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 int (*fork_event)(struct perf_sched *sched, struct perf_evsel *evsel, 113 struct perf_sample *sample); 114 115 int (*migrate_task_event)(struct perf_sched *sched, 116 struct perf_evsel *evsel, 117 struct perf_sample *sample, 118 struct machine *machine); 119 }; 120 121 struct perf_sched { 122 struct perf_tool tool; 123 const char *sort_order; 124 unsigned long nr_tasks; 125 struct task_desc *pid_to_task[MAX_PID]; 126 struct task_desc **tasks; 127 const struct trace_sched_handler *tp_handler; 128 pthread_mutex_t start_work_mutex; 129 pthread_mutex_t work_done_wait_mutex; 130 int profile_cpu; 131 /* 132 * Track the current task - that way we can know whether there's any 133 * weird events, such as a task being switched away that is not current. 134 */ 135 int max_cpu; 136 u32 curr_pid[MAX_CPUS]; 137 struct thread *curr_thread[MAX_CPUS]; 138 char next_shortname1; 139 char next_shortname2; 140 unsigned int replay_repeat; 141 unsigned long nr_run_events; 142 unsigned long nr_sleep_events; 143 unsigned long nr_wakeup_events; 144 unsigned long nr_sleep_corrections; 145 unsigned long nr_run_events_optimized; 146 unsigned long targetless_wakeups; 147 unsigned long multitarget_wakeups; 148 unsigned long nr_runs; 149 unsigned long nr_timestamps; 150 unsigned long nr_unordered_timestamps; 151 unsigned long nr_state_machine_bugs; 152 unsigned long nr_context_switch_bugs; 153 unsigned long nr_events; 154 unsigned long nr_lost_chunks; 155 unsigned long nr_lost_events; 156 u64 run_measurement_overhead; 157 u64 sleep_measurement_overhead; 158 u64 start_time; 159 u64 cpu_usage; 160 u64 runavg_cpu_usage; 161 u64 parent_cpu_usage; 162 u64 runavg_parent_cpu_usage; 163 u64 sum_runtime; 164 u64 sum_fluct; 165 u64 run_avg; 166 u64 all_runtime; 167 u64 all_count; 168 u64 cpu_last_switched[MAX_CPUS]; 169 struct rb_root atom_root, sorted_atom_root; 170 struct list_head sort_list, cmp_pid; 171 }; 172 173 static u64 get_nsecs(void) 174 { 175 struct timespec ts; 176 177 clock_gettime(CLOCK_MONOTONIC, &ts); 178 179 return ts.tv_sec * 1000000000ULL + ts.tv_nsec; 180 } 181 182 static void burn_nsecs(struct perf_sched *sched, u64 nsecs) 183 { 184 u64 T0 = get_nsecs(), T1; 185 186 do { 187 T1 = get_nsecs(); 188 } while (T1 + sched->run_measurement_overhead < T0 + nsecs); 189 } 190 191 static void sleep_nsecs(u64 nsecs) 192 { 193 struct timespec ts; 194 195 ts.tv_nsec = nsecs % 999999999; 196 ts.tv_sec = nsecs / 999999999; 197 198 nanosleep(&ts, NULL); 199 } 200 201 static void calibrate_run_measurement_overhead(struct perf_sched *sched) 202 { 203 u64 T0, T1, delta, min_delta = 1000000000ULL; 204 int i; 205 206 for (i = 0; i < 10; i++) { 207 T0 = get_nsecs(); 208 burn_nsecs(sched, 0); 209 T1 = get_nsecs(); 210 delta = T1-T0; 211 min_delta = min(min_delta, delta); 212 } 213 sched->run_measurement_overhead = min_delta; 214 215 printf("run measurement overhead: %" PRIu64 " nsecs\n", min_delta); 216 } 217 218 static void calibrate_sleep_measurement_overhead(struct perf_sched *sched) 219 { 220 u64 T0, T1, delta, min_delta = 1000000000ULL; 221 int i; 222 223 for (i = 0; i < 10; i++) { 224 T0 = get_nsecs(); 225 sleep_nsecs(10000); 226 T1 = get_nsecs(); 227 delta = T1-T0; 228 min_delta = min(min_delta, delta); 229 } 230 min_delta -= 10000; 231 sched->sleep_measurement_overhead = min_delta; 232 233 printf("sleep measurement overhead: %" PRIu64 " nsecs\n", min_delta); 234 } 235 236 static struct sched_atom * 237 get_new_event(struct task_desc *task, u64 timestamp) 238 { 239 struct sched_atom *event = zalloc(sizeof(*event)); 240 unsigned long idx = task->nr_events; 241 size_t size; 242 243 event->timestamp = timestamp; 244 event->nr = idx; 245 246 task->nr_events++; 247 size = sizeof(struct sched_atom *) * task->nr_events; 248 task->atoms = realloc(task->atoms, size); 249 BUG_ON(!task->atoms); 250 251 task->atoms[idx] = event; 252 253 return event; 254 } 255 256 static struct sched_atom *last_event(struct task_desc *task) 257 { 258 if (!task->nr_events) 259 return NULL; 260 261 return task->atoms[task->nr_events - 1]; 262 } 263 264 static void add_sched_event_run(struct perf_sched *sched, struct task_desc *task, 265 u64 timestamp, u64 duration) 266 { 267 struct sched_atom *event, *curr_event = last_event(task); 268 269 /* 270 * optimize an existing RUN event by merging this one 271 * to it: 272 */ 273 if (curr_event && curr_event->type == SCHED_EVENT_RUN) { 274 sched->nr_run_events_optimized++; 275 curr_event->duration += duration; 276 return; 277 } 278 279 event = get_new_event(task, timestamp); 280 281 event->type = SCHED_EVENT_RUN; 282 event->duration = duration; 283 284 sched->nr_run_events++; 285 } 286 287 static void add_sched_event_wakeup(struct perf_sched *sched, struct task_desc *task, 288 u64 timestamp, struct task_desc *wakee) 289 { 290 struct sched_atom *event, *wakee_event; 291 292 event = get_new_event(task, timestamp); 293 event->type = SCHED_EVENT_WAKEUP; 294 event->wakee = wakee; 295 296 wakee_event = last_event(wakee); 297 if (!wakee_event || wakee_event->type != SCHED_EVENT_SLEEP) { 298 sched->targetless_wakeups++; 299 return; 300 } 301 if (wakee_event->wait_sem) { 302 sched->multitarget_wakeups++; 303 return; 304 } 305 306 wakee_event->wait_sem = zalloc(sizeof(*wakee_event->wait_sem)); 307 sem_init(wakee_event->wait_sem, 0, 0); 308 wakee_event->specific_wait = 1; 309 event->wait_sem = wakee_event->wait_sem; 310 311 sched->nr_wakeup_events++; 312 } 313 314 static void add_sched_event_sleep(struct perf_sched *sched, struct task_desc *task, 315 u64 timestamp, u64 task_state __maybe_unused) 316 { 317 struct sched_atom *event = get_new_event(task, timestamp); 318 319 event->type = SCHED_EVENT_SLEEP; 320 321 sched->nr_sleep_events++; 322 } 323 324 static struct task_desc *register_pid(struct perf_sched *sched, 325 unsigned long pid, const char *comm) 326 { 327 struct task_desc *task; 328 329 BUG_ON(pid >= MAX_PID); 330 331 task = sched->pid_to_task[pid]; 332 333 if (task) 334 return task; 335 336 task = zalloc(sizeof(*task)); 337 task->pid = pid; 338 task->nr = sched->nr_tasks; 339 strcpy(task->comm, comm); 340 /* 341 * every task starts in sleeping state - this gets ignored 342 * if there's no wakeup pointing to this sleep state: 343 */ 344 add_sched_event_sleep(sched, task, 0, 0); 345 346 sched->pid_to_task[pid] = task; 347 sched->nr_tasks++; 348 sched->tasks = realloc(sched->tasks, sched->nr_tasks * sizeof(struct task_task *)); 349 BUG_ON(!sched->tasks); 350 sched->tasks[task->nr] = task; 351 352 if (verbose) 353 printf("registered task #%ld, PID %ld (%s)\n", sched->nr_tasks, pid, comm); 354 355 return task; 356 } 357 358 359 static void print_task_traces(struct perf_sched *sched) 360 { 361 struct task_desc *task; 362 unsigned long i; 363 364 for (i = 0; i < sched->nr_tasks; i++) { 365 task = sched->tasks[i]; 366 printf("task %6ld (%20s:%10ld), nr_events: %ld\n", 367 task->nr, task->comm, task->pid, task->nr_events); 368 } 369 } 370 371 static void add_cross_task_wakeups(struct perf_sched *sched) 372 { 373 struct task_desc *task1, *task2; 374 unsigned long i, j; 375 376 for (i = 0; i < sched->nr_tasks; i++) { 377 task1 = sched->tasks[i]; 378 j = i + 1; 379 if (j == sched->nr_tasks) 380 j = 0; 381 task2 = sched->tasks[j]; 382 add_sched_event_wakeup(sched, task1, 0, task2); 383 } 384 } 385 386 static void perf_sched__process_event(struct perf_sched *sched, 387 struct sched_atom *atom) 388 { 389 int ret = 0; 390 391 switch (atom->type) { 392 case SCHED_EVENT_RUN: 393 burn_nsecs(sched, atom->duration); 394 break; 395 case SCHED_EVENT_SLEEP: 396 if (atom->wait_sem) 397 ret = sem_wait(atom->wait_sem); 398 BUG_ON(ret); 399 break; 400 case SCHED_EVENT_WAKEUP: 401 if (atom->wait_sem) 402 ret = sem_post(atom->wait_sem); 403 BUG_ON(ret); 404 break; 405 case SCHED_EVENT_MIGRATION: 406 break; 407 default: 408 BUG_ON(1); 409 } 410 } 411 412 static u64 get_cpu_usage_nsec_parent(void) 413 { 414 struct rusage ru; 415 u64 sum; 416 int err; 417 418 err = getrusage(RUSAGE_SELF, &ru); 419 BUG_ON(err); 420 421 sum = ru.ru_utime.tv_sec*1e9 + ru.ru_utime.tv_usec*1e3; 422 sum += ru.ru_stime.tv_sec*1e9 + ru.ru_stime.tv_usec*1e3; 423 424 return sum; 425 } 426 427 static int self_open_counters(void) 428 { 429 struct perf_event_attr attr; 430 int fd; 431 432 memset(&attr, 0, sizeof(attr)); 433 434 attr.type = PERF_TYPE_SOFTWARE; 435 attr.config = PERF_COUNT_SW_TASK_CLOCK; 436 437 fd = sys_perf_event_open(&attr, 0, -1, -1, 0); 438 439 if (fd < 0) 440 pr_err("Error: sys_perf_event_open() syscall returned " 441 "with %d (%s)\n", fd, strerror(errno)); 442 return fd; 443 } 444 445 static u64 get_cpu_usage_nsec_self(int fd) 446 { 447 u64 runtime; 448 int ret; 449 450 ret = read(fd, &runtime, sizeof(runtime)); 451 BUG_ON(ret != sizeof(runtime)); 452 453 return runtime; 454 } 455 456 struct sched_thread_parms { 457 struct task_desc *task; 458 struct perf_sched *sched; 459 }; 460 461 static void *thread_func(void *ctx) 462 { 463 struct sched_thread_parms *parms = ctx; 464 struct task_desc *this_task = parms->task; 465 struct perf_sched *sched = parms->sched; 466 u64 cpu_usage_0, cpu_usage_1; 467 unsigned long i, ret; 468 char comm2[22]; 469 int fd; 470 471 free(parms); 472 473 sprintf(comm2, ":%s", this_task->comm); 474 prctl(PR_SET_NAME, comm2); 475 fd = self_open_counters(); 476 if (fd < 0) 477 return NULL; 478 again: 479 ret = sem_post(&this_task->ready_for_work); 480 BUG_ON(ret); 481 ret = pthread_mutex_lock(&sched->start_work_mutex); 482 BUG_ON(ret); 483 ret = pthread_mutex_unlock(&sched->start_work_mutex); 484 BUG_ON(ret); 485 486 cpu_usage_0 = get_cpu_usage_nsec_self(fd); 487 488 for (i = 0; i < this_task->nr_events; i++) { 489 this_task->curr_event = i; 490 perf_sched__process_event(sched, this_task->atoms[i]); 491 } 492 493 cpu_usage_1 = get_cpu_usage_nsec_self(fd); 494 this_task->cpu_usage = cpu_usage_1 - cpu_usage_0; 495 ret = sem_post(&this_task->work_done_sem); 496 BUG_ON(ret); 497 498 ret = pthread_mutex_lock(&sched->work_done_wait_mutex); 499 BUG_ON(ret); 500 ret = pthread_mutex_unlock(&sched->work_done_wait_mutex); 501 BUG_ON(ret); 502 503 goto again; 504 } 505 506 static void create_tasks(struct perf_sched *sched) 507 { 508 struct task_desc *task; 509 pthread_attr_t attr; 510 unsigned long i; 511 int err; 512 513 err = pthread_attr_init(&attr); 514 BUG_ON(err); 515 err = pthread_attr_setstacksize(&attr, 516 (size_t) max(16 * 1024, PTHREAD_STACK_MIN)); 517 BUG_ON(err); 518 err = pthread_mutex_lock(&sched->start_work_mutex); 519 BUG_ON(err); 520 err = pthread_mutex_lock(&sched->work_done_wait_mutex); 521 BUG_ON(err); 522 for (i = 0; i < sched->nr_tasks; i++) { 523 struct sched_thread_parms *parms = malloc(sizeof(*parms)); 524 BUG_ON(parms == NULL); 525 parms->task = task = sched->tasks[i]; 526 parms->sched = sched; 527 sem_init(&task->sleep_sem, 0, 0); 528 sem_init(&task->ready_for_work, 0, 0); 529 sem_init(&task->work_done_sem, 0, 0); 530 task->curr_event = 0; 531 err = pthread_create(&task->thread, &attr, thread_func, parms); 532 BUG_ON(err); 533 } 534 } 535 536 static void wait_for_tasks(struct perf_sched *sched) 537 { 538 u64 cpu_usage_0, cpu_usage_1; 539 struct task_desc *task; 540 unsigned long i, ret; 541 542 sched->start_time = get_nsecs(); 543 sched->cpu_usage = 0; 544 pthread_mutex_unlock(&sched->work_done_wait_mutex); 545 546 for (i = 0; i < sched->nr_tasks; i++) { 547 task = sched->tasks[i]; 548 ret = sem_wait(&task->ready_for_work); 549 BUG_ON(ret); 550 sem_init(&task->ready_for_work, 0, 0); 551 } 552 ret = pthread_mutex_lock(&sched->work_done_wait_mutex); 553 BUG_ON(ret); 554 555 cpu_usage_0 = get_cpu_usage_nsec_parent(); 556 557 pthread_mutex_unlock(&sched->start_work_mutex); 558 559 for (i = 0; i < sched->nr_tasks; i++) { 560 task = sched->tasks[i]; 561 ret = sem_wait(&task->work_done_sem); 562 BUG_ON(ret); 563 sem_init(&task->work_done_sem, 0, 0); 564 sched->cpu_usage += task->cpu_usage; 565 task->cpu_usage = 0; 566 } 567 568 cpu_usage_1 = get_cpu_usage_nsec_parent(); 569 if (!sched->runavg_cpu_usage) 570 sched->runavg_cpu_usage = sched->cpu_usage; 571 sched->runavg_cpu_usage = (sched->runavg_cpu_usage * 9 + sched->cpu_usage) / 10; 572 573 sched->parent_cpu_usage = cpu_usage_1 - cpu_usage_0; 574 if (!sched->runavg_parent_cpu_usage) 575 sched->runavg_parent_cpu_usage = sched->parent_cpu_usage; 576 sched->runavg_parent_cpu_usage = (sched->runavg_parent_cpu_usage * 9 + 577 sched->parent_cpu_usage)/10; 578 579 ret = pthread_mutex_lock(&sched->start_work_mutex); 580 BUG_ON(ret); 581 582 for (i = 0; i < sched->nr_tasks; i++) { 583 task = sched->tasks[i]; 584 sem_init(&task->sleep_sem, 0, 0); 585 task->curr_event = 0; 586 } 587 } 588 589 static void run_one_test(struct perf_sched *sched) 590 { 591 u64 T0, T1, delta, avg_delta, fluct; 592 593 T0 = get_nsecs(); 594 wait_for_tasks(sched); 595 T1 = get_nsecs(); 596 597 delta = T1 - T0; 598 sched->sum_runtime += delta; 599 sched->nr_runs++; 600 601 avg_delta = sched->sum_runtime / sched->nr_runs; 602 if (delta < avg_delta) 603 fluct = avg_delta - delta; 604 else 605 fluct = delta - avg_delta; 606 sched->sum_fluct += fluct; 607 if (!sched->run_avg) 608 sched->run_avg = delta; 609 sched->run_avg = (sched->run_avg * 9 + delta) / 10; 610 611 printf("#%-3ld: %0.3f, ", sched->nr_runs, (double)delta / 1000000.0); 612 613 printf("ravg: %0.2f, ", (double)sched->run_avg / 1e6); 614 615 printf("cpu: %0.2f / %0.2f", 616 (double)sched->cpu_usage / 1e6, (double)sched->runavg_cpu_usage / 1e6); 617 618 #if 0 619 /* 620 * rusage statistics done by the parent, these are less 621 * accurate than the sched->sum_exec_runtime based statistics: 622 */ 623 printf(" [%0.2f / %0.2f]", 624 (double)sched->parent_cpu_usage/1e6, 625 (double)sched->runavg_parent_cpu_usage/1e6); 626 #endif 627 628 printf("\n"); 629 630 if (sched->nr_sleep_corrections) 631 printf(" (%ld sleep corrections)\n", sched->nr_sleep_corrections); 632 sched->nr_sleep_corrections = 0; 633 } 634 635 static void test_calibrations(struct perf_sched *sched) 636 { 637 u64 T0, T1; 638 639 T0 = get_nsecs(); 640 burn_nsecs(sched, 1e6); 641 T1 = get_nsecs(); 642 643 printf("the run test took %" PRIu64 " nsecs\n", T1 - T0); 644 645 T0 = get_nsecs(); 646 sleep_nsecs(1e6); 647 T1 = get_nsecs(); 648 649 printf("the sleep test took %" PRIu64 " nsecs\n", T1 - T0); 650 } 651 652 static int 653 replay_wakeup_event(struct perf_sched *sched, 654 struct perf_evsel *evsel, struct perf_sample *sample, 655 struct machine *machine __maybe_unused) 656 { 657 const char *comm = perf_evsel__strval(evsel, sample, "comm"); 658 const u32 pid = perf_evsel__intval(evsel, sample, "pid"); 659 struct task_desc *waker, *wakee; 660 661 if (verbose) { 662 printf("sched_wakeup event %p\n", evsel); 663 664 printf(" ... pid %d woke up %s/%d\n", sample->tid, comm, pid); 665 } 666 667 waker = register_pid(sched, sample->tid, "<unknown>"); 668 wakee = register_pid(sched, pid, comm); 669 670 add_sched_event_wakeup(sched, waker, sample->time, wakee); 671 return 0; 672 } 673 674 static int replay_switch_event(struct perf_sched *sched, 675 struct perf_evsel *evsel, 676 struct perf_sample *sample, 677 struct machine *machine __maybe_unused) 678 { 679 const char *prev_comm = perf_evsel__strval(evsel, sample, "prev_comm"), 680 *next_comm = perf_evsel__strval(evsel, sample, "next_comm"); 681 const u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"), 682 next_pid = perf_evsel__intval(evsel, sample, "next_pid"); 683 const u64 prev_state = perf_evsel__intval(evsel, sample, "prev_state"); 684 struct task_desc *prev, __maybe_unused *next; 685 u64 timestamp0, timestamp = sample->time; 686 int cpu = sample->cpu; 687 s64 delta; 688 689 if (verbose) 690 printf("sched_switch event %p\n", evsel); 691 692 if (cpu >= MAX_CPUS || cpu < 0) 693 return 0; 694 695 timestamp0 = sched->cpu_last_switched[cpu]; 696 if (timestamp0) 697 delta = timestamp - timestamp0; 698 else 699 delta = 0; 700 701 if (delta < 0) { 702 pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta); 703 return -1; 704 } 705 706 pr_debug(" ... switch from %s/%d to %s/%d [ran %" PRIu64 " nsecs]\n", 707 prev_comm, prev_pid, next_comm, next_pid, delta); 708 709 prev = register_pid(sched, prev_pid, prev_comm); 710 next = register_pid(sched, next_pid, next_comm); 711 712 sched->cpu_last_switched[cpu] = timestamp; 713 714 add_sched_event_run(sched, prev, timestamp, delta); 715 add_sched_event_sleep(sched, prev, timestamp, prev_state); 716 717 return 0; 718 } 719 720 static int replay_fork_event(struct perf_sched *sched, struct perf_evsel *evsel, 721 struct perf_sample *sample) 722 { 723 const char *parent_comm = perf_evsel__strval(evsel, sample, "parent_comm"), 724 *child_comm = perf_evsel__strval(evsel, sample, "child_comm"); 725 const u32 parent_pid = perf_evsel__intval(evsel, sample, "parent_pid"), 726 child_pid = perf_evsel__intval(evsel, sample, "child_pid"); 727 728 if (verbose) { 729 printf("sched_fork event %p\n", evsel); 730 printf("... parent: %s/%d\n", parent_comm, parent_pid); 731 printf("... child: %s/%d\n", child_comm, child_pid); 732 } 733 734 register_pid(sched, parent_pid, parent_comm); 735 register_pid(sched, child_pid, child_comm); 736 return 0; 737 } 738 739 struct sort_dimension { 740 const char *name; 741 sort_fn_t cmp; 742 struct list_head list; 743 }; 744 745 static int 746 thread_lat_cmp(struct list_head *list, struct work_atoms *l, struct work_atoms *r) 747 { 748 struct sort_dimension *sort; 749 int ret = 0; 750 751 BUG_ON(list_empty(list)); 752 753 list_for_each_entry(sort, list, list) { 754 ret = sort->cmp(l, r); 755 if (ret) 756 return ret; 757 } 758 759 return ret; 760 } 761 762 static struct work_atoms * 763 thread_atoms_search(struct rb_root *root, struct thread *thread, 764 struct list_head *sort_list) 765 { 766 struct rb_node *node = root->rb_node; 767 struct work_atoms key = { .thread = thread }; 768 769 while (node) { 770 struct work_atoms *atoms; 771 int cmp; 772 773 atoms = container_of(node, struct work_atoms, node); 774 775 cmp = thread_lat_cmp(sort_list, &key, atoms); 776 if (cmp > 0) 777 node = node->rb_left; 778 else if (cmp < 0) 779 node = node->rb_right; 780 else { 781 BUG_ON(thread != atoms->thread); 782 return atoms; 783 } 784 } 785 return NULL; 786 } 787 788 static void 789 __thread_latency_insert(struct rb_root *root, struct work_atoms *data, 790 struct list_head *sort_list) 791 { 792 struct rb_node **new = &(root->rb_node), *parent = NULL; 793 794 while (*new) { 795 struct work_atoms *this; 796 int cmp; 797 798 this = container_of(*new, struct work_atoms, node); 799 parent = *new; 800 801 cmp = thread_lat_cmp(sort_list, data, this); 802 803 if (cmp > 0) 804 new = &((*new)->rb_left); 805 else 806 new = &((*new)->rb_right); 807 } 808 809 rb_link_node(&data->node, parent, new); 810 rb_insert_color(&data->node, root); 811 } 812 813 static int thread_atoms_insert(struct perf_sched *sched, struct thread *thread) 814 { 815 struct work_atoms *atoms = zalloc(sizeof(*atoms)); 816 if (!atoms) { 817 pr_err("No memory at %s\n", __func__); 818 return -1; 819 } 820 821 atoms->thread = thread; 822 INIT_LIST_HEAD(&atoms->work_list); 823 __thread_latency_insert(&sched->atom_root, atoms, &sched->cmp_pid); 824 return 0; 825 } 826 827 static int latency_fork_event(struct perf_sched *sched __maybe_unused, 828 struct perf_evsel *evsel __maybe_unused, 829 struct perf_sample *sample __maybe_unused) 830 { 831 /* should insert the newcomer */ 832 return 0; 833 } 834 835 static char sched_out_state(u64 prev_state) 836 { 837 const char *str = TASK_STATE_TO_CHAR_STR; 838 839 return str[prev_state]; 840 } 841 842 static int 843 add_sched_out_event(struct work_atoms *atoms, 844 char run_state, 845 u64 timestamp) 846 { 847 struct work_atom *atom = zalloc(sizeof(*atom)); 848 if (!atom) { 849 pr_err("Non memory at %s", __func__); 850 return -1; 851 } 852 853 atom->sched_out_time = timestamp; 854 855 if (run_state == 'R') { 856 atom->state = THREAD_WAIT_CPU; 857 atom->wake_up_time = atom->sched_out_time; 858 } 859 860 list_add_tail(&atom->list, &atoms->work_list); 861 return 0; 862 } 863 864 static void 865 add_runtime_event(struct work_atoms *atoms, u64 delta, 866 u64 timestamp __maybe_unused) 867 { 868 struct work_atom *atom; 869 870 BUG_ON(list_empty(&atoms->work_list)); 871 872 atom = list_entry(atoms->work_list.prev, struct work_atom, list); 873 874 atom->runtime += delta; 875 atoms->total_runtime += delta; 876 } 877 878 static void 879 add_sched_in_event(struct work_atoms *atoms, u64 timestamp) 880 { 881 struct work_atom *atom; 882 u64 delta; 883 884 if (list_empty(&atoms->work_list)) 885 return; 886 887 atom = list_entry(atoms->work_list.prev, struct work_atom, list); 888 889 if (atom->state != THREAD_WAIT_CPU) 890 return; 891 892 if (timestamp < atom->wake_up_time) { 893 atom->state = THREAD_IGNORE; 894 return; 895 } 896 897 atom->state = THREAD_SCHED_IN; 898 atom->sched_in_time = timestamp; 899 900 delta = atom->sched_in_time - atom->wake_up_time; 901 atoms->total_lat += delta; 902 if (delta > atoms->max_lat) { 903 atoms->max_lat = delta; 904 atoms->max_lat_at = timestamp; 905 } 906 atoms->nb_atoms++; 907 } 908 909 static int latency_switch_event(struct perf_sched *sched, 910 struct perf_evsel *evsel, 911 struct perf_sample *sample, 912 struct machine *machine) 913 { 914 const u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"), 915 next_pid = perf_evsel__intval(evsel, sample, "next_pid"); 916 const u64 prev_state = perf_evsel__intval(evsel, sample, "prev_state"); 917 struct work_atoms *out_events, *in_events; 918 struct thread *sched_out, *sched_in; 919 u64 timestamp0, timestamp = sample->time; 920 int cpu = sample->cpu; 921 s64 delta; 922 923 BUG_ON(cpu >= MAX_CPUS || cpu < 0); 924 925 timestamp0 = sched->cpu_last_switched[cpu]; 926 sched->cpu_last_switched[cpu] = timestamp; 927 if (timestamp0) 928 delta = timestamp - timestamp0; 929 else 930 delta = 0; 931 932 if (delta < 0) { 933 pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta); 934 return -1; 935 } 936 937 sched_out = machine__findnew_thread(machine, prev_pid); 938 sched_in = machine__findnew_thread(machine, next_pid); 939 940 out_events = thread_atoms_search(&sched->atom_root, sched_out, &sched->cmp_pid); 941 if (!out_events) { 942 if (thread_atoms_insert(sched, sched_out)) 943 return -1; 944 out_events = thread_atoms_search(&sched->atom_root, sched_out, &sched->cmp_pid); 945 if (!out_events) { 946 pr_err("out-event: Internal tree error"); 947 return -1; 948 } 949 } 950 if (add_sched_out_event(out_events, sched_out_state(prev_state), timestamp)) 951 return -1; 952 953 in_events = thread_atoms_search(&sched->atom_root, sched_in, &sched->cmp_pid); 954 if (!in_events) { 955 if (thread_atoms_insert(sched, sched_in)) 956 return -1; 957 in_events = thread_atoms_search(&sched->atom_root, sched_in, &sched->cmp_pid); 958 if (!in_events) { 959 pr_err("in-event: Internal tree error"); 960 return -1; 961 } 962 /* 963 * Take came in we have not heard about yet, 964 * add in an initial atom in runnable state: 965 */ 966 if (add_sched_out_event(in_events, 'R', timestamp)) 967 return -1; 968 } 969 add_sched_in_event(in_events, timestamp); 970 971 return 0; 972 } 973 974 static int latency_runtime_event(struct perf_sched *sched, 975 struct perf_evsel *evsel, 976 struct perf_sample *sample, 977 struct machine *machine) 978 { 979 const u32 pid = perf_evsel__intval(evsel, sample, "pid"); 980 const u64 runtime = perf_evsel__intval(evsel, sample, "runtime"); 981 struct thread *thread = machine__findnew_thread(machine, pid); 982 struct work_atoms *atoms = thread_atoms_search(&sched->atom_root, thread, &sched->cmp_pid); 983 u64 timestamp = sample->time; 984 int cpu = sample->cpu; 985 986 BUG_ON(cpu >= MAX_CPUS || cpu < 0); 987 if (!atoms) { 988 if (thread_atoms_insert(sched, thread)) 989 return -1; 990 atoms = thread_atoms_search(&sched->atom_root, thread, &sched->cmp_pid); 991 if (!atoms) { 992 pr_err("in-event: Internal tree error"); 993 return -1; 994 } 995 if (add_sched_out_event(atoms, 'R', timestamp)) 996 return -1; 997 } 998 999 add_runtime_event(atoms, runtime, timestamp); 1000 return 0; 1001 } 1002 1003 static int latency_wakeup_event(struct perf_sched *sched, 1004 struct perf_evsel *evsel, 1005 struct perf_sample *sample, 1006 struct machine *machine) 1007 { 1008 const u32 pid = perf_evsel__intval(evsel, sample, "pid"), 1009 success = perf_evsel__intval(evsel, sample, "success"); 1010 struct work_atoms *atoms; 1011 struct work_atom *atom; 1012 struct thread *wakee; 1013 u64 timestamp = sample->time; 1014 1015 /* Note for later, it may be interesting to observe the failing cases */ 1016 if (!success) 1017 return 0; 1018 1019 wakee = machine__findnew_thread(machine, pid); 1020 atoms = thread_atoms_search(&sched->atom_root, wakee, &sched->cmp_pid); 1021 if (!atoms) { 1022 if (thread_atoms_insert(sched, wakee)) 1023 return -1; 1024 atoms = thread_atoms_search(&sched->atom_root, wakee, &sched->cmp_pid); 1025 if (!atoms) { 1026 pr_err("wakeup-event: Internal tree error"); 1027 return -1; 1028 } 1029 if (add_sched_out_event(atoms, 'S', timestamp)) 1030 return -1; 1031 } 1032 1033 BUG_ON(list_empty(&atoms->work_list)); 1034 1035 atom = list_entry(atoms->work_list.prev, struct work_atom, list); 1036 1037 /* 1038 * You WILL be missing events if you've recorded only 1039 * one CPU, or are only looking at only one, so don't 1040 * make useless noise. 1041 */ 1042 if (sched->profile_cpu == -1 && atom->state != THREAD_SLEEPING) 1043 sched->nr_state_machine_bugs++; 1044 1045 sched->nr_timestamps++; 1046 if (atom->sched_out_time > timestamp) { 1047 sched->nr_unordered_timestamps++; 1048 return 0; 1049 } 1050 1051 atom->state = THREAD_WAIT_CPU; 1052 atom->wake_up_time = timestamp; 1053 return 0; 1054 } 1055 1056 static int latency_migrate_task_event(struct perf_sched *sched, 1057 struct perf_evsel *evsel, 1058 struct perf_sample *sample, 1059 struct machine *machine) 1060 { 1061 const u32 pid = perf_evsel__intval(evsel, sample, "pid"); 1062 u64 timestamp = sample->time; 1063 struct work_atoms *atoms; 1064 struct work_atom *atom; 1065 struct thread *migrant; 1066 1067 /* 1068 * Only need to worry about migration when profiling one CPU. 1069 */ 1070 if (sched->profile_cpu == -1) 1071 return 0; 1072 1073 migrant = machine__findnew_thread(machine, pid); 1074 atoms = thread_atoms_search(&sched->atom_root, migrant, &sched->cmp_pid); 1075 if (!atoms) { 1076 if (thread_atoms_insert(sched, migrant)) 1077 return -1; 1078 register_pid(sched, migrant->pid, migrant->comm); 1079 atoms = thread_atoms_search(&sched->atom_root, migrant, &sched->cmp_pid); 1080 if (!atoms) { 1081 pr_err("migration-event: Internal tree error"); 1082 return -1; 1083 } 1084 if (add_sched_out_event(atoms, 'R', timestamp)) 1085 return -1; 1086 } 1087 1088 BUG_ON(list_empty(&atoms->work_list)); 1089 1090 atom = list_entry(atoms->work_list.prev, struct work_atom, list); 1091 atom->sched_in_time = atom->sched_out_time = atom->wake_up_time = timestamp; 1092 1093 sched->nr_timestamps++; 1094 1095 if (atom->sched_out_time > timestamp) 1096 sched->nr_unordered_timestamps++; 1097 1098 return 0; 1099 } 1100 1101 static void output_lat_thread(struct perf_sched *sched, struct work_atoms *work_list) 1102 { 1103 int i; 1104 int ret; 1105 u64 avg; 1106 1107 if (!work_list->nb_atoms) 1108 return; 1109 /* 1110 * Ignore idle threads: 1111 */ 1112 if (!strcmp(work_list->thread->comm, "swapper")) 1113 return; 1114 1115 sched->all_runtime += work_list->total_runtime; 1116 sched->all_count += work_list->nb_atoms; 1117 1118 ret = printf(" %s:%d ", work_list->thread->comm, work_list->thread->pid); 1119 1120 for (i = 0; i < 24 - ret; i++) 1121 printf(" "); 1122 1123 avg = work_list->total_lat / work_list->nb_atoms; 1124 1125 printf("|%11.3f ms |%9" PRIu64 " | avg:%9.3f ms | max:%9.3f ms | max at: %9.6f s\n", 1126 (double)work_list->total_runtime / 1e6, 1127 work_list->nb_atoms, (double)avg / 1e6, 1128 (double)work_list->max_lat / 1e6, 1129 (double)work_list->max_lat_at / 1e9); 1130 } 1131 1132 static int pid_cmp(struct work_atoms *l, struct work_atoms *r) 1133 { 1134 if (l->thread->pid < r->thread->pid) 1135 return -1; 1136 if (l->thread->pid > r->thread->pid) 1137 return 1; 1138 1139 return 0; 1140 } 1141 1142 static int avg_cmp(struct work_atoms *l, struct work_atoms *r) 1143 { 1144 u64 avgl, avgr; 1145 1146 if (!l->nb_atoms) 1147 return -1; 1148 1149 if (!r->nb_atoms) 1150 return 1; 1151 1152 avgl = l->total_lat / l->nb_atoms; 1153 avgr = r->total_lat / r->nb_atoms; 1154 1155 if (avgl < avgr) 1156 return -1; 1157 if (avgl > avgr) 1158 return 1; 1159 1160 return 0; 1161 } 1162 1163 static int max_cmp(struct work_atoms *l, struct work_atoms *r) 1164 { 1165 if (l->max_lat < r->max_lat) 1166 return -1; 1167 if (l->max_lat > r->max_lat) 1168 return 1; 1169 1170 return 0; 1171 } 1172 1173 static int switch_cmp(struct work_atoms *l, struct work_atoms *r) 1174 { 1175 if (l->nb_atoms < r->nb_atoms) 1176 return -1; 1177 if (l->nb_atoms > r->nb_atoms) 1178 return 1; 1179 1180 return 0; 1181 } 1182 1183 static int runtime_cmp(struct work_atoms *l, struct work_atoms *r) 1184 { 1185 if (l->total_runtime < r->total_runtime) 1186 return -1; 1187 if (l->total_runtime > r->total_runtime) 1188 return 1; 1189 1190 return 0; 1191 } 1192 1193 static int sort_dimension__add(const char *tok, struct list_head *list) 1194 { 1195 size_t i; 1196 static struct sort_dimension avg_sort_dimension = { 1197 .name = "avg", 1198 .cmp = avg_cmp, 1199 }; 1200 static struct sort_dimension max_sort_dimension = { 1201 .name = "max", 1202 .cmp = max_cmp, 1203 }; 1204 static struct sort_dimension pid_sort_dimension = { 1205 .name = "pid", 1206 .cmp = pid_cmp, 1207 }; 1208 static struct sort_dimension runtime_sort_dimension = { 1209 .name = "runtime", 1210 .cmp = runtime_cmp, 1211 }; 1212 static struct sort_dimension switch_sort_dimension = { 1213 .name = "switch", 1214 .cmp = switch_cmp, 1215 }; 1216 struct sort_dimension *available_sorts[] = { 1217 &pid_sort_dimension, 1218 &avg_sort_dimension, 1219 &max_sort_dimension, 1220 &switch_sort_dimension, 1221 &runtime_sort_dimension, 1222 }; 1223 1224 for (i = 0; i < ARRAY_SIZE(available_sorts); i++) { 1225 if (!strcmp(available_sorts[i]->name, tok)) { 1226 list_add_tail(&available_sorts[i]->list, list); 1227 1228 return 0; 1229 } 1230 } 1231 1232 return -1; 1233 } 1234 1235 static void perf_sched__sort_lat(struct perf_sched *sched) 1236 { 1237 struct rb_node *node; 1238 1239 for (;;) { 1240 struct work_atoms *data; 1241 node = rb_first(&sched->atom_root); 1242 if (!node) 1243 break; 1244 1245 rb_erase(node, &sched->atom_root); 1246 data = rb_entry(node, struct work_atoms, node); 1247 __thread_latency_insert(&sched->sorted_atom_root, data, &sched->sort_list); 1248 } 1249 } 1250 1251 static int process_sched_wakeup_event(struct perf_tool *tool, 1252 struct perf_evsel *evsel, 1253 struct perf_sample *sample, 1254 struct machine *machine) 1255 { 1256 struct perf_sched *sched = container_of(tool, struct perf_sched, tool); 1257 1258 if (sched->tp_handler->wakeup_event) 1259 return sched->tp_handler->wakeup_event(sched, evsel, sample, machine); 1260 1261 return 0; 1262 } 1263 1264 static int map_switch_event(struct perf_sched *sched, struct perf_evsel *evsel, 1265 struct perf_sample *sample, struct machine *machine) 1266 { 1267 const u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"), 1268 next_pid = perf_evsel__intval(evsel, sample, "next_pid"); 1269 struct thread *sched_out __maybe_unused, *sched_in; 1270 int new_shortname; 1271 u64 timestamp0, timestamp = sample->time; 1272 s64 delta; 1273 int cpu, this_cpu = sample->cpu; 1274 1275 BUG_ON(this_cpu >= MAX_CPUS || this_cpu < 0); 1276 1277 if (this_cpu > sched->max_cpu) 1278 sched->max_cpu = this_cpu; 1279 1280 timestamp0 = sched->cpu_last_switched[this_cpu]; 1281 sched->cpu_last_switched[this_cpu] = timestamp; 1282 if (timestamp0) 1283 delta = timestamp - timestamp0; 1284 else 1285 delta = 0; 1286 1287 if (delta < 0) { 1288 pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta); 1289 return -1; 1290 } 1291 1292 sched_out = machine__findnew_thread(machine, prev_pid); 1293 sched_in = machine__findnew_thread(machine, next_pid); 1294 1295 sched->curr_thread[this_cpu] = sched_in; 1296 1297 printf(" "); 1298 1299 new_shortname = 0; 1300 if (!sched_in->shortname[0]) { 1301 sched_in->shortname[0] = sched->next_shortname1; 1302 sched_in->shortname[1] = sched->next_shortname2; 1303 1304 if (sched->next_shortname1 < 'Z') { 1305 sched->next_shortname1++; 1306 } else { 1307 sched->next_shortname1='A'; 1308 if (sched->next_shortname2 < '9') { 1309 sched->next_shortname2++; 1310 } else { 1311 sched->next_shortname2='0'; 1312 } 1313 } 1314 new_shortname = 1; 1315 } 1316 1317 for (cpu = 0; cpu <= sched->max_cpu; cpu++) { 1318 if (cpu != this_cpu) 1319 printf(" "); 1320 else 1321 printf("*"); 1322 1323 if (sched->curr_thread[cpu]) { 1324 if (sched->curr_thread[cpu]->pid) 1325 printf("%2s ", sched->curr_thread[cpu]->shortname); 1326 else 1327 printf(". "); 1328 } else 1329 printf(" "); 1330 } 1331 1332 printf(" %12.6f secs ", (double)timestamp/1e9); 1333 if (new_shortname) { 1334 printf("%s => %s:%d\n", 1335 sched_in->shortname, sched_in->comm, sched_in->pid); 1336 } else { 1337 printf("\n"); 1338 } 1339 1340 return 0; 1341 } 1342 1343 static int process_sched_switch_event(struct perf_tool *tool, 1344 struct perf_evsel *evsel, 1345 struct perf_sample *sample, 1346 struct machine *machine) 1347 { 1348 struct perf_sched *sched = container_of(tool, struct perf_sched, tool); 1349 int this_cpu = sample->cpu, err = 0; 1350 u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"), 1351 next_pid = perf_evsel__intval(evsel, sample, "next_pid"); 1352 1353 if (sched->curr_pid[this_cpu] != (u32)-1) { 1354 /* 1355 * Are we trying to switch away a PID that is 1356 * not current? 1357 */ 1358 if (sched->curr_pid[this_cpu] != prev_pid) 1359 sched->nr_context_switch_bugs++; 1360 } 1361 1362 if (sched->tp_handler->switch_event) 1363 err = sched->tp_handler->switch_event(sched, evsel, sample, machine); 1364 1365 sched->curr_pid[this_cpu] = next_pid; 1366 return err; 1367 } 1368 1369 static int process_sched_runtime_event(struct perf_tool *tool, 1370 struct perf_evsel *evsel, 1371 struct perf_sample *sample, 1372 struct machine *machine) 1373 { 1374 struct perf_sched *sched = container_of(tool, struct perf_sched, tool); 1375 1376 if (sched->tp_handler->runtime_event) 1377 return sched->tp_handler->runtime_event(sched, evsel, sample, machine); 1378 1379 return 0; 1380 } 1381 1382 static int process_sched_fork_event(struct perf_tool *tool, 1383 struct perf_evsel *evsel, 1384 struct perf_sample *sample, 1385 struct machine *machine __maybe_unused) 1386 { 1387 struct perf_sched *sched = container_of(tool, struct perf_sched, tool); 1388 1389 if (sched->tp_handler->fork_event) 1390 return sched->tp_handler->fork_event(sched, evsel, sample); 1391 1392 return 0; 1393 } 1394 1395 static int process_sched_exit_event(struct perf_tool *tool __maybe_unused, 1396 struct perf_evsel *evsel, 1397 struct perf_sample *sample __maybe_unused, 1398 struct machine *machine __maybe_unused) 1399 { 1400 pr_debug("sched_exit event %p\n", evsel); 1401 return 0; 1402 } 1403 1404 static int process_sched_migrate_task_event(struct perf_tool *tool, 1405 struct perf_evsel *evsel, 1406 struct perf_sample *sample, 1407 struct machine *machine) 1408 { 1409 struct perf_sched *sched = container_of(tool, struct perf_sched, tool); 1410 1411 if (sched->tp_handler->migrate_task_event) 1412 return sched->tp_handler->migrate_task_event(sched, evsel, sample, machine); 1413 1414 return 0; 1415 } 1416 1417 typedef int (*tracepoint_handler)(struct perf_tool *tool, 1418 struct perf_evsel *evsel, 1419 struct perf_sample *sample, 1420 struct machine *machine); 1421 1422 static int perf_sched__process_tracepoint_sample(struct perf_tool *tool __maybe_unused, 1423 union perf_event *event __maybe_unused, 1424 struct perf_sample *sample, 1425 struct perf_evsel *evsel, 1426 struct machine *machine) 1427 { 1428 struct thread *thread = machine__findnew_thread(machine, sample->tid); 1429 int err = 0; 1430 1431 if (thread == NULL) { 1432 pr_debug("problem processing %s event, skipping it.\n", 1433 perf_evsel__name(evsel)); 1434 return -1; 1435 } 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, bool destroy, 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_process_fork", process_sched_fork_event, }, 1457 { "sched:sched_process_exit", process_sched_exit_event, }, 1458 { "sched:sched_migrate_task", process_sched_migrate_task_event, }, 1459 }; 1460 struct perf_session *session; 1461 1462 session = perf_session__new(input_name, O_RDONLY, 0, false, &sched->tool); 1463 if (session == NULL) { 1464 pr_debug("No Memory for session\n"); 1465 return -1; 1466 } 1467 1468 if (perf_session__set_tracepoints_handlers(session, handlers)) 1469 goto out_delete; 1470 1471 if (perf_session__has_traces(session, "record -R")) { 1472 int err = perf_session__process_events(session, &sched->tool); 1473 if (err) { 1474 pr_err("Failed to process events, error %d", err); 1475 goto out_delete; 1476 } 1477 1478 sched->nr_events = session->stats.nr_events[0]; 1479 sched->nr_lost_events = session->stats.total_lost; 1480 sched->nr_lost_chunks = session->stats.nr_events[PERF_RECORD_LOST]; 1481 } 1482 1483 if (destroy) 1484 perf_session__delete(session); 1485 1486 if (psession) 1487 *psession = session; 1488 1489 return 0; 1490 1491 out_delete: 1492 perf_session__delete(session); 1493 return -1; 1494 } 1495 1496 static void print_bad_events(struct perf_sched *sched) 1497 { 1498 if (sched->nr_unordered_timestamps && sched->nr_timestamps) { 1499 printf(" INFO: %.3f%% unordered timestamps (%ld out of %ld)\n", 1500 (double)sched->nr_unordered_timestamps/(double)sched->nr_timestamps*100.0, 1501 sched->nr_unordered_timestamps, sched->nr_timestamps); 1502 } 1503 if (sched->nr_lost_events && sched->nr_events) { 1504 printf(" INFO: %.3f%% lost events (%ld out of %ld, in %ld chunks)\n", 1505 (double)sched->nr_lost_events/(double)sched->nr_events * 100.0, 1506 sched->nr_lost_events, sched->nr_events, sched->nr_lost_chunks); 1507 } 1508 if (sched->nr_state_machine_bugs && sched->nr_timestamps) { 1509 printf(" INFO: %.3f%% state machine bugs (%ld out of %ld)", 1510 (double)sched->nr_state_machine_bugs/(double)sched->nr_timestamps*100.0, 1511 sched->nr_state_machine_bugs, sched->nr_timestamps); 1512 if (sched->nr_lost_events) 1513 printf(" (due to lost events?)"); 1514 printf("\n"); 1515 } 1516 if (sched->nr_context_switch_bugs && sched->nr_timestamps) { 1517 printf(" INFO: %.3f%% context switch bugs (%ld out of %ld)", 1518 (double)sched->nr_context_switch_bugs/(double)sched->nr_timestamps*100.0, 1519 sched->nr_context_switch_bugs, sched->nr_timestamps); 1520 if (sched->nr_lost_events) 1521 printf(" (due to lost events?)"); 1522 printf("\n"); 1523 } 1524 } 1525 1526 static int perf_sched__lat(struct perf_sched *sched) 1527 { 1528 struct rb_node *next; 1529 struct perf_session *session; 1530 1531 setup_pager(); 1532 if (perf_sched__read_events(sched, false, &session)) 1533 return -1; 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, true, 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, true, 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 "-f", 1636 "-m", "1024", 1637 "-c", "1", 1638 "-e", "sched:sched_switch", 1639 "-e", "sched:sched_stat_wait", 1640 "-e", "sched:sched_stat_sleep", 1641 "-e", "sched:sched_stat_iowait", 1642 "-e", "sched:sched_stat_runtime", 1643 "-e", "sched:sched_process_exit", 1644 "-e", "sched:sched_process_fork", 1645 "-e", "sched:sched_wakeup", 1646 "-e", "sched:sched_migrate_task", 1647 }; 1648 1649 rec_argc = ARRAY_SIZE(record_args) + argc - 1; 1650 rec_argv = calloc(rec_argc + 1, sizeof(char *)); 1651 1652 if (rec_argv == NULL) 1653 return -ENOMEM; 1654 1655 for (i = 0; i < ARRAY_SIZE(record_args); i++) 1656 rec_argv[i] = strdup(record_args[i]); 1657 1658 for (j = 1; j < (unsigned int)argc; j++, i++) 1659 rec_argv[i] = argv[j]; 1660 1661 BUG_ON(i != rec_argc); 1662 1663 return cmd_record(i, rec_argv, NULL); 1664 } 1665 1666 int cmd_sched(int argc, const char **argv, const char *prefix __maybe_unused) 1667 { 1668 const char default_sort_order[] = "avg, max, switch, runtime"; 1669 struct perf_sched sched = { 1670 .tool = { 1671 .sample = perf_sched__process_tracepoint_sample, 1672 .comm = perf_event__process_comm, 1673 .lost = perf_event__process_lost, 1674 .exit = perf_event__process_exit, 1675 .fork = perf_event__process_fork, 1676 .ordered_samples = true, 1677 }, 1678 .cmp_pid = LIST_HEAD_INIT(sched.cmp_pid), 1679 .sort_list = LIST_HEAD_INIT(sched.sort_list), 1680 .start_work_mutex = PTHREAD_MUTEX_INITIALIZER, 1681 .work_done_wait_mutex = PTHREAD_MUTEX_INITIALIZER, 1682 .curr_pid = { [0 ... MAX_CPUS - 1] = -1 }, 1683 .sort_order = default_sort_order, 1684 .replay_repeat = 10, 1685 .profile_cpu = -1, 1686 .next_shortname1 = 'A', 1687 .next_shortname2 = '0', 1688 }; 1689 const struct option latency_options[] = { 1690 OPT_STRING('s', "sort", &sched.sort_order, "key[,key2...]", 1691 "sort by key(s): runtime, switch, avg, max"), 1692 OPT_INCR('v', "verbose", &verbose, 1693 "be more verbose (show symbol address, etc)"), 1694 OPT_INTEGER('C', "CPU", &sched.profile_cpu, 1695 "CPU to profile on"), 1696 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, 1697 "dump raw trace in ASCII"), 1698 OPT_END() 1699 }; 1700 const struct option replay_options[] = { 1701 OPT_UINTEGER('r', "repeat", &sched.replay_repeat, 1702 "repeat the workload replay N times (-1: infinite)"), 1703 OPT_INCR('v', "verbose", &verbose, 1704 "be more verbose (show symbol address, etc)"), 1705 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, 1706 "dump raw trace in ASCII"), 1707 OPT_END() 1708 }; 1709 const struct option sched_options[] = { 1710 OPT_STRING('i', "input", &input_name, "file", 1711 "input file name"), 1712 OPT_INCR('v', "verbose", &verbose, 1713 "be more verbose (show symbol address, etc)"), 1714 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, 1715 "dump raw trace in ASCII"), 1716 OPT_END() 1717 }; 1718 const char * const latency_usage[] = { 1719 "perf sched latency [<options>]", 1720 NULL 1721 }; 1722 const char * const replay_usage[] = { 1723 "perf sched replay [<options>]", 1724 NULL 1725 }; 1726 const char * const sched_usage[] = { 1727 "perf sched [<options>] {record|latency|map|replay|script}", 1728 NULL 1729 }; 1730 struct trace_sched_handler lat_ops = { 1731 .wakeup_event = latency_wakeup_event, 1732 .switch_event = latency_switch_event, 1733 .runtime_event = latency_runtime_event, 1734 .fork_event = latency_fork_event, 1735 .migrate_task_event = latency_migrate_task_event, 1736 }; 1737 struct trace_sched_handler map_ops = { 1738 .switch_event = map_switch_event, 1739 }; 1740 struct trace_sched_handler replay_ops = { 1741 .wakeup_event = replay_wakeup_event, 1742 .switch_event = replay_switch_event, 1743 .fork_event = replay_fork_event, 1744 }; 1745 1746 argc = parse_options(argc, argv, sched_options, sched_usage, 1747 PARSE_OPT_STOP_AT_NON_OPTION); 1748 if (!argc) 1749 usage_with_options(sched_usage, sched_options); 1750 1751 /* 1752 * Aliased to 'perf script' for now: 1753 */ 1754 if (!strcmp(argv[0], "script")) 1755 return cmd_script(argc, argv, prefix); 1756 1757 symbol__init(); 1758 if (!strncmp(argv[0], "rec", 3)) { 1759 return __cmd_record(argc, argv); 1760 } else if (!strncmp(argv[0], "lat", 3)) { 1761 sched.tp_handler = &lat_ops; 1762 if (argc > 1) { 1763 argc = parse_options(argc, argv, latency_options, latency_usage, 0); 1764 if (argc) 1765 usage_with_options(latency_usage, latency_options); 1766 } 1767 setup_sorting(&sched, latency_options, latency_usage); 1768 return perf_sched__lat(&sched); 1769 } else if (!strcmp(argv[0], "map")) { 1770 sched.tp_handler = &map_ops; 1771 setup_sorting(&sched, latency_options, latency_usage); 1772 return perf_sched__map(&sched); 1773 } else if (!strncmp(argv[0], "rep", 3)) { 1774 sched.tp_handler = &replay_ops; 1775 if (argc) { 1776 argc = parse_options(argc, argv, replay_options, replay_usage, 0); 1777 if (argc) 1778 usage_with_options(replay_usage, replay_options); 1779 } 1780 return perf_sched__replay(&sched); 1781 } else { 1782 usage_with_options(sched_usage, sched_options); 1783 } 1784 1785 return 0; 1786 } 1787