xref: /linux/tools/perf/builtin-sched.c (revision c34e9ab9a612ee8b18273398ef75c207b01f516d)
1 // SPDX-License-Identifier: GPL-2.0
2 #include "builtin.h"
3 #include "perf-sys.h"
4 
5 #include "util/cpumap.h"
6 #include "util/evlist.h"
7 #include "util/evsel.h"
8 #include "util/evsel_fprintf.h"
9 #include "util/mutex.h"
10 #include "util/symbol.h"
11 #include "util/thread.h"
12 #include "util/header.h"
13 #include "util/session.h"
14 #include "util/tool.h"
15 #include "util/cloexec.h"
16 #include "util/thread_map.h"
17 #include "util/color.h"
18 #include "util/stat.h"
19 #include "util/string2.h"
20 #include "util/callchain.h"
21 #include "util/time-utils.h"
22 
23 #include <subcmd/pager.h>
24 #include <subcmd/parse-options.h>
25 #include "util/trace-event.h"
26 
27 #include "util/debug.h"
28 #include "util/event.h"
29 #include "util/util.h"
30 
31 #include <linux/kernel.h>
32 #include <linux/log2.h>
33 #include <linux/zalloc.h>
34 #include <sys/prctl.h>
35 #include <sys/resource.h>
36 #include <inttypes.h>
37 
38 #include <errno.h>
39 #include <semaphore.h>
40 #include <pthread.h>
41 #include <math.h>
42 #include <api/fs/fs.h>
43 #include <perf/cpumap.h>
44 #include <linux/time64.h>
45 #include <linux/err.h>
46 
47 #include <linux/ctype.h>
48 
49 #define PR_SET_NAME		15               /* Set process name */
50 #define MAX_CPUS		4096
51 #define COMM_LEN		20
52 #define SYM_LEN			129
53 #define MAX_PID			1024000
54 #define MAX_PRIO		140
55 
56 static const char *cpu_list;
57 static DECLARE_BITMAP(cpu_bitmap, MAX_NR_CPUS);
58 
59 struct sched_atom;
60 
61 struct task_desc {
62 	unsigned long		nr;
63 	unsigned long		pid;
64 	char			comm[COMM_LEN];
65 
66 	unsigned long		nr_events;
67 	unsigned long		curr_event;
68 	struct sched_atom	**atoms;
69 
70 	pthread_t		thread;
71 
72 	sem_t			ready_for_work;
73 	sem_t			work_done_sem;
74 
75 	u64			cpu_usage;
76 };
77 
78 enum sched_event_type {
79 	SCHED_EVENT_RUN,
80 	SCHED_EVENT_SLEEP,
81 	SCHED_EVENT_WAKEUP,
82 };
83 
84 struct sched_atom {
85 	enum sched_event_type	type;
86 	u64			timestamp;
87 	u64			duration;
88 	unsigned long		nr;
89 	sem_t			*wait_sem;
90 	struct task_desc	*wakee;
91 };
92 
93 enum thread_state {
94 	THREAD_SLEEPING = 0,
95 	THREAD_WAIT_CPU,
96 	THREAD_SCHED_IN,
97 	THREAD_IGNORE
98 };
99 
100 struct work_atom {
101 	struct list_head	list;
102 	enum thread_state	state;
103 	u64			sched_out_time;
104 	u64			wake_up_time;
105 	u64			sched_in_time;
106 	u64			runtime;
107 };
108 
109 struct work_atoms {
110 	struct list_head	work_list;
111 	struct thread		*thread;
112 	struct rb_node		node;
113 	u64			max_lat;
114 	u64			max_lat_start;
115 	u64			max_lat_end;
116 	u64			total_lat;
117 	u64			nb_atoms;
118 	u64			total_runtime;
119 	int			num_merged;
120 };
121 
122 typedef int (*sort_fn_t)(struct work_atoms *, struct work_atoms *);
123 
124 struct perf_sched;
125 
126 struct trace_sched_handler {
127 	int (*switch_event)(struct perf_sched *sched, struct evsel *evsel,
128 			    struct perf_sample *sample, struct machine *machine);
129 
130 	int (*runtime_event)(struct perf_sched *sched, struct evsel *evsel,
131 			     struct perf_sample *sample, struct machine *machine);
132 
133 	int (*wakeup_event)(struct perf_sched *sched, struct evsel *evsel,
134 			    struct perf_sample *sample, struct machine *machine);
135 
136 	/* PERF_RECORD_FORK event, not sched_process_fork tracepoint */
137 	int (*fork_event)(struct perf_sched *sched, union perf_event *event,
138 			  struct machine *machine);
139 
140 	int (*migrate_task_event)(struct perf_sched *sched,
141 				  struct evsel *evsel,
142 				  struct perf_sample *sample,
143 				  struct machine *machine);
144 };
145 
146 #define COLOR_PIDS PERF_COLOR_BLUE
147 #define COLOR_CPUS PERF_COLOR_BG_RED
148 
149 struct perf_sched_map {
150 	DECLARE_BITMAP(comp_cpus_mask, MAX_CPUS);
151 	struct perf_cpu		*comp_cpus;
152 	bool			 comp;
153 	struct perf_thread_map *color_pids;
154 	const char		*color_pids_str;
155 	struct perf_cpu_map	*color_cpus;
156 	const char		*color_cpus_str;
157 	const char		*task_name;
158 	struct strlist		*task_names;
159 	bool			fuzzy;
160 	struct perf_cpu_map	*cpus;
161 	const char		*cpus_str;
162 };
163 
164 struct perf_sched {
165 	struct perf_tool tool;
166 	const char	 *sort_order;
167 	unsigned long	 nr_tasks;
168 	struct task_desc **pid_to_task;
169 	struct task_desc **tasks;
170 	const struct trace_sched_handler *tp_handler;
171 	struct mutex	 start_work_mutex;
172 	struct mutex	 work_done_wait_mutex;
173 	int		 profile_cpu;
174 /*
175  * Track the current task - that way we can know whether there's any
176  * weird events, such as a task being switched away that is not current.
177  */
178 	struct perf_cpu	 max_cpu;
179 	u32		 *curr_pid;
180 	struct thread	 **curr_thread;
181 	struct thread	 **curr_out_thread;
182 	char		 next_shortname1;
183 	char		 next_shortname2;
184 	unsigned int	 replay_repeat;
185 	unsigned long	 nr_run_events;
186 	unsigned long	 nr_sleep_events;
187 	unsigned long	 nr_wakeup_events;
188 	unsigned long	 nr_sleep_corrections;
189 	unsigned long	 nr_run_events_optimized;
190 	unsigned long	 targetless_wakeups;
191 	unsigned long	 multitarget_wakeups;
192 	unsigned long	 nr_runs;
193 	unsigned long	 nr_timestamps;
194 	unsigned long	 nr_unordered_timestamps;
195 	unsigned long	 nr_context_switch_bugs;
196 	unsigned long	 nr_events;
197 	unsigned long	 nr_lost_chunks;
198 	unsigned long	 nr_lost_events;
199 	u64		 run_measurement_overhead;
200 	u64		 sleep_measurement_overhead;
201 	u64		 start_time;
202 	u64		 cpu_usage;
203 	u64		 runavg_cpu_usage;
204 	u64		 parent_cpu_usage;
205 	u64		 runavg_parent_cpu_usage;
206 	u64		 sum_runtime;
207 	u64		 sum_fluct;
208 	u64		 run_avg;
209 	u64		 all_runtime;
210 	u64		 all_count;
211 	u64		 *cpu_last_switched;
212 	struct rb_root_cached atom_root, sorted_atom_root, merged_atom_root;
213 	struct list_head sort_list, cmp_pid;
214 	bool force;
215 	bool skip_merge;
216 	struct perf_sched_map map;
217 
218 	/* options for timehist command */
219 	bool		summary;
220 	bool		summary_only;
221 	bool		idle_hist;
222 	bool		show_callchain;
223 	unsigned int	max_stack;
224 	bool		show_cpu_visual;
225 	bool		show_wakeups;
226 	bool		show_next;
227 	bool		show_migrations;
228 	bool		pre_migrations;
229 	bool		show_state;
230 	bool		show_prio;
231 	u64		skipped_samples;
232 	const char	*time_str;
233 	struct perf_time_interval ptime;
234 	struct perf_time_interval hist_time;
235 	volatile bool   thread_funcs_exit;
236 	const char	*prio_str;
237 	DECLARE_BITMAP(prio_bitmap, MAX_PRIO);
238 };
239 
240 /* per thread run time data */
241 struct thread_runtime {
242 	u64 last_time;      /* time of previous sched in/out event */
243 	u64 dt_run;         /* run time */
244 	u64 dt_sleep;       /* time between CPU access by sleep (off cpu) */
245 	u64 dt_iowait;      /* time between CPU access by iowait (off cpu) */
246 	u64 dt_preempt;     /* time between CPU access by preempt (off cpu) */
247 	u64 dt_delay;       /* time between wakeup and sched-in */
248 	u64 dt_pre_mig;     /* time between migration and wakeup */
249 	u64 ready_to_run;   /* time of wakeup */
250 	u64 migrated;	    /* time when a thread is migrated */
251 
252 	struct stats run_stats;
253 	u64 total_run_time;
254 	u64 total_sleep_time;
255 	u64 total_iowait_time;
256 	u64 total_preempt_time;
257 	u64 total_delay_time;
258 	u64 total_pre_mig_time;
259 
260 	char last_state;
261 
262 	char shortname[3];
263 	bool comm_changed;
264 
265 	u64 migrations;
266 
267 	int prio;
268 };
269 
270 /* per event run time data */
271 struct evsel_runtime {
272 	u64 *last_time; /* time this event was last seen per cpu */
273 	u32 ncpu;       /* highest cpu slot allocated */
274 };
275 
276 /* per cpu idle time data */
277 struct idle_thread_runtime {
278 	struct thread_runtime	tr;
279 	struct thread		*last_thread;
280 	struct rb_root_cached	sorted_root;
281 	struct callchain_root	callchain;
282 	struct callchain_cursor	cursor;
283 };
284 
285 /* track idle times per cpu */
286 static struct thread **idle_threads;
287 static int idle_max_cpu;
288 static char idle_comm[] = "<idle>";
289 
290 static u64 get_nsecs(void)
291 {
292 	struct timespec ts;
293 
294 	clock_gettime(CLOCK_MONOTONIC, &ts);
295 
296 	return ts.tv_sec * NSEC_PER_SEC + ts.tv_nsec;
297 }
298 
299 static void burn_nsecs(struct perf_sched *sched, u64 nsecs)
300 {
301 	u64 T0 = get_nsecs(), T1;
302 
303 	do {
304 		T1 = get_nsecs();
305 	} while (T1 + sched->run_measurement_overhead < T0 + nsecs);
306 }
307 
308 static void sleep_nsecs(u64 nsecs)
309 {
310 	struct timespec ts;
311 
312 	ts.tv_nsec = nsecs % 999999999;
313 	ts.tv_sec = nsecs / 999999999;
314 
315 	nanosleep(&ts, NULL);
316 }
317 
318 static void calibrate_run_measurement_overhead(struct perf_sched *sched)
319 {
320 	u64 T0, T1, delta, min_delta = NSEC_PER_SEC;
321 	int i;
322 
323 	for (i = 0; i < 10; i++) {
324 		T0 = get_nsecs();
325 		burn_nsecs(sched, 0);
326 		T1 = get_nsecs();
327 		delta = T1-T0;
328 		min_delta = min(min_delta, delta);
329 	}
330 	sched->run_measurement_overhead = min_delta;
331 
332 	printf("run measurement overhead: %" PRIu64 " nsecs\n", min_delta);
333 }
334 
335 static void calibrate_sleep_measurement_overhead(struct perf_sched *sched)
336 {
337 	u64 T0, T1, delta, min_delta = NSEC_PER_SEC;
338 	int i;
339 
340 	for (i = 0; i < 10; i++) {
341 		T0 = get_nsecs();
342 		sleep_nsecs(10000);
343 		T1 = get_nsecs();
344 		delta = T1-T0;
345 		min_delta = min(min_delta, delta);
346 	}
347 	min_delta -= 10000;
348 	sched->sleep_measurement_overhead = min_delta;
349 
350 	printf("sleep measurement overhead: %" PRIu64 " nsecs\n", min_delta);
351 }
352 
353 static struct sched_atom *
354 get_new_event(struct task_desc *task, u64 timestamp)
355 {
356 	struct sched_atom *event = zalloc(sizeof(*event));
357 	unsigned long idx = task->nr_events;
358 	size_t size;
359 
360 	event->timestamp = timestamp;
361 	event->nr = idx;
362 
363 	task->nr_events++;
364 	size = sizeof(struct sched_atom *) * task->nr_events;
365 	task->atoms = realloc(task->atoms, size);
366 	BUG_ON(!task->atoms);
367 
368 	task->atoms[idx] = event;
369 
370 	return event;
371 }
372 
373 static struct sched_atom *last_event(struct task_desc *task)
374 {
375 	if (!task->nr_events)
376 		return NULL;
377 
378 	return task->atoms[task->nr_events - 1];
379 }
380 
381 static void add_sched_event_run(struct perf_sched *sched, struct task_desc *task,
382 				u64 timestamp, u64 duration)
383 {
384 	struct sched_atom *event, *curr_event = last_event(task);
385 
386 	/*
387 	 * optimize an existing RUN event by merging this one
388 	 * to it:
389 	 */
390 	if (curr_event && curr_event->type == SCHED_EVENT_RUN) {
391 		sched->nr_run_events_optimized++;
392 		curr_event->duration += duration;
393 		return;
394 	}
395 
396 	event = get_new_event(task, timestamp);
397 
398 	event->type = SCHED_EVENT_RUN;
399 	event->duration = duration;
400 
401 	sched->nr_run_events++;
402 }
403 
404 static void add_sched_event_wakeup(struct perf_sched *sched, struct task_desc *task,
405 				   u64 timestamp, struct task_desc *wakee)
406 {
407 	struct sched_atom *event, *wakee_event;
408 
409 	event = get_new_event(task, timestamp);
410 	event->type = SCHED_EVENT_WAKEUP;
411 	event->wakee = wakee;
412 
413 	wakee_event = last_event(wakee);
414 	if (!wakee_event || wakee_event->type != SCHED_EVENT_SLEEP) {
415 		sched->targetless_wakeups++;
416 		return;
417 	}
418 	if (wakee_event->wait_sem) {
419 		sched->multitarget_wakeups++;
420 		return;
421 	}
422 
423 	wakee_event->wait_sem = zalloc(sizeof(*wakee_event->wait_sem));
424 	sem_init(wakee_event->wait_sem, 0, 0);
425 	event->wait_sem = wakee_event->wait_sem;
426 
427 	sched->nr_wakeup_events++;
428 }
429 
430 static void add_sched_event_sleep(struct perf_sched *sched, struct task_desc *task,
431 				  u64 timestamp)
432 {
433 	struct sched_atom *event = get_new_event(task, timestamp);
434 
435 	event->type = SCHED_EVENT_SLEEP;
436 
437 	sched->nr_sleep_events++;
438 }
439 
440 static struct task_desc *register_pid(struct perf_sched *sched,
441 				      unsigned long pid, const char *comm)
442 {
443 	struct task_desc *task;
444 	static int pid_max;
445 
446 	if (sched->pid_to_task == NULL) {
447 		if (sysctl__read_int("kernel/pid_max", &pid_max) < 0)
448 			pid_max = MAX_PID;
449 		BUG_ON((sched->pid_to_task = calloc(pid_max, sizeof(struct task_desc *))) == NULL);
450 	}
451 	if (pid >= (unsigned long)pid_max) {
452 		BUG_ON((sched->pid_to_task = realloc(sched->pid_to_task, (pid + 1) *
453 			sizeof(struct task_desc *))) == NULL);
454 		while (pid >= (unsigned long)pid_max)
455 			sched->pid_to_task[pid_max++] = NULL;
456 	}
457 
458 	task = sched->pid_to_task[pid];
459 
460 	if (task)
461 		return task;
462 
463 	task = zalloc(sizeof(*task));
464 	task->pid = pid;
465 	task->nr = sched->nr_tasks;
466 	strcpy(task->comm, comm);
467 	/*
468 	 * every task starts in sleeping state - this gets ignored
469 	 * if there's no wakeup pointing to this sleep state:
470 	 */
471 	add_sched_event_sleep(sched, task, 0);
472 
473 	sched->pid_to_task[pid] = task;
474 	sched->nr_tasks++;
475 	sched->tasks = realloc(sched->tasks, sched->nr_tasks * sizeof(struct task_desc *));
476 	BUG_ON(!sched->tasks);
477 	sched->tasks[task->nr] = task;
478 
479 	if (verbose > 0)
480 		printf("registered task #%ld, PID %ld (%s)\n", sched->nr_tasks, pid, comm);
481 
482 	return task;
483 }
484 
485 
486 static void print_task_traces(struct perf_sched *sched)
487 {
488 	struct task_desc *task;
489 	unsigned long i;
490 
491 	for (i = 0; i < sched->nr_tasks; i++) {
492 		task = sched->tasks[i];
493 		printf("task %6ld (%20s:%10ld), nr_events: %ld\n",
494 			task->nr, task->comm, task->pid, task->nr_events);
495 	}
496 }
497 
498 static void add_cross_task_wakeups(struct perf_sched *sched)
499 {
500 	struct task_desc *task1, *task2;
501 	unsigned long i, j;
502 
503 	for (i = 0; i < sched->nr_tasks; i++) {
504 		task1 = sched->tasks[i];
505 		j = i + 1;
506 		if (j == sched->nr_tasks)
507 			j = 0;
508 		task2 = sched->tasks[j];
509 		add_sched_event_wakeup(sched, task1, 0, task2);
510 	}
511 }
512 
513 static void perf_sched__process_event(struct perf_sched *sched,
514 				      struct sched_atom *atom)
515 {
516 	int ret = 0;
517 
518 	switch (atom->type) {
519 		case SCHED_EVENT_RUN:
520 			burn_nsecs(sched, atom->duration);
521 			break;
522 		case SCHED_EVENT_SLEEP:
523 			if (atom->wait_sem)
524 				ret = sem_wait(atom->wait_sem);
525 			BUG_ON(ret);
526 			break;
527 		case SCHED_EVENT_WAKEUP:
528 			if (atom->wait_sem)
529 				ret = sem_post(atom->wait_sem);
530 			BUG_ON(ret);
531 			break;
532 		default:
533 			BUG_ON(1);
534 	}
535 }
536 
537 static u64 get_cpu_usage_nsec_parent(void)
538 {
539 	struct rusage ru;
540 	u64 sum;
541 	int err;
542 
543 	err = getrusage(RUSAGE_SELF, &ru);
544 	BUG_ON(err);
545 
546 	sum =  ru.ru_utime.tv_sec * NSEC_PER_SEC + ru.ru_utime.tv_usec * NSEC_PER_USEC;
547 	sum += ru.ru_stime.tv_sec * NSEC_PER_SEC + ru.ru_stime.tv_usec * NSEC_PER_USEC;
548 
549 	return sum;
550 }
551 
552 static int self_open_counters(struct perf_sched *sched, unsigned long cur_task)
553 {
554 	struct perf_event_attr attr;
555 	char sbuf[STRERR_BUFSIZE], info[STRERR_BUFSIZE];
556 	int fd;
557 	struct rlimit limit;
558 	bool need_privilege = false;
559 
560 	memset(&attr, 0, sizeof(attr));
561 
562 	attr.type = PERF_TYPE_SOFTWARE;
563 	attr.config = PERF_COUNT_SW_TASK_CLOCK;
564 
565 force_again:
566 	fd = sys_perf_event_open(&attr, 0, -1, -1,
567 				 perf_event_open_cloexec_flag());
568 
569 	if (fd < 0) {
570 		if (errno == EMFILE) {
571 			if (sched->force) {
572 				BUG_ON(getrlimit(RLIMIT_NOFILE, &limit) == -1);
573 				limit.rlim_cur += sched->nr_tasks - cur_task;
574 				if (limit.rlim_cur > limit.rlim_max) {
575 					limit.rlim_max = limit.rlim_cur;
576 					need_privilege = true;
577 				}
578 				if (setrlimit(RLIMIT_NOFILE, &limit) == -1) {
579 					if (need_privilege && errno == EPERM)
580 						strcpy(info, "Need privilege\n");
581 				} else
582 					goto force_again;
583 			} else
584 				strcpy(info, "Have a try with -f option\n");
585 		}
586 		pr_err("Error: sys_perf_event_open() syscall returned "
587 		       "with %d (%s)\n%s", fd,
588 		       str_error_r(errno, sbuf, sizeof(sbuf)), info);
589 		exit(EXIT_FAILURE);
590 	}
591 	return fd;
592 }
593 
594 static u64 get_cpu_usage_nsec_self(int fd)
595 {
596 	u64 runtime;
597 	int ret;
598 
599 	ret = read(fd, &runtime, sizeof(runtime));
600 	BUG_ON(ret != sizeof(runtime));
601 
602 	return runtime;
603 }
604 
605 struct sched_thread_parms {
606 	struct task_desc  *task;
607 	struct perf_sched *sched;
608 	int fd;
609 };
610 
611 static void *thread_func(void *ctx)
612 {
613 	struct sched_thread_parms *parms = ctx;
614 	struct task_desc *this_task = parms->task;
615 	struct perf_sched *sched = parms->sched;
616 	u64 cpu_usage_0, cpu_usage_1;
617 	unsigned long i, ret;
618 	char comm2[22];
619 	int fd = parms->fd;
620 
621 	zfree(&parms);
622 
623 	sprintf(comm2, ":%s", this_task->comm);
624 	prctl(PR_SET_NAME, comm2);
625 	if (fd < 0)
626 		return NULL;
627 
628 	while (!sched->thread_funcs_exit) {
629 		ret = sem_post(&this_task->ready_for_work);
630 		BUG_ON(ret);
631 		mutex_lock(&sched->start_work_mutex);
632 		mutex_unlock(&sched->start_work_mutex);
633 
634 		cpu_usage_0 = get_cpu_usage_nsec_self(fd);
635 
636 		for (i = 0; i < this_task->nr_events; i++) {
637 			this_task->curr_event = i;
638 			perf_sched__process_event(sched, this_task->atoms[i]);
639 		}
640 
641 		cpu_usage_1 = get_cpu_usage_nsec_self(fd);
642 		this_task->cpu_usage = cpu_usage_1 - cpu_usage_0;
643 		ret = sem_post(&this_task->work_done_sem);
644 		BUG_ON(ret);
645 
646 		mutex_lock(&sched->work_done_wait_mutex);
647 		mutex_unlock(&sched->work_done_wait_mutex);
648 	}
649 	return NULL;
650 }
651 
652 static void create_tasks(struct perf_sched *sched)
653 	EXCLUSIVE_LOCK_FUNCTION(sched->start_work_mutex)
654 	EXCLUSIVE_LOCK_FUNCTION(sched->work_done_wait_mutex)
655 {
656 	struct task_desc *task;
657 	pthread_attr_t attr;
658 	unsigned long i;
659 	int err;
660 
661 	err = pthread_attr_init(&attr);
662 	BUG_ON(err);
663 	err = pthread_attr_setstacksize(&attr,
664 			(size_t) max(16 * 1024, (int)PTHREAD_STACK_MIN));
665 	BUG_ON(err);
666 	mutex_lock(&sched->start_work_mutex);
667 	mutex_lock(&sched->work_done_wait_mutex);
668 	for (i = 0; i < sched->nr_tasks; i++) {
669 		struct sched_thread_parms *parms = malloc(sizeof(*parms));
670 		BUG_ON(parms == NULL);
671 		parms->task = task = sched->tasks[i];
672 		parms->sched = sched;
673 		parms->fd = self_open_counters(sched, i);
674 		sem_init(&task->ready_for_work, 0, 0);
675 		sem_init(&task->work_done_sem, 0, 0);
676 		task->curr_event = 0;
677 		err = pthread_create(&task->thread, &attr, thread_func, parms);
678 		BUG_ON(err);
679 	}
680 }
681 
682 static void destroy_tasks(struct perf_sched *sched)
683 	UNLOCK_FUNCTION(sched->start_work_mutex)
684 	UNLOCK_FUNCTION(sched->work_done_wait_mutex)
685 {
686 	struct task_desc *task;
687 	unsigned long i;
688 	int err;
689 
690 	mutex_unlock(&sched->start_work_mutex);
691 	mutex_unlock(&sched->work_done_wait_mutex);
692 	/* Get rid of threads so they won't be upset by mutex destrunction */
693 	for (i = 0; i < sched->nr_tasks; i++) {
694 		task = sched->tasks[i];
695 		err = pthread_join(task->thread, NULL);
696 		BUG_ON(err);
697 		sem_destroy(&task->ready_for_work);
698 		sem_destroy(&task->work_done_sem);
699 	}
700 }
701 
702 static void wait_for_tasks(struct perf_sched *sched)
703 	EXCLUSIVE_LOCKS_REQUIRED(sched->work_done_wait_mutex)
704 	EXCLUSIVE_LOCKS_REQUIRED(sched->start_work_mutex)
705 {
706 	u64 cpu_usage_0, cpu_usage_1;
707 	struct task_desc *task;
708 	unsigned long i, ret;
709 
710 	sched->start_time = get_nsecs();
711 	sched->cpu_usage = 0;
712 	mutex_unlock(&sched->work_done_wait_mutex);
713 
714 	for (i = 0; i < sched->nr_tasks; i++) {
715 		task = sched->tasks[i];
716 		ret = sem_wait(&task->ready_for_work);
717 		BUG_ON(ret);
718 		sem_init(&task->ready_for_work, 0, 0);
719 	}
720 	mutex_lock(&sched->work_done_wait_mutex);
721 
722 	cpu_usage_0 = get_cpu_usage_nsec_parent();
723 
724 	mutex_unlock(&sched->start_work_mutex);
725 
726 	for (i = 0; i < sched->nr_tasks; i++) {
727 		task = sched->tasks[i];
728 		ret = sem_wait(&task->work_done_sem);
729 		BUG_ON(ret);
730 		sem_init(&task->work_done_sem, 0, 0);
731 		sched->cpu_usage += task->cpu_usage;
732 		task->cpu_usage = 0;
733 	}
734 
735 	cpu_usage_1 = get_cpu_usage_nsec_parent();
736 	if (!sched->runavg_cpu_usage)
737 		sched->runavg_cpu_usage = sched->cpu_usage;
738 	sched->runavg_cpu_usage = (sched->runavg_cpu_usage * (sched->replay_repeat - 1) + sched->cpu_usage) / sched->replay_repeat;
739 
740 	sched->parent_cpu_usage = cpu_usage_1 - cpu_usage_0;
741 	if (!sched->runavg_parent_cpu_usage)
742 		sched->runavg_parent_cpu_usage = sched->parent_cpu_usage;
743 	sched->runavg_parent_cpu_usage = (sched->runavg_parent_cpu_usage * (sched->replay_repeat - 1) +
744 					 sched->parent_cpu_usage)/sched->replay_repeat;
745 
746 	mutex_lock(&sched->start_work_mutex);
747 
748 	for (i = 0; i < sched->nr_tasks; i++) {
749 		task = sched->tasks[i];
750 		task->curr_event = 0;
751 	}
752 }
753 
754 static void run_one_test(struct perf_sched *sched)
755 	EXCLUSIVE_LOCKS_REQUIRED(sched->work_done_wait_mutex)
756 	EXCLUSIVE_LOCKS_REQUIRED(sched->start_work_mutex)
757 {
758 	u64 T0, T1, delta, avg_delta, fluct;
759 
760 	T0 = get_nsecs();
761 	wait_for_tasks(sched);
762 	T1 = get_nsecs();
763 
764 	delta = T1 - T0;
765 	sched->sum_runtime += delta;
766 	sched->nr_runs++;
767 
768 	avg_delta = sched->sum_runtime / sched->nr_runs;
769 	if (delta < avg_delta)
770 		fluct = avg_delta - delta;
771 	else
772 		fluct = delta - avg_delta;
773 	sched->sum_fluct += fluct;
774 	if (!sched->run_avg)
775 		sched->run_avg = delta;
776 	sched->run_avg = (sched->run_avg * (sched->replay_repeat - 1) + delta) / sched->replay_repeat;
777 
778 	printf("#%-3ld: %0.3f, ", sched->nr_runs, (double)delta / NSEC_PER_MSEC);
779 
780 	printf("ravg: %0.2f, ", (double)sched->run_avg / NSEC_PER_MSEC);
781 
782 	printf("cpu: %0.2f / %0.2f",
783 		(double)sched->cpu_usage / NSEC_PER_MSEC, (double)sched->runavg_cpu_usage / NSEC_PER_MSEC);
784 
785 #if 0
786 	/*
787 	 * rusage statistics done by the parent, these are less
788 	 * accurate than the sched->sum_exec_runtime based statistics:
789 	 */
790 	printf(" [%0.2f / %0.2f]",
791 		(double)sched->parent_cpu_usage / NSEC_PER_MSEC,
792 		(double)sched->runavg_parent_cpu_usage / NSEC_PER_MSEC);
793 #endif
794 
795 	printf("\n");
796 
797 	if (sched->nr_sleep_corrections)
798 		printf(" (%ld sleep corrections)\n", sched->nr_sleep_corrections);
799 	sched->nr_sleep_corrections = 0;
800 }
801 
802 static void test_calibrations(struct perf_sched *sched)
803 {
804 	u64 T0, T1;
805 
806 	T0 = get_nsecs();
807 	burn_nsecs(sched, NSEC_PER_MSEC);
808 	T1 = get_nsecs();
809 
810 	printf("the run test took %" PRIu64 " nsecs\n", T1 - T0);
811 
812 	T0 = get_nsecs();
813 	sleep_nsecs(NSEC_PER_MSEC);
814 	T1 = get_nsecs();
815 
816 	printf("the sleep test took %" PRIu64 " nsecs\n", T1 - T0);
817 }
818 
819 static int
820 replay_wakeup_event(struct perf_sched *sched,
821 		    struct evsel *evsel, struct perf_sample *sample,
822 		    struct machine *machine __maybe_unused)
823 {
824 	const char *comm = evsel__strval(evsel, sample, "comm");
825 	const u32 pid	 = evsel__intval(evsel, sample, "pid");
826 	struct task_desc *waker, *wakee;
827 
828 	if (verbose > 0) {
829 		printf("sched_wakeup event %p\n", evsel);
830 
831 		printf(" ... pid %d woke up %s/%d\n", sample->tid, comm, pid);
832 	}
833 
834 	waker = register_pid(sched, sample->tid, "<unknown>");
835 	wakee = register_pid(sched, pid, comm);
836 
837 	add_sched_event_wakeup(sched, waker, sample->time, wakee);
838 	return 0;
839 }
840 
841 static int replay_switch_event(struct perf_sched *sched,
842 			       struct evsel *evsel,
843 			       struct perf_sample *sample,
844 			       struct machine *machine __maybe_unused)
845 {
846 	const char *prev_comm  = evsel__strval(evsel, sample, "prev_comm"),
847 		   *next_comm  = evsel__strval(evsel, sample, "next_comm");
848 	const u32 prev_pid = evsel__intval(evsel, sample, "prev_pid"),
849 		  next_pid = evsel__intval(evsel, sample, "next_pid");
850 	struct task_desc *prev, __maybe_unused *next;
851 	u64 timestamp0, timestamp = sample->time;
852 	int cpu = sample->cpu;
853 	s64 delta;
854 
855 	if (verbose > 0)
856 		printf("sched_switch event %p\n", evsel);
857 
858 	if (cpu >= MAX_CPUS || cpu < 0)
859 		return 0;
860 
861 	timestamp0 = sched->cpu_last_switched[cpu];
862 	if (timestamp0)
863 		delta = timestamp - timestamp0;
864 	else
865 		delta = 0;
866 
867 	if (delta < 0) {
868 		pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta);
869 		return -1;
870 	}
871 
872 	pr_debug(" ... switch from %s/%d to %s/%d [ran %" PRIu64 " nsecs]\n",
873 		 prev_comm, prev_pid, next_comm, next_pid, delta);
874 
875 	prev = register_pid(sched, prev_pid, prev_comm);
876 	next = register_pid(sched, next_pid, next_comm);
877 
878 	sched->cpu_last_switched[cpu] = timestamp;
879 
880 	add_sched_event_run(sched, prev, timestamp, delta);
881 	add_sched_event_sleep(sched, prev, timestamp);
882 
883 	return 0;
884 }
885 
886 static int replay_fork_event(struct perf_sched *sched,
887 			     union perf_event *event,
888 			     struct machine *machine)
889 {
890 	struct thread *child, *parent;
891 
892 	child = machine__findnew_thread(machine, event->fork.pid,
893 					event->fork.tid);
894 	parent = machine__findnew_thread(machine, event->fork.ppid,
895 					 event->fork.ptid);
896 
897 	if (child == NULL || parent == NULL) {
898 		pr_debug("thread does not exist on fork event: child %p, parent %p\n",
899 				 child, parent);
900 		goto out_put;
901 	}
902 
903 	if (verbose > 0) {
904 		printf("fork event\n");
905 		printf("... parent: %s/%d\n", thread__comm_str(parent), thread__tid(parent));
906 		printf("...  child: %s/%d\n", thread__comm_str(child), thread__tid(child));
907 	}
908 
909 	register_pid(sched, thread__tid(parent), thread__comm_str(parent));
910 	register_pid(sched, thread__tid(child), thread__comm_str(child));
911 out_put:
912 	thread__put(child);
913 	thread__put(parent);
914 	return 0;
915 }
916 
917 struct sort_dimension {
918 	const char		*name;
919 	sort_fn_t		cmp;
920 	struct list_head	list;
921 };
922 
923 static inline void init_prio(struct thread_runtime *r)
924 {
925 	r->prio = -1;
926 }
927 
928 /*
929  * handle runtime stats saved per thread
930  */
931 static struct thread_runtime *thread__init_runtime(struct thread *thread)
932 {
933 	struct thread_runtime *r;
934 
935 	r = zalloc(sizeof(struct thread_runtime));
936 	if (!r)
937 		return NULL;
938 
939 	init_stats(&r->run_stats);
940 	init_prio(r);
941 	thread__set_priv(thread, r);
942 
943 	return r;
944 }
945 
946 static struct thread_runtime *thread__get_runtime(struct thread *thread)
947 {
948 	struct thread_runtime *tr;
949 
950 	tr = thread__priv(thread);
951 	if (tr == NULL) {
952 		tr = thread__init_runtime(thread);
953 		if (tr == NULL)
954 			pr_debug("Failed to malloc memory for runtime data.\n");
955 	}
956 
957 	return tr;
958 }
959 
960 static int
961 thread_lat_cmp(struct list_head *list, struct work_atoms *l, struct work_atoms *r)
962 {
963 	struct sort_dimension *sort;
964 	int ret = 0;
965 
966 	BUG_ON(list_empty(list));
967 
968 	list_for_each_entry(sort, list, list) {
969 		ret = sort->cmp(l, r);
970 		if (ret)
971 			return ret;
972 	}
973 
974 	return ret;
975 }
976 
977 static struct work_atoms *
978 thread_atoms_search(struct rb_root_cached *root, struct thread *thread,
979 			 struct list_head *sort_list)
980 {
981 	struct rb_node *node = root->rb_root.rb_node;
982 	struct work_atoms key = { .thread = thread };
983 
984 	while (node) {
985 		struct work_atoms *atoms;
986 		int cmp;
987 
988 		atoms = container_of(node, struct work_atoms, node);
989 
990 		cmp = thread_lat_cmp(sort_list, &key, atoms);
991 		if (cmp > 0)
992 			node = node->rb_left;
993 		else if (cmp < 0)
994 			node = node->rb_right;
995 		else {
996 			BUG_ON(thread != atoms->thread);
997 			return atoms;
998 		}
999 	}
1000 	return NULL;
1001 }
1002 
1003 static void
1004 __thread_latency_insert(struct rb_root_cached *root, struct work_atoms *data,
1005 			 struct list_head *sort_list)
1006 {
1007 	struct rb_node **new = &(root->rb_root.rb_node), *parent = NULL;
1008 	bool leftmost = true;
1009 
1010 	while (*new) {
1011 		struct work_atoms *this;
1012 		int cmp;
1013 
1014 		this = container_of(*new, struct work_atoms, node);
1015 		parent = *new;
1016 
1017 		cmp = thread_lat_cmp(sort_list, data, this);
1018 
1019 		if (cmp > 0)
1020 			new = &((*new)->rb_left);
1021 		else {
1022 			new = &((*new)->rb_right);
1023 			leftmost = false;
1024 		}
1025 	}
1026 
1027 	rb_link_node(&data->node, parent, new);
1028 	rb_insert_color_cached(&data->node, root, leftmost);
1029 }
1030 
1031 static int thread_atoms_insert(struct perf_sched *sched, struct thread *thread)
1032 {
1033 	struct work_atoms *atoms = zalloc(sizeof(*atoms));
1034 	if (!atoms) {
1035 		pr_err("No memory at %s\n", __func__);
1036 		return -1;
1037 	}
1038 
1039 	atoms->thread = thread__get(thread);
1040 	INIT_LIST_HEAD(&atoms->work_list);
1041 	__thread_latency_insert(&sched->atom_root, atoms, &sched->cmp_pid);
1042 	return 0;
1043 }
1044 
1045 static int
1046 add_sched_out_event(struct work_atoms *atoms,
1047 		    char run_state,
1048 		    u64 timestamp)
1049 {
1050 	struct work_atom *atom = zalloc(sizeof(*atom));
1051 	if (!atom) {
1052 		pr_err("Non memory at %s", __func__);
1053 		return -1;
1054 	}
1055 
1056 	atom->sched_out_time = timestamp;
1057 
1058 	if (run_state == 'R') {
1059 		atom->state = THREAD_WAIT_CPU;
1060 		atom->wake_up_time = atom->sched_out_time;
1061 	}
1062 
1063 	list_add_tail(&atom->list, &atoms->work_list);
1064 	return 0;
1065 }
1066 
1067 static void
1068 add_runtime_event(struct work_atoms *atoms, u64 delta,
1069 		  u64 timestamp __maybe_unused)
1070 {
1071 	struct work_atom *atom;
1072 
1073 	BUG_ON(list_empty(&atoms->work_list));
1074 
1075 	atom = list_entry(atoms->work_list.prev, struct work_atom, list);
1076 
1077 	atom->runtime += delta;
1078 	atoms->total_runtime += delta;
1079 }
1080 
1081 static void
1082 add_sched_in_event(struct work_atoms *atoms, u64 timestamp)
1083 {
1084 	struct work_atom *atom;
1085 	u64 delta;
1086 
1087 	if (list_empty(&atoms->work_list))
1088 		return;
1089 
1090 	atom = list_entry(atoms->work_list.prev, struct work_atom, list);
1091 
1092 	if (atom->state != THREAD_WAIT_CPU)
1093 		return;
1094 
1095 	if (timestamp < atom->wake_up_time) {
1096 		atom->state = THREAD_IGNORE;
1097 		return;
1098 	}
1099 
1100 	atom->state = THREAD_SCHED_IN;
1101 	atom->sched_in_time = timestamp;
1102 
1103 	delta = atom->sched_in_time - atom->wake_up_time;
1104 	atoms->total_lat += delta;
1105 	if (delta > atoms->max_lat) {
1106 		atoms->max_lat = delta;
1107 		atoms->max_lat_start = atom->wake_up_time;
1108 		atoms->max_lat_end = timestamp;
1109 	}
1110 	atoms->nb_atoms++;
1111 }
1112 
1113 static int latency_switch_event(struct perf_sched *sched,
1114 				struct evsel *evsel,
1115 				struct perf_sample *sample,
1116 				struct machine *machine)
1117 {
1118 	const u32 prev_pid = evsel__intval(evsel, sample, "prev_pid"),
1119 		  next_pid = evsel__intval(evsel, sample, "next_pid");
1120 	const char prev_state = evsel__taskstate(evsel, sample, "prev_state");
1121 	struct work_atoms *out_events, *in_events;
1122 	struct thread *sched_out, *sched_in;
1123 	u64 timestamp0, timestamp = sample->time;
1124 	int cpu = sample->cpu, err = -1;
1125 	s64 delta;
1126 
1127 	BUG_ON(cpu >= MAX_CPUS || cpu < 0);
1128 
1129 	timestamp0 = sched->cpu_last_switched[cpu];
1130 	sched->cpu_last_switched[cpu] = timestamp;
1131 	if (timestamp0)
1132 		delta = timestamp - timestamp0;
1133 	else
1134 		delta = 0;
1135 
1136 	if (delta < 0) {
1137 		pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta);
1138 		return -1;
1139 	}
1140 
1141 	sched_out = machine__findnew_thread(machine, -1, prev_pid);
1142 	sched_in = machine__findnew_thread(machine, -1, next_pid);
1143 	if (sched_out == NULL || sched_in == NULL)
1144 		goto out_put;
1145 
1146 	out_events = thread_atoms_search(&sched->atom_root, sched_out, &sched->cmp_pid);
1147 	if (!out_events) {
1148 		if (thread_atoms_insert(sched, sched_out))
1149 			goto out_put;
1150 		out_events = thread_atoms_search(&sched->atom_root, sched_out, &sched->cmp_pid);
1151 		if (!out_events) {
1152 			pr_err("out-event: Internal tree error");
1153 			goto out_put;
1154 		}
1155 	}
1156 	if (add_sched_out_event(out_events, prev_state, timestamp))
1157 		return -1;
1158 
1159 	in_events = thread_atoms_search(&sched->atom_root, sched_in, &sched->cmp_pid);
1160 	if (!in_events) {
1161 		if (thread_atoms_insert(sched, sched_in))
1162 			goto out_put;
1163 		in_events = thread_atoms_search(&sched->atom_root, sched_in, &sched->cmp_pid);
1164 		if (!in_events) {
1165 			pr_err("in-event: Internal tree error");
1166 			goto out_put;
1167 		}
1168 		/*
1169 		 * Take came in we have not heard about yet,
1170 		 * add in an initial atom in runnable state:
1171 		 */
1172 		if (add_sched_out_event(in_events, 'R', timestamp))
1173 			goto out_put;
1174 	}
1175 	add_sched_in_event(in_events, timestamp);
1176 	err = 0;
1177 out_put:
1178 	thread__put(sched_out);
1179 	thread__put(sched_in);
1180 	return err;
1181 }
1182 
1183 static int latency_runtime_event(struct perf_sched *sched,
1184 				 struct evsel *evsel,
1185 				 struct perf_sample *sample,
1186 				 struct machine *machine)
1187 {
1188 	const u32 pid	   = evsel__intval(evsel, sample, "pid");
1189 	const u64 runtime  = evsel__intval(evsel, sample, "runtime");
1190 	struct thread *thread = machine__findnew_thread(machine, -1, pid);
1191 	struct work_atoms *atoms = thread_atoms_search(&sched->atom_root, thread, &sched->cmp_pid);
1192 	u64 timestamp = sample->time;
1193 	int cpu = sample->cpu, err = -1;
1194 
1195 	if (thread == NULL)
1196 		return -1;
1197 
1198 	BUG_ON(cpu >= MAX_CPUS || cpu < 0);
1199 	if (!atoms) {
1200 		if (thread_atoms_insert(sched, thread))
1201 			goto out_put;
1202 		atoms = thread_atoms_search(&sched->atom_root, thread, &sched->cmp_pid);
1203 		if (!atoms) {
1204 			pr_err("in-event: Internal tree error");
1205 			goto out_put;
1206 		}
1207 		if (add_sched_out_event(atoms, 'R', timestamp))
1208 			goto out_put;
1209 	}
1210 
1211 	add_runtime_event(atoms, runtime, timestamp);
1212 	err = 0;
1213 out_put:
1214 	thread__put(thread);
1215 	return err;
1216 }
1217 
1218 static int latency_wakeup_event(struct perf_sched *sched,
1219 				struct evsel *evsel,
1220 				struct perf_sample *sample,
1221 				struct machine *machine)
1222 {
1223 	const u32 pid	  = evsel__intval(evsel, sample, "pid");
1224 	struct work_atoms *atoms;
1225 	struct work_atom *atom;
1226 	struct thread *wakee;
1227 	u64 timestamp = sample->time;
1228 	int err = -1;
1229 
1230 	wakee = machine__findnew_thread(machine, -1, pid);
1231 	if (wakee == NULL)
1232 		return -1;
1233 	atoms = thread_atoms_search(&sched->atom_root, wakee, &sched->cmp_pid);
1234 	if (!atoms) {
1235 		if (thread_atoms_insert(sched, wakee))
1236 			goto out_put;
1237 		atoms = thread_atoms_search(&sched->atom_root, wakee, &sched->cmp_pid);
1238 		if (!atoms) {
1239 			pr_err("wakeup-event: Internal tree error");
1240 			goto out_put;
1241 		}
1242 		if (add_sched_out_event(atoms, 'S', timestamp))
1243 			goto out_put;
1244 	}
1245 
1246 	BUG_ON(list_empty(&atoms->work_list));
1247 
1248 	atom = list_entry(atoms->work_list.prev, struct work_atom, list);
1249 
1250 	/*
1251 	 * As we do not guarantee the wakeup event happens when
1252 	 * task is out of run queue, also may happen when task is
1253 	 * on run queue and wakeup only change ->state to TASK_RUNNING,
1254 	 * then we should not set the ->wake_up_time when wake up a
1255 	 * task which is on run queue.
1256 	 *
1257 	 * You WILL be missing events if you've recorded only
1258 	 * one CPU, or are only looking at only one, so don't
1259 	 * skip in this case.
1260 	 */
1261 	if (sched->profile_cpu == -1 && atom->state != THREAD_SLEEPING)
1262 		goto out_ok;
1263 
1264 	sched->nr_timestamps++;
1265 	if (atom->sched_out_time > timestamp) {
1266 		sched->nr_unordered_timestamps++;
1267 		goto out_ok;
1268 	}
1269 
1270 	atom->state = THREAD_WAIT_CPU;
1271 	atom->wake_up_time = timestamp;
1272 out_ok:
1273 	err = 0;
1274 out_put:
1275 	thread__put(wakee);
1276 	return err;
1277 }
1278 
1279 static int latency_migrate_task_event(struct perf_sched *sched,
1280 				      struct evsel *evsel,
1281 				      struct perf_sample *sample,
1282 				      struct machine *machine)
1283 {
1284 	const u32 pid = evsel__intval(evsel, sample, "pid");
1285 	u64 timestamp = sample->time;
1286 	struct work_atoms *atoms;
1287 	struct work_atom *atom;
1288 	struct thread *migrant;
1289 	int err = -1;
1290 
1291 	/*
1292 	 * Only need to worry about migration when profiling one CPU.
1293 	 */
1294 	if (sched->profile_cpu == -1)
1295 		return 0;
1296 
1297 	migrant = machine__findnew_thread(machine, -1, pid);
1298 	if (migrant == NULL)
1299 		return -1;
1300 	atoms = thread_atoms_search(&sched->atom_root, migrant, &sched->cmp_pid);
1301 	if (!atoms) {
1302 		if (thread_atoms_insert(sched, migrant))
1303 			goto out_put;
1304 		register_pid(sched, thread__tid(migrant), thread__comm_str(migrant));
1305 		atoms = thread_atoms_search(&sched->atom_root, migrant, &sched->cmp_pid);
1306 		if (!atoms) {
1307 			pr_err("migration-event: Internal tree error");
1308 			goto out_put;
1309 		}
1310 		if (add_sched_out_event(atoms, 'R', timestamp))
1311 			goto out_put;
1312 	}
1313 
1314 	BUG_ON(list_empty(&atoms->work_list));
1315 
1316 	atom = list_entry(atoms->work_list.prev, struct work_atom, list);
1317 	atom->sched_in_time = atom->sched_out_time = atom->wake_up_time = timestamp;
1318 
1319 	sched->nr_timestamps++;
1320 
1321 	if (atom->sched_out_time > timestamp)
1322 		sched->nr_unordered_timestamps++;
1323 	err = 0;
1324 out_put:
1325 	thread__put(migrant);
1326 	return err;
1327 }
1328 
1329 static void output_lat_thread(struct perf_sched *sched, struct work_atoms *work_list)
1330 {
1331 	int i;
1332 	int ret;
1333 	u64 avg;
1334 	char max_lat_start[32], max_lat_end[32];
1335 
1336 	if (!work_list->nb_atoms)
1337 		return;
1338 	/*
1339 	 * Ignore idle threads:
1340 	 */
1341 	if (!strcmp(thread__comm_str(work_list->thread), "swapper"))
1342 		return;
1343 
1344 	sched->all_runtime += work_list->total_runtime;
1345 	sched->all_count   += work_list->nb_atoms;
1346 
1347 	if (work_list->num_merged > 1) {
1348 		ret = printf("  %s:(%d) ", thread__comm_str(work_list->thread),
1349 			     work_list->num_merged);
1350 	} else {
1351 		ret = printf("  %s:%d ", thread__comm_str(work_list->thread),
1352 			     thread__tid(work_list->thread));
1353 	}
1354 
1355 	for (i = 0; i < 24 - ret; i++)
1356 		printf(" ");
1357 
1358 	avg = work_list->total_lat / work_list->nb_atoms;
1359 	timestamp__scnprintf_usec(work_list->max_lat_start, max_lat_start, sizeof(max_lat_start));
1360 	timestamp__scnprintf_usec(work_list->max_lat_end, max_lat_end, sizeof(max_lat_end));
1361 
1362 	printf("|%11.3f ms |%9" PRIu64 " | avg:%8.3f ms | max:%8.3f ms | max start: %12s s | max end: %12s s\n",
1363 	      (double)work_list->total_runtime / NSEC_PER_MSEC,
1364 		 work_list->nb_atoms, (double)avg / NSEC_PER_MSEC,
1365 		 (double)work_list->max_lat / NSEC_PER_MSEC,
1366 		 max_lat_start, max_lat_end);
1367 }
1368 
1369 static int pid_cmp(struct work_atoms *l, struct work_atoms *r)
1370 {
1371 	pid_t l_tid, r_tid;
1372 
1373 	if (RC_CHK_EQUAL(l->thread, r->thread))
1374 		return 0;
1375 	l_tid = thread__tid(l->thread);
1376 	r_tid = thread__tid(r->thread);
1377 	if (l_tid < r_tid)
1378 		return -1;
1379 	if (l_tid > r_tid)
1380 		return 1;
1381 	return (int)(RC_CHK_ACCESS(l->thread) - RC_CHK_ACCESS(r->thread));
1382 }
1383 
1384 static int avg_cmp(struct work_atoms *l, struct work_atoms *r)
1385 {
1386 	u64 avgl, avgr;
1387 
1388 	if (!l->nb_atoms)
1389 		return -1;
1390 
1391 	if (!r->nb_atoms)
1392 		return 1;
1393 
1394 	avgl = l->total_lat / l->nb_atoms;
1395 	avgr = r->total_lat / r->nb_atoms;
1396 
1397 	if (avgl < avgr)
1398 		return -1;
1399 	if (avgl > avgr)
1400 		return 1;
1401 
1402 	return 0;
1403 }
1404 
1405 static int max_cmp(struct work_atoms *l, struct work_atoms *r)
1406 {
1407 	if (l->max_lat < r->max_lat)
1408 		return -1;
1409 	if (l->max_lat > r->max_lat)
1410 		return 1;
1411 
1412 	return 0;
1413 }
1414 
1415 static int switch_cmp(struct work_atoms *l, struct work_atoms *r)
1416 {
1417 	if (l->nb_atoms < r->nb_atoms)
1418 		return -1;
1419 	if (l->nb_atoms > r->nb_atoms)
1420 		return 1;
1421 
1422 	return 0;
1423 }
1424 
1425 static int runtime_cmp(struct work_atoms *l, struct work_atoms *r)
1426 {
1427 	if (l->total_runtime < r->total_runtime)
1428 		return -1;
1429 	if (l->total_runtime > r->total_runtime)
1430 		return 1;
1431 
1432 	return 0;
1433 }
1434 
1435 static int sort_dimension__add(const char *tok, struct list_head *list)
1436 {
1437 	size_t i;
1438 	static struct sort_dimension avg_sort_dimension = {
1439 		.name = "avg",
1440 		.cmp  = avg_cmp,
1441 	};
1442 	static struct sort_dimension max_sort_dimension = {
1443 		.name = "max",
1444 		.cmp  = max_cmp,
1445 	};
1446 	static struct sort_dimension pid_sort_dimension = {
1447 		.name = "pid",
1448 		.cmp  = pid_cmp,
1449 	};
1450 	static struct sort_dimension runtime_sort_dimension = {
1451 		.name = "runtime",
1452 		.cmp  = runtime_cmp,
1453 	};
1454 	static struct sort_dimension switch_sort_dimension = {
1455 		.name = "switch",
1456 		.cmp  = switch_cmp,
1457 	};
1458 	struct sort_dimension *available_sorts[] = {
1459 		&pid_sort_dimension,
1460 		&avg_sort_dimension,
1461 		&max_sort_dimension,
1462 		&switch_sort_dimension,
1463 		&runtime_sort_dimension,
1464 	};
1465 
1466 	for (i = 0; i < ARRAY_SIZE(available_sorts); i++) {
1467 		if (!strcmp(available_sorts[i]->name, tok)) {
1468 			list_add_tail(&available_sorts[i]->list, list);
1469 
1470 			return 0;
1471 		}
1472 	}
1473 
1474 	return -1;
1475 }
1476 
1477 static void perf_sched__sort_lat(struct perf_sched *sched)
1478 {
1479 	struct rb_node *node;
1480 	struct rb_root_cached *root = &sched->atom_root;
1481 again:
1482 	for (;;) {
1483 		struct work_atoms *data;
1484 		node = rb_first_cached(root);
1485 		if (!node)
1486 			break;
1487 
1488 		rb_erase_cached(node, root);
1489 		data = rb_entry(node, struct work_atoms, node);
1490 		__thread_latency_insert(&sched->sorted_atom_root, data, &sched->sort_list);
1491 	}
1492 	if (root == &sched->atom_root) {
1493 		root = &sched->merged_atom_root;
1494 		goto again;
1495 	}
1496 }
1497 
1498 static int process_sched_wakeup_event(const struct perf_tool *tool,
1499 				      struct evsel *evsel,
1500 				      struct perf_sample *sample,
1501 				      struct machine *machine)
1502 {
1503 	struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
1504 
1505 	if (sched->tp_handler->wakeup_event)
1506 		return sched->tp_handler->wakeup_event(sched, evsel, sample, machine);
1507 
1508 	return 0;
1509 }
1510 
1511 static int process_sched_wakeup_ignore(const struct perf_tool *tool __maybe_unused,
1512 				      struct evsel *evsel __maybe_unused,
1513 				      struct perf_sample *sample __maybe_unused,
1514 				      struct machine *machine __maybe_unused)
1515 {
1516 	return 0;
1517 }
1518 
1519 union map_priv {
1520 	void	*ptr;
1521 	bool	 color;
1522 };
1523 
1524 static bool thread__has_color(struct thread *thread)
1525 {
1526 	union map_priv priv = {
1527 		.ptr = thread__priv(thread),
1528 	};
1529 
1530 	return priv.color;
1531 }
1532 
1533 static struct thread*
1534 map__findnew_thread(struct perf_sched *sched, struct machine *machine, pid_t pid, pid_t tid)
1535 {
1536 	struct thread *thread = machine__findnew_thread(machine, pid, tid);
1537 	union map_priv priv = {
1538 		.color = false,
1539 	};
1540 
1541 	if (!sched->map.color_pids || !thread || thread__priv(thread))
1542 		return thread;
1543 
1544 	if (thread_map__has(sched->map.color_pids, tid))
1545 		priv.color = true;
1546 
1547 	thread__set_priv(thread, priv.ptr);
1548 	return thread;
1549 }
1550 
1551 static bool sched_match_task(struct perf_sched *sched, const char *comm_str)
1552 {
1553 	bool fuzzy_match = sched->map.fuzzy;
1554 	struct strlist *task_names = sched->map.task_names;
1555 	struct str_node *node;
1556 
1557 	strlist__for_each_entry(node, task_names) {
1558 		bool match_found = fuzzy_match ? !!strstr(comm_str, node->s) :
1559 							!strcmp(comm_str, node->s);
1560 		if (match_found)
1561 			return true;
1562 	}
1563 
1564 	return false;
1565 }
1566 
1567 static void print_sched_map(struct perf_sched *sched, struct perf_cpu this_cpu, int cpus_nr,
1568 								const char *color, bool sched_out)
1569 {
1570 	for (int i = 0; i < cpus_nr; i++) {
1571 		struct perf_cpu cpu = {
1572 			.cpu = sched->map.comp ? sched->map.comp_cpus[i].cpu : i,
1573 		};
1574 		struct thread *curr_thread = sched->curr_thread[cpu.cpu];
1575 		struct thread *curr_out_thread = sched->curr_out_thread[cpu.cpu];
1576 		struct thread_runtime *curr_tr;
1577 		const char *pid_color = color;
1578 		const char *cpu_color = color;
1579 		char symbol = ' ';
1580 		struct thread *thread_to_check = sched_out ? curr_out_thread : curr_thread;
1581 
1582 		if (thread_to_check && thread__has_color(thread_to_check))
1583 			pid_color = COLOR_PIDS;
1584 
1585 		if (sched->map.color_cpus && perf_cpu_map__has(sched->map.color_cpus, cpu))
1586 			cpu_color = COLOR_CPUS;
1587 
1588 		if (cpu.cpu == this_cpu.cpu)
1589 			symbol = '*';
1590 
1591 		color_fprintf(stdout, cpu.cpu != this_cpu.cpu ? color : cpu_color, "%c", symbol);
1592 
1593 		thread_to_check = sched_out ? sched->curr_out_thread[cpu.cpu] :
1594 								sched->curr_thread[cpu.cpu];
1595 
1596 		if (thread_to_check) {
1597 			curr_tr = thread__get_runtime(thread_to_check);
1598 			if (curr_tr == NULL)
1599 				return;
1600 
1601 			if (sched_out) {
1602 				if (cpu.cpu == this_cpu.cpu)
1603 					color_fprintf(stdout, color, "-  ");
1604 				else {
1605 					curr_tr = thread__get_runtime(sched->curr_thread[cpu.cpu]);
1606 					if (curr_tr != NULL)
1607 						color_fprintf(stdout, pid_color, "%2s ",
1608 										curr_tr->shortname);
1609 				}
1610 			} else
1611 				color_fprintf(stdout, pid_color, "%2s ", curr_tr->shortname);
1612 		} else
1613 			color_fprintf(stdout, color, "   ");
1614 	}
1615 }
1616 
1617 static int map_switch_event(struct perf_sched *sched, struct evsel *evsel,
1618 			    struct perf_sample *sample, struct machine *machine)
1619 {
1620 	const u32 next_pid = evsel__intval(evsel, sample, "next_pid");
1621 	const u32 prev_pid = evsel__intval(evsel, sample, "prev_pid");
1622 	struct thread *sched_in, *sched_out;
1623 	struct thread_runtime *tr;
1624 	int new_shortname;
1625 	u64 timestamp0, timestamp = sample->time;
1626 	s64 delta;
1627 	struct perf_cpu this_cpu = {
1628 		.cpu = sample->cpu,
1629 	};
1630 	int cpus_nr;
1631 	int proceed;
1632 	bool new_cpu = false;
1633 	const char *color = PERF_COLOR_NORMAL;
1634 	char stimestamp[32];
1635 	const char *str;
1636 
1637 	BUG_ON(this_cpu.cpu >= MAX_CPUS || this_cpu.cpu < 0);
1638 
1639 	if (this_cpu.cpu > sched->max_cpu.cpu)
1640 		sched->max_cpu = this_cpu;
1641 
1642 	if (sched->map.comp) {
1643 		cpus_nr = bitmap_weight(sched->map.comp_cpus_mask, MAX_CPUS);
1644 		if (!__test_and_set_bit(this_cpu.cpu, sched->map.comp_cpus_mask)) {
1645 			sched->map.comp_cpus[cpus_nr++] = this_cpu;
1646 			new_cpu = true;
1647 		}
1648 	} else
1649 		cpus_nr = sched->max_cpu.cpu;
1650 
1651 	timestamp0 = sched->cpu_last_switched[this_cpu.cpu];
1652 	sched->cpu_last_switched[this_cpu.cpu] = timestamp;
1653 	if (timestamp0)
1654 		delta = timestamp - timestamp0;
1655 	else
1656 		delta = 0;
1657 
1658 	if (delta < 0) {
1659 		pr_err("hm, delta: %" PRIu64 " < 0 ?\n", delta);
1660 		return -1;
1661 	}
1662 
1663 	sched_in = map__findnew_thread(sched, machine, -1, next_pid);
1664 	sched_out = map__findnew_thread(sched, machine, -1, prev_pid);
1665 	if (sched_in == NULL || sched_out == NULL)
1666 		return -1;
1667 
1668 	tr = thread__get_runtime(sched_in);
1669 	if (tr == NULL) {
1670 		thread__put(sched_in);
1671 		return -1;
1672 	}
1673 
1674 	sched->curr_thread[this_cpu.cpu] = thread__get(sched_in);
1675 	sched->curr_out_thread[this_cpu.cpu] = thread__get(sched_out);
1676 
1677 	str = thread__comm_str(sched_in);
1678 	new_shortname = 0;
1679 	if (!tr->shortname[0]) {
1680 		if (!strcmp(thread__comm_str(sched_in), "swapper")) {
1681 			/*
1682 			 * Don't allocate a letter-number for swapper:0
1683 			 * as a shortname. Instead, we use '.' for it.
1684 			 */
1685 			tr->shortname[0] = '.';
1686 			tr->shortname[1] = ' ';
1687 		} else if (!sched->map.task_name || sched_match_task(sched, str)) {
1688 			tr->shortname[0] = sched->next_shortname1;
1689 			tr->shortname[1] = sched->next_shortname2;
1690 
1691 			if (sched->next_shortname1 < 'Z') {
1692 				sched->next_shortname1++;
1693 			} else {
1694 				sched->next_shortname1 = 'A';
1695 				if (sched->next_shortname2 < '9')
1696 					sched->next_shortname2++;
1697 				else
1698 					sched->next_shortname2 = '0';
1699 			}
1700 		} else {
1701 			tr->shortname[0] = '-';
1702 			tr->shortname[1] = ' ';
1703 		}
1704 		new_shortname = 1;
1705 	}
1706 
1707 	if (sched->map.cpus && !perf_cpu_map__has(sched->map.cpus, this_cpu))
1708 		goto out;
1709 
1710 	proceed = 0;
1711 	str = thread__comm_str(sched_in);
1712 	/*
1713 	 * Check which of sched_in and sched_out matches the passed --task-name
1714 	 * arguments and call the corresponding print_sched_map.
1715 	 */
1716 	if (sched->map.task_name && !sched_match_task(sched, str)) {
1717 		if (!sched_match_task(sched, thread__comm_str(sched_out)))
1718 			goto out;
1719 		else
1720 			goto sched_out;
1721 
1722 	} else {
1723 		str = thread__comm_str(sched_out);
1724 		if (!(sched->map.task_name && !sched_match_task(sched, str)))
1725 			proceed = 1;
1726 	}
1727 
1728 	printf("  ");
1729 
1730 	print_sched_map(sched, this_cpu, cpus_nr, color, false);
1731 
1732 	timestamp__scnprintf_usec(timestamp, stimestamp, sizeof(stimestamp));
1733 	color_fprintf(stdout, color, "  %12s secs ", stimestamp);
1734 	if (new_shortname || tr->comm_changed || (verbose > 0 && thread__tid(sched_in))) {
1735 		const char *pid_color = color;
1736 
1737 		if (thread__has_color(sched_in))
1738 			pid_color = COLOR_PIDS;
1739 
1740 		color_fprintf(stdout, pid_color, "%s => %s:%d",
1741 			tr->shortname, thread__comm_str(sched_in), thread__tid(sched_in));
1742 		tr->comm_changed = false;
1743 	}
1744 
1745 	if (sched->map.comp && new_cpu)
1746 		color_fprintf(stdout, color, " (CPU %d)", this_cpu.cpu);
1747 
1748 	if (proceed != 1) {
1749 		color_fprintf(stdout, color, "\n");
1750 		goto out;
1751 	}
1752 
1753 sched_out:
1754 	if (sched->map.task_name) {
1755 		tr = thread__get_runtime(sched->curr_out_thread[this_cpu.cpu]);
1756 		if (strcmp(tr->shortname, "") == 0)
1757 			goto out;
1758 
1759 		if (proceed == 1)
1760 			color_fprintf(stdout, color, "\n");
1761 
1762 		printf("  ");
1763 		print_sched_map(sched, this_cpu, cpus_nr, color, true);
1764 		timestamp__scnprintf_usec(timestamp, stimestamp, sizeof(stimestamp));
1765 		color_fprintf(stdout, color, "  %12s secs ", stimestamp);
1766 	}
1767 
1768 	color_fprintf(stdout, color, "\n");
1769 
1770 out:
1771 	if (sched->map.task_name)
1772 		thread__put(sched_out);
1773 
1774 	thread__put(sched_in);
1775 
1776 	return 0;
1777 }
1778 
1779 static int process_sched_switch_event(const struct perf_tool *tool,
1780 				      struct evsel *evsel,
1781 				      struct perf_sample *sample,
1782 				      struct machine *machine)
1783 {
1784 	struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
1785 	int this_cpu = sample->cpu, err = 0;
1786 	u32 prev_pid = evsel__intval(evsel, sample, "prev_pid"),
1787 	    next_pid = evsel__intval(evsel, sample, "next_pid");
1788 
1789 	if (sched->curr_pid[this_cpu] != (u32)-1) {
1790 		/*
1791 		 * Are we trying to switch away a PID that is
1792 		 * not current?
1793 		 */
1794 		if (sched->curr_pid[this_cpu] != prev_pid)
1795 			sched->nr_context_switch_bugs++;
1796 	}
1797 
1798 	if (sched->tp_handler->switch_event)
1799 		err = sched->tp_handler->switch_event(sched, evsel, sample, machine);
1800 
1801 	sched->curr_pid[this_cpu] = next_pid;
1802 	return err;
1803 }
1804 
1805 static int process_sched_runtime_event(const struct perf_tool *tool,
1806 				       struct evsel *evsel,
1807 				       struct perf_sample *sample,
1808 				       struct machine *machine)
1809 {
1810 	struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
1811 
1812 	if (sched->tp_handler->runtime_event)
1813 		return sched->tp_handler->runtime_event(sched, evsel, sample, machine);
1814 
1815 	return 0;
1816 }
1817 
1818 static int perf_sched__process_fork_event(const struct perf_tool *tool,
1819 					  union perf_event *event,
1820 					  struct perf_sample *sample,
1821 					  struct machine *machine)
1822 {
1823 	struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
1824 
1825 	/* run the fork event through the perf machinery */
1826 	perf_event__process_fork(tool, event, sample, machine);
1827 
1828 	/* and then run additional processing needed for this command */
1829 	if (sched->tp_handler->fork_event)
1830 		return sched->tp_handler->fork_event(sched, event, machine);
1831 
1832 	return 0;
1833 }
1834 
1835 static int process_sched_migrate_task_event(const struct perf_tool *tool,
1836 					    struct evsel *evsel,
1837 					    struct perf_sample *sample,
1838 					    struct machine *machine)
1839 {
1840 	struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
1841 
1842 	if (sched->tp_handler->migrate_task_event)
1843 		return sched->tp_handler->migrate_task_event(sched, evsel, sample, machine);
1844 
1845 	return 0;
1846 }
1847 
1848 typedef int (*tracepoint_handler)(const struct perf_tool *tool,
1849 				  struct evsel *evsel,
1850 				  struct perf_sample *sample,
1851 				  struct machine *machine);
1852 
1853 static int perf_sched__process_tracepoint_sample(const struct perf_tool *tool __maybe_unused,
1854 						 union perf_event *event __maybe_unused,
1855 						 struct perf_sample *sample,
1856 						 struct evsel *evsel,
1857 						 struct machine *machine)
1858 {
1859 	int err = 0;
1860 
1861 	if (evsel->handler != NULL) {
1862 		tracepoint_handler f = evsel->handler;
1863 		err = f(tool, evsel, sample, machine);
1864 	}
1865 
1866 	return err;
1867 }
1868 
1869 static int perf_sched__process_comm(const struct perf_tool *tool __maybe_unused,
1870 				    union perf_event *event,
1871 				    struct perf_sample *sample,
1872 				    struct machine *machine)
1873 {
1874 	struct thread *thread;
1875 	struct thread_runtime *tr;
1876 	int err;
1877 
1878 	err = perf_event__process_comm(tool, event, sample, machine);
1879 	if (err)
1880 		return err;
1881 
1882 	thread = machine__find_thread(machine, sample->pid, sample->tid);
1883 	if (!thread) {
1884 		pr_err("Internal error: can't find thread\n");
1885 		return -1;
1886 	}
1887 
1888 	tr = thread__get_runtime(thread);
1889 	if (tr == NULL) {
1890 		thread__put(thread);
1891 		return -1;
1892 	}
1893 
1894 	tr->comm_changed = true;
1895 	thread__put(thread);
1896 
1897 	return 0;
1898 }
1899 
1900 static int perf_sched__read_events(struct perf_sched *sched)
1901 {
1902 	struct evsel_str_handler handlers[] = {
1903 		{ "sched:sched_switch",	      process_sched_switch_event, },
1904 		{ "sched:sched_stat_runtime", process_sched_runtime_event, },
1905 		{ "sched:sched_wakeup",	      process_sched_wakeup_event, },
1906 		{ "sched:sched_waking",	      process_sched_wakeup_event, },
1907 		{ "sched:sched_wakeup_new",   process_sched_wakeup_event, },
1908 		{ "sched:sched_migrate_task", process_sched_migrate_task_event, },
1909 	};
1910 	struct perf_session *session;
1911 	struct perf_data data = {
1912 		.path  = input_name,
1913 		.mode  = PERF_DATA_MODE_READ,
1914 		.force = sched->force,
1915 	};
1916 	int rc = -1;
1917 
1918 	session = perf_session__new(&data, &sched->tool);
1919 	if (IS_ERR(session)) {
1920 		pr_debug("Error creating perf session");
1921 		return PTR_ERR(session);
1922 	}
1923 
1924 	symbol__init(&session->header.env);
1925 
1926 	/* prefer sched_waking if it is captured */
1927 	if (evlist__find_tracepoint_by_name(session->evlist, "sched:sched_waking"))
1928 		handlers[2].handler = process_sched_wakeup_ignore;
1929 
1930 	if (perf_session__set_tracepoints_handlers(session, handlers))
1931 		goto out_delete;
1932 
1933 	if (perf_session__has_traces(session, "record -R")) {
1934 		int err = perf_session__process_events(session);
1935 		if (err) {
1936 			pr_err("Failed to process events, error %d", err);
1937 			goto out_delete;
1938 		}
1939 
1940 		sched->nr_events      = session->evlist->stats.nr_events[0];
1941 		sched->nr_lost_events = session->evlist->stats.total_lost;
1942 		sched->nr_lost_chunks = session->evlist->stats.nr_events[PERF_RECORD_LOST];
1943 	}
1944 
1945 	rc = 0;
1946 out_delete:
1947 	perf_session__delete(session);
1948 	return rc;
1949 }
1950 
1951 /*
1952  * scheduling times are printed as msec.usec
1953  */
1954 static inline void print_sched_time(unsigned long long nsecs, int width)
1955 {
1956 	unsigned long msecs;
1957 	unsigned long usecs;
1958 
1959 	msecs  = nsecs / NSEC_PER_MSEC;
1960 	nsecs -= msecs * NSEC_PER_MSEC;
1961 	usecs  = nsecs / NSEC_PER_USEC;
1962 	printf("%*lu.%03lu ", width, msecs, usecs);
1963 }
1964 
1965 /*
1966  * returns runtime data for event, allocating memory for it the
1967  * first time it is used.
1968  */
1969 static struct evsel_runtime *evsel__get_runtime(struct evsel *evsel)
1970 {
1971 	struct evsel_runtime *r = evsel->priv;
1972 
1973 	if (r == NULL) {
1974 		r = zalloc(sizeof(struct evsel_runtime));
1975 		evsel->priv = r;
1976 	}
1977 
1978 	return r;
1979 }
1980 
1981 /*
1982  * save last time event was seen per cpu
1983  */
1984 static void evsel__save_time(struct evsel *evsel, u64 timestamp, u32 cpu)
1985 {
1986 	struct evsel_runtime *r = evsel__get_runtime(evsel);
1987 
1988 	if (r == NULL)
1989 		return;
1990 
1991 	if ((cpu >= r->ncpu) || (r->last_time == NULL)) {
1992 		int i, n = __roundup_pow_of_two(cpu+1);
1993 		void *p = r->last_time;
1994 
1995 		p = realloc(r->last_time, n * sizeof(u64));
1996 		if (!p)
1997 			return;
1998 
1999 		r->last_time = p;
2000 		for (i = r->ncpu; i < n; ++i)
2001 			r->last_time[i] = (u64) 0;
2002 
2003 		r->ncpu = n;
2004 	}
2005 
2006 	r->last_time[cpu] = timestamp;
2007 }
2008 
2009 /* returns last time this event was seen on the given cpu */
2010 static u64 evsel__get_time(struct evsel *evsel, u32 cpu)
2011 {
2012 	struct evsel_runtime *r = evsel__get_runtime(evsel);
2013 
2014 	if ((r == NULL) || (r->last_time == NULL) || (cpu >= r->ncpu))
2015 		return 0;
2016 
2017 	return r->last_time[cpu];
2018 }
2019 
2020 static int comm_width = 30;
2021 
2022 static char *timehist_get_commstr(struct thread *thread)
2023 {
2024 	static char str[32];
2025 	const char *comm = thread__comm_str(thread);
2026 	pid_t tid = thread__tid(thread);
2027 	pid_t pid = thread__pid(thread);
2028 	int n;
2029 
2030 	if (pid == 0)
2031 		n = scnprintf(str, sizeof(str), "%s", comm);
2032 
2033 	else if (tid != pid)
2034 		n = scnprintf(str, sizeof(str), "%s[%d/%d]", comm, tid, pid);
2035 
2036 	else
2037 		n = scnprintf(str, sizeof(str), "%s[%d]", comm, tid);
2038 
2039 	if (n > comm_width)
2040 		comm_width = n;
2041 
2042 	return str;
2043 }
2044 
2045 /* prio field format: xxx or xxx->yyy */
2046 #define MAX_PRIO_STR_LEN 8
2047 static char *timehist_get_priostr(struct evsel *evsel,
2048 				  struct thread *thread,
2049 				  struct perf_sample *sample)
2050 {
2051 	static char prio_str[16];
2052 	int prev_prio = (int)evsel__intval(evsel, sample, "prev_prio");
2053 	struct thread_runtime *tr = thread__priv(thread);
2054 
2055 	if (tr->prio != prev_prio && tr->prio != -1)
2056 		scnprintf(prio_str, sizeof(prio_str), "%d->%d", tr->prio, prev_prio);
2057 	else
2058 		scnprintf(prio_str, sizeof(prio_str), "%d", prev_prio);
2059 
2060 	return prio_str;
2061 }
2062 
2063 static void timehist_header(struct perf_sched *sched)
2064 {
2065 	u32 ncpus = sched->max_cpu.cpu + 1;
2066 	u32 i, j;
2067 
2068 	printf("%15s %6s ", "time", "cpu");
2069 
2070 	if (sched->show_cpu_visual) {
2071 		printf(" ");
2072 		for (i = 0, j = 0; i < ncpus; ++i) {
2073 			printf("%x", j++);
2074 			if (j > 15)
2075 				j = 0;
2076 		}
2077 		printf(" ");
2078 	}
2079 
2080 	printf(" %-*s", comm_width, "task name");
2081 
2082 	if (sched->show_prio)
2083 		printf("  %-*s", MAX_PRIO_STR_LEN, "prio");
2084 
2085 	printf("  %9s  %9s  %9s", "wait time", "sch delay", "run time");
2086 
2087 	if (sched->pre_migrations)
2088 		printf("  %9s", "pre-mig time");
2089 
2090 	if (sched->show_state)
2091 		printf("  %s", "state");
2092 
2093 	printf("\n");
2094 
2095 	/*
2096 	 * units row
2097 	 */
2098 	printf("%15s %-6s ", "", "");
2099 
2100 	if (sched->show_cpu_visual)
2101 		printf(" %*s ", ncpus, "");
2102 
2103 	printf(" %-*s", comm_width, "[tid/pid]");
2104 
2105 	if (sched->show_prio)
2106 		printf("  %-*s", MAX_PRIO_STR_LEN, "");
2107 
2108 	printf("  %9s  %9s  %9s", "(msec)", "(msec)", "(msec)");
2109 
2110 	if (sched->pre_migrations)
2111 		printf("  %9s", "(msec)");
2112 
2113 	printf("\n");
2114 
2115 	/*
2116 	 * separator
2117 	 */
2118 	printf("%.15s %.6s ", graph_dotted_line, graph_dotted_line);
2119 
2120 	if (sched->show_cpu_visual)
2121 		printf(" %.*s ", ncpus, graph_dotted_line);
2122 
2123 	printf(" %.*s", comm_width, graph_dotted_line);
2124 
2125 	if (sched->show_prio)
2126 		printf("  %.*s", MAX_PRIO_STR_LEN, graph_dotted_line);
2127 
2128 	printf("  %.9s  %.9s  %.9s", graph_dotted_line, graph_dotted_line, graph_dotted_line);
2129 
2130 	if (sched->pre_migrations)
2131 		printf("  %.9s", graph_dotted_line);
2132 
2133 	if (sched->show_state)
2134 		printf("  %.5s", graph_dotted_line);
2135 
2136 	printf("\n");
2137 }
2138 
2139 static void timehist_print_sample(struct perf_sched *sched,
2140 				  struct evsel *evsel,
2141 				  struct perf_sample *sample,
2142 				  struct addr_location *al,
2143 				  struct thread *thread,
2144 				  u64 t, const char state)
2145 {
2146 	struct thread_runtime *tr = thread__priv(thread);
2147 	const char *next_comm = evsel__strval(evsel, sample, "next_comm");
2148 	const u32 next_pid = evsel__intval(evsel, sample, "next_pid");
2149 	u32 max_cpus = sched->max_cpu.cpu + 1;
2150 	char tstr[64];
2151 	char nstr[30];
2152 	u64 wait_time;
2153 
2154 	if (cpu_list && !test_bit(sample->cpu, cpu_bitmap))
2155 		return;
2156 
2157 	timestamp__scnprintf_usec(t, tstr, sizeof(tstr));
2158 	printf("%15s [%04d] ", tstr, sample->cpu);
2159 
2160 	if (sched->show_cpu_visual) {
2161 		u32 i;
2162 		char c;
2163 
2164 		printf(" ");
2165 		for (i = 0; i < max_cpus; ++i) {
2166 			/* flag idle times with 'i'; others are sched events */
2167 			if (i == sample->cpu)
2168 				c = (thread__tid(thread) == 0) ? 'i' : 's';
2169 			else
2170 				c = ' ';
2171 			printf("%c", c);
2172 		}
2173 		printf(" ");
2174 	}
2175 
2176 	printf(" %-*s ", comm_width, timehist_get_commstr(thread));
2177 
2178 	if (sched->show_prio)
2179 		printf(" %-*s ", MAX_PRIO_STR_LEN, timehist_get_priostr(evsel, thread, sample));
2180 
2181 	wait_time = tr->dt_sleep + tr->dt_iowait + tr->dt_preempt;
2182 	print_sched_time(wait_time, 6);
2183 
2184 	print_sched_time(tr->dt_delay, 6);
2185 	print_sched_time(tr->dt_run, 6);
2186 	if (sched->pre_migrations)
2187 		print_sched_time(tr->dt_pre_mig, 6);
2188 
2189 	if (sched->show_state)
2190 		printf(" %5c ", thread__tid(thread) == 0 ? 'I' : state);
2191 
2192 	if (sched->show_next) {
2193 		snprintf(nstr, sizeof(nstr), "next: %s[%d]", next_comm, next_pid);
2194 		printf(" %-*s", comm_width, nstr);
2195 	}
2196 
2197 	if (sched->show_wakeups && !sched->show_next)
2198 		printf("  %-*s", comm_width, "");
2199 
2200 	if (thread__tid(thread) == 0)
2201 		goto out;
2202 
2203 	if (sched->show_callchain)
2204 		printf("  ");
2205 
2206 	sample__fprintf_sym(sample, al, 0,
2207 			    EVSEL__PRINT_SYM | EVSEL__PRINT_ONELINE |
2208 			    EVSEL__PRINT_CALLCHAIN_ARROW |
2209 			    EVSEL__PRINT_SKIP_IGNORED,
2210 			    get_tls_callchain_cursor(), symbol_conf.bt_stop_list,  stdout);
2211 
2212 out:
2213 	printf("\n");
2214 }
2215 
2216 /*
2217  * Explanation of delta-time stats:
2218  *
2219  *            t = time of current schedule out event
2220  *        tprev = time of previous sched out event
2221  *                also time of schedule-in event for current task
2222  *    last_time = time of last sched change event for current task
2223  *                (i.e, time process was last scheduled out)
2224  * ready_to_run = time of wakeup for current task
2225  *     migrated = time of task migration to another CPU
2226  *
2227  * -----|-------------|-------------|-------------|-------------|-----
2228  *    last         ready         migrated       tprev           t
2229  *    time         to run
2230  *
2231  *      |---------------- dt_wait ----------------|
2232  *                   |--------- dt_delay ---------|-- dt_run --|
2233  *                   |- dt_pre_mig -|
2234  *
2235  *     dt_run = run time of current task
2236  *    dt_wait = time between last schedule out event for task and tprev
2237  *              represents time spent off the cpu
2238  *   dt_delay = time between wakeup and schedule-in of task
2239  * dt_pre_mig = time between wakeup and migration to another CPU
2240  */
2241 
2242 static void timehist_update_runtime_stats(struct thread_runtime *r,
2243 					 u64 t, u64 tprev)
2244 {
2245 	r->dt_delay   = 0;
2246 	r->dt_sleep   = 0;
2247 	r->dt_iowait  = 0;
2248 	r->dt_preempt = 0;
2249 	r->dt_run     = 0;
2250 	r->dt_pre_mig = 0;
2251 
2252 	if (tprev) {
2253 		r->dt_run = t - tprev;
2254 		if (r->ready_to_run) {
2255 			if (r->ready_to_run > tprev)
2256 				pr_debug("time travel: wakeup time for task > previous sched_switch event\n");
2257 			else
2258 				r->dt_delay = tprev - r->ready_to_run;
2259 
2260 			if ((r->migrated > r->ready_to_run) && (r->migrated < tprev))
2261 				r->dt_pre_mig = r->migrated - r->ready_to_run;
2262 		}
2263 
2264 		if (r->last_time > tprev)
2265 			pr_debug("time travel: last sched out time for task > previous sched_switch event\n");
2266 		else if (r->last_time) {
2267 			u64 dt_wait = tprev - r->last_time;
2268 
2269 			if (r->last_state == 'R')
2270 				r->dt_preempt = dt_wait;
2271 			else if (r->last_state == 'D')
2272 				r->dt_iowait = dt_wait;
2273 			else
2274 				r->dt_sleep = dt_wait;
2275 		}
2276 	}
2277 
2278 	update_stats(&r->run_stats, r->dt_run);
2279 
2280 	r->total_run_time     += r->dt_run;
2281 	r->total_delay_time   += r->dt_delay;
2282 	r->total_sleep_time   += r->dt_sleep;
2283 	r->total_iowait_time  += r->dt_iowait;
2284 	r->total_preempt_time += r->dt_preempt;
2285 	r->total_pre_mig_time += r->dt_pre_mig;
2286 }
2287 
2288 static bool is_idle_sample(struct perf_sample *sample,
2289 			   struct evsel *evsel)
2290 {
2291 	/* pid 0 == swapper == idle task */
2292 	if (evsel__name_is(evsel, "sched:sched_switch"))
2293 		return evsel__intval(evsel, sample, "prev_pid") == 0;
2294 
2295 	return sample->pid == 0;
2296 }
2297 
2298 static void save_task_callchain(struct perf_sched *sched,
2299 				struct perf_sample *sample,
2300 				struct evsel *evsel,
2301 				struct machine *machine)
2302 {
2303 	struct callchain_cursor *cursor;
2304 	struct thread *thread;
2305 
2306 	/* want main thread for process - has maps */
2307 	thread = machine__findnew_thread(machine, sample->pid, sample->pid);
2308 	if (thread == NULL) {
2309 		pr_debug("Failed to get thread for pid %d.\n", sample->pid);
2310 		return;
2311 	}
2312 
2313 	if (!sched->show_callchain || sample->callchain == NULL)
2314 		return;
2315 
2316 	cursor = get_tls_callchain_cursor();
2317 
2318 	if (thread__resolve_callchain(thread, cursor, evsel, sample,
2319 				      NULL, NULL, sched->max_stack + 2) != 0) {
2320 		if (verbose > 0)
2321 			pr_err("Failed to resolve callchain. Skipping\n");
2322 
2323 		return;
2324 	}
2325 
2326 	callchain_cursor_commit(cursor);
2327 
2328 	while (true) {
2329 		struct callchain_cursor_node *node;
2330 		struct symbol *sym;
2331 
2332 		node = callchain_cursor_current(cursor);
2333 		if (node == NULL)
2334 			break;
2335 
2336 		sym = node->ms.sym;
2337 		if (sym) {
2338 			if (!strcmp(sym->name, "schedule") ||
2339 			    !strcmp(sym->name, "__schedule") ||
2340 			    !strcmp(sym->name, "preempt_schedule"))
2341 				sym->ignore = 1;
2342 		}
2343 
2344 		callchain_cursor_advance(cursor);
2345 	}
2346 }
2347 
2348 static int init_idle_thread(struct thread *thread)
2349 {
2350 	struct idle_thread_runtime *itr;
2351 
2352 	thread__set_comm(thread, idle_comm, 0);
2353 
2354 	itr = zalloc(sizeof(*itr));
2355 	if (itr == NULL)
2356 		return -ENOMEM;
2357 
2358 	init_prio(&itr->tr);
2359 	init_stats(&itr->tr.run_stats);
2360 	callchain_init(&itr->callchain);
2361 	callchain_cursor_reset(&itr->cursor);
2362 	thread__set_priv(thread, itr);
2363 
2364 	return 0;
2365 }
2366 
2367 /*
2368  * Track idle stats per cpu by maintaining a local thread
2369  * struct for the idle task on each cpu.
2370  */
2371 static int init_idle_threads(int ncpu)
2372 {
2373 	int i, ret;
2374 
2375 	idle_threads = zalloc(ncpu * sizeof(struct thread *));
2376 	if (!idle_threads)
2377 		return -ENOMEM;
2378 
2379 	idle_max_cpu = ncpu;
2380 
2381 	/* allocate the actual thread struct if needed */
2382 	for (i = 0; i < ncpu; ++i) {
2383 		idle_threads[i] = thread__new(0, 0);
2384 		if (idle_threads[i] == NULL)
2385 			return -ENOMEM;
2386 
2387 		ret = init_idle_thread(idle_threads[i]);
2388 		if (ret < 0)
2389 			return ret;
2390 	}
2391 
2392 	return 0;
2393 }
2394 
2395 static void free_idle_threads(void)
2396 {
2397 	int i;
2398 
2399 	if (idle_threads == NULL)
2400 		return;
2401 
2402 	for (i = 0; i < idle_max_cpu; ++i) {
2403 		if ((idle_threads[i]))
2404 			thread__delete(idle_threads[i]);
2405 	}
2406 
2407 	free(idle_threads);
2408 }
2409 
2410 static struct thread *get_idle_thread(int cpu)
2411 {
2412 	/*
2413 	 * expand/allocate array of pointers to local thread
2414 	 * structs if needed
2415 	 */
2416 	if ((cpu >= idle_max_cpu) || (idle_threads == NULL)) {
2417 		int i, j = __roundup_pow_of_two(cpu+1);
2418 		void *p;
2419 
2420 		p = realloc(idle_threads, j * sizeof(struct thread *));
2421 		if (!p)
2422 			return NULL;
2423 
2424 		idle_threads = (struct thread **) p;
2425 		for (i = idle_max_cpu; i < j; ++i)
2426 			idle_threads[i] = NULL;
2427 
2428 		idle_max_cpu = j;
2429 	}
2430 
2431 	/* allocate a new thread struct if needed */
2432 	if (idle_threads[cpu] == NULL) {
2433 		idle_threads[cpu] = thread__new(0, 0);
2434 
2435 		if (idle_threads[cpu]) {
2436 			if (init_idle_thread(idle_threads[cpu]) < 0)
2437 				return NULL;
2438 		}
2439 	}
2440 
2441 	return idle_threads[cpu];
2442 }
2443 
2444 static void save_idle_callchain(struct perf_sched *sched,
2445 				struct idle_thread_runtime *itr,
2446 				struct perf_sample *sample)
2447 {
2448 	struct callchain_cursor *cursor;
2449 
2450 	if (!sched->show_callchain || sample->callchain == NULL)
2451 		return;
2452 
2453 	cursor = get_tls_callchain_cursor();
2454 	if (cursor == NULL)
2455 		return;
2456 
2457 	callchain_cursor__copy(&itr->cursor, cursor);
2458 }
2459 
2460 static struct thread *timehist_get_thread(struct perf_sched *sched,
2461 					  struct perf_sample *sample,
2462 					  struct machine *machine,
2463 					  struct evsel *evsel)
2464 {
2465 	struct thread *thread;
2466 
2467 	if (is_idle_sample(sample, evsel)) {
2468 		thread = get_idle_thread(sample->cpu);
2469 		if (thread == NULL)
2470 			pr_err("Failed to get idle thread for cpu %d.\n", sample->cpu);
2471 
2472 	} else {
2473 		/* there were samples with tid 0 but non-zero pid */
2474 		thread = machine__findnew_thread(machine, sample->pid,
2475 						 sample->tid ?: sample->pid);
2476 		if (thread == NULL) {
2477 			pr_debug("Failed to get thread for tid %d. skipping sample.\n",
2478 				 sample->tid);
2479 		}
2480 
2481 		save_task_callchain(sched, sample, evsel, machine);
2482 		if (sched->idle_hist) {
2483 			struct thread *idle;
2484 			struct idle_thread_runtime *itr;
2485 
2486 			idle = get_idle_thread(sample->cpu);
2487 			if (idle == NULL) {
2488 				pr_err("Failed to get idle thread for cpu %d.\n", sample->cpu);
2489 				return NULL;
2490 			}
2491 
2492 			itr = thread__priv(idle);
2493 			if (itr == NULL)
2494 				return NULL;
2495 
2496 			itr->last_thread = thread;
2497 
2498 			/* copy task callchain when entering to idle */
2499 			if (evsel__intval(evsel, sample, "next_pid") == 0)
2500 				save_idle_callchain(sched, itr, sample);
2501 		}
2502 	}
2503 
2504 	return thread;
2505 }
2506 
2507 static bool timehist_skip_sample(struct perf_sched *sched,
2508 				 struct thread *thread,
2509 				 struct evsel *evsel,
2510 				 struct perf_sample *sample)
2511 {
2512 	bool rc = false;
2513 	int prio = -1;
2514 	struct thread_runtime *tr = NULL;
2515 
2516 	if (thread__is_filtered(thread)) {
2517 		rc = true;
2518 		sched->skipped_samples++;
2519 	}
2520 
2521 	if (sched->prio_str) {
2522 		/*
2523 		 * Because priority may be changed during task execution,
2524 		 * first read priority from prev sched_in event for current task.
2525 		 * If prev sched_in event is not saved, then read priority from
2526 		 * current task sched_out event.
2527 		 */
2528 		tr = thread__get_runtime(thread);
2529 		if (tr && tr->prio != -1)
2530 			prio = tr->prio;
2531 		else if (evsel__name_is(evsel, "sched:sched_switch"))
2532 			prio = evsel__intval(evsel, sample, "prev_prio");
2533 
2534 		if (prio != -1 && !test_bit(prio, sched->prio_bitmap)) {
2535 			rc = true;
2536 			sched->skipped_samples++;
2537 		}
2538 	}
2539 
2540 	if (sched->idle_hist) {
2541 		if (!evsel__name_is(evsel, "sched:sched_switch"))
2542 			rc = true;
2543 		else if (evsel__intval(evsel, sample, "prev_pid") != 0 &&
2544 			 evsel__intval(evsel, sample, "next_pid") != 0)
2545 			rc = true;
2546 	}
2547 
2548 	return rc;
2549 }
2550 
2551 static void timehist_print_wakeup_event(struct perf_sched *sched,
2552 					struct evsel *evsel,
2553 					struct perf_sample *sample,
2554 					struct machine *machine,
2555 					struct thread *awakened)
2556 {
2557 	struct thread *thread;
2558 	char tstr[64];
2559 
2560 	thread = machine__findnew_thread(machine, sample->pid, sample->tid);
2561 	if (thread == NULL)
2562 		return;
2563 
2564 	/* show wakeup unless both awakee and awaker are filtered */
2565 	if (timehist_skip_sample(sched, thread, evsel, sample) &&
2566 	    timehist_skip_sample(sched, awakened, evsel, sample)) {
2567 		return;
2568 	}
2569 
2570 	timestamp__scnprintf_usec(sample->time, tstr, sizeof(tstr));
2571 	printf("%15s [%04d] ", tstr, sample->cpu);
2572 	if (sched->show_cpu_visual)
2573 		printf(" %*s ", sched->max_cpu.cpu + 1, "");
2574 
2575 	printf(" %-*s ", comm_width, timehist_get_commstr(thread));
2576 
2577 	/* dt spacer */
2578 	printf("  %9s  %9s  %9s ", "", "", "");
2579 
2580 	printf("awakened: %s", timehist_get_commstr(awakened));
2581 
2582 	printf("\n");
2583 }
2584 
2585 static int timehist_sched_wakeup_ignore(const struct perf_tool *tool __maybe_unused,
2586 					union perf_event *event __maybe_unused,
2587 					struct evsel *evsel __maybe_unused,
2588 					struct perf_sample *sample __maybe_unused,
2589 					struct machine *machine __maybe_unused)
2590 {
2591 	return 0;
2592 }
2593 
2594 static int timehist_sched_wakeup_event(const struct perf_tool *tool,
2595 				       union perf_event *event __maybe_unused,
2596 				       struct evsel *evsel,
2597 				       struct perf_sample *sample,
2598 				       struct machine *machine)
2599 {
2600 	struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
2601 	struct thread *thread;
2602 	struct thread_runtime *tr = NULL;
2603 	/* want pid of awakened task not pid in sample */
2604 	const u32 pid = evsel__intval(evsel, sample, "pid");
2605 
2606 	thread = machine__findnew_thread(machine, 0, pid);
2607 	if (thread == NULL)
2608 		return -1;
2609 
2610 	tr = thread__get_runtime(thread);
2611 	if (tr == NULL)
2612 		return -1;
2613 
2614 	if (tr->ready_to_run == 0)
2615 		tr->ready_to_run = sample->time;
2616 
2617 	/* show wakeups if requested */
2618 	if (sched->show_wakeups &&
2619 	    !perf_time__skip_sample(&sched->ptime, sample->time))
2620 		timehist_print_wakeup_event(sched, evsel, sample, machine, thread);
2621 
2622 	return 0;
2623 }
2624 
2625 static void timehist_print_migration_event(struct perf_sched *sched,
2626 					struct evsel *evsel,
2627 					struct perf_sample *sample,
2628 					struct machine *machine,
2629 					struct thread *migrated)
2630 {
2631 	struct thread *thread;
2632 	char tstr[64];
2633 	u32 max_cpus;
2634 	u32 ocpu, dcpu;
2635 
2636 	if (sched->summary_only)
2637 		return;
2638 
2639 	max_cpus = sched->max_cpu.cpu + 1;
2640 	ocpu = evsel__intval(evsel, sample, "orig_cpu");
2641 	dcpu = evsel__intval(evsel, sample, "dest_cpu");
2642 
2643 	thread = machine__findnew_thread(machine, sample->pid, sample->tid);
2644 	if (thread == NULL)
2645 		return;
2646 
2647 	if (timehist_skip_sample(sched, thread, evsel, sample) &&
2648 	    timehist_skip_sample(sched, migrated, evsel, sample)) {
2649 		return;
2650 	}
2651 
2652 	timestamp__scnprintf_usec(sample->time, tstr, sizeof(tstr));
2653 	printf("%15s [%04d] ", tstr, sample->cpu);
2654 
2655 	if (sched->show_cpu_visual) {
2656 		u32 i;
2657 		char c;
2658 
2659 		printf("  ");
2660 		for (i = 0; i < max_cpus; ++i) {
2661 			c = (i == sample->cpu) ? 'm' : ' ';
2662 			printf("%c", c);
2663 		}
2664 		printf("  ");
2665 	}
2666 
2667 	printf(" %-*s ", comm_width, timehist_get_commstr(thread));
2668 
2669 	/* dt spacer */
2670 	printf("  %9s  %9s  %9s ", "", "", "");
2671 
2672 	printf("migrated: %s", timehist_get_commstr(migrated));
2673 	printf(" cpu %d => %d", ocpu, dcpu);
2674 
2675 	printf("\n");
2676 }
2677 
2678 static int timehist_migrate_task_event(const struct perf_tool *tool,
2679 				       union perf_event *event __maybe_unused,
2680 				       struct evsel *evsel,
2681 				       struct perf_sample *sample,
2682 				       struct machine *machine)
2683 {
2684 	struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
2685 	struct thread *thread;
2686 	struct thread_runtime *tr = NULL;
2687 	/* want pid of migrated task not pid in sample */
2688 	const u32 pid = evsel__intval(evsel, sample, "pid");
2689 
2690 	thread = machine__findnew_thread(machine, 0, pid);
2691 	if (thread == NULL)
2692 		return -1;
2693 
2694 	tr = thread__get_runtime(thread);
2695 	if (tr == NULL)
2696 		return -1;
2697 
2698 	tr->migrations++;
2699 	tr->migrated = sample->time;
2700 
2701 	/* show migrations if requested */
2702 	if (sched->show_migrations) {
2703 		timehist_print_migration_event(sched, evsel, sample,
2704 							machine, thread);
2705 	}
2706 
2707 	return 0;
2708 }
2709 
2710 static void timehist_update_task_prio(struct evsel *evsel,
2711 				      struct perf_sample *sample,
2712 				      struct machine *machine)
2713 {
2714 	struct thread *thread;
2715 	struct thread_runtime *tr = NULL;
2716 	const u32 next_pid = evsel__intval(evsel, sample, "next_pid");
2717 	const u32 next_prio = evsel__intval(evsel, sample, "next_prio");
2718 
2719 	if (next_pid == 0)
2720 		thread = get_idle_thread(sample->cpu);
2721 	else
2722 		thread = machine__findnew_thread(machine, -1, next_pid);
2723 
2724 	if (thread == NULL)
2725 		return;
2726 
2727 	tr = thread__get_runtime(thread);
2728 	if (tr == NULL)
2729 		return;
2730 
2731 	tr->prio = next_prio;
2732 }
2733 
2734 static int timehist_sched_change_event(const struct perf_tool *tool,
2735 				       union perf_event *event,
2736 				       struct evsel *evsel,
2737 				       struct perf_sample *sample,
2738 				       struct machine *machine)
2739 {
2740 	struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
2741 	struct perf_time_interval *ptime = &sched->ptime;
2742 	struct addr_location al;
2743 	struct thread *thread;
2744 	struct thread_runtime *tr = NULL;
2745 	u64 tprev, t = sample->time;
2746 	int rc = 0;
2747 	const char state = evsel__taskstate(evsel, sample, "prev_state");
2748 
2749 	addr_location__init(&al);
2750 	if (machine__resolve(machine, &al, sample) < 0) {
2751 		pr_err("problem processing %d event. skipping it\n",
2752 		       event->header.type);
2753 		rc = -1;
2754 		goto out;
2755 	}
2756 
2757 	if (sched->show_prio || sched->prio_str)
2758 		timehist_update_task_prio(evsel, sample, machine);
2759 
2760 	thread = timehist_get_thread(sched, sample, machine, evsel);
2761 	if (thread == NULL) {
2762 		rc = -1;
2763 		goto out;
2764 	}
2765 
2766 	if (timehist_skip_sample(sched, thread, evsel, sample))
2767 		goto out;
2768 
2769 	tr = thread__get_runtime(thread);
2770 	if (tr == NULL) {
2771 		rc = -1;
2772 		goto out;
2773 	}
2774 
2775 	tprev = evsel__get_time(evsel, sample->cpu);
2776 
2777 	/*
2778 	 * If start time given:
2779 	 * - sample time is under window user cares about - skip sample
2780 	 * - tprev is under window user cares about  - reset to start of window
2781 	 */
2782 	if (ptime->start && ptime->start > t)
2783 		goto out;
2784 
2785 	if (tprev && ptime->start > tprev)
2786 		tprev = ptime->start;
2787 
2788 	/*
2789 	 * If end time given:
2790 	 * - previous sched event is out of window - we are done
2791 	 * - sample time is beyond window user cares about - reset it
2792 	 *   to close out stats for time window interest
2793 	 * - If tprev is 0, that is, sched_in event for current task is
2794 	 *   not recorded, cannot determine whether sched_in event is
2795 	 *   within time window interest - ignore it
2796 	 */
2797 	if (ptime->end) {
2798 		if (!tprev || tprev > ptime->end)
2799 			goto out;
2800 
2801 		if (t > ptime->end)
2802 			t = ptime->end;
2803 	}
2804 
2805 	if (!sched->idle_hist || thread__tid(thread) == 0) {
2806 		if (!cpu_list || test_bit(sample->cpu, cpu_bitmap))
2807 			timehist_update_runtime_stats(tr, t, tprev);
2808 
2809 		if (sched->idle_hist) {
2810 			struct idle_thread_runtime *itr = (void *)tr;
2811 			struct thread_runtime *last_tr;
2812 
2813 			if (itr->last_thread == NULL)
2814 				goto out;
2815 
2816 			/* add current idle time as last thread's runtime */
2817 			last_tr = thread__get_runtime(itr->last_thread);
2818 			if (last_tr == NULL)
2819 				goto out;
2820 
2821 			timehist_update_runtime_stats(last_tr, t, tprev);
2822 			/*
2823 			 * remove delta time of last thread as it's not updated
2824 			 * and otherwise it will show an invalid value next
2825 			 * time.  we only care total run time and run stat.
2826 			 */
2827 			last_tr->dt_run = 0;
2828 			last_tr->dt_delay = 0;
2829 			last_tr->dt_sleep = 0;
2830 			last_tr->dt_iowait = 0;
2831 			last_tr->dt_preempt = 0;
2832 
2833 			if (itr->cursor.nr)
2834 				callchain_append(&itr->callchain, &itr->cursor, t - tprev);
2835 
2836 			itr->last_thread = NULL;
2837 		}
2838 
2839 		if (!sched->summary_only)
2840 			timehist_print_sample(sched, evsel, sample, &al, thread, t, state);
2841 	}
2842 
2843 out:
2844 	if (sched->hist_time.start == 0 && t >= ptime->start)
2845 		sched->hist_time.start = t;
2846 	if (ptime->end == 0 || t <= ptime->end)
2847 		sched->hist_time.end = t;
2848 
2849 	if (tr) {
2850 		/* time of this sched_switch event becomes last time task seen */
2851 		tr->last_time = sample->time;
2852 
2853 		/* last state is used to determine where to account wait time */
2854 		tr->last_state = state;
2855 
2856 		/* sched out event for task so reset ready to run time and migrated time */
2857 		if (state == 'R')
2858 			tr->ready_to_run = t;
2859 		else
2860 			tr->ready_to_run = 0;
2861 
2862 		tr->migrated = 0;
2863 	}
2864 
2865 	evsel__save_time(evsel, sample->time, sample->cpu);
2866 
2867 	addr_location__exit(&al);
2868 	return rc;
2869 }
2870 
2871 static int timehist_sched_switch_event(const struct perf_tool *tool,
2872 			     union perf_event *event,
2873 			     struct evsel *evsel,
2874 			     struct perf_sample *sample,
2875 			     struct machine *machine __maybe_unused)
2876 {
2877 	return timehist_sched_change_event(tool, event, evsel, sample, machine);
2878 }
2879 
2880 static int process_lost(const struct perf_tool *tool __maybe_unused,
2881 			union perf_event *event,
2882 			struct perf_sample *sample,
2883 			struct machine *machine __maybe_unused)
2884 {
2885 	char tstr[64];
2886 
2887 	timestamp__scnprintf_usec(sample->time, tstr, sizeof(tstr));
2888 	printf("%15s ", tstr);
2889 	printf("lost %" PRI_lu64 " events on cpu %d\n", event->lost.lost, sample->cpu);
2890 
2891 	return 0;
2892 }
2893 
2894 
2895 static void print_thread_runtime(struct thread *t,
2896 				 struct thread_runtime *r)
2897 {
2898 	double mean = avg_stats(&r->run_stats);
2899 	float stddev;
2900 
2901 	printf("%*s   %5d  %9" PRIu64 " ",
2902 	       comm_width, timehist_get_commstr(t), thread__ppid(t),
2903 	       (u64) r->run_stats.n);
2904 
2905 	print_sched_time(r->total_run_time, 8);
2906 	stddev = rel_stddev_stats(stddev_stats(&r->run_stats), mean);
2907 	print_sched_time(r->run_stats.min, 6);
2908 	printf(" ");
2909 	print_sched_time((u64) mean, 6);
2910 	printf(" ");
2911 	print_sched_time(r->run_stats.max, 6);
2912 	printf("  ");
2913 	printf("%5.2f", stddev);
2914 	printf("   %5" PRIu64, r->migrations);
2915 	printf("\n");
2916 }
2917 
2918 static void print_thread_waittime(struct thread *t,
2919 				  struct thread_runtime *r)
2920 {
2921 	printf("%*s   %5d  %9" PRIu64 " ",
2922 	       comm_width, timehist_get_commstr(t), thread__ppid(t),
2923 	       (u64) r->run_stats.n);
2924 
2925 	print_sched_time(r->total_run_time, 8);
2926 	print_sched_time(r->total_sleep_time, 6);
2927 	printf(" ");
2928 	print_sched_time(r->total_iowait_time, 6);
2929 	printf(" ");
2930 	print_sched_time(r->total_preempt_time, 6);
2931 	printf(" ");
2932 	print_sched_time(r->total_delay_time, 6);
2933 	printf("\n");
2934 }
2935 
2936 struct total_run_stats {
2937 	struct perf_sched *sched;
2938 	u64  sched_count;
2939 	u64  task_count;
2940 	u64  total_run_time;
2941 };
2942 
2943 static int show_thread_runtime(struct thread *t, void *priv)
2944 {
2945 	struct total_run_stats *stats = priv;
2946 	struct thread_runtime *r;
2947 
2948 	if (thread__is_filtered(t))
2949 		return 0;
2950 
2951 	r = thread__priv(t);
2952 	if (r && r->run_stats.n) {
2953 		stats->task_count++;
2954 		stats->sched_count += r->run_stats.n;
2955 		stats->total_run_time += r->total_run_time;
2956 
2957 		if (stats->sched->show_state)
2958 			print_thread_waittime(t, r);
2959 		else
2960 			print_thread_runtime(t, r);
2961 	}
2962 
2963 	return 0;
2964 }
2965 
2966 static size_t callchain__fprintf_folded(FILE *fp, struct callchain_node *node)
2967 {
2968 	const char *sep = " <- ";
2969 	struct callchain_list *chain;
2970 	size_t ret = 0;
2971 	char bf[1024];
2972 	bool first;
2973 
2974 	if (node == NULL)
2975 		return 0;
2976 
2977 	ret = callchain__fprintf_folded(fp, node->parent);
2978 	first = (ret == 0);
2979 
2980 	list_for_each_entry(chain, &node->val, list) {
2981 		if (chain->ip >= PERF_CONTEXT_MAX)
2982 			continue;
2983 		if (chain->ms.sym && chain->ms.sym->ignore)
2984 			continue;
2985 		ret += fprintf(fp, "%s%s", first ? "" : sep,
2986 			       callchain_list__sym_name(chain, bf, sizeof(bf),
2987 							false));
2988 		first = false;
2989 	}
2990 
2991 	return ret;
2992 }
2993 
2994 static size_t timehist_print_idlehist_callchain(struct rb_root_cached *root)
2995 {
2996 	size_t ret = 0;
2997 	FILE *fp = stdout;
2998 	struct callchain_node *chain;
2999 	struct rb_node *rb_node = rb_first_cached(root);
3000 
3001 	printf("  %16s  %8s  %s\n", "Idle time (msec)", "Count", "Callchains");
3002 	printf("  %.16s  %.8s  %.50s\n", graph_dotted_line, graph_dotted_line,
3003 	       graph_dotted_line);
3004 
3005 	while (rb_node) {
3006 		chain = rb_entry(rb_node, struct callchain_node, rb_node);
3007 		rb_node = rb_next(rb_node);
3008 
3009 		ret += fprintf(fp, "  ");
3010 		print_sched_time(chain->hit, 12);
3011 		ret += 16;  /* print_sched_time returns 2nd arg + 4 */
3012 		ret += fprintf(fp, " %8d  ", chain->count);
3013 		ret += callchain__fprintf_folded(fp, chain);
3014 		ret += fprintf(fp, "\n");
3015 	}
3016 
3017 	return ret;
3018 }
3019 
3020 static void timehist_print_summary(struct perf_sched *sched,
3021 				   struct perf_session *session)
3022 {
3023 	struct machine *m = &session->machines.host;
3024 	struct total_run_stats totals;
3025 	u64 task_count;
3026 	struct thread *t;
3027 	struct thread_runtime *r;
3028 	int i;
3029 	u64 hist_time = sched->hist_time.end - sched->hist_time.start;
3030 
3031 	memset(&totals, 0, sizeof(totals));
3032 	totals.sched = sched;
3033 
3034 	if (sched->idle_hist) {
3035 		printf("\nIdle-time summary\n");
3036 		printf("%*s  parent  sched-out  ", comm_width, "comm");
3037 		printf("  idle-time   min-idle    avg-idle    max-idle  stddev  migrations\n");
3038 	} else if (sched->show_state) {
3039 		printf("\nWait-time summary\n");
3040 		printf("%*s  parent   sched-in  ", comm_width, "comm");
3041 		printf("   run-time      sleep      iowait     preempt       delay\n");
3042 	} else {
3043 		printf("\nRuntime summary\n");
3044 		printf("%*s  parent   sched-in  ", comm_width, "comm");
3045 		printf("   run-time    min-run     avg-run     max-run  stddev  migrations\n");
3046 	}
3047 	printf("%*s            (count)  ", comm_width, "");
3048 	printf("     (msec)     (msec)      (msec)      (msec)       %s\n",
3049 	       sched->show_state ? "(msec)" : "%");
3050 	printf("%.117s\n", graph_dotted_line);
3051 
3052 	machine__for_each_thread(m, show_thread_runtime, &totals);
3053 	task_count = totals.task_count;
3054 	if (!task_count)
3055 		printf("<no still running tasks>\n");
3056 
3057 	/* CPU idle stats not tracked when samples were skipped */
3058 	if (sched->skipped_samples && !sched->idle_hist)
3059 		return;
3060 
3061 	printf("\nIdle stats:\n");
3062 	for (i = 0; i < idle_max_cpu; ++i) {
3063 		if (cpu_list && !test_bit(i, cpu_bitmap))
3064 			continue;
3065 
3066 		t = idle_threads[i];
3067 		if (!t)
3068 			continue;
3069 
3070 		r = thread__priv(t);
3071 		if (r && r->run_stats.n) {
3072 			totals.sched_count += r->run_stats.n;
3073 			printf("    CPU %2d idle for ", i);
3074 			print_sched_time(r->total_run_time, 6);
3075 			printf(" msec  (%6.2f%%)\n", 100.0 * r->total_run_time / hist_time);
3076 		} else
3077 			printf("    CPU %2d idle entire time window\n", i);
3078 	}
3079 
3080 	if (sched->idle_hist && sched->show_callchain) {
3081 		callchain_param.mode  = CHAIN_FOLDED;
3082 		callchain_param.value = CCVAL_PERIOD;
3083 
3084 		callchain_register_param(&callchain_param);
3085 
3086 		printf("\nIdle stats by callchain:\n");
3087 		for (i = 0; i < idle_max_cpu; ++i) {
3088 			struct idle_thread_runtime *itr;
3089 
3090 			t = idle_threads[i];
3091 			if (!t)
3092 				continue;
3093 
3094 			itr = thread__priv(t);
3095 			if (itr == NULL)
3096 				continue;
3097 
3098 			callchain_param.sort(&itr->sorted_root.rb_root, &itr->callchain,
3099 					     0, &callchain_param);
3100 
3101 			printf("  CPU %2d:", i);
3102 			print_sched_time(itr->tr.total_run_time, 6);
3103 			printf(" msec\n");
3104 			timehist_print_idlehist_callchain(&itr->sorted_root);
3105 			printf("\n");
3106 		}
3107 	}
3108 
3109 	printf("\n"
3110 	       "    Total number of unique tasks: %" PRIu64 "\n"
3111 	       "Total number of context switches: %" PRIu64 "\n",
3112 	       totals.task_count, totals.sched_count);
3113 
3114 	printf("           Total run time (msec): ");
3115 	print_sched_time(totals.total_run_time, 2);
3116 	printf("\n");
3117 
3118 	printf("    Total scheduling time (msec): ");
3119 	print_sched_time(hist_time, 2);
3120 	printf(" (x %d)\n", sched->max_cpu.cpu);
3121 }
3122 
3123 typedef int (*sched_handler)(const struct perf_tool *tool,
3124 			  union perf_event *event,
3125 			  struct evsel *evsel,
3126 			  struct perf_sample *sample,
3127 			  struct machine *machine);
3128 
3129 static int perf_timehist__process_sample(const struct perf_tool *tool,
3130 					 union perf_event *event,
3131 					 struct perf_sample *sample,
3132 					 struct evsel *evsel,
3133 					 struct machine *machine)
3134 {
3135 	struct perf_sched *sched = container_of(tool, struct perf_sched, tool);
3136 	int err = 0;
3137 	struct perf_cpu this_cpu = {
3138 		.cpu = sample->cpu,
3139 	};
3140 
3141 	if (this_cpu.cpu > sched->max_cpu.cpu)
3142 		sched->max_cpu = this_cpu;
3143 
3144 	if (evsel->handler != NULL) {
3145 		sched_handler f = evsel->handler;
3146 
3147 		err = f(tool, event, evsel, sample, machine);
3148 	}
3149 
3150 	return err;
3151 }
3152 
3153 static int timehist_check_attr(struct perf_sched *sched,
3154 			       struct evlist *evlist)
3155 {
3156 	struct evsel *evsel;
3157 	struct evsel_runtime *er;
3158 
3159 	list_for_each_entry(evsel, &evlist->core.entries, core.node) {
3160 		er = evsel__get_runtime(evsel);
3161 		if (er == NULL) {
3162 			pr_err("Failed to allocate memory for evsel runtime data\n");
3163 			return -1;
3164 		}
3165 
3166 		/* only need to save callchain related to sched_switch event */
3167 		if (sched->show_callchain &&
3168 		    evsel__name_is(evsel, "sched:sched_switch") &&
3169 		    !evsel__has_callchain(evsel)) {
3170 			pr_info("Samples of sched_switch event do not have callchains.\n");
3171 			sched->show_callchain = 0;
3172 			symbol_conf.use_callchain = 0;
3173 		}
3174 	}
3175 
3176 	return 0;
3177 }
3178 
3179 static int timehist_parse_prio_str(struct perf_sched *sched)
3180 {
3181 	char *p;
3182 	unsigned long start_prio, end_prio;
3183 	const char *str = sched->prio_str;
3184 
3185 	if (!str)
3186 		return 0;
3187 
3188 	while (isdigit(*str)) {
3189 		p = NULL;
3190 		start_prio = strtoul(str, &p, 0);
3191 		if (start_prio >= MAX_PRIO || (*p != '\0' && *p != ',' && *p != '-'))
3192 			return -1;
3193 
3194 		if (*p == '-') {
3195 			str = ++p;
3196 			p = NULL;
3197 			end_prio = strtoul(str, &p, 0);
3198 
3199 			if (end_prio >= MAX_PRIO || (*p != '\0' && *p != ','))
3200 				return -1;
3201 
3202 			if (end_prio < start_prio)
3203 				return -1;
3204 		} else {
3205 			end_prio = start_prio;
3206 		}
3207 
3208 		for (; start_prio <= end_prio; start_prio++)
3209 			__set_bit(start_prio, sched->prio_bitmap);
3210 
3211 		if (*p)
3212 			++p;
3213 
3214 		str = p;
3215 	}
3216 
3217 	return 0;
3218 }
3219 
3220 static int perf_sched__timehist(struct perf_sched *sched)
3221 {
3222 	struct evsel_str_handler handlers[] = {
3223 		{ "sched:sched_switch",       timehist_sched_switch_event, },
3224 		{ "sched:sched_wakeup",	      timehist_sched_wakeup_event, },
3225 		{ "sched:sched_waking",       timehist_sched_wakeup_event, },
3226 		{ "sched:sched_wakeup_new",   timehist_sched_wakeup_event, },
3227 	};
3228 	const struct evsel_str_handler migrate_handlers[] = {
3229 		{ "sched:sched_migrate_task", timehist_migrate_task_event, },
3230 	};
3231 	struct perf_data data = {
3232 		.path  = input_name,
3233 		.mode  = PERF_DATA_MODE_READ,
3234 		.force = sched->force,
3235 	};
3236 
3237 	struct perf_session *session;
3238 	struct evlist *evlist;
3239 	int err = -1;
3240 
3241 	/*
3242 	 * event handlers for timehist option
3243 	 */
3244 	sched->tool.sample	 = perf_timehist__process_sample;
3245 	sched->tool.mmap	 = perf_event__process_mmap;
3246 	sched->tool.comm	 = perf_event__process_comm;
3247 	sched->tool.exit	 = perf_event__process_exit;
3248 	sched->tool.fork	 = perf_event__process_fork;
3249 	sched->tool.lost	 = process_lost;
3250 	sched->tool.attr	 = perf_event__process_attr;
3251 	sched->tool.tracing_data = perf_event__process_tracing_data;
3252 	sched->tool.build_id	 = perf_event__process_build_id;
3253 
3254 	sched->tool.ordering_requires_timestamps = true;
3255 
3256 	symbol_conf.use_callchain = sched->show_callchain;
3257 
3258 	session = perf_session__new(&data, &sched->tool);
3259 	if (IS_ERR(session))
3260 		return PTR_ERR(session);
3261 
3262 	if (cpu_list) {
3263 		err = perf_session__cpu_bitmap(session, cpu_list, cpu_bitmap);
3264 		if (err < 0)
3265 			goto out;
3266 	}
3267 
3268 	evlist = session->evlist;
3269 
3270 	symbol__init(&session->header.env);
3271 
3272 	if (perf_time__parse_str(&sched->ptime, sched->time_str) != 0) {
3273 		pr_err("Invalid time string\n");
3274 		err = -EINVAL;
3275 		goto out;
3276 	}
3277 
3278 	if (timehist_check_attr(sched, evlist) != 0)
3279 		goto out;
3280 
3281 	if (timehist_parse_prio_str(sched) != 0) {
3282 		pr_err("Invalid prio string\n");
3283 		goto out;
3284 	}
3285 
3286 	setup_pager();
3287 
3288 	/* prefer sched_waking if it is captured */
3289 	if (evlist__find_tracepoint_by_name(session->evlist, "sched:sched_waking"))
3290 		handlers[1].handler = timehist_sched_wakeup_ignore;
3291 
3292 	/* setup per-evsel handlers */
3293 	if (perf_session__set_tracepoints_handlers(session, handlers))
3294 		goto out;
3295 
3296 	/* sched_switch event at a minimum needs to exist */
3297 	if (!evlist__find_tracepoint_by_name(session->evlist, "sched:sched_switch")) {
3298 		pr_err("No sched_switch events found. Have you run 'perf sched record'?\n");
3299 		goto out;
3300 	}
3301 
3302 	if ((sched->show_migrations || sched->pre_migrations) &&
3303 		perf_session__set_tracepoints_handlers(session, migrate_handlers))
3304 		goto out;
3305 
3306 	/* pre-allocate struct for per-CPU idle stats */
3307 	sched->max_cpu.cpu = session->header.env.nr_cpus_online;
3308 	if (sched->max_cpu.cpu == 0)
3309 		sched->max_cpu.cpu = 4;
3310 	if (init_idle_threads(sched->max_cpu.cpu))
3311 		goto out;
3312 
3313 	/* summary_only implies summary option, but don't overwrite summary if set */
3314 	if (sched->summary_only)
3315 		sched->summary = sched->summary_only;
3316 
3317 	if (!sched->summary_only)
3318 		timehist_header(sched);
3319 
3320 	err = perf_session__process_events(session);
3321 	if (err) {
3322 		pr_err("Failed to process events, error %d", err);
3323 		goto out;
3324 	}
3325 
3326 	sched->nr_events      = evlist->stats.nr_events[0];
3327 	sched->nr_lost_events = evlist->stats.total_lost;
3328 	sched->nr_lost_chunks = evlist->stats.nr_events[PERF_RECORD_LOST];
3329 
3330 	if (sched->summary)
3331 		timehist_print_summary(sched, session);
3332 
3333 out:
3334 	free_idle_threads();
3335 	perf_session__delete(session);
3336 
3337 	return err;
3338 }
3339 
3340 
3341 static void print_bad_events(struct perf_sched *sched)
3342 {
3343 	if (sched->nr_unordered_timestamps && sched->nr_timestamps) {
3344 		printf("  INFO: %.3f%% unordered timestamps (%ld out of %ld)\n",
3345 			(double)sched->nr_unordered_timestamps/(double)sched->nr_timestamps*100.0,
3346 			sched->nr_unordered_timestamps, sched->nr_timestamps);
3347 	}
3348 	if (sched->nr_lost_events && sched->nr_events) {
3349 		printf("  INFO: %.3f%% lost events (%ld out of %ld, in %ld chunks)\n",
3350 			(double)sched->nr_lost_events/(double)sched->nr_events * 100.0,
3351 			sched->nr_lost_events, sched->nr_events, sched->nr_lost_chunks);
3352 	}
3353 	if (sched->nr_context_switch_bugs && sched->nr_timestamps) {
3354 		printf("  INFO: %.3f%% context switch bugs (%ld out of %ld)",
3355 			(double)sched->nr_context_switch_bugs/(double)sched->nr_timestamps*100.0,
3356 			sched->nr_context_switch_bugs, sched->nr_timestamps);
3357 		if (sched->nr_lost_events)
3358 			printf(" (due to lost events?)");
3359 		printf("\n");
3360 	}
3361 }
3362 
3363 static void __merge_work_atoms(struct rb_root_cached *root, struct work_atoms *data)
3364 {
3365 	struct rb_node **new = &(root->rb_root.rb_node), *parent = NULL;
3366 	struct work_atoms *this;
3367 	const char *comm = thread__comm_str(data->thread), *this_comm;
3368 	bool leftmost = true;
3369 
3370 	while (*new) {
3371 		int cmp;
3372 
3373 		this = container_of(*new, struct work_atoms, node);
3374 		parent = *new;
3375 
3376 		this_comm = thread__comm_str(this->thread);
3377 		cmp = strcmp(comm, this_comm);
3378 		if (cmp > 0) {
3379 			new = &((*new)->rb_left);
3380 		} else if (cmp < 0) {
3381 			new = &((*new)->rb_right);
3382 			leftmost = false;
3383 		} else {
3384 			this->num_merged++;
3385 			this->total_runtime += data->total_runtime;
3386 			this->nb_atoms += data->nb_atoms;
3387 			this->total_lat += data->total_lat;
3388 			list_splice(&data->work_list, &this->work_list);
3389 			if (this->max_lat < data->max_lat) {
3390 				this->max_lat = data->max_lat;
3391 				this->max_lat_start = data->max_lat_start;
3392 				this->max_lat_end = data->max_lat_end;
3393 			}
3394 			zfree(&data);
3395 			return;
3396 		}
3397 	}
3398 
3399 	data->num_merged++;
3400 	rb_link_node(&data->node, parent, new);
3401 	rb_insert_color_cached(&data->node, root, leftmost);
3402 }
3403 
3404 static void perf_sched__merge_lat(struct perf_sched *sched)
3405 {
3406 	struct work_atoms *data;
3407 	struct rb_node *node;
3408 
3409 	if (sched->skip_merge)
3410 		return;
3411 
3412 	while ((node = rb_first_cached(&sched->atom_root))) {
3413 		rb_erase_cached(node, &sched->atom_root);
3414 		data = rb_entry(node, struct work_atoms, node);
3415 		__merge_work_atoms(&sched->merged_atom_root, data);
3416 	}
3417 }
3418 
3419 static int setup_cpus_switch_event(struct perf_sched *sched)
3420 {
3421 	unsigned int i;
3422 
3423 	sched->cpu_last_switched = calloc(MAX_CPUS, sizeof(*(sched->cpu_last_switched)));
3424 	if (!sched->cpu_last_switched)
3425 		return -1;
3426 
3427 	sched->curr_pid = malloc(MAX_CPUS * sizeof(*(sched->curr_pid)));
3428 	if (!sched->curr_pid) {
3429 		zfree(&sched->cpu_last_switched);
3430 		return -1;
3431 	}
3432 
3433 	for (i = 0; i < MAX_CPUS; i++)
3434 		sched->curr_pid[i] = -1;
3435 
3436 	return 0;
3437 }
3438 
3439 static void free_cpus_switch_event(struct perf_sched *sched)
3440 {
3441 	zfree(&sched->curr_pid);
3442 	zfree(&sched->cpu_last_switched);
3443 }
3444 
3445 static int perf_sched__lat(struct perf_sched *sched)
3446 {
3447 	int rc = -1;
3448 	struct rb_node *next;
3449 
3450 	setup_pager();
3451 
3452 	if (setup_cpus_switch_event(sched))
3453 		return rc;
3454 
3455 	if (perf_sched__read_events(sched))
3456 		goto out_free_cpus_switch_event;
3457 
3458 	perf_sched__merge_lat(sched);
3459 	perf_sched__sort_lat(sched);
3460 
3461 	printf("\n -------------------------------------------------------------------------------------------------------------------------------------------\n");
3462 	printf("  Task                  |   Runtime ms  |  Count   | Avg delay ms    | Max delay ms    | Max delay start           | Max delay end          |\n");
3463 	printf(" -------------------------------------------------------------------------------------------------------------------------------------------\n");
3464 
3465 	next = rb_first_cached(&sched->sorted_atom_root);
3466 
3467 	while (next) {
3468 		struct work_atoms *work_list;
3469 
3470 		work_list = rb_entry(next, struct work_atoms, node);
3471 		output_lat_thread(sched, work_list);
3472 		next = rb_next(next);
3473 		thread__zput(work_list->thread);
3474 	}
3475 
3476 	printf(" -----------------------------------------------------------------------------------------------------------------\n");
3477 	printf("  TOTAL:                |%11.3f ms |%9" PRIu64 " |\n",
3478 		(double)sched->all_runtime / NSEC_PER_MSEC, sched->all_count);
3479 
3480 	printf(" ---------------------------------------------------\n");
3481 
3482 	print_bad_events(sched);
3483 	printf("\n");
3484 
3485 	rc = 0;
3486 
3487 out_free_cpus_switch_event:
3488 	free_cpus_switch_event(sched);
3489 	return rc;
3490 }
3491 
3492 static int setup_map_cpus(struct perf_sched *sched)
3493 {
3494 	sched->max_cpu.cpu  = sysconf(_SC_NPROCESSORS_CONF);
3495 
3496 	if (sched->map.comp) {
3497 		sched->map.comp_cpus = zalloc(sched->max_cpu.cpu * sizeof(int));
3498 		if (!sched->map.comp_cpus)
3499 			return -1;
3500 	}
3501 
3502 	if (sched->map.cpus_str) {
3503 		sched->map.cpus = perf_cpu_map__new(sched->map.cpus_str);
3504 		if (!sched->map.cpus) {
3505 			pr_err("failed to get cpus map from %s\n", sched->map.cpus_str);
3506 			zfree(&sched->map.comp_cpus);
3507 			return -1;
3508 		}
3509 	}
3510 
3511 	return 0;
3512 }
3513 
3514 static int setup_color_pids(struct perf_sched *sched)
3515 {
3516 	struct perf_thread_map *map;
3517 
3518 	if (!sched->map.color_pids_str)
3519 		return 0;
3520 
3521 	map = thread_map__new_by_tid_str(sched->map.color_pids_str);
3522 	if (!map) {
3523 		pr_err("failed to get thread map from %s\n", sched->map.color_pids_str);
3524 		return -1;
3525 	}
3526 
3527 	sched->map.color_pids = map;
3528 	return 0;
3529 }
3530 
3531 static int setup_color_cpus(struct perf_sched *sched)
3532 {
3533 	struct perf_cpu_map *map;
3534 
3535 	if (!sched->map.color_cpus_str)
3536 		return 0;
3537 
3538 	map = perf_cpu_map__new(sched->map.color_cpus_str);
3539 	if (!map) {
3540 		pr_err("failed to get thread map from %s\n", sched->map.color_cpus_str);
3541 		return -1;
3542 	}
3543 
3544 	sched->map.color_cpus = map;
3545 	return 0;
3546 }
3547 
3548 static int perf_sched__map(struct perf_sched *sched)
3549 {
3550 	int rc = -1;
3551 
3552 	sched->curr_thread = calloc(MAX_CPUS, sizeof(*(sched->curr_thread)));
3553 	if (!sched->curr_thread)
3554 		return rc;
3555 
3556 	sched->curr_out_thread = calloc(MAX_CPUS, sizeof(*(sched->curr_out_thread)));
3557 	if (!sched->curr_out_thread)
3558 		return rc;
3559 
3560 	if (setup_cpus_switch_event(sched))
3561 		goto out_free_curr_thread;
3562 
3563 	if (setup_map_cpus(sched))
3564 		goto out_free_cpus_switch_event;
3565 
3566 	if (setup_color_pids(sched))
3567 		goto out_put_map_cpus;
3568 
3569 	if (setup_color_cpus(sched))
3570 		goto out_put_color_pids;
3571 
3572 	setup_pager();
3573 	if (perf_sched__read_events(sched))
3574 		goto out_put_color_cpus;
3575 
3576 	rc = 0;
3577 	print_bad_events(sched);
3578 
3579 out_put_color_cpus:
3580 	perf_cpu_map__put(sched->map.color_cpus);
3581 
3582 out_put_color_pids:
3583 	perf_thread_map__put(sched->map.color_pids);
3584 
3585 out_put_map_cpus:
3586 	zfree(&sched->map.comp_cpus);
3587 	perf_cpu_map__put(sched->map.cpus);
3588 
3589 out_free_cpus_switch_event:
3590 	free_cpus_switch_event(sched);
3591 
3592 out_free_curr_thread:
3593 	zfree(&sched->curr_thread);
3594 	return rc;
3595 }
3596 
3597 static int perf_sched__replay(struct perf_sched *sched)
3598 {
3599 	int ret;
3600 	unsigned long i;
3601 
3602 	mutex_init(&sched->start_work_mutex);
3603 	mutex_init(&sched->work_done_wait_mutex);
3604 
3605 	ret = setup_cpus_switch_event(sched);
3606 	if (ret)
3607 		goto out_mutex_destroy;
3608 
3609 	calibrate_run_measurement_overhead(sched);
3610 	calibrate_sleep_measurement_overhead(sched);
3611 
3612 	test_calibrations(sched);
3613 
3614 	ret = perf_sched__read_events(sched);
3615 	if (ret)
3616 		goto out_free_cpus_switch_event;
3617 
3618 	printf("nr_run_events:        %ld\n", sched->nr_run_events);
3619 	printf("nr_sleep_events:      %ld\n", sched->nr_sleep_events);
3620 	printf("nr_wakeup_events:     %ld\n", sched->nr_wakeup_events);
3621 
3622 	if (sched->targetless_wakeups)
3623 		printf("target-less wakeups:  %ld\n", sched->targetless_wakeups);
3624 	if (sched->multitarget_wakeups)
3625 		printf("multi-target wakeups: %ld\n", sched->multitarget_wakeups);
3626 	if (sched->nr_run_events_optimized)
3627 		printf("run atoms optimized: %ld\n",
3628 			sched->nr_run_events_optimized);
3629 
3630 	print_task_traces(sched);
3631 	add_cross_task_wakeups(sched);
3632 
3633 	sched->thread_funcs_exit = false;
3634 	create_tasks(sched);
3635 	printf("------------------------------------------------------------\n");
3636 	if (sched->replay_repeat == 0)
3637 		sched->replay_repeat = UINT_MAX;
3638 
3639 	for (i = 0; i < sched->replay_repeat; i++)
3640 		run_one_test(sched);
3641 
3642 	sched->thread_funcs_exit = true;
3643 	destroy_tasks(sched);
3644 
3645 out_free_cpus_switch_event:
3646 	free_cpus_switch_event(sched);
3647 
3648 out_mutex_destroy:
3649 	mutex_destroy(&sched->start_work_mutex);
3650 	mutex_destroy(&sched->work_done_wait_mutex);
3651 	return ret;
3652 }
3653 
3654 static void setup_sorting(struct perf_sched *sched, const struct option *options,
3655 			  const char * const usage_msg[])
3656 {
3657 	char *tmp, *tok, *str = strdup(sched->sort_order);
3658 
3659 	for (tok = strtok_r(str, ", ", &tmp);
3660 			tok; tok = strtok_r(NULL, ", ", &tmp)) {
3661 		if (sort_dimension__add(tok, &sched->sort_list) < 0) {
3662 			usage_with_options_msg(usage_msg, options,
3663 					"Unknown --sort key: `%s'", tok);
3664 		}
3665 	}
3666 
3667 	free(str);
3668 
3669 	sort_dimension__add("pid", &sched->cmp_pid);
3670 }
3671 
3672 static bool schedstat_events_exposed(void)
3673 {
3674 	/*
3675 	 * Select "sched:sched_stat_wait" event to check
3676 	 * whether schedstat tracepoints are exposed.
3677 	 */
3678 	return IS_ERR(trace_event__tp_format("sched", "sched_stat_wait")) ?
3679 		false : true;
3680 }
3681 
3682 static int __cmd_record(int argc, const char **argv)
3683 {
3684 	unsigned int rec_argc, i, j;
3685 	char **rec_argv;
3686 	const char **rec_argv_copy;
3687 	const char * const record_args[] = {
3688 		"record",
3689 		"-a",
3690 		"-R",
3691 		"-m", "1024",
3692 		"-c", "1",
3693 		"-e", "sched:sched_switch",
3694 		"-e", "sched:sched_stat_runtime",
3695 		"-e", "sched:sched_process_fork",
3696 		"-e", "sched:sched_wakeup_new",
3697 		"-e", "sched:sched_migrate_task",
3698 	};
3699 
3700 	/*
3701 	 * The tracepoints trace_sched_stat_{wait, sleep, iowait}
3702 	 * are not exposed to user if CONFIG_SCHEDSTATS is not set,
3703 	 * to prevent "perf sched record" execution failure, determine
3704 	 * whether to record schedstat events according to actual situation.
3705 	 */
3706 	const char * const schedstat_args[] = {
3707 		"-e", "sched:sched_stat_wait",
3708 		"-e", "sched:sched_stat_sleep",
3709 		"-e", "sched:sched_stat_iowait",
3710 	};
3711 	unsigned int schedstat_argc = schedstat_events_exposed() ?
3712 		ARRAY_SIZE(schedstat_args) : 0;
3713 
3714 	struct tep_event *waking_event;
3715 	int ret;
3716 
3717 	/*
3718 	 * +2 for either "-e", "sched:sched_wakeup" or
3719 	 * "-e", "sched:sched_waking"
3720 	 */
3721 	rec_argc = ARRAY_SIZE(record_args) + 2 + schedstat_argc + argc - 1;
3722 	rec_argv = calloc(rec_argc + 1, sizeof(char *));
3723 	if (rec_argv == NULL)
3724 		return -ENOMEM;
3725 	rec_argv_copy = calloc(rec_argc + 1, sizeof(char *));
3726 	if (rec_argv_copy == NULL) {
3727 		free(rec_argv);
3728 		return -ENOMEM;
3729 	}
3730 
3731 	for (i = 0; i < ARRAY_SIZE(record_args); i++)
3732 		rec_argv[i] = strdup(record_args[i]);
3733 
3734 	rec_argv[i++] = strdup("-e");
3735 	waking_event = trace_event__tp_format("sched", "sched_waking");
3736 	if (!IS_ERR(waking_event))
3737 		rec_argv[i++] = strdup("sched:sched_waking");
3738 	else
3739 		rec_argv[i++] = strdup("sched:sched_wakeup");
3740 
3741 	for (j = 0; j < schedstat_argc; j++)
3742 		rec_argv[i++] = strdup(schedstat_args[j]);
3743 
3744 	for (j = 1; j < (unsigned int)argc; j++, i++)
3745 		rec_argv[i] = strdup(argv[j]);
3746 
3747 	BUG_ON(i != rec_argc);
3748 
3749 	memcpy(rec_argv_copy, rec_argv, sizeof(char *) * rec_argc);
3750 	ret = cmd_record(rec_argc, rec_argv_copy);
3751 
3752 	for (i = 0; i < rec_argc; i++)
3753 		free(rec_argv[i]);
3754 	free(rec_argv);
3755 	free(rec_argv_copy);
3756 
3757 	return ret;
3758 }
3759 
3760 int cmd_sched(int argc, const char **argv)
3761 {
3762 	static const char default_sort_order[] = "avg, max, switch, runtime";
3763 	struct perf_sched sched = {
3764 		.cmp_pid	      = LIST_HEAD_INIT(sched.cmp_pid),
3765 		.sort_list	      = LIST_HEAD_INIT(sched.sort_list),
3766 		.sort_order	      = default_sort_order,
3767 		.replay_repeat	      = 10,
3768 		.profile_cpu	      = -1,
3769 		.next_shortname1      = 'A',
3770 		.next_shortname2      = '0',
3771 		.skip_merge           = 0,
3772 		.show_callchain	      = 1,
3773 		.max_stack            = 5,
3774 	};
3775 	const struct option sched_options[] = {
3776 	OPT_STRING('i', "input", &input_name, "file",
3777 		    "input file name"),
3778 	OPT_INCR('v', "verbose", &verbose,
3779 		    "be more verbose (show symbol address, etc)"),
3780 	OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace,
3781 		    "dump raw trace in ASCII"),
3782 	OPT_BOOLEAN('f', "force", &sched.force, "don't complain, do it"),
3783 	OPT_END()
3784 	};
3785 	const struct option latency_options[] = {
3786 	OPT_STRING('s', "sort", &sched.sort_order, "key[,key2...]",
3787 		   "sort by key(s): runtime, switch, avg, max"),
3788 	OPT_INTEGER('C', "CPU", &sched.profile_cpu,
3789 		    "CPU to profile on"),
3790 	OPT_BOOLEAN('p', "pids", &sched.skip_merge,
3791 		    "latency stats per pid instead of per comm"),
3792 	OPT_PARENT(sched_options)
3793 	};
3794 	const struct option replay_options[] = {
3795 	OPT_UINTEGER('r', "repeat", &sched.replay_repeat,
3796 		     "repeat the workload replay N times (0: infinite)"),
3797 	OPT_PARENT(sched_options)
3798 	};
3799 	const struct option map_options[] = {
3800 	OPT_BOOLEAN(0, "compact", &sched.map.comp,
3801 		    "map output in compact mode"),
3802 	OPT_STRING(0, "color-pids", &sched.map.color_pids_str, "pids",
3803 		   "highlight given pids in map"),
3804 	OPT_STRING(0, "color-cpus", &sched.map.color_cpus_str, "cpus",
3805                     "highlight given CPUs in map"),
3806 	OPT_STRING(0, "cpus", &sched.map.cpus_str, "cpus",
3807                     "display given CPUs in map"),
3808 	OPT_STRING(0, "task-name", &sched.map.task_name, "task",
3809 		"map output only for the given task name(s)."),
3810 	OPT_BOOLEAN(0, "fuzzy-name", &sched.map.fuzzy,
3811 		"given command name can be partially matched (fuzzy matching)"),
3812 	OPT_PARENT(sched_options)
3813 	};
3814 	const struct option timehist_options[] = {
3815 	OPT_STRING('k', "vmlinux", &symbol_conf.vmlinux_name,
3816 		   "file", "vmlinux pathname"),
3817 	OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
3818 		   "file", "kallsyms pathname"),
3819 	OPT_BOOLEAN('g', "call-graph", &sched.show_callchain,
3820 		    "Display call chains if present (default on)"),
3821 	OPT_UINTEGER(0, "max-stack", &sched.max_stack,
3822 		   "Maximum number of functions to display backtrace."),
3823 	OPT_STRING(0, "symfs", &symbol_conf.symfs, "directory",
3824 		    "Look for files with symbols relative to this directory"),
3825 	OPT_BOOLEAN('s', "summary", &sched.summary_only,
3826 		    "Show only syscall summary with statistics"),
3827 	OPT_BOOLEAN('S', "with-summary", &sched.summary,
3828 		    "Show all syscalls and summary with statistics"),
3829 	OPT_BOOLEAN('w', "wakeups", &sched.show_wakeups, "Show wakeup events"),
3830 	OPT_BOOLEAN('n', "next", &sched.show_next, "Show next task"),
3831 	OPT_BOOLEAN('M', "migrations", &sched.show_migrations, "Show migration events"),
3832 	OPT_BOOLEAN('V', "cpu-visual", &sched.show_cpu_visual, "Add CPU visual"),
3833 	OPT_BOOLEAN('I', "idle-hist", &sched.idle_hist, "Show idle events only"),
3834 	OPT_STRING(0, "time", &sched.time_str, "str",
3835 		   "Time span for analysis (start,stop)"),
3836 	OPT_BOOLEAN(0, "state", &sched.show_state, "Show task state when sched-out"),
3837 	OPT_STRING('p', "pid", &symbol_conf.pid_list_str, "pid[,pid...]",
3838 		   "analyze events only for given process id(s)"),
3839 	OPT_STRING('t', "tid", &symbol_conf.tid_list_str, "tid[,tid...]",
3840 		   "analyze events only for given thread id(s)"),
3841 	OPT_STRING('C', "cpu", &cpu_list, "cpu", "list of cpus to profile"),
3842 	OPT_BOOLEAN(0, "show-prio", &sched.show_prio, "Show task priority"),
3843 	OPT_STRING(0, "prio", &sched.prio_str, "prio",
3844 		   "analyze events only for given task priority(ies)"),
3845 	OPT_BOOLEAN('P', "pre-migrations", &sched.pre_migrations, "Show pre-migration wait time"),
3846 	OPT_PARENT(sched_options)
3847 	};
3848 
3849 	const char * const latency_usage[] = {
3850 		"perf sched latency [<options>]",
3851 		NULL
3852 	};
3853 	const char * const replay_usage[] = {
3854 		"perf sched replay [<options>]",
3855 		NULL
3856 	};
3857 	const char * const map_usage[] = {
3858 		"perf sched map [<options>]",
3859 		NULL
3860 	};
3861 	const char * const timehist_usage[] = {
3862 		"perf sched timehist [<options>]",
3863 		NULL
3864 	};
3865 	const char *const sched_subcommands[] = { "record", "latency", "map",
3866 						  "replay", "script",
3867 						  "timehist", NULL };
3868 	const char *sched_usage[] = {
3869 		NULL,
3870 		NULL
3871 	};
3872 	struct trace_sched_handler lat_ops  = {
3873 		.wakeup_event	    = latency_wakeup_event,
3874 		.switch_event	    = latency_switch_event,
3875 		.runtime_event	    = latency_runtime_event,
3876 		.migrate_task_event = latency_migrate_task_event,
3877 	};
3878 	struct trace_sched_handler map_ops  = {
3879 		.switch_event	    = map_switch_event,
3880 	};
3881 	struct trace_sched_handler replay_ops  = {
3882 		.wakeup_event	    = replay_wakeup_event,
3883 		.switch_event	    = replay_switch_event,
3884 		.fork_event	    = replay_fork_event,
3885 	};
3886 	int ret;
3887 
3888 	perf_tool__init(&sched.tool, /*ordered_events=*/true);
3889 	sched.tool.sample	 = perf_sched__process_tracepoint_sample;
3890 	sched.tool.comm		 = perf_sched__process_comm;
3891 	sched.tool.namespaces	 = perf_event__process_namespaces;
3892 	sched.tool.lost		 = perf_event__process_lost;
3893 	sched.tool.fork		 = perf_sched__process_fork_event;
3894 
3895 	argc = parse_options_subcommand(argc, argv, sched_options, sched_subcommands,
3896 					sched_usage, PARSE_OPT_STOP_AT_NON_OPTION);
3897 	if (!argc)
3898 		usage_with_options(sched_usage, sched_options);
3899 
3900 	/*
3901 	 * Aliased to 'perf script' for now:
3902 	 */
3903 	if (!strcmp(argv[0], "script")) {
3904 		return cmd_script(argc, argv);
3905 	} else if (strlen(argv[0]) > 2 && strstarts("record", argv[0])) {
3906 		return __cmd_record(argc, argv);
3907 	} else if (strlen(argv[0]) > 2 && strstarts("latency", argv[0])) {
3908 		sched.tp_handler = &lat_ops;
3909 		if (argc > 1) {
3910 			argc = parse_options(argc, argv, latency_options, latency_usage, 0);
3911 			if (argc)
3912 				usage_with_options(latency_usage, latency_options);
3913 		}
3914 		setup_sorting(&sched, latency_options, latency_usage);
3915 		return perf_sched__lat(&sched);
3916 	} else if (!strcmp(argv[0], "map")) {
3917 		if (argc) {
3918 			argc = parse_options(argc, argv, map_options, map_usage, 0);
3919 			if (argc)
3920 				usage_with_options(map_usage, map_options);
3921 
3922 			if (sched.map.task_name) {
3923 				sched.map.task_names = strlist__new(sched.map.task_name, NULL);
3924 				if (sched.map.task_names == NULL) {
3925 					fprintf(stderr, "Failed to parse task names\n");
3926 					return -1;
3927 				}
3928 			}
3929 		}
3930 		sched.tp_handler = &map_ops;
3931 		setup_sorting(&sched, latency_options, latency_usage);
3932 		return perf_sched__map(&sched);
3933 	} else if (strlen(argv[0]) > 2 && strstarts("replay", argv[0])) {
3934 		sched.tp_handler = &replay_ops;
3935 		if (argc) {
3936 			argc = parse_options(argc, argv, replay_options, replay_usage, 0);
3937 			if (argc)
3938 				usage_with_options(replay_usage, replay_options);
3939 		}
3940 		return perf_sched__replay(&sched);
3941 	} else if (!strcmp(argv[0], "timehist")) {
3942 		if (argc) {
3943 			argc = parse_options(argc, argv, timehist_options,
3944 					     timehist_usage, 0);
3945 			if (argc)
3946 				usage_with_options(timehist_usage, timehist_options);
3947 		}
3948 		if ((sched.show_wakeups || sched.show_next) &&
3949 		    sched.summary_only) {
3950 			pr_err(" Error: -s and -[n|w] are mutually exclusive.\n");
3951 			parse_options_usage(timehist_usage, timehist_options, "s", true);
3952 			if (sched.show_wakeups)
3953 				parse_options_usage(NULL, timehist_options, "w", true);
3954 			if (sched.show_next)
3955 				parse_options_usage(NULL, timehist_options, "n", true);
3956 			return -EINVAL;
3957 		}
3958 		ret = symbol__validate_sym_arguments();
3959 		if (ret)
3960 			return ret;
3961 
3962 		return perf_sched__timehist(&sched);
3963 	} else {
3964 		usage_with_options(sched_usage, sched_options);
3965 	}
3966 
3967 	/* free usage string allocated by parse_options_subcommand */
3968 	free((void *)sched_usage[0]);
3969 
3970 	return 0;
3971 }
3972