1 // SPDX-License-Identifier: GPL-2.0 2 #include "builtin.h" 3 #include "perf.h" 4 5 #include "util/util.h" 6 #include "util/evlist.h" 7 #include "util/cache.h" 8 #include "util/evsel.h" 9 #include "util/symbol.h" 10 #include "util/thread.h" 11 #include "util/header.h" 12 #include "util/session.h" 13 #include "util/tool.h" 14 #include "util/cloexec.h" 15 #include "util/thread_map.h" 16 #include "util/color.h" 17 #include "util/stat.h" 18 #include "util/callchain.h" 19 #include "util/time-utils.h" 20 21 #include <subcmd/parse-options.h> 22 #include "util/trace-event.h" 23 24 #include "util/debug.h" 25 26 #include <linux/kernel.h> 27 #include <linux/log2.h> 28 #include <sys/prctl.h> 29 #include <sys/resource.h> 30 #include <inttypes.h> 31 32 #include <errno.h> 33 #include <semaphore.h> 34 #include <pthread.h> 35 #include <math.h> 36 #include <api/fs/fs.h> 37 #include <linux/time64.h> 38 39 #include "sane_ctype.h" 40 41 #define PR_SET_NAME 15 /* Set process name */ 42 #define MAX_CPUS 4096 43 #define COMM_LEN 20 44 #define SYM_LEN 129 45 #define MAX_PID 1024000 46 47 struct sched_atom; 48 49 struct task_desc { 50 unsigned long nr; 51 unsigned long pid; 52 char comm[COMM_LEN]; 53 54 unsigned long nr_events; 55 unsigned long curr_event; 56 struct sched_atom **atoms; 57 58 pthread_t thread; 59 sem_t sleep_sem; 60 61 sem_t ready_for_work; 62 sem_t work_done_sem; 63 64 u64 cpu_usage; 65 }; 66 67 enum sched_event_type { 68 SCHED_EVENT_RUN, 69 SCHED_EVENT_SLEEP, 70 SCHED_EVENT_WAKEUP, 71 SCHED_EVENT_MIGRATION, 72 }; 73 74 struct sched_atom { 75 enum sched_event_type type; 76 int specific_wait; 77 u64 timestamp; 78 u64 duration; 79 unsigned long nr; 80 sem_t *wait_sem; 81 struct task_desc *wakee; 82 }; 83 84 #define TASK_STATE_TO_CHAR_STR "RSDTtZXxKWP" 85 86 /* task state bitmask, copied from include/linux/sched.h */ 87 #define TASK_RUNNING 0 88 #define TASK_INTERRUPTIBLE 1 89 #define TASK_UNINTERRUPTIBLE 2 90 #define __TASK_STOPPED 4 91 #define __TASK_TRACED 8 92 /* in tsk->exit_state */ 93 #define EXIT_DEAD 16 94 #define EXIT_ZOMBIE 32 95 #define EXIT_TRACE (EXIT_ZOMBIE | EXIT_DEAD) 96 /* in tsk->state again */ 97 #define TASK_DEAD 64 98 #define TASK_WAKEKILL 128 99 #define TASK_WAKING 256 100 #define TASK_PARKED 512 101 102 enum thread_state { 103 THREAD_SLEEPING = 0, 104 THREAD_WAIT_CPU, 105 THREAD_SCHED_IN, 106 THREAD_IGNORE 107 }; 108 109 struct work_atom { 110 struct list_head list; 111 enum thread_state state; 112 u64 sched_out_time; 113 u64 wake_up_time; 114 u64 sched_in_time; 115 u64 runtime; 116 }; 117 118 struct work_atoms { 119 struct list_head work_list; 120 struct thread *thread; 121 struct rb_node node; 122 u64 max_lat; 123 u64 max_lat_at; 124 u64 total_lat; 125 u64 nb_atoms; 126 u64 total_runtime; 127 int num_merged; 128 }; 129 130 typedef int (*sort_fn_t)(struct work_atoms *, struct work_atoms *); 131 132 struct perf_sched; 133 134 struct trace_sched_handler { 135 int (*switch_event)(struct perf_sched *sched, struct perf_evsel *evsel, 136 struct perf_sample *sample, struct machine *machine); 137 138 int (*runtime_event)(struct perf_sched *sched, struct perf_evsel *evsel, 139 struct perf_sample *sample, struct machine *machine); 140 141 int (*wakeup_event)(struct perf_sched *sched, struct perf_evsel *evsel, 142 struct perf_sample *sample, struct machine *machine); 143 144 /* PERF_RECORD_FORK event, not sched_process_fork tracepoint */ 145 int (*fork_event)(struct perf_sched *sched, union perf_event *event, 146 struct machine *machine); 147 148 int (*migrate_task_event)(struct perf_sched *sched, 149 struct perf_evsel *evsel, 150 struct perf_sample *sample, 151 struct machine *machine); 152 }; 153 154 #define COLOR_PIDS PERF_COLOR_BLUE 155 #define COLOR_CPUS PERF_COLOR_BG_RED 156 157 struct perf_sched_map { 158 DECLARE_BITMAP(comp_cpus_mask, MAX_CPUS); 159 int *comp_cpus; 160 bool comp; 161 struct thread_map *color_pids; 162 const char *color_pids_str; 163 struct cpu_map *color_cpus; 164 const char *color_cpus_str; 165 struct cpu_map *cpus; 166 const char *cpus_str; 167 }; 168 169 struct perf_sched { 170 struct perf_tool tool; 171 const char *sort_order; 172 unsigned long nr_tasks; 173 struct task_desc **pid_to_task; 174 struct task_desc **tasks; 175 const struct trace_sched_handler *tp_handler; 176 pthread_mutex_t start_work_mutex; 177 pthread_mutex_t work_done_wait_mutex; 178 int profile_cpu; 179 /* 180 * Track the current task - that way we can know whether there's any 181 * weird events, such as a task being switched away that is not current. 182 */ 183 int max_cpu; 184 u32 curr_pid[MAX_CPUS]; 185 struct thread *curr_thread[MAX_CPUS]; 186 char next_shortname1; 187 char next_shortname2; 188 unsigned int replay_repeat; 189 unsigned long nr_run_events; 190 unsigned long nr_sleep_events; 191 unsigned long nr_wakeup_events; 192 unsigned long nr_sleep_corrections; 193 unsigned long nr_run_events_optimized; 194 unsigned long targetless_wakeups; 195 unsigned long multitarget_wakeups; 196 unsigned long nr_runs; 197 unsigned long nr_timestamps; 198 unsigned long nr_unordered_timestamps; 199 unsigned long nr_context_switch_bugs; 200 unsigned long nr_events; 201 unsigned long nr_lost_chunks; 202 unsigned long nr_lost_events; 203 u64 run_measurement_overhead; 204 u64 sleep_measurement_overhead; 205 u64 start_time; 206 u64 cpu_usage; 207 u64 runavg_cpu_usage; 208 u64 parent_cpu_usage; 209 u64 runavg_parent_cpu_usage; 210 u64 sum_runtime; 211 u64 sum_fluct; 212 u64 run_avg; 213 u64 all_runtime; 214 u64 all_count; 215 u64 cpu_last_switched[MAX_CPUS]; 216 struct rb_root atom_root, sorted_atom_root, merged_atom_root; 217 struct list_head sort_list, cmp_pid; 218 bool force; 219 bool skip_merge; 220 struct perf_sched_map map; 221 222 /* options for timehist command */ 223 bool summary; 224 bool summary_only; 225 bool idle_hist; 226 bool show_callchain; 227 unsigned int max_stack; 228 bool show_cpu_visual; 229 bool show_wakeups; 230 bool show_next; 231 bool show_migrations; 232 bool show_state; 233 u64 skipped_samples; 234 const char *time_str; 235 struct perf_time_interval ptime; 236 struct perf_time_interval hist_time; 237 }; 238 239 /* per thread run time data */ 240 struct thread_runtime { 241 u64 last_time; /* time of previous sched in/out event */ 242 u64 dt_run; /* run time */ 243 u64 dt_sleep; /* time between CPU access by sleep (off cpu) */ 244 u64 dt_iowait; /* time between CPU access by iowait (off cpu) */ 245 u64 dt_preempt; /* time between CPU access by preempt (off cpu) */ 246 u64 dt_delay; /* time between wakeup and sched-in */ 247 u64 ready_to_run; /* time of wakeup */ 248 249 struct stats run_stats; 250 u64 total_run_time; 251 u64 total_sleep_time; 252 u64 total_iowait_time; 253 u64 total_preempt_time; 254 u64 total_delay_time; 255 256 int last_state; 257 258 char shortname[3]; 259 bool comm_changed; 260 261 u64 migrations; 262 }; 263 264 /* per event run time data */ 265 struct evsel_runtime { 266 u64 *last_time; /* time this event was last seen per cpu */ 267 u32 ncpu; /* highest cpu slot allocated */ 268 }; 269 270 /* per cpu idle time data */ 271 struct idle_thread_runtime { 272 struct thread_runtime tr; 273 struct thread *last_thread; 274 struct rb_root sorted_root; 275 struct callchain_root callchain; 276 struct callchain_cursor cursor; 277 }; 278 279 /* track idle times per cpu */ 280 static struct thread **idle_threads; 281 static int idle_max_cpu; 282 static char idle_comm[] = "<idle>"; 283 284 static u64 get_nsecs(void) 285 { 286 struct timespec ts; 287 288 clock_gettime(CLOCK_MONOTONIC, &ts); 289 290 return ts.tv_sec * NSEC_PER_SEC + ts.tv_nsec; 291 } 292 293 static void burn_nsecs(struct perf_sched *sched, u64 nsecs) 294 { 295 u64 T0 = get_nsecs(), T1; 296 297 do { 298 T1 = get_nsecs(); 299 } while (T1 + sched->run_measurement_overhead < T0 + nsecs); 300 } 301 302 static void sleep_nsecs(u64 nsecs) 303 { 304 struct timespec ts; 305 306 ts.tv_nsec = nsecs % 999999999; 307 ts.tv_sec = nsecs / 999999999; 308 309 nanosleep(&ts, NULL); 310 } 311 312 static void calibrate_run_measurement_overhead(struct perf_sched *sched) 313 { 314 u64 T0, T1, delta, min_delta = NSEC_PER_SEC; 315 int i; 316 317 for (i = 0; i < 10; i++) { 318 T0 = get_nsecs(); 319 burn_nsecs(sched, 0); 320 T1 = get_nsecs(); 321 delta = T1-T0; 322 min_delta = min(min_delta, delta); 323 } 324 sched->run_measurement_overhead = min_delta; 325 326 printf("run measurement overhead: %" PRIu64 " nsecs\n", min_delta); 327 } 328 329 static void calibrate_sleep_measurement_overhead(struct perf_sched *sched) 330 { 331 u64 T0, T1, delta, min_delta = NSEC_PER_SEC; 332 int i; 333 334 for (i = 0; i < 10; i++) { 335 T0 = get_nsecs(); 336 sleep_nsecs(10000); 337 T1 = get_nsecs(); 338 delta = T1-T0; 339 min_delta = min(min_delta, delta); 340 } 341 min_delta -= 10000; 342 sched->sleep_measurement_overhead = min_delta; 343 344 printf("sleep measurement overhead: %" PRIu64 " nsecs\n", min_delta); 345 } 346 347 static struct sched_atom * 348 get_new_event(struct task_desc *task, u64 timestamp) 349 { 350 struct sched_atom *event = zalloc(sizeof(*event)); 351 unsigned long idx = task->nr_events; 352 size_t size; 353 354 event->timestamp = timestamp; 355 event->nr = idx; 356 357 task->nr_events++; 358 size = sizeof(struct sched_atom *) * task->nr_events; 359 task->atoms = realloc(task->atoms, size); 360 BUG_ON(!task->atoms); 361 362 task->atoms[idx] = event; 363 364 return event; 365 } 366 367 static struct sched_atom *last_event(struct task_desc *task) 368 { 369 if (!task->nr_events) 370 return NULL; 371 372 return task->atoms[task->nr_events - 1]; 373 } 374 375 static void add_sched_event_run(struct perf_sched *sched, struct task_desc *task, 376 u64 timestamp, u64 duration) 377 { 378 struct sched_atom *event, *curr_event = last_event(task); 379 380 /* 381 * optimize an existing RUN event by merging this one 382 * to it: 383 */ 384 if (curr_event && curr_event->type == SCHED_EVENT_RUN) { 385 sched->nr_run_events_optimized++; 386 curr_event->duration += duration; 387 return; 388 } 389 390 event = get_new_event(task, timestamp); 391 392 event->type = SCHED_EVENT_RUN; 393 event->duration = duration; 394 395 sched->nr_run_events++; 396 } 397 398 static void add_sched_event_wakeup(struct perf_sched *sched, struct task_desc *task, 399 u64 timestamp, struct task_desc *wakee) 400 { 401 struct sched_atom *event, *wakee_event; 402 403 event = get_new_event(task, timestamp); 404 event->type = SCHED_EVENT_WAKEUP; 405 event->wakee = wakee; 406 407 wakee_event = last_event(wakee); 408 if (!wakee_event || wakee_event->type != SCHED_EVENT_SLEEP) { 409 sched->targetless_wakeups++; 410 return; 411 } 412 if (wakee_event->wait_sem) { 413 sched->multitarget_wakeups++; 414 return; 415 } 416 417 wakee_event->wait_sem = zalloc(sizeof(*wakee_event->wait_sem)); 418 sem_init(wakee_event->wait_sem, 0, 0); 419 wakee_event->specific_wait = 1; 420 event->wait_sem = wakee_event->wait_sem; 421 422 sched->nr_wakeup_events++; 423 } 424 425 static void add_sched_event_sleep(struct perf_sched *sched, struct task_desc *task, 426 u64 timestamp, u64 task_state __maybe_unused) 427 { 428 struct sched_atom *event = get_new_event(task, timestamp); 429 430 event->type = SCHED_EVENT_SLEEP; 431 432 sched->nr_sleep_events++; 433 } 434 435 static struct task_desc *register_pid(struct perf_sched *sched, 436 unsigned long pid, const char *comm) 437 { 438 struct task_desc *task; 439 static int pid_max; 440 441 if (sched->pid_to_task == NULL) { 442 if (sysctl__read_int("kernel/pid_max", &pid_max) < 0) 443 pid_max = MAX_PID; 444 BUG_ON((sched->pid_to_task = calloc(pid_max, sizeof(struct task_desc *))) == NULL); 445 } 446 if (pid >= (unsigned long)pid_max) { 447 BUG_ON((sched->pid_to_task = realloc(sched->pid_to_task, (pid + 1) * 448 sizeof(struct task_desc *))) == NULL); 449 while (pid >= (unsigned long)pid_max) 450 sched->pid_to_task[pid_max++] = NULL; 451 } 452 453 task = sched->pid_to_task[pid]; 454 455 if (task) 456 return task; 457 458 task = zalloc(sizeof(*task)); 459 task->pid = pid; 460 task->nr = sched->nr_tasks; 461 strcpy(task->comm, comm); 462 /* 463 * every task starts in sleeping state - this gets ignored 464 * if there's no wakeup pointing to this sleep state: 465 */ 466 add_sched_event_sleep(sched, task, 0, 0); 467 468 sched->pid_to_task[pid] = task; 469 sched->nr_tasks++; 470 sched->tasks = realloc(sched->tasks, sched->nr_tasks * sizeof(struct task_desc *)); 471 BUG_ON(!sched->tasks); 472 sched->tasks[task->nr] = task; 473 474 if (verbose > 0) 475 printf("registered task #%ld, PID %ld (%s)\n", sched->nr_tasks, pid, comm); 476 477 return task; 478 } 479 480 481 static void print_task_traces(struct perf_sched *sched) 482 { 483 struct task_desc *task; 484 unsigned long i; 485 486 for (i = 0; i < sched->nr_tasks; i++) { 487 task = sched->tasks[i]; 488 printf("task %6ld (%20s:%10ld), nr_events: %ld\n", 489 task->nr, task->comm, task->pid, task->nr_events); 490 } 491 } 492 493 static void add_cross_task_wakeups(struct perf_sched *sched) 494 { 495 struct task_desc *task1, *task2; 496 unsigned long i, j; 497 498 for (i = 0; i < sched->nr_tasks; i++) { 499 task1 = sched->tasks[i]; 500 j = i + 1; 501 if (j == sched->nr_tasks) 502 j = 0; 503 task2 = sched->tasks[j]; 504 add_sched_event_wakeup(sched, task1, 0, task2); 505 } 506 } 507 508 static void perf_sched__process_event(struct perf_sched *sched, 509 struct sched_atom *atom) 510 { 511 int ret = 0; 512 513 switch (atom->type) { 514 case SCHED_EVENT_RUN: 515 burn_nsecs(sched, atom->duration); 516 break; 517 case SCHED_EVENT_SLEEP: 518 if (atom->wait_sem) 519 ret = sem_wait(atom->wait_sem); 520 BUG_ON(ret); 521 break; 522 case SCHED_EVENT_WAKEUP: 523 if (atom->wait_sem) 524 ret = sem_post(atom->wait_sem); 525 BUG_ON(ret); 526 break; 527 case SCHED_EVENT_MIGRATION: 528 break; 529 default: 530 BUG_ON(1); 531 } 532 } 533 534 static u64 get_cpu_usage_nsec_parent(void) 535 { 536 struct rusage ru; 537 u64 sum; 538 int err; 539 540 err = getrusage(RUSAGE_SELF, &ru); 541 BUG_ON(err); 542 543 sum = ru.ru_utime.tv_sec * NSEC_PER_SEC + ru.ru_utime.tv_usec * NSEC_PER_USEC; 544 sum += ru.ru_stime.tv_sec * NSEC_PER_SEC + ru.ru_stime.tv_usec * NSEC_PER_USEC; 545 546 return sum; 547 } 548 549 static int self_open_counters(struct perf_sched *sched, unsigned long cur_task) 550 { 551 struct perf_event_attr attr; 552 char sbuf[STRERR_BUFSIZE], info[STRERR_BUFSIZE]; 553 int fd; 554 struct rlimit limit; 555 bool need_privilege = false; 556 557 memset(&attr, 0, sizeof(attr)); 558 559 attr.type = PERF_TYPE_SOFTWARE; 560 attr.config = PERF_COUNT_SW_TASK_CLOCK; 561 562 force_again: 563 fd = sys_perf_event_open(&attr, 0, -1, -1, 564 perf_event_open_cloexec_flag()); 565 566 if (fd < 0) { 567 if (errno == EMFILE) { 568 if (sched->force) { 569 BUG_ON(getrlimit(RLIMIT_NOFILE, &limit) == -1); 570 limit.rlim_cur += sched->nr_tasks - cur_task; 571 if (limit.rlim_cur > limit.rlim_max) { 572 limit.rlim_max = limit.rlim_cur; 573 need_privilege = true; 574 } 575 if (setrlimit(RLIMIT_NOFILE, &limit) == -1) { 576 if (need_privilege && errno == EPERM) 577 strcpy(info, "Need privilege\n"); 578 } else 579 goto force_again; 580 } else 581 strcpy(info, "Have a try with -f option\n"); 582 } 583 pr_err("Error: sys_perf_event_open() syscall returned " 584 "with %d (%s)\n%s", fd, 585 str_error_r(errno, sbuf, sizeof(sbuf)), info); 586 exit(EXIT_FAILURE); 587 } 588 return fd; 589 } 590 591 static u64 get_cpu_usage_nsec_self(int fd) 592 { 593 u64 runtime; 594 int ret; 595 596 ret = read(fd, &runtime, sizeof(runtime)); 597 BUG_ON(ret != sizeof(runtime)); 598 599 return runtime; 600 } 601 602 struct sched_thread_parms { 603 struct task_desc *task; 604 struct perf_sched *sched; 605 int fd; 606 }; 607 608 static void *thread_func(void *ctx) 609 { 610 struct sched_thread_parms *parms = ctx; 611 struct task_desc *this_task = parms->task; 612 struct perf_sched *sched = parms->sched; 613 u64 cpu_usage_0, cpu_usage_1; 614 unsigned long i, ret; 615 char comm2[22]; 616 int fd = parms->fd; 617 618 zfree(&parms); 619 620 sprintf(comm2, ":%s", this_task->comm); 621 prctl(PR_SET_NAME, comm2); 622 if (fd < 0) 623 return NULL; 624 again: 625 ret = sem_post(&this_task->ready_for_work); 626 BUG_ON(ret); 627 ret = pthread_mutex_lock(&sched->start_work_mutex); 628 BUG_ON(ret); 629 ret = pthread_mutex_unlock(&sched->start_work_mutex); 630 BUG_ON(ret); 631 632 cpu_usage_0 = get_cpu_usage_nsec_self(fd); 633 634 for (i = 0; i < this_task->nr_events; i++) { 635 this_task->curr_event = i; 636 perf_sched__process_event(sched, this_task->atoms[i]); 637 } 638 639 cpu_usage_1 = get_cpu_usage_nsec_self(fd); 640 this_task->cpu_usage = cpu_usage_1 - cpu_usage_0; 641 ret = sem_post(&this_task->work_done_sem); 642 BUG_ON(ret); 643 644 ret = pthread_mutex_lock(&sched->work_done_wait_mutex); 645 BUG_ON(ret); 646 ret = pthread_mutex_unlock(&sched->work_done_wait_mutex); 647 BUG_ON(ret); 648 649 goto again; 650 } 651 652 static void create_tasks(struct perf_sched *sched) 653 { 654 struct task_desc *task; 655 pthread_attr_t attr; 656 unsigned long i; 657 int err; 658 659 err = pthread_attr_init(&attr); 660 BUG_ON(err); 661 err = pthread_attr_setstacksize(&attr, 662 (size_t) max(16 * 1024, PTHREAD_STACK_MIN)); 663 BUG_ON(err); 664 err = pthread_mutex_lock(&sched->start_work_mutex); 665 BUG_ON(err); 666 err = pthread_mutex_lock(&sched->work_done_wait_mutex); 667 BUG_ON(err); 668 for (i = 0; i < sched->nr_tasks; i++) { 669 struct sched_thread_parms *parms = malloc(sizeof(*parms)); 670 BUG_ON(parms == NULL); 671 parms->task = task = sched->tasks[i]; 672 parms->sched = sched; 673 parms->fd = self_open_counters(sched, i); 674 sem_init(&task->sleep_sem, 0, 0); 675 sem_init(&task->ready_for_work, 0, 0); 676 sem_init(&task->work_done_sem, 0, 0); 677 task->curr_event = 0; 678 err = pthread_create(&task->thread, &attr, thread_func, parms); 679 BUG_ON(err); 680 } 681 } 682 683 static void wait_for_tasks(struct perf_sched *sched) 684 { 685 u64 cpu_usage_0, cpu_usage_1; 686 struct task_desc *task; 687 unsigned long i, ret; 688 689 sched->start_time = get_nsecs(); 690 sched->cpu_usage = 0; 691 pthread_mutex_unlock(&sched->work_done_wait_mutex); 692 693 for (i = 0; i < sched->nr_tasks; i++) { 694 task = sched->tasks[i]; 695 ret = sem_wait(&task->ready_for_work); 696 BUG_ON(ret); 697 sem_init(&task->ready_for_work, 0, 0); 698 } 699 ret = pthread_mutex_lock(&sched->work_done_wait_mutex); 700 BUG_ON(ret); 701 702 cpu_usage_0 = get_cpu_usage_nsec_parent(); 703 704 pthread_mutex_unlock(&sched->start_work_mutex); 705 706 for (i = 0; i < sched->nr_tasks; i++) { 707 task = sched->tasks[i]; 708 ret = sem_wait(&task->work_done_sem); 709 BUG_ON(ret); 710 sem_init(&task->work_done_sem, 0, 0); 711 sched->cpu_usage += task->cpu_usage; 712 task->cpu_usage = 0; 713 } 714 715 cpu_usage_1 = get_cpu_usage_nsec_parent(); 716 if (!sched->runavg_cpu_usage) 717 sched->runavg_cpu_usage = sched->cpu_usage; 718 sched->runavg_cpu_usage = (sched->runavg_cpu_usage * (sched->replay_repeat - 1) + sched->cpu_usage) / sched->replay_repeat; 719 720 sched->parent_cpu_usage = cpu_usage_1 - cpu_usage_0; 721 if (!sched->runavg_parent_cpu_usage) 722 sched->runavg_parent_cpu_usage = sched->parent_cpu_usage; 723 sched->runavg_parent_cpu_usage = (sched->runavg_parent_cpu_usage * (sched->replay_repeat - 1) + 724 sched->parent_cpu_usage)/sched->replay_repeat; 725 726 ret = pthread_mutex_lock(&sched->start_work_mutex); 727 BUG_ON(ret); 728 729 for (i = 0; i < sched->nr_tasks; i++) { 730 task = sched->tasks[i]; 731 sem_init(&task->sleep_sem, 0, 0); 732 task->curr_event = 0; 733 } 734 } 735 736 static void run_one_test(struct perf_sched *sched) 737 { 738 u64 T0, T1, delta, avg_delta, fluct; 739 740 T0 = get_nsecs(); 741 wait_for_tasks(sched); 742 T1 = get_nsecs(); 743 744 delta = T1 - T0; 745 sched->sum_runtime += delta; 746 sched->nr_runs++; 747 748 avg_delta = sched->sum_runtime / sched->nr_runs; 749 if (delta < avg_delta) 750 fluct = avg_delta - delta; 751 else 752 fluct = delta - avg_delta; 753 sched->sum_fluct += fluct; 754 if (!sched->run_avg) 755 sched->run_avg = delta; 756 sched->run_avg = (sched->run_avg * (sched->replay_repeat - 1) + delta) / sched->replay_repeat; 757 758 printf("#%-3ld: %0.3f, ", sched->nr_runs, (double)delta / NSEC_PER_MSEC); 759 760 printf("ravg: %0.2f, ", (double)sched->run_avg / NSEC_PER_MSEC); 761 762 printf("cpu: %0.2f / %0.2f", 763 (double)sched->cpu_usage / NSEC_PER_MSEC, (double)sched->runavg_cpu_usage / NSEC_PER_MSEC); 764 765 #if 0 766 /* 767 * rusage statistics done by the parent, these are less 768 * accurate than the sched->sum_exec_runtime based statistics: 769 */ 770 printf(" [%0.2f / %0.2f]", 771 (double)sched->parent_cpu_usage / NSEC_PER_MSEC, 772 (double)sched->runavg_parent_cpu_usage / NSEC_PER_MSEC); 773 #endif 774 775 printf("\n"); 776 777 if (sched->nr_sleep_corrections) 778 printf(" (%ld sleep corrections)\n", sched->nr_sleep_corrections); 779 sched->nr_sleep_corrections = 0; 780 } 781 782 static void test_calibrations(struct perf_sched *sched) 783 { 784 u64 T0, T1; 785 786 T0 = get_nsecs(); 787 burn_nsecs(sched, NSEC_PER_MSEC); 788 T1 = get_nsecs(); 789 790 printf("the run test took %" PRIu64 " nsecs\n", T1 - T0); 791 792 T0 = get_nsecs(); 793 sleep_nsecs(NSEC_PER_MSEC); 794 T1 = get_nsecs(); 795 796 printf("the sleep test took %" PRIu64 " nsecs\n", T1 - T0); 797 } 798 799 static int 800 replay_wakeup_event(struct perf_sched *sched, 801 struct perf_evsel *evsel, struct perf_sample *sample, 802 struct machine *machine __maybe_unused) 803 { 804 const char *comm = perf_evsel__strval(evsel, sample, "comm"); 805 const u32 pid = perf_evsel__intval(evsel, sample, "pid"); 806 struct task_desc *waker, *wakee; 807 808 if (verbose > 0) { 809 printf("sched_wakeup event %p\n", evsel); 810 811 printf(" ... pid %d woke up %s/%d\n", sample->tid, comm, pid); 812 } 813 814 waker = register_pid(sched, sample->tid, "<unknown>"); 815 wakee = register_pid(sched, pid, comm); 816 817 add_sched_event_wakeup(sched, waker, sample->time, wakee); 818 return 0; 819 } 820 821 static int replay_switch_event(struct perf_sched *sched, 822 struct perf_evsel *evsel, 823 struct perf_sample *sample, 824 struct machine *machine __maybe_unused) 825 { 826 const char *prev_comm = perf_evsel__strval(evsel, sample, "prev_comm"), 827 *next_comm = perf_evsel__strval(evsel, sample, "next_comm"); 828 const u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"), 829 next_pid = perf_evsel__intval(evsel, sample, "next_pid"); 830 const u64 prev_state = perf_evsel__intval(evsel, sample, "prev_state"); 831 struct task_desc *prev, __maybe_unused *next; 832 u64 timestamp0, timestamp = sample->time; 833 int cpu = sample->cpu; 834 s64 delta; 835 836 if (verbose > 0) 837 printf("sched_switch event %p\n", evsel); 838 839 if (cpu >= MAX_CPUS || cpu < 0) 840 return 0; 841 842 timestamp0 = sched->cpu_last_switched[cpu]; 843 if (timestamp0) 844 delta = timestamp - timestamp0; 845 else 846 delta = 0; 847 848 if (delta < 0) { 849 pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta); 850 return -1; 851 } 852 853 pr_debug(" ... switch from %s/%d to %s/%d [ran %" PRIu64 " nsecs]\n", 854 prev_comm, prev_pid, next_comm, next_pid, delta); 855 856 prev = register_pid(sched, prev_pid, prev_comm); 857 next = register_pid(sched, next_pid, next_comm); 858 859 sched->cpu_last_switched[cpu] = timestamp; 860 861 add_sched_event_run(sched, prev, timestamp, delta); 862 add_sched_event_sleep(sched, prev, timestamp, prev_state); 863 864 return 0; 865 } 866 867 static int replay_fork_event(struct perf_sched *sched, 868 union perf_event *event, 869 struct machine *machine) 870 { 871 struct thread *child, *parent; 872 873 child = machine__findnew_thread(machine, event->fork.pid, 874 event->fork.tid); 875 parent = machine__findnew_thread(machine, event->fork.ppid, 876 event->fork.ptid); 877 878 if (child == NULL || parent == NULL) { 879 pr_debug("thread does not exist on fork event: child %p, parent %p\n", 880 child, parent); 881 goto out_put; 882 } 883 884 if (verbose > 0) { 885 printf("fork event\n"); 886 printf("... parent: %s/%d\n", thread__comm_str(parent), parent->tid); 887 printf("... child: %s/%d\n", thread__comm_str(child), child->tid); 888 } 889 890 register_pid(sched, parent->tid, thread__comm_str(parent)); 891 register_pid(sched, child->tid, thread__comm_str(child)); 892 out_put: 893 thread__put(child); 894 thread__put(parent); 895 return 0; 896 } 897 898 struct sort_dimension { 899 const char *name; 900 sort_fn_t cmp; 901 struct list_head list; 902 }; 903 904 /* 905 * handle runtime stats saved per thread 906 */ 907 static struct thread_runtime *thread__init_runtime(struct thread *thread) 908 { 909 struct thread_runtime *r; 910 911 r = zalloc(sizeof(struct thread_runtime)); 912 if (!r) 913 return NULL; 914 915 init_stats(&r->run_stats); 916 thread__set_priv(thread, r); 917 918 return r; 919 } 920 921 static struct thread_runtime *thread__get_runtime(struct thread *thread) 922 { 923 struct thread_runtime *tr; 924 925 tr = thread__priv(thread); 926 if (tr == NULL) { 927 tr = thread__init_runtime(thread); 928 if (tr == NULL) 929 pr_debug("Failed to malloc memory for runtime data.\n"); 930 } 931 932 return tr; 933 } 934 935 static int 936 thread_lat_cmp(struct list_head *list, struct work_atoms *l, struct work_atoms *r) 937 { 938 struct sort_dimension *sort; 939 int ret = 0; 940 941 BUG_ON(list_empty(list)); 942 943 list_for_each_entry(sort, list, list) { 944 ret = sort->cmp(l, r); 945 if (ret) 946 return ret; 947 } 948 949 return ret; 950 } 951 952 static struct work_atoms * 953 thread_atoms_search(struct rb_root *root, struct thread *thread, 954 struct list_head *sort_list) 955 { 956 struct rb_node *node = root->rb_node; 957 struct work_atoms key = { .thread = thread }; 958 959 while (node) { 960 struct work_atoms *atoms; 961 int cmp; 962 963 atoms = container_of(node, struct work_atoms, node); 964 965 cmp = thread_lat_cmp(sort_list, &key, atoms); 966 if (cmp > 0) 967 node = node->rb_left; 968 else if (cmp < 0) 969 node = node->rb_right; 970 else { 971 BUG_ON(thread != atoms->thread); 972 return atoms; 973 } 974 } 975 return NULL; 976 } 977 978 static void 979 __thread_latency_insert(struct rb_root *root, struct work_atoms *data, 980 struct list_head *sort_list) 981 { 982 struct rb_node **new = &(root->rb_node), *parent = NULL; 983 984 while (*new) { 985 struct work_atoms *this; 986 int cmp; 987 988 this = container_of(*new, struct work_atoms, node); 989 parent = *new; 990 991 cmp = thread_lat_cmp(sort_list, data, this); 992 993 if (cmp > 0) 994 new = &((*new)->rb_left); 995 else 996 new = &((*new)->rb_right); 997 } 998 999 rb_link_node(&data->node, parent, new); 1000 rb_insert_color(&data->node, root); 1001 } 1002 1003 static int thread_atoms_insert(struct perf_sched *sched, struct thread *thread) 1004 { 1005 struct work_atoms *atoms = zalloc(sizeof(*atoms)); 1006 if (!atoms) { 1007 pr_err("No memory at %s\n", __func__); 1008 return -1; 1009 } 1010 1011 atoms->thread = thread__get(thread); 1012 INIT_LIST_HEAD(&atoms->work_list); 1013 __thread_latency_insert(&sched->atom_root, atoms, &sched->cmp_pid); 1014 return 0; 1015 } 1016 1017 static char sched_out_state(u64 prev_state) 1018 { 1019 const char *str = TASK_STATE_TO_CHAR_STR; 1020 1021 return str[prev_state]; 1022 } 1023 1024 static int 1025 add_sched_out_event(struct work_atoms *atoms, 1026 char run_state, 1027 u64 timestamp) 1028 { 1029 struct work_atom *atom = zalloc(sizeof(*atom)); 1030 if (!atom) { 1031 pr_err("Non memory at %s", __func__); 1032 return -1; 1033 } 1034 1035 atom->sched_out_time = timestamp; 1036 1037 if (run_state == 'R') { 1038 atom->state = THREAD_WAIT_CPU; 1039 atom->wake_up_time = atom->sched_out_time; 1040 } 1041 1042 list_add_tail(&atom->list, &atoms->work_list); 1043 return 0; 1044 } 1045 1046 static void 1047 add_runtime_event(struct work_atoms *atoms, u64 delta, 1048 u64 timestamp __maybe_unused) 1049 { 1050 struct work_atom *atom; 1051 1052 BUG_ON(list_empty(&atoms->work_list)); 1053 1054 atom = list_entry(atoms->work_list.prev, struct work_atom, list); 1055 1056 atom->runtime += delta; 1057 atoms->total_runtime += delta; 1058 } 1059 1060 static void 1061 add_sched_in_event(struct work_atoms *atoms, u64 timestamp) 1062 { 1063 struct work_atom *atom; 1064 u64 delta; 1065 1066 if (list_empty(&atoms->work_list)) 1067 return; 1068 1069 atom = list_entry(atoms->work_list.prev, struct work_atom, list); 1070 1071 if (atom->state != THREAD_WAIT_CPU) 1072 return; 1073 1074 if (timestamp < atom->wake_up_time) { 1075 atom->state = THREAD_IGNORE; 1076 return; 1077 } 1078 1079 atom->state = THREAD_SCHED_IN; 1080 atom->sched_in_time = timestamp; 1081 1082 delta = atom->sched_in_time - atom->wake_up_time; 1083 atoms->total_lat += delta; 1084 if (delta > atoms->max_lat) { 1085 atoms->max_lat = delta; 1086 atoms->max_lat_at = timestamp; 1087 } 1088 atoms->nb_atoms++; 1089 } 1090 1091 static int latency_switch_event(struct perf_sched *sched, 1092 struct perf_evsel *evsel, 1093 struct perf_sample *sample, 1094 struct machine *machine) 1095 { 1096 const u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"), 1097 next_pid = perf_evsel__intval(evsel, sample, "next_pid"); 1098 const u64 prev_state = perf_evsel__intval(evsel, sample, "prev_state"); 1099 struct work_atoms *out_events, *in_events; 1100 struct thread *sched_out, *sched_in; 1101 u64 timestamp0, timestamp = sample->time; 1102 int cpu = sample->cpu, err = -1; 1103 s64 delta; 1104 1105 BUG_ON(cpu >= MAX_CPUS || cpu < 0); 1106 1107 timestamp0 = sched->cpu_last_switched[cpu]; 1108 sched->cpu_last_switched[cpu] = timestamp; 1109 if (timestamp0) 1110 delta = timestamp - timestamp0; 1111 else 1112 delta = 0; 1113 1114 if (delta < 0) { 1115 pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta); 1116 return -1; 1117 } 1118 1119 sched_out = machine__findnew_thread(machine, -1, prev_pid); 1120 sched_in = machine__findnew_thread(machine, -1, next_pid); 1121 if (sched_out == NULL || sched_in == NULL) 1122 goto out_put; 1123 1124 out_events = thread_atoms_search(&sched->atom_root, sched_out, &sched->cmp_pid); 1125 if (!out_events) { 1126 if (thread_atoms_insert(sched, sched_out)) 1127 goto out_put; 1128 out_events = thread_atoms_search(&sched->atom_root, sched_out, &sched->cmp_pid); 1129 if (!out_events) { 1130 pr_err("out-event: Internal tree error"); 1131 goto out_put; 1132 } 1133 } 1134 if (add_sched_out_event(out_events, sched_out_state(prev_state), timestamp)) 1135 return -1; 1136 1137 in_events = thread_atoms_search(&sched->atom_root, sched_in, &sched->cmp_pid); 1138 if (!in_events) { 1139 if (thread_atoms_insert(sched, sched_in)) 1140 goto out_put; 1141 in_events = thread_atoms_search(&sched->atom_root, sched_in, &sched->cmp_pid); 1142 if (!in_events) { 1143 pr_err("in-event: Internal tree error"); 1144 goto out_put; 1145 } 1146 /* 1147 * Take came in we have not heard about yet, 1148 * add in an initial atom in runnable state: 1149 */ 1150 if (add_sched_out_event(in_events, 'R', timestamp)) 1151 goto out_put; 1152 } 1153 add_sched_in_event(in_events, timestamp); 1154 err = 0; 1155 out_put: 1156 thread__put(sched_out); 1157 thread__put(sched_in); 1158 return err; 1159 } 1160 1161 static int latency_runtime_event(struct perf_sched *sched, 1162 struct perf_evsel *evsel, 1163 struct perf_sample *sample, 1164 struct machine *machine) 1165 { 1166 const u32 pid = perf_evsel__intval(evsel, sample, "pid"); 1167 const u64 runtime = perf_evsel__intval(evsel, sample, "runtime"); 1168 struct thread *thread = machine__findnew_thread(machine, -1, pid); 1169 struct work_atoms *atoms = thread_atoms_search(&sched->atom_root, thread, &sched->cmp_pid); 1170 u64 timestamp = sample->time; 1171 int cpu = sample->cpu, err = -1; 1172 1173 if (thread == NULL) 1174 return -1; 1175 1176 BUG_ON(cpu >= MAX_CPUS || cpu < 0); 1177 if (!atoms) { 1178 if (thread_atoms_insert(sched, thread)) 1179 goto out_put; 1180 atoms = thread_atoms_search(&sched->atom_root, thread, &sched->cmp_pid); 1181 if (!atoms) { 1182 pr_err("in-event: Internal tree error"); 1183 goto out_put; 1184 } 1185 if (add_sched_out_event(atoms, 'R', timestamp)) 1186 goto out_put; 1187 } 1188 1189 add_runtime_event(atoms, runtime, timestamp); 1190 err = 0; 1191 out_put: 1192 thread__put(thread); 1193 return err; 1194 } 1195 1196 static int latency_wakeup_event(struct perf_sched *sched, 1197 struct perf_evsel *evsel, 1198 struct perf_sample *sample, 1199 struct machine *machine) 1200 { 1201 const u32 pid = perf_evsel__intval(evsel, sample, "pid"); 1202 struct work_atoms *atoms; 1203 struct work_atom *atom; 1204 struct thread *wakee; 1205 u64 timestamp = sample->time; 1206 int err = -1; 1207 1208 wakee = machine__findnew_thread(machine, -1, pid); 1209 if (wakee == NULL) 1210 return -1; 1211 atoms = thread_atoms_search(&sched->atom_root, wakee, &sched->cmp_pid); 1212 if (!atoms) { 1213 if (thread_atoms_insert(sched, wakee)) 1214 goto out_put; 1215 atoms = thread_atoms_search(&sched->atom_root, wakee, &sched->cmp_pid); 1216 if (!atoms) { 1217 pr_err("wakeup-event: Internal tree error"); 1218 goto out_put; 1219 } 1220 if (add_sched_out_event(atoms, 'S', timestamp)) 1221 goto out_put; 1222 } 1223 1224 BUG_ON(list_empty(&atoms->work_list)); 1225 1226 atom = list_entry(atoms->work_list.prev, struct work_atom, list); 1227 1228 /* 1229 * As we do not guarantee the wakeup event happens when 1230 * task is out of run queue, also may happen when task is 1231 * on run queue and wakeup only change ->state to TASK_RUNNING, 1232 * then we should not set the ->wake_up_time when wake up a 1233 * task which is on run queue. 1234 * 1235 * You WILL be missing events if you've recorded only 1236 * one CPU, or are only looking at only one, so don't 1237 * skip in this case. 1238 */ 1239 if (sched->profile_cpu == -1 && atom->state != THREAD_SLEEPING) 1240 goto out_ok; 1241 1242 sched->nr_timestamps++; 1243 if (atom->sched_out_time > timestamp) { 1244 sched->nr_unordered_timestamps++; 1245 goto out_ok; 1246 } 1247 1248 atom->state = THREAD_WAIT_CPU; 1249 atom->wake_up_time = timestamp; 1250 out_ok: 1251 err = 0; 1252 out_put: 1253 thread__put(wakee); 1254 return err; 1255 } 1256 1257 static int latency_migrate_task_event(struct perf_sched *sched, 1258 struct perf_evsel *evsel, 1259 struct perf_sample *sample, 1260 struct machine *machine) 1261 { 1262 const u32 pid = perf_evsel__intval(evsel, sample, "pid"); 1263 u64 timestamp = sample->time; 1264 struct work_atoms *atoms; 1265 struct work_atom *atom; 1266 struct thread *migrant; 1267 int err = -1; 1268 1269 /* 1270 * Only need to worry about migration when profiling one CPU. 1271 */ 1272 if (sched->profile_cpu == -1) 1273 return 0; 1274 1275 migrant = machine__findnew_thread(machine, -1, pid); 1276 if (migrant == NULL) 1277 return -1; 1278 atoms = thread_atoms_search(&sched->atom_root, migrant, &sched->cmp_pid); 1279 if (!atoms) { 1280 if (thread_atoms_insert(sched, migrant)) 1281 goto out_put; 1282 register_pid(sched, migrant->tid, thread__comm_str(migrant)); 1283 atoms = thread_atoms_search(&sched->atom_root, migrant, &sched->cmp_pid); 1284 if (!atoms) { 1285 pr_err("migration-event: Internal tree error"); 1286 goto out_put; 1287 } 1288 if (add_sched_out_event(atoms, 'R', timestamp)) 1289 goto out_put; 1290 } 1291 1292 BUG_ON(list_empty(&atoms->work_list)); 1293 1294 atom = list_entry(atoms->work_list.prev, struct work_atom, list); 1295 atom->sched_in_time = atom->sched_out_time = atom->wake_up_time = timestamp; 1296 1297 sched->nr_timestamps++; 1298 1299 if (atom->sched_out_time > timestamp) 1300 sched->nr_unordered_timestamps++; 1301 err = 0; 1302 out_put: 1303 thread__put(migrant); 1304 return err; 1305 } 1306 1307 static void output_lat_thread(struct perf_sched *sched, struct work_atoms *work_list) 1308 { 1309 int i; 1310 int ret; 1311 u64 avg; 1312 char max_lat_at[32]; 1313 1314 if (!work_list->nb_atoms) 1315 return; 1316 /* 1317 * Ignore idle threads: 1318 */ 1319 if (!strcmp(thread__comm_str(work_list->thread), "swapper")) 1320 return; 1321 1322 sched->all_runtime += work_list->total_runtime; 1323 sched->all_count += work_list->nb_atoms; 1324 1325 if (work_list->num_merged > 1) 1326 ret = printf(" %s:(%d) ", thread__comm_str(work_list->thread), work_list->num_merged); 1327 else 1328 ret = printf(" %s:%d ", thread__comm_str(work_list->thread), work_list->thread->tid); 1329 1330 for (i = 0; i < 24 - ret; i++) 1331 printf(" "); 1332 1333 avg = work_list->total_lat / work_list->nb_atoms; 1334 timestamp__scnprintf_usec(work_list->max_lat_at, max_lat_at, sizeof(max_lat_at)); 1335 1336 printf("|%11.3f ms |%9" PRIu64 " | avg:%9.3f ms | max:%9.3f ms | max at: %13s s\n", 1337 (double)work_list->total_runtime / NSEC_PER_MSEC, 1338 work_list->nb_atoms, (double)avg / NSEC_PER_MSEC, 1339 (double)work_list->max_lat / NSEC_PER_MSEC, 1340 max_lat_at); 1341 } 1342 1343 static int pid_cmp(struct work_atoms *l, struct work_atoms *r) 1344 { 1345 if (l->thread == r->thread) 1346 return 0; 1347 if (l->thread->tid < r->thread->tid) 1348 return -1; 1349 if (l->thread->tid > r->thread->tid) 1350 return 1; 1351 return (int)(l->thread - r->thread); 1352 } 1353 1354 static int avg_cmp(struct work_atoms *l, struct work_atoms *r) 1355 { 1356 u64 avgl, avgr; 1357 1358 if (!l->nb_atoms) 1359 return -1; 1360 1361 if (!r->nb_atoms) 1362 return 1; 1363 1364 avgl = l->total_lat / l->nb_atoms; 1365 avgr = r->total_lat / r->nb_atoms; 1366 1367 if (avgl < avgr) 1368 return -1; 1369 if (avgl > avgr) 1370 return 1; 1371 1372 return 0; 1373 } 1374 1375 static int max_cmp(struct work_atoms *l, struct work_atoms *r) 1376 { 1377 if (l->max_lat < r->max_lat) 1378 return -1; 1379 if (l->max_lat > r->max_lat) 1380 return 1; 1381 1382 return 0; 1383 } 1384 1385 static int switch_cmp(struct work_atoms *l, struct work_atoms *r) 1386 { 1387 if (l->nb_atoms < r->nb_atoms) 1388 return -1; 1389 if (l->nb_atoms > r->nb_atoms) 1390 return 1; 1391 1392 return 0; 1393 } 1394 1395 static int runtime_cmp(struct work_atoms *l, struct work_atoms *r) 1396 { 1397 if (l->total_runtime < r->total_runtime) 1398 return -1; 1399 if (l->total_runtime > r->total_runtime) 1400 return 1; 1401 1402 return 0; 1403 } 1404 1405 static int sort_dimension__add(const char *tok, struct list_head *list) 1406 { 1407 size_t i; 1408 static struct sort_dimension avg_sort_dimension = { 1409 .name = "avg", 1410 .cmp = avg_cmp, 1411 }; 1412 static struct sort_dimension max_sort_dimension = { 1413 .name = "max", 1414 .cmp = max_cmp, 1415 }; 1416 static struct sort_dimension pid_sort_dimension = { 1417 .name = "pid", 1418 .cmp = pid_cmp, 1419 }; 1420 static struct sort_dimension runtime_sort_dimension = { 1421 .name = "runtime", 1422 .cmp = runtime_cmp, 1423 }; 1424 static struct sort_dimension switch_sort_dimension = { 1425 .name = "switch", 1426 .cmp = switch_cmp, 1427 }; 1428 struct sort_dimension *available_sorts[] = { 1429 &pid_sort_dimension, 1430 &avg_sort_dimension, 1431 &max_sort_dimension, 1432 &switch_sort_dimension, 1433 &runtime_sort_dimension, 1434 }; 1435 1436 for (i = 0; i < ARRAY_SIZE(available_sorts); i++) { 1437 if (!strcmp(available_sorts[i]->name, tok)) { 1438 list_add_tail(&available_sorts[i]->list, list); 1439 1440 return 0; 1441 } 1442 } 1443 1444 return -1; 1445 } 1446 1447 static void perf_sched__sort_lat(struct perf_sched *sched) 1448 { 1449 struct rb_node *node; 1450 struct rb_root *root = &sched->atom_root; 1451 again: 1452 for (;;) { 1453 struct work_atoms *data; 1454 node = rb_first(root); 1455 if (!node) 1456 break; 1457 1458 rb_erase(node, root); 1459 data = rb_entry(node, struct work_atoms, node); 1460 __thread_latency_insert(&sched->sorted_atom_root, data, &sched->sort_list); 1461 } 1462 if (root == &sched->atom_root) { 1463 root = &sched->merged_atom_root; 1464 goto again; 1465 } 1466 } 1467 1468 static int process_sched_wakeup_event(struct perf_tool *tool, 1469 struct perf_evsel *evsel, 1470 struct perf_sample *sample, 1471 struct machine *machine) 1472 { 1473 struct perf_sched *sched = container_of(tool, struct perf_sched, tool); 1474 1475 if (sched->tp_handler->wakeup_event) 1476 return sched->tp_handler->wakeup_event(sched, evsel, sample, machine); 1477 1478 return 0; 1479 } 1480 1481 union map_priv { 1482 void *ptr; 1483 bool color; 1484 }; 1485 1486 static bool thread__has_color(struct thread *thread) 1487 { 1488 union map_priv priv = { 1489 .ptr = thread__priv(thread), 1490 }; 1491 1492 return priv.color; 1493 } 1494 1495 static struct thread* 1496 map__findnew_thread(struct perf_sched *sched, struct machine *machine, pid_t pid, pid_t tid) 1497 { 1498 struct thread *thread = machine__findnew_thread(machine, pid, tid); 1499 union map_priv priv = { 1500 .color = false, 1501 }; 1502 1503 if (!sched->map.color_pids || !thread || thread__priv(thread)) 1504 return thread; 1505 1506 if (thread_map__has(sched->map.color_pids, tid)) 1507 priv.color = true; 1508 1509 thread__set_priv(thread, priv.ptr); 1510 return thread; 1511 } 1512 1513 static int map_switch_event(struct perf_sched *sched, struct perf_evsel *evsel, 1514 struct perf_sample *sample, struct machine *machine) 1515 { 1516 const u32 next_pid = perf_evsel__intval(evsel, sample, "next_pid"); 1517 struct thread *sched_in; 1518 struct thread_runtime *tr; 1519 int new_shortname; 1520 u64 timestamp0, timestamp = sample->time; 1521 s64 delta; 1522 int i, this_cpu = sample->cpu; 1523 int cpus_nr; 1524 bool new_cpu = false; 1525 const char *color = PERF_COLOR_NORMAL; 1526 char stimestamp[32]; 1527 1528 BUG_ON(this_cpu >= MAX_CPUS || this_cpu < 0); 1529 1530 if (this_cpu > sched->max_cpu) 1531 sched->max_cpu = this_cpu; 1532 1533 if (sched->map.comp) { 1534 cpus_nr = bitmap_weight(sched->map.comp_cpus_mask, MAX_CPUS); 1535 if (!test_and_set_bit(this_cpu, sched->map.comp_cpus_mask)) { 1536 sched->map.comp_cpus[cpus_nr++] = this_cpu; 1537 new_cpu = true; 1538 } 1539 } else 1540 cpus_nr = sched->max_cpu; 1541 1542 timestamp0 = sched->cpu_last_switched[this_cpu]; 1543 sched->cpu_last_switched[this_cpu] = timestamp; 1544 if (timestamp0) 1545 delta = timestamp - timestamp0; 1546 else 1547 delta = 0; 1548 1549 if (delta < 0) { 1550 pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta); 1551 return -1; 1552 } 1553 1554 sched_in = map__findnew_thread(sched, machine, -1, next_pid); 1555 if (sched_in == NULL) 1556 return -1; 1557 1558 tr = thread__get_runtime(sched_in); 1559 if (tr == NULL) { 1560 thread__put(sched_in); 1561 return -1; 1562 } 1563 1564 sched->curr_thread[this_cpu] = thread__get(sched_in); 1565 1566 printf(" "); 1567 1568 new_shortname = 0; 1569 if (!tr->shortname[0]) { 1570 if (!strcmp(thread__comm_str(sched_in), "swapper")) { 1571 /* 1572 * Don't allocate a letter-number for swapper:0 1573 * as a shortname. Instead, we use '.' for it. 1574 */ 1575 tr->shortname[0] = '.'; 1576 tr->shortname[1] = ' '; 1577 } else { 1578 tr->shortname[0] = sched->next_shortname1; 1579 tr->shortname[1] = sched->next_shortname2; 1580 1581 if (sched->next_shortname1 < 'Z') { 1582 sched->next_shortname1++; 1583 } else { 1584 sched->next_shortname1 = 'A'; 1585 if (sched->next_shortname2 < '9') 1586 sched->next_shortname2++; 1587 else 1588 sched->next_shortname2 = '0'; 1589 } 1590 } 1591 new_shortname = 1; 1592 } 1593 1594 for (i = 0; i < cpus_nr; i++) { 1595 int cpu = sched->map.comp ? sched->map.comp_cpus[i] : i; 1596 struct thread *curr_thread = sched->curr_thread[cpu]; 1597 struct thread_runtime *curr_tr; 1598 const char *pid_color = color; 1599 const char *cpu_color = color; 1600 1601 if (curr_thread && thread__has_color(curr_thread)) 1602 pid_color = COLOR_PIDS; 1603 1604 if (sched->map.cpus && !cpu_map__has(sched->map.cpus, cpu)) 1605 continue; 1606 1607 if (sched->map.color_cpus && cpu_map__has(sched->map.color_cpus, cpu)) 1608 cpu_color = COLOR_CPUS; 1609 1610 if (cpu != this_cpu) 1611 color_fprintf(stdout, color, " "); 1612 else 1613 color_fprintf(stdout, cpu_color, "*"); 1614 1615 if (sched->curr_thread[cpu]) { 1616 curr_tr = thread__get_runtime(sched->curr_thread[cpu]); 1617 if (curr_tr == NULL) { 1618 thread__put(sched_in); 1619 return -1; 1620 } 1621 color_fprintf(stdout, pid_color, "%2s ", curr_tr->shortname); 1622 } else 1623 color_fprintf(stdout, color, " "); 1624 } 1625 1626 if (sched->map.cpus && !cpu_map__has(sched->map.cpus, this_cpu)) 1627 goto out; 1628 1629 timestamp__scnprintf_usec(timestamp, stimestamp, sizeof(stimestamp)); 1630 color_fprintf(stdout, color, " %12s secs ", stimestamp); 1631 if (new_shortname || tr->comm_changed || (verbose > 0 && sched_in->tid)) { 1632 const char *pid_color = color; 1633 1634 if (thread__has_color(sched_in)) 1635 pid_color = COLOR_PIDS; 1636 1637 color_fprintf(stdout, pid_color, "%s => %s:%d", 1638 tr->shortname, thread__comm_str(sched_in), sched_in->tid); 1639 tr->comm_changed = false; 1640 } 1641 1642 if (sched->map.comp && new_cpu) 1643 color_fprintf(stdout, color, " (CPU %d)", this_cpu); 1644 1645 out: 1646 color_fprintf(stdout, color, "\n"); 1647 1648 thread__put(sched_in); 1649 1650 return 0; 1651 } 1652 1653 static int process_sched_switch_event(struct perf_tool *tool, 1654 struct perf_evsel *evsel, 1655 struct perf_sample *sample, 1656 struct machine *machine) 1657 { 1658 struct perf_sched *sched = container_of(tool, struct perf_sched, tool); 1659 int this_cpu = sample->cpu, err = 0; 1660 u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"), 1661 next_pid = perf_evsel__intval(evsel, sample, "next_pid"); 1662 1663 if (sched->curr_pid[this_cpu] != (u32)-1) { 1664 /* 1665 * Are we trying to switch away a PID that is 1666 * not current? 1667 */ 1668 if (sched->curr_pid[this_cpu] != prev_pid) 1669 sched->nr_context_switch_bugs++; 1670 } 1671 1672 if (sched->tp_handler->switch_event) 1673 err = sched->tp_handler->switch_event(sched, evsel, sample, machine); 1674 1675 sched->curr_pid[this_cpu] = next_pid; 1676 return err; 1677 } 1678 1679 static int process_sched_runtime_event(struct perf_tool *tool, 1680 struct perf_evsel *evsel, 1681 struct perf_sample *sample, 1682 struct machine *machine) 1683 { 1684 struct perf_sched *sched = container_of(tool, struct perf_sched, tool); 1685 1686 if (sched->tp_handler->runtime_event) 1687 return sched->tp_handler->runtime_event(sched, evsel, sample, machine); 1688 1689 return 0; 1690 } 1691 1692 static int perf_sched__process_fork_event(struct perf_tool *tool, 1693 union perf_event *event, 1694 struct perf_sample *sample, 1695 struct machine *machine) 1696 { 1697 struct perf_sched *sched = container_of(tool, struct perf_sched, tool); 1698 1699 /* run the fork event through the perf machineruy */ 1700 perf_event__process_fork(tool, event, sample, machine); 1701 1702 /* and then run additional processing needed for this command */ 1703 if (sched->tp_handler->fork_event) 1704 return sched->tp_handler->fork_event(sched, event, machine); 1705 1706 return 0; 1707 } 1708 1709 static int process_sched_migrate_task_event(struct perf_tool *tool, 1710 struct perf_evsel *evsel, 1711 struct perf_sample *sample, 1712 struct machine *machine) 1713 { 1714 struct perf_sched *sched = container_of(tool, struct perf_sched, tool); 1715 1716 if (sched->tp_handler->migrate_task_event) 1717 return sched->tp_handler->migrate_task_event(sched, evsel, sample, machine); 1718 1719 return 0; 1720 } 1721 1722 typedef int (*tracepoint_handler)(struct perf_tool *tool, 1723 struct perf_evsel *evsel, 1724 struct perf_sample *sample, 1725 struct machine *machine); 1726 1727 static int perf_sched__process_tracepoint_sample(struct perf_tool *tool __maybe_unused, 1728 union perf_event *event __maybe_unused, 1729 struct perf_sample *sample, 1730 struct perf_evsel *evsel, 1731 struct machine *machine) 1732 { 1733 int err = 0; 1734 1735 if (evsel->handler != NULL) { 1736 tracepoint_handler f = evsel->handler; 1737 err = f(tool, evsel, sample, machine); 1738 } 1739 1740 return err; 1741 } 1742 1743 static int perf_sched__process_comm(struct perf_tool *tool __maybe_unused, 1744 union perf_event *event, 1745 struct perf_sample *sample, 1746 struct machine *machine) 1747 { 1748 struct thread *thread; 1749 struct thread_runtime *tr; 1750 int err; 1751 1752 err = perf_event__process_comm(tool, event, sample, machine); 1753 if (err) 1754 return err; 1755 1756 thread = machine__find_thread(machine, sample->pid, sample->tid); 1757 if (!thread) { 1758 pr_err("Internal error: can't find thread\n"); 1759 return -1; 1760 } 1761 1762 tr = thread__get_runtime(thread); 1763 if (tr == NULL) { 1764 thread__put(thread); 1765 return -1; 1766 } 1767 1768 tr->comm_changed = true; 1769 thread__put(thread); 1770 1771 return 0; 1772 } 1773 1774 static int perf_sched__read_events(struct perf_sched *sched) 1775 { 1776 const struct perf_evsel_str_handler handlers[] = { 1777 { "sched:sched_switch", process_sched_switch_event, }, 1778 { "sched:sched_stat_runtime", process_sched_runtime_event, }, 1779 { "sched:sched_wakeup", process_sched_wakeup_event, }, 1780 { "sched:sched_wakeup_new", process_sched_wakeup_event, }, 1781 { "sched:sched_migrate_task", process_sched_migrate_task_event, }, 1782 }; 1783 struct perf_session *session; 1784 struct perf_data data = { 1785 .file = { 1786 .path = input_name, 1787 }, 1788 .mode = PERF_DATA_MODE_READ, 1789 .force = sched->force, 1790 }; 1791 int rc = -1; 1792 1793 session = perf_session__new(&data, false, &sched->tool); 1794 if (session == NULL) { 1795 pr_debug("No Memory for session\n"); 1796 return -1; 1797 } 1798 1799 symbol__init(&session->header.env); 1800 1801 if (perf_session__set_tracepoints_handlers(session, handlers)) 1802 goto out_delete; 1803 1804 if (perf_session__has_traces(session, "record -R")) { 1805 int err = perf_session__process_events(session); 1806 if (err) { 1807 pr_err("Failed to process events, error %d", err); 1808 goto out_delete; 1809 } 1810 1811 sched->nr_events = session->evlist->stats.nr_events[0]; 1812 sched->nr_lost_events = session->evlist->stats.total_lost; 1813 sched->nr_lost_chunks = session->evlist->stats.nr_events[PERF_RECORD_LOST]; 1814 } 1815 1816 rc = 0; 1817 out_delete: 1818 perf_session__delete(session); 1819 return rc; 1820 } 1821 1822 /* 1823 * scheduling times are printed as msec.usec 1824 */ 1825 static inline void print_sched_time(unsigned long long nsecs, int width) 1826 { 1827 unsigned long msecs; 1828 unsigned long usecs; 1829 1830 msecs = nsecs / NSEC_PER_MSEC; 1831 nsecs -= msecs * NSEC_PER_MSEC; 1832 usecs = nsecs / NSEC_PER_USEC; 1833 printf("%*lu.%03lu ", width, msecs, usecs); 1834 } 1835 1836 /* 1837 * returns runtime data for event, allocating memory for it the 1838 * first time it is used. 1839 */ 1840 static struct evsel_runtime *perf_evsel__get_runtime(struct perf_evsel *evsel) 1841 { 1842 struct evsel_runtime *r = evsel->priv; 1843 1844 if (r == NULL) { 1845 r = zalloc(sizeof(struct evsel_runtime)); 1846 evsel->priv = r; 1847 } 1848 1849 return r; 1850 } 1851 1852 /* 1853 * save last time event was seen per cpu 1854 */ 1855 static void perf_evsel__save_time(struct perf_evsel *evsel, 1856 u64 timestamp, u32 cpu) 1857 { 1858 struct evsel_runtime *r = perf_evsel__get_runtime(evsel); 1859 1860 if (r == NULL) 1861 return; 1862 1863 if ((cpu >= r->ncpu) || (r->last_time == NULL)) { 1864 int i, n = __roundup_pow_of_two(cpu+1); 1865 void *p = r->last_time; 1866 1867 p = realloc(r->last_time, n * sizeof(u64)); 1868 if (!p) 1869 return; 1870 1871 r->last_time = p; 1872 for (i = r->ncpu; i < n; ++i) 1873 r->last_time[i] = (u64) 0; 1874 1875 r->ncpu = n; 1876 } 1877 1878 r->last_time[cpu] = timestamp; 1879 } 1880 1881 /* returns last time this event was seen on the given cpu */ 1882 static u64 perf_evsel__get_time(struct perf_evsel *evsel, u32 cpu) 1883 { 1884 struct evsel_runtime *r = perf_evsel__get_runtime(evsel); 1885 1886 if ((r == NULL) || (r->last_time == NULL) || (cpu >= r->ncpu)) 1887 return 0; 1888 1889 return r->last_time[cpu]; 1890 } 1891 1892 static int comm_width = 30; 1893 1894 static char *timehist_get_commstr(struct thread *thread) 1895 { 1896 static char str[32]; 1897 const char *comm = thread__comm_str(thread); 1898 pid_t tid = thread->tid; 1899 pid_t pid = thread->pid_; 1900 int n; 1901 1902 if (pid == 0) 1903 n = scnprintf(str, sizeof(str), "%s", comm); 1904 1905 else if (tid != pid) 1906 n = scnprintf(str, sizeof(str), "%s[%d/%d]", comm, tid, pid); 1907 1908 else 1909 n = scnprintf(str, sizeof(str), "%s[%d]", comm, tid); 1910 1911 if (n > comm_width) 1912 comm_width = n; 1913 1914 return str; 1915 } 1916 1917 static void timehist_header(struct perf_sched *sched) 1918 { 1919 u32 ncpus = sched->max_cpu + 1; 1920 u32 i, j; 1921 1922 printf("%15s %6s ", "time", "cpu"); 1923 1924 if (sched->show_cpu_visual) { 1925 printf(" "); 1926 for (i = 0, j = 0; i < ncpus; ++i) { 1927 printf("%x", j++); 1928 if (j > 15) 1929 j = 0; 1930 } 1931 printf(" "); 1932 } 1933 1934 printf(" %-*s %9s %9s %9s", comm_width, 1935 "task name", "wait time", "sch delay", "run time"); 1936 1937 if (sched->show_state) 1938 printf(" %s", "state"); 1939 1940 printf("\n"); 1941 1942 /* 1943 * units row 1944 */ 1945 printf("%15s %-6s ", "", ""); 1946 1947 if (sched->show_cpu_visual) 1948 printf(" %*s ", ncpus, ""); 1949 1950 printf(" %-*s %9s %9s %9s", comm_width, 1951 "[tid/pid]", "(msec)", "(msec)", "(msec)"); 1952 1953 if (sched->show_state) 1954 printf(" %5s", ""); 1955 1956 printf("\n"); 1957 1958 /* 1959 * separator 1960 */ 1961 printf("%.15s %.6s ", graph_dotted_line, graph_dotted_line); 1962 1963 if (sched->show_cpu_visual) 1964 printf(" %.*s ", ncpus, graph_dotted_line); 1965 1966 printf(" %.*s %.9s %.9s %.9s", comm_width, 1967 graph_dotted_line, graph_dotted_line, graph_dotted_line, 1968 graph_dotted_line); 1969 1970 if (sched->show_state) 1971 printf(" %.5s", graph_dotted_line); 1972 1973 printf("\n"); 1974 } 1975 1976 static char task_state_char(struct thread *thread, int state) 1977 { 1978 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR; 1979 unsigned bit = state ? ffs(state) : 0; 1980 1981 /* 'I' for idle */ 1982 if (thread->tid == 0) 1983 return 'I'; 1984 1985 return bit < sizeof(state_to_char) - 1 ? state_to_char[bit] : '?'; 1986 } 1987 1988 static void timehist_print_sample(struct perf_sched *sched, 1989 struct perf_evsel *evsel, 1990 struct perf_sample *sample, 1991 struct addr_location *al, 1992 struct thread *thread, 1993 u64 t, int state) 1994 { 1995 struct thread_runtime *tr = thread__priv(thread); 1996 const char *next_comm = perf_evsel__strval(evsel, sample, "next_comm"); 1997 const u32 next_pid = perf_evsel__intval(evsel, sample, "next_pid"); 1998 u32 max_cpus = sched->max_cpu + 1; 1999 char tstr[64]; 2000 char nstr[30]; 2001 u64 wait_time; 2002 2003 timestamp__scnprintf_usec(t, tstr, sizeof(tstr)); 2004 printf("%15s [%04d] ", tstr, sample->cpu); 2005 2006 if (sched->show_cpu_visual) { 2007 u32 i; 2008 char c; 2009 2010 printf(" "); 2011 for (i = 0; i < max_cpus; ++i) { 2012 /* flag idle times with 'i'; others are sched events */ 2013 if (i == sample->cpu) 2014 c = (thread->tid == 0) ? 'i' : 's'; 2015 else 2016 c = ' '; 2017 printf("%c", c); 2018 } 2019 printf(" "); 2020 } 2021 2022 printf(" %-*s ", comm_width, timehist_get_commstr(thread)); 2023 2024 wait_time = tr->dt_sleep + tr->dt_iowait + tr->dt_preempt; 2025 print_sched_time(wait_time, 6); 2026 2027 print_sched_time(tr->dt_delay, 6); 2028 print_sched_time(tr->dt_run, 6); 2029 2030 if (sched->show_state) 2031 printf(" %5c ", task_state_char(thread, state)); 2032 2033 if (sched->show_next) { 2034 snprintf(nstr, sizeof(nstr), "next: %s[%d]", next_comm, next_pid); 2035 printf(" %-*s", comm_width, nstr); 2036 } 2037 2038 if (sched->show_wakeups && !sched->show_next) 2039 printf(" %-*s", comm_width, ""); 2040 2041 if (thread->tid == 0) 2042 goto out; 2043 2044 if (sched->show_callchain) 2045 printf(" "); 2046 2047 sample__fprintf_sym(sample, al, 0, 2048 EVSEL__PRINT_SYM | EVSEL__PRINT_ONELINE | 2049 EVSEL__PRINT_CALLCHAIN_ARROW | 2050 EVSEL__PRINT_SKIP_IGNORED, 2051 &callchain_cursor, stdout); 2052 2053 out: 2054 printf("\n"); 2055 } 2056 2057 /* 2058 * Explanation of delta-time stats: 2059 * 2060 * t = time of current schedule out event 2061 * tprev = time of previous sched out event 2062 * also time of schedule-in event for current task 2063 * last_time = time of last sched change event for current task 2064 * (i.e, time process was last scheduled out) 2065 * ready_to_run = time of wakeup for current task 2066 * 2067 * -----|------------|------------|------------|------ 2068 * last ready tprev t 2069 * time to run 2070 * 2071 * |-------- dt_wait --------| 2072 * |- dt_delay -|-- dt_run --| 2073 * 2074 * dt_run = run time of current task 2075 * dt_wait = time between last schedule out event for task and tprev 2076 * represents time spent off the cpu 2077 * dt_delay = time between wakeup and schedule-in of task 2078 */ 2079 2080 static void timehist_update_runtime_stats(struct thread_runtime *r, 2081 u64 t, u64 tprev) 2082 { 2083 r->dt_delay = 0; 2084 r->dt_sleep = 0; 2085 r->dt_iowait = 0; 2086 r->dt_preempt = 0; 2087 r->dt_run = 0; 2088 2089 if (tprev) { 2090 r->dt_run = t - tprev; 2091 if (r->ready_to_run) { 2092 if (r->ready_to_run > tprev) 2093 pr_debug("time travel: wakeup time for task > previous sched_switch event\n"); 2094 else 2095 r->dt_delay = tprev - r->ready_to_run; 2096 } 2097 2098 if (r->last_time > tprev) 2099 pr_debug("time travel: last sched out time for task > previous sched_switch event\n"); 2100 else if (r->last_time) { 2101 u64 dt_wait = tprev - r->last_time; 2102 2103 if (r->last_state == TASK_RUNNING) 2104 r->dt_preempt = dt_wait; 2105 else if (r->last_state == TASK_UNINTERRUPTIBLE) 2106 r->dt_iowait = dt_wait; 2107 else 2108 r->dt_sleep = dt_wait; 2109 } 2110 } 2111 2112 update_stats(&r->run_stats, r->dt_run); 2113 2114 r->total_run_time += r->dt_run; 2115 r->total_delay_time += r->dt_delay; 2116 r->total_sleep_time += r->dt_sleep; 2117 r->total_iowait_time += r->dt_iowait; 2118 r->total_preempt_time += r->dt_preempt; 2119 } 2120 2121 static bool is_idle_sample(struct perf_sample *sample, 2122 struct perf_evsel *evsel) 2123 { 2124 /* pid 0 == swapper == idle task */ 2125 if (strcmp(perf_evsel__name(evsel), "sched:sched_switch") == 0) 2126 return perf_evsel__intval(evsel, sample, "prev_pid") == 0; 2127 2128 return sample->pid == 0; 2129 } 2130 2131 static void save_task_callchain(struct perf_sched *sched, 2132 struct perf_sample *sample, 2133 struct perf_evsel *evsel, 2134 struct machine *machine) 2135 { 2136 struct callchain_cursor *cursor = &callchain_cursor; 2137 struct thread *thread; 2138 2139 /* want main thread for process - has maps */ 2140 thread = machine__findnew_thread(machine, sample->pid, sample->pid); 2141 if (thread == NULL) { 2142 pr_debug("Failed to get thread for pid %d.\n", sample->pid); 2143 return; 2144 } 2145 2146 if (!symbol_conf.use_callchain || sample->callchain == NULL) 2147 return; 2148 2149 if (thread__resolve_callchain(thread, cursor, evsel, sample, 2150 NULL, NULL, sched->max_stack + 2) != 0) { 2151 if (verbose > 0) 2152 pr_err("Failed to resolve callchain. Skipping\n"); 2153 2154 return; 2155 } 2156 2157 callchain_cursor_commit(cursor); 2158 2159 while (true) { 2160 struct callchain_cursor_node *node; 2161 struct symbol *sym; 2162 2163 node = callchain_cursor_current(cursor); 2164 if (node == NULL) 2165 break; 2166 2167 sym = node->sym; 2168 if (sym) { 2169 if (!strcmp(sym->name, "schedule") || 2170 !strcmp(sym->name, "__schedule") || 2171 !strcmp(sym->name, "preempt_schedule")) 2172 sym->ignore = 1; 2173 } 2174 2175 callchain_cursor_advance(cursor); 2176 } 2177 } 2178 2179 static int init_idle_thread(struct thread *thread) 2180 { 2181 struct idle_thread_runtime *itr; 2182 2183 thread__set_comm(thread, idle_comm, 0); 2184 2185 itr = zalloc(sizeof(*itr)); 2186 if (itr == NULL) 2187 return -ENOMEM; 2188 2189 init_stats(&itr->tr.run_stats); 2190 callchain_init(&itr->callchain); 2191 callchain_cursor_reset(&itr->cursor); 2192 thread__set_priv(thread, itr); 2193 2194 return 0; 2195 } 2196 2197 /* 2198 * Track idle stats per cpu by maintaining a local thread 2199 * struct for the idle task on each cpu. 2200 */ 2201 static int init_idle_threads(int ncpu) 2202 { 2203 int i, ret; 2204 2205 idle_threads = zalloc(ncpu * sizeof(struct thread *)); 2206 if (!idle_threads) 2207 return -ENOMEM; 2208 2209 idle_max_cpu = ncpu; 2210 2211 /* allocate the actual thread struct if needed */ 2212 for (i = 0; i < ncpu; ++i) { 2213 idle_threads[i] = thread__new(0, 0); 2214 if (idle_threads[i] == NULL) 2215 return -ENOMEM; 2216 2217 ret = init_idle_thread(idle_threads[i]); 2218 if (ret < 0) 2219 return ret; 2220 } 2221 2222 return 0; 2223 } 2224 2225 static void free_idle_threads(void) 2226 { 2227 int i; 2228 2229 if (idle_threads == NULL) 2230 return; 2231 2232 for (i = 0; i < idle_max_cpu; ++i) { 2233 if ((idle_threads[i])) 2234 thread__delete(idle_threads[i]); 2235 } 2236 2237 free(idle_threads); 2238 } 2239 2240 static struct thread *get_idle_thread(int cpu) 2241 { 2242 /* 2243 * expand/allocate array of pointers to local thread 2244 * structs if needed 2245 */ 2246 if ((cpu >= idle_max_cpu) || (idle_threads == NULL)) { 2247 int i, j = __roundup_pow_of_two(cpu+1); 2248 void *p; 2249 2250 p = realloc(idle_threads, j * sizeof(struct thread *)); 2251 if (!p) 2252 return NULL; 2253 2254 idle_threads = (struct thread **) p; 2255 for (i = idle_max_cpu; i < j; ++i) 2256 idle_threads[i] = NULL; 2257 2258 idle_max_cpu = j; 2259 } 2260 2261 /* allocate a new thread struct if needed */ 2262 if (idle_threads[cpu] == NULL) { 2263 idle_threads[cpu] = thread__new(0, 0); 2264 2265 if (idle_threads[cpu]) { 2266 if (init_idle_thread(idle_threads[cpu]) < 0) 2267 return NULL; 2268 } 2269 } 2270 2271 return idle_threads[cpu]; 2272 } 2273 2274 static void save_idle_callchain(struct idle_thread_runtime *itr, 2275 struct perf_sample *sample) 2276 { 2277 if (!symbol_conf.use_callchain || sample->callchain == NULL) 2278 return; 2279 2280 callchain_cursor__copy(&itr->cursor, &callchain_cursor); 2281 } 2282 2283 static struct thread *timehist_get_thread(struct perf_sched *sched, 2284 struct perf_sample *sample, 2285 struct machine *machine, 2286 struct perf_evsel *evsel) 2287 { 2288 struct thread *thread; 2289 2290 if (is_idle_sample(sample, evsel)) { 2291 thread = get_idle_thread(sample->cpu); 2292 if (thread == NULL) 2293 pr_err("Failed to get idle thread for cpu %d.\n", sample->cpu); 2294 2295 } else { 2296 /* there were samples with tid 0 but non-zero pid */ 2297 thread = machine__findnew_thread(machine, sample->pid, 2298 sample->tid ?: sample->pid); 2299 if (thread == NULL) { 2300 pr_debug("Failed to get thread for tid %d. skipping sample.\n", 2301 sample->tid); 2302 } 2303 2304 save_task_callchain(sched, sample, evsel, machine); 2305 if (sched->idle_hist) { 2306 struct thread *idle; 2307 struct idle_thread_runtime *itr; 2308 2309 idle = get_idle_thread(sample->cpu); 2310 if (idle == NULL) { 2311 pr_err("Failed to get idle thread for cpu %d.\n", sample->cpu); 2312 return NULL; 2313 } 2314 2315 itr = thread__priv(idle); 2316 if (itr == NULL) 2317 return NULL; 2318 2319 itr->last_thread = thread; 2320 2321 /* copy task callchain when entering to idle */ 2322 if (perf_evsel__intval(evsel, sample, "next_pid") == 0) 2323 save_idle_callchain(itr, sample); 2324 } 2325 } 2326 2327 return thread; 2328 } 2329 2330 static bool timehist_skip_sample(struct perf_sched *sched, 2331 struct thread *thread, 2332 struct perf_evsel *evsel, 2333 struct perf_sample *sample) 2334 { 2335 bool rc = false; 2336 2337 if (thread__is_filtered(thread)) { 2338 rc = true; 2339 sched->skipped_samples++; 2340 } 2341 2342 if (sched->idle_hist) { 2343 if (strcmp(perf_evsel__name(evsel), "sched:sched_switch")) 2344 rc = true; 2345 else if (perf_evsel__intval(evsel, sample, "prev_pid") != 0 && 2346 perf_evsel__intval(evsel, sample, "next_pid") != 0) 2347 rc = true; 2348 } 2349 2350 return rc; 2351 } 2352 2353 static void timehist_print_wakeup_event(struct perf_sched *sched, 2354 struct perf_evsel *evsel, 2355 struct perf_sample *sample, 2356 struct machine *machine, 2357 struct thread *awakened) 2358 { 2359 struct thread *thread; 2360 char tstr[64]; 2361 2362 thread = machine__findnew_thread(machine, sample->pid, sample->tid); 2363 if (thread == NULL) 2364 return; 2365 2366 /* show wakeup unless both awakee and awaker are filtered */ 2367 if (timehist_skip_sample(sched, thread, evsel, sample) && 2368 timehist_skip_sample(sched, awakened, evsel, sample)) { 2369 return; 2370 } 2371 2372 timestamp__scnprintf_usec(sample->time, tstr, sizeof(tstr)); 2373 printf("%15s [%04d] ", tstr, sample->cpu); 2374 if (sched->show_cpu_visual) 2375 printf(" %*s ", sched->max_cpu + 1, ""); 2376 2377 printf(" %-*s ", comm_width, timehist_get_commstr(thread)); 2378 2379 /* dt spacer */ 2380 printf(" %9s %9s %9s ", "", "", ""); 2381 2382 printf("awakened: %s", timehist_get_commstr(awakened)); 2383 2384 printf("\n"); 2385 } 2386 2387 static int timehist_sched_wakeup_event(struct perf_tool *tool, 2388 union perf_event *event __maybe_unused, 2389 struct perf_evsel *evsel, 2390 struct perf_sample *sample, 2391 struct machine *machine) 2392 { 2393 struct perf_sched *sched = container_of(tool, struct perf_sched, tool); 2394 struct thread *thread; 2395 struct thread_runtime *tr = NULL; 2396 /* want pid of awakened task not pid in sample */ 2397 const u32 pid = perf_evsel__intval(evsel, sample, "pid"); 2398 2399 thread = machine__findnew_thread(machine, 0, pid); 2400 if (thread == NULL) 2401 return -1; 2402 2403 tr = thread__get_runtime(thread); 2404 if (tr == NULL) 2405 return -1; 2406 2407 if (tr->ready_to_run == 0) 2408 tr->ready_to_run = sample->time; 2409 2410 /* show wakeups if requested */ 2411 if (sched->show_wakeups && 2412 !perf_time__skip_sample(&sched->ptime, sample->time)) 2413 timehist_print_wakeup_event(sched, evsel, sample, machine, thread); 2414 2415 return 0; 2416 } 2417 2418 static void timehist_print_migration_event(struct perf_sched *sched, 2419 struct perf_evsel *evsel, 2420 struct perf_sample *sample, 2421 struct machine *machine, 2422 struct thread *migrated) 2423 { 2424 struct thread *thread; 2425 char tstr[64]; 2426 u32 max_cpus = sched->max_cpu + 1; 2427 u32 ocpu, dcpu; 2428 2429 if (sched->summary_only) 2430 return; 2431 2432 max_cpus = sched->max_cpu + 1; 2433 ocpu = perf_evsel__intval(evsel, sample, "orig_cpu"); 2434 dcpu = perf_evsel__intval(evsel, sample, "dest_cpu"); 2435 2436 thread = machine__findnew_thread(machine, sample->pid, sample->tid); 2437 if (thread == NULL) 2438 return; 2439 2440 if (timehist_skip_sample(sched, thread, evsel, sample) && 2441 timehist_skip_sample(sched, migrated, evsel, sample)) { 2442 return; 2443 } 2444 2445 timestamp__scnprintf_usec(sample->time, tstr, sizeof(tstr)); 2446 printf("%15s [%04d] ", tstr, sample->cpu); 2447 2448 if (sched->show_cpu_visual) { 2449 u32 i; 2450 char c; 2451 2452 printf(" "); 2453 for (i = 0; i < max_cpus; ++i) { 2454 c = (i == sample->cpu) ? 'm' : ' '; 2455 printf("%c", c); 2456 } 2457 printf(" "); 2458 } 2459 2460 printf(" %-*s ", comm_width, timehist_get_commstr(thread)); 2461 2462 /* dt spacer */ 2463 printf(" %9s %9s %9s ", "", "", ""); 2464 2465 printf("migrated: %s", timehist_get_commstr(migrated)); 2466 printf(" cpu %d => %d", ocpu, dcpu); 2467 2468 printf("\n"); 2469 } 2470 2471 static int timehist_migrate_task_event(struct perf_tool *tool, 2472 union perf_event *event __maybe_unused, 2473 struct perf_evsel *evsel, 2474 struct perf_sample *sample, 2475 struct machine *machine) 2476 { 2477 struct perf_sched *sched = container_of(tool, struct perf_sched, tool); 2478 struct thread *thread; 2479 struct thread_runtime *tr = NULL; 2480 /* want pid of migrated task not pid in sample */ 2481 const u32 pid = perf_evsel__intval(evsel, sample, "pid"); 2482 2483 thread = machine__findnew_thread(machine, 0, pid); 2484 if (thread == NULL) 2485 return -1; 2486 2487 tr = thread__get_runtime(thread); 2488 if (tr == NULL) 2489 return -1; 2490 2491 tr->migrations++; 2492 2493 /* show migrations if requested */ 2494 timehist_print_migration_event(sched, evsel, sample, machine, thread); 2495 2496 return 0; 2497 } 2498 2499 static int timehist_sched_change_event(struct perf_tool *tool, 2500 union perf_event *event, 2501 struct perf_evsel *evsel, 2502 struct perf_sample *sample, 2503 struct machine *machine) 2504 { 2505 struct perf_sched *sched = container_of(tool, struct perf_sched, tool); 2506 struct perf_time_interval *ptime = &sched->ptime; 2507 struct addr_location al; 2508 struct thread *thread; 2509 struct thread_runtime *tr = NULL; 2510 u64 tprev, t = sample->time; 2511 int rc = 0; 2512 int state = perf_evsel__intval(evsel, sample, "prev_state"); 2513 2514 2515 if (machine__resolve(machine, &al, sample) < 0) { 2516 pr_err("problem processing %d event. skipping it\n", 2517 event->header.type); 2518 rc = -1; 2519 goto out; 2520 } 2521 2522 thread = timehist_get_thread(sched, sample, machine, evsel); 2523 if (thread == NULL) { 2524 rc = -1; 2525 goto out; 2526 } 2527 2528 if (timehist_skip_sample(sched, thread, evsel, sample)) 2529 goto out; 2530 2531 tr = thread__get_runtime(thread); 2532 if (tr == NULL) { 2533 rc = -1; 2534 goto out; 2535 } 2536 2537 tprev = perf_evsel__get_time(evsel, sample->cpu); 2538 2539 /* 2540 * If start time given: 2541 * - sample time is under window user cares about - skip sample 2542 * - tprev is under window user cares about - reset to start of window 2543 */ 2544 if (ptime->start && ptime->start > t) 2545 goto out; 2546 2547 if (tprev && ptime->start > tprev) 2548 tprev = ptime->start; 2549 2550 /* 2551 * If end time given: 2552 * - previous sched event is out of window - we are done 2553 * - sample time is beyond window user cares about - reset it 2554 * to close out stats for time window interest 2555 */ 2556 if (ptime->end) { 2557 if (tprev > ptime->end) 2558 goto out; 2559 2560 if (t > ptime->end) 2561 t = ptime->end; 2562 } 2563 2564 if (!sched->idle_hist || thread->tid == 0) { 2565 timehist_update_runtime_stats(tr, t, tprev); 2566 2567 if (sched->idle_hist) { 2568 struct idle_thread_runtime *itr = (void *)tr; 2569 struct thread_runtime *last_tr; 2570 2571 BUG_ON(thread->tid != 0); 2572 2573 if (itr->last_thread == NULL) 2574 goto out; 2575 2576 /* add current idle time as last thread's runtime */ 2577 last_tr = thread__get_runtime(itr->last_thread); 2578 if (last_tr == NULL) 2579 goto out; 2580 2581 timehist_update_runtime_stats(last_tr, t, tprev); 2582 /* 2583 * remove delta time of last thread as it's not updated 2584 * and otherwise it will show an invalid value next 2585 * time. we only care total run time and run stat. 2586 */ 2587 last_tr->dt_run = 0; 2588 last_tr->dt_delay = 0; 2589 last_tr->dt_sleep = 0; 2590 last_tr->dt_iowait = 0; 2591 last_tr->dt_preempt = 0; 2592 2593 if (itr->cursor.nr) 2594 callchain_append(&itr->callchain, &itr->cursor, t - tprev); 2595 2596 itr->last_thread = NULL; 2597 } 2598 } 2599 2600 if (!sched->summary_only) 2601 timehist_print_sample(sched, evsel, sample, &al, thread, t, state); 2602 2603 out: 2604 if (sched->hist_time.start == 0 && t >= ptime->start) 2605 sched->hist_time.start = t; 2606 if (ptime->end == 0 || t <= ptime->end) 2607 sched->hist_time.end = t; 2608 2609 if (tr) { 2610 /* time of this sched_switch event becomes last time task seen */ 2611 tr->last_time = sample->time; 2612 2613 /* last state is used to determine where to account wait time */ 2614 tr->last_state = state; 2615 2616 /* sched out event for task so reset ready to run time */ 2617 tr->ready_to_run = 0; 2618 } 2619 2620 perf_evsel__save_time(evsel, sample->time, sample->cpu); 2621 2622 return rc; 2623 } 2624 2625 static int timehist_sched_switch_event(struct perf_tool *tool, 2626 union perf_event *event, 2627 struct perf_evsel *evsel, 2628 struct perf_sample *sample, 2629 struct machine *machine __maybe_unused) 2630 { 2631 return timehist_sched_change_event(tool, event, evsel, sample, machine); 2632 } 2633 2634 static int process_lost(struct perf_tool *tool __maybe_unused, 2635 union perf_event *event, 2636 struct perf_sample *sample, 2637 struct machine *machine __maybe_unused) 2638 { 2639 char tstr[64]; 2640 2641 timestamp__scnprintf_usec(sample->time, tstr, sizeof(tstr)); 2642 printf("%15s ", tstr); 2643 printf("lost %" PRIu64 " events on cpu %d\n", event->lost.lost, sample->cpu); 2644 2645 return 0; 2646 } 2647 2648 2649 static void print_thread_runtime(struct thread *t, 2650 struct thread_runtime *r) 2651 { 2652 double mean = avg_stats(&r->run_stats); 2653 float stddev; 2654 2655 printf("%*s %5d %9" PRIu64 " ", 2656 comm_width, timehist_get_commstr(t), t->ppid, 2657 (u64) r->run_stats.n); 2658 2659 print_sched_time(r->total_run_time, 8); 2660 stddev = rel_stddev_stats(stddev_stats(&r->run_stats), mean); 2661 print_sched_time(r->run_stats.min, 6); 2662 printf(" "); 2663 print_sched_time((u64) mean, 6); 2664 printf(" "); 2665 print_sched_time(r->run_stats.max, 6); 2666 printf(" "); 2667 printf("%5.2f", stddev); 2668 printf(" %5" PRIu64, r->migrations); 2669 printf("\n"); 2670 } 2671 2672 static void print_thread_waittime(struct thread *t, 2673 struct thread_runtime *r) 2674 { 2675 printf("%*s %5d %9" PRIu64 " ", 2676 comm_width, timehist_get_commstr(t), t->ppid, 2677 (u64) r->run_stats.n); 2678 2679 print_sched_time(r->total_run_time, 8); 2680 print_sched_time(r->total_sleep_time, 6); 2681 printf(" "); 2682 print_sched_time(r->total_iowait_time, 6); 2683 printf(" "); 2684 print_sched_time(r->total_preempt_time, 6); 2685 printf(" "); 2686 print_sched_time(r->total_delay_time, 6); 2687 printf("\n"); 2688 } 2689 2690 struct total_run_stats { 2691 struct perf_sched *sched; 2692 u64 sched_count; 2693 u64 task_count; 2694 u64 total_run_time; 2695 }; 2696 2697 static int __show_thread_runtime(struct thread *t, void *priv) 2698 { 2699 struct total_run_stats *stats = priv; 2700 struct thread_runtime *r; 2701 2702 if (thread__is_filtered(t)) 2703 return 0; 2704 2705 r = thread__priv(t); 2706 if (r && r->run_stats.n) { 2707 stats->task_count++; 2708 stats->sched_count += r->run_stats.n; 2709 stats->total_run_time += r->total_run_time; 2710 2711 if (stats->sched->show_state) 2712 print_thread_waittime(t, r); 2713 else 2714 print_thread_runtime(t, r); 2715 } 2716 2717 return 0; 2718 } 2719 2720 static int show_thread_runtime(struct thread *t, void *priv) 2721 { 2722 if (t->dead) 2723 return 0; 2724 2725 return __show_thread_runtime(t, priv); 2726 } 2727 2728 static int show_deadthread_runtime(struct thread *t, void *priv) 2729 { 2730 if (!t->dead) 2731 return 0; 2732 2733 return __show_thread_runtime(t, priv); 2734 } 2735 2736 static size_t callchain__fprintf_folded(FILE *fp, struct callchain_node *node) 2737 { 2738 const char *sep = " <- "; 2739 struct callchain_list *chain; 2740 size_t ret = 0; 2741 char bf[1024]; 2742 bool first; 2743 2744 if (node == NULL) 2745 return 0; 2746 2747 ret = callchain__fprintf_folded(fp, node->parent); 2748 first = (ret == 0); 2749 2750 list_for_each_entry(chain, &node->val, list) { 2751 if (chain->ip >= PERF_CONTEXT_MAX) 2752 continue; 2753 if (chain->ms.sym && chain->ms.sym->ignore) 2754 continue; 2755 ret += fprintf(fp, "%s%s", first ? "" : sep, 2756 callchain_list__sym_name(chain, bf, sizeof(bf), 2757 false)); 2758 first = false; 2759 } 2760 2761 return ret; 2762 } 2763 2764 static size_t timehist_print_idlehist_callchain(struct rb_root *root) 2765 { 2766 size_t ret = 0; 2767 FILE *fp = stdout; 2768 struct callchain_node *chain; 2769 struct rb_node *rb_node = rb_first(root); 2770 2771 printf(" %16s %8s %s\n", "Idle time (msec)", "Count", "Callchains"); 2772 printf(" %.16s %.8s %.50s\n", graph_dotted_line, graph_dotted_line, 2773 graph_dotted_line); 2774 2775 while (rb_node) { 2776 chain = rb_entry(rb_node, struct callchain_node, rb_node); 2777 rb_node = rb_next(rb_node); 2778 2779 ret += fprintf(fp, " "); 2780 print_sched_time(chain->hit, 12); 2781 ret += 16; /* print_sched_time returns 2nd arg + 4 */ 2782 ret += fprintf(fp, " %8d ", chain->count); 2783 ret += callchain__fprintf_folded(fp, chain); 2784 ret += fprintf(fp, "\n"); 2785 } 2786 2787 return ret; 2788 } 2789 2790 static void timehist_print_summary(struct perf_sched *sched, 2791 struct perf_session *session) 2792 { 2793 struct machine *m = &session->machines.host; 2794 struct total_run_stats totals; 2795 u64 task_count; 2796 struct thread *t; 2797 struct thread_runtime *r; 2798 int i; 2799 u64 hist_time = sched->hist_time.end - sched->hist_time.start; 2800 2801 memset(&totals, 0, sizeof(totals)); 2802 totals.sched = sched; 2803 2804 if (sched->idle_hist) { 2805 printf("\nIdle-time summary\n"); 2806 printf("%*s parent sched-out ", comm_width, "comm"); 2807 printf(" idle-time min-idle avg-idle max-idle stddev migrations\n"); 2808 } else if (sched->show_state) { 2809 printf("\nWait-time summary\n"); 2810 printf("%*s parent sched-in ", comm_width, "comm"); 2811 printf(" run-time sleep iowait preempt delay\n"); 2812 } else { 2813 printf("\nRuntime summary\n"); 2814 printf("%*s parent sched-in ", comm_width, "comm"); 2815 printf(" run-time min-run avg-run max-run stddev migrations\n"); 2816 } 2817 printf("%*s (count) ", comm_width, ""); 2818 printf(" (msec) (msec) (msec) (msec) %s\n", 2819 sched->show_state ? "(msec)" : "%"); 2820 printf("%.117s\n", graph_dotted_line); 2821 2822 machine__for_each_thread(m, show_thread_runtime, &totals); 2823 task_count = totals.task_count; 2824 if (!task_count) 2825 printf("<no still running tasks>\n"); 2826 2827 printf("\nTerminated tasks:\n"); 2828 machine__for_each_thread(m, show_deadthread_runtime, &totals); 2829 if (task_count == totals.task_count) 2830 printf("<no terminated tasks>\n"); 2831 2832 /* CPU idle stats not tracked when samples were skipped */ 2833 if (sched->skipped_samples && !sched->idle_hist) 2834 return; 2835 2836 printf("\nIdle stats:\n"); 2837 for (i = 0; i < idle_max_cpu; ++i) { 2838 t = idle_threads[i]; 2839 if (!t) 2840 continue; 2841 2842 r = thread__priv(t); 2843 if (r && r->run_stats.n) { 2844 totals.sched_count += r->run_stats.n; 2845 printf(" CPU %2d idle for ", i); 2846 print_sched_time(r->total_run_time, 6); 2847 printf(" msec (%6.2f%%)\n", 100.0 * r->total_run_time / hist_time); 2848 } else 2849 printf(" CPU %2d idle entire time window\n", i); 2850 } 2851 2852 if (sched->idle_hist && symbol_conf.use_callchain) { 2853 callchain_param.mode = CHAIN_FOLDED; 2854 callchain_param.value = CCVAL_PERIOD; 2855 2856 callchain_register_param(&callchain_param); 2857 2858 printf("\nIdle stats by callchain:\n"); 2859 for (i = 0; i < idle_max_cpu; ++i) { 2860 struct idle_thread_runtime *itr; 2861 2862 t = idle_threads[i]; 2863 if (!t) 2864 continue; 2865 2866 itr = thread__priv(t); 2867 if (itr == NULL) 2868 continue; 2869 2870 callchain_param.sort(&itr->sorted_root, &itr->callchain, 2871 0, &callchain_param); 2872 2873 printf(" CPU %2d:", i); 2874 print_sched_time(itr->tr.total_run_time, 6); 2875 printf(" msec\n"); 2876 timehist_print_idlehist_callchain(&itr->sorted_root); 2877 printf("\n"); 2878 } 2879 } 2880 2881 printf("\n" 2882 " Total number of unique tasks: %" PRIu64 "\n" 2883 "Total number of context switches: %" PRIu64 "\n", 2884 totals.task_count, totals.sched_count); 2885 2886 printf(" Total run time (msec): "); 2887 print_sched_time(totals.total_run_time, 2); 2888 printf("\n"); 2889 2890 printf(" Total scheduling time (msec): "); 2891 print_sched_time(hist_time, 2); 2892 printf(" (x %d)\n", sched->max_cpu); 2893 } 2894 2895 typedef int (*sched_handler)(struct perf_tool *tool, 2896 union perf_event *event, 2897 struct perf_evsel *evsel, 2898 struct perf_sample *sample, 2899 struct machine *machine); 2900 2901 static int perf_timehist__process_sample(struct perf_tool *tool, 2902 union perf_event *event, 2903 struct perf_sample *sample, 2904 struct perf_evsel *evsel, 2905 struct machine *machine) 2906 { 2907 struct perf_sched *sched = container_of(tool, struct perf_sched, tool); 2908 int err = 0; 2909 int this_cpu = sample->cpu; 2910 2911 if (this_cpu > sched->max_cpu) 2912 sched->max_cpu = this_cpu; 2913 2914 if (evsel->handler != NULL) { 2915 sched_handler f = evsel->handler; 2916 2917 err = f(tool, event, evsel, sample, machine); 2918 } 2919 2920 return err; 2921 } 2922 2923 static int timehist_check_attr(struct perf_sched *sched, 2924 struct perf_evlist *evlist) 2925 { 2926 struct perf_evsel *evsel; 2927 struct evsel_runtime *er; 2928 2929 list_for_each_entry(evsel, &evlist->entries, node) { 2930 er = perf_evsel__get_runtime(evsel); 2931 if (er == NULL) { 2932 pr_err("Failed to allocate memory for evsel runtime data\n"); 2933 return -1; 2934 } 2935 2936 if (sched->show_callchain && 2937 !(evsel->attr.sample_type & PERF_SAMPLE_CALLCHAIN)) { 2938 pr_info("Samples do not have callchains.\n"); 2939 sched->show_callchain = 0; 2940 symbol_conf.use_callchain = 0; 2941 } 2942 } 2943 2944 return 0; 2945 } 2946 2947 static int perf_sched__timehist(struct perf_sched *sched) 2948 { 2949 const struct perf_evsel_str_handler handlers[] = { 2950 { "sched:sched_switch", timehist_sched_switch_event, }, 2951 { "sched:sched_wakeup", timehist_sched_wakeup_event, }, 2952 { "sched:sched_wakeup_new", timehist_sched_wakeup_event, }, 2953 }; 2954 const struct perf_evsel_str_handler migrate_handlers[] = { 2955 { "sched:sched_migrate_task", timehist_migrate_task_event, }, 2956 }; 2957 struct perf_data data = { 2958 .file = { 2959 .path = input_name, 2960 }, 2961 .mode = PERF_DATA_MODE_READ, 2962 .force = sched->force, 2963 }; 2964 2965 struct perf_session *session; 2966 struct perf_evlist *evlist; 2967 int err = -1; 2968 2969 /* 2970 * event handlers for timehist option 2971 */ 2972 sched->tool.sample = perf_timehist__process_sample; 2973 sched->tool.mmap = perf_event__process_mmap; 2974 sched->tool.comm = perf_event__process_comm; 2975 sched->tool.exit = perf_event__process_exit; 2976 sched->tool.fork = perf_event__process_fork; 2977 sched->tool.lost = process_lost; 2978 sched->tool.attr = perf_event__process_attr; 2979 sched->tool.tracing_data = perf_event__process_tracing_data; 2980 sched->tool.build_id = perf_event__process_build_id; 2981 2982 sched->tool.ordered_events = true; 2983 sched->tool.ordering_requires_timestamps = true; 2984 2985 symbol_conf.use_callchain = sched->show_callchain; 2986 2987 session = perf_session__new(&data, false, &sched->tool); 2988 if (session == NULL) 2989 return -ENOMEM; 2990 2991 evlist = session->evlist; 2992 2993 symbol__init(&session->header.env); 2994 2995 if (perf_time__parse_str(&sched->ptime, sched->time_str) != 0) { 2996 pr_err("Invalid time string\n"); 2997 return -EINVAL; 2998 } 2999 3000 if (timehist_check_attr(sched, evlist) != 0) 3001 goto out; 3002 3003 setup_pager(); 3004 3005 /* setup per-evsel handlers */ 3006 if (perf_session__set_tracepoints_handlers(session, handlers)) 3007 goto out; 3008 3009 /* sched_switch event at a minimum needs to exist */ 3010 if (!perf_evlist__find_tracepoint_by_name(session->evlist, 3011 "sched:sched_switch")) { 3012 pr_err("No sched_switch events found. Have you run 'perf sched record'?\n"); 3013 goto out; 3014 } 3015 3016 if (sched->show_migrations && 3017 perf_session__set_tracepoints_handlers(session, migrate_handlers)) 3018 goto out; 3019 3020 /* pre-allocate struct for per-CPU idle stats */ 3021 sched->max_cpu = session->header.env.nr_cpus_online; 3022 if (sched->max_cpu == 0) 3023 sched->max_cpu = 4; 3024 if (init_idle_threads(sched->max_cpu)) 3025 goto out; 3026 3027 /* summary_only implies summary option, but don't overwrite summary if set */ 3028 if (sched->summary_only) 3029 sched->summary = sched->summary_only; 3030 3031 if (!sched->summary_only) 3032 timehist_header(sched); 3033 3034 err = perf_session__process_events(session); 3035 if (err) { 3036 pr_err("Failed to process events, error %d", err); 3037 goto out; 3038 } 3039 3040 sched->nr_events = evlist->stats.nr_events[0]; 3041 sched->nr_lost_events = evlist->stats.total_lost; 3042 sched->nr_lost_chunks = evlist->stats.nr_events[PERF_RECORD_LOST]; 3043 3044 if (sched->summary) 3045 timehist_print_summary(sched, session); 3046 3047 out: 3048 free_idle_threads(); 3049 perf_session__delete(session); 3050 3051 return err; 3052 } 3053 3054 3055 static void print_bad_events(struct perf_sched *sched) 3056 { 3057 if (sched->nr_unordered_timestamps && sched->nr_timestamps) { 3058 printf(" INFO: %.3f%% unordered timestamps (%ld out of %ld)\n", 3059 (double)sched->nr_unordered_timestamps/(double)sched->nr_timestamps*100.0, 3060 sched->nr_unordered_timestamps, sched->nr_timestamps); 3061 } 3062 if (sched->nr_lost_events && sched->nr_events) { 3063 printf(" INFO: %.3f%% lost events (%ld out of %ld, in %ld chunks)\n", 3064 (double)sched->nr_lost_events/(double)sched->nr_events * 100.0, 3065 sched->nr_lost_events, sched->nr_events, sched->nr_lost_chunks); 3066 } 3067 if (sched->nr_context_switch_bugs && sched->nr_timestamps) { 3068 printf(" INFO: %.3f%% context switch bugs (%ld out of %ld)", 3069 (double)sched->nr_context_switch_bugs/(double)sched->nr_timestamps*100.0, 3070 sched->nr_context_switch_bugs, sched->nr_timestamps); 3071 if (sched->nr_lost_events) 3072 printf(" (due to lost events?)"); 3073 printf("\n"); 3074 } 3075 } 3076 3077 static void __merge_work_atoms(struct rb_root *root, struct work_atoms *data) 3078 { 3079 struct rb_node **new = &(root->rb_node), *parent = NULL; 3080 struct work_atoms *this; 3081 const char *comm = thread__comm_str(data->thread), *this_comm; 3082 3083 while (*new) { 3084 int cmp; 3085 3086 this = container_of(*new, struct work_atoms, node); 3087 parent = *new; 3088 3089 this_comm = thread__comm_str(this->thread); 3090 cmp = strcmp(comm, this_comm); 3091 if (cmp > 0) { 3092 new = &((*new)->rb_left); 3093 } else if (cmp < 0) { 3094 new = &((*new)->rb_right); 3095 } else { 3096 this->num_merged++; 3097 this->total_runtime += data->total_runtime; 3098 this->nb_atoms += data->nb_atoms; 3099 this->total_lat += data->total_lat; 3100 list_splice(&data->work_list, &this->work_list); 3101 if (this->max_lat < data->max_lat) { 3102 this->max_lat = data->max_lat; 3103 this->max_lat_at = data->max_lat_at; 3104 } 3105 zfree(&data); 3106 return; 3107 } 3108 } 3109 3110 data->num_merged++; 3111 rb_link_node(&data->node, parent, new); 3112 rb_insert_color(&data->node, root); 3113 } 3114 3115 static void perf_sched__merge_lat(struct perf_sched *sched) 3116 { 3117 struct work_atoms *data; 3118 struct rb_node *node; 3119 3120 if (sched->skip_merge) 3121 return; 3122 3123 while ((node = rb_first(&sched->atom_root))) { 3124 rb_erase(node, &sched->atom_root); 3125 data = rb_entry(node, struct work_atoms, node); 3126 __merge_work_atoms(&sched->merged_atom_root, data); 3127 } 3128 } 3129 3130 static int perf_sched__lat(struct perf_sched *sched) 3131 { 3132 struct rb_node *next; 3133 3134 setup_pager(); 3135 3136 if (perf_sched__read_events(sched)) 3137 return -1; 3138 3139 perf_sched__merge_lat(sched); 3140 perf_sched__sort_lat(sched); 3141 3142 printf("\n -----------------------------------------------------------------------------------------------------------------\n"); 3143 printf(" Task | Runtime ms | Switches | Average delay ms | Maximum delay ms | Maximum delay at |\n"); 3144 printf(" -----------------------------------------------------------------------------------------------------------------\n"); 3145 3146 next = rb_first(&sched->sorted_atom_root); 3147 3148 while (next) { 3149 struct work_atoms *work_list; 3150 3151 work_list = rb_entry(next, struct work_atoms, node); 3152 output_lat_thread(sched, work_list); 3153 next = rb_next(next); 3154 thread__zput(work_list->thread); 3155 } 3156 3157 printf(" -----------------------------------------------------------------------------------------------------------------\n"); 3158 printf(" TOTAL: |%11.3f ms |%9" PRIu64 " |\n", 3159 (double)sched->all_runtime / NSEC_PER_MSEC, sched->all_count); 3160 3161 printf(" ---------------------------------------------------\n"); 3162 3163 print_bad_events(sched); 3164 printf("\n"); 3165 3166 return 0; 3167 } 3168 3169 static int setup_map_cpus(struct perf_sched *sched) 3170 { 3171 struct cpu_map *map; 3172 3173 sched->max_cpu = sysconf(_SC_NPROCESSORS_CONF); 3174 3175 if (sched->map.comp) { 3176 sched->map.comp_cpus = zalloc(sched->max_cpu * sizeof(int)); 3177 if (!sched->map.comp_cpus) 3178 return -1; 3179 } 3180 3181 if (!sched->map.cpus_str) 3182 return 0; 3183 3184 map = cpu_map__new(sched->map.cpus_str); 3185 if (!map) { 3186 pr_err("failed to get cpus map from %s\n", sched->map.cpus_str); 3187 return -1; 3188 } 3189 3190 sched->map.cpus = map; 3191 return 0; 3192 } 3193 3194 static int setup_color_pids(struct perf_sched *sched) 3195 { 3196 struct thread_map *map; 3197 3198 if (!sched->map.color_pids_str) 3199 return 0; 3200 3201 map = thread_map__new_by_tid_str(sched->map.color_pids_str); 3202 if (!map) { 3203 pr_err("failed to get thread map from %s\n", sched->map.color_pids_str); 3204 return -1; 3205 } 3206 3207 sched->map.color_pids = map; 3208 return 0; 3209 } 3210 3211 static int setup_color_cpus(struct perf_sched *sched) 3212 { 3213 struct cpu_map *map; 3214 3215 if (!sched->map.color_cpus_str) 3216 return 0; 3217 3218 map = cpu_map__new(sched->map.color_cpus_str); 3219 if (!map) { 3220 pr_err("failed to get thread map from %s\n", sched->map.color_cpus_str); 3221 return -1; 3222 } 3223 3224 sched->map.color_cpus = map; 3225 return 0; 3226 } 3227 3228 static int perf_sched__map(struct perf_sched *sched) 3229 { 3230 if (setup_map_cpus(sched)) 3231 return -1; 3232 3233 if (setup_color_pids(sched)) 3234 return -1; 3235 3236 if (setup_color_cpus(sched)) 3237 return -1; 3238 3239 setup_pager(); 3240 if (perf_sched__read_events(sched)) 3241 return -1; 3242 print_bad_events(sched); 3243 return 0; 3244 } 3245 3246 static int perf_sched__replay(struct perf_sched *sched) 3247 { 3248 unsigned long i; 3249 3250 calibrate_run_measurement_overhead(sched); 3251 calibrate_sleep_measurement_overhead(sched); 3252 3253 test_calibrations(sched); 3254 3255 if (perf_sched__read_events(sched)) 3256 return -1; 3257 3258 printf("nr_run_events: %ld\n", sched->nr_run_events); 3259 printf("nr_sleep_events: %ld\n", sched->nr_sleep_events); 3260 printf("nr_wakeup_events: %ld\n", sched->nr_wakeup_events); 3261 3262 if (sched->targetless_wakeups) 3263 printf("target-less wakeups: %ld\n", sched->targetless_wakeups); 3264 if (sched->multitarget_wakeups) 3265 printf("multi-target wakeups: %ld\n", sched->multitarget_wakeups); 3266 if (sched->nr_run_events_optimized) 3267 printf("run atoms optimized: %ld\n", 3268 sched->nr_run_events_optimized); 3269 3270 print_task_traces(sched); 3271 add_cross_task_wakeups(sched); 3272 3273 create_tasks(sched); 3274 printf("------------------------------------------------------------\n"); 3275 for (i = 0; i < sched->replay_repeat; i++) 3276 run_one_test(sched); 3277 3278 return 0; 3279 } 3280 3281 static void setup_sorting(struct perf_sched *sched, const struct option *options, 3282 const char * const usage_msg[]) 3283 { 3284 char *tmp, *tok, *str = strdup(sched->sort_order); 3285 3286 for (tok = strtok_r(str, ", ", &tmp); 3287 tok; tok = strtok_r(NULL, ", ", &tmp)) { 3288 if (sort_dimension__add(tok, &sched->sort_list) < 0) { 3289 usage_with_options_msg(usage_msg, options, 3290 "Unknown --sort key: `%s'", tok); 3291 } 3292 } 3293 3294 free(str); 3295 3296 sort_dimension__add("pid", &sched->cmp_pid); 3297 } 3298 3299 static int __cmd_record(int argc, const char **argv) 3300 { 3301 unsigned int rec_argc, i, j; 3302 const char **rec_argv; 3303 const char * const record_args[] = { 3304 "record", 3305 "-a", 3306 "-R", 3307 "-m", "1024", 3308 "-c", "1", 3309 "-e", "sched:sched_switch", 3310 "-e", "sched:sched_stat_wait", 3311 "-e", "sched:sched_stat_sleep", 3312 "-e", "sched:sched_stat_iowait", 3313 "-e", "sched:sched_stat_runtime", 3314 "-e", "sched:sched_process_fork", 3315 "-e", "sched:sched_wakeup", 3316 "-e", "sched:sched_wakeup_new", 3317 "-e", "sched:sched_migrate_task", 3318 }; 3319 3320 rec_argc = ARRAY_SIZE(record_args) + argc - 1; 3321 rec_argv = calloc(rec_argc + 1, sizeof(char *)); 3322 3323 if (rec_argv == NULL) 3324 return -ENOMEM; 3325 3326 for (i = 0; i < ARRAY_SIZE(record_args); i++) 3327 rec_argv[i] = strdup(record_args[i]); 3328 3329 for (j = 1; j < (unsigned int)argc; j++, i++) 3330 rec_argv[i] = argv[j]; 3331 3332 BUG_ON(i != rec_argc); 3333 3334 return cmd_record(i, rec_argv); 3335 } 3336 3337 int cmd_sched(int argc, const char **argv) 3338 { 3339 const char default_sort_order[] = "avg, max, switch, runtime"; 3340 struct perf_sched sched = { 3341 .tool = { 3342 .sample = perf_sched__process_tracepoint_sample, 3343 .comm = perf_sched__process_comm, 3344 .namespaces = perf_event__process_namespaces, 3345 .lost = perf_event__process_lost, 3346 .fork = perf_sched__process_fork_event, 3347 .ordered_events = true, 3348 }, 3349 .cmp_pid = LIST_HEAD_INIT(sched.cmp_pid), 3350 .sort_list = LIST_HEAD_INIT(sched.sort_list), 3351 .start_work_mutex = PTHREAD_MUTEX_INITIALIZER, 3352 .work_done_wait_mutex = PTHREAD_MUTEX_INITIALIZER, 3353 .sort_order = default_sort_order, 3354 .replay_repeat = 10, 3355 .profile_cpu = -1, 3356 .next_shortname1 = 'A', 3357 .next_shortname2 = '0', 3358 .skip_merge = 0, 3359 .show_callchain = 1, 3360 .max_stack = 5, 3361 }; 3362 const struct option sched_options[] = { 3363 OPT_STRING('i', "input", &input_name, "file", 3364 "input file name"), 3365 OPT_INCR('v', "verbose", &verbose, 3366 "be more verbose (show symbol address, etc)"), 3367 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, 3368 "dump raw trace in ASCII"), 3369 OPT_BOOLEAN('f', "force", &sched.force, "don't complain, do it"), 3370 OPT_END() 3371 }; 3372 const struct option latency_options[] = { 3373 OPT_STRING('s', "sort", &sched.sort_order, "key[,key2...]", 3374 "sort by key(s): runtime, switch, avg, max"), 3375 OPT_INTEGER('C', "CPU", &sched.profile_cpu, 3376 "CPU to profile on"), 3377 OPT_BOOLEAN('p', "pids", &sched.skip_merge, 3378 "latency stats per pid instead of per comm"), 3379 OPT_PARENT(sched_options) 3380 }; 3381 const struct option replay_options[] = { 3382 OPT_UINTEGER('r', "repeat", &sched.replay_repeat, 3383 "repeat the workload replay N times (-1: infinite)"), 3384 OPT_PARENT(sched_options) 3385 }; 3386 const struct option map_options[] = { 3387 OPT_BOOLEAN(0, "compact", &sched.map.comp, 3388 "map output in compact mode"), 3389 OPT_STRING(0, "color-pids", &sched.map.color_pids_str, "pids", 3390 "highlight given pids in map"), 3391 OPT_STRING(0, "color-cpus", &sched.map.color_cpus_str, "cpus", 3392 "highlight given CPUs in map"), 3393 OPT_STRING(0, "cpus", &sched.map.cpus_str, "cpus", 3394 "display given CPUs in map"), 3395 OPT_PARENT(sched_options) 3396 }; 3397 const struct option timehist_options[] = { 3398 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name, 3399 "file", "vmlinux pathname"), 3400 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name, 3401 "file", "kallsyms pathname"), 3402 OPT_BOOLEAN('g', "call-graph", &sched.show_callchain, 3403 "Display call chains if present (default on)"), 3404 OPT_UINTEGER(0, "max-stack", &sched.max_stack, 3405 "Maximum number of functions to display backtrace."), 3406 OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory", 3407 "Look for files with symbols relative to this directory"), 3408 OPT_BOOLEAN('s', "summary", &sched.summary_only, 3409 "Show only syscall summary with statistics"), 3410 OPT_BOOLEAN('S', "with-summary", &sched.summary, 3411 "Show all syscalls and summary with statistics"), 3412 OPT_BOOLEAN('w', "wakeups", &sched.show_wakeups, "Show wakeup events"), 3413 OPT_BOOLEAN('n', "next", &sched.show_next, "Show next task"), 3414 OPT_BOOLEAN('M', "migrations", &sched.show_migrations, "Show migration events"), 3415 OPT_BOOLEAN('V', "cpu-visual", &sched.show_cpu_visual, "Add CPU visual"), 3416 OPT_BOOLEAN('I', "idle-hist", &sched.idle_hist, "Show idle events only"), 3417 OPT_STRING(0, "time", &sched.time_str, "str", 3418 "Time span for analysis (start,stop)"), 3419 OPT_BOOLEAN(0, "state", &sched.show_state, "Show task state when sched-out"), 3420 OPT_STRING('p', "pid", &symbol_conf.pid_list_str, "pid[,pid...]", 3421 "analyze events only for given process id(s)"), 3422 OPT_STRING('t', "tid", &symbol_conf.tid_list_str, "tid[,tid...]", 3423 "analyze events only for given thread id(s)"), 3424 OPT_PARENT(sched_options) 3425 }; 3426 3427 const char * const latency_usage[] = { 3428 "perf sched latency [<options>]", 3429 NULL 3430 }; 3431 const char * const replay_usage[] = { 3432 "perf sched replay [<options>]", 3433 NULL 3434 }; 3435 const char * const map_usage[] = { 3436 "perf sched map [<options>]", 3437 NULL 3438 }; 3439 const char * const timehist_usage[] = { 3440 "perf sched timehist [<options>]", 3441 NULL 3442 }; 3443 const char *const sched_subcommands[] = { "record", "latency", "map", 3444 "replay", "script", 3445 "timehist", NULL }; 3446 const char *sched_usage[] = { 3447 NULL, 3448 NULL 3449 }; 3450 struct trace_sched_handler lat_ops = { 3451 .wakeup_event = latency_wakeup_event, 3452 .switch_event = latency_switch_event, 3453 .runtime_event = latency_runtime_event, 3454 .migrate_task_event = latency_migrate_task_event, 3455 }; 3456 struct trace_sched_handler map_ops = { 3457 .switch_event = map_switch_event, 3458 }; 3459 struct trace_sched_handler replay_ops = { 3460 .wakeup_event = replay_wakeup_event, 3461 .switch_event = replay_switch_event, 3462 .fork_event = replay_fork_event, 3463 }; 3464 unsigned int i; 3465 3466 for (i = 0; i < ARRAY_SIZE(sched.curr_pid); i++) 3467 sched.curr_pid[i] = -1; 3468 3469 argc = parse_options_subcommand(argc, argv, sched_options, sched_subcommands, 3470 sched_usage, PARSE_OPT_STOP_AT_NON_OPTION); 3471 if (!argc) 3472 usage_with_options(sched_usage, sched_options); 3473 3474 /* 3475 * Aliased to 'perf script' for now: 3476 */ 3477 if (!strcmp(argv[0], "script")) 3478 return cmd_script(argc, argv); 3479 3480 if (!strncmp(argv[0], "rec", 3)) { 3481 return __cmd_record(argc, argv); 3482 } else if (!strncmp(argv[0], "lat", 3)) { 3483 sched.tp_handler = &lat_ops; 3484 if (argc > 1) { 3485 argc = parse_options(argc, argv, latency_options, latency_usage, 0); 3486 if (argc) 3487 usage_with_options(latency_usage, latency_options); 3488 } 3489 setup_sorting(&sched, latency_options, latency_usage); 3490 return perf_sched__lat(&sched); 3491 } else if (!strcmp(argv[0], "map")) { 3492 if (argc) { 3493 argc = parse_options(argc, argv, map_options, map_usage, 0); 3494 if (argc) 3495 usage_with_options(map_usage, map_options); 3496 } 3497 sched.tp_handler = &map_ops; 3498 setup_sorting(&sched, latency_options, latency_usage); 3499 return perf_sched__map(&sched); 3500 } else if (!strncmp(argv[0], "rep", 3)) { 3501 sched.tp_handler = &replay_ops; 3502 if (argc) { 3503 argc = parse_options(argc, argv, replay_options, replay_usage, 0); 3504 if (argc) 3505 usage_with_options(replay_usage, replay_options); 3506 } 3507 return perf_sched__replay(&sched); 3508 } else if (!strcmp(argv[0], "timehist")) { 3509 if (argc) { 3510 argc = parse_options(argc, argv, timehist_options, 3511 timehist_usage, 0); 3512 if (argc) 3513 usage_with_options(timehist_usage, timehist_options); 3514 } 3515 if ((sched.show_wakeups || sched.show_next) && 3516 sched.summary_only) { 3517 pr_err(" Error: -s and -[n|w] are mutually exclusive.\n"); 3518 parse_options_usage(timehist_usage, timehist_options, "s", true); 3519 if (sched.show_wakeups) 3520 parse_options_usage(NULL, timehist_options, "w", true); 3521 if (sched.show_next) 3522 parse_options_usage(NULL, timehist_options, "n", true); 3523 return -EINVAL; 3524 } 3525 3526 return perf_sched__timehist(&sched); 3527 } else { 3528 usage_with_options(sched_usage, sched_options); 3529 } 3530 3531 return 0; 3532 } 3533