xref: /linux/kernel/trace/ftrace.c (revision 2dbf708448c836754d25fe6108c5bfe1f5697c95)
1 /*
2  * Infrastructure for profiling code inserted by 'gcc -pg'.
3  *
4  * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5  * Copyright (C) 2004-2008 Ingo Molnar <mingo@redhat.com>
6  *
7  * Originally ported from the -rt patch by:
8  *   Copyright (C) 2007 Arnaldo Carvalho de Melo <acme@redhat.com>
9  *
10  * Based on code in the latency_tracer, that is:
11  *
12  *  Copyright (C) 2004-2006 Ingo Molnar
13  *  Copyright (C) 2004 William Lee Irwin III
14  */
15 
16 #include <linux/stop_machine.h>
17 #include <linux/clocksource.h>
18 #include <linux/kallsyms.h>
19 #include <linux/seq_file.h>
20 #include <linux/suspend.h>
21 #include <linux/debugfs.h>
22 #include <linux/hardirq.h>
23 #include <linux/kthread.h>
24 #include <linux/uaccess.h>
25 #include <linux/bsearch.h>
26 #include <linux/module.h>
27 #include <linux/ftrace.h>
28 #include <linux/sysctl.h>
29 #include <linux/slab.h>
30 #include <linux/ctype.h>
31 #include <linux/sort.h>
32 #include <linux/list.h>
33 #include <linux/hash.h>
34 #include <linux/rcupdate.h>
35 
36 #include <trace/events/sched.h>
37 
38 #include <asm/setup.h>
39 
40 #include "trace_output.h"
41 #include "trace_stat.h"
42 
43 #define FTRACE_WARN_ON(cond)			\
44 	({					\
45 		int ___r = cond;		\
46 		if (WARN_ON(___r))		\
47 			ftrace_kill();		\
48 		___r;				\
49 	})
50 
51 #define FTRACE_WARN_ON_ONCE(cond)		\
52 	({					\
53 		int ___r = cond;		\
54 		if (WARN_ON_ONCE(___r))		\
55 			ftrace_kill();		\
56 		___r;				\
57 	})
58 
59 /* hash bits for specific function selection */
60 #define FTRACE_HASH_BITS 7
61 #define FTRACE_FUNC_HASHSIZE (1 << FTRACE_HASH_BITS)
62 #define FTRACE_HASH_DEFAULT_BITS 10
63 #define FTRACE_HASH_MAX_BITS 12
64 
65 #define FL_GLOBAL_CONTROL_MASK (FTRACE_OPS_FL_GLOBAL | FTRACE_OPS_FL_CONTROL)
66 
67 /* ftrace_enabled is a method to turn ftrace on or off */
68 int ftrace_enabled __read_mostly;
69 static int last_ftrace_enabled;
70 
71 /* Quick disabling of function tracer. */
72 int function_trace_stop;
73 
74 /* List for set_ftrace_pid's pids. */
75 LIST_HEAD(ftrace_pids);
76 struct ftrace_pid {
77 	struct list_head list;
78 	struct pid *pid;
79 };
80 
81 /*
82  * ftrace_disabled is set when an anomaly is discovered.
83  * ftrace_disabled is much stronger than ftrace_enabled.
84  */
85 static int ftrace_disabled __read_mostly;
86 
87 static DEFINE_MUTEX(ftrace_lock);
88 
89 static struct ftrace_ops ftrace_list_end __read_mostly = {
90 	.func		= ftrace_stub,
91 };
92 
93 static struct ftrace_ops *ftrace_global_list __read_mostly = &ftrace_list_end;
94 static struct ftrace_ops *ftrace_control_list __read_mostly = &ftrace_list_end;
95 static struct ftrace_ops *ftrace_ops_list __read_mostly = &ftrace_list_end;
96 ftrace_func_t ftrace_trace_function __read_mostly = ftrace_stub;
97 static ftrace_func_t __ftrace_trace_function_delay __read_mostly = ftrace_stub;
98 ftrace_func_t __ftrace_trace_function __read_mostly = ftrace_stub;
99 ftrace_func_t ftrace_pid_function __read_mostly = ftrace_stub;
100 static struct ftrace_ops global_ops;
101 static struct ftrace_ops control_ops;
102 
103 static void
104 ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip);
105 
106 /*
107  * Traverse the ftrace_global_list, invoking all entries.  The reason that we
108  * can use rcu_dereference_raw() is that elements removed from this list
109  * are simply leaked, so there is no need to interact with a grace-period
110  * mechanism.  The rcu_dereference_raw() calls are needed to handle
111  * concurrent insertions into the ftrace_global_list.
112  *
113  * Silly Alpha and silly pointer-speculation compiler optimizations!
114  */
115 static void ftrace_global_list_func(unsigned long ip,
116 				    unsigned long parent_ip)
117 {
118 	struct ftrace_ops *op;
119 
120 	if (unlikely(trace_recursion_test(TRACE_GLOBAL_BIT)))
121 		return;
122 
123 	trace_recursion_set(TRACE_GLOBAL_BIT);
124 	op = rcu_dereference_raw(ftrace_global_list); /*see above*/
125 	while (op != &ftrace_list_end) {
126 		op->func(ip, parent_ip);
127 		op = rcu_dereference_raw(op->next); /*see above*/
128 	};
129 	trace_recursion_clear(TRACE_GLOBAL_BIT);
130 }
131 
132 static void ftrace_pid_func(unsigned long ip, unsigned long parent_ip)
133 {
134 	if (!test_tsk_trace_trace(current))
135 		return;
136 
137 	ftrace_pid_function(ip, parent_ip);
138 }
139 
140 static void set_ftrace_pid_function(ftrace_func_t func)
141 {
142 	/* do not set ftrace_pid_function to itself! */
143 	if (func != ftrace_pid_func)
144 		ftrace_pid_function = func;
145 }
146 
147 /**
148  * clear_ftrace_function - reset the ftrace function
149  *
150  * This NULLs the ftrace function and in essence stops
151  * tracing.  There may be lag
152  */
153 void clear_ftrace_function(void)
154 {
155 	ftrace_trace_function = ftrace_stub;
156 	__ftrace_trace_function = ftrace_stub;
157 	__ftrace_trace_function_delay = ftrace_stub;
158 	ftrace_pid_function = ftrace_stub;
159 }
160 
161 #ifndef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
162 /*
163  * For those archs that do not test ftrace_trace_stop in their
164  * mcount call site, we need to do it from C.
165  */
166 static void ftrace_test_stop_func(unsigned long ip, unsigned long parent_ip)
167 {
168 	if (function_trace_stop)
169 		return;
170 
171 	__ftrace_trace_function(ip, parent_ip);
172 }
173 #endif
174 
175 static void control_ops_disable_all(struct ftrace_ops *ops)
176 {
177 	int cpu;
178 
179 	for_each_possible_cpu(cpu)
180 		*per_cpu_ptr(ops->disabled, cpu) = 1;
181 }
182 
183 static int control_ops_alloc(struct ftrace_ops *ops)
184 {
185 	int __percpu *disabled;
186 
187 	disabled = alloc_percpu(int);
188 	if (!disabled)
189 		return -ENOMEM;
190 
191 	ops->disabled = disabled;
192 	control_ops_disable_all(ops);
193 	return 0;
194 }
195 
196 static void control_ops_free(struct ftrace_ops *ops)
197 {
198 	free_percpu(ops->disabled);
199 }
200 
201 static void update_global_ops(void)
202 {
203 	ftrace_func_t func;
204 
205 	/*
206 	 * If there's only one function registered, then call that
207 	 * function directly. Otherwise, we need to iterate over the
208 	 * registered callers.
209 	 */
210 	if (ftrace_global_list == &ftrace_list_end ||
211 	    ftrace_global_list->next == &ftrace_list_end)
212 		func = ftrace_global_list->func;
213 	else
214 		func = ftrace_global_list_func;
215 
216 	/* If we filter on pids, update to use the pid function */
217 	if (!list_empty(&ftrace_pids)) {
218 		set_ftrace_pid_function(func);
219 		func = ftrace_pid_func;
220 	}
221 
222 	global_ops.func = func;
223 }
224 
225 static void update_ftrace_function(void)
226 {
227 	ftrace_func_t func;
228 
229 	update_global_ops();
230 
231 	/*
232 	 * If we are at the end of the list and this ops is
233 	 * not dynamic, then have the mcount trampoline call
234 	 * the function directly
235 	 */
236 	if (ftrace_ops_list == &ftrace_list_end ||
237 	    (ftrace_ops_list->next == &ftrace_list_end &&
238 	     !(ftrace_ops_list->flags & FTRACE_OPS_FL_DYNAMIC)))
239 		func = ftrace_ops_list->func;
240 	else
241 		func = ftrace_ops_list_func;
242 
243 #ifdef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
244 	ftrace_trace_function = func;
245 #else
246 #ifdef CONFIG_DYNAMIC_FTRACE
247 	/* do not update till all functions have been modified */
248 	__ftrace_trace_function_delay = func;
249 #else
250 	__ftrace_trace_function = func;
251 #endif
252 	ftrace_trace_function =
253 		(func == ftrace_stub) ? func : ftrace_test_stop_func;
254 #endif
255 }
256 
257 static void add_ftrace_ops(struct ftrace_ops **list, struct ftrace_ops *ops)
258 {
259 	ops->next = *list;
260 	/*
261 	 * We are entering ops into the list but another
262 	 * CPU might be walking that list. We need to make sure
263 	 * the ops->next pointer is valid before another CPU sees
264 	 * the ops pointer included into the list.
265 	 */
266 	rcu_assign_pointer(*list, ops);
267 }
268 
269 static int remove_ftrace_ops(struct ftrace_ops **list, struct ftrace_ops *ops)
270 {
271 	struct ftrace_ops **p;
272 
273 	/*
274 	 * If we are removing the last function, then simply point
275 	 * to the ftrace_stub.
276 	 */
277 	if (*list == ops && ops->next == &ftrace_list_end) {
278 		*list = &ftrace_list_end;
279 		return 0;
280 	}
281 
282 	for (p = list; *p != &ftrace_list_end; p = &(*p)->next)
283 		if (*p == ops)
284 			break;
285 
286 	if (*p != ops)
287 		return -1;
288 
289 	*p = (*p)->next;
290 	return 0;
291 }
292 
293 static void add_ftrace_list_ops(struct ftrace_ops **list,
294 				struct ftrace_ops *main_ops,
295 				struct ftrace_ops *ops)
296 {
297 	int first = *list == &ftrace_list_end;
298 	add_ftrace_ops(list, ops);
299 	if (first)
300 		add_ftrace_ops(&ftrace_ops_list, main_ops);
301 }
302 
303 static int remove_ftrace_list_ops(struct ftrace_ops **list,
304 				  struct ftrace_ops *main_ops,
305 				  struct ftrace_ops *ops)
306 {
307 	int ret = remove_ftrace_ops(list, ops);
308 	if (!ret && *list == &ftrace_list_end)
309 		ret = remove_ftrace_ops(&ftrace_ops_list, main_ops);
310 	return ret;
311 }
312 
313 static int __register_ftrace_function(struct ftrace_ops *ops)
314 {
315 	if (ftrace_disabled)
316 		return -ENODEV;
317 
318 	if (FTRACE_WARN_ON(ops == &global_ops))
319 		return -EINVAL;
320 
321 	if (WARN_ON(ops->flags & FTRACE_OPS_FL_ENABLED))
322 		return -EBUSY;
323 
324 	/* We don't support both control and global flags set. */
325 	if ((ops->flags & FL_GLOBAL_CONTROL_MASK) == FL_GLOBAL_CONTROL_MASK)
326 		return -EINVAL;
327 
328 	if (!core_kernel_data((unsigned long)ops))
329 		ops->flags |= FTRACE_OPS_FL_DYNAMIC;
330 
331 	if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
332 		add_ftrace_list_ops(&ftrace_global_list, &global_ops, ops);
333 		ops->flags |= FTRACE_OPS_FL_ENABLED;
334 	} else if (ops->flags & FTRACE_OPS_FL_CONTROL) {
335 		if (control_ops_alloc(ops))
336 			return -ENOMEM;
337 		add_ftrace_list_ops(&ftrace_control_list, &control_ops, ops);
338 	} else
339 		add_ftrace_ops(&ftrace_ops_list, ops);
340 
341 	if (ftrace_enabled)
342 		update_ftrace_function();
343 
344 	return 0;
345 }
346 
347 static int __unregister_ftrace_function(struct ftrace_ops *ops)
348 {
349 	int ret;
350 
351 	if (ftrace_disabled)
352 		return -ENODEV;
353 
354 	if (WARN_ON(!(ops->flags & FTRACE_OPS_FL_ENABLED)))
355 		return -EBUSY;
356 
357 	if (FTRACE_WARN_ON(ops == &global_ops))
358 		return -EINVAL;
359 
360 	if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
361 		ret = remove_ftrace_list_ops(&ftrace_global_list,
362 					     &global_ops, ops);
363 		if (!ret)
364 			ops->flags &= ~FTRACE_OPS_FL_ENABLED;
365 	} else if (ops->flags & FTRACE_OPS_FL_CONTROL) {
366 		ret = remove_ftrace_list_ops(&ftrace_control_list,
367 					     &control_ops, ops);
368 		if (!ret) {
369 			/*
370 			 * The ftrace_ops is now removed from the list,
371 			 * so there'll be no new users. We must ensure
372 			 * all current users are done before we free
373 			 * the control data.
374 			 */
375 			synchronize_sched();
376 			control_ops_free(ops);
377 		}
378 	} else
379 		ret = remove_ftrace_ops(&ftrace_ops_list, ops);
380 
381 	if (ret < 0)
382 		return ret;
383 
384 	if (ftrace_enabled)
385 		update_ftrace_function();
386 
387 	/*
388 	 * Dynamic ops may be freed, we must make sure that all
389 	 * callers are done before leaving this function.
390 	 */
391 	if (ops->flags & FTRACE_OPS_FL_DYNAMIC)
392 		synchronize_sched();
393 
394 	return 0;
395 }
396 
397 static void ftrace_update_pid_func(void)
398 {
399 	/* Only do something if we are tracing something */
400 	if (ftrace_trace_function == ftrace_stub)
401 		return;
402 
403 	update_ftrace_function();
404 }
405 
406 #ifdef CONFIG_FUNCTION_PROFILER
407 struct ftrace_profile {
408 	struct hlist_node		node;
409 	unsigned long			ip;
410 	unsigned long			counter;
411 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
412 	unsigned long long		time;
413 	unsigned long long		time_squared;
414 #endif
415 };
416 
417 struct ftrace_profile_page {
418 	struct ftrace_profile_page	*next;
419 	unsigned long			index;
420 	struct ftrace_profile		records[];
421 };
422 
423 struct ftrace_profile_stat {
424 	atomic_t			disabled;
425 	struct hlist_head		*hash;
426 	struct ftrace_profile_page	*pages;
427 	struct ftrace_profile_page	*start;
428 	struct tracer_stat		stat;
429 };
430 
431 #define PROFILE_RECORDS_SIZE						\
432 	(PAGE_SIZE - offsetof(struct ftrace_profile_page, records))
433 
434 #define PROFILES_PER_PAGE					\
435 	(PROFILE_RECORDS_SIZE / sizeof(struct ftrace_profile))
436 
437 static int ftrace_profile_bits __read_mostly;
438 static int ftrace_profile_enabled __read_mostly;
439 
440 /* ftrace_profile_lock - synchronize the enable and disable of the profiler */
441 static DEFINE_MUTEX(ftrace_profile_lock);
442 
443 static DEFINE_PER_CPU(struct ftrace_profile_stat, ftrace_profile_stats);
444 
445 #define FTRACE_PROFILE_HASH_SIZE 1024 /* must be power of 2 */
446 
447 static void *
448 function_stat_next(void *v, int idx)
449 {
450 	struct ftrace_profile *rec = v;
451 	struct ftrace_profile_page *pg;
452 
453 	pg = (struct ftrace_profile_page *)((unsigned long)rec & PAGE_MASK);
454 
455  again:
456 	if (idx != 0)
457 		rec++;
458 
459 	if ((void *)rec >= (void *)&pg->records[pg->index]) {
460 		pg = pg->next;
461 		if (!pg)
462 			return NULL;
463 		rec = &pg->records[0];
464 		if (!rec->counter)
465 			goto again;
466 	}
467 
468 	return rec;
469 }
470 
471 static void *function_stat_start(struct tracer_stat *trace)
472 {
473 	struct ftrace_profile_stat *stat =
474 		container_of(trace, struct ftrace_profile_stat, stat);
475 
476 	if (!stat || !stat->start)
477 		return NULL;
478 
479 	return function_stat_next(&stat->start->records[0], 0);
480 }
481 
482 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
483 /* function graph compares on total time */
484 static int function_stat_cmp(void *p1, void *p2)
485 {
486 	struct ftrace_profile *a = p1;
487 	struct ftrace_profile *b = p2;
488 
489 	if (a->time < b->time)
490 		return -1;
491 	if (a->time > b->time)
492 		return 1;
493 	else
494 		return 0;
495 }
496 #else
497 /* not function graph compares against hits */
498 static int function_stat_cmp(void *p1, void *p2)
499 {
500 	struct ftrace_profile *a = p1;
501 	struct ftrace_profile *b = p2;
502 
503 	if (a->counter < b->counter)
504 		return -1;
505 	if (a->counter > b->counter)
506 		return 1;
507 	else
508 		return 0;
509 }
510 #endif
511 
512 static int function_stat_headers(struct seq_file *m)
513 {
514 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
515 	seq_printf(m, "  Function                               "
516 		   "Hit    Time            Avg             s^2\n"
517 		      "  --------                               "
518 		   "---    ----            ---             ---\n");
519 #else
520 	seq_printf(m, "  Function                               Hit\n"
521 		      "  --------                               ---\n");
522 #endif
523 	return 0;
524 }
525 
526 static int function_stat_show(struct seq_file *m, void *v)
527 {
528 	struct ftrace_profile *rec = v;
529 	char str[KSYM_SYMBOL_LEN];
530 	int ret = 0;
531 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
532 	static struct trace_seq s;
533 	unsigned long long avg;
534 	unsigned long long stddev;
535 #endif
536 	mutex_lock(&ftrace_profile_lock);
537 
538 	/* we raced with function_profile_reset() */
539 	if (unlikely(rec->counter == 0)) {
540 		ret = -EBUSY;
541 		goto out;
542 	}
543 
544 	kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
545 	seq_printf(m, "  %-30.30s  %10lu", str, rec->counter);
546 
547 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
548 	seq_printf(m, "    ");
549 	avg = rec->time;
550 	do_div(avg, rec->counter);
551 
552 	/* Sample standard deviation (s^2) */
553 	if (rec->counter <= 1)
554 		stddev = 0;
555 	else {
556 		stddev = rec->time_squared - rec->counter * avg * avg;
557 		/*
558 		 * Divide only 1000 for ns^2 -> us^2 conversion.
559 		 * trace_print_graph_duration will divide 1000 again.
560 		 */
561 		do_div(stddev, (rec->counter - 1) * 1000);
562 	}
563 
564 	trace_seq_init(&s);
565 	trace_print_graph_duration(rec->time, &s);
566 	trace_seq_puts(&s, "    ");
567 	trace_print_graph_duration(avg, &s);
568 	trace_seq_puts(&s, "    ");
569 	trace_print_graph_duration(stddev, &s);
570 	trace_print_seq(m, &s);
571 #endif
572 	seq_putc(m, '\n');
573 out:
574 	mutex_unlock(&ftrace_profile_lock);
575 
576 	return ret;
577 }
578 
579 static void ftrace_profile_reset(struct ftrace_profile_stat *stat)
580 {
581 	struct ftrace_profile_page *pg;
582 
583 	pg = stat->pages = stat->start;
584 
585 	while (pg) {
586 		memset(pg->records, 0, PROFILE_RECORDS_SIZE);
587 		pg->index = 0;
588 		pg = pg->next;
589 	}
590 
591 	memset(stat->hash, 0,
592 	       FTRACE_PROFILE_HASH_SIZE * sizeof(struct hlist_head));
593 }
594 
595 int ftrace_profile_pages_init(struct ftrace_profile_stat *stat)
596 {
597 	struct ftrace_profile_page *pg;
598 	int functions;
599 	int pages;
600 	int i;
601 
602 	/* If we already allocated, do nothing */
603 	if (stat->pages)
604 		return 0;
605 
606 	stat->pages = (void *)get_zeroed_page(GFP_KERNEL);
607 	if (!stat->pages)
608 		return -ENOMEM;
609 
610 #ifdef CONFIG_DYNAMIC_FTRACE
611 	functions = ftrace_update_tot_cnt;
612 #else
613 	/*
614 	 * We do not know the number of functions that exist because
615 	 * dynamic tracing is what counts them. With past experience
616 	 * we have around 20K functions. That should be more than enough.
617 	 * It is highly unlikely we will execute every function in
618 	 * the kernel.
619 	 */
620 	functions = 20000;
621 #endif
622 
623 	pg = stat->start = stat->pages;
624 
625 	pages = DIV_ROUND_UP(functions, PROFILES_PER_PAGE);
626 
627 	for (i = 0; i < pages; i++) {
628 		pg->next = (void *)get_zeroed_page(GFP_KERNEL);
629 		if (!pg->next)
630 			goto out_free;
631 		pg = pg->next;
632 	}
633 
634 	return 0;
635 
636  out_free:
637 	pg = stat->start;
638 	while (pg) {
639 		unsigned long tmp = (unsigned long)pg;
640 
641 		pg = pg->next;
642 		free_page(tmp);
643 	}
644 
645 	free_page((unsigned long)stat->pages);
646 	stat->pages = NULL;
647 	stat->start = NULL;
648 
649 	return -ENOMEM;
650 }
651 
652 static int ftrace_profile_init_cpu(int cpu)
653 {
654 	struct ftrace_profile_stat *stat;
655 	int size;
656 
657 	stat = &per_cpu(ftrace_profile_stats, cpu);
658 
659 	if (stat->hash) {
660 		/* If the profile is already created, simply reset it */
661 		ftrace_profile_reset(stat);
662 		return 0;
663 	}
664 
665 	/*
666 	 * We are profiling all functions, but usually only a few thousand
667 	 * functions are hit. We'll make a hash of 1024 items.
668 	 */
669 	size = FTRACE_PROFILE_HASH_SIZE;
670 
671 	stat->hash = kzalloc(sizeof(struct hlist_head) * size, GFP_KERNEL);
672 
673 	if (!stat->hash)
674 		return -ENOMEM;
675 
676 	if (!ftrace_profile_bits) {
677 		size--;
678 
679 		for (; size; size >>= 1)
680 			ftrace_profile_bits++;
681 	}
682 
683 	/* Preallocate the function profiling pages */
684 	if (ftrace_profile_pages_init(stat) < 0) {
685 		kfree(stat->hash);
686 		stat->hash = NULL;
687 		return -ENOMEM;
688 	}
689 
690 	return 0;
691 }
692 
693 static int ftrace_profile_init(void)
694 {
695 	int cpu;
696 	int ret = 0;
697 
698 	for_each_online_cpu(cpu) {
699 		ret = ftrace_profile_init_cpu(cpu);
700 		if (ret)
701 			break;
702 	}
703 
704 	return ret;
705 }
706 
707 /* interrupts must be disabled */
708 static struct ftrace_profile *
709 ftrace_find_profiled_func(struct ftrace_profile_stat *stat, unsigned long ip)
710 {
711 	struct ftrace_profile *rec;
712 	struct hlist_head *hhd;
713 	struct hlist_node *n;
714 	unsigned long key;
715 
716 	key = hash_long(ip, ftrace_profile_bits);
717 	hhd = &stat->hash[key];
718 
719 	if (hlist_empty(hhd))
720 		return NULL;
721 
722 	hlist_for_each_entry_rcu(rec, n, hhd, node) {
723 		if (rec->ip == ip)
724 			return rec;
725 	}
726 
727 	return NULL;
728 }
729 
730 static void ftrace_add_profile(struct ftrace_profile_stat *stat,
731 			       struct ftrace_profile *rec)
732 {
733 	unsigned long key;
734 
735 	key = hash_long(rec->ip, ftrace_profile_bits);
736 	hlist_add_head_rcu(&rec->node, &stat->hash[key]);
737 }
738 
739 /*
740  * The memory is already allocated, this simply finds a new record to use.
741  */
742 static struct ftrace_profile *
743 ftrace_profile_alloc(struct ftrace_profile_stat *stat, unsigned long ip)
744 {
745 	struct ftrace_profile *rec = NULL;
746 
747 	/* prevent recursion (from NMIs) */
748 	if (atomic_inc_return(&stat->disabled) != 1)
749 		goto out;
750 
751 	/*
752 	 * Try to find the function again since an NMI
753 	 * could have added it
754 	 */
755 	rec = ftrace_find_profiled_func(stat, ip);
756 	if (rec)
757 		goto out;
758 
759 	if (stat->pages->index == PROFILES_PER_PAGE) {
760 		if (!stat->pages->next)
761 			goto out;
762 		stat->pages = stat->pages->next;
763 	}
764 
765 	rec = &stat->pages->records[stat->pages->index++];
766 	rec->ip = ip;
767 	ftrace_add_profile(stat, rec);
768 
769  out:
770 	atomic_dec(&stat->disabled);
771 
772 	return rec;
773 }
774 
775 static void
776 function_profile_call(unsigned long ip, unsigned long parent_ip)
777 {
778 	struct ftrace_profile_stat *stat;
779 	struct ftrace_profile *rec;
780 	unsigned long flags;
781 
782 	if (!ftrace_profile_enabled)
783 		return;
784 
785 	local_irq_save(flags);
786 
787 	stat = &__get_cpu_var(ftrace_profile_stats);
788 	if (!stat->hash || !ftrace_profile_enabled)
789 		goto out;
790 
791 	rec = ftrace_find_profiled_func(stat, ip);
792 	if (!rec) {
793 		rec = ftrace_profile_alloc(stat, ip);
794 		if (!rec)
795 			goto out;
796 	}
797 
798 	rec->counter++;
799  out:
800 	local_irq_restore(flags);
801 }
802 
803 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
804 static int profile_graph_entry(struct ftrace_graph_ent *trace)
805 {
806 	function_profile_call(trace->func, 0);
807 	return 1;
808 }
809 
810 static void profile_graph_return(struct ftrace_graph_ret *trace)
811 {
812 	struct ftrace_profile_stat *stat;
813 	unsigned long long calltime;
814 	struct ftrace_profile *rec;
815 	unsigned long flags;
816 
817 	local_irq_save(flags);
818 	stat = &__get_cpu_var(ftrace_profile_stats);
819 	if (!stat->hash || !ftrace_profile_enabled)
820 		goto out;
821 
822 	/* If the calltime was zero'd ignore it */
823 	if (!trace->calltime)
824 		goto out;
825 
826 	calltime = trace->rettime - trace->calltime;
827 
828 	if (!(trace_flags & TRACE_ITER_GRAPH_TIME)) {
829 		int index;
830 
831 		index = trace->depth;
832 
833 		/* Append this call time to the parent time to subtract */
834 		if (index)
835 			current->ret_stack[index - 1].subtime += calltime;
836 
837 		if (current->ret_stack[index].subtime < calltime)
838 			calltime -= current->ret_stack[index].subtime;
839 		else
840 			calltime = 0;
841 	}
842 
843 	rec = ftrace_find_profiled_func(stat, trace->func);
844 	if (rec) {
845 		rec->time += calltime;
846 		rec->time_squared += calltime * calltime;
847 	}
848 
849  out:
850 	local_irq_restore(flags);
851 }
852 
853 static int register_ftrace_profiler(void)
854 {
855 	return register_ftrace_graph(&profile_graph_return,
856 				     &profile_graph_entry);
857 }
858 
859 static void unregister_ftrace_profiler(void)
860 {
861 	unregister_ftrace_graph();
862 }
863 #else
864 static struct ftrace_ops ftrace_profile_ops __read_mostly = {
865 	.func		= function_profile_call,
866 };
867 
868 static int register_ftrace_profiler(void)
869 {
870 	return register_ftrace_function(&ftrace_profile_ops);
871 }
872 
873 static void unregister_ftrace_profiler(void)
874 {
875 	unregister_ftrace_function(&ftrace_profile_ops);
876 }
877 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
878 
879 static ssize_t
880 ftrace_profile_write(struct file *filp, const char __user *ubuf,
881 		     size_t cnt, loff_t *ppos)
882 {
883 	unsigned long val;
884 	int ret;
885 
886 	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
887 	if (ret)
888 		return ret;
889 
890 	val = !!val;
891 
892 	mutex_lock(&ftrace_profile_lock);
893 	if (ftrace_profile_enabled ^ val) {
894 		if (val) {
895 			ret = ftrace_profile_init();
896 			if (ret < 0) {
897 				cnt = ret;
898 				goto out;
899 			}
900 
901 			ret = register_ftrace_profiler();
902 			if (ret < 0) {
903 				cnt = ret;
904 				goto out;
905 			}
906 			ftrace_profile_enabled = 1;
907 		} else {
908 			ftrace_profile_enabled = 0;
909 			/*
910 			 * unregister_ftrace_profiler calls stop_machine
911 			 * so this acts like an synchronize_sched.
912 			 */
913 			unregister_ftrace_profiler();
914 		}
915 	}
916  out:
917 	mutex_unlock(&ftrace_profile_lock);
918 
919 	*ppos += cnt;
920 
921 	return cnt;
922 }
923 
924 static ssize_t
925 ftrace_profile_read(struct file *filp, char __user *ubuf,
926 		     size_t cnt, loff_t *ppos)
927 {
928 	char buf[64];		/* big enough to hold a number */
929 	int r;
930 
931 	r = sprintf(buf, "%u\n", ftrace_profile_enabled);
932 	return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
933 }
934 
935 static const struct file_operations ftrace_profile_fops = {
936 	.open		= tracing_open_generic,
937 	.read		= ftrace_profile_read,
938 	.write		= ftrace_profile_write,
939 	.llseek		= default_llseek,
940 };
941 
942 /* used to initialize the real stat files */
943 static struct tracer_stat function_stats __initdata = {
944 	.name		= "functions",
945 	.stat_start	= function_stat_start,
946 	.stat_next	= function_stat_next,
947 	.stat_cmp	= function_stat_cmp,
948 	.stat_headers	= function_stat_headers,
949 	.stat_show	= function_stat_show
950 };
951 
952 static __init void ftrace_profile_debugfs(struct dentry *d_tracer)
953 {
954 	struct ftrace_profile_stat *stat;
955 	struct dentry *entry;
956 	char *name;
957 	int ret;
958 	int cpu;
959 
960 	for_each_possible_cpu(cpu) {
961 		stat = &per_cpu(ftrace_profile_stats, cpu);
962 
963 		/* allocate enough for function name + cpu number */
964 		name = kmalloc(32, GFP_KERNEL);
965 		if (!name) {
966 			/*
967 			 * The files created are permanent, if something happens
968 			 * we still do not free memory.
969 			 */
970 			WARN(1,
971 			     "Could not allocate stat file for cpu %d\n",
972 			     cpu);
973 			return;
974 		}
975 		stat->stat = function_stats;
976 		snprintf(name, 32, "function%d", cpu);
977 		stat->stat.name = name;
978 		ret = register_stat_tracer(&stat->stat);
979 		if (ret) {
980 			WARN(1,
981 			     "Could not register function stat for cpu %d\n",
982 			     cpu);
983 			kfree(name);
984 			return;
985 		}
986 	}
987 
988 	entry = debugfs_create_file("function_profile_enabled", 0644,
989 				    d_tracer, NULL, &ftrace_profile_fops);
990 	if (!entry)
991 		pr_warning("Could not create debugfs "
992 			   "'function_profile_enabled' entry\n");
993 }
994 
995 #else /* CONFIG_FUNCTION_PROFILER */
996 static __init void ftrace_profile_debugfs(struct dentry *d_tracer)
997 {
998 }
999 #endif /* CONFIG_FUNCTION_PROFILER */
1000 
1001 static struct pid * const ftrace_swapper_pid = &init_struct_pid;
1002 
1003 #ifdef CONFIG_DYNAMIC_FTRACE
1004 
1005 #ifndef CONFIG_FTRACE_MCOUNT_RECORD
1006 # error Dynamic ftrace depends on MCOUNT_RECORD
1007 #endif
1008 
1009 static struct hlist_head ftrace_func_hash[FTRACE_FUNC_HASHSIZE] __read_mostly;
1010 
1011 struct ftrace_func_probe {
1012 	struct hlist_node	node;
1013 	struct ftrace_probe_ops	*ops;
1014 	unsigned long		flags;
1015 	unsigned long		ip;
1016 	void			*data;
1017 	struct rcu_head		rcu;
1018 };
1019 
1020 struct ftrace_func_entry {
1021 	struct hlist_node hlist;
1022 	unsigned long ip;
1023 };
1024 
1025 struct ftrace_hash {
1026 	unsigned long		size_bits;
1027 	struct hlist_head	*buckets;
1028 	unsigned long		count;
1029 	struct rcu_head		rcu;
1030 };
1031 
1032 /*
1033  * We make these constant because no one should touch them,
1034  * but they are used as the default "empty hash", to avoid allocating
1035  * it all the time. These are in a read only section such that if
1036  * anyone does try to modify it, it will cause an exception.
1037  */
1038 static const struct hlist_head empty_buckets[1];
1039 static const struct ftrace_hash empty_hash = {
1040 	.buckets = (struct hlist_head *)empty_buckets,
1041 };
1042 #define EMPTY_HASH	((struct ftrace_hash *)&empty_hash)
1043 
1044 static struct ftrace_ops global_ops = {
1045 	.func			= ftrace_stub,
1046 	.notrace_hash		= EMPTY_HASH,
1047 	.filter_hash		= EMPTY_HASH,
1048 };
1049 
1050 static DEFINE_MUTEX(ftrace_regex_lock);
1051 
1052 struct ftrace_page {
1053 	struct ftrace_page	*next;
1054 	struct dyn_ftrace	*records;
1055 	int			index;
1056 	int			size;
1057 };
1058 
1059 static struct ftrace_page *ftrace_new_pgs;
1060 
1061 #define ENTRY_SIZE sizeof(struct dyn_ftrace)
1062 #define ENTRIES_PER_PAGE (PAGE_SIZE / ENTRY_SIZE)
1063 
1064 /* estimate from running different kernels */
1065 #define NR_TO_INIT		10000
1066 
1067 static struct ftrace_page	*ftrace_pages_start;
1068 static struct ftrace_page	*ftrace_pages;
1069 
1070 static bool ftrace_hash_empty(struct ftrace_hash *hash)
1071 {
1072 	return !hash || !hash->count;
1073 }
1074 
1075 static struct ftrace_func_entry *
1076 ftrace_lookup_ip(struct ftrace_hash *hash, unsigned long ip)
1077 {
1078 	unsigned long key;
1079 	struct ftrace_func_entry *entry;
1080 	struct hlist_head *hhd;
1081 	struct hlist_node *n;
1082 
1083 	if (ftrace_hash_empty(hash))
1084 		return NULL;
1085 
1086 	if (hash->size_bits > 0)
1087 		key = hash_long(ip, hash->size_bits);
1088 	else
1089 		key = 0;
1090 
1091 	hhd = &hash->buckets[key];
1092 
1093 	hlist_for_each_entry_rcu(entry, n, hhd, hlist) {
1094 		if (entry->ip == ip)
1095 			return entry;
1096 	}
1097 	return NULL;
1098 }
1099 
1100 static void __add_hash_entry(struct ftrace_hash *hash,
1101 			     struct ftrace_func_entry *entry)
1102 {
1103 	struct hlist_head *hhd;
1104 	unsigned long key;
1105 
1106 	if (hash->size_bits)
1107 		key = hash_long(entry->ip, hash->size_bits);
1108 	else
1109 		key = 0;
1110 
1111 	hhd = &hash->buckets[key];
1112 	hlist_add_head(&entry->hlist, hhd);
1113 	hash->count++;
1114 }
1115 
1116 static int add_hash_entry(struct ftrace_hash *hash, unsigned long ip)
1117 {
1118 	struct ftrace_func_entry *entry;
1119 
1120 	entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1121 	if (!entry)
1122 		return -ENOMEM;
1123 
1124 	entry->ip = ip;
1125 	__add_hash_entry(hash, entry);
1126 
1127 	return 0;
1128 }
1129 
1130 static void
1131 free_hash_entry(struct ftrace_hash *hash,
1132 		  struct ftrace_func_entry *entry)
1133 {
1134 	hlist_del(&entry->hlist);
1135 	kfree(entry);
1136 	hash->count--;
1137 }
1138 
1139 static void
1140 remove_hash_entry(struct ftrace_hash *hash,
1141 		  struct ftrace_func_entry *entry)
1142 {
1143 	hlist_del(&entry->hlist);
1144 	hash->count--;
1145 }
1146 
1147 static void ftrace_hash_clear(struct ftrace_hash *hash)
1148 {
1149 	struct hlist_head *hhd;
1150 	struct hlist_node *tp, *tn;
1151 	struct ftrace_func_entry *entry;
1152 	int size = 1 << hash->size_bits;
1153 	int i;
1154 
1155 	if (!hash->count)
1156 		return;
1157 
1158 	for (i = 0; i < size; i++) {
1159 		hhd = &hash->buckets[i];
1160 		hlist_for_each_entry_safe(entry, tp, tn, hhd, hlist)
1161 			free_hash_entry(hash, entry);
1162 	}
1163 	FTRACE_WARN_ON(hash->count);
1164 }
1165 
1166 static void free_ftrace_hash(struct ftrace_hash *hash)
1167 {
1168 	if (!hash || hash == EMPTY_HASH)
1169 		return;
1170 	ftrace_hash_clear(hash);
1171 	kfree(hash->buckets);
1172 	kfree(hash);
1173 }
1174 
1175 static void __free_ftrace_hash_rcu(struct rcu_head *rcu)
1176 {
1177 	struct ftrace_hash *hash;
1178 
1179 	hash = container_of(rcu, struct ftrace_hash, rcu);
1180 	free_ftrace_hash(hash);
1181 }
1182 
1183 static void free_ftrace_hash_rcu(struct ftrace_hash *hash)
1184 {
1185 	if (!hash || hash == EMPTY_HASH)
1186 		return;
1187 	call_rcu_sched(&hash->rcu, __free_ftrace_hash_rcu);
1188 }
1189 
1190 void ftrace_free_filter(struct ftrace_ops *ops)
1191 {
1192 	free_ftrace_hash(ops->filter_hash);
1193 	free_ftrace_hash(ops->notrace_hash);
1194 }
1195 
1196 static struct ftrace_hash *alloc_ftrace_hash(int size_bits)
1197 {
1198 	struct ftrace_hash *hash;
1199 	int size;
1200 
1201 	hash = kzalloc(sizeof(*hash), GFP_KERNEL);
1202 	if (!hash)
1203 		return NULL;
1204 
1205 	size = 1 << size_bits;
1206 	hash->buckets = kcalloc(size, sizeof(*hash->buckets), GFP_KERNEL);
1207 
1208 	if (!hash->buckets) {
1209 		kfree(hash);
1210 		return NULL;
1211 	}
1212 
1213 	hash->size_bits = size_bits;
1214 
1215 	return hash;
1216 }
1217 
1218 static struct ftrace_hash *
1219 alloc_and_copy_ftrace_hash(int size_bits, struct ftrace_hash *hash)
1220 {
1221 	struct ftrace_func_entry *entry;
1222 	struct ftrace_hash *new_hash;
1223 	struct hlist_node *tp;
1224 	int size;
1225 	int ret;
1226 	int i;
1227 
1228 	new_hash = alloc_ftrace_hash(size_bits);
1229 	if (!new_hash)
1230 		return NULL;
1231 
1232 	/* Empty hash? */
1233 	if (ftrace_hash_empty(hash))
1234 		return new_hash;
1235 
1236 	size = 1 << hash->size_bits;
1237 	for (i = 0; i < size; i++) {
1238 		hlist_for_each_entry(entry, tp, &hash->buckets[i], hlist) {
1239 			ret = add_hash_entry(new_hash, entry->ip);
1240 			if (ret < 0)
1241 				goto free_hash;
1242 		}
1243 	}
1244 
1245 	FTRACE_WARN_ON(new_hash->count != hash->count);
1246 
1247 	return new_hash;
1248 
1249  free_hash:
1250 	free_ftrace_hash(new_hash);
1251 	return NULL;
1252 }
1253 
1254 static void
1255 ftrace_hash_rec_disable(struct ftrace_ops *ops, int filter_hash);
1256 static void
1257 ftrace_hash_rec_enable(struct ftrace_ops *ops, int filter_hash);
1258 
1259 static int
1260 ftrace_hash_move(struct ftrace_ops *ops, int enable,
1261 		 struct ftrace_hash **dst, struct ftrace_hash *src)
1262 {
1263 	struct ftrace_func_entry *entry;
1264 	struct hlist_node *tp, *tn;
1265 	struct hlist_head *hhd;
1266 	struct ftrace_hash *old_hash;
1267 	struct ftrace_hash *new_hash;
1268 	unsigned long key;
1269 	int size = src->count;
1270 	int bits = 0;
1271 	int ret;
1272 	int i;
1273 
1274 	/*
1275 	 * Remove the current set, update the hash and add
1276 	 * them back.
1277 	 */
1278 	ftrace_hash_rec_disable(ops, enable);
1279 
1280 	/*
1281 	 * If the new source is empty, just free dst and assign it
1282 	 * the empty_hash.
1283 	 */
1284 	if (!src->count) {
1285 		free_ftrace_hash_rcu(*dst);
1286 		rcu_assign_pointer(*dst, EMPTY_HASH);
1287 		/* still need to update the function records */
1288 		ret = 0;
1289 		goto out;
1290 	}
1291 
1292 	/*
1293 	 * Make the hash size about 1/2 the # found
1294 	 */
1295 	for (size /= 2; size; size >>= 1)
1296 		bits++;
1297 
1298 	/* Don't allocate too much */
1299 	if (bits > FTRACE_HASH_MAX_BITS)
1300 		bits = FTRACE_HASH_MAX_BITS;
1301 
1302 	ret = -ENOMEM;
1303 	new_hash = alloc_ftrace_hash(bits);
1304 	if (!new_hash)
1305 		goto out;
1306 
1307 	size = 1 << src->size_bits;
1308 	for (i = 0; i < size; i++) {
1309 		hhd = &src->buckets[i];
1310 		hlist_for_each_entry_safe(entry, tp, tn, hhd, hlist) {
1311 			if (bits > 0)
1312 				key = hash_long(entry->ip, bits);
1313 			else
1314 				key = 0;
1315 			remove_hash_entry(src, entry);
1316 			__add_hash_entry(new_hash, entry);
1317 		}
1318 	}
1319 
1320 	old_hash = *dst;
1321 	rcu_assign_pointer(*dst, new_hash);
1322 	free_ftrace_hash_rcu(old_hash);
1323 
1324 	ret = 0;
1325  out:
1326 	/*
1327 	 * Enable regardless of ret:
1328 	 *  On success, we enable the new hash.
1329 	 *  On failure, we re-enable the original hash.
1330 	 */
1331 	ftrace_hash_rec_enable(ops, enable);
1332 
1333 	return ret;
1334 }
1335 
1336 /*
1337  * Test the hashes for this ops to see if we want to call
1338  * the ops->func or not.
1339  *
1340  * It's a match if the ip is in the ops->filter_hash or
1341  * the filter_hash does not exist or is empty,
1342  *  AND
1343  * the ip is not in the ops->notrace_hash.
1344  *
1345  * This needs to be called with preemption disabled as
1346  * the hashes are freed with call_rcu_sched().
1347  */
1348 static int
1349 ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip)
1350 {
1351 	struct ftrace_hash *filter_hash;
1352 	struct ftrace_hash *notrace_hash;
1353 	int ret;
1354 
1355 	filter_hash = rcu_dereference_raw(ops->filter_hash);
1356 	notrace_hash = rcu_dereference_raw(ops->notrace_hash);
1357 
1358 	if ((ftrace_hash_empty(filter_hash) ||
1359 	     ftrace_lookup_ip(filter_hash, ip)) &&
1360 	    (ftrace_hash_empty(notrace_hash) ||
1361 	     !ftrace_lookup_ip(notrace_hash, ip)))
1362 		ret = 1;
1363 	else
1364 		ret = 0;
1365 
1366 	return ret;
1367 }
1368 
1369 /*
1370  * This is a double for. Do not use 'break' to break out of the loop,
1371  * you must use a goto.
1372  */
1373 #define do_for_each_ftrace_rec(pg, rec)					\
1374 	for (pg = ftrace_pages_start; pg; pg = pg->next) {		\
1375 		int _____i;						\
1376 		for (_____i = 0; _____i < pg->index; _____i++) {	\
1377 			rec = &pg->records[_____i];
1378 
1379 #define while_for_each_ftrace_rec()		\
1380 		}				\
1381 	}
1382 
1383 
1384 static int ftrace_cmp_recs(const void *a, const void *b)
1385 {
1386 	const struct dyn_ftrace *reca = a;
1387 	const struct dyn_ftrace *recb = b;
1388 
1389 	if (reca->ip > recb->ip)
1390 		return 1;
1391 	if (reca->ip < recb->ip)
1392 		return -1;
1393 	return 0;
1394 }
1395 
1396 /**
1397  * ftrace_location - return true if the ip giving is a traced location
1398  * @ip: the instruction pointer to check
1399  *
1400  * Returns 1 if @ip given is a pointer to a ftrace location.
1401  * That is, the instruction that is either a NOP or call to
1402  * the function tracer. It checks the ftrace internal tables to
1403  * determine if the address belongs or not.
1404  */
1405 int ftrace_location(unsigned long ip)
1406 {
1407 	struct ftrace_page *pg;
1408 	struct dyn_ftrace *rec;
1409 	struct dyn_ftrace key;
1410 
1411 	key.ip = ip;
1412 
1413 	for (pg = ftrace_pages_start; pg; pg = pg->next) {
1414 		rec = bsearch(&key, pg->records, pg->index,
1415 			      sizeof(struct dyn_ftrace),
1416 			      ftrace_cmp_recs);
1417 		if (rec)
1418 			return 1;
1419 	}
1420 
1421 	return 0;
1422 }
1423 
1424 static void __ftrace_hash_rec_update(struct ftrace_ops *ops,
1425 				     int filter_hash,
1426 				     bool inc)
1427 {
1428 	struct ftrace_hash *hash;
1429 	struct ftrace_hash *other_hash;
1430 	struct ftrace_page *pg;
1431 	struct dyn_ftrace *rec;
1432 	int count = 0;
1433 	int all = 0;
1434 
1435 	/* Only update if the ops has been registered */
1436 	if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1437 		return;
1438 
1439 	/*
1440 	 * In the filter_hash case:
1441 	 *   If the count is zero, we update all records.
1442 	 *   Otherwise we just update the items in the hash.
1443 	 *
1444 	 * In the notrace_hash case:
1445 	 *   We enable the update in the hash.
1446 	 *   As disabling notrace means enabling the tracing,
1447 	 *   and enabling notrace means disabling, the inc variable
1448 	 *   gets inversed.
1449 	 */
1450 	if (filter_hash) {
1451 		hash = ops->filter_hash;
1452 		other_hash = ops->notrace_hash;
1453 		if (ftrace_hash_empty(hash))
1454 			all = 1;
1455 	} else {
1456 		inc = !inc;
1457 		hash = ops->notrace_hash;
1458 		other_hash = ops->filter_hash;
1459 		/*
1460 		 * If the notrace hash has no items,
1461 		 * then there's nothing to do.
1462 		 */
1463 		if (ftrace_hash_empty(hash))
1464 			return;
1465 	}
1466 
1467 	do_for_each_ftrace_rec(pg, rec) {
1468 		int in_other_hash = 0;
1469 		int in_hash = 0;
1470 		int match = 0;
1471 
1472 		if (all) {
1473 			/*
1474 			 * Only the filter_hash affects all records.
1475 			 * Update if the record is not in the notrace hash.
1476 			 */
1477 			if (!other_hash || !ftrace_lookup_ip(other_hash, rec->ip))
1478 				match = 1;
1479 		} else {
1480 			in_hash = !!ftrace_lookup_ip(hash, rec->ip);
1481 			in_other_hash = !!ftrace_lookup_ip(other_hash, rec->ip);
1482 
1483 			/*
1484 			 *
1485 			 */
1486 			if (filter_hash && in_hash && !in_other_hash)
1487 				match = 1;
1488 			else if (!filter_hash && in_hash &&
1489 				 (in_other_hash || ftrace_hash_empty(other_hash)))
1490 				match = 1;
1491 		}
1492 		if (!match)
1493 			continue;
1494 
1495 		if (inc) {
1496 			rec->flags++;
1497 			if (FTRACE_WARN_ON((rec->flags & ~FTRACE_FL_MASK) == FTRACE_REF_MAX))
1498 				return;
1499 		} else {
1500 			if (FTRACE_WARN_ON((rec->flags & ~FTRACE_FL_MASK) == 0))
1501 				return;
1502 			rec->flags--;
1503 		}
1504 		count++;
1505 		/* Shortcut, if we handled all records, we are done. */
1506 		if (!all && count == hash->count)
1507 			return;
1508 	} while_for_each_ftrace_rec();
1509 }
1510 
1511 static void ftrace_hash_rec_disable(struct ftrace_ops *ops,
1512 				    int filter_hash)
1513 {
1514 	__ftrace_hash_rec_update(ops, filter_hash, 0);
1515 }
1516 
1517 static void ftrace_hash_rec_enable(struct ftrace_ops *ops,
1518 				   int filter_hash)
1519 {
1520 	__ftrace_hash_rec_update(ops, filter_hash, 1);
1521 }
1522 
1523 static struct dyn_ftrace *ftrace_alloc_dyn_node(unsigned long ip)
1524 {
1525 	if (ftrace_pages->index == ftrace_pages->size) {
1526 		/* We should have allocated enough */
1527 		if (WARN_ON(!ftrace_pages->next))
1528 			return NULL;
1529 		ftrace_pages = ftrace_pages->next;
1530 	}
1531 
1532 	return &ftrace_pages->records[ftrace_pages->index++];
1533 }
1534 
1535 static struct dyn_ftrace *
1536 ftrace_record_ip(unsigned long ip)
1537 {
1538 	struct dyn_ftrace *rec;
1539 
1540 	if (ftrace_disabled)
1541 		return NULL;
1542 
1543 	rec = ftrace_alloc_dyn_node(ip);
1544 	if (!rec)
1545 		return NULL;
1546 
1547 	rec->ip = ip;
1548 
1549 	return rec;
1550 }
1551 
1552 static void print_ip_ins(const char *fmt, unsigned char *p)
1553 {
1554 	int i;
1555 
1556 	printk(KERN_CONT "%s", fmt);
1557 
1558 	for (i = 0; i < MCOUNT_INSN_SIZE; i++)
1559 		printk(KERN_CONT "%s%02x", i ? ":" : "", p[i]);
1560 }
1561 
1562 /**
1563  * ftrace_bug - report and shutdown function tracer
1564  * @failed: The failed type (EFAULT, EINVAL, EPERM)
1565  * @ip: The address that failed
1566  *
1567  * The arch code that enables or disables the function tracing
1568  * can call ftrace_bug() when it has detected a problem in
1569  * modifying the code. @failed should be one of either:
1570  * EFAULT - if the problem happens on reading the @ip address
1571  * EINVAL - if what is read at @ip is not what was expected
1572  * EPERM - if the problem happens on writting to the @ip address
1573  */
1574 void ftrace_bug(int failed, unsigned long ip)
1575 {
1576 	switch (failed) {
1577 	case -EFAULT:
1578 		FTRACE_WARN_ON_ONCE(1);
1579 		pr_info("ftrace faulted on modifying ");
1580 		print_ip_sym(ip);
1581 		break;
1582 	case -EINVAL:
1583 		FTRACE_WARN_ON_ONCE(1);
1584 		pr_info("ftrace failed to modify ");
1585 		print_ip_sym(ip);
1586 		print_ip_ins(" actual: ", (unsigned char *)ip);
1587 		printk(KERN_CONT "\n");
1588 		break;
1589 	case -EPERM:
1590 		FTRACE_WARN_ON_ONCE(1);
1591 		pr_info("ftrace faulted on writing ");
1592 		print_ip_sym(ip);
1593 		break;
1594 	default:
1595 		FTRACE_WARN_ON_ONCE(1);
1596 		pr_info("ftrace faulted on unknown error ");
1597 		print_ip_sym(ip);
1598 	}
1599 }
1600 
1601 
1602 /* Return 1 if the address range is reserved for ftrace */
1603 int ftrace_text_reserved(void *start, void *end)
1604 {
1605 	struct dyn_ftrace *rec;
1606 	struct ftrace_page *pg;
1607 
1608 	do_for_each_ftrace_rec(pg, rec) {
1609 		if (rec->ip <= (unsigned long)end &&
1610 		    rec->ip + MCOUNT_INSN_SIZE > (unsigned long)start)
1611 			return 1;
1612 	} while_for_each_ftrace_rec();
1613 	return 0;
1614 }
1615 
1616 static int ftrace_check_record(struct dyn_ftrace *rec, int enable, int update)
1617 {
1618 	unsigned long flag = 0UL;
1619 
1620 	/*
1621 	 * If we are updating calls:
1622 	 *
1623 	 *   If the record has a ref count, then we need to enable it
1624 	 *   because someone is using it.
1625 	 *
1626 	 *   Otherwise we make sure its disabled.
1627 	 *
1628 	 * If we are disabling calls, then disable all records that
1629 	 * are enabled.
1630 	 */
1631 	if (enable && (rec->flags & ~FTRACE_FL_MASK))
1632 		flag = FTRACE_FL_ENABLED;
1633 
1634 	/* If the state of this record hasn't changed, then do nothing */
1635 	if ((rec->flags & FTRACE_FL_ENABLED) == flag)
1636 		return FTRACE_UPDATE_IGNORE;
1637 
1638 	if (flag) {
1639 		if (update)
1640 			rec->flags |= FTRACE_FL_ENABLED;
1641 		return FTRACE_UPDATE_MAKE_CALL;
1642 	}
1643 
1644 	if (update)
1645 		rec->flags &= ~FTRACE_FL_ENABLED;
1646 
1647 	return FTRACE_UPDATE_MAKE_NOP;
1648 }
1649 
1650 /**
1651  * ftrace_update_record, set a record that now is tracing or not
1652  * @rec: the record to update
1653  * @enable: set to 1 if the record is tracing, zero to force disable
1654  *
1655  * The records that represent all functions that can be traced need
1656  * to be updated when tracing has been enabled.
1657  */
1658 int ftrace_update_record(struct dyn_ftrace *rec, int enable)
1659 {
1660 	return ftrace_check_record(rec, enable, 1);
1661 }
1662 
1663 /**
1664  * ftrace_test_record, check if the record has been enabled or not
1665  * @rec: the record to test
1666  * @enable: set to 1 to check if enabled, 0 if it is disabled
1667  *
1668  * The arch code may need to test if a record is already set to
1669  * tracing to determine how to modify the function code that it
1670  * represents.
1671  */
1672 int ftrace_test_record(struct dyn_ftrace *rec, int enable)
1673 {
1674 	return ftrace_check_record(rec, enable, 0);
1675 }
1676 
1677 static int
1678 __ftrace_replace_code(struct dyn_ftrace *rec, int enable)
1679 {
1680 	unsigned long ftrace_addr;
1681 	int ret;
1682 
1683 	ftrace_addr = (unsigned long)FTRACE_ADDR;
1684 
1685 	ret = ftrace_update_record(rec, enable);
1686 
1687 	switch (ret) {
1688 	case FTRACE_UPDATE_IGNORE:
1689 		return 0;
1690 
1691 	case FTRACE_UPDATE_MAKE_CALL:
1692 		return ftrace_make_call(rec, ftrace_addr);
1693 
1694 	case FTRACE_UPDATE_MAKE_NOP:
1695 		return ftrace_make_nop(NULL, rec, ftrace_addr);
1696 	}
1697 
1698 	return -1; /* unknow ftrace bug */
1699 }
1700 
1701 static void ftrace_replace_code(int update)
1702 {
1703 	struct dyn_ftrace *rec;
1704 	struct ftrace_page *pg;
1705 	int failed;
1706 
1707 	if (unlikely(ftrace_disabled))
1708 		return;
1709 
1710 	do_for_each_ftrace_rec(pg, rec) {
1711 		failed = __ftrace_replace_code(rec, update);
1712 		if (failed) {
1713 			ftrace_bug(failed, rec->ip);
1714 			/* Stop processing */
1715 			return;
1716 		}
1717 	} while_for_each_ftrace_rec();
1718 }
1719 
1720 struct ftrace_rec_iter {
1721 	struct ftrace_page	*pg;
1722 	int			index;
1723 };
1724 
1725 /**
1726  * ftrace_rec_iter_start, start up iterating over traced functions
1727  *
1728  * Returns an iterator handle that is used to iterate over all
1729  * the records that represent address locations where functions
1730  * are traced.
1731  *
1732  * May return NULL if no records are available.
1733  */
1734 struct ftrace_rec_iter *ftrace_rec_iter_start(void)
1735 {
1736 	/*
1737 	 * We only use a single iterator.
1738 	 * Protected by the ftrace_lock mutex.
1739 	 */
1740 	static struct ftrace_rec_iter ftrace_rec_iter;
1741 	struct ftrace_rec_iter *iter = &ftrace_rec_iter;
1742 
1743 	iter->pg = ftrace_pages_start;
1744 	iter->index = 0;
1745 
1746 	/* Could have empty pages */
1747 	while (iter->pg && !iter->pg->index)
1748 		iter->pg = iter->pg->next;
1749 
1750 	if (!iter->pg)
1751 		return NULL;
1752 
1753 	return iter;
1754 }
1755 
1756 /**
1757  * ftrace_rec_iter_next, get the next record to process.
1758  * @iter: The handle to the iterator.
1759  *
1760  * Returns the next iterator after the given iterator @iter.
1761  */
1762 struct ftrace_rec_iter *ftrace_rec_iter_next(struct ftrace_rec_iter *iter)
1763 {
1764 	iter->index++;
1765 
1766 	if (iter->index >= iter->pg->index) {
1767 		iter->pg = iter->pg->next;
1768 		iter->index = 0;
1769 
1770 		/* Could have empty pages */
1771 		while (iter->pg && !iter->pg->index)
1772 			iter->pg = iter->pg->next;
1773 	}
1774 
1775 	if (!iter->pg)
1776 		return NULL;
1777 
1778 	return iter;
1779 }
1780 
1781 /**
1782  * ftrace_rec_iter_record, get the record at the iterator location
1783  * @iter: The current iterator location
1784  *
1785  * Returns the record that the current @iter is at.
1786  */
1787 struct dyn_ftrace *ftrace_rec_iter_record(struct ftrace_rec_iter *iter)
1788 {
1789 	return &iter->pg->records[iter->index];
1790 }
1791 
1792 static int
1793 ftrace_code_disable(struct module *mod, struct dyn_ftrace *rec)
1794 {
1795 	unsigned long ip;
1796 	int ret;
1797 
1798 	ip = rec->ip;
1799 
1800 	if (unlikely(ftrace_disabled))
1801 		return 0;
1802 
1803 	ret = ftrace_make_nop(mod, rec, MCOUNT_ADDR);
1804 	if (ret) {
1805 		ftrace_bug(ret, ip);
1806 		return 0;
1807 	}
1808 	return 1;
1809 }
1810 
1811 /*
1812  * archs can override this function if they must do something
1813  * before the modifying code is performed.
1814  */
1815 int __weak ftrace_arch_code_modify_prepare(void)
1816 {
1817 	return 0;
1818 }
1819 
1820 /*
1821  * archs can override this function if they must do something
1822  * after the modifying code is performed.
1823  */
1824 int __weak ftrace_arch_code_modify_post_process(void)
1825 {
1826 	return 0;
1827 }
1828 
1829 static int __ftrace_modify_code(void *data)
1830 {
1831 	int *command = data;
1832 
1833 	if (*command & FTRACE_UPDATE_CALLS)
1834 		ftrace_replace_code(1);
1835 	else if (*command & FTRACE_DISABLE_CALLS)
1836 		ftrace_replace_code(0);
1837 
1838 	if (*command & FTRACE_UPDATE_TRACE_FUNC)
1839 		ftrace_update_ftrace_func(ftrace_trace_function);
1840 
1841 	if (*command & FTRACE_START_FUNC_RET)
1842 		ftrace_enable_ftrace_graph_caller();
1843 	else if (*command & FTRACE_STOP_FUNC_RET)
1844 		ftrace_disable_ftrace_graph_caller();
1845 
1846 	return 0;
1847 }
1848 
1849 /**
1850  * ftrace_run_stop_machine, go back to the stop machine method
1851  * @command: The command to tell ftrace what to do
1852  *
1853  * If an arch needs to fall back to the stop machine method, the
1854  * it can call this function.
1855  */
1856 void ftrace_run_stop_machine(int command)
1857 {
1858 	stop_machine(__ftrace_modify_code, &command, NULL);
1859 }
1860 
1861 /**
1862  * arch_ftrace_update_code, modify the code to trace or not trace
1863  * @command: The command that needs to be done
1864  *
1865  * Archs can override this function if it does not need to
1866  * run stop_machine() to modify code.
1867  */
1868 void __weak arch_ftrace_update_code(int command)
1869 {
1870 	ftrace_run_stop_machine(command);
1871 }
1872 
1873 static void ftrace_run_update_code(int command)
1874 {
1875 	int ret;
1876 
1877 	ret = ftrace_arch_code_modify_prepare();
1878 	FTRACE_WARN_ON(ret);
1879 	if (ret)
1880 		return;
1881 	/*
1882 	 * Do not call function tracer while we update the code.
1883 	 * We are in stop machine.
1884 	 */
1885 	function_trace_stop++;
1886 
1887 	/*
1888 	 * By default we use stop_machine() to modify the code.
1889 	 * But archs can do what ever they want as long as it
1890 	 * is safe. The stop_machine() is the safest, but also
1891 	 * produces the most overhead.
1892 	 */
1893 	arch_ftrace_update_code(command);
1894 
1895 #ifndef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
1896 	/*
1897 	 * For archs that call ftrace_test_stop_func(), we must
1898 	 * wait till after we update all the function callers
1899 	 * before we update the callback. This keeps different
1900 	 * ops that record different functions from corrupting
1901 	 * each other.
1902 	 */
1903 	__ftrace_trace_function = __ftrace_trace_function_delay;
1904 #endif
1905 	function_trace_stop--;
1906 
1907 	ret = ftrace_arch_code_modify_post_process();
1908 	FTRACE_WARN_ON(ret);
1909 }
1910 
1911 static ftrace_func_t saved_ftrace_func;
1912 static int ftrace_start_up;
1913 static int global_start_up;
1914 
1915 static void ftrace_startup_enable(int command)
1916 {
1917 	if (saved_ftrace_func != ftrace_trace_function) {
1918 		saved_ftrace_func = ftrace_trace_function;
1919 		command |= FTRACE_UPDATE_TRACE_FUNC;
1920 	}
1921 
1922 	if (!command || !ftrace_enabled)
1923 		return;
1924 
1925 	ftrace_run_update_code(command);
1926 }
1927 
1928 static int ftrace_startup(struct ftrace_ops *ops, int command)
1929 {
1930 	bool hash_enable = true;
1931 
1932 	if (unlikely(ftrace_disabled))
1933 		return -ENODEV;
1934 
1935 	ftrace_start_up++;
1936 	command |= FTRACE_UPDATE_CALLS;
1937 
1938 	/* ops marked global share the filter hashes */
1939 	if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
1940 		ops = &global_ops;
1941 		/* Don't update hash if global is already set */
1942 		if (global_start_up)
1943 			hash_enable = false;
1944 		global_start_up++;
1945 	}
1946 
1947 	ops->flags |= FTRACE_OPS_FL_ENABLED;
1948 	if (hash_enable)
1949 		ftrace_hash_rec_enable(ops, 1);
1950 
1951 	ftrace_startup_enable(command);
1952 
1953 	return 0;
1954 }
1955 
1956 static void ftrace_shutdown(struct ftrace_ops *ops, int command)
1957 {
1958 	bool hash_disable = true;
1959 
1960 	if (unlikely(ftrace_disabled))
1961 		return;
1962 
1963 	ftrace_start_up--;
1964 	/*
1965 	 * Just warn in case of unbalance, no need to kill ftrace, it's not
1966 	 * critical but the ftrace_call callers may be never nopped again after
1967 	 * further ftrace uses.
1968 	 */
1969 	WARN_ON_ONCE(ftrace_start_up < 0);
1970 
1971 	if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
1972 		ops = &global_ops;
1973 		global_start_up--;
1974 		WARN_ON_ONCE(global_start_up < 0);
1975 		/* Don't update hash if global still has users */
1976 		if (global_start_up) {
1977 			WARN_ON_ONCE(!ftrace_start_up);
1978 			hash_disable = false;
1979 		}
1980 	}
1981 
1982 	if (hash_disable)
1983 		ftrace_hash_rec_disable(ops, 1);
1984 
1985 	if (ops != &global_ops || !global_start_up)
1986 		ops->flags &= ~FTRACE_OPS_FL_ENABLED;
1987 
1988 	command |= FTRACE_UPDATE_CALLS;
1989 
1990 	if (saved_ftrace_func != ftrace_trace_function) {
1991 		saved_ftrace_func = ftrace_trace_function;
1992 		command |= FTRACE_UPDATE_TRACE_FUNC;
1993 	}
1994 
1995 	if (!command || !ftrace_enabled)
1996 		return;
1997 
1998 	ftrace_run_update_code(command);
1999 }
2000 
2001 static void ftrace_startup_sysctl(void)
2002 {
2003 	if (unlikely(ftrace_disabled))
2004 		return;
2005 
2006 	/* Force update next time */
2007 	saved_ftrace_func = NULL;
2008 	/* ftrace_start_up is true if we want ftrace running */
2009 	if (ftrace_start_up)
2010 		ftrace_run_update_code(FTRACE_UPDATE_CALLS);
2011 }
2012 
2013 static void ftrace_shutdown_sysctl(void)
2014 {
2015 	if (unlikely(ftrace_disabled))
2016 		return;
2017 
2018 	/* ftrace_start_up is true if ftrace is running */
2019 	if (ftrace_start_up)
2020 		ftrace_run_update_code(FTRACE_DISABLE_CALLS);
2021 }
2022 
2023 static cycle_t		ftrace_update_time;
2024 static unsigned long	ftrace_update_cnt;
2025 unsigned long		ftrace_update_tot_cnt;
2026 
2027 static int ops_traces_mod(struct ftrace_ops *ops)
2028 {
2029 	struct ftrace_hash *hash;
2030 
2031 	hash = ops->filter_hash;
2032 	return ftrace_hash_empty(hash);
2033 }
2034 
2035 static int ftrace_update_code(struct module *mod)
2036 {
2037 	struct ftrace_page *pg;
2038 	struct dyn_ftrace *p;
2039 	cycle_t start, stop;
2040 	unsigned long ref = 0;
2041 	int i;
2042 
2043 	/*
2044 	 * When adding a module, we need to check if tracers are
2045 	 * currently enabled and if they are set to trace all functions.
2046 	 * If they are, we need to enable the module functions as well
2047 	 * as update the reference counts for those function records.
2048 	 */
2049 	if (mod) {
2050 		struct ftrace_ops *ops;
2051 
2052 		for (ops = ftrace_ops_list;
2053 		     ops != &ftrace_list_end; ops = ops->next) {
2054 			if (ops->flags & FTRACE_OPS_FL_ENABLED &&
2055 			    ops_traces_mod(ops))
2056 				ref++;
2057 		}
2058 	}
2059 
2060 	start = ftrace_now(raw_smp_processor_id());
2061 	ftrace_update_cnt = 0;
2062 
2063 	for (pg = ftrace_new_pgs; pg; pg = pg->next) {
2064 
2065 		for (i = 0; i < pg->index; i++) {
2066 			/* If something went wrong, bail without enabling anything */
2067 			if (unlikely(ftrace_disabled))
2068 				return -1;
2069 
2070 			p = &pg->records[i];
2071 			p->flags = ref;
2072 
2073 			/*
2074 			 * Do the initial record conversion from mcount jump
2075 			 * to the NOP instructions.
2076 			 */
2077 			if (!ftrace_code_disable(mod, p))
2078 				break;
2079 
2080 			ftrace_update_cnt++;
2081 
2082 			/*
2083 			 * If the tracing is enabled, go ahead and enable the record.
2084 			 *
2085 			 * The reason not to enable the record immediatelly is the
2086 			 * inherent check of ftrace_make_nop/ftrace_make_call for
2087 			 * correct previous instructions.  Making first the NOP
2088 			 * conversion puts the module to the correct state, thus
2089 			 * passing the ftrace_make_call check.
2090 			 */
2091 			if (ftrace_start_up && ref) {
2092 				int failed = __ftrace_replace_code(p, 1);
2093 				if (failed)
2094 					ftrace_bug(failed, p->ip);
2095 			}
2096 		}
2097 	}
2098 
2099 	ftrace_new_pgs = NULL;
2100 
2101 	stop = ftrace_now(raw_smp_processor_id());
2102 	ftrace_update_time = stop - start;
2103 	ftrace_update_tot_cnt += ftrace_update_cnt;
2104 
2105 	return 0;
2106 }
2107 
2108 static int ftrace_allocate_records(struct ftrace_page *pg, int count)
2109 {
2110 	int order;
2111 	int cnt;
2112 
2113 	if (WARN_ON(!count))
2114 		return -EINVAL;
2115 
2116 	order = get_count_order(DIV_ROUND_UP(count, ENTRIES_PER_PAGE));
2117 
2118 	/*
2119 	 * We want to fill as much as possible. No more than a page
2120 	 * may be empty.
2121 	 */
2122 	while ((PAGE_SIZE << order) / ENTRY_SIZE >= count + ENTRIES_PER_PAGE)
2123 		order--;
2124 
2125  again:
2126 	pg->records = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
2127 
2128 	if (!pg->records) {
2129 		/* if we can't allocate this size, try something smaller */
2130 		if (!order)
2131 			return -ENOMEM;
2132 		order >>= 1;
2133 		goto again;
2134 	}
2135 
2136 	cnt = (PAGE_SIZE << order) / ENTRY_SIZE;
2137 	pg->size = cnt;
2138 
2139 	if (cnt > count)
2140 		cnt = count;
2141 
2142 	return cnt;
2143 }
2144 
2145 static struct ftrace_page *
2146 ftrace_allocate_pages(unsigned long num_to_init)
2147 {
2148 	struct ftrace_page *start_pg;
2149 	struct ftrace_page *pg;
2150 	int order;
2151 	int cnt;
2152 
2153 	if (!num_to_init)
2154 		return 0;
2155 
2156 	start_pg = pg = kzalloc(sizeof(*pg), GFP_KERNEL);
2157 	if (!pg)
2158 		return NULL;
2159 
2160 	/*
2161 	 * Try to allocate as much as possible in one continues
2162 	 * location that fills in all of the space. We want to
2163 	 * waste as little space as possible.
2164 	 */
2165 	for (;;) {
2166 		cnt = ftrace_allocate_records(pg, num_to_init);
2167 		if (cnt < 0)
2168 			goto free_pages;
2169 
2170 		num_to_init -= cnt;
2171 		if (!num_to_init)
2172 			break;
2173 
2174 		pg->next = kzalloc(sizeof(*pg), GFP_KERNEL);
2175 		if (!pg->next)
2176 			goto free_pages;
2177 
2178 		pg = pg->next;
2179 	}
2180 
2181 	return start_pg;
2182 
2183  free_pages:
2184 	while (start_pg) {
2185 		order = get_count_order(pg->size / ENTRIES_PER_PAGE);
2186 		free_pages((unsigned long)pg->records, order);
2187 		start_pg = pg->next;
2188 		kfree(pg);
2189 		pg = start_pg;
2190 	}
2191 	pr_info("ftrace: FAILED to allocate memory for functions\n");
2192 	return NULL;
2193 }
2194 
2195 static int __init ftrace_dyn_table_alloc(unsigned long num_to_init)
2196 {
2197 	int cnt;
2198 
2199 	if (!num_to_init) {
2200 		pr_info("ftrace: No functions to be traced?\n");
2201 		return -1;
2202 	}
2203 
2204 	cnt = num_to_init / ENTRIES_PER_PAGE;
2205 	pr_info("ftrace: allocating %ld entries in %d pages\n",
2206 		num_to_init, cnt + 1);
2207 
2208 	return 0;
2209 }
2210 
2211 #define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
2212 
2213 struct ftrace_iterator {
2214 	loff_t				pos;
2215 	loff_t				func_pos;
2216 	struct ftrace_page		*pg;
2217 	struct dyn_ftrace		*func;
2218 	struct ftrace_func_probe	*probe;
2219 	struct trace_parser		parser;
2220 	struct ftrace_hash		*hash;
2221 	struct ftrace_ops		*ops;
2222 	int				hidx;
2223 	int				idx;
2224 	unsigned			flags;
2225 };
2226 
2227 static void *
2228 t_hash_next(struct seq_file *m, loff_t *pos)
2229 {
2230 	struct ftrace_iterator *iter = m->private;
2231 	struct hlist_node *hnd = NULL;
2232 	struct hlist_head *hhd;
2233 
2234 	(*pos)++;
2235 	iter->pos = *pos;
2236 
2237 	if (iter->probe)
2238 		hnd = &iter->probe->node;
2239  retry:
2240 	if (iter->hidx >= FTRACE_FUNC_HASHSIZE)
2241 		return NULL;
2242 
2243 	hhd = &ftrace_func_hash[iter->hidx];
2244 
2245 	if (hlist_empty(hhd)) {
2246 		iter->hidx++;
2247 		hnd = NULL;
2248 		goto retry;
2249 	}
2250 
2251 	if (!hnd)
2252 		hnd = hhd->first;
2253 	else {
2254 		hnd = hnd->next;
2255 		if (!hnd) {
2256 			iter->hidx++;
2257 			goto retry;
2258 		}
2259 	}
2260 
2261 	if (WARN_ON_ONCE(!hnd))
2262 		return NULL;
2263 
2264 	iter->probe = hlist_entry(hnd, struct ftrace_func_probe, node);
2265 
2266 	return iter;
2267 }
2268 
2269 static void *t_hash_start(struct seq_file *m, loff_t *pos)
2270 {
2271 	struct ftrace_iterator *iter = m->private;
2272 	void *p = NULL;
2273 	loff_t l;
2274 
2275 	if (!(iter->flags & FTRACE_ITER_DO_HASH))
2276 		return NULL;
2277 
2278 	if (iter->func_pos > *pos)
2279 		return NULL;
2280 
2281 	iter->hidx = 0;
2282 	for (l = 0; l <= (*pos - iter->func_pos); ) {
2283 		p = t_hash_next(m, &l);
2284 		if (!p)
2285 			break;
2286 	}
2287 	if (!p)
2288 		return NULL;
2289 
2290 	/* Only set this if we have an item */
2291 	iter->flags |= FTRACE_ITER_HASH;
2292 
2293 	return iter;
2294 }
2295 
2296 static int
2297 t_hash_show(struct seq_file *m, struct ftrace_iterator *iter)
2298 {
2299 	struct ftrace_func_probe *rec;
2300 
2301 	rec = iter->probe;
2302 	if (WARN_ON_ONCE(!rec))
2303 		return -EIO;
2304 
2305 	if (rec->ops->print)
2306 		return rec->ops->print(m, rec->ip, rec->ops, rec->data);
2307 
2308 	seq_printf(m, "%ps:%ps", (void *)rec->ip, (void *)rec->ops->func);
2309 
2310 	if (rec->data)
2311 		seq_printf(m, ":%p", rec->data);
2312 	seq_putc(m, '\n');
2313 
2314 	return 0;
2315 }
2316 
2317 static void *
2318 t_next(struct seq_file *m, void *v, loff_t *pos)
2319 {
2320 	struct ftrace_iterator *iter = m->private;
2321 	struct ftrace_ops *ops = iter->ops;
2322 	struct dyn_ftrace *rec = NULL;
2323 
2324 	if (unlikely(ftrace_disabled))
2325 		return NULL;
2326 
2327 	if (iter->flags & FTRACE_ITER_HASH)
2328 		return t_hash_next(m, pos);
2329 
2330 	(*pos)++;
2331 	iter->pos = iter->func_pos = *pos;
2332 
2333 	if (iter->flags & FTRACE_ITER_PRINTALL)
2334 		return t_hash_start(m, pos);
2335 
2336  retry:
2337 	if (iter->idx >= iter->pg->index) {
2338 		if (iter->pg->next) {
2339 			iter->pg = iter->pg->next;
2340 			iter->idx = 0;
2341 			goto retry;
2342 		}
2343 	} else {
2344 		rec = &iter->pg->records[iter->idx++];
2345 		if (((iter->flags & FTRACE_ITER_FILTER) &&
2346 		     !(ftrace_lookup_ip(ops->filter_hash, rec->ip))) ||
2347 
2348 		    ((iter->flags & FTRACE_ITER_NOTRACE) &&
2349 		     !ftrace_lookup_ip(ops->notrace_hash, rec->ip)) ||
2350 
2351 		    ((iter->flags & FTRACE_ITER_ENABLED) &&
2352 		     !(rec->flags & ~FTRACE_FL_MASK))) {
2353 
2354 			rec = NULL;
2355 			goto retry;
2356 		}
2357 	}
2358 
2359 	if (!rec)
2360 		return t_hash_start(m, pos);
2361 
2362 	iter->func = rec;
2363 
2364 	return iter;
2365 }
2366 
2367 static void reset_iter_read(struct ftrace_iterator *iter)
2368 {
2369 	iter->pos = 0;
2370 	iter->func_pos = 0;
2371 	iter->flags &= ~(FTRACE_ITER_PRINTALL & FTRACE_ITER_HASH);
2372 }
2373 
2374 static void *t_start(struct seq_file *m, loff_t *pos)
2375 {
2376 	struct ftrace_iterator *iter = m->private;
2377 	struct ftrace_ops *ops = iter->ops;
2378 	void *p = NULL;
2379 	loff_t l;
2380 
2381 	mutex_lock(&ftrace_lock);
2382 
2383 	if (unlikely(ftrace_disabled))
2384 		return NULL;
2385 
2386 	/*
2387 	 * If an lseek was done, then reset and start from beginning.
2388 	 */
2389 	if (*pos < iter->pos)
2390 		reset_iter_read(iter);
2391 
2392 	/*
2393 	 * For set_ftrace_filter reading, if we have the filter
2394 	 * off, we can short cut and just print out that all
2395 	 * functions are enabled.
2396 	 */
2397 	if (iter->flags & FTRACE_ITER_FILTER &&
2398 	    ftrace_hash_empty(ops->filter_hash)) {
2399 		if (*pos > 0)
2400 			return t_hash_start(m, pos);
2401 		iter->flags |= FTRACE_ITER_PRINTALL;
2402 		/* reset in case of seek/pread */
2403 		iter->flags &= ~FTRACE_ITER_HASH;
2404 		return iter;
2405 	}
2406 
2407 	if (iter->flags & FTRACE_ITER_HASH)
2408 		return t_hash_start(m, pos);
2409 
2410 	/*
2411 	 * Unfortunately, we need to restart at ftrace_pages_start
2412 	 * every time we let go of the ftrace_mutex. This is because
2413 	 * those pointers can change without the lock.
2414 	 */
2415 	iter->pg = ftrace_pages_start;
2416 	iter->idx = 0;
2417 	for (l = 0; l <= *pos; ) {
2418 		p = t_next(m, p, &l);
2419 		if (!p)
2420 			break;
2421 	}
2422 
2423 	if (!p)
2424 		return t_hash_start(m, pos);
2425 
2426 	return iter;
2427 }
2428 
2429 static void t_stop(struct seq_file *m, void *p)
2430 {
2431 	mutex_unlock(&ftrace_lock);
2432 }
2433 
2434 static int t_show(struct seq_file *m, void *v)
2435 {
2436 	struct ftrace_iterator *iter = m->private;
2437 	struct dyn_ftrace *rec;
2438 
2439 	if (iter->flags & FTRACE_ITER_HASH)
2440 		return t_hash_show(m, iter);
2441 
2442 	if (iter->flags & FTRACE_ITER_PRINTALL) {
2443 		seq_printf(m, "#### all functions enabled ####\n");
2444 		return 0;
2445 	}
2446 
2447 	rec = iter->func;
2448 
2449 	if (!rec)
2450 		return 0;
2451 
2452 	seq_printf(m, "%ps", (void *)rec->ip);
2453 	if (iter->flags & FTRACE_ITER_ENABLED)
2454 		seq_printf(m, " (%ld)",
2455 			   rec->flags & ~FTRACE_FL_MASK);
2456 	seq_printf(m, "\n");
2457 
2458 	return 0;
2459 }
2460 
2461 static const struct seq_operations show_ftrace_seq_ops = {
2462 	.start = t_start,
2463 	.next = t_next,
2464 	.stop = t_stop,
2465 	.show = t_show,
2466 };
2467 
2468 static int
2469 ftrace_avail_open(struct inode *inode, struct file *file)
2470 {
2471 	struct ftrace_iterator *iter;
2472 	int ret;
2473 
2474 	if (unlikely(ftrace_disabled))
2475 		return -ENODEV;
2476 
2477 	iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2478 	if (!iter)
2479 		return -ENOMEM;
2480 
2481 	iter->pg = ftrace_pages_start;
2482 	iter->ops = &global_ops;
2483 
2484 	ret = seq_open(file, &show_ftrace_seq_ops);
2485 	if (!ret) {
2486 		struct seq_file *m = file->private_data;
2487 
2488 		m->private = iter;
2489 	} else {
2490 		kfree(iter);
2491 	}
2492 
2493 	return ret;
2494 }
2495 
2496 static int
2497 ftrace_enabled_open(struct inode *inode, struct file *file)
2498 {
2499 	struct ftrace_iterator *iter;
2500 	int ret;
2501 
2502 	if (unlikely(ftrace_disabled))
2503 		return -ENODEV;
2504 
2505 	iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2506 	if (!iter)
2507 		return -ENOMEM;
2508 
2509 	iter->pg = ftrace_pages_start;
2510 	iter->flags = FTRACE_ITER_ENABLED;
2511 	iter->ops = &global_ops;
2512 
2513 	ret = seq_open(file, &show_ftrace_seq_ops);
2514 	if (!ret) {
2515 		struct seq_file *m = file->private_data;
2516 
2517 		m->private = iter;
2518 	} else {
2519 		kfree(iter);
2520 	}
2521 
2522 	return ret;
2523 }
2524 
2525 static void ftrace_filter_reset(struct ftrace_hash *hash)
2526 {
2527 	mutex_lock(&ftrace_lock);
2528 	ftrace_hash_clear(hash);
2529 	mutex_unlock(&ftrace_lock);
2530 }
2531 
2532 /**
2533  * ftrace_regex_open - initialize function tracer filter files
2534  * @ops: The ftrace_ops that hold the hash filters
2535  * @flag: The type of filter to process
2536  * @inode: The inode, usually passed in to your open routine
2537  * @file: The file, usually passed in to your open routine
2538  *
2539  * ftrace_regex_open() initializes the filter files for the
2540  * @ops. Depending on @flag it may process the filter hash or
2541  * the notrace hash of @ops. With this called from the open
2542  * routine, you can use ftrace_filter_write() for the write
2543  * routine if @flag has FTRACE_ITER_FILTER set, or
2544  * ftrace_notrace_write() if @flag has FTRACE_ITER_NOTRACE set.
2545  * ftrace_regex_lseek() should be used as the lseek routine, and
2546  * release must call ftrace_regex_release().
2547  */
2548 int
2549 ftrace_regex_open(struct ftrace_ops *ops, int flag,
2550 		  struct inode *inode, struct file *file)
2551 {
2552 	struct ftrace_iterator *iter;
2553 	struct ftrace_hash *hash;
2554 	int ret = 0;
2555 
2556 	if (unlikely(ftrace_disabled))
2557 		return -ENODEV;
2558 
2559 	iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2560 	if (!iter)
2561 		return -ENOMEM;
2562 
2563 	if (trace_parser_get_init(&iter->parser, FTRACE_BUFF_MAX)) {
2564 		kfree(iter);
2565 		return -ENOMEM;
2566 	}
2567 
2568 	if (flag & FTRACE_ITER_NOTRACE)
2569 		hash = ops->notrace_hash;
2570 	else
2571 		hash = ops->filter_hash;
2572 
2573 	iter->ops = ops;
2574 	iter->flags = flag;
2575 
2576 	if (file->f_mode & FMODE_WRITE) {
2577 		mutex_lock(&ftrace_lock);
2578 		iter->hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, hash);
2579 		mutex_unlock(&ftrace_lock);
2580 
2581 		if (!iter->hash) {
2582 			trace_parser_put(&iter->parser);
2583 			kfree(iter);
2584 			return -ENOMEM;
2585 		}
2586 	}
2587 
2588 	mutex_lock(&ftrace_regex_lock);
2589 
2590 	if ((file->f_mode & FMODE_WRITE) &&
2591 	    (file->f_flags & O_TRUNC))
2592 		ftrace_filter_reset(iter->hash);
2593 
2594 	if (file->f_mode & FMODE_READ) {
2595 		iter->pg = ftrace_pages_start;
2596 
2597 		ret = seq_open(file, &show_ftrace_seq_ops);
2598 		if (!ret) {
2599 			struct seq_file *m = file->private_data;
2600 			m->private = iter;
2601 		} else {
2602 			/* Failed */
2603 			free_ftrace_hash(iter->hash);
2604 			trace_parser_put(&iter->parser);
2605 			kfree(iter);
2606 		}
2607 	} else
2608 		file->private_data = iter;
2609 	mutex_unlock(&ftrace_regex_lock);
2610 
2611 	return ret;
2612 }
2613 
2614 static int
2615 ftrace_filter_open(struct inode *inode, struct file *file)
2616 {
2617 	return ftrace_regex_open(&global_ops,
2618 			FTRACE_ITER_FILTER | FTRACE_ITER_DO_HASH,
2619 			inode, file);
2620 }
2621 
2622 static int
2623 ftrace_notrace_open(struct inode *inode, struct file *file)
2624 {
2625 	return ftrace_regex_open(&global_ops, FTRACE_ITER_NOTRACE,
2626 				 inode, file);
2627 }
2628 
2629 loff_t
2630 ftrace_regex_lseek(struct file *file, loff_t offset, int origin)
2631 {
2632 	loff_t ret;
2633 
2634 	if (file->f_mode & FMODE_READ)
2635 		ret = seq_lseek(file, offset, origin);
2636 	else
2637 		file->f_pos = ret = 1;
2638 
2639 	return ret;
2640 }
2641 
2642 static int ftrace_match(char *str, char *regex, int len, int type)
2643 {
2644 	int matched = 0;
2645 	int slen;
2646 
2647 	switch (type) {
2648 	case MATCH_FULL:
2649 		if (strcmp(str, regex) == 0)
2650 			matched = 1;
2651 		break;
2652 	case MATCH_FRONT_ONLY:
2653 		if (strncmp(str, regex, len) == 0)
2654 			matched = 1;
2655 		break;
2656 	case MATCH_MIDDLE_ONLY:
2657 		if (strstr(str, regex))
2658 			matched = 1;
2659 		break;
2660 	case MATCH_END_ONLY:
2661 		slen = strlen(str);
2662 		if (slen >= len && memcmp(str + slen - len, regex, len) == 0)
2663 			matched = 1;
2664 		break;
2665 	}
2666 
2667 	return matched;
2668 }
2669 
2670 static int
2671 enter_record(struct ftrace_hash *hash, struct dyn_ftrace *rec, int not)
2672 {
2673 	struct ftrace_func_entry *entry;
2674 	int ret = 0;
2675 
2676 	entry = ftrace_lookup_ip(hash, rec->ip);
2677 	if (not) {
2678 		/* Do nothing if it doesn't exist */
2679 		if (!entry)
2680 			return 0;
2681 
2682 		free_hash_entry(hash, entry);
2683 	} else {
2684 		/* Do nothing if it exists */
2685 		if (entry)
2686 			return 0;
2687 
2688 		ret = add_hash_entry(hash, rec->ip);
2689 	}
2690 	return ret;
2691 }
2692 
2693 static int
2694 ftrace_match_record(struct dyn_ftrace *rec, char *mod,
2695 		    char *regex, int len, int type)
2696 {
2697 	char str[KSYM_SYMBOL_LEN];
2698 	char *modname;
2699 
2700 	kallsyms_lookup(rec->ip, NULL, NULL, &modname, str);
2701 
2702 	if (mod) {
2703 		/* module lookup requires matching the module */
2704 		if (!modname || strcmp(modname, mod))
2705 			return 0;
2706 
2707 		/* blank search means to match all funcs in the mod */
2708 		if (!len)
2709 			return 1;
2710 	}
2711 
2712 	return ftrace_match(str, regex, len, type);
2713 }
2714 
2715 static int
2716 match_records(struct ftrace_hash *hash, char *buff,
2717 	      int len, char *mod, int not)
2718 {
2719 	unsigned search_len = 0;
2720 	struct ftrace_page *pg;
2721 	struct dyn_ftrace *rec;
2722 	int type = MATCH_FULL;
2723 	char *search = buff;
2724 	int found = 0;
2725 	int ret;
2726 
2727 	if (len) {
2728 		type = filter_parse_regex(buff, len, &search, &not);
2729 		search_len = strlen(search);
2730 	}
2731 
2732 	mutex_lock(&ftrace_lock);
2733 
2734 	if (unlikely(ftrace_disabled))
2735 		goto out_unlock;
2736 
2737 	do_for_each_ftrace_rec(pg, rec) {
2738 		if (ftrace_match_record(rec, mod, search, search_len, type)) {
2739 			ret = enter_record(hash, rec, not);
2740 			if (ret < 0) {
2741 				found = ret;
2742 				goto out_unlock;
2743 			}
2744 			found = 1;
2745 		}
2746 	} while_for_each_ftrace_rec();
2747  out_unlock:
2748 	mutex_unlock(&ftrace_lock);
2749 
2750 	return found;
2751 }
2752 
2753 static int
2754 ftrace_match_records(struct ftrace_hash *hash, char *buff, int len)
2755 {
2756 	return match_records(hash, buff, len, NULL, 0);
2757 }
2758 
2759 static int
2760 ftrace_match_module_records(struct ftrace_hash *hash, char *buff, char *mod)
2761 {
2762 	int not = 0;
2763 
2764 	/* blank or '*' mean the same */
2765 	if (strcmp(buff, "*") == 0)
2766 		buff[0] = 0;
2767 
2768 	/* handle the case of 'dont filter this module' */
2769 	if (strcmp(buff, "!") == 0 || strcmp(buff, "!*") == 0) {
2770 		buff[0] = 0;
2771 		not = 1;
2772 	}
2773 
2774 	return match_records(hash, buff, strlen(buff), mod, not);
2775 }
2776 
2777 /*
2778  * We register the module command as a template to show others how
2779  * to register the a command as well.
2780  */
2781 
2782 static int
2783 ftrace_mod_callback(struct ftrace_hash *hash,
2784 		    char *func, char *cmd, char *param, int enable)
2785 {
2786 	char *mod;
2787 	int ret = -EINVAL;
2788 
2789 	/*
2790 	 * cmd == 'mod' because we only registered this func
2791 	 * for the 'mod' ftrace_func_command.
2792 	 * But if you register one func with multiple commands,
2793 	 * you can tell which command was used by the cmd
2794 	 * parameter.
2795 	 */
2796 
2797 	/* we must have a module name */
2798 	if (!param)
2799 		return ret;
2800 
2801 	mod = strsep(&param, ":");
2802 	if (!strlen(mod))
2803 		return ret;
2804 
2805 	ret = ftrace_match_module_records(hash, func, mod);
2806 	if (!ret)
2807 		ret = -EINVAL;
2808 	if (ret < 0)
2809 		return ret;
2810 
2811 	return 0;
2812 }
2813 
2814 static struct ftrace_func_command ftrace_mod_cmd = {
2815 	.name			= "mod",
2816 	.func			= ftrace_mod_callback,
2817 };
2818 
2819 static int __init ftrace_mod_cmd_init(void)
2820 {
2821 	return register_ftrace_command(&ftrace_mod_cmd);
2822 }
2823 device_initcall(ftrace_mod_cmd_init);
2824 
2825 static void
2826 function_trace_probe_call(unsigned long ip, unsigned long parent_ip)
2827 {
2828 	struct ftrace_func_probe *entry;
2829 	struct hlist_head *hhd;
2830 	struct hlist_node *n;
2831 	unsigned long key;
2832 
2833 	key = hash_long(ip, FTRACE_HASH_BITS);
2834 
2835 	hhd = &ftrace_func_hash[key];
2836 
2837 	if (hlist_empty(hhd))
2838 		return;
2839 
2840 	/*
2841 	 * Disable preemption for these calls to prevent a RCU grace
2842 	 * period. This syncs the hash iteration and freeing of items
2843 	 * on the hash. rcu_read_lock is too dangerous here.
2844 	 */
2845 	preempt_disable_notrace();
2846 	hlist_for_each_entry_rcu(entry, n, hhd, node) {
2847 		if (entry->ip == ip)
2848 			entry->ops->func(ip, parent_ip, &entry->data);
2849 	}
2850 	preempt_enable_notrace();
2851 }
2852 
2853 static struct ftrace_ops trace_probe_ops __read_mostly =
2854 {
2855 	.func		= function_trace_probe_call,
2856 };
2857 
2858 static int ftrace_probe_registered;
2859 
2860 static void __enable_ftrace_function_probe(void)
2861 {
2862 	int ret;
2863 	int i;
2864 
2865 	if (ftrace_probe_registered)
2866 		return;
2867 
2868 	for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
2869 		struct hlist_head *hhd = &ftrace_func_hash[i];
2870 		if (hhd->first)
2871 			break;
2872 	}
2873 	/* Nothing registered? */
2874 	if (i == FTRACE_FUNC_HASHSIZE)
2875 		return;
2876 
2877 	ret = __register_ftrace_function(&trace_probe_ops);
2878 	if (!ret)
2879 		ret = ftrace_startup(&trace_probe_ops, 0);
2880 
2881 	ftrace_probe_registered = 1;
2882 }
2883 
2884 static void __disable_ftrace_function_probe(void)
2885 {
2886 	int ret;
2887 	int i;
2888 
2889 	if (!ftrace_probe_registered)
2890 		return;
2891 
2892 	for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
2893 		struct hlist_head *hhd = &ftrace_func_hash[i];
2894 		if (hhd->first)
2895 			return;
2896 	}
2897 
2898 	/* no more funcs left */
2899 	ret = __unregister_ftrace_function(&trace_probe_ops);
2900 	if (!ret)
2901 		ftrace_shutdown(&trace_probe_ops, 0);
2902 
2903 	ftrace_probe_registered = 0;
2904 }
2905 
2906 
2907 static void ftrace_free_entry_rcu(struct rcu_head *rhp)
2908 {
2909 	struct ftrace_func_probe *entry =
2910 		container_of(rhp, struct ftrace_func_probe, rcu);
2911 
2912 	if (entry->ops->free)
2913 		entry->ops->free(&entry->data);
2914 	kfree(entry);
2915 }
2916 
2917 
2918 int
2919 register_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
2920 			      void *data)
2921 {
2922 	struct ftrace_func_probe *entry;
2923 	struct ftrace_page *pg;
2924 	struct dyn_ftrace *rec;
2925 	int type, len, not;
2926 	unsigned long key;
2927 	int count = 0;
2928 	char *search;
2929 
2930 	type = filter_parse_regex(glob, strlen(glob), &search, &not);
2931 	len = strlen(search);
2932 
2933 	/* we do not support '!' for function probes */
2934 	if (WARN_ON(not))
2935 		return -EINVAL;
2936 
2937 	mutex_lock(&ftrace_lock);
2938 
2939 	if (unlikely(ftrace_disabled))
2940 		goto out_unlock;
2941 
2942 	do_for_each_ftrace_rec(pg, rec) {
2943 
2944 		if (!ftrace_match_record(rec, NULL, search, len, type))
2945 			continue;
2946 
2947 		entry = kmalloc(sizeof(*entry), GFP_KERNEL);
2948 		if (!entry) {
2949 			/* If we did not process any, then return error */
2950 			if (!count)
2951 				count = -ENOMEM;
2952 			goto out_unlock;
2953 		}
2954 
2955 		count++;
2956 
2957 		entry->data = data;
2958 
2959 		/*
2960 		 * The caller might want to do something special
2961 		 * for each function we find. We call the callback
2962 		 * to give the caller an opportunity to do so.
2963 		 */
2964 		if (ops->callback) {
2965 			if (ops->callback(rec->ip, &entry->data) < 0) {
2966 				/* caller does not like this func */
2967 				kfree(entry);
2968 				continue;
2969 			}
2970 		}
2971 
2972 		entry->ops = ops;
2973 		entry->ip = rec->ip;
2974 
2975 		key = hash_long(entry->ip, FTRACE_HASH_BITS);
2976 		hlist_add_head_rcu(&entry->node, &ftrace_func_hash[key]);
2977 
2978 	} while_for_each_ftrace_rec();
2979 	__enable_ftrace_function_probe();
2980 
2981  out_unlock:
2982 	mutex_unlock(&ftrace_lock);
2983 
2984 	return count;
2985 }
2986 
2987 enum {
2988 	PROBE_TEST_FUNC		= 1,
2989 	PROBE_TEST_DATA		= 2
2990 };
2991 
2992 static void
2993 __unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
2994 				  void *data, int flags)
2995 {
2996 	struct ftrace_func_probe *entry;
2997 	struct hlist_node *n, *tmp;
2998 	char str[KSYM_SYMBOL_LEN];
2999 	int type = MATCH_FULL;
3000 	int i, len = 0;
3001 	char *search;
3002 
3003 	if (glob && (strcmp(glob, "*") == 0 || !strlen(glob)))
3004 		glob = NULL;
3005 	else if (glob) {
3006 		int not;
3007 
3008 		type = filter_parse_regex(glob, strlen(glob), &search, &not);
3009 		len = strlen(search);
3010 
3011 		/* we do not support '!' for function probes */
3012 		if (WARN_ON(not))
3013 			return;
3014 	}
3015 
3016 	mutex_lock(&ftrace_lock);
3017 	for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
3018 		struct hlist_head *hhd = &ftrace_func_hash[i];
3019 
3020 		hlist_for_each_entry_safe(entry, n, tmp, hhd, node) {
3021 
3022 			/* break up if statements for readability */
3023 			if ((flags & PROBE_TEST_FUNC) && entry->ops != ops)
3024 				continue;
3025 
3026 			if ((flags & PROBE_TEST_DATA) && entry->data != data)
3027 				continue;
3028 
3029 			/* do this last, since it is the most expensive */
3030 			if (glob) {
3031 				kallsyms_lookup(entry->ip, NULL, NULL,
3032 						NULL, str);
3033 				if (!ftrace_match(str, glob, len, type))
3034 					continue;
3035 			}
3036 
3037 			hlist_del(&entry->node);
3038 			call_rcu(&entry->rcu, ftrace_free_entry_rcu);
3039 		}
3040 	}
3041 	__disable_ftrace_function_probe();
3042 	mutex_unlock(&ftrace_lock);
3043 }
3044 
3045 void
3046 unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
3047 				void *data)
3048 {
3049 	__unregister_ftrace_function_probe(glob, ops, data,
3050 					  PROBE_TEST_FUNC | PROBE_TEST_DATA);
3051 }
3052 
3053 void
3054 unregister_ftrace_function_probe_func(char *glob, struct ftrace_probe_ops *ops)
3055 {
3056 	__unregister_ftrace_function_probe(glob, ops, NULL, PROBE_TEST_FUNC);
3057 }
3058 
3059 void unregister_ftrace_function_probe_all(char *glob)
3060 {
3061 	__unregister_ftrace_function_probe(glob, NULL, NULL, 0);
3062 }
3063 
3064 static LIST_HEAD(ftrace_commands);
3065 static DEFINE_MUTEX(ftrace_cmd_mutex);
3066 
3067 int register_ftrace_command(struct ftrace_func_command *cmd)
3068 {
3069 	struct ftrace_func_command *p;
3070 	int ret = 0;
3071 
3072 	mutex_lock(&ftrace_cmd_mutex);
3073 	list_for_each_entry(p, &ftrace_commands, list) {
3074 		if (strcmp(cmd->name, p->name) == 0) {
3075 			ret = -EBUSY;
3076 			goto out_unlock;
3077 		}
3078 	}
3079 	list_add(&cmd->list, &ftrace_commands);
3080  out_unlock:
3081 	mutex_unlock(&ftrace_cmd_mutex);
3082 
3083 	return ret;
3084 }
3085 
3086 int unregister_ftrace_command(struct ftrace_func_command *cmd)
3087 {
3088 	struct ftrace_func_command *p, *n;
3089 	int ret = -ENODEV;
3090 
3091 	mutex_lock(&ftrace_cmd_mutex);
3092 	list_for_each_entry_safe(p, n, &ftrace_commands, list) {
3093 		if (strcmp(cmd->name, p->name) == 0) {
3094 			ret = 0;
3095 			list_del_init(&p->list);
3096 			goto out_unlock;
3097 		}
3098 	}
3099  out_unlock:
3100 	mutex_unlock(&ftrace_cmd_mutex);
3101 
3102 	return ret;
3103 }
3104 
3105 static int ftrace_process_regex(struct ftrace_hash *hash,
3106 				char *buff, int len, int enable)
3107 {
3108 	char *func, *command, *next = buff;
3109 	struct ftrace_func_command *p;
3110 	int ret = -EINVAL;
3111 
3112 	func = strsep(&next, ":");
3113 
3114 	if (!next) {
3115 		ret = ftrace_match_records(hash, func, len);
3116 		if (!ret)
3117 			ret = -EINVAL;
3118 		if (ret < 0)
3119 			return ret;
3120 		return 0;
3121 	}
3122 
3123 	/* command found */
3124 
3125 	command = strsep(&next, ":");
3126 
3127 	mutex_lock(&ftrace_cmd_mutex);
3128 	list_for_each_entry(p, &ftrace_commands, list) {
3129 		if (strcmp(p->name, command) == 0) {
3130 			ret = p->func(hash, func, command, next, enable);
3131 			goto out_unlock;
3132 		}
3133 	}
3134  out_unlock:
3135 	mutex_unlock(&ftrace_cmd_mutex);
3136 
3137 	return ret;
3138 }
3139 
3140 static ssize_t
3141 ftrace_regex_write(struct file *file, const char __user *ubuf,
3142 		   size_t cnt, loff_t *ppos, int enable)
3143 {
3144 	struct ftrace_iterator *iter;
3145 	struct trace_parser *parser;
3146 	ssize_t ret, read;
3147 
3148 	if (!cnt)
3149 		return 0;
3150 
3151 	mutex_lock(&ftrace_regex_lock);
3152 
3153 	ret = -ENODEV;
3154 	if (unlikely(ftrace_disabled))
3155 		goto out_unlock;
3156 
3157 	if (file->f_mode & FMODE_READ) {
3158 		struct seq_file *m = file->private_data;
3159 		iter = m->private;
3160 	} else
3161 		iter = file->private_data;
3162 
3163 	parser = &iter->parser;
3164 	read = trace_get_user(parser, ubuf, cnt, ppos);
3165 
3166 	if (read >= 0 && trace_parser_loaded(parser) &&
3167 	    !trace_parser_cont(parser)) {
3168 		ret = ftrace_process_regex(iter->hash, parser->buffer,
3169 					   parser->idx, enable);
3170 		trace_parser_clear(parser);
3171 		if (ret)
3172 			goto out_unlock;
3173 	}
3174 
3175 	ret = read;
3176 out_unlock:
3177 	mutex_unlock(&ftrace_regex_lock);
3178 
3179 	return ret;
3180 }
3181 
3182 ssize_t
3183 ftrace_filter_write(struct file *file, const char __user *ubuf,
3184 		    size_t cnt, loff_t *ppos)
3185 {
3186 	return ftrace_regex_write(file, ubuf, cnt, ppos, 1);
3187 }
3188 
3189 ssize_t
3190 ftrace_notrace_write(struct file *file, const char __user *ubuf,
3191 		     size_t cnt, loff_t *ppos)
3192 {
3193 	return ftrace_regex_write(file, ubuf, cnt, ppos, 0);
3194 }
3195 
3196 static int
3197 ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len,
3198 		 int reset, int enable)
3199 {
3200 	struct ftrace_hash **orig_hash;
3201 	struct ftrace_hash *hash;
3202 	int ret;
3203 
3204 	/* All global ops uses the global ops filters */
3205 	if (ops->flags & FTRACE_OPS_FL_GLOBAL)
3206 		ops = &global_ops;
3207 
3208 	if (unlikely(ftrace_disabled))
3209 		return -ENODEV;
3210 
3211 	if (enable)
3212 		orig_hash = &ops->filter_hash;
3213 	else
3214 		orig_hash = &ops->notrace_hash;
3215 
3216 	hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
3217 	if (!hash)
3218 		return -ENOMEM;
3219 
3220 	mutex_lock(&ftrace_regex_lock);
3221 	if (reset)
3222 		ftrace_filter_reset(hash);
3223 	if (buf && !ftrace_match_records(hash, buf, len)) {
3224 		ret = -EINVAL;
3225 		goto out_regex_unlock;
3226 	}
3227 
3228 	mutex_lock(&ftrace_lock);
3229 	ret = ftrace_hash_move(ops, enable, orig_hash, hash);
3230 	if (!ret && ops->flags & FTRACE_OPS_FL_ENABLED
3231 	    && ftrace_enabled)
3232 		ftrace_run_update_code(FTRACE_UPDATE_CALLS);
3233 
3234 	mutex_unlock(&ftrace_lock);
3235 
3236  out_regex_unlock:
3237 	mutex_unlock(&ftrace_regex_lock);
3238 
3239 	free_ftrace_hash(hash);
3240 	return ret;
3241 }
3242 
3243 /**
3244  * ftrace_set_filter - set a function to filter on in ftrace
3245  * @ops - the ops to set the filter with
3246  * @buf - the string that holds the function filter text.
3247  * @len - the length of the string.
3248  * @reset - non zero to reset all filters before applying this filter.
3249  *
3250  * Filters denote which functions should be enabled when tracing is enabled.
3251  * If @buf is NULL and reset is set, all functions will be enabled for tracing.
3252  */
3253 int ftrace_set_filter(struct ftrace_ops *ops, unsigned char *buf,
3254 		       int len, int reset)
3255 {
3256 	return ftrace_set_regex(ops, buf, len, reset, 1);
3257 }
3258 EXPORT_SYMBOL_GPL(ftrace_set_filter);
3259 
3260 /**
3261  * ftrace_set_notrace - set a function to not trace in ftrace
3262  * @ops - the ops to set the notrace filter with
3263  * @buf - the string that holds the function notrace text.
3264  * @len - the length of the string.
3265  * @reset - non zero to reset all filters before applying this filter.
3266  *
3267  * Notrace Filters denote which functions should not be enabled when tracing
3268  * is enabled. If @buf is NULL and reset is set, all functions will be enabled
3269  * for tracing.
3270  */
3271 int ftrace_set_notrace(struct ftrace_ops *ops, unsigned char *buf,
3272 			int len, int reset)
3273 {
3274 	return ftrace_set_regex(ops, buf, len, reset, 0);
3275 }
3276 EXPORT_SYMBOL_GPL(ftrace_set_notrace);
3277 /**
3278  * ftrace_set_filter - set a function to filter on in ftrace
3279  * @ops - the ops to set the filter with
3280  * @buf - the string that holds the function filter text.
3281  * @len - the length of the string.
3282  * @reset - non zero to reset all filters before applying this filter.
3283  *
3284  * Filters denote which functions should be enabled when tracing is enabled.
3285  * If @buf is NULL and reset is set, all functions will be enabled for tracing.
3286  */
3287 void ftrace_set_global_filter(unsigned char *buf, int len, int reset)
3288 {
3289 	ftrace_set_regex(&global_ops, buf, len, reset, 1);
3290 }
3291 EXPORT_SYMBOL_GPL(ftrace_set_global_filter);
3292 
3293 /**
3294  * ftrace_set_notrace - set a function to not trace in ftrace
3295  * @ops - the ops to set the notrace filter with
3296  * @buf - the string that holds the function notrace text.
3297  * @len - the length of the string.
3298  * @reset - non zero to reset all filters before applying this filter.
3299  *
3300  * Notrace Filters denote which functions should not be enabled when tracing
3301  * is enabled. If @buf is NULL and reset is set, all functions will be enabled
3302  * for tracing.
3303  */
3304 void ftrace_set_global_notrace(unsigned char *buf, int len, int reset)
3305 {
3306 	ftrace_set_regex(&global_ops, buf, len, reset, 0);
3307 }
3308 EXPORT_SYMBOL_GPL(ftrace_set_global_notrace);
3309 
3310 /*
3311  * command line interface to allow users to set filters on boot up.
3312  */
3313 #define FTRACE_FILTER_SIZE		COMMAND_LINE_SIZE
3314 static char ftrace_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
3315 static char ftrace_filter_buf[FTRACE_FILTER_SIZE] __initdata;
3316 
3317 static int __init set_ftrace_notrace(char *str)
3318 {
3319 	strncpy(ftrace_notrace_buf, str, FTRACE_FILTER_SIZE);
3320 	return 1;
3321 }
3322 __setup("ftrace_notrace=", set_ftrace_notrace);
3323 
3324 static int __init set_ftrace_filter(char *str)
3325 {
3326 	strncpy(ftrace_filter_buf, str, FTRACE_FILTER_SIZE);
3327 	return 1;
3328 }
3329 __setup("ftrace_filter=", set_ftrace_filter);
3330 
3331 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3332 static char ftrace_graph_buf[FTRACE_FILTER_SIZE] __initdata;
3333 static int ftrace_set_func(unsigned long *array, int *idx, char *buffer);
3334 
3335 static int __init set_graph_function(char *str)
3336 {
3337 	strlcpy(ftrace_graph_buf, str, FTRACE_FILTER_SIZE);
3338 	return 1;
3339 }
3340 __setup("ftrace_graph_filter=", set_graph_function);
3341 
3342 static void __init set_ftrace_early_graph(char *buf)
3343 {
3344 	int ret;
3345 	char *func;
3346 
3347 	while (buf) {
3348 		func = strsep(&buf, ",");
3349 		/* we allow only one expression at a time */
3350 		ret = ftrace_set_func(ftrace_graph_funcs, &ftrace_graph_count,
3351 				      func);
3352 		if (ret)
3353 			printk(KERN_DEBUG "ftrace: function %s not "
3354 					  "traceable\n", func);
3355 	}
3356 }
3357 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
3358 
3359 void __init
3360 ftrace_set_early_filter(struct ftrace_ops *ops, char *buf, int enable)
3361 {
3362 	char *func;
3363 
3364 	while (buf) {
3365 		func = strsep(&buf, ",");
3366 		ftrace_set_regex(ops, func, strlen(func), 0, enable);
3367 	}
3368 }
3369 
3370 static void __init set_ftrace_early_filters(void)
3371 {
3372 	if (ftrace_filter_buf[0])
3373 		ftrace_set_early_filter(&global_ops, ftrace_filter_buf, 1);
3374 	if (ftrace_notrace_buf[0])
3375 		ftrace_set_early_filter(&global_ops, ftrace_notrace_buf, 0);
3376 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3377 	if (ftrace_graph_buf[0])
3378 		set_ftrace_early_graph(ftrace_graph_buf);
3379 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
3380 }
3381 
3382 int ftrace_regex_release(struct inode *inode, struct file *file)
3383 {
3384 	struct seq_file *m = (struct seq_file *)file->private_data;
3385 	struct ftrace_iterator *iter;
3386 	struct ftrace_hash **orig_hash;
3387 	struct trace_parser *parser;
3388 	int filter_hash;
3389 	int ret;
3390 
3391 	mutex_lock(&ftrace_regex_lock);
3392 	if (file->f_mode & FMODE_READ) {
3393 		iter = m->private;
3394 
3395 		seq_release(inode, file);
3396 	} else
3397 		iter = file->private_data;
3398 
3399 	parser = &iter->parser;
3400 	if (trace_parser_loaded(parser)) {
3401 		parser->buffer[parser->idx] = 0;
3402 		ftrace_match_records(iter->hash, parser->buffer, parser->idx);
3403 	}
3404 
3405 	trace_parser_put(parser);
3406 
3407 	if (file->f_mode & FMODE_WRITE) {
3408 		filter_hash = !!(iter->flags & FTRACE_ITER_FILTER);
3409 
3410 		if (filter_hash)
3411 			orig_hash = &iter->ops->filter_hash;
3412 		else
3413 			orig_hash = &iter->ops->notrace_hash;
3414 
3415 		mutex_lock(&ftrace_lock);
3416 		ret = ftrace_hash_move(iter->ops, filter_hash,
3417 				       orig_hash, iter->hash);
3418 		if (!ret && (iter->ops->flags & FTRACE_OPS_FL_ENABLED)
3419 		    && ftrace_enabled)
3420 			ftrace_run_update_code(FTRACE_UPDATE_CALLS);
3421 
3422 		mutex_unlock(&ftrace_lock);
3423 	}
3424 	free_ftrace_hash(iter->hash);
3425 	kfree(iter);
3426 
3427 	mutex_unlock(&ftrace_regex_lock);
3428 	return 0;
3429 }
3430 
3431 static const struct file_operations ftrace_avail_fops = {
3432 	.open = ftrace_avail_open,
3433 	.read = seq_read,
3434 	.llseek = seq_lseek,
3435 	.release = seq_release_private,
3436 };
3437 
3438 static const struct file_operations ftrace_enabled_fops = {
3439 	.open = ftrace_enabled_open,
3440 	.read = seq_read,
3441 	.llseek = seq_lseek,
3442 	.release = seq_release_private,
3443 };
3444 
3445 static const struct file_operations ftrace_filter_fops = {
3446 	.open = ftrace_filter_open,
3447 	.read = seq_read,
3448 	.write = ftrace_filter_write,
3449 	.llseek = ftrace_regex_lseek,
3450 	.release = ftrace_regex_release,
3451 };
3452 
3453 static const struct file_operations ftrace_notrace_fops = {
3454 	.open = ftrace_notrace_open,
3455 	.read = seq_read,
3456 	.write = ftrace_notrace_write,
3457 	.llseek = ftrace_regex_lseek,
3458 	.release = ftrace_regex_release,
3459 };
3460 
3461 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3462 
3463 static DEFINE_MUTEX(graph_lock);
3464 
3465 int ftrace_graph_count;
3466 int ftrace_graph_filter_enabled;
3467 unsigned long ftrace_graph_funcs[FTRACE_GRAPH_MAX_FUNCS] __read_mostly;
3468 
3469 static void *
3470 __g_next(struct seq_file *m, loff_t *pos)
3471 {
3472 	if (*pos >= ftrace_graph_count)
3473 		return NULL;
3474 	return &ftrace_graph_funcs[*pos];
3475 }
3476 
3477 static void *
3478 g_next(struct seq_file *m, void *v, loff_t *pos)
3479 {
3480 	(*pos)++;
3481 	return __g_next(m, pos);
3482 }
3483 
3484 static void *g_start(struct seq_file *m, loff_t *pos)
3485 {
3486 	mutex_lock(&graph_lock);
3487 
3488 	/* Nothing, tell g_show to print all functions are enabled */
3489 	if (!ftrace_graph_filter_enabled && !*pos)
3490 		return (void *)1;
3491 
3492 	return __g_next(m, pos);
3493 }
3494 
3495 static void g_stop(struct seq_file *m, void *p)
3496 {
3497 	mutex_unlock(&graph_lock);
3498 }
3499 
3500 static int g_show(struct seq_file *m, void *v)
3501 {
3502 	unsigned long *ptr = v;
3503 
3504 	if (!ptr)
3505 		return 0;
3506 
3507 	if (ptr == (unsigned long *)1) {
3508 		seq_printf(m, "#### all functions enabled ####\n");
3509 		return 0;
3510 	}
3511 
3512 	seq_printf(m, "%ps\n", (void *)*ptr);
3513 
3514 	return 0;
3515 }
3516 
3517 static const struct seq_operations ftrace_graph_seq_ops = {
3518 	.start = g_start,
3519 	.next = g_next,
3520 	.stop = g_stop,
3521 	.show = g_show,
3522 };
3523 
3524 static int
3525 ftrace_graph_open(struct inode *inode, struct file *file)
3526 {
3527 	int ret = 0;
3528 
3529 	if (unlikely(ftrace_disabled))
3530 		return -ENODEV;
3531 
3532 	mutex_lock(&graph_lock);
3533 	if ((file->f_mode & FMODE_WRITE) &&
3534 	    (file->f_flags & O_TRUNC)) {
3535 		ftrace_graph_filter_enabled = 0;
3536 		ftrace_graph_count = 0;
3537 		memset(ftrace_graph_funcs, 0, sizeof(ftrace_graph_funcs));
3538 	}
3539 	mutex_unlock(&graph_lock);
3540 
3541 	if (file->f_mode & FMODE_READ)
3542 		ret = seq_open(file, &ftrace_graph_seq_ops);
3543 
3544 	return ret;
3545 }
3546 
3547 static int
3548 ftrace_graph_release(struct inode *inode, struct file *file)
3549 {
3550 	if (file->f_mode & FMODE_READ)
3551 		seq_release(inode, file);
3552 	return 0;
3553 }
3554 
3555 static int
3556 ftrace_set_func(unsigned long *array, int *idx, char *buffer)
3557 {
3558 	struct dyn_ftrace *rec;
3559 	struct ftrace_page *pg;
3560 	int search_len;
3561 	int fail = 1;
3562 	int type, not;
3563 	char *search;
3564 	bool exists;
3565 	int i;
3566 
3567 	/* decode regex */
3568 	type = filter_parse_regex(buffer, strlen(buffer), &search, &not);
3569 	if (!not && *idx >= FTRACE_GRAPH_MAX_FUNCS)
3570 		return -EBUSY;
3571 
3572 	search_len = strlen(search);
3573 
3574 	mutex_lock(&ftrace_lock);
3575 
3576 	if (unlikely(ftrace_disabled)) {
3577 		mutex_unlock(&ftrace_lock);
3578 		return -ENODEV;
3579 	}
3580 
3581 	do_for_each_ftrace_rec(pg, rec) {
3582 
3583 		if (ftrace_match_record(rec, NULL, search, search_len, type)) {
3584 			/* if it is in the array */
3585 			exists = false;
3586 			for (i = 0; i < *idx; i++) {
3587 				if (array[i] == rec->ip) {
3588 					exists = true;
3589 					break;
3590 				}
3591 			}
3592 
3593 			if (!not) {
3594 				fail = 0;
3595 				if (!exists) {
3596 					array[(*idx)++] = rec->ip;
3597 					if (*idx >= FTRACE_GRAPH_MAX_FUNCS)
3598 						goto out;
3599 				}
3600 			} else {
3601 				if (exists) {
3602 					array[i] = array[--(*idx)];
3603 					array[*idx] = 0;
3604 					fail = 0;
3605 				}
3606 			}
3607 		}
3608 	} while_for_each_ftrace_rec();
3609 out:
3610 	mutex_unlock(&ftrace_lock);
3611 
3612 	if (fail)
3613 		return -EINVAL;
3614 
3615 	ftrace_graph_filter_enabled = 1;
3616 	return 0;
3617 }
3618 
3619 static ssize_t
3620 ftrace_graph_write(struct file *file, const char __user *ubuf,
3621 		   size_t cnt, loff_t *ppos)
3622 {
3623 	struct trace_parser parser;
3624 	ssize_t read, ret;
3625 
3626 	if (!cnt)
3627 		return 0;
3628 
3629 	mutex_lock(&graph_lock);
3630 
3631 	if (trace_parser_get_init(&parser, FTRACE_BUFF_MAX)) {
3632 		ret = -ENOMEM;
3633 		goto out_unlock;
3634 	}
3635 
3636 	read = trace_get_user(&parser, ubuf, cnt, ppos);
3637 
3638 	if (read >= 0 && trace_parser_loaded((&parser))) {
3639 		parser.buffer[parser.idx] = 0;
3640 
3641 		/* we allow only one expression at a time */
3642 		ret = ftrace_set_func(ftrace_graph_funcs, &ftrace_graph_count,
3643 					parser.buffer);
3644 		if (ret)
3645 			goto out_free;
3646 	}
3647 
3648 	ret = read;
3649 
3650 out_free:
3651 	trace_parser_put(&parser);
3652 out_unlock:
3653 	mutex_unlock(&graph_lock);
3654 
3655 	return ret;
3656 }
3657 
3658 static const struct file_operations ftrace_graph_fops = {
3659 	.open		= ftrace_graph_open,
3660 	.read		= seq_read,
3661 	.write		= ftrace_graph_write,
3662 	.release	= ftrace_graph_release,
3663 	.llseek		= seq_lseek,
3664 };
3665 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
3666 
3667 static __init int ftrace_init_dyn_debugfs(struct dentry *d_tracer)
3668 {
3669 
3670 	trace_create_file("available_filter_functions", 0444,
3671 			d_tracer, NULL, &ftrace_avail_fops);
3672 
3673 	trace_create_file("enabled_functions", 0444,
3674 			d_tracer, NULL, &ftrace_enabled_fops);
3675 
3676 	trace_create_file("set_ftrace_filter", 0644, d_tracer,
3677 			NULL, &ftrace_filter_fops);
3678 
3679 	trace_create_file("set_ftrace_notrace", 0644, d_tracer,
3680 				    NULL, &ftrace_notrace_fops);
3681 
3682 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3683 	trace_create_file("set_graph_function", 0444, d_tracer,
3684 				    NULL,
3685 				    &ftrace_graph_fops);
3686 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
3687 
3688 	return 0;
3689 }
3690 
3691 static void ftrace_swap_recs(void *a, void *b, int size)
3692 {
3693 	struct dyn_ftrace *reca = a;
3694 	struct dyn_ftrace *recb = b;
3695 	struct dyn_ftrace t;
3696 
3697 	t = *reca;
3698 	*reca = *recb;
3699 	*recb = t;
3700 }
3701 
3702 static int ftrace_process_locs(struct module *mod,
3703 			       unsigned long *start,
3704 			       unsigned long *end)
3705 {
3706 	struct ftrace_page *pg;
3707 	unsigned long count;
3708 	unsigned long *p;
3709 	unsigned long addr;
3710 	unsigned long flags = 0; /* Shut up gcc */
3711 	int ret = -ENOMEM;
3712 
3713 	count = end - start;
3714 
3715 	if (!count)
3716 		return 0;
3717 
3718 	pg = ftrace_allocate_pages(count);
3719 	if (!pg)
3720 		return -ENOMEM;
3721 
3722 	mutex_lock(&ftrace_lock);
3723 
3724 	/*
3725 	 * Core and each module needs their own pages, as
3726 	 * modules will free them when they are removed.
3727 	 * Force a new page to be allocated for modules.
3728 	 */
3729 	if (!mod) {
3730 		WARN_ON(ftrace_pages || ftrace_pages_start);
3731 		/* First initialization */
3732 		ftrace_pages = ftrace_pages_start = pg;
3733 	} else {
3734 		if (!ftrace_pages)
3735 			goto out;
3736 
3737 		if (WARN_ON(ftrace_pages->next)) {
3738 			/* Hmm, we have free pages? */
3739 			while (ftrace_pages->next)
3740 				ftrace_pages = ftrace_pages->next;
3741 		}
3742 
3743 		ftrace_pages->next = pg;
3744 		ftrace_pages = pg;
3745 	}
3746 
3747 	p = start;
3748 	while (p < end) {
3749 		addr = ftrace_call_adjust(*p++);
3750 		/*
3751 		 * Some architecture linkers will pad between
3752 		 * the different mcount_loc sections of different
3753 		 * object files to satisfy alignments.
3754 		 * Skip any NULL pointers.
3755 		 */
3756 		if (!addr)
3757 			continue;
3758 		if (!ftrace_record_ip(addr))
3759 			break;
3760 	}
3761 
3762 	/* These new locations need to be initialized */
3763 	ftrace_new_pgs = pg;
3764 
3765 	/* Make each individual set of pages sorted by ips */
3766 	for (; pg; pg = pg->next)
3767 		sort(pg->records, pg->index, sizeof(struct dyn_ftrace),
3768 		     ftrace_cmp_recs, ftrace_swap_recs);
3769 
3770 	/*
3771 	 * We only need to disable interrupts on start up
3772 	 * because we are modifying code that an interrupt
3773 	 * may execute, and the modification is not atomic.
3774 	 * But for modules, nothing runs the code we modify
3775 	 * until we are finished with it, and there's no
3776 	 * reason to cause large interrupt latencies while we do it.
3777 	 */
3778 	if (!mod)
3779 		local_irq_save(flags);
3780 	ftrace_update_code(mod);
3781 	if (!mod)
3782 		local_irq_restore(flags);
3783 	ret = 0;
3784  out:
3785 	mutex_unlock(&ftrace_lock);
3786 
3787 	return ret;
3788 }
3789 
3790 #ifdef CONFIG_MODULES
3791 
3792 #define next_to_ftrace_page(p) container_of(p, struct ftrace_page, next)
3793 
3794 void ftrace_release_mod(struct module *mod)
3795 {
3796 	struct dyn_ftrace *rec;
3797 	struct ftrace_page **last_pg;
3798 	struct ftrace_page *pg;
3799 	int order;
3800 
3801 	mutex_lock(&ftrace_lock);
3802 
3803 	if (ftrace_disabled)
3804 		goto out_unlock;
3805 
3806 	/*
3807 	 * Each module has its own ftrace_pages, remove
3808 	 * them from the list.
3809 	 */
3810 	last_pg = &ftrace_pages_start;
3811 	for (pg = ftrace_pages_start; pg; pg = *last_pg) {
3812 		rec = &pg->records[0];
3813 		if (within_module_core(rec->ip, mod)) {
3814 			/*
3815 			 * As core pages are first, the first
3816 			 * page should never be a module page.
3817 			 */
3818 			if (WARN_ON(pg == ftrace_pages_start))
3819 				goto out_unlock;
3820 
3821 			/* Check if we are deleting the last page */
3822 			if (pg == ftrace_pages)
3823 				ftrace_pages = next_to_ftrace_page(last_pg);
3824 
3825 			*last_pg = pg->next;
3826 			order = get_count_order(pg->size / ENTRIES_PER_PAGE);
3827 			free_pages((unsigned long)pg->records, order);
3828 			kfree(pg);
3829 		} else
3830 			last_pg = &pg->next;
3831 	}
3832  out_unlock:
3833 	mutex_unlock(&ftrace_lock);
3834 }
3835 
3836 static void ftrace_init_module(struct module *mod,
3837 			       unsigned long *start, unsigned long *end)
3838 {
3839 	if (ftrace_disabled || start == end)
3840 		return;
3841 	ftrace_process_locs(mod, start, end);
3842 }
3843 
3844 static int ftrace_module_notify(struct notifier_block *self,
3845 				unsigned long val, void *data)
3846 {
3847 	struct module *mod = data;
3848 
3849 	switch (val) {
3850 	case MODULE_STATE_COMING:
3851 		ftrace_init_module(mod, mod->ftrace_callsites,
3852 				   mod->ftrace_callsites +
3853 				   mod->num_ftrace_callsites);
3854 		break;
3855 	case MODULE_STATE_GOING:
3856 		ftrace_release_mod(mod);
3857 		break;
3858 	}
3859 
3860 	return 0;
3861 }
3862 #else
3863 static int ftrace_module_notify(struct notifier_block *self,
3864 				unsigned long val, void *data)
3865 {
3866 	return 0;
3867 }
3868 #endif /* CONFIG_MODULES */
3869 
3870 struct notifier_block ftrace_module_nb = {
3871 	.notifier_call = ftrace_module_notify,
3872 	.priority = 0,
3873 };
3874 
3875 extern unsigned long __start_mcount_loc[];
3876 extern unsigned long __stop_mcount_loc[];
3877 
3878 void __init ftrace_init(void)
3879 {
3880 	unsigned long count, addr, flags;
3881 	int ret;
3882 
3883 	/* Keep the ftrace pointer to the stub */
3884 	addr = (unsigned long)ftrace_stub;
3885 
3886 	local_irq_save(flags);
3887 	ftrace_dyn_arch_init(&addr);
3888 	local_irq_restore(flags);
3889 
3890 	/* ftrace_dyn_arch_init places the return code in addr */
3891 	if (addr)
3892 		goto failed;
3893 
3894 	count = __stop_mcount_loc - __start_mcount_loc;
3895 
3896 	ret = ftrace_dyn_table_alloc(count);
3897 	if (ret)
3898 		goto failed;
3899 
3900 	last_ftrace_enabled = ftrace_enabled = 1;
3901 
3902 	ret = ftrace_process_locs(NULL,
3903 				  __start_mcount_loc,
3904 				  __stop_mcount_loc);
3905 
3906 	ret = register_module_notifier(&ftrace_module_nb);
3907 	if (ret)
3908 		pr_warning("Failed to register trace ftrace module notifier\n");
3909 
3910 	set_ftrace_early_filters();
3911 
3912 	return;
3913  failed:
3914 	ftrace_disabled = 1;
3915 }
3916 
3917 #else
3918 
3919 static struct ftrace_ops global_ops = {
3920 	.func			= ftrace_stub,
3921 };
3922 
3923 static int __init ftrace_nodyn_init(void)
3924 {
3925 	ftrace_enabled = 1;
3926 	return 0;
3927 }
3928 device_initcall(ftrace_nodyn_init);
3929 
3930 static inline int ftrace_init_dyn_debugfs(struct dentry *d_tracer) { return 0; }
3931 static inline void ftrace_startup_enable(int command) { }
3932 /* Keep as macros so we do not need to define the commands */
3933 # define ftrace_startup(ops, command)			\
3934 	({						\
3935 		(ops)->flags |= FTRACE_OPS_FL_ENABLED;	\
3936 		0;					\
3937 	})
3938 # define ftrace_shutdown(ops, command)	do { } while (0)
3939 # define ftrace_startup_sysctl()	do { } while (0)
3940 # define ftrace_shutdown_sysctl()	do { } while (0)
3941 
3942 static inline int
3943 ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip)
3944 {
3945 	return 1;
3946 }
3947 
3948 #endif /* CONFIG_DYNAMIC_FTRACE */
3949 
3950 static void
3951 ftrace_ops_control_func(unsigned long ip, unsigned long parent_ip)
3952 {
3953 	struct ftrace_ops *op;
3954 
3955 	if (unlikely(trace_recursion_test(TRACE_CONTROL_BIT)))
3956 		return;
3957 
3958 	/*
3959 	 * Some of the ops may be dynamically allocated,
3960 	 * they must be freed after a synchronize_sched().
3961 	 */
3962 	preempt_disable_notrace();
3963 	trace_recursion_set(TRACE_CONTROL_BIT);
3964 	op = rcu_dereference_raw(ftrace_control_list);
3965 	while (op != &ftrace_list_end) {
3966 		if (!ftrace_function_local_disabled(op) &&
3967 		    ftrace_ops_test(op, ip))
3968 			op->func(ip, parent_ip);
3969 
3970 		op = rcu_dereference_raw(op->next);
3971 	};
3972 	trace_recursion_clear(TRACE_CONTROL_BIT);
3973 	preempt_enable_notrace();
3974 }
3975 
3976 static struct ftrace_ops control_ops = {
3977 	.func = ftrace_ops_control_func,
3978 };
3979 
3980 static void
3981 ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip)
3982 {
3983 	struct ftrace_ops *op;
3984 
3985 	if (unlikely(trace_recursion_test(TRACE_INTERNAL_BIT)))
3986 		return;
3987 
3988 	trace_recursion_set(TRACE_INTERNAL_BIT);
3989 	/*
3990 	 * Some of the ops may be dynamically allocated,
3991 	 * they must be freed after a synchronize_sched().
3992 	 */
3993 	preempt_disable_notrace();
3994 	op = rcu_dereference_raw(ftrace_ops_list);
3995 	while (op != &ftrace_list_end) {
3996 		if (ftrace_ops_test(op, ip))
3997 			op->func(ip, parent_ip);
3998 		op = rcu_dereference_raw(op->next);
3999 	};
4000 	preempt_enable_notrace();
4001 	trace_recursion_clear(TRACE_INTERNAL_BIT);
4002 }
4003 
4004 static void clear_ftrace_swapper(void)
4005 {
4006 	struct task_struct *p;
4007 	int cpu;
4008 
4009 	get_online_cpus();
4010 	for_each_online_cpu(cpu) {
4011 		p = idle_task(cpu);
4012 		clear_tsk_trace_trace(p);
4013 	}
4014 	put_online_cpus();
4015 }
4016 
4017 static void set_ftrace_swapper(void)
4018 {
4019 	struct task_struct *p;
4020 	int cpu;
4021 
4022 	get_online_cpus();
4023 	for_each_online_cpu(cpu) {
4024 		p = idle_task(cpu);
4025 		set_tsk_trace_trace(p);
4026 	}
4027 	put_online_cpus();
4028 }
4029 
4030 static void clear_ftrace_pid(struct pid *pid)
4031 {
4032 	struct task_struct *p;
4033 
4034 	rcu_read_lock();
4035 	do_each_pid_task(pid, PIDTYPE_PID, p) {
4036 		clear_tsk_trace_trace(p);
4037 	} while_each_pid_task(pid, PIDTYPE_PID, p);
4038 	rcu_read_unlock();
4039 
4040 	put_pid(pid);
4041 }
4042 
4043 static void set_ftrace_pid(struct pid *pid)
4044 {
4045 	struct task_struct *p;
4046 
4047 	rcu_read_lock();
4048 	do_each_pid_task(pid, PIDTYPE_PID, p) {
4049 		set_tsk_trace_trace(p);
4050 	} while_each_pid_task(pid, PIDTYPE_PID, p);
4051 	rcu_read_unlock();
4052 }
4053 
4054 static void clear_ftrace_pid_task(struct pid *pid)
4055 {
4056 	if (pid == ftrace_swapper_pid)
4057 		clear_ftrace_swapper();
4058 	else
4059 		clear_ftrace_pid(pid);
4060 }
4061 
4062 static void set_ftrace_pid_task(struct pid *pid)
4063 {
4064 	if (pid == ftrace_swapper_pid)
4065 		set_ftrace_swapper();
4066 	else
4067 		set_ftrace_pid(pid);
4068 }
4069 
4070 static int ftrace_pid_add(int p)
4071 {
4072 	struct pid *pid;
4073 	struct ftrace_pid *fpid;
4074 	int ret = -EINVAL;
4075 
4076 	mutex_lock(&ftrace_lock);
4077 
4078 	if (!p)
4079 		pid = ftrace_swapper_pid;
4080 	else
4081 		pid = find_get_pid(p);
4082 
4083 	if (!pid)
4084 		goto out;
4085 
4086 	ret = 0;
4087 
4088 	list_for_each_entry(fpid, &ftrace_pids, list)
4089 		if (fpid->pid == pid)
4090 			goto out_put;
4091 
4092 	ret = -ENOMEM;
4093 
4094 	fpid = kmalloc(sizeof(*fpid), GFP_KERNEL);
4095 	if (!fpid)
4096 		goto out_put;
4097 
4098 	list_add(&fpid->list, &ftrace_pids);
4099 	fpid->pid = pid;
4100 
4101 	set_ftrace_pid_task(pid);
4102 
4103 	ftrace_update_pid_func();
4104 	ftrace_startup_enable(0);
4105 
4106 	mutex_unlock(&ftrace_lock);
4107 	return 0;
4108 
4109 out_put:
4110 	if (pid != ftrace_swapper_pid)
4111 		put_pid(pid);
4112 
4113 out:
4114 	mutex_unlock(&ftrace_lock);
4115 	return ret;
4116 }
4117 
4118 static void ftrace_pid_reset(void)
4119 {
4120 	struct ftrace_pid *fpid, *safe;
4121 
4122 	mutex_lock(&ftrace_lock);
4123 	list_for_each_entry_safe(fpid, safe, &ftrace_pids, list) {
4124 		struct pid *pid = fpid->pid;
4125 
4126 		clear_ftrace_pid_task(pid);
4127 
4128 		list_del(&fpid->list);
4129 		kfree(fpid);
4130 	}
4131 
4132 	ftrace_update_pid_func();
4133 	ftrace_startup_enable(0);
4134 
4135 	mutex_unlock(&ftrace_lock);
4136 }
4137 
4138 static void *fpid_start(struct seq_file *m, loff_t *pos)
4139 {
4140 	mutex_lock(&ftrace_lock);
4141 
4142 	if (list_empty(&ftrace_pids) && (!*pos))
4143 		return (void *) 1;
4144 
4145 	return seq_list_start(&ftrace_pids, *pos);
4146 }
4147 
4148 static void *fpid_next(struct seq_file *m, void *v, loff_t *pos)
4149 {
4150 	if (v == (void *)1)
4151 		return NULL;
4152 
4153 	return seq_list_next(v, &ftrace_pids, pos);
4154 }
4155 
4156 static void fpid_stop(struct seq_file *m, void *p)
4157 {
4158 	mutex_unlock(&ftrace_lock);
4159 }
4160 
4161 static int fpid_show(struct seq_file *m, void *v)
4162 {
4163 	const struct ftrace_pid *fpid = list_entry(v, struct ftrace_pid, list);
4164 
4165 	if (v == (void *)1) {
4166 		seq_printf(m, "no pid\n");
4167 		return 0;
4168 	}
4169 
4170 	if (fpid->pid == ftrace_swapper_pid)
4171 		seq_printf(m, "swapper tasks\n");
4172 	else
4173 		seq_printf(m, "%u\n", pid_vnr(fpid->pid));
4174 
4175 	return 0;
4176 }
4177 
4178 static const struct seq_operations ftrace_pid_sops = {
4179 	.start = fpid_start,
4180 	.next = fpid_next,
4181 	.stop = fpid_stop,
4182 	.show = fpid_show,
4183 };
4184 
4185 static int
4186 ftrace_pid_open(struct inode *inode, struct file *file)
4187 {
4188 	int ret = 0;
4189 
4190 	if ((file->f_mode & FMODE_WRITE) &&
4191 	    (file->f_flags & O_TRUNC))
4192 		ftrace_pid_reset();
4193 
4194 	if (file->f_mode & FMODE_READ)
4195 		ret = seq_open(file, &ftrace_pid_sops);
4196 
4197 	return ret;
4198 }
4199 
4200 static ssize_t
4201 ftrace_pid_write(struct file *filp, const char __user *ubuf,
4202 		   size_t cnt, loff_t *ppos)
4203 {
4204 	char buf[64], *tmp;
4205 	long val;
4206 	int ret;
4207 
4208 	if (cnt >= sizeof(buf))
4209 		return -EINVAL;
4210 
4211 	if (copy_from_user(&buf, ubuf, cnt))
4212 		return -EFAULT;
4213 
4214 	buf[cnt] = 0;
4215 
4216 	/*
4217 	 * Allow "echo > set_ftrace_pid" or "echo -n '' > set_ftrace_pid"
4218 	 * to clean the filter quietly.
4219 	 */
4220 	tmp = strstrip(buf);
4221 	if (strlen(tmp) == 0)
4222 		return 1;
4223 
4224 	ret = strict_strtol(tmp, 10, &val);
4225 	if (ret < 0)
4226 		return ret;
4227 
4228 	ret = ftrace_pid_add(val);
4229 
4230 	return ret ? ret : cnt;
4231 }
4232 
4233 static int
4234 ftrace_pid_release(struct inode *inode, struct file *file)
4235 {
4236 	if (file->f_mode & FMODE_READ)
4237 		seq_release(inode, file);
4238 
4239 	return 0;
4240 }
4241 
4242 static const struct file_operations ftrace_pid_fops = {
4243 	.open		= ftrace_pid_open,
4244 	.write		= ftrace_pid_write,
4245 	.read		= seq_read,
4246 	.llseek		= seq_lseek,
4247 	.release	= ftrace_pid_release,
4248 };
4249 
4250 static __init int ftrace_init_debugfs(void)
4251 {
4252 	struct dentry *d_tracer;
4253 
4254 	d_tracer = tracing_init_dentry();
4255 	if (!d_tracer)
4256 		return 0;
4257 
4258 	ftrace_init_dyn_debugfs(d_tracer);
4259 
4260 	trace_create_file("set_ftrace_pid", 0644, d_tracer,
4261 			    NULL, &ftrace_pid_fops);
4262 
4263 	ftrace_profile_debugfs(d_tracer);
4264 
4265 	return 0;
4266 }
4267 fs_initcall(ftrace_init_debugfs);
4268 
4269 /**
4270  * ftrace_kill - kill ftrace
4271  *
4272  * This function should be used by panic code. It stops ftrace
4273  * but in a not so nice way. If you need to simply kill ftrace
4274  * from a non-atomic section, use ftrace_kill.
4275  */
4276 void ftrace_kill(void)
4277 {
4278 	ftrace_disabled = 1;
4279 	ftrace_enabled = 0;
4280 	clear_ftrace_function();
4281 }
4282 
4283 /**
4284  * Test if ftrace is dead or not.
4285  */
4286 int ftrace_is_dead(void)
4287 {
4288 	return ftrace_disabled;
4289 }
4290 
4291 /**
4292  * register_ftrace_function - register a function for profiling
4293  * @ops - ops structure that holds the function for profiling.
4294  *
4295  * Register a function to be called by all functions in the
4296  * kernel.
4297  *
4298  * Note: @ops->func and all the functions it calls must be labeled
4299  *       with "notrace", otherwise it will go into a
4300  *       recursive loop.
4301  */
4302 int register_ftrace_function(struct ftrace_ops *ops)
4303 {
4304 	int ret = -1;
4305 
4306 	mutex_lock(&ftrace_lock);
4307 
4308 	if (unlikely(ftrace_disabled))
4309 		goto out_unlock;
4310 
4311 	ret = __register_ftrace_function(ops);
4312 	if (!ret)
4313 		ret = ftrace_startup(ops, 0);
4314 
4315 
4316  out_unlock:
4317 	mutex_unlock(&ftrace_lock);
4318 	return ret;
4319 }
4320 EXPORT_SYMBOL_GPL(register_ftrace_function);
4321 
4322 /**
4323  * unregister_ftrace_function - unregister a function for profiling.
4324  * @ops - ops structure that holds the function to unregister
4325  *
4326  * Unregister a function that was added to be called by ftrace profiling.
4327  */
4328 int unregister_ftrace_function(struct ftrace_ops *ops)
4329 {
4330 	int ret;
4331 
4332 	mutex_lock(&ftrace_lock);
4333 	ret = __unregister_ftrace_function(ops);
4334 	if (!ret)
4335 		ftrace_shutdown(ops, 0);
4336 	mutex_unlock(&ftrace_lock);
4337 
4338 	return ret;
4339 }
4340 EXPORT_SYMBOL_GPL(unregister_ftrace_function);
4341 
4342 int
4343 ftrace_enable_sysctl(struct ctl_table *table, int write,
4344 		     void __user *buffer, size_t *lenp,
4345 		     loff_t *ppos)
4346 {
4347 	int ret = -ENODEV;
4348 
4349 	mutex_lock(&ftrace_lock);
4350 
4351 	if (unlikely(ftrace_disabled))
4352 		goto out;
4353 
4354 	ret = proc_dointvec(table, write, buffer, lenp, ppos);
4355 
4356 	if (ret || !write || (last_ftrace_enabled == !!ftrace_enabled))
4357 		goto out;
4358 
4359 	last_ftrace_enabled = !!ftrace_enabled;
4360 
4361 	if (ftrace_enabled) {
4362 
4363 		ftrace_startup_sysctl();
4364 
4365 		/* we are starting ftrace again */
4366 		if (ftrace_ops_list != &ftrace_list_end) {
4367 			if (ftrace_ops_list->next == &ftrace_list_end)
4368 				ftrace_trace_function = ftrace_ops_list->func;
4369 			else
4370 				ftrace_trace_function = ftrace_ops_list_func;
4371 		}
4372 
4373 	} else {
4374 		/* stopping ftrace calls (just send to ftrace_stub) */
4375 		ftrace_trace_function = ftrace_stub;
4376 
4377 		ftrace_shutdown_sysctl();
4378 	}
4379 
4380  out:
4381 	mutex_unlock(&ftrace_lock);
4382 	return ret;
4383 }
4384 
4385 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
4386 
4387 static int ftrace_graph_active;
4388 static struct notifier_block ftrace_suspend_notifier;
4389 
4390 int ftrace_graph_entry_stub(struct ftrace_graph_ent *trace)
4391 {
4392 	return 0;
4393 }
4394 
4395 /* The callbacks that hook a function */
4396 trace_func_graph_ret_t ftrace_graph_return =
4397 			(trace_func_graph_ret_t)ftrace_stub;
4398 trace_func_graph_ent_t ftrace_graph_entry = ftrace_graph_entry_stub;
4399 
4400 /* Try to assign a return stack array on FTRACE_RETSTACK_ALLOC_SIZE tasks. */
4401 static int alloc_retstack_tasklist(struct ftrace_ret_stack **ret_stack_list)
4402 {
4403 	int i;
4404 	int ret = 0;
4405 	unsigned long flags;
4406 	int start = 0, end = FTRACE_RETSTACK_ALLOC_SIZE;
4407 	struct task_struct *g, *t;
4408 
4409 	for (i = 0; i < FTRACE_RETSTACK_ALLOC_SIZE; i++) {
4410 		ret_stack_list[i] = kmalloc(FTRACE_RETFUNC_DEPTH
4411 					* sizeof(struct ftrace_ret_stack),
4412 					GFP_KERNEL);
4413 		if (!ret_stack_list[i]) {
4414 			start = 0;
4415 			end = i;
4416 			ret = -ENOMEM;
4417 			goto free;
4418 		}
4419 	}
4420 
4421 	read_lock_irqsave(&tasklist_lock, flags);
4422 	do_each_thread(g, t) {
4423 		if (start == end) {
4424 			ret = -EAGAIN;
4425 			goto unlock;
4426 		}
4427 
4428 		if (t->ret_stack == NULL) {
4429 			atomic_set(&t->tracing_graph_pause, 0);
4430 			atomic_set(&t->trace_overrun, 0);
4431 			t->curr_ret_stack = -1;
4432 			/* Make sure the tasks see the -1 first: */
4433 			smp_wmb();
4434 			t->ret_stack = ret_stack_list[start++];
4435 		}
4436 	} while_each_thread(g, t);
4437 
4438 unlock:
4439 	read_unlock_irqrestore(&tasklist_lock, flags);
4440 free:
4441 	for (i = start; i < end; i++)
4442 		kfree(ret_stack_list[i]);
4443 	return ret;
4444 }
4445 
4446 static void
4447 ftrace_graph_probe_sched_switch(void *ignore,
4448 			struct task_struct *prev, struct task_struct *next)
4449 {
4450 	unsigned long long timestamp;
4451 	int index;
4452 
4453 	/*
4454 	 * Does the user want to count the time a function was asleep.
4455 	 * If so, do not update the time stamps.
4456 	 */
4457 	if (trace_flags & TRACE_ITER_SLEEP_TIME)
4458 		return;
4459 
4460 	timestamp = trace_clock_local();
4461 
4462 	prev->ftrace_timestamp = timestamp;
4463 
4464 	/* only process tasks that we timestamped */
4465 	if (!next->ftrace_timestamp)
4466 		return;
4467 
4468 	/*
4469 	 * Update all the counters in next to make up for the
4470 	 * time next was sleeping.
4471 	 */
4472 	timestamp -= next->ftrace_timestamp;
4473 
4474 	for (index = next->curr_ret_stack; index >= 0; index--)
4475 		next->ret_stack[index].calltime += timestamp;
4476 }
4477 
4478 /* Allocate a return stack for each task */
4479 static int start_graph_tracing(void)
4480 {
4481 	struct ftrace_ret_stack **ret_stack_list;
4482 	int ret, cpu;
4483 
4484 	ret_stack_list = kmalloc(FTRACE_RETSTACK_ALLOC_SIZE *
4485 				sizeof(struct ftrace_ret_stack *),
4486 				GFP_KERNEL);
4487 
4488 	if (!ret_stack_list)
4489 		return -ENOMEM;
4490 
4491 	/* The cpu_boot init_task->ret_stack will never be freed */
4492 	for_each_online_cpu(cpu) {
4493 		if (!idle_task(cpu)->ret_stack)
4494 			ftrace_graph_init_idle_task(idle_task(cpu), cpu);
4495 	}
4496 
4497 	do {
4498 		ret = alloc_retstack_tasklist(ret_stack_list);
4499 	} while (ret == -EAGAIN);
4500 
4501 	if (!ret) {
4502 		ret = register_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
4503 		if (ret)
4504 			pr_info("ftrace_graph: Couldn't activate tracepoint"
4505 				" probe to kernel_sched_switch\n");
4506 	}
4507 
4508 	kfree(ret_stack_list);
4509 	return ret;
4510 }
4511 
4512 /*
4513  * Hibernation protection.
4514  * The state of the current task is too much unstable during
4515  * suspend/restore to disk. We want to protect against that.
4516  */
4517 static int
4518 ftrace_suspend_notifier_call(struct notifier_block *bl, unsigned long state,
4519 							void *unused)
4520 {
4521 	switch (state) {
4522 	case PM_HIBERNATION_PREPARE:
4523 		pause_graph_tracing();
4524 		break;
4525 
4526 	case PM_POST_HIBERNATION:
4527 		unpause_graph_tracing();
4528 		break;
4529 	}
4530 	return NOTIFY_DONE;
4531 }
4532 
4533 int register_ftrace_graph(trace_func_graph_ret_t retfunc,
4534 			trace_func_graph_ent_t entryfunc)
4535 {
4536 	int ret = 0;
4537 
4538 	mutex_lock(&ftrace_lock);
4539 
4540 	/* we currently allow only one tracer registered at a time */
4541 	if (ftrace_graph_active) {
4542 		ret = -EBUSY;
4543 		goto out;
4544 	}
4545 
4546 	ftrace_suspend_notifier.notifier_call = ftrace_suspend_notifier_call;
4547 	register_pm_notifier(&ftrace_suspend_notifier);
4548 
4549 	ftrace_graph_active++;
4550 	ret = start_graph_tracing();
4551 	if (ret) {
4552 		ftrace_graph_active--;
4553 		goto out;
4554 	}
4555 
4556 	ftrace_graph_return = retfunc;
4557 	ftrace_graph_entry = entryfunc;
4558 
4559 	ret = ftrace_startup(&global_ops, FTRACE_START_FUNC_RET);
4560 
4561 out:
4562 	mutex_unlock(&ftrace_lock);
4563 	return ret;
4564 }
4565 
4566 void unregister_ftrace_graph(void)
4567 {
4568 	mutex_lock(&ftrace_lock);
4569 
4570 	if (unlikely(!ftrace_graph_active))
4571 		goto out;
4572 
4573 	ftrace_graph_active--;
4574 	ftrace_graph_return = (trace_func_graph_ret_t)ftrace_stub;
4575 	ftrace_graph_entry = ftrace_graph_entry_stub;
4576 	ftrace_shutdown(&global_ops, FTRACE_STOP_FUNC_RET);
4577 	unregister_pm_notifier(&ftrace_suspend_notifier);
4578 	unregister_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
4579 
4580  out:
4581 	mutex_unlock(&ftrace_lock);
4582 }
4583 
4584 static DEFINE_PER_CPU(struct ftrace_ret_stack *, idle_ret_stack);
4585 
4586 static void
4587 graph_init_task(struct task_struct *t, struct ftrace_ret_stack *ret_stack)
4588 {
4589 	atomic_set(&t->tracing_graph_pause, 0);
4590 	atomic_set(&t->trace_overrun, 0);
4591 	t->ftrace_timestamp = 0;
4592 	/* make curr_ret_stack visible before we add the ret_stack */
4593 	smp_wmb();
4594 	t->ret_stack = ret_stack;
4595 }
4596 
4597 /*
4598  * Allocate a return stack for the idle task. May be the first
4599  * time through, or it may be done by CPU hotplug online.
4600  */
4601 void ftrace_graph_init_idle_task(struct task_struct *t, int cpu)
4602 {
4603 	t->curr_ret_stack = -1;
4604 	/*
4605 	 * The idle task has no parent, it either has its own
4606 	 * stack or no stack at all.
4607 	 */
4608 	if (t->ret_stack)
4609 		WARN_ON(t->ret_stack != per_cpu(idle_ret_stack, cpu));
4610 
4611 	if (ftrace_graph_active) {
4612 		struct ftrace_ret_stack *ret_stack;
4613 
4614 		ret_stack = per_cpu(idle_ret_stack, cpu);
4615 		if (!ret_stack) {
4616 			ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
4617 					    * sizeof(struct ftrace_ret_stack),
4618 					    GFP_KERNEL);
4619 			if (!ret_stack)
4620 				return;
4621 			per_cpu(idle_ret_stack, cpu) = ret_stack;
4622 		}
4623 		graph_init_task(t, ret_stack);
4624 	}
4625 }
4626 
4627 /* Allocate a return stack for newly created task */
4628 void ftrace_graph_init_task(struct task_struct *t)
4629 {
4630 	/* Make sure we do not use the parent ret_stack */
4631 	t->ret_stack = NULL;
4632 	t->curr_ret_stack = -1;
4633 
4634 	if (ftrace_graph_active) {
4635 		struct ftrace_ret_stack *ret_stack;
4636 
4637 		ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
4638 				* sizeof(struct ftrace_ret_stack),
4639 				GFP_KERNEL);
4640 		if (!ret_stack)
4641 			return;
4642 		graph_init_task(t, ret_stack);
4643 	}
4644 }
4645 
4646 void ftrace_graph_exit_task(struct task_struct *t)
4647 {
4648 	struct ftrace_ret_stack	*ret_stack = t->ret_stack;
4649 
4650 	t->ret_stack = NULL;
4651 	/* NULL must become visible to IRQs before we free it: */
4652 	barrier();
4653 
4654 	kfree(ret_stack);
4655 }
4656 
4657 void ftrace_graph_stop(void)
4658 {
4659 	ftrace_stop();
4660 }
4661 #endif
4662