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 */
 16TRACE_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 */
 38TRACE_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 */
 64TRACE_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 */
 93TRACE_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 */
119TRACE_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 */
141DECLARE_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 */
170DEFINE_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 */
178DEFINE_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 */
185DEFINE_EVENT(sched_wakeup_template, sched_wakeup_new,
186	     TP_PROTO(struct task_struct *p),
187	     TP_ARGS(p));
188
189#ifdef CREATE_TRACE_POINTS
190static 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#ifdef CONFIG_SCHED_DEBUG
197	BUG_ON(p != current);
198#endif /* CONFIG_SCHED_DEBUG */
199
200	/*
201	 * Preemption ignores task state, therefore preempted tasks are always
202	 * RUNNING (we will not have dequeued if state != RUNNING).
203	 */
204	if (preempt)
205		return TASK_REPORT_MAX;
206
 
207	/*
208	 * task_state_index() uses fls() and returns a value from 0-8 range.
209	 * Decrement it by 1 (except TASK_RUNNING state i.e 0) before using
210	 * it for left shift operation to get the correct task->state
211	 * mapping.
212	 */
213	state = __task_state_index(prev_state, p->exit_state);
 
 
214
215	return state ? (1 << (state - 1)) : state;
216}
217#endif /* CREATE_TRACE_POINTS */
218
219/*
220 * Tracepoint for task switches, performed by the scheduler:
221 */
222TRACE_EVENT(sched_switch,
223
224	TP_PROTO(bool preempt,
225		 struct task_struct *prev,
226		 struct task_struct *next,
227		 unsigned int prev_state),
228
229	TP_ARGS(preempt, prev, next, prev_state),
230
231	TP_STRUCT__entry(
232		__array(	char,	prev_comm,	TASK_COMM_LEN	)
233		__field(	pid_t,	prev_pid			)
234		__field(	int,	prev_prio			)
235		__field(	long,	prev_state			)
236		__array(	char,	next_comm,	TASK_COMM_LEN	)
237		__field(	pid_t,	next_pid			)
238		__field(	int,	next_prio			)
239	),
240
241	TP_fast_assign(
242		memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN);
243		__entry->prev_pid	= prev->pid;
244		__entry->prev_prio	= prev->prio;
245		__entry->prev_state	= __trace_sched_switch_state(preempt, prev_state, prev);
246		memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN);
247		__entry->next_pid	= next->pid;
248		__entry->next_prio	= next->prio;
249		/* XXX SCHED_DEADLINE */
250	),
251
252	TP_printk("prev_comm=%s prev_pid=%d prev_prio=%d prev_state=%s%s ==> next_comm=%s next_pid=%d next_prio=%d",
253		__entry->prev_comm, __entry->prev_pid, __entry->prev_prio,
254
255		(__entry->prev_state & (TASK_REPORT_MAX - 1)) ?
256		  __print_flags(__entry->prev_state & (TASK_REPORT_MAX - 1), "|",
257				{ TASK_INTERRUPTIBLE, "S" },
258				{ TASK_UNINTERRUPTIBLE, "D" },
259				{ __TASK_STOPPED, "T" },
260				{ __TASK_TRACED, "t" },
261				{ EXIT_DEAD, "X" },
262				{ EXIT_ZOMBIE, "Z" },
263				{ TASK_PARKED, "P" },
264				{ TASK_DEAD, "I" }) :
265		  "R",
266
267		__entry->prev_state & TASK_REPORT_MAX ? "+" : "",
268		__entry->next_comm, __entry->next_pid, __entry->next_prio)
269);
270
271/*
272 * Tracepoint for a task being migrated:
273 */
274TRACE_EVENT(sched_migrate_task,
275
276	TP_PROTO(struct task_struct *p, int dest_cpu),
277
278	TP_ARGS(p, dest_cpu),
279
280	TP_STRUCT__entry(
281		__array(	char,	comm,	TASK_COMM_LEN	)
282		__field(	pid_t,	pid			)
283		__field(	int,	prio			)
284		__field(	int,	orig_cpu		)
285		__field(	int,	dest_cpu		)
286	),
287
288	TP_fast_assign(
289		memcpy(__entry->comm, p->comm, TASK_COMM_LEN);
290		__entry->pid		= p->pid;
291		__entry->prio		= p->prio; /* XXX SCHED_DEADLINE */
292		__entry->orig_cpu	= task_cpu(p);
293		__entry->dest_cpu	= dest_cpu;
294	),
295
296	TP_printk("comm=%s pid=%d prio=%d orig_cpu=%d dest_cpu=%d",
297		  __entry->comm, __entry->pid, __entry->prio,
298		  __entry->orig_cpu, __entry->dest_cpu)
299);
300
301DECLARE_EVENT_CLASS(sched_process_template,
302
303	TP_PROTO(struct task_struct *p),
304
305	TP_ARGS(p),
306
307	TP_STRUCT__entry(
308		__array(	char,	comm,	TASK_COMM_LEN	)
309		__field(	pid_t,	pid			)
310		__field(	int,	prio			)
311	),
312
313	TP_fast_assign(
314		memcpy(__entry->comm, p->comm, TASK_COMM_LEN);
315		__entry->pid		= p->pid;
316		__entry->prio		= p->prio; /* XXX SCHED_DEADLINE */
317	),
318
319	TP_printk("comm=%s pid=%d prio=%d",
320		  __entry->comm, __entry->pid, __entry->prio)
321);
322
323/*
324 * Tracepoint for freeing a task:
325 */
326DEFINE_EVENT(sched_process_template, sched_process_free,
327	     TP_PROTO(struct task_struct *p),
328	     TP_ARGS(p));
 
329
330/*
331 * Tracepoint for a task exiting:
332 */
333DEFINE_EVENT(sched_process_template, sched_process_exit,
334	     TP_PROTO(struct task_struct *p),
335	     TP_ARGS(p));
336
337/*
338 * Tracepoint for waiting on task to unschedule:
339 */
340DEFINE_EVENT(sched_process_template, sched_wait_task,
341	TP_PROTO(struct task_struct *p),
342	TP_ARGS(p));
343
344/*
345 * Tracepoint for a waiting task:
346 */
347TRACE_EVENT(sched_process_wait,
348
349	TP_PROTO(struct pid *pid),
350
351	TP_ARGS(pid),
352
353	TP_STRUCT__entry(
354		__array(	char,	comm,	TASK_COMM_LEN	)
355		__field(	pid_t,	pid			)
356		__field(	int,	prio			)
357	),
358
359	TP_fast_assign(
360		memcpy(__entry->comm, current->comm, TASK_COMM_LEN);
361		__entry->pid		= pid_nr(pid);
362		__entry->prio		= current->prio; /* XXX SCHED_DEADLINE */
363	),
364
365	TP_printk("comm=%s pid=%d prio=%d",
366		  __entry->comm, __entry->pid, __entry->prio)
367);
368
369/*
370 * Tracepoint for kernel_clone:
371 */
372TRACE_EVENT(sched_process_fork,
373
374	TP_PROTO(struct task_struct *parent, struct task_struct *child),
375
376	TP_ARGS(parent, child),
377
378	TP_STRUCT__entry(
379		__array(	char,	parent_comm,	TASK_COMM_LEN	)
380		__field(	pid_t,	parent_pid			)
381		__array(	char,	child_comm,	TASK_COMM_LEN	)
382		__field(	pid_t,	child_pid			)
383	),
384
385	TP_fast_assign(
386		memcpy(__entry->parent_comm, parent->comm, TASK_COMM_LEN);
387		__entry->parent_pid	= parent->pid;
388		memcpy(__entry->child_comm, child->comm, TASK_COMM_LEN);
389		__entry->child_pid	= child->pid;
390	),
391
392	TP_printk("comm=%s pid=%d child_comm=%s child_pid=%d",
393		__entry->parent_comm, __entry->parent_pid,
394		__entry->child_comm, __entry->child_pid)
395);
396
397/*
398 * Tracepoint for exec:
399 */
400TRACE_EVENT(sched_process_exec,
401
402	TP_PROTO(struct task_struct *p, pid_t old_pid,
403		 struct linux_binprm *bprm),
404
405	TP_ARGS(p, old_pid, bprm),
406
407	TP_STRUCT__entry(
408		__string(	filename,	bprm->filename	)
409		__field(	pid_t,		pid		)
410		__field(	pid_t,		old_pid		)
411	),
412
413	TP_fast_assign(
414		__assign_str(filename);
415		__entry->pid		= p->pid;
416		__entry->old_pid	= old_pid;
417	),
418
419	TP_printk("filename=%s pid=%d old_pid=%d", __get_str(filename),
420		  __entry->pid, __entry->old_pid)
421);
422
423/**
424 * sched_prepare_exec - called before setting up new exec
425 * @task:	pointer to the current task
426 * @bprm:	pointer to linux_binprm used for new exec
427 *
428 * Called before flushing the old exec, where @task is still unchanged, but at
429 * the point of no return during switching to the new exec. At the point it is
430 * called the exec will either succeed, or on failure terminate the task. Also
431 * see the "sched_process_exec" tracepoint, which is called right after @task
432 * has successfully switched to the new exec.
433 */
434TRACE_EVENT(sched_prepare_exec,
435
436	TP_PROTO(struct task_struct *task, struct linux_binprm *bprm),
437
438	TP_ARGS(task, bprm),
439
440	TP_STRUCT__entry(
441		__string(	interp,		bprm->interp	)
442		__string(	filename,	bprm->filename	)
443		__field(	pid_t,		pid		)
444		__string(	comm,		task->comm	)
445	),
446
447	TP_fast_assign(
448		__assign_str(interp);
449		__assign_str(filename);
450		__entry->pid = task->pid;
451		__assign_str(comm);
452	),
453
454	TP_printk("interp=%s filename=%s pid=%d comm=%s",
455		  __get_str(interp), __get_str(filename),
456		  __entry->pid, __get_str(comm))
457);
458
459#ifdef CONFIG_SCHEDSTATS
460#define DEFINE_EVENT_SCHEDSTAT DEFINE_EVENT
461#define DECLARE_EVENT_CLASS_SCHEDSTAT DECLARE_EVENT_CLASS
462#else
463#define DEFINE_EVENT_SCHEDSTAT DEFINE_EVENT_NOP
464#define DECLARE_EVENT_CLASS_SCHEDSTAT DECLARE_EVENT_CLASS_NOP
465#endif
466
467/*
468 * XXX the below sched_stat tracepoints only apply to SCHED_OTHER/BATCH/IDLE
469 *     adding sched_stat support to SCHED_FIFO/RR would be welcome.
470 */
471DECLARE_EVENT_CLASS_SCHEDSTAT(sched_stat_template,
472
473	TP_PROTO(struct task_struct *tsk, u64 delay),
474
475	TP_ARGS(__perf_task(tsk), __perf_count(delay)),
476
477	TP_STRUCT__entry(
478		__array( char,	comm,	TASK_COMM_LEN	)
479		__field( pid_t,	pid			)
480		__field( u64,	delay			)
481	),
482
483	TP_fast_assign(
484		memcpy(__entry->comm, tsk->comm, TASK_COMM_LEN);
485		__entry->pid	= tsk->pid;
486		__entry->delay	= delay;
 
 
 
487	),
488
489	TP_printk("comm=%s pid=%d delay=%Lu [ns]",
490			__entry->comm, __entry->pid,
491			(unsigned long long)__entry->delay)
492);
493
 
494/*
495 * Tracepoint for accounting wait time (time the task is runnable
496 * but not actually running due to scheduler contention).
497 */
498DEFINE_EVENT_SCHEDSTAT(sched_stat_template, sched_stat_wait,
499	     TP_PROTO(struct task_struct *tsk, u64 delay),
500	     TP_ARGS(tsk, delay));
501
502/*
503 * Tracepoint for accounting sleep time (time the task is not runnable,
504 * including iowait, see below).
505 */
506DEFINE_EVENT_SCHEDSTAT(sched_stat_template, sched_stat_sleep,
507	     TP_PROTO(struct task_struct *tsk, u64 delay),
508	     TP_ARGS(tsk, delay));
509
510/*
511 * Tracepoint for accounting iowait time (time the task is not runnable
512 * due to waiting on IO to complete).
513 */
514DEFINE_EVENT_SCHEDSTAT(sched_stat_template, sched_stat_iowait,
515	     TP_PROTO(struct task_struct *tsk, u64 delay),
516	     TP_ARGS(tsk, delay));
517
518/*
519 * Tracepoint for accounting blocked time (time the task is in uninterruptible).
520 */
521DEFINE_EVENT_SCHEDSTAT(sched_stat_template, sched_stat_blocked,
522	     TP_PROTO(struct task_struct *tsk, u64 delay),
523	     TP_ARGS(tsk, delay));
524
525/*
526 * Tracepoint for accounting runtime (time the task is executing
527 * on a CPU).
528 */
529DECLARE_EVENT_CLASS(sched_stat_runtime,
530
531	TP_PROTO(struct task_struct *tsk, u64 runtime),
532
533	TP_ARGS(tsk, __perf_count(runtime)),
534
535	TP_STRUCT__entry(
536		__array( char,	comm,	TASK_COMM_LEN	)
537		__field( pid_t,	pid			)
538		__field( u64,	runtime			)
 
539	),
540
541	TP_fast_assign(
542		memcpy(__entry->comm, tsk->comm, TASK_COMM_LEN);
543		__entry->pid		= tsk->pid;
544		__entry->runtime	= runtime;
 
 
 
 
545	),
546
547	TP_printk("comm=%s pid=%d runtime=%Lu [ns]",
548			__entry->comm, __entry->pid,
549			(unsigned long long)__entry->runtime)
 
550);
551
552DEFINE_EVENT(sched_stat_runtime, sched_stat_runtime,
553	     TP_PROTO(struct task_struct *tsk, u64 runtime),
554	     TP_ARGS(tsk, runtime));
555
556/*
557 * Tracepoint for showing priority inheritance modifying a tasks
558 * priority.
559 */
560TRACE_EVENT(sched_pi_setprio,
561
562	TP_PROTO(struct task_struct *tsk, struct task_struct *pi_task),
563
564	TP_ARGS(tsk, pi_task),
565
566	TP_STRUCT__entry(
567		__array( char,	comm,	TASK_COMM_LEN	)
568		__field( pid_t,	pid			)
569		__field( int,	oldprio			)
570		__field( int,	newprio			)
571	),
572
573	TP_fast_assign(
574		memcpy(__entry->comm, tsk->comm, TASK_COMM_LEN);
575		__entry->pid		= tsk->pid;
576		__entry->oldprio	= tsk->prio;
577		__entry->newprio	= pi_task ?
578				min(tsk->normal_prio, pi_task->prio) :
579				tsk->normal_prio;
580		/* XXX SCHED_DEADLINE bits missing */
581	),
582
583	TP_printk("comm=%s pid=%d oldprio=%d newprio=%d",
584			__entry->comm, __entry->pid,
585			__entry->oldprio, __entry->newprio)
586);
587
588#ifdef CONFIG_DETECT_HUNG_TASK
589TRACE_EVENT(sched_process_hang,
590	TP_PROTO(struct task_struct *tsk),
591	TP_ARGS(tsk),
592
593	TP_STRUCT__entry(
594		__array( char,	comm,	TASK_COMM_LEN	)
595		__field( pid_t,	pid			)
596	),
597
598	TP_fast_assign(
599		memcpy(__entry->comm, tsk->comm, TASK_COMM_LEN);
600		__entry->pid = tsk->pid;
601	),
602
603	TP_printk("comm=%s pid=%d", __entry->comm, __entry->pid)
604);
605#endif /* CONFIG_DETECT_HUNG_TASK */
606
607/*
608 * Tracks migration of tasks from one runqueue to another. Can be used to
609 * detect if automatic NUMA balancing is bouncing between nodes.
610 */
611TRACE_EVENT(sched_move_numa,
612
613	TP_PROTO(struct task_struct *tsk, int src_cpu, int dst_cpu),
614
615	TP_ARGS(tsk, src_cpu, dst_cpu),
616
617	TP_STRUCT__entry(
618		__field( pid_t,	pid			)
619		__field( pid_t,	tgid			)
620		__field( pid_t,	ngid			)
621		__field( int,	src_cpu			)
622		__field( int,	src_nid			)
623		__field( int,	dst_cpu			)
624		__field( int,	dst_nid			)
625	),
626
627	TP_fast_assign(
628		__entry->pid		= task_pid_nr(tsk);
629		__entry->tgid		= task_tgid_nr(tsk);
630		__entry->ngid		= task_numa_group_id(tsk);
631		__entry->src_cpu	= src_cpu;
632		__entry->src_nid	= cpu_to_node(src_cpu);
633		__entry->dst_cpu	= dst_cpu;
634		__entry->dst_nid	= cpu_to_node(dst_cpu);
635	),
636
637	TP_printk("pid=%d tgid=%d ngid=%d src_cpu=%d src_nid=%d dst_cpu=%d dst_nid=%d",
638			__entry->pid, __entry->tgid, __entry->ngid,
639			__entry->src_cpu, __entry->src_nid,
640			__entry->dst_cpu, __entry->dst_nid)
641);
642
643DECLARE_EVENT_CLASS(sched_numa_pair_template,
644
645	TP_PROTO(struct task_struct *src_tsk, int src_cpu,
646		 struct task_struct *dst_tsk, int dst_cpu),
647
648	TP_ARGS(src_tsk, src_cpu, dst_tsk, dst_cpu),
649
650	TP_STRUCT__entry(
651		__field( pid_t,	src_pid			)
652		__field( pid_t,	src_tgid		)
653		__field( pid_t,	src_ngid		)
654		__field( int,	src_cpu			)
655		__field( int,	src_nid			)
656		__field( pid_t,	dst_pid			)
657		__field( pid_t,	dst_tgid		)
658		__field( pid_t,	dst_ngid		)
659		__field( int,	dst_cpu			)
660		__field( int,	dst_nid			)
661	),
662
663	TP_fast_assign(
664		__entry->src_pid	= task_pid_nr(src_tsk);
665		__entry->src_tgid	= task_tgid_nr(src_tsk);
666		__entry->src_ngid	= task_numa_group_id(src_tsk);
667		__entry->src_cpu	= src_cpu;
668		__entry->src_nid	= cpu_to_node(src_cpu);
669		__entry->dst_pid	= dst_tsk ? task_pid_nr(dst_tsk) : 0;
670		__entry->dst_tgid	= dst_tsk ? task_tgid_nr(dst_tsk) : 0;
671		__entry->dst_ngid	= dst_tsk ? task_numa_group_id(dst_tsk) : 0;
672		__entry->dst_cpu	= dst_cpu;
673		__entry->dst_nid	= dst_cpu >= 0 ? cpu_to_node(dst_cpu) : -1;
674	),
675
676	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",
677			__entry->src_pid, __entry->src_tgid, __entry->src_ngid,
678			__entry->src_cpu, __entry->src_nid,
679			__entry->dst_pid, __entry->dst_tgid, __entry->dst_ngid,
680			__entry->dst_cpu, __entry->dst_nid)
681);
682
683DEFINE_EVENT(sched_numa_pair_template, sched_stick_numa,
684
685	TP_PROTO(struct task_struct *src_tsk, int src_cpu,
686		 struct task_struct *dst_tsk, int dst_cpu),
687
688	TP_ARGS(src_tsk, src_cpu, dst_tsk, dst_cpu)
689);
690
691DEFINE_EVENT(sched_numa_pair_template, sched_swap_numa,
692
693	TP_PROTO(struct task_struct *src_tsk, int src_cpu,
694		 struct task_struct *dst_tsk, int dst_cpu),
695
696	TP_ARGS(src_tsk, src_cpu, dst_tsk, dst_cpu)
697);
698
699#ifdef CONFIG_NUMA_BALANCING
700#define NUMAB_SKIP_REASON					\
701	EM( NUMAB_SKIP_UNSUITABLE,		"unsuitable" )	\
702	EM( NUMAB_SKIP_SHARED_RO,		"shared_ro" )	\
703	EM( NUMAB_SKIP_INACCESSIBLE,		"inaccessible" )	\
704	EM( NUMAB_SKIP_SCAN_DELAY,		"scan_delay" )	\
705	EM( NUMAB_SKIP_PID_INACTIVE,		"pid_inactive" )	\
706	EM( NUMAB_SKIP_IGNORE_PID,		"ignore_pid_inactive" )		\
707	EMe(NUMAB_SKIP_SEQ_COMPLETED,		"seq_completed" )
708
709/* Redefine for export. */
710#undef EM
711#undef EMe
712#define EM(a, b)	TRACE_DEFINE_ENUM(a);
713#define EMe(a, b)	TRACE_DEFINE_ENUM(a);
714
715NUMAB_SKIP_REASON
716
717/* Redefine for symbolic printing. */
718#undef EM
719#undef EMe
720#define EM(a, b)	{ a, b },
721#define EMe(a, b)	{ a, b }
722
723TRACE_EVENT(sched_skip_vma_numa,
724
725	TP_PROTO(struct mm_struct *mm, struct vm_area_struct *vma,
726		 enum numa_vmaskip_reason reason),
727
728	TP_ARGS(mm, vma, reason),
729
730	TP_STRUCT__entry(
731		__field(unsigned long, numa_scan_offset)
732		__field(unsigned long, vm_start)
733		__field(unsigned long, vm_end)
734		__field(enum numa_vmaskip_reason, reason)
735	),
736
737	TP_fast_assign(
738		__entry->numa_scan_offset	= mm->numa_scan_offset;
739		__entry->vm_start		= vma->vm_start;
740		__entry->vm_end			= vma->vm_end;
741		__entry->reason			= reason;
742	),
743
744	TP_printk("numa_scan_offset=%lX vm_start=%lX vm_end=%lX reason=%s",
745		  __entry->numa_scan_offset,
746		  __entry->vm_start,
747		  __entry->vm_end,
748		  __print_symbolic(__entry->reason, NUMAB_SKIP_REASON))
749);
750#endif /* CONFIG_NUMA_BALANCING */
751
752/*
753 * Tracepoint for waking a polling cpu without an IPI.
754 */
755TRACE_EVENT(sched_wake_idle_without_ipi,
756
757	TP_PROTO(int cpu),
758
759	TP_ARGS(cpu),
760
761	TP_STRUCT__entry(
762		__field(	int,	cpu	)
763	),
764
765	TP_fast_assign(
766		__entry->cpu	= cpu;
767	),
768
769	TP_printk("cpu=%d", __entry->cpu)
770);
771
772/*
773 * Following tracepoints are not exported in tracefs and provide hooking
774 * mechanisms only for testing and debugging purposes.
775 *
776 * Postfixed with _tp to make them easily identifiable in the code.
777 */
778DECLARE_TRACE(pelt_cfs_tp,
779	TP_PROTO(struct cfs_rq *cfs_rq),
780	TP_ARGS(cfs_rq));
781
782DECLARE_TRACE(pelt_rt_tp,
783	TP_PROTO(struct rq *rq),
784	TP_ARGS(rq));
785
786DECLARE_TRACE(pelt_dl_tp,
787	TP_PROTO(struct rq *rq),
788	TP_ARGS(rq));
789
790DECLARE_TRACE(pelt_hw_tp,
791	TP_PROTO(struct rq *rq),
792	TP_ARGS(rq));
793
794DECLARE_TRACE(pelt_irq_tp,
795	TP_PROTO(struct rq *rq),
796	TP_ARGS(rq));
797
798DECLARE_TRACE(pelt_se_tp,
799	TP_PROTO(struct sched_entity *se),
800	TP_ARGS(se));
801
802DECLARE_TRACE(sched_cpu_capacity_tp,
803	TP_PROTO(struct rq *rq),
804	TP_ARGS(rq));
805
806DECLARE_TRACE(sched_overutilized_tp,
807	TP_PROTO(struct root_domain *rd, bool overutilized),
808	TP_ARGS(rd, overutilized));
809
810DECLARE_TRACE(sched_util_est_cfs_tp,
811	TP_PROTO(struct cfs_rq *cfs_rq),
812	TP_ARGS(cfs_rq));
813
814DECLARE_TRACE(sched_util_est_se_tp,
815	TP_PROTO(struct sched_entity *se),
816	TP_ARGS(se));
817
818DECLARE_TRACE(sched_update_nr_running_tp,
819	TP_PROTO(struct rq *rq, int change),
820	TP_ARGS(rq, change));
821
822DECLARE_TRACE(sched_compute_energy_tp,
823	TP_PROTO(struct task_struct *p, int dst_cpu, unsigned long energy,
824		 unsigned long max_util, unsigned long busy_time),
825	TP_ARGS(p, dst_cpu, energy, max_util, busy_time));
826
827#endif /* _TRACE_SCHED_H */
828
829/* This part must be outside protection */
830#include <trace/define_trace.h>