xref: /linux/kernel/trace/trace_irqsoff.c (revision 60e13231561b3a4c5269bfa1ef6c0569ad6f28ec)
1 /*
2  * trace irqs off critical timings
3  *
4  * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5  * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
6  *
7  * From code in the latency_tracer, that is:
8  *
9  *  Copyright (C) 2004-2006 Ingo Molnar
10  *  Copyright (C) 2004 William Lee Irwin III
11  */
12 #include <linux/kallsyms.h>
13 #include <linux/debugfs.h>
14 #include <linux/uaccess.h>
15 #include <linux/module.h>
16 #include <linux/ftrace.h>
17 #include <linux/fs.h>
18 
19 #include "trace.h"
20 
21 static struct trace_array		*irqsoff_trace __read_mostly;
22 static int				tracer_enabled __read_mostly;
23 
24 static DEFINE_PER_CPU(int, tracing_cpu);
25 
26 static DEFINE_SPINLOCK(max_trace_lock);
27 
28 enum {
29 	TRACER_IRQS_OFF		= (1 << 1),
30 	TRACER_PREEMPT_OFF	= (1 << 2),
31 };
32 
33 static int trace_type __read_mostly;
34 
35 static int save_lat_flag;
36 
37 static void stop_irqsoff_tracer(struct trace_array *tr, int graph);
38 static int start_irqsoff_tracer(struct trace_array *tr, int graph);
39 
40 #ifdef CONFIG_PREEMPT_TRACER
41 static inline int
42 preempt_trace(void)
43 {
44 	return ((trace_type & TRACER_PREEMPT_OFF) && preempt_count());
45 }
46 #else
47 # define preempt_trace() (0)
48 #endif
49 
50 #ifdef CONFIG_IRQSOFF_TRACER
51 static inline int
52 irq_trace(void)
53 {
54 	return ((trace_type & TRACER_IRQS_OFF) &&
55 		irqs_disabled());
56 }
57 #else
58 # define irq_trace() (0)
59 #endif
60 
61 #define TRACE_DISPLAY_GRAPH	1
62 
63 static struct tracer_opt trace_opts[] = {
64 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
65 	/* display latency trace as call graph */
66 	{ TRACER_OPT(display-graph, TRACE_DISPLAY_GRAPH) },
67 #endif
68 	{ } /* Empty entry */
69 };
70 
71 static struct tracer_flags tracer_flags = {
72 	.val  = 0,
73 	.opts = trace_opts,
74 };
75 
76 #define is_graph() (tracer_flags.val & TRACE_DISPLAY_GRAPH)
77 
78 /*
79  * Sequence count - we record it when starting a measurement and
80  * skip the latency if the sequence has changed - some other section
81  * did a maximum and could disturb our measurement with serial console
82  * printouts, etc. Truly coinciding maximum latencies should be rare
83  * and what happens together happens separately as well, so this doesn't
84  * decrease the validity of the maximum found:
85  */
86 static __cacheline_aligned_in_smp	unsigned long max_sequence;
87 
88 #ifdef CONFIG_FUNCTION_TRACER
89 /*
90  * Prologue for the preempt and irqs off function tracers.
91  *
92  * Returns 1 if it is OK to continue, and data->disabled is
93  *            incremented.
94  *         0 if the trace is to be ignored, and data->disabled
95  *            is kept the same.
96  *
97  * Note, this function is also used outside this ifdef but
98  *  inside the #ifdef of the function graph tracer below.
99  *  This is OK, since the function graph tracer is
100  *  dependent on the function tracer.
101  */
102 static int func_prolog_dec(struct trace_array *tr,
103 			   struct trace_array_cpu **data,
104 			   unsigned long *flags)
105 {
106 	long disabled;
107 	int cpu;
108 
109 	/*
110 	 * Does not matter if we preempt. We test the flags
111 	 * afterward, to see if irqs are disabled or not.
112 	 * If we preempt and get a false positive, the flags
113 	 * test will fail.
114 	 */
115 	cpu = raw_smp_processor_id();
116 	if (likely(!per_cpu(tracing_cpu, cpu)))
117 		return 0;
118 
119 	local_save_flags(*flags);
120 	/* slight chance to get a false positive on tracing_cpu */
121 	if (!irqs_disabled_flags(*flags))
122 		return 0;
123 
124 	*data = tr->data[cpu];
125 	disabled = atomic_inc_return(&(*data)->disabled);
126 
127 	if (likely(disabled == 1))
128 		return 1;
129 
130 	atomic_dec(&(*data)->disabled);
131 
132 	return 0;
133 }
134 
135 /*
136  * irqsoff uses its own tracer function to keep the overhead down:
137  */
138 static void
139 irqsoff_tracer_call(unsigned long ip, unsigned long parent_ip)
140 {
141 	struct trace_array *tr = irqsoff_trace;
142 	struct trace_array_cpu *data;
143 	unsigned long flags;
144 
145 	if (!func_prolog_dec(tr, &data, &flags))
146 		return;
147 
148 	trace_function(tr, ip, parent_ip, flags, preempt_count());
149 
150 	atomic_dec(&data->disabled);
151 }
152 
153 static struct ftrace_ops trace_ops __read_mostly =
154 {
155 	.func = irqsoff_tracer_call,
156 	.flags = FTRACE_OPS_FL_GLOBAL,
157 };
158 #endif /* CONFIG_FUNCTION_TRACER */
159 
160 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
161 static int irqsoff_set_flag(u32 old_flags, u32 bit, int set)
162 {
163 	int cpu;
164 
165 	if (!(bit & TRACE_DISPLAY_GRAPH))
166 		return -EINVAL;
167 
168 	if (!(is_graph() ^ set))
169 		return 0;
170 
171 	stop_irqsoff_tracer(irqsoff_trace, !set);
172 
173 	for_each_possible_cpu(cpu)
174 		per_cpu(tracing_cpu, cpu) = 0;
175 
176 	tracing_max_latency = 0;
177 	tracing_reset_online_cpus(irqsoff_trace);
178 
179 	return start_irqsoff_tracer(irqsoff_trace, set);
180 }
181 
182 static int irqsoff_graph_entry(struct ftrace_graph_ent *trace)
183 {
184 	struct trace_array *tr = irqsoff_trace;
185 	struct trace_array_cpu *data;
186 	unsigned long flags;
187 	int ret;
188 	int pc;
189 
190 	if (!func_prolog_dec(tr, &data, &flags))
191 		return 0;
192 
193 	pc = preempt_count();
194 	ret = __trace_graph_entry(tr, trace, flags, pc);
195 	atomic_dec(&data->disabled);
196 
197 	return ret;
198 }
199 
200 static void irqsoff_graph_return(struct ftrace_graph_ret *trace)
201 {
202 	struct trace_array *tr = irqsoff_trace;
203 	struct trace_array_cpu *data;
204 	unsigned long flags;
205 	int pc;
206 
207 	if (!func_prolog_dec(tr, &data, &flags))
208 		return;
209 
210 	pc = preempt_count();
211 	__trace_graph_return(tr, trace, flags, pc);
212 	atomic_dec(&data->disabled);
213 }
214 
215 static void irqsoff_trace_open(struct trace_iterator *iter)
216 {
217 	if (is_graph())
218 		graph_trace_open(iter);
219 
220 }
221 
222 static void irqsoff_trace_close(struct trace_iterator *iter)
223 {
224 	if (iter->private)
225 		graph_trace_close(iter);
226 }
227 
228 #define GRAPH_TRACER_FLAGS (TRACE_GRAPH_PRINT_CPU | \
229 			    TRACE_GRAPH_PRINT_PROC | \
230 			    TRACE_GRAPH_PRINT_ABS_TIME | \
231 			    TRACE_GRAPH_PRINT_DURATION)
232 
233 static enum print_line_t irqsoff_print_line(struct trace_iterator *iter)
234 {
235 	/*
236 	 * In graph mode call the graph tracer output function,
237 	 * otherwise go with the TRACE_FN event handler
238 	 */
239 	if (is_graph())
240 		return print_graph_function_flags(iter, GRAPH_TRACER_FLAGS);
241 
242 	return TRACE_TYPE_UNHANDLED;
243 }
244 
245 static void irqsoff_print_header(struct seq_file *s)
246 {
247 	if (is_graph())
248 		print_graph_headers_flags(s, GRAPH_TRACER_FLAGS);
249 	else
250 		trace_default_header(s);
251 }
252 
253 static void
254 __trace_function(struct trace_array *tr,
255 		 unsigned long ip, unsigned long parent_ip,
256 		 unsigned long flags, int pc)
257 {
258 	if (is_graph())
259 		trace_graph_function(tr, ip, parent_ip, flags, pc);
260 	else
261 		trace_function(tr, ip, parent_ip, flags, pc);
262 }
263 
264 #else
265 #define __trace_function trace_function
266 
267 static int irqsoff_set_flag(u32 old_flags, u32 bit, int set)
268 {
269 	return -EINVAL;
270 }
271 
272 static int irqsoff_graph_entry(struct ftrace_graph_ent *trace)
273 {
274 	return -1;
275 }
276 
277 static enum print_line_t irqsoff_print_line(struct trace_iterator *iter)
278 {
279 	return TRACE_TYPE_UNHANDLED;
280 }
281 
282 static void irqsoff_graph_return(struct ftrace_graph_ret *trace) { }
283 static void irqsoff_print_header(struct seq_file *s) { }
284 static void irqsoff_trace_open(struct trace_iterator *iter) { }
285 static void irqsoff_trace_close(struct trace_iterator *iter) { }
286 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
287 
288 /*
289  * Should this new latency be reported/recorded?
290  */
291 static int report_latency(cycle_t delta)
292 {
293 	if (tracing_thresh) {
294 		if (delta < tracing_thresh)
295 			return 0;
296 	} else {
297 		if (delta <= tracing_max_latency)
298 			return 0;
299 	}
300 	return 1;
301 }
302 
303 static void
304 check_critical_timing(struct trace_array *tr,
305 		      struct trace_array_cpu *data,
306 		      unsigned long parent_ip,
307 		      int cpu)
308 {
309 	cycle_t T0, T1, delta;
310 	unsigned long flags;
311 	int pc;
312 
313 	T0 = data->preempt_timestamp;
314 	T1 = ftrace_now(cpu);
315 	delta = T1-T0;
316 
317 	local_save_flags(flags);
318 
319 	pc = preempt_count();
320 
321 	if (!report_latency(delta))
322 		goto out;
323 
324 	spin_lock_irqsave(&max_trace_lock, flags);
325 
326 	/* check if we are still the max latency */
327 	if (!report_latency(delta))
328 		goto out_unlock;
329 
330 	__trace_function(tr, CALLER_ADDR0, parent_ip, flags, pc);
331 	/* Skip 5 functions to get to the irq/preempt enable function */
332 	__trace_stack(tr, flags, 5, pc);
333 
334 	if (data->critical_sequence != max_sequence)
335 		goto out_unlock;
336 
337 	data->critical_end = parent_ip;
338 
339 	if (likely(!is_tracing_stopped())) {
340 		tracing_max_latency = delta;
341 		update_max_tr_single(tr, current, cpu);
342 	}
343 
344 	max_sequence++;
345 
346 out_unlock:
347 	spin_unlock_irqrestore(&max_trace_lock, flags);
348 
349 out:
350 	data->critical_sequence = max_sequence;
351 	data->preempt_timestamp = ftrace_now(cpu);
352 	__trace_function(tr, CALLER_ADDR0, parent_ip, flags, pc);
353 }
354 
355 static inline void
356 start_critical_timing(unsigned long ip, unsigned long parent_ip)
357 {
358 	int cpu;
359 	struct trace_array *tr = irqsoff_trace;
360 	struct trace_array_cpu *data;
361 	unsigned long flags;
362 
363 	if (likely(!tracer_enabled))
364 		return;
365 
366 	cpu = raw_smp_processor_id();
367 
368 	if (per_cpu(tracing_cpu, cpu))
369 		return;
370 
371 	data = tr->data[cpu];
372 
373 	if (unlikely(!data) || atomic_read(&data->disabled))
374 		return;
375 
376 	atomic_inc(&data->disabled);
377 
378 	data->critical_sequence = max_sequence;
379 	data->preempt_timestamp = ftrace_now(cpu);
380 	data->critical_start = parent_ip ? : ip;
381 
382 	local_save_flags(flags);
383 
384 	__trace_function(tr, ip, parent_ip, flags, preempt_count());
385 
386 	per_cpu(tracing_cpu, cpu) = 1;
387 
388 	atomic_dec(&data->disabled);
389 }
390 
391 static inline void
392 stop_critical_timing(unsigned long ip, unsigned long parent_ip)
393 {
394 	int cpu;
395 	struct trace_array *tr = irqsoff_trace;
396 	struct trace_array_cpu *data;
397 	unsigned long flags;
398 
399 	cpu = raw_smp_processor_id();
400 	/* Always clear the tracing cpu on stopping the trace */
401 	if (unlikely(per_cpu(tracing_cpu, cpu)))
402 		per_cpu(tracing_cpu, cpu) = 0;
403 	else
404 		return;
405 
406 	if (!tracer_enabled)
407 		return;
408 
409 	data = tr->data[cpu];
410 
411 	if (unlikely(!data) ||
412 	    !data->critical_start || atomic_read(&data->disabled))
413 		return;
414 
415 	atomic_inc(&data->disabled);
416 
417 	local_save_flags(flags);
418 	__trace_function(tr, ip, parent_ip, flags, preempt_count());
419 	check_critical_timing(tr, data, parent_ip ? : ip, cpu);
420 	data->critical_start = 0;
421 	atomic_dec(&data->disabled);
422 }
423 
424 /* start and stop critical timings used to for stoppage (in idle) */
425 void start_critical_timings(void)
426 {
427 	if (preempt_trace() || irq_trace())
428 		start_critical_timing(CALLER_ADDR0, CALLER_ADDR1);
429 }
430 EXPORT_SYMBOL_GPL(start_critical_timings);
431 
432 void stop_critical_timings(void)
433 {
434 	if (preempt_trace() || irq_trace())
435 		stop_critical_timing(CALLER_ADDR0, CALLER_ADDR1);
436 }
437 EXPORT_SYMBOL_GPL(stop_critical_timings);
438 
439 #ifdef CONFIG_IRQSOFF_TRACER
440 #ifdef CONFIG_PROVE_LOCKING
441 void time_hardirqs_on(unsigned long a0, unsigned long a1)
442 {
443 	if (!preempt_trace() && irq_trace())
444 		stop_critical_timing(a0, a1);
445 }
446 
447 void time_hardirqs_off(unsigned long a0, unsigned long a1)
448 {
449 	if (!preempt_trace() && irq_trace())
450 		start_critical_timing(a0, a1);
451 }
452 
453 #else /* !CONFIG_PROVE_LOCKING */
454 
455 /*
456  * Stubs:
457  */
458 
459 void trace_softirqs_on(unsigned long ip)
460 {
461 }
462 
463 void trace_softirqs_off(unsigned long ip)
464 {
465 }
466 
467 inline void print_irqtrace_events(struct task_struct *curr)
468 {
469 }
470 
471 /*
472  * We are only interested in hardirq on/off events:
473  */
474 void trace_hardirqs_on(void)
475 {
476 	if (!preempt_trace() && irq_trace())
477 		stop_critical_timing(CALLER_ADDR0, CALLER_ADDR1);
478 }
479 EXPORT_SYMBOL(trace_hardirqs_on);
480 
481 void trace_hardirqs_off(void)
482 {
483 	if (!preempt_trace() && irq_trace())
484 		start_critical_timing(CALLER_ADDR0, CALLER_ADDR1);
485 }
486 EXPORT_SYMBOL(trace_hardirqs_off);
487 
488 void trace_hardirqs_on_caller(unsigned long caller_addr)
489 {
490 	if (!preempt_trace() && irq_trace())
491 		stop_critical_timing(CALLER_ADDR0, caller_addr);
492 }
493 EXPORT_SYMBOL(trace_hardirqs_on_caller);
494 
495 void trace_hardirqs_off_caller(unsigned long caller_addr)
496 {
497 	if (!preempt_trace() && irq_trace())
498 		start_critical_timing(CALLER_ADDR0, caller_addr);
499 }
500 EXPORT_SYMBOL(trace_hardirqs_off_caller);
501 
502 #endif /* CONFIG_PROVE_LOCKING */
503 #endif /*  CONFIG_IRQSOFF_TRACER */
504 
505 #ifdef CONFIG_PREEMPT_TRACER
506 void trace_preempt_on(unsigned long a0, unsigned long a1)
507 {
508 	if (preempt_trace())
509 		stop_critical_timing(a0, a1);
510 }
511 
512 void trace_preempt_off(unsigned long a0, unsigned long a1)
513 {
514 	if (preempt_trace())
515 		start_critical_timing(a0, a1);
516 }
517 #endif /* CONFIG_PREEMPT_TRACER */
518 
519 static int start_irqsoff_tracer(struct trace_array *tr, int graph)
520 {
521 	int ret = 0;
522 
523 	if (!graph)
524 		ret = register_ftrace_function(&trace_ops);
525 	else
526 		ret = register_ftrace_graph(&irqsoff_graph_return,
527 					    &irqsoff_graph_entry);
528 
529 	if (!ret && tracing_is_enabled())
530 		tracer_enabled = 1;
531 	else
532 		tracer_enabled = 0;
533 
534 	return ret;
535 }
536 
537 static void stop_irqsoff_tracer(struct trace_array *tr, int graph)
538 {
539 	tracer_enabled = 0;
540 
541 	if (!graph)
542 		unregister_ftrace_function(&trace_ops);
543 	else
544 		unregister_ftrace_graph();
545 }
546 
547 static void __irqsoff_tracer_init(struct trace_array *tr)
548 {
549 	save_lat_flag = trace_flags & TRACE_ITER_LATENCY_FMT;
550 	trace_flags |= TRACE_ITER_LATENCY_FMT;
551 
552 	tracing_max_latency = 0;
553 	irqsoff_trace = tr;
554 	/* make sure that the tracer is visible */
555 	smp_wmb();
556 	tracing_reset_online_cpus(tr);
557 
558 	if (start_irqsoff_tracer(tr, is_graph()))
559 		printk(KERN_ERR "failed to start irqsoff tracer\n");
560 }
561 
562 static void irqsoff_tracer_reset(struct trace_array *tr)
563 {
564 	stop_irqsoff_tracer(tr, is_graph());
565 
566 	if (!save_lat_flag)
567 		trace_flags &= ~TRACE_ITER_LATENCY_FMT;
568 }
569 
570 static void irqsoff_tracer_start(struct trace_array *tr)
571 {
572 	tracer_enabled = 1;
573 }
574 
575 static void irqsoff_tracer_stop(struct trace_array *tr)
576 {
577 	tracer_enabled = 0;
578 }
579 
580 #ifdef CONFIG_IRQSOFF_TRACER
581 static int irqsoff_tracer_init(struct trace_array *tr)
582 {
583 	trace_type = TRACER_IRQS_OFF;
584 
585 	__irqsoff_tracer_init(tr);
586 	return 0;
587 }
588 static struct tracer irqsoff_tracer __read_mostly =
589 {
590 	.name		= "irqsoff",
591 	.init		= irqsoff_tracer_init,
592 	.reset		= irqsoff_tracer_reset,
593 	.start		= irqsoff_tracer_start,
594 	.stop		= irqsoff_tracer_stop,
595 	.print_max	= 1,
596 	.print_header   = irqsoff_print_header,
597 	.print_line     = irqsoff_print_line,
598 	.flags		= &tracer_flags,
599 	.set_flag	= irqsoff_set_flag,
600 #ifdef CONFIG_FTRACE_SELFTEST
601 	.selftest    = trace_selftest_startup_irqsoff,
602 #endif
603 	.open           = irqsoff_trace_open,
604 	.close          = irqsoff_trace_close,
605 	.use_max_tr	= 1,
606 };
607 # define register_irqsoff(trace) register_tracer(&trace)
608 #else
609 # define register_irqsoff(trace) do { } while (0)
610 #endif
611 
612 #ifdef CONFIG_PREEMPT_TRACER
613 static int preemptoff_tracer_init(struct trace_array *tr)
614 {
615 	trace_type = TRACER_PREEMPT_OFF;
616 
617 	__irqsoff_tracer_init(tr);
618 	return 0;
619 }
620 
621 static struct tracer preemptoff_tracer __read_mostly =
622 {
623 	.name		= "preemptoff",
624 	.init		= preemptoff_tracer_init,
625 	.reset		= irqsoff_tracer_reset,
626 	.start		= irqsoff_tracer_start,
627 	.stop		= irqsoff_tracer_stop,
628 	.print_max	= 1,
629 	.print_header   = irqsoff_print_header,
630 	.print_line     = irqsoff_print_line,
631 	.flags		= &tracer_flags,
632 	.set_flag	= irqsoff_set_flag,
633 #ifdef CONFIG_FTRACE_SELFTEST
634 	.selftest    = trace_selftest_startup_preemptoff,
635 #endif
636 	.open		= irqsoff_trace_open,
637 	.close		= irqsoff_trace_close,
638 	.use_max_tr	= 1,
639 };
640 # define register_preemptoff(trace) register_tracer(&trace)
641 #else
642 # define register_preemptoff(trace) do { } while (0)
643 #endif
644 
645 #if defined(CONFIG_IRQSOFF_TRACER) && \
646 	defined(CONFIG_PREEMPT_TRACER)
647 
648 static int preemptirqsoff_tracer_init(struct trace_array *tr)
649 {
650 	trace_type = TRACER_IRQS_OFF | TRACER_PREEMPT_OFF;
651 
652 	__irqsoff_tracer_init(tr);
653 	return 0;
654 }
655 
656 static struct tracer preemptirqsoff_tracer __read_mostly =
657 {
658 	.name		= "preemptirqsoff",
659 	.init		= preemptirqsoff_tracer_init,
660 	.reset		= irqsoff_tracer_reset,
661 	.start		= irqsoff_tracer_start,
662 	.stop		= irqsoff_tracer_stop,
663 	.print_max	= 1,
664 	.print_header   = irqsoff_print_header,
665 	.print_line     = irqsoff_print_line,
666 	.flags		= &tracer_flags,
667 	.set_flag	= irqsoff_set_flag,
668 #ifdef CONFIG_FTRACE_SELFTEST
669 	.selftest    = trace_selftest_startup_preemptirqsoff,
670 #endif
671 	.open		= irqsoff_trace_open,
672 	.close		= irqsoff_trace_close,
673 	.use_max_tr	= 1,
674 };
675 
676 # define register_preemptirqsoff(trace) register_tracer(&trace)
677 #else
678 # define register_preemptirqsoff(trace) do { } while (0)
679 #endif
680 
681 __init static int init_irqsoff_tracer(void)
682 {
683 	register_irqsoff(irqsoff_tracer);
684 	register_preemptoff(preemptoff_tracer);
685 	register_preemptirqsoff(preemptirqsoff_tracer);
686 
687 	return 0;
688 }
689 device_initcall(init_irqsoff_tracer);
690