Root/tools/perf/builtin-sched.c

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

Archive Download this file



interactive