4 #include "util/cache.h"
5 #include "util/symbol.h"
6 #include "util/thread.h"
7 #include "util/header.h"
9 #include "util/parse-options.h"
12 #include "util/debug.h"
14 #include "util/trace-event.h"
15 #include <sys/types.h>
20 static char const *input_name = "perf.data";
22 static unsigned long page_size;
23 static unsigned long mmap_window = 32;
25 static unsigned long total_comm = 0;
27 static struct rb_root threads;
28 static struct thread *last_match;
30 static struct perf_header *header;
31 static u64 sample_type;
33 static int replay_mode;
38 * Scheduler benchmarks
40 #include <sys/resource.h>
41 #include <sys/types.h>
44 #include <sys/prctl.h>
46 #include <linux/unistd.h>
48 #include <semaphore.h>
62 #define PR_SET_NAME 15 /* Set process name */
64 #define BUG_ON(x) assert(!(x))
68 typedef unsigned long long nsec_t;
70 static nsec_t run_measurement_overhead;
71 static nsec_t sleep_measurement_overhead;
73 static nsec_t get_nsecs(void)
77 clock_gettime(CLOCK_MONOTONIC, &ts);
79 return ts.tv_sec * 1000000000ULL + ts.tv_nsec;
82 static void burn_nsecs(nsec_t nsecs)
84 nsec_t T0 = get_nsecs(), T1;
88 } while (T1 + run_measurement_overhead < T0 + nsecs);
91 static void sleep_nsecs(nsec_t nsecs)
95 ts.tv_nsec = nsecs % 999999999;
96 ts.tv_sec = nsecs / 999999999;
101 static void calibrate_run_measurement_overhead(void)
103 nsec_t T0, T1, delta, min_delta = 1000000000ULL;
106 for (i = 0; i < 10; i++) {
111 min_delta = min(min_delta, delta);
113 run_measurement_overhead = min_delta;
115 printf("run measurement overhead: %Ld nsecs\n", min_delta);
118 static void calibrate_sleep_measurement_overhead(void)
120 nsec_t T0, T1, delta, min_delta = 1000000000ULL;
123 for (i = 0; i < 10; i++) {
128 min_delta = min(min_delta, delta);
131 sleep_measurement_overhead = min_delta;
133 printf("sleep measurement overhead: %Ld nsecs\n", min_delta);
139 #define MAX_PID 65536
141 static unsigned long nr_tasks;
150 unsigned long nr_events;
151 unsigned long curr_event;
152 struct sched_event **events;
157 sem_t ready_for_work;
163 enum sched_event_type {
170 enum sched_event_type type;
176 struct task_desc *wakee;
179 static struct task_desc *pid_to_task[MAX_PID];
181 static struct task_desc **tasks;
183 static pthread_mutex_t start_work_mutex = PTHREAD_MUTEX_INITIALIZER;
184 static nsec_t start_time;
186 static pthread_mutex_t work_done_wait_mutex = PTHREAD_MUTEX_INITIALIZER;
188 static unsigned long nr_run_events;
189 static unsigned long nr_sleep_events;
190 static unsigned long nr_wakeup_events;
192 static unsigned long nr_sleep_corrections;
193 static unsigned long nr_run_events_optimized;
195 static struct sched_event *
196 get_new_event(struct task_desc *task, nsec_t timestamp)
198 struct sched_event *event = calloc(1, sizeof(*event));
199 unsigned long idx = task->nr_events;
202 event->timestamp = timestamp;
206 size = sizeof(struct sched_event *) * task->nr_events;
207 task->events = realloc(task->events, size);
208 BUG_ON(!task->events);
210 task->events[idx] = event;
215 static struct sched_event *last_event(struct task_desc *task)
217 if (!task->nr_events)
220 return task->events[task->nr_events - 1];
224 add_sched_event_run(struct task_desc *task, nsec_t timestamp, u64 duration)
226 struct sched_event *event, *curr_event = last_event(task);
229 * optimize an existing RUN event by merging this one
232 if (curr_event && curr_event->type == SCHED_EVENT_RUN) {
233 nr_run_events_optimized++;
234 curr_event->duration += duration;
238 event = get_new_event(task, timestamp);
240 event->type = SCHED_EVENT_RUN;
241 event->duration = duration;
246 static unsigned long targetless_wakeups;
247 static unsigned long multitarget_wakeups;
250 add_sched_event_wakeup(struct task_desc *task, nsec_t timestamp,
251 struct task_desc *wakee)
253 struct sched_event *event, *wakee_event;
255 event = get_new_event(task, timestamp);
256 event->type = SCHED_EVENT_WAKEUP;
257 event->wakee = wakee;
259 wakee_event = last_event(wakee);
260 if (!wakee_event || wakee_event->type != SCHED_EVENT_SLEEP) {
261 targetless_wakeups++;
264 if (wakee_event->wait_sem) {
265 multitarget_wakeups++;
269 wakee_event->wait_sem = calloc(1, sizeof(*wakee_event->wait_sem));
270 sem_init(wakee_event->wait_sem, 0, 0);
271 wakee_event->specific_wait = 1;
272 event->wait_sem = wakee_event->wait_sem;
278 add_sched_event_sleep(struct task_desc *task, nsec_t timestamp,
279 u64 task_state __used)
281 struct sched_event *event = get_new_event(task, timestamp);
283 event->type = SCHED_EVENT_SLEEP;
288 static struct task_desc *register_pid(unsigned long pid, const char *comm)
290 struct task_desc *task;
292 BUG_ON(pid >= MAX_PID);
294 task = pid_to_task[pid];
299 task = calloc(1, sizeof(*task));
302 strcpy(task->comm, comm);
304 * every task starts in sleeping state - this gets ignored
305 * if there's no wakeup pointing to this sleep state:
307 add_sched_event_sleep(task, 0, 0);
309 pid_to_task[pid] = task;
311 tasks = realloc(tasks, nr_tasks*sizeof(struct task_task *));
313 tasks[task->nr] = task;
316 printf("registered task #%ld, PID %ld (%s)\n", nr_tasks, pid, comm);
322 static void print_task_traces(void)
324 struct task_desc *task;
327 for (i = 0; i < nr_tasks; i++) {
329 printf("task %6ld (%20s:%10ld), nr_events: %ld\n",
330 task->nr, task->comm, task->pid, task->nr_events);
334 static void add_cross_task_wakeups(void)
336 struct task_desc *task1, *task2;
339 for (i = 0; i < nr_tasks; i++) {
345 add_sched_event_wakeup(task1, 0, task2);
350 process_sched_event(struct task_desc *this_task __used, struct sched_event *event)
357 delta = start_time + event->timestamp - now;
359 switch (event->type) {
360 case SCHED_EVENT_RUN:
361 burn_nsecs(event->duration);
363 case SCHED_EVENT_SLEEP:
365 ret = sem_wait(event->wait_sem);
368 case SCHED_EVENT_WAKEUP:
370 ret = sem_post(event->wait_sem);
378 static nsec_t get_cpu_usage_nsec_parent(void)
384 err = getrusage(RUSAGE_SELF, &ru);
387 sum = ru.ru_utime.tv_sec*1e9 + ru.ru_utime.tv_usec*1e3;
388 sum += ru.ru_stime.tv_sec*1e9 + ru.ru_stime.tv_usec*1e3;
393 static nsec_t get_cpu_usage_nsec_self(void)
395 char filename [] = "/proc/1234567890/sched";
396 unsigned long msecs, nsecs;
404 sprintf(filename, "/proc/%d/sched", getpid());
405 file = fopen(filename, "r");
408 while ((chars = getline(&line, &len, file)) != -1) {
409 ret = sscanf(line, "se.sum_exec_runtime : %ld.%06ld\n",
412 total = msecs*1e6 + nsecs;
423 static void *thread_func(void *ctx)
425 struct task_desc *this_task = ctx;
426 nsec_t cpu_usage_0, cpu_usage_1;
427 unsigned long i, ret;
430 sprintf(comm2, ":%s", this_task->comm);
431 prctl(PR_SET_NAME, comm2);
434 ret = sem_post(&this_task->ready_for_work);
436 ret = pthread_mutex_lock(&start_work_mutex);
438 ret = pthread_mutex_unlock(&start_work_mutex);
441 cpu_usage_0 = get_cpu_usage_nsec_self();
443 for (i = 0; i < this_task->nr_events; i++) {
444 this_task->curr_event = i;
445 process_sched_event(this_task, this_task->events[i]);
448 cpu_usage_1 = get_cpu_usage_nsec_self();
449 this_task->cpu_usage = cpu_usage_1 - cpu_usage_0;
451 ret = sem_post(&this_task->work_done_sem);
454 ret = pthread_mutex_lock(&work_done_wait_mutex);
456 ret = pthread_mutex_unlock(&work_done_wait_mutex);
462 static void create_tasks(void)
464 struct task_desc *task;
469 err = pthread_attr_init(&attr);
471 err = pthread_attr_setstacksize(&attr, (size_t)(16*1024));
473 err = pthread_mutex_lock(&start_work_mutex);
475 err = pthread_mutex_lock(&work_done_wait_mutex);
477 for (i = 0; i < nr_tasks; i++) {
479 sem_init(&task->sleep_sem, 0, 0);
480 sem_init(&task->ready_for_work, 0, 0);
481 sem_init(&task->work_done_sem, 0, 0);
482 task->curr_event = 0;
483 err = pthread_create(&task->thread, &attr, thread_func, task);
488 static nsec_t cpu_usage;
489 static nsec_t runavg_cpu_usage;
490 static nsec_t parent_cpu_usage;
491 static nsec_t runavg_parent_cpu_usage;
493 static void wait_for_tasks(void)
495 nsec_t cpu_usage_0, cpu_usage_1;
496 struct task_desc *task;
497 unsigned long i, ret;
499 start_time = get_nsecs();
501 pthread_mutex_unlock(&work_done_wait_mutex);
503 for (i = 0; i < nr_tasks; i++) {
505 ret = sem_wait(&task->ready_for_work);
507 sem_init(&task->ready_for_work, 0, 0);
509 ret = pthread_mutex_lock(&work_done_wait_mutex);
512 cpu_usage_0 = get_cpu_usage_nsec_parent();
514 pthread_mutex_unlock(&start_work_mutex);
516 for (i = 0; i < nr_tasks; i++) {
518 ret = sem_wait(&task->work_done_sem);
520 sem_init(&task->work_done_sem, 0, 0);
521 cpu_usage += task->cpu_usage;
525 cpu_usage_1 = get_cpu_usage_nsec_parent();
526 if (!runavg_cpu_usage)
527 runavg_cpu_usage = cpu_usage;
528 runavg_cpu_usage = (runavg_cpu_usage*9 + cpu_usage)/10;
530 parent_cpu_usage = cpu_usage_1 - cpu_usage_0;
531 if (!runavg_parent_cpu_usage)
532 runavg_parent_cpu_usage = parent_cpu_usage;
533 runavg_parent_cpu_usage = (runavg_parent_cpu_usage*9 +
534 parent_cpu_usage)/10;
536 ret = pthread_mutex_lock(&start_work_mutex);
539 for (i = 0; i < nr_tasks; i++) {
541 sem_init(&task->sleep_sem, 0, 0);
542 task->curr_event = 0;
546 static int read_events(void);
548 static unsigned long nr_runs;
549 static nsec_t sum_runtime;
550 static nsec_t sum_fluct;
551 static nsec_t run_avg;
553 static void run_one_test(void)
555 nsec_t T0, T1, delta, avg_delta, fluct, std_dev;
562 sum_runtime += delta;
565 avg_delta = sum_runtime / nr_runs;
566 if (delta < avg_delta)
567 fluct = avg_delta - delta;
569 fluct = delta - avg_delta;
571 std_dev = sum_fluct / nr_runs / sqrt(nr_runs);
574 run_avg = (run_avg*9 + delta)/10;
576 printf("#%-3ld: %0.3f, ",
577 nr_runs, (double)delta/1000000.0);
580 printf("%0.2f +- %0.2f, ",
581 (double)avg_delta/1e6, (double)std_dev/1e6);
583 printf("ravg: %0.2f, ",
584 (double)run_avg/1e6);
586 printf("cpu: %0.2f / %0.2f",
587 (double)cpu_usage/1e6, (double)runavg_cpu_usage/1e6);
591 * rusage statistics done by the parent, these are less
592 * accurate than the sum_exec_runtime based statistics:
594 printf(" [%0.2f / %0.2f]",
595 (double)parent_cpu_usage/1e6,
596 (double)runavg_parent_cpu_usage/1e6);
601 if (nr_sleep_corrections)
602 printf(" (%ld sleep corrections)\n", nr_sleep_corrections);
603 nr_sleep_corrections = 0;
606 static void test_calibrations(void)
614 printf("the run test took %Ld nsecs\n", T1-T0);
620 printf("the sleep test took %Ld nsecs\n", T1-T0);
623 static void __cmd_replay(void)
625 long nr_iterations = 10, i;
627 calibrate_run_measurement_overhead();
628 calibrate_sleep_measurement_overhead();
634 printf("nr_run_events: %ld\n", nr_run_events);
635 printf("nr_sleep_events: %ld\n", nr_sleep_events);
636 printf("nr_wakeup_events: %ld\n", nr_wakeup_events);
638 if (targetless_wakeups)
639 printf("target-less wakeups: %ld\n", targetless_wakeups);
640 if (multitarget_wakeups)
641 printf("multi-target wakeups: %ld\n", multitarget_wakeups);
642 if (nr_run_events_optimized)
643 printf("run events optimized: %ld\n",
644 nr_run_events_optimized);
647 add_cross_task_wakeups();
650 printf("------------------------------------------------------------\n");
651 for (i = 0; i < nr_iterations; i++)
656 process_comm_event(event_t *event, unsigned long offset, unsigned long head)
658 struct thread *thread;
660 thread = threads__findnew(event->comm.pid, &threads, &last_match);
662 dump_printf("%p [%p]: PERF_EVENT_COMM: %s:%d\n",
663 (void *)(offset + head),
664 (void *)(long)(event->header.size),
665 event->comm.comm, event->comm.pid);
667 if (thread == NULL ||
668 thread__set_comm(thread, event->comm.comm)) {
669 dump_printf("problem processing PERF_EVENT_COMM, skipping event.\n");
678 struct raw_event_sample {
683 #define FILL_FIELD(ptr, field, event, data) \
684 ptr.field = (typeof(ptr.field)) raw_field_value(event, #field, data)
686 #define FILL_ARRAY(ptr, array, event, data) \
688 void *__array = raw_field_ptr(event, #array, data); \
689 memcpy(ptr.array, __array, sizeof(ptr.array)); \
692 #define FILL_COMMON_FIELDS(ptr, event, data) \
694 FILL_FIELD(ptr, common_type, event, data); \
695 FILL_FIELD(ptr, common_flags, event, data); \
696 FILL_FIELD(ptr, common_preempt_count, event, data); \
697 FILL_FIELD(ptr, common_pid, event, data); \
698 FILL_FIELD(ptr, common_tgid, event, data); \
703 struct trace_switch_event {
708 u8 common_preempt_count;
722 struct trace_wakeup_event {
727 u8 common_preempt_count;
739 struct trace_fork_event {
744 u8 common_preempt_count;
748 char parent_comm[16];
754 struct trace_sched_handler {
755 void (*switch_event)(struct trace_switch_event *,
759 struct thread *thread);
761 void (*wakeup_event)(struct trace_wakeup_event *,
765 struct thread *thread);
767 void (*fork_event)(struct trace_fork_event *,
771 struct thread *thread);
776 replay_wakeup_event(struct trace_wakeup_event *wakeup_event,
779 u64 timestamp __used,
780 struct thread *thread __used)
782 struct task_desc *waker, *wakee;
785 printf("sched_wakeup event %p\n", event);
787 printf(" ... pid %d woke up %s/%d\n",
788 wakeup_event->common_pid,
793 waker = register_pid(wakeup_event->common_pid, "<unknown>");
794 wakee = register_pid(wakeup_event->pid, wakeup_event->comm);
796 add_sched_event_wakeup(waker, timestamp, wakee);
799 static unsigned long cpu_last_switched[MAX_CPUS];
802 replay_switch_event(struct trace_switch_event *switch_event,
806 struct thread *thread __used)
808 struct task_desc *prev, *next;
813 printf("sched_switch event %p\n", event);
815 if (cpu >= MAX_CPUS || cpu < 0)
818 timestamp0 = cpu_last_switched[cpu];
820 delta = timestamp - timestamp0;
825 die("hm, delta: %Ld < 0 ?\n", delta);
828 printf(" ... switch from %s/%d to %s/%d [ran %Ld nsecs]\n",
829 switch_event->prev_comm, switch_event->prev_pid,
830 switch_event->next_comm, switch_event->next_pid,
834 prev = register_pid(switch_event->prev_pid, switch_event->prev_comm);
835 next = register_pid(switch_event->next_pid, switch_event->next_comm);
837 cpu_last_switched[cpu] = timestamp;
839 add_sched_event_run(prev, timestamp, delta);
840 add_sched_event_sleep(prev, timestamp, switch_event->prev_state);
845 replay_fork_event(struct trace_fork_event *fork_event,
848 u64 timestamp __used,
849 struct thread *thread __used)
852 printf("sched_fork event %p\n", event);
853 printf("... parent: %s/%d\n", fork_event->parent_comm, fork_event->parent_pid);
854 printf("... child: %s/%d\n", fork_event->child_comm, fork_event->child_pid);
856 register_pid(fork_event->parent_pid, fork_event->parent_comm);
857 register_pid(fork_event->child_pid, fork_event->child_comm);
860 static struct trace_sched_handler replay_ops = {
861 .wakeup_event = replay_wakeup_event,
862 .switch_event = replay_switch_event,
863 .fork_event = replay_fork_event,
866 #define TASK_STATE_TO_CHAR_STR "RSDTtZX"
876 struct list_head list;
877 enum thread_state state;
884 struct list_head snapshot_list;
885 struct thread *thread;
893 static struct rb_root lat_snapshot_root;
895 static struct task_atoms *
896 thread_atom_list_search(struct rb_root *root, struct thread *thread)
898 struct rb_node *node = root->rb_node;
901 struct task_atoms *atoms;
903 atoms = container_of(node, struct task_atoms, node);
904 if (thread->pid < atoms->thread->pid)
905 node = node->rb_left;
906 else if (thread->pid > atoms->thread->pid)
907 node = node->rb_right;
916 __thread_latency_insert(struct rb_root *root, struct task_atoms *data)
918 struct rb_node **new = &(root->rb_node), *parent = NULL;
921 struct task_atoms *this;
923 this = container_of(*new, struct task_atoms, node);
925 if (data->thread->pid < this->thread->pid)
926 new = &((*new)->rb_left);
927 else if (data->thread->pid > this->thread->pid)
928 new = &((*new)->rb_right);
930 die("Double thread insertion\n");
933 rb_link_node(&data->node, parent, new);
934 rb_insert_color(&data->node, root);
937 static void thread_atom_list_insert(struct thread *thread)
939 struct task_atoms *atoms;
940 atoms = calloc(sizeof(*atoms), 1);
944 atoms->thread = thread;
945 INIT_LIST_HEAD(&atoms->snapshot_list);
946 __thread_latency_insert(&lat_snapshot_root, atoms);
950 latency_fork_event(struct trace_fork_event *fork_event __used,
951 struct event *event __used,
953 u64 timestamp __used,
954 struct thread *thread __used)
956 /* should insert the newcomer */
960 static char sched_out_state(struct trace_switch_event *switch_event)
962 const char *str = TASK_STATE_TO_CHAR_STR;
964 return str[switch_event->prev_state];
968 lat_sched_out(struct task_atoms *atoms,
969 struct trace_switch_event *switch_event __used,
973 struct work_atom *snapshot;
975 snapshot = calloc(sizeof(*snapshot), 1);
979 if (sched_out_state(switch_event) == 'R') {
980 snapshot->state = THREAD_WAIT_CPU;
981 snapshot->wake_up_time = timestamp;
984 snapshot->runtime = delta;
985 list_add_tail(&snapshot->list, &atoms->snapshot_list);
989 lat_sched_in(struct task_atoms *atoms, u64 timestamp)
991 struct work_atom *snapshot;
994 if (list_empty(&atoms->snapshot_list))
997 snapshot = list_entry(atoms->snapshot_list.prev, struct work_atom,
1000 if (snapshot->state != THREAD_WAIT_CPU)
1003 if (timestamp < snapshot->wake_up_time) {
1004 snapshot->state = THREAD_IGNORE;
1008 snapshot->state = THREAD_SCHED_IN;
1009 snapshot->sched_in_time = timestamp;
1011 delta = snapshot->sched_in_time - snapshot->wake_up_time;
1012 atoms->total_lat += delta;
1013 if (delta > atoms->max_lat)
1014 atoms->max_lat = delta;
1016 atoms->total_runtime += snapshot->runtime;
1020 latency_switch_event(struct trace_switch_event *switch_event,
1021 struct event *event __used,
1024 struct thread *thread __used)
1026 struct task_atoms *out_atoms, *in_atoms;
1027 struct thread *sched_out, *sched_in;
1031 if (cpu >= MAX_CPUS || cpu < 0)
1034 timestamp0 = cpu_last_switched[cpu];
1035 cpu_last_switched[cpu] = timestamp;
1037 delta = timestamp - timestamp0;
1042 die("hm, delta: %Ld < 0 ?\n", delta);
1045 sched_out = threads__findnew(switch_event->prev_pid, &threads, &last_match);
1046 sched_in = threads__findnew(switch_event->next_pid, &threads, &last_match);
1048 in_atoms = thread_atom_list_search(&lat_snapshot_root, sched_in);
1050 thread_atom_list_insert(sched_in);
1051 in_atoms = thread_atom_list_search(&lat_snapshot_root, sched_in);
1053 die("Internal latency tree error");
1056 out_atoms = thread_atom_list_search(&lat_snapshot_root, sched_out);
1058 thread_atom_list_insert(sched_out);
1059 out_atoms = thread_atom_list_search(&lat_snapshot_root, sched_out);
1061 die("Internal latency tree error");
1064 lat_sched_in(in_atoms, timestamp);
1065 lat_sched_out(out_atoms, switch_event, delta, timestamp);
1069 latency_wakeup_event(struct trace_wakeup_event *wakeup_event,
1070 struct event *event __used,
1073 struct thread *thread __used)
1075 struct task_atoms *atoms;
1076 struct work_atom *snapshot;
1077 struct thread *wakee;
1079 /* Note for later, it may be interesting to observe the failing cases */
1080 if (!wakeup_event->success)
1083 wakee = threads__findnew(wakeup_event->pid, &threads, &last_match);
1084 atoms = thread_atom_list_search(&lat_snapshot_root, wakee);
1086 thread_atom_list_insert(wakee);
1090 if (list_empty(&atoms->snapshot_list))
1093 snapshot = list_entry(atoms->snapshot_list.prev, struct work_atom,
1096 if (snapshot->state != THREAD_SLEEPING)
1099 snapshot->state = THREAD_WAIT_CPU;
1100 snapshot->wake_up_time = timestamp;
1103 static struct trace_sched_handler lat_ops = {
1104 .wakeup_event = latency_wakeup_event,
1105 .switch_event = latency_switch_event,
1106 .fork_event = latency_fork_event,
1109 static u64 all_runtime;
1110 static u64 all_count;
1112 static void output_lat_thread(struct task_atoms *atom_list)
1118 if (!atom_list->nb_atoms)
1121 all_runtime += atom_list->total_runtime;
1122 all_count += atom_list->nb_atoms;
1124 ret = printf(" %s ", atom_list->thread->comm);
1126 for (i = 0; i < 19 - ret; i++)
1129 avg = atom_list->total_lat / atom_list->nb_atoms;
1131 printf("|%9.3f ms |%9llu | avg:%9.3f ms | max:%9.3f ms |\n",
1132 (double)atom_list->total_runtime / 1e6,
1133 atom_list->nb_atoms, (double)avg / 1e6,
1134 (double)atom_list->max_lat / 1e6);
1137 static void __cmd_lat(void)
1139 struct rb_node *next;
1144 printf("-----------------------------------------------------------------------------------\n");
1145 printf(" Task | Runtime ms | Switches | Average delay ms | Maximum delay ms |\n");
1146 printf("-----------------------------------------------------------------------------------\n");
1148 next = rb_first(&lat_snapshot_root);
1151 struct task_atoms *atom_list;
1153 atom_list = rb_entry(next, struct task_atoms, node);
1154 output_lat_thread(atom_list);
1155 next = rb_next(next);
1158 printf("-----------------------------------------------------------------------------------\n");
1159 printf(" TOTAL: |%9.3f ms |%9Ld |\n",
1160 (double)all_runtime/1e6, all_count);
1161 printf("---------------------------------------------\n");
1164 static struct trace_sched_handler *trace_handler;
1167 process_sched_wakeup_event(struct raw_event_sample *raw,
1168 struct event *event,
1170 u64 timestamp __used,
1171 struct thread *thread __used)
1173 struct trace_wakeup_event wakeup_event;
1175 FILL_COMMON_FIELDS(wakeup_event, event, raw->data);
1177 FILL_ARRAY(wakeup_event, comm, event, raw->data);
1178 FILL_FIELD(wakeup_event, pid, event, raw->data);
1179 FILL_FIELD(wakeup_event, prio, event, raw->data);
1180 FILL_FIELD(wakeup_event, success, event, raw->data);
1181 FILL_FIELD(wakeup_event, cpu, event, raw->data);
1183 trace_handler->wakeup_event(&wakeup_event, event, cpu, timestamp, thread);
1187 process_sched_switch_event(struct raw_event_sample *raw,
1188 struct event *event,
1190 u64 timestamp __used,
1191 struct thread *thread __used)
1193 struct trace_switch_event switch_event;
1195 FILL_COMMON_FIELDS(switch_event, event, raw->data);
1197 FILL_ARRAY(switch_event, prev_comm, event, raw->data);
1198 FILL_FIELD(switch_event, prev_pid, event, raw->data);
1199 FILL_FIELD(switch_event, prev_prio, event, raw->data);
1200 FILL_FIELD(switch_event, prev_state, event, raw->data);
1201 FILL_ARRAY(switch_event, next_comm, event, raw->data);
1202 FILL_FIELD(switch_event, next_pid, event, raw->data);
1203 FILL_FIELD(switch_event, next_prio, event, raw->data);
1205 trace_handler->switch_event(&switch_event, event, cpu, timestamp, thread);
1209 process_sched_fork_event(struct raw_event_sample *raw,
1210 struct event *event,
1212 u64 timestamp __used,
1213 struct thread *thread __used)
1215 struct trace_fork_event fork_event;
1217 FILL_COMMON_FIELDS(fork_event, event, raw->data);
1219 FILL_ARRAY(fork_event, parent_comm, event, raw->data);
1220 FILL_FIELD(fork_event, parent_pid, event, raw->data);
1221 FILL_ARRAY(fork_event, child_comm, event, raw->data);
1222 FILL_FIELD(fork_event, child_pid, event, raw->data);
1224 trace_handler->fork_event(&fork_event, event, cpu, timestamp, thread);
1228 process_sched_exit_event(struct event *event,
1230 u64 timestamp __used,
1231 struct thread *thread __used)
1234 printf("sched_exit event %p\n", event);
1238 process_raw_event(event_t *raw_event __used, void *more_data,
1239 int cpu, u64 timestamp, struct thread *thread)
1241 struct raw_event_sample *raw = more_data;
1242 struct event *event;
1245 type = trace_parse_common_type(raw->data);
1246 event = trace_find_event(type);
1248 if (!strcmp(event->name, "sched_switch"))
1249 process_sched_switch_event(raw, event, cpu, timestamp, thread);
1250 if (!strcmp(event->name, "sched_wakeup"))
1251 process_sched_wakeup_event(raw, event, cpu, timestamp, thread);
1252 if (!strcmp(event->name, "sched_wakeup_new"))
1253 process_sched_wakeup_event(raw, event, cpu, timestamp, thread);
1254 if (!strcmp(event->name, "sched_process_fork"))
1255 process_sched_fork_event(raw, event, cpu, timestamp, thread);
1256 if (!strcmp(event->name, "sched_process_exit"))
1257 process_sched_exit_event(event, cpu, timestamp, thread);
1261 process_sample_event(event_t *event, unsigned long offset, unsigned long head)
1265 struct dso *dso = NULL;
1266 struct thread *thread;
1267 u64 ip = event->ip.ip;
1271 void *more_data = event->ip.__more_data;
1274 thread = threads__findnew(event->ip.pid, &threads, &last_match);
1276 if (sample_type & PERF_SAMPLE_TIME) {
1277 timestamp = *(u64 *)more_data;
1278 more_data += sizeof(u64);
1281 if (sample_type & PERF_SAMPLE_CPU) {
1282 cpu = *(u32 *)more_data;
1283 more_data += sizeof(u32);
1284 more_data += sizeof(u32); /* reserved */
1287 if (sample_type & PERF_SAMPLE_PERIOD) {
1288 period = *(u64 *)more_data;
1289 more_data += sizeof(u64);
1292 dump_printf("%p [%p]: PERF_EVENT_SAMPLE (IP, %d): %d/%d: %p period: %Ld\n",
1293 (void *)(offset + head),
1294 (void *)(long)(event->header.size),
1296 event->ip.pid, event->ip.tid,
1300 dump_printf(" ... thread: %s:%d\n", thread->comm, thread->pid);
1302 if (thread == NULL) {
1303 eprintf("problem processing %d event, skipping it.\n",
1304 event->header.type);
1308 cpumode = event->header.misc & PERF_EVENT_MISC_CPUMODE_MASK;
1310 if (cpumode == PERF_EVENT_MISC_KERNEL) {
1316 dump_printf(" ...... dso: %s\n", dso->name);
1318 } else if (cpumode == PERF_EVENT_MISC_USER) {
1327 dso = hypervisor_dso;
1329 dump_printf(" ...... dso: [hypervisor]\n");
1332 if (sample_type & PERF_SAMPLE_RAW)
1333 process_raw_event(event, more_data, cpu, timestamp, thread);
1339 process_event(event_t *event, unsigned long offset, unsigned long head)
1343 switch (event->header.type) {
1344 case PERF_EVENT_MMAP ... PERF_EVENT_LOST:
1347 case PERF_EVENT_COMM:
1348 return process_comm_event(event, offset, head);
1350 case PERF_EVENT_EXIT ... PERF_EVENT_READ:
1353 case PERF_EVENT_SAMPLE:
1354 return process_sample_event(event, offset, head);
1356 case PERF_EVENT_MAX:
1364 static int read_events(void)
1366 int ret, rc = EXIT_FAILURE;
1367 unsigned long offset = 0;
1368 unsigned long head = 0;
1369 struct stat perf_stat;
1375 register_idle_thread(&threads, &last_match);
1377 input = open(input_name, O_RDONLY);
1379 perror("failed to open file");
1383 ret = fstat(input, &perf_stat);
1385 perror("failed to stat file");
1389 if (!perf_stat.st_size) {
1390 fprintf(stderr, "zero-sized file, nothing to do!\n");
1393 header = perf_header__read(input);
1394 head = header->data_offset;
1395 sample_type = perf_header__sample_type(header);
1397 if (!(sample_type & PERF_SAMPLE_RAW))
1398 die("No trace sample to read. Did you call perf record "
1401 if (load_kernel() < 0) {
1402 perror("failed to load kernel symbols");
1403 return EXIT_FAILURE;
1407 buf = (char *)mmap(NULL, page_size * mmap_window, PROT_READ,
1408 MAP_SHARED, input, offset);
1409 if (buf == MAP_FAILED) {
1410 perror("failed to mmap file");
1415 event = (event_t *)(buf + head);
1417 size = event->header.size;
1421 if (head + event->header.size >= page_size * mmap_window) {
1422 unsigned long shift = page_size * (head / page_size);
1425 res = munmap(buf, page_size * mmap_window);
1433 size = event->header.size;
1436 if (!size || process_event(event, offset, head) < 0) {
1439 * assume we lost track of the stream, check alignment, and
1440 * increment a single u64 in the hope to catch on again 'soon'.
1443 if (unlikely(head & 7))
1451 if (offset + head < (unsigned long)perf_stat.st_size)
1460 static const char * const sched_usage[] = {
1461 "perf sched [<options>] <command>",
1465 static const struct option options[] = {
1466 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
1467 "dump raw trace in ASCII"),
1468 OPT_BOOLEAN('r', "replay", &replay_mode,
1469 "replay sched behaviour from traces"),
1470 OPT_BOOLEAN('l', "latency", &lat_mode,
1471 "measure various latencies"),
1472 OPT_BOOLEAN('v', "verbose", &verbose,
1473 "be more verbose (show symbol address, etc)"),
1477 int cmd_sched(int argc, const char **argv, const char *prefix __used)
1480 page_size = getpagesize();
1482 argc = parse_options(argc, argv, options, sched_usage, 0);
1485 * Special case: if there's an argument left then assume tha
1486 * it's a symbol filter:
1489 usage_with_options(sched_usage, options);
1493 trace_handler = &replay_ops;
1495 trace_handler = &lat_ops;
1497 usage_with_options(sched_usage, options);