5 #include "util/cache.h"
6 #include "util/symbol.h"
7 #include "util/thread.h"
8 #include "util/header.h"
9 #include "util/session.h"
11 #include "util/parse-options.h"
12 #include "util/trace-event.h"
14 #include "util/debug.h"
16 #include <sys/prctl.h>
18 #include <semaphore.h>
22 static char const *input_name
= "perf.data";
24 static char default_sort_order
[] = "avg, max, switch, runtime";
25 static const char *sort_order
= default_sort_order
;
27 static int profile_cpu
= -1;
29 #define PR_SET_NAME 15 /* Set process name */
32 static u64 run_measurement_overhead
;
33 static u64 sleep_measurement_overhead
;
40 static unsigned long nr_tasks
;
49 unsigned long nr_events
;
50 unsigned long curr_event
;
51 struct sched_atom
**atoms
;
62 enum sched_event_type
{
66 SCHED_EVENT_MIGRATION
,
70 enum sched_event_type type
;
76 struct task_desc
*wakee
;
79 static struct task_desc
*pid_to_task
[MAX_PID
];
81 static struct task_desc
**tasks
;
83 static pthread_mutex_t start_work_mutex
= PTHREAD_MUTEX_INITIALIZER
;
84 static u64 start_time
;
86 static pthread_mutex_t work_done_wait_mutex
= PTHREAD_MUTEX_INITIALIZER
;
88 static unsigned long nr_run_events
;
89 static unsigned long nr_sleep_events
;
90 static unsigned long nr_wakeup_events
;
92 static unsigned long nr_sleep_corrections
;
93 static unsigned long nr_run_events_optimized
;
95 static unsigned long targetless_wakeups
;
96 static unsigned long multitarget_wakeups
;
99 static u64 runavg_cpu_usage
;
100 static u64 parent_cpu_usage
;
101 static u64 runavg_parent_cpu_usage
;
103 static unsigned long nr_runs
;
104 static u64 sum_runtime
;
105 static u64 sum_fluct
;
108 static unsigned int replay_repeat
= 10;
109 static unsigned long nr_timestamps
;
110 static unsigned long nr_unordered_timestamps
;
111 static unsigned long nr_state_machine_bugs
;
112 static unsigned long nr_context_switch_bugs
;
113 static unsigned long nr_events
;
114 static unsigned long nr_lost_chunks
;
115 static unsigned long nr_lost_events
;
117 #define TASK_STATE_TO_CHAR_STR "RSDTtZX"
127 struct list_head list
;
128 enum thread_state state
;
136 struct list_head work_list
;
137 struct thread
*thread
;
146 typedef int (*sort_fn_t
)(struct work_atoms
*, struct work_atoms
*);
148 static struct rb_root atom_root
, sorted_atom_root
;
150 static u64 all_runtime
;
151 static u64 all_count
;
154 static u64
get_nsecs(void)
158 clock_gettime(CLOCK_MONOTONIC
, &ts
);
160 return ts
.tv_sec
* 1000000000ULL + ts
.tv_nsec
;
163 static void burn_nsecs(u64 nsecs
)
165 u64 T0
= get_nsecs(), T1
;
169 } while (T1
+ run_measurement_overhead
< T0
+ nsecs
);
172 static void sleep_nsecs(u64 nsecs
)
176 ts
.tv_nsec
= nsecs
% 999999999;
177 ts
.tv_sec
= nsecs
/ 999999999;
179 nanosleep(&ts
, NULL
);
182 static void calibrate_run_measurement_overhead(void)
184 u64 T0
, T1
, delta
, min_delta
= 1000000000ULL;
187 for (i
= 0; i
< 10; i
++) {
192 min_delta
= min(min_delta
, delta
);
194 run_measurement_overhead
= min_delta
;
196 printf("run measurement overhead: %Ld nsecs\n", min_delta
);
199 static void calibrate_sleep_measurement_overhead(void)
201 u64 T0
, T1
, delta
, min_delta
= 1000000000ULL;
204 for (i
= 0; i
< 10; i
++) {
209 min_delta
= min(min_delta
, delta
);
212 sleep_measurement_overhead
= min_delta
;
214 printf("sleep measurement overhead: %Ld nsecs\n", min_delta
);
217 static struct sched_atom
*
218 get_new_event(struct task_desc
*task
, u64 timestamp
)
220 struct sched_atom
*event
= zalloc(sizeof(*event
));
221 unsigned long idx
= task
->nr_events
;
224 event
->timestamp
= timestamp
;
228 size
= sizeof(struct sched_atom
*) * task
->nr_events
;
229 task
->atoms
= realloc(task
->atoms
, size
);
230 BUG_ON(!task
->atoms
);
232 task
->atoms
[idx
] = event
;
237 static struct sched_atom
*last_event(struct task_desc
*task
)
239 if (!task
->nr_events
)
242 return task
->atoms
[task
->nr_events
- 1];
246 add_sched_event_run(struct task_desc
*task
, u64 timestamp
, u64 duration
)
248 struct sched_atom
*event
, *curr_event
= last_event(task
);
251 * optimize an existing RUN event by merging this one
254 if (curr_event
&& curr_event
->type
== SCHED_EVENT_RUN
) {
255 nr_run_events_optimized
++;
256 curr_event
->duration
+= duration
;
260 event
= get_new_event(task
, timestamp
);
262 event
->type
= SCHED_EVENT_RUN
;
263 event
->duration
= duration
;
269 add_sched_event_wakeup(struct task_desc
*task
, u64 timestamp
,
270 struct task_desc
*wakee
)
272 struct sched_atom
*event
, *wakee_event
;
274 event
= get_new_event(task
, timestamp
);
275 event
->type
= SCHED_EVENT_WAKEUP
;
276 event
->wakee
= wakee
;
278 wakee_event
= last_event(wakee
);
279 if (!wakee_event
|| wakee_event
->type
!= SCHED_EVENT_SLEEP
) {
280 targetless_wakeups
++;
283 if (wakee_event
->wait_sem
) {
284 multitarget_wakeups
++;
288 wakee_event
->wait_sem
= zalloc(sizeof(*wakee_event
->wait_sem
));
289 sem_init(wakee_event
->wait_sem
, 0, 0);
290 wakee_event
->specific_wait
= 1;
291 event
->wait_sem
= wakee_event
->wait_sem
;
297 add_sched_event_sleep(struct task_desc
*task
, u64 timestamp
,
298 u64 task_state __used
)
300 struct sched_atom
*event
= get_new_event(task
, timestamp
);
302 event
->type
= SCHED_EVENT_SLEEP
;
307 static struct task_desc
*register_pid(unsigned long pid
, const char *comm
)
309 struct task_desc
*task
;
311 BUG_ON(pid
>= MAX_PID
);
313 task
= pid_to_task
[pid
];
318 task
= zalloc(sizeof(*task
));
321 strcpy(task
->comm
, comm
);
323 * every task starts in sleeping state - this gets ignored
324 * if there's no wakeup pointing to this sleep state:
326 add_sched_event_sleep(task
, 0, 0);
328 pid_to_task
[pid
] = task
;
330 tasks
= realloc(tasks
, nr_tasks
*sizeof(struct task_task
*));
332 tasks
[task
->nr
] = task
;
335 printf("registered task #%ld, PID %ld (%s)\n", nr_tasks
, pid
, comm
);
341 static void print_task_traces(void)
343 struct task_desc
*task
;
346 for (i
= 0; i
< nr_tasks
; i
++) {
348 printf("task %6ld (%20s:%10ld), nr_events: %ld\n",
349 task
->nr
, task
->comm
, task
->pid
, task
->nr_events
);
353 static void add_cross_task_wakeups(void)
355 struct task_desc
*task1
, *task2
;
358 for (i
= 0; i
< nr_tasks
; i
++) {
364 add_sched_event_wakeup(task1
, 0, task2
);
369 process_sched_event(struct task_desc
*this_task __used
, struct sched_atom
*atom
)
376 delta
= start_time
+ atom
->timestamp
- now
;
378 switch (atom
->type
) {
379 case SCHED_EVENT_RUN
:
380 burn_nsecs(atom
->duration
);
382 case SCHED_EVENT_SLEEP
:
384 ret
= sem_wait(atom
->wait_sem
);
387 case SCHED_EVENT_WAKEUP
:
389 ret
= sem_post(atom
->wait_sem
);
392 case SCHED_EVENT_MIGRATION
:
399 static u64
get_cpu_usage_nsec_parent(void)
405 err
= getrusage(RUSAGE_SELF
, &ru
);
408 sum
= ru
.ru_utime
.tv_sec
*1e9
+ ru
.ru_utime
.tv_usec
*1e3
;
409 sum
+= ru
.ru_stime
.tv_sec
*1e9
+ ru
.ru_stime
.tv_usec
*1e3
;
414 static int self_open_counters(void)
416 struct perf_event_attr attr
;
419 memset(&attr
, 0, sizeof(attr
));
421 attr
.type
= PERF_TYPE_SOFTWARE
;
422 attr
.config
= PERF_COUNT_SW_TASK_CLOCK
;
424 fd
= sys_perf_event_open(&attr
, 0, -1, -1, 0);
427 die("Error: sys_perf_event_open() syscall returned"
428 "with %d (%s)\n", fd
, strerror(errno
));
432 static u64
get_cpu_usage_nsec_self(int fd
)
437 ret
= read(fd
, &runtime
, sizeof(runtime
));
438 BUG_ON(ret
!= sizeof(runtime
));
443 static void *thread_func(void *ctx
)
445 struct task_desc
*this_task
= ctx
;
446 u64 cpu_usage_0
, cpu_usage_1
;
447 unsigned long i
, ret
;
451 sprintf(comm2
, ":%s", this_task
->comm
);
452 prctl(PR_SET_NAME
, comm2
);
453 fd
= self_open_counters();
456 ret
= sem_post(&this_task
->ready_for_work
);
458 ret
= pthread_mutex_lock(&start_work_mutex
);
460 ret
= pthread_mutex_unlock(&start_work_mutex
);
463 cpu_usage_0
= get_cpu_usage_nsec_self(fd
);
465 for (i
= 0; i
< this_task
->nr_events
; i
++) {
466 this_task
->curr_event
= i
;
467 process_sched_event(this_task
, this_task
->atoms
[i
]);
470 cpu_usage_1
= get_cpu_usage_nsec_self(fd
);
471 this_task
->cpu_usage
= cpu_usage_1
- cpu_usage_0
;
472 ret
= sem_post(&this_task
->work_done_sem
);
475 ret
= pthread_mutex_lock(&work_done_wait_mutex
);
477 ret
= pthread_mutex_unlock(&work_done_wait_mutex
);
483 static void create_tasks(void)
485 struct task_desc
*task
;
490 err
= pthread_attr_init(&attr
);
492 err
= pthread_attr_setstacksize(&attr
,
493 (size_t) max(16 * 1024, PTHREAD_STACK_MIN
));
495 err
= pthread_mutex_lock(&start_work_mutex
);
497 err
= pthread_mutex_lock(&work_done_wait_mutex
);
499 for (i
= 0; i
< nr_tasks
; i
++) {
501 sem_init(&task
->sleep_sem
, 0, 0);
502 sem_init(&task
->ready_for_work
, 0, 0);
503 sem_init(&task
->work_done_sem
, 0, 0);
504 task
->curr_event
= 0;
505 err
= pthread_create(&task
->thread
, &attr
, thread_func
, task
);
510 static void wait_for_tasks(void)
512 u64 cpu_usage_0
, cpu_usage_1
;
513 struct task_desc
*task
;
514 unsigned long i
, ret
;
516 start_time
= get_nsecs();
518 pthread_mutex_unlock(&work_done_wait_mutex
);
520 for (i
= 0; i
< nr_tasks
; i
++) {
522 ret
= sem_wait(&task
->ready_for_work
);
524 sem_init(&task
->ready_for_work
, 0, 0);
526 ret
= pthread_mutex_lock(&work_done_wait_mutex
);
529 cpu_usage_0
= get_cpu_usage_nsec_parent();
531 pthread_mutex_unlock(&start_work_mutex
);
533 for (i
= 0; i
< nr_tasks
; i
++) {
535 ret
= sem_wait(&task
->work_done_sem
);
537 sem_init(&task
->work_done_sem
, 0, 0);
538 cpu_usage
+= task
->cpu_usage
;
542 cpu_usage_1
= get_cpu_usage_nsec_parent();
543 if (!runavg_cpu_usage
)
544 runavg_cpu_usage
= cpu_usage
;
545 runavg_cpu_usage
= (runavg_cpu_usage
*9 + cpu_usage
)/10;
547 parent_cpu_usage
= cpu_usage_1
- cpu_usage_0
;
548 if (!runavg_parent_cpu_usage
)
549 runavg_parent_cpu_usage
= parent_cpu_usage
;
550 runavg_parent_cpu_usage
= (runavg_parent_cpu_usage
*9 +
551 parent_cpu_usage
)/10;
553 ret
= pthread_mutex_lock(&start_work_mutex
);
556 for (i
= 0; i
< nr_tasks
; i
++) {
558 sem_init(&task
->sleep_sem
, 0, 0);
559 task
->curr_event
= 0;
563 static void run_one_test(void)
565 u64 T0
, T1
, delta
, avg_delta
, fluct
, std_dev
;
572 sum_runtime
+= delta
;
575 avg_delta
= sum_runtime
/ nr_runs
;
576 if (delta
< avg_delta
)
577 fluct
= avg_delta
- delta
;
579 fluct
= delta
- avg_delta
;
581 std_dev
= sum_fluct
/ nr_runs
/ sqrt(nr_runs
);
584 run_avg
= (run_avg
*9 + delta
)/10;
586 printf("#%-3ld: %0.3f, ",
587 nr_runs
, (double)delta
/1000000.0);
589 printf("ravg: %0.2f, ",
590 (double)run_avg
/1e6
);
592 printf("cpu: %0.2f / %0.2f",
593 (double)cpu_usage
/1e6
, (double)runavg_cpu_usage
/1e6
);
597 * rusage statistics done by the parent, these are less
598 * accurate than the sum_exec_runtime based statistics:
600 printf(" [%0.2f / %0.2f]",
601 (double)parent_cpu_usage
/1e6
,
602 (double)runavg_parent_cpu_usage
/1e6
);
607 if (nr_sleep_corrections
)
608 printf(" (%ld sleep corrections)\n", nr_sleep_corrections
);
609 nr_sleep_corrections
= 0;
612 static void test_calibrations(void)
620 printf("the run test took %Ld nsecs\n", T1
-T0
);
626 printf("the sleep test took %Ld nsecs\n", T1
-T0
);
629 #define FILL_FIELD(ptr, field, event, data) \
630 ptr.field = (typeof(ptr.field)) raw_field_value(event, #field, data)
632 #define FILL_ARRAY(ptr, array, event, data) \
634 void *__array = raw_field_ptr(event, #array, data); \
635 memcpy(ptr.array, __array, sizeof(ptr.array)); \
638 #define FILL_COMMON_FIELDS(ptr, event, data) \
640 FILL_FIELD(ptr, common_type, event, data); \
641 FILL_FIELD(ptr, common_flags, event, data); \
642 FILL_FIELD(ptr, common_preempt_count, event, data); \
643 FILL_FIELD(ptr, common_pid, event, data); \
644 FILL_FIELD(ptr, common_tgid, event, data); \
649 struct trace_switch_event
{
654 u8 common_preempt_count
;
667 struct trace_runtime_event
{
672 u8 common_preempt_count
;
682 struct trace_wakeup_event
{
687 u8 common_preempt_count
;
699 struct trace_fork_event
{
704 u8 common_preempt_count
;
708 char parent_comm
[16];
714 struct trace_migrate_task_event
{
719 u8 common_preempt_count
;
730 struct trace_sched_handler
{
731 void (*switch_event
)(struct trace_switch_event
*,
732 struct perf_session
*,
736 struct thread
*thread
);
738 void (*runtime_event
)(struct trace_runtime_event
*,
739 struct perf_session
*,
743 struct thread
*thread
);
745 void (*wakeup_event
)(struct trace_wakeup_event
*,
746 struct perf_session
*,
750 struct thread
*thread
);
752 void (*fork_event
)(struct trace_fork_event
*,
756 struct thread
*thread
);
758 void (*migrate_task_event
)(struct trace_migrate_task_event
*,
759 struct perf_session
*session
,
763 struct thread
*thread
);
768 replay_wakeup_event(struct trace_wakeup_event
*wakeup_event
,
769 struct perf_session
*session __used
,
772 u64 timestamp __used
,
773 struct thread
*thread __used
)
775 struct task_desc
*waker
, *wakee
;
778 printf("sched_wakeup event %p\n", event
);
780 printf(" ... pid %d woke up %s/%d\n",
781 wakeup_event
->common_pid
,
786 waker
= register_pid(wakeup_event
->common_pid
, "<unknown>");
787 wakee
= register_pid(wakeup_event
->pid
, wakeup_event
->comm
);
789 add_sched_event_wakeup(waker
, timestamp
, wakee
);
792 static u64 cpu_last_switched
[MAX_CPUS
];
795 replay_switch_event(struct trace_switch_event
*switch_event
,
796 struct perf_session
*session __used
,
800 struct thread
*thread __used
)
802 struct task_desc
*prev
, *next
;
807 printf("sched_switch event %p\n", event
);
809 if (cpu
>= MAX_CPUS
|| cpu
< 0)
812 timestamp0
= cpu_last_switched
[cpu
];
814 delta
= timestamp
- timestamp0
;
819 die("hm, delta: %Ld < 0 ?\n", delta
);
822 printf(" ... switch from %s/%d to %s/%d [ran %Ld nsecs]\n",
823 switch_event
->prev_comm
, switch_event
->prev_pid
,
824 switch_event
->next_comm
, switch_event
->next_pid
,
828 prev
= register_pid(switch_event
->prev_pid
, switch_event
->prev_comm
);
829 next
= register_pid(switch_event
->next_pid
, switch_event
->next_comm
);
831 cpu_last_switched
[cpu
] = timestamp
;
833 add_sched_event_run(prev
, timestamp
, delta
);
834 add_sched_event_sleep(prev
, timestamp
, switch_event
->prev_state
);
839 replay_fork_event(struct trace_fork_event
*fork_event
,
842 u64 timestamp __used
,
843 struct thread
*thread __used
)
846 printf("sched_fork event %p\n", event
);
847 printf("... parent: %s/%d\n", fork_event
->parent_comm
, fork_event
->parent_pid
);
848 printf("... child: %s/%d\n", fork_event
->child_comm
, fork_event
->child_pid
);
850 register_pid(fork_event
->parent_pid
, fork_event
->parent_comm
);
851 register_pid(fork_event
->child_pid
, fork_event
->child_comm
);
854 static struct trace_sched_handler replay_ops
= {
855 .wakeup_event
= replay_wakeup_event
,
856 .switch_event
= replay_switch_event
,
857 .fork_event
= replay_fork_event
,
860 struct sort_dimension
{
863 struct list_head list
;
866 static LIST_HEAD(cmp_pid
);
869 thread_lat_cmp(struct list_head
*list
, struct work_atoms
*l
, struct work_atoms
*r
)
871 struct sort_dimension
*sort
;
874 BUG_ON(list_empty(list
));
876 list_for_each_entry(sort
, list
, list
) {
877 ret
= sort
->cmp(l
, r
);
885 static struct work_atoms
*
886 thread_atoms_search(struct rb_root
*root
, struct thread
*thread
,
887 struct list_head
*sort_list
)
889 struct rb_node
*node
= root
->rb_node
;
890 struct work_atoms key
= { .thread
= thread
};
893 struct work_atoms
*atoms
;
896 atoms
= container_of(node
, struct work_atoms
, node
);
898 cmp
= thread_lat_cmp(sort_list
, &key
, atoms
);
900 node
= node
->rb_left
;
902 node
= node
->rb_right
;
904 BUG_ON(thread
!= atoms
->thread
);
912 __thread_latency_insert(struct rb_root
*root
, struct work_atoms
*data
,
913 struct list_head
*sort_list
)
915 struct rb_node
**new = &(root
->rb_node
), *parent
= NULL
;
918 struct work_atoms
*this;
921 this = container_of(*new, struct work_atoms
, node
);
924 cmp
= thread_lat_cmp(sort_list
, data
, this);
927 new = &((*new)->rb_left
);
929 new = &((*new)->rb_right
);
932 rb_link_node(&data
->node
, parent
, new);
933 rb_insert_color(&data
->node
, root
);
936 static void thread_atoms_insert(struct thread
*thread
)
938 struct work_atoms
*atoms
= zalloc(sizeof(*atoms
));
942 atoms
->thread
= thread
;
943 INIT_LIST_HEAD(&atoms
->work_list
);
944 __thread_latency_insert(&atom_root
, atoms
, &cmp_pid
);
948 latency_fork_event(struct trace_fork_event
*fork_event __used
,
949 struct event
*event __used
,
951 u64 timestamp __used
,
952 struct thread
*thread __used
)
954 /* should insert the newcomer */
958 static char sched_out_state(struct trace_switch_event
*switch_event
)
960 const char *str
= TASK_STATE_TO_CHAR_STR
;
962 return str
[switch_event
->prev_state
];
966 add_sched_out_event(struct work_atoms
*atoms
,
970 struct work_atom
*atom
= zalloc(sizeof(*atom
));
974 atom
->sched_out_time
= timestamp
;
976 if (run_state
== 'R') {
977 atom
->state
= THREAD_WAIT_CPU
;
978 atom
->wake_up_time
= atom
->sched_out_time
;
981 list_add_tail(&atom
->list
, &atoms
->work_list
);
985 add_runtime_event(struct work_atoms
*atoms
, u64 delta
, u64 timestamp __used
)
987 struct work_atom
*atom
;
989 BUG_ON(list_empty(&atoms
->work_list
));
991 atom
= list_entry(atoms
->work_list
.prev
, struct work_atom
, list
);
993 atom
->runtime
+= delta
;
994 atoms
->total_runtime
+= delta
;
998 add_sched_in_event(struct work_atoms
*atoms
, u64 timestamp
)
1000 struct work_atom
*atom
;
1003 if (list_empty(&atoms
->work_list
))
1006 atom
= list_entry(atoms
->work_list
.prev
, struct work_atom
, list
);
1008 if (atom
->state
!= THREAD_WAIT_CPU
)
1011 if (timestamp
< atom
->wake_up_time
) {
1012 atom
->state
= THREAD_IGNORE
;
1016 atom
->state
= THREAD_SCHED_IN
;
1017 atom
->sched_in_time
= timestamp
;
1019 delta
= atom
->sched_in_time
- atom
->wake_up_time
;
1020 atoms
->total_lat
+= delta
;
1021 if (delta
> atoms
->max_lat
) {
1022 atoms
->max_lat
= delta
;
1023 atoms
->max_lat_at
= timestamp
;
1029 latency_switch_event(struct trace_switch_event
*switch_event
,
1030 struct perf_session
*session
,
1031 struct event
*event __used
,
1034 struct thread
*thread __used
)
1036 struct work_atoms
*out_events
, *in_events
;
1037 struct thread
*sched_out
, *sched_in
;
1041 BUG_ON(cpu
>= MAX_CPUS
|| cpu
< 0);
1043 timestamp0
= cpu_last_switched
[cpu
];
1044 cpu_last_switched
[cpu
] = timestamp
;
1046 delta
= timestamp
- timestamp0
;
1051 die("hm, delta: %Ld < 0 ?\n", delta
);
1054 sched_out
= perf_session__findnew(session
, switch_event
->prev_pid
);
1055 sched_in
= perf_session__findnew(session
, switch_event
->next_pid
);
1057 out_events
= thread_atoms_search(&atom_root
, sched_out
, &cmp_pid
);
1059 thread_atoms_insert(sched_out
);
1060 out_events
= thread_atoms_search(&atom_root
, sched_out
, &cmp_pid
);
1062 die("out-event: Internal tree error");
1064 add_sched_out_event(out_events
, sched_out_state(switch_event
), timestamp
);
1066 in_events
= thread_atoms_search(&atom_root
, sched_in
, &cmp_pid
);
1068 thread_atoms_insert(sched_in
);
1069 in_events
= thread_atoms_search(&atom_root
, sched_in
, &cmp_pid
);
1071 die("in-event: Internal tree error");
1073 * Take came in we have not heard about yet,
1074 * add in an initial atom in runnable state:
1076 add_sched_out_event(in_events
, 'R', timestamp
);
1078 add_sched_in_event(in_events
, timestamp
);
1082 latency_runtime_event(struct trace_runtime_event
*runtime_event
,
1083 struct perf_session
*session
,
1084 struct event
*event __used
,
1087 struct thread
*this_thread __used
)
1089 struct thread
*thread
= perf_session__findnew(session
, runtime_event
->pid
);
1090 struct work_atoms
*atoms
= thread_atoms_search(&atom_root
, thread
, &cmp_pid
);
1092 BUG_ON(cpu
>= MAX_CPUS
|| cpu
< 0);
1094 thread_atoms_insert(thread
);
1095 atoms
= thread_atoms_search(&atom_root
, thread
, &cmp_pid
);
1097 die("in-event: Internal tree error");
1098 add_sched_out_event(atoms
, 'R', timestamp
);
1101 add_runtime_event(atoms
, runtime_event
->runtime
, timestamp
);
1105 latency_wakeup_event(struct trace_wakeup_event
*wakeup_event
,
1106 struct perf_session
*session
,
1107 struct event
*__event __used
,
1110 struct thread
*thread __used
)
1112 struct work_atoms
*atoms
;
1113 struct work_atom
*atom
;
1114 struct thread
*wakee
;
1116 /* Note for later, it may be interesting to observe the failing cases */
1117 if (!wakeup_event
->success
)
1120 wakee
= perf_session__findnew(session
, wakeup_event
->pid
);
1121 atoms
= thread_atoms_search(&atom_root
, wakee
, &cmp_pid
);
1123 thread_atoms_insert(wakee
);
1124 atoms
= thread_atoms_search(&atom_root
, wakee
, &cmp_pid
);
1126 die("wakeup-event: Internal tree error");
1127 add_sched_out_event(atoms
, 'S', timestamp
);
1130 BUG_ON(list_empty(&atoms
->work_list
));
1132 atom
= list_entry(atoms
->work_list
.prev
, struct work_atom
, list
);
1135 * You WILL be missing events if you've recorded only
1136 * one CPU, or are only looking at only one, so don't
1137 * make useless noise.
1139 if (profile_cpu
== -1 && atom
->state
!= THREAD_SLEEPING
)
1140 nr_state_machine_bugs
++;
1143 if (atom
->sched_out_time
> timestamp
) {
1144 nr_unordered_timestamps
++;
1148 atom
->state
= THREAD_WAIT_CPU
;
1149 atom
->wake_up_time
= timestamp
;
1153 latency_migrate_task_event(struct trace_migrate_task_event
*migrate_task_event
,
1154 struct perf_session
*session
,
1155 struct event
*__event __used
,
1158 struct thread
*thread __used
)
1160 struct work_atoms
*atoms
;
1161 struct work_atom
*atom
;
1162 struct thread
*migrant
;
1165 * Only need to worry about migration when profiling one CPU.
1167 if (profile_cpu
== -1)
1170 migrant
= perf_session__findnew(session
, migrate_task_event
->pid
);
1171 atoms
= thread_atoms_search(&atom_root
, migrant
, &cmp_pid
);
1173 thread_atoms_insert(migrant
);
1174 register_pid(migrant
->pid
, migrant
->comm
);
1175 atoms
= thread_atoms_search(&atom_root
, migrant
, &cmp_pid
);
1177 die("migration-event: Internal tree error");
1178 add_sched_out_event(atoms
, 'R', timestamp
);
1181 BUG_ON(list_empty(&atoms
->work_list
));
1183 atom
= list_entry(atoms
->work_list
.prev
, struct work_atom
, list
);
1184 atom
->sched_in_time
= atom
->sched_out_time
= atom
->wake_up_time
= timestamp
;
1188 if (atom
->sched_out_time
> timestamp
)
1189 nr_unordered_timestamps
++;
1192 static struct trace_sched_handler lat_ops
= {
1193 .wakeup_event
= latency_wakeup_event
,
1194 .switch_event
= latency_switch_event
,
1195 .runtime_event
= latency_runtime_event
,
1196 .fork_event
= latency_fork_event
,
1197 .migrate_task_event
= latency_migrate_task_event
,
1200 static void output_lat_thread(struct work_atoms
*work_list
)
1206 if (!work_list
->nb_atoms
)
1209 * Ignore idle threads:
1211 if (!strcmp(work_list
->thread
->comm
, "swapper"))
1214 all_runtime
+= work_list
->total_runtime
;
1215 all_count
+= work_list
->nb_atoms
;
1217 ret
= printf(" %s:%d ", work_list
->thread
->comm
, work_list
->thread
->pid
);
1219 for (i
= 0; i
< 24 - ret
; i
++)
1222 avg
= work_list
->total_lat
/ work_list
->nb_atoms
;
1224 printf("|%11.3f ms |%9llu | avg:%9.3f ms | max:%9.3f ms | max at: %9.6f s\n",
1225 (double)work_list
->total_runtime
/ 1e6
,
1226 work_list
->nb_atoms
, (double)avg
/ 1e6
,
1227 (double)work_list
->max_lat
/ 1e6
,
1228 (double)work_list
->max_lat_at
/ 1e9
);
1231 static int pid_cmp(struct work_atoms
*l
, struct work_atoms
*r
)
1233 if (l
->thread
->pid
< r
->thread
->pid
)
1235 if (l
->thread
->pid
> r
->thread
->pid
)
1241 static struct sort_dimension pid_sort_dimension
= {
1246 static int avg_cmp(struct work_atoms
*l
, struct work_atoms
*r
)
1256 avgl
= l
->total_lat
/ l
->nb_atoms
;
1257 avgr
= r
->total_lat
/ r
->nb_atoms
;
1267 static struct sort_dimension avg_sort_dimension
= {
1272 static int max_cmp(struct work_atoms
*l
, struct work_atoms
*r
)
1274 if (l
->max_lat
< r
->max_lat
)
1276 if (l
->max_lat
> r
->max_lat
)
1282 static struct sort_dimension max_sort_dimension
= {
1287 static int switch_cmp(struct work_atoms
*l
, struct work_atoms
*r
)
1289 if (l
->nb_atoms
< r
->nb_atoms
)
1291 if (l
->nb_atoms
> r
->nb_atoms
)
1297 static struct sort_dimension switch_sort_dimension
= {
1302 static int runtime_cmp(struct work_atoms
*l
, struct work_atoms
*r
)
1304 if (l
->total_runtime
< r
->total_runtime
)
1306 if (l
->total_runtime
> r
->total_runtime
)
1312 static struct sort_dimension runtime_sort_dimension
= {
1317 static struct sort_dimension
*available_sorts
[] = {
1318 &pid_sort_dimension
,
1319 &avg_sort_dimension
,
1320 &max_sort_dimension
,
1321 &switch_sort_dimension
,
1322 &runtime_sort_dimension
,
1325 #define NB_AVAILABLE_SORTS (int)(sizeof(available_sorts) / sizeof(struct sort_dimension *))
1327 static LIST_HEAD(sort_list
);
1329 static int sort_dimension__add(const char *tok
, struct list_head
*list
)
1333 for (i
= 0; i
< NB_AVAILABLE_SORTS
; i
++) {
1334 if (!strcmp(available_sorts
[i
]->name
, tok
)) {
1335 list_add_tail(&available_sorts
[i
]->list
, list
);
1344 static void setup_sorting(void);
1346 static void sort_lat(void)
1348 struct rb_node
*node
;
1351 struct work_atoms
*data
;
1352 node
= rb_first(&atom_root
);
1356 rb_erase(node
, &atom_root
);
1357 data
= rb_entry(node
, struct work_atoms
, node
);
1358 __thread_latency_insert(&sorted_atom_root
, data
, &sort_list
);
1362 static struct trace_sched_handler
*trace_handler
;
1365 process_sched_wakeup_event(void *data
, struct perf_session
*session
,
1366 struct event
*event
,
1368 u64 timestamp __used
,
1369 struct thread
*thread __used
)
1371 struct trace_wakeup_event wakeup_event
;
1373 FILL_COMMON_FIELDS(wakeup_event
, event
, data
);
1375 FILL_ARRAY(wakeup_event
, comm
, event
, data
);
1376 FILL_FIELD(wakeup_event
, pid
, event
, data
);
1377 FILL_FIELD(wakeup_event
, prio
, event
, data
);
1378 FILL_FIELD(wakeup_event
, success
, event
, data
);
1379 FILL_FIELD(wakeup_event
, cpu
, event
, data
);
1381 if (trace_handler
->wakeup_event
)
1382 trace_handler
->wakeup_event(&wakeup_event
, session
, event
,
1383 cpu
, timestamp
, thread
);
1387 * Track the current task - that way we can know whether there's any
1388 * weird events, such as a task being switched away that is not current.
1392 static u32 curr_pid
[MAX_CPUS
] = { [0 ... MAX_CPUS
-1] = -1 };
1394 static struct thread
*curr_thread
[MAX_CPUS
];
1396 static char next_shortname1
= 'A';
1397 static char next_shortname2
= '0';
1400 map_switch_event(struct trace_switch_event
*switch_event
,
1401 struct perf_session
*session
,
1402 struct event
*event __used
,
1405 struct thread
*thread __used
)
1407 struct thread
*sched_out
, *sched_in
;
1413 BUG_ON(this_cpu
>= MAX_CPUS
|| this_cpu
< 0);
1415 if (this_cpu
> max_cpu
)
1418 timestamp0
= cpu_last_switched
[this_cpu
];
1419 cpu_last_switched
[this_cpu
] = timestamp
;
1421 delta
= timestamp
- timestamp0
;
1426 die("hm, delta: %Ld < 0 ?\n", delta
);
1429 sched_out
= perf_session__findnew(session
, switch_event
->prev_pid
);
1430 sched_in
= perf_session__findnew(session
, switch_event
->next_pid
);
1432 curr_thread
[this_cpu
] = sched_in
;
1437 if (!sched_in
->shortname
[0]) {
1438 sched_in
->shortname
[0] = next_shortname1
;
1439 sched_in
->shortname
[1] = next_shortname2
;
1441 if (next_shortname1
< 'Z') {
1444 next_shortname1
='A';
1445 if (next_shortname2
< '9') {
1448 next_shortname2
='0';
1454 for (cpu
= 0; cpu
<= max_cpu
; cpu
++) {
1455 if (cpu
!= this_cpu
)
1460 if (curr_thread
[cpu
]) {
1461 if (curr_thread
[cpu
]->pid
)
1462 printf("%2s ", curr_thread
[cpu
]->shortname
);
1469 printf(" %12.6f secs ", (double)timestamp
/1e9
);
1470 if (new_shortname
) {
1471 printf("%s => %s:%d\n",
1472 sched_in
->shortname
, sched_in
->comm
, sched_in
->pid
);
1480 process_sched_switch_event(void *data
, struct perf_session
*session
,
1481 struct event
*event
,
1483 u64 timestamp __used
,
1484 struct thread
*thread __used
)
1486 struct trace_switch_event switch_event
;
1488 FILL_COMMON_FIELDS(switch_event
, event
, data
);
1490 FILL_ARRAY(switch_event
, prev_comm
, event
, data
);
1491 FILL_FIELD(switch_event
, prev_pid
, event
, data
);
1492 FILL_FIELD(switch_event
, prev_prio
, event
, data
);
1493 FILL_FIELD(switch_event
, prev_state
, event
, data
);
1494 FILL_ARRAY(switch_event
, next_comm
, event
, data
);
1495 FILL_FIELD(switch_event
, next_pid
, event
, data
);
1496 FILL_FIELD(switch_event
, next_prio
, event
, data
);
1498 if (curr_pid
[this_cpu
] != (u32
)-1) {
1500 * Are we trying to switch away a PID that is
1503 if (curr_pid
[this_cpu
] != switch_event
.prev_pid
)
1504 nr_context_switch_bugs
++;
1506 if (trace_handler
->switch_event
)
1507 trace_handler
->switch_event(&switch_event
, session
, event
,
1508 this_cpu
, timestamp
, thread
);
1510 curr_pid
[this_cpu
] = switch_event
.next_pid
;
1514 process_sched_runtime_event(void *data
, struct perf_session
*session
,
1515 struct event
*event
,
1517 u64 timestamp __used
,
1518 struct thread
*thread __used
)
1520 struct trace_runtime_event runtime_event
;
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
);
1527 if (trace_handler
->runtime_event
)
1528 trace_handler
->runtime_event(&runtime_event
, session
, event
, cpu
, timestamp
, thread
);
1532 process_sched_fork_event(void *data
,
1533 struct event
*event
,
1535 u64 timestamp __used
,
1536 struct thread
*thread __used
)
1538 struct trace_fork_event fork_event
;
1540 FILL_COMMON_FIELDS(fork_event
, event
, data
);
1542 FILL_ARRAY(fork_event
, parent_comm
, event
, data
);
1543 FILL_FIELD(fork_event
, parent_pid
, event
, data
);
1544 FILL_ARRAY(fork_event
, child_comm
, event
, data
);
1545 FILL_FIELD(fork_event
, child_pid
, event
, data
);
1547 if (trace_handler
->fork_event
)
1548 trace_handler
->fork_event(&fork_event
, event
,
1549 cpu
, timestamp
, thread
);
1553 process_sched_exit_event(struct event
*event
,
1555 u64 timestamp __used
,
1556 struct thread
*thread __used
)
1559 printf("sched_exit event %p\n", event
);
1563 process_sched_migrate_task_event(void *data
, struct perf_session
*session
,
1564 struct event
*event
,
1566 u64 timestamp __used
,
1567 struct thread
*thread __used
)
1569 struct trace_migrate_task_event migrate_task_event
;
1571 FILL_COMMON_FIELDS(migrate_task_event
, event
, data
);
1573 FILL_ARRAY(migrate_task_event
, comm
, event
, data
);
1574 FILL_FIELD(migrate_task_event
, pid
, event
, data
);
1575 FILL_FIELD(migrate_task_event
, prio
, event
, data
);
1576 FILL_FIELD(migrate_task_event
, cpu
, event
, data
);
1578 if (trace_handler
->migrate_task_event
)
1579 trace_handler
->migrate_task_event(&migrate_task_event
, session
,
1580 event
, cpu
, timestamp
, thread
);
1584 process_raw_event(event_t
*raw_event __used
, struct perf_session
*session
,
1585 void *data
, int cpu
, u64 timestamp
, struct thread
*thread
)
1587 struct event
*event
;
1591 type
= trace_parse_common_type(data
);
1592 event
= trace_find_event(type
);
1594 if (!strcmp(event
->name
, "sched_switch"))
1595 process_sched_switch_event(data
, session
, event
, cpu
, timestamp
, thread
);
1596 if (!strcmp(event
->name
, "sched_stat_runtime"))
1597 process_sched_runtime_event(data
, session
, event
, cpu
, timestamp
, thread
);
1598 if (!strcmp(event
->name
, "sched_wakeup"))
1599 process_sched_wakeup_event(data
, session
, event
, cpu
, timestamp
, thread
);
1600 if (!strcmp(event
->name
, "sched_wakeup_new"))
1601 process_sched_wakeup_event(data
, session
, event
, cpu
, timestamp
, thread
);
1602 if (!strcmp(event
->name
, "sched_process_fork"))
1603 process_sched_fork_event(data
, event
, cpu
, timestamp
, thread
);
1604 if (!strcmp(event
->name
, "sched_process_exit"))
1605 process_sched_exit_event(event
, cpu
, timestamp
, thread
);
1606 if (!strcmp(event
->name
, "sched_migrate_task"))
1607 process_sched_migrate_task_event(data
, session
, event
, cpu
, timestamp
, thread
);
1610 static int process_sample_event(event_t
*event
, struct sample_data
*sample
,
1611 struct perf_session
*session
)
1613 struct thread
*thread
;
1615 if (!(session
->sample_type
& PERF_SAMPLE_RAW
))
1618 thread
= perf_session__findnew(session
, sample
->pid
);
1619 if (thread
== NULL
) {
1620 pr_debug("problem processing %d event, skipping it.\n",
1621 event
->header
.type
);
1625 dump_printf(" ... thread: %s:%d\n", thread
->comm
, thread
->pid
);
1627 if (profile_cpu
!= -1 && profile_cpu
!= (int)sample
->cpu
)
1630 process_raw_event(event
, session
, sample
->raw_data
, sample
->cpu
,
1631 sample
->time
, thread
);
1636 static struct perf_event_ops event_ops
= {
1637 .sample
= process_sample_event
,
1638 .comm
= event__process_comm
,
1639 .lost
= event__process_lost
,
1640 .fork
= event__process_task
,
1641 .ordered_samples
= true,
1644 static int read_events(void)
1647 struct perf_session
*session
= perf_session__new(input_name
, O_RDONLY
,
1648 0, false, &event_ops
);
1649 if (session
== NULL
)
1652 if (perf_session__has_traces(session
, "record -R")) {
1653 err
= perf_session__process_events(session
, &event_ops
);
1654 nr_events
= session
->hists
.stats
.nr_events
[0];
1655 nr_lost_events
= session
->hists
.stats
.total_lost
;
1656 nr_lost_chunks
= session
->hists
.stats
.nr_events
[PERF_RECORD_LOST
];
1659 perf_session__delete(session
);
1663 static void print_bad_events(void)
1665 if (nr_unordered_timestamps
&& nr_timestamps
) {
1666 printf(" INFO: %.3f%% unordered timestamps (%ld out of %ld)\n",
1667 (double)nr_unordered_timestamps
/(double)nr_timestamps
*100.0,
1668 nr_unordered_timestamps
, nr_timestamps
);
1670 if (nr_lost_events
&& nr_events
) {
1671 printf(" INFO: %.3f%% lost events (%ld out of %ld, in %ld chunks)\n",
1672 (double)nr_lost_events
/(double)nr_events
*100.0,
1673 nr_lost_events
, nr_events
, nr_lost_chunks
);
1675 if (nr_state_machine_bugs
&& nr_timestamps
) {
1676 printf(" INFO: %.3f%% state machine bugs (%ld out of %ld)",
1677 (double)nr_state_machine_bugs
/(double)nr_timestamps
*100.0,
1678 nr_state_machine_bugs
, nr_timestamps
);
1680 printf(" (due to lost events?)");
1683 if (nr_context_switch_bugs
&& nr_timestamps
) {
1684 printf(" INFO: %.3f%% context switch bugs (%ld out of %ld)",
1685 (double)nr_context_switch_bugs
/(double)nr_timestamps
*100.0,
1686 nr_context_switch_bugs
, nr_timestamps
);
1688 printf(" (due to lost events?)");
1693 static void __cmd_lat(void)
1695 struct rb_node
*next
;
1701 printf("\n ---------------------------------------------------------------------------------------------------------------\n");
1702 printf(" Task | Runtime ms | Switches | Average delay ms | Maximum delay ms | Maximum delay at |\n");
1703 printf(" ---------------------------------------------------------------------------------------------------------------\n");
1705 next
= rb_first(&sorted_atom_root
);
1708 struct work_atoms
*work_list
;
1710 work_list
= rb_entry(next
, struct work_atoms
, node
);
1711 output_lat_thread(work_list
);
1712 next
= rb_next(next
);
1715 printf(" -----------------------------------------------------------------------------------------\n");
1716 printf(" TOTAL: |%11.3f ms |%9Ld |\n",
1717 (double)all_runtime
/1e6
, all_count
);
1719 printf(" ---------------------------------------------------\n");
1726 static struct trace_sched_handler map_ops
= {
1727 .wakeup_event
= NULL
,
1728 .switch_event
= map_switch_event
,
1729 .runtime_event
= NULL
,
1733 static void __cmd_map(void)
1735 max_cpu
= sysconf(_SC_NPROCESSORS_CONF
);
1742 static void __cmd_replay(void)
1746 calibrate_run_measurement_overhead();
1747 calibrate_sleep_measurement_overhead();
1749 test_calibrations();
1753 printf("nr_run_events: %ld\n", nr_run_events
);
1754 printf("nr_sleep_events: %ld\n", nr_sleep_events
);
1755 printf("nr_wakeup_events: %ld\n", nr_wakeup_events
);
1757 if (targetless_wakeups
)
1758 printf("target-less wakeups: %ld\n", targetless_wakeups
);
1759 if (multitarget_wakeups
)
1760 printf("multi-target wakeups: %ld\n", multitarget_wakeups
);
1761 if (nr_run_events_optimized
)
1762 printf("run atoms optimized: %ld\n",
1763 nr_run_events_optimized
);
1765 print_task_traces();
1766 add_cross_task_wakeups();
1769 printf("------------------------------------------------------------\n");
1770 for (i
= 0; i
< replay_repeat
; i
++)
1775 static const char * const sched_usage
[] = {
1776 "perf sched [<options>] {record|latency|map|replay|trace}",
1780 static const struct option sched_options
[] = {
1781 OPT_STRING('i', "input", &input_name
, "file",
1783 OPT_INCR('v', "verbose", &verbose
,
1784 "be more verbose (show symbol address, etc)"),
1785 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace
,
1786 "dump raw trace in ASCII"),
1790 static const char * const latency_usage
[] = {
1791 "perf sched latency [<options>]",
1795 static const struct option latency_options
[] = {
1796 OPT_STRING('s', "sort", &sort_order
, "key[,key2...]",
1797 "sort by key(s): runtime, switch, avg, max"),
1798 OPT_INCR('v', "verbose", &verbose
,
1799 "be more verbose (show symbol address, etc)"),
1800 OPT_INTEGER('C', "CPU", &profile_cpu
,
1801 "CPU to profile on"),
1802 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace
,
1803 "dump raw trace in ASCII"),
1807 static const char * const replay_usage
[] = {
1808 "perf sched replay [<options>]",
1812 static const struct option replay_options
[] = {
1813 OPT_UINTEGER('r', "repeat", &replay_repeat
,
1814 "repeat the workload replay N times (-1: infinite)"),
1815 OPT_INCR('v', "verbose", &verbose
,
1816 "be more verbose (show symbol address, etc)"),
1817 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace
,
1818 "dump raw trace in ASCII"),
1822 static void setup_sorting(void)
1824 char *tmp
, *tok
, *str
= strdup(sort_order
);
1826 for (tok
= strtok_r(str
, ", ", &tmp
);
1827 tok
; tok
= strtok_r(NULL
, ", ", &tmp
)) {
1828 if (sort_dimension__add(tok
, &sort_list
) < 0) {
1829 error("Unknown --sort key: `%s'", tok
);
1830 usage_with_options(latency_usage
, latency_options
);
1836 sort_dimension__add("pid", &cmp_pid
);
1839 static const char *record_args
[] = {
1846 "-e", "sched:sched_switch",
1847 "-e", "sched:sched_stat_wait",
1848 "-e", "sched:sched_stat_sleep",
1849 "-e", "sched:sched_stat_iowait",
1850 "-e", "sched:sched_stat_runtime",
1851 "-e", "sched:sched_process_exit",
1852 "-e", "sched:sched_process_fork",
1853 "-e", "sched:sched_wakeup",
1854 "-e", "sched:sched_migrate_task",
1857 static int __cmd_record(int argc
, const char **argv
)
1859 unsigned int rec_argc
, i
, j
;
1860 const char **rec_argv
;
1862 rec_argc
= ARRAY_SIZE(record_args
) + argc
- 1;
1863 rec_argv
= calloc(rec_argc
+ 1, sizeof(char *));
1865 if (rec_argv
== NULL
)
1868 for (i
= 0; i
< ARRAY_SIZE(record_args
); i
++)
1869 rec_argv
[i
] = strdup(record_args
[i
]);
1871 for (j
= 1; j
< (unsigned int)argc
; j
++, i
++)
1872 rec_argv
[i
] = argv
[j
];
1874 BUG_ON(i
!= rec_argc
);
1876 return cmd_record(i
, rec_argv
, NULL
);
1879 int cmd_sched(int argc
, const char **argv
, const char *prefix __used
)
1881 argc
= parse_options(argc
, argv
, sched_options
, sched_usage
,
1882 PARSE_OPT_STOP_AT_NON_OPTION
);
1884 usage_with_options(sched_usage
, sched_options
);
1887 * Aliased to 'perf script' for now:
1889 if (!strcmp(argv
[0], "script"))
1890 return cmd_script(argc
, argv
, prefix
);
1893 if (!strncmp(argv
[0], "rec", 3)) {
1894 return __cmd_record(argc
, argv
);
1895 } else if (!strncmp(argv
[0], "lat", 3)) {
1896 trace_handler
= &lat_ops
;
1898 argc
= parse_options(argc
, argv
, latency_options
, latency_usage
, 0);
1900 usage_with_options(latency_usage
, latency_options
);
1904 } else if (!strcmp(argv
[0], "map")) {
1905 trace_handler
= &map_ops
;
1908 } else if (!strncmp(argv
[0], "rep", 3)) {
1909 trace_handler
= &replay_ops
;
1911 argc
= parse_options(argc
, argv
, replay_options
, replay_usage
, 0);
1913 usage_with_options(replay_usage
, replay_options
);
1917 usage_with_options(sched_usage
, sched_options
);