SAFE public projects git trees. - safe/jmp/linux-2.6/blob - tools/perf/builtin-sched.c

   1 #include "builtin.h"
   2 #include "perf.h"
   3
   4 #include "util/util.h"
   5 #include "util/cache.h"
   6 #include "util/symbol.h"
   7 #include "util/thread.h"
   8 #include "util/header.h"
   9
  10 #include "util/parse-options.h"
  11 #include "util/trace-event.h"
  12
  13 #include "util/debug.h"
  14
  15 #include <sys/types.h>
  16 #include <sys/prctl.h>
  17
  18 #include <semaphore.h>
  19 #include <pthread.h>
  20 #include <math.h>
  21
  22 static char                     const *input_name = "perf.data";
  23 static int                      input;
  24 static unsigned long            page_size;
  25 static unsigned long            mmap_window = 32;
  26
  27 static unsigned long            total_comm = 0;
  28
  29 static struct rb_root           threads;
  30 static struct thread            *last_match;
  31
  32 static struct perf_header       *header;
  33 static u64                      sample_type;
  34
  35 static char                     default_sort_order[] = "avg, max, switch, runtime";
  36 static char                     *sort_order = default_sort_order;
  37
  38 #define PR_SET_NAME             15               /* Set process name */
  39 #define MAX_CPUS                4096
  40
  41 #define BUG_ON(x)               assert(!(x))
  42
  43 static u64                      run_measurement_overhead;
  44 static u64                      sleep_measurement_overhead;
  45
  46 #define COMM_LEN                20
  47 #define SYM_LEN                 129
  48
  49 #define MAX_PID                 65536
  50
  51 static unsigned long            nr_tasks;
  52
  53 struct sched_atom;
  54
  55 struct task_desc {
  56         unsigned long           nr;
  57         unsigned long           pid;
  58         char                    comm[COMM_LEN];
  59
  60         unsigned long           nr_events;
  61         unsigned long           curr_event;
  62         struct sched_atom       **atoms;
  63
  64         pthread_t               thread;
  65         sem_t                   sleep_sem;
  66
  67         sem_t                   ready_for_work;
  68         sem_t                   work_done_sem;
  69
  70         u64                     cpu_usage;
  71 };
  72
  73 enum sched_event_type {
  74         SCHED_EVENT_RUN,
  75         SCHED_EVENT_SLEEP,
  76         SCHED_EVENT_WAKEUP,
  77 };
  78
  79 struct sched_atom {
  80         enum sched_event_type   type;
  81         u64                     timestamp;
  82         u64                     duration;
  83         unsigned long           nr;
  84         int                     specific_wait;
  85         sem_t                   *wait_sem;
  86         struct task_desc        *wakee;
  87 };
  88
  89 static struct task_desc         *pid_to_task[MAX_PID];
  90
  91 static struct task_desc         **tasks;
  92
  93 static pthread_mutex_t          start_work_mutex = PTHREAD_MUTEX_INITIALIZER;
  94 static u64                      start_time;
  95
  96 static pthread_mutex_t          work_done_wait_mutex = PTHREAD_MUTEX_INITIALIZER;
  97
  98 static unsigned long            nr_run_events;
  99 static unsigned long            nr_sleep_events;
 100 static unsigned long            nr_wakeup_events;
 101
 102 static unsigned long            nr_sleep_corrections;
 103 static unsigned long            nr_run_events_optimized;
 104
 105 static unsigned long            targetless_wakeups;
 106 static unsigned long            multitarget_wakeups;
 107
 108 static u64                      cpu_usage;
 109 static u64                      runavg_cpu_usage;
 110 static u64                      parent_cpu_usage;
 111 static u64                      runavg_parent_cpu_usage;
 112
 113 static unsigned long            nr_runs;
 114 static u64                      sum_runtime;
 115 static u64                      sum_fluct;
 116 static u64                      run_avg;
 117
 118 static unsigned long            replay_repeat = 10;
 119 static unsigned long            nr_timestamps;
 120 static unsigned long            nr_unordered_timestamps;
 121 static unsigned long            nr_state_machine_bugs;
 122 static unsigned long            nr_context_switch_bugs;
 123 static unsigned long            nr_events;
 124 static unsigned long            nr_lost_chunks;
 125 static unsigned long            nr_lost_events;
 126
 127 #define TASK_STATE_TO_CHAR_STR "RSDTtZX"
 128
 129 enum thread_state {
 130         THREAD_SLEEPING = 0,
 131         THREAD_WAIT_CPU,
 132         THREAD_SCHED_IN,
 133         THREAD_IGNORE
 134 };
 135
 136 struct work_atom {
 137         struct list_head        list;
 138         enum thread_state       state;
 139         u64                     sched_out_time;
 140         u64                     wake_up_time;
 141         u64                     sched_in_time;
 142         u64                     runtime;
 143 };
 144
 145 struct work_atoms {
 146         struct list_head        work_list;
 147         struct thread           *thread;
 148         struct rb_node          node;
 149         u64                     max_lat;
 150         u64                     total_lat;
 151         u64                     nb_atoms;
 152         u64                     total_runtime;
 153 };
 154
 155 typedef int (*sort_fn_t)(struct work_atoms *, struct work_atoms *);
 156
 157 static struct rb_root           atom_root, sorted_atom_root;
 158
 159 static u64                      all_runtime;
 160 static u64                      all_count;
 161
 162
 163 static u64 get_nsecs(void)
 164 {
 165         struct timespec ts;
 166
 167         clock_gettime(CLOCK_MONOTONIC, &ts);
 168
 169         return ts.tv_sec * 1000000000ULL + ts.tv_nsec;
 170 }
 171
 172 static void burn_nsecs(u64 nsecs)
 173 {
 174         u64 T0 = get_nsecs(), T1;
 175
 176         do {
 177                 T1 = get_nsecs();
 178         } while (T1 + run_measurement_overhead < T0 + nsecs);
 179 }
 180
 181 static void sleep_nsecs(u64 nsecs)
 182 {
 183         struct timespec ts;
 184
 185         ts.tv_nsec = nsecs % 999999999;
 186         ts.tv_sec = nsecs / 999999999;
 187
 188         nanosleep(&ts, NULL);
 189 }
 190
 191 static void calibrate_run_measurement_overhead(void)
 192 {
 193         u64 T0, T1, delta, min_delta = 1000000000ULL;
 194         int i;
 195
 196         for (i = 0; i < 10; i++) {
 197                 T0 = get_nsecs();
 198                 burn_nsecs(0);
 199                 T1 = get_nsecs();
 200                 delta = T1-T0;
 201                 min_delta = min(min_delta, delta);
 202         }
 203         run_measurement_overhead = min_delta;
 204
 205         printf("run measurement overhead: %Ld nsecs\n", min_delta);
 206 }
 207
 208 static void calibrate_sleep_measurement_overhead(void)
 209 {
 210         u64 T0, T1, delta, min_delta = 1000000000ULL;
 211         int i;
 212
 213         for (i = 0; i < 10; i++) {
 214                 T0 = get_nsecs();
 215                 sleep_nsecs(10000);
 216                 T1 = get_nsecs();
 217                 delta = T1-T0;
 218                 min_delta = min(min_delta, delta);
 219         }
 220         min_delta -= 10000;
 221         sleep_measurement_overhead = min_delta;
 222
 223         printf("sleep measurement overhead: %Ld nsecs\n", min_delta);
 224 }
 225
 226 static struct sched_atom *
 227 get_new_event(struct task_desc *task, u64 timestamp)
 228 {
 229         struct sched_atom *event = calloc(1, sizeof(*event));
 230         unsigned long idx = task->nr_events;
 231         size_t size;
 232
 233         event->timestamp = timestamp;
 234         event->nr = idx;
 235
 236         task->nr_events++;
 237         size = sizeof(struct sched_atom *) * task->nr_events;
 238         task->atoms = realloc(task->atoms, size);
 239         BUG_ON(!task->atoms);
 240
 241         task->atoms[idx] = event;
 242
 243         return event;
 244 }
 245
 246 static struct sched_atom *last_event(struct task_desc *task)
 247 {
 248         if (!task->nr_events)
 249                 return NULL;
 250
 251         return task->atoms[task->nr_events - 1];
 252 }
 253
 254 static void
 255 add_sched_event_run(struct task_desc *task, u64 timestamp, u64 duration)
 256 {
 257         struct sched_atom *event, *curr_event = last_event(task);
 258
 259         /*
 260          * optimize an existing RUN event by merging this one
 261          * to it:
 262          */
 263         if (curr_event && curr_event->type == SCHED_EVENT_RUN) {
 264                 nr_run_events_optimized++;
 265                 curr_event->duration += duration;
 266                 return;
 267         }
 268
 269         event = get_new_event(task, timestamp);
 270
 271         event->type = SCHED_EVENT_RUN;
 272         event->duration = duration;
 273
 274         nr_run_events++;
 275 }
 276
 277 static void
 278 add_sched_event_wakeup(struct task_desc *task, u64 timestamp,
 279                        struct task_desc *wakee)
 280 {
 281         struct sched_atom *event, *wakee_event;
 282
 283         event = get_new_event(task, timestamp);
 284         event->type = SCHED_EVENT_WAKEUP;
 285         event->wakee = wakee;
 286
 287         wakee_event = last_event(wakee);
 288         if (!wakee_event || wakee_event->type != SCHED_EVENT_SLEEP) {
 289                 targetless_wakeups++;
 290                 return;
 291         }
 292         if (wakee_event->wait_sem) {
 293                 multitarget_wakeups++;
 294                 return;
 295         }
 296
 297         wakee_event->wait_sem = calloc(1, sizeof(*wakee_event->wait_sem));
 298         sem_init(wakee_event->wait_sem, 0, 0);
 299         wakee_event->specific_wait = 1;
 300         event->wait_sem = wakee_event->wait_sem;
 301
 302         nr_wakeup_events++;
 303 }
 304
 305 static void
 306 add_sched_event_sleep(struct task_desc *task, u64 timestamp,
 307                       u64 task_state __used)
 308 {
 309         struct sched_atom *event = get_new_event(task, timestamp);
 310
 311         event->type = SCHED_EVENT_SLEEP;
 312
 313         nr_sleep_events++;
 314 }
 315
 316 static struct task_desc *register_pid(unsigned long pid, const char *comm)
 317 {
 318         struct task_desc *task;
 319
 320         BUG_ON(pid >= MAX_PID);
 321
 322         task = pid_to_task[pid];
 323
 324         if (task)
 325                 return task;
 326
 327         task = calloc(1, sizeof(*task));
 328         task->pid = pid;
 329         task->nr = nr_tasks;
 330         strcpy(task->comm, comm);
 331         /*
 332          * every task starts in sleeping state - this gets ignored
 333          * if there's no wakeup pointing to this sleep state:
 334          */
 335         add_sched_event_sleep(task, 0, 0);
 336
 337         pid_to_task[pid] = task;
 338         nr_tasks++;
 339         tasks = realloc(tasks, nr_tasks*sizeof(struct task_task *));
 340         BUG_ON(!tasks);
 341         tasks[task->nr] = task;
 342
 343         if (verbose)
 344                 printf("registered task #%ld, PID %ld (%s)\n", nr_tasks, pid, comm);
 345
 346         return task;
 347 }
 348
 349
 350 static void print_task_traces(void)
 351 {
 352         struct task_desc *task;
 353         unsigned long i;
 354
 355         for (i = 0; i < nr_tasks; i++) {
 356                 task = tasks[i];
 357                 printf("task %6ld (%20s:%10ld), nr_events: %ld\n",
 358                         task->nr, task->comm, task->pid, task->nr_events);
 359         }
 360 }
 361
 362 static void add_cross_task_wakeups(void)
 363 {
 364         struct task_desc *task1, *task2;
 365         unsigned long i, j;
 366
 367         for (i = 0; i < nr_tasks; i++) {
 368                 task1 = tasks[i];
 369                 j = i + 1;
 370                 if (j == nr_tasks)
 371                         j = 0;
 372                 task2 = tasks[j];
 373                 add_sched_event_wakeup(task1, 0, task2);
 374         }
 375 }
 376
 377 static void
 378 process_sched_event(struct task_desc *this_task __used, struct sched_atom *atom)
 379 {
 380         int ret = 0;
 381         u64 now;
 382         long long delta;
 383
 384         now = get_nsecs();
 385         delta = start_time + atom->timestamp - now;
 386
 387         switch (atom->type) {
 388                 case SCHED_EVENT_RUN:
 389                         burn_nsecs(atom->duration);
 390                         break;
 391                 case SCHED_EVENT_SLEEP:
 392                         if (atom->wait_sem)
 393                                 ret = sem_wait(atom->wait_sem);
 394                         BUG_ON(ret);
 395                         break;
 396                 case SCHED_EVENT_WAKEUP:
 397                         if (atom->wait_sem)
 398                                 ret = sem_post(atom->wait_sem);
 399                         BUG_ON(ret);
 400                         break;
 401                 default:
 402                         BUG_ON(1);
 403         }
 404 }
 405
 406 static u64 get_cpu_usage_nsec_parent(void)
 407 {
 408         struct rusage ru;
 409         u64 sum;
 410         int err;
 411
 412         err = getrusage(RUSAGE_SELF, &ru);
 413         BUG_ON(err);
 414
 415         sum =  ru.ru_utime.tv_sec*1e9 + ru.ru_utime.tv_usec*1e3;
 416         sum += ru.ru_stime.tv_sec*1e9 + ru.ru_stime.tv_usec*1e3;
 417
 418         return sum;
 419 }
 420
 421 static u64 get_cpu_usage_nsec_self(void)
 422 {
 423         char filename [] = "/proc/1234567890/sched";
 424         unsigned long msecs, nsecs;
 425         char *line = NULL;
 426         u64 total = 0;
 427         size_t len = 0;
 428         ssize_t chars;
 429         FILE *file;
 430         int ret;
 431
 432         sprintf(filename, "/proc/%d/sched", getpid());
 433         file = fopen(filename, "r");
 434         BUG_ON(!file);
 435
 436         while ((chars = getline(&line, &len, file)) != -1) {
 437                 ret = sscanf(line, "se.sum_exec_runtime : %ld.%06ld\n",
 438                         &msecs, &nsecs);
 439                 if (ret == 2) {
 440                         total = msecs*1e6 + nsecs;
 441                         break;
 442                 }
 443         }
 444         if (line)
 445                 free(line);
 446         fclose(file);
 447
 448         return total;
 449 }
 450
 451 static void *thread_func(void *ctx)
 452 {
 453         struct task_desc *this_task = ctx;
 454         u64 cpu_usage_0, cpu_usage_1;
 455         unsigned long i, ret;
 456         char comm2[22];
 457
 458         sprintf(comm2, ":%s", this_task->comm);
 459         prctl(PR_SET_NAME, comm2);
 460
 461 again:
 462         ret = sem_post(&this_task->ready_for_work);
 463         BUG_ON(ret);
 464         ret = pthread_mutex_lock(&start_work_mutex);
 465         BUG_ON(ret);
 466         ret = pthread_mutex_unlock(&start_work_mutex);
 467         BUG_ON(ret);
 468
 469         cpu_usage_0 = get_cpu_usage_nsec_self();
 470
 471         for (i = 0; i < this_task->nr_events; i++) {
 472                 this_task->curr_event = i;
 473                 process_sched_event(this_task, this_task->atoms[i]);
 474         }
 475
 476         cpu_usage_1 = get_cpu_usage_nsec_self();
 477         this_task->cpu_usage = cpu_usage_1 - cpu_usage_0;
 478
 479         ret = sem_post(&this_task->work_done_sem);
 480         BUG_ON(ret);
 481
 482         ret = pthread_mutex_lock(&work_done_wait_mutex);
 483         BUG_ON(ret);
 484         ret = pthread_mutex_unlock(&work_done_wait_mutex);
 485         BUG_ON(ret);
 486
 487         goto again;
 488 }
 489
 490 static void create_tasks(void)
 491 {
 492         struct task_desc *task;
 493         pthread_attr_t attr;
 494         unsigned long i;
 495         int err;
 496
 497         err = pthread_attr_init(&attr);
 498         BUG_ON(err);
 499         err = pthread_attr_setstacksize(&attr, (size_t)(16*1024));
 500         BUG_ON(err);
 501         err = pthread_mutex_lock(&start_work_mutex);
 502         BUG_ON(err);
 503         err = pthread_mutex_lock(&work_done_wait_mutex);
 504         BUG_ON(err);
 505         for (i = 0; i < nr_tasks; i++) {
 506                 task = tasks[i];
 507                 sem_init(&task->sleep_sem, 0, 0);
 508                 sem_init(&task->ready_for_work, 0, 0);
 509                 sem_init(&task->work_done_sem, 0, 0);
 510                 task->curr_event = 0;
 511                 err = pthread_create(&task->thread, &attr, thread_func, task);
 512                 BUG_ON(err);
 513         }
 514 }
 515
 516 static void wait_for_tasks(void)
 517 {
 518         u64 cpu_usage_0, cpu_usage_1;
 519         struct task_desc *task;
 520         unsigned long i, ret;
 521
 522         start_time = get_nsecs();
 523         cpu_usage = 0;
 524         pthread_mutex_unlock(&work_done_wait_mutex);
 525
 526         for (i = 0; i < nr_tasks; i++) {
 527                 task = tasks[i];
 528                 ret = sem_wait(&task->ready_for_work);
 529                 BUG_ON(ret);
 530                 sem_init(&task->ready_for_work, 0, 0);
 531         }
 532         ret = pthread_mutex_lock(&work_done_wait_mutex);
 533         BUG_ON(ret);
 534
 535         cpu_usage_0 = get_cpu_usage_nsec_parent();
 536
 537         pthread_mutex_unlock(&start_work_mutex);
 538
 539         for (i = 0; i < nr_tasks; i++) {
 540                 task = tasks[i];
 541                 ret = sem_wait(&task->work_done_sem);
 542                 BUG_ON(ret);
 543                 sem_init(&task->work_done_sem, 0, 0);
 544                 cpu_usage += task->cpu_usage;
 545                 task->cpu_usage = 0;
 546         }
 547
 548         cpu_usage_1 = get_cpu_usage_nsec_parent();
 549         if (!runavg_cpu_usage)
 550                 runavg_cpu_usage = cpu_usage;
 551         runavg_cpu_usage = (runavg_cpu_usage*9 + cpu_usage)/10;
 552
 553         parent_cpu_usage = cpu_usage_1 - cpu_usage_0;
 554         if (!runavg_parent_cpu_usage)
 555                 runavg_parent_cpu_usage = parent_cpu_usage;
 556         runavg_parent_cpu_usage = (runavg_parent_cpu_usage*9 +
 557                                    parent_cpu_usage)/10;
 558
 559         ret = pthread_mutex_lock(&start_work_mutex);
 560         BUG_ON(ret);
 561
 562         for (i = 0; i < nr_tasks; i++) {
 563                 task = tasks[i];
 564                 sem_init(&task->sleep_sem, 0, 0);
 565                 task->curr_event = 0;
 566         }
 567 }
 568
 569 static void run_one_test(void)
 570 {
 571         u64 T0, T1, delta, avg_delta, fluct, std_dev;
 572
 573         T0 = get_nsecs();
 574         wait_for_tasks();
 575         T1 = get_nsecs();
 576
 577         delta = T1 - T0;
 578         sum_runtime += delta;
 579         nr_runs++;
 580
 581         avg_delta = sum_runtime / nr_runs;
 582         if (delta < avg_delta)
 583                 fluct = avg_delta - delta;
 584         else
 585                 fluct = delta - avg_delta;
 586         sum_fluct += fluct;
 587         std_dev = sum_fluct / nr_runs / sqrt(nr_runs);
 588         if (!run_avg)
 589                 run_avg = delta;
 590         run_avg = (run_avg*9 + delta)/10;
 591
 592         printf("#%-3ld: %0.3f, ",
 593                 nr_runs, (double)delta/1000000.0);
 594
 595         printf("ravg: %0.2f, ",
 596                 (double)run_avg/1e6);
 597
 598         printf("cpu: %0.2f / %0.2f",
 599                 (double)cpu_usage/1e6, (double)runavg_cpu_usage/1e6);
 600
 601 #if 0
 602         /*
 603          * rusage statistics done by the parent, these are less
 604          * accurate than the sum_exec_runtime based statistics:
 605          */
 606         printf(" [%0.2f / %0.2f]",
 607                 (double)parent_cpu_usage/1e6,
 608                 (double)runavg_parent_cpu_usage/1e6);
 609 #endif
 610
 611         printf("\n");
 612
 613         if (nr_sleep_corrections)
 614                 printf(" (%ld sleep corrections)\n", nr_sleep_corrections);
 615         nr_sleep_corrections = 0;
 616 }
 617
 618 static void test_calibrations(void)
 619 {
 620         u64 T0, T1;
 621
 622         T0 = get_nsecs();
 623         burn_nsecs(1e6);
 624         T1 = get_nsecs();
 625
 626         printf("the run test took %Ld nsecs\n", T1-T0);
 627
 628         T0 = get_nsecs();
 629         sleep_nsecs(1e6);
 630         T1 = get_nsecs();
 631
 632         printf("the sleep test took %Ld nsecs\n", T1-T0);
 633 }
 634
 635 static int
 636 process_comm_event(event_t *event, unsigned long offset, unsigned long head)
 637 {
 638         struct thread *thread;
 639
 640         thread = threads__findnew(event->comm.pid, &threads, &last_match);
 641
 642         dump_printf("%p [%p]: perf_event_comm: %s:%d\n",
 643                 (void *)(offset + head),
 644                 (void *)(long)(event->header.size),
 645                 event->comm.comm, event->comm.pid);
 646
 647         if (thread == NULL ||
 648             thread__set_comm(thread, event->comm.comm)) {
 649                 dump_printf("problem processing perf_event_comm, skipping event.\n");
 650                 return -1;
 651         }
 652         total_comm++;
 653
 654         return 0;
 655 }
 656
 657
 658 struct raw_event_sample {
 659         u32 size;
 660         char data[0];
 661 };
 662
 663 #define FILL_FIELD(ptr, field, event, data)     \
 664         ptr.field = (typeof(ptr.field)) raw_field_value(event, #field, data)
 665
 666 #define FILL_ARRAY(ptr, array, event, data)                     \
 667 do {                                                            \
 668         void *__array = raw_field_ptr(event, #array, data);     \
 669         memcpy(ptr.array, __array, sizeof(ptr.array));  \
 670 } while(0)
 671
 672 #define FILL_COMMON_FIELDS(ptr, event, data)                    \
 673 do {                                                            \
 674         FILL_FIELD(ptr, common_type, event, data);              \
 675         FILL_FIELD(ptr, common_flags, event, data);             \
 676         FILL_FIELD(ptr, common_preempt_count, event, data);     \
 677         FILL_FIELD(ptr, common_pid, event, data);               \
 678         FILL_FIELD(ptr, common_tgid, event, data);              \
 679 } while (0)
 680
 681
 682
 683 struct trace_switch_event {
 684         u32 size;
 685
 686         u16 common_type;
 687         u8 common_flags;
 688         u8 common_preempt_count;
 689         u32 common_pid;
 690         u32 common_tgid;
 691
 692         char prev_comm[16];
 693         u32 prev_pid;
 694         u32 prev_prio;
 695         u64 prev_state;
 696         char next_comm[16];
 697         u32 next_pid;
 698         u32 next_prio;
 699 };
 700
 701 struct trace_runtime_event {
 702         u32 size;
 703
 704         u16 common_type;
 705         u8 common_flags;
 706         u8 common_preempt_count;
 707         u32 common_pid;
 708         u32 common_tgid;
 709
 710         char comm[16];
 711         u32 pid;
 712         u64 runtime;
 713         u64 vruntime;
 714 };
 715
 716 struct trace_wakeup_event {
 717         u32 size;
 718
 719         u16 common_type;
 720         u8 common_flags;
 721         u8 common_preempt_count;
 722         u32 common_pid;
 723         u32 common_tgid;
 724
 725         char comm[16];
 726         u32 pid;
 727
 728         u32 prio;
 729         u32 success;
 730         u32 cpu;
 731 };
 732
 733 struct trace_fork_event {
 734         u32 size;
 735
 736         u16 common_type;
 737         u8 common_flags;
 738         u8 common_preempt_count;
 739         u32 common_pid;
 740         u32 common_tgid;
 741
 742         char parent_comm[16];
 743         u32 parent_pid;
 744         char child_comm[16];
 745         u32 child_pid;
 746 };
 747
 748 struct trace_sched_handler {
 749         void (*switch_event)(struct trace_switch_event *,
 750                              struct event *,
 751                              int cpu,
 752                              u64 timestamp,
 753                              struct thread *thread);
 754
 755         void (*runtime_event)(struct trace_runtime_event *,
 756                               struct event *,
 757                               int cpu,
 758                               u64 timestamp,
 759                               struct thread *thread);
 760
 761         void (*wakeup_event)(struct trace_wakeup_event *,
 762                              struct event *,
 763                              int cpu,
 764                              u64 timestamp,
 765                              struct thread *thread);
 766
 767         void (*fork_event)(struct trace_fork_event *,
 768                            struct event *,
 769                            int cpu,
 770                            u64 timestamp,
 771                            struct thread *thread);
 772 };
 773
 774
 775 static void
 776 replay_wakeup_event(struct trace_wakeup_event *wakeup_event,
 777                     struct event *event,
 778                     int cpu __used,
 779                     u64 timestamp __used,
 780                     struct thread *thread __used)
 781 {
 782         struct task_desc *waker, *wakee;
 783
 784         if (verbose) {
 785                 printf("sched_wakeup event %p\n", event);
 786
 787                 printf(" ... pid %d woke up %s/%d\n",
 788                         wakeup_event->common_pid,
 789                         wakeup_event->comm,
 790                         wakeup_event->pid);
 791         }
 792
 793         waker = register_pid(wakeup_event->common_pid, "<unknown>");
 794         wakee = register_pid(wakeup_event->pid, wakeup_event->comm);
 795
 796         add_sched_event_wakeup(waker, timestamp, wakee);
 797 }
 798
 799 static u64 cpu_last_switched[MAX_CPUS];
 800
 801 static void
 802 replay_switch_event(struct trace_switch_event *switch_event,
 803                     struct event *event,
 804                     int cpu,
 805                     u64 timestamp,
 806                     struct thread *thread __used)
 807 {
 808         struct task_desc *prev, *next;
 809         u64 timestamp0;
 810         s64 delta;
 811
 812         if (verbose)
 813                 printf("sched_switch event %p\n", event);
 814
 815         if (cpu >= MAX_CPUS || cpu < 0)
 816                 return;
 817
 818         timestamp0 = cpu_last_switched[cpu];
 819         if (timestamp0)
 820                 delta = timestamp - timestamp0;
 821         else
 822                 delta = 0;
 823
 824         if (delta < 0)
 825                 die("hm, delta: %Ld < 0 ?\n", delta);
 826
 827         if (verbose) {
 828                 printf(" ... switch from %s/%d to %s/%d [ran %Ld nsecs]\n",
 829                         switch_event->prev_comm, switch_event->prev_pid,
 830                         switch_event->next_comm, switch_event->next_pid,
 831                         delta);
 832         }
 833
 834         prev = register_pid(switch_event->prev_pid, switch_event->prev_comm);
 835         next = register_pid(switch_event->next_pid, switch_event->next_comm);
 836
 837         cpu_last_switched[cpu] = timestamp;
 838
 839         add_sched_event_run(prev, timestamp, delta);
 840         add_sched_event_sleep(prev, timestamp, switch_event->prev_state);
 841 }
 842
 843
 844 static void
 845 replay_fork_event(struct trace_fork_event *fork_event,
 846                   struct event *event,
 847                   int cpu __used,
 848                   u64 timestamp __used,
 849                   struct thread *thread __used)
 850 {
 851         if (verbose) {
 852                 printf("sched_fork event %p\n", event);
 853                 printf("... parent: %s/%d\n", fork_event->parent_comm, fork_event->parent_pid);
 854                 printf("...  child: %s/%d\n", fork_event->child_comm, fork_event->child_pid);
 855         }
 856         register_pid(fork_event->parent_pid, fork_event->parent_comm);
 857         register_pid(fork_event->child_pid, fork_event->child_comm);
 858 }
 859
 860 static struct trace_sched_handler replay_ops  = {
 861         .wakeup_event           = replay_wakeup_event,
 862         .switch_event           = replay_switch_event,
 863         .fork_event             = replay_fork_event,
 864 };
 865
 866 struct sort_dimension {
 867         const char              *name;
 868         sort_fn_t               cmp;
 869         struct list_head        list;
 870 };
 871
 872 static LIST_HEAD(cmp_pid);
 873
 874 static int
 875 thread_lat_cmp(struct list_head *list, struct work_atoms *l, struct work_atoms *r)
 876 {
 877         struct sort_dimension *sort;
 878         int ret = 0;
 879
 880         BUG_ON(list_empty(list));
 881
 882         list_for_each_entry(sort, list, list) {
 883                 ret = sort->cmp(l, r);
 884                 if (ret)
 885                         return ret;
 886         }
 887
 888         return ret;
 889 }
 890
 891 static struct work_atoms *
 892 thread_atoms_search(struct rb_root *root, struct thread *thread,
 893                          struct list_head *sort_list)
 894 {
 895         struct rb_node *node = root->rb_node;
 896         struct work_atoms key = { .thread = thread };
 897
 898         while (node) {
 899                 struct work_atoms *atoms;
 900                 int cmp;
 901
 902                 atoms = container_of(node, struct work_atoms, node);
 903
 904                 cmp = thread_lat_cmp(sort_list, &key, atoms);
 905                 if (cmp > 0)
 906                         node = node->rb_left;
 907                 else if (cmp < 0)
 908                         node = node->rb_right;
 909                 else {
 910                         BUG_ON(thread != atoms->thread);
 911                         return atoms;
 912                 }
 913         }
 914         return NULL;
 915 }
 916
 917 static void
 918 __thread_latency_insert(struct rb_root *root, struct work_atoms *data,
 919                          struct list_head *sort_list)
 920 {
 921         struct rb_node **new = &(root->rb_node), *parent = NULL;
 922
 923         while (*new) {
 924                 struct work_atoms *this;
 925                 int cmp;
 926
 927                 this = container_of(*new, struct work_atoms, node);
 928                 parent = *new;
 929
 930                 cmp = thread_lat_cmp(sort_list, data, this);
 931
 932                 if (cmp > 0)
 933                         new = &((*new)->rb_left);
 934                 else
 935                         new = &((*new)->rb_right);
 936         }
 937
 938         rb_link_node(&data->node, parent, new);
 939         rb_insert_color(&data->node, root);
 940 }
 941
 942 static void thread_atoms_insert(struct thread *thread)
 943 {
 944         struct work_atoms *atoms;
 945
 946         atoms = calloc(sizeof(*atoms), 1);
 947         if (!atoms)
 948                 die("No memory");
 949
 950         atoms->thread = thread;
 951         INIT_LIST_HEAD(&atoms->work_list);
 952         __thread_latency_insert(&atom_root, atoms, &cmp_pid);
 953 }
 954
 955 static void
 956 latency_fork_event(struct trace_fork_event *fork_event __used,
 957                    struct event *event __used,
 958                    int cpu __used,
 959                    u64 timestamp __used,
 960                    struct thread *thread __used)
 961 {
 962         /* should insert the newcomer */
 963 }
 964
 965 __used
 966 static char sched_out_state(struct trace_switch_event *switch_event)
 967 {
 968         const char *str = TASK_STATE_TO_CHAR_STR;
 969
 970         return str[switch_event->prev_state];
 971 }
 972
 973 static void
 974 add_sched_out_event(struct work_atoms *atoms,
 975                     char run_state,
 976                     u64 timestamp)
 977 {
 978         struct work_atom *atom;
 979
 980         atom = calloc(sizeof(*atom), 1);
 981         if (!atom)
 982                 die("Non memory");
 983
 984         atom->sched_out_time = timestamp;
 985
 986         if (run_state == 'R') {
 987                 atom->state = THREAD_WAIT_CPU;
 988                 atom->wake_up_time = atom->sched_out_time;
 989         }
 990
 991         list_add_tail(&atom->list, &atoms->work_list);
 992 }
 993
 994 static void
 995 add_runtime_event(struct work_atoms *atoms, u64 delta, u64 timestamp __used)
 996 {
 997         struct work_atom *atom;
 998
 999         BUG_ON(list_empty(&atoms->work_list));
1000
1001         atom = list_entry(atoms->work_list.prev, struct work_atom, list);
1002
1003         atom->runtime += delta;
1004         atoms->total_runtime += delta;
1005 }
1006
1007 static void
1008 add_sched_in_event(struct work_atoms *atoms, u64 timestamp)
1009 {
1010         struct work_atom *atom;
1011         u64 delta;
1012
1013         if (list_empty(&atoms->work_list))
1014                 return;
1015
1016         atom = list_entry(atoms->work_list.prev, struct work_atom, list);
1017
1018         if (atom->state != THREAD_WAIT_CPU)
1019                 return;
1020
1021         if (timestamp < atom->wake_up_time) {
1022                 atom->state = THREAD_IGNORE;
1023                 return;
1024         }
1025
1026         atom->state = THREAD_SCHED_IN;
1027         atom->sched_in_time = timestamp;
1028
1029         delta = atom->sched_in_time - atom->wake_up_time;
1030         atoms->total_lat += delta;
1031         if (delta > atoms->max_lat)
1032                 atoms->max_lat = delta;
1033         atoms->nb_atoms++;
1034 }
1035
1036 static void
1037 latency_switch_event(struct trace_switch_event *switch_event,
1038                      struct event *event __used,
1039                      int cpu,
1040                      u64 timestamp,
1041                      struct thread *thread __used)
1042 {
1043         struct work_atoms *out_events, *in_events;
1044         struct thread *sched_out, *sched_in;
1045         u64 timestamp0;
1046         s64 delta;
1047
1048         BUG_ON(cpu >= MAX_CPUS || cpu < 0);
1049
1050         timestamp0 = cpu_last_switched[cpu];
1051         cpu_last_switched[cpu] = timestamp;
1052         if (timestamp0)
1053                 delta = timestamp - timestamp0;
1054         else
1055                 delta = 0;
1056
1057         if (delta < 0)
1058                 die("hm, delta: %Ld < 0 ?\n", delta);
1059
1060
1061         sched_out = threads__findnew(switch_event->prev_pid, &threads, &last_match);
1062         sched_in = threads__findnew(switch_event->next_pid, &threads, &last_match);
1063
1064         out_events = thread_atoms_search(&atom_root, sched_out, &cmp_pid);
1065         if (!out_events) {
1066                 thread_atoms_insert(sched_out);
1067                 out_events = thread_atoms_search(&atom_root, sched_out, &cmp_pid);
1068                 if (!out_events)
1069                         die("out-event: Internal tree error");
1070         }
1071         add_sched_out_event(out_events, sched_out_state(switch_event), timestamp);
1072
1073         in_events = thread_atoms_search(&atom_root, sched_in, &cmp_pid);
1074         if (!in_events) {
1075                 thread_atoms_insert(sched_in);
1076                 in_events = thread_atoms_search(&atom_root, sched_in, &cmp_pid);
1077                 if (!in_events)
1078                         die("in-event: Internal tree error");
1079                 /*
1080                  * Take came in we have not heard about yet,
1081                  * add in an initial atom in runnable state:
1082                  */
1083                 add_sched_out_event(in_events, 'R', timestamp);
1084         }
1085         add_sched_in_event(in_events, timestamp);
1086 }
1087
1088 static void
1089 latency_runtime_event(struct trace_runtime_event *runtime_event,
1090                      struct event *event __used,
1091                      int cpu,
1092                      u64 timestamp,
1093                      struct thread *this_thread __used)
1094 {
1095         struct work_atoms *atoms;
1096         struct thread *thread;
1097
1098         BUG_ON(cpu >= MAX_CPUS || cpu < 0);
1099
1100         thread = threads__findnew(runtime_event->pid, &threads, &last_match);
1101         atoms = thread_atoms_search(&atom_root, thread, &cmp_pid);
1102         if (!atoms) {
1103                 thread_atoms_insert(thread);
1104                 atoms = thread_atoms_search(&atom_root, thread, &cmp_pid);
1105                 if (!atoms)
1106                         die("in-event: Internal tree error");
1107                 add_sched_out_event(atoms, 'R', timestamp);
1108         }
1109
1110         add_runtime_event(atoms, runtime_event->runtime, timestamp);
1111 }
1112
1113 static void
1114 latency_wakeup_event(struct trace_wakeup_event *wakeup_event,
1115                      struct event *__event __used,
1116                      int cpu __used,
1117                      u64 timestamp,
1118                      struct thread *thread __used)
1119 {
1120         struct work_atoms *atoms;
1121         struct work_atom *atom;
1122         struct thread *wakee;
1123
1124         /* Note for later, it may be interesting to observe the failing cases */
1125         if (!wakeup_event->success)
1126                 return;
1127
1128         wakee = threads__findnew(wakeup_event->pid, &threads, &last_match);
1129         atoms = thread_atoms_search(&atom_root, wakee, &cmp_pid);
1130         if (!atoms) {
1131                 thread_atoms_insert(wakee);
1132                 atoms = thread_atoms_search(&atom_root, wakee, &cmp_pid);
1133                 if (!atoms)
1134                         die("wakeup-event: Internal tree error");
1135                 add_sched_out_event(atoms, 'S', timestamp);
1136         }
1137
1138         BUG_ON(list_empty(&atoms->work_list));
1139
1140         atom = list_entry(atoms->work_list.prev, struct work_atom, list);
1141
1142         if (atom->state != THREAD_SLEEPING)
1143                 nr_state_machine_bugs++;
1144
1145         nr_timestamps++;
1146         if (atom->sched_out_time > timestamp) {
1147                 nr_unordered_timestamps++;
1148                 return;
1149         }
1150
1151         atom->state = THREAD_WAIT_CPU;
1152         atom->wake_up_time = timestamp;
1153 }
1154
1155 static struct trace_sched_handler lat_ops  = {
1156         .wakeup_event           = latency_wakeup_event,
1157         .switch_event           = latency_switch_event,
1158         .runtime_event          = latency_runtime_event,
1159         .fork_event             = latency_fork_event,
1160 };
1161
1162 static void output_lat_thread(struct work_atoms *work_list)
1163 {
1164         int i;
1165         int ret;
1166         u64 avg;
1167
1168         if (!work_list->nb_atoms)
1169                 return;
1170         /*
1171          * Ignore idle threads:
1172          */
1173         if (!strcmp(work_list->thread->comm, "swapper"))
1174                 return;
1175
1176         all_runtime += work_list->total_runtime;
1177         all_count += work_list->nb_atoms;
1178
1179         ret = printf("  %s:%d ", work_list->thread->comm, work_list->thread->pid);
1180
1181         for (i = 0; i < 24 - ret; i++)
1182                 printf(" ");
1183
1184         avg = work_list->total_lat / work_list->nb_atoms;
1185
1186         printf("|%11.3f ms |%9llu | avg:%9.3f ms | max:%9.3f ms |\n",
1187               (double)work_list->total_runtime / 1e6,
1188                  work_list->nb_atoms, (double)avg / 1e6,
1189                  (double)work_list->max_lat / 1e6);
1190 }
1191
1192 static int pid_cmp(struct work_atoms *l, struct work_atoms *r)
1193 {
1194         if (l->thread->pid < r->thread->pid)
1195                 return -1;
1196         if (l->thread->pid > r->thread->pid)
1197                 return 1;
1198
1199         return 0;
1200 }
1201
1202 static struct sort_dimension pid_sort_dimension = {
1203         .name                   = "pid",
1204         .cmp                    = pid_cmp,
1205 };
1206
1207 static int avg_cmp(struct work_atoms *l, struct work_atoms *r)
1208 {
1209         u64 avgl, avgr;
1210
1211         if (!l->nb_atoms)
1212                 return -1;
1213
1214         if (!r->nb_atoms)
1215                 return 1;
1216
1217         avgl = l->total_lat / l->nb_atoms;
1218         avgr = r->total_lat / r->nb_atoms;
1219
1220         if (avgl < avgr)
1221                 return -1;
1222         if (avgl > avgr)
1223                 return 1;
1224
1225         return 0;
1226 }
1227
1228 static struct sort_dimension avg_sort_dimension = {
1229         .name                   = "avg",
1230         .cmp                    = avg_cmp,
1231 };
1232
1233 static int max_cmp(struct work_atoms *l, struct work_atoms *r)
1234 {
1235         if (l->max_lat < r->max_lat)
1236                 return -1;
1237         if (l->max_lat > r->max_lat)
1238                 return 1;
1239
1240         return 0;
1241 }
1242
1243 static struct sort_dimension max_sort_dimension = {
1244         .name                   = "max",
1245         .cmp                    = max_cmp,
1246 };
1247
1248 static int switch_cmp(struct work_atoms *l, struct work_atoms *r)
1249 {
1250         if (l->nb_atoms < r->nb_atoms)
1251                 return -1;
1252         if (l->nb_atoms > r->nb_atoms)
1253                 return 1;
1254
1255         return 0;
1256 }
1257
1258 static struct sort_dimension switch_sort_dimension = {
1259         .name                   = "switch",
1260         .cmp                    = switch_cmp,
1261 };
1262
1263 static int runtime_cmp(struct work_atoms *l, struct work_atoms *r)
1264 {
1265         if (l->total_runtime < r->total_runtime)
1266                 return -1;
1267         if (l->total_runtime > r->total_runtime)
1268                 return 1;
1269
1270         return 0;
1271 }
1272
1273 static struct sort_dimension runtime_sort_dimension = {
1274         .name                   = "runtime",
1275         .cmp                    = runtime_cmp,
1276 };
1277
1278 static struct sort_dimension *available_sorts[] = {
1279         &pid_sort_dimension,
1280         &avg_sort_dimension,
1281         &max_sort_dimension,
1282         &switch_sort_dimension,
1283         &runtime_sort_dimension,
1284 };
1285
1286 #define NB_AVAILABLE_SORTS      (int)(sizeof(available_sorts) / sizeof(struct sort_dimension *))
1287
1288 static LIST_HEAD(sort_list);
1289
1290 static int sort_dimension__add(char *tok, struct list_head *list)
1291 {
1292         int i;
1293
1294         for (i = 0; i < NB_AVAILABLE_SORTS; i++) {
1295                 if (!strcmp(available_sorts[i]->name, tok)) {
1296                         list_add_tail(&available_sorts[i]->list, list);
1297
1298                         return 0;
1299                 }
1300         }
1301
1302         return -1;
1303 }
1304
1305 static void setup_sorting(void);
1306
1307 static void sort_lat(void)
1308 {
1309         struct rb_node *node;
1310
1311         for (;;) {
1312                 struct work_atoms *data;
1313                 node = rb_first(&atom_root);
1314                 if (!node)
1315                         break;
1316
1317                 rb_erase(node, &atom_root);
1318                 data = rb_entry(node, struct work_atoms, node);
1319                 __thread_latency_insert(&sorted_atom_root, data, &sort_list);
1320         }
1321 }
1322
1323 static struct trace_sched_handler *trace_handler;
1324
1325 static void
1326 process_sched_wakeup_event(struct raw_event_sample *raw,
1327                            struct event *event,
1328                            int cpu __used,
1329                            u64 timestamp __used,
1330                            struct thread *thread __used)
1331 {
1332         struct trace_wakeup_event wakeup_event;
1333
1334         FILL_COMMON_FIELDS(wakeup_event, event, raw->data);
1335
1336         FILL_ARRAY(wakeup_event, comm, event, raw->data);
1337         FILL_FIELD(wakeup_event, pid, event, raw->data);
1338         FILL_FIELD(wakeup_event, prio, event, raw->data);
1339         FILL_FIELD(wakeup_event, success, event, raw->data);
1340         FILL_FIELD(wakeup_event, cpu, event, raw->data);
1341
1342         if (trace_handler->wakeup_event)
1343                 trace_handler->wakeup_event(&wakeup_event, event, cpu, timestamp, thread);
1344 }
1345
1346 /*
1347  * Track the current task - that way we can know whether there's any
1348  * weird events, such as a task being switched away that is not current.
1349  */
1350 static int max_cpu;
1351
1352 static u32 curr_pid[MAX_CPUS] = { [0 ... MAX_CPUS-1] = -1 };
1353
1354 static struct thread *curr_thread[MAX_CPUS];
1355
1356 static char next_shortname1 = 'A';
1357 static char next_shortname2 = '0';
1358
1359 static void
1360 map_switch_event(struct trace_switch_event *switch_event,
1361                  struct event *event __used,
1362                  int this_cpu,
1363                  u64 timestamp,
1364                  struct thread *thread __used)
1365 {
1366         struct thread *sched_out, *sched_in;
1367         int new_shortname;
1368         u64 timestamp0;
1369         s64 delta;
1370         int cpu;
1371
1372         BUG_ON(this_cpu >= MAX_CPUS || this_cpu < 0);
1373
1374         if (this_cpu > max_cpu)
1375                 max_cpu = this_cpu;
1376
1377         timestamp0 = cpu_last_switched[this_cpu];
1378         cpu_last_switched[this_cpu] = timestamp;
1379         if (timestamp0)
1380                 delta = timestamp - timestamp0;
1381         else
1382                 delta = 0;
1383
1384         if (delta < 0)
1385                 die("hm, delta: %Ld < 0 ?\n", delta);
1386
1387
1388         sched_out = threads__findnew(switch_event->prev_pid, &threads, &last_match);
1389         sched_in = threads__findnew(switch_event->next_pid, &threads, &last_match);
1390
1391         curr_thread[this_cpu] = sched_in;
1392
1393         printf("  ");
1394
1395         new_shortname = 0;
1396         if (!sched_in->shortname[0]) {
1397                 sched_in->shortname[0] = next_shortname1;
1398                 sched_in->shortname[1] = next_shortname2;
1399
1400                 if (next_shortname1 < 'Z') {
1401                         next_shortname1++;
1402                 } else {
1403                         next_shortname1='A';
1404                         if (next_shortname2 < '9') {
1405                                 next_shortname2++;
1406                         } else {
1407                                 next_shortname2='0';
1408                         }
1409                 }
1410                 new_shortname = 1;
1411         }
1412
1413         for (cpu = 0; cpu <= max_cpu; cpu++) {
1414                 if (cpu != this_cpu)
1415                         printf(" ");
1416                 else
1417                         printf("*");
1418
1419                 if (curr_thread[cpu]) {
1420                         if (curr_thread[cpu]->pid)
1421                                 printf("%2s ", curr_thread[cpu]->shortname);
1422                         else
1423                                 printf(".  ");
1424                 } else
1425                         printf("   ");
1426         }
1427
1428         printf("  %12.6f secs ", (double)timestamp/1e9);
1429         if (new_shortname) {
1430                 printf("%s => %s:%d\n",
1431                         sched_in->shortname, sched_in->comm, sched_in->pid);
1432         } else {
1433                 printf("\n");
1434         }
1435 }
1436
1437
1438 static void
1439 process_sched_switch_event(struct raw_event_sample *raw,
1440                            struct event *event,
1441                            int this_cpu,
1442                            u64 timestamp __used,
1443                            struct thread *thread __used)
1444 {
1445         struct trace_switch_event switch_event;
1446
1447         FILL_COMMON_FIELDS(switch_event, event, raw->data);
1448
1449         FILL_ARRAY(switch_event, prev_comm, event, raw->data);
1450         FILL_FIELD(switch_event, prev_pid, event, raw->data);
1451         FILL_FIELD(switch_event, prev_prio, event, raw->data);
1452         FILL_FIELD(switch_event, prev_state, event, raw->data);
1453         FILL_ARRAY(switch_event, next_comm, event, raw->data);
1454         FILL_FIELD(switch_event, next_pid, event, raw->data);
1455         FILL_FIELD(switch_event, next_prio, event, raw->data);
1456
1457         if (curr_pid[this_cpu] != (u32)-1) {
1458                 /*
1459                  * Are we trying to switch away a PID that is
1460                  * not current?
1461                  */
1462                 if (curr_pid[this_cpu] != switch_event.prev_pid)
1463                         nr_context_switch_bugs++;
1464         }
1465         if (trace_handler->switch_event)
1466                 trace_handler->switch_event(&switch_event, event, this_cpu, timestamp, thread);
1467
1468         curr_pid[this_cpu] = switch_event.next_pid;
1469 }
1470
1471 static void
1472 process_sched_runtime_event(struct raw_event_sample *raw,
1473                            struct event *event,
1474                            int cpu __used,
1475                            u64 timestamp __used,
1476                            struct thread *thread __used)
1477 {
1478         struct trace_runtime_event runtime_event;
1479
1480         FILL_ARRAY(runtime_event, comm, event, raw->data);
1481         FILL_FIELD(runtime_event, pid, event, raw->data);
1482         FILL_FIELD(runtime_event, runtime, event, raw->data);
1483         FILL_FIELD(runtime_event, vruntime, event, raw->data);
1484
1485         if (trace_handler->runtime_event)
1486                 trace_handler->runtime_event(&runtime_event, event, cpu, timestamp, thread);
1487 }
1488
1489 static void
1490 process_sched_fork_event(struct raw_event_sample *raw,
1491                          struct event *event,
1492                          int cpu __used,
1493                          u64 timestamp __used,
1494                          struct thread *thread __used)
1495 {
1496         struct trace_fork_event fork_event;
1497
1498         FILL_COMMON_FIELDS(fork_event, event, raw->data);
1499
1500         FILL_ARRAY(fork_event, parent_comm, event, raw->data);
1501         FILL_FIELD(fork_event, parent_pid, event, raw->data);
1502         FILL_ARRAY(fork_event, child_comm, event, raw->data);
1503         FILL_FIELD(fork_event, child_pid, event, raw->data);
1504
1505         if (trace_handler->fork_event)
1506                 trace_handler->fork_event(&fork_event, event, cpu, timestamp, thread);
1507 }
1508
1509 static void
1510 process_sched_exit_event(struct event *event,
1511                          int cpu __used,
1512                          u64 timestamp __used,
1513                          struct thread *thread __used)
1514 {
1515         if (verbose)
1516                 printf("sched_exit event %p\n", event);
1517 }
1518
1519 static void
1520 process_raw_event(event_t *raw_event __used, void *more_data,
1521                   int cpu, u64 timestamp, struct thread *thread)
1522 {
1523         struct raw_event_sample *raw = more_data;
1524         struct event *event;
1525         int type;
1526
1527         type = trace_parse_common_type(raw->data);
1528         event = trace_find_event(type);
1529
1530         if (!strcmp(event->name, "sched_switch"))
1531                 process_sched_switch_event(raw, event, cpu, timestamp, thread);
1532         if (!strcmp(event->name, "sched_stat_runtime"))
1533                 process_sched_runtime_event(raw, event, cpu, timestamp, thread);
1534         if (!strcmp(event->name, "sched_wakeup"))
1535                 process_sched_wakeup_event(raw, event, cpu, timestamp, thread);
1536         if (!strcmp(event->name, "sched_wakeup_new"))
1537                 process_sched_wakeup_event(raw, event, cpu, timestamp, thread);
1538         if (!strcmp(event->name, "sched_process_fork"))
1539                 process_sched_fork_event(raw, event, cpu, timestamp, thread);
1540         if (!strcmp(event->name, "sched_process_exit"))
1541                 process_sched_exit_event(event, cpu, timestamp, thread);
1542 }
1543
1544 static int
1545 process_sample_event(event_t *event, unsigned long offset, unsigned long head)
1546 {
1547         char level;
1548         int show = 0;
1549         struct dso *dso = NULL;
1550         struct thread *thread;
1551         u64 ip = event->ip.ip;
1552         u64 timestamp = -1;
1553         u32 cpu = -1;
1554         u64 period = 1;
1555         void *more_data = event->ip.__more_data;
1556         int cpumode;
1557
1558         thread = threads__findnew(event->ip.pid, &threads, &last_match);
1559
1560         if (sample_type & PERF_SAMPLE_TIME) {
1561                 timestamp = *(u64 *)more_data;
1562                 more_data += sizeof(u64);
1563         }
1564
1565         if (sample_type & PERF_SAMPLE_CPU) {
1566                 cpu = *(u32 *)more_data;
1567                 more_data += sizeof(u32);
1568                 more_data += sizeof(u32); /* reserved */
1569         }
1570
1571         if (sample_type & PERF_SAMPLE_PERIOD) {
1572                 period = *(u64 *)more_data;
1573                 more_data += sizeof(u64);
1574         }
1575
1576         dump_printf("%p [%p]: PERF_EVENT_SAMPLE (IP, %d): %d/%d: %p period: %Ld\n",
1577                 (void *)(offset + head),
1578                 (void *)(long)(event->header.size),
1579                 event->header.misc,
1580                 event->ip.pid, event->ip.tid,
1581                 (void *)(long)ip,
1582                 (long long)period);
1583
1584         dump_printf(" ... thread: %s:%d\n", thread->comm, thread->pid);
1585
1586         if (thread == NULL) {
1587                 eprintf("problem processing %d event, skipping it.\n",
1588                         event->header.type);
1589                 return -1;
1590         }
1591
1592         cpumode = event->header.misc & PERF_EVENT_MISC_CPUMODE_MASK;
1593
1594         if (cpumode == PERF_EVENT_MISC_KERNEL) {
1595                 show = SHOW_KERNEL;
1596                 level = 'k';
1597
1598                 dso = kernel_dso;
1599
1600                 dump_printf(" ...... dso: %s\n", dso->name);
1601
1602         } else if (cpumode == PERF_EVENT_MISC_USER) {
1603
1604                 show = SHOW_USER;
1605                 level = '.';
1606
1607         } else {
1608                 show = SHOW_HV;
1609                 level = 'H';
1610
1611                 dso = hypervisor_dso;
1612
1613                 dump_printf(" ...... dso: [hypervisor]\n");
1614         }
1615
1616         if (sample_type & PERF_SAMPLE_RAW)
1617                 process_raw_event(event, more_data, cpu, timestamp, thread);
1618
1619         return 0;
1620 }
1621
1622 static int
1623 process_event(event_t *event, unsigned long offset, unsigned long head)
1624 {
1625         trace_event(event);
1626
1627         nr_events++;
1628         switch (event->header.type) {
1629         case PERF_EVENT_MMAP:
1630                 return 0;
1631         case PERF_EVENT_LOST:
1632                 nr_lost_chunks++;
1633                 nr_lost_events += event->lost.lost;
1634                 return 0;
1635
1636         case PERF_EVENT_COMM:
1637                 return process_comm_event(event, offset, head);
1638
1639         case PERF_EVENT_EXIT ... PERF_EVENT_READ:
1640                 return 0;
1641
1642         case PERF_EVENT_SAMPLE:
1643                 return process_sample_event(event, offset, head);
1644
1645         case PERF_EVENT_MAX:
1646         default:
1647                 return -1;
1648         }
1649
1650         return 0;
1651 }
1652
1653 static int read_events(void)
1654 {
1655         int ret, rc = EXIT_FAILURE;
1656         unsigned long offset = 0;
1657         unsigned long head = 0;
1658         struct stat perf_stat;
1659         event_t *event;
1660         uint32_t size;
1661         char *buf;
1662
1663         trace_report();
1664         register_idle_thread(&threads, &last_match);
1665
1666         input = open(input_name, O_RDONLY);
1667         if (input < 0) {
1668                 perror("failed to open file");
1669                 exit(-1);
1670         }
1671
1672         ret = fstat(input, &perf_stat);
1673         if (ret < 0) {
1674                 perror("failed to stat file");
1675                 exit(-1);
1676         }
1677
1678         if (!perf_stat.st_size) {
1679                 fprintf(stderr, "zero-sized file, nothing to do!\n");
1680                 exit(0);
1681         }
1682         header = perf_header__read(input);
1683         head = header->data_offset;
1684         sample_type = perf_header__sample_type(header);
1685
1686         if (!(sample_type & PERF_SAMPLE_RAW))
1687                 die("No trace sample to read. Did you call perf record "
1688                     "without -R?");
1689
1690         if (load_kernel() < 0) {
1691                 perror("failed to load kernel symbols");
1692                 return EXIT_FAILURE;
1693         }
1694
1695 remap:
1696         buf = (char *)mmap(NULL, page_size * mmap_window, PROT_READ,
1697                            MAP_SHARED, input, offset);
1698         if (buf == MAP_FAILED) {
1699                 perror("failed to mmap file");
1700                 exit(-1);
1701         }
1702
1703 more:
1704         event = (event_t *)(buf + head);
1705
1706         size = event->header.size;
1707         if (!size)
1708                 size = 8;
1709
1710         if (head + event->header.size >= page_size * mmap_window) {
1711                 unsigned long shift = page_size * (head / page_size);
1712                 int res;
1713
1714                 res = munmap(buf, page_size * mmap_window);
1715                 assert(res == 0);
1716
1717                 offset += shift;
1718                 head -= shift;
1719                 goto remap;
1720         }
1721
1722         size = event->header.size;
1723
1724
1725         if (!size || process_event(event, offset, head) < 0) {
1726
1727                 /*
1728                  * assume we lost track of the stream, check alignment, and
1729                  * increment a single u64 in the hope to catch on again 'soon'.
1730                  */
1731
1732                 if (unlikely(head & 7))
1733                         head &= ~7ULL;
1734
1735                 size = 8;
1736         }
1737
1738         head += size;
1739
1740         if (offset + head < (unsigned long)perf_stat.st_size)
1741                 goto more;
1742
1743         rc = EXIT_SUCCESS;
1744         close(input);
1745
1746         return rc;
1747 }
1748
1749 static void print_bad_events(void)
1750 {
1751         if (nr_unordered_timestamps && nr_timestamps) {
1752                 printf("  INFO: %.3f%% unordered timestamps (%ld out of %ld)\n",
1753                         (double)nr_unordered_timestamps/(double)nr_timestamps*100.0,
1754                         nr_unordered_timestamps, nr_timestamps);
1755         }
1756         if (nr_lost_events && nr_events) {
1757                 printf("  INFO: %.3f%% lost events (%ld out of %ld, in %ld chunks)\n",
1758                         (double)nr_lost_events/(double)nr_events*100.0,
1759                         nr_lost_events, nr_events, nr_lost_chunks);
1760         }
1761         if (nr_state_machine_bugs && nr_timestamps) {
1762                 printf("  INFO: %.3f%% state machine bugs (%ld out of %ld)",
1763                         (double)nr_state_machine_bugs/(double)nr_timestamps*100.0,
1764                         nr_state_machine_bugs, nr_timestamps);
1765                 if (nr_lost_events)
1766                         printf(" (due to lost events?)");
1767                 printf("\n");
1768         }
1769         if (nr_context_switch_bugs && nr_timestamps) {
1770                 printf("  INFO: %.3f%% context switch bugs (%ld out of %ld)",
1771                         (double)nr_context_switch_bugs/(double)nr_timestamps*100.0,
1772                         nr_context_switch_bugs, nr_timestamps);
1773                 if (nr_lost_events)
1774                         printf(" (due to lost events?)");
1775                 printf("\n");
1776         }
1777 }
1778
1779 static void __cmd_lat(void)
1780 {
1781         struct rb_node *next;
1782
1783         setup_pager();
1784         read_events();
1785         sort_lat();
1786
1787         printf("\n -----------------------------------------------------------------------------------------\n");
1788         printf("  Task                  |   Runtime ms  | Switches | Average delay ms | Maximum delay ms |\n");
1789         printf(" -----------------------------------------------------------------------------------------\n");
1790
1791         next = rb_first(&sorted_atom_root);
1792
1793         while (next) {
1794                 struct work_atoms *work_list;
1795
1796                 work_list = rb_entry(next, struct work_atoms, node);
1797                 output_lat_thread(work_list);
1798                 next = rb_next(next);
1799         }
1800
1801         printf(" -----------------------------------------------------------------------------------------\n");
1802         printf("  TOTAL:                |%11.3f ms |%9Ld |\n",
1803                 (double)all_runtime/1e6, all_count);
1804
1805         printf(" ---------------------------------------------------\n");
1806
1807         print_bad_events();
1808         printf("\n");
1809
1810 }
1811
1812 static struct trace_sched_handler map_ops  = {
1813         .wakeup_event           = NULL,
1814         .switch_event           = map_switch_event,
1815         .runtime_event          = NULL,
1816         .fork_event             = NULL,
1817 };
1818
1819 static void __cmd_map(void)
1820 {
1821         max_cpu = sysconf(_SC_NPROCESSORS_CONF);
1822
1823         setup_pager();
1824         read_events();
1825         print_bad_events();
1826 }
1827
1828 static void __cmd_replay(void)
1829 {
1830         unsigned long i;
1831
1832         calibrate_run_measurement_overhead();
1833         calibrate_sleep_measurement_overhead();
1834
1835         test_calibrations();
1836
1837         read_events();
1838
1839         printf("nr_run_events:        %ld\n", nr_run_events);
1840         printf("nr_sleep_events:      %ld\n", nr_sleep_events);
1841         printf("nr_wakeup_events:     %ld\n", nr_wakeup_events);
1842
1843         if (targetless_wakeups)
1844                 printf("target-less wakeups:  %ld\n", targetless_wakeups);
1845         if (multitarget_wakeups)
1846                 printf("multi-target wakeups: %ld\n", multitarget_wakeups);
1847         if (nr_run_events_optimized)
1848                 printf("run atoms optimized: %ld\n",
1849                         nr_run_events_optimized);
1850
1851         print_task_traces();
1852         add_cross_task_wakeups();
1853
1854         create_tasks();
1855         printf("------------------------------------------------------------\n");
1856         for (i = 0; i < replay_repeat; i++)
1857                 run_one_test();
1858 }
1859
1860
1861 static const char * const sched_usage[] = {
1862         "perf sched [<options>] {record|latency|replay|trace}",
1863         NULL
1864 };
1865
1866 static const struct option sched_options[] = {
1867         OPT_BOOLEAN('v', "verbose", &verbose,
1868                     "be more verbose (show symbol address, etc)"),
1869         OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
1870                     "dump raw trace in ASCII"),
1871         OPT_END()
1872 };
1873
1874 static const char * const latency_usage[] = {
1875         "perf sched latency [<options>]",
1876         NULL
1877 };
1878
1879 static const struct option latency_options[] = {
1880         OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1881                    "sort by key(s): runtime, switch, avg, max"),
1882         OPT_BOOLEAN('v', "verbose", &verbose,
1883                     "be more verbose (show symbol address, etc)"),
1884         OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
1885                     "dump raw trace in ASCII"),
1886         OPT_END()
1887 };
1888
1889 static const char * const replay_usage[] = {
1890         "perf sched replay [<options>]",
1891         NULL
1892 };
1893
1894 static const struct option replay_options[] = {
1895         OPT_INTEGER('r', "repeat", &replay_repeat,
1896                     "repeat the workload replay N times (-1: infinite)"),
1897         OPT_BOOLEAN('v', "verbose", &verbose,
1898                     "be more verbose (show symbol address, etc)"),
1899         OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
1900                     "dump raw trace in ASCII"),
1901         OPT_END()
1902 };
1903
1904 static void setup_sorting(void)
1905 {
1906         char *tmp, *tok, *str = strdup(sort_order);
1907
1908         for (tok = strtok_r(str, ", ", &tmp);
1909                         tok; tok = strtok_r(NULL, ", ", &tmp)) {
1910                 if (sort_dimension__add(tok, &sort_list) < 0) {
1911                         error("Unknown --sort key: `%s'", tok);
1912                         usage_with_options(latency_usage, latency_options);
1913                 }
1914         }
1915
1916         free(str);
1917
1918         sort_dimension__add((char *)"pid", &cmp_pid);
1919 }
1920
1921 static const char *record_args[] = {
1922         "record",
1923         "-a",
1924         "-R",
1925         "-M",
1926         "-f",
1927         "-m", "1024",
1928         "-c", "1",
1929         "-e", "sched:sched_switch:r",
1930         "-e", "sched:sched_stat_wait:r",
1931         "-e", "sched:sched_stat_sleep:r",
1932         "-e", "sched:sched_stat_iowait:r",
1933         "-e", "sched:sched_stat_runtime:r",
1934         "-e", "sched:sched_process_exit:r",
1935         "-e", "sched:sched_process_fork:r",
1936         "-e", "sched:sched_wakeup:r",
1937         "-e", "sched:sched_migrate_task:r",
1938 };
1939
1940 static int __cmd_record(int argc, const char **argv)
1941 {
1942         unsigned int rec_argc, i, j;
1943         const char **rec_argv;
1944
1945         rec_argc = ARRAY_SIZE(record_args) + argc - 1;
1946         rec_argv = calloc(rec_argc + 1, sizeof(char *));
1947
1948         for (i = 0; i < ARRAY_SIZE(record_args); i++)
1949                 rec_argv[i] = strdup(record_args[i]);
1950
1951         for (j = 1; j < (unsigned int)argc; j++, i++)
1952                 rec_argv[i] = argv[j];
1953
1954         BUG_ON(i != rec_argc);
1955
1956         return cmd_record(i, rec_argv, NULL);
1957 }
1958
1959 int cmd_sched(int argc, const char **argv, const char *prefix __used)
1960 {
1961         symbol__init();
1962         page_size = getpagesize();
1963
1964         argc = parse_options(argc, argv, sched_options, sched_usage,
1965                              PARSE_OPT_STOP_AT_NON_OPTION);
1966         if (!argc)
1967                 usage_with_options(sched_usage, sched_options);
1968
1969         if (!strncmp(argv[0], "rec", 3)) {
1970                 return __cmd_record(argc, argv);
1971         } else if (!strncmp(argv[0], "lat", 3)) {
1972                 trace_handler = &lat_ops;
1973                 if (argc > 1) {
1974                         argc = parse_options(argc, argv, latency_options, latency_usage, 0);
1975                         if (argc)
1976                                 usage_with_options(latency_usage, latency_options);
1977                 }
1978                 setup_sorting();
1979                 __cmd_lat();
1980         } else if (!strcmp(argv[0], "map")) {
1981                 trace_handler = &map_ops;
1982                 setup_sorting();
1983                 __cmd_map();
1984         } else if (!strncmp(argv[0], "rep", 3)) {
1985                 trace_handler = &replay_ops;
1986                 if (argc) {
1987                         argc = parse_options(argc, argv, replay_options, replay_usage, 0);
1988                         if (argc)
1989                                 usage_with_options(replay_usage, replay_options);
1990                 }
1991                 __cmd_replay();
1992         } else if (!strcmp(argv[0], "trace")) {
1993                 /*
1994                  * Aliased to 'perf trace' for now:
1995                  */
1996                 return cmd_trace(argc, argv, prefix);
1997         } else {
1998                 usage_with_options(sched_usage, sched_options);
1999         }
2000
2001         return 0;
2002 }