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#ifdef CONFIG_SCHED_DEBUG
111 BUG_ON(p != current);
112#endif /* CONFIG_SCHED_DEBUG */
113
114 /*
115 * Preemption ignores task state, therefore preempted tasks are always
116 * RUNNING (we will not have dequeued if state != RUNNING).
117 */
118 if (preempt)
119 return TASK_REPORT_MAX;
120
121 return 1 << task_state_index(p);
122}
123#endif /* CREATE_TRACE_POINTS */
124
125/*
126 * Tracepoint for task switches, performed by the scheduler:
127 */
128TRACE_EVENT(sched_switch,
129
130 TP_PROTO(bool preempt,
131 struct task_struct *prev,
132 struct task_struct *next),
133
134 TP_ARGS(preempt, prev, next),
135
136 TP_STRUCT__entry(
137 __array( char, prev_comm, TASK_COMM_LEN )
138 __field( pid_t, prev_pid )
139 __field( int, prev_prio )
140 __field( long, prev_state )
141 __array( char, next_comm, TASK_COMM_LEN )
142 __field( pid_t, next_pid )
143 __field( int, next_prio )
144 ),
145
146 TP_fast_assign(
147 memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN);
148 __entry->prev_pid = prev->pid;
149 __entry->prev_prio = prev->prio;
150 __entry->prev_state = __trace_sched_switch_state(preempt, prev);
151 memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN);
152 __entry->next_pid = next->pid;
153 __entry->next_prio = next->prio;
154 /* XXX SCHED_DEADLINE */
155 ),
156
157 TP_printk("prev_comm=%s prev_pid=%d prev_prio=%d prev_state=%s%s ==> next_comm=%s next_pid=%d next_prio=%d",
158 __entry->prev_comm, __entry->prev_pid, __entry->prev_prio,
159
160 (__entry->prev_state & (TASK_REPORT_MAX - 1)) ?
161 __print_flags(__entry->prev_state & (TASK_REPORT_MAX - 1), "|",
162 { 0x01, "S" }, { 0x02, "D" }, { 0x04, "T" },
163 { 0x08, "t" }, { 0x10, "X" }, { 0x20, "Z" },
164 { 0x40, "P" }, { 0x80, "I" }) :
165 "R",
166
167 __entry->prev_state & TASK_REPORT_MAX ? "+" : "",
168 __entry->next_comm, __entry->next_pid, __entry->next_prio)
169);
170
171/*
172 * Tracepoint for a task being migrated:
173 */
174TRACE_EVENT(sched_migrate_task,
175
176 TP_PROTO(struct task_struct *p, int dest_cpu),
177
178 TP_ARGS(p, dest_cpu),
179
180 TP_STRUCT__entry(
181 __array( char, comm, TASK_COMM_LEN )
182 __field( pid_t, pid )
183 __field( int, prio )
184 __field( int, orig_cpu )
185 __field( int, dest_cpu )
186 ),
187
188 TP_fast_assign(
189 memcpy(__entry->comm, p->comm, TASK_COMM_LEN);
190 __entry->pid = p->pid;
191 __entry->prio = p->prio; /* XXX SCHED_DEADLINE */
192 __entry->orig_cpu = task_cpu(p);
193 __entry->dest_cpu = dest_cpu;
194 ),
195
196 TP_printk("comm=%s pid=%d prio=%d orig_cpu=%d dest_cpu=%d",
197 __entry->comm, __entry->pid, __entry->prio,
198 __entry->orig_cpu, __entry->dest_cpu)
199);
200
201DECLARE_EVENT_CLASS(sched_process_template,
202
203 TP_PROTO(struct task_struct *p),
204
205 TP_ARGS(p),
206
207 TP_STRUCT__entry(
208 __array( char, comm, TASK_COMM_LEN )
209 __field( pid_t, pid )
210 __field( int, prio )
211 ),
212
213 TP_fast_assign(
214 memcpy(__entry->comm, p->comm, TASK_COMM_LEN);
215 __entry->pid = p->pid;
216 __entry->prio = p->prio; /* XXX SCHED_DEADLINE */
217 ),
218
219 TP_printk("comm=%s pid=%d prio=%d",
220 __entry->comm, __entry->pid, __entry->prio)
221);
222
223/*
224 * Tracepoint for freeing a task:
225 */
226DEFINE_EVENT(sched_process_template, sched_process_free,
227 TP_PROTO(struct task_struct *p),
228 TP_ARGS(p));
229
230
231/*
232 * Tracepoint for a task exiting:
233 */
234DEFINE_EVENT(sched_process_template, sched_process_exit,
235 TP_PROTO(struct task_struct *p),
236 TP_ARGS(p));
237
238/*
239 * Tracepoint for waiting on task to unschedule:
240 */
241DEFINE_EVENT(sched_process_template, sched_wait_task,
242 TP_PROTO(struct task_struct *p),
243 TP_ARGS(p));
244
245/*
246 * Tracepoint for a waiting task:
247 */
248TRACE_EVENT(sched_process_wait,
249
250 TP_PROTO(struct pid *pid),
251
252 TP_ARGS(pid),
253
254 TP_STRUCT__entry(
255 __array( char, comm, TASK_COMM_LEN )
256 __field( pid_t, pid )
257 __field( int, prio )
258 ),
259
260 TP_fast_assign(
261 memcpy(__entry->comm, current->comm, TASK_COMM_LEN);
262 __entry->pid = pid_nr(pid);
263 __entry->prio = current->prio; /* XXX SCHED_DEADLINE */
264 ),
265
266 TP_printk("comm=%s pid=%d prio=%d",
267 __entry->comm, __entry->pid, __entry->prio)
268);
269
270/*
271 * Tracepoint for do_fork:
272 */
273TRACE_EVENT(sched_process_fork,
274
275 TP_PROTO(struct task_struct *parent, struct task_struct *child),
276
277 TP_ARGS(parent, child),
278
279 TP_STRUCT__entry(
280 __array( char, parent_comm, TASK_COMM_LEN )
281 __field( pid_t, parent_pid )
282 __array( char, child_comm, TASK_COMM_LEN )
283 __field( pid_t, child_pid )
284 ),
285
286 TP_fast_assign(
287 memcpy(__entry->parent_comm, parent->comm, TASK_COMM_LEN);
288 __entry->parent_pid = parent->pid;
289 memcpy(__entry->child_comm, child->comm, TASK_COMM_LEN);
290 __entry->child_pid = child->pid;
291 ),
292
293 TP_printk("comm=%s pid=%d child_comm=%s child_pid=%d",
294 __entry->parent_comm, __entry->parent_pid,
295 __entry->child_comm, __entry->child_pid)
296);
297
298/*
299 * Tracepoint for exec:
300 */
301TRACE_EVENT(sched_process_exec,
302
303 TP_PROTO(struct task_struct *p, pid_t old_pid,
304 struct linux_binprm *bprm),
305
306 TP_ARGS(p, old_pid, bprm),
307
308 TP_STRUCT__entry(
309 __string( filename, bprm->filename )
310 __field( pid_t, pid )
311 __field( pid_t, old_pid )
312 ),
313
314 TP_fast_assign(
315 __assign_str(filename, bprm->filename);
316 __entry->pid = p->pid;
317 __entry->old_pid = old_pid;
318 ),
319
320 TP_printk("filename=%s pid=%d old_pid=%d", __get_str(filename),
321 __entry->pid, __entry->old_pid)
322);
323
324/*
325 * XXX the below sched_stat tracepoints only apply to SCHED_OTHER/BATCH/IDLE
326 * adding sched_stat support to SCHED_FIFO/RR would be welcome.
327 */
328DECLARE_EVENT_CLASS(sched_stat_template,
329
330 TP_PROTO(struct task_struct *tsk, u64 delay),
331
332 TP_ARGS(__perf_task(tsk), __perf_count(delay)),
333
334 TP_STRUCT__entry(
335 __array( char, comm, TASK_COMM_LEN )
336 __field( pid_t, pid )
337 __field( u64, delay )
338 ),
339
340 TP_fast_assign(
341 memcpy(__entry->comm, tsk->comm, TASK_COMM_LEN);
342 __entry->pid = tsk->pid;
343 __entry->delay = delay;
344 ),
345
346 TP_printk("comm=%s pid=%d delay=%Lu [ns]",
347 __entry->comm, __entry->pid,
348 (unsigned long long)__entry->delay)
349);
350
351
352/*
353 * Tracepoint for accounting wait time (time the task is runnable
354 * but not actually running due to scheduler contention).
355 */
356DEFINE_EVENT(sched_stat_template, sched_stat_wait,
357 TP_PROTO(struct task_struct *tsk, u64 delay),
358 TP_ARGS(tsk, delay));
359
360/*
361 * Tracepoint for accounting sleep time (time the task is not runnable,
362 * including iowait, see below).
363 */
364DEFINE_EVENT(sched_stat_template, sched_stat_sleep,
365 TP_PROTO(struct task_struct *tsk, u64 delay),
366 TP_ARGS(tsk, delay));
367
368/*
369 * Tracepoint for accounting iowait time (time the task is not runnable
370 * due to waiting on IO to complete).
371 */
372DEFINE_EVENT(sched_stat_template, sched_stat_iowait,
373 TP_PROTO(struct task_struct *tsk, u64 delay),
374 TP_ARGS(tsk, delay));
375
376/*
377 * Tracepoint for accounting blocked time (time the task is in uninterruptible).
378 */
379DEFINE_EVENT(sched_stat_template, sched_stat_blocked,
380 TP_PROTO(struct task_struct *tsk, u64 delay),
381 TP_ARGS(tsk, delay));
382
383/*
384 * Tracepoint for accounting runtime (time the task is executing
385 * on a CPU).
386 */
387DECLARE_EVENT_CLASS(sched_stat_runtime,
388
389 TP_PROTO(struct task_struct *tsk, u64 runtime, u64 vruntime),
390
391 TP_ARGS(tsk, __perf_count(runtime), vruntime),
392
393 TP_STRUCT__entry(
394 __array( char, comm, TASK_COMM_LEN )
395 __field( pid_t, pid )
396 __field( u64, runtime )
397 __field( u64, vruntime )
398 ),
399
400 TP_fast_assign(
401 memcpy(__entry->comm, tsk->comm, TASK_COMM_LEN);
402 __entry->pid = tsk->pid;
403 __entry->runtime = runtime;
404 __entry->vruntime = vruntime;
405 ),
406
407 TP_printk("comm=%s pid=%d runtime=%Lu [ns] vruntime=%Lu [ns]",
408 __entry->comm, __entry->pid,
409 (unsigned long long)__entry->runtime,
410 (unsigned long long)__entry->vruntime)
411);
412
413DEFINE_EVENT(sched_stat_runtime, sched_stat_runtime,
414 TP_PROTO(struct task_struct *tsk, u64 runtime, u64 vruntime),
415 TP_ARGS(tsk, runtime, vruntime));
416
417/*
418 * Tracepoint for showing priority inheritance modifying a tasks
419 * priority.
420 */
421TRACE_EVENT(sched_pi_setprio,
422
423 TP_PROTO(struct task_struct *tsk, struct task_struct *pi_task),
424
425 TP_ARGS(tsk, pi_task),
426
427 TP_STRUCT__entry(
428 __array( char, comm, TASK_COMM_LEN )
429 __field( pid_t, pid )
430 __field( int, oldprio )
431 __field( int, newprio )
432 ),
433
434 TP_fast_assign(
435 memcpy(__entry->comm, tsk->comm, TASK_COMM_LEN);
436 __entry->pid = tsk->pid;
437 __entry->oldprio = tsk->prio;
438 __entry->newprio = pi_task ?
439 min(tsk->normal_prio, pi_task->prio) :
440 tsk->normal_prio;
441 /* XXX SCHED_DEADLINE bits missing */
442 ),
443
444 TP_printk("comm=%s pid=%d oldprio=%d newprio=%d",
445 __entry->comm, __entry->pid,
446 __entry->oldprio, __entry->newprio)
447);
448
449#ifdef CONFIG_DETECT_HUNG_TASK
450TRACE_EVENT(sched_process_hang,
451 TP_PROTO(struct task_struct *tsk),
452 TP_ARGS(tsk),
453
454 TP_STRUCT__entry(
455 __array( char, comm, TASK_COMM_LEN )
456 __field( pid_t, pid )
457 ),
458
459 TP_fast_assign(
460 memcpy(__entry->comm, tsk->comm, TASK_COMM_LEN);
461 __entry->pid = tsk->pid;
462 ),
463
464 TP_printk("comm=%s pid=%d", __entry->comm, __entry->pid)
465);
466#endif /* CONFIG_DETECT_HUNG_TASK */
467
468DECLARE_EVENT_CLASS(sched_move_task_template,
469
470 TP_PROTO(struct task_struct *tsk, int src_cpu, int dst_cpu),
471
472 TP_ARGS(tsk, src_cpu, dst_cpu),
473
474 TP_STRUCT__entry(
475 __field( pid_t, pid )
476 __field( pid_t, tgid )
477 __field( pid_t, ngid )
478 __field( int, src_cpu )
479 __field( int, src_nid )
480 __field( int, dst_cpu )
481 __field( int, dst_nid )
482 ),
483
484 TP_fast_assign(
485 __entry->pid = task_pid_nr(tsk);
486 __entry->tgid = task_tgid_nr(tsk);
487 __entry->ngid = task_numa_group_id(tsk);
488 __entry->src_cpu = src_cpu;
489 __entry->src_nid = cpu_to_node(src_cpu);
490 __entry->dst_cpu = dst_cpu;
491 __entry->dst_nid = cpu_to_node(dst_cpu);
492 ),
493
494 TP_printk("pid=%d tgid=%d ngid=%d src_cpu=%d src_nid=%d dst_cpu=%d dst_nid=%d",
495 __entry->pid, __entry->tgid, __entry->ngid,
496 __entry->src_cpu, __entry->src_nid,
497 __entry->dst_cpu, __entry->dst_nid)
498);
499
500/*
501 * Tracks migration of tasks from one runqueue to another. Can be used to
502 * detect if automatic NUMA balancing is bouncing between nodes
503 */
504DEFINE_EVENT(sched_move_task_template, sched_move_numa,
505 TP_PROTO(struct task_struct *tsk, int src_cpu, int dst_cpu),
506
507 TP_ARGS(tsk, src_cpu, dst_cpu)
508);
509
510DEFINE_EVENT(sched_move_task_template, sched_stick_numa,
511 TP_PROTO(struct task_struct *tsk, int src_cpu, int dst_cpu),
512
513 TP_ARGS(tsk, src_cpu, dst_cpu)
514);
515
516TRACE_EVENT(sched_swap_numa,
517
518 TP_PROTO(struct task_struct *src_tsk, int src_cpu,
519 struct task_struct *dst_tsk, int dst_cpu),
520
521 TP_ARGS(src_tsk, src_cpu, dst_tsk, dst_cpu),
522
523 TP_STRUCT__entry(
524 __field( pid_t, src_pid )
525 __field( pid_t, src_tgid )
526 __field( pid_t, src_ngid )
527 __field( int, src_cpu )
528 __field( int, src_nid )
529 __field( pid_t, dst_pid )
530 __field( pid_t, dst_tgid )
531 __field( pid_t, dst_ngid )
532 __field( int, dst_cpu )
533 __field( int, dst_nid )
534 ),
535
536 TP_fast_assign(
537 __entry->src_pid = task_pid_nr(src_tsk);
538 __entry->src_tgid = task_tgid_nr(src_tsk);
539 __entry->src_ngid = task_numa_group_id(src_tsk);
540 __entry->src_cpu = src_cpu;
541 __entry->src_nid = cpu_to_node(src_cpu);
542 __entry->dst_pid = task_pid_nr(dst_tsk);
543 __entry->dst_tgid = task_tgid_nr(dst_tsk);
544 __entry->dst_ngid = task_numa_group_id(dst_tsk);
545 __entry->dst_cpu = dst_cpu;
546 __entry->dst_nid = cpu_to_node(dst_cpu);
547 ),
548
549 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",
550 __entry->src_pid, __entry->src_tgid, __entry->src_ngid,
551 __entry->src_cpu, __entry->src_nid,
552 __entry->dst_pid, __entry->dst_tgid, __entry->dst_ngid,
553 __entry->dst_cpu, __entry->dst_nid)
554);
555
556/*
557 * Tracepoint for waking a polling cpu without an IPI.
558 */
559TRACE_EVENT(sched_wake_idle_without_ipi,
560
561 TP_PROTO(int cpu),
562
563 TP_ARGS(cpu),
564
565 TP_STRUCT__entry(
566 __field( int, cpu )
567 ),
568
569 TP_fast_assign(
570 __entry->cpu = cpu;
571 ),
572
573 TP_printk("cpu=%d", __entry->cpu)
574);
575#endif /* _TRACE_SCHED_H */
576
577/* This part must be outside protection */
578#include <trace/define_trace.h>
579