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