projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
perf sched timehist: Split is_idle_sample()
[linux-2.6-block.git] / tools / perf / builtin-sched.c
CommitLineData
0a02ad93 1#include "builtin.h"
b1ffe8f3 2#include "perf.h"
0a02ad93
IM		 3
4#include "util/util.h"
ee29be62 5#include "util/evlist.h"
0a02ad93 6#include "util/cache.h"
e3f42609 7#include "util/evsel.h"
0a02ad93
IM		 8#include "util/symbol.h"
9#include "util/thread.h"
10#include "util/header.h"
94c744b6 11#include "util/session.h"
45694aa7 12#include "util/tool.h"
57480d2c 13#include "util/cloexec.h"
a151a37a 14#include "util/thread_map.h"
8cd91195 15#include "util/color.h"
49394a2a 16#include "util/stat.h"
6c973c90 17#include "util/callchain.h"
853b7407 18#include "util/time-utils.h"
0a02ad93 19
4b6ab94e 20#include <subcmd/parse-options.h>
b1ffe8f3 21#include "util/trace-event.h"
0a02ad93 22
0a02ad93
IM		 23#include "util/debug.h"
24
49394a2a 25#include <linux/log2.h>
b1ffe8f3 26#include <sys/prctl.h>
7b78f136 27#include <sys/resource.h>
0a02ad93 28
b1ffe8f3
IM		 29#include <semaphore.h>
30#include <pthread.h>
31#include <math.h>
cb06ac25 32#include <api/fs/fs.h>
4fc76e49 33#include <linux/time64.h>
419ab0d6 34
b1ffe8f3
IM		 35#define PR_SET_NAME 15 /* Set process name */
36#define MAX_CPUS 4096
b1ffe8f3
IM		 37#define COMM_LEN 20
38#define SYM_LEN 129
a35e27d0 39#define MAX_PID 1024000
ec156764 40
39aeb52f 41struct sched_atom;
ec156764 42
b1ffe8f3
IM		 43struct task_desc {
44 unsigned long nr;
45 unsigned long pid;
46 char comm[COMM_LEN];
ec156764 47
b1ffe8f3
IM		 48 unsigned long nr_events;
49 unsigned long curr_event;
39aeb52f 50 struct sched_atom **atoms;
b1ffe8f3
IM		 51
52 pthread_t thread;
53 sem_t sleep_sem;
ec156764 54
b1ffe8f3
IM		 55 sem_t ready_for_work;
56 sem_t work_done_sem;
57
58 u64 cpu_usage;
59};
60
61enum sched_event_type {
62 SCHED_EVENT_RUN,
63 SCHED_EVENT_SLEEP,
64 SCHED_EVENT_WAKEUP,
55ffb7a6 65 SCHED_EVENT_MIGRATION,
b1ffe8f3
IM		 66};
67
39aeb52f 68struct sched_atom {
b1ffe8f3 69 enum sched_event_type type;
eed05fe7 70 int specific_wait;
b1ffe8f3
IM		 71 u64 timestamp;
72 u64 duration;
73 unsigned long nr;
b1ffe8f3
IM		 74 sem_t *wait_sem;
75 struct task_desc *wakee;
76};
77
e936e8e4 78#define TASK_STATE_TO_CHAR_STR "RSDTtZXxKWP"
b1ffe8f3
IM		 79
80enum thread_state {
81 THREAD_SLEEPING = 0,
82 THREAD_WAIT_CPU,
83 THREAD_SCHED_IN,
84 THREAD_IGNORE
85};
86
87struct work_atom {
88 struct list_head list;
89 enum thread_state state;
aa1ab9d2 90 u64 sched_out_time;
b1ffe8f3
IM		 91 u64 wake_up_time;
92 u64 sched_in_time;
93 u64 runtime;
94};
95
39aeb52f 96struct work_atoms {
97 struct list_head work_list;
b1ffe8f3
IM		 98 struct thread *thread;
99 struct rb_node node;
100 u64 max_lat;
3786310a 101 u64 max_lat_at;
b1ffe8f3
IM		 102 u64 total_lat;
103 u64 nb_atoms;
104 u64 total_runtime;
2f80dd44 105 int num_merged;
b1ffe8f3
IM		 106};
107
39aeb52f 108typedef int (*sort_fn_t)(struct work_atoms *, struct work_atoms *);
b1ffe8f3 109
9ec3f4e4 110struct perf_sched;
0e9b07e5 111
9ec3f4e4
ACM		 112struct trace_sched_handler {
113 int (*switch_event)(struct perf_sched *sched, struct perf_evsel *evsel,
114 struct perf_sample *sample, struct machine *machine);
0e9b07e5 115
9ec3f4e4
ACM		 116 int (*runtime_event)(struct perf_sched *sched, struct perf_evsel *evsel,
117 struct perf_sample *sample, struct machine *machine);
0e9b07e5 118
9ec3f4e4
ACM		 119 int (*wakeup_event)(struct perf_sched *sched, struct perf_evsel *evsel,
120 struct perf_sample *sample, struct machine *machine);
0e9b07e5 121
cb627505
DA		 122 /* PERF_RECORD_FORK event, not sched_process_fork tracepoint */
123 int (*fork_event)(struct perf_sched *sched, union perf_event *event,
124 struct machine *machine);
0e9b07e5
ACM		 125
126 int (*migrate_task_event)(struct perf_sched *sched,
9ec3f4e4
ACM		 127 struct perf_evsel *evsel,
128 struct perf_sample *sample,
129 struct machine *machine);
0e9b07e5
ACM		 130};
131
a151a37a 132#define COLOR_PIDS PERF_COLOR_BLUE
cf294f24 133#define COLOR_CPUS PERF_COLOR_BG_RED
a151a37a 134
99623c62
JO		 135struct perf_sched_map {
136 DECLARE_BITMAP(comp_cpus_mask, MAX_CPUS);
137 int *comp_cpus;
138 bool comp;
a151a37a
JO		 139 struct thread_map *color_pids;
140 const char *color_pids_str;
cf294f24
JO		 141 struct cpu_map *color_cpus;
142 const char *color_cpus_str;
73643bb6
JO		 143 struct cpu_map *cpus;
144 const char *cpus_str;
99623c62
JO		 145};
146
0e9b07e5
ACM		 147struct perf_sched {
148 struct perf_tool tool;
0e9b07e5
ACM		 149 const char *sort_order;
150 unsigned long nr_tasks;
cb06ac25 151 struct task_desc **pid_to_task;
0e9b07e5
ACM		 152 struct task_desc **tasks;
153 const struct trace_sched_handler *tp_handler;
154 pthread_mutex_t start_work_mutex;
155 pthread_mutex_t work_done_wait_mutex;
156 int profile_cpu;
157/*
158 * Track the current task - that way we can know whether there's any
159 * weird events, such as a task being switched away that is not current.
160 */
161 int max_cpu;
162 u32 curr_pid[MAX_CPUS];
163 struct thread *curr_thread[MAX_CPUS];
164 char next_shortname1;
165 char next_shortname2;
166 unsigned int replay_repeat;
167 unsigned long nr_run_events;
168 unsigned long nr_sleep_events;
169 unsigned long nr_wakeup_events;
170 unsigned long nr_sleep_corrections;
171 unsigned long nr_run_events_optimized;
172 unsigned long targetless_wakeups;
173 unsigned long multitarget_wakeups;
174 unsigned long nr_runs;
175 unsigned long nr_timestamps;
176 unsigned long nr_unordered_timestamps;
0e9b07e5
ACM		 177 unsigned long nr_context_switch_bugs;
178 unsigned long nr_events;
179 unsigned long nr_lost_chunks;
180 unsigned long nr_lost_events;
181 u64 run_measurement_overhead;
182 u64 sleep_measurement_overhead;
183 u64 start_time;
184 u64 cpu_usage;
185 u64 runavg_cpu_usage;
186 u64 parent_cpu_usage;
187 u64 runavg_parent_cpu_usage;
188 u64 sum_runtime;
189 u64 sum_fluct;
190 u64 run_avg;
191 u64 all_runtime;
192 u64 all_count;
193 u64 cpu_last_switched[MAX_CPUS];
2f80dd44 194 struct rb_root atom_root, sorted_atom_root, merged_atom_root;
0e9b07e5 195 struct list_head sort_list, cmp_pid;
939cda52 196 bool force;
2f80dd44 197 bool skip_merge;
99623c62 198 struct perf_sched_map map;
52df138c
DA		 199
200 /* options for timehist command */
201 bool summary;
202 bool summary_only;
6c973c90
DA		 203 bool show_callchain;
204 unsigned int max_stack;
a407b067 205 bool show_cpu_visual;
fc1469f1 206 bool show_wakeups;
350f54fa 207 bool show_migrations;
52df138c 208 u64 skipped_samples;
853b7407
DA		 209 const char *time_str;
210 struct perf_time_interval ptime;
0e9b07e5 211};
b1ffe8f3 212
49394a2a
DA		 213/* per thread run time data */
214struct thread_runtime {
215 u64 last_time; /* time of previous sched in/out event */
216 u64 dt_run; /* run time */
217 u64 dt_wait; /* time between CPU access (off cpu) */
218 u64 dt_delay; /* time between wakeup and sched-in */
219 u64 ready_to_run; /* time of wakeup */
220
221 struct stats run_stats;
222 u64 total_run_time;
350f54fa
DA		 223
224 u64 migrations;
49394a2a
DA		 225};
226
227/* per event run time data */
228struct evsel_runtime {
229 u64 *last_time; /* time this event was last seen per cpu */
230 u32 ncpu; /* highest cpu slot allocated */
231};
232
233/* track idle times per cpu */
234static struct thread **idle_threads;
235static int idle_max_cpu;
236static char idle_comm[] = "<idle>";
237
b1ffe8f3 238static u64 get_nsecs(void)
ec156764
IM		 239{
240 struct timespec ts;
241
242 clock_gettime(CLOCK_MONOTONIC, &ts);
243
4fc76e49 244 return ts.tv_sec * NSEC_PER_SEC + ts.tv_nsec;
ec156764
IM		 245}
246
0e9b07e5 247static void burn_nsecs(struct perf_sched *sched, u64 nsecs)
ec156764 248{
b1ffe8f3 249 u64 T0 = get_nsecs(), T1;
ec156764
IM		 250
251 do {
252 T1 = get_nsecs();
0e9b07e5 253 } while (T1 + sched->run_measurement_overhead < T0 + nsecs);
ec156764
IM		 254}
255
b1ffe8f3 256static void sleep_nsecs(u64 nsecs)
ec156764
IM		 257{
258 struct timespec ts;
259
260 ts.tv_nsec = nsecs % 999999999;
261 ts.tv_sec = nsecs / 999999999;
262
263 nanosleep(&ts, NULL);
264}
265
0e9b07e5 266static void calibrate_run_measurement_overhead(struct perf_sched *sched)
ec156764 267{
4fc76e49 268 u64 T0, T1, delta, min_delta = NSEC_PER_SEC;
ec156764
IM		 269 int i;
270
271 for (i = 0; i < 10; i++) {
272 T0 = get_nsecs();
0e9b07e5 273 burn_nsecs(sched, 0);
ec156764
IM		 274 T1 = get_nsecs();
275 delta = T1-T0;
276 min_delta = min(min_delta, delta);
277 }
0e9b07e5 278 sched->run_measurement_overhead = min_delta;
ec156764 279
9486aa38 280 printf("run measurement overhead: %" PRIu64 " nsecs\n", min_delta);
ec156764
IM		 281}
282
0e9b07e5 283static void calibrate_sleep_measurement_overhead(struct perf_sched *sched)
ec156764 284{
4fc76e49 285 u64 T0, T1, delta, min_delta = NSEC_PER_SEC;
ec156764
IM		 286 int i;
287
288 for (i = 0; i < 10; i++) {
289 T0 = get_nsecs();
290 sleep_nsecs(10000);
291 T1 = get_nsecs();
292 delta = T1-T0;
293 min_delta = min(min_delta, delta);
294 }
295 min_delta -= 10000;
0e9b07e5 296 sched->sleep_measurement_overhead = min_delta;
ec156764 297
9486aa38 298 printf("sleep measurement overhead: %" PRIu64 " nsecs\n", min_delta);
ec156764
IM		 299}
300
39aeb52f 301static struct sched_atom *
b1ffe8f3 302get_new_event(struct task_desc *task, u64 timestamp)
ec156764 303{
36479484 304 struct sched_atom *event = zalloc(sizeof(*event));
ec156764
IM		 305 unsigned long idx = task->nr_events;
306 size_t size;
307
308 event->timestamp = timestamp;
309 event->nr = idx;
310
311 task->nr_events++;
39aeb52f 312 size = sizeof(struct sched_atom *) * task->nr_events;
313 task->atoms = realloc(task->atoms, size);
314 BUG_ON(!task->atoms);
ec156764 315
39aeb52f 316 task->atoms[idx] = event;
ec156764
IM		 317
318 return event;
319}
320
39aeb52f 321static struct sched_atom *last_event(struct task_desc *task)
ec156764
IM		 322{
323 if (!task->nr_events)
324 return NULL;
325
39aeb52f 326 return task->atoms[task->nr_events - 1];
ec156764
IM		 327}
328
0e9b07e5
ACM		 329static void add_sched_event_run(struct perf_sched *sched, struct task_desc *task,
330 u64 timestamp, u64 duration)
ec156764 331{
39aeb52f 332 struct sched_atom *event, *curr_event = last_event(task);
ec156764
IM		 333
334 /*
fbf94829
IM		 335 * optimize an existing RUN event by merging this one
336 * to it:
337 */
ec156764 338 if (curr_event && curr_event->type == SCHED_EVENT_RUN) {
0e9b07e5 339 sched->nr_run_events_optimized++;
ec156764
IM		 340 curr_event->duration += duration;
341 return;
342 }
343
344 event = get_new_event(task, timestamp);
345
346 event->type = SCHED_EVENT_RUN;
347 event->duration = duration;
348
0e9b07e5 349 sched->nr_run_events++;
ec156764
IM		 350}
351
0e9b07e5
ACM		 352static void add_sched_event_wakeup(struct perf_sched *sched, struct task_desc *task,
353 u64 timestamp, struct task_desc *wakee)
ec156764 354{
39aeb52f 355 struct sched_atom *event, *wakee_event;
ec156764
IM		 356
357 event = get_new_event(task, timestamp);
358 event->type = SCHED_EVENT_WAKEUP;
359 event->wakee = wakee;
360
361 wakee_event = last_event(wakee);
362 if (!wakee_event || wakee_event->type != SCHED_EVENT_SLEEP) {
0e9b07e5 363 sched->targetless_wakeups++;
ec156764
IM		 364 return;
365 }
366 if (wakee_event->wait_sem) {
0e9b07e5 367 sched->multitarget_wakeups++;
ec156764
IM		 368 return;
369 }
370
36479484 371 wakee_event->wait_sem = zalloc(sizeof(*wakee_event->wait_sem));
ec156764
IM		 372 sem_init(wakee_event->wait_sem, 0, 0);
373 wakee_event->specific_wait = 1;
374 event->wait_sem = wakee_event->wait_sem;
375
0e9b07e5 376 sched->nr_wakeup_events++;
ec156764
IM		 377}
378
0e9b07e5
ACM		 379static void add_sched_event_sleep(struct perf_sched *sched, struct task_desc *task,
380 u64 timestamp, u64 task_state __maybe_unused)
ec156764 381{
39aeb52f 382 struct sched_atom *event = get_new_event(task, timestamp);
ec156764
IM		 383
384 event->type = SCHED_EVENT_SLEEP;
385
0e9b07e5 386 sched->nr_sleep_events++;
ec156764
IM		 387}
388
0e9b07e5
ACM		 389static struct task_desc *register_pid(struct perf_sched *sched,
390 unsigned long pid, const char *comm)
ec156764
IM		 391{
392 struct task_desc *task;
cb06ac25 393 static int pid_max;
ec156764 394
cb06ac25
YS		 395 if (sched->pid_to_task == NULL) {
396 if (sysctl__read_int("kernel/pid_max", &pid_max) < 0)
397 pid_max = MAX_PID;
398 BUG_ON((sched->pid_to_task = calloc(pid_max, sizeof(struct task_desc *))) == NULL);
399 }
3a423a5c
YS		 400 if (pid >= (unsigned long)pid_max) {
401 BUG_ON((sched->pid_to_task = realloc(sched->pid_to_task, (pid + 1) *
402 sizeof(struct task_desc *))) == NULL);
403 while (pid >= (unsigned long)pid_max)
404 sched->pid_to_task[pid_max++] = NULL;
405 }
ec156764 406
0e9b07e5 407 task = sched->pid_to_task[pid];
ec156764
IM		 408
409 if (task)
410 return task;
411
36479484 412 task = zalloc(sizeof(*task));
ec156764 413 task->pid = pid;
0e9b07e5 414 task->nr = sched->nr_tasks;
ec156764
IM		 415 strcpy(task->comm, comm);
416 /*
417 * every task starts in sleeping state - this gets ignored
418 * if there's no wakeup pointing to this sleep state:
419 */
0e9b07e5 420 add_sched_event_sleep(sched, task, 0, 0);
ec156764 421
0e9b07e5
ACM		 422 sched->pid_to_task[pid] = task;
423 sched->nr_tasks++;
0755bc4d 424 sched->tasks = realloc(sched->tasks, sched->nr_tasks * sizeof(struct task_desc *));
0e9b07e5
ACM		 425 BUG_ON(!sched->tasks);
426 sched->tasks[task->nr] = task;
ec156764 427
ad236fd2 428 if (verbose)
0e9b07e5 429 printf("registered task #%ld, PID %ld (%s)\n", sched->nr_tasks, pid, comm);
ec156764
IM		 430
431 return task;
432}
433
434
0e9b07e5 435static void print_task_traces(struct perf_sched *sched)
ec156764
IM		 436{
437 struct task_desc *task;
438 unsigned long i;
439
0e9b07e5
ACM		 440 for (i = 0; i < sched->nr_tasks; i++) {
441 task = sched->tasks[i];
ad236fd2 442 printf("task %6ld (%20s:%10ld), nr_events: %ld\n",
ec156764
IM		 443 task->nr, task->comm, task->pid, task->nr_events);
444 }
445}
446
0e9b07e5 447static void add_cross_task_wakeups(struct perf_sched *sched)
ec156764
IM		 448{
449 struct task_desc *task1, *task2;
450 unsigned long i, j;
451
0e9b07e5
ACM		 452 for (i = 0; i < sched->nr_tasks; i++) {
453 task1 = sched->tasks[i];
ec156764 454 j = i + 1;
0e9b07e5 455 if (j == sched->nr_tasks)
ec156764 456 j = 0;
0e9b07e5
ACM		 457 task2 = sched->tasks[j];
458 add_sched_event_wakeup(sched, task1, 0, task2);
ec156764
IM		 459 }
460}
461
0e9b07e5
ACM		 462static void perf_sched__process_event(struct perf_sched *sched,
463 struct sched_atom *atom)
ec156764
IM		 464{
465 int ret = 0;
ec156764 466
39aeb52f 467 switch (atom->type) {
ec156764 468 case SCHED_EVENT_RUN:
0e9b07e5 469 burn_nsecs(sched, atom->duration);
ec156764
IM		 470 break;
471 case SCHED_EVENT_SLEEP:
39aeb52f 472 if (atom->wait_sem)
473 ret = sem_wait(atom->wait_sem);
ec156764
IM		 474 BUG_ON(ret);
475 break;
476 case SCHED_EVENT_WAKEUP:
39aeb52f 477 if (atom->wait_sem)
478 ret = sem_post(atom->wait_sem);
ec156764
IM		 479 BUG_ON(ret);
480 break;
55ffb7a6
MG		 481 case SCHED_EVENT_MIGRATION:
482 break;
ec156764
IM		 483 default:
484 BUG_ON(1);
485 }
486}
487
b1ffe8f3 488static u64 get_cpu_usage_nsec_parent(void)
ec156764
IM		 489{
490 struct rusage ru;
b1ffe8f3 491 u64 sum;
ec156764
IM		 492 int err;
493
494 err = getrusage(RUSAGE_SELF, &ru);
495 BUG_ON(err);
496
4fc76e49
ACM		 497 sum = ru.ru_utime.tv_sec * NSEC_PER_SEC + ru.ru_utime.tv_usec * NSEC_PER_USEC;
498 sum += ru.ru_stime.tv_sec * NSEC_PER_SEC + ru.ru_stime.tv_usec * NSEC_PER_USEC;
ec156764
IM		 499
500 return sum;
501}
502
939cda52 503static int self_open_counters(struct perf_sched *sched, unsigned long cur_task)
ec156764 504{
c0c9e721 505 struct perf_event_attr attr;
939cda52 506 char sbuf[STRERR_BUFSIZE], info[STRERR_BUFSIZE];
c0c9e721 507 int fd;
939cda52
YS		 508 struct rlimit limit;
509 bool need_privilege = false;
ec156764 510
c0c9e721 511 memset(&attr, 0, sizeof(attr));
ec156764 512
c0c9e721
XG		 513 attr.type = PERF_TYPE_SOFTWARE;
514 attr.config = PERF_COUNT_SW_TASK_CLOCK;
ec156764 515
939cda52 516force_again:
57480d2c
YD		 517 fd = sys_perf_event_open(&attr, 0, -1, -1,
518 perf_event_open_cloexec_flag());
c0c9e721 519
1aff59be 520 if (fd < 0) {
939cda52
YS		 521 if (errno == EMFILE) {
522 if (sched->force) {
523 BUG_ON(getrlimit(RLIMIT_NOFILE, &limit) == -1);
524 limit.rlim_cur += sched->nr_tasks - cur_task;
525 if (limit.rlim_cur > limit.rlim_max) {
526 limit.rlim_max = limit.rlim_cur;
527 need_privilege = true;
528 }
529 if (setrlimit(RLIMIT_NOFILE, &limit) == -1) {
530 if (need_privilege && errno == EPERM)
531 strcpy(info, "Need privilege\n");
532 } else
533 goto force_again;
534 } else
535 strcpy(info, "Have a try with -f option\n");
536 }
60b7d14a 537 pr_err("Error: sys_perf_event_open() syscall returned "
939cda52 538 "with %d (%s)\n%s", fd,
c8b5f2c9 539 str_error_r(errno, sbuf, sizeof(sbuf)), info);
1aff59be
YS		 540 exit(EXIT_FAILURE);
541 }
c0c9e721
XG		 542 return fd;
543}
544
545static u64 get_cpu_usage_nsec_self(int fd)
546{
547 u64 runtime;
548 int ret;
549
550 ret = read(fd, &runtime, sizeof(runtime));
551 BUG_ON(ret != sizeof(runtime));
552
553 return runtime;
ec156764
IM		 554}
555
0e9b07e5
ACM		 556struct sched_thread_parms {
557 struct task_desc *task;
558 struct perf_sched *sched;
08097abc 559 int fd;
0e9b07e5
ACM		 560};
561
ec156764
IM		 562static void *thread_func(void *ctx)
563{
0e9b07e5
ACM		 564 struct sched_thread_parms *parms = ctx;
565 struct task_desc *this_task = parms->task;
566 struct perf_sched *sched = parms->sched;
b1ffe8f3 567 u64 cpu_usage_0, cpu_usage_1;
ec156764
IM		 568 unsigned long i, ret;
569 char comm2[22];
08097abc 570 int fd = parms->fd;
ec156764 571
74cf249d 572 zfree(&parms);
0e9b07e5 573
ec156764
IM		 574 sprintf(comm2, ":%s", this_task->comm);
575 prctl(PR_SET_NAME, comm2);
a116e05d
ACM		 576 if (fd < 0)
577 return NULL;
ec156764
IM		 578again:
579 ret = sem_post(&this_task->ready_for_work);
580 BUG_ON(ret);
0e9b07e5 581 ret = pthread_mutex_lock(&sched->start_work_mutex);
ec156764 582 BUG_ON(ret);
0e9b07e5 583 ret = pthread_mutex_unlock(&sched->start_work_mutex);
ec156764 584 BUG_ON(ret);
ec156764 585
c0c9e721 586 cpu_usage_0 = get_cpu_usage_nsec_self(fd);
ec156764
IM		 587
588 for (i = 0; i < this_task->nr_events; i++) {
589 this_task->curr_event = i;
0e9b07e5 590 perf_sched__process_event(sched, this_task->atoms[i]);
ec156764
IM		 591 }
592
c0c9e721 593 cpu_usage_1 = get_cpu_usage_nsec_self(fd);
ec156764 594 this_task->cpu_usage = cpu_usage_1 - cpu_usage_0;
ec156764
IM		 595 ret = sem_post(&this_task->work_done_sem);
596 BUG_ON(ret);
ec156764 597
0e9b07e5 598 ret = pthread_mutex_lock(&sched->work_done_wait_mutex);
ec156764 599 BUG_ON(ret);
0e9b07e5 600 ret = pthread_mutex_unlock(&sched->work_done_wait_mutex);
ec156764 601 BUG_ON(ret);
ec156764
IM		 602
603 goto again;
604}
605
0e9b07e5 606static void create_tasks(struct perf_sched *sched)
ec156764
IM		 607{
608 struct task_desc *task;
609 pthread_attr_t attr;
610 unsigned long i;
611 int err;
612
613 err = pthread_attr_init(&attr);
614 BUG_ON(err);
12f7e036
JP		 615 err = pthread_attr_setstacksize(&attr,
616 (size_t) max(16 * 1024, PTHREAD_STACK_MIN));
ec156764 617 BUG_ON(err);
0e9b07e5 618 err = pthread_mutex_lock(&sched->start_work_mutex);
ec156764 619 BUG_ON(err);
0e9b07e5 620 err = pthread_mutex_lock(&sched->work_done_wait_mutex);
ec156764 621 BUG_ON(err);
0e9b07e5
ACM		 622 for (i = 0; i < sched->nr_tasks; i++) {
623 struct sched_thread_parms *parms = malloc(sizeof(*parms));
624 BUG_ON(parms == NULL);
625 parms->task = task = sched->tasks[i];
626 parms->sched = sched;
939cda52 627 parms->fd = self_open_counters(sched, i);
ec156764
IM		 628 sem_init(&task->sleep_sem, 0, 0);
629 sem_init(&task->ready_for_work, 0, 0);
630 sem_init(&task->work_done_sem, 0, 0);
631 task->curr_event = 0;
0e9b07e5 632 err = pthread_create(&task->thread, &attr, thread_func, parms);
ec156764
IM		 633 BUG_ON(err);
634 }
635}
636
0e9b07e5 637static void wait_for_tasks(struct perf_sched *sched)
ec156764 638{
b1ffe8f3 639 u64 cpu_usage_0, cpu_usage_1;
ec156764
IM		 640 struct task_desc *task;
641 unsigned long i, ret;
642
0e9b07e5
ACM		 643 sched->start_time = get_nsecs();
644 sched->cpu_usage = 0;
645 pthread_mutex_unlock(&sched->work_done_wait_mutex);
ec156764 646
0e9b07e5
ACM		 647 for (i = 0; i < sched->nr_tasks; i++) {
648 task = sched->tasks[i];
ec156764
IM		 649 ret = sem_wait(&task->ready_for_work);
650 BUG_ON(ret);
651 sem_init(&task->ready_for_work, 0, 0);
652 }
0e9b07e5 653 ret = pthread_mutex_lock(&sched->work_done_wait_mutex);
ec156764
IM		 654 BUG_ON(ret);
655
656 cpu_usage_0 = get_cpu_usage_nsec_parent();
657
0e9b07e5 658 pthread_mutex_unlock(&sched->start_work_mutex);
ec156764 659
0e9b07e5
ACM		 660 for (i = 0; i < sched->nr_tasks; i++) {
661 task = sched->tasks[i];
ec156764
IM		 662 ret = sem_wait(&task->work_done_sem);
663 BUG_ON(ret);
664 sem_init(&task->work_done_sem, 0, 0);
0e9b07e5 665 sched->cpu_usage += task->cpu_usage;
ec156764
IM		 666 task->cpu_usage = 0;
667 }
668
669 cpu_usage_1 = get_cpu_usage_nsec_parent();
0e9b07e5
ACM		 670 if (!sched->runavg_cpu_usage)
671 sched->runavg_cpu_usage = sched->cpu_usage;
ff5f3bbd 672 sched->runavg_cpu_usage = (sched->runavg_cpu_usage * (sched->replay_repeat - 1) + sched->cpu_usage) / sched->replay_repeat;
ec156764 673
0e9b07e5
ACM		 674 sched->parent_cpu_usage = cpu_usage_1 - cpu_usage_0;
675 if (!sched->runavg_parent_cpu_usage)
676 sched->runavg_parent_cpu_usage = sched->parent_cpu_usage;
ff5f3bbd
YS		 677 sched->runavg_parent_cpu_usage = (sched->runavg_parent_cpu_usage * (sched->replay_repeat - 1) +
678 sched->parent_cpu_usage)/sched->replay_repeat;
ec156764 679
0e9b07e5 680 ret = pthread_mutex_lock(&sched->start_work_mutex);
ec156764
IM		 681 BUG_ON(ret);
682
0e9b07e5
ACM		 683 for (i = 0; i < sched->nr_tasks; i++) {
684 task = sched->tasks[i];
ec156764
IM		 685 sem_init(&task->sleep_sem, 0, 0);
686 task->curr_event = 0;
687 }
688}
689
0e9b07e5 690static void run_one_test(struct perf_sched *sched)
ec156764 691{
fb7d0b3c 692 u64 T0, T1, delta, avg_delta, fluct;
ec156764
IM		 693
694 T0 = get_nsecs();
0e9b07e5 695 wait_for_tasks(sched);
ec156764
IM		 696 T1 = get_nsecs();
697
698 delta = T1 - T0;
0e9b07e5
ACM		 699 sched->sum_runtime += delta;
700 sched->nr_runs++;
ec156764 701
0e9b07e5 702 avg_delta = sched->sum_runtime / sched->nr_runs;
ec156764
IM		 703 if (delta < avg_delta)
704 fluct = avg_delta - delta;
705 else
706 fluct = delta - avg_delta;
0e9b07e5
ACM		 707 sched->sum_fluct += fluct;
708 if (!sched->run_avg)
709 sched->run_avg = delta;
ff5f3bbd 710 sched->run_avg = (sched->run_avg * (sched->replay_repeat - 1) + delta) / sched->replay_repeat;
ec156764 711
4fc76e49 712 printf("#%-3ld: %0.3f, ", sched->nr_runs, (double)delta / NSEC_PER_MSEC);
ec156764 713
4fc76e49 714 printf("ravg: %0.2f, ", (double)sched->run_avg / NSEC_PER_MSEC);
ec156764 715
ad236fd2 716 printf("cpu: %0.2f / %0.2f",
4fc76e49 717 (double)sched->cpu_usage / NSEC_PER_MSEC, (double)sched->runavg_cpu_usage / NSEC_PER_MSEC);
ec156764
IM		 718
719#if 0
720 /*
fbf94829 721 * rusage statistics done by the parent, these are less
0e9b07e5 722 * accurate than the sched->sum_exec_runtime based statistics:
fbf94829 723 */
ad236fd2 724 printf(" [%0.2f / %0.2f]",
4fc76e49
ACM		 725 (double)sched->parent_cpu_usage / NSEC_PER_MSEC,
726 (double)sched->runavg_parent_cpu_usage / NSEC_PER_MSEC);
ec156764
IM		 727#endif
728
ad236fd2 729 printf("\n");
ec156764 730
0e9b07e5
ACM		 731 if (sched->nr_sleep_corrections)
732 printf(" (%ld sleep corrections)\n", sched->nr_sleep_corrections);
733 sched->nr_sleep_corrections = 0;
ec156764
IM		 734}
735
0e9b07e5 736static void test_calibrations(struct perf_sched *sched)
ec156764 737{
b1ffe8f3 738 u64 T0, T1;
ec156764
IM		 739
740 T0 = get_nsecs();
4fc76e49 741 burn_nsecs(sched, NSEC_PER_MSEC);
ec156764
IM		 742 T1 = get_nsecs();
743
9486aa38 744 printf("the run test took %" PRIu64 " nsecs\n", T1 - T0);
ec156764
IM		 745
746 T0 = get_nsecs();
4fc76e49 747 sleep_nsecs(NSEC_PER_MSEC);
ec156764
IM		 748 T1 = get_nsecs();
749
9486aa38 750 printf("the sleep test took %" PRIu64 " nsecs\n", T1 - T0);
ec156764
IM		 751}
752
a116e05d 753static int
0e9b07e5 754replay_wakeup_event(struct perf_sched *sched,
9ec3f4e4
ACM		 755 struct perf_evsel *evsel, struct perf_sample *sample,
756 struct machine *machine __maybe_unused)
419ab0d6 757{
9ec3f4e4
ACM		 758 const char *comm = perf_evsel__strval(evsel, sample, "comm");
759 const u32 pid = perf_evsel__intval(evsel, sample, "pid");
419ab0d6 760 struct task_desc *waker, *wakee;
fbf94829 761
ad236fd2 762 if (verbose) {
2b7fcbc5 763 printf("sched_wakeup event %p\n", evsel);
fbf94829 764
9ec3f4e4 765 printf(" ... pid %d woke up %s/%d\n", sample->tid, comm, pid);
ad236fd2 766 }
fbf94829 767
2b7fcbc5 768 waker = register_pid(sched, sample->tid, "<unknown>");
9ec3f4e4 769 wakee = register_pid(sched, pid, comm);
fbf94829 770
0e9b07e5 771 add_sched_event_wakeup(sched, waker, sample->time, wakee);
a116e05d 772 return 0;
ec156764
IM		 773}
774
9ec3f4e4
ACM		 775static int replay_switch_event(struct perf_sched *sched,
776 struct perf_evsel *evsel,
777 struct perf_sample *sample,
778 struct machine *machine __maybe_unused)
ec156764 779{
9ec3f4e4
ACM		 780 const char *prev_comm = perf_evsel__strval(evsel, sample, "prev_comm"),
781 *next_comm = perf_evsel__strval(evsel, sample, "next_comm");
782 const u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"),
783 next_pid = perf_evsel__intval(evsel, sample, "next_pid");
784 const u64 prev_state = perf_evsel__intval(evsel, sample, "prev_state");
1d037ca1 785 struct task_desc *prev, __maybe_unused *next;
7f7f8d0b
ACM		 786 u64 timestamp0, timestamp = sample->time;
787 int cpu = sample->cpu;
fbf94829
IM		 788 s64 delta;
789
ad236fd2 790 if (verbose)
2b7fcbc5 791 printf("sched_switch event %p\n", evsel);
ad236fd2 792
fbf94829 793 if (cpu >= MAX_CPUS || cpu < 0)
a116e05d 794 return 0;
fbf94829 795
0e9b07e5 796 timestamp0 = sched->cpu_last_switched[cpu];
fbf94829
IM		 797 if (timestamp0)
798 delta = timestamp - timestamp0;
799 else
800 delta = 0;
801
a116e05d 802 if (delta < 0) {
60b7d14a 803 pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta);
a116e05d
ACM		 804 return -1;
805 }
fbf94829 806
9ec3f4e4
ACM		 807 pr_debug(" ... switch from %s/%d to %s/%d [ran %" PRIu64 " nsecs]\n",
808 prev_comm, prev_pid, next_comm, next_pid, delta);
fbf94829 809
9ec3f4e4
ACM		 810 prev = register_pid(sched, prev_pid, prev_comm);
811 next = register_pid(sched, next_pid, next_comm);
fbf94829 812
0e9b07e5 813 sched->cpu_last_switched[cpu] = timestamp;
fbf94829 814
0e9b07e5 815 add_sched_event_run(sched, prev, timestamp, delta);
9ec3f4e4 816 add_sched_event_sleep(sched, prev, timestamp, prev_state);
a116e05d
ACM		 817
818 return 0;
fbf94829
IM		 819}
820
cb627505
DA		 821static int replay_fork_event(struct perf_sched *sched,
822 union perf_event *event,
823 struct machine *machine)
419ab0d6 824{
cb627505
DA		 825 struct thread *child, *parent;
826
314add6b
AH		 827 child = machine__findnew_thread(machine, event->fork.pid,
828 event->fork.tid);
829 parent = machine__findnew_thread(machine, event->fork.ppid,
830 event->fork.ptid);
cb627505
DA		 831
832 if (child == NULL || parent == NULL) {
833 pr_debug("thread does not exist on fork event: child %p, parent %p\n",
834 child, parent);
b91fc39f 835 goto out_put;
cb627505 836 }
9ec3f4e4 837
419ab0d6 838 if (verbose) {
cb627505 839 printf("fork event\n");
b9c5143a
FW		 840 printf("... parent: %s/%d\n", thread__comm_str(parent), parent->tid);
841 printf("... child: %s/%d\n", thread__comm_str(child), child->tid);
419ab0d6 842 }
9ec3f4e4 843
b9c5143a
FW		 844 register_pid(sched, parent->tid, thread__comm_str(parent));
845 register_pid(sched, child->tid, thread__comm_str(child));
b91fc39f
ACM		 846out_put:
847 thread__put(child);
848 thread__put(parent);
a116e05d 849 return 0;
419ab0d6 850}
fbf94829 851
b1ffe8f3
IM		 852struct sort_dimension {
853 const char *name;
b5fae128 854 sort_fn_t cmp;
b1ffe8f3
IM		 855 struct list_head list;
856};
857
daa1d7a5 858static int
39aeb52f 859thread_lat_cmp(struct list_head *list, struct work_atoms *l, struct work_atoms *r)
daa1d7a5
FW		 860{
861 struct sort_dimension *sort;
862 int ret = 0;
863
b5fae128
IM		 864 BUG_ON(list_empty(list));
865
daa1d7a5
FW		 866 list_for_each_entry(sort, list, list) {
867 ret = sort->cmp(l, r);
868 if (ret)
869 return ret;
870 }
871
872 return ret;
873}
874
39aeb52f 875static struct work_atoms *
b5fae128
IM		 876thread_atoms_search(struct rb_root *root, struct thread *thread,
877 struct list_head *sort_list)
878{
879 struct rb_node *node = root->rb_node;
39aeb52f 880 struct work_atoms key = { .thread = thread };
b5fae128
IM		 881
882 while (node) {
39aeb52f 883 struct work_atoms *atoms;
b5fae128
IM		 884 int cmp;
885
39aeb52f 886 atoms = container_of(node, struct work_atoms, node);
b5fae128
IM		 887
888 cmp = thread_lat_cmp(sort_list, &key, atoms);
889 if (cmp > 0)
890 node = node->rb_left;
891 else if (cmp < 0)
892 node = node->rb_right;
893 else {
894 BUG_ON(thread != atoms->thread);
895 return atoms;
896 }
897 }
898 return NULL;
899}
900
cdce9d73 901static void
39aeb52f 902__thread_latency_insert(struct rb_root *root, struct work_atoms *data,
daa1d7a5 903 struct list_head *sort_list)
cdce9d73
FW		 904{
905 struct rb_node **new = &(root->rb_node), *parent = NULL;
906
907 while (*new) {
39aeb52f 908 struct work_atoms *this;
daa1d7a5 909 int cmp;
cdce9d73 910
39aeb52f 911 this = container_of(*new, struct work_atoms, node);
cdce9d73 912 parent = *new;
daa1d7a5
FW		 913
914 cmp = thread_lat_cmp(sort_list, data, this);
915
916 if (cmp > 0)
cdce9d73 917 new = &((*new)->rb_left);
cdce9d73 918 else
daa1d7a5 919 new = &((*new)->rb_right);
cdce9d73
FW		 920 }
921
922 rb_link_node(&data->node, parent, new);
923 rb_insert_color(&data->node, root);
924}
925
0e9b07e5 926static int thread_atoms_insert(struct perf_sched *sched, struct thread *thread)
cdce9d73 927{
36479484 928 struct work_atoms *atoms = zalloc(sizeof(*atoms));
a116e05d
ACM		 929 if (!atoms) {
930 pr_err("No memory at %s\n", __func__);
931 return -1;
932 }
cdce9d73 933
f3b623b8 934 atoms->thread = thread__get(thread);
39aeb52f 935 INIT_LIST_HEAD(&atoms->work_list);
0e9b07e5 936 __thread_latency_insert(&sched->atom_root, atoms, &sched->cmp_pid);
a116e05d 937 return 0;
cdce9d73
FW		 938}
939
9ec3f4e4 940static char sched_out_state(u64 prev_state)
cdce9d73
FW		 941{
942 const char *str = TASK_STATE_TO_CHAR_STR;
943
9ec3f4e4 944 return str[prev_state];
cdce9d73
FW		 945}
946
a116e05d 947static int
39aeb52f 948add_sched_out_event(struct work_atoms *atoms,
949 char run_state,
950 u64 timestamp)
cdce9d73 951{
36479484 952 struct work_atom *atom = zalloc(sizeof(*atom));
a116e05d
ACM		 953 if (!atom) {
954 pr_err("Non memory at %s", __func__);
955 return -1;
956 }
cdce9d73 957
aa1ab9d2
FW		 958 atom->sched_out_time = timestamp;
959
39aeb52f 960 if (run_state == 'R') {
b1ffe8f3 961 atom->state = THREAD_WAIT_CPU;
aa1ab9d2 962 atom->wake_up_time = atom->sched_out_time;
c6ced611
FW		 963 }
964
39aeb52f 965 list_add_tail(&atom->list, &atoms->work_list);
a116e05d 966 return 0;
cdce9d73
FW		 967}
968
969static void
1d037ca1
IT		 970add_runtime_event(struct work_atoms *atoms, u64 delta,
971 u64 timestamp __maybe_unused)
39aeb52f 972{
973 struct work_atom *atom;
974
975 BUG_ON(list_empty(&atoms->work_list));
976
977 atom = list_entry(atoms->work_list.prev, struct work_atom, list);
978
979 atom->runtime += delta;
980 atoms->total_runtime += delta;
981}
982
983static void
984add_sched_in_event(struct work_atoms *atoms, u64 timestamp)
cdce9d73 985{
b1ffe8f3 986 struct work_atom *atom;
66685678 987 u64 delta;
cdce9d73 988
39aeb52f 989 if (list_empty(&atoms->work_list))
cdce9d73
FW		 990 return;
991
39aeb52f 992 atom = list_entry(atoms->work_list.prev, struct work_atom, list);
cdce9d73 993
b1ffe8f3 994 if (atom->state != THREAD_WAIT_CPU)
cdce9d73
FW		 995 return;
996
b1ffe8f3
IM		 997 if (timestamp < atom->wake_up_time) {
998 atom->state = THREAD_IGNORE;
cdce9d73
FW		 999 return;
1000 }
1001
b1ffe8f3
IM		 1002 atom->state = THREAD_SCHED_IN;
1003 atom->sched_in_time = timestamp;
66685678 1004
b1ffe8f3 1005 delta = atom->sched_in_time - atom->wake_up_time;
66685678 1006 atoms->total_lat += delta;
3786310a 1007 if (delta > atoms->max_lat) {
66685678 1008 atoms->max_lat = delta;
3786310a
FW		 1009 atoms->max_lat_at = timestamp;
1010 }
66685678 1011 atoms->nb_atoms++;
cdce9d73
FW		 1012}
1013
9ec3f4e4
ACM		 1014static int latency_switch_event(struct perf_sched *sched,
1015 struct perf_evsel *evsel,
1016 struct perf_sample *sample,
1017 struct machine *machine)
cdce9d73 1018{
9ec3f4e4
ACM		 1019 const u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"),
1020 next_pid = perf_evsel__intval(evsel, sample, "next_pid");
1021 const u64 prev_state = perf_evsel__intval(evsel, sample, "prev_state");
39aeb52f 1022 struct work_atoms *out_events, *in_events;
cdce9d73 1023 struct thread *sched_out, *sched_in;
7f7f8d0b 1024 u64 timestamp0, timestamp = sample->time;
b91fc39f 1025 int cpu = sample->cpu, err = -1;
ea92ed5a
IM		 1026 s64 delta;
1027
39aeb52f 1028 BUG_ON(cpu >= MAX_CPUS || cpu < 0);
ea92ed5a 1029
0e9b07e5
ACM		 1030 timestamp0 = sched->cpu_last_switched[cpu];
1031 sched->cpu_last_switched[cpu] = timestamp;
ea92ed5a
IM		 1032 if (timestamp0)
1033 delta = timestamp - timestamp0;
1034 else
1035 delta = 0;
1036
a116e05d
ACM		 1037 if (delta < 0) {
1038 pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta);
1039 return -1;
1040 }
cdce9d73 1041
1fcb8768
AH		 1042 sched_out = machine__findnew_thread(machine, -1, prev_pid);
1043 sched_in = machine__findnew_thread(machine, -1, next_pid);
b91fc39f
ACM		 1044 if (sched_out == NULL || sched_in == NULL)
1045 goto out_put;
cdce9d73 1046
0e9b07e5 1047 out_events = thread_atoms_search(&sched->atom_root, sched_out, &sched->cmp_pid);
39aeb52f 1048 if (!out_events) {
0e9b07e5 1049 if (thread_atoms_insert(sched, sched_out))
b91fc39f 1050 goto out_put;
0e9b07e5 1051 out_events = thread_atoms_search(&sched->atom_root, sched_out, &sched->cmp_pid);
a116e05d
ACM		 1052 if (!out_events) {
1053 pr_err("out-event: Internal tree error");
b91fc39f 1054 goto out_put;
a116e05d 1055 }
39aeb52f 1056 }
9ec3f4e4 1057 if (add_sched_out_event(out_events, sched_out_state(prev_state), timestamp))
a116e05d 1058 return -1;
39aeb52f 1059
0e9b07e5 1060 in_events = thread_atoms_search(&sched->atom_root, sched_in, &sched->cmp_pid);
39aeb52f 1061 if (!in_events) {
0e9b07e5 1062 if (thread_atoms_insert(sched, sched_in))
b91fc39f 1063 goto out_put;
0e9b07e5 1064 in_events = thread_atoms_search(&sched->atom_root, sched_in, &sched->cmp_pid);
a116e05d
ACM		 1065 if (!in_events) {
1066 pr_err("in-event: Internal tree error");
b91fc39f 1067 goto out_put;
a116e05d 1068 }
39aeb52f 1069 /*
1070 * Take came in we have not heard about yet,
1071 * add in an initial atom in runnable state:
1072 */
a116e05d 1073 if (add_sched_out_event(in_events, 'R', timestamp))
b91fc39f 1074 goto out_put;
cdce9d73 1075 }
39aeb52f 1076 add_sched_in_event(in_events, timestamp);
b91fc39f
ACM		 1077 err = 0;
1078out_put:
1079 thread__put(sched_out);
1080 thread__put(sched_in);
1081 return err;
39aeb52f 1082}
cdce9d73 1083
9ec3f4e4
ACM		 1084static int latency_runtime_event(struct perf_sched *sched,
1085 struct perf_evsel *evsel,
1086 struct perf_sample *sample,
1087 struct machine *machine)
39aeb52f 1088{
9ec3f4e4
ACM		 1089 const u32 pid = perf_evsel__intval(evsel, sample, "pid");
1090 const u64 runtime = perf_evsel__intval(evsel, sample, "runtime");
1fcb8768 1091 struct thread *thread = machine__findnew_thread(machine, -1, pid);
0e9b07e5 1092 struct work_atoms *atoms = thread_atoms_search(&sched->atom_root, thread, &sched->cmp_pid);
7f7f8d0b 1093 u64 timestamp = sample->time;
b91fc39f
ACM		 1094 int cpu = sample->cpu, err = -1;
1095
1096 if (thread == NULL)
1097 return -1;
39aeb52f 1098
1099 BUG_ON(cpu >= MAX_CPUS || cpu < 0);
39aeb52f 1100 if (!atoms) {
0e9b07e5 1101 if (thread_atoms_insert(sched, thread))
b91fc39f 1102 goto out_put;
0e9b07e5 1103 atoms = thread_atoms_search(&sched->atom_root, thread, &sched->cmp_pid);
a116e05d 1104 if (!atoms) {
60b7d14a 1105 pr_err("in-event: Internal tree error");
b91fc39f 1106 goto out_put;
a116e05d
ACM		 1107 }
1108 if (add_sched_out_event(atoms, 'R', timestamp))
b91fc39f 1109 goto out_put;
cdce9d73
FW		 1110 }
1111
9ec3f4e4 1112 add_runtime_event(atoms, runtime, timestamp);
b91fc39f
ACM		 1113 err = 0;
1114out_put:
1115 thread__put(thread);
1116 return err;
cdce9d73
FW		 1117}
1118
9ec3f4e4
ACM		 1119static int latency_wakeup_event(struct perf_sched *sched,
1120 struct perf_evsel *evsel,
1121 struct perf_sample *sample,
1122 struct machine *machine)
cdce9d73 1123{
0680ee7d 1124 const u32 pid = perf_evsel__intval(evsel, sample, "pid");
39aeb52f 1125 struct work_atoms *atoms;
b1ffe8f3 1126 struct work_atom *atom;
cdce9d73 1127 struct thread *wakee;
7f7f8d0b 1128 u64 timestamp = sample->time;
b91fc39f 1129 int err = -1;
cdce9d73 1130
1fcb8768 1131 wakee = machine__findnew_thread(machine, -1, pid);
b91fc39f
ACM		 1132 if (wakee == NULL)
1133 return -1;
0e9b07e5 1134 atoms = thread_atoms_search(&sched->atom_root, wakee, &sched->cmp_pid);
17562205 1135 if (!atoms) {
0e9b07e5 1136 if (thread_atoms_insert(sched, wakee))
b91fc39f 1137 goto out_put;
0e9b07e5 1138 atoms = thread_atoms_search(&sched->atom_root, wakee, &sched->cmp_pid);
a116e05d 1139 if (!atoms) {
60b7d14a 1140 pr_err("wakeup-event: Internal tree error");
b91fc39f 1141 goto out_put;
a116e05d
ACM		 1142 }
1143 if (add_sched_out_event(atoms, 'S', timestamp))
b91fc39f 1144 goto out_put;
cdce9d73
FW		 1145 }
1146
39aeb52f 1147 BUG_ON(list_empty(&atoms->work_list));
cdce9d73 1148
39aeb52f 1149 atom = list_entry(atoms->work_list.prev, struct work_atom, list);
cdce9d73 1150
55ffb7a6 1151 /*
67d6259d
DY		 1152 * As we do not guarantee the wakeup event happens when
1153 * task is out of run queue, also may happen when task is
1154 * on run queue and wakeup only change ->state to TASK_RUNNING,
1155 * then we should not set the ->wake_up_time when wake up a
1156 * task which is on run queue.
1157 *
55ffb7a6
MG		 1158 * You WILL be missing events if you've recorded only
1159 * one CPU, or are only looking at only one, so don't
67d6259d 1160 * skip in this case.
55ffb7a6 1161 */
0e9b07e5 1162 if (sched->profile_cpu == -1 && atom->state != THREAD_SLEEPING)
b91fc39f 1163 goto out_ok;
cdce9d73 1164
0e9b07e5 1165 sched->nr_timestamps++;
ea57c4f5 1166 if (atom->sched_out_time > timestamp) {
0e9b07e5 1167 sched->nr_unordered_timestamps++;
b91fc39f 1168 goto out_ok;
ea57c4f5 1169 }
aa1ab9d2 1170
b1ffe8f3
IM		 1171 atom->state = THREAD_WAIT_CPU;
1172 atom->wake_up_time = timestamp;
b91fc39f
ACM		 1173out_ok:
1174 err = 0;
1175out_put:
1176 thread__put(wakee);
1177 return err;
cdce9d73
FW		 1178}
1179
9ec3f4e4
ACM		 1180static int latency_migrate_task_event(struct perf_sched *sched,
1181 struct perf_evsel *evsel,
1182 struct perf_sample *sample,
1183 struct machine *machine)
55ffb7a6 1184{
9ec3f4e4 1185 const u32 pid = perf_evsel__intval(evsel, sample, "pid");
7f7f8d0b 1186 u64 timestamp = sample->time;
55ffb7a6
MG		 1187 struct work_atoms *atoms;
1188 struct work_atom *atom;
1189 struct thread *migrant;
b91fc39f 1190 int err = -1;
55ffb7a6
MG		 1191
1192 /*
1193 * Only need to worry about migration when profiling one CPU.
1194 */
0e9b07e5 1195 if (sched->profile_cpu == -1)
a116e05d 1196 return 0;
55ffb7a6 1197
1fcb8768 1198 migrant = machine__findnew_thread(machine, -1, pid);
b91fc39f
ACM		 1199 if (migrant == NULL)
1200 return -1;
0e9b07e5 1201 atoms = thread_atoms_search(&sched->atom_root, migrant, &sched->cmp_pid);
55ffb7a6 1202 if (!atoms) {
0e9b07e5 1203 if (thread_atoms_insert(sched, migrant))
b91fc39f 1204 goto out_put;
b9c5143a 1205 register_pid(sched, migrant->tid, thread__comm_str(migrant));
0e9b07e5 1206 atoms = thread_atoms_search(&sched->atom_root, migrant, &sched->cmp_pid);
a116e05d 1207 if (!atoms) {
60b7d14a 1208 pr_err("migration-event: Internal tree error");
b91fc39f 1209 goto out_put;
a116e05d
ACM		 1210 }
1211 if (add_sched_out_event(atoms, 'R', timestamp))
b91fc39f 1212 goto out_put;
55ffb7a6
MG		 1213 }
1214
1215 BUG_ON(list_empty(&atoms->work_list));
1216
1217 atom = list_entry(atoms->work_list.prev, struct work_atom, list);
1218 atom->sched_in_time = atom->sched_out_time = atom->wake_up_time = timestamp;
1219
0e9b07e5 1220 sched->nr_timestamps++;
55ffb7a6
MG		 1221
1222 if (atom->sched_out_time > timestamp)
0e9b07e5 1223 sched->nr_unordered_timestamps++;
b91fc39f
ACM		 1224 err = 0;
1225out_put:
1226 thread__put(migrant);
1227 return err;
55ffb7a6
MG		 1228}
1229
0e9b07e5 1230static void output_lat_thread(struct perf_sched *sched, struct work_atoms *work_list)
cdce9d73 1231{
cdce9d73
FW		 1232 int i;
1233 int ret;
66685678 1234 u64 avg;
99620a5d 1235 char max_lat_at[32];
cdce9d73 1236
39aeb52f 1237 if (!work_list->nb_atoms)
cdce9d73 1238 return;
ea57c4f5
IM		 1239 /*
1240 * Ignore idle threads:
1241 */
b9c5143a 1242 if (!strcmp(thread__comm_str(work_list->thread), "swapper"))
ea57c4f5 1243 return;
cdce9d73 1244
0e9b07e5
ACM		 1245 sched->all_runtime += work_list->total_runtime;
1246 sched->all_count += work_list->nb_atoms;
66685678 1247
2f80dd44
JB		 1248 if (work_list->num_merged > 1)
1249 ret = printf(" %s:(%d) ", thread__comm_str(work_list->thread), work_list->num_merged);
1250 else
1251 ret = printf(" %s:%d ", thread__comm_str(work_list->thread), work_list->thread->tid);
cdce9d73 1252
08f69e6c 1253 for (i = 0; i < 24 - ret; i++)
cdce9d73
FW		 1254 printf(" ");
1255
39aeb52f 1256 avg = work_list->total_lat / work_list->nb_atoms;
99620a5d 1257 timestamp__scnprintf_usec(work_list->max_lat_at, max_lat_at, sizeof(max_lat_at));
cdce9d73 1258
99620a5d 1259 printf("|%11.3f ms |%9" PRIu64 " | avg:%9.3f ms | max:%9.3f ms | max at: %13s s\n",
4fc76e49
ACM		 1260 (double)work_list->total_runtime / NSEC_PER_MSEC,
1261 work_list->nb_atoms, (double)avg / NSEC_PER_MSEC,
1262 (double)work_list->max_lat / NSEC_PER_MSEC,
99620a5d 1263 max_lat_at);
cdce9d73
FW		 1264}
1265
39aeb52f 1266static int pid_cmp(struct work_atoms *l, struct work_atoms *r)
daa1d7a5 1267{
0014de17
JO		 1268 if (l->thread == r->thread)
1269 return 0;
38051234 1270 if (l->thread->tid < r->thread->tid)
daa1d7a5 1271 return -1;
38051234 1272 if (l->thread->tid > r->thread->tid)
daa1d7a5 1273 return 1;
0014de17 1274 return (int)(l->thread - r->thread);
daa1d7a5
FW		 1275}
1276
39aeb52f 1277static int avg_cmp(struct work_atoms *l, struct work_atoms *r)
daa1d7a5
FW		 1278{
1279 u64 avgl, avgr;
1280
1281 if (!l->nb_atoms)
1282 return -1;
1283
1284 if (!r->nb_atoms)
1285 return 1;
1286
1287 avgl = l->total_lat / l->nb_atoms;
1288 avgr = r->total_lat / r->nb_atoms;
1289
1290 if (avgl < avgr)
1291 return -1;
1292 if (avgl > avgr)
1293 return 1;
1294
1295 return 0;
1296}
1297
39aeb52f 1298static int max_cmp(struct work_atoms *l, struct work_atoms *r)
daa1d7a5
FW		 1299{
1300 if (l->max_lat < r->max_lat)
1301 return -1;
1302 if (l->max_lat > r->max_lat)
1303 return 1;
1304
1305 return 0;
1306}
1307
39aeb52f 1308static int switch_cmp(struct work_atoms *l, struct work_atoms *r)
daa1d7a5
FW		 1309{
1310 if (l->nb_atoms < r->nb_atoms)
1311 return -1;
1312 if (l->nb_atoms > r->nb_atoms)
1313 return 1;
1314
1315 return 0;
1316}
1317
39aeb52f 1318static int runtime_cmp(struct work_atoms *l, struct work_atoms *r)
daa1d7a5
FW		 1319{
1320 if (l->total_runtime < r->total_runtime)
1321 return -1;
1322 if (l->total_runtime > r->total_runtime)
1323 return 1;
1324
1325 return 0;
1326}
1327
cbef79a8 1328static int sort_dimension__add(const char *tok, struct list_head *list)
daa1d7a5 1329{
0e9b07e5
ACM		 1330 size_t i;
1331 static struct sort_dimension avg_sort_dimension = {
1332 .name = "avg",
1333 .cmp = avg_cmp,
1334 };
1335 static struct sort_dimension max_sort_dimension = {
1336 .name = "max",
1337 .cmp = max_cmp,
1338 };
1339 static struct sort_dimension pid_sort_dimension = {
1340 .name = "pid",
1341 .cmp = pid_cmp,
1342 };
1343 static struct sort_dimension runtime_sort_dimension = {
1344 .name = "runtime",
1345 .cmp = runtime_cmp,
1346 };
1347 static struct sort_dimension switch_sort_dimension = {
1348 .name = "switch",
1349 .cmp = switch_cmp,
1350 };
1351 struct sort_dimension *available_sorts[] = {
1352 &pid_sort_dimension,
1353 &avg_sort_dimension,
1354 &max_sort_dimension,
1355 &switch_sort_dimension,
1356 &runtime_sort_dimension,
1357 };
daa1d7a5 1358
0e9b07e5 1359 for (i = 0; i < ARRAY_SIZE(available_sorts); i++) {
daa1d7a5
FW		 1360 if (!strcmp(available_sorts[i]->name, tok)) {
1361 list_add_tail(&available_sorts[i]->list, list);
1362
1363 return 0;
1364 }
1365 }
1366
1367 return -1;
1368}
1369
0e9b07e5 1370static void perf_sched__sort_lat(struct perf_sched *sched)
daa1d7a5
FW		 1371{
1372 struct rb_node *node;
2f80dd44
JB		 1373 struct rb_root *root = &sched->atom_root;
1374again:
daa1d7a5 1375 for (;;) {
39aeb52f 1376 struct work_atoms *data;
2f80dd44 1377 node = rb_first(root);
daa1d7a5
FW		 1378 if (!node)
1379 break;
1380
2f80dd44 1381 rb_erase(node, root);
39aeb52f 1382 data = rb_entry(node, struct work_atoms, node);
0e9b07e5 1383 __thread_latency_insert(&sched->sorted_atom_root, data, &sched->sort_list);
daa1d7a5 1384 }
2f80dd44
JB		 1385 if (root == &sched->atom_root) {
1386 root = &sched->merged_atom_root;
1387 goto again;
1388 }
daa1d7a5
FW		 1389}
1390
0e9b07e5 1391static int process_sched_wakeup_event(struct perf_tool *tool,
2b7fcbc5 1392 struct perf_evsel *evsel,
1d037ca1 1393 struct perf_sample *sample,
4218e673 1394 struct machine *machine)
419ab0d6 1395{
0e9b07e5 1396 struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
419ab0d6 1397
9ec3f4e4
ACM		 1398 if (sched->tp_handler->wakeup_event)
1399 return sched->tp_handler->wakeup_event(sched, evsel, sample, machine);
a116e05d 1400
2b7fcbc5 1401 return 0;
419ab0d6
FW		 1402}
1403
a151a37a
JO		 1404union map_priv {
1405 void *ptr;
1406 bool color;
1407};
1408
1409static bool thread__has_color(struct thread *thread)
1410{
1411 union map_priv priv = {
1412 .ptr = thread__priv(thread),
1413 };
1414
1415 return priv.color;
1416}
1417
1418static struct thread*
1419map__findnew_thread(struct perf_sched *sched, struct machine *machine, pid_t pid, pid_t tid)
1420{
1421 struct thread *thread = machine__findnew_thread(machine, pid, tid);
1422 union map_priv priv = {
1423 .color = false,
1424 };
1425
1426 if (!sched->map.color_pids || !thread || thread__priv(thread))
1427 return thread;
1428
1429 if (thread_map__has(sched->map.color_pids, tid))
1430 priv.color = true;
1431
1432 thread__set_priv(thread, priv.ptr);
1433 return thread;
1434}
1435
9ec3f4e4
ACM		 1436static int map_switch_event(struct perf_sched *sched, struct perf_evsel *evsel,
1437 struct perf_sample *sample, struct machine *machine)
0ec04e16 1438{
9d372ca5
DY		 1439 const u32 next_pid = perf_evsel__intval(evsel, sample, "next_pid");
1440 struct thread *sched_in;
0ec04e16 1441 int new_shortname;
7f7f8d0b 1442 u64 timestamp0, timestamp = sample->time;
0ec04e16 1443 s64 delta;
99623c62
JO		 1444 int i, this_cpu = sample->cpu;
1445 int cpus_nr;
1446 bool new_cpu = false;
8cd91195 1447 const char *color = PERF_COLOR_NORMAL;
99620a5d 1448 char stimestamp[32];
0ec04e16
IM		 1449
1450 BUG_ON(this_cpu >= MAX_CPUS || this_cpu < 0);
1451
0e9b07e5
ACM		 1452 if (this_cpu > sched->max_cpu)
1453 sched->max_cpu = this_cpu;
0ec04e16 1454
99623c62
JO		 1455 if (sched->map.comp) {
1456 cpus_nr = bitmap_weight(sched->map.comp_cpus_mask, MAX_CPUS);
1457 if (!test_and_set_bit(this_cpu, sched->map.comp_cpus_mask)) {
1458 sched->map.comp_cpus[cpus_nr++] = this_cpu;
1459 new_cpu = true;
1460 }
1461 } else
1462 cpus_nr = sched->max_cpu;
1463
0e9b07e5
ACM		 1464 timestamp0 = sched->cpu_last_switched[this_cpu];
1465 sched->cpu_last_switched[this_cpu] = timestamp;
0ec04e16
IM		 1466 if (timestamp0)
1467 delta = timestamp - timestamp0;
1468 else
1469 delta = 0;
1470
a116e05d 1471 if (delta < 0) {
60b7d14a 1472 pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta);
a116e05d
ACM		 1473 return -1;
1474 }
0ec04e16 1475
a151a37a 1476 sched_in = map__findnew_thread(sched, machine, -1, next_pid);
b91fc39f
ACM		 1477 if (sched_in == NULL)
1478 return -1;
0ec04e16 1479
b91fc39f 1480 sched->curr_thread[this_cpu] = thread__get(sched_in);
0ec04e16
IM		 1481
1482 printf(" ");
1483
1484 new_shortname = 0;
1485 if (!sched_in->shortname[0]) {
6bcab4e1
D		 1486 if (!strcmp(thread__comm_str(sched_in), "swapper")) {
1487 /*
1488 * Don't allocate a letter-number for swapper:0
1489 * as a shortname. Instead, we use '.' for it.
1490 */
1491 sched_in->shortname[0] = '.';
1492 sched_in->shortname[1] = ' ';
0ec04e16 1493 } else {
6bcab4e1
D		 1494 sched_in->shortname[0] = sched->next_shortname1;
1495 sched_in->shortname[1] = sched->next_shortname2;
1496
1497 if (sched->next_shortname1 < 'Z') {
1498 sched->next_shortname1++;
0ec04e16 1499 } else {
6bcab4e1
D		 1500 sched->next_shortname1 = 'A';
1501 if (sched->next_shortname2 < '9')
1502 sched->next_shortname2++;
1503 else
1504 sched->next_shortname2 = '0';
0ec04e16
IM		 1505 }
1506 }
1507 new_shortname = 1;
1508 }
1509
99623c62
JO		 1510 for (i = 0; i < cpus_nr; i++) {
1511 int cpu = sched->map.comp ? sched->map.comp_cpus[i] : i;
a151a37a
JO		 1512 struct thread *curr_thread = sched->curr_thread[cpu];
1513 const char *pid_color = color;
cf294f24 1514 const char *cpu_color = color;
a151a37a
JO		 1515
1516 if (curr_thread && thread__has_color(curr_thread))
1517 pid_color = COLOR_PIDS;
99623c62 1518
73643bb6
JO		 1519 if (sched->map.cpus && !cpu_map__has(sched->map.cpus, cpu))
1520 continue;
1521
cf294f24
JO		 1522 if (sched->map.color_cpus && cpu_map__has(sched->map.color_cpus, cpu))
1523 cpu_color = COLOR_CPUS;
1524
0ec04e16 1525 if (cpu != this_cpu)
1208bb27 1526 color_fprintf(stdout, color, " ");
0ec04e16 1527 else
cf294f24 1528 color_fprintf(stdout, cpu_color, "*");
0ec04e16 1529
6bcab4e1 1530 if (sched->curr_thread[cpu])
a151a37a 1531 color_fprintf(stdout, pid_color, "%2s ", sched->curr_thread[cpu]->shortname);
6bcab4e1 1532 else
8cd91195 1533 color_fprintf(stdout, color, " ");
0ec04e16
IM		 1534 }
1535
73643bb6
JO		 1536 if (sched->map.cpus && !cpu_map__has(sched->map.cpus, this_cpu))
1537 goto out;
1538
99620a5d
NK		 1539 timestamp__scnprintf_usec(timestamp, stimestamp, sizeof(stimestamp));
1540 color_fprintf(stdout, color, " %12s secs ", stimestamp);
e107f129 1541 if (new_shortname || (verbose && sched_in->tid)) {
a151a37a
JO		 1542 const char *pid_color = color;
1543
1544 if (thread__has_color(sched_in))
1545 pid_color = COLOR_PIDS;
1546
1547 color_fprintf(stdout, pid_color, "%s => %s:%d",
b9c5143a 1548 sched_in->shortname, thread__comm_str(sched_in), sched_in->tid);
0ec04e16 1549 }
a116e05d 1550
99623c62 1551 if (sched->map.comp && new_cpu)
8cd91195 1552 color_fprintf(stdout, color, " (CPU %d)", this_cpu);
99623c62 1553
73643bb6 1554out:
8cd91195 1555 color_fprintf(stdout, color, "\n");
99623c62 1556
b91fc39f
ACM		 1557 thread__put(sched_in);
1558
a116e05d 1559 return 0;
0ec04e16
IM		 1560}
1561
0e9b07e5 1562static int process_sched_switch_event(struct perf_tool *tool,
2b7fcbc5 1563 struct perf_evsel *evsel,
1d037ca1 1564 struct perf_sample *sample,
4218e673 1565 struct machine *machine)
419ab0d6 1566{
0e9b07e5 1567 struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
a116e05d 1568 int this_cpu = sample->cpu, err = 0;
2b7fcbc5
ACM		 1569 u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"),
1570 next_pid = perf_evsel__intval(evsel, sample, "next_pid");
419ab0d6 1571
0e9b07e5 1572 if (sched->curr_pid[this_cpu] != (u32)-1) {
c8a37751
IM		 1573 /*
1574 * Are we trying to switch away a PID that is
1575 * not current?
1576 */
2b7fcbc5 1577 if (sched->curr_pid[this_cpu] != prev_pid)
0e9b07e5 1578 sched->nr_context_switch_bugs++;
c8a37751 1579 }
c8a37751 1580
9ec3f4e4
ACM		 1581 if (sched->tp_handler->switch_event)
1582 err = sched->tp_handler->switch_event(sched, evsel, sample, machine);
2b7fcbc5
ACM		 1583
1584 sched->curr_pid[this_cpu] = next_pid;
a116e05d 1585 return err;
419ab0d6
FW		 1586}
1587
0e9b07e5 1588static int process_sched_runtime_event(struct perf_tool *tool,
2b7fcbc5 1589 struct perf_evsel *evsel,
1d037ca1 1590 struct perf_sample *sample,
4218e673 1591 struct machine *machine)
39aeb52f 1592{
0e9b07e5 1593 struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
39aeb52f 1594
9ec3f4e4
ACM		 1595 if (sched->tp_handler->runtime_event)
1596 return sched->tp_handler->runtime_event(sched, evsel, sample, machine);
a116e05d 1597
2b7fcbc5 1598 return 0;
39aeb52f 1599}
1600
cb627505
DA		 1601static int perf_sched__process_fork_event(struct perf_tool *tool,
1602 union perf_event *event,
1603 struct perf_sample *sample,
1604 struct machine *machine)
fbf94829 1605{
0e9b07e5 1606 struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
46538818 1607
cb627505
DA		 1608 /* run the fork event through the perf machineruy */
1609 perf_event__process_fork(tool, event, sample, machine);
1610
1611 /* and then run additional processing needed for this command */
9ec3f4e4 1612 if (sched->tp_handler->fork_event)
cb627505 1613 return sched->tp_handler->fork_event(sched, event, machine);
a116e05d 1614
2b7fcbc5 1615 return 0;
fbf94829
IM		 1616}
1617
0e9b07e5 1618static int process_sched_migrate_task_event(struct perf_tool *tool,
2b7fcbc5 1619 struct perf_evsel *evsel,
1d037ca1 1620 struct perf_sample *sample,
4218e673 1621 struct machine *machine)
55ffb7a6 1622{
0e9b07e5 1623 struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
55ffb7a6 1624
9ec3f4e4
ACM		 1625 if (sched->tp_handler->migrate_task_event)
1626 return sched->tp_handler->migrate_task_event(sched, evsel, sample, machine);
a116e05d 1627
2b7fcbc5 1628 return 0;
55ffb7a6
MG		 1629}
1630
a116e05d 1631typedef int (*tracepoint_handler)(struct perf_tool *tool,
2b7fcbc5 1632 struct perf_evsel *evsel,
a116e05d 1633 struct perf_sample *sample,
4218e673 1634 struct machine *machine);
ec156764 1635
1d037ca1
IT		 1636static int perf_sched__process_tracepoint_sample(struct perf_tool *tool __maybe_unused,
1637 union perf_event *event __maybe_unused,
ee29be62
ACM		 1638 struct perf_sample *sample,
1639 struct perf_evsel *evsel,
1640 struct machine *machine)
0a02ad93 1641{
a116e05d 1642 int err = 0;
0a02ad93 1643
744a9719
ACM		 1644 if (evsel->handler != NULL) {
1645 tracepoint_handler f = evsel->handler;
2b7fcbc5 1646 err = f(tool, evsel, sample, machine);
ee29be62 1647 }
0a02ad93 1648
a116e05d 1649 return err;
0a02ad93
IM		 1650}
1651
ae536acf 1652static int perf_sched__read_events(struct perf_sched *sched)
0a02ad93 1653{
ee29be62
ACM		 1654 const struct perf_evsel_str_handler handlers[] = {
1655 { "sched:sched_switch", process_sched_switch_event, },
1656 { "sched:sched_stat_runtime", process_sched_runtime_event, },
1657 { "sched:sched_wakeup", process_sched_wakeup_event, },
1658 { "sched:sched_wakeup_new", process_sched_wakeup_event, },
ee29be62
ACM		 1659 { "sched:sched_migrate_task", process_sched_migrate_task_event, },
1660 };
da378962 1661 struct perf_session *session;
f5fc1412
JO		 1662 struct perf_data_file file = {
1663 .path = input_name,
1664 .mode = PERF_DATA_MODE_READ,
f0dd330f 1665 .force = sched->force,
f5fc1412 1666 };
ae536acf 1667 int rc = -1;
da378962 1668
f5fc1412 1669 session = perf_session__new(&file, false, &sched->tool);
a116e05d
ACM		 1670 if (session == NULL) {
1671 pr_debug("No Memory for session\n");
1672 return -1;
1673 }
94c744b6 1674
0a7e6d1b 1675 symbol__init(&session->header.env);
04934106 1676
a116e05d
ACM		 1677 if (perf_session__set_tracepoints_handlers(session, handlers))
1678 goto out_delete;
ee29be62 1679
cee75ac7 1680 if (perf_session__has_traces(session, "record -R")) {
b7b61cbe 1681 int err = perf_session__process_events(session);
a116e05d
ACM		 1682 if (err) {
1683 pr_err("Failed to process events, error %d", err);
1684 goto out_delete;
1685 }
4c09bafa 1686
75be989a
ACM		 1687 sched->nr_events = session->evlist->stats.nr_events[0];
1688 sched->nr_lost_events = session->evlist->stats.total_lost;
1689 sched->nr_lost_chunks = session->evlist->stats.nr_events[PERF_RECORD_LOST];
cee75ac7 1690 }
d549c769 1691
ae536acf 1692 rc = 0;
a116e05d
ACM		 1693out_delete:
1694 perf_session__delete(session);
ae536acf 1695 return rc;
0a02ad93
IM		 1696}
1697
49394a2a
DA		 1698/*
1699 * scheduling times are printed as msec.usec
1700 */
1701static inline void print_sched_time(unsigned long long nsecs, int width)
1702{
1703 unsigned long msecs;
1704 unsigned long usecs;
1705
1706 msecs = nsecs / NSEC_PER_MSEC;
1707 nsecs -= msecs * NSEC_PER_MSEC;
1708 usecs = nsecs / NSEC_PER_USEC;
1709 printf("%*lu.%03lu ", width, msecs, usecs);
1710}
1711
1712/*
1713 * returns runtime data for event, allocating memory for it the
1714 * first time it is used.
1715 */
1716static struct evsel_runtime *perf_evsel__get_runtime(struct perf_evsel *evsel)
1717{
1718 struct evsel_runtime *r = evsel->priv;
1719
1720 if (r == NULL) {
1721 r = zalloc(sizeof(struct evsel_runtime));
1722 evsel->priv = r;
1723 }
1724
1725 return r;
1726}
1727
1728/*
1729 * save last time event was seen per cpu
1730 */
1731static void perf_evsel__save_time(struct perf_evsel *evsel,
1732 u64 timestamp, u32 cpu)
1733{
1734 struct evsel_runtime *r = perf_evsel__get_runtime(evsel);
1735
1736 if (r == NULL)
1737 return;
1738
1739 if ((cpu >= r->ncpu) || (r->last_time == NULL)) {
1740 int i, n = __roundup_pow_of_two(cpu+1);
1741 void *p = r->last_time;
1742
1743 p = realloc(r->last_time, n * sizeof(u64));
1744 if (!p)
1745 return;
1746
1747 r->last_time = p;
1748 for (i = r->ncpu; i < n; ++i)
1749 r->last_time[i] = (u64) 0;
1750
1751 r->ncpu = n;
1752 }
1753
1754 r->last_time[cpu] = timestamp;
1755}
1756
1757/* returns last time this event was seen on the given cpu */
1758static u64 perf_evsel__get_time(struct perf_evsel *evsel, u32 cpu)
1759{
1760 struct evsel_runtime *r = perf_evsel__get_runtime(evsel);
1761
1762 if ((r == NULL) || (r->last_time == NULL) || (cpu >= r->ncpu))
1763 return 0;
1764
1765 return r->last_time[cpu];
1766}
1767
1768static int comm_width = 20;
1769
1770static char *timehist_get_commstr(struct thread *thread)
1771{
1772 static char str[32];
1773 const char *comm = thread__comm_str(thread);
1774 pid_t tid = thread->tid;
1775 pid_t pid = thread->pid_;
1776 int n;
1777
1778 if (pid == 0)
1779 n = scnprintf(str, sizeof(str), "%s", comm);
1780
1781 else if (tid != pid)
1782 n = scnprintf(str, sizeof(str), "%s[%d/%d]", comm, tid, pid);
1783
1784 else
1785 n = scnprintf(str, sizeof(str), "%s[%d]", comm, tid);
1786
1787 if (n > comm_width)
1788 comm_width = n;
1789
1790 return str;
1791}
1792
a407b067 1793static void timehist_header(struct perf_sched *sched)
49394a2a 1794{
a407b067
DA		 1795 u32 ncpus = sched->max_cpu + 1;
1796 u32 i, j;
1797
49394a2a
DA		 1798 printf("%15s %6s ", "time", "cpu");
1799
a407b067
DA		 1800 if (sched->show_cpu_visual) {
1801 printf(" ");
1802 for (i = 0, j = 0; i < ncpus; ++i) {
1803 printf("%x", j++);
1804 if (j > 15)
1805 j = 0;
1806 }
1807 printf(" ");
1808 }
1809
49394a2a
DA		 1810 printf(" %-20s %9s %9s %9s",
1811 "task name", "wait time", "sch delay", "run time");
1812
1813 printf("\n");
1814
1815 /*
1816 * units row
1817 */
1818 printf("%15s %-6s ", "", "");
1819
a407b067
DA		 1820 if (sched->show_cpu_visual)
1821 printf(" %*s ", ncpus, "");
1822
49394a2a
DA		 1823 printf(" %-20s %9s %9s %9s\n", "[tid/pid]", "(msec)", "(msec)", "(msec)");
1824
1825 /*
1826 * separator
1827 */
1828 printf("%.15s %.6s ", graph_dotted_line, graph_dotted_line);
1829
a407b067
DA		 1830 if (sched->show_cpu_visual)
1831 printf(" %.*s ", ncpus, graph_dotted_line);
1832
49394a2a
DA		 1833 printf(" %.20s %.9s %.9s %.9s",
1834 graph_dotted_line, graph_dotted_line, graph_dotted_line,
1835 graph_dotted_line);
1836
1837 printf("\n");
1838}
1839
fc1469f1
DA		 1840static void timehist_print_sample(struct perf_sched *sched,
1841 struct perf_sample *sample,
6c973c90 1842 struct addr_location *al,
853b7407
DA		 1843 struct thread *thread,
1844 u64 t)
49394a2a
DA		 1845{
1846 struct thread_runtime *tr = thread__priv(thread);
a407b067 1847 u32 max_cpus = sched->max_cpu + 1;
49394a2a
DA		 1848 char tstr[64];
1849
853b7407 1850 timestamp__scnprintf_usec(t, tstr, sizeof(tstr));
49394a2a
DA		 1851 printf("%15s [%04d] ", tstr, sample->cpu);
1852
a407b067
DA		 1853 if (sched->show_cpu_visual) {
1854 u32 i;
1855 char c;
1856
1857 printf(" ");
1858 for (i = 0; i < max_cpus; ++i) {
1859 /* flag idle times with 'i'; others are sched events */
1860 if (i == sample->cpu)
1861 c = (thread->tid == 0) ? 'i' : 's';
1862 else
1863 c = ' ';
1864 printf("%c", c);
1865 }
1866 printf(" ");
1867 }
1868
49394a2a
DA		 1869 printf(" %-*s ", comm_width, timehist_get_commstr(thread));
1870
1871 print_sched_time(tr->dt_wait, 6);
1872 print_sched_time(tr->dt_delay, 6);
1873 print_sched_time(tr->dt_run, 6);
fc1469f1
DA		 1874
1875 if (sched->show_wakeups)
1876 printf(" %-*s", comm_width, "");
1877
6c973c90
DA		 1878 if (thread->tid == 0)
1879 goto out;
1880
1881 if (sched->show_callchain)
1882 printf(" ");
1883
1884 sample__fprintf_sym(sample, al, 0,
1885 EVSEL__PRINT_SYM | EVSEL__PRINT_ONELINE |
2d9bbf6e
NK		 1886 EVSEL__PRINT_CALLCHAIN_ARROW |
1887 EVSEL__PRINT_SKIP_IGNORED,
6c973c90
DA		 1888 &callchain_cursor, stdout);
1889
1890out:
49394a2a
DA		 1891 printf("\n");
1892}
1893
1894/*
1895 * Explanation of delta-time stats:
1896 *
1897 * t = time of current schedule out event
1898 * tprev = time of previous sched out event
1899 * also time of schedule-in event for current task
1900 * last_time = time of last sched change event for current task
1901 * (i.e, time process was last scheduled out)
1902 * ready_to_run = time of wakeup for current task
1903 *
1904 * -----|------------|------------|------------|------
1905 * last ready tprev t
1906 * time to run
1907 *
1908 * |-------- dt_wait --------|
1909 * |- dt_delay -|-- dt_run --|
1910 *
1911 * dt_run = run time of current task
1912 * dt_wait = time between last schedule out event for task and tprev
1913 * represents time spent off the cpu
1914 * dt_delay = time between wakeup and schedule-in of task
1915 */
1916
1917static void timehist_update_runtime_stats(struct thread_runtime *r,
1918 u64 t, u64 tprev)
1919{
1920 r->dt_delay = 0;
1921 r->dt_wait = 0;
1922 r->dt_run = 0;
1923 if (tprev) {
1924 r->dt_run = t - tprev;
1925 if (r->ready_to_run) {
1926 if (r->ready_to_run > tprev)
1927 pr_debug("time travel: wakeup time for task > previous sched_switch event\n");
1928 else
1929 r->dt_delay = tprev - r->ready_to_run;
1930 }
1931
1932 if (r->last_time > tprev)
1933 pr_debug("time travel: last sched out time for task > previous sched_switch event\n");
1934 else if (r->last_time)
1935 r->dt_wait = tprev - r->last_time;
1936 }
1937
1938 update_stats(&r->run_stats, r->dt_run);
1939 r->total_run_time += r->dt_run;
1940}
1941
96039c7c
NK		 1942static bool is_idle_sample(struct perf_sample *sample,
1943 struct perf_evsel *evsel)
49394a2a
DA		 1944{
1945 /* pid 0 == swapper == idle task */
96039c7c
NK		 1946 if (strcmp(perf_evsel__name(evsel), "sched:sched_switch") == 0)
1947 return perf_evsel__intval(evsel, sample, "prev_pid") == 0;
49394a2a 1948
96039c7c
NK		 1949 return sample->pid == 0;
1950}
1951
1952static void save_task_callchain(struct perf_sched *sched,
1953 struct perf_sample *sample,
1954 struct perf_evsel *evsel,
1955 struct machine *machine)
1956{
1957 struct callchain_cursor *cursor = &callchain_cursor;
1958 struct thread *thread;
6c973c90
DA		 1959
1960 /* want main thread for process - has maps */
1961 thread = machine__findnew_thread(machine, sample->pid, sample->pid);
1962 if (thread == NULL) {
1963 pr_debug("Failed to get thread for pid %d.\n", sample->pid);
96039c7c 1964 return;
6c973c90
DA		 1965 }
1966
1967 if (!symbol_conf.use_callchain || sample->callchain == NULL)
96039c7c 1968 return;
6c973c90
DA		 1969
1970 if (thread__resolve_callchain(thread, cursor, evsel, sample,
8388deb3 1971 NULL, NULL, sched->max_stack + 2) != 0) {
6c973c90
DA		 1972 if (verbose)
1973 error("Failed to resolve callchain. Skipping\n");
1974
96039c7c 1975 return;
6c973c90 1976 }
cdeb01bf 1977
6c973c90 1978 callchain_cursor_commit(cursor);
cdeb01bf
NK		 1979
1980 while (true) {
1981 struct callchain_cursor_node *node;
1982 struct symbol *sym;
1983
1984 node = callchain_cursor_current(cursor);
1985 if (node == NULL)
1986 break;
1987
1988 sym = node->sym;
1989 if (sym && sym->name) {
1990 if (!strcmp(sym->name, "schedule") ||
1991 !strcmp(sym->name, "__schedule") ||
1992 !strcmp(sym->name, "preempt_schedule"))
1993 sym->ignore = 1;
1994 }
1995
1996 callchain_cursor_advance(cursor);
1997 }
49394a2a
DA		 1998}
1999
2000/*
2001 * Track idle stats per cpu by maintaining a local thread
2002 * struct for the idle task on each cpu.
2003 */
2004static int init_idle_threads(int ncpu)
2005{
2006 int i;
2007
2008 idle_threads = zalloc(ncpu * sizeof(struct thread *));
2009 if (!idle_threads)
2010 return -ENOMEM;
2011
b336352b 2012 idle_max_cpu = ncpu;
49394a2a
DA		 2013
2014 /* allocate the actual thread struct if needed */
2015 for (i = 0; i < ncpu; ++i) {
2016 idle_threads[i] = thread__new(0, 0);
2017 if (idle_threads[i] == NULL)
2018 return -ENOMEM;
2019
2020 thread__set_comm(idle_threads[i], idle_comm, 0);
2021 }
2022
2023 return 0;
2024}
2025
2026static void free_idle_threads(void)
2027{
2028 int i;
2029
2030 if (idle_threads == NULL)
2031 return;
2032
b336352b 2033 for (i = 0; i < idle_max_cpu; ++i) {
49394a2a
DA		 2034 if ((idle_threads[i]))
2035 thread__delete(idle_threads[i]);
2036 }
2037
2038 free(idle_threads);
2039}
2040
2041static struct thread *get_idle_thread(int cpu)
2042{
2043 /*
2044 * expand/allocate array of pointers to local thread
2045 * structs if needed
2046 */
2047 if ((cpu >= idle_max_cpu) || (idle_threads == NULL)) {
2048 int i, j = __roundup_pow_of_two(cpu+1);
2049 void *p;
2050
2051 p = realloc(idle_threads, j * sizeof(struct thread *));
2052 if (!p)
2053 return NULL;
2054
2055 idle_threads = (struct thread **) p;
b336352b 2056 for (i = idle_max_cpu; i < j; ++i)
49394a2a
DA		 2057 idle_threads[i] = NULL;
2058
2059 idle_max_cpu = j;
2060 }
2061
2062 /* allocate a new thread struct if needed */
2063 if (idle_threads[cpu] == NULL) {
2064 idle_threads[cpu] = thread__new(0, 0);
2065
2066 if (idle_threads[cpu]) {
2067 idle_threads[cpu]->tid = 0;
2068 thread__set_comm(idle_threads[cpu], idle_comm, 0);
2069 }
2070 }
2071
2072 return idle_threads[cpu];
2073}
2074
2075/*
2076 * handle runtime stats saved per thread
2077 */
2078static struct thread_runtime *thread__init_runtime(struct thread *thread)
2079{
2080 struct thread_runtime *r;
2081
2082 r = zalloc(sizeof(struct thread_runtime));
2083 if (!r)
2084 return NULL;
2085
2086 init_stats(&r->run_stats);
2087 thread__set_priv(thread, r);
2088
2089 return r;
2090}
2091
2092static struct thread_runtime *thread__get_runtime(struct thread *thread)
2093{
2094 struct thread_runtime *tr;
2095
2096 tr = thread__priv(thread);
2097 if (tr == NULL) {
2098 tr = thread__init_runtime(thread);
2099 if (tr == NULL)
2100 pr_debug("Failed to malloc memory for runtime data.\n");
2101 }
2102
2103 return tr;
2104}
2105
6c973c90
DA		 2106static struct thread *timehist_get_thread(struct perf_sched *sched,
2107 struct perf_sample *sample,
49394a2a
DA		 2108 struct machine *machine,
2109 struct perf_evsel *evsel)
2110{
2111 struct thread *thread;
2112
96039c7c 2113 if (is_idle_sample(sample, evsel)) {
49394a2a
DA		 2114 thread = get_idle_thread(sample->cpu);
2115 if (thread == NULL)
2116 pr_err("Failed to get idle thread for cpu %d.\n", sample->cpu);
2117
2118 } else {
5d92d96a
NK		 2119 /* there were samples with tid 0 but non-zero pid */
2120 thread = machine__findnew_thread(machine, sample->pid,
2121 sample->tid ?: sample->pid);
49394a2a
DA		 2122 if (thread == NULL) {
2123 pr_debug("Failed to get thread for tid %d. skipping sample.\n",
2124 sample->tid);
2125 }
96039c7c
NK		 2126
2127 save_task_callchain(sched, sample, evsel, machine);
49394a2a
DA		 2128 }
2129
2130 return thread;
2131}
2132
52df138c
DA		 2133static bool timehist_skip_sample(struct perf_sched *sched,
2134 struct thread *thread)
49394a2a
DA		 2135{
2136 bool rc = false;
2137
52df138c 2138 if (thread__is_filtered(thread)) {
49394a2a 2139 rc = true;
52df138c
DA		 2140 sched->skipped_samples++;
2141 }
49394a2a
DA		 2142
2143 return rc;
2144}
2145
fc1469f1
DA		 2146static void timehist_print_wakeup_event(struct perf_sched *sched,
2147 struct perf_sample *sample,
2148 struct machine *machine,
2149 struct thread *awakened)
2150{
2151 struct thread *thread;
2152 char tstr[64];
2153
2154 thread = machine__findnew_thread(machine, sample->pid, sample->tid);
2155 if (thread == NULL)
2156 return;
2157
2158 /* show wakeup unless both awakee and awaker are filtered */
2159 if (timehist_skip_sample(sched, thread) &&
2160 timehist_skip_sample(sched, awakened)) {
2161 return;
2162 }
2163
2164 timestamp__scnprintf_usec(sample->time, tstr, sizeof(tstr));
2165 printf("%15s [%04d] ", tstr, sample->cpu);
a407b067
DA		 2166 if (sched->show_cpu_visual)
2167 printf(" %*s ", sched->max_cpu + 1, "");
fc1469f1
DA		 2168
2169 printf(" %-*s ", comm_width, timehist_get_commstr(thread));
2170
2171 /* dt spacer */
2172 printf(" %9s %9s %9s ", "", "", "");
2173
2174 printf("awakened: %s", timehist_get_commstr(awakened));
2175
2176 printf("\n");
2177}
2178
2179static int timehist_sched_wakeup_event(struct perf_tool *tool,
49394a2a
DA		 2180 union perf_event *event __maybe_unused,
2181 struct perf_evsel *evsel,
2182 struct perf_sample *sample,
2183 struct machine *machine)
2184{
fc1469f1 2185 struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
49394a2a
DA		 2186 struct thread *thread;
2187 struct thread_runtime *tr = NULL;
2188 /* want pid of awakened task not pid in sample */
2189 const u32 pid = perf_evsel__intval(evsel, sample, "pid");
2190
2191 thread = machine__findnew_thread(machine, 0, pid);
2192 if (thread == NULL)
2193 return -1;
2194
2195 tr = thread__get_runtime(thread);
2196 if (tr == NULL)
2197 return -1;
2198
2199 if (tr->ready_to_run == 0)
2200 tr->ready_to_run = sample->time;
2201
fc1469f1 2202 /* show wakeups if requested */
853b7407
DA		 2203 if (sched->show_wakeups &&
2204 !perf_time__skip_sample(&sched->ptime, sample->time))
fc1469f1
DA		 2205 timehist_print_wakeup_event(sched, sample, machine, thread);
2206
49394a2a
DA		 2207 return 0;
2208}
2209
350f54fa
DA		 2210static void timehist_print_migration_event(struct perf_sched *sched,
2211 struct perf_evsel *evsel,
2212 struct perf_sample *sample,
2213 struct machine *machine,
2214 struct thread *migrated)
2215{
2216 struct thread *thread;
2217 char tstr[64];
2218 u32 max_cpus = sched->max_cpu + 1;
2219 u32 ocpu, dcpu;
2220
2221 if (sched->summary_only)
2222 return;
2223
2224 max_cpus = sched->max_cpu + 1;
2225 ocpu = perf_evsel__intval(evsel, sample, "orig_cpu");
2226 dcpu = perf_evsel__intval(evsel, sample, "dest_cpu");
2227
2228 thread = machine__findnew_thread(machine, sample->pid, sample->tid);
2229 if (thread == NULL)
2230 return;
2231
2232 if (timehist_skip_sample(sched, thread) &&
2233 timehist_skip_sample(sched, migrated)) {
2234 return;
2235 }
2236
2237 timestamp__scnprintf_usec(sample->time, tstr, sizeof(tstr));
2238 printf("%15s [%04d] ", tstr, sample->cpu);
2239
2240 if (sched->show_cpu_visual) {
2241 u32 i;
2242 char c;
2243
2244 printf(" ");
2245 for (i = 0; i < max_cpus; ++i) {
2246 c = (i == sample->cpu) ? 'm' : ' ';
2247 printf("%c", c);
2248 }
2249 printf(" ");
2250 }
2251
2252 printf(" %-*s ", comm_width, timehist_get_commstr(thread));
2253
2254 /* dt spacer */
2255 printf(" %9s %9s %9s ", "", "", "");
2256
2257 printf("migrated: %s", timehist_get_commstr(migrated));
2258 printf(" cpu %d => %d", ocpu, dcpu);
2259
2260 printf("\n");
2261}
2262
2263static int timehist_migrate_task_event(struct perf_tool *tool,
2264 union perf_event *event __maybe_unused,
2265 struct perf_evsel *evsel,
2266 struct perf_sample *sample,
2267 struct machine *machine)
2268{
2269 struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
2270 struct thread *thread;
2271 struct thread_runtime *tr = NULL;
2272 /* want pid of migrated task not pid in sample */
2273 const u32 pid = perf_evsel__intval(evsel, sample, "pid");
2274
2275 thread = machine__findnew_thread(machine, 0, pid);
2276 if (thread == NULL)
2277 return -1;
2278
2279 tr = thread__get_runtime(thread);
2280 if (tr == NULL)
2281 return -1;
2282
2283 tr->migrations++;
2284
2285 /* show migrations if requested */
2286 timehist_print_migration_event(sched, evsel, sample, machine, thread);
2287
2288 return 0;
2289}
2290
52df138c 2291static int timehist_sched_change_event(struct perf_tool *tool,
49394a2a
DA		 2292 union perf_event *event,
2293 struct perf_evsel *evsel,
2294 struct perf_sample *sample,
2295 struct machine *machine)
2296{
fc1469f1 2297 struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
853b7407 2298 struct perf_time_interval *ptime = &sched->ptime;
49394a2a
DA		 2299 struct addr_location al;
2300 struct thread *thread;
2301 struct thread_runtime *tr = NULL;
853b7407 2302 u64 tprev, t = sample->time;
49394a2a
DA		 2303 int rc = 0;
2304
2305 if (machine__resolve(machine, &al, sample) < 0) {
2306 pr_err("problem processing %d event. skipping it\n",
2307 event->header.type);
2308 rc = -1;
2309 goto out;
2310 }
2311
6c973c90 2312 thread = timehist_get_thread(sched, sample, machine, evsel);
49394a2a
DA		 2313 if (thread == NULL) {
2314 rc = -1;
2315 goto out;
2316 }
2317
52df138c 2318 if (timehist_skip_sample(sched, thread))
49394a2a
DA		 2319 goto out;
2320
2321 tr = thread__get_runtime(thread);
2322 if (tr == NULL) {
2323 rc = -1;
2324 goto out;
2325 }
2326
2327 tprev = perf_evsel__get_time(evsel, sample->cpu);
2328
853b7407
DA		 2329 /*
2330 * If start time given:
2331 * - sample time is under window user cares about - skip sample
2332 * - tprev is under window user cares about - reset to start of window
2333 */
2334 if (ptime->start && ptime->start > t)
2335 goto out;
2336
2337 if (ptime->start > tprev)
2338 tprev = ptime->start;
2339
2340 /*
2341 * If end time given:
2342 * - previous sched event is out of window - we are done
2343 * - sample time is beyond window user cares about - reset it
2344 * to close out stats for time window interest
2345 */
2346 if (ptime->end) {
2347 if (tprev > ptime->end)
2348 goto out;
2349
2350 if (t > ptime->end)
2351 t = ptime->end;
2352 }
2353
2354 timehist_update_runtime_stats(tr, t, tprev);
2355
52df138c 2356 if (!sched->summary_only)
853b7407 2357 timehist_print_sample(sched, sample, &al, thread, t);
49394a2a
DA		 2358
2359out:
2360 if (tr) {
2361 /* time of this sched_switch event becomes last time task seen */
2362 tr->last_time = sample->time;
2363
2364 /* sched out event for task so reset ready to run time */
2365 tr->ready_to_run = 0;
2366 }
2367
2368 perf_evsel__save_time(evsel, sample->time, sample->cpu);
2369
2370 return rc;
2371}
2372
2373static int timehist_sched_switch_event(struct perf_tool *tool,
2374 union perf_event *event,
2375 struct perf_evsel *evsel,
2376 struct perf_sample *sample,
2377 struct machine *machine __maybe_unused)
2378{
2379 return timehist_sched_change_event(tool, event, evsel, sample, machine);
2380}
2381
2382static int process_lost(struct perf_tool *tool __maybe_unused,
2383 union perf_event *event,
2384 struct perf_sample *sample,
2385 struct machine *machine __maybe_unused)
2386{
2387 char tstr[64];
2388
2389 timestamp__scnprintf_usec(sample->time, tstr, sizeof(tstr));
2390 printf("%15s ", tstr);
2391 printf("lost %" PRIu64 " events on cpu %d\n", event->lost.lost, sample->cpu);
2392
2393 return 0;
2394}
2395
2396
52df138c
DA		 2397static void print_thread_runtime(struct thread *t,
2398 struct thread_runtime *r)
2399{
2400 double mean = avg_stats(&r->run_stats);
2401 float stddev;
2402
2403 printf("%*s %5d %9" PRIu64 " ",
2404 comm_width, timehist_get_commstr(t), t->ppid,
2405 (u64) r->run_stats.n);
2406
2407 print_sched_time(r->total_run_time, 8);
2408 stddev = rel_stddev_stats(stddev_stats(&r->run_stats), mean);
2409 print_sched_time(r->run_stats.min, 6);
2410 printf(" ");
2411 print_sched_time((u64) mean, 6);
2412 printf(" ");
2413 print_sched_time(r->run_stats.max, 6);
2414 printf(" ");
2415 printf("%5.2f", stddev);
350f54fa 2416 printf(" %5" PRIu64, r->migrations);
52df138c
DA		 2417 printf("\n");
2418}
2419
2420struct total_run_stats {
2421 u64 sched_count;
2422 u64 task_count;
2423 u64 total_run_time;
2424};
2425
2426static int __show_thread_runtime(struct thread *t, void *priv)
2427{
2428 struct total_run_stats *stats = priv;
2429 struct thread_runtime *r;
2430
2431 if (thread__is_filtered(t))
2432 return 0;
2433
2434 r = thread__priv(t);
2435 if (r && r->run_stats.n) {
2436 stats->task_count++;
2437 stats->sched_count += r->run_stats.n;
2438 stats->total_run_time += r->total_run_time;
2439 print_thread_runtime(t, r);
2440 }
2441
2442 return 0;
2443}
2444
2445static int show_thread_runtime(struct thread *t, void *priv)
2446{
2447 if (t->dead)
2448 return 0;
2449
2450 return __show_thread_runtime(t, priv);
2451}
2452
2453static int show_deadthread_runtime(struct thread *t, void *priv)
2454{
2455 if (!t->dead)
2456 return 0;
2457
2458 return __show_thread_runtime(t, priv);
2459}
2460
2461static void timehist_print_summary(struct perf_sched *sched,
2462 struct perf_session *session)
2463{
2464 struct machine *m = &session->machines.host;
2465 struct total_run_stats totals;
2466 u64 task_count;
2467 struct thread *t;
2468 struct thread_runtime *r;
2469 int i;
2470
2471 memset(&totals, 0, sizeof(totals));
2472
2473 if (comm_width < 30)
2474 comm_width = 30;
2475
2476 printf("\nRuntime summary\n");
2477 printf("%*s parent sched-in ", comm_width, "comm");
350f54fa 2478 printf(" run-time min-run avg-run max-run stddev migrations\n");
52df138c
DA		 2479 printf("%*s (count) ", comm_width, "");
2480 printf(" (msec) (msec) (msec) (msec) %%\n");
350f54fa 2481 printf("%.117s\n", graph_dotted_line);
52df138c
DA		 2482
2483 machine__for_each_thread(m, show_thread_runtime, &totals);
2484 task_count = totals.task_count;
2485 if (!task_count)
2486 printf("<no still running tasks>\n");
2487
2488 printf("\nTerminated tasks:\n");
2489 machine__for_each_thread(m, show_deadthread_runtime, &totals);
2490 if (task_count == totals.task_count)
2491 printf("<no terminated tasks>\n");
2492
2493 /* CPU idle stats not tracked when samples were skipped */
2494 if (sched->skipped_samples)
2495 return;
2496
2497 printf("\nIdle stats:\n");
b336352b 2498 for (i = 0; i < idle_max_cpu; ++i) {
52df138c
DA		 2499 t = idle_threads[i];
2500 if (!t)
2501 continue;
2502
2503 r = thread__priv(t);
2504 if (r && r->run_stats.n) {
2505 totals.sched_count += r->run_stats.n;
2506 printf(" CPU %2d idle for ", i);
2507 print_sched_time(r->total_run_time, 6);
2508 printf(" msec\n");
2509 } else
2510 printf(" CPU %2d idle entire time window\n", i);
2511 }
2512
2513 printf("\n"
2514 " Total number of unique tasks: %" PRIu64 "\n"
2515 "Total number of context switches: %" PRIu64 "\n"
2516 " Total run time (msec): ",
2517 totals.task_count, totals.sched_count);
2518
2519 print_sched_time(totals.total_run_time, 2);
2520 printf("\n");
2521}
2522
49394a2a
DA		 2523typedef int (*sched_handler)(struct perf_tool *tool,
2524 union perf_event *event,
2525 struct perf_evsel *evsel,
2526 struct perf_sample *sample,
2527 struct machine *machine);
2528
2529static int perf_timehist__process_sample(struct perf_tool *tool,
2530 union perf_event *event,
2531 struct perf_sample *sample,
2532 struct perf_evsel *evsel,
2533 struct machine *machine)
2534{
2535 struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
2536 int err = 0;
2537 int this_cpu = sample->cpu;
2538
2539 if (this_cpu > sched->max_cpu)
2540 sched->max_cpu = this_cpu;
2541
2542 if (evsel->handler != NULL) {
2543 sched_handler f = evsel->handler;
2544
2545 err = f(tool, event, evsel, sample, machine);
2546 }
2547
2548 return err;
2549}
2550
6c973c90
DA		 2551static int timehist_check_attr(struct perf_sched *sched,
2552 struct perf_evlist *evlist)
2553{
2554 struct perf_evsel *evsel;
2555 struct evsel_runtime *er;
2556
2557 list_for_each_entry(evsel, &evlist->entries, node) {
2558 er = perf_evsel__get_runtime(evsel);
2559 if (er == NULL) {
2560 pr_err("Failed to allocate memory for evsel runtime data\n");
2561 return -1;
2562 }
2563
2564 if (sched->show_callchain &&
2565 !(evsel->attr.sample_type & PERF_SAMPLE_CALLCHAIN)) {
2566 pr_info("Samples do not have callchains.\n");
2567 sched->show_callchain = 0;
2568 symbol_conf.use_callchain = 0;
2569 }
2570 }
2571
2572 return 0;
2573}
2574
49394a2a
DA		 2575static int perf_sched__timehist(struct perf_sched *sched)
2576{
2577 const struct perf_evsel_str_handler handlers[] = {
2578 { "sched:sched_switch", timehist_sched_switch_event, },
2579 { "sched:sched_wakeup", timehist_sched_wakeup_event, },
2580 { "sched:sched_wakeup_new", timehist_sched_wakeup_event, },
2581 };
350f54fa
DA		 2582 const struct perf_evsel_str_handler migrate_handlers[] = {
2583 { "sched:sched_migrate_task", timehist_migrate_task_event, },
2584 };
49394a2a
DA		 2585 struct perf_data_file file = {
2586 .path = input_name,
2587 .mode = PERF_DATA_MODE_READ,
6fa94258 2588 .force = sched->force,
49394a2a
DA		 2589 };
2590
2591 struct perf_session *session;
52df138c 2592 struct perf_evlist *evlist;
49394a2a
DA		 2593 int err = -1;
2594
2595 /*
2596 * event handlers for timehist option
2597 */
2598 sched->tool.sample = perf_timehist__process_sample;
2599 sched->tool.mmap = perf_event__process_mmap;
2600 sched->tool.comm = perf_event__process_comm;
2601 sched->tool.exit = perf_event__process_exit;
2602 sched->tool.fork = perf_event__process_fork;
2603 sched->tool.lost = process_lost;
2604 sched->tool.attr = perf_event__process_attr;
2605 sched->tool.tracing_data = perf_event__process_tracing_data;
2606 sched->tool.build_id = perf_event__process_build_id;
2607
2608 sched->tool.ordered_events = true;
2609 sched->tool.ordering_requires_timestamps = true;
2610
6c973c90
DA		 2611 symbol_conf.use_callchain = sched->show_callchain;
2612
49394a2a
DA		 2613 session = perf_session__new(&file, false, &sched->tool);
2614 if (session == NULL)
2615 return -ENOMEM;
2616
52df138c
DA		 2617 evlist = session->evlist;
2618
49394a2a
DA		 2619 symbol__init(&session->header.env);
2620
853b7407
DA		 2621 if (perf_time__parse_str(&sched->ptime, sched->time_str) != 0) {
2622 pr_err("Invalid time string\n");
2623 return -EINVAL;
2624 }
2625
6c973c90
DA		 2626 if (timehist_check_attr(sched, evlist) != 0)
2627 goto out;
2628
49394a2a
DA		 2629 setup_pager();
2630
2631 /* setup per-evsel handlers */
2632 if (perf_session__set_tracepoints_handlers(session, handlers))
2633 goto out;
2634
f45bf8d3
DA		 2635 /* sched_switch event at a minimum needs to exist */
2636 if (!perf_evlist__find_tracepoint_by_name(session->evlist,
2637 "sched:sched_switch")) {
2638 pr_err("No sched_switch events found. Have you run 'perf sched record'?\n");
49394a2a 2639 goto out;
f45bf8d3 2640 }
49394a2a 2641
350f54fa
DA		 2642 if (sched->show_migrations &&
2643 perf_session__set_tracepoints_handlers(session, migrate_handlers))
2644 goto out;
2645
49394a2a
DA		 2646 /* pre-allocate struct for per-CPU idle stats */
2647 sched->max_cpu = session->header.env.nr_cpus_online;
2648 if (sched->max_cpu == 0)
2649 sched->max_cpu = 4;
2650 if (init_idle_threads(sched->max_cpu))
2651 goto out;
2652
52df138c
DA		 2653 /* summary_only implies summary option, but don't overwrite summary if set */
2654 if (sched->summary_only)
2655 sched->summary = sched->summary_only;
2656
2657 if (!sched->summary_only)
a407b067 2658 timehist_header(sched);
49394a2a
DA		 2659
2660 err = perf_session__process_events(session);
2661 if (err) {
2662 pr_err("Failed to process events, error %d", err);
2663 goto out;
2664 }
2665
52df138c
DA		 2666 sched->nr_events = evlist->stats.nr_events[0];
2667 sched->nr_lost_events = evlist->stats.total_lost;
2668 sched->nr_lost_chunks = evlist->stats.nr_events[PERF_RECORD_LOST];
2669
2670 if (sched->summary)
2671 timehist_print_summary(sched, session);
2672
49394a2a
DA		 2673out:
2674 free_idle_threads();
2675 perf_session__delete(session);
2676
2677 return err;
2678}
2679
2680
0e9b07e5 2681static void print_bad_events(struct perf_sched *sched)
0ec04e16 2682{
0e9b07e5 2683 if (sched->nr_unordered_timestamps && sched->nr_timestamps) {
0ec04e16 2684 printf(" INFO: %.3f%% unordered timestamps (%ld out of %ld)\n",
0e9b07e5
ACM		 2685 (double)sched->nr_unordered_timestamps/(double)sched->nr_timestamps*100.0,
2686 sched->nr_unordered_timestamps, sched->nr_timestamps);
0ec04e16 2687 }
0e9b07e5 2688 if (sched->nr_lost_events && sched->nr_events) {
0ec04e16 2689 printf(" INFO: %.3f%% lost events (%ld out of %ld, in %ld chunks)\n",
0e9b07e5
ACM		 2690 (double)sched->nr_lost_events/(double)sched->nr_events * 100.0,
2691 sched->nr_lost_events, sched->nr_events, sched->nr_lost_chunks);
0ec04e16 2692 }
0e9b07e5 2693 if (sched->nr_context_switch_bugs && sched->nr_timestamps) {
0ec04e16 2694 printf(" INFO: %.3f%% context switch bugs (%ld out of %ld)",
0e9b07e5
ACM		 2695 (double)sched->nr_context_switch_bugs/(double)sched->nr_timestamps*100.0,
2696 sched->nr_context_switch_bugs, sched->nr_timestamps);
2697 if (sched->nr_lost_events)
0ec04e16
IM		 2698 printf(" (due to lost events?)");
2699 printf("\n");
2700 }
2701}
2702
2f80dd44
JB		 2703static void __merge_work_atoms(struct rb_root *root, struct work_atoms *data)
2704{
2705 struct rb_node **new = &(root->rb_node), *parent = NULL;
2706 struct work_atoms *this;
2707 const char *comm = thread__comm_str(data->thread), *this_comm;
2708
2709 while (*new) {
2710 int cmp;
2711
2712 this = container_of(*new, struct work_atoms, node);
2713 parent = *new;
2714
2715 this_comm = thread__comm_str(this->thread);
2716 cmp = strcmp(comm, this_comm);
2717 if (cmp > 0) {
2718 new = &((*new)->rb_left);
2719 } else if (cmp < 0) {
2720 new = &((*new)->rb_right);
2721 } else {
2722 this->num_merged++;
2723 this->total_runtime += data->total_runtime;
2724 this->nb_atoms += data->nb_atoms;
2725 this->total_lat += data->total_lat;
2726 list_splice(&data->work_list, &this->work_list);
2727 if (this->max_lat < data->max_lat) {
2728 this->max_lat = data->max_lat;
2729 this->max_lat_at = data->max_lat_at;
2730 }
2731 zfree(&data);
2732 return;
2733 }
2734 }
2735
2736 data->num_merged++;
2737 rb_link_node(&data->node, parent, new);
2738 rb_insert_color(&data->node, root);
2739}
2740
2741static void perf_sched__merge_lat(struct perf_sched *sched)
2742{
2743 struct work_atoms *data;
2744 struct rb_node *node;
2745
2746 if (sched->skip_merge)
2747 return;
2748
2749 while ((node = rb_first(&sched->atom_root))) {
2750 rb_erase(node, &sched->atom_root);
2751 data = rb_entry(node, struct work_atoms, node);
2752 __merge_work_atoms(&sched->merged_atom_root, data);
2753 }
2754}
2755
0e9b07e5 2756static int perf_sched__lat(struct perf_sched *sched)
0ec04e16
IM		 2757{
2758 struct rb_node *next;
2759
2760 setup_pager();
ad9def7c 2761
ae536acf 2762 if (perf_sched__read_events(sched))
a116e05d 2763 return -1;
ad9def7c 2764
2f80dd44 2765 perf_sched__merge_lat(sched);
0e9b07e5 2766 perf_sched__sort_lat(sched);
0ec04e16 2767
80790e0b
RR		 2768 printf("\n -----------------------------------------------------------------------------------------------------------------\n");
2769 printf(" Task | Runtime ms | Switches | Average delay ms | Maximum delay ms | Maximum delay at |\n");
2770 printf(" -----------------------------------------------------------------------------------------------------------------\n");
0ec04e16 2771
0e9b07e5 2772 next = rb_first(&sched->sorted_atom_root);
0ec04e16
IM		 2773
2774 while (next) {
2775 struct work_atoms *work_list;
2776
2777 work_list = rb_entry(next, struct work_atoms, node);
0e9b07e5 2778 output_lat_thread(sched, work_list);
0ec04e16 2779 next = rb_next(next);
ae536acf 2780 thread__zput(work_list->thread);
0ec04e16
IM		 2781 }
2782
80790e0b 2783 printf(" -----------------------------------------------------------------------------------------------------------------\n");
9486aa38 2784 printf(" TOTAL: |%11.3f ms |%9" PRIu64 " |\n",
4fc76e49 2785 (double)sched->all_runtime / NSEC_PER_MSEC, sched->all_count);
0ec04e16
IM		 2786
2787 printf(" ---------------------------------------------------\n");
2788
0e9b07e5 2789 print_bad_events(sched);
0ec04e16
IM		 2790 printf("\n");
2791
a116e05d 2792 return 0;
0ec04e16
IM		 2793}
2794
99623c62
JO		 2795static int setup_map_cpus(struct perf_sched *sched)
2796{
73643bb6
JO		 2797 struct cpu_map *map;
2798
99623c62
JO		 2799 sched->max_cpu = sysconf(_SC_NPROCESSORS_CONF);
2800
2801 if (sched->map.comp) {
2802 sched->map.comp_cpus = zalloc(sched->max_cpu * sizeof(int));
cf294f24
JO		 2803 if (!sched->map.comp_cpus)
2804 return -1;
99623c62
JO		 2805 }
2806
73643bb6
JO		 2807 if (!sched->map.cpus_str)
2808 return 0;
2809
2810 map = cpu_map__new(sched->map.cpus_str);
2811 if (!map) {
2812 pr_err("failed to get cpus map from %s\n", sched->map.cpus_str);
2813 return -1;
2814 }
2815
2816 sched->map.cpus = map;
99623c62
JO		 2817 return 0;
2818}
2819
a151a37a
JO		 2820static int setup_color_pids(struct perf_sched *sched)
2821{
2822 struct thread_map *map;
2823
2824 if (!sched->map.color_pids_str)
2825 return 0;
2826
2827 map = thread_map__new_by_tid_str(sched->map.color_pids_str);
2828 if (!map) {
2829 pr_err("failed to get thread map from %s\n", sched->map.color_pids_str);
2830 return -1;
2831 }
2832
2833 sched->map.color_pids = map;
2834 return 0;
2835}
2836
cf294f24
JO		 2837static int setup_color_cpus(struct perf_sched *sched)
2838{
2839 struct cpu_map *map;
2840
2841 if (!sched->map.color_cpus_str)
2842 return 0;
2843
2844 map = cpu_map__new(sched->map.color_cpus_str);
2845 if (!map) {
2846 pr_err("failed to get thread map from %s\n", sched->map.color_cpus_str);
2847 return -1;
2848 }
2849
2850 sched->map.color_cpus = map;
2851 return 0;
2852}
2853
0e9b07e5 2854static int perf_sched__map(struct perf_sched *sched)
0ec04e16 2855{
99623c62
JO		 2856 if (setup_map_cpus(sched))
2857 return -1;
40749d0f 2858
a151a37a
JO		 2859 if (setup_color_pids(sched))
2860 return -1;
2861
cf294f24
JO		 2862 if (setup_color_cpus(sched))
2863 return -1;
2864
0ec04e16 2865 setup_pager();
ae536acf 2866 if (perf_sched__read_events(sched))
a116e05d 2867 return -1;
0e9b07e5 2868 print_bad_events(sched);
a116e05d 2869 return 0;
0ec04e16
IM		 2870}
2871
0e9b07e5 2872static int perf_sched__replay(struct perf_sched *sched)
0ec04e16
IM		 2873{
2874 unsigned long i;
2875
0e9b07e5
ACM		 2876 calibrate_run_measurement_overhead(sched);
2877 calibrate_sleep_measurement_overhead(sched);
0ec04e16 2878
0e9b07e5 2879 test_calibrations(sched);
0ec04e16 2880
ae536acf 2881 if (perf_sched__read_events(sched))
a116e05d 2882 return -1;
0ec04e16 2883
0e9b07e5
ACM		 2884 printf("nr_run_events: %ld\n", sched->nr_run_events);
2885 printf("nr_sleep_events: %ld\n", sched->nr_sleep_events);
2886 printf("nr_wakeup_events: %ld\n", sched->nr_wakeup_events);
0ec04e16 2887
0e9b07e5
ACM		 2888 if (sched->targetless_wakeups)
2889 printf("target-less wakeups: %ld\n", sched->targetless_wakeups);
2890 if (sched->multitarget_wakeups)
2891 printf("multi-target wakeups: %ld\n", sched->multitarget_wakeups);
2892 if (sched->nr_run_events_optimized)
0ec04e16 2893 printf("run atoms optimized: %ld\n",
0e9b07e5 2894 sched->nr_run_events_optimized);
0ec04e16 2895
0e9b07e5
ACM		 2896 print_task_traces(sched);
2897 add_cross_task_wakeups(sched);
0ec04e16 2898
0e9b07e5 2899 create_tasks(sched);
0ec04e16 2900 printf("------------------------------------------------------------\n");
0e9b07e5
ACM		 2901 for (i = 0; i < sched->replay_repeat; i++)
2902 run_one_test(sched);
a116e05d
ACM		 2903
2904 return 0;
0ec04e16
IM		 2905}
2906
0e9b07e5
ACM		 2907static void setup_sorting(struct perf_sched *sched, const struct option *options,
2908 const char * const usage_msg[])
daa1d7a5 2909{
0e9b07e5 2910 char *tmp, *tok, *str = strdup(sched->sort_order);
daa1d7a5
FW		 2911
2912 for (tok = strtok_r(str, ", ", &tmp);
2913 tok; tok = strtok_r(NULL, ", ", &tmp)) {
0e9b07e5 2914 if (sort_dimension__add(tok, &sched->sort_list) < 0) {
c7118369
NK		 2915 usage_with_options_msg(usage_msg, options,
2916 "Unknown --sort key: `%s'", tok);
daa1d7a5
FW		 2917 }
2918 }
2919
2920 free(str);
2921
0e9b07e5 2922 sort_dimension__add("pid", &sched->cmp_pid);
daa1d7a5
FW		 2923}
2924
1fc35b29
IM		 2925static int __cmd_record(int argc, const char **argv)
2926{
2927 unsigned int rec_argc, i, j;
2928 const char **rec_argv;
0e9b07e5
ACM		 2929 const char * const record_args[] = {
2930 "record",
2931 "-a",
2932 "-R",
0e9b07e5
ACM		 2933 "-m", "1024",
2934 "-c", "1",
2935 "-e", "sched:sched_switch",
2936 "-e", "sched:sched_stat_wait",
2937 "-e", "sched:sched_stat_sleep",
2938 "-e", "sched:sched_stat_iowait",
2939 "-e", "sched:sched_stat_runtime",
0e9b07e5
ACM		 2940 "-e", "sched:sched_process_fork",
2941 "-e", "sched:sched_wakeup",
7fff9597 2942 "-e", "sched:sched_wakeup_new",
0e9b07e5
ACM		 2943 "-e", "sched:sched_migrate_task",
2944 };
1fc35b29
IM		 2945
2946 rec_argc = ARRAY_SIZE(record_args) + argc - 1;
2947 rec_argv = calloc(rec_argc + 1, sizeof(char *));
2948
e462dc55 2949 if (rec_argv == NULL)
ce47dc56
CS		 2950 return -ENOMEM;
2951
1fc35b29
IM		 2952 for (i = 0; i < ARRAY_SIZE(record_args); i++)
2953 rec_argv[i] = strdup(record_args[i]);
2954
2955 for (j = 1; j < (unsigned int)argc; j++, i++)
2956 rec_argv[i] = argv[j];
2957
2958 BUG_ON(i != rec_argc);
2959
2960 return cmd_record(i, rec_argv, NULL);
2961}
2962
1d037ca1 2963int cmd_sched(int argc, const char **argv, const char *prefix __maybe_unused)
0a02ad93 2964{
8a39df8f
AH		 2965 const char default_sort_order[] = "avg, max, switch, runtime";
2966 struct perf_sched sched = {
2967 .tool = {
2968 .sample = perf_sched__process_tracepoint_sample,
2969 .comm = perf_event__process_comm,
2970 .lost = perf_event__process_lost,
2971 .fork = perf_sched__process_fork_event,
0a8cb85c 2972 .ordered_events = true,
8a39df8f
AH		 2973 },
2974 .cmp_pid = LIST_HEAD_INIT(sched.cmp_pid),
2975 .sort_list = LIST_HEAD_INIT(sched.sort_list),
2976 .start_work_mutex = PTHREAD_MUTEX_INITIALIZER,
2977 .work_done_wait_mutex = PTHREAD_MUTEX_INITIALIZER,
8a39df8f
AH		 2978 .sort_order = default_sort_order,
2979 .replay_repeat = 10,
2980 .profile_cpu = -1,
2981 .next_shortname1 = 'A',
2982 .next_shortname2 = '0',
2f80dd44 2983 .skip_merge = 0,
6c973c90
DA		 2984 .show_callchain = 1,
2985 .max_stack = 5,
8a39df8f 2986 };
77f02f44
NK		 2987 const struct option sched_options[] = {
2988 OPT_STRING('i', "input", &input_name, "file",
2989 "input file name"),
2990 OPT_INCR('v', "verbose", &verbose,
2991 "be more verbose (show symbol address, etc)"),
2992 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
2993 "dump raw trace in ASCII"),
6fa94258 2994 OPT_BOOLEAN('f', "force", &sched.force, "don't complain, do it"),
77f02f44
NK		 2995 OPT_END()
2996 };
0e9b07e5
ACM		 2997 const struct option latency_options[] = {
2998 OPT_STRING('s', "sort", &sched.sort_order, "key[,key2...]",
2999 "sort by key(s): runtime, switch, avg, max"),
0e9b07e5
ACM		 3000 OPT_INTEGER('C', "CPU", &sched.profile_cpu,
3001 "CPU to profile on"),
2f80dd44
JB		 3002 OPT_BOOLEAN('p', "pids", &sched.skip_merge,
3003 "latency stats per pid instead of per comm"),
77f02f44 3004 OPT_PARENT(sched_options)
0e9b07e5
ACM		 3005 };
3006 const struct option replay_options[] = {
3007 OPT_UINTEGER('r', "repeat", &sched.replay_repeat,
3008 "repeat the workload replay N times (-1: infinite)"),
77f02f44 3009 OPT_PARENT(sched_options)
0e9b07e5 3010 };
99623c62
JO		 3011 const struct option map_options[] = {
3012 OPT_BOOLEAN(0, "compact", &sched.map.comp,
3013 "map output in compact mode"),
a151a37a
JO		 3014 OPT_STRING(0, "color-pids", &sched.map.color_pids_str, "pids",
3015 "highlight given pids in map"),
cf294f24
JO		 3016 OPT_STRING(0, "color-cpus", &sched.map.color_cpus_str, "cpus",
3017 "highlight given CPUs in map"),
73643bb6
JO		 3018 OPT_STRING(0, "cpus", &sched.map.cpus_str, "cpus",
3019 "display given CPUs in map"),
77f02f44 3020 OPT_PARENT(sched_options)
99623c62 3021 };
49394a2a
DA		 3022 const struct option timehist_options[] = {
3023 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
3024 "file", "vmlinux pathname"),
3025 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
3026 "file", "kallsyms pathname"),
6c973c90
DA		 3027 OPT_BOOLEAN('g', "call-graph", &sched.show_callchain,
3028 "Display call chains if present (default on)"),
3029 OPT_UINTEGER(0, "max-stack", &sched.max_stack,
3030 "Maximum number of functions to display backtrace."),
49394a2a
DA		 3031 OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory",
3032 "Look for files with symbols relative to this directory"),
52df138c
DA		 3033 OPT_BOOLEAN('s', "summary", &sched.summary_only,
3034 "Show only syscall summary with statistics"),
3035 OPT_BOOLEAN('S', "with-summary", &sched.summary,
3036 "Show all syscalls and summary with statistics"),
fc1469f1 3037 OPT_BOOLEAN('w', "wakeups", &sched.show_wakeups, "Show wakeup events"),
350f54fa 3038 OPT_BOOLEAN('M', "migrations", &sched.show_migrations, "Show migration events"),
a407b067 3039 OPT_BOOLEAN('V', "cpu-visual", &sched.show_cpu_visual, "Add CPU visual"),
853b7407
DA		 3040 OPT_STRING(0, "time", &sched.time_str, "str",
3041 "Time span for analysis (start,stop)"),
49394a2a
DA		 3042 OPT_PARENT(sched_options)
3043 };
3044
0e9b07e5
ACM		 3045 const char * const latency_usage[] = {
3046 "perf sched latency [<options>]",
3047 NULL
3048 };
3049 const char * const replay_usage[] = {
3050 "perf sched replay [<options>]",
3051 NULL
3052 };
99623c62
JO		 3053 const char * const map_usage[] = {
3054 "perf sched map [<options>]",
3055 NULL
3056 };
49394a2a
DA		 3057 const char * const timehist_usage[] = {
3058 "perf sched timehist [<options>]",
3059 NULL
3060 };
a83edb2d 3061 const char *const sched_subcommands[] = { "record", "latency", "map",
49394a2a
DA		 3062 "replay", "script",
3063 "timehist", NULL };
a83edb2d
RR		 3064 const char *sched_usage[] = {
3065 NULL,
0e9b07e5
ACM		 3066 NULL
3067 };
3068 struct trace_sched_handler lat_ops = {
3069 .wakeup_event = latency_wakeup_event,
3070 .switch_event = latency_switch_event,
3071 .runtime_event = latency_runtime_event,
0e9b07e5
ACM		 3072 .migrate_task_event = latency_migrate_task_event,
3073 };
3074 struct trace_sched_handler map_ops = {
3075 .switch_event = map_switch_event,
3076 };
3077 struct trace_sched_handler replay_ops = {
3078 .wakeup_event = replay_wakeup_event,
3079 .switch_event = replay_switch_event,
3080 .fork_event = replay_fork_event,
3081 };
156a2b02
AH		 3082 unsigned int i;
3083
3084 for (i = 0; i < ARRAY_SIZE(sched.curr_pid); i++)
3085 sched.curr_pid[i] = -1;
0e9b07e5 3086
a83edb2d
RR		 3087 argc = parse_options_subcommand(argc, argv, sched_options, sched_subcommands,
3088 sched_usage, PARSE_OPT_STOP_AT_NON_OPTION);
f2858d8a
IM		 3089 if (!argc)
3090 usage_with_options(sched_usage, sched_options);
0a02ad93 3091
c0777c5a 3092 /*
133dc4c3 3093 * Aliased to 'perf script' for now:
c0777c5a 3094 */
133dc4c3
IM		 3095 if (!strcmp(argv[0], "script"))
3096 return cmd_script(argc, argv, prefix);
c0777c5a 3097
1fc35b29
IM		 3098 if (!strncmp(argv[0], "rec", 3)) {
3099 return __cmd_record(argc, argv);
3100 } else if (!strncmp(argv[0], "lat", 3)) {
0e9b07e5 3101 sched.tp_handler = &lat_ops;
f2858d8a
IM		 3102 if (argc > 1) {
3103 argc = parse_options(argc, argv, latency_options, latency_usage, 0);
3104 if (argc)
3105 usage_with_options(latency_usage, latency_options);
f2858d8a 3106 }
0e9b07e5
ACM		 3107 setup_sorting(&sched, latency_options, latency_usage);
3108 return perf_sched__lat(&sched);
0ec04e16 3109 } else if (!strcmp(argv[0], "map")) {
99623c62 3110 if (argc) {
a151a37a 3111 argc = parse_options(argc, argv, map_options, map_usage, 0);
99623c62
JO		 3112 if (argc)
3113 usage_with_options(map_usage, map_options);
3114 }
0e9b07e5
ACM		 3115 sched.tp_handler = &map_ops;
3116 setup_sorting(&sched, latency_options, latency_usage);
3117 return perf_sched__map(&sched);
f2858d8a 3118 } else if (!strncmp(argv[0], "rep", 3)) {
0e9b07e5 3119 sched.tp_handler = &replay_ops;
f2858d8a
IM		 3120 if (argc) {
3121 argc = parse_options(argc, argv, replay_options, replay_usage, 0);
3122 if (argc)
3123 usage_with_options(replay_usage, replay_options);
3124 }
0e9b07e5 3125 return perf_sched__replay(&sched);
49394a2a
DA		 3126 } else if (!strcmp(argv[0], "timehist")) {
3127 if (argc) {
3128 argc = parse_options(argc, argv, timehist_options,
3129 timehist_usage, 0);
3130 if (argc)
3131 usage_with_options(timehist_usage, timehist_options);
3132 }
fc1469f1
DA		 3133 if (sched.show_wakeups && sched.summary_only) {
3134 pr_err(" Error: -s and -w are mutually exclusive.\n");
3135 parse_options_usage(timehist_usage, timehist_options, "s", true);
3136 parse_options_usage(NULL, timehist_options, "w", true);
3137 return -EINVAL;
3138 }
3139
49394a2a 3140 return perf_sched__timehist(&sched);
f2858d8a
IM		 3141 } else {
3142 usage_with_options(sched_usage, sched_options);
3143 }
3144
ec156764 3145 return 0;
0a02ad93 3146}
Linux 6.x block layer and io_uring tree(s)