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