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/*
247 * Tracepoint for a task exiting:
248 */
249DEFINE_EVENT(sched_process_template, sched_process_exit,
250 TP_PROTO(struct task_struct *p),
251 TP_ARGS(p));
252
253/*
254 * Tracepoint for waiting on task to unschedule:
255 */
256DEFINE_EVENT(sched_process_template, sched_wait_task,
257 TP_PROTO(struct task_struct *p),
258 TP_ARGS(p));
259
260/*
261 * Tracepoint for a waiting task:
262 */
263TRACE_EVENT(sched_process_wait,
264
265 TP_PROTO(struct pid *pid),
266
267 TP_ARGS(pid),
268
269 TP_STRUCT__entry(
270 __array( char, comm, TASK_COMM_LEN )
271 __field( pid_t, pid )
272 __field( int, prio )
273 ),
274
275 TP_fast_assign(
276 memcpy(__entry->comm, current->comm, TASK_COMM_LEN);
277 __entry->pid = pid_nr(pid);
278 __entry->prio = current->prio; /* XXX SCHED_DEADLINE */
279 ),
280
281 TP_printk("comm=%s pid=%d prio=%d",
282 __entry->comm, __entry->pid, __entry->prio)
283);
284
285/*
286 * Tracepoint for do_fork:
287 */
288TRACE_EVENT(sched_process_fork,
289
290 TP_PROTO(struct task_struct *parent, struct task_struct *child),
291
292 TP_ARGS(parent, child),
293
294 TP_STRUCT__entry(
295 __array( char, parent_comm, TASK_COMM_LEN )
296 __field( pid_t, parent_pid )
297 __array( char, child_comm, TASK_COMM_LEN )
298 __field( pid_t, child_pid )
299 ),
300
301 TP_fast_assign(
302 memcpy(__entry->parent_comm, parent->comm, TASK_COMM_LEN);
303 __entry->parent_pid = parent->pid;
304 memcpy(__entry->child_comm, child->comm, TASK_COMM_LEN);
305 __entry->child_pid = child->pid;
306 ),
307
308 TP_printk("comm=%s pid=%d child_comm=%s child_pid=%d",
309 __entry->parent_comm, __entry->parent_pid,
310 __entry->child_comm, __entry->child_pid)
311);
312
313/*
314 * Tracepoint for exec:
315 */
316TRACE_EVENT(sched_process_exec,
317
318 TP_PROTO(struct task_struct *p, pid_t old_pid,
319 struct linux_binprm *bprm),
320
321 TP_ARGS(p, old_pid, bprm),
322
323 TP_STRUCT__entry(
324 __string( filename, bprm->filename )
325 __field( pid_t, pid )
326 __field( pid_t, old_pid )
327 ),
328
329 TP_fast_assign(
330 __assign_str(filename, bprm->filename);
331 __entry->pid = p->pid;
332 __entry->old_pid = old_pid;
333 ),
334
335 TP_printk("filename=%s pid=%d old_pid=%d", __get_str(filename),
336 __entry->pid, __entry->old_pid)
337);
338
339/*
340 * XXX the below sched_stat tracepoints only apply to SCHED_OTHER/BATCH/IDLE
341 * adding sched_stat support to SCHED_FIFO/RR would be welcome.
342 */
343DECLARE_EVENT_CLASS(sched_stat_template,
344
345 TP_PROTO(struct task_struct *tsk, u64 delay),
346
347 TP_ARGS(__perf_task(tsk), __perf_count(delay)),
348
349 TP_STRUCT__entry(
350 __array( char, comm, TASK_COMM_LEN )
351 __field( pid_t, pid )
352 __field( u64, delay )
353 ),
354
355 TP_fast_assign(
356 memcpy(__entry->comm, tsk->comm, TASK_COMM_LEN);
357 __entry->pid = tsk->pid;
358 __entry->delay = delay;
359 ),
360
361 TP_printk("comm=%s pid=%d delay=%Lu [ns]",
362 __entry->comm, __entry->pid,
363 (unsigned long long)__entry->delay)
364);
365
366
367/*
368 * Tracepoint for accounting wait time (time the task is runnable
369 * but not actually running due to scheduler contention).
370 */
371DEFINE_EVENT(sched_stat_template, sched_stat_wait,
372 TP_PROTO(struct task_struct *tsk, u64 delay),
373 TP_ARGS(tsk, delay));
374
375/*
376 * Tracepoint for accounting sleep time (time the task is not runnable,
377 * including iowait, see below).
378 */
379DEFINE_EVENT(sched_stat_template, sched_stat_sleep,
380 TP_PROTO(struct task_struct *tsk, u64 delay),
381 TP_ARGS(tsk, delay));
382
383/*
384 * Tracepoint for accounting iowait time (time the task is not runnable
385 * due to waiting on IO to complete).
386 */
387DEFINE_EVENT(sched_stat_template, sched_stat_iowait,
388 TP_PROTO(struct task_struct *tsk, u64 delay),
389 TP_ARGS(tsk, delay));
390
391/*
392 * Tracepoint for accounting blocked time (time the task is in uninterruptible).
393 */
394DEFINE_EVENT(sched_stat_template, sched_stat_blocked,
395 TP_PROTO(struct task_struct *tsk, u64 delay),
396 TP_ARGS(tsk, delay));
397
398/*
399 * Tracepoint for accounting runtime (time the task is executing
400 * on a CPU).
401 */
402DECLARE_EVENT_CLASS(sched_stat_runtime,
403
404 TP_PROTO(struct task_struct *tsk, u64 runtime, u64 vruntime),
405
406 TP_ARGS(tsk, __perf_count(runtime), vruntime),
407
408 TP_STRUCT__entry(
409 __array( char, comm, TASK_COMM_LEN )
410 __field( pid_t, pid )
411 __field( u64, runtime )
412 __field( u64, vruntime )
413 ),
414
415 TP_fast_assign(
416 memcpy(__entry->comm, tsk->comm, TASK_COMM_LEN);
417 __entry->pid = tsk->pid;
418 __entry->runtime = runtime;
419 __entry->vruntime = vruntime;
420 ),
421
422 TP_printk("comm=%s pid=%d runtime=%Lu [ns] vruntime=%Lu [ns]",
423 __entry->comm, __entry->pid,
424 (unsigned long long)__entry->runtime,
425 (unsigned long long)__entry->vruntime)
426);
427
428DEFINE_EVENT(sched_stat_runtime, sched_stat_runtime,
429 TP_PROTO(struct task_struct *tsk, u64 runtime, u64 vruntime),
430 TP_ARGS(tsk, runtime, vruntime));
431
432/*
433 * Tracepoint for showing priority inheritance modifying a tasks
434 * priority.
435 */
436TRACE_EVENT(sched_pi_setprio,
437
438 TP_PROTO(struct task_struct *tsk, struct task_struct *pi_task),
439
440 TP_ARGS(tsk, pi_task),
441
442 TP_STRUCT__entry(
443 __array( char, comm, TASK_COMM_LEN )
444 __field( pid_t, pid )
445 __field( int, oldprio )
446 __field( int, newprio )
447 ),
448
449 TP_fast_assign(
450 memcpy(__entry->comm, tsk->comm, TASK_COMM_LEN);
451 __entry->pid = tsk->pid;
452 __entry->oldprio = tsk->prio;
453 __entry->newprio = pi_task ?
454 min(tsk->normal_prio, pi_task->prio) :
455 tsk->normal_prio;
456 /* XXX SCHED_DEADLINE bits missing */
457 ),
458
459 TP_printk("comm=%s pid=%d oldprio=%d newprio=%d",
460 __entry->comm, __entry->pid,
461 __entry->oldprio, __entry->newprio)
462);
463
464#ifdef CONFIG_DETECT_HUNG_TASK
465TRACE_EVENT(sched_process_hang,
466 TP_PROTO(struct task_struct *tsk),
467 TP_ARGS(tsk),
468
469 TP_STRUCT__entry(
470 __array( char, comm, TASK_COMM_LEN )
471 __field( pid_t, pid )
472 ),
473
474 TP_fast_assign(
475 memcpy(__entry->comm, tsk->comm, TASK_COMM_LEN);
476 __entry->pid = tsk->pid;
477 ),
478
479 TP_printk("comm=%s pid=%d", __entry->comm, __entry->pid)
480);
481#endif /* CONFIG_DETECT_HUNG_TASK */
482
483DECLARE_EVENT_CLASS(sched_move_task_template,
484
485 TP_PROTO(struct task_struct *tsk, int src_cpu, int dst_cpu),
486
487 TP_ARGS(tsk, src_cpu, dst_cpu),
488
489 TP_STRUCT__entry(
490 __field( pid_t, pid )
491 __field( pid_t, tgid )
492 __field( pid_t, ngid )
493 __field( int, src_cpu )
494 __field( int, src_nid )
495 __field( int, dst_cpu )
496 __field( int, dst_nid )
497 ),
498
499 TP_fast_assign(
500 __entry->pid = task_pid_nr(tsk);
501 __entry->tgid = task_tgid_nr(tsk);
502 __entry->ngid = task_numa_group_id(tsk);
503 __entry->src_cpu = src_cpu;
504 __entry->src_nid = cpu_to_node(src_cpu);
505 __entry->dst_cpu = dst_cpu;
506 __entry->dst_nid = cpu_to_node(dst_cpu);
507 ),
508
509 TP_printk("pid=%d tgid=%d ngid=%d src_cpu=%d src_nid=%d dst_cpu=%d dst_nid=%d",
510 __entry->pid, __entry->tgid, __entry->ngid,
511 __entry->src_cpu, __entry->src_nid,
512 __entry->dst_cpu, __entry->dst_nid)
513);
514
515/*
516 * Tracks migration of tasks from one runqueue to another. Can be used to
517 * detect if automatic NUMA balancing is bouncing between nodes
518 */
519DEFINE_EVENT(sched_move_task_template, sched_move_numa,
520 TP_PROTO(struct task_struct *tsk, int src_cpu, int dst_cpu),
521
522 TP_ARGS(tsk, src_cpu, dst_cpu)
523);
524
525DEFINE_EVENT(sched_move_task_template, sched_stick_numa,
526 TP_PROTO(struct task_struct *tsk, int src_cpu, int dst_cpu),
527
528 TP_ARGS(tsk, src_cpu, dst_cpu)
529);
530
531TRACE_EVENT(sched_swap_numa,
532
533 TP_PROTO(struct task_struct *src_tsk, int src_cpu,
534 struct task_struct *dst_tsk, int dst_cpu),
535
536 TP_ARGS(src_tsk, src_cpu, dst_tsk, dst_cpu),
537
538 TP_STRUCT__entry(
539 __field( pid_t, src_pid )
540 __field( pid_t, src_tgid )
541 __field( pid_t, src_ngid )
542 __field( int, src_cpu )
543 __field( int, src_nid )
544 __field( pid_t, dst_pid )
545 __field( pid_t, dst_tgid )
546 __field( pid_t, dst_ngid )
547 __field( int, dst_cpu )
548 __field( int, dst_nid )
549 ),
550
551 TP_fast_assign(
552 __entry->src_pid = task_pid_nr(src_tsk);
553 __entry->src_tgid = task_tgid_nr(src_tsk);
554 __entry->src_ngid = task_numa_group_id(src_tsk);
555 __entry->src_cpu = src_cpu;
556 __entry->src_nid = cpu_to_node(src_cpu);
557 __entry->dst_pid = task_pid_nr(dst_tsk);
558 __entry->dst_tgid = task_tgid_nr(dst_tsk);
559 __entry->dst_ngid = task_numa_group_id(dst_tsk);
560 __entry->dst_cpu = dst_cpu;
561 __entry->dst_nid = cpu_to_node(dst_cpu);
562 ),
563
564 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",
565 __entry->src_pid, __entry->src_tgid, __entry->src_ngid,
566 __entry->src_cpu, __entry->src_nid,
567 __entry->dst_pid, __entry->dst_tgid, __entry->dst_ngid,
568 __entry->dst_cpu, __entry->dst_nid)
569);
570
571/*
572 * Tracepoint for waking a polling cpu without an IPI.
573 */
574TRACE_EVENT(sched_wake_idle_without_ipi,
575
576 TP_PROTO(int cpu),
577
578 TP_ARGS(cpu),
579
580 TP_STRUCT__entry(
581 __field( int, cpu )
582 ),
583
584 TP_fast_assign(
585 __entry->cpu = cpu;
586 ),
587
588 TP_printk("cpu=%d", __entry->cpu)
589);
590#endif /* _TRACE_SCHED_H */
591
592/* This part must be outside protection */
593#include <trace/define_trace.h>
594