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/sched/numa_balancing.h>
9 #include <linux/tracepoint.h>
10 #include <linux/binfmts.h>
13 * Tracepoint for calling kthread_stop, performed to end a kthread:
15 TRACE_EVENT(sched_kthread_stop
,
17 TP_PROTO(struct task_struct
*t
),
22 __array( char, comm
, TASK_COMM_LEN
)
27 memcpy(__entry
->comm
, t
->comm
, TASK_COMM_LEN
);
28 __entry
->pid
= t
->pid
;
31 TP_printk("comm=%s pid=%d", __entry
->comm
, __entry
->pid
)
35 * Tracepoint for the return value of the kthread stopping:
37 TRACE_EVENT(sched_kthread_stop_ret
,
51 TP_printk("ret=%d", __entry
->ret
)
55 * Tracepoint for waking up a task:
57 DECLARE_EVENT_CLASS(sched_wakeup_template
,
59 TP_PROTO(struct task_struct
*p
),
61 TP_ARGS(__perf_task(p
)),
64 __array( char, comm
, TASK_COMM_LEN
)
67 __field( int, success
)
68 __field( int, target_cpu
)
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
);
79 TP_printk("comm=%s pid=%d prio=%d target_cpu=%03d",
80 __entry
->comm
, __entry
->pid
, __entry
->prio
,
85 * Tracepoint called when waking a task; this tracepoint is guaranteed to be
86 * called from the waking context.
88 DEFINE_EVENT(sched_wakeup_template
, sched_waking
,
89 TP_PROTO(struct task_struct
*p
),
93 * Tracepoint called when the task is actually woken; p->state == TASK_RUNNNG.
94 * It it not always called from the waking context.
96 DEFINE_EVENT(sched_wakeup_template
, sched_wakeup
,
97 TP_PROTO(struct task_struct
*p
),
101 * Tracepoint for waking up a new task:
103 DEFINE_EVENT(sched_wakeup_template
, sched_wakeup_new
,
104 TP_PROTO(struct task_struct
*p
),
107 #ifdef CREATE_TRACE_POINTS
108 static inline long __trace_sched_switch_state(bool preempt
, struct task_struct
*p
)
112 #ifdef CONFIG_SCHED_DEBUG
113 BUG_ON(p
!= current
);
114 #endif /* CONFIG_SCHED_DEBUG */
117 * Preemption ignores task state, therefore preempted tasks are always
118 * RUNNING (we will not have dequeued if state != RUNNING).
121 return TASK_REPORT_MAX
;
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
129 state
= task_state_index(p
);
131 return state
? (1 << (state
- 1)) : state
;
133 #endif /* CREATE_TRACE_POINTS */
136 * Tracepoint for task switches, performed by the scheduler:
138 TRACE_EVENT(sched_switch
,
140 TP_PROTO(bool preempt
,
141 struct task_struct
*prev
,
142 struct task_struct
*next
),
144 TP_ARGS(preempt
, prev
, next
),
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
)
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 */
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
,
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" },
177 { EXIT_ZOMBIE
, "Z" },
178 { TASK_PARKED
, "P" },
179 { TASK_DEAD
, "I" }) :
182 __entry
->prev_state
& TASK_REPORT_MAX
? "+" : "",
183 __entry
->next_comm
, __entry
->next_pid
, __entry
->next_prio
)
187 * Tracepoint for a task being migrated:
189 TRACE_EVENT(sched_migrate_task
,
191 TP_PROTO(struct task_struct
*p
, int dest_cpu
),
193 TP_ARGS(p
, dest_cpu
),
196 __array( char, comm
, TASK_COMM_LEN
)
197 __field( pid_t
, pid
)
199 __field( int, orig_cpu
)
200 __field( int, dest_cpu
)
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
;
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
)
216 DECLARE_EVENT_CLASS(sched_process_template
,
218 TP_PROTO(struct task_struct
*p
),
223 __array( char, comm
, TASK_COMM_LEN
)
224 __field( pid_t
, pid
)
229 memcpy(__entry
->comm
, p
->comm
, TASK_COMM_LEN
);
230 __entry
->pid
= p
->pid
;
231 __entry
->prio
= p
->prio
; /* XXX SCHED_DEADLINE */
234 TP_printk("comm=%s pid=%d prio=%d",
235 __entry
->comm
, __entry
->pid
, __entry
->prio
)
239 * Tracepoint for freeing a task:
241 DEFINE_EVENT(sched_process_template
, sched_process_free
,
242 TP_PROTO(struct task_struct
*p
),
247 * Tracepoint for a task exiting:
249 DEFINE_EVENT(sched_process_template
, sched_process_exit
,
250 TP_PROTO(struct task_struct
*p
),
254 * Tracepoint for waiting on task to unschedule:
256 DEFINE_EVENT(sched_process_template
, sched_wait_task
,
257 TP_PROTO(struct task_struct
*p
),
261 * Tracepoint for a waiting task:
263 TRACE_EVENT(sched_process_wait
,
265 TP_PROTO(struct pid
*pid
),
270 __array( char, comm
, TASK_COMM_LEN
)
271 __field( pid_t
, pid
)
276 memcpy(__entry
->comm
, current
->comm
, TASK_COMM_LEN
);
277 __entry
->pid
= pid_nr(pid
);
278 __entry
->prio
= current
->prio
; /* XXX SCHED_DEADLINE */
281 TP_printk("comm=%s pid=%d prio=%d",
282 __entry
->comm
, __entry
->pid
, __entry
->prio
)
286 * Tracepoint for do_fork:
288 TRACE_EVENT(sched_process_fork
,
290 TP_PROTO(struct task_struct
*parent
, struct task_struct
*child
),
292 TP_ARGS(parent
, child
),
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
)
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
;
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
)
314 * Tracepoint for exec:
316 TRACE_EVENT(sched_process_exec
,
318 TP_PROTO(struct task_struct
*p
, pid_t old_pid
,
319 struct linux_binprm
*bprm
),
321 TP_ARGS(p
, old_pid
, bprm
),
324 __string( filename
, bprm
->filename
)
325 __field( pid_t
, pid
)
326 __field( pid_t
, old_pid
)
330 __assign_str(filename
, bprm
->filename
);
331 __entry
->pid
= p
->pid
;
332 __entry
->old_pid
= old_pid
;
335 TP_printk("filename=%s pid=%d old_pid=%d", __get_str(filename
),
336 __entry
->pid
, __entry
->old_pid
)
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.
343 DECLARE_EVENT_CLASS(sched_stat_template
,
345 TP_PROTO(struct task_struct
*tsk
, u64 delay
),
347 TP_ARGS(__perf_task(tsk
), __perf_count(delay
)),
350 __array( char, comm
, TASK_COMM_LEN
)
351 __field( pid_t
, pid
)
352 __field( u64
, delay
)
356 memcpy(__entry
->comm
, tsk
->comm
, TASK_COMM_LEN
);
357 __entry
->pid
= tsk
->pid
;
358 __entry
->delay
= delay
;
361 TP_printk("comm=%s pid=%d delay=%Lu [ns]",
362 __entry
->comm
, __entry
->pid
,
363 (unsigned long long)__entry
->delay
)
368 * Tracepoint for accounting wait time (time the task is runnable
369 * but not actually running due to scheduler contention).
371 DEFINE_EVENT(sched_stat_template
, sched_stat_wait
,
372 TP_PROTO(struct task_struct
*tsk
, u64 delay
),
373 TP_ARGS(tsk
, delay
));
376 * Tracepoint for accounting sleep time (time the task is not runnable,
377 * including iowait, see below).
379 DEFINE_EVENT(sched_stat_template
, sched_stat_sleep
,
380 TP_PROTO(struct task_struct
*tsk
, u64 delay
),
381 TP_ARGS(tsk
, delay
));
384 * Tracepoint for accounting iowait time (time the task is not runnable
385 * due to waiting on IO to complete).
387 DEFINE_EVENT(sched_stat_template
, sched_stat_iowait
,
388 TP_PROTO(struct task_struct
*tsk
, u64 delay
),
389 TP_ARGS(tsk
, delay
));
392 * Tracepoint for accounting blocked time (time the task is in uninterruptible).
394 DEFINE_EVENT(sched_stat_template
, sched_stat_blocked
,
395 TP_PROTO(struct task_struct
*tsk
, u64 delay
),
396 TP_ARGS(tsk
, delay
));
399 * Tracepoint for accounting runtime (time the task is executing
402 DECLARE_EVENT_CLASS(sched_stat_runtime
,
404 TP_PROTO(struct task_struct
*tsk
, u64 runtime
, u64 vruntime
),
406 TP_ARGS(tsk
, __perf_count(runtime
), vruntime
),
409 __array( char, comm
, TASK_COMM_LEN
)
410 __field( pid_t
, pid
)
411 __field( u64
, runtime
)
412 __field( u64
, vruntime
)
416 memcpy(__entry
->comm
, tsk
->comm
, TASK_COMM_LEN
);
417 __entry
->pid
= tsk
->pid
;
418 __entry
->runtime
= runtime
;
419 __entry
->vruntime
= vruntime
;
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
)
428 DEFINE_EVENT(sched_stat_runtime
, sched_stat_runtime
,
429 TP_PROTO(struct task_struct
*tsk
, u64 runtime
, u64 vruntime
),
430 TP_ARGS(tsk
, runtime
, vruntime
));
433 * Tracepoint for showing priority inheritance modifying a tasks
436 TRACE_EVENT(sched_pi_setprio
,
438 TP_PROTO(struct task_struct
*tsk
, struct task_struct
*pi_task
),
440 TP_ARGS(tsk
, pi_task
),
443 __array( char, comm
, TASK_COMM_LEN
)
444 __field( pid_t
, pid
)
445 __field( int, oldprio
)
446 __field( int, newprio
)
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
) :
456 /* XXX SCHED_DEADLINE bits missing */
459 TP_printk("comm=%s pid=%d oldprio=%d newprio=%d",
460 __entry
->comm
, __entry
->pid
,
461 __entry
->oldprio
, __entry
->newprio
)
464 #ifdef CONFIG_DETECT_HUNG_TASK
465 TRACE_EVENT(sched_process_hang
,
466 TP_PROTO(struct task_struct
*tsk
),
470 __array( char, comm
, TASK_COMM_LEN
)
471 __field( pid_t
, pid
)
475 memcpy(__entry
->comm
, tsk
->comm
, TASK_COMM_LEN
);
476 __entry
->pid
= tsk
->pid
;
479 TP_printk("comm=%s pid=%d", __entry
->comm
, __entry
->pid
)
481 #endif /* CONFIG_DETECT_HUNG_TASK */
483 DECLARE_EVENT_CLASS(sched_move_task_template
,
485 TP_PROTO(struct task_struct
*tsk
, int src_cpu
, int dst_cpu
),
487 TP_ARGS(tsk
, src_cpu
, dst_cpu
),
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
)
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
);
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
)
516 * Tracks migration of tasks from one runqueue to another. Can be used to
517 * detect if automatic NUMA balancing is bouncing between nodes
519 DEFINE_EVENT(sched_move_task_template
, sched_move_numa
,
520 TP_PROTO(struct task_struct
*tsk
, int src_cpu
, int dst_cpu
),
522 TP_ARGS(tsk
, src_cpu
, dst_cpu
)
525 DEFINE_EVENT(sched_move_task_template
, sched_stick_numa
,
526 TP_PROTO(struct task_struct
*tsk
, int src_cpu
, int dst_cpu
),
528 TP_ARGS(tsk
, src_cpu
, dst_cpu
)
531 TRACE_EVENT(sched_swap_numa
,
533 TP_PROTO(struct task_struct
*src_tsk
, int src_cpu
,
534 struct task_struct
*dst_tsk
, int dst_cpu
),
536 TP_ARGS(src_tsk
, src_cpu
, dst_tsk
, dst_cpu
),
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
)
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
);
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
)
572 * Tracepoint for waking a polling cpu without an IPI.
574 TRACE_EVENT(sched_wake_idle_without_ipi
,
588 TP_printk("cpu=%d", __entry
->cpu
)
590 #endif /* _TRACE_SCHED_H */
592 /* This part must be outside protection */
593 #include <trace/define_trace.h>