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 #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 */
222 TRACE_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 */
274 TRACE_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
301 DECLARE_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 */
326 DEFINE_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 */
333 DEFINE_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 */
340 DEFINE_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 */
347 TRACE_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 */
372 TRACE_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 */
400 TRACE_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 */
434 TRACE_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 */
471 DECLARE_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 */
498 DEFINE_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 */
506 DEFINE_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 */
514 DEFINE_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 */
521 DEFINE_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 recording the cause of uninterruptible sleep.
527 */
528 TRACE_EVENT(sched_blocked_reason,
529
530 TP_PROTO(struct task_struct *tsk, void *blocked_func),
531
532 TP_ARGS(tsk, blocked_func),
533
534 TP_STRUCT__entry(
535 __field( pid_t, pid )
536 __field( void*, caller )
537 __field( bool, io_wait )
538 ),
539
540 TP_fast_assign(
541 __entry->pid = tsk->pid;
542 __entry->caller = blocked_func;
543 __entry->io_wait = tsk->in_iowait;
544 ),
545
546 TP_printk("pid=%d iowait=%d caller=%pS", __entry->pid, __entry->io_wait, __entry->caller)
547 );
548
549 /*
550 * Tracepoint for accounting runtime (time the task is executing
551 * on a CPU).
552 */
553 DECLARE_EVENT_CLASS(sched_stat_runtime,
554
555 TP_PROTO(struct task_struct *tsk, u64 runtime),
556
557 TP_ARGS(tsk, __perf_count(runtime)),
558
559 TP_STRUCT__entry(
560 __array( char, comm, TASK_COMM_LEN )
561 __field( pid_t, pid )
562 __field( u64, runtime )
563 ),
564
565 TP_fast_assign(
566 memcpy(__entry->comm, tsk->comm, TASK_COMM_LEN);
567 __entry->pid = tsk->pid;
568 __entry->runtime = runtime;
569 ),
570
571 TP_printk("comm=%s pid=%d runtime=%Lu [ns]",
572 __entry->comm, __entry->pid,
573 (unsigned long long)__entry->runtime)
574 );
575
576 DEFINE_EVENT(sched_stat_runtime, sched_stat_runtime,
577 TP_PROTO(struct task_struct *tsk, u64 runtime),
578 TP_ARGS(tsk, runtime));
579
580 /*
581 * Tracepoint for showing priority inheritance modifying a tasks
582 * priority.
583 */
584 TRACE_EVENT(sched_pi_setprio,
585
586 TP_PROTO(struct task_struct *tsk, struct task_struct *pi_task),
587
588 TP_ARGS(tsk, pi_task),
589
590 TP_STRUCT__entry(
591 __array( char, comm, TASK_COMM_LEN )
592 __field( pid_t, pid )
593 __field( int, oldprio )
594 __field( int, newprio )
595 ),
596
597 TP_fast_assign(
598 memcpy(__entry->comm, tsk->comm, TASK_COMM_LEN);
599 __entry->pid = tsk->pid;
600 __entry->oldprio = tsk->prio;
601 __entry->newprio = pi_task ?
602 min(tsk->normal_prio, pi_task->prio) :
603 tsk->normal_prio;
604 /* XXX SCHED_DEADLINE bits missing */
605 ),
606
607 TP_printk("comm=%s pid=%d oldprio=%d newprio=%d",
608 __entry->comm, __entry->pid,
609 __entry->oldprio, __entry->newprio)
610 );
611
612 #ifdef CONFIG_DETECT_HUNG_TASK
613 TRACE_EVENT(sched_process_hang,
614 TP_PROTO(struct task_struct *tsk),
615 TP_ARGS(tsk),
616
617 TP_STRUCT__entry(
618 __array( char, comm, TASK_COMM_LEN )
619 __field( pid_t, pid )
620 ),
621
622 TP_fast_assign(
623 memcpy(__entry->comm, tsk->comm, TASK_COMM_LEN);
624 __entry->pid = tsk->pid;
625 ),
626
627 TP_printk("comm=%s pid=%d", __entry->comm, __entry->pid)
628 );
629 #endif /* CONFIG_DETECT_HUNG_TASK */
630
631 /*
632 * Tracks migration of tasks from one runqueue to another. Can be used to
633 * detect if automatic NUMA balancing is bouncing between nodes.
634 */
635 TRACE_EVENT(sched_move_numa,
636
637 TP_PROTO(struct task_struct *tsk, int src_cpu, int dst_cpu),
638
639 TP_ARGS(tsk, src_cpu, dst_cpu),
640
641 TP_STRUCT__entry(
642 __field( pid_t, pid )
643 __field( pid_t, tgid )
644 __field( pid_t, ngid )
645 __field( int, src_cpu )
646 __field( int, src_nid )
647 __field( int, dst_cpu )
648 __field( int, dst_nid )
649 ),
650
651 TP_fast_assign(
652 __entry->pid = task_pid_nr(tsk);
653 __entry->tgid = task_tgid_nr(tsk);
654 __entry->ngid = task_numa_group_id(tsk);
655 __entry->src_cpu = src_cpu;
656 __entry->src_nid = cpu_to_node(src_cpu);
657 __entry->dst_cpu = dst_cpu;
658 __entry->dst_nid = cpu_to_node(dst_cpu);
659 ),
660
661 TP_printk("pid=%d tgid=%d ngid=%d src_cpu=%d src_nid=%d dst_cpu=%d dst_nid=%d",
662 __entry->pid, __entry->tgid, __entry->ngid,
663 __entry->src_cpu, __entry->src_nid,
664 __entry->dst_cpu, __entry->dst_nid)
665 );
666
667 DECLARE_EVENT_CLASS(sched_numa_pair_template,
668
669 TP_PROTO(struct task_struct *src_tsk, int src_cpu,
670 struct task_struct *dst_tsk, int dst_cpu),
671
672 TP_ARGS(src_tsk, src_cpu, dst_tsk, dst_cpu),
673
674 TP_STRUCT__entry(
675 __field( pid_t, src_pid )
676 __field( pid_t, src_tgid )
677 __field( pid_t, src_ngid )
678 __field( int, src_cpu )
679 __field( int, src_nid )
680 __field( pid_t, dst_pid )
681 __field( pid_t, dst_tgid )
682 __field( pid_t, dst_ngid )
683 __field( int, dst_cpu )
684 __field( int, dst_nid )
685 ),
686
687 TP_fast_assign(
688 __entry->src_pid = task_pid_nr(src_tsk);
689 __entry->src_tgid = task_tgid_nr(src_tsk);
690 __entry->src_ngid = task_numa_group_id(src_tsk);
691 __entry->src_cpu = src_cpu;
692 __entry->src_nid = cpu_to_node(src_cpu);
693 __entry->dst_pid = dst_tsk ? task_pid_nr(dst_tsk) : 0;
694 __entry->dst_tgid = dst_tsk ? task_tgid_nr(dst_tsk) : 0;
695 __entry->dst_ngid = dst_tsk ? task_numa_group_id(dst_tsk) : 0;
696 __entry->dst_cpu = dst_cpu;
697 __entry->dst_nid = dst_cpu >= 0 ? cpu_to_node(dst_cpu) : -1;
698 ),
699
700 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",
701 __entry->src_pid, __entry->src_tgid, __entry->src_ngid,
702 __entry->src_cpu, __entry->src_nid,
703 __entry->dst_pid, __entry->dst_tgid, __entry->dst_ngid,
704 __entry->dst_cpu, __entry->dst_nid)
705 );
706
707 DEFINE_EVENT(sched_numa_pair_template, sched_stick_numa,
708
709 TP_PROTO(struct task_struct *src_tsk, int src_cpu,
710 struct task_struct *dst_tsk, int dst_cpu),
711
712 TP_ARGS(src_tsk, src_cpu, dst_tsk, dst_cpu)
713 );
714
715 DEFINE_EVENT(sched_numa_pair_template, sched_swap_numa,
716
717 TP_PROTO(struct task_struct *src_tsk, int src_cpu,
718 struct task_struct *dst_tsk, int dst_cpu),
719
720 TP_ARGS(src_tsk, src_cpu, dst_tsk, dst_cpu)
721 );
722
723 #ifdef CONFIG_NUMA_BALANCING
724 #define NUMAB_SKIP_REASON \
725 EM( NUMAB_SKIP_UNSUITABLE, "unsuitable" ) \
726 EM( NUMAB_SKIP_SHARED_RO, "shared_ro" ) \
727 EM( NUMAB_SKIP_INACCESSIBLE, "inaccessible" ) \
728 EM( NUMAB_SKIP_SCAN_DELAY, "scan_delay" ) \
729 EM( NUMAB_SKIP_PID_INACTIVE, "pid_inactive" ) \
730 EM( NUMAB_SKIP_IGNORE_PID, "ignore_pid_inactive" ) \
731 EMe(NUMAB_SKIP_SEQ_COMPLETED, "seq_completed" )
732
733 /* Redefine for export. */
734 #undef EM
735 #undef EMe
736 #define EM(a, b) TRACE_DEFINE_ENUM(a);
737 #define EMe(a, b) TRACE_DEFINE_ENUM(a);
738
739 NUMAB_SKIP_REASON
740
741 /* Redefine for symbolic printing. */
742 #undef EM
743 #undef EMe
744 #define EM(a, b) { a, b },
745 #define EMe(a, b) { a, b }
746
747 TRACE_EVENT(sched_skip_vma_numa,
748
749 TP_PROTO(struct mm_struct *mm, struct vm_area_struct *vma,
750 enum numa_vmaskip_reason reason),
751
752 TP_ARGS(mm, vma, reason),
753
754 TP_STRUCT__entry(
755 __field(unsigned long, numa_scan_offset)
756 __field(unsigned long, vm_start)
757 __field(unsigned long, vm_end)
758 __field(enum numa_vmaskip_reason, reason)
759 ),
760
761 TP_fast_assign(
762 __entry->numa_scan_offset = mm->numa_scan_offset;
763 __entry->vm_start = vma->vm_start;
764 __entry->vm_end = vma->vm_end;
765 __entry->reason = reason;
766 ),
767
768 TP_printk("numa_scan_offset=%lX vm_start=%lX vm_end=%lX reason=%s",
769 __entry->numa_scan_offset,
770 __entry->vm_start,
771 __entry->vm_end,
772 __print_symbolic(__entry->reason, NUMAB_SKIP_REASON))
773 );
774 #endif /* CONFIG_NUMA_BALANCING */
775
776 /*
777 * Tracepoint for waking a polling cpu without an IPI.
778 */
779 TRACE_EVENT(sched_wake_idle_without_ipi,
780
781 TP_PROTO(int cpu),
782
783 TP_ARGS(cpu),
784
785 TP_STRUCT__entry(
786 __field( int, cpu )
787 ),
788
789 TP_fast_assign(
790 __entry->cpu = cpu;
791 ),
792
793 TP_printk("cpu=%d", __entry->cpu)
794 );
795
796 #ifdef CONFIG_SCHED_PROXY_EXEC
797 /**
798 * sched_pe_enqueue_sleeping_task - called when a task is enqueued on wait
799 * queue of a sleeping task (mutex owner).
800 * @mutex_owner: pointer to struct task_struct
801 * @blocked: pointer to struct task_struct
802 */
803 TRACE_EVENT(sched_pe_enqueue_sleeping_task,
804
805 TP_PROTO(struct task_struct *mutex_owner, struct task_struct *blocked),
806
807 TP_ARGS(mutex_owner, blocked),
808
809 TP_STRUCT__entry(
810 __array(char, owner_comm, TASK_COMM_LEN )
811 __field(pid_t, owner_pid )
812 __field(int, owner_prio )
813 __field(int, owner_cpu )
814 __array(char, blocked_comm, TASK_COMM_LEN )
815 __field(pid_t, blocked_pid )
816 __field(int, blocked_prio )
817 __field(int, blocked_cpu )
818 ),
819
820 TP_fast_assign(
821 strscpy(__entry->owner_comm, mutex_owner->comm, TASK_COMM_LEN);
822 __entry->owner_pid = mutex_owner->pid;
823 __entry->owner_prio = mutex_owner->prio; /* XXX SCHED_DEADLINE */
824 __entry->owner_cpu = task_cpu(mutex_owner);
825
826 strscpy(__entry->blocked_comm, blocked->comm, TASK_COMM_LEN);
827 __entry->blocked_pid = blocked->pid;
828 __entry->blocked_prio = blocked->prio; /* XXX SCHED_DEADLINE */
829 __entry->blocked_cpu = task_cpu(blocked);
830 ),
831
832 TP_printk("task=%s pid=%d prio=%d cpu=%d blocked_on owner_task=%s owner_pid=%d owner_prio=%d owner_cpu=%d",
833 __entry->blocked_comm, __entry->blocked_pid,
834 __entry->blocked_prio, __entry->blocked_cpu,
835 __entry->owner_comm, __entry->owner_pid,
836 __entry->owner_prio, __entry->owner_cpu)
837 );
838
839 /**
840 * sched_pe_activate_blocked_entity - called when a blocked entity is activated
841 * when mutex owner wakes
842 * @owner: pointer to struct task_struct
843 * @blocked: pointer to struct task_struct
844 */
845 TRACE_EVENT(sched_pe_activate_blocked_entity,
846
847 TP_PROTO(struct task_struct *owner, struct task_struct *blocked),
848
849 TP_ARGS(owner, blocked),
850
851 TP_STRUCT__entry(
852 __array(char, owner_comm, TASK_COMM_LEN )
853 __field(pid_t, owner_pid )
854 __field(int, owner_prio )
855 __field(int, owner_cpu )
856 __array(char, blocked_comm, TASK_COMM_LEN )
857 __field(pid_t, blocked_pid )
858 __field(int, blocked_prio )
859 __field(int, blocked_cpu )
860 ),
861
862 TP_fast_assign(
863 strscpy(__entry->owner_comm, owner->comm, TASK_COMM_LEN);
864 __entry->owner_pid = owner->pid;
865 __entry->owner_prio = owner->prio; /* XXX SCHED_DEADLINE */
866 __entry->owner_cpu = task_cpu(owner);
867
868 strscpy(__entry->blocked_comm, blocked->comm, TASK_COMM_LEN);
869 __entry->blocked_pid = blocked->pid;
870 __entry->blocked_prio = blocked->prio; /* XXX SCHED_DEADLINE */
871 __entry->blocked_cpu = task_cpu(blocked);
872 ),
873
874 TP_printk("activating task=%s pid=%d prio=%d cpu=%d, blocked_on owner=%s pid=%d prio=%d cpu=%d",
875 __entry->blocked_comm, __entry->blocked_pid,
876 __entry->blocked_prio, __entry->blocked_cpu,
877 __entry->owner_comm, __entry->owner_pid,
878 __entry->owner_prio, __entry->owner_cpu)
879 );
880
881 /**
882 * sched_pe_cross_remote_cpu - called when dependency chain crosses remote CPU
883 * @p: pointer to struct task_struct
884 */
885 TRACE_EVENT(sched_pe_migration,
886
887 TP_PROTO(struct task_struct *next, struct task_struct *owner),
888
889 TP_ARGS(next, owner),
890
891 TP_STRUCT__entry(
892 __array(char, next_comm, TASK_COMM_LEN )
893 __field(pid_t, next_pid )
894 __field(int, next_prio )
895 __array(char, owner_comm, TASK_COMM_LEN )
896 __field(pid_t, owner_pid )
897 __field(int, owner_prio )
898 __field(int, remote_cpu )
899 ),
900
901 TP_fast_assign(
902 strscpy(__entry->next_comm, next->comm, TASK_COMM_LEN);
903 __entry->next_pid = next->pid;
904 __entry->next_prio = next->prio; /* XXX SCHED_DEADLINE */
905 strscpy(__entry->owner_comm, owner->comm, TASK_COMM_LEN);
906 __entry->owner_pid = owner->pid;
907 __entry->owner_prio = owner->prio; /* XXX SCHED_DEADLINE */
908 __entry->remote_cpu = task_cpu(owner);
909 ),
910
911 TP_printk("next=%s pid=%d prio=%d owner=%s pid=%d prio=%d remote_cpu=%d",
912 __entry->next_comm, __entry->next_pid, __entry->next_prio,
913 __entry->owner_comm, __entry->owner_pid, __entry->owner_prio,
914 __entry->remote_cpu)
915 );
916
917 /**
918 * sched_pe_task_is_migrating - called when mutex owner is in migrating state
919 * @p: pointer to struct task_struct
920 */
921 TRACE_EVENT(sched_pe_owner_is_migrating,
922
923 TP_PROTO(struct task_struct *owner, struct task_struct *waiter),
924
925 TP_ARGS(owner, waiter),
926
927 TP_STRUCT__entry(
928 __array(char, owner_comm, TASK_COMM_LEN )
929 __field(pid_t, owner_pid )
930 __field(int, owner_prio )
931 __array(char, waiter_comm, TASK_COMM_LEN )
932 __field(pid_t, waiter_pid )
933 __field(int, waiter_prio )
934 ),
935
936 TP_fast_assign(
937 strscpy(__entry->owner_comm, owner->comm, TASK_COMM_LEN);
938 __entry->owner_pid = owner->pid;
939 __entry->owner_prio = owner->prio; /* XXX SCHED_DEADLINE */
940 strscpy(__entry->waiter_comm, waiter->comm, TASK_COMM_LEN);
941 __entry->waiter_pid = waiter->pid;
942 __entry->waiter_prio = waiter->prio; /* XXX SCHED_DEADLINE */
943 ),
944
945 TP_printk("owner comm=%s pid=%d prio=%d waiter comm=%s pid=%d prio=%d",
946 __entry->owner_comm, __entry->owner_pid, __entry->owner_prio,
947 __entry->waiter_comm, __entry->waiter_pid, __entry->waiter_prio)
948 );
949 #endif /* CONFIG_SCHED_PROXY_EXEC */
950
951 /**
952 * sched_pe_cross_remote_cpu - called when dependency chain crosses remote CPU
953 * @p: pointer to struct task_struct
954 */
955 TRACE_EVENT(sched_pe_return_migration,
956
957 TP_PROTO(struct task_struct *task, int target_cpu),
958
959 TP_ARGS(task, target_cpu),
960
961 TP_STRUCT__entry(
962 __array(char, comm, TASK_COMM_LEN )
963 __field(pid_t, pid )
964 __field(int, prio )
965 __field(int, target_cpu )
966 ),
967
968 TP_fast_assign(
969 strscpy(__entry->comm, task->comm, TASK_COMM_LEN);
970 __entry->pid = task->pid;
971 __entry->prio = task->prio; /* XXX SCHED_DEADLINE */
972 __entry->target_cpu = target_cpu;
973 ),
974
975 TP_printk("task=%s pid=%d prio=%d target_cpu=%d",
976 __entry->comm, __entry->pid, __entry->prio,
977 __entry->target_cpu)
978 );
979
980
981 TRACE_EVENT(sched_start_task_selection,
982
983 TP_PROTO(struct task_struct *prev, int cpu, bool is_blocked),
984
985 TP_ARGS(prev, cpu, is_blocked),
986
987 TP_STRUCT__entry(
988 __array(char, comm, TASK_COMM_LEN )
989 __field(pid_t, pid )
990 __field(int, prio )
991 __field(bool, is_blocked )
992 ),
993
994 TP_fast_assign(
995 strscpy(__entry->comm, prev->comm, TASK_COMM_LEN);
996 __entry->pid = prev->pid;
997 __entry->prio = prev->prio; /* XXX SCHED_DEADLINE */
998 __entry->is_blocked = is_blocked;
999 ),
1000
1001 TP_printk("prev=%s pid=%d prio=%d is_blocked: %i",
1002 __entry->comm, __entry->pid, __entry->prio, __entry->is_blocked)
1003 );
1004
1005 TRACE_EVENT(sched_finish_task_selection,
1006
1007 TP_PROTO(struct task_struct *donor, struct task_struct *next, int cpu),
1008
1009 TP_ARGS(donor, next, cpu),
1010
1011 TP_STRUCT__entry(
1012 __array(char, donor_comm, TASK_COMM_LEN )
1013 __field(pid_t, donor_pid )
1014 __field(int, donor_prio )
1015 __array(char, next_comm, TASK_COMM_LEN )
1016 __field(pid_t, next_pid )
1017 __field(int, next_prio )
1018 __field(int, cpu )
1019 ),
1020
1021 TP_fast_assign(
1022 strscpy(__entry->donor_comm, donor->comm, TASK_COMM_LEN);
1023 __entry->donor_pid = donor->pid;
1024 __entry->donor_prio = donor->prio; /* XXX SCHED_DEADLINE */
1025 strscpy(__entry->next_comm, next->comm, TASK_COMM_LEN);
1026 __entry->next_pid = next->pid;
1027 __entry->next_prio = next->prio; /* XXX SCHED_DEADLINE */
1028 __entry->cpu = cpu;
1029 ),
1030
1031 TP_printk("donor=%s pid=%d prio=%d next=%s pid=%d prio=%d cpu=%d",
1032 __entry->donor_comm, __entry->donor_pid, __entry->donor_prio,
1033 __entry->next_comm, __entry->next_pid, __entry->next_prio,
1034 __entry->cpu)
1035 );
1036
1037 /*
1038 * Following tracepoints are not exported in tracefs and provide hooking
1039 * mechanisms only for testing and debugging purposes.
1040 *
1041 * Postfixed with _tp to make them easily identifiable in the code.
1042 */
1043 DECLARE_TRACE(pelt_cfs_tp,
1044 TP_PROTO(struct cfs_rq *cfs_rq),
1045 TP_ARGS(cfs_rq));
1046
1047 DECLARE_TRACE(pelt_rt_tp,
1048 TP_PROTO(struct rq *rq),
1049 TP_ARGS(rq));
1050
1051 DECLARE_TRACE(pelt_dl_tp,
1052 TP_PROTO(struct rq *rq),
1053 TP_ARGS(rq));
1054
1055 DECLARE_TRACE(pelt_hw_tp,
1056 TP_PROTO(struct rq *rq),
1057 TP_ARGS(rq));
1058
1059 DECLARE_TRACE(pelt_irq_tp,
1060 TP_PROTO(struct rq *rq),
1061 TP_ARGS(rq));
1062
1063 DECLARE_TRACE(pelt_se_tp,
1064 TP_PROTO(struct sched_entity *se),
1065 TP_ARGS(se));
1066
1067 DECLARE_TRACE(sched_cpu_capacity_tp,
1068 TP_PROTO(struct rq *rq),
1069 TP_ARGS(rq));
1070
1071 DECLARE_TRACE(sched_overutilized_tp,
1072 TP_PROTO(struct root_domain *rd, bool overutilized),
1073 TP_ARGS(rd, overutilized));
1074
1075 DECLARE_TRACE(sched_util_est_cfs_tp,
1076 TP_PROTO(struct cfs_rq *cfs_rq),
1077 TP_ARGS(cfs_rq));
1078
1079 DECLARE_TRACE(sched_util_est_se_tp,
1080 TP_PROTO(struct sched_entity *se),
1081 TP_ARGS(se));
1082
1083 DECLARE_TRACE(sched_update_nr_running_tp,
1084 TP_PROTO(struct rq *rq, int change),
1085 TP_ARGS(rq, change));
1086
1087 DECLARE_TRACE(sched_compute_energy_tp,
1088 TP_PROTO(struct task_struct *p, int dst_cpu, unsigned long energy,
1089 unsigned long max_util, unsigned long busy_time),
1090 TP_ARGS(p, dst_cpu, energy, max_util, busy_time));
1091
1092 #endif /* _TRACE_SCHED_H */
1093
1094 /* This part must be outside protection */
1095 #include <trace/define_trace.h>
1096