]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - tools/perf/builtin-sched.c
projects / mirror_ubuntu-artful-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
perf pmu: Pass pmu as a parameter to get_cpuid_str()
[mirror_ubuntu-artful-kernel.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
877a7a11 25#include <linux/kernel.h>
49394a2a 26#include <linux/log2.h>
b1ffe8f3 27#include <sys/prctl.h>
7b78f136 28#include <sys/resource.h>
fd20e811 29#include <inttypes.h>
0a02ad93 30
a43783ae 31#include <errno.h>
b1ffe8f3
IM		 32#include <semaphore.h>
33#include <pthread.h>
34#include <math.h>
cb06ac25 35#include <api/fs/fs.h>
4fc76e49 36#include <linux/time64.h>
419ab0d6 37
3d689ed6
ACM		 38#include "sane_ctype.h"
39
b1ffe8f3
IM		 40#define PR_SET_NAME 15 /* Set process name */
41#define MAX_CPUS 4096
b1ffe8f3
IM		 42#define COMM_LEN 20
43#define SYM_LEN 129
a35e27d0 44#define MAX_PID 1024000
ec156764 45
39aeb52f 46struct sched_atom;
ec156764 47
b1ffe8f3
IM		 48struct task_desc {
49 unsigned long nr;
50 unsigned long pid;
51 char comm[COMM_LEN];
ec156764 52
b1ffe8f3
IM		 53 unsigned long nr_events;
54 unsigned long curr_event;
39aeb52f 55 struct sched_atom **atoms;
b1ffe8f3
IM		 56
57 pthread_t thread;
58 sem_t sleep_sem;
ec156764 59
b1ffe8f3
IM		 60 sem_t ready_for_work;
61 sem_t work_done_sem;
62
63 u64 cpu_usage;
64};
65
66enum sched_event_type {
67 SCHED_EVENT_RUN,
68 SCHED_EVENT_SLEEP,
69 SCHED_EVENT_WAKEUP,
55ffb7a6 70 SCHED_EVENT_MIGRATION,
b1ffe8f3
IM		 71};
72
39aeb52f 73struct sched_atom {
b1ffe8f3 74 enum sched_event_type type;
eed05fe7 75 int specific_wait;
b1ffe8f3
IM		 76 u64 timestamp;
77 u64 duration;
78 unsigned long nr;
b1ffe8f3
IM		 79 sem_t *wait_sem;
80 struct task_desc *wakee;
81};
82
e936e8e4 83#define TASK_STATE_TO_CHAR_STR "RSDTtZXxKWP"
b1ffe8f3 84
941bdea7
NK		 85/* task state bitmask, copied from include/linux/sched.h */
86#define TASK_RUNNING 0
87#define TASK_INTERRUPTIBLE 1
88#define TASK_UNINTERRUPTIBLE 2
89#define __TASK_STOPPED 4
90#define __TASK_TRACED 8
91/* in tsk->exit_state */
92#define EXIT_DEAD 16
93#define EXIT_ZOMBIE 32
94#define EXIT_TRACE (EXIT_ZOMBIE | EXIT_DEAD)
95/* in tsk->state again */
96#define TASK_DEAD 64
97#define TASK_WAKEKILL 128
98#define TASK_WAKING 256
99#define TASK_PARKED 512
100
b1ffe8f3
IM		 101enum thread_state {
102 THREAD_SLEEPING = 0,
103 THREAD_WAIT_CPU,
104 THREAD_SCHED_IN,
105 THREAD_IGNORE
106};
107
108struct work_atom {
109 struct list_head list;
110 enum thread_state state;
aa1ab9d2 111 u64 sched_out_time;
b1ffe8f3
IM		 112 u64 wake_up_time;
113 u64 sched_in_time;
114 u64 runtime;
115};
116
39aeb52f 117struct work_atoms {
118 struct list_head work_list;
b1ffe8f3
IM		 119 struct thread *thread;
120 struct rb_node node;
121 u64 max_lat;
3786310a 122 u64 max_lat_at;
b1ffe8f3
IM		 123 u64 total_lat;
124 u64 nb_atoms;
125 u64 total_runtime;
2f80dd44 126 int num_merged;
b1ffe8f3
IM		 127};
128
39aeb52f 129typedef int (*sort_fn_t)(struct work_atoms *, struct work_atoms *);
b1ffe8f3 130
9ec3f4e4 131struct perf_sched;
0e9b07e5 132
9ec3f4e4
ACM		 133struct trace_sched_handler {
134 int (*switch_event)(struct perf_sched *sched, struct perf_evsel *evsel,
135 struct perf_sample *sample, struct machine *machine);
0e9b07e5 136
9ec3f4e4
ACM		 137 int (*runtime_event)(struct perf_sched *sched, struct perf_evsel *evsel,
138 struct perf_sample *sample, struct machine *machine);
0e9b07e5 139
9ec3f4e4
ACM		 140 int (*wakeup_event)(struct perf_sched *sched, struct perf_evsel *evsel,
141 struct perf_sample *sample, struct machine *machine);
0e9b07e5 142
cb627505
DA		 143 /* PERF_RECORD_FORK event, not sched_process_fork tracepoint */
144 int (*fork_event)(struct perf_sched *sched, union perf_event *event,
145 struct machine *machine);
0e9b07e5
ACM		 146
147 int (*migrate_task_event)(struct perf_sched *sched,
9ec3f4e4
ACM		 148 struct perf_evsel *evsel,
149 struct perf_sample *sample,
150 struct machine *machine);
0e9b07e5
ACM		 151};
152
a151a37a 153#define COLOR_PIDS PERF_COLOR_BLUE
cf294f24 154#define COLOR_CPUS PERF_COLOR_BG_RED
a151a37a 155
99623c62
JO		 156struct perf_sched_map {
157 DECLARE_BITMAP(comp_cpus_mask, MAX_CPUS);
158 int *comp_cpus;
159 bool comp;
a151a37a
JO		 160 struct thread_map *color_pids;
161 const char *color_pids_str;
cf294f24
JO		 162 struct cpu_map *color_cpus;
163 const char *color_cpus_str;
73643bb6
JO		 164 struct cpu_map *cpus;
165 const char *cpus_str;
99623c62
JO		 166};
167
0e9b07e5
ACM		 168struct perf_sched {
169 struct perf_tool tool;
0e9b07e5
ACM		 170 const char *sort_order;
171 unsigned long nr_tasks;
cb06ac25 172 struct task_desc **pid_to_task;
0e9b07e5
ACM		 173 struct task_desc **tasks;
174 const struct trace_sched_handler *tp_handler;
175 pthread_mutex_t start_work_mutex;
176 pthread_mutex_t work_done_wait_mutex;
177 int profile_cpu;
178/*
179 * Track the current task - that way we can know whether there's any
180 * weird events, such as a task being switched away that is not current.
181 */
182 int max_cpu;
183 u32 curr_pid[MAX_CPUS];
184 struct thread *curr_thread[MAX_CPUS];
185 char next_shortname1;
186 char next_shortname2;
187 unsigned int replay_repeat;
188 unsigned long nr_run_events;
189 unsigned long nr_sleep_events;
190 unsigned long nr_wakeup_events;
191 unsigned long nr_sleep_corrections;
192 unsigned long nr_run_events_optimized;
193 unsigned long targetless_wakeups;
194 unsigned long multitarget_wakeups;
195 unsigned long nr_runs;
196 unsigned long nr_timestamps;
197 unsigned long nr_unordered_timestamps;
0e9b07e5
ACM		 198 unsigned long nr_context_switch_bugs;
199 unsigned long nr_events;
200 unsigned long nr_lost_chunks;
201 unsigned long nr_lost_events;
202 u64 run_measurement_overhead;
203 u64 sleep_measurement_overhead;
204 u64 start_time;
205 u64 cpu_usage;
206 u64 runavg_cpu_usage;
207 u64 parent_cpu_usage;
208 u64 runavg_parent_cpu_usage;
209 u64 sum_runtime;
210 u64 sum_fluct;
211 u64 run_avg;
212 u64 all_runtime;
213 u64 all_count;
214 u64 cpu_last_switched[MAX_CPUS];
2f80dd44 215 struct rb_root atom_root, sorted_atom_root, merged_atom_root;
0e9b07e5 216 struct list_head sort_list, cmp_pid;
939cda52 217 bool force;
2f80dd44 218 bool skip_merge;
99623c62 219 struct perf_sched_map map;
52df138c
DA		 220
221 /* options for timehist command */
222 bool summary;
223 bool summary_only;
699b5b92 224 bool idle_hist;
6c973c90
DA		 225 bool show_callchain;
226 unsigned int max_stack;
a407b067 227 bool show_cpu_visual;
fc1469f1 228 bool show_wakeups;
292c4a8f 229 bool show_next;
350f54fa 230 bool show_migrations;
414e050c 231 bool show_state;
52df138c 232 u64 skipped_samples;
853b7407
DA		 233 const char *time_str;
234 struct perf_time_interval ptime;
9396c9cb 235 struct perf_time_interval hist_time;
0e9b07e5 236};
b1ffe8f3 237
49394a2a
DA		 238/* per thread run time data */
239struct thread_runtime {
240 u64 last_time; /* time of previous sched in/out event */
241 u64 dt_run; /* run time */
941bdea7
NK		 242 u64 dt_sleep; /* time between CPU access by sleep (off cpu) */
243 u64 dt_iowait; /* time between CPU access by iowait (off cpu) */
244 u64 dt_preempt; /* time between CPU access by preempt (off cpu) */
49394a2a
DA		 245 u64 dt_delay; /* time between wakeup and sched-in */
246 u64 ready_to_run; /* time of wakeup */
247
248 struct stats run_stats;
249 u64 total_run_time;
587782c5
NK		 250 u64 total_sleep_time;
251 u64 total_iowait_time;
252 u64 total_preempt_time;
253 u64 total_delay_time;
350f54fa 254
941bdea7 255 int last_state;
350f54fa 256 u64 migrations;
49394a2a
DA		 257};
258
259/* per event run time data */
260struct evsel_runtime {
261 u64 *last_time; /* time this event was last seen per cpu */
262 u32 ncpu; /* highest cpu slot allocated */
263};
264
3bc2fa9c
NK		 265/* per cpu idle time data */
266struct idle_thread_runtime {
267 struct thread_runtime tr;
268 struct thread *last_thread;
269 struct rb_root sorted_root;
270 struct callchain_root callchain;
271 struct callchain_cursor cursor;
272};
273
49394a2a
DA		 274/* track idle times per cpu */
275static struct thread **idle_threads;
276static int idle_max_cpu;
277static char idle_comm[] = "<idle>";
278
b1ffe8f3 279static u64 get_nsecs(void)
ec156764
IM		 280{
281 struct timespec ts;
282
283 clock_gettime(CLOCK_MONOTONIC, &ts);
284
4fc76e49 285 return ts.tv_sec * NSEC_PER_SEC + ts.tv_nsec;
ec156764
IM		 286}
287
0e9b07e5 288static void burn_nsecs(struct perf_sched *sched, u64 nsecs)
ec156764 289{
b1ffe8f3 290 u64 T0 = get_nsecs(), T1;
ec156764
IM		 291
292 do {
293 T1 = get_nsecs();
0e9b07e5 294 } while (T1 + sched->run_measurement_overhead < T0 + nsecs);
ec156764
IM		 295}
296
b1ffe8f3 297static void sleep_nsecs(u64 nsecs)
ec156764
IM		 298{
299 struct timespec ts;
300
301 ts.tv_nsec = nsecs % 999999999;
302 ts.tv_sec = nsecs / 999999999;
303
304 nanosleep(&ts, NULL);
305}
306
0e9b07e5 307static void calibrate_run_measurement_overhead(struct perf_sched *sched)
ec156764 308{
4fc76e49 309 u64 T0, T1, delta, min_delta = NSEC_PER_SEC;
ec156764
IM		 310 int i;
311
312 for (i = 0; i < 10; i++) {
313 T0 = get_nsecs();
0e9b07e5 314 burn_nsecs(sched, 0);
ec156764
IM		 315 T1 = get_nsecs();
316 delta = T1-T0;
317 min_delta = min(min_delta, delta);
318 }
0e9b07e5 319 sched->run_measurement_overhead = min_delta;
ec156764 320
9486aa38 321 printf("run measurement overhead: %" PRIu64 " nsecs\n", min_delta);
ec156764
IM		 322}
323
0e9b07e5 324static void calibrate_sleep_measurement_overhead(struct perf_sched *sched)
ec156764 325{
4fc76e49 326 u64 T0, T1, delta, min_delta = NSEC_PER_SEC;
ec156764
IM		 327 int i;
328
329 for (i = 0; i < 10; i++) {
330 T0 = get_nsecs();
331 sleep_nsecs(10000);
332 T1 = get_nsecs();
333 delta = T1-T0;
334 min_delta = min(min_delta, delta);
335 }
336 min_delta -= 10000;
0e9b07e5 337 sched->sleep_measurement_overhead = min_delta;
ec156764 338
9486aa38 339 printf("sleep measurement overhead: %" PRIu64 " nsecs\n", min_delta);
ec156764
IM		 340}
341
39aeb52f 342static struct sched_atom *
b1ffe8f3 343get_new_event(struct task_desc *task, u64 timestamp)
ec156764 344{
36479484 345 struct sched_atom *event = zalloc(sizeof(*event));
ec156764
IM		 346 unsigned long idx = task->nr_events;
347 size_t size;
348
349 event->timestamp = timestamp;
350 event->nr = idx;
351
352 task->nr_events++;
39aeb52f 353 size = sizeof(struct sched_atom *) * task->nr_events;
354 task->atoms = realloc(task->atoms, size);
355 BUG_ON(!task->atoms);
ec156764 356
39aeb52f 357 task->atoms[idx] = event;
ec156764
IM		 358
359 return event;
360}
361
39aeb52f 362static struct sched_atom *last_event(struct task_desc *task)
ec156764
IM		 363{
364 if (!task->nr_events)
365 return NULL;
366
39aeb52f 367 return task->atoms[task->nr_events - 1];
ec156764
IM		 368}
369
0e9b07e5
ACM		 370static void add_sched_event_run(struct perf_sched *sched, struct task_desc *task,
371 u64 timestamp, u64 duration)
ec156764 372{
39aeb52f 373 struct sched_atom *event, *curr_event = last_event(task);
ec156764
IM		 374
375 /*
fbf94829
IM		 376 * optimize an existing RUN event by merging this one
377 * to it:
378 */
ec156764 379 if (curr_event && curr_event->type == SCHED_EVENT_RUN) {
0e9b07e5 380 sched->nr_run_events_optimized++;
ec156764
IM		 381 curr_event->duration += duration;
382 return;
383 }
384
385 event = get_new_event(task, timestamp);
386
387 event->type = SCHED_EVENT_RUN;
388 event->duration = duration;
389
0e9b07e5 390 sched->nr_run_events++;
ec156764
IM		 391}
392
0e9b07e5
ACM		 393static void add_sched_event_wakeup(struct perf_sched *sched, struct task_desc *task,
394 u64 timestamp, struct task_desc *wakee)
ec156764 395{
39aeb52f 396 struct sched_atom *event, *wakee_event;
ec156764
IM		 397
398 event = get_new_event(task, timestamp);
399 event->type = SCHED_EVENT_WAKEUP;
400 event->wakee = wakee;
401
402 wakee_event = last_event(wakee);
403 if (!wakee_event || wakee_event->type != SCHED_EVENT_SLEEP) {
0e9b07e5 404 sched->targetless_wakeups++;
ec156764
IM		 405 return;
406 }
407 if (wakee_event->wait_sem) {
0e9b07e5 408 sched->multitarget_wakeups++;
ec156764
IM		 409 return;
410 }
411
36479484 412 wakee_event->wait_sem = zalloc(sizeof(*wakee_event->wait_sem));
ec156764
IM		 413 sem_init(wakee_event->wait_sem, 0, 0);
414 wakee_event->specific_wait = 1;
415 event->wait_sem = wakee_event->wait_sem;
416
0e9b07e5 417 sched->nr_wakeup_events++;
ec156764
IM		 418}
419
0e9b07e5
ACM		 420static void add_sched_event_sleep(struct perf_sched *sched, struct task_desc *task,
421 u64 timestamp, u64 task_state __maybe_unused)
ec156764 422{
39aeb52f 423 struct sched_atom *event = get_new_event(task, timestamp);
ec156764
IM		 424
425 event->type = SCHED_EVENT_SLEEP;
426
0e9b07e5 427 sched->nr_sleep_events++;
ec156764
IM		 428}
429
0e9b07e5
ACM		 430static struct task_desc *register_pid(struct perf_sched *sched,
431 unsigned long pid, const char *comm)
ec156764
IM		 432{
433 struct task_desc *task;
cb06ac25 434 static int pid_max;
ec156764 435
cb06ac25
YS		 436 if (sched->pid_to_task == NULL) {
437 if (sysctl__read_int("kernel/pid_max", &pid_max) < 0)
438 pid_max = MAX_PID;
439 BUG_ON((sched->pid_to_task = calloc(pid_max, sizeof(struct task_desc *))) == NULL);
440 }
3a423a5c
YS		 441 if (pid >= (unsigned long)pid_max) {
442 BUG_ON((sched->pid_to_task = realloc(sched->pid_to_task, (pid + 1) *
443 sizeof(struct task_desc *))) == NULL);
444 while (pid >= (unsigned long)pid_max)
445 sched->pid_to_task[pid_max++] = NULL;
446 }
ec156764 447
0e9b07e5 448 task = sched->pid_to_task[pid];
ec156764
IM		 449
450 if (task)
451 return task;
452
36479484 453 task = zalloc(sizeof(*task));
ec156764 454 task->pid = pid;
0e9b07e5 455 task->nr = sched->nr_tasks;
ec156764
IM		 456 strcpy(task->comm, comm);
457 /*
458 * every task starts in sleeping state - this gets ignored
459 * if there's no wakeup pointing to this sleep state:
460 */
0e9b07e5 461 add_sched_event_sleep(sched, task, 0, 0);
ec156764 462
0e9b07e5
ACM		 463 sched->pid_to_task[pid] = task;
464 sched->nr_tasks++;
0755bc4d 465 sched->tasks = realloc(sched->tasks, sched->nr_tasks * sizeof(struct task_desc *));
0e9b07e5
ACM		 466 BUG_ON(!sched->tasks);
467 sched->tasks[task->nr] = task;
ec156764 468
bb963e16 469 if (verbose > 0)
0e9b07e5 470 printf("registered task #%ld, PID %ld (%s)\n", sched->nr_tasks, pid, comm);
ec156764
IM		 471
472 return task;
473}
474
475
0e9b07e5 476static void print_task_traces(struct perf_sched *sched)
ec156764
IM		 477{
478 struct task_desc *task;
479 unsigned long i;
480
0e9b07e5
ACM		 481 for (i = 0; i < sched->nr_tasks; i++) {
482 task = sched->tasks[i];
ad236fd2 483 printf("task %6ld (%20s:%10ld), nr_events: %ld\n",
ec156764
IM		 484 task->nr, task->comm, task->pid, task->nr_events);
485 }
486}
487
0e9b07e5 488static void add_cross_task_wakeups(struct perf_sched *sched)
ec156764
IM		 489{
490 struct task_desc *task1, *task2;
491 unsigned long i, j;
492
0e9b07e5
ACM		 493 for (i = 0; i < sched->nr_tasks; i++) {
494 task1 = sched->tasks[i];
ec156764 495 j = i + 1;
0e9b07e5 496 if (j == sched->nr_tasks)
ec156764 497 j = 0;
0e9b07e5
ACM		 498 task2 = sched->tasks[j];
499 add_sched_event_wakeup(sched, task1, 0, task2);
ec156764
IM		 500 }
501}
502
0e9b07e5
ACM		 503static void perf_sched__process_event(struct perf_sched *sched,
504 struct sched_atom *atom)
ec156764
IM		 505{
506 int ret = 0;
ec156764 507
39aeb52f 508 switch (atom->type) {
ec156764 509 case SCHED_EVENT_RUN:
0e9b07e5 510 burn_nsecs(sched, atom->duration);
ec156764
IM		 511 break;
512 case SCHED_EVENT_SLEEP:
39aeb52f 513 if (atom->wait_sem)
514 ret = sem_wait(atom->wait_sem);
ec156764
IM		 515 BUG_ON(ret);
516 break;
517 case SCHED_EVENT_WAKEUP:
39aeb52f 518 if (atom->wait_sem)
519 ret = sem_post(atom->wait_sem);
ec156764
IM		 520 BUG_ON(ret);
521 break;
55ffb7a6
MG		 522 case SCHED_EVENT_MIGRATION:
523 break;
ec156764
IM		 524 default:
525 BUG_ON(1);
526 }
527}
528
b1ffe8f3 529static u64 get_cpu_usage_nsec_parent(void)
ec156764
IM		 530{
531 struct rusage ru;
b1ffe8f3 532 u64 sum;
ec156764
IM		 533 int err;
534
535 err = getrusage(RUSAGE_SELF, &ru);
536 BUG_ON(err);
537
4fc76e49
ACM		 538 sum = ru.ru_utime.tv_sec * NSEC_PER_SEC + ru.ru_utime.tv_usec * NSEC_PER_USEC;
539 sum += ru.ru_stime.tv_sec * NSEC_PER_SEC + ru.ru_stime.tv_usec * NSEC_PER_USEC;
ec156764
IM		 540
541 return sum;
542}
543
939cda52 544static int self_open_counters(struct perf_sched *sched, unsigned long cur_task)
ec156764 545{
c0c9e721 546 struct perf_event_attr attr;
939cda52 547 char sbuf[STRERR_BUFSIZE], info[STRERR_BUFSIZE];
c0c9e721 548 int fd;
939cda52
YS		 549 struct rlimit limit;
550 bool need_privilege = false;
ec156764 551
c0c9e721 552 memset(&attr, 0, sizeof(attr));
ec156764 553
c0c9e721
XG		 554 attr.type = PERF_TYPE_SOFTWARE;
555 attr.config = PERF_COUNT_SW_TASK_CLOCK;
ec156764 556
939cda52 557force_again:
57480d2c
YD		 558 fd = sys_perf_event_open(&attr, 0, -1, -1,
559 perf_event_open_cloexec_flag());
c0c9e721 560
1aff59be 561 if (fd < 0) {
939cda52
YS		 562 if (errno == EMFILE) {
563 if (sched->force) {
564 BUG_ON(getrlimit(RLIMIT_NOFILE, &limit) == -1);
565 limit.rlim_cur += sched->nr_tasks - cur_task;
566 if (limit.rlim_cur > limit.rlim_max) {
567 limit.rlim_max = limit.rlim_cur;
568 need_privilege = true;
569 }
570 if (setrlimit(RLIMIT_NOFILE, &limit) == -1) {
571 if (need_privilege && errno == EPERM)
572 strcpy(info, "Need privilege\n");
573 } else
574 goto force_again;
575 } else
576 strcpy(info, "Have a try with -f option\n");
577 }
60b7d14a 578 pr_err("Error: sys_perf_event_open() syscall returned "
939cda52 579 "with %d (%s)\n%s", fd,
c8b5f2c9 580 str_error_r(errno, sbuf, sizeof(sbuf)), info);
1aff59be
YS		 581 exit(EXIT_FAILURE);
582 }
c0c9e721
XG		 583 return fd;
584}
585
586static u64 get_cpu_usage_nsec_self(int fd)
587{
588 u64 runtime;
589 int ret;
590
591 ret = read(fd, &runtime, sizeof(runtime));
592 BUG_ON(ret != sizeof(runtime));
593
594 return runtime;
ec156764
IM		 595}
596
0e9b07e5
ACM		 597struct sched_thread_parms {
598 struct task_desc *task;
599 struct perf_sched *sched;
08097abc 600 int fd;
0e9b07e5
ACM		 601};
602
ec156764
IM		 603static void *thread_func(void *ctx)
604{
0e9b07e5
ACM		 605 struct sched_thread_parms *parms = ctx;
606 struct task_desc *this_task = parms->task;
607 struct perf_sched *sched = parms->sched;
b1ffe8f3 608 u64 cpu_usage_0, cpu_usage_1;
ec156764
IM		 609 unsigned long i, ret;
610 char comm2[22];
08097abc 611 int fd = parms->fd;
ec156764 612
74cf249d 613 zfree(&parms);
0e9b07e5 614
ec156764
IM		 615 sprintf(comm2, ":%s", this_task->comm);
616 prctl(PR_SET_NAME, comm2);
a116e05d
ACM		 617 if (fd < 0)
618 return NULL;
ec156764
IM		 619again:
620 ret = sem_post(&this_task->ready_for_work);
621 BUG_ON(ret);
0e9b07e5 622 ret = pthread_mutex_lock(&sched->start_work_mutex);
ec156764 623 BUG_ON(ret);
0e9b07e5 624 ret = pthread_mutex_unlock(&sched->start_work_mutex);
ec156764 625 BUG_ON(ret);
ec156764 626
c0c9e721 627 cpu_usage_0 = get_cpu_usage_nsec_self(fd);
ec156764
IM		 628
629 for (i = 0; i < this_task->nr_events; i++) {
630 this_task->curr_event = i;
0e9b07e5 631 perf_sched__process_event(sched, this_task->atoms[i]);
ec156764
IM		 632 }
633
c0c9e721 634 cpu_usage_1 = get_cpu_usage_nsec_self(fd);
ec156764 635 this_task->cpu_usage = cpu_usage_1 - cpu_usage_0;
ec156764
IM		 636 ret = sem_post(&this_task->work_done_sem);
637 BUG_ON(ret);
ec156764 638
0e9b07e5 639 ret = pthread_mutex_lock(&sched->work_done_wait_mutex);
ec156764 640 BUG_ON(ret);
0e9b07e5 641 ret = pthread_mutex_unlock(&sched->work_done_wait_mutex);
ec156764 642 BUG_ON(ret);
ec156764
IM		 643
644 goto again;
645}
646
0e9b07e5 647static void create_tasks(struct perf_sched *sched)
ec156764
IM		 648{
649 struct task_desc *task;
650 pthread_attr_t attr;
651 unsigned long i;
652 int err;
653
654 err = pthread_attr_init(&attr);
655 BUG_ON(err);
12f7e036
JP		 656 err = pthread_attr_setstacksize(&attr,
657 (size_t) max(16 * 1024, PTHREAD_STACK_MIN));
ec156764 658 BUG_ON(err);
0e9b07e5 659 err = pthread_mutex_lock(&sched->start_work_mutex);
ec156764 660 BUG_ON(err);
0e9b07e5 661 err = pthread_mutex_lock(&sched->work_done_wait_mutex);
ec156764 662 BUG_ON(err);
0e9b07e5
ACM		 663 for (i = 0; i < sched->nr_tasks; i++) {
664 struct sched_thread_parms *parms = malloc(sizeof(*parms));
665 BUG_ON(parms == NULL);
666 parms->task = task = sched->tasks[i];
667 parms->sched = sched;
939cda52 668 parms->fd = self_open_counters(sched, i);
ec156764
IM		 669 sem_init(&task->sleep_sem, 0, 0);
670 sem_init(&task->ready_for_work, 0, 0);
671 sem_init(&task->work_done_sem, 0, 0);
672 task->curr_event = 0;
0e9b07e5 673 err = pthread_create(&task->thread, &attr, thread_func, parms);
ec156764
IM		 674 BUG_ON(err);
675 }
676}
677
0e9b07e5 678static void wait_for_tasks(struct perf_sched *sched)
ec156764 679{
b1ffe8f3 680 u64 cpu_usage_0, cpu_usage_1;
ec156764
IM		 681 struct task_desc *task;
682 unsigned long i, ret;
683
0e9b07e5
ACM		 684 sched->start_time = get_nsecs();
685 sched->cpu_usage = 0;
686 pthread_mutex_unlock(&sched->work_done_wait_mutex);
ec156764 687
0e9b07e5
ACM		 688 for (i = 0; i < sched->nr_tasks; i++) {
689 task = sched->tasks[i];
ec156764
IM		 690 ret = sem_wait(&task->ready_for_work);
691 BUG_ON(ret);
692 sem_init(&task->ready_for_work, 0, 0);
693 }
0e9b07e5 694 ret = pthread_mutex_lock(&sched->work_done_wait_mutex);
ec156764
IM		 695 BUG_ON(ret);
696
697 cpu_usage_0 = get_cpu_usage_nsec_parent();
698
0e9b07e5 699 pthread_mutex_unlock(&sched->start_work_mutex);
ec156764 700
0e9b07e5
ACM		 701 for (i = 0; i < sched->nr_tasks; i++) {
702 task = sched->tasks[i];
ec156764
IM		 703 ret = sem_wait(&task->work_done_sem);
704 BUG_ON(ret);
705 sem_init(&task->work_done_sem, 0, 0);
0e9b07e5 706 sched->cpu_usage += task->cpu_usage;
ec156764
IM		 707 task->cpu_usage = 0;
708 }
709
710 cpu_usage_1 = get_cpu_usage_nsec_parent();
0e9b07e5
ACM		 711 if (!sched->runavg_cpu_usage)
712 sched->runavg_cpu_usage = sched->cpu_usage;
ff5f3bbd 713 sched->runavg_cpu_usage = (sched->runavg_cpu_usage * (sched->replay_repeat - 1) + sched->cpu_usage) / sched->replay_repeat;
ec156764 714
0e9b07e5
ACM		 715 sched->parent_cpu_usage = cpu_usage_1 - cpu_usage_0;
716 if (!sched->runavg_parent_cpu_usage)
717 sched->runavg_parent_cpu_usage = sched->parent_cpu_usage;
ff5f3bbd
YS		 718 sched->runavg_parent_cpu_usage = (sched->runavg_parent_cpu_usage * (sched->replay_repeat - 1) +
719 sched->parent_cpu_usage)/sched->replay_repeat;
ec156764 720
0e9b07e5 721 ret = pthread_mutex_lock(&sched->start_work_mutex);
ec156764
IM		 722 BUG_ON(ret);
723
0e9b07e5
ACM		 724 for (i = 0; i < sched->nr_tasks; i++) {
725 task = sched->tasks[i];
ec156764
IM		 726 sem_init(&task->sleep_sem, 0, 0);
727 task->curr_event = 0;
728 }
729}
730
0e9b07e5 731static void run_one_test(struct perf_sched *sched)
ec156764 732{
fb7d0b3c 733 u64 T0, T1, delta, avg_delta, fluct;
ec156764
IM		 734
735 T0 = get_nsecs();
0e9b07e5 736 wait_for_tasks(sched);
ec156764
IM		 737 T1 = get_nsecs();
738
739 delta = T1 - T0;
0e9b07e5
ACM		 740 sched->sum_runtime += delta;
741 sched->nr_runs++;
ec156764 742
0e9b07e5 743 avg_delta = sched->sum_runtime / sched->nr_runs;
ec156764
IM		 744 if (delta < avg_delta)
745 fluct = avg_delta - delta;
746 else
747 fluct = delta - avg_delta;
0e9b07e5
ACM		 748 sched->sum_fluct += fluct;
749 if (!sched->run_avg)
750 sched->run_avg = delta;
ff5f3bbd 751 sched->run_avg = (sched->run_avg * (sched->replay_repeat - 1) + delta) / sched->replay_repeat;
ec156764 752
4fc76e49 753 printf("#%-3ld: %0.3f, ", sched->nr_runs, (double)delta / NSEC_PER_MSEC);
ec156764 754
4fc76e49 755 printf("ravg: %0.2f, ", (double)sched->run_avg / NSEC_PER_MSEC);
ec156764 756
ad236fd2 757 printf("cpu: %0.2f / %0.2f",
4fc76e49 758 (double)sched->cpu_usage / NSEC_PER_MSEC, (double)sched->runavg_cpu_usage / NSEC_PER_MSEC);
ec156764
IM		 759
760#if 0
761 /*
fbf94829 762 * rusage statistics done by the parent, these are less
0e9b07e5 763 * accurate than the sched->sum_exec_runtime based statistics:
fbf94829 764 */
ad236fd2 765 printf(" [%0.2f / %0.2f]",
4fc76e49
ACM		 766 (double)sched->parent_cpu_usage / NSEC_PER_MSEC,
767 (double)sched->runavg_parent_cpu_usage / NSEC_PER_MSEC);
ec156764
IM		 768#endif
769
ad236fd2 770 printf("\n");
ec156764 771
0e9b07e5
ACM		 772 if (sched->nr_sleep_corrections)
773 printf(" (%ld sleep corrections)\n", sched->nr_sleep_corrections);
774 sched->nr_sleep_corrections = 0;
ec156764
IM		 775}
776
0e9b07e5 777static void test_calibrations(struct perf_sched *sched)
ec156764 778{
b1ffe8f3 779 u64 T0, T1;
ec156764
IM		 780
781 T0 = get_nsecs();
4fc76e49 782 burn_nsecs(sched, NSEC_PER_MSEC);
ec156764
IM		 783 T1 = get_nsecs();
784
9486aa38 785 printf("the run test took %" PRIu64 " nsecs\n", T1 - T0);
ec156764
IM		 786
787 T0 = get_nsecs();
4fc76e49 788 sleep_nsecs(NSEC_PER_MSEC);
ec156764
IM		 789 T1 = get_nsecs();
790
9486aa38 791 printf("the sleep test took %" PRIu64 " nsecs\n", T1 - T0);
ec156764
IM		 792}
793
a116e05d 794static int
0e9b07e5 795replay_wakeup_event(struct perf_sched *sched,
9ec3f4e4
ACM		 796 struct perf_evsel *evsel, struct perf_sample *sample,
797 struct machine *machine __maybe_unused)
419ab0d6 798{
9ec3f4e4
ACM		 799 const char *comm = perf_evsel__strval(evsel, sample, "comm");
800 const u32 pid = perf_evsel__intval(evsel, sample, "pid");
419ab0d6 801 struct task_desc *waker, *wakee;
fbf94829 802
bb963e16 803 if (verbose > 0) {
2b7fcbc5 804 printf("sched_wakeup event %p\n", evsel);
fbf94829 805
9ec3f4e4 806 printf(" ... pid %d woke up %s/%d\n", sample->tid, comm, pid);
ad236fd2 807 }
fbf94829 808
2b7fcbc5 809 waker = register_pid(sched, sample->tid, "<unknown>");
9ec3f4e4 810 wakee = register_pid(sched, pid, comm);
fbf94829 811
0e9b07e5 812 add_sched_event_wakeup(sched, waker, sample->time, wakee);
a116e05d 813 return 0;
ec156764
IM		 814}
815
9ec3f4e4
ACM		 816static int replay_switch_event(struct perf_sched *sched,
817 struct perf_evsel *evsel,
818 struct perf_sample *sample,
819 struct machine *machine __maybe_unused)
ec156764 820{
9ec3f4e4
ACM		 821 const char *prev_comm = perf_evsel__strval(evsel, sample, "prev_comm"),
822 *next_comm = perf_evsel__strval(evsel, sample, "next_comm");
823 const u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"),
824 next_pid = perf_evsel__intval(evsel, sample, "next_pid");
825 const u64 prev_state = perf_evsel__intval(evsel, sample, "prev_state");
1d037ca1 826 struct task_desc *prev, __maybe_unused *next;
7f7f8d0b
ACM		 827 u64 timestamp0, timestamp = sample->time;
828 int cpu = sample->cpu;
fbf94829
IM		 829 s64 delta;
830
bb963e16 831 if (verbose > 0)
2b7fcbc5 832 printf("sched_switch event %p\n", evsel);
ad236fd2 833
fbf94829 834 if (cpu >= MAX_CPUS || cpu < 0)
a116e05d 835 return 0;
fbf94829 836
0e9b07e5 837 timestamp0 = sched->cpu_last_switched[cpu];
fbf94829
IM		 838 if (timestamp0)
839 delta = timestamp - timestamp0;
840 else
841 delta = 0;
842
a116e05d 843 if (delta < 0) {
60b7d14a 844 pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta);
a116e05d
ACM		 845 return -1;
846 }
fbf94829 847
9ec3f4e4
ACM		 848 pr_debug(" ... switch from %s/%d to %s/%d [ran %" PRIu64 " nsecs]\n",
849 prev_comm, prev_pid, next_comm, next_pid, delta);
fbf94829 850
9ec3f4e4
ACM		 851 prev = register_pid(sched, prev_pid, prev_comm);
852 next = register_pid(sched, next_pid, next_comm);
fbf94829 853
0e9b07e5 854 sched->cpu_last_switched[cpu] = timestamp;
fbf94829 855
0e9b07e5 856 add_sched_event_run(sched, prev, timestamp, delta);
9ec3f4e4 857 add_sched_event_sleep(sched, prev, timestamp, prev_state);
a116e05d
ACM		 858
859 return 0;
fbf94829
IM		 860}
861
cb627505
DA		 862static int replay_fork_event(struct perf_sched *sched,
863 union perf_event *event,
864 struct machine *machine)
419ab0d6 865{
cb627505
DA		 866 struct thread *child, *parent;
867
314add6b
AH		 868 child = machine__findnew_thread(machine, event->fork.pid,
869 event->fork.tid);
870 parent = machine__findnew_thread(machine, event->fork.ppid,
871 event->fork.ptid);
cb627505
DA		 872
873 if (child == NULL || parent == NULL) {
874 pr_debug("thread does not exist on fork event: child %p, parent %p\n",
875 child, parent);
b91fc39f 876 goto out_put;
cb627505 877 }
9ec3f4e4 878
bb963e16 879 if (verbose > 0) {
cb627505 880 printf("fork event\n");
b9c5143a
FW		 881 printf("... parent: %s/%d\n", thread__comm_str(parent), parent->tid);
882 printf("... child: %s/%d\n", thread__comm_str(child), child->tid);
419ab0d6 883 }
9ec3f4e4 884
b9c5143a
FW		 885 register_pid(sched, parent->tid, thread__comm_str(parent));
886 register_pid(sched, child->tid, thread__comm_str(child));
b91fc39f
ACM		 887out_put:
888 thread__put(child);
889 thread__put(parent);
a116e05d 890 return 0;
419ab0d6 891}
fbf94829 892
b1ffe8f3
IM		 893struct sort_dimension {
894 const char *name;
b5fae128 895 sort_fn_t cmp;
b1ffe8f3
IM		 896 struct list_head list;
897};
898
daa1d7a5 899static int
39aeb52f 900thread_lat_cmp(struct list_head *list, struct work_atoms *l, struct work_atoms *r)
daa1d7a5
FW		 901{
902 struct sort_dimension *sort;
903 int ret = 0;
904
b5fae128
IM		 905 BUG_ON(list_empty(list));
906
daa1d7a5
FW		 907 list_for_each_entry(sort, list, list) {
908 ret = sort->cmp(l, r);
909 if (ret)
910 return ret;
911 }
912
913 return ret;
914}
915
39aeb52f 916static struct work_atoms *
b5fae128
IM		 917thread_atoms_search(struct rb_root *root, struct thread *thread,
918 struct list_head *sort_list)
919{
920 struct rb_node *node = root->rb_node;
39aeb52f 921 struct work_atoms key = { .thread = thread };
b5fae128
IM		 922
923 while (node) {
39aeb52f 924 struct work_atoms *atoms;
b5fae128
IM		 925 int cmp;
926
39aeb52f 927 atoms = container_of(node, struct work_atoms, node);
b5fae128
IM		 928
929 cmp = thread_lat_cmp(sort_list, &key, atoms);
930 if (cmp > 0)
931 node = node->rb_left;
932 else if (cmp < 0)
933 node = node->rb_right;
934 else {
935 BUG_ON(thread != atoms->thread);
936 return atoms;
937 }
938 }
939 return NULL;
940}
941
cdce9d73 942static void
39aeb52f 943__thread_latency_insert(struct rb_root *root, struct work_atoms *data,
daa1d7a5 944 struct list_head *sort_list)
cdce9d73
FW		 945{
946 struct rb_node **new = &(root->rb_node), *parent = NULL;
947
948 while (*new) {
39aeb52f 949 struct work_atoms *this;
daa1d7a5 950 int cmp;
cdce9d73 951
39aeb52f 952 this = container_of(*new, struct work_atoms, node);
cdce9d73 953 parent = *new;
daa1d7a5
FW		 954
955 cmp = thread_lat_cmp(sort_list, data, this);
956
957 if (cmp > 0)
cdce9d73 958 new = &((*new)->rb_left);
cdce9d73 959 else
daa1d7a5 960 new = &((*new)->rb_right);
cdce9d73
FW		 961 }
962
963 rb_link_node(&data->node, parent, new);
964 rb_insert_color(&data->node, root);
965}
966
0e9b07e5 967static int thread_atoms_insert(struct perf_sched *sched, struct thread *thread)
cdce9d73 968{
36479484 969 struct work_atoms *atoms = zalloc(sizeof(*atoms));
a116e05d
ACM		 970 if (!atoms) {
971 pr_err("No memory at %s\n", __func__);
972 return -1;
973 }
cdce9d73 974
f3b623b8 975 atoms->thread = thread__get(thread);
39aeb52f 976 INIT_LIST_HEAD(&atoms->work_list);
0e9b07e5 977 __thread_latency_insert(&sched->atom_root, atoms, &sched->cmp_pid);
a116e05d 978 return 0;
cdce9d73
FW		 979}
980
9ec3f4e4 981static char sched_out_state(u64 prev_state)
cdce9d73
FW		 982{
983 const char *str = TASK_STATE_TO_CHAR_STR;
984
9ec3f4e4 985 return str[prev_state];
cdce9d73
FW		 986}
987
a116e05d 988static int
39aeb52f 989add_sched_out_event(struct work_atoms *atoms,
990 char run_state,
991 u64 timestamp)
cdce9d73 992{
36479484 993 struct work_atom *atom = zalloc(sizeof(*atom));
a116e05d
ACM		 994 if (!atom) {
995 pr_err("Non memory at %s", __func__);
996 return -1;
997 }
cdce9d73 998
aa1ab9d2
FW		 999 atom->sched_out_time = timestamp;
1000
39aeb52f 1001 if (run_state == 'R') {
b1ffe8f3 1002 atom->state = THREAD_WAIT_CPU;
aa1ab9d2 1003 atom->wake_up_time = atom->sched_out_time;
c6ced611
FW		 1004 }
1005
39aeb52f 1006 list_add_tail(&atom->list, &atoms->work_list);
a116e05d 1007 return 0;
cdce9d73
FW		 1008}
1009
1010static void
1d037ca1
IT		 1011add_runtime_event(struct work_atoms *atoms, u64 delta,
1012 u64 timestamp __maybe_unused)
39aeb52f 1013{
1014 struct work_atom *atom;
1015
1016 BUG_ON(list_empty(&atoms->work_list));
1017
1018 atom = list_entry(atoms->work_list.prev, struct work_atom, list);
1019
1020 atom->runtime += delta;
1021 atoms->total_runtime += delta;
1022}
1023
1024static void
1025add_sched_in_event(struct work_atoms *atoms, u64 timestamp)
cdce9d73 1026{
b1ffe8f3 1027 struct work_atom *atom;
66685678 1028 u64 delta;
cdce9d73 1029
39aeb52f 1030 if (list_empty(&atoms->work_list))
cdce9d73
FW		 1031 return;
1032
39aeb52f 1033 atom = list_entry(atoms->work_list.prev, struct work_atom, list);
cdce9d73 1034
b1ffe8f3 1035 if (atom->state != THREAD_WAIT_CPU)
cdce9d73
FW		 1036 return;
1037
b1ffe8f3
IM		 1038 if (timestamp < atom->wake_up_time) {
1039 atom->state = THREAD_IGNORE;
cdce9d73
FW		 1040 return;
1041 }
1042
b1ffe8f3
IM		 1043 atom->state = THREAD_SCHED_IN;
1044 atom->sched_in_time = timestamp;
66685678 1045
b1ffe8f3 1046 delta = atom->sched_in_time - atom->wake_up_time;
66685678 1047 atoms->total_lat += delta;
3786310a 1048 if (delta > atoms->max_lat) {
66685678 1049 atoms->max_lat = delta;
3786310a
FW		 1050 atoms->max_lat_at = timestamp;
1051 }
66685678 1052 atoms->nb_atoms++;
cdce9d73
FW		 1053}
1054
9ec3f4e4
ACM		 1055static int latency_switch_event(struct perf_sched *sched,
1056 struct perf_evsel *evsel,
1057 struct perf_sample *sample,
1058 struct machine *machine)
cdce9d73 1059{
9ec3f4e4
ACM		 1060 const u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"),
1061 next_pid = perf_evsel__intval(evsel, sample, "next_pid");
1062 const u64 prev_state = perf_evsel__intval(evsel, sample, "prev_state");
39aeb52f 1063 struct work_atoms *out_events, *in_events;
cdce9d73 1064 struct thread *sched_out, *sched_in;
7f7f8d0b 1065 u64 timestamp0, timestamp = sample->time;
b91fc39f 1066 int cpu = sample->cpu, err = -1;
ea92ed5a
IM		 1067 s64 delta;
1068
39aeb52f 1069 BUG_ON(cpu >= MAX_CPUS || cpu < 0);
ea92ed5a 1070
0e9b07e5
ACM		 1071 timestamp0 = sched->cpu_last_switched[cpu];
1072 sched->cpu_last_switched[cpu] = timestamp;
ea92ed5a
IM		 1073 if (timestamp0)
1074 delta = timestamp - timestamp0;
1075 else
1076 delta = 0;
1077
a116e05d
ACM		 1078 if (delta < 0) {
1079 pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta);
1080 return -1;
1081 }
cdce9d73 1082
1fcb8768
AH		 1083 sched_out = machine__findnew_thread(machine, -1, prev_pid);
1084 sched_in = machine__findnew_thread(machine, -1, next_pid);
b91fc39f
ACM		 1085 if (sched_out == NULL || sched_in == NULL)
1086 goto out_put;
cdce9d73 1087
0e9b07e5 1088 out_events = thread_atoms_search(&sched->atom_root, sched_out, &sched->cmp_pid);
39aeb52f 1089 if (!out_events) {
0e9b07e5 1090 if (thread_atoms_insert(sched, sched_out))
b91fc39f 1091 goto out_put;
0e9b07e5 1092 out_events = thread_atoms_search(&sched->atom_root, sched_out, &sched->cmp_pid);
a116e05d
ACM		 1093 if (!out_events) {
1094 pr_err("out-event: Internal tree error");
b91fc39f 1095 goto out_put;
a116e05d 1096 }
39aeb52f 1097 }
9ec3f4e4 1098 if (add_sched_out_event(out_events, sched_out_state(prev_state), timestamp))
a116e05d 1099 return -1;
39aeb52f 1100
0e9b07e5 1101 in_events = thread_atoms_search(&sched->atom_root, sched_in, &sched->cmp_pid);
39aeb52f 1102 if (!in_events) {
0e9b07e5 1103 if (thread_atoms_insert(sched, sched_in))
b91fc39f 1104 goto out_put;
0e9b07e5 1105 in_events = thread_atoms_search(&sched->atom_root, sched_in, &sched->cmp_pid);
a116e05d
ACM		 1106 if (!in_events) {
1107 pr_err("in-event: Internal tree error");
b91fc39f 1108 goto out_put;
a116e05d 1109 }
39aeb52f 1110 /*
1111 * Take came in we have not heard about yet,
1112 * add in an initial atom in runnable state:
1113 */
a116e05d 1114 if (add_sched_out_event(in_events, 'R', timestamp))
b91fc39f 1115 goto out_put;
cdce9d73 1116 }
39aeb52f 1117 add_sched_in_event(in_events, timestamp);
b91fc39f
ACM		 1118 err = 0;
1119out_put:
1120 thread__put(sched_out);
1121 thread__put(sched_in);
1122 return err;
39aeb52f 1123}
cdce9d73 1124
9ec3f4e4
ACM		 1125static int latency_runtime_event(struct perf_sched *sched,
1126 struct perf_evsel *evsel,
1127 struct perf_sample *sample,
1128 struct machine *machine)
39aeb52f 1129{
9ec3f4e4
ACM		 1130 const u32 pid = perf_evsel__intval(evsel, sample, "pid");
1131 const u64 runtime = perf_evsel__intval(evsel, sample, "runtime");
1fcb8768 1132 struct thread *thread = machine__findnew_thread(machine, -1, pid);
0e9b07e5 1133 struct work_atoms *atoms = thread_atoms_search(&sched->atom_root, thread, &sched->cmp_pid);
7f7f8d0b 1134 u64 timestamp = sample->time;
b91fc39f
ACM		 1135 int cpu = sample->cpu, err = -1;
1136
1137 if (thread == NULL)
1138 return -1;
39aeb52f 1139
1140 BUG_ON(cpu >= MAX_CPUS || cpu < 0);
39aeb52f 1141 if (!atoms) {
0e9b07e5 1142 if (thread_atoms_insert(sched, thread))
b91fc39f 1143 goto out_put;
0e9b07e5 1144 atoms = thread_atoms_search(&sched->atom_root, thread, &sched->cmp_pid);
a116e05d 1145 if (!atoms) {
60b7d14a 1146 pr_err("in-event: Internal tree error");
b91fc39f 1147 goto out_put;
a116e05d
ACM		 1148 }
1149 if (add_sched_out_event(atoms, 'R', timestamp))
b91fc39f 1150 goto out_put;
cdce9d73
FW		 1151 }
1152
9ec3f4e4 1153 add_runtime_event(atoms, runtime, timestamp);
b91fc39f
ACM		 1154 err = 0;
1155out_put:
1156 thread__put(thread);
1157 return err;
cdce9d73
FW		 1158}
1159
9ec3f4e4
ACM		 1160static int latency_wakeup_event(struct perf_sched *sched,
1161 struct perf_evsel *evsel,
1162 struct perf_sample *sample,
1163 struct machine *machine)
cdce9d73 1164{
0680ee7d 1165 const u32 pid = perf_evsel__intval(evsel, sample, "pid");
39aeb52f 1166 struct work_atoms *atoms;
b1ffe8f3 1167 struct work_atom *atom;
cdce9d73 1168 struct thread *wakee;
7f7f8d0b 1169 u64 timestamp = sample->time;
b91fc39f 1170 int err = -1;
cdce9d73 1171
1fcb8768 1172 wakee = machine__findnew_thread(machine, -1, pid);
b91fc39f
ACM		 1173 if (wakee == NULL)
1174 return -1;
0e9b07e5 1175 atoms = thread_atoms_search(&sched->atom_root, wakee, &sched->cmp_pid);
17562205 1176 if (!atoms) {
0e9b07e5 1177 if (thread_atoms_insert(sched, wakee))
b91fc39f 1178 goto out_put;
0e9b07e5 1179 atoms = thread_atoms_search(&sched->atom_root, wakee, &sched->cmp_pid);
a116e05d 1180 if (!atoms) {
60b7d14a 1181 pr_err("wakeup-event: Internal tree error");
b91fc39f 1182 goto out_put;
a116e05d
ACM		 1183 }
1184 if (add_sched_out_event(atoms, 'S', timestamp))
b91fc39f 1185 goto out_put;
cdce9d73
FW		 1186 }
1187
39aeb52f 1188 BUG_ON(list_empty(&atoms->work_list));
cdce9d73 1189
39aeb52f 1190 atom = list_entry(atoms->work_list.prev, struct work_atom, list);
cdce9d73 1191
55ffb7a6 1192 /*
67d6259d
DY		 1193 * As we do not guarantee the wakeup event happens when
1194 * task is out of run queue, also may happen when task is
1195 * on run queue and wakeup only change ->state to TASK_RUNNING,
1196 * then we should not set the ->wake_up_time when wake up a
1197 * task which is on run queue.
1198 *
55ffb7a6
MG		 1199 * You WILL be missing events if you've recorded only
1200 * one CPU, or are only looking at only one, so don't
67d6259d 1201 * skip in this case.
55ffb7a6 1202 */
0e9b07e5 1203 if (sched->profile_cpu == -1 && atom->state != THREAD_SLEEPING)
b91fc39f 1204 goto out_ok;
cdce9d73 1205
0e9b07e5 1206 sched->nr_timestamps++;
ea57c4f5 1207 if (atom->sched_out_time > timestamp) {
0e9b07e5 1208 sched->nr_unordered_timestamps++;
b91fc39f 1209 goto out_ok;
ea57c4f5 1210 }
aa1ab9d2 1211
b1ffe8f3
IM		 1212 atom->state = THREAD_WAIT_CPU;
1213 atom->wake_up_time = timestamp;
b91fc39f
ACM		 1214out_ok:
1215 err = 0;
1216out_put:
1217 thread__put(wakee);
1218 return err;
cdce9d73
FW		 1219}
1220
9ec3f4e4
ACM		 1221static int latency_migrate_task_event(struct perf_sched *sched,
1222 struct perf_evsel *evsel,
1223 struct perf_sample *sample,
1224 struct machine *machine)
55ffb7a6 1225{
9ec3f4e4 1226 const u32 pid = perf_evsel__intval(evsel, sample, "pid");
7f7f8d0b 1227 u64 timestamp = sample->time;
55ffb7a6
MG		 1228 struct work_atoms *atoms;
1229 struct work_atom *atom;
1230 struct thread *migrant;
b91fc39f 1231 int err = -1;
55ffb7a6
MG		 1232
1233 /*
1234 * Only need to worry about migration when profiling one CPU.
1235 */
0e9b07e5 1236 if (sched->profile_cpu == -1)
a116e05d 1237 return 0;
55ffb7a6 1238
1fcb8768 1239 migrant = machine__findnew_thread(machine, -1, pid);
b91fc39f
ACM		 1240 if (migrant == NULL)
1241 return -1;
0e9b07e5 1242 atoms = thread_atoms_search(&sched->atom_root, migrant, &sched->cmp_pid);
55ffb7a6 1243 if (!atoms) {
0e9b07e5 1244 if (thread_atoms_insert(sched, migrant))
b91fc39f 1245 goto out_put;
b9c5143a 1246 register_pid(sched, migrant->tid, thread__comm_str(migrant));
0e9b07e5 1247 atoms = thread_atoms_search(&sched->atom_root, migrant, &sched->cmp_pid);
a116e05d 1248 if (!atoms) {
60b7d14a 1249 pr_err("migration-event: Internal tree error");
b91fc39f 1250 goto out_put;
a116e05d
ACM		 1251 }
1252 if (add_sched_out_event(atoms, 'R', timestamp))
b91fc39f 1253 goto out_put;
55ffb7a6
MG		 1254 }
1255
1256 BUG_ON(list_empty(&atoms->work_list));
1257
1258 atom = list_entry(atoms->work_list.prev, struct work_atom, list);
1259 atom->sched_in_time = atom->sched_out_time = atom->wake_up_time = timestamp;
1260
0e9b07e5 1261 sched->nr_timestamps++;
55ffb7a6
MG		 1262
1263 if (atom->sched_out_time > timestamp)
0e9b07e5 1264 sched->nr_unordered_timestamps++;
b91fc39f
ACM		 1265 err = 0;
1266out_put:
1267 thread__put(migrant);
1268 return err;
55ffb7a6
MG		 1269}
1270
0e9b07e5 1271static void output_lat_thread(struct perf_sched *sched, struct work_atoms *work_list)
cdce9d73 1272{
cdce9d73
FW		 1273 int i;
1274 int ret;
66685678 1275 u64 avg;
99620a5d 1276 char max_lat_at[32];
cdce9d73 1277
39aeb52f 1278 if (!work_list->nb_atoms)
cdce9d73 1279 return;
ea57c4f5
IM		 1280 /*
1281 * Ignore idle threads:
1282 */
b9c5143a 1283 if (!strcmp(thread__comm_str(work_list->thread), "swapper"))
ea57c4f5 1284 return;
cdce9d73 1285
0e9b07e5
ACM		 1286 sched->all_runtime += work_list->total_runtime;
1287 sched->all_count += work_list->nb_atoms;
66685678 1288
2f80dd44
JB		 1289 if (work_list->num_merged > 1)
1290 ret = printf(" %s:(%d) ", thread__comm_str(work_list->thread), work_list->num_merged);
1291 else
1292 ret = printf(" %s:%d ", thread__comm_str(work_list->thread), work_list->thread->tid);
cdce9d73 1293
08f69e6c 1294 for (i = 0; i < 24 - ret; i++)
cdce9d73
FW		 1295 printf(" ");
1296
39aeb52f 1297 avg = work_list->total_lat / work_list->nb_atoms;
99620a5d 1298 timestamp__scnprintf_usec(work_list->max_lat_at, max_lat_at, sizeof(max_lat_at));
cdce9d73 1299
99620a5d 1300 printf("|%11.3f ms |%9" PRIu64 " | avg:%9.3f ms | max:%9.3f ms | max at: %13s s\n",
4fc76e49
ACM		 1301 (double)work_list->total_runtime / NSEC_PER_MSEC,
1302 work_list->nb_atoms, (double)avg / NSEC_PER_MSEC,
1303 (double)work_list->max_lat / NSEC_PER_MSEC,
99620a5d 1304 max_lat_at);
cdce9d73
FW		 1305}
1306
39aeb52f 1307static int pid_cmp(struct work_atoms *l, struct work_atoms *r)
daa1d7a5 1308{
0014de17
JO		 1309 if (l->thread == r->thread)
1310 return 0;
38051234 1311 if (l->thread->tid < r->thread->tid)
daa1d7a5 1312 return -1;
38051234 1313 if (l->thread->tid > r->thread->tid)
daa1d7a5 1314 return 1;
0014de17 1315 return (int)(l->thread - r->thread);
daa1d7a5
FW		 1316}
1317
39aeb52f 1318static int avg_cmp(struct work_atoms *l, struct work_atoms *r)
daa1d7a5
FW		 1319{
1320 u64 avgl, avgr;
1321
1322 if (!l->nb_atoms)
1323 return -1;
1324
1325 if (!r->nb_atoms)
1326 return 1;
1327
1328 avgl = l->total_lat / l->nb_atoms;
1329 avgr = r->total_lat / r->nb_atoms;
1330
1331 if (avgl < avgr)
1332 return -1;
1333 if (avgl > avgr)
1334 return 1;
1335
1336 return 0;
1337}
1338
39aeb52f 1339static int max_cmp(struct work_atoms *l, struct work_atoms *r)
daa1d7a5
FW		 1340{
1341 if (l->max_lat < r->max_lat)
1342 return -1;
1343 if (l->max_lat > r->max_lat)
1344 return 1;
1345
1346 return 0;
1347}
1348
39aeb52f 1349static int switch_cmp(struct work_atoms *l, struct work_atoms *r)
daa1d7a5
FW		 1350{
1351 if (l->nb_atoms < r->nb_atoms)
1352 return -1;
1353 if (l->nb_atoms > r->nb_atoms)
1354 return 1;
1355
1356 return 0;
1357}
1358
39aeb52f 1359static int runtime_cmp(struct work_atoms *l, struct work_atoms *r)
daa1d7a5
FW		 1360{
1361 if (l->total_runtime < r->total_runtime)
1362 return -1;
1363 if (l->total_runtime > r->total_runtime)
1364 return 1;
1365
1366 return 0;
1367}
1368
cbef79a8 1369static int sort_dimension__add(const char *tok, struct list_head *list)
daa1d7a5 1370{
0e9b07e5
ACM		 1371 size_t i;
1372 static struct sort_dimension avg_sort_dimension = {
1373 .name = "avg",
1374 .cmp = avg_cmp,
1375 };
1376 static struct sort_dimension max_sort_dimension = {
1377 .name = "max",
1378 .cmp = max_cmp,
1379 };
1380 static struct sort_dimension pid_sort_dimension = {
1381 .name = "pid",
1382 .cmp = pid_cmp,
1383 };
1384 static struct sort_dimension runtime_sort_dimension = {
1385 .name = "runtime",
1386 .cmp = runtime_cmp,
1387 };
1388 static struct sort_dimension switch_sort_dimension = {
1389 .name = "switch",
1390 .cmp = switch_cmp,
1391 };
1392 struct sort_dimension *available_sorts[] = {
1393 &pid_sort_dimension,
1394 &avg_sort_dimension,
1395 &max_sort_dimension,
1396 &switch_sort_dimension,
1397 &runtime_sort_dimension,
1398 };
daa1d7a5 1399
0e9b07e5 1400 for (i = 0; i < ARRAY_SIZE(available_sorts); i++) {
daa1d7a5
FW		 1401 if (!strcmp(available_sorts[i]->name, tok)) {
1402 list_add_tail(&available_sorts[i]->list, list);
1403
1404 return 0;
1405 }
1406 }
1407
1408 return -1;
1409}
1410
0e9b07e5 1411static void perf_sched__sort_lat(struct perf_sched *sched)
daa1d7a5
FW		 1412{
1413 struct rb_node *node;
2f80dd44
JB		 1414 struct rb_root *root = &sched->atom_root;
1415again:
daa1d7a5 1416 for (;;) {
39aeb52f 1417 struct work_atoms *data;
2f80dd44 1418 node = rb_first(root);
daa1d7a5
FW		 1419 if (!node)
1420 break;
1421
2f80dd44 1422 rb_erase(node, root);
39aeb52f 1423 data = rb_entry(node, struct work_atoms, node);
0e9b07e5 1424 __thread_latency_insert(&sched->sorted_atom_root, data, &sched->sort_list);
daa1d7a5 1425 }
2f80dd44
JB		 1426 if (root == &sched->atom_root) {
1427 root = &sched->merged_atom_root;
1428 goto again;
1429 }
daa1d7a5
FW		 1430}
1431
0e9b07e5 1432static int process_sched_wakeup_event(struct perf_tool *tool,
2b7fcbc5 1433 struct perf_evsel *evsel,
1d037ca1 1434 struct perf_sample *sample,
4218e673 1435 struct machine *machine)
419ab0d6 1436{
0e9b07e5 1437 struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
419ab0d6 1438
9ec3f4e4
ACM		 1439 if (sched->tp_handler->wakeup_event)
1440 return sched->tp_handler->wakeup_event(sched, evsel, sample, machine);
a116e05d 1441
2b7fcbc5 1442 return 0;
419ab0d6
FW		 1443}
1444
a151a37a
JO		 1445union map_priv {
1446 void *ptr;
1447 bool color;
1448};
1449
1450static bool thread__has_color(struct thread *thread)
1451{
1452 union map_priv priv = {
1453 .ptr = thread__priv(thread),
1454 };
1455
1456 return priv.color;
1457}
1458
1459static struct thread*
1460map__findnew_thread(struct perf_sched *sched, struct machine *machine, pid_t pid, pid_t tid)
1461{
1462 struct thread *thread = machine__findnew_thread(machine, pid, tid);
1463 union map_priv priv = {
1464 .color = false,
1465 };
1466
1467 if (!sched->map.color_pids || !thread || thread__priv(thread))
1468 return thread;
1469
1470 if (thread_map__has(sched->map.color_pids, tid))
1471 priv.color = true;
1472
1473 thread__set_priv(thread, priv.ptr);
1474 return thread;
1475}
1476
9ec3f4e4
ACM		 1477static int map_switch_event(struct perf_sched *sched, struct perf_evsel *evsel,
1478 struct perf_sample *sample, struct machine *machine)
0ec04e16 1479{
9d372ca5
DY		 1480 const u32 next_pid = perf_evsel__intval(evsel, sample, "next_pid");
1481 struct thread *sched_in;
0ec04e16 1482 int new_shortname;
7f7f8d0b 1483 u64 timestamp0, timestamp = sample->time;
0ec04e16 1484 s64 delta;
99623c62
JO		 1485 int i, this_cpu = sample->cpu;
1486 int cpus_nr;
1487 bool new_cpu = false;
8cd91195 1488 const char *color = PERF_COLOR_NORMAL;
99620a5d 1489 char stimestamp[32];
0ec04e16
IM		 1490
1491 BUG_ON(this_cpu >= MAX_CPUS || this_cpu < 0);
1492
0e9b07e5
ACM		 1493 if (this_cpu > sched->max_cpu)
1494 sched->max_cpu = this_cpu;
0ec04e16 1495
99623c62
JO		 1496 if (sched->map.comp) {
1497 cpus_nr = bitmap_weight(sched->map.comp_cpus_mask, MAX_CPUS);
1498 if (!test_and_set_bit(this_cpu, sched->map.comp_cpus_mask)) {
1499 sched->map.comp_cpus[cpus_nr++] = this_cpu;
1500 new_cpu = true;
1501 }
1502 } else
1503 cpus_nr = sched->max_cpu;
1504
0e9b07e5
ACM		 1505 timestamp0 = sched->cpu_last_switched[this_cpu];
1506 sched->cpu_last_switched[this_cpu] = timestamp;
0ec04e16
IM		 1507 if (timestamp0)
1508 delta = timestamp - timestamp0;
1509 else
1510 delta = 0;
1511
a116e05d 1512 if (delta < 0) {
60b7d14a 1513 pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta);
a116e05d
ACM		 1514 return -1;
1515 }
0ec04e16 1516
a151a37a 1517 sched_in = map__findnew_thread(sched, machine, -1, next_pid);
b91fc39f
ACM		 1518 if (sched_in == NULL)
1519 return -1;
0ec04e16 1520
b91fc39f 1521 sched->curr_thread[this_cpu] = thread__get(sched_in);
0ec04e16
IM		 1522
1523 printf(" ");
1524
1525 new_shortname = 0;
1526 if (!sched_in->shortname[0]) {
6bcab4e1
D		 1527 if (!strcmp(thread__comm_str(sched_in), "swapper")) {
1528 /*
1529 * Don't allocate a letter-number for swapper:0
1530 * as a shortname. Instead, we use '.' for it.
1531 */
1532 sched_in->shortname[0] = '.';
1533 sched_in->shortname[1] = ' ';
0ec04e16 1534 } else {
6bcab4e1
D		 1535 sched_in->shortname[0] = sched->next_shortname1;
1536 sched_in->shortname[1] = sched->next_shortname2;
1537
1538 if (sched->next_shortname1 < 'Z') {
1539 sched->next_shortname1++;
0ec04e16 1540 } else {
6bcab4e1
D		 1541 sched->next_shortname1 = 'A';
1542 if (sched->next_shortname2 < '9')
1543 sched->next_shortname2++;
1544 else
1545 sched->next_shortname2 = '0';
0ec04e16
IM		 1546 }
1547 }
1548 new_shortname = 1;
1549 }
1550
99623c62
JO		 1551 for (i = 0; i < cpus_nr; i++) {
1552 int cpu = sched->map.comp ? sched->map.comp_cpus[i] : i;
a151a37a
JO		 1553 struct thread *curr_thread = sched->curr_thread[cpu];
1554 const char *pid_color = color;
cf294f24 1555 const char *cpu_color = color;
a151a37a
JO		 1556
1557 if (curr_thread && thread__has_color(curr_thread))
1558 pid_color = COLOR_PIDS;
99623c62 1559
73643bb6
JO		 1560 if (sched->map.cpus && !cpu_map__has(sched->map.cpus, cpu))
1561 continue;
1562
cf294f24
JO		 1563 if (sched->map.color_cpus && cpu_map__has(sched->map.color_cpus, cpu))
1564 cpu_color = COLOR_CPUS;
1565
0ec04e16 1566 if (cpu != this_cpu)
1208bb27 1567 color_fprintf(stdout, color, " ");
0ec04e16 1568 else
cf294f24 1569 color_fprintf(stdout, cpu_color, "*");
0ec04e16 1570
6bcab4e1 1571 if (sched->curr_thread[cpu])
a151a37a 1572 color_fprintf(stdout, pid_color, "%2s ", sched->curr_thread[cpu]->shortname);
6bcab4e1 1573 else
8cd91195 1574 color_fprintf(stdout, color, " ");
0ec04e16
IM		 1575 }
1576
73643bb6
JO		 1577 if (sched->map.cpus && !cpu_map__has(sched->map.cpus, this_cpu))
1578 goto out;
1579
99620a5d
NK		 1580 timestamp__scnprintf_usec(timestamp, stimestamp, sizeof(stimestamp));
1581 color_fprintf(stdout, color, " %12s secs ", stimestamp);
bb963e16 1582 if (new_shortname || (verbose > 0 && sched_in->tid)) {
a151a37a
JO		 1583 const char *pid_color = color;
1584
1585 if (thread__has_color(sched_in))
1586 pid_color = COLOR_PIDS;
1587
1588 color_fprintf(stdout, pid_color, "%s => %s:%d",
b9c5143a 1589 sched_in->shortname, thread__comm_str(sched_in), sched_in->tid);
0ec04e16 1590 }
a116e05d 1591
99623c62 1592 if (sched->map.comp && new_cpu)
8cd91195 1593 color_fprintf(stdout, color, " (CPU %d)", this_cpu);
99623c62 1594
73643bb6 1595out:
8cd91195 1596 color_fprintf(stdout, color, "\n");
99623c62 1597
b91fc39f
ACM		 1598 thread__put(sched_in);
1599
a116e05d 1600 return 0;
0ec04e16
IM		 1601}
1602
0e9b07e5 1603static int process_sched_switch_event(struct perf_tool *tool,
2b7fcbc5 1604 struct perf_evsel *evsel,
1d037ca1 1605 struct perf_sample *sample,
4218e673 1606 struct machine *machine)
419ab0d6 1607{
0e9b07e5 1608 struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
a116e05d 1609 int this_cpu = sample->cpu, err = 0;
2b7fcbc5
ACM		 1610 u32 prev_pid = perf_evsel__intval(evsel, sample, "prev_pid"),
1611 next_pid = perf_evsel__intval(evsel, sample, "next_pid");
419ab0d6 1612
0e9b07e5 1613 if (sched->curr_pid[this_cpu] != (u32)-1) {
c8a37751
IM		 1614 /*
1615 * Are we trying to switch away a PID that is
1616 * not current?
1617 */
2b7fcbc5 1618 if (sched->curr_pid[this_cpu] != prev_pid)
0e9b07e5 1619 sched->nr_context_switch_bugs++;
c8a37751 1620 }
c8a37751 1621
9ec3f4e4
ACM		 1622 if (sched->tp_handler->switch_event)
1623 err = sched->tp_handler->switch_event(sched, evsel, sample, machine);
2b7fcbc5
ACM		 1624
1625 sched->curr_pid[this_cpu] = next_pid;
a116e05d 1626 return err;
419ab0d6
FW		 1627}
1628
0e9b07e5 1629static int process_sched_runtime_event(struct perf_tool *tool,
2b7fcbc5 1630 struct perf_evsel *evsel,
1d037ca1 1631 struct perf_sample *sample,
4218e673 1632 struct machine *machine)
39aeb52f 1633{
0e9b07e5 1634 struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
39aeb52f 1635
9ec3f4e4
ACM		 1636 if (sched->tp_handler->runtime_event)
1637 return sched->tp_handler->runtime_event(sched, evsel, sample, machine);
a116e05d 1638
2b7fcbc5 1639 return 0;
39aeb52f 1640}
1641
cb627505
DA		 1642static int perf_sched__process_fork_event(struct perf_tool *tool,
1643 union perf_event *event,
1644 struct perf_sample *sample,
1645 struct machine *machine)
fbf94829 1646{
0e9b07e5 1647 struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
46538818 1648
cb627505
DA		 1649 /* run the fork event through the perf machineruy */
1650 perf_event__process_fork(tool, event, sample, machine);
1651
1652 /* and then run additional processing needed for this command */
9ec3f4e4 1653 if (sched->tp_handler->fork_event)
cb627505 1654 return sched->tp_handler->fork_event(sched, event, machine);
a116e05d 1655
2b7fcbc5 1656 return 0;
fbf94829
IM		 1657}
1658
0e9b07e5 1659static int process_sched_migrate_task_event(struct perf_tool *tool,
2b7fcbc5 1660 struct perf_evsel *evsel,
1d037ca1 1661 struct perf_sample *sample,
4218e673 1662 struct machine *machine)
55ffb7a6 1663{
0e9b07e5 1664 struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
55ffb7a6 1665
9ec3f4e4
ACM		 1666 if (sched->tp_handler->migrate_task_event)
1667 return sched->tp_handler->migrate_task_event(sched, evsel, sample, machine);
a116e05d 1668
2b7fcbc5 1669 return 0;
55ffb7a6
MG		 1670}
1671
a116e05d 1672typedef int (*tracepoint_handler)(struct perf_tool *tool,
2b7fcbc5 1673 struct perf_evsel *evsel,
a116e05d 1674 struct perf_sample *sample,
4218e673 1675 struct machine *machine);
ec156764 1676
1d037ca1
IT		 1677static int perf_sched__process_tracepoint_sample(struct perf_tool *tool __maybe_unused,
1678 union perf_event *event __maybe_unused,
ee29be62
ACM		 1679 struct perf_sample *sample,
1680 struct perf_evsel *evsel,
1681 struct machine *machine)
0a02ad93 1682{
a116e05d 1683 int err = 0;
0a02ad93 1684
744a9719
ACM		 1685 if (evsel->handler != NULL) {
1686 tracepoint_handler f = evsel->handler;
2b7fcbc5 1687 err = f(tool, evsel, sample, machine);
ee29be62 1688 }
0a02ad93 1689
a116e05d 1690 return err;
0a02ad93
IM		 1691}
1692
ae536acf 1693static int perf_sched__read_events(struct perf_sched *sched)
0a02ad93 1694{
ee29be62
ACM		 1695 const struct perf_evsel_str_handler handlers[] = {
1696 { "sched:sched_switch", process_sched_switch_event, },
1697 { "sched:sched_stat_runtime", process_sched_runtime_event, },
1698 { "sched:sched_wakeup", process_sched_wakeup_event, },
1699 { "sched:sched_wakeup_new", process_sched_wakeup_event, },
ee29be62
ACM		 1700 { "sched:sched_migrate_task", process_sched_migrate_task_event, },
1701 };
da378962 1702 struct perf_session *session;
f5fc1412
JO		 1703 struct perf_data_file file = {
1704 .path = input_name,
1705 .mode = PERF_DATA_MODE_READ,
f0dd330f 1706 .force = sched->force,
f5fc1412 1707 };
ae536acf 1708 int rc = -1;
da378962 1709
f5fc1412 1710 session = perf_session__new(&file, false, &sched->tool);
a116e05d
ACM		 1711 if (session == NULL) {
1712 pr_debug("No Memory for session\n");
1713 return -1;
1714 }
94c744b6 1715
0a7e6d1b 1716 symbol__init(&session->header.env);
04934106 1717
a116e05d
ACM		 1718 if (perf_session__set_tracepoints_handlers(session, handlers))
1719 goto out_delete;
ee29be62 1720
cee75ac7 1721 if (perf_session__has_traces(session, "record -R")) {
b7b61cbe 1722 int err = perf_session__process_events(session);
a116e05d
ACM		 1723 if (err) {
1724 pr_err("Failed to process events, error %d", err);
1725 goto out_delete;
1726 }
4c09bafa 1727
75be989a
ACM		 1728 sched->nr_events = session->evlist->stats.nr_events[0];
1729 sched->nr_lost_events = session->evlist->stats.total_lost;
1730 sched->nr_lost_chunks = session->evlist->stats.nr_events[PERF_RECORD_LOST];
cee75ac7 1731 }
d549c769 1732
ae536acf 1733 rc = 0;
a116e05d
ACM		 1734out_delete:
1735 perf_session__delete(session);
ae536acf 1736 return rc;
0a02ad93
IM		 1737}
1738
49394a2a
DA		 1739/*
1740 * scheduling times are printed as msec.usec
1741 */
1742static inline void print_sched_time(unsigned long long nsecs, int width)
1743{
1744 unsigned long msecs;
1745 unsigned long usecs;
1746
1747 msecs = nsecs / NSEC_PER_MSEC;
1748 nsecs -= msecs * NSEC_PER_MSEC;
1749 usecs = nsecs / NSEC_PER_USEC;
1750 printf("%*lu.%03lu ", width, msecs, usecs);
1751}
1752
1753/*
1754 * returns runtime data for event, allocating memory for it the
1755 * first time it is used.
1756 */
1757static struct evsel_runtime *perf_evsel__get_runtime(struct perf_evsel *evsel)
1758{
1759 struct evsel_runtime *r = evsel->priv;
1760
1761 if (r == NULL) {
1762 r = zalloc(sizeof(struct evsel_runtime));
1763 evsel->priv = r;
1764 }
1765
1766 return r;
1767}
1768
1769/*
1770 * save last time event was seen per cpu
1771 */
1772static void perf_evsel__save_time(struct perf_evsel *evsel,
1773 u64 timestamp, u32 cpu)
1774{
1775 struct evsel_runtime *r = perf_evsel__get_runtime(evsel);
1776
1777 if (r == NULL)
1778 return;
1779
1780 if ((cpu >= r->ncpu) || (r->last_time == NULL)) {
1781 int i, n = __roundup_pow_of_two(cpu+1);
1782 void *p = r->last_time;
1783
1784 p = realloc(r->last_time, n * sizeof(u64));
1785 if (!p)
1786 return;
1787
1788 r->last_time = p;
1789 for (i = r->ncpu; i < n; ++i)
1790 r->last_time[i] = (u64) 0;
1791
1792 r->ncpu = n;
1793 }
1794
1795 r->last_time[cpu] = timestamp;
1796}
1797
1798/* returns last time this event was seen on the given cpu */
1799static u64 perf_evsel__get_time(struct perf_evsel *evsel, u32 cpu)
1800{
1801 struct evsel_runtime *r = perf_evsel__get_runtime(evsel);
1802
1803 if ((r == NULL) || (r->last_time == NULL) || (cpu >= r->ncpu))
1804 return 0;
1805
1806 return r->last_time[cpu];
1807}
1808
9b8087d7 1809static int comm_width = 30;
49394a2a
DA		 1810
1811static char *timehist_get_commstr(struct thread *thread)
1812{
1813 static char str[32];
1814 const char *comm = thread__comm_str(thread);
1815 pid_t tid = thread->tid;
1816 pid_t pid = thread->pid_;
1817 int n;
1818
1819 if (pid == 0)
1820 n = scnprintf(str, sizeof(str), "%s", comm);
1821
1822 else if (tid != pid)
1823 n = scnprintf(str, sizeof(str), "%s[%d/%d]", comm, tid, pid);
1824
1825 else
1826 n = scnprintf(str, sizeof(str), "%s[%d]", comm, tid);
1827
1828 if (n > comm_width)
1829 comm_width = n;
1830
1831 return str;
1832}
1833
a407b067 1834static void timehist_header(struct perf_sched *sched)
49394a2a 1835{
a407b067
DA		 1836 u32 ncpus = sched->max_cpu + 1;
1837 u32 i, j;
1838
49394a2a
DA		 1839 printf("%15s %6s ", "time", "cpu");
1840
a407b067
DA		 1841 if (sched->show_cpu_visual) {
1842 printf(" ");
1843 for (i = 0, j = 0; i < ncpus; ++i) {
1844 printf("%x", j++);
1845 if (j > 15)
1846 j = 0;
1847 }
1848 printf(" ");
1849 }
1850
0e6758e8 1851 printf(" %-*s %9s %9s %9s", comm_width,
49394a2a
DA		 1852 "task name", "wait time", "sch delay", "run time");
1853
414e050c
NK		 1854 if (sched->show_state)
1855 printf(" %s", "state");
1856
49394a2a
DA		 1857 printf("\n");
1858
1859 /*
1860 * units row
1861 */
1862 printf("%15s %-6s ", "", "");
1863
a407b067
DA		 1864 if (sched->show_cpu_visual)
1865 printf(" %*s ", ncpus, "");
1866
414e050c 1867 printf(" %-*s %9s %9s %9s", comm_width,
0e6758e8 1868 "[tid/pid]", "(msec)", "(msec)", "(msec)");
49394a2a 1869
414e050c
NK		 1870 if (sched->show_state)
1871 printf(" %5s", "");
1872
1873 printf("\n");
1874
49394a2a
DA		 1875 /*
1876 * separator
1877 */
1878 printf("%.15s %.6s ", graph_dotted_line, graph_dotted_line);
1879
a407b067
DA		 1880 if (sched->show_cpu_visual)
1881 printf(" %.*s ", ncpus, graph_dotted_line);
1882
0e6758e8 1883 printf(" %.*s %.9s %.9s %.9s", comm_width,
49394a2a
DA		 1884 graph_dotted_line, graph_dotted_line, graph_dotted_line,
1885 graph_dotted_line);
1886
414e050c
NK		 1887 if (sched->show_state)
1888 printf(" %.5s", graph_dotted_line);
1889
49394a2a
DA		 1890 printf("\n");
1891}
1892
414e050c
NK		 1893static char task_state_char(struct thread *thread, int state)
1894{
1895 static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1896 unsigned bit = state ? ffs(state) : 0;
1897
1898 /* 'I' for idle */
1899 if (thread->tid == 0)
1900 return 'I';
1901
1902 return bit < sizeof(state_to_char) - 1 ? state_to_char[bit] : '?';
1903}
1904
fc1469f1 1905static void timehist_print_sample(struct perf_sched *sched,
292c4a8f 1906 struct perf_evsel *evsel,
fc1469f1 1907 struct perf_sample *sample,
6c973c90 1908 struct addr_location *al,
853b7407 1909 struct thread *thread,
414e050c 1910 u64 t, int state)
49394a2a
DA		 1911{
1912 struct thread_runtime *tr = thread__priv(thread);
292c4a8f
BG		 1913 const char *next_comm = perf_evsel__strval(evsel, sample, "next_comm");
1914 const u32 next_pid = perf_evsel__intval(evsel, sample, "next_pid");
a407b067 1915 u32 max_cpus = sched->max_cpu + 1;
49394a2a 1916 char tstr[64];
292c4a8f 1917 char nstr[30];
941bdea7 1918 u64 wait_time;
49394a2a 1919
853b7407 1920 timestamp__scnprintf_usec(t, tstr, sizeof(tstr));
49394a2a
DA		 1921 printf("%15s [%04d] ", tstr, sample->cpu);
1922
a407b067
DA		 1923 if (sched->show_cpu_visual) {
1924 u32 i;
1925 char c;
1926
1927 printf(" ");
1928 for (i = 0; i < max_cpus; ++i) {
1929 /* flag idle times with 'i'; others are sched events */
1930 if (i == sample->cpu)
1931 c = (thread->tid == 0) ? 'i' : 's';
1932 else
1933 c = ' ';
1934 printf("%c", c);
1935 }
1936 printf(" ");
1937 }
1938
49394a2a
DA		 1939 printf(" %-*s ", comm_width, timehist_get_commstr(thread));
1940
941bdea7
NK		 1941 wait_time = tr->dt_sleep + tr->dt_iowait + tr->dt_preempt;
1942 print_sched_time(wait_time, 6);
1943
49394a2a
DA		 1944 print_sched_time(tr->dt_delay, 6);
1945 print_sched_time(tr->dt_run, 6);
fc1469f1 1946
414e050c
NK		 1947 if (sched->show_state)
1948 printf(" %5c ", task_state_char(thread, state));
1949
292c4a8f
BG		 1950 if (sched->show_next) {
1951 snprintf(nstr, sizeof(nstr), "next: %s[%d]", next_comm, next_pid);
1952 printf(" %-*s", comm_width, nstr);
1953 }
1954
1955 if (sched->show_wakeups && !sched->show_next)
fc1469f1
DA		 1956 printf(" %-*s", comm_width, "");
1957
6c973c90
DA		 1958 if (thread->tid == 0)
1959 goto out;
1960
1961 if (sched->show_callchain)
1962 printf(" ");
1963
1964 sample__fprintf_sym(sample, al, 0,
1965 EVSEL__PRINT_SYM | EVSEL__PRINT_ONELINE |
2d9bbf6e
NK		 1966 EVSEL__PRINT_CALLCHAIN_ARROW |
1967 EVSEL__PRINT_SKIP_IGNORED,
6c973c90
DA		 1968 &callchain_cursor, stdout);
1969
1970out:
49394a2a
DA		 1971 printf("\n");
1972}
1973
1974/*
1975 * Explanation of delta-time stats:
1976 *
1977 * t = time of current schedule out event
1978 * tprev = time of previous sched out event
1979 * also time of schedule-in event for current task
1980 * last_time = time of last sched change event for current task
1981 * (i.e, time process was last scheduled out)
1982 * ready_to_run = time of wakeup for current task
1983 *
1984 * -----|------------|------------|------------|------
1985 * last ready tprev t
1986 * time to run
1987 *
1988 * |-------- dt_wait --------|
1989 * |- dt_delay -|-- dt_run --|
1990 *
1991 * dt_run = run time of current task
1992 * dt_wait = time between last schedule out event for task and tprev
1993 * represents time spent off the cpu
1994 * dt_delay = time between wakeup and schedule-in of task
1995 */
1996
1997static void timehist_update_runtime_stats(struct thread_runtime *r,
1998 u64 t, u64 tprev)
1999{
2000 r->dt_delay = 0;
941bdea7
NK		 2001 r->dt_sleep = 0;
2002 r->dt_iowait = 0;
2003 r->dt_preempt = 0;
49394a2a 2004 r->dt_run = 0;
941bdea7 2005
49394a2a
DA		 2006 if (tprev) {
2007 r->dt_run = t - tprev;
2008 if (r->ready_to_run) {
2009 if (r->ready_to_run > tprev)
2010 pr_debug("time travel: wakeup time for task > previous sched_switch event\n");
2011 else
2012 r->dt_delay = tprev - r->ready_to_run;
2013 }
2014
2015 if (r->last_time > tprev)
2016 pr_debug("time travel: last sched out time for task > previous sched_switch event\n");
941bdea7
NK		 2017 else if (r->last_time) {
2018 u64 dt_wait = tprev - r->last_time;
2019
2020 if (r->last_state == TASK_RUNNING)
2021 r->dt_preempt = dt_wait;
2022 else if (r->last_state == TASK_UNINTERRUPTIBLE)
2023 r->dt_iowait = dt_wait;
2024 else
2025 r->dt_sleep = dt_wait;
2026 }
49394a2a
DA		 2027 }
2028
2029 update_stats(&r->run_stats, r->dt_run);
587782c5
NK		 2030
2031 r->total_run_time += r->dt_run;
2032 r->total_delay_time += r->dt_delay;
2033 r->total_sleep_time += r->dt_sleep;
2034 r->total_iowait_time += r->dt_iowait;
2035 r->total_preempt_time += r->dt_preempt;
49394a2a
DA		 2036}
2037
96039c7c
NK		 2038static bool is_idle_sample(struct perf_sample *sample,
2039 struct perf_evsel *evsel)
49394a2a
DA		 2040{
2041 /* pid 0 == swapper == idle task */
96039c7c
NK		 2042 if (strcmp(perf_evsel__name(evsel), "sched:sched_switch") == 0)
2043 return perf_evsel__intval(evsel, sample, "prev_pid") == 0;
49394a2a 2044
96039c7c
NK		 2045 return sample->pid == 0;
2046}
2047
2048static void save_task_callchain(struct perf_sched *sched,
2049 struct perf_sample *sample,
2050 struct perf_evsel *evsel,
2051 struct machine *machine)
2052{
2053 struct callchain_cursor *cursor = &callchain_cursor;
2054 struct thread *thread;
6c973c90
DA		 2055
2056 /* want main thread for process - has maps */
2057 thread = machine__findnew_thread(machine, sample->pid, sample->pid);
2058 if (thread == NULL) {
2059 pr_debug("Failed to get thread for pid %d.\n", sample->pid);
96039c7c 2060 return;
6c973c90
DA		 2061 }
2062
2063 if (!symbol_conf.use_callchain || sample->callchain == NULL)
96039c7c 2064 return;
6c973c90
DA		 2065
2066 if (thread__resolve_callchain(thread, cursor, evsel, sample,
8388deb3 2067 NULL, NULL, sched->max_stack + 2) != 0) {
bb963e16 2068 if (verbose > 0)
62d94b00 2069 pr_err("Failed to resolve callchain. Skipping\n");
6c973c90 2070
96039c7c 2071 return;
6c973c90 2072 }
cdeb01bf 2073
6c973c90 2074 callchain_cursor_commit(cursor);
cdeb01bf
NK		 2075
2076 while (true) {
2077 struct callchain_cursor_node *node;
2078 struct symbol *sym;
2079
2080 node = callchain_cursor_current(cursor);
2081 if (node == NULL)
2082 break;
2083
2084 sym = node->sym;
a7c3899c 2085 if (sym) {
cdeb01bf
NK		 2086 if (!strcmp(sym->name, "schedule") ||
2087 !strcmp(sym->name, "__schedule") ||
2088 !strcmp(sym->name, "preempt_schedule"))
2089 sym->ignore = 1;
2090 }
2091
2092 callchain_cursor_advance(cursor);
2093 }
49394a2a
DA		 2094}
2095
3bc2fa9c
NK		 2096static int init_idle_thread(struct thread *thread)
2097{
2098 struct idle_thread_runtime *itr;
2099
2100 thread__set_comm(thread, idle_comm, 0);
2101
2102 itr = zalloc(sizeof(*itr));
2103 if (itr == NULL)
2104 return -ENOMEM;
2105
2106 init_stats(&itr->tr.run_stats);
2107 callchain_init(&itr->callchain);
2108 callchain_cursor_reset(&itr->cursor);
2109 thread__set_priv(thread, itr);
2110
2111 return 0;
2112}
2113
49394a2a
DA		 2114/*
2115 * Track idle stats per cpu by maintaining a local thread
2116 * struct for the idle task on each cpu.
2117 */
2118static int init_idle_threads(int ncpu)
2119{
3bc2fa9c 2120 int i, ret;
49394a2a
DA		 2121
2122 idle_threads = zalloc(ncpu * sizeof(struct thread *));
2123 if (!idle_threads)
2124 return -ENOMEM;
2125
b336352b 2126 idle_max_cpu = ncpu;
49394a2a
DA		 2127
2128 /* allocate the actual thread struct if needed */
2129 for (i = 0; i < ncpu; ++i) {
2130 idle_threads[i] = thread__new(0, 0);
2131 if (idle_threads[i] == NULL)
2132 return -ENOMEM;
2133
3bc2fa9c
NK		 2134 ret = init_idle_thread(idle_threads[i]);
2135 if (ret < 0)
2136 return ret;
49394a2a
DA		 2137 }
2138
2139 return 0;
2140}
2141
2142static void free_idle_threads(void)
2143{
2144 int i;
2145
2146 if (idle_threads == NULL)
2147 return;
2148
b336352b 2149 for (i = 0; i < idle_max_cpu; ++i) {
49394a2a
DA		 2150 if ((idle_threads[i]))
2151 thread__delete(idle_threads[i]);
2152 }
2153
2154 free(idle_threads);
2155}
2156
2157static struct thread *get_idle_thread(int cpu)
2158{
2159 /*
2160 * expand/allocate array of pointers to local thread
2161 * structs if needed
2162 */
2163 if ((cpu >= idle_max_cpu) || (idle_threads == NULL)) {
2164 int i, j = __roundup_pow_of_two(cpu+1);
2165 void *p;
2166
2167 p = realloc(idle_threads, j * sizeof(struct thread *));
2168 if (!p)
2169 return NULL;
2170
2171 idle_threads = (struct thread **) p;
b336352b 2172 for (i = idle_max_cpu; i < j; ++i)
49394a2a
DA		 2173 idle_threads[i] = NULL;
2174
2175 idle_max_cpu = j;
2176 }
2177
2178 /* allocate a new thread struct if needed */
2179 if (idle_threads[cpu] == NULL) {
2180 idle_threads[cpu] = thread__new(0, 0);
2181
2182 if (idle_threads[cpu]) {
3bc2fa9c
NK		 2183 if (init_idle_thread(idle_threads[cpu]) < 0)
2184 return NULL;
49394a2a
DA		 2185 }
2186 }
2187
2188 return idle_threads[cpu];
2189}
2190
699b5b92
NK		 2191static void save_idle_callchain(struct idle_thread_runtime *itr,
2192 struct perf_sample *sample)
2193{
2194 if (!symbol_conf.use_callchain || sample->callchain == NULL)
2195 return;
2196
2197 callchain_cursor__copy(&itr->cursor, &callchain_cursor);
2198}
2199
49394a2a
DA		 2200/*
2201 * handle runtime stats saved per thread
2202 */
2203static struct thread_runtime *thread__init_runtime(struct thread *thread)
2204{
2205 struct thread_runtime *r;
2206
2207 r = zalloc(sizeof(struct thread_runtime));
2208 if (!r)
2209 return NULL;
2210
2211 init_stats(&r->run_stats);
2212 thread__set_priv(thread, r);
2213
2214 return r;
2215}
2216
2217static struct thread_runtime *thread__get_runtime(struct thread *thread)
2218{
2219 struct thread_runtime *tr;
2220
2221 tr = thread__priv(thread);
2222 if (tr == NULL) {
2223 tr = thread__init_runtime(thread);
2224 if (tr == NULL)
2225 pr_debug("Failed to malloc memory for runtime data.\n");
2226 }
2227
2228 return tr;
2229}
2230
6c973c90
DA		 2231static struct thread *timehist_get_thread(struct perf_sched *sched,
2232 struct perf_sample *sample,
49394a2a
DA		 2233 struct machine *machine,
2234 struct perf_evsel *evsel)
2235{
2236 struct thread *thread;
2237
96039c7c 2238 if (is_idle_sample(sample, evsel)) {
49394a2a
DA		 2239 thread = get_idle_thread(sample->cpu);
2240 if (thread == NULL)
2241 pr_err("Failed to get idle thread for cpu %d.\n", sample->cpu);
2242
2243 } else {
5d92d96a
NK		 2244 /* there were samples with tid 0 but non-zero pid */
2245 thread = machine__findnew_thread(machine, sample->pid,
2246 sample->tid ?: sample->pid);
49394a2a
DA		 2247 if (thread == NULL) {
2248 pr_debug("Failed to get thread for tid %d. skipping sample.\n",
2249 sample->tid);
2250 }
96039c7c
NK		 2251
2252 save_task_callchain(sched, sample, evsel, machine);
699b5b92
NK		 2253 if (sched->idle_hist) {
2254 struct thread *idle;
2255 struct idle_thread_runtime *itr;
2256
2257 idle = get_idle_thread(sample->cpu);
2258 if (idle == NULL) {
2259 pr_err("Failed to get idle thread for cpu %d.\n", sample->cpu);
2260 return NULL;
2261 }
2262
2263 itr = thread__priv(idle);
2264 if (itr == NULL)
2265 return NULL;
2266
2267 itr->last_thread = thread;
2268
2269 /* copy task callchain when entering to idle */
2270 if (perf_evsel__intval(evsel, sample, "next_pid") == 0)
2271 save_idle_callchain(itr, sample);
2272 }
49394a2a
DA		 2273 }
2274
2275 return thread;
2276}
2277
52df138c 2278static bool timehist_skip_sample(struct perf_sched *sched,
a4b2b6f5
NK		 2279 struct thread *thread,
2280 struct perf_evsel *evsel,
2281 struct perf_sample *sample)
49394a2a
DA		 2282{
2283 bool rc = false;
2284
52df138c 2285 if (thread__is_filtered(thread)) {
49394a2a 2286 rc = true;
52df138c
DA		 2287 sched->skipped_samples++;
2288 }
49394a2a 2289
a4b2b6f5
NK		 2290 if (sched->idle_hist) {
2291 if (strcmp(perf_evsel__name(evsel), "sched:sched_switch"))
2292 rc = true;
2293 else if (perf_evsel__intval(evsel, sample, "prev_pid") != 0 &&
2294 perf_evsel__intval(evsel, sample, "next_pid") != 0)
2295 rc = true;
2296 }
2297
49394a2a
DA		 2298 return rc;
2299}
2300
fc1469f1 2301static void timehist_print_wakeup_event(struct perf_sched *sched,
a4b2b6f5 2302 struct perf_evsel *evsel,
fc1469f1
DA		 2303 struct perf_sample *sample,
2304 struct machine *machine,
2305 struct thread *awakened)
2306{
2307 struct thread *thread;
2308 char tstr[64];
2309
2310 thread = machine__findnew_thread(machine, sample->pid, sample->tid);
2311 if (thread == NULL)
2312 return;
2313
2314 /* show wakeup unless both awakee and awaker are filtered */
a4b2b6f5
NK		 2315 if (timehist_skip_sample(sched, thread, evsel, sample) &&
2316 timehist_skip_sample(sched, awakened, evsel, sample)) {
fc1469f1
DA		 2317 return;
2318 }
2319
2320 timestamp__scnprintf_usec(sample->time, tstr, sizeof(tstr));
2321 printf("%15s [%04d] ", tstr, sample->cpu);
a407b067
DA		 2322 if (sched->show_cpu_visual)
2323 printf(" %*s ", sched->max_cpu + 1, "");
fc1469f1
DA		 2324
2325 printf(" %-*s ", comm_width, timehist_get_commstr(thread));
2326
2327 /* dt spacer */
2328 printf(" %9s %9s %9s ", "", "", "");
2329
2330 printf("awakened: %s", timehist_get_commstr(awakened));
2331
2332 printf("\n");
2333}
2334
2335static int timehist_sched_wakeup_event(struct perf_tool *tool,
49394a2a
DA		 2336 union perf_event *event __maybe_unused,
2337 struct perf_evsel *evsel,
2338 struct perf_sample *sample,
2339 struct machine *machine)
2340{
fc1469f1 2341 struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
49394a2a
DA		 2342 struct thread *thread;
2343 struct thread_runtime *tr = NULL;
2344 /* want pid of awakened task not pid in sample */
2345 const u32 pid = perf_evsel__intval(evsel, sample, "pid");
2346
2347 thread = machine__findnew_thread(machine, 0, pid);
2348 if (thread == NULL)
2349 return -1;
2350
2351 tr = thread__get_runtime(thread);
2352 if (tr == NULL)
2353 return -1;
2354
2355 if (tr->ready_to_run == 0)
2356 tr->ready_to_run = sample->time;
2357
fc1469f1 2358 /* show wakeups if requested */
853b7407
DA		 2359 if (sched->show_wakeups &&
2360 !perf_time__skip_sample(&sched->ptime, sample->time))
a4b2b6f5 2361 timehist_print_wakeup_event(sched, evsel, sample, machine, thread);
fc1469f1 2362
49394a2a
DA		 2363 return 0;
2364}
2365
350f54fa
DA		 2366static void timehist_print_migration_event(struct perf_sched *sched,
2367 struct perf_evsel *evsel,
2368 struct perf_sample *sample,
2369 struct machine *machine,
2370 struct thread *migrated)
2371{
2372 struct thread *thread;
2373 char tstr[64];
2374 u32 max_cpus = sched->max_cpu + 1;
2375 u32 ocpu, dcpu;
2376
2377 if (sched->summary_only)
2378 return;
2379
2380 max_cpus = sched->max_cpu + 1;
2381 ocpu = perf_evsel__intval(evsel, sample, "orig_cpu");
2382 dcpu = perf_evsel__intval(evsel, sample, "dest_cpu");
2383
2384 thread = machine__findnew_thread(machine, sample->pid, sample->tid);
2385 if (thread == NULL)
2386 return;
2387
a4b2b6f5
NK		 2388 if (timehist_skip_sample(sched, thread, evsel, sample) &&
2389 timehist_skip_sample(sched, migrated, evsel, sample)) {
350f54fa
DA		 2390 return;
2391 }
2392
2393 timestamp__scnprintf_usec(sample->time, tstr, sizeof(tstr));
2394 printf("%15s [%04d] ", tstr, sample->cpu);
2395
2396 if (sched->show_cpu_visual) {
2397 u32 i;
2398 char c;
2399
2400 printf(" ");
2401 for (i = 0; i < max_cpus; ++i) {
2402 c = (i == sample->cpu) ? 'm' : ' ';
2403 printf("%c", c);
2404 }
2405 printf(" ");
2406 }
2407
2408 printf(" %-*s ", comm_width, timehist_get_commstr(thread));
2409
2410 /* dt spacer */
2411 printf(" %9s %9s %9s ", "", "", "");
2412
2413 printf("migrated: %s", timehist_get_commstr(migrated));
2414 printf(" cpu %d => %d", ocpu, dcpu);
2415
2416 printf("\n");
2417}
2418
2419static int timehist_migrate_task_event(struct perf_tool *tool,
2420 union perf_event *event __maybe_unused,
2421 struct perf_evsel *evsel,
2422 struct perf_sample *sample,
2423 struct machine *machine)
2424{
2425 struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
2426 struct thread *thread;
2427 struct thread_runtime *tr = NULL;
2428 /* want pid of migrated task not pid in sample */
2429 const u32 pid = perf_evsel__intval(evsel, sample, "pid");
2430
2431 thread = machine__findnew_thread(machine, 0, pid);
2432 if (thread == NULL)
2433 return -1;
2434
2435 tr = thread__get_runtime(thread);
2436 if (tr == NULL)
2437 return -1;
2438
2439 tr->migrations++;
2440
2441 /* show migrations if requested */
2442 timehist_print_migration_event(sched, evsel, sample, machine, thread);
2443
2444 return 0;
2445}
2446
52df138c 2447static int timehist_sched_change_event(struct perf_tool *tool,
49394a2a
DA		 2448 union perf_event *event,
2449 struct perf_evsel *evsel,
2450 struct perf_sample *sample,
2451 struct machine *machine)
2452{
fc1469f1 2453 struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
853b7407 2454 struct perf_time_interval *ptime = &sched->ptime;
49394a2a
DA		 2455 struct addr_location al;
2456 struct thread *thread;
2457 struct thread_runtime *tr = NULL;
853b7407 2458 u64 tprev, t = sample->time;
49394a2a 2459 int rc = 0;
414e050c
NK		 2460 int state = perf_evsel__intval(evsel, sample, "prev_state");
2461
49394a2a
DA		 2462
2463 if (machine__resolve(machine, &al, sample) < 0) {
2464 pr_err("problem processing %d event. skipping it\n",
2465 event->header.type);
2466 rc = -1;
2467 goto out;
2468 }
2469
6c973c90 2470 thread = timehist_get_thread(sched, sample, machine, evsel);
49394a2a
DA		 2471 if (thread == NULL) {
2472 rc = -1;
2473 goto out;
2474 }
2475
a4b2b6f5 2476 if (timehist_skip_sample(sched, thread, evsel, sample))
49394a2a
DA		 2477 goto out;
2478
2479 tr = thread__get_runtime(thread);
2480 if (tr == NULL) {
2481 rc = -1;
2482 goto out;
2483 }
2484
2485 tprev = perf_evsel__get_time(evsel, sample->cpu);
2486
853b7407
DA		 2487 /*
2488 * If start time given:
2489 * - sample time is under window user cares about - skip sample
2490 * - tprev is under window user cares about - reset to start of window
2491 */
2492 if (ptime->start && ptime->start > t)
2493 goto out;
2494
bdd75729 2495 if (tprev && ptime->start > tprev)
853b7407
DA		 2496 tprev = ptime->start;
2497
2498 /*
2499 * If end time given:
2500 * - previous sched event is out of window - we are done
2501 * - sample time is beyond window user cares about - reset it
2502 * to close out stats for time window interest
2503 */
2504 if (ptime->end) {
2505 if (tprev > ptime->end)
2506 goto out;
2507
2508 if (t > ptime->end)
2509 t = ptime->end;
2510 }
2511
07235f84
NK		 2512 if (!sched->idle_hist || thread->tid == 0) {
2513 timehist_update_runtime_stats(tr, t, tprev);
2514
2515 if (sched->idle_hist) {
2516 struct idle_thread_runtime *itr = (void *)tr;
2517 struct thread_runtime *last_tr;
2518
2519 BUG_ON(thread->tid != 0);
2520
2521 if (itr->last_thread == NULL)
2522 goto out;
2523
2524 /* add current idle time as last thread's runtime */
2525 last_tr = thread__get_runtime(itr->last_thread);
2526 if (last_tr == NULL)
2527 goto out;
2528
2529 timehist_update_runtime_stats(last_tr, t, tprev);
2530 /*
2531 * remove delta time of last thread as it's not updated
2532 * and otherwise it will show an invalid value next
2533 * time. we only care total run time and run stat.
2534 */
2535 last_tr->dt_run = 0;
07235f84 2536 last_tr->dt_delay = 0;
941bdea7
NK		 2537 last_tr->dt_sleep = 0;
2538 last_tr->dt_iowait = 0;
2539 last_tr->dt_preempt = 0;
07235f84 2540
ba957ebb
NK		 2541 if (itr->cursor.nr)
2542 callchain_append(&itr->callchain, &itr->cursor, t - tprev);
2543
07235f84
NK		 2544 itr->last_thread = NULL;
2545 }
2546 }
853b7407 2547
52df138c 2548 if (!sched->summary_only)
292c4a8f 2549 timehist_print_sample(sched, evsel, sample, &al, thread, t, state);
49394a2a
DA		 2550
2551out:
9396c9cb
NK		 2552 if (sched->hist_time.start == 0 && t >= ptime->start)
2553 sched->hist_time.start = t;
2554 if (ptime->end == 0 || t <= ptime->end)
2555 sched->hist_time.end = t;
2556
49394a2a
DA		 2557 if (tr) {
2558 /* time of this sched_switch event becomes last time task seen */
2559 tr->last_time = sample->time;
2560
941bdea7 2561 /* last state is used to determine where to account wait time */
414e050c 2562 tr->last_state = state;
941bdea7 2563
49394a2a
DA		 2564 /* sched out event for task so reset ready to run time */
2565 tr->ready_to_run = 0;
2566 }
2567
2568 perf_evsel__save_time(evsel, sample->time, sample->cpu);
2569
2570 return rc;
2571}
2572
2573static int timehist_sched_switch_event(struct perf_tool *tool,
2574 union perf_event *event,
2575 struct perf_evsel *evsel,
2576 struct perf_sample *sample,
2577 struct machine *machine __maybe_unused)
2578{
2579 return timehist_sched_change_event(tool, event, evsel, sample, machine);
2580}
2581
2582static int process_lost(struct perf_tool *tool __maybe_unused,
2583 union perf_event *event,
2584 struct perf_sample *sample,
2585 struct machine *machine __maybe_unused)
2586{
2587 char tstr[64];
2588
2589 timestamp__scnprintf_usec(sample->time, tstr, sizeof(tstr));
2590 printf("%15s ", tstr);
2591 printf("lost %" PRIu64 " events on cpu %d\n", event->lost.lost, sample->cpu);
2592
2593 return 0;
2594}
2595
2596
52df138c
DA		 2597static void print_thread_runtime(struct thread *t,
2598 struct thread_runtime *r)
2599{
2600 double mean = avg_stats(&r->run_stats);
2601 float stddev;
2602
2603 printf("%*s %5d %9" PRIu64 " ",
2604 comm_width, timehist_get_commstr(t), t->ppid,
2605 (u64) r->run_stats.n);
2606
2607 print_sched_time(r->total_run_time, 8);
2608 stddev = rel_stddev_stats(stddev_stats(&r->run_stats), mean);
2609 print_sched_time(r->run_stats.min, 6);
2610 printf(" ");
2611 print_sched_time((u64) mean, 6);
2612 printf(" ");
2613 print_sched_time(r->run_stats.max, 6);
2614 printf(" ");
2615 printf("%5.2f", stddev);
350f54fa 2616 printf(" %5" PRIu64, r->migrations);
52df138c
DA		 2617 printf("\n");
2618}
2619
587782c5
NK		 2620static void print_thread_waittime(struct thread *t,
2621 struct thread_runtime *r)
2622{
2623 printf("%*s %5d %9" PRIu64 " ",
2624 comm_width, timehist_get_commstr(t), t->ppid,
2625 (u64) r->run_stats.n);
2626
2627 print_sched_time(r->total_run_time, 8);
2628 print_sched_time(r->total_sleep_time, 6);
2629 printf(" ");
2630 print_sched_time(r->total_iowait_time, 6);
2631 printf(" ");
2632 print_sched_time(r->total_preempt_time, 6);
2633 printf(" ");
2634 print_sched_time(r->total_delay_time, 6);
2635 printf("\n");
2636}
2637
52df138c 2638struct total_run_stats {
587782c5 2639 struct perf_sched *sched;
52df138c
DA		 2640 u64 sched_count;
2641 u64 task_count;
2642 u64 total_run_time;
2643};
2644
2645static int __show_thread_runtime(struct thread *t, void *priv)
2646{
2647 struct total_run_stats *stats = priv;
2648 struct thread_runtime *r;
2649
2650 if (thread__is_filtered(t))
2651 return 0;
2652
2653 r = thread__priv(t);
2654 if (r && r->run_stats.n) {
2655 stats->task_count++;
2656 stats->sched_count += r->run_stats.n;
2657 stats->total_run_time += r->total_run_time;
587782c5
NK		 2658
2659 if (stats->sched->show_state)
2660 print_thread_waittime(t, r);
2661 else
2662 print_thread_runtime(t, r);
52df138c
DA		 2663 }
2664
2665 return 0;
2666}
2667
2668static int show_thread_runtime(struct thread *t, void *priv)
2669{
2670 if (t->dead)
2671 return 0;
2672
2673 return __show_thread_runtime(t, priv);
2674}
2675
2676static int show_deadthread_runtime(struct thread *t, void *priv)
2677{
2678 if (!t->dead)
2679 return 0;
2680
2681 return __show_thread_runtime(t, priv);
2682}
2683
ba957ebb
NK		 2684static size_t callchain__fprintf_folded(FILE *fp, struct callchain_node *node)
2685{
2686 const char *sep = " <- ";
2687 struct callchain_list *chain;
2688 size_t ret = 0;
2689 char bf[1024];
2690 bool first;
2691
2692 if (node == NULL)
2693 return 0;
2694
2695 ret = callchain__fprintf_folded(fp, node->parent);
2696 first = (ret == 0);
2697
2698 list_for_each_entry(chain, &node->val, list) {
2699 if (chain->ip >= PERF_CONTEXT_MAX)
2700 continue;
2701 if (chain->ms.sym && chain->ms.sym->ignore)
2702 continue;
2703 ret += fprintf(fp, "%s%s", first ? "" : sep,
2704 callchain_list__sym_name(chain, bf, sizeof(bf),
2705 false));
2706 first = false;
2707 }
2708
2709 return ret;
2710}
2711
2712static size_t timehist_print_idlehist_callchain(struct rb_root *root)
2713{
2714 size_t ret = 0;
2715 FILE *fp = stdout;
2716 struct callchain_node *chain;
2717 struct rb_node *rb_node = rb_first(root);
2718
2719 printf(" %16s %8s %s\n", "Idle time (msec)", "Count", "Callchains");
2720 printf(" %.16s %.8s %.50s\n", graph_dotted_line, graph_dotted_line,
2721 graph_dotted_line);
2722
2723 while (rb_node) {
2724 chain = rb_entry(rb_node, struct callchain_node, rb_node);
2725 rb_node = rb_next(rb_node);
2726
2727 ret += fprintf(fp, " ");
2728 print_sched_time(chain->hit, 12);
2729 ret += 16; /* print_sched_time returns 2nd arg + 4 */
2730 ret += fprintf(fp, " %8d ", chain->count);
2731 ret += callchain__fprintf_folded(fp, chain);
2732 ret += fprintf(fp, "\n");
2733 }
2734
2735 return ret;
2736}
2737
52df138c
DA		 2738static void timehist_print_summary(struct perf_sched *sched,
2739 struct perf_session *session)
2740{
2741 struct machine *m = &session->machines.host;
2742 struct total_run_stats totals;
2743 u64 task_count;
2744 struct thread *t;
2745 struct thread_runtime *r;
2746 int i;
9396c9cb 2747 u64 hist_time = sched->hist_time.end - sched->hist_time.start;
52df138c
DA		 2748
2749 memset(&totals, 0, sizeof(totals));
587782c5 2750 totals.sched = sched;
52df138c 2751
07235f84
NK		 2752 if (sched->idle_hist) {
2753 printf("\nIdle-time summary\n");
2754 printf("%*s parent sched-out ", comm_width, "comm");
2755 printf(" idle-time min-idle avg-idle max-idle stddev migrations\n");
587782c5
NK		 2756 } else if (sched->show_state) {
2757 printf("\nWait-time summary\n");
2758 printf("%*s parent sched-in ", comm_width, "comm");
2759 printf(" run-time sleep iowait preempt delay\n");
07235f84
NK		 2760 } else {
2761 printf("\nRuntime summary\n");
2762 printf("%*s parent sched-in ", comm_width, "comm");
2763 printf(" run-time min-run avg-run max-run stddev migrations\n");
2764 }
52df138c 2765 printf("%*s (count) ", comm_width, "");
587782c5
NK		 2766 printf(" (msec) (msec) (msec) (msec) %s\n",
2767 sched->show_state ? "(msec)" : "%");
350f54fa 2768 printf("%.117s\n", graph_dotted_line);
52df138c
DA		 2769
2770 machine__for_each_thread(m, show_thread_runtime, &totals);
2771 task_count = totals.task_count;
2772 if (!task_count)
2773 printf("<no still running tasks>\n");
2774
2775 printf("\nTerminated tasks:\n");
2776 machine__for_each_thread(m, show_deadthread_runtime, &totals);
2777 if (task_count == totals.task_count)
2778 printf("<no terminated tasks>\n");
2779
2780 /* CPU idle stats not tracked when samples were skipped */
07235f84 2781 if (sched->skipped_samples && !sched->idle_hist)
52df138c
DA		 2782 return;
2783
2784 printf("\nIdle stats:\n");
b336352b 2785 for (i = 0; i < idle_max_cpu; ++i) {
52df138c
DA		 2786 t = idle_threads[i];
2787 if (!t)
2788 continue;
2789
2790 r = thread__priv(t);
2791 if (r && r->run_stats.n) {
2792 totals.sched_count += r->run_stats.n;
2793 printf(" CPU %2d idle for ", i);
2794 print_sched_time(r->total_run_time, 6);
9396c9cb 2795 printf(" msec (%6.2f%%)\n", 100.0 * r->total_run_time / hist_time);
52df138c
DA		 2796 } else
2797 printf(" CPU %2d idle entire time window\n", i);
2798 }
2799
ba957ebb
NK		 2800 if (sched->idle_hist && symbol_conf.use_callchain) {
2801 callchain_param.mode = CHAIN_FOLDED;
2802 callchain_param.value = CCVAL_PERIOD;
2803
2804 callchain_register_param(&callchain_param);
2805
2806 printf("\nIdle stats by callchain:\n");
2807 for (i = 0; i < idle_max_cpu; ++i) {
2808 struct idle_thread_runtime *itr;
2809
2810 t = idle_threads[i];
2811 if (!t)
2812 continue;
2813
2814 itr = thread__priv(t);
2815 if (itr == NULL)
2816 continue;
2817
2818 callchain_param.sort(&itr->sorted_root, &itr->callchain,
2819 0, &callchain_param);
2820
2821 printf(" CPU %2d:", i);
2822 print_sched_time(itr->tr.total_run_time, 6);
2823 printf(" msec\n");
2824 timehist_print_idlehist_callchain(&itr->sorted_root);
2825 printf("\n");
2826 }
2827 }
2828
52df138c
DA		 2829 printf("\n"
2830 " Total number of unique tasks: %" PRIu64 "\n"
9396c9cb 2831 "Total number of context switches: %" PRIu64 "\n",
52df138c
DA		 2832 totals.task_count, totals.sched_count);
2833
9396c9cb 2834 printf(" Total run time (msec): ");
52df138c
DA		 2835 print_sched_time(totals.total_run_time, 2);
2836 printf("\n");
9396c9cb
NK		 2837
2838 printf(" Total scheduling time (msec): ");
2839 print_sched_time(hist_time, 2);
2840 printf(" (x %d)\n", sched->max_cpu);
52df138c
DA		 2841}
2842
49394a2a
DA		 2843typedef int (*sched_handler)(struct perf_tool *tool,
2844 union perf_event *event,
2845 struct perf_evsel *evsel,
2846 struct perf_sample *sample,
2847 struct machine *machine);
2848
2849static int perf_timehist__process_sample(struct perf_tool *tool,
2850 union perf_event *event,
2851 struct perf_sample *sample,
2852 struct perf_evsel *evsel,
2853 struct machine *machine)
2854{
2855 struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
2856 int err = 0;
2857 int this_cpu = sample->cpu;
2858
2859 if (this_cpu > sched->max_cpu)
2860 sched->max_cpu = this_cpu;
2861
2862 if (evsel->handler != NULL) {
2863 sched_handler f = evsel->handler;
2864
2865 err = f(tool, event, evsel, sample, machine);
2866 }
2867
2868 return err;
2869}
2870
6c973c90
DA		 2871static int timehist_check_attr(struct perf_sched *sched,
2872 struct perf_evlist *evlist)
2873{
2874 struct perf_evsel *evsel;
2875 struct evsel_runtime *er;
2876
2877 list_for_each_entry(evsel, &evlist->entries, node) {
2878 er = perf_evsel__get_runtime(evsel);
2879 if (er == NULL) {
2880 pr_err("Failed to allocate memory for evsel runtime data\n");
2881 return -1;
2882 }
2883
2884 if (sched->show_callchain &&
2885 !(evsel->attr.sample_type & PERF_SAMPLE_CALLCHAIN)) {
2886 pr_info("Samples do not have callchains.\n");
2887 sched->show_callchain = 0;
2888 symbol_conf.use_callchain = 0;
2889 }
2890 }
2891
2892 return 0;
2893}
2894
49394a2a
DA		 2895static int perf_sched__timehist(struct perf_sched *sched)
2896{
2897 const struct perf_evsel_str_handler handlers[] = {
2898 { "sched:sched_switch", timehist_sched_switch_event, },
2899 { "sched:sched_wakeup", timehist_sched_wakeup_event, },
2900 { "sched:sched_wakeup_new", timehist_sched_wakeup_event, },
2901 };
350f54fa
DA		 2902 const struct perf_evsel_str_handler migrate_handlers[] = {
2903 { "sched:sched_migrate_task", timehist_migrate_task_event, },
2904 };
49394a2a
DA		 2905 struct perf_data_file file = {
2906 .path = input_name,
2907 .mode = PERF_DATA_MODE_READ,
6fa94258 2908 .force = sched->force,
49394a2a
DA		 2909 };
2910
2911 struct perf_session *session;
52df138c 2912 struct perf_evlist *evlist;
49394a2a
DA		 2913 int err = -1;
2914
2915 /*
2916 * event handlers for timehist option
2917 */
2918 sched->tool.sample = perf_timehist__process_sample;
2919 sched->tool.mmap = perf_event__process_mmap;
2920 sched->tool.comm = perf_event__process_comm;
2921 sched->tool.exit = perf_event__process_exit;
2922 sched->tool.fork = perf_event__process_fork;
2923 sched->tool.lost = process_lost;
2924 sched->tool.attr = perf_event__process_attr;
2925 sched->tool.tracing_data = perf_event__process_tracing_data;
2926 sched->tool.build_id = perf_event__process_build_id;
2927
2928 sched->tool.ordered_events = true;
2929 sched->tool.ordering_requires_timestamps = true;
2930
6c973c90
DA		 2931 symbol_conf.use_callchain = sched->show_callchain;
2932
49394a2a
DA		 2933 session = perf_session__new(&file, false, &sched->tool);
2934 if (session == NULL)
2935 return -ENOMEM;
2936
52df138c
DA		 2937 evlist = session->evlist;
2938
49394a2a
DA		 2939 symbol__init(&session->header.env);
2940
853b7407
DA		 2941 if (perf_time__parse_str(&sched->ptime, sched->time_str) != 0) {
2942 pr_err("Invalid time string\n");
2943 return -EINVAL;
2944 }
2945
6c973c90
DA		 2946 if (timehist_check_attr(sched, evlist) != 0)
2947 goto out;
2948
49394a2a
DA		 2949 setup_pager();
2950
2951 /* setup per-evsel handlers */
2952 if (perf_session__set_tracepoints_handlers(session, handlers))
2953 goto out;
2954
f45bf8d3
DA		 2955 /* sched_switch event at a minimum needs to exist */
2956 if (!perf_evlist__find_tracepoint_by_name(session->evlist,
2957 "sched:sched_switch")) {
2958 pr_err("No sched_switch events found. Have you run 'perf sched record'?\n");
49394a2a 2959 goto out;
f45bf8d3 2960 }
49394a2a 2961
350f54fa
DA		 2962 if (sched->show_migrations &&
2963 perf_session__set_tracepoints_handlers(session, migrate_handlers))
2964 goto out;
2965
49394a2a
DA		 2966 /* pre-allocate struct for per-CPU idle stats */
2967 sched->max_cpu = session->header.env.nr_cpus_online;
2968 if (sched->max_cpu == 0)
2969 sched->max_cpu = 4;
2970 if (init_idle_threads(sched->max_cpu))
2971 goto out;
2972
52df138c
DA		 2973 /* summary_only implies summary option, but don't overwrite summary if set */
2974 if (sched->summary_only)
2975 sched->summary = sched->summary_only;
2976
2977 if (!sched->summary_only)
a407b067 2978 timehist_header(sched);
49394a2a
DA		 2979
2980 err = perf_session__process_events(session);
2981 if (err) {
2982 pr_err("Failed to process events, error %d", err);
2983 goto out;
2984 }
2985
52df138c
DA		 2986 sched->nr_events = evlist->stats.nr_events[0];
2987 sched->nr_lost_events = evlist->stats.total_lost;
2988 sched->nr_lost_chunks = evlist->stats.nr_events[PERF_RECORD_LOST];
2989
2990 if (sched->summary)
2991 timehist_print_summary(sched, session);
2992
49394a2a
DA		 2993out:
2994 free_idle_threads();
2995 perf_session__delete(session);
2996
2997 return err;
2998}
2999
3000
0e9b07e5 3001static void print_bad_events(struct perf_sched *sched)
0ec04e16 3002{
0e9b07e5 3003 if (sched->nr_unordered_timestamps && sched->nr_timestamps) {
0ec04e16 3004 printf(" INFO: %.3f%% unordered timestamps (%ld out of %ld)\n",
0e9b07e5
ACM		 3005 (double)sched->nr_unordered_timestamps/(double)sched->nr_timestamps*100.0,
3006 sched->nr_unordered_timestamps, sched->nr_timestamps);
0ec04e16 3007 }
0e9b07e5 3008 if (sched->nr_lost_events && sched->nr_events) {
0ec04e16 3009 printf(" INFO: %.3f%% lost events (%ld out of %ld, in %ld chunks)\n",
0e9b07e5
ACM		 3010 (double)sched->nr_lost_events/(double)sched->nr_events * 100.0,
3011 sched->nr_lost_events, sched->nr_events, sched->nr_lost_chunks);
0ec04e16 3012 }
0e9b07e5 3013 if (sched->nr_context_switch_bugs && sched->nr_timestamps) {
0ec04e16 3014 printf(" INFO: %.3f%% context switch bugs (%ld out of %ld)",
0e9b07e5
ACM		 3015 (double)sched->nr_context_switch_bugs/(double)sched->nr_timestamps*100.0,
3016 sched->nr_context_switch_bugs, sched->nr_timestamps);
3017 if (sched->nr_lost_events)
0ec04e16
IM		 3018 printf(" (due to lost events?)");
3019 printf("\n");
3020 }
3021}
3022
2f80dd44
JB		 3023static void __merge_work_atoms(struct rb_root *root, struct work_atoms *data)
3024{
3025 struct rb_node **new = &(root->rb_node), *parent = NULL;
3026 struct work_atoms *this;
3027 const char *comm = thread__comm_str(data->thread), *this_comm;
3028
3029 while (*new) {
3030 int cmp;
3031
3032 this = container_of(*new, struct work_atoms, node);
3033 parent = *new;
3034
3035 this_comm = thread__comm_str(this->thread);
3036 cmp = strcmp(comm, this_comm);
3037 if (cmp > 0) {
3038 new = &((*new)->rb_left);
3039 } else if (cmp < 0) {
3040 new = &((*new)->rb_right);
3041 } else {
3042 this->num_merged++;
3043 this->total_runtime += data->total_runtime;
3044 this->nb_atoms += data->nb_atoms;
3045 this->total_lat += data->total_lat;
3046 list_splice(&data->work_list, &this->work_list);
3047 if (this->max_lat < data->max_lat) {
3048 this->max_lat = data->max_lat;
3049 this->max_lat_at = data->max_lat_at;
3050 }
3051 zfree(&data);
3052 return;
3053 }
3054 }
3055
3056 data->num_merged++;
3057 rb_link_node(&data->node, parent, new);
3058 rb_insert_color(&data->node, root);
3059}
3060
3061static void perf_sched__merge_lat(struct perf_sched *sched)
3062{
3063 struct work_atoms *data;
3064 struct rb_node *node;
3065
3066 if (sched->skip_merge)
3067 return;
3068
3069 while ((node = rb_first(&sched->atom_root))) {
3070 rb_erase(node, &sched->atom_root);
3071 data = rb_entry(node, struct work_atoms, node);
3072 __merge_work_atoms(&sched->merged_atom_root, data);
3073 }
3074}
3075
0e9b07e5 3076static int perf_sched__lat(struct perf_sched *sched)
0ec04e16
IM		 3077{
3078 struct rb_node *next;
3079
3080 setup_pager();
ad9def7c 3081
ae536acf 3082 if (perf_sched__read_events(sched))
a116e05d 3083 return -1;
ad9def7c 3084
2f80dd44 3085 perf_sched__merge_lat(sched);
0e9b07e5 3086 perf_sched__sort_lat(sched);
0ec04e16 3087
80790e0b
RR		 3088 printf("\n -----------------------------------------------------------------------------------------------------------------\n");
3089 printf(" Task | Runtime ms | Switches | Average delay ms | Maximum delay ms | Maximum delay at |\n");
3090 printf(" -----------------------------------------------------------------------------------------------------------------\n");
0ec04e16 3091
0e9b07e5 3092 next = rb_first(&sched->sorted_atom_root);
0ec04e16
IM		 3093
3094 while (next) {
3095 struct work_atoms *work_list;
3096
3097 work_list = rb_entry(next, struct work_atoms, node);
0e9b07e5 3098 output_lat_thread(sched, work_list);
0ec04e16 3099 next = rb_next(next);
ae536acf 3100 thread__zput(work_list->thread);
0ec04e16
IM		 3101 }
3102
80790e0b 3103 printf(" -----------------------------------------------------------------------------------------------------------------\n");
9486aa38 3104 printf(" TOTAL: |%11.3f ms |%9" PRIu64 " |\n",
4fc76e49 3105 (double)sched->all_runtime / NSEC_PER_MSEC, sched->all_count);
0ec04e16
IM		 3106
3107 printf(" ---------------------------------------------------\n");
3108
0e9b07e5 3109 print_bad_events(sched);
0ec04e16
IM		 3110 printf("\n");
3111
a116e05d 3112 return 0;
0ec04e16
IM		 3113}
3114
99623c62
JO		 3115static int setup_map_cpus(struct perf_sched *sched)
3116{
73643bb6
JO		 3117 struct cpu_map *map;
3118
99623c62
JO		 3119 sched->max_cpu = sysconf(_SC_NPROCESSORS_CONF);
3120
3121 if (sched->map.comp) {
3122 sched->map.comp_cpus = zalloc(sched->max_cpu * sizeof(int));
cf294f24
JO		 3123 if (!sched->map.comp_cpus)
3124 return -1;
99623c62
JO		 3125 }
3126
73643bb6
JO		 3127 if (!sched->map.cpus_str)
3128 return 0;
3129
3130 map = cpu_map__new(sched->map.cpus_str);
3131 if (!map) {
3132 pr_err("failed to get cpus map from %s\n", sched->map.cpus_str);
3133 return -1;
3134 }
3135
3136 sched->map.cpus = map;
99623c62
JO		 3137 return 0;
3138}
3139
a151a37a
JO		 3140static int setup_color_pids(struct perf_sched *sched)
3141{
3142 struct thread_map *map;
3143
3144 if (!sched->map.color_pids_str)
3145 return 0;
3146
3147 map = thread_map__new_by_tid_str(sched->map.color_pids_str);
3148 if (!map) {
3149 pr_err("failed to get thread map from %s\n", sched->map.color_pids_str);
3150 return -1;
3151 }
3152
3153 sched->map.color_pids = map;
3154 return 0;
3155}
3156
cf294f24
JO		 3157static int setup_color_cpus(struct perf_sched *sched)
3158{
3159 struct cpu_map *map;
3160
3161 if (!sched->map.color_cpus_str)
3162 return 0;
3163
3164 map = cpu_map__new(sched->map.color_cpus_str);
3165 if (!map) {
3166 pr_err("failed to get thread map from %s\n", sched->map.color_cpus_str);
3167 return -1;
3168 }
3169
3170 sched->map.color_cpus = map;
3171 return 0;
3172}
3173
0e9b07e5 3174static int perf_sched__map(struct perf_sched *sched)
0ec04e16 3175{
99623c62
JO		 3176 if (setup_map_cpus(sched))
3177 return -1;
40749d0f 3178
a151a37a
JO		 3179 if (setup_color_pids(sched))
3180 return -1;
3181
cf294f24
JO		 3182 if (setup_color_cpus(sched))
3183 return -1;
3184
0ec04e16 3185 setup_pager();
ae536acf 3186 if (perf_sched__read_events(sched))
a116e05d 3187 return -1;
0e9b07e5 3188 print_bad_events(sched);
a116e05d 3189 return 0;
0ec04e16
IM		 3190}
3191
0e9b07e5 3192static int perf_sched__replay(struct perf_sched *sched)
0ec04e16
IM		 3193{
3194 unsigned long i;
3195
0e9b07e5
ACM		 3196 calibrate_run_measurement_overhead(sched);
3197 calibrate_sleep_measurement_overhead(sched);
0ec04e16 3198
0e9b07e5 3199 test_calibrations(sched);
0ec04e16 3200
ae536acf 3201 if (perf_sched__read_events(sched))
a116e05d 3202 return -1;
0ec04e16 3203
0e9b07e5
ACM		 3204 printf("nr_run_events: %ld\n", sched->nr_run_events);
3205 printf("nr_sleep_events: %ld\n", sched->nr_sleep_events);
3206 printf("nr_wakeup_events: %ld\n", sched->nr_wakeup_events);
0ec04e16 3207
0e9b07e5
ACM		 3208 if (sched->targetless_wakeups)
3209 printf("target-less wakeups: %ld\n", sched->targetless_wakeups);
3210 if (sched->multitarget_wakeups)
3211 printf("multi-target wakeups: %ld\n", sched->multitarget_wakeups);
3212 if (sched->nr_run_events_optimized)
0ec04e16 3213 printf("run atoms optimized: %ld\n",
0e9b07e5 3214 sched->nr_run_events_optimized);
0ec04e16 3215
0e9b07e5
ACM		 3216 print_task_traces(sched);
3217 add_cross_task_wakeups(sched);
0ec04e16 3218
0e9b07e5 3219 create_tasks(sched);
0ec04e16 3220 printf("------------------------------------------------------------\n");
0e9b07e5
ACM		 3221 for (i = 0; i < sched->replay_repeat; i++)
3222 run_one_test(sched);
a116e05d
ACM		 3223
3224 return 0;
0ec04e16
IM		 3225}
3226
0e9b07e5
ACM		 3227static void setup_sorting(struct perf_sched *sched, const struct option *options,
3228 const char * const usage_msg[])
daa1d7a5 3229{
0e9b07e5 3230 char *tmp, *tok, *str = strdup(sched->sort_order);
daa1d7a5
FW		 3231
3232 for (tok = strtok_r(str, ", ", &tmp);
3233 tok; tok = strtok_r(NULL, ", ", &tmp)) {
0e9b07e5 3234 if (sort_dimension__add(tok, &sched->sort_list) < 0) {
c7118369
NK		 3235 usage_with_options_msg(usage_msg, options,
3236 "Unknown --sort key: `%s'", tok);
daa1d7a5
FW		 3237 }
3238 }
3239
3240 free(str);
3241
0e9b07e5 3242 sort_dimension__add("pid", &sched->cmp_pid);
daa1d7a5
FW		 3243}
3244
1fc35b29
IM		 3245static int __cmd_record(int argc, const char **argv)
3246{
3247 unsigned int rec_argc, i, j;
3248 const char **rec_argv;
0e9b07e5
ACM		 3249 const char * const record_args[] = {
3250 "record",
3251 "-a",
3252 "-R",
0e9b07e5
ACM		 3253 "-m", "1024",
3254 "-c", "1",
3255 "-e", "sched:sched_switch",
3256 "-e", "sched:sched_stat_wait",
3257 "-e", "sched:sched_stat_sleep",
3258 "-e", "sched:sched_stat_iowait",
3259 "-e", "sched:sched_stat_runtime",
0e9b07e5
ACM		 3260 "-e", "sched:sched_process_fork",
3261 "-e", "sched:sched_wakeup",
7fff9597 3262 "-e", "sched:sched_wakeup_new",
0e9b07e5
ACM		 3263 "-e", "sched:sched_migrate_task",
3264 };
1fc35b29
IM		 3265
3266 rec_argc = ARRAY_SIZE(record_args) + argc - 1;
3267 rec_argv = calloc(rec_argc + 1, sizeof(char *));
3268
e462dc55 3269 if (rec_argv == NULL)
ce47dc56
CS		 3270 return -ENOMEM;
3271
1fc35b29
IM		 3272 for (i = 0; i < ARRAY_SIZE(record_args); i++)
3273 rec_argv[i] = strdup(record_args[i]);
3274
3275 for (j = 1; j < (unsigned int)argc; j++, i++)
3276 rec_argv[i] = argv[j];
3277
3278 BUG_ON(i != rec_argc);
3279
b0ad8ea6 3280 return cmd_record(i, rec_argv);
1fc35b29
IM		 3281}
3282
b0ad8ea6 3283int cmd_sched(int argc, const char **argv)
0a02ad93 3284{
8a39df8f
AH		 3285 const char default_sort_order[] = "avg, max, switch, runtime";
3286 struct perf_sched sched = {
3287 .tool = {
3288 .sample = perf_sched__process_tracepoint_sample,
3289 .comm = perf_event__process_comm,
f3b3614a 3290 .namespaces = perf_event__process_namespaces,
8a39df8f
AH		 3291 .lost = perf_event__process_lost,
3292 .fork = perf_sched__process_fork_event,
0a8cb85c 3293 .ordered_events = true,
8a39df8f
AH		 3294 },
3295 .cmp_pid = LIST_HEAD_INIT(sched.cmp_pid),
3296 .sort_list = LIST_HEAD_INIT(sched.sort_list),
3297 .start_work_mutex = PTHREAD_MUTEX_INITIALIZER,
3298 .work_done_wait_mutex = PTHREAD_MUTEX_INITIALIZER,
8a39df8f
AH		 3299 .sort_order = default_sort_order,
3300 .replay_repeat = 10,
3301 .profile_cpu = -1,
3302 .next_shortname1 = 'A',
3303 .next_shortname2 = '0',
2f80dd44 3304 .skip_merge = 0,
6c973c90
DA		 3305 .show_callchain = 1,
3306 .max_stack = 5,
8a39df8f 3307 };
77f02f44
NK		 3308 const struct option sched_options[] = {
3309 OPT_STRING('i', "input", &input_name, "file",
3310 "input file name"),
3311 OPT_INCR('v', "verbose", &verbose,
3312 "be more verbose (show symbol address, etc)"),
3313 OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
3314 "dump raw trace in ASCII"),
6fa94258 3315 OPT_BOOLEAN('f', "force", &sched.force, "don't complain, do it"),
77f02f44
NK		 3316 OPT_END()
3317 };
0e9b07e5
ACM		 3318 const struct option latency_options[] = {
3319 OPT_STRING('s', "sort", &sched.sort_order, "key[,key2...]",
3320 "sort by key(s): runtime, switch, avg, max"),
0e9b07e5
ACM		 3321 OPT_INTEGER('C', "CPU", &sched.profile_cpu,
3322 "CPU to profile on"),
2f80dd44
JB		 3323 OPT_BOOLEAN('p', "pids", &sched.skip_merge,
3324 "latency stats per pid instead of per comm"),
77f02f44 3325 OPT_PARENT(sched_options)
0e9b07e5
ACM		 3326 };
3327 const struct option replay_options[] = {
3328 OPT_UINTEGER('r', "repeat", &sched.replay_repeat,
3329 "repeat the workload replay N times (-1: infinite)"),
77f02f44 3330 OPT_PARENT(sched_options)
0e9b07e5 3331 };
99623c62
JO		 3332 const struct option map_options[] = {
3333 OPT_BOOLEAN(0, "compact", &sched.map.comp,
3334 "map output in compact mode"),
a151a37a
JO		 3335 OPT_STRING(0, "color-pids", &sched.map.color_pids_str, "pids",
3336 "highlight given pids in map"),
cf294f24
JO		 3337 OPT_STRING(0, "color-cpus", &sched.map.color_cpus_str, "cpus",
3338 "highlight given CPUs in map"),
73643bb6
JO		 3339 OPT_STRING(0, "cpus", &sched.map.cpus_str, "cpus",
3340 "display given CPUs in map"),
77f02f44 3341 OPT_PARENT(sched_options)
99623c62 3342 };
49394a2a
DA		 3343 const struct option timehist_options[] = {
3344 OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
3345 "file", "vmlinux pathname"),
3346 OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
3347 "file", "kallsyms pathname"),
6c973c90
DA		 3348 OPT_BOOLEAN('g', "call-graph", &sched.show_callchain,
3349 "Display call chains if present (default on)"),
3350 OPT_UINTEGER(0, "max-stack", &sched.max_stack,
3351 "Maximum number of functions to display backtrace."),
49394a2a
DA		 3352 OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory",
3353 "Look for files with symbols relative to this directory"),
52df138c
DA		 3354 OPT_BOOLEAN('s', "summary", &sched.summary_only,
3355 "Show only syscall summary with statistics"),
3356 OPT_BOOLEAN('S', "with-summary", &sched.summary,
3357 "Show all syscalls and summary with statistics"),
fc1469f1 3358 OPT_BOOLEAN('w', "wakeups", &sched.show_wakeups, "Show wakeup events"),
292c4a8f 3359 OPT_BOOLEAN('n', "next", &sched.show_next, "Show next task"),
350f54fa 3360 OPT_BOOLEAN('M', "migrations", &sched.show_migrations, "Show migration events"),
a407b067 3361 OPT_BOOLEAN('V', "cpu-visual", &sched.show_cpu_visual, "Add CPU visual"),
07235f84 3362 OPT_BOOLEAN('I', "idle-hist", &sched.idle_hist, "Show idle events only"),
853b7407
DA		 3363 OPT_STRING(0, "time", &sched.time_str, "str",
3364 "Time span for analysis (start,stop)"),
414e050c 3365 OPT_BOOLEAN(0, "state", &sched.show_state, "Show task state when sched-out"),
49394a2a
DA		 3366 OPT_PARENT(sched_options)
3367 };
3368
0e9b07e5
ACM		 3369 const char * const latency_usage[] = {
3370 "perf sched latency [<options>]",
3371 NULL
3372 };
3373 const char * const replay_usage[] = {
3374 "perf sched replay [<options>]",
3375 NULL
3376 };
99623c62
JO		 3377 const char * const map_usage[] = {
3378 "perf sched map [<options>]",
3379 NULL
3380 };
49394a2a
DA		 3381 const char * const timehist_usage[] = {
3382 "perf sched timehist [<options>]",
3383 NULL
3384 };
a83edb2d 3385 const char *const sched_subcommands[] = { "record", "latency", "map",
49394a2a
DA		 3386 "replay", "script",
3387 "timehist", NULL };
a83edb2d
RR		 3388 const char *sched_usage[] = {
3389 NULL,
0e9b07e5
ACM		 3390 NULL
3391 };
3392 struct trace_sched_handler lat_ops = {
3393 .wakeup_event = latency_wakeup_event,
3394 .switch_event = latency_switch_event,
3395 .runtime_event = latency_runtime_event,
0e9b07e5
ACM		 3396 .migrate_task_event = latency_migrate_task_event,
3397 };
3398 struct trace_sched_handler map_ops = {
3399 .switch_event = map_switch_event,
3400 };
3401 struct trace_sched_handler replay_ops = {
3402 .wakeup_event = replay_wakeup_event,
3403 .switch_event = replay_switch_event,
3404 .fork_event = replay_fork_event,
3405 };
156a2b02
AH		 3406 unsigned int i;
3407
3408 for (i = 0; i < ARRAY_SIZE(sched.curr_pid); i++)
3409 sched.curr_pid[i] = -1;
0e9b07e5 3410
a83edb2d
RR		 3411 argc = parse_options_subcommand(argc, argv, sched_options, sched_subcommands,
3412 sched_usage, PARSE_OPT_STOP_AT_NON_OPTION);
f2858d8a
IM		 3413 if (!argc)
3414 usage_with_options(sched_usage, sched_options);
0a02ad93 3415
c0777c5a 3416 /*
133dc4c3 3417 * Aliased to 'perf script' for now:
c0777c5a 3418 */
133dc4c3 3419 if (!strcmp(argv[0], "script"))
b0ad8ea6 3420 return cmd_script(argc, argv);
c0777c5a 3421
1fc35b29
IM		 3422 if (!strncmp(argv[0], "rec", 3)) {
3423 return __cmd_record(argc, argv);
3424 } else if (!strncmp(argv[0], "lat", 3)) {
0e9b07e5 3425 sched.tp_handler = &lat_ops;
f2858d8a
IM		 3426 if (argc > 1) {
3427 argc = parse_options(argc, argv, latency_options, latency_usage, 0);
3428 if (argc)
3429 usage_with_options(latency_usage, latency_options);
f2858d8a 3430 }
0e9b07e5
ACM		 3431 setup_sorting(&sched, latency_options, latency_usage);
3432 return perf_sched__lat(&sched);
0ec04e16 3433 } else if (!strcmp(argv[0], "map")) {
99623c62 3434 if (argc) {
a151a37a 3435 argc = parse_options(argc, argv, map_options, map_usage, 0);
99623c62
JO		 3436 if (argc)
3437 usage_with_options(map_usage, map_options);
3438 }
0e9b07e5
ACM		 3439 sched.tp_handler = &map_ops;
3440 setup_sorting(&sched, latency_options, latency_usage);
3441 return perf_sched__map(&sched);
f2858d8a 3442 } else if (!strncmp(argv[0], "rep", 3)) {
0e9b07e5 3443 sched.tp_handler = &replay_ops;
f2858d8a
IM		 3444 if (argc) {
3445 argc = parse_options(argc, argv, replay_options, replay_usage, 0);
3446 if (argc)
3447 usage_with_options(replay_usage, replay_options);
3448 }
0e9b07e5 3449 return perf_sched__replay(&sched);
49394a2a
DA		 3450 } else if (!strcmp(argv[0], "timehist")) {
3451 if (argc) {
3452 argc = parse_options(argc, argv, timehist_options,
3453 timehist_usage, 0);
3454 if (argc)
3455 usage_with_options(timehist_usage, timehist_options);
3456 }
292c4a8f
BG		 3457 if ((sched.show_wakeups || sched.show_next) &&
3458 sched.summary_only) {
3459 pr_err(" Error: -s and -[n|w] are mutually exclusive.\n");
fc1469f1 3460 parse_options_usage(timehist_usage, timehist_options, "s", true);
292c4a8f
BG		 3461 if (sched.show_wakeups)
3462 parse_options_usage(NULL, timehist_options, "w", true);
3463 if (sched.show_next)
3464 parse_options_usage(NULL, timehist_options, "n", true);
fc1469f1
DA		 3465 return -EINVAL;
3466 }
3467
49394a2a 3468 return perf_sched__timehist(&sched);
f2858d8a
IM		 3469 } else {
3470 usage_with_options(sched_usage, sched_options);
3471 }
3472
ec156764 3473 return 0;
0a02ad93 3474}
Ubuntu Artful kernel mirror