1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *
4  * Function graph tracer.
5  * Copyright (c) 2008-2009 Frederic Weisbecker <[email protected]>
6  * Mostly borrowed from function tracer which
7  * is Copyright (c) Steven Rostedt <[email protected]>
8  *
9  */
10 #include <linux/uaccess.h>
11 #include <linux/ftrace.h>
12 #include <linux/interrupt.h>
13 #include <linux/slab.h>
14 #include <linux/fs.h>
15 
16 #include "trace.h"
17 #include "trace_output.h"
18 
19 /* When set, irq functions will be ignored */
20 static int ftrace_graph_skip_irqs;
21 
22 struct fgraph_cpu_data {
23 	pid_t		last_pid;
24 	int		depth;
25 	int		depth_irq;
26 	int		ignore;
27 	unsigned long	enter_funcs[FTRACE_RETFUNC_DEPTH];
28 };
29 
30 struct fgraph_data {
31 	struct fgraph_cpu_data __percpu *cpu_data;
32 
33 	/* Place to preserve last processed entry. */
34 	union {
35 		struct ftrace_graph_ent_entry	ent;
36 		struct fgraph_retaddr_ent_entry	rent;
37 	} ent;
38 	struct ftrace_graph_ret_entry	ret;
39 	int				failed;
40 	int				cpu;
41 };
42 
43 #define TRACE_GRAPH_INDENT	2
44 
45 unsigned int fgraph_max_depth;
46 
47 static struct tracer_opt trace_opts[] = {
48 	/* Display overruns? (for self-debug purpose) */
49 	{ TRACER_OPT(funcgraph-overrun, TRACE_GRAPH_PRINT_OVERRUN) },
50 	/* Display CPU ? */
51 	{ TRACER_OPT(funcgraph-cpu, TRACE_GRAPH_PRINT_CPU) },
52 	/* Display Overhead ? */
53 	{ TRACER_OPT(funcgraph-overhead, TRACE_GRAPH_PRINT_OVERHEAD) },
54 	/* Display proc name/pid */
55 	{ TRACER_OPT(funcgraph-proc, TRACE_GRAPH_PRINT_PROC) },
56 	/* Display duration of execution */
57 	{ TRACER_OPT(funcgraph-duration, TRACE_GRAPH_PRINT_DURATION) },
58 	/* Display absolute time of an entry */
59 	{ TRACER_OPT(funcgraph-abstime, TRACE_GRAPH_PRINT_ABS_TIME) },
60 	/* Display interrupts */
61 	{ TRACER_OPT(funcgraph-irqs, TRACE_GRAPH_PRINT_IRQS) },
62 	/* Display function name after trailing } */
63 	{ TRACER_OPT(funcgraph-tail, TRACE_GRAPH_PRINT_TAIL) },
64 #ifdef CONFIG_FUNCTION_GRAPH_RETVAL
65 	/* Display function return value ? */
66 	{ TRACER_OPT(funcgraph-retval, TRACE_GRAPH_PRINT_RETVAL) },
67 	/* Display function return value in hexadecimal format ? */
68 	{ TRACER_OPT(funcgraph-retval-hex, TRACE_GRAPH_PRINT_RETVAL_HEX) },
69 #endif
70 #ifdef CONFIG_FUNCTION_GRAPH_RETADDR
71 	/* Display function return address ? */
72 	{ TRACER_OPT(funcgraph-retaddr, TRACE_GRAPH_PRINT_RETADDR) },
73 #endif
74 	/* Include sleep time (scheduled out) between entry and return */
75 	{ TRACER_OPT(sleep-time, TRACE_GRAPH_SLEEP_TIME) },
76 
77 #ifdef CONFIG_FUNCTION_PROFILER
78 	/* Include time within nested functions */
79 	{ TRACER_OPT(graph-time, TRACE_GRAPH_GRAPH_TIME) },
80 #endif
81 
82 	{ } /* Empty entry */
83 };
84 
85 static struct tracer_flags tracer_flags = {
86 	/* Don't display overruns, proc, or tail by default */
87 	.val = TRACE_GRAPH_PRINT_CPU | TRACE_GRAPH_PRINT_OVERHEAD |
88 	       TRACE_GRAPH_PRINT_DURATION | TRACE_GRAPH_PRINT_IRQS |
89 	       TRACE_GRAPH_SLEEP_TIME | TRACE_GRAPH_GRAPH_TIME,
90 	.opts = trace_opts
91 };
92 
93 static bool tracer_flags_is_set(u32 flags)
94 {
95 	return (tracer_flags.val & flags) == flags;
96 }
97 
98 /*
99  * DURATION column is being also used to display IRQ signs,
100  * following values are used by print_graph_irq and others
101  * to fill in space into DURATION column.
102  */
103 enum {
104 	FLAGS_FILL_FULL  = 1 << TRACE_GRAPH_PRINT_FILL_SHIFT,
105 	FLAGS_FILL_START = 2 << TRACE_GRAPH_PRINT_FILL_SHIFT,
106 	FLAGS_FILL_END   = 3 << TRACE_GRAPH_PRINT_FILL_SHIFT,
107 };
108 
109 static void
110 print_graph_duration(struct trace_array *tr, unsigned long long duration,
111 		     struct trace_seq *s, u32 flags);
112 
113 int __trace_graph_entry(struct trace_array *tr,
114 				struct ftrace_graph_ent *trace,
115 				unsigned int trace_ctx)
116 {
117 	struct trace_event_call *call = &event_funcgraph_entry;
118 	struct ring_buffer_event *event;
119 	struct trace_buffer *buffer = tr->array_buffer.buffer;
120 	struct ftrace_graph_ent_entry *entry;
121 
122 	event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_ENT,
123 					  sizeof(*entry), trace_ctx);
124 	if (!event)
125 		return 0;
126 	entry	= ring_buffer_event_data(event);
127 	entry->graph_ent			= *trace;
128 	if (!call_filter_check_discard(call, entry, buffer, event))
129 		trace_buffer_unlock_commit_nostack(buffer, event);
130 
131 	return 1;
132 }
133 
134 #ifdef CONFIG_FUNCTION_GRAPH_RETADDR
135 int __trace_graph_retaddr_entry(struct trace_array *tr,
136 				struct ftrace_graph_ent *trace,
137 				unsigned int trace_ctx,
138 				unsigned long retaddr)
139 {
140 	struct trace_event_call *call = &event_fgraph_retaddr_entry;
141 	struct ring_buffer_event *event;
142 	struct trace_buffer *buffer = tr->array_buffer.buffer;
143 	struct fgraph_retaddr_ent_entry *entry;
144 
145 	event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_RETADDR_ENT,
146 					  sizeof(*entry), trace_ctx);
147 	if (!event)
148 		return 0;
149 	entry	= ring_buffer_event_data(event);
150 	entry->graph_ent.func = trace->func;
151 	entry->graph_ent.depth = trace->depth;
152 	entry->graph_ent.retaddr = retaddr;
153 	if (!call_filter_check_discard(call, entry, buffer, event))
154 		trace_buffer_unlock_commit_nostack(buffer, event);
155 
156 	return 1;
157 }
158 #else
159 int __trace_graph_retaddr_entry(struct trace_array *tr,
160 				struct ftrace_graph_ent *trace,
161 				unsigned int trace_ctx,
162 				unsigned long retaddr)
163 {
164 	return 1;
165 }
166 #endif
167 
168 static inline int ftrace_graph_ignore_irqs(void)
169 {
170 	if (!ftrace_graph_skip_irqs || trace_recursion_test(TRACE_IRQ_BIT))
171 		return 0;
172 
173 	return in_hardirq();
174 }
175 
176 struct fgraph_times {
177 	unsigned long long		calltime;
178 	unsigned long long		sleeptime; /* may be optional! */
179 };
180 
181 int trace_graph_entry(struct ftrace_graph_ent *trace,
182 		      struct fgraph_ops *gops)
183 {
184 	unsigned long *task_var = fgraph_get_task_var(gops);
185 	struct trace_array *tr = gops->private;
186 	struct trace_array_cpu *data;
187 	struct fgraph_times *ftimes;
188 	unsigned long flags;
189 	unsigned int trace_ctx;
190 	long disabled;
191 	int ret;
192 	int cpu;
193 
194 	if (*task_var & TRACE_GRAPH_NOTRACE)
195 		return 0;
196 
197 	/*
198 	 * Do not trace a function if it's filtered by set_graph_notrace.
199 	 * Make the index of ret stack negative to indicate that it should
200 	 * ignore further functions.  But it needs its own ret stack entry
201 	 * to recover the original index in order to continue tracing after
202 	 * returning from the function.
203 	 */
204 	if (ftrace_graph_notrace_addr(trace->func)) {
205 		*task_var |= TRACE_GRAPH_NOTRACE_BIT;
206 		/*
207 		 * Need to return 1 to have the return called
208 		 * that will clear the NOTRACE bit.
209 		 */
210 		return 1;
211 	}
212 
213 	if (!ftrace_trace_task(tr))
214 		return 0;
215 
216 	if (ftrace_graph_ignore_func(gops, trace))
217 		return 0;
218 
219 	if (ftrace_graph_ignore_irqs())
220 		return 0;
221 
222 	if (fgraph_sleep_time) {
223 		/* Only need to record the calltime */
224 		ftimes = fgraph_reserve_data(gops->idx, sizeof(ftimes->calltime));
225 	} else {
226 		ftimes = fgraph_reserve_data(gops->idx, sizeof(*ftimes));
227 		if (ftimes)
228 			ftimes->sleeptime = current->ftrace_sleeptime;
229 	}
230 	if (!ftimes)
231 		return 0;
232 
233 	ftimes->calltime = trace_clock_local();
234 
235 	/*
236 	 * Stop here if tracing_threshold is set. We only write function return
237 	 * events to the ring buffer.
238 	 */
239 	if (tracing_thresh)
240 		return 1;
241 
242 	local_irq_save(flags);
243 	cpu = raw_smp_processor_id();
244 	data = per_cpu_ptr(tr->array_buffer.data, cpu);
245 	disabled = atomic_inc_return(&data->disabled);
246 	if (likely(disabled == 1)) {
247 		trace_ctx = tracing_gen_ctx_flags(flags);
248 		if (unlikely(IS_ENABLED(CONFIG_FUNCTION_GRAPH_RETADDR) &&
249 			tracer_flags_is_set(TRACE_GRAPH_PRINT_RETADDR))) {
250 			unsigned long retaddr = ftrace_graph_top_ret_addr(current);
251 
252 			ret = __trace_graph_retaddr_entry(tr, trace, trace_ctx, retaddr);
253 		} else
254 			ret = __trace_graph_entry(tr, trace, trace_ctx);
255 	} else {
256 		ret = 0;
257 	}
258 
259 	atomic_dec(&data->disabled);
260 	local_irq_restore(flags);
261 
262 	return ret;
263 }
264 
265 static void
266 __trace_graph_function(struct trace_array *tr,
267 		unsigned long ip, unsigned int trace_ctx)
268 {
269 	u64 time = trace_clock_local();
270 	struct ftrace_graph_ent ent = {
271 		.func  = ip,
272 		.depth = 0,
273 	};
274 	struct ftrace_graph_ret ret = {
275 		.func     = ip,
276 		.depth    = 0,
277 		.calltime = time,
278 		.rettime  = time,
279 	};
280 
281 	__trace_graph_entry(tr, &ent, trace_ctx);
282 	__trace_graph_return(tr, &ret, trace_ctx);
283 }
284 
285 void
286 trace_graph_function(struct trace_array *tr,
287 		unsigned long ip, unsigned long parent_ip,
288 		unsigned int trace_ctx)
289 {
290 	__trace_graph_function(tr, ip, trace_ctx);
291 }
292 
293 void __trace_graph_return(struct trace_array *tr,
294 				struct ftrace_graph_ret *trace,
295 				unsigned int trace_ctx)
296 {
297 	struct trace_event_call *call = &event_funcgraph_exit;
298 	struct ring_buffer_event *event;
299 	struct trace_buffer *buffer = tr->array_buffer.buffer;
300 	struct ftrace_graph_ret_entry *entry;
301 
302 	event = trace_buffer_lock_reserve(buffer, TRACE_GRAPH_RET,
303 					  sizeof(*entry), trace_ctx);
304 	if (!event)
305 		return;
306 	entry	= ring_buffer_event_data(event);
307 	entry->ret				= *trace;
308 	if (!call_filter_check_discard(call, entry, buffer, event))
309 		trace_buffer_unlock_commit_nostack(buffer, event);
310 }
311 
312 static void handle_nosleeptime(struct ftrace_graph_ret *trace,
313 			       struct fgraph_times *ftimes,
314 			       int size)
315 {
316 	if (fgraph_sleep_time || size < sizeof(*ftimes))
317 		return;
318 
319 	ftimes->calltime += current->ftrace_sleeptime - ftimes->sleeptime;
320 }
321 
322 void trace_graph_return(struct ftrace_graph_ret *trace,
323 			struct fgraph_ops *gops)
324 {
325 	unsigned long *task_var = fgraph_get_task_var(gops);
326 	struct trace_array *tr = gops->private;
327 	struct trace_array_cpu *data;
328 	struct fgraph_times *ftimes;
329 	unsigned long flags;
330 	unsigned int trace_ctx;
331 	long disabled;
332 	int size;
333 	int cpu;
334 
335 	ftrace_graph_addr_finish(gops, trace);
336 
337 	if (*task_var & TRACE_GRAPH_NOTRACE) {
338 		*task_var &= ~TRACE_GRAPH_NOTRACE;
339 		return;
340 	}
341 
342 	ftimes = fgraph_retrieve_data(gops->idx, &size);
343 	if (!ftimes)
344 		return;
345 
346 	handle_nosleeptime(trace, ftimes, size);
347 
348 	trace->calltime = ftimes->calltime;
349 
350 	local_irq_save(flags);
351 	cpu = raw_smp_processor_id();
352 	data = per_cpu_ptr(tr->array_buffer.data, cpu);
353 	disabled = atomic_inc_return(&data->disabled);
354 	if (likely(disabled == 1)) {
355 		trace_ctx = tracing_gen_ctx_flags(flags);
356 		__trace_graph_return(tr, trace, trace_ctx);
357 	}
358 	atomic_dec(&data->disabled);
359 	local_irq_restore(flags);
360 }
361 
362 static void trace_graph_thresh_return(struct ftrace_graph_ret *trace,
363 				      struct fgraph_ops *gops)
364 {
365 	struct fgraph_times *ftimes;
366 	int size;
367 
368 	ftrace_graph_addr_finish(gops, trace);
369 
370 	if (trace_recursion_test(TRACE_GRAPH_NOTRACE_BIT)) {
371 		trace_recursion_clear(TRACE_GRAPH_NOTRACE_BIT);
372 		return;
373 	}
374 
375 	ftimes = fgraph_retrieve_data(gops->idx, &size);
376 	if (!ftimes)
377 		return;
378 
379 	handle_nosleeptime(trace, ftimes, size);
380 
381 	trace->calltime = ftimes->calltime;
382 
383 	if (tracing_thresh &&
384 	    (trace->rettime - ftimes->calltime < tracing_thresh))
385 		return;
386 	else
387 		trace_graph_return(trace, gops);
388 }
389 
390 static struct fgraph_ops funcgraph_ops = {
391 	.entryfunc = &trace_graph_entry,
392 	.retfunc = &trace_graph_return,
393 };
394 
395 int allocate_fgraph_ops(struct trace_array *tr, struct ftrace_ops *ops)
396 {
397 	struct fgraph_ops *gops;
398 
399 	gops = kzalloc(sizeof(*gops), GFP_KERNEL);
400 	if (!gops)
401 		return -ENOMEM;
402 
403 	gops->entryfunc = &trace_graph_entry;
404 	gops->retfunc = &trace_graph_return;
405 
406 	tr->gops = gops;
407 	gops->private = tr;
408 
409 	fgraph_init_ops(&gops->ops, ops);
410 
411 	return 0;
412 }
413 
414 void free_fgraph_ops(struct trace_array *tr)
415 {
416 	kfree(tr->gops);
417 }
418 
419 __init void init_array_fgraph_ops(struct trace_array *tr, struct ftrace_ops *ops)
420 {
421 	tr->gops = &funcgraph_ops;
422 	funcgraph_ops.private = tr;
423 	fgraph_init_ops(&tr->gops->ops, ops);
424 }
425 
426 static int graph_trace_init(struct trace_array *tr)
427 {
428 	int ret;
429 
430 	tr->gops->entryfunc = trace_graph_entry;
431 
432 	if (tracing_thresh)
433 		tr->gops->retfunc = trace_graph_thresh_return;
434 	else
435 		tr->gops->retfunc = trace_graph_return;
436 
437 	/* Make gops functions are visible before we start tracing */
438 	smp_mb();
439 
440 	ret = register_ftrace_graph(tr->gops);
441 	if (ret)
442 		return ret;
443 	tracing_start_cmdline_record();
444 
445 	return 0;
446 }
447 
448 static void graph_trace_reset(struct trace_array *tr)
449 {
450 	tracing_stop_cmdline_record();
451 	unregister_ftrace_graph(tr->gops);
452 }
453 
454 static int graph_trace_update_thresh(struct trace_array *tr)
455 {
456 	graph_trace_reset(tr);
457 	return graph_trace_init(tr);
458 }
459 
460 static int max_bytes_for_cpu;
461 
462 static void print_graph_cpu(struct trace_seq *s, int cpu)
463 {
464 	/*
465 	 * Start with a space character - to make it stand out
466 	 * to the right a bit when trace output is pasted into
467 	 * email:
468 	 */
469 	trace_seq_printf(s, " %*d) ", max_bytes_for_cpu, cpu);
470 }
471 
472 #define TRACE_GRAPH_PROCINFO_LENGTH	14
473 
474 static void print_graph_proc(struct trace_seq *s, pid_t pid)
475 {
476 	char comm[TASK_COMM_LEN];
477 	/* sign + log10(MAX_INT) + '\0' */
478 	char pid_str[11];
479 	int spaces = 0;
480 	int len;
481 	int i;
482 
483 	trace_find_cmdline(pid, comm);
484 	comm[7] = '\0';
485 	sprintf(pid_str, "%d", pid);
486 
487 	/* 1 stands for the "-" character */
488 	len = strlen(comm) + strlen(pid_str) + 1;
489 
490 	if (len < TRACE_GRAPH_PROCINFO_LENGTH)
491 		spaces = TRACE_GRAPH_PROCINFO_LENGTH - len;
492 
493 	/* First spaces to align center */
494 	for (i = 0; i < spaces / 2; i++)
495 		trace_seq_putc(s, ' ');
496 
497 	trace_seq_printf(s, "%s-%s", comm, pid_str);
498 
499 	/* Last spaces to align center */
500 	for (i = 0; i < spaces - (spaces / 2); i++)
501 		trace_seq_putc(s, ' ');
502 }
503 
504 
505 static void print_graph_lat_fmt(struct trace_seq *s, struct trace_entry *entry)
506 {
507 	trace_seq_putc(s, ' ');
508 	trace_print_lat_fmt(s, entry);
509 	trace_seq_puts(s, " | ");
510 }
511 
512 /* If the pid changed since the last trace, output this event */
513 static void
514 verif_pid(struct trace_seq *s, pid_t pid, int cpu, struct fgraph_data *data)
515 {
516 	pid_t prev_pid;
517 	pid_t *last_pid;
518 
519 	if (!data)
520 		return;
521 
522 	last_pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid);
523 
524 	if (*last_pid == pid)
525 		return;
526 
527 	prev_pid = *last_pid;
528 	*last_pid = pid;
529 
530 	if (prev_pid == -1)
531 		return;
532 /*
533  * Context-switch trace line:
534 
535  ------------------------------------------
536  | 1)  migration/0--1  =>  sshd-1755
537  ------------------------------------------
538 
539  */
540 	trace_seq_puts(s, " ------------------------------------------\n");
541 	print_graph_cpu(s, cpu);
542 	print_graph_proc(s, prev_pid);
543 	trace_seq_puts(s, " => ");
544 	print_graph_proc(s, pid);
545 	trace_seq_puts(s, "\n ------------------------------------------\n\n");
546 }
547 
548 static struct ftrace_graph_ret_entry *
549 get_return_for_leaf(struct trace_iterator *iter,
550 		struct ftrace_graph_ent_entry *curr)
551 {
552 	struct fgraph_data *data = iter->private;
553 	struct ring_buffer_iter *ring_iter = NULL;
554 	struct ring_buffer_event *event;
555 	struct ftrace_graph_ret_entry *next;
556 
557 	/*
558 	 * If the previous output failed to write to the seq buffer,
559 	 * then we just reuse the data from before.
560 	 */
561 	if (data && data->failed) {
562 		curr = &data->ent.ent;
563 		next = &data->ret;
564 	} else {
565 
566 		ring_iter = trace_buffer_iter(iter, iter->cpu);
567 
568 		/* First peek to compare current entry and the next one */
569 		if (ring_iter)
570 			event = ring_buffer_iter_peek(ring_iter, NULL);
571 		else {
572 			/*
573 			 * We need to consume the current entry to see
574 			 * the next one.
575 			 */
576 			ring_buffer_consume(iter->array_buffer->buffer, iter->cpu,
577 					    NULL, NULL);
578 			event = ring_buffer_peek(iter->array_buffer->buffer, iter->cpu,
579 						 NULL, NULL);
580 		}
581 
582 		if (!event)
583 			return NULL;
584 
585 		next = ring_buffer_event_data(event);
586 
587 		if (data) {
588 			/*
589 			 * Save current and next entries for later reference
590 			 * if the output fails.
591 			 */
592 			if (unlikely(curr->ent.type == TRACE_GRAPH_RETADDR_ENT))
593 				data->ent.rent = *(struct fgraph_retaddr_ent_entry *)curr;
594 			else
595 				data->ent.ent = *curr;
596 			/*
597 			 * If the next event is not a return type, then
598 			 * we only care about what type it is. Otherwise we can
599 			 * safely copy the entire event.
600 			 */
601 			if (next->ent.type == TRACE_GRAPH_RET)
602 				data->ret = *next;
603 			else
604 				data->ret.ent.type = next->ent.type;
605 		}
606 	}
607 
608 	if (next->ent.type != TRACE_GRAPH_RET)
609 		return NULL;
610 
611 	if (curr->ent.pid != next->ent.pid ||
612 			curr->graph_ent.func != next->ret.func)
613 		return NULL;
614 
615 	/* this is a leaf, now advance the iterator */
616 	if (ring_iter)
617 		ring_buffer_iter_advance(ring_iter);
618 
619 	return next;
620 }
621 
622 static void print_graph_abs_time(u64 t, struct trace_seq *s)
623 {
624 	unsigned long usecs_rem;
625 
626 	usecs_rem = do_div(t, NSEC_PER_SEC);
627 	usecs_rem /= 1000;
628 
629 	trace_seq_printf(s, "%5lu.%06lu |  ",
630 			 (unsigned long)t, usecs_rem);
631 }
632 
633 static void
634 print_graph_rel_time(struct trace_iterator *iter, struct trace_seq *s)
635 {
636 	unsigned long long usecs;
637 
638 	usecs = iter->ts - iter->array_buffer->time_start;
639 	do_div(usecs, NSEC_PER_USEC);
640 
641 	trace_seq_printf(s, "%9llu us |  ", usecs);
642 }
643 
644 static void
645 print_graph_irq(struct trace_iterator *iter, unsigned long addr,
646 		enum trace_type type, int cpu, pid_t pid, u32 flags)
647 {
648 	struct trace_array *tr = iter->tr;
649 	struct trace_seq *s = &iter->seq;
650 	struct trace_entry *ent = iter->ent;
651 
652 	addr += iter->tr->text_delta;
653 
654 	if (addr < (unsigned long)__irqentry_text_start ||
655 		addr >= (unsigned long)__irqentry_text_end)
656 		return;
657 
658 	if (tr->trace_flags & TRACE_ITER_CONTEXT_INFO) {
659 		/* Absolute time */
660 		if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
661 			print_graph_abs_time(iter->ts, s);
662 
663 		/* Relative time */
664 		if (flags & TRACE_GRAPH_PRINT_REL_TIME)
665 			print_graph_rel_time(iter, s);
666 
667 		/* Cpu */
668 		if (flags & TRACE_GRAPH_PRINT_CPU)
669 			print_graph_cpu(s, cpu);
670 
671 		/* Proc */
672 		if (flags & TRACE_GRAPH_PRINT_PROC) {
673 			print_graph_proc(s, pid);
674 			trace_seq_puts(s, " | ");
675 		}
676 
677 		/* Latency format */
678 		if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
679 			print_graph_lat_fmt(s, ent);
680 	}
681 
682 	/* No overhead */
683 	print_graph_duration(tr, 0, s, flags | FLAGS_FILL_START);
684 
685 	if (type == TRACE_GRAPH_ENT)
686 		trace_seq_puts(s, "==========>");
687 	else
688 		trace_seq_puts(s, "<==========");
689 
690 	print_graph_duration(tr, 0, s, flags | FLAGS_FILL_END);
691 	trace_seq_putc(s, '\n');
692 }
693 
694 void
695 trace_print_graph_duration(unsigned long long duration, struct trace_seq *s)
696 {
697 	unsigned long nsecs_rem = do_div(duration, 1000);
698 	/* log10(ULONG_MAX) + '\0' */
699 	char usecs_str[21];
700 	char nsecs_str[5];
701 	int len;
702 	int i;
703 
704 	sprintf(usecs_str, "%lu", (unsigned long) duration);
705 
706 	/* Print msecs */
707 	trace_seq_printf(s, "%s", usecs_str);
708 
709 	len = strlen(usecs_str);
710 
711 	/* Print nsecs (we don't want to exceed 7 numbers) */
712 	if (len < 7) {
713 		size_t slen = min_t(size_t, sizeof(nsecs_str), 8UL - len);
714 
715 		snprintf(nsecs_str, slen, "%03lu", nsecs_rem);
716 		trace_seq_printf(s, ".%s", nsecs_str);
717 		len += strlen(nsecs_str) + 1;
718 	}
719 
720 	trace_seq_puts(s, " us ");
721 
722 	/* Print remaining spaces to fit the row's width */
723 	for (i = len; i < 8; i++)
724 		trace_seq_putc(s, ' ');
725 }
726 
727 static void
728 print_graph_duration(struct trace_array *tr, unsigned long long duration,
729 		     struct trace_seq *s, u32 flags)
730 {
731 	if (!(flags & TRACE_GRAPH_PRINT_DURATION) ||
732 	    !(tr->trace_flags & TRACE_ITER_CONTEXT_INFO))
733 		return;
734 
735 	/* No real adata, just filling the column with spaces */
736 	switch (flags & TRACE_GRAPH_PRINT_FILL_MASK) {
737 	case FLAGS_FILL_FULL:
738 		trace_seq_puts(s, "              |  ");
739 		return;
740 	case FLAGS_FILL_START:
741 		trace_seq_puts(s, "  ");
742 		return;
743 	case FLAGS_FILL_END:
744 		trace_seq_puts(s, " |");
745 		return;
746 	}
747 
748 	/* Signal a overhead of time execution to the output */
749 	if (flags & TRACE_GRAPH_PRINT_OVERHEAD)
750 		trace_seq_printf(s, "%c ", trace_find_mark(duration));
751 	else
752 		trace_seq_puts(s, "  ");
753 
754 	trace_print_graph_duration(duration, s);
755 	trace_seq_puts(s, "|  ");
756 }
757 
758 #ifdef CONFIG_FUNCTION_GRAPH_RETVAL
759 #define __TRACE_GRAPH_PRINT_RETVAL TRACE_GRAPH_PRINT_RETVAL
760 #else
761 #define __TRACE_GRAPH_PRINT_RETVAL 0
762 #endif
763 
764 #ifdef CONFIG_FUNCTION_GRAPH_RETADDR
765 #define __TRACE_GRAPH_PRINT_RETADDR TRACE_GRAPH_PRINT_RETADDR
766 static void print_graph_retaddr(struct trace_seq *s, struct fgraph_retaddr_ent_entry *entry,
767 				u32 trace_flags, bool comment)
768 {
769 	if (comment)
770 		trace_seq_puts(s, " /*");
771 
772 	trace_seq_puts(s, " <-");
773 	seq_print_ip_sym(s, entry->graph_ent.retaddr, trace_flags | TRACE_ITER_SYM_OFFSET);
774 
775 	if (comment)
776 		trace_seq_puts(s, " */");
777 }
778 #else
779 #define __TRACE_GRAPH_PRINT_RETADDR 0
780 #define print_graph_retaddr(_seq, _entry, _tflags, _comment)		do { } while (0)
781 #endif
782 
783 #if defined(CONFIG_FUNCTION_GRAPH_RETVAL) || defined(CONFIG_FUNCTION_GRAPH_RETADDR)
784 
785 static void print_graph_retval(struct trace_seq *s, struct ftrace_graph_ent_entry *entry,
786 				struct ftrace_graph_ret *graph_ret, void *func,
787 				u32 opt_flags, u32 trace_flags)
788 {
789 	unsigned long err_code = 0;
790 	unsigned long retval = 0;
791 	bool print_retaddr = false;
792 	bool print_retval = false;
793 	bool hex_format = !!(opt_flags & TRACE_GRAPH_PRINT_RETVAL_HEX);
794 
795 #ifdef CONFIG_FUNCTION_GRAPH_RETVAL
796 	retval = graph_ret->retval;
797 	print_retval = !!(opt_flags & TRACE_GRAPH_PRINT_RETVAL);
798 #endif
799 
800 #ifdef CONFIG_FUNCTION_GRAPH_RETADDR
801 	print_retaddr = !!(opt_flags & TRACE_GRAPH_PRINT_RETADDR);
802 #endif
803 
804 	if (print_retval && retval && !hex_format) {
805 		/* Check if the return value matches the negative format */
806 		if (IS_ENABLED(CONFIG_64BIT) && (retval & BIT(31)) &&
807 			(((u64)retval) >> 32) == 0) {
808 			err_code = sign_extend64(retval, 31);
809 		} else {
810 			err_code = retval;
811 		}
812 
813 		if (!IS_ERR_VALUE(err_code))
814 			err_code = 0;
815 	}
816 
817 	if (entry) {
818 		if (entry->ent.type != TRACE_GRAPH_RETADDR_ENT)
819 			print_retaddr = false;
820 
821 		trace_seq_printf(s, "%ps();", func);
822 		if (print_retval || print_retaddr)
823 			trace_seq_puts(s, " /*");
824 		else
825 			trace_seq_putc(s, '\n');
826 	} else {
827 		print_retaddr = false;
828 		trace_seq_printf(s, "} /* %ps", func);
829 	}
830 
831 	if (print_retaddr)
832 		print_graph_retaddr(s, (struct fgraph_retaddr_ent_entry *)entry,
833 				    trace_flags, false);
834 
835 	if (print_retval) {
836 		if (hex_format || (err_code == 0))
837 			trace_seq_printf(s, " ret=0x%lx", retval);
838 		else
839 			trace_seq_printf(s, " ret=%ld", err_code);
840 	}
841 
842 	if (!entry || print_retval || print_retaddr)
843 		trace_seq_puts(s, " */\n");
844 }
845 
846 #else
847 
848 #define print_graph_retval(_seq, _ent, _ret, _func, _opt_flags, _trace_flags) do {} while (0)
849 
850 #endif
851 
852 /* Case of a leaf function on its call entry */
853 static enum print_line_t
854 print_graph_entry_leaf(struct trace_iterator *iter,
855 		struct ftrace_graph_ent_entry *entry,
856 		struct ftrace_graph_ret_entry *ret_entry,
857 		struct trace_seq *s, u32 flags)
858 {
859 	struct fgraph_data *data = iter->private;
860 	struct trace_array *tr = iter->tr;
861 	struct ftrace_graph_ret *graph_ret;
862 	struct ftrace_graph_ent *call;
863 	unsigned long long duration;
864 	unsigned long func;
865 	int cpu = iter->cpu;
866 	int i;
867 
868 	graph_ret = &ret_entry->ret;
869 	call = &entry->graph_ent;
870 	duration = graph_ret->rettime - graph_ret->calltime;
871 
872 	func = call->func + iter->tr->text_delta;
873 
874 	if (data) {
875 		struct fgraph_cpu_data *cpu_data;
876 
877 		cpu_data = per_cpu_ptr(data->cpu_data, cpu);
878 
879 		/*
880 		 * Comments display at + 1 to depth. Since
881 		 * this is a leaf function, keep the comments
882 		 * equal to this depth.
883 		 */
884 		cpu_data->depth = call->depth - 1;
885 
886 		/* No need to keep this function around for this depth */
887 		if (call->depth < FTRACE_RETFUNC_DEPTH &&
888 		    !WARN_ON_ONCE(call->depth < 0))
889 			cpu_data->enter_funcs[call->depth] = 0;
890 	}
891 
892 	/* Overhead and duration */
893 	print_graph_duration(tr, duration, s, flags);
894 
895 	/* Function */
896 	for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++)
897 		trace_seq_putc(s, ' ');
898 
899 	/*
900 	 * Write out the function return value or return address
901 	 */
902 	if (flags & (__TRACE_GRAPH_PRINT_RETVAL | __TRACE_GRAPH_PRINT_RETADDR)) {
903 		print_graph_retval(s, entry, graph_ret,
904 				   (void *)graph_ret->func + iter->tr->text_delta,
905 				   flags, tr->trace_flags);
906 	} else {
907 		trace_seq_printf(s, "%ps();\n", (void *)func);
908 	}
909 
910 	print_graph_irq(iter, graph_ret->func, TRACE_GRAPH_RET,
911 			cpu, iter->ent->pid, flags);
912 
913 	return trace_handle_return(s);
914 }
915 
916 static enum print_line_t
917 print_graph_entry_nested(struct trace_iterator *iter,
918 			 struct ftrace_graph_ent_entry *entry,
919 			 struct trace_seq *s, int cpu, u32 flags)
920 {
921 	struct ftrace_graph_ent *call = &entry->graph_ent;
922 	struct fgraph_data *data = iter->private;
923 	struct trace_array *tr = iter->tr;
924 	unsigned long func;
925 	int i;
926 
927 	if (data) {
928 		struct fgraph_cpu_data *cpu_data;
929 		int cpu = iter->cpu;
930 
931 		cpu_data = per_cpu_ptr(data->cpu_data, cpu);
932 		cpu_data->depth = call->depth;
933 
934 		/* Save this function pointer to see if the exit matches */
935 		if (call->depth < FTRACE_RETFUNC_DEPTH &&
936 		    !WARN_ON_ONCE(call->depth < 0))
937 			cpu_data->enter_funcs[call->depth] = call->func;
938 	}
939 
940 	/* No time */
941 	print_graph_duration(tr, 0, s, flags | FLAGS_FILL_FULL);
942 
943 	/* Function */
944 	for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++)
945 		trace_seq_putc(s, ' ');
946 
947 	func = call->func + iter->tr->text_delta;
948 
949 	trace_seq_printf(s, "%ps() {", (void *)func);
950 	if (flags & __TRACE_GRAPH_PRINT_RETADDR  &&
951 		entry->ent.type == TRACE_GRAPH_RETADDR_ENT)
952 		print_graph_retaddr(s, (struct fgraph_retaddr_ent_entry *)entry,
953 			tr->trace_flags, true);
954 	trace_seq_putc(s, '\n');
955 
956 	if (trace_seq_has_overflowed(s))
957 		return TRACE_TYPE_PARTIAL_LINE;
958 
959 	/*
960 	 * we already consumed the current entry to check the next one
961 	 * and see if this is a leaf.
962 	 */
963 	return TRACE_TYPE_NO_CONSUME;
964 }
965 
966 static void
967 print_graph_prologue(struct trace_iterator *iter, struct trace_seq *s,
968 		     int type, unsigned long addr, u32 flags)
969 {
970 	struct fgraph_data *data = iter->private;
971 	struct trace_entry *ent = iter->ent;
972 	struct trace_array *tr = iter->tr;
973 	int cpu = iter->cpu;
974 
975 	/* Pid */
976 	verif_pid(s, ent->pid, cpu, data);
977 
978 	if (type)
979 		/* Interrupt */
980 		print_graph_irq(iter, addr, type, cpu, ent->pid, flags);
981 
982 	if (!(tr->trace_flags & TRACE_ITER_CONTEXT_INFO))
983 		return;
984 
985 	/* Absolute time */
986 	if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
987 		print_graph_abs_time(iter->ts, s);
988 
989 	/* Relative time */
990 	if (flags & TRACE_GRAPH_PRINT_REL_TIME)
991 		print_graph_rel_time(iter, s);
992 
993 	/* Cpu */
994 	if (flags & TRACE_GRAPH_PRINT_CPU)
995 		print_graph_cpu(s, cpu);
996 
997 	/* Proc */
998 	if (flags & TRACE_GRAPH_PRINT_PROC) {
999 		print_graph_proc(s, ent->pid);
1000 		trace_seq_puts(s, " | ");
1001 	}
1002 
1003 	/* Latency format */
1004 	if (tr->trace_flags & TRACE_ITER_LATENCY_FMT)
1005 		print_graph_lat_fmt(s, ent);
1006 
1007 	return;
1008 }
1009 
1010 /*
1011  * Entry check for irq code
1012  *
1013  * returns 1 if
1014  *  - we are inside irq code
1015  *  - we just entered irq code
1016  *
1017  * returns 0 if
1018  *  - funcgraph-interrupts option is set
1019  *  - we are not inside irq code
1020  */
1021 static int
1022 check_irq_entry(struct trace_iterator *iter, u32 flags,
1023 		unsigned long addr, int depth)
1024 {
1025 	int cpu = iter->cpu;
1026 	int *depth_irq;
1027 	struct fgraph_data *data = iter->private;
1028 
1029 	addr += iter->tr->text_delta;
1030 
1031 	/*
1032 	 * If we are either displaying irqs, or we got called as
1033 	 * a graph event and private data does not exist,
1034 	 * then we bypass the irq check.
1035 	 */
1036 	if ((flags & TRACE_GRAPH_PRINT_IRQS) ||
1037 	    (!data))
1038 		return 0;
1039 
1040 	depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
1041 
1042 	/*
1043 	 * We are inside the irq code
1044 	 */
1045 	if (*depth_irq >= 0)
1046 		return 1;
1047 
1048 	if ((addr < (unsigned long)__irqentry_text_start) ||
1049 	    (addr >= (unsigned long)__irqentry_text_end))
1050 		return 0;
1051 
1052 	/*
1053 	 * We are entering irq code.
1054 	 */
1055 	*depth_irq = depth;
1056 	return 1;
1057 }
1058 
1059 /*
1060  * Return check for irq code
1061  *
1062  * returns 1 if
1063  *  - we are inside irq code
1064  *  - we just left irq code
1065  *
1066  * returns 0 if
1067  *  - funcgraph-interrupts option is set
1068  *  - we are not inside irq code
1069  */
1070 static int
1071 check_irq_return(struct trace_iterator *iter, u32 flags, int depth)
1072 {
1073 	int cpu = iter->cpu;
1074 	int *depth_irq;
1075 	struct fgraph_data *data = iter->private;
1076 
1077 	/*
1078 	 * If we are either displaying irqs, or we got called as
1079 	 * a graph event and private data does not exist,
1080 	 * then we bypass the irq check.
1081 	 */
1082 	if ((flags & TRACE_GRAPH_PRINT_IRQS) ||
1083 	    (!data))
1084 		return 0;
1085 
1086 	depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
1087 
1088 	/*
1089 	 * We are not inside the irq code.
1090 	 */
1091 	if (*depth_irq == -1)
1092 		return 0;
1093 
1094 	/*
1095 	 * We are inside the irq code, and this is returning entry.
1096 	 * Let's not trace it and clear the entry depth, since
1097 	 * we are out of irq code.
1098 	 *
1099 	 * This condition ensures that we 'leave the irq code' once
1100 	 * we are out of the entry depth. Thus protecting us from
1101 	 * the RETURN entry loss.
1102 	 */
1103 	if (*depth_irq >= depth) {
1104 		*depth_irq = -1;
1105 		return 1;
1106 	}
1107 
1108 	/*
1109 	 * We are inside the irq code, and this is not the entry.
1110 	 */
1111 	return 1;
1112 }
1113 
1114 static enum print_line_t
1115 print_graph_entry(struct ftrace_graph_ent_entry *field, struct trace_seq *s,
1116 			struct trace_iterator *iter, u32 flags)
1117 {
1118 	struct fgraph_data *data = iter->private;
1119 	struct ftrace_graph_ent *call = &field->graph_ent;
1120 	struct ftrace_graph_ret_entry *leaf_ret;
1121 	static enum print_line_t ret;
1122 	int cpu = iter->cpu;
1123 
1124 	if (check_irq_entry(iter, flags, call->func, call->depth))
1125 		return TRACE_TYPE_HANDLED;
1126 
1127 	print_graph_prologue(iter, s, TRACE_GRAPH_ENT, call->func, flags);
1128 
1129 	leaf_ret = get_return_for_leaf(iter, field);
1130 	if (leaf_ret)
1131 		ret = print_graph_entry_leaf(iter, field, leaf_ret, s, flags);
1132 	else
1133 		ret = print_graph_entry_nested(iter, field, s, cpu, flags);
1134 
1135 	if (data) {
1136 		/*
1137 		 * If we failed to write our output, then we need to make
1138 		 * note of it. Because we already consumed our entry.
1139 		 */
1140 		if (s->full) {
1141 			data->failed = 1;
1142 			data->cpu = cpu;
1143 		} else
1144 			data->failed = 0;
1145 	}
1146 
1147 	return ret;
1148 }
1149 
1150 static enum print_line_t
1151 print_graph_return(struct ftrace_graph_ret *trace, struct trace_seq *s,
1152 		   struct trace_entry *ent, struct trace_iterator *iter,
1153 		   u32 flags)
1154 {
1155 	unsigned long long duration = trace->rettime - trace->calltime;
1156 	struct fgraph_data *data = iter->private;
1157 	struct trace_array *tr = iter->tr;
1158 	unsigned long func;
1159 	pid_t pid = ent->pid;
1160 	int cpu = iter->cpu;
1161 	int func_match = 1;
1162 	int i;
1163 
1164 	func = trace->func + iter->tr->text_delta;
1165 
1166 	if (check_irq_return(iter, flags, trace->depth))
1167 		return TRACE_TYPE_HANDLED;
1168 
1169 	if (data) {
1170 		struct fgraph_cpu_data *cpu_data;
1171 		int cpu = iter->cpu;
1172 
1173 		cpu_data = per_cpu_ptr(data->cpu_data, cpu);
1174 
1175 		/*
1176 		 * Comments display at + 1 to depth. This is the
1177 		 * return from a function, we now want the comments
1178 		 * to display at the same level of the bracket.
1179 		 */
1180 		cpu_data->depth = trace->depth - 1;
1181 
1182 		if (trace->depth < FTRACE_RETFUNC_DEPTH &&
1183 		    !WARN_ON_ONCE(trace->depth < 0)) {
1184 			if (cpu_data->enter_funcs[trace->depth] != trace->func)
1185 				func_match = 0;
1186 			cpu_data->enter_funcs[trace->depth] = 0;
1187 		}
1188 	}
1189 
1190 	print_graph_prologue(iter, s, 0, 0, flags);
1191 
1192 	/* Overhead and duration */
1193 	print_graph_duration(tr, duration, s, flags);
1194 
1195 	/* Closing brace */
1196 	for (i = 0; i < trace->depth * TRACE_GRAPH_INDENT; i++)
1197 		trace_seq_putc(s, ' ');
1198 
1199 	/*
1200 	 * Always write out the function name and its return value if the
1201 	 * funcgraph-retval option is enabled.
1202 	 */
1203 	if (flags & __TRACE_GRAPH_PRINT_RETVAL) {
1204 		print_graph_retval(s, NULL, trace, (void *)func, flags, tr->trace_flags);
1205 	} else {
1206 		/*
1207 		 * If the return function does not have a matching entry,
1208 		 * then the entry was lost. Instead of just printing
1209 		 * the '}' and letting the user guess what function this
1210 		 * belongs to, write out the function name. Always do
1211 		 * that if the funcgraph-tail option is enabled.
1212 		 */
1213 		if (func_match && !(flags & TRACE_GRAPH_PRINT_TAIL))
1214 			trace_seq_puts(s, "}\n");
1215 		else
1216 			trace_seq_printf(s, "} /* %ps */\n", (void *)func);
1217 	}
1218 
1219 	/* Overrun */
1220 	if (flags & TRACE_GRAPH_PRINT_OVERRUN)
1221 		trace_seq_printf(s, " (Overruns: %u)\n",
1222 				 trace->overrun);
1223 
1224 	print_graph_irq(iter, trace->func, TRACE_GRAPH_RET,
1225 			cpu, pid, flags);
1226 
1227 	return trace_handle_return(s);
1228 }
1229 
1230 static enum print_line_t
1231 print_graph_comment(struct trace_seq *s, struct trace_entry *ent,
1232 		    struct trace_iterator *iter, u32 flags)
1233 {
1234 	struct trace_array *tr = iter->tr;
1235 	unsigned long sym_flags = (tr->trace_flags & TRACE_ITER_SYM_MASK);
1236 	struct fgraph_data *data = iter->private;
1237 	struct trace_event *event;
1238 	int depth = 0;
1239 	int ret;
1240 	int i;
1241 
1242 	if (data)
1243 		depth = per_cpu_ptr(data->cpu_data, iter->cpu)->depth;
1244 
1245 	print_graph_prologue(iter, s, 0, 0, flags);
1246 
1247 	/* No time */
1248 	print_graph_duration(tr, 0, s, flags | FLAGS_FILL_FULL);
1249 
1250 	/* Indentation */
1251 	if (depth > 0)
1252 		for (i = 0; i < (depth + 1) * TRACE_GRAPH_INDENT; i++)
1253 			trace_seq_putc(s, ' ');
1254 
1255 	/* The comment */
1256 	trace_seq_puts(s, "/* ");
1257 
1258 	switch (iter->ent->type) {
1259 	case TRACE_BPUTS:
1260 		ret = trace_print_bputs_msg_only(iter);
1261 		if (ret != TRACE_TYPE_HANDLED)
1262 			return ret;
1263 		break;
1264 	case TRACE_BPRINT:
1265 		ret = trace_print_bprintk_msg_only(iter);
1266 		if (ret != TRACE_TYPE_HANDLED)
1267 			return ret;
1268 		break;
1269 	case TRACE_PRINT:
1270 		ret = trace_print_printk_msg_only(iter);
1271 		if (ret != TRACE_TYPE_HANDLED)
1272 			return ret;
1273 		break;
1274 	default:
1275 		event = ftrace_find_event(ent->type);
1276 		if (!event)
1277 			return TRACE_TYPE_UNHANDLED;
1278 
1279 		ret = event->funcs->trace(iter, sym_flags, event);
1280 		if (ret != TRACE_TYPE_HANDLED)
1281 			return ret;
1282 	}
1283 
1284 	if (trace_seq_has_overflowed(s))
1285 		goto out;
1286 
1287 	/* Strip ending newline */
1288 	if (s->buffer[s->seq.len - 1] == '\n') {
1289 		s->buffer[s->seq.len - 1] = '\0';
1290 		s->seq.len--;
1291 	}
1292 
1293 	trace_seq_puts(s, " */\n");
1294  out:
1295 	return trace_handle_return(s);
1296 }
1297 
1298 
1299 enum print_line_t
1300 print_graph_function_flags(struct trace_iterator *iter, u32 flags)
1301 {
1302 	struct ftrace_graph_ent_entry *field;
1303 	struct fgraph_data *data = iter->private;
1304 	struct trace_entry *entry = iter->ent;
1305 	struct trace_seq *s = &iter->seq;
1306 	int cpu = iter->cpu;
1307 	int ret;
1308 
1309 	if (data && per_cpu_ptr(data->cpu_data, cpu)->ignore) {
1310 		per_cpu_ptr(data->cpu_data, cpu)->ignore = 0;
1311 		return TRACE_TYPE_HANDLED;
1312 	}
1313 
1314 	/*
1315 	 * If the last output failed, there's a possibility we need
1316 	 * to print out the missing entry which would never go out.
1317 	 */
1318 	if (data && data->failed) {
1319 		field = &data->ent.ent;
1320 		iter->cpu = data->cpu;
1321 		ret = print_graph_entry(field, s, iter, flags);
1322 		if (ret == TRACE_TYPE_HANDLED && iter->cpu != cpu) {
1323 			per_cpu_ptr(data->cpu_data, iter->cpu)->ignore = 1;
1324 			ret = TRACE_TYPE_NO_CONSUME;
1325 		}
1326 		iter->cpu = cpu;
1327 		return ret;
1328 	}
1329 
1330 	switch (entry->type) {
1331 	case TRACE_GRAPH_ENT: {
1332 		/*
1333 		 * print_graph_entry() may consume the current event,
1334 		 * thus @field may become invalid, so we need to save it.
1335 		 * sizeof(struct ftrace_graph_ent_entry) is very small,
1336 		 * it can be safely saved at the stack.
1337 		 */
1338 		struct ftrace_graph_ent_entry saved;
1339 		trace_assign_type(field, entry);
1340 		saved = *field;
1341 		return print_graph_entry(&saved, s, iter, flags);
1342 	}
1343 #ifdef CONFIG_FUNCTION_GRAPH_RETADDR
1344 	case TRACE_GRAPH_RETADDR_ENT: {
1345 		struct fgraph_retaddr_ent_entry saved;
1346 		struct fgraph_retaddr_ent_entry *rfield;
1347 
1348 		trace_assign_type(rfield, entry);
1349 		saved = *rfield;
1350 		return print_graph_entry((struct ftrace_graph_ent_entry *)&saved, s, iter, flags);
1351 	}
1352 #endif
1353 	case TRACE_GRAPH_RET: {
1354 		struct ftrace_graph_ret_entry *field;
1355 		trace_assign_type(field, entry);
1356 		return print_graph_return(&field->ret, s, entry, iter, flags);
1357 	}
1358 	case TRACE_STACK:
1359 	case TRACE_FN:
1360 		/* dont trace stack and functions as comments */
1361 		return TRACE_TYPE_UNHANDLED;
1362 
1363 	default:
1364 		return print_graph_comment(s, entry, iter, flags);
1365 	}
1366 
1367 	return TRACE_TYPE_HANDLED;
1368 }
1369 
1370 static enum print_line_t
1371 print_graph_function(struct trace_iterator *iter)
1372 {
1373 	return print_graph_function_flags(iter, tracer_flags.val);
1374 }
1375 
1376 static enum print_line_t
1377 print_graph_function_event(struct trace_iterator *iter, int flags,
1378 			   struct trace_event *event)
1379 {
1380 	return print_graph_function(iter);
1381 }
1382 
1383 static void print_lat_header(struct seq_file *s, u32 flags)
1384 {
1385 	static const char spaces[] = "                "	/* 16 spaces */
1386 		"    "					/* 4 spaces */
1387 		"                 ";			/* 17 spaces */
1388 	int size = 0;
1389 
1390 	if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1391 		size += 16;
1392 	if (flags & TRACE_GRAPH_PRINT_REL_TIME)
1393 		size += 16;
1394 	if (flags & TRACE_GRAPH_PRINT_CPU)
1395 		size += 4;
1396 	if (flags & TRACE_GRAPH_PRINT_PROC)
1397 		size += 17;
1398 
1399 	seq_printf(s, "#%.*s  _-----=> irqs-off        \n", size, spaces);
1400 	seq_printf(s, "#%.*s / _----=> need-resched    \n", size, spaces);
1401 	seq_printf(s, "#%.*s| / _---=> hardirq/softirq \n", size, spaces);
1402 	seq_printf(s, "#%.*s|| / _--=> preempt-depth   \n", size, spaces);
1403 	seq_printf(s, "#%.*s||| /                      \n", size, spaces);
1404 }
1405 
1406 static void __print_graph_headers_flags(struct trace_array *tr,
1407 					struct seq_file *s, u32 flags)
1408 {
1409 	int lat = tr->trace_flags & TRACE_ITER_LATENCY_FMT;
1410 
1411 	if (lat)
1412 		print_lat_header(s, flags);
1413 
1414 	/* 1st line */
1415 	seq_putc(s, '#');
1416 	if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1417 		seq_puts(s, "     TIME       ");
1418 	if (flags & TRACE_GRAPH_PRINT_REL_TIME)
1419 		seq_puts(s, "   REL TIME     ");
1420 	if (flags & TRACE_GRAPH_PRINT_CPU)
1421 		seq_puts(s, " CPU");
1422 	if (flags & TRACE_GRAPH_PRINT_PROC)
1423 		seq_puts(s, "  TASK/PID       ");
1424 	if (lat)
1425 		seq_puts(s, "||||   ");
1426 	if (flags & TRACE_GRAPH_PRINT_DURATION)
1427 		seq_puts(s, "  DURATION   ");
1428 	seq_puts(s, "               FUNCTION CALLS\n");
1429 
1430 	/* 2nd line */
1431 	seq_putc(s, '#');
1432 	if (flags & TRACE_GRAPH_PRINT_ABS_TIME)
1433 		seq_puts(s, "      |         ");
1434 	if (flags & TRACE_GRAPH_PRINT_REL_TIME)
1435 		seq_puts(s, "      |         ");
1436 	if (flags & TRACE_GRAPH_PRINT_CPU)
1437 		seq_puts(s, " |  ");
1438 	if (flags & TRACE_GRAPH_PRINT_PROC)
1439 		seq_puts(s, "   |    |        ");
1440 	if (lat)
1441 		seq_puts(s, "||||   ");
1442 	if (flags & TRACE_GRAPH_PRINT_DURATION)
1443 		seq_puts(s, "   |   |      ");
1444 	seq_puts(s, "               |   |   |   |\n");
1445 }
1446 
1447 static void print_graph_headers(struct seq_file *s)
1448 {
1449 	print_graph_headers_flags(s, tracer_flags.val);
1450 }
1451 
1452 void print_graph_headers_flags(struct seq_file *s, u32 flags)
1453 {
1454 	struct trace_iterator *iter = s->private;
1455 	struct trace_array *tr = iter->tr;
1456 
1457 	if (!(tr->trace_flags & TRACE_ITER_CONTEXT_INFO))
1458 		return;
1459 
1460 	if (tr->trace_flags & TRACE_ITER_LATENCY_FMT) {
1461 		/* print nothing if the buffers are empty */
1462 		if (trace_empty(iter))
1463 			return;
1464 
1465 		print_trace_header(s, iter);
1466 	}
1467 
1468 	__print_graph_headers_flags(tr, s, flags);
1469 }
1470 
1471 void graph_trace_open(struct trace_iterator *iter)
1472 {
1473 	/* pid and depth on the last trace processed */
1474 	struct fgraph_data *data;
1475 	gfp_t gfpflags;
1476 	int cpu;
1477 
1478 	iter->private = NULL;
1479 
1480 	/* We can be called in atomic context via ftrace_dump() */
1481 	gfpflags = (in_atomic() || irqs_disabled()) ? GFP_ATOMIC : GFP_KERNEL;
1482 
1483 	data = kzalloc(sizeof(*data), gfpflags);
1484 	if (!data)
1485 		goto out_err;
1486 
1487 	data->cpu_data = alloc_percpu_gfp(struct fgraph_cpu_data, gfpflags);
1488 	if (!data->cpu_data)
1489 		goto out_err_free;
1490 
1491 	for_each_possible_cpu(cpu) {
1492 		pid_t *pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid);
1493 		int *depth = &(per_cpu_ptr(data->cpu_data, cpu)->depth);
1494 		int *ignore = &(per_cpu_ptr(data->cpu_data, cpu)->ignore);
1495 		int *depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
1496 
1497 		*pid = -1;
1498 		*depth = 0;
1499 		*ignore = 0;
1500 		*depth_irq = -1;
1501 	}
1502 
1503 	iter->private = data;
1504 
1505 	return;
1506 
1507  out_err_free:
1508 	kfree(data);
1509  out_err:
1510 	pr_warn("function graph tracer: not enough memory\n");
1511 }
1512 
1513 void graph_trace_close(struct trace_iterator *iter)
1514 {
1515 	struct fgraph_data *data = iter->private;
1516 
1517 	if (data) {
1518 		free_percpu(data->cpu_data);
1519 		kfree(data);
1520 	}
1521 }
1522 
1523 static int
1524 func_graph_set_flag(struct trace_array *tr, u32 old_flags, u32 bit, int set)
1525 {
1526 	if (bit == TRACE_GRAPH_PRINT_IRQS)
1527 		ftrace_graph_skip_irqs = !set;
1528 
1529 	if (bit == TRACE_GRAPH_SLEEP_TIME)
1530 		ftrace_graph_sleep_time_control(set);
1531 
1532 	if (bit == TRACE_GRAPH_GRAPH_TIME)
1533 		ftrace_graph_graph_time_control(set);
1534 
1535 	return 0;
1536 }
1537 
1538 static struct trace_event_functions graph_functions = {
1539 	.trace		= print_graph_function_event,
1540 };
1541 
1542 static struct trace_event graph_trace_entry_event = {
1543 	.type		= TRACE_GRAPH_ENT,
1544 	.funcs		= &graph_functions,
1545 };
1546 
1547 #ifdef CONFIG_FUNCTION_GRAPH_RETADDR
1548 static struct trace_event graph_trace_retaddr_entry_event = {
1549 	.type		= TRACE_GRAPH_RETADDR_ENT,
1550 	.funcs		= &graph_functions,
1551 };
1552 #endif
1553 
1554 static struct trace_event graph_trace_ret_event = {
1555 	.type		= TRACE_GRAPH_RET,
1556 	.funcs		= &graph_functions
1557 };
1558 
1559 static struct tracer graph_trace __tracer_data = {
1560 	.name		= "function_graph",
1561 	.update_thresh	= graph_trace_update_thresh,
1562 	.open		= graph_trace_open,
1563 	.pipe_open	= graph_trace_open,
1564 	.close		= graph_trace_close,
1565 	.pipe_close	= graph_trace_close,
1566 	.init		= graph_trace_init,
1567 	.reset		= graph_trace_reset,
1568 	.print_line	= print_graph_function,
1569 	.print_header	= print_graph_headers,
1570 	.flags		= &tracer_flags,
1571 	.set_flag	= func_graph_set_flag,
1572 	.allow_instances = true,
1573 #ifdef CONFIG_FTRACE_SELFTEST
1574 	.selftest	= trace_selftest_startup_function_graph,
1575 #endif
1576 };
1577 
1578 
1579 static ssize_t
1580 graph_depth_write(struct file *filp, const char __user *ubuf, size_t cnt,
1581 		  loff_t *ppos)
1582 {
1583 	unsigned long val;
1584 	int ret;
1585 
1586 	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
1587 	if (ret)
1588 		return ret;
1589 
1590 	fgraph_max_depth = val;
1591 
1592 	*ppos += cnt;
1593 
1594 	return cnt;
1595 }
1596 
1597 static ssize_t
1598 graph_depth_read(struct file *filp, char __user *ubuf, size_t cnt,
1599 		 loff_t *ppos)
1600 {
1601 	char buf[15]; /* More than enough to hold UINT_MAX + "\n"*/
1602 	int n;
1603 
1604 	n = sprintf(buf, "%d\n", fgraph_max_depth);
1605 
1606 	return simple_read_from_buffer(ubuf, cnt, ppos, buf, n);
1607 }
1608 
1609 static const struct file_operations graph_depth_fops = {
1610 	.open		= tracing_open_generic,
1611 	.write		= graph_depth_write,
1612 	.read		= graph_depth_read,
1613 	.llseek		= generic_file_llseek,
1614 };
1615 
1616 static __init int init_graph_tracefs(void)
1617 {
1618 	int ret;
1619 
1620 	ret = tracing_init_dentry();
1621 	if (ret)
1622 		return 0;
1623 
1624 	trace_create_file("max_graph_depth", TRACE_MODE_WRITE, NULL,
1625 			  NULL, &graph_depth_fops);
1626 
1627 	return 0;
1628 }
1629 fs_initcall(init_graph_tracefs);
1630 
1631 static __init int init_graph_trace(void)
1632 {
1633 	max_bytes_for_cpu = snprintf(NULL, 0, "%u", nr_cpu_ids - 1);
1634 
1635 	if (!register_trace_event(&graph_trace_entry_event)) {
1636 		pr_warn("Warning: could not register graph trace events\n");
1637 		return 1;
1638 	}
1639 
1640 #ifdef CONFIG_FUNCTION_GRAPH_RETADDR
1641 	if (!register_trace_event(&graph_trace_retaddr_entry_event)) {
1642 		pr_warn("Warning: could not register graph trace retaddr events\n");
1643 		return 1;
1644 	}
1645 #endif
1646 
1647 	if (!register_trace_event(&graph_trace_ret_event)) {
1648 		pr_warn("Warning: could not register graph trace events\n");
1649 		return 1;
1650 	}
1651 
1652 	return register_tracer(&graph_trace);
1653 }
1654 
1655 core_initcall(init_graph_trace);
1656