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 struct timespec ts; 177 178 clock_gettime(CLOCK_MONOTONIC, &ts); 179 180 return ts.tv_sec * 1000000000ULL + ts.tv_nsec; 181 } 182 183 static void burn_nsecs(struct perf_sched *sched, u64 nsecs) 184 { 185 u64 T0 = get_nsecs(), T1; 186 187 do { 188 T1 = get_nsecs(); 189 } while (T1 + sched->run_measurement_overhead < T0 + nsecs); 190 } 191 192 static void sleep_nsecs(u64 nsecs) 193 { 194 struct timespec ts; 195 196 ts.tv_nsec = nsecs % 999999999; 197 ts.tv_sec = nsecs / 999999999; 198 199 nanosleep(&ts, NULL); 200 } 201 202 static void calibrate_run_measurement_overhead(struct perf_sched *sched) 203 { 204 u64 T0, T1, delta, min_delta = 1000000000ULL; 205 int i; 206 207 for (i = 0; i < 10; i++) { 208 T0 = get_nsecs(); 209 burn_nsecs(sched, 0); 210 T1 = get_nsecs(); 211 delta = T1-T0; 212 min_delta = min(min_delta, delta); 213 } 214 sched->run_measurement_overhead = min_delta; 215 216 printf("run measurement overhead: %" PRIu64 " nsecs\n", min_delta); 217 } 218 219 static void calibrate_sleep_measurement_overhead(struct perf_sched *sched) 220 { 221 u64 T0, T1, delta, min_delta = 1000000000ULL; 222 int i; 223 224 for (i = 0; i < 10; i++) { 225 T0 = get_nsecs(); 226 sleep_nsecs(10000); 227 T1 = get_nsecs(); 228 delta = T1-T0; 229 min_delta = min(min_delta, delta); 230 } 231 min_delta -= 10000; 232 sched->sleep_measurement_overhead = min_delta; 233 234 printf("sleep measurement overhead: %" PRIu64 " nsecs\n", min_delta); 235 } 236 237 static struct sched_atom * 238 get_new_event(struct task_desc *task, u64 timestamp) 239 { 240 struct sched_atom *event = zalloc(sizeof(*event)); 241 unsigned long idx = task->nr_events; 242 size_t size; 243 244 event->timestamp = timestamp; 245 event->nr = idx; 246 247 task->nr_events++; 248 size = sizeof(struct sched_atom *) * task->nr_events; 249 task->atoms = realloc(task->atoms, size); 250 BUG_ON(!task->atoms); 251 252 task->atoms[idx] = event; 253 254 return event; 255 } 256 257 static struct sched_atom *last_event(struct task_desc *task) 258 { 259 if (!task->nr_events) 260 return NULL; 261 262 return task->atoms[task->nr_events - 1]; 263 } 264 265 static void add_sched_event_run(struct perf_sched *sched, struct task_desc *task, 266 u64 timestamp, u64 duration) 267 { 268 struct sched_atom *event, *curr_event = last_event(task); 269 270 /* 271 * optimize an existing RUN event by merging this one 272 * to it: 273 */ 274 if (curr_event && curr_event->type == SCHED_EVENT_RUN) { 275 sched->nr_run_events_optimized++; 276 curr_event->duration += duration; 277 return; 278 } 279 280 event = get_new_event(task, timestamp); 281 282 event->type = SCHED_EVENT_RUN; 283 event->duration = duration; 284 285 sched->nr_run_events++; 286 } 287 288 static void add_sched_event_wakeup(struct perf_sched *sched, struct task_desc *task, 289 u64 timestamp, struct task_desc *wakee) 290 { 291 struct sched_atom *event, *wakee_event; 292 293 event = get_new_event(task, timestamp); 294 event->type = SCHED_EVENT_WAKEUP; 295 event->wakee = wakee; 296 297 wakee_event = last_event(wakee); 298 if (!wakee_event || wakee_event->type != SCHED_EVENT_SLEEP) { 299 sched->targetless_wakeups++; 300 return; 301 } 302 if (wakee_event->wait_sem) { 303 sched->multitarget_wakeups++; 304 return; 305 } 306 307 wakee_event->wait_sem = zalloc(sizeof(*wakee_event->wait_sem)); 308 sem_init(wakee_event->wait_sem, 0, 0); 309 wakee_event->specific_wait = 1; 310 event->wait_sem = wakee_event->wait_sem; 311 312 sched->nr_wakeup_events++; 313 } 314 315 static void add_sched_event_sleep(struct perf_sched *sched, struct task_desc *task, 316 u64 timestamp, u64 task_state __maybe_unused) 317 { 318 struct sched_atom *event = get_new_event(task, timestamp); 319 320 event->type = SCHED_EVENT_SLEEP; 321 322 sched->nr_sleep_events++; 323 } 324 325 static struct task_desc *register_pid(struct perf_sched *sched, 326 unsigned long pid, const char *comm) 327 { 328 struct task_desc *task; 329 330 BUG_ON(pid >= MAX_PID); 331 332 task = sched->pid_to_task[pid]; 333 334 if (task) 335 return task; 336 337 task = zalloc(sizeof(*task)); 338 task->pid = pid; 339 task->nr = sched->nr_tasks; 340 strcpy(task->comm, comm); 341 /* 342 * every task starts in sleeping state - this gets ignored 343 * if there's no wakeup pointing to this sleep state: 344 */ 345 add_sched_event_sleep(sched, task, 0, 0); 346 347 sched->pid_to_task[pid] = task; 348 sched->nr_tasks++; 349 sched->tasks = realloc(sched->tasks, sched->nr_tasks * sizeof(struct task_task *)); 350 BUG_ON(!sched->tasks); 351 sched->tasks[task->nr] = task; 352 353 if (verbose) 354 printf("registered task #%ld, PID %ld (%s)\n", sched->nr_tasks, pid, comm); 355 356 return task; 357 } 358 359 360 static void print_task_traces(struct perf_sched *sched) 361 { 362 struct task_desc *task; 363 unsigned long i; 364 365 for (i = 0; i < sched->nr_tasks; i++) { 366 task = sched->tasks[i]; 367 printf("task %6ld (%20s:%10ld), nr_events: %ld\n", 368 task->nr, task->comm, task->pid, task->nr_events); 369 } 370 } 371 372 static void add_cross_task_wakeups(struct perf_sched *sched) 373 { 374 struct task_desc *task1, *task2; 375 unsigned long i, j; 376 377 for (i = 0; i < sched->nr_tasks; i++) { 378 task1 = sched->tasks[i]; 379 j = i + 1; 380 if (j == sched->nr_tasks) 381 j = 0; 382 task2 = sched->tasks[j]; 383 add_sched_event_wakeup(sched, task1, 0, task2); 384 } 385 } 386 387 static void perf_sched__process_event(struct perf_sched *sched, 388 struct sched_atom *atom) 389 { 390 int ret = 0; 391 392 switch (atom->type) { 393 case SCHED_EVENT_RUN: 394 burn_nsecs(sched, atom->duration); 395 break; 396 case SCHED_EVENT_SLEEP: 397 if (atom->wait_sem) 398 ret = sem_wait(atom->wait_sem); 399 BUG_ON(ret); 400 break; 401 case SCHED_EVENT_WAKEUP: 402 if (atom->wait_sem) 403 ret = sem_post(atom->wait_sem); 404 BUG_ON(ret); 405 break; 406 case SCHED_EVENT_MIGRATION: 407 break; 408 default: 409 BUG_ON(1); 410 } 411 } 412 413 static u64 get_cpu_usage_nsec_parent(void) 414 { 415 struct rusage ru; 416 u64 sum; 417 int err; 418 419 err = getrusage(RUSAGE_SELF, &ru); 420 BUG_ON(err); 421 422 sum = ru.ru_utime.tv_sec*1e9 + ru.ru_utime.tv_usec*1e3; 423 sum += ru.ru_stime.tv_sec*1e9 + ru.ru_stime.tv_usec*1e3; 424 425 return sum; 426 } 427 428 static int self_open_counters(void) 429 { 430 struct perf_event_attr attr; 431 int fd; 432 433 memset(&attr, 0, sizeof(attr)); 434 435 attr.type = PERF_TYPE_SOFTWARE; 436 attr.config = PERF_COUNT_SW_TASK_CLOCK; 437 438 fd = sys_perf_event_open(&attr, 0, -1, -1, 0); 439 440 if (fd < 0) 441 pr_err("Error: sys_perf_event_open() syscall returned " 442 "with %d (%s)\n", fd, strerror(errno)); 443 return fd; 444 } 445 446 static u64 get_cpu_usage_nsec_self(int fd) 447 { 448 u64 runtime; 449 int ret; 450 451 ret = read(fd, &runtime, sizeof(runtime)); 452 BUG_ON(ret != sizeof(runtime)); 453 454 return runtime; 455 } 456 457 struct sched_thread_parms { 458 struct task_desc *task; 459 struct perf_sched *sched; 460 }; 461 462 static void *thread_func(void *ctx) 463 { 464 struct sched_thread_parms *parms = ctx; 465 struct task_desc *this_task = parms->task; 466 struct perf_sched *sched = parms->sched; 467 u64 cpu_usage_0, cpu_usage_1; 468 unsigned long i, ret; 469 char comm2[22]; 470 int fd; 471 472 zfree(&parms); 473 474 sprintf(comm2, ":%s", this_task->comm); 475 prctl(PR_SET_NAME, comm2); 476 fd = self_open_counters(); 477 if (fd < 0) 478 return NULL; 479 again: 480 ret = sem_post(&this_task->ready_for_work); 481 BUG_ON(ret); 482 ret = pthread_mutex_lock(&sched->start_work_mutex); 483 BUG_ON(ret); 484 ret = pthread_mutex_unlock(&sched->start_work_mutex); 485 BUG_ON(ret); 486 487 cpu_usage_0 = get_cpu_usage_nsec_self(fd); 488 489 for (i = 0; i < this_task->nr_events; i++) { 490 this_task->curr_event = i; 491 perf_sched__process_event(sched, this_task->atoms[i]); 492 } 493 494 cpu_usage_1 = get_cpu_usage_nsec_self(fd); 495 this_task->cpu_usage = cpu_usage_1 - cpu_usage_0; 496 ret = sem_post(&this_task->work_done_sem); 497 BUG_ON(ret); 498 499 ret = pthread_mutex_lock(&sched->work_done_wait_mutex); 500 BUG_ON(ret); 501 ret = pthread_mutex_unlock(&sched->work_done_wait_mutex); 502 BUG_ON(ret); 503 504 goto again; 505 } 506 507 static void create_tasks(struct perf_sched *sched) 508 { 509 struct task_desc *task; 510 pthread_attr_t attr; 511 unsigned long i; 512 int err; 513 514 err = pthread_attr_init(&attr); 515 BUG_ON(err); 516 err = pthread_attr_setstacksize(&attr, 517 (size_t) max(16 * 1024, PTHREAD_STACK_MIN)); 518 BUG_ON(err); 519 err = pthread_mutex_lock(&sched->start_work_mutex); 520 BUG_ON(err); 521 err = pthread_mutex_lock(&sched->work_done_wait_mutex); 522 BUG_ON(err); 523 for (i = 0; i < sched->nr_tasks; i++) { 524 struct sched_thread_parms *parms = malloc(sizeof(*parms)); 525 BUG_ON(parms == NULL); 526 parms->task = task = sched->tasks[i]; 527 parms->sched = sched; 528 sem_init(&task->sleep_sem, 0, 0); 529 sem_init(&task->ready_for_work, 0, 0); 530 sem_init(&task->work_done_sem, 0, 0); 531 task->curr_event = 0; 532 err = pthread_create(&task->thread, &attr, thread_func, parms); 533 BUG_ON(err); 534 } 535 } 536 537 static void wait_for_tasks(struct perf_sched *sched) 538 { 539 u64 cpu_usage_0, cpu_usage_1; 540 struct task_desc *task; 541 unsigned long i, ret; 542 543 sched->start_time = get_nsecs(); 544 sched->cpu_usage = 0; 545 pthread_mutex_unlock(&sched->work_done_wait_mutex); 546 547 for (i = 0; i < sched->nr_tasks; i++) { 548 task = sched->tasks[i]; 549 ret = sem_wait(&task->ready_for_work); 550 BUG_ON(ret); 551 sem_init(&task->ready_for_work, 0, 0); 552 } 553 ret = pthread_mutex_lock(&sched->work_done_wait_mutex); 554 BUG_ON(ret); 555 556 cpu_usage_0 = get_cpu_usage_nsec_parent(); 557 558 pthread_mutex_unlock(&sched->start_work_mutex); 559 560 for (i = 0; i < sched->nr_tasks; i++) { 561 task = sched->tasks[i]; 562 ret = sem_wait(&task->work_done_sem); 563 BUG_ON(ret); 564 sem_init(&task->work_done_sem, 0, 0); 565 sched->cpu_usage += task->cpu_usage; 566 task->cpu_usage = 0; 567 } 568 569 cpu_usage_1 = get_cpu_usage_nsec_parent(); 570 if (!sched->runavg_cpu_usage) 571 sched->runavg_cpu_usage = sched->cpu_usage; 572 sched->runavg_cpu_usage = (sched->runavg_cpu_usage * 9 + sched->cpu_usage) / 10; 573 574 sched->parent_cpu_usage = cpu_usage_1 - cpu_usage_0; 575 if (!sched->runavg_parent_cpu_usage) 576 sched->runavg_parent_cpu_usage = sched->parent_cpu_usage; 577 sched->runavg_parent_cpu_usage = (sched->runavg_parent_cpu_usage * 9 + 578 sched->parent_cpu_usage)/10; 579 580 ret = pthread_mutex_lock(&sched->start_work_mutex); 581 BUG_ON(ret); 582 583 for (i = 0; i < sched->nr_tasks; i++) { 584 task = sched->tasks[i]; 585 sem_init(&task->sleep_sem, 0, 0); 586 task->curr_event = 0; 587 } 588 } 589 590 static void run_one_test(struct perf_sched *sched) 591 { 592 u64 T0, T1, delta, avg_delta, fluct; 593 594 T0 = get_nsecs(); 595 wait_for_tasks(sched); 596 T1 = get_nsecs(); 597 598 delta = T1 - T0; 599 sched->sum_runtime += delta; 600 sched->nr_runs++; 601 602 avg_delta = sched->sum_runtime / sched->nr_runs; 603 if (delta < avg_delta) 604 fluct = avg_delta - delta; 605 else 606 fluct = delta - avg_delta; 607 sched->sum_fluct += fluct; 608 if (!sched->run_avg) 609 sched->run_avg = delta; 610 sched->run_avg = (sched->run_avg * 9 + delta) / 10; 611 612 printf("#%-3ld: %0.3f, ", sched->nr_runs, (double)delta / 1000000.0); 613 614 printf("ravg: %0.2f, ", (double)sched->run_avg / 1e6); 615 616 printf("cpu: %0.2f / %0.2f", 617 (double)sched->cpu_usage / 1e6, (double)sched->runavg_cpu_usage / 1e6); 618 619 #if 0 620 /* 621 * rusage statistics done by the parent, these are less 622 * accurate than the sched->sum_exec_runtime based statistics: 623 */ 624 printf(" [%0.2f / %0.2f]", 625 (double)sched->parent_cpu_usage/1e6, 626 (double)sched->runavg_parent_cpu_usage/1e6); 627 #endif 628 629 printf("\n"); 630 631 if (sched->nr_sleep_corrections) 632 printf(" (%ld sleep corrections)\n", sched->nr_sleep_corrections); 633 sched->nr_sleep_corrections = 0; 634 } 635 636 static void test_calibrations(struct perf_sched *sched) 637 { 638 u64 T0, T1; 639 640 T0 = get_nsecs(); 641 burn_nsecs(sched, 1e6); 642 T1 = get_nsecs(); 643 644 printf("the run test took %" PRIu64 " nsecs\n", T1 - T0); 645 646 T0 = get_nsecs(); 647 sleep_nsecs(1e6); 648 T1 = get_nsecs(); 649 650 printf("the sleep test took %" PRIu64 " nsecs\n", T1 - T0); 651 } 652 653 static int 654 replay_wakeup_event(struct perf_sched *sched, 655 struct perf_evsel *evsel, struct perf_sample *sample, 656 struct machine *machine __maybe_unused) 657 { 658 const char *comm = perf_evsel__strval(evsel, sample, "comm"); 659 const u32 pid = perf_evsel__intval(evsel, sample, "pid"); 660 struct task_desc *waker, *wakee; 661 662 if (verbose) { 663 printf("sched_wakeup event %p\n", evsel); 664 665 printf(" ... pid %d woke up %s/%d\n", sample->tid, comm, pid); 666 } 667 668 waker = register_pid(sched, sample->tid, "<unknown>"); 669 wakee = register_pid(sched, pid, comm); 670 671 add_sched_event_wakeup(sched, waker, sample->time, wakee); 672 return 0; 673 } 674 675 static int replay_switch_event(struct perf_sched *sched, 676 struct perf_evsel *evsel, 677 struct perf_sample *sample, 678 struct machine *machine __maybe_unused) 679 { 680 const char *prev_comm = perf_evsel__strval(evsel, sample, "prev_comm"), 681 *next_comm = perf_evsel__strval(evsel, sample, "next_comm"); 682 const u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"), 683 next_pid = perf_evsel__intval(evsel, sample, "next_pid"); 684 const u64 prev_state = perf_evsel__intval(evsel, sample, "prev_state"); 685 struct task_desc *prev, __maybe_unused *next; 686 u64 timestamp0, timestamp = sample->time; 687 int cpu = sample->cpu; 688 s64 delta; 689 690 if (verbose) 691 printf("sched_switch event %p\n", evsel); 692 693 if (cpu >= MAX_CPUS || cpu < 0) 694 return 0; 695 696 timestamp0 = sched->cpu_last_switched[cpu]; 697 if (timestamp0) 698 delta = timestamp - timestamp0; 699 else 700 delta = 0; 701 702 if (delta < 0) { 703 pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta); 704 return -1; 705 } 706 707 pr_debug(" ... switch from %s/%d to %s/%d [ran %" PRIu64 " nsecs]\n", 708 prev_comm, prev_pid, next_comm, next_pid, delta); 709 710 prev = register_pid(sched, prev_pid, prev_comm); 711 next = register_pid(sched, next_pid, next_comm); 712 713 sched->cpu_last_switched[cpu] = timestamp; 714 715 add_sched_event_run(sched, prev, timestamp, delta); 716 add_sched_event_sleep(sched, prev, timestamp, prev_state); 717 718 return 0; 719 } 720 721 static int replay_fork_event(struct perf_sched *sched, 722 union perf_event *event, 723 struct machine *machine) 724 { 725 struct thread *child, *parent; 726 727 child = machine__findnew_thread(machine, event->fork.pid, 728 event->fork.tid); 729 parent = machine__findnew_thread(machine, event->fork.ppid, 730 event->fork.ptid); 731 732 if (child == NULL || parent == NULL) { 733 pr_debug("thread does not exist on fork event: child %p, parent %p\n", 734 child, parent); 735 return 0; 736 } 737 738 if (verbose) { 739 printf("fork event\n"); 740 printf("... parent: %s/%d\n", thread__comm_str(parent), parent->tid); 741 printf("... child: %s/%d\n", thread__comm_str(child), child->tid); 742 } 743 744 register_pid(sched, parent->tid, thread__comm_str(parent)); 745 register_pid(sched, child->tid, thread__comm_str(child)); 746 return 0; 747 } 748 749 struct sort_dimension { 750 const char *name; 751 sort_fn_t cmp; 752 struct list_head list; 753 }; 754 755 static int 756 thread_lat_cmp(struct list_head *list, struct work_atoms *l, struct work_atoms *r) 757 { 758 struct sort_dimension *sort; 759 int ret = 0; 760 761 BUG_ON(list_empty(list)); 762 763 list_for_each_entry(sort, list, list) { 764 ret = sort->cmp(l, r); 765 if (ret) 766 return ret; 767 } 768 769 return ret; 770 } 771 772 static struct work_atoms * 773 thread_atoms_search(struct rb_root *root, struct thread *thread, 774 struct list_head *sort_list) 775 { 776 struct rb_node *node = root->rb_node; 777 struct work_atoms key = { .thread = thread }; 778 779 while (node) { 780 struct work_atoms *atoms; 781 int cmp; 782 783 atoms = container_of(node, struct work_atoms, node); 784 785 cmp = thread_lat_cmp(sort_list, &key, atoms); 786 if (cmp > 0) 787 node = node->rb_left; 788 else if (cmp < 0) 789 node = node->rb_right; 790 else { 791 BUG_ON(thread != atoms->thread); 792 return atoms; 793 } 794 } 795 return NULL; 796 } 797 798 static void 799 __thread_latency_insert(struct rb_root *root, struct work_atoms *data, 800 struct list_head *sort_list) 801 { 802 struct rb_node **new = &(root->rb_node), *parent = NULL; 803 804 while (*new) { 805 struct work_atoms *this; 806 int cmp; 807 808 this = container_of(*new, struct work_atoms, node); 809 parent = *new; 810 811 cmp = thread_lat_cmp(sort_list, data, this); 812 813 if (cmp > 0) 814 new = &((*new)->rb_left); 815 else 816 new = &((*new)->rb_right); 817 } 818 819 rb_link_node(&data->node, parent, new); 820 rb_insert_color(&data->node, root); 821 } 822 823 static int thread_atoms_insert(struct perf_sched *sched, struct thread *thread) 824 { 825 struct work_atoms *atoms = zalloc(sizeof(*atoms)); 826 if (!atoms) { 827 pr_err("No memory at %s\n", __func__); 828 return -1; 829 } 830 831 atoms->thread = thread; 832 INIT_LIST_HEAD(&atoms->work_list); 833 __thread_latency_insert(&sched->atom_root, atoms, &sched->cmp_pid); 834 return 0; 835 } 836 837 static char sched_out_state(u64 prev_state) 838 { 839 const char *str = TASK_STATE_TO_CHAR_STR; 840 841 return str[prev_state]; 842 } 843 844 static int 845 add_sched_out_event(struct work_atoms *atoms, 846 char run_state, 847 u64 timestamp) 848 { 849 struct work_atom *atom = zalloc(sizeof(*atom)); 850 if (!atom) { 851 pr_err("Non memory at %s", __func__); 852 return -1; 853 } 854 855 atom->sched_out_time = timestamp; 856 857 if (run_state == 'R') { 858 atom->state = THREAD_WAIT_CPU; 859 atom->wake_up_time = atom->sched_out_time; 860 } 861 862 list_add_tail(&atom->list, &atoms->work_list); 863 return 0; 864 } 865 866 static void 867 add_runtime_event(struct work_atoms *atoms, u64 delta, 868 u64 timestamp __maybe_unused) 869 { 870 struct work_atom *atom; 871 872 BUG_ON(list_empty(&atoms->work_list)); 873 874 atom = list_entry(atoms->work_list.prev, struct work_atom, list); 875 876 atom->runtime += delta; 877 atoms->total_runtime += delta; 878 } 879 880 static void 881 add_sched_in_event(struct work_atoms *atoms, u64 timestamp) 882 { 883 struct work_atom *atom; 884 u64 delta; 885 886 if (list_empty(&atoms->work_list)) 887 return; 888 889 atom = list_entry(atoms->work_list.prev, struct work_atom, list); 890 891 if (atom->state != THREAD_WAIT_CPU) 892 return; 893 894 if (timestamp < atom->wake_up_time) { 895 atom->state = THREAD_IGNORE; 896 return; 897 } 898 899 atom->state = THREAD_SCHED_IN; 900 atom->sched_in_time = timestamp; 901 902 delta = atom->sched_in_time - atom->wake_up_time; 903 atoms->total_lat += delta; 904 if (delta > atoms->max_lat) { 905 atoms->max_lat = delta; 906 atoms->max_lat_at = timestamp; 907 } 908 atoms->nb_atoms++; 909 } 910 911 static int latency_switch_event(struct perf_sched *sched, 912 struct perf_evsel *evsel, 913 struct perf_sample *sample, 914 struct machine *machine) 915 { 916 const u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"), 917 next_pid = perf_evsel__intval(evsel, sample, "next_pid"); 918 const u64 prev_state = perf_evsel__intval(evsel, sample, "prev_state"); 919 struct work_atoms *out_events, *in_events; 920 struct thread *sched_out, *sched_in; 921 u64 timestamp0, timestamp = sample->time; 922 int cpu = sample->cpu; 923 s64 delta; 924 925 BUG_ON(cpu >= MAX_CPUS || cpu < 0); 926 927 timestamp0 = sched->cpu_last_switched[cpu]; 928 sched->cpu_last_switched[cpu] = timestamp; 929 if (timestamp0) 930 delta = timestamp - timestamp0; 931 else 932 delta = 0; 933 934 if (delta < 0) { 935 pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta); 936 return -1; 937 } 938 939 sched_out = machine__findnew_thread(machine, 0, prev_pid); 940 sched_in = machine__findnew_thread(machine, 0, next_pid); 941 942 out_events = thread_atoms_search(&sched->atom_root, sched_out, &sched->cmp_pid); 943 if (!out_events) { 944 if (thread_atoms_insert(sched, sched_out)) 945 return -1; 946 out_events = thread_atoms_search(&sched->atom_root, sched_out, &sched->cmp_pid); 947 if (!out_events) { 948 pr_err("out-event: Internal tree error"); 949 return -1; 950 } 951 } 952 if (add_sched_out_event(out_events, sched_out_state(prev_state), timestamp)) 953 return -1; 954 955 in_events = thread_atoms_search(&sched->atom_root, sched_in, &sched->cmp_pid); 956 if (!in_events) { 957 if (thread_atoms_insert(sched, sched_in)) 958 return -1; 959 in_events = thread_atoms_search(&sched->atom_root, sched_in, &sched->cmp_pid); 960 if (!in_events) { 961 pr_err("in-event: Internal tree error"); 962 return -1; 963 } 964 /* 965 * Take came in we have not heard about yet, 966 * add in an initial atom in runnable state: 967 */ 968 if (add_sched_out_event(in_events, 'R', timestamp)) 969 return -1; 970 } 971 add_sched_in_event(in_events, timestamp); 972 973 return 0; 974 } 975 976 static int latency_runtime_event(struct perf_sched *sched, 977 struct perf_evsel *evsel, 978 struct perf_sample *sample, 979 struct machine *machine) 980 { 981 const u32 pid = perf_evsel__intval(evsel, sample, "pid"); 982 const u64 runtime = perf_evsel__intval(evsel, sample, "runtime"); 983 struct thread *thread = machine__findnew_thread(machine, 0, pid); 984 struct work_atoms *atoms = thread_atoms_search(&sched->atom_root, thread, &sched->cmp_pid); 985 u64 timestamp = sample->time; 986 int cpu = sample->cpu; 987 988 BUG_ON(cpu >= MAX_CPUS || cpu < 0); 989 if (!atoms) { 990 if (thread_atoms_insert(sched, thread)) 991 return -1; 992 atoms = thread_atoms_search(&sched->atom_root, thread, &sched->cmp_pid); 993 if (!atoms) { 994 pr_err("in-event: Internal tree error"); 995 return -1; 996 } 997 if (add_sched_out_event(atoms, 'R', timestamp)) 998 return -1; 999 } 1000 1001 add_runtime_event(atoms, runtime, timestamp); 1002 return 0; 1003 } 1004 1005 static int latency_wakeup_event(struct perf_sched *sched, 1006 struct perf_evsel *evsel, 1007 struct perf_sample *sample, 1008 struct machine *machine) 1009 { 1010 const u32 pid = perf_evsel__intval(evsel, sample, "pid"), 1011 success = perf_evsel__intval(evsel, sample, "success"); 1012 struct work_atoms *atoms; 1013 struct work_atom *atom; 1014 struct thread *wakee; 1015 u64 timestamp = sample->time; 1016 1017 /* Note for later, it may be interesting to observe the failing cases */ 1018 if (!success) 1019 return 0; 1020 1021 wakee = machine__findnew_thread(machine, 0, pid); 1022 atoms = thread_atoms_search(&sched->atom_root, wakee, &sched->cmp_pid); 1023 if (!atoms) { 1024 if (thread_atoms_insert(sched, wakee)) 1025 return -1; 1026 atoms = thread_atoms_search(&sched->atom_root, wakee, &sched->cmp_pid); 1027 if (!atoms) { 1028 pr_err("wakeup-event: Internal tree error"); 1029 return -1; 1030 } 1031 if (add_sched_out_event(atoms, 'S', timestamp)) 1032 return -1; 1033 } 1034 1035 BUG_ON(list_empty(&atoms->work_list)); 1036 1037 atom = list_entry(atoms->work_list.prev, struct work_atom, list); 1038 1039 /* 1040 * You WILL be missing events if you've recorded only 1041 * one CPU, or are only looking at only one, so don't 1042 * make useless noise. 1043 */ 1044 if (sched->profile_cpu == -1 && atom->state != THREAD_SLEEPING) 1045 sched->nr_state_machine_bugs++; 1046 1047 sched->nr_timestamps++; 1048 if (atom->sched_out_time > timestamp) { 1049 sched->nr_unordered_timestamps++; 1050 return 0; 1051 } 1052 1053 atom->state = THREAD_WAIT_CPU; 1054 atom->wake_up_time = timestamp; 1055 return 0; 1056 } 1057 1058 static int latency_migrate_task_event(struct perf_sched *sched, 1059 struct perf_evsel *evsel, 1060 struct perf_sample *sample, 1061 struct machine *machine) 1062 { 1063 const u32 pid = perf_evsel__intval(evsel, sample, "pid"); 1064 u64 timestamp = sample->time; 1065 struct work_atoms *atoms; 1066 struct work_atom *atom; 1067 struct thread *migrant; 1068 1069 /* 1070 * Only need to worry about migration when profiling one CPU. 1071 */ 1072 if (sched->profile_cpu == -1) 1073 return 0; 1074 1075 migrant = machine__findnew_thread(machine, 0, pid); 1076 atoms = thread_atoms_search(&sched->atom_root, migrant, &sched->cmp_pid); 1077 if (!atoms) { 1078 if (thread_atoms_insert(sched, migrant)) 1079 return -1; 1080 register_pid(sched, migrant->tid, thread__comm_str(migrant)); 1081 atoms = thread_atoms_search(&sched->atom_root, migrant, &sched->cmp_pid); 1082 if (!atoms) { 1083 pr_err("migration-event: Internal tree error"); 1084 return -1; 1085 } 1086 if (add_sched_out_event(atoms, 'R', timestamp)) 1087 return -1; 1088 } 1089 1090 BUG_ON(list_empty(&atoms->work_list)); 1091 1092 atom = list_entry(atoms->work_list.prev, struct work_atom, list); 1093 atom->sched_in_time = atom->sched_out_time = atom->wake_up_time = timestamp; 1094 1095 sched->nr_timestamps++; 1096 1097 if (atom->sched_out_time > timestamp) 1098 sched->nr_unordered_timestamps++; 1099 1100 return 0; 1101 } 1102 1103 static void output_lat_thread(struct perf_sched *sched, struct work_atoms *work_list) 1104 { 1105 int i; 1106 int ret; 1107 u64 avg; 1108 1109 if (!work_list->nb_atoms) 1110 return; 1111 /* 1112 * Ignore idle threads: 1113 */ 1114 if (!strcmp(thread__comm_str(work_list->thread), "swapper")) 1115 return; 1116 1117 sched->all_runtime += work_list->total_runtime; 1118 sched->all_count += work_list->nb_atoms; 1119 1120 ret = printf(" %s:%d ", thread__comm_str(work_list->thread), work_list->thread->tid); 1121 1122 for (i = 0; i < 24 - ret; i++) 1123 printf(" "); 1124 1125 avg = work_list->total_lat / work_list->nb_atoms; 1126 1127 printf("|%11.3f ms |%9" PRIu64 " | avg:%9.3f ms | max:%9.3f ms | max at: %13.6f s\n", 1128 (double)work_list->total_runtime / 1e6, 1129 work_list->nb_atoms, (double)avg / 1e6, 1130 (double)work_list->max_lat / 1e6, 1131 (double)work_list->max_lat_at / 1e9); 1132 } 1133 1134 static int pid_cmp(struct work_atoms *l, struct work_atoms *r) 1135 { 1136 if (l->thread->tid < r->thread->tid) 1137 return -1; 1138 if (l->thread->tid > r->thread->tid) 1139 return 1; 1140 1141 return 0; 1142 } 1143 1144 static int avg_cmp(struct work_atoms *l, struct work_atoms *r) 1145 { 1146 u64 avgl, avgr; 1147 1148 if (!l->nb_atoms) 1149 return -1; 1150 1151 if (!r->nb_atoms) 1152 return 1; 1153 1154 avgl = l->total_lat / l->nb_atoms; 1155 avgr = r->total_lat / r->nb_atoms; 1156 1157 if (avgl < avgr) 1158 return -1; 1159 if (avgl > avgr) 1160 return 1; 1161 1162 return 0; 1163 } 1164 1165 static int max_cmp(struct work_atoms *l, struct work_atoms *r) 1166 { 1167 if (l->max_lat < r->max_lat) 1168 return -1; 1169 if (l->max_lat > r->max_lat) 1170 return 1; 1171 1172 return 0; 1173 } 1174 1175 static int switch_cmp(struct work_atoms *l, struct work_atoms *r) 1176 { 1177 if (l->nb_atoms < r->nb_atoms) 1178 return -1; 1179 if (l->nb_atoms > r->nb_atoms) 1180 return 1; 1181 1182 return 0; 1183 } 1184 1185 static int runtime_cmp(struct work_atoms *l, struct work_atoms *r) 1186 { 1187 if (l->total_runtime < r->total_runtime) 1188 return -1; 1189 if (l->total_runtime > r->total_runtime) 1190 return 1; 1191 1192 return 0; 1193 } 1194 1195 static int sort_dimension__add(const char *tok, struct list_head *list) 1196 { 1197 size_t i; 1198 static struct sort_dimension avg_sort_dimension = { 1199 .name = "avg", 1200 .cmp = avg_cmp, 1201 }; 1202 static struct sort_dimension max_sort_dimension = { 1203 .name = "max", 1204 .cmp = max_cmp, 1205 }; 1206 static struct sort_dimension pid_sort_dimension = { 1207 .name = "pid", 1208 .cmp = pid_cmp, 1209 }; 1210 static struct sort_dimension runtime_sort_dimension = { 1211 .name = "runtime", 1212 .cmp = runtime_cmp, 1213 }; 1214 static struct sort_dimension switch_sort_dimension = { 1215 .name = "switch", 1216 .cmp = switch_cmp, 1217 }; 1218 struct sort_dimension *available_sorts[] = { 1219 &pid_sort_dimension, 1220 &avg_sort_dimension, 1221 &max_sort_dimension, 1222 &switch_sort_dimension, 1223 &runtime_sort_dimension, 1224 }; 1225 1226 for (i = 0; i < ARRAY_SIZE(available_sorts); i++) { 1227 if (!strcmp(available_sorts[i]->name, tok)) { 1228 list_add_tail(&available_sorts[i]->list, list); 1229 1230 return 0; 1231 } 1232 } 1233 1234 return -1; 1235 } 1236 1237 static void perf_sched__sort_lat(struct perf_sched *sched) 1238 { 1239 struct rb_node *node; 1240 1241 for (;;) { 1242 struct work_atoms *data; 1243 node = rb_first(&sched->atom_root); 1244 if (!node) 1245 break; 1246 1247 rb_erase(node, &sched->atom_root); 1248 data = rb_entry(node, struct work_atoms, node); 1249 __thread_latency_insert(&sched->sorted_atom_root, data, &sched->sort_list); 1250 } 1251 } 1252 1253 static int process_sched_wakeup_event(struct perf_tool *tool, 1254 struct perf_evsel *evsel, 1255 struct perf_sample *sample, 1256 struct machine *machine) 1257 { 1258 struct perf_sched *sched = container_of(tool, struct perf_sched, tool); 1259 1260 if (sched->tp_handler->wakeup_event) 1261 return sched->tp_handler->wakeup_event(sched, evsel, sample, machine); 1262 1263 return 0; 1264 } 1265 1266 static int map_switch_event(struct perf_sched *sched, struct perf_evsel *evsel, 1267 struct perf_sample *sample, struct machine *machine) 1268 { 1269 const u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"), 1270 next_pid = perf_evsel__intval(evsel, sample, "next_pid"); 1271 struct thread *sched_out __maybe_unused, *sched_in; 1272 int new_shortname; 1273 u64 timestamp0, timestamp = sample->time; 1274 s64 delta; 1275 int cpu, this_cpu = sample->cpu; 1276 1277 BUG_ON(this_cpu >= MAX_CPUS || this_cpu < 0); 1278 1279 if (this_cpu > sched->max_cpu) 1280 sched->max_cpu = this_cpu; 1281 1282 timestamp0 = sched->cpu_last_switched[this_cpu]; 1283 sched->cpu_last_switched[this_cpu] = timestamp; 1284 if (timestamp0) 1285 delta = timestamp - timestamp0; 1286 else 1287 delta = 0; 1288 1289 if (delta < 0) { 1290 pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta); 1291 return -1; 1292 } 1293 1294 sched_out = machine__findnew_thread(machine, 0, prev_pid); 1295 sched_in = machine__findnew_thread(machine, 0, next_pid); 1296 1297 sched->curr_thread[this_cpu] = sched_in; 1298 1299 printf(" "); 1300 1301 new_shortname = 0; 1302 if (!sched_in->shortname[0]) { 1303 sched_in->shortname[0] = sched->next_shortname1; 1304 sched_in->shortname[1] = sched->next_shortname2; 1305 1306 if (sched->next_shortname1 < 'Z') { 1307 sched->next_shortname1++; 1308 } else { 1309 sched->next_shortname1='A'; 1310 if (sched->next_shortname2 < '9') { 1311 sched->next_shortname2++; 1312 } else { 1313 sched->next_shortname2='0'; 1314 } 1315 } 1316 new_shortname = 1; 1317 } 1318 1319 for (cpu = 0; cpu <= sched->max_cpu; cpu++) { 1320 if (cpu != this_cpu) 1321 printf(" "); 1322 else 1323 printf("*"); 1324 1325 if (sched->curr_thread[cpu]) { 1326 if (sched->curr_thread[cpu]->tid) 1327 printf("%2s ", sched->curr_thread[cpu]->shortname); 1328 else 1329 printf(". "); 1330 } else 1331 printf(" "); 1332 } 1333 1334 printf(" %12.6f secs ", (double)timestamp/1e9); 1335 if (new_shortname) { 1336 printf("%s => %s:%d\n", 1337 sched_in->shortname, thread__comm_str(sched_in), sched_in->tid); 1338 } else { 1339 printf("\n"); 1340 } 1341 1342 return 0; 1343 } 1344 1345 static int process_sched_switch_event(struct perf_tool *tool, 1346 struct perf_evsel *evsel, 1347 struct perf_sample *sample, 1348 struct machine *machine) 1349 { 1350 struct perf_sched *sched = container_of(tool, struct perf_sched, tool); 1351 int this_cpu = sample->cpu, err = 0; 1352 u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"), 1353 next_pid = perf_evsel__intval(evsel, sample, "next_pid"); 1354 1355 if (sched->curr_pid[this_cpu] != (u32)-1) { 1356 /* 1357 * Are we trying to switch away a PID that is 1358 * not current? 1359 */ 1360 if (sched->curr_pid[this_cpu] != prev_pid) 1361 sched->nr_context_switch_bugs++; 1362 } 1363 1364 if (sched->tp_handler->switch_event) 1365 err = sched->tp_handler->switch_event(sched, evsel, sample, machine); 1366 1367 sched->curr_pid[this_cpu] = next_pid; 1368 return err; 1369 } 1370 1371 static int process_sched_runtime_event(struct perf_tool *tool, 1372 struct perf_evsel *evsel, 1373 struct perf_sample *sample, 1374 struct machine *machine) 1375 { 1376 struct perf_sched *sched = container_of(tool, struct perf_sched, tool); 1377 1378 if (sched->tp_handler->runtime_event) 1379 return sched->tp_handler->runtime_event(sched, evsel, sample, machine); 1380 1381 return 0; 1382 } 1383 1384 static int perf_sched__process_fork_event(struct perf_tool *tool, 1385 union perf_event *event, 1386 struct perf_sample *sample, 1387 struct machine *machine) 1388 { 1389 struct perf_sched *sched = container_of(tool, struct perf_sched, tool); 1390 1391 /* run the fork event through the perf machineruy */ 1392 perf_event__process_fork(tool, event, sample, machine); 1393 1394 /* and then run additional processing needed for this command */ 1395 if (sched->tp_handler->fork_event) 1396 return sched->tp_handler->fork_event(sched, event, machine); 1397 1398 return 0; 1399 } 1400 1401 static int process_sched_migrate_task_event(struct perf_tool *tool, 1402 struct perf_evsel *evsel, 1403 struct perf_sample *sample, 1404 struct machine *machine) 1405 { 1406 struct perf_sched *sched = container_of(tool, struct perf_sched, tool); 1407 1408 if (sched->tp_handler->migrate_task_event) 1409 return sched->tp_handler->migrate_task_event(sched, evsel, sample, machine); 1410 1411 return 0; 1412 } 1413 1414 typedef int (*tracepoint_handler)(struct perf_tool *tool, 1415 struct perf_evsel *evsel, 1416 struct perf_sample *sample, 1417 struct machine *machine); 1418 1419 static int perf_sched__process_tracepoint_sample(struct perf_tool *tool __maybe_unused, 1420 union perf_event *event __maybe_unused, 1421 struct perf_sample *sample, 1422 struct perf_evsel *evsel, 1423 struct machine *machine) 1424 { 1425 int err = 0; 1426 1427 evsel->hists.stats.total_period += sample->period; 1428 hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE); 1429 1430 if (evsel->handler != NULL) { 1431 tracepoint_handler f = evsel->handler; 1432 err = f(tool, evsel, sample, machine); 1433 } 1434 1435 return err; 1436 } 1437 1438 static int perf_sched__read_events(struct perf_sched *sched, 1439 struct perf_session **psession) 1440 { 1441 const struct perf_evsel_str_handler handlers[] = { 1442 { "sched:sched_switch", process_sched_switch_event, }, 1443 { "sched:sched_stat_runtime", process_sched_runtime_event, }, 1444 { "sched:sched_wakeup", process_sched_wakeup_event, }, 1445 { "sched:sched_wakeup_new", process_sched_wakeup_event, }, 1446 { "sched:sched_migrate_task", process_sched_migrate_task_event, }, 1447 }; 1448 struct perf_session *session; 1449 struct perf_data_file file = { 1450 .path = input_name, 1451 .mode = PERF_DATA_MODE_READ, 1452 }; 1453 1454 session = perf_session__new(&file, false, &sched->tool); 1455 if (session == NULL) { 1456 pr_debug("No Memory for session\n"); 1457 return -1; 1458 } 1459 1460 if (perf_session__set_tracepoints_handlers(session, handlers)) 1461 goto out_delete; 1462 1463 if (perf_session__has_traces(session, "record -R")) { 1464 int err = perf_session__process_events(session, &sched->tool); 1465 if (err) { 1466 pr_err("Failed to process events, error %d", err); 1467 goto out_delete; 1468 } 1469 1470 sched->nr_events = session->stats.nr_events[0]; 1471 sched->nr_lost_events = session->stats.total_lost; 1472 sched->nr_lost_chunks = session->stats.nr_events[PERF_RECORD_LOST]; 1473 } 1474 1475 if (psession) 1476 *psession = session; 1477 else 1478 perf_session__delete(session); 1479 1480 return 0; 1481 1482 out_delete: 1483 perf_session__delete(session); 1484 return -1; 1485 } 1486 1487 static void print_bad_events(struct perf_sched *sched) 1488 { 1489 if (sched->nr_unordered_timestamps && sched->nr_timestamps) { 1490 printf(" INFO: %.3f%% unordered timestamps (%ld out of %ld)\n", 1491 (double)sched->nr_unordered_timestamps/(double)sched->nr_timestamps*100.0, 1492 sched->nr_unordered_timestamps, sched->nr_timestamps); 1493 } 1494 if (sched->nr_lost_events && sched->nr_events) { 1495 printf(" INFO: %.3f%% lost events (%ld out of %ld, in %ld chunks)\n", 1496 (double)sched->nr_lost_events/(double)sched->nr_events * 100.0, 1497 sched->nr_lost_events, sched->nr_events, sched->nr_lost_chunks); 1498 } 1499 if (sched->nr_state_machine_bugs && sched->nr_timestamps) { 1500 printf(" INFO: %.3f%% state machine bugs (%ld out of %ld)", 1501 (double)sched->nr_state_machine_bugs/(double)sched->nr_timestamps*100.0, 1502 sched->nr_state_machine_bugs, sched->nr_timestamps); 1503 if (sched->nr_lost_events) 1504 printf(" (due to lost events?)"); 1505 printf("\n"); 1506 } 1507 if (sched->nr_context_switch_bugs && sched->nr_timestamps) { 1508 printf(" INFO: %.3f%% context switch bugs (%ld out of %ld)", 1509 (double)sched->nr_context_switch_bugs/(double)sched->nr_timestamps*100.0, 1510 sched->nr_context_switch_bugs, sched->nr_timestamps); 1511 if (sched->nr_lost_events) 1512 printf(" (due to lost events?)"); 1513 printf("\n"); 1514 } 1515 } 1516 1517 static int perf_sched__lat(struct perf_sched *sched) 1518 { 1519 struct rb_node *next; 1520 struct perf_session *session; 1521 1522 setup_pager(); 1523 1524 /* save session -- references to threads are held in work_list */ 1525 if (perf_sched__read_events(sched, &session)) 1526 return -1; 1527 1528 perf_sched__sort_lat(sched); 1529 1530 printf("\n -----------------------------------------------------------------------------------------------------------------\n"); 1531 printf(" Task | Runtime ms | Switches | Average delay ms | Maximum delay ms | Maximum delay at |\n"); 1532 printf(" -----------------------------------------------------------------------------------------------------------------\n"); 1533 1534 next = rb_first(&sched->sorted_atom_root); 1535 1536 while (next) { 1537 struct work_atoms *work_list; 1538 1539 work_list = rb_entry(next, struct work_atoms, node); 1540 output_lat_thread(sched, work_list); 1541 next = rb_next(next); 1542 } 1543 1544 printf(" -----------------------------------------------------------------------------------------------------------------\n"); 1545 printf(" TOTAL: |%11.3f ms |%9" PRIu64 " |\n", 1546 (double)sched->all_runtime / 1e6, sched->all_count); 1547 1548 printf(" ---------------------------------------------------\n"); 1549 1550 print_bad_events(sched); 1551 printf("\n"); 1552 1553 perf_session__delete(session); 1554 return 0; 1555 } 1556 1557 static int perf_sched__map(struct perf_sched *sched) 1558 { 1559 sched->max_cpu = sysconf(_SC_NPROCESSORS_CONF); 1560 1561 setup_pager(); 1562 if (perf_sched__read_events(sched, NULL)) 1563 return -1; 1564 print_bad_events(sched); 1565 return 0; 1566 } 1567 1568 static int perf_sched__replay(struct perf_sched *sched) 1569 { 1570 unsigned long i; 1571 1572 calibrate_run_measurement_overhead(sched); 1573 calibrate_sleep_measurement_overhead(sched); 1574 1575 test_calibrations(sched); 1576 1577 if (perf_sched__read_events(sched, NULL)) 1578 return -1; 1579 1580 printf("nr_run_events: %ld\n", sched->nr_run_events); 1581 printf("nr_sleep_events: %ld\n", sched->nr_sleep_events); 1582 printf("nr_wakeup_events: %ld\n", sched->nr_wakeup_events); 1583 1584 if (sched->targetless_wakeups) 1585 printf("target-less wakeups: %ld\n", sched->targetless_wakeups); 1586 if (sched->multitarget_wakeups) 1587 printf("multi-target wakeups: %ld\n", sched->multitarget_wakeups); 1588 if (sched->nr_run_events_optimized) 1589 printf("run atoms optimized: %ld\n", 1590 sched->nr_run_events_optimized); 1591 1592 print_task_traces(sched); 1593 add_cross_task_wakeups(sched); 1594 1595 create_tasks(sched); 1596 printf("------------------------------------------------------------\n"); 1597 for (i = 0; i < sched->replay_repeat; i++) 1598 run_one_test(sched); 1599 1600 return 0; 1601 } 1602 1603 static void setup_sorting(struct perf_sched *sched, const struct option *options, 1604 const char * const usage_msg[]) 1605 { 1606 char *tmp, *tok, *str = strdup(sched->sort_order); 1607 1608 for (tok = strtok_r(str, ", ", &tmp); 1609 tok; tok = strtok_r(NULL, ", ", &tmp)) { 1610 if (sort_dimension__add(tok, &sched->sort_list) < 0) { 1611 error("Unknown --sort key: `%s'", tok); 1612 usage_with_options(usage_msg, options); 1613 } 1614 } 1615 1616 free(str); 1617 1618 sort_dimension__add("pid", &sched->cmp_pid); 1619 } 1620 1621 static int __cmd_record(int argc, const char **argv) 1622 { 1623 unsigned int rec_argc, i, j; 1624 const char **rec_argv; 1625 const char * const record_args[] = { 1626 "record", 1627 "-a", 1628 "-R", 1629 "-m", "1024", 1630 "-c", "1", 1631 "-e", "sched:sched_switch", 1632 "-e", "sched:sched_stat_wait", 1633 "-e", "sched:sched_stat_sleep", 1634 "-e", "sched:sched_stat_iowait", 1635 "-e", "sched:sched_stat_runtime", 1636 "-e", "sched:sched_process_fork", 1637 "-e", "sched:sched_wakeup", 1638 "-e", "sched:sched_migrate_task", 1639 }; 1640 1641 rec_argc = ARRAY_SIZE(record_args) + argc - 1; 1642 rec_argv = calloc(rec_argc + 1, sizeof(char *)); 1643 1644 if (rec_argv == NULL) 1645 return -ENOMEM; 1646 1647 for (i = 0; i < ARRAY_SIZE(record_args); i++) 1648 rec_argv[i] = strdup(record_args[i]); 1649 1650 for (j = 1; j < (unsigned int)argc; j++, i++) 1651 rec_argv[i] = argv[j]; 1652 1653 BUG_ON(i != rec_argc); 1654 1655 return cmd_record(i, rec_argv, NULL); 1656 } 1657 1658 int cmd_sched(int argc, const char **argv, const char *prefix __maybe_unused) 1659 { 1660 const char default_sort_order[] = "avg, max, switch, runtime"; 1661 struct perf_sched sched = { 1662 .tool = { 1663 .sample = perf_sched__process_tracepoint_sample, 1664 .comm = perf_event__process_comm, 1665 .lost = perf_event__process_lost, 1666 .fork = perf_sched__process_fork_event, 1667 .ordered_samples = true, 1668 }, 1669 .cmp_pid = LIST_HEAD_INIT(sched.cmp_pid), 1670 .sort_list = LIST_HEAD_INIT(sched.sort_list), 1671 .start_work_mutex = PTHREAD_MUTEX_INITIALIZER, 1672 .work_done_wait_mutex = PTHREAD_MUTEX_INITIALIZER, 1673 .sort_order = default_sort_order, 1674 .replay_repeat = 10, 1675 .profile_cpu = -1, 1676 .next_shortname1 = 'A', 1677 .next_shortname2 = '0', 1678 }; 1679 const struct option latency_options[] = { 1680 OPT_STRING('s', "sort", &sched.sort_order, "key[,key2...]", 1681 "sort by key(s): runtime, switch, avg, max"), 1682 OPT_INCR('v', "verbose", &verbose, 1683 "be more verbose (show symbol address, etc)"), 1684 OPT_INTEGER('C', "CPU", &sched.profile_cpu, 1685 "CPU to profile on"), 1686 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, 1687 "dump raw trace in ASCII"), 1688 OPT_END() 1689 }; 1690 const struct option replay_options[] = { 1691 OPT_UINTEGER('r', "repeat", &sched.replay_repeat, 1692 "repeat the workload replay N times (-1: infinite)"), 1693 OPT_INCR('v', "verbose", &verbose, 1694 "be more verbose (show symbol address, etc)"), 1695 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, 1696 "dump raw trace in ASCII"), 1697 OPT_END() 1698 }; 1699 const struct option sched_options[] = { 1700 OPT_STRING('i', "input", &input_name, "file", 1701 "input file name"), 1702 OPT_INCR('v', "verbose", &verbose, 1703 "be more verbose (show symbol address, etc)"), 1704 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, 1705 "dump raw trace in ASCII"), 1706 OPT_END() 1707 }; 1708 const char * const latency_usage[] = { 1709 "perf sched latency [<options>]", 1710 NULL 1711 }; 1712 const char * const replay_usage[] = { 1713 "perf sched replay [<options>]", 1714 NULL 1715 }; 1716 const char * const sched_usage[] = { 1717 "perf sched [<options>] {record|latency|map|replay|script}", 1718 NULL 1719 }; 1720 struct trace_sched_handler lat_ops = { 1721 .wakeup_event = latency_wakeup_event, 1722 .switch_event = latency_switch_event, 1723 .runtime_event = latency_runtime_event, 1724 .migrate_task_event = latency_migrate_task_event, 1725 }; 1726 struct trace_sched_handler map_ops = { 1727 .switch_event = map_switch_event, 1728 }; 1729 struct trace_sched_handler replay_ops = { 1730 .wakeup_event = replay_wakeup_event, 1731 .switch_event = replay_switch_event, 1732 .fork_event = replay_fork_event, 1733 }; 1734 unsigned int i; 1735 1736 for (i = 0; i < ARRAY_SIZE(sched.curr_pid); i++) 1737 sched.curr_pid[i] = -1; 1738 1739 argc = parse_options(argc, argv, sched_options, sched_usage, 1740 PARSE_OPT_STOP_AT_NON_OPTION); 1741 if (!argc) 1742 usage_with_options(sched_usage, sched_options); 1743 1744 /* 1745 * Aliased to 'perf script' for now: 1746 */ 1747 if (!strcmp(argv[0], "script")) 1748 return cmd_script(argc, argv, prefix); 1749 1750 symbol__init(); 1751 if (!strncmp(argv[0], "rec", 3)) { 1752 return __cmd_record(argc, argv); 1753 } else if (!strncmp(argv[0], "lat", 3)) { 1754 sched.tp_handler = &lat_ops; 1755 if (argc > 1) { 1756 argc = parse_options(argc, argv, latency_options, latency_usage, 0); 1757 if (argc) 1758 usage_with_options(latency_usage, latency_options); 1759 } 1760 setup_sorting(&sched, latency_options, latency_usage); 1761 return perf_sched__lat(&sched); 1762 } else if (!strcmp(argv[0], "map")) { 1763 sched.tp_handler = &map_ops; 1764 setup_sorting(&sched, latency_options, latency_usage); 1765 return perf_sched__map(&sched); 1766 } else if (!strncmp(argv[0], "rep", 3)) { 1767 sched.tp_handler = &replay_ops; 1768 if (argc) { 1769 argc = parse_options(argc, argv, replay_options, replay_usage, 0); 1770 if (argc) 1771 usage_with_options(replay_usage, replay_options); 1772 } 1773 return perf_sched__replay(&sched); 1774 } else { 1775 usage_with_options(sched_usage, sched_options); 1776 } 1777 1778 return 0; 1779 } 1780