xref: /linux/kernel/trace/ftrace.c (revision 0d456bad36d42d16022be045c8a53ddbb59ee478)
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 Nadia Yvette Chambers
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 static struct ftrace_ops ftrace_list_end __read_mostly = {
68 	.func		= ftrace_stub,
69 	.flags		= FTRACE_OPS_FL_RECURSION_SAFE,
70 };
71 
72 /* ftrace_enabled is a method to turn ftrace on or off */
73 int ftrace_enabled __read_mostly;
74 static int last_ftrace_enabled;
75 
76 /* Quick disabling of function tracer. */
77 int function_trace_stop __read_mostly;
78 
79 /* Current function tracing op */
80 struct ftrace_ops *function_trace_op __read_mostly = &ftrace_list_end;
81 
82 /* List for set_ftrace_pid's pids. */
83 LIST_HEAD(ftrace_pids);
84 struct ftrace_pid {
85 	struct list_head list;
86 	struct pid *pid;
87 };
88 
89 /*
90  * ftrace_disabled is set when an anomaly is discovered.
91  * ftrace_disabled is much stronger than ftrace_enabled.
92  */
93 static int ftrace_disabled __read_mostly;
94 
95 static DEFINE_MUTEX(ftrace_lock);
96 
97 static struct ftrace_ops *ftrace_global_list __read_mostly = &ftrace_list_end;
98 static struct ftrace_ops *ftrace_control_list __read_mostly = &ftrace_list_end;
99 static struct ftrace_ops *ftrace_ops_list __read_mostly = &ftrace_list_end;
100 ftrace_func_t ftrace_trace_function __read_mostly = ftrace_stub;
101 ftrace_func_t ftrace_pid_function __read_mostly = ftrace_stub;
102 static struct ftrace_ops global_ops;
103 static struct ftrace_ops control_ops;
104 
105 #if ARCH_SUPPORTS_FTRACE_OPS
106 static void ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
107 				 struct ftrace_ops *op, struct pt_regs *regs);
108 #else
109 /* See comment below, where ftrace_ops_list_func is defined */
110 static void ftrace_ops_no_ops(unsigned long ip, unsigned long parent_ip);
111 #define ftrace_ops_list_func ((ftrace_func_t)ftrace_ops_no_ops)
112 #endif
113 
114 /**
115  * ftrace_nr_registered_ops - return number of ops registered
116  *
117  * Returns the number of ftrace_ops registered and tracing functions
118  */
119 int ftrace_nr_registered_ops(void)
120 {
121 	struct ftrace_ops *ops;
122 	int cnt = 0;
123 
124 	mutex_lock(&ftrace_lock);
125 
126 	for (ops = ftrace_ops_list;
127 	     ops != &ftrace_list_end; ops = ops->next)
128 		cnt++;
129 
130 	mutex_unlock(&ftrace_lock);
131 
132 	return cnt;
133 }
134 
135 /*
136  * Traverse the ftrace_global_list, invoking all entries.  The reason that we
137  * can use rcu_dereference_raw() is that elements removed from this list
138  * are simply leaked, so there is no need to interact with a grace-period
139  * mechanism.  The rcu_dereference_raw() calls are needed to handle
140  * concurrent insertions into the ftrace_global_list.
141  *
142  * Silly Alpha and silly pointer-speculation compiler optimizations!
143  */
144 static void
145 ftrace_global_list_func(unsigned long ip, unsigned long parent_ip,
146 			struct ftrace_ops *op, struct pt_regs *regs)
147 {
148 	if (unlikely(trace_recursion_test(TRACE_GLOBAL_BIT)))
149 		return;
150 
151 	trace_recursion_set(TRACE_GLOBAL_BIT);
152 	op = rcu_dereference_raw(ftrace_global_list); /*see above*/
153 	while (op != &ftrace_list_end) {
154 		op->func(ip, parent_ip, op, regs);
155 		op = rcu_dereference_raw(op->next); /*see above*/
156 	};
157 	trace_recursion_clear(TRACE_GLOBAL_BIT);
158 }
159 
160 static void ftrace_pid_func(unsigned long ip, unsigned long parent_ip,
161 			    struct ftrace_ops *op, struct pt_regs *regs)
162 {
163 	if (!test_tsk_trace_trace(current))
164 		return;
165 
166 	ftrace_pid_function(ip, parent_ip, op, regs);
167 }
168 
169 static void set_ftrace_pid_function(ftrace_func_t func)
170 {
171 	/* do not set ftrace_pid_function to itself! */
172 	if (func != ftrace_pid_func)
173 		ftrace_pid_function = func;
174 }
175 
176 /**
177  * clear_ftrace_function - reset the ftrace function
178  *
179  * This NULLs the ftrace function and in essence stops
180  * tracing.  There may be lag
181  */
182 void clear_ftrace_function(void)
183 {
184 	ftrace_trace_function = ftrace_stub;
185 	ftrace_pid_function = ftrace_stub;
186 }
187 
188 static void control_ops_disable_all(struct ftrace_ops *ops)
189 {
190 	int cpu;
191 
192 	for_each_possible_cpu(cpu)
193 		*per_cpu_ptr(ops->disabled, cpu) = 1;
194 }
195 
196 static int control_ops_alloc(struct ftrace_ops *ops)
197 {
198 	int __percpu *disabled;
199 
200 	disabled = alloc_percpu(int);
201 	if (!disabled)
202 		return -ENOMEM;
203 
204 	ops->disabled = disabled;
205 	control_ops_disable_all(ops);
206 	return 0;
207 }
208 
209 static void control_ops_free(struct ftrace_ops *ops)
210 {
211 	free_percpu(ops->disabled);
212 }
213 
214 static void update_global_ops(void)
215 {
216 	ftrace_func_t func;
217 
218 	/*
219 	 * If there's only one function registered, then call that
220 	 * function directly. Otherwise, we need to iterate over the
221 	 * registered callers.
222 	 */
223 	if (ftrace_global_list == &ftrace_list_end ||
224 	    ftrace_global_list->next == &ftrace_list_end)
225 		func = ftrace_global_list->func;
226 	else
227 		func = ftrace_global_list_func;
228 
229 	/* If we filter on pids, update to use the pid function */
230 	if (!list_empty(&ftrace_pids)) {
231 		set_ftrace_pid_function(func);
232 		func = ftrace_pid_func;
233 	}
234 
235 	global_ops.func = func;
236 }
237 
238 static void update_ftrace_function(void)
239 {
240 	ftrace_func_t func;
241 
242 	update_global_ops();
243 
244 	/*
245 	 * If we are at the end of the list and this ops is
246 	 * recursion safe and not dynamic and the arch supports passing ops,
247 	 * then have the mcount trampoline call the function directly.
248 	 */
249 	if (ftrace_ops_list == &ftrace_list_end ||
250 	    (ftrace_ops_list->next == &ftrace_list_end &&
251 	     !(ftrace_ops_list->flags & FTRACE_OPS_FL_DYNAMIC) &&
252 	     (ftrace_ops_list->flags & FTRACE_OPS_FL_RECURSION_SAFE) &&
253 	     !FTRACE_FORCE_LIST_FUNC)) {
254 		/* Set the ftrace_ops that the arch callback uses */
255 		if (ftrace_ops_list == &global_ops)
256 			function_trace_op = ftrace_global_list;
257 		else
258 			function_trace_op = ftrace_ops_list;
259 		func = ftrace_ops_list->func;
260 	} else {
261 		/* Just use the default ftrace_ops */
262 		function_trace_op = &ftrace_list_end;
263 		func = ftrace_ops_list_func;
264 	}
265 
266 	ftrace_trace_function = func;
267 }
268 
269 static void add_ftrace_ops(struct ftrace_ops **list, struct ftrace_ops *ops)
270 {
271 	ops->next = *list;
272 	/*
273 	 * We are entering ops into the list but another
274 	 * CPU might be walking that list. We need to make sure
275 	 * the ops->next pointer is valid before another CPU sees
276 	 * the ops pointer included into the list.
277 	 */
278 	rcu_assign_pointer(*list, ops);
279 }
280 
281 static int remove_ftrace_ops(struct ftrace_ops **list, struct ftrace_ops *ops)
282 {
283 	struct ftrace_ops **p;
284 
285 	/*
286 	 * If we are removing the last function, then simply point
287 	 * to the ftrace_stub.
288 	 */
289 	if (*list == ops && ops->next == &ftrace_list_end) {
290 		*list = &ftrace_list_end;
291 		return 0;
292 	}
293 
294 	for (p = list; *p != &ftrace_list_end; p = &(*p)->next)
295 		if (*p == ops)
296 			break;
297 
298 	if (*p != ops)
299 		return -1;
300 
301 	*p = (*p)->next;
302 	return 0;
303 }
304 
305 static void add_ftrace_list_ops(struct ftrace_ops **list,
306 				struct ftrace_ops *main_ops,
307 				struct ftrace_ops *ops)
308 {
309 	int first = *list == &ftrace_list_end;
310 	add_ftrace_ops(list, ops);
311 	if (first)
312 		add_ftrace_ops(&ftrace_ops_list, main_ops);
313 }
314 
315 static int remove_ftrace_list_ops(struct ftrace_ops **list,
316 				  struct ftrace_ops *main_ops,
317 				  struct ftrace_ops *ops)
318 {
319 	int ret = remove_ftrace_ops(list, ops);
320 	if (!ret && *list == &ftrace_list_end)
321 		ret = remove_ftrace_ops(&ftrace_ops_list, main_ops);
322 	return ret;
323 }
324 
325 static int __register_ftrace_function(struct ftrace_ops *ops)
326 {
327 	if (unlikely(ftrace_disabled))
328 		return -ENODEV;
329 
330 	if (FTRACE_WARN_ON(ops == &global_ops))
331 		return -EINVAL;
332 
333 	if (WARN_ON(ops->flags & FTRACE_OPS_FL_ENABLED))
334 		return -EBUSY;
335 
336 	/* We don't support both control and global flags set. */
337 	if ((ops->flags & FL_GLOBAL_CONTROL_MASK) == FL_GLOBAL_CONTROL_MASK)
338 		return -EINVAL;
339 
340 #ifndef ARCH_SUPPORTS_FTRACE_SAVE_REGS
341 	/*
342 	 * If the ftrace_ops specifies SAVE_REGS, then it only can be used
343 	 * if the arch supports it, or SAVE_REGS_IF_SUPPORTED is also set.
344 	 * Setting SAVE_REGS_IF_SUPPORTED makes SAVE_REGS irrelevant.
345 	 */
346 	if (ops->flags & FTRACE_OPS_FL_SAVE_REGS &&
347 	    !(ops->flags & FTRACE_OPS_FL_SAVE_REGS_IF_SUPPORTED))
348 		return -EINVAL;
349 
350 	if (ops->flags & FTRACE_OPS_FL_SAVE_REGS_IF_SUPPORTED)
351 		ops->flags |= FTRACE_OPS_FL_SAVE_REGS;
352 #endif
353 
354 	if (!core_kernel_data((unsigned long)ops))
355 		ops->flags |= FTRACE_OPS_FL_DYNAMIC;
356 
357 	if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
358 		add_ftrace_list_ops(&ftrace_global_list, &global_ops, ops);
359 		ops->flags |= FTRACE_OPS_FL_ENABLED;
360 	} else if (ops->flags & FTRACE_OPS_FL_CONTROL) {
361 		if (control_ops_alloc(ops))
362 			return -ENOMEM;
363 		add_ftrace_list_ops(&ftrace_control_list, &control_ops, ops);
364 	} else
365 		add_ftrace_ops(&ftrace_ops_list, ops);
366 
367 	if (ftrace_enabled)
368 		update_ftrace_function();
369 
370 	return 0;
371 }
372 
373 static int __unregister_ftrace_function(struct ftrace_ops *ops)
374 {
375 	int ret;
376 
377 	if (ftrace_disabled)
378 		return -ENODEV;
379 
380 	if (WARN_ON(!(ops->flags & FTRACE_OPS_FL_ENABLED)))
381 		return -EBUSY;
382 
383 	if (FTRACE_WARN_ON(ops == &global_ops))
384 		return -EINVAL;
385 
386 	if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
387 		ret = remove_ftrace_list_ops(&ftrace_global_list,
388 					     &global_ops, ops);
389 		if (!ret)
390 			ops->flags &= ~FTRACE_OPS_FL_ENABLED;
391 	} else if (ops->flags & FTRACE_OPS_FL_CONTROL) {
392 		ret = remove_ftrace_list_ops(&ftrace_control_list,
393 					     &control_ops, ops);
394 		if (!ret) {
395 			/*
396 			 * The ftrace_ops is now removed from the list,
397 			 * so there'll be no new users. We must ensure
398 			 * all current users are done before we free
399 			 * the control data.
400 			 */
401 			synchronize_sched();
402 			control_ops_free(ops);
403 		}
404 	} else
405 		ret = remove_ftrace_ops(&ftrace_ops_list, ops);
406 
407 	if (ret < 0)
408 		return ret;
409 
410 	if (ftrace_enabled)
411 		update_ftrace_function();
412 
413 	/*
414 	 * Dynamic ops may be freed, we must make sure that all
415 	 * callers are done before leaving this function.
416 	 */
417 	if (ops->flags & FTRACE_OPS_FL_DYNAMIC)
418 		synchronize_sched();
419 
420 	return 0;
421 }
422 
423 static void ftrace_update_pid_func(void)
424 {
425 	/* Only do something if we are tracing something */
426 	if (ftrace_trace_function == ftrace_stub)
427 		return;
428 
429 	update_ftrace_function();
430 }
431 
432 #ifdef CONFIG_FUNCTION_PROFILER
433 struct ftrace_profile {
434 	struct hlist_node		node;
435 	unsigned long			ip;
436 	unsigned long			counter;
437 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
438 	unsigned long long		time;
439 	unsigned long long		time_squared;
440 #endif
441 };
442 
443 struct ftrace_profile_page {
444 	struct ftrace_profile_page	*next;
445 	unsigned long			index;
446 	struct ftrace_profile		records[];
447 };
448 
449 struct ftrace_profile_stat {
450 	atomic_t			disabled;
451 	struct hlist_head		*hash;
452 	struct ftrace_profile_page	*pages;
453 	struct ftrace_profile_page	*start;
454 	struct tracer_stat		stat;
455 };
456 
457 #define PROFILE_RECORDS_SIZE						\
458 	(PAGE_SIZE - offsetof(struct ftrace_profile_page, records))
459 
460 #define PROFILES_PER_PAGE					\
461 	(PROFILE_RECORDS_SIZE / sizeof(struct ftrace_profile))
462 
463 static int ftrace_profile_bits __read_mostly;
464 static int ftrace_profile_enabled __read_mostly;
465 
466 /* ftrace_profile_lock - synchronize the enable and disable of the profiler */
467 static DEFINE_MUTEX(ftrace_profile_lock);
468 
469 static DEFINE_PER_CPU(struct ftrace_profile_stat, ftrace_profile_stats);
470 
471 #define FTRACE_PROFILE_HASH_SIZE 1024 /* must be power of 2 */
472 
473 static void *
474 function_stat_next(void *v, int idx)
475 {
476 	struct ftrace_profile *rec = v;
477 	struct ftrace_profile_page *pg;
478 
479 	pg = (struct ftrace_profile_page *)((unsigned long)rec & PAGE_MASK);
480 
481  again:
482 	if (idx != 0)
483 		rec++;
484 
485 	if ((void *)rec >= (void *)&pg->records[pg->index]) {
486 		pg = pg->next;
487 		if (!pg)
488 			return NULL;
489 		rec = &pg->records[0];
490 		if (!rec->counter)
491 			goto again;
492 	}
493 
494 	return rec;
495 }
496 
497 static void *function_stat_start(struct tracer_stat *trace)
498 {
499 	struct ftrace_profile_stat *stat =
500 		container_of(trace, struct ftrace_profile_stat, stat);
501 
502 	if (!stat || !stat->start)
503 		return NULL;
504 
505 	return function_stat_next(&stat->start->records[0], 0);
506 }
507 
508 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
509 /* function graph compares on total time */
510 static int function_stat_cmp(void *p1, void *p2)
511 {
512 	struct ftrace_profile *a = p1;
513 	struct ftrace_profile *b = p2;
514 
515 	if (a->time < b->time)
516 		return -1;
517 	if (a->time > b->time)
518 		return 1;
519 	else
520 		return 0;
521 }
522 #else
523 /* not function graph compares against hits */
524 static int function_stat_cmp(void *p1, void *p2)
525 {
526 	struct ftrace_profile *a = p1;
527 	struct ftrace_profile *b = p2;
528 
529 	if (a->counter < b->counter)
530 		return -1;
531 	if (a->counter > b->counter)
532 		return 1;
533 	else
534 		return 0;
535 }
536 #endif
537 
538 static int function_stat_headers(struct seq_file *m)
539 {
540 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
541 	seq_printf(m, "  Function                               "
542 		   "Hit    Time            Avg             s^2\n"
543 		      "  --------                               "
544 		   "---    ----            ---             ---\n");
545 #else
546 	seq_printf(m, "  Function                               Hit\n"
547 		      "  --------                               ---\n");
548 #endif
549 	return 0;
550 }
551 
552 static int function_stat_show(struct seq_file *m, void *v)
553 {
554 	struct ftrace_profile *rec = v;
555 	char str[KSYM_SYMBOL_LEN];
556 	int ret = 0;
557 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
558 	static struct trace_seq s;
559 	unsigned long long avg;
560 	unsigned long long stddev;
561 #endif
562 	mutex_lock(&ftrace_profile_lock);
563 
564 	/* we raced with function_profile_reset() */
565 	if (unlikely(rec->counter == 0)) {
566 		ret = -EBUSY;
567 		goto out;
568 	}
569 
570 	kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
571 	seq_printf(m, "  %-30.30s  %10lu", str, rec->counter);
572 
573 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
574 	seq_printf(m, "    ");
575 	avg = rec->time;
576 	do_div(avg, rec->counter);
577 
578 	/* Sample standard deviation (s^2) */
579 	if (rec->counter <= 1)
580 		stddev = 0;
581 	else {
582 		stddev = rec->time_squared - rec->counter * avg * avg;
583 		/*
584 		 * Divide only 1000 for ns^2 -> us^2 conversion.
585 		 * trace_print_graph_duration will divide 1000 again.
586 		 */
587 		do_div(stddev, (rec->counter - 1) * 1000);
588 	}
589 
590 	trace_seq_init(&s);
591 	trace_print_graph_duration(rec->time, &s);
592 	trace_seq_puts(&s, "    ");
593 	trace_print_graph_duration(avg, &s);
594 	trace_seq_puts(&s, "    ");
595 	trace_print_graph_duration(stddev, &s);
596 	trace_print_seq(m, &s);
597 #endif
598 	seq_putc(m, '\n');
599 out:
600 	mutex_unlock(&ftrace_profile_lock);
601 
602 	return ret;
603 }
604 
605 static void ftrace_profile_reset(struct ftrace_profile_stat *stat)
606 {
607 	struct ftrace_profile_page *pg;
608 
609 	pg = stat->pages = stat->start;
610 
611 	while (pg) {
612 		memset(pg->records, 0, PROFILE_RECORDS_SIZE);
613 		pg->index = 0;
614 		pg = pg->next;
615 	}
616 
617 	memset(stat->hash, 0,
618 	       FTRACE_PROFILE_HASH_SIZE * sizeof(struct hlist_head));
619 }
620 
621 int ftrace_profile_pages_init(struct ftrace_profile_stat *stat)
622 {
623 	struct ftrace_profile_page *pg;
624 	int functions;
625 	int pages;
626 	int i;
627 
628 	/* If we already allocated, do nothing */
629 	if (stat->pages)
630 		return 0;
631 
632 	stat->pages = (void *)get_zeroed_page(GFP_KERNEL);
633 	if (!stat->pages)
634 		return -ENOMEM;
635 
636 #ifdef CONFIG_DYNAMIC_FTRACE
637 	functions = ftrace_update_tot_cnt;
638 #else
639 	/*
640 	 * We do not know the number of functions that exist because
641 	 * dynamic tracing is what counts them. With past experience
642 	 * we have around 20K functions. That should be more than enough.
643 	 * It is highly unlikely we will execute every function in
644 	 * the kernel.
645 	 */
646 	functions = 20000;
647 #endif
648 
649 	pg = stat->start = stat->pages;
650 
651 	pages = DIV_ROUND_UP(functions, PROFILES_PER_PAGE);
652 
653 	for (i = 0; i < pages; i++) {
654 		pg->next = (void *)get_zeroed_page(GFP_KERNEL);
655 		if (!pg->next)
656 			goto out_free;
657 		pg = pg->next;
658 	}
659 
660 	return 0;
661 
662  out_free:
663 	pg = stat->start;
664 	while (pg) {
665 		unsigned long tmp = (unsigned long)pg;
666 
667 		pg = pg->next;
668 		free_page(tmp);
669 	}
670 
671 	free_page((unsigned long)stat->pages);
672 	stat->pages = NULL;
673 	stat->start = NULL;
674 
675 	return -ENOMEM;
676 }
677 
678 static int ftrace_profile_init_cpu(int cpu)
679 {
680 	struct ftrace_profile_stat *stat;
681 	int size;
682 
683 	stat = &per_cpu(ftrace_profile_stats, cpu);
684 
685 	if (stat->hash) {
686 		/* If the profile is already created, simply reset it */
687 		ftrace_profile_reset(stat);
688 		return 0;
689 	}
690 
691 	/*
692 	 * We are profiling all functions, but usually only a few thousand
693 	 * functions are hit. We'll make a hash of 1024 items.
694 	 */
695 	size = FTRACE_PROFILE_HASH_SIZE;
696 
697 	stat->hash = kzalloc(sizeof(struct hlist_head) * size, GFP_KERNEL);
698 
699 	if (!stat->hash)
700 		return -ENOMEM;
701 
702 	if (!ftrace_profile_bits) {
703 		size--;
704 
705 		for (; size; size >>= 1)
706 			ftrace_profile_bits++;
707 	}
708 
709 	/* Preallocate the function profiling pages */
710 	if (ftrace_profile_pages_init(stat) < 0) {
711 		kfree(stat->hash);
712 		stat->hash = NULL;
713 		return -ENOMEM;
714 	}
715 
716 	return 0;
717 }
718 
719 static int ftrace_profile_init(void)
720 {
721 	int cpu;
722 	int ret = 0;
723 
724 	for_each_online_cpu(cpu) {
725 		ret = ftrace_profile_init_cpu(cpu);
726 		if (ret)
727 			break;
728 	}
729 
730 	return ret;
731 }
732 
733 /* interrupts must be disabled */
734 static struct ftrace_profile *
735 ftrace_find_profiled_func(struct ftrace_profile_stat *stat, unsigned long ip)
736 {
737 	struct ftrace_profile *rec;
738 	struct hlist_head *hhd;
739 	struct hlist_node *n;
740 	unsigned long key;
741 
742 	key = hash_long(ip, ftrace_profile_bits);
743 	hhd = &stat->hash[key];
744 
745 	if (hlist_empty(hhd))
746 		return NULL;
747 
748 	hlist_for_each_entry_rcu(rec, n, hhd, node) {
749 		if (rec->ip == ip)
750 			return rec;
751 	}
752 
753 	return NULL;
754 }
755 
756 static void ftrace_add_profile(struct ftrace_profile_stat *stat,
757 			       struct ftrace_profile *rec)
758 {
759 	unsigned long key;
760 
761 	key = hash_long(rec->ip, ftrace_profile_bits);
762 	hlist_add_head_rcu(&rec->node, &stat->hash[key]);
763 }
764 
765 /*
766  * The memory is already allocated, this simply finds a new record to use.
767  */
768 static struct ftrace_profile *
769 ftrace_profile_alloc(struct ftrace_profile_stat *stat, unsigned long ip)
770 {
771 	struct ftrace_profile *rec = NULL;
772 
773 	/* prevent recursion (from NMIs) */
774 	if (atomic_inc_return(&stat->disabled) != 1)
775 		goto out;
776 
777 	/*
778 	 * Try to find the function again since an NMI
779 	 * could have added it
780 	 */
781 	rec = ftrace_find_profiled_func(stat, ip);
782 	if (rec)
783 		goto out;
784 
785 	if (stat->pages->index == PROFILES_PER_PAGE) {
786 		if (!stat->pages->next)
787 			goto out;
788 		stat->pages = stat->pages->next;
789 	}
790 
791 	rec = &stat->pages->records[stat->pages->index++];
792 	rec->ip = ip;
793 	ftrace_add_profile(stat, rec);
794 
795  out:
796 	atomic_dec(&stat->disabled);
797 
798 	return rec;
799 }
800 
801 static void
802 function_profile_call(unsigned long ip, unsigned long parent_ip,
803 		      struct ftrace_ops *ops, struct pt_regs *regs)
804 {
805 	struct ftrace_profile_stat *stat;
806 	struct ftrace_profile *rec;
807 	unsigned long flags;
808 
809 	if (!ftrace_profile_enabled)
810 		return;
811 
812 	local_irq_save(flags);
813 
814 	stat = &__get_cpu_var(ftrace_profile_stats);
815 	if (!stat->hash || !ftrace_profile_enabled)
816 		goto out;
817 
818 	rec = ftrace_find_profiled_func(stat, ip);
819 	if (!rec) {
820 		rec = ftrace_profile_alloc(stat, ip);
821 		if (!rec)
822 			goto out;
823 	}
824 
825 	rec->counter++;
826  out:
827 	local_irq_restore(flags);
828 }
829 
830 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
831 static int profile_graph_entry(struct ftrace_graph_ent *trace)
832 {
833 	function_profile_call(trace->func, 0, NULL, NULL);
834 	return 1;
835 }
836 
837 static void profile_graph_return(struct ftrace_graph_ret *trace)
838 {
839 	struct ftrace_profile_stat *stat;
840 	unsigned long long calltime;
841 	struct ftrace_profile *rec;
842 	unsigned long flags;
843 
844 	local_irq_save(flags);
845 	stat = &__get_cpu_var(ftrace_profile_stats);
846 	if (!stat->hash || !ftrace_profile_enabled)
847 		goto out;
848 
849 	/* If the calltime was zero'd ignore it */
850 	if (!trace->calltime)
851 		goto out;
852 
853 	calltime = trace->rettime - trace->calltime;
854 
855 	if (!(trace_flags & TRACE_ITER_GRAPH_TIME)) {
856 		int index;
857 
858 		index = trace->depth;
859 
860 		/* Append this call time to the parent time to subtract */
861 		if (index)
862 			current->ret_stack[index - 1].subtime += calltime;
863 
864 		if (current->ret_stack[index].subtime < calltime)
865 			calltime -= current->ret_stack[index].subtime;
866 		else
867 			calltime = 0;
868 	}
869 
870 	rec = ftrace_find_profiled_func(stat, trace->func);
871 	if (rec) {
872 		rec->time += calltime;
873 		rec->time_squared += calltime * calltime;
874 	}
875 
876  out:
877 	local_irq_restore(flags);
878 }
879 
880 static int register_ftrace_profiler(void)
881 {
882 	return register_ftrace_graph(&profile_graph_return,
883 				     &profile_graph_entry);
884 }
885 
886 static void unregister_ftrace_profiler(void)
887 {
888 	unregister_ftrace_graph();
889 }
890 #else
891 static struct ftrace_ops ftrace_profile_ops __read_mostly = {
892 	.func		= function_profile_call,
893 	.flags		= FTRACE_OPS_FL_RECURSION_SAFE,
894 };
895 
896 static int register_ftrace_profiler(void)
897 {
898 	return register_ftrace_function(&ftrace_profile_ops);
899 }
900 
901 static void unregister_ftrace_profiler(void)
902 {
903 	unregister_ftrace_function(&ftrace_profile_ops);
904 }
905 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
906 
907 static ssize_t
908 ftrace_profile_write(struct file *filp, const char __user *ubuf,
909 		     size_t cnt, loff_t *ppos)
910 {
911 	unsigned long val;
912 	int ret;
913 
914 	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
915 	if (ret)
916 		return ret;
917 
918 	val = !!val;
919 
920 	mutex_lock(&ftrace_profile_lock);
921 	if (ftrace_profile_enabled ^ val) {
922 		if (val) {
923 			ret = ftrace_profile_init();
924 			if (ret < 0) {
925 				cnt = ret;
926 				goto out;
927 			}
928 
929 			ret = register_ftrace_profiler();
930 			if (ret < 0) {
931 				cnt = ret;
932 				goto out;
933 			}
934 			ftrace_profile_enabled = 1;
935 		} else {
936 			ftrace_profile_enabled = 0;
937 			/*
938 			 * unregister_ftrace_profiler calls stop_machine
939 			 * so this acts like an synchronize_sched.
940 			 */
941 			unregister_ftrace_profiler();
942 		}
943 	}
944  out:
945 	mutex_unlock(&ftrace_profile_lock);
946 
947 	*ppos += cnt;
948 
949 	return cnt;
950 }
951 
952 static ssize_t
953 ftrace_profile_read(struct file *filp, char __user *ubuf,
954 		     size_t cnt, loff_t *ppos)
955 {
956 	char buf[64];		/* big enough to hold a number */
957 	int r;
958 
959 	r = sprintf(buf, "%u\n", ftrace_profile_enabled);
960 	return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
961 }
962 
963 static const struct file_operations ftrace_profile_fops = {
964 	.open		= tracing_open_generic,
965 	.read		= ftrace_profile_read,
966 	.write		= ftrace_profile_write,
967 	.llseek		= default_llseek,
968 };
969 
970 /* used to initialize the real stat files */
971 static struct tracer_stat function_stats __initdata = {
972 	.name		= "functions",
973 	.stat_start	= function_stat_start,
974 	.stat_next	= function_stat_next,
975 	.stat_cmp	= function_stat_cmp,
976 	.stat_headers	= function_stat_headers,
977 	.stat_show	= function_stat_show
978 };
979 
980 static __init void ftrace_profile_debugfs(struct dentry *d_tracer)
981 {
982 	struct ftrace_profile_stat *stat;
983 	struct dentry *entry;
984 	char *name;
985 	int ret;
986 	int cpu;
987 
988 	for_each_possible_cpu(cpu) {
989 		stat = &per_cpu(ftrace_profile_stats, cpu);
990 
991 		/* allocate enough for function name + cpu number */
992 		name = kmalloc(32, GFP_KERNEL);
993 		if (!name) {
994 			/*
995 			 * The files created are permanent, if something happens
996 			 * we still do not free memory.
997 			 */
998 			WARN(1,
999 			     "Could not allocate stat file for cpu %d\n",
1000 			     cpu);
1001 			return;
1002 		}
1003 		stat->stat = function_stats;
1004 		snprintf(name, 32, "function%d", cpu);
1005 		stat->stat.name = name;
1006 		ret = register_stat_tracer(&stat->stat);
1007 		if (ret) {
1008 			WARN(1,
1009 			     "Could not register function stat for cpu %d\n",
1010 			     cpu);
1011 			kfree(name);
1012 			return;
1013 		}
1014 	}
1015 
1016 	entry = debugfs_create_file("function_profile_enabled", 0644,
1017 				    d_tracer, NULL, &ftrace_profile_fops);
1018 	if (!entry)
1019 		pr_warning("Could not create debugfs "
1020 			   "'function_profile_enabled' entry\n");
1021 }
1022 
1023 #else /* CONFIG_FUNCTION_PROFILER */
1024 static __init void ftrace_profile_debugfs(struct dentry *d_tracer)
1025 {
1026 }
1027 #endif /* CONFIG_FUNCTION_PROFILER */
1028 
1029 static struct pid * const ftrace_swapper_pid = &init_struct_pid;
1030 
1031 #ifdef CONFIG_DYNAMIC_FTRACE
1032 
1033 #ifndef CONFIG_FTRACE_MCOUNT_RECORD
1034 # error Dynamic ftrace depends on MCOUNT_RECORD
1035 #endif
1036 
1037 static struct hlist_head ftrace_func_hash[FTRACE_FUNC_HASHSIZE] __read_mostly;
1038 
1039 struct ftrace_func_probe {
1040 	struct hlist_node	node;
1041 	struct ftrace_probe_ops	*ops;
1042 	unsigned long		flags;
1043 	unsigned long		ip;
1044 	void			*data;
1045 	struct rcu_head		rcu;
1046 };
1047 
1048 struct ftrace_func_entry {
1049 	struct hlist_node hlist;
1050 	unsigned long ip;
1051 };
1052 
1053 struct ftrace_hash {
1054 	unsigned long		size_bits;
1055 	struct hlist_head	*buckets;
1056 	unsigned long		count;
1057 	struct rcu_head		rcu;
1058 };
1059 
1060 /*
1061  * We make these constant because no one should touch them,
1062  * but they are used as the default "empty hash", to avoid allocating
1063  * it all the time. These are in a read only section such that if
1064  * anyone does try to modify it, it will cause an exception.
1065  */
1066 static const struct hlist_head empty_buckets[1];
1067 static const struct ftrace_hash empty_hash = {
1068 	.buckets = (struct hlist_head *)empty_buckets,
1069 };
1070 #define EMPTY_HASH	((struct ftrace_hash *)&empty_hash)
1071 
1072 static struct ftrace_ops global_ops = {
1073 	.func			= ftrace_stub,
1074 	.notrace_hash		= EMPTY_HASH,
1075 	.filter_hash		= EMPTY_HASH,
1076 	.flags			= FTRACE_OPS_FL_RECURSION_SAFE,
1077 };
1078 
1079 static DEFINE_MUTEX(ftrace_regex_lock);
1080 
1081 struct ftrace_page {
1082 	struct ftrace_page	*next;
1083 	struct dyn_ftrace	*records;
1084 	int			index;
1085 	int			size;
1086 };
1087 
1088 static struct ftrace_page *ftrace_new_pgs;
1089 
1090 #define ENTRY_SIZE sizeof(struct dyn_ftrace)
1091 #define ENTRIES_PER_PAGE (PAGE_SIZE / ENTRY_SIZE)
1092 
1093 /* estimate from running different kernels */
1094 #define NR_TO_INIT		10000
1095 
1096 static struct ftrace_page	*ftrace_pages_start;
1097 static struct ftrace_page	*ftrace_pages;
1098 
1099 static bool ftrace_hash_empty(struct ftrace_hash *hash)
1100 {
1101 	return !hash || !hash->count;
1102 }
1103 
1104 static struct ftrace_func_entry *
1105 ftrace_lookup_ip(struct ftrace_hash *hash, unsigned long ip)
1106 {
1107 	unsigned long key;
1108 	struct ftrace_func_entry *entry;
1109 	struct hlist_head *hhd;
1110 	struct hlist_node *n;
1111 
1112 	if (ftrace_hash_empty(hash))
1113 		return NULL;
1114 
1115 	if (hash->size_bits > 0)
1116 		key = hash_long(ip, hash->size_bits);
1117 	else
1118 		key = 0;
1119 
1120 	hhd = &hash->buckets[key];
1121 
1122 	hlist_for_each_entry_rcu(entry, n, hhd, hlist) {
1123 		if (entry->ip == ip)
1124 			return entry;
1125 	}
1126 	return NULL;
1127 }
1128 
1129 static void __add_hash_entry(struct ftrace_hash *hash,
1130 			     struct ftrace_func_entry *entry)
1131 {
1132 	struct hlist_head *hhd;
1133 	unsigned long key;
1134 
1135 	if (hash->size_bits)
1136 		key = hash_long(entry->ip, hash->size_bits);
1137 	else
1138 		key = 0;
1139 
1140 	hhd = &hash->buckets[key];
1141 	hlist_add_head(&entry->hlist, hhd);
1142 	hash->count++;
1143 }
1144 
1145 static int add_hash_entry(struct ftrace_hash *hash, unsigned long ip)
1146 {
1147 	struct ftrace_func_entry *entry;
1148 
1149 	entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1150 	if (!entry)
1151 		return -ENOMEM;
1152 
1153 	entry->ip = ip;
1154 	__add_hash_entry(hash, entry);
1155 
1156 	return 0;
1157 }
1158 
1159 static void
1160 free_hash_entry(struct ftrace_hash *hash,
1161 		  struct ftrace_func_entry *entry)
1162 {
1163 	hlist_del(&entry->hlist);
1164 	kfree(entry);
1165 	hash->count--;
1166 }
1167 
1168 static void
1169 remove_hash_entry(struct ftrace_hash *hash,
1170 		  struct ftrace_func_entry *entry)
1171 {
1172 	hlist_del(&entry->hlist);
1173 	hash->count--;
1174 }
1175 
1176 static void ftrace_hash_clear(struct ftrace_hash *hash)
1177 {
1178 	struct hlist_head *hhd;
1179 	struct hlist_node *tp, *tn;
1180 	struct ftrace_func_entry *entry;
1181 	int size = 1 << hash->size_bits;
1182 	int i;
1183 
1184 	if (!hash->count)
1185 		return;
1186 
1187 	for (i = 0; i < size; i++) {
1188 		hhd = &hash->buckets[i];
1189 		hlist_for_each_entry_safe(entry, tp, tn, hhd, hlist)
1190 			free_hash_entry(hash, entry);
1191 	}
1192 	FTRACE_WARN_ON(hash->count);
1193 }
1194 
1195 static void free_ftrace_hash(struct ftrace_hash *hash)
1196 {
1197 	if (!hash || hash == EMPTY_HASH)
1198 		return;
1199 	ftrace_hash_clear(hash);
1200 	kfree(hash->buckets);
1201 	kfree(hash);
1202 }
1203 
1204 static void __free_ftrace_hash_rcu(struct rcu_head *rcu)
1205 {
1206 	struct ftrace_hash *hash;
1207 
1208 	hash = container_of(rcu, struct ftrace_hash, rcu);
1209 	free_ftrace_hash(hash);
1210 }
1211 
1212 static void free_ftrace_hash_rcu(struct ftrace_hash *hash)
1213 {
1214 	if (!hash || hash == EMPTY_HASH)
1215 		return;
1216 	call_rcu_sched(&hash->rcu, __free_ftrace_hash_rcu);
1217 }
1218 
1219 void ftrace_free_filter(struct ftrace_ops *ops)
1220 {
1221 	free_ftrace_hash(ops->filter_hash);
1222 	free_ftrace_hash(ops->notrace_hash);
1223 }
1224 
1225 static struct ftrace_hash *alloc_ftrace_hash(int size_bits)
1226 {
1227 	struct ftrace_hash *hash;
1228 	int size;
1229 
1230 	hash = kzalloc(sizeof(*hash), GFP_KERNEL);
1231 	if (!hash)
1232 		return NULL;
1233 
1234 	size = 1 << size_bits;
1235 	hash->buckets = kcalloc(size, sizeof(*hash->buckets), GFP_KERNEL);
1236 
1237 	if (!hash->buckets) {
1238 		kfree(hash);
1239 		return NULL;
1240 	}
1241 
1242 	hash->size_bits = size_bits;
1243 
1244 	return hash;
1245 }
1246 
1247 static struct ftrace_hash *
1248 alloc_and_copy_ftrace_hash(int size_bits, struct ftrace_hash *hash)
1249 {
1250 	struct ftrace_func_entry *entry;
1251 	struct ftrace_hash *new_hash;
1252 	struct hlist_node *tp;
1253 	int size;
1254 	int ret;
1255 	int i;
1256 
1257 	new_hash = alloc_ftrace_hash(size_bits);
1258 	if (!new_hash)
1259 		return NULL;
1260 
1261 	/* Empty hash? */
1262 	if (ftrace_hash_empty(hash))
1263 		return new_hash;
1264 
1265 	size = 1 << hash->size_bits;
1266 	for (i = 0; i < size; i++) {
1267 		hlist_for_each_entry(entry, tp, &hash->buckets[i], hlist) {
1268 			ret = add_hash_entry(new_hash, entry->ip);
1269 			if (ret < 0)
1270 				goto free_hash;
1271 		}
1272 	}
1273 
1274 	FTRACE_WARN_ON(new_hash->count != hash->count);
1275 
1276 	return new_hash;
1277 
1278  free_hash:
1279 	free_ftrace_hash(new_hash);
1280 	return NULL;
1281 }
1282 
1283 static void
1284 ftrace_hash_rec_disable(struct ftrace_ops *ops, int filter_hash);
1285 static void
1286 ftrace_hash_rec_enable(struct ftrace_ops *ops, int filter_hash);
1287 
1288 static int
1289 ftrace_hash_move(struct ftrace_ops *ops, int enable,
1290 		 struct ftrace_hash **dst, struct ftrace_hash *src)
1291 {
1292 	struct ftrace_func_entry *entry;
1293 	struct hlist_node *tp, *tn;
1294 	struct hlist_head *hhd;
1295 	struct ftrace_hash *old_hash;
1296 	struct ftrace_hash *new_hash;
1297 	unsigned long key;
1298 	int size = src->count;
1299 	int bits = 0;
1300 	int ret;
1301 	int i;
1302 
1303 	/*
1304 	 * Remove the current set, update the hash and add
1305 	 * them back.
1306 	 */
1307 	ftrace_hash_rec_disable(ops, enable);
1308 
1309 	/*
1310 	 * If the new source is empty, just free dst and assign it
1311 	 * the empty_hash.
1312 	 */
1313 	if (!src->count) {
1314 		free_ftrace_hash_rcu(*dst);
1315 		rcu_assign_pointer(*dst, EMPTY_HASH);
1316 		/* still need to update the function records */
1317 		ret = 0;
1318 		goto out;
1319 	}
1320 
1321 	/*
1322 	 * Make the hash size about 1/2 the # found
1323 	 */
1324 	for (size /= 2; size; size >>= 1)
1325 		bits++;
1326 
1327 	/* Don't allocate too much */
1328 	if (bits > FTRACE_HASH_MAX_BITS)
1329 		bits = FTRACE_HASH_MAX_BITS;
1330 
1331 	ret = -ENOMEM;
1332 	new_hash = alloc_ftrace_hash(bits);
1333 	if (!new_hash)
1334 		goto out;
1335 
1336 	size = 1 << src->size_bits;
1337 	for (i = 0; i < size; i++) {
1338 		hhd = &src->buckets[i];
1339 		hlist_for_each_entry_safe(entry, tp, tn, hhd, hlist) {
1340 			if (bits > 0)
1341 				key = hash_long(entry->ip, bits);
1342 			else
1343 				key = 0;
1344 			remove_hash_entry(src, entry);
1345 			__add_hash_entry(new_hash, entry);
1346 		}
1347 	}
1348 
1349 	old_hash = *dst;
1350 	rcu_assign_pointer(*dst, new_hash);
1351 	free_ftrace_hash_rcu(old_hash);
1352 
1353 	ret = 0;
1354  out:
1355 	/*
1356 	 * Enable regardless of ret:
1357 	 *  On success, we enable the new hash.
1358 	 *  On failure, we re-enable the original hash.
1359 	 */
1360 	ftrace_hash_rec_enable(ops, enable);
1361 
1362 	return ret;
1363 }
1364 
1365 /*
1366  * Test the hashes for this ops to see if we want to call
1367  * the ops->func or not.
1368  *
1369  * It's a match if the ip is in the ops->filter_hash or
1370  * the filter_hash does not exist or is empty,
1371  *  AND
1372  * the ip is not in the ops->notrace_hash.
1373  *
1374  * This needs to be called with preemption disabled as
1375  * the hashes are freed with call_rcu_sched().
1376  */
1377 static int
1378 ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip)
1379 {
1380 	struct ftrace_hash *filter_hash;
1381 	struct ftrace_hash *notrace_hash;
1382 	int ret;
1383 
1384 	filter_hash = rcu_dereference_raw(ops->filter_hash);
1385 	notrace_hash = rcu_dereference_raw(ops->notrace_hash);
1386 
1387 	if ((ftrace_hash_empty(filter_hash) ||
1388 	     ftrace_lookup_ip(filter_hash, ip)) &&
1389 	    (ftrace_hash_empty(notrace_hash) ||
1390 	     !ftrace_lookup_ip(notrace_hash, ip)))
1391 		ret = 1;
1392 	else
1393 		ret = 0;
1394 
1395 	return ret;
1396 }
1397 
1398 /*
1399  * This is a double for. Do not use 'break' to break out of the loop,
1400  * you must use a goto.
1401  */
1402 #define do_for_each_ftrace_rec(pg, rec)					\
1403 	for (pg = ftrace_pages_start; pg; pg = pg->next) {		\
1404 		int _____i;						\
1405 		for (_____i = 0; _____i < pg->index; _____i++) {	\
1406 			rec = &pg->records[_____i];
1407 
1408 #define while_for_each_ftrace_rec()		\
1409 		}				\
1410 	}
1411 
1412 
1413 static int ftrace_cmp_recs(const void *a, const void *b)
1414 {
1415 	const struct dyn_ftrace *key = a;
1416 	const struct dyn_ftrace *rec = b;
1417 
1418 	if (key->flags < rec->ip)
1419 		return -1;
1420 	if (key->ip >= rec->ip + MCOUNT_INSN_SIZE)
1421 		return 1;
1422 	return 0;
1423 }
1424 
1425 static unsigned long ftrace_location_range(unsigned long start, unsigned long end)
1426 {
1427 	struct ftrace_page *pg;
1428 	struct dyn_ftrace *rec;
1429 	struct dyn_ftrace key;
1430 
1431 	key.ip = start;
1432 	key.flags = end;	/* overload flags, as it is unsigned long */
1433 
1434 	for (pg = ftrace_pages_start; pg; pg = pg->next) {
1435 		if (end < pg->records[0].ip ||
1436 		    start >= (pg->records[pg->index - 1].ip + MCOUNT_INSN_SIZE))
1437 			continue;
1438 		rec = bsearch(&key, pg->records, pg->index,
1439 			      sizeof(struct dyn_ftrace),
1440 			      ftrace_cmp_recs);
1441 		if (rec)
1442 			return rec->ip;
1443 	}
1444 
1445 	return 0;
1446 }
1447 
1448 /**
1449  * ftrace_location - return true if the ip giving is a traced location
1450  * @ip: the instruction pointer to check
1451  *
1452  * Returns rec->ip if @ip given is a pointer to a ftrace location.
1453  * That is, the instruction that is either a NOP or call to
1454  * the function tracer. It checks the ftrace internal tables to
1455  * determine if the address belongs or not.
1456  */
1457 unsigned long ftrace_location(unsigned long ip)
1458 {
1459 	return ftrace_location_range(ip, ip);
1460 }
1461 
1462 /**
1463  * ftrace_text_reserved - return true if range contains an ftrace location
1464  * @start: start of range to search
1465  * @end: end of range to search (inclusive). @end points to the last byte to check.
1466  *
1467  * Returns 1 if @start and @end contains a ftrace location.
1468  * That is, the instruction that is either a NOP or call to
1469  * the function tracer. It checks the ftrace internal tables to
1470  * determine if the address belongs or not.
1471  */
1472 int ftrace_text_reserved(void *start, void *end)
1473 {
1474 	unsigned long ret;
1475 
1476 	ret = ftrace_location_range((unsigned long)start,
1477 				    (unsigned long)end);
1478 
1479 	return (int)!!ret;
1480 }
1481 
1482 static void __ftrace_hash_rec_update(struct ftrace_ops *ops,
1483 				     int filter_hash,
1484 				     bool inc)
1485 {
1486 	struct ftrace_hash *hash;
1487 	struct ftrace_hash *other_hash;
1488 	struct ftrace_page *pg;
1489 	struct dyn_ftrace *rec;
1490 	int count = 0;
1491 	int all = 0;
1492 
1493 	/* Only update if the ops has been registered */
1494 	if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1495 		return;
1496 
1497 	/*
1498 	 * In the filter_hash case:
1499 	 *   If the count is zero, we update all records.
1500 	 *   Otherwise we just update the items in the hash.
1501 	 *
1502 	 * In the notrace_hash case:
1503 	 *   We enable the update in the hash.
1504 	 *   As disabling notrace means enabling the tracing,
1505 	 *   and enabling notrace means disabling, the inc variable
1506 	 *   gets inversed.
1507 	 */
1508 	if (filter_hash) {
1509 		hash = ops->filter_hash;
1510 		other_hash = ops->notrace_hash;
1511 		if (ftrace_hash_empty(hash))
1512 			all = 1;
1513 	} else {
1514 		inc = !inc;
1515 		hash = ops->notrace_hash;
1516 		other_hash = ops->filter_hash;
1517 		/*
1518 		 * If the notrace hash has no items,
1519 		 * then there's nothing to do.
1520 		 */
1521 		if (ftrace_hash_empty(hash))
1522 			return;
1523 	}
1524 
1525 	do_for_each_ftrace_rec(pg, rec) {
1526 		int in_other_hash = 0;
1527 		int in_hash = 0;
1528 		int match = 0;
1529 
1530 		if (all) {
1531 			/*
1532 			 * Only the filter_hash affects all records.
1533 			 * Update if the record is not in the notrace hash.
1534 			 */
1535 			if (!other_hash || !ftrace_lookup_ip(other_hash, rec->ip))
1536 				match = 1;
1537 		} else {
1538 			in_hash = !!ftrace_lookup_ip(hash, rec->ip);
1539 			in_other_hash = !!ftrace_lookup_ip(other_hash, rec->ip);
1540 
1541 			/*
1542 			 *
1543 			 */
1544 			if (filter_hash && in_hash && !in_other_hash)
1545 				match = 1;
1546 			else if (!filter_hash && in_hash &&
1547 				 (in_other_hash || ftrace_hash_empty(other_hash)))
1548 				match = 1;
1549 		}
1550 		if (!match)
1551 			continue;
1552 
1553 		if (inc) {
1554 			rec->flags++;
1555 			if (FTRACE_WARN_ON((rec->flags & ~FTRACE_FL_MASK) == FTRACE_REF_MAX))
1556 				return;
1557 			/*
1558 			 * If any ops wants regs saved for this function
1559 			 * then all ops will get saved regs.
1560 			 */
1561 			if (ops->flags & FTRACE_OPS_FL_SAVE_REGS)
1562 				rec->flags |= FTRACE_FL_REGS;
1563 		} else {
1564 			if (FTRACE_WARN_ON((rec->flags & ~FTRACE_FL_MASK) == 0))
1565 				return;
1566 			rec->flags--;
1567 		}
1568 		count++;
1569 		/* Shortcut, if we handled all records, we are done. */
1570 		if (!all && count == hash->count)
1571 			return;
1572 	} while_for_each_ftrace_rec();
1573 }
1574 
1575 static void ftrace_hash_rec_disable(struct ftrace_ops *ops,
1576 				    int filter_hash)
1577 {
1578 	__ftrace_hash_rec_update(ops, filter_hash, 0);
1579 }
1580 
1581 static void ftrace_hash_rec_enable(struct ftrace_ops *ops,
1582 				   int filter_hash)
1583 {
1584 	__ftrace_hash_rec_update(ops, filter_hash, 1);
1585 }
1586 
1587 static void print_ip_ins(const char *fmt, unsigned char *p)
1588 {
1589 	int i;
1590 
1591 	printk(KERN_CONT "%s", fmt);
1592 
1593 	for (i = 0; i < MCOUNT_INSN_SIZE; i++)
1594 		printk(KERN_CONT "%s%02x", i ? ":" : "", p[i]);
1595 }
1596 
1597 /**
1598  * ftrace_bug - report and shutdown function tracer
1599  * @failed: The failed type (EFAULT, EINVAL, EPERM)
1600  * @ip: The address that failed
1601  *
1602  * The arch code that enables or disables the function tracing
1603  * can call ftrace_bug() when it has detected a problem in
1604  * modifying the code. @failed should be one of either:
1605  * EFAULT - if the problem happens on reading the @ip address
1606  * EINVAL - if what is read at @ip is not what was expected
1607  * EPERM - if the problem happens on writting to the @ip address
1608  */
1609 void ftrace_bug(int failed, unsigned long ip)
1610 {
1611 	switch (failed) {
1612 	case -EFAULT:
1613 		FTRACE_WARN_ON_ONCE(1);
1614 		pr_info("ftrace faulted on modifying ");
1615 		print_ip_sym(ip);
1616 		break;
1617 	case -EINVAL:
1618 		FTRACE_WARN_ON_ONCE(1);
1619 		pr_info("ftrace failed to modify ");
1620 		print_ip_sym(ip);
1621 		print_ip_ins(" actual: ", (unsigned char *)ip);
1622 		printk(KERN_CONT "\n");
1623 		break;
1624 	case -EPERM:
1625 		FTRACE_WARN_ON_ONCE(1);
1626 		pr_info("ftrace faulted on writing ");
1627 		print_ip_sym(ip);
1628 		break;
1629 	default:
1630 		FTRACE_WARN_ON_ONCE(1);
1631 		pr_info("ftrace faulted on unknown error ");
1632 		print_ip_sym(ip);
1633 	}
1634 }
1635 
1636 static int ftrace_check_record(struct dyn_ftrace *rec, int enable, int update)
1637 {
1638 	unsigned long flag = 0UL;
1639 
1640 	/*
1641 	 * If we are updating calls:
1642 	 *
1643 	 *   If the record has a ref count, then we need to enable it
1644 	 *   because someone is using it.
1645 	 *
1646 	 *   Otherwise we make sure its disabled.
1647 	 *
1648 	 * If we are disabling calls, then disable all records that
1649 	 * are enabled.
1650 	 */
1651 	if (enable && (rec->flags & ~FTRACE_FL_MASK))
1652 		flag = FTRACE_FL_ENABLED;
1653 
1654 	/*
1655 	 * If enabling and the REGS flag does not match the REGS_EN, then
1656 	 * do not ignore this record. Set flags to fail the compare against
1657 	 * ENABLED.
1658 	 */
1659 	if (flag &&
1660 	    (!(rec->flags & FTRACE_FL_REGS) != !(rec->flags & FTRACE_FL_REGS_EN)))
1661 		flag |= FTRACE_FL_REGS;
1662 
1663 	/* If the state of this record hasn't changed, then do nothing */
1664 	if ((rec->flags & FTRACE_FL_ENABLED) == flag)
1665 		return FTRACE_UPDATE_IGNORE;
1666 
1667 	if (flag) {
1668 		/* Save off if rec is being enabled (for return value) */
1669 		flag ^= rec->flags & FTRACE_FL_ENABLED;
1670 
1671 		if (update) {
1672 			rec->flags |= FTRACE_FL_ENABLED;
1673 			if (flag & FTRACE_FL_REGS) {
1674 				if (rec->flags & FTRACE_FL_REGS)
1675 					rec->flags |= FTRACE_FL_REGS_EN;
1676 				else
1677 					rec->flags &= ~FTRACE_FL_REGS_EN;
1678 			}
1679 		}
1680 
1681 		/*
1682 		 * If this record is being updated from a nop, then
1683 		 *   return UPDATE_MAKE_CALL.
1684 		 * Otherwise, if the EN flag is set, then return
1685 		 *   UPDATE_MODIFY_CALL_REGS to tell the caller to convert
1686 		 *   from the non-save regs, to a save regs function.
1687 		 * Otherwise,
1688 		 *   return UPDATE_MODIFY_CALL to tell the caller to convert
1689 		 *   from the save regs, to a non-save regs function.
1690 		 */
1691 		if (flag & FTRACE_FL_ENABLED)
1692 			return FTRACE_UPDATE_MAKE_CALL;
1693 		else if (rec->flags & FTRACE_FL_REGS_EN)
1694 			return FTRACE_UPDATE_MODIFY_CALL_REGS;
1695 		else
1696 			return FTRACE_UPDATE_MODIFY_CALL;
1697 	}
1698 
1699 	if (update) {
1700 		/* If there's no more users, clear all flags */
1701 		if (!(rec->flags & ~FTRACE_FL_MASK))
1702 			rec->flags = 0;
1703 		else
1704 			/* Just disable the record (keep REGS state) */
1705 			rec->flags &= ~FTRACE_FL_ENABLED;
1706 	}
1707 
1708 	return FTRACE_UPDATE_MAKE_NOP;
1709 }
1710 
1711 /**
1712  * ftrace_update_record, set a record that now is tracing or not
1713  * @rec: the record to update
1714  * @enable: set to 1 if the record is tracing, zero to force disable
1715  *
1716  * The records that represent all functions that can be traced need
1717  * to be updated when tracing has been enabled.
1718  */
1719 int ftrace_update_record(struct dyn_ftrace *rec, int enable)
1720 {
1721 	return ftrace_check_record(rec, enable, 1);
1722 }
1723 
1724 /**
1725  * ftrace_test_record, check if the record has been enabled or not
1726  * @rec: the record to test
1727  * @enable: set to 1 to check if enabled, 0 if it is disabled
1728  *
1729  * The arch code may need to test if a record is already set to
1730  * tracing to determine how to modify the function code that it
1731  * represents.
1732  */
1733 int ftrace_test_record(struct dyn_ftrace *rec, int enable)
1734 {
1735 	return ftrace_check_record(rec, enable, 0);
1736 }
1737 
1738 static int
1739 __ftrace_replace_code(struct dyn_ftrace *rec, int enable)
1740 {
1741 	unsigned long ftrace_old_addr;
1742 	unsigned long ftrace_addr;
1743 	int ret;
1744 
1745 	ret = ftrace_update_record(rec, enable);
1746 
1747 	if (rec->flags & FTRACE_FL_REGS)
1748 		ftrace_addr = (unsigned long)FTRACE_REGS_ADDR;
1749 	else
1750 		ftrace_addr = (unsigned long)FTRACE_ADDR;
1751 
1752 	switch (ret) {
1753 	case FTRACE_UPDATE_IGNORE:
1754 		return 0;
1755 
1756 	case FTRACE_UPDATE_MAKE_CALL:
1757 		return ftrace_make_call(rec, ftrace_addr);
1758 
1759 	case FTRACE_UPDATE_MAKE_NOP:
1760 		return ftrace_make_nop(NULL, rec, ftrace_addr);
1761 
1762 	case FTRACE_UPDATE_MODIFY_CALL_REGS:
1763 	case FTRACE_UPDATE_MODIFY_CALL:
1764 		if (rec->flags & FTRACE_FL_REGS)
1765 			ftrace_old_addr = (unsigned long)FTRACE_ADDR;
1766 		else
1767 			ftrace_old_addr = (unsigned long)FTRACE_REGS_ADDR;
1768 
1769 		return ftrace_modify_call(rec, ftrace_old_addr, ftrace_addr);
1770 	}
1771 
1772 	return -1; /* unknow ftrace bug */
1773 }
1774 
1775 void __weak ftrace_replace_code(int enable)
1776 {
1777 	struct dyn_ftrace *rec;
1778 	struct ftrace_page *pg;
1779 	int failed;
1780 
1781 	if (unlikely(ftrace_disabled))
1782 		return;
1783 
1784 	do_for_each_ftrace_rec(pg, rec) {
1785 		failed = __ftrace_replace_code(rec, enable);
1786 		if (failed) {
1787 			ftrace_bug(failed, rec->ip);
1788 			/* Stop processing */
1789 			return;
1790 		}
1791 	} while_for_each_ftrace_rec();
1792 }
1793 
1794 struct ftrace_rec_iter {
1795 	struct ftrace_page	*pg;
1796 	int			index;
1797 };
1798 
1799 /**
1800  * ftrace_rec_iter_start, start up iterating over traced functions
1801  *
1802  * Returns an iterator handle that is used to iterate over all
1803  * the records that represent address locations where functions
1804  * are traced.
1805  *
1806  * May return NULL if no records are available.
1807  */
1808 struct ftrace_rec_iter *ftrace_rec_iter_start(void)
1809 {
1810 	/*
1811 	 * We only use a single iterator.
1812 	 * Protected by the ftrace_lock mutex.
1813 	 */
1814 	static struct ftrace_rec_iter ftrace_rec_iter;
1815 	struct ftrace_rec_iter *iter = &ftrace_rec_iter;
1816 
1817 	iter->pg = ftrace_pages_start;
1818 	iter->index = 0;
1819 
1820 	/* Could have empty pages */
1821 	while (iter->pg && !iter->pg->index)
1822 		iter->pg = iter->pg->next;
1823 
1824 	if (!iter->pg)
1825 		return NULL;
1826 
1827 	return iter;
1828 }
1829 
1830 /**
1831  * ftrace_rec_iter_next, get the next record to process.
1832  * @iter: The handle to the iterator.
1833  *
1834  * Returns the next iterator after the given iterator @iter.
1835  */
1836 struct ftrace_rec_iter *ftrace_rec_iter_next(struct ftrace_rec_iter *iter)
1837 {
1838 	iter->index++;
1839 
1840 	if (iter->index >= iter->pg->index) {
1841 		iter->pg = iter->pg->next;
1842 		iter->index = 0;
1843 
1844 		/* Could have empty pages */
1845 		while (iter->pg && !iter->pg->index)
1846 			iter->pg = iter->pg->next;
1847 	}
1848 
1849 	if (!iter->pg)
1850 		return NULL;
1851 
1852 	return iter;
1853 }
1854 
1855 /**
1856  * ftrace_rec_iter_record, get the record at the iterator location
1857  * @iter: The current iterator location
1858  *
1859  * Returns the record that the current @iter is at.
1860  */
1861 struct dyn_ftrace *ftrace_rec_iter_record(struct ftrace_rec_iter *iter)
1862 {
1863 	return &iter->pg->records[iter->index];
1864 }
1865 
1866 static int
1867 ftrace_code_disable(struct module *mod, struct dyn_ftrace *rec)
1868 {
1869 	unsigned long ip;
1870 	int ret;
1871 
1872 	ip = rec->ip;
1873 
1874 	if (unlikely(ftrace_disabled))
1875 		return 0;
1876 
1877 	ret = ftrace_make_nop(mod, rec, MCOUNT_ADDR);
1878 	if (ret) {
1879 		ftrace_bug(ret, ip);
1880 		return 0;
1881 	}
1882 	return 1;
1883 }
1884 
1885 /*
1886  * archs can override this function if they must do something
1887  * before the modifying code is performed.
1888  */
1889 int __weak ftrace_arch_code_modify_prepare(void)
1890 {
1891 	return 0;
1892 }
1893 
1894 /*
1895  * archs can override this function if they must do something
1896  * after the modifying code is performed.
1897  */
1898 int __weak ftrace_arch_code_modify_post_process(void)
1899 {
1900 	return 0;
1901 }
1902 
1903 void ftrace_modify_all_code(int command)
1904 {
1905 	if (command & FTRACE_UPDATE_CALLS)
1906 		ftrace_replace_code(1);
1907 	else if (command & FTRACE_DISABLE_CALLS)
1908 		ftrace_replace_code(0);
1909 
1910 	if (command & FTRACE_UPDATE_TRACE_FUNC)
1911 		ftrace_update_ftrace_func(ftrace_trace_function);
1912 
1913 	if (command & FTRACE_START_FUNC_RET)
1914 		ftrace_enable_ftrace_graph_caller();
1915 	else if (command & FTRACE_STOP_FUNC_RET)
1916 		ftrace_disable_ftrace_graph_caller();
1917 }
1918 
1919 static int __ftrace_modify_code(void *data)
1920 {
1921 	int *command = data;
1922 
1923 	ftrace_modify_all_code(*command);
1924 
1925 	return 0;
1926 }
1927 
1928 /**
1929  * ftrace_run_stop_machine, go back to the stop machine method
1930  * @command: The command to tell ftrace what to do
1931  *
1932  * If an arch needs to fall back to the stop machine method, the
1933  * it can call this function.
1934  */
1935 void ftrace_run_stop_machine(int command)
1936 {
1937 	stop_machine(__ftrace_modify_code, &command, NULL);
1938 }
1939 
1940 /**
1941  * arch_ftrace_update_code, modify the code to trace or not trace
1942  * @command: The command that needs to be done
1943  *
1944  * Archs can override this function if it does not need to
1945  * run stop_machine() to modify code.
1946  */
1947 void __weak arch_ftrace_update_code(int command)
1948 {
1949 	ftrace_run_stop_machine(command);
1950 }
1951 
1952 static void ftrace_run_update_code(int command)
1953 {
1954 	int ret;
1955 
1956 	ret = ftrace_arch_code_modify_prepare();
1957 	FTRACE_WARN_ON(ret);
1958 	if (ret)
1959 		return;
1960 	/*
1961 	 * Do not call function tracer while we update the code.
1962 	 * We are in stop machine.
1963 	 */
1964 	function_trace_stop++;
1965 
1966 	/*
1967 	 * By default we use stop_machine() to modify the code.
1968 	 * But archs can do what ever they want as long as it
1969 	 * is safe. The stop_machine() is the safest, but also
1970 	 * produces the most overhead.
1971 	 */
1972 	arch_ftrace_update_code(command);
1973 
1974 	function_trace_stop--;
1975 
1976 	ret = ftrace_arch_code_modify_post_process();
1977 	FTRACE_WARN_ON(ret);
1978 }
1979 
1980 static ftrace_func_t saved_ftrace_func;
1981 static int ftrace_start_up;
1982 static int global_start_up;
1983 
1984 static void ftrace_startup_enable(int command)
1985 {
1986 	if (saved_ftrace_func != ftrace_trace_function) {
1987 		saved_ftrace_func = ftrace_trace_function;
1988 		command |= FTRACE_UPDATE_TRACE_FUNC;
1989 	}
1990 
1991 	if (!command || !ftrace_enabled)
1992 		return;
1993 
1994 	ftrace_run_update_code(command);
1995 }
1996 
1997 static int ftrace_startup(struct ftrace_ops *ops, int command)
1998 {
1999 	bool hash_enable = true;
2000 
2001 	if (unlikely(ftrace_disabled))
2002 		return -ENODEV;
2003 
2004 	ftrace_start_up++;
2005 	command |= FTRACE_UPDATE_CALLS;
2006 
2007 	/* ops marked global share the filter hashes */
2008 	if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
2009 		ops = &global_ops;
2010 		/* Don't update hash if global is already set */
2011 		if (global_start_up)
2012 			hash_enable = false;
2013 		global_start_up++;
2014 	}
2015 
2016 	ops->flags |= FTRACE_OPS_FL_ENABLED;
2017 	if (hash_enable)
2018 		ftrace_hash_rec_enable(ops, 1);
2019 
2020 	ftrace_startup_enable(command);
2021 
2022 	return 0;
2023 }
2024 
2025 static void ftrace_shutdown(struct ftrace_ops *ops, int command)
2026 {
2027 	bool hash_disable = true;
2028 
2029 	if (unlikely(ftrace_disabled))
2030 		return;
2031 
2032 	ftrace_start_up--;
2033 	/*
2034 	 * Just warn in case of unbalance, no need to kill ftrace, it's not
2035 	 * critical but the ftrace_call callers may be never nopped again after
2036 	 * further ftrace uses.
2037 	 */
2038 	WARN_ON_ONCE(ftrace_start_up < 0);
2039 
2040 	if (ops->flags & FTRACE_OPS_FL_GLOBAL) {
2041 		ops = &global_ops;
2042 		global_start_up--;
2043 		WARN_ON_ONCE(global_start_up < 0);
2044 		/* Don't update hash if global still has users */
2045 		if (global_start_up) {
2046 			WARN_ON_ONCE(!ftrace_start_up);
2047 			hash_disable = false;
2048 		}
2049 	}
2050 
2051 	if (hash_disable)
2052 		ftrace_hash_rec_disable(ops, 1);
2053 
2054 	if (ops != &global_ops || !global_start_up)
2055 		ops->flags &= ~FTRACE_OPS_FL_ENABLED;
2056 
2057 	command |= FTRACE_UPDATE_CALLS;
2058 
2059 	if (saved_ftrace_func != ftrace_trace_function) {
2060 		saved_ftrace_func = ftrace_trace_function;
2061 		command |= FTRACE_UPDATE_TRACE_FUNC;
2062 	}
2063 
2064 	if (!command || !ftrace_enabled)
2065 		return;
2066 
2067 	ftrace_run_update_code(command);
2068 }
2069 
2070 static void ftrace_startup_sysctl(void)
2071 {
2072 	if (unlikely(ftrace_disabled))
2073 		return;
2074 
2075 	/* Force update next time */
2076 	saved_ftrace_func = NULL;
2077 	/* ftrace_start_up is true if we want ftrace running */
2078 	if (ftrace_start_up)
2079 		ftrace_run_update_code(FTRACE_UPDATE_CALLS);
2080 }
2081 
2082 static void ftrace_shutdown_sysctl(void)
2083 {
2084 	if (unlikely(ftrace_disabled))
2085 		return;
2086 
2087 	/* ftrace_start_up is true if ftrace is running */
2088 	if (ftrace_start_up)
2089 		ftrace_run_update_code(FTRACE_DISABLE_CALLS);
2090 }
2091 
2092 static cycle_t		ftrace_update_time;
2093 static unsigned long	ftrace_update_cnt;
2094 unsigned long		ftrace_update_tot_cnt;
2095 
2096 static int ops_traces_mod(struct ftrace_ops *ops)
2097 {
2098 	struct ftrace_hash *hash;
2099 
2100 	hash = ops->filter_hash;
2101 	return ftrace_hash_empty(hash);
2102 }
2103 
2104 static int ftrace_update_code(struct module *mod)
2105 {
2106 	struct ftrace_page *pg;
2107 	struct dyn_ftrace *p;
2108 	cycle_t start, stop;
2109 	unsigned long ref = 0;
2110 	int i;
2111 
2112 	/*
2113 	 * When adding a module, we need to check if tracers are
2114 	 * currently enabled and if they are set to trace all functions.
2115 	 * If they are, we need to enable the module functions as well
2116 	 * as update the reference counts for those function records.
2117 	 */
2118 	if (mod) {
2119 		struct ftrace_ops *ops;
2120 
2121 		for (ops = ftrace_ops_list;
2122 		     ops != &ftrace_list_end; ops = ops->next) {
2123 			if (ops->flags & FTRACE_OPS_FL_ENABLED &&
2124 			    ops_traces_mod(ops))
2125 				ref++;
2126 		}
2127 	}
2128 
2129 	start = ftrace_now(raw_smp_processor_id());
2130 	ftrace_update_cnt = 0;
2131 
2132 	for (pg = ftrace_new_pgs; pg; pg = pg->next) {
2133 
2134 		for (i = 0; i < pg->index; i++) {
2135 			/* If something went wrong, bail without enabling anything */
2136 			if (unlikely(ftrace_disabled))
2137 				return -1;
2138 
2139 			p = &pg->records[i];
2140 			p->flags = ref;
2141 
2142 			/*
2143 			 * Do the initial record conversion from mcount jump
2144 			 * to the NOP instructions.
2145 			 */
2146 			if (!ftrace_code_disable(mod, p))
2147 				break;
2148 
2149 			ftrace_update_cnt++;
2150 
2151 			/*
2152 			 * If the tracing is enabled, go ahead and enable the record.
2153 			 *
2154 			 * The reason not to enable the record immediatelly is the
2155 			 * inherent check of ftrace_make_nop/ftrace_make_call for
2156 			 * correct previous instructions.  Making first the NOP
2157 			 * conversion puts the module to the correct state, thus
2158 			 * passing the ftrace_make_call check.
2159 			 */
2160 			if (ftrace_start_up && ref) {
2161 				int failed = __ftrace_replace_code(p, 1);
2162 				if (failed)
2163 					ftrace_bug(failed, p->ip);
2164 			}
2165 		}
2166 	}
2167 
2168 	ftrace_new_pgs = NULL;
2169 
2170 	stop = ftrace_now(raw_smp_processor_id());
2171 	ftrace_update_time = stop - start;
2172 	ftrace_update_tot_cnt += ftrace_update_cnt;
2173 
2174 	return 0;
2175 }
2176 
2177 static int ftrace_allocate_records(struct ftrace_page *pg, int count)
2178 {
2179 	int order;
2180 	int cnt;
2181 
2182 	if (WARN_ON(!count))
2183 		return -EINVAL;
2184 
2185 	order = get_count_order(DIV_ROUND_UP(count, ENTRIES_PER_PAGE));
2186 
2187 	/*
2188 	 * We want to fill as much as possible. No more than a page
2189 	 * may be empty.
2190 	 */
2191 	while ((PAGE_SIZE << order) / ENTRY_SIZE >= count + ENTRIES_PER_PAGE)
2192 		order--;
2193 
2194  again:
2195 	pg->records = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
2196 
2197 	if (!pg->records) {
2198 		/* if we can't allocate this size, try something smaller */
2199 		if (!order)
2200 			return -ENOMEM;
2201 		order >>= 1;
2202 		goto again;
2203 	}
2204 
2205 	cnt = (PAGE_SIZE << order) / ENTRY_SIZE;
2206 	pg->size = cnt;
2207 
2208 	if (cnt > count)
2209 		cnt = count;
2210 
2211 	return cnt;
2212 }
2213 
2214 static struct ftrace_page *
2215 ftrace_allocate_pages(unsigned long num_to_init)
2216 {
2217 	struct ftrace_page *start_pg;
2218 	struct ftrace_page *pg;
2219 	int order;
2220 	int cnt;
2221 
2222 	if (!num_to_init)
2223 		return 0;
2224 
2225 	start_pg = pg = kzalloc(sizeof(*pg), GFP_KERNEL);
2226 	if (!pg)
2227 		return NULL;
2228 
2229 	/*
2230 	 * Try to allocate as much as possible in one continues
2231 	 * location that fills in all of the space. We want to
2232 	 * waste as little space as possible.
2233 	 */
2234 	for (;;) {
2235 		cnt = ftrace_allocate_records(pg, num_to_init);
2236 		if (cnt < 0)
2237 			goto free_pages;
2238 
2239 		num_to_init -= cnt;
2240 		if (!num_to_init)
2241 			break;
2242 
2243 		pg->next = kzalloc(sizeof(*pg), GFP_KERNEL);
2244 		if (!pg->next)
2245 			goto free_pages;
2246 
2247 		pg = pg->next;
2248 	}
2249 
2250 	return start_pg;
2251 
2252  free_pages:
2253 	while (start_pg) {
2254 		order = get_count_order(pg->size / ENTRIES_PER_PAGE);
2255 		free_pages((unsigned long)pg->records, order);
2256 		start_pg = pg->next;
2257 		kfree(pg);
2258 		pg = start_pg;
2259 	}
2260 	pr_info("ftrace: FAILED to allocate memory for functions\n");
2261 	return NULL;
2262 }
2263 
2264 static int __init ftrace_dyn_table_alloc(unsigned long num_to_init)
2265 {
2266 	int cnt;
2267 
2268 	if (!num_to_init) {
2269 		pr_info("ftrace: No functions to be traced?\n");
2270 		return -1;
2271 	}
2272 
2273 	cnt = num_to_init / ENTRIES_PER_PAGE;
2274 	pr_info("ftrace: allocating %ld entries in %d pages\n",
2275 		num_to_init, cnt + 1);
2276 
2277 	return 0;
2278 }
2279 
2280 #define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
2281 
2282 struct ftrace_iterator {
2283 	loff_t				pos;
2284 	loff_t				func_pos;
2285 	struct ftrace_page		*pg;
2286 	struct dyn_ftrace		*func;
2287 	struct ftrace_func_probe	*probe;
2288 	struct trace_parser		parser;
2289 	struct ftrace_hash		*hash;
2290 	struct ftrace_ops		*ops;
2291 	int				hidx;
2292 	int				idx;
2293 	unsigned			flags;
2294 };
2295 
2296 static void *
2297 t_hash_next(struct seq_file *m, loff_t *pos)
2298 {
2299 	struct ftrace_iterator *iter = m->private;
2300 	struct hlist_node *hnd = NULL;
2301 	struct hlist_head *hhd;
2302 
2303 	(*pos)++;
2304 	iter->pos = *pos;
2305 
2306 	if (iter->probe)
2307 		hnd = &iter->probe->node;
2308  retry:
2309 	if (iter->hidx >= FTRACE_FUNC_HASHSIZE)
2310 		return NULL;
2311 
2312 	hhd = &ftrace_func_hash[iter->hidx];
2313 
2314 	if (hlist_empty(hhd)) {
2315 		iter->hidx++;
2316 		hnd = NULL;
2317 		goto retry;
2318 	}
2319 
2320 	if (!hnd)
2321 		hnd = hhd->first;
2322 	else {
2323 		hnd = hnd->next;
2324 		if (!hnd) {
2325 			iter->hidx++;
2326 			goto retry;
2327 		}
2328 	}
2329 
2330 	if (WARN_ON_ONCE(!hnd))
2331 		return NULL;
2332 
2333 	iter->probe = hlist_entry(hnd, struct ftrace_func_probe, node);
2334 
2335 	return iter;
2336 }
2337 
2338 static void *t_hash_start(struct seq_file *m, loff_t *pos)
2339 {
2340 	struct ftrace_iterator *iter = m->private;
2341 	void *p = NULL;
2342 	loff_t l;
2343 
2344 	if (!(iter->flags & FTRACE_ITER_DO_HASH))
2345 		return NULL;
2346 
2347 	if (iter->func_pos > *pos)
2348 		return NULL;
2349 
2350 	iter->hidx = 0;
2351 	for (l = 0; l <= (*pos - iter->func_pos); ) {
2352 		p = t_hash_next(m, &l);
2353 		if (!p)
2354 			break;
2355 	}
2356 	if (!p)
2357 		return NULL;
2358 
2359 	/* Only set this if we have an item */
2360 	iter->flags |= FTRACE_ITER_HASH;
2361 
2362 	return iter;
2363 }
2364 
2365 static int
2366 t_hash_show(struct seq_file *m, struct ftrace_iterator *iter)
2367 {
2368 	struct ftrace_func_probe *rec;
2369 
2370 	rec = iter->probe;
2371 	if (WARN_ON_ONCE(!rec))
2372 		return -EIO;
2373 
2374 	if (rec->ops->print)
2375 		return rec->ops->print(m, rec->ip, rec->ops, rec->data);
2376 
2377 	seq_printf(m, "%ps:%ps", (void *)rec->ip, (void *)rec->ops->func);
2378 
2379 	if (rec->data)
2380 		seq_printf(m, ":%p", rec->data);
2381 	seq_putc(m, '\n');
2382 
2383 	return 0;
2384 }
2385 
2386 static void *
2387 t_next(struct seq_file *m, void *v, loff_t *pos)
2388 {
2389 	struct ftrace_iterator *iter = m->private;
2390 	struct ftrace_ops *ops = iter->ops;
2391 	struct dyn_ftrace *rec = NULL;
2392 
2393 	if (unlikely(ftrace_disabled))
2394 		return NULL;
2395 
2396 	if (iter->flags & FTRACE_ITER_HASH)
2397 		return t_hash_next(m, pos);
2398 
2399 	(*pos)++;
2400 	iter->pos = iter->func_pos = *pos;
2401 
2402 	if (iter->flags & FTRACE_ITER_PRINTALL)
2403 		return t_hash_start(m, pos);
2404 
2405  retry:
2406 	if (iter->idx >= iter->pg->index) {
2407 		if (iter->pg->next) {
2408 			iter->pg = iter->pg->next;
2409 			iter->idx = 0;
2410 			goto retry;
2411 		}
2412 	} else {
2413 		rec = &iter->pg->records[iter->idx++];
2414 		if (((iter->flags & FTRACE_ITER_FILTER) &&
2415 		     !(ftrace_lookup_ip(ops->filter_hash, rec->ip))) ||
2416 
2417 		    ((iter->flags & FTRACE_ITER_NOTRACE) &&
2418 		     !ftrace_lookup_ip(ops->notrace_hash, rec->ip)) ||
2419 
2420 		    ((iter->flags & FTRACE_ITER_ENABLED) &&
2421 		     !(rec->flags & ~FTRACE_FL_MASK))) {
2422 
2423 			rec = NULL;
2424 			goto retry;
2425 		}
2426 	}
2427 
2428 	if (!rec)
2429 		return t_hash_start(m, pos);
2430 
2431 	iter->func = rec;
2432 
2433 	return iter;
2434 }
2435 
2436 static void reset_iter_read(struct ftrace_iterator *iter)
2437 {
2438 	iter->pos = 0;
2439 	iter->func_pos = 0;
2440 	iter->flags &= ~(FTRACE_ITER_PRINTALL | FTRACE_ITER_HASH);
2441 }
2442 
2443 static void *t_start(struct seq_file *m, loff_t *pos)
2444 {
2445 	struct ftrace_iterator *iter = m->private;
2446 	struct ftrace_ops *ops = iter->ops;
2447 	void *p = NULL;
2448 	loff_t l;
2449 
2450 	mutex_lock(&ftrace_lock);
2451 
2452 	if (unlikely(ftrace_disabled))
2453 		return NULL;
2454 
2455 	/*
2456 	 * If an lseek was done, then reset and start from beginning.
2457 	 */
2458 	if (*pos < iter->pos)
2459 		reset_iter_read(iter);
2460 
2461 	/*
2462 	 * For set_ftrace_filter reading, if we have the filter
2463 	 * off, we can short cut and just print out that all
2464 	 * functions are enabled.
2465 	 */
2466 	if (iter->flags & FTRACE_ITER_FILTER &&
2467 	    ftrace_hash_empty(ops->filter_hash)) {
2468 		if (*pos > 0)
2469 			return t_hash_start(m, pos);
2470 		iter->flags |= FTRACE_ITER_PRINTALL;
2471 		/* reset in case of seek/pread */
2472 		iter->flags &= ~FTRACE_ITER_HASH;
2473 		return iter;
2474 	}
2475 
2476 	if (iter->flags & FTRACE_ITER_HASH)
2477 		return t_hash_start(m, pos);
2478 
2479 	/*
2480 	 * Unfortunately, we need to restart at ftrace_pages_start
2481 	 * every time we let go of the ftrace_mutex. This is because
2482 	 * those pointers can change without the lock.
2483 	 */
2484 	iter->pg = ftrace_pages_start;
2485 	iter->idx = 0;
2486 	for (l = 0; l <= *pos; ) {
2487 		p = t_next(m, p, &l);
2488 		if (!p)
2489 			break;
2490 	}
2491 
2492 	if (!p)
2493 		return t_hash_start(m, pos);
2494 
2495 	return iter;
2496 }
2497 
2498 static void t_stop(struct seq_file *m, void *p)
2499 {
2500 	mutex_unlock(&ftrace_lock);
2501 }
2502 
2503 static int t_show(struct seq_file *m, void *v)
2504 {
2505 	struct ftrace_iterator *iter = m->private;
2506 	struct dyn_ftrace *rec;
2507 
2508 	if (iter->flags & FTRACE_ITER_HASH)
2509 		return t_hash_show(m, iter);
2510 
2511 	if (iter->flags & FTRACE_ITER_PRINTALL) {
2512 		seq_printf(m, "#### all functions enabled ####\n");
2513 		return 0;
2514 	}
2515 
2516 	rec = iter->func;
2517 
2518 	if (!rec)
2519 		return 0;
2520 
2521 	seq_printf(m, "%ps", (void *)rec->ip);
2522 	if (iter->flags & FTRACE_ITER_ENABLED)
2523 		seq_printf(m, " (%ld)%s",
2524 			   rec->flags & ~FTRACE_FL_MASK,
2525 			   rec->flags & FTRACE_FL_REGS ? " R" : "");
2526 	seq_printf(m, "\n");
2527 
2528 	return 0;
2529 }
2530 
2531 static const struct seq_operations show_ftrace_seq_ops = {
2532 	.start = t_start,
2533 	.next = t_next,
2534 	.stop = t_stop,
2535 	.show = t_show,
2536 };
2537 
2538 static int
2539 ftrace_avail_open(struct inode *inode, struct file *file)
2540 {
2541 	struct ftrace_iterator *iter;
2542 
2543 	if (unlikely(ftrace_disabled))
2544 		return -ENODEV;
2545 
2546 	iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
2547 	if (iter) {
2548 		iter->pg = ftrace_pages_start;
2549 		iter->ops = &global_ops;
2550 	}
2551 
2552 	return iter ? 0 : -ENOMEM;
2553 }
2554 
2555 static int
2556 ftrace_enabled_open(struct inode *inode, struct file *file)
2557 {
2558 	struct ftrace_iterator *iter;
2559 
2560 	if (unlikely(ftrace_disabled))
2561 		return -ENODEV;
2562 
2563 	iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
2564 	if (iter) {
2565 		iter->pg = ftrace_pages_start;
2566 		iter->flags = FTRACE_ITER_ENABLED;
2567 		iter->ops = &global_ops;
2568 	}
2569 
2570 	return iter ? 0 : -ENOMEM;
2571 }
2572 
2573 static void ftrace_filter_reset(struct ftrace_hash *hash)
2574 {
2575 	mutex_lock(&ftrace_lock);
2576 	ftrace_hash_clear(hash);
2577 	mutex_unlock(&ftrace_lock);
2578 }
2579 
2580 /**
2581  * ftrace_regex_open - initialize function tracer filter files
2582  * @ops: The ftrace_ops that hold the hash filters
2583  * @flag: The type of filter to process
2584  * @inode: The inode, usually passed in to your open routine
2585  * @file: The file, usually passed in to your open routine
2586  *
2587  * ftrace_regex_open() initializes the filter files for the
2588  * @ops. Depending on @flag it may process the filter hash or
2589  * the notrace hash of @ops. With this called from the open
2590  * routine, you can use ftrace_filter_write() for the write
2591  * routine if @flag has FTRACE_ITER_FILTER set, or
2592  * ftrace_notrace_write() if @flag has FTRACE_ITER_NOTRACE set.
2593  * ftrace_regex_lseek() should be used as the lseek routine, and
2594  * release must call ftrace_regex_release().
2595  */
2596 int
2597 ftrace_regex_open(struct ftrace_ops *ops, int flag,
2598 		  struct inode *inode, struct file *file)
2599 {
2600 	struct ftrace_iterator *iter;
2601 	struct ftrace_hash *hash;
2602 	int ret = 0;
2603 
2604 	if (unlikely(ftrace_disabled))
2605 		return -ENODEV;
2606 
2607 	iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2608 	if (!iter)
2609 		return -ENOMEM;
2610 
2611 	if (trace_parser_get_init(&iter->parser, FTRACE_BUFF_MAX)) {
2612 		kfree(iter);
2613 		return -ENOMEM;
2614 	}
2615 
2616 	if (flag & FTRACE_ITER_NOTRACE)
2617 		hash = ops->notrace_hash;
2618 	else
2619 		hash = ops->filter_hash;
2620 
2621 	iter->ops = ops;
2622 	iter->flags = flag;
2623 
2624 	if (file->f_mode & FMODE_WRITE) {
2625 		mutex_lock(&ftrace_lock);
2626 		iter->hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, hash);
2627 		mutex_unlock(&ftrace_lock);
2628 
2629 		if (!iter->hash) {
2630 			trace_parser_put(&iter->parser);
2631 			kfree(iter);
2632 			return -ENOMEM;
2633 		}
2634 	}
2635 
2636 	mutex_lock(&ftrace_regex_lock);
2637 
2638 	if ((file->f_mode & FMODE_WRITE) &&
2639 	    (file->f_flags & O_TRUNC))
2640 		ftrace_filter_reset(iter->hash);
2641 
2642 	if (file->f_mode & FMODE_READ) {
2643 		iter->pg = ftrace_pages_start;
2644 
2645 		ret = seq_open(file, &show_ftrace_seq_ops);
2646 		if (!ret) {
2647 			struct seq_file *m = file->private_data;
2648 			m->private = iter;
2649 		} else {
2650 			/* Failed */
2651 			free_ftrace_hash(iter->hash);
2652 			trace_parser_put(&iter->parser);
2653 			kfree(iter);
2654 		}
2655 	} else
2656 		file->private_data = iter;
2657 	mutex_unlock(&ftrace_regex_lock);
2658 
2659 	return ret;
2660 }
2661 
2662 static int
2663 ftrace_filter_open(struct inode *inode, struct file *file)
2664 {
2665 	return ftrace_regex_open(&global_ops,
2666 			FTRACE_ITER_FILTER | FTRACE_ITER_DO_HASH,
2667 			inode, file);
2668 }
2669 
2670 static int
2671 ftrace_notrace_open(struct inode *inode, struct file *file)
2672 {
2673 	return ftrace_regex_open(&global_ops, FTRACE_ITER_NOTRACE,
2674 				 inode, file);
2675 }
2676 
2677 loff_t
2678 ftrace_regex_lseek(struct file *file, loff_t offset, int whence)
2679 {
2680 	loff_t ret;
2681 
2682 	if (file->f_mode & FMODE_READ)
2683 		ret = seq_lseek(file, offset, whence);
2684 	else
2685 		file->f_pos = ret = 1;
2686 
2687 	return ret;
2688 }
2689 
2690 static int ftrace_match(char *str, char *regex, int len, int type)
2691 {
2692 	int matched = 0;
2693 	int slen;
2694 
2695 	switch (type) {
2696 	case MATCH_FULL:
2697 		if (strcmp(str, regex) == 0)
2698 			matched = 1;
2699 		break;
2700 	case MATCH_FRONT_ONLY:
2701 		if (strncmp(str, regex, len) == 0)
2702 			matched = 1;
2703 		break;
2704 	case MATCH_MIDDLE_ONLY:
2705 		if (strstr(str, regex))
2706 			matched = 1;
2707 		break;
2708 	case MATCH_END_ONLY:
2709 		slen = strlen(str);
2710 		if (slen >= len && memcmp(str + slen - len, regex, len) == 0)
2711 			matched = 1;
2712 		break;
2713 	}
2714 
2715 	return matched;
2716 }
2717 
2718 static int
2719 enter_record(struct ftrace_hash *hash, struct dyn_ftrace *rec, int not)
2720 {
2721 	struct ftrace_func_entry *entry;
2722 	int ret = 0;
2723 
2724 	entry = ftrace_lookup_ip(hash, rec->ip);
2725 	if (not) {
2726 		/* Do nothing if it doesn't exist */
2727 		if (!entry)
2728 			return 0;
2729 
2730 		free_hash_entry(hash, entry);
2731 	} else {
2732 		/* Do nothing if it exists */
2733 		if (entry)
2734 			return 0;
2735 
2736 		ret = add_hash_entry(hash, rec->ip);
2737 	}
2738 	return ret;
2739 }
2740 
2741 static int
2742 ftrace_match_record(struct dyn_ftrace *rec, char *mod,
2743 		    char *regex, int len, int type)
2744 {
2745 	char str[KSYM_SYMBOL_LEN];
2746 	char *modname;
2747 
2748 	kallsyms_lookup(rec->ip, NULL, NULL, &modname, str);
2749 
2750 	if (mod) {
2751 		/* module lookup requires matching the module */
2752 		if (!modname || strcmp(modname, mod))
2753 			return 0;
2754 
2755 		/* blank search means to match all funcs in the mod */
2756 		if (!len)
2757 			return 1;
2758 	}
2759 
2760 	return ftrace_match(str, regex, len, type);
2761 }
2762 
2763 static int
2764 match_records(struct ftrace_hash *hash, char *buff,
2765 	      int len, char *mod, int not)
2766 {
2767 	unsigned search_len = 0;
2768 	struct ftrace_page *pg;
2769 	struct dyn_ftrace *rec;
2770 	int type = MATCH_FULL;
2771 	char *search = buff;
2772 	int found = 0;
2773 	int ret;
2774 
2775 	if (len) {
2776 		type = filter_parse_regex(buff, len, &search, &not);
2777 		search_len = strlen(search);
2778 	}
2779 
2780 	mutex_lock(&ftrace_lock);
2781 
2782 	if (unlikely(ftrace_disabled))
2783 		goto out_unlock;
2784 
2785 	do_for_each_ftrace_rec(pg, rec) {
2786 		if (ftrace_match_record(rec, mod, search, search_len, type)) {
2787 			ret = enter_record(hash, rec, not);
2788 			if (ret < 0) {
2789 				found = ret;
2790 				goto out_unlock;
2791 			}
2792 			found = 1;
2793 		}
2794 	} while_for_each_ftrace_rec();
2795  out_unlock:
2796 	mutex_unlock(&ftrace_lock);
2797 
2798 	return found;
2799 }
2800 
2801 static int
2802 ftrace_match_records(struct ftrace_hash *hash, char *buff, int len)
2803 {
2804 	return match_records(hash, buff, len, NULL, 0);
2805 }
2806 
2807 static int
2808 ftrace_match_module_records(struct ftrace_hash *hash, char *buff, char *mod)
2809 {
2810 	int not = 0;
2811 
2812 	/* blank or '*' mean the same */
2813 	if (strcmp(buff, "*") == 0)
2814 		buff[0] = 0;
2815 
2816 	/* handle the case of 'dont filter this module' */
2817 	if (strcmp(buff, "!") == 0 || strcmp(buff, "!*") == 0) {
2818 		buff[0] = 0;
2819 		not = 1;
2820 	}
2821 
2822 	return match_records(hash, buff, strlen(buff), mod, not);
2823 }
2824 
2825 /*
2826  * We register the module command as a template to show others how
2827  * to register the a command as well.
2828  */
2829 
2830 static int
2831 ftrace_mod_callback(struct ftrace_hash *hash,
2832 		    char *func, char *cmd, char *param, int enable)
2833 {
2834 	char *mod;
2835 	int ret = -EINVAL;
2836 
2837 	/*
2838 	 * cmd == 'mod' because we only registered this func
2839 	 * for the 'mod' ftrace_func_command.
2840 	 * But if you register one func with multiple commands,
2841 	 * you can tell which command was used by the cmd
2842 	 * parameter.
2843 	 */
2844 
2845 	/* we must have a module name */
2846 	if (!param)
2847 		return ret;
2848 
2849 	mod = strsep(&param, ":");
2850 	if (!strlen(mod))
2851 		return ret;
2852 
2853 	ret = ftrace_match_module_records(hash, func, mod);
2854 	if (!ret)
2855 		ret = -EINVAL;
2856 	if (ret < 0)
2857 		return ret;
2858 
2859 	return 0;
2860 }
2861 
2862 static struct ftrace_func_command ftrace_mod_cmd = {
2863 	.name			= "mod",
2864 	.func			= ftrace_mod_callback,
2865 };
2866 
2867 static int __init ftrace_mod_cmd_init(void)
2868 {
2869 	return register_ftrace_command(&ftrace_mod_cmd);
2870 }
2871 core_initcall(ftrace_mod_cmd_init);
2872 
2873 static void function_trace_probe_call(unsigned long ip, unsigned long parent_ip,
2874 				      struct ftrace_ops *op, struct pt_regs *pt_regs)
2875 {
2876 	struct ftrace_func_probe *entry;
2877 	struct hlist_head *hhd;
2878 	struct hlist_node *n;
2879 	unsigned long key;
2880 
2881 	key = hash_long(ip, FTRACE_HASH_BITS);
2882 
2883 	hhd = &ftrace_func_hash[key];
2884 
2885 	if (hlist_empty(hhd))
2886 		return;
2887 
2888 	/*
2889 	 * Disable preemption for these calls to prevent a RCU grace
2890 	 * period. This syncs the hash iteration and freeing of items
2891 	 * on the hash. rcu_read_lock is too dangerous here.
2892 	 */
2893 	preempt_disable_notrace();
2894 	hlist_for_each_entry_rcu(entry, n, hhd, node) {
2895 		if (entry->ip == ip)
2896 			entry->ops->func(ip, parent_ip, &entry->data);
2897 	}
2898 	preempt_enable_notrace();
2899 }
2900 
2901 static struct ftrace_ops trace_probe_ops __read_mostly =
2902 {
2903 	.func		= function_trace_probe_call,
2904 };
2905 
2906 static int ftrace_probe_registered;
2907 
2908 static void __enable_ftrace_function_probe(void)
2909 {
2910 	int ret;
2911 	int i;
2912 
2913 	if (ftrace_probe_registered)
2914 		return;
2915 
2916 	for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
2917 		struct hlist_head *hhd = &ftrace_func_hash[i];
2918 		if (hhd->first)
2919 			break;
2920 	}
2921 	/* Nothing registered? */
2922 	if (i == FTRACE_FUNC_HASHSIZE)
2923 		return;
2924 
2925 	ret = __register_ftrace_function(&trace_probe_ops);
2926 	if (!ret)
2927 		ret = ftrace_startup(&trace_probe_ops, 0);
2928 
2929 	ftrace_probe_registered = 1;
2930 }
2931 
2932 static void __disable_ftrace_function_probe(void)
2933 {
2934 	int ret;
2935 	int i;
2936 
2937 	if (!ftrace_probe_registered)
2938 		return;
2939 
2940 	for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
2941 		struct hlist_head *hhd = &ftrace_func_hash[i];
2942 		if (hhd->first)
2943 			return;
2944 	}
2945 
2946 	/* no more funcs left */
2947 	ret = __unregister_ftrace_function(&trace_probe_ops);
2948 	if (!ret)
2949 		ftrace_shutdown(&trace_probe_ops, 0);
2950 
2951 	ftrace_probe_registered = 0;
2952 }
2953 
2954 
2955 static void ftrace_free_entry_rcu(struct rcu_head *rhp)
2956 {
2957 	struct ftrace_func_probe *entry =
2958 		container_of(rhp, struct ftrace_func_probe, rcu);
2959 
2960 	if (entry->ops->free)
2961 		entry->ops->free(&entry->data);
2962 	kfree(entry);
2963 }
2964 
2965 
2966 int
2967 register_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
2968 			      void *data)
2969 {
2970 	struct ftrace_func_probe *entry;
2971 	struct ftrace_page *pg;
2972 	struct dyn_ftrace *rec;
2973 	int type, len, not;
2974 	unsigned long key;
2975 	int count = 0;
2976 	char *search;
2977 
2978 	type = filter_parse_regex(glob, strlen(glob), &search, &not);
2979 	len = strlen(search);
2980 
2981 	/* we do not support '!' for function probes */
2982 	if (WARN_ON(not))
2983 		return -EINVAL;
2984 
2985 	mutex_lock(&ftrace_lock);
2986 
2987 	if (unlikely(ftrace_disabled))
2988 		goto out_unlock;
2989 
2990 	do_for_each_ftrace_rec(pg, rec) {
2991 
2992 		if (!ftrace_match_record(rec, NULL, search, len, type))
2993 			continue;
2994 
2995 		entry = kmalloc(sizeof(*entry), GFP_KERNEL);
2996 		if (!entry) {
2997 			/* If we did not process any, then return error */
2998 			if (!count)
2999 				count = -ENOMEM;
3000 			goto out_unlock;
3001 		}
3002 
3003 		count++;
3004 
3005 		entry->data = data;
3006 
3007 		/*
3008 		 * The caller might want to do something special
3009 		 * for each function we find. We call the callback
3010 		 * to give the caller an opportunity to do so.
3011 		 */
3012 		if (ops->callback) {
3013 			if (ops->callback(rec->ip, &entry->data) < 0) {
3014 				/* caller does not like this func */
3015 				kfree(entry);
3016 				continue;
3017 			}
3018 		}
3019 
3020 		entry->ops = ops;
3021 		entry->ip = rec->ip;
3022 
3023 		key = hash_long(entry->ip, FTRACE_HASH_BITS);
3024 		hlist_add_head_rcu(&entry->node, &ftrace_func_hash[key]);
3025 
3026 	} while_for_each_ftrace_rec();
3027 	__enable_ftrace_function_probe();
3028 
3029  out_unlock:
3030 	mutex_unlock(&ftrace_lock);
3031 
3032 	return count;
3033 }
3034 
3035 enum {
3036 	PROBE_TEST_FUNC		= 1,
3037 	PROBE_TEST_DATA		= 2
3038 };
3039 
3040 static void
3041 __unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
3042 				  void *data, int flags)
3043 {
3044 	struct ftrace_func_probe *entry;
3045 	struct hlist_node *n, *tmp;
3046 	char str[KSYM_SYMBOL_LEN];
3047 	int type = MATCH_FULL;
3048 	int i, len = 0;
3049 	char *search;
3050 
3051 	if (glob && (strcmp(glob, "*") == 0 || !strlen(glob)))
3052 		glob = NULL;
3053 	else if (glob) {
3054 		int not;
3055 
3056 		type = filter_parse_regex(glob, strlen(glob), &search, &not);
3057 		len = strlen(search);
3058 
3059 		/* we do not support '!' for function probes */
3060 		if (WARN_ON(not))
3061 			return;
3062 	}
3063 
3064 	mutex_lock(&ftrace_lock);
3065 	for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
3066 		struct hlist_head *hhd = &ftrace_func_hash[i];
3067 
3068 		hlist_for_each_entry_safe(entry, n, tmp, hhd, node) {
3069 
3070 			/* break up if statements for readability */
3071 			if ((flags & PROBE_TEST_FUNC) && entry->ops != ops)
3072 				continue;
3073 
3074 			if ((flags & PROBE_TEST_DATA) && entry->data != data)
3075 				continue;
3076 
3077 			/* do this last, since it is the most expensive */
3078 			if (glob) {
3079 				kallsyms_lookup(entry->ip, NULL, NULL,
3080 						NULL, str);
3081 				if (!ftrace_match(str, glob, len, type))
3082 					continue;
3083 			}
3084 
3085 			hlist_del(&entry->node);
3086 			call_rcu(&entry->rcu, ftrace_free_entry_rcu);
3087 		}
3088 	}
3089 	__disable_ftrace_function_probe();
3090 	mutex_unlock(&ftrace_lock);
3091 }
3092 
3093 void
3094 unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
3095 				void *data)
3096 {
3097 	__unregister_ftrace_function_probe(glob, ops, data,
3098 					  PROBE_TEST_FUNC | PROBE_TEST_DATA);
3099 }
3100 
3101 void
3102 unregister_ftrace_function_probe_func(char *glob, struct ftrace_probe_ops *ops)
3103 {
3104 	__unregister_ftrace_function_probe(glob, ops, NULL, PROBE_TEST_FUNC);
3105 }
3106 
3107 void unregister_ftrace_function_probe_all(char *glob)
3108 {
3109 	__unregister_ftrace_function_probe(glob, NULL, NULL, 0);
3110 }
3111 
3112 static LIST_HEAD(ftrace_commands);
3113 static DEFINE_MUTEX(ftrace_cmd_mutex);
3114 
3115 int register_ftrace_command(struct ftrace_func_command *cmd)
3116 {
3117 	struct ftrace_func_command *p;
3118 	int ret = 0;
3119 
3120 	mutex_lock(&ftrace_cmd_mutex);
3121 	list_for_each_entry(p, &ftrace_commands, list) {
3122 		if (strcmp(cmd->name, p->name) == 0) {
3123 			ret = -EBUSY;
3124 			goto out_unlock;
3125 		}
3126 	}
3127 	list_add(&cmd->list, &ftrace_commands);
3128  out_unlock:
3129 	mutex_unlock(&ftrace_cmd_mutex);
3130 
3131 	return ret;
3132 }
3133 
3134 int unregister_ftrace_command(struct ftrace_func_command *cmd)
3135 {
3136 	struct ftrace_func_command *p, *n;
3137 	int ret = -ENODEV;
3138 
3139 	mutex_lock(&ftrace_cmd_mutex);
3140 	list_for_each_entry_safe(p, n, &ftrace_commands, list) {
3141 		if (strcmp(cmd->name, p->name) == 0) {
3142 			ret = 0;
3143 			list_del_init(&p->list);
3144 			goto out_unlock;
3145 		}
3146 	}
3147  out_unlock:
3148 	mutex_unlock(&ftrace_cmd_mutex);
3149 
3150 	return ret;
3151 }
3152 
3153 static int ftrace_process_regex(struct ftrace_hash *hash,
3154 				char *buff, int len, int enable)
3155 {
3156 	char *func, *command, *next = buff;
3157 	struct ftrace_func_command *p;
3158 	int ret = -EINVAL;
3159 
3160 	func = strsep(&next, ":");
3161 
3162 	if (!next) {
3163 		ret = ftrace_match_records(hash, func, len);
3164 		if (!ret)
3165 			ret = -EINVAL;
3166 		if (ret < 0)
3167 			return ret;
3168 		return 0;
3169 	}
3170 
3171 	/* command found */
3172 
3173 	command = strsep(&next, ":");
3174 
3175 	mutex_lock(&ftrace_cmd_mutex);
3176 	list_for_each_entry(p, &ftrace_commands, list) {
3177 		if (strcmp(p->name, command) == 0) {
3178 			ret = p->func(hash, func, command, next, enable);
3179 			goto out_unlock;
3180 		}
3181 	}
3182  out_unlock:
3183 	mutex_unlock(&ftrace_cmd_mutex);
3184 
3185 	return ret;
3186 }
3187 
3188 static ssize_t
3189 ftrace_regex_write(struct file *file, const char __user *ubuf,
3190 		   size_t cnt, loff_t *ppos, int enable)
3191 {
3192 	struct ftrace_iterator *iter;
3193 	struct trace_parser *parser;
3194 	ssize_t ret, read;
3195 
3196 	if (!cnt)
3197 		return 0;
3198 
3199 	mutex_lock(&ftrace_regex_lock);
3200 
3201 	ret = -ENODEV;
3202 	if (unlikely(ftrace_disabled))
3203 		goto out_unlock;
3204 
3205 	if (file->f_mode & FMODE_READ) {
3206 		struct seq_file *m = file->private_data;
3207 		iter = m->private;
3208 	} else
3209 		iter = file->private_data;
3210 
3211 	parser = &iter->parser;
3212 	read = trace_get_user(parser, ubuf, cnt, ppos);
3213 
3214 	if (read >= 0 && trace_parser_loaded(parser) &&
3215 	    !trace_parser_cont(parser)) {
3216 		ret = ftrace_process_regex(iter->hash, parser->buffer,
3217 					   parser->idx, enable);
3218 		trace_parser_clear(parser);
3219 		if (ret)
3220 			goto out_unlock;
3221 	}
3222 
3223 	ret = read;
3224 out_unlock:
3225 	mutex_unlock(&ftrace_regex_lock);
3226 
3227 	return ret;
3228 }
3229 
3230 ssize_t
3231 ftrace_filter_write(struct file *file, const char __user *ubuf,
3232 		    size_t cnt, loff_t *ppos)
3233 {
3234 	return ftrace_regex_write(file, ubuf, cnt, ppos, 1);
3235 }
3236 
3237 ssize_t
3238 ftrace_notrace_write(struct file *file, const char __user *ubuf,
3239 		     size_t cnt, loff_t *ppos)
3240 {
3241 	return ftrace_regex_write(file, ubuf, cnt, ppos, 0);
3242 }
3243 
3244 static int
3245 ftrace_match_addr(struct ftrace_hash *hash, unsigned long ip, int remove)
3246 {
3247 	struct ftrace_func_entry *entry;
3248 
3249 	if (!ftrace_location(ip))
3250 		return -EINVAL;
3251 
3252 	if (remove) {
3253 		entry = ftrace_lookup_ip(hash, ip);
3254 		if (!entry)
3255 			return -ENOENT;
3256 		free_hash_entry(hash, entry);
3257 		return 0;
3258 	}
3259 
3260 	return add_hash_entry(hash, ip);
3261 }
3262 
3263 static int
3264 ftrace_set_hash(struct ftrace_ops *ops, unsigned char *buf, int len,
3265 		unsigned long ip, int remove, int reset, int enable)
3266 {
3267 	struct ftrace_hash **orig_hash;
3268 	struct ftrace_hash *hash;
3269 	int ret;
3270 
3271 	/* All global ops uses the global ops filters */
3272 	if (ops->flags & FTRACE_OPS_FL_GLOBAL)
3273 		ops = &global_ops;
3274 
3275 	if (unlikely(ftrace_disabled))
3276 		return -ENODEV;
3277 
3278 	if (enable)
3279 		orig_hash = &ops->filter_hash;
3280 	else
3281 		orig_hash = &ops->notrace_hash;
3282 
3283 	hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
3284 	if (!hash)
3285 		return -ENOMEM;
3286 
3287 	mutex_lock(&ftrace_regex_lock);
3288 	if (reset)
3289 		ftrace_filter_reset(hash);
3290 	if (buf && !ftrace_match_records(hash, buf, len)) {
3291 		ret = -EINVAL;
3292 		goto out_regex_unlock;
3293 	}
3294 	if (ip) {
3295 		ret = ftrace_match_addr(hash, ip, remove);
3296 		if (ret < 0)
3297 			goto out_regex_unlock;
3298 	}
3299 
3300 	mutex_lock(&ftrace_lock);
3301 	ret = ftrace_hash_move(ops, enable, orig_hash, hash);
3302 	if (!ret && ops->flags & FTRACE_OPS_FL_ENABLED
3303 	    && ftrace_enabled)
3304 		ftrace_run_update_code(FTRACE_UPDATE_CALLS);
3305 
3306 	mutex_unlock(&ftrace_lock);
3307 
3308  out_regex_unlock:
3309 	mutex_unlock(&ftrace_regex_lock);
3310 
3311 	free_ftrace_hash(hash);
3312 	return ret;
3313 }
3314 
3315 static int
3316 ftrace_set_addr(struct ftrace_ops *ops, unsigned long ip, int remove,
3317 		int reset, int enable)
3318 {
3319 	return ftrace_set_hash(ops, 0, 0, ip, remove, reset, enable);
3320 }
3321 
3322 /**
3323  * ftrace_set_filter_ip - set a function to filter on in ftrace by address
3324  * @ops - the ops to set the filter with
3325  * @ip - the address to add to or remove from the filter.
3326  * @remove - non zero to remove the ip from the filter
3327  * @reset - non zero to reset all filters before applying this filter.
3328  *
3329  * Filters denote which functions should be enabled when tracing is enabled
3330  * If @ip is NULL, it failes to update filter.
3331  */
3332 int ftrace_set_filter_ip(struct ftrace_ops *ops, unsigned long ip,
3333 			 int remove, int reset)
3334 {
3335 	return ftrace_set_addr(ops, ip, remove, reset, 1);
3336 }
3337 EXPORT_SYMBOL_GPL(ftrace_set_filter_ip);
3338 
3339 static int
3340 ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len,
3341 		 int reset, int enable)
3342 {
3343 	return ftrace_set_hash(ops, buf, len, 0, 0, reset, enable);
3344 }
3345 
3346 /**
3347  * ftrace_set_filter - set a function to filter on in ftrace
3348  * @ops - the ops to set the filter with
3349  * @buf - the string that holds the function filter text.
3350  * @len - the length of the string.
3351  * @reset - non zero to reset all filters before applying this filter.
3352  *
3353  * Filters denote which functions should be enabled when tracing is enabled.
3354  * If @buf is NULL and reset is set, all functions will be enabled for tracing.
3355  */
3356 int ftrace_set_filter(struct ftrace_ops *ops, unsigned char *buf,
3357 		       int len, int reset)
3358 {
3359 	return ftrace_set_regex(ops, buf, len, reset, 1);
3360 }
3361 EXPORT_SYMBOL_GPL(ftrace_set_filter);
3362 
3363 /**
3364  * ftrace_set_notrace - set a function to not trace in ftrace
3365  * @ops - the ops to set the notrace filter with
3366  * @buf - the string that holds the function notrace text.
3367  * @len - the length of the string.
3368  * @reset - non zero to reset all filters before applying this filter.
3369  *
3370  * Notrace Filters denote which functions should not be enabled when tracing
3371  * is enabled. If @buf is NULL and reset is set, all functions will be enabled
3372  * for tracing.
3373  */
3374 int ftrace_set_notrace(struct ftrace_ops *ops, unsigned char *buf,
3375 			int len, int reset)
3376 {
3377 	return ftrace_set_regex(ops, buf, len, reset, 0);
3378 }
3379 EXPORT_SYMBOL_GPL(ftrace_set_notrace);
3380 /**
3381  * ftrace_set_filter - set a function to filter on in ftrace
3382  * @ops - the ops to set the filter with
3383  * @buf - the string that holds the function filter text.
3384  * @len - the length of the string.
3385  * @reset - non zero to reset all filters before applying this filter.
3386  *
3387  * Filters denote which functions should be enabled when tracing is enabled.
3388  * If @buf is NULL and reset is set, all functions will be enabled for tracing.
3389  */
3390 void ftrace_set_global_filter(unsigned char *buf, int len, int reset)
3391 {
3392 	ftrace_set_regex(&global_ops, buf, len, reset, 1);
3393 }
3394 EXPORT_SYMBOL_GPL(ftrace_set_global_filter);
3395 
3396 /**
3397  * ftrace_set_notrace - set a function to not trace in ftrace
3398  * @ops - the ops to set the notrace filter with
3399  * @buf - the string that holds the function notrace text.
3400  * @len - the length of the string.
3401  * @reset - non zero to reset all filters before applying this filter.
3402  *
3403  * Notrace Filters denote which functions should not be enabled when tracing
3404  * is enabled. If @buf is NULL and reset is set, all functions will be enabled
3405  * for tracing.
3406  */
3407 void ftrace_set_global_notrace(unsigned char *buf, int len, int reset)
3408 {
3409 	ftrace_set_regex(&global_ops, buf, len, reset, 0);
3410 }
3411 EXPORT_SYMBOL_GPL(ftrace_set_global_notrace);
3412 
3413 /*
3414  * command line interface to allow users to set filters on boot up.
3415  */
3416 #define FTRACE_FILTER_SIZE		COMMAND_LINE_SIZE
3417 static char ftrace_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
3418 static char ftrace_filter_buf[FTRACE_FILTER_SIZE] __initdata;
3419 
3420 static int __init set_ftrace_notrace(char *str)
3421 {
3422 	strncpy(ftrace_notrace_buf, str, FTRACE_FILTER_SIZE);
3423 	return 1;
3424 }
3425 __setup("ftrace_notrace=", set_ftrace_notrace);
3426 
3427 static int __init set_ftrace_filter(char *str)
3428 {
3429 	strncpy(ftrace_filter_buf, str, FTRACE_FILTER_SIZE);
3430 	return 1;
3431 }
3432 __setup("ftrace_filter=", set_ftrace_filter);
3433 
3434 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3435 static char ftrace_graph_buf[FTRACE_FILTER_SIZE] __initdata;
3436 static int ftrace_set_func(unsigned long *array, int *idx, char *buffer);
3437 
3438 static int __init set_graph_function(char *str)
3439 {
3440 	strlcpy(ftrace_graph_buf, str, FTRACE_FILTER_SIZE);
3441 	return 1;
3442 }
3443 __setup("ftrace_graph_filter=", set_graph_function);
3444 
3445 static void __init set_ftrace_early_graph(char *buf)
3446 {
3447 	int ret;
3448 	char *func;
3449 
3450 	while (buf) {
3451 		func = strsep(&buf, ",");
3452 		/* we allow only one expression at a time */
3453 		ret = ftrace_set_func(ftrace_graph_funcs, &ftrace_graph_count,
3454 				      func);
3455 		if (ret)
3456 			printk(KERN_DEBUG "ftrace: function %s not "
3457 					  "traceable\n", func);
3458 	}
3459 }
3460 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
3461 
3462 void __init
3463 ftrace_set_early_filter(struct ftrace_ops *ops, char *buf, int enable)
3464 {
3465 	char *func;
3466 
3467 	while (buf) {
3468 		func = strsep(&buf, ",");
3469 		ftrace_set_regex(ops, func, strlen(func), 0, enable);
3470 	}
3471 }
3472 
3473 static void __init set_ftrace_early_filters(void)
3474 {
3475 	if (ftrace_filter_buf[0])
3476 		ftrace_set_early_filter(&global_ops, ftrace_filter_buf, 1);
3477 	if (ftrace_notrace_buf[0])
3478 		ftrace_set_early_filter(&global_ops, ftrace_notrace_buf, 0);
3479 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3480 	if (ftrace_graph_buf[0])
3481 		set_ftrace_early_graph(ftrace_graph_buf);
3482 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
3483 }
3484 
3485 int ftrace_regex_release(struct inode *inode, struct file *file)
3486 {
3487 	struct seq_file *m = (struct seq_file *)file->private_data;
3488 	struct ftrace_iterator *iter;
3489 	struct ftrace_hash **orig_hash;
3490 	struct trace_parser *parser;
3491 	int filter_hash;
3492 	int ret;
3493 
3494 	mutex_lock(&ftrace_regex_lock);
3495 	if (file->f_mode & FMODE_READ) {
3496 		iter = m->private;
3497 
3498 		seq_release(inode, file);
3499 	} else
3500 		iter = file->private_data;
3501 
3502 	parser = &iter->parser;
3503 	if (trace_parser_loaded(parser)) {
3504 		parser->buffer[parser->idx] = 0;
3505 		ftrace_match_records(iter->hash, parser->buffer, parser->idx);
3506 	}
3507 
3508 	trace_parser_put(parser);
3509 
3510 	if (file->f_mode & FMODE_WRITE) {
3511 		filter_hash = !!(iter->flags & FTRACE_ITER_FILTER);
3512 
3513 		if (filter_hash)
3514 			orig_hash = &iter->ops->filter_hash;
3515 		else
3516 			orig_hash = &iter->ops->notrace_hash;
3517 
3518 		mutex_lock(&ftrace_lock);
3519 		ret = ftrace_hash_move(iter->ops, filter_hash,
3520 				       orig_hash, iter->hash);
3521 		if (!ret && (iter->ops->flags & FTRACE_OPS_FL_ENABLED)
3522 		    && ftrace_enabled)
3523 			ftrace_run_update_code(FTRACE_UPDATE_CALLS);
3524 
3525 		mutex_unlock(&ftrace_lock);
3526 	}
3527 	free_ftrace_hash(iter->hash);
3528 	kfree(iter);
3529 
3530 	mutex_unlock(&ftrace_regex_lock);
3531 	return 0;
3532 }
3533 
3534 static const struct file_operations ftrace_avail_fops = {
3535 	.open = ftrace_avail_open,
3536 	.read = seq_read,
3537 	.llseek = seq_lseek,
3538 	.release = seq_release_private,
3539 };
3540 
3541 static const struct file_operations ftrace_enabled_fops = {
3542 	.open = ftrace_enabled_open,
3543 	.read = seq_read,
3544 	.llseek = seq_lseek,
3545 	.release = seq_release_private,
3546 };
3547 
3548 static const struct file_operations ftrace_filter_fops = {
3549 	.open = ftrace_filter_open,
3550 	.read = seq_read,
3551 	.write = ftrace_filter_write,
3552 	.llseek = ftrace_regex_lseek,
3553 	.release = ftrace_regex_release,
3554 };
3555 
3556 static const struct file_operations ftrace_notrace_fops = {
3557 	.open = ftrace_notrace_open,
3558 	.read = seq_read,
3559 	.write = ftrace_notrace_write,
3560 	.llseek = ftrace_regex_lseek,
3561 	.release = ftrace_regex_release,
3562 };
3563 
3564 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3565 
3566 static DEFINE_MUTEX(graph_lock);
3567 
3568 int ftrace_graph_count;
3569 int ftrace_graph_filter_enabled;
3570 unsigned long ftrace_graph_funcs[FTRACE_GRAPH_MAX_FUNCS] __read_mostly;
3571 
3572 static void *
3573 __g_next(struct seq_file *m, loff_t *pos)
3574 {
3575 	if (*pos >= ftrace_graph_count)
3576 		return NULL;
3577 	return &ftrace_graph_funcs[*pos];
3578 }
3579 
3580 static void *
3581 g_next(struct seq_file *m, void *v, loff_t *pos)
3582 {
3583 	(*pos)++;
3584 	return __g_next(m, pos);
3585 }
3586 
3587 static void *g_start(struct seq_file *m, loff_t *pos)
3588 {
3589 	mutex_lock(&graph_lock);
3590 
3591 	/* Nothing, tell g_show to print all functions are enabled */
3592 	if (!ftrace_graph_filter_enabled && !*pos)
3593 		return (void *)1;
3594 
3595 	return __g_next(m, pos);
3596 }
3597 
3598 static void g_stop(struct seq_file *m, void *p)
3599 {
3600 	mutex_unlock(&graph_lock);
3601 }
3602 
3603 static int g_show(struct seq_file *m, void *v)
3604 {
3605 	unsigned long *ptr = v;
3606 
3607 	if (!ptr)
3608 		return 0;
3609 
3610 	if (ptr == (unsigned long *)1) {
3611 		seq_printf(m, "#### all functions enabled ####\n");
3612 		return 0;
3613 	}
3614 
3615 	seq_printf(m, "%ps\n", (void *)*ptr);
3616 
3617 	return 0;
3618 }
3619 
3620 static const struct seq_operations ftrace_graph_seq_ops = {
3621 	.start = g_start,
3622 	.next = g_next,
3623 	.stop = g_stop,
3624 	.show = g_show,
3625 };
3626 
3627 static int
3628 ftrace_graph_open(struct inode *inode, struct file *file)
3629 {
3630 	int ret = 0;
3631 
3632 	if (unlikely(ftrace_disabled))
3633 		return -ENODEV;
3634 
3635 	mutex_lock(&graph_lock);
3636 	if ((file->f_mode & FMODE_WRITE) &&
3637 	    (file->f_flags & O_TRUNC)) {
3638 		ftrace_graph_filter_enabled = 0;
3639 		ftrace_graph_count = 0;
3640 		memset(ftrace_graph_funcs, 0, sizeof(ftrace_graph_funcs));
3641 	}
3642 	mutex_unlock(&graph_lock);
3643 
3644 	if (file->f_mode & FMODE_READ)
3645 		ret = seq_open(file, &ftrace_graph_seq_ops);
3646 
3647 	return ret;
3648 }
3649 
3650 static int
3651 ftrace_graph_release(struct inode *inode, struct file *file)
3652 {
3653 	if (file->f_mode & FMODE_READ)
3654 		seq_release(inode, file);
3655 	return 0;
3656 }
3657 
3658 static int
3659 ftrace_set_func(unsigned long *array, int *idx, char *buffer)
3660 {
3661 	struct dyn_ftrace *rec;
3662 	struct ftrace_page *pg;
3663 	int search_len;
3664 	int fail = 1;
3665 	int type, not;
3666 	char *search;
3667 	bool exists;
3668 	int i;
3669 
3670 	/* decode regex */
3671 	type = filter_parse_regex(buffer, strlen(buffer), &search, &not);
3672 	if (!not && *idx >= FTRACE_GRAPH_MAX_FUNCS)
3673 		return -EBUSY;
3674 
3675 	search_len = strlen(search);
3676 
3677 	mutex_lock(&ftrace_lock);
3678 
3679 	if (unlikely(ftrace_disabled)) {
3680 		mutex_unlock(&ftrace_lock);
3681 		return -ENODEV;
3682 	}
3683 
3684 	do_for_each_ftrace_rec(pg, rec) {
3685 
3686 		if (ftrace_match_record(rec, NULL, search, search_len, type)) {
3687 			/* if it is in the array */
3688 			exists = false;
3689 			for (i = 0; i < *idx; i++) {
3690 				if (array[i] == rec->ip) {
3691 					exists = true;
3692 					break;
3693 				}
3694 			}
3695 
3696 			if (!not) {
3697 				fail = 0;
3698 				if (!exists) {
3699 					array[(*idx)++] = rec->ip;
3700 					if (*idx >= FTRACE_GRAPH_MAX_FUNCS)
3701 						goto out;
3702 				}
3703 			} else {
3704 				if (exists) {
3705 					array[i] = array[--(*idx)];
3706 					array[*idx] = 0;
3707 					fail = 0;
3708 				}
3709 			}
3710 		}
3711 	} while_for_each_ftrace_rec();
3712 out:
3713 	mutex_unlock(&ftrace_lock);
3714 
3715 	if (fail)
3716 		return -EINVAL;
3717 
3718 	ftrace_graph_filter_enabled = 1;
3719 	return 0;
3720 }
3721 
3722 static ssize_t
3723 ftrace_graph_write(struct file *file, const char __user *ubuf,
3724 		   size_t cnt, loff_t *ppos)
3725 {
3726 	struct trace_parser parser;
3727 	ssize_t read, ret;
3728 
3729 	if (!cnt)
3730 		return 0;
3731 
3732 	mutex_lock(&graph_lock);
3733 
3734 	if (trace_parser_get_init(&parser, FTRACE_BUFF_MAX)) {
3735 		ret = -ENOMEM;
3736 		goto out_unlock;
3737 	}
3738 
3739 	read = trace_get_user(&parser, ubuf, cnt, ppos);
3740 
3741 	if (read >= 0 && trace_parser_loaded((&parser))) {
3742 		parser.buffer[parser.idx] = 0;
3743 
3744 		/* we allow only one expression at a time */
3745 		ret = ftrace_set_func(ftrace_graph_funcs, &ftrace_graph_count,
3746 					parser.buffer);
3747 		if (ret)
3748 			goto out_free;
3749 	}
3750 
3751 	ret = read;
3752 
3753 out_free:
3754 	trace_parser_put(&parser);
3755 out_unlock:
3756 	mutex_unlock(&graph_lock);
3757 
3758 	return ret;
3759 }
3760 
3761 static const struct file_operations ftrace_graph_fops = {
3762 	.open		= ftrace_graph_open,
3763 	.read		= seq_read,
3764 	.write		= ftrace_graph_write,
3765 	.release	= ftrace_graph_release,
3766 	.llseek		= seq_lseek,
3767 };
3768 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
3769 
3770 static __init int ftrace_init_dyn_debugfs(struct dentry *d_tracer)
3771 {
3772 
3773 	trace_create_file("available_filter_functions", 0444,
3774 			d_tracer, NULL, &ftrace_avail_fops);
3775 
3776 	trace_create_file("enabled_functions", 0444,
3777 			d_tracer, NULL, &ftrace_enabled_fops);
3778 
3779 	trace_create_file("set_ftrace_filter", 0644, d_tracer,
3780 			NULL, &ftrace_filter_fops);
3781 
3782 	trace_create_file("set_ftrace_notrace", 0644, d_tracer,
3783 				    NULL, &ftrace_notrace_fops);
3784 
3785 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
3786 	trace_create_file("set_graph_function", 0444, d_tracer,
3787 				    NULL,
3788 				    &ftrace_graph_fops);
3789 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
3790 
3791 	return 0;
3792 }
3793 
3794 static int ftrace_cmp_ips(const void *a, const void *b)
3795 {
3796 	const unsigned long *ipa = a;
3797 	const unsigned long *ipb = b;
3798 
3799 	if (*ipa > *ipb)
3800 		return 1;
3801 	if (*ipa < *ipb)
3802 		return -1;
3803 	return 0;
3804 }
3805 
3806 static void ftrace_swap_ips(void *a, void *b, int size)
3807 {
3808 	unsigned long *ipa = a;
3809 	unsigned long *ipb = b;
3810 	unsigned long t;
3811 
3812 	t = *ipa;
3813 	*ipa = *ipb;
3814 	*ipb = t;
3815 }
3816 
3817 static int ftrace_process_locs(struct module *mod,
3818 			       unsigned long *start,
3819 			       unsigned long *end)
3820 {
3821 	struct ftrace_page *start_pg;
3822 	struct ftrace_page *pg;
3823 	struct dyn_ftrace *rec;
3824 	unsigned long count;
3825 	unsigned long *p;
3826 	unsigned long addr;
3827 	unsigned long flags = 0; /* Shut up gcc */
3828 	int ret = -ENOMEM;
3829 
3830 	count = end - start;
3831 
3832 	if (!count)
3833 		return 0;
3834 
3835 	sort(start, count, sizeof(*start),
3836 	     ftrace_cmp_ips, ftrace_swap_ips);
3837 
3838 	start_pg = ftrace_allocate_pages(count);
3839 	if (!start_pg)
3840 		return -ENOMEM;
3841 
3842 	mutex_lock(&ftrace_lock);
3843 
3844 	/*
3845 	 * Core and each module needs their own pages, as
3846 	 * modules will free them when they are removed.
3847 	 * Force a new page to be allocated for modules.
3848 	 */
3849 	if (!mod) {
3850 		WARN_ON(ftrace_pages || ftrace_pages_start);
3851 		/* First initialization */
3852 		ftrace_pages = ftrace_pages_start = start_pg;
3853 	} else {
3854 		if (!ftrace_pages)
3855 			goto out;
3856 
3857 		if (WARN_ON(ftrace_pages->next)) {
3858 			/* Hmm, we have free pages? */
3859 			while (ftrace_pages->next)
3860 				ftrace_pages = ftrace_pages->next;
3861 		}
3862 
3863 		ftrace_pages->next = start_pg;
3864 	}
3865 
3866 	p = start;
3867 	pg = start_pg;
3868 	while (p < end) {
3869 		addr = ftrace_call_adjust(*p++);
3870 		/*
3871 		 * Some architecture linkers will pad between
3872 		 * the different mcount_loc sections of different
3873 		 * object files to satisfy alignments.
3874 		 * Skip any NULL pointers.
3875 		 */
3876 		if (!addr)
3877 			continue;
3878 
3879 		if (pg->index == pg->size) {
3880 			/* We should have allocated enough */
3881 			if (WARN_ON(!pg->next))
3882 				break;
3883 			pg = pg->next;
3884 		}
3885 
3886 		rec = &pg->records[pg->index++];
3887 		rec->ip = addr;
3888 	}
3889 
3890 	/* We should have used all pages */
3891 	WARN_ON(pg->next);
3892 
3893 	/* Assign the last page to ftrace_pages */
3894 	ftrace_pages = pg;
3895 
3896 	/* These new locations need to be initialized */
3897 	ftrace_new_pgs = start_pg;
3898 
3899 	/*
3900 	 * We only need to disable interrupts on start up
3901 	 * because we are modifying code that an interrupt
3902 	 * may execute, and the modification is not atomic.
3903 	 * But for modules, nothing runs the code we modify
3904 	 * until we are finished with it, and there's no
3905 	 * reason to cause large interrupt latencies while we do it.
3906 	 */
3907 	if (!mod)
3908 		local_irq_save(flags);
3909 	ftrace_update_code(mod);
3910 	if (!mod)
3911 		local_irq_restore(flags);
3912 	ret = 0;
3913  out:
3914 	mutex_unlock(&ftrace_lock);
3915 
3916 	return ret;
3917 }
3918 
3919 #ifdef CONFIG_MODULES
3920 
3921 #define next_to_ftrace_page(p) container_of(p, struct ftrace_page, next)
3922 
3923 void ftrace_release_mod(struct module *mod)
3924 {
3925 	struct dyn_ftrace *rec;
3926 	struct ftrace_page **last_pg;
3927 	struct ftrace_page *pg;
3928 	int order;
3929 
3930 	mutex_lock(&ftrace_lock);
3931 
3932 	if (ftrace_disabled)
3933 		goto out_unlock;
3934 
3935 	/*
3936 	 * Each module has its own ftrace_pages, remove
3937 	 * them from the list.
3938 	 */
3939 	last_pg = &ftrace_pages_start;
3940 	for (pg = ftrace_pages_start; pg; pg = *last_pg) {
3941 		rec = &pg->records[0];
3942 		if (within_module_core(rec->ip, mod)) {
3943 			/*
3944 			 * As core pages are first, the first
3945 			 * page should never be a module page.
3946 			 */
3947 			if (WARN_ON(pg == ftrace_pages_start))
3948 				goto out_unlock;
3949 
3950 			/* Check if we are deleting the last page */
3951 			if (pg == ftrace_pages)
3952 				ftrace_pages = next_to_ftrace_page(last_pg);
3953 
3954 			*last_pg = pg->next;
3955 			order = get_count_order(pg->size / ENTRIES_PER_PAGE);
3956 			free_pages((unsigned long)pg->records, order);
3957 			kfree(pg);
3958 		} else
3959 			last_pg = &pg->next;
3960 	}
3961  out_unlock:
3962 	mutex_unlock(&ftrace_lock);
3963 }
3964 
3965 static void ftrace_init_module(struct module *mod,
3966 			       unsigned long *start, unsigned long *end)
3967 {
3968 	if (ftrace_disabled || start == end)
3969 		return;
3970 	ftrace_process_locs(mod, start, end);
3971 }
3972 
3973 static int ftrace_module_notify(struct notifier_block *self,
3974 				unsigned long val, void *data)
3975 {
3976 	struct module *mod = data;
3977 
3978 	switch (val) {
3979 	case MODULE_STATE_COMING:
3980 		ftrace_init_module(mod, mod->ftrace_callsites,
3981 				   mod->ftrace_callsites +
3982 				   mod->num_ftrace_callsites);
3983 		break;
3984 	case MODULE_STATE_GOING:
3985 		ftrace_release_mod(mod);
3986 		break;
3987 	}
3988 
3989 	return 0;
3990 }
3991 #else
3992 static int ftrace_module_notify(struct notifier_block *self,
3993 				unsigned long val, void *data)
3994 {
3995 	return 0;
3996 }
3997 #endif /* CONFIG_MODULES */
3998 
3999 struct notifier_block ftrace_module_nb = {
4000 	.notifier_call = ftrace_module_notify,
4001 	.priority = 0,
4002 };
4003 
4004 extern unsigned long __start_mcount_loc[];
4005 extern unsigned long __stop_mcount_loc[];
4006 
4007 void __init ftrace_init(void)
4008 {
4009 	unsigned long count, addr, flags;
4010 	int ret;
4011 
4012 	/* Keep the ftrace pointer to the stub */
4013 	addr = (unsigned long)ftrace_stub;
4014 
4015 	local_irq_save(flags);
4016 	ftrace_dyn_arch_init(&addr);
4017 	local_irq_restore(flags);
4018 
4019 	/* ftrace_dyn_arch_init places the return code in addr */
4020 	if (addr)
4021 		goto failed;
4022 
4023 	count = __stop_mcount_loc - __start_mcount_loc;
4024 
4025 	ret = ftrace_dyn_table_alloc(count);
4026 	if (ret)
4027 		goto failed;
4028 
4029 	last_ftrace_enabled = ftrace_enabled = 1;
4030 
4031 	ret = ftrace_process_locs(NULL,
4032 				  __start_mcount_loc,
4033 				  __stop_mcount_loc);
4034 
4035 	ret = register_module_notifier(&ftrace_module_nb);
4036 	if (ret)
4037 		pr_warning("Failed to register trace ftrace module notifier\n");
4038 
4039 	set_ftrace_early_filters();
4040 
4041 	return;
4042  failed:
4043 	ftrace_disabled = 1;
4044 }
4045 
4046 #else
4047 
4048 static struct ftrace_ops global_ops = {
4049 	.func			= ftrace_stub,
4050 	.flags			= FTRACE_OPS_FL_RECURSION_SAFE,
4051 };
4052 
4053 static int __init ftrace_nodyn_init(void)
4054 {
4055 	ftrace_enabled = 1;
4056 	return 0;
4057 }
4058 core_initcall(ftrace_nodyn_init);
4059 
4060 static inline int ftrace_init_dyn_debugfs(struct dentry *d_tracer) { return 0; }
4061 static inline void ftrace_startup_enable(int command) { }
4062 /* Keep as macros so we do not need to define the commands */
4063 # define ftrace_startup(ops, command)			\
4064 	({						\
4065 		(ops)->flags |= FTRACE_OPS_FL_ENABLED;	\
4066 		0;					\
4067 	})
4068 # define ftrace_shutdown(ops, command)	do { } while (0)
4069 # define ftrace_startup_sysctl()	do { } while (0)
4070 # define ftrace_shutdown_sysctl()	do { } while (0)
4071 
4072 static inline int
4073 ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip)
4074 {
4075 	return 1;
4076 }
4077 
4078 #endif /* CONFIG_DYNAMIC_FTRACE */
4079 
4080 static void
4081 ftrace_ops_control_func(unsigned long ip, unsigned long parent_ip,
4082 			struct ftrace_ops *op, struct pt_regs *regs)
4083 {
4084 	if (unlikely(trace_recursion_test(TRACE_CONTROL_BIT)))
4085 		return;
4086 
4087 	/*
4088 	 * Some of the ops may be dynamically allocated,
4089 	 * they must be freed after a synchronize_sched().
4090 	 */
4091 	preempt_disable_notrace();
4092 	trace_recursion_set(TRACE_CONTROL_BIT);
4093 	op = rcu_dereference_raw(ftrace_control_list);
4094 	while (op != &ftrace_list_end) {
4095 		if (!ftrace_function_local_disabled(op) &&
4096 		    ftrace_ops_test(op, ip))
4097 			op->func(ip, parent_ip, op, regs);
4098 
4099 		op = rcu_dereference_raw(op->next);
4100 	};
4101 	trace_recursion_clear(TRACE_CONTROL_BIT);
4102 	preempt_enable_notrace();
4103 }
4104 
4105 static struct ftrace_ops control_ops = {
4106 	.func = ftrace_ops_control_func,
4107 	.flags = FTRACE_OPS_FL_RECURSION_SAFE,
4108 };
4109 
4110 static inline void
4111 __ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
4112 		       struct ftrace_ops *ignored, struct pt_regs *regs)
4113 {
4114 	struct ftrace_ops *op;
4115 
4116 	if (function_trace_stop)
4117 		return;
4118 
4119 	if (unlikely(trace_recursion_test(TRACE_INTERNAL_BIT)))
4120 		return;
4121 
4122 	trace_recursion_set(TRACE_INTERNAL_BIT);
4123 	/*
4124 	 * Some of the ops may be dynamically allocated,
4125 	 * they must be freed after a synchronize_sched().
4126 	 */
4127 	preempt_disable_notrace();
4128 	op = rcu_dereference_raw(ftrace_ops_list);
4129 	while (op != &ftrace_list_end) {
4130 		if (ftrace_ops_test(op, ip))
4131 			op->func(ip, parent_ip, op, regs);
4132 		op = rcu_dereference_raw(op->next);
4133 	};
4134 	preempt_enable_notrace();
4135 	trace_recursion_clear(TRACE_INTERNAL_BIT);
4136 }
4137 
4138 /*
4139  * Some archs only support passing ip and parent_ip. Even though
4140  * the list function ignores the op parameter, we do not want any
4141  * C side effects, where a function is called without the caller
4142  * sending a third parameter.
4143  * Archs are to support both the regs and ftrace_ops at the same time.
4144  * If they support ftrace_ops, it is assumed they support regs.
4145  * If call backs want to use regs, they must either check for regs
4146  * being NULL, or ARCH_SUPPORTS_FTRACE_SAVE_REGS.
4147  * Note, ARCH_SUPPORT_SAVE_REGS expects a full regs to be saved.
4148  * An architecture can pass partial regs with ftrace_ops and still
4149  * set the ARCH_SUPPORT_FTARCE_OPS.
4150  */
4151 #if ARCH_SUPPORTS_FTRACE_OPS
4152 static void ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
4153 				 struct ftrace_ops *op, struct pt_regs *regs)
4154 {
4155 	__ftrace_ops_list_func(ip, parent_ip, NULL, regs);
4156 }
4157 #else
4158 static void ftrace_ops_no_ops(unsigned long ip, unsigned long parent_ip)
4159 {
4160 	__ftrace_ops_list_func(ip, parent_ip, NULL, NULL);
4161 }
4162 #endif
4163 
4164 static void clear_ftrace_swapper(void)
4165 {
4166 	struct task_struct *p;
4167 	int cpu;
4168 
4169 	get_online_cpus();
4170 	for_each_online_cpu(cpu) {
4171 		p = idle_task(cpu);
4172 		clear_tsk_trace_trace(p);
4173 	}
4174 	put_online_cpus();
4175 }
4176 
4177 static void set_ftrace_swapper(void)
4178 {
4179 	struct task_struct *p;
4180 	int cpu;
4181 
4182 	get_online_cpus();
4183 	for_each_online_cpu(cpu) {
4184 		p = idle_task(cpu);
4185 		set_tsk_trace_trace(p);
4186 	}
4187 	put_online_cpus();
4188 }
4189 
4190 static void clear_ftrace_pid(struct pid *pid)
4191 {
4192 	struct task_struct *p;
4193 
4194 	rcu_read_lock();
4195 	do_each_pid_task(pid, PIDTYPE_PID, p) {
4196 		clear_tsk_trace_trace(p);
4197 	} while_each_pid_task(pid, PIDTYPE_PID, p);
4198 	rcu_read_unlock();
4199 
4200 	put_pid(pid);
4201 }
4202 
4203 static void set_ftrace_pid(struct pid *pid)
4204 {
4205 	struct task_struct *p;
4206 
4207 	rcu_read_lock();
4208 	do_each_pid_task(pid, PIDTYPE_PID, p) {
4209 		set_tsk_trace_trace(p);
4210 	} while_each_pid_task(pid, PIDTYPE_PID, p);
4211 	rcu_read_unlock();
4212 }
4213 
4214 static void clear_ftrace_pid_task(struct pid *pid)
4215 {
4216 	if (pid == ftrace_swapper_pid)
4217 		clear_ftrace_swapper();
4218 	else
4219 		clear_ftrace_pid(pid);
4220 }
4221 
4222 static void set_ftrace_pid_task(struct pid *pid)
4223 {
4224 	if (pid == ftrace_swapper_pid)
4225 		set_ftrace_swapper();
4226 	else
4227 		set_ftrace_pid(pid);
4228 }
4229 
4230 static int ftrace_pid_add(int p)
4231 {
4232 	struct pid *pid;
4233 	struct ftrace_pid *fpid;
4234 	int ret = -EINVAL;
4235 
4236 	mutex_lock(&ftrace_lock);
4237 
4238 	if (!p)
4239 		pid = ftrace_swapper_pid;
4240 	else
4241 		pid = find_get_pid(p);
4242 
4243 	if (!pid)
4244 		goto out;
4245 
4246 	ret = 0;
4247 
4248 	list_for_each_entry(fpid, &ftrace_pids, list)
4249 		if (fpid->pid == pid)
4250 			goto out_put;
4251 
4252 	ret = -ENOMEM;
4253 
4254 	fpid = kmalloc(sizeof(*fpid), GFP_KERNEL);
4255 	if (!fpid)
4256 		goto out_put;
4257 
4258 	list_add(&fpid->list, &ftrace_pids);
4259 	fpid->pid = pid;
4260 
4261 	set_ftrace_pid_task(pid);
4262 
4263 	ftrace_update_pid_func();
4264 	ftrace_startup_enable(0);
4265 
4266 	mutex_unlock(&ftrace_lock);
4267 	return 0;
4268 
4269 out_put:
4270 	if (pid != ftrace_swapper_pid)
4271 		put_pid(pid);
4272 
4273 out:
4274 	mutex_unlock(&ftrace_lock);
4275 	return ret;
4276 }
4277 
4278 static void ftrace_pid_reset(void)
4279 {
4280 	struct ftrace_pid *fpid, *safe;
4281 
4282 	mutex_lock(&ftrace_lock);
4283 	list_for_each_entry_safe(fpid, safe, &ftrace_pids, list) {
4284 		struct pid *pid = fpid->pid;
4285 
4286 		clear_ftrace_pid_task(pid);
4287 
4288 		list_del(&fpid->list);
4289 		kfree(fpid);
4290 	}
4291 
4292 	ftrace_update_pid_func();
4293 	ftrace_startup_enable(0);
4294 
4295 	mutex_unlock(&ftrace_lock);
4296 }
4297 
4298 static void *fpid_start(struct seq_file *m, loff_t *pos)
4299 {
4300 	mutex_lock(&ftrace_lock);
4301 
4302 	if (list_empty(&ftrace_pids) && (!*pos))
4303 		return (void *) 1;
4304 
4305 	return seq_list_start(&ftrace_pids, *pos);
4306 }
4307 
4308 static void *fpid_next(struct seq_file *m, void *v, loff_t *pos)
4309 {
4310 	if (v == (void *)1)
4311 		return NULL;
4312 
4313 	return seq_list_next(v, &ftrace_pids, pos);
4314 }
4315 
4316 static void fpid_stop(struct seq_file *m, void *p)
4317 {
4318 	mutex_unlock(&ftrace_lock);
4319 }
4320 
4321 static int fpid_show(struct seq_file *m, void *v)
4322 {
4323 	const struct ftrace_pid *fpid = list_entry(v, struct ftrace_pid, list);
4324 
4325 	if (v == (void *)1) {
4326 		seq_printf(m, "no pid\n");
4327 		return 0;
4328 	}
4329 
4330 	if (fpid->pid == ftrace_swapper_pid)
4331 		seq_printf(m, "swapper tasks\n");
4332 	else
4333 		seq_printf(m, "%u\n", pid_vnr(fpid->pid));
4334 
4335 	return 0;
4336 }
4337 
4338 static const struct seq_operations ftrace_pid_sops = {
4339 	.start = fpid_start,
4340 	.next = fpid_next,
4341 	.stop = fpid_stop,
4342 	.show = fpid_show,
4343 };
4344 
4345 static int
4346 ftrace_pid_open(struct inode *inode, struct file *file)
4347 {
4348 	int ret = 0;
4349 
4350 	if ((file->f_mode & FMODE_WRITE) &&
4351 	    (file->f_flags & O_TRUNC))
4352 		ftrace_pid_reset();
4353 
4354 	if (file->f_mode & FMODE_READ)
4355 		ret = seq_open(file, &ftrace_pid_sops);
4356 
4357 	return ret;
4358 }
4359 
4360 static ssize_t
4361 ftrace_pid_write(struct file *filp, const char __user *ubuf,
4362 		   size_t cnt, loff_t *ppos)
4363 {
4364 	char buf[64], *tmp;
4365 	long val;
4366 	int ret;
4367 
4368 	if (cnt >= sizeof(buf))
4369 		return -EINVAL;
4370 
4371 	if (copy_from_user(&buf, ubuf, cnt))
4372 		return -EFAULT;
4373 
4374 	buf[cnt] = 0;
4375 
4376 	/*
4377 	 * Allow "echo > set_ftrace_pid" or "echo -n '' > set_ftrace_pid"
4378 	 * to clean the filter quietly.
4379 	 */
4380 	tmp = strstrip(buf);
4381 	if (strlen(tmp) == 0)
4382 		return 1;
4383 
4384 	ret = kstrtol(tmp, 10, &val);
4385 	if (ret < 0)
4386 		return ret;
4387 
4388 	ret = ftrace_pid_add(val);
4389 
4390 	return ret ? ret : cnt;
4391 }
4392 
4393 static int
4394 ftrace_pid_release(struct inode *inode, struct file *file)
4395 {
4396 	if (file->f_mode & FMODE_READ)
4397 		seq_release(inode, file);
4398 
4399 	return 0;
4400 }
4401 
4402 static const struct file_operations ftrace_pid_fops = {
4403 	.open		= ftrace_pid_open,
4404 	.write		= ftrace_pid_write,
4405 	.read		= seq_read,
4406 	.llseek		= seq_lseek,
4407 	.release	= ftrace_pid_release,
4408 };
4409 
4410 static __init int ftrace_init_debugfs(void)
4411 {
4412 	struct dentry *d_tracer;
4413 
4414 	d_tracer = tracing_init_dentry();
4415 	if (!d_tracer)
4416 		return 0;
4417 
4418 	ftrace_init_dyn_debugfs(d_tracer);
4419 
4420 	trace_create_file("set_ftrace_pid", 0644, d_tracer,
4421 			    NULL, &ftrace_pid_fops);
4422 
4423 	ftrace_profile_debugfs(d_tracer);
4424 
4425 	return 0;
4426 }
4427 fs_initcall(ftrace_init_debugfs);
4428 
4429 /**
4430  * ftrace_kill - kill ftrace
4431  *
4432  * This function should be used by panic code. It stops ftrace
4433  * but in a not so nice way. If you need to simply kill ftrace
4434  * from a non-atomic section, use ftrace_kill.
4435  */
4436 void ftrace_kill(void)
4437 {
4438 	ftrace_disabled = 1;
4439 	ftrace_enabled = 0;
4440 	clear_ftrace_function();
4441 }
4442 
4443 /**
4444  * Test if ftrace is dead or not.
4445  */
4446 int ftrace_is_dead(void)
4447 {
4448 	return ftrace_disabled;
4449 }
4450 
4451 /**
4452  * register_ftrace_function - register a function for profiling
4453  * @ops - ops structure that holds the function for profiling.
4454  *
4455  * Register a function to be called by all functions in the
4456  * kernel.
4457  *
4458  * Note: @ops->func and all the functions it calls must be labeled
4459  *       with "notrace", otherwise it will go into a
4460  *       recursive loop.
4461  */
4462 int register_ftrace_function(struct ftrace_ops *ops)
4463 {
4464 	int ret = -1;
4465 
4466 	mutex_lock(&ftrace_lock);
4467 
4468 	ret = __register_ftrace_function(ops);
4469 	if (!ret)
4470 		ret = ftrace_startup(ops, 0);
4471 
4472 	mutex_unlock(&ftrace_lock);
4473 
4474 	return ret;
4475 }
4476 EXPORT_SYMBOL_GPL(register_ftrace_function);
4477 
4478 /**
4479  * unregister_ftrace_function - unregister a function for profiling.
4480  * @ops - ops structure that holds the function to unregister
4481  *
4482  * Unregister a function that was added to be called by ftrace profiling.
4483  */
4484 int unregister_ftrace_function(struct ftrace_ops *ops)
4485 {
4486 	int ret;
4487 
4488 	mutex_lock(&ftrace_lock);
4489 	ret = __unregister_ftrace_function(ops);
4490 	if (!ret)
4491 		ftrace_shutdown(ops, 0);
4492 	mutex_unlock(&ftrace_lock);
4493 
4494 	return ret;
4495 }
4496 EXPORT_SYMBOL_GPL(unregister_ftrace_function);
4497 
4498 int
4499 ftrace_enable_sysctl(struct ctl_table *table, int write,
4500 		     void __user *buffer, size_t *lenp,
4501 		     loff_t *ppos)
4502 {
4503 	int ret = -ENODEV;
4504 
4505 	mutex_lock(&ftrace_lock);
4506 
4507 	if (unlikely(ftrace_disabled))
4508 		goto out;
4509 
4510 	ret = proc_dointvec(table, write, buffer, lenp, ppos);
4511 
4512 	if (ret || !write || (last_ftrace_enabled == !!ftrace_enabled))
4513 		goto out;
4514 
4515 	last_ftrace_enabled = !!ftrace_enabled;
4516 
4517 	if (ftrace_enabled) {
4518 
4519 		ftrace_startup_sysctl();
4520 
4521 		/* we are starting ftrace again */
4522 		if (ftrace_ops_list != &ftrace_list_end) {
4523 			if (ftrace_ops_list->next == &ftrace_list_end)
4524 				ftrace_trace_function = ftrace_ops_list->func;
4525 			else
4526 				ftrace_trace_function = ftrace_ops_list_func;
4527 		}
4528 
4529 	} else {
4530 		/* stopping ftrace calls (just send to ftrace_stub) */
4531 		ftrace_trace_function = ftrace_stub;
4532 
4533 		ftrace_shutdown_sysctl();
4534 	}
4535 
4536  out:
4537 	mutex_unlock(&ftrace_lock);
4538 	return ret;
4539 }
4540 
4541 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
4542 
4543 static int ftrace_graph_active;
4544 static struct notifier_block ftrace_suspend_notifier;
4545 
4546 int ftrace_graph_entry_stub(struct ftrace_graph_ent *trace)
4547 {
4548 	return 0;
4549 }
4550 
4551 /* The callbacks that hook a function */
4552 trace_func_graph_ret_t ftrace_graph_return =
4553 			(trace_func_graph_ret_t)ftrace_stub;
4554 trace_func_graph_ent_t ftrace_graph_entry = ftrace_graph_entry_stub;
4555 
4556 /* Try to assign a return stack array on FTRACE_RETSTACK_ALLOC_SIZE tasks. */
4557 static int alloc_retstack_tasklist(struct ftrace_ret_stack **ret_stack_list)
4558 {
4559 	int i;
4560 	int ret = 0;
4561 	unsigned long flags;
4562 	int start = 0, end = FTRACE_RETSTACK_ALLOC_SIZE;
4563 	struct task_struct *g, *t;
4564 
4565 	for (i = 0; i < FTRACE_RETSTACK_ALLOC_SIZE; i++) {
4566 		ret_stack_list[i] = kmalloc(FTRACE_RETFUNC_DEPTH
4567 					* sizeof(struct ftrace_ret_stack),
4568 					GFP_KERNEL);
4569 		if (!ret_stack_list[i]) {
4570 			start = 0;
4571 			end = i;
4572 			ret = -ENOMEM;
4573 			goto free;
4574 		}
4575 	}
4576 
4577 	read_lock_irqsave(&tasklist_lock, flags);
4578 	do_each_thread(g, t) {
4579 		if (start == end) {
4580 			ret = -EAGAIN;
4581 			goto unlock;
4582 		}
4583 
4584 		if (t->ret_stack == NULL) {
4585 			atomic_set(&t->tracing_graph_pause, 0);
4586 			atomic_set(&t->trace_overrun, 0);
4587 			t->curr_ret_stack = -1;
4588 			/* Make sure the tasks see the -1 first: */
4589 			smp_wmb();
4590 			t->ret_stack = ret_stack_list[start++];
4591 		}
4592 	} while_each_thread(g, t);
4593 
4594 unlock:
4595 	read_unlock_irqrestore(&tasklist_lock, flags);
4596 free:
4597 	for (i = start; i < end; i++)
4598 		kfree(ret_stack_list[i]);
4599 	return ret;
4600 }
4601 
4602 static void
4603 ftrace_graph_probe_sched_switch(void *ignore,
4604 			struct task_struct *prev, struct task_struct *next)
4605 {
4606 	unsigned long long timestamp;
4607 	int index;
4608 
4609 	/*
4610 	 * Does the user want to count the time a function was asleep.
4611 	 * If so, do not update the time stamps.
4612 	 */
4613 	if (trace_flags & TRACE_ITER_SLEEP_TIME)
4614 		return;
4615 
4616 	timestamp = trace_clock_local();
4617 
4618 	prev->ftrace_timestamp = timestamp;
4619 
4620 	/* only process tasks that we timestamped */
4621 	if (!next->ftrace_timestamp)
4622 		return;
4623 
4624 	/*
4625 	 * Update all the counters in next to make up for the
4626 	 * time next was sleeping.
4627 	 */
4628 	timestamp -= next->ftrace_timestamp;
4629 
4630 	for (index = next->curr_ret_stack; index >= 0; index--)
4631 		next->ret_stack[index].calltime += timestamp;
4632 }
4633 
4634 /* Allocate a return stack for each task */
4635 static int start_graph_tracing(void)
4636 {
4637 	struct ftrace_ret_stack **ret_stack_list;
4638 	int ret, cpu;
4639 
4640 	ret_stack_list = kmalloc(FTRACE_RETSTACK_ALLOC_SIZE *
4641 				sizeof(struct ftrace_ret_stack *),
4642 				GFP_KERNEL);
4643 
4644 	if (!ret_stack_list)
4645 		return -ENOMEM;
4646 
4647 	/* The cpu_boot init_task->ret_stack will never be freed */
4648 	for_each_online_cpu(cpu) {
4649 		if (!idle_task(cpu)->ret_stack)
4650 			ftrace_graph_init_idle_task(idle_task(cpu), cpu);
4651 	}
4652 
4653 	do {
4654 		ret = alloc_retstack_tasklist(ret_stack_list);
4655 	} while (ret == -EAGAIN);
4656 
4657 	if (!ret) {
4658 		ret = register_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
4659 		if (ret)
4660 			pr_info("ftrace_graph: Couldn't activate tracepoint"
4661 				" probe to kernel_sched_switch\n");
4662 	}
4663 
4664 	kfree(ret_stack_list);
4665 	return ret;
4666 }
4667 
4668 /*
4669  * Hibernation protection.
4670  * The state of the current task is too much unstable during
4671  * suspend/restore to disk. We want to protect against that.
4672  */
4673 static int
4674 ftrace_suspend_notifier_call(struct notifier_block *bl, unsigned long state,
4675 							void *unused)
4676 {
4677 	switch (state) {
4678 	case PM_HIBERNATION_PREPARE:
4679 		pause_graph_tracing();
4680 		break;
4681 
4682 	case PM_POST_HIBERNATION:
4683 		unpause_graph_tracing();
4684 		break;
4685 	}
4686 	return NOTIFY_DONE;
4687 }
4688 
4689 int register_ftrace_graph(trace_func_graph_ret_t retfunc,
4690 			trace_func_graph_ent_t entryfunc)
4691 {
4692 	int ret = 0;
4693 
4694 	mutex_lock(&ftrace_lock);
4695 
4696 	/* we currently allow only one tracer registered at a time */
4697 	if (ftrace_graph_active) {
4698 		ret = -EBUSY;
4699 		goto out;
4700 	}
4701 
4702 	ftrace_suspend_notifier.notifier_call = ftrace_suspend_notifier_call;
4703 	register_pm_notifier(&ftrace_suspend_notifier);
4704 
4705 	ftrace_graph_active++;
4706 	ret = start_graph_tracing();
4707 	if (ret) {
4708 		ftrace_graph_active--;
4709 		goto out;
4710 	}
4711 
4712 	ftrace_graph_return = retfunc;
4713 	ftrace_graph_entry = entryfunc;
4714 
4715 	ret = ftrace_startup(&global_ops, FTRACE_START_FUNC_RET);
4716 
4717 out:
4718 	mutex_unlock(&ftrace_lock);
4719 	return ret;
4720 }
4721 
4722 void unregister_ftrace_graph(void)
4723 {
4724 	mutex_lock(&ftrace_lock);
4725 
4726 	if (unlikely(!ftrace_graph_active))
4727 		goto out;
4728 
4729 	ftrace_graph_active--;
4730 	ftrace_graph_return = (trace_func_graph_ret_t)ftrace_stub;
4731 	ftrace_graph_entry = ftrace_graph_entry_stub;
4732 	ftrace_shutdown(&global_ops, FTRACE_STOP_FUNC_RET);
4733 	unregister_pm_notifier(&ftrace_suspend_notifier);
4734 	unregister_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
4735 
4736  out:
4737 	mutex_unlock(&ftrace_lock);
4738 }
4739 
4740 static DEFINE_PER_CPU(struct ftrace_ret_stack *, idle_ret_stack);
4741 
4742 static void
4743 graph_init_task(struct task_struct *t, struct ftrace_ret_stack *ret_stack)
4744 {
4745 	atomic_set(&t->tracing_graph_pause, 0);
4746 	atomic_set(&t->trace_overrun, 0);
4747 	t->ftrace_timestamp = 0;
4748 	/* make curr_ret_stack visible before we add the ret_stack */
4749 	smp_wmb();
4750 	t->ret_stack = ret_stack;
4751 }
4752 
4753 /*
4754  * Allocate a return stack for the idle task. May be the first
4755  * time through, or it may be done by CPU hotplug online.
4756  */
4757 void ftrace_graph_init_idle_task(struct task_struct *t, int cpu)
4758 {
4759 	t->curr_ret_stack = -1;
4760 	/*
4761 	 * The idle task has no parent, it either has its own
4762 	 * stack or no stack at all.
4763 	 */
4764 	if (t->ret_stack)
4765 		WARN_ON(t->ret_stack != per_cpu(idle_ret_stack, cpu));
4766 
4767 	if (ftrace_graph_active) {
4768 		struct ftrace_ret_stack *ret_stack;
4769 
4770 		ret_stack = per_cpu(idle_ret_stack, cpu);
4771 		if (!ret_stack) {
4772 			ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
4773 					    * sizeof(struct ftrace_ret_stack),
4774 					    GFP_KERNEL);
4775 			if (!ret_stack)
4776 				return;
4777 			per_cpu(idle_ret_stack, cpu) = ret_stack;
4778 		}
4779 		graph_init_task(t, ret_stack);
4780 	}
4781 }
4782 
4783 /* Allocate a return stack for newly created task */
4784 void ftrace_graph_init_task(struct task_struct *t)
4785 {
4786 	/* Make sure we do not use the parent ret_stack */
4787 	t->ret_stack = NULL;
4788 	t->curr_ret_stack = -1;
4789 
4790 	if (ftrace_graph_active) {
4791 		struct ftrace_ret_stack *ret_stack;
4792 
4793 		ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
4794 				* sizeof(struct ftrace_ret_stack),
4795 				GFP_KERNEL);
4796 		if (!ret_stack)
4797 			return;
4798 		graph_init_task(t, ret_stack);
4799 	}
4800 }
4801 
4802 void ftrace_graph_exit_task(struct task_struct *t)
4803 {
4804 	struct ftrace_ret_stack	*ret_stack = t->ret_stack;
4805 
4806 	t->ret_stack = NULL;
4807 	/* NULL must become visible to IRQs before we free it: */
4808 	barrier();
4809 
4810 	kfree(ret_stack);
4811 }
4812 
4813 void ftrace_graph_stop(void)
4814 {
4815 	ftrace_stop();
4816 }
4817 #endif
4818