2 * trace task wakeup timings
4 * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
7 * Based on code from the latency_tracer, that is:
9 * Copyright (C) 2004-2006 Ingo Molnar
10 * Copyright (C) 2004 Nadia Yvette Chambers
12 #include <linux/module.h>
13 #include <linux/kallsyms.h>
14 #include <linux/uaccess.h>
15 #include <linux/ftrace.h>
16 #include <linux/sched/rt.h>
17 #include <linux/sched/deadline.h>
18 #include <trace/events/sched.h>
21 static struct trace_array
*wakeup_trace
;
22 static int __read_mostly tracer_enabled
;
24 static struct task_struct
*wakeup_task
;
25 static int wakeup_cpu
;
26 static int wakeup_current_cpu
;
27 static unsigned wakeup_prio
= -1;
30 static int tracing_dl
= 0;
32 static arch_spinlock_t wakeup_lock
=
33 (arch_spinlock_t
)__ARCH_SPIN_LOCK_UNLOCKED
;
35 static void wakeup_reset(struct trace_array
*tr
);
36 static void __wakeup_reset(struct trace_array
*tr
);
38 static int save_flags
;
40 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
41 static int wakeup_display_graph(struct trace_array
*tr
, int set
);
42 # define is_graph(tr) ((tr)->trace_flags & TRACE_ITER_DISPLAY_GRAPH)
44 static inline int wakeup_display_graph(struct trace_array
*tr
, int set
)
48 # define is_graph(tr) false
52 #ifdef CONFIG_FUNCTION_TRACER
54 static int wakeup_graph_entry(struct ftrace_graph_ent
*trace
);
55 static void wakeup_graph_return(struct ftrace_graph_ret
*trace
);
57 static bool function_enabled
;
60 * Prologue for the wakeup function tracers.
62 * Returns 1 if it is OK to continue, and preemption
63 * is disabled and data->disabled is incremented.
64 * 0 if the trace is to be ignored, and preemption
65 * is not disabled and data->disabled is
68 * Note, this function is also used outside this ifdef but
69 * inside the #ifdef of the function graph tracer below.
70 * This is OK, since the function graph tracer is
71 * dependent on the function tracer.
74 func_prolog_preempt_disable(struct trace_array
*tr
,
75 struct trace_array_cpu
**data
,
81 if (likely(!wakeup_task
))
84 *pc
= preempt_count();
85 preempt_disable_notrace();
87 cpu
= raw_smp_processor_id();
88 if (cpu
!= wakeup_current_cpu
)
91 *data
= per_cpu_ptr(tr
->trace_buffer
.data
, cpu
);
92 disabled
= atomic_inc_return(&(*data
)->disabled
);
93 if (unlikely(disabled
!= 1))
99 atomic_dec(&(*data
)->disabled
);
102 preempt_enable_notrace();
107 * wakeup uses its own tracer function to keep the overhead down:
110 wakeup_tracer_call(unsigned long ip
, unsigned long parent_ip
,
111 struct ftrace_ops
*op
, struct pt_regs
*pt_regs
)
113 struct trace_array
*tr
= wakeup_trace
;
114 struct trace_array_cpu
*data
;
118 if (!func_prolog_preempt_disable(tr
, &data
, &pc
))
121 local_irq_save(flags
);
122 trace_function(tr
, ip
, parent_ip
, flags
, pc
);
123 local_irq_restore(flags
);
125 atomic_dec(&data
->disabled
);
126 preempt_enable_notrace();
129 static int register_wakeup_function(struct trace_array
*tr
, int graph
, int set
)
133 /* 'set' is set if TRACE_ITER_FUNCTION is about to be set */
134 if (function_enabled
|| (!set
&& !(tr
->trace_flags
& TRACE_ITER_FUNCTION
)))
138 ret
= register_ftrace_graph(&wakeup_graph_return
,
139 &wakeup_graph_entry
);
141 ret
= register_ftrace_function(tr
->ops
);
144 function_enabled
= true;
149 static void unregister_wakeup_function(struct trace_array
*tr
, int graph
)
151 if (!function_enabled
)
155 unregister_ftrace_graph();
157 unregister_ftrace_function(tr
->ops
);
159 function_enabled
= false;
162 static int wakeup_function_set(struct trace_array
*tr
, u32 mask
, int set
)
164 if (!(mask
& TRACE_ITER_FUNCTION
))
168 register_wakeup_function(tr
, is_graph(tr
), 1);
170 unregister_wakeup_function(tr
, is_graph(tr
));
174 static int register_wakeup_function(struct trace_array
*tr
, int graph
, int set
)
178 static void unregister_wakeup_function(struct trace_array
*tr
, int graph
) { }
179 static int wakeup_function_set(struct trace_array
*tr
, u32 mask
, int set
)
183 #endif /* CONFIG_FUNCTION_TRACER */
185 static int wakeup_flag_changed(struct trace_array
*tr
, u32 mask
, int set
)
187 struct tracer
*tracer
= tr
->current_trace
;
189 if (wakeup_function_set(tr
, mask
, set
))
192 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
193 if (mask
& TRACE_ITER_DISPLAY_GRAPH
)
194 return wakeup_display_graph(tr
, set
);
197 return trace_keep_overwrite(tracer
, mask
, set
);
200 static int start_func_tracer(struct trace_array
*tr
, int graph
)
204 ret
= register_wakeup_function(tr
, graph
, 0);
206 if (!ret
&& tracing_is_enabled())
214 static void stop_func_tracer(struct trace_array
*tr
, int graph
)
218 unregister_wakeup_function(tr
, graph
);
221 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
222 static int wakeup_display_graph(struct trace_array
*tr
, int set
)
224 if (!(is_graph(tr
) ^ set
))
227 stop_func_tracer(tr
, !set
);
229 wakeup_reset(wakeup_trace
);
232 return start_func_tracer(tr
, set
);
235 static int wakeup_graph_entry(struct ftrace_graph_ent
*trace
)
237 struct trace_array
*tr
= wakeup_trace
;
238 struct trace_array_cpu
*data
;
242 if (!func_prolog_preempt_disable(tr
, &data
, &pc
))
245 local_save_flags(flags
);
246 ret
= __trace_graph_entry(tr
, trace
, flags
, pc
);
247 atomic_dec(&data
->disabled
);
248 preempt_enable_notrace();
253 static void wakeup_graph_return(struct ftrace_graph_ret
*trace
)
255 struct trace_array
*tr
= wakeup_trace
;
256 struct trace_array_cpu
*data
;
260 if (!func_prolog_preempt_disable(tr
, &data
, &pc
))
263 local_save_flags(flags
);
264 __trace_graph_return(tr
, trace
, flags
, pc
);
265 atomic_dec(&data
->disabled
);
267 preempt_enable_notrace();
271 static void wakeup_trace_open(struct trace_iterator
*iter
)
273 if (is_graph(iter
->tr
))
274 graph_trace_open(iter
);
277 static void wakeup_trace_close(struct trace_iterator
*iter
)
280 graph_trace_close(iter
);
283 #define GRAPH_TRACER_FLAGS (TRACE_GRAPH_PRINT_PROC | \
284 TRACE_GRAPH_PRINT_ABS_TIME | \
285 TRACE_GRAPH_PRINT_DURATION)
287 static enum print_line_t
wakeup_print_line(struct trace_iterator
*iter
)
290 * In graph mode call the graph tracer output function,
291 * otherwise go with the TRACE_FN event handler
293 if (is_graph(iter
->tr
))
294 return print_graph_function_flags(iter
, GRAPH_TRACER_FLAGS
);
296 return TRACE_TYPE_UNHANDLED
;
299 static void wakeup_print_header(struct seq_file
*s
)
301 if (is_graph(wakeup_trace
))
302 print_graph_headers_flags(s
, GRAPH_TRACER_FLAGS
);
304 trace_default_header(s
);
308 __trace_function(struct trace_array
*tr
,
309 unsigned long ip
, unsigned long parent_ip
,
310 unsigned long flags
, int pc
)
313 trace_graph_function(tr
, ip
, parent_ip
, flags
, pc
);
315 trace_function(tr
, ip
, parent_ip
, flags
, pc
);
318 #define __trace_function trace_function
320 static enum print_line_t
wakeup_print_line(struct trace_iterator
*iter
)
322 return TRACE_TYPE_UNHANDLED
;
325 static void wakeup_trace_open(struct trace_iterator
*iter
) { }
326 static void wakeup_trace_close(struct trace_iterator
*iter
) { }
328 #ifdef CONFIG_FUNCTION_TRACER
329 static int wakeup_graph_entry(struct ftrace_graph_ent
*trace
)
333 static void wakeup_graph_return(struct ftrace_graph_ret
*trace
) { }
334 static void wakeup_print_header(struct seq_file
*s
)
336 trace_default_header(s
);
339 static void wakeup_print_header(struct seq_file
*s
)
341 trace_latency_header(s
);
343 #endif /* CONFIG_FUNCTION_TRACER */
344 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
347 * Should this new latency be reported/recorded?
349 static bool report_latency(struct trace_array
*tr
, cycle_t delta
)
351 if (tracing_thresh
) {
352 if (delta
< tracing_thresh
)
355 if (delta
<= tr
->max_latency
)
362 probe_wakeup_migrate_task(void *ignore
, struct task_struct
*task
, int cpu
)
364 if (task
!= wakeup_task
)
367 wakeup_current_cpu
= cpu
;
371 tracing_sched_switch_trace(struct trace_array
*tr
,
372 struct task_struct
*prev
,
373 struct task_struct
*next
,
374 unsigned long flags
, int pc
)
376 struct trace_event_call
*call
= &event_context_switch
;
377 struct ring_buffer
*buffer
= tr
->trace_buffer
.buffer
;
378 struct ring_buffer_event
*event
;
379 struct ctx_switch_entry
*entry
;
381 event
= trace_buffer_lock_reserve(buffer
, TRACE_CTX
,
382 sizeof(*entry
), flags
, pc
);
385 entry
= ring_buffer_event_data(event
);
386 entry
->prev_pid
= prev
->pid
;
387 entry
->prev_prio
= prev
->prio
;
388 entry
->prev_state
= prev
->state
;
389 entry
->next_pid
= next
->pid
;
390 entry
->next_prio
= next
->prio
;
391 entry
->next_state
= next
->state
;
392 entry
->next_cpu
= task_cpu(next
);
394 if (!call_filter_check_discard(call
, entry
, buffer
, event
))
395 trace_buffer_unlock_commit(tr
, buffer
, event
, flags
, pc
);
399 tracing_sched_wakeup_trace(struct trace_array
*tr
,
400 struct task_struct
*wakee
,
401 struct task_struct
*curr
,
402 unsigned long flags
, int pc
)
404 struct trace_event_call
*call
= &event_wakeup
;
405 struct ring_buffer_event
*event
;
406 struct ctx_switch_entry
*entry
;
407 struct ring_buffer
*buffer
= tr
->trace_buffer
.buffer
;
409 event
= trace_buffer_lock_reserve(buffer
, TRACE_WAKE
,
410 sizeof(*entry
), flags
, pc
);
413 entry
= ring_buffer_event_data(event
);
414 entry
->prev_pid
= curr
->pid
;
415 entry
->prev_prio
= curr
->prio
;
416 entry
->prev_state
= curr
->state
;
417 entry
->next_pid
= wakee
->pid
;
418 entry
->next_prio
= wakee
->prio
;
419 entry
->next_state
= wakee
->state
;
420 entry
->next_cpu
= task_cpu(wakee
);
422 if (!call_filter_check_discard(call
, entry
, buffer
, event
))
423 trace_buffer_unlock_commit(tr
, buffer
, event
, flags
, pc
);
427 probe_wakeup_sched_switch(void *ignore
, bool preempt
,
428 struct task_struct
*prev
, struct task_struct
*next
)
430 struct trace_array_cpu
*data
;
431 cycle_t T0
, T1
, delta
;
437 tracing_record_cmdline(prev
);
439 if (unlikely(!tracer_enabled
))
443 * When we start a new trace, we set wakeup_task to NULL
444 * and then set tracer_enabled = 1. We want to make sure
445 * that another CPU does not see the tracer_enabled = 1
446 * and the wakeup_task with an older task, that might
447 * actually be the same as next.
451 if (next
!= wakeup_task
)
454 pc
= preempt_count();
456 /* disable local data, not wakeup_cpu data */
457 cpu
= raw_smp_processor_id();
458 disabled
= atomic_inc_return(&per_cpu_ptr(wakeup_trace
->trace_buffer
.data
, cpu
)->disabled
);
459 if (likely(disabled
!= 1))
462 local_irq_save(flags
);
463 arch_spin_lock(&wakeup_lock
);
465 /* We could race with grabbing wakeup_lock */
466 if (unlikely(!tracer_enabled
|| next
!= wakeup_task
))
469 /* The task we are waiting for is waking up */
470 data
= per_cpu_ptr(wakeup_trace
->trace_buffer
.data
, wakeup_cpu
);
472 __trace_function(wakeup_trace
, CALLER_ADDR0
, CALLER_ADDR1
, flags
, pc
);
473 tracing_sched_switch_trace(wakeup_trace
, prev
, next
, flags
, pc
);
475 T0
= data
->preempt_timestamp
;
476 T1
= ftrace_now(cpu
);
479 if (!report_latency(wakeup_trace
, delta
))
482 if (likely(!is_tracing_stopped())) {
483 wakeup_trace
->max_latency
= delta
;
484 update_max_tr(wakeup_trace
, wakeup_task
, wakeup_cpu
);
488 __wakeup_reset(wakeup_trace
);
489 arch_spin_unlock(&wakeup_lock
);
490 local_irq_restore(flags
);
492 atomic_dec(&per_cpu_ptr(wakeup_trace
->trace_buffer
.data
, cpu
)->disabled
);
495 static void __wakeup_reset(struct trace_array
*tr
)
502 put_task_struct(wakeup_task
);
507 static void wakeup_reset(struct trace_array
*tr
)
511 tracing_reset_online_cpus(&tr
->trace_buffer
);
513 local_irq_save(flags
);
514 arch_spin_lock(&wakeup_lock
);
516 arch_spin_unlock(&wakeup_lock
);
517 local_irq_restore(flags
);
521 probe_wakeup(void *ignore
, struct task_struct
*p
)
523 struct trace_array_cpu
*data
;
524 int cpu
= smp_processor_id();
529 if (likely(!tracer_enabled
))
532 tracing_record_cmdline(p
);
533 tracing_record_cmdline(current
);
536 * Semantic is like this:
537 * - wakeup tracer handles all tasks in the system, independently
538 * from their scheduling class;
539 * - wakeup_rt tracer handles tasks belonging to sched_dl and
541 * - wakeup_dl handles tasks belonging to sched_dl class only.
543 if (tracing_dl
|| (wakeup_dl
&& !dl_task(p
)) ||
544 (wakeup_rt
&& !dl_task(p
) && !rt_task(p
)) ||
545 (!dl_task(p
) && (p
->prio
>= wakeup_prio
|| p
->prio
>= current
->prio
)))
548 pc
= preempt_count();
549 disabled
= atomic_inc_return(&per_cpu_ptr(wakeup_trace
->trace_buffer
.data
, cpu
)->disabled
);
550 if (unlikely(disabled
!= 1))
553 /* interrupts should be off from try_to_wake_up */
554 arch_spin_lock(&wakeup_lock
);
556 /* check for races. */
557 if (!tracer_enabled
|| tracing_dl
||
558 (!dl_task(p
) && p
->prio
>= wakeup_prio
))
561 /* reset the trace */
562 __wakeup_reset(wakeup_trace
);
564 wakeup_cpu
= task_cpu(p
);
565 wakeup_current_cpu
= wakeup_cpu
;
566 wakeup_prio
= p
->prio
;
569 * Once you start tracing a -deadline task, don't bother tracing
570 * another task until the first one wakes up.
578 get_task_struct(wakeup_task
);
580 local_save_flags(flags
);
582 data
= per_cpu_ptr(wakeup_trace
->trace_buffer
.data
, wakeup_cpu
);
583 data
->preempt_timestamp
= ftrace_now(cpu
);
584 tracing_sched_wakeup_trace(wakeup_trace
, p
, current
, flags
, pc
);
587 * We must be careful in using CALLER_ADDR2. But since wake_up
588 * is not called by an assembly function (where as schedule is)
589 * it should be safe to use it here.
591 __trace_function(wakeup_trace
, CALLER_ADDR1
, CALLER_ADDR2
, flags
, pc
);
594 arch_spin_unlock(&wakeup_lock
);
596 atomic_dec(&per_cpu_ptr(wakeup_trace
->trace_buffer
.data
, cpu
)->disabled
);
599 static void start_wakeup_tracer(struct trace_array
*tr
)
603 ret
= register_trace_sched_wakeup(probe_wakeup
, NULL
);
605 pr_info("wakeup trace: Couldn't activate tracepoint"
606 " probe to kernel_sched_wakeup\n");
610 ret
= register_trace_sched_wakeup_new(probe_wakeup
, NULL
);
612 pr_info("wakeup trace: Couldn't activate tracepoint"
613 " probe to kernel_sched_wakeup_new\n");
617 ret
= register_trace_sched_switch(probe_wakeup_sched_switch
, NULL
);
619 pr_info("sched trace: Couldn't activate tracepoint"
620 " probe to kernel_sched_switch\n");
621 goto fail_deprobe_wake_new
;
624 ret
= register_trace_sched_migrate_task(probe_wakeup_migrate_task
, NULL
);
626 pr_info("wakeup trace: Couldn't activate tracepoint"
627 " probe to kernel_sched_migrate_task\n");
634 * Don't let the tracer_enabled = 1 show up before
635 * the wakeup_task is reset. This may be overkill since
636 * wakeup_reset does a spin_unlock after setting the
637 * wakeup_task to NULL, but I want to be safe.
638 * This is a slow path anyway.
642 if (start_func_tracer(tr
, is_graph(tr
)))
643 printk(KERN_ERR
"failed to start wakeup tracer\n");
646 fail_deprobe_wake_new
:
647 unregister_trace_sched_wakeup_new(probe_wakeup
, NULL
);
649 unregister_trace_sched_wakeup(probe_wakeup
, NULL
);
652 static void stop_wakeup_tracer(struct trace_array
*tr
)
655 stop_func_tracer(tr
, is_graph(tr
));
656 unregister_trace_sched_switch(probe_wakeup_sched_switch
, NULL
);
657 unregister_trace_sched_wakeup_new(probe_wakeup
, NULL
);
658 unregister_trace_sched_wakeup(probe_wakeup
, NULL
);
659 unregister_trace_sched_migrate_task(probe_wakeup_migrate_task
, NULL
);
662 static bool wakeup_busy
;
664 static int __wakeup_tracer_init(struct trace_array
*tr
)
666 save_flags
= tr
->trace_flags
;
668 /* non overwrite screws up the latency tracers */
669 set_tracer_flag(tr
, TRACE_ITER_OVERWRITE
, 1);
670 set_tracer_flag(tr
, TRACE_ITER_LATENCY_FMT
, 1);
674 ftrace_init_array_ops(tr
, wakeup_tracer_call
);
675 start_wakeup_tracer(tr
);
681 static int wakeup_tracer_init(struct trace_array
*tr
)
688 return __wakeup_tracer_init(tr
);
691 static int wakeup_rt_tracer_init(struct trace_array
*tr
)
698 return __wakeup_tracer_init(tr
);
701 static int wakeup_dl_tracer_init(struct trace_array
*tr
)
708 return __wakeup_tracer_init(tr
);
711 static void wakeup_tracer_reset(struct trace_array
*tr
)
713 int lat_flag
= save_flags
& TRACE_ITER_LATENCY_FMT
;
714 int overwrite_flag
= save_flags
& TRACE_ITER_OVERWRITE
;
716 stop_wakeup_tracer(tr
);
717 /* make sure we put back any tasks we are tracing */
720 set_tracer_flag(tr
, TRACE_ITER_LATENCY_FMT
, lat_flag
);
721 set_tracer_flag(tr
, TRACE_ITER_OVERWRITE
, overwrite_flag
);
722 ftrace_reset_array_ops(tr
);
726 static void wakeup_tracer_start(struct trace_array
*tr
)
732 static void wakeup_tracer_stop(struct trace_array
*tr
)
737 static struct tracer wakeup_tracer __read_mostly
=
740 .init
= wakeup_tracer_init
,
741 .reset
= wakeup_tracer_reset
,
742 .start
= wakeup_tracer_start
,
743 .stop
= wakeup_tracer_stop
,
745 .print_header
= wakeup_print_header
,
746 .print_line
= wakeup_print_line
,
747 .flag_changed
= wakeup_flag_changed
,
748 #ifdef CONFIG_FTRACE_SELFTEST
749 .selftest
= trace_selftest_startup_wakeup
,
751 .open
= wakeup_trace_open
,
752 .close
= wakeup_trace_close
,
753 .allow_instances
= true,
757 static struct tracer wakeup_rt_tracer __read_mostly
=
760 .init
= wakeup_rt_tracer_init
,
761 .reset
= wakeup_tracer_reset
,
762 .start
= wakeup_tracer_start
,
763 .stop
= wakeup_tracer_stop
,
765 .print_header
= wakeup_print_header
,
766 .print_line
= wakeup_print_line
,
767 .flag_changed
= wakeup_flag_changed
,
768 #ifdef CONFIG_FTRACE_SELFTEST
769 .selftest
= trace_selftest_startup_wakeup
,
771 .open
= wakeup_trace_open
,
772 .close
= wakeup_trace_close
,
773 .allow_instances
= true,
777 static struct tracer wakeup_dl_tracer __read_mostly
=
780 .init
= wakeup_dl_tracer_init
,
781 .reset
= wakeup_tracer_reset
,
782 .start
= wakeup_tracer_start
,
783 .stop
= wakeup_tracer_stop
,
785 .print_header
= wakeup_print_header
,
786 .print_line
= wakeup_print_line
,
787 .flag_changed
= wakeup_flag_changed
,
788 #ifdef CONFIG_FTRACE_SELFTEST
789 .selftest
= trace_selftest_startup_wakeup
,
791 .open
= wakeup_trace_open
,
792 .close
= wakeup_trace_close
,
796 __init
static int init_wakeup_tracer(void)
800 ret
= register_tracer(&wakeup_tracer
);
804 ret
= register_tracer(&wakeup_rt_tracer
);
808 ret
= register_tracer(&wakeup_dl_tracer
);
814 core_initcall(init_wakeup_tracer
);