1 // SPDX-License-Identifier: GPL-2.0
3 * trace task wakeup timings
5 * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
6 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
8 * Based on code from the latency_tracer, that is:
10 * Copyright (C) 2004-2006 Ingo Molnar
11 * Copyright (C) 2004 Nadia Yvette Chambers
13 #include <linux/module.h>
14 #include <linux/kallsyms.h>
15 #include <linux/uaccess.h>
16 #include <linux/ftrace.h>
17 #include <linux/sched/rt.h>
18 #include <linux/sched/deadline.h>
19 #include <trace/events/sched.h>
22 static struct trace_array *wakeup_trace;
23 static int __read_mostly tracer_enabled;
25 static struct task_struct *wakeup_task;
26 static int wakeup_cpu;
27 static int wakeup_current_cpu;
28 static unsigned wakeup_prio = -1;
29 static bool wakeup_rt;
30 static bool wakeup_dl;
31 static bool tracing_dl;
33 static arch_spinlock_t wakeup_lock =
34 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
36 static void wakeup_reset(struct trace_array *tr);
37 static void __wakeup_reset(struct trace_array *tr);
38 static int start_func_tracer(struct trace_array *tr, int graph);
39 static void stop_func_tracer(struct trace_array *tr, int graph);
41 static int save_flags;
43 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
44 # define is_graph(tr) ((tr)->trace_flags & TRACE_ITER_DISPLAY_GRAPH)
46 # define is_graph(tr) false
49 #ifdef CONFIG_FUNCTION_TRACER
51 static bool function_enabled;
54 * Prologue for the wakeup function tracers.
56 * Returns 1 if it is OK to continue, and preemption
57 * is disabled and data->disabled is incremented.
58 * 0 if the trace is to be ignored, and preemption
59 * is not disabled and data->disabled is
62 * Note, this function is also used outside this ifdef but
63 * inside the #ifdef of the function graph tracer below.
64 * This is OK, since the function graph tracer is
65 * dependent on the function tracer.
68 func_prolog_preempt_disable(struct trace_array *tr,
69 struct trace_array_cpu **data,
70 unsigned int *trace_ctx)
75 if (likely(!wakeup_task))
78 *trace_ctx = tracing_gen_ctx();
79 preempt_disable_notrace();
81 cpu = raw_smp_processor_id();
82 if (cpu != wakeup_current_cpu)
85 *data = per_cpu_ptr(tr->array_buffer.data, cpu);
86 disabled = atomic_inc_return(&(*data)->disabled);
87 if (unlikely(disabled != 1))
93 atomic_dec(&(*data)->disabled);
96 preempt_enable_notrace();
100 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
102 static int wakeup_display_graph(struct trace_array *tr, int set)
104 if (!(is_graph(tr) ^ set))
107 stop_func_tracer(tr, !set);
109 wakeup_reset(wakeup_trace);
112 return start_func_tracer(tr, set);
115 static int wakeup_graph_entry(struct ftrace_graph_ent *trace)
117 struct trace_array *tr = wakeup_trace;
118 struct trace_array_cpu *data;
119 unsigned int trace_ctx;
122 if (ftrace_graph_ignore_func(trace))
125 * Do not trace a function if it's filtered by set_graph_notrace.
126 * Make the index of ret stack negative to indicate that it should
127 * ignore further functions. But it needs its own ret stack entry
128 * to recover the original index in order to continue tracing after
129 * returning from the function.
131 if (ftrace_graph_notrace_addr(trace->func))
134 if (!func_prolog_preempt_disable(tr, &data, &trace_ctx))
137 ret = __trace_graph_entry(tr, trace, trace_ctx);
138 atomic_dec(&data->disabled);
139 preempt_enable_notrace();
144 static void wakeup_graph_return(struct ftrace_graph_ret *trace)
146 struct trace_array *tr = wakeup_trace;
147 struct trace_array_cpu *data;
148 unsigned int trace_ctx;
150 ftrace_graph_addr_finish(trace);
152 if (!func_prolog_preempt_disable(tr, &data, &trace_ctx))
155 __trace_graph_return(tr, trace, trace_ctx);
156 atomic_dec(&data->disabled);
158 preempt_enable_notrace();
162 static struct fgraph_ops fgraph_wakeup_ops = {
163 .entryfunc = &wakeup_graph_entry,
164 .retfunc = &wakeup_graph_return,
167 static void wakeup_trace_open(struct trace_iterator *iter)
169 if (is_graph(iter->tr))
170 graph_trace_open(iter);
172 iter->private = NULL;
175 static void wakeup_trace_close(struct trace_iterator *iter)
178 graph_trace_close(iter);
181 #define GRAPH_TRACER_FLAGS (TRACE_GRAPH_PRINT_PROC | \
182 TRACE_GRAPH_PRINT_CPU | \
183 TRACE_GRAPH_PRINT_REL_TIME | \
184 TRACE_GRAPH_PRINT_DURATION | \
185 TRACE_GRAPH_PRINT_OVERHEAD | \
186 TRACE_GRAPH_PRINT_IRQS)
188 static enum print_line_t wakeup_print_line(struct trace_iterator *iter)
191 * In graph mode call the graph tracer output function,
192 * otherwise go with the TRACE_FN event handler
194 if (is_graph(iter->tr))
195 return print_graph_function_flags(iter, GRAPH_TRACER_FLAGS);
197 return TRACE_TYPE_UNHANDLED;
200 static void wakeup_print_header(struct seq_file *s)
202 if (is_graph(wakeup_trace))
203 print_graph_headers_flags(s, GRAPH_TRACER_FLAGS);
205 trace_default_header(s);
207 #endif /* else CONFIG_FUNCTION_GRAPH_TRACER */
210 * wakeup uses its own tracer function to keep the overhead down:
213 wakeup_tracer_call(unsigned long ip, unsigned long parent_ip,
214 struct ftrace_ops *op, struct ftrace_regs *fregs)
216 struct trace_array *tr = wakeup_trace;
217 struct trace_array_cpu *data;
219 unsigned int trace_ctx;
221 if (!func_prolog_preempt_disable(tr, &data, &trace_ctx))
224 local_irq_save(flags);
225 trace_function(tr, ip, parent_ip, trace_ctx);
226 local_irq_restore(flags);
228 atomic_dec(&data->disabled);
229 preempt_enable_notrace();
232 static int register_wakeup_function(struct trace_array *tr, int graph, int set)
236 /* 'set' is set if TRACE_ITER_FUNCTION is about to be set */
237 if (function_enabled || (!set && !(tr->trace_flags & TRACE_ITER_FUNCTION)))
241 ret = register_ftrace_graph(&fgraph_wakeup_ops);
243 ret = register_ftrace_function(tr->ops);
246 function_enabled = true;
251 static void unregister_wakeup_function(struct trace_array *tr, int graph)
253 if (!function_enabled)
257 unregister_ftrace_graph(&fgraph_wakeup_ops);
259 unregister_ftrace_function(tr->ops);
261 function_enabled = false;
264 static int wakeup_function_set(struct trace_array *tr, u32 mask, int set)
266 if (!(mask & TRACE_ITER_FUNCTION))
270 register_wakeup_function(tr, is_graph(tr), 1);
272 unregister_wakeup_function(tr, is_graph(tr));
275 #else /* CONFIG_FUNCTION_TRACER */
276 static int register_wakeup_function(struct trace_array *tr, int graph, int set)
280 static void unregister_wakeup_function(struct trace_array *tr, int graph) { }
281 static int wakeup_function_set(struct trace_array *tr, u32 mask, int set)
285 #endif /* else CONFIG_FUNCTION_TRACER */
287 #ifndef CONFIG_FUNCTION_GRAPH_TRACER
288 static enum print_line_t wakeup_print_line(struct trace_iterator *iter)
290 return TRACE_TYPE_UNHANDLED;
293 static void wakeup_trace_open(struct trace_iterator *iter) { }
294 static void wakeup_trace_close(struct trace_iterator *iter) { }
296 static void wakeup_print_header(struct seq_file *s)
298 trace_default_header(s);
300 #endif /* !CONFIG_FUNCTION_GRAPH_TRACER */
303 __trace_function(struct trace_array *tr,
304 unsigned long ip, unsigned long parent_ip,
305 unsigned int trace_ctx)
308 trace_graph_function(tr, ip, parent_ip, trace_ctx);
310 trace_function(tr, ip, parent_ip, trace_ctx);
313 static int wakeup_flag_changed(struct trace_array *tr, u32 mask, int set)
315 struct tracer *tracer = tr->current_trace;
317 if (wakeup_function_set(tr, mask, set))
320 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
321 if (mask & TRACE_ITER_DISPLAY_GRAPH)
322 return wakeup_display_graph(tr, set);
325 return trace_keep_overwrite(tracer, mask, set);
328 static int start_func_tracer(struct trace_array *tr, int graph)
332 ret = register_wakeup_function(tr, graph, 0);
334 if (!ret && tracing_is_enabled())
342 static void stop_func_tracer(struct trace_array *tr, int graph)
346 unregister_wakeup_function(tr, graph);
350 * Should this new latency be reported/recorded?
352 static bool report_latency(struct trace_array *tr, u64 delta)
354 if (tracing_thresh) {
355 if (delta < tracing_thresh)
358 if (delta <= tr->max_latency)
365 probe_wakeup_migrate_task(void *ignore, struct task_struct *task, int cpu)
367 if (task != wakeup_task)
370 wakeup_current_cpu = cpu;
374 tracing_sched_switch_trace(struct trace_array *tr,
375 struct task_struct *prev,
376 struct task_struct *next,
377 unsigned int trace_ctx)
379 struct trace_event_call *call = &event_context_switch;
380 struct trace_buffer *buffer = tr->array_buffer.buffer;
381 struct ring_buffer_event *event;
382 struct ctx_switch_entry *entry;
384 event = trace_buffer_lock_reserve(buffer, TRACE_CTX,
385 sizeof(*entry), trace_ctx);
388 entry = ring_buffer_event_data(event);
389 entry->prev_pid = prev->pid;
390 entry->prev_prio = prev->prio;
391 entry->prev_state = task_state_index(prev);
392 entry->next_pid = next->pid;
393 entry->next_prio = next->prio;
394 entry->next_state = task_state_index(next);
395 entry->next_cpu = task_cpu(next);
397 if (!call_filter_check_discard(call, entry, buffer, event))
398 trace_buffer_unlock_commit(tr, buffer, event, trace_ctx);
402 tracing_sched_wakeup_trace(struct trace_array *tr,
403 struct task_struct *wakee,
404 struct task_struct *curr,
405 unsigned int trace_ctx)
407 struct trace_event_call *call = &event_wakeup;
408 struct ring_buffer_event *event;
409 struct ctx_switch_entry *entry;
410 struct trace_buffer *buffer = tr->array_buffer.buffer;
412 event = trace_buffer_lock_reserve(buffer, TRACE_WAKE,
413 sizeof(*entry), trace_ctx);
416 entry = ring_buffer_event_data(event);
417 entry->prev_pid = curr->pid;
418 entry->prev_prio = curr->prio;
419 entry->prev_state = task_state_index(curr);
420 entry->next_pid = wakee->pid;
421 entry->next_prio = wakee->prio;
422 entry->next_state = task_state_index(wakee);
423 entry->next_cpu = task_cpu(wakee);
425 if (!call_filter_check_discard(call, entry, buffer, event))
426 trace_buffer_unlock_commit(tr, buffer, event, trace_ctx);
430 probe_wakeup_sched_switch(void *ignore, bool preempt,
431 struct task_struct *prev, struct task_struct *next,
432 unsigned int prev_state)
434 struct trace_array_cpu *data;
439 unsigned int trace_ctx;
441 tracing_record_cmdline(prev);
443 if (unlikely(!tracer_enabled))
447 * When we start a new trace, we set wakeup_task to NULL
448 * and then set tracer_enabled = 1. We want to make sure
449 * that another CPU does not see the tracer_enabled = 1
450 * and the wakeup_task with an older task, that might
451 * actually be the same as next.
455 if (next != wakeup_task)
458 /* disable local data, not wakeup_cpu data */
459 cpu = raw_smp_processor_id();
460 disabled = atomic_inc_return(&per_cpu_ptr(wakeup_trace->array_buffer.data, cpu)->disabled);
461 if (likely(disabled != 1))
464 local_irq_save(flags);
465 trace_ctx = tracing_gen_ctx_flags(flags);
467 arch_spin_lock(&wakeup_lock);
469 /* We could race with grabbing wakeup_lock */
470 if (unlikely(!tracer_enabled || next != wakeup_task))
473 /* The task we are waiting for is waking up */
474 data = per_cpu_ptr(wakeup_trace->array_buffer.data, wakeup_cpu);
476 __trace_function(wakeup_trace, CALLER_ADDR0, CALLER_ADDR1, trace_ctx);
477 tracing_sched_switch_trace(wakeup_trace, prev, next, trace_ctx);
478 __trace_stack(wakeup_trace, trace_ctx, 0);
480 T0 = data->preempt_timestamp;
481 T1 = ftrace_now(cpu);
484 if (!report_latency(wakeup_trace, delta))
487 if (likely(!is_tracing_stopped())) {
488 wakeup_trace->max_latency = delta;
489 update_max_tr(wakeup_trace, wakeup_task, wakeup_cpu, NULL);
493 __wakeup_reset(wakeup_trace);
494 arch_spin_unlock(&wakeup_lock);
495 local_irq_restore(flags);
497 atomic_dec(&per_cpu_ptr(wakeup_trace->array_buffer.data, cpu)->disabled);
500 static void __wakeup_reset(struct trace_array *tr)
507 put_task_struct(wakeup_task);
512 static void wakeup_reset(struct trace_array *tr)
516 tracing_reset_online_cpus(&tr->array_buffer);
518 local_irq_save(flags);
519 arch_spin_lock(&wakeup_lock);
521 arch_spin_unlock(&wakeup_lock);
522 local_irq_restore(flags);
526 probe_wakeup(void *ignore, struct task_struct *p)
528 struct trace_array_cpu *data;
529 int cpu = smp_processor_id();
531 unsigned int trace_ctx;
533 if (likely(!tracer_enabled))
536 tracing_record_cmdline(p);
537 tracing_record_cmdline(current);
540 * Semantic is like this:
541 * - wakeup tracer handles all tasks in the system, independently
542 * from their scheduling class;
543 * - wakeup_rt tracer handles tasks belonging to sched_dl and
545 * - wakeup_dl handles tasks belonging to sched_dl class only.
547 if (tracing_dl || (wakeup_dl && !dl_task(p)) ||
548 (wakeup_rt && !dl_task(p) && !rt_task(p)) ||
549 (!dl_task(p) && (p->prio >= wakeup_prio || p->prio >= current->prio)))
552 disabled = atomic_inc_return(&per_cpu_ptr(wakeup_trace->array_buffer.data, cpu)->disabled);
553 if (unlikely(disabled != 1))
556 trace_ctx = tracing_gen_ctx();
558 /* interrupts should be off from try_to_wake_up */
559 arch_spin_lock(&wakeup_lock);
561 /* check for races. */
562 if (!tracer_enabled || tracing_dl ||
563 (!dl_task(p) && p->prio >= wakeup_prio))
566 /* reset the trace */
567 __wakeup_reset(wakeup_trace);
569 wakeup_cpu = task_cpu(p);
570 wakeup_current_cpu = wakeup_cpu;
571 wakeup_prio = p->prio;
574 * Once you start tracing a -deadline task, don't bother tracing
575 * another task until the first one wakes up.
582 wakeup_task = get_task_struct(p);
584 data = per_cpu_ptr(wakeup_trace->array_buffer.data, wakeup_cpu);
585 data->preempt_timestamp = ftrace_now(cpu);
586 tracing_sched_wakeup_trace(wakeup_trace, p, current, trace_ctx);
587 __trace_stack(wakeup_trace, trace_ctx, 0);
590 * We must be careful in using CALLER_ADDR2. But since wake_up
591 * is not called by an assembly function (where as schedule is)
592 * it should be safe to use it here.
594 __trace_function(wakeup_trace, CALLER_ADDR1, CALLER_ADDR2, trace_ctx);
597 arch_spin_unlock(&wakeup_lock);
599 atomic_dec(&per_cpu_ptr(wakeup_trace->array_buffer.data, cpu)->disabled);
602 static void start_wakeup_tracer(struct trace_array *tr)
606 ret = register_trace_sched_wakeup(probe_wakeup, NULL);
608 pr_info("wakeup trace: Couldn't activate tracepoint"
609 " probe to kernel_sched_wakeup\n");
613 ret = register_trace_sched_wakeup_new(probe_wakeup, NULL);
615 pr_info("wakeup trace: Couldn't activate tracepoint"
616 " probe to kernel_sched_wakeup_new\n");
620 ret = register_trace_sched_switch(probe_wakeup_sched_switch, NULL);
622 pr_info("sched trace: Couldn't activate tracepoint"
623 " probe to kernel_sched_switch\n");
624 goto fail_deprobe_wake_new;
627 ret = register_trace_sched_migrate_task(probe_wakeup_migrate_task, NULL);
629 pr_info("wakeup trace: Couldn't activate tracepoint"
630 " probe to kernel_sched_migrate_task\n");
631 goto fail_deprobe_sched_switch;
637 * Don't let the tracer_enabled = 1 show up before
638 * the wakeup_task is reset. This may be overkill since
639 * wakeup_reset does a spin_unlock after setting the
640 * wakeup_task to NULL, but I want to be safe.
641 * This is a slow path anyway.
645 if (start_func_tracer(tr, is_graph(tr)))
646 printk(KERN_ERR "failed to start wakeup tracer\n");
649 fail_deprobe_sched_switch:
650 unregister_trace_sched_switch(probe_wakeup_sched_switch, NULL);
651 fail_deprobe_wake_new:
652 unregister_trace_sched_wakeup_new(probe_wakeup, NULL);
654 unregister_trace_sched_wakeup(probe_wakeup, NULL);
657 static void stop_wakeup_tracer(struct trace_array *tr)
660 stop_func_tracer(tr, is_graph(tr));
661 unregister_trace_sched_switch(probe_wakeup_sched_switch, NULL);
662 unregister_trace_sched_wakeup_new(probe_wakeup, NULL);
663 unregister_trace_sched_wakeup(probe_wakeup, NULL);
664 unregister_trace_sched_migrate_task(probe_wakeup_migrate_task, NULL);
667 static bool wakeup_busy;
669 static int __wakeup_tracer_init(struct trace_array *tr)
671 save_flags = tr->trace_flags;
673 /* non overwrite screws up the latency tracers */
674 set_tracer_flag(tr, TRACE_ITER_OVERWRITE, 1);
675 set_tracer_flag(tr, TRACE_ITER_LATENCY_FMT, 1);
679 ftrace_init_array_ops(tr, wakeup_tracer_call);
680 start_wakeup_tracer(tr);
686 static int wakeup_tracer_init(struct trace_array *tr)
693 return __wakeup_tracer_init(tr);
696 static int wakeup_rt_tracer_init(struct trace_array *tr)
703 return __wakeup_tracer_init(tr);
706 static int wakeup_dl_tracer_init(struct trace_array *tr)
713 return __wakeup_tracer_init(tr);
716 static void wakeup_tracer_reset(struct trace_array *tr)
718 int lat_flag = save_flags & TRACE_ITER_LATENCY_FMT;
719 int overwrite_flag = save_flags & TRACE_ITER_OVERWRITE;
721 stop_wakeup_tracer(tr);
722 /* make sure we put back any tasks we are tracing */
725 set_tracer_flag(tr, TRACE_ITER_LATENCY_FMT, lat_flag);
726 set_tracer_flag(tr, TRACE_ITER_OVERWRITE, overwrite_flag);
727 ftrace_reset_array_ops(tr);
731 static void wakeup_tracer_start(struct trace_array *tr)
737 static void wakeup_tracer_stop(struct trace_array *tr)
742 static struct tracer wakeup_tracer __read_mostly =
745 .init = wakeup_tracer_init,
746 .reset = wakeup_tracer_reset,
747 .start = wakeup_tracer_start,
748 .stop = wakeup_tracer_stop,
750 .print_header = wakeup_print_header,
751 .print_line = wakeup_print_line,
752 .flag_changed = wakeup_flag_changed,
753 #ifdef CONFIG_FTRACE_SELFTEST
754 .selftest = trace_selftest_startup_wakeup,
756 .open = wakeup_trace_open,
757 .close = wakeup_trace_close,
758 .allow_instances = true,
762 static struct tracer wakeup_rt_tracer __read_mostly =
765 .init = wakeup_rt_tracer_init,
766 .reset = wakeup_tracer_reset,
767 .start = wakeup_tracer_start,
768 .stop = wakeup_tracer_stop,
770 .print_header = wakeup_print_header,
771 .print_line = wakeup_print_line,
772 .flag_changed = wakeup_flag_changed,
773 #ifdef CONFIG_FTRACE_SELFTEST
774 .selftest = trace_selftest_startup_wakeup,
776 .open = wakeup_trace_open,
777 .close = wakeup_trace_close,
778 .allow_instances = true,
782 static struct tracer wakeup_dl_tracer __read_mostly =
785 .init = wakeup_dl_tracer_init,
786 .reset = wakeup_tracer_reset,
787 .start = wakeup_tracer_start,
788 .stop = wakeup_tracer_stop,
790 .print_header = wakeup_print_header,
791 .print_line = wakeup_print_line,
792 .flag_changed = wakeup_flag_changed,
793 #ifdef CONFIG_FTRACE_SELFTEST
794 .selftest = trace_selftest_startup_wakeup,
796 .open = wakeup_trace_open,
797 .close = wakeup_trace_close,
798 .allow_instances = true,
802 __init static int init_wakeup_tracer(void)
806 ret = register_tracer(&wakeup_tracer);
810 ret = register_tracer(&wakeup_rt_tracer);
814 ret = register_tracer(&wakeup_dl_tracer);
820 core_initcall(init_wakeup_tracer);