xref: /linux/kernel/trace/ftrace.c (revision 9ffc93f203c18a70623f21950f1dd473c9ec48cd)
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 = ftrace_test_stop_func;
253 #endif
254 }
255 
256 static void add_ftrace_ops(struct ftrace_ops **list, struct ftrace_ops *ops)
257 {
258 	ops->next = *list;
259 	/*
260 	 * We are entering ops into the list but another
261 	 * CPU might be walking that list. We need to make sure
262 	 * the ops->next pointer is valid before another CPU sees
263 	 * the ops pointer included into the list.
264 	 */
265 	rcu_assign_pointer(*list, ops);
266 }
267 
268 static int remove_ftrace_ops(struct ftrace_ops **list, struct ftrace_ops *ops)
269 {
270 	struct ftrace_ops **p;
271 
272 	/*
273 	 * If we are removing the last function, then simply point
274 	 * to the ftrace_stub.
275 	 */
276 	if (*list == ops && ops->next == &ftrace_list_end) {
277 		*list = &ftrace_list_end;
278 		return 0;
279 	}
280 
281 	for (p = list; *p != &ftrace_list_end; p = &(*p)->next)
282 		if (*p == ops)
283 			break;
284 
285 	if (*p != ops)
286 		return -1;
287 
288 	*p = (*p)->next;
289 	return 0;
290 }
291 
292 static void add_ftrace_list_ops(struct ftrace_ops **list,
293 				struct ftrace_ops *main_ops,
294 				struct ftrace_ops *ops)
295 {
296 	int first = *list == &ftrace_list_end;
297 	add_ftrace_ops(list, ops);
298 	if (first)
299 		add_ftrace_ops(&ftrace_ops_list, main_ops);
300 }
301 
302 static int remove_ftrace_list_ops(struct ftrace_ops **list,
303 				  struct ftrace_ops *main_ops,
304 				  struct ftrace_ops *ops)
305 {
306 	int ret = remove_ftrace_ops(list, ops);
307 	if (!ret && *list == &ftrace_list_end)
308 		ret = remove_ftrace_ops(&ftrace_ops_list, main_ops);
309 	return ret;
310 }
311 
312 static int __register_ftrace_function(struct ftrace_ops *ops)
313 {
314 	if (ftrace_disabled)
315 		return -ENODEV;
316 
317 	if (FTRACE_WARN_ON(ops == &global_ops))
318 		return -EINVAL;
319 
320 	if (WARN_ON(ops->flags & FTRACE_OPS_FL_ENABLED))
321 		return -EBUSY;
322 
323 	/* We don't support both control and global flags set. */
324 	if ((ops->flags & FL_GLOBAL_CONTROL_MASK) == FL_GLOBAL_CONTROL_MASK)
325 		return -EINVAL;
326 
327 	if (!core_kernel_data((unsigned long)ops))
328 		ops->flags |= FTRACE_OPS_FL_DYNAMIC;
329 
330 	if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
331 		add_ftrace_list_ops(&ftrace_global_list, &global_ops, ops);
332 		ops->flags |= FTRACE_OPS_FL_ENABLED;
333 	} else if (ops->flags & FTRACE_OPS_FL_CONTROL) {
334 		if (control_ops_alloc(ops))
335 			return -ENOMEM;
336 		add_ftrace_list_ops(&ftrace_control_list, &control_ops, ops);
337 	} else
338 		add_ftrace_ops(&ftrace_ops_list, ops);
339 
340 	if (ftrace_enabled)
341 		update_ftrace_function();
342 
343 	return 0;
344 }
345 
346 static int __unregister_ftrace_function(struct ftrace_ops *ops)
347 {
348 	int ret;
349 
350 	if (ftrace_disabled)
351 		return -ENODEV;
352 
353 	if (WARN_ON(!(ops->flags & FTRACE_OPS_FL_ENABLED)))
354 		return -EBUSY;
355 
356 	if (FTRACE_WARN_ON(ops == &global_ops))
357 		return -EINVAL;
358 
359 	if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
360 		ret = remove_ftrace_list_ops(&ftrace_global_list,
361 					     &global_ops, ops);
362 		if (!ret)
363 			ops->flags &= ~FTRACE_OPS_FL_ENABLED;
364 	} else if (ops->flags & FTRACE_OPS_FL_CONTROL) {
365 		ret = remove_ftrace_list_ops(&ftrace_control_list,
366 					     &control_ops, ops);
367 		if (!ret) {
368 			/*
369 			 * The ftrace_ops is now removed from the list,
370 			 * so there'll be no new users. We must ensure
371 			 * all current users are done before we free
372 			 * the control data.
373 			 */
374 			synchronize_sched();
375 			control_ops_free(ops);
376 		}
377 	} else
378 		ret = remove_ftrace_ops(&ftrace_ops_list, ops);
379 
380 	if (ret < 0)
381 		return ret;
382 
383 	if (ftrace_enabled)
384 		update_ftrace_function();
385 
386 	/*
387 	 * Dynamic ops may be freed, we must make sure that all
388 	 * callers are done before leaving this function.
389 	 */
390 	if (ops->flags & FTRACE_OPS_FL_DYNAMIC)
391 		synchronize_sched();
392 
393 	return 0;
394 }
395 
396 static void ftrace_update_pid_func(void)
397 {
398 	/* Only do something if we are tracing something */
399 	if (ftrace_trace_function == ftrace_stub)
400 		return;
401 
402 	update_ftrace_function();
403 }
404 
405 #ifdef CONFIG_FUNCTION_PROFILER
406 struct ftrace_profile {
407 	struct hlist_node		node;
408 	unsigned long			ip;
409 	unsigned long			counter;
410 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
411 	unsigned long long		time;
412 	unsigned long long		time_squared;
413 #endif
414 };
415 
416 struct ftrace_profile_page {
417 	struct ftrace_profile_page	*next;
418 	unsigned long			index;
419 	struct ftrace_profile		records[];
420 };
421 
422 struct ftrace_profile_stat {
423 	atomic_t			disabled;
424 	struct hlist_head		*hash;
425 	struct ftrace_profile_page	*pages;
426 	struct ftrace_profile_page	*start;
427 	struct tracer_stat		stat;
428 };
429 
430 #define PROFILE_RECORDS_SIZE						\
431 	(PAGE_SIZE - offsetof(struct ftrace_profile_page, records))
432 
433 #define PROFILES_PER_PAGE					\
434 	(PROFILE_RECORDS_SIZE / sizeof(struct ftrace_profile))
435 
436 static int ftrace_profile_bits __read_mostly;
437 static int ftrace_profile_enabled __read_mostly;
438 
439 /* ftrace_profile_lock - synchronize the enable and disable of the profiler */
440 static DEFINE_MUTEX(ftrace_profile_lock);
441 
442 static DEFINE_PER_CPU(struct ftrace_profile_stat, ftrace_profile_stats);
443 
444 #define FTRACE_PROFILE_HASH_SIZE 1024 /* must be power of 2 */
445 
446 static void *
447 function_stat_next(void *v, int idx)
448 {
449 	struct ftrace_profile *rec = v;
450 	struct ftrace_profile_page *pg;
451 
452 	pg = (struct ftrace_profile_page *)((unsigned long)rec & PAGE_MASK);
453 
454  again:
455 	if (idx != 0)
456 		rec++;
457 
458 	if ((void *)rec >= (void *)&pg->records[pg->index]) {
459 		pg = pg->next;
460 		if (!pg)
461 			return NULL;
462 		rec = &pg->records[0];
463 		if (!rec->counter)
464 			goto again;
465 	}
466 
467 	return rec;
468 }
469 
470 static void *function_stat_start(struct tracer_stat *trace)
471 {
472 	struct ftrace_profile_stat *stat =
473 		container_of(trace, struct ftrace_profile_stat, stat);
474 
475 	if (!stat || !stat->start)
476 		return NULL;
477 
478 	return function_stat_next(&stat->start->records[0], 0);
479 }
480 
481 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
482 /* function graph compares on total time */
483 static int function_stat_cmp(void *p1, void *p2)
484 {
485 	struct ftrace_profile *a = p1;
486 	struct ftrace_profile *b = p2;
487 
488 	if (a->time < b->time)
489 		return -1;
490 	if (a->time > b->time)
491 		return 1;
492 	else
493 		return 0;
494 }
495 #else
496 /* not function graph compares against hits */
497 static int function_stat_cmp(void *p1, void *p2)
498 {
499 	struct ftrace_profile *a = p1;
500 	struct ftrace_profile *b = p2;
501 
502 	if (a->counter < b->counter)
503 		return -1;
504 	if (a->counter > b->counter)
505 		return 1;
506 	else
507 		return 0;
508 }
509 #endif
510 
511 static int function_stat_headers(struct seq_file *m)
512 {
513 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
514 	seq_printf(m, "  Function                               "
515 		   "Hit    Time            Avg             s^2\n"
516 		      "  --------                               "
517 		   "---    ----            ---             ---\n");
518 #else
519 	seq_printf(m, "  Function                               Hit\n"
520 		      "  --------                               ---\n");
521 #endif
522 	return 0;
523 }
524 
525 static int function_stat_show(struct seq_file *m, void *v)
526 {
527 	struct ftrace_profile *rec = v;
528 	char str[KSYM_SYMBOL_LEN];
529 	int ret = 0;
530 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
531 	static struct trace_seq s;
532 	unsigned long long avg;
533 	unsigned long long stddev;
534 #endif
535 	mutex_lock(&ftrace_profile_lock);
536 
537 	/* we raced with function_profile_reset() */
538 	if (unlikely(rec->counter == 0)) {
539 		ret = -EBUSY;
540 		goto out;
541 	}
542 
543 	kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
544 	seq_printf(m, "  %-30.30s  %10lu", str, rec->counter);
545 
546 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
547 	seq_printf(m, "    ");
548 	avg = rec->time;
549 	do_div(avg, rec->counter);
550 
551 	/* Sample standard deviation (s^2) */
552 	if (rec->counter <= 1)
553 		stddev = 0;
554 	else {
555 		stddev = rec->time_squared - rec->counter * avg * avg;
556 		/*
557 		 * Divide only 1000 for ns^2 -> us^2 conversion.
558 		 * trace_print_graph_duration will divide 1000 again.
559 		 */
560 		do_div(stddev, (rec->counter - 1) * 1000);
561 	}
562 
563 	trace_seq_init(&s);
564 	trace_print_graph_duration(rec->time, &s);
565 	trace_seq_puts(&s, "    ");
566 	trace_print_graph_duration(avg, &s);
567 	trace_seq_puts(&s, "    ");
568 	trace_print_graph_duration(stddev, &s);
569 	trace_print_seq(m, &s);
570 #endif
571 	seq_putc(m, '\n');
572 out:
573 	mutex_unlock(&ftrace_profile_lock);
574 
575 	return ret;
576 }
577 
578 static void ftrace_profile_reset(struct ftrace_profile_stat *stat)
579 {
580 	struct ftrace_profile_page *pg;
581 
582 	pg = stat->pages = stat->start;
583 
584 	while (pg) {
585 		memset(pg->records, 0, PROFILE_RECORDS_SIZE);
586 		pg->index = 0;
587 		pg = pg->next;
588 	}
589 
590 	memset(stat->hash, 0,
591 	       FTRACE_PROFILE_HASH_SIZE * sizeof(struct hlist_head));
592 }
593 
594 int ftrace_profile_pages_init(struct ftrace_profile_stat *stat)
595 {
596 	struct ftrace_profile_page *pg;
597 	int functions;
598 	int pages;
599 	int i;
600 
601 	/* If we already allocated, do nothing */
602 	if (stat->pages)
603 		return 0;
604 
605 	stat->pages = (void *)get_zeroed_page(GFP_KERNEL);
606 	if (!stat->pages)
607 		return -ENOMEM;
608 
609 #ifdef CONFIG_DYNAMIC_FTRACE
610 	functions = ftrace_update_tot_cnt;
611 #else
612 	/*
613 	 * We do not know the number of functions that exist because
614 	 * dynamic tracing is what counts them. With past experience
615 	 * we have around 20K functions. That should be more than enough.
616 	 * It is highly unlikely we will execute every function in
617 	 * the kernel.
618 	 */
619 	functions = 20000;
620 #endif
621 
622 	pg = stat->start = stat->pages;
623 
624 	pages = DIV_ROUND_UP(functions, PROFILES_PER_PAGE);
625 
626 	for (i = 0; i < pages; i++) {
627 		pg->next = (void *)get_zeroed_page(GFP_KERNEL);
628 		if (!pg->next)
629 			goto out_free;
630 		pg = pg->next;
631 	}
632 
633 	return 0;
634 
635  out_free:
636 	pg = stat->start;
637 	while (pg) {
638 		unsigned long tmp = (unsigned long)pg;
639 
640 		pg = pg->next;
641 		free_page(tmp);
642 	}
643 
644 	free_page((unsigned long)stat->pages);
645 	stat->pages = NULL;
646 	stat->start = NULL;
647 
648 	return -ENOMEM;
649 }
650 
651 static int ftrace_profile_init_cpu(int cpu)
652 {
653 	struct ftrace_profile_stat *stat;
654 	int size;
655 
656 	stat = &per_cpu(ftrace_profile_stats, cpu);
657 
658 	if (stat->hash) {
659 		/* If the profile is already created, simply reset it */
660 		ftrace_profile_reset(stat);
661 		return 0;
662 	}
663 
664 	/*
665 	 * We are profiling all functions, but usually only a few thousand
666 	 * functions are hit. We'll make a hash of 1024 items.
667 	 */
668 	size = FTRACE_PROFILE_HASH_SIZE;
669 
670 	stat->hash = kzalloc(sizeof(struct hlist_head) * size, GFP_KERNEL);
671 
672 	if (!stat->hash)
673 		return -ENOMEM;
674 
675 	if (!ftrace_profile_bits) {
676 		size--;
677 
678 		for (; size; size >>= 1)
679 			ftrace_profile_bits++;
680 	}
681 
682 	/* Preallocate the function profiling pages */
683 	if (ftrace_profile_pages_init(stat) < 0) {
684 		kfree(stat->hash);
685 		stat->hash = NULL;
686 		return -ENOMEM;
687 	}
688 
689 	return 0;
690 }
691 
692 static int ftrace_profile_init(void)
693 {
694 	int cpu;
695 	int ret = 0;
696 
697 	for_each_online_cpu(cpu) {
698 		ret = ftrace_profile_init_cpu(cpu);
699 		if (ret)
700 			break;
701 	}
702 
703 	return ret;
704 }
705 
706 /* interrupts must be disabled */
707 static struct ftrace_profile *
708 ftrace_find_profiled_func(struct ftrace_profile_stat *stat, unsigned long ip)
709 {
710 	struct ftrace_profile *rec;
711 	struct hlist_head *hhd;
712 	struct hlist_node *n;
713 	unsigned long key;
714 
715 	key = hash_long(ip, ftrace_profile_bits);
716 	hhd = &stat->hash[key];
717 
718 	if (hlist_empty(hhd))
719 		return NULL;
720 
721 	hlist_for_each_entry_rcu(rec, n, hhd, node) {
722 		if (rec->ip == ip)
723 			return rec;
724 	}
725 
726 	return NULL;
727 }
728 
729 static void ftrace_add_profile(struct ftrace_profile_stat *stat,
730 			       struct ftrace_profile *rec)
731 {
732 	unsigned long key;
733 
734 	key = hash_long(rec->ip, ftrace_profile_bits);
735 	hlist_add_head_rcu(&rec->node, &stat->hash[key]);
736 }
737 
738 /*
739  * The memory is already allocated, this simply finds a new record to use.
740  */
741 static struct ftrace_profile *
742 ftrace_profile_alloc(struct ftrace_profile_stat *stat, unsigned long ip)
743 {
744 	struct ftrace_profile *rec = NULL;
745 
746 	/* prevent recursion (from NMIs) */
747 	if (atomic_inc_return(&stat->disabled) != 1)
748 		goto out;
749 
750 	/*
751 	 * Try to find the function again since an NMI
752 	 * could have added it
753 	 */
754 	rec = ftrace_find_profiled_func(stat, ip);
755 	if (rec)
756 		goto out;
757 
758 	if (stat->pages->index == PROFILES_PER_PAGE) {
759 		if (!stat->pages->next)
760 			goto out;
761 		stat->pages = stat->pages->next;
762 	}
763 
764 	rec = &stat->pages->records[stat->pages->index++];
765 	rec->ip = ip;
766 	ftrace_add_profile(stat, rec);
767 
768  out:
769 	atomic_dec(&stat->disabled);
770 
771 	return rec;
772 }
773 
774 static void
775 function_profile_call(unsigned long ip, unsigned long parent_ip)
776 {
777 	struct ftrace_profile_stat *stat;
778 	struct ftrace_profile *rec;
779 	unsigned long flags;
780 
781 	if (!ftrace_profile_enabled)
782 		return;
783 
784 	local_irq_save(flags);
785 
786 	stat = &__get_cpu_var(ftrace_profile_stats);
787 	if (!stat->hash || !ftrace_profile_enabled)
788 		goto out;
789 
790 	rec = ftrace_find_profiled_func(stat, ip);
791 	if (!rec) {
792 		rec = ftrace_profile_alloc(stat, ip);
793 		if (!rec)
794 			goto out;
795 	}
796 
797 	rec->counter++;
798  out:
799 	local_irq_restore(flags);
800 }
801 
802 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
803 static int profile_graph_entry(struct ftrace_graph_ent *trace)
804 {
805 	function_profile_call(trace->func, 0);
806 	return 1;
807 }
808 
809 static void profile_graph_return(struct ftrace_graph_ret *trace)
810 {
811 	struct ftrace_profile_stat *stat;
812 	unsigned long long calltime;
813 	struct ftrace_profile *rec;
814 	unsigned long flags;
815 
816 	local_irq_save(flags);
817 	stat = &__get_cpu_var(ftrace_profile_stats);
818 	if (!stat->hash || !ftrace_profile_enabled)
819 		goto out;
820 
821 	/* If the calltime was zero'd ignore it */
822 	if (!trace->calltime)
823 		goto out;
824 
825 	calltime = trace->rettime - trace->calltime;
826 
827 	if (!(trace_flags & TRACE_ITER_GRAPH_TIME)) {
828 		int index;
829 
830 		index = trace->depth;
831 
832 		/* Append this call time to the parent time to subtract */
833 		if (index)
834 			current->ret_stack[index - 1].subtime += calltime;
835 
836 		if (current->ret_stack[index].subtime < calltime)
837 			calltime -= current->ret_stack[index].subtime;
838 		else
839 			calltime = 0;
840 	}
841 
842 	rec = ftrace_find_profiled_func(stat, trace->func);
843 	if (rec) {
844 		rec->time += calltime;
845 		rec->time_squared += calltime * calltime;
846 	}
847 
848  out:
849 	local_irq_restore(flags);
850 }
851 
852 static int register_ftrace_profiler(void)
853 {
854 	return register_ftrace_graph(&profile_graph_return,
855 				     &profile_graph_entry);
856 }
857 
858 static void unregister_ftrace_profiler(void)
859 {
860 	unregister_ftrace_graph();
861 }
862 #else
863 static struct ftrace_ops ftrace_profile_ops __read_mostly = {
864 	.func		= function_profile_call,
865 };
866 
867 static int register_ftrace_profiler(void)
868 {
869 	return register_ftrace_function(&ftrace_profile_ops);
870 }
871 
872 static void unregister_ftrace_profiler(void)
873 {
874 	unregister_ftrace_function(&ftrace_profile_ops);
875 }
876 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
877 
878 static ssize_t
879 ftrace_profile_write(struct file *filp, const char __user *ubuf,
880 		     size_t cnt, loff_t *ppos)
881 {
882 	unsigned long val;
883 	int ret;
884 
885 	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
886 	if (ret)
887 		return ret;
888 
889 	val = !!val;
890 
891 	mutex_lock(&ftrace_profile_lock);
892 	if (ftrace_profile_enabled ^ val) {
893 		if (val) {
894 			ret = ftrace_profile_init();
895 			if (ret < 0) {
896 				cnt = ret;
897 				goto out;
898 			}
899 
900 			ret = register_ftrace_profiler();
901 			if (ret < 0) {
902 				cnt = ret;
903 				goto out;
904 			}
905 			ftrace_profile_enabled = 1;
906 		} else {
907 			ftrace_profile_enabled = 0;
908 			/*
909 			 * unregister_ftrace_profiler calls stop_machine
910 			 * so this acts like an synchronize_sched.
911 			 */
912 			unregister_ftrace_profiler();
913 		}
914 	}
915  out:
916 	mutex_unlock(&ftrace_profile_lock);
917 
918 	*ppos += cnt;
919 
920 	return cnt;
921 }
922 
923 static ssize_t
924 ftrace_profile_read(struct file *filp, char __user *ubuf,
925 		     size_t cnt, loff_t *ppos)
926 {
927 	char buf[64];		/* big enough to hold a number */
928 	int r;
929 
930 	r = sprintf(buf, "%u\n", ftrace_profile_enabled);
931 	return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
932 }
933 
934 static const struct file_operations ftrace_profile_fops = {
935 	.open		= tracing_open_generic,
936 	.read		= ftrace_profile_read,
937 	.write		= ftrace_profile_write,
938 	.llseek		= default_llseek,
939 };
940 
941 /* used to initialize the real stat files */
942 static struct tracer_stat function_stats __initdata = {
943 	.name		= "functions",
944 	.stat_start	= function_stat_start,
945 	.stat_next	= function_stat_next,
946 	.stat_cmp	= function_stat_cmp,
947 	.stat_headers	= function_stat_headers,
948 	.stat_show	= function_stat_show
949 };
950 
951 static __init void ftrace_profile_debugfs(struct dentry *d_tracer)
952 {
953 	struct ftrace_profile_stat *stat;
954 	struct dentry *entry;
955 	char *name;
956 	int ret;
957 	int cpu;
958 
959 	for_each_possible_cpu(cpu) {
960 		stat = &per_cpu(ftrace_profile_stats, cpu);
961 
962 		/* allocate enough for function name + cpu number */
963 		name = kmalloc(32, GFP_KERNEL);
964 		if (!name) {
965 			/*
966 			 * The files created are permanent, if something happens
967 			 * we still do not free memory.
968 			 */
969 			WARN(1,
970 			     "Could not allocate stat file for cpu %d\n",
971 			     cpu);
972 			return;
973 		}
974 		stat->stat = function_stats;
975 		snprintf(name, 32, "function%d", cpu);
976 		stat->stat.name = name;
977 		ret = register_stat_tracer(&stat->stat);
978 		if (ret) {
979 			WARN(1,
980 			     "Could not register function stat for cpu %d\n",
981 			     cpu);
982 			kfree(name);
983 			return;
984 		}
985 	}
986 
987 	entry = debugfs_create_file("function_profile_enabled", 0644,
988 				    d_tracer, NULL, &ftrace_profile_fops);
989 	if (!entry)
990 		pr_warning("Could not create debugfs "
991 			   "'function_profile_enabled' entry\n");
992 }
993 
994 #else /* CONFIG_FUNCTION_PROFILER */
995 static __init void ftrace_profile_debugfs(struct dentry *d_tracer)
996 {
997 }
998 #endif /* CONFIG_FUNCTION_PROFILER */
999 
1000 static struct pid * const ftrace_swapper_pid = &init_struct_pid;
1001 
1002 #ifdef CONFIG_DYNAMIC_FTRACE
1003 
1004 #ifndef CONFIG_FTRACE_MCOUNT_RECORD
1005 # error Dynamic ftrace depends on MCOUNT_RECORD
1006 #endif
1007 
1008 static struct hlist_head ftrace_func_hash[FTRACE_FUNC_HASHSIZE] __read_mostly;
1009 
1010 struct ftrace_func_probe {
1011 	struct hlist_node	node;
1012 	struct ftrace_probe_ops	*ops;
1013 	unsigned long		flags;
1014 	unsigned long		ip;
1015 	void			*data;
1016 	struct rcu_head		rcu;
1017 };
1018 
1019 struct ftrace_func_entry {
1020 	struct hlist_node hlist;
1021 	unsigned long ip;
1022 };
1023 
1024 struct ftrace_hash {
1025 	unsigned long		size_bits;
1026 	struct hlist_head	*buckets;
1027 	unsigned long		count;
1028 	struct rcu_head		rcu;
1029 };
1030 
1031 /*
1032  * We make these constant because no one should touch them,
1033  * but they are used as the default "empty hash", to avoid allocating
1034  * it all the time. These are in a read only section such that if
1035  * anyone does try to modify it, it will cause an exception.
1036  */
1037 static const struct hlist_head empty_buckets[1];
1038 static const struct ftrace_hash empty_hash = {
1039 	.buckets = (struct hlist_head *)empty_buckets,
1040 };
1041 #define EMPTY_HASH	((struct ftrace_hash *)&empty_hash)
1042 
1043 static struct ftrace_ops global_ops = {
1044 	.func			= ftrace_stub,
1045 	.notrace_hash		= EMPTY_HASH,
1046 	.filter_hash		= EMPTY_HASH,
1047 };
1048 
1049 static DEFINE_MUTEX(ftrace_regex_lock);
1050 
1051 struct ftrace_page {
1052 	struct ftrace_page	*next;
1053 	struct dyn_ftrace	*records;
1054 	int			index;
1055 	int			size;
1056 };
1057 
1058 static struct ftrace_page *ftrace_new_pgs;
1059 
1060 #define ENTRY_SIZE sizeof(struct dyn_ftrace)
1061 #define ENTRIES_PER_PAGE (PAGE_SIZE / ENTRY_SIZE)
1062 
1063 /* estimate from running different kernels */
1064 #define NR_TO_INIT		10000
1065 
1066 static struct ftrace_page	*ftrace_pages_start;
1067 static struct ftrace_page	*ftrace_pages;
1068 
1069 static bool ftrace_hash_empty(struct ftrace_hash *hash)
1070 {
1071 	return !hash || !hash->count;
1072 }
1073 
1074 static struct ftrace_func_entry *
1075 ftrace_lookup_ip(struct ftrace_hash *hash, unsigned long ip)
1076 {
1077 	unsigned long key;
1078 	struct ftrace_func_entry *entry;
1079 	struct hlist_head *hhd;
1080 	struct hlist_node *n;
1081 
1082 	if (ftrace_hash_empty(hash))
1083 		return NULL;
1084 
1085 	if (hash->size_bits > 0)
1086 		key = hash_long(ip, hash->size_bits);
1087 	else
1088 		key = 0;
1089 
1090 	hhd = &hash->buckets[key];
1091 
1092 	hlist_for_each_entry_rcu(entry, n, hhd, hlist) {
1093 		if (entry->ip == ip)
1094 			return entry;
1095 	}
1096 	return NULL;
1097 }
1098 
1099 static void __add_hash_entry(struct ftrace_hash *hash,
1100 			     struct ftrace_func_entry *entry)
1101 {
1102 	struct hlist_head *hhd;
1103 	unsigned long key;
1104 
1105 	if (hash->size_bits)
1106 		key = hash_long(entry->ip, hash->size_bits);
1107 	else
1108 		key = 0;
1109 
1110 	hhd = &hash->buckets[key];
1111 	hlist_add_head(&entry->hlist, hhd);
1112 	hash->count++;
1113 }
1114 
1115 static int add_hash_entry(struct ftrace_hash *hash, unsigned long ip)
1116 {
1117 	struct ftrace_func_entry *entry;
1118 
1119 	entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1120 	if (!entry)
1121 		return -ENOMEM;
1122 
1123 	entry->ip = ip;
1124 	__add_hash_entry(hash, entry);
1125 
1126 	return 0;
1127 }
1128 
1129 static void
1130 free_hash_entry(struct ftrace_hash *hash,
1131 		  struct ftrace_func_entry *entry)
1132 {
1133 	hlist_del(&entry->hlist);
1134 	kfree(entry);
1135 	hash->count--;
1136 }
1137 
1138 static void
1139 remove_hash_entry(struct ftrace_hash *hash,
1140 		  struct ftrace_func_entry *entry)
1141 {
1142 	hlist_del(&entry->hlist);
1143 	hash->count--;
1144 }
1145 
1146 static void ftrace_hash_clear(struct ftrace_hash *hash)
1147 {
1148 	struct hlist_head *hhd;
1149 	struct hlist_node *tp, *tn;
1150 	struct ftrace_func_entry *entry;
1151 	int size = 1 << hash->size_bits;
1152 	int i;
1153 
1154 	if (!hash->count)
1155 		return;
1156 
1157 	for (i = 0; i < size; i++) {
1158 		hhd = &hash->buckets[i];
1159 		hlist_for_each_entry_safe(entry, tp, tn, hhd, hlist)
1160 			free_hash_entry(hash, entry);
1161 	}
1162 	FTRACE_WARN_ON(hash->count);
1163 }
1164 
1165 static void free_ftrace_hash(struct ftrace_hash *hash)
1166 {
1167 	if (!hash || hash == EMPTY_HASH)
1168 		return;
1169 	ftrace_hash_clear(hash);
1170 	kfree(hash->buckets);
1171 	kfree(hash);
1172 }
1173 
1174 static void __free_ftrace_hash_rcu(struct rcu_head *rcu)
1175 {
1176 	struct ftrace_hash *hash;
1177 
1178 	hash = container_of(rcu, struct ftrace_hash, rcu);
1179 	free_ftrace_hash(hash);
1180 }
1181 
1182 static void free_ftrace_hash_rcu(struct ftrace_hash *hash)
1183 {
1184 	if (!hash || hash == EMPTY_HASH)
1185 		return;
1186 	call_rcu_sched(&hash->rcu, __free_ftrace_hash_rcu);
1187 }
1188 
1189 void ftrace_free_filter(struct ftrace_ops *ops)
1190 {
1191 	free_ftrace_hash(ops->filter_hash);
1192 	free_ftrace_hash(ops->notrace_hash);
1193 }
1194 
1195 static struct ftrace_hash *alloc_ftrace_hash(int size_bits)
1196 {
1197 	struct ftrace_hash *hash;
1198 	int size;
1199 
1200 	hash = kzalloc(sizeof(*hash), GFP_KERNEL);
1201 	if (!hash)
1202 		return NULL;
1203 
1204 	size = 1 << size_bits;
1205 	hash->buckets = kcalloc(size, sizeof(*hash->buckets), GFP_KERNEL);
1206 
1207 	if (!hash->buckets) {
1208 		kfree(hash);
1209 		return NULL;
1210 	}
1211 
1212 	hash->size_bits = size_bits;
1213 
1214 	return hash;
1215 }
1216 
1217 static struct ftrace_hash *
1218 alloc_and_copy_ftrace_hash(int size_bits, struct ftrace_hash *hash)
1219 {
1220 	struct ftrace_func_entry *entry;
1221 	struct ftrace_hash *new_hash;
1222 	struct hlist_node *tp;
1223 	int size;
1224 	int ret;
1225 	int i;
1226 
1227 	new_hash = alloc_ftrace_hash(size_bits);
1228 	if (!new_hash)
1229 		return NULL;
1230 
1231 	/* Empty hash? */
1232 	if (ftrace_hash_empty(hash))
1233 		return new_hash;
1234 
1235 	size = 1 << hash->size_bits;
1236 	for (i = 0; i < size; i++) {
1237 		hlist_for_each_entry(entry, tp, &hash->buckets[i], hlist) {
1238 			ret = add_hash_entry(new_hash, entry->ip);
1239 			if (ret < 0)
1240 				goto free_hash;
1241 		}
1242 	}
1243 
1244 	FTRACE_WARN_ON(new_hash->count != hash->count);
1245 
1246 	return new_hash;
1247 
1248  free_hash:
1249 	free_ftrace_hash(new_hash);
1250 	return NULL;
1251 }
1252 
1253 static void
1254 ftrace_hash_rec_disable(struct ftrace_ops *ops, int filter_hash);
1255 static void
1256 ftrace_hash_rec_enable(struct ftrace_ops *ops, int filter_hash);
1257 
1258 static int
1259 ftrace_hash_move(struct ftrace_ops *ops, int enable,
1260 		 struct ftrace_hash **dst, struct ftrace_hash *src)
1261 {
1262 	struct ftrace_func_entry *entry;
1263 	struct hlist_node *tp, *tn;
1264 	struct hlist_head *hhd;
1265 	struct ftrace_hash *old_hash;
1266 	struct ftrace_hash *new_hash;
1267 	unsigned long key;
1268 	int size = src->count;
1269 	int bits = 0;
1270 	int ret;
1271 	int i;
1272 
1273 	/*
1274 	 * Remove the current set, update the hash and add
1275 	 * them back.
1276 	 */
1277 	ftrace_hash_rec_disable(ops, enable);
1278 
1279 	/*
1280 	 * If the new source is empty, just free dst and assign it
1281 	 * the empty_hash.
1282 	 */
1283 	if (!src->count) {
1284 		free_ftrace_hash_rcu(*dst);
1285 		rcu_assign_pointer(*dst, EMPTY_HASH);
1286 		/* still need to update the function records */
1287 		ret = 0;
1288 		goto out;
1289 	}
1290 
1291 	/*
1292 	 * Make the hash size about 1/2 the # found
1293 	 */
1294 	for (size /= 2; size; size >>= 1)
1295 		bits++;
1296 
1297 	/* Don't allocate too much */
1298 	if (bits > FTRACE_HASH_MAX_BITS)
1299 		bits = FTRACE_HASH_MAX_BITS;
1300 
1301 	ret = -ENOMEM;
1302 	new_hash = alloc_ftrace_hash(bits);
1303 	if (!new_hash)
1304 		goto out;
1305 
1306 	size = 1 << src->size_bits;
1307 	for (i = 0; i < size; i++) {
1308 		hhd = &src->buckets[i];
1309 		hlist_for_each_entry_safe(entry, tp, tn, hhd, hlist) {
1310 			if (bits > 0)
1311 				key = hash_long(entry->ip, bits);
1312 			else
1313 				key = 0;
1314 			remove_hash_entry(src, entry);
1315 			__add_hash_entry(new_hash, entry);
1316 		}
1317 	}
1318 
1319 	old_hash = *dst;
1320 	rcu_assign_pointer(*dst, new_hash);
1321 	free_ftrace_hash_rcu(old_hash);
1322 
1323 	ret = 0;
1324  out:
1325 	/*
1326 	 * Enable regardless of ret:
1327 	 *  On success, we enable the new hash.
1328 	 *  On failure, we re-enable the original hash.
1329 	 */
1330 	ftrace_hash_rec_enable(ops, enable);
1331 
1332 	return ret;
1333 }
1334 
1335 /*
1336  * Test the hashes for this ops to see if we want to call
1337  * the ops->func or not.
1338  *
1339  * It's a match if the ip is in the ops->filter_hash or
1340  * the filter_hash does not exist or is empty,
1341  *  AND
1342  * the ip is not in the ops->notrace_hash.
1343  *
1344  * This needs to be called with preemption disabled as
1345  * the hashes are freed with call_rcu_sched().
1346  */
1347 static int
1348 ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip)
1349 {
1350 	struct ftrace_hash *filter_hash;
1351 	struct ftrace_hash *notrace_hash;
1352 	int ret;
1353 
1354 	filter_hash = rcu_dereference_raw(ops->filter_hash);
1355 	notrace_hash = rcu_dereference_raw(ops->notrace_hash);
1356 
1357 	if ((ftrace_hash_empty(filter_hash) ||
1358 	     ftrace_lookup_ip(filter_hash, ip)) &&
1359 	    (ftrace_hash_empty(notrace_hash) ||
1360 	     !ftrace_lookup_ip(notrace_hash, ip)))
1361 		ret = 1;
1362 	else
1363 		ret = 0;
1364 
1365 	return ret;
1366 }
1367 
1368 /*
1369  * This is a double for. Do not use 'break' to break out of the loop,
1370  * you must use a goto.
1371  */
1372 #define do_for_each_ftrace_rec(pg, rec)					\
1373 	for (pg = ftrace_pages_start; pg; pg = pg->next) {		\
1374 		int _____i;						\
1375 		for (_____i = 0; _____i < pg->index; _____i++) {	\
1376 			rec = &pg->records[_____i];
1377 
1378 #define while_for_each_ftrace_rec()		\
1379 		}				\
1380 	}
1381 
1382 
1383 static int ftrace_cmp_recs(const void *a, const void *b)
1384 {
1385 	const struct dyn_ftrace *reca = a;
1386 	const struct dyn_ftrace *recb = b;
1387 
1388 	if (reca->ip > recb->ip)
1389 		return 1;
1390 	if (reca->ip < recb->ip)
1391 		return -1;
1392 	return 0;
1393 }
1394 
1395 /**
1396  * ftrace_location - return true if the ip giving is a traced location
1397  * @ip: the instruction pointer to check
1398  *
1399  * Returns 1 if @ip given is a pointer to a ftrace location.
1400  * That is, the instruction that is either a NOP or call to
1401  * the function tracer. It checks the ftrace internal tables to
1402  * determine if the address belongs or not.
1403  */
1404 int ftrace_location(unsigned long ip)
1405 {
1406 	struct ftrace_page *pg;
1407 	struct dyn_ftrace *rec;
1408 	struct dyn_ftrace key;
1409 
1410 	key.ip = ip;
1411 
1412 	for (pg = ftrace_pages_start; pg; pg = pg->next) {
1413 		rec = bsearch(&key, pg->records, pg->index,
1414 			      sizeof(struct dyn_ftrace),
1415 			      ftrace_cmp_recs);
1416 		if (rec)
1417 			return 1;
1418 	}
1419 
1420 	return 0;
1421 }
1422 
1423 static void __ftrace_hash_rec_update(struct ftrace_ops *ops,
1424 				     int filter_hash,
1425 				     bool inc)
1426 {
1427 	struct ftrace_hash *hash;
1428 	struct ftrace_hash *other_hash;
1429 	struct ftrace_page *pg;
1430 	struct dyn_ftrace *rec;
1431 	int count = 0;
1432 	int all = 0;
1433 
1434 	/* Only update if the ops has been registered */
1435 	if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1436 		return;
1437 
1438 	/*
1439 	 * In the filter_hash case:
1440 	 *   If the count is zero, we update all records.
1441 	 *   Otherwise we just update the items in the hash.
1442 	 *
1443 	 * In the notrace_hash case:
1444 	 *   We enable the update in the hash.
1445 	 *   As disabling notrace means enabling the tracing,
1446 	 *   and enabling notrace means disabling, the inc variable
1447 	 *   gets inversed.
1448 	 */
1449 	if (filter_hash) {
1450 		hash = ops->filter_hash;
1451 		other_hash = ops->notrace_hash;
1452 		if (ftrace_hash_empty(hash))
1453 			all = 1;
1454 	} else {
1455 		inc = !inc;
1456 		hash = ops->notrace_hash;
1457 		other_hash = ops->filter_hash;
1458 		/*
1459 		 * If the notrace hash has no items,
1460 		 * then there's nothing to do.
1461 		 */
1462 		if (ftrace_hash_empty(hash))
1463 			return;
1464 	}
1465 
1466 	do_for_each_ftrace_rec(pg, rec) {
1467 		int in_other_hash = 0;
1468 		int in_hash = 0;
1469 		int match = 0;
1470 
1471 		if (all) {
1472 			/*
1473 			 * Only the filter_hash affects all records.
1474 			 * Update if the record is not in the notrace hash.
1475 			 */
1476 			if (!other_hash || !ftrace_lookup_ip(other_hash, rec->ip))
1477 				match = 1;
1478 		} else {
1479 			in_hash = !!ftrace_lookup_ip(hash, rec->ip);
1480 			in_other_hash = !!ftrace_lookup_ip(other_hash, rec->ip);
1481 
1482 			/*
1483 			 *
1484 			 */
1485 			if (filter_hash && in_hash && !in_other_hash)
1486 				match = 1;
1487 			else if (!filter_hash && in_hash &&
1488 				 (in_other_hash || ftrace_hash_empty(other_hash)))
1489 				match = 1;
1490 		}
1491 		if (!match)
1492 			continue;
1493 
1494 		if (inc) {
1495 			rec->flags++;
1496 			if (FTRACE_WARN_ON((rec->flags & ~FTRACE_FL_MASK) == FTRACE_REF_MAX))
1497 				return;
1498 		} else {
1499 			if (FTRACE_WARN_ON((rec->flags & ~FTRACE_FL_MASK) == 0))
1500 				return;
1501 			rec->flags--;
1502 		}
1503 		count++;
1504 		/* Shortcut, if we handled all records, we are done. */
1505 		if (!all && count == hash->count)
1506 			return;
1507 	} while_for_each_ftrace_rec();
1508 }
1509 
1510 static void ftrace_hash_rec_disable(struct ftrace_ops *ops,
1511 				    int filter_hash)
1512 {
1513 	__ftrace_hash_rec_update(ops, filter_hash, 0);
1514 }
1515 
1516 static void ftrace_hash_rec_enable(struct ftrace_ops *ops,
1517 				   int filter_hash)
1518 {
1519 	__ftrace_hash_rec_update(ops, filter_hash, 1);
1520 }
1521 
1522 static struct dyn_ftrace *ftrace_alloc_dyn_node(unsigned long ip)
1523 {
1524 	if (ftrace_pages->index == ftrace_pages->size) {
1525 		/* We should have allocated enough */
1526 		if (WARN_ON(!ftrace_pages->next))
1527 			return NULL;
1528 		ftrace_pages = ftrace_pages->next;
1529 	}
1530 
1531 	return &ftrace_pages->records[ftrace_pages->index++];
1532 }
1533 
1534 static struct dyn_ftrace *
1535 ftrace_record_ip(unsigned long ip)
1536 {
1537 	struct dyn_ftrace *rec;
1538 
1539 	if (ftrace_disabled)
1540 		return NULL;
1541 
1542 	rec = ftrace_alloc_dyn_node(ip);
1543 	if (!rec)
1544 		return NULL;
1545 
1546 	rec->ip = ip;
1547 
1548 	return rec;
1549 }
1550 
1551 static void print_ip_ins(const char *fmt, unsigned char *p)
1552 {
1553 	int i;
1554 
1555 	printk(KERN_CONT "%s", fmt);
1556 
1557 	for (i = 0; i < MCOUNT_INSN_SIZE; i++)
1558 		printk(KERN_CONT "%s%02x", i ? ":" : "", p[i]);
1559 }
1560 
1561 /**
1562  * ftrace_bug - report and shutdown function tracer
1563  * @failed: The failed type (EFAULT, EINVAL, EPERM)
1564  * @ip: The address that failed
1565  *
1566  * The arch code that enables or disables the function tracing
1567  * can call ftrace_bug() when it has detected a problem in
1568  * modifying the code. @failed should be one of either:
1569  * EFAULT - if the problem happens on reading the @ip address
1570  * EINVAL - if what is read at @ip is not what was expected
1571  * EPERM - if the problem happens on writting to the @ip address
1572  */
1573 void ftrace_bug(int failed, unsigned long ip)
1574 {
1575 	switch (failed) {
1576 	case -EFAULT:
1577 		FTRACE_WARN_ON_ONCE(1);
1578 		pr_info("ftrace faulted on modifying ");
1579 		print_ip_sym(ip);
1580 		break;
1581 	case -EINVAL:
1582 		FTRACE_WARN_ON_ONCE(1);
1583 		pr_info("ftrace failed to modify ");
1584 		print_ip_sym(ip);
1585 		print_ip_ins(" actual: ", (unsigned char *)ip);
1586 		printk(KERN_CONT "\n");
1587 		break;
1588 	case -EPERM:
1589 		FTRACE_WARN_ON_ONCE(1);
1590 		pr_info("ftrace faulted on writing ");
1591 		print_ip_sym(ip);
1592 		break;
1593 	default:
1594 		FTRACE_WARN_ON_ONCE(1);
1595 		pr_info("ftrace faulted on unknown error ");
1596 		print_ip_sym(ip);
1597 	}
1598 }
1599 
1600 
1601 /* Return 1 if the address range is reserved for ftrace */
1602 int ftrace_text_reserved(void *start, void *end)
1603 {
1604 	struct dyn_ftrace *rec;
1605 	struct ftrace_page *pg;
1606 
1607 	do_for_each_ftrace_rec(pg, rec) {
1608 		if (rec->ip <= (unsigned long)end &&
1609 		    rec->ip + MCOUNT_INSN_SIZE > (unsigned long)start)
1610 			return 1;
1611 	} while_for_each_ftrace_rec();
1612 	return 0;
1613 }
1614 
1615 static int ftrace_check_record(struct dyn_ftrace *rec, int enable, int update)
1616 {
1617 	unsigned long flag = 0UL;
1618 
1619 	/*
1620 	 * If we are updating calls:
1621 	 *
1622 	 *   If the record has a ref count, then we need to enable it
1623 	 *   because someone is using it.
1624 	 *
1625 	 *   Otherwise we make sure its disabled.
1626 	 *
1627 	 * If we are disabling calls, then disable all records that
1628 	 * are enabled.
1629 	 */
1630 	if (enable && (rec->flags & ~FTRACE_FL_MASK))
1631 		flag = FTRACE_FL_ENABLED;
1632 
1633 	/* If the state of this record hasn't changed, then do nothing */
1634 	if ((rec->flags & FTRACE_FL_ENABLED) == flag)
1635 		return FTRACE_UPDATE_IGNORE;
1636 
1637 	if (flag) {
1638 		if (update)
1639 			rec->flags |= FTRACE_FL_ENABLED;
1640 		return FTRACE_UPDATE_MAKE_CALL;
1641 	}
1642 
1643 	if (update)
1644 		rec->flags &= ~FTRACE_FL_ENABLED;
1645 
1646 	return FTRACE_UPDATE_MAKE_NOP;
1647 }
1648 
1649 /**
1650  * ftrace_update_record, set a record that now is tracing or not
1651  * @rec: the record to update
1652  * @enable: set to 1 if the record is tracing, zero to force disable
1653  *
1654  * The records that represent all functions that can be traced need
1655  * to be updated when tracing has been enabled.
1656  */
1657 int ftrace_update_record(struct dyn_ftrace *rec, int enable)
1658 {
1659 	return ftrace_check_record(rec, enable, 1);
1660 }
1661 
1662 /**
1663  * ftrace_test_record, check if the record has been enabled or not
1664  * @rec: the record to test
1665  * @enable: set to 1 to check if enabled, 0 if it is disabled
1666  *
1667  * The arch code may need to test if a record is already set to
1668  * tracing to determine how to modify the function code that it
1669  * represents.
1670  */
1671 int ftrace_test_record(struct dyn_ftrace *rec, int enable)
1672 {
1673 	return ftrace_check_record(rec, enable, 0);
1674 }
1675 
1676 static int
1677 __ftrace_replace_code(struct dyn_ftrace *rec, int enable)
1678 {
1679 	unsigned long ftrace_addr;
1680 	int ret;
1681 
1682 	ftrace_addr = (unsigned long)FTRACE_ADDR;
1683 
1684 	ret = ftrace_update_record(rec, enable);
1685 
1686 	switch (ret) {
1687 	case FTRACE_UPDATE_IGNORE:
1688 		return 0;
1689 
1690 	case FTRACE_UPDATE_MAKE_CALL:
1691 		return ftrace_make_call(rec, ftrace_addr);
1692 
1693 	case FTRACE_UPDATE_MAKE_NOP:
1694 		return ftrace_make_nop(NULL, rec, ftrace_addr);
1695 	}
1696 
1697 	return -1; /* unknow ftrace bug */
1698 }
1699 
1700 static void ftrace_replace_code(int update)
1701 {
1702 	struct dyn_ftrace *rec;
1703 	struct ftrace_page *pg;
1704 	int failed;
1705 
1706 	if (unlikely(ftrace_disabled))
1707 		return;
1708 
1709 	do_for_each_ftrace_rec(pg, rec) {
1710 		failed = __ftrace_replace_code(rec, update);
1711 		if (failed) {
1712 			ftrace_bug(failed, rec->ip);
1713 			/* Stop processing */
1714 			return;
1715 		}
1716 	} while_for_each_ftrace_rec();
1717 }
1718 
1719 struct ftrace_rec_iter {
1720 	struct ftrace_page	*pg;
1721 	int			index;
1722 };
1723 
1724 /**
1725  * ftrace_rec_iter_start, start up iterating over traced functions
1726  *
1727  * Returns an iterator handle that is used to iterate over all
1728  * the records that represent address locations where functions
1729  * are traced.
1730  *
1731  * May return NULL if no records are available.
1732  */
1733 struct ftrace_rec_iter *ftrace_rec_iter_start(void)
1734 {
1735 	/*
1736 	 * We only use a single iterator.
1737 	 * Protected by the ftrace_lock mutex.
1738 	 */
1739 	static struct ftrace_rec_iter ftrace_rec_iter;
1740 	struct ftrace_rec_iter *iter = &ftrace_rec_iter;
1741 
1742 	iter->pg = ftrace_pages_start;
1743 	iter->index = 0;
1744 
1745 	/* Could have empty pages */
1746 	while (iter->pg && !iter->pg->index)
1747 		iter->pg = iter->pg->next;
1748 
1749 	if (!iter->pg)
1750 		return NULL;
1751 
1752 	return iter;
1753 }
1754 
1755 /**
1756  * ftrace_rec_iter_next, get the next record to process.
1757  * @iter: The handle to the iterator.
1758  *
1759  * Returns the next iterator after the given iterator @iter.
1760  */
1761 struct ftrace_rec_iter *ftrace_rec_iter_next(struct ftrace_rec_iter *iter)
1762 {
1763 	iter->index++;
1764 
1765 	if (iter->index >= iter->pg->index) {
1766 		iter->pg = iter->pg->next;
1767 		iter->index = 0;
1768 
1769 		/* Could have empty pages */
1770 		while (iter->pg && !iter->pg->index)
1771 			iter->pg = iter->pg->next;
1772 	}
1773 
1774 	if (!iter->pg)
1775 		return NULL;
1776 
1777 	return iter;
1778 }
1779 
1780 /**
1781  * ftrace_rec_iter_record, get the record at the iterator location
1782  * @iter: The current iterator location
1783  *
1784  * Returns the record that the current @iter is at.
1785  */
1786 struct dyn_ftrace *ftrace_rec_iter_record(struct ftrace_rec_iter *iter)
1787 {
1788 	return &iter->pg->records[iter->index];
1789 }
1790 
1791 static int
1792 ftrace_code_disable(struct module *mod, struct dyn_ftrace *rec)
1793 {
1794 	unsigned long ip;
1795 	int ret;
1796 
1797 	ip = rec->ip;
1798 
1799 	if (unlikely(ftrace_disabled))
1800 		return 0;
1801 
1802 	ret = ftrace_make_nop(mod, rec, MCOUNT_ADDR);
1803 	if (ret) {
1804 		ftrace_bug(ret, ip);
1805 		return 0;
1806 	}
1807 	return 1;
1808 }
1809 
1810 /*
1811  * archs can override this function if they must do something
1812  * before the modifying code is performed.
1813  */
1814 int __weak ftrace_arch_code_modify_prepare(void)
1815 {
1816 	return 0;
1817 }
1818 
1819 /*
1820  * archs can override this function if they must do something
1821  * after the modifying code is performed.
1822  */
1823 int __weak ftrace_arch_code_modify_post_process(void)
1824 {
1825 	return 0;
1826 }
1827 
1828 static int __ftrace_modify_code(void *data)
1829 {
1830 	int *command = data;
1831 
1832 	if (*command & FTRACE_UPDATE_CALLS)
1833 		ftrace_replace_code(1);
1834 	else if (*command & FTRACE_DISABLE_CALLS)
1835 		ftrace_replace_code(0);
1836 
1837 	if (*command & FTRACE_UPDATE_TRACE_FUNC)
1838 		ftrace_update_ftrace_func(ftrace_trace_function);
1839 
1840 	if (*command & FTRACE_START_FUNC_RET)
1841 		ftrace_enable_ftrace_graph_caller();
1842 	else if (*command & FTRACE_STOP_FUNC_RET)
1843 		ftrace_disable_ftrace_graph_caller();
1844 
1845 	return 0;
1846 }
1847 
1848 /**
1849  * ftrace_run_stop_machine, go back to the stop machine method
1850  * @command: The command to tell ftrace what to do
1851  *
1852  * If an arch needs to fall back to the stop machine method, the
1853  * it can call this function.
1854  */
1855 void ftrace_run_stop_machine(int command)
1856 {
1857 	stop_machine(__ftrace_modify_code, &command, NULL);
1858 }
1859 
1860 /**
1861  * arch_ftrace_update_code, modify the code to trace or not trace
1862  * @command: The command that needs to be done
1863  *
1864  * Archs can override this function if it does not need to
1865  * run stop_machine() to modify code.
1866  */
1867 void __weak arch_ftrace_update_code(int command)
1868 {
1869 	ftrace_run_stop_machine(command);
1870 }
1871 
1872 static void ftrace_run_update_code(int command)
1873 {
1874 	int ret;
1875 
1876 	ret = ftrace_arch_code_modify_prepare();
1877 	FTRACE_WARN_ON(ret);
1878 	if (ret)
1879 		return;
1880 	/*
1881 	 * Do not call function tracer while we update the code.
1882 	 * We are in stop machine.
1883 	 */
1884 	function_trace_stop++;
1885 
1886 	/*
1887 	 * By default we use stop_machine() to modify the code.
1888 	 * But archs can do what ever they want as long as it
1889 	 * is safe. The stop_machine() is the safest, but also
1890 	 * produces the most overhead.
1891 	 */
1892 	arch_ftrace_update_code(command);
1893 
1894 #ifndef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
1895 	/*
1896 	 * For archs that call ftrace_test_stop_func(), we must
1897 	 * wait till after we update all the function callers
1898 	 * before we update the callback. This keeps different
1899 	 * ops that record different functions from corrupting
1900 	 * each other.
1901 	 */
1902 	__ftrace_trace_function = __ftrace_trace_function_delay;
1903 #endif
1904 	function_trace_stop--;
1905 
1906 	ret = ftrace_arch_code_modify_post_process();
1907 	FTRACE_WARN_ON(ret);
1908 }
1909 
1910 static ftrace_func_t saved_ftrace_func;
1911 static int ftrace_start_up;
1912 static int global_start_up;
1913 
1914 static void ftrace_startup_enable(int command)
1915 {
1916 	if (saved_ftrace_func != ftrace_trace_function) {
1917 		saved_ftrace_func = ftrace_trace_function;
1918 		command |= FTRACE_UPDATE_TRACE_FUNC;
1919 	}
1920 
1921 	if (!command || !ftrace_enabled)
1922 		return;
1923 
1924 	ftrace_run_update_code(command);
1925 }
1926 
1927 static int ftrace_startup(struct ftrace_ops *ops, int command)
1928 {
1929 	bool hash_enable = true;
1930 
1931 	if (unlikely(ftrace_disabled))
1932 		return -ENODEV;
1933 
1934 	ftrace_start_up++;
1935 	command |= FTRACE_UPDATE_CALLS;
1936 
1937 	/* ops marked global share the filter hashes */
1938 	if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
1939 		ops = &global_ops;
1940 		/* Don't update hash if global is already set */
1941 		if (global_start_up)
1942 			hash_enable = false;
1943 		global_start_up++;
1944 	}
1945 
1946 	ops->flags |= FTRACE_OPS_FL_ENABLED;
1947 	if (hash_enable)
1948 		ftrace_hash_rec_enable(ops, 1);
1949 
1950 	ftrace_startup_enable(command);
1951 
1952 	return 0;
1953 }
1954 
1955 static void ftrace_shutdown(struct ftrace_ops *ops, int command)
1956 {
1957 	bool hash_disable = true;
1958 
1959 	if (unlikely(ftrace_disabled))
1960 		return;
1961 
1962 	ftrace_start_up--;
1963 	/*
1964 	 * Just warn in case of unbalance, no need to kill ftrace, it's not
1965 	 * critical but the ftrace_call callers may be never nopped again after
1966 	 * further ftrace uses.
1967 	 */
1968 	WARN_ON_ONCE(ftrace_start_up < 0);
1969 
1970 	if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
1971 		ops = &global_ops;
1972 		global_start_up--;
1973 		WARN_ON_ONCE(global_start_up < 0);
1974 		/* Don't update hash if global still has users */
1975 		if (global_start_up) {
1976 			WARN_ON_ONCE(!ftrace_start_up);
1977 			hash_disable = false;
1978 		}
1979 	}
1980 
1981 	if (hash_disable)
1982 		ftrace_hash_rec_disable(ops, 1);
1983 
1984 	if (ops != &global_ops || !global_start_up)
1985 		ops->flags &= ~FTRACE_OPS_FL_ENABLED;
1986 
1987 	command |= FTRACE_UPDATE_CALLS;
1988 
1989 	if (saved_ftrace_func != ftrace_trace_function) {
1990 		saved_ftrace_func = ftrace_trace_function;
1991 		command |= FTRACE_UPDATE_TRACE_FUNC;
1992 	}
1993 
1994 	if (!command || !ftrace_enabled)
1995 		return;
1996 
1997 	ftrace_run_update_code(command);
1998 }
1999 
2000 static void ftrace_startup_sysctl(void)
2001 {
2002 	if (unlikely(ftrace_disabled))
2003 		return;
2004 
2005 	/* Force update next time */
2006 	saved_ftrace_func = NULL;
2007 	/* ftrace_start_up is true if we want ftrace running */
2008 	if (ftrace_start_up)
2009 		ftrace_run_update_code(FTRACE_UPDATE_CALLS);
2010 }
2011 
2012 static void ftrace_shutdown_sysctl(void)
2013 {
2014 	if (unlikely(ftrace_disabled))
2015 		return;
2016 
2017 	/* ftrace_start_up is true if ftrace is running */
2018 	if (ftrace_start_up)
2019 		ftrace_run_update_code(FTRACE_DISABLE_CALLS);
2020 }
2021 
2022 static cycle_t		ftrace_update_time;
2023 static unsigned long	ftrace_update_cnt;
2024 unsigned long		ftrace_update_tot_cnt;
2025 
2026 static int ops_traces_mod(struct ftrace_ops *ops)
2027 {
2028 	struct ftrace_hash *hash;
2029 
2030 	hash = ops->filter_hash;
2031 	return ftrace_hash_empty(hash);
2032 }
2033 
2034 static int ftrace_update_code(struct module *mod)
2035 {
2036 	struct ftrace_page *pg;
2037 	struct dyn_ftrace *p;
2038 	cycle_t start, stop;
2039 	unsigned long ref = 0;
2040 	int i;
2041 
2042 	/*
2043 	 * When adding a module, we need to check if tracers are
2044 	 * currently enabled and if they are set to trace all functions.
2045 	 * If they are, we need to enable the module functions as well
2046 	 * as update the reference counts for those function records.
2047 	 */
2048 	if (mod) {
2049 		struct ftrace_ops *ops;
2050 
2051 		for (ops = ftrace_ops_list;
2052 		     ops != &ftrace_list_end; ops = ops->next) {
2053 			if (ops->flags & FTRACE_OPS_FL_ENABLED &&
2054 			    ops_traces_mod(ops))
2055 				ref++;
2056 		}
2057 	}
2058 
2059 	start = ftrace_now(raw_smp_processor_id());
2060 	ftrace_update_cnt = 0;
2061 
2062 	for (pg = ftrace_new_pgs; pg; pg = pg->next) {
2063 
2064 		for (i = 0; i < pg->index; i++) {
2065 			/* If something went wrong, bail without enabling anything */
2066 			if (unlikely(ftrace_disabled))
2067 				return -1;
2068 
2069 			p = &pg->records[i];
2070 			p->flags = ref;
2071 
2072 			/*
2073 			 * Do the initial record conversion from mcount jump
2074 			 * to the NOP instructions.
2075 			 */
2076 			if (!ftrace_code_disable(mod, p))
2077 				break;
2078 
2079 			ftrace_update_cnt++;
2080 
2081 			/*
2082 			 * If the tracing is enabled, go ahead and enable the record.
2083 			 *
2084 			 * The reason not to enable the record immediatelly is the
2085 			 * inherent check of ftrace_make_nop/ftrace_make_call for
2086 			 * correct previous instructions.  Making first the NOP
2087 			 * conversion puts the module to the correct state, thus
2088 			 * passing the ftrace_make_call check.
2089 			 */
2090 			if (ftrace_start_up && ref) {
2091 				int failed = __ftrace_replace_code(p, 1);
2092 				if (failed)
2093 					ftrace_bug(failed, p->ip);
2094 			}
2095 		}
2096 	}
2097 
2098 	ftrace_new_pgs = NULL;
2099 
2100 	stop = ftrace_now(raw_smp_processor_id());
2101 	ftrace_update_time = stop - start;
2102 	ftrace_update_tot_cnt += ftrace_update_cnt;
2103 
2104 	return 0;
2105 }
2106 
2107 static int ftrace_allocate_records(struct ftrace_page *pg, int count)
2108 {
2109 	int order;
2110 	int cnt;
2111 
2112 	if (WARN_ON(!count))
2113 		return -EINVAL;
2114 
2115 	order = get_count_order(DIV_ROUND_UP(count, ENTRIES_PER_PAGE));
2116 
2117 	/*
2118 	 * We want to fill as much as possible. No more than a page
2119 	 * may be empty.
2120 	 */
2121 	while ((PAGE_SIZE << order) / ENTRY_SIZE >= count + ENTRIES_PER_PAGE)
2122 		order--;
2123 
2124  again:
2125 	pg->records = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
2126 
2127 	if (!pg->records) {
2128 		/* if we can't allocate this size, try something smaller */
2129 		if (!order)
2130 			return -ENOMEM;
2131 		order >>= 1;
2132 		goto again;
2133 	}
2134 
2135 	cnt = (PAGE_SIZE << order) / ENTRY_SIZE;
2136 	pg->size = cnt;
2137 
2138 	if (cnt > count)
2139 		cnt = count;
2140 
2141 	return cnt;
2142 }
2143 
2144 static struct ftrace_page *
2145 ftrace_allocate_pages(unsigned long num_to_init)
2146 {
2147 	struct ftrace_page *start_pg;
2148 	struct ftrace_page *pg;
2149 	int order;
2150 	int cnt;
2151 
2152 	if (!num_to_init)
2153 		return 0;
2154 
2155 	start_pg = pg = kzalloc(sizeof(*pg), GFP_KERNEL);
2156 	if (!pg)
2157 		return NULL;
2158 
2159 	/*
2160 	 * Try to allocate as much as possible in one continues
2161 	 * location that fills in all of the space. We want to
2162 	 * waste as little space as possible.
2163 	 */
2164 	for (;;) {
2165 		cnt = ftrace_allocate_records(pg, num_to_init);
2166 		if (cnt < 0)
2167 			goto free_pages;
2168 
2169 		num_to_init -= cnt;
2170 		if (!num_to_init)
2171 			break;
2172 
2173 		pg->next = kzalloc(sizeof(*pg), GFP_KERNEL);
2174 		if (!pg->next)
2175 			goto free_pages;
2176 
2177 		pg = pg->next;
2178 	}
2179 
2180 	return start_pg;
2181 
2182  free_pages:
2183 	while (start_pg) {
2184 		order = get_count_order(pg->size / ENTRIES_PER_PAGE);
2185 		free_pages((unsigned long)pg->records, order);
2186 		start_pg = pg->next;
2187 		kfree(pg);
2188 		pg = start_pg;
2189 	}
2190 	pr_info("ftrace: FAILED to allocate memory for functions\n");
2191 	return NULL;
2192 }
2193 
2194 static int __init ftrace_dyn_table_alloc(unsigned long num_to_init)
2195 {
2196 	int cnt;
2197 
2198 	if (!num_to_init) {
2199 		pr_info("ftrace: No functions to be traced?\n");
2200 		return -1;
2201 	}
2202 
2203 	cnt = num_to_init / ENTRIES_PER_PAGE;
2204 	pr_info("ftrace: allocating %ld entries in %d pages\n",
2205 		num_to_init, cnt + 1);
2206 
2207 	return 0;
2208 }
2209 
2210 #define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
2211 
2212 struct ftrace_iterator {
2213 	loff_t				pos;
2214 	loff_t				func_pos;
2215 	struct ftrace_page		*pg;
2216 	struct dyn_ftrace		*func;
2217 	struct ftrace_func_probe	*probe;
2218 	struct trace_parser		parser;
2219 	struct ftrace_hash		*hash;
2220 	struct ftrace_ops		*ops;
2221 	int				hidx;
2222 	int				idx;
2223 	unsigned			flags;
2224 };
2225 
2226 static void *
2227 t_hash_next(struct seq_file *m, loff_t *pos)
2228 {
2229 	struct ftrace_iterator *iter = m->private;
2230 	struct hlist_node *hnd = NULL;
2231 	struct hlist_head *hhd;
2232 
2233 	(*pos)++;
2234 	iter->pos = *pos;
2235 
2236 	if (iter->probe)
2237 		hnd = &iter->probe->node;
2238  retry:
2239 	if (iter->hidx >= FTRACE_FUNC_HASHSIZE)
2240 		return NULL;
2241 
2242 	hhd = &ftrace_func_hash[iter->hidx];
2243 
2244 	if (hlist_empty(hhd)) {
2245 		iter->hidx++;
2246 		hnd = NULL;
2247 		goto retry;
2248 	}
2249 
2250 	if (!hnd)
2251 		hnd = hhd->first;
2252 	else {
2253 		hnd = hnd->next;
2254 		if (!hnd) {
2255 			iter->hidx++;
2256 			goto retry;
2257 		}
2258 	}
2259 
2260 	if (WARN_ON_ONCE(!hnd))
2261 		return NULL;
2262 
2263 	iter->probe = hlist_entry(hnd, struct ftrace_func_probe, node);
2264 
2265 	return iter;
2266 }
2267 
2268 static void *t_hash_start(struct seq_file *m, loff_t *pos)
2269 {
2270 	struct ftrace_iterator *iter = m->private;
2271 	void *p = NULL;
2272 	loff_t l;
2273 
2274 	if (!(iter->flags & FTRACE_ITER_DO_HASH))
2275 		return NULL;
2276 
2277 	if (iter->func_pos > *pos)
2278 		return NULL;
2279 
2280 	iter->hidx = 0;
2281 	for (l = 0; l <= (*pos - iter->func_pos); ) {
2282 		p = t_hash_next(m, &l);
2283 		if (!p)
2284 			break;
2285 	}
2286 	if (!p)
2287 		return NULL;
2288 
2289 	/* Only set this if we have an item */
2290 	iter->flags |= FTRACE_ITER_HASH;
2291 
2292 	return iter;
2293 }
2294 
2295 static int
2296 t_hash_show(struct seq_file *m, struct ftrace_iterator *iter)
2297 {
2298 	struct ftrace_func_probe *rec;
2299 
2300 	rec = iter->probe;
2301 	if (WARN_ON_ONCE(!rec))
2302 		return -EIO;
2303 
2304 	if (rec->ops->print)
2305 		return rec->ops->print(m, rec->ip, rec->ops, rec->data);
2306 
2307 	seq_printf(m, "%ps:%ps", (void *)rec->ip, (void *)rec->ops->func);
2308 
2309 	if (rec->data)
2310 		seq_printf(m, ":%p", rec->data);
2311 	seq_putc(m, '\n');
2312 
2313 	return 0;
2314 }
2315 
2316 static void *
2317 t_next(struct seq_file *m, void *v, loff_t *pos)
2318 {
2319 	struct ftrace_iterator *iter = m->private;
2320 	struct ftrace_ops *ops = iter->ops;
2321 	struct dyn_ftrace *rec = NULL;
2322 
2323 	if (unlikely(ftrace_disabled))
2324 		return NULL;
2325 
2326 	if (iter->flags & FTRACE_ITER_HASH)
2327 		return t_hash_next(m, pos);
2328 
2329 	(*pos)++;
2330 	iter->pos = iter->func_pos = *pos;
2331 
2332 	if (iter->flags & FTRACE_ITER_PRINTALL)
2333 		return t_hash_start(m, pos);
2334 
2335  retry:
2336 	if (iter->idx >= iter->pg->index) {
2337 		if (iter->pg->next) {
2338 			iter->pg = iter->pg->next;
2339 			iter->idx = 0;
2340 			goto retry;
2341 		}
2342 	} else {
2343 		rec = &iter->pg->records[iter->idx++];
2344 		if (((iter->flags & FTRACE_ITER_FILTER) &&
2345 		     !(ftrace_lookup_ip(ops->filter_hash, rec->ip))) ||
2346 
2347 		    ((iter->flags & FTRACE_ITER_NOTRACE) &&
2348 		     !ftrace_lookup_ip(ops->notrace_hash, rec->ip)) ||
2349 
2350 		    ((iter->flags & FTRACE_ITER_ENABLED) &&
2351 		     !(rec->flags & ~FTRACE_FL_MASK))) {
2352 
2353 			rec = NULL;
2354 			goto retry;
2355 		}
2356 	}
2357 
2358 	if (!rec)
2359 		return t_hash_start(m, pos);
2360 
2361 	iter->func = rec;
2362 
2363 	return iter;
2364 }
2365 
2366 static void reset_iter_read(struct ftrace_iterator *iter)
2367 {
2368 	iter->pos = 0;
2369 	iter->func_pos = 0;
2370 	iter->flags &= ~(FTRACE_ITER_PRINTALL & FTRACE_ITER_HASH);
2371 }
2372 
2373 static void *t_start(struct seq_file *m, loff_t *pos)
2374 {
2375 	struct ftrace_iterator *iter = m->private;
2376 	struct ftrace_ops *ops = iter->ops;
2377 	void *p = NULL;
2378 	loff_t l;
2379 
2380 	mutex_lock(&ftrace_lock);
2381 
2382 	if (unlikely(ftrace_disabled))
2383 		return NULL;
2384 
2385 	/*
2386 	 * If an lseek was done, then reset and start from beginning.
2387 	 */
2388 	if (*pos < iter->pos)
2389 		reset_iter_read(iter);
2390 
2391 	/*
2392 	 * For set_ftrace_filter reading, if we have the filter
2393 	 * off, we can short cut and just print out that all
2394 	 * functions are enabled.
2395 	 */
2396 	if (iter->flags & FTRACE_ITER_FILTER &&
2397 	    ftrace_hash_empty(ops->filter_hash)) {
2398 		if (*pos > 0)
2399 			return t_hash_start(m, pos);
2400 		iter->flags |= FTRACE_ITER_PRINTALL;
2401 		/* reset in case of seek/pread */
2402 		iter->flags &= ~FTRACE_ITER_HASH;
2403 		return iter;
2404 	}
2405 
2406 	if (iter->flags & FTRACE_ITER_HASH)
2407 		return t_hash_start(m, pos);
2408 
2409 	/*
2410 	 * Unfortunately, we need to restart at ftrace_pages_start
2411 	 * every time we let go of the ftrace_mutex. This is because
2412 	 * those pointers can change without the lock.
2413 	 */
2414 	iter->pg = ftrace_pages_start;
2415 	iter->idx = 0;
2416 	for (l = 0; l <= *pos; ) {
2417 		p = t_next(m, p, &l);
2418 		if (!p)
2419 			break;
2420 	}
2421 
2422 	if (!p)
2423 		return t_hash_start(m, pos);
2424 
2425 	return iter;
2426 }
2427 
2428 static void t_stop(struct seq_file *m, void *p)
2429 {
2430 	mutex_unlock(&ftrace_lock);
2431 }
2432 
2433 static int t_show(struct seq_file *m, void *v)
2434 {
2435 	struct ftrace_iterator *iter = m->private;
2436 	struct dyn_ftrace *rec;
2437 
2438 	if (iter->flags & FTRACE_ITER_HASH)
2439 		return t_hash_show(m, iter);
2440 
2441 	if (iter->flags & FTRACE_ITER_PRINTALL) {
2442 		seq_printf(m, "#### all functions enabled ####\n");
2443 		return 0;
2444 	}
2445 
2446 	rec = iter->func;
2447 
2448 	if (!rec)
2449 		return 0;
2450 
2451 	seq_printf(m, "%ps", (void *)rec->ip);
2452 	if (iter->flags & FTRACE_ITER_ENABLED)
2453 		seq_printf(m, " (%ld)",
2454 			   rec->flags & ~FTRACE_FL_MASK);
2455 	seq_printf(m, "\n");
2456 
2457 	return 0;
2458 }
2459 
2460 static const struct seq_operations show_ftrace_seq_ops = {
2461 	.start = t_start,
2462 	.next = t_next,
2463 	.stop = t_stop,
2464 	.show = t_show,
2465 };
2466 
2467 static int
2468 ftrace_avail_open(struct inode *inode, struct file *file)
2469 {
2470 	struct ftrace_iterator *iter;
2471 	int ret;
2472 
2473 	if (unlikely(ftrace_disabled))
2474 		return -ENODEV;
2475 
2476 	iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2477 	if (!iter)
2478 		return -ENOMEM;
2479 
2480 	iter->pg = ftrace_pages_start;
2481 	iter->ops = &global_ops;
2482 
2483 	ret = seq_open(file, &show_ftrace_seq_ops);
2484 	if (!ret) {
2485 		struct seq_file *m = file->private_data;
2486 
2487 		m->private = iter;
2488 	} else {
2489 		kfree(iter);
2490 	}
2491 
2492 	return ret;
2493 }
2494 
2495 static int
2496 ftrace_enabled_open(struct inode *inode, struct file *file)
2497 {
2498 	struct ftrace_iterator *iter;
2499 	int ret;
2500 
2501 	if (unlikely(ftrace_disabled))
2502 		return -ENODEV;
2503 
2504 	iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2505 	if (!iter)
2506 		return -ENOMEM;
2507 
2508 	iter->pg = ftrace_pages_start;
2509 	iter->flags = FTRACE_ITER_ENABLED;
2510 	iter->ops = &global_ops;
2511 
2512 	ret = seq_open(file, &show_ftrace_seq_ops);
2513 	if (!ret) {
2514 		struct seq_file *m = file->private_data;
2515 
2516 		m->private = iter;
2517 	} else {
2518 		kfree(iter);
2519 	}
2520 
2521 	return ret;
2522 }
2523 
2524 static void ftrace_filter_reset(struct ftrace_hash *hash)
2525 {
2526 	mutex_lock(&ftrace_lock);
2527 	ftrace_hash_clear(hash);
2528 	mutex_unlock(&ftrace_lock);
2529 }
2530 
2531 /**
2532  * ftrace_regex_open - initialize function tracer filter files
2533  * @ops: The ftrace_ops that hold the hash filters
2534  * @flag: The type of filter to process
2535  * @inode: The inode, usually passed in to your open routine
2536  * @file: The file, usually passed in to your open routine
2537  *
2538  * ftrace_regex_open() initializes the filter files for the
2539  * @ops. Depending on @flag it may process the filter hash or
2540  * the notrace hash of @ops. With this called from the open
2541  * routine, you can use ftrace_filter_write() for the write
2542  * routine if @flag has FTRACE_ITER_FILTER set, or
2543  * ftrace_notrace_write() if @flag has FTRACE_ITER_NOTRACE set.
2544  * ftrace_regex_lseek() should be used as the lseek routine, and
2545  * release must call ftrace_regex_release().
2546  */
2547 int
2548 ftrace_regex_open(struct ftrace_ops *ops, int flag,
2549 		  struct inode *inode, struct file *file)
2550 {
2551 	struct ftrace_iterator *iter;
2552 	struct ftrace_hash *hash;
2553 	int ret = 0;
2554 
2555 	if (unlikely(ftrace_disabled))
2556 		return -ENODEV;
2557 
2558 	iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2559 	if (!iter)
2560 		return -ENOMEM;
2561 
2562 	if (trace_parser_get_init(&iter->parser, FTRACE_BUFF_MAX)) {
2563 		kfree(iter);
2564 		return -ENOMEM;
2565 	}
2566 
2567 	if (flag & FTRACE_ITER_NOTRACE)
2568 		hash = ops->notrace_hash;
2569 	else
2570 		hash = ops->filter_hash;
2571 
2572 	iter->ops = ops;
2573 	iter->flags = flag;
2574 
2575 	if (file->f_mode & FMODE_WRITE) {
2576 		mutex_lock(&ftrace_lock);
2577 		iter->hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, hash);
2578 		mutex_unlock(&ftrace_lock);
2579 
2580 		if (!iter->hash) {
2581 			trace_parser_put(&iter->parser);
2582 			kfree(iter);
2583 			return -ENOMEM;
2584 		}
2585 	}
2586 
2587 	mutex_lock(&ftrace_regex_lock);
2588 
2589 	if ((file->f_mode & FMODE_WRITE) &&
2590 	    (file->f_flags & O_TRUNC))
2591 		ftrace_filter_reset(iter->hash);
2592 
2593 	if (file->f_mode & FMODE_READ) {
2594 		iter->pg = ftrace_pages_start;
2595 
2596 		ret = seq_open(file, &show_ftrace_seq_ops);
2597 		if (!ret) {
2598 			struct seq_file *m = file->private_data;
2599 			m->private = iter;
2600 		} else {
2601 			/* Failed */
2602 			free_ftrace_hash(iter->hash);
2603 			trace_parser_put(&iter->parser);
2604 			kfree(iter);
2605 		}
2606 	} else
2607 		file->private_data = iter;
2608 	mutex_unlock(&ftrace_regex_lock);
2609 
2610 	return ret;
2611 }
2612 
2613 static int
2614 ftrace_filter_open(struct inode *inode, struct file *file)
2615 {
2616 	return ftrace_regex_open(&global_ops,
2617 			FTRACE_ITER_FILTER | FTRACE_ITER_DO_HASH,
2618 			inode, file);
2619 }
2620 
2621 static int
2622 ftrace_notrace_open(struct inode *inode, struct file *file)
2623 {
2624 	return ftrace_regex_open(&global_ops, FTRACE_ITER_NOTRACE,
2625 				 inode, file);
2626 }
2627 
2628 loff_t
2629 ftrace_regex_lseek(struct file *file, loff_t offset, int origin)
2630 {
2631 	loff_t ret;
2632 
2633 	if (file->f_mode & FMODE_READ)
2634 		ret = seq_lseek(file, offset, origin);
2635 	else
2636 		file->f_pos = ret = 1;
2637 
2638 	return ret;
2639 }
2640 
2641 static int ftrace_match(char *str, char *regex, int len, int type)
2642 {
2643 	int matched = 0;
2644 	int slen;
2645 
2646 	switch (type) {
2647 	case MATCH_FULL:
2648 		if (strcmp(str, regex) == 0)
2649 			matched = 1;
2650 		break;
2651 	case MATCH_FRONT_ONLY:
2652 		if (strncmp(str, regex, len) == 0)
2653 			matched = 1;
2654 		break;
2655 	case MATCH_MIDDLE_ONLY:
2656 		if (strstr(str, regex))
2657 			matched = 1;
2658 		break;
2659 	case MATCH_END_ONLY:
2660 		slen = strlen(str);
2661 		if (slen >= len && memcmp(str + slen - len, regex, len) == 0)
2662 			matched = 1;
2663 		break;
2664 	}
2665 
2666 	return matched;
2667 }
2668 
2669 static int
2670 enter_record(struct ftrace_hash *hash, struct dyn_ftrace *rec, int not)
2671 {
2672 	struct ftrace_func_entry *entry;
2673 	int ret = 0;
2674 
2675 	entry = ftrace_lookup_ip(hash, rec->ip);
2676 	if (not) {
2677 		/* Do nothing if it doesn't exist */
2678 		if (!entry)
2679 			return 0;
2680 
2681 		free_hash_entry(hash, entry);
2682 	} else {
2683 		/* Do nothing if it exists */
2684 		if (entry)
2685 			return 0;
2686 
2687 		ret = add_hash_entry(hash, rec->ip);
2688 	}
2689 	return ret;
2690 }
2691 
2692 static int
2693 ftrace_match_record(struct dyn_ftrace *rec, char *mod,
2694 		    char *regex, int len, int type)
2695 {
2696 	char str[KSYM_SYMBOL_LEN];
2697 	char *modname;
2698 
2699 	kallsyms_lookup(rec->ip, NULL, NULL, &modname, str);
2700 
2701 	if (mod) {
2702 		/* module lookup requires matching the module */
2703 		if (!modname || strcmp(modname, mod))
2704 			return 0;
2705 
2706 		/* blank search means to match all funcs in the mod */
2707 		if (!len)
2708 			return 1;
2709 	}
2710 
2711 	return ftrace_match(str, regex, len, type);
2712 }
2713 
2714 static int
2715 match_records(struct ftrace_hash *hash, char *buff,
2716 	      int len, char *mod, int not)
2717 {
2718 	unsigned search_len = 0;
2719 	struct ftrace_page *pg;
2720 	struct dyn_ftrace *rec;
2721 	int type = MATCH_FULL;
2722 	char *search = buff;
2723 	int found = 0;
2724 	int ret;
2725 
2726 	if (len) {
2727 		type = filter_parse_regex(buff, len, &search, &not);
2728 		search_len = strlen(search);
2729 	}
2730 
2731 	mutex_lock(&ftrace_lock);
2732 
2733 	if (unlikely(ftrace_disabled))
2734 		goto out_unlock;
2735 
2736 	do_for_each_ftrace_rec(pg, rec) {
2737 		if (ftrace_match_record(rec, mod, search, search_len, type)) {
2738 			ret = enter_record(hash, rec, not);
2739 			if (ret < 0) {
2740 				found = ret;
2741 				goto out_unlock;
2742 			}
2743 			found = 1;
2744 		}
2745 	} while_for_each_ftrace_rec();
2746  out_unlock:
2747 	mutex_unlock(&ftrace_lock);
2748 
2749 	return found;
2750 }
2751 
2752 static int
2753 ftrace_match_records(struct ftrace_hash *hash, char *buff, int len)
2754 {
2755 	return match_records(hash, buff, len, NULL, 0);
2756 }
2757 
2758 static int
2759 ftrace_match_module_records(struct ftrace_hash *hash, char *buff, char *mod)
2760 {
2761 	int not = 0;
2762 
2763 	/* blank or '*' mean the same */
2764 	if (strcmp(buff, "*") == 0)
2765 		buff[0] = 0;
2766 
2767 	/* handle the case of 'dont filter this module' */
2768 	if (strcmp(buff, "!") == 0 || strcmp(buff, "!*") == 0) {
2769 		buff[0] = 0;
2770 		not = 1;
2771 	}
2772 
2773 	return match_records(hash, buff, strlen(buff), mod, not);
2774 }
2775 
2776 /*
2777  * We register the module command as a template to show others how
2778  * to register the a command as well.
2779  */
2780 
2781 static int
2782 ftrace_mod_callback(struct ftrace_hash *hash,
2783 		    char *func, char *cmd, char *param, int enable)
2784 {
2785 	char *mod;
2786 	int ret = -EINVAL;
2787 
2788 	/*
2789 	 * cmd == 'mod' because we only registered this func
2790 	 * for the 'mod' ftrace_func_command.
2791 	 * But if you register one func with multiple commands,
2792 	 * you can tell which command was used by the cmd
2793 	 * parameter.
2794 	 */
2795 
2796 	/* we must have a module name */
2797 	if (!param)
2798 		return ret;
2799 
2800 	mod = strsep(&param, ":");
2801 	if (!strlen(mod))
2802 		return ret;
2803 
2804 	ret = ftrace_match_module_records(hash, func, mod);
2805 	if (!ret)
2806 		ret = -EINVAL;
2807 	if (ret < 0)
2808 		return ret;
2809 
2810 	return 0;
2811 }
2812 
2813 static struct ftrace_func_command ftrace_mod_cmd = {
2814 	.name			= "mod",
2815 	.func			= ftrace_mod_callback,
2816 };
2817 
2818 static int __init ftrace_mod_cmd_init(void)
2819 {
2820 	return register_ftrace_command(&ftrace_mod_cmd);
2821 }
2822 device_initcall(ftrace_mod_cmd_init);
2823 
2824 static void
2825 function_trace_probe_call(unsigned long ip, unsigned long parent_ip)
2826 {
2827 	struct ftrace_func_probe *entry;
2828 	struct hlist_head *hhd;
2829 	struct hlist_node *n;
2830 	unsigned long key;
2831 
2832 	key = hash_long(ip, FTRACE_HASH_BITS);
2833 
2834 	hhd = &ftrace_func_hash[key];
2835 
2836 	if (hlist_empty(hhd))
2837 		return;
2838 
2839 	/*
2840 	 * Disable preemption for these calls to prevent a RCU grace
2841 	 * period. This syncs the hash iteration and freeing of items
2842 	 * on the hash. rcu_read_lock is too dangerous here.
2843 	 */
2844 	preempt_disable_notrace();
2845 	hlist_for_each_entry_rcu(entry, n, hhd, node) {
2846 		if (entry->ip == ip)
2847 			entry->ops->func(ip, parent_ip, &entry->data);
2848 	}
2849 	preempt_enable_notrace();
2850 }
2851 
2852 static struct ftrace_ops trace_probe_ops __read_mostly =
2853 {
2854 	.func		= function_trace_probe_call,
2855 };
2856 
2857 static int ftrace_probe_registered;
2858 
2859 static void __enable_ftrace_function_probe(void)
2860 {
2861 	int ret;
2862 	int i;
2863 
2864 	if (ftrace_probe_registered)
2865 		return;
2866 
2867 	for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
2868 		struct hlist_head *hhd = &ftrace_func_hash[i];
2869 		if (hhd->first)
2870 			break;
2871 	}
2872 	/* Nothing registered? */
2873 	if (i == FTRACE_FUNC_HASHSIZE)
2874 		return;
2875 
2876 	ret = __register_ftrace_function(&trace_probe_ops);
2877 	if (!ret)
2878 		ret = ftrace_startup(&trace_probe_ops, 0);
2879 
2880 	ftrace_probe_registered = 1;
2881 }
2882 
2883 static void __disable_ftrace_function_probe(void)
2884 {
2885 	int ret;
2886 	int i;
2887 
2888 	if (!ftrace_probe_registered)
2889 		return;
2890 
2891 	for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
2892 		struct hlist_head *hhd = &ftrace_func_hash[i];
2893 		if (hhd->first)
2894 			return;
2895 	}
2896 
2897 	/* no more funcs left */
2898 	ret = __unregister_ftrace_function(&trace_probe_ops);
2899 	if (!ret)
2900 		ftrace_shutdown(&trace_probe_ops, 0);
2901 
2902 	ftrace_probe_registered = 0;
2903 }
2904 
2905 
2906 static void ftrace_free_entry_rcu(struct rcu_head *rhp)
2907 {
2908 	struct ftrace_func_probe *entry =
2909 		container_of(rhp, struct ftrace_func_probe, rcu);
2910 
2911 	if (entry->ops->free)
2912 		entry->ops->free(&entry->data);
2913 	kfree(entry);
2914 }
2915 
2916 
2917 int
2918 register_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
2919 			      void *data)
2920 {
2921 	struct ftrace_func_probe *entry;
2922 	struct ftrace_page *pg;
2923 	struct dyn_ftrace *rec;
2924 	int type, len, not;
2925 	unsigned long key;
2926 	int count = 0;
2927 	char *search;
2928 
2929 	type = filter_parse_regex(glob, strlen(glob), &search, &not);
2930 	len = strlen(search);
2931 
2932 	/* we do not support '!' for function probes */
2933 	if (WARN_ON(not))
2934 		return -EINVAL;
2935 
2936 	mutex_lock(&ftrace_lock);
2937 
2938 	if (unlikely(ftrace_disabled))
2939 		goto out_unlock;
2940 
2941 	do_for_each_ftrace_rec(pg, rec) {
2942 
2943 		if (!ftrace_match_record(rec, NULL, search, len, type))
2944 			continue;
2945 
2946 		entry = kmalloc(sizeof(*entry), GFP_KERNEL);
2947 		if (!entry) {
2948 			/* If we did not process any, then return error */
2949 			if (!count)
2950 				count = -ENOMEM;
2951 			goto out_unlock;
2952 		}
2953 
2954 		count++;
2955 
2956 		entry->data = data;
2957 
2958 		/*
2959 		 * The caller might want to do something special
2960 		 * for each function we find. We call the callback
2961 		 * to give the caller an opportunity to do so.
2962 		 */
2963 		if (ops->callback) {
2964 			if (ops->callback(rec->ip, &entry->data) < 0) {
2965 				/* caller does not like this func */
2966 				kfree(entry);
2967 				continue;
2968 			}
2969 		}
2970 
2971 		entry->ops = ops;
2972 		entry->ip = rec->ip;
2973 
2974 		key = hash_long(entry->ip, FTRACE_HASH_BITS);
2975 		hlist_add_head_rcu(&entry->node, &ftrace_func_hash[key]);
2976 
2977 	} while_for_each_ftrace_rec();
2978 	__enable_ftrace_function_probe();
2979 
2980  out_unlock:
2981 	mutex_unlock(&ftrace_lock);
2982 
2983 	return count;
2984 }
2985 
2986 enum {
2987 	PROBE_TEST_FUNC		= 1,
2988 	PROBE_TEST_DATA		= 2
2989 };
2990 
2991 static void
2992 __unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
2993 				  void *data, int flags)
2994 {
2995 	struct ftrace_func_probe *entry;
2996 	struct hlist_node *n, *tmp;
2997 	char str[KSYM_SYMBOL_LEN];
2998 	int type = MATCH_FULL;
2999 	int i, len = 0;
3000 	char *search;
3001 
3002 	if (glob && (strcmp(glob, "*") == 0 || !strlen(glob)))
3003 		glob = NULL;
3004 	else if (glob) {
3005 		int not;
3006 
3007 		type = filter_parse_regex(glob, strlen(glob), &search, &not);
3008 		len = strlen(search);
3009 
3010 		/* we do not support '!' for function probes */
3011 		if (WARN_ON(not))
3012 			return;
3013 	}
3014 
3015 	mutex_lock(&ftrace_lock);
3016 	for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
3017 		struct hlist_head *hhd = &ftrace_func_hash[i];
3018 
3019 		hlist_for_each_entry_safe(entry, n, tmp, hhd, node) {
3020 
3021 			/* break up if statements for readability */
3022 			if ((flags & PROBE_TEST_FUNC) && entry->ops != ops)
3023 				continue;
3024 
3025 			if ((flags & PROBE_TEST_DATA) && entry->data != data)
3026 				continue;
3027 
3028 			/* do this last, since it is the most expensive */
3029 			if (glob) {
3030 				kallsyms_lookup(entry->ip, NULL, NULL,
3031 						NULL, str);
3032 				if (!ftrace_match(str, glob, len, type))
3033 					continue;
3034 			}
3035 
3036 			hlist_del(&entry->node);
3037 			call_rcu(&entry->rcu, ftrace_free_entry_rcu);
3038 		}
3039 	}
3040 	__disable_ftrace_function_probe();
3041 	mutex_unlock(&ftrace_lock);
3042 }
3043 
3044 void
3045 unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
3046 				void *data)
3047 {
3048 	__unregister_ftrace_function_probe(glob, ops, data,
3049 					  PROBE_TEST_FUNC | PROBE_TEST_DATA);
3050 }
3051 
3052 void
3053 unregister_ftrace_function_probe_func(char *glob, struct ftrace_probe_ops *ops)
3054 {
3055 	__unregister_ftrace_function_probe(glob, ops, NULL, PROBE_TEST_FUNC);
3056 }
3057 
3058 void unregister_ftrace_function_probe_all(char *glob)
3059 {
3060 	__unregister_ftrace_function_probe(glob, NULL, NULL, 0);
3061 }
3062 
3063 static LIST_HEAD(ftrace_commands);
3064 static DEFINE_MUTEX(ftrace_cmd_mutex);
3065 
3066 int register_ftrace_command(struct ftrace_func_command *cmd)
3067 {
3068 	struct ftrace_func_command *p;
3069 	int ret = 0;
3070 
3071 	mutex_lock(&ftrace_cmd_mutex);
3072 	list_for_each_entry(p, &ftrace_commands, list) {
3073 		if (strcmp(cmd->name, p->name) == 0) {
3074 			ret = -EBUSY;
3075 			goto out_unlock;
3076 		}
3077 	}
3078 	list_add(&cmd->list, &ftrace_commands);
3079  out_unlock:
3080 	mutex_unlock(&ftrace_cmd_mutex);
3081 
3082 	return ret;
3083 }
3084 
3085 int unregister_ftrace_command(struct ftrace_func_command *cmd)
3086 {
3087 	struct ftrace_func_command *p, *n;
3088 	int ret = -ENODEV;
3089 
3090 	mutex_lock(&ftrace_cmd_mutex);
3091 	list_for_each_entry_safe(p, n, &ftrace_commands, list) {
3092 		if (strcmp(cmd->name, p->name) == 0) {
3093 			ret = 0;
3094 			list_del_init(&p->list);
3095 			goto out_unlock;
3096 		}
3097 	}
3098  out_unlock:
3099 	mutex_unlock(&ftrace_cmd_mutex);
3100 
3101 	return ret;
3102 }
3103 
3104 static int ftrace_process_regex(struct ftrace_hash *hash,
3105 				char *buff, int len, int enable)
3106 {
3107 	char *func, *command, *next = buff;
3108 	struct ftrace_func_command *p;
3109 	int ret = -EINVAL;
3110 
3111 	func = strsep(&next, ":");
3112 
3113 	if (!next) {
3114 		ret = ftrace_match_records(hash, func, len);
3115 		if (!ret)
3116 			ret = -EINVAL;
3117 		if (ret < 0)
3118 			return ret;
3119 		return 0;
3120 	}
3121 
3122 	/* command found */
3123 
3124 	command = strsep(&next, ":");
3125 
3126 	mutex_lock(&ftrace_cmd_mutex);
3127 	list_for_each_entry(p, &ftrace_commands, list) {
3128 		if (strcmp(p->name, command) == 0) {
3129 			ret = p->func(hash, func, command, next, enable);
3130 			goto out_unlock;
3131 		}
3132 	}
3133  out_unlock:
3134 	mutex_unlock(&ftrace_cmd_mutex);
3135 
3136 	return ret;
3137 }
3138 
3139 static ssize_t
3140 ftrace_regex_write(struct file *file, const char __user *ubuf,
3141 		   size_t cnt, loff_t *ppos, int enable)
3142 {
3143 	struct ftrace_iterator *iter;
3144 	struct trace_parser *parser;
3145 	ssize_t ret, read;
3146 
3147 	if (!cnt)
3148 		return 0;
3149 
3150 	mutex_lock(&ftrace_regex_lock);
3151 
3152 	ret = -ENODEV;
3153 	if (unlikely(ftrace_disabled))
3154 		goto out_unlock;
3155 
3156 	if (file->f_mode & FMODE_READ) {
3157 		struct seq_file *m = file->private_data;
3158 		iter = m->private;
3159 	} else
3160 		iter = file->private_data;
3161 
3162 	parser = &iter->parser;
3163 	read = trace_get_user(parser, ubuf, cnt, ppos);
3164 
3165 	if (read >= 0 && trace_parser_loaded(parser) &&
3166 	    !trace_parser_cont(parser)) {
3167 		ret = ftrace_process_regex(iter->hash, parser->buffer,
3168 					   parser->idx, enable);
3169 		trace_parser_clear(parser);
3170 		if (ret)
3171 			goto out_unlock;
3172 	}
3173 
3174 	ret = read;
3175 out_unlock:
3176 	mutex_unlock(&ftrace_regex_lock);
3177 
3178 	return ret;
3179 }
3180 
3181 ssize_t
3182 ftrace_filter_write(struct file *file, const char __user *ubuf,
3183 		    size_t cnt, loff_t *ppos)
3184 {
3185 	return ftrace_regex_write(file, ubuf, cnt, ppos, 1);
3186 }
3187 
3188 ssize_t
3189 ftrace_notrace_write(struct file *file, const char __user *ubuf,
3190 		     size_t cnt, loff_t *ppos)
3191 {
3192 	return ftrace_regex_write(file, ubuf, cnt, ppos, 0);
3193 }
3194 
3195 static int
3196 ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len,
3197 		 int reset, int enable)
3198 {
3199 	struct ftrace_hash **orig_hash;
3200 	struct ftrace_hash *hash;
3201 	int ret;
3202 
3203 	/* All global ops uses the global ops filters */
3204 	if (ops->flags & FTRACE_OPS_FL_GLOBAL)
3205 		ops = &global_ops;
3206 
3207 	if (unlikely(ftrace_disabled))
3208 		return -ENODEV;
3209 
3210 	if (enable)
3211 		orig_hash = &ops->filter_hash;
3212 	else
3213 		orig_hash = &ops->notrace_hash;
3214 
3215 	hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
3216 	if (!hash)
3217 		return -ENOMEM;
3218 
3219 	mutex_lock(&ftrace_regex_lock);
3220 	if (reset)
3221 		ftrace_filter_reset(hash);
3222 	if (buf && !ftrace_match_records(hash, buf, len)) {
3223 		ret = -EINVAL;
3224 		goto out_regex_unlock;
3225 	}
3226 
3227 	mutex_lock(&ftrace_lock);
3228 	ret = ftrace_hash_move(ops, enable, orig_hash, hash);
3229 	if (!ret && ops->flags & FTRACE_OPS_FL_ENABLED
3230 	    && ftrace_enabled)
3231 		ftrace_run_update_code(FTRACE_UPDATE_CALLS);
3232 
3233 	mutex_unlock(&ftrace_lock);
3234 
3235  out_regex_unlock:
3236 	mutex_unlock(&ftrace_regex_lock);
3237 
3238 	free_ftrace_hash(hash);
3239 	return ret;
3240 }
3241 
3242 /**
3243  * ftrace_set_filter - set a function to filter on in ftrace
3244  * @ops - the ops to set the filter with
3245  * @buf - the string that holds the function filter text.
3246  * @len - the length of the string.
3247  * @reset - non zero to reset all filters before applying this filter.
3248  *
3249  * Filters denote which functions should be enabled when tracing is enabled.
3250  * If @buf is NULL and reset is set, all functions will be enabled for tracing.
3251  */
3252 int ftrace_set_filter(struct ftrace_ops *ops, unsigned char *buf,
3253 		       int len, int reset)
3254 {
3255 	return ftrace_set_regex(ops, buf, len, reset, 1);
3256 }
3257 EXPORT_SYMBOL_GPL(ftrace_set_filter);
3258 
3259 /**
3260  * ftrace_set_notrace - set a function to not trace in ftrace
3261  * @ops - the ops to set the notrace filter with
3262  * @buf - the string that holds the function notrace text.
3263  * @len - the length of the string.
3264  * @reset - non zero to reset all filters before applying this filter.
3265  *
3266  * Notrace Filters denote which functions should not be enabled when tracing
3267  * is enabled. If @buf is NULL and reset is set, all functions will be enabled
3268  * for tracing.
3269  */
3270 int ftrace_set_notrace(struct ftrace_ops *ops, unsigned char *buf,
3271 			int len, int reset)
3272 {
3273 	return ftrace_set_regex(ops, buf, len, reset, 0);
3274 }
3275 EXPORT_SYMBOL_GPL(ftrace_set_notrace);
3276 /**
3277  * ftrace_set_filter - set a function to filter on in ftrace
3278  * @ops - the ops to set the filter with
3279  * @buf - the string that holds the function filter text.
3280  * @len - the length of the string.
3281  * @reset - non zero to reset all filters before applying this filter.
3282  *
3283  * Filters denote which functions should be enabled when tracing is enabled.
3284  * If @buf is NULL and reset is set, all functions will be enabled for tracing.
3285  */
3286 void ftrace_set_global_filter(unsigned char *buf, int len, int reset)
3287 {
3288 	ftrace_set_regex(&global_ops, buf, len, reset, 1);
3289 }
3290 EXPORT_SYMBOL_GPL(ftrace_set_global_filter);
3291 
3292 /**
3293  * ftrace_set_notrace - set a function to not trace in ftrace
3294  * @ops - the ops to set the notrace filter with
3295  * @buf - the string that holds the function notrace text.
3296  * @len - the length of the string.
3297  * @reset - non zero to reset all filters before applying this filter.
3298  *
3299  * Notrace Filters denote which functions should not be enabled when tracing
3300  * is enabled. If @buf is NULL and reset is set, all functions will be enabled
3301  * for tracing.
3302  */
3303 void ftrace_set_global_notrace(unsigned char *buf, int len, int reset)
3304 {
3305 	ftrace_set_regex(&global_ops, buf, len, reset, 0);
3306 }
3307 EXPORT_SYMBOL_GPL(ftrace_set_global_notrace);
3308 
3309 /*
3310  * command line interface to allow users to set filters on boot up.
3311  */
3312 #define FTRACE_FILTER_SIZE		COMMAND_LINE_SIZE
3313 static char ftrace_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
3314 static char ftrace_filter_buf[FTRACE_FILTER_SIZE] __initdata;
3315 
3316 static int __init set_ftrace_notrace(char *str)
3317 {
3318 	strncpy(ftrace_notrace_buf, str, FTRACE_FILTER_SIZE);
3319 	return 1;
3320 }
3321 __setup("ftrace_notrace=", set_ftrace_notrace);
3322 
3323 static int __init set_ftrace_filter(char *str)
3324 {
3325 	strncpy(ftrace_filter_buf, str, FTRACE_FILTER_SIZE);
3326 	return 1;
3327 }
3328 __setup("ftrace_filter=", set_ftrace_filter);
3329 
3330 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3331 static char ftrace_graph_buf[FTRACE_FILTER_SIZE] __initdata;
3332 static int ftrace_set_func(unsigned long *array, int *idx, char *buffer);
3333 
3334 static int __init set_graph_function(char *str)
3335 {
3336 	strlcpy(ftrace_graph_buf, str, FTRACE_FILTER_SIZE);
3337 	return 1;
3338 }
3339 __setup("ftrace_graph_filter=", set_graph_function);
3340 
3341 static void __init set_ftrace_early_graph(char *buf)
3342 {
3343 	int ret;
3344 	char *func;
3345 
3346 	while (buf) {
3347 		func = strsep(&buf, ",");
3348 		/* we allow only one expression at a time */
3349 		ret = ftrace_set_func(ftrace_graph_funcs, &ftrace_graph_count,
3350 				      func);
3351 		if (ret)
3352 			printk(KERN_DEBUG "ftrace: function %s not "
3353 					  "traceable\n", func);
3354 	}
3355 }
3356 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
3357 
3358 void __init
3359 ftrace_set_early_filter(struct ftrace_ops *ops, char *buf, int enable)
3360 {
3361 	char *func;
3362 
3363 	while (buf) {
3364 		func = strsep(&buf, ",");
3365 		ftrace_set_regex(ops, func, strlen(func), 0, enable);
3366 	}
3367 }
3368 
3369 static void __init set_ftrace_early_filters(void)
3370 {
3371 	if (ftrace_filter_buf[0])
3372 		ftrace_set_early_filter(&global_ops, ftrace_filter_buf, 1);
3373 	if (ftrace_notrace_buf[0])
3374 		ftrace_set_early_filter(&global_ops, ftrace_notrace_buf, 0);
3375 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3376 	if (ftrace_graph_buf[0])
3377 		set_ftrace_early_graph(ftrace_graph_buf);
3378 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
3379 }
3380 
3381 int ftrace_regex_release(struct inode *inode, struct file *file)
3382 {
3383 	struct seq_file *m = (struct seq_file *)file->private_data;
3384 	struct ftrace_iterator *iter;
3385 	struct ftrace_hash **orig_hash;
3386 	struct trace_parser *parser;
3387 	int filter_hash;
3388 	int ret;
3389 
3390 	mutex_lock(&ftrace_regex_lock);
3391 	if (file->f_mode & FMODE_READ) {
3392 		iter = m->private;
3393 
3394 		seq_release(inode, file);
3395 	} else
3396 		iter = file->private_data;
3397 
3398 	parser = &iter->parser;
3399 	if (trace_parser_loaded(parser)) {
3400 		parser->buffer[parser->idx] = 0;
3401 		ftrace_match_records(iter->hash, parser->buffer, parser->idx);
3402 	}
3403 
3404 	trace_parser_put(parser);
3405 
3406 	if (file->f_mode & FMODE_WRITE) {
3407 		filter_hash = !!(iter->flags & FTRACE_ITER_FILTER);
3408 
3409 		if (filter_hash)
3410 			orig_hash = &iter->ops->filter_hash;
3411 		else
3412 			orig_hash = &iter->ops->notrace_hash;
3413 
3414 		mutex_lock(&ftrace_lock);
3415 		ret = ftrace_hash_move(iter->ops, filter_hash,
3416 				       orig_hash, iter->hash);
3417 		if (!ret && (iter->ops->flags & FTRACE_OPS_FL_ENABLED)
3418 		    && ftrace_enabled)
3419 			ftrace_run_update_code(FTRACE_UPDATE_CALLS);
3420 
3421 		mutex_unlock(&ftrace_lock);
3422 	}
3423 	free_ftrace_hash(iter->hash);
3424 	kfree(iter);
3425 
3426 	mutex_unlock(&ftrace_regex_lock);
3427 	return 0;
3428 }
3429 
3430 static const struct file_operations ftrace_avail_fops = {
3431 	.open = ftrace_avail_open,
3432 	.read = seq_read,
3433 	.llseek = seq_lseek,
3434 	.release = seq_release_private,
3435 };
3436 
3437 static const struct file_operations ftrace_enabled_fops = {
3438 	.open = ftrace_enabled_open,
3439 	.read = seq_read,
3440 	.llseek = seq_lseek,
3441 	.release = seq_release_private,
3442 };
3443 
3444 static const struct file_operations ftrace_filter_fops = {
3445 	.open = ftrace_filter_open,
3446 	.read = seq_read,
3447 	.write = ftrace_filter_write,
3448 	.llseek = ftrace_regex_lseek,
3449 	.release = ftrace_regex_release,
3450 };
3451 
3452 static const struct file_operations ftrace_notrace_fops = {
3453 	.open = ftrace_notrace_open,
3454 	.read = seq_read,
3455 	.write = ftrace_notrace_write,
3456 	.llseek = ftrace_regex_lseek,
3457 	.release = ftrace_regex_release,
3458 };
3459 
3460 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3461 
3462 static DEFINE_MUTEX(graph_lock);
3463 
3464 int ftrace_graph_count;
3465 int ftrace_graph_filter_enabled;
3466 unsigned long ftrace_graph_funcs[FTRACE_GRAPH_MAX_FUNCS] __read_mostly;
3467 
3468 static void *
3469 __g_next(struct seq_file *m, loff_t *pos)
3470 {
3471 	if (*pos >= ftrace_graph_count)
3472 		return NULL;
3473 	return &ftrace_graph_funcs[*pos];
3474 }
3475 
3476 static void *
3477 g_next(struct seq_file *m, void *v, loff_t *pos)
3478 {
3479 	(*pos)++;
3480 	return __g_next(m, pos);
3481 }
3482 
3483 static void *g_start(struct seq_file *m, loff_t *pos)
3484 {
3485 	mutex_lock(&graph_lock);
3486 
3487 	/* Nothing, tell g_show to print all functions are enabled */
3488 	if (!ftrace_graph_filter_enabled && !*pos)
3489 		return (void *)1;
3490 
3491 	return __g_next(m, pos);
3492 }
3493 
3494 static void g_stop(struct seq_file *m, void *p)
3495 {
3496 	mutex_unlock(&graph_lock);
3497 }
3498 
3499 static int g_show(struct seq_file *m, void *v)
3500 {
3501 	unsigned long *ptr = v;
3502 
3503 	if (!ptr)
3504 		return 0;
3505 
3506 	if (ptr == (unsigned long *)1) {
3507 		seq_printf(m, "#### all functions enabled ####\n");
3508 		return 0;
3509 	}
3510 
3511 	seq_printf(m, "%ps\n", (void *)*ptr);
3512 
3513 	return 0;
3514 }
3515 
3516 static const struct seq_operations ftrace_graph_seq_ops = {
3517 	.start = g_start,
3518 	.next = g_next,
3519 	.stop = g_stop,
3520 	.show = g_show,
3521 };
3522 
3523 static int
3524 ftrace_graph_open(struct inode *inode, struct file *file)
3525 {
3526 	int ret = 0;
3527 
3528 	if (unlikely(ftrace_disabled))
3529 		return -ENODEV;
3530 
3531 	mutex_lock(&graph_lock);
3532 	if ((file->f_mode & FMODE_WRITE) &&
3533 	    (file->f_flags & O_TRUNC)) {
3534 		ftrace_graph_filter_enabled = 0;
3535 		ftrace_graph_count = 0;
3536 		memset(ftrace_graph_funcs, 0, sizeof(ftrace_graph_funcs));
3537 	}
3538 	mutex_unlock(&graph_lock);
3539 
3540 	if (file->f_mode & FMODE_READ)
3541 		ret = seq_open(file, &ftrace_graph_seq_ops);
3542 
3543 	return ret;
3544 }
3545 
3546 static int
3547 ftrace_graph_release(struct inode *inode, struct file *file)
3548 {
3549 	if (file->f_mode & FMODE_READ)
3550 		seq_release(inode, file);
3551 	return 0;
3552 }
3553 
3554 static int
3555 ftrace_set_func(unsigned long *array, int *idx, char *buffer)
3556 {
3557 	struct dyn_ftrace *rec;
3558 	struct ftrace_page *pg;
3559 	int search_len;
3560 	int fail = 1;
3561 	int type, not;
3562 	char *search;
3563 	bool exists;
3564 	int i;
3565 
3566 	/* decode regex */
3567 	type = filter_parse_regex(buffer, strlen(buffer), &search, &not);
3568 	if (!not && *idx >= FTRACE_GRAPH_MAX_FUNCS)
3569 		return -EBUSY;
3570 
3571 	search_len = strlen(search);
3572 
3573 	mutex_lock(&ftrace_lock);
3574 
3575 	if (unlikely(ftrace_disabled)) {
3576 		mutex_unlock(&ftrace_lock);
3577 		return -ENODEV;
3578 	}
3579 
3580 	do_for_each_ftrace_rec(pg, rec) {
3581 
3582 		if (ftrace_match_record(rec, NULL, search, search_len, type)) {
3583 			/* if it is in the array */
3584 			exists = false;
3585 			for (i = 0; i < *idx; i++) {
3586 				if (array[i] == rec->ip) {
3587 					exists = true;
3588 					break;
3589 				}
3590 			}
3591 
3592 			if (!not) {
3593 				fail = 0;
3594 				if (!exists) {
3595 					array[(*idx)++] = rec->ip;
3596 					if (*idx >= FTRACE_GRAPH_MAX_FUNCS)
3597 						goto out;
3598 				}
3599 			} else {
3600 				if (exists) {
3601 					array[i] = array[--(*idx)];
3602 					array[*idx] = 0;
3603 					fail = 0;
3604 				}
3605 			}
3606 		}
3607 	} while_for_each_ftrace_rec();
3608 out:
3609 	mutex_unlock(&ftrace_lock);
3610 
3611 	if (fail)
3612 		return -EINVAL;
3613 
3614 	ftrace_graph_filter_enabled = 1;
3615 	return 0;
3616 }
3617 
3618 static ssize_t
3619 ftrace_graph_write(struct file *file, const char __user *ubuf,
3620 		   size_t cnt, loff_t *ppos)
3621 {
3622 	struct trace_parser parser;
3623 	ssize_t read, ret;
3624 
3625 	if (!cnt)
3626 		return 0;
3627 
3628 	mutex_lock(&graph_lock);
3629 
3630 	if (trace_parser_get_init(&parser, FTRACE_BUFF_MAX)) {
3631 		ret = -ENOMEM;
3632 		goto out_unlock;
3633 	}
3634 
3635 	read = trace_get_user(&parser, ubuf, cnt, ppos);
3636 
3637 	if (read >= 0 && trace_parser_loaded((&parser))) {
3638 		parser.buffer[parser.idx] = 0;
3639 
3640 		/* we allow only one expression at a time */
3641 		ret = ftrace_set_func(ftrace_graph_funcs, &ftrace_graph_count,
3642 					parser.buffer);
3643 		if (ret)
3644 			goto out_free;
3645 	}
3646 
3647 	ret = read;
3648 
3649 out_free:
3650 	trace_parser_put(&parser);
3651 out_unlock:
3652 	mutex_unlock(&graph_lock);
3653 
3654 	return ret;
3655 }
3656 
3657 static const struct file_operations ftrace_graph_fops = {
3658 	.open		= ftrace_graph_open,
3659 	.read		= seq_read,
3660 	.write		= ftrace_graph_write,
3661 	.release	= ftrace_graph_release,
3662 	.llseek		= seq_lseek,
3663 };
3664 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
3665 
3666 static __init int ftrace_init_dyn_debugfs(struct dentry *d_tracer)
3667 {
3668 
3669 	trace_create_file("available_filter_functions", 0444,
3670 			d_tracer, NULL, &ftrace_avail_fops);
3671 
3672 	trace_create_file("enabled_functions", 0444,
3673 			d_tracer, NULL, &ftrace_enabled_fops);
3674 
3675 	trace_create_file("set_ftrace_filter", 0644, d_tracer,
3676 			NULL, &ftrace_filter_fops);
3677 
3678 	trace_create_file("set_ftrace_notrace", 0644, d_tracer,
3679 				    NULL, &ftrace_notrace_fops);
3680 
3681 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3682 	trace_create_file("set_graph_function", 0444, d_tracer,
3683 				    NULL,
3684 				    &ftrace_graph_fops);
3685 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
3686 
3687 	return 0;
3688 }
3689 
3690 static void ftrace_swap_recs(void *a, void *b, int size)
3691 {
3692 	struct dyn_ftrace *reca = a;
3693 	struct dyn_ftrace *recb = b;
3694 	struct dyn_ftrace t;
3695 
3696 	t = *reca;
3697 	*reca = *recb;
3698 	*recb = t;
3699 }
3700 
3701 static int ftrace_process_locs(struct module *mod,
3702 			       unsigned long *start,
3703 			       unsigned long *end)
3704 {
3705 	struct ftrace_page *pg;
3706 	unsigned long count;
3707 	unsigned long *p;
3708 	unsigned long addr;
3709 	unsigned long flags = 0; /* Shut up gcc */
3710 	int ret = -ENOMEM;
3711 
3712 	count = end - start;
3713 
3714 	if (!count)
3715 		return 0;
3716 
3717 	pg = ftrace_allocate_pages(count);
3718 	if (!pg)
3719 		return -ENOMEM;
3720 
3721 	mutex_lock(&ftrace_lock);
3722 
3723 	/*
3724 	 * Core and each module needs their own pages, as
3725 	 * modules will free them when they are removed.
3726 	 * Force a new page to be allocated for modules.
3727 	 */
3728 	if (!mod) {
3729 		WARN_ON(ftrace_pages || ftrace_pages_start);
3730 		/* First initialization */
3731 		ftrace_pages = ftrace_pages_start = pg;
3732 	} else {
3733 		if (!ftrace_pages)
3734 			goto out;
3735 
3736 		if (WARN_ON(ftrace_pages->next)) {
3737 			/* Hmm, we have free pages? */
3738 			while (ftrace_pages->next)
3739 				ftrace_pages = ftrace_pages->next;
3740 		}
3741 
3742 		ftrace_pages->next = pg;
3743 		ftrace_pages = pg;
3744 	}
3745 
3746 	p = start;
3747 	while (p < end) {
3748 		addr = ftrace_call_adjust(*p++);
3749 		/*
3750 		 * Some architecture linkers will pad between
3751 		 * the different mcount_loc sections of different
3752 		 * object files to satisfy alignments.
3753 		 * Skip any NULL pointers.
3754 		 */
3755 		if (!addr)
3756 			continue;
3757 		if (!ftrace_record_ip(addr))
3758 			break;
3759 	}
3760 
3761 	/* These new locations need to be initialized */
3762 	ftrace_new_pgs = pg;
3763 
3764 	/* Make each individual set of pages sorted by ips */
3765 	for (; pg; pg = pg->next)
3766 		sort(pg->records, pg->index, sizeof(struct dyn_ftrace),
3767 		     ftrace_cmp_recs, ftrace_swap_recs);
3768 
3769 	/*
3770 	 * We only need to disable interrupts on start up
3771 	 * because we are modifying code that an interrupt
3772 	 * may execute, and the modification is not atomic.
3773 	 * But for modules, nothing runs the code we modify
3774 	 * until we are finished with it, and there's no
3775 	 * reason to cause large interrupt latencies while we do it.
3776 	 */
3777 	if (!mod)
3778 		local_irq_save(flags);
3779 	ftrace_update_code(mod);
3780 	if (!mod)
3781 		local_irq_restore(flags);
3782 	ret = 0;
3783  out:
3784 	mutex_unlock(&ftrace_lock);
3785 
3786 	return ret;
3787 }
3788 
3789 #ifdef CONFIG_MODULES
3790 
3791 #define next_to_ftrace_page(p) container_of(p, struct ftrace_page, next)
3792 
3793 void ftrace_release_mod(struct module *mod)
3794 {
3795 	struct dyn_ftrace *rec;
3796 	struct ftrace_page **last_pg;
3797 	struct ftrace_page *pg;
3798 	int order;
3799 
3800 	mutex_lock(&ftrace_lock);
3801 
3802 	if (ftrace_disabled)
3803 		goto out_unlock;
3804 
3805 	/*
3806 	 * Each module has its own ftrace_pages, remove
3807 	 * them from the list.
3808 	 */
3809 	last_pg = &ftrace_pages_start;
3810 	for (pg = ftrace_pages_start; pg; pg = *last_pg) {
3811 		rec = &pg->records[0];
3812 		if (within_module_core(rec->ip, mod)) {
3813 			/*
3814 			 * As core pages are first, the first
3815 			 * page should never be a module page.
3816 			 */
3817 			if (WARN_ON(pg == ftrace_pages_start))
3818 				goto out_unlock;
3819 
3820 			/* Check if we are deleting the last page */
3821 			if (pg == ftrace_pages)
3822 				ftrace_pages = next_to_ftrace_page(last_pg);
3823 
3824 			*last_pg = pg->next;
3825 			order = get_count_order(pg->size / ENTRIES_PER_PAGE);
3826 			free_pages((unsigned long)pg->records, order);
3827 			kfree(pg);
3828 		} else
3829 			last_pg = &pg->next;
3830 	}
3831  out_unlock:
3832 	mutex_unlock(&ftrace_lock);
3833 }
3834 
3835 static void ftrace_init_module(struct module *mod,
3836 			       unsigned long *start, unsigned long *end)
3837 {
3838 	if (ftrace_disabled || start == end)
3839 		return;
3840 	ftrace_process_locs(mod, start, end);
3841 }
3842 
3843 static int ftrace_module_notify(struct notifier_block *self,
3844 				unsigned long val, void *data)
3845 {
3846 	struct module *mod = data;
3847 
3848 	switch (val) {
3849 	case MODULE_STATE_COMING:
3850 		ftrace_init_module(mod, mod->ftrace_callsites,
3851 				   mod->ftrace_callsites +
3852 				   mod->num_ftrace_callsites);
3853 		break;
3854 	case MODULE_STATE_GOING:
3855 		ftrace_release_mod(mod);
3856 		break;
3857 	}
3858 
3859 	return 0;
3860 }
3861 #else
3862 static int ftrace_module_notify(struct notifier_block *self,
3863 				unsigned long val, void *data)
3864 {
3865 	return 0;
3866 }
3867 #endif /* CONFIG_MODULES */
3868 
3869 struct notifier_block ftrace_module_nb = {
3870 	.notifier_call = ftrace_module_notify,
3871 	.priority = 0,
3872 };
3873 
3874 extern unsigned long __start_mcount_loc[];
3875 extern unsigned long __stop_mcount_loc[];
3876 
3877 void __init ftrace_init(void)
3878 {
3879 	unsigned long count, addr, flags;
3880 	int ret;
3881 
3882 	/* Keep the ftrace pointer to the stub */
3883 	addr = (unsigned long)ftrace_stub;
3884 
3885 	local_irq_save(flags);
3886 	ftrace_dyn_arch_init(&addr);
3887 	local_irq_restore(flags);
3888 
3889 	/* ftrace_dyn_arch_init places the return code in addr */
3890 	if (addr)
3891 		goto failed;
3892 
3893 	count = __stop_mcount_loc - __start_mcount_loc;
3894 
3895 	ret = ftrace_dyn_table_alloc(count);
3896 	if (ret)
3897 		goto failed;
3898 
3899 	last_ftrace_enabled = ftrace_enabled = 1;
3900 
3901 	ret = ftrace_process_locs(NULL,
3902 				  __start_mcount_loc,
3903 				  __stop_mcount_loc);
3904 
3905 	ret = register_module_notifier(&ftrace_module_nb);
3906 	if (ret)
3907 		pr_warning("Failed to register trace ftrace module notifier\n");
3908 
3909 	set_ftrace_early_filters();
3910 
3911 	return;
3912  failed:
3913 	ftrace_disabled = 1;
3914 }
3915 
3916 #else
3917 
3918 static struct ftrace_ops global_ops = {
3919 	.func			= ftrace_stub,
3920 };
3921 
3922 static int __init ftrace_nodyn_init(void)
3923 {
3924 	ftrace_enabled = 1;
3925 	return 0;
3926 }
3927 device_initcall(ftrace_nodyn_init);
3928 
3929 static inline int ftrace_init_dyn_debugfs(struct dentry *d_tracer) { return 0; }
3930 static inline void ftrace_startup_enable(int command) { }
3931 /* Keep as macros so we do not need to define the commands */
3932 # define ftrace_startup(ops, command)			\
3933 	({						\
3934 		(ops)->flags |= FTRACE_OPS_FL_ENABLED;	\
3935 		0;					\
3936 	})
3937 # define ftrace_shutdown(ops, command)	do { } while (0)
3938 # define ftrace_startup_sysctl()	do { } while (0)
3939 # define ftrace_shutdown_sysctl()	do { } while (0)
3940 
3941 static inline int
3942 ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip)
3943 {
3944 	return 1;
3945 }
3946 
3947 #endif /* CONFIG_DYNAMIC_FTRACE */
3948 
3949 static void
3950 ftrace_ops_control_func(unsigned long ip, unsigned long parent_ip)
3951 {
3952 	struct ftrace_ops *op;
3953 
3954 	if (unlikely(trace_recursion_test(TRACE_CONTROL_BIT)))
3955 		return;
3956 
3957 	/*
3958 	 * Some of the ops may be dynamically allocated,
3959 	 * they must be freed after a synchronize_sched().
3960 	 */
3961 	preempt_disable_notrace();
3962 	trace_recursion_set(TRACE_CONTROL_BIT);
3963 	op = rcu_dereference_raw(ftrace_control_list);
3964 	while (op != &ftrace_list_end) {
3965 		if (!ftrace_function_local_disabled(op) &&
3966 		    ftrace_ops_test(op, ip))
3967 			op->func(ip, parent_ip);
3968 
3969 		op = rcu_dereference_raw(op->next);
3970 	};
3971 	trace_recursion_clear(TRACE_CONTROL_BIT);
3972 	preempt_enable_notrace();
3973 }
3974 
3975 static struct ftrace_ops control_ops = {
3976 	.func = ftrace_ops_control_func,
3977 };
3978 
3979 static void
3980 ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip)
3981 {
3982 	struct ftrace_ops *op;
3983 
3984 	if (unlikely(trace_recursion_test(TRACE_INTERNAL_BIT)))
3985 		return;
3986 
3987 	trace_recursion_set(TRACE_INTERNAL_BIT);
3988 	/*
3989 	 * Some of the ops may be dynamically allocated,
3990 	 * they must be freed after a synchronize_sched().
3991 	 */
3992 	preempt_disable_notrace();
3993 	op = rcu_dereference_raw(ftrace_ops_list);
3994 	while (op != &ftrace_list_end) {
3995 		if (ftrace_ops_test(op, ip))
3996 			op->func(ip, parent_ip);
3997 		op = rcu_dereference_raw(op->next);
3998 	};
3999 	preempt_enable_notrace();
4000 	trace_recursion_clear(TRACE_INTERNAL_BIT);
4001 }
4002 
4003 static void clear_ftrace_swapper(void)
4004 {
4005 	struct task_struct *p;
4006 	int cpu;
4007 
4008 	get_online_cpus();
4009 	for_each_online_cpu(cpu) {
4010 		p = idle_task(cpu);
4011 		clear_tsk_trace_trace(p);
4012 	}
4013 	put_online_cpus();
4014 }
4015 
4016 static void set_ftrace_swapper(void)
4017 {
4018 	struct task_struct *p;
4019 	int cpu;
4020 
4021 	get_online_cpus();
4022 	for_each_online_cpu(cpu) {
4023 		p = idle_task(cpu);
4024 		set_tsk_trace_trace(p);
4025 	}
4026 	put_online_cpus();
4027 }
4028 
4029 static void clear_ftrace_pid(struct pid *pid)
4030 {
4031 	struct task_struct *p;
4032 
4033 	rcu_read_lock();
4034 	do_each_pid_task(pid, PIDTYPE_PID, p) {
4035 		clear_tsk_trace_trace(p);
4036 	} while_each_pid_task(pid, PIDTYPE_PID, p);
4037 	rcu_read_unlock();
4038 
4039 	put_pid(pid);
4040 }
4041 
4042 static void set_ftrace_pid(struct pid *pid)
4043 {
4044 	struct task_struct *p;
4045 
4046 	rcu_read_lock();
4047 	do_each_pid_task(pid, PIDTYPE_PID, p) {
4048 		set_tsk_trace_trace(p);
4049 	} while_each_pid_task(pid, PIDTYPE_PID, p);
4050 	rcu_read_unlock();
4051 }
4052 
4053 static void clear_ftrace_pid_task(struct pid *pid)
4054 {
4055 	if (pid == ftrace_swapper_pid)
4056 		clear_ftrace_swapper();
4057 	else
4058 		clear_ftrace_pid(pid);
4059 }
4060 
4061 static void set_ftrace_pid_task(struct pid *pid)
4062 {
4063 	if (pid == ftrace_swapper_pid)
4064 		set_ftrace_swapper();
4065 	else
4066 		set_ftrace_pid(pid);
4067 }
4068 
4069 static int ftrace_pid_add(int p)
4070 {
4071 	struct pid *pid;
4072 	struct ftrace_pid *fpid;
4073 	int ret = -EINVAL;
4074 
4075 	mutex_lock(&ftrace_lock);
4076 
4077 	if (!p)
4078 		pid = ftrace_swapper_pid;
4079 	else
4080 		pid = find_get_pid(p);
4081 
4082 	if (!pid)
4083 		goto out;
4084 
4085 	ret = 0;
4086 
4087 	list_for_each_entry(fpid, &ftrace_pids, list)
4088 		if (fpid->pid == pid)
4089 			goto out_put;
4090 
4091 	ret = -ENOMEM;
4092 
4093 	fpid = kmalloc(sizeof(*fpid), GFP_KERNEL);
4094 	if (!fpid)
4095 		goto out_put;
4096 
4097 	list_add(&fpid->list, &ftrace_pids);
4098 	fpid->pid = pid;
4099 
4100 	set_ftrace_pid_task(pid);
4101 
4102 	ftrace_update_pid_func();
4103 	ftrace_startup_enable(0);
4104 
4105 	mutex_unlock(&ftrace_lock);
4106 	return 0;
4107 
4108 out_put:
4109 	if (pid != ftrace_swapper_pid)
4110 		put_pid(pid);
4111 
4112 out:
4113 	mutex_unlock(&ftrace_lock);
4114 	return ret;
4115 }
4116 
4117 static void ftrace_pid_reset(void)
4118 {
4119 	struct ftrace_pid *fpid, *safe;
4120 
4121 	mutex_lock(&ftrace_lock);
4122 	list_for_each_entry_safe(fpid, safe, &ftrace_pids, list) {
4123 		struct pid *pid = fpid->pid;
4124 
4125 		clear_ftrace_pid_task(pid);
4126 
4127 		list_del(&fpid->list);
4128 		kfree(fpid);
4129 	}
4130 
4131 	ftrace_update_pid_func();
4132 	ftrace_startup_enable(0);
4133 
4134 	mutex_unlock(&ftrace_lock);
4135 }
4136 
4137 static void *fpid_start(struct seq_file *m, loff_t *pos)
4138 {
4139 	mutex_lock(&ftrace_lock);
4140 
4141 	if (list_empty(&ftrace_pids) && (!*pos))
4142 		return (void *) 1;
4143 
4144 	return seq_list_start(&ftrace_pids, *pos);
4145 }
4146 
4147 static void *fpid_next(struct seq_file *m, void *v, loff_t *pos)
4148 {
4149 	if (v == (void *)1)
4150 		return NULL;
4151 
4152 	return seq_list_next(v, &ftrace_pids, pos);
4153 }
4154 
4155 static void fpid_stop(struct seq_file *m, void *p)
4156 {
4157 	mutex_unlock(&ftrace_lock);
4158 }
4159 
4160 static int fpid_show(struct seq_file *m, void *v)
4161 {
4162 	const struct ftrace_pid *fpid = list_entry(v, struct ftrace_pid, list);
4163 
4164 	if (v == (void *)1) {
4165 		seq_printf(m, "no pid\n");
4166 		return 0;
4167 	}
4168 
4169 	if (fpid->pid == ftrace_swapper_pid)
4170 		seq_printf(m, "swapper tasks\n");
4171 	else
4172 		seq_printf(m, "%u\n", pid_vnr(fpid->pid));
4173 
4174 	return 0;
4175 }
4176 
4177 static const struct seq_operations ftrace_pid_sops = {
4178 	.start = fpid_start,
4179 	.next = fpid_next,
4180 	.stop = fpid_stop,
4181 	.show = fpid_show,
4182 };
4183 
4184 static int
4185 ftrace_pid_open(struct inode *inode, struct file *file)
4186 {
4187 	int ret = 0;
4188 
4189 	if ((file->f_mode & FMODE_WRITE) &&
4190 	    (file->f_flags & O_TRUNC))
4191 		ftrace_pid_reset();
4192 
4193 	if (file->f_mode & FMODE_READ)
4194 		ret = seq_open(file, &ftrace_pid_sops);
4195 
4196 	return ret;
4197 }
4198 
4199 static ssize_t
4200 ftrace_pid_write(struct file *filp, const char __user *ubuf,
4201 		   size_t cnt, loff_t *ppos)
4202 {
4203 	char buf[64], *tmp;
4204 	long val;
4205 	int ret;
4206 
4207 	if (cnt >= sizeof(buf))
4208 		return -EINVAL;
4209 
4210 	if (copy_from_user(&buf, ubuf, cnt))
4211 		return -EFAULT;
4212 
4213 	buf[cnt] = 0;
4214 
4215 	/*
4216 	 * Allow "echo > set_ftrace_pid" or "echo -n '' > set_ftrace_pid"
4217 	 * to clean the filter quietly.
4218 	 */
4219 	tmp = strstrip(buf);
4220 	if (strlen(tmp) == 0)
4221 		return 1;
4222 
4223 	ret = strict_strtol(tmp, 10, &val);
4224 	if (ret < 0)
4225 		return ret;
4226 
4227 	ret = ftrace_pid_add(val);
4228 
4229 	return ret ? ret : cnt;
4230 }
4231 
4232 static int
4233 ftrace_pid_release(struct inode *inode, struct file *file)
4234 {
4235 	if (file->f_mode & FMODE_READ)
4236 		seq_release(inode, file);
4237 
4238 	return 0;
4239 }
4240 
4241 static const struct file_operations ftrace_pid_fops = {
4242 	.open		= ftrace_pid_open,
4243 	.write		= ftrace_pid_write,
4244 	.read		= seq_read,
4245 	.llseek		= seq_lseek,
4246 	.release	= ftrace_pid_release,
4247 };
4248 
4249 static __init int ftrace_init_debugfs(void)
4250 {
4251 	struct dentry *d_tracer;
4252 
4253 	d_tracer = tracing_init_dentry();
4254 	if (!d_tracer)
4255 		return 0;
4256 
4257 	ftrace_init_dyn_debugfs(d_tracer);
4258 
4259 	trace_create_file("set_ftrace_pid", 0644, d_tracer,
4260 			    NULL, &ftrace_pid_fops);
4261 
4262 	ftrace_profile_debugfs(d_tracer);
4263 
4264 	return 0;
4265 }
4266 fs_initcall(ftrace_init_debugfs);
4267 
4268 /**
4269  * ftrace_kill - kill ftrace
4270  *
4271  * This function should be used by panic code. It stops ftrace
4272  * but in a not so nice way. If you need to simply kill ftrace
4273  * from a non-atomic section, use ftrace_kill.
4274  */
4275 void ftrace_kill(void)
4276 {
4277 	ftrace_disabled = 1;
4278 	ftrace_enabled = 0;
4279 	clear_ftrace_function();
4280 }
4281 
4282 /**
4283  * Test if ftrace is dead or not.
4284  */
4285 int ftrace_is_dead(void)
4286 {
4287 	return ftrace_disabled;
4288 }
4289 
4290 /**
4291  * register_ftrace_function - register a function for profiling
4292  * @ops - ops structure that holds the function for profiling.
4293  *
4294  * Register a function to be called by all functions in the
4295  * kernel.
4296  *
4297  * Note: @ops->func and all the functions it calls must be labeled
4298  *       with "notrace", otherwise it will go into a
4299  *       recursive loop.
4300  */
4301 int register_ftrace_function(struct ftrace_ops *ops)
4302 {
4303 	int ret = -1;
4304 
4305 	mutex_lock(&ftrace_lock);
4306 
4307 	if (unlikely(ftrace_disabled))
4308 		goto out_unlock;
4309 
4310 	ret = __register_ftrace_function(ops);
4311 	if (!ret)
4312 		ret = ftrace_startup(ops, 0);
4313 
4314 
4315  out_unlock:
4316 	mutex_unlock(&ftrace_lock);
4317 	return ret;
4318 }
4319 EXPORT_SYMBOL_GPL(register_ftrace_function);
4320 
4321 /**
4322  * unregister_ftrace_function - unregister a function for profiling.
4323  * @ops - ops structure that holds the function to unregister
4324  *
4325  * Unregister a function that was added to be called by ftrace profiling.
4326  */
4327 int unregister_ftrace_function(struct ftrace_ops *ops)
4328 {
4329 	int ret;
4330 
4331 	mutex_lock(&ftrace_lock);
4332 	ret = __unregister_ftrace_function(ops);
4333 	if (!ret)
4334 		ftrace_shutdown(ops, 0);
4335 	mutex_unlock(&ftrace_lock);
4336 
4337 	return ret;
4338 }
4339 EXPORT_SYMBOL_GPL(unregister_ftrace_function);
4340 
4341 int
4342 ftrace_enable_sysctl(struct ctl_table *table, int write,
4343 		     void __user *buffer, size_t *lenp,
4344 		     loff_t *ppos)
4345 {
4346 	int ret = -ENODEV;
4347 
4348 	mutex_lock(&ftrace_lock);
4349 
4350 	if (unlikely(ftrace_disabled))
4351 		goto out;
4352 
4353 	ret = proc_dointvec(table, write, buffer, lenp, ppos);
4354 
4355 	if (ret || !write || (last_ftrace_enabled == !!ftrace_enabled))
4356 		goto out;
4357 
4358 	last_ftrace_enabled = !!ftrace_enabled;
4359 
4360 	if (ftrace_enabled) {
4361 
4362 		ftrace_startup_sysctl();
4363 
4364 		/* we are starting ftrace again */
4365 		if (ftrace_ops_list != &ftrace_list_end) {
4366 			if (ftrace_ops_list->next == &ftrace_list_end)
4367 				ftrace_trace_function = ftrace_ops_list->func;
4368 			else
4369 				ftrace_trace_function = ftrace_ops_list_func;
4370 		}
4371 
4372 	} else {
4373 		/* stopping ftrace calls (just send to ftrace_stub) */
4374 		ftrace_trace_function = ftrace_stub;
4375 
4376 		ftrace_shutdown_sysctl();
4377 	}
4378 
4379  out:
4380 	mutex_unlock(&ftrace_lock);
4381 	return ret;
4382 }
4383 
4384 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
4385 
4386 static int ftrace_graph_active;
4387 static struct notifier_block ftrace_suspend_notifier;
4388 
4389 int ftrace_graph_entry_stub(struct ftrace_graph_ent *trace)
4390 {
4391 	return 0;
4392 }
4393 
4394 /* The callbacks that hook a function */
4395 trace_func_graph_ret_t ftrace_graph_return =
4396 			(trace_func_graph_ret_t)ftrace_stub;
4397 trace_func_graph_ent_t ftrace_graph_entry = ftrace_graph_entry_stub;
4398 
4399 /* Try to assign a return stack array on FTRACE_RETSTACK_ALLOC_SIZE tasks. */
4400 static int alloc_retstack_tasklist(struct ftrace_ret_stack **ret_stack_list)
4401 {
4402 	int i;
4403 	int ret = 0;
4404 	unsigned long flags;
4405 	int start = 0, end = FTRACE_RETSTACK_ALLOC_SIZE;
4406 	struct task_struct *g, *t;
4407 
4408 	for (i = 0; i < FTRACE_RETSTACK_ALLOC_SIZE; i++) {
4409 		ret_stack_list[i] = kmalloc(FTRACE_RETFUNC_DEPTH
4410 					* sizeof(struct ftrace_ret_stack),
4411 					GFP_KERNEL);
4412 		if (!ret_stack_list[i]) {
4413 			start = 0;
4414 			end = i;
4415 			ret = -ENOMEM;
4416 			goto free;
4417 		}
4418 	}
4419 
4420 	read_lock_irqsave(&tasklist_lock, flags);
4421 	do_each_thread(g, t) {
4422 		if (start == end) {
4423 			ret = -EAGAIN;
4424 			goto unlock;
4425 		}
4426 
4427 		if (t->ret_stack == NULL) {
4428 			atomic_set(&t->tracing_graph_pause, 0);
4429 			atomic_set(&t->trace_overrun, 0);
4430 			t->curr_ret_stack = -1;
4431 			/* Make sure the tasks see the -1 first: */
4432 			smp_wmb();
4433 			t->ret_stack = ret_stack_list[start++];
4434 		}
4435 	} while_each_thread(g, t);
4436 
4437 unlock:
4438 	read_unlock_irqrestore(&tasklist_lock, flags);
4439 free:
4440 	for (i = start; i < end; i++)
4441 		kfree(ret_stack_list[i]);
4442 	return ret;
4443 }
4444 
4445 static void
4446 ftrace_graph_probe_sched_switch(void *ignore,
4447 			struct task_struct *prev, struct task_struct *next)
4448 {
4449 	unsigned long long timestamp;
4450 	int index;
4451 
4452 	/*
4453 	 * Does the user want to count the time a function was asleep.
4454 	 * If so, do not update the time stamps.
4455 	 */
4456 	if (trace_flags & TRACE_ITER_SLEEP_TIME)
4457 		return;
4458 
4459 	timestamp = trace_clock_local();
4460 
4461 	prev->ftrace_timestamp = timestamp;
4462 
4463 	/* only process tasks that we timestamped */
4464 	if (!next->ftrace_timestamp)
4465 		return;
4466 
4467 	/*
4468 	 * Update all the counters in next to make up for the
4469 	 * time next was sleeping.
4470 	 */
4471 	timestamp -= next->ftrace_timestamp;
4472 
4473 	for (index = next->curr_ret_stack; index >= 0; index--)
4474 		next->ret_stack[index].calltime += timestamp;
4475 }
4476 
4477 /* Allocate a return stack for each task */
4478 static int start_graph_tracing(void)
4479 {
4480 	struct ftrace_ret_stack **ret_stack_list;
4481 	int ret, cpu;
4482 
4483 	ret_stack_list = kmalloc(FTRACE_RETSTACK_ALLOC_SIZE *
4484 				sizeof(struct ftrace_ret_stack *),
4485 				GFP_KERNEL);
4486 
4487 	if (!ret_stack_list)
4488 		return -ENOMEM;
4489 
4490 	/* The cpu_boot init_task->ret_stack will never be freed */
4491 	for_each_online_cpu(cpu) {
4492 		if (!idle_task(cpu)->ret_stack)
4493 			ftrace_graph_init_idle_task(idle_task(cpu), cpu);
4494 	}
4495 
4496 	do {
4497 		ret = alloc_retstack_tasklist(ret_stack_list);
4498 	} while (ret == -EAGAIN);
4499 
4500 	if (!ret) {
4501 		ret = register_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
4502 		if (ret)
4503 			pr_info("ftrace_graph: Couldn't activate tracepoint"
4504 				" probe to kernel_sched_switch\n");
4505 	}
4506 
4507 	kfree(ret_stack_list);
4508 	return ret;
4509 }
4510 
4511 /*
4512  * Hibernation protection.
4513  * The state of the current task is too much unstable during
4514  * suspend/restore to disk. We want to protect against that.
4515  */
4516 static int
4517 ftrace_suspend_notifier_call(struct notifier_block *bl, unsigned long state,
4518 							void *unused)
4519 {
4520 	switch (state) {
4521 	case PM_HIBERNATION_PREPARE:
4522 		pause_graph_tracing();
4523 		break;
4524 
4525 	case PM_POST_HIBERNATION:
4526 		unpause_graph_tracing();
4527 		break;
4528 	}
4529 	return NOTIFY_DONE;
4530 }
4531 
4532 int register_ftrace_graph(trace_func_graph_ret_t retfunc,
4533 			trace_func_graph_ent_t entryfunc)
4534 {
4535 	int ret = 0;
4536 
4537 	mutex_lock(&ftrace_lock);
4538 
4539 	/* we currently allow only one tracer registered at a time */
4540 	if (ftrace_graph_active) {
4541 		ret = -EBUSY;
4542 		goto out;
4543 	}
4544 
4545 	ftrace_suspend_notifier.notifier_call = ftrace_suspend_notifier_call;
4546 	register_pm_notifier(&ftrace_suspend_notifier);
4547 
4548 	ftrace_graph_active++;
4549 	ret = start_graph_tracing();
4550 	if (ret) {
4551 		ftrace_graph_active--;
4552 		goto out;
4553 	}
4554 
4555 	ftrace_graph_return = retfunc;
4556 	ftrace_graph_entry = entryfunc;
4557 
4558 	ret = ftrace_startup(&global_ops, FTRACE_START_FUNC_RET);
4559 
4560 out:
4561 	mutex_unlock(&ftrace_lock);
4562 	return ret;
4563 }
4564 
4565 void unregister_ftrace_graph(void)
4566 {
4567 	mutex_lock(&ftrace_lock);
4568 
4569 	if (unlikely(!ftrace_graph_active))
4570 		goto out;
4571 
4572 	ftrace_graph_active--;
4573 	ftrace_graph_return = (trace_func_graph_ret_t)ftrace_stub;
4574 	ftrace_graph_entry = ftrace_graph_entry_stub;
4575 	ftrace_shutdown(&global_ops, FTRACE_STOP_FUNC_RET);
4576 	unregister_pm_notifier(&ftrace_suspend_notifier);
4577 	unregister_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
4578 
4579  out:
4580 	mutex_unlock(&ftrace_lock);
4581 }
4582 
4583 static DEFINE_PER_CPU(struct ftrace_ret_stack *, idle_ret_stack);
4584 
4585 static void
4586 graph_init_task(struct task_struct *t, struct ftrace_ret_stack *ret_stack)
4587 {
4588 	atomic_set(&t->tracing_graph_pause, 0);
4589 	atomic_set(&t->trace_overrun, 0);
4590 	t->ftrace_timestamp = 0;
4591 	/* make curr_ret_stack visible before we add the ret_stack */
4592 	smp_wmb();
4593 	t->ret_stack = ret_stack;
4594 }
4595 
4596 /*
4597  * Allocate a return stack for the idle task. May be the first
4598  * time through, or it may be done by CPU hotplug online.
4599  */
4600 void ftrace_graph_init_idle_task(struct task_struct *t, int cpu)
4601 {
4602 	t->curr_ret_stack = -1;
4603 	/*
4604 	 * The idle task has no parent, it either has its own
4605 	 * stack or no stack at all.
4606 	 */
4607 	if (t->ret_stack)
4608 		WARN_ON(t->ret_stack != per_cpu(idle_ret_stack, cpu));
4609 
4610 	if (ftrace_graph_active) {
4611 		struct ftrace_ret_stack *ret_stack;
4612 
4613 		ret_stack = per_cpu(idle_ret_stack, cpu);
4614 		if (!ret_stack) {
4615 			ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
4616 					    * sizeof(struct ftrace_ret_stack),
4617 					    GFP_KERNEL);
4618 			if (!ret_stack)
4619 				return;
4620 			per_cpu(idle_ret_stack, cpu) = ret_stack;
4621 		}
4622 		graph_init_task(t, ret_stack);
4623 	}
4624 }
4625 
4626 /* Allocate a return stack for newly created task */
4627 void ftrace_graph_init_task(struct task_struct *t)
4628 {
4629 	/* Make sure we do not use the parent ret_stack */
4630 	t->ret_stack = NULL;
4631 	t->curr_ret_stack = -1;
4632 
4633 	if (ftrace_graph_active) {
4634 		struct ftrace_ret_stack *ret_stack;
4635 
4636 		ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
4637 				* sizeof(struct ftrace_ret_stack),
4638 				GFP_KERNEL);
4639 		if (!ret_stack)
4640 			return;
4641 		graph_init_task(t, ret_stack);
4642 	}
4643 }
4644 
4645 void ftrace_graph_exit_task(struct task_struct *t)
4646 {
4647 	struct ftrace_ret_stack	*ret_stack = t->ret_stack;
4648 
4649 	t->ret_stack = NULL;
4650 	/* NULL must become visible to IRQs before we free it: */
4651 	barrier();
4652 
4653 	kfree(ret_stack);
4654 }
4655 
4656 void ftrace_graph_stop(void)
4657 {
4658 	ftrace_stop();
4659 }
4660 #endif
4661