← Back to branch summary

~ubuntu-branches/ubuntu/precise/linux-lowlatency/precise

« back to all changes in this revision

Viewing changes to tools/perf/builtin-sched.c

  • Committer: Package Import Robot
  • Author(s): Alessio Igor Bogani
  • Date: 2011-10-26 11:13:05 UTC
  • Revision ID: package-import@ubuntu.com-20111026111305-tz023xykf0i6eosh
Tags: upstream-3.2.0
Import upstream version 3.2.0

Show diffs side-by-side

added added

removed removed

Lines of Context:
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
 
 
22
static char                     const *input_name = "perf.data";
 
23
 
 
24
static char                     default_sort_order[] = "avg, max, switch, runtime";
 
25
static const char               *sort_order = default_sort_order;
 
26
 
 
27
static int                      profile_cpu = -1;
 
28
 
 
29
#define PR_SET_NAME             15               /* Set process name */
 
30
#define MAX_CPUS                4096
 
31
 
 
32
static u64                      run_measurement_overhead;
 
33
static u64                      sleep_measurement_overhead;
 
34
 
 
35
#define COMM_LEN                20
 
36
#define SYM_LEN                 129
 
37
 
 
38
#define MAX_PID                 65536
 
39
 
 
40
static unsigned long            nr_tasks;
 
41
 
 
42
struct sched_atom;
 
43
 
 
44
struct 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
 
 
62
enum sched_event_type {
 
63
        SCHED_EVENT_RUN,
 
64
        SCHED_EVENT_SLEEP,
 
65
        SCHED_EVENT_WAKEUP,
 
66
        SCHED_EVENT_MIGRATION,
 
67
};
 
68
 
 
69
struct 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
 
 
79
static struct task_desc         *pid_to_task[MAX_PID];
 
80
 
 
81
static struct task_desc         **tasks;
 
82
 
 
83
static pthread_mutex_t          start_work_mutex = PTHREAD_MUTEX_INITIALIZER;
 
84
static u64                      start_time;
 
85
 
 
86
static pthread_mutex_t          work_done_wait_mutex = PTHREAD_MUTEX_INITIALIZER;
 
87
 
 
88
static unsigned long            nr_run_events;
 
89
static unsigned long            nr_sleep_events;
 
90
static unsigned long            nr_wakeup_events;
 
91
 
 
92
static unsigned long            nr_sleep_corrections;
 
93
static unsigned long            nr_run_events_optimized;
 
94
 
 
95
static unsigned long            targetless_wakeups;
 
96
static unsigned long            multitarget_wakeups;
 
97
 
 
98
static u64                      cpu_usage;
 
99
static u64                      runavg_cpu_usage;
 
100
static u64                      parent_cpu_usage;
 
101
static u64                      runavg_parent_cpu_usage;
 
102
 
 
103
static unsigned long            nr_runs;
 
104
static u64                      sum_runtime;
 
105
static u64                      sum_fluct;
 
106
static u64                      run_avg;
 
107
 
 
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;
 
116
 
 
117
#define TASK_STATE_TO_CHAR_STR "RSDTtZX"
 
118
 
 
119
enum thread_state {
 
120
        THREAD_SLEEPING = 0,
 
121
        THREAD_WAIT_CPU,
 
122
        THREAD_SCHED_IN,
 
123
        THREAD_IGNORE
 
124
};
 
125
 
 
126
struct 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
 
 
135
struct 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
 
 
146
typedef int (*sort_fn_t)(struct work_atoms *, struct work_atoms *);
 
147
 
 
148
static struct rb_root           atom_root, sorted_atom_root;
 
149
 
 
150
static u64                      all_runtime;
 
151
static u64                      all_count;
 
152
 
 
153
 
 
154
static 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
 
 
163
static 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
 
 
172
static 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
 
 
182
static 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
 
 
199
static 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
 
 
217
static struct sched_atom *
 
218
get_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
 
 
237
static 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
 
 
245
static void
 
246
add_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
 
 
268
static void
 
269
add_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
 
 
296
static void
 
297
add_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
 
 
307
static 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
 
 
341
static 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
 
 
353
static 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
 
 
368
static void
 
369
process_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
 
 
394
static 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
 
 
409
static 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
 
 
427
static 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
 
 
438
static 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
 
 
450
again:
 
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
 
 
478
static 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
 
 
505
static 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
 
 
558
static 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
 
 
606
static 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)                     \
 
627
do {                                                            \
 
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)                    \
 
633
do {                                                            \
 
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
 
 
643
struct 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
 
 
661
struct 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
 
 
676
struct 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
 
 
693
struct 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
 
 
708
struct 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
 
 
724
struct 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
 
 
761
static void
 
762
replay_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
 
 
786
static u64 cpu_last_switched[MAX_CPUS];
 
787
 
 
788
static void
 
789
replay_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
 
 
832
static void
 
833
replay_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
 
 
848
static 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
 
 
854
struct sort_dimension {
 
855
        const char              *name;
 
856
        sort_fn_t               cmp;
 
857
        struct list_head        list;
 
858
};
 
859
 
 
860
static LIST_HEAD(cmp_pid);
 
861
 
 
862
static int
 
863
thread_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
 
 
879
static struct work_atoms *
 
880
thread_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
 
 
905
static 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
 
 
930
static 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
 
 
941
static void
 
942
latency_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
 
952
static 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
 
 
959
static void
 
960
add_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
 
 
978
static void
 
979
add_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
 
 
991
static void
 
992
add_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
 
 
1022
static void
 
1023
latency_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
 
 
1075
static void
 
1076
latency_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
 
 
1098
static void
 
1099
latency_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
 
 
1146
static void
 
1147
latency_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
 
 
1186
static 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
 
 
1194
static 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
 
 
1225
static 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
 
 
1235
static struct sort_dimension pid_sort_dimension = {
 
1236
        .name                   = "pid",
 
1237
        .cmp                    = pid_cmp,
 
1238
};
 
1239
 
 
1240
static 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
 
 
1261
static struct sort_dimension avg_sort_dimension = {
 
1262
        .name                   = "avg",
 
1263
        .cmp                    = avg_cmp,
 
1264
};
 
1265
 
 
1266
static 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
 
 
1276
static struct sort_dimension max_sort_dimension = {
 
1277
        .name                   = "max",
 
1278
        .cmp                    = max_cmp,
 
1279
};
 
1280
 
 
1281
static 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
 
 
1291
static struct sort_dimension switch_sort_dimension = {
 
1292
        .name                   = "switch",
 
1293
        .cmp                    = switch_cmp,
 
1294
};
 
1295
 
 
1296
static 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
 
 
1306
static struct sort_dimension runtime_sort_dimension = {
 
1307
        .name                   = "runtime",
 
1308
        .cmp                    = runtime_cmp,
 
1309
};
 
1310
 
 
1311
static 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
 
 
1321
static LIST_HEAD(sort_list);
 
1322
 
 
1323
static 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
 
 
1338
static void setup_sorting(void);
 
1339
 
 
1340
static 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
 
 
1356
static struct trace_sched_handler *trace_handler;
 
1357
 
 
1358
static void
 
1359
process_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
 */
 
1384
static int max_cpu;
 
1385
 
 
1386
static u32 curr_pid[MAX_CPUS] = { [0 ... MAX_CPUS-1] = -1 };
 
1387
 
 
1388
static struct thread *curr_thread[MAX_CPUS];
 
1389
 
 
1390
static char next_shortname1 = 'A';
 
1391
static char next_shortname2 = '0';
 
1392
 
 
1393
static void
 
1394
map_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
 
 
1473
static void
 
1474
process_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
 
 
1507
static void
 
1508
process_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
 
 
1525
static void
 
1526
process_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
 
 
1546
static void
 
1547
process_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
 
 
1556
static void
 
1557
process_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
 
 
1577
static 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
 
 
1604
static 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
 
 
1632
static 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
 
 
1640
static void read_events(bool destroy, struct perf_session **psession)
 
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
                die("No Memory");
 
1647
 
 
1648
        if (perf_session__has_traces(session, "record -R")) {
 
1649
                err = perf_session__process_events(session, &event_ops);
 
1650
                if (err)
 
1651
                        die("Failed to process events, error %d", err);
 
1652
 
 
1653
                nr_events      = session->hists.stats.nr_events[0];
 
1654
                nr_lost_events = session->hists.stats.total_lost;
 
1655
                nr_lost_chunks = session->hists.stats.nr_events[PERF_RECORD_LOST];
 
1656
        }
 
1657
 
 
1658
        if (destroy)
 
1659
                perf_session__delete(session);
 
1660
 
 
1661
        if (psession)
 
1662
                *psession = session;
 
1663
}
 
1664
 
 
1665
static void print_bad_events(void)
 
1666
{
 
1667
        if (nr_unordered_timestamps && nr_timestamps) {
 
1668
                printf("  INFO: %.3f%% unordered timestamps (%ld out of %ld)\n",
 
1669
                        (double)nr_unordered_timestamps/(double)nr_timestamps*100.0,
 
1670
                        nr_unordered_timestamps, nr_timestamps);
 
1671
        }
 
1672
        if (nr_lost_events && nr_events) {
 
1673
                printf("  INFO: %.3f%% lost events (%ld out of %ld, in %ld chunks)\n",
 
1674
                        (double)nr_lost_events/(double)nr_events*100.0,
 
1675
                        nr_lost_events, nr_events, nr_lost_chunks);
 
1676
        }
 
1677
        if (nr_state_machine_bugs && nr_timestamps) {
 
1678
                printf("  INFO: %.3f%% state machine bugs (%ld out of %ld)",
 
1679
                        (double)nr_state_machine_bugs/(double)nr_timestamps*100.0,
 
1680
                        nr_state_machine_bugs, nr_timestamps);
 
1681
                if (nr_lost_events)
 
1682
                        printf(" (due to lost events?)");
 
1683
                printf("\n");
 
1684
        }
 
1685
        if (nr_context_switch_bugs && nr_timestamps) {
 
1686
                printf("  INFO: %.3f%% context switch bugs (%ld out of %ld)",
 
1687
                        (double)nr_context_switch_bugs/(double)nr_timestamps*100.0,
 
1688
                        nr_context_switch_bugs, nr_timestamps);
 
1689
                if (nr_lost_events)
 
1690
                        printf(" (due to lost events?)");
 
1691
                printf("\n");
 
1692
        }
 
1693
}
 
1694
 
 
1695
static void __cmd_lat(void)
 
1696
{
 
1697
        struct rb_node *next;
 
1698
        struct perf_session *session;
 
1699
 
 
1700
        setup_pager();
 
1701
        read_events(false, &session);
 
1702
        sort_lat();
 
1703
 
 
1704
        printf("\n ---------------------------------------------------------------------------------------------------------------\n");
 
1705
        printf("  Task                  |   Runtime ms  | Switches | Average delay ms | Maximum delay ms | Maximum delay at     |\n");
 
1706
        printf(" ---------------------------------------------------------------------------------------------------------------\n");
 
1707
 
 
1708
        next = rb_first(&sorted_atom_root);
 
1709
 
 
1710
        while (next) {
 
1711
                struct work_atoms *work_list;
 
1712
 
 
1713
                work_list = rb_entry(next, struct work_atoms, node);
 
1714
                output_lat_thread(work_list);
 
1715
                next = rb_next(next);
 
1716
        }
 
1717
 
 
1718
        printf(" -----------------------------------------------------------------------------------------\n");
 
1719
        printf("  TOTAL:                |%11.3f ms |%9" PRIu64 " |\n",
 
1720
                (double)all_runtime/1e6, all_count);
 
1721
 
 
1722
        printf(" ---------------------------------------------------\n");
 
1723
 
 
1724
        print_bad_events();
 
1725
        printf("\n");
 
1726
 
 
1727
        perf_session__delete(session);
 
1728
}
 
1729
 
 
1730
static struct trace_sched_handler map_ops  = {
 
1731
        .wakeup_event           = NULL,
 
1732
        .switch_event           = map_switch_event,
 
1733
        .runtime_event          = NULL,
 
1734
        .fork_event             = NULL,
 
1735
};
 
1736
 
 
1737
static void __cmd_map(void)
 
1738
{
 
1739
        max_cpu = sysconf(_SC_NPROCESSORS_CONF);
 
1740
 
 
1741
        setup_pager();
 
1742
        read_events(true, NULL);
 
1743
        print_bad_events();
 
1744
}
 
1745
 
 
1746
static void __cmd_replay(void)
 
1747
{
 
1748
        unsigned long i;
 
1749
 
 
1750
        calibrate_run_measurement_overhead();
 
1751
        calibrate_sleep_measurement_overhead();
 
1752
 
 
1753
        test_calibrations();
 
1754
 
 
1755
        read_events(true, NULL);
 
1756
 
 
1757
        printf("nr_run_events:        %ld\n", nr_run_events);
 
1758
        printf("nr_sleep_events:      %ld\n", nr_sleep_events);
 
1759
        printf("nr_wakeup_events:     %ld\n", nr_wakeup_events);
 
1760
 
 
1761
        if (targetless_wakeups)
 
1762
                printf("target-less wakeups:  %ld\n", targetless_wakeups);
 
1763
        if (multitarget_wakeups)
 
1764
                printf("multi-target wakeups: %ld\n", multitarget_wakeups);
 
1765
        if (nr_run_events_optimized)
 
1766
                printf("run atoms optimized: %ld\n",
 
1767
                        nr_run_events_optimized);
 
1768
 
 
1769
        print_task_traces();
 
1770
        add_cross_task_wakeups();
 
1771
 
 
1772
        create_tasks();
 
1773
        printf("------------------------------------------------------------\n");
 
1774
        for (i = 0; i < replay_repeat; i++)
 
1775
                run_one_test();
 
1776
}
 
1777
 
 
1778
 
 
1779
static const char * const sched_usage[] = {
 
1780
        "perf sched [<options>] {record|latency|map|replay|script}",
 
1781
        NULL
 
1782
};
 
1783
 
 
1784
static const struct option sched_options[] = {
 
1785
        OPT_STRING('i', "input", &input_name, "file",
 
1786
                    "input file name"),
 
1787
        OPT_INCR('v', "verbose", &verbose,
 
1788
                    "be more verbose (show symbol address, etc)"),
 
1789
        OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
 
1790
                    "dump raw trace in ASCII"),
 
1791
        OPT_END()
 
1792
};
 
1793
 
 
1794
static const char * const latency_usage[] = {
 
1795
        "perf sched latency [<options>]",
 
1796
        NULL
 
1797
};
 
1798
 
 
1799
static const struct option latency_options[] = {
 
1800
        OPT_STRING('s', "sort", &sort_order, "key[,key2...]",
 
1801
                   "sort by key(s): runtime, switch, avg, max"),
 
1802
        OPT_INCR('v', "verbose", &verbose,
 
1803
                    "be more verbose (show symbol address, etc)"),
 
1804
        OPT_INTEGER('C', "CPU", &profile_cpu,
 
1805
                    "CPU to profile on"),
 
1806
        OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
 
1807
                    "dump raw trace in ASCII"),
 
1808
        OPT_END()
 
1809
};
 
1810
 
 
1811
static const char * const replay_usage[] = {
 
1812
        "perf sched replay [<options>]",
 
1813
        NULL
 
1814
};
 
1815
 
 
1816
static const struct option replay_options[] = {
 
1817
        OPT_UINTEGER('r', "repeat", &replay_repeat,
 
1818
                     "repeat the workload replay N times (-1: infinite)"),
 
1819
        OPT_INCR('v', "verbose", &verbose,
 
1820
                    "be more verbose (show symbol address, etc)"),
 
1821
        OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
 
1822
                    "dump raw trace in ASCII"),
 
1823
        OPT_END()
 
1824
};
 
1825
 
 
1826
static void setup_sorting(void)
 
1827
{
 
1828
        char *tmp, *tok, *str = strdup(sort_order);
 
1829
 
 
1830
        for (tok = strtok_r(str, ", ", &tmp);
 
1831
                        tok; tok = strtok_r(NULL, ", ", &tmp)) {
 
1832
                if (sort_dimension__add(tok, &sort_list) < 0) {
 
1833
                        error("Unknown --sort key: `%s'", tok);
 
1834
                        usage_with_options(latency_usage, latency_options);
 
1835
                }
 
1836
        }
 
1837
 
 
1838
        free(str);
 
1839
 
 
1840
        sort_dimension__add("pid", &cmp_pid);
 
1841
}
 
1842
 
 
1843
static const char *record_args[] = {
 
1844
        "record",
 
1845
        "-a",
 
1846
        "-R",
 
1847
        "-f",
 
1848
        "-m", "1024",
 
1849
        "-c", "1",
 
1850
        "-e", "sched:sched_switch",
 
1851
        "-e", "sched:sched_stat_wait",
 
1852
        "-e", "sched:sched_stat_sleep",
 
1853
        "-e", "sched:sched_stat_iowait",
 
1854
        "-e", "sched:sched_stat_runtime",
 
1855
        "-e", "sched:sched_process_exit",
 
1856
        "-e", "sched:sched_process_fork",
 
1857
        "-e", "sched:sched_wakeup",
 
1858
        "-e", "sched:sched_migrate_task",
 
1859
};
 
1860
 
 
1861
static int __cmd_record(int argc, const char **argv)
 
1862
{
 
1863
        unsigned int rec_argc, i, j;
 
1864
        const char **rec_argv;
 
1865
 
 
1866
        rec_argc = ARRAY_SIZE(record_args) + argc - 1;
 
1867
        rec_argv = calloc(rec_argc + 1, sizeof(char *));
 
1868
 
 
1869
        if (rec_argv == NULL)
 
1870
                return -ENOMEM;
 
1871
 
 
1872
        for (i = 0; i < ARRAY_SIZE(record_args); i++)
 
1873
                rec_argv[i] = strdup(record_args[i]);
 
1874
 
 
1875
        for (j = 1; j < (unsigned int)argc; j++, i++)
 
1876
                rec_argv[i] = argv[j];
 
1877
 
 
1878
        BUG_ON(i != rec_argc);
 
1879
 
 
1880
        return cmd_record(i, rec_argv, NULL);
 
1881
}
 
1882
 
 
1883
int cmd_sched(int argc, const char **argv, const char *prefix __used)
 
1884
{
 
1885
        argc = parse_options(argc, argv, sched_options, sched_usage,
 
1886
                             PARSE_OPT_STOP_AT_NON_OPTION);
 
1887
        if (!argc)
 
1888
                usage_with_options(sched_usage, sched_options);
 
1889
 
 
1890
        /*
 
1891
         * Aliased to 'perf script' for now:
 
1892
         */
 
1893
        if (!strcmp(argv[0], "script"))
 
1894
                return cmd_script(argc, argv, prefix);
 
1895
 
 
1896
        symbol__init();
 
1897
        if (!strncmp(argv[0], "rec", 3)) {
 
1898
                return __cmd_record(argc, argv);
 
1899
        } else if (!strncmp(argv[0], "lat", 3)) {
 
1900
                trace_handler = &lat_ops;
 
1901
                if (argc > 1) {
 
1902
                        argc = parse_options(argc, argv, latency_options, latency_usage, 0);
 
1903
                        if (argc)
 
1904
                                usage_with_options(latency_usage, latency_options);
 
1905
                }
 
1906
                setup_sorting();
 
1907
                __cmd_lat();
 
1908
        } else if (!strcmp(argv[0], "map")) {
 
1909
                trace_handler = &map_ops;
 
1910
                setup_sorting();
 
1911
                __cmd_map();
 
1912
        } else if (!strncmp(argv[0], "rep", 3)) {
 
1913
                trace_handler = &replay_ops;
 
1914
                if (argc) {
 
1915
                        argc = parse_options(argc, argv, replay_options, replay_usage, 0);
 
1916
                        if (argc)
 
1917
                                usage_with_options(replay_usage, replay_options);
 
1918
                }
 
1919
                __cmd_replay();
 
1920
        } else {
 
1921
                usage_with_options(sched_usage, sched_options);
 
1922
        }
 
1923
 
 
1924
        return 0;
 
1925
}