Loading...
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 */
15TRACE_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 */
37TRACE_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 */
57DECLARE_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 */
88DEFINE_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 */
96DEFINE_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 */
103DEFINE_EVENT(sched_wakeup_template, sched_wakeup_new,
104 TP_PROTO(struct task_struct *p),
105 TP_ARGS(p));
106
107#ifdef CREATE_TRACE_POINTS
108static 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 */
138TRACE_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 */
189TRACE_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
216DECLARE_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 */
241DEFINE_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 */
248DEFINE_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 */
255DEFINE_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 */
262TRACE_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 */
287TRACE_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 */
315TRACE_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 */
351DECLARE_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 */
378DEFINE_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 */
386DEFINE_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 */
394DEFINE_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 */
401DEFINE_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 accounting runtime (time the task is executing
407 * on a CPU).
408 */
409DECLARE_EVENT_CLASS(sched_stat_runtime,
410
411 TP_PROTO(struct task_struct *tsk, u64 runtime, u64 vruntime),
412
413 TP_ARGS(tsk, __perf_count(runtime), vruntime),
414
415 TP_STRUCT__entry(
416 __array( char, comm, TASK_COMM_LEN )
417 __field( pid_t, pid )
418 __field( u64, runtime )
419 __field( u64, vruntime )
420 ),
421
422 TP_fast_assign(
423 memcpy(__entry->comm, tsk->comm, TASK_COMM_LEN);
424 __entry->pid = tsk->pid;
425 __entry->runtime = runtime;
426 __entry->vruntime = vruntime;
427 ),
428
429 TP_printk("comm=%s pid=%d runtime=%Lu [ns] vruntime=%Lu [ns]",
430 __entry->comm, __entry->pid,
431 (unsigned long long)__entry->runtime,
432 (unsigned long long)__entry->vruntime)
433);
434
435DEFINE_EVENT(sched_stat_runtime, sched_stat_runtime,
436 TP_PROTO(struct task_struct *tsk, u64 runtime, u64 vruntime),
437 TP_ARGS(tsk, runtime, vruntime));
438
439/*
440 * Tracepoint for showing priority inheritance modifying a tasks
441 * priority.
442 */
443TRACE_EVENT(sched_pi_setprio,
444
445 TP_PROTO(struct task_struct *tsk, struct task_struct *pi_task),
446
447 TP_ARGS(tsk, pi_task),
448
449 TP_STRUCT__entry(
450 __array( char, comm, TASK_COMM_LEN )
451 __field( pid_t, pid )
452 __field( int, oldprio )
453 __field( int, newprio )
454 ),
455
456 TP_fast_assign(
457 memcpy(__entry->comm, tsk->comm, TASK_COMM_LEN);
458 __entry->pid = tsk->pid;
459 __entry->oldprio = tsk->prio;
460 __entry->newprio = pi_task ?
461 min(tsk->normal_prio, pi_task->prio) :
462 tsk->normal_prio;
463 /* XXX SCHED_DEADLINE bits missing */
464 ),
465
466 TP_printk("comm=%s pid=%d oldprio=%d newprio=%d",
467 __entry->comm, __entry->pid,
468 __entry->oldprio, __entry->newprio)
469);
470
471#ifdef CONFIG_DETECT_HUNG_TASK
472TRACE_EVENT(sched_process_hang,
473 TP_PROTO(struct task_struct *tsk),
474 TP_ARGS(tsk),
475
476 TP_STRUCT__entry(
477 __array( char, comm, TASK_COMM_LEN )
478 __field( pid_t, pid )
479 ),
480
481 TP_fast_assign(
482 memcpy(__entry->comm, tsk->comm, TASK_COMM_LEN);
483 __entry->pid = tsk->pid;
484 ),
485
486 TP_printk("comm=%s pid=%d", __entry->comm, __entry->pid)
487);
488#endif /* CONFIG_DETECT_HUNG_TASK */
489
490DECLARE_EVENT_CLASS(sched_move_task_template,
491
492 TP_PROTO(struct task_struct *tsk, int src_cpu, int dst_cpu),
493
494 TP_ARGS(tsk, src_cpu, dst_cpu),
495
496 TP_STRUCT__entry(
497 __field( pid_t, pid )
498 __field( pid_t, tgid )
499 __field( pid_t, ngid )
500 __field( int, src_cpu )
501 __field( int, src_nid )
502 __field( int, dst_cpu )
503 __field( int, dst_nid )
504 ),
505
506 TP_fast_assign(
507 __entry->pid = task_pid_nr(tsk);
508 __entry->tgid = task_tgid_nr(tsk);
509 __entry->ngid = task_numa_group_id(tsk);
510 __entry->src_cpu = src_cpu;
511 __entry->src_nid = cpu_to_node(src_cpu);
512 __entry->dst_cpu = dst_cpu;
513 __entry->dst_nid = cpu_to_node(dst_cpu);
514 ),
515
516 TP_printk("pid=%d tgid=%d ngid=%d src_cpu=%d src_nid=%d dst_cpu=%d dst_nid=%d",
517 __entry->pid, __entry->tgid, __entry->ngid,
518 __entry->src_cpu, __entry->src_nid,
519 __entry->dst_cpu, __entry->dst_nid)
520);
521
522/*
523 * Tracks migration of tasks from one runqueue to another. Can be used to
524 * detect if automatic NUMA balancing is bouncing between nodes
525 */
526DEFINE_EVENT(sched_move_task_template, sched_move_numa,
527 TP_PROTO(struct task_struct *tsk, int src_cpu, int dst_cpu),
528
529 TP_ARGS(tsk, src_cpu, dst_cpu)
530);
531
532DEFINE_EVENT(sched_move_task_template, sched_stick_numa,
533 TP_PROTO(struct task_struct *tsk, int src_cpu, int dst_cpu),
534
535 TP_ARGS(tsk, src_cpu, dst_cpu)
536);
537
538TRACE_EVENT(sched_swap_numa,
539
540 TP_PROTO(struct task_struct *src_tsk, int src_cpu,
541 struct task_struct *dst_tsk, int dst_cpu),
542
543 TP_ARGS(src_tsk, src_cpu, dst_tsk, dst_cpu),
544
545 TP_STRUCT__entry(
546 __field( pid_t, src_pid )
547 __field( pid_t, src_tgid )
548 __field( pid_t, src_ngid )
549 __field( int, src_cpu )
550 __field( int, src_nid )
551 __field( pid_t, dst_pid )
552 __field( pid_t, dst_tgid )
553 __field( pid_t, dst_ngid )
554 __field( int, dst_cpu )
555 __field( int, dst_nid )
556 ),
557
558 TP_fast_assign(
559 __entry->src_pid = task_pid_nr(src_tsk);
560 __entry->src_tgid = task_tgid_nr(src_tsk);
561 __entry->src_ngid = task_numa_group_id(src_tsk);
562 __entry->src_cpu = src_cpu;
563 __entry->src_nid = cpu_to_node(src_cpu);
564 __entry->dst_pid = task_pid_nr(dst_tsk);
565 __entry->dst_tgid = task_tgid_nr(dst_tsk);
566 __entry->dst_ngid = task_numa_group_id(dst_tsk);
567 __entry->dst_cpu = dst_cpu;
568 __entry->dst_nid = cpu_to_node(dst_cpu);
569 ),
570
571 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",
572 __entry->src_pid, __entry->src_tgid, __entry->src_ngid,
573 __entry->src_cpu, __entry->src_nid,
574 __entry->dst_pid, __entry->dst_tgid, __entry->dst_ngid,
575 __entry->dst_cpu, __entry->dst_nid)
576);
577
578/*
579 * Tracepoint for waking a polling cpu without an IPI.
580 */
581TRACE_EVENT(sched_wake_idle_without_ipi,
582
583 TP_PROTO(int cpu),
584
585 TP_ARGS(cpu),
586
587 TP_STRUCT__entry(
588 __field( int, cpu )
589 ),
590
591 TP_fast_assign(
592 __entry->cpu = cpu;
593 ),
594
595 TP_printk("cpu=%d", __entry->cpu)
596);
597
598/*
599 * Following tracepoints are not exported in tracefs and provide hooking
600 * mechanisms only for testing and debugging purposes.
601 *
602 * Postfixed with _tp to make them easily identifiable in the code.
603 */
604DECLARE_TRACE(pelt_cfs_tp,
605 TP_PROTO(struct cfs_rq *cfs_rq),
606 TP_ARGS(cfs_rq));
607
608DECLARE_TRACE(pelt_rt_tp,
609 TP_PROTO(struct rq *rq),
610 TP_ARGS(rq));
611
612DECLARE_TRACE(pelt_dl_tp,
613 TP_PROTO(struct rq *rq),
614 TP_ARGS(rq));
615
616DECLARE_TRACE(pelt_irq_tp,
617 TP_PROTO(struct rq *rq),
618 TP_ARGS(rq));
619
620DECLARE_TRACE(pelt_se_tp,
621 TP_PROTO(struct sched_entity *se),
622 TP_ARGS(se));
623
624DECLARE_TRACE(sched_overutilized_tp,
625 TP_PROTO(struct root_domain *rd, bool overutilized),
626 TP_ARGS(rd, overutilized));
627
628#endif /* _TRACE_SCHED_H */
629
630/* This part must be outside protection */
631#include <trace/define_trace.h>
1#undef TRACE_SYSTEM
2#define TRACE_SYSTEM sched
3
4#if !defined(_TRACE_SCHED_H) || defined(TRACE_HEADER_MULTI_READ)
5#define _TRACE_SCHED_H
6
7#include <linux/sched.h>
8#include <linux/tracepoint.h>
9#include <linux/binfmts.h>
10
11/*
12 * Tracepoint for calling kthread_stop, performed to end a kthread:
13 */
14TRACE_EVENT(sched_kthread_stop,
15
16 TP_PROTO(struct task_struct *t),
17
18 TP_ARGS(t),
19
20 TP_STRUCT__entry(
21 __array( char, comm, TASK_COMM_LEN )
22 __field( pid_t, pid )
23 ),
24
25 TP_fast_assign(
26 memcpy(__entry->comm, t->comm, TASK_COMM_LEN);
27 __entry->pid = t->pid;
28 ),
29
30 TP_printk("comm=%s pid=%d", __entry->comm, __entry->pid)
31);
32
33/*
34 * Tracepoint for the return value of the kthread stopping:
35 */
36TRACE_EVENT(sched_kthread_stop_ret,
37
38 TP_PROTO(int ret),
39
40 TP_ARGS(ret),
41
42 TP_STRUCT__entry(
43 __field( int, ret )
44 ),
45
46 TP_fast_assign(
47 __entry->ret = ret;
48 ),
49
50 TP_printk("ret=%d", __entry->ret)
51);
52
53/*
54 * Tracepoint for waking up a task:
55 */
56DECLARE_EVENT_CLASS(sched_wakeup_template,
57
58 TP_PROTO(struct task_struct *p),
59
60 TP_ARGS(__perf_task(p)),
61
62 TP_STRUCT__entry(
63 __array( char, comm, TASK_COMM_LEN )
64 __field( pid_t, pid )
65 __field( int, prio )
66 __field( int, success )
67 __field( int, target_cpu )
68 ),
69
70 TP_fast_assign(
71 memcpy(__entry->comm, p->comm, TASK_COMM_LEN);
72 __entry->pid = p->pid;
73 __entry->prio = p->prio;
74 __entry->success = 1; /* rudiment, kill when possible */
75 __entry->target_cpu = task_cpu(p);
76 ),
77
78 TP_printk("comm=%s pid=%d prio=%d target_cpu=%03d",
79 __entry->comm, __entry->pid, __entry->prio,
80 __entry->target_cpu)
81);
82
83/*
84 * Tracepoint called when waking a task; this tracepoint is guaranteed to be
85 * called from the waking context.
86 */
87DEFINE_EVENT(sched_wakeup_template, sched_waking,
88 TP_PROTO(struct task_struct *p),
89 TP_ARGS(p));
90
91/*
92 * Tracepoint called when the task is actually woken; p->state == TASK_RUNNNG.
93 * It it not always called from the waking context.
94 */
95DEFINE_EVENT(sched_wakeup_template, sched_wakeup,
96 TP_PROTO(struct task_struct *p),
97 TP_ARGS(p));
98
99/*
100 * Tracepoint for waking up a new task:
101 */
102DEFINE_EVENT(sched_wakeup_template, sched_wakeup_new,
103 TP_PROTO(struct task_struct *p),
104 TP_ARGS(p));
105
106#ifdef CREATE_TRACE_POINTS
107static inline long __trace_sched_switch_state(bool preempt, struct task_struct *p)
108{
109#ifdef CONFIG_SCHED_DEBUG
110 BUG_ON(p != current);
111#endif /* CONFIG_SCHED_DEBUG */
112
113 /*
114 * Preemption ignores task state, therefore preempted tasks are always
115 * RUNNING (we will not have dequeued if state != RUNNING).
116 */
117 return preempt ? TASK_RUNNING | TASK_STATE_MAX : p->state;
118}
119#endif /* CREATE_TRACE_POINTS */
120
121/*
122 * Tracepoint for task switches, performed by the scheduler:
123 */
124TRACE_EVENT(sched_switch,
125
126 TP_PROTO(bool preempt,
127 struct task_struct *prev,
128 struct task_struct *next),
129
130 TP_ARGS(preempt, prev, next),
131
132 TP_STRUCT__entry(
133 __array( char, prev_comm, TASK_COMM_LEN )
134 __field( pid_t, prev_pid )
135 __field( int, prev_prio )
136 __field( long, prev_state )
137 __array( char, next_comm, TASK_COMM_LEN )
138 __field( pid_t, next_pid )
139 __field( int, next_prio )
140 ),
141
142 TP_fast_assign(
143 memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN);
144 __entry->prev_pid = prev->pid;
145 __entry->prev_prio = prev->prio;
146 __entry->prev_state = __trace_sched_switch_state(preempt, prev);
147 memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN);
148 __entry->next_pid = next->pid;
149 __entry->next_prio = next->prio;
150 ),
151
152 TP_printk("prev_comm=%s prev_pid=%d prev_prio=%d prev_state=%s%s ==> next_comm=%s next_pid=%d next_prio=%d",
153 __entry->prev_comm, __entry->prev_pid, __entry->prev_prio,
154 __entry->prev_state & (TASK_STATE_MAX-1) ?
155 __print_flags(__entry->prev_state & (TASK_STATE_MAX-1), "|",
156 { 1, "S"} , { 2, "D" }, { 4, "T" }, { 8, "t" },
157 { 16, "Z" }, { 32, "X" }, { 64, "x" },
158 { 128, "K" }, { 256, "W" }, { 512, "P" },
159 { 1024, "N" }) : "R",
160 __entry->prev_state & TASK_STATE_MAX ? "+" : "",
161 __entry->next_comm, __entry->next_pid, __entry->next_prio)
162);
163
164/*
165 * Tracepoint for a task being migrated:
166 */
167TRACE_EVENT(sched_migrate_task,
168
169 TP_PROTO(struct task_struct *p, int dest_cpu),
170
171 TP_ARGS(p, dest_cpu),
172
173 TP_STRUCT__entry(
174 __array( char, comm, TASK_COMM_LEN )
175 __field( pid_t, pid )
176 __field( int, prio )
177 __field( int, orig_cpu )
178 __field( int, dest_cpu )
179 ),
180
181 TP_fast_assign(
182 memcpy(__entry->comm, p->comm, TASK_COMM_LEN);
183 __entry->pid = p->pid;
184 __entry->prio = p->prio;
185 __entry->orig_cpu = task_cpu(p);
186 __entry->dest_cpu = dest_cpu;
187 ),
188
189 TP_printk("comm=%s pid=%d prio=%d orig_cpu=%d dest_cpu=%d",
190 __entry->comm, __entry->pid, __entry->prio,
191 __entry->orig_cpu, __entry->dest_cpu)
192);
193
194DECLARE_EVENT_CLASS(sched_process_template,
195
196 TP_PROTO(struct task_struct *p),
197
198 TP_ARGS(p),
199
200 TP_STRUCT__entry(
201 __array( char, comm, TASK_COMM_LEN )
202 __field( pid_t, pid )
203 __field( int, prio )
204 ),
205
206 TP_fast_assign(
207 memcpy(__entry->comm, p->comm, TASK_COMM_LEN);
208 __entry->pid = p->pid;
209 __entry->prio = p->prio;
210 ),
211
212 TP_printk("comm=%s pid=%d prio=%d",
213 __entry->comm, __entry->pid, __entry->prio)
214);
215
216/*
217 * Tracepoint for freeing a task:
218 */
219DEFINE_EVENT(sched_process_template, sched_process_free,
220 TP_PROTO(struct task_struct *p),
221 TP_ARGS(p));
222
223
224/*
225 * Tracepoint for a task exiting:
226 */
227DEFINE_EVENT(sched_process_template, sched_process_exit,
228 TP_PROTO(struct task_struct *p),
229 TP_ARGS(p));
230
231/*
232 * Tracepoint for waiting on task to unschedule:
233 */
234DEFINE_EVENT(sched_process_template, sched_wait_task,
235 TP_PROTO(struct task_struct *p),
236 TP_ARGS(p));
237
238/*
239 * Tracepoint for a waiting task:
240 */
241TRACE_EVENT(sched_process_wait,
242
243 TP_PROTO(struct pid *pid),
244
245 TP_ARGS(pid),
246
247 TP_STRUCT__entry(
248 __array( char, comm, TASK_COMM_LEN )
249 __field( pid_t, pid )
250 __field( int, prio )
251 ),
252
253 TP_fast_assign(
254 memcpy(__entry->comm, current->comm, TASK_COMM_LEN);
255 __entry->pid = pid_nr(pid);
256 __entry->prio = current->prio;
257 ),
258
259 TP_printk("comm=%s pid=%d prio=%d",
260 __entry->comm, __entry->pid, __entry->prio)
261);
262
263/*
264 * Tracepoint for do_fork:
265 */
266TRACE_EVENT(sched_process_fork,
267
268 TP_PROTO(struct task_struct *parent, struct task_struct *child),
269
270 TP_ARGS(parent, child),
271
272 TP_STRUCT__entry(
273 __array( char, parent_comm, TASK_COMM_LEN )
274 __field( pid_t, parent_pid )
275 __array( char, child_comm, TASK_COMM_LEN )
276 __field( pid_t, child_pid )
277 ),
278
279 TP_fast_assign(
280 memcpy(__entry->parent_comm, parent->comm, TASK_COMM_LEN);
281 __entry->parent_pid = parent->pid;
282 memcpy(__entry->child_comm, child->comm, TASK_COMM_LEN);
283 __entry->child_pid = child->pid;
284 ),
285
286 TP_printk("comm=%s pid=%d child_comm=%s child_pid=%d",
287 __entry->parent_comm, __entry->parent_pid,
288 __entry->child_comm, __entry->child_pid)
289);
290
291/*
292 * Tracepoint for exec:
293 */
294TRACE_EVENT(sched_process_exec,
295
296 TP_PROTO(struct task_struct *p, pid_t old_pid,
297 struct linux_binprm *bprm),
298
299 TP_ARGS(p, old_pid, bprm),
300
301 TP_STRUCT__entry(
302 __string( filename, bprm->filename )
303 __field( pid_t, pid )
304 __field( pid_t, old_pid )
305 ),
306
307 TP_fast_assign(
308 __assign_str(filename, bprm->filename);
309 __entry->pid = p->pid;
310 __entry->old_pid = old_pid;
311 ),
312
313 TP_printk("filename=%s pid=%d old_pid=%d", __get_str(filename),
314 __entry->pid, __entry->old_pid)
315);
316
317/*
318 * XXX the below sched_stat tracepoints only apply to SCHED_OTHER/BATCH/IDLE
319 * adding sched_stat support to SCHED_FIFO/RR would be welcome.
320 */
321DECLARE_EVENT_CLASS(sched_stat_template,
322
323 TP_PROTO(struct task_struct *tsk, u64 delay),
324
325 TP_ARGS(__perf_task(tsk), __perf_count(delay)),
326
327 TP_STRUCT__entry(
328 __array( char, comm, TASK_COMM_LEN )
329 __field( pid_t, pid )
330 __field( u64, delay )
331 ),
332
333 TP_fast_assign(
334 memcpy(__entry->comm, tsk->comm, TASK_COMM_LEN);
335 __entry->pid = tsk->pid;
336 __entry->delay = delay;
337 ),
338
339 TP_printk("comm=%s pid=%d delay=%Lu [ns]",
340 __entry->comm, __entry->pid,
341 (unsigned long long)__entry->delay)
342);
343
344
345/*
346 * Tracepoint for accounting wait time (time the task is runnable
347 * but not actually running due to scheduler contention).
348 */
349DEFINE_EVENT(sched_stat_template, sched_stat_wait,
350 TP_PROTO(struct task_struct *tsk, u64 delay),
351 TP_ARGS(tsk, delay));
352
353/*
354 * Tracepoint for accounting sleep time (time the task is not runnable,
355 * including iowait, see below).
356 */
357DEFINE_EVENT(sched_stat_template, sched_stat_sleep,
358 TP_PROTO(struct task_struct *tsk, u64 delay),
359 TP_ARGS(tsk, delay));
360
361/*
362 * Tracepoint for accounting iowait time (time the task is not runnable
363 * due to waiting on IO to complete).
364 */
365DEFINE_EVENT(sched_stat_template, sched_stat_iowait,
366 TP_PROTO(struct task_struct *tsk, u64 delay),
367 TP_ARGS(tsk, delay));
368
369/*
370 * Tracepoint for accounting blocked time (time the task is in uninterruptible).
371 */
372DEFINE_EVENT(sched_stat_template, sched_stat_blocked,
373 TP_PROTO(struct task_struct *tsk, u64 delay),
374 TP_ARGS(tsk, delay));
375
376/*
377 * Tracepoint for accounting runtime (time the task is executing
378 * on a CPU).
379 */
380DECLARE_EVENT_CLASS(sched_stat_runtime,
381
382 TP_PROTO(struct task_struct *tsk, u64 runtime, u64 vruntime),
383
384 TP_ARGS(tsk, __perf_count(runtime), vruntime),
385
386 TP_STRUCT__entry(
387 __array( char, comm, TASK_COMM_LEN )
388 __field( pid_t, pid )
389 __field( u64, runtime )
390 __field( u64, vruntime )
391 ),
392
393 TP_fast_assign(
394 memcpy(__entry->comm, tsk->comm, TASK_COMM_LEN);
395 __entry->pid = tsk->pid;
396 __entry->runtime = runtime;
397 __entry->vruntime = vruntime;
398 ),
399
400 TP_printk("comm=%s pid=%d runtime=%Lu [ns] vruntime=%Lu [ns]",
401 __entry->comm, __entry->pid,
402 (unsigned long long)__entry->runtime,
403 (unsigned long long)__entry->vruntime)
404);
405
406DEFINE_EVENT(sched_stat_runtime, sched_stat_runtime,
407 TP_PROTO(struct task_struct *tsk, u64 runtime, u64 vruntime),
408 TP_ARGS(tsk, runtime, vruntime));
409
410/*
411 * Tracepoint for showing priority inheritance modifying a tasks
412 * priority.
413 */
414TRACE_EVENT(sched_pi_setprio,
415
416 TP_PROTO(struct task_struct *tsk, int newprio),
417
418 TP_ARGS(tsk, newprio),
419
420 TP_STRUCT__entry(
421 __array( char, comm, TASK_COMM_LEN )
422 __field( pid_t, pid )
423 __field( int, oldprio )
424 __field( int, newprio )
425 ),
426
427 TP_fast_assign(
428 memcpy(__entry->comm, tsk->comm, TASK_COMM_LEN);
429 __entry->pid = tsk->pid;
430 __entry->oldprio = tsk->prio;
431 __entry->newprio = newprio;
432 ),
433
434 TP_printk("comm=%s pid=%d oldprio=%d newprio=%d",
435 __entry->comm, __entry->pid,
436 __entry->oldprio, __entry->newprio)
437);
438
439#ifdef CONFIG_DETECT_HUNG_TASK
440TRACE_EVENT(sched_process_hang,
441 TP_PROTO(struct task_struct *tsk),
442 TP_ARGS(tsk),
443
444 TP_STRUCT__entry(
445 __array( char, comm, TASK_COMM_LEN )
446 __field( pid_t, pid )
447 ),
448
449 TP_fast_assign(
450 memcpy(__entry->comm, tsk->comm, TASK_COMM_LEN);
451 __entry->pid = tsk->pid;
452 ),
453
454 TP_printk("comm=%s pid=%d", __entry->comm, __entry->pid)
455);
456#endif /* CONFIG_DETECT_HUNG_TASK */
457
458DECLARE_EVENT_CLASS(sched_move_task_template,
459
460 TP_PROTO(struct task_struct *tsk, int src_cpu, int dst_cpu),
461
462 TP_ARGS(tsk, src_cpu, dst_cpu),
463
464 TP_STRUCT__entry(
465 __field( pid_t, pid )
466 __field( pid_t, tgid )
467 __field( pid_t, ngid )
468 __field( int, src_cpu )
469 __field( int, src_nid )
470 __field( int, dst_cpu )
471 __field( int, dst_nid )
472 ),
473
474 TP_fast_assign(
475 __entry->pid = task_pid_nr(tsk);
476 __entry->tgid = task_tgid_nr(tsk);
477 __entry->ngid = task_numa_group_id(tsk);
478 __entry->src_cpu = src_cpu;
479 __entry->src_nid = cpu_to_node(src_cpu);
480 __entry->dst_cpu = dst_cpu;
481 __entry->dst_nid = cpu_to_node(dst_cpu);
482 ),
483
484 TP_printk("pid=%d tgid=%d ngid=%d src_cpu=%d src_nid=%d dst_cpu=%d dst_nid=%d",
485 __entry->pid, __entry->tgid, __entry->ngid,
486 __entry->src_cpu, __entry->src_nid,
487 __entry->dst_cpu, __entry->dst_nid)
488);
489
490/*
491 * Tracks migration of tasks from one runqueue to another. Can be used to
492 * detect if automatic NUMA balancing is bouncing between nodes
493 */
494DEFINE_EVENT(sched_move_task_template, sched_move_numa,
495 TP_PROTO(struct task_struct *tsk, int src_cpu, int dst_cpu),
496
497 TP_ARGS(tsk, src_cpu, dst_cpu)
498);
499
500DEFINE_EVENT(sched_move_task_template, sched_stick_numa,
501 TP_PROTO(struct task_struct *tsk, int src_cpu, int dst_cpu),
502
503 TP_ARGS(tsk, src_cpu, dst_cpu)
504);
505
506TRACE_EVENT(sched_swap_numa,
507
508 TP_PROTO(struct task_struct *src_tsk, int src_cpu,
509 struct task_struct *dst_tsk, int dst_cpu),
510
511 TP_ARGS(src_tsk, src_cpu, dst_tsk, dst_cpu),
512
513 TP_STRUCT__entry(
514 __field( pid_t, src_pid )
515 __field( pid_t, src_tgid )
516 __field( pid_t, src_ngid )
517 __field( int, src_cpu )
518 __field( int, src_nid )
519 __field( pid_t, dst_pid )
520 __field( pid_t, dst_tgid )
521 __field( pid_t, dst_ngid )
522 __field( int, dst_cpu )
523 __field( int, dst_nid )
524 ),
525
526 TP_fast_assign(
527 __entry->src_pid = task_pid_nr(src_tsk);
528 __entry->src_tgid = task_tgid_nr(src_tsk);
529 __entry->src_ngid = task_numa_group_id(src_tsk);
530 __entry->src_cpu = src_cpu;
531 __entry->src_nid = cpu_to_node(src_cpu);
532 __entry->dst_pid = task_pid_nr(dst_tsk);
533 __entry->dst_tgid = task_tgid_nr(dst_tsk);
534 __entry->dst_ngid = task_numa_group_id(dst_tsk);
535 __entry->dst_cpu = dst_cpu;
536 __entry->dst_nid = cpu_to_node(dst_cpu);
537 ),
538
539 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",
540 __entry->src_pid, __entry->src_tgid, __entry->src_ngid,
541 __entry->src_cpu, __entry->src_nid,
542 __entry->dst_pid, __entry->dst_tgid, __entry->dst_ngid,
543 __entry->dst_cpu, __entry->dst_nid)
544);
545
546/*
547 * Tracepoint for waking a polling cpu without an IPI.
548 */
549TRACE_EVENT(sched_wake_idle_without_ipi,
550
551 TP_PROTO(int cpu),
552
553 TP_ARGS(cpu),
554
555 TP_STRUCT__entry(
556 __field( int, cpu )
557 ),
558
559 TP_fast_assign(
560 __entry->cpu = cpu;
561 ),
562
563 TP_printk("cpu=%d", __entry->cpu)
564);
565#endif /* _TRACE_SCHED_H */
566
567/* This part must be outside protection */
568#include <trace/define_trace.h>