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,unsigned int prev_state,struct task_struct * p)190 static inline long __trace_sched_switch_state(bool preempt,
191 					      unsigned int prev_state,
192 					      struct task_struct *p)
193 {
194 	unsigned int state;
195 
196 	BUG_ON(p != current);
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(prev_state, p->exit_state);
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 		 unsigned int prev_state),
226 
227 	TP_ARGS(preempt, prev, next, prev_state),
228 
229 	TP_STRUCT__entry(
230 		__array(	char,	prev_comm,	TASK_COMM_LEN	)
231 		__field(	pid_t,	prev_pid			)
232 		__field(	int,	prev_prio			)
233 		__field(	long,	prev_state			)
234 		__array(	char,	next_comm,	TASK_COMM_LEN	)
235 		__field(	pid_t,	next_pid			)
236 		__field(	int,	next_prio			)
237 	),
238 
239 	TP_fast_assign(
240 		memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN);
241 		__entry->prev_pid	= prev->pid;
242 		__entry->prev_prio	= prev->prio;
243 		__entry->prev_state	= __trace_sched_switch_state(preempt, prev_state, prev);
244 		memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN);
245 		__entry->next_pid	= next->pid;
246 		__entry->next_prio	= next->prio;
247 		/* XXX SCHED_DEADLINE */
248 	),
249 
250 	TP_printk("prev_comm=%s prev_pid=%d prev_prio=%d prev_state=%s%s ==> next_comm=%s next_pid=%d next_prio=%d",
251 		__entry->prev_comm, __entry->prev_pid, __entry->prev_prio,
252 
253 		(__entry->prev_state & (TASK_REPORT_MAX - 1)) ?
254 		  __print_flags(__entry->prev_state & (TASK_REPORT_MAX - 1), "|",
255 				{ TASK_INTERRUPTIBLE, "S" },
256 				{ TASK_UNINTERRUPTIBLE, "D" },
257 				{ __TASK_STOPPED, "T" },
258 				{ __TASK_TRACED, "t" },
259 				{ EXIT_DEAD, "X" },
260 				{ EXIT_ZOMBIE, "Z" },
261 				{ TASK_PARKED, "P" },
262 				{ TASK_DEAD, "I" }) :
263 		  "R",
264 
265 		__entry->prev_state & TASK_REPORT_MAX ? "+" : "",
266 		__entry->next_comm, __entry->next_pid, __entry->next_prio)
267 );
268 
269 /*
270  * Tracepoint for a task being migrated:
271  */
272 TRACE_EVENT(sched_migrate_task,
273 
274 	TP_PROTO(struct task_struct *p, int dest_cpu),
275 
276 	TP_ARGS(p, dest_cpu),
277 
278 	TP_STRUCT__entry(
279 		__array(	char,	comm,	TASK_COMM_LEN	)
280 		__field(	pid_t,	pid			)
281 		__field(	int,	prio			)
282 		__field(	int,	orig_cpu		)
283 		__field(	int,	dest_cpu		)
284 	),
285 
286 	TP_fast_assign(
287 		memcpy(__entry->comm, p->comm, TASK_COMM_LEN);
288 		__entry->pid		= p->pid;
289 		__entry->prio		= p->prio; /* XXX SCHED_DEADLINE */
290 		__entry->orig_cpu	= task_cpu(p);
291 		__entry->dest_cpu	= dest_cpu;
292 	),
293 
294 	TP_printk("comm=%s pid=%d prio=%d orig_cpu=%d dest_cpu=%d",
295 		  __entry->comm, __entry->pid, __entry->prio,
296 		  __entry->orig_cpu, __entry->dest_cpu)
297 );
298 
299 DECLARE_EVENT_CLASS(sched_process_template,
300 
301 	TP_PROTO(struct task_struct *p),
302 
303 	TP_ARGS(p),
304 
305 	TP_STRUCT__entry(
306 		__array(	char,	comm,	TASK_COMM_LEN	)
307 		__field(	pid_t,	pid			)
308 		__field(	int,	prio			)
309 	),
310 
311 	TP_fast_assign(
312 		memcpy(__entry->comm, p->comm, TASK_COMM_LEN);
313 		__entry->pid		= p->pid;
314 		__entry->prio		= p->prio; /* XXX SCHED_DEADLINE */
315 	),
316 
317 	TP_printk("comm=%s pid=%d prio=%d",
318 		  __entry->comm, __entry->pid, __entry->prio)
319 );
320 
321 /*
322  * Tracepoint for freeing a task:
323  */
324 DEFINE_EVENT(sched_process_template, sched_process_free,
325 	     TP_PROTO(struct task_struct *p),
326 	     TP_ARGS(p));
327 
328 /*
329  * Tracepoint for a task exiting:
330  */
331 DEFINE_EVENT(sched_process_template, sched_process_exit,
332 	     TP_PROTO(struct task_struct *p),
333 	     TP_ARGS(p));
334 
335 /*
336  * Tracepoint for waiting on task to unschedule:
337  */
338 DEFINE_EVENT(sched_process_template, sched_wait_task,
339 	TP_PROTO(struct task_struct *p),
340 	TP_ARGS(p));
341 
342 /*
343  * Tracepoint for a waiting task:
344  */
345 TRACE_EVENT(sched_process_wait,
346 
347 	TP_PROTO(struct pid *pid),
348 
349 	TP_ARGS(pid),
350 
351 	TP_STRUCT__entry(
352 		__array(	char,	comm,	TASK_COMM_LEN	)
353 		__field(	pid_t,	pid			)
354 		__field(	int,	prio			)
355 	),
356 
357 	TP_fast_assign(
358 		memcpy(__entry->comm, current->comm, TASK_COMM_LEN);
359 		__entry->pid		= pid_nr(pid);
360 		__entry->prio		= current->prio; /* XXX SCHED_DEADLINE */
361 	),
362 
363 	TP_printk("comm=%s pid=%d prio=%d",
364 		  __entry->comm, __entry->pid, __entry->prio)
365 );
366 
367 /*
368  * Tracepoint for kernel_clone:
369  */
370 TRACE_EVENT(sched_process_fork,
371 
372 	TP_PROTO(struct task_struct *parent, struct task_struct *child),
373 
374 	TP_ARGS(parent, child),
375 
376 	TP_STRUCT__entry(
377 		__array(	char,	parent_comm,	TASK_COMM_LEN	)
378 		__field(	pid_t,	parent_pid			)
379 		__array(	char,	child_comm,	TASK_COMM_LEN	)
380 		__field(	pid_t,	child_pid			)
381 	),
382 
383 	TP_fast_assign(
384 		memcpy(__entry->parent_comm, parent->comm, TASK_COMM_LEN);
385 		__entry->parent_pid	= parent->pid;
386 		memcpy(__entry->child_comm, child->comm, TASK_COMM_LEN);
387 		__entry->child_pid	= child->pid;
388 	),
389 
390 	TP_printk("comm=%s pid=%d child_comm=%s child_pid=%d",
391 		__entry->parent_comm, __entry->parent_pid,
392 		__entry->child_comm, __entry->child_pid)
393 );
394 
395 /*
396  * Tracepoint for exec:
397  */
398 TRACE_EVENT(sched_process_exec,
399 
400 	TP_PROTO(struct task_struct *p, pid_t old_pid,
401 		 struct linux_binprm *bprm),
402 
403 	TP_ARGS(p, old_pid, bprm),
404 
405 	TP_STRUCT__entry(
406 		__string(	filename,	bprm->filename	)
407 		__field(	pid_t,		pid		)
408 		__field(	pid_t,		old_pid		)
409 	),
410 
411 	TP_fast_assign(
412 		__assign_str(filename);
413 		__entry->pid		= p->pid;
414 		__entry->old_pid	= old_pid;
415 	),
416 
417 	TP_printk("filename=%s pid=%d old_pid=%d", __get_str(filename),
418 		  __entry->pid, __entry->old_pid)
419 );
420 
421 /**
422  * sched_prepare_exec - called before setting up new exec
423  * @task:	pointer to the current task
424  * @bprm:	pointer to linux_binprm used for new exec
425  *
426  * Called before flushing the old exec, where @task is still unchanged, but at
427  * the point of no return during switching to the new exec. At the point it is
428  * called the exec will either succeed, or on failure terminate the task. Also
429  * see the "sched_process_exec" tracepoint, which is called right after @task
430  * has successfully switched to the new exec.
431  */
432 TRACE_EVENT(sched_prepare_exec,
433 
434 	TP_PROTO(struct task_struct *task, struct linux_binprm *bprm),
435 
436 	TP_ARGS(task, bprm),
437 
438 	TP_STRUCT__entry(
439 		__string(	interp,		bprm->interp	)
440 		__string(	filename,	bprm->filename	)
441 		__field(	pid_t,		pid		)
442 		__string(	comm,		task->comm	)
443 	),
444 
445 	TP_fast_assign(
446 		__assign_str(interp);
447 		__assign_str(filename);
448 		__entry->pid = task->pid;
449 		__assign_str(comm);
450 	),
451 
452 	TP_printk("interp=%s filename=%s pid=%d comm=%s",
453 		  __get_str(interp), __get_str(filename),
454 		  __entry->pid, __get_str(comm))
455 );
456 
457 #ifdef CONFIG_SCHEDSTATS
458 #define DEFINE_EVENT_SCHEDSTAT DEFINE_EVENT
459 #define DECLARE_EVENT_CLASS_SCHEDSTAT DECLARE_EVENT_CLASS
460 #else
461 #define DEFINE_EVENT_SCHEDSTAT DEFINE_EVENT_NOP
462 #define DECLARE_EVENT_CLASS_SCHEDSTAT DECLARE_EVENT_CLASS_NOP
463 #endif
464 
465 /*
466  * XXX the below sched_stat tracepoints only apply to SCHED_OTHER/BATCH/IDLE
467  *     adding sched_stat support to SCHED_FIFO/RR would be welcome.
468  */
469 DECLARE_EVENT_CLASS_SCHEDSTAT(sched_stat_template,
470 
471 	TP_PROTO(struct task_struct *tsk, u64 delay),
472 
473 	TP_ARGS(__perf_task(tsk), __perf_count(delay)),
474 
475 	TP_STRUCT__entry(
476 		__array( char,	comm,	TASK_COMM_LEN	)
477 		__field( pid_t,	pid			)
478 		__field( u64,	delay			)
479 	),
480 
481 	TP_fast_assign(
482 		memcpy(__entry->comm, tsk->comm, TASK_COMM_LEN);
483 		__entry->pid	= tsk->pid;
484 		__entry->delay	= delay;
485 	),
486 
487 	TP_printk("comm=%s pid=%d delay=%Lu [ns]",
488 			__entry->comm, __entry->pid,
489 			(unsigned long long)__entry->delay)
490 );
491 
492 /*
493  * Tracepoint for accounting wait time (time the task is runnable
494  * but not actually running due to scheduler contention).
495  */
496 DEFINE_EVENT_SCHEDSTAT(sched_stat_template, sched_stat_wait,
497 	     TP_PROTO(struct task_struct *tsk, u64 delay),
498 	     TP_ARGS(tsk, delay));
499 
500 /*
501  * Tracepoint for accounting sleep time (time the task is not runnable,
502  * including iowait, see below).
503  */
504 DEFINE_EVENT_SCHEDSTAT(sched_stat_template, sched_stat_sleep,
505 	     TP_PROTO(struct task_struct *tsk, u64 delay),
506 	     TP_ARGS(tsk, delay));
507 
508 /*
509  * Tracepoint for accounting iowait time (time the task is not runnable
510  * due to waiting on IO to complete).
511  */
512 DEFINE_EVENT_SCHEDSTAT(sched_stat_template, sched_stat_iowait,
513 	     TP_PROTO(struct task_struct *tsk, u64 delay),
514 	     TP_ARGS(tsk, delay));
515 
516 /*
517  * Tracepoint for accounting blocked time (time the task is in uninterruptible).
518  */
519 DEFINE_EVENT_SCHEDSTAT(sched_stat_template, sched_stat_blocked,
520 	     TP_PROTO(struct task_struct *tsk, u64 delay),
521 	     TP_ARGS(tsk, delay));
522 
523 /*
524  * Tracepoint for accounting runtime (time the task is executing
525  * on a CPU).
526  */
527 DECLARE_EVENT_CLASS(sched_stat_runtime,
528 
529 	TP_PROTO(struct task_struct *tsk, u64 runtime),
530 
531 	TP_ARGS(tsk, __perf_count(runtime)),
532 
533 	TP_STRUCT__entry(
534 		__array( char,	comm,	TASK_COMM_LEN	)
535 		__field( pid_t,	pid			)
536 		__field( u64,	runtime			)
537 	),
538 
539 	TP_fast_assign(
540 		memcpy(__entry->comm, tsk->comm, TASK_COMM_LEN);
541 		__entry->pid		= tsk->pid;
542 		__entry->runtime	= runtime;
543 	),
544 
545 	TP_printk("comm=%s pid=%d runtime=%Lu [ns]",
546 			__entry->comm, __entry->pid,
547 			(unsigned long long)__entry->runtime)
548 );
549 
550 DEFINE_EVENT(sched_stat_runtime, sched_stat_runtime,
551 	     TP_PROTO(struct task_struct *tsk, u64 runtime),
552 	     TP_ARGS(tsk, runtime));
553 
554 /*
555  * Tracepoint for showing priority inheritance modifying a tasks
556  * priority.
557  */
558 TRACE_EVENT(sched_pi_setprio,
559 
560 	TP_PROTO(struct task_struct *tsk, struct task_struct *pi_task),
561 
562 	TP_ARGS(tsk, pi_task),
563 
564 	TP_STRUCT__entry(
565 		__array( char,	comm,	TASK_COMM_LEN	)
566 		__field( pid_t,	pid			)
567 		__field( int,	oldprio			)
568 		__field( int,	newprio			)
569 	),
570 
571 	TP_fast_assign(
572 		memcpy(__entry->comm, tsk->comm, TASK_COMM_LEN);
573 		__entry->pid		= tsk->pid;
574 		__entry->oldprio	= tsk->prio;
575 		__entry->newprio	= pi_task ?
576 				min(tsk->normal_prio, pi_task->prio) :
577 				tsk->normal_prio;
578 		/* XXX SCHED_DEADLINE bits missing */
579 	),
580 
581 	TP_printk("comm=%s pid=%d oldprio=%d newprio=%d",
582 			__entry->comm, __entry->pid,
583 			__entry->oldprio, __entry->newprio)
584 );
585 
586 #ifdef CONFIG_DETECT_HUNG_TASK
587 TRACE_EVENT(sched_process_hang,
588 	TP_PROTO(struct task_struct *tsk),
589 	TP_ARGS(tsk),
590 
591 	TP_STRUCT__entry(
592 		__array( char,	comm,	TASK_COMM_LEN	)
593 		__field( pid_t,	pid			)
594 	),
595 
596 	TP_fast_assign(
597 		memcpy(__entry->comm, tsk->comm, TASK_COMM_LEN);
598 		__entry->pid = tsk->pid;
599 	),
600 
601 	TP_printk("comm=%s pid=%d", __entry->comm, __entry->pid)
602 );
603 #endif /* CONFIG_DETECT_HUNG_TASK */
604 
605 /*
606  * Tracks migration of tasks from one runqueue to another. Can be used to
607  * detect if automatic NUMA balancing is bouncing between nodes.
608  */
609 TRACE_EVENT(sched_move_numa,
610 
611 	TP_PROTO(struct task_struct *tsk, int src_cpu, int dst_cpu),
612 
613 	TP_ARGS(tsk, src_cpu, dst_cpu),
614 
615 	TP_STRUCT__entry(
616 		__field( pid_t,	pid			)
617 		__field( pid_t,	tgid			)
618 		__field( pid_t,	ngid			)
619 		__field( int,	src_cpu			)
620 		__field( int,	src_nid			)
621 		__field( int,	dst_cpu			)
622 		__field( int,	dst_nid			)
623 	),
624 
625 	TP_fast_assign(
626 		__entry->pid		= task_pid_nr(tsk);
627 		__entry->tgid		= task_tgid_nr(tsk);
628 		__entry->ngid		= task_numa_group_id(tsk);
629 		__entry->src_cpu	= src_cpu;
630 		__entry->src_nid	= cpu_to_node(src_cpu);
631 		__entry->dst_cpu	= dst_cpu;
632 		__entry->dst_nid	= cpu_to_node(dst_cpu);
633 	),
634 
635 	TP_printk("pid=%d tgid=%d ngid=%d src_cpu=%d src_nid=%d dst_cpu=%d dst_nid=%d",
636 			__entry->pid, __entry->tgid, __entry->ngid,
637 			__entry->src_cpu, __entry->src_nid,
638 			__entry->dst_cpu, __entry->dst_nid)
639 );
640 
641 DECLARE_EVENT_CLASS(sched_numa_pair_template,
642 
643 	TP_PROTO(struct task_struct *src_tsk, int src_cpu,
644 		 struct task_struct *dst_tsk, int dst_cpu),
645 
646 	TP_ARGS(src_tsk, src_cpu, dst_tsk, dst_cpu),
647 
648 	TP_STRUCT__entry(
649 		__field( pid_t,	src_pid			)
650 		__field( pid_t,	src_tgid		)
651 		__field( pid_t,	src_ngid		)
652 		__field( int,	src_cpu			)
653 		__field( int,	src_nid			)
654 		__field( pid_t,	dst_pid			)
655 		__field( pid_t,	dst_tgid		)
656 		__field( pid_t,	dst_ngid		)
657 		__field( int,	dst_cpu			)
658 		__field( int,	dst_nid			)
659 	),
660 
661 	TP_fast_assign(
662 		__entry->src_pid	= task_pid_nr(src_tsk);
663 		__entry->src_tgid	= task_tgid_nr(src_tsk);
664 		__entry->src_ngid	= task_numa_group_id(src_tsk);
665 		__entry->src_cpu	= src_cpu;
666 		__entry->src_nid	= cpu_to_node(src_cpu);
667 		__entry->dst_pid	= dst_tsk ? task_pid_nr(dst_tsk) : 0;
668 		__entry->dst_tgid	= dst_tsk ? task_tgid_nr(dst_tsk) : 0;
669 		__entry->dst_ngid	= dst_tsk ? task_numa_group_id(dst_tsk) : 0;
670 		__entry->dst_cpu	= dst_cpu;
671 		__entry->dst_nid	= dst_cpu >= 0 ? cpu_to_node(dst_cpu) : -1;
672 	),
673 
674 	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",
675 			__entry->src_pid, __entry->src_tgid, __entry->src_ngid,
676 			__entry->src_cpu, __entry->src_nid,
677 			__entry->dst_pid, __entry->dst_tgid, __entry->dst_ngid,
678 			__entry->dst_cpu, __entry->dst_nid)
679 );
680 
681 DEFINE_EVENT(sched_numa_pair_template, sched_stick_numa,
682 
683 	TP_PROTO(struct task_struct *src_tsk, int src_cpu,
684 		 struct task_struct *dst_tsk, int dst_cpu),
685 
686 	TP_ARGS(src_tsk, src_cpu, dst_tsk, dst_cpu)
687 );
688 
689 DEFINE_EVENT(sched_numa_pair_template, sched_swap_numa,
690 
691 	TP_PROTO(struct task_struct *src_tsk, int src_cpu,
692 		 struct task_struct *dst_tsk, int dst_cpu),
693 
694 	TP_ARGS(src_tsk, src_cpu, dst_tsk, dst_cpu)
695 );
696 
697 #ifdef CONFIG_NUMA_BALANCING
698 #define NUMAB_SKIP_REASON					\
699 	EM( NUMAB_SKIP_UNSUITABLE,		"unsuitable" )	\
700 	EM( NUMAB_SKIP_SHARED_RO,		"shared_ro" )	\
701 	EM( NUMAB_SKIP_INACCESSIBLE,		"inaccessible" )	\
702 	EM( NUMAB_SKIP_SCAN_DELAY,		"scan_delay" )	\
703 	EM( NUMAB_SKIP_PID_INACTIVE,		"pid_inactive" )	\
704 	EM( NUMAB_SKIP_IGNORE_PID,		"ignore_pid_inactive" )		\
705 	EMe(NUMAB_SKIP_SEQ_COMPLETED,		"seq_completed" )
706 
707 /* Redefine for export. */
708 #undef EM
709 #undef EMe
710 #define EM(a, b)	TRACE_DEFINE_ENUM(a);
711 #define EMe(a, b)	TRACE_DEFINE_ENUM(a);
712 
713 NUMAB_SKIP_REASON
714 
715 /* Redefine for symbolic printing. */
716 #undef EM
717 #undef EMe
718 #define EM(a, b)	{ a, b },
719 #define EMe(a, b)	{ a, b }
720 
721 TRACE_EVENT(sched_skip_vma_numa,
722 
723 	TP_PROTO(struct mm_struct *mm, struct vm_area_struct *vma,
724 		 enum numa_vmaskip_reason reason),
725 
726 	TP_ARGS(mm, vma, reason),
727 
728 	TP_STRUCT__entry(
729 		__field(unsigned long, numa_scan_offset)
730 		__field(unsigned long, vm_start)
731 		__field(unsigned long, vm_end)
732 		__field(enum numa_vmaskip_reason, reason)
733 	),
734 
735 	TP_fast_assign(
736 		__entry->numa_scan_offset	= mm->numa_scan_offset;
737 		__entry->vm_start		= vma->vm_start;
738 		__entry->vm_end			= vma->vm_end;
739 		__entry->reason			= reason;
740 	),
741 
742 	TP_printk("numa_scan_offset=%lX vm_start=%lX vm_end=%lX reason=%s",
743 		  __entry->numa_scan_offset,
744 		  __entry->vm_start,
745 		  __entry->vm_end,
746 		  __print_symbolic(__entry->reason, NUMAB_SKIP_REASON))
747 );
748 #endif /* CONFIG_NUMA_BALANCING */
749 
750 /*
751  * Tracepoint for waking a polling cpu without an IPI.
752  */
753 TRACE_EVENT(sched_wake_idle_without_ipi,
754 
755 	TP_PROTO(int cpu),
756 
757 	TP_ARGS(cpu),
758 
759 	TP_STRUCT__entry(
760 		__field(	int,	cpu	)
761 	),
762 
763 	TP_fast_assign(
764 		__entry->cpu	= cpu;
765 	),
766 
767 	TP_printk("cpu=%d", __entry->cpu)
768 );
769 
770 /*
771  * Following tracepoints are not exported in tracefs and provide hooking
772  * mechanisms only for testing and debugging purposes.
773  *
774  * Postfixed with _tp to make them easily identifiable in the code.
775  */
776 DECLARE_TRACE(pelt_cfs_tp,
777 	TP_PROTO(struct cfs_rq *cfs_rq),
778 	TP_ARGS(cfs_rq));
779 
780 DECLARE_TRACE(pelt_rt_tp,
781 	TP_PROTO(struct rq *rq),
782 	TP_ARGS(rq));
783 
784 DECLARE_TRACE(pelt_dl_tp,
785 	TP_PROTO(struct rq *rq),
786 	TP_ARGS(rq));
787 
788 DECLARE_TRACE(pelt_hw_tp,
789 	TP_PROTO(struct rq *rq),
790 	TP_ARGS(rq));
791 
792 DECLARE_TRACE(pelt_irq_tp,
793 	TP_PROTO(struct rq *rq),
794 	TP_ARGS(rq));
795 
796 DECLARE_TRACE(pelt_se_tp,
797 	TP_PROTO(struct sched_entity *se),
798 	TP_ARGS(se));
799 
800 DECLARE_TRACE(sched_cpu_capacity_tp,
801 	TP_PROTO(struct rq *rq),
802 	TP_ARGS(rq));
803 
804 DECLARE_TRACE(sched_overutilized_tp,
805 	TP_PROTO(struct root_domain *rd, bool overutilized),
806 	TP_ARGS(rd, overutilized));
807 
808 DECLARE_TRACE(sched_util_est_cfs_tp,
809 	TP_PROTO(struct cfs_rq *cfs_rq),
810 	TP_ARGS(cfs_rq));
811 
812 DECLARE_TRACE(sched_util_est_se_tp,
813 	TP_PROTO(struct sched_entity *se),
814 	TP_ARGS(se));
815 
816 DECLARE_TRACE(sched_update_nr_running_tp,
817 	TP_PROTO(struct rq *rq, int change),
818 	TP_ARGS(rq, change));
819 
820 DECLARE_TRACE(sched_compute_energy_tp,
821 	TP_PROTO(struct task_struct *p, int dst_cpu, unsigned long energy,
822 		 unsigned long max_util, unsigned long busy_time),
823 	TP_ARGS(p, dst_cpu, energy, max_util, busy_time));
824 
825 DECLARE_TRACE(sched_entry_tp,
826 	TP_PROTO(bool preempt, unsigned long ip),
827 	TP_ARGS(preempt, ip));
828 
829 DECLARE_TRACE(sched_exit_tp,
830 	TP_PROTO(bool is_switch, unsigned long ip),
831 	TP_ARGS(is_switch, ip));
832 
833 DECLARE_TRACE_CONDITION(sched_set_state_tp,
834 	TP_PROTO(struct task_struct *tsk, int state),
835 	TP_ARGS(tsk, state),
836 	TP_CONDITION(!!(tsk->__state) != !!state));
837 
838 #endif /* _TRACE_SCHED_H */
839 
840 /* This part must be outside protection */
841 #include <trace/define_trace.h>
842