1 #include "builtin.h" 2 #include "perf.h" 3 4 #include "util/util.h" 5 #include "util/evlist.h" 6 #include "util/cache.h" 7 #include "util/evsel.h" 8 #include "util/symbol.h" 9 #include "util/thread.h" 10 #include "util/header.h" 11 #include "util/session.h" 12 #include "util/tool.h" 13 14 #include "util/parse-options.h" 15 #include "util/trace-event.h" 16 17 #include "util/debug.h" 18 19 #include <sys/prctl.h> 20 #include <sys/resource.h> 21 22 #include <semaphore.h> 23 #include <pthread.h> 24 #include <math.h> 25 26 #define PR_SET_NAME 15 /* Set process name */ 27 #define MAX_CPUS 4096 28 #define COMM_LEN 20 29 #define SYM_LEN 129 30 #define MAX_PID 65536 31 32 struct sched_atom; 33 34 struct task_desc { 35 unsigned long nr; 36 unsigned long pid; 37 char comm[COMM_LEN]; 38 39 unsigned long nr_events; 40 unsigned long curr_event; 41 struct sched_atom **atoms; 42 43 pthread_t thread; 44 sem_t sleep_sem; 45 46 sem_t ready_for_work; 47 sem_t work_done_sem; 48 49 u64 cpu_usage; 50 }; 51 52 enum sched_event_type { 53 SCHED_EVENT_RUN, 54 SCHED_EVENT_SLEEP, 55 SCHED_EVENT_WAKEUP, 56 SCHED_EVENT_MIGRATION, 57 }; 58 59 struct sched_atom { 60 enum sched_event_type type; 61 int specific_wait; 62 u64 timestamp; 63 u64 duration; 64 unsigned long nr; 65 sem_t *wait_sem; 66 struct task_desc *wakee; 67 }; 68 69 #define TASK_STATE_TO_CHAR_STR "RSDTtZX" 70 71 enum thread_state { 72 THREAD_SLEEPING = 0, 73 THREAD_WAIT_CPU, 74 THREAD_SCHED_IN, 75 THREAD_IGNORE 76 }; 77 78 struct work_atom { 79 struct list_head list; 80 enum thread_state state; 81 u64 sched_out_time; 82 u64 wake_up_time; 83 u64 sched_in_time; 84 u64 runtime; 85 }; 86 87 struct work_atoms { 88 struct list_head work_list; 89 struct thread *thread; 90 struct rb_node node; 91 u64 max_lat; 92 u64 max_lat_at; 93 u64 total_lat; 94 u64 nb_atoms; 95 u64 total_runtime; 96 }; 97 98 typedef int (*sort_fn_t)(struct work_atoms *, struct work_atoms *); 99 100 struct perf_sched; 101 102 struct trace_sched_handler { 103 int (*switch_event)(struct perf_sched *sched, struct perf_evsel *evsel, 104 struct perf_sample *sample, struct machine *machine); 105 106 int (*runtime_event)(struct perf_sched *sched, struct perf_evsel *evsel, 107 struct perf_sample *sample, struct machine *machine); 108 109 int (*wakeup_event)(struct perf_sched *sched, struct perf_evsel *evsel, 110 struct perf_sample *sample, struct machine *machine); 111 112 /* PERF_RECORD_FORK event, not sched_process_fork tracepoint */ 113 int (*fork_event)(struct perf_sched *sched, union perf_event *event, 114 struct machine *machine); 115 116 int (*migrate_task_event)(struct perf_sched *sched, 117 struct perf_evsel *evsel, 118 struct perf_sample *sample, 119 struct machine *machine); 120 }; 121 122 struct perf_sched { 123 struct perf_tool tool; 124 const char *sort_order; 125 unsigned long nr_tasks; 126 struct task_desc *pid_to_task[MAX_PID]; 127 struct task_desc **tasks; 128 const struct trace_sched_handler *tp_handler; 129 pthread_mutex_t start_work_mutex; 130 pthread_mutex_t work_done_wait_mutex; 131 int profile_cpu; 132 /* 133 * Track the current task - that way we can know whether there's any 134 * weird events, such as a task being switched away that is not current. 135 */ 136 int max_cpu; 137 u32 curr_pid[MAX_CPUS]; 138 struct thread *curr_thread[MAX_CPUS]; 139 char next_shortname1; 140 char next_shortname2; 141 unsigned int replay_repeat; 142 unsigned long nr_run_events; 143 unsigned long nr_sleep_events; 144 unsigned long nr_wakeup_events; 145 unsigned long nr_sleep_corrections; 146 unsigned long nr_run_events_optimized; 147 unsigned long targetless_wakeups; 148 unsigned long multitarget_wakeups; 149 unsigned long nr_runs; 150 unsigned long nr_timestamps; 151 unsigned long nr_unordered_timestamps; 152 unsigned long nr_state_machine_bugs; 153 unsigned long nr_context_switch_bugs; 154 unsigned long nr_events; 155 unsigned long nr_lost_chunks; 156 unsigned long nr_lost_events; 157 u64 run_measurement_overhead; 158 u64 sleep_measurement_overhead; 159 u64 start_time; 160 u64 cpu_usage; 161 u64 runavg_cpu_usage; 162 u64 parent_cpu_usage; 163 u64 runavg_parent_cpu_usage; 164 u64 sum_runtime; 165 u64 sum_fluct; 166 u64 run_avg; 167 u64 all_runtime; 168 u64 all_count; 169 u64 cpu_last_switched[MAX_CPUS]; 170 struct rb_root atom_root, sorted_atom_root; 171 struct list_head sort_list, cmp_pid; 172 }; 173 174 static u64 get_nsecs(void) 175 { 176 #ifndef __APPLE__ 177 struct timespec ts; 178 179 clock_gettime(CLOCK_MONOTONIC, &ts); 180 181 return ts.tv_sec * 1000000000ULL + ts.tv_nsec; 182 #else 183 perror("get_nsecs not supported on MacOS"); 184 return 0; 185 #endif 186 } 187 188 static void burn_nsecs(struct perf_sched *sched, u64 nsecs) 189 { 190 u64 T0 = get_nsecs(), T1; 191 192 do { 193 T1 = get_nsecs(); 194 } while (T1 + sched->run_measurement_overhead < T0 + nsecs); 195 } 196 197 static void sleep_nsecs(u64 nsecs) 198 { 199 struct timespec ts; 200 201 ts.tv_nsec = nsecs % 999999999; 202 ts.tv_sec = nsecs / 999999999; 203 204 nanosleep(&ts, NULL); 205 } 206 207 static void calibrate_run_measurement_overhead(struct perf_sched *sched) 208 { 209 u64 T0, T1, delta, min_delta = 1000000000ULL; 210 int i; 211 212 for (i = 0; i < 10; i++) { 213 T0 = get_nsecs(); 214 burn_nsecs(sched, 0); 215 T1 = get_nsecs(); 216 delta = T1-T0; 217 min_delta = min(min_delta, delta); 218 } 219 sched->run_measurement_overhead = min_delta; 220 221 printf("run measurement overhead: %" PRIu64 " nsecs\n", min_delta); 222 } 223 224 static void calibrate_sleep_measurement_overhead(struct perf_sched *sched) 225 { 226 u64 T0, T1, delta, min_delta = 1000000000ULL; 227 int i; 228 229 for (i = 0; i < 10; i++) { 230 T0 = get_nsecs(); 231 sleep_nsecs(10000); 232 T1 = get_nsecs(); 233 delta = T1-T0; 234 min_delta = min(min_delta, delta); 235 } 236 min_delta -= 10000; 237 sched->sleep_measurement_overhead = min_delta; 238 239 printf("sleep measurement overhead: %" PRIu64 " nsecs\n", min_delta); 240 } 241 242 static struct sched_atom * 243 get_new_event(struct task_desc *task, u64 timestamp) 244 { 245 struct sched_atom *event = zalloc(sizeof(*event)); 246 unsigned long idx = task->nr_events; 247 size_t size; 248 249 event->timestamp = timestamp; 250 event->nr = idx; 251 252 task->nr_events++; 253 size = sizeof(struct sched_atom *) * task->nr_events; 254 task->atoms = realloc(task->atoms, size); 255 BUG_ON(!task->atoms); 256 257 task->atoms[idx] = event; 258 259 return event; 260 } 261 262 static struct sched_atom *last_event(struct task_desc *task) 263 { 264 if (!task->nr_events) 265 return NULL; 266 267 return task->atoms[task->nr_events - 1]; 268 } 269 270 static void add_sched_event_run(struct perf_sched *sched, struct task_desc *task, 271 u64 timestamp, u64 duration) 272 { 273 struct sched_atom *event, *curr_event = last_event(task); 274 275 /* 276 * optimize an existing RUN event by merging this one 277 * to it: 278 */ 279 if (curr_event && curr_event->type == SCHED_EVENT_RUN) { 280 sched->nr_run_events_optimized++; 281 curr_event->duration += duration; 282 return; 283 } 284 285 event = get_new_event(task, timestamp); 286 287 event->type = SCHED_EVENT_RUN; 288 event->duration = duration; 289 290 sched->nr_run_events++; 291 } 292 293 static void add_sched_event_wakeup(struct perf_sched *sched, struct task_desc *task, 294 u64 timestamp, struct task_desc *wakee) 295 { 296 struct sched_atom *event, *wakee_event; 297 298 event = get_new_event(task, timestamp); 299 event->type = SCHED_EVENT_WAKEUP; 300 event->wakee = wakee; 301 302 wakee_event = last_event(wakee); 303 if (!wakee_event || wakee_event->type != SCHED_EVENT_SLEEP) { 304 sched->targetless_wakeups++; 305 return; 306 } 307 if (wakee_event->wait_sem) { 308 sched->multitarget_wakeups++; 309 return; 310 } 311 312 wakee_event->wait_sem = zalloc(sizeof(*wakee_event->wait_sem)); 313 sem_init(wakee_event->wait_sem, 0, 0); 314 wakee_event->specific_wait = 1; 315 event->wait_sem = wakee_event->wait_sem; 316 317 sched->nr_wakeup_events++; 318 } 319 320 static void add_sched_event_sleep(struct perf_sched *sched, struct task_desc *task, 321 u64 timestamp, u64 task_state __maybe_unused) 322 { 323 struct sched_atom *event = get_new_event(task, timestamp); 324 325 event->type = SCHED_EVENT_SLEEP; 326 327 sched->nr_sleep_events++; 328 } 329 330 static struct task_desc *register_pid(struct perf_sched *sched, 331 unsigned long pid, const char *comm) 332 { 333 struct task_desc *task; 334 335 BUG_ON(pid >= MAX_PID); 336 337 task = sched->pid_to_task[pid]; 338 339 if (task) 340 return task; 341 342 task = zalloc(sizeof(*task)); 343 task->pid = pid; 344 task->nr = sched->nr_tasks; 345 strcpy(task->comm, comm); 346 /* 347 * every task starts in sleeping state - this gets ignored 348 * if there's no wakeup pointing to this sleep state: 349 */ 350 add_sched_event_sleep(sched, task, 0, 0); 351 352 sched->pid_to_task[pid] = task; 353 sched->nr_tasks++; 354 sched->tasks = realloc(sched->tasks, sched->nr_tasks * sizeof(struct task_task *)); 355 BUG_ON(!sched->tasks); 356 sched->tasks[task->nr] = task; 357 358 if (verbose) 359 printf("registered task #%ld, PID %ld (%s)\n", sched->nr_tasks, pid, comm); 360 361 return task; 362 } 363 364 365 static void print_task_traces(struct perf_sched *sched) 366 { 367 struct task_desc *task; 368 unsigned long i; 369 370 for (i = 0; i < sched->nr_tasks; i++) { 371 task = sched->tasks[i]; 372 printf("task %6ld (%20s:%10ld), nr_events: %ld\n", 373 task->nr, task->comm, task->pid, task->nr_events); 374 } 375 } 376 377 static void add_cross_task_wakeups(struct perf_sched *sched) 378 { 379 struct task_desc *task1, *task2; 380 unsigned long i, j; 381 382 for (i = 0; i < sched->nr_tasks; i++) { 383 task1 = sched->tasks[i]; 384 j = i + 1; 385 if (j == sched->nr_tasks) 386 j = 0; 387 task2 = sched->tasks[j]; 388 add_sched_event_wakeup(sched, task1, 0, task2); 389 } 390 } 391 392 static void perf_sched__process_event(struct perf_sched *sched, 393 struct sched_atom *atom) 394 { 395 int ret = 0; 396 397 switch (atom->type) { 398 case SCHED_EVENT_RUN: 399 burn_nsecs(sched, atom->duration); 400 break; 401 case SCHED_EVENT_SLEEP: 402 if (atom->wait_sem) 403 ret = sem_wait(atom->wait_sem); 404 BUG_ON(ret); 405 break; 406 case SCHED_EVENT_WAKEUP: 407 if (atom->wait_sem) 408 ret = sem_post(atom->wait_sem); 409 BUG_ON(ret); 410 break; 411 case SCHED_EVENT_MIGRATION: 412 break; 413 default: 414 BUG_ON(1); 415 } 416 } 417 418 static u64 get_cpu_usage_nsec_parent(void) 419 { 420 struct rusage ru; 421 u64 sum; 422 int err; 423 424 err = getrusage(RUSAGE_SELF, &ru); 425 BUG_ON(err); 426 427 sum = ru.ru_utime.tv_sec*1e9 + ru.ru_utime.tv_usec*1e3; 428 sum += ru.ru_stime.tv_sec*1e9 + ru.ru_stime.tv_usec*1e3; 429 430 return sum; 431 } 432 433 static int self_open_counters(void) 434 { 435 struct perf_event_attr attr; 436 int fd; 437 438 memset(&attr, 0, sizeof(attr)); 439 440 attr.type = PERF_TYPE_SOFTWARE; 441 attr.config = PERF_COUNT_SW_TASK_CLOCK; 442 443 fd = sys_perf_event_open(&attr, 0, -1, -1, 0); 444 445 if (fd < 0) 446 pr_err("Error: sys_perf_event_open() syscall returned " 447 "with %d (%s)\n", fd, strerror(errno)); 448 return fd; 449 } 450 451 static u64 get_cpu_usage_nsec_self(int fd) 452 { 453 u64 runtime; 454 int ret; 455 456 ret = read(fd, &runtime, sizeof(runtime)); 457 BUG_ON(ret != sizeof(runtime)); 458 459 return runtime; 460 } 461 462 struct sched_thread_parms { 463 struct task_desc *task; 464 struct perf_sched *sched; 465 }; 466 467 static void *thread_func(void *ctx) 468 { 469 #ifndef __APPLE__ 470 struct sched_thread_parms *parms = ctx; 471 struct task_desc *this_task = parms->task; 472 struct perf_sched *sched = parms->sched; 473 u64 cpu_usage_0, cpu_usage_1; 474 unsigned long i, ret; 475 char comm2[22]; 476 int fd; 477 478 free(parms); 479 480 sprintf(comm2, ":%s", this_task->comm); 481 prctl(PR_SET_NAME, comm2); 482 fd = self_open_counters(); 483 if (fd < 0) 484 return NULL; 485 again: 486 ret = sem_post(&this_task->ready_for_work); 487 BUG_ON(ret); 488 ret = pthread_mutex_lock(&sched->start_work_mutex); 489 BUG_ON(ret); 490 ret = pthread_mutex_unlock(&sched->start_work_mutex); 491 BUG_ON(ret); 492 493 cpu_usage_0 = get_cpu_usage_nsec_self(fd); 494 495 for (i = 0; i < this_task->nr_events; i++) { 496 this_task->curr_event = i; 497 perf_sched__process_event(sched, this_task->atoms[i]); 498 } 499 500 cpu_usage_1 = get_cpu_usage_nsec_self(fd); 501 this_task->cpu_usage = cpu_usage_1 - cpu_usage_0; 502 ret = sem_post(&this_task->work_done_sem); 503 BUG_ON(ret); 504 505 ret = pthread_mutex_lock(&sched->work_done_wait_mutex); 506 BUG_ON(ret); 507 ret = pthread_mutex_unlock(&sched->work_done_wait_mutex); 508 BUG_ON(ret); 509 510 goto again; 511 #else 512 perror("thread_func not supported on MacOS"); 513 return NULL; 514 #endif 515 } 516 517 static void create_tasks(struct perf_sched *sched) 518 { 519 struct task_desc *task; 520 pthread_attr_t attr; 521 unsigned long i; 522 int err; 523 524 err = pthread_attr_init(&attr); 525 BUG_ON(err); 526 err = pthread_attr_setstacksize(&attr, 527 (size_t) max(16 * 1024, PTHREAD_STACK_MIN)); 528 BUG_ON(err); 529 err = pthread_mutex_lock(&sched->start_work_mutex); 530 BUG_ON(err); 531 err = pthread_mutex_lock(&sched->work_done_wait_mutex); 532 BUG_ON(err); 533 for (i = 0; i < sched->nr_tasks; i++) { 534 struct sched_thread_parms *parms = malloc(sizeof(*parms)); 535 BUG_ON(parms == NULL); 536 parms->task = task = sched->tasks[i]; 537 parms->sched = sched; 538 sem_init(&task->sleep_sem, 0, 0); 539 sem_init(&task->ready_for_work, 0, 0); 540 sem_init(&task->work_done_sem, 0, 0); 541 task->curr_event = 0; 542 err = pthread_create(&task->thread, &attr, thread_func, parms); 543 BUG_ON(err); 544 } 545 } 546 547 static void wait_for_tasks(struct perf_sched *sched) 548 { 549 u64 cpu_usage_0, cpu_usage_1; 550 struct task_desc *task; 551 unsigned long i, ret; 552 553 sched->start_time = get_nsecs(); 554 sched->cpu_usage = 0; 555 pthread_mutex_unlock(&sched->work_done_wait_mutex); 556 557 for (i = 0; i < sched->nr_tasks; i++) { 558 task = sched->tasks[i]; 559 ret = sem_wait(&task->ready_for_work); 560 BUG_ON(ret); 561 sem_init(&task->ready_for_work, 0, 0); 562 } 563 ret = pthread_mutex_lock(&sched->work_done_wait_mutex); 564 BUG_ON(ret); 565 566 cpu_usage_0 = get_cpu_usage_nsec_parent(); 567 568 pthread_mutex_unlock(&sched->start_work_mutex); 569 570 for (i = 0; i < sched->nr_tasks; i++) { 571 task = sched->tasks[i]; 572 ret = sem_wait(&task->work_done_sem); 573 BUG_ON(ret); 574 sem_init(&task->work_done_sem, 0, 0); 575 sched->cpu_usage += task->cpu_usage; 576 task->cpu_usage = 0; 577 } 578 579 cpu_usage_1 = get_cpu_usage_nsec_parent(); 580 if (!sched->runavg_cpu_usage) 581 sched->runavg_cpu_usage = sched->cpu_usage; 582 sched->runavg_cpu_usage = (sched->runavg_cpu_usage * 9 + sched->cpu_usage) / 10; 583 584 sched->parent_cpu_usage = cpu_usage_1 - cpu_usage_0; 585 if (!sched->runavg_parent_cpu_usage) 586 sched->runavg_parent_cpu_usage = sched->parent_cpu_usage; 587 sched->runavg_parent_cpu_usage = (sched->runavg_parent_cpu_usage * 9 + 588 sched->parent_cpu_usage)/10; 589 590 ret = pthread_mutex_lock(&sched->start_work_mutex); 591 BUG_ON(ret); 592 593 for (i = 0; i < sched->nr_tasks; i++) { 594 task = sched->tasks[i]; 595 sem_init(&task->sleep_sem, 0, 0); 596 task->curr_event = 0; 597 } 598 } 599 600 static void run_one_test(struct perf_sched *sched) 601 { 602 u64 T0, T1, delta, avg_delta, fluct; 603 604 T0 = get_nsecs(); 605 wait_for_tasks(sched); 606 T1 = get_nsecs(); 607 608 delta = T1 - T0; 609 sched->sum_runtime += delta; 610 sched->nr_runs++; 611 612 avg_delta = sched->sum_runtime / sched->nr_runs; 613 if (delta < avg_delta) 614 fluct = avg_delta - delta; 615 else 616 fluct = delta - avg_delta; 617 sched->sum_fluct += fluct; 618 if (!sched->run_avg) 619 sched->run_avg = delta; 620 sched->run_avg = (sched->run_avg * 9 + delta) / 10; 621 622 printf("#%-3ld: %0.3f, ", sched->nr_runs, (double)delta / 1000000.0); 623 624 printf("ravg: %0.2f, ", (double)sched->run_avg / 1e6); 625 626 printf("cpu: %0.2f / %0.2f", 627 (double)sched->cpu_usage / 1e6, (double)sched->runavg_cpu_usage / 1e6); 628 629 #if 0 630 /* 631 * rusage statistics done by the parent, these are less 632 * accurate than the sched->sum_exec_runtime based statistics: 633 */ 634 printf(" [%0.2f / %0.2f]", 635 (double)sched->parent_cpu_usage/1e6, 636 (double)sched->runavg_parent_cpu_usage/1e6); 637 #endif 638 639 printf("\n"); 640 641 if (sched->nr_sleep_corrections) 642 printf(" (%ld sleep corrections)\n", sched->nr_sleep_corrections); 643 sched->nr_sleep_corrections = 0; 644 } 645 646 static void test_calibrations(struct perf_sched *sched) 647 { 648 u64 T0, T1; 649 650 T0 = get_nsecs(); 651 burn_nsecs(sched, 1e6); 652 T1 = get_nsecs(); 653 654 printf("the run test took %" PRIu64 " nsecs\n", T1 - T0); 655 656 T0 = get_nsecs(); 657 sleep_nsecs(1e6); 658 T1 = get_nsecs(); 659 660 printf("the sleep test took %" PRIu64 " nsecs\n", T1 - T0); 661 } 662 663 static int 664 replay_wakeup_event(struct perf_sched *sched, 665 struct perf_evsel *evsel, struct perf_sample *sample, 666 struct machine *machine __maybe_unused) 667 { 668 const char *comm = perf_evsel__strval(evsel, sample, "comm"); 669 const u32 pid = perf_evsel__intval(evsel, sample, "pid"); 670 struct task_desc *waker, *wakee; 671 672 if (verbose) { 673 printf("sched_wakeup event %p\n", evsel); 674 675 printf(" ... pid %d woke up %s/%d\n", sample->tid, comm, pid); 676 } 677 678 waker = register_pid(sched, sample->tid, "<unknown>"); 679 wakee = register_pid(sched, pid, comm); 680 681 add_sched_event_wakeup(sched, waker, sample->time, wakee); 682 return 0; 683 } 684 685 static int replay_switch_event(struct perf_sched *sched, 686 struct perf_evsel *evsel, 687 struct perf_sample *sample, 688 struct machine *machine __maybe_unused) 689 { 690 const char *prev_comm = perf_evsel__strval(evsel, sample, "prev_comm"), 691 *next_comm = perf_evsel__strval(evsel, sample, "next_comm"); 692 const u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"), 693 next_pid = perf_evsel__intval(evsel, sample, "next_pid"); 694 const u64 prev_state = perf_evsel__intval(evsel, sample, "prev_state"); 695 struct task_desc *prev, __maybe_unused *next; 696 u64 timestamp0, timestamp = sample->time; 697 int cpu = sample->cpu; 698 s64 delta; 699 700 if (verbose) 701 printf("sched_switch event %p\n", evsel); 702 703 if (cpu >= MAX_CPUS || cpu < 0) 704 return 0; 705 706 timestamp0 = sched->cpu_last_switched[cpu]; 707 if (timestamp0) 708 delta = timestamp - timestamp0; 709 else 710 delta = 0; 711 712 if (delta < 0) { 713 pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta); 714 return -1; 715 } 716 717 pr_debug(" ... switch from %s/%d to %s/%d [ran %" PRIu64 " nsecs]\n", 718 prev_comm, prev_pid, next_comm, next_pid, delta); 719 720 prev = register_pid(sched, prev_pid, prev_comm); 721 next = register_pid(sched, next_pid, next_comm); 722 723 sched->cpu_last_switched[cpu] = timestamp; 724 725 add_sched_event_run(sched, prev, timestamp, delta); 726 add_sched_event_sleep(sched, prev, timestamp, prev_state); 727 728 return 0; 729 } 730 731 static int replay_fork_event(struct perf_sched *sched, 732 union perf_event *event, 733 struct machine *machine) 734 { 735 struct thread *child, *parent; 736 737 child = machine__findnew_thread(machine, event->fork.pid, 738 event->fork.tid); 739 parent = machine__findnew_thread(machine, event->fork.ppid, 740 event->fork.ptid); 741 742 if (child == NULL || parent == NULL) { 743 pr_debug("thread does not exist on fork event: child %p, parent %p\n", 744 child, parent); 745 return 0; 746 } 747 748 if (verbose) { 749 printf("fork event\n"); 750 printf("... parent: %s/%d\n", parent->comm, parent->tid); 751 printf("... child: %s/%d\n", child->comm, child->tid); 752 } 753 754 register_pid(sched, parent->tid, parent->comm); 755 register_pid(sched, child->tid, child->comm); 756 return 0; 757 } 758 759 struct sort_dimension { 760 const char *name; 761 sort_fn_t cmp; 762 struct list_head list; 763 }; 764 765 static int 766 thread_lat_cmp(struct list_head *list, struct work_atoms *l, struct work_atoms *r) 767 { 768 struct sort_dimension *sort; 769 int ret = 0; 770 771 BUG_ON(list_empty(list)); 772 773 list_for_each_entry(sort, list, list) { 774 ret = sort->cmp(l, r); 775 if (ret) 776 return ret; 777 } 778 779 return ret; 780 } 781 782 static struct work_atoms * 783 thread_atoms_search(struct rb_root *root, struct thread *thread, 784 struct list_head *sort_list) 785 { 786 struct rb_node *node = root->rb_node; 787 struct work_atoms key = { .thread = thread }; 788 789 while (node) { 790 struct work_atoms *atoms; 791 int cmp; 792 793 atoms = container_of(node, struct work_atoms, node); 794 795 cmp = thread_lat_cmp(sort_list, &key, atoms); 796 if (cmp > 0) 797 node = node->rb_left; 798 else if (cmp < 0) 799 node = node->rb_right; 800 else { 801 BUG_ON(thread != atoms->thread); 802 return atoms; 803 } 804 } 805 return NULL; 806 } 807 808 static void 809 __thread_latency_insert(struct rb_root *root, struct work_atoms *data, 810 struct list_head *sort_list) 811 { 812 struct rb_node **new = &(root->rb_node), *parent = NULL; 813 814 while (*new) { 815 struct work_atoms *this; 816 int cmp; 817 818 this = container_of(*new, struct work_atoms, node); 819 parent = *new; 820 821 cmp = thread_lat_cmp(sort_list, data, this); 822 823 if (cmp > 0) 824 new = &((*new)->rb_left); 825 else 826 new = &((*new)->rb_right); 827 } 828 829 rb_link_node(&data->node, parent, new); 830 rb_insert_color(&data->node, root); 831 } 832 833 static int thread_atoms_insert(struct perf_sched *sched, struct thread *thread) 834 { 835 struct work_atoms *atoms = zalloc(sizeof(*atoms)); 836 if (!atoms) { 837 pr_err("No memory at %s\n", __func__); 838 return -1; 839 } 840 841 atoms->thread = thread; 842 INIT_LIST_HEAD(&atoms->work_list); 843 __thread_latency_insert(&sched->atom_root, atoms, &sched->cmp_pid); 844 return 0; 845 } 846 847 static char sched_out_state(u64 prev_state) 848 { 849 const char *str = TASK_STATE_TO_CHAR_STR; 850 851 return str[prev_state]; 852 } 853 854 static int 855 add_sched_out_event(struct work_atoms *atoms, 856 char run_state, 857 u64 timestamp) 858 { 859 struct work_atom *atom = zalloc(sizeof(*atom)); 860 if (!atom) { 861 pr_err("Non memory at %s", __func__); 862 return -1; 863 } 864 865 atom->sched_out_time = timestamp; 866 867 if (run_state == 'R') { 868 atom->state = THREAD_WAIT_CPU; 869 atom->wake_up_time = atom->sched_out_time; 870 } 871 872 list_add_tail(&atom->list, &atoms->work_list); 873 return 0; 874 } 875 876 static void 877 add_runtime_event(struct work_atoms *atoms, u64 delta, 878 u64 timestamp __maybe_unused) 879 { 880 struct work_atom *atom; 881 882 BUG_ON(list_empty(&atoms->work_list)); 883 884 atom = list_entry(atoms->work_list.prev, struct work_atom, list); 885 886 atom->runtime += delta; 887 atoms->total_runtime += delta; 888 } 889 890 static void 891 add_sched_in_event(struct work_atoms *atoms, u64 timestamp) 892 { 893 struct work_atom *atom; 894 u64 delta; 895 896 if (list_empty(&atoms->work_list)) 897 return; 898 899 atom = list_entry(atoms->work_list.prev, struct work_atom, list); 900 901 if (atom->state != THREAD_WAIT_CPU) 902 return; 903 904 if (timestamp < atom->wake_up_time) { 905 atom->state = THREAD_IGNORE; 906 return; 907 } 908 909 atom->state = THREAD_SCHED_IN; 910 atom->sched_in_time = timestamp; 911 912 delta = atom->sched_in_time - atom->wake_up_time; 913 atoms->total_lat += delta; 914 if (delta > atoms->max_lat) { 915 atoms->max_lat = delta; 916 atoms->max_lat_at = timestamp; 917 } 918 atoms->nb_atoms++; 919 } 920 921 static int latency_switch_event(struct perf_sched *sched, 922 struct perf_evsel *evsel, 923 struct perf_sample *sample, 924 struct machine *machine) 925 { 926 const u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"), 927 next_pid = perf_evsel__intval(evsel, sample, "next_pid"); 928 const u64 prev_state = perf_evsel__intval(evsel, sample, "prev_state"); 929 struct work_atoms *out_events, *in_events; 930 struct thread *sched_out, *sched_in; 931 u64 timestamp0, timestamp = sample->time; 932 int cpu = sample->cpu; 933 s64 delta; 934 935 BUG_ON(cpu >= MAX_CPUS || cpu < 0); 936 937 timestamp0 = sched->cpu_last_switched[cpu]; 938 sched->cpu_last_switched[cpu] = timestamp; 939 if (timestamp0) 940 delta = timestamp - timestamp0; 941 else 942 delta = 0; 943 944 if (delta < 0) { 945 pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta); 946 return -1; 947 } 948 949 sched_out = machine__findnew_thread(machine, 0, prev_pid); 950 sched_in = machine__findnew_thread(machine, 0, next_pid); 951 952 out_events = thread_atoms_search(&sched->atom_root, sched_out, &sched->cmp_pid); 953 if (!out_events) { 954 if (thread_atoms_insert(sched, sched_out)) 955 return -1; 956 out_events = thread_atoms_search(&sched->atom_root, sched_out, &sched->cmp_pid); 957 if (!out_events) { 958 pr_err("out-event: Internal tree error"); 959 return -1; 960 } 961 } 962 if (add_sched_out_event(out_events, sched_out_state(prev_state), timestamp)) 963 return -1; 964 965 in_events = thread_atoms_search(&sched->atom_root, sched_in, &sched->cmp_pid); 966 if (!in_events) { 967 if (thread_atoms_insert(sched, sched_in)) 968 return -1; 969 in_events = thread_atoms_search(&sched->atom_root, sched_in, &sched->cmp_pid); 970 if (!in_events) { 971 pr_err("in-event: Internal tree error"); 972 return -1; 973 } 974 /* 975 * Take came in we have not heard about yet, 976 * add in an initial atom in runnable state: 977 */ 978 if (add_sched_out_event(in_events, 'R', timestamp)) 979 return -1; 980 } 981 add_sched_in_event(in_events, timestamp); 982 983 return 0; 984 } 985 986 static int latency_runtime_event(struct perf_sched *sched, 987 struct perf_evsel *evsel, 988 struct perf_sample *sample, 989 struct machine *machine) 990 { 991 const u32 pid = perf_evsel__intval(evsel, sample, "pid"); 992 const u64 runtime = perf_evsel__intval(evsel, sample, "runtime"); 993 struct thread *thread = machine__findnew_thread(machine, 0, pid); 994 struct work_atoms *atoms = thread_atoms_search(&sched->atom_root, thread, &sched->cmp_pid); 995 u64 timestamp = sample->time; 996 int cpu = sample->cpu; 997 998 BUG_ON(cpu >= MAX_CPUS || cpu < 0); 999 if (!atoms) { 1000 if (thread_atoms_insert(sched, thread)) 1001 return -1; 1002 atoms = thread_atoms_search(&sched->atom_root, thread, &sched->cmp_pid); 1003 if (!atoms) { 1004 pr_err("in-event: Internal tree error"); 1005 return -1; 1006 } 1007 if (add_sched_out_event(atoms, 'R', timestamp)) 1008 return -1; 1009 } 1010 1011 add_runtime_event(atoms, runtime, timestamp); 1012 return 0; 1013 } 1014 1015 static int latency_wakeup_event(struct perf_sched *sched, 1016 struct perf_evsel *evsel, 1017 struct perf_sample *sample, 1018 struct machine *machine) 1019 { 1020 const u32 pid = perf_evsel__intval(evsel, sample, "pid"), 1021 success = perf_evsel__intval(evsel, sample, "success"); 1022 struct work_atoms *atoms; 1023 struct work_atom *atom; 1024 struct thread *wakee; 1025 u64 timestamp = sample->time; 1026 1027 /* Note for later, it may be interesting to observe the failing cases */ 1028 if (!success) 1029 return 0; 1030 1031 wakee = machine__findnew_thread(machine, 0, pid); 1032 atoms = thread_atoms_search(&sched->atom_root, wakee, &sched->cmp_pid); 1033 if (!atoms) { 1034 if (thread_atoms_insert(sched, wakee)) 1035 return -1; 1036 atoms = thread_atoms_search(&sched->atom_root, wakee, &sched->cmp_pid); 1037 if (!atoms) { 1038 pr_err("wakeup-event: Internal tree error"); 1039 return -1; 1040 } 1041 if (add_sched_out_event(atoms, 'S', timestamp)) 1042 return -1; 1043 } 1044 1045 BUG_ON(list_empty(&atoms->work_list)); 1046 1047 atom = list_entry(atoms->work_list.prev, struct work_atom, list); 1048 1049 /* 1050 * You WILL be missing events if you've recorded only 1051 * one CPU, or are only looking at only one, so don't 1052 * make useless noise. 1053 */ 1054 if (sched->profile_cpu == -1 && atom->state != THREAD_SLEEPING) 1055 sched->nr_state_machine_bugs++; 1056 1057 sched->nr_timestamps++; 1058 if (atom->sched_out_time > timestamp) { 1059 sched->nr_unordered_timestamps++; 1060 return 0; 1061 } 1062 1063 atom->state = THREAD_WAIT_CPU; 1064 atom->wake_up_time = timestamp; 1065 return 0; 1066 } 1067 1068 static int latency_migrate_task_event(struct perf_sched *sched, 1069 struct perf_evsel *evsel, 1070 struct perf_sample *sample, 1071 struct machine *machine) 1072 { 1073 const u32 pid = perf_evsel__intval(evsel, sample, "pid"); 1074 u64 timestamp = sample->time; 1075 struct work_atoms *atoms; 1076 struct work_atom *atom; 1077 struct thread *migrant; 1078 1079 /* 1080 * Only need to worry about migration when profiling one CPU. 1081 */ 1082 if (sched->profile_cpu == -1) 1083 return 0; 1084 1085 migrant = machine__findnew_thread(machine, 0, pid); 1086 atoms = thread_atoms_search(&sched->atom_root, migrant, &sched->cmp_pid); 1087 if (!atoms) { 1088 if (thread_atoms_insert(sched, migrant)) 1089 return -1; 1090 register_pid(sched, migrant->tid, migrant->comm); 1091 atoms = thread_atoms_search(&sched->atom_root, migrant, &sched->cmp_pid); 1092 if (!atoms) { 1093 pr_err("migration-event: Internal tree error"); 1094 return -1; 1095 } 1096 if (add_sched_out_event(atoms, 'R', timestamp)) 1097 return -1; 1098 } 1099 1100 BUG_ON(list_empty(&atoms->work_list)); 1101 1102 atom = list_entry(atoms->work_list.prev, struct work_atom, list); 1103 atom->sched_in_time = atom->sched_out_time = atom->wake_up_time = timestamp; 1104 1105 sched->nr_timestamps++; 1106 1107 if (atom->sched_out_time > timestamp) 1108 sched->nr_unordered_timestamps++; 1109 1110 return 0; 1111 } 1112 1113 static void output_lat_thread(struct perf_sched *sched, struct work_atoms *work_list) 1114 { 1115 int i; 1116 int ret; 1117 u64 avg; 1118 1119 if (!work_list->nb_atoms) 1120 return; 1121 /* 1122 * Ignore idle threads: 1123 */ 1124 if (!strcmp(work_list->thread->comm, "swapper")) 1125 return; 1126 1127 sched->all_runtime += work_list->total_runtime; 1128 sched->all_count += work_list->nb_atoms; 1129 1130 ret = printf(" %s:%d ", work_list->thread->comm, work_list->thread->tid); 1131 1132 for (i = 0; i < 24 - ret; i++) 1133 printf(" "); 1134 1135 avg = work_list->total_lat / work_list->nb_atoms; 1136 1137 printf("|%11.3f ms |%9" PRIu64 " | avg:%9.3f ms | max:%9.3f ms | max at: %9.6f s\n", 1138 (double)work_list->total_runtime / 1e6, 1139 work_list->nb_atoms, (double)avg / 1e6, 1140 (double)work_list->max_lat / 1e6, 1141 (double)work_list->max_lat_at / 1e9); 1142 } 1143 1144 static int pid_cmp(struct work_atoms *l, struct work_atoms *r) 1145 { 1146 if (l->thread->tid < r->thread->tid) 1147 return -1; 1148 if (l->thread->tid > r->thread->tid) 1149 return 1; 1150 1151 return 0; 1152 } 1153 1154 static int avg_cmp(struct work_atoms *l, struct work_atoms *r) 1155 { 1156 u64 avgl, avgr; 1157 1158 if (!l->nb_atoms) 1159 return -1; 1160 1161 if (!r->nb_atoms) 1162 return 1; 1163 1164 avgl = l->total_lat / l->nb_atoms; 1165 avgr = r->total_lat / r->nb_atoms; 1166 1167 if (avgl < avgr) 1168 return -1; 1169 if (avgl > avgr) 1170 return 1; 1171 1172 return 0; 1173 } 1174 1175 static int max_cmp(struct work_atoms *l, struct work_atoms *r) 1176 { 1177 if (l->max_lat < r->max_lat) 1178 return -1; 1179 if (l->max_lat > r->max_lat) 1180 return 1; 1181 1182 return 0; 1183 } 1184 1185 static int switch_cmp(struct work_atoms *l, struct work_atoms *r) 1186 { 1187 if (l->nb_atoms < r->nb_atoms) 1188 return -1; 1189 if (l->nb_atoms > r->nb_atoms) 1190 return 1; 1191 1192 return 0; 1193 } 1194 1195 static int runtime_cmp(struct work_atoms *l, struct work_atoms *r) 1196 { 1197 if (l->total_runtime < r->total_runtime) 1198 return -1; 1199 if (l->total_runtime > r->total_runtime) 1200 return 1; 1201 1202 return 0; 1203 } 1204 1205 static int sort_dimension__add(const char *tok, struct list_head *list) 1206 { 1207 size_t i; 1208 static struct sort_dimension avg_sort_dimension = { 1209 .name = "avg", 1210 .cmp = avg_cmp, 1211 }; 1212 static struct sort_dimension max_sort_dimension = { 1213 .name = "max", 1214 .cmp = max_cmp, 1215 }; 1216 static struct sort_dimension pid_sort_dimension = { 1217 .name = "pid", 1218 .cmp = pid_cmp, 1219 }; 1220 static struct sort_dimension runtime_sort_dimension = { 1221 .name = "runtime", 1222 .cmp = runtime_cmp, 1223 }; 1224 static struct sort_dimension switch_sort_dimension = { 1225 .name = "switch", 1226 .cmp = switch_cmp, 1227 }; 1228 struct sort_dimension *available_sorts[] = { 1229 &pid_sort_dimension, 1230 &avg_sort_dimension, 1231 &max_sort_dimension, 1232 &switch_sort_dimension, 1233 &runtime_sort_dimension, 1234 }; 1235 1236 for (i = 0; i < ARRAY_SIZE(available_sorts); i++) { 1237 if (!strcmp(available_sorts[i]->name, tok)) { 1238 list_add_tail(&available_sorts[i]->list, list); 1239 1240 return 0; 1241 } 1242 } 1243 1244 return -1; 1245 } 1246 1247 static void perf_sched__sort_lat(struct perf_sched *sched) 1248 { 1249 struct rb_node *node; 1250 1251 for (;;) { 1252 struct work_atoms *data; 1253 node = rb_first(&sched->atom_root); 1254 if (!node) 1255 break; 1256 1257 rb_erase(node, &sched->atom_root); 1258 data = rb_entry(node, struct work_atoms, node); 1259 __thread_latency_insert(&sched->sorted_atom_root, data, &sched->sort_list); 1260 } 1261 } 1262 1263 static int process_sched_wakeup_event(struct perf_tool *tool, 1264 struct perf_evsel *evsel, 1265 struct perf_sample *sample, 1266 struct machine *machine) 1267 { 1268 struct perf_sched *sched = container_of(tool, struct perf_sched, tool); 1269 1270 if (sched->tp_handler->wakeup_event) 1271 return sched->tp_handler->wakeup_event(sched, evsel, sample, machine); 1272 1273 return 0; 1274 } 1275 1276 static int map_switch_event(struct perf_sched *sched, struct perf_evsel *evsel, 1277 struct perf_sample *sample, struct machine *machine) 1278 { 1279 const u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"), 1280 next_pid = perf_evsel__intval(evsel, sample, "next_pid"); 1281 struct thread *sched_out __maybe_unused, *sched_in; 1282 int new_shortname; 1283 u64 timestamp0, timestamp = sample->time; 1284 s64 delta; 1285 int cpu, this_cpu = sample->cpu; 1286 1287 BUG_ON(this_cpu >= MAX_CPUS || this_cpu < 0); 1288 1289 if (this_cpu > sched->max_cpu) 1290 sched->max_cpu = this_cpu; 1291 1292 timestamp0 = sched->cpu_last_switched[this_cpu]; 1293 sched->cpu_last_switched[this_cpu] = timestamp; 1294 if (timestamp0) 1295 delta = timestamp - timestamp0; 1296 else 1297 delta = 0; 1298 1299 if (delta < 0) { 1300 pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta); 1301 return -1; 1302 } 1303 1304 sched_out = machine__findnew_thread(machine, 0, prev_pid); 1305 sched_in = machine__findnew_thread(machine, 0, next_pid); 1306 1307 sched->curr_thread[this_cpu] = sched_in; 1308 1309 printf(" "); 1310 1311 new_shortname = 0; 1312 if (!sched_in->shortname[0]) { 1313 sched_in->shortname[0] = sched->next_shortname1; 1314 sched_in->shortname[1] = sched->next_shortname2; 1315 1316 if (sched->next_shortname1 < 'Z') { 1317 sched->next_shortname1++; 1318 } else { 1319 sched->next_shortname1='A'; 1320 if (sched->next_shortname2 < '9') { 1321 sched->next_shortname2++; 1322 } else { 1323 sched->next_shortname2='0'; 1324 } 1325 } 1326 new_shortname = 1; 1327 } 1328 1329 for (cpu = 0; cpu <= sched->max_cpu; cpu++) { 1330 if (cpu != this_cpu) 1331 printf(" "); 1332 else 1333 printf("*"); 1334 1335 if (sched->curr_thread[cpu]) { 1336 if (sched->curr_thread[cpu]->tid) 1337 printf("%2s ", sched->curr_thread[cpu]->shortname); 1338 else 1339 printf(". "); 1340 } else 1341 printf(" "); 1342 } 1343 1344 printf(" %12.6f secs ", (double)timestamp/1e9); 1345 if (new_shortname) { 1346 printf("%s => %s:%d\n", 1347 sched_in->shortname, sched_in->comm, sched_in->tid); 1348 } else { 1349 printf("\n"); 1350 } 1351 1352 return 0; 1353 } 1354 1355 static int process_sched_switch_event(struct perf_tool *tool, 1356 struct perf_evsel *evsel, 1357 struct perf_sample *sample, 1358 struct machine *machine) 1359 { 1360 struct perf_sched *sched = container_of(tool, struct perf_sched, tool); 1361 int this_cpu = sample->cpu, err = 0; 1362 u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"), 1363 next_pid = perf_evsel__intval(evsel, sample, "next_pid"); 1364 1365 if (sched->curr_pid[this_cpu] != (u32)-1) { 1366 /* 1367 * Are we trying to switch away a PID that is 1368 * not current? 1369 */ 1370 if (sched->curr_pid[this_cpu] != prev_pid) 1371 sched->nr_context_switch_bugs++; 1372 } 1373 1374 if (sched->tp_handler->switch_event) 1375 err = sched->tp_handler->switch_event(sched, evsel, sample, machine); 1376 1377 sched->curr_pid[this_cpu] = next_pid; 1378 return err; 1379 } 1380 1381 static int process_sched_runtime_event(struct perf_tool *tool, 1382 struct perf_evsel *evsel, 1383 struct perf_sample *sample, 1384 struct machine *machine) 1385 { 1386 struct perf_sched *sched = container_of(tool, struct perf_sched, tool); 1387 1388 if (sched->tp_handler->runtime_event) 1389 return sched->tp_handler->runtime_event(sched, evsel, sample, machine); 1390 1391 return 0; 1392 } 1393 1394 static int perf_sched__process_fork_event(struct perf_tool *tool, 1395 union perf_event *event, 1396 struct perf_sample *sample, 1397 struct machine *machine) 1398 { 1399 struct perf_sched *sched = container_of(tool, struct perf_sched, tool); 1400 1401 /* run the fork event through the perf machineruy */ 1402 perf_event__process_fork(tool, event, sample, machine); 1403 1404 /* and then run additional processing needed for this command */ 1405 if (sched->tp_handler->fork_event) 1406 return sched->tp_handler->fork_event(sched, event, machine); 1407 1408 return 0; 1409 } 1410 1411 static int process_sched_migrate_task_event(struct perf_tool *tool, 1412 struct perf_evsel *evsel, 1413 struct perf_sample *sample, 1414 struct machine *machine) 1415 { 1416 struct perf_sched *sched = container_of(tool, struct perf_sched, tool); 1417 1418 if (sched->tp_handler->migrate_task_event) 1419 return sched->tp_handler->migrate_task_event(sched, evsel, sample, machine); 1420 1421 return 0; 1422 } 1423 1424 typedef int (*tracepoint_handler)(struct perf_tool *tool, 1425 struct perf_evsel *evsel, 1426 struct perf_sample *sample, 1427 struct machine *machine); 1428 1429 static int perf_sched__process_tracepoint_sample(struct perf_tool *tool __maybe_unused, 1430 union perf_event *event __maybe_unused, 1431 struct perf_sample *sample, 1432 struct perf_evsel *evsel, 1433 struct machine *machine) 1434 { 1435 int err = 0; 1436 1437 evsel->hists.stats.total_period += sample->period; 1438 hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE); 1439 1440 if (evsel->handler.func != NULL) { 1441 tracepoint_handler f = evsel->handler.func; 1442 err = f(tool, evsel, sample, machine); 1443 } 1444 1445 return err; 1446 } 1447 1448 static int perf_sched__read_events(struct perf_sched *sched, 1449 struct perf_session **psession) 1450 { 1451 const struct perf_evsel_str_handler handlers[] = { 1452 { "sched:sched_switch", process_sched_switch_event, }, 1453 { "sched:sched_stat_runtime", process_sched_runtime_event, }, 1454 { "sched:sched_wakeup", process_sched_wakeup_event, }, 1455 { "sched:sched_wakeup_new", process_sched_wakeup_event, }, 1456 { "sched:sched_migrate_task", process_sched_migrate_task_event, }, 1457 }; 1458 struct perf_session *session; 1459 1460 session = perf_session__new(input_name, O_RDONLY, 0, false, &sched->tool); 1461 if (session == NULL) { 1462 pr_debug("No Memory for session\n"); 1463 return -1; 1464 } 1465 1466 if (perf_session__set_tracepoints_handlers(session, handlers)) 1467 goto out_delete; 1468 1469 if (perf_session__has_traces(session, "record -R")) { 1470 int err = perf_session__process_events(session, &sched->tool); 1471 if (err) { 1472 pr_err("Failed to process events, error %d", err); 1473 goto out_delete; 1474 } 1475 1476 sched->nr_events = session->stats.nr_events[0]; 1477 sched->nr_lost_events = session->stats.total_lost; 1478 sched->nr_lost_chunks = session->stats.nr_events[PERF_RECORD_LOST]; 1479 } 1480 1481 if (psession) 1482 *psession = session; 1483 else 1484 perf_session__delete(session); 1485 1486 return 0; 1487 1488 out_delete: 1489 perf_session__delete(session); 1490 return -1; 1491 } 1492 1493 static void print_bad_events(struct perf_sched *sched) 1494 { 1495 if (sched->nr_unordered_timestamps && sched->nr_timestamps) { 1496 printf(" INFO: %.3f%% unordered timestamps (%ld out of %ld)\n", 1497 (double)sched->nr_unordered_timestamps/(double)sched->nr_timestamps*100.0, 1498 sched->nr_unordered_timestamps, sched->nr_timestamps); 1499 } 1500 if (sched->nr_lost_events && sched->nr_events) { 1501 printf(" INFO: %.3f%% lost events (%ld out of %ld, in %ld chunks)\n", 1502 (double)sched->nr_lost_events/(double)sched->nr_events * 100.0, 1503 sched->nr_lost_events, sched->nr_events, sched->nr_lost_chunks); 1504 } 1505 if (sched->nr_state_machine_bugs && sched->nr_timestamps) { 1506 printf(" INFO: %.3f%% state machine bugs (%ld out of %ld)", 1507 (double)sched->nr_state_machine_bugs/(double)sched->nr_timestamps*100.0, 1508 sched->nr_state_machine_bugs, sched->nr_timestamps); 1509 if (sched->nr_lost_events) 1510 printf(" (due to lost events?)"); 1511 printf("\n"); 1512 } 1513 if (sched->nr_context_switch_bugs && sched->nr_timestamps) { 1514 printf(" INFO: %.3f%% context switch bugs (%ld out of %ld)", 1515 (double)sched->nr_context_switch_bugs/(double)sched->nr_timestamps*100.0, 1516 sched->nr_context_switch_bugs, sched->nr_timestamps); 1517 if (sched->nr_lost_events) 1518 printf(" (due to lost events?)"); 1519 printf("\n"); 1520 } 1521 } 1522 1523 static int perf_sched__lat(struct perf_sched *sched) 1524 { 1525 struct rb_node *next; 1526 struct perf_session *session; 1527 1528 setup_pager(); 1529 1530 /* save session -- references to threads are held in work_list */ 1531 if (perf_sched__read_events(sched, &session)) 1532 return -1; 1533 1534 perf_sched__sort_lat(sched); 1535 1536 printf("\n ---------------------------------------------------------------------------------------------------------------\n"); 1537 printf(" Task | Runtime ms | Switches | Average delay ms | Maximum delay ms | Maximum delay at |\n"); 1538 printf(" ---------------------------------------------------------------------------------------------------------------\n"); 1539 1540 next = rb_first(&sched->sorted_atom_root); 1541 1542 while (next) { 1543 struct work_atoms *work_list; 1544 1545 work_list = rb_entry(next, struct work_atoms, node); 1546 output_lat_thread(sched, work_list); 1547 next = rb_next(next); 1548 } 1549 1550 printf(" -----------------------------------------------------------------------------------------\n"); 1551 printf(" TOTAL: |%11.3f ms |%9" PRIu64 " |\n", 1552 (double)sched->all_runtime / 1e6, sched->all_count); 1553 1554 printf(" ---------------------------------------------------\n"); 1555 1556 print_bad_events(sched); 1557 printf("\n"); 1558 1559 perf_session__delete(session); 1560 return 0; 1561 } 1562 1563 static int perf_sched__map(struct perf_sched *sched) 1564 { 1565 sched->max_cpu = sysconf(_SC_NPROCESSORS_CONF); 1566 1567 setup_pager(); 1568 if (perf_sched__read_events(sched, NULL)) 1569 return -1; 1570 print_bad_events(sched); 1571 return 0; 1572 } 1573 1574 static int perf_sched__replay(struct perf_sched *sched) 1575 { 1576 unsigned long i; 1577 1578 calibrate_run_measurement_overhead(sched); 1579 calibrate_sleep_measurement_overhead(sched); 1580 1581 test_calibrations(sched); 1582 1583 if (perf_sched__read_events(sched, NULL)) 1584 return -1; 1585 1586 printf("nr_run_events: %ld\n", sched->nr_run_events); 1587 printf("nr_sleep_events: %ld\n", sched->nr_sleep_events); 1588 printf("nr_wakeup_events: %ld\n", sched->nr_wakeup_events); 1589 1590 if (sched->targetless_wakeups) 1591 printf("target-less wakeups: %ld\n", sched->targetless_wakeups); 1592 if (sched->multitarget_wakeups) 1593 printf("multi-target wakeups: %ld\n", sched->multitarget_wakeups); 1594 if (sched->nr_run_events_optimized) 1595 printf("run atoms optimized: %ld\n", 1596 sched->nr_run_events_optimized); 1597 1598 print_task_traces(sched); 1599 add_cross_task_wakeups(sched); 1600 1601 create_tasks(sched); 1602 printf("------------------------------------------------------------\n"); 1603 for (i = 0; i < sched->replay_repeat; i++) 1604 run_one_test(sched); 1605 1606 return 0; 1607 } 1608 1609 static void setup_sorting(struct perf_sched *sched, const struct option *options, 1610 const char * const usage_msg[]) 1611 { 1612 char *tmp, *tok, *str = strdup(sched->sort_order); 1613 1614 for (tok = strtok_r(str, ", ", &tmp); 1615 tok; tok = strtok_r(NULL, ", ", &tmp)) { 1616 if (sort_dimension__add(tok, &sched->sort_list) < 0) { 1617 error("Unknown --sort key: `%s'", tok); 1618 usage_with_options(usage_msg, options); 1619 } 1620 } 1621 1622 free(str); 1623 1624 sort_dimension__add("pid", &sched->cmp_pid); 1625 } 1626 1627 static int __cmd_record(int argc, const char **argv) 1628 { 1629 unsigned int rec_argc, i, j; 1630 const char **rec_argv; 1631 const char * const record_args[] = { 1632 "record", 1633 "-a", 1634 "-R", 1635 "-m", "1024", 1636 "-c", "1", 1637 "-e", "sched:sched_switch", 1638 "-e", "sched:sched_stat_wait", 1639 "-e", "sched:sched_stat_sleep", 1640 "-e", "sched:sched_stat_iowait", 1641 "-e", "sched:sched_stat_runtime", 1642 "-e", "sched:sched_process_fork", 1643 "-e", "sched:sched_wakeup", 1644 "-e", "sched:sched_migrate_task", 1645 }; 1646 1647 rec_argc = ARRAY_SIZE(record_args) + argc - 1; 1648 rec_argv = calloc(rec_argc + 1, sizeof(char *)); 1649 1650 if (rec_argv == NULL) 1651 return -ENOMEM; 1652 1653 for (i = 0; i < ARRAY_SIZE(record_args); i++) 1654 rec_argv[i] = strdup(record_args[i]); 1655 1656 for (j = 1; j < (unsigned int)argc; j++, i++) 1657 rec_argv[i] = argv[j]; 1658 1659 BUG_ON(i != rec_argc); 1660 1661 return cmd_record(i, rec_argv, NULL); 1662 } 1663 1664 static const char default_sort_order[] = "avg, max, switch, runtime"; 1665 static struct perf_sched sched = { 1666 .tool = { 1667 .sample = perf_sched__process_tracepoint_sample, 1668 .comm = perf_event__process_comm, 1669 .lost = perf_event__process_lost, 1670 .fork = perf_sched__process_fork_event, 1671 .ordered_samples = true, 1672 }, 1673 .cmp_pid = LIST_HEAD_INIT(sched.cmp_pid), 1674 .sort_list = LIST_HEAD_INIT(sched.sort_list), 1675 .start_work_mutex = PTHREAD_MUTEX_INITIALIZER, 1676 .work_done_wait_mutex = PTHREAD_MUTEX_INITIALIZER, 1677 .curr_pid = { [0 ... MAX_CPUS - 1] = -1 }, 1678 .sort_order = default_sort_order, 1679 .replay_repeat = 10, 1680 .profile_cpu = -1, 1681 .next_shortname1 = 'A', 1682 .next_shortname2 = '0', 1683 }; 1684 1685 int cmd_sched(int argc, const char **argv, const char *prefix __maybe_unused) 1686 { 1687 const struct option latency_options[] = { 1688 OPT_STRING('s', "sort", &sched.sort_order, "key[,key2...]", 1689 "sort by key(s): runtime, switch, avg, max"), 1690 OPT_INCR('v', "verbose", &verbose, 1691 "be more verbose (show symbol address, etc)"), 1692 OPT_INTEGER('C', "CPU", &sched.profile_cpu, 1693 "CPU to profile on"), 1694 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, 1695 "dump raw trace in ASCII"), 1696 OPT_END() 1697 }; 1698 const struct option replay_options[] = { 1699 OPT_UINTEGER('r', "repeat", &sched.replay_repeat, 1700 "repeat the workload replay N times (-1: infinite)"), 1701 OPT_INCR('v', "verbose", &verbose, 1702 "be more verbose (show symbol address, etc)"), 1703 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, 1704 "dump raw trace in ASCII"), 1705 OPT_END() 1706 }; 1707 const struct option sched_options[] = { 1708 OPT_STRING('i', "input", &input_name, "file", 1709 "input file name"), 1710 OPT_INCR('v', "verbose", &verbose, 1711 "be more verbose (show symbol address, etc)"), 1712 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, 1713 "dump raw trace in ASCII"), 1714 OPT_END() 1715 }; 1716 const char * const latency_usage[] = { 1717 "perf sched latency [<options>]", 1718 NULL 1719 }; 1720 const char * const replay_usage[] = { 1721 "perf sched replay [<options>]", 1722 NULL 1723 }; 1724 const char * const sched_usage[] = { 1725 "perf sched [<options>] {record|latency|map|replay|script}", 1726 NULL 1727 }; 1728 struct trace_sched_handler lat_ops = { 1729 .wakeup_event = latency_wakeup_event, 1730 .switch_event = latency_switch_event, 1731 .runtime_event = latency_runtime_event, 1732 .migrate_task_event = latency_migrate_task_event, 1733 }; 1734 struct trace_sched_handler map_ops = { 1735 .switch_event = map_switch_event, 1736 }; 1737 struct trace_sched_handler replay_ops = { 1738 .wakeup_event = replay_wakeup_event, 1739 .switch_event = replay_switch_event, 1740 .fork_event = replay_fork_event, 1741 }; 1742 1743 argc = parse_options(argc, argv, sched_options, sched_usage, 1744 PARSE_OPT_STOP_AT_NON_OPTION); 1745 if (!argc) 1746 usage_with_options(sched_usage, sched_options); 1747 1748 /* 1749 * Aliased to 'perf script' for now: 1750 */ 1751 if (!strcmp(argv[0], "script")) 1752 return cmd_script(argc, argv, prefix); 1753 1754 symbol__init(); 1755 if (!strncmp(argv[0], "rec", 3)) { 1756 return __cmd_record(argc, argv); 1757 } else if (!strncmp(argv[0], "lat", 3)) { 1758 sched.tp_handler = &lat_ops; 1759 if (argc > 1) { 1760 argc = parse_options(argc, argv, latency_options, latency_usage, 0); 1761 if (argc) 1762 usage_with_options(latency_usage, latency_options); 1763 } 1764 setup_sorting(&sched, latency_options, latency_usage); 1765 return perf_sched__lat(&sched); 1766 } else if (!strcmp(argv[0], "map")) { 1767 sched.tp_handler = &map_ops; 1768 setup_sorting(&sched, latency_options, latency_usage); 1769 return perf_sched__map(&sched); 1770 } else if (!strncmp(argv[0], "rep", 3)) { 1771 sched.tp_handler = &replay_ops; 1772 if (argc) { 1773 argc = parse_options(argc, argv, replay_options, replay_usage, 0); 1774 if (argc) 1775 usage_with_options(replay_usage, replay_options); 1776 } 1777 return perf_sched__replay(&sched); 1778 } else { 1779 usage_with_options(sched_usage, sched_options); 1780 } 1781 1782 return 0; 1783 } 1784