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, success
)
152 __field( int, target_cpu
)
156 memcpy(__entry
->comm
, p
->comm
, TASK_COMM_LEN
);
157 __entry
->pid
= p
->pid
;
158 __entry
->prio
= p
->prio
; /* XXX SCHED_DEADLINE */
159 __entry
->success
= 1; /* rudiment, kill when possible */
160 __entry
->target_cpu
= task_cpu(p
);
163 TP_printk("comm=%s pid=%d prio=%d target_cpu=%03d",
164 __entry
->comm
, __entry
->pid
, __entry
->prio
,
169 * Tracepoint called when waking a task; this tracepoint is guaranteed to be
170 * called from the waking context.
172 DEFINE_EVENT(sched_wakeup_template
, sched_waking
,
173 TP_PROTO(struct task_struct
*p
),
177 * Tracepoint called when the task is actually woken; p->state == TASK_RUNNNG.
178 * It is not always called from the waking context.
180 DEFINE_EVENT(sched_wakeup_template
, sched_wakeup
,
181 TP_PROTO(struct task_struct
*p
),
185 * Tracepoint for waking up a new task:
187 DEFINE_EVENT(sched_wakeup_template
, sched_wakeup_new
,
188 TP_PROTO(struct task_struct
*p
),
191 #ifdef CREATE_TRACE_POINTS
192 static inline long __trace_sched_switch_state(bool preempt
, struct task_struct
*p
)
196 #ifdef CONFIG_SCHED_DEBUG
197 BUG_ON(p
!= current
);
198 #endif /* CONFIG_SCHED_DEBUG */
201 * Preemption ignores task state, therefore preempted tasks are always
202 * RUNNING (we will not have dequeued if state != RUNNING).
205 return TASK_REPORT_MAX
;
208 * task_state_index() uses fls() and returns a value from 0-8 range.
209 * Decrement it by 1 (except TASK_RUNNING state i.e 0) before using
210 * it for left shift operation to get the correct task->state
213 state
= task_state_index(p
);
215 return state
? (1 << (state
- 1)) : state
;
217 #endif /* CREATE_TRACE_POINTS */
220 * Tracepoint for task switches, performed by the scheduler:
222 TRACE_EVENT(sched_switch
,
224 TP_PROTO(bool preempt
,
225 struct task_struct
*prev
,
226 struct task_struct
*next
),
228 TP_ARGS(preempt
, prev
, next
),
231 __array( char, prev_comm
, TASK_COMM_LEN
)
232 __field( pid_t
, prev_pid
)
233 __field( int, prev_prio
)
234 __field( long, prev_state
)
235 __array( char, next_comm
, TASK_COMM_LEN
)
236 __field( pid_t
, next_pid
)
237 __field( int, next_prio
)
241 memcpy(__entry
->next_comm
, next
->comm
, TASK_COMM_LEN
);
242 __entry
->prev_pid
= prev
->pid
;
243 __entry
->prev_prio
= prev
->prio
;
244 __entry
->prev_state
= __trace_sched_switch_state(preempt
, prev
);
245 memcpy(__entry
->prev_comm
, prev
->comm
, TASK_COMM_LEN
);
246 __entry
->next_pid
= next
->pid
;
247 __entry
->next_prio
= next
->prio
;
248 /* XXX SCHED_DEADLINE */
251 TP_printk("prev_comm=%s prev_pid=%d prev_prio=%d prev_state=%s%s ==> next_comm=%s next_pid=%d next_prio=%d",
252 __entry
->prev_comm
, __entry
->prev_pid
, __entry
->prev_prio
,
254 (__entry
->prev_state
& (TASK_REPORT_MAX
- 1)) ?
255 __print_flags(__entry
->prev_state
& (TASK_REPORT_MAX
- 1), "|",
256 { TASK_INTERRUPTIBLE
, "S" },
257 { TASK_UNINTERRUPTIBLE
, "D" },
258 { __TASK_STOPPED
, "T" },
259 { __TASK_TRACED
, "t" },
261 { EXIT_ZOMBIE
, "Z" },
262 { TASK_PARKED
, "P" },
263 { TASK_DEAD
, "I" }) :
266 __entry
->prev_state
& TASK_REPORT_MAX
? "+" : "",
267 __entry
->next_comm
, __entry
->next_pid
, __entry
->next_prio
)
271 * Tracepoint for a task being migrated:
273 TRACE_EVENT(sched_migrate_task
,
275 TP_PROTO(struct task_struct
*p
, int dest_cpu
),
277 TP_ARGS(p
, dest_cpu
),
280 __array( char, comm
, TASK_COMM_LEN
)
281 __field( pid_t
, pid
)
283 __field( int, orig_cpu
)
284 __field( int, dest_cpu
)
288 memcpy(__entry
->comm
, p
->comm
, TASK_COMM_LEN
);
289 __entry
->pid
= p
->pid
;
290 __entry
->prio
= p
->prio
; /* XXX SCHED_DEADLINE */
291 __entry
->orig_cpu
= task_cpu(p
);
292 __entry
->dest_cpu
= dest_cpu
;
295 TP_printk("comm=%s pid=%d prio=%d orig_cpu=%d dest_cpu=%d",
296 __entry
->comm
, __entry
->pid
, __entry
->prio
,
297 __entry
->orig_cpu
, __entry
->dest_cpu
)
300 DECLARE_EVENT_CLASS(sched_process_template
,
302 TP_PROTO(struct task_struct
*p
),
307 __array( char, comm
, TASK_COMM_LEN
)
308 __field( pid_t
, pid
)
313 memcpy(__entry
->comm
, p
->comm
, TASK_COMM_LEN
);
314 __entry
->pid
= p
->pid
;
315 __entry
->prio
= p
->prio
; /* XXX SCHED_DEADLINE */
318 TP_printk("comm=%s pid=%d prio=%d",
319 __entry
->comm
, __entry
->pid
, __entry
->prio
)
323 * Tracepoint for freeing a task:
325 DEFINE_EVENT(sched_process_template
, sched_process_free
,
326 TP_PROTO(struct task_struct
*p
),
330 * Tracepoint for a task exiting:
332 DEFINE_EVENT(sched_process_template
, sched_process_exit
,
333 TP_PROTO(struct task_struct
*p
),
337 * Tracepoint for waiting on task to unschedule:
339 DEFINE_EVENT(sched_process_template
, sched_wait_task
,
340 TP_PROTO(struct task_struct
*p
),
344 * Tracepoint for a waiting task:
346 TRACE_EVENT(sched_process_wait
,
348 TP_PROTO(struct pid
*pid
),
353 __array( char, comm
, TASK_COMM_LEN
)
354 __field( pid_t
, pid
)
359 memcpy(__entry
->comm
, current
->comm
, TASK_COMM_LEN
);
360 __entry
->pid
= pid_nr(pid
);
361 __entry
->prio
= current
->prio
; /* XXX SCHED_DEADLINE */
364 TP_printk("comm=%s pid=%d prio=%d",
365 __entry
->comm
, __entry
->pid
, __entry
->prio
)
369 * Tracepoint for do_fork:
371 TRACE_EVENT(sched_process_fork
,
373 TP_PROTO(struct task_struct
*parent
, struct task_struct
*child
),
375 TP_ARGS(parent
, child
),
378 __array( char, parent_comm
, TASK_COMM_LEN
)
379 __field( pid_t
, parent_pid
)
380 __array( char, child_comm
, TASK_COMM_LEN
)
381 __field( pid_t
, child_pid
)
385 memcpy(__entry
->parent_comm
, parent
->comm
, TASK_COMM_LEN
);
386 __entry
->parent_pid
= parent
->pid
;
387 memcpy(__entry
->child_comm
, child
->comm
, TASK_COMM_LEN
);
388 __entry
->child_pid
= child
->pid
;
391 TP_printk("comm=%s pid=%d child_comm=%s child_pid=%d",
392 __entry
->parent_comm
, __entry
->parent_pid
,
393 __entry
->child_comm
, __entry
->child_pid
)
397 * Tracepoint for exec:
399 TRACE_EVENT(sched_process_exec
,
401 TP_PROTO(struct task_struct
*p
, pid_t old_pid
,
402 struct linux_binprm
*bprm
),
404 TP_ARGS(p
, old_pid
, bprm
),
407 __string( filename
, bprm
->filename
)
408 __field( pid_t
, pid
)
409 __field( pid_t
, old_pid
)
413 __assign_str(filename
, bprm
->filename
);
414 __entry
->pid
= p
->pid
;
415 __entry
->old_pid
= old_pid
;
418 TP_printk("filename=%s pid=%d old_pid=%d", __get_str(filename
),
419 __entry
->pid
, __entry
->old_pid
)
423 #ifdef CONFIG_SCHEDSTATS
424 #define DEFINE_EVENT_SCHEDSTAT DEFINE_EVENT
425 #define DECLARE_EVENT_CLASS_SCHEDSTAT DECLARE_EVENT_CLASS
427 #define DEFINE_EVENT_SCHEDSTAT DEFINE_EVENT_NOP
428 #define DECLARE_EVENT_CLASS_SCHEDSTAT DECLARE_EVENT_CLASS_NOP
432 * XXX the below sched_stat tracepoints only apply to SCHED_OTHER/BATCH/IDLE
433 * adding sched_stat support to SCHED_FIFO/RR would be welcome.
435 DECLARE_EVENT_CLASS_SCHEDSTAT(sched_stat_template
,
437 TP_PROTO(struct task_struct
*tsk
, u64 delay
),
439 TP_ARGS(__perf_task(tsk
), __perf_count(delay
)),
442 __array( char, comm
, TASK_COMM_LEN
)
443 __field( pid_t
, pid
)
444 __field( u64
, delay
)
448 memcpy(__entry
->comm
, tsk
->comm
, TASK_COMM_LEN
);
449 __entry
->pid
= tsk
->pid
;
450 __entry
->delay
= delay
;
453 TP_printk("comm=%s pid=%d delay=%Lu [ns]",
454 __entry
->comm
, __entry
->pid
,
455 (unsigned long long)__entry
->delay
)
459 * Tracepoint for accounting wait time (time the task is runnable
460 * but not actually running due to scheduler contention).
462 DEFINE_EVENT_SCHEDSTAT(sched_stat_template
, sched_stat_wait
,
463 TP_PROTO(struct task_struct
*tsk
, u64 delay
),
464 TP_ARGS(tsk
, delay
));
467 * Tracepoint for accounting sleep time (time the task is not runnable,
468 * including iowait, see below).
470 DEFINE_EVENT_SCHEDSTAT(sched_stat_template
, sched_stat_sleep
,
471 TP_PROTO(struct task_struct
*tsk
, u64 delay
),
472 TP_ARGS(tsk
, delay
));
475 * Tracepoint for accounting iowait time (time the task is not runnable
476 * due to waiting on IO to complete).
478 DEFINE_EVENT_SCHEDSTAT(sched_stat_template
, sched_stat_iowait
,
479 TP_PROTO(struct task_struct
*tsk
, u64 delay
),
480 TP_ARGS(tsk
, delay
));
483 * Tracepoint for accounting blocked time (time the task is in uninterruptible).
485 DEFINE_EVENT_SCHEDSTAT(sched_stat_template
, sched_stat_blocked
,
486 TP_PROTO(struct task_struct
*tsk
, u64 delay
),
487 TP_ARGS(tsk
, delay
));
490 * Tracepoint for accounting runtime (time the task is executing
493 DECLARE_EVENT_CLASS(sched_stat_runtime
,
495 TP_PROTO(struct task_struct
*tsk
, u64 runtime
, u64 vruntime
),
497 TP_ARGS(tsk
, __perf_count(runtime
), vruntime
),
500 __array( char, comm
, TASK_COMM_LEN
)
501 __field( pid_t
, pid
)
502 __field( u64
, runtime
)
503 __field( u64
, vruntime
)
507 memcpy(__entry
->comm
, tsk
->comm
, TASK_COMM_LEN
);
508 __entry
->pid
= tsk
->pid
;
509 __entry
->runtime
= runtime
;
510 __entry
->vruntime
= vruntime
;
513 TP_printk("comm=%s pid=%d runtime=%Lu [ns] vruntime=%Lu [ns]",
514 __entry
->comm
, __entry
->pid
,
515 (unsigned long long)__entry
->runtime
,
516 (unsigned long long)__entry
->vruntime
)
519 DEFINE_EVENT(sched_stat_runtime
, sched_stat_runtime
,
520 TP_PROTO(struct task_struct
*tsk
, u64 runtime
, u64 vruntime
),
521 TP_ARGS(tsk
, runtime
, vruntime
));
524 * Tracepoint for showing priority inheritance modifying a tasks
527 TRACE_EVENT(sched_pi_setprio
,
529 TP_PROTO(struct task_struct
*tsk
, struct task_struct
*pi_task
),
531 TP_ARGS(tsk
, pi_task
),
534 __array( char, comm
, TASK_COMM_LEN
)
535 __field( pid_t
, pid
)
536 __field( int, oldprio
)
537 __field( int, newprio
)
541 memcpy(__entry
->comm
, tsk
->comm
, TASK_COMM_LEN
);
542 __entry
->pid
= tsk
->pid
;
543 __entry
->oldprio
= tsk
->prio
;
544 __entry
->newprio
= pi_task
?
545 min(tsk
->normal_prio
, pi_task
->prio
) :
547 /* XXX SCHED_DEADLINE bits missing */
550 TP_printk("comm=%s pid=%d oldprio=%d newprio=%d",
551 __entry
->comm
, __entry
->pid
,
552 __entry
->oldprio
, __entry
->newprio
)
555 #ifdef CONFIG_DETECT_HUNG_TASK
556 TRACE_EVENT(sched_process_hang
,
557 TP_PROTO(struct task_struct
*tsk
),
561 __array( char, comm
, TASK_COMM_LEN
)
562 __field( pid_t
, pid
)
566 memcpy(__entry
->comm
, tsk
->comm
, TASK_COMM_LEN
);
567 __entry
->pid
= tsk
->pid
;
570 TP_printk("comm=%s pid=%d", __entry
->comm
, __entry
->pid
)
572 #endif /* CONFIG_DETECT_HUNG_TASK */
575 * Tracks migration of tasks from one runqueue to another. Can be used to
576 * detect if automatic NUMA balancing is bouncing between nodes.
578 TRACE_EVENT(sched_move_numa
,
580 TP_PROTO(struct task_struct
*tsk
, int src_cpu
, int dst_cpu
),
582 TP_ARGS(tsk
, src_cpu
, dst_cpu
),
585 __field( pid_t
, pid
)
586 __field( pid_t
, tgid
)
587 __field( pid_t
, ngid
)
588 __field( int, src_cpu
)
589 __field( int, src_nid
)
590 __field( int, dst_cpu
)
591 __field( int, dst_nid
)
595 __entry
->pid
= task_pid_nr(tsk
);
596 __entry
->tgid
= task_tgid_nr(tsk
);
597 __entry
->ngid
= task_numa_group_id(tsk
);
598 __entry
->src_cpu
= src_cpu
;
599 __entry
->src_nid
= cpu_to_node(src_cpu
);
600 __entry
->dst_cpu
= dst_cpu
;
601 __entry
->dst_nid
= cpu_to_node(dst_cpu
);
604 TP_printk("pid=%d tgid=%d ngid=%d src_cpu=%d src_nid=%d dst_cpu=%d dst_nid=%d",
605 __entry
->pid
, __entry
->tgid
, __entry
->ngid
,
606 __entry
->src_cpu
, __entry
->src_nid
,
607 __entry
->dst_cpu
, __entry
->dst_nid
)
610 DECLARE_EVENT_CLASS(sched_numa_pair_template
,
612 TP_PROTO(struct task_struct
*src_tsk
, int src_cpu
,
613 struct task_struct
*dst_tsk
, int dst_cpu
),
615 TP_ARGS(src_tsk
, src_cpu
, dst_tsk
, dst_cpu
),
618 __field( pid_t
, src_pid
)
619 __field( pid_t
, src_tgid
)
620 __field( pid_t
, src_ngid
)
621 __field( int, src_cpu
)
622 __field( int, src_nid
)
623 __field( pid_t
, dst_pid
)
624 __field( pid_t
, dst_tgid
)
625 __field( pid_t
, dst_ngid
)
626 __field( int, dst_cpu
)
627 __field( int, dst_nid
)
631 __entry
->src_pid
= task_pid_nr(src_tsk
);
632 __entry
->src_tgid
= task_tgid_nr(src_tsk
);
633 __entry
->src_ngid
= task_numa_group_id(src_tsk
);
634 __entry
->src_cpu
= src_cpu
;
635 __entry
->src_nid
= cpu_to_node(src_cpu
);
636 __entry
->dst_pid
= dst_tsk
? task_pid_nr(dst_tsk
) : 0;
637 __entry
->dst_tgid
= dst_tsk
? task_tgid_nr(dst_tsk
) : 0;
638 __entry
->dst_ngid
= dst_tsk
? task_numa_group_id(dst_tsk
) : 0;
639 __entry
->dst_cpu
= dst_cpu
;
640 __entry
->dst_nid
= dst_cpu
>= 0 ? cpu_to_node(dst_cpu
) : -1;
643 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",
644 __entry
->src_pid
, __entry
->src_tgid
, __entry
->src_ngid
,
645 __entry
->src_cpu
, __entry
->src_nid
,
646 __entry
->dst_pid
, __entry
->dst_tgid
, __entry
->dst_ngid
,
647 __entry
->dst_cpu
, __entry
->dst_nid
)
650 DEFINE_EVENT(sched_numa_pair_template
, sched_stick_numa
,
652 TP_PROTO(struct task_struct
*src_tsk
, int src_cpu
,
653 struct task_struct
*dst_tsk
, int dst_cpu
),
655 TP_ARGS(src_tsk
, src_cpu
, dst_tsk
, dst_cpu
)
658 DEFINE_EVENT(sched_numa_pair_template
, sched_swap_numa
,
660 TP_PROTO(struct task_struct
*src_tsk
, int src_cpu
,
661 struct task_struct
*dst_tsk
, int dst_cpu
),
663 TP_ARGS(src_tsk
, src_cpu
, dst_tsk
, dst_cpu
)
668 * Tracepoint for waking a polling cpu without an IPI.
670 TRACE_EVENT(sched_wake_idle_without_ipi
,
684 TP_printk("cpu=%d", __entry
->cpu
)
688 * Following tracepoints are not exported in tracefs and provide hooking
689 * mechanisms only for testing and debugging purposes.
691 * Postfixed with _tp to make them easily identifiable in the code.
693 DECLARE_TRACE(pelt_cfs_tp
,
694 TP_PROTO(struct cfs_rq
*cfs_rq
),
697 DECLARE_TRACE(pelt_rt_tp
,
698 TP_PROTO(struct rq
*rq
),
701 DECLARE_TRACE(pelt_dl_tp
,
702 TP_PROTO(struct rq
*rq
),
705 DECLARE_TRACE(pelt_thermal_tp
,
706 TP_PROTO(struct rq
*rq
),
709 DECLARE_TRACE(pelt_irq_tp
,
710 TP_PROTO(struct rq
*rq
),
713 DECLARE_TRACE(pelt_se_tp
,
714 TP_PROTO(struct sched_entity
*se
),
717 DECLARE_TRACE(sched_cpu_capacity_tp
,
718 TP_PROTO(struct rq
*rq
),
721 DECLARE_TRACE(sched_overutilized_tp
,
722 TP_PROTO(struct root_domain
*rd
, bool overutilized
),
723 TP_ARGS(rd
, overutilized
));
725 DECLARE_TRACE(sched_util_est_cfs_tp
,
726 TP_PROTO(struct cfs_rq
*cfs_rq
),
729 DECLARE_TRACE(sched_util_est_se_tp
,
730 TP_PROTO(struct sched_entity
*se
),
733 DECLARE_TRACE(sched_update_nr_running_tp
,
734 TP_PROTO(struct rq
*rq
, int change
),
735 TP_ARGS(rq
, change
));
737 #endif /* _TRACE_SCHED_H */
739 /* This part must be outside protection */
740 #include <trace/define_trace.h>