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