct perf_session *session, 2308 struct ui_progress *prog) 2309 { 2310 u64 size; 2311 int err = READER_OK; 2312 union perf_event *event; 2313 s64 skip; 2314 2315 event = fetch_mmaped_event(rd->head, rd->mmap_size, rd->mmap_cur, 2316 session->header.needs_swap); 2317 if (IS_ERR(event)) 2318 return PTR_ERR(event); 2319 2320 if (!event) 2321 return READER_NODATA; 2322 2323 size = event->header.size; 2324 2325 skip = -EINVAL; 2326 2327 if (size < sizeof(struct perf_event_header) || 2328 (skip = rd->process(session, event, rd->file_pos, rd->path)) < 0) { 2329 pr_err("%#" PRIx64 " [%#x]: failed to process type: %d [%s]\n", 2330 rd->file_offset + rd->head, event->header.size, 2331 event->header.type, strerror(-skip)); 2332 err = skip; 2333 goto out; 2334 } 2335 2336 if (skip) 2337 size += skip; 2338 2339 rd->size += size; 2340 rd->head += size; 2341 rd->file_pos += size; 2342 2343 err = __perf_session__process_decomp_events(session); 2344 if (err) 2345 goto out; 2346 2347 ui_progress__update(prog, size); 2348 2349 out: 2350 return err; 2351 } 2352 2353 static inline bool 2354 reader__eof(struct reader *rd) 2355 { 2356 return (rd->file_pos >= rd->data_size + rd->data_offset); 2357 } 2358 2359 static int 2360 reader__process_events(struct reader *rd, struct perf_session *session, 2361 struct ui_progress *prog) 2362 { 2363 int err; 2364 2365 err = reader__init(rd, &session->one_mmap); 2366 if (err) 2367 goto out; 2368 2369 session->active_decomp = &rd->decomp_data; 2370 2371 remap: 2372 err = reader__mmap(rd, session); 2373 if (err) 2374 goto out; 2375 2376 more: 2377 err = reader__read_event(rd, session, prog); 2378 if (err < 0) 2379 goto out; 2380 else if (err == READER_NODATA) 2381 goto remap; 2382 2383 if (session_done()) 2384 goto out; 2385 2386 if (!reader__eof(rd)) 2387 goto more; 2388 2389 out: 2390 session->active_decomp = &session->decomp_data; 2391 return err; 2392 } 2393 2394 static s64 process_simple(struct perf_session *session, 2395 union perf_event *event, 2396 u64 file_offset, 2397 const char *file_path) 2398 { 2399 return perf_session__process_event(session, event, file_offset, file_path); 2400 } 2401 2402 static int __perf_session__process_events(struct perf_session *session) 2403 { 2404 struct reader rd = { 2405 .fd = perf_data__fd(session->data), 2406 .path = session->data->file.path, 2407 .data_size = session->header.data_size, 2408 .data_offset = session->header.data_offset, 2409 .process = process_simple, 2410 .in_place_update = session->data->in_place_update, 2411 }; 2412 struct ordered_events *oe = &session->ordered_events; 2413 const struct perf_tool *tool = session->tool; 2414 struct ui_progress prog; 2415 int err; 2416 2417 if (rd.data_size == 0) 2418 return -1; 2419 2420 ui_progress__init_size(&prog, rd.data_size, "Processing events..."); 2421 2422 err = reader__process_events(&rd, session, &prog); 2423 if (err) 2424 goto out_err; 2425 /* do the final flush for ordered samples */ 2426 err = ordered_events__flush(oe, OE_FLUSH__FINAL); 2427 if (err) 2428 goto out_err; 2429 err = auxtrace__flush_events(session, tool); 2430 if (err) 2431 goto out_err; 2432 err = session__flush_deferred_samples(session, tool); 2433 if (err) 2434 goto out_err; 2435 err = perf_session__flush_thread_stacks(session); 2436 out_err: 2437 ui_progress__finish(); 2438 if (!tool->no_warn) 2439 perf_session__warn_about_errors(session); 2440 /* 2441 * We may switching perf.data output, make ordered_events 2442 * reusable. 2443 */ 2444 ordered_events__reinit(&session->ordered_events); 2445 auxtrace__free_events(session); 2446 reader__release_decomp(&rd); 2447 session->one_mmap = false; 2448 return err; 2449 } 2450 2451 /* 2452 * Processing 2 MB of data from each reader in sequence, 2453 * because that's the way the ordered events sorting works 2454 * most efficiently. 2455 */ 2456 #define READER_MAX_SIZE (2 * 1024 * 1024) 2457 2458 /* 2459 * This function reads, merge and process directory data. 2460 * It assumens the version 1 of directory data, where each 2461 * data file holds per-cpu data, already sorted by kernel. 2462 */ 2463 static int __perf_session__process_dir_events(struct perf_session *session) 2464 { 2465 struct perf_data *data = session->data; 2466 const struct perf_tool *tool = session->tool; 2467 int i, ret, readers, nr_readers; 2468 struct ui_progress prog; 2469 u64 total_size = perf_data__size(session->data); 2470 struct reader *rd; 2471 2472 ui_progress__init_size(&prog, total_size, "Processing events..."); 2473 2474 nr_readers = 1; 2475 for (i = 0; i < data->dir.nr; i++) { 2476 if (data->dir.files[i].size) 2477 nr_readers++; 2478 } 2479 2480 rd = zalloc(nr_readers * sizeof(struct reader)); 2481 if (!rd) 2482 return -ENOMEM; 2483 2484 rd[0] = (struct reader) { 2485 .fd = perf_data__fd(session->data), 2486 .path = session->data->file.path, 2487 .data_size = session->header.data_size, 2488 .data_offset = session->header.data_offset, 2489 .process = process_simple, 2490 .in_place_update = session->data->in_place_update, 2491 }; 2492 ret = reader__init(&rd[0], NULL); 2493 if (ret) 2494 goto out_err; 2495 ret = reader__mmap(&rd[0], session); 2496 if (ret) 2497 goto out_err; 2498 readers = 1; 2499 2500 for (i = 0; i < data->dir.nr; i++) { 2501 if (!data->dir.files[i].size) 2502 continue; 2503 rd[readers] = (struct reader) { 2504 .fd = data->dir.files[i].fd, 2505 .path = data->dir.files[i].path, 2506 .data_size = data->dir.files[i].size, 2507 .data_offset = 0, 2508 .process = process_simple, 2509 .in_place_update = session->data->in_place_update, 2510 }; 2511 ret = reader__init(&rd[readers], NULL); 2512 if (ret) 2513 goto out_err; 2514 ret = reader__mmap(&rd[readers], session); 2515 if (ret) 2516 goto out_err; 2517 readers++; 2518 } 2519 2520 i = 0; 2521 while (readers) { 2522 if (session_done()) 2523 break; 2524 2525 if (rd[i].done) { 2526 i = (i + 1) % nr_readers; 2527 continue; 2528 } 2529 if (reader__eof(&rd[i])) { 2530 rd[i].done = true; 2531 readers--; 2532 continue; 2533 } 2534 2535 session->active_decomp = &rd[i].decomp_data; 2536 ret = reader__read_event(&rd[i], session, &prog); 2537 if (ret < 0) { 2538 goto out_err; 2539 } else if (ret == READER_NODATA) { 2540 ret = reader__mmap(&rd[i], session); 2541 if (ret) 2542 goto out_err; 2543 } 2544 2545 if (rd[i].size >= READER_MAX_SIZE) { 2546 rd[i].size = 0; 2547 i = (i + 1) % nr_readers; 2548 } 2549 } 2550 2551 ret = ordered_events__flush(&session->ordered_events, OE_FLUSH__FINAL); 2552 if (ret) 2553 goto out_err; 2554 2555 ret = session__flush_deferred_samples(session, tool); 2556 if (ret) 2557 goto out_err; 2558 2559 ret = perf_session__flush_thread_stacks(session); 2560 out_err: 2561 ui_progress__finish(); 2562 2563 if (!tool->no_warn) 2564 perf_session__warn_about_errors(session); 2565 2566 /* 2567 * We may switching perf.data output, make ordered_events 2568 * reusable. 2569 */ 2570 ordered_events__reinit(&session->ordered_events); 2571 2572 session->one_mmap = false; 2573 2574 session->active_decomp = &session->decomp_data; 2575 for (i = 0; i < nr_readers; i++) 2576 reader__release_decomp(&rd[i]); 2577 zfree(&rd); 2578 2579 return ret; 2580 } 2581 2582 int perf_session__process_events(struct perf_session *session) 2583 { 2584 if (perf_session__register_idle_thread(session) < 0) 2585 return -ENOMEM; 2586 2587 if (perf_data__is_pipe(session->data)) 2588 return __perf_session__process_pipe_events(session); 2589 2590 if (perf_data__is_dir(session->data) && session->data->dir.nr) 2591 return __perf_session__process_dir_events(session); 2592 2593 return __perf_session__process_events(session); 2594 } 2595 2596 bool perf_session__has_traces(struct perf_session *session, const char *msg) 2597 { 2598 struct evsel *evsel; 2599 2600 evlist__for_each_entry(session->evlist, evsel) { 2601 if (evsel->core.attr.type == PERF_TYPE_TRACEPOINT) 2602 return true; 2603 } 2604 2605 pr_err("No trace sample to read. Did you call 'perf %s'?\n", msg); 2606 return false; 2607 } 2608 2609 bool perf_session__has_switch_events(struct perf_session *session) 2610 { 2611 struct evsel *evsel; 2612 2613 evlist__for_each_entry(session->evlist, evsel) { 2614 if (evsel->core.attr.context_switch) 2615 return true; 2616 } 2617 2618 return false; 2619 } 2620 2621 int map__set_kallsyms_ref_reloc_sym(struct map *map, const char *symbol_name, u64 addr) 2622 { 2623 char *bracket; 2624 struct ref_reloc_sym *ref; 2625 struct kmap *kmap; 2626 2627 ref = zalloc(sizeof(struct ref_reloc_sym)); 2628 if (ref == NULL) 2629 return -ENOMEM; 2630 2631 ref->name = strdup(symbol_name); 2632 if (ref->name == NULL) { 2633 free(ref); 2634 return -ENOMEM; 2635 } 2636 2637 bracket = strchr(ref->name, ']'); 2638 if (bracket) 2639 *bracket = '\0'; 2640 2641 ref->addr = addr; 2642 2643 kmap = map__kmap(map); 2644 if (kmap) 2645 kmap->ref_reloc_sym = ref; 2646 2647 return 0; 2648 } 2649 2650 size_t perf_session__fprintf_dsos(struct perf_session *session, FILE *fp) 2651 { 2652 return machines__fprintf_dsos(&session->machines, fp); 2653 } 2654 2655 size_t perf_session__fprintf_dsos_buildid(struct perf_session *session, FILE *fp, 2656 bool (skip)(struct dso *dso, int parm), int parm) 2657 { 2658 return machines__fprintf_dsos_buildid(&session->machines, fp, skip, parm); 2659 } 2660 2661 size_t perf_session__fprintf_nr_events(struct perf_session *session, FILE *fp) 2662 { 2663 size_t ret; 2664 const char *msg = ""; 2665 2666 if (perf_header__has_feat(&session->header, HEADER_AUXTRACE)) 2667 msg = " (excludes AUX area (e.g. instruction trace) decoded / synthesized events)"; 2668 2669 ret = fprintf(fp, "\nAggregated stats:%s\n", msg); 2670 2671 ret += events_stats__fprintf(&session->evlist->stats, fp); 2672 return ret; 2673 } 2674 2675 size_t perf_session__fprintf(struct perf_session *session, FILE *fp) 2676 { 2677 /* 2678 * FIXME: Here we have to actually print all the machines in this 2679 * session, not just the host... 2680 */ 2681 return machine__fprintf(&session->machines.host, fp); 2682 } 2683 2684 void perf_session__dump_kmaps(struct perf_session *session) 2685 { 2686 int save_verbose = verbose; 2687 2688 fflush(stdout); 2689 fprintf(stderr, "Kernel and module maps:\n"); 2690 verbose = 0; /* Suppress verbose to print a summary only */ 2691 maps__fprintf(machine__kernel_maps(&session->machines.host), stderr); 2692 verbose = save_verbose; 2693 } 2694 2695 struct evsel *perf_session__find_first_evtype(struct perf_session *session, 2696 unsigned int type) 2697 { 2698 struct evsel *pos; 2699 2700 evlist__for_each_entry(session->evlist, pos) { 2701 if (pos->core.attr.type == type) 2702 return pos; 2703 } 2704 return NULL; 2705 } 2706 2707 int perf_session__cpu_bitmap(struct perf_session *session, 2708 const char *cpu_list, unsigned long *cpu_bitmap) 2709 { 2710 int i, err = -1; 2711 struct perf_cpu_map *map; 2712 int nr_cpus = min(perf_session__env(session)->nr_cpus_avail, MAX_NR_CPUS); 2713 struct perf_cpu cpu; 2714 2715 for (i = 0; i < PERF_TYPE_MAX; ++i) { 2716 struct evsel *evsel; 2717 2718 evsel = perf_session__find_first_evtype(session, i); 2719 if (!evsel) 2720 continue; 2721 2722 if (!(evsel->core.attr.sample_type & PERF_SAMPLE_CPU)) { 2723 pr_err("File does not contain CPU events. " 2724 "Remove -C option to proceed.\n"); 2725 return -1; 2726 } 2727 } 2728 2729 map = perf_cpu_map__new(cpu_list); 2730 if (map == NULL) { 2731 pr_err("Invalid cpu_list\n"); 2732 return -1; 2733 } 2734 2735 perf_cpu_map__for_each_cpu(cpu, i, map) { 2736 if (cpu.cpu >= nr_cpus) { 2737 pr_err("Requested CPU %d too large. " 2738 "Consider raising MAX_NR_CPUS\n", cpu.cpu); 2739 goto out_delete_map; 2740 } 2741 2742 __set_bit(cpu.cpu, cpu_bitmap); 2743 } 2744 2745 err = 0; 2746 2747 out_delete_map: 2748 perf_cpu_map__put(map); 2749 return err; 2750 } 2751 2752 void perf_session__fprintf_info(struct perf_session *session, FILE *fp, 2753 bool full) 2754 { 2755 if (session == NULL || fp == NULL) 2756 return; 2757 2758 fprintf(fp, "# ========\n"); 2759 perf_header__fprintf_info(session, fp, full); 2760 fprintf(fp, "# ========\n#\n"); 2761 } 2762 2763 static int perf_session__register_guest(struct perf_session *session, pid_t machine_pid) 2764 { 2765 struct machine *machine = machines__findnew(&session->machines, machine_pid); 2766 struct thread *thread; 2767 2768 if (!machine) 2769 return -ENOMEM; 2770 2771 machine->single_address_space = session->machines.host.single_address_space; 2772 2773 thread = machine__idle_thread(machine); 2774 if (!thread) 2775 return -ENOMEM; 2776 thread__put(thread); 2777 2778 machine->kallsyms_filename = perf_data__guest_kallsyms_name(session->data, machine_pid); 2779 2780 return 0; 2781 } 2782 2783 static int perf_session__set_guest_cpu(struct perf_session *session, pid_t pid, 2784 pid_t tid, int guest_cpu) 2785 { 2786 struct machine *machine = &session->machines.host; 2787 struct thread *thread = machine__findnew_thread(machine, pid, tid); 2788 2789 if (!thread) 2790 return -ENOMEM; 2791 thread__set_guest_cpu(thread, guest_cpu); 2792 thread__put(thread); 2793 2794 return 0; 2795 } 2796 2797 int perf_event__process_id_index(const struct perf_tool *tool __maybe_unused, 2798 struct perf_session *session, 2799 union perf_event *event) 2800 { 2801 struct evlist *evlist = session->evlist; 2802 struct perf_record_id_index *ie = &event->id_index; 2803 size_t sz = ie->header.size - sizeof(*ie); 2804 size_t i, nr, max_nr; 2805 size_t e1_sz = sizeof(struct id_index_entry); 2806 size_t e2_sz = sizeof(struct id_index_entry_2); 2807 size_t etot_sz = e1_sz + e2_sz; 2808 struct id_index_entry_2 *e2; 2809 pid_t last_pid = 0; 2810 2811 max_nr = sz / e1_sz; 2812 nr = ie->nr; 2813 if (nr > max_nr) { 2814 printf("Too big: nr %zu max_nr %zu\n", nr, max_nr); 2815 return -EINVAL; 2816 } 2817 2818 if (sz >= nr * etot_sz) { 2819 max_nr = sz / etot_sz; 2820 if (nr > max_nr) { 2821 printf("Too big2: nr %zu max_nr %zu\n", nr, max_nr); 2822 return -EINVAL; 2823 } 2824 e2 = (void *)ie + sizeof(*ie) + nr * e1_sz; 2825 } else { 2826 e2 = NULL; 2827 } 2828 2829 if (dump_trace) 2830 fprintf(stdout, " nr: %zu\n", nr); 2831 2832 for (i = 0; i < nr; i++, (e2 ? e2++ : 0)) { 2833 struct id_index_entry *e = &ie->entries[i]; 2834 struct perf_sample_id *sid; 2835 int ret; 2836 2837 if (dump_trace) { 2838 fprintf(stdout, " ... id: %"PRI_lu64, e->id); 2839 fprintf(stdout, " idx: %"PRI_lu64, e->idx); 2840 fprintf(stdout, " cpu: %"PRI_ld64, e->cpu); 2841 fprintf(stdout, " tid: %"PRI_ld64, e->tid); 2842 if (e2) { 2843 fprintf(stdout, " machine_pid: %"PRI_ld64, e2->machine_pid); 2844 fprintf(stdout, " vcpu: %"PRI_lu64"\n", e2->vcpu); 2845 } else { 2846 fprintf(stdout, "\n"); 2847 } 2848 } 2849 2850 sid = evlist__id2sid(evlist, e->id); 2851 if (!sid) 2852 return -ENOENT; 2853 2854 sid->idx = e->idx; 2855 sid->cpu.cpu = e->cpu; 2856 sid->tid = e->tid; 2857 2858 if (!e2) 2859 continue; 2860 2861 sid->machine_pid = e2->machine_pid; 2862 sid->vcpu.cpu = e2->vcpu; 2863 2864 if (!sid->machine_pid) 2865 continue; 2866 2867 if (sid->machine_pid != last_pid) { 2868 ret = perf_session__register_guest(session, sid->machine_pid); 2869 if (ret) 2870 return ret; 2871 last_pid = sid->machine_pid; 2872 perf_guest = true; 2873 } 2874 2875 ret = perf_session__set_guest_cpu(session, sid->machine_pid, e->tid, e2->vcpu); 2876 if (ret) 2877 return ret; 2878 } 2879 return 0; 2880 } 2881 2882 int perf_session__dsos_hit_all(struct perf_session *session) 2883 { 2884 struct rb_node *nd; 2885 int err; 2886 2887 err = machine__hit_all_dsos(&session->machines.host); 2888 if (err) 2889 return err; 2890 2891 for (nd = rb_first_cached(&session->machines.guests); nd; 2892 nd = rb_next(nd)) { 2893 struct machine *pos = rb_entry(nd, struct machine, rb_node); 2894 2895 err = machine__hit_all_dsos(pos); 2896 if (err) 2897 return err; 2898 } 2899 2900 return 0; 2901 } 2902 2903 struct perf_env *perf_session__env(struct perf_session *session) 2904 { 2905 return &session->header.env; 2906 } 2907