1 /* SPDX-License-Identifier: GPL-2.0 */
3 #define TRACE_SYSTEM sched
5 #if !defined(_TRACE_SCHED_H) || defined(TRACE_HEADER_MULTI_READ)
8 #include <linux/kthread.h>
9 #include <linux/sched/numa_balancing.h>
10 #include <linux/tracepoint.h>
11 #include <linux/binfmts.h>
14 * Tracepoint for calling kthread_stop, performed to end a kthread:
16 TRACE_EVENT(sched_kthread_stop,
18 TP_PROTO(struct task_struct *t),
23 __array( char, comm, TASK_COMM_LEN )
28 memcpy(__entry->comm, t->comm, TASK_COMM_LEN);
29 __entry->pid = t->pid;
32 TP_printk("comm=%s pid=%d", __entry->comm, __entry->pid)
36 * Tracepoint for the return value of the kthread stopping:
38 TRACE_EVENT(sched_kthread_stop_ret,
52 TP_printk("ret=%d", __entry->ret)
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
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
64 TRACE_EVENT(sched_kthread_work_queue_work,
66 TP_PROTO(struct kthread_worker *worker,
67 struct kthread_work *work),
69 TP_ARGS(worker, work),
72 __field( void *, work )
73 __field( void *, function)
74 __field( void *, worker)
79 __entry->function = work->func;
80 __entry->worker = worker;
83 TP_printk("work struct=%p function=%ps worker=%p",
84 __entry->work, __entry->function, __entry->worker)
88 * sched_kthread_work_execute_start - called immediately before the work callback
89 * @work: pointer to struct kthread_work
91 * Allows to track kthread work execution.
93 TRACE_EVENT(sched_kthread_work_execute_start,
95 TP_PROTO(struct kthread_work *work),
100 __field( void *, work )
101 __field( void *, function)
105 __entry->work = work;
106 __entry->function = work->func;
109 TP_printk("work struct %p: function %ps", __entry->work, __entry->function)
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
117 * Allows to track workqueue execution.
119 TRACE_EVENT(sched_kthread_work_execute_end,
121 TP_PROTO(struct kthread_work *work, kthread_work_func_t function),
123 TP_ARGS(work, function),
126 __field( void *, work )
127 __field( void *, function)
131 __entry->work = work;
132 __entry->function = function;
135 TP_printk("work struct %p: function %ps", __entry->work, __entry->function)
139 * Tracepoint for waking up a task:
141 DECLARE_EVENT_CLASS(sched_wakeup_template,
143 TP_PROTO(struct task_struct *p),
145 TP_ARGS(__perf_task(p)),
148 __array( char, comm, TASK_COMM_LEN )
149 __field( pid_t, pid )
151 __field( int, target_cpu )
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);
161 TP_printk("comm=%s pid=%d prio=%d target_cpu=%03d",
162 __entry->comm, __entry->pid, __entry->prio,
167 * Tracepoint called when waking a task; this tracepoint is guaranteed to be
168 * called from the waking context.
170 DEFINE_EVENT(sched_wakeup_template, sched_waking,
171 TP_PROTO(struct task_struct *p),
175 * Tracepoint called when the task is actually woken; p->state == TASK_RUNNING.
176 * It is not always called from the waking context.
178 DEFINE_EVENT(sched_wakeup_template, sched_wakeup,
179 TP_PROTO(struct task_struct *p),
183 * Tracepoint for waking up a new task:
185 DEFINE_EVENT(sched_wakeup_template, sched_wakeup_new,
186 TP_PROTO(struct task_struct *p),
189 #ifdef CREATE_TRACE_POINTS
190 static inline long __trace_sched_switch_state(bool preempt, struct task_struct *p)
194 #ifdef CONFIG_SCHED_DEBUG
195 BUG_ON(p != current);
196 #endif /* CONFIG_SCHED_DEBUG */
199 * Preemption ignores task state, therefore preempted tasks are always
200 * RUNNING (we will not have dequeued if state != RUNNING).
203 return TASK_REPORT_MAX;
206 * task_state_index() uses fls() and returns a value from 0-8 range.
207 * Decrement it by 1 (except TASK_RUNNING state i.e 0) before using
208 * it for left shift operation to get the correct task->state
211 state = task_state_index(p);
213 return state ? (1 << (state - 1)) : state;
215 #endif /* CREATE_TRACE_POINTS */
218 * Tracepoint for task switches, performed by the scheduler:
220 TRACE_EVENT(sched_switch,
222 TP_PROTO(bool preempt,
223 struct task_struct *prev,
224 struct task_struct *next),
226 TP_ARGS(preempt, prev, next),
229 __array( char, prev_comm, TASK_COMM_LEN )
230 __field( pid_t, prev_pid )
231 __field( int, prev_prio )
232 __field( long, prev_state )
233 __array( char, next_comm, TASK_COMM_LEN )
234 __field( pid_t, next_pid )
235 __field( int, next_prio )
239 memcpy(__entry->next_comm, next->comm, TASK_COMM_LEN);
240 __entry->prev_pid = prev->pid;
241 __entry->prev_prio = prev->prio;
242 __entry->prev_state = __trace_sched_switch_state(preempt, prev);
243 memcpy(__entry->prev_comm, prev->comm, TASK_COMM_LEN);
244 __entry->next_pid = next->pid;
245 __entry->next_prio = next->prio;
246 /* XXX SCHED_DEADLINE */
249 TP_printk("prev_comm=%s prev_pid=%d prev_prio=%d prev_state=%s%s ==> next_comm=%s next_pid=%d next_prio=%d",
250 __entry->prev_comm, __entry->prev_pid, __entry->prev_prio,
252 (__entry->prev_state & (TASK_REPORT_MAX - 1)) ?
253 __print_flags(__entry->prev_state & (TASK_REPORT_MAX - 1), "|",
254 { TASK_INTERRUPTIBLE, "S" },
255 { TASK_UNINTERRUPTIBLE, "D" },
256 { __TASK_STOPPED, "T" },
257 { __TASK_TRACED, "t" },
259 { EXIT_ZOMBIE, "Z" },
260 { TASK_PARKED, "P" },
261 { TASK_DEAD, "I" }) :
264 __entry->prev_state & TASK_REPORT_MAX ? "+" : "",
265 __entry->next_comm, __entry->next_pid, __entry->next_prio)
269 * Tracepoint for a task being migrated:
271 TRACE_EVENT(sched_migrate_task,
273 TP_PROTO(struct task_struct *p, int dest_cpu),
275 TP_ARGS(p, dest_cpu),
278 __array( char, comm, TASK_COMM_LEN )
279 __field( pid_t, pid )
281 __field( int, orig_cpu )
282 __field( int, dest_cpu )
286 memcpy(__entry->comm, p->comm, TASK_COMM_LEN);
287 __entry->pid = p->pid;
288 __entry->prio = p->prio; /* XXX SCHED_DEADLINE */
289 __entry->orig_cpu = task_cpu(p);
290 __entry->dest_cpu = dest_cpu;
293 TP_printk("comm=%s pid=%d prio=%d orig_cpu=%d dest_cpu=%d",
294 __entry->comm, __entry->pid, __entry->prio,
295 __entry->orig_cpu, __entry->dest_cpu)
298 DECLARE_EVENT_CLASS(sched_process_template,
300 TP_PROTO(struct task_struct *p),
305 __array( char, comm, TASK_COMM_LEN )
306 __field( pid_t, pid )
311 memcpy(__entry->comm, p->comm, TASK_COMM_LEN);
312 __entry->pid = p->pid;
313 __entry->prio = p->prio; /* XXX SCHED_DEADLINE */
316 TP_printk("comm=%s pid=%d prio=%d",
317 __entry->comm, __entry->pid, __entry->prio)
321 * Tracepoint for freeing a task:
323 DEFINE_EVENT(sched_process_template, sched_process_free,
324 TP_PROTO(struct task_struct *p),
328 * Tracepoint for a task exiting:
330 DEFINE_EVENT(sched_process_template, sched_process_exit,
331 TP_PROTO(struct task_struct *p),
335 * Tracepoint for waiting on task to unschedule:
337 DEFINE_EVENT(sched_process_template, sched_wait_task,
338 TP_PROTO(struct task_struct *p),
342 * Tracepoint for a waiting task:
344 TRACE_EVENT(sched_process_wait,
346 TP_PROTO(struct pid *pid),
351 __array( char, comm, TASK_COMM_LEN )
352 __field( pid_t, pid )
357 memcpy(__entry->comm, current->comm, TASK_COMM_LEN);
358 __entry->pid = pid_nr(pid);
359 __entry->prio = current->prio; /* XXX SCHED_DEADLINE */
362 TP_printk("comm=%s pid=%d prio=%d",
363 __entry->comm, __entry->pid, __entry->prio)
367 * Tracepoint for kernel_clone:
369 TRACE_EVENT(sched_process_fork,
371 TP_PROTO(struct task_struct *parent, struct task_struct *child),
373 TP_ARGS(parent, child),
376 __array( char, parent_comm, TASK_COMM_LEN )
377 __field( pid_t, parent_pid )
378 __array( char, child_comm, TASK_COMM_LEN )
379 __field( pid_t, child_pid )
383 memcpy(__entry->parent_comm, parent->comm, TASK_COMM_LEN);
384 __entry->parent_pid = parent->pid;
385 memcpy(__entry->child_comm, child->comm, TASK_COMM_LEN);
386 __entry->child_pid = child->pid;
389 TP_printk("comm=%s pid=%d child_comm=%s child_pid=%d",
390 __entry->parent_comm, __entry->parent_pid,
391 __entry->child_comm, __entry->child_pid)
395 * Tracepoint for exec:
397 TRACE_EVENT(sched_process_exec,
399 TP_PROTO(struct task_struct *p, pid_t old_pid,
400 struct linux_binprm *bprm),
402 TP_ARGS(p, old_pid, bprm),
405 __string( filename, bprm->filename )
406 __field( pid_t, pid )
407 __field( pid_t, old_pid )
411 __assign_str(filename, bprm->filename);
412 __entry->pid = p->pid;
413 __entry->old_pid = old_pid;
416 TP_printk("filename=%s pid=%d old_pid=%d", __get_str(filename),
417 __entry->pid, __entry->old_pid)
421 #ifdef CONFIG_SCHEDSTATS
422 #define DEFINE_EVENT_SCHEDSTAT DEFINE_EVENT
423 #define DECLARE_EVENT_CLASS_SCHEDSTAT DECLARE_EVENT_CLASS
425 #define DEFINE_EVENT_SCHEDSTAT DEFINE_EVENT_NOP
426 #define DECLARE_EVENT_CLASS_SCHEDSTAT DECLARE_EVENT_CLASS_NOP
430 * XXX the below sched_stat tracepoints only apply to SCHED_OTHER/BATCH/IDLE
431 * adding sched_stat support to SCHED_FIFO/RR would be welcome.
433 DECLARE_EVENT_CLASS_SCHEDSTAT(sched_stat_template,
435 TP_PROTO(struct task_struct *tsk, u64 delay),
437 TP_ARGS(__perf_task(tsk), __perf_count(delay)),
440 __array( char, comm, TASK_COMM_LEN )
441 __field( pid_t, pid )
442 __field( u64, delay )
446 memcpy(__entry->comm, tsk->comm, TASK_COMM_LEN);
447 __entry->pid = tsk->pid;
448 __entry->delay = delay;
451 TP_printk("comm=%s pid=%d delay=%Lu [ns]",
452 __entry->comm, __entry->pid,
453 (unsigned long long)__entry->delay)
457 * Tracepoint for accounting wait time (time the task is runnable
458 * but not actually running due to scheduler contention).
460 DEFINE_EVENT_SCHEDSTAT(sched_stat_template, sched_stat_wait,
461 TP_PROTO(struct task_struct *tsk, u64 delay),
462 TP_ARGS(tsk, delay));
465 * Tracepoint for accounting sleep time (time the task is not runnable,
466 * including iowait, see below).
468 DEFINE_EVENT_SCHEDSTAT(sched_stat_template, sched_stat_sleep,
469 TP_PROTO(struct task_struct *tsk, u64 delay),
470 TP_ARGS(tsk, delay));
473 * Tracepoint for accounting iowait time (time the task is not runnable
474 * due to waiting on IO to complete).
476 DEFINE_EVENT_SCHEDSTAT(sched_stat_template, sched_stat_iowait,
477 TP_PROTO(struct task_struct *tsk, u64 delay),
478 TP_ARGS(tsk, delay));
481 * Tracepoint for accounting blocked time (time the task is in uninterruptible).
483 DEFINE_EVENT_SCHEDSTAT(sched_stat_template, sched_stat_blocked,
484 TP_PROTO(struct task_struct *tsk, u64 delay),
485 TP_ARGS(tsk, delay));
488 * Tracepoint for accounting runtime (time the task is executing
491 DECLARE_EVENT_CLASS(sched_stat_runtime,
493 TP_PROTO(struct task_struct *tsk, u64 runtime, u64 vruntime),
495 TP_ARGS(tsk, __perf_count(runtime), vruntime),
498 __array( char, comm, TASK_COMM_LEN )
499 __field( pid_t, pid )
500 __field( u64, runtime )
501 __field( u64, vruntime )
505 memcpy(__entry->comm, tsk->comm, TASK_COMM_LEN);
506 __entry->pid = tsk->pid;
507 __entry->runtime = runtime;
508 __entry->vruntime = vruntime;
511 TP_printk("comm=%s pid=%d runtime=%Lu [ns] vruntime=%Lu [ns]",
512 __entry->comm, __entry->pid,
513 (unsigned long long)__entry->runtime,
514 (unsigned long long)__entry->vruntime)
517 DEFINE_EVENT(sched_stat_runtime, sched_stat_runtime,
518 TP_PROTO(struct task_struct *tsk, u64 runtime, u64 vruntime),
519 TP_ARGS(tsk, runtime, vruntime));
522 * Tracepoint for showing priority inheritance modifying a tasks
525 TRACE_EVENT(sched_pi_setprio,
527 TP_PROTO(struct task_struct *tsk, struct task_struct *pi_task),
529 TP_ARGS(tsk, pi_task),
532 __array( char, comm, TASK_COMM_LEN )
533 __field( pid_t, pid )
534 __field( int, oldprio )
535 __field( int, newprio )
539 memcpy(__entry->comm, tsk->comm, TASK_COMM_LEN);
540 __entry->pid = tsk->pid;
541 __entry->oldprio = tsk->prio;
542 __entry->newprio = pi_task ?
543 min(tsk->normal_prio, pi_task->prio) :
545 /* XXX SCHED_DEADLINE bits missing */
548 TP_printk("comm=%s pid=%d oldprio=%d newprio=%d",
549 __entry->comm, __entry->pid,
550 __entry->oldprio, __entry->newprio)
553 #ifdef CONFIG_DETECT_HUNG_TASK
554 TRACE_EVENT(sched_process_hang,
555 TP_PROTO(struct task_struct *tsk),
559 __array( char, comm, TASK_COMM_LEN )
560 __field( pid_t, pid )
564 memcpy(__entry->comm, tsk->comm, TASK_COMM_LEN);
565 __entry->pid = tsk->pid;
568 TP_printk("comm=%s pid=%d", __entry->comm, __entry->pid)
570 #endif /* CONFIG_DETECT_HUNG_TASK */
573 * Tracks migration of tasks from one runqueue to another. Can be used to
574 * detect if automatic NUMA balancing is bouncing between nodes.
576 TRACE_EVENT(sched_move_numa,
578 TP_PROTO(struct task_struct *tsk, int src_cpu, int dst_cpu),
580 TP_ARGS(tsk, src_cpu, dst_cpu),
583 __field( pid_t, pid )
584 __field( pid_t, tgid )
585 __field( pid_t, ngid )
586 __field( int, src_cpu )
587 __field( int, src_nid )
588 __field( int, dst_cpu )
589 __field( int, dst_nid )
593 __entry->pid = task_pid_nr(tsk);
594 __entry->tgid = task_tgid_nr(tsk);
595 __entry->ngid = task_numa_group_id(tsk);
596 __entry->src_cpu = src_cpu;
597 __entry->src_nid = cpu_to_node(src_cpu);
598 __entry->dst_cpu = dst_cpu;
599 __entry->dst_nid = cpu_to_node(dst_cpu);
602 TP_printk("pid=%d tgid=%d ngid=%d src_cpu=%d src_nid=%d dst_cpu=%d dst_nid=%d",
603 __entry->pid, __entry->tgid, __entry->ngid,
604 __entry->src_cpu, __entry->src_nid,
605 __entry->dst_cpu, __entry->dst_nid)
608 DECLARE_EVENT_CLASS(sched_numa_pair_template,
610 TP_PROTO(struct task_struct *src_tsk, int src_cpu,
611 struct task_struct *dst_tsk, int dst_cpu),
613 TP_ARGS(src_tsk, src_cpu, dst_tsk, dst_cpu),
616 __field( pid_t, src_pid )
617 __field( pid_t, src_tgid )
618 __field( pid_t, src_ngid )
619 __field( int, src_cpu )
620 __field( int, src_nid )
621 __field( pid_t, dst_pid )
622 __field( pid_t, dst_tgid )
623 __field( pid_t, dst_ngid )
624 __field( int, dst_cpu )
625 __field( int, dst_nid )
629 __entry->src_pid = task_pid_nr(src_tsk);
630 __entry->src_tgid = task_tgid_nr(src_tsk);
631 __entry->src_ngid = task_numa_group_id(src_tsk);
632 __entry->src_cpu = src_cpu;
633 __entry->src_nid = cpu_to_node(src_cpu);
634 __entry->dst_pid = dst_tsk ? task_pid_nr(dst_tsk) : 0;
635 __entry->dst_tgid = dst_tsk ? task_tgid_nr(dst_tsk) : 0;
636 __entry->dst_ngid = dst_tsk ? task_numa_group_id(dst_tsk) : 0;
637 __entry->dst_cpu = dst_cpu;
638 __entry->dst_nid = dst_cpu >= 0 ? cpu_to_node(dst_cpu) : -1;
641 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",
642 __entry->src_pid, __entry->src_tgid, __entry->src_ngid,
643 __entry->src_cpu, __entry->src_nid,
644 __entry->dst_pid, __entry->dst_tgid, __entry->dst_ngid,
645 __entry->dst_cpu, __entry->dst_nid)
648 DEFINE_EVENT(sched_numa_pair_template, sched_stick_numa,
650 TP_PROTO(struct task_struct *src_tsk, int src_cpu,
651 struct task_struct *dst_tsk, int dst_cpu),
653 TP_ARGS(src_tsk, src_cpu, dst_tsk, dst_cpu)
656 DEFINE_EVENT(sched_numa_pair_template, sched_swap_numa,
658 TP_PROTO(struct task_struct *src_tsk, int src_cpu,
659 struct task_struct *dst_tsk, int dst_cpu),
661 TP_ARGS(src_tsk, src_cpu, dst_tsk, dst_cpu)
666 * Tracepoint for waking a polling cpu without an IPI.
668 TRACE_EVENT(sched_wake_idle_without_ipi,
682 TP_printk("cpu=%d", __entry->cpu)
686 * Following tracepoints are not exported in tracefs and provide hooking
687 * mechanisms only for testing and debugging purposes.
689 * Postfixed with _tp to make them easily identifiable in the code.
691 DECLARE_TRACE(pelt_cfs_tp,
692 TP_PROTO(struct cfs_rq *cfs_rq),
695 DECLARE_TRACE(pelt_rt_tp,
696 TP_PROTO(struct rq *rq),
699 DECLARE_TRACE(pelt_dl_tp,
700 TP_PROTO(struct rq *rq),
703 DECLARE_TRACE(pelt_thermal_tp,
704 TP_PROTO(struct rq *rq),
707 DECLARE_TRACE(pelt_irq_tp,
708 TP_PROTO(struct rq *rq),
711 DECLARE_TRACE(pelt_se_tp,
712 TP_PROTO(struct sched_entity *se),
715 DECLARE_TRACE(sched_cpu_capacity_tp,
716 TP_PROTO(struct rq *rq),
719 DECLARE_TRACE(sched_overutilized_tp,
720 TP_PROTO(struct root_domain *rd, bool overutilized),
721 TP_ARGS(rd, overutilized));
723 DECLARE_TRACE(sched_util_est_cfs_tp,
724 TP_PROTO(struct cfs_rq *cfs_rq),
727 DECLARE_TRACE(sched_util_est_se_tp,
728 TP_PROTO(struct sched_entity *se),
731 DECLARE_TRACE(sched_update_nr_running_tp,
732 TP_PROTO(struct rq *rq, int change),
733 TP_ARGS(rq, change));
735 #endif /* _TRACE_SCHED_H */
737 /* This part must be outside protection */
738 #include <trace/define_trace.h>