• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #undef TRACE_SYSTEM
3 #define TRACE_SYSTEM sched
4 
5 #if !defined(_TRACE_SCHED_H) || defined(TRACE_HEADER_MULTI_READ)
6 #define _TRACE_SCHED_H
7 
8 #include <linux/kthread.h>
9 #include <linux/sched/numa_balancing.h>
10 #include <linux/tracepoint.h>
11 #include <linux/binfmts.h>
12 
13 /*
14  * Tracepoint for calling kthread_stop, performed to end a kthread:
15  */
16 TRACE_EVENT(sched_kthread_stop,
17 
18 	TP_PROTO(struct task_struct *t),
19 
20 	TP_ARGS(t),
21 
22 	TP_STRUCT__entry(
23 		__array(	char,	comm,	TASK_COMM_LEN	)
24 		__field(	pid_t,	pid			)
25 	),
26 
27 	TP_fast_assign(
28 		memcpy(__entry->comm, t->comm, TASK_COMM_LEN);
29 		__entry->pid	= t->pid;
30 	),
31 
32 	TP_printk("comm=%s pid=%d", __entry->comm, __entry->pid)
33 );
34 
35 /*
36  * Tracepoint for the return value of the kthread stopping:
37  */
38 TRACE_EVENT(sched_kthread_stop_ret,
39 
40 	TP_PROTO(int ret),
41 
42 	TP_ARGS(ret),
43 
44 	TP_STRUCT__entry(
45 		__field(	int,	ret	)
46 	),
47 
48 	TP_fast_assign(
49 		__entry->ret	= ret;
50 	),
51 
52 	TP_printk("ret=%d", __entry->ret)
53 );
54 
55 /**
56  * sched_kthread_work_queue_work - called when a work gets queued
57  * @worker:	pointer to the kthread_worker
58  * @work:	pointer to struct kthread_work
59  *
60  * This event occurs when a work is queued immediately or once a
61  * delayed work is actually queued (ie: once the delay has been
62  * reached).
63  */
64 TRACE_EVENT(sched_kthread_work_queue_work,
65 
66 	TP_PROTO(struct kthread_worker *worker,
67 		 struct kthread_work *work),
68 
69 	TP_ARGS(worker, work),
70 
71 	TP_STRUCT__entry(
72 		__field( void *,	work	)
73 		__field( void *,	function)
74 		__field( void *,	worker)
75 	),
76 
77 	TP_fast_assign(
78 		__entry->work		= work;
79 		__entry->function	= work->func;
80 		__entry->worker		= worker;
81 	),
82 
83 	TP_printk("work struct=%p function=%ps worker=%p",
84 		  __entry->work, __entry->function, __entry->worker)
85 );
86 
87 /**
88  * sched_kthread_work_execute_start - called immediately before the work callback
89  * @work:	pointer to struct kthread_work
90  *
91  * Allows to track kthread work execution.
92  */
93 TRACE_EVENT(sched_kthread_work_execute_start,
94 
95 	TP_PROTO(struct kthread_work *work),
96 
97 	TP_ARGS(work),
98 
99 	TP_STRUCT__entry(
100 		__field( void *,	work	)
101 		__field( void *,	function)
102 	),
103 
104 	TP_fast_assign(
105 		__entry->work		= work;
106 		__entry->function	= work->func;
107 	),
108 
109 	TP_printk("work struct %p: function %ps", __entry->work, __entry->function)
110 );
111 
112 /**
113  * sched_kthread_work_execute_end - called immediately after the work callback
114  * @work:	pointer to struct work_struct
115  * @function:   pointer to worker function
116  *
117  * Allows to track workqueue execution.
118  */
119 TRACE_EVENT(sched_kthread_work_execute_end,
120 
121 	TP_PROTO(struct kthread_work *work, kthread_work_func_t function),
122 
123 	TP_ARGS(work, function),
124 
125 	TP_STRUCT__entry(
126 		__field( void *,	work	)
127 		__field( void *,	function)
128 	),
129 
130 	TP_fast_assign(
131 		__entry->work		= work;
132 		__entry->function	= function;
133 	),
134 
135 	TP_printk("work struct %p: function %ps", __entry->work, __entry->function)
136 );
137 
138 /*
139  * Tracepoint for waking up a task:
140  */
141 DECLARE_EVENT_CLASS(sched_wakeup_template,
142 
143 	TP_PROTO(struct task_struct *p),
144 
145 	TP_ARGS(__perf_task(p)),
146 
147 	TP_STRUCT__entry(
148 		__array(	char,	comm,	TASK_COMM_LEN	)
149 		__field(	pid_t,	pid			)
150 		__field(	int,	prio			)
151 		__field(	int,	target_cpu		)
152 	),
153 
154 	TP_fast_assign(
155 		memcpy(__entry->comm, p->comm, TASK_COMM_LEN);
156 		__entry->pid		= p->pid;
157 		__entry->prio		= p->prio; /* XXX SCHED_DEADLINE */
158 		__entry->target_cpu	= task_cpu(p);
159 	),
160 
161 	TP_printk("comm=%s pid=%d prio=%d target_cpu=%03d",
162 		  __entry->comm, __entry->pid, __entry->prio,
163 		  __entry->target_cpu)
164 );
165 
166 /*
167  * Tracepoint called when waking a task; this tracepoint is guaranteed to be
168  * called from the waking context.
169  */
170 DEFINE_EVENT(sched_wakeup_template, sched_waking,
171 	     TP_PROTO(struct task_struct *p),
172 	     TP_ARGS(p));
173 
174 /*
175  * Tracepoint called when the task is actually woken; p->state == TASK_RUNNING.
176  * It is not always called from the waking context.
177  */
178 DEFINE_EVENT(sched_wakeup_template, sched_wakeup,
179 	     TP_PROTO(struct task_struct *p),
180 	     TP_ARGS(p));
181 
182 /*
183  * Tracepoint for waking up a new task:
184  */
185 DEFINE_EVENT(sched_wakeup_template, sched_wakeup_new,
186 	     TP_PROTO(struct task_struct *p),
187 	     TP_ARGS(p));
188 
189 #ifdef CREATE_TRACE_POINTS
__trace_sched_switch_state(bool preempt,struct task_struct * p)190 static inline long __trace_sched_switch_state(bool preempt, struct task_struct *p)
191 {
192 	unsigned int state;
193 
194 #ifdef CONFIG_SCHED_DEBUG
195 	BUG_ON(p != current);
196 #endif /* CONFIG_SCHED_DEBUG */
197 
198 	/*
199 	 * Preemption ignores task state, therefore preempted tasks are always
200 	 * RUNNING (we will not have dequeued if state != RUNNING).
201 	 */
202 	if (preempt)
203 		return TASK_REPORT_MAX;
204 
205 	/*
206 	 * task_state_index() uses fls() and returns a value from 0-8 range.
207 	 * Decrement it by 1 (except TASK_RUNNING state i.e 0) before using
208 	 * it for left shift operation to get the correct task->state
209 	 * mapping.
210 	 */
211 	state = task_state_index(p);
212 
213 	return state ? (1 << (state - 1)) : state;
214 }
215 #endif /* CREATE_TRACE_POINTS */
216 
217 /*
218  * Tracepoint for task switches, performed by the scheduler:
219  */
220 TRACE_EVENT(sched_switch,
221 
222 	TP_PROTO(bool preempt,
223 		 struct task_struct *prev,
224 		 struct task_struct *next),
225 
226 	TP_ARGS(preempt, prev, next),
227 
228 	TP_STRUCT__entry(
229 		__array(	char,	prev_comm,	TASK_COMM_LEN	)
230 		__field(	pid_t,	prev_pid			)
231 		__field(	int,	prev_prio			)
232 		__field(	long,	prev_state			)
233 		__array(	char,	next_comm,	TASK_COMM_LEN	)
234 		__field(	pid_t,	next_pid			)
235 		__field(	int,	next_prio			)
236 	),
237 
238 	TP_fast_assign(
239 		memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN);
240 		__entry->prev_pid	= prev->pid;
241 		__entry->prev_prio	= prev->prio;
242 		__entry->prev_state	= __trace_sched_switch_state(preempt, prev);
243 		memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN);
244 		__entry->next_pid	= next->pid;
245 		__entry->next_prio	= next->prio;
246 		/* XXX SCHED_DEADLINE */
247 	),
248 
249 	TP_printk("prev_comm=%s prev_pid=%d prev_prio=%d prev_state=%s%s ==> next_comm=%s next_pid=%d next_prio=%d",
250 		__entry->prev_comm, __entry->prev_pid, __entry->prev_prio,
251 
252 		(__entry->prev_state & (TASK_REPORT_MAX - 1)) ?
253 		  __print_flags(__entry->prev_state & (TASK_REPORT_MAX - 1), "|",
254 				{ TASK_INTERRUPTIBLE, "S" },
255 				{ TASK_UNINTERRUPTIBLE, "D" },
256 				{ __TASK_STOPPED, "T" },
257 				{ __TASK_TRACED, "t" },
258 				{ EXIT_DEAD, "X" },
259 				{ EXIT_ZOMBIE, "Z" },
260 				{ TASK_PARKED, "P" },
261 				{ TASK_DEAD, "I" }) :
262 		  "R",
263 
264 		__entry->prev_state & TASK_REPORT_MAX ? "+" : "",
265 		__entry->next_comm, __entry->next_pid, __entry->next_prio)
266 );
267 
268 /*
269  * Tracepoint for a task being migrated:
270  */
271 TRACE_EVENT(sched_migrate_task,
272 
273 	TP_PROTO(struct task_struct *p, int dest_cpu),
274 
275 	TP_ARGS(p, dest_cpu),
276 
277 	TP_STRUCT__entry(
278 		__array(	char,	comm,	TASK_COMM_LEN	)
279 		__field(	pid_t,	pid			)
280 		__field(	int,	prio			)
281 		__field(	int,	orig_cpu		)
282 		__field(	int,	dest_cpu		)
283 	),
284 
285 	TP_fast_assign(
286 		memcpy(__entry->comm, p->comm, TASK_COMM_LEN);
287 		__entry->pid		= p->pid;
288 		__entry->prio		= p->prio; /* XXX SCHED_DEADLINE */
289 		__entry->orig_cpu	= task_cpu(p);
290 		__entry->dest_cpu	= dest_cpu;
291 	),
292 
293 	TP_printk("comm=%s pid=%d prio=%d orig_cpu=%d dest_cpu=%d",
294 		  __entry->comm, __entry->pid, __entry->prio,
295 		  __entry->orig_cpu, __entry->dest_cpu)
296 );
297 
298 DECLARE_EVENT_CLASS(sched_process_template,
299 
300 	TP_PROTO(struct task_struct *p),
301 
302 	TP_ARGS(p),
303 
304 	TP_STRUCT__entry(
305 		__array(	char,	comm,	TASK_COMM_LEN	)
306 		__field(	pid_t,	pid			)
307 		__field(	int,	prio			)
308 	),
309 
310 	TP_fast_assign(
311 		memcpy(__entry->comm, p->comm, TASK_COMM_LEN);
312 		__entry->pid		= p->pid;
313 		__entry->prio		= p->prio; /* XXX SCHED_DEADLINE */
314 	),
315 
316 	TP_printk("comm=%s pid=%d prio=%d",
317 		  __entry->comm, __entry->pid, __entry->prio)
318 );
319 
320 /*
321  * Tracepoint for freeing a task:
322  */
323 DEFINE_EVENT(sched_process_template, sched_process_free,
324 	     TP_PROTO(struct task_struct *p),
325 	     TP_ARGS(p));
326 
327 /*
328  * Tracepoint for a task exiting:
329  */
330 DEFINE_EVENT(sched_process_template, sched_process_exit,
331 	     TP_PROTO(struct task_struct *p),
332 	     TP_ARGS(p));
333 
334 /*
335  * Tracepoint for waiting on task to unschedule:
336  */
337 DEFINE_EVENT(sched_process_template, sched_wait_task,
338 	TP_PROTO(struct task_struct *p),
339 	TP_ARGS(p));
340 
341 /*
342  * Tracepoint for a waiting task:
343  */
344 TRACE_EVENT(sched_process_wait,
345 
346 	TP_PROTO(struct pid *pid),
347 
348 	TP_ARGS(pid),
349 
350 	TP_STRUCT__entry(
351 		__array(	char,	comm,	TASK_COMM_LEN	)
352 		__field(	pid_t,	pid			)
353 		__field(	int,	prio			)
354 	),
355 
356 	TP_fast_assign(
357 		memcpy(__entry->comm, current->comm, TASK_COMM_LEN);
358 		__entry->pid		= pid_nr(pid);
359 		__entry->prio		= current->prio; /* XXX SCHED_DEADLINE */
360 	),
361 
362 	TP_printk("comm=%s pid=%d prio=%d",
363 		  __entry->comm, __entry->pid, __entry->prio)
364 );
365 
366 /*
367  * Tracepoint for kernel_clone:
368  */
369 TRACE_EVENT(sched_process_fork,
370 
371 	TP_PROTO(struct task_struct *parent, struct task_struct *child),
372 
373 	TP_ARGS(parent, child),
374 
375 	TP_STRUCT__entry(
376 		__array(	char,	parent_comm,	TASK_COMM_LEN	)
377 		__field(	pid_t,	parent_pid			)
378 		__array(	char,	child_comm,	TASK_COMM_LEN	)
379 		__field(	pid_t,	child_pid			)
380 	),
381 
382 	TP_fast_assign(
383 		memcpy(__entry->parent_comm, parent->comm, TASK_COMM_LEN);
384 		__entry->parent_pid	= parent->pid;
385 		memcpy(__entry->child_comm, child->comm, TASK_COMM_LEN);
386 		__entry->child_pid	= child->pid;
387 	),
388 
389 	TP_printk("comm=%s pid=%d child_comm=%s child_pid=%d",
390 		__entry->parent_comm, __entry->parent_pid,
391 		__entry->child_comm, __entry->child_pid)
392 );
393 
394 /*
395  * Tracepoint for exec:
396  */
397 TRACE_EVENT(sched_process_exec,
398 
399 	TP_PROTO(struct task_struct *p, pid_t old_pid,
400 		 struct linux_binprm *bprm),
401 
402 	TP_ARGS(p, old_pid, bprm),
403 
404 	TP_STRUCT__entry(
405 		__string(	filename,	bprm->filename	)
406 		__field(	pid_t,		pid		)
407 		__field(	pid_t,		old_pid		)
408 	),
409 
410 	TP_fast_assign(
411 		__assign_str(filename, bprm->filename);
412 		__entry->pid		= p->pid;
413 		__entry->old_pid	= old_pid;
414 	),
415 
416 	TP_printk("filename=%s pid=%d old_pid=%d", __get_str(filename),
417 		  __entry->pid, __entry->old_pid)
418 );
419 
420 
421 #ifdef CONFIG_SCHEDSTATS
422 #define DEFINE_EVENT_SCHEDSTAT DEFINE_EVENT
423 #define DECLARE_EVENT_CLASS_SCHEDSTAT DECLARE_EVENT_CLASS
424 #else
425 #define DEFINE_EVENT_SCHEDSTAT DEFINE_EVENT_NOP
426 #define DECLARE_EVENT_CLASS_SCHEDSTAT DECLARE_EVENT_CLASS_NOP
427 #endif
428 
429 /*
430  * XXX the below sched_stat tracepoints only apply to SCHED_OTHER/BATCH/IDLE
431  *     adding sched_stat support to SCHED_FIFO/RR would be welcome.
432  */
433 DECLARE_EVENT_CLASS_SCHEDSTAT(sched_stat_template,
434 
435 	TP_PROTO(struct task_struct *tsk, u64 delay),
436 
437 	TP_ARGS(__perf_task(tsk), __perf_count(delay)),
438 
439 	TP_STRUCT__entry(
440 		__array( char,	comm,	TASK_COMM_LEN	)
441 		__field( pid_t,	pid			)
442 		__field( u64,	delay			)
443 	),
444 
445 	TP_fast_assign(
446 		memcpy(__entry->comm, tsk->comm, TASK_COMM_LEN);
447 		__entry->pid	= tsk->pid;
448 		__entry->delay	= delay;
449 	),
450 
451 	TP_printk("comm=%s pid=%d delay=%Lu [ns]",
452 			__entry->comm, __entry->pid,
453 			(unsigned long long)__entry->delay)
454 );
455 
456 /*
457  * Tracepoint for accounting wait time (time the task is runnable
458  * but not actually running due to scheduler contention).
459  */
460 DEFINE_EVENT_SCHEDSTAT(sched_stat_template, sched_stat_wait,
461 	     TP_PROTO(struct task_struct *tsk, u64 delay),
462 	     TP_ARGS(tsk, delay));
463 
464 /*
465  * Tracepoint for accounting sleep time (time the task is not runnable,
466  * including iowait, see below).
467  */
468 DEFINE_EVENT_SCHEDSTAT(sched_stat_template, sched_stat_sleep,
469 	     TP_PROTO(struct task_struct *tsk, u64 delay),
470 	     TP_ARGS(tsk, delay));
471 
472 /*
473  * Tracepoint for accounting iowait time (time the task is not runnable
474  * due to waiting on IO to complete).
475  */
476 DEFINE_EVENT_SCHEDSTAT(sched_stat_template, sched_stat_iowait,
477 	     TP_PROTO(struct task_struct *tsk, u64 delay),
478 	     TP_ARGS(tsk, delay));
479 
480 /*
481  * Tracepoint for accounting blocked time (time the task is in uninterruptible).
482  */
483 DEFINE_EVENT_SCHEDSTAT(sched_stat_template, sched_stat_blocked,
484 	     TP_PROTO(struct task_struct *tsk, u64 delay),
485 	     TP_ARGS(tsk, delay));
486 
487 /*
488  * Tracepoint for recording the cause of uninterruptible sleep.
489  */
490 TRACE_EVENT(sched_blocked_reason,
491 
492 	TP_PROTO(struct task_struct *tsk),
493 
494 	TP_ARGS(tsk),
495 
496 	TP_STRUCT__entry(
497 		__field( pid_t,	pid	)
498 		__field( void*, caller	)
499 		__field( bool, io_wait	)
500 	),
501 
502 	TP_fast_assign(
503 		__entry->pid	= tsk->pid;
504 		__entry->caller = (void *)get_wchan(tsk);
505 		__entry->io_wait = tsk->in_iowait;
506 	),
507 
508 	TP_printk("pid=%d iowait=%d caller=%pS", __entry->pid, __entry->io_wait, __entry->caller)
509 );
510 
511 /*
512  * Tracepoint for accounting runtime (time the task is executing
513  * on a CPU).
514  */
515 DECLARE_EVENT_CLASS(sched_stat_runtime,
516 
517 	TP_PROTO(struct task_struct *tsk, u64 runtime, u64 vruntime),
518 
519 	TP_ARGS(tsk, __perf_count(runtime), vruntime),
520 
521 	TP_STRUCT__entry(
522 		__array( char,	comm,	TASK_COMM_LEN	)
523 		__field( pid_t,	pid			)
524 		__field( u64,	runtime			)
525 		__field( u64,	vruntime			)
526 	),
527 
528 	TP_fast_assign(
529 		memcpy(__entry->comm, tsk->comm, TASK_COMM_LEN);
530 		__entry->pid		= tsk->pid;
531 		__entry->runtime	= runtime;
532 		__entry->vruntime	= vruntime;
533 	),
534 
535 	TP_printk("comm=%s pid=%d runtime=%Lu [ns] vruntime=%Lu [ns]",
536 			__entry->comm, __entry->pid,
537 			(unsigned long long)__entry->runtime,
538 			(unsigned long long)__entry->vruntime)
539 );
540 
541 DEFINE_EVENT(sched_stat_runtime, sched_stat_runtime,
542 	     TP_PROTO(struct task_struct *tsk, u64 runtime, u64 vruntime),
543 	     TP_ARGS(tsk, runtime, vruntime));
544 
545 /*
546  * Tracepoint for showing priority inheritance modifying a tasks
547  * priority.
548  */
549 TRACE_EVENT(sched_pi_setprio,
550 
551 	TP_PROTO(struct task_struct *tsk, struct task_struct *pi_task),
552 
553 	TP_ARGS(tsk, pi_task),
554 
555 	TP_STRUCT__entry(
556 		__array( char,	comm,	TASK_COMM_LEN	)
557 		__field( pid_t,	pid			)
558 		__field( int,	oldprio			)
559 		__field( int,	newprio			)
560 	),
561 
562 	TP_fast_assign(
563 		memcpy(__entry->comm, tsk->comm, TASK_COMM_LEN);
564 		__entry->pid		= tsk->pid;
565 		__entry->oldprio	= tsk->prio;
566 		__entry->newprio	= pi_task ?
567 				min(tsk->normal_prio, pi_task->prio) :
568 				tsk->normal_prio;
569 		/* XXX SCHED_DEADLINE bits missing */
570 	),
571 
572 	TP_printk("comm=%s pid=%d oldprio=%d newprio=%d",
573 			__entry->comm, __entry->pid,
574 			__entry->oldprio, __entry->newprio)
575 );
576 
577 #ifdef CONFIG_DETECT_HUNG_TASK
578 TRACE_EVENT(sched_process_hang,
579 	TP_PROTO(struct task_struct *tsk),
580 	TP_ARGS(tsk),
581 
582 	TP_STRUCT__entry(
583 		__array( char,	comm,	TASK_COMM_LEN	)
584 		__field( pid_t,	pid			)
585 	),
586 
587 	TP_fast_assign(
588 		memcpy(__entry->comm, tsk->comm, TASK_COMM_LEN);
589 		__entry->pid = tsk->pid;
590 	),
591 
592 	TP_printk("comm=%s pid=%d", __entry->comm, __entry->pid)
593 );
594 #endif /* CONFIG_DETECT_HUNG_TASK */
595 
596 /*
597  * Tracks migration of tasks from one runqueue to another. Can be used to
598  * detect if automatic NUMA balancing is bouncing between nodes.
599  */
600 TRACE_EVENT(sched_move_numa,
601 
602 	TP_PROTO(struct task_struct *tsk, int src_cpu, int dst_cpu),
603 
604 	TP_ARGS(tsk, src_cpu, dst_cpu),
605 
606 	TP_STRUCT__entry(
607 		__field( pid_t,	pid			)
608 		__field( pid_t,	tgid			)
609 		__field( pid_t,	ngid			)
610 		__field( int,	src_cpu			)
611 		__field( int,	src_nid			)
612 		__field( int,	dst_cpu			)
613 		__field( int,	dst_nid			)
614 	),
615 
616 	TP_fast_assign(
617 		__entry->pid		= task_pid_nr(tsk);
618 		__entry->tgid		= task_tgid_nr(tsk);
619 		__entry->ngid		= task_numa_group_id(tsk);
620 		__entry->src_cpu	= src_cpu;
621 		__entry->src_nid	= cpu_to_node(src_cpu);
622 		__entry->dst_cpu	= dst_cpu;
623 		__entry->dst_nid	= cpu_to_node(dst_cpu);
624 	),
625 
626 	TP_printk("pid=%d tgid=%d ngid=%d src_cpu=%d src_nid=%d dst_cpu=%d dst_nid=%d",
627 			__entry->pid, __entry->tgid, __entry->ngid,
628 			__entry->src_cpu, __entry->src_nid,
629 			__entry->dst_cpu, __entry->dst_nid)
630 );
631 
632 DECLARE_EVENT_CLASS(sched_numa_pair_template,
633 
634 	TP_PROTO(struct task_struct *src_tsk, int src_cpu,
635 		 struct task_struct *dst_tsk, int dst_cpu),
636 
637 	TP_ARGS(src_tsk, src_cpu, dst_tsk, dst_cpu),
638 
639 	TP_STRUCT__entry(
640 		__field( pid_t,	src_pid			)
641 		__field( pid_t,	src_tgid		)
642 		__field( pid_t,	src_ngid		)
643 		__field( int,	src_cpu			)
644 		__field( int,	src_nid			)
645 		__field( pid_t,	dst_pid			)
646 		__field( pid_t,	dst_tgid		)
647 		__field( pid_t,	dst_ngid		)
648 		__field( int,	dst_cpu			)
649 		__field( int,	dst_nid			)
650 	),
651 
652 	TP_fast_assign(
653 		__entry->src_pid	= task_pid_nr(src_tsk);
654 		__entry->src_tgid	= task_tgid_nr(src_tsk);
655 		__entry->src_ngid	= task_numa_group_id(src_tsk);
656 		__entry->src_cpu	= src_cpu;
657 		__entry->src_nid	= cpu_to_node(src_cpu);
658 		__entry->dst_pid	= dst_tsk ? task_pid_nr(dst_tsk) : 0;
659 		__entry->dst_tgid	= dst_tsk ? task_tgid_nr(dst_tsk) : 0;
660 		__entry->dst_ngid	= dst_tsk ? task_numa_group_id(dst_tsk) : 0;
661 		__entry->dst_cpu	= dst_cpu;
662 		__entry->dst_nid	= dst_cpu >= 0 ? cpu_to_node(dst_cpu) : -1;
663 	),
664 
665 	TP_printk("src_pid=%d src_tgid=%d src_ngid=%d src_cpu=%d src_nid=%d dst_pid=%d dst_tgid=%d dst_ngid=%d dst_cpu=%d dst_nid=%d",
666 			__entry->src_pid, __entry->src_tgid, __entry->src_ngid,
667 			__entry->src_cpu, __entry->src_nid,
668 			__entry->dst_pid, __entry->dst_tgid, __entry->dst_ngid,
669 			__entry->dst_cpu, __entry->dst_nid)
670 );
671 
672 DEFINE_EVENT(sched_numa_pair_template, sched_stick_numa,
673 
674 	TP_PROTO(struct task_struct *src_tsk, int src_cpu,
675 		 struct task_struct *dst_tsk, int dst_cpu),
676 
677 	TP_ARGS(src_tsk, src_cpu, dst_tsk, dst_cpu)
678 );
679 
680 DEFINE_EVENT(sched_numa_pair_template, sched_swap_numa,
681 
682 	TP_PROTO(struct task_struct *src_tsk, int src_cpu,
683 		 struct task_struct *dst_tsk, int dst_cpu),
684 
685 	TP_ARGS(src_tsk, src_cpu, dst_tsk, dst_cpu)
686 );
687 
688 
689 /*
690  * Tracepoint for waking a polling cpu without an IPI.
691  */
692 TRACE_EVENT(sched_wake_idle_without_ipi,
693 
694 	TP_PROTO(int cpu),
695 
696 	TP_ARGS(cpu),
697 
698 	TP_STRUCT__entry(
699 		__field(	int,	cpu	)
700 	),
701 
702 	TP_fast_assign(
703 		__entry->cpu	= cpu;
704 	),
705 
706 	TP_printk("cpu=%d", __entry->cpu)
707 );
708 
709 /*
710  * Following tracepoints are not exported in tracefs and provide hooking
711  * mechanisms only for testing and debugging purposes.
712  *
713  * Postfixed with _tp to make them easily identifiable in the code.
714  */
715 DECLARE_TRACE(pelt_cfs_tp,
716 	TP_PROTO(struct cfs_rq *cfs_rq),
717 	TP_ARGS(cfs_rq));
718 
719 DECLARE_TRACE(pelt_rt_tp,
720 	TP_PROTO(struct rq *rq),
721 	TP_ARGS(rq));
722 
723 DECLARE_TRACE(pelt_dl_tp,
724 	TP_PROTO(struct rq *rq),
725 	TP_ARGS(rq));
726 
727 DECLARE_TRACE(pelt_thermal_tp,
728 	TP_PROTO(struct rq *rq),
729 	TP_ARGS(rq));
730 
731 DECLARE_TRACE(pelt_irq_tp,
732 	TP_PROTO(struct rq *rq),
733 	TP_ARGS(rq));
734 
735 DECLARE_TRACE(pelt_se_tp,
736 	TP_PROTO(struct sched_entity *se),
737 	TP_ARGS(se));
738 
739 DECLARE_TRACE(sched_cpu_capacity_tp,
740 	TP_PROTO(struct rq *rq),
741 	TP_ARGS(rq));
742 
743 DECLARE_TRACE(sched_overutilized_tp,
744 	TP_PROTO(struct root_domain *rd, bool overutilized),
745 	TP_ARGS(rd, overutilized));
746 
747 DECLARE_TRACE(sched_util_est_cfs_tp,
748 	TP_PROTO(struct cfs_rq *cfs_rq),
749 	TP_ARGS(cfs_rq));
750 
751 DECLARE_TRACE(sched_util_est_se_tp,
752 	TP_PROTO(struct sched_entity *se),
753 	TP_ARGS(se));
754 
755 DECLARE_TRACE(sched_update_nr_running_tp,
756 	TP_PROTO(struct rq *rq, int change),
757 	TP_ARGS(rq, change));
758 
759 #endif /* _TRACE_SCHED_H */
760 
761 /* This part must be outside protection */
762 #include <trace/define_trace.h>
763