xref: /openbmc/linux/kernel/trace/ftrace.c (revision 93032e31)
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/tracefs.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 #ifdef CONFIG_DYNAMIC_FTRACE
66 #define INIT_OPS_HASH(opsname)	\
67 	.func_hash		= &opsname.local_hash,			\
68 	.local_hash.regex_lock	= __MUTEX_INITIALIZER(opsname.local_hash.regex_lock),
69 #define ASSIGN_OPS_HASH(opsname, val) \
70 	.func_hash		= val, \
71 	.local_hash.regex_lock	= __MUTEX_INITIALIZER(opsname.local_hash.regex_lock),
72 #else
73 #define INIT_OPS_HASH(opsname)
74 #define ASSIGN_OPS_HASH(opsname, val)
75 #endif
76 
77 static struct ftrace_ops ftrace_list_end __read_mostly = {
78 	.func		= ftrace_stub,
79 	.flags		= FTRACE_OPS_FL_RECURSION_SAFE | FTRACE_OPS_FL_STUB,
80 	INIT_OPS_HASH(ftrace_list_end)
81 };
82 
83 /* ftrace_enabled is a method to turn ftrace on or off */
84 int ftrace_enabled __read_mostly;
85 static int last_ftrace_enabled;
86 
87 /* Current function tracing op */
88 struct ftrace_ops *function_trace_op __read_mostly = &ftrace_list_end;
89 /* What to set function_trace_op to */
90 static struct ftrace_ops *set_function_trace_op;
91 
92 static bool ftrace_pids_enabled(struct ftrace_ops *ops)
93 {
94 	struct trace_array *tr;
95 
96 	if (!(ops->flags & FTRACE_OPS_FL_PID) || !ops->private)
97 		return false;
98 
99 	tr = ops->private;
100 
101 	return tr->function_pids != NULL;
102 }
103 
104 static void ftrace_update_trampoline(struct ftrace_ops *ops);
105 
106 /*
107  * ftrace_disabled is set when an anomaly is discovered.
108  * ftrace_disabled is much stronger than ftrace_enabled.
109  */
110 static int ftrace_disabled __read_mostly;
111 
112 static DEFINE_MUTEX(ftrace_lock);
113 
114 static struct ftrace_ops *ftrace_ops_list __read_mostly = &ftrace_list_end;
115 ftrace_func_t ftrace_trace_function __read_mostly = ftrace_stub;
116 static struct ftrace_ops global_ops;
117 
118 #if ARCH_SUPPORTS_FTRACE_OPS
119 static void ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
120 				 struct ftrace_ops *op, struct pt_regs *regs);
121 #else
122 /* See comment below, where ftrace_ops_list_func is defined */
123 static void ftrace_ops_no_ops(unsigned long ip, unsigned long parent_ip);
124 #define ftrace_ops_list_func ((ftrace_func_t)ftrace_ops_no_ops)
125 #endif
126 
127 /*
128  * Traverse the ftrace_global_list, invoking all entries.  The reason that we
129  * can use rcu_dereference_raw_notrace() is that elements removed from this list
130  * are simply leaked, so there is no need to interact with a grace-period
131  * mechanism.  The rcu_dereference_raw_notrace() calls are needed to handle
132  * concurrent insertions into the ftrace_global_list.
133  *
134  * Silly Alpha and silly pointer-speculation compiler optimizations!
135  */
136 #define do_for_each_ftrace_op(op, list)			\
137 	op = rcu_dereference_raw_notrace(list);			\
138 	do
139 
140 /*
141  * Optimized for just a single item in the list (as that is the normal case).
142  */
143 #define while_for_each_ftrace_op(op)				\
144 	while (likely(op = rcu_dereference_raw_notrace((op)->next)) &&	\
145 	       unlikely((op) != &ftrace_list_end))
146 
147 static inline void ftrace_ops_init(struct ftrace_ops *ops)
148 {
149 #ifdef CONFIG_DYNAMIC_FTRACE
150 	if (!(ops->flags & FTRACE_OPS_FL_INITIALIZED)) {
151 		mutex_init(&ops->local_hash.regex_lock);
152 		ops->func_hash = &ops->local_hash;
153 		ops->flags |= FTRACE_OPS_FL_INITIALIZED;
154 	}
155 #endif
156 }
157 
158 /**
159  * ftrace_nr_registered_ops - return number of ops registered
160  *
161  * Returns the number of ftrace_ops registered and tracing functions
162  */
163 int ftrace_nr_registered_ops(void)
164 {
165 	struct ftrace_ops *ops;
166 	int cnt = 0;
167 
168 	mutex_lock(&ftrace_lock);
169 
170 	for (ops = ftrace_ops_list;
171 	     ops != &ftrace_list_end; ops = ops->next)
172 		cnt++;
173 
174 	mutex_unlock(&ftrace_lock);
175 
176 	return cnt;
177 }
178 
179 static void ftrace_pid_func(unsigned long ip, unsigned long parent_ip,
180 			    struct ftrace_ops *op, struct pt_regs *regs)
181 {
182 	struct trace_array *tr = op->private;
183 
184 	if (tr && this_cpu_read(tr->trace_buffer.data->ftrace_ignore_pid))
185 		return;
186 
187 	op->saved_func(ip, parent_ip, op, regs);
188 }
189 
190 /**
191  * clear_ftrace_function - reset the ftrace function
192  *
193  * This NULLs the ftrace function and in essence stops
194  * tracing.  There may be lag
195  */
196 void clear_ftrace_function(void)
197 {
198 	ftrace_trace_function = ftrace_stub;
199 }
200 
201 static void per_cpu_ops_disable_all(struct ftrace_ops *ops)
202 {
203 	int cpu;
204 
205 	for_each_possible_cpu(cpu)
206 		*per_cpu_ptr(ops->disabled, cpu) = 1;
207 }
208 
209 static int per_cpu_ops_alloc(struct ftrace_ops *ops)
210 {
211 	int __percpu *disabled;
212 
213 	if (WARN_ON_ONCE(!(ops->flags & FTRACE_OPS_FL_PER_CPU)))
214 		return -EINVAL;
215 
216 	disabled = alloc_percpu(int);
217 	if (!disabled)
218 		return -ENOMEM;
219 
220 	ops->disabled = disabled;
221 	per_cpu_ops_disable_all(ops);
222 	return 0;
223 }
224 
225 static void ftrace_sync(struct work_struct *work)
226 {
227 	/*
228 	 * This function is just a stub to implement a hard force
229 	 * of synchronize_sched(). This requires synchronizing
230 	 * tasks even in userspace and idle.
231 	 *
232 	 * Yes, function tracing is rude.
233 	 */
234 }
235 
236 static void ftrace_sync_ipi(void *data)
237 {
238 	/* Probably not needed, but do it anyway */
239 	smp_rmb();
240 }
241 
242 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
243 static void update_function_graph_func(void);
244 
245 /* Both enabled by default (can be cleared by function_graph tracer flags */
246 static bool fgraph_sleep_time = true;
247 static bool fgraph_graph_time = true;
248 
249 #else
250 static inline void update_function_graph_func(void) { }
251 #endif
252 
253 
254 static ftrace_func_t ftrace_ops_get_list_func(struct ftrace_ops *ops)
255 {
256 	/*
257 	 * If this is a dynamic, RCU, or per CPU ops, or we force list func,
258 	 * then it needs to call the list anyway.
259 	 */
260 	if (ops->flags & (FTRACE_OPS_FL_DYNAMIC | FTRACE_OPS_FL_PER_CPU |
261 			  FTRACE_OPS_FL_RCU) || FTRACE_FORCE_LIST_FUNC)
262 		return ftrace_ops_list_func;
263 
264 	return ftrace_ops_get_func(ops);
265 }
266 
267 static void update_ftrace_function(void)
268 {
269 	ftrace_func_t func;
270 
271 	/*
272 	 * Prepare the ftrace_ops that the arch callback will use.
273 	 * If there's only one ftrace_ops registered, the ftrace_ops_list
274 	 * will point to the ops we want.
275 	 */
276 	set_function_trace_op = ftrace_ops_list;
277 
278 	/* If there's no ftrace_ops registered, just call the stub function */
279 	if (ftrace_ops_list == &ftrace_list_end) {
280 		func = ftrace_stub;
281 
282 	/*
283 	 * If we are at the end of the list and this ops is
284 	 * recursion safe and not dynamic and the arch supports passing ops,
285 	 * then have the mcount trampoline call the function directly.
286 	 */
287 	} else if (ftrace_ops_list->next == &ftrace_list_end) {
288 		func = ftrace_ops_get_list_func(ftrace_ops_list);
289 
290 	} else {
291 		/* Just use the default ftrace_ops */
292 		set_function_trace_op = &ftrace_list_end;
293 		func = ftrace_ops_list_func;
294 	}
295 
296 	update_function_graph_func();
297 
298 	/* If there's no change, then do nothing more here */
299 	if (ftrace_trace_function == func)
300 		return;
301 
302 	/*
303 	 * If we are using the list function, it doesn't care
304 	 * about the function_trace_ops.
305 	 */
306 	if (func == ftrace_ops_list_func) {
307 		ftrace_trace_function = func;
308 		/*
309 		 * Don't even bother setting function_trace_ops,
310 		 * it would be racy to do so anyway.
311 		 */
312 		return;
313 	}
314 
315 #ifndef CONFIG_DYNAMIC_FTRACE
316 	/*
317 	 * For static tracing, we need to be a bit more careful.
318 	 * The function change takes affect immediately. Thus,
319 	 * we need to coorditate the setting of the function_trace_ops
320 	 * with the setting of the ftrace_trace_function.
321 	 *
322 	 * Set the function to the list ops, which will call the
323 	 * function we want, albeit indirectly, but it handles the
324 	 * ftrace_ops and doesn't depend on function_trace_op.
325 	 */
326 	ftrace_trace_function = ftrace_ops_list_func;
327 	/*
328 	 * Make sure all CPUs see this. Yes this is slow, but static
329 	 * tracing is slow and nasty to have enabled.
330 	 */
331 	schedule_on_each_cpu(ftrace_sync);
332 	/* Now all cpus are using the list ops. */
333 	function_trace_op = set_function_trace_op;
334 	/* Make sure the function_trace_op is visible on all CPUs */
335 	smp_wmb();
336 	/* Nasty way to force a rmb on all cpus */
337 	smp_call_function(ftrace_sync_ipi, NULL, 1);
338 	/* OK, we are all set to update the ftrace_trace_function now! */
339 #endif /* !CONFIG_DYNAMIC_FTRACE */
340 
341 	ftrace_trace_function = func;
342 }
343 
344 int using_ftrace_ops_list_func(void)
345 {
346 	return ftrace_trace_function == ftrace_ops_list_func;
347 }
348 
349 static void add_ftrace_ops(struct ftrace_ops **list, struct ftrace_ops *ops)
350 {
351 	ops->next = *list;
352 	/*
353 	 * We are entering ops into the list but another
354 	 * CPU might be walking that list. We need to make sure
355 	 * the ops->next pointer is valid before another CPU sees
356 	 * the ops pointer included into the list.
357 	 */
358 	rcu_assign_pointer(*list, ops);
359 }
360 
361 static int remove_ftrace_ops(struct ftrace_ops **list, struct ftrace_ops *ops)
362 {
363 	struct ftrace_ops **p;
364 
365 	/*
366 	 * If we are removing the last function, then simply point
367 	 * to the ftrace_stub.
368 	 */
369 	if (*list == ops && ops->next == &ftrace_list_end) {
370 		*list = &ftrace_list_end;
371 		return 0;
372 	}
373 
374 	for (p = list; *p != &ftrace_list_end; p = &(*p)->next)
375 		if (*p == ops)
376 			break;
377 
378 	if (*p != ops)
379 		return -1;
380 
381 	*p = (*p)->next;
382 	return 0;
383 }
384 
385 static void ftrace_update_trampoline(struct ftrace_ops *ops);
386 
387 static int __register_ftrace_function(struct ftrace_ops *ops)
388 {
389 	if (ops->flags & FTRACE_OPS_FL_DELETED)
390 		return -EINVAL;
391 
392 	if (WARN_ON(ops->flags & FTRACE_OPS_FL_ENABLED))
393 		return -EBUSY;
394 
395 #ifndef CONFIG_DYNAMIC_FTRACE_WITH_REGS
396 	/*
397 	 * If the ftrace_ops specifies SAVE_REGS, then it only can be used
398 	 * if the arch supports it, or SAVE_REGS_IF_SUPPORTED is also set.
399 	 * Setting SAVE_REGS_IF_SUPPORTED makes SAVE_REGS irrelevant.
400 	 */
401 	if (ops->flags & FTRACE_OPS_FL_SAVE_REGS &&
402 	    !(ops->flags & FTRACE_OPS_FL_SAVE_REGS_IF_SUPPORTED))
403 		return -EINVAL;
404 
405 	if (ops->flags & FTRACE_OPS_FL_SAVE_REGS_IF_SUPPORTED)
406 		ops->flags |= FTRACE_OPS_FL_SAVE_REGS;
407 #endif
408 
409 	if (!core_kernel_data((unsigned long)ops))
410 		ops->flags |= FTRACE_OPS_FL_DYNAMIC;
411 
412 	if (ops->flags & FTRACE_OPS_FL_PER_CPU) {
413 		if (per_cpu_ops_alloc(ops))
414 			return -ENOMEM;
415 	}
416 
417 	add_ftrace_ops(&ftrace_ops_list, ops);
418 
419 	/* Always save the function, and reset at unregistering */
420 	ops->saved_func = ops->func;
421 
422 	if (ftrace_pids_enabled(ops))
423 		ops->func = ftrace_pid_func;
424 
425 	ftrace_update_trampoline(ops);
426 
427 	if (ftrace_enabled)
428 		update_ftrace_function();
429 
430 	return 0;
431 }
432 
433 static int __unregister_ftrace_function(struct ftrace_ops *ops)
434 {
435 	int ret;
436 
437 	if (WARN_ON(!(ops->flags & FTRACE_OPS_FL_ENABLED)))
438 		return -EBUSY;
439 
440 	ret = remove_ftrace_ops(&ftrace_ops_list, ops);
441 
442 	if (ret < 0)
443 		return ret;
444 
445 	if (ftrace_enabled)
446 		update_ftrace_function();
447 
448 	ops->func = ops->saved_func;
449 
450 	return 0;
451 }
452 
453 static void ftrace_update_pid_func(void)
454 {
455 	struct ftrace_ops *op;
456 
457 	/* Only do something if we are tracing something */
458 	if (ftrace_trace_function == ftrace_stub)
459 		return;
460 
461 	do_for_each_ftrace_op(op, ftrace_ops_list) {
462 		if (op->flags & FTRACE_OPS_FL_PID) {
463 			op->func = ftrace_pids_enabled(op) ?
464 				ftrace_pid_func : op->saved_func;
465 			ftrace_update_trampoline(op);
466 		}
467 	} while_for_each_ftrace_op(op);
468 
469 	update_ftrace_function();
470 }
471 
472 #ifdef CONFIG_FUNCTION_PROFILER
473 struct ftrace_profile {
474 	struct hlist_node		node;
475 	unsigned long			ip;
476 	unsigned long			counter;
477 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
478 	unsigned long long		time;
479 	unsigned long long		time_squared;
480 #endif
481 };
482 
483 struct ftrace_profile_page {
484 	struct ftrace_profile_page	*next;
485 	unsigned long			index;
486 	struct ftrace_profile		records[];
487 };
488 
489 struct ftrace_profile_stat {
490 	atomic_t			disabled;
491 	struct hlist_head		*hash;
492 	struct ftrace_profile_page	*pages;
493 	struct ftrace_profile_page	*start;
494 	struct tracer_stat		stat;
495 };
496 
497 #define PROFILE_RECORDS_SIZE						\
498 	(PAGE_SIZE - offsetof(struct ftrace_profile_page, records))
499 
500 #define PROFILES_PER_PAGE					\
501 	(PROFILE_RECORDS_SIZE / sizeof(struct ftrace_profile))
502 
503 static int ftrace_profile_enabled __read_mostly;
504 
505 /* ftrace_profile_lock - synchronize the enable and disable of the profiler */
506 static DEFINE_MUTEX(ftrace_profile_lock);
507 
508 static DEFINE_PER_CPU(struct ftrace_profile_stat, ftrace_profile_stats);
509 
510 #define FTRACE_PROFILE_HASH_BITS 10
511 #define FTRACE_PROFILE_HASH_SIZE (1 << FTRACE_PROFILE_HASH_BITS)
512 
513 static void *
514 function_stat_next(void *v, int idx)
515 {
516 	struct ftrace_profile *rec = v;
517 	struct ftrace_profile_page *pg;
518 
519 	pg = (struct ftrace_profile_page *)((unsigned long)rec & PAGE_MASK);
520 
521  again:
522 	if (idx != 0)
523 		rec++;
524 
525 	if ((void *)rec >= (void *)&pg->records[pg->index]) {
526 		pg = pg->next;
527 		if (!pg)
528 			return NULL;
529 		rec = &pg->records[0];
530 		if (!rec->counter)
531 			goto again;
532 	}
533 
534 	return rec;
535 }
536 
537 static void *function_stat_start(struct tracer_stat *trace)
538 {
539 	struct ftrace_profile_stat *stat =
540 		container_of(trace, struct ftrace_profile_stat, stat);
541 
542 	if (!stat || !stat->start)
543 		return NULL;
544 
545 	return function_stat_next(&stat->start->records[0], 0);
546 }
547 
548 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
549 /* function graph compares on total time */
550 static int function_stat_cmp(void *p1, void *p2)
551 {
552 	struct ftrace_profile *a = p1;
553 	struct ftrace_profile *b = p2;
554 
555 	if (a->time < b->time)
556 		return -1;
557 	if (a->time > b->time)
558 		return 1;
559 	else
560 		return 0;
561 }
562 #else
563 /* not function graph compares against hits */
564 static int function_stat_cmp(void *p1, void *p2)
565 {
566 	struct ftrace_profile *a = p1;
567 	struct ftrace_profile *b = p2;
568 
569 	if (a->counter < b->counter)
570 		return -1;
571 	if (a->counter > b->counter)
572 		return 1;
573 	else
574 		return 0;
575 }
576 #endif
577 
578 static int function_stat_headers(struct seq_file *m)
579 {
580 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
581 	seq_puts(m, "  Function                               "
582 		 "Hit    Time            Avg             s^2\n"
583 		    "  --------                               "
584 		 "---    ----            ---             ---\n");
585 #else
586 	seq_puts(m, "  Function                               Hit\n"
587 		    "  --------                               ---\n");
588 #endif
589 	return 0;
590 }
591 
592 static int function_stat_show(struct seq_file *m, void *v)
593 {
594 	struct ftrace_profile *rec = v;
595 	char str[KSYM_SYMBOL_LEN];
596 	int ret = 0;
597 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
598 	static struct trace_seq s;
599 	unsigned long long avg;
600 	unsigned long long stddev;
601 #endif
602 	mutex_lock(&ftrace_profile_lock);
603 
604 	/* we raced with function_profile_reset() */
605 	if (unlikely(rec->counter == 0)) {
606 		ret = -EBUSY;
607 		goto out;
608 	}
609 
610 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
611 	avg = rec->time;
612 	do_div(avg, rec->counter);
613 	if (tracing_thresh && (avg < tracing_thresh))
614 		goto out;
615 #endif
616 
617 	kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
618 	seq_printf(m, "  %-30.30s  %10lu", str, rec->counter);
619 
620 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
621 	seq_puts(m, "    ");
622 
623 	/* Sample standard deviation (s^2) */
624 	if (rec->counter <= 1)
625 		stddev = 0;
626 	else {
627 		/*
628 		 * Apply Welford's method:
629 		 * s^2 = 1 / (n * (n-1)) * (n * \Sum (x_i)^2 - (\Sum x_i)^2)
630 		 */
631 		stddev = rec->counter * rec->time_squared -
632 			 rec->time * rec->time;
633 
634 		/*
635 		 * Divide only 1000 for ns^2 -> us^2 conversion.
636 		 * trace_print_graph_duration will divide 1000 again.
637 		 */
638 		do_div(stddev, rec->counter * (rec->counter - 1) * 1000);
639 	}
640 
641 	trace_seq_init(&s);
642 	trace_print_graph_duration(rec->time, &s);
643 	trace_seq_puts(&s, "    ");
644 	trace_print_graph_duration(avg, &s);
645 	trace_seq_puts(&s, "    ");
646 	trace_print_graph_duration(stddev, &s);
647 	trace_print_seq(m, &s);
648 #endif
649 	seq_putc(m, '\n');
650 out:
651 	mutex_unlock(&ftrace_profile_lock);
652 
653 	return ret;
654 }
655 
656 static void ftrace_profile_reset(struct ftrace_profile_stat *stat)
657 {
658 	struct ftrace_profile_page *pg;
659 
660 	pg = stat->pages = stat->start;
661 
662 	while (pg) {
663 		memset(pg->records, 0, PROFILE_RECORDS_SIZE);
664 		pg->index = 0;
665 		pg = pg->next;
666 	}
667 
668 	memset(stat->hash, 0,
669 	       FTRACE_PROFILE_HASH_SIZE * sizeof(struct hlist_head));
670 }
671 
672 int ftrace_profile_pages_init(struct ftrace_profile_stat *stat)
673 {
674 	struct ftrace_profile_page *pg;
675 	int functions;
676 	int pages;
677 	int i;
678 
679 	/* If we already allocated, do nothing */
680 	if (stat->pages)
681 		return 0;
682 
683 	stat->pages = (void *)get_zeroed_page(GFP_KERNEL);
684 	if (!stat->pages)
685 		return -ENOMEM;
686 
687 #ifdef CONFIG_DYNAMIC_FTRACE
688 	functions = ftrace_update_tot_cnt;
689 #else
690 	/*
691 	 * We do not know the number of functions that exist because
692 	 * dynamic tracing is what counts them. With past experience
693 	 * we have around 20K functions. That should be more than enough.
694 	 * It is highly unlikely we will execute every function in
695 	 * the kernel.
696 	 */
697 	functions = 20000;
698 #endif
699 
700 	pg = stat->start = stat->pages;
701 
702 	pages = DIV_ROUND_UP(functions, PROFILES_PER_PAGE);
703 
704 	for (i = 1; i < pages; i++) {
705 		pg->next = (void *)get_zeroed_page(GFP_KERNEL);
706 		if (!pg->next)
707 			goto out_free;
708 		pg = pg->next;
709 	}
710 
711 	return 0;
712 
713  out_free:
714 	pg = stat->start;
715 	while (pg) {
716 		unsigned long tmp = (unsigned long)pg;
717 
718 		pg = pg->next;
719 		free_page(tmp);
720 	}
721 
722 	stat->pages = NULL;
723 	stat->start = NULL;
724 
725 	return -ENOMEM;
726 }
727 
728 static int ftrace_profile_init_cpu(int cpu)
729 {
730 	struct ftrace_profile_stat *stat;
731 	int size;
732 
733 	stat = &per_cpu(ftrace_profile_stats, cpu);
734 
735 	if (stat->hash) {
736 		/* If the profile is already created, simply reset it */
737 		ftrace_profile_reset(stat);
738 		return 0;
739 	}
740 
741 	/*
742 	 * We are profiling all functions, but usually only a few thousand
743 	 * functions are hit. We'll make a hash of 1024 items.
744 	 */
745 	size = FTRACE_PROFILE_HASH_SIZE;
746 
747 	stat->hash = kzalloc(sizeof(struct hlist_head) * size, GFP_KERNEL);
748 
749 	if (!stat->hash)
750 		return -ENOMEM;
751 
752 	/* Preallocate the function profiling pages */
753 	if (ftrace_profile_pages_init(stat) < 0) {
754 		kfree(stat->hash);
755 		stat->hash = NULL;
756 		return -ENOMEM;
757 	}
758 
759 	return 0;
760 }
761 
762 static int ftrace_profile_init(void)
763 {
764 	int cpu;
765 	int ret = 0;
766 
767 	for_each_possible_cpu(cpu) {
768 		ret = ftrace_profile_init_cpu(cpu);
769 		if (ret)
770 			break;
771 	}
772 
773 	return ret;
774 }
775 
776 /* interrupts must be disabled */
777 static struct ftrace_profile *
778 ftrace_find_profiled_func(struct ftrace_profile_stat *stat, unsigned long ip)
779 {
780 	struct ftrace_profile *rec;
781 	struct hlist_head *hhd;
782 	unsigned long key;
783 
784 	key = hash_long(ip, FTRACE_PROFILE_HASH_BITS);
785 	hhd = &stat->hash[key];
786 
787 	if (hlist_empty(hhd))
788 		return NULL;
789 
790 	hlist_for_each_entry_rcu_notrace(rec, hhd, node) {
791 		if (rec->ip == ip)
792 			return rec;
793 	}
794 
795 	return NULL;
796 }
797 
798 static void ftrace_add_profile(struct ftrace_profile_stat *stat,
799 			       struct ftrace_profile *rec)
800 {
801 	unsigned long key;
802 
803 	key = hash_long(rec->ip, FTRACE_PROFILE_HASH_BITS);
804 	hlist_add_head_rcu(&rec->node, &stat->hash[key]);
805 }
806 
807 /*
808  * The memory is already allocated, this simply finds a new record to use.
809  */
810 static struct ftrace_profile *
811 ftrace_profile_alloc(struct ftrace_profile_stat *stat, unsigned long ip)
812 {
813 	struct ftrace_profile *rec = NULL;
814 
815 	/* prevent recursion (from NMIs) */
816 	if (atomic_inc_return(&stat->disabled) != 1)
817 		goto out;
818 
819 	/*
820 	 * Try to find the function again since an NMI
821 	 * could have added it
822 	 */
823 	rec = ftrace_find_profiled_func(stat, ip);
824 	if (rec)
825 		goto out;
826 
827 	if (stat->pages->index == PROFILES_PER_PAGE) {
828 		if (!stat->pages->next)
829 			goto out;
830 		stat->pages = stat->pages->next;
831 	}
832 
833 	rec = &stat->pages->records[stat->pages->index++];
834 	rec->ip = ip;
835 	ftrace_add_profile(stat, rec);
836 
837  out:
838 	atomic_dec(&stat->disabled);
839 
840 	return rec;
841 }
842 
843 static void
844 function_profile_call(unsigned long ip, unsigned long parent_ip,
845 		      struct ftrace_ops *ops, struct pt_regs *regs)
846 {
847 	struct ftrace_profile_stat *stat;
848 	struct ftrace_profile *rec;
849 	unsigned long flags;
850 
851 	if (!ftrace_profile_enabled)
852 		return;
853 
854 	local_irq_save(flags);
855 
856 	stat = this_cpu_ptr(&ftrace_profile_stats);
857 	if (!stat->hash || !ftrace_profile_enabled)
858 		goto out;
859 
860 	rec = ftrace_find_profiled_func(stat, ip);
861 	if (!rec) {
862 		rec = ftrace_profile_alloc(stat, ip);
863 		if (!rec)
864 			goto out;
865 	}
866 
867 	rec->counter++;
868  out:
869 	local_irq_restore(flags);
870 }
871 
872 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
873 static int profile_graph_entry(struct ftrace_graph_ent *trace)
874 {
875 	int index = trace->depth;
876 
877 	function_profile_call(trace->func, 0, NULL, NULL);
878 
879 	if (index >= 0 && index < FTRACE_RETFUNC_DEPTH)
880 		current->ret_stack[index].subtime = 0;
881 
882 	return 1;
883 }
884 
885 static void profile_graph_return(struct ftrace_graph_ret *trace)
886 {
887 	struct ftrace_profile_stat *stat;
888 	unsigned long long calltime;
889 	struct ftrace_profile *rec;
890 	unsigned long flags;
891 
892 	local_irq_save(flags);
893 	stat = this_cpu_ptr(&ftrace_profile_stats);
894 	if (!stat->hash || !ftrace_profile_enabled)
895 		goto out;
896 
897 	/* If the calltime was zero'd ignore it */
898 	if (!trace->calltime)
899 		goto out;
900 
901 	calltime = trace->rettime - trace->calltime;
902 
903 	if (!fgraph_graph_time) {
904 		int index;
905 
906 		index = trace->depth;
907 
908 		/* Append this call time to the parent time to subtract */
909 		if (index)
910 			current->ret_stack[index - 1].subtime += calltime;
911 
912 		if (current->ret_stack[index].subtime < calltime)
913 			calltime -= current->ret_stack[index].subtime;
914 		else
915 			calltime = 0;
916 	}
917 
918 	rec = ftrace_find_profiled_func(stat, trace->func);
919 	if (rec) {
920 		rec->time += calltime;
921 		rec->time_squared += calltime * calltime;
922 	}
923 
924  out:
925 	local_irq_restore(flags);
926 }
927 
928 static int register_ftrace_profiler(void)
929 {
930 	return register_ftrace_graph(&profile_graph_return,
931 				     &profile_graph_entry);
932 }
933 
934 static void unregister_ftrace_profiler(void)
935 {
936 	unregister_ftrace_graph();
937 }
938 #else
939 static struct ftrace_ops ftrace_profile_ops __read_mostly = {
940 	.func		= function_profile_call,
941 	.flags		= FTRACE_OPS_FL_RECURSION_SAFE | FTRACE_OPS_FL_INITIALIZED,
942 	INIT_OPS_HASH(ftrace_profile_ops)
943 };
944 
945 static int register_ftrace_profiler(void)
946 {
947 	return register_ftrace_function(&ftrace_profile_ops);
948 }
949 
950 static void unregister_ftrace_profiler(void)
951 {
952 	unregister_ftrace_function(&ftrace_profile_ops);
953 }
954 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
955 
956 static ssize_t
957 ftrace_profile_write(struct file *filp, const char __user *ubuf,
958 		     size_t cnt, loff_t *ppos)
959 {
960 	unsigned long val;
961 	int ret;
962 
963 	ret = kstrtoul_from_user(ubuf, cnt, 10, &val);
964 	if (ret)
965 		return ret;
966 
967 	val = !!val;
968 
969 	mutex_lock(&ftrace_profile_lock);
970 	if (ftrace_profile_enabled ^ val) {
971 		if (val) {
972 			ret = ftrace_profile_init();
973 			if (ret < 0) {
974 				cnt = ret;
975 				goto out;
976 			}
977 
978 			ret = register_ftrace_profiler();
979 			if (ret < 0) {
980 				cnt = ret;
981 				goto out;
982 			}
983 			ftrace_profile_enabled = 1;
984 		} else {
985 			ftrace_profile_enabled = 0;
986 			/*
987 			 * unregister_ftrace_profiler calls stop_machine
988 			 * so this acts like an synchronize_sched.
989 			 */
990 			unregister_ftrace_profiler();
991 		}
992 	}
993  out:
994 	mutex_unlock(&ftrace_profile_lock);
995 
996 	*ppos += cnt;
997 
998 	return cnt;
999 }
1000 
1001 static ssize_t
1002 ftrace_profile_read(struct file *filp, char __user *ubuf,
1003 		     size_t cnt, loff_t *ppos)
1004 {
1005 	char buf[64];		/* big enough to hold a number */
1006 	int r;
1007 
1008 	r = sprintf(buf, "%u\n", ftrace_profile_enabled);
1009 	return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
1010 }
1011 
1012 static const struct file_operations ftrace_profile_fops = {
1013 	.open		= tracing_open_generic,
1014 	.read		= ftrace_profile_read,
1015 	.write		= ftrace_profile_write,
1016 	.llseek		= default_llseek,
1017 };
1018 
1019 /* used to initialize the real stat files */
1020 static struct tracer_stat function_stats __initdata = {
1021 	.name		= "functions",
1022 	.stat_start	= function_stat_start,
1023 	.stat_next	= function_stat_next,
1024 	.stat_cmp	= function_stat_cmp,
1025 	.stat_headers	= function_stat_headers,
1026 	.stat_show	= function_stat_show
1027 };
1028 
1029 static __init void ftrace_profile_tracefs(struct dentry *d_tracer)
1030 {
1031 	struct ftrace_profile_stat *stat;
1032 	struct dentry *entry;
1033 	char *name;
1034 	int ret;
1035 	int cpu;
1036 
1037 	for_each_possible_cpu(cpu) {
1038 		stat = &per_cpu(ftrace_profile_stats, cpu);
1039 
1040 		name = kasprintf(GFP_KERNEL, "function%d", cpu);
1041 		if (!name) {
1042 			/*
1043 			 * The files created are permanent, if something happens
1044 			 * we still do not free memory.
1045 			 */
1046 			WARN(1,
1047 			     "Could not allocate stat file for cpu %d\n",
1048 			     cpu);
1049 			return;
1050 		}
1051 		stat->stat = function_stats;
1052 		stat->stat.name = name;
1053 		ret = register_stat_tracer(&stat->stat);
1054 		if (ret) {
1055 			WARN(1,
1056 			     "Could not register function stat for cpu %d\n",
1057 			     cpu);
1058 			kfree(name);
1059 			return;
1060 		}
1061 	}
1062 
1063 	entry = tracefs_create_file("function_profile_enabled", 0644,
1064 				    d_tracer, NULL, &ftrace_profile_fops);
1065 	if (!entry)
1066 		pr_warn("Could not create tracefs 'function_profile_enabled' entry\n");
1067 }
1068 
1069 #else /* CONFIG_FUNCTION_PROFILER */
1070 static __init void ftrace_profile_tracefs(struct dentry *d_tracer)
1071 {
1072 }
1073 #endif /* CONFIG_FUNCTION_PROFILER */
1074 
1075 static struct pid * const ftrace_swapper_pid = &init_struct_pid;
1076 
1077 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
1078 static int ftrace_graph_active;
1079 #else
1080 # define ftrace_graph_active 0
1081 #endif
1082 
1083 #ifdef CONFIG_DYNAMIC_FTRACE
1084 
1085 static struct ftrace_ops *removed_ops;
1086 
1087 /*
1088  * Set when doing a global update, like enabling all recs or disabling them.
1089  * It is not set when just updating a single ftrace_ops.
1090  */
1091 static bool update_all_ops;
1092 
1093 #ifndef CONFIG_FTRACE_MCOUNT_RECORD
1094 # error Dynamic ftrace depends on MCOUNT_RECORD
1095 #endif
1096 
1097 static struct hlist_head ftrace_func_hash[FTRACE_FUNC_HASHSIZE] __read_mostly;
1098 
1099 struct ftrace_func_probe {
1100 	struct hlist_node	node;
1101 	struct ftrace_probe_ops	*ops;
1102 	unsigned long		flags;
1103 	unsigned long		ip;
1104 	void			*data;
1105 	struct list_head	free_list;
1106 };
1107 
1108 struct ftrace_func_entry {
1109 	struct hlist_node hlist;
1110 	unsigned long ip;
1111 };
1112 
1113 struct ftrace_hash {
1114 	unsigned long		size_bits;
1115 	struct hlist_head	*buckets;
1116 	unsigned long		count;
1117 	struct rcu_head		rcu;
1118 };
1119 
1120 /*
1121  * We make these constant because no one should touch them,
1122  * but they are used as the default "empty hash", to avoid allocating
1123  * it all the time. These are in a read only section such that if
1124  * anyone does try to modify it, it will cause an exception.
1125  */
1126 static const struct hlist_head empty_buckets[1];
1127 static const struct ftrace_hash empty_hash = {
1128 	.buckets = (struct hlist_head *)empty_buckets,
1129 };
1130 #define EMPTY_HASH	((struct ftrace_hash *)&empty_hash)
1131 
1132 static struct ftrace_ops global_ops = {
1133 	.func				= ftrace_stub,
1134 	.local_hash.notrace_hash	= EMPTY_HASH,
1135 	.local_hash.filter_hash		= EMPTY_HASH,
1136 	INIT_OPS_HASH(global_ops)
1137 	.flags				= FTRACE_OPS_FL_RECURSION_SAFE |
1138 					  FTRACE_OPS_FL_INITIALIZED |
1139 					  FTRACE_OPS_FL_PID,
1140 };
1141 
1142 /*
1143  * This is used by __kernel_text_address() to return true if the
1144  * address is on a dynamically allocated trampoline that would
1145  * not return true for either core_kernel_text() or
1146  * is_module_text_address().
1147  */
1148 bool is_ftrace_trampoline(unsigned long addr)
1149 {
1150 	struct ftrace_ops *op;
1151 	bool ret = false;
1152 
1153 	/*
1154 	 * Some of the ops may be dynamically allocated,
1155 	 * they are freed after a synchronize_sched().
1156 	 */
1157 	preempt_disable_notrace();
1158 
1159 	do_for_each_ftrace_op(op, ftrace_ops_list) {
1160 		/*
1161 		 * This is to check for dynamically allocated trampolines.
1162 		 * Trampolines that are in kernel text will have
1163 		 * core_kernel_text() return true.
1164 		 */
1165 		if (op->trampoline && op->trampoline_size)
1166 			if (addr >= op->trampoline &&
1167 			    addr < op->trampoline + op->trampoline_size) {
1168 				ret = true;
1169 				goto out;
1170 			}
1171 	} while_for_each_ftrace_op(op);
1172 
1173  out:
1174 	preempt_enable_notrace();
1175 
1176 	return ret;
1177 }
1178 
1179 struct ftrace_page {
1180 	struct ftrace_page	*next;
1181 	struct dyn_ftrace	*records;
1182 	int			index;
1183 	int			size;
1184 };
1185 
1186 #define ENTRY_SIZE sizeof(struct dyn_ftrace)
1187 #define ENTRIES_PER_PAGE (PAGE_SIZE / ENTRY_SIZE)
1188 
1189 /* estimate from running different kernels */
1190 #define NR_TO_INIT		10000
1191 
1192 static struct ftrace_page	*ftrace_pages_start;
1193 static struct ftrace_page	*ftrace_pages;
1194 
1195 static bool __always_inline ftrace_hash_empty(struct ftrace_hash *hash)
1196 {
1197 	return !hash || !hash->count;
1198 }
1199 
1200 static struct ftrace_func_entry *
1201 ftrace_lookup_ip(struct ftrace_hash *hash, unsigned long ip)
1202 {
1203 	unsigned long key;
1204 	struct ftrace_func_entry *entry;
1205 	struct hlist_head *hhd;
1206 
1207 	if (ftrace_hash_empty(hash))
1208 		return NULL;
1209 
1210 	if (hash->size_bits > 0)
1211 		key = hash_long(ip, hash->size_bits);
1212 	else
1213 		key = 0;
1214 
1215 	hhd = &hash->buckets[key];
1216 
1217 	hlist_for_each_entry_rcu_notrace(entry, hhd, hlist) {
1218 		if (entry->ip == ip)
1219 			return entry;
1220 	}
1221 	return NULL;
1222 }
1223 
1224 static void __add_hash_entry(struct ftrace_hash *hash,
1225 			     struct ftrace_func_entry *entry)
1226 {
1227 	struct hlist_head *hhd;
1228 	unsigned long key;
1229 
1230 	if (hash->size_bits)
1231 		key = hash_long(entry->ip, hash->size_bits);
1232 	else
1233 		key = 0;
1234 
1235 	hhd = &hash->buckets[key];
1236 	hlist_add_head(&entry->hlist, hhd);
1237 	hash->count++;
1238 }
1239 
1240 static int add_hash_entry(struct ftrace_hash *hash, unsigned long ip)
1241 {
1242 	struct ftrace_func_entry *entry;
1243 
1244 	entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1245 	if (!entry)
1246 		return -ENOMEM;
1247 
1248 	entry->ip = ip;
1249 	__add_hash_entry(hash, entry);
1250 
1251 	return 0;
1252 }
1253 
1254 static void
1255 free_hash_entry(struct ftrace_hash *hash,
1256 		  struct ftrace_func_entry *entry)
1257 {
1258 	hlist_del(&entry->hlist);
1259 	kfree(entry);
1260 	hash->count--;
1261 }
1262 
1263 static void
1264 remove_hash_entry(struct ftrace_hash *hash,
1265 		  struct ftrace_func_entry *entry)
1266 {
1267 	hlist_del(&entry->hlist);
1268 	hash->count--;
1269 }
1270 
1271 static void ftrace_hash_clear(struct ftrace_hash *hash)
1272 {
1273 	struct hlist_head *hhd;
1274 	struct hlist_node *tn;
1275 	struct ftrace_func_entry *entry;
1276 	int size = 1 << hash->size_bits;
1277 	int i;
1278 
1279 	if (!hash->count)
1280 		return;
1281 
1282 	for (i = 0; i < size; i++) {
1283 		hhd = &hash->buckets[i];
1284 		hlist_for_each_entry_safe(entry, tn, hhd, hlist)
1285 			free_hash_entry(hash, entry);
1286 	}
1287 	FTRACE_WARN_ON(hash->count);
1288 }
1289 
1290 static void free_ftrace_hash(struct ftrace_hash *hash)
1291 {
1292 	if (!hash || hash == EMPTY_HASH)
1293 		return;
1294 	ftrace_hash_clear(hash);
1295 	kfree(hash->buckets);
1296 	kfree(hash);
1297 }
1298 
1299 static void __free_ftrace_hash_rcu(struct rcu_head *rcu)
1300 {
1301 	struct ftrace_hash *hash;
1302 
1303 	hash = container_of(rcu, struct ftrace_hash, rcu);
1304 	free_ftrace_hash(hash);
1305 }
1306 
1307 static void free_ftrace_hash_rcu(struct ftrace_hash *hash)
1308 {
1309 	if (!hash || hash == EMPTY_HASH)
1310 		return;
1311 	call_rcu_sched(&hash->rcu, __free_ftrace_hash_rcu);
1312 }
1313 
1314 void ftrace_free_filter(struct ftrace_ops *ops)
1315 {
1316 	ftrace_ops_init(ops);
1317 	free_ftrace_hash(ops->func_hash->filter_hash);
1318 	free_ftrace_hash(ops->func_hash->notrace_hash);
1319 }
1320 
1321 static struct ftrace_hash *alloc_ftrace_hash(int size_bits)
1322 {
1323 	struct ftrace_hash *hash;
1324 	int size;
1325 
1326 	hash = kzalloc(sizeof(*hash), GFP_KERNEL);
1327 	if (!hash)
1328 		return NULL;
1329 
1330 	size = 1 << size_bits;
1331 	hash->buckets = kcalloc(size, sizeof(*hash->buckets), GFP_KERNEL);
1332 
1333 	if (!hash->buckets) {
1334 		kfree(hash);
1335 		return NULL;
1336 	}
1337 
1338 	hash->size_bits = size_bits;
1339 
1340 	return hash;
1341 }
1342 
1343 static struct ftrace_hash *
1344 alloc_and_copy_ftrace_hash(int size_bits, struct ftrace_hash *hash)
1345 {
1346 	struct ftrace_func_entry *entry;
1347 	struct ftrace_hash *new_hash;
1348 	int size;
1349 	int ret;
1350 	int i;
1351 
1352 	new_hash = alloc_ftrace_hash(size_bits);
1353 	if (!new_hash)
1354 		return NULL;
1355 
1356 	/* Empty hash? */
1357 	if (ftrace_hash_empty(hash))
1358 		return new_hash;
1359 
1360 	size = 1 << hash->size_bits;
1361 	for (i = 0; i < size; i++) {
1362 		hlist_for_each_entry(entry, &hash->buckets[i], hlist) {
1363 			ret = add_hash_entry(new_hash, entry->ip);
1364 			if (ret < 0)
1365 				goto free_hash;
1366 		}
1367 	}
1368 
1369 	FTRACE_WARN_ON(new_hash->count != hash->count);
1370 
1371 	return new_hash;
1372 
1373  free_hash:
1374 	free_ftrace_hash(new_hash);
1375 	return NULL;
1376 }
1377 
1378 static void
1379 ftrace_hash_rec_disable_modify(struct ftrace_ops *ops, int filter_hash);
1380 static void
1381 ftrace_hash_rec_enable_modify(struct ftrace_ops *ops, int filter_hash);
1382 
1383 static int ftrace_hash_ipmodify_update(struct ftrace_ops *ops,
1384 				       struct ftrace_hash *new_hash);
1385 
1386 static int
1387 ftrace_hash_move(struct ftrace_ops *ops, int enable,
1388 		 struct ftrace_hash **dst, struct ftrace_hash *src)
1389 {
1390 	struct ftrace_func_entry *entry;
1391 	struct hlist_node *tn;
1392 	struct hlist_head *hhd;
1393 	struct ftrace_hash *new_hash;
1394 	int size = src->count;
1395 	int bits = 0;
1396 	int ret;
1397 	int i;
1398 
1399 	/* Reject setting notrace hash on IPMODIFY ftrace_ops */
1400 	if (ops->flags & FTRACE_OPS_FL_IPMODIFY && !enable)
1401 		return -EINVAL;
1402 
1403 	/*
1404 	 * If the new source is empty, just free dst and assign it
1405 	 * the empty_hash.
1406 	 */
1407 	if (!src->count) {
1408 		new_hash = EMPTY_HASH;
1409 		goto update;
1410 	}
1411 
1412 	/*
1413 	 * Make the hash size about 1/2 the # found
1414 	 */
1415 	for (size /= 2; size; size >>= 1)
1416 		bits++;
1417 
1418 	/* Don't allocate too much */
1419 	if (bits > FTRACE_HASH_MAX_BITS)
1420 		bits = FTRACE_HASH_MAX_BITS;
1421 
1422 	new_hash = alloc_ftrace_hash(bits);
1423 	if (!new_hash)
1424 		return -ENOMEM;
1425 
1426 	size = 1 << src->size_bits;
1427 	for (i = 0; i < size; i++) {
1428 		hhd = &src->buckets[i];
1429 		hlist_for_each_entry_safe(entry, tn, hhd, hlist) {
1430 			remove_hash_entry(src, entry);
1431 			__add_hash_entry(new_hash, entry);
1432 		}
1433 	}
1434 
1435 update:
1436 	/* Make sure this can be applied if it is IPMODIFY ftrace_ops */
1437 	if (enable) {
1438 		/* IPMODIFY should be updated only when filter_hash updating */
1439 		ret = ftrace_hash_ipmodify_update(ops, new_hash);
1440 		if (ret < 0) {
1441 			free_ftrace_hash(new_hash);
1442 			return ret;
1443 		}
1444 	}
1445 
1446 	/*
1447 	 * Remove the current set, update the hash and add
1448 	 * them back.
1449 	 */
1450 	ftrace_hash_rec_disable_modify(ops, enable);
1451 
1452 	rcu_assign_pointer(*dst, new_hash);
1453 
1454 	ftrace_hash_rec_enable_modify(ops, enable);
1455 
1456 	return 0;
1457 }
1458 
1459 static bool hash_contains_ip(unsigned long ip,
1460 			     struct ftrace_ops_hash *hash)
1461 {
1462 	/*
1463 	 * The function record is a match if it exists in the filter
1464 	 * hash and not in the notrace hash. Note, an emty hash is
1465 	 * considered a match for the filter hash, but an empty
1466 	 * notrace hash is considered not in the notrace hash.
1467 	 */
1468 	return (ftrace_hash_empty(hash->filter_hash) ||
1469 		ftrace_lookup_ip(hash->filter_hash, ip)) &&
1470 		(ftrace_hash_empty(hash->notrace_hash) ||
1471 		 !ftrace_lookup_ip(hash->notrace_hash, ip));
1472 }
1473 
1474 /*
1475  * Test the hashes for this ops to see if we want to call
1476  * the ops->func or not.
1477  *
1478  * It's a match if the ip is in the ops->filter_hash or
1479  * the filter_hash does not exist or is empty,
1480  *  AND
1481  * the ip is not in the ops->notrace_hash.
1482  *
1483  * This needs to be called with preemption disabled as
1484  * the hashes are freed with call_rcu_sched().
1485  */
1486 static int
1487 ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip, void *regs)
1488 {
1489 	struct ftrace_ops_hash hash;
1490 	int ret;
1491 
1492 #ifdef CONFIG_DYNAMIC_FTRACE_WITH_REGS
1493 	/*
1494 	 * There's a small race when adding ops that the ftrace handler
1495 	 * that wants regs, may be called without them. We can not
1496 	 * allow that handler to be called if regs is NULL.
1497 	 */
1498 	if (regs == NULL && (ops->flags & FTRACE_OPS_FL_SAVE_REGS))
1499 		return 0;
1500 #endif
1501 
1502 	hash.filter_hash = rcu_dereference_raw_notrace(ops->func_hash->filter_hash);
1503 	hash.notrace_hash = rcu_dereference_raw_notrace(ops->func_hash->notrace_hash);
1504 
1505 	if (hash_contains_ip(ip, &hash))
1506 		ret = 1;
1507 	else
1508 		ret = 0;
1509 
1510 	return ret;
1511 }
1512 
1513 /*
1514  * This is a double for. Do not use 'break' to break out of the loop,
1515  * you must use a goto.
1516  */
1517 #define do_for_each_ftrace_rec(pg, rec)					\
1518 	for (pg = ftrace_pages_start; pg; pg = pg->next) {		\
1519 		int _____i;						\
1520 		for (_____i = 0; _____i < pg->index; _____i++) {	\
1521 			rec = &pg->records[_____i];
1522 
1523 #define while_for_each_ftrace_rec()		\
1524 		}				\
1525 	}
1526 
1527 
1528 static int ftrace_cmp_recs(const void *a, const void *b)
1529 {
1530 	const struct dyn_ftrace *key = a;
1531 	const struct dyn_ftrace *rec = b;
1532 
1533 	if (key->flags < rec->ip)
1534 		return -1;
1535 	if (key->ip >= rec->ip + MCOUNT_INSN_SIZE)
1536 		return 1;
1537 	return 0;
1538 }
1539 
1540 /**
1541  * ftrace_location_range - return the first address of a traced location
1542  *	if it touches the given ip range
1543  * @start: start of range to search.
1544  * @end: end of range to search (inclusive). @end points to the last byte
1545  *	to check.
1546  *
1547  * Returns rec->ip if the related ftrace location is a least partly within
1548  * the given address range. That is, the first address of the instruction
1549  * that is either a NOP or call to the function tracer. It checks the ftrace
1550  * internal tables to determine if the address belongs or not.
1551  */
1552 unsigned long ftrace_location_range(unsigned long start, unsigned long end)
1553 {
1554 	struct ftrace_page *pg;
1555 	struct dyn_ftrace *rec;
1556 	struct dyn_ftrace key;
1557 
1558 	key.ip = start;
1559 	key.flags = end;	/* overload flags, as it is unsigned long */
1560 
1561 	for (pg = ftrace_pages_start; pg; pg = pg->next) {
1562 		if (end < pg->records[0].ip ||
1563 		    start >= (pg->records[pg->index - 1].ip + MCOUNT_INSN_SIZE))
1564 			continue;
1565 		rec = bsearch(&key, pg->records, pg->index,
1566 			      sizeof(struct dyn_ftrace),
1567 			      ftrace_cmp_recs);
1568 		if (rec)
1569 			return rec->ip;
1570 	}
1571 
1572 	return 0;
1573 }
1574 
1575 /**
1576  * ftrace_location - return true if the ip giving is a traced location
1577  * @ip: the instruction pointer to check
1578  *
1579  * Returns rec->ip if @ip given is a pointer to a ftrace location.
1580  * That is, the instruction that is either a NOP or call to
1581  * the function tracer. It checks the ftrace internal tables to
1582  * determine if the address belongs or not.
1583  */
1584 unsigned long ftrace_location(unsigned long ip)
1585 {
1586 	return ftrace_location_range(ip, ip);
1587 }
1588 
1589 /**
1590  * ftrace_text_reserved - return true if range contains an ftrace location
1591  * @start: start of range to search
1592  * @end: end of range to search (inclusive). @end points to the last byte to check.
1593  *
1594  * Returns 1 if @start and @end contains a ftrace location.
1595  * That is, the instruction that is either a NOP or call to
1596  * the function tracer. It checks the ftrace internal tables to
1597  * determine if the address belongs or not.
1598  */
1599 int ftrace_text_reserved(const void *start, const void *end)
1600 {
1601 	unsigned long ret;
1602 
1603 	ret = ftrace_location_range((unsigned long)start,
1604 				    (unsigned long)end);
1605 
1606 	return (int)!!ret;
1607 }
1608 
1609 /* Test if ops registered to this rec needs regs */
1610 static bool test_rec_ops_needs_regs(struct dyn_ftrace *rec)
1611 {
1612 	struct ftrace_ops *ops;
1613 	bool keep_regs = false;
1614 
1615 	for (ops = ftrace_ops_list;
1616 	     ops != &ftrace_list_end; ops = ops->next) {
1617 		/* pass rec in as regs to have non-NULL val */
1618 		if (ftrace_ops_test(ops, rec->ip, rec)) {
1619 			if (ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
1620 				keep_regs = true;
1621 				break;
1622 			}
1623 		}
1624 	}
1625 
1626 	return  keep_regs;
1627 }
1628 
1629 static bool __ftrace_hash_rec_update(struct ftrace_ops *ops,
1630 				     int filter_hash,
1631 				     bool inc)
1632 {
1633 	struct ftrace_hash *hash;
1634 	struct ftrace_hash *other_hash;
1635 	struct ftrace_page *pg;
1636 	struct dyn_ftrace *rec;
1637 	bool update = false;
1638 	int count = 0;
1639 	int all = 0;
1640 
1641 	/* Only update if the ops has been registered */
1642 	if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1643 		return false;
1644 
1645 	/*
1646 	 * In the filter_hash case:
1647 	 *   If the count is zero, we update all records.
1648 	 *   Otherwise we just update the items in the hash.
1649 	 *
1650 	 * In the notrace_hash case:
1651 	 *   We enable the update in the hash.
1652 	 *   As disabling notrace means enabling the tracing,
1653 	 *   and enabling notrace means disabling, the inc variable
1654 	 *   gets inversed.
1655 	 */
1656 	if (filter_hash) {
1657 		hash = ops->func_hash->filter_hash;
1658 		other_hash = ops->func_hash->notrace_hash;
1659 		if (ftrace_hash_empty(hash))
1660 			all = 1;
1661 	} else {
1662 		inc = !inc;
1663 		hash = ops->func_hash->notrace_hash;
1664 		other_hash = ops->func_hash->filter_hash;
1665 		/*
1666 		 * If the notrace hash has no items,
1667 		 * then there's nothing to do.
1668 		 */
1669 		if (ftrace_hash_empty(hash))
1670 			return false;
1671 	}
1672 
1673 	do_for_each_ftrace_rec(pg, rec) {
1674 		int in_other_hash = 0;
1675 		int in_hash = 0;
1676 		int match = 0;
1677 
1678 		if (rec->flags & FTRACE_FL_DISABLED)
1679 			continue;
1680 
1681 		if (all) {
1682 			/*
1683 			 * Only the filter_hash affects all records.
1684 			 * Update if the record is not in the notrace hash.
1685 			 */
1686 			if (!other_hash || !ftrace_lookup_ip(other_hash, rec->ip))
1687 				match = 1;
1688 		} else {
1689 			in_hash = !!ftrace_lookup_ip(hash, rec->ip);
1690 			in_other_hash = !!ftrace_lookup_ip(other_hash, rec->ip);
1691 
1692 			/*
1693 			 * If filter_hash is set, we want to match all functions
1694 			 * that are in the hash but not in the other hash.
1695 			 *
1696 			 * If filter_hash is not set, then we are decrementing.
1697 			 * That means we match anything that is in the hash
1698 			 * and also in the other_hash. That is, we need to turn
1699 			 * off functions in the other hash because they are disabled
1700 			 * by this hash.
1701 			 */
1702 			if (filter_hash && in_hash && !in_other_hash)
1703 				match = 1;
1704 			else if (!filter_hash && in_hash &&
1705 				 (in_other_hash || ftrace_hash_empty(other_hash)))
1706 				match = 1;
1707 		}
1708 		if (!match)
1709 			continue;
1710 
1711 		if (inc) {
1712 			rec->flags++;
1713 			if (FTRACE_WARN_ON(ftrace_rec_count(rec) == FTRACE_REF_MAX))
1714 				return false;
1715 
1716 			/*
1717 			 * If there's only a single callback registered to a
1718 			 * function, and the ops has a trampoline registered
1719 			 * for it, then we can call it directly.
1720 			 */
1721 			if (ftrace_rec_count(rec) == 1 && ops->trampoline)
1722 				rec->flags |= FTRACE_FL_TRAMP;
1723 			else
1724 				/*
1725 				 * If we are adding another function callback
1726 				 * to this function, and the previous had a
1727 				 * custom trampoline in use, then we need to go
1728 				 * back to the default trampoline.
1729 				 */
1730 				rec->flags &= ~FTRACE_FL_TRAMP;
1731 
1732 			/*
1733 			 * If any ops wants regs saved for this function
1734 			 * then all ops will get saved regs.
1735 			 */
1736 			if (ops->flags & FTRACE_OPS_FL_SAVE_REGS)
1737 				rec->flags |= FTRACE_FL_REGS;
1738 		} else {
1739 			if (FTRACE_WARN_ON(ftrace_rec_count(rec) == 0))
1740 				return false;
1741 			rec->flags--;
1742 
1743 			/*
1744 			 * If the rec had REGS enabled and the ops that is
1745 			 * being removed had REGS set, then see if there is
1746 			 * still any ops for this record that wants regs.
1747 			 * If not, we can stop recording them.
1748 			 */
1749 			if (ftrace_rec_count(rec) > 0 &&
1750 			    rec->flags & FTRACE_FL_REGS &&
1751 			    ops->flags & FTRACE_OPS_FL_SAVE_REGS) {
1752 				if (!test_rec_ops_needs_regs(rec))
1753 					rec->flags &= ~FTRACE_FL_REGS;
1754 			}
1755 
1756 			/*
1757 			 * If the rec had TRAMP enabled, then it needs to
1758 			 * be cleared. As TRAMP can only be enabled iff
1759 			 * there is only a single ops attached to it.
1760 			 * In otherwords, always disable it on decrementing.
1761 			 * In the future, we may set it if rec count is
1762 			 * decremented to one, and the ops that is left
1763 			 * has a trampoline.
1764 			 */
1765 			rec->flags &= ~FTRACE_FL_TRAMP;
1766 
1767 			/*
1768 			 * flags will be cleared in ftrace_check_record()
1769 			 * if rec count is zero.
1770 			 */
1771 		}
1772 		count++;
1773 
1774 		/* Must match FTRACE_UPDATE_CALLS in ftrace_modify_all_code() */
1775 		update |= ftrace_test_record(rec, 1) != FTRACE_UPDATE_IGNORE;
1776 
1777 		/* Shortcut, if we handled all records, we are done. */
1778 		if (!all && count == hash->count)
1779 			return update;
1780 	} while_for_each_ftrace_rec();
1781 
1782 	return update;
1783 }
1784 
1785 static bool ftrace_hash_rec_disable(struct ftrace_ops *ops,
1786 				    int filter_hash)
1787 {
1788 	return __ftrace_hash_rec_update(ops, filter_hash, 0);
1789 }
1790 
1791 static bool ftrace_hash_rec_enable(struct ftrace_ops *ops,
1792 				   int filter_hash)
1793 {
1794 	return __ftrace_hash_rec_update(ops, filter_hash, 1);
1795 }
1796 
1797 static void ftrace_hash_rec_update_modify(struct ftrace_ops *ops,
1798 					  int filter_hash, int inc)
1799 {
1800 	struct ftrace_ops *op;
1801 
1802 	__ftrace_hash_rec_update(ops, filter_hash, inc);
1803 
1804 	if (ops->func_hash != &global_ops.local_hash)
1805 		return;
1806 
1807 	/*
1808 	 * If the ops shares the global_ops hash, then we need to update
1809 	 * all ops that are enabled and use this hash.
1810 	 */
1811 	do_for_each_ftrace_op(op, ftrace_ops_list) {
1812 		/* Already done */
1813 		if (op == ops)
1814 			continue;
1815 		if (op->func_hash == &global_ops.local_hash)
1816 			__ftrace_hash_rec_update(op, filter_hash, inc);
1817 	} while_for_each_ftrace_op(op);
1818 }
1819 
1820 static void ftrace_hash_rec_disable_modify(struct ftrace_ops *ops,
1821 					   int filter_hash)
1822 {
1823 	ftrace_hash_rec_update_modify(ops, filter_hash, 0);
1824 }
1825 
1826 static void ftrace_hash_rec_enable_modify(struct ftrace_ops *ops,
1827 					  int filter_hash)
1828 {
1829 	ftrace_hash_rec_update_modify(ops, filter_hash, 1);
1830 }
1831 
1832 /*
1833  * Try to update IPMODIFY flag on each ftrace_rec. Return 0 if it is OK
1834  * or no-needed to update, -EBUSY if it detects a conflict of the flag
1835  * on a ftrace_rec, and -EINVAL if the new_hash tries to trace all recs.
1836  * Note that old_hash and new_hash has below meanings
1837  *  - If the hash is NULL, it hits all recs (if IPMODIFY is set, this is rejected)
1838  *  - If the hash is EMPTY_HASH, it hits nothing
1839  *  - Anything else hits the recs which match the hash entries.
1840  */
1841 static int __ftrace_hash_update_ipmodify(struct ftrace_ops *ops,
1842 					 struct ftrace_hash *old_hash,
1843 					 struct ftrace_hash *new_hash)
1844 {
1845 	struct ftrace_page *pg;
1846 	struct dyn_ftrace *rec, *end = NULL;
1847 	int in_old, in_new;
1848 
1849 	/* Only update if the ops has been registered */
1850 	if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
1851 		return 0;
1852 
1853 	if (!(ops->flags & FTRACE_OPS_FL_IPMODIFY))
1854 		return 0;
1855 
1856 	/*
1857 	 * Since the IPMODIFY is a very address sensitive action, we do not
1858 	 * allow ftrace_ops to set all functions to new hash.
1859 	 */
1860 	if (!new_hash || !old_hash)
1861 		return -EINVAL;
1862 
1863 	/* Update rec->flags */
1864 	do_for_each_ftrace_rec(pg, rec) {
1865 		/* We need to update only differences of filter_hash */
1866 		in_old = !!ftrace_lookup_ip(old_hash, rec->ip);
1867 		in_new = !!ftrace_lookup_ip(new_hash, rec->ip);
1868 		if (in_old == in_new)
1869 			continue;
1870 
1871 		if (in_new) {
1872 			/* New entries must ensure no others are using it */
1873 			if (rec->flags & FTRACE_FL_IPMODIFY)
1874 				goto rollback;
1875 			rec->flags |= FTRACE_FL_IPMODIFY;
1876 		} else /* Removed entry */
1877 			rec->flags &= ~FTRACE_FL_IPMODIFY;
1878 	} while_for_each_ftrace_rec();
1879 
1880 	return 0;
1881 
1882 rollback:
1883 	end = rec;
1884 
1885 	/* Roll back what we did above */
1886 	do_for_each_ftrace_rec(pg, rec) {
1887 		if (rec == end)
1888 			goto err_out;
1889 
1890 		in_old = !!ftrace_lookup_ip(old_hash, rec->ip);
1891 		in_new = !!ftrace_lookup_ip(new_hash, rec->ip);
1892 		if (in_old == in_new)
1893 			continue;
1894 
1895 		if (in_new)
1896 			rec->flags &= ~FTRACE_FL_IPMODIFY;
1897 		else
1898 			rec->flags |= FTRACE_FL_IPMODIFY;
1899 	} while_for_each_ftrace_rec();
1900 
1901 err_out:
1902 	return -EBUSY;
1903 }
1904 
1905 static int ftrace_hash_ipmodify_enable(struct ftrace_ops *ops)
1906 {
1907 	struct ftrace_hash *hash = ops->func_hash->filter_hash;
1908 
1909 	if (ftrace_hash_empty(hash))
1910 		hash = NULL;
1911 
1912 	return __ftrace_hash_update_ipmodify(ops, EMPTY_HASH, hash);
1913 }
1914 
1915 /* Disabling always succeeds */
1916 static void ftrace_hash_ipmodify_disable(struct ftrace_ops *ops)
1917 {
1918 	struct ftrace_hash *hash = ops->func_hash->filter_hash;
1919 
1920 	if (ftrace_hash_empty(hash))
1921 		hash = NULL;
1922 
1923 	__ftrace_hash_update_ipmodify(ops, hash, EMPTY_HASH);
1924 }
1925 
1926 static int ftrace_hash_ipmodify_update(struct ftrace_ops *ops,
1927 				       struct ftrace_hash *new_hash)
1928 {
1929 	struct ftrace_hash *old_hash = ops->func_hash->filter_hash;
1930 
1931 	if (ftrace_hash_empty(old_hash))
1932 		old_hash = NULL;
1933 
1934 	if (ftrace_hash_empty(new_hash))
1935 		new_hash = NULL;
1936 
1937 	return __ftrace_hash_update_ipmodify(ops, old_hash, new_hash);
1938 }
1939 
1940 static void print_ip_ins(const char *fmt, const unsigned char *p)
1941 {
1942 	int i;
1943 
1944 	printk(KERN_CONT "%s", fmt);
1945 
1946 	for (i = 0; i < MCOUNT_INSN_SIZE; i++)
1947 		printk(KERN_CONT "%s%02x", i ? ":" : "", p[i]);
1948 }
1949 
1950 static struct ftrace_ops *
1951 ftrace_find_tramp_ops_any(struct dyn_ftrace *rec);
1952 static struct ftrace_ops *
1953 ftrace_find_tramp_ops_next(struct dyn_ftrace *rec, struct ftrace_ops *ops);
1954 
1955 enum ftrace_bug_type ftrace_bug_type;
1956 const void *ftrace_expected;
1957 
1958 static void print_bug_type(void)
1959 {
1960 	switch (ftrace_bug_type) {
1961 	case FTRACE_BUG_UNKNOWN:
1962 		break;
1963 	case FTRACE_BUG_INIT:
1964 		pr_info("Initializing ftrace call sites\n");
1965 		break;
1966 	case FTRACE_BUG_NOP:
1967 		pr_info("Setting ftrace call site to NOP\n");
1968 		break;
1969 	case FTRACE_BUG_CALL:
1970 		pr_info("Setting ftrace call site to call ftrace function\n");
1971 		break;
1972 	case FTRACE_BUG_UPDATE:
1973 		pr_info("Updating ftrace call site to call a different ftrace function\n");
1974 		break;
1975 	}
1976 }
1977 
1978 /**
1979  * ftrace_bug - report and shutdown function tracer
1980  * @failed: The failed type (EFAULT, EINVAL, EPERM)
1981  * @rec: The record that failed
1982  *
1983  * The arch code that enables or disables the function tracing
1984  * can call ftrace_bug() when it has detected a problem in
1985  * modifying the code. @failed should be one of either:
1986  * EFAULT - if the problem happens on reading the @ip address
1987  * EINVAL - if what is read at @ip is not what was expected
1988  * EPERM - if the problem happens on writting to the @ip address
1989  */
1990 void ftrace_bug(int failed, struct dyn_ftrace *rec)
1991 {
1992 	unsigned long ip = rec ? rec->ip : 0;
1993 
1994 	switch (failed) {
1995 	case -EFAULT:
1996 		FTRACE_WARN_ON_ONCE(1);
1997 		pr_info("ftrace faulted on modifying ");
1998 		print_ip_sym(ip);
1999 		break;
2000 	case -EINVAL:
2001 		FTRACE_WARN_ON_ONCE(1);
2002 		pr_info("ftrace failed to modify ");
2003 		print_ip_sym(ip);
2004 		print_ip_ins(" actual:   ", (unsigned char *)ip);
2005 		pr_cont("\n");
2006 		if (ftrace_expected) {
2007 			print_ip_ins(" expected: ", ftrace_expected);
2008 			pr_cont("\n");
2009 		}
2010 		break;
2011 	case -EPERM:
2012 		FTRACE_WARN_ON_ONCE(1);
2013 		pr_info("ftrace faulted on writing ");
2014 		print_ip_sym(ip);
2015 		break;
2016 	default:
2017 		FTRACE_WARN_ON_ONCE(1);
2018 		pr_info("ftrace faulted on unknown error ");
2019 		print_ip_sym(ip);
2020 	}
2021 	print_bug_type();
2022 	if (rec) {
2023 		struct ftrace_ops *ops = NULL;
2024 
2025 		pr_info("ftrace record flags: %lx\n", rec->flags);
2026 		pr_cont(" (%ld)%s", ftrace_rec_count(rec),
2027 			rec->flags & FTRACE_FL_REGS ? " R" : "  ");
2028 		if (rec->flags & FTRACE_FL_TRAMP_EN) {
2029 			ops = ftrace_find_tramp_ops_any(rec);
2030 			if (ops) {
2031 				do {
2032 					pr_cont("\ttramp: %pS (%pS)",
2033 						(void *)ops->trampoline,
2034 						(void *)ops->func);
2035 					ops = ftrace_find_tramp_ops_next(rec, ops);
2036 				} while (ops);
2037 			} else
2038 				pr_cont("\ttramp: ERROR!");
2039 
2040 		}
2041 		ip = ftrace_get_addr_curr(rec);
2042 		pr_cont("\n expected tramp: %lx\n", ip);
2043 	}
2044 }
2045 
2046 static int ftrace_check_record(struct dyn_ftrace *rec, int enable, int update)
2047 {
2048 	unsigned long flag = 0UL;
2049 
2050 	ftrace_bug_type = FTRACE_BUG_UNKNOWN;
2051 
2052 	if (rec->flags & FTRACE_FL_DISABLED)
2053 		return FTRACE_UPDATE_IGNORE;
2054 
2055 	/*
2056 	 * If we are updating calls:
2057 	 *
2058 	 *   If the record has a ref count, then we need to enable it
2059 	 *   because someone is using it.
2060 	 *
2061 	 *   Otherwise we make sure its disabled.
2062 	 *
2063 	 * If we are disabling calls, then disable all records that
2064 	 * are enabled.
2065 	 */
2066 	if (enable && ftrace_rec_count(rec))
2067 		flag = FTRACE_FL_ENABLED;
2068 
2069 	/*
2070 	 * If enabling and the REGS flag does not match the REGS_EN, or
2071 	 * the TRAMP flag doesn't match the TRAMP_EN, then do not ignore
2072 	 * this record. Set flags to fail the compare against ENABLED.
2073 	 */
2074 	if (flag) {
2075 		if (!(rec->flags & FTRACE_FL_REGS) !=
2076 		    !(rec->flags & FTRACE_FL_REGS_EN))
2077 			flag |= FTRACE_FL_REGS;
2078 
2079 		if (!(rec->flags & FTRACE_FL_TRAMP) !=
2080 		    !(rec->flags & FTRACE_FL_TRAMP_EN))
2081 			flag |= FTRACE_FL_TRAMP;
2082 	}
2083 
2084 	/* If the state of this record hasn't changed, then do nothing */
2085 	if ((rec->flags & FTRACE_FL_ENABLED) == flag)
2086 		return FTRACE_UPDATE_IGNORE;
2087 
2088 	if (flag) {
2089 		/* Save off if rec is being enabled (for return value) */
2090 		flag ^= rec->flags & FTRACE_FL_ENABLED;
2091 
2092 		if (update) {
2093 			rec->flags |= FTRACE_FL_ENABLED;
2094 			if (flag & FTRACE_FL_REGS) {
2095 				if (rec->flags & FTRACE_FL_REGS)
2096 					rec->flags |= FTRACE_FL_REGS_EN;
2097 				else
2098 					rec->flags &= ~FTRACE_FL_REGS_EN;
2099 			}
2100 			if (flag & FTRACE_FL_TRAMP) {
2101 				if (rec->flags & FTRACE_FL_TRAMP)
2102 					rec->flags |= FTRACE_FL_TRAMP_EN;
2103 				else
2104 					rec->flags &= ~FTRACE_FL_TRAMP_EN;
2105 			}
2106 		}
2107 
2108 		/*
2109 		 * If this record is being updated from a nop, then
2110 		 *   return UPDATE_MAKE_CALL.
2111 		 * Otherwise,
2112 		 *   return UPDATE_MODIFY_CALL to tell the caller to convert
2113 		 *   from the save regs, to a non-save regs function or
2114 		 *   vice versa, or from a trampoline call.
2115 		 */
2116 		if (flag & FTRACE_FL_ENABLED) {
2117 			ftrace_bug_type = FTRACE_BUG_CALL;
2118 			return FTRACE_UPDATE_MAKE_CALL;
2119 		}
2120 
2121 		ftrace_bug_type = FTRACE_BUG_UPDATE;
2122 		return FTRACE_UPDATE_MODIFY_CALL;
2123 	}
2124 
2125 	if (update) {
2126 		/* If there's no more users, clear all flags */
2127 		if (!ftrace_rec_count(rec))
2128 			rec->flags = 0;
2129 		else
2130 			/*
2131 			 * Just disable the record, but keep the ops TRAMP
2132 			 * and REGS states. The _EN flags must be disabled though.
2133 			 */
2134 			rec->flags &= ~(FTRACE_FL_ENABLED | FTRACE_FL_TRAMP_EN |
2135 					FTRACE_FL_REGS_EN);
2136 	}
2137 
2138 	ftrace_bug_type = FTRACE_BUG_NOP;
2139 	return FTRACE_UPDATE_MAKE_NOP;
2140 }
2141 
2142 /**
2143  * ftrace_update_record, set a record that now is tracing or not
2144  * @rec: the record to update
2145  * @enable: set to 1 if the record is tracing, zero to force disable
2146  *
2147  * The records that represent all functions that can be traced need
2148  * to be updated when tracing has been enabled.
2149  */
2150 int ftrace_update_record(struct dyn_ftrace *rec, int enable)
2151 {
2152 	return ftrace_check_record(rec, enable, 1);
2153 }
2154 
2155 /**
2156  * ftrace_test_record, check if the record has been enabled or not
2157  * @rec: the record to test
2158  * @enable: set to 1 to check if enabled, 0 if it is disabled
2159  *
2160  * The arch code may need to test if a record is already set to
2161  * tracing to determine how to modify the function code that it
2162  * represents.
2163  */
2164 int ftrace_test_record(struct dyn_ftrace *rec, int enable)
2165 {
2166 	return ftrace_check_record(rec, enable, 0);
2167 }
2168 
2169 static struct ftrace_ops *
2170 ftrace_find_tramp_ops_any(struct dyn_ftrace *rec)
2171 {
2172 	struct ftrace_ops *op;
2173 	unsigned long ip = rec->ip;
2174 
2175 	do_for_each_ftrace_op(op, ftrace_ops_list) {
2176 
2177 		if (!op->trampoline)
2178 			continue;
2179 
2180 		if (hash_contains_ip(ip, op->func_hash))
2181 			return op;
2182 	} while_for_each_ftrace_op(op);
2183 
2184 	return NULL;
2185 }
2186 
2187 static struct ftrace_ops *
2188 ftrace_find_tramp_ops_next(struct dyn_ftrace *rec,
2189 			   struct ftrace_ops *op)
2190 {
2191 	unsigned long ip = rec->ip;
2192 
2193 	while_for_each_ftrace_op(op) {
2194 
2195 		if (!op->trampoline)
2196 			continue;
2197 
2198 		if (hash_contains_ip(ip, op->func_hash))
2199 			return op;
2200 	}
2201 
2202 	return NULL;
2203 }
2204 
2205 static struct ftrace_ops *
2206 ftrace_find_tramp_ops_curr(struct dyn_ftrace *rec)
2207 {
2208 	struct ftrace_ops *op;
2209 	unsigned long ip = rec->ip;
2210 
2211 	/*
2212 	 * Need to check removed ops first.
2213 	 * If they are being removed, and this rec has a tramp,
2214 	 * and this rec is in the ops list, then it would be the
2215 	 * one with the tramp.
2216 	 */
2217 	if (removed_ops) {
2218 		if (hash_contains_ip(ip, &removed_ops->old_hash))
2219 			return removed_ops;
2220 	}
2221 
2222 	/*
2223 	 * Need to find the current trampoline for a rec.
2224 	 * Now, a trampoline is only attached to a rec if there
2225 	 * was a single 'ops' attached to it. But this can be called
2226 	 * when we are adding another op to the rec or removing the
2227 	 * current one. Thus, if the op is being added, we can
2228 	 * ignore it because it hasn't attached itself to the rec
2229 	 * yet.
2230 	 *
2231 	 * If an ops is being modified (hooking to different functions)
2232 	 * then we don't care about the new functions that are being
2233 	 * added, just the old ones (that are probably being removed).
2234 	 *
2235 	 * If we are adding an ops to a function that already is using
2236 	 * a trampoline, it needs to be removed (trampolines are only
2237 	 * for single ops connected), then an ops that is not being
2238 	 * modified also needs to be checked.
2239 	 */
2240 	do_for_each_ftrace_op(op, ftrace_ops_list) {
2241 
2242 		if (!op->trampoline)
2243 			continue;
2244 
2245 		/*
2246 		 * If the ops is being added, it hasn't gotten to
2247 		 * the point to be removed from this tree yet.
2248 		 */
2249 		if (op->flags & FTRACE_OPS_FL_ADDING)
2250 			continue;
2251 
2252 
2253 		/*
2254 		 * If the ops is being modified and is in the old
2255 		 * hash, then it is probably being removed from this
2256 		 * function.
2257 		 */
2258 		if ((op->flags & FTRACE_OPS_FL_MODIFYING) &&
2259 		    hash_contains_ip(ip, &op->old_hash))
2260 			return op;
2261 		/*
2262 		 * If the ops is not being added or modified, and it's
2263 		 * in its normal filter hash, then this must be the one
2264 		 * we want!
2265 		 */
2266 		if (!(op->flags & FTRACE_OPS_FL_MODIFYING) &&
2267 		    hash_contains_ip(ip, op->func_hash))
2268 			return op;
2269 
2270 	} while_for_each_ftrace_op(op);
2271 
2272 	return NULL;
2273 }
2274 
2275 static struct ftrace_ops *
2276 ftrace_find_tramp_ops_new(struct dyn_ftrace *rec)
2277 {
2278 	struct ftrace_ops *op;
2279 	unsigned long ip = rec->ip;
2280 
2281 	do_for_each_ftrace_op(op, ftrace_ops_list) {
2282 		/* pass rec in as regs to have non-NULL val */
2283 		if (hash_contains_ip(ip, op->func_hash))
2284 			return op;
2285 	} while_for_each_ftrace_op(op);
2286 
2287 	return NULL;
2288 }
2289 
2290 /**
2291  * ftrace_get_addr_new - Get the call address to set to
2292  * @rec:  The ftrace record descriptor
2293  *
2294  * If the record has the FTRACE_FL_REGS set, that means that it
2295  * wants to convert to a callback that saves all regs. If FTRACE_FL_REGS
2296  * is not not set, then it wants to convert to the normal callback.
2297  *
2298  * Returns the address of the trampoline to set to
2299  */
2300 unsigned long ftrace_get_addr_new(struct dyn_ftrace *rec)
2301 {
2302 	struct ftrace_ops *ops;
2303 
2304 	/* Trampolines take precedence over regs */
2305 	if (rec->flags & FTRACE_FL_TRAMP) {
2306 		ops = ftrace_find_tramp_ops_new(rec);
2307 		if (FTRACE_WARN_ON(!ops || !ops->trampoline)) {
2308 			pr_warn("Bad trampoline accounting at: %p (%pS) (%lx)\n",
2309 				(void *)rec->ip, (void *)rec->ip, rec->flags);
2310 			/* Ftrace is shutting down, return anything */
2311 			return (unsigned long)FTRACE_ADDR;
2312 		}
2313 		return ops->trampoline;
2314 	}
2315 
2316 	if (rec->flags & FTRACE_FL_REGS)
2317 		return (unsigned long)FTRACE_REGS_ADDR;
2318 	else
2319 		return (unsigned long)FTRACE_ADDR;
2320 }
2321 
2322 /**
2323  * ftrace_get_addr_curr - Get the call address that is already there
2324  * @rec:  The ftrace record descriptor
2325  *
2326  * The FTRACE_FL_REGS_EN is set when the record already points to
2327  * a function that saves all the regs. Basically the '_EN' version
2328  * represents the current state of the function.
2329  *
2330  * Returns the address of the trampoline that is currently being called
2331  */
2332 unsigned long ftrace_get_addr_curr(struct dyn_ftrace *rec)
2333 {
2334 	struct ftrace_ops *ops;
2335 
2336 	/* Trampolines take precedence over regs */
2337 	if (rec->flags & FTRACE_FL_TRAMP_EN) {
2338 		ops = ftrace_find_tramp_ops_curr(rec);
2339 		if (FTRACE_WARN_ON(!ops)) {
2340 			pr_warn("Bad trampoline accounting at: %p (%pS)\n",
2341 				(void *)rec->ip, (void *)rec->ip);
2342 			/* Ftrace is shutting down, return anything */
2343 			return (unsigned long)FTRACE_ADDR;
2344 		}
2345 		return ops->trampoline;
2346 	}
2347 
2348 	if (rec->flags & FTRACE_FL_REGS_EN)
2349 		return (unsigned long)FTRACE_REGS_ADDR;
2350 	else
2351 		return (unsigned long)FTRACE_ADDR;
2352 }
2353 
2354 static int
2355 __ftrace_replace_code(struct dyn_ftrace *rec, int enable)
2356 {
2357 	unsigned long ftrace_old_addr;
2358 	unsigned long ftrace_addr;
2359 	int ret;
2360 
2361 	ftrace_addr = ftrace_get_addr_new(rec);
2362 
2363 	/* This needs to be done before we call ftrace_update_record */
2364 	ftrace_old_addr = ftrace_get_addr_curr(rec);
2365 
2366 	ret = ftrace_update_record(rec, enable);
2367 
2368 	ftrace_bug_type = FTRACE_BUG_UNKNOWN;
2369 
2370 	switch (ret) {
2371 	case FTRACE_UPDATE_IGNORE:
2372 		return 0;
2373 
2374 	case FTRACE_UPDATE_MAKE_CALL:
2375 		ftrace_bug_type = FTRACE_BUG_CALL;
2376 		return ftrace_make_call(rec, ftrace_addr);
2377 
2378 	case FTRACE_UPDATE_MAKE_NOP:
2379 		ftrace_bug_type = FTRACE_BUG_NOP;
2380 		return ftrace_make_nop(NULL, rec, ftrace_old_addr);
2381 
2382 	case FTRACE_UPDATE_MODIFY_CALL:
2383 		ftrace_bug_type = FTRACE_BUG_UPDATE;
2384 		return ftrace_modify_call(rec, ftrace_old_addr, ftrace_addr);
2385 	}
2386 
2387 	return -1; /* unknow ftrace bug */
2388 }
2389 
2390 void __weak ftrace_replace_code(int enable)
2391 {
2392 	struct dyn_ftrace *rec;
2393 	struct ftrace_page *pg;
2394 	int failed;
2395 
2396 	if (unlikely(ftrace_disabled))
2397 		return;
2398 
2399 	do_for_each_ftrace_rec(pg, rec) {
2400 		failed = __ftrace_replace_code(rec, enable);
2401 		if (failed) {
2402 			ftrace_bug(failed, rec);
2403 			/* Stop processing */
2404 			return;
2405 		}
2406 	} while_for_each_ftrace_rec();
2407 }
2408 
2409 struct ftrace_rec_iter {
2410 	struct ftrace_page	*pg;
2411 	int			index;
2412 };
2413 
2414 /**
2415  * ftrace_rec_iter_start, start up iterating over traced functions
2416  *
2417  * Returns an iterator handle that is used to iterate over all
2418  * the records that represent address locations where functions
2419  * are traced.
2420  *
2421  * May return NULL if no records are available.
2422  */
2423 struct ftrace_rec_iter *ftrace_rec_iter_start(void)
2424 {
2425 	/*
2426 	 * We only use a single iterator.
2427 	 * Protected by the ftrace_lock mutex.
2428 	 */
2429 	static struct ftrace_rec_iter ftrace_rec_iter;
2430 	struct ftrace_rec_iter *iter = &ftrace_rec_iter;
2431 
2432 	iter->pg = ftrace_pages_start;
2433 	iter->index = 0;
2434 
2435 	/* Could have empty pages */
2436 	while (iter->pg && !iter->pg->index)
2437 		iter->pg = iter->pg->next;
2438 
2439 	if (!iter->pg)
2440 		return NULL;
2441 
2442 	return iter;
2443 }
2444 
2445 /**
2446  * ftrace_rec_iter_next, get the next record to process.
2447  * @iter: The handle to the iterator.
2448  *
2449  * Returns the next iterator after the given iterator @iter.
2450  */
2451 struct ftrace_rec_iter *ftrace_rec_iter_next(struct ftrace_rec_iter *iter)
2452 {
2453 	iter->index++;
2454 
2455 	if (iter->index >= iter->pg->index) {
2456 		iter->pg = iter->pg->next;
2457 		iter->index = 0;
2458 
2459 		/* Could have empty pages */
2460 		while (iter->pg && !iter->pg->index)
2461 			iter->pg = iter->pg->next;
2462 	}
2463 
2464 	if (!iter->pg)
2465 		return NULL;
2466 
2467 	return iter;
2468 }
2469 
2470 /**
2471  * ftrace_rec_iter_record, get the record at the iterator location
2472  * @iter: The current iterator location
2473  *
2474  * Returns the record that the current @iter is at.
2475  */
2476 struct dyn_ftrace *ftrace_rec_iter_record(struct ftrace_rec_iter *iter)
2477 {
2478 	return &iter->pg->records[iter->index];
2479 }
2480 
2481 static int
2482 ftrace_code_disable(struct module *mod, struct dyn_ftrace *rec)
2483 {
2484 	int ret;
2485 
2486 	if (unlikely(ftrace_disabled))
2487 		return 0;
2488 
2489 	ret = ftrace_make_nop(mod, rec, MCOUNT_ADDR);
2490 	if (ret) {
2491 		ftrace_bug_type = FTRACE_BUG_INIT;
2492 		ftrace_bug(ret, rec);
2493 		return 0;
2494 	}
2495 	return 1;
2496 }
2497 
2498 /*
2499  * archs can override this function if they must do something
2500  * before the modifying code is performed.
2501  */
2502 int __weak ftrace_arch_code_modify_prepare(void)
2503 {
2504 	return 0;
2505 }
2506 
2507 /*
2508  * archs can override this function if they must do something
2509  * after the modifying code is performed.
2510  */
2511 int __weak ftrace_arch_code_modify_post_process(void)
2512 {
2513 	return 0;
2514 }
2515 
2516 void ftrace_modify_all_code(int command)
2517 {
2518 	int update = command & FTRACE_UPDATE_TRACE_FUNC;
2519 	int err = 0;
2520 
2521 	/*
2522 	 * If the ftrace_caller calls a ftrace_ops func directly,
2523 	 * we need to make sure that it only traces functions it
2524 	 * expects to trace. When doing the switch of functions,
2525 	 * we need to update to the ftrace_ops_list_func first
2526 	 * before the transition between old and new calls are set,
2527 	 * as the ftrace_ops_list_func will check the ops hashes
2528 	 * to make sure the ops are having the right functions
2529 	 * traced.
2530 	 */
2531 	if (update) {
2532 		err = ftrace_update_ftrace_func(ftrace_ops_list_func);
2533 		if (FTRACE_WARN_ON(err))
2534 			return;
2535 	}
2536 
2537 	if (command & FTRACE_UPDATE_CALLS)
2538 		ftrace_replace_code(1);
2539 	else if (command & FTRACE_DISABLE_CALLS)
2540 		ftrace_replace_code(0);
2541 
2542 	if (update && ftrace_trace_function != ftrace_ops_list_func) {
2543 		function_trace_op = set_function_trace_op;
2544 		smp_wmb();
2545 		/* If irqs are disabled, we are in stop machine */
2546 		if (!irqs_disabled())
2547 			smp_call_function(ftrace_sync_ipi, NULL, 1);
2548 		err = ftrace_update_ftrace_func(ftrace_trace_function);
2549 		if (FTRACE_WARN_ON(err))
2550 			return;
2551 	}
2552 
2553 	if (command & FTRACE_START_FUNC_RET)
2554 		err = ftrace_enable_ftrace_graph_caller();
2555 	else if (command & FTRACE_STOP_FUNC_RET)
2556 		err = ftrace_disable_ftrace_graph_caller();
2557 	FTRACE_WARN_ON(err);
2558 }
2559 
2560 static int __ftrace_modify_code(void *data)
2561 {
2562 	int *command = data;
2563 
2564 	ftrace_modify_all_code(*command);
2565 
2566 	return 0;
2567 }
2568 
2569 /**
2570  * ftrace_run_stop_machine, go back to the stop machine method
2571  * @command: The command to tell ftrace what to do
2572  *
2573  * If an arch needs to fall back to the stop machine method, the
2574  * it can call this function.
2575  */
2576 void ftrace_run_stop_machine(int command)
2577 {
2578 	stop_machine(__ftrace_modify_code, &command, NULL);
2579 }
2580 
2581 /**
2582  * arch_ftrace_update_code, modify the code to trace or not trace
2583  * @command: The command that needs to be done
2584  *
2585  * Archs can override this function if it does not need to
2586  * run stop_machine() to modify code.
2587  */
2588 void __weak arch_ftrace_update_code(int command)
2589 {
2590 	ftrace_run_stop_machine(command);
2591 }
2592 
2593 static void ftrace_run_update_code(int command)
2594 {
2595 	int ret;
2596 
2597 	ret = ftrace_arch_code_modify_prepare();
2598 	FTRACE_WARN_ON(ret);
2599 	if (ret)
2600 		return;
2601 
2602 	/*
2603 	 * By default we use stop_machine() to modify the code.
2604 	 * But archs can do what ever they want as long as it
2605 	 * is safe. The stop_machine() is the safest, but also
2606 	 * produces the most overhead.
2607 	 */
2608 	arch_ftrace_update_code(command);
2609 
2610 	ret = ftrace_arch_code_modify_post_process();
2611 	FTRACE_WARN_ON(ret);
2612 }
2613 
2614 static void ftrace_run_modify_code(struct ftrace_ops *ops, int command,
2615 				   struct ftrace_ops_hash *old_hash)
2616 {
2617 	ops->flags |= FTRACE_OPS_FL_MODIFYING;
2618 	ops->old_hash.filter_hash = old_hash->filter_hash;
2619 	ops->old_hash.notrace_hash = old_hash->notrace_hash;
2620 	ftrace_run_update_code(command);
2621 	ops->old_hash.filter_hash = NULL;
2622 	ops->old_hash.notrace_hash = NULL;
2623 	ops->flags &= ~FTRACE_OPS_FL_MODIFYING;
2624 }
2625 
2626 static ftrace_func_t saved_ftrace_func;
2627 static int ftrace_start_up;
2628 
2629 void __weak arch_ftrace_trampoline_free(struct ftrace_ops *ops)
2630 {
2631 }
2632 
2633 static void per_cpu_ops_free(struct ftrace_ops *ops)
2634 {
2635 	free_percpu(ops->disabled);
2636 }
2637 
2638 static void ftrace_startup_enable(int command)
2639 {
2640 	if (saved_ftrace_func != ftrace_trace_function) {
2641 		saved_ftrace_func = ftrace_trace_function;
2642 		command |= FTRACE_UPDATE_TRACE_FUNC;
2643 	}
2644 
2645 	if (!command || !ftrace_enabled)
2646 		return;
2647 
2648 	ftrace_run_update_code(command);
2649 }
2650 
2651 static void ftrace_startup_all(int command)
2652 {
2653 	update_all_ops = true;
2654 	ftrace_startup_enable(command);
2655 	update_all_ops = false;
2656 }
2657 
2658 static int ftrace_startup(struct ftrace_ops *ops, int command)
2659 {
2660 	int ret;
2661 
2662 	if (unlikely(ftrace_disabled))
2663 		return -ENODEV;
2664 
2665 	ret = __register_ftrace_function(ops);
2666 	if (ret)
2667 		return ret;
2668 
2669 	ftrace_start_up++;
2670 
2671 	/*
2672 	 * Note that ftrace probes uses this to start up
2673 	 * and modify functions it will probe. But we still
2674 	 * set the ADDING flag for modification, as probes
2675 	 * do not have trampolines. If they add them in the
2676 	 * future, then the probes will need to distinguish
2677 	 * between adding and updating probes.
2678 	 */
2679 	ops->flags |= FTRACE_OPS_FL_ENABLED | FTRACE_OPS_FL_ADDING;
2680 
2681 	ret = ftrace_hash_ipmodify_enable(ops);
2682 	if (ret < 0) {
2683 		/* Rollback registration process */
2684 		__unregister_ftrace_function(ops);
2685 		ftrace_start_up--;
2686 		ops->flags &= ~FTRACE_OPS_FL_ENABLED;
2687 		return ret;
2688 	}
2689 
2690 	if (ftrace_hash_rec_enable(ops, 1))
2691 		command |= FTRACE_UPDATE_CALLS;
2692 
2693 	ftrace_startup_enable(command);
2694 
2695 	ops->flags &= ~FTRACE_OPS_FL_ADDING;
2696 
2697 	return 0;
2698 }
2699 
2700 static int ftrace_shutdown(struct ftrace_ops *ops, int command)
2701 {
2702 	int ret;
2703 
2704 	if (unlikely(ftrace_disabled))
2705 		return -ENODEV;
2706 
2707 	ret = __unregister_ftrace_function(ops);
2708 	if (ret)
2709 		return ret;
2710 
2711 	ftrace_start_up--;
2712 	/*
2713 	 * Just warn in case of unbalance, no need to kill ftrace, it's not
2714 	 * critical but the ftrace_call callers may be never nopped again after
2715 	 * further ftrace uses.
2716 	 */
2717 	WARN_ON_ONCE(ftrace_start_up < 0);
2718 
2719 	/* Disabling ipmodify never fails */
2720 	ftrace_hash_ipmodify_disable(ops);
2721 
2722 	if (ftrace_hash_rec_disable(ops, 1))
2723 		command |= FTRACE_UPDATE_CALLS;
2724 
2725 	ops->flags &= ~FTRACE_OPS_FL_ENABLED;
2726 
2727 	if (saved_ftrace_func != ftrace_trace_function) {
2728 		saved_ftrace_func = ftrace_trace_function;
2729 		command |= FTRACE_UPDATE_TRACE_FUNC;
2730 	}
2731 
2732 	if (!command || !ftrace_enabled) {
2733 		/*
2734 		 * If these are per_cpu ops, they still need their
2735 		 * per_cpu field freed. Since, function tracing is
2736 		 * not currently active, we can just free them
2737 		 * without synchronizing all CPUs.
2738 		 */
2739 		if (ops->flags & FTRACE_OPS_FL_PER_CPU)
2740 			per_cpu_ops_free(ops);
2741 		return 0;
2742 	}
2743 
2744 	/*
2745 	 * If the ops uses a trampoline, then it needs to be
2746 	 * tested first on update.
2747 	 */
2748 	ops->flags |= FTRACE_OPS_FL_REMOVING;
2749 	removed_ops = ops;
2750 
2751 	/* The trampoline logic checks the old hashes */
2752 	ops->old_hash.filter_hash = ops->func_hash->filter_hash;
2753 	ops->old_hash.notrace_hash = ops->func_hash->notrace_hash;
2754 
2755 	ftrace_run_update_code(command);
2756 
2757 	/*
2758 	 * If there's no more ops registered with ftrace, run a
2759 	 * sanity check to make sure all rec flags are cleared.
2760 	 */
2761 	if (ftrace_ops_list == &ftrace_list_end) {
2762 		struct ftrace_page *pg;
2763 		struct dyn_ftrace *rec;
2764 
2765 		do_for_each_ftrace_rec(pg, rec) {
2766 			if (FTRACE_WARN_ON_ONCE(rec->flags))
2767 				pr_warn("  %pS flags:%lx\n",
2768 					(void *)rec->ip, rec->flags);
2769 		} while_for_each_ftrace_rec();
2770 	}
2771 
2772 	ops->old_hash.filter_hash = NULL;
2773 	ops->old_hash.notrace_hash = NULL;
2774 
2775 	removed_ops = NULL;
2776 	ops->flags &= ~FTRACE_OPS_FL_REMOVING;
2777 
2778 	/*
2779 	 * Dynamic ops may be freed, we must make sure that all
2780 	 * callers are done before leaving this function.
2781 	 * The same goes for freeing the per_cpu data of the per_cpu
2782 	 * ops.
2783 	 *
2784 	 * Again, normal synchronize_sched() is not good enough.
2785 	 * We need to do a hard force of sched synchronization.
2786 	 * This is because we use preempt_disable() to do RCU, but
2787 	 * the function tracers can be called where RCU is not watching
2788 	 * (like before user_exit()). We can not rely on the RCU
2789 	 * infrastructure to do the synchronization, thus we must do it
2790 	 * ourselves.
2791 	 */
2792 	if (ops->flags & (FTRACE_OPS_FL_DYNAMIC | FTRACE_OPS_FL_PER_CPU)) {
2793 		schedule_on_each_cpu(ftrace_sync);
2794 
2795 		arch_ftrace_trampoline_free(ops);
2796 
2797 		if (ops->flags & FTRACE_OPS_FL_PER_CPU)
2798 			per_cpu_ops_free(ops);
2799 	}
2800 
2801 	return 0;
2802 }
2803 
2804 static void ftrace_startup_sysctl(void)
2805 {
2806 	int command;
2807 
2808 	if (unlikely(ftrace_disabled))
2809 		return;
2810 
2811 	/* Force update next time */
2812 	saved_ftrace_func = NULL;
2813 	/* ftrace_start_up is true if we want ftrace running */
2814 	if (ftrace_start_up) {
2815 		command = FTRACE_UPDATE_CALLS;
2816 		if (ftrace_graph_active)
2817 			command |= FTRACE_START_FUNC_RET;
2818 		ftrace_startup_enable(command);
2819 	}
2820 }
2821 
2822 static void ftrace_shutdown_sysctl(void)
2823 {
2824 	int command;
2825 
2826 	if (unlikely(ftrace_disabled))
2827 		return;
2828 
2829 	/* ftrace_start_up is true if ftrace is running */
2830 	if (ftrace_start_up) {
2831 		command = FTRACE_DISABLE_CALLS;
2832 		if (ftrace_graph_active)
2833 			command |= FTRACE_STOP_FUNC_RET;
2834 		ftrace_run_update_code(command);
2835 	}
2836 }
2837 
2838 static cycle_t		ftrace_update_time;
2839 unsigned long		ftrace_update_tot_cnt;
2840 
2841 static inline int ops_traces_mod(struct ftrace_ops *ops)
2842 {
2843 	/*
2844 	 * Filter_hash being empty will default to trace module.
2845 	 * But notrace hash requires a test of individual module functions.
2846 	 */
2847 	return ftrace_hash_empty(ops->func_hash->filter_hash) &&
2848 		ftrace_hash_empty(ops->func_hash->notrace_hash);
2849 }
2850 
2851 /*
2852  * Check if the current ops references the record.
2853  *
2854  * If the ops traces all functions, then it was already accounted for.
2855  * If the ops does not trace the current record function, skip it.
2856  * If the ops ignores the function via notrace filter, skip it.
2857  */
2858 static inline bool
2859 ops_references_rec(struct ftrace_ops *ops, struct dyn_ftrace *rec)
2860 {
2861 	/* If ops isn't enabled, ignore it */
2862 	if (!(ops->flags & FTRACE_OPS_FL_ENABLED))
2863 		return 0;
2864 
2865 	/* If ops traces all then it includes this function */
2866 	if (ops_traces_mod(ops))
2867 		return 1;
2868 
2869 	/* The function must be in the filter */
2870 	if (!ftrace_hash_empty(ops->func_hash->filter_hash) &&
2871 	    !ftrace_lookup_ip(ops->func_hash->filter_hash, rec->ip))
2872 		return 0;
2873 
2874 	/* If in notrace hash, we ignore it too */
2875 	if (ftrace_lookup_ip(ops->func_hash->notrace_hash, rec->ip))
2876 		return 0;
2877 
2878 	return 1;
2879 }
2880 
2881 static int ftrace_update_code(struct module *mod, struct ftrace_page *new_pgs)
2882 {
2883 	struct ftrace_page *pg;
2884 	struct dyn_ftrace *p;
2885 	cycle_t start, stop;
2886 	unsigned long update_cnt = 0;
2887 	unsigned long rec_flags = 0;
2888 	int i;
2889 
2890 	start = ftrace_now(raw_smp_processor_id());
2891 
2892 	/*
2893 	 * When a module is loaded, this function is called to convert
2894 	 * the calls to mcount in its text to nops, and also to create
2895 	 * an entry in the ftrace data. Now, if ftrace is activated
2896 	 * after this call, but before the module sets its text to
2897 	 * read-only, the modification of enabling ftrace can fail if
2898 	 * the read-only is done while ftrace is converting the calls.
2899 	 * To prevent this, the module's records are set as disabled
2900 	 * and will be enabled after the call to set the module's text
2901 	 * to read-only.
2902 	 */
2903 	if (mod)
2904 		rec_flags |= FTRACE_FL_DISABLED;
2905 
2906 	for (pg = new_pgs; pg; pg = pg->next) {
2907 
2908 		for (i = 0; i < pg->index; i++) {
2909 
2910 			/* If something went wrong, bail without enabling anything */
2911 			if (unlikely(ftrace_disabled))
2912 				return -1;
2913 
2914 			p = &pg->records[i];
2915 			p->flags = rec_flags;
2916 
2917 			/*
2918 			 * Do the initial record conversion from mcount jump
2919 			 * to the NOP instructions.
2920 			 */
2921 			if (!ftrace_code_disable(mod, p))
2922 				break;
2923 
2924 			update_cnt++;
2925 		}
2926 	}
2927 
2928 	stop = ftrace_now(raw_smp_processor_id());
2929 	ftrace_update_time = stop - start;
2930 	ftrace_update_tot_cnt += update_cnt;
2931 
2932 	return 0;
2933 }
2934 
2935 static int ftrace_allocate_records(struct ftrace_page *pg, int count)
2936 {
2937 	int order;
2938 	int cnt;
2939 
2940 	if (WARN_ON(!count))
2941 		return -EINVAL;
2942 
2943 	order = get_count_order(DIV_ROUND_UP(count, ENTRIES_PER_PAGE));
2944 
2945 	/*
2946 	 * We want to fill as much as possible. No more than a page
2947 	 * may be empty.
2948 	 */
2949 	while ((PAGE_SIZE << order) / ENTRY_SIZE >= count + ENTRIES_PER_PAGE)
2950 		order--;
2951 
2952  again:
2953 	pg->records = (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, order);
2954 
2955 	if (!pg->records) {
2956 		/* if we can't allocate this size, try something smaller */
2957 		if (!order)
2958 			return -ENOMEM;
2959 		order >>= 1;
2960 		goto again;
2961 	}
2962 
2963 	cnt = (PAGE_SIZE << order) / ENTRY_SIZE;
2964 	pg->size = cnt;
2965 
2966 	if (cnt > count)
2967 		cnt = count;
2968 
2969 	return cnt;
2970 }
2971 
2972 static struct ftrace_page *
2973 ftrace_allocate_pages(unsigned long num_to_init)
2974 {
2975 	struct ftrace_page *start_pg;
2976 	struct ftrace_page *pg;
2977 	int order;
2978 	int cnt;
2979 
2980 	if (!num_to_init)
2981 		return 0;
2982 
2983 	start_pg = pg = kzalloc(sizeof(*pg), GFP_KERNEL);
2984 	if (!pg)
2985 		return NULL;
2986 
2987 	/*
2988 	 * Try to allocate as much as possible in one continues
2989 	 * location that fills in all of the space. We want to
2990 	 * waste as little space as possible.
2991 	 */
2992 	for (;;) {
2993 		cnt = ftrace_allocate_records(pg, num_to_init);
2994 		if (cnt < 0)
2995 			goto free_pages;
2996 
2997 		num_to_init -= cnt;
2998 		if (!num_to_init)
2999 			break;
3000 
3001 		pg->next = kzalloc(sizeof(*pg), GFP_KERNEL);
3002 		if (!pg->next)
3003 			goto free_pages;
3004 
3005 		pg = pg->next;
3006 	}
3007 
3008 	return start_pg;
3009 
3010  free_pages:
3011 	pg = start_pg;
3012 	while (pg) {
3013 		order = get_count_order(pg->size / ENTRIES_PER_PAGE);
3014 		free_pages((unsigned long)pg->records, order);
3015 		start_pg = pg->next;
3016 		kfree(pg);
3017 		pg = start_pg;
3018 	}
3019 	pr_info("ftrace: FAILED to allocate memory for functions\n");
3020 	return NULL;
3021 }
3022 
3023 #define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
3024 
3025 struct ftrace_iterator {
3026 	loff_t				pos;
3027 	loff_t				func_pos;
3028 	struct ftrace_page		*pg;
3029 	struct dyn_ftrace		*func;
3030 	struct ftrace_func_probe	*probe;
3031 	struct trace_parser		parser;
3032 	struct ftrace_hash		*hash;
3033 	struct ftrace_ops		*ops;
3034 	int				hidx;
3035 	int				idx;
3036 	unsigned			flags;
3037 };
3038 
3039 static void *
3040 t_hash_next(struct seq_file *m, loff_t *pos)
3041 {
3042 	struct ftrace_iterator *iter = m->private;
3043 	struct hlist_node *hnd = NULL;
3044 	struct hlist_head *hhd;
3045 
3046 	(*pos)++;
3047 	iter->pos = *pos;
3048 
3049 	if (iter->probe)
3050 		hnd = &iter->probe->node;
3051  retry:
3052 	if (iter->hidx >= FTRACE_FUNC_HASHSIZE)
3053 		return NULL;
3054 
3055 	hhd = &ftrace_func_hash[iter->hidx];
3056 
3057 	if (hlist_empty(hhd)) {
3058 		iter->hidx++;
3059 		hnd = NULL;
3060 		goto retry;
3061 	}
3062 
3063 	if (!hnd)
3064 		hnd = hhd->first;
3065 	else {
3066 		hnd = hnd->next;
3067 		if (!hnd) {
3068 			iter->hidx++;
3069 			goto retry;
3070 		}
3071 	}
3072 
3073 	if (WARN_ON_ONCE(!hnd))
3074 		return NULL;
3075 
3076 	iter->probe = hlist_entry(hnd, struct ftrace_func_probe, node);
3077 
3078 	return iter;
3079 }
3080 
3081 static void *t_hash_start(struct seq_file *m, loff_t *pos)
3082 {
3083 	struct ftrace_iterator *iter = m->private;
3084 	void *p = NULL;
3085 	loff_t l;
3086 
3087 	if (!(iter->flags & FTRACE_ITER_DO_HASH))
3088 		return NULL;
3089 
3090 	if (iter->func_pos > *pos)
3091 		return NULL;
3092 
3093 	iter->hidx = 0;
3094 	for (l = 0; l <= (*pos - iter->func_pos); ) {
3095 		p = t_hash_next(m, &l);
3096 		if (!p)
3097 			break;
3098 	}
3099 	if (!p)
3100 		return NULL;
3101 
3102 	/* Only set this if we have an item */
3103 	iter->flags |= FTRACE_ITER_HASH;
3104 
3105 	return iter;
3106 }
3107 
3108 static int
3109 t_hash_show(struct seq_file *m, struct ftrace_iterator *iter)
3110 {
3111 	struct ftrace_func_probe *rec;
3112 
3113 	rec = iter->probe;
3114 	if (WARN_ON_ONCE(!rec))
3115 		return -EIO;
3116 
3117 	if (rec->ops->print)
3118 		return rec->ops->print(m, rec->ip, rec->ops, rec->data);
3119 
3120 	seq_printf(m, "%ps:%ps", (void *)rec->ip, (void *)rec->ops->func);
3121 
3122 	if (rec->data)
3123 		seq_printf(m, ":%p", rec->data);
3124 	seq_putc(m, '\n');
3125 
3126 	return 0;
3127 }
3128 
3129 static void *
3130 t_next(struct seq_file *m, void *v, loff_t *pos)
3131 {
3132 	struct ftrace_iterator *iter = m->private;
3133 	struct ftrace_ops *ops = iter->ops;
3134 	struct dyn_ftrace *rec = NULL;
3135 
3136 	if (unlikely(ftrace_disabled))
3137 		return NULL;
3138 
3139 	if (iter->flags & FTRACE_ITER_HASH)
3140 		return t_hash_next(m, pos);
3141 
3142 	(*pos)++;
3143 	iter->pos = iter->func_pos = *pos;
3144 
3145 	if (iter->flags & FTRACE_ITER_PRINTALL)
3146 		return t_hash_start(m, pos);
3147 
3148  retry:
3149 	if (iter->idx >= iter->pg->index) {
3150 		if (iter->pg->next) {
3151 			iter->pg = iter->pg->next;
3152 			iter->idx = 0;
3153 			goto retry;
3154 		}
3155 	} else {
3156 		rec = &iter->pg->records[iter->idx++];
3157 		if (((iter->flags & FTRACE_ITER_FILTER) &&
3158 		     !(ftrace_lookup_ip(ops->func_hash->filter_hash, rec->ip))) ||
3159 
3160 		    ((iter->flags & FTRACE_ITER_NOTRACE) &&
3161 		     !ftrace_lookup_ip(ops->func_hash->notrace_hash, rec->ip)) ||
3162 
3163 		    ((iter->flags & FTRACE_ITER_ENABLED) &&
3164 		     !(rec->flags & FTRACE_FL_ENABLED))) {
3165 
3166 			rec = NULL;
3167 			goto retry;
3168 		}
3169 	}
3170 
3171 	if (!rec)
3172 		return t_hash_start(m, pos);
3173 
3174 	iter->func = rec;
3175 
3176 	return iter;
3177 }
3178 
3179 static void reset_iter_read(struct ftrace_iterator *iter)
3180 {
3181 	iter->pos = 0;
3182 	iter->func_pos = 0;
3183 	iter->flags &= ~(FTRACE_ITER_PRINTALL | FTRACE_ITER_HASH);
3184 }
3185 
3186 static void *t_start(struct seq_file *m, loff_t *pos)
3187 {
3188 	struct ftrace_iterator *iter = m->private;
3189 	struct ftrace_ops *ops = iter->ops;
3190 	void *p = NULL;
3191 	loff_t l;
3192 
3193 	mutex_lock(&ftrace_lock);
3194 
3195 	if (unlikely(ftrace_disabled))
3196 		return NULL;
3197 
3198 	/*
3199 	 * If an lseek was done, then reset and start from beginning.
3200 	 */
3201 	if (*pos < iter->pos)
3202 		reset_iter_read(iter);
3203 
3204 	/*
3205 	 * For set_ftrace_filter reading, if we have the filter
3206 	 * off, we can short cut and just print out that all
3207 	 * functions are enabled.
3208 	 */
3209 	if ((iter->flags & FTRACE_ITER_FILTER &&
3210 	     ftrace_hash_empty(ops->func_hash->filter_hash)) ||
3211 	    (iter->flags & FTRACE_ITER_NOTRACE &&
3212 	     ftrace_hash_empty(ops->func_hash->notrace_hash))) {
3213 		if (*pos > 0)
3214 			return t_hash_start(m, pos);
3215 		iter->flags |= FTRACE_ITER_PRINTALL;
3216 		/* reset in case of seek/pread */
3217 		iter->flags &= ~FTRACE_ITER_HASH;
3218 		return iter;
3219 	}
3220 
3221 	if (iter->flags & FTRACE_ITER_HASH)
3222 		return t_hash_start(m, pos);
3223 
3224 	/*
3225 	 * Unfortunately, we need to restart at ftrace_pages_start
3226 	 * every time we let go of the ftrace_mutex. This is because
3227 	 * those pointers can change without the lock.
3228 	 */
3229 	iter->pg = ftrace_pages_start;
3230 	iter->idx = 0;
3231 	for (l = 0; l <= *pos; ) {
3232 		p = t_next(m, p, &l);
3233 		if (!p)
3234 			break;
3235 	}
3236 
3237 	if (!p)
3238 		return t_hash_start(m, pos);
3239 
3240 	return iter;
3241 }
3242 
3243 static void t_stop(struct seq_file *m, void *p)
3244 {
3245 	mutex_unlock(&ftrace_lock);
3246 }
3247 
3248 void * __weak
3249 arch_ftrace_trampoline_func(struct ftrace_ops *ops, struct dyn_ftrace *rec)
3250 {
3251 	return NULL;
3252 }
3253 
3254 static void add_trampoline_func(struct seq_file *m, struct ftrace_ops *ops,
3255 				struct dyn_ftrace *rec)
3256 {
3257 	void *ptr;
3258 
3259 	ptr = arch_ftrace_trampoline_func(ops, rec);
3260 	if (ptr)
3261 		seq_printf(m, " ->%pS", ptr);
3262 }
3263 
3264 static int t_show(struct seq_file *m, void *v)
3265 {
3266 	struct ftrace_iterator *iter = m->private;
3267 	struct dyn_ftrace *rec;
3268 
3269 	if (iter->flags & FTRACE_ITER_HASH)
3270 		return t_hash_show(m, iter);
3271 
3272 	if (iter->flags & FTRACE_ITER_PRINTALL) {
3273 		if (iter->flags & FTRACE_ITER_NOTRACE)
3274 			seq_puts(m, "#### no functions disabled ####\n");
3275 		else
3276 			seq_puts(m, "#### all functions enabled ####\n");
3277 		return 0;
3278 	}
3279 
3280 	rec = iter->func;
3281 
3282 	if (!rec)
3283 		return 0;
3284 
3285 	seq_printf(m, "%ps", (void *)rec->ip);
3286 	if (iter->flags & FTRACE_ITER_ENABLED) {
3287 		struct ftrace_ops *ops;
3288 
3289 		seq_printf(m, " (%ld)%s%s",
3290 			   ftrace_rec_count(rec),
3291 			   rec->flags & FTRACE_FL_REGS ? " R" : "  ",
3292 			   rec->flags & FTRACE_FL_IPMODIFY ? " I" : "  ");
3293 		if (rec->flags & FTRACE_FL_TRAMP_EN) {
3294 			ops = ftrace_find_tramp_ops_any(rec);
3295 			if (ops) {
3296 				do {
3297 					seq_printf(m, "\ttramp: %pS (%pS)",
3298 						   (void *)ops->trampoline,
3299 						   (void *)ops->func);
3300 					add_trampoline_func(m, ops, rec);
3301 					ops = ftrace_find_tramp_ops_next(rec, ops);
3302 				} while (ops);
3303 			} else
3304 				seq_puts(m, "\ttramp: ERROR!");
3305 		} else {
3306 			add_trampoline_func(m, NULL, rec);
3307 		}
3308 	}
3309 
3310 	seq_putc(m, '\n');
3311 
3312 	return 0;
3313 }
3314 
3315 static const struct seq_operations show_ftrace_seq_ops = {
3316 	.start = t_start,
3317 	.next = t_next,
3318 	.stop = t_stop,
3319 	.show = t_show,
3320 };
3321 
3322 static int
3323 ftrace_avail_open(struct inode *inode, struct file *file)
3324 {
3325 	struct ftrace_iterator *iter;
3326 
3327 	if (unlikely(ftrace_disabled))
3328 		return -ENODEV;
3329 
3330 	iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
3331 	if (iter) {
3332 		iter->pg = ftrace_pages_start;
3333 		iter->ops = &global_ops;
3334 	}
3335 
3336 	return iter ? 0 : -ENOMEM;
3337 }
3338 
3339 static int
3340 ftrace_enabled_open(struct inode *inode, struct file *file)
3341 {
3342 	struct ftrace_iterator *iter;
3343 
3344 	iter = __seq_open_private(file, &show_ftrace_seq_ops, sizeof(*iter));
3345 	if (iter) {
3346 		iter->pg = ftrace_pages_start;
3347 		iter->flags = FTRACE_ITER_ENABLED;
3348 		iter->ops = &global_ops;
3349 	}
3350 
3351 	return iter ? 0 : -ENOMEM;
3352 }
3353 
3354 /**
3355  * ftrace_regex_open - initialize function tracer filter files
3356  * @ops: The ftrace_ops that hold the hash filters
3357  * @flag: The type of filter to process
3358  * @inode: The inode, usually passed in to your open routine
3359  * @file: The file, usually passed in to your open routine
3360  *
3361  * ftrace_regex_open() initializes the filter files for the
3362  * @ops. Depending on @flag it may process the filter hash or
3363  * the notrace hash of @ops. With this called from the open
3364  * routine, you can use ftrace_filter_write() for the write
3365  * routine if @flag has FTRACE_ITER_FILTER set, or
3366  * ftrace_notrace_write() if @flag has FTRACE_ITER_NOTRACE set.
3367  * tracing_lseek() should be used as the lseek routine, and
3368  * release must call ftrace_regex_release().
3369  */
3370 int
3371 ftrace_regex_open(struct ftrace_ops *ops, int flag,
3372 		  struct inode *inode, struct file *file)
3373 {
3374 	struct ftrace_iterator *iter;
3375 	struct ftrace_hash *hash;
3376 	int ret = 0;
3377 
3378 	ftrace_ops_init(ops);
3379 
3380 	if (unlikely(ftrace_disabled))
3381 		return -ENODEV;
3382 
3383 	iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3384 	if (!iter)
3385 		return -ENOMEM;
3386 
3387 	if (trace_parser_get_init(&iter->parser, FTRACE_BUFF_MAX)) {
3388 		kfree(iter);
3389 		return -ENOMEM;
3390 	}
3391 
3392 	iter->ops = ops;
3393 	iter->flags = flag;
3394 
3395 	mutex_lock(&ops->func_hash->regex_lock);
3396 
3397 	if (flag & FTRACE_ITER_NOTRACE)
3398 		hash = ops->func_hash->notrace_hash;
3399 	else
3400 		hash = ops->func_hash->filter_hash;
3401 
3402 	if (file->f_mode & FMODE_WRITE) {
3403 		const int size_bits = FTRACE_HASH_DEFAULT_BITS;
3404 
3405 		if (file->f_flags & O_TRUNC)
3406 			iter->hash = alloc_ftrace_hash(size_bits);
3407 		else
3408 			iter->hash = alloc_and_copy_ftrace_hash(size_bits, hash);
3409 
3410 		if (!iter->hash) {
3411 			trace_parser_put(&iter->parser);
3412 			kfree(iter);
3413 			ret = -ENOMEM;
3414 			goto out_unlock;
3415 		}
3416 	}
3417 
3418 	if (file->f_mode & FMODE_READ) {
3419 		iter->pg = ftrace_pages_start;
3420 
3421 		ret = seq_open(file, &show_ftrace_seq_ops);
3422 		if (!ret) {
3423 			struct seq_file *m = file->private_data;
3424 			m->private = iter;
3425 		} else {
3426 			/* Failed */
3427 			free_ftrace_hash(iter->hash);
3428 			trace_parser_put(&iter->parser);
3429 			kfree(iter);
3430 		}
3431 	} else
3432 		file->private_data = iter;
3433 
3434  out_unlock:
3435 	mutex_unlock(&ops->func_hash->regex_lock);
3436 
3437 	return ret;
3438 }
3439 
3440 static int
3441 ftrace_filter_open(struct inode *inode, struct file *file)
3442 {
3443 	struct ftrace_ops *ops = inode->i_private;
3444 
3445 	return ftrace_regex_open(ops,
3446 			FTRACE_ITER_FILTER | FTRACE_ITER_DO_HASH,
3447 			inode, file);
3448 }
3449 
3450 static int
3451 ftrace_notrace_open(struct inode *inode, struct file *file)
3452 {
3453 	struct ftrace_ops *ops = inode->i_private;
3454 
3455 	return ftrace_regex_open(ops, FTRACE_ITER_NOTRACE,
3456 				 inode, file);
3457 }
3458 
3459 /* Type for quick search ftrace basic regexes (globs) from filter_parse_regex */
3460 struct ftrace_glob {
3461 	char *search;
3462 	unsigned len;
3463 	int type;
3464 };
3465 
3466 /*
3467  * If symbols in an architecture don't correspond exactly to the user-visible
3468  * name of what they represent, it is possible to define this function to
3469  * perform the necessary adjustments.
3470 */
3471 char * __weak arch_ftrace_match_adjust(char *str, const char *search)
3472 {
3473 	return str;
3474 }
3475 
3476 static int ftrace_match(char *str, struct ftrace_glob *g)
3477 {
3478 	int matched = 0;
3479 	int slen;
3480 
3481 	str = arch_ftrace_match_adjust(str, g->search);
3482 
3483 	switch (g->type) {
3484 	case MATCH_FULL:
3485 		if (strcmp(str, g->search) == 0)
3486 			matched = 1;
3487 		break;
3488 	case MATCH_FRONT_ONLY:
3489 		if (strncmp(str, g->search, g->len) == 0)
3490 			matched = 1;
3491 		break;
3492 	case MATCH_MIDDLE_ONLY:
3493 		if (strstr(str, g->search))
3494 			matched = 1;
3495 		break;
3496 	case MATCH_END_ONLY:
3497 		slen = strlen(str);
3498 		if (slen >= g->len &&
3499 		    memcmp(str + slen - g->len, g->search, g->len) == 0)
3500 			matched = 1;
3501 		break;
3502 	}
3503 
3504 	return matched;
3505 }
3506 
3507 static int
3508 enter_record(struct ftrace_hash *hash, struct dyn_ftrace *rec, int clear_filter)
3509 {
3510 	struct ftrace_func_entry *entry;
3511 	int ret = 0;
3512 
3513 	entry = ftrace_lookup_ip(hash, rec->ip);
3514 	if (clear_filter) {
3515 		/* Do nothing if it doesn't exist */
3516 		if (!entry)
3517 			return 0;
3518 
3519 		free_hash_entry(hash, entry);
3520 	} else {
3521 		/* Do nothing if it exists */
3522 		if (entry)
3523 			return 0;
3524 
3525 		ret = add_hash_entry(hash, rec->ip);
3526 	}
3527 	return ret;
3528 }
3529 
3530 static int
3531 ftrace_match_record(struct dyn_ftrace *rec, struct ftrace_glob *func_g,
3532 		struct ftrace_glob *mod_g, int exclude_mod)
3533 {
3534 	char str[KSYM_SYMBOL_LEN];
3535 	char *modname;
3536 
3537 	kallsyms_lookup(rec->ip, NULL, NULL, &modname, str);
3538 
3539 	if (mod_g) {
3540 		int mod_matches = (modname) ? ftrace_match(modname, mod_g) : 0;
3541 
3542 		/* blank module name to match all modules */
3543 		if (!mod_g->len) {
3544 			/* blank module globbing: modname xor exclude_mod */
3545 			if ((!exclude_mod) != (!modname))
3546 				goto func_match;
3547 			return 0;
3548 		}
3549 
3550 		/* not matching the module */
3551 		if (!modname || !mod_matches) {
3552 			if (exclude_mod)
3553 				goto func_match;
3554 			else
3555 				return 0;
3556 		}
3557 
3558 		if (mod_matches && exclude_mod)
3559 			return 0;
3560 
3561 func_match:
3562 		/* blank search means to match all funcs in the mod */
3563 		if (!func_g->len)
3564 			return 1;
3565 	}
3566 
3567 	return ftrace_match(str, func_g);
3568 }
3569 
3570 static int
3571 match_records(struct ftrace_hash *hash, char *func, int len, char *mod)
3572 {
3573 	struct ftrace_page *pg;
3574 	struct dyn_ftrace *rec;
3575 	struct ftrace_glob func_g = { .type = MATCH_FULL };
3576 	struct ftrace_glob mod_g = { .type = MATCH_FULL };
3577 	struct ftrace_glob *mod_match = (mod) ? &mod_g : NULL;
3578 	int exclude_mod = 0;
3579 	int found = 0;
3580 	int ret;
3581 	int clear_filter;
3582 
3583 	if (func) {
3584 		func_g.type = filter_parse_regex(func, len, &func_g.search,
3585 						 &clear_filter);
3586 		func_g.len = strlen(func_g.search);
3587 	}
3588 
3589 	if (mod) {
3590 		mod_g.type = filter_parse_regex(mod, strlen(mod),
3591 				&mod_g.search, &exclude_mod);
3592 		mod_g.len = strlen(mod_g.search);
3593 	}
3594 
3595 	mutex_lock(&ftrace_lock);
3596 
3597 	if (unlikely(ftrace_disabled))
3598 		goto out_unlock;
3599 
3600 	do_for_each_ftrace_rec(pg, rec) {
3601 		if (ftrace_match_record(rec, &func_g, mod_match, exclude_mod)) {
3602 			ret = enter_record(hash, rec, clear_filter);
3603 			if (ret < 0) {
3604 				found = ret;
3605 				goto out_unlock;
3606 			}
3607 			found = 1;
3608 		}
3609 	} while_for_each_ftrace_rec();
3610  out_unlock:
3611 	mutex_unlock(&ftrace_lock);
3612 
3613 	return found;
3614 }
3615 
3616 static int
3617 ftrace_match_records(struct ftrace_hash *hash, char *buff, int len)
3618 {
3619 	return match_records(hash, buff, len, NULL);
3620 }
3621 
3622 
3623 /*
3624  * We register the module command as a template to show others how
3625  * to register the a command as well.
3626  */
3627 
3628 static int
3629 ftrace_mod_callback(struct ftrace_hash *hash,
3630 		    char *func, char *cmd, char *module, int enable)
3631 {
3632 	int ret;
3633 
3634 	/*
3635 	 * cmd == 'mod' because we only registered this func
3636 	 * for the 'mod' ftrace_func_command.
3637 	 * But if you register one func with multiple commands,
3638 	 * you can tell which command was used by the cmd
3639 	 * parameter.
3640 	 */
3641 	ret = match_records(hash, func, strlen(func), module);
3642 	if (!ret)
3643 		return -EINVAL;
3644 	if (ret < 0)
3645 		return ret;
3646 	return 0;
3647 }
3648 
3649 static struct ftrace_func_command ftrace_mod_cmd = {
3650 	.name			= "mod",
3651 	.func			= ftrace_mod_callback,
3652 };
3653 
3654 static int __init ftrace_mod_cmd_init(void)
3655 {
3656 	return register_ftrace_command(&ftrace_mod_cmd);
3657 }
3658 core_initcall(ftrace_mod_cmd_init);
3659 
3660 static void function_trace_probe_call(unsigned long ip, unsigned long parent_ip,
3661 				      struct ftrace_ops *op, struct pt_regs *pt_regs)
3662 {
3663 	struct ftrace_func_probe *entry;
3664 	struct hlist_head *hhd;
3665 	unsigned long key;
3666 
3667 	key = hash_long(ip, FTRACE_HASH_BITS);
3668 
3669 	hhd = &ftrace_func_hash[key];
3670 
3671 	if (hlist_empty(hhd))
3672 		return;
3673 
3674 	/*
3675 	 * Disable preemption for these calls to prevent a RCU grace
3676 	 * period. This syncs the hash iteration and freeing of items
3677 	 * on the hash. rcu_read_lock is too dangerous here.
3678 	 */
3679 	preempt_disable_notrace();
3680 	hlist_for_each_entry_rcu_notrace(entry, hhd, node) {
3681 		if (entry->ip == ip)
3682 			entry->ops->func(ip, parent_ip, &entry->data);
3683 	}
3684 	preempt_enable_notrace();
3685 }
3686 
3687 static struct ftrace_ops trace_probe_ops __read_mostly =
3688 {
3689 	.func		= function_trace_probe_call,
3690 	.flags		= FTRACE_OPS_FL_INITIALIZED,
3691 	INIT_OPS_HASH(trace_probe_ops)
3692 };
3693 
3694 static int ftrace_probe_registered;
3695 
3696 static void __enable_ftrace_function_probe(struct ftrace_ops_hash *old_hash)
3697 {
3698 	int ret;
3699 	int i;
3700 
3701 	if (ftrace_probe_registered) {
3702 		/* still need to update the function call sites */
3703 		if (ftrace_enabled)
3704 			ftrace_run_modify_code(&trace_probe_ops, FTRACE_UPDATE_CALLS,
3705 					       old_hash);
3706 		return;
3707 	}
3708 
3709 	for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
3710 		struct hlist_head *hhd = &ftrace_func_hash[i];
3711 		if (hhd->first)
3712 			break;
3713 	}
3714 	/* Nothing registered? */
3715 	if (i == FTRACE_FUNC_HASHSIZE)
3716 		return;
3717 
3718 	ret = ftrace_startup(&trace_probe_ops, 0);
3719 
3720 	ftrace_probe_registered = 1;
3721 }
3722 
3723 static void __disable_ftrace_function_probe(void)
3724 {
3725 	int i;
3726 
3727 	if (!ftrace_probe_registered)
3728 		return;
3729 
3730 	for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
3731 		struct hlist_head *hhd = &ftrace_func_hash[i];
3732 		if (hhd->first)
3733 			return;
3734 	}
3735 
3736 	/* no more funcs left */
3737 	ftrace_shutdown(&trace_probe_ops, 0);
3738 
3739 	ftrace_probe_registered = 0;
3740 }
3741 
3742 
3743 static void ftrace_free_entry(struct ftrace_func_probe *entry)
3744 {
3745 	if (entry->ops->free)
3746 		entry->ops->free(entry->ops, entry->ip, &entry->data);
3747 	kfree(entry);
3748 }
3749 
3750 int
3751 register_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
3752 			      void *data)
3753 {
3754 	struct ftrace_ops_hash old_hash_ops;
3755 	struct ftrace_func_probe *entry;
3756 	struct ftrace_glob func_g;
3757 	struct ftrace_hash **orig_hash = &trace_probe_ops.func_hash->filter_hash;
3758 	struct ftrace_hash *old_hash = *orig_hash;
3759 	struct ftrace_hash *hash;
3760 	struct ftrace_page *pg;
3761 	struct dyn_ftrace *rec;
3762 	int not;
3763 	unsigned long key;
3764 	int count = 0;
3765 	int ret;
3766 
3767 	func_g.type = filter_parse_regex(glob, strlen(glob),
3768 			&func_g.search, &not);
3769 	func_g.len = strlen(func_g.search);
3770 
3771 	/* we do not support '!' for function probes */
3772 	if (WARN_ON(not))
3773 		return -EINVAL;
3774 
3775 	mutex_lock(&trace_probe_ops.func_hash->regex_lock);
3776 
3777 	old_hash_ops.filter_hash = old_hash;
3778 	/* Probes only have filters */
3779 	old_hash_ops.notrace_hash = NULL;
3780 
3781 	hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, old_hash);
3782 	if (!hash) {
3783 		count = -ENOMEM;
3784 		goto out;
3785 	}
3786 
3787 	if (unlikely(ftrace_disabled)) {
3788 		count = -ENODEV;
3789 		goto out;
3790 	}
3791 
3792 	mutex_lock(&ftrace_lock);
3793 
3794 	do_for_each_ftrace_rec(pg, rec) {
3795 
3796 		if (!ftrace_match_record(rec, &func_g, NULL, 0))
3797 			continue;
3798 
3799 		entry = kmalloc(sizeof(*entry), GFP_KERNEL);
3800 		if (!entry) {
3801 			/* If we did not process any, then return error */
3802 			if (!count)
3803 				count = -ENOMEM;
3804 			goto out_unlock;
3805 		}
3806 
3807 		count++;
3808 
3809 		entry->data = data;
3810 
3811 		/*
3812 		 * The caller might want to do something special
3813 		 * for each function we find. We call the callback
3814 		 * to give the caller an opportunity to do so.
3815 		 */
3816 		if (ops->init) {
3817 			if (ops->init(ops, rec->ip, &entry->data) < 0) {
3818 				/* caller does not like this func */
3819 				kfree(entry);
3820 				continue;
3821 			}
3822 		}
3823 
3824 		ret = enter_record(hash, rec, 0);
3825 		if (ret < 0) {
3826 			kfree(entry);
3827 			count = ret;
3828 			goto out_unlock;
3829 		}
3830 
3831 		entry->ops = ops;
3832 		entry->ip = rec->ip;
3833 
3834 		key = hash_long(entry->ip, FTRACE_HASH_BITS);
3835 		hlist_add_head_rcu(&entry->node, &ftrace_func_hash[key]);
3836 
3837 	} while_for_each_ftrace_rec();
3838 
3839 	ret = ftrace_hash_move(&trace_probe_ops, 1, orig_hash, hash);
3840 
3841 	__enable_ftrace_function_probe(&old_hash_ops);
3842 
3843 	if (!ret)
3844 		free_ftrace_hash_rcu(old_hash);
3845 	else
3846 		count = ret;
3847 
3848  out_unlock:
3849 	mutex_unlock(&ftrace_lock);
3850  out:
3851 	mutex_unlock(&trace_probe_ops.func_hash->regex_lock);
3852 	free_ftrace_hash(hash);
3853 
3854 	return count;
3855 }
3856 
3857 enum {
3858 	PROBE_TEST_FUNC		= 1,
3859 	PROBE_TEST_DATA		= 2
3860 };
3861 
3862 static void
3863 __unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
3864 				  void *data, int flags)
3865 {
3866 	struct ftrace_func_entry *rec_entry;
3867 	struct ftrace_func_probe *entry;
3868 	struct ftrace_func_probe *p;
3869 	struct ftrace_glob func_g;
3870 	struct ftrace_hash **orig_hash = &trace_probe_ops.func_hash->filter_hash;
3871 	struct ftrace_hash *old_hash = *orig_hash;
3872 	struct list_head free_list;
3873 	struct ftrace_hash *hash;
3874 	struct hlist_node *tmp;
3875 	char str[KSYM_SYMBOL_LEN];
3876 	int i, ret;
3877 
3878 	if (glob && (strcmp(glob, "*") == 0 || !strlen(glob)))
3879 		func_g.search = NULL;
3880 	else if (glob) {
3881 		int not;
3882 
3883 		func_g.type = filter_parse_regex(glob, strlen(glob),
3884 						 &func_g.search, &not);
3885 		func_g.len = strlen(func_g.search);
3886 		func_g.search = glob;
3887 
3888 		/* we do not support '!' for function probes */
3889 		if (WARN_ON(not))
3890 			return;
3891 	}
3892 
3893 	mutex_lock(&trace_probe_ops.func_hash->regex_lock);
3894 
3895 	hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
3896 	if (!hash)
3897 		/* Hmm, should report this somehow */
3898 		goto out_unlock;
3899 
3900 	INIT_LIST_HEAD(&free_list);
3901 
3902 	for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
3903 		struct hlist_head *hhd = &ftrace_func_hash[i];
3904 
3905 		hlist_for_each_entry_safe(entry, tmp, hhd, node) {
3906 
3907 			/* break up if statements for readability */
3908 			if ((flags & PROBE_TEST_FUNC) && entry->ops != ops)
3909 				continue;
3910 
3911 			if ((flags & PROBE_TEST_DATA) && entry->data != data)
3912 				continue;
3913 
3914 			/* do this last, since it is the most expensive */
3915 			if (func_g.search) {
3916 				kallsyms_lookup(entry->ip, NULL, NULL,
3917 						NULL, str);
3918 				if (!ftrace_match(str, &func_g))
3919 					continue;
3920 			}
3921 
3922 			rec_entry = ftrace_lookup_ip(hash, entry->ip);
3923 			/* It is possible more than one entry had this ip */
3924 			if (rec_entry)
3925 				free_hash_entry(hash, rec_entry);
3926 
3927 			hlist_del_rcu(&entry->node);
3928 			list_add(&entry->free_list, &free_list);
3929 		}
3930 	}
3931 	mutex_lock(&ftrace_lock);
3932 	__disable_ftrace_function_probe();
3933 	/*
3934 	 * Remove after the disable is called. Otherwise, if the last
3935 	 * probe is removed, a null hash means *all enabled*.
3936 	 */
3937 	ret = ftrace_hash_move(&trace_probe_ops, 1, orig_hash, hash);
3938 	synchronize_sched();
3939 	if (!ret)
3940 		free_ftrace_hash_rcu(old_hash);
3941 
3942 	list_for_each_entry_safe(entry, p, &free_list, free_list) {
3943 		list_del(&entry->free_list);
3944 		ftrace_free_entry(entry);
3945 	}
3946 	mutex_unlock(&ftrace_lock);
3947 
3948  out_unlock:
3949 	mutex_unlock(&trace_probe_ops.func_hash->regex_lock);
3950 	free_ftrace_hash(hash);
3951 }
3952 
3953 void
3954 unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
3955 				void *data)
3956 {
3957 	__unregister_ftrace_function_probe(glob, ops, data,
3958 					  PROBE_TEST_FUNC | PROBE_TEST_DATA);
3959 }
3960 
3961 void
3962 unregister_ftrace_function_probe_func(char *glob, struct ftrace_probe_ops *ops)
3963 {
3964 	__unregister_ftrace_function_probe(glob, ops, NULL, PROBE_TEST_FUNC);
3965 }
3966 
3967 void unregister_ftrace_function_probe_all(char *glob)
3968 {
3969 	__unregister_ftrace_function_probe(glob, NULL, NULL, 0);
3970 }
3971 
3972 static LIST_HEAD(ftrace_commands);
3973 static DEFINE_MUTEX(ftrace_cmd_mutex);
3974 
3975 /*
3976  * Currently we only register ftrace commands from __init, so mark this
3977  * __init too.
3978  */
3979 __init int register_ftrace_command(struct ftrace_func_command *cmd)
3980 {
3981 	struct ftrace_func_command *p;
3982 	int ret = 0;
3983 
3984 	mutex_lock(&ftrace_cmd_mutex);
3985 	list_for_each_entry(p, &ftrace_commands, list) {
3986 		if (strcmp(cmd->name, p->name) == 0) {
3987 			ret = -EBUSY;
3988 			goto out_unlock;
3989 		}
3990 	}
3991 	list_add(&cmd->list, &ftrace_commands);
3992  out_unlock:
3993 	mutex_unlock(&ftrace_cmd_mutex);
3994 
3995 	return ret;
3996 }
3997 
3998 /*
3999  * Currently we only unregister ftrace commands from __init, so mark
4000  * this __init too.
4001  */
4002 __init int unregister_ftrace_command(struct ftrace_func_command *cmd)
4003 {
4004 	struct ftrace_func_command *p, *n;
4005 	int ret = -ENODEV;
4006 
4007 	mutex_lock(&ftrace_cmd_mutex);
4008 	list_for_each_entry_safe(p, n, &ftrace_commands, list) {
4009 		if (strcmp(cmd->name, p->name) == 0) {
4010 			ret = 0;
4011 			list_del_init(&p->list);
4012 			goto out_unlock;
4013 		}
4014 	}
4015  out_unlock:
4016 	mutex_unlock(&ftrace_cmd_mutex);
4017 
4018 	return ret;
4019 }
4020 
4021 static int ftrace_process_regex(struct ftrace_hash *hash,
4022 				char *buff, int len, int enable)
4023 {
4024 	char *func, *command, *next = buff;
4025 	struct ftrace_func_command *p;
4026 	int ret = -EINVAL;
4027 
4028 	func = strsep(&next, ":");
4029 
4030 	if (!next) {
4031 		ret = ftrace_match_records(hash, func, len);
4032 		if (!ret)
4033 			ret = -EINVAL;
4034 		if (ret < 0)
4035 			return ret;
4036 		return 0;
4037 	}
4038 
4039 	/* command found */
4040 
4041 	command = strsep(&next, ":");
4042 
4043 	mutex_lock(&ftrace_cmd_mutex);
4044 	list_for_each_entry(p, &ftrace_commands, list) {
4045 		if (strcmp(p->name, command) == 0) {
4046 			ret = p->func(hash, func, command, next, enable);
4047 			goto out_unlock;
4048 		}
4049 	}
4050  out_unlock:
4051 	mutex_unlock(&ftrace_cmd_mutex);
4052 
4053 	return ret;
4054 }
4055 
4056 static ssize_t
4057 ftrace_regex_write(struct file *file, const char __user *ubuf,
4058 		   size_t cnt, loff_t *ppos, int enable)
4059 {
4060 	struct ftrace_iterator *iter;
4061 	struct trace_parser *parser;
4062 	ssize_t ret, read;
4063 
4064 	if (!cnt)
4065 		return 0;
4066 
4067 	if (file->f_mode & FMODE_READ) {
4068 		struct seq_file *m = file->private_data;
4069 		iter = m->private;
4070 	} else
4071 		iter = file->private_data;
4072 
4073 	if (unlikely(ftrace_disabled))
4074 		return -ENODEV;
4075 
4076 	/* iter->hash is a local copy, so we don't need regex_lock */
4077 
4078 	parser = &iter->parser;
4079 	read = trace_get_user(parser, ubuf, cnt, ppos);
4080 
4081 	if (read >= 0 && trace_parser_loaded(parser) &&
4082 	    !trace_parser_cont(parser)) {
4083 		ret = ftrace_process_regex(iter->hash, parser->buffer,
4084 					   parser->idx, enable);
4085 		trace_parser_clear(parser);
4086 		if (ret < 0)
4087 			goto out;
4088 	}
4089 
4090 	ret = read;
4091  out:
4092 	return ret;
4093 }
4094 
4095 ssize_t
4096 ftrace_filter_write(struct file *file, const char __user *ubuf,
4097 		    size_t cnt, loff_t *ppos)
4098 {
4099 	return ftrace_regex_write(file, ubuf, cnt, ppos, 1);
4100 }
4101 
4102 ssize_t
4103 ftrace_notrace_write(struct file *file, const char __user *ubuf,
4104 		     size_t cnt, loff_t *ppos)
4105 {
4106 	return ftrace_regex_write(file, ubuf, cnt, ppos, 0);
4107 }
4108 
4109 static int
4110 ftrace_match_addr(struct ftrace_hash *hash, unsigned long ip, int remove)
4111 {
4112 	struct ftrace_func_entry *entry;
4113 
4114 	if (!ftrace_location(ip))
4115 		return -EINVAL;
4116 
4117 	if (remove) {
4118 		entry = ftrace_lookup_ip(hash, ip);
4119 		if (!entry)
4120 			return -ENOENT;
4121 		free_hash_entry(hash, entry);
4122 		return 0;
4123 	}
4124 
4125 	return add_hash_entry(hash, ip);
4126 }
4127 
4128 static void ftrace_ops_update_code(struct ftrace_ops *ops,
4129 				   struct ftrace_ops_hash *old_hash)
4130 {
4131 	struct ftrace_ops *op;
4132 
4133 	if (!ftrace_enabled)
4134 		return;
4135 
4136 	if (ops->flags & FTRACE_OPS_FL_ENABLED) {
4137 		ftrace_run_modify_code(ops, FTRACE_UPDATE_CALLS, old_hash);
4138 		return;
4139 	}
4140 
4141 	/*
4142 	 * If this is the shared global_ops filter, then we need to
4143 	 * check if there is another ops that shares it, is enabled.
4144 	 * If so, we still need to run the modify code.
4145 	 */
4146 	if (ops->func_hash != &global_ops.local_hash)
4147 		return;
4148 
4149 	do_for_each_ftrace_op(op, ftrace_ops_list) {
4150 		if (op->func_hash == &global_ops.local_hash &&
4151 		    op->flags & FTRACE_OPS_FL_ENABLED) {
4152 			ftrace_run_modify_code(op, FTRACE_UPDATE_CALLS, old_hash);
4153 			/* Only need to do this once */
4154 			return;
4155 		}
4156 	} while_for_each_ftrace_op(op);
4157 }
4158 
4159 static int
4160 ftrace_set_hash(struct ftrace_ops *ops, unsigned char *buf, int len,
4161 		unsigned long ip, int remove, int reset, int enable)
4162 {
4163 	struct ftrace_hash **orig_hash;
4164 	struct ftrace_ops_hash old_hash_ops;
4165 	struct ftrace_hash *old_hash;
4166 	struct ftrace_hash *hash;
4167 	int ret;
4168 
4169 	if (unlikely(ftrace_disabled))
4170 		return -ENODEV;
4171 
4172 	mutex_lock(&ops->func_hash->regex_lock);
4173 
4174 	if (enable)
4175 		orig_hash = &ops->func_hash->filter_hash;
4176 	else
4177 		orig_hash = &ops->func_hash->notrace_hash;
4178 
4179 	if (reset)
4180 		hash = alloc_ftrace_hash(FTRACE_HASH_DEFAULT_BITS);
4181 	else
4182 		hash = alloc_and_copy_ftrace_hash(FTRACE_HASH_DEFAULT_BITS, *orig_hash);
4183 
4184 	if (!hash) {
4185 		ret = -ENOMEM;
4186 		goto out_regex_unlock;
4187 	}
4188 
4189 	if (buf && !ftrace_match_records(hash, buf, len)) {
4190 		ret = -EINVAL;
4191 		goto out_regex_unlock;
4192 	}
4193 	if (ip) {
4194 		ret = ftrace_match_addr(hash, ip, remove);
4195 		if (ret < 0)
4196 			goto out_regex_unlock;
4197 	}
4198 
4199 	mutex_lock(&ftrace_lock);
4200 	old_hash = *orig_hash;
4201 	old_hash_ops.filter_hash = ops->func_hash->filter_hash;
4202 	old_hash_ops.notrace_hash = ops->func_hash->notrace_hash;
4203 	ret = ftrace_hash_move(ops, enable, orig_hash, hash);
4204 	if (!ret) {
4205 		ftrace_ops_update_code(ops, &old_hash_ops);
4206 		free_ftrace_hash_rcu(old_hash);
4207 	}
4208 	mutex_unlock(&ftrace_lock);
4209 
4210  out_regex_unlock:
4211 	mutex_unlock(&ops->func_hash->regex_lock);
4212 
4213 	free_ftrace_hash(hash);
4214 	return ret;
4215 }
4216 
4217 static int
4218 ftrace_set_addr(struct ftrace_ops *ops, unsigned long ip, int remove,
4219 		int reset, int enable)
4220 {
4221 	return ftrace_set_hash(ops, 0, 0, ip, remove, reset, enable);
4222 }
4223 
4224 /**
4225  * ftrace_set_filter_ip - set a function to filter on in ftrace by address
4226  * @ops - the ops to set the filter with
4227  * @ip - the address to add to or remove from the filter.
4228  * @remove - non zero to remove the ip from the filter
4229  * @reset - non zero to reset all filters before applying this filter.
4230  *
4231  * Filters denote which functions should be enabled when tracing is enabled
4232  * If @ip is NULL, it failes to update filter.
4233  */
4234 int ftrace_set_filter_ip(struct ftrace_ops *ops, unsigned long ip,
4235 			 int remove, int reset)
4236 {
4237 	ftrace_ops_init(ops);
4238 	return ftrace_set_addr(ops, ip, remove, reset, 1);
4239 }
4240 EXPORT_SYMBOL_GPL(ftrace_set_filter_ip);
4241 
4242 static int
4243 ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len,
4244 		 int reset, int enable)
4245 {
4246 	return ftrace_set_hash(ops, buf, len, 0, 0, reset, enable);
4247 }
4248 
4249 /**
4250  * ftrace_set_filter - set a function to filter on in ftrace
4251  * @ops - the ops to set the filter with
4252  * @buf - the string that holds the function filter text.
4253  * @len - the length of the string.
4254  * @reset - non zero to reset all filters before applying this filter.
4255  *
4256  * Filters denote which functions should be enabled when tracing is enabled.
4257  * If @buf is NULL and reset is set, all functions will be enabled for tracing.
4258  */
4259 int ftrace_set_filter(struct ftrace_ops *ops, unsigned char *buf,
4260 		       int len, int reset)
4261 {
4262 	ftrace_ops_init(ops);
4263 	return ftrace_set_regex(ops, buf, len, reset, 1);
4264 }
4265 EXPORT_SYMBOL_GPL(ftrace_set_filter);
4266 
4267 /**
4268  * ftrace_set_notrace - set a function to not trace in ftrace
4269  * @ops - the ops to set the notrace filter with
4270  * @buf - the string that holds the function notrace text.
4271  * @len - the length of the string.
4272  * @reset - non zero to reset all filters before applying this filter.
4273  *
4274  * Notrace Filters denote which functions should not be enabled when tracing
4275  * is enabled. If @buf is NULL and reset is set, all functions will be enabled
4276  * for tracing.
4277  */
4278 int ftrace_set_notrace(struct ftrace_ops *ops, unsigned char *buf,
4279 			int len, int reset)
4280 {
4281 	ftrace_ops_init(ops);
4282 	return ftrace_set_regex(ops, buf, len, reset, 0);
4283 }
4284 EXPORT_SYMBOL_GPL(ftrace_set_notrace);
4285 /**
4286  * ftrace_set_global_filter - set a function to filter on with global tracers
4287  * @buf - the string that holds the function filter text.
4288  * @len - the length of the string.
4289  * @reset - non zero to reset all filters before applying this filter.
4290  *
4291  * Filters denote which functions should be enabled when tracing is enabled.
4292  * If @buf is NULL and reset is set, all functions will be enabled for tracing.
4293  */
4294 void ftrace_set_global_filter(unsigned char *buf, int len, int reset)
4295 {
4296 	ftrace_set_regex(&global_ops, buf, len, reset, 1);
4297 }
4298 EXPORT_SYMBOL_GPL(ftrace_set_global_filter);
4299 
4300 /**
4301  * ftrace_set_global_notrace - set a function to not trace with global tracers
4302  * @buf - the string that holds the function notrace text.
4303  * @len - the length of the string.
4304  * @reset - non zero to reset all filters before applying this filter.
4305  *
4306  * Notrace Filters denote which functions should not be enabled when tracing
4307  * is enabled. If @buf is NULL and reset is set, all functions will be enabled
4308  * for tracing.
4309  */
4310 void ftrace_set_global_notrace(unsigned char *buf, int len, int reset)
4311 {
4312 	ftrace_set_regex(&global_ops, buf, len, reset, 0);
4313 }
4314 EXPORT_SYMBOL_GPL(ftrace_set_global_notrace);
4315 
4316 /*
4317  * command line interface to allow users to set filters on boot up.
4318  */
4319 #define FTRACE_FILTER_SIZE		COMMAND_LINE_SIZE
4320 static char ftrace_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
4321 static char ftrace_filter_buf[FTRACE_FILTER_SIZE] __initdata;
4322 
4323 /* Used by function selftest to not test if filter is set */
4324 bool ftrace_filter_param __initdata;
4325 
4326 static int __init set_ftrace_notrace(char *str)
4327 {
4328 	ftrace_filter_param = true;
4329 	strlcpy(ftrace_notrace_buf, str, FTRACE_FILTER_SIZE);
4330 	return 1;
4331 }
4332 __setup("ftrace_notrace=", set_ftrace_notrace);
4333 
4334 static int __init set_ftrace_filter(char *str)
4335 {
4336 	ftrace_filter_param = true;
4337 	strlcpy(ftrace_filter_buf, str, FTRACE_FILTER_SIZE);
4338 	return 1;
4339 }
4340 __setup("ftrace_filter=", set_ftrace_filter);
4341 
4342 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
4343 static char ftrace_graph_buf[FTRACE_FILTER_SIZE] __initdata;
4344 static char ftrace_graph_notrace_buf[FTRACE_FILTER_SIZE] __initdata;
4345 static int ftrace_set_func(unsigned long *array, int *idx, int size, char *buffer);
4346 
4347 static unsigned long save_global_trampoline;
4348 static unsigned long save_global_flags;
4349 
4350 static int __init set_graph_function(char *str)
4351 {
4352 	strlcpy(ftrace_graph_buf, str, FTRACE_FILTER_SIZE);
4353 	return 1;
4354 }
4355 __setup("ftrace_graph_filter=", set_graph_function);
4356 
4357 static int __init set_graph_notrace_function(char *str)
4358 {
4359 	strlcpy(ftrace_graph_notrace_buf, str, FTRACE_FILTER_SIZE);
4360 	return 1;
4361 }
4362 __setup("ftrace_graph_notrace=", set_graph_notrace_function);
4363 
4364 static void __init set_ftrace_early_graph(char *buf, int enable)
4365 {
4366 	int ret;
4367 	char *func;
4368 	unsigned long *table = ftrace_graph_funcs;
4369 	int *count = &ftrace_graph_count;
4370 
4371 	if (!enable) {
4372 		table = ftrace_graph_notrace_funcs;
4373 		count = &ftrace_graph_notrace_count;
4374 	}
4375 
4376 	while (buf) {
4377 		func = strsep(&buf, ",");
4378 		/* we allow only one expression at a time */
4379 		ret = ftrace_set_func(table, count, FTRACE_GRAPH_MAX_FUNCS, func);
4380 		if (ret)
4381 			printk(KERN_DEBUG "ftrace: function %s not "
4382 					  "traceable\n", func);
4383 	}
4384 }
4385 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
4386 
4387 void __init
4388 ftrace_set_early_filter(struct ftrace_ops *ops, char *buf, int enable)
4389 {
4390 	char *func;
4391 
4392 	ftrace_ops_init(ops);
4393 
4394 	while (buf) {
4395 		func = strsep(&buf, ",");
4396 		ftrace_set_regex(ops, func, strlen(func), 0, enable);
4397 	}
4398 }
4399 
4400 static void __init set_ftrace_early_filters(void)
4401 {
4402 	if (ftrace_filter_buf[0])
4403 		ftrace_set_early_filter(&global_ops, ftrace_filter_buf, 1);
4404 	if (ftrace_notrace_buf[0])
4405 		ftrace_set_early_filter(&global_ops, ftrace_notrace_buf, 0);
4406 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
4407 	if (ftrace_graph_buf[0])
4408 		set_ftrace_early_graph(ftrace_graph_buf, 1);
4409 	if (ftrace_graph_notrace_buf[0])
4410 		set_ftrace_early_graph(ftrace_graph_notrace_buf, 0);
4411 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
4412 }
4413 
4414 int ftrace_regex_release(struct inode *inode, struct file *file)
4415 {
4416 	struct seq_file *m = (struct seq_file *)file->private_data;
4417 	struct ftrace_ops_hash old_hash_ops;
4418 	struct ftrace_iterator *iter;
4419 	struct ftrace_hash **orig_hash;
4420 	struct ftrace_hash *old_hash;
4421 	struct trace_parser *parser;
4422 	int filter_hash;
4423 	int ret;
4424 
4425 	if (file->f_mode & FMODE_READ) {
4426 		iter = m->private;
4427 		seq_release(inode, file);
4428 	} else
4429 		iter = file->private_data;
4430 
4431 	parser = &iter->parser;
4432 	if (trace_parser_loaded(parser)) {
4433 		parser->buffer[parser->idx] = 0;
4434 		ftrace_match_records(iter->hash, parser->buffer, parser->idx);
4435 	}
4436 
4437 	trace_parser_put(parser);
4438 
4439 	mutex_lock(&iter->ops->func_hash->regex_lock);
4440 
4441 	if (file->f_mode & FMODE_WRITE) {
4442 		filter_hash = !!(iter->flags & FTRACE_ITER_FILTER);
4443 
4444 		if (filter_hash)
4445 			orig_hash = &iter->ops->func_hash->filter_hash;
4446 		else
4447 			orig_hash = &iter->ops->func_hash->notrace_hash;
4448 
4449 		mutex_lock(&ftrace_lock);
4450 		old_hash = *orig_hash;
4451 		old_hash_ops.filter_hash = iter->ops->func_hash->filter_hash;
4452 		old_hash_ops.notrace_hash = iter->ops->func_hash->notrace_hash;
4453 		ret = ftrace_hash_move(iter->ops, filter_hash,
4454 				       orig_hash, iter->hash);
4455 		if (!ret) {
4456 			ftrace_ops_update_code(iter->ops, &old_hash_ops);
4457 			free_ftrace_hash_rcu(old_hash);
4458 		}
4459 		mutex_unlock(&ftrace_lock);
4460 	}
4461 
4462 	mutex_unlock(&iter->ops->func_hash->regex_lock);
4463 	free_ftrace_hash(iter->hash);
4464 	kfree(iter);
4465 
4466 	return 0;
4467 }
4468 
4469 static const struct file_operations ftrace_avail_fops = {
4470 	.open = ftrace_avail_open,
4471 	.read = seq_read,
4472 	.llseek = seq_lseek,
4473 	.release = seq_release_private,
4474 };
4475 
4476 static const struct file_operations ftrace_enabled_fops = {
4477 	.open = ftrace_enabled_open,
4478 	.read = seq_read,
4479 	.llseek = seq_lseek,
4480 	.release = seq_release_private,
4481 };
4482 
4483 static const struct file_operations ftrace_filter_fops = {
4484 	.open = ftrace_filter_open,
4485 	.read = seq_read,
4486 	.write = ftrace_filter_write,
4487 	.llseek = tracing_lseek,
4488 	.release = ftrace_regex_release,
4489 };
4490 
4491 static const struct file_operations ftrace_notrace_fops = {
4492 	.open = ftrace_notrace_open,
4493 	.read = seq_read,
4494 	.write = ftrace_notrace_write,
4495 	.llseek = tracing_lseek,
4496 	.release = ftrace_regex_release,
4497 };
4498 
4499 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
4500 
4501 static DEFINE_MUTEX(graph_lock);
4502 
4503 int ftrace_graph_count;
4504 int ftrace_graph_notrace_count;
4505 unsigned long ftrace_graph_funcs[FTRACE_GRAPH_MAX_FUNCS] __read_mostly;
4506 unsigned long ftrace_graph_notrace_funcs[FTRACE_GRAPH_MAX_FUNCS] __read_mostly;
4507 
4508 struct ftrace_graph_data {
4509 	unsigned long *table;
4510 	size_t size;
4511 	int *count;
4512 	const struct seq_operations *seq_ops;
4513 };
4514 
4515 static void *
4516 __g_next(struct seq_file *m, loff_t *pos)
4517 {
4518 	struct ftrace_graph_data *fgd = m->private;
4519 
4520 	if (*pos >= *fgd->count)
4521 		return NULL;
4522 	return &fgd->table[*pos];
4523 }
4524 
4525 static void *
4526 g_next(struct seq_file *m, void *v, loff_t *pos)
4527 {
4528 	(*pos)++;
4529 	return __g_next(m, pos);
4530 }
4531 
4532 static void *g_start(struct seq_file *m, loff_t *pos)
4533 {
4534 	struct ftrace_graph_data *fgd = m->private;
4535 
4536 	mutex_lock(&graph_lock);
4537 
4538 	/* Nothing, tell g_show to print all functions are enabled */
4539 	if (!*fgd->count && !*pos)
4540 		return (void *)1;
4541 
4542 	return __g_next(m, pos);
4543 }
4544 
4545 static void g_stop(struct seq_file *m, void *p)
4546 {
4547 	mutex_unlock(&graph_lock);
4548 }
4549 
4550 static int g_show(struct seq_file *m, void *v)
4551 {
4552 	unsigned long *ptr = v;
4553 
4554 	if (!ptr)
4555 		return 0;
4556 
4557 	if (ptr == (unsigned long *)1) {
4558 		struct ftrace_graph_data *fgd = m->private;
4559 
4560 		if (fgd->table == ftrace_graph_funcs)
4561 			seq_puts(m, "#### all functions enabled ####\n");
4562 		else
4563 			seq_puts(m, "#### no functions disabled ####\n");
4564 		return 0;
4565 	}
4566 
4567 	seq_printf(m, "%ps\n", (void *)*ptr);
4568 
4569 	return 0;
4570 }
4571 
4572 static const struct seq_operations ftrace_graph_seq_ops = {
4573 	.start = g_start,
4574 	.next = g_next,
4575 	.stop = g_stop,
4576 	.show = g_show,
4577 };
4578 
4579 static int
4580 __ftrace_graph_open(struct inode *inode, struct file *file,
4581 		    struct ftrace_graph_data *fgd)
4582 {
4583 	int ret = 0;
4584 
4585 	mutex_lock(&graph_lock);
4586 	if ((file->f_mode & FMODE_WRITE) &&
4587 	    (file->f_flags & O_TRUNC)) {
4588 		*fgd->count = 0;
4589 		memset(fgd->table, 0, fgd->size * sizeof(*fgd->table));
4590 	}
4591 	mutex_unlock(&graph_lock);
4592 
4593 	if (file->f_mode & FMODE_READ) {
4594 		ret = seq_open(file, fgd->seq_ops);
4595 		if (!ret) {
4596 			struct seq_file *m = file->private_data;
4597 			m->private = fgd;
4598 		}
4599 	} else
4600 		file->private_data = fgd;
4601 
4602 	return ret;
4603 }
4604 
4605 static int
4606 ftrace_graph_open(struct inode *inode, struct file *file)
4607 {
4608 	struct ftrace_graph_data *fgd;
4609 
4610 	if (unlikely(ftrace_disabled))
4611 		return -ENODEV;
4612 
4613 	fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
4614 	if (fgd == NULL)
4615 		return -ENOMEM;
4616 
4617 	fgd->table = ftrace_graph_funcs;
4618 	fgd->size = FTRACE_GRAPH_MAX_FUNCS;
4619 	fgd->count = &ftrace_graph_count;
4620 	fgd->seq_ops = &ftrace_graph_seq_ops;
4621 
4622 	return __ftrace_graph_open(inode, file, fgd);
4623 }
4624 
4625 static int
4626 ftrace_graph_notrace_open(struct inode *inode, struct file *file)
4627 {
4628 	struct ftrace_graph_data *fgd;
4629 
4630 	if (unlikely(ftrace_disabled))
4631 		return -ENODEV;
4632 
4633 	fgd = kmalloc(sizeof(*fgd), GFP_KERNEL);
4634 	if (fgd == NULL)
4635 		return -ENOMEM;
4636 
4637 	fgd->table = ftrace_graph_notrace_funcs;
4638 	fgd->size = FTRACE_GRAPH_MAX_FUNCS;
4639 	fgd->count = &ftrace_graph_notrace_count;
4640 	fgd->seq_ops = &ftrace_graph_seq_ops;
4641 
4642 	return __ftrace_graph_open(inode, file, fgd);
4643 }
4644 
4645 static int
4646 ftrace_graph_release(struct inode *inode, struct file *file)
4647 {
4648 	if (file->f_mode & FMODE_READ) {
4649 		struct seq_file *m = file->private_data;
4650 
4651 		kfree(m->private);
4652 		seq_release(inode, file);
4653 	} else {
4654 		kfree(file->private_data);
4655 	}
4656 
4657 	return 0;
4658 }
4659 
4660 static int
4661 ftrace_set_func(unsigned long *array, int *idx, int size, char *buffer)
4662 {
4663 	struct ftrace_glob func_g;
4664 	struct dyn_ftrace *rec;
4665 	struct ftrace_page *pg;
4666 	int fail = 1;
4667 	int not;
4668 	bool exists;
4669 	int i;
4670 
4671 	/* decode regex */
4672 	func_g.type = filter_parse_regex(buffer, strlen(buffer),
4673 					 &func_g.search, &not);
4674 	if (!not && *idx >= size)
4675 		return -EBUSY;
4676 
4677 	func_g.len = strlen(func_g.search);
4678 
4679 	mutex_lock(&ftrace_lock);
4680 
4681 	if (unlikely(ftrace_disabled)) {
4682 		mutex_unlock(&ftrace_lock);
4683 		return -ENODEV;
4684 	}
4685 
4686 	do_for_each_ftrace_rec(pg, rec) {
4687 
4688 		if (ftrace_match_record(rec, &func_g, NULL, 0)) {
4689 			/* if it is in the array */
4690 			exists = false;
4691 			for (i = 0; i < *idx; i++) {
4692 				if (array[i] == rec->ip) {
4693 					exists = true;
4694 					break;
4695 				}
4696 			}
4697 
4698 			if (!not) {
4699 				fail = 0;
4700 				if (!exists) {
4701 					array[(*idx)++] = rec->ip;
4702 					if (*idx >= size)
4703 						goto out;
4704 				}
4705 			} else {
4706 				if (exists) {
4707 					array[i] = array[--(*idx)];
4708 					array[*idx] = 0;
4709 					fail = 0;
4710 				}
4711 			}
4712 		}
4713 	} while_for_each_ftrace_rec();
4714 out:
4715 	mutex_unlock(&ftrace_lock);
4716 
4717 	if (fail)
4718 		return -EINVAL;
4719 
4720 	return 0;
4721 }
4722 
4723 static ssize_t
4724 ftrace_graph_write(struct file *file, const char __user *ubuf,
4725 		   size_t cnt, loff_t *ppos)
4726 {
4727 	struct trace_parser parser;
4728 	ssize_t read, ret = 0;
4729 	struct ftrace_graph_data *fgd = file->private_data;
4730 
4731 	if (!cnt)
4732 		return 0;
4733 
4734 	if (trace_parser_get_init(&parser, FTRACE_BUFF_MAX))
4735 		return -ENOMEM;
4736 
4737 	read = trace_get_user(&parser, ubuf, cnt, ppos);
4738 
4739 	if (read >= 0 && trace_parser_loaded((&parser))) {
4740 		parser.buffer[parser.idx] = 0;
4741 
4742 		mutex_lock(&graph_lock);
4743 
4744 		/* we allow only one expression at a time */
4745 		ret = ftrace_set_func(fgd->table, fgd->count, fgd->size,
4746 				      parser.buffer);
4747 
4748 		mutex_unlock(&graph_lock);
4749 	}
4750 
4751 	if (!ret)
4752 		ret = read;
4753 
4754 	trace_parser_put(&parser);
4755 
4756 	return ret;
4757 }
4758 
4759 static const struct file_operations ftrace_graph_fops = {
4760 	.open		= ftrace_graph_open,
4761 	.read		= seq_read,
4762 	.write		= ftrace_graph_write,
4763 	.llseek		= tracing_lseek,
4764 	.release	= ftrace_graph_release,
4765 };
4766 
4767 static const struct file_operations ftrace_graph_notrace_fops = {
4768 	.open		= ftrace_graph_notrace_open,
4769 	.read		= seq_read,
4770 	.write		= ftrace_graph_write,
4771 	.llseek		= tracing_lseek,
4772 	.release	= ftrace_graph_release,
4773 };
4774 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
4775 
4776 void ftrace_create_filter_files(struct ftrace_ops *ops,
4777 				struct dentry *parent)
4778 {
4779 
4780 	trace_create_file("set_ftrace_filter", 0644, parent,
4781 			  ops, &ftrace_filter_fops);
4782 
4783 	trace_create_file("set_ftrace_notrace", 0644, parent,
4784 			  ops, &ftrace_notrace_fops);
4785 }
4786 
4787 /*
4788  * The name "destroy_filter_files" is really a misnomer. Although
4789  * in the future, it may actualy delete the files, but this is
4790  * really intended to make sure the ops passed in are disabled
4791  * and that when this function returns, the caller is free to
4792  * free the ops.
4793  *
4794  * The "destroy" name is only to match the "create" name that this
4795  * should be paired with.
4796  */
4797 void ftrace_destroy_filter_files(struct ftrace_ops *ops)
4798 {
4799 	mutex_lock(&ftrace_lock);
4800 	if (ops->flags & FTRACE_OPS_FL_ENABLED)
4801 		ftrace_shutdown(ops, 0);
4802 	ops->flags |= FTRACE_OPS_FL_DELETED;
4803 	mutex_unlock(&ftrace_lock);
4804 }
4805 
4806 static __init int ftrace_init_dyn_tracefs(struct dentry *d_tracer)
4807 {
4808 
4809 	trace_create_file("available_filter_functions", 0444,
4810 			d_tracer, NULL, &ftrace_avail_fops);
4811 
4812 	trace_create_file("enabled_functions", 0444,
4813 			d_tracer, NULL, &ftrace_enabled_fops);
4814 
4815 	ftrace_create_filter_files(&global_ops, d_tracer);
4816 
4817 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
4818 	trace_create_file("set_graph_function", 0444, d_tracer,
4819 				    NULL,
4820 				    &ftrace_graph_fops);
4821 	trace_create_file("set_graph_notrace", 0444, d_tracer,
4822 				    NULL,
4823 				    &ftrace_graph_notrace_fops);
4824 #endif /* CONFIG_FUNCTION_GRAPH_TRACER */
4825 
4826 	return 0;
4827 }
4828 
4829 static int ftrace_cmp_ips(const void *a, const void *b)
4830 {
4831 	const unsigned long *ipa = a;
4832 	const unsigned long *ipb = b;
4833 
4834 	if (*ipa > *ipb)
4835 		return 1;
4836 	if (*ipa < *ipb)
4837 		return -1;
4838 	return 0;
4839 }
4840 
4841 static int ftrace_process_locs(struct module *mod,
4842 			       unsigned long *start,
4843 			       unsigned long *end)
4844 {
4845 	struct ftrace_page *start_pg;
4846 	struct ftrace_page *pg;
4847 	struct dyn_ftrace *rec;
4848 	unsigned long count;
4849 	unsigned long *p;
4850 	unsigned long addr;
4851 	unsigned long flags = 0; /* Shut up gcc */
4852 	int ret = -ENOMEM;
4853 
4854 	count = end - start;
4855 
4856 	if (!count)
4857 		return 0;
4858 
4859 	sort(start, count, sizeof(*start),
4860 	     ftrace_cmp_ips, NULL);
4861 
4862 	start_pg = ftrace_allocate_pages(count);
4863 	if (!start_pg)
4864 		return -ENOMEM;
4865 
4866 	mutex_lock(&ftrace_lock);
4867 
4868 	/*
4869 	 * Core and each module needs their own pages, as
4870 	 * modules will free them when they are removed.
4871 	 * Force a new page to be allocated for modules.
4872 	 */
4873 	if (!mod) {
4874 		WARN_ON(ftrace_pages || ftrace_pages_start);
4875 		/* First initialization */
4876 		ftrace_pages = ftrace_pages_start = start_pg;
4877 	} else {
4878 		if (!ftrace_pages)
4879 			goto out;
4880 
4881 		if (WARN_ON(ftrace_pages->next)) {
4882 			/* Hmm, we have free pages? */
4883 			while (ftrace_pages->next)
4884 				ftrace_pages = ftrace_pages->next;
4885 		}
4886 
4887 		ftrace_pages->next = start_pg;
4888 	}
4889 
4890 	p = start;
4891 	pg = start_pg;
4892 	while (p < end) {
4893 		addr = ftrace_call_adjust(*p++);
4894 		/*
4895 		 * Some architecture linkers will pad between
4896 		 * the different mcount_loc sections of different
4897 		 * object files to satisfy alignments.
4898 		 * Skip any NULL pointers.
4899 		 */
4900 		if (!addr)
4901 			continue;
4902 
4903 		if (pg->index == pg->size) {
4904 			/* We should have allocated enough */
4905 			if (WARN_ON(!pg->next))
4906 				break;
4907 			pg = pg->next;
4908 		}
4909 
4910 		rec = &pg->records[pg->index++];
4911 		rec->ip = addr;
4912 	}
4913 
4914 	/* We should have used all pages */
4915 	WARN_ON(pg->next);
4916 
4917 	/* Assign the last page to ftrace_pages */
4918 	ftrace_pages = pg;
4919 
4920 	/*
4921 	 * We only need to disable interrupts on start up
4922 	 * because we are modifying code that an interrupt
4923 	 * may execute, and the modification is not atomic.
4924 	 * But for modules, nothing runs the code we modify
4925 	 * until we are finished with it, and there's no
4926 	 * reason to cause large interrupt latencies while we do it.
4927 	 */
4928 	if (!mod)
4929 		local_irq_save(flags);
4930 	ftrace_update_code(mod, start_pg);
4931 	if (!mod)
4932 		local_irq_restore(flags);
4933 	ret = 0;
4934  out:
4935 	mutex_unlock(&ftrace_lock);
4936 
4937 	return ret;
4938 }
4939 
4940 #ifdef CONFIG_MODULES
4941 
4942 #define next_to_ftrace_page(p) container_of(p, struct ftrace_page, next)
4943 
4944 static int referenced_filters(struct dyn_ftrace *rec)
4945 {
4946 	struct ftrace_ops *ops;
4947 	int cnt = 0;
4948 
4949 	for (ops = ftrace_ops_list; ops != &ftrace_list_end; ops = ops->next) {
4950 		if (ops_references_rec(ops, rec))
4951 		    cnt++;
4952 	}
4953 
4954 	return cnt;
4955 }
4956 
4957 void ftrace_release_mod(struct module *mod)
4958 {
4959 	struct dyn_ftrace *rec;
4960 	struct ftrace_page **last_pg;
4961 	struct ftrace_page *pg;
4962 	int order;
4963 
4964 	mutex_lock(&ftrace_lock);
4965 
4966 	if (ftrace_disabled)
4967 		goto out_unlock;
4968 
4969 	/*
4970 	 * Each module has its own ftrace_pages, remove
4971 	 * them from the list.
4972 	 */
4973 	last_pg = &ftrace_pages_start;
4974 	for (pg = ftrace_pages_start; pg; pg = *last_pg) {
4975 		rec = &pg->records[0];
4976 		if (within_module_core(rec->ip, mod)) {
4977 			/*
4978 			 * As core pages are first, the first
4979 			 * page should never be a module page.
4980 			 */
4981 			if (WARN_ON(pg == ftrace_pages_start))
4982 				goto out_unlock;
4983 
4984 			/* Check if we are deleting the last page */
4985 			if (pg == ftrace_pages)
4986 				ftrace_pages = next_to_ftrace_page(last_pg);
4987 
4988 			*last_pg = pg->next;
4989 			order = get_count_order(pg->size / ENTRIES_PER_PAGE);
4990 			free_pages((unsigned long)pg->records, order);
4991 			kfree(pg);
4992 		} else
4993 			last_pg = &pg->next;
4994 	}
4995  out_unlock:
4996 	mutex_unlock(&ftrace_lock);
4997 }
4998 
4999 void ftrace_module_enable(struct module *mod)
5000 {
5001 	struct dyn_ftrace *rec;
5002 	struct ftrace_page *pg;
5003 
5004 	mutex_lock(&ftrace_lock);
5005 
5006 	if (ftrace_disabled)
5007 		goto out_unlock;
5008 
5009 	/*
5010 	 * If the tracing is enabled, go ahead and enable the record.
5011 	 *
5012 	 * The reason not to enable the record immediatelly is the
5013 	 * inherent check of ftrace_make_nop/ftrace_make_call for
5014 	 * correct previous instructions.  Making first the NOP
5015 	 * conversion puts the module to the correct state, thus
5016 	 * passing the ftrace_make_call check.
5017 	 *
5018 	 * We also delay this to after the module code already set the
5019 	 * text to read-only, as we now need to set it back to read-write
5020 	 * so that we can modify the text.
5021 	 */
5022 	if (ftrace_start_up)
5023 		ftrace_arch_code_modify_prepare();
5024 
5025 	do_for_each_ftrace_rec(pg, rec) {
5026 		int cnt;
5027 		/*
5028 		 * do_for_each_ftrace_rec() is a double loop.
5029 		 * module text shares the pg. If a record is
5030 		 * not part of this module, then skip this pg,
5031 		 * which the "break" will do.
5032 		 */
5033 		if (!within_module_core(rec->ip, mod))
5034 			break;
5035 
5036 		cnt = 0;
5037 
5038 		/*
5039 		 * When adding a module, we need to check if tracers are
5040 		 * currently enabled and if they are, and can trace this record,
5041 		 * we need to enable the module functions as well as update the
5042 		 * reference counts for those function records.
5043 		 */
5044 		if (ftrace_start_up)
5045 			cnt += referenced_filters(rec);
5046 
5047 		/* This clears FTRACE_FL_DISABLED */
5048 		rec->flags = cnt;
5049 
5050 		if (ftrace_start_up && cnt) {
5051 			int failed = __ftrace_replace_code(rec, 1);
5052 			if (failed) {
5053 				ftrace_bug(failed, rec);
5054 				goto out_loop;
5055 			}
5056 		}
5057 
5058 	} while_for_each_ftrace_rec();
5059 
5060  out_loop:
5061 	if (ftrace_start_up)
5062 		ftrace_arch_code_modify_post_process();
5063 
5064  out_unlock:
5065 	mutex_unlock(&ftrace_lock);
5066 }
5067 
5068 void ftrace_module_init(struct module *mod)
5069 {
5070 	if (ftrace_disabled || !mod->num_ftrace_callsites)
5071 		return;
5072 
5073 	ftrace_process_locs(mod, mod->ftrace_callsites,
5074 			    mod->ftrace_callsites + mod->num_ftrace_callsites);
5075 }
5076 #endif /* CONFIG_MODULES */
5077 
5078 void __init ftrace_init(void)
5079 {
5080 	extern unsigned long __start_mcount_loc[];
5081 	extern unsigned long __stop_mcount_loc[];
5082 	unsigned long count, flags;
5083 	int ret;
5084 
5085 	local_irq_save(flags);
5086 	ret = ftrace_dyn_arch_init();
5087 	local_irq_restore(flags);
5088 	if (ret)
5089 		goto failed;
5090 
5091 	count = __stop_mcount_loc - __start_mcount_loc;
5092 	if (!count) {
5093 		pr_info("ftrace: No functions to be traced?\n");
5094 		goto failed;
5095 	}
5096 
5097 	pr_info("ftrace: allocating %ld entries in %ld pages\n",
5098 		count, count / ENTRIES_PER_PAGE + 1);
5099 
5100 	last_ftrace_enabled = ftrace_enabled = 1;
5101 
5102 	ret = ftrace_process_locs(NULL,
5103 				  __start_mcount_loc,
5104 				  __stop_mcount_loc);
5105 
5106 	set_ftrace_early_filters();
5107 
5108 	return;
5109  failed:
5110 	ftrace_disabled = 1;
5111 }
5112 
5113 /* Do nothing if arch does not support this */
5114 void __weak arch_ftrace_update_trampoline(struct ftrace_ops *ops)
5115 {
5116 }
5117 
5118 static void ftrace_update_trampoline(struct ftrace_ops *ops)
5119 {
5120 
5121 /*
5122  * Currently there's no safe way to free a trampoline when the kernel
5123  * is configured with PREEMPT. That is because a task could be preempted
5124  * when it jumped to the trampoline, it may be preempted for a long time
5125  * depending on the system load, and currently there's no way to know
5126  * when it will be off the trampoline. If the trampoline is freed
5127  * too early, when the task runs again, it will be executing on freed
5128  * memory and crash.
5129  */
5130 #ifdef CONFIG_PREEMPT
5131 	/* Currently, only non dynamic ops can have a trampoline */
5132 	if (ops->flags & FTRACE_OPS_FL_DYNAMIC)
5133 		return;
5134 #endif
5135 
5136 	arch_ftrace_update_trampoline(ops);
5137 }
5138 
5139 #else
5140 
5141 static struct ftrace_ops global_ops = {
5142 	.func			= ftrace_stub,
5143 	.flags			= FTRACE_OPS_FL_RECURSION_SAFE |
5144 				  FTRACE_OPS_FL_INITIALIZED |
5145 				  FTRACE_OPS_FL_PID,
5146 };
5147 
5148 static int __init ftrace_nodyn_init(void)
5149 {
5150 	ftrace_enabled = 1;
5151 	return 0;
5152 }
5153 core_initcall(ftrace_nodyn_init);
5154 
5155 static inline int ftrace_init_dyn_tracefs(struct dentry *d_tracer) { return 0; }
5156 static inline void ftrace_startup_enable(int command) { }
5157 static inline void ftrace_startup_all(int command) { }
5158 /* Keep as macros so we do not need to define the commands */
5159 # define ftrace_startup(ops, command)					\
5160 	({								\
5161 		int ___ret = __register_ftrace_function(ops);		\
5162 		if (!___ret)						\
5163 			(ops)->flags |= FTRACE_OPS_FL_ENABLED;		\
5164 		___ret;							\
5165 	})
5166 # define ftrace_shutdown(ops, command)					\
5167 	({								\
5168 		int ___ret = __unregister_ftrace_function(ops);		\
5169 		if (!___ret)						\
5170 			(ops)->flags &= ~FTRACE_OPS_FL_ENABLED;		\
5171 		___ret;							\
5172 	})
5173 
5174 # define ftrace_startup_sysctl()	do { } while (0)
5175 # define ftrace_shutdown_sysctl()	do { } while (0)
5176 
5177 static inline int
5178 ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip, void *regs)
5179 {
5180 	return 1;
5181 }
5182 
5183 static void ftrace_update_trampoline(struct ftrace_ops *ops)
5184 {
5185 }
5186 
5187 #endif /* CONFIG_DYNAMIC_FTRACE */
5188 
5189 __init void ftrace_init_global_array_ops(struct trace_array *tr)
5190 {
5191 	tr->ops = &global_ops;
5192 	tr->ops->private = tr;
5193 }
5194 
5195 void ftrace_init_array_ops(struct trace_array *tr, ftrace_func_t func)
5196 {
5197 	/* If we filter on pids, update to use the pid function */
5198 	if (tr->flags & TRACE_ARRAY_FL_GLOBAL) {
5199 		if (WARN_ON(tr->ops->func != ftrace_stub))
5200 			printk("ftrace ops had %pS for function\n",
5201 			       tr->ops->func);
5202 	}
5203 	tr->ops->func = func;
5204 	tr->ops->private = tr;
5205 }
5206 
5207 void ftrace_reset_array_ops(struct trace_array *tr)
5208 {
5209 	tr->ops->func = ftrace_stub;
5210 }
5211 
5212 static inline void
5213 __ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
5214 		       struct ftrace_ops *ignored, struct pt_regs *regs)
5215 {
5216 	struct ftrace_ops *op;
5217 	int bit;
5218 
5219 	bit = trace_test_and_set_recursion(TRACE_LIST_START, TRACE_LIST_MAX);
5220 	if (bit < 0)
5221 		return;
5222 
5223 	/*
5224 	 * Some of the ops may be dynamically allocated,
5225 	 * they must be freed after a synchronize_sched().
5226 	 */
5227 	preempt_disable_notrace();
5228 
5229 	do_for_each_ftrace_op(op, ftrace_ops_list) {
5230 		/*
5231 		 * Check the following for each ops before calling their func:
5232 		 *  if RCU flag is set, then rcu_is_watching() must be true
5233 		 *  if PER_CPU is set, then ftrace_function_local_disable()
5234 		 *                          must be false
5235 		 *  Otherwise test if the ip matches the ops filter
5236 		 *
5237 		 * If any of the above fails then the op->func() is not executed.
5238 		 */
5239 		if ((!(op->flags & FTRACE_OPS_FL_RCU) || rcu_is_watching()) &&
5240 		    (!(op->flags & FTRACE_OPS_FL_PER_CPU) ||
5241 		     !ftrace_function_local_disabled(op)) &&
5242 		    ftrace_ops_test(op, ip, regs)) {
5243 
5244 			if (FTRACE_WARN_ON(!op->func)) {
5245 				pr_warn("op=%p %pS\n", op, op);
5246 				goto out;
5247 			}
5248 			op->func(ip, parent_ip, op, regs);
5249 		}
5250 	} while_for_each_ftrace_op(op);
5251 out:
5252 	preempt_enable_notrace();
5253 	trace_clear_recursion(bit);
5254 }
5255 
5256 /*
5257  * Some archs only support passing ip and parent_ip. Even though
5258  * the list function ignores the op parameter, we do not want any
5259  * C side effects, where a function is called without the caller
5260  * sending a third parameter.
5261  * Archs are to support both the regs and ftrace_ops at the same time.
5262  * If they support ftrace_ops, it is assumed they support regs.
5263  * If call backs want to use regs, they must either check for regs
5264  * being NULL, or CONFIG_DYNAMIC_FTRACE_WITH_REGS.
5265  * Note, CONFIG_DYNAMIC_FTRACE_WITH_REGS expects a full regs to be saved.
5266  * An architecture can pass partial regs with ftrace_ops and still
5267  * set the ARCH_SUPPORTS_FTRACE_OPS.
5268  */
5269 #if ARCH_SUPPORTS_FTRACE_OPS
5270 static void ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
5271 				 struct ftrace_ops *op, struct pt_regs *regs)
5272 {
5273 	__ftrace_ops_list_func(ip, parent_ip, NULL, regs);
5274 }
5275 #else
5276 static void ftrace_ops_no_ops(unsigned long ip, unsigned long parent_ip)
5277 {
5278 	__ftrace_ops_list_func(ip, parent_ip, NULL, NULL);
5279 }
5280 #endif
5281 
5282 /*
5283  * If there's only one function registered but it does not support
5284  * recursion, needs RCU protection and/or requires per cpu handling, then
5285  * this function will be called by the mcount trampoline.
5286  */
5287 static void ftrace_ops_assist_func(unsigned long ip, unsigned long parent_ip,
5288 				   struct ftrace_ops *op, struct pt_regs *regs)
5289 {
5290 	int bit;
5291 
5292 	if ((op->flags & FTRACE_OPS_FL_RCU) && !rcu_is_watching())
5293 		return;
5294 
5295 	bit = trace_test_and_set_recursion(TRACE_LIST_START, TRACE_LIST_MAX);
5296 	if (bit < 0)
5297 		return;
5298 
5299 	preempt_disable_notrace();
5300 
5301 	if (!(op->flags & FTRACE_OPS_FL_PER_CPU) ||
5302 	    !ftrace_function_local_disabled(op)) {
5303 		op->func(ip, parent_ip, op, regs);
5304 	}
5305 
5306 	preempt_enable_notrace();
5307 	trace_clear_recursion(bit);
5308 }
5309 
5310 /**
5311  * ftrace_ops_get_func - get the function a trampoline should call
5312  * @ops: the ops to get the function for
5313  *
5314  * Normally the mcount trampoline will call the ops->func, but there
5315  * are times that it should not. For example, if the ops does not
5316  * have its own recursion protection, then it should call the
5317  * ftrace_ops_recurs_func() instead.
5318  *
5319  * Returns the function that the trampoline should call for @ops.
5320  */
5321 ftrace_func_t ftrace_ops_get_func(struct ftrace_ops *ops)
5322 {
5323 	/*
5324 	 * If the function does not handle recursion, needs to be RCU safe,
5325 	 * or does per cpu logic, then we need to call the assist handler.
5326 	 */
5327 	if (!(ops->flags & FTRACE_OPS_FL_RECURSION_SAFE) ||
5328 	    ops->flags & (FTRACE_OPS_FL_RCU | FTRACE_OPS_FL_PER_CPU))
5329 		return ftrace_ops_assist_func;
5330 
5331 	return ops->func;
5332 }
5333 
5334 static void
5335 ftrace_filter_pid_sched_switch_probe(void *data, bool preempt,
5336 		    struct task_struct *prev, struct task_struct *next)
5337 {
5338 	struct trace_array *tr = data;
5339 	struct trace_pid_list *pid_list;
5340 
5341 	pid_list = rcu_dereference_sched(tr->function_pids);
5342 
5343 	this_cpu_write(tr->trace_buffer.data->ftrace_ignore_pid,
5344 		       trace_ignore_this_task(pid_list, next));
5345 }
5346 
5347 static void clear_ftrace_pids(struct trace_array *tr)
5348 {
5349 	struct trace_pid_list *pid_list;
5350 	int cpu;
5351 
5352 	pid_list = rcu_dereference_protected(tr->function_pids,
5353 					     lockdep_is_held(&ftrace_lock));
5354 	if (!pid_list)
5355 		return;
5356 
5357 	unregister_trace_sched_switch(ftrace_filter_pid_sched_switch_probe, tr);
5358 
5359 	for_each_possible_cpu(cpu)
5360 		per_cpu_ptr(tr->trace_buffer.data, cpu)->ftrace_ignore_pid = false;
5361 
5362 	rcu_assign_pointer(tr->function_pids, NULL);
5363 
5364 	/* Wait till all users are no longer using pid filtering */
5365 	synchronize_sched();
5366 
5367 	trace_free_pid_list(pid_list);
5368 }
5369 
5370 static void ftrace_pid_reset(struct trace_array *tr)
5371 {
5372 	mutex_lock(&ftrace_lock);
5373 	clear_ftrace_pids(tr);
5374 
5375 	ftrace_update_pid_func();
5376 	ftrace_startup_all(0);
5377 
5378 	mutex_unlock(&ftrace_lock);
5379 }
5380 
5381 /* Greater than any max PID */
5382 #define FTRACE_NO_PIDS		(void *)(PID_MAX_LIMIT + 1)
5383 
5384 static void *fpid_start(struct seq_file *m, loff_t *pos)
5385 	__acquires(RCU)
5386 {
5387 	struct trace_pid_list *pid_list;
5388 	struct trace_array *tr = m->private;
5389 
5390 	mutex_lock(&ftrace_lock);
5391 	rcu_read_lock_sched();
5392 
5393 	pid_list = rcu_dereference_sched(tr->function_pids);
5394 
5395 	if (!pid_list)
5396 		return !(*pos) ? FTRACE_NO_PIDS : NULL;
5397 
5398 	return trace_pid_start(pid_list, pos);
5399 }
5400 
5401 static void *fpid_next(struct seq_file *m, void *v, loff_t *pos)
5402 {
5403 	struct trace_array *tr = m->private;
5404 	struct trace_pid_list *pid_list = rcu_dereference_sched(tr->function_pids);
5405 
5406 	if (v == FTRACE_NO_PIDS)
5407 		return NULL;
5408 
5409 	return trace_pid_next(pid_list, v, pos);
5410 }
5411 
5412 static void fpid_stop(struct seq_file *m, void *p)
5413 	__releases(RCU)
5414 {
5415 	rcu_read_unlock_sched();
5416 	mutex_unlock(&ftrace_lock);
5417 }
5418 
5419 static int fpid_show(struct seq_file *m, void *v)
5420 {
5421 	if (v == FTRACE_NO_PIDS) {
5422 		seq_puts(m, "no pid\n");
5423 		return 0;
5424 	}
5425 
5426 	return trace_pid_show(m, v);
5427 }
5428 
5429 static const struct seq_operations ftrace_pid_sops = {
5430 	.start = fpid_start,
5431 	.next = fpid_next,
5432 	.stop = fpid_stop,
5433 	.show = fpid_show,
5434 };
5435 
5436 static int
5437 ftrace_pid_open(struct inode *inode, struct file *file)
5438 {
5439 	struct trace_array *tr = inode->i_private;
5440 	struct seq_file *m;
5441 	int ret = 0;
5442 
5443 	if (trace_array_get(tr) < 0)
5444 		return -ENODEV;
5445 
5446 	if ((file->f_mode & FMODE_WRITE) &&
5447 	    (file->f_flags & O_TRUNC))
5448 		ftrace_pid_reset(tr);
5449 
5450 	ret = seq_open(file, &ftrace_pid_sops);
5451 	if (ret < 0) {
5452 		trace_array_put(tr);
5453 	} else {
5454 		m = file->private_data;
5455 		/* copy tr over to seq ops */
5456 		m->private = tr;
5457 	}
5458 
5459 	return ret;
5460 }
5461 
5462 static void ignore_task_cpu(void *data)
5463 {
5464 	struct trace_array *tr = data;
5465 	struct trace_pid_list *pid_list;
5466 
5467 	/*
5468 	 * This function is called by on_each_cpu() while the
5469 	 * event_mutex is held.
5470 	 */
5471 	pid_list = rcu_dereference_protected(tr->function_pids,
5472 					     mutex_is_locked(&ftrace_lock));
5473 
5474 	this_cpu_write(tr->trace_buffer.data->ftrace_ignore_pid,
5475 		       trace_ignore_this_task(pid_list, current));
5476 }
5477 
5478 static ssize_t
5479 ftrace_pid_write(struct file *filp, const char __user *ubuf,
5480 		   size_t cnt, loff_t *ppos)
5481 {
5482 	struct seq_file *m = filp->private_data;
5483 	struct trace_array *tr = m->private;
5484 	struct trace_pid_list *filtered_pids = NULL;
5485 	struct trace_pid_list *pid_list;
5486 	ssize_t ret;
5487 
5488 	if (!cnt)
5489 		return 0;
5490 
5491 	mutex_lock(&ftrace_lock);
5492 
5493 	filtered_pids = rcu_dereference_protected(tr->function_pids,
5494 					     lockdep_is_held(&ftrace_lock));
5495 
5496 	ret = trace_pid_write(filtered_pids, &pid_list, ubuf, cnt);
5497 	if (ret < 0)
5498 		goto out;
5499 
5500 	rcu_assign_pointer(tr->function_pids, pid_list);
5501 
5502 	if (filtered_pids) {
5503 		synchronize_sched();
5504 		trace_free_pid_list(filtered_pids);
5505 	} else if (pid_list) {
5506 		/* Register a probe to set whether to ignore the tracing of a task */
5507 		register_trace_sched_switch(ftrace_filter_pid_sched_switch_probe, tr);
5508 	}
5509 
5510 	/*
5511 	 * Ignoring of pids is done at task switch. But we have to
5512 	 * check for those tasks that are currently running.
5513 	 * Always do this in case a pid was appended or removed.
5514 	 */
5515 	on_each_cpu(ignore_task_cpu, tr, 1);
5516 
5517 	ftrace_update_pid_func();
5518 	ftrace_startup_all(0);
5519  out:
5520 	mutex_unlock(&ftrace_lock);
5521 
5522 	if (ret > 0)
5523 		*ppos += ret;
5524 
5525 	return ret;
5526 }
5527 
5528 static int
5529 ftrace_pid_release(struct inode *inode, struct file *file)
5530 {
5531 	struct trace_array *tr = inode->i_private;
5532 
5533 	trace_array_put(tr);
5534 
5535 	return seq_release(inode, file);
5536 }
5537 
5538 static const struct file_operations ftrace_pid_fops = {
5539 	.open		= ftrace_pid_open,
5540 	.write		= ftrace_pid_write,
5541 	.read		= seq_read,
5542 	.llseek		= tracing_lseek,
5543 	.release	= ftrace_pid_release,
5544 };
5545 
5546 void ftrace_init_tracefs(struct trace_array *tr, struct dentry *d_tracer)
5547 {
5548 	trace_create_file("set_ftrace_pid", 0644, d_tracer,
5549 			    tr, &ftrace_pid_fops);
5550 }
5551 
5552 void __init ftrace_init_tracefs_toplevel(struct trace_array *tr,
5553 					 struct dentry *d_tracer)
5554 {
5555 	/* Only the top level directory has the dyn_tracefs and profile */
5556 	WARN_ON(!(tr->flags & TRACE_ARRAY_FL_GLOBAL));
5557 
5558 	ftrace_init_dyn_tracefs(d_tracer);
5559 	ftrace_profile_tracefs(d_tracer);
5560 }
5561 
5562 /**
5563  * ftrace_kill - kill ftrace
5564  *
5565  * This function should be used by panic code. It stops ftrace
5566  * but in a not so nice way. If you need to simply kill ftrace
5567  * from a non-atomic section, use ftrace_kill.
5568  */
5569 void ftrace_kill(void)
5570 {
5571 	ftrace_disabled = 1;
5572 	ftrace_enabled = 0;
5573 	clear_ftrace_function();
5574 }
5575 
5576 /**
5577  * Test if ftrace is dead or not.
5578  */
5579 int ftrace_is_dead(void)
5580 {
5581 	return ftrace_disabled;
5582 }
5583 
5584 /**
5585  * register_ftrace_function - register a function for profiling
5586  * @ops - ops structure that holds the function for profiling.
5587  *
5588  * Register a function to be called by all functions in the
5589  * kernel.
5590  *
5591  * Note: @ops->func and all the functions it calls must be labeled
5592  *       with "notrace", otherwise it will go into a
5593  *       recursive loop.
5594  */
5595 int register_ftrace_function(struct ftrace_ops *ops)
5596 {
5597 	int ret = -1;
5598 
5599 	ftrace_ops_init(ops);
5600 
5601 	mutex_lock(&ftrace_lock);
5602 
5603 	ret = ftrace_startup(ops, 0);
5604 
5605 	mutex_unlock(&ftrace_lock);
5606 
5607 	return ret;
5608 }
5609 EXPORT_SYMBOL_GPL(register_ftrace_function);
5610 
5611 /**
5612  * unregister_ftrace_function - unregister a function for profiling.
5613  * @ops - ops structure that holds the function to unregister
5614  *
5615  * Unregister a function that was added to be called by ftrace profiling.
5616  */
5617 int unregister_ftrace_function(struct ftrace_ops *ops)
5618 {
5619 	int ret;
5620 
5621 	mutex_lock(&ftrace_lock);
5622 	ret = ftrace_shutdown(ops, 0);
5623 	mutex_unlock(&ftrace_lock);
5624 
5625 	return ret;
5626 }
5627 EXPORT_SYMBOL_GPL(unregister_ftrace_function);
5628 
5629 int
5630 ftrace_enable_sysctl(struct ctl_table *table, int write,
5631 		     void __user *buffer, size_t *lenp,
5632 		     loff_t *ppos)
5633 {
5634 	int ret = -ENODEV;
5635 
5636 	mutex_lock(&ftrace_lock);
5637 
5638 	if (unlikely(ftrace_disabled))
5639 		goto out;
5640 
5641 	ret = proc_dointvec(table, write, buffer, lenp, ppos);
5642 
5643 	if (ret || !write || (last_ftrace_enabled == !!ftrace_enabled))
5644 		goto out;
5645 
5646 	last_ftrace_enabled = !!ftrace_enabled;
5647 
5648 	if (ftrace_enabled) {
5649 
5650 		/* we are starting ftrace again */
5651 		if (ftrace_ops_list != &ftrace_list_end)
5652 			update_ftrace_function();
5653 
5654 		ftrace_startup_sysctl();
5655 
5656 	} else {
5657 		/* stopping ftrace calls (just send to ftrace_stub) */
5658 		ftrace_trace_function = ftrace_stub;
5659 
5660 		ftrace_shutdown_sysctl();
5661 	}
5662 
5663  out:
5664 	mutex_unlock(&ftrace_lock);
5665 	return ret;
5666 }
5667 
5668 #ifdef CONFIG_FUNCTION_GRAPH_TRACER
5669 
5670 static struct ftrace_ops graph_ops = {
5671 	.func			= ftrace_stub,
5672 	.flags			= FTRACE_OPS_FL_RECURSION_SAFE |
5673 				   FTRACE_OPS_FL_INITIALIZED |
5674 				   FTRACE_OPS_FL_PID |
5675 				   FTRACE_OPS_FL_STUB,
5676 #ifdef FTRACE_GRAPH_TRAMP_ADDR
5677 	.trampoline		= FTRACE_GRAPH_TRAMP_ADDR,
5678 	/* trampoline_size is only needed for dynamically allocated tramps */
5679 #endif
5680 	ASSIGN_OPS_HASH(graph_ops, &global_ops.local_hash)
5681 };
5682 
5683 void ftrace_graph_sleep_time_control(bool enable)
5684 {
5685 	fgraph_sleep_time = enable;
5686 }
5687 
5688 void ftrace_graph_graph_time_control(bool enable)
5689 {
5690 	fgraph_graph_time = enable;
5691 }
5692 
5693 int ftrace_graph_entry_stub(struct ftrace_graph_ent *trace)
5694 {
5695 	return 0;
5696 }
5697 
5698 /* The callbacks that hook a function */
5699 trace_func_graph_ret_t ftrace_graph_return =
5700 			(trace_func_graph_ret_t)ftrace_stub;
5701 trace_func_graph_ent_t ftrace_graph_entry = ftrace_graph_entry_stub;
5702 static trace_func_graph_ent_t __ftrace_graph_entry = ftrace_graph_entry_stub;
5703 
5704 /* Try to assign a return stack array on FTRACE_RETSTACK_ALLOC_SIZE tasks. */
5705 static int alloc_retstack_tasklist(struct ftrace_ret_stack **ret_stack_list)
5706 {
5707 	int i;
5708 	int ret = 0;
5709 	int start = 0, end = FTRACE_RETSTACK_ALLOC_SIZE;
5710 	struct task_struct *g, *t;
5711 
5712 	for (i = 0; i < FTRACE_RETSTACK_ALLOC_SIZE; i++) {
5713 		ret_stack_list[i] = kmalloc(FTRACE_RETFUNC_DEPTH
5714 					* sizeof(struct ftrace_ret_stack),
5715 					GFP_KERNEL);
5716 		if (!ret_stack_list[i]) {
5717 			start = 0;
5718 			end = i;
5719 			ret = -ENOMEM;
5720 			goto free;
5721 		}
5722 	}
5723 
5724 	read_lock(&tasklist_lock);
5725 	do_each_thread(g, t) {
5726 		if (start == end) {
5727 			ret = -EAGAIN;
5728 			goto unlock;
5729 		}
5730 
5731 		if (t->ret_stack == NULL) {
5732 			atomic_set(&t->tracing_graph_pause, 0);
5733 			atomic_set(&t->trace_overrun, 0);
5734 			t->curr_ret_stack = -1;
5735 			/* Make sure the tasks see the -1 first: */
5736 			smp_wmb();
5737 			t->ret_stack = ret_stack_list[start++];
5738 		}
5739 	} while_each_thread(g, t);
5740 
5741 unlock:
5742 	read_unlock(&tasklist_lock);
5743 free:
5744 	for (i = start; i < end; i++)
5745 		kfree(ret_stack_list[i]);
5746 	return ret;
5747 }
5748 
5749 static void
5750 ftrace_graph_probe_sched_switch(void *ignore, bool preempt,
5751 			struct task_struct *prev, struct task_struct *next)
5752 {
5753 	unsigned long long timestamp;
5754 	int index;
5755 
5756 	/*
5757 	 * Does the user want to count the time a function was asleep.
5758 	 * If so, do not update the time stamps.
5759 	 */
5760 	if (fgraph_sleep_time)
5761 		return;
5762 
5763 	timestamp = trace_clock_local();
5764 
5765 	prev->ftrace_timestamp = timestamp;
5766 
5767 	/* only process tasks that we timestamped */
5768 	if (!next->ftrace_timestamp)
5769 		return;
5770 
5771 	/*
5772 	 * Update all the counters in next to make up for the
5773 	 * time next was sleeping.
5774 	 */
5775 	timestamp -= next->ftrace_timestamp;
5776 
5777 	for (index = next->curr_ret_stack; index >= 0; index--)
5778 		next->ret_stack[index].calltime += timestamp;
5779 }
5780 
5781 /* Allocate a return stack for each task */
5782 static int start_graph_tracing(void)
5783 {
5784 	struct ftrace_ret_stack **ret_stack_list;
5785 	int ret, cpu;
5786 
5787 	ret_stack_list = kmalloc(FTRACE_RETSTACK_ALLOC_SIZE *
5788 				sizeof(struct ftrace_ret_stack *),
5789 				GFP_KERNEL);
5790 
5791 	if (!ret_stack_list)
5792 		return -ENOMEM;
5793 
5794 	/* The cpu_boot init_task->ret_stack will never be freed */
5795 	for_each_online_cpu(cpu) {
5796 		if (!idle_task(cpu)->ret_stack)
5797 			ftrace_graph_init_idle_task(idle_task(cpu), cpu);
5798 	}
5799 
5800 	do {
5801 		ret = alloc_retstack_tasklist(ret_stack_list);
5802 	} while (ret == -EAGAIN);
5803 
5804 	if (!ret) {
5805 		ret = register_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
5806 		if (ret)
5807 			pr_info("ftrace_graph: Couldn't activate tracepoint"
5808 				" probe to kernel_sched_switch\n");
5809 	}
5810 
5811 	kfree(ret_stack_list);
5812 	return ret;
5813 }
5814 
5815 /*
5816  * Hibernation protection.
5817  * The state of the current task is too much unstable during
5818  * suspend/restore to disk. We want to protect against that.
5819  */
5820 static int
5821 ftrace_suspend_notifier_call(struct notifier_block *bl, unsigned long state,
5822 							void *unused)
5823 {
5824 	switch (state) {
5825 	case PM_HIBERNATION_PREPARE:
5826 		pause_graph_tracing();
5827 		break;
5828 
5829 	case PM_POST_HIBERNATION:
5830 		unpause_graph_tracing();
5831 		break;
5832 	}
5833 	return NOTIFY_DONE;
5834 }
5835 
5836 static int ftrace_graph_entry_test(struct ftrace_graph_ent *trace)
5837 {
5838 	if (!ftrace_ops_test(&global_ops, trace->func, NULL))
5839 		return 0;
5840 	return __ftrace_graph_entry(trace);
5841 }
5842 
5843 /*
5844  * The function graph tracer should only trace the functions defined
5845  * by set_ftrace_filter and set_ftrace_notrace. If another function
5846  * tracer ops is registered, the graph tracer requires testing the
5847  * function against the global ops, and not just trace any function
5848  * that any ftrace_ops registered.
5849  */
5850 static void update_function_graph_func(void)
5851 {
5852 	struct ftrace_ops *op;
5853 	bool do_test = false;
5854 
5855 	/*
5856 	 * The graph and global ops share the same set of functions
5857 	 * to test. If any other ops is on the list, then
5858 	 * the graph tracing needs to test if its the function
5859 	 * it should call.
5860 	 */
5861 	do_for_each_ftrace_op(op, ftrace_ops_list) {
5862 		if (op != &global_ops && op != &graph_ops &&
5863 		    op != &ftrace_list_end) {
5864 			do_test = true;
5865 			/* in double loop, break out with goto */
5866 			goto out;
5867 		}
5868 	} while_for_each_ftrace_op(op);
5869  out:
5870 	if (do_test)
5871 		ftrace_graph_entry = ftrace_graph_entry_test;
5872 	else
5873 		ftrace_graph_entry = __ftrace_graph_entry;
5874 }
5875 
5876 static struct notifier_block ftrace_suspend_notifier = {
5877 	.notifier_call = ftrace_suspend_notifier_call,
5878 };
5879 
5880 int register_ftrace_graph(trace_func_graph_ret_t retfunc,
5881 			trace_func_graph_ent_t entryfunc)
5882 {
5883 	int ret = 0;
5884 
5885 	mutex_lock(&ftrace_lock);
5886 
5887 	/* we currently allow only one tracer registered at a time */
5888 	if (ftrace_graph_active) {
5889 		ret = -EBUSY;
5890 		goto out;
5891 	}
5892 
5893 	register_pm_notifier(&ftrace_suspend_notifier);
5894 
5895 	ftrace_graph_active++;
5896 	ret = start_graph_tracing();
5897 	if (ret) {
5898 		ftrace_graph_active--;
5899 		goto out;
5900 	}
5901 
5902 	ftrace_graph_return = retfunc;
5903 
5904 	/*
5905 	 * Update the indirect function to the entryfunc, and the
5906 	 * function that gets called to the entry_test first. Then
5907 	 * call the update fgraph entry function to determine if
5908 	 * the entryfunc should be called directly or not.
5909 	 */
5910 	__ftrace_graph_entry = entryfunc;
5911 	ftrace_graph_entry = ftrace_graph_entry_test;
5912 	update_function_graph_func();
5913 
5914 	ret = ftrace_startup(&graph_ops, FTRACE_START_FUNC_RET);
5915 out:
5916 	mutex_unlock(&ftrace_lock);
5917 	return ret;
5918 }
5919 
5920 void unregister_ftrace_graph(void)
5921 {
5922 	mutex_lock(&ftrace_lock);
5923 
5924 	if (unlikely(!ftrace_graph_active))
5925 		goto out;
5926 
5927 	ftrace_graph_active--;
5928 	ftrace_graph_return = (trace_func_graph_ret_t)ftrace_stub;
5929 	ftrace_graph_entry = ftrace_graph_entry_stub;
5930 	__ftrace_graph_entry = ftrace_graph_entry_stub;
5931 	ftrace_shutdown(&graph_ops, FTRACE_STOP_FUNC_RET);
5932 	unregister_pm_notifier(&ftrace_suspend_notifier);
5933 	unregister_trace_sched_switch(ftrace_graph_probe_sched_switch, NULL);
5934 
5935 #ifdef CONFIG_DYNAMIC_FTRACE
5936 	/*
5937 	 * Function graph does not allocate the trampoline, but
5938 	 * other global_ops do. We need to reset the ALLOC_TRAMP flag
5939 	 * if one was used.
5940 	 */
5941 	global_ops.trampoline = save_global_trampoline;
5942 	if (save_global_flags & FTRACE_OPS_FL_ALLOC_TRAMP)
5943 		global_ops.flags |= FTRACE_OPS_FL_ALLOC_TRAMP;
5944 #endif
5945 
5946  out:
5947 	mutex_unlock(&ftrace_lock);
5948 }
5949 
5950 static DEFINE_PER_CPU(struct ftrace_ret_stack *, idle_ret_stack);
5951 
5952 static void
5953 graph_init_task(struct task_struct *t, struct ftrace_ret_stack *ret_stack)
5954 {
5955 	atomic_set(&t->tracing_graph_pause, 0);
5956 	atomic_set(&t->trace_overrun, 0);
5957 	t->ftrace_timestamp = 0;
5958 	/* make curr_ret_stack visible before we add the ret_stack */
5959 	smp_wmb();
5960 	t->ret_stack = ret_stack;
5961 }
5962 
5963 /*
5964  * Allocate a return stack for the idle task. May be the first
5965  * time through, or it may be done by CPU hotplug online.
5966  */
5967 void ftrace_graph_init_idle_task(struct task_struct *t, int cpu)
5968 {
5969 	t->curr_ret_stack = -1;
5970 	/*
5971 	 * The idle task has no parent, it either has its own
5972 	 * stack or no stack at all.
5973 	 */
5974 	if (t->ret_stack)
5975 		WARN_ON(t->ret_stack != per_cpu(idle_ret_stack, cpu));
5976 
5977 	if (ftrace_graph_active) {
5978 		struct ftrace_ret_stack *ret_stack;
5979 
5980 		ret_stack = per_cpu(idle_ret_stack, cpu);
5981 		if (!ret_stack) {
5982 			ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
5983 					    * sizeof(struct ftrace_ret_stack),
5984 					    GFP_KERNEL);
5985 			if (!ret_stack)
5986 				return;
5987 			per_cpu(idle_ret_stack, cpu) = ret_stack;
5988 		}
5989 		graph_init_task(t, ret_stack);
5990 	}
5991 }
5992 
5993 /* Allocate a return stack for newly created task */
5994 void ftrace_graph_init_task(struct task_struct *t)
5995 {
5996 	/* Make sure we do not use the parent ret_stack */
5997 	t->ret_stack = NULL;
5998 	t->curr_ret_stack = -1;
5999 
6000 	if (ftrace_graph_active) {
6001 		struct ftrace_ret_stack *ret_stack;
6002 
6003 		ret_stack = kmalloc(FTRACE_RETFUNC_DEPTH
6004 				* sizeof(struct ftrace_ret_stack),
6005 				GFP_KERNEL);
6006 		if (!ret_stack)
6007 			return;
6008 		graph_init_task(t, ret_stack);
6009 	}
6010 }
6011 
6012 void ftrace_graph_exit_task(struct task_struct *t)
6013 {
6014 	struct ftrace_ret_stack	*ret_stack = t->ret_stack;
6015 
6016 	t->ret_stack = NULL;
6017 	/* NULL must become visible to IRQs before we free it: */
6018 	barrier();
6019 
6020 	kfree(ret_stack);
6021 }
6022 #endif
6023