xref: /linux/kernel/trace/trace_functions_graph.c (revision 5499b45190237ca90dd2ac86395cf464fe1f4cc7)
1 /*
2  *
3  * Function graph tracer.
4  * Copyright (c) 2008-2009 Frederic Weisbecker <fweisbec@gmail.com>
5  * Mostly borrowed from function tracer which
6  * is Copyright (c) Steven Rostedt <srostedt@redhat.com>
7  *
8  */
9 #include <linux/debugfs.h>
10 #include <linux/uaccess.h>
11 #include <linux/ftrace.h>
12 #include <linux/fs.h>
13 
14 #include "trace.h"
15 #include "trace_output.h"
16 
17 struct fgraph_cpu_data {
18 	pid_t		last_pid;
19 	int		depth;
20 	int		ignore;
21 	unsigned long	enter_funcs[FTRACE_RETFUNC_DEPTH];
22 };
23 
24 struct fgraph_data {
25 	struct fgraph_cpu_data		*cpu_data;
26 
27 	/* Place to preserve last processed entry. */
28 	struct ftrace_graph_ent_entry	ent;
29 	struct ftrace_graph_ret_entry	ret;
30 	int				failed;
31 	int				cpu;
32 };
33 
34 #define TRACE_GRAPH_INDENT	2
35 
36 /* Flag options */
37 #define TRACE_GRAPH_PRINT_OVERRUN	0x1
38 #define TRACE_GRAPH_PRINT_CPU		0x2
39 #define TRACE_GRAPH_PRINT_OVERHEAD	0x4
40 #define TRACE_GRAPH_PRINT_PROC		0x8
41 #define TRACE_GRAPH_PRINT_DURATION	0x10
42 #define TRACE_GRAPH_PRINT_ABS_TIME	0X20
43 
44 static struct tracer_opt trace_opts[] = {
45 	/* Display overruns? (for self-debug purpose) */
46 	{ TRACER_OPT(funcgraph-overrun, TRACE_GRAPH_PRINT_OVERRUN) },
47 	/* Display CPU ? */
48 	{ TRACER_OPT(funcgraph-cpu, TRACE_GRAPH_PRINT_CPU) },
49 	/* Display Overhead ? */
50 	{ TRACER_OPT(funcgraph-overhead, TRACE_GRAPH_PRINT_OVERHEAD) },
51 	/* Display proc name/pid */
52 	{ TRACER_OPT(funcgraph-proc, TRACE_GRAPH_PRINT_PROC) },
53 	/* Display duration of execution */
54 	{ TRACER_OPT(funcgraph-duration, TRACE_GRAPH_PRINT_DURATION) },
55 	/* Display absolute time of an entry */
56 	{ TRACER_OPT(funcgraph-abstime, TRACE_GRAPH_PRINT_ABS_TIME) },
57 	{ } /* Empty entry */
58 };
59 
60 static struct tracer_flags tracer_flags = {
61 	/* Don't display overruns and proc by default */
62 	.val = TRACE_GRAPH_PRINT_CPU | TRACE_GRAPH_PRINT_OVERHEAD |
63 	       TRACE_GRAPH_PRINT_DURATION,
64 	.opts = trace_opts
65 };
66 
67 static struct trace_array *graph_array;
68 
69 
70 /* Add a function return address to the trace stack on thread info.*/
71 int
72 ftrace_push_return_trace(unsigned long ret, unsigned long func, int *depth,
73 			 unsigned long frame_pointer)
74 {
75 	unsigned long long calltime;
76 	int index;
77 
78 	if (!current->ret_stack)
79 		return -EBUSY;
80 
81 	/*
82 	 * We must make sure the ret_stack is tested before we read
83 	 * anything else.
84 	 */
85 	smp_rmb();
86 
87 	/* The return trace stack is full */
88 	if (current->curr_ret_stack == FTRACE_RETFUNC_DEPTH - 1) {
89 		atomic_inc(&current->trace_overrun);
90 		return -EBUSY;
91 	}
92 
93 	calltime = trace_clock_local();
94 
95 	index = ++current->curr_ret_stack;
96 	barrier();
97 	current->ret_stack[index].ret = ret;
98 	current->ret_stack[index].func = func;
99 	current->ret_stack[index].calltime = calltime;
100 	current->ret_stack[index].subtime = 0;
101 	current->ret_stack[index].fp = frame_pointer;
102 	*depth = index;
103 
104 	return 0;
105 }
106 
107 /* Retrieve a function return address to the trace stack on thread info.*/
108 static void
109 ftrace_pop_return_trace(struct ftrace_graph_ret *trace, unsigned long *ret,
110 			unsigned long frame_pointer)
111 {
112 	int index;
113 
114 	index = current->curr_ret_stack;
115 
116 	if (unlikely(index < 0)) {
117 		ftrace_graph_stop();
118 		WARN_ON(1);
119 		/* Might as well panic, otherwise we have no where to go */
120 		*ret = (unsigned long)panic;
121 		return;
122 	}
123 
124 #ifdef CONFIG_HAVE_FUNCTION_GRAPH_FP_TEST
125 	/*
126 	 * The arch may choose to record the frame pointer used
127 	 * and check it here to make sure that it is what we expect it
128 	 * to be. If gcc does not set the place holder of the return
129 	 * address in the frame pointer, and does a copy instead, then
130 	 * the function graph trace will fail. This test detects this
131 	 * case.
132 	 *
133 	 * Currently, x86_32 with optimize for size (-Os) makes the latest
134 	 * gcc do the above.
135 	 */
136 	if (unlikely(current->ret_stack[index].fp != frame_pointer)) {
137 		ftrace_graph_stop();
138 		WARN(1, "Bad frame pointer: expected %lx, received %lx\n"
139 		     "  from func %ps return to %lx\n",
140 		     current->ret_stack[index].fp,
141 		     frame_pointer,
142 		     (void *)current->ret_stack[index].func,
143 		     current->ret_stack[index].ret);
144 		*ret = (unsigned long)panic;
145 		return;
146 	}
147 #endif
148 
149 	*ret = current->ret_stack[index].ret;
150 	trace->func = current->ret_stack[index].func;
151 	trace->calltime = current->ret_stack[index].calltime;
152 	trace->overrun = atomic_read(&current->trace_overrun);
153 	trace->depth = index;
154 }
155 
156 /*
157  * Send the trace to the ring-buffer.
158  * @return the original return address.
159  */
160 unsigned long ftrace_return_to_handler(unsigned long frame_pointer)
161 {
162 	struct ftrace_graph_ret trace;
163 	unsigned long ret;
164 
165 	ftrace_pop_return_trace(&trace, &ret, frame_pointer);
166 	trace.rettime = trace_clock_local();
167 	ftrace_graph_return(&trace);
168 	barrier();
169 	current->curr_ret_stack--;
170 
171 	if (unlikely(!ret)) {
172 		ftrace_graph_stop();
173 		WARN_ON(1);
174 		/* Might as well panic. What else to do? */
175 		ret = (unsigned long)panic;
176 	}
177 
178 	return ret;
179 }
180 
181 static int __trace_graph_entry(struct trace_array *tr,
182 				struct ftrace_graph_ent *trace,
183 				unsigned long flags,
184 				int pc)
185 {
186 	struct ftrace_event_call *call = &event_funcgraph_entry;
187 	struct ring_buffer_event *event;
188 	struct ring_buffer *buffer = tr->buffer;
189 	struct ftrace_graph_ent_entry *entry;
190 
191 	if (unlikely(__this_cpu_read(per_cpu_var(ftrace_cpu_disabled))))
192 		return 0;
193 
194 	event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_ENT,
195 					  sizeof(*entry), flags, pc);
196 	if (!event)
197 		return 0;
198 	entry	= ring_buffer_event_data(event);
199 	entry->graph_ent			= *trace;
200 	if (!filter_current_check_discard(buffer, call, entry, event))
201 		ring_buffer_unlock_commit(buffer, event);
202 
203 	return 1;
204 }
205 
206 int trace_graph_entry(struct ftrace_graph_ent *trace)
207 {
208 	struct trace_array *tr = graph_array;
209 	struct trace_array_cpu *data;
210 	unsigned long flags;
211 	long disabled;
212 	int ret;
213 	int cpu;
214 	int pc;
215 
216 	if (!ftrace_trace_task(current))
217 		return 0;
218 
219 	/* trace it when it is-nested-in or is a function enabled. */
220 	if (!(trace->depth || ftrace_graph_addr(trace->func)))
221 		return 0;
222 
223 	local_irq_save(flags);
224 	cpu = raw_smp_processor_id();
225 	data = tr->data[cpu];
226 	disabled = atomic_inc_return(&data->disabled);
227 	if (likely(disabled == 1)) {
228 		pc = preempt_count();
229 		ret = __trace_graph_entry(tr, trace, flags, pc);
230 	} else {
231 		ret = 0;
232 	}
233 
234 	atomic_dec(&data->disabled);
235 	local_irq_restore(flags);
236 
237 	return ret;
238 }
239 
240 static void __trace_graph_return(struct trace_array *tr,
241 				struct ftrace_graph_ret *trace,
242 				unsigned long flags,
243 				int pc)
244 {
245 	struct ftrace_event_call *call = &event_funcgraph_exit;
246 	struct ring_buffer_event *event;
247 	struct ring_buffer *buffer = tr->buffer;
248 	struct ftrace_graph_ret_entry *entry;
249 
250 	if (unlikely(__this_cpu_read(per_cpu_var(ftrace_cpu_disabled))))
251 		return;
252 
253 	event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_RET,
254 					  sizeof(*entry), flags, pc);
255 	if (!event)
256 		return;
257 	entry	= ring_buffer_event_data(event);
258 	entry->ret				= *trace;
259 	if (!filter_current_check_discard(buffer, call, entry, event))
260 		ring_buffer_unlock_commit(buffer, event);
261 }
262 
263 void trace_graph_return(struct ftrace_graph_ret *trace)
264 {
265 	struct trace_array *tr = graph_array;
266 	struct trace_array_cpu *data;
267 	unsigned long flags;
268 	long disabled;
269 	int cpu;
270 	int pc;
271 
272 	local_irq_save(flags);
273 	cpu = raw_smp_processor_id();
274 	data = tr->data[cpu];
275 	disabled = atomic_inc_return(&data->disabled);
276 	if (likely(disabled == 1)) {
277 		pc = preempt_count();
278 		__trace_graph_return(tr, trace, flags, pc);
279 	}
280 	atomic_dec(&data->disabled);
281 	local_irq_restore(flags);
282 }
283 
284 void set_graph_array(struct trace_array *tr)
285 {
286 	graph_array = tr;
287 
288 	/* Make graph_array visible before we start tracing */
289 
290 	smp_mb();
291 }
292 
293 static int graph_trace_init(struct trace_array *tr)
294 {
295 	int ret;
296 
297 	set_graph_array(tr);
298 	ret = register_ftrace_graph(&trace_graph_return,
299 				    &trace_graph_entry);
300 	if (ret)
301 		return ret;
302 	tracing_start_cmdline_record();
303 
304 	return 0;
305 }
306 
307 static void graph_trace_reset(struct trace_array *tr)
308 {
309 	tracing_stop_cmdline_record();
310 	unregister_ftrace_graph();
311 }
312 
313 static int max_bytes_for_cpu;
314 
315 static enum print_line_t
316 print_graph_cpu(struct trace_seq *s, int cpu)
317 {
318 	int ret;
319 
320 	/*
321 	 * Start with a space character - to make it stand out
322 	 * to the right a bit when trace output is pasted into
323 	 * email:
324 	 */
325 	ret = trace_seq_printf(s, " %*d) ", max_bytes_for_cpu, cpu);
326 	if (!ret)
327 		return TRACE_TYPE_PARTIAL_LINE;
328 
329 	return TRACE_TYPE_HANDLED;
330 }
331 
332 #define TRACE_GRAPH_PROCINFO_LENGTH	14
333 
334 static enum print_line_t
335 print_graph_proc(struct trace_seq *s, pid_t pid)
336 {
337 	char comm[TASK_COMM_LEN];
338 	/* sign + log10(MAX_INT) + '\0' */
339 	char pid_str[11];
340 	int spaces = 0;
341 	int ret;
342 	int len;
343 	int i;
344 
345 	trace_find_cmdline(pid, comm);
346 	comm[7] = '\0';
347 	sprintf(pid_str, "%d", pid);
348 
349 	/* 1 stands for the "-" character */
350 	len = strlen(comm) + strlen(pid_str) + 1;
351 
352 	if (len < TRACE_GRAPH_PROCINFO_LENGTH)
353 		spaces = TRACE_GRAPH_PROCINFO_LENGTH - len;
354 
355 	/* First spaces to align center */
356 	for (i = 0; i < spaces / 2; i++) {
357 		ret = trace_seq_printf(s, " ");
358 		if (!ret)
359 			return TRACE_TYPE_PARTIAL_LINE;
360 	}
361 
362 	ret = trace_seq_printf(s, "%s-%s", comm, pid_str);
363 	if (!ret)
364 		return TRACE_TYPE_PARTIAL_LINE;
365 
366 	/* Last spaces to align center */
367 	for (i = 0; i < spaces - (spaces / 2); i++) {
368 		ret = trace_seq_printf(s, " ");
369 		if (!ret)
370 			return TRACE_TYPE_PARTIAL_LINE;
371 	}
372 	return TRACE_TYPE_HANDLED;
373 }
374 
375 
376 static enum print_line_t
377 print_graph_lat_fmt(struct trace_seq *s, struct trace_entry *entry)
378 {
379 	if (!trace_seq_putc(s, ' '))
380 		return 0;
381 
382 	return trace_print_lat_fmt(s, entry);
383 }
384 
385 /* If the pid changed since the last trace, output this event */
386 static enum print_line_t
387 verif_pid(struct trace_seq *s, pid_t pid, int cpu, struct fgraph_data *data)
388 {
389 	pid_t prev_pid;
390 	pid_t *last_pid;
391 	int ret;
392 
393 	if (!data)
394 		return TRACE_TYPE_HANDLED;
395 
396 	last_pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid);
397 
398 	if (*last_pid == pid)
399 		return TRACE_TYPE_HANDLED;
400 
401 	prev_pid = *last_pid;
402 	*last_pid = pid;
403 
404 	if (prev_pid == -1)
405 		return TRACE_TYPE_HANDLED;
406 /*
407  * Context-switch trace line:
408 
409  ------------------------------------------
410  | 1)  migration/0--1  =>  sshd-1755
411  ------------------------------------------
412 
413  */
414 	ret = trace_seq_printf(s,
415 		" ------------------------------------------\n");
416 	if (!ret)
417 		return TRACE_TYPE_PARTIAL_LINE;
418 
419 	ret = print_graph_cpu(s, cpu);
420 	if (ret == TRACE_TYPE_PARTIAL_LINE)
421 		return TRACE_TYPE_PARTIAL_LINE;
422 
423 	ret = print_graph_proc(s, prev_pid);
424 	if (ret == TRACE_TYPE_PARTIAL_LINE)
425 		return TRACE_TYPE_PARTIAL_LINE;
426 
427 	ret = trace_seq_printf(s, " => ");
428 	if (!ret)
429 		return TRACE_TYPE_PARTIAL_LINE;
430 
431 	ret = print_graph_proc(s, pid);
432 	if (ret == TRACE_TYPE_PARTIAL_LINE)
433 		return TRACE_TYPE_PARTIAL_LINE;
434 
435 	ret = trace_seq_printf(s,
436 		"\n ------------------------------------------\n\n");
437 	if (!ret)
438 		return TRACE_TYPE_PARTIAL_LINE;
439 
440 	return TRACE_TYPE_HANDLED;
441 }
442 
443 static struct ftrace_graph_ret_entry *
444 get_return_for_leaf(struct trace_iterator *iter,
445 		struct ftrace_graph_ent_entry *curr)
446 {
447 	struct fgraph_data *data = iter->private;
448 	struct ring_buffer_iter *ring_iter = NULL;
449 	struct ring_buffer_event *event;
450 	struct ftrace_graph_ret_entry *next;
451 
452 	/*
453 	 * If the previous output failed to write to the seq buffer,
454 	 * then we just reuse the data from before.
455 	 */
456 	if (data && data->failed) {
457 		curr = &data->ent;
458 		next = &data->ret;
459 	} else {
460 
461 		ring_iter = iter->buffer_iter[iter->cpu];
462 
463 		/* First peek to compare current entry and the next one */
464 		if (ring_iter)
465 			event = ring_buffer_iter_peek(ring_iter, NULL);
466 		else {
467 			/*
468 			 * We need to consume the current entry to see
469 			 * the next one.
470 			 */
471 			ring_buffer_consume(iter->tr->buffer, iter->cpu, NULL);
472 			event = ring_buffer_peek(iter->tr->buffer, iter->cpu,
473 						 NULL);
474 		}
475 
476 		if (!event)
477 			return NULL;
478 
479 		next = ring_buffer_event_data(event);
480 
481 		if (data) {
482 			/*
483 			 * Save current and next entries for later reference
484 			 * if the output fails.
485 			 */
486 			data->ent = *curr;
487 			data->ret = *next;
488 		}
489 	}
490 
491 	if (next->ent.type != TRACE_GRAPH_RET)
492 		return NULL;
493 
494 	if (curr->ent.pid != next->ent.pid ||
495 			curr->graph_ent.func != next->ret.func)
496 		return NULL;
497 
498 	/* this is a leaf, now advance the iterator */
499 	if (ring_iter)
500 		ring_buffer_read(ring_iter, NULL);
501 
502 	return next;
503 }
504 
505 /* Signal a overhead of time execution to the output */
506 static int
507 print_graph_overhead(unsigned long long duration, struct trace_seq *s)
508 {
509 	/* If duration disappear, we don't need anything */
510 	if (!(tracer_flags.val & TRACE_GRAPH_PRINT_DURATION))
511 		return 1;
512 
513 	/* Non nested entry or return */
514 	if (duration == -1)
515 		return trace_seq_printf(s, "  ");
516 
517 	if (tracer_flags.val & TRACE_GRAPH_PRINT_OVERHEAD) {
518 		/* Duration exceeded 100 msecs */
519 		if (duration > 100000ULL)
520 			return trace_seq_printf(s, "! ");
521 
522 		/* Duration exceeded 10 msecs */
523 		if (duration > 10000ULL)
524 			return trace_seq_printf(s, "+ ");
525 	}
526 
527 	return trace_seq_printf(s, "  ");
528 }
529 
530 static int print_graph_abs_time(u64 t, struct trace_seq *s)
531 {
532 	unsigned long usecs_rem;
533 
534 	usecs_rem = do_div(t, NSEC_PER_SEC);
535 	usecs_rem /= 1000;
536 
537 	return trace_seq_printf(s, "%5lu.%06lu |  ",
538 			(unsigned long)t, usecs_rem);
539 }
540 
541 static enum print_line_t
542 print_graph_irq(struct trace_iterator *iter, unsigned long addr,
543 		enum trace_type type, int cpu, pid_t pid)
544 {
545 	int ret;
546 	struct trace_seq *s = &iter->seq;
547 
548 	if (addr < (unsigned long)__irqentry_text_start ||
549 		addr >= (unsigned long)__irqentry_text_end)
550 		return TRACE_TYPE_UNHANDLED;
551 
552 	/* Absolute time */
553 	if (tracer_flags.val & TRACE_GRAPH_PRINT_ABS_TIME) {
554 		ret = print_graph_abs_time(iter->ts, s);
555 		if (!ret)
556 			return TRACE_TYPE_PARTIAL_LINE;
557 	}
558 
559 	/* Cpu */
560 	if (tracer_flags.val & TRACE_GRAPH_PRINT_CPU) {
561 		ret = print_graph_cpu(s, cpu);
562 		if (ret == TRACE_TYPE_PARTIAL_LINE)
563 			return TRACE_TYPE_PARTIAL_LINE;
564 	}
565 
566 	/* Proc */
567 	if (tracer_flags.val & TRACE_GRAPH_PRINT_PROC) {
568 		ret = print_graph_proc(s, pid);
569 		if (ret == TRACE_TYPE_PARTIAL_LINE)
570 			return TRACE_TYPE_PARTIAL_LINE;
571 		ret = trace_seq_printf(s, " | ");
572 		if (!ret)
573 			return TRACE_TYPE_PARTIAL_LINE;
574 	}
575 
576 	/* No overhead */
577 	ret = print_graph_overhead(-1, s);
578 	if (!ret)
579 		return TRACE_TYPE_PARTIAL_LINE;
580 
581 	if (type == TRACE_GRAPH_ENT)
582 		ret = trace_seq_printf(s, "==========>");
583 	else
584 		ret = trace_seq_printf(s, "<==========");
585 
586 	if (!ret)
587 		return TRACE_TYPE_PARTIAL_LINE;
588 
589 	/* Don't close the duration column if haven't one */
590 	if (tracer_flags.val & TRACE_GRAPH_PRINT_DURATION)
591 		trace_seq_printf(s, " |");
592 	ret = trace_seq_printf(s, "\n");
593 
594 	if (!ret)
595 		return TRACE_TYPE_PARTIAL_LINE;
596 	return TRACE_TYPE_HANDLED;
597 }
598 
599 enum print_line_t
600 trace_print_graph_duration(unsigned long long duration, struct trace_seq *s)
601 {
602 	unsigned long nsecs_rem = do_div(duration, 1000);
603 	/* log10(ULONG_MAX) + '\0' */
604 	char msecs_str[21];
605 	char nsecs_str[5];
606 	int ret, len;
607 	int i;
608 
609 	sprintf(msecs_str, "%lu", (unsigned long) duration);
610 
611 	/* Print msecs */
612 	ret = trace_seq_printf(s, "%s", msecs_str);
613 	if (!ret)
614 		return TRACE_TYPE_PARTIAL_LINE;
615 
616 	len = strlen(msecs_str);
617 
618 	/* Print nsecs (we don't want to exceed 7 numbers) */
619 	if (len < 7) {
620 		snprintf(nsecs_str, 8 - len, "%03lu", nsecs_rem);
621 		ret = trace_seq_printf(s, ".%s", nsecs_str);
622 		if (!ret)
623 			return TRACE_TYPE_PARTIAL_LINE;
624 		len += strlen(nsecs_str);
625 	}
626 
627 	ret = trace_seq_printf(s, " us ");
628 	if (!ret)
629 		return TRACE_TYPE_PARTIAL_LINE;
630 
631 	/* Print remaining spaces to fit the row's width */
632 	for (i = len; i < 7; i++) {
633 		ret = trace_seq_printf(s, " ");
634 		if (!ret)
635 			return TRACE_TYPE_PARTIAL_LINE;
636 	}
637 	return TRACE_TYPE_HANDLED;
638 }
639 
640 static enum print_line_t
641 print_graph_duration(unsigned long long duration, struct trace_seq *s)
642 {
643 	int ret;
644 
645 	ret = trace_print_graph_duration(duration, s);
646 	if (ret != TRACE_TYPE_HANDLED)
647 		return ret;
648 
649 	ret = trace_seq_printf(s, "|  ");
650 	if (!ret)
651 		return TRACE_TYPE_PARTIAL_LINE;
652 
653 	return TRACE_TYPE_HANDLED;
654 }
655 
656 /* Case of a leaf function on its call entry */
657 static enum print_line_t
658 print_graph_entry_leaf(struct trace_iterator *iter,
659 		struct ftrace_graph_ent_entry *entry,
660 		struct ftrace_graph_ret_entry *ret_entry, struct trace_seq *s)
661 {
662 	struct fgraph_data *data = iter->private;
663 	struct ftrace_graph_ret *graph_ret;
664 	struct ftrace_graph_ent *call;
665 	unsigned long long duration;
666 	int ret;
667 	int i;
668 
669 	graph_ret = &ret_entry->ret;
670 	call = &entry->graph_ent;
671 	duration = graph_ret->rettime - graph_ret->calltime;
672 
673 	if (data) {
674 		struct fgraph_cpu_data *cpu_data;
675 		int cpu = iter->cpu;
676 
677 		cpu_data = per_cpu_ptr(data->cpu_data, cpu);
678 
679 		/*
680 		 * Comments display at + 1 to depth. Since
681 		 * this is a leaf function, keep the comments
682 		 * equal to this depth.
683 		 */
684 		cpu_data->depth = call->depth - 1;
685 
686 		/* No need to keep this function around for this depth */
687 		if (call->depth < FTRACE_RETFUNC_DEPTH)
688 			cpu_data->enter_funcs[call->depth] = 0;
689 	}
690 
691 	/* Overhead */
692 	ret = print_graph_overhead(duration, s);
693 	if (!ret)
694 		return TRACE_TYPE_PARTIAL_LINE;
695 
696 	/* Duration */
697 	if (tracer_flags.val & TRACE_GRAPH_PRINT_DURATION) {
698 		ret = print_graph_duration(duration, s);
699 		if (ret == TRACE_TYPE_PARTIAL_LINE)
700 			return TRACE_TYPE_PARTIAL_LINE;
701 	}
702 
703 	/* Function */
704 	for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++) {
705 		ret = trace_seq_printf(s, " ");
706 		if (!ret)
707 			return TRACE_TYPE_PARTIAL_LINE;
708 	}
709 
710 	ret = trace_seq_printf(s, "%ps();\n", (void *)call->func);
711 	if (!ret)
712 		return TRACE_TYPE_PARTIAL_LINE;
713 
714 	return TRACE_TYPE_HANDLED;
715 }
716 
717 static enum print_line_t
718 print_graph_entry_nested(struct trace_iterator *iter,
719 			 struct ftrace_graph_ent_entry *entry,
720 			 struct trace_seq *s, int cpu)
721 {
722 	struct ftrace_graph_ent *call = &entry->graph_ent;
723 	struct fgraph_data *data = iter->private;
724 	int ret;
725 	int i;
726 
727 	if (data) {
728 		struct fgraph_cpu_data *cpu_data;
729 		int cpu = iter->cpu;
730 
731 		cpu_data = per_cpu_ptr(data->cpu_data, cpu);
732 		cpu_data->depth = call->depth;
733 
734 		/* Save this function pointer to see if the exit matches */
735 		if (call->depth < FTRACE_RETFUNC_DEPTH)
736 			cpu_data->enter_funcs[call->depth] = call->func;
737 	}
738 
739 	/* No overhead */
740 	ret = print_graph_overhead(-1, s);
741 	if (!ret)
742 		return TRACE_TYPE_PARTIAL_LINE;
743 
744 	/* No time */
745 	if (tracer_flags.val & TRACE_GRAPH_PRINT_DURATION) {
746 		ret = trace_seq_printf(s, "            |  ");
747 		if (!ret)
748 			return TRACE_TYPE_PARTIAL_LINE;
749 	}
750 
751 	/* Function */
752 	for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++) {
753 		ret = trace_seq_printf(s, " ");
754 		if (!ret)
755 			return TRACE_TYPE_PARTIAL_LINE;
756 	}
757 
758 	ret = trace_seq_printf(s, "%ps() {\n", (void *)call->func);
759 	if (!ret)
760 		return TRACE_TYPE_PARTIAL_LINE;
761 
762 	/*
763 	 * we already consumed the current entry to check the next one
764 	 * and see if this is a leaf.
765 	 */
766 	return TRACE_TYPE_NO_CONSUME;
767 }
768 
769 static enum print_line_t
770 print_graph_prologue(struct trace_iterator *iter, struct trace_seq *s,
771 		     int type, unsigned long addr)
772 {
773 	struct fgraph_data *data = iter->private;
774 	struct trace_entry *ent = iter->ent;
775 	int cpu = iter->cpu;
776 	int ret;
777 
778 	/* Pid */
779 	if (verif_pid(s, ent->pid, cpu, data) == TRACE_TYPE_PARTIAL_LINE)
780 		return TRACE_TYPE_PARTIAL_LINE;
781 
782 	if (type) {
783 		/* Interrupt */
784 		ret = print_graph_irq(iter, addr, type, cpu, ent->pid);
785 		if (ret == TRACE_TYPE_PARTIAL_LINE)
786 			return TRACE_TYPE_PARTIAL_LINE;
787 	}
788 
789 	/* Absolute time */
790 	if (tracer_flags.val & TRACE_GRAPH_PRINT_ABS_TIME) {
791 		ret = print_graph_abs_time(iter->ts, s);
792 		if (!ret)
793 			return TRACE_TYPE_PARTIAL_LINE;
794 	}
795 
796 	/* Cpu */
797 	if (tracer_flags.val & TRACE_GRAPH_PRINT_CPU) {
798 		ret = print_graph_cpu(s, cpu);
799 		if (ret == TRACE_TYPE_PARTIAL_LINE)
800 			return TRACE_TYPE_PARTIAL_LINE;
801 	}
802 
803 	/* Proc */
804 	if (tracer_flags.val & TRACE_GRAPH_PRINT_PROC) {
805 		ret = print_graph_proc(s, ent->pid);
806 		if (ret == TRACE_TYPE_PARTIAL_LINE)
807 			return TRACE_TYPE_PARTIAL_LINE;
808 
809 		ret = trace_seq_printf(s, " | ");
810 		if (!ret)
811 			return TRACE_TYPE_PARTIAL_LINE;
812 	}
813 
814 	/* Latency format */
815 	if (trace_flags & TRACE_ITER_LATENCY_FMT) {
816 		ret = print_graph_lat_fmt(s, ent);
817 		if (ret == TRACE_TYPE_PARTIAL_LINE)
818 			return TRACE_TYPE_PARTIAL_LINE;
819 	}
820 
821 	return 0;
822 }
823 
824 static enum print_line_t
825 print_graph_entry(struct ftrace_graph_ent_entry *field, struct trace_seq *s,
826 			struct trace_iterator *iter)
827 {
828 	struct fgraph_data *data = iter->private;
829 	struct ftrace_graph_ent *call = &field->graph_ent;
830 	struct ftrace_graph_ret_entry *leaf_ret;
831 	static enum print_line_t ret;
832 	int cpu = iter->cpu;
833 
834 	if (print_graph_prologue(iter, s, TRACE_GRAPH_ENT, call->func))
835 		return TRACE_TYPE_PARTIAL_LINE;
836 
837 	leaf_ret = get_return_for_leaf(iter, field);
838 	if (leaf_ret)
839 		ret = print_graph_entry_leaf(iter, field, leaf_ret, s);
840 	else
841 		ret = print_graph_entry_nested(iter, field, s, cpu);
842 
843 	if (data) {
844 		/*
845 		 * If we failed to write our output, then we need to make
846 		 * note of it. Because we already consumed our entry.
847 		 */
848 		if (s->full) {
849 			data->failed = 1;
850 			data->cpu = cpu;
851 		} else
852 			data->failed = 0;
853 	}
854 
855 	return ret;
856 }
857 
858 static enum print_line_t
859 print_graph_return(struct ftrace_graph_ret *trace, struct trace_seq *s,
860 		   struct trace_entry *ent, struct trace_iterator *iter)
861 {
862 	unsigned long long duration = trace->rettime - trace->calltime;
863 	struct fgraph_data *data = iter->private;
864 	pid_t pid = ent->pid;
865 	int cpu = iter->cpu;
866 	int func_match = 1;
867 	int ret;
868 	int i;
869 
870 	if (data) {
871 		struct fgraph_cpu_data *cpu_data;
872 		int cpu = iter->cpu;
873 
874 		cpu_data = per_cpu_ptr(data->cpu_data, cpu);
875 
876 		/*
877 		 * Comments display at + 1 to depth. This is the
878 		 * return from a function, we now want the comments
879 		 * to display at the same level of the bracket.
880 		 */
881 		cpu_data->depth = trace->depth - 1;
882 
883 		if (trace->depth < FTRACE_RETFUNC_DEPTH) {
884 			if (cpu_data->enter_funcs[trace->depth] != trace->func)
885 				func_match = 0;
886 			cpu_data->enter_funcs[trace->depth] = 0;
887 		}
888 	}
889 
890 	if (print_graph_prologue(iter, s, 0, 0))
891 		return TRACE_TYPE_PARTIAL_LINE;
892 
893 	/* Overhead */
894 	ret = print_graph_overhead(duration, s);
895 	if (!ret)
896 		return TRACE_TYPE_PARTIAL_LINE;
897 
898 	/* Duration */
899 	if (tracer_flags.val & TRACE_GRAPH_PRINT_DURATION) {
900 		ret = print_graph_duration(duration, s);
901 		if (ret == TRACE_TYPE_PARTIAL_LINE)
902 			return TRACE_TYPE_PARTIAL_LINE;
903 	}
904 
905 	/* Closing brace */
906 	for (i = 0; i < trace->depth * TRACE_GRAPH_INDENT; i++) {
907 		ret = trace_seq_printf(s, " ");
908 		if (!ret)
909 			return TRACE_TYPE_PARTIAL_LINE;
910 	}
911 
912 	/*
913 	 * If the return function does not have a matching entry,
914 	 * then the entry was lost. Instead of just printing
915 	 * the '}' and letting the user guess what function this
916 	 * belongs to, write out the function name.
917 	 */
918 	if (func_match) {
919 		ret = trace_seq_printf(s, "}\n");
920 		if (!ret)
921 			return TRACE_TYPE_PARTIAL_LINE;
922 	} else {
923 		ret = trace_seq_printf(s, "} (%ps)\n", (void *)trace->func);
924 		if (!ret)
925 			return TRACE_TYPE_PARTIAL_LINE;
926 	}
927 
928 	/* Overrun */
929 	if (tracer_flags.val & TRACE_GRAPH_PRINT_OVERRUN) {
930 		ret = trace_seq_printf(s, " (Overruns: %lu)\n",
931 					trace->overrun);
932 		if (!ret)
933 			return TRACE_TYPE_PARTIAL_LINE;
934 	}
935 
936 	ret = print_graph_irq(iter, trace->func, TRACE_GRAPH_RET, cpu, pid);
937 	if (ret == TRACE_TYPE_PARTIAL_LINE)
938 		return TRACE_TYPE_PARTIAL_LINE;
939 
940 	return TRACE_TYPE_HANDLED;
941 }
942 
943 static enum print_line_t
944 print_graph_comment(struct trace_seq *s,  struct trace_entry *ent,
945 		    struct trace_iterator *iter)
946 {
947 	unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
948 	struct fgraph_data *data = iter->private;
949 	struct trace_event *event;
950 	int depth = 0;
951 	int ret;
952 	int i;
953 
954 	if (data)
955 		depth = per_cpu_ptr(data->cpu_data, iter->cpu)->depth;
956 
957 	if (print_graph_prologue(iter, s, 0, 0))
958 		return TRACE_TYPE_PARTIAL_LINE;
959 
960 	/* No overhead */
961 	ret = print_graph_overhead(-1, s);
962 	if (!ret)
963 		return TRACE_TYPE_PARTIAL_LINE;
964 
965 	/* No time */
966 	if (tracer_flags.val & TRACE_GRAPH_PRINT_DURATION) {
967 		ret = trace_seq_printf(s, "            |  ");
968 		if (!ret)
969 			return TRACE_TYPE_PARTIAL_LINE;
970 	}
971 
972 	/* Indentation */
973 	if (depth > 0)
974 		for (i = 0; i < (depth + 1) * TRACE_GRAPH_INDENT; i++) {
975 			ret = trace_seq_printf(s, " ");
976 			if (!ret)
977 				return TRACE_TYPE_PARTIAL_LINE;
978 		}
979 
980 	/* The comment */
981 	ret = trace_seq_printf(s, "/* ");
982 	if (!ret)
983 		return TRACE_TYPE_PARTIAL_LINE;
984 
985 	switch (iter->ent->type) {
986 	case TRACE_BPRINT:
987 		ret = trace_print_bprintk_msg_only(iter);
988 		if (ret != TRACE_TYPE_HANDLED)
989 			return ret;
990 		break;
991 	case TRACE_PRINT:
992 		ret = trace_print_printk_msg_only(iter);
993 		if (ret != TRACE_TYPE_HANDLED)
994 			return ret;
995 		break;
996 	default:
997 		event = ftrace_find_event(ent->type);
998 		if (!event)
999 			return TRACE_TYPE_UNHANDLED;
1000 
1001 		ret = event->trace(iter, sym_flags);
1002 		if (ret != TRACE_TYPE_HANDLED)
1003 			return ret;
1004 	}
1005 
1006 	/* Strip ending newline */
1007 	if (s->buffer[s->len - 1] == '\n') {
1008 		s->buffer[s->len - 1] = '\0';
1009 		s->len--;
1010 	}
1011 
1012 	ret = trace_seq_printf(s, " */\n");
1013 	if (!ret)
1014 		return TRACE_TYPE_PARTIAL_LINE;
1015 
1016 	return TRACE_TYPE_HANDLED;
1017 }
1018 
1019 
1020 enum print_line_t
1021 print_graph_function(struct trace_iterator *iter)
1022 {
1023 	struct ftrace_graph_ent_entry *field;
1024 	struct fgraph_data *data = iter->private;
1025 	struct trace_entry *entry = iter->ent;
1026 	struct trace_seq *s = &iter->seq;
1027 	int cpu = iter->cpu;
1028 	int ret;
1029 
1030 	if (data && per_cpu_ptr(data->cpu_data, cpu)->ignore) {
1031 		per_cpu_ptr(data->cpu_data, cpu)->ignore = 0;
1032 		return TRACE_TYPE_HANDLED;
1033 	}
1034 
1035 	/*
1036 	 * If the last output failed, there's a possibility we need
1037 	 * to print out the missing entry which would never go out.
1038 	 */
1039 	if (data && data->failed) {
1040 		field = &data->ent;
1041 		iter->cpu = data->cpu;
1042 		ret = print_graph_entry(field, s, iter);
1043 		if (ret == TRACE_TYPE_HANDLED && iter->cpu != cpu) {
1044 			per_cpu_ptr(data->cpu_data, iter->cpu)->ignore = 1;
1045 			ret = TRACE_TYPE_NO_CONSUME;
1046 		}
1047 		iter->cpu = cpu;
1048 		return ret;
1049 	}
1050 
1051 	switch (entry->type) {
1052 	case TRACE_GRAPH_ENT: {
1053 		/*
1054 		 * print_graph_entry() may consume the current event,
1055 		 * thus @field may become invalid, so we need to save it.
1056 		 * sizeof(struct ftrace_graph_ent_entry) is very small,
1057 		 * it can be safely saved at the stack.
1058 		 */
1059 		struct ftrace_graph_ent_entry saved;
1060 		trace_assign_type(field, entry);
1061 		saved = *field;
1062 		return print_graph_entry(&saved, s, iter);
1063 	}
1064 	case TRACE_GRAPH_RET: {
1065 		struct ftrace_graph_ret_entry *field;
1066 		trace_assign_type(field, entry);
1067 		return print_graph_return(&field->ret, s, entry, iter);
1068 	}
1069 	default:
1070 		return print_graph_comment(s, entry, iter);
1071 	}
1072 
1073 	return TRACE_TYPE_HANDLED;
1074 }
1075 
1076 static void print_lat_header(struct seq_file *s)
1077 {
1078 	static const char spaces[] = "                "	/* 16 spaces */
1079 		"    "					/* 4 spaces */
1080 		"                 ";			/* 17 spaces */
1081 	int size = 0;
1082 
1083 	if (tracer_flags.val & TRACE_GRAPH_PRINT_ABS_TIME)
1084 		size += 16;
1085 	if (tracer_flags.val & TRACE_GRAPH_PRINT_CPU)
1086 		size += 4;
1087 	if (tracer_flags.val & TRACE_GRAPH_PRINT_PROC)
1088 		size += 17;
1089 
1090 	seq_printf(s, "#%.*s  _-----=> irqs-off        \n", size, spaces);
1091 	seq_printf(s, "#%.*s / _----=> need-resched    \n", size, spaces);
1092 	seq_printf(s, "#%.*s| / _---=> hardirq/softirq \n", size, spaces);
1093 	seq_printf(s, "#%.*s|| / _--=> preempt-depth   \n", size, spaces);
1094 	seq_printf(s, "#%.*s||| / _-=> lock-depth      \n", size, spaces);
1095 	seq_printf(s, "#%.*s|||| /                     \n", size, spaces);
1096 }
1097 
1098 static void print_graph_headers(struct seq_file *s)
1099 {
1100 	int lat = trace_flags & TRACE_ITER_LATENCY_FMT;
1101 
1102 	if (lat)
1103 		print_lat_header(s);
1104 
1105 	/* 1st line */
1106 	seq_printf(s, "#");
1107 	if (tracer_flags.val & TRACE_GRAPH_PRINT_ABS_TIME)
1108 		seq_printf(s, "     TIME       ");
1109 	if (tracer_flags.val & TRACE_GRAPH_PRINT_CPU)
1110 		seq_printf(s, " CPU");
1111 	if (tracer_flags.val & TRACE_GRAPH_PRINT_PROC)
1112 		seq_printf(s, "  TASK/PID       ");
1113 	if (lat)
1114 		seq_printf(s, "|||||");
1115 	if (tracer_flags.val & TRACE_GRAPH_PRINT_DURATION)
1116 		seq_printf(s, "  DURATION   ");
1117 	seq_printf(s, "               FUNCTION CALLS\n");
1118 
1119 	/* 2nd line */
1120 	seq_printf(s, "#");
1121 	if (tracer_flags.val & TRACE_GRAPH_PRINT_ABS_TIME)
1122 		seq_printf(s, "      |         ");
1123 	if (tracer_flags.val & TRACE_GRAPH_PRINT_CPU)
1124 		seq_printf(s, " |  ");
1125 	if (tracer_flags.val & TRACE_GRAPH_PRINT_PROC)
1126 		seq_printf(s, "   |    |        ");
1127 	if (lat)
1128 		seq_printf(s, "|||||");
1129 	if (tracer_flags.val & TRACE_GRAPH_PRINT_DURATION)
1130 		seq_printf(s, "   |   |      ");
1131 	seq_printf(s, "               |   |   |   |\n");
1132 }
1133 
1134 static void graph_trace_open(struct trace_iterator *iter)
1135 {
1136 	/* pid and depth on the last trace processed */
1137 	struct fgraph_data *data;
1138 	int cpu;
1139 
1140 	iter->private = NULL;
1141 
1142 	data = kzalloc(sizeof(*data), GFP_KERNEL);
1143 	if (!data)
1144 		goto out_err;
1145 
1146 	data->cpu_data = alloc_percpu(struct fgraph_cpu_data);
1147 	if (!data->cpu_data)
1148 		goto out_err_free;
1149 
1150 	for_each_possible_cpu(cpu) {
1151 		pid_t *pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid);
1152 		int *depth = &(per_cpu_ptr(data->cpu_data, cpu)->depth);
1153 		int *ignore = &(per_cpu_ptr(data->cpu_data, cpu)->ignore);
1154 		*pid = -1;
1155 		*depth = 0;
1156 		*ignore = 0;
1157 	}
1158 
1159 	iter->private = data;
1160 
1161 	return;
1162 
1163  out_err_free:
1164 	kfree(data);
1165  out_err:
1166 	pr_warning("function graph tracer: not enough memory\n");
1167 }
1168 
1169 static void graph_trace_close(struct trace_iterator *iter)
1170 {
1171 	struct fgraph_data *data = iter->private;
1172 
1173 	if (data) {
1174 		free_percpu(data->cpu_data);
1175 		kfree(data);
1176 	}
1177 }
1178 
1179 static struct tracer graph_trace __read_mostly = {
1180 	.name		= "function_graph",
1181 	.open		= graph_trace_open,
1182 	.pipe_open	= graph_trace_open,
1183 	.close		= graph_trace_close,
1184 	.pipe_close	= graph_trace_close,
1185 	.wait_pipe	= poll_wait_pipe,
1186 	.init		= graph_trace_init,
1187 	.reset		= graph_trace_reset,
1188 	.print_line	= print_graph_function,
1189 	.print_header	= print_graph_headers,
1190 	.flags		= &tracer_flags,
1191 #ifdef CONFIG_FTRACE_SELFTEST
1192 	.selftest	= trace_selftest_startup_function_graph,
1193 #endif
1194 };
1195 
1196 static __init int init_graph_trace(void)
1197 {
1198 	max_bytes_for_cpu = snprintf(NULL, 0, "%d", nr_cpu_ids - 1);
1199 
1200 	return register_tracer(&graph_trace);
1201 }
1202 
1203 device_initcall(init_graph_trace);
1204