perf tools: Add x86 RDPMC, RDTSC test
[~shefty/rdma-dev.git] / tools / perf / builtin-sched.c
1 #include "builtin.h"
2 #include "perf.h"
3
4 #include "util/util.h"
5 #include "util/evlist.h"
6 #include "util/cache.h"
7 #include "util/evsel.h"
8 #include "util/symbol.h"
9 #include "util/thread.h"
10 #include "util/header.h"
11 #include "util/session.h"
12 #include "util/tool.h"
13
14 #include "util/parse-options.h"
15 #include "util/trace-event.h"
16
17 #include "util/debug.h"
18
19 #include <sys/prctl.h>
20
21 #include <semaphore.h>
22 #include <pthread.h>
23 #include <math.h>
24
25 static char                     const *input_name = "perf.data";
26
27 static char                     default_sort_order[] = "avg, max, switch, runtime";
28 static const char               *sort_order = default_sort_order;
29
30 static int                      profile_cpu = -1;
31
32 #define PR_SET_NAME             15               /* Set process name */
33 #define MAX_CPUS                4096
34
35 static u64                      run_measurement_overhead;
36 static u64                      sleep_measurement_overhead;
37
38 #define COMM_LEN                20
39 #define SYM_LEN                 129
40
41 #define MAX_PID                 65536
42
43 static unsigned long            nr_tasks;
44
45 struct sched_atom;
46
47 struct task_desc {
48         unsigned long           nr;
49         unsigned long           pid;
50         char                    comm[COMM_LEN];
51
52         unsigned long           nr_events;
53         unsigned long           curr_event;
54         struct sched_atom       **atoms;
55
56         pthread_t               thread;
57         sem_t                   sleep_sem;
58
59         sem_t                   ready_for_work;
60         sem_t                   work_done_sem;
61
62         u64                     cpu_usage;
63 };
64
65 enum sched_event_type {
66         SCHED_EVENT_RUN,
67         SCHED_EVENT_SLEEP,
68         SCHED_EVENT_WAKEUP,
69         SCHED_EVENT_MIGRATION,
70 };
71
72 struct sched_atom {
73         enum sched_event_type   type;
74         int                     specific_wait;
75         u64                     timestamp;
76         u64                     duration;
77         unsigned long           nr;
78         sem_t                   *wait_sem;
79         struct task_desc        *wakee;
80 };
81
82 static struct task_desc         *pid_to_task[MAX_PID];
83
84 static struct task_desc         **tasks;
85
86 static pthread_mutex_t          start_work_mutex = PTHREAD_MUTEX_INITIALIZER;
87 static u64                      start_time;
88
89 static pthread_mutex_t          work_done_wait_mutex = PTHREAD_MUTEX_INITIALIZER;
90
91 static unsigned long            nr_run_events;
92 static unsigned long            nr_sleep_events;
93 static unsigned long            nr_wakeup_events;
94
95 static unsigned long            nr_sleep_corrections;
96 static unsigned long            nr_run_events_optimized;
97
98 static unsigned long            targetless_wakeups;
99 static unsigned long            multitarget_wakeups;
100
101 static u64                      cpu_usage;
102 static u64                      runavg_cpu_usage;
103 static u64                      parent_cpu_usage;
104 static u64                      runavg_parent_cpu_usage;
105
106 static unsigned long            nr_runs;
107 static u64                      sum_runtime;
108 static u64                      sum_fluct;
109 static u64                      run_avg;
110
111 static unsigned int             replay_repeat = 10;
112 static unsigned long            nr_timestamps;
113 static unsigned long            nr_unordered_timestamps;
114 static unsigned long            nr_state_machine_bugs;
115 static unsigned long            nr_context_switch_bugs;
116 static unsigned long            nr_events;
117 static unsigned long            nr_lost_chunks;
118 static unsigned long            nr_lost_events;
119
120 #define TASK_STATE_TO_CHAR_STR "RSDTtZX"
121
122 enum thread_state {
123         THREAD_SLEEPING = 0,
124         THREAD_WAIT_CPU,
125         THREAD_SCHED_IN,
126         THREAD_IGNORE
127 };
128
129 struct work_atom {
130         struct list_head        list;
131         enum thread_state       state;
132         u64                     sched_out_time;
133         u64                     wake_up_time;
134         u64                     sched_in_time;
135         u64                     runtime;
136 };
137
138 struct work_atoms {
139         struct list_head        work_list;
140         struct thread           *thread;
141         struct rb_node          node;
142         u64                     max_lat;
143         u64                     max_lat_at;
144         u64                     total_lat;
145         u64                     nb_atoms;
146         u64                     total_runtime;
147 };
148
149 typedef int (*sort_fn_t)(struct work_atoms *, struct work_atoms *);
150
151 static struct rb_root           atom_root, sorted_atom_root;
152
153 static u64                      all_runtime;
154 static u64                      all_count;
155
156
157 static u64 get_nsecs(void)
158 {
159         struct timespec ts;
160
161         clock_gettime(CLOCK_MONOTONIC, &ts);
162
163         return ts.tv_sec * 1000000000ULL + ts.tv_nsec;
164 }
165
166 static void burn_nsecs(u64 nsecs)
167 {
168         u64 T0 = get_nsecs(), T1;
169
170         do {
171                 T1 = get_nsecs();
172         } while (T1 + run_measurement_overhead < T0 + nsecs);
173 }
174
175 static void sleep_nsecs(u64 nsecs)
176 {
177         struct timespec ts;
178
179         ts.tv_nsec = nsecs % 999999999;
180         ts.tv_sec = nsecs / 999999999;
181
182         nanosleep(&ts, NULL);
183 }
184
185 static void calibrate_run_measurement_overhead(void)
186 {
187         u64 T0, T1, delta, min_delta = 1000000000ULL;
188         int i;
189
190         for (i = 0; i < 10; i++) {
191                 T0 = get_nsecs();
192                 burn_nsecs(0);
193                 T1 = get_nsecs();
194                 delta = T1-T0;
195                 min_delta = min(min_delta, delta);
196         }
197         run_measurement_overhead = min_delta;
198
199         printf("run measurement overhead: %" PRIu64 " nsecs\n", min_delta);
200 }
201
202 static void calibrate_sleep_measurement_overhead(void)
203 {
204         u64 T0, T1, delta, min_delta = 1000000000ULL;
205         int i;
206
207         for (i = 0; i < 10; i++) {
208                 T0 = get_nsecs();
209                 sleep_nsecs(10000);
210                 T1 = get_nsecs();
211                 delta = T1-T0;
212                 min_delta = min(min_delta, delta);
213         }
214         min_delta -= 10000;
215         sleep_measurement_overhead = min_delta;
216
217         printf("sleep measurement overhead: %" PRIu64 " nsecs\n", min_delta);
218 }
219
220 static struct sched_atom *
221 get_new_event(struct task_desc *task, u64 timestamp)
222 {
223         struct sched_atom *event = zalloc(sizeof(*event));
224         unsigned long idx = task->nr_events;
225         size_t size;
226
227         event->timestamp = timestamp;
228         event->nr = idx;
229
230         task->nr_events++;
231         size = sizeof(struct sched_atom *) * task->nr_events;
232         task->atoms = realloc(task->atoms, size);
233         BUG_ON(!task->atoms);
234
235         task->atoms[idx] = event;
236
237         return event;
238 }
239
240 static struct sched_atom *last_event(struct task_desc *task)
241 {
242         if (!task->nr_events)
243                 return NULL;
244
245         return task->atoms[task->nr_events - 1];
246 }
247
248 static void
249 add_sched_event_run(struct task_desc *task, u64 timestamp, u64 duration)
250 {
251         struct sched_atom *event, *curr_event = last_event(task);
252
253         /*
254          * optimize an existing RUN event by merging this one
255          * to it:
256          */
257         if (curr_event && curr_event->type == SCHED_EVENT_RUN) {
258                 nr_run_events_optimized++;
259                 curr_event->duration += duration;
260                 return;
261         }
262
263         event = get_new_event(task, timestamp);
264
265         event->type = SCHED_EVENT_RUN;
266         event->duration = duration;
267
268         nr_run_events++;
269 }
270
271 static void
272 add_sched_event_wakeup(struct task_desc *task, u64 timestamp,
273                        struct task_desc *wakee)
274 {
275         struct sched_atom *event, *wakee_event;
276
277         event = get_new_event(task, timestamp);
278         event->type = SCHED_EVENT_WAKEUP;
279         event->wakee = wakee;
280
281         wakee_event = last_event(wakee);
282         if (!wakee_event || wakee_event->type != SCHED_EVENT_SLEEP) {
283                 targetless_wakeups++;
284                 return;
285         }
286         if (wakee_event->wait_sem) {
287                 multitarget_wakeups++;
288                 return;
289         }
290
291         wakee_event->wait_sem = zalloc(sizeof(*wakee_event->wait_sem));
292         sem_init(wakee_event->wait_sem, 0, 0);
293         wakee_event->specific_wait = 1;
294         event->wait_sem = wakee_event->wait_sem;
295
296         nr_wakeup_events++;
297 }
298
299 static void
300 add_sched_event_sleep(struct task_desc *task, u64 timestamp,
301                       u64 task_state __used)
302 {
303         struct sched_atom *event = get_new_event(task, timestamp);
304
305         event->type = SCHED_EVENT_SLEEP;
306
307         nr_sleep_events++;
308 }
309
310 static struct task_desc *register_pid(unsigned long pid, const char *comm)
311 {
312         struct task_desc *task;
313
314         BUG_ON(pid >= MAX_PID);
315
316         task = pid_to_task[pid];
317
318         if (task)
319                 return task;
320
321         task = zalloc(sizeof(*task));
322         task->pid = pid;
323         task->nr = nr_tasks;
324         strcpy(task->comm, comm);
325         /*
326          * every task starts in sleeping state - this gets ignored
327          * if there's no wakeup pointing to this sleep state:
328          */
329         add_sched_event_sleep(task, 0, 0);
330
331         pid_to_task[pid] = task;
332         nr_tasks++;
333         tasks = realloc(tasks, nr_tasks*sizeof(struct task_task *));
334         BUG_ON(!tasks);
335         tasks[task->nr] = task;
336
337         if (verbose)
338                 printf("registered task #%ld, PID %ld (%s)\n", nr_tasks, pid, comm);
339
340         return task;
341 }
342
343
344 static void print_task_traces(void)
345 {
346         struct task_desc *task;
347         unsigned long i;
348
349         for (i = 0; i < nr_tasks; i++) {
350                 task = tasks[i];
351                 printf("task %6ld (%20s:%10ld), nr_events: %ld\n",
352                         task->nr, task->comm, task->pid, task->nr_events);
353         }
354 }
355
356 static void add_cross_task_wakeups(void)
357 {
358         struct task_desc *task1, *task2;
359         unsigned long i, j;
360
361         for (i = 0; i < nr_tasks; i++) {
362                 task1 = tasks[i];
363                 j = i + 1;
364                 if (j == nr_tasks)
365                         j = 0;
366                 task2 = tasks[j];
367                 add_sched_event_wakeup(task1, 0, task2);
368         }
369 }
370
371 static void
372 process_sched_event(struct task_desc *this_task __used, struct sched_atom *atom)
373 {
374         int ret = 0;
375
376         switch (atom->type) {
377                 case SCHED_EVENT_RUN:
378                         burn_nsecs(atom->duration);
379                         break;
380                 case SCHED_EVENT_SLEEP:
381                         if (atom->wait_sem)
382                                 ret = sem_wait(atom->wait_sem);
383                         BUG_ON(ret);
384                         break;
385                 case SCHED_EVENT_WAKEUP:
386                         if (atom->wait_sem)
387                                 ret = sem_post(atom->wait_sem);
388                         BUG_ON(ret);
389                         break;
390                 case SCHED_EVENT_MIGRATION:
391                         break;
392                 default:
393                         BUG_ON(1);
394         }
395 }
396
397 static u64 get_cpu_usage_nsec_parent(void)
398 {
399         struct rusage ru;
400         u64 sum;
401         int err;
402
403         err = getrusage(RUSAGE_SELF, &ru);
404         BUG_ON(err);
405
406         sum =  ru.ru_utime.tv_sec*1e9 + ru.ru_utime.tv_usec*1e3;
407         sum += ru.ru_stime.tv_sec*1e9 + ru.ru_stime.tv_usec*1e3;
408
409         return sum;
410 }
411
412 static int self_open_counters(void)
413 {
414         struct perf_event_attr attr;
415         int fd;
416
417         memset(&attr, 0, sizeof(attr));
418
419         attr.type = PERF_TYPE_SOFTWARE;
420         attr.config = PERF_COUNT_SW_TASK_CLOCK;
421
422         fd = sys_perf_event_open(&attr, 0, -1, -1, 0);
423
424         if (fd < 0)
425                 die("Error: sys_perf_event_open() syscall returned"
426                     "with %d (%s)\n", fd, strerror(errno));
427         return fd;
428 }
429
430 static u64 get_cpu_usage_nsec_self(int fd)
431 {
432         u64 runtime;
433         int ret;
434
435         ret = read(fd, &runtime, sizeof(runtime));
436         BUG_ON(ret != sizeof(runtime));
437
438         return runtime;
439 }
440
441 static void *thread_func(void *ctx)
442 {
443         struct task_desc *this_task = ctx;
444         u64 cpu_usage_0, cpu_usage_1;
445         unsigned long i, ret;
446         char comm2[22];
447         int fd;
448
449         sprintf(comm2, ":%s", this_task->comm);
450         prctl(PR_SET_NAME, comm2);
451         fd = self_open_counters();
452
453 again:
454         ret = sem_post(&this_task->ready_for_work);
455         BUG_ON(ret);
456         ret = pthread_mutex_lock(&start_work_mutex);
457         BUG_ON(ret);
458         ret = pthread_mutex_unlock(&start_work_mutex);
459         BUG_ON(ret);
460
461         cpu_usage_0 = get_cpu_usage_nsec_self(fd);
462
463         for (i = 0; i < this_task->nr_events; i++) {
464                 this_task->curr_event = i;
465                 process_sched_event(this_task, this_task->atoms[i]);
466         }
467
468         cpu_usage_1 = get_cpu_usage_nsec_self(fd);
469         this_task->cpu_usage = cpu_usage_1 - cpu_usage_0;
470         ret = sem_post(&this_task->work_done_sem);
471         BUG_ON(ret);
472
473         ret = pthread_mutex_lock(&work_done_wait_mutex);
474         BUG_ON(ret);
475         ret = pthread_mutex_unlock(&work_done_wait_mutex);
476         BUG_ON(ret);
477
478         goto again;
479 }
480
481 static void create_tasks(void)
482 {
483         struct task_desc *task;
484         pthread_attr_t attr;
485         unsigned long i;
486         int err;
487
488         err = pthread_attr_init(&attr);
489         BUG_ON(err);
490         err = pthread_attr_setstacksize(&attr,
491                         (size_t) max(16 * 1024, PTHREAD_STACK_MIN));
492         BUG_ON(err);
493         err = pthread_mutex_lock(&start_work_mutex);
494         BUG_ON(err);
495         err = pthread_mutex_lock(&work_done_wait_mutex);
496         BUG_ON(err);
497         for (i = 0; i < nr_tasks; i++) {
498                 task = tasks[i];
499                 sem_init(&task->sleep_sem, 0, 0);
500                 sem_init(&task->ready_for_work, 0, 0);
501                 sem_init(&task->work_done_sem, 0, 0);
502                 task->curr_event = 0;
503                 err = pthread_create(&task->thread, &attr, thread_func, task);
504                 BUG_ON(err);
505         }
506 }
507
508 static void wait_for_tasks(void)
509 {
510         u64 cpu_usage_0, cpu_usage_1;
511         struct task_desc *task;
512         unsigned long i, ret;
513
514         start_time = get_nsecs();
515         cpu_usage = 0;
516         pthread_mutex_unlock(&work_done_wait_mutex);
517
518         for (i = 0; i < nr_tasks; i++) {
519                 task = tasks[i];
520                 ret = sem_wait(&task->ready_for_work);
521                 BUG_ON(ret);
522                 sem_init(&task->ready_for_work, 0, 0);
523         }
524         ret = pthread_mutex_lock(&work_done_wait_mutex);
525         BUG_ON(ret);
526
527         cpu_usage_0 = get_cpu_usage_nsec_parent();
528
529         pthread_mutex_unlock(&start_work_mutex);
530
531         for (i = 0; i < nr_tasks; i++) {
532                 task = tasks[i];
533                 ret = sem_wait(&task->work_done_sem);
534                 BUG_ON(ret);
535                 sem_init(&task->work_done_sem, 0, 0);
536                 cpu_usage += task->cpu_usage;
537                 task->cpu_usage = 0;
538         }
539
540         cpu_usage_1 = get_cpu_usage_nsec_parent();
541         if (!runavg_cpu_usage)
542                 runavg_cpu_usage = cpu_usage;
543         runavg_cpu_usage = (runavg_cpu_usage*9 + cpu_usage)/10;
544
545         parent_cpu_usage = cpu_usage_1 - cpu_usage_0;
546         if (!runavg_parent_cpu_usage)
547                 runavg_parent_cpu_usage = parent_cpu_usage;
548         runavg_parent_cpu_usage = (runavg_parent_cpu_usage*9 +
549                                    parent_cpu_usage)/10;
550
551         ret = pthread_mutex_lock(&start_work_mutex);
552         BUG_ON(ret);
553
554         for (i = 0; i < nr_tasks; i++) {
555                 task = tasks[i];
556                 sem_init(&task->sleep_sem, 0, 0);
557                 task->curr_event = 0;
558         }
559 }
560
561 static void run_one_test(void)
562 {
563         u64 T0, T1, delta, avg_delta, fluct;
564
565         T0 = get_nsecs();
566         wait_for_tasks();
567         T1 = get_nsecs();
568
569         delta = T1 - T0;
570         sum_runtime += delta;
571         nr_runs++;
572
573         avg_delta = sum_runtime / nr_runs;
574         if (delta < avg_delta)
575                 fluct = avg_delta - delta;
576         else
577                 fluct = delta - avg_delta;
578         sum_fluct += fluct;
579         if (!run_avg)
580                 run_avg = delta;
581         run_avg = (run_avg*9 + delta)/10;
582
583         printf("#%-3ld: %0.3f, ",
584                 nr_runs, (double)delta/1000000.0);
585
586         printf("ravg: %0.2f, ",
587                 (double)run_avg/1e6);
588
589         printf("cpu: %0.2f / %0.2f",
590                 (double)cpu_usage/1e6, (double)runavg_cpu_usage/1e6);
591
592 #if 0
593         /*
594          * rusage statistics done by the parent, these are less
595          * accurate than the sum_exec_runtime based statistics:
596          */
597         printf(" [%0.2f / %0.2f]",
598                 (double)parent_cpu_usage/1e6,
599                 (double)runavg_parent_cpu_usage/1e6);
600 #endif
601
602         printf("\n");
603
604         if (nr_sleep_corrections)
605                 printf(" (%ld sleep corrections)\n", nr_sleep_corrections);
606         nr_sleep_corrections = 0;
607 }
608
609 static void test_calibrations(void)
610 {
611         u64 T0, T1;
612
613         T0 = get_nsecs();
614         burn_nsecs(1e6);
615         T1 = get_nsecs();
616
617         printf("the run test took %" PRIu64 " nsecs\n", T1 - T0);
618
619         T0 = get_nsecs();
620         sleep_nsecs(1e6);
621         T1 = get_nsecs();
622
623         printf("the sleep test took %" PRIu64 " nsecs\n", T1 - T0);
624 }
625
626 #define FILL_FIELD(ptr, field, event, data)     \
627         ptr.field = (typeof(ptr.field)) raw_field_value(event, #field, data)
628
629 #define FILL_ARRAY(ptr, array, event, data)                     \
630 do {                                                            \
631         void *__array = raw_field_ptr(event, #array, data);     \
632         memcpy(ptr.array, __array, sizeof(ptr.array));  \
633 } while(0)
634
635 #define FILL_COMMON_FIELDS(ptr, event, data)                    \
636 do {                                                            \
637         FILL_FIELD(ptr, common_type, event, data);              \
638         FILL_FIELD(ptr, common_flags, event, data);             \
639         FILL_FIELD(ptr, common_preempt_count, event, data);     \
640         FILL_FIELD(ptr, common_pid, event, data);               \
641         FILL_FIELD(ptr, common_tgid, event, data);              \
642 } while (0)
643
644
645
646 struct trace_switch_event {
647         u32 size;
648
649         u16 common_type;
650         u8 common_flags;
651         u8 common_preempt_count;
652         u32 common_pid;
653         u32 common_tgid;
654
655         char prev_comm[16];
656         u32 prev_pid;
657         u32 prev_prio;
658         u64 prev_state;
659         char next_comm[16];
660         u32 next_pid;
661         u32 next_prio;
662 };
663
664 struct trace_runtime_event {
665         u32 size;
666
667         u16 common_type;
668         u8 common_flags;
669         u8 common_preempt_count;
670         u32 common_pid;
671         u32 common_tgid;
672
673         char comm[16];
674         u32 pid;
675         u64 runtime;
676         u64 vruntime;
677 };
678
679 struct trace_wakeup_event {
680         u32 size;
681
682         u16 common_type;
683         u8 common_flags;
684         u8 common_preempt_count;
685         u32 common_pid;
686         u32 common_tgid;
687
688         char comm[16];
689         u32 pid;
690
691         u32 prio;
692         u32 success;
693         u32 cpu;
694 };
695
696 struct trace_fork_event {
697         u32 size;
698
699         u16 common_type;
700         u8 common_flags;
701         u8 common_preempt_count;
702         u32 common_pid;
703         u32 common_tgid;
704
705         char parent_comm[16];
706         u32 parent_pid;
707         char child_comm[16];
708         u32 child_pid;
709 };
710
711 struct trace_migrate_task_event {
712         u32 size;
713
714         u16 common_type;
715         u8 common_flags;
716         u8 common_preempt_count;
717         u32 common_pid;
718         u32 common_tgid;
719
720         char comm[16];
721         u32 pid;
722
723         u32 prio;
724         u32 cpu;
725 };
726
727 struct trace_sched_handler {
728         void (*switch_event)(struct trace_switch_event *,
729                              struct machine *,
730                              struct event *,
731                              int cpu,
732                              u64 timestamp,
733                              struct thread *thread);
734
735         void (*runtime_event)(struct trace_runtime_event *,
736                               struct machine *,
737                               struct event *,
738                               int cpu,
739                               u64 timestamp,
740                               struct thread *thread);
741
742         void (*wakeup_event)(struct trace_wakeup_event *,
743                              struct machine *,
744                              struct event *,
745                              int cpu,
746                              u64 timestamp,
747                              struct thread *thread);
748
749         void (*fork_event)(struct trace_fork_event *,
750                            struct event *,
751                            int cpu,
752                            u64 timestamp,
753                            struct thread *thread);
754
755         void (*migrate_task_event)(struct trace_migrate_task_event *,
756                            struct machine *machine,
757                            struct event *,
758                            int cpu,
759                            u64 timestamp,
760                            struct thread *thread);
761 };
762
763
764 static void
765 replay_wakeup_event(struct trace_wakeup_event *wakeup_event,
766                     struct machine *machine __used,
767                     struct event *event,
768                     int cpu __used,
769                     u64 timestamp __used,
770                     struct thread *thread __used)
771 {
772         struct task_desc *waker, *wakee;
773
774         if (verbose) {
775                 printf("sched_wakeup event %p\n", event);
776
777                 printf(" ... pid %d woke up %s/%d\n",
778                         wakeup_event->common_pid,
779                         wakeup_event->comm,
780                         wakeup_event->pid);
781         }
782
783         waker = register_pid(wakeup_event->common_pid, "<unknown>");
784         wakee = register_pid(wakeup_event->pid, wakeup_event->comm);
785
786         add_sched_event_wakeup(waker, timestamp, wakee);
787 }
788
789 static u64 cpu_last_switched[MAX_CPUS];
790
791 static void
792 replay_switch_event(struct trace_switch_event *switch_event,
793                     struct machine *machine __used,
794                     struct event *event,
795                     int cpu,
796                     u64 timestamp,
797                     struct thread *thread __used)
798 {
799         struct task_desc *prev, __used *next;
800         u64 timestamp0;
801         s64 delta;
802
803         if (verbose)
804                 printf("sched_switch event %p\n", event);
805
806         if (cpu >= MAX_CPUS || cpu < 0)
807                 return;
808
809         timestamp0 = cpu_last_switched[cpu];
810         if (timestamp0)
811                 delta = timestamp - timestamp0;
812         else
813                 delta = 0;
814
815         if (delta < 0)
816                 die("hm, delta: %" PRIu64 " < 0 ?\n", delta);
817
818         if (verbose) {
819                 printf(" ... switch from %s/%d to %s/%d [ran %" PRIu64 " nsecs]\n",
820                         switch_event->prev_comm, switch_event->prev_pid,
821                         switch_event->next_comm, switch_event->next_pid,
822                         delta);
823         }
824
825         prev = register_pid(switch_event->prev_pid, switch_event->prev_comm);
826         next = register_pid(switch_event->next_pid, switch_event->next_comm);
827
828         cpu_last_switched[cpu] = timestamp;
829
830         add_sched_event_run(prev, timestamp, delta);
831         add_sched_event_sleep(prev, timestamp, switch_event->prev_state);
832 }
833
834
835 static void
836 replay_fork_event(struct trace_fork_event *fork_event,
837                   struct event *event,
838                   int cpu __used,
839                   u64 timestamp __used,
840                   struct thread *thread __used)
841 {
842         if (verbose) {
843                 printf("sched_fork event %p\n", event);
844                 printf("... parent: %s/%d\n", fork_event->parent_comm, fork_event->parent_pid);
845                 printf("...  child: %s/%d\n", fork_event->child_comm, fork_event->child_pid);
846         }
847         register_pid(fork_event->parent_pid, fork_event->parent_comm);
848         register_pid(fork_event->child_pid, fork_event->child_comm);
849 }
850
851 static struct trace_sched_handler replay_ops  = {
852         .wakeup_event           = replay_wakeup_event,
853         .switch_event           = replay_switch_event,
854         .fork_event             = replay_fork_event,
855 };
856
857 struct sort_dimension {
858         const char              *name;
859         sort_fn_t               cmp;
860         struct list_head        list;
861 };
862
863 static LIST_HEAD(cmp_pid);
864
865 static int
866 thread_lat_cmp(struct list_head *list, struct work_atoms *l, struct work_atoms *r)
867 {
868         struct sort_dimension *sort;
869         int ret = 0;
870
871         BUG_ON(list_empty(list));
872
873         list_for_each_entry(sort, list, list) {
874                 ret = sort->cmp(l, r);
875                 if (ret)
876                         return ret;
877         }
878
879         return ret;
880 }
881
882 static struct work_atoms *
883 thread_atoms_search(struct rb_root *root, struct thread *thread,
884                          struct list_head *sort_list)
885 {
886         struct rb_node *node = root->rb_node;
887         struct work_atoms key = { .thread = thread };
888
889         while (node) {
890                 struct work_atoms *atoms;
891                 int cmp;
892
893                 atoms = container_of(node, struct work_atoms, node);
894
895                 cmp = thread_lat_cmp(sort_list, &key, atoms);
896                 if (cmp > 0)
897                         node = node->rb_left;
898                 else if (cmp < 0)
899                         node = node->rb_right;
900                 else {
901                         BUG_ON(thread != atoms->thread);
902                         return atoms;
903                 }
904         }
905         return NULL;
906 }
907
908 static void
909 __thread_latency_insert(struct rb_root *root, struct work_atoms *data,
910                          struct list_head *sort_list)
911 {
912         struct rb_node **new = &(root->rb_node), *parent = NULL;
913
914         while (*new) {
915                 struct work_atoms *this;
916                 int cmp;
917
918                 this = container_of(*new, struct work_atoms, node);
919                 parent = *new;
920
921                 cmp = thread_lat_cmp(sort_list, data, this);
922
923                 if (cmp > 0)
924                         new = &((*new)->rb_left);
925                 else
926                         new = &((*new)->rb_right);
927         }
928
929         rb_link_node(&data->node, parent, new);
930         rb_insert_color(&data->node, root);
931 }
932
933 static void thread_atoms_insert(struct thread *thread)
934 {
935         struct work_atoms *atoms = zalloc(sizeof(*atoms));
936         if (!atoms)
937                 die("No memory");
938
939         atoms->thread = thread;
940         INIT_LIST_HEAD(&atoms->work_list);
941         __thread_latency_insert(&atom_root, atoms, &cmp_pid);
942 }
943
944 static void
945 latency_fork_event(struct trace_fork_event *fork_event __used,
946                    struct event *event __used,
947                    int cpu __used,
948                    u64 timestamp __used,
949                    struct thread *thread __used)
950 {
951         /* should insert the newcomer */
952 }
953
954 __used
955 static char sched_out_state(struct trace_switch_event *switch_event)
956 {
957         const char *str = TASK_STATE_TO_CHAR_STR;
958
959         return str[switch_event->prev_state];
960 }
961
962 static void
963 add_sched_out_event(struct work_atoms *atoms,
964                     char run_state,
965                     u64 timestamp)
966 {
967         struct work_atom *atom = zalloc(sizeof(*atom));
968         if (!atom)
969                 die("Non memory");
970
971         atom->sched_out_time = timestamp;
972
973         if (run_state == 'R') {
974                 atom->state = THREAD_WAIT_CPU;
975                 atom->wake_up_time = atom->sched_out_time;
976         }
977
978         list_add_tail(&atom->list, &atoms->work_list);
979 }
980
981 static void
982 add_runtime_event(struct work_atoms *atoms, u64 delta, u64 timestamp __used)
983 {
984         struct work_atom *atom;
985
986         BUG_ON(list_empty(&atoms->work_list));
987
988         atom = list_entry(atoms->work_list.prev, struct work_atom, list);
989
990         atom->runtime += delta;
991         atoms->total_runtime += delta;
992 }
993
994 static void
995 add_sched_in_event(struct work_atoms *atoms, u64 timestamp)
996 {
997         struct work_atom *atom;
998         u64 delta;
999
1000         if (list_empty(&atoms->work_list))
1001                 return;
1002
1003         atom = list_entry(atoms->work_list.prev, struct work_atom, list);
1004
1005         if (atom->state != THREAD_WAIT_CPU)
1006                 return;
1007
1008         if (timestamp < atom->wake_up_time) {
1009                 atom->state = THREAD_IGNORE;
1010                 return;
1011         }
1012
1013         atom->state = THREAD_SCHED_IN;
1014         atom->sched_in_time = timestamp;
1015
1016         delta = atom->sched_in_time - atom->wake_up_time;
1017         atoms->total_lat += delta;
1018         if (delta > atoms->max_lat) {
1019                 atoms->max_lat = delta;
1020                 atoms->max_lat_at = timestamp;
1021         }
1022         atoms->nb_atoms++;
1023 }
1024
1025 static void
1026 latency_switch_event(struct trace_switch_event *switch_event,
1027                      struct machine *machine,
1028                      struct event *event __used,
1029                      int cpu,
1030                      u64 timestamp,
1031                      struct thread *thread __used)
1032 {
1033         struct work_atoms *out_events, *in_events;
1034         struct thread *sched_out, *sched_in;
1035         u64 timestamp0;
1036         s64 delta;
1037
1038         BUG_ON(cpu >= MAX_CPUS || cpu < 0);
1039
1040         timestamp0 = cpu_last_switched[cpu];
1041         cpu_last_switched[cpu] = timestamp;
1042         if (timestamp0)
1043                 delta = timestamp - timestamp0;
1044         else
1045                 delta = 0;
1046
1047         if (delta < 0)
1048                 die("hm, delta: %" PRIu64 " < 0 ?\n", delta);
1049
1050
1051         sched_out = machine__findnew_thread(machine, switch_event->prev_pid);
1052         sched_in = machine__findnew_thread(machine, switch_event->next_pid);
1053
1054         out_events = thread_atoms_search(&atom_root, sched_out, &cmp_pid);
1055         if (!out_events) {
1056                 thread_atoms_insert(sched_out);
1057                 out_events = thread_atoms_search(&atom_root, sched_out, &cmp_pid);
1058                 if (!out_events)
1059                         die("out-event: Internal tree error");
1060         }
1061         add_sched_out_event(out_events, sched_out_state(switch_event), timestamp);
1062
1063         in_events = thread_atoms_search(&atom_root, sched_in, &cmp_pid);
1064         if (!in_events) {
1065                 thread_atoms_insert(sched_in);
1066                 in_events = thread_atoms_search(&atom_root, sched_in, &cmp_pid);
1067                 if (!in_events)
1068                         die("in-event: Internal tree error");
1069                 /*
1070                  * Take came in we have not heard about yet,
1071                  * add in an initial atom in runnable state:
1072                  */
1073                 add_sched_out_event(in_events, 'R', timestamp);
1074         }
1075         add_sched_in_event(in_events, timestamp);
1076 }
1077
1078 static void
1079 latency_runtime_event(struct trace_runtime_event *runtime_event,
1080                      struct machine *machine,
1081                      struct event *event __used,
1082                      int cpu,
1083                      u64 timestamp,
1084                      struct thread *this_thread __used)
1085 {
1086         struct thread *thread = machine__findnew_thread(machine, runtime_event->pid);
1087         struct work_atoms *atoms = thread_atoms_search(&atom_root, thread, &cmp_pid);
1088
1089         BUG_ON(cpu >= MAX_CPUS || cpu < 0);
1090         if (!atoms) {
1091                 thread_atoms_insert(thread);
1092                 atoms = thread_atoms_search(&atom_root, thread, &cmp_pid);
1093                 if (!atoms)
1094                         die("in-event: Internal tree error");
1095                 add_sched_out_event(atoms, 'R', timestamp);
1096         }
1097
1098         add_runtime_event(atoms, runtime_event->runtime, timestamp);
1099 }
1100
1101 static void
1102 latency_wakeup_event(struct trace_wakeup_event *wakeup_event,
1103                      struct machine *machine,
1104                      struct event *__event __used,
1105                      int cpu __used,
1106                      u64 timestamp,
1107                      struct thread *thread __used)
1108 {
1109         struct work_atoms *atoms;
1110         struct work_atom *atom;
1111         struct thread *wakee;
1112
1113         /* Note for later, it may be interesting to observe the failing cases */
1114         if (!wakeup_event->success)
1115                 return;
1116
1117         wakee = machine__findnew_thread(machine, wakeup_event->pid);
1118         atoms = thread_atoms_search(&atom_root, wakee, &cmp_pid);
1119         if (!atoms) {
1120                 thread_atoms_insert(wakee);
1121                 atoms = thread_atoms_search(&atom_root, wakee, &cmp_pid);
1122                 if (!atoms)
1123                         die("wakeup-event: Internal tree error");
1124                 add_sched_out_event(atoms, 'S', timestamp);
1125         }
1126
1127         BUG_ON(list_empty(&atoms->work_list));
1128
1129         atom = list_entry(atoms->work_list.prev, struct work_atom, list);
1130
1131         /*
1132          * You WILL be missing events if you've recorded only
1133          * one CPU, or are only looking at only one, so don't
1134          * make useless noise.
1135          */
1136         if (profile_cpu == -1 && atom->state != THREAD_SLEEPING)
1137                 nr_state_machine_bugs++;
1138
1139         nr_timestamps++;
1140         if (atom->sched_out_time > timestamp) {
1141                 nr_unordered_timestamps++;
1142                 return;
1143         }
1144
1145         atom->state = THREAD_WAIT_CPU;
1146         atom->wake_up_time = timestamp;
1147 }
1148
1149 static void
1150 latency_migrate_task_event(struct trace_migrate_task_event *migrate_task_event,
1151                      struct machine *machine,
1152                      struct event *__event __used,
1153                      int cpu __used,
1154                      u64 timestamp,
1155                      struct thread *thread __used)
1156 {
1157         struct work_atoms *atoms;
1158         struct work_atom *atom;
1159         struct thread *migrant;
1160
1161         /*
1162          * Only need to worry about migration when profiling one CPU.
1163          */
1164         if (profile_cpu == -1)
1165                 return;
1166
1167         migrant = machine__findnew_thread(machine, migrate_task_event->pid);
1168         atoms = thread_atoms_search(&atom_root, migrant, &cmp_pid);
1169         if (!atoms) {
1170                 thread_atoms_insert(migrant);
1171                 register_pid(migrant->pid, migrant->comm);
1172                 atoms = thread_atoms_search(&atom_root, migrant, &cmp_pid);
1173                 if (!atoms)
1174                         die("migration-event: Internal tree error");
1175                 add_sched_out_event(atoms, 'R', timestamp);
1176         }
1177
1178         BUG_ON(list_empty(&atoms->work_list));
1179
1180         atom = list_entry(atoms->work_list.prev, struct work_atom, list);
1181         atom->sched_in_time = atom->sched_out_time = atom->wake_up_time = timestamp;
1182
1183         nr_timestamps++;
1184
1185         if (atom->sched_out_time > timestamp)
1186                 nr_unordered_timestamps++;
1187 }
1188
1189 static struct trace_sched_handler lat_ops  = {
1190         .wakeup_event           = latency_wakeup_event,
1191         .switch_event           = latency_switch_event,
1192         .runtime_event          = latency_runtime_event,
1193         .fork_event             = latency_fork_event,
1194         .migrate_task_event     = latency_migrate_task_event,
1195 };
1196
1197 static void output_lat_thread(struct work_atoms *work_list)
1198 {
1199         int i;
1200         int ret;
1201         u64 avg;
1202
1203         if (!work_list->nb_atoms)
1204                 return;
1205         /*
1206          * Ignore idle threads:
1207          */
1208         if (!strcmp(work_list->thread->comm, "swapper"))
1209                 return;
1210
1211         all_runtime += work_list->total_runtime;
1212         all_count += work_list->nb_atoms;
1213
1214         ret = printf("  %s:%d ", work_list->thread->comm, work_list->thread->pid);
1215
1216         for (i = 0; i < 24 - ret; i++)
1217                 printf(" ");
1218
1219         avg = work_list->total_lat / work_list->nb_atoms;
1220
1221         printf("|%11.3f ms |%9" PRIu64 " | avg:%9.3f ms | max:%9.3f ms | max at: %9.6f s\n",
1222               (double)work_list->total_runtime / 1e6,
1223                  work_list->nb_atoms, (double)avg / 1e6,
1224                  (double)work_list->max_lat / 1e6,
1225                  (double)work_list->max_lat_at / 1e9);
1226 }
1227
1228 static int pid_cmp(struct work_atoms *l, struct work_atoms *r)
1229 {
1230         if (l->thread->pid < r->thread->pid)
1231                 return -1;
1232         if (l->thread->pid > r->thread->pid)
1233                 return 1;
1234
1235         return 0;
1236 }
1237
1238 static struct sort_dimension pid_sort_dimension = {
1239         .name                   = "pid",
1240         .cmp                    = pid_cmp,
1241 };
1242
1243 static int avg_cmp(struct work_atoms *l, struct work_atoms *r)
1244 {
1245         u64 avgl, avgr;
1246
1247         if (!l->nb_atoms)
1248                 return -1;
1249
1250         if (!r->nb_atoms)
1251                 return 1;
1252
1253         avgl = l->total_lat / l->nb_atoms;
1254         avgr = r->total_lat / r->nb_atoms;
1255
1256         if (avgl < avgr)
1257                 return -1;
1258         if (avgl > avgr)
1259                 return 1;
1260
1261         return 0;
1262 }
1263
1264 static struct sort_dimension avg_sort_dimension = {
1265         .name                   = "avg",
1266         .cmp                    = avg_cmp,
1267 };
1268
1269 static int max_cmp(struct work_atoms *l, struct work_atoms *r)
1270 {
1271         if (l->max_lat < r->max_lat)
1272                 return -1;
1273         if (l->max_lat > r->max_lat)
1274                 return 1;
1275
1276         return 0;
1277 }
1278
1279 static struct sort_dimension max_sort_dimension = {
1280         .name                   = "max",
1281         .cmp                    = max_cmp,
1282 };
1283
1284 static int switch_cmp(struct work_atoms *l, struct work_atoms *r)
1285 {
1286         if (l->nb_atoms < r->nb_atoms)
1287                 return -1;
1288         if (l->nb_atoms > r->nb_atoms)
1289                 return 1;
1290
1291         return 0;
1292 }
1293
1294 static struct sort_dimension switch_sort_dimension = {
1295         .name                   = "switch",
1296         .cmp                    = switch_cmp,
1297 };
1298
1299 static int runtime_cmp(struct work_atoms *l, struct work_atoms *r)
1300 {
1301         if (l->total_runtime < r->total_runtime)
1302                 return -1;
1303         if (l->total_runtime > r->total_runtime)
1304                 return 1;
1305
1306         return 0;
1307 }
1308
1309 static struct sort_dimension runtime_sort_dimension = {
1310         .name                   = "runtime",
1311         .cmp                    = runtime_cmp,
1312 };
1313
1314 static struct sort_dimension *available_sorts[] = {
1315         &pid_sort_dimension,
1316         &avg_sort_dimension,
1317         &max_sort_dimension,
1318         &switch_sort_dimension,
1319         &runtime_sort_dimension,
1320 };
1321
1322 #define NB_AVAILABLE_SORTS      (int)(sizeof(available_sorts) / sizeof(struct sort_dimension *))
1323
1324 static LIST_HEAD(sort_list);
1325
1326 static int sort_dimension__add(const char *tok, struct list_head *list)
1327 {
1328         int i;
1329
1330         for (i = 0; i < NB_AVAILABLE_SORTS; i++) {
1331                 if (!strcmp(available_sorts[i]->name, tok)) {
1332                         list_add_tail(&available_sorts[i]->list, list);
1333
1334                         return 0;
1335                 }
1336         }
1337
1338         return -1;
1339 }
1340
1341 static void setup_sorting(void);
1342
1343 static void sort_lat(void)
1344 {
1345         struct rb_node *node;
1346
1347         for (;;) {
1348                 struct work_atoms *data;
1349                 node = rb_first(&atom_root);
1350                 if (!node)
1351                         break;
1352
1353                 rb_erase(node, &atom_root);
1354                 data = rb_entry(node, struct work_atoms, node);
1355                 __thread_latency_insert(&sorted_atom_root, data, &sort_list);
1356         }
1357 }
1358
1359 static struct trace_sched_handler *trace_handler;
1360
1361 static void
1362 process_sched_wakeup_event(struct perf_tool *tool __used,
1363                            struct event *event,
1364                            struct perf_sample *sample,
1365                            struct machine *machine,
1366                            struct thread *thread)
1367 {
1368         void *data = sample->raw_data;
1369         struct trace_wakeup_event wakeup_event;
1370
1371         FILL_COMMON_FIELDS(wakeup_event, event, data);
1372
1373         FILL_ARRAY(wakeup_event, comm, event, data);
1374         FILL_FIELD(wakeup_event, pid, event, data);
1375         FILL_FIELD(wakeup_event, prio, event, data);
1376         FILL_FIELD(wakeup_event, success, event, data);
1377         FILL_FIELD(wakeup_event, cpu, event, data);
1378
1379         if (trace_handler->wakeup_event)
1380                 trace_handler->wakeup_event(&wakeup_event, machine, event,
1381                                             sample->cpu, sample->time, thread);
1382 }
1383
1384 /*
1385  * Track the current task - that way we can know whether there's any
1386  * weird events, such as a task being switched away that is not current.
1387  */
1388 static int max_cpu;
1389
1390 static u32 curr_pid[MAX_CPUS] = { [0 ... MAX_CPUS-1] = -1 };
1391
1392 static struct thread *curr_thread[MAX_CPUS];
1393
1394 static char next_shortname1 = 'A';
1395 static char next_shortname2 = '0';
1396
1397 static void
1398 map_switch_event(struct trace_switch_event *switch_event,
1399                  struct machine *machine,
1400                  struct event *event __used,
1401                  int this_cpu,
1402                  u64 timestamp,
1403                  struct thread *thread __used)
1404 {
1405         struct thread *sched_out __used, *sched_in;
1406         int new_shortname;
1407         u64 timestamp0;
1408         s64 delta;
1409         int cpu;
1410
1411         BUG_ON(this_cpu >= MAX_CPUS || this_cpu < 0);
1412
1413         if (this_cpu > max_cpu)
1414                 max_cpu = this_cpu;
1415
1416         timestamp0 = cpu_last_switched[this_cpu];
1417         cpu_last_switched[this_cpu] = timestamp;
1418         if (timestamp0)
1419                 delta = timestamp - timestamp0;
1420         else
1421                 delta = 0;
1422
1423         if (delta < 0)
1424                 die("hm, delta: %" PRIu64 " < 0 ?\n", delta);
1425
1426
1427         sched_out = machine__findnew_thread(machine, switch_event->prev_pid);
1428         sched_in = machine__findnew_thread(machine, switch_event->next_pid);
1429
1430         curr_thread[this_cpu] = sched_in;
1431
1432         printf("  ");
1433
1434         new_shortname = 0;
1435         if (!sched_in->shortname[0]) {
1436                 sched_in->shortname[0] = next_shortname1;
1437                 sched_in->shortname[1] = next_shortname2;
1438
1439                 if (next_shortname1 < 'Z') {
1440                         next_shortname1++;
1441                 } else {
1442                         next_shortname1='A';
1443                         if (next_shortname2 < '9') {
1444                                 next_shortname2++;
1445                         } else {
1446                                 next_shortname2='0';
1447                         }
1448                 }
1449                 new_shortname = 1;
1450         }
1451
1452         for (cpu = 0; cpu <= max_cpu; cpu++) {
1453                 if (cpu != this_cpu)
1454                         printf(" ");
1455                 else
1456                         printf("*");
1457
1458                 if (curr_thread[cpu]) {
1459                         if (curr_thread[cpu]->pid)
1460                                 printf("%2s ", curr_thread[cpu]->shortname);
1461                         else
1462                                 printf(".  ");
1463                 } else
1464                         printf("   ");
1465         }
1466
1467         printf("  %12.6f secs ", (double)timestamp/1e9);
1468         if (new_shortname) {
1469                 printf("%s => %s:%d\n",
1470                         sched_in->shortname, sched_in->comm, sched_in->pid);
1471         } else {
1472                 printf("\n");
1473         }
1474 }
1475
1476 static void
1477 process_sched_switch_event(struct perf_tool *tool __used,
1478                            struct event *event,
1479                            struct perf_sample *sample,
1480                            struct machine *machine,
1481                            struct thread *thread)
1482 {
1483         int this_cpu = sample->cpu;
1484         void *data = sample->raw_data;
1485         struct trace_switch_event switch_event;
1486
1487         FILL_COMMON_FIELDS(switch_event, event, data);
1488
1489         FILL_ARRAY(switch_event, prev_comm, event, data);
1490         FILL_FIELD(switch_event, prev_pid, event, data);
1491         FILL_FIELD(switch_event, prev_prio, event, data);
1492         FILL_FIELD(switch_event, prev_state, event, data);
1493         FILL_ARRAY(switch_event, next_comm, event, data);
1494         FILL_FIELD(switch_event, next_pid, event, data);
1495         FILL_FIELD(switch_event, next_prio, event, data);
1496
1497         if (curr_pid[this_cpu] != (u32)-1) {
1498                 /*
1499                  * Are we trying to switch away a PID that is
1500                  * not current?
1501                  */
1502                 if (curr_pid[this_cpu] != switch_event.prev_pid)
1503                         nr_context_switch_bugs++;
1504         }
1505         if (trace_handler->switch_event)
1506                 trace_handler->switch_event(&switch_event, machine, event,
1507                                             this_cpu, sample->time, thread);
1508
1509         curr_pid[this_cpu] = switch_event.next_pid;
1510 }
1511
1512 static void
1513 process_sched_runtime_event(struct perf_tool *tool __used,
1514                             struct event *event,
1515                             struct perf_sample *sample,
1516                             struct machine *machine,
1517                             struct thread *thread)
1518 {
1519         void *data = sample->raw_data;
1520         struct trace_runtime_event runtime_event;
1521
1522         FILL_ARRAY(runtime_event, comm, event, data);
1523         FILL_FIELD(runtime_event, pid, event, data);
1524         FILL_FIELD(runtime_event, runtime, event, data);
1525         FILL_FIELD(runtime_event, vruntime, event, data);
1526
1527         if (trace_handler->runtime_event)
1528                 trace_handler->runtime_event(&runtime_event, machine, event,
1529                                              sample->cpu, sample->time, thread);
1530 }
1531
1532 static void
1533 process_sched_fork_event(struct perf_tool *tool __used,
1534                          struct event *event,
1535                          struct perf_sample *sample,
1536                          struct machine *machine __used,
1537                          struct thread *thread)
1538 {
1539         void *data = sample->raw_data;
1540         struct trace_fork_event fork_event;
1541
1542         FILL_COMMON_FIELDS(fork_event, event, data);
1543
1544         FILL_ARRAY(fork_event, parent_comm, event, data);
1545         FILL_FIELD(fork_event, parent_pid, event, data);
1546         FILL_ARRAY(fork_event, child_comm, event, data);
1547         FILL_FIELD(fork_event, child_pid, event, data);
1548
1549         if (trace_handler->fork_event)
1550                 trace_handler->fork_event(&fork_event, event,
1551                                           sample->cpu, sample->time, thread);
1552 }
1553
1554 static void
1555 process_sched_exit_event(struct perf_tool *tool __used,
1556                          struct event *event,
1557                          struct perf_sample *sample __used,
1558                          struct machine *machine __used,
1559                          struct thread *thread __used)
1560 {
1561         if (verbose)
1562                 printf("sched_exit event %p\n", event);
1563 }
1564
1565 static void
1566 process_sched_migrate_task_event(struct perf_tool *tool __used,
1567                                  struct event *event,
1568                                  struct perf_sample *sample,
1569                                  struct machine *machine,
1570                                  struct thread *thread)
1571 {
1572         void *data = sample->raw_data;
1573         struct trace_migrate_task_event migrate_task_event;
1574
1575         FILL_COMMON_FIELDS(migrate_task_event, event, data);
1576
1577         FILL_ARRAY(migrate_task_event, comm, event, data);
1578         FILL_FIELD(migrate_task_event, pid, event, data);
1579         FILL_FIELD(migrate_task_event, prio, event, data);
1580         FILL_FIELD(migrate_task_event, cpu, event, data);
1581
1582         if (trace_handler->migrate_task_event)
1583                 trace_handler->migrate_task_event(&migrate_task_event, machine,
1584                                                   event, sample->cpu,
1585                                                   sample->time, thread);
1586 }
1587
1588 typedef void (*tracepoint_handler)(struct perf_tool *tool, struct event *event,
1589                                    struct perf_sample *sample,
1590                                    struct machine *machine,
1591                                    struct thread *thread);
1592
1593 static int perf_sched__process_tracepoint_sample(struct perf_tool *tool,
1594                                                  union perf_event *event __used,
1595                                                  struct perf_sample *sample,
1596                                                  struct perf_evsel *evsel,
1597                                                  struct machine *machine)
1598 {
1599         struct thread *thread = machine__findnew_thread(machine, sample->pid);
1600
1601         if (thread == NULL) {
1602                 pr_debug("problem processing %s event, skipping it.\n",
1603                          evsel->name);
1604                 return -1;
1605         }
1606
1607         evsel->hists.stats.total_period += sample->period;
1608         hists__inc_nr_events(&evsel->hists, PERF_RECORD_SAMPLE);
1609
1610         if (evsel->handler.func != NULL) {
1611                 tracepoint_handler f = evsel->handler.func;
1612
1613                 if (evsel->handler.data == NULL)
1614                         evsel->handler.data = trace_find_event(evsel->attr.config);
1615
1616                 f(tool, evsel->handler.data, sample, machine, thread);
1617         }
1618
1619         return 0;
1620 }
1621
1622 static struct perf_tool perf_sched = {
1623         .sample                 = perf_sched__process_tracepoint_sample,
1624         .comm                   = perf_event__process_comm,
1625         .lost                   = perf_event__process_lost,
1626         .fork                   = perf_event__process_task,
1627         .ordered_samples        = true,
1628 };
1629
1630 static void read_events(bool destroy, struct perf_session **psession)
1631 {
1632         int err = -EINVAL;
1633         const struct perf_evsel_str_handler handlers[] = {
1634                 { "sched:sched_switch",       process_sched_switch_event, },
1635                 { "sched:sched_stat_runtime", process_sched_runtime_event, },
1636                 { "sched:sched_wakeup",       process_sched_wakeup_event, },
1637                 { "sched:sched_wakeup_new",   process_sched_wakeup_event, },
1638                 { "sched:sched_process_fork", process_sched_fork_event, },
1639                 { "sched:sched_process_exit", process_sched_exit_event, },
1640                 { "sched:sched_migrate_task", process_sched_migrate_task_event, },
1641         };
1642         struct perf_session *session = perf_session__new(input_name, O_RDONLY,
1643                                                          0, false, &perf_sched);
1644         if (session == NULL)
1645                 die("No Memory");
1646
1647         err = perf_evlist__set_tracepoints_handlers_array(session->evlist, handlers);
1648         assert(err == 0);
1649
1650         if (perf_session__has_traces(session, "record -R")) {
1651                 err = perf_session__process_events(session, &perf_sched);
1652                 if (err)
1653                         die("Failed to process events, error %d", err);
1654
1655                 nr_events      = session->hists.stats.nr_events[0];
1656                 nr_lost_events = session->hists.stats.total_lost;
1657                 nr_lost_chunks = session->hists.stats.nr_events[PERF_RECORD_LOST];
1658         }
1659
1660         if (destroy)
1661                 perf_session__delete(session);
1662
1663         if (psession)
1664                 *psession = session;
1665 }
1666
1667 static void print_bad_events(void)
1668 {
1669         if (nr_unordered_timestamps && nr_timestamps) {
1670                 printf("  INFO: %.3f%% unordered timestamps (%ld out of %ld)\n",
1671                         (double)nr_unordered_timestamps/(double)nr_timestamps*100.0,
1672                         nr_unordered_timestamps, nr_timestamps);
1673         }
1674         if (nr_lost_events && nr_events) {
1675                 printf("  INFO: %.3f%% lost events (%ld out of %ld, in %ld chunks)\n",
1676                         (double)nr_lost_events/(double)nr_events*100.0,
1677                         nr_lost_events, nr_events, nr_lost_chunks);
1678         }
1679         if (nr_state_machine_bugs && nr_timestamps) {
1680                 printf("  INFO: %.3f%% state machine bugs (%ld out of %ld)",
1681                         (double)nr_state_machine_bugs/(double)nr_timestamps*100.0,
1682                         nr_state_machine_bugs, nr_timestamps);
1683                 if (nr_lost_events)
1684                         printf(" (due to lost events?)");
1685                 printf("\n");
1686         }
1687         if (nr_context_switch_bugs && nr_timestamps) {
1688                 printf("  INFO: %.3f%% context switch bugs (%ld out of %ld)",
1689                         (double)nr_context_switch_bugs/(double)nr_timestamps*100.0,
1690                         nr_context_switch_bugs, nr_timestamps);
1691                 if (nr_lost_events)
1692                         printf(" (due to lost events?)");
1693                 printf("\n");
1694         }
1695 }
1696
1697 static void __cmd_lat(void)
1698 {
1699         struct rb_node *next;
1700         struct perf_session *session;
1701
1702         setup_pager();
1703         read_events(false, &session);
1704         sort_lat();
1705
1706         printf("\n ---------------------------------------------------------------------------------------------------------------\n");
1707         printf("  Task                  |   Runtime ms  | Switches | Average delay ms | Maximum delay ms | Maximum delay at     |\n");
1708         printf(" ---------------------------------------------------------------------------------------------------------------\n");
1709
1710         next = rb_first(&sorted_atom_root);
1711
1712         while (next) {
1713                 struct work_atoms *work_list;
1714
1715                 work_list = rb_entry(next, struct work_atoms, node);
1716                 output_lat_thread(work_list);
1717                 next = rb_next(next);
1718         }
1719
1720         printf(" -----------------------------------------------------------------------------------------\n");
1721         printf("  TOTAL:                |%11.3f ms |%9" PRIu64 " |\n",
1722                 (double)all_runtime/1e6, all_count);
1723
1724         printf(" ---------------------------------------------------\n");
1725
1726         print_bad_events();
1727         printf("\n");
1728
1729         perf_session__delete(session);
1730 }
1731
1732 static struct trace_sched_handler map_ops  = {
1733         .wakeup_event           = NULL,
1734         .switch_event           = map_switch_event,
1735         .runtime_event          = NULL,
1736         .fork_event             = NULL,
1737 };
1738
1739 static void __cmd_map(void)
1740 {
1741         max_cpu = sysconf(_SC_NPROCESSORS_CONF);
1742
1743         setup_pager();
1744         read_events(true, NULL);
1745         print_bad_events();
1746 }
1747
1748 static void __cmd_replay(void)
1749 {
1750         unsigned long i;
1751
1752         calibrate_run_measurement_overhead();
1753         calibrate_sleep_measurement_overhead();
1754
1755         test_calibrations();
1756
1757         read_events(true, NULL);
1758
1759         printf("nr_run_events:        %ld\n", nr_run_events);
1760         printf("nr_sleep_events:      %ld\n", nr_sleep_events);
1761         printf("nr_wakeup_events:     %ld\n", nr_wakeup_events);
1762
1763         if (targetless_wakeups)
1764                 printf("target-less wakeups:  %ld\n", targetless_wakeups);
1765         if (multitarget_wakeups)
1766                 printf("multi-target wakeups: %ld\n", multitarget_wakeups);
1767         if (nr_run_events_optimized)
1768                 printf("run atoms optimized: %ld\n",
1769                         nr_run_events_optimized);
1770
1771         print_task_traces();
1772         add_cross_task_wakeups();
1773
1774         create_tasks();
1775         printf("------------------------------------------------------------\n");
1776         for (i = 0; i < replay_repeat; i++)
1777                 run_one_test();
1778 }
1779
1780
1781 static const char * const sched_usage[] = {
1782         "perf sched [<options>] {record|latency|map|replay|script}",
1783         NULL
1784 };
1785
1786 static const struct option sched_options[] = {
1787         OPT_STRING('i', "input", &input_name, "file",
1788                     "input file name"),
1789         OPT_INCR('v', "verbose", &verbose,
1790                     "be more verbose (show symbol address, etc)"),
1791         OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
1792                     "dump raw trace in ASCII"),
1793         OPT_END()
1794 };
1795
1796 static const char * const latency_usage[] = {
1797         "perf sched latency [<options>]",
1798         NULL
1799 };
1800
1801 static const struct option latency_options[] = {
1802         OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
1803                    "sort by key(s): runtime, switch, avg, max"),
1804         OPT_INCR('v', "verbose", &verbose,
1805                     "be more verbose (show symbol address, etc)"),
1806         OPT_INTEGER('C', "CPU", &profile_cpu,
1807                     "CPU to profile on"),
1808         OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
1809                     "dump raw trace in ASCII"),
1810         OPT_END()
1811 };
1812
1813 static const char * const replay_usage[] = {
1814         "perf sched replay [<options>]",
1815         NULL
1816 };
1817
1818 static const struct option replay_options[] = {
1819         OPT_UINTEGER('r', "repeat", &replay_repeat,
1820                      "repeat the workload replay N times (-1: infinite)"),
1821         OPT_INCR('v', "verbose", &verbose,
1822                     "be more verbose (show symbol address, etc)"),
1823         OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
1824                     "dump raw trace in ASCII"),
1825         OPT_END()
1826 };
1827
1828 static void setup_sorting(void)
1829 {
1830         char *tmp, *tok, *str = strdup(sort_order);
1831
1832         for (tok = strtok_r(str, ", ", &tmp);
1833                         tok; tok = strtok_r(NULL, ", ", &tmp)) {
1834                 if (sort_dimension__add(tok, &sort_list) < 0) {
1835                         error("Unknown --sort key: `%s'", tok);
1836                         usage_with_options(latency_usage, latency_options);
1837                 }
1838         }
1839
1840         free(str);
1841
1842         sort_dimension__add("pid", &cmp_pid);
1843 }
1844
1845 static const char *record_args[] = {
1846         "record",
1847         "-a",
1848         "-R",
1849         "-f",
1850         "-m", "1024",
1851         "-c", "1",
1852         "-e", "sched:sched_switch",
1853         "-e", "sched:sched_stat_wait",
1854         "-e", "sched:sched_stat_sleep",
1855         "-e", "sched:sched_stat_iowait",
1856         "-e", "sched:sched_stat_runtime",
1857         "-e", "sched:sched_process_exit",
1858         "-e", "sched:sched_process_fork",
1859         "-e", "sched:sched_wakeup",
1860         "-e", "sched:sched_migrate_task",
1861 };
1862
1863 static int __cmd_record(int argc, const char **argv)
1864 {
1865         unsigned int rec_argc, i, j;
1866         const char **rec_argv;
1867
1868         rec_argc = ARRAY_SIZE(record_args) + argc - 1;
1869         rec_argv = calloc(rec_argc + 1, sizeof(char *));
1870
1871         if (rec_argv == NULL)
1872                 return -ENOMEM;
1873
1874         for (i = 0; i < ARRAY_SIZE(record_args); i++)
1875                 rec_argv[i] = strdup(record_args[i]);
1876
1877         for (j = 1; j < (unsigned int)argc; j++, i++)
1878                 rec_argv[i] = argv[j];
1879
1880         BUG_ON(i != rec_argc);
1881
1882         return cmd_record(i, rec_argv, NULL);
1883 }
1884
1885 int cmd_sched(int argc, const char **argv, const char *prefix __used)
1886 {
1887         argc = parse_options(argc, argv, sched_options, sched_usage,
1888                              PARSE_OPT_STOP_AT_NON_OPTION);
1889         if (!argc)
1890                 usage_with_options(sched_usage, sched_options);
1891
1892         /*
1893          * Aliased to 'perf script' for now:
1894          */
1895         if (!strcmp(argv[0], "script"))
1896                 return cmd_script(argc, argv, prefix);
1897
1898         symbol__init();
1899         if (!strncmp(argv[0], "rec", 3)) {
1900                 return __cmd_record(argc, argv);
1901         } else if (!strncmp(argv[0], "lat", 3)) {
1902                 trace_handler = &lat_ops;
1903                 if (argc > 1) {
1904                         argc = parse_options(argc, argv, latency_options, latency_usage, 0);
1905                         if (argc)
1906                                 usage_with_options(latency_usage, latency_options);
1907                 }
1908                 setup_sorting();
1909                 __cmd_lat();
1910         } else if (!strcmp(argv[0], "map")) {
1911                 trace_handler = &map_ops;
1912                 setup_sorting();
1913                 __cmd_map();
1914         } else if (!strncmp(argv[0], "rep", 3)) {
1915                 trace_handler = &replay_ops;
1916                 if (argc) {
1917                         argc = parse_options(argc, argv, replay_options, replay_usage, 0);
1918                         if (argc)
1919                                 usage_with_options(replay_usage, replay_options);
1920                 }
1921                 __cmd_replay();
1922         } else {
1923                 usage_with_options(sched_usage, sched_options);
1924         }
1925
1926         return 0;
1927 }